Merge branch 'master' into dev
diff --git a/.gitattributes b/.gitattributes
new file mode 100644
index 0000000..4ad67d3
--- /dev/null
+++ b/.gitattributes
@@ -0,0 +1,4 @@
+/.travis.yml export-ignore
+/appveyor.yml export-ignore
+/ci export-ignore
+/.gitattributes export-ignore
diff --git a/.gitignore b/.gitignore
deleted file mode 100644
index db7a0c4..0000000
--- a/.gitignore
+++ /dev/null
@@ -1,14 +0,0 @@
-.DS_Store
-Makefile.in
-aclocal.m4
-ar-lib
-autom4te.cache
-compile
-config.guess
-config.h.in
-config.sub
-configure
-depcomp
-install-sh
-ltmain.sh
-missing
diff --git a/.travis.yml b/.travis.yml
index c1e36bd..73001cf 100644
--- a/.travis.yml
+++ b/.travis.yml
@@ -11,35 +11,41 @@
env: BUILD_OFFICIAL=1
osx_image: xcode7.3
- os: linux
+ dist: trusty
compiler: clang
env:
- CFLAGS="-O1 -g -fsanitize=address -fno-omit-frame-pointer"
- CONFIGURE_FLAGS="--disable-shared"
+ CMAKE_BUILD_TYPE=RelWithDebInfo
+ CFLAGS_RELWITHDEBINFO="-O1 -g -fsanitize=address -fno-omit-frame-pointer"
+ CMAKE_FLAGS="-DENABLE_SHARED=0"
ASAN_OPTIONS="detect_leaks=1 symbolize=1"
addons:
apt:
packages:
- nasm
- os: linux
+ dist: trusty
compiler: gcc
- env: CONFIGURE_FLAGS="--with-12bit"
+ env: CMAKE_FLAGS="-DWITH_12BIT=1"
- os: linux
+ dist: trusty
compiler: gcc
- env: CONFIGURE_FLAGS="--with-jpeg7"
+ env: CMAKE_FLAGS="-DWITH_JPEG7=1"
addons:
apt:
packages:
- nasm
- os: linux
+ dist: trusty
compiler: gcc
- env: CONFIGURE_FLAGS="--with-jpeg8"
+ env: CMAKE_FLAGS="-DWITH_JPEG8=1"
addons:
apt:
packages:
- nasm
- os: linux
+ dist: trusty
compiler: gcc
- env: CONFIGURE_FLAGS="--without-simd"
+ env: CMAKE_FLAGS="-DWITH_SIMD=0"
before_install:
- if [ "$TRAVIS_OS_NAME" = "osx" ]; then
@@ -80,17 +86,16 @@
fi
fi
- if [ "${BUILD_OFFICIAL:-}" == "" ]; then
- autoreconf -fiv &&
mkdir build &&
pushd build &&
- ../configure ${CONFIGURE_FLAGS} &&
+ cmake -G"Unix Makefiles" -DCMAKE_BUILD_TYPE=$CMAKE_BUILD_TYPE "-DCMAKE_C_FLAGS_RELWITHDEBINFO=$CFLAGS_RELWITHDEBINFO" $CMAKE_FLAGS .. &&
make -j &&
- if [[ "${CONFIGURE_FLAGS}" =~ "with-12bit" ||
- "${CONFIGURE_FLAGS}" =~ "without-simd" ]]; then
- make test FLOATTEST=32bit;
- else
- make test FLOATTEST=sse &&
- JSIMD_FORCENONE=1 make test FLOATTEST=32bit;
+ make test &&
+ if [[ ! "${CMAKE_FLAGS[0]}" =~ "WITH_12BIT" &&
+ ! "${CMAKE_FLAGS[0]}" =~ "WITH_SIMD" ]]; then
+ JSIMD_FORCESSE2=1 make test &&
+ cmake -DFLOATTEST=32bit .. &&
+ JSIMD_FORCENONE=1 make test;
fi &&
popd;
fi
diff --git a/BUILDING.md b/BUILDING.md
index 2725f30..c9f6468 100644
--- a/BUILDING.md
+++ b/BUILDING.md
@@ -1,26 +1,24 @@
-Un*x Platforms (including Mac and Cygwin)
-=========================================
+Building libjpeg-turbo
+======================
Build Requirements
------------------
-- autoconf 2.56 or later
-- automake 1.7 or later
-- libtool 1.4 or later
- * If using Xcode 4.3 or later on OS X, autoconf and automake are no longer
- provided. The easiest way to obtain them is from
- [MacPorts](http://www.MacPorts.org) or [Homebrew](http://brew.sh/).
+
+### All Systems
+
+- [CMake](http://www.cmake.org) v2.8.12 or later
- [NASM](http://www.nasm.us) or [YASM](http://yasm.tortall.net)
(if building x86 or x86-64 SIMD extensions)
- * If using NASM, 0.98, or 2.01 or later is required for an x86 build (0.99
- and 2.00 do not work properly with libjpeg-turbo's x86 SIMD code.)
- * If using NASM, 2.00 or later is required for an x86-64 build.
- * If using NASM, 2.07 or later (except 2.11.08) is required for an x86-64
- Mac build (2.11.08 does not work properly with libjpeg-turbo's x86-64 SIMD
- code when building macho64 objects.) NASM or YASM can be obtained from
+ * If using NASM, 2.10 or later is required.
+ * If using NASM, 2.10 or later (except 2.11.08) is required for an x86-64 Mac
+ build (2.11.08 does not work properly with libjpeg-turbo's x86-64 SIMD code
+ when building macho64 objects.) NASM or YASM can be obtained from
[MacPorts](http://www.macports.org/) or [Homebrew](http://brew.sh/).
+ * If using YASM, 1.2.0 or later is required.
+ * If building on Windows, **nasm.exe**/**yasm.exe** should be in your `PATH`.
The binary RPMs released by the NASM project do not work on older Linux
systems, such as Red Hat Enterprise Linux 5. On such systems, you can easily
@@ -36,6 +34,9 @@
NOTE: the NASM build will fail if texinfo is not installed.
+
+### Un*x Platforms (including Linux, Mac, FreeBSD, Solaris, and Cygwin)
+
- GCC v4.1 (or later) or Clang recommended for best performance
- If building the TurboJPEG Java wrapper, JDK or OpenJDK 1.5 or later is
@@ -47,509 +48,7 @@
<http://www.java.com>.
-Out-of-Tree Builds
-------------------
-
-Binary objects, libraries, and executables are generated in the directory from
-which `configure` is executed (the "binary directory"), and this directory need
-not necessarily be the same as the libjpeg-turbo source directory. You can
-create multiple independent binary directories, in which different versions of
-libjpeg-turbo can be built from the same source tree using different compilers
-or settings. In the sections below, *{build_directory}* refers to the binary
-directory, whereas *{source_directory}* refers to the libjpeg-turbo source
-directory. For in-tree builds, these directories are the same.
-
-
-Build Procedure
----------------
-
-The following procedure will build libjpeg-turbo on Unix and Unix-like systems.
-(On Solaris, this generates a 32-bit build. See "Build Recipes" below for
-64-bit build instructions.)
-
- cd {source_directory}
- autoreconf -fiv
- cd {build_directory}
- sh {source_directory}/configure [additional configure flags]
- make
-
-NOTE: Running autoreconf in the source directory is not necessary if building
-libjpeg-turbo from one of the official release tarballs.
-
-This will generate the following files under **.libs/**:
-
-**libjpeg.a**<br>
-Static link library for the libjpeg API
-
-**libjpeg.so.{version}** (Linux, Unix)<br>
-**libjpeg.{version}.dylib** (Mac)<br>
-**cygjpeg-{version}.dll** (Cygwin)<br>
-Shared library for the libjpeg API
-
-By default, *{version}* is 62.2.0, 7.2.0, or 8.1.2, depending on whether
-libjpeg v6b (default), v7, or v8 emulation is enabled. If using Cygwin,
-*{version}* is 62, 7, or 8.
-
-**libjpeg.so** (Linux, Unix)<br>
-**libjpeg.dylib** (Mac)<br>
-Development symlink for the libjpeg API
-
-**libjpeg.dll.a** (Cygwin)<br>
-Import library for the libjpeg API
-
-**libturbojpeg.a**<br>
-Static link library for the TurboJPEG API
-
-**libturbojpeg.so.0.1.0** (Linux, Unix)<br>
-**libturbojpeg.0.1.0.dylib** (Mac)<br>
-**cygturbojpeg-0.dll** (Cygwin)<br>
-Shared library for the TurboJPEG API
-
-**libturbojpeg.so** (Linux, Unix)<br>
-**libturbojpeg.dylib** (Mac)<br>
-Development symlink for the TurboJPEG API
-
-**libturbojpeg.dll.a** (Cygwin)<br>
-Import library for the TurboJPEG API
-
-
-### libjpeg v7 or v8 API/ABI Emulation
-
-Add `--with-jpeg7` to the `configure` command line to build a version of
-libjpeg-turbo that is API/ABI-compatible with libjpeg v7. Add `--with-jpeg8`
-to the `configure` command to build a version of libjpeg-turbo that is
-API/ABI-compatible with libjpeg v8. See [README.md](README.md) for more
-information about libjpeg v7 and v8 emulation.
-
-
-### In-Memory Source/Destination Managers
-
-When using libjpeg v6b or v7 API/ABI emulation, add `--without-mem-srcdst` to
-the `configure` command line to build a version of libjpeg-turbo that lacks the
-`jpeg_mem_src()` and `jpeg_mem_dest()` functions. These functions were not
-part of the original libjpeg v6b and v7 APIs, so removing them ensures strict
-conformance with those APIs. See [README.md](README.md) for more information.
-
-
-### Arithmetic Coding Support
-
-Since the patent on arithmetic coding has expired, this functionality has been
-included in this release of libjpeg-turbo. libjpeg-turbo's implementation is
-based on the implementation in libjpeg v8, but it works when emulating libjpeg
-v7 or v6b as well. The default is to enable both arithmetic encoding and
-decoding, but those who have philosophical objections to arithmetic coding can
-add `--without-arith-enc` or `--without-arith-dec` to the `configure` command
-line to disable encoding or decoding (respectively.)
-
-
-### TurboJPEG Java Wrapper
-
-Add `--with-java` to the `configure` command line to incorporate an optional
-Java Native Interface (JNI) wrapper into the TurboJPEG shared library and build
-the Java front-end classes to support it. This allows the TurboJPEG shared
-library to be used directly from Java applications. See
-[java/README](java/README) for more details.
-
-You can set the `JAVAC`, `JAR`, and `JAVA` configure variables to specify
-alternate commands for javac, jar, and java (respectively.) You can also
-set the `JAVACFLAGS` configure variable to specify arguments that should be
-passed to the Java compiler when building the TurboJPEG classes, and
-`JNI_CFLAGS` to specify arguments that should be passed to the C compiler when
-building the JNI wrapper. Run `configure --help` for more details.
-
-
-Build Recipes
--------------
-
-
-### 32-bit Build on 64-bit Linux
-
-Add
-
- --host i686-pc-linux-gnu CFLAGS='-O3 -m32' LDFLAGS=-m32
-
-to the `configure` command line.
-
-
-### 64-bit Build on 64-bit OS X
-
-Add
-
- --host x86_64-apple-darwin NASM=/opt/local/bin/nasm
-
-to the `configure` command line. NASM 2.07 or later from MacPorts or Homebrew
-must be installed. If using Homebrew, then replace `/opt/local` with
-`/usr/local`.
-
-
-### 32-bit Build on 64-bit OS X
-
-Add
-
- --host i686-apple-darwin CFLAGS='-O3 -m32' LDFLAGS=-m32
-
-to the `configure` command line.
-
-
-### 64-bit Backward-Compatible Build on 64-bit OS X
-
-Add
-
- --host x86_64-apple-darwin NASM=/opt/local/bin/nasm \
- CFLAGS='-mmacosx-version-min=10.5 -O3' \
- LDFLAGS='-mmacosx-version-min=10.5'
-
-to the `configure` command line. NASM 2.07 or later from MacPorts or Homebrew
-must be installed. If using Homebrew, then replace `/opt/local` with
-`/usr/local`.
-
-
-### 32-bit Backward-Compatible Build on OS X
-
-Add
-
- --host i686-apple-darwin \
- CFLAGS='-mmacosx-version-min=10.5 -O3 -m32' \
- LDFLAGS='-mmacosx-version-min=10.5 -m32'
-
-to the `configure` command line.
-
-
-### 64-bit Build on 64-bit Solaris
-
-Add
-
- --host x86_64-pc-solaris CFLAGS='-O3 -m64' LDFLAGS=-m64
-
-to the `configure` command line.
-
-
-### 32-bit Build on 64-bit FreeBSD
-
-Add
-
- --host i386-unknown-freebsd CFLAGS='-O3 -m32' LDFLAGS=-m32
-
-to the `configure` command line. NASM 2.07 or later from FreeBSD ports must be
-installed.
-
-
-### Oracle Solaris Studio
-
-Add
-
- CC=cc
-
-to the `configure` command line. libjpeg-turbo will automatically be built
-with the maximum optimization level (-xO5) unless you override `CFLAGS`.
-
-To build a 64-bit version of libjpeg-turbo using Oracle Solaris Studio, add
-
- --host x86_64-pc-solaris CC=cc CFLAGS='-xO5 -m64' LDFLAGS=-m64
-
-to the `configure` command line.
-
-
-### MinGW Build on Cygwin
-
-Use CMake (see recipes below)
-
-
-Building libjpeg-turbo for iOS
-------------------------------
-
-iOS platforms, such as the iPhone and iPad, use ARM processors, and all
-currently supported models include NEON instructions. Thus, they can take
-advantage of libjpeg-turbo's SIMD extensions to significantly accelerate JPEG
-compression/decompression. This section describes how to build libjpeg-turbo
-for these platforms.
-
-
-### Additional build requirements
-
-- For configurations that require [gas-preprocessor.pl]
- (https://raw.githubusercontent.com/libjpeg-turbo/gas-preprocessor/master/gas-preprocessor.pl),
- it should be installed in your `PATH`.
-
-
-### ARMv7 (32-bit)
-
-**gas-preprocessor.pl required**
-
-The following scripts demonstrate how to build libjpeg-turbo to run on the
-iPhone 3GS-4S/iPad 1st-3rd Generation and newer:
-
-#### Xcode 4.2 and earlier (LLVM-GCC)
-
- IOS_PLATFORMDIR=/Developer/Platforms/iPhoneOS.platform
- IOS_SYSROOT=($IOS_PLATFORMDIR/Developer/SDKs/iPhoneOS*.sdk)
-
- export host_alias=arm-apple-darwin10
- export CC=${IOS_PLATFORMDIR}/Developer/usr/bin/arm-apple-darwin10-llvm-gcc-4.2
- export CFLAGS="-mfloat-abi=softfp -isysroot ${IOS_SYSROOT[0]} -O3 -march=armv7 -mcpu=cortex-a8 -mtune=cortex-a8 -mfpu=neon -miphoneos-version-min=3.0"
-
- cd {build_directory}
- sh {source_directory}/configure [additional configure flags]
- make
-
-#### Xcode 4.3-4.6 (LLVM-GCC)
-
-Same as above, but replace the first line with:
-
- IOS_PLATFORMDIR=/Applications/Xcode.app/Contents/Developer/Platforms/iPhoneOS.platform
-
-#### Xcode 5 and later (Clang)
-
- IOS_PLATFORMDIR=/Applications/Xcode.app/Contents/Developer/Platforms/iPhoneOS.platform
- IOS_SYSROOT=($IOS_PLATFORMDIR/Developer/SDKs/iPhoneOS*.sdk)
-
- export host_alias=arm-apple-darwin10
- export CC=/Applications/Xcode.app/Contents/Developer/Toolchains/XcodeDefault.xctoolchain/usr/bin/clang
- export CFLAGS="-mfloat-abi=softfp -isysroot ${IOS_SYSROOT[0]} -O3 -arch armv7 -miphoneos-version-min=3.0"
- export CCASFLAGS="$CFLAGS -no-integrated-as"
-
- cd {build_directory}
- sh {source_directory}/configure [additional configure flags]
- make
-
-
-### ARMv7s (32-bit)
-
-**gas-preprocessor.pl required**
-
-The following scripts demonstrate how to build libjpeg-turbo to run on the
-iPhone 5/iPad 4th Generation and newer:
-
-#### Xcode 4.5-4.6 (LLVM-GCC)
-
- IOS_PLATFORMDIR=/Applications/Xcode.app/Contents/Developer/Platforms/iPhoneOS.platform
- IOS_SYSROOT=($IOS_PLATFORMDIR/Developer/SDKs/iPhoneOS*.sdk)
-
- export host_alias=arm-apple-darwin10
- export CC=${IOS_PLATFORMDIR}/Developer/usr/bin/arm-apple-darwin10-llvm-gcc-4.2
- export CFLAGS="-mfloat-abi=softfp -isysroot ${IOS_SYSROOT[0]} -O3 -march=armv7s -mcpu=swift -mtune=swift -mfpu=neon -miphoneos-version-min=6.0"
-
- cd {build_directory}
- sh {source_directory}/configure [additional configure flags]
- make
-
-#### Xcode 5 and later (Clang)
-
-Same as the ARMv7 build procedure for Xcode 5 and later, except replace the
-compiler flags as follows:
-
- export CFLAGS="-mfloat-abi=softfp -isysroot ${IOS_SYSROOT[0]} -O3 -arch armv7s -miphoneos-version-min=6.0"
-
-
-### ARMv8 (64-bit)
-
-**gas-preprocessor.pl required if using Xcode < 6**
-
-The following script demonstrates how to build libjpeg-turbo to run on the
-iPhone 5S/iPad Mini 2/iPad Air and newer.
-
- IOS_PLATFORMDIR=/Applications/Xcode.app/Contents/Developer/Platforms/iPhoneOS.platform
- IOS_SYSROOT=($IOS_PLATFORMDIR/Developer/SDKs/iPhoneOS*.sdk)
-
- export host_alias=aarch64-apple-darwin
- export CC=/Applications/Xcode.app/Contents/Developer/Toolchains/XcodeDefault.xctoolchain/usr/bin/clang
- export CFLAGS="-isysroot ${IOS_SYSROOT[0]} -O3 -arch arm64 -miphoneos-version-min=7.0 -funwind-tables"
-
- cd {build_directory}
- sh {source_directory}/configure [additional configure flags]
- make
-
-Once built, lipo can be used to combine the ARMv7, v7s, and/or v8 variants into
-a universal library.
-
-
-Building libjpeg-turbo for Android
-----------------------------------
-
-Building libjpeg-turbo for Android platforms requires the
-[Android NDK](https://developer.android.com/tools/sdk/ndk) and autotools.
-
-
-### ARMv7 (32-bit)
-
-The following is a general recipe script that can be modified for your specific
-needs.
-
- # Set these variables to suit your needs
- NDK_PATH={full path to the "ndk" directory-- for example, /opt/android/sdk/ndk-bundle}
- BUILD_PLATFORM={the platform name for the NDK package you installed--
- for example, "windows-x86" or "linux-x86_64" or "darwin-x86_64"}
- TOOLCHAIN_VERSION={"4.8", "4.9", "clang3.5", etc. This corresponds to a
- toolchain directory under ${NDK_PATH}/toolchains/.}
- ANDROID_VERSION={The minimum version of Android to support-- for example,
- "16", "19", etc.}
-
- # It should not be necessary to modify the rest
- HOST=arm-linux-androideabi
- SYSROOT=${NDK_PATH}/platforms/android-${ANDROID_VERSION}/arch-arm
- ANDROID_CFLAGS="-march=armv7-a -mfloat-abi=softfp -fprefetch-loop-arrays \
- --sysroot=${SYSROOT}"
-
- TOOLCHAIN=${NDK_PATH}/toolchains/${HOST}-${TOOLCHAIN_VERSION}/prebuilt/${BUILD_PLATFORM}
- export CPP=${TOOLCHAIN}/bin/${HOST}-cpp
- export AR=${TOOLCHAIN}/bin/${HOST}-ar
- export NM=${TOOLCHAIN}/bin/${HOST}-nm
- export CC=${TOOLCHAIN}/bin/${HOST}-gcc
- export LD=${TOOLCHAIN}/bin/${HOST}-ld
- export RANLIB=${TOOLCHAIN}/bin/${HOST}-ranlib
- export OBJDUMP=${TOOLCHAIN}/bin/${HOST}-objdump
- export STRIP=${TOOLCHAIN}/bin/${HOST}-strip
- cd {build_directory}
- sh {source_directory}/configure --host=${HOST} \
- CFLAGS="${ANDROID_CFLAGS} -O3 -fPIE" \
- CPPFLAGS="${ANDROID_CFLAGS}" \
- LDFLAGS="${ANDROID_CFLAGS} -pie" --with-simd ${1+"$@"}
- make
-
-
-### ARMv8 (64-bit)
-
-The following is a general recipe script that can be modified for your specific
-needs.
-
- # Set these variables to suit your needs
- NDK_PATH={full path to the "ndk" directory-- for example, /opt/android/sdk/ndk-bundle}
- BUILD_PLATFORM={the platform name for the NDK package you installed--
- for example, "windows-x86" or "linux-x86_64" or "darwin-x86_64"}
- TOOLCHAIN_VERSION={"4.8", "4.9", "clang3.5", etc. This corresponds to a
- toolchain directory under ${NDK_PATH}/toolchains/.}
- ANDROID_VERSION={The minimum version of Android to support. "21" or later
- is required for a 64-bit build.}
-
- # It should not be necessary to modify the rest
- HOST=aarch64-linux-android
- SYSROOT=${NDK_PATH}/platforms/android-${ANDROID_VERSION}/arch-arm64
- ANDROID_CFLAGS="--sysroot=${SYSROOT}"
-
- TOOLCHAIN=${NDK_PATH}/toolchains/${HOST}-${TOOLCHAIN_VERSION}/prebuilt/${BUILD_PLATFORM}
- export CPP=${TOOLCHAIN}/bin/${HOST}-cpp
- export AR=${TOOLCHAIN}/bin/${HOST}-ar
- export NM=${TOOLCHAIN}/bin/${HOST}-nm
- export CC=${TOOLCHAIN}/bin/${HOST}-gcc
- export LD=${TOOLCHAIN}/bin/${HOST}-ld
- export RANLIB=${TOOLCHAIN}/bin/${HOST}-ranlib
- export OBJDUMP=${TOOLCHAIN}/bin/${HOST}-objdump
- export STRIP=${TOOLCHAIN}/bin/${HOST}-strip
- cd {build_directory}
- sh {source_directory}/configure --host=${HOST} \
- CFLAGS="${ANDROID_CFLAGS} -O3 -fPIE" \
- CPPFLAGS="${ANDROID_CFLAGS}" \
- LDFLAGS="${ANDROID_CFLAGS} -pie" --with-simd ${1+"$@"}
- make
-
-
-### x86 (32-bit)
-
-The following is a general recipe script that can be modified for your specific
-needs.
-
- # Set these variables to suit your needs
- NDK_PATH={full path to the "ndk" directory-- for example, /opt/android/sdk/ndk-bundle}
- BUILD_PLATFORM={the platform name for the NDK package you installed--
- for example, "windows-x86" or "linux-x86_64" or "darwin-x86_64"}
- TOOLCHAIN_VERSION={"4.8", "4.9", "clang3.5", etc. This corresponds to a
- toolchain directory under ${NDK_PATH}/toolchains/.}
- ANDROID_VERSION={The minimum version of Android to support-- for example,
- "16", "19", etc.}
-
- # It should not be necessary to modify the rest
- HOST=i686-linux-android
- SYSROOT=${NDK_PATH}/platforms/android-${ANDROID_VERSION}/arch-x86
- ANDROID_CFLAGS="--sysroot=${SYSROOT}"
-
- TOOLCHAIN=${NDK_PATH}/toolchains/x86-${TOOLCHAIN_VERSION}/prebuilt/${BUILD_PLATFORM}
- export CPP=${TOOLCHAIN}/bin/${HOST}-cpp
- export AR=${TOOLCHAIN}/bin/${HOST}-ar
- export NM=${TOOLCHAIN}/bin/${HOST}-nm
- export CC=${TOOLCHAIN}/bin/${HOST}-gcc
- export LD=${TOOLCHAIN}/bin/${HOST}-ld
- export RANLIB=${TOOLCHAIN}/bin/${HOST}-ranlib
- export OBJDUMP=${TOOLCHAIN}/bin/${HOST}-objdump
- export STRIP=${TOOLCHAIN}/bin/${HOST}-strip
- cd {build_directory}
- sh {source_directory}/configure --host=${HOST} \
- CFLAGS="${ANDROID_CFLAGS} -O3 -fPIE" \
- CPPFLAGS="${ANDROID_CFLAGS}" \
- LDFLAGS="${ANDROID_CFLAGS} -pie" --with-simd ${1+"$@"}
- make
-
-
-### x86-64 (64-bit)
-
-The following is a general recipe script that can be modified for your specific
-needs.
-
- # Set these variables to suit your needs
- NDK_PATH={full path to the "ndk" directory-- for example, /opt/android/sdk/ndk-bundle}
- BUILD_PLATFORM={the platform name for the NDK package you installed--
- for example, "windows-x86" or "linux-x86_64" or "darwin-x86_64"}
- TOOLCHAIN_VERSION={"4.8", "4.9", "clang3.5", etc. This corresponds to a
- toolchain directory under ${NDK_PATH}/toolchains/.}
- ANDROID_VERSION={The minimum version of Android to support. "21" or later
- is required for a 64-bit build.}
-
- # It should not be necessary to modify the rest
- HOST=x86_64-linux-android
- SYSROOT=${NDK_PATH}/platforms/android-${ANDROID_VERSION}/arch-x86_64
- ANDROID_CFLAGS="--sysroot=${SYSROOT}"
-
- TOOLCHAIN=${NDK_PATH}/toolchains/x86_64-${TOOLCHAIN_VERSION}/prebuilt/${BUILD_PLATFORM}
- export CPP=${TOOLCHAIN}/bin/${HOST}-cpp
- export AR=${TOOLCHAIN}/bin/${HOST}-ar
- export NM=${TOOLCHAIN}/bin/${HOST}-nm
- export CC=${TOOLCHAIN}/bin/${HOST}-gcc
- export LD=${TOOLCHAIN}/bin/${HOST}-ld
- export RANLIB=${TOOLCHAIN}/bin/${HOST}-ranlib
- export OBJDUMP=${TOOLCHAIN}/bin/${HOST}-objdump
- export STRIP=${TOOLCHAIN}/bin/${HOST}-strip
- cd {build_directory}
- sh {source_directory}/configure --host=${HOST} \
- CFLAGS="${ANDROID_CFLAGS} -O3 -fPIE" \
- CPPFLAGS="${ANDROID_CFLAGS}" \
- LDFLAGS="${ANDROID_CFLAGS} -pie" --with-simd ${1+"$@"}
- make
-
-
-If building for Android 4.0.x (API level < 16) or earlier, remove `-fPIE` from
-`CFLAGS` and `-pie` from `LDFLAGS`.
-
-
-Installing libjpeg-turbo
-------------------------
-
-To install libjpeg-turbo after it is built, replace `make` in the build
-instructions with `make install`.
-
-The `--prefix` argument to configure (or the `prefix` configure variable) can
-be used to specify an installation directory of your choosing. If you don't
-specify an installation directory, then the default is to install libjpeg-turbo
-under **/opt/libjpeg-turbo** and to place the libraries in
-**/opt/libjpeg-turbo/lib32** (32-bit) or **/opt/libjpeg-turbo/lib64** (64-bit.)
-
-The `bindir`, `datadir`, `docdir`, `includedir`, `libdir`, and `mandir`
-configure variables allow a finer degree of control over where specific files in
-the libjpeg-turbo distribution should be installed. These variables can either
-be specified at configure time or passed as arguments to `make install`.
-
-
-Windows (Visual C++ or MinGW)
-=============================
-
-
-Build Requirements
-------------------
-
-- [CMake](http://www.cmake.org) v2.8.11 or later
-
-- [NASM](http://www.nasm.us) or [YASM](http://yasm.tortall.net)
- * If using NASM, 0.98 or later is required for an x86 build.
- * If using NASM, 2.05 or later is required for an x86-64 build.
- * **nasm.exe**/**yasm.exe** should be in your `PATH`.
+### Windows
- Microsoft Visual C++ 2005 or later
@@ -605,6 +104,53 @@
well.
+### Un*x
+
+The following procedure will build libjpeg-turbo on Unix and Unix-like systems.
+(On Solaris, this generates a 32-bit build. See "Build Recipes" below for
+64-bit build instructions.)
+
+ cd {build_directory}
+ cmake -G"Unix Makefiles" [additional CMake flags] {source_directory}
+ make
+
+This will generate the following files under *{build_directory}*:
+
+**libjpeg.a**<br>
+Static link library for the libjpeg API
+
+**libjpeg.so.{version}** (Linux, Unix)<br>
+**libjpeg.{version}.dylib** (Mac)<br>
+**cygjpeg-{version}.dll** (Cygwin)<br>
+Shared library for the libjpeg API
+
+By default, *{version}* is 62.2.0, 7.2.0, or 8.1.2, depending on whether
+libjpeg v6b (default), v7, or v8 emulation is enabled. If using Cygwin,
+*{version}* is 62, 7, or 8.
+
+**libjpeg.so** (Linux, Unix)<br>
+**libjpeg.dylib** (Mac)<br>
+Development symlink for the libjpeg API
+
+**libjpeg.dll.a** (Cygwin)<br>
+Import library for the libjpeg API
+
+**libturbojpeg.a**<br>
+Static link library for the TurboJPEG API
+
+**libturbojpeg.so.0.2.0** (Linux, Unix)<br>
+**libturbojpeg.0.2.0.dylib** (Mac)<br>
+**cygturbojpeg-0.dll** (Cygwin)<br>
+Shared library for the TurboJPEG API
+
+**libturbojpeg.so** (Linux, Unix)<br>
+**libturbojpeg.dylib** (Mac)<br>
+Development symlink for the TurboJPEG API
+
+**libturbojpeg.dll.a** (Cygwin)<br>
+Import library for the TurboJPEG API
+
+
### Visual C++ (Command Line)
cd {build_directory}
@@ -619,10 +165,10 @@
**jpeg-static.lib**<br>
Static link library for the libjpeg API
-**sharedlib/jpeg{version}.dll**<br>
+**jpeg{version}.dll**<br>
DLL for the libjpeg API
-**sharedlib/jpeg.lib**<br>
+**jpeg.lib**<br>
Import library for the libjpeg API
**turbojpeg-static.lib**<br>
@@ -647,9 +193,9 @@
cd {build_directory}
cmake -G"Visual Studio 10" [additional CMake flags] {source_directory}
-NOTE: Add "Win64" to the generator name (for example, "Visual Studio 10
-Win64") to build a 64-bit version of libjpeg-turbo. A separate build directory
-must be used for 32-bit and 64-bit builds.
+NOTE: Add "Win64" to the generator name (for example, "Visual Studio 10 Win64")
+to build a 64-bit version of libjpeg-turbo. A separate build directory must be
+used for 32-bit and 64-bit builds.
You can then open **ALL_BUILD.vcproj** in Visual Studio and build one of the
configurations in that project ("Debug", "Release", etc.) to generate a full
@@ -660,10 +206,10 @@
**{configuration}/jpeg-static.lib**<br>
Static link library for the libjpeg API
-**sharedlib/{configuration}/jpeg{version}.dll**<br>
+**{configuration}/jpeg{version}.dll**<br>
DLL for the libjpeg API
-**sharedlib/{configuration}/jpeg.lib**<br>
+**{configuration}/jpeg.lib**<br>
Import library for the libjpeg API
**{configuration}/turbojpeg-static.lib**<br>
@@ -695,10 +241,10 @@
**libjpeg.a**<br>
Static link library for the libjpeg API
-**sharedlib/libjpeg-{version}.dll**<br>
+**libjpeg-{version}.dll**<br>
DLL for the libjpeg API
-**sharedlib/libjpeg.dll.a**<br>
+**libjpeg.dll.a**<br>
Import library for the libjpeg API
**libturbojpeg.a**<br>
@@ -763,15 +309,42 @@
environment variable to the location of the JDK that you wish to use. The
`Java_JAVAC_EXECUTABLE`, `Java_JAVA_EXECUTABLE`, and `Java_JAR_EXECUTABLE`
CMake variables can also be used to specify alternate commands or locations for
-javac, jar, and java (respectively.) You can also set the `JAVACFLAGS` CMake
+javac, jar, and java (respectively.) You can also set the
+`CMAKE_JAVA_COMPILE_FLAGS` CMake variable or the `JAVAFLAGS` environment
variable to specify arguments that should be passed to the Java compiler when
-building the TurboJPEG classes.
+building the TurboJPEG classes, and the `JAVAARGS` CMake variable to specify
+arguments that should be passed to the JRE when running the TurboJPEG Java unit
+tests.
Build Recipes
-------------
+### 32-bit Build on 64-bit Linux/Unix/Mac
+
+Use export/setenv to set the following environment variables before running
+CMake:
+
+ CFLAGS=-m32
+ LDFLAGS=-m32
+
+
+### 64-bit Build on Solaris
+
+Use export/setenv to set the following environment variables before running
+CMake:
+
+ CFLAGS=-m64
+ LDFLAGS=-m64
+
+
+### Other Compilers
+
+On Un*x systems, prior to running CMake, you can set the `CC` environment
+variable to the command used to invoke the C compiler.
+
+
### 32-bit MinGW Build on Un*x (including Mac and Cygwin)
Create a file called **toolchain.cmake** under *{build_directory}*, with the
@@ -810,8 +383,306 @@
make
+Building libjpeg-turbo for iOS
+------------------------------
+
+iOS platforms, such as the iPhone and iPad, use ARM processors, and all
+currently supported models include NEON instructions. Thus, they can take
+advantage of libjpeg-turbo's SIMD extensions to significantly accelerate JPEG
+compression/decompression. This section describes how to build libjpeg-turbo
+for these platforms.
+
+
+### Additional build requirements
+
+- For configurations that require [gas-preprocessor.pl]
+ (https://raw.githubusercontent.com/libjpeg-turbo/gas-preprocessor/master/gas-preprocessor.pl),
+ it should be installed in your `PATH`.
+
+
+### ARMv7 (32-bit)
+
+**gas-preprocessor.pl required**
+
+The following scripts demonstrate how to build libjpeg-turbo to run on the
+iPhone 3GS-4S/iPad 1st-3rd Generation and newer:
+
+#### Xcode 4.2 and earlier (LLVM-GCC)
+
+ IOS_PLATFORMDIR=/Developer/Platforms/iPhoneOS.platform
+ IOS_SYSROOT=($IOS_PLATFORMDIR/Developer/SDKs/iPhoneOS*.sdk)
+ export CFLAGS="-mfloat-abi=softfp -march=armv7 -mcpu=cortex-a8 -mtune=cortex-a8 -mfpu=neon -miphoneos-version-min=3.0"
+
+ cat <<EOF >toolchain.cmake
+ set(CMAKE_SYSTEM_NAME Darwin)
+ set(CMAKE_SYSTEM_PROCESSOR arm)
+ set(CMAKE_C_COMPILER ${IOS_PLATFORMDIR}/Developer/usr/bin/arm-apple-darwin10-llvm-gcc-4.2)
+ EOF
+
+ cd {build_directory}
+ cmake -G"Unix Makefiles" -DCMAKE_TOOLCHAIN_FILE=toolchain.cmake \
+ -DCMAKE_OSX_SYSROOT=${IOS_SYSROOT[0]} \
+ [additional CMake flags] {source_directory}
+ make
+
+#### Xcode 4.3-4.6 (LLVM-GCC)
+
+Same as above, but replace the first line with:
+
+ IOS_PLATFORMDIR=/Applications/Xcode.app/Contents/Developer/Platforms/iPhoneOS.platform
+
+#### Xcode 5 and later (Clang)
+
+ IOS_PLATFORMDIR=/Applications/Xcode.app/Contents/Developer/Platforms/iPhoneOS.platform
+ IOS_SYSROOT=($IOS_PLATFORMDIR/Developer/SDKs/iPhoneOS*.sdk)
+ export CFLAGS="--mfloat-abi=softfp -arch armv7 -miphoneos-version-min=3.0"
+ export ASMFLAGS="-no-integrated-as"
+
+ cat <<EOF >toolchain.cmake
+ set(CMAKE_SYSTEM_NAME Darwin)
+ set(CMAKE_SYSTEM_PROCESSOR arm)
+ set(CMAKE_C_COMPILER /Applications/Xcode.app/Contents/Developer/Toolchains/XcodeDefault.xctoolchain/usr/bin/clang)
+ EOF
+
+ cd {build_directory}
+ cmake -G"Unix Makefiles" -DCMAKE_TOOLCHAIN_FILE=toolchain.cmake \
+ -DCMAKE_OSX_SYSROOT=${IOS_SYSROOT[0]} \
+ [additional CMake flags] {source_directory}
+ make
+
+
+### ARMv7s (32-bit)
+
+**gas-preprocessor.pl required**
+
+The following scripts demonstrate how to build libjpeg-turbo to run on the
+iPhone 5/iPad 4th Generation and newer:
+
+#### Xcode 4.5-4.6 (LLVM-GCC)
+
+ IOS_PLATFORMDIR=/Applications/Xcode.app/Contents/Developer/Platforms/iPhoneOS.platform
+ IOS_SYSROOT=($IOS_PLATFORMDIR/Developer/SDKs/iPhoneOS*.sdk)
+ export CFLAGS="-Wall -mfloat-abi=softfp -march=armv7s -mcpu=swift -mtune=swift -mfpu=neon -miphoneos-version-min=6.0"
+
+ cat <<EOF >toolchain.cmake
+ set(CMAKE_SYSTEM_NAME Darwin)
+ set(CMAKE_SYSTEM_PROCESSOR arm)
+ set(CMAKE_C_COMPILER ${IOS_PLATFORMDIR}/Developer/usr/bin/arm-apple-darwin10-llvm-gcc-4.2)
+ EOF
+
+ cd {build_directory}
+ cmake -G"Unix Makefiles" -DCMAKE_TOOLCHAIN_FILE=toolchain.cmake \
+ -DCMAKE_OSX_SYSROOT=${IOS_SYSROOT[0]} \
+ [additional CMake flags] {source_directory}
+ make
+
+#### Xcode 5 and later (Clang)
+
+Same as the ARMv7 build procedure for Xcode 5 and later, except replace the
+compiler flags as follows:
+
+ export CFLAGS="-Wall -mfloat-abi=softfp -arch armv7s -miphoneos-version-min=6.0"
+
+
+### ARMv8 (64-bit)
+
+**gas-preprocessor.pl required if using Xcode < 6**
+
+The following script demonstrates how to build libjpeg-turbo to run on the
+iPhone 5S/iPad Mini 2/iPad Air and newer.
+
+ IOS_PLATFORMDIR=/Applications/Xcode.app/Contents/Developer/Platforms/iPhoneOS.platform
+ IOS_SYSROOT=($IOS_PLATFORMDIR/Developer/SDKs/iPhoneOS*.sdk)
+ export CFLAGS="-Wall -arch arm64 -miphoneos-version-min=7.0 -funwind-tables"
+
+ cat <<EOF >toolchain.cmake
+ set(CMAKE_SYSTEM_NAME Darwin)
+ set(CMAKE_SYSTEM_PROCESSOR aarch64)
+ set(CMAKE_C_COMPILER /Applications/Xcode.app/Contents/Developer/Toolchains/XcodeDefault.xctoolchain/usr/bin/clang)
+ EOF
+
+ cd {build_directory}
+ cmake -G"Unix Makefiles" -DCMAKE_TOOLCHAIN_FILE=toolchain.cmake \
+ -DCMAKE_OSX_SYSROOT=${IOS_SYSROOT[0]} \
+ [additional CMake flags] {source_directory}
+ make
+
+Once built, lipo can be used to combine the ARMv7, v7s, and/or v8 variants into
+a universal library.
+
+
+Building libjpeg-turbo for Android
+----------------------------------
+
+Building libjpeg-turbo for Android platforms requires the
+[Android NDK](https://developer.android.com/tools/sdk/ndk).
+
+
+### ARMv7 (32-bit)
+
+The following is a general recipe script that can be modified for your specific
+needs.
+
+ # Set these variables to suit your needs
+ NDK_PATH={full path to the "ndk" directory-- for example, /opt/android/sdk/ndk-bundle}
+ BUILD_PLATFORM={the platform name for the NDK package you installed--
+ for example, "windows-x86" or "linux-x86_64" or "darwin-x86_64"}
+ TOOLCHAIN_VERSION={"4.8", "4.9", "clang3.5", etc. This corresponds to a
+ toolchain directory under ${NDK_PATH}/toolchains/.}
+ ANDROID_VERSION={The minimum version of Android to support-- for example,
+ "16", "19", etc.}
+
+ # It should not be necessary to modify the rest
+ HOST=arm-linux-androideabi
+ SYSROOT=${NDK_PATH}/platforms/android-${ANDROID_VERSION}/arch-arm
+ export CFLAGS="-march=armv7-a -mfloat-abi=softfp -fprefetch-loop-arrays --sysroot=${SYSROOT}"
+ export LDFLAGS=-pie
+ TOOLCHAIN=${NDK_PATH}/toolchains/${HOST}-${TOOLCHAIN_VERSION}/prebuilt/${BUILD_PLATFORM}
+
+ cat <<EOF >toolchain.cmake
+ set(CMAKE_SYSTEM_NAME Linux)
+ set(CMAKE_SYSTEM_PROCESSOR arm)
+ set(CMAKE_C_COMPILER ${TOOLCHAIN}/bin/${HOST}-gcc)
+ set(CMAKE_FIND_ROOT_PATH ${TOOLCHAIN}/${HOST})
+ EOF
+
+ cd {build_directory}
+ cmake -G"Unix Makefiles" -DCMAKE_TOOLCHAIN_FILE=toolchain.cmake \
+ -DCMAKE_POSITION_INDEPENDENT_CODE=1 \
+ [additional CMake flags] {source_directory}
+ make
+
+
+### ARMv8 (64-bit)
+
+The following is a general recipe script that can be modified for your specific
+needs.
+
+ # Set these variables to suit your needs
+ NDK_PATH={full path to the "ndk" directory-- for example, /opt/android/sdk/ndk-bundle}
+ BUILD_PLATFORM={the platform name for the NDK package you installed--
+ for example, "windows-x86" or "linux-x86_64" or "darwin-x86_64"}
+ TOOLCHAIN_VERSION={"4.8", "4.9", "clang3.5", etc. This corresponds to a
+ toolchain directory under ${NDK_PATH}/toolchains/.}
+ ANDROID_VERSION={The minimum version of Android to support. "21" or later
+ is required for a 64-bit build.}
+
+ # It should not be necessary to modify the rest
+ HOST=aarch64-linux-android
+ SYSROOT=${NDK_PATH}/platforms/android-${ANDROID_VERSION}/arch-arm64
+ export CFLAGS="--sysroot=${SYSROOT}"
+ export LDFLAGS=-pie
+ TOOLCHAIN=${NDK_PATH}/toolchains/${HOST}-${TOOLCHAIN_VERSION}/prebuilt/${BUILD_PLATFORM}
+
+ cat <<EOF >toolchain.cmake
+ set(CMAKE_SYSTEM_NAME Linux)
+ set(CMAKE_SYSTEM_PROCESSOR aarch64)
+ set(CMAKE_C_COMPILER ${TOOLCHAIN}/bin/${HOST}-gcc)
+ set(CMAKE_FIND_ROOT_PATH ${TOOLCHAIN}/${HOST})
+ EOF
+
+ cd {build_directory}
+ cmake -G"Unix Makefiles" -DCMAKE_TOOLCHAIN_FILE=toolchain.cmake \
+ -DCMAKE_POSITION_INDEPENDENT_CODE=1 \
+ [additional CMake flags] {source_directory}
+ make
+
+
+### x86 (32-bit)
+
+The following is a general recipe script that can be modified for your specific
+needs.
+
+ # Set these variables to suit your needs
+ NDK_PATH={full path to the "ndk" directory-- for example, /opt/android/sdk/ndk-bundle}
+ BUILD_PLATFORM={the platform name for the NDK package you installed--
+ for example, "windows-x86" or "linux-x86_64" or "darwin-x86_64"}
+ TOOLCHAIN_VERSION={"4.8", "4.9", "clang3.5", etc. This corresponds to a
+ toolchain directory under ${NDK_PATH}/toolchains/.}
+ ANDROID_VERSION={The minimum version of Android to support-- for example,
+ "16", "19", etc.}
+
+ # It should not be necessary to modify the rest
+ HOST=i686-linux-android
+ SYSROOT=${NDK_PATH}/platforms/android-${ANDROID_VERSION}/arch-x86
+ export CFLAGS="--sysroot=${SYSROOT}"
+ export LDFLAGS=-pie
+ TOOLCHAIN=${NDK_PATH}/toolchains/x86-${TOOLCHAIN_VERSION}/prebuilt/${BUILD_PLATFORM}
+
+ cat <<EOF >toolchain.cmake
+ set(CMAKE_SYSTEM_NAME Linux)
+ set(CMAKE_SYSTEM_PROCESSOR i386)
+ set(CMAKE_C_COMPILER ${TOOLCHAIN}/bin/${HOST}-gcc)
+ set(CMAKE_FIND_ROOT_PATH ${TOOLCHAIN}/${HOST})
+ EOF
+
+ cd {build_directory}
+ cmake -G"Unix Makefiles" -DCMAKE_TOOLCHAIN_FILE=toolchain.cmake \
+ -DCMAKE_POSITION_INDEPENDENT_CODE=1 \
+ [additional CMake flags] {source_directory}
+ make
+
+
+### x86-64 (64-bit)
+
+The following is a general recipe script that can be modified for your specific
+needs.
+
+ # Set these variables to suit your needs
+ NDK_PATH={full path to the "ndk" directory-- for example, /opt/android/sdk/ndk-bundle}
+ BUILD_PLATFORM={the platform name for the NDK package you installed--
+ for example, "windows-x86" or "linux-x86_64" or "darwin-x86_64"}
+ TOOLCHAIN_VERSION={"4.8", "4.9", "clang3.5", etc. This corresponds to a
+ toolchain directory under ${NDK_PATH}/toolchains/.}
+ ANDROID_VERSION={The minimum version of Android to support. "21" or later
+ is required for a 64-bit build.}
+
+ # It should not be necessary to modify the rest
+ HOST=x86_64-linux-android
+ SYSROOT=${NDK_PATH}/platforms/android-${ANDROID_VERSION}/arch-x86_64
+ export CFLAGS="--sysroot=${SYSROOT}"
+ export LDFLAGS=-pie
+ TOOLCHAIN=${NDK_PATH}/toolchains/x86_64-${TOOLCHAIN_VERSION}/prebuilt/${BUILD_PLATFORM}
+
+ cat <<EOF >toolchain.cmake
+ set(CMAKE_SYSTEM_NAME Linux)
+ set(CMAKE_SYSTEM_PROCESSOR x86_64)
+ set(CMAKE_C_COMPILER ${TOOLCHAIN}/bin/${HOST}-gcc)
+ set(CMAKE_FIND_ROOT_PATH ${TOOLCHAIN}/${HOST})
+ EOF
+
+ cd {build_directory}
+ cmake -G"Unix Makefiles" -DCMAKE_TOOLCHAIN_FILE=toolchain.cmake \
+ -DCMAKE_POSITION_INDEPENDENT_CODE=1 \
+ [additional CMake flags] {source_directory}
+ make
+
+
+If building for Android 4.0.x (API level < 16) or earlier, remove
+`-DCMAKE_POSITION_INDEPENDENT_CODE=1` from the CMake arguments and `-pie` from
+`LDFLAGS`.
+
+
+Advanced CMake Options
+----------------------
+
+To list and configure other CMake options not specifically mentioned in this
+guide, run
+
+ ccmake {source_directory}
+
+or
+
+ cmake-gui {source_directory}
+
+from the build directory after initially configuring the build. CCMake is a
+text-based interactive version of CMake, and CMake-GUI is a GUI version. Both
+will display all variables that are relevant to the libjpeg-turbo build, their
+current values, and a help string describing what they do.
+
+
Installing libjpeg-turbo
-------------------------
+========================
You can use the build system to install libjpeg-turbo (as opposed to creating
an installer package.) To do this, run `make install` or `nmake install`
@@ -835,6 +706,37 @@
**c:\libjpeg-turbo-gcc64**<br>
MinGW 64-bit build
+**/opt/libjpeg-turbo**<br>
+Un*x
+
+The default value of `CMAKE_INSTALL_PREFIX` causes the libjpeg-turbo files to
+be installed with a directory structure resembling that of the official
+libjpeg-turbo binary packages. Changing the value of `CMAKE_INSTALL_PREFIX`
+(for instance, to **/usr/local**) causes the libjpeg-turbo files to be
+installed with a directory structure that conforms to GNU standards.
+
+The `CMAKE_INSTALL_BINDIR`, `CMAKE_INSTALL_DATAROOTDIR`,
+`CMAKE_INSTALL_DOCDIR`, `CMAKE_INSTALL_INCLUDEDIR`, `CMAKE_INSTALL_JAVADIR`,
+`CMAKE_INSTALL_LIBDIR`, and `CMAKE_INSTALL_MANDIR` CMake variables allow a
+finer degree of control over where specific files in the libjpeg-turbo
+distribution should be installed. These directory variables can either be
+specified as absolute paths or as paths relative to `CMAKE_INSTALL_PREFIX` (for
+instance, setting `CMAKE_INSTALL_DOCDIR` to **doc** would cause the
+documentation to be installed in **${CMAKE\_INSTALL\_PREFIX}/doc**.) If a
+directory variable contains the name of another directory variable in angle
+brackets, then its final value will depend on the final value of that other
+variable. For instance, the default value of `CMAKE_INSTALL_MANDIR` is
+**\<CMAKE\_INSTALL\_DATAROOTDIR\>/man**.
+
+NOTE: If setting one of these directory variables to a relative path using the
+CMake command line, you must specify that the variable is of type `PATH`.
+For example:
+
+ cmake -G"{generator type}" -DCMAKE_INSTALL_LIBDIR:PATH=lib {source_directory}
+
+Otherwise, CMake will assume that the path is relative to the build directory
+rather than the install directory.
+
Creating Distribution Packages
==============================
@@ -871,40 +773,35 @@
Packages built in this manner can be installed on OS X 10.5 and later, but they
must be built on OS X 10.6 or later.
- make udmg [BUILDDIR32={32-bit build directory}]
+ make udmg
-On 64-bit OS X systems, this creates a Mac package/disk image that contains
-universal i386/x86-64 binaries. You should first configure a 32-bit
-out-of-tree build of libjpeg-turbo, then configure a 64-bit out-of-tree build,
-then run `make udmg` from the 64-bit build directory. The build system will
-look for the 32-bit build under *{source_directory}*/osxx86 by default, but you
-can override this by setting the `BUILDDIR32` variable on the make command line
-as shown above.
+This creates a Mac package/disk image that contains universal x86-64/i386/ARM
+binaries. The following CMake variables control which architectures are
+included in the universal binaries. Setting any of these variables to an empty
+string excludes that architecture from the package.
- make iosdmg [BUILDDIR32={32-bit build directory}] \
- [BUILDDIRARMV7={ARMv7 build directory}] \
- [BUILDDIRARMV7S={ARMv7s build directory}] \
- [BUILDDIRARMV8={ARMv8 build directory}]
+* `OSX_32BIT_BUILD`: Directory containing an i386 (32-bit) Mac build of
+ libjpeg-turbo (default: *{source_directory}*/osxx86)
+* `IOS_ARMV7_BUILD`: Directory containing an ARMv7 (32-bit) iOS build of
+ libjpeg-turbo (default: *{source_directory}*/iosarmv7)
+* `IOS_ARMV7S_BUILD`: Directory containing an ARMv7s (32-bit) iOS build of
+ libjpeg-turbo (default: *{source_directory}*/iosarmv7s)
+* `IOS_ARMV8_BUILD`: Directory containing an ARMv8 (64-bit) iOS build of
+ libjpeg-turbo (default: *{source_directory}*/iosarmv8)
-This creates a Mac package/disk image in which the libjpeg-turbo libraries
-contain ARM architectures necessary to build iOS applications. If building on
-an x86-64 system, the binaries will also contain the i386 architecture, as with
-`make udmg` above. You should first configure ARMv7, ARMv7s, and/or ARMv8
-out-of-tree builds of libjpeg-turbo (see "Building libjpeg-turbo for iOS"
-above.) If you are building an x86-64 version of libjpeg-turbo, you should
-configure a 32-bit out-of-tree build as well. Next, build libjpeg-turbo as you
-would normally, using an out-of-tree build. When it is built, run `make
-iosdmg` from the build directory. The build system will look for the ARMv7
-build under *{source_directory}*/iosarmv7 by default, the ARMv7s build under
-*{source_directory}*/iosarmv7s by default, the ARMv8 build under
-*{source_directory}*/iosarmv8 by default, and (if applicable) the 32-bit build
-under *{source_directory}*/osxx86 by default, but you can override this by
-setting the `BUILDDIR32`, `BUILDDIRARMV7`, `BUILDDIRARMV7S`, and/or
-`BUILDDIRARMV8` variables on the `make` command line as shown above.
+You should first use CMake to configure i386, ARMv7, ARMv7s, and/or ARMv8
+sub-builds of libjpeg-turbo (see "Build Recipes" and "Building libjpeg-turbo
+for iOS" above) in build directories that match those specified in the
+aforementioned CMake variables. Next, configure the primary build of
+libjpeg-turbo as an out-of-tree build, and build it. Once the primary build
+has been built, run `make udmg` from the build directory. The packaging system
+will build the sub-builds, use lipo to combine them into a single set of
+universal binaries, then package the universal binaries in the same manner as
+`make dmg`.
-NOTE: If including an ARMv8 build in the package, then you may need to use
-Xcode's version of lipo instead of the operating system's. To do this, pass
-an argument of `LIPO="xcrun lipo"` on the make command line.
+
+Cygwin
+------
make cygwinpkg
diff --git a/CMakeLists.txt b/CMakeLists.txt
index fb5e182..a2a4573 100644
--- a/CMakeLists.txt
+++ b/CMakeLists.txt
@@ -1,15 +1,7 @@
-#
-# Setup
-#
-
-cmake_minimum_required(VERSION 2.8.11)
-# Use LINK_INTERFACE_LIBRARIES instead of INTERFACE_LINK_LIBRARIES
-if(POLICY CMP0022)
- cmake_policy(SET CMP0022 OLD)
-endif()
+cmake_minimum_required(VERSION 2.8.12)
project(libjpeg-turbo C)
-set(VERSION 1.5.3)
+set(VERSION 1.5.80)
string(REPLACE "." ";" VERSION_TRIPLET ${VERSION})
list(GET VERSION_TRIPLET 0 VERSION_MAJOR)
list(GET VERSION_TRIPLET 1 VERSION_MINOR)
@@ -29,116 +21,281 @@
pad_number(VERSION_REVISION 3)
set(LIBJPEG_TURBO_VERSION_NUMBER ${VERSION_MAJOR}${VERSION_MINOR}${VERSION_REVISION})
-if(NOT WIN32)
- message(FATAL_ERROR "Platform not supported by this build system. Use autotools instead.")
-endif()
+string(TIMESTAMP DEFAULT_BUILD "%Y%m%d")
+set(BUILD ${DEFAULT_BUILD} CACHE STRING "Build string (default: ${DEFAULT_BUILD})")
-string(TIMESTAMP BUILD "%Y%m%d")
-
-# This does nothing except when using MinGW. CMAKE_BUILD_TYPE has no meaning
-# in Visual Studio, and it always defaults to Debug when using NMake.
+# NOTE: On Windows, this does nothing except when using MinGW or Cygwin.
+# CMAKE_BUILD_TYPE has no meaning in Visual Studio, and it always defaults to
+# Debug when using NMake.
if(NOT CMAKE_BUILD_TYPE)
set(CMAKE_BUILD_TYPE Release)
endif()
-
message(STATUS "CMAKE_BUILD_TYPE = ${CMAKE_BUILD_TYPE}")
-# This only works if building from the command line. There is currently no way
-# to set a variable's value based on the build type when using Visual Studio.
-if(CMAKE_BUILD_TYPE STREQUAL "Debug")
- set(BUILD "${BUILD}d")
-endif()
-
message(STATUS "VERSION = ${VERSION}, BUILD = ${BUILD}")
-option(WITH_SIMD "Include SIMD extensions" TRUE)
-option(WITH_ARITH_ENC "Include arithmetic encoding support when emulating the libjpeg v6b API/ABI" TRUE)
-option(WITH_ARITH_DEC "Include arithmetic decoding support when emulating the libjpeg v6b API/ABI" TRUE)
-option(WITH_JPEG7 "Emulate libjpeg v7 API/ABI (this makes libjpeg-turbo backward incompatible with libjpeg v6b)" FALSE)
-option(WITH_JPEG8 "Emulate libjpeg v8 API/ABI (this makes libjpeg-turbo backward incompatible with libjpeg v6b)" FALSE)
-option(WITH_MEM_SRCDST "Include in-memory source/destination manager functions when emulating the libjpeg v6b or v7 API/ABI" TRUE)
-option(WITH_TURBOJPEG "Include the TurboJPEG wrapper library and associated test programs" TRUE)
-option(WITH_JAVA "Build Java wrapper for the TurboJPEG library" FALSE)
-option(WITH_12BIT "Encode/decode JPEG images with 12-bit samples (implies WITH_SIMD=0 WITH_TURBOJPEG=0 WITH_ARITH_ENC=0 WITH_ARITH_DEC=0)" FALSE)
-option(ENABLE_STATIC "Build static libraries" TRUE)
+# Detect CPU type and whether we're building 64-bit or 32-bit code
+math(EXPR BITS "${CMAKE_SIZEOF_VOID_P} * 8")
+string(TOLOWER ${CMAKE_SYSTEM_PROCESSOR} CMAKE_SYSTEM_PROCESSOR_LC)
+if(CMAKE_SYSTEM_PROCESSOR_LC MATCHES "x86_64" OR
+ CMAKE_SYSTEM_PROCESSOR_LC MATCHES "amd64" OR
+ CMAKE_SYSTEM_PROCESSOR_LC MATCHES "i[0-9]86" OR
+ CMAKE_SYSTEM_PROCESSOR_LC MATCHES "x86" OR
+ CMAKE_SYSTEM_PROCESSOR_LC MATCHES "ia32")
+ if(BITS EQUAL 64)
+ set(CPU_TYPE x86_64)
+ else()
+ set(CPU_TYPE i386)
+ endif()
+ if(NOT CMAKE_SYSTEM_PROCESSOR STREQUAL ${CPU_TYPE})
+ set(CMAKE_SYSTEM_PROCESSOR ${CPU_TYPE})
+ endif()
+elseif(CMAKE_SYSTEM_PROCESSOR_LC STREQUAL "aarch64" OR
+ CMAKE_SYSTEM_PROCESSOR_LC MATCHES "arm*64*")
+ set(CPU_TYPE arm64)
+elseif(CMAKE_SYSTEM_PROCESSOR_LC MATCHES "arm*")
+ set(CPU_TYPE arm)
+elseif(CMAKE_SYSTEM_PROCESSOR_LC MATCHES "ppc*" OR
+ CMAKE_SYSTEM_PROCESSOR_LC MATCHES "powerpc*")
+ set(CPU_TYPE powerpc)
+else()
+ set(CPU_TYPE ${CMAKE_SYSTEM_PROCESSOR_LC})
+endif()
+message(STATUS "${BITS}-bit build (${CPU_TYPE})")
+
+
+###############################################################################
+# INSTALL DIRECTORIES
+###############################################################################
+
+if(WIN32)
+ if(MSVC)
+ set(CMAKE_INSTALL_DEFAULT_PREFIX "c:/${CMAKE_PROJECT_NAME}")
+ else()
+ set(CMAKE_INSTALL_DEFAULT_PREFIX "c:/${CMAKE_PROJECT_NAME}-gcc")
+ endif()
+ if(BITS EQUAL 64)
+ set(CMAKE_INSTALL_DEFAULT_PREFIX "${CMAKE_INSTALL_DEFAULT_PREFIX}64")
+ endif()
+else()
+ set(CMAKE_INSTALL_DEFAULT_PREFIX /opt/${CMAKE_PROJECT_NAME})
+endif()
+if(CMAKE_INSTALL_PREFIX_INITIALIZED_TO_DEFAULT)
+ set(CMAKE_INSTALL_PREFIX "${CMAKE_INSTALL_DEFAULT_PREFIX}" CACHE PATH
+ "Directory into which to install ${CMAKE_PROJECT_NAME} (default: ${CMAKE_INSTALL_DEFAULT_PREFIX})"
+ FORCE)
+endif()
+message(STATUS "CMAKE_INSTALL_PREFIX = ${CMAKE_INSTALL_PREFIX}")
+
+# When the prefix is /opt/${CMAKE_PROJECT_NAME}, we assume that an "official"
+# build is being created, and thus we install things into specific locations.
+
+if(CMAKE_INSTALL_PREFIX STREQUAL "${CMAKE_INSTALL_DEFAULT_PREFIX}")
+ set(CMAKE_INSTALL_DEFAULT_DATAROOTDIR "")
+ set(CMAKE_INSTALL_DEFAULT_DOCDIR "<CMAKE_INSTALL_DATAROOTDIR>/doc")
+ set(CMAKE_INSTALL_DEFAULT_JAVADIR "<CMAKE_INSTALL_DATAROOTDIR>/classes")
+ if(UNIX AND NOT APPLE)
+ if(BITS EQUAL 64)
+ set(CMAKE_INSTALL_DEFAULT_LIBDIR "lib64")
+ else()
+ set(CMAKE_INSTALL_DEFAULT_LIBDIR "lib32")
+ endif()
+ endif()
+endif()
+
+include(cmakescripts/GNUInstallDirs.cmake)
+
+set(CMAKE_INSTALL_RPATH ${CMAKE_INSTALL_FULL_LIBDIR})
+
+macro(report_directory var)
+ if(CMAKE_INSTALL_${var} STREQUAL CMAKE_INSTALL_FULL_${var})
+ message(STATUS "CMAKE_INSTALL_${var} = ${CMAKE_INSTALL_${var}}")
+ else()
+ message(STATUS "CMAKE_INSTALL_${var} = ${CMAKE_INSTALL_${var}} (${CMAKE_INSTALL_FULL_${var}})")
+ endif()
+ mark_as_advanced(CLEAR CMAKE_INSTALL_${var})
+endmacro()
+
+set(DIRLIST "BINDIR;DATAROOTDIR;DOCDIR;INCLUDEDIR;LIBDIR")
+if(UNIX)
+ list(APPEND DIRLIST "MANDIR")
+endif()
+foreach(dir ${DIRLIST})
+ report_directory(${dir})
+endforeach()
+
+
+###############################################################################
+# CONFIGURATION OPTIONS
+###############################################################################
+
+macro(boolean_number var)
+ if(${var})
+ set(${var} 1)
+ else()
+ set(${var} 0)
+ endif()
+endmacro()
+
option(ENABLE_SHARED "Build shared libraries" TRUE)
+boolean_number(ENABLE_SHARED)
+option(ENABLE_STATIC "Build static libraries" TRUE)
+boolean_number(ENABLE_STATIC)
+option(REQUIRE_SIMD "Generate a fatal error if SIMD extensions are not available for this platform (default is to fall back to a non-SIMD build)" FALSE)
+boolean_number(REQUIRE_SIMD)
+option(WITH_12BIT "Encode/decode JPEG images with 12-bit samples (implies WITH_ARITH_DEC=0 WITH_ARITH_ENC=0 WITH_JAVA=0 WITH_SIMD=0 WITH_TURBOJPEG=0 )" FALSE)
+boolean_number(WITH_12BIT)
+option(WITH_ARITH_DEC "Include arithmetic decoding support when emulating the libjpeg v6b API/ABI" TRUE)
+boolean_number(WITH_ARITH_DEC)
+option(WITH_ARITH_ENC "Include arithmetic encoding support when emulating the libjpeg v6b API/ABI" TRUE)
+boolean_number(WITH_ARITH_ENC)
+option(WITH_JAVA "Build Java wrapper for the TurboJPEG API library (implies ENABLE_SHARED=1)" FALSE)
+boolean_number(WITH_JAVA)
+option(WITH_JPEG7 "Emulate libjpeg v7 API/ABI (this makes ${CMAKE_PROJECT_NAME} backward-incompatible with libjpeg v6b)" FALSE)
+boolean_number(WITH_JPEG7)
+option(WITH_JPEG8 "Emulate libjpeg v8 API/ABI (this makes ${CMAKE_PROJECT_NAME} backward-incompatible with libjpeg v6b)" FALSE)
+boolean_number(WITH_JPEG8)
+option(WITH_MEM_SRCDST "Include in-memory source/destination manager functions when emulating the libjpeg v6b or v7 API/ABI" TRUE)
+boolean_number(WITH_MEM_SRCDST)
+option(WITH_SIMD "Include SIMD extensions, if available for this platform" TRUE)
+boolean_number(WITH_SIMD)
+option(WITH_TURBOJPEG "Include the TurboJPEG API library and associated test programs" TRUE)
+boolean_number(WITH_TURBOJPEG)
+
+macro(report_option var desc)
+ if(${var})
+ message(STATUS "${desc} enabled (${var} = ${${var}})")
+ else()
+ message(STATUS "${desc} disabled (${var} = ${${var}})")
+ endif()
+endmacro()
+
+if(WITH_JAVA)
+ set(ENABLE_SHARED 1)
+endif()
+
+# Explicitly setting CMAKE_POSITION_INDEPENDENT_CODE=FALSE disables PIC for all
+# targets, which will cause the shared library builds to fail. Thus, if shared
+# libraries are enabled and CMAKE_POSITION_INDEPENDENT_CODE is explicitly set
+# to FALSE, we need to unset it, thus restoring the default behavior
+# (automatically using PIC for shared library targets.)
+if(DEFINED CMAKE_POSITION_INDEPENDENT_CODE AND
+ NOT CMAKE_POSITION_INDEPENDENT_CODE AND ENABLE_SHARED)
+ unset(CMAKE_POSITION_INDEPENDENT_CODE CACHE)
+endif()
+
+report_option(ENABLE_SHARED "Shared libraries")
+report_option(ENABLE_STATIC "Static libraries")
if(WITH_12BIT)
- set(WITH_SIMD FALSE)
- set(WITH_TURBOJPEG FALSE)
- set(WITH_JAVA FALSE)
- set(WITH_ARITH_ENC FALSE)
- set(WITH_ARITH_DEC FALSE)
+ set(WITH_ARITH_DEC 0)
+ set(WITH_ARITH_ENC 0)
+ set(WITH_JAVA 0)
+ set(WITH_SIMD 0)
+ set(WITH_TURBOJPEG 0)
set(BITS_IN_JSAMPLE 12)
- message(STATUS "12-bit JPEG support enabled")
else()
set(BITS_IN_JSAMPLE 8)
endif()
+report_option(WITH_12BIT "12-bit JPEG support")
if(WITH_JPEG8 OR WITH_JPEG7)
set(WITH_ARITH_ENC 1)
set(WITH_ARITH_DEC 1)
endif()
if(WITH_JPEG8)
- set(WITH_MEM_SRCDST 1)
-endif()
-
-if(WITH_ARITH_ENC)
- set(C_ARITH_CODING_SUPPORTED 1)
- message(STATUS "Arithmetic encoding support enabled")
-else()
- message(STATUS "Arithmetic encoding support disabled")
+ set(WITH_MEM_SRCDST 0)
endif()
if(WITH_ARITH_DEC)
set(D_ARITH_CODING_SUPPORTED 1)
- message(STATUS "Arithmetic decoding support enabled")
-else()
- message(STATUS "Arithmetic decoding support disabled")
+endif()
+if(NOT WITH_12BIT)
+ report_option(WITH_ARITH_DEC "Arithmetic decoding support")
endif()
-if(WITH_TURBOJPEG)
- message(STATUS "TurboJPEG C wrapper enabled")
-else()
- message(STATUS "TurboJPEG C wrapper disabled")
+if(WITH_ARITH_ENC)
+ set(C_ARITH_CODING_SUPPORTED 1)
+endif()
+if(NOT WITH_12BIT)
+ report_option(WITH_ARITH_ENC "Arithmetic encoding support")
endif()
-if(WITH_JAVA)
- message(STATUS "TurboJPEG Java wrapper enabled")
-else()
- message(STATUS "TurboJPEG Java wrapper disabled")
+if(NOT WITH_12BIT)
+ report_option(WITH_TURBOJPEG "TurboJPEG API library")
+ report_option(WITH_JAVA "TurboJPEG Java wrapper")
endif()
-set(SO_AGE 0)
-if(WITH_MEM_SRCDST)
- set(SO_AGE 1)
-endif()
-
-set(JPEG_LIB_VERSION 62)
-set(DLL_VERSION ${JPEG_LIB_VERSION})
-set(FULLVERSION ${DLL_VERSION}.${SO_AGE}.0)
-if(WITH_JPEG8)
- set(JPEG_LIB_VERSION 80)
- set(DLL_VERSION 8)
- set(FULLVERSION ${DLL_VERSION}.0.2)
- message(STATUS "Emulating libjpeg v8 API/ABI")
-elseif(WITH_JPEG7)
- set(JPEG_LIB_VERSION 70)
- set(DLL_VERSION 7)
- set(FULLVERSION ${DLL_VERSION}.${SO_AGE}.0)
- message(STATUS "Emulating libjpeg v7 API/ABI")
-endif(WITH_JPEG8)
-
if(WITH_MEM_SRCDST)
set(MEM_SRCDST_SUPPORTED 1)
- message(STATUS "In-memory source/destination managers enabled")
-else()
- message(STATUS "In-memory source/destination managers disabled")
+ set(MEM_SRCDST_FUNCTIONS "global: jpeg_mem_dest; jpeg_mem_src;")
endif()
+if(NOT WITH_JPEG8)
+ report_option(WITH_MEM_SRCDST "In-memory source/destination managers")
+endif()
+
+set(SO_AGE 2)
+if(WITH_MEM_SRCDST)
+ set(SO_AGE 3)
+endif()
+
+if(WITH_JPEG8)
+ set(JPEG_LIB_VERSION 80)
+elseif(WITH_JPEG7)
+ set(JPEG_LIB_VERSION 70)
+else()
+ set(JPEG_LIB_VERSION 62)
+endif()
+
+math(EXPR JPEG_LIB_VERSION_DIV10 "${JPEG_LIB_VERSION} / 10")
+math(EXPR JPEG_LIB_VERSION_MOD10 "${JPEG_LIB_VERSION} % 10")
+if(JPEG_LIB_VERSION STREQUAL "62")
+ set(DEFAULT_SO_MAJOR_VERSION ${JPEG_LIB_VERSION})
+else()
+ set(DEFAULT_SO_MAJOR_VERSION ${JPEG_LIB_VERSION_DIV10})
+endif()
+if(JPEG_LIB_VERSION STREQUAL "80")
+ set(DEFAULT_SO_MINOR_VERSION 2)
+else()
+ set(DEFAULT_SO_MINOR_VERSION 0)
+endif()
+
+# This causes SO_MAJOR_VERSION/SO_MINOR_VERSION to reset to defaults if
+# WITH_JPEG7 or WITH_JPEG8 has changed.
+if((DEFINED WITH_JPEG7_INT AND NOT WITH_JPEG7 EQUAL WITH_JPEG7_INT) OR
+ (DEFINED WITH_JPEG8_INT AND NOT WITH_JPEG8 EQUAL WITH_JPEG8_INT))
+ set(FORCE_SO_VERSION "FORCE")
+endif()
+set(WITH_JPEG7_INT ${WITH_JPEG7} CACHE INTERNAL "")
+set(WITH_JPEG8_INT ${WITH_JPEG8} CACHE INTERNAL "")
+
+set(SO_MAJOR_VERSION ${DEFAULT_SO_MAJOR_VERSION} CACHE STRING
+ "Major version of the libjpeg API shared library (default: ${DEFAULT_SO_MAJOR_VERSION})"
+ ${FORCE_SO_VERSION})
+set(SO_MINOR_VERSION ${DEFAULT_SO_MINOR_VERSION} CACHE STRING
+ "Minor version of the libjpeg API shared library (default: ${DEFAULT_SO_MINOR_VERSION})"
+ ${FORCE_SO_VERSION})
+
+set(JPEG_LIB_VERSION_DECIMAL "${JPEG_LIB_VERSION_DIV10}.${JPEG_LIB_VERSION_MOD10}")
+message(STATUS "Emulating libjpeg API/ABI v${JPEG_LIB_VERSION_DECIMAL} (WITH_JPEG7 = ${WITH_JPEG7}, WITH_JPEG8 = ${WITH_JPEG8})")
+message(STATUS "libjpeg API shared library version = ${SO_MAJOR_VERSION}.${SO_AGE}.${SO_MINOR_VERSION}")
+
+# Because the TurboJPEG API library uses versioned symbols and changes the
+# names of functions whenever they are modified in a backward-incompatible
+# manner, it is always backward-ABI-compatible with itself, so the major and
+# minor SO versions don't change. However, we increase the middle number (the
+# SO "age") whenever functions are added to the API.
+set(TURBOJPEG_SO_MAJOR_VERSION 0)
+set(TURBOJPEG_SO_VERSION 0.2.0)
+
+
+###############################################################################
+# COMPILER SETTINGS
+###############################################################################
if(MSVC)
option(WITH_CRT_DLL
- "Link all libjpeg-turbo libraries and executables with the C run-time DLL (msvcr*.dll) instead of the static C run-time library (libcmt*.lib.) The default is to use the C run-time DLL only with the libraries and executables that need it."
+ "Link all ${CMAKE_PROJECT_NAME} libraries and executables with the C run-time DLL (msvcr*.dll) instead of the static C run-time library (libcmt*.lib.) The default is to use the C run-time DLL only with the libraries and executables that need it."
FALSE)
if(NOT WITH_CRT_DLL)
# Use the static C library for all build types
@@ -149,45 +306,32 @@
endif()
endforeach()
endif()
- add_definitions(-W3 -wd4996)
+ set(CMAKE_C_FLAGS "${CMAKE_C_FLAGS} /W3 /wd4996")
endif()
-# Detect whether compiler is 64-bit
-if(MSVC AND CMAKE_CL_64)
- set(SIMD_X86_64 1)
- set(64BIT 1)
-elseif(CMAKE_SIZEOF_VOID_P MATCHES 8)
- set(SIMD_X86_64 1)
- set(64BIT 1)
+if(CMAKE_COMPILER_IS_GNUCC OR CMAKE_C_COMPILER_ID STREQUAL "Clang")
+ # Use the maximum optimization level for release builds
+ foreach(var CMAKE_C_FLAGS_RELEASE CMAKE_C_FLAGS_RELWITHDEBINFO)
+ if(${var} MATCHES "-O2")
+ string(REGEX REPLACE "-O2" "-O3" ${var} "${${var}}")
+ endif()
+ endforeach()
endif()
-if(64BIT)
- message(STATUS "64-bit build")
-else()
- message(STATUS "32-bit build")
-endif()
-
-if(CMAKE_INSTALL_PREFIX_INITIALIZED_TO_DEFAULT)
- if(MSVC)
- set(CMAKE_INSTALL_PREFIX_DEFAULT ${CMAKE_PROJECT_NAME})
- else()
- set(CMAKE_INSTALL_PREFIX_DEFAULT ${CMAKE_PROJECT_NAME}-gcc)
+if(CMAKE_SYSTEM_NAME STREQUAL "SunOS")
+ if(CMAKE_C_COMPILER_ID MATCHES "SunPro")
+ # Use the maximum optimization level for release builds
+ foreach(var CMAKE_C_FLAGS_RELEASE CMAKE_C_FLAGS_RELWITHDEBINFO)
+ if(${var} MATCHES "-xO3")
+ string(REGEX REPLACE "-xO3" "-xO5" ${var} "${${var}}")
+ endif()
+ if(${var} MATCHES "-xO2")
+ string(REGEX REPLACE "-xO2" "-xO5" ${var} "${${var}}")
+ endif()
+ endforeach()
endif()
- if(64BIT)
- set(CMAKE_INSTALL_PREFIX_DEFAULT ${CMAKE_INSTALL_PREFIX_DEFAULT}64)
- endif()
- set(CMAKE_INSTALL_PREFIX "c:/${CMAKE_INSTALL_PREFIX_DEFAULT}" CACHE PATH
- "Directory into which to install libjpeg-turbo (default: c:/${CMAKE_INSTALL_PREFIX_DEFAULT})"
- FORCE)
endif()
-message(STATUS "Install directory = ${CMAKE_INSTALL_PREFIX}")
-
-configure_file(win/jconfig.h.in jconfig.h)
-configure_file(win/jconfigint.h.in jconfigint.h)
-
-include_directories(${CMAKE_CURRENT_BINARY_DIR} ${CMAKE_SOURCE_DIR})
-
string(TOUPPER ${CMAKE_BUILD_TYPE} CMAKE_BUILD_TYPE_UC)
set(EFFECTIVE_C_FLAGS "${CMAKE_C_FLAGS} ${CMAKE_C_FLAGS_${CMAKE_BUILD_TYPE_UC}}")
@@ -196,26 +340,161 @@
set(EFFECTIVE_LD_FLAGS "${CMAKE_EXE_LINKER_FLAGS} ${CMAKE_EXE_LINKER_FLAGS_${CMAKE_BUILD_TYPE_UC}}")
message(STATUS "Linker flags = ${EFFECTIVE_LD_FLAGS}")
-if(WITH_JAVA)
- find_package(Java)
- find_package(JNI)
- if(DEFINED JAVACFLAGS)
- message(STATUS "Java compiler flags = ${JAVACFLAGS}")
+include(CheckCSourceCompiles)
+
+if(UNIX)
+ # Check for headers
+ include(CheckIncludeFiles)
+ check_include_files(locale.h HAVE_LOCALE_H)
+ check_include_files(stddef.h HAVE_STDDEF_H)
+ check_include_files(stdlib.h HAVE_STDLIB_H)
+ check_include_files(sys/types.h NEED_SYS_TYPES_H)
+
+ # Check for functions
+ include(CheckFunctionExists)
+ check_function_exists(memset HAVE_MEMSET)
+ check_function_exists(memcpy HAVE_MEMCPY)
+ if(NOT HAVE_MEMSET AND NOT HAVE_MEMCPY)
+ set(NEED_BSD_STRINGS 1)
+ endif()
+
+ # Check for types
+ include(CheckTypeSize)
+ check_type_size("unsigned char" UNSIGNED_CHAR)
+ check_type_size("unsigned short" UNSIGNED_SHORT)
+ check_type_size("size_t" SIZE_T)
+
+ # Check for compiler features
+ check_c_source_compiles("int main(void) { typedef struct undefined_structure *undef_struct_ptr; }"
+ INCOMPLETE_TYPES)
+ if(INCOMPLETE_TYPES)
+ message(STATUS "Compiler supports pointers to undefined structures.")
+ else()
+ set(INCOMPLETE_TYPES_BROKEN 1)
+ message(STATUS "Compiler does not support pointers to undefined structures.")
+ endif()
+
+ if(CMAKE_CROSSCOMPILING)
+ set(RIGHT_SHIFT_IS_UNSIGNED 0)
+ else()
+ include(CheckCSourceRuns)
+ check_c_source_runs("
+ #include <stdio.h>
+ #include <stdlib.h>
+ int is_shifting_signed (long arg) {
+ long res = arg >> 4;
+ if (res == -0x7F7E80CL)
+ return 1; /* right shift is signed */
+ /* see if unsigned-shift hack will fix it. */
+ /* we can't just test exact value since it depends on width of long... */
+ res |= (~0L) << (32-4);
+ if (res == -0x7F7E80CL)
+ return 0; /* right shift is unsigned */
+ printf(\"Right shift isn't acting as I expect it to.\\\\n\");
+ printf(\"I fear the JPEG software will not work at all.\\\\n\\\\n\");
+ return 0; /* try it with unsigned anyway */
+ }
+ int main (void) {
+ exit(is_shifting_signed(-0x7F7E80B1L));
+ }" RIGHT_SHIFT_IS_UNSIGNED)
+ endif()
+
+ if(CMAKE_CROSSCOMPILING)
+ set(__CHAR_UNSIGNED__ 0)
+ else()
+ check_c_source_runs("int main(void) { return ((char) -1 < 0); }"
+ __CHAR_UNSIGNED__)
endif()
endif()
+if(MSVC)
+ set(INLINE_OPTIONS "__inline;inline")
+else()
+ set(INLINE_OPTIONS "__inline__;inline")
+endif()
+option(FORCE_INLINE "Force function inlining" TRUE)
+boolean_number(FORCE_INLINE)
+if(FORCE_INLINE)
+ if(MSVC)
+ list(INSERT INLINE_OPTIONS 0 "__forceinline")
+ else()
+ list(INSERT INLINE_OPTIONS 0 "inline __attribute__((always_inline))")
+ list(INSERT INLINE_OPTIONS 0 "__inline__ __attribute__((always_inline))")
+ endif()
+endif()
+foreach(inline ${INLINE_OPTIONS})
+ check_c_source_compiles("${inline} static void foo(void) {} int main(void) { foo(); }"
+ INLINE_WORKS)
+ if(INLINE_WORKS)
+ set(INLINE ${inline})
+ break()
+ endif()
+endforeach()
+if(NOT INLINE_WORKS)
+ message(FATAL_ERROR "Could not determine how to inline functions.")
+endif()
+message(STATUS "INLINE = ${INLINE} (FORCE_INLINE = ${FORCE_INLINE})")
-#
-# Targets
-#
+if(UNIX AND NOT APPLE)
+ file(WRITE ${CMAKE_CURRENT_BINARY_DIR}/conftest.map "VERS_1 { global: *; };")
+ set(CMAKE_REQUIRED_FLAGS
+ "-Wl,--version-script,${CMAKE_CURRENT_BINARY_DIR}/conftest.map")
+ check_c_source_compiles("int main(void) { return 0; }" HAVE_VERSION_SCRIPT)
+ set(CMAKE_REQUIRED_FLAGS)
+ file(REMOVE ${CMAKE_CURRENT_BINARY_DIR}/conftest.map)
+ if(HAVE_VERSION_SCRIPT)
+ message(STATUS "Linker supports GNU-style version scripts")
+ set(MAPFLAG "-Wl,--version-script,")
+ set(TJMAPFLAG "-Wl,--version-script,")
+ else()
+ message(STATUS "Linker does not support GNU-style version scripts")
+ if(CMAKE_SYSTEM_NAME STREQUAL "SunOS")
+ # The Solaris linker doesn't like our version script for the libjpeg API
+ # library, but the version script for the TurboJPEG API library should
+ # still work.
+ file(WRITE ${CMAKE_CURRENT_BINARY_DIR}/conftest.map
+ "VERS_1 { global: foo; local: *; }; VERS_2 { global: foo2; } VERS_1;")
+ set(CMAKE_REQUIRED_FLAGS "-Wl,-M,${CMAKE_CURRENT_BINARY_DIR}/conftest.map")
+ check_c_source_compiles("void foo() {} void foo2() {} int main(void) { return 0; }"
+ HAVE_MAPFILE)
+ set(CMAKE_REQUIRED_FLAGS)
+ file(REMOVE ${CMAKE_CURRENT_BINARY_DIR}/conftest.map)
+ if(HAVE_MAPFILE)
+ message(STATUS "Linker supports mapfiles")
+ set(TJMAPFLAG "-Wl,-M,")
+ else()
+ message(STATUS "Linker does not support mapfiles")
+ endif()
+ endif()
+ endif()
+endif()
+
+# Generate files
+if(WIN32)
+ configure_file(win/jconfig.h.in jconfig.h)
+else()
+ configure_file(jconfig.h.in jconfig.h)
+endif()
+configure_file(jconfigint.h.in jconfigint.h)
+if(UNIX)
+ configure_file(libjpeg.map.in libjpeg.map)
+endif()
+
+# Include directories and compiler definitions
+include_directories(${CMAKE_CURRENT_BINARY_DIR} ${CMAKE_CURRENT_SOURCE_DIR})
+
+
+###############################################################################
+# TARGETS
+###############################################################################
set(JPEG_SOURCES jcapimin.c jcapistd.c jccoefct.c jccolor.c jcdctmgr.c jchuff.c
- jcinit.c jcmainct.c jcmarker.c jcmaster.c jcomapi.c jcparam.c jcphuff.c
- jcprepct.c jcsample.c jctrans.c jdapimin.c jdapistd.c jdatadst.c jdatasrc.c
- jdcoefct.c jdcolor.c jddctmgr.c jdhuff.c jdinput.c jdmainct.c jdmarker.c
- jdmaster.c jdmerge.c jdphuff.c jdpostct.c jdsample.c jdtrans.c jerror.c
- jfdctflt.c jfdctfst.c jfdctint.c jidctflt.c jidctfst.c jidctint.c jidctred.c
- jquant1.c jquant2.c jutils.c jmemmgr.c jmemnobs.c)
+ jcicc.c jcinit.c jcmainct.c jcmarker.c jcmaster.c jcomapi.c jcparam.c
+ jcphuff.c jcprepct.c jcsample.c jctrans.c jdapimin.c jdapistd.c jdatadst.c
+ jdatasrc.c jdcoefct.c jdcolor.c jddctmgr.c jdhuff.c jdicc.c jdinput.c
+ jdmainct.c jdmarker.c jdmaster.c jdmerge.c jdphuff.c jdpostct.c jdsample.c
+ jdtrans.c jerror.c jfdctflt.c jfdctfst.c jfdctint.c jidctflt.c jidctfst.c
+ jidctint.c jidctred.c jquant1.c jquant2.c jutils.c jmemmgr.c jmemnobs.c)
if(WITH_ARITH_ENC OR WITH_ARITH_DEC)
set(JPEG_SOURCES ${JPEG_SOURCES} jaricom.c)
@@ -230,74 +509,81 @@
endif()
if(WITH_SIMD)
- add_definitions(-DWITH_SIMD)
add_subdirectory(simd)
- if(SIMD_X86_64)
- set(JPEG_SOURCES ${JPEG_SOURCES} simd/jsimd_x86_64.c)
- else()
- set(JPEG_SOURCES ${JPEG_SOURCES} simd/jsimd_i386.c)
+elseif(NOT WITH_12BIT)
+ message(STATUS "SIMD extensions: None (WITH_SIMD = ${WITH_SIMD})")
+endif()
+if(WITH_SIMD)
+ message(STATUS "SIMD extensions: ${CPU_TYPE} (WITH_SIMD = ${WITH_SIMD})")
+ if(MSVC_IDE)
+ set_source_files_properties(${SIMD_OBJS} PROPERTIES GENERATED 1)
endif()
- # This tells CMake that the "source" files haven't been generated yet
- set_source_files_properties(${SIMD_OBJS} PROPERTIES GENERATED 1)
else()
- set(JPEG_SOURCES ${JPEG_SOURCES} jsimd_none.c)
- message(STATUS "Not using SIMD acceleration")
+ add_library(simd OBJECT jsimd_none.c)
endif()
if(WITH_JAVA)
add_subdirectory(java)
- set(ENABLE_SHARED TRUE)
endif()
if(ENABLE_SHARED)
add_subdirectory(sharedlib)
endif()
-if(ENABLE_STATIC OR WITH_TURBOJPEG)
- add_library(jpeg-static STATIC ${JPEG_SOURCES} ${SIMD_OBJS})
+if(ENABLE_STATIC)
+ add_library(jpeg-static STATIC ${JPEG_SOURCES} $<TARGET_OBJECTS:simd>
+ ${SIMD_OBJS})
if(NOT MSVC)
set_target_properties(jpeg-static PROPERTIES OUTPUT_NAME jpeg)
endif()
- if(WITH_SIMD)
- add_dependencies(jpeg-static simd)
- endif()
endif()
if(WITH_TURBOJPEG)
- set(TURBOJPEG_SOURCES turbojpeg.c transupp.c jdatadst-tj.c jdatasrc-tj.c)
- if(WITH_JAVA)
- set(TURBOJPEG_SOURCES ${TURBOJPEG_SOURCES} turbojpeg-jni.c)
- include_directories(${JAVA_INCLUDE_PATH} ${JAVA_INCLUDE_PATH2})
- endif()
-
if(ENABLE_SHARED)
+ set(TURBOJPEG_SOURCES ${JPEG_SOURCES} $<TARGET_OBJECTS:simd> ${SIMD_OBJS}
+ turbojpeg.c transupp.c jdatadst-tj.c jdatasrc-tj.c)
+ set(TJMAPFILE ${CMAKE_CURRENT_SOURCE_DIR}/turbojpeg-mapfile)
+ if(WITH_JAVA)
+ set(TURBOJPEG_SOURCES ${TURBOJPEG_SOURCES} turbojpeg-jni.c)
+ include_directories(${JAVA_INCLUDE_PATH} ${JAVA_INCLUDE_PATH2})
+ set(TJMAPFILE ${CMAKE_CURRENT_SOURCE_DIR}/turbojpeg-mapfile.jni)
+ endif()
add_library(turbojpeg SHARED ${TURBOJPEG_SOURCES})
- set_target_properties(turbojpeg PROPERTIES DEFINE_SYMBOL DLLDEFINE)
+ if(WIN32)
+ set_target_properties(turbojpeg PROPERTIES DEFINE_SYMBOL DLLDEFINE)
+ endif()
if(MINGW)
set_target_properties(turbojpeg PROPERTIES LINK_FLAGS -Wl,--kill-at)
endif()
- target_link_libraries(turbojpeg jpeg-static)
- set_target_properties(turbojpeg PROPERTIES LINK_INTERFACE_LIBRARIES "")
+ if(APPLE)
+ set_target_properties(turbojpeg PROPERTIES MACOSX_RPATH 1)
+ endif()
+ set_target_properties(turbojpeg PROPERTIES
+ SOVERSION ${TURBOJPEG_SO_MAJOR_VERSION} VERSION ${TURBOJPEG_SO_VERSION})
+ if(TJMAPFLAG)
+ set_target_properties(turbojpeg PROPERTIES
+ LINK_FLAGS "${TJMAPFLAG}${TJMAPFILE}")
+ endif()
add_executable(tjunittest tjunittest.c tjutil.c)
target_link_libraries(tjunittest turbojpeg)
add_executable(tjbench tjbench.c bmp.c tjutil.c rdbmp.c rdppm.c wrbmp.c
wrppm.c)
- target_link_libraries(tjbench turbojpeg jpeg-static)
+ target_link_libraries(tjbench turbojpeg jpeg)
+ if(UNIX)
+ target_link_libraries(tjbench m)
+ endif()
set_property(TARGET tjbench PROPERTY COMPILE_FLAGS
"-DBMP_SUPPORTED -DPPM_SUPPORTED")
endif()
if(ENABLE_STATIC)
- add_library(turbojpeg-static STATIC ${JPEG_SOURCES} ${SIMD_OBJS}
- turbojpeg.c transupp.c jdatadst-tj.c jdatasrc-tj.c)
+ add_library(turbojpeg-static STATIC ${JPEG_SOURCES} $<TARGET_OBJECTS:simd>
+ ${SIMD_OBJS} turbojpeg.c transupp.c jdatadst-tj.c jdatasrc-tj.c)
if(NOT MSVC)
set_target_properties(turbojpeg-static PROPERTIES OUTPUT_NAME turbojpeg)
endif()
- if(WITH_SIMD)
- add_dependencies(turbojpeg-static simd)
- endif()
add_executable(tjunittest-static tjunittest.c tjutil.c)
target_link_libraries(tjunittest-static turbojpeg-static)
@@ -305,15 +591,21 @@
add_executable(tjbench-static tjbench.c bmp.c tjutil.c rdbmp.c rdppm.c
wrbmp.c wrppm.c)
target_link_libraries(tjbench-static turbojpeg-static jpeg-static)
+ if(UNIX)
+ target_link_libraries(tjbench-static m)
+ endif()
set_property(TARGET tjbench-static PROPERTY COMPILE_FLAGS
"-DBMP_SUPPORTED -DPPM_SUPPORTED")
endif()
endif()
+if(WIN32)
+ set(USE_SETMODE "-DUSE_SETMODE")
+endif()
if(WITH_12BIT)
- set(COMPILE_FLAGS "-DGIF_SUPPORTED -DPPM_SUPPORTED -DUSE_SETMODE")
+ set(COMPILE_FLAGS "-DGIF_SUPPORTED -DPPM_SUPPORTED ${USE_SETMODE}")
else()
- set(COMPILE_FLAGS "-DBMP_SUPPORTED -DGIF_SUPPORTED -DPPM_SUPPORTED -DTARGA_SUPPORTED -DUSE_SETMODE")
+ set(COMPILE_FLAGS "-DBMP_SUPPORTED -DGIF_SUPPORTED -DPPM_SUPPORTED -DTARGA_SUPPORTED ${USE_SETMODE}")
set(CJPEG_BMP_SOURCES rdbmp.c rdtarga.c)
set(DJPEG_BMP_SOURCES wrbmp.c wrtarga.c)
endif()
@@ -331,7 +623,7 @@
add_executable(jpegtran-static jpegtran.c cdjpeg.c rdswitch.c transupp.c)
target_link_libraries(jpegtran-static jpeg-static)
- set_property(TARGET jpegtran-static PROPERTY COMPILE_FLAGS "-DUSE_SETMODE")
+ set_property(TARGET jpegtran-static PROPERTY COMPILE_FLAGS "${USE_SETMODE}")
endif()
add_executable(rdjpgcom rdjpgcom.c)
@@ -339,9 +631,9 @@
add_executable(wrjpgcom wrjpgcom.c)
-#
-# Tests
-#
+###############################################################################
+# TESTS
+###############################################################################
add_subdirectory(md5)
@@ -355,7 +647,8 @@
if(WITH_12BIT)
set(TESTORIG testorig12.jpg)
- set(MD5_JPEG_RGB_ISLOW 9620f424569594bb9242b48498ad801f)
+ set(MD5_JPEG_RGB_ISLOW 9d7369207c520d37f2c1cbfcb82b2964)
+ set(MD5_JPEG_RGB_ISLOW2 a00bd20d8ae49684640ef7177d2e0b64)
set(MD5_PPM_RGB_ISLOW f3301d2219783b8b3d942b7239fa50c0)
set(MD5_JPEG_422_IFAST_OPT 7322e3bd2f127f7de4b40d4480ce60e4)
set(MD5_PPM_422_IFAST 79807fa552899e66a04708f533e16950)
@@ -367,20 +660,20 @@
set(MD5_PPM_GRAY_ISLOW 7213c10af507ad467da5578ca5ee1fca)
set(MD5_PPM_GRAY_ISLOW_RGB e96ee81c30a6ed422d466338bd3de65d)
set(MD5_JPEG_420S_IFAST_OPT 7af8e60be4d9c227ec63ac9b6630855e)
- if(64BIT)
- # Windows/x64 uses SSE for floating point
- set(MD5_JPEG_3x2_FLOAT_PROG a8c17daf77b457725ec929e215b603f8)
- set(MD5_PPM_3x2_FLOAT 42876ab9e5c2f76a87d08db5fbd57956)
- else()
- # Windows/x86 uses the 387 FPU for floating point
- if(MSVC)
- set(MD5_JPEG_3x2_FLOAT_PROG e27840755870fa849872e58aa0cd1400)
- set(MD5_PPM_3x2_FLOAT 6c2880b83bb1aa41dfe330e7a9768690)
- else()
- set(MD5_JPEG_3x2_FLOAT_PROG bc6dbbefac2872f6b9d6c4a0ae60c3c0)
- set(MD5_PPM_3x2_FLOAT f58119ee294198ac9b4a9f5645a34266)
- endif()
- endif()
+
+ set(MD5_JPEG_3x2_FLOAT_PROG_SSE a8c17daf77b457725ec929e215b603f8)
+ set(MD5_PPM_3x2_FLOAT_SSE 42876ab9e5c2f76a87d08db5fbd57956)
+ set(MD5_JPEG_3x2_FLOAT_PROG_32BIT a8c17daf77b457725ec929e215b603f8)
+ set(MD5_PPM_3x2_FLOAT_32BIT ${MD5_PPM_3x2_FLOAT_SSE})
+ set(MD5_JPEG_3x2_FLOAT_PROG_64BIT ${MD5_JPEG_3x2_FLOAT_PROG_32BIT})
+ set(MD5_PPM_3x2_FLOAT_64BIT ${MD5_PPM_3x2_FLOAT_SSE})
+ set(MD5_JPEG_3x2_FLOAT_PROG_387 bc6dbbefac2872f6b9d6c4a0ae60c3c0)
+ set(MD5_PPM_3x2_FLOAT_387 bcc5723c61560463ac60f772e742d092)
+ set(MD5_JPEG_3x2_FLOAT_PROG_MSVC e27840755870fa849872e58aa0cd1400)
+ set(MD5_PPM_3x2_FLOAT_MSVC 6c2880b83bb1aa41dfe330e7a9768690)
+
+ set(MD5_JPEG_3x2_IFAST_PROG 1396cc2b7185cfe943d408c9d305339e)
+ set(MD5_PPM_3x2_IFAST 3975985ef6eeb0a2cdc58daa651ccc00)
set(MD5_PPM_420M_ISLOW_2_1 4ca6be2a6f326ff9eaab63e70a8259c0)
set(MD5_PPM_420M_ISLOW_15_8 12aa9f9534c1b3d7ba047322226365eb)
set(MD5_PPM_420M_ISLOW_13_8 f7e22817c7b25e1393e4ec101e9d4e96)
@@ -400,7 +693,8 @@
set(MD5_JPEG_CROP cdb35ff4b4519392690ea040c56ea99c)
else()
set(TESTORIG testorig.jpg)
- set(MD5_JPEG_RGB_ISLOW 768e970dd57b340ff1b83c9d3d47c77b)
+ set(MD5_JPEG_RGB_ISLOW 1d44a406f61da743b5fd31c0a9abdca3)
+ set(MD5_JPEG_RGB_ISLOW2 31d121e57b6c2934c890a7fc7763bcd4)
set(MD5_PPM_RGB_ISLOW 00a257f5393fef8821f2b88ac7421291)
set(MD5_BMP_RGB_ISLOW_565 f07d2e75073e4bb10f6c6f4d36e2e3be)
set(MD5_BMP_RGB_ISLOW_565D 4cfa0928ef3e6bb626d7728c924cfda4)
@@ -418,23 +712,20 @@
set(MD5_BMP_GRAY_ISLOW_565 12f78118e56a2f48b966f792fedf23cc)
set(MD5_BMP_GRAY_ISLOW_565D bdbbd616441a24354c98553df5dc82db)
set(MD5_JPEG_420S_IFAST_OPT 388708217ac46273ca33086b22827ed8)
- if(WITH_SIMD)
- set(MD5_JPEG_3x2_FLOAT_PROG 343e3f8caf8af5986ebaf0bdc13b5c71)
- set(MD5_PPM_3x2_FLOAT 1a75f36e5904d6fc3a85a43da9ad89bb)
- else()
- if(64BIT)
- set(MD5_JPEG_3x2_FLOAT_PROG 9bca803d2042bd1eb03819e2bf92b3e5)
- set(MD5_PPM_3x2_FLOAT f6bfab038438ed8f5522fbd33595dcdc)
- else()
- if(MSVC)
- set(MD5_JPEG_3x2_FLOAT_PROG 7999ce9cd0ee9b6c7043b7351ab7639d)
- set(MD5_PPM_3x2_FLOAT 28cdc448a6b75e97892f0e0f8d4b21f3)
- else()
- set(MD5_JPEG_3x2_FLOAT_PROG 1657664a410e0822c924b54f6f65e6e9)
- set(MD5_PPM_3x2_FLOAT cb0a1f027f3d2917c902b5640214e025)
- endif()
- endif()
- endif()
+
+ set(MD5_JPEG_3x2_FLOAT_PROG_SSE 343e3f8caf8af5986ebaf0bdc13b5c71)
+ set(MD5_PPM_3x2_FLOAT_SSE 1a75f36e5904d6fc3a85a43da9ad89bb)
+ set(MD5_JPEG_3x2_FLOAT_PROG_32BIT 9bca803d2042bd1eb03819e2bf92b3e5)
+ set(MD5_PPM_3x2_FLOAT_32BIT f6bfab038438ed8f5522fbd33595dcdc)
+ set(MD5_JPEG_3x2_FLOAT_PROG_64BIT ${MD5_JPEG_3x2_FLOAT_PROG_32BIT})
+ set(MD5_PPM_3x2_FLOAT_64BIT 0e917a34193ef976b679a6b069b1be26)
+ set(MD5_JPEG_3x2_FLOAT_PROG_387 1657664a410e0822c924b54f6f65e6e9)
+ set(MD5_PPM_3x2_FLOAT_387 cb0a1f027f3d2917c902b5640214e025)
+ set(MD5_JPEG_3x2_FLOAT_PROG_MSVC 7999ce9cd0ee9b6c7043b7351ab7639d)
+ set(MD5_PPM_3x2_FLOAT_MSVC 28cdc448a6b75e97892f0e0f8d4b21f3)
+
+ set(MD5_JPEG_3x2_IFAST_PROG 1ee5d2c1a77f2da495f993c8c7cceca5)
+ set(MD5_PPM_3x2_IFAST fd283664b3b49127984af0a7f118fccd)
set(MD5_JPEG_420_ISLOW_ARI e986fb0a637a8d833d96e8a6d6d84ea1)
set(MD5_JPEG_444_ISLOW_PROGARI 0a8f1c8f66e113c3cf635df0a475a617)
set(MD5_PPM_420M_IFAST_ARI 72b59a99bcf1de24c5b27d151bde2437)
@@ -468,27 +759,27 @@
if(WITH_JAVA)
add_test(TJUnitTest
- ${JAVA_RUNTIME} -cp java/${OBJDIR}turbojpeg.jar
+ ${Java_JAVA_EXECUTABLE} ${JAVAARGS} -cp java/turbojpeg.jar
-Djava.library.path=${CMAKE_CURRENT_BINARY_DIR}/${OBJDIR}
TJUnitTest)
add_test(TJUnitTest-yuv
- ${JAVA_RUNTIME} -cp java/${OBJDIR}turbojpeg.jar
+ ${Java_JAVA_EXECUTABLE} ${JAVAARGS} -cp java/turbojpeg.jar
-Djava.library.path=${CMAKE_CURRENT_BINARY_DIR}/${OBJDIR}
TJUnitTest -yuv)
add_test(TJUnitTest-yuv-nopad
- ${JAVA_RUNTIME} -cp java/${OBJDIR}turbojpeg.jar
+ ${Java_JAVA_EXECUTABLE} ${JAVAARGS} -cp java/turbojpeg.jar
-Djava.library.path=${CMAKE_CURRENT_BINARY_DIR}/${OBJDIR}
TJUnitTest -yuv -noyuvpad)
add_test(TJUnitTest-bi
- ${JAVA_RUNTIME} -cp java/${OBJDIR}turbojpeg.jar
+ ${Java_JAVA_EXECUTABLE} ${JAVAARGS} -cp java/turbojpeg.jar
-Djava.library.path=${CMAKE_CURRENT_BINARY_DIR}/${OBJDIR}
TJUnitTest -bi)
add_test(TJUnitTest-bi-yuv
- ${JAVA_RUNTIME} -cp java/${OBJDIR}turbojpeg.jar
+ ${Java_JAVA_EXECUTABLE} ${JAVAARGS} -cp java/turbojpeg.jar
-Djava.library.path=${CMAKE_CURRENT_BINARY_DIR}/${OBJDIR}
TJUnitTest -bi -yuv)
add_test(TJUnitTest-bi-yuv-nopad
- ${JAVA_RUNTIME} -cp java/${OBJDIR}turbojpeg.jar
+ ${Java_JAVA_EXECUTABLE} ${JAVAARGS} -cp java/turbojpeg.jar
-Djava.library.path=${CMAKE_CURRENT_BINARY_DIR}/${OBJDIR}
TJUnitTest -bi -yuv -noyuvpad)
endif()
@@ -501,214 +792,324 @@
set(TEST_LIBTYPES ${TEST_LIBTYPES} static)
endif()
-set(TESTIMAGES ${CMAKE_SOURCE_DIR}/testimages)
+set(TESTIMAGES ${CMAKE_CURRENT_SOURCE_DIR}/testimages)
set(MD5CMP ${CMAKE_CURRENT_BINARY_DIR}/md5/md5cmp)
if(CMAKE_CROSSCOMPILING)
file(RELATIVE_PATH TESTIMAGES ${CMAKE_CURRENT_BINARY_DIR} ${TESTIMAGES})
file(RELATIVE_PATH MD5CMP ${CMAKE_CURRENT_BINARY_DIR} ${MD5CMP})
endif()
+# The output of the floating point DCT/IDCT algorithms differs depending on the
+# type of floating point math used, so the FLOATTEST CMake variable must be
+# set in order to tell the testing system which floating point results it
+# should expect:
+#
+# sse = validate against the expected results from the libjpeg-turbo SSE SIMD
+# extensions
+# 32bit = validate against the expected results from the C code when running on
+# a 32-bit FPU (or when SSE is being used for floating point math,
+# which is generally the default with x86-64 compilers)
+# 64bit = validate against the expected results from the C code when running
+# on a 64-bit FPU
+# 387 = validate against the expected results from the C code when the 387 FPU
+# is being used for floating point math (which is generally the default
+# with x86 compilers)
+# msvc = validate against the expected results from the C code when compiled
+# with a 32-bit version of Visual C++
+
+if(CPU_TYPE STREQUAL "x86_64" OR CPU_TYPE STREQUAL "i386")
+ if(WITH_SIMD)
+ set(DEFAULT_FLOATTEST sse)
+ elseif(CPU_TYPE STREQUAL "x86_64")
+ set(DEFAULT_FLOATTEST 32bit)
+ elseif(CPU_TYPE STREQUAL "i386" AND MSVC)
+ set(DEFAULT_FLOATTEST msvc)
+ endif()
+else()
+ if(BITS EQUAL 64)
+ set(DEFAULT_FLOATTEST 64bit)
+ elseif(BITS EQUAL 32)
+ set(DEFAULT_FLOATTEST 32bit)
+ endif()
+endif()
+
+# This causes FLOATTEST to reset to the default value if WITH_SIMD has
+# changed.
+if(DEFINED WITH_SIMD_INT AND NOT WITH_SIMD EQUAL WITH_SIMD_INT)
+ set(FORCE_FLOATTEST "FORCE")
+endif()
+set(WITH_SIMD_INT ${WITH_SIMD} CACHE INTERNAL "")
+set(FLOATTEST ${DEFAULT_FLOATTEST} CACHE STRING
+ "The type of floating point math used by the floating point DCT/IDCT algorithms. This tells the testing system which numerical results it should expect from those tests. [sse = libjpeg-turbo x86/x86-64 SIMD extensions, 32bit = generic 32-bit FPU or SSE, 64bit = generic 64-bit FPU, 387 = 387 FPU, msvc = 32-bit Visual Studio] (default = ${DEFAULT_FLOATTEST})"
+ ${FORCE_FLOATTEST})
+message(STATUS "FLOATTEST = ${FLOATTEST}")
+
+if(FLOATTEST)
+ string(TOUPPER ${FLOATTEST} FLOATTEST_UC)
+ string(TOLOWER ${FLOATTEST} FLOATTEST)
+ if(NOT FLOATTEST STREQUAL "sse" AND NOT FLOATTEST STREQUAL "32bit" AND
+ NOT FLOATTEST STREQUAL "64bit" AND NOT FLOATTEST STREQUAL "387" AND
+ NOT FLOATTEST STREQUAL "msvc")
+ message(FATAL_ERROR "\"${FLOATTEST}\" is not a valid value for FLOATTEST.")
+ endif()
+endif()
+
foreach(libtype ${TEST_LIBTYPES})
- if(libtype STREQUAL "shared")
- set(dir sharedlib/)
- else()
- set(dir "")
+ if(libtype STREQUAL "static")
set(suffix -static)
endif()
if(WITH_TURBOJPEG)
- add_test(tjunittest${suffix} tjunittest${suffix})
- add_test(tjunittest${suffix}-alloc tjunittest${suffix} -alloc)
- add_test(tjunittest${suffix}-yuv tjunittest${suffix} -yuv)
- add_test(tjunittest${suffix}-yuv-alloc tjunittest${suffix} -yuv -alloc)
- add_test(tjunittest${suffix}-yuv-nopad tjunittest${suffix} -yuv -noyuvpad)
+ add_test(tjunittest-${libtype} tjunittest${suffix})
+ add_test(tjunittest-${libtype}-alloc tjunittest${suffix} -alloc)
+ add_test(tjunittest-${libtype}-yuv tjunittest${suffix} -yuv)
+ add_test(tjunittest-${libtype}-yuv-alloc tjunittest${suffix} -yuv -alloc)
+ add_test(tjunittest-${libtype}-yuv-nopad tjunittest${suffix} -yuv -noyuvpad)
+
+ set(MD5_PPM_GRAY_TILE 89d3ca21213d9d864b50b4e4e7de4ca6)
+ set(MD5_PPM_420_8x8_TILE 847fceab15c5b7b911cb986cf0f71de3)
+ set(MD5_PPM_420_16x16_TILE ca45552a93687e078f7137cc4126a7b0)
+ set(MD5_PPM_420_32x32_TILE d8676f1d6b68df358353bba9844f4a00)
+ set(MD5_PPM_420_64x64_TILE 4e4c1a3d7ea4bace4f868bcbe83b7050)
+ set(MD5_PPM_420_128x128_TILE f24c3429c52265832beab9df72a0ceae)
+ set(MD5_PPM_420M_8x8_TILE bc25320e1f4c31ce2e610e43e9fd173c)
+ set(MD5_PPM_420M_TILE 75ffdf14602258c5c189522af57fa605)
+ set(MD5_PPM_422_8x8_TILE d83dacd9fc73b0a6f10c09acad64eb1e)
+ set(MD5_PPM_422_16x16_TILE 35077fb610d72dd743b1eb0cbcfe10fb)
+ set(MD5_PPM_422_32x32_TILE e6902ed8a449ecc0f0d6f2bf945f65f7)
+ set(MD5_PPM_422_64x64_TILE 2b4502a8f316cedbde1da7bce3d2231e)
+ set(MD5_PPM_422_128x128_TILE f0b5617d578f5e13c8eee215d64d4877)
+ set(MD5_PPM_422M_8x8_TILE 828941d7f41cd6283abd6beffb7fd51d)
+ set(MD5_PPM_422M_TILE e877ae1324c4a280b95376f7f018172f)
+ set(MD5_PPM_444_TILE 7964e41e67cfb8d0a587c0aa4798f9c3)
+
+ # Test compressing from/decompressing to an arbitrary subregion of a larger
+ # image buffer
+ add_test(tjbench-${libtype}-tile-cp
+ ${CMAKE_COMMAND} -E copy_if_different ${TESTIMAGES}/testorig.ppm
+ testout_tile.ppm)
+ add_test(tjbench-${libtype}-tile
+ tjbench${suffix} testout_tile.ppm 95 -rgb -quiet -tile -benchtime 0.01
+ -warmup 0)
+ set_tests_properties(tjbench-${libtype}-tile
+ PROPERTIES DEPENDS tjbench-${libtype}-tile-cp)
+
+ foreach(tile 8 16 32 64 128)
+ add_test(tjbench-${libtype}-tile-gray-${tile}x${tile}-cmp
+ ${MD5CMP} ${MD5_PPM_GRAY_TILE}
+ testout_tile_GRAY_Q95_${tile}x${tile}.ppm)
+ foreach(subsamp 420 422)
+ add_test(tjbench-${libtype}-tile-${subsamp}-${tile}x${tile}-cmp
+ ${MD5CMP} ${MD5_PPM_${subsamp}_${tile}x${tile}_TILE}
+ testout_tile_${subsamp}_Q95_${tile}x${tile}.ppm)
+ endforeach()
+ add_test(tjbench-${libtype}-tile-444-${tile}x${tile}-cmp
+ ${MD5CMP} ${MD5_PPM_444_TILE}
+ testout_tile_444_Q95_${tile}x${tile}.ppm)
+ foreach(subsamp gray 420 422 444)
+ set_tests_properties(tjbench-${libtype}-tile-${subsamp}-${tile}x${tile}-cmp
+ PROPERTIES DEPENDS tjbench-${libtype}-tile)
+ endforeach()
+ endforeach()
+
+ add_test(tjbench-${libtype}-tilem-cp
+ ${CMAKE_COMMAND} -E copy_if_different ${TESTIMAGES}/testorig.ppm
+ testout_tilem.ppm)
+ add_test(tjbench-${libtype}-tilem
+ tjbench${suffix} testout_tilem.ppm 95 -rgb -fastupsample -quiet -tile
+ -benchtime 0.01 -warmup 0)
+ set_tests_properties(tjbench-${libtype}-tilem
+ PROPERTIES DEPENDS tjbench-${libtype}-tilem-cp)
+
+ add_test(tjbench-${libtype}-tile-420m-8x8-cmp
+ ${MD5CMP} ${MD5_PPM_420M_8x8_TILE} testout_tilem_420_Q95_8x8.ppm)
+ add_test(tjbench-${libtype}-tile-422m-8x8-cmp
+ ${MD5CMP} ${MD5_PPM_422M_8x8_TILE} testout_tilem_422_Q95_8x8.ppm)
+ foreach(tile 16 32 64 128)
+ foreach(subsamp 420 422)
+ add_test(tjbench-${libtype}-tile-${subsamp}m-${tile}x${tile}-cmp
+ ${MD5CMP} ${MD5_PPM_${subsamp}M_TILE}
+ testout_tilem_${subsamp}_Q95_${tile}x${tile}.ppm)
+ endforeach()
+ endforeach()
+ foreach(tile 8 16 32 64 128)
+ foreach(subsamp 420 422)
+ set_tests_properties(tjbench-${libtype}-tile-${subsamp}m-${tile}x${tile}-cmp
+ PROPERTIES DEPENDS tjbench-${libtype}-tilem)
+ endforeach()
+ endforeach()
endif()
- # These tests are carefully chosen to provide full coverage of as many of the
- # underlying algorithms as possible (including all of the SIMD-accelerated
- # ones.)
+ # These tests are carefully crafted to provide full coverage of as many of
+ # the underlying algorithms as possible (including all of the
+ # SIMD-accelerated ones.)
+
+ macro(add_bittest PROG NAME ARGS OUTFILE INFILE MD5SUM)
+ add_test(${PROG}-${libtype}-${NAME}
+ ${PROG}${suffix} ${ARGS} -outfile ${OUTFILE} ${INFILE})
+ add_test(${PROG}-${libtype}-${NAME}-cmp
+ ${MD5CMP} ${MD5SUM} ${OUTFILE})
+ set_tests_properties(${PROG}-${libtype}-${NAME}-cmp PROPERTIES
+ DEPENDS ${PROG}-${libtype}-${NAME})
+ if(${ARGC} GREATER 6)
+ set(DEPENDS ${ARGN})
+ set_tests_properties(${PROG}-${libtype}-${NAME} PROPERTIES
+ DEPENDS ${DEPENDS})
+ endif()
+ endmacro()
# CC: null SAMP: fullsize FDCT: islow ENT: huff
- add_test(cjpeg${suffix}-rgb-islow
- ${dir}cjpeg${suffix} -rgb -dct int
- -outfile testout_rgb_islow.jpg ${TESTIMAGES}/testorig.ppm)
- add_test(cjpeg${suffix}-rgb-islow-cmp
- ${MD5CMP} ${MD5_JPEG_RGB_ISLOW} testout_rgb_islow.jpg)
+ add_bittest(cjpeg rgb-islow "-rgb;-dct;int;-icc;${TESTIMAGES}/test1.icc"
+ testout_rgb_islow.jpg ${TESTIMAGES}/testorig.ppm
+ ${MD5_JPEG_RGB_ISLOW})
# CC: null SAMP: fullsize IDCT: islow ENT: huff
- add_test(djpeg${suffix}-rgb-islow
- ${dir}djpeg${suffix} -dct int -ppm
- -outfile testout_rgb_islow.ppm testout_rgb_islow.jpg)
- add_test(djpeg${suffix}-rgb-islow-cmp
- ${MD5CMP} ${MD5_PPM_RGB_ISLOW} testout_rgb_islow.ppm)
+ add_bittest(djpeg rgb-islow "-dct;int;-ppm;-icc;testout_rgb_islow.icc"
+ testout_rgb_islow.ppm testout_rgb_islow.jpg
+ ${MD5_PPM_RGB_ISLOW} cjpeg-${libtype}-rgb-islow)
+
+ add_test(djpeg-${libtype}-rgb-islow-icc-cmp
+ ${MD5CMP} b06a39d730129122e85c1363ed1bbc9e testout_rgb_islow.icc)
+
+ add_bittest(jpegtran icc "-copy;all;-icc;${TESTIMAGES}/test2.icc"
+ testout_rgb_islow2.jpg testout_rgb_islow.jpg ${MD5_JPEG_RGB_ISLOW2})
if(NOT WITH_12BIT)
# CC: RGB->RGB565 SAMP: fullsize IDCT: islow ENT: huff
- add_test(djpeg${suffix}-rgb-islow-565
- ${dir}djpeg${suffix} -dct int -rgb565 -dither none -bmp
- -outfile testout_rgb_islow_565.bmp testout_rgb_islow.jpg)
- add_test(djpeg${suffix}-rgb-islow-565-cmp
- ${MD5CMP} ${MD5_BMP_RGB_ISLOW_565} testout_rgb_islow_565.bmp)
+ add_bittest(djpeg rgb-islow-565 "-dct;int;-rgb565;-dither;none;-bmp"
+ testout_rgb_islow_565.bmp testout_rgb_islow.jpg
+ ${MD5_BMP_RGB_ISLOW_565} cjpeg-${libtype}-rgb-islow)
# CC: RGB->RGB565 (dithered) SAMP: fullsize IDCT: islow ENT: huff
- add_test(djpeg${suffix}-rgb-islow-565D
- ${dir}djpeg${suffix} -dct int -rgb565 -bmp
- -outfile testout_rgb_islow_565D.bmp testout_rgb_islow.jpg)
- add_test(djpeg${suffix}-rgb-islow-565D-cmp
- ${MD5CMP} ${MD5_BMP_RGB_ISLOW_565D} testout_rgb_islow_565D.bmp)
+ add_bittest(djpeg rgb-islow-565D "-dct;int;-rgb565;-bmp"
+ testout_rgb_islow_565D.bmp testout_rgb_islow.jpg
+ ${MD5_BMP_RGB_ISLOW_565D} cjpeg-${libtype}-rgb-islow)
endif()
# CC: RGB->YCC SAMP: fullsize/h2v1 FDCT: ifast ENT: 2-pass huff
- add_test(cjpeg${suffix}-422-ifast-opt
- ${dir}cjpeg${suffix} -sample 2x1 -dct fast -opt
- -outfile testout_422_ifast_opt.jpg ${TESTIMAGES}/testorig.ppm)
- add_test(cjpeg${suffix}-422-ifast-opt-cmp
- ${MD5CMP} ${MD5_JPEG_422_IFAST_OPT} testout_422_ifast_opt.jpg)
+ add_bittest(cjpeg 422-ifast-opt "-sample;2x1;-dct;fast;-opt"
+ testout_422_ifast_opt.jpg ${TESTIMAGES}/testorig.ppm
+ ${MD5_JPEG_422_IFAST_OPT})
# CC: YCC->RGB SAMP: fullsize/h2v1 fancy IDCT: ifast ENT: huff
- add_test(djpeg${suffix}-422-ifast
- ${dir}djpeg${suffix} -dct fast
- -outfile testout_422_ifast.ppm testout_422_ifast_opt.jpg)
- add_test(djpeg${suffix}-422-ifast-cmp
- ${MD5CMP} ${MD5_PPM_422_IFAST} testout_422_ifast.ppm)
+ add_bittest(djpeg 422-ifast "-dct;fast"
+ testout_422_ifast.ppm testout_422_ifast_opt.jpg
+ ${MD5_PPM_422_IFAST} cjpeg-${libtype}-422-ifast-opt)
# CC: YCC->RGB SAMP: h2v1 merged IDCT: ifast ENT: huff
- add_test(djpeg${suffix}-422m-ifast
- ${dir}djpeg${suffix} -dct fast -nosmooth
- -outfile testout_422m_ifast.ppm testout_422_ifast_opt.jpg)
- add_test(djpeg${suffix}-422m-ifast-cmp
- ${MD5CMP} ${MD5_PPM_422M_IFAST} testout_422m_ifast.ppm)
+ add_bittest(djpeg 422m-ifast "-dct;fast;-nosmooth"
+ testout_422m_ifast.ppm testout_422_ifast_opt.jpg
+ ${MD5_PPM_422M_IFAST} cjpeg-${libtype}-422-ifast-opt)
if(NOT WITH_12BIT)
# CC: YCC->RGB565 SAMP: h2v1 merged IDCT: ifast ENT: huff
- add_test(djpeg${suffix}-422m-ifast-565
- ${dir}djpeg${suffix} -dct int -nosmooth -rgb565 -dither none -bmp
- -outfile testout_422m_ifast_565.bmp testout_422_ifast_opt.jpg)
- add_test(djpeg${suffix}-422m-ifast-565-cmp
- ${MD5CMP} ${MD5_BMP_422M_IFAST_565} testout_422m_ifast_565.bmp)
+ add_bittest(djpeg 422m-ifast-565
+ "-dct;int;-nosmooth;-rgb565;-dither;none;-bmp"
+ testout_422m_ifast_565.bmp testout_422_ifast_opt.jpg
+ ${MD5_BMP_422M_IFAST_565} cjpeg-${libtype}-422-ifast-opt)
# CC: YCC->RGB565 (dithered) SAMP: h2v1 merged IDCT: ifast ENT: huff
- add_test(djpeg${suffix}-422m-ifast-565D
- ${dir}djpeg${suffix} -dct int -nosmooth -rgb565 -bmp
- -outfile testout_422m_ifast_565D.bmp testout_422_ifast_opt.jpg)
- add_test(djpeg${suffix}-422m-ifast-565D-cmp
- ${MD5CMP} ${MD5_BMP_422M_IFAST_565D} testout_422m_ifast_565D.bmp)
+ add_bittest(djpeg 422m-ifast-565D "-dct;int;-nosmooth;-rgb565;-bmp"
+ testout_422m_ifast_565D.bmp testout_422_ifast_opt.jpg
+ ${MD5_BMP_422M_IFAST_565D} cjpeg-${libtype}-422-ifast-opt)
endif()
# CC: RGB->YCC SAMP: fullsize/h2v2 FDCT: ifast ENT: prog huff
- add_test(cjpeg${suffix}-420-q100-ifast-prog
- ${dir}cjpeg${suffix} -sample 2x2 -quality 100 -dct fast -prog
- -outfile testout_420_q100_ifast_prog.jpg ${TESTIMAGES}/testorig.ppm)
- add_test(cjpeg${suffix}-420-q100-ifast-prog-cmp
- ${MD5CMP} ${MD5_JPEG_420_IFAST_Q100_PROG} testout_420_q100_ifast_prog.jpg)
+ add_bittest(cjpeg 420-q100-ifast-prog
+ "-sample;2x2;-quality;100;-dct;fast;-prog"
+ testout_420_q100_ifast_prog.jpg ${TESTIMAGES}/testorig.ppm
+ ${MD5_JPEG_420_IFAST_Q100_PROG})
# CC: YCC->RGB SAMP: fullsize/h2v2 fancy IDCT: ifast ENT: prog huff
- add_test(djpeg${suffix}-420-q100-ifast-prog
- ${dir}djpeg${suffix} -dct fast
- -outfile testout_420_q100_ifast.ppm testout_420_q100_ifast_prog.jpg)
- add_test(djpeg${suffix}-420-q100-ifast-prog-cmp
- ${MD5CMP} ${MD5_PPM_420_Q100_IFAST} testout_420_q100_ifast.ppm)
+ add_bittest(djpeg 420-q100-ifast-prog "-dct;fast"
+ testout_420_q100_ifast.ppm testout_420_q100_ifast_prog.jpg
+ ${MD5_PPM_420_Q100_IFAST} cjpeg-${libtype}-420-q100-ifast-prog)
# CC: YCC->RGB SAMP: h2v2 merged IDCT: ifast ENT: prog huff
- add_test(djpeg${suffix}-420m-q100-ifast-prog
- ${dir}djpeg${suffix} -dct fast -nosmooth
- -outfile testout_420m_q100_ifast.ppm testout_420_q100_ifast_prog.jpg)
- add_test(djpeg${suffix}-420m-q100-ifast-prog-cmp
- ${MD5CMP} ${MD5_PPM_420M_Q100_IFAST} testout_420m_q100_ifast.ppm)
+ add_bittest(djpeg 420m-q100-ifast-prog "-dct;fast;-nosmooth"
+ testout_420m_q100_ifast.ppm testout_420_q100_ifast_prog.jpg
+ ${MD5_PPM_420M_Q100_IFAST} cjpeg-${libtype}-420-q100-ifast-prog)
# CC: RGB->Gray SAMP: fullsize FDCT: islow ENT: huff
- add_test(cjpeg${suffix}-gray-islow
- ${dir}cjpeg${suffix} -gray -dct int
- -outfile testout_gray_islow.jpg ${TESTIMAGES}/testorig.ppm)
- add_test(cjpeg${suffix}-gray-islow-cmp
- ${MD5CMP} ${MD5_JPEG_GRAY_ISLOW} testout_gray_islow.jpg)
+ add_bittest(cjpeg gray-islow "-gray;-dct;int"
+ testout_gray_islow.jpg ${TESTIMAGES}/testorig.ppm
+ ${MD5_JPEG_GRAY_ISLOW})
# CC: Gray->Gray SAMP: fullsize IDCT: islow ENT: huff
- add_test(djpeg${suffix}-gray-islow
- ${dir}djpeg${suffix} -dct int
- -outfile testout_gray_islow.ppm testout_gray_islow.jpg)
- add_test(djpeg${suffix}-gray-islow-cmp
- ${MD5CMP} ${MD5_PPM_GRAY_ISLOW} testout_gray_islow.ppm)
+ add_bittest(djpeg gray-islow "-dct;int"
+ testout_gray_islow.ppm testout_gray_islow.jpg
+ ${MD5_PPM_GRAY_ISLOW} cjpeg-${libtype}-gray-islow)
# CC: Gray->RGB SAMP: fullsize IDCT: islow ENT: huff
- add_test(djpeg${suffix}-gray-islow-rgb
- ${dir}djpeg${suffix} -dct int -rgb
- -outfile testout_gray_islow_rgb.ppm testout_gray_islow.jpg)
- add_test(djpeg${suffix}-gray-islow-rgb-cmp
- ${MD5CMP} ${MD5_PPM_GRAY_ISLOW_RGB} testout_gray_islow_rgb.ppm)
+ add_bittest(djpeg gray-islow-rgb "-dct;int;-rgb"
+ testout_gray_islow_rgb.ppm testout_gray_islow.jpg
+ ${MD5_PPM_GRAY_ISLOW_RGB} cjpeg-${libtype}-gray-islow)
if(NOT WITH_12BIT)
# CC: Gray->RGB565 SAMP: fullsize IDCT: islow ENT: huff
- add_test(djpeg${suffix}-gray-islow-565
- ${dir}djpeg${suffix} -dct int -rgb565 -dither none -bmp
- -outfile testout_gray_islow_565.bmp testout_gray_islow.jpg)
- add_test(djpeg${suffix}-gray-islow-565-cmp
- ${MD5CMP} ${MD5_BMP_GRAY_ISLOW_565} testout_gray_islow_565.bmp)
+ add_bittest(djpeg gray-islow-565 "-dct;int;-rgb565;-dither;none;-bmp"
+ testout_gray_islow_565.bmp testout_gray_islow.jpg
+ ${MD5_BMP_GRAY_ISLOW_565} cjpeg-${libtype}-gray-islow)
# CC: Gray->RGB565 (dithered) SAMP: fullsize IDCT: islow ENT: huff
- add_test(djpeg${suffix}-gray-islow-565D
- ${dir}djpeg${suffix} -dct int -rgb565 -bmp
- -outfile testout_gray_islow_565D.bmp testout_gray_islow.jpg)
- add_test(djpeg${suffix}-gray-islow-565D-cmp
- ${MD5CMP} ${MD5_BMP_GRAY_ISLOW_565D} testout_gray_islow_565D.bmp)
+ add_bittest(djpeg gray-islow-565D "-dct;int;-rgb565;-bmp"
+ testout_gray_islow_565D.bmp testout_gray_islow.jpg
+ ${MD5_BMP_GRAY_ISLOW_565D} cjpeg-${libtype}-gray-islow)
endif()
# CC: RGB->YCC SAMP: fullsize smooth/h2v2 smooth FDCT: islow
# ENT: 2-pass huff
- add_test(cjpeg${suffix}-420s-ifast-opt
- ${dir}cjpeg${suffix} -sample 2x2 -smooth 1 -dct int -opt
- -outfile testout_420s_ifast_opt.jpg ${TESTIMAGES}/testorig.ppm)
- add_test(cjpeg${suffix}-420s-ifast-opt-cmp
- ${MD5CMP} ${MD5_JPEG_420S_IFAST_OPT} testout_420s_ifast_opt.jpg)
+ add_bittest(cjpeg 420s-ifast-opt "-sample;2x2;-smooth;1;-dct;int;-opt"
+ testout_420s_ifast_opt.jpg ${TESTIMAGES}/testorig.ppm
+ ${MD5_JPEG_420S_IFAST_OPT})
- # CC: RGB->YCC SAMP: fullsize/int FDCT: float ENT: prog huff
- add_test(cjpeg${suffix}-3x2-float-prog
- ${dir}cjpeg${suffix} -sample 3x2 -dct float -prog
- -outfile testout_3x2_float_prog.jpg ${TESTIMAGES}/testorig.ppm)
- add_test(cjpeg${suffix}-3x2-float-prog-cmp
- ${MD5CMP} ${MD5_JPEG_3x2_FLOAT_PROG} testout_3x2_float_prog.jpg)
+ if(FLOATTEST)
+ # CC: RGB->YCC SAMP: fullsize/int FDCT: float ENT: prog huff
+ add_bittest(cjpeg 3x2-float-prog "-sample;3x2;-dct;float;-prog"
+ testout_3x2_float_prog.jpg ${TESTIMAGES}/testorig.ppm
+ ${MD5_JPEG_3x2_FLOAT_PROG_${FLOATTEST_UC}})
- # CC: YCC->RGB SAMP: fullsize/int IDCT: float ENT: prog huff
- add_test(djpeg${suffix}-3x2-float-prog
- ${dir}djpeg${suffix} -dct float
- -outfile testout_3x2_float.ppm testout_3x2_float_prog.jpg)
- add_test(djpeg${suffix}-3x2-float-prog-cmp
- ${MD5CMP} ${MD5_PPM_3x2_FLOAT} testout_3x2_float.ppm)
+ # CC: YCC->RGB SAMP: fullsize/int IDCT: float ENT: prog huff
+ add_bittest(djpeg 3x2-float-prog "-dct;float"
+ testout_3x2_float.ppm testout_3x2_float_prog.jpg
+ ${MD5_PPM_3x2_FLOAT_${FLOATTEST_UC}} cjpeg-${libtype}-3x2-float-prog)
+ endif()
+
+ # CC: RGB->YCC SAMP: fullsize/int FDCT: ifast ENT: prog huff
+ add_bittest(cjpeg 3x2-ifast-prog "-sample;3x2;-dct;fast;-prog"
+ testout_3x2_ifast_prog.jpg ${TESTIMAGES}/testorig.ppm
+ ${MD5_JPEG_3x2_IFAST_PROG})
+
+ # CC: YCC->RGB SAMP: fullsize/int IDCT: ifast ENT: prog huff
+ add_bittest(djpeg 3x2-ifast-prog "-dct;fast"
+ testout_3x2_ifast.ppm testout_3x2_ifast_prog.jpg
+ ${MD5_PPM_3x2_IFAST} cjpeg-${libtype}-3x2-ifast-prog)
if(WITH_ARITH_ENC)
# CC: YCC->RGB SAMP: fullsize/h2v2 FDCT: islow ENT: arith
- add_test(cjpeg${suffix}-420-islow-ari
- ${dir}cjpeg${suffix} -dct int -arithmetic
- -outfile testout_420_islow_ari.jpg ${TESTIMAGES}/testorig.ppm)
- add_test(cjpeg${suffix}-420-islow-ari-cmp
- ${MD5CMP} ${MD5_JPEG_420_ISLOW_ARI} testout_420_islow_ari.jpg)
+ add_bittest(cjpeg 420-islow-ari "-dct;int;-arithmetic"
+ testout_420_islow_ari.jpg ${TESTIMAGES}/testorig.ppm
+ ${MD5_JPEG_420_ISLOW_ARI})
- add_test(jpegtran${suffix}-420-islow-ari
- ${dir}jpegtran${suffix} -arithmetic
- -outfile testout_420_islow_ari.jpg ${TESTIMAGES}/testimgint.jpg)
- add_test(jpegtran${suffix}-420-islow-ari-cmp
- ${MD5CMP} ${MD5_JPEG_420_ISLOW_ARI} testout_420_islow_ari.jpg)
+ add_bittest(jpegtran 420-islow-ari "-arithmetic"
+ testout_420_islow_ari2.jpg ${TESTIMAGES}/testimgint.jpg
+ ${MD5_JPEG_420_ISLOW_ARI})
# CC: YCC->RGB SAMP: fullsize FDCT: islow ENT: prog arith
- add_test(cjpeg${suffix}-444-islow-progari
- ${dir}cjpeg${suffix} -sample 1x1 -dct int -prog -arithmetic
- -outfile testout_444_islow_progari.jpg ${TESTIMAGES}/testorig.ppm)
- add_test(cjpeg${suffix}-444-islow-progari-cmp
- ${MD5CMP} ${MD5_JPEG_444_ISLOW_PROGARI} testout_444_islow_progari.jpg)
+ add_bittest(cjpeg 444-islow-progari
+ "-sample;1x1;-dct;int;-prog;-arithmetic"
+ testout_444_islow_progari.jpg ${TESTIMAGES}/testorig.ppm
+ ${MD5_JPEG_444_ISLOW_PROGARI})
endif()
if(WITH_ARITH_DEC)
# CC: RGB->YCC SAMP: h2v2 merged IDCT: ifast ENT: arith
- add_test(djpeg${suffix}-420m-ifast-ari
- ${dir}djpeg${suffix} -fast -ppm
- -outfile testout_420m_ifast_ari.ppm ${TESTIMAGES}/testimgari.jpg)
- add_test(djpeg${suffix}-420m-ifast-ari-cmp
- ${MD5CMP} ${MD5_PPM_420M_IFAST_ARI} testout_420m_ifast_ari.ppm)
+ add_bittest(djpeg 420m-ifast-ari "-fast;-ppm"
+ testout_420m_ifast_ari.ppm ${TESTIMAGES}/testimgari.jpg
+ ${MD5_PPM_420M_IFAST_ARI})
- add_test(jpegtran${suffix}-420-islow
- ${dir}jpegtran${suffix}
- -outfile testout_420_islow.jpg ${TESTIMAGES}/testimgari.jpg)
- add_test(jpegtran${suffix}-420-islow-cmp
- ${MD5CMP} ${MD5_JPEG_420_ISLOW} testout_420_islow.jpg)
+ add_bittest(jpegtran 420-islow ""
+ testout_420_islow.jpg ${TESTIMAGES}/testimgari.jpg
+ ${MD5_JPEG_420_ISLOW})
endif()
# 2/1-- CC: YCC->RGB SAMP: h2v2 merged IDCT: 16x16 islow ENT: huff
@@ -732,231 +1133,205 @@
# ENT: huff
foreach(scale 2_1 15_8 13_8 11_8 9_8 7_8 3_4 5_8 1_2 3_8 1_4 1_8)
string(REGEX REPLACE "_" "/" scalearg ${scale})
- add_test(djpeg${suffix}-420m-islow-${scale}
- ${dir}djpeg${suffix} -dct int -scale ${scalearg} -nosmooth -ppm
- -outfile testout_420m_islow_${scale}.ppm ${TESTIMAGES}/${TESTORIG})
- add_test(djpeg${suffix}-420m-islow-${scale}-cmp
- ${MD5CMP} ${MD5_PPM_420M_ISLOW_${scale}} testout_420m_islow_${scale}.ppm)
+ add_bittest(djpeg 420m-islow-${scale}
+ "-dct;int;-scale;${scalearg};-nosmooth;-ppm"
+ testout_420m_islow_${scale}.ppm ${TESTIMAGES}/${TESTORIG}
+ ${MD5_PPM_420M_ISLOW_${scale}})
endforeach()
if(NOT WITH_12BIT)
# CC: YCC->RGB (dithered) SAMP: h2v2 fancy IDCT: islow ENT: huff
- add_test(djpeg${suffix}-420-islow-256
- ${dir}djpeg${suffix} -dct int -colors 256 -bmp
- -outfile testout_420_islow_256.bmp ${TESTIMAGES}/${TESTORIG})
- add_test(djpeg${suffix}-420-islow-256-cmp
- ${MD5CMP} ${MD5_BMP_420_ISLOW_256} testout_420_islow_256.bmp)
+ add_bittest(djpeg 420-islow-256 "-dct;int;-colors;256;-bmp"
+ testout_420_islow_256.bmp ${TESTIMAGES}/${TESTORIG}
+ ${MD5_BMP_420_ISLOW_256})
# CC: YCC->RGB565 SAMP: h2v2 fancy IDCT: islow ENT: huff
- add_test(djpeg${suffix}-420-islow-565
- ${dir}djpeg${suffix} -dct int -rgb565 -dither none -bmp
- -outfile testout_420_islow_565.bmp ${TESTIMAGES}/${TESTORIG})
- add_test(djpeg${suffix}-420-islow-565-cmp
- ${MD5CMP} ${MD5_BMP_420_ISLOW_565} testout_420_islow_565.bmp)
+ add_bittest(djpeg 420-islow-565 "-dct;int;-rgb565;-dither;none;-bmp"
+ testout_420_islow_565.bmp ${TESTIMAGES}/${TESTORIG}
+ ${MD5_BMP_420_ISLOW_565})
# CC: YCC->RGB565 (dithered) SAMP: h2v2 fancy IDCT: islow ENT: huff
- add_test(djpeg${suffix}-420-islow-565D
- ${dir}djpeg${suffix} -dct int -rgb565 -bmp
- -outfile testout_420_islow_565D.bmp ${TESTIMAGES}/${TESTORIG})
- add_test(djpeg${suffix}-420-islow-565D-cmp
- ${MD5CMP} ${MD5_BMP_420_ISLOW_565D} testout_420_islow_565D.bmp)
+ add_bittest(djpeg 420-islow-565D "-dct;int;-rgb565;-bmp"
+ testout_420_islow_565D.bmp ${TESTIMAGES}/${TESTORIG}
+ ${MD5_BMP_420_ISLOW_565D})
# CC: YCC->RGB565 SAMP: h2v2 merged IDCT: islow ENT: huff
- add_test(djpeg${suffix}-420m-islow-565
- ${dir}djpeg${suffix} -dct int -nosmooth -rgb565 -dither none -bmp
- -outfile testout_420m_islow_565.bmp ${TESTIMAGES}/${TESTORIG})
- add_test(djpeg${suffix}-420m-islow-565-cmp
- ${MD5CMP} ${MD5_BMP_420M_ISLOW_565} testout_420m_islow_565.bmp)
+ add_bittest(djpeg 420m-islow-565
+ "-dct;int;-nosmooth;-rgb565;-dither;none;-bmp"
+ testout_420m_islow_565.bmp ${TESTIMAGES}/${TESTORIG}
+ ${MD5_BMP_420M_ISLOW_565})
# CC: YCC->RGB565 (dithered) SAMP: h2v2 merged IDCT: islow ENT: huff
- add_test(djpeg${suffix}-420m-islow-565D
- ${dir}djpeg${suffix} -dct int -nosmooth -rgb565 -bmp
- -outfile testout_420m_islow_565D.bmp ${TESTIMAGES}/${TESTORIG})
- add_test(djpeg${suffix}-420m-islow-565D-cmp
- ${MD5CMP} ${MD5_BMP_420M_ISLOW_565D} testout_420m_islow_565D.bmp)
+ add_bittest(djpeg 420m-islow-565D "-dct;int;-nosmooth;-rgb565;-bmp"
+ testout_420m_islow_565D.bmp ${TESTIMAGES}/${TESTORIG}
+ ${MD5_BMP_420M_ISLOW_565D})
endif()
# Partial decode tests. These tests are designed to cover all of the
# possible code paths in jpeg_skip_scanlines().
# Context rows: Yes Intra-iMCU row: Yes iMCU row prefetch: No ENT: huff
- add_test(djpeg${suffix}-420-islow-skip15_31
- ${dir}djpeg${suffix} -dct int -skip 15,31 -ppm
- -outfile testout_420_islow_skip15,31.ppm ${TESTIMAGES}/${TESTORIG})
- add_test(djpeg${suffix}-420-islow-skip15_31-cmp
- ${MD5CMP} ${MD5_PPM_420_ISLOW_SKIP15_31} testout_420_islow_skip15,31.ppm)
+ add_bittest(djpeg 420-islow-skip15_31 "-dct;int;-skip;15,31;-ppm"
+ testout_420_islow_skip15,31.ppm ${TESTIMAGES}/${TESTORIG}
+ ${MD5_PPM_420_ISLOW_SKIP15_31})
# Context rows: Yes Intra-iMCU row: No iMCU row prefetch: Yes ENT: arith
if(WITH_ARITH_DEC)
- add_test(djpeg${suffix}-420-islow-ari-skip16_139
- ${dir}djpeg${suffix} -dct int -skip 16,139 -ppm
- -outfile testout_420_islow_ari_skip16,139.ppm
- ${TESTIMAGES}/testimgari.jpg)
- add_test(djpeg${suffix}-420-islow-ari_skip16_139-cmp
- ${MD5CMP} ${MD5_PPM_420_ISLOW_ARI_SKIP16_139}
- testout_420_islow_ari_skip16,139.ppm)
+ add_bittest(djpeg 420-islow-ari-skip16_139 "-dct;int;-skip;16,139;-ppm"
+ testout_420_islow_ari_skip16,139.ppm ${TESTIMAGES}/testimgari.jpg
+ ${MD5_PPM_420_ISLOW_ARI_SKIP16_139})
endif()
# Context rows: Yes Intra-iMCU row: No iMCU row prefetch: No ENT: prog huff
- add_test(cjpeg${suffix}-420-islow-prog
- ${dir}cjpeg${suffix} -dct int -prog
+ add_test(cjpeg-${libtype}-420-islow-prog
+ cjpeg${suffix} -dct int -prog
-outfile testout_420_islow_prog.jpg ${TESTIMAGES}/testorig.ppm)
- add_test(djpeg${suffix}-420-islow-prog-crop62x62_71_71
- ${dir}djpeg${suffix} -dct int -crop 62x62+71+71 -ppm
- -outfile testout_420_islow_prog_crop62x62,71,71.ppm
- testout_420_islow_prog.jpg)
- add_test(djpeg${suffix}-420-islow-prog-crop62x62_71_71-cmp
- ${MD5CMP} ${MD5_PPM_420_ISLOW_PROG_CROP62x62_71_71}
- testout_420_islow_prog_crop62x62,71,71.ppm)
+ add_bittest(djpeg 420-islow-prog-crop62x62_71_71
+ "-dct;int;-crop;62x62+71+71;-ppm"
+ testout_420_islow_prog_crop62x62,71,71.ppm testout_420_islow_prog.jpg
+ ${MD5_PPM_420_ISLOW_PROG_CROP62x62_71_71} cjpeg-${libtype}-420-islow-prog)
# Context rows: Yes Intra-iMCU row: No iMCU row prefetch: No ENT: arith
if(WITH_ARITH_DEC)
- add_test(djpeg${suffix}-420-islow-ari-crop53x53_4_4
- ${dir}djpeg${suffix} -dct int -crop 53x53+4+4 -ppm
- -outfile testout_420_islow_ari_crop53x53,4,4.ppm
- ${TESTIMAGES}/testimgari.jpg)
- add_test(djpeg${suffix}-420-islow-ari-crop53x53_4_4-cmp
- ${MD5CMP} ${MD5_PPM_420_ISLOW_ARI_CROP53x53_4_4}
- testout_420_islow_ari_crop53x53,4,4.ppm)
+ add_bittest(djpeg 420-islow-ari-crop53x53_4_4
+ "-dct;int;-crop;53x53+4+4;-ppm"
+ testout_420_islow_ari_crop53x53,4,4.ppm ${TESTIMAGES}/testimgari.jpg
+ ${MD5_PPM_420_ISLOW_ARI_CROP53x53_4_4})
endif()
# Context rows: No Intra-iMCU row: Yes ENT: huff
- add_test(cjpeg${suffix}-444-islow
- ${dir}cjpeg${suffix} -dct int -sample 1x1
+ add_test(cjpeg-${libtype}-444-islow
+ cjpeg${suffix} -dct int -sample 1x1
-outfile testout_444_islow.jpg ${TESTIMAGES}/testorig.ppm)
- add_test(djpeg${suffix}-444-islow-skip1_6
- ${dir}djpeg${suffix} -dct int -skip 1,6 -ppm
- -outfile testout_444_islow_skip1,6.ppm testout_444_islow.jpg)
- add_test(djpeg${suffix}-444-islow-skip1_6-cmp
- ${MD5CMP} ${MD5_PPM_444_ISLOW_SKIP1_6} testout_444_islow_skip1,6.ppm)
+ add_bittest(djpeg 444-islow-skip1_6 "-dct;int;-skip;1,6;-ppm"
+ testout_444_islow_skip1,6.ppm testout_444_islow.jpg
+ ${MD5_PPM_444_ISLOW_SKIP1_6} cjpeg-${libtype}-444-islow)
# Context rows: No Intra-iMCU row: No ENT: prog huff
- add_test(cjpeg${suffix}-444-islow-prog
- ${dir}cjpeg${suffix} -dct int -prog -sample 1x1
+ add_test(cjpeg-${libtype}-444-islow-prog
+ cjpeg${suffix} -dct int -prog -sample 1x1
-outfile testout_444_islow_prog.jpg ${TESTIMAGES}/testorig.ppm)
- add_test(djpeg${suffix}-444-islow-prog-crop98x98_13_13
- ${dir}djpeg${suffix} -dct int -crop 98x98+13+13 -ppm
- -outfile testout_444_islow_prog_crop98x98,13,13.ppm
- testout_444_islow_prog.jpg)
- add_test(djpeg${suffix}-444-islow-prog_crop98x98_13_13-cmp
- ${MD5CMP} ${MD5_PPM_444_ISLOW_PROG_CROP98x98_13_13}
- testout_444_islow_prog_crop98x98,13,13.ppm)
+ add_bittest(djpeg 444-islow-prog-crop98x98_13_13
+ "-dct;int;-crop;98x98+13+13;-ppm"
+ testout_444_islow_prog_crop98x98,13,13.ppm testout_444_islow_prog.jpg
+ ${MD5_PPM_444_ISLOW_PROG_CROP98x98_13_13} cjpeg-${libtype}-444-islow-prog)
# Context rows: No Intra-iMCU row: No ENT: arith
if(WITH_ARITH_ENC)
- add_test(cjpeg${suffix}-444-islow-ari
- ${dir}cjpeg${suffix} -dct int -arithmetic -sample 1x1
+ add_test(cjpeg-${libtype}-444-islow-ari
+ cjpeg${suffix} -dct int -arithmetic -sample 1x1
-outfile testout_444_islow_ari.jpg ${TESTIMAGES}/testorig.ppm)
if(WITH_ARITH_DEC)
- add_test(djpeg${suffix}-444-islow-ari-crop37x37_0_0
- ${dir}djpeg${suffix} -dct int -crop 37x37+0+0 -ppm
- -outfile testout_444_islow_ari_crop37x37,0,0.ppm
- testout_444_islow_ari.jpg)
- add_test(djpeg${suffix}-444-islow-ari-crop37x37_0_0-cmp
- ${MD5CMP} ${MD5_PPM_444_ISLOW_ARI_CROP37x37_0_0}
- testout_444_islow_ari_crop37x37,0,0.ppm)
+ add_bittest(djpeg 444-islow-ari-crop37x37_0_0
+ "-dct;int;-crop;37x37+0+0;-ppm"
+ testout_444_islow_ari_crop37x37,0,0.ppm testout_444_islow_ari.jpg
+ ${MD5_PPM_444_ISLOW_ARI_CROP37x37_0_0} cjpeg-${libtype}-444-islow-ari)
endif()
endif()
- add_test(jpegtran${suffix}-crop
- ${dir}jpegtran${suffix} -crop 120x90+20+50 -transpose -perfect
- -outfile testout_crop.jpg ${TESTIMAGES}/${TESTORIG})
- add_test(jpegtran${suffix}-crop-cmp
- ${MD5CMP} ${MD5_JPEG_CROP} testout_crop.jpg)
+ add_bittest(jpegtran crop "-crop;120x90+20+50;-transpose;-perfect"
+ testout_crop.jpg ${TESTIMAGES}/${TESTORIG}
+ ${MD5_JPEG_CROP})
endforeach()
add_custom_target(testclean COMMAND ${CMAKE_COMMAND} -P
- ${CMAKE_SOURCE_DIR}/cmakescripts/testclean.cmake)
+ ${CMAKE_CURRENT_SOURCE_DIR}/cmakescripts/testclean.cmake)
-
-#
-# Installer
-#
-
-if(MSVC)
- set(INST_PLATFORM "Visual C++")
- set(INST_NAME ${CMAKE_PROJECT_NAME}-${VERSION}-vc)
- set(INST_REG_NAME ${CMAKE_PROJECT_NAME})
-elseif(MINGW)
- set(INST_PLATFORM GCC)
- set(INST_NAME ${CMAKE_PROJECT_NAME}-${VERSION}-gcc)
- set(INST_REG_NAME ${CMAKE_PROJECT_NAME}-gcc)
- set(INST_DEFS -DGCC)
+if(WITH_TURBOJPEG)
+ configure_file(tjbenchtest.in tjbenchtest @ONLY)
+ if(WITH_JAVA)
+ configure_file(tjbenchtest.java.in tjbenchtest.java @ONLY)
+ configure_file(tjexampletest.in tjexampletest @ONLY)
+ add_custom_target(tjtest
+ echo ./tjbenchtest COMMAND ./tjbenchtest
+ COMMAND echo ./tjbenchtest -alloc COMMAND ./tjbenchtest -alloc
+ COMMAND echo ./tjbenchtest -yuv COMMAND ./tjbenchtest -yuv
+ COMMAND echo ./tjbenchtest -yuv -alloc COMMAND ./tjbenchtest -yuv -alloc
+ COMMAND echo ./tjbenchtest.java COMMAND ./tjbenchtest.java
+ COMMAND echo ./tjbenchtest.java -yuv COMMAND ./tjbenchtest.java -yuv
+ COMMAND echo ./tjexampletest COMMAND ./tjexampletest
+ DEPENDS ${CMAKE_CURRENT_BINARY_DIR}/tjbenchtest
+ ${CMAKE_CURRENT_BINARY_DIR}/tjbenchtest.java
+ ${CMAKE_CURRENT_BINARY_DIR}/tjexampletest)
+ else()
+ add_custom_target(tjtest
+ echo ./tjbenchtest COMMAND ./tjbenchtest
+ COMMAND echo ./tjbenchtest -alloc COMMAND ./tjbenchtest -alloc
+ COMMAND echo ./tjbenchtest -yuv COMMAND ./tjbenchtest -yuv
+ COMMAND echo ./tjbenchtest -yuv -alloc COMMAND ./tjbenchtest -yuv -alloc
+ DEPENDS ${CMAKE_CURRENT_BINARY_DIR}/tjbenchtest)
+ endif()
endif()
-if(64BIT)
- set(INST_PLATFORM "${INST_PLATFORM} 64-bit")
- set(INST_NAME ${INST_NAME}64)
- set(INST_REG_NAME ${INST_DIR}64)
- set(INST_DEFS ${INST_DEFS} -DWIN64)
+
+###############################################################################
+# INSTALLATION
+###############################################################################
+
+if(WIN32)
+ set(EXE ".exe")
endif()
-if(WITH_JAVA)
- set(INST_DEFS ${INST_DEFS} -DJAVA)
-endif()
-
-if(MSVC_IDE)
- set(INST_DEFS ${INST_DEFS} "-DBUILDDIR=${CMAKE_CFG_INTDIR}\\")
-else()
- set(INST_DEFS ${INST_DEFS} "-DBUILDDIR=")
-endif()
-
-STRING(REGEX REPLACE "/" "\\\\" INST_DIR ${CMAKE_INSTALL_PREFIX})
-
-configure_file(release/libjpeg-turbo.nsi.in libjpeg-turbo.nsi @ONLY)
-
-if(WITH_JAVA)
- set(JAVA_DEPEND java)
-endif()
-add_custom_target(installer
- makensis -nocd ${INST_DEFS} libjpeg-turbo.nsi
- DEPENDS jpeg jpeg-static turbojpeg turbojpeg-static rdjpgcom wrjpgcom
- cjpeg djpeg jpegtran tjbench ${JAVA_DEPEND}
- SOURCES libjpeg-turbo.nsi)
-
if(WITH_TURBOJPEG)
if(ENABLE_SHARED)
install(TARGETS turbojpeg tjbench
- ARCHIVE DESTINATION lib
- LIBRARY DESTINATION lib
- RUNTIME DESTINATION bin)
+ ARCHIVE DESTINATION ${CMAKE_INSTALL_LIBDIR}
+ LIBRARY DESTINATION ${CMAKE_INSTALL_LIBDIR}
+ RUNTIME DESTINATION ${CMAKE_INSTALL_BINDIR})
endif()
if(ENABLE_STATIC)
- install(TARGETS turbojpeg-static ARCHIVE DESTINATION lib)
+ install(TARGETS turbojpeg-static ARCHIVE
+ DESTINATION ${CMAKE_INSTALL_LIBDIR})
if(NOT ENABLE_SHARED)
- install(PROGRAMS ${CMAKE_CURRENT_BINARY_DIR}/tjbench-static.exe
- DESTINATION bin RENAME tjbench.exe)
+ install(PROGRAMS ${CMAKE_CURRENT_BINARY_DIR}/tjbench-static${EXE}
+ DESTINATION ${CMAKE_INSTALL_BINDIR} RENAME tjbench${EXE})
endif()
endif()
- install(FILES ${CMAKE_SOURCE_DIR}/turbojpeg.h DESTINATION include)
+ install(FILES ${CMAKE_CURRENT_SOURCE_DIR}/turbojpeg.h
+ DESTINATION ${CMAKE_INSTALL_INCLUDEDIR})
endif()
if(ENABLE_STATIC)
- install(TARGETS jpeg-static ARCHIVE DESTINATION lib)
+ install(TARGETS jpeg-static ARCHIVE DESTINATION ${CMAKE_INSTALL_LIBDIR})
if(NOT ENABLE_SHARED)
- install(PROGRAMS ${CMAKE_CURRENT_BINARY_DIR}/cjpeg-static.exe
- DESTINATION bin RENAME cjpeg.exe)
- install(PROGRAMS ${CMAKE_CURRENT_BINARY_DIR}/djpeg-static.exe
- DESTINATION bin RENAME djpeg.exe)
- install(PROGRAMS ${CMAKE_CURRENT_BINARY_DIR}/jpegtran-static.exe
- DESTINATION bin RENAME jpegtran.exe)
+ install(PROGRAMS ${CMAKE_CURRENT_BINARY_DIR}/cjpeg-static${EXE}
+ DESTINATION ${CMAKE_INSTALL_BINDIR} RENAME cjpeg${EXE})
+ install(PROGRAMS ${CMAKE_CURRENT_BINARY_DIR}/djpeg-static${EXE}
+ DESTINATION ${CMAKE_INSTALL_BINDIR} RENAME djpeg${EXE})
+ install(PROGRAMS ${CMAKE_CURRENT_BINARY_DIR}/jpegtran-static${EXE}
+ DESTINATION ${CMAKE_INSTALL_BINDIR} RENAME jpegtran${EXE})
endif()
endif()
-install(TARGETS rdjpgcom wrjpgcom RUNTIME DESTINATION bin)
+install(TARGETS rdjpgcom wrjpgcom RUNTIME DESTINATION ${CMAKE_INSTALL_BINDIR})
-install(FILES ${CMAKE_SOURCE_DIR}/README.ijg ${CMAKE_SOURCE_DIR}/README.md
- ${CMAKE_SOURCE_DIR}/example.c ${CMAKE_SOURCE_DIR}/libjpeg.txt
- ${CMAKE_SOURCE_DIR}/structure.txt ${CMAKE_SOURCE_DIR}/usage.txt
- ${CMAKE_SOURCE_DIR}/wizard.txt
- DESTINATION doc)
+install(FILES ${CMAKE_CURRENT_SOURCE_DIR}/README.ijg
+ ${CMAKE_CURRENT_SOURCE_DIR}/README.md ${CMAKE_CURRENT_SOURCE_DIR}/example.c
+ ${CMAKE_CURRENT_SOURCE_DIR}/libjpeg.txt
+ ${CMAKE_CURRENT_SOURCE_DIR}/structure.txt
+ ${CMAKE_CURRENT_SOURCE_DIR}/usage.txt ${CMAKE_CURRENT_SOURCE_DIR}/wizard.txt
+ ${CMAKE_CURRENT_SOURCE_DIR}/LICENSE.md DESTINATION ${CMAKE_INSTALL_DOCDIR})
-install(FILES ${CMAKE_BINARY_DIR}/jconfig.h ${CMAKE_SOURCE_DIR}/jerror.h
- ${CMAKE_SOURCE_DIR}/jmorecfg.h ${CMAKE_SOURCE_DIR}/jpeglib.h
- DESTINATION include)
+if(UNIX)
+ install(FILES ${CMAKE_CURRENT_SOURCE_DIR}/cjpeg.1
+ ${CMAKE_CURRENT_SOURCE_DIR}/djpeg.1 ${CMAKE_CURRENT_SOURCE_DIR}/jpegtran.1
+ ${CMAKE_CURRENT_SOURCE_DIR}/rdjpgcom.1
+ ${CMAKE_CURRENT_SOURCE_DIR}/wrjpgcom.1
+ DESTINATION ${CMAKE_INSTALL_MANDIR}/man1)
+ install(FILES ${CMAKE_CURRENT_BINARY_DIR}/pkgscripts/libjpeg.pc
+ ${CMAKE_CURRENT_BINARY_DIR}/pkgscripts/libturbojpeg.pc
+ DESTINATION ${CMAKE_INSTALL_LIBDIR}/pkgconfig)
+endif()
-configure_file("${CMAKE_SOURCE_DIR}/cmakescripts/cmake_uninstall.cmake.in"
+install(FILES ${CMAKE_CURRENT_BINARY_DIR}/jconfig.h
+ ${CMAKE_CURRENT_SOURCE_DIR}/jerror.h ${CMAKE_CURRENT_SOURCE_DIR}/jmorecfg.h
+ ${CMAKE_CURRENT_SOURCE_DIR}/jpeglib.h
+ DESTINATION ${CMAKE_INSTALL_INCLUDEDIR})
+
+include(cmakescripts/BuildPackages.cmake)
+
+configure_file("${CMAKE_CURRENT_SOURCE_DIR}/cmakescripts/cmake_uninstall.cmake.in"
"cmake_uninstall.cmake" IMMEDIATE @ONLY)
add_custom_target(uninstall COMMAND ${CMAKE_COMMAND} -P cmake_uninstall.cmake)
diff --git a/ChangeLog.md b/ChangeLog.md
index 1f9473f..f9b8bd7 100644
--- a/ChangeLog.md
+++ b/ChangeLog.md
@@ -1,13 +1,83 @@
-1.5.3
-=====
+1.6 pre-beta
+============
### Significant changes relative to 1.5.2:
-1. Fixed a NullPointerException in the TurboJPEG Java wrapper that occurred
+1. Added AVX2 SIMD implementations of the colorspace conversion, chroma
+downsampling and upsampling, and integer quantization algorithms. This speeds
+up the compression of RGB images by approximately 10-20% when using 64-bit code
+and 8-16% when using 32-bit code, and the decompression of RGB images by
+approximately 6-15% when using 64-bit code and 4-12% when using 32-bit code.
+(As tested on a 3 GHz Intel Core i7. Actual mileage may vary.)
+
+2. Overhauled the build system to use CMake on all platforms, and removed the
+autotools-based build system. This decision resulted from extensive
+discussions within the libjpeg-turbo community. libjpeg-turbo traditionally
+used CMake only for Windows builds, but there was an increasing amount of
+demand to extend CMake support to other platforms. However, because of the
+unique nature of our code base (the need to support different assemblers on
+each platform, the need for Java support, etc.), providing dual build systems
+as other OSS imaging libraries do (including libpng and libtiff) would have
+created a maintenance burden. The use of CMake greatly simplifies some aspects
+of our build system, owing to CMake's built-in support for various assemblers,
+Java, and unit testing, as well as generally fewer quirks that have to be
+worked around in order to implement our packaging system. Eliminating
+autotools puts our project slightly at odds with the traditional practices of
+the OSS community, since most "system libraries" tend to be built with
+autotools, but it is believed that the benefits of this move outweigh the
+risks. In addition to providing a unified build environment, switching to
+CMake allows for the use of various build tools and IDEs that aren't supported
+under autotools, including XCode, Ninja, and Eclipse. It also eliminates the
+need to install autotools via MacPorts/Homebrew on OS X and allows
+libjpeg-turbo to be configured without the use of a terminal/command prompt.
+Extensive testing was conducted to ensure that all features provided by the
+autotools-based build system are provided by the new build system.
+
+3. The libjpeg API in this version of libjpeg-turbo now includes two additional
+functions, `jpeg_read_icc_profile()` and `jpeg_write_icc_profile()`, that can
+be used to extract ICC profile data from a JPEG file while decompressing or to
+embed ICC profile data in a JPEG file while compressing or transforming. This
+eliminates the need for downstream projects, such as color management libraries
+and browsers, to include their own glueware for accomplishing this.
+
+4. Improved error handling in the TurboJPEG API library:
+
+ - Introduced a new function (`tjGetErrorStr2()`) in the TurboJPEG C API
+that allows compression/decompression/transform error messages to be retrieved
+in a thread-safe manner. Retrieving error messages from global functions, such
+as `tjInitCompress()` or `tjBufSize()`, is still thread-unsafe, but since those
+functions will only throw errors if passed an invalid argument or if a memory
+allocation failure occurs, thread safety is not as much of a concern.
+ - Introduced a new function (`tjGetErrorCode()`) in the TurboJPEG C API
+and a new method (`TJException.getErrorCode()`) in the TurboJPEG Java API that
+can be used to determine the severity of the last
+compression/decompression/transform error. This allows applications to
+choose whether to ignore warnings (non-fatal errors) from the underlying
+libjpeg API or to treat them as fatal.
+ - Introduced a new flag (`TJFLAG_STOPONWARNING` in the TurboJPEG C API and
+`TJ.FLAG_STOPONWARNING` in the TurboJPEG Java API) that causes the library to
+immediately halt a compression/decompression/transform operation if it
+encounters a warning from the underlying libjpeg API (the default behavior is
+to allow the operation to complete unless a fatal error is encountered.)
+
+5. Introduced a new flag in the TurboJPEG C and Java APIs (`TJFLAG_PROGRESSIVE`
+and `TJ.FLAG_PROGRESSIVE`, respectively) that causes the library to use
+progressive entropy coding in JPEG images generated by compression and
+transform operations. Additionally, a new transform option
+(`TJXOPT_PROGRESSIVE` in the C API and `TJTransform.OPT_PROGRESSIVE` in the
+Java API) has been introduced, allowing progressive entropy coding to be
+enabled for selected transforms in a multi-transform operation.
+
+6. Introduced a new transform option in the TurboJPEG API (`TJXOPT_COPYNONE` in
+the C API and `TJTransform.OPT_COPYNONE` in the Java API) that allows the
+copying of markers (including EXIF and ICC profile data) to be disabled for a
+particular transform.
+
+7. Fixed a NullPointerException in the TurboJPEG Java wrapper that occurred
when using the YUVImage constructor that creates an instance backed by separate
image planes and allocates memory for the image planes.
-2. Fixed an issue whereby the Java version of TJUnitTest would fail when
+8. Fixed an issue whereby the Java version of TJUnitTest would fail when
testing BufferedImage encoding/decoding on big endian systems.
diff --git a/LICENSE.md b/LICENSE.md
index 0572390..8dfc045 100644
--- a/LICENSE.md
+++ b/LICENSE.md
@@ -11,7 +11,8 @@
- The Modified (3-clause) BSD License, which is listed below
- This license covers the TurboJPEG API library and associated programs.
+ This license covers the TurboJPEG API library and associated programs, as
+ well as the build system.
- The zlib License, which is listed below
diff --git a/Makefile.am b/Makefile.am
deleted file mode 100644
index eb099ea..0000000
--- a/Makefile.am
+++ /dev/null
@@ -1,792 +0,0 @@
-lib_LTLIBRARIES = libjpeg.la
-libjpeg_la_LDFLAGS = -version-info ${LIBTOOL_CURRENT}:${SO_MINOR_VERSION}:${SO_AGE} -no-undefined
-include_HEADERS = jerror.h jmorecfg.h jpeglib.h
-
-if WITH_TURBOJPEG
-lib_LTLIBRARIES += libturbojpeg.la
-libturbojpeg_la_LDFLAGS = -version-info 1:0:1 -no-undefined
-include_HEADERS += turbojpeg.h
-endif
-
-nodist_include_HEADERS = jconfig.h
-
-pkgconfigdir = $(libdir)/pkgconfig
-pkgconfig_DATA = pkgscripts/libjpeg.pc
-if WITH_TURBOJPEG
-pkgconfig_DATA += pkgscripts/libturbojpeg.pc
-endif
-
-HDRS = jchuff.h jdct.h jdhuff.h jerror.h jinclude.h jmemsys.h jmorecfg.h \
- jpegint.h jpeglib.h jversion.h jsimd.h jsimddct.h jpegcomp.h \
- jpeg_nbits_table.h
-
-libjpeg_la_SOURCES = $(HDRS) jcapimin.c jcapistd.c jccoefct.c jccolor.c \
- jcdctmgr.c jchuff.c jcinit.c jcmainct.c jcmarker.c jcmaster.c \
- jcomapi.c jcparam.c jcphuff.c jcprepct.c jcsample.c jctrans.c \
- jdapimin.c jdapistd.c jdatadst.c jdatasrc.c jdcoefct.c jdcolor.c \
- jddctmgr.c jdhuff.c jdinput.c jdmainct.c jdmarker.c jdmaster.c \
- jdmerge.c jdphuff.c jdpostct.c jdsample.c jdtrans.c jerror.c \
- jfdctflt.c jfdctfst.c jfdctint.c jidctflt.c jidctfst.c jidctint.c \
- jidctred.c jquant1.c jquant2.c jutils.c jmemmgr.c jmemnobs.c
-
-if WITH_ARITH
-libjpeg_la_SOURCES += jaricom.c
-endif
-
-if WITH_ARITH_ENC
-libjpeg_la_SOURCES += jcarith.c
-endif
-
-if WITH_ARITH_DEC
-libjpeg_la_SOURCES += jdarith.c
-endif
-
-
-SUBDIRS = java
-
-
-if WITH_TURBOJPEG
-
-libturbojpeg_la_SOURCES = $(libjpeg_la_SOURCES) turbojpeg.c turbojpeg.h \
- transupp.c transupp.h jdatadst-tj.c jdatasrc-tj.c
-
-if WITH_JAVA
-
-libturbojpeg_la_SOURCES += turbojpeg-jni.c
-libturbojpeg_la_CFLAGS = ${JNI_CFLAGS}
-TJMAPFILE = turbojpeg-mapfile.jni
-
-else
-
-TJMAPFILE = turbojpeg-mapfile
-
-endif
-
-libturbojpeg_la_SOURCES += $(TJMAPFILE)
-
-if VERSION_SCRIPT
-libturbojpeg_la_LDFLAGS += $(VERSION_SCRIPT_FLAG)$(srcdir)/$(TJMAPFILE)
-endif
-
-endif
-
-
-if VERSION_SCRIPT
-libjpeg_la_LDFLAGS += $(VERSION_SCRIPT_FLAG)libjpeg.map
-endif
-
-
-if WITH_SIMD
-
-SUBDIRS += simd
-libjpeg_la_LIBADD = simd/libsimd.la
-libturbojpeg_la_LIBADD = simd/libsimd.la
-
-else
-
-libjpeg_la_SOURCES += jsimd_none.c
-
-endif
-
-
-bin_PROGRAMS = cjpeg djpeg jpegtran rdjpgcom wrjpgcom
-noinst_PROGRAMS = jcstest
-
-
-if WITH_TURBOJPEG
-
-bin_PROGRAMS += tjbench
-
-noinst_PROGRAMS += tjunittest
-
-tjbench_SOURCES = tjbench.c bmp.h bmp.c tjutil.h tjutil.c rdbmp.c rdppm.c \
- wrbmp.c wrppm.c
-
-tjbench_LDADD = libturbojpeg.la libjpeg.la -lm
-
-tjbench_CFLAGS = -DBMP_SUPPORTED -DPPM_SUPPORTED
-
-tjunittest_SOURCES = tjunittest.c tjutil.h tjutil.c
-
-tjunittest_LDADD = libturbojpeg.la
-
-endif
-
-
-cjpeg_SOURCES = cdjpeg.h cderror.h cdjpeg.c cjpeg.c rdgif.c rdppm.c rdswitch.c
-if WITH_12BIT
-else
-cjpeg_SOURCES += rdbmp.c rdtarga.c
-endif
-
-cjpeg_LDADD = libjpeg.la
-
-cjpeg_CFLAGS = -DGIF_SUPPORTED -DPPM_SUPPORTED
-if WITH_12BIT
-else
-cjpeg_CFLAGS += -DBMP_SUPPORTED -DTARGA_SUPPORTED
-endif
-
-djpeg_SOURCES = cdjpeg.h cderror.h cdjpeg.c djpeg.c rdcolmap.c rdswitch.c \
- wrgif.c wrppm.c
-if WITH_12BIT
-else
-djpeg_SOURCES += wrbmp.c wrtarga.c
-endif
-
-djpeg_LDADD = libjpeg.la
-
-djpeg_CFLAGS = -DGIF_SUPPORTED -DPPM_SUPPORTED
-if WITH_12BIT
-else
-djpeg_CFLAGS += -DBMP_SUPPORTED -DTARGA_SUPPORTED
-endif
-
-jpegtran_SOURCES = jpegtran.c rdswitch.c cdjpeg.c transupp.c transupp.h
-
-jpegtran_LDADD = libjpeg.la
-
-rdjpgcom_SOURCES = rdjpgcom.c
-
-rdjpgcom_LDADD = libjpeg.la
-
-wrjpgcom_SOURCES = wrjpgcom.c
-
-wrjpgcom_LDADD = libjpeg.la
-
-jcstest_SOURCES = jcstest.c
-
-jcstest_LDADD = libjpeg.la
-
-dist_man1_MANS = cjpeg.1 djpeg.1 jpegtran.1 rdjpgcom.1 wrjpgcom.1
-
-DOCS= coderules.txt jconfig.txt change.log rdrle.c wrrle.c BUILDING.md \
- ChangeLog.md
-
-dist_doc_DATA = README.ijg README.md libjpeg.txt structure.txt usage.txt \
- wizard.txt LICENSE.md
-
-exampledir = $(docdir)
-dist_example_DATA = example.c
-
-
-EXTRA_DIST = win release $(DOCS) testimages CMakeLists.txt \
- sharedlib/CMakeLists.txt cmakescripts libjpeg.map.in doc doxygen.config \
- doxygen-extra.css jccolext.c jdcolext.c jdcol565.c jdmrgext.c jdmrg565.c \
- jstdhuff.c jdcoefct.h jdmainct.h jdmaster.h jdsample.h wrppm.h \
- md5/CMakeLists.txt
-
-dist-hook:
- rm -rf `find $(distdir) -name .svn`
-
-
-SUBDIRS += md5
-
-if WITH_12BIT
-
-TESTORIG = testorig12.jpg
-MD5_JPEG_RGB_ISLOW = 9620f424569594bb9242b48498ad801f
-MD5_PPM_RGB_ISLOW = f3301d2219783b8b3d942b7239fa50c0
-MD5_JPEG_422_IFAST_OPT = 7322e3bd2f127f7de4b40d4480ce60e4
-MD5_PPM_422_IFAST = 79807fa552899e66a04708f533e16950
-MD5_PPM_422M_IFAST = 07737bfe8a7c1c87aaa393a0098d16b0
-MD5_JPEG_420_IFAST_Q100_PROG = a1da220b5604081863a504297ed59e55
-MD5_PPM_420_Q100_IFAST = 1b3730122709f53d007255e8dfd3305e
-MD5_PPM_420M_Q100_IFAST = 980a1a3c5bf9510022869d30b7d26566
-MD5_JPEG_GRAY_ISLOW = 235c90707b16e2e069f37c888b2636d9
-MD5_PPM_GRAY_ISLOW = 7213c10af507ad467da5578ca5ee1fca
-MD5_PPM_GRAY_ISLOW_RGB = e96ee81c30a6ed422d466338bd3de65d
-MD5_JPEG_420S_IFAST_OPT = 7af8e60be4d9c227ec63ac9b6630855e
-MD5_JPEG_3x2_FLOAT_PROG_SSE = a8c17daf77b457725ec929e215b603f8
-MD5_PPM_3x2_FLOAT_SSE = 42876ab9e5c2f76a87d08db5fbd57956
-MD5_JPEG_3x2_FLOAT_PROG_32BIT = a8c17daf77b457725ec929e215b603f8
-MD5_PPM_3x2_FLOAT_32BIT = 42876ab9e5c2f76a87d08db5fbd57956
-MD5_PPM_3x2_FLOAT_64BIT = d6fbc71153b3d8ded484dbc17c7b9cf4
-MD5_JPEG_3x2_FLOAT_PROG_387 = bc6dbbefac2872f6b9d6c4a0ae60c3c0
-MD5_PPM_3x2_FLOAT_387 = bcc5723c61560463ac60f772e742d092
-MD5_JPEG_3x2_IFAST_PROG = 1396cc2b7185cfe943d408c9d305339e
-MD5_PPM_3x2_IFAST = 3975985ef6eeb0a2cdc58daa651ccc00
-MD5_PPM_420M_ISLOW_2_1 = 4ca6be2a6f326ff9eaab63e70a8259c0
-MD5_PPM_420M_ISLOW_15_8 = 12aa9f9534c1b3d7ba047322226365eb
-MD5_PPM_420M_ISLOW_13_8 = f7e22817c7b25e1393e4ec101e9d4e96
-MD5_PPM_420M_ISLOW_11_8 = 800a16f9f4dc9b293197bfe11be10a82
-MD5_PPM_420M_ISLOW_9_8 = 06b7a92a9bc69f4dc36ec40f1937d55c
-MD5_PPM_420M_ISLOW_7_8 = 3ec444a14a4ab4eab88ffc49c48eca43
-MD5_PPM_420M_ISLOW_3_4 = 3e726b7ea872445b19437d1c1d4f0d93
-MD5_PPM_420M_ISLOW_5_8 = a8a771abdc94301d20ffac119b2caccd
-MD5_PPM_420M_ISLOW_1_2 = b419124dd5568b085787234866102866
-MD5_PPM_420M_ISLOW_3_8 = 343d19015531b7bbe746124127244fa8
-MD5_PPM_420M_ISLOW_1_4 = 35fd59d866e44659edfa3c18db2a3edb
-MD5_PPM_420M_ISLOW_1_8 = ccaed48ac0aedefda5d4abe4013f4ad7
-MD5_PPM_420_ISLOW_SKIP15_31 = 86664cd9dc956536409e44e244d20a97
-MD5_PPM_420_ISLOW_PROG_CROP62x62_71_71 = 452a21656115a163029cfba5c04fa76a
-MD5_PPM_444_ISLOW_SKIP1_6 = ef63901f71ef7a75cd78253fc0914f84
-MD5_PPM_444_ISLOW_PROG_CROP98x98_13_13 = 15b173fb5872d9575572fbcc1b05956f
-MD5_JPEG_CROP = cdb35ff4b4519392690ea040c56ea99c
-
-else
-
-TESTORIG = testorig.jpg
-MD5_JPEG_RGB_ISLOW = 768e970dd57b340ff1b83c9d3d47c77b
-MD5_PPM_RGB_ISLOW = 00a257f5393fef8821f2b88ac7421291
-MD5_BMP_RGB_ISLOW_565 = f07d2e75073e4bb10f6c6f4d36e2e3be
-MD5_BMP_RGB_ISLOW_565D = 4cfa0928ef3e6bb626d7728c924cfda4
-MD5_JPEG_422_IFAST_OPT = 2540287b79d913f91665e660303ab2c8
-MD5_PPM_422_IFAST = 35bd6b3f833bad23de82acea847129fa
-MD5_PPM_422M_IFAST = 8dbc65323d62cca7c91ba02dd1cfa81d
-MD5_BMP_422M_IFAST_565 = 3294bd4d9a1f2b3d08ea6020d0db7065
-MD5_BMP_422M_IFAST_565D = da98c9c7b6039511be4a79a878a9abc1
-MD5_JPEG_420_IFAST_Q100_PROG = 990cbe0329c882420a2094da7e5adade
-MD5_PPM_420_Q100_IFAST = 5a732542015c278ff43635e473a8a294
-MD5_PPM_420M_Q100_IFAST = ff692ee9323a3b424894862557c092f1
-MD5_JPEG_GRAY_ISLOW = 72b51f894b8f4a10b3ee3066770aa38d
-MD5_PPM_GRAY_ISLOW = 8d3596c56eace32f205deccc229aa5ed
-MD5_PPM_GRAY_ISLOW_RGB = 116424ac07b79e5e801f00508eab48ec
-MD5_BMP_GRAY_ISLOW_565 = 12f78118e56a2f48b966f792fedf23cc
-MD5_BMP_GRAY_ISLOW_565D = bdbbd616441a24354c98553df5dc82db
-MD5_JPEG_420S_IFAST_OPT = 388708217ac46273ca33086b22827ed8
-# See README.md for more details on why this next bit is necessary.
-MD5_JPEG_3x2_FLOAT_PROG_SSE = 343e3f8caf8af5986ebaf0bdc13b5c71
-MD5_PPM_3x2_FLOAT_SSE = 1a75f36e5904d6fc3a85a43da9ad89bb
-MD5_JPEG_3x2_FLOAT_PROG_32BIT = 9bca803d2042bd1eb03819e2bf92b3e5
-MD5_PPM_3x2_FLOAT_32BIT = f6bfab038438ed8f5522fbd33595dcdc
-MD5_PPM_3x2_FLOAT_64BIT = 0e917a34193ef976b679a6b069b1be26
-MD5_JPEG_3x2_FLOAT_PROG_387 = 1657664a410e0822c924b54f6f65e6e9
-MD5_PPM_3x2_FLOAT_387 = cb0a1f027f3d2917c902b5640214e025
-MD5_JPEG_3x2_IFAST_PROG = 1ee5d2c1a77f2da495f993c8c7cceca5
-MD5_PPM_3x2_IFAST = fd283664b3b49127984af0a7f118fccd
-MD5_JPEG_420_ISLOW_ARI = e986fb0a637a8d833d96e8a6d6d84ea1
-MD5_JPEG_444_ISLOW_PROGARI = 0a8f1c8f66e113c3cf635df0a475a617
-MD5_PPM_420M_IFAST_ARI = 72b59a99bcf1de24c5b27d151bde2437
-MD5_JPEG_420_ISLOW = 9a68f56bc76e466aa7e52f415d0f4a5f
-MD5_PPM_420M_ISLOW_2_1 = 9f9de8c0612f8d06869b960b05abf9c9
-MD5_PPM_420M_ISLOW_15_8 = b6875bc070720b899566cc06459b63b7
-MD5_PPM_420M_ISLOW_13_8 = bc3452573c8152f6ae552939ee19f82f
-MD5_PPM_420M_ISLOW_11_8 = d8cc73c0aaacd4556569b59437ba00a5
-MD5_PPM_420M_ISLOW_9_8 = d25e61bc7eac0002f5b393aa223747b6
-MD5_PPM_420M_ISLOW_7_8 = ddb564b7c74a09494016d6cd7502a946
-MD5_PPM_420M_ISLOW_3_4 = 8ed8e68808c3fbc4ea764fc9d2968646
-MD5_PPM_420M_ISLOW_5_8 = a3363274999da2366a024efae6d16c9b
-MD5_PPM_420M_ISLOW_1_2 = e692a315cea26b988c8e8b29a5dbcd81
-MD5_PPM_420M_ISLOW_3_8 = 79eca9175652ced755155c90e785a996
-MD5_PPM_420M_ISLOW_1_4 = 79cd778f8bf1a117690052cacdd54eca
-MD5_PPM_420M_ISLOW_1_8 = 391b3d4aca640c8567d6f8745eb2142f
-MD5_BMP_420_ISLOW_256 = 4980185e3776e89bd931736e1cddeee6
-MD5_BMP_420_ISLOW_565 = bf9d13e16c4923b92e1faa604d7922cb
-MD5_BMP_420_ISLOW_565D = 6bde71526acc44bcff76f696df8638d2
-MD5_BMP_420M_ISLOW_565 = 8dc0185245353cfa32ad97027342216f
-MD5_BMP_420M_ISLOW_565D =d1be3a3339166255e76fa50a0d70d73e
-MD5_PPM_420_ISLOW_SKIP15_31 = c4c65c1e43d7275cd50328a61e6534f0
-MD5_PPM_420_ISLOW_ARI_SKIP16_139 = 087c6b123db16ac00cb88c5b590bb74a
-MD5_PPM_420_ISLOW_PROG_CROP62x62_71_71 = 26eb36ccc7d1f0cb80cdabb0ac8b5d99
-MD5_PPM_420_ISLOW_ARI_CROP53x53_4_4 = 886c6775af22370257122f8b16207e6d
-MD5_PPM_444_ISLOW_SKIP1_6 = 5606f86874cf26b8fcee1117a0a436a6
-MD5_PPM_444_ISLOW_PROG_CROP98x98_13_13 = db87dc7ce26bcdc7a6b56239ce2b9d6c
-MD5_PPM_444_ISLOW_ARI_CROP37x37_0_0 = cb57b32bd6d03e35432362f7bf184b6d
-MD5_JPEG_CROP = b4197f377e621c4e9b1d20471432610d
-
-endif
-
-.PHONY: test
-test: tjquicktest tjbittest bittest
-
-if CROSS_COMPILING
-tjquicktest: testclean
-else
-tjquicktest: testclean all
-endif
-
-if WITH_TURBOJPEG
-if WITH_JAVA
- $(JAVA) -cp java/turbojpeg.jar -Djava.library.path=.libs TJUnitTest
- $(JAVA) -cp java/turbojpeg.jar -Djava.library.path=.libs TJUnitTest -bi
- $(JAVA) -cp java/turbojpeg.jar -Djava.library.path=.libs TJUnitTest -yuv
- $(JAVA) -cp java/turbojpeg.jar -Djava.library.path=.libs TJUnitTest -yuv -noyuvpad
- $(JAVA) -cp java/turbojpeg.jar -Djava.library.path=.libs TJUnitTest -yuv -bi
- $(JAVA) -cp java/turbojpeg.jar -Djava.library.path=.libs TJUnitTest -yuv -bi -noyuvpad
-endif
- ./tjunittest
- ./tjunittest -alloc
- ./tjunittest -yuv
- ./tjunittest -yuv -alloc
- ./tjunittest -yuv -noyuvpad
-endif
- echo GREAT SUCCESS!
-
-if CROSS_COMPILING
-tjbittest: testclean
-else
-tjbittest: testclean all
-endif
-
-if WITH_TURBOJPEG
-
-MD5_PPM_GRAY_TILE = 89d3ca21213d9d864b50b4e4e7de4ca6
-MD5_PPM_420_8x8_TILE = 847fceab15c5b7b911cb986cf0f71de3
-MD5_PPM_420_16x16_TILE = ca45552a93687e078f7137cc4126a7b0
-MD5_PPM_420_32x32_TILE = d8676f1d6b68df358353bba9844f4a00
-MD5_PPM_420_64x64_TILE = 4e4c1a3d7ea4bace4f868bcbe83b7050
-MD5_PPM_420_128x128_TILE = f24c3429c52265832beab9df72a0ceae
-MD5_PPM_420M_8x8_TILE = bc25320e1f4c31ce2e610e43e9fd173c
-MD5_PPM_420M_TILE = 75ffdf14602258c5c189522af57fa605
-MD5_PPM_422_8x8_TILE = d83dacd9fc73b0a6f10c09acad64eb1e
-MD5_PPM_422_16x16_TILE = 35077fb610d72dd743b1eb0cbcfe10fb
-MD5_PPM_422_32x32_TILE = e6902ed8a449ecc0f0d6f2bf945f65f7
-MD5_PPM_422_64x64_TILE = 2b4502a8f316cedbde1da7bce3d2231e
-MD5_PPM_422_128x128_TILE = f0b5617d578f5e13c8eee215d64d4877
-MD5_PPM_422M_8x8_TILE = 828941d7f41cd6283abd6beffb7fd51d
-MD5_PPM_422M_TILE = e877ae1324c4a280b95376f7f018172f
-MD5_PPM_444_TILE = 7964e41e67cfb8d0a587c0aa4798f9c3
-
-# Test compressing from/decompressing to an arbitrary subregion of a larger
-# image buffer
- cp $(srcdir)/testimages/testorig.ppm testout_tile.ppm
- ./tjbench testout_tile.ppm 95 -rgb -quiet -tile -benchtime 0.01 -warmup 0 >/dev/null 2>&1
- for i in 8 16 32 64 128; do \
- md5/md5cmp $(MD5_PPM_GRAY_TILE) testout_tile_GRAY_Q95_$$i\x$$i.ppm; \
- done
- md5/md5cmp $(MD5_PPM_420_8x8_TILE) testout_tile_420_Q95_8x8.ppm
- md5/md5cmp $(MD5_PPM_420_16x16_TILE) testout_tile_420_Q95_16x16.ppm
- md5/md5cmp $(MD5_PPM_420_32x32_TILE) testout_tile_420_Q95_32x32.ppm
- md5/md5cmp $(MD5_PPM_420_64x64_TILE) testout_tile_420_Q95_64x64.ppm
- md5/md5cmp $(MD5_PPM_420_128x128_TILE) testout_tile_420_Q95_128x128.ppm
- md5/md5cmp $(MD5_PPM_422_8x8_TILE) testout_tile_422_Q95_8x8.ppm
- md5/md5cmp $(MD5_PPM_422_16x16_TILE) testout_tile_422_Q95_16x16.ppm
- md5/md5cmp $(MD5_PPM_422_32x32_TILE) testout_tile_422_Q95_32x32.ppm
- md5/md5cmp $(MD5_PPM_422_64x64_TILE) testout_tile_422_Q95_64x64.ppm
- md5/md5cmp $(MD5_PPM_422_128x128_TILE) testout_tile_422_Q95_128x128.ppm
- for i in 8 16 32 64 128; do \
- md5/md5cmp $(MD5_PPM_444_TILE) testout_tile_444_Q95_$$i\x$$i.ppm; \
- done
- rm -f testout_tile_GRAY_* testout_tile_420_* testout_tile_422_* testout_tile_444_*
-
- ./tjbench testout_tile.ppm 95 -rgb -fastupsample -quiet -tile -benchtime 0.01 -warmup 0 >/dev/null 2>&1
- md5/md5cmp $(MD5_PPM_420M_8x8_TILE) testout_tile_420_Q95_8x8.ppm
- for i in 16 32 64 128; do \
- md5/md5cmp $(MD5_PPM_420M_TILE) testout_tile_420_Q95_$$i\x$$i.ppm; \
- done
- md5/md5cmp $(MD5_PPM_422M_8x8_TILE) testout_tile_422_Q95_8x8.ppm
- for i in 16 32 64 128; do \
- md5/md5cmp $(MD5_PPM_422M_TILE) testout_tile_422_Q95_$$i\x$$i.ppm; \
- done
- rm -f testout_tile_GRAY_* testout_tile_420_* testout_tile_422_* testout_tile_444_* testout_tile.ppm
- echo GREAT SUCCESS!
-
-endif
-
-if CROSS_COMPILING
-bittest: testclean
-else
-bittest: testclean all
-endif
-
-# These tests are carefully crafted to provide full coverage of as many of the
-# underlying algorithms as possible (including all of the SIMD-accelerated
-# ones.)
-
-# CC: null SAMP: fullsize FDCT: islow ENT: huff
- ./cjpeg -rgb -dct int -outfile testout_rgb_islow.jpg $(srcdir)/testimages/testorig.ppm
- md5/md5cmp $(MD5_JPEG_RGB_ISLOW) testout_rgb_islow.jpg
-# CC: null SAMP: fullsize IDCT: islow ENT: huff
- ./djpeg -dct int -ppm -outfile testout_rgb_islow.ppm testout_rgb_islow.jpg
- md5/md5cmp $(MD5_PPM_RGB_ISLOW) testout_rgb_islow.ppm
- rm -f testout_rgb_islow.ppm
-if WITH_12BIT
- rm -f testout_rgb_islow.jpg
-else
-# CC: RGB->RGB565 SAMP: fullsize IDCT: islow ENT: huff
- ./djpeg -dct int -rgb565 -dither none -bmp -outfile testout_rgb_islow_565.bmp testout_rgb_islow.jpg
- md5/md5cmp $(MD5_BMP_RGB_ISLOW_565) testout_rgb_islow_565.bmp
- rm -f testout_rgb_islow_565.bmp
-# CC: RGB->RGB565 (dithered) SAMP: fullsize IDCT: islow ENT: huff
- ./djpeg -dct int -rgb565 -bmp -outfile testout_rgb_islow_565D.bmp testout_rgb_islow.jpg
- md5/md5cmp $(MD5_BMP_RGB_ISLOW_565D) testout_rgb_islow_565D.bmp
- rm -f testout_rgb_islow_565D.bmp testout_rgb_islow.jpg
-endif
-
-# CC: RGB->YCC SAMP: fullsize/h2v1 FDCT: ifast ENT: 2-pass huff
- ./cjpeg -sample 2x1 -dct fast -opt -outfile testout_422_ifast_opt.jpg $(srcdir)/testimages/testorig.ppm
- md5/md5cmp $(MD5_JPEG_422_IFAST_OPT) testout_422_ifast_opt.jpg
-# CC: YCC->RGB SAMP: fullsize/h2v1 fancy IDCT: ifast ENT: huff
- ./djpeg -dct fast -outfile testout_422_ifast.ppm testout_422_ifast_opt.jpg
- md5/md5cmp $(MD5_PPM_422_IFAST) testout_422_ifast.ppm
- rm -f testout_422_ifast.ppm
-# CC: YCC->RGB SAMP: h2v1 merged IDCT: ifast ENT: huff
- ./djpeg -dct fast -nosmooth -outfile testout_422m_ifast.ppm testout_422_ifast_opt.jpg
- md5/md5cmp $(MD5_PPM_422M_IFAST) testout_422m_ifast.ppm
- rm -f testout_422m_ifast.ppm
-if WITH_12BIT
- rm -f testout_422_ifast_opt.jpg
-else
-# CC: YCC->RGB565 SAMP: h2v1 merged IDCT: ifast ENT: huff
- ./djpeg -dct int -nosmooth -rgb565 -dither none -bmp -outfile testout_422m_ifast_565.bmp testout_422_ifast_opt.jpg
- md5/md5cmp $(MD5_BMP_422M_IFAST_565) testout_422m_ifast_565.bmp
- rm -f testout_422m_ifast_565.bmp
-# CC: YCC->RGB565 (dithered) SAMP: h2v1 merged IDCT: ifast ENT: huff
- ./djpeg -dct int -nosmooth -rgb565 -bmp -outfile testout_422m_ifast_565D.bmp testout_422_ifast_opt.jpg
- md5/md5cmp $(MD5_BMP_422M_IFAST_565D) testout_422m_ifast_565D.bmp
- rm -f testout_422m_ifast_565D.bmp testout_422_ifast_opt.jpg
-endif
-
-# CC: RGB->YCC SAMP: fullsize/h2v2 FDCT: ifast ENT: prog huff
- ./cjpeg -sample 2x2 -quality 100 -dct fast -prog -outfile testout_420_q100_ifast_prog.jpg $(srcdir)/testimages/testorig.ppm
- md5/md5cmp $(MD5_JPEG_420_IFAST_Q100_PROG) testout_420_q100_ifast_prog.jpg
-# CC: YCC->RGB SAMP: fullsize/h2v2 fancy IDCT: ifast ENT: prog huff
- ./djpeg -dct fast -outfile testout_420_q100_ifast.ppm testout_420_q100_ifast_prog.jpg
- md5/md5cmp $(MD5_PPM_420_Q100_IFAST) testout_420_q100_ifast.ppm
- rm -f testout_420_q100_ifast.ppm
-# CC: YCC->RGB SAMP: h2v2 merged IDCT: ifast ENT: prog huff
- ./djpeg -dct fast -nosmooth -outfile testout_420m_q100_ifast.ppm testout_420_q100_ifast_prog.jpg
- md5/md5cmp $(MD5_PPM_420M_Q100_IFAST) testout_420m_q100_ifast.ppm
- rm -f testout_420m_q100_ifast.ppm testout_420_q100_ifast_prog.jpg
-
-# CC: RGB->Gray SAMP: fullsize FDCT: islow ENT: huff
- ./cjpeg -gray -dct int -outfile testout_gray_islow.jpg $(srcdir)/testimages/testorig.ppm
- md5/md5cmp $(MD5_JPEG_GRAY_ISLOW) testout_gray_islow.jpg
-# CC: Gray->Gray SAMP: fullsize IDCT: islow ENT: huff
- ./djpeg -dct int -outfile testout_gray_islow.ppm testout_gray_islow.jpg
- md5/md5cmp $(MD5_PPM_GRAY_ISLOW) testout_gray_islow.ppm
- rm -f testout_gray_islow.ppm
-# CC: Gray->RGB SAMP: fullsize IDCT: islow ENT: huff
- ./djpeg -dct int -rgb -outfile testout_gray_islow_rgb.ppm testout_gray_islow.jpg
- md5/md5cmp $(MD5_PPM_GRAY_ISLOW_RGB) testout_gray_islow_rgb.ppm
- rm -f testout_gray_islow_rgb.ppm
-if WITH_12BIT
- rm -f testout_gray_islow.jpg
-else
-# CC: Gray->RGB565 SAMP: fullsize IDCT: islow ENT: huff
- ./djpeg -dct int -rgb565 -dither none -bmp -outfile testout_gray_islow_565.bmp testout_gray_islow.jpg
- md5/md5cmp $(MD5_BMP_GRAY_ISLOW_565) testout_gray_islow_565.bmp
- rm -f testout_gray_islow_565.bmp
-# CC: Gray->RGB565 (dithered) SAMP: fullsize IDCT: islow ENT: huff
- ./djpeg -dct int -rgb565 -bmp -outfile testout_gray_islow_565D.bmp testout_gray_islow.jpg
- md5/md5cmp $(MD5_BMP_GRAY_ISLOW_565D) testout_gray_islow_565D.bmp
- rm -f testout_gray_islow_565D.bmp testout_gray_islow.jpg
-endif
-
-# CC: RGB->YCC SAMP: fullsize smooth/h2v2 smooth FDCT: islow
-# ENT: 2-pass huff
- ./cjpeg -sample 2x2 -smooth 1 -dct int -opt -outfile testout_420s_ifast_opt.jpg $(srcdir)/testimages/testorig.ppm
- md5/md5cmp $(MD5_JPEG_420S_IFAST_OPT) testout_420s_ifast_opt.jpg
- rm -f testout_420s_ifast_opt.jpg
-
-# The output of the floating point tests is not validated by default, because
-# the output differs depending on the type of floating point math used, and
-# this is only deterministic if the DCT/IDCT are implemented using SIMD
-# instructions on a particular platform. Pass one of the following on the make
-# command line to validate the floating point tests against one of the expected
-# results:
-#
-# FLOATTEST=sse validate against the expected results from the libjpeg-turbo
-# SSE SIMD extensions
-# FLOATTEST=32bit validate against the expected results from the C code
-# when running on a 32-bit FPU (or when SSE is being used for
-# floating point math, which is generally the default with
-# x86-64 compilers)
-# FLOATTEST=64bit validate against the exepected results from the C code
-# when running on a 64-bit FPU
-# FLOATTEST=387 validate against the expected results from the C code when
-# the 387 FPU is being used for floating point math (which is
-# generally the default with x86 compilers)
-
-# CC: RGB->YCC SAMP: fullsize/int FDCT: float ENT: prog huff
- ./cjpeg -sample 3x2 -dct float -prog -outfile testout_3x2_float_prog.jpg $(srcdir)/testimages/testorig.ppm
- if [ "${FLOATTEST}" = "sse" ]; then \
- md5/md5cmp $(MD5_JPEG_3x2_FLOAT_PROG_SSE) testout_3x2_float_prog.jpg; \
- elif [ "${FLOATTEST}" = "32bit" -o "${FLOATTEST}" = "64bit" ]; then \
- md5/md5cmp $(MD5_JPEG_3x2_FLOAT_PROG_32BIT) testout_3x2_float_prog.jpg; \
- elif [ "${FLOATTEST}" = "387" ]; then \
- md5/md5cmp $(MD5_JPEG_3x2_FLOAT_PROG_387) testout_3x2_float_prog.jpg; \
- fi
-# CC: YCC->RGB SAMP: fullsize/int IDCT: float ENT: prog huff
- ./djpeg -dct float -outfile testout_3x2_float.ppm testout_3x2_float_prog.jpg
- if [ "${FLOATTEST}" = "sse" ]; then \
- md5/md5cmp $(MD5_PPM_3x2_FLOAT_SSE) testout_3x2_float.ppm; \
- elif [ "${FLOATTEST}" = "32bit" ]; then \
- md5/md5cmp $(MD5_PPM_3x2_FLOAT_32BIT) testout_3x2_float.ppm; \
- elif [ "${FLOATTEST}" = "64bit" ]; then \
- md5/md5cmp $(MD5_PPM_3x2_FLOAT_64BIT) testout_3x2_float.ppm; \
- elif [ "${FLOATTEST}" = "387" ]; then \
- md5/md5cmp $(MD5_PPM_3x2_FLOAT_387) testout_3x2_float.ppm; \
- fi
- rm -f testout_3x2_float.ppm testout_3x2_float_prog.jpg
-
-# CC: RGB->YCC SAMP: fullsize/int FDCT: ifast ENT: prog huff
- ./cjpeg -sample 3x2 -dct fast -prog -outfile testout_3x2_ifast_prog.jpg $(srcdir)/testimages/testorig.ppm
- md5/md5cmp $(MD5_JPEG_3x2_IFAST_PROG) testout_3x2_ifast_prog.jpg
-# CC: YCC->RGB SAMP: fullsize/int IDCT: ifast ENT: prog huff
- ./djpeg -dct fast -outfile testout_3x2_ifast.ppm testout_3x2_ifast_prog.jpg
- md5/md5cmp $(MD5_PPM_3x2_IFAST) testout_3x2_ifast.ppm
- rm -f testout_3x2_ifast.ppm testout_3x2_ifast_prog.jpg
-
-if WITH_ARITH_ENC
-# CC: YCC->RGB SAMP: fullsize/h2v2 FDCT: islow ENT: arith
- ./cjpeg -dct int -arithmetic -outfile testout_420_islow_ari.jpg $(srcdir)/testimages/testorig.ppm
- md5/md5cmp $(MD5_JPEG_420_ISLOW_ARI) testout_420_islow_ari.jpg
- rm -f testout_420_islow_ari.jpg
- ./jpegtran -arithmetic -outfile testout_420_islow_ari.jpg $(srcdir)/testimages/testimgint.jpg
- md5/md5cmp $(MD5_JPEG_420_ISLOW_ARI) testout_420_islow_ari.jpg
- rm -f testout_420_islow_ari.jpg
-# CC: YCC->RGB SAMP: fullsize FDCT: islow ENT: prog arith
- ./cjpeg -sample 1x1 -dct int -prog -arithmetic -outfile testout_444_islow_progari.jpg $(srcdir)/testimages/testorig.ppm
- md5/md5cmp $(MD5_JPEG_444_ISLOW_PROGARI) testout_444_islow_progari.jpg
- rm -f testout_444_islow_progari.jpg
-endif
-if WITH_ARITH_DEC
-# CC: RGB->YCC SAMP: h2v2 merged IDCT: ifast ENT: arith
- ./djpeg -fast -ppm -outfile testout_420m_ifast_ari.ppm $(srcdir)/testimages/testimgari.jpg
- md5/md5cmp $(MD5_PPM_420M_IFAST_ARI) testout_420m_ifast_ari.ppm
- rm -f testout_420m_ifast_ari.ppm
- ./jpegtran -outfile testout_420_islow.jpg $(srcdir)/testimages/testimgari.jpg
- md5/md5cmp $(MD5_JPEG_420_ISLOW) testout_420_islow.jpg
- rm -f testout_420_islow.jpg
-endif
-
-# CC: YCC->RGB SAMP: h2v2 merged IDCT: 16x16 islow ENT: huff
- ./djpeg -dct int -scale 2/1 -nosmooth -ppm -outfile testout_420m_islow_2_1.ppm $(srcdir)/testimages/$(TESTORIG)
- md5/md5cmp $(MD5_PPM_420M_ISLOW_2_1) testout_420m_islow_2_1.ppm
- rm -f testout_420m_islow_2_1.ppm
-# CC: YCC->RGB SAMP: h2v2 merged IDCT: 15x15 islow ENT: huff
- ./djpeg -dct int -scale 15/8 -nosmooth -ppm -outfile testout_420m_islow_15_8.ppm $(srcdir)/testimages/$(TESTORIG)
- md5/md5cmp $(MD5_PPM_420M_ISLOW_15_8) testout_420m_islow_15_8.ppm
- rm -f testout_420m_islow_15_8.ppm
-# CC: YCC->RGB SAMP: h2v2 merged IDCT: 13x13 islow ENT: huff
- ./djpeg -dct int -scale 13/8 -nosmooth -ppm -outfile testout_420m_islow_13_8.ppm $(srcdir)/testimages/$(TESTORIG)
- md5/md5cmp $(MD5_PPM_420M_ISLOW_13_8) testout_420m_islow_13_8.ppm
- rm -f testout_420m_islow_13_8.ppm
-# CC: YCC->RGB SAMP: h2v2 merged IDCT: 11x11 islow ENT: huff
- ./djpeg -dct int -scale 11/8 -nosmooth -ppm -outfile testout_420m_islow_11_8.ppm $(srcdir)/testimages/$(TESTORIG)
- md5/md5cmp $(MD5_PPM_420M_ISLOW_11_8) testout_420m_islow_11_8.ppm
- rm -f testout_420m_islow_11_8.ppm
-# CC: YCC->RGB SAMP: h2v2 merged IDCT: 9x9 islow ENT: huff
- ./djpeg -dct int -scale 9/8 -nosmooth -ppm -outfile testout_420m_islow_9_8.ppm $(srcdir)/testimages/$(TESTORIG)
- md5/md5cmp $(MD5_PPM_420M_ISLOW_9_8) testout_420m_islow_9_8.ppm
- rm -f testout_420m_islow_9_8.ppm
-# CC: YCC->RGB SAMP: h2v2 merged IDCT: 7x7 islow/14x14 islow ENT: huff
- ./djpeg -dct int -scale 7/8 -nosmooth -ppm -outfile testout_420m_islow_7_8.ppm $(srcdir)/testimages/$(TESTORIG)
- md5/md5cmp $(MD5_PPM_420M_ISLOW_7_8) testout_420m_islow_7_8.ppm
- rm -f testout_420m_islow_7_8.ppm
-# CC: YCC->RGB SAMP: h2v2 merged IDCT: 6x6 islow/12x12 islow ENT: huff
- ./djpeg -dct int -scale 3/4 -nosmooth -ppm -outfile testout_420m_islow_3_4.ppm $(srcdir)/testimages/$(TESTORIG)
- md5/md5cmp $(MD5_PPM_420M_ISLOW_3_4) testout_420m_islow_3_4.ppm
- rm -f testout_420m_islow_3_4.ppm
-# CC: YCC->RGB SAMP: h2v2 merged IDCT: 5x5 islow/10x10 islow ENT: huff
- ./djpeg -dct int -scale 5/8 -nosmooth -ppm -outfile testout_420m_islow_5_8.ppm $(srcdir)/testimages/$(TESTORIG)
- md5/md5cmp $(MD5_PPM_420M_ISLOW_5_8) testout_420m_islow_5_8.ppm
- rm -f testout_420m_islow_5_8.ppm
-# CC: YCC->RGB SAMP: h2v2 merged IDCT: 4x4 islow/8x8 islow ENT: huff
- ./djpeg -dct int -scale 1/2 -nosmooth -ppm -outfile testout_420m_islow_1_2.ppm $(srcdir)/testimages/$(TESTORIG)
- md5/md5cmp $(MD5_PPM_420M_ISLOW_1_2) testout_420m_islow_1_2.ppm
- rm -f testout_420m_islow_1_2.ppm
-# CC: YCC->RGB SAMP: h2v2 merged IDCT: 3x3 islow/6x6 islow ENT: huff
- ./djpeg -dct int -scale 3/8 -nosmooth -ppm -outfile testout_420m_islow_3_8.ppm $(srcdir)/testimages/$(TESTORIG)
- md5/md5cmp $(MD5_PPM_420M_ISLOW_3_8) testout_420m_islow_3_8.ppm
- rm -f testout_420m_islow_3_8.ppm
-# CC: YCC->RGB SAMP: h2v2 merged IDCT: 2x2 islow/4x4 islow ENT: huff
- ./djpeg -dct int -scale 1/4 -nosmooth -ppm -outfile testout_420m_islow_1_4.ppm $(srcdir)/testimages/$(TESTORIG)
- md5/md5cmp $(MD5_PPM_420M_ISLOW_1_4) testout_420m_islow_1_4.ppm
- rm -f testout_420m_islow_1_4.ppm
-# CC: YCC->RGB SAMP: h2v2 merged IDCT: 1x1 islow/2x2 islow ENT: huff
- ./djpeg -dct int -scale 1/8 -nosmooth -ppm -outfile testout_420m_islow_1_8.ppm $(srcdir)/testimages/$(TESTORIG)
- md5/md5cmp $(MD5_PPM_420M_ISLOW_1_8) testout_420m_islow_1_8.ppm
- rm -f testout_420m_islow_1_8.ppm
-if WITH_12BIT
-else
-# CC: YCC->RGB (dithered) SAMP: h2v2 fancy IDCT: islow ENT: huff
- ./djpeg -dct int -colors 256 -bmp -outfile testout_420_islow_256.bmp $(srcdir)/testimages/$(TESTORIG)
- md5/md5cmp $(MD5_BMP_420_ISLOW_256) testout_420_islow_256.bmp
- rm -f testout_420_islow_256.bmp
-# CC: YCC->RGB565 SAMP: h2v2 fancy IDCT: islow ENT: huff
- ./djpeg -dct int -rgb565 -dither none -bmp -outfile testout_420_islow_565.bmp $(srcdir)/testimages/$(TESTORIG)
- md5/md5cmp $(MD5_BMP_420_ISLOW_565) testout_420_islow_565.bmp
- rm -f testout_420_islow_565.bmp
-# CC: YCC->RGB565 (dithered) SAMP: h2v2 fancy IDCT: islow ENT: huff
- ./djpeg -dct int -rgb565 -bmp -outfile testout_420_islow_565D.bmp $(srcdir)/testimages/$(TESTORIG)
- md5/md5cmp $(MD5_BMP_420_ISLOW_565D) testout_420_islow_565D.bmp
- rm -f testout_420_islow_565D.bmp
-# CC: YCC->RGB565 SAMP: h2v2 merged IDCT: islow ENT: huff
- ./djpeg -dct int -nosmooth -rgb565 -dither none -bmp -outfile testout_420m_islow_565.bmp $(srcdir)/testimages/$(TESTORIG)
- md5/md5cmp $(MD5_BMP_420M_ISLOW_565) testout_420m_islow_565.bmp
- rm -f testout_420m_islow_565.bmp
-# CC: YCC->RGB565 (dithered) SAMP: h2v2 merged IDCT: islow ENT: huff
- ./djpeg -dct int -nosmooth -rgb565 -bmp -outfile testout_420m_islow_565D.bmp $(srcdir)/testimages/$(TESTORIG)
- md5/md5cmp $(MD5_BMP_420M_ISLOW_565D) testout_420m_islow_565D.bmp
- rm -f testout_420m_islow_565D.bmp
-endif
-
-# Partial decode tests. These tests are designed to cover all of the possible
-# code paths in jpeg_skip_scanlines().
-
-# Context rows: Yes Intra-iMCU row: Yes iMCU row prefetch: No ENT: huff
- ./djpeg -dct int -skip 15,31 -ppm -outfile testout_420_islow_skip15,31.ppm $(srcdir)/testimages/$(TESTORIG)
- md5/md5cmp $(MD5_PPM_420_ISLOW_SKIP15_31) testout_420_islow_skip15,31.ppm
- rm -f testout_420_islow_skip15,31.ppm
-# Context rows: Yes Intra-iMCU row: No iMCU row prefetch: Yes ENT: arith
-if WITH_ARITH_DEC
- ./djpeg -dct int -skip 16,139 -ppm -outfile testout_420_islow_ari_skip16,139.ppm $(srcdir)/testimages/testimgari.jpg
- md5/md5cmp $(MD5_PPM_420_ISLOW_ARI_SKIP16_139) testout_420_islow_ari_skip16,139.ppm
- rm -f testout_420_islow_ari_skip16,139.ppm
-endif
-# Context rows: Yes Intra-iMCU row: No iMCU row prefetch: No ENT: prog huff
- ./cjpeg -dct int -prog -outfile testout_420_islow_prog.jpg $(srcdir)/testimages/testorig.ppm
- ./djpeg -dct int -crop 62x62+71+71 -ppm -outfile testout_420_islow_prog_crop62x62,71,71.ppm testout_420_islow_prog.jpg
- md5/md5cmp $(MD5_PPM_420_ISLOW_PROG_CROP62x62_71_71) testout_420_islow_prog_crop62x62,71,71.ppm
- rm -f testout_420_islow_prog_crop62x62,71,71.ppm testout_420_islow_prog.jpg
-# Context rows: Yes Intra-iMCU row: No iMCU row prefetch: No ENT: arith
-if WITH_ARITH_DEC
- ./djpeg -dct int -crop 53x53+4+4 -ppm -outfile testout_420_islow_ari_crop53x53,4,4.ppm $(srcdir)/testimages/testimgari.jpg
- md5/md5cmp $(MD5_PPM_420_ISLOW_ARI_CROP53x53_4_4) testout_420_islow_ari_crop53x53,4,4.ppm
- rm -f testout_420_islow_ari_crop53x53,4,4.ppm
-endif
-# Context rows: No Intra-iMCU row: Yes ENT: huff
- ./cjpeg -dct int -sample 1x1 -outfile testout_444_islow.jpg $(srcdir)/testimages/testorig.ppm
- ./djpeg -dct int -skip 1,6 -ppm -outfile testout_444_islow_skip1,6.ppm testout_444_islow.jpg
- md5/md5cmp $(MD5_PPM_444_ISLOW_SKIP1_6) testout_444_islow_skip1,6.ppm
- rm -f testout_444_islow_skip1,6.ppm testout_444_islow.jpg
-# Context rows: No Intra-iMCU row: No ENT: prog huff
- ./cjpeg -dct int -prog -sample 1x1 -outfile testout_444_islow_prog.jpg $(srcdir)/testimages/testorig.ppm
- ./djpeg -dct int -crop 98x98+13+13 -ppm -outfile testout_444_islow_prog_crop98x98,13,13.ppm testout_444_islow_prog.jpg
- md5/md5cmp $(MD5_PPM_444_ISLOW_PROG_CROP98x98_13_13) testout_444_islow_prog_crop98x98,13,13.ppm
- rm -f testout_444_islow_prog_crop98x98,13,13.ppm testout_444_islow_prog.jpg
-# Context rows: No Intra-iMCU row: No ENT: arith
-if WITH_ARITH_ENC
- ./cjpeg -dct int -arithmetic -sample 1x1 -outfile testout_444_islow_ari.jpg $(srcdir)/testimages/testorig.ppm
-if WITH_ARITH_DEC
- ./djpeg -dct int -crop 37x37+0+0 -ppm -outfile testout_444_islow_ari_crop37x37,0,0.ppm testout_444_islow_ari.jpg
- md5/md5cmp $(MD5_PPM_444_ISLOW_ARI_CROP37x37_0_0) testout_444_islow_ari_crop37x37,0,0.ppm
- rm -f testout_444_islow_ari_crop37x37,0,0.ppm
-endif
- rm -f testout_444_islow_ari.jpg
-endif
-
- ./jpegtran -crop 120x90+20+50 -transpose -perfect -outfile testout_crop.jpg $(srcdir)/testimages/$(TESTORIG)
- md5/md5cmp $(MD5_JPEG_CROP) testout_crop.jpg
- rm -f testout_crop.jpg
- echo GREAT SUCCESS!
-
-
-testclean:
- rm -f testout*
- rm -f *_GRAY_*.bmp
- rm -f *_GRAY_*.png
- rm -f *_GRAY_*.ppm
- rm -f *_GRAY_*.jpg
- rm -f *_GRAY.yuv
- rm -f *_420_*.bmp
- rm -f *_420_*.png
- rm -f *_420_*.ppm
- rm -f *_420_*.jpg
- rm -f *_420.yuv
- rm -f *_422_*.bmp
- rm -f *_422_*.png
- rm -f *_422_*.ppm
- rm -f *_422_*.jpg
- rm -f *_422.yuv
- rm -f *_444_*.bmp
- rm -f *_444_*.png
- rm -f *_444_*.ppm
- rm -f *_444_*.jpg
- rm -f *_444.yuv
- rm -f *_440_*.bmp
- rm -f *_440_*.png
- rm -f *_440_*.ppm
- rm -f *_440_*.jpg
- rm -f *_440.yuv
- rm -f *_411_*.bmp
- rm -f *_411_*.png
- rm -f *_411_*.ppm
- rm -f *_411_*.jpg
- rm -f *_411.yuv
-
-
-tjtest:
- sh ./tjbenchtest
- sh ./tjbenchtest -alloc
- sh ./tjbenchtest -yuv
- sh ./tjbenchtest -yuv -alloc
-if WITH_JAVA
- sh ./tjbenchtest.java
- sh ./tjbenchtest.java -yuv
-endif
-
-
-pkgscripts/libjpeg-turbo.spec: pkgscripts/libjpeg-turbo.spec.tmpl
- cat pkgscripts/libjpeg-turbo.spec.tmpl | sed s@%{__prefix}@$(prefix)@g | \
- sed s@%{__bindir}@$(bindir)@g | sed s@%{__datadir}@$(datadir)@g | \
- sed s@%{__docdir}@$(docdir)@g | sed s@%{__includedir}@$(includedir)@g | \
- sed s@%{__libdir}@$(libdir)@g | sed s@%{__mandir}@$(mandir)@g \
- > pkgscripts/libjpeg-turbo.spec
-
-rpm: all pkgscripts/libjpeg-turbo.spec
- TMPDIR=`mktemp -d /tmp/${PACKAGE_NAME}-build.XXXXXX`; \
- mkdir -p $$TMPDIR/RPMS; \
- ln -fs `pwd` $$TMPDIR/BUILD; \
- rm -f ${PKGNAME}-${VERSION}.${RPMARCH}.rpm; \
- rpmbuild -bb --define "_blddir $$TMPDIR/buildroot" \
- --define "_topdir $$TMPDIR" \
- --target ${RPMARCH} pkgscripts/libjpeg-turbo.spec; \
- cp $$TMPDIR/RPMS/${RPMARCH}/${PKGNAME}-${VERSION}-${BUILD}.${RPMARCH}.rpm \
- ${PKGNAME}-${VERSION}.${RPMARCH}.rpm; \
- rm -rf $$TMPDIR
-
-srpm: dist-gzip pkgscripts/libjpeg-turbo.spec
- TMPDIR=`mktemp -d /tmp/${PACKAGE_NAME}-build.XXXXXX`; \
- mkdir -p $$TMPDIR/RPMS; \
- mkdir -p $$TMPDIR/SRPMS; \
- mkdir -p $$TMPDIR/BUILD; \
- mkdir -p $$TMPDIR/SOURCES; \
- mkdir -p $$TMPDIR/SPECS; \
- rm -f ${PKGNAME}-${VERSION}.src.rpm; \
- cp ${PACKAGE_NAME}-${VERSION}.tar.gz $$TMPDIR/SOURCES; \
- cat pkgscripts/libjpeg-turbo.spec | sed s/%{_blddir}/%{_tmppath}/g \
- | sed s/#--\>//g \
- > $$TMPDIR/SPECS/libjpeg-turbo.spec; \
- rpmbuild -bs --define "_topdir $$TMPDIR" $$TMPDIR/SPECS/libjpeg-turbo.spec; \
- cp $$TMPDIR/SRPMS/${PKGNAME}-${VERSION}-${BUILD}.src.rpm \
- ${PKGNAME}-${VERSION}.src.rpm; \
- rm -rf $$TMPDIR
-
-pkgscripts/makedpkg: pkgscripts/makedpkg.tmpl
- cat pkgscripts/makedpkg.tmpl | sed s@%{__prefix}@$(prefix)@g | \
- sed s@%{__docdir}@$(docdir)@g | sed s@%{__libdir}@$(libdir)@g \
- > pkgscripts/makedpkg
-
-deb: all pkgscripts/makedpkg
- sh pkgscripts/makedpkg
-
-pkgscripts/uninstall: pkgscripts/uninstall.tmpl
- cat pkgscripts/uninstall.tmpl | sed s@%{__prefix}@$(prefix)@g | \
- sed s@%{__bindir}@$(bindir)@g | sed s@%{__datadir}@$(datadir)@g | \
- sed s@%{__includedir}@$(includedir)@g | sed s@%{__libdir}@$(libdir)@g | \
- sed s@%{__mandir}@$(mandir)@g > pkgscripts/uninstall
-
-pkgscripts/makemacpkg: pkgscripts/makemacpkg.tmpl
- cat pkgscripts/makemacpkg.tmpl | sed s@%{__prefix}@$(prefix)@g | \
- sed s@%{__bindir}@$(bindir)@g | sed s@%{__docdir}@$(docdir)@g | \
- sed s@%{__libdir}@$(libdir)@g > pkgscripts/makemacpkg
-
-if X86_64
-
-udmg: all pkgscripts/makemacpkg pkgscripts/uninstall
- sh pkgscripts/makemacpkg -build32 ${BUILDDIR32}
-
-iosdmg: all pkgscripts/makemacpkg pkgscripts/uninstall
- sh pkgscripts/makemacpkg -build32 ${BUILDDIR32} -buildarmv7 ${BUILDDIRARMV7} -buildarmv7s ${BUILDDIRARMV7S} -buildarmv8 ${BUILDDIRARMV8} -lipo "${LIPO}"
-
-else
-
-iosdmg: all pkgscripts/makemacpkg pkgscripts/uninstall
- sh pkgscripts/makemacpkg -buildarmv7 ${BUILDDIRARMV7} -buildarmv7s ${BUILDDIRARMV7S} -buildarmv8 ${BUILDDIRARMV8} -lipo "${LIPO}"
-
-endif
-
-dmg: all pkgscripts/makemacpkg pkgscripts/uninstall
- sh pkgscripts/makemacpkg
-
-pkgscripts/makecygwinpkg: pkgscripts/makecygwinpkg.tmpl
- cat pkgscripts/makecygwinpkg.tmpl | sed s@%{__prefix}@$(prefix)@g | \
- sed s@%{__docdir}@$(docdir)@g | sed s@%{__libdir}@$(libdir)@g \
- > pkgscripts/makecygwinpkg
-
-cygwinpkg: all pkgscripts/makecygwinpkg
- sh pkgscripts/makecygwinpkg
diff --git a/README.ijg b/README.ijg
index 9c450ce..d9d81b8 100644
--- a/README.ijg
+++ b/README.ijg
@@ -159,12 +159,6 @@
assumed by the product vendor.
-The Unix configuration script "configure" was produced with GNU Autoconf.
-It is copyright by the Free Software Foundation but is freely distributable.
-The same holds for its supporting scripts (config.guess, config.sub,
-ltmain.sh). Another support script, install-sh, is copyright by X Consortium
-but is also freely distributable.
-
The IJG distribution formerly included code to read and write GIF files.
To avoid entanglement with the Unisys LZW patent (now expired), GIF reading
support has been removed altogether, and the GIF writer has been simplified
diff --git a/README.md b/README.md
index 74e6eac..d54feb7 100755
--- a/README.md
+++ b/README.md
@@ -2,8 +2,8 @@
==========
libjpeg-turbo is a JPEG image codec that uses SIMD instructions (MMX, SSE2,
-NEON, AltiVec) to accelerate baseline JPEG compression and decompression on
-x86, x86-64, ARM, and PowerPC systems. On such systems, libjpeg-turbo is
+AVX2, NEON, AltiVec) to accelerate baseline JPEG compression and decompression
+on x86, x86-64, ARM, and PowerPC systems. On such systems, libjpeg-turbo is
generally 2-6x as fast as libjpeg, all else being equal. On other types of
systems, libjpeg-turbo can still outperform libjpeg by a significant amount, by
virtue of its highly-optimized Huffman coding routines. In many cases, the
diff --git a/acinclude.m4 b/acinclude.m4
deleted file mode 100644
index 113169f..0000000
--- a/acinclude.m4
+++ /dev/null
@@ -1,287 +0,0 @@
-# AC_PROG_NASM
-# --------------------------
-# Check that NASM exists and determine flags
-AC_DEFUN([AC_PROG_NASM],[
-
-AC_ARG_VAR(NASM, [NASM command (used to build the x86/x86-64 SIMD code)])
-if test "x$NASM" = "x"; then
- AC_CHECK_PROGS(NASM, [nasm nasmw yasm])
- test -z "$NASM" && AC_MSG_ERROR([no nasm (Netwide Assembler) found])
-fi
-
-AC_MSG_CHECKING([for object file format of host system])
-case "$host_os" in
- cygwin* | mingw* | pw32* | interix*)
- case "$host_cpu" in
- x86_64)
- objfmt='Win64-COFF'
- ;;
- *)
- objfmt='Win32-COFF'
- ;;
- esac
- ;;
- msdosdjgpp* | go32*)
- objfmt='COFF'
- ;;
- os2-emx*) # not tested
- objfmt='MSOMF' # obj
- ;;
- linux*coff* | linux*oldld*)
- objfmt='COFF' # ???
- ;;
- linux*aout*)
- objfmt='a.out'
- ;;
- linux*)
- case "$host_cpu" in
- x86_64)
- objfmt='ELF64'
- ;;
- *)
- objfmt='ELF'
- ;;
- esac
- ;;
- kfreebsd* | freebsd* | netbsd* | openbsd*)
- if echo __ELF__ | $CC -E - | grep __ELF__ > /dev/null; then
- objfmt='BSD-a.out'
- else
- case "$host_cpu" in
- x86_64 | amd64)
- objfmt='ELF64'
- ;;
- *)
- objfmt='ELF'
- ;;
- esac
- fi
- ;;
- solaris* | sunos* | sysv* | sco*)
- case "$host_cpu" in
- x86_64)
- objfmt='ELF64'
- ;;
- *)
- objfmt='ELF'
- ;;
- esac
- ;;
- darwin* | rhapsody* | nextstep* | openstep* | macos*)
- case "$host_cpu" in
- x86_64)
- objfmt='Mach-O64'
- ;;
- *)
- objfmt='Mach-O'
- ;;
- esac
- ;;
- *)
- objfmt='ELF ?'
- ;;
-esac
-
-AC_MSG_RESULT([$objfmt])
-if test "$objfmt" = 'ELF ?'; then
- objfmt='ELF'
- AC_MSG_WARN([unexpected host system. assumed that the format is $objfmt.])
-fi
-
-AC_MSG_CHECKING([for object file format specifier (NAFLAGS) ])
-case "$objfmt" in
- MSOMF) NAFLAGS='-fobj -DOBJ32';;
- Win32-COFF) NAFLAGS='-fwin32 -DWIN32';;
- Win64-COFF) NAFLAGS='-fwin64 -DWIN64 -D__x86_64__';;
- COFF) NAFLAGS='-fcoff -DCOFF';;
- a.out) NAFLAGS='-faout -DAOUT';;
- BSD-a.out) NAFLAGS='-faoutb -DAOUT';;
- ELF) NAFLAGS='-felf -DELF';;
- ELF64) NAFLAGS='-felf64 -DELF -D__x86_64__';;
- RDF) NAFLAGS='-frdf -DRDF';;
- Mach-O) NAFLAGS='-fmacho -DMACHO';;
- Mach-O64) NAFLAGS='-fmacho64 -DMACHO -D__x86_64__';;
-esac
-AC_MSG_RESULT([$NAFLAGS])
-AC_SUBST([NAFLAGS])
-
-AC_MSG_CHECKING([whether the assembler ($NASM $NAFLAGS) works])
-cat > conftest.asm <<EOF
-[%line __oline__ "configure"
- section .text
- global _main,main
-_main:
-main: xor eax,eax
- ret
-]EOF
-try_nasm='$NASM $NAFLAGS -o conftest.o conftest.asm'
-if AC_TRY_EVAL(try_nasm) && test -s conftest.o; then
- AC_MSG_RESULT(yes)
-else
- echo "configure: failed program was:" >&AC_FD_CC
- cat conftest.asm >&AC_FD_CC
- rm -rf conftest*
- AC_MSG_RESULT(no)
- AC_MSG_ERROR([installation or configuration problem: assembler cannot create object files.])
-fi
-
-AC_MSG_CHECKING([whether the linker accepts assembler output])
-try_nasm='${CC-cc} -o conftest${ac_exeext} $LDFLAGS conftest.o $LIBS 1>&AC_FD_CC'
-if AC_TRY_EVAL(try_nasm) && test -s conftest${ac_exeext}; then
- rm -rf conftest*
- AC_MSG_RESULT(yes)
-else
- rm -rf conftest*
- AC_MSG_RESULT(no)
- AC_MSG_ERROR([configuration problem: maybe object file format mismatch.])
-fi
-
-])
-
-# AC_CHECK_COMPATIBLE_ARM_ASSEMBLER_IFELSE
-# --------------------------
-# Test whether the assembler is suitable and supports NEON instructions
-AC_DEFUN([AC_CHECK_COMPATIBLE_ARM_ASSEMBLER_IFELSE],[
- ac_good_gnu_arm_assembler=no
- ac_save_CC="$CC"
- ac_save_CFLAGS="$CFLAGS"
- CFLAGS="$CCASFLAGS -x assembler-with-cpp"
- CC="$CCAS"
- AC_COMPILE_IFELSE([AC_LANG_SOURCE([[
- .text
- .fpu neon
- .arch armv7a
- .object_arch armv4
- .arm
- pld [r0]
- vmovn.u16 d0, q0]])], ac_good_gnu_arm_assembler=yes)
-
- ac_use_gas_preprocessor=no
- if test "x$ac_good_gnu_arm_assembler" = "xno" ; then
- CC="gas-preprocessor.pl $CCAS"
- AC_COMPILE_IFELSE([AC_LANG_SOURCE([[
- .text
- .fpu neon
- .arch armv7a
- .object_arch armv4
- .arm
- pld [r0]
- vmovn.u16 d0, q0]])], ac_use_gas_preprocessor=yes)
- fi
- CFLAGS="$ac_save_CFLAGS"
- CC="$ac_save_CC"
-
- if test "x$ac_use_gas_preprocessor" = "xyes" ; then
- CCAS="gas-preprocessor.pl $CCAS"
- AC_SUBST([CCAS])
- ac_good_gnu_arm_assembler=yes
- fi
-
- if test "x$ac_good_gnu_arm_assembler" = "xyes" ; then
- $1
- else
- $2
- fi
-])
-
-# AC_CHECK_COMPATIBLE_MIPSEL_ASSEMBLER_IFELSE
-# --------------------------
-# Test whether the assembler is suitable and supports MIPS instructions
-AC_DEFUN([AC_CHECK_COMPATIBLE_MIPS_ASSEMBLER_IFELSE],[
- have_mips_dspr2=no
- ac_save_CFLAGS="$CFLAGS"
- CFLAGS="$CCASFLAGS -mdspr2"
-
- AC_COMPILE_IFELSE([AC_LANG_SOURCE([[
-
- int main ()
- {
- int c = 0, a = 0, b = 0;
- __asm__ __volatile__ (
- "precr.qb.ph %[c], %[a], %[b] \n\t"
- : [c] "=r" (c)
- : [a] "r" (a), [b] "r" (b)
- );
- return c;
- }
- ]])], have_mips_dspr2=yes)
- CFLAGS=$ac_save_CFLAGS
-
- if test "x$have_mips_dspr2" = "xyes" ; then
- $1
- else
- $2
- fi
-])
-
-AC_DEFUN([AC_CHECK_COMPATIBLE_ARM64_ASSEMBLER_IFELSE],[
- ac_good_gnu_arm_assembler=no
- ac_save_CC="$CC"
- ac_save_CFLAGS="$CFLAGS"
- CFLAGS="$CCASFLAGS -x assembler-with-cpp"
- CC="$CCAS"
- AC_COMPILE_IFELSE([AC_LANG_SOURCE([[
- .text
- MYVAR .req x0
- movi v0.16b, #100
- mov MYVAR, #100
- .unreq MYVAR]])], ac_good_gnu_arm_assembler=yes)
-
- ac_use_gas_preprocessor=no
- if test "x$ac_good_gnu_arm_assembler" = "xno" ; then
- CC="gas-preprocessor.pl $CCAS"
- AC_COMPILE_IFELSE([AC_LANG_SOURCE([[
- .text
- MYVAR .req x0
- movi v0.16b, #100
- mov MYVAR, #100
- .unreq MYVAR]])], ac_use_gas_preprocessor=yes)
- fi
- CFLAGS="$ac_save_CFLAGS"
- CC="$ac_save_CC"
-
- if test "x$ac_use_gas_preprocessor" = "xyes" ; then
- CCAS="gas-preprocessor.pl $CCAS"
- AC_SUBST([CCAS])
- ac_good_gnu_arm_assembler=yes
- fi
-
- if test "x$ac_good_gnu_arm_assembler" = "xyes" ; then
- $1
- else
- $2
- fi
-])
-
-# AC_CHECK_ALTIVEC
-# ----------------
-# Test whether AltiVec intrinsics are supported
-AC_DEFUN([AC_CHECK_ALTIVEC],[
- ac_save_CFLAGS="$CFLAGS"
- CFLAGS="$CFLAGS -maltivec"
- AC_COMPILE_IFELSE([AC_LANG_SOURCE([[
- #include <altivec.h>
- int main(void) {
- __vector int vi = { 0, 0, 0, 0 };
- int i[4];
- vec_st(vi, 0, i);
- return i[0];
- }]])], ac_has_altivec=yes)
- CFLAGS="$ac_save_CFLAGS"
- if test "x$ac_has_altivec" = "xyes" ; then
- $1
- else
- $2
- fi
-])
-
-AC_DEFUN([AC_NO_SIMD],[
- AC_MSG_RESULT([no ("$1")])
- with_simd=no;
- if test "x${require_simd}" = "xyes"; then
- AC_MSG_ERROR([SIMD support not available for this CPU.])
- else
- AC_MSG_WARN([SIMD support not available for this CPU. Performance will\
- suffer.])
- fi
-])
diff --git a/appveyor.yml b/appveyor.yml
index 4f2d6cc..0acc329 100644
--- a/appveyor.yml
+++ b/appveyor.yml
@@ -16,11 +16,11 @@
set MSYSTEM=MINGW32
- bash -c "pacman --noconfirm -S autoconf automake libtool zip"
+ bash -c "pacman --noconfirm -S zip"
mklink /d "%ProgramData%\Oracle\Java32" "c:\Program Files (x86)\Java\jdk1.6.0"
- git clone --depth=1 https://github.com/libjpeg-turbo/buildscripts.git c:/buildscripts
+ git clone --depth=1 https://github.com/libjpeg-turbo/buildscripts.git -b %APPVEYOR_REPO_BRANCH% c:/buildscripts
build_script:
- cmd: >-
diff --git a/cjpeg.1 b/cjpeg.1
index 283fc81..557b0b0 100644
--- a/cjpeg.1
+++ b/cjpeg.1
@@ -187,6 +187,9 @@
roundoff behavior, whereas the integer methods should give the same results on
all machines.
.TP
+.BI \-icc " file"
+Embed ICC color management profile contained in the specified file.
+.TP
.BI \-restart " N"
Emit a JPEG restart marker every N MCU rows, or every N MCU blocks if "B" is
attached to the number.
diff --git a/cjpeg.c b/cjpeg.c
index 9d282b8..da9af73 100644
--- a/cjpeg.c
+++ b/cjpeg.c
@@ -31,6 +31,11 @@
#include "jversion.h" /* for version message */
#include "jconfigint.h"
+#ifndef HAVE_STDLIB_H /* <stdlib.h> should declare malloc(),free() */
+extern void *malloc (size_t size);
+extern void free (void *ptr);
+#endif
+
#ifdef USE_CCOMMAND /* command-line reader for Macintosh */
#ifdef __MWERKS__
#include <SIOUX.h> /* Metrowerks needs this */
@@ -139,6 +144,7 @@
static const char *progname; /* program name for error messages */
+static char *icc_filename; /* for -icc switch */
static char *outfilename; /* for -outfile switch */
boolean memdst; /* for -memdst switch */
@@ -184,6 +190,7 @@
fprintf(stderr, " -dct float Use floating-point DCT method%s\n",
(JDCT_DEFAULT == JDCT_FLOAT ? " (default)" : ""));
#endif
+ fprintf(stderr, " -icc FILE Embed ICC profile contained in FILE\n");
fprintf(stderr, " -restart N Set restart interval in rows, or in blocks with B\n");
#ifdef INPUT_SMOOTHING_SUPPORTED
fprintf(stderr, " -smooth N Smooth dithered input (N=1..100 is strength)\n");
@@ -234,6 +241,7 @@
force_baseline = FALSE; /* by default, allow 16-bit quantizers */
simple_progressive = FALSE;
is_targa = FALSE;
+ icc_filename = NULL;
outfilename = NULL;
memdst = FALSE;
cinfo->err->trace_level = 0;
@@ -307,6 +315,12 @@
/* Force an RGB JPEG file to be generated. */
jpeg_set_colorspace(cinfo, JCS_RGB);
+ } else if (keymatch(arg, "icc", 1)) {
+ /* Set ICC filename. */
+ if (++argn >= argc) /* advance to next argument */
+ usage();
+ icc_filename = argv[argn];
+
} else if (keymatch(arg, "maxmemory", 3)) {
/* Maximum memory in Kb (or Mb with 'm'). */
long lval;
@@ -496,6 +510,9 @@
int file_index;
cjpeg_source_ptr src_mgr;
FILE *input_file;
+ FILE *icc_file;
+ JOCTET *icc_profile = NULL;
+ long icc_len = 0;
FILE *output_file = NULL;
unsigned char *outbuffer = NULL;
unsigned long outsize = 0;
@@ -583,6 +600,33 @@
output_file = write_stdout();
}
+ if (icc_filename != NULL) {
+ if ((icc_file = fopen(icc_filename, READ_BINARY)) == NULL) {
+ fprintf(stderr, "%s: can't open %s\n", progname, icc_filename);
+ exit(EXIT_FAILURE);
+ }
+ if (fseek(icc_file, 0, SEEK_END) < 0 ||
+ (icc_len = ftell(icc_file)) < 1 ||
+ fseek(icc_file, 0, SEEK_SET) < 0) {
+ fprintf(stderr, "%s: can't determine size of %s\n", progname,
+ icc_filename);
+ exit(EXIT_FAILURE);
+ }
+ if ((icc_profile = (JOCTET *)malloc(icc_len)) == NULL) {
+ fprintf(stderr, "%s: can't allocate memory for ICC profile\n", progname);
+ fclose(icc_file);
+ exit(EXIT_FAILURE);
+ }
+ if (fread(icc_profile, icc_len, 1, icc_file) < 1) {
+ fprintf(stderr, "%s: can't read ICC profile from %s\n", progname,
+ icc_filename);
+ free(icc_profile);
+ fclose(icc_file);
+ exit(EXIT_FAILURE);
+ }
+ fclose(icc_file);
+ }
+
#ifdef PROGRESS_REPORT
start_progress_monitor((j_common_ptr) &cinfo, &progress);
#endif
@@ -611,6 +655,9 @@
/* Start compressor */
jpeg_start_compress(&cinfo, TRUE);
+ if (icc_profile != NULL)
+ jpeg_write_icc_profile(&cinfo, icc_profile, (unsigned int)icc_len);
+
/* Process data */
while (cinfo.next_scanline < cinfo.image_height) {
num_scanlines = (*src_mgr->get_pixel_rows) (&cinfo, src_mgr);
@@ -638,6 +685,9 @@
free(outbuffer);
}
+ if (icc_profile != NULL)
+ free(icc_profile);
+
/* All done. */
exit(jerr.num_warnings ? EXIT_WARNING : EXIT_SUCCESS);
return 0; /* suppress no-return-value warnings */
diff --git a/cmakescripts/BuildPackages.cmake b/cmakescripts/BuildPackages.cmake
new file mode 100644
index 0000000..5c03998
--- /dev/null
+++ b/cmakescripts/BuildPackages.cmake
@@ -0,0 +1,174 @@
+# This file is included from the top-level CMakeLists.txt. We just store it
+# here to avoid cluttering up that file.
+
+set(PKGNAME ${CMAKE_PROJECT_NAME} CACHE STRING
+ "Distribution package name (default: ${CMAKE_PROJECT_NAME})")
+set(PKGVENDOR "The ${CMAKE_PROJECT_NAME} Project" CACHE STRING
+ "Vendor name to be included in distribution package descriptions (default: The ${CMAKE_PROJECT_NAME} Project)")
+set(PKGURL "http://www.${CMAKE_PROJECT_NAME}.org" CACHE STRING
+ "URL of project web site to be included in distribution package descriptions (default: http://www.${CMAKE_PROJECT_NAME}.org)")
+set(PKGEMAIL "information@${CMAKE_PROJECT_NAME}.org" CACHE STRING
+ "E-mail of project maintainer to be included in distribution package descriptions (default: information@${CMAKE_PROJECT_NAME}.org")
+set(PKGID "com.${CMAKE_PROJECT_NAME}.${PKGNAME}" CACHE STRING
+ "Globally unique package identifier (reverse DNS notation) (default: com.${CMAKE_PROJECT_NAME}.${PKGNAME})")
+
+
+###############################################################################
+# Linux RPM and DEB
+###############################################################################
+
+if(CMAKE_SYSTEM_NAME STREQUAL "Linux")
+
+set(RPMARCH ${CMAKE_SYSTEM_PROCESSOR})
+if(CPU_TYPE STREQUAL "x86_64")
+ set(DEBARCH amd64)
+elseif(CMAKE_SYSTEM_PROCESSOR MATCHES "armv7*")
+ set(DEBARCH armhf)
+elseif(CPU_TYPE STREQUAL "arm64")
+ set(DEBARCH ${CPU_TYPE})
+elseif(CPU_TYPE STREQUAL "arm")
+ set(DEBARCH armel)
+elseif(CMAKE_SYSTEM_PROCESSOR_LC STREQUAL "ppc64le")
+ set(DEBARCH ppc64el)
+else()
+ set(DEBARCH ${CMAKE_SYSTEM_PROCESSOR})
+endif()
+message(STATUS "RPM architecture = ${RPMARCH}, DEB architecture = ${DEBARCH}")
+
+# Re-set CMAKE_POSITION_INDEPENDENT_CODE so that the RPM spec file works
+# properly
+boolean_number(CMAKE_POSITION_INDEPENDENT_CODE)
+
+configure_file(release/makerpm.in pkgscripts/makerpm)
+configure_file(release/rpm.spec.in pkgscripts/rpm.spec @ONLY)
+
+add_custom_target(rpm sh pkgscripts/makerpm
+ SOURCES pkgscripts/makerpm)
+
+configure_file(release/makesrpm.in pkgscripts/makesrpm)
+
+add_custom_target(srpm sh pkgscripts/makesrpm
+ SOURCES pkgscripts/makesrpm
+ DEPENDS dist)
+
+configure_file(release/makedpkg.in pkgscripts/makedpkg)
+configure_file(release/deb-control.in pkgscripts/deb-control)
+
+add_custom_target(deb sh pkgscripts/makedpkg
+ SOURCES pkgscripts/makedpkg)
+
+endif() # Linux
+
+
+###############################################################################
+# Windows installer (NullSoft Installer)
+###############################################################################
+
+if(WIN32)
+
+if(MSVC)
+ set(INST_PLATFORM "Visual C++")
+ set(INST_NAME ${CMAKE_PROJECT_NAME}-${VERSION}-vc)
+ set(INST_REG_NAME ${CMAKE_PROJECT_NAME})
+elseif(MINGW)
+ set(INST_PLATFORM GCC)
+ set(INST_NAME ${CMAKE_PROJECT_NAME}-${VERSION}-gcc)
+ set(INST_REG_NAME ${CMAKE_PROJECT_NAME}-gcc)
+ set(INST_DEFS -DGCC)
+endif()
+
+if(BITS EQUAL 64)
+ set(INST_PLATFORM "${INST_PLATFORM} 64-bit")
+ set(INST_NAME ${INST_NAME}64)
+ set(INST_REG_NAME ${INST_DIR}64)
+ set(INST_DEFS ${INST_DEFS} -DWIN64)
+endif()
+
+if(WITH_JAVA)
+ set(INST_DEFS ${INST_DEFS} -DJAVA)
+endif()
+
+if(MSVC_IDE)
+ set(INST_DEFS ${INST_DEFS} "-DBUILDDIR=${CMAKE_CFG_INTDIR}\\")
+else()
+ set(INST_DEFS ${INST_DEFS} "-DBUILDDIR=")
+endif()
+
+string(REGEX REPLACE "/" "\\\\" INST_DIR ${CMAKE_INSTALL_PREFIX})
+
+configure_file(release/installer.nsi.in installer.nsi @ONLY)
+
+if(WITH_JAVA)
+ set(JAVA_DEPEND turbojpeg-java)
+endif()
+add_custom_target(installer
+ makensis -nocd ${INST_DEFS} installer.nsi
+ DEPENDS jpeg jpeg-static turbojpeg turbojpeg-static rdjpgcom wrjpgcom
+ cjpeg djpeg jpegtran tjbench ${JAVA_DEPEND}
+ SOURCES installer.nsi)
+
+endif() # WIN32
+
+
+###############################################################################
+# Cygwin Package
+###############################################################################
+
+if(CYGWIN)
+
+configure_file(release/makecygwinpkg.in pkgscripts/makecygwinpkg)
+
+add_custom_target(cygwinpkg sh pkgscripts/makecygwinpkg)
+
+endif() # CYGWIN
+
+
+###############################################################################
+# Mac DMG
+###############################################################################
+
+if(APPLE)
+
+set(DEFAULT_OSX_32BIT_BUILD ${CMAKE_SOURCE_DIR}/osxx86)
+set(OSX_32BIT_BUILD ${DEFAULT_OSX_32BIT_BUILD} CACHE PATH
+ "Directory containing 32-bit (i386) Mac build to include in universal binaries (default: ${DEFAULT_OSX_32BIT_BUILD})")
+set(DEFAULT_IOS_ARMV7_BUILD ${CMAKE_SOURCE_DIR}/iosarmv7)
+set(IOS_ARMV7_BUILD ${DEFAULT_IOS_ARMV7_BUILD} CACHE PATH
+ "Directory containing ARMv7 iOS build to include in universal binaries (default: ${DEFAULT_IOS_ARMV7_BUILD})")
+set(DEFAULT_IOS_ARMV7S_BUILD ${CMAKE_SOURCE_DIR}/iosarmv7s)
+set(IOS_ARMV7S_BUILD ${DEFAULT_IOS_ARMV7S_BUILD} CACHE PATH
+ "Directory containing ARMv7s iOS build to include in universal binaries (default: ${DEFAULT_IOS_ARMV7S_BUILD})")
+set(DEFAULT_IOS_ARMV8_BUILD ${CMAKE_SOURCE_DIR}/iosarmv8)
+set(IOS_ARMV8_BUILD ${DEFAULT_IOS_ARMV8_BUILD} CACHE PATH
+ "Directory containing ARMv8 iOS build to include in universal binaries (default: ${DEFAULT_IOS_ARMV8_BUILD})")
+
+configure_file(release/makemacpkg.in pkgscripts/makemacpkg)
+configure_file(release/Distribution.xml.in pkgscripts/Distribution.xml)
+configure_file(release/uninstall.in pkgscripts/uninstall)
+
+add_custom_target(dmg sh pkgscripts/makemacpkg
+ SOURCES pkgscripts/makemacpkg)
+
+add_custom_target(udmg sh pkgscripts/makemacpkg universal
+ SOURCES pkgscripts/makemacpkg)
+
+endif() # APPLE
+
+
+###############################################################################
+# Generic
+###############################################################################
+
+add_custom_target(dist
+ COMMAND git archive --prefix=${CMAKE_PROJECT_NAME}-${VERSION}/ HEAD |
+ gzip > ${CMAKE_CURRENT_BINARY_DIR}/${CMAKE_PROJECT_NAME}-${VERSION}.tar.gz
+ WORKING_DIRECTORY ${CMAKE_CURRENT_SOURCE_DIR})
+
+configure_file(release/maketarball.in pkgscripts/maketarball)
+
+add_custom_target(tarball sh pkgscripts/maketarball
+ SOURCES pkgscripts/maketarball)
+
+configure_file(release/libjpeg.pc.in pkgscripts/libjpeg.pc @ONLY)
+
+configure_file(release/libturbojpeg.pc.in pkgscripts/libturbojpeg.pc @ONLY)
diff --git a/cmakescripts/GNUInstallDirs.cmake b/cmakescripts/GNUInstallDirs.cmake
new file mode 100644
index 0000000..ef564bb
--- /dev/null
+++ b/cmakescripts/GNUInstallDirs.cmake
@@ -0,0 +1,416 @@
+#.rst:
+# GNUInstallDirs
+# --------------
+#
+# Define GNU standard installation directories
+#
+# Provides install directory variables as defined by the
+# `GNU Coding Standards`_.
+#
+# .. _`GNU Coding Standards`: https://www.gnu.org/prep/standards/html_node/Directory-Variables.html
+#
+# Result Variables
+# ^^^^^^^^^^^^^^^^
+#
+# Inclusion of this module defines the following variables:
+#
+# ``CMAKE_INSTALL_<dir>``
+#
+# Destination for files of a given type. This value may be passed to
+# the ``DESTINATION`` options of :command:`install` commands for the
+# corresponding file type.
+#
+# ``CMAKE_INSTALL_FULL_<dir>``
+#
+# The absolute path generated from the corresponding ``CMAKE_INSTALL_<dir>``
+# value. If the value is not already an absolute path, an absolute path
+# is constructed typically by prepending the value of the
+# :variable:`CMAKE_INSTALL_PREFIX` variable. However, there are some
+# `special cases`_ as documented below.
+#
+# where ``<dir>`` is one of:
+#
+# ``BINDIR``
+# user executables (``bin``)
+# ``SBINDIR``
+# system admin executables (``sbin``)
+# ``LIBEXECDIR``
+# program executables (``libexec``)
+# ``SYSCONFDIR``
+# read-only single-machine data (``etc``)
+# ``SHAREDSTATEDIR``
+# modifiable architecture-independent data (``com``)
+# ``LOCALSTATEDIR``
+# modifiable single-machine data (``var``)
+# ``LIBDIR``
+# object code libraries (``lib`` or ``lib64``
+# or ``lib/<multiarch-tuple>`` on Debian)
+# ``INCLUDEDIR``
+# C header files (``include``)
+# ``OLDINCLUDEDIR``
+# C header files for non-gcc (``/usr/include``)
+# ``DATAROOTDIR``
+# read-only architecture-independent data root (``share``)
+# ``DATADIR``
+# read-only architecture-independent data (``DATAROOTDIR``)
+# ``INFODIR``
+# info documentation (``DATAROOTDIR/info``)
+# ``LOCALEDIR``
+# locale-dependent data (``DATAROOTDIR/locale``)
+# ``MANDIR``
+# man documentation (``DATAROOTDIR/man``)
+# ``DOCDIR``
+# documentation root (``DATAROOTDIR/doc/PROJECT_NAME``)
+#
+# If the includer does not define a value the above-shown default will be
+# used and the value will appear in the cache for editing by the user.
+#
+# Special Cases
+# ^^^^^^^^^^^^^
+#
+# The following values of :variable:`CMAKE_INSTALL_PREFIX` are special:
+#
+# ``/``
+#
+# For ``<dir>`` other than the ``SYSCONFDIR`` and ``LOCALSTATEDIR``,
+# the value of ``CMAKE_INSTALL_<dir>`` is prefixed with ``usr/`` if
+# it is not user-specified as an absolute path. For example, the
+# ``INCLUDEDIR`` value ``include`` becomes ``usr/include``.
+# This is required by the `GNU Coding Standards`_, which state:
+#
+# When building the complete GNU system, the prefix will be empty
+# and ``/usr`` will be a symbolic link to ``/``.
+#
+# ``/usr``
+#
+# For ``<dir>`` equal to ``SYSCONFDIR`` or ``LOCALSTATEDIR``, the
+# ``CMAKE_INSTALL_FULL_<dir>`` is computed by prepending just ``/``
+# to the value of ``CMAKE_INSTALL_<dir>`` if it is not user-specified
+# as an absolute path. For example, the ``SYSCONFDIR`` value ``etc``
+# becomes ``/etc``. This is required by the `GNU Coding Standards`_.
+#
+# ``/opt/...``
+#
+# For ``<dir>`` equal to ``SYSCONFDIR`` or ``LOCALSTATEDIR``, the
+# ``CMAKE_INSTALL_FULL_<dir>`` is computed by *appending* the prefix
+# to the value of ``CMAKE_INSTALL_<dir>`` if it is not user-specified
+# as an absolute path. For example, the ``SYSCONFDIR`` value ``etc``
+# becomes ``/etc/opt/...``. This is defined by the
+# `Filesystem Hierarchy Standard`_.
+#
+# .. _`Filesystem Hierarchy Standard`: https://refspecs.linuxfoundation.org/FHS_3.0/fhs/index.html
+#
+# Macros
+# ^^^^^^
+#
+# .. command:: GNUInstallDirs_get_absolute_install_dir
+#
+# ::
+#
+# GNUInstallDirs_get_absolute_install_dir(absvar var)
+#
+# Set the given variable ``absvar`` to the absolute path contained
+# within the variable ``var``. This is to allow the computation of an
+# absolute path, accounting for all the special cases documented
+# above. While this macro is used to compute the various
+# ``CMAKE_INSTALL_FULL_<dir>`` variables, it is exposed publicly to
+# allow users who create additional path variables to also compute
+# absolute paths where necessary, using the same logic.
+
+#=============================================================================
+# Copyright 2016 D. R. Commander
+# Copyright 2016 Dmitry Marakasov
+# Copyright 2016 Roger Leigh
+# Copyright 2015 Alex Turbov
+# Copyright 2014 Rolf Eike Beer
+# Copyright 2014 Daniele E. Domenichelli
+# Copyright 2013 Dimitri John Ledkov
+# Copyright 2011 Alex Neundorf
+# Copyright 2011 Eric NOULARD
+# Copyright 2011, 2013-2015 Kitware, Inc.
+# Copyright 2011 Nikita Krupen'ko
+#
+# Redistribution and use in source and binary forms, with or without
+# modification, are permitted provided that the following conditions
+# are met:
+#
+# * Redistributions of source code must retain the above copyright
+# notice, this list of conditions and the following disclaimer.
+#
+# * Redistributions in binary form must reproduce the above copyright
+# notice, this list of conditions and the following disclaimer in the
+# documentation and/or other materials provided with the distribution.
+#
+# * Neither the names of Kitware, Inc., the Insight Software Consortium,
+# nor the names of their contributors may be used to endorse or promote
+# products derived from this software without specific prior written
+# permission.
+#
+# THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
+# "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
+# LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR
+# A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT
+# HOLDER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
+# SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT
+# LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
+# DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
+# THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
+# (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
+# OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
+#=============================================================================
+
+# Installation directories
+#
+
+macro(GNUInstallDirs_set_install_dir var docstring)
+ # If CMAKE_INSTALL_PREFIX changes and CMAKE_INSTALL_*DIR is still set to the
+ # default value, then modify it accordingly. This presumes that the default
+ # value may change based on the prefix.
+
+ set(_GNUInstallDirs_CMAKE_INSTALL_FORCE_${var} "")
+ if(NOT DEFINED CMAKE_INSTALL_${var})
+ set(_GNUInstallDirs_CMAKE_INSTALL_DEFAULT_${var} 1 CACHE INTERNAL
+ "CMAKE_INSTALL_${var} has default value")
+ elseif(DEFINED _GNUInstallDirs_CMAKE_INSTALL_LAST_DEFAULT_${var} AND
+ NOT "${_GNUInstallDirs_CMAKE_INSTALL_LAST_DEFAULT_${var}}" STREQUAL
+ "${CMAKE_INSTALL_DEFAULT_${var}}" AND
+ _GNUInstallDirs_CMAKE_INSTALL_DEFAULT_${var} AND
+ "${_GNUInstallDirs_CMAKE_INSTALL_LAST_${var}}" STREQUAL
+ "${CMAKE_INSTALL_${var}}")
+ set(_GNUInstallDirs_CMAKE_INSTALL_FORCE_${var} "FORCE")
+ endif()
+
+ set(CMAKE_INSTALL_${var} "${CMAKE_INSTALL_DEFAULT_${var}}" CACHE PATH
+ "${docstring} (Default: ${CMAKE_INSTALL_DEFAULT_${var}})"
+ ${_GNUInstallDirs_CMAKE_INSTALL_FORCE_${var}})
+
+ if(NOT "${CMAKE_INSTALL_${var}}" STREQUAL "${CMAKE_INSTALL_DEFAULT_${var}}")
+ unset(_GNUInstallDirs_CMAKE_INSTALL_DEFAULT_${var} CACHE)
+ endif()
+
+ # Save for next run
+ set(_GNUInstallDirs_CMAKE_INSTALL_LAST_${var} "${CMAKE_INSTALL_${var}}"
+ CACHE INTERNAL "CMAKE_INSTALL_${var} during last run")
+ set(_GNUInstallDirs_CMAKE_INSTALL_LAST_DEFAULT_${var}
+ "${CMAKE_INSTALL_DEFAULT_${var}}" CACHE INTERNAL
+ "CMAKE_INSTALL_DEFAULT_${var} during last run")
+endmacro()
+
+if(NOT DEFINED CMAKE_INSTALL_DEFAULT_BINDIR)
+ set(CMAKE_INSTALL_DEFAULT_BINDIR "bin")
+endif()
+GNUInstallDirs_set_install_dir(BINDIR
+ "Directory into which user executables should be installed")
+
+if(NOT DEFINED CMAKE_INSTALL_DEFAULT_SBINDIR)
+ set(CMAKE_INSTALL_DEFAULT_SBINDIR "sbin")
+endif()
+GNUInstallDirs_set_install_dir(SBINDIR
+ "Directory into which system admin executables should be installed")
+
+if(NOT DEFINED CMAKE_INSTALL_DEFAULT_LIBEXECDIR)
+ set(CMAKE_INSTALL_DEFAULT_LIBEXECDIR "libexec")
+endif()
+GNUInstallDirs_set_install_dir(LIBEXECDIR
+ "Directory under which executables run by other programs should be installed")
+
+if(NOT DEFINED CMAKE_INSTALL_DEFAULT_SYSCONFDIR)
+ set(CMAKE_INSTALL_DEFAULT_SYSCONFDIR "etc")
+endif()
+GNUInstallDirs_set_install_dir(SYSCONFDIR
+ "Directory into which machine-specific read-only ASCII data and configuration files should be installed")
+
+if(NOT DEFINED CMAKE_INSTALL_DEFAULT_SHAREDSTATEDIR)
+ set(CMAKE_INSTALL_DEFAULT_SHAREDSTATEDIR "com")
+endif()
+GNUInstallDirs_set_install_dir(SHAREDSTATEDIR
+ "Directory into which architecture-independent run-time-modifiable data files should be installed")
+
+if(NOT DEFINED CMAKE_INSTALL_DEFAULT_LOCALSTATEDIR)
+ set(CMAKE_INSTALL_DEFAULT_LOCALSTATEDIR "var")
+endif()
+GNUInstallDirs_set_install_dir(LOCALSTATEDIR
+ "Directory into which machine-specific run-time-modifiable data files should be installed")
+
+if(NOT DEFINED CMAKE_INSTALL_DEFAULT_LIBDIR)
+ set(CMAKE_INSTALL_DEFAULT_LIBDIR "lib")
+ # Override this default 'lib' with 'lib64' iff:
+ # - we are on Linux system but NOT cross-compiling
+ # - we are NOT on debian
+ # - we are on a 64 bits system
+ # reason is: amd64 ABI: http://www.x86-64.org/documentation/abi.pdf
+ # For Debian with multiarch, use 'lib/${CMAKE_LIBRARY_ARCHITECTURE}' if
+ # CMAKE_LIBRARY_ARCHITECTURE is set (which contains e.g. "i386-linux-gnu"
+ # and CMAKE_INSTALL_PREFIX is "/usr"
+ # See http://wiki.debian.org/Multiarch
+ if(CMAKE_SYSTEM_NAME MATCHES "^(Linux|kFreeBSD|GNU)$"
+ AND NOT CMAKE_CROSSCOMPILING)
+ if (EXISTS "/etc/debian_version") # is this a debian system ?
+ if(CMAKE_LIBRARY_ARCHITECTURE)
+ if("${CMAKE_INSTALL_PREFIX}" MATCHES "^/usr/?$")
+ set(CMAKE_INSTALL_DEFAULT_LIBDIR "lib/${CMAKE_LIBRARY_ARCHITECTURE}")
+ endif()
+ endif()
+ else() # not debian, rely on CMAKE_SIZEOF_VOID_P:
+ if(NOT DEFINED CMAKE_SIZEOF_VOID_P)
+ message(AUTHOR_WARNING
+ "Unable to determine default CMAKE_INSTALL_LIBDIR directory because no target architecture is known. "
+ "Please enable at least one language before including GNUInstallDirs.")
+ else()
+ if("${CMAKE_SIZEOF_VOID_P}" EQUAL "8")
+ set(CMAKE_INSTALL_DEFAULT_LIBDIR "lib64")
+ endif()
+ endif()
+ endif()
+ endif()
+endif()
+GNUInstallDirs_set_install_dir(LIBDIR
+ "Directory into which object files and object code libraries should be installed")
+
+if(NOT DEFINED CMAKE_INSTALL_DEFAULT_INCLUDEDIR)
+ set(CMAKE_INSTALL_DEFAULT_INCLUDEDIR "include")
+endif()
+GNUInstallDirs_set_install_dir(INCLUDEDIR
+ "Directory into which C header files should be installed")
+
+if(NOT DEFINED CMAKE_INSTALL_DEFAULT_OLDINCLUDEDIR)
+ set(CMAKE_INSTALL_DEFAULT_OLDINCLUDEDIR "/usr/include")
+endif()
+GNUInstallDirs_set_install_dir(OLDINCLUDEDIR
+ PATH "Directory into which C header files for non-GCC compilers should be installed")
+
+if(NOT DEFINED CMAKE_INSTALL_DEFAULT_DATAROOTDIR)
+ set(CMAKE_INSTALL_DEFAULT_DATAROOTDIR "share")
+endif()
+GNUInstallDirs_set_install_dir(DATAROOTDIR
+ "The root of the directory tree for read-only architecture-independent data files")
+
+#-----------------------------------------------------------------------------
+# Values whose defaults are relative to DATAROOTDIR. Store empty values in
+# the cache and store the defaults in local variables if the cache values are
+# not set explicitly. This auto-updates the defaults as DATAROOTDIR changes.
+
+if(NOT DEFINED CMAKE_INSTALL_DEFAULT_DATADIR)
+ set(CMAKE_INSTALL_DEFAULT_DATADIR "<CMAKE_INSTALL_DATAROOTDIR>")
+endif()
+GNUInstallDirs_set_install_dir(DATADIR
+ "The directory under which read-only architecture-independent data files should be installed")
+
+if(NOT DEFINED CMAKE_INSTALL_DEFAULT_INFODIR)
+ if(CMAKE_SYSTEM_NAME MATCHES "^(.*BSD|DragonFly)$")
+ set(CMAKE_INSTALL_DEFAULT_INFODIR "info")
+ else()
+ set(CMAKE_INSTALL_DEFAULT_INFODIR "<CMAKE_INSTALL_DATAROOTDIR>/info")
+ endif()
+endif()
+GNUInstallDirs_set_install_dir(INFODIR
+ "The directory into which info documentation files should be installed")
+
+if(NOT DEFINED CMAKE_INSTALL_DEFAULT_MANDIR)
+ if(CMAKE_SYSTEM_NAME MATCHES "^(.*BSD|DragonFly)$")
+ set(CMAKE_INSTALL_DEFAULT_MANDIR "man")
+ else()
+ set(CMAKE_INSTALL_DEFAULT_MANDIR "<CMAKE_INSTALL_DATAROOTDIR>/man")
+ endif()
+endif()
+GNUInstallDirs_set_install_dir(MANDIR
+ "The directory under which man pages should be installed")
+
+if(NOT DEFINED CMAKE_INSTALL_DEFAULT_LOCALEDIR)
+ set(CMAKE_INSTALL_DEFAULT_LOCALEDIR "<CMAKE_INSTALL_DATAROOTDIR>/locale")
+endif()
+GNUInstallDirs_set_install_dir(LOCALEDIR
+ "The directory under which locale-specific message catalogs should be installed")
+
+if(NOT DEFINED CMAKE_INSTALL_DEFAULT_DOCDIR)
+ set(CMAKE_INSTALL_DEFAULT_DOCDIR "<CMAKE_INSTALL_DATAROOTDIR>/doc/${PROJECT_NAME}")
+endif()
+GNUInstallDirs_set_install_dir(DOCDIR
+ "The directory into which documentation files (other than info files) should be installed")
+
+#-----------------------------------------------------------------------------
+
+mark_as_advanced(
+ CMAKE_INSTALL_BINDIR
+ CMAKE_INSTALL_SBINDIR
+ CMAKE_INSTALL_LIBEXECDIR
+ CMAKE_INSTALL_SYSCONFDIR
+ CMAKE_INSTALL_SHAREDSTATEDIR
+ CMAKE_INSTALL_LOCALSTATEDIR
+ CMAKE_INSTALL_LIBDIR
+ CMAKE_INSTALL_INCLUDEDIR
+ CMAKE_INSTALL_OLDINCLUDEDIR
+ CMAKE_INSTALL_DATAROOTDIR
+ CMAKE_INSTALL_DATADIR
+ CMAKE_INSTALL_INFODIR
+ CMAKE_INSTALL_LOCALEDIR
+ CMAKE_INSTALL_MANDIR
+ CMAKE_INSTALL_DOCDIR
+ )
+
+macro(GNUInstallDirs_get_absolute_install_dir absvar var)
+ string(REGEX REPLACE "[<>]" "@" ${var} "${${var}}")
+ # Handle the specific case of an empty CMAKE_INSTALL_DATAROOTDIR
+ if(NOT CMAKE_INSTALL_DATAROOTDIR AND
+ ${var} MATCHES "\@CMAKE_INSTALL_DATAROOTDIR\@/")
+ string(CONFIGURE "${${var}}" ${var} @ONLY)
+ string(REGEX REPLACE "^/" "" ${var} "${${var}}")
+ else()
+ string(CONFIGURE "${${var}}" ${var} @ONLY)
+ endif()
+ if(NOT IS_ABSOLUTE "${${var}}")
+ # Handle special cases:
+ # - CMAKE_INSTALL_PREFIX == /
+ # - CMAKE_INSTALL_PREFIX == /usr
+ # - CMAKE_INSTALL_PREFIX == /opt/...
+ if("${CMAKE_INSTALL_PREFIX}" STREQUAL "/")
+ if("${dir}" STREQUAL "SYSCONFDIR" OR "${dir}" STREQUAL "LOCALSTATEDIR")
+ set(${absvar} "/${${var}}")
+ else()
+ if (NOT "${${var}}" MATCHES "^usr/")
+ set(${var} "usr/${${var}}")
+ endif()
+ set(${absvar} "/${${var}}")
+ endif()
+ elseif("${CMAKE_INSTALL_PREFIX}" MATCHES "^/usr/?$")
+ if("${dir}" STREQUAL "SYSCONFDIR" OR "${dir}" STREQUAL "LOCALSTATEDIR")
+ set(${absvar} "/${${var}}")
+ else()
+ set(${absvar} "${CMAKE_INSTALL_PREFIX}/${${var}}")
+ endif()
+ elseif("${CMAKE_INSTALL_PREFIX}" MATCHES "^/opt/.*")
+ if("${dir}" STREQUAL "SYSCONFDIR" OR "${dir}" STREQUAL "LOCALSTATEDIR")
+ set(${absvar} "/${${var}}${CMAKE_INSTALL_PREFIX}")
+ else()
+ set(${absvar} "${CMAKE_INSTALL_PREFIX}/${${var}}")
+ endif()
+ else()
+ set(${absvar} "${CMAKE_INSTALL_PREFIX}/${${var}}")
+ endif()
+ else()
+ set(${absvar} "${${var}}")
+ endif()
+ string(REGEX REPLACE "/$" "" ${absvar} "${${absvar}}")
+endmacro()
+
+# Result directories
+#
+foreach(dir
+ BINDIR
+ SBINDIR
+ LIBEXECDIR
+ SYSCONFDIR
+ SHAREDSTATEDIR
+ LOCALSTATEDIR
+ LIBDIR
+ INCLUDEDIR
+ OLDINCLUDEDIR
+ DATAROOTDIR
+ DATADIR
+ INFODIR
+ LOCALEDIR
+ MANDIR
+ DOCDIR
+ )
+ GNUInstallDirs_get_absolute_install_dir(CMAKE_INSTALL_FULL_${dir} CMAKE_INSTALL_${dir})
+endforeach()
diff --git a/cmakescripts/cmake_uninstall.cmake.in b/cmakescripts/cmake_uninstall.cmake.in
index b35d100..6726a0d 100644
--- a/cmakescripts/cmake_uninstall.cmake.in
+++ b/cmakescripts/cmake_uninstall.cmake.in
@@ -1,10 +1,10 @@
# This code is from the CMake FAQ
-if (NOT EXISTS "@CMAKE_CURRENT_BINARY_DIR@/install_manifest.txt")
- message(FATAL_ERROR "Cannot find install manifest: \"@CMAKE_CURRENT_BINARY_DIR@/install_manifest.txt\"")
-endif(NOT EXISTS "@CMAKE_CURRENT_BINARY_DIR@/install_manifest.txt")
+if (NOT EXISTS "@CMAKE_BINARY_DIR@/install_manifest.txt")
+ message(FATAL_ERROR "Cannot find install manifest: \"@CMAKE_BINARY_DIR@/install_manifest.txt\"")
+endif(NOT EXISTS "@CMAKE_BINARY_DIR@/install_manifest.txt")
-file(READ "@CMAKE_CURRENT_BINARY_DIR@/install_manifest.txt" files)
+file(READ "@CMAKE_BINARY_DIR@/install_manifest.txt" files)
string(REGEX REPLACE "\n" ";" files "${files}")
list(REVERSE files)
foreach (file ${files})
diff --git a/configure.ac b/configure.ac
deleted file mode 100644
index 3b0151c..0000000
--- a/configure.ac
+++ /dev/null
@@ -1,611 +0,0 @@
-# -*- Autoconf -*-
-# Process this file with autoconf to produce a configure script.
-
-AC_PREREQ([2.56])
-AC_INIT([libjpeg-turbo], [1.5.3])
-
-AM_INIT_AUTOMAKE([-Wall foreign dist-bzip2])
-AC_PREFIX_DEFAULT(/opt/libjpeg-turbo)
-
-m4_ifdef([AM_SILENT_RULES], [AM_SILENT_RULES([yes])])
-
-# Checks for programs.
-SAVED_CFLAGS=${CFLAGS}
-SAVED_CPPFLAGS=${CPPFLAGS}
-AC_PROG_CPP
-AC_PROG_CC
-m4_ifdef([AM_PROG_AR], [AM_PROG_AR])
-AM_PROG_AS
-AM_PROG_CC_C_O
-AC_PROG_INSTALL
-AC_PROG_LIBTOOL
-AC_PROG_LN_S
-
-AC_ARG_WITH([build-date], [Use custom build string to enable reproducible builds (default: YYMMDD)],
- [BUILD="$with_build_date"],
- [BUILD=`date +%Y%m%d`])
-
-PKG_PROG_PKG_CONFIG
-
-# When the prefix is /opt/libjpeg-turbo, we assume that an "official" binary is
-# being created, and thus we install things into specific locations.
-
-old_prefix=${prefix}
-if test "x$prefix" = "xNONE" -a "x$ac_default_prefix" != "x"; then
- prefix=$ac_default_prefix
-fi
-DATADIR=`eval echo ${datadir}`
-DATADIR=`eval echo $DATADIR`
-if test "$DATADIR" = "/opt/libjpeg-turbo/share"; then
- datadir='${prefix}'
-fi
-DATADIR=`eval echo ${datarootdir}`
-DATADIR=`eval echo $DATADIR`
-if test "$DATADIR" = "/opt/libjpeg-turbo/share"; then
- datarootdir='${prefix}'
-fi
-DOCDIR=`eval echo ${docdir}`
-DOCDIR=`eval echo $DOCDIR`
-if test "$DOCDIR" = "/opt/libjpeg-turbo/doc/libjpeg-turbo"; then
- docdir='${datadir}/doc'
-fi
-
-old_exec_prefix=${exec_prefix}
-if test "x$exec_prefix" = "xNONE"; then
- exec_prefix=${prefix}
-fi
-
-AC_CHECK_SIZEOF(size_t)
-
-if test "x${libdir}" = 'x${exec_prefix}/lib' -o "x${libdir}" = 'x${prefix}/lib'; then
- LIBDIR=`eval echo ${libdir}`
- LIBDIR=`eval echo $LIBDIR`
- if test "$LIBDIR" = "/opt/libjpeg-turbo/lib"; then
- case $host_os in
- darwin*)
- ;;
- *)
- if test "${ac_cv_sizeof_size_t}" = "8"; then
- libdir='${exec_prefix}/lib64'
- elif test "${ac_cv_sizeof_size_t}" = "4"; then
- libdir='${exec_prefix}/lib32'
- fi
- ;;
- esac
- fi
-fi
-exec_prefix=${old_exec_prefix}
-prefix=${old_prefix}
-
-# Check whether compiler supports pointers to undefined structures
-AC_MSG_CHECKING(whether compiler supports pointers to undefined structures)
-AC_TRY_COMPILE([ typedef struct undefined_structure *undef_struct_ptr; ], ,
- AC_MSG_RESULT(yes),
- [AC_MSG_RESULT(no)
- AC_DEFINE([INCOMPLETE_TYPES_BROKEN], [1],
- [Compiler does not support pointers to undefined structures.])])
-
-if test "x${GCC}" = "xyes"; then
- if test "x${SAVED_CFLAGS}" = "x"; then
- CFLAGS=-O3
- fi
- if test "x${SAVED_CPPFLAGS}" = "x"; then
- CPPFLAGS=-Wall
- fi
-fi
-
-AC_CHECK_DECL([__SUNPRO_C], [SUNCC="yes"], [SUNCC="no"])
-if test "x${SUNCC}" = "xyes"; then
- if test "x${SAVED_CFLAGS}" = "x"; then
- CFLAGS=-xO5
- fi
-fi
-
-# Checks for libraries.
-
-# Checks for header files.
-AC_HEADER_STDC
-AC_CHECK_HEADERS([stddef.h stdlib.h locale.h string.h])
-AC_CHECK_HEADER([sys/types.h],
- AC_DEFINE([NEED_SYS_TYPES_H], 1, [Define if you need to include <sys/types.h> to get size_t.]))
-
-# Checks for typedefs, structures, and compiler characteristics.
-AC_C_CONST
-AC_C_CHAR_UNSIGNED
-AC_C_INLINE
-AC_TYPE_SIZE_T
-AC_CHECK_TYPES([unsigned char, unsigned short])
-
-AC_MSG_CHECKING([if right shift is signed])
-AC_TRY_RUN(
- [#include <stdio.h>
- int is_shifting_signed (long arg) {
- long res = arg >> 4;
-
- if (res == -0x7F7E80CL)
- return 1; /* right shift is signed */
-
- /* see if unsigned-shift hack will fix it. */
- /* we can't just test exact value since it depends on width of long... */
- res |= (~0L) << (32-4);
- if (res == -0x7F7E80CL)
- return 0; /* right shift is unsigned */
-
- printf("Right shift isn't acting as I expect it to.\n");
- printf("I fear the JPEG software will not work at all.\n\n");
- return 0; /* try it with unsigned anyway */
- }
- int main (void) {
- exit(is_shifting_signed(-0x7F7E80B1L));
- }],
- [AC_MSG_RESULT(no)
- AC_DEFINE([RIGHT_SHIFT_IS_UNSIGNED], 1,
- [Define if your (broken) compiler shifts signed values as if they were unsigned.])],
- [AC_MSG_RESULT(yes)],
- [AC_MSG_RESULT(Assuming that right shift is signed on target machine.)])
-
-# Checks for library functions.
-AC_CHECK_FUNCS([memset memcpy], [],
- [AC_DEFINE([NEED_BSD_STRINGS], 1,
- [Define if you have BSD-like bzero and bcopy in <strings.h> rather than memset/memcpy in <string.h>.])])
-
-AC_MSG_CHECKING([libjpeg API version])
-AC_ARG_VAR(JPEG_LIB_VERSION, [libjpeg API version (62, 70, or 80)])
-if test "x$JPEG_LIB_VERSION" = "x"; then
- AC_ARG_WITH([jpeg7],
- AC_HELP_STRING([--with-jpeg7],
- [Emulate libjpeg v7 API/ABI (this makes libjpeg-turbo backward incompatible with libjpeg v6b.)]))
- AC_ARG_WITH([jpeg8],
- AC_HELP_STRING([--with-jpeg8],
- [Emulate libjpeg v8 API/ABI (this makes libjpeg-turbo backward incompatible with libjpeg v6b.)]))
- if test "x${with_jpeg8}" = "xyes"; then
- JPEG_LIB_VERSION=80
- else
- if test "x${with_jpeg7}" = "xyes"; then
- JPEG_LIB_VERSION=70
- else
- JPEG_LIB_VERSION=62
- fi
- fi
-fi
-JPEG_LIB_VERSION_DECIMAL=`expr $JPEG_LIB_VERSION / 10`.`expr $JPEG_LIB_VERSION % 10`
-AC_SUBST(JPEG_LIB_VERSION_DECIMAL)
-AC_MSG_RESULT([$JPEG_LIB_VERSION_DECIMAL])
-AC_DEFINE_UNQUOTED(JPEG_LIB_VERSION, [$JPEG_LIB_VERSION],
- [libjpeg API version])
-
-AC_ARG_VAR(SO_MAJOR_VERSION,
- [Major version of the libjpeg-turbo shared library (default is determined by the API version)])
-AC_ARG_VAR(SO_MINOR_VERSION,
- [Minor version of the libjpeg-turbo shared library (default is determined by the API version)])
-if test "x$SO_MAJOR_VERSION" = "x"; then
- case "$JPEG_LIB_VERSION" in
- 62) SO_MAJOR_VERSION=$JPEG_LIB_VERSION ;;
- *) SO_MAJOR_VERSION=`expr $JPEG_LIB_VERSION / 10` ;;
- esac
-fi
-if test "x$SO_MINOR_VERSION" = "x"; then
- case "$JPEG_LIB_VERSION" in
- 80) SO_MINOR_VERSION=2 ;;
- *) SO_MINOR_VERSION=0 ;;
- esac
-fi
-
-RPM_CONFIG_ARGS=
-
-# Memory source/destination managers
-SO_AGE=1
-MEM_SRCDST_FUNCTIONS=
-if test "x${with_jpeg8}" != "xyes"; then
- AC_MSG_CHECKING([whether to include in-memory source/destination managers])
- AC_ARG_WITH([mem-srcdst],
- AC_HELP_STRING([--without-mem-srcdst],
- [Do not include in-memory source/destination manager functions when emulating the libjpeg v6b or v7 API/ABI]))
- if test "x$with_mem_srcdst" != "xno"; then
- AC_MSG_RESULT(yes)
- AC_DEFINE([MEM_SRCDST_SUPPORTED], [1],
- [Support in-memory source/destination managers])
- SO_AGE=2
- MEM_SRCDST_FUNCTIONS="global: jpeg_mem_dest; jpeg_mem_src;";
- else
- AC_MSG_RESULT(no)
- RPM_CONFIG_ARGS="$RPM_CONFIG_ARGS --without-mem-srcdst"
- fi
-fi
-
-AC_MSG_CHECKING([libjpeg shared library version])
-AC_MSG_RESULT([$SO_MAJOR_VERSION.$SO_AGE.$SO_MINOR_VERSION])
-LIBTOOL_CURRENT=`expr $SO_MAJOR_VERSION + $SO_AGE`
-AC_SUBST(LIBTOOL_CURRENT)
-AC_SUBST(SO_MAJOR_VERSION)
-AC_SUBST(SO_MINOR_VERSION)
-AC_SUBST(SO_AGE)
-AC_SUBST(MEM_SRCDST_FUNCTIONS)
-
-AC_DEFINE_UNQUOTED(LIBJPEG_TURBO_VERSION, [$VERSION], [libjpeg-turbo version])
-
-m4_define(version_triplet,m4_split(AC_PACKAGE_VERSION,[[.]]))
-m4_define(version_major,m4_car(m4_shiftn(1,[],version_triplet)))
-m4_define(version_minor,m4_car(m4_shiftn(2,[],version_triplet)))
-m4_define(version_revision,m4_car(m4_shiftn(3,[],version_triplet)))
-VERSION_MAJOR=version_major
-VERSION_MINOR=version_minor
-VERSION_REVISION=version_revision
-LIBJPEG_TURBO_VERSION_NUMBER=`printf "%d%03d%03d" $VERSION_MAJOR $VERSION_MINOR $VERSION_REVISION`
-AC_DEFINE_UNQUOTED(LIBJPEG_TURBO_VERSION_NUMBER, [$LIBJPEG_TURBO_VERSION_NUMBER], [libjpeg-turbo version in integer form])
-
-VERSION_SCRIPT=yes
-AC_ARG_ENABLE([ld-version-script],
- AS_HELP_STRING([--disable-ld-version-script],
- [Disable linker version script for libjpeg-turbo (default is to use linker version script if the linker supports it)]),
- [VERSION_SCRIPT=$enableval], [])
-
-AC_MSG_CHECKING([whether the linker supports version scripts])
-SAVED_LDFLAGS="$LDFLAGS"
-LDFLAGS="$LDFLAGS -Wl,--version-script,conftest.map"
-cat > conftest.map <<EOF
-VERS_1 {
- global: *;
-};
-EOF
-AC_LINK_IFELSE([AC_LANG_PROGRAM([], [])],
- [VERSION_SCRIPT_FLAG=-Wl,--version-script,;
- AC_MSG_RESULT([yes (GNU style)])],
- [])
-if test "x$VERSION_SCRIPT_FLAG" = "x"; then
- LDFLAGS="$SAVED_LDFLAGS -Wl,-M,conftest.map"
- AC_LINK_IFELSE([AC_LANG_PROGRAM([], [])],
- [VERSION_SCRIPT_FLAG=-Wl,-M,;
- AC_MSG_RESULT([yes (Sun style)])],
- [])
-fi
-if test "x$VERSION_SCRIPT_FLAG" = "x"; then
- VERSION_SCRIPT=no
- AC_MSG_RESULT(no)
-fi
-LDFLAGS="$SAVED_LDFLAGS"
-
-AC_MSG_CHECKING([whether to use version script when building libjpeg-turbo])
-AC_MSG_RESULT($VERSION_SCRIPT)
-
-AM_CONDITIONAL(VERSION_SCRIPT, test "x$VERSION_SCRIPT" = "xyes")
-AC_SUBST(VERSION_SCRIPT_FLAG)
-
-# Check for non-broken inline under various spellings
-AC_MSG_CHECKING(for inline)
-ljt_cv_inline=""
-AC_TRY_COMPILE(, [} inline __attribute__((always_inline)) int foo() { return 0; }
-int bar() { return foo();], ljt_cv_inline="inline __attribute__((always_inline))",
-AC_TRY_COMPILE(, [} __inline__ int foo() { return 0; }
-int bar() { return foo();], ljt_cv_inline="__inline__",
-AC_TRY_COMPILE(, [} __inline int foo() { return 0; }
-int bar() { return foo();], ljt_cv_inline="__inline",
-AC_TRY_COMPILE(, [} inline int foo() { return 0; }
-int bar() { return foo();], ljt_cv_inline="inline"))))
-AC_MSG_RESULT($ljt_cv_inline)
-AC_DEFINE_UNQUOTED([INLINE],[$ljt_cv_inline],[How to obtain function inlining.])
-
-# Arithmetic coding support
-AC_MSG_CHECKING([whether to include arithmetic encoding support])
-AC_ARG_WITH([arith-enc],
- AC_HELP_STRING([--without-arith-enc],
- [Do not include arithmetic encoding support when emulating the libjpeg v6b API/ABI]))
-if test "x$with_12bit" = "xyes"; then
- with_arith_enc=no
-fi
-if test "x${with_jpeg8}" = "xyes" -o "x${with_jpeg7}" = "xyes"; then
- with_arith_enc=yes
-fi
-if test "x$with_arith_enc" = "xno"; then
- AC_MSG_RESULT(no)
- RPM_CONFIG_ARGS="$RPM_CONFIG_ARGS --without-arith-enc"
-else
- AC_DEFINE([C_ARITH_CODING_SUPPORTED], [1], [Support arithmetic encoding])
- AC_MSG_RESULT(yes)
-fi
-AM_CONDITIONAL([WITH_ARITH_ENC], [test "x$with_arith_enc" != "xno"])
-
-AC_MSG_CHECKING([whether to include arithmetic decoding support])
-AC_ARG_WITH([arith-dec],
- AC_HELP_STRING([--without-arith-dec],
- [Do not include arithmetic decoding support when emulating the libjpeg v6b API/ABI]))
-if test "x$with_12bit" = "xyes"; then
- with_arith_dec=no
-fi
-if test "x${with_jpeg8}" = "xyes" -o "x${with_jpeg7}" = "xyes"; then
- with_arith_dec=yes
-fi
-if test "x$with_arith_dec" = "xno"; then
- AC_MSG_RESULT(no)
- RPM_CONFIG_ARGS="$RPM_CONFIG_ARGS --without-arith-dec"
-else
- AC_DEFINE([D_ARITH_CODING_SUPPORTED], [1], [Support arithmetic decoding])
- AC_MSG_RESULT(yes)
-fi
-AM_CONDITIONAL([WITH_ARITH_DEC], [test "x$with_arith_dec" != "xno"])
-
-AM_CONDITIONAL([WITH_ARITH],
- [test "x$with_arith_dec" != "xno" -o "x$with_arith_enc" != "xno"])
-
-# 12-bit component support
-AC_MSG_CHECKING([whether to use 12-bit samples])
-AC_ARG_WITH([12bit],
- AC_HELP_STRING([--with-12bit], [Encode/decode JPEG images with 12-bit samples (implies --without-simd --without-turbojpeg --without-arith-dec --without-arith-enc)]))
-if test "x$with_12bit" = "xyes"; then
- AC_DEFINE([BITS_IN_JSAMPLE], [12], [use 8 or 12])
- AC_MSG_RESULT(yes)
-else
- AC_MSG_RESULT(no)
-fi
-AM_CONDITIONAL([WITH_12BIT], [test "x$with_12bit" = "xyes"])
-
-# TurboJPEG support
-AC_MSG_CHECKING([whether to build TurboJPEG C wrapper])
-AC_ARG_WITH([turbojpeg],
- AC_HELP_STRING([--without-turbojpeg],
- [Do not include the TurboJPEG wrapper library and associated test programs]))
-if test "x$with_12bit" = "xyes"; then
- with_turbojpeg=no
-fi
-if test "x$with_turbojpeg" = "xno"; then
- AC_MSG_RESULT(no)
- RPM_CONFIG_ARGS="$RPM_CONFIG_ARGS --without-turbojpeg"
-else
- AC_MSG_RESULT(yes)
-fi
-
-# Java support
-AC_ARG_VAR(JAVAC, [Java compiler command (default: javac)])
-if test "x$JAVAC" = "x"; then
- JAVAC=javac
-fi
-AC_SUBST(JAVAC)
-AC_ARG_VAR(JAVACFLAGS, [Java compiler flags])
-AC_SUBST(JAVACFLAGS)
-AC_ARG_VAR(JAR, [Java archive command (default: jar)])
-if test "x$JAR" = "x"; then
- JAR=jar
-fi
-AC_SUBST(JAR)
-AC_ARG_VAR(JAVA, [Java runtime command (default: java)])
-if test "x$JAVA" = "x"; then
- JAVA=java
-fi
-AC_SUBST(JAVA)
-AC_ARG_VAR(JNI_CFLAGS,
- [C compiler flags needed to include jni.h (default: -I/System/Library/Frameworks/JavaVM.framework/Headers on OS X, '-I/usr/java/include -I/usr/java/include/solaris' on Solaris, and '-I/usr/java/default/include -I/usr/java/default/include/linux' on Linux)])
-
-AC_MSG_CHECKING([whether to build TurboJPEG Java wrapper])
-AC_ARG_WITH([java],
- AC_HELP_STRING([--with-java], [Build Java wrapper for the TurboJPEG library]))
-if test "x$with_12bit" = "xyes" -o "x$with_turbojpeg" = "xno"; then
- with_java=no
-fi
-
-WITH_JAVA=0
-if test "x$with_java" = "xyes"; then
- AC_MSG_RESULT(yes)
-
- case $host_os in
- darwin*)
- DEFAULT_JNI_CFLAGS=-I/System/Library/Frameworks/JavaVM.framework/Headers
- ;;
- solaris*)
- DEFAULT_JNI_CFLAGS='-I/usr/java/include -I/usr/java/include/solaris'
- ;;
- linux*)
- DEFAULT_JNI_CFLAGS='-I/usr/java/default/include -I/usr/java/default/include/linux'
- ;;
- esac
- if test "x$JNI_CFLAGS" = "x"; then
- JNI_CFLAGS=$DEFAULT_JNI_CFLAGS
- fi
-
- SAVE_CPPFLAGS=${CPPFLAGS}
- CPPFLAGS="${CPPFLAGS} ${JNI_CFLAGS}"
- AC_CHECK_HEADERS([jni.h], [DUMMY=1],
- [AC_MSG_ERROR([Could not find JNI header file])])
- CPPFLAGS=${SAVE_CPPFLAGS}
- AC_SUBST(JNI_CFLAGS)
-
- RPM_CONFIG_ARGS="$RPM_CONFIG_ARGS --with-java"
- JAVA_RPM_CONTENTS_1='%dir %{_datadir}/classes'
- JAVA_RPM_CONTENTS_2=%{_datadir}/classes/turbojpeg.jar
- WITH_JAVA=1
-else
- AC_MSG_RESULT(no)
-fi
-AM_CONDITIONAL([WITH_JAVA], [test "x$with_java" = "xyes"])
-AC_SUBST(WITH_JAVA)
-AC_SUBST(JAVA_RPM_CONTENTS_1)
-AC_SUBST(JAVA_RPM_CONTENTS_2)
-
-# optionally force using gas-preprocessor.pl for compatibility testing
-AC_ARG_WITH([gas-preprocessor],
- AC_HELP_STRING([--with-gas-preprocessor],
- [Force using gas-preprocessor.pl on ARM.]))
-if test "x${with_gas_preprocessor}" = "xyes"; then
- case $host_os in
- darwin*)
- CCAS="gas-preprocessor.pl -fix-unreq $CC"
- ;;
- *)
- CCAS="gas-preprocessor.pl -no-fix-unreq $CC"
- ;;
- esac
- AC_SUBST([CCAS])
-fi
-
-# SIMD is optional
-AC_ARG_WITH([simd],
- AC_HELP_STRING([--without-simd], [Do not include SIMD extensions]))
-if test "x$with_12bit" = "xyes"; then
- with_simd=no
-fi
-if test "x${with_simd}" != "xno"; then
- require_simd=no
- if test "x${with_simd}" = "xyes"; then
- require_simd=yes
- fi
- # Check if we're on a supported CPU
- AC_MSG_CHECKING([if we have SIMD optimisations for cpu type])
- case "$host_cpu" in
- x86_64 | amd64)
- AC_MSG_RESULT([yes (x86_64)])
- AC_PROG_NASM
- simd_arch=x86_64
- ;;
- i*86 | x86 | ia32)
- AC_MSG_RESULT([yes (i386)])
- AC_PROG_NASM
- simd_arch=i386
- ;;
- arm*)
- AC_MSG_RESULT([yes (arm)])
- AC_MSG_CHECKING([if the assembler is GNU-compatible and can be used])
- AC_CHECK_COMPATIBLE_ARM_ASSEMBLER_IFELSE(
- [if test "x$ac_use_gas_preprocessor" = "xyes"; then
- AC_MSG_RESULT([yes (with gas-preprocessor)])
- else
- AC_MSG_RESULT([yes])
- fi
- simd_arch=arm],
- [AC_MSG_RESULT([no])
- with_simd=no])
- if test "x${with_simd}" = "xno"; then
- if test "x${require_simd}" = "xyes"; then
- AC_MSG_ERROR([SIMD support can't be enabled.])
- else
- AC_MSG_WARN([SIMD support can't be enabled. Performance will suffer.])
- fi
- fi
- ;;
- aarch64*)
- AC_MSG_RESULT([yes (arm64)])
- AC_MSG_CHECKING([if the assembler is GNU-compatible and can be used])
- AC_CHECK_COMPATIBLE_ARM64_ASSEMBLER_IFELSE(
- [if test "x$ac_use_gas_preprocessor" = "xyes"; then
- AC_MSG_RESULT([yes (with gas-preprocessor)])
- else
- AC_MSG_RESULT([yes])
- fi
- simd_arch=aarch64],
- [AC_MSG_RESULT([no])
- with_simd=no])
- if test "x${with_simd}" = "xno"; then
- if test "x${require_simd}" = "xyes"; then
- AC_MSG_ERROR([SIMD support can't be enabled.])
- else
- AC_MSG_WARN([SIMD support can't be enabled. Performance will suffer.])
- fi
- fi
- ;;
- mips*)
- AC_MSG_RESULT([yes (mips)])
- AC_MSG_CHECKING([if the assembler is GNU-compatible and can be used])
- AC_CHECK_COMPATIBLE_MIPS_ASSEMBLER_IFELSE(
- [AC_MSG_RESULT([yes])
- simd_arch=mips],
- [AC_MSG_RESULT([no])
- with_simd=no])
- if test "x${with_simd}" = "xno"; then
- if test "x${require_simd}" = "xyes"; then
- AC_MSG_ERROR([SIMD support can't be enabled.])
- else
- AC_MSG_WARN([SIMD support can't be enabled. Performance will suffer.])
- fi
- fi
- ;;
- powerpc*)
- AC_CHECK_ALTIVEC(
- [AC_MSG_RESULT([yes (powerpc)])
- simd_arch=powerpc],
- [AC_NO_SIMD(PowerPC SPE)])
- ;;
- *)
- AC_NO_SIMD($host_cpu)
- ;;
- esac
-
- if test "x${with_simd}" != "xno"; then
- AC_DEFINE([WITH_SIMD], [1], [Use accelerated SIMD routines.])
- fi
-else
- RPM_CONFIG_ARGS="$RPM_CONFIG_ARGS --without-simd"
-fi
-
-AM_CONDITIONAL([WITH_SIMD], [test "x$with_simd" != "xno"])
-AM_CONDITIONAL([WITH_SSE_FLOAT_DCT], [test "x$simd_arch" = "xx86_64" -o "x$simd_arch" = "xi386"])
-AM_CONDITIONAL([SIMD_I386], [test "x$simd_arch" = "xi386"])
-AM_CONDITIONAL([SIMD_X86_64], [test "x$simd_arch" = "xx86_64"])
-AM_CONDITIONAL([SIMD_ARM], [test "x$simd_arch" = "xarm"])
-AM_CONDITIONAL([SIMD_ARM_64], [test "x$simd_arch" = "xaarch64"])
-AM_CONDITIONAL([SIMD_MIPS], [test "x$simd_arch" = "xmips"])
-AM_CONDITIONAL([SIMD_POWERPC], [test "x$simd_arch" = "xpowerpc"])
-AM_CONDITIONAL([X86_64], [test "x$host_cpu" = "xx86_64" -o "x$host_cpu" = "xamd64"])
-AM_CONDITIONAL([WITH_TURBOJPEG], [test "x$with_turbojpeg" != "xno"])
-AM_CONDITIONAL([CROSS_COMPILING], [test "x$cross_compiling" = "xyes"])
-
-AC_ARG_VAR(PKGNAME, [distribution package name (default: libjpeg-turbo)])
-if test "x$PKGNAME" = "x"; then
- PKGNAME=$PACKAGE_NAME
-fi
-AC_SUBST(PKGNAME)
-
-case "$host_cpu" in
- x86_64)
- RPMARCH=x86_64
- DEBARCH=amd64
- ;;
- i*86 | x86 | ia32)
- RPMARCH=i386
- DEBARCH=i386
- ;;
- powerpc64le)
- RPMARCH=`uname -m`
- DEBARCH=ppc64el
- ;;
- *)
- RPMARCH=`uname -m`
- DEBARCH=$RPMARCH
- ;;
-esac
-
-if test "${docdir}" = ""; then
- docdir=${datadir}/doc
- AC_SUBST(docdir)
-fi
-
-AC_SUBST(RPMARCH)
-AC_SUBST(RPM_CONFIG_ARGS)
-AC_SUBST(DEBARCH)
-AC_SUBST(BUILD)
-AC_DEFINE_UNQUOTED([BUILD], "$BUILD", [libjpeg-turbo build number])
-
-# NOTE: autoheader automatically modifies the input file of the first
-# invocation of AC_CONFIG_HEADERS, so we put config.h first to prevent
-# jconfig.h.in from being clobbered. config.h is used only internally, whereas
-# jconfig.h contains macros that are relevant to external programs (macros that
-# specify which features were built into the library.)
-AC_CONFIG_HEADERS([config.h])
-AC_CONFIG_HEADERS([jconfig.h])
-AC_CONFIG_HEADERS([jconfigint.h])
-AC_CONFIG_FILES([pkgscripts/libjpeg-turbo.spec.tmpl:release/libjpeg-turbo.spec.in])
-AC_CONFIG_FILES([pkgscripts/makecygwinpkg.tmpl:release/makecygwinpkg.in])
-AC_CONFIG_FILES([pkgscripts/makedpkg.tmpl:release/makedpkg.in])
-AC_CONFIG_FILES([pkgscripts/makemacpkg.tmpl:release/makemacpkg.in])
-AC_CONFIG_FILES([pkgscripts/uninstall.tmpl:release/uninstall.in])
-AC_CONFIG_FILES([pkgscripts/libjpeg.pc:release/libjpeg.pc.in])
-AC_CONFIG_FILES([pkgscripts/libturbojpeg.pc:release/libturbojpeg.pc.in])
-if test "x$with_turbojpeg" != "xno"; then
- AC_CONFIG_FILES([tjbenchtest])
-fi
-if test "x$with_java" = "xyes"; then
- AC_CONFIG_FILES([tjbenchtest.java])
- AC_CONFIG_FILES([tjexampletest])
-fi
-AC_CONFIG_FILES([libjpeg.map])
-AC_CONFIG_FILES([Makefile simd/Makefile])
-AC_CONFIG_FILES([java/Makefile])
-AC_CONFIG_FILES([md5/Makefile])
-AC_OUTPUT
diff --git a/djpeg.1 b/djpeg.1
index b1e7f76..1b13b7e 100644
--- a/djpeg.1
+++ b/djpeg.1
@@ -1,4 +1,4 @@
-.TH DJPEG 1 "18 March 2017"
+TH DJPEG 1 "18 March 2017"
.SH NAME
djpeg \- decompress a JPEG file to an image file
.SH SYNOPSIS
@@ -157,6 +157,9 @@
.B \-onepass
mode.
.TP
+.BI \-icc " file"
+Extract ICC color management profile to the specified file.
+.TP
.BI \-map " file"
Quantize to the colors used in the specified image file. This is useful for
producing multiple files with identical color maps, or for forcing a
diff --git a/djpeg.c b/djpeg.c
index 54cd525..53a8009 100644
--- a/djpeg.c
+++ b/djpeg.c
@@ -5,7 +5,7 @@
* Copyright (C) 1991-1997, Thomas G. Lane.
* Modified 2013 by Guido Vollbeding.
* libjpeg-turbo Modifications:
- * Copyright (C) 2010-2011, 2013-2016, D. R. Commander.
+ * Copyright (C) 2010-2011, 2013-2017, D. R. Commander.
* Copyright (C) 2015, Google, Inc.
* For conditions of distribution and use, see the accompanying README.ijg
* file.
@@ -33,6 +33,10 @@
#include "jconfigint.h"
#include "wrppm.h"
+#ifndef HAVE_STDLIB_H /* <stdlib.h> should declare free() */
+extern void free (void *ptr);
+#endif
+
#include <ctype.h> /* to declare isprint() */
#ifdef USE_CCOMMAND /* command-line reader for Macintosh */
@@ -90,6 +94,7 @@
static const char *progname; /* program name for error messages */
+static char *icc_filename; /* for -icc switch */
static char *outfilename; /* for -outfile switch */
boolean memsrc; /* for -memsrc switch */
boolean skip, crop;
@@ -158,6 +163,7 @@
fprintf(stderr, " -dither fs Use F-S dithering (default)\n");
fprintf(stderr, " -dither none Don't use dithering in quantization\n");
fprintf(stderr, " -dither ordered Use ordered dither (medium speed, quality)\n");
+ fprintf(stderr, " -icc FILE Extract ICC profile to FILE\n");
#ifdef QUANT_2PASS_SUPPORTED
fprintf(stderr, " -map FILE Map to colors used in named image file\n");
#endif
@@ -196,6 +202,7 @@
/* Set up default JPEG parameters. */
requested_fmt = DEFAULT_FMT; /* set default output file format */
+ icc_filename = NULL;
outfilename = NULL;
memsrc = FALSE;
skip = FALSE;
@@ -303,6 +310,13 @@
/* Force RGB565 output. */
cinfo->out_color_space = JCS_RGB565;
+ } else if (keymatch(arg, "icc", 1)) {
+ /* Set ICC filename. */
+ if (++argn >= argc) /* advance to next argument */
+ usage();
+ icc_filename = argv[argn];
+ jpeg_save_markers(cinfo, JPEG_APP0 + 2, 0xFFFF);
+
} else if (keymatch(arg, "map", 3)) {
/* Quantize to a color map taken from an input file. */
if (++argn >= argc) /* advance to next argument */
@@ -754,6 +768,29 @@
progress.pub.completed_passes = progress.pub.total_passes;
#endif
+ if (icc_filename != NULL) {
+ FILE *icc_file;
+ JOCTET *icc_profile;
+ unsigned int icc_len;
+
+ if ((icc_file = fopen(icc_filename, WRITE_BINARY)) == NULL) {
+ fprintf(stderr, "%s: can't open %s\n", progname, icc_filename);
+ exit(EXIT_FAILURE);
+ }
+ if (jpeg_read_icc_profile(&cinfo, &icc_profile, &icc_len)) {
+ if (fwrite(icc_profile, icc_len, 1, icc_file) < 1) {
+ fprintf(stderr, "%s: can't read ICC profile from %s\n", progname,
+ icc_filename);
+ free(icc_profile);
+ fclose(icc_file);
+ exit(EXIT_FAILURE);
+ }
+ free(icc_profile);
+ fclose(icc_file);
+ } else if (cinfo.err->msg_code != JWRN_BOGUS_ICC)
+ fprintf(stderr, "%s: no ICC profile data in JPEG file\n", progname);
+ }
+
/* Finish decompression and release memory.
* I must do it in this order because output module has allocated memory
* of lifespan JPOOL_IMAGE; it needs to finish before releasing memory.
diff --git a/doc/html/annotated.html b/doc/html/annotated.html
index d0b0e1e..8172f55 100644
--- a/doc/html/annotated.html
+++ b/doc/html/annotated.html
@@ -24,7 +24,7 @@
<tr style="height: 56px;">
<td style="padding-left: 0.5em;">
<div id="projectname">TurboJPEG
-  <span id="projectnumber">1.5</span>
+  <span id="projectnumber">1.6</span>
</div>
</td>
</tr>
diff --git a/doc/html/classes.html b/doc/html/classes.html
index 275e96d..bc23e51 100644
--- a/doc/html/classes.html
+++ b/doc/html/classes.html
@@ -24,7 +24,7 @@
<tr style="height: 56px;">
<td style="padding-left: 0.5em;">
<div id="projectname">TurboJPEG
-  <span id="projectnumber">1.5</span>
+  <span id="projectnumber">1.6</span>
</div>
</td>
</tr>
diff --git a/doc/html/functions.html b/doc/html/functions.html
index 31d78f5..570debd 100644
--- a/doc/html/functions.html
+++ b/doc/html/functions.html
@@ -24,7 +24,7 @@
<tr style="height: 56px;">
<td style="padding-left: 0.5em;">
<div id="projectname">TurboJPEG
-  <span id="projectnumber">1.5</span>
+  <span id="projectnumber">1.6</span>
</div>
</td>
</tr>
diff --git a/doc/html/functions_vars.html b/doc/html/functions_vars.html
index 8373eac..7d45f87 100644
--- a/doc/html/functions_vars.html
+++ b/doc/html/functions_vars.html
@@ -24,7 +24,7 @@
<tr style="height: 56px;">
<td style="padding-left: 0.5em;">
<div id="projectname">TurboJPEG
-  <span id="projectnumber">1.5</span>
+  <span id="projectnumber">1.6</span>
</div>
</td>
</tr>
diff --git a/doc/html/group___turbo_j_p_e_g.html b/doc/html/group___turbo_j_p_e_g.html
index 8d7626d..cdaab54 100644
--- a/doc/html/group___turbo_j_p_e_g.html
+++ b/doc/html/group___turbo_j_p_e_g.html
@@ -24,7 +24,7 @@
<tr style="height: 56px;">
<td style="padding-left: 0.5em;">
<div id="projectname">TurboJPEG
-  <span id="projectnumber">1.5</span>
+  <span id="projectnumber">1.6</span>
</div>
</td>
</tr>
@@ -128,6 +128,15 @@
<tr class="memitem:gacb233cfd722d66d1ccbf48a7de81f0e0"><td class="memItemLeft" align="right" valign="top">#define </td><td class="memItemRight" valign="bottom"><a class="el" href="group___turbo_j_p_e_g.html#gacb233cfd722d66d1ccbf48a7de81f0e0">TJFLAG_ACCURATEDCT</a></td></tr>
<tr class="memdesc:gacb233cfd722d66d1ccbf48a7de81f0e0"><td class="mdescLeft"> </td><td class="mdescRight">Use the most accurate DCT/IDCT algorithm available in the underlying codec. <a href="#gacb233cfd722d66d1ccbf48a7de81f0e0">More...</a><br/></td></tr>
<tr class="separator:gacb233cfd722d66d1ccbf48a7de81f0e0"><td class="memSeparator" colspan="2"> </td></tr>
+<tr class="memitem:ga519cfa4ef6c18d9e5b455fdf59306a3a"><td class="memItemLeft" align="right" valign="top">#define </td><td class="memItemRight" valign="bottom"><a class="el" href="group___turbo_j_p_e_g.html#ga519cfa4ef6c18d9e5b455fdf59306a3a">TJFLAG_STOPONWARNING</a></td></tr>
+<tr class="memdesc:ga519cfa4ef6c18d9e5b455fdf59306a3a"><td class="mdescLeft"> </td><td class="mdescRight">Immediately discontinue the current compression/decompression/transform operation if the underlying codec throws a warning (non-fatal error). <a href="#ga519cfa4ef6c18d9e5b455fdf59306a3a">More...</a><br/></td></tr>
+<tr class="separator:ga519cfa4ef6c18d9e5b455fdf59306a3a"><td class="memSeparator" colspan="2"> </td></tr>
+<tr class="memitem:ga43b426750b46190a25d34a67ef76df1b"><td class="memItemLeft" align="right" valign="top">#define </td><td class="memItemRight" valign="bottom"><a class="el" href="group___turbo_j_p_e_g.html#ga43b426750b46190a25d34a67ef76df1b">TJFLAG_PROGRESSIVE</a></td></tr>
+<tr class="memdesc:ga43b426750b46190a25d34a67ef76df1b"><td class="mdescLeft"> </td><td class="mdescRight">Use progressive entropy coding in JPEG images generated by the compression and transform functions. <a href="#ga43b426750b46190a25d34a67ef76df1b">More...</a><br/></td></tr>
+<tr class="separator:ga43b426750b46190a25d34a67ef76df1b"><td class="memSeparator" colspan="2"> </td></tr>
+<tr class="memitem:ga79bde1b4a3e2351e00887e47781b966e"><td class="memItemLeft" align="right" valign="top">#define </td><td class="memItemRight" valign="bottom"><a class="el" href="group___turbo_j_p_e_g.html#ga79bde1b4a3e2351e00887e47781b966e">TJ_NUMERR</a></td></tr>
+<tr class="memdesc:ga79bde1b4a3e2351e00887e47781b966e"><td class="mdescLeft"> </td><td class="mdescRight">The number of error codes. <a href="#ga79bde1b4a3e2351e00887e47781b966e">More...</a><br/></td></tr>
+<tr class="separator:ga79bde1b4a3e2351e00887e47781b966e"><td class="memSeparator" colspan="2"> </td></tr>
<tr class="memitem:ga0f6dbd18adf38b7d46ac547f0f4d562c"><td class="memItemLeft" align="right" valign="top">#define </td><td class="memItemRight" valign="bottom"><a class="el" href="group___turbo_j_p_e_g.html#ga0f6dbd18adf38b7d46ac547f0f4d562c">TJ_NUMXOP</a></td></tr>
<tr class="memdesc:ga0f6dbd18adf38b7d46ac547f0f4d562c"><td class="mdescLeft"> </td><td class="mdescRight">The number of transform operations. <a href="#ga0f6dbd18adf38b7d46ac547f0f4d562c">More...</a><br/></td></tr>
<tr class="separator:ga0f6dbd18adf38b7d46ac547f0f4d562c"><td class="memSeparator" colspan="2"> </td></tr>
@@ -146,6 +155,12 @@
<tr class="memitem:gafbf992bbf6e006705886333703ffab31"><td class="memItemLeft" align="right" valign="top">#define </td><td class="memItemRight" valign="bottom"><a class="el" href="group___turbo_j_p_e_g.html#gafbf992bbf6e006705886333703ffab31">TJXOPT_NOOUTPUT</a></td></tr>
<tr class="memdesc:gafbf992bbf6e006705886333703ffab31"><td class="mdescLeft"> </td><td class="mdescRight">This option will prevent <a class="el" href="group___turbo_j_p_e_g.html#gad02cd42b69f193a0623a9c801788df3a" title="Losslessly transform a JPEG image into another JPEG image.">tjTransform()</a> from outputting a JPEG image for this particular transform (this can be used in conjunction with a custom filter to capture the transformed DCT coefficients without transcoding them.) <a href="#gafbf992bbf6e006705886333703ffab31">More...</a><br/></td></tr>
<tr class="separator:gafbf992bbf6e006705886333703ffab31"><td class="memSeparator" colspan="2"> </td></tr>
+<tr class="memitem:gad2371c80674584ecc1a7d75e564cf026"><td class="memItemLeft" align="right" valign="top">#define </td><td class="memItemRight" valign="bottom"><a class="el" href="group___turbo_j_p_e_g.html#gad2371c80674584ecc1a7d75e564cf026">TJXOPT_PROGRESSIVE</a></td></tr>
+<tr class="memdesc:gad2371c80674584ecc1a7d75e564cf026"><td class="mdescLeft"> </td><td class="mdescRight">This option will enable progressive entropy coding in the output image generated by this particular transform. <a href="#gad2371c80674584ecc1a7d75e564cf026">More...</a><br/></td></tr>
+<tr class="separator:gad2371c80674584ecc1a7d75e564cf026"><td class="memSeparator" colspan="2"> </td></tr>
+<tr class="memitem:ga153b468cfb905d0de61706c838986fe8"><td class="memItemLeft" align="right" valign="top">#define </td><td class="memItemRight" valign="bottom"><a class="el" href="group___turbo_j_p_e_g.html#ga153b468cfb905d0de61706c838986fe8">TJXOPT_COPYNONE</a></td></tr>
+<tr class="memdesc:ga153b468cfb905d0de61706c838986fe8"><td class="mdescLeft"> </td><td class="mdescRight">This option will prevent <a class="el" href="group___turbo_j_p_e_g.html#gad02cd42b69f193a0623a9c801788df3a" title="Losslessly transform a JPEG image into another JPEG image.">tjTransform()</a> from copying any extra markers (including EXIF and ICC profile data) from the source image to the output image. <a href="#ga153b468cfb905d0de61706c838986fe8">More...</a><br/></td></tr>
+<tr class="separator:ga153b468cfb905d0de61706c838986fe8"><td class="memSeparator" colspan="2"> </td></tr>
<tr class="memitem:ga0aba955473315e405295d978f0c16511"><td class="memItemLeft" align="right" valign="top">#define </td><td class="memItemRight" valign="bottom"><a class="el" href="group___turbo_j_p_e_g.html#ga0aba955473315e405295d978f0c16511">TJPAD</a>(width)</td></tr>
<tr class="memdesc:ga0aba955473315e405295d978f0c16511"><td class="mdescLeft"> </td><td class="mdescRight">Pad the given width to the nearest 32-bit boundary. <a href="#ga0aba955473315e405295d978f0c16511">More...</a><br/></td></tr>
<tr class="separator:ga0aba955473315e405295d978f0c16511"><td class="memSeparator" colspan="2"> </td></tr>
@@ -206,6 +221,11 @@
}</td></tr>
<tr class="memdesc:ga4f83ad3368e0e29d1957be0efa7c3720"><td class="mdescLeft"> </td><td class="mdescRight">JPEG colorspaces. <a href="group___turbo_j_p_e_g.html#ga4f83ad3368e0e29d1957be0efa7c3720">More...</a><br/></td></tr>
<tr class="separator:ga4f83ad3368e0e29d1957be0efa7c3720"><td class="memSeparator" colspan="2"> </td></tr>
+<tr class="memitem:gafbc17cfa57d0d5d11fea35ac025950fe"><td class="memItemLeft" align="right" valign="top">enum  </td><td class="memItemRight" valign="bottom"><a class="el" href="group___turbo_j_p_e_g.html#gafbc17cfa57d0d5d11fea35ac025950fe">TJERR</a> { <a class="el" href="group___turbo_j_p_e_g.html#ggafbc17cfa57d0d5d11fea35ac025950fea342dd6e2aedb47bb257b4e7568329b59">TJERR_WARNING</a>,
+<a class="el" href="group___turbo_j_p_e_g.html#ggafbc17cfa57d0d5d11fea35ac025950feafc9cceeada13122b09e4851e3788039a">TJERR_FATAL</a>
+ }</td></tr>
+<tr class="memdesc:gafbc17cfa57d0d5d11fea35ac025950fe"><td class="mdescLeft"> </td><td class="mdescRight">Error codes. <a href="group___turbo_j_p_e_g.html#gafbc17cfa57d0d5d11fea35ac025950fe">More...</a><br/></td></tr>
+<tr class="separator:gafbc17cfa57d0d5d11fea35ac025950fe"><td class="memSeparator" colspan="2"> </td></tr>
<tr class="memitem:ga2de531af4e7e6c4f124908376b354866"><td class="memItemLeft" align="right" valign="top">enum  </td><td class="memItemRight" valign="bottom"><a class="el" href="group___turbo_j_p_e_g.html#ga2de531af4e7e6c4f124908376b354866">TJXOP</a> { <br/>
  <a class="el" href="group___turbo_j_p_e_g.html#gga2de531af4e7e6c4f124908376b354866aad88c0366cd3f7d0eac9d7a3fa1c2c27">TJXOP_NONE</a>,
<a class="el" href="group___turbo_j_p_e_g.html#gga2de531af4e7e6c4f124908376b354866aa0df69776caa30f0fa28e26332d311ce">TJXOP_HFLIP</a>,
@@ -295,9 +315,12 @@
<tr class="memitem:ga8c4a1231dc06a450514c835f6471f137"><td class="memItemLeft" align="right" valign="top">DLLEXPORT void DLLCALL </td><td class="memItemRight" valign="bottom"><a class="el" href="group___turbo_j_p_e_g.html#ga8c4a1231dc06a450514c835f6471f137">tjFree</a> (unsigned char *buffer)</td></tr>
<tr class="memdesc:ga8c4a1231dc06a450514c835f6471f137"><td class="mdescLeft"> </td><td class="mdescRight">Free an image buffer previously allocated by TurboJPEG. <a href="#ga8c4a1231dc06a450514c835f6471f137">More...</a><br/></td></tr>
<tr class="separator:ga8c4a1231dc06a450514c835f6471f137"><td class="memSeparator" colspan="2"> </td></tr>
-<tr class="memitem:ga9af79c908ec131b1ae8d52fe40375abf"><td class="memItemLeft" align="right" valign="top">DLLEXPORT char *DLLCALL </td><td class="memItemRight" valign="bottom"><a class="el" href="group___turbo_j_p_e_g.html#ga9af79c908ec131b1ae8d52fe40375abf">tjGetErrorStr</a> (void)</td></tr>
-<tr class="memdesc:ga9af79c908ec131b1ae8d52fe40375abf"><td class="mdescLeft"> </td><td class="mdescRight">Returns a descriptive error message explaining why the last command failed. <a href="#ga9af79c908ec131b1ae8d52fe40375abf">More...</a><br/></td></tr>
-<tr class="separator:ga9af79c908ec131b1ae8d52fe40375abf"><td class="memSeparator" colspan="2"> </td></tr>
+<tr class="memitem:ga94a235bd4f1088f61ad87b4eadb64c9c"><td class="memItemLeft" align="right" valign="top">DLLEXPORT char *DLLCALL </td><td class="memItemRight" valign="bottom"><a class="el" href="group___turbo_j_p_e_g.html#ga94a235bd4f1088f61ad87b4eadb64c9c">tjGetErrorStr2</a> (<a class="el" href="group___turbo_j_p_e_g.html#ga758d2634ecb4949de7815cba621f5763">tjhandle</a> handle)</td></tr>
+<tr class="memdesc:ga94a235bd4f1088f61ad87b4eadb64c9c"><td class="mdescLeft"> </td><td class="mdescRight">Returns a descriptive error message explaining why the last command failed. <a href="#ga94a235bd4f1088f61ad87b4eadb64c9c">More...</a><br/></td></tr>
+<tr class="separator:ga94a235bd4f1088f61ad87b4eadb64c9c"><td class="memSeparator" colspan="2"> </td></tr>
+<tr class="memitem:ga0be00a62bd1be897f170fa1fed5fb4cb"><td class="memItemLeft" align="right" valign="top">DLLEXPORT int DLLCALL </td><td class="memItemRight" valign="bottom"><a class="el" href="group___turbo_j_p_e_g.html#ga0be00a62bd1be897f170fa1fed5fb4cb">tjGetErrorCode</a> (<a class="el" href="group___turbo_j_p_e_g.html#ga758d2634ecb4949de7815cba621f5763">tjhandle</a> handle)</td></tr>
+<tr class="memdesc:ga0be00a62bd1be897f170fa1fed5fb4cb"><td class="mdescLeft"> </td><td class="mdescRight">Returns a code indicating the severity of the last error. <a href="#ga0be00a62bd1be897f170fa1fed5fb4cb">More...</a><br/></td></tr>
+<tr class="separator:ga0be00a62bd1be897f170fa1fed5fb4cb"><td class="memSeparator" colspan="2"> </td></tr>
</table><table class="memberdecls">
<tr class="heading"><td colspan="2"><h2 class="groupheader"><a name="var-members"></a>
Variables</h2></td></tr>
@@ -343,6 +366,20 @@
</div>
</div>
+<a class="anchor" id="ga79bde1b4a3e2351e00887e47781b966e"></a>
+<div class="memitem">
+<div class="memproto">
+ <table class="memname">
+ <tr>
+ <td class="memname">#define TJ_NUMERR</td>
+ </tr>
+ </table>
+</div><div class="memdoc">
+
+<p>The number of error codes. </p>
+
+</div>
+</div>
<a class="anchor" id="ga7010a4402f54a45ba822ad8675a4655e"></a>
<div class="memitem">
<div class="memproto">
@@ -459,6 +496,36 @@
</div>
</div>
+<a class="anchor" id="ga43b426750b46190a25d34a67ef76df1b"></a>
+<div class="memitem">
+<div class="memproto">
+ <table class="memname">
+ <tr>
+ <td class="memname">#define TJFLAG_PROGRESSIVE</td>
+ </tr>
+ </table>
+</div><div class="memdoc">
+
+<p>Use progressive entropy coding in JPEG images generated by the compression and transform functions. </p>
+<p>Progressive entropy coding will generally improve compression relative to baseline entropy coding (the default), but it will reduce compression and decompression performance considerably. </p>
+
+</div>
+</div>
+<a class="anchor" id="ga519cfa4ef6c18d9e5b455fdf59306a3a"></a>
+<div class="memitem">
+<div class="memproto">
+ <table class="memname">
+ <tr>
+ <td class="memname">#define TJFLAG_STOPONWARNING</td>
+ </tr>
+ </table>
+</div><div class="memdoc">
+
+<p>Immediately discontinue the current compression/decompression/transform operation if the underlying codec throws a warning (non-fatal error). </p>
+<p>The default behavior is to allow the operation to complete unless a fatal error is encountered. </p>
+
+</div>
+</div>
<a class="anchor" id="ga0aba955473315e405295d978f0c16511"></a>
<div class="memitem">
<div class="memproto">
@@ -506,6 +573,20 @@
</div>
</div>
+<a class="anchor" id="ga153b468cfb905d0de61706c838986fe8"></a>
+<div class="memitem">
+<div class="memproto">
+ <table class="memname">
+ <tr>
+ <td class="memname">#define TJXOPT_COPYNONE</td>
+ </tr>
+ </table>
+</div><div class="memdoc">
+
+<p>This option will prevent <a class="el" href="group___turbo_j_p_e_g.html#gad02cd42b69f193a0623a9c801788df3a" title="Losslessly transform a JPEG image into another JPEG image.">tjTransform()</a> from copying any extra markers (including EXIF and ICC profile data) from the source image to the output image. </p>
+
+</div>
+</div>
<a class="anchor" id="ga9c771a757fc1294add611906b89ab2d2"></a>
<div class="memitem">
<div class="memproto">
@@ -564,6 +645,21 @@
</div>
</div>
+<a class="anchor" id="gad2371c80674584ecc1a7d75e564cf026"></a>
+<div class="memitem">
+<div class="memproto">
+ <table class="memname">
+ <tr>
+ <td class="memname">#define TJXOPT_PROGRESSIVE</td>
+ </tr>
+ </table>
+</div><div class="memdoc">
+
+<p>This option will enable progressive entropy coding in the output image generated by this particular transform. </p>
+<p>Progressive entropy coding will generally improve compression relative to baseline entropy coding (the default), but it will reduce compression and decompression performance considerably. </p>
+
+</div>
+</div>
<a class="anchor" id="ga319826b7eb1583c0595bbe7b95428709"></a>
<div class="memitem">
<div class="memproto">
@@ -644,6 +740,28 @@
</div>
</div>
+<a class="anchor" id="gafbc17cfa57d0d5d11fea35ac025950fe"></a>
+<div class="memitem">
+<div class="memproto">
+ <table class="memname">
+ <tr>
+ <td class="memname">enum <a class="el" href="group___turbo_j_p_e_g.html#gafbc17cfa57d0d5d11fea35ac025950fe">TJERR</a></td>
+ </tr>
+ </table>
+</div><div class="memdoc">
+
+<p>Error codes. </p>
+<table class="fieldtable">
+<tr><th colspan="2">Enumerator</th></tr><tr><td class="fieldname"><em><a class="anchor" id="ggafbc17cfa57d0d5d11fea35ac025950fea342dd6e2aedb47bb257b4e7568329b59"></a>TJERR_WARNING</em> </td><td class="fielddoc">
+<p>The error was non-fatal and recoverable, but the image may still be corrupt. </p>
+</td></tr>
+<tr><td class="fieldname"><em><a class="anchor" id="ggafbc17cfa57d0d5d11fea35ac025950feafc9cceeada13122b09e4851e3788039a"></a>TJERR_FATAL</em> </td><td class="fielddoc">
+<p>The error was fatal and non-recoverable. </p>
+</td></tr>
+</table>
+
+</div>
+</div>
<a class="anchor" id="gac916144e26c3817ac514e64ae5d12e2a"></a>
<div class="memitem">
<div class="memproto">
@@ -1018,7 +1136,7 @@
</table>
</dd>
</dl>
-<dl class="section return"><dt>Returns</dt><dd>0 if successful, or -1 if an error occurred (see <a class="el" href="group___turbo_j_p_e_g.html#ga9af79c908ec131b1ae8d52fe40375abf" title="Returns a descriptive error message explaining why the last command failed.">tjGetErrorStr()</a>.) </dd></dl>
+<dl class="section return"><dt>Returns</dt><dd>0 if successful, or -1 if an error occurred (see <a class="el" href="group___turbo_j_p_e_g.html#ga94a235bd4f1088f61ad87b4eadb64c9c" title="Returns a descriptive error message explaining why the last command failed.">tjGetErrorStr2()</a> and <a class="el" href="group___turbo_j_p_e_g.html#ga0be00a62bd1be897f170fa1fed5fb4cb" title="Returns a code indicating the severity of the last error.">tjGetErrorCode()</a>.) </dd></dl>
</div>
</div>
@@ -1115,7 +1233,7 @@
</table>
</dd>
</dl>
-<dl class="section return"><dt>Returns</dt><dd>0 if successful, or -1 if an error occurred (see <a class="el" href="group___turbo_j_p_e_g.html#ga9af79c908ec131b1ae8d52fe40375abf" title="Returns a descriptive error message explaining why the last command failed.">tjGetErrorStr()</a>.) </dd></dl>
+<dl class="section return"><dt>Returns</dt><dd>0 if successful, or -1 if an error occurred (see <a class="el" href="group___turbo_j_p_e_g.html#ga94a235bd4f1088f61ad87b4eadb64c9c" title="Returns a descriptive error message explaining why the last command failed.">tjGetErrorStr2()</a> and <a class="el" href="group___turbo_j_p_e_g.html#ga0be00a62bd1be897f170fa1fed5fb4cb" title="Returns a code indicating the severity of the last error.">tjGetErrorCode()</a>.) </dd></dl>
</div>
</div>
@@ -1212,7 +1330,7 @@
</table>
</dd>
</dl>
-<dl class="section return"><dt>Returns</dt><dd>0 if successful, or -1 if an error occurred (see <a class="el" href="group___turbo_j_p_e_g.html#ga9af79c908ec131b1ae8d52fe40375abf" title="Returns a descriptive error message explaining why the last command failed.">tjGetErrorStr()</a>.) </dd></dl>
+<dl class="section return"><dt>Returns</dt><dd>0 if successful, or -1 if an error occurred (see <a class="el" href="group___turbo_j_p_e_g.html#ga94a235bd4f1088f61ad87b4eadb64c9c" title="Returns a descriptive error message explaining why the last command failed.">tjGetErrorStr2()</a> and <a class="el" href="group___turbo_j_p_e_g.html#ga0be00a62bd1be897f170fa1fed5fb4cb" title="Returns a code indicating the severity of the last error.">tjGetErrorCode()</a>.) </dd></dl>
</div>
</div>
@@ -1305,7 +1423,7 @@
</table>
</dd>
</dl>
-<dl class="section return"><dt>Returns</dt><dd>0 if successful, or -1 if an error occurred (see <a class="el" href="group___turbo_j_p_e_g.html#ga9af79c908ec131b1ae8d52fe40375abf" title="Returns a descriptive error message explaining why the last command failed.">tjGetErrorStr()</a>.) </dd></dl>
+<dl class="section return"><dt>Returns</dt><dd>0 if successful, or -1 if an error occurred (see <a class="el" href="group___turbo_j_p_e_g.html#ga94a235bd4f1088f61ad87b4eadb64c9c" title="Returns a descriptive error message explaining why the last command failed.">tjGetErrorStr2()</a> and <a class="el" href="group___turbo_j_p_e_g.html#ga0be00a62bd1be897f170fa1fed5fb4cb" title="Returns a code indicating the severity of the last error.">tjGetErrorCode()</a>.) </dd></dl>
</div>
</div>
@@ -1398,7 +1516,7 @@
</table>
</dd>
</dl>
-<dl class="section return"><dt>Returns</dt><dd>0 if successful, or -1 if an error occurred (see <a class="el" href="group___turbo_j_p_e_g.html#ga9af79c908ec131b1ae8d52fe40375abf" title="Returns a descriptive error message explaining why the last command failed.">tjGetErrorStr()</a>.) </dd></dl>
+<dl class="section return"><dt>Returns</dt><dd>0 if successful, or -1 if an error occurred (see <a class="el" href="group___turbo_j_p_e_g.html#ga94a235bd4f1088f61ad87b4eadb64c9c" title="Returns a descriptive error message explaining why the last command failed.">tjGetErrorStr2()</a> and <a class="el" href="group___turbo_j_p_e_g.html#ga0be00a62bd1be897f170fa1fed5fb4cb" title="Returns a code indicating the severity of the last error.">tjGetErrorCode()</a>.) </dd></dl>
</div>
</div>
@@ -1483,7 +1601,7 @@
</table>
</dd>
</dl>
-<dl class="section return"><dt>Returns</dt><dd>0 if successful, or -1 if an error occurred (see <a class="el" href="group___turbo_j_p_e_g.html#ga9af79c908ec131b1ae8d52fe40375abf" title="Returns a descriptive error message explaining why the last command failed.">tjGetErrorStr()</a>.) </dd></dl>
+<dl class="section return"><dt>Returns</dt><dd>0 if successful, or -1 if an error occurred (see <a class="el" href="group___turbo_j_p_e_g.html#ga94a235bd4f1088f61ad87b4eadb64c9c" title="Returns a descriptive error message explaining why the last command failed.">tjGetErrorStr2()</a> and <a class="el" href="group___turbo_j_p_e_g.html#ga0be00a62bd1be897f170fa1fed5fb4cb" title="Returns a code indicating the severity of the last error.">tjGetErrorCode()</a>.) </dd></dl>
</div>
</div>
@@ -1554,7 +1672,7 @@
</table>
</dd>
</dl>
-<dl class="section return"><dt>Returns</dt><dd>0 if successful, or -1 if an error occurred (see <a class="el" href="group___turbo_j_p_e_g.html#ga9af79c908ec131b1ae8d52fe40375abf" title="Returns a descriptive error message explaining why the last command failed.">tjGetErrorStr()</a>.) </dd></dl>
+<dl class="section return"><dt>Returns</dt><dd>0 if successful, or -1 if an error occurred (see <a class="el" href="group___turbo_j_p_e_g.html#ga94a235bd4f1088f61ad87b4eadb64c9c" title="Returns a descriptive error message explaining why the last command failed.">tjGetErrorStr2()</a> and <a class="el" href="group___turbo_j_p_e_g.html#ga0be00a62bd1be897f170fa1fed5fb4cb" title="Returns a code indicating the severity of the last error.">tjGetErrorCode()</a>.) </dd></dl>
</div>
</div>
@@ -1633,7 +1751,7 @@
</table>
</dd>
</dl>
-<dl class="section return"><dt>Returns</dt><dd>0 if successful, or -1 if an error occurred (see <a class="el" href="group___turbo_j_p_e_g.html#ga9af79c908ec131b1ae8d52fe40375abf" title="Returns a descriptive error message explaining why the last command failed.">tjGetErrorStr()</a>.) </dd></dl>
+<dl class="section return"><dt>Returns</dt><dd>0 if successful, or -1 if an error occurred (see <a class="el" href="group___turbo_j_p_e_g.html#ga94a235bd4f1088f61ad87b4eadb64c9c" title="Returns a descriptive error message explaining why the last command failed.">tjGetErrorStr2()</a> and <a class="el" href="group___turbo_j_p_e_g.html#ga0be00a62bd1be897f170fa1fed5fb4cb" title="Returns a code indicating the severity of the last error.">tjGetErrorCode()</a>.) </dd></dl>
</div>
</div>
@@ -1712,7 +1830,7 @@
</table>
</dd>
</dl>
-<dl class="section return"><dt>Returns</dt><dd>0 if successful, or -1 if an error occurred (see <a class="el" href="group___turbo_j_p_e_g.html#ga9af79c908ec131b1ae8d52fe40375abf" title="Returns a descriptive error message explaining why the last command failed.">tjGetErrorStr()</a>.) </dd></dl>
+<dl class="section return"><dt>Returns</dt><dd>0 if successful, or -1 if an error occurred (see <a class="el" href="group___turbo_j_p_e_g.html#ga94a235bd4f1088f61ad87b4eadb64c9c" title="Returns a descriptive error message explaining why the last command failed.">tjGetErrorStr2()</a> and <a class="el" href="group___turbo_j_p_e_g.html#ga0be00a62bd1be897f170fa1fed5fb4cb" title="Returns a code indicating the severity of the last error.">tjGetErrorCode()</a>.) </dd></dl>
</div>
</div>
@@ -1737,7 +1855,7 @@
</table>
</dd>
</dl>
-<dl class="section return"><dt>Returns</dt><dd>0 if successful, or -1 if an error occurred (see <a class="el" href="group___turbo_j_p_e_g.html#ga9af79c908ec131b1ae8d52fe40375abf" title="Returns a descriptive error message explaining why the last command failed.">tjGetErrorStr()</a>.) </dd></dl>
+<dl class="section return"><dt>Returns</dt><dd>0 if successful, or -1 if an error occurred (see <a class="el" href="group___turbo_j_p_e_g.html#ga94a235bd4f1088f61ad87b4eadb64c9c" title="Returns a descriptive error message explaining why the last command failed.">tjGetErrorStr2()</a>.) </dd></dl>
</div>
</div>
@@ -1830,7 +1948,7 @@
</table>
</dd>
</dl>
-<dl class="section return"><dt>Returns</dt><dd>0 if successful, or -1 if an error occurred (see <a class="el" href="group___turbo_j_p_e_g.html#ga9af79c908ec131b1ae8d52fe40375abf" title="Returns a descriptive error message explaining why the last command failed.">tjGetErrorStr()</a>.) </dd></dl>
+<dl class="section return"><dt>Returns</dt><dd>0 if successful, or -1 if an error occurred (see <a class="el" href="group___turbo_j_p_e_g.html#ga94a235bd4f1088f61ad87b4eadb64c9c" title="Returns a descriptive error message explaining why the last command failed.">tjGetErrorStr2()</a> and <a class="el" href="group___turbo_j_p_e_g.html#ga0be00a62bd1be897f170fa1fed5fb4cb" title="Returns a code indicating the severity of the last error.">tjGetErrorCode()</a>.) </dd></dl>
</div>
</div>
@@ -1923,7 +2041,7 @@
</table>
</dd>
</dl>
-<dl class="section return"><dt>Returns</dt><dd>0 if successful, or -1 if an error occurred (see <a class="el" href="group___turbo_j_p_e_g.html#ga9af79c908ec131b1ae8d52fe40375abf" title="Returns a descriptive error message explaining why the last command failed.">tjGetErrorStr()</a>.) </dd></dl>
+<dl class="section return"><dt>Returns</dt><dd>0 if successful, or -1 if an error occurred (see <a class="el" href="group___turbo_j_p_e_g.html#ga94a235bd4f1088f61ad87b4eadb64c9c" title="Returns a descriptive error message explaining why the last command failed.">tjGetErrorStr2()</a> and <a class="el" href="group___turbo_j_p_e_g.html#ga0be00a62bd1be897f170fa1fed5fb4cb" title="Returns a code indicating the severity of the last error.">tjGetErrorCode()</a>.) </dd></dl>
</div>
</div>
@@ -1953,21 +2071,53 @@
</div>
</div>
-<a class="anchor" id="ga9af79c908ec131b1ae8d52fe40375abf"></a>
+<a class="anchor" id="ga0be00a62bd1be897f170fa1fed5fb4cb"></a>
<div class="memitem">
<div class="memproto">
<table class="memname">
<tr>
- <td class="memname">DLLEXPORT char* DLLCALL tjGetErrorStr </td>
+ <td class="memname">DLLEXPORT int DLLCALL tjGetErrorCode </td>
<td>(</td>
- <td class="paramtype">void </td>
- <td class="paramname"></td><td>)</td>
+ <td class="paramtype"><a class="el" href="group___turbo_j_p_e_g.html#ga758d2634ecb4949de7815cba621f5763">tjhandle</a> </td>
+ <td class="paramname"><em>handle</em></td><td>)</td>
+ <td></td>
+ </tr>
+ </table>
+</div><div class="memdoc">
+
+<p>Returns a code indicating the severity of the last error. </p>
+<p>See <a class="el" href="group___turbo_j_p_e_g.html#gafbc17cfa57d0d5d11fea35ac025950fe">Error codes</a>.</p>
+<dl class="params"><dt>Parameters</dt><dd>
+ <table class="params">
+ <tr><td class="paramname">handle</td><td>a handle to a TurboJPEG compressor, decompressor or transformer instance</td></tr>
+ </table>
+ </dd>
+</dl>
+<dl class="section return"><dt>Returns</dt><dd>a code indicating the severity of the last error. See <a class="el" href="group___turbo_j_p_e_g.html#gafbc17cfa57d0d5d11fea35ac025950fe">Error codes</a>. </dd></dl>
+
+</div>
+</div>
+<a class="anchor" id="ga94a235bd4f1088f61ad87b4eadb64c9c"></a>
+<div class="memitem">
+<div class="memproto">
+ <table class="memname">
+ <tr>
+ <td class="memname">DLLEXPORT char* DLLCALL tjGetErrorStr2 </td>
+ <td>(</td>
+ <td class="paramtype"><a class="el" href="group___turbo_j_p_e_g.html#ga758d2634ecb4949de7815cba621f5763">tjhandle</a> </td>
+ <td class="paramname"><em>handle</em></td><td>)</td>
<td></td>
</tr>
</table>
</div><div class="memdoc">
<p>Returns a descriptive error message explaining why the last command failed. </p>
+<dl class="params"><dt>Parameters</dt><dd>
+ <table class="params">
+ <tr><td class="paramname">handle</td><td>a handle to a TurboJPEG compressor, decompressor, or transformer instance, or NULL if the error was generated by a global function (but note that retrieving the error message for a global function is not thread-safe.)</td></tr>
+ </table>
+ </dd>
+</dl>
<dl class="section return"><dt>Returns</dt><dd>a descriptive error message explaining why the last command failed. </dd></dl>
</div>
@@ -1993,7 +2143,7 @@
</table>
</dd>
</dl>
-<dl class="section return"><dt>Returns</dt><dd>a pointer to a list of fractional scaling factors, or NULL if an error is encountered (see <a class="el" href="group___turbo_j_p_e_g.html#ga9af79c908ec131b1ae8d52fe40375abf" title="Returns a descriptive error message explaining why the last command failed.">tjGetErrorStr()</a>.) </dd></dl>
+<dl class="section return"><dt>Returns</dt><dd>a pointer to a list of fractional scaling factors, or NULL if an error is encountered (see <a class="el" href="group___turbo_j_p_e_g.html#ga94a235bd4f1088f61ad87b4eadb64c9c" title="Returns a descriptive error message explaining why the last command failed.">tjGetErrorStr2()</a>.) </dd></dl>
</div>
</div>
@@ -2012,7 +2162,7 @@
</div><div class="memdoc">
<p>Create a TurboJPEG compressor instance. </p>
-<dl class="section return"><dt>Returns</dt><dd>a handle to the newly-created instance, or NULL if an error occurred (see <a class="el" href="group___turbo_j_p_e_g.html#ga9af79c908ec131b1ae8d52fe40375abf" title="Returns a descriptive error message explaining why the last command failed.">tjGetErrorStr()</a>.) </dd></dl>
+<dl class="section return"><dt>Returns</dt><dd>a handle to the newly-created instance, or NULL if an error occurred (see <a class="el" href="group___turbo_j_p_e_g.html#ga94a235bd4f1088f61ad87b4eadb64c9c" title="Returns a descriptive error message explaining why the last command failed.">tjGetErrorStr2()</a>.) </dd></dl>
</div>
</div>
@@ -2031,7 +2181,7 @@
</div><div class="memdoc">
<p>Create a TurboJPEG decompressor instance. </p>
-<dl class="section return"><dt>Returns</dt><dd>a handle to the newly-created instance, or NULL if an error occurred (see <a class="el" href="group___turbo_j_p_e_g.html#ga9af79c908ec131b1ae8d52fe40375abf" title="Returns a descriptive error message explaining why the last command failed.">tjGetErrorStr()</a>.) </dd></dl>
+<dl class="section return"><dt>Returns</dt><dd>a handle to the newly-created instance, or NULL if an error occurred (see <a class="el" href="group___turbo_j_p_e_g.html#ga94a235bd4f1088f61ad87b4eadb64c9c" title="Returns a descriptive error message explaining why the last command failed.">tjGetErrorStr2()</a>.) </dd></dl>
</div>
</div>
@@ -2050,7 +2200,7 @@
</div><div class="memdoc">
<p>Create a new TurboJPEG transformer instance. </p>
-<dl class="section return"><dt>Returns</dt><dd>a handle to the newly-created instance, or NULL if an error occurred (see <a class="el" href="group___turbo_j_p_e_g.html#ga9af79c908ec131b1ae8d52fe40375abf" title="Returns a descriptive error message explaining why the last command failed.">tjGetErrorStr()</a>.) </dd></dl>
+<dl class="section return"><dt>Returns</dt><dd>a handle to the newly-created instance, or NULL if an error occurred (see <a class="el" href="group___turbo_j_p_e_g.html#ga94a235bd4f1088f61ad87b4eadb64c9c" title="Returns a descriptive error message explaining why the last command failed.">tjGetErrorStr2()</a>.) </dd></dl>
</div>
</div>
@@ -2279,7 +2429,7 @@
</table>
</dd>
</dl>
-<dl class="section return"><dt>Returns</dt><dd>0 if successful, or -1 if an error occurred (see <a class="el" href="group___turbo_j_p_e_g.html#ga9af79c908ec131b1ae8d52fe40375abf" title="Returns a descriptive error message explaining why the last command failed.">tjGetErrorStr()</a>.) </dd></dl>
+<dl class="section return"><dt>Returns</dt><dd>0 if successful, or -1 if an error occurred (see <a class="el" href="group___turbo_j_p_e_g.html#ga94a235bd4f1088f61ad87b4eadb64c9c" title="Returns a descriptive error message explaining why the last command failed.">tjGetErrorStr2()</a> and <a class="el" href="group___turbo_j_p_e_g.html#ga0be00a62bd1be897f170fa1fed5fb4cb" title="Returns a code indicating the severity of the last error.">tjGetErrorCode()</a>.) </dd></dl>
</div>
</div>
diff --git a/doc/html/index.html b/doc/html/index.html
index 3cc1b3e..6b27f31 100644
--- a/doc/html/index.html
+++ b/doc/html/index.html
@@ -24,7 +24,7 @@
<tr style="height: 56px;">
<td style="padding-left: 0.5em;">
<div id="projectname">TurboJPEG
-  <span id="projectnumber">1.5</span>
+  <span id="projectnumber">1.6</span>
</div>
</td>
</tr>
diff --git a/doc/html/modules.html b/doc/html/modules.html
index 8e6f815..8b38151 100644
--- a/doc/html/modules.html
+++ b/doc/html/modules.html
@@ -24,7 +24,7 @@
<tr style="height: 56px;">
<td style="padding-left: 0.5em;">
<div id="projectname">TurboJPEG
-  <span id="projectnumber">1.5</span>
+  <span id="projectnumber">1.6</span>
</div>
</td>
</tr>
diff --git a/doc/html/search/all_74.js b/doc/html/search/all_74.js
index 444aaef..a211e3d 100644
--- a/doc/html/search/all_74.js
+++ b/doc/html/search/all_74.js
@@ -1,6 +1,7 @@
var searchData=
[
['tj_5fnumcs',['TJ_NUMCS',['../group___turbo_j_p_e_g.html#ga39f57a6fb02d9cf32e7b6890099b5a71',1,'turbojpeg.h']]],
+ ['tj_5fnumerr',['TJ_NUMERR',['../group___turbo_j_p_e_g.html#ga79bde1b4a3e2351e00887e47781b966e',1,'turbojpeg.h']]],
['tj_5fnumpf',['TJ_NUMPF',['../group___turbo_j_p_e_g.html#ga7010a4402f54a45ba822ad8675a4655e',1,'turbojpeg.h']]],
['tj_5fnumsamp',['TJ_NUMSAMP',['../group___turbo_j_p_e_g.html#ga5ef3d169162ce77ce348e292a0b7477c',1,'turbojpeg.h']]],
['tj_5fnumxop',['TJ_NUMXOP',['../group___turbo_j_p_e_g.html#ga0f6dbd18adf38b7d46ac547f0f4d562c',1,'turbojpeg.h']]],
@@ -26,13 +27,19 @@
['tjdestroy',['tjDestroy',['../group___turbo_j_p_e_g.html#ga674adee917b95ad4a896f1ba39e12540',1,'turbojpeg.h']]],
['tjencodeyuv3',['tjEncodeYUV3',['../group___turbo_j_p_e_g.html#gaabe05acd734990053ad1294b5ef239aa',1,'turbojpeg.h']]],
['tjencodeyuvplanes',['tjEncodeYUVPlanes',['../group___turbo_j_p_e_g.html#ga8a65ed3bd12df57c219d46afbc9008f1',1,'turbojpeg.h']]],
+ ['tjerr',['TJERR',['../group___turbo_j_p_e_g.html#gafbc17cfa57d0d5d11fea35ac025950fe',1,'turbojpeg.h']]],
+ ['tjerr_5ffatal',['TJERR_FATAL',['../group___turbo_j_p_e_g.html#ggafbc17cfa57d0d5d11fea35ac025950feafc9cceeada13122b09e4851e3788039a',1,'turbojpeg.h']]],
+ ['tjerr_5fwarning',['TJERR_WARNING',['../group___turbo_j_p_e_g.html#ggafbc17cfa57d0d5d11fea35ac025950fea342dd6e2aedb47bb257b4e7568329b59',1,'turbojpeg.h']]],
['tjflag_5faccuratedct',['TJFLAG_ACCURATEDCT',['../group___turbo_j_p_e_g.html#gacb233cfd722d66d1ccbf48a7de81f0e0',1,'turbojpeg.h']]],
['tjflag_5fbottomup',['TJFLAG_BOTTOMUP',['../group___turbo_j_p_e_g.html#ga72ecf4ebe6eb702d3c6f5ca27455e1ec',1,'turbojpeg.h']]],
['tjflag_5ffastdct',['TJFLAG_FASTDCT',['../group___turbo_j_p_e_g.html#gaabce235db80d3f698b27f36cbd453da2',1,'turbojpeg.h']]],
['tjflag_5ffastupsample',['TJFLAG_FASTUPSAMPLE',['../group___turbo_j_p_e_g.html#ga4ee4506c81177a06f77e2504a22efd2d',1,'turbojpeg.h']]],
['tjflag_5fnorealloc',['TJFLAG_NOREALLOC',['../group___turbo_j_p_e_g.html#ga8808d403c68b62aaa58a4c1e58e98963',1,'turbojpeg.h']]],
+ ['tjflag_5fprogressive',['TJFLAG_PROGRESSIVE',['../group___turbo_j_p_e_g.html#ga43b426750b46190a25d34a67ef76df1b',1,'turbojpeg.h']]],
+ ['tjflag_5fstoponwarning',['TJFLAG_STOPONWARNING',['../group___turbo_j_p_e_g.html#ga519cfa4ef6c18d9e5b455fdf59306a3a',1,'turbojpeg.h']]],
['tjfree',['tjFree',['../group___turbo_j_p_e_g.html#ga8c4a1231dc06a450514c835f6471f137',1,'turbojpeg.h']]],
- ['tjgeterrorstr',['tjGetErrorStr',['../group___turbo_j_p_e_g.html#ga9af79c908ec131b1ae8d52fe40375abf',1,'turbojpeg.h']]],
+ ['tjgeterrorcode',['tjGetErrorCode',['../group___turbo_j_p_e_g.html#ga0be00a62bd1be897f170fa1fed5fb4cb',1,'turbojpeg.h']]],
+ ['tjgeterrorstr2',['tjGetErrorStr2',['../group___turbo_j_p_e_g.html#ga94a235bd4f1088f61ad87b4eadb64c9c',1,'turbojpeg.h']]],
['tjgetscalingfactors',['tjGetScalingFactors',['../group___turbo_j_p_e_g.html#ga6449044b9af402999ccf52f401333be8',1,'turbojpeg.h']]],
['tjgreenoffset',['tjGreenOffset',['../group___turbo_j_p_e_g.html#ga82d6e35da441112a411da41923c0ba2f',1,'turbojpeg.h']]],
['tjhandle',['tjhandle',['../group___turbo_j_p_e_g.html#ga758d2634ecb4949de7815cba621f5763',1,'turbojpeg.h']]],
@@ -70,7 +77,7 @@
['tjsamp_5fgray',['TJSAMP_GRAY',['../group___turbo_j_p_e_g.html#gga1d047060ea80bb9820d540bb928e9074a3f1c9504842ddc7a48d0f690754b6248',1,'turbojpeg.h']]],
['tjscaled',['TJSCALED',['../group___turbo_j_p_e_g.html#ga84878bb65404204743aa18cac02781df',1,'turbojpeg.h']]],
['tjscalingfactor',['tjscalingfactor',['../structtjscalingfactor.html',1,'']]],
- ['tjtransform',['tjtransform',['../structtjtransform.html',1,'tjtransform'],['../group___turbo_j_p_e_g.html#gad02cd42b69f193a0623a9c801788df3a',1,'tjTransform(tjhandle handle, const unsigned char *jpegBuf, unsigned long jpegSize, int n, unsigned char **dstBufs, unsigned long *dstSizes, tjtransform *transforms, int flags): turbojpeg.h'],['../group___turbo_j_p_e_g.html#gaa29f3189c41be12ec5dee7caec318a31',1,'tjtransform(): turbojpeg.h']]],
+ ['tjtransform',['tjtransform',['../structtjtransform.html',1,'tjtransform'],['../group___turbo_j_p_e_g.html#gaa29f3189c41be12ec5dee7caec318a31',1,'tjtransform(): turbojpeg.h'],['../group___turbo_j_p_e_g.html#gad02cd42b69f193a0623a9c801788df3a',1,'tjTransform(tjhandle handle, const unsigned char *jpegBuf, unsigned long jpegSize, int n, unsigned char **dstBufs, unsigned long *dstSizes, tjtransform *transforms, int flags): turbojpeg.h']]],
['tjxop',['TJXOP',['../group___turbo_j_p_e_g.html#ga2de531af4e7e6c4f124908376b354866',1,'turbojpeg.h']]],
['tjxop_5fhflip',['TJXOP_HFLIP',['../group___turbo_j_p_e_g.html#gga2de531af4e7e6c4f124908376b354866aa0df69776caa30f0fa28e26332d311ce',1,'turbojpeg.h']]],
['tjxop_5fnone',['TJXOP_NONE',['../group___turbo_j_p_e_g.html#gga2de531af4e7e6c4f124908376b354866aad88c0366cd3f7d0eac9d7a3fa1c2c27',1,'turbojpeg.h']]],
@@ -80,10 +87,12 @@
['tjxop_5ftranspose',['TJXOP_TRANSPOSE',['../group___turbo_j_p_e_g.html#gga2de531af4e7e6c4f124908376b354866a31060aed199f886afdd417f80499c32d',1,'turbojpeg.h']]],
['tjxop_5ftransverse',['TJXOP_TRANSVERSE',['../group___turbo_j_p_e_g.html#gga2de531af4e7e6c4f124908376b354866af3b14d488aea6ece9e5b3df73a74d6a4',1,'turbojpeg.h']]],
['tjxop_5fvflip',['TJXOP_VFLIP',['../group___turbo_j_p_e_g.html#gga2de531af4e7e6c4f124908376b354866a324eddfbec53b7e691f61e56929d0d5d',1,'turbojpeg.h']]],
+ ['tjxopt_5fcopynone',['TJXOPT_COPYNONE',['../group___turbo_j_p_e_g.html#ga153b468cfb905d0de61706c838986fe8',1,'turbojpeg.h']]],
['tjxopt_5fcrop',['TJXOPT_CROP',['../group___turbo_j_p_e_g.html#ga9c771a757fc1294add611906b89ab2d2',1,'turbojpeg.h']]],
['tjxopt_5fgray',['TJXOPT_GRAY',['../group___turbo_j_p_e_g.html#ga3acee7b48ade1b99e5588736007c2589',1,'turbojpeg.h']]],
['tjxopt_5fnooutput',['TJXOPT_NOOUTPUT',['../group___turbo_j_p_e_g.html#gafbf992bbf6e006705886333703ffab31',1,'turbojpeg.h']]],
['tjxopt_5fperfect',['TJXOPT_PERFECT',['../group___turbo_j_p_e_g.html#ga50e03cb5ed115330e212417429600b00',1,'turbojpeg.h']]],
+ ['tjxopt_5fprogressive',['TJXOPT_PROGRESSIVE',['../group___turbo_j_p_e_g.html#gad2371c80674584ecc1a7d75e564cf026',1,'turbojpeg.h']]],
['tjxopt_5ftrim',['TJXOPT_TRIM',['../group___turbo_j_p_e_g.html#ga319826b7eb1583c0595bbe7b95428709',1,'turbojpeg.h']]],
['turbojpeg',['TurboJPEG',['../group___turbo_j_p_e_g.html',1,'']]]
];
diff --git a/doc/html/search/enums_74.js b/doc/html/search/enums_74.js
index 276aa24..19c20cf 100644
--- a/doc/html/search/enums_74.js
+++ b/doc/html/search/enums_74.js
@@ -1,6 +1,7 @@
var searchData=
[
['tjcs',['TJCS',['../group___turbo_j_p_e_g.html#ga4f83ad3368e0e29d1957be0efa7c3720',1,'turbojpeg.h']]],
+ ['tjerr',['TJERR',['../group___turbo_j_p_e_g.html#gafbc17cfa57d0d5d11fea35ac025950fe',1,'turbojpeg.h']]],
['tjpf',['TJPF',['../group___turbo_j_p_e_g.html#gac916144e26c3817ac514e64ae5d12e2a',1,'turbojpeg.h']]],
['tjsamp',['TJSAMP',['../group___turbo_j_p_e_g.html#ga1d047060ea80bb9820d540bb928e9074',1,'turbojpeg.h']]],
['tjxop',['TJXOP',['../group___turbo_j_p_e_g.html#ga2de531af4e7e6c4f124908376b354866',1,'turbojpeg.h']]]
diff --git a/doc/html/search/enumvalues_74.js b/doc/html/search/enumvalues_74.js
index 7dc2f8d..799f9ac 100644
--- a/doc/html/search/enumvalues_74.js
+++ b/doc/html/search/enumvalues_74.js
@@ -5,6 +5,8 @@
['tjcs_5frgb',['TJCS_RGB',['../group___turbo_j_p_e_g.html#gga4f83ad3368e0e29d1957be0efa7c3720a677cb7ccb85c4038ac41964a2e09e555',1,'turbojpeg.h']]],
['tjcs_5fycbcr',['TJCS_YCbCr',['../group___turbo_j_p_e_g.html#gga4f83ad3368e0e29d1957be0efa7c3720a7389b8f65bb387ffedce3efd0d78ec75',1,'turbojpeg.h']]],
['tjcs_5fycck',['TJCS_YCCK',['../group___turbo_j_p_e_g.html#gga4f83ad3368e0e29d1957be0efa7c3720a53839e0fe867b76b58d16b0a1a7c598e',1,'turbojpeg.h']]],
+ ['tjerr_5ffatal',['TJERR_FATAL',['../group___turbo_j_p_e_g.html#ggafbc17cfa57d0d5d11fea35ac025950feafc9cceeada13122b09e4851e3788039a',1,'turbojpeg.h']]],
+ ['tjerr_5fwarning',['TJERR_WARNING',['../group___turbo_j_p_e_g.html#ggafbc17cfa57d0d5d11fea35ac025950fea342dd6e2aedb47bb257b4e7568329b59',1,'turbojpeg.h']]],
['tjpf_5fabgr',['TJPF_ABGR',['../group___turbo_j_p_e_g.html#ggac916144e26c3817ac514e64ae5d12e2aa1ba1a7f1631dbeaa49a0a85fc4a40081',1,'turbojpeg.h']]],
['tjpf_5fargb',['TJPF_ARGB',['../group___turbo_j_p_e_g.html#ggac916144e26c3817ac514e64ae5d12e2aae8f846ed9d9de99b6e1dfe448848765c',1,'turbojpeg.h']]],
['tjpf_5fbgr',['TJPF_BGR',['../group___turbo_j_p_e_g.html#ggac916144e26c3817ac514e64ae5d12e2aab10624437fb8ef495a0b153e65749839',1,'turbojpeg.h']]],
diff --git a/doc/html/search/functions_74.js b/doc/html/search/functions_74.js
index 69410b0..d2164bb 100644
--- a/doc/html/search/functions_74.js
+++ b/doc/html/search/functions_74.js
@@ -16,7 +16,8 @@
['tjencodeyuv3',['tjEncodeYUV3',['../group___turbo_j_p_e_g.html#gaabe05acd734990053ad1294b5ef239aa',1,'turbojpeg.h']]],
['tjencodeyuvplanes',['tjEncodeYUVPlanes',['../group___turbo_j_p_e_g.html#ga8a65ed3bd12df57c219d46afbc9008f1',1,'turbojpeg.h']]],
['tjfree',['tjFree',['../group___turbo_j_p_e_g.html#ga8c4a1231dc06a450514c835f6471f137',1,'turbojpeg.h']]],
- ['tjgeterrorstr',['tjGetErrorStr',['../group___turbo_j_p_e_g.html#ga9af79c908ec131b1ae8d52fe40375abf',1,'turbojpeg.h']]],
+ ['tjgeterrorcode',['tjGetErrorCode',['../group___turbo_j_p_e_g.html#ga0be00a62bd1be897f170fa1fed5fb4cb',1,'turbojpeg.h']]],
+ ['tjgeterrorstr2',['tjGetErrorStr2',['../group___turbo_j_p_e_g.html#ga94a235bd4f1088f61ad87b4eadb64c9c',1,'turbojpeg.h']]],
['tjgetscalingfactors',['tjGetScalingFactors',['../group___turbo_j_p_e_g.html#ga6449044b9af402999ccf52f401333be8',1,'turbojpeg.h']]],
['tjinitcompress',['tjInitCompress',['../group___turbo_j_p_e_g.html#ga3d10c47fbe4a2489a2b30c931551d01a',1,'turbojpeg.h']]],
['tjinitdecompress',['tjInitDecompress',['../group___turbo_j_p_e_g.html#gae5408179d041e2a2f7199c8283cf649e',1,'turbojpeg.h']]],
diff --git a/doc/html/structtjregion.html b/doc/html/structtjregion.html
index af2a473..36c0afa 100644
--- a/doc/html/structtjregion.html
+++ b/doc/html/structtjregion.html
@@ -24,7 +24,7 @@
<tr style="height: 56px;">
<td style="padding-left: 0.5em;">
<div id="projectname">TurboJPEG
-  <span id="projectnumber">1.5</span>
+  <span id="projectnumber">1.6</span>
</div>
</td>
</tr>
diff --git a/doc/html/structtjscalingfactor.html b/doc/html/structtjscalingfactor.html
index 3bb50f5..2f54900 100644
--- a/doc/html/structtjscalingfactor.html
+++ b/doc/html/structtjscalingfactor.html
@@ -24,7 +24,7 @@
<tr style="height: 56px;">
<td style="padding-left: 0.5em;">
<div id="projectname">TurboJPEG
-  <span id="projectnumber">1.5</span>
+  <span id="projectnumber">1.6</span>
</div>
</td>
</tr>
diff --git a/doc/html/structtjtransform.html b/doc/html/structtjtransform.html
index 9fd97f7..d9536a3 100644
--- a/doc/html/structtjtransform.html
+++ b/doc/html/structtjtransform.html
@@ -24,7 +24,7 @@
<tr style="height: 56px;">
<td style="padding-left: 0.5em;">
<div id="projectname">TurboJPEG
-  <span id="projectnumber">1.5</span>
+  <span id="projectnumber">1.6</span>
</div>
</td>
</tr>
diff --git a/doxygen.config b/doxygen.config
index 1723123..2f2ab66 100644
--- a/doxygen.config
+++ b/doxygen.config
@@ -1,5 +1,5 @@
PROJECT_NAME = TurboJPEG
-PROJECT_NUMBER = 1.5
+PROJECT_NUMBER = 1.6
OUTPUT_DIRECTORY = doc/
USE_WINDOWS_ENCODING = NO
OPTIMIZE_OUTPUT_FOR_C = YES
diff --git a/java/CMakeLists.txt b/java/CMakeLists.txt
index 0af8ae1..75a0b24 100644
--- a/java/CMakeLists.txt
+++ b/java/CMakeLists.txt
@@ -1,57 +1,78 @@
-set(JAR_FILE turbojpeg.jar)
-set(MANIFEST_FILE ${CMAKE_CURRENT_SOURCE_DIR}/MANIFEST.MF)
+find_package(Java REQUIRED)
+find_package(JNI REQUIRED)
-set(JAVA_CLASSNAMES org/libjpegturbo/turbojpeg/TJ
- org/libjpegturbo/turbojpeg/TJCompressor
- org/libjpegturbo/turbojpeg/TJCustomFilter
- org/libjpegturbo/turbojpeg/TJDecompressor
- org/libjpegturbo/turbojpeg/TJException
- org/libjpegturbo/turbojpeg/TJScalingFactor
- org/libjpegturbo/turbojpeg/TJTransform
- org/libjpegturbo/turbojpeg/TJTransformer
- org/libjpegturbo/turbojpeg/YUVImage
- TJUnitTest
- TJExample
- TJBench)
-
-if(MSVC_IDE)
- set(OBJDIR "${CMAKE_CURRENT_BINARY_DIR}/${CMAKE_CFG_INTDIR}")
-else()
- set(OBJDIR ${CMAKE_CURRENT_BINARY_DIR})
+# Allow the Java compiler flags to be set using an environment variable
+if(NOT DEFINED CMAKE_JAVA_COMPILE_FLAGS AND DEFINED ENV{JAVAFLAGS})
+ set(CMAKE_JAVA_COMPILE_FLAGS $ENV{JAVAFLAGS})
endif()
+include(UseJava)
+
+set(CMAKE_JAVA_COMPILE_FLAGS "${CMAKE_JAVA_COMPILE_FLAGS} -J-Dfile.encoding=UTF8")
+message(STATUS "CMAKE_JAVA_COMPILE_FLAGS = ${CMAKE_JAVA_COMPILE_FLAGS}")
+string(REGEX REPLACE " " ";" CMAKE_JAVA_COMPILE_FLAGS "${CMAKE_JAVA_COMPILE_FLAGS}")
+
+set(JAVAARGS "" CACHE STRING "Additional arguments to pass to java when running unit tests (example: -d32)")
+message(STATUS "JAVAARGS = ${JAVAARGS}")
+
+set(JAVA_SOURCES org/libjpegturbo/turbojpeg/TJ.java
+ org/libjpegturbo/turbojpeg/TJCompressor.java
+ org/libjpegturbo/turbojpeg/TJCustomFilter.java
+ org/libjpegturbo/turbojpeg/TJDecompressor.java
+ org/libjpegturbo/turbojpeg/TJException.java
+ org/libjpegturbo/turbojpeg/TJScalingFactor.java
+ org/libjpegturbo/turbojpeg/TJTransform.java
+ org/libjpegturbo/turbojpeg/TJTransformer.java
+ org/libjpegturbo/turbojpeg/YUVImage.java
+ TJUnitTest.java
+ TJExample.java
+ TJBench.java)
+
set(TURBOJPEG_DLL_NAME "turbojpeg")
if(MINGW)
set(TURBOJPEG_DLL_NAME "libturbojpeg")
endif()
-configure_file(org/libjpegturbo/turbojpeg/TJLoader.java.in
- ${CMAKE_CURRENT_BINARY_DIR}/org/libjpegturbo/turbojpeg/TJLoader.java)
-
-set(JAVA_SOURCES "")
-set(JAVA_CLASSES "")
-set(JAVA_CLASSES_FULL "")
-foreach(class ${JAVA_CLASSNAMES})
- set(JAVA_SOURCES ${JAVA_SOURCES} ${CMAKE_CURRENT_SOURCE_DIR}/${class}.java)
- set(JAVA_CLASSES ${JAVA_CLASSES} ${class}.class)
- set(JAVA_CLASSES_FULL ${JAVA_CLASSES_FULL} ${OBJDIR}/${class}.class)
-endforeach()
-
+if(WIN32)
+ configure_file(org/libjpegturbo/turbojpeg/TJLoader-win.java.in
+ ${CMAKE_CURRENT_BINARY_DIR}/org/libjpegturbo/turbojpeg/TJLoader.java)
+else()
+ configure_file(org/libjpegturbo/turbojpeg/TJLoader-unix.java.in
+ ${CMAKE_CURRENT_BINARY_DIR}/org/libjpegturbo/turbojpeg/TJLoader.java)
+endif()
set(JAVA_SOURCES ${JAVA_SOURCES}
${CMAKE_CURRENT_BINARY_DIR}/org/libjpegturbo/turbojpeg/TJLoader.java)
-set(JAVA_CLASSES ${JAVA_CLASSES}
- org/libjpegturbo/turbojpeg/TJLoader.class)
-set(JAVA_CLASSES_FULL ${JAVA_CLASSES_FULL}
- ${OBJDIR}/org/libjpegturbo/turbojpeg/TJLoader.class)
-string(REGEX REPLACE " " ";" JAVACFLAGS "${JAVACFLAGS}")
-add_custom_command(OUTPUT ${JAVA_CLASSES_FULL} DEPENDS ${JAVA_SOURCES}
- COMMAND ${JAVA_COMPILE} ARGS ${JAVACFLAGS} -d ${OBJDIR} ${JAVA_SOURCES})
+if(MSYS)
+ # UGLY HACK ALERT: If we don't do this, then UseJava.cmake will separate
+ # class path members with a semicolon, which is interpreted as a command
+ # separator by the MSYS shell.
+ set(CMAKE_HOST_SYSTEM_NAME_BAK ${CMAKE_HOST_SYSTEM_NAME})
+ set(CMAKE_HOST_SYSTEM_NAME "MSYS")
+endif()
+add_jar(turbojpeg-java ${JAVA_SOURCES} OUTPUT_NAME turbojpeg
+ ENTRY_POINT TJExample)
+if(MSYS)
+ set(CMAKE_HOST_SYSTEM_NAME ${CMAKE_HOST_SYSTEM_NAME_BAK})
+endif()
-add_custom_command(OUTPUT ${JAR_FILE} DEPENDS ${JAVA_CLASSES_FULL}
- ${MANIFEST_FILE}
- COMMAND ${JAVA_ARCHIVE} cfm ${JAR_FILE} ${MANIFEST_FILE} ${JAVA_CLASSES}
- WORKING_DIRECTORY ${OBJDIR})
+add_custom_target(javadoc COMMAND
+ javadoc -notimestamp -d ${CMAKE_CURRENT_SOURCE_DIR}/doc -sourcepath ${CMAKE_CURRENT_SOURCE_DIR} org.libjpegturbo.turbojpeg)
+set(JAVACLASSPATH ${CMAKE_CURRENT_BINARY_DIR}/${CMAKE_FILES_DIRECTORY}/turbojpeg-java.dir)
+add_custom_target(javah
+ COMMAND javah -d ${CMAKE_CURRENT_SOURCE_DIR} -classpath ${JAVACLASSPATH} org.libjpegturbo.turbojpeg.TJ
+ COMMAND javah -d ${CMAKE_CURRENT_SOURCE_DIR} -classpath ${JAVACLASSPATH} org.libjpegturbo.turbojpeg.TJCompressor
+ COMMAND javah -d ${CMAKE_CURRENT_SOURCE_DIR} -classpath ${JAVACLASSPATH} org.libjpegturbo.turbojpeg.TJDecompressor
+ COMMAND javah -d ${CMAKE_CURRENT_SOURCE_DIR} -classpath ${JAVACLASSPATH} org.libjpegturbo.turbojpeg.TJTransformer)
-add_custom_target(java ALL DEPENDS ${JAR_FILE})
-
-install(FILES ${CMAKE_CURRENT_BINARY_DIR}/${JAR_FILE} DESTINATION classes)
+if(NOT DEFINED CMAKE_INSTALL_DEFAULT_JAVADIR)
+ set(CMAKE_INSTALL_DEFAULT_JAVADIR "<CMAKE_INSTALL_DATAROOTDIR>/java")
+endif()
+GNUInstallDirs_set_install_dir(JAVADIR
+ "The directory into which Java classes should be installed")
+GNUInstallDirs_get_absolute_install_dir(CMAKE_INSTALL_FULL_JAVADIR
+ CMAKE_INSTALL_JAVADIR)
+set(CMAKE_INSTALL_JAVADIR ${CMAKE_INSTALL_JAVADIR} PARENT_SCOPE)
+set(CMAKE_INSTALL_FULL_JAVADIR ${CMAKE_INSTALL_FULL_JAVADIR} PARENT_SCOPE)
+report_directory(JAVADIR)
+install_jar(turbojpeg-java ${CMAKE_INSTALL_JAVADIR})
+mark_as_advanced(CLEAR CMAKE_INSTALL_JAVADIR)
diff --git a/java/Makefile.am b/java/Makefile.am
deleted file mode 100644
index d3fc59c..0000000
--- a/java/Makefile.am
+++ /dev/null
@@ -1,75 +0,0 @@
-JAVAROOT = .
-
-org/libjpegturbo/turbojpeg/TJLoader.java: $(srcdir)/org/libjpegturbo/turbojpeg/TJLoader.java.tmpl
- mkdir -p org/libjpegturbo/turbojpeg; \
- cat $(srcdir)/org/libjpegturbo/turbojpeg/TJLoader.java.tmpl | \
- sed s@%{__libdir}@$(libdir)@g > org/libjpegturbo/turbojpeg/TJLoader.java
-
-
-JAVASOURCES = org/libjpegturbo/turbojpeg/TJ.java \
- org/libjpegturbo/turbojpeg/TJCompressor.java \
- org/libjpegturbo/turbojpeg/TJCustomFilter.java \
- org/libjpegturbo/turbojpeg/TJDecompressor.java \
- org/libjpegturbo/turbojpeg/TJException.java \
- org/libjpegturbo/turbojpeg/TJScalingFactor.java \
- org/libjpegturbo/turbojpeg/TJTransform.java \
- org/libjpegturbo/turbojpeg/TJTransformer.java \
- org/libjpegturbo/turbojpeg/YUVImage.java \
- TJExample.java \
- TJUnitTest.java \
- TJBench.java
-
-JNIHEADERS = org_libjpegturbo_turbojpeg_TJ.h \
- org_libjpegturbo_turbojpeg_TJCompressor.h \
- org_libjpegturbo_turbojpeg_TJDecompressor.h \
- org_libjpegturbo_turbojpeg_TJTransformer.h
-
-if WITH_JAVA
-
-nodist_noinst_JAVA = ${JAVASOURCES} org/libjpegturbo/turbojpeg/TJLoader.java
-
-JAVA_CLASSES = org/libjpegturbo/turbojpeg/TJ.class \
- org/libjpegturbo/turbojpeg/TJCompressor.class \
- org/libjpegturbo/turbojpeg/TJCustomFilter.class \
- org/libjpegturbo/turbojpeg/TJDecompressor.class \
- org/libjpegturbo/turbojpeg/TJException.class \
- org/libjpegturbo/turbojpeg/TJLoader.class \
- org/libjpegturbo/turbojpeg/TJScalingFactor.class \
- org/libjpegturbo/turbojpeg/TJTransform.class \
- org/libjpegturbo/turbojpeg/TJTransformer.class \
- org/libjpegturbo/turbojpeg/YUVImage.class \
- TJExample.class \
- TJUnitTest.class \
- TJBench.class
-
-all: all-am turbojpeg.jar
-
-turbojpeg.jar: classnoinst.stamp ${srcdir}/MANIFEST.MF
- $(JAR) cfm turbojpeg.jar ${srcdir}/MANIFEST.MF $(JAVA_CLASSES)
-
-clean-local:
- rm -f turbojpeg.jar
-
-install-exec-local: turbojpeg.jar
- mkdir -p $(DESTDIR)/$(datadir)/classes
- $(INSTALL) -m 644 turbojpeg.jar $(DESTDIR)/$(datadir)/classes/
-
-uninstall-local:
- rm -f $(DESTDIR)/$(datadir)/classes/turbojpeg.jar
- if [ -d $(DESTDIR)/$(datadir)/classes ]; then rmdir $(DESTDIR)/$(datadir)/classes; fi
-
-headers: all
- javah -d ${srcdir} org.libjpegturbo.turbojpeg.TJ; \
- javah -d ${srcdir} org.libjpegturbo.turbojpeg.TJCompressor; \
- javah -d ${srcdir} org.libjpegturbo.turbojpeg.TJDecompressor; \
- javah -d ${srcdir} org.libjpegturbo.turbojpeg.TJTransformer
-
-docs: all
- mkdir -p ${srcdir}/doc; \
- javadoc -notimestamp -d ${srcdir}/doc -sourcepath ${srcdir} org.libjpegturbo.turbojpeg
-
-endif
-
-EXTRA_DIST = MANIFEST.MF ${JAVASOURCES} ${JNIHEADERS} doc CMakeLists.txt \
- org/libjpegturbo/turbojpeg/TJLoader.java.tmpl \
- org/libjpegturbo/turbojpeg/TJLoader.java.in
diff --git a/java/TJBench.java b/java/TJBench.java
index 5185d65..0733816 100644
--- a/java/TJBench.java
+++ b/java/TJBench.java
@@ -63,6 +63,26 @@
}
+ static String tjErrorMsg;
+ static int tjErrorCode = -1;
+
+ static void handleTJException(TJException e) throws TJException {
+ String _tjErrorMsg = e.getMessage();
+ int _tjErrorCode = e.getErrorCode();
+
+ if ((flags & TJ.FLAG_STOPONWARNING) == 0 &&
+ _tjErrorCode == TJ.ERR_WARNING) {
+ if (tjErrorMsg == null || !tjErrorMsg.equals(_tjErrorMsg) ||
+ tjErrorCode != _tjErrorCode) {
+ tjErrorMsg = _tjErrorMsg;
+ tjErrorCode = _tjErrorCode;
+ System.out.println("WARNING: " + _tjErrorMsg);
+ }
+ } else
+ throw e;
+ }
+
+
static String formatName(int subsamp, int cs) {
if (cs == TJ.CS_YCbCr)
return subNameLong[subsamp];
@@ -174,14 +194,21 @@
tjd.setSourceImage(jpegBuf[tile], jpegSize[tile]);
if (doYUV) {
yuvImage.setBuf(yuvImage.getBuf(), width, yuvpad, height, subsamp);
- tjd.decompressToYUV(yuvImage, flags);
+ try {
+ tjd.decompressToYUV(yuvImage, flags);
+ } catch (TJException e) { handleTJException(e); }
double startDecode = getTime();
tjd.setSourceImage(yuvImage);
- tjd.decompress(dstBuf, x, y, width, pitch, height, pf, flags);
+ try {
+ tjd.decompress(dstBuf, x, y, width, pitch, height, pf, flags);
+ } catch (TJException e) { handleTJException(e); }
if (iter >= 0)
elapsedDecode += getTime() - startDecode;
- } else
- tjd.decompress(dstBuf, x, y, width, pitch, height, pf, flags);
+ } else {
+ try {
+ tjd.decompress(dstBuf, x, y, width, pitch, height, pf, flags);
+ } catch (TJException e) { handleTJException(e); }
+ }
}
}
elapsed += getTime() - start;
@@ -636,6 +663,8 @@
System.out.println(" codec");
System.out.println("-accuratedct = Use the most accurate DCT/IDCT algorithms available in the");
System.out.println(" underlying codec");
+ System.out.println("-progressive = Use progressive entropy coding in JPEG images generated by");
+ System.out.println(" compression and transform operations.");
System.out.println("-subsamp <s> = When testing JPEG compression, this option specifies the level");
System.out.println(" of chrominance subsampling to use (<s> = 444, 422, 440, 420, 411, or");
System.out.println(" GRAY). The default is to test Grayscale, 4:2:0, 4:2:2, and 4:4:4 in");
@@ -667,13 +696,18 @@
System.out.println(" decompression (these options are mutually exclusive)");
System.out.println("-grayscale = Perform lossless grayscale conversion prior to decompression");
System.out.println(" test (can be combined with the other transforms above)");
+ System.out.println("-copynone = Do not copy any extra markers (including EXIF and ICC profile data)");
+ System.out.println(" when transforming the image.");
System.out.println("-benchtime <t> = Run each benchmark for at least <t> seconds (default = 5.0)");
System.out.println("-warmup <t> = Run each benchmark for <t> seconds (default = 1.0) prior to");
System.out.println(" starting the timer, in order to prime the caches and thus improve the");
System.out.println(" consistency of the results.");
System.out.println("-componly = Stop after running compression tests. Do not test decompression.");
System.out.println("-nowrite = Do not write reference or output images (improves consistency");
- System.out.println(" of performance measurements.)\n");
+ System.out.println(" of performance measurements.)");
+ System.out.println("-stoponwarning = Immediately discontinue the current");
+ System.out.println(" compression/decompression/transform operation if the underlying codec");
+ System.out.println(" throws a warning (non-fatal error)\n");
System.out.println("NOTE: If the quality is specified as a range (e.g. 90-100), a separate");
System.out.println("test will be performed for all quality values in the range.\n");
System.exit(1);
@@ -733,6 +767,10 @@
System.out.println("Using most accurate DCT/IDCT algorithm\n");
flags |= TJ.FLAG_ACCURATEDCT;
}
+ if (argv[i].equalsIgnoreCase("-progressive")) {
+ System.out.println("Using progressive entropy coding\n");
+ flags |= TJ.FLAG_PROGRESSIVE;
+ }
if (argv[i].equalsIgnoreCase("-rgb"))
pf = TJ.PF_RGB;
if (argv[i].equalsIgnoreCase("-rgbx"))
@@ -792,6 +830,8 @@
xformOpt |= TJTransform.OPT_GRAY;
if (argv[i].equalsIgnoreCase("-nooutput"))
xformOpt |= TJTransform.OPT_NOOUTPUT;
+ if (argv[i].equalsIgnoreCase("-copynone"))
+ xformOpt |= TJTransform.OPT_COPYNONE;
if (argv[i].equalsIgnoreCase("-benchtime") && i < argv.length - 1) {
double temp = -1;
try {
@@ -844,6 +884,8 @@
} else
usage();
}
+ if (argv[i].equalsIgnoreCase("-stoponwarning"))
+ flags |= TJ.FLAG_STOPONWARNING;
if (argv[i].equalsIgnoreCase("-?"))
usage();
}
@@ -916,7 +958,12 @@
}
} catch (Exception e) {
- System.out.println("ERROR: " + e.getMessage());
+ if (e instanceof TJException) {
+ TJException tje = (TJException)e;
+ System.out.println((tje.getErrorCode() == TJ.ERR_WARNING ?
+ "WARNING: " : "ERROR: ") + tje.getMessage());
+ } else
+ System.out.println("ERROR: " + e.getMessage());
e.printStackTrace();
retval = -1;
}
diff --git a/java/doc/constant-values.html b/java/doc/constant-values.html
index ec1b21d..985784e 100644
--- a/java/doc/constant-values.html
+++ b/java/doc/constant-values.html
@@ -115,6 +115,20 @@
<td class="colLast"><code>4</code></td>
</tr>
<tr class="rowColor">
+<td class="colFirst"><a name="org.libjpegturbo.turbojpeg.TJ.ERR_FATAL">
+<!-- -->
+</a><code>public static final int</code></td>
+<td><code><a href="org/libjpegturbo/turbojpeg/TJ.html#ERR_FATAL">ERR_FATAL</a></code></td>
+<td class="colLast"><code>1</code></td>
+</tr>
+<tr class="altColor">
+<td class="colFirst"><a name="org.libjpegturbo.turbojpeg.TJ.ERR_WARNING">
+<!-- -->
+</a><code>public static final int</code></td>
+<td><code><a href="org/libjpegturbo/turbojpeg/TJ.html#ERR_WARNING">ERR_WARNING</a></code></td>
+<td class="colLast"><code>0</code></td>
+</tr>
+<tr class="rowColor">
<td class="colFirst"><a name="org.libjpegturbo.turbojpeg.TJ.FLAG_ACCURATEDCT">
<!-- -->
</a><code>public static final int</code></td>
@@ -171,6 +185,20 @@
<td class="colLast"><code>128</code></td>
</tr>
<tr class="rowColor">
+<td class="colFirst"><a name="org.libjpegturbo.turbojpeg.TJ.FLAG_PROGRESSIVE">
+<!-- -->
+</a><code>public static final int</code></td>
+<td><code><a href="org/libjpegturbo/turbojpeg/TJ.html#FLAG_PROGRESSIVE">FLAG_PROGRESSIVE</a></code></td>
+<td class="colLast"><code>16384</code></td>
+</tr>
+<tr class="altColor">
+<td class="colFirst"><a name="org.libjpegturbo.turbojpeg.TJ.FLAG_STOPONWARNING">
+<!-- -->
+</a><code>public static final int</code></td>
+<td><code><a href="org/libjpegturbo/turbojpeg/TJ.html#FLAG_STOPONWARNING">FLAG_STOPONWARNING</a></code></td>
+<td class="colLast"><code>8192</code></td>
+</tr>
+<tr class="rowColor">
<td class="colFirst"><a name="org.libjpegturbo.turbojpeg.TJ.NUMCS">
<!-- -->
</a><code>public static final int</code></td>
@@ -178,139 +206,146 @@
<td class="colLast"><code>5</code></td>
</tr>
<tr class="altColor">
+<td class="colFirst"><a name="org.libjpegturbo.turbojpeg.TJ.NUMERR">
+<!-- -->
+</a><code>public static final int</code></td>
+<td><code><a href="org/libjpegturbo/turbojpeg/TJ.html#NUMERR">NUMERR</a></code></td>
+<td class="colLast"><code>2</code></td>
+</tr>
+<tr class="rowColor">
<td class="colFirst"><a name="org.libjpegturbo.turbojpeg.TJ.NUMPF">
<!-- -->
</a><code>public static final int</code></td>
<td><code><a href="org/libjpegturbo/turbojpeg/TJ.html#NUMPF">NUMPF</a></code></td>
<td class="colLast"><code>12</code></td>
</tr>
-<tr class="rowColor">
+<tr class="altColor">
<td class="colFirst"><a name="org.libjpegturbo.turbojpeg.TJ.NUMSAMP">
<!-- -->
</a><code>public static final int</code></td>
<td><code><a href="org/libjpegturbo/turbojpeg/TJ.html#NUMSAMP">NUMSAMP</a></code></td>
<td class="colLast"><code>6</code></td>
</tr>
-<tr class="altColor">
+<tr class="rowColor">
<td class="colFirst"><a name="org.libjpegturbo.turbojpeg.TJ.PF_ABGR">
<!-- -->
</a><code>public static final int</code></td>
<td><code><a href="org/libjpegturbo/turbojpeg/TJ.html#PF_ABGR">PF_ABGR</a></code></td>
<td class="colLast"><code>9</code></td>
</tr>
-<tr class="rowColor">
+<tr class="altColor">
<td class="colFirst"><a name="org.libjpegturbo.turbojpeg.TJ.PF_ARGB">
<!-- -->
</a><code>public static final int</code></td>
<td><code><a href="org/libjpegturbo/turbojpeg/TJ.html#PF_ARGB">PF_ARGB</a></code></td>
<td class="colLast"><code>10</code></td>
</tr>
-<tr class="altColor">
+<tr class="rowColor">
<td class="colFirst"><a name="org.libjpegturbo.turbojpeg.TJ.PF_BGR">
<!-- -->
</a><code>public static final int</code></td>
<td><code><a href="org/libjpegturbo/turbojpeg/TJ.html#PF_BGR">PF_BGR</a></code></td>
<td class="colLast"><code>1</code></td>
</tr>
-<tr class="rowColor">
+<tr class="altColor">
<td class="colFirst"><a name="org.libjpegturbo.turbojpeg.TJ.PF_BGRA">
<!-- -->
</a><code>public static final int</code></td>
<td><code><a href="org/libjpegturbo/turbojpeg/TJ.html#PF_BGRA">PF_BGRA</a></code></td>
<td class="colLast"><code>8</code></td>
</tr>
-<tr class="altColor">
+<tr class="rowColor">
<td class="colFirst"><a name="org.libjpegturbo.turbojpeg.TJ.PF_BGRX">
<!-- -->
</a><code>public static final int</code></td>
<td><code><a href="org/libjpegturbo/turbojpeg/TJ.html#PF_BGRX">PF_BGRX</a></code></td>
<td class="colLast"><code>3</code></td>
</tr>
-<tr class="rowColor">
+<tr class="altColor">
<td class="colFirst"><a name="org.libjpegturbo.turbojpeg.TJ.PF_CMYK">
<!-- -->
</a><code>public static final int</code></td>
<td><code><a href="org/libjpegturbo/turbojpeg/TJ.html#PF_CMYK">PF_CMYK</a></code></td>
<td class="colLast"><code>11</code></td>
</tr>
-<tr class="altColor">
+<tr class="rowColor">
<td class="colFirst"><a name="org.libjpegturbo.turbojpeg.TJ.PF_GRAY">
<!-- -->
</a><code>public static final int</code></td>
<td><code><a href="org/libjpegturbo/turbojpeg/TJ.html#PF_GRAY">PF_GRAY</a></code></td>
<td class="colLast"><code>6</code></td>
</tr>
-<tr class="rowColor">
+<tr class="altColor">
<td class="colFirst"><a name="org.libjpegturbo.turbojpeg.TJ.PF_RGB">
<!-- -->
</a><code>public static final int</code></td>
<td><code><a href="org/libjpegturbo/turbojpeg/TJ.html#PF_RGB">PF_RGB</a></code></td>
<td class="colLast"><code>0</code></td>
</tr>
-<tr class="altColor">
+<tr class="rowColor">
<td class="colFirst"><a name="org.libjpegturbo.turbojpeg.TJ.PF_RGBA">
<!-- -->
</a><code>public static final int</code></td>
<td><code><a href="org/libjpegturbo/turbojpeg/TJ.html#PF_RGBA">PF_RGBA</a></code></td>
<td class="colLast"><code>7</code></td>
</tr>
-<tr class="rowColor">
+<tr class="altColor">
<td class="colFirst"><a name="org.libjpegturbo.turbojpeg.TJ.PF_RGBX">
<!-- -->
</a><code>public static final int</code></td>
<td><code><a href="org/libjpegturbo/turbojpeg/TJ.html#PF_RGBX">PF_RGBX</a></code></td>
<td class="colLast"><code>2</code></td>
</tr>
-<tr class="altColor">
+<tr class="rowColor">
<td class="colFirst"><a name="org.libjpegturbo.turbojpeg.TJ.PF_XBGR">
<!-- -->
</a><code>public static final int</code></td>
<td><code><a href="org/libjpegturbo/turbojpeg/TJ.html#PF_XBGR">PF_XBGR</a></code></td>
<td class="colLast"><code>4</code></td>
</tr>
-<tr class="rowColor">
+<tr class="altColor">
<td class="colFirst"><a name="org.libjpegturbo.turbojpeg.TJ.PF_XRGB">
<!-- -->
</a><code>public static final int</code></td>
<td><code><a href="org/libjpegturbo/turbojpeg/TJ.html#PF_XRGB">PF_XRGB</a></code></td>
<td class="colLast"><code>5</code></td>
</tr>
-<tr class="altColor">
+<tr class="rowColor">
<td class="colFirst"><a name="org.libjpegturbo.turbojpeg.TJ.SAMP_411">
<!-- -->
</a><code>public static final int</code></td>
<td><code><a href="org/libjpegturbo/turbojpeg/TJ.html#SAMP_411">SAMP_411</a></code></td>
<td class="colLast"><code>5</code></td>
</tr>
-<tr class="rowColor">
+<tr class="altColor">
<td class="colFirst"><a name="org.libjpegturbo.turbojpeg.TJ.SAMP_420">
<!-- -->
</a><code>public static final int</code></td>
<td><code><a href="org/libjpegturbo/turbojpeg/TJ.html#SAMP_420">SAMP_420</a></code></td>
<td class="colLast"><code>2</code></td>
</tr>
-<tr class="altColor">
+<tr class="rowColor">
<td class="colFirst"><a name="org.libjpegturbo.turbojpeg.TJ.SAMP_422">
<!-- -->
</a><code>public static final int</code></td>
<td><code><a href="org/libjpegturbo/turbojpeg/TJ.html#SAMP_422">SAMP_422</a></code></td>
<td class="colLast"><code>1</code></td>
</tr>
-<tr class="rowColor">
+<tr class="altColor">
<td class="colFirst"><a name="org.libjpegturbo.turbojpeg.TJ.SAMP_440">
<!-- -->
</a><code>public static final int</code></td>
<td><code><a href="org/libjpegturbo/turbojpeg/TJ.html#SAMP_440">SAMP_440</a></code></td>
<td class="colLast"><code>4</code></td>
</tr>
-<tr class="altColor">
+<tr class="rowColor">
<td class="colFirst"><a name="org.libjpegturbo.turbojpeg.TJ.SAMP_444">
<!-- -->
</a><code>public static final int</code></td>
<td><code><a href="org/libjpegturbo/turbojpeg/TJ.html#SAMP_444">SAMP_444</a></code></td>
<td class="colLast"><code>0</code></td>
</tr>
-<tr class="rowColor">
+<tr class="altColor">
<td class="colFirst"><a name="org.libjpegturbo.turbojpeg.TJ.SAMP_GRAY">
<!-- -->
</a><code>public static final int</code></td>
@@ -393,33 +428,47 @@
<td class="colLast"><code>2</code></td>
</tr>
<tr class="rowColor">
+<td class="colFirst"><a name="org.libjpegturbo.turbojpeg.TJTransform.OPT_COPYNONE">
+<!-- -->
+</a><code>public static final int</code></td>
+<td><code><a href="org/libjpegturbo/turbojpeg/TJTransform.html#OPT_COPYNONE">OPT_COPYNONE</a></code></td>
+<td class="colLast"><code>64</code></td>
+</tr>
+<tr class="altColor">
<td class="colFirst"><a name="org.libjpegturbo.turbojpeg.TJTransform.OPT_CROP">
<!-- -->
</a><code>public static final int</code></td>
<td><code><a href="org/libjpegturbo/turbojpeg/TJTransform.html#OPT_CROP">OPT_CROP</a></code></td>
<td class="colLast"><code>4</code></td>
</tr>
-<tr class="altColor">
+<tr class="rowColor">
<td class="colFirst"><a name="org.libjpegturbo.turbojpeg.TJTransform.OPT_GRAY">
<!-- -->
</a><code>public static final int</code></td>
<td><code><a href="org/libjpegturbo/turbojpeg/TJTransform.html#OPT_GRAY">OPT_GRAY</a></code></td>
<td class="colLast"><code>8</code></td>
</tr>
-<tr class="rowColor">
+<tr class="altColor">
<td class="colFirst"><a name="org.libjpegturbo.turbojpeg.TJTransform.OPT_NOOUTPUT">
<!-- -->
</a><code>public static final int</code></td>
<td><code><a href="org/libjpegturbo/turbojpeg/TJTransform.html#OPT_NOOUTPUT">OPT_NOOUTPUT</a></code></td>
<td class="colLast"><code>16</code></td>
</tr>
-<tr class="altColor">
+<tr class="rowColor">
<td class="colFirst"><a name="org.libjpegturbo.turbojpeg.TJTransform.OPT_PERFECT">
<!-- -->
</a><code>public static final int</code></td>
<td><code><a href="org/libjpegturbo/turbojpeg/TJTransform.html#OPT_PERFECT">OPT_PERFECT</a></code></td>
<td class="colLast"><code>1</code></td>
</tr>
+<tr class="altColor">
+<td class="colFirst"><a name="org.libjpegturbo.turbojpeg.TJTransform.OPT_PROGRESSIVE">
+<!-- -->
+</a><code>public static final int</code></td>
+<td><code><a href="org/libjpegturbo/turbojpeg/TJTransform.html#OPT_PROGRESSIVE">OPT_PROGRESSIVE</a></code></td>
+<td class="colLast"><code>32</code></td>
+</tr>
<tr class="rowColor">
<td class="colFirst"><a name="org.libjpegturbo.turbojpeg.TJTransform.OPT_TRIM">
<!-- -->
diff --git a/java/doc/index-all.html b/java/doc/index-all.html
index a02d9c4..a9d139a 100644
--- a/java/doc/index-all.html
+++ b/java/doc/index-all.html
@@ -278,6 +278,15 @@
<div class="block">Returns true or false, depending on whether this instance and
<code>other</code> have the same numerator and denominator.</div>
</dd>
+<dt><span class="strong"><a href="./org/libjpegturbo/turbojpeg/TJ.html#ERR_FATAL">ERR_FATAL</a></span> - Static variable in class org.libjpegturbo.turbojpeg.<a href="./org/libjpegturbo/turbojpeg/TJ.html" title="class in org.libjpegturbo.turbojpeg">TJ</a></dt>
+<dd>
+<div class="block">The error was fatal and non-recoverable.</div>
+</dd>
+<dt><span class="strong"><a href="./org/libjpegturbo/turbojpeg/TJ.html#ERR_WARNING">ERR_WARNING</a></span> - Static variable in class org.libjpegturbo.turbojpeg.<a href="./org/libjpegturbo/turbojpeg/TJ.html" title="class in org.libjpegturbo.turbojpeg">TJ</a></dt>
+<dd>
+<div class="block">The error was non-fatal and recoverable, but the image may still be
+ corrupt.</div>
+</dd>
</dl>
<a name="_F_">
<!-- -->
@@ -324,6 +333,16 @@
<dd>
<div class="block"><span class="strong">Deprecated.</span></div>
</dd>
+<dt><span class="strong"><a href="./org/libjpegturbo/turbojpeg/TJ.html#FLAG_PROGRESSIVE">FLAG_PROGRESSIVE</a></span> - Static variable in class org.libjpegturbo.turbojpeg.<a href="./org/libjpegturbo/turbojpeg/TJ.html" title="class in org.libjpegturbo.turbojpeg">TJ</a></dt>
+<dd>
+<div class="block">Use progressive entropy coding in JPEG images generated by compression and
+ transform operations.</div>
+</dd>
+<dt><span class="strong"><a href="./org/libjpegturbo/turbojpeg/TJ.html#FLAG_STOPONWARNING">FLAG_STOPONWARNING</a></span> - Static variable in class org.libjpegturbo.turbojpeg.<a href="./org/libjpegturbo/turbojpeg/TJ.html" title="class in org.libjpegturbo.turbojpeg">TJ</a></dt>
+<dd>
+<div class="block">Immediately discontinue the current compression/decompression/transform
+ operation if the underlying codec throws a warning (non-fatal error).</div>
+</dd>
</dl>
<a name="_G_">
<!-- -->
@@ -354,6 +373,11 @@
<dd>
<div class="block">Returns denominator</div>
</dd>
+<dt><span class="strong"><a href="./org/libjpegturbo/turbojpeg/TJException.html#getErrorCode()">getErrorCode()</a></span> - Method in exception org.libjpegturbo.turbojpeg.<a href="./org/libjpegturbo/turbojpeg/TJException.html" title="class in org.libjpegturbo.turbojpeg">TJException</a></dt>
+<dd>
+<div class="block">Returns a code (one of <a href="./org/libjpegturbo/turbojpeg/TJ.html" title="class in org.libjpegturbo.turbojpeg"><code>TJ.ERR_*</code></a>) indicating the severity of the
+ last error.</div>
+</dd>
<dt><span class="strong"><a href="./org/libjpegturbo/turbojpeg/TJ.html#getGreenOffset(int)">getGreenOffset(int)</a></span> - Static method in class org.libjpegturbo.turbojpeg.<a href="./org/libjpegturbo/turbojpeg/TJ.html" title="class in org.libjpegturbo.turbojpeg">TJ</a></dt>
<dd>
<div class="block">For the given pixel format, returns the number of bytes that the green
@@ -516,6 +540,10 @@
<dd>
<div class="block">The number of JPEG colorspaces</div>
</dd>
+<dt><span class="strong"><a href="./org/libjpegturbo/turbojpeg/TJ.html#NUMERR">NUMERR</a></span> - Static variable in class org.libjpegturbo.turbojpeg.<a href="./org/libjpegturbo/turbojpeg/TJ.html" title="class in org.libjpegturbo.turbojpeg">TJ</a></dt>
+<dd>
+<div class="block">The number of error codes</div>
+</dd>
<dt><span class="strong"><a href="./org/libjpegturbo/turbojpeg/TJTransform.html#NUMOP">NUMOP</a></span> - Static variable in class org.libjpegturbo.turbojpeg.<a href="./org/libjpegturbo/turbojpeg/TJTransform.html" title="class in org.libjpegturbo.turbojpeg">TJTransform</a></dt>
<dd>
<div class="block">The number of lossless transform operations</div>
@@ -571,6 +599,11 @@
<dd>
<div class="block">Flip (mirror) image vertically.</div>
</dd>
+<dt><span class="strong"><a href="./org/libjpegturbo/turbojpeg/TJTransform.html#OPT_COPYNONE">OPT_COPYNONE</a></span> - Static variable in class org.libjpegturbo.turbojpeg.<a href="./org/libjpegturbo/turbojpeg/TJTransform.html" title="class in org.libjpegturbo.turbojpeg">TJTransform</a></dt>
+<dd>
+<div class="block">This option will prevent <a href="./org/libjpegturbo/turbojpeg/TJTransformer.html#transform(byte[][],%20org.libjpegturbo.turbojpeg.TJTransform[],%20int)"><code>TJTransformer.transform()</code></a> from copying any extra markers (including EXIF
+ and ICC profile data) from the source image to the output image.</div>
+</dd>
<dt><span class="strong"><a href="./org/libjpegturbo/turbojpeg/TJTransform.html#OPT_CROP">OPT_CROP</a></span> - Static variable in class org.libjpegturbo.turbojpeg.<a href="./org/libjpegturbo/turbojpeg/TJTransform.html" title="class in org.libjpegturbo.turbojpeg">TJTransform</a></dt>
<dd>
<div class="block">This option will enable lossless cropping.</div>
@@ -590,6 +623,11 @@
<div class="block">This option will cause <a href="./org/libjpegturbo/turbojpeg/TJTransformer.html#transform(byte[][],%20org.libjpegturbo.turbojpeg.TJTransform[],%20int)"><code>TJTransformer.transform()</code></a> to throw an exception if the transform is not
perfect.</div>
</dd>
+<dt><span class="strong"><a href="./org/libjpegturbo/turbojpeg/TJTransform.html#OPT_PROGRESSIVE">OPT_PROGRESSIVE</a></span> - Static variable in class org.libjpegturbo.turbojpeg.<a href="./org/libjpegturbo/turbojpeg/TJTransform.html" title="class in org.libjpegturbo.turbojpeg">TJTransform</a></dt>
+<dd>
+<div class="block">This option will enable progressive entropy coding in the output image
+ generated by this particular transform.</div>
+</dd>
<dt><span class="strong"><a href="./org/libjpegturbo/turbojpeg/TJTransform.html#OPT_TRIM">OPT_TRIM</a></span> - Static variable in class org.libjpegturbo.turbojpeg.<a href="./org/libjpegturbo/turbojpeg/TJTransform.html" title="class in org.libjpegturbo.turbojpeg">TJTransform</a></dt>
<dd>
<div class="block">This option will discard any partial MCU blocks that cannot be
@@ -829,6 +867,8 @@
<dd> </dd>
<dt><span class="strong"><a href="./org/libjpegturbo/turbojpeg/TJException.html#TJException(java.lang.String)">TJException(String)</a></span> - Constructor for exception org.libjpegturbo.turbojpeg.<a href="./org/libjpegturbo/turbojpeg/TJException.html" title="class in org.libjpegturbo.turbojpeg">TJException</a></dt>
<dd> </dd>
+<dt><span class="strong"><a href="./org/libjpegturbo/turbojpeg/TJException.html#TJException(java.lang.String,%20int)">TJException(String, int)</a></span> - Constructor for exception org.libjpegturbo.turbojpeg.<a href="./org/libjpegturbo/turbojpeg/TJException.html" title="class in org.libjpegturbo.turbojpeg">TJException</a></dt>
+<dd> </dd>
<dt><span class="strong"><a href="./org/libjpegturbo/turbojpeg/TJException.html#TJException(java.lang.Throwable)">TJException(Throwable)</a></span> - Constructor for exception org.libjpegturbo.turbojpeg.<a href="./org/libjpegturbo/turbojpeg/TJException.html" title="class in org.libjpegturbo.turbojpeg">TJException</a></dt>
<dd> </dd>
<dt><a href="./org/libjpegturbo/turbojpeg/TJScalingFactor.html" title="class in org.libjpegturbo.turbojpeg"><span class="strong">TJScalingFactor</span></a> - Class in <a href="./org/libjpegturbo/turbojpeg/package-summary.html">org.libjpegturbo.turbojpeg</a></dt>
diff --git a/java/doc/org/libjpegturbo/turbojpeg/TJ.html b/java/doc/org/libjpegturbo/turbojpeg/TJ.html
index ffef657..d569126 100644
--- a/java/doc/org/libjpegturbo/turbojpeg/TJ.html
+++ b/java/doc/org/libjpegturbo/turbojpeg/TJ.html
@@ -145,6 +145,19 @@
</tr>
<tr class="rowColor">
<td class="colFirst"><code>static int</code></td>
+<td class="colLast"><code><strong><a href="../../../org/libjpegturbo/turbojpeg/TJ.html#ERR_FATAL">ERR_FATAL</a></strong></code>
+<div class="block">The error was fatal and non-recoverable.</div>
+</td>
+</tr>
+<tr class="altColor">
+<td class="colFirst"><code>static int</code></td>
+<td class="colLast"><code><strong><a href="../../../org/libjpegturbo/turbojpeg/TJ.html#ERR_WARNING">ERR_WARNING</a></strong></code>
+<div class="block">The error was non-fatal and recoverable, but the image may still be
+ corrupt.</div>
+</td>
+</tr>
+<tr class="rowColor">
+<td class="colFirst"><code>static int</code></td>
<td class="colLast"><code><strong><a href="../../../org/libjpegturbo/turbojpeg/TJ.html#FLAG_ACCURATEDCT">FLAG_ACCURATEDCT</a></strong></code>
<div class="block">Use the most accurate DCT/IDCT algorithm available in the underlying
codec.</div>
@@ -197,125 +210,145 @@
</tr>
<tr class="rowColor">
<td class="colFirst"><code>static int</code></td>
+<td class="colLast"><code><strong><a href="../../../org/libjpegturbo/turbojpeg/TJ.html#FLAG_PROGRESSIVE">FLAG_PROGRESSIVE</a></strong></code>
+<div class="block">Use progressive entropy coding in JPEG images generated by compression and
+ transform operations.</div>
+</td>
+</tr>
+<tr class="altColor">
+<td class="colFirst"><code>static int</code></td>
+<td class="colLast"><code><strong><a href="../../../org/libjpegturbo/turbojpeg/TJ.html#FLAG_STOPONWARNING">FLAG_STOPONWARNING</a></strong></code>
+<div class="block">Immediately discontinue the current compression/decompression/transform
+ operation if the underlying codec throws a warning (non-fatal error).</div>
+</td>
+</tr>
+<tr class="rowColor">
+<td class="colFirst"><code>static int</code></td>
<td class="colLast"><code><strong><a href="../../../org/libjpegturbo/turbojpeg/TJ.html#NUMCS">NUMCS</a></strong></code>
<div class="block">The number of JPEG colorspaces</div>
</td>
</tr>
<tr class="altColor">
<td class="colFirst"><code>static int</code></td>
+<td class="colLast"><code><strong><a href="../../../org/libjpegturbo/turbojpeg/TJ.html#NUMERR">NUMERR</a></strong></code>
+<div class="block">The number of error codes</div>
+</td>
+</tr>
+<tr class="rowColor">
+<td class="colFirst"><code>static int</code></td>
<td class="colLast"><code><strong><a href="../../../org/libjpegturbo/turbojpeg/TJ.html#NUMPF">NUMPF</a></strong></code>
<div class="block">The number of pixel formats</div>
</td>
</tr>
-<tr class="rowColor">
+<tr class="altColor">
<td class="colFirst"><code>static int</code></td>
<td class="colLast"><code><strong><a href="../../../org/libjpegturbo/turbojpeg/TJ.html#NUMSAMP">NUMSAMP</a></strong></code>
<div class="block">The number of chrominance subsampling options</div>
</td>
</tr>
-<tr class="altColor">
+<tr class="rowColor">
<td class="colFirst"><code>static int</code></td>
<td class="colLast"><code><strong><a href="../../../org/libjpegturbo/turbojpeg/TJ.html#PF_ABGR">PF_ABGR</a></strong></code>
<div class="block">ABGR pixel format.</div>
</td>
</tr>
-<tr class="rowColor">
+<tr class="altColor">
<td class="colFirst"><code>static int</code></td>
<td class="colLast"><code><strong><a href="../../../org/libjpegturbo/turbojpeg/TJ.html#PF_ARGB">PF_ARGB</a></strong></code>
<div class="block">ARGB pixel format.</div>
</td>
</tr>
-<tr class="altColor">
+<tr class="rowColor">
<td class="colFirst"><code>static int</code></td>
<td class="colLast"><code><strong><a href="../../../org/libjpegturbo/turbojpeg/TJ.html#PF_BGR">PF_BGR</a></strong></code>
<div class="block">BGR pixel format.</div>
</td>
</tr>
-<tr class="rowColor">
+<tr class="altColor">
<td class="colFirst"><code>static int</code></td>
<td class="colLast"><code><strong><a href="../../../org/libjpegturbo/turbojpeg/TJ.html#PF_BGRA">PF_BGRA</a></strong></code>
<div class="block">BGRA pixel format.</div>
</td>
</tr>
-<tr class="altColor">
+<tr class="rowColor">
<td class="colFirst"><code>static int</code></td>
<td class="colLast"><code><strong><a href="../../../org/libjpegturbo/turbojpeg/TJ.html#PF_BGRX">PF_BGRX</a></strong></code>
<div class="block">BGRX pixel format.</div>
</td>
</tr>
-<tr class="rowColor">
+<tr class="altColor">
<td class="colFirst"><code>static int</code></td>
<td class="colLast"><code><strong><a href="../../../org/libjpegturbo/turbojpeg/TJ.html#PF_CMYK">PF_CMYK</a></strong></code>
<div class="block">CMYK pixel format.</div>
</td>
</tr>
-<tr class="altColor">
+<tr class="rowColor">
<td class="colFirst"><code>static int</code></td>
<td class="colLast"><code><strong><a href="../../../org/libjpegturbo/turbojpeg/TJ.html#PF_GRAY">PF_GRAY</a></strong></code>
<div class="block">Grayscale pixel format.</div>
</td>
</tr>
-<tr class="rowColor">
+<tr class="altColor">
<td class="colFirst"><code>static int</code></td>
<td class="colLast"><code><strong><a href="../../../org/libjpegturbo/turbojpeg/TJ.html#PF_RGB">PF_RGB</a></strong></code>
<div class="block">RGB pixel format.</div>
</td>
</tr>
-<tr class="altColor">
+<tr class="rowColor">
<td class="colFirst"><code>static int</code></td>
<td class="colLast"><code><strong><a href="../../../org/libjpegturbo/turbojpeg/TJ.html#PF_RGBA">PF_RGBA</a></strong></code>
<div class="block">RGBA pixel format.</div>
</td>
</tr>
-<tr class="rowColor">
+<tr class="altColor">
<td class="colFirst"><code>static int</code></td>
<td class="colLast"><code><strong><a href="../../../org/libjpegturbo/turbojpeg/TJ.html#PF_RGBX">PF_RGBX</a></strong></code>
<div class="block">RGBX pixel format.</div>
</td>
</tr>
-<tr class="altColor">
+<tr class="rowColor">
<td class="colFirst"><code>static int</code></td>
<td class="colLast"><code><strong><a href="../../../org/libjpegturbo/turbojpeg/TJ.html#PF_XBGR">PF_XBGR</a></strong></code>
<div class="block">XBGR pixel format.</div>
</td>
</tr>
-<tr class="rowColor">
+<tr class="altColor">
<td class="colFirst"><code>static int</code></td>
<td class="colLast"><code><strong><a href="../../../org/libjpegturbo/turbojpeg/TJ.html#PF_XRGB">PF_XRGB</a></strong></code>
<div class="block">XRGB pixel format.</div>
</td>
</tr>
-<tr class="altColor">
+<tr class="rowColor">
<td class="colFirst"><code>static int</code></td>
<td class="colLast"><code><strong><a href="../../../org/libjpegturbo/turbojpeg/TJ.html#SAMP_411">SAMP_411</a></strong></code>
<div class="block">4:1:1 chrominance subsampling.</div>
</td>
</tr>
-<tr class="rowColor">
+<tr class="altColor">
<td class="colFirst"><code>static int</code></td>
<td class="colLast"><code><strong><a href="../../../org/libjpegturbo/turbojpeg/TJ.html#SAMP_420">SAMP_420</a></strong></code>
<div class="block">4:2:0 chrominance subsampling.</div>
</td>
</tr>
-<tr class="altColor">
+<tr class="rowColor">
<td class="colFirst"><code>static int</code></td>
<td class="colLast"><code><strong><a href="../../../org/libjpegturbo/turbojpeg/TJ.html#SAMP_422">SAMP_422</a></strong></code>
<div class="block">4:2:2 chrominance subsampling.</div>
</td>
</tr>
-<tr class="rowColor">
+<tr class="altColor">
<td class="colFirst"><code>static int</code></td>
<td class="colLast"><code><strong><a href="../../../org/libjpegturbo/turbojpeg/TJ.html#SAMP_440">SAMP_440</a></strong></code>
<div class="block">4:4:0 chrominance subsampling.</div>
</td>
</tr>
-<tr class="altColor">
+<tr class="rowColor">
<td class="colFirst"><code>static int</code></td>
<td class="colLast"><code><strong><a href="../../../org/libjpegturbo/turbojpeg/TJ.html#SAMP_444">SAMP_444</a></strong></code>
<div class="block">4:4:4 chrominance subsampling (no chrominance subsampling).</div>
</td>
</tr>
-<tr class="rowColor">
+<tr class="altColor">
<td class="colFirst"><code>static int</code></td>
<td class="colLast"><code><strong><a href="../../../org/libjpegturbo/turbojpeg/TJ.html#SAMP_GRAY">SAMP_GRAY</a></strong></code>
<div class="block">Grayscale.</div>
@@ -938,7 +971,7 @@
<a name="FLAG_ACCURATEDCT">
<!-- -->
</a>
-<ul class="blockListLast">
+<ul class="blockList">
<li class="blockList">
<h4>FLAG_ACCURATEDCT</h4>
<pre>public static final int FLAG_ACCURATEDCT</pre>
@@ -952,6 +985,78 @@
<dl><dt><span class="strong">See Also:</span></dt><dd><a href="../../../constant-values.html#org.libjpegturbo.turbojpeg.TJ.FLAG_ACCURATEDCT">Constant Field Values</a></dd></dl>
</li>
</ul>
+<a name="FLAG_STOPONWARNING">
+<!-- -->
+</a>
+<ul class="blockList">
+<li class="blockList">
+<h4>FLAG_STOPONWARNING</h4>
+<pre>public static final int FLAG_STOPONWARNING</pre>
+<div class="block">Immediately discontinue the current compression/decompression/transform
+ operation if the underlying codec throws a warning (non-fatal error). The
+ default behavior is to allow the operation to complete unless a fatal
+ error is encountered.
+ <p>
+ NOTE: due to the design of the TurboJPEG Java API, only certain methods
+ (specifically, <a href="../../../org/libjpegturbo/turbojpeg/TJDecompressor.html" title="class in org.libjpegturbo.turbojpeg"><code>TJDecompressor.decompress*()</code></a> methods
+ with a void return type) will complete and leave the output image in a
+ fully recoverable state after a non-fatal error occurs.</div>
+<dl><dt><span class="strong">See Also:</span></dt><dd><a href="../../../constant-values.html#org.libjpegturbo.turbojpeg.TJ.FLAG_STOPONWARNING">Constant Field Values</a></dd></dl>
+</li>
+</ul>
+<a name="FLAG_PROGRESSIVE">
+<!-- -->
+</a>
+<ul class="blockList">
+<li class="blockList">
+<h4>FLAG_PROGRESSIVE</h4>
+<pre>public static final int FLAG_PROGRESSIVE</pre>
+<div class="block">Use progressive entropy coding in JPEG images generated by compression and
+ transform operations. Progressive entropy coding will generally improve
+ compression relative to baseline entropy coding (the default), but it will
+ reduce compression and decompression performance considerably.</div>
+<dl><dt><span class="strong">See Also:</span></dt><dd><a href="../../../constant-values.html#org.libjpegturbo.turbojpeg.TJ.FLAG_PROGRESSIVE">Constant Field Values</a></dd></dl>
+</li>
+</ul>
+<a name="NUMERR">
+<!-- -->
+</a>
+<ul class="blockList">
+<li class="blockList">
+<h4>NUMERR</h4>
+<pre>public static final int NUMERR</pre>
+<div class="block">The number of error codes</div>
+<dl><dt><span class="strong">See Also:</span></dt><dd><a href="../../../constant-values.html#org.libjpegturbo.turbojpeg.TJ.NUMERR">Constant Field Values</a></dd></dl>
+</li>
+</ul>
+<a name="ERR_WARNING">
+<!-- -->
+</a>
+<ul class="blockList">
+<li class="blockList">
+<h4>ERR_WARNING</h4>
+<pre>public static final int ERR_WARNING</pre>
+<div class="block">The error was non-fatal and recoverable, but the image may still be
+ corrupt.
+ <p>
+ NOTE: due to the design of the TurboJPEG Java API, only certain methods
+ (specifically, <a href="../../../org/libjpegturbo/turbojpeg/TJDecompressor.html" title="class in org.libjpegturbo.turbojpeg"><code>TJDecompressor.decompress*()</code></a> methods
+ with a void return type) will complete and leave the output image in a
+ fully recoverable state after a non-fatal error occurs.</div>
+<dl><dt><span class="strong">See Also:</span></dt><dd><a href="../../../constant-values.html#org.libjpegturbo.turbojpeg.TJ.ERR_WARNING">Constant Field Values</a></dd></dl>
+</li>
+</ul>
+<a name="ERR_FATAL">
+<!-- -->
+</a>
+<ul class="blockListLast">
+<li class="blockList">
+<h4>ERR_FATAL</h4>
+<pre>public static final int ERR_FATAL</pre>
+<div class="block">The error was fatal and non-recoverable.</div>
+<dl><dt><span class="strong">See Also:</span></dt><dd><a href="../../../constant-values.html#org.libjpegturbo.turbojpeg.TJ.ERR_FATAL">Constant Field Values</a></dd></dl>
+</li>
+</ul>
</li>
</ul>
<!-- ========= CONSTRUCTOR DETAIL ======== -->
diff --git a/java/doc/org/libjpegturbo/turbojpeg/TJDecompressor.html b/java/doc/org/libjpegturbo/turbojpeg/TJDecompressor.html
index a914de9..8e33249 100644
--- a/java/doc/org/libjpegturbo/turbojpeg/TJDecompressor.html
+++ b/java/doc/org/libjpegturbo/turbojpeg/TJDecompressor.html
@@ -785,7 +785,11 @@
throws <a href="../../../org/libjpegturbo/turbojpeg/TJException.html" title="class in org.libjpegturbo.turbojpeg">TJException</a></pre>
<div class="block">Decompress the JPEG source image or decode the YUV source image associated
with this decompressor instance and output a grayscale, RGB, or CMYK image
- to the given destination buffer.</div>
+ to the given destination buffer.
+ <p>
+ NOTE: The output image is fully recoverable if this method throws a
+ non-fatal <a href="../../../org/libjpegturbo/turbojpeg/TJException.html" title="class in org.libjpegturbo.turbojpeg"><code>TJException</code></a> (unless
+ <a href="../../../org/libjpegturbo/turbojpeg/TJ.html#FLAG_STOPONWARNING"><code>TJ.FLAG_STOPONWARNING</code></a> is specified.)</div>
<dl><dt><span class="strong">Parameters:</span></dt><dd><code>dstBuf</code> - buffer that will receive the decompressed/decoded image.
If the source image is a JPEG image, then this buffer should normally be
<code>pitch * scaledHeight</code> bytes in size, where
@@ -895,7 +899,11 @@
<code>YUVImage</code> instance. This method performs JPEG decompression
but leaves out the color conversion step, so a planar YUV image is
generated instead of an RGB or grayscale image. This method cannot be
- used to decompress JPEG source images with the CMYK or YCCK colorspace.</div>
+ used to decompress JPEG source images with the CMYK or YCCK colorspace.
+ <p>
+ NOTE: The YUV planar output image is fully recoverable if this method
+ throws a non-fatal <a href="../../../org/libjpegturbo/turbojpeg/TJException.html" title="class in org.libjpegturbo.turbojpeg"><code>TJException</code></a> (unless
+ <a href="../../../org/libjpegturbo/turbojpeg/TJ.html#FLAG_STOPONWARNING"><code>TJ.FLAG_STOPONWARNING</code></a> is specified.)</div>
<dl><dt><span class="strong">Parameters:</span></dt><dd><code>dstImage</code> - <a href="../../../org/libjpegturbo/turbojpeg/YUVImage.html" title="class in org.libjpegturbo.turbojpeg"><code>YUVImage</code></a> instance that will receive the YUV planar
image. The level of subsampling specified in this <code>YUVImage</code>
instance must match that of the JPEG image, and the width and height
@@ -1035,7 +1043,11 @@
throws <a href="../../../org/libjpegturbo/turbojpeg/TJException.html" title="class in org.libjpegturbo.turbojpeg">TJException</a></pre>
<div class="block">Decompress the JPEG source image or decode the YUV source image associated
with this decompressor instance and output a grayscale, RGB, or CMYK image
- to the given destination buffer.</div>
+ to the given destination buffer.
+ <p>
+ NOTE: The output image is fully recoverable if this method throws a
+ non-fatal <a href="../../../org/libjpegturbo/turbojpeg/TJException.html" title="class in org.libjpegturbo.turbojpeg"><code>TJException</code></a> (unless
+ <a href="../../../org/libjpegturbo/turbojpeg/TJ.html#FLAG_STOPONWARNING"><code>TJ.FLAG_STOPONWARNING</code></a> is specified.)</div>
<dl><dt><span class="strong">Parameters:</span></dt><dd><code>dstBuf</code> - buffer that will receive the decompressed/decoded image.
If the source image is a JPEG image, then this buffer should normally be
<code>stride * scaledHeight</code> pixels in size, where
@@ -1092,7 +1104,11 @@
throws <a href="../../../org/libjpegturbo/turbojpeg/TJException.html" title="class in org.libjpegturbo.turbojpeg">TJException</a></pre>
<div class="block">Decompress the JPEG source image or decode the YUV source image associated
with this decompressor instance and output a decompressed/decoded image to
- the given <code>BufferedImage</code> instance.</div>
+ the given <code>BufferedImage</code> instance.
+ <p>
+ NOTE: The output image is fully recoverable if this method throws a
+ non-fatal <a href="../../../org/libjpegturbo/turbojpeg/TJException.html" title="class in org.libjpegturbo.turbojpeg"><code>TJException</code></a> (unless
+ <a href="../../../org/libjpegturbo/turbojpeg/TJ.html#FLAG_STOPONWARNING"><code>TJ.FLAG_STOPONWARNING</code></a> is specified.)</div>
<dl><dt><span class="strong">Parameters:</span></dt><dd><code>dstImage</code> - a <code>BufferedImage</code> instance that will receive
the decompressed/decoded image. If the source image is a JPEG image, then
the width and height of the <code>BufferedImage</code> instance must match
diff --git a/java/doc/org/libjpegturbo/turbojpeg/TJException.html b/java/doc/org/libjpegturbo/turbojpeg/TJException.html
index 6088066..a57b351 100644
--- a/java/doc/org/libjpegturbo/turbojpeg/TJException.html
+++ b/java/doc/org/libjpegturbo/turbojpeg/TJException.html
@@ -60,13 +60,13 @@
<li>Nested | </li>
<li>Field | </li>
<li><a href="#constructor_summary">Constr</a> | </li>
-<li><a href="#methods_inherited_from_class_java.lang.Throwable">Method</a></li>
+<li><a href="#method_summary">Method</a></li>
</ul>
<ul class="subNavList">
<li>Detail: </li>
<li>Field | </li>
<li><a href="#constructor_detail">Constr</a> | </li>
-<li>Method</li>
+<li><a href="#method_detail">Method</a></li>
</ul>
</div>
<a name="skip-navbar_top">
@@ -138,10 +138,14 @@
<td class="colOne"><code><strong><a href="../../../org/libjpegturbo/turbojpeg/TJException.html#TJException(java.lang.String)">TJException</a></strong>(java.lang.String message)</code> </td>
</tr>
<tr class="altColor">
+<td class="colOne"><code><strong><a href="../../../org/libjpegturbo/turbojpeg/TJException.html#TJException(java.lang.String,%20int)">TJException</a></strong>(java.lang.String message,
+ int code)</code> </td>
+</tr>
+<tr class="rowColor">
<td class="colOne"><code><strong><a href="../../../org/libjpegturbo/turbojpeg/TJException.html#TJException(java.lang.String,%20java.lang.Throwable)">TJException</a></strong>(java.lang.String message,
java.lang.Throwable cause)</code> </td>
</tr>
-<tr class="rowColor">
+<tr class="altColor">
<td class="colOne"><code><strong><a href="../../../org/libjpegturbo/turbojpeg/TJException.html#TJException(java.lang.Throwable)">TJException</a></strong>(java.lang.Throwable cause)</code> </td>
</tr>
</table>
@@ -153,6 +157,20 @@
<!-- -->
</a>
<h3>Method Summary</h3>
+<table class="overviewSummary" border="0" cellpadding="3" cellspacing="0" summary="Method Summary table, listing methods, and an explanation">
+<caption><span>Methods</span><span class="tabEnd"> </span></caption>
+<tr>
+<th class="colFirst" scope="col">Modifier and Type</th>
+<th class="colLast" scope="col">Method and Description</th>
+</tr>
+<tr class="altColor">
+<td class="colFirst"><code>int</code></td>
+<td class="colLast"><code><strong><a href="../../../org/libjpegturbo/turbojpeg/TJException.html#getErrorCode()">getErrorCode</a></strong>()</code>
+<div class="block">Returns a code (one of <a href="../../../org/libjpegturbo/turbojpeg/TJ.html" title="class in org.libjpegturbo.turbojpeg"><code>TJ.ERR_*</code></a>) indicating the severity of the
+ last error.</div>
+</td>
+</tr>
+</table>
<ul class="blockList">
<li class="blockList"><a name="methods_inherited_from_class_java.lang.Throwable">
<!-- -->
@@ -209,6 +227,16 @@
<pre>public TJException(java.lang.String message)</pre>
</li>
</ul>
+<a name="TJException(java.lang.String, int)">
+<!-- -->
+</a>
+<ul class="blockList">
+<li class="blockList">
+<h4>TJException</h4>
+<pre>public TJException(java.lang.String message,
+ int code)</pre>
+</li>
+</ul>
<a name="TJException(java.lang.Throwable)">
<!-- -->
</a>
@@ -220,6 +248,27 @@
</ul>
</li>
</ul>
+<!-- ============ METHOD DETAIL ========== -->
+<ul class="blockList">
+<li class="blockList"><a name="method_detail">
+<!-- -->
+</a>
+<h3>Method Detail</h3>
+<a name="getErrorCode()">
+<!-- -->
+</a>
+<ul class="blockListLast">
+<li class="blockList">
+<h4>getErrorCode</h4>
+<pre>public int getErrorCode()</pre>
+<div class="block">Returns a code (one of <a href="../../../org/libjpegturbo/turbojpeg/TJ.html" title="class in org.libjpegturbo.turbojpeg"><code>TJ.ERR_*</code></a>) indicating the severity of the
+ last error.</div>
+<dl><dt><span class="strong">Returns:</span></dt><dd>a code (one of <a href="../../../org/libjpegturbo/turbojpeg/TJ.html" title="class in org.libjpegturbo.turbojpeg"><code>TJ.ERR_*</code></a>) indicating the severity of the
+ last error.</dd></dl>
+</li>
+</ul>
+</li>
+</ul>
</li>
</ul>
</div>
@@ -270,13 +319,13 @@
<li>Nested | </li>
<li>Field | </li>
<li><a href="#constructor_summary">Constr</a> | </li>
-<li><a href="#methods_inherited_from_class_java.lang.Throwable">Method</a></li>
+<li><a href="#method_summary">Method</a></li>
</ul>
<ul class="subNavList">
<li>Detail: </li>
<li>Field | </li>
<li><a href="#constructor_detail">Constr</a> | </li>
-<li>Method</li>
+<li><a href="#method_detail">Method</a></li>
</ul>
</div>
<a name="skip-navbar_bottom">
diff --git a/java/doc/org/libjpegturbo/turbojpeg/TJTransform.html b/java/doc/org/libjpegturbo/turbojpeg/TJTransform.html
index cf65bd2..86fafa3 100644
--- a/java/doc/org/libjpegturbo/turbojpeg/TJTransform.html
+++ b/java/doc/org/libjpegturbo/turbojpeg/TJTransform.html
@@ -217,31 +217,45 @@
</tr>
<tr class="rowColor">
<td class="colFirst"><code>static int</code></td>
+<td class="colLast"><code><strong><a href="../../../org/libjpegturbo/turbojpeg/TJTransform.html#OPT_COPYNONE">OPT_COPYNONE</a></strong></code>
+<div class="block">This option will prevent <a href="../../../org/libjpegturbo/turbojpeg/TJTransformer.html#transform(byte[][],%20org.libjpegturbo.turbojpeg.TJTransform[],%20int)"><code>TJTransformer.transform()</code></a> from copying any extra markers (including EXIF
+ and ICC profile data) from the source image to the output image.</div>
+</td>
+</tr>
+<tr class="altColor">
+<td class="colFirst"><code>static int</code></td>
<td class="colLast"><code><strong><a href="../../../org/libjpegturbo/turbojpeg/TJTransform.html#OPT_CROP">OPT_CROP</a></strong></code>
<div class="block">This option will enable lossless cropping.</div>
</td>
</tr>
-<tr class="altColor">
+<tr class="rowColor">
<td class="colFirst"><code>static int</code></td>
<td class="colLast"><code><strong><a href="../../../org/libjpegturbo/turbojpeg/TJTransform.html#OPT_GRAY">OPT_GRAY</a></strong></code>
<div class="block">This option will discard the color data in the input image and produce
a grayscale output image.</div>
</td>
</tr>
-<tr class="rowColor">
+<tr class="altColor">
<td class="colFirst"><code>static int</code></td>
<td class="colLast"><code><strong><a href="../../../org/libjpegturbo/turbojpeg/TJTransform.html#OPT_NOOUTPUT">OPT_NOOUTPUT</a></strong></code>
<div class="block">This option will prevent <a href="../../../org/libjpegturbo/turbojpeg/TJTransformer.html#transform(byte[][],%20org.libjpegturbo.turbojpeg.TJTransform[],%20int)"><code>TJTransformer.transform()</code></a> from outputting a JPEG image for this
particular transform.</div>
</td>
</tr>
-<tr class="altColor">
+<tr class="rowColor">
<td class="colFirst"><code>static int</code></td>
<td class="colLast"><code><strong><a href="../../../org/libjpegturbo/turbojpeg/TJTransform.html#OPT_PERFECT">OPT_PERFECT</a></strong></code>
<div class="block">This option will cause <a href="../../../org/libjpegturbo/turbojpeg/TJTransformer.html#transform(byte[][],%20org.libjpegturbo.turbojpeg.TJTransform[],%20int)"><code>TJTransformer.transform()</code></a> to throw an exception if the transform is not
perfect.</div>
</td>
</tr>
+<tr class="altColor">
+<td class="colFirst"><code>static int</code></td>
+<td class="colLast"><code><strong><a href="../../../org/libjpegturbo/turbojpeg/TJTransform.html#OPT_PROGRESSIVE">OPT_PROGRESSIVE</a></strong></code>
+<div class="block">This option will enable progressive entropy coding in the output image
+ generated by this particular transform.</div>
+</td>
+</tr>
<tr class="rowColor">
<td class="colFirst"><code>static int</code></td>
<td class="colLast"><code><strong><a href="../../../org/libjpegturbo/turbojpeg/TJTransform.html#OPT_TRIM">OPT_TRIM</a></strong></code>
@@ -548,6 +562,33 @@
<dl><dt><span class="strong">See Also:</span></dt><dd><a href="../../../constant-values.html#org.libjpegturbo.turbojpeg.TJTransform.OPT_NOOUTPUT">Constant Field Values</a></dd></dl>
</li>
</ul>
+<a name="OPT_PROGRESSIVE">
+<!-- -->
+</a>
+<ul class="blockList">
+<li class="blockList">
+<h4>OPT_PROGRESSIVE</h4>
+<pre>public static final int OPT_PROGRESSIVE</pre>
+<div class="block">This option will enable progressive entropy coding in the output image
+ generated by this particular transform. Progressive entropy coding will
+ generally improve compression relative to baseline entropy coding (the
+ default), but it will reduce compression and decompression performance
+ considerably.</div>
+<dl><dt><span class="strong">See Also:</span></dt><dd><a href="../../../constant-values.html#org.libjpegturbo.turbojpeg.TJTransform.OPT_PROGRESSIVE">Constant Field Values</a></dd></dl>
+</li>
+</ul>
+<a name="OPT_COPYNONE">
+<!-- -->
+</a>
+<ul class="blockList">
+<li class="blockList">
+<h4>OPT_COPYNONE</h4>
+<pre>public static final int OPT_COPYNONE</pre>
+<div class="block">This option will prevent <a href="../../../org/libjpegturbo/turbojpeg/TJTransformer.html#transform(byte[][],%20org.libjpegturbo.turbojpeg.TJTransform[],%20int)"><code>TJTransformer.transform()</code></a> from copying any extra markers (including EXIF
+ and ICC profile data) from the source image to the output image.</div>
+<dl><dt><span class="strong">See Also:</span></dt><dd><a href="../../../constant-values.html#org.libjpegturbo.turbojpeg.TJTransform.OPT_COPYNONE">Constant Field Values</a></dd></dl>
+</li>
+</ul>
<a name="op">
<!-- -->
</a>
diff --git a/java/doc/script.js b/java/doc/script.js
new file mode 100644
index 0000000..b346356
--- /dev/null
+++ b/java/doc/script.js
@@ -0,0 +1,30 @@
+function show(type)
+{
+ count = 0;
+ for (var key in methods) {
+ var row = document.getElementById(key);
+ if ((methods[key] & type) != 0) {
+ row.style.display = '';
+ row.className = (count++ % 2) ? rowColor : altColor;
+ }
+ else
+ row.style.display = 'none';
+ }
+ updateTabs(type);
+}
+
+function updateTabs(type)
+{
+ for (var value in tabs) {
+ var sNode = document.getElementById(tabs[value][0]);
+ var spanNode = sNode.firstChild;
+ if (value == type) {
+ sNode.className = activeTableTab;
+ spanNode.innerHTML = tabs[value][1];
+ }
+ else {
+ sNode.className = tableTab;
+ spanNode.innerHTML = "<a href=\"javascript:show("+ value + ");\">" + tabs[value][1] + "</a>";
+ }
+ }
+}
diff --git a/java/doc/serialized-form.html b/java/doc/serialized-form.html
index 846cabc..9ba3fd9 100644
--- a/java/doc/serialized-form.html
+++ b/java/doc/serialized-form.html
@@ -74,6 +74,19 @@
<dt>serialVersionUID:</dt>
<dd>1L</dd>
</dl>
+<ul class="blockList">
+<li class="blockList"><a name="serializedForm">
+<!-- -->
+</a>
+<h3>Serialized Fields</h3>
+<ul class="blockList">
+<li class="blockListLast">
+<h4>errorCode</h4>
+<pre>int errorCode</pre>
+</li>
+</ul>
+</li>
+</ul>
</li>
<li class="blockList"><a name="org.libjpegturbo.turbojpeg.TJTransform">
<!-- -->
diff --git a/java/org/libjpegturbo/turbojpeg/TJ.java b/java/org/libjpegturbo/turbojpeg/TJ.java
index 02d14c0..ee000a9 100644
--- a/java/org/libjpegturbo/turbojpeg/TJ.java
+++ b/java/org/libjpegturbo/turbojpeg/TJ.java
@@ -1,5 +1,5 @@
/*
- * Copyright (C)2011-2013 D. R. Commander. All Rights Reserved.
+ * Copyright (C)2011-2013, 2017 D. R. Commander. All Rights Reserved.
* Copyright (C)2015 Viktor Szathmáry. All Rights Reserved.
*
* Redistribution and use in source and binary forms, with or without
@@ -348,16 +348,16 @@
* The uncompressed source/destination image is stored in bottom-up (Windows,
* OpenGL) order, not top-down (X11) order.
*/
- public static final int FLAG_BOTTOMUP = 2;
+ public static final int FLAG_BOTTOMUP = 2;
@Deprecated
- public static final int FLAG_FORCEMMX = 8;
+ public static final int FLAG_FORCEMMX = 8;
@Deprecated
- public static final int FLAG_FORCESSE = 16;
+ public static final int FLAG_FORCESSE = 16;
@Deprecated
- public static final int FLAG_FORCESSE2 = 32;
+ public static final int FLAG_FORCESSE2 = 32;
@Deprecated
- public static final int FLAG_FORCESSE3 = 128;
+ public static final int FLAG_FORCESSE3 = 128;
/**
* When decompressing an image that was compressed using chrominance
@@ -366,7 +366,7 @@
* creates a smooth transition between neighboring chrominance components in
* order to reduce upsampling artifacts in the decompressed image.
*/
- public static final int FLAG_FASTUPSAMPLE = 256;
+ public static final int FLAG_FASTUPSAMPLE = 256;
/**
* Use the fastest DCT/IDCT algorithm available in the underlying codec. The
* default if this flag is not specified is implementation-specific. For
@@ -375,7 +375,7 @@
* only a very slight effect on accuracy, but it uses the accurate algorithm
* when decompressing, because this has been shown to have a larger effect.
*/
- public static final int FLAG_FASTDCT = 2048;
+ public static final int FLAG_FASTDCT = 2048;
/**
* Use the most accurate DCT/IDCT algorithm available in the underlying
* codec. The default if this flag is not specified is
@@ -385,7 +385,46 @@
* but it uses the accurate algorithm when decompressing, because this has
* been shown to have a larger effect.
*/
- public static final int FLAG_ACCURATEDCT = 4096;
+ public static final int FLAG_ACCURATEDCT = 4096;
+ /**
+ * Immediately discontinue the current compression/decompression/transform
+ * operation if the underlying codec throws a warning (non-fatal error). The
+ * default behavior is to allow the operation to complete unless a fatal
+ * error is encountered.
+ * <p>
+ * NOTE: due to the design of the TurboJPEG Java API, only certain methods
+ * (specifically, {@link TJDecompressor TJDecompressor.decompress*()} methods
+ * with a void return type) will complete and leave the output image in a
+ * fully recoverable state after a non-fatal error occurs.
+ */
+ public static final int FLAG_STOPONWARNING = 8192;
+ /**
+ * Use progressive entropy coding in JPEG images generated by compression and
+ * transform operations. Progressive entropy coding will generally improve
+ * compression relative to baseline entropy coding (the default), but it will
+ * reduce compression and decompression performance considerably.
+ */
+ public static final int FLAG_PROGRESSIVE = 16384;
+
+
+ /**
+ * The number of error codes
+ */
+ public static final int NUMERR = 2;
+ /**
+ * The error was non-fatal and recoverable, but the image may still be
+ * corrupt.
+ * <p>
+ * NOTE: due to the design of the TurboJPEG Java API, only certain methods
+ * (specifically, {@link TJDecompressor TJDecompressor.decompress*()} methods
+ * with a void return type) will complete and leave the output image in a
+ * fully recoverable state after a non-fatal error occurs.
+ */
+ public static final int ERR_WARNING = 0;
+ /**
+ * The error was fatal and non-recoverable.
+ */
+ public static final int ERR_FATAL = 1;
/**
diff --git a/java/org/libjpegturbo/turbojpeg/TJDecompressor.java b/java/org/libjpegturbo/turbojpeg/TJDecompressor.java
index bd0e694..be71cb0 100644
--- a/java/org/libjpegturbo/turbojpeg/TJDecompressor.java
+++ b/java/org/libjpegturbo/turbojpeg/TJDecompressor.java
@@ -308,6 +308,10 @@
* Decompress the JPEG source image or decode the YUV source image associated
* with this decompressor instance and output a grayscale, RGB, or CMYK image
* to the given destination buffer.
+ * <p>
+ * NOTE: The output image is fully recoverable if this method throws a
+ * non-fatal {@link TJException} (unless
+ * {@link TJ#FLAG_STOPONWARNING TJ.FLAG_STOPONWARNING} is specified.)
*
* @param dstBuf buffer that will receive the decompressed/decoded image.
* If the source image is a JPEG image, then this buffer should normally be
@@ -451,6 +455,10 @@
* but leaves out the color conversion step, so a planar YUV image is
* generated instead of an RGB or grayscale image. This method cannot be
* used to decompress JPEG source images with the CMYK or YCCK colorspace.
+ * <p>
+ * NOTE: The YUV planar output image is fully recoverable if this method
+ * throws a non-fatal {@link TJException} (unless
+ * {@link TJ#FLAG_STOPONWARNING TJ.FLAG_STOPONWARNING} is specified.)
*
* @param dstImage {@link YUVImage} instance that will receive the YUV planar
* image. The level of subsampling specified in this <code>YUVImage</code>
@@ -618,6 +626,10 @@
* Decompress the JPEG source image or decode the YUV source image associated
* with this decompressor instance and output a grayscale, RGB, or CMYK image
* to the given destination buffer.
+ * <p>
+ * NOTE: The output image is fully recoverable if this method throws a
+ * non-fatal {@link TJException} (unless
+ * {@link TJ#FLAG_STOPONWARNING TJ.FLAG_STOPONWARNING} is specified.)
*
* @param dstBuf buffer that will receive the decompressed/decoded image.
* If the source image is a JPEG image, then this buffer should normally be
@@ -699,6 +711,10 @@
* Decompress the JPEG source image or decode the YUV source image associated
* with this decompressor instance and output a decompressed/decoded image to
* the given <code>BufferedImage</code> instance.
+ * <p>
+ * NOTE: The output image is fully recoverable if this method throws a
+ * non-fatal {@link TJException} (unless
+ * {@link TJ#FLAG_STOPONWARNING TJ.FLAG_STOPONWARNING} is specified.)
*
* @param dstImage a <code>BufferedImage</code> instance that will receive
* the decompressed/decoded image. If the source image is a JPEG image, then
diff --git a/java/org/libjpegturbo/turbojpeg/TJException.java b/java/org/libjpegturbo/turbojpeg/TJException.java
index 59c2041..97659d4 100644
--- a/java/org/libjpegturbo/turbojpeg/TJException.java
+++ b/java/org/libjpegturbo/turbojpeg/TJException.java
@@ -1,5 +1,6 @@
/*
* Copyright (C)2015 Viktor Szathmáry. All Rights Reserved.
+ * Copyright (C)2017 D. R. Commander. All Rights Reserved.
*
* Redistribution and use in source and binary forms, with or without
* modification, are permitted provided that the following conditions are met:
@@ -46,8 +47,26 @@
super(message);
}
+ public TJException(String message, int code) {
+ super(message);
+ if (errorCode >= 0 && errorCode < TJ.NUMERR)
+ errorCode = code;
+ }
+
public TJException(Throwable cause) {
super(cause);
}
+ /**
+ * Returns a code (one of {@link TJ TJ.ERR_*}) indicating the severity of the
+ * last error.
+ *
+ * @return a code (one of {@link TJ TJ.ERR_*}) indicating the severity of the
+ * last error.
+ */
+ public int getErrorCode() {
+ return errorCode;
+ }
+
+ private int errorCode = TJ.ERR_FATAL;
}
diff --git a/java/org/libjpegturbo/turbojpeg/TJLoader.java.tmpl b/java/org/libjpegturbo/turbojpeg/TJLoader-unix.java.in
similarity index 77%
rename from java/org/libjpegturbo/turbojpeg/TJLoader.java.tmpl
rename to java/org/libjpegturbo/turbojpeg/TJLoader-unix.java.in
index 5ef3118..65884e8 100644
--- a/java/org/libjpegturbo/turbojpeg/TJLoader.java.tmpl
+++ b/java/org/libjpegturbo/turbojpeg/TJLoader-unix.java.in
@@ -1,5 +1,5 @@
/*
- * Copyright (C)2011-2013 D. R. Commander. All Rights Reserved.
+ * Copyright (C)2011-2013, 2016 D. R. Commander. All Rights Reserved.
*
* Redistribution and use in source and binary forms, with or without
* modification, are permitted provided that the following conditions are met:
@@ -36,19 +36,19 @@
String os = System.getProperty("os.name").toLowerCase();
if (os.indexOf("mac") >= 0) {
try {
- System.load("%{__libdir}/libturbojpeg.jnilib");
+ System.load("@CMAKE_INSTALL_FULL_LIBDIR@/libturbojpeg.jnilib");
} catch (java.lang.UnsatisfiedLinkError e2) {
System.load("/usr/lib/libturbojpeg.jnilib");
}
} else {
try {
- System.load("%{__libdir}/libturbojpeg.so");
+ System.load("@CMAKE_INSTALL_FULL_LIBDIR@/libturbojpeg.so");
} catch (java.lang.UnsatisfiedLinkError e3) {
- String libdir = "%{__libdir}";
- if (libdir.equals("/opt/libjpeg-turbo/lib64")) {
- System.load("/opt/libjpeg-turbo/lib32/libturbojpeg.so");
- } else if (libdir.equals("/opt/libjpeg-turbo/lib32")) {
- System.load("/opt/libjpeg-turbo/lib64/libturbojpeg.so");
+ String libdir = "@CMAKE_INSTALL_FULL_LIBDIR@";
+ if (libdir.equals("@CMAKE_INSTALL_DEFAULT_PREFIX@/lib64")) {
+ System.load("@CMAKE_INSTALL_DEFAULT_PREFIX@/lib32/libturbojpeg.so");
+ } else if (libdir.equals("@CMAKE_INSTALL_DEFAULT_PREFIX@/lib32")) {
+ System.load("@CMAKE_INSTALL_DEFAULT_PREFIX@/lib64/libturbojpeg.so");
} else {
throw e3;
}
diff --git a/java/org/libjpegturbo/turbojpeg/TJLoader.java.in b/java/org/libjpegturbo/turbojpeg/TJLoader-win.java.in
similarity index 100%
rename from java/org/libjpegturbo/turbojpeg/TJLoader.java.in
rename to java/org/libjpegturbo/turbojpeg/TJLoader-win.java.in
diff --git a/java/org/libjpegturbo/turbojpeg/TJTransform.java b/java/org/libjpegturbo/turbojpeg/TJTransform.java
index 7381f36..1b10d3d 100644
--- a/java/org/libjpegturbo/turbojpeg/TJTransform.java
+++ b/java/org/libjpegturbo/turbojpeg/TJTransform.java
@@ -103,21 +103,21 @@
* partial MCU blocks that cannot be transformed will be left in place, which
* will create odd-looking strips on the right or bottom edge of the image.
*/
- public static final int OPT_PERFECT = 1;
+ public static final int OPT_PERFECT = 1;
/**
* This option will discard any partial MCU blocks that cannot be
* transformed.
*/
- public static final int OPT_TRIM = 2;
+ public static final int OPT_TRIM = 2;
/**
* This option will enable lossless cropping.
*/
- public static final int OPT_CROP = 4;
+ public static final int OPT_CROP = 4;
/**
* This option will discard the color data in the input image and produce
* a grayscale output image.
*/
- public static final int OPT_GRAY = 8;
+ public static final int OPT_GRAY = 8;
/**
* This option will prevent {@link TJTransformer#transform
* TJTransformer.transform()} from outputting a JPEG image for this
@@ -125,7 +125,21 @@
* filter to capture the transformed DCT coefficients without transcoding
* them.
*/
- public static final int OPT_NOOUTPUT = 16;
+ public static final int OPT_NOOUTPUT = 16;
+ /**
+ * This option will enable progressive entropy coding in the output image
+ * generated by this particular transform. Progressive entropy coding will
+ * generally improve compression relative to baseline entropy coding (the
+ * default), but it will reduce compression and decompression performance
+ * considerably.
+ */
+ public static final int OPT_PROGRESSIVE = 32;
+ /**
+ * This option will prevent {@link TJTransformer#transform
+ * TJTransformer.transform()} from copying any extra markers (including EXIF
+ * and ICC profile data) from the source image to the output image.
+ */
+ public static final int OPT_COPYNONE = 64;
/**
diff --git a/jcicc.c b/jcicc.c
new file mode 100644
index 0000000..0613779
--- /dev/null
+++ b/jcicc.c
@@ -0,0 +1,105 @@
+/*
+ * jcicc.c
+ *
+ * Copyright (C) 1997-1998, Thomas G. Lane, Todd Newman.
+ * Copyright (C) 2017, D. R. Commander.
+ * For conditions of distribution and use, see the accompanying README.ijg
+ * file.
+ *
+ * This file provides code to write International Color Consortium (ICC) device
+ * profiles embedded in JFIF JPEG image files. The ICC has defined a standard
+ * for including such data in JPEG "APP2" markers. The code given here does
+ * not know anything about the internal structure of the ICC profile data; it
+ * just knows how to embed the profile data in a JPEG file while writing it.
+ */
+
+#define JPEG_INTERNALS
+#include "jinclude.h"
+#include "jpeglib.h"
+#include "jerror.h"
+
+
+/*
+ * Since an ICC profile can be larger than the maximum size of a JPEG marker
+ * (64K), we need provisions to split it into multiple markers. The format
+ * defined by the ICC specifies one or more APP2 markers containing the
+ * following data:
+ * Identifying string ASCII "ICC_PROFILE\0" (12 bytes)
+ * Marker sequence number 1 for first APP2, 2 for next, etc (1 byte)
+ * Number of markers Total number of APP2's used (1 byte)
+ * Profile data (remainder of APP2 data)
+ * Decoders should use the marker sequence numbers to reassemble the profile,
+ * rather than assuming that the APP2 markers appear in the correct sequence.
+ */
+
+#define ICC_MARKER (JPEG_APP0 + 2) /* JPEG marker code for ICC */
+#define ICC_OVERHEAD_LEN 14 /* size of non-profile data in APP2 */
+#define MAX_BYTES_IN_MARKER 65533 /* maximum data len of a JPEG marker */
+#define MAX_DATA_BYTES_IN_MARKER (MAX_BYTES_IN_MARKER - ICC_OVERHEAD_LEN)
+
+
+/*
+ * This routine writes the given ICC profile data into a JPEG file. It *must*
+ * be called AFTER calling jpeg_start_compress() and BEFORE the first call to
+ * jpeg_write_scanlines(). (This ordering ensures that the APP2 marker(s) will
+ * appear after the SOI and JFIF or Adobe markers, but before all else.)
+ */
+
+GLOBAL(void)
+jpeg_write_icc_profile (j_compress_ptr cinfo, const JOCTET *icc_data_ptr,
+ unsigned int icc_data_len)
+{
+ unsigned int num_markers; /* total number of markers we'll write */
+ int cur_marker = 1; /* per spec, counting starts at 1 */
+ unsigned int length; /* number of bytes to write in this marker */
+
+ if (icc_data_ptr == NULL || icc_data_len == 0)
+ ERREXIT(cinfo, JERR_BUFFER_SIZE);
+ if (cinfo->global_state < CSTATE_SCANNING)
+ ERREXIT1(cinfo, JERR_BAD_STATE, cinfo->global_state);
+
+ /* Calculate the number of markers we'll need, rounding up of course */
+ num_markers = icc_data_len / MAX_DATA_BYTES_IN_MARKER;
+ if (num_markers * MAX_DATA_BYTES_IN_MARKER != icc_data_len)
+ num_markers++;
+
+ while (icc_data_len > 0) {
+ /* length of profile to put in this marker */
+ length = icc_data_len;
+ if (length > MAX_DATA_BYTES_IN_MARKER)
+ length = MAX_DATA_BYTES_IN_MARKER;
+ icc_data_len -= length;
+
+ /* Write the JPEG marker header (APP2 code and marker length) */
+ jpeg_write_m_header(cinfo, ICC_MARKER,
+ (unsigned int) (length + ICC_OVERHEAD_LEN));
+
+ /* Write the marker identifying string "ICC_PROFILE" (null-terminated). We
+ * code it in this less-than-transparent way so that the code works even if
+ * the local character set is not ASCII.
+ */
+ jpeg_write_m_byte(cinfo, 0x49);
+ jpeg_write_m_byte(cinfo, 0x43);
+ jpeg_write_m_byte(cinfo, 0x43);
+ jpeg_write_m_byte(cinfo, 0x5F);
+ jpeg_write_m_byte(cinfo, 0x50);
+ jpeg_write_m_byte(cinfo, 0x52);
+ jpeg_write_m_byte(cinfo, 0x4F);
+ jpeg_write_m_byte(cinfo, 0x46);
+ jpeg_write_m_byte(cinfo, 0x49);
+ jpeg_write_m_byte(cinfo, 0x4C);
+ jpeg_write_m_byte(cinfo, 0x45);
+ jpeg_write_m_byte(cinfo, 0x0);
+
+ /* Add the sequencing info */
+ jpeg_write_m_byte(cinfo, cur_marker);
+ jpeg_write_m_byte(cinfo, (int) num_markers);
+
+ /* Add the profile data */
+ while (length--) {
+ jpeg_write_m_byte(cinfo, *icc_data_ptr);
+ icc_data_ptr++;
+ }
+ cur_marker++;
+ }
+}
diff --git a/jconfig.h.in b/jconfig.h.in
index 02c12cc..3a47c18 100644
--- a/jconfig.h.in
+++ b/jconfig.h.in
@@ -1,19 +1,25 @@
/* Version ID for the JPEG library.
* Might be useful for tests like "#if JPEG_LIB_VERSION >= 60".
*/
-#define JPEG_LIB_VERSION 62 /* Version 6b */
+#define JPEG_LIB_VERSION @JPEG_LIB_VERSION@
/* libjpeg-turbo version */
-#define LIBJPEG_TURBO_VERSION 0
+#define LIBJPEG_TURBO_VERSION @VERSION@
/* libjpeg-turbo version in integer form */
-#define LIBJPEG_TURBO_VERSION_NUMBER 0
+#define LIBJPEG_TURBO_VERSION_NUMBER @LIBJPEG_TURBO_VERSION_NUMBER@
/* Support arithmetic encoding */
-#undef C_ARITH_CODING_SUPPORTED
+#cmakedefine C_ARITH_CODING_SUPPORTED
/* Support arithmetic decoding */
-#undef D_ARITH_CODING_SUPPORTED
+#cmakedefine D_ARITH_CODING_SUPPORTED
+
+/* Support in-memory source/destination managers */
+#cmakedefine MEM_SRCDST_SUPPORTED
+
+/* Use accelerated SIMD routines. */
+#cmakedefine WITH_SIMD
/*
* Define BITS_IN_JSAMPLE as either
@@ -24,50 +30,44 @@
* We do not support run-time selection of data precision, sorry.
*/
-#define BITS_IN_JSAMPLE 8 /* use 8 or 12 */
+#define BITS_IN_JSAMPLE @BITS_IN_JSAMPLE@ /* use 8 or 12 */
/* Define to 1 if you have the <locale.h> header file. */
-#undef HAVE_LOCALE_H
+#cmakedefine HAVE_LOCALE_H
/* Define to 1 if you have the <stddef.h> header file. */
-#undef HAVE_STDDEF_H
+#cmakedefine HAVE_STDDEF_H
/* Define to 1 if you have the <stdlib.h> header file. */
-#undef HAVE_STDLIB_H
+#cmakedefine HAVE_STDLIB_H
-/* Define to 1 if the system has the type `unsigned char'. */
-#undef HAVE_UNSIGNED_CHAR
-
-/* Define to 1 if the system has the type `unsigned short'. */
-#undef HAVE_UNSIGNED_SHORT
-
-/* Compiler does not support pointers to undefined structures. */
-#undef INCOMPLETE_TYPES_BROKEN
-
-/* Support in-memory source/destination managers */
-#undef MEM_SRCDST_SUPPORTED
+/* Define if you need to include <sys/types.h> to get size_t. */
+#cmakedefine NEED_SYS_TYPES_H
/* Define if you have BSD-like bzero and bcopy in <strings.h> rather than
memset/memcpy in <string.h>. */
-#undef NEED_BSD_STRINGS
+#cmakedefine NEED_BSD_STRINGS
-/* Define if you need to include <sys/types.h> to get size_t. */
-#undef NEED_SYS_TYPES_H
+/* Define to 1 if the system has the type `unsigned char'. */
+#cmakedefine HAVE_UNSIGNED_CHAR
+
+/* Define to 1 if the system has the type `unsigned short'. */
+#cmakedefine HAVE_UNSIGNED_SHORT
+
+/* Compiler does not support pointers to undefined structures. */
+#cmakedefine INCOMPLETE_TYPES_BROKEN
/* Define if your (broken) compiler shifts signed values as if they were
unsigned. */
-#undef RIGHT_SHIFT_IS_UNSIGNED
-
-/* Use accelerated SIMD routines. */
-#undef WITH_SIMD
+#cmakedefine RIGHT_SHIFT_IS_UNSIGNED
/* Define to 1 if type `char' is unsigned and you are not using gcc. */
#ifndef __CHAR_UNSIGNED__
-# undef __CHAR_UNSIGNED__
+ #cmakedefine __CHAR_UNSIGNED__
#endif
/* Define to empty if `const' does not conform to ANSI C. */
-#undef const
+/* #undef const */
/* Define to `unsigned int' if <sys/types.h> does not define. */
-#undef size_t
+/* #undef size_t */
diff --git a/jconfigint.h.in b/jconfigint.h.in
index 963e760..835f7c4 100644
--- a/jconfigint.h.in
+++ b/jconfigint.h.in
@@ -1,17 +1,21 @@
/* libjpeg-turbo build number */
-#undef BUILD
+#define BUILD "@BUILD@"
/* Compiler's inline keyword */
#undef inline
/* How to obtain function inlining. */
-#undef INLINE
+#define INLINE @INLINE@
/* Define to the full name of this package. */
-#undef PACKAGE_NAME
+#define PACKAGE_NAME "@CMAKE_PROJECT_NAME@"
/* Version number of package */
-#undef VERSION
+#define VERSION "@VERSION@"
+
+#ifndef _WIN32
/* The size of `size_t', as computed by sizeof. */
-#undef SIZEOF_SIZE_T
+#define SIZEOF_SIZE_T @SIZE_T@
+
+#endif
diff --git a/jdicc.c b/jdicc.c
new file mode 100644
index 0000000..b233736
--- /dev/null
+++ b/jdicc.c
@@ -0,0 +1,171 @@
+/*
+ * jdicc.c
+ *
+ * Copyright (C) 1997-1998, Thomas G. Lane, Todd Newman.
+ * Copyright (C) 2017, D. R. Commander.
+ * For conditions of distribution and use, see the accompanying README.ijg
+ * file.
+ *
+ * This file provides code to read International Color Consortium (ICC) device
+ * profiles embedded in JFIF JPEG image files. The ICC has defined a standard
+ * for including such data in JPEG "APP2" markers. The code given here does
+ * not know anything about the internal structure of the ICC profile data; it
+ * just knows how to get the profile data from a JPEG file while reading it.
+ */
+
+#define JPEG_INTERNALS
+#include "jinclude.h"
+#include "jpeglib.h"
+#include "jerror.h"
+
+#ifndef HAVE_STDLIB_H /* <stdlib.h> should declare malloc() */
+extern void *malloc (size_t size);
+#endif
+
+
+#define ICC_MARKER (JPEG_APP0 + 2) /* JPEG marker code for ICC */
+#define ICC_OVERHEAD_LEN 14 /* size of non-profile data in APP2 */
+
+
+/*
+ * Handy subroutine to test whether a saved marker is an ICC profile marker.
+ */
+
+LOCAL(boolean)
+marker_is_icc (jpeg_saved_marker_ptr marker)
+{
+ return
+ marker->marker == ICC_MARKER &&
+ marker->data_length >= ICC_OVERHEAD_LEN &&
+ /* verify the identifying string */
+ GETJOCTET(marker->data[0]) == 0x49 &&
+ GETJOCTET(marker->data[1]) == 0x43 &&
+ GETJOCTET(marker->data[2]) == 0x43 &&
+ GETJOCTET(marker->data[3]) == 0x5F &&
+ GETJOCTET(marker->data[4]) == 0x50 &&
+ GETJOCTET(marker->data[5]) == 0x52 &&
+ GETJOCTET(marker->data[6]) == 0x4F &&
+ GETJOCTET(marker->data[7]) == 0x46 &&
+ GETJOCTET(marker->data[8]) == 0x49 &&
+ GETJOCTET(marker->data[9]) == 0x4C &&
+ GETJOCTET(marker->data[10]) == 0x45 &&
+ GETJOCTET(marker->data[11]) == 0x0;
+}
+
+
+/*
+ * See if there was an ICC profile in the JPEG file being read; if so,
+ * reassemble and return the profile data.
+ *
+ * TRUE is returned if an ICC profile was found, FALSE if not. If TRUE is
+ * returned, *icc_data_ptr is set to point to the returned data, and
+ * *icc_data_len is set to its length.
+ *
+ * IMPORTANT: the data at *icc_data_ptr is allocated with malloc() and must be
+ * freed by the caller with free() when the caller no longer needs it.
+ * (Alternatively, we could write this routine to use the IJG library's memory
+ * allocator, so that the data would be freed implicitly when
+ * jpeg_finish_decompress() is called. But it seems likely that many
+ * applications will prefer to have the data stick around after decompression
+ * finishes.)
+ */
+
+GLOBAL(boolean)
+jpeg_read_icc_profile (j_decompress_ptr cinfo, JOCTET **icc_data_ptr,
+ unsigned int *icc_data_len)
+{
+ jpeg_saved_marker_ptr marker;
+ int num_markers = 0;
+ int seq_no;
+ JOCTET *icc_data;
+ unsigned int total_length;
+#define MAX_SEQ_NO 255 /* sufficient since marker numbers are bytes */
+ char marker_present[MAX_SEQ_NO+1]; /* 1 if marker found */
+ unsigned int data_length[MAX_SEQ_NO+1]; /* size of profile data in marker */
+ unsigned int data_offset[MAX_SEQ_NO+1]; /* offset for data in marker */
+
+ if (icc_data_ptr == NULL || icc_data_len == NULL)
+ ERREXIT(cinfo, JERR_BUFFER_SIZE);
+ if (cinfo->global_state < DSTATE_READY)
+ ERREXIT1(cinfo, JERR_BAD_STATE, cinfo->global_state);
+
+ *icc_data_ptr = NULL; /* avoid confusion if FALSE return */
+ *icc_data_len = 0;
+
+ /* This first pass over the saved markers discovers whether there are
+ * any ICC markers and verifies the consistency of the marker numbering.
+ */
+
+ for (seq_no = 1; seq_no <= MAX_SEQ_NO; seq_no++)
+ marker_present[seq_no] = 0;
+
+ for (marker = cinfo->marker_list; marker != NULL; marker = marker->next) {
+ if (marker_is_icc(marker)) {
+ if (num_markers == 0)
+ num_markers = GETJOCTET(marker->data[13]);
+ else if (num_markers != GETJOCTET(marker->data[13])) {
+ WARNMS(cinfo, JWRN_BOGUS_ICC); /* inconsistent num_markers fields */
+ return FALSE;
+ }
+ seq_no = GETJOCTET(marker->data[12]);
+ if (seq_no <= 0 || seq_no > num_markers) {
+ WARNMS(cinfo, JWRN_BOGUS_ICC); /* bogus sequence number */
+ return FALSE;
+ }
+ if (marker_present[seq_no]) {
+ WARNMS(cinfo, JWRN_BOGUS_ICC); /* duplicate sequence numbers */
+ return FALSE;
+ }
+ marker_present[seq_no] = 1;
+ data_length[seq_no] = marker->data_length - ICC_OVERHEAD_LEN;
+ }
+ }
+
+ if (num_markers == 0)
+ return FALSE;
+
+ /* Check for missing markers, count total space needed,
+ * compute offset of each marker's part of the data.
+ */
+
+ total_length = 0;
+ for (seq_no = 1; seq_no <= num_markers; seq_no++) {
+ if (marker_present[seq_no] == 0) {
+ WARNMS(cinfo, JWRN_BOGUS_ICC); /* missing sequence number */
+ return FALSE;
+ }
+ data_offset[seq_no] = total_length;
+ total_length += data_length[seq_no];
+ }
+
+ if (total_length == 0) {
+ WARNMS(cinfo, JWRN_BOGUS_ICC); /* found only empty markers? */
+ return FALSE;
+ }
+
+ /* Allocate space for assembled data */
+ icc_data = (JOCTET *) malloc(total_length * sizeof(JOCTET));
+ if (icc_data == NULL)
+ ERREXIT1(cinfo, JERR_OUT_OF_MEMORY, 11); /* oops, out of memory */
+
+ /* and fill it in */
+ for (marker = cinfo->marker_list; marker != NULL; marker = marker->next) {
+ if (marker_is_icc(marker)) {
+ JOCTET FAR *src_ptr;
+ JOCTET *dst_ptr;
+ unsigned int length;
+ seq_no = GETJOCTET(marker->data[12]);
+ dst_ptr = icc_data + data_offset[seq_no];
+ src_ptr = marker->data + ICC_OVERHEAD_LEN;
+ length = data_length[seq_no];
+ while (length--) {
+ *dst_ptr++ = *src_ptr++;
+ }
+ }
+ }
+
+ *icc_data_ptr = icc_data;
+ *icc_data_len = total_length;
+
+ return TRUE;
+}
diff --git a/jerror.h b/jerror.h
index 11a07cb..a79586f 100644
--- a/jerror.h
+++ b/jerror.h
@@ -5,7 +5,7 @@
* Copyright (C) 1994-1997, Thomas G. Lane.
* Modified 1997-2009 by Guido Vollbeding.
* libjpeg-turbo Modifications:
- * Copyright (C) 2014, D. R. Commander.
+ * Copyright (C) 2014, 2017, D. R. Commander.
* For conditions of distribution and use, see the accompanying README.ijg
* file.
*
@@ -208,6 +208,7 @@
JMESSAGE(JWRN_ARITH_BAD_CODE, "Corrupt JPEG data: bad arithmetic code")
#endif
#endif
+JMESSAGE(JWRN_BOGUS_ICC, "Corrupt JPEG data: bad ICC marker")
#ifdef JMAKE_ENUM_LIST
diff --git a/jmemmgr.c b/jmemmgr.c
index 8dfb633..4f74870 100644
--- a/jmemmgr.c
+++ b/jmemmgr.c
@@ -89,7 +89,9 @@
#ifndef WITH_SIMD
#define ALIGN_SIZE sizeof(double)
#else
-#define ALIGN_SIZE 16 /* Most SIMD implementations require this */
+#define ALIGN_SIZE 32 /* Most of the SIMD instructions we support require
+ 16-byte (128-bit) alignment, but AVX2 requires
+ 32-byte alignment. */
#endif
#endif
diff --git a/jpeglib.h b/jpeglib.h
index 6c63f58..de6883e 100644
--- a/jpeglib.h
+++ b/jpeglib.h
@@ -5,7 +5,7 @@
* Copyright (C) 1991-1998, Thomas G. Lane.
* Modified 2002-2009 by Guido Vollbeding.
* libjpeg-turbo Modifications:
- * Copyright (C) 2009-2011, 2013-2014, 2016, D. R. Commander.
+ * Copyright (C) 2009-2011, 2013-2014, 2016-2017, D. R. Commander.
* Copyright (C) 2015, Google, Inc.
* For conditions of distribution and use, see the accompanying README.ijg
* file.
@@ -976,6 +976,12 @@
/* Alternate compression function: just write an abbreviated table file */
EXTERN(void) jpeg_write_tables (j_compress_ptr cinfo);
+/* Write ICC profile. See libjpeg.txt for usage information. */
+EXTERN(void) jpeg_write_icc_profile (j_compress_ptr cinfo,
+ const JOCTET *icc_data_ptr,
+ unsigned int icc_data_len);
+
+
/* Decompression startup: read start of JPEG datastream to see what's there */
EXTERN(int) jpeg_read_header (j_decompress_ptr cinfo, boolean require_image);
/* Return value is one of: */
@@ -1057,6 +1063,11 @@
/* Default restart-marker-resync procedure for use by data source modules */
EXTERN(boolean) jpeg_resync_to_restart (j_decompress_ptr cinfo, int desired);
+/* Read ICC profile. See libjpeg.txt for usage information. */
+EXTERN(boolean) jpeg_read_icc_profile (j_decompress_ptr cinfo,
+ JOCTET **icc_data_ptr,
+ unsigned int *icc_data_len);
+
/* These marker codes are exported since applications and data source modules
* are likely to want to use them.
diff --git a/jpegtran.1 b/jpegtran.1
index 631455b..2efb264 100644
--- a/jpegtran.1
+++ b/jpegtran.1
@@ -217,6 +217,11 @@
.PP
Additional switches recognized by jpegtran are:
.TP
+.BI \-icc " file"
+Embed ICC color management profile contained in the specified file. Note that
+this will cause \fBjpegtran\fR to ignore any APP2 markers in the input file,
+even if \fB-copy all\fR is specified.
+.TP
.BI \-maxmemory " N"
Set limit for amount of memory to use in processing large images. Value is
in thousands of bytes, or millions of bytes if "M" is attached to the
diff --git a/jpegtran.c b/jpegtran.c
index 6f8fd5b..ebd751e 100644
--- a/jpegtran.c
+++ b/jpegtran.c
@@ -40,6 +40,7 @@
static const char *progname; /* program name for error messages */
+static char *icc_filename; /* for -icc switch */
static char *outfilename; /* for -outfile switch */
static JCOPY_OPTION copyoption; /* -copy switch */
static jpeg_transform_info transformoption; /* image transformation options */
@@ -83,6 +84,7 @@
#ifdef C_ARITH_CODING_SUPPORTED
fprintf(stderr, " -arithmetic Use arithmetic coding\n");
#endif
+ fprintf(stderr, " -icc FILE Embed ICC profile contained in FILE\n");
fprintf(stderr, " -restart N Set restart interval in rows, or in blocks with B\n");
fprintf(stderr, " -maxmemory N Maximum memory to use (in kbytes)\n");
fprintf(stderr, " -outfile name Specify name for output file\n");
@@ -138,6 +140,7 @@
/* Set up default JPEG parameters. */
simple_progressive = FALSE;
+ icc_filename = NULL;
outfilename = NULL;
copyoption = JCOPYOPT_DEFAULT;
transformoption.transform = JXFORM_NONE;
@@ -238,6 +241,12 @@
select_transform(JXFORM_NONE); /* force an error */
#endif
+ } else if (keymatch(arg, "icc", 1)) {
+ /* Set ICC filename. */
+ if (++argn >= argc) /* advance to next argument */
+ usage();
+ icc_filename = argv[argn];
+
} else if (keymatch(arg, "maxmemory", 3)) {
/* Maximum memory in Kb (or Mb with 'm'). */
long lval;
@@ -385,6 +394,9 @@
* single file pointer for sequential input and output operation.
*/
FILE *fp;
+ FILE *icc_file;
+ JOCTET *icc_profile = NULL;
+ long icc_len = 0;
/* On Mac, fetch a command line. */
#ifdef USE_CCOMMAND
@@ -449,6 +461,35 @@
fp = read_stdin();
}
+ if (icc_filename != NULL) {
+ if ((icc_file = fopen(icc_filename, READ_BINARY)) == NULL) {
+ fprintf(stderr, "%s: can't open %s\n", progname, icc_filename);
+ exit(EXIT_FAILURE);
+ }
+ if (fseek(icc_file, 0, SEEK_END) < 0 ||
+ (icc_len = ftell(icc_file)) < 1 ||
+ fseek(icc_file, 0, SEEK_SET) < 0) {
+ fprintf(stderr, "%s: can't determine size of %s\n", progname,
+ icc_filename);
+ exit(EXIT_FAILURE);
+ }
+ if ((icc_profile = (JOCTET *)malloc(icc_len)) == NULL) {
+ fprintf(stderr, "%s: can't allocate memory for ICC profile\n", progname);
+ fclose(icc_file);
+ exit(EXIT_FAILURE);
+ }
+ if (fread(icc_profile, icc_len, 1, icc_file) < 1) {
+ fprintf(stderr, "%s: can't read ICC profile from %s\n", progname,
+ icc_filename);
+ free(icc_profile);
+ fclose(icc_file);
+ exit(EXIT_FAILURE);
+ }
+ fclose(icc_file);
+ if (copyoption == JCOPYOPT_ALL)
+ copyoption = JCOPYOPT_ALL_EXCEPT_ICC;
+ }
+
#ifdef PROGRESS_REPORT
start_progress_monitor((j_common_ptr) &dstinfo, &progress);
#endif
@@ -524,6 +565,9 @@
/* Copy to the output file any extra markers that we want to preserve */
jcopy_markers_execute(&srcinfo, &dstinfo, copyoption);
+ if (icc_profile != NULL)
+ jpeg_write_icc_profile(&dstinfo, icc_profile, (unsigned int)icc_len);
+
/* Execute image transformation, if any */
#if TRANSFORMS_SUPPORTED
jtransform_execute_transformation(&srcinfo, &dstinfo,
@@ -545,6 +589,9 @@
end_progress_monitor((j_common_ptr) &dstinfo);
#endif
+ if (icc_profile != NULL)
+ free(icc_profile);
+
/* All done. */
exit(jsrcerr.num_warnings + jdsterr.num_warnings ?EXIT_WARNING:EXIT_SUCCESS);
return 0; /* suppress no-return-value warnings */
diff --git a/jversion.h b/jversion.h
index 7e44eaa..7bb8ff4 100644
--- a/jversion.h
+++ b/jversion.h
@@ -38,6 +38,7 @@
#define JCOPYRIGHT "Copyright (C) 2009-2017 D. R. Commander\n" \
"Copyright (C) 2011-2016 Siarhei Siamashka\n" \
"Copyright (C) 2015-2016 Matthieu Darbois\n" \
+ "Copyright (C) 2015 Intel Corporation\n" \
"Copyright (C) 2015 Google, Inc.\n" \
"Copyright (C) 2013-2014 MIPS Technologies, Inc.\n" \
"Copyright (C) 2013 Linaro Limited\n" \
diff --git a/libjpeg.txt b/libjpeg.txt
index 5181afc..13094d4 100644
--- a/libjpeg.txt
+++ b/libjpeg.txt
@@ -47,6 +47,7 @@
Buffered-image mode
Abbreviated datastreams and multiple images
Special markers
+ ICC profiles
Raw (downsampled) image data
Really raw data: DCT coefficients
Progress monitoring
@@ -2633,6 +2634,44 @@
Also, see jpegtran.c for an example of using jpeg_save_markers.
+ICC profiles
+------------
+
+Two functions are provided for writing and reading International Color
+Consortium (ICC) device profiles embedded in JFIF JPEG image files:
+
+ void jpeg_write_icc_profile (j_compress_ptr cinfo,
+ const JOCTET *icc_data_ptr,
+ unsigned int icc_data_len);
+ boolean jpeg_read_icc_profile (j_decompress_ptr cinfo,
+ JOCTET **icc_data_ptr,
+ unsigned int *icc_data_len);
+
+The ICC has defined a standard for including such data in JPEG "APP2" markers.
+The aforementioned functions do not know anything about the internal structure
+of the ICC profile data; they just know how to embed the profile data into a
+JPEG file while writing it, or to extract the profile data from a JPEG file
+while reading it.
+
+jpeg_write_icc_profile() must be called after calling jpeg_start_compress() and
+before the first call to jpeg_write_scanlines() or jpeg_write_raw_data(). This
+ordering ensures that the APP2 marker(s) will appear after the SOI and JFIF or
+Adobe markers, but before all other data.
+
+jpeg_read_icc_profile() returns TRUE if an ICC profile was found and FALSE
+otherwise. If an ICC profile was found, then the function will allocate a
+memory region containing the profile and will return a pointer to that memory
+region in *icc_data_ptr, as well as the length of the region in *icc_data_len.
+This memory region is allocated by the library using malloc() and must be freed
+by the caller using free() when the memory region is no longer needed. Callers
+wishing to use jpeg_read_icc_profile() must call
+
+ jpeg_save_markers(cinfo, JPEG_APP0 + 2, 0xFFFF);
+
+prior to calling jpeg_read_header(). jpeg_read_icc_profile() can be called at
+any point between jpeg_read_header() and jpeg_finish_decompress().
+
+
Raw (downsampled) image data
----------------------------
diff --git a/md5/Makefile.am b/md5/Makefile.am
deleted file mode 100644
index b36f019..0000000
--- a/md5/Makefile.am
+++ /dev/null
@@ -1,4 +0,0 @@
-noinst_PROGRAMS = md5cmp
-
-md5cmp_SOURCES = md5cmp.c md5.c md5hl.c md5.h
-md5cmp_CFLAGS = -I$(srcdir)
diff --git a/release/Distribution.xml b/release/Distribution.xml.in
similarity index 63%
rename from release/Distribution.xml
rename to release/Distribution.xml.in
index ee73ab0..e1f79ee 100644
--- a/release/Distribution.xml
+++ b/release/Distribution.xml.in
@@ -1,6 +1,6 @@
<?xml version="1.0" encoding="utf-8"?>
<installer-gui-script minSpecVersion="1">
- <title>libjpeg-turbo</title>
+ <title>@CMAKE_PROJECT_NAME@</title>
<welcome file="Welcome.rtf" />
<readme file="ReadMe.txt" />
<license file="License.rtf" />
@@ -12,13 +12,13 @@
<options customize="never" />
<choices-outline>
<line choice="default">
- <line choice="com.libjpeg-turbo.libjpeg-turbo"/>
+ <line choice="@PKGID@"/>
</line>
</choices-outline>
<choice id="default"/>
- <choice id="com.libjpeg-turbo.libjpeg-turbo" visible="false">
- <pkg-ref id="com.libjpeg-turbo.libjpeg-turbo"/>
+ <choice id="@PKGID@" visible="false">
+ <pkg-ref id="@PKGID@"/>
</choice>
<pkg-ref auth="root"
- id="com.libjpeg-turbo.libjpeg-turbo">libjpeg-turbo.pkg</pkg-ref>
+ id="@PKGID@">@PKGNAME@.pkg</pkg-ref>
</installer-gui-script>
diff --git a/release/ReadMe.txt b/release/ReadMe.txt
index 7fb8d0f..a3bafd9 100644
--- a/release/ReadMe.txt
+++ b/release/ReadMe.txt
@@ -1,4 +1,4 @@
-libjpeg-turbo is a JPEG image codec that uses SIMD instructions (MMX, SSE2, NEON, AltiVec) to accelerate baseline JPEG compression and decompression on x86, x86-64, ARM, and PowerPC systems. On such systems, libjpeg-turbo is generally 2-6x as fast as libjpeg, all else being equal. On other types of systems, libjpeg-turbo can still outperform libjpeg by a significant amount, by virtue of its highly-optimized Huffman coding routines. In many cases, the performance of libjpeg-turbo rivals that of proprietary high-speed JPEG codecs.
+libjpeg-turbo is a JPEG image codec that uses SIMD instructions (MMX, SSE2, AVX2, NEON, AltiVec) to accelerate baseline JPEG compression and decompression on x86, x86-64, ARM, and PowerPC systems. On such systems, libjpeg-turbo is generally 2-6x as fast as libjpeg, all else being equal. On other types of systems, libjpeg-turbo can still outperform libjpeg by a significant amount, by virtue of its highly-optimized Huffman coding routines. In many cases, the performance of libjpeg-turbo rivals that of proprietary high-speed JPEG codecs.
libjpeg-turbo implements both the traditional libjpeg API as well as the less powerful but more straightforward TurboJPEG API. libjpeg-turbo also features colorspace extensions that allow it to compress from/decompress to 32-bit and big-endian pixel buffers (RGBX, XBGR, etc.), as well as a full-featured Java interface.
diff --git a/release/deb-control.tmpl b/release/deb-control.in
similarity index 81%
rename from release/deb-control.tmpl
rename to release/deb-control.in
index 681721d..8c54dd9 100644
--- a/release/deb-control.tmpl
+++ b/release/deb-control.in
@@ -1,16 +1,16 @@
Package: {__PKGNAME}
-Version: {__VERSION}-{__BUILD}
+Version: @VERSION@-@BUILD@
Section: misc
Priority: optional
Architecture: {__ARCH}
Essential: no
-Maintainer: The libjpeg-turbo Project <information@libjpeg-turbo.org>
-Homepage: http://www.libjpeg-turbo.org
+Maintainer: @PKGVENDOR@ <@PKGEMAIL@>
+Homepage: @PKGURL@
Installed-Size: {__SIZE}
Description: A SIMD-accelerated JPEG codec that provides both the libjpeg and TurboJPEG APIs
libjpeg-turbo is a JPEG image codec that uses SIMD instructions (MMX, SSE2,
- NEON, AltiVec) to accelerate baseline JPEG compression and decompression on
- x86, x86-64, ARM, and PowerPC systems. On such systems, libjpeg-turbo is
+ AVX2, NEON, AltiVec) to accelerate baseline JPEG compression and decompression
+ on x86, x86-64, ARM, and PowerPC systems. On such systems, libjpeg-turbo is
generally 2-6x as fast as libjpeg, all else being equal. On other types of
systems, libjpeg-turbo can still outperform libjpeg by a significant amount,
by virtue of its highly-optimized Huffman coding routines. In many cases, the
diff --git a/release/libjpeg-turbo.nsi.in b/release/installer.nsi.in
similarity index 63%
rename from release/libjpeg-turbo.nsi.in
rename to release/installer.nsi.in
index f458b81..b413838 100755
--- a/release/libjpeg-turbo.nsi.in
+++ b/release/installer.nsi.in
@@ -1,6 +1,6 @@
!include x64.nsh
Name "@CMAKE_PROJECT_NAME@ SDK for @INST_PLATFORM@"
-OutFile "@CMAKE_BINARY_DIR@\${BUILDDIR}@INST_NAME@.exe"
+OutFile "@CMAKE_CURRENT_BINARY_DIR@\${BUILDDIR}@INST_NAME@.exe"
InstallDir @INST_DIR@
SetCompressor bzip2
@@ -35,58 +35,58 @@
notexists:
SetOutPath $SYSDIR
!ifdef GCC
- File "@CMAKE_BINARY_DIR@\libturbojpeg.dll"
+ File "@CMAKE_CURRENT_BINARY_DIR@\libturbojpeg.dll"
!else
- File "@CMAKE_BINARY_DIR@\${BUILDDIR}turbojpeg.dll"
+ File "@CMAKE_CURRENT_BINARY_DIR@\${BUILDDIR}turbojpeg.dll"
!endif
SetOutPath $INSTDIR\bin
!ifdef GCC
- File "@CMAKE_BINARY_DIR@\libturbojpeg.dll"
+ File "@CMAKE_CURRENT_BINARY_DIR@\libturbojpeg.dll"
!else
- File "@CMAKE_BINARY_DIR@\${BUILDDIR}turbojpeg.dll"
+ File "@CMAKE_CURRENT_BINARY_DIR@\${BUILDDIR}turbojpeg.dll"
!endif
!ifdef GCC
- File "/oname=libjpeg-@DLL_VERSION@.dll" "@CMAKE_BINARY_DIR@\sharedlib\libjpeg-*.dll"
+ File "@CMAKE_CURRENT_BINARY_DIR@\libjpeg-@SO_MAJOR_VERSION@.dll"
!else
- File "@CMAKE_BINARY_DIR@\sharedlib\${BUILDDIR}jpeg@DLL_VERSION@.dll"
+ File "@CMAKE_CURRENT_BINARY_DIR@\${BUILDDIR}jpeg@SO_MAJOR_VERSION@.dll"
!endif
- File "@CMAKE_BINARY_DIR@\sharedlib\${BUILDDIR}cjpeg.exe"
- File "@CMAKE_BINARY_DIR@\sharedlib\${BUILDDIR}djpeg.exe"
- File "@CMAKE_BINARY_DIR@\sharedlib\${BUILDDIR}jpegtran.exe"
- File "@CMAKE_BINARY_DIR@\${BUILDDIR}tjbench.exe"
- File "@CMAKE_BINARY_DIR@\${BUILDDIR}rdjpgcom.exe"
- File "@CMAKE_BINARY_DIR@\${BUILDDIR}wrjpgcom.exe"
+ File "@CMAKE_CURRENT_BINARY_DIR@\${BUILDDIR}cjpeg.exe"
+ File "@CMAKE_CURRENT_BINARY_DIR@\${BUILDDIR}djpeg.exe"
+ File "@CMAKE_CURRENT_BINARY_DIR@\${BUILDDIR}jpegtran.exe"
+ File "@CMAKE_CURRENT_BINARY_DIR@\${BUILDDIR}tjbench.exe"
+ File "@CMAKE_CURRENT_BINARY_DIR@\${BUILDDIR}rdjpgcom.exe"
+ File "@CMAKE_CURRENT_BINARY_DIR@\${BUILDDIR}wrjpgcom.exe"
SetOutPath $INSTDIR\lib
!ifdef GCC
- File "@CMAKE_BINARY_DIR@\libturbojpeg.dll.a"
- File "@CMAKE_BINARY_DIR@\libturbojpeg.a"
- File "@CMAKE_BINARY_DIR@\sharedlib\libjpeg.dll.a"
- File "@CMAKE_BINARY_DIR@\libjpeg.a"
+ File "@CMAKE_CURRENT_BINARY_DIR@\libturbojpeg.dll.a"
+ File "@CMAKE_CURRENT_BINARY_DIR@\libturbojpeg.a"
+ File "@CMAKE_CURRENT_BINARY_DIR@\libjpeg.dll.a"
+ File "@CMAKE_CURRENT_BINARY_DIR@\libjpeg.a"
!else
- File "@CMAKE_BINARY_DIR@\${BUILDDIR}turbojpeg.lib"
- File "@CMAKE_BINARY_DIR@\${BUILDDIR}turbojpeg-static.lib"
- File "@CMAKE_BINARY_DIR@\sharedlib\${BUILDDIR}jpeg.lib"
- File "@CMAKE_BINARY_DIR@\${BUILDDIR}jpeg-static.lib"
+ File "@CMAKE_CURRENT_BINARY_DIR@\${BUILDDIR}turbojpeg.lib"
+ File "@CMAKE_CURRENT_BINARY_DIR@\${BUILDDIR}turbojpeg-static.lib"
+ File "@CMAKE_CURRENT_BINARY_DIR@\${BUILDDIR}jpeg.lib"
+ File "@CMAKE_CURRENT_BINARY_DIR@\${BUILDDIR}jpeg-static.lib"
!endif
!ifdef JAVA
SetOutPath $INSTDIR\classes
- File "@CMAKE_BINARY_DIR@\java\${BUILDDIR}turbojpeg.jar"
+ File "@CMAKE_CURRENT_BINARY_DIR@\java\turbojpeg.jar"
!endif
SetOutPath $INSTDIR\include
- File "@CMAKE_BINARY_DIR@\jconfig.h"
- File "@CMAKE_SOURCE_DIR@\jerror.h"
- File "@CMAKE_SOURCE_DIR@\jmorecfg.h"
- File "@CMAKE_SOURCE_DIR@\jpeglib.h"
- File "@CMAKE_SOURCE_DIR@\turbojpeg.h"
+ File "@CMAKE_CURRENT_BINARY_DIR@\jconfig.h"
+ File "@CMAKE_CURRENT_SOURCE_DIR@\jerror.h"
+ File "@CMAKE_CURRENT_SOURCE_DIR@\jmorecfg.h"
+ File "@CMAKE_CURRENT_SOURCE_DIR@\jpeglib.h"
+ File "@CMAKE_CURRENT_SOURCE_DIR@\turbojpeg.h"
SetOutPath $INSTDIR\doc
- File "@CMAKE_SOURCE_DIR@\README.ijg"
- File "@CMAKE_SOURCE_DIR@\README.md"
- File "@CMAKE_SOURCE_DIR@\LICENSE.md"
- File "@CMAKE_SOURCE_DIR@\example.c"
- File "@CMAKE_SOURCE_DIR@\libjpeg.txt"
- File "@CMAKE_SOURCE_DIR@\structure.txt"
- File "@CMAKE_SOURCE_DIR@\usage.txt"
- File "@CMAKE_SOURCE_DIR@\wizard.txt"
+ File "@CMAKE_CURRENT_SOURCE_DIR@\README.ijg"
+ File "@CMAKE_CURRENT_SOURCE_DIR@\README.md"
+ File "@CMAKE_CURRENT_SOURCE_DIR@\LICENSE.md"
+ File "@CMAKE_CURRENT_SOURCE_DIR@\example.c"
+ File "@CMAKE_CURRENT_SOURCE_DIR@\libjpeg.txt"
+ File "@CMAKE_CURRENT_SOURCE_DIR@\structure.txt"
+ File "@CMAKE_CURRENT_SOURCE_DIR@\usage.txt"
+ File "@CMAKE_CURRENT_SOURCE_DIR@\wizard.txt"
WriteRegStr HKLM "SOFTWARE\@INST_REG_NAME@ @VERSION@" "Install_Dir" "$INSTDIR"
@@ -110,7 +110,7 @@
DeleteRegKey HKLM "SOFTWARE\@INST_REG_NAME@ @VERSION@"
!ifdef GCC
- Delete $INSTDIR\bin\libjpeg-@DLL_VERSION@.dll
+ Delete $INSTDIR\bin\libjpeg-@SO_MAJOR_VERSION@.dll
Delete $INSTDIR\bin\libturbojpeg.dll
Delete $SYSDIR\libturbojpeg.dll
Delete $INSTDIR\lib\libturbojpeg.dll.a"
@@ -118,7 +118,7 @@
Delete $INSTDIR\lib\libjpeg.dll.a"
Delete $INSTDIR\lib\libjpeg.a"
!else
- Delete $INSTDIR\bin\jpeg@DLL_VERSION@.dll
+ Delete $INSTDIR\bin\jpeg@SO_MAJOR_VERSION@.dll
Delete $INSTDIR\bin\turbojpeg.dll
Delete $SYSDIR\turbojpeg.dll
Delete $INSTDIR\lib\jpeg.lib
diff --git a/release/libjpeg-turbo.spec.in b/release/libjpeg-turbo.spec.in
deleted file mode 100644
index 4b792d7..0000000
--- a/release/libjpeg-turbo.spec.in
+++ /dev/null
@@ -1,153 +0,0 @@
-# Path under which libjpeg-turbo should be installed
-%define _prefix %{__prefix}
-
-# Path under which executables should be installed
-%define _bindir %{__bindir}
-
-# Path under which Java classes and man pages should be installed
-%define _datadir %{__datadir}
-
-# Path under which docs should be installed
-%define _docdir /usr/share/doc/%{name}-%{version}
-
-# Path under which headers should be installed
-%define _includedir %{__includedir}
-
-# _libdir is set to %{_prefix}/%{_lib} by default
-%ifarch x86_64
-%define _lib lib64
-%else
-%if "%{_prefix}" == "/opt/libjpeg-turbo"
-%define _lib lib32
-%endif
-%endif
-
-# Path under which man pages should be installed
-%define _mandir %{__mandir}
-
-Summary: A SIMD-accelerated JPEG codec that provides both the libjpeg and TurboJPEG APIs
-Name: @PKGNAME@
-Version: @VERSION@
-Vendor: The libjpeg-turbo Project
-URL: http://www.libjpeg-turbo.org
-Group: System Environment/Libraries
-#-->Source0: http://prdownloads.sourceforge.net/libjpeg-turbo/libjpeg-turbo-%{version}.tar.gz
-Release: @BUILD@
-License: BSD-style
-BuildRoot: %{_blddir}/%{name}-buildroot-%{version}-%{release}
-Prereq: /sbin/ldconfig
-%ifarch x86_64
-Provides: %{name} = %{version}-%{release}, @PACKAGE_NAME@ = %{version}-%{release}, libturbojpeg.so()(64bit)
-%else
-Provides: %{name} = %{version}-%{release}, @PACKAGE_NAME@ = %{version}-%{release}, libturbojpeg.so
-%endif
-
-%description
-libjpeg-turbo is a JPEG image codec that uses SIMD instructions (MMX, SSE2,
-NEON, AltiVec) to accelerate baseline JPEG compression and decompression on
-x86, x86-64, ARM, and PowerPC systems. On such systems, libjpeg-turbo is
-generally 2-6x as fast as libjpeg, all else being equal. On other types of
-systems, libjpeg-turbo can still outperform libjpeg by a significant amount, by
-virtue of its highly-optimized Huffman coding routines. In many cases, the
-performance of libjpeg-turbo rivals that of proprietary high-speed JPEG codecs.
-
-libjpeg-turbo implements both the traditional libjpeg API as well as the less
-powerful but more straightforward TurboJPEG API. libjpeg-turbo also features
-colorspace extensions that allow it to compress from/decompress to 32-bit and
-big-endian pixel buffers (RGBX, XBGR, etc.), as well as a full-featured Java
-interface.
-
-libjpeg-turbo was originally based on libjpeg/SIMD, an MMX-accelerated
-derivative of libjpeg v6b developed by Miyasaka Masaru. The TigerVNC and
-VirtualGL projects made numerous enhancements to the codec in 2009, and in
-early 2010, libjpeg-turbo spun off into an independent project, with the goal
-of making high-speed JPEG compression/decompression technology available to a
-broader range of users and developers.
-
-#-->%prep
-#-->%setup -q -n libjpeg-turbo-%{version}
-
-#-->%build
-#-->./configure prefix=%{_prefix} bindir=%{_bindir} datadir=%{_datadir} \
-#--> docdir=%{_docdir} includedir=%{_includedir} libdir=%{_libdir} \
-#--> mandir=%{_mandir} JPEG_LIB_VERSION=@JPEG_LIB_VERSION@ \
-#--> SO_MAJOR_VERSION=@SO_MAJOR_VERSION@ SO_MINOR_VERSION=@SO_MINOR_VERSION@ \
-#--> --with-pic @RPM_CONFIG_ARGS@
-#-->make DESTDIR=$RPM_BUILD_ROOT
-
-%install
-
-rm -rf $RPM_BUILD_ROOT
-make install DESTDIR=$RPM_BUILD_ROOT docdir=%{_docdir} exampledir=%{_docdir}
-rm -f $RPM_BUILD_ROOT%{_libdir}/*.la
-/sbin/ldconfig -n $RPM_BUILD_ROOT%{_libdir}
-
-#-->%if 0
-
-LJT_LIBDIR=%{__libdir}
-if [ ! "$LJT_LIBDIR" = "%{_libdir}" ]; then
- echo ERROR: libjpeg-turbo must be configured with libdir=%{_prefix}/%{_lib} when generating an in-tree RPM for this architecture.
- exit 1
-fi
-
-#-->%endif
-
-LJT_DOCDIR=%{__docdir}
-if [ "%{_prefix}" = "/opt/libjpeg-turbo" -a "$LJT_DOCDIR" = "/opt/libjpeg-turbo/doc" ]; then
- ln -fs %{_docdir} $RPM_BUILD_ROOT/$LJT_DOCDIR
-fi
-
-%post -p /sbin/ldconfig
-
-%postun -p /sbin/ldconfig
-
-%clean
-rm -rf $RPM_BUILD_ROOT
-
-%files
-%defattr(-,root,root)
-%dir %{_docdir}
-%doc %{_docdir}/*
-%dir %{_prefix}
-%if "%{_prefix}" == "/opt/libjpeg-turbo" && "%{_docdir}" != "%{_prefix}/doc"
- %{_prefix}/doc
-%endif
-%dir %{_bindir}
-%{_bindir}/cjpeg
-%{_bindir}/djpeg
-%{_bindir}/jpegtran
-%{_bindir}/tjbench
-%{_bindir}/rdjpgcom
-%{_bindir}/wrjpgcom
-%dir %{_libdir}
-%{_libdir}/libjpeg.so.@SO_MAJOR_VERSION@.@SO_AGE@.@SO_MINOR_VERSION@
-%{_libdir}/libjpeg.so.@SO_MAJOR_VERSION@
-%{_libdir}/libjpeg.so
-%{_libdir}/libjpeg.a
-%{_libdir}/pkgconfig
-%{_libdir}/pkgconfig/libjpeg.pc
-%{_libdir}/libturbojpeg.so.0.1.0
-%{_libdir}/libturbojpeg.so.0
-%{_libdir}/libturbojpeg.so
-%{_libdir}/libturbojpeg.a
-%{_libdir}/pkgconfig/libturbojpeg.pc
-%dir %{_includedir}
-%{_includedir}/jconfig.h
-%{_includedir}/jerror.h
-%{_includedir}/jmorecfg.h
-%{_includedir}/jpeglib.h
-%{_includedir}/turbojpeg.h
-%dir %{_mandir}
-%dir %{_mandir}/man1
-%{_mandir}/man1/cjpeg.1*
-%{_mandir}/man1/djpeg.1*
-%{_mandir}/man1/jpegtran.1*
-%{_mandir}/man1/rdjpgcom.1*
-%{_mandir}/man1/wrjpgcom.1*
-%if "%{_prefix}" != "%{_datadir}"
- %dir %{_datadir}
-%endif
-@JAVA_RPM_CONTENTS_1@
-@JAVA_RPM_CONTENTS_2@
-
-%changelog
diff --git a/release/libjpeg.pc.in b/release/libjpeg.pc.in
index 40795f7..74fb7fc 100644
--- a/release/libjpeg.pc.in
+++ b/release/libjpeg.pc.in
@@ -1,10 +1,10 @@
-prefix=@prefix@
-exec_prefix=@exec_prefix@
-libdir=@libdir@
-includedir=@includedir@
+prefix=@CMAKE_INSTALL_PREFIX@
+exec_prefix=@CMAKE_INSTALL_PREFIX@
+libdir=@CMAKE_INSTALL_FULL_LIBDIR@
+includedir=@CMAKE_INSTALL_FULL_INCLUDEDIR@
Name: libjpeg
Description: A SIMD-accelerated JPEG codec that provides the libjpeg API
-Version: @PACKAGE_VERSION@
+Version: @VERSION@
Libs: -L${libdir} -ljpeg
Cflags: -I${includedir}
diff --git a/release/libturbojpeg.pc.in b/release/libturbojpeg.pc.in
index 7d4b656..81a0063 100644
--- a/release/libturbojpeg.pc.in
+++ b/release/libturbojpeg.pc.in
@@ -1,10 +1,10 @@
-prefix=@prefix@
-exec_prefix=@exec_prefix@
-libdir=@libdir@
-includedir=@includedir@
+prefix=@CMAKE_INSTALL_PREFIX@
+exec_prefix=@CMAKE_INSTALL_PREFIX@
+libdir=@CMAKE_INSTALL_FULL_LIBDIR@
+includedir=@CMAKE_INSTALL_FULL_INCLUDEDIR@
Name: libturbojpeg
Description: A SIMD-accelerated JPEG codec that provides the TurboJPEG API
-Version: @PACKAGE_VERSION@
+Version: @VERSION@
Libs: -L${libdir} -lturbojpeg
Cflags: -I${includedir}
diff --git a/release/makecygwinpkg.in b/release/makecygwinpkg.in
index f303546..b7f353e 100755
--- a/release/makecygwinpkg.in
+++ b/release/makecygwinpkg.in
@@ -15,27 +15,51 @@
fi
}
-PACKAGE_NAME=@PKGNAME@
+safedirmove ()
+{
+ if [ "$1" = "$2" ]; then
+ return 0
+ fi
+ if [ "$1" = "" -o ! -d "$1" ]; then
+ echo safedirmove: source dir $1 is not valid
+ return 1
+ fi
+ if [ "$2" = "" -o -e "$2" ]; then
+ echo safedirmove: dest dir $2 is not valid
+ return 1
+ fi
+ if [ "$3" = "" -o -e "$3" ]; then
+ echo safedirmove: tmp dir $3 is not valid
+ return 1
+ fi
+ mkdir -p $3
+ mv $1/* $3/
+ rmdir $1
+ mkdir -p $2
+ mv $3/* $2/
+ rmdir $3
+ return 0
+}
+
+PKGNAME=@PKGNAME@
VERSION=@VERSION@
BUILD=@BUILD@
-SRCDIR=@abs_top_srcdir@
-PREFIX=%{__prefix}
-DOCDIR=%{__docdir}
-LIBDIR=%{__libdir}
+PREFIX=@CMAKE_INSTALL_PREFIX@
+DOCDIR=@CMAKE_INSTALL_FULL_DOCDIR@
+LIBDIR=@CMAKE_INSTALL_FULL_LIBDIR@
umask 022
-rm -f $PACKAGE_NAME-$VERSION-$BUILD.tar.bz2
+rm -f $PKGNAME-$VERSION-$BUILD.tar.bz2
TMPDIR=`mktemp -d /tmp/ljtbuild.XXXXXX`
__PWD=`pwd`
-make install DESTDIR=$TMPDIR/pkg docdir=/usr/share/doc/$PACKAGE_NAME-$VERSION \
- exampledir=/usr/share/doc/$PACKAGE_NAME-$VERSION
-rm $TMPDIR/pkg$LIBDIR/*.la
-if [ "$PREFIX" = "/opt/libjpeg-turbo" -a "$DOCDIR" = "/opt/libjpeg-turbo/doc" ]; then
- ln -fs /usr/share/doc/$PACKAGE_NAME-$VERSION $TMPDIR/pkg$DOCDIR
+make install DESTDIR=$TMPDIR/pkg
+if [ "$PREFIX" = "@CMAKE_INSTALL_DEFAULT_PREFIX@" -a "$DOCDIR" = "@CMAKE_INSTALL_DEFAULT_PREFIX@/doc" ]; then
+ safedirmove $TMPDIR/pkg$DOCDIR $TMPDIR/pkg/usr/share/doc/$PKGNAME-$VERSION $TMPDIR/__tmpdoc
+ ln -fs /usr/share/doc/$PKGNAME-$VERSION $TMPDIR/pkg$DOCDIR
fi
cd $TMPDIR/pkg
-tar cfj ../$PACKAGE_NAME-$VERSION-$BUILD.tar.bz2 *
+tar cfj ../$PKGNAME-$VERSION-$BUILD.tar.bz2 *
cd $__PWD
mv $TMPDIR/*.tar.bz2 .
diff --git a/release/makedpkg.in b/release/makedpkg.in
index 80cc89b..77836dd 100644
--- a/release/makedpkg.in
+++ b/release/makedpkg.in
@@ -21,56 +21,89 @@
id | cut -f2 -d = | cut -f1 -d \(;
}
+safedirmove ()
+{
+ if [ "$1" = "$2" ]; then
+ return 0
+ fi
+ if [ "$1" = "" -o ! -d "$1" ]; then
+ echo safedirmove: source dir $1 is not valid
+ return 1
+ fi
+ if [ "$2" = "" -o -e "$2" ]; then
+ echo safedirmove: dest dir $2 is not valid
+ return 1
+ fi
+ if [ "$3" = "" -o -e "$3" ]; then
+ echo safedirmove: tmp dir $3 is not valid
+ return 1
+ fi
+ mkdir -p $3
+ mv $1/* $3/
+ rmdir $1
+ mkdir -p $2
+ mv $3/* $2/
+ rmdir $3
+ return 0
+}
+
makedeb()
{
SUPPLEMENT=$1
- DIRNAME=$PACKAGE_NAME
+ DIRNAME=$PKGNAME
if [ $SUPPLEMENT = 1 ]; then
- PACKAGE_NAME=$PACKAGE_NAME\32
+ PKGNAME=$PKGNAME\32
DEBARCH=amd64
fi
umask 022
- rm -f $PACKAGE_NAME\_$VERSION\_$DEBARCH.deb
- TMPDIR=`mktemp -d /tmp/$PACKAGE_NAME-build.XXXXXX`
+ rm -f $PKGNAME\_$VERSION\_$DEBARCH.deb
+ TMPDIR=`mktemp -d /tmp/$PKGNAME-build.XXXXXX`
mkdir $TMPDIR/DEBIAN
if [ $SUPPLEMENT = 1 ]; then
- make install DESTDIR=$TMPDIR bindir=/dummy/bin datadir=/dummy/data \
- docdir=/dummy/doc includedir=/dummy/include mandir=/dummy/man
- rm -f $TMPDIR$LIBDIR/*.la
- rm -rf $TMPDIR/dummy
+ make install DESTDIR=$TMPDIR
+ rm -rf $TMPDIR$BINDIR
+ if [ "$DATAROOTDIR" != "$PREFIX" ]; then
+ rm -rf $TMPDIR$DATAROOTDIR
+ fi
+ if [ "$JAVADIR" != "" ]; then
+ rm -rf $TMPDIR$JAVADIR
+ fi
+ rm -rf $TMPDIR$DOCDIR
+ rm -rf $TMPDIR$INCLUDEDIR
+ rm -rf $TMPDIR$MANDIR
else
- make install DESTDIR=$TMPDIR docdir=/usr/share/doc/$DIRNAME-$VERSION \
- exampledir=/usr/share/doc/$DIRNAME-$VERSION
- rm -f $TMPDIR$LIBDIR/*.la
- if [ "$PREFIX" = "/opt/libjpeg-turbo" -a "$DOCDIR" = "/opt/libjpeg-turbo/doc" ]; then
+ make install DESTDIR=$TMPDIR
+ if [ "$PREFIX" = "@CMAKE_INSTALL_DEFAULT_PREFIX@" -a "$DOCDIR" = "@CMAKE_INSTALL_DEFAULT_PREFIX@/doc" ]; then
+ safedirmove $TMPDIR/$DOCDIR $TMPDIR/usr/share/doc/$PKGNAME-$VERSION $TMPDIR/__tmpdoc
ln -fs /usr/share/doc/$DIRNAME-$VERSION $TMPDIR$DOCDIR
fi
fi
SIZE=`du -s $TMPDIR | cut -f1`
- (cat $SRCDIR/release/deb-control.tmpl | sed s/{__PKGNAME}/$PACKAGE_NAME/g \
- | sed s/{__VERSION}/$VERSION/g | sed s/{__BUILD}/$BUILD/g \
+ (cat pkgscripts/deb-control | sed s/{__PKGNAME}/$PKGNAME/g \
| sed s/{__ARCH}/$DEBARCH/g | sed s/{__SIZE}/$SIZE/g \
> $TMPDIR/DEBIAN/control)
-
/sbin/ldconfig -n $TMPDIR$LIBDIR
$SUDO chown -Rh root:root $TMPDIR/*
- dpkg -b $TMPDIR $PACKAGE_NAME\_$VERSION\_$DEBARCH.deb
+ dpkg -b $TMPDIR $PKGNAME\_$VERSION\_$DEBARCH.deb
}
-PACKAGE_NAME=@PKGNAME@
+PKGNAME=@PKGNAME@
VERSION=@VERSION@
-BUILD=@BUILD@
DEBARCH=@DEBARCH@
-SRCDIR=@abs_top_srcdir@
-PREFIX=%{__prefix}
-DOCDIR=%{__docdir}
-LIBDIR=%{__libdir}
+PREFIX=@CMAKE_INSTALL_PREFIX@
+BINDIR=@CMAKE_INSTALL_FULL_BINDIR@
+DATAROOTDIR=@CMAKE_INSTALL_FULL_DATAROOTDIR@
+DOCDIR=@CMAKE_INSTALL_FULL_DOCDIR@
+INCLUDEDIR=@CMAKE_INSTALL_FULL_INCLUDEDIR@
+JAVADIR=@CMAKE_INSTALL_FULL_JAVADIR@
+LIBDIR=@CMAKE_INSTALL_FULL_LIBDIR@
+MANDIR=@CMAKE_INSTALL_FULL_MANDIR@
if [ ! `uid` -eq 0 ]; then
SUDO=sudo
diff --git a/release/makemacpkg.in b/release/makemacpkg.in
index 7163757..b0a2e23 100644
--- a/release/makemacpkg.in
+++ b/release/makemacpkg.in
@@ -15,76 +15,63 @@
fi
}
+safedirmove ()
+{
+ if [ "$1" = "$2" ]; then
+ return 0
+ fi
+ if [ "$1" = "" -o ! -d "$1" ]; then
+ echo safedirmove: source dir $1 is not valid
+ return 1
+ fi
+ if [ "$2" = "" -o -e "$2" ]; then
+ echo safedirmove: dest dir $2 is not valid
+ return 1
+ fi
+ if [ "$3" = "" -o -e "$3" ]; then
+ echo safedirmove: tmp dir $3 is not valid
+ return 1
+ fi
+ mkdir -p $3
+ mv $1/* $3/
+ rmdir $1
+ mkdir -p $2
+ mv $3/* $2/
+ rmdir $3
+ return 0
+}
+
usage()
{
- echo "$0 [-build32 [32-bit build dir]] [-buildarmv6 [ARMv6 build dir]] [-buildarmv7 [ARMv7 build dir]] [-buildarmv7s [ARMv7s build dir] [-buildarmv8 [ARMv8 build dir]] [-lipo [path to lipo]]"
+ echo "$0 [universal] [-lipo [path to lipo]]"
exit 1
}
-PACKAGE_NAME=@PKGNAME@
+UNIVERSAL=0
+
+PKGNAME=@PKGNAME@
VERSION=@VERSION@
BUILD=@BUILD@
-SRCDIR=@abs_top_srcdir@
-BUILDDIR32=@abs_top_srcdir@/osxx86
-BUILD32=0
-BUILDDIRARMV6=@abs_top_srcdir@/iosarmv6
-BUILDARMV6=0
-BUILDDIRARMV7=@abs_top_srcdir@/iosarmv7
-BUILDARMV7=0
-BUILDDIRARMV7S=@abs_top_srcdir@/iosarmv7s
-BUILDARMV7S=0
-BUILDDIRARMV8=@abs_top_srcdir@/iosarmv8
-BUILDARMV8=0
+SRCDIR=@CMAKE_CURRENT_SOURCE_DIR@
+BUILDDIR32=@OSX_32BIT_BUILD@
+BUILDDIRARMV7=@IOS_ARMV7_BUILD@
+BUILDDIRARMV7S=@IOS_ARMV7S_BUILD@
+BUILDDIRARMV8=@IOS_ARMV8_BUILD@
WITH_JAVA=@WITH_JAVA@
LIPO=lipo
-PREFIX=%{__prefix}
-BINDIR=%{__bindir}
-DOCDIR=%{__docdir}
-LIBDIR=%{__libdir}
+PREFIX=@CMAKE_INSTALL_PREFIX@
+BINDIR=@CMAKE_INSTALL_FULL_BINDIR@
+DOCDIR=@CMAKE_INSTALL_FULL_DOCDIR@
+LIBDIR=@CMAKE_INSTALL_FULL_LIBDIR@
+
+LIBJPEG_DSO_NAME=libjpeg.@SO_MAJOR_VERSION@.@SO_AGE@.@SO_MINOR_VERSION@.dylib
+TURBOJPEG_DSO_NAME=libturbojpeg.@TURBOJPEG_SO_VERSION@.dylib
while [ $# -gt 0 ]; do
case $1 in
- -h*) usage 0 ;;
- -build32)
- BUILD32=1
- if [ $# -gt 1 ]; then
- if [[ ! "$2" =~ -.* ]]; then
- BUILDDIR32=$2; shift
- fi
- fi
- ;;
- -buildarmv6)
- BUILDARMV6=1
- if [ $# -gt 1 ]; then
- if [[ ! "$2" =~ -.* ]]; then
- BUILDDIRARMV6=$2; shift
- fi
- fi
- ;;
- -buildarmv7)
- BUILDARMV7=1
- if [ $# -gt 1 ]; then
- if [[ ! "$2" =~ -.* ]]; then
- BUILDDIRARMV7=$2; shift
- fi
- fi
- ;;
- -buildarmv7s)
- BUILDARMV7S=1
- if [ $# -gt 1 ]; then
- if [[ ! "$2" =~ -.* ]]; then
- BUILDDIRARMV7S=$2; shift
- fi
- fi
- ;;
- -buildarmv8)
- BUILDARMV8=1
- if [ $# -gt 1 ]; then
- if [[ ! "$2" =~ -.* ]]; then
- BUILDDIRARMV8=$2; shift
- fi
- fi
+ -h*)
+ usage 0
;;
-lipo)
if [ $# -gt 1 ]; then
@@ -93,27 +80,31 @@
fi
fi
;;
+ universal)
+ UNIVERSAL=1
+ ;;
esac
shift
done
-if [ -f $PACKAGE_NAME-$VERSION.dmg ]; then
- rm -f $PACKAGE_NAME-$VERSION.dmg
+if [ -f $PKGNAME-$VERSION.dmg ]; then
+ rm -f $PKGNAME-$VERSION.dmg
fi
umask 022
-TMPDIR=`mktemp -d /tmp/$PACKAGE_NAME-build.XXXXXX`
+TMPDIR=`mktemp -d /tmp/$PKGNAME-build.XXXXXX`
PKGROOT=$TMPDIR/pkg/Package_Root
mkdir -p $PKGROOT
-make install DESTDIR=$PKGROOT docdir=/Library/Documentation/$PACKAGE_NAME \
- exampledir=/Library/Documentation/$PACKAGE_NAME
-rm -f $PKGROOT$LIBDIR/*.la
-if [ "$PREFIX" = "/opt/libjpeg-turbo" -a "$DOCDIR" = "/opt/libjpeg-turbo/doc" ]; then
- ln -fs /Library/Documentation/$PACKAGE_NAME $PKGROOT$DOCDIR
+make install DESTDIR=$PKGROOT
+
+if [ "$PREFIX" = "@CMAKE_INSTALL_DEFAULT_PREFIX@" -a "$DOCDIR" = "@CMAKE_INSTALL_DEFAULT_PREFIX@/doc" ]; then
+ mkdir -p $PKGROOT/Library/Documentation
+ safedirmove $PKGROOT$DOCDIR $PKGROOT/Library/Documentation/$PKGNAME $TMPDIR/__tmpdoc
+ ln -fs /Library/Documentation/$PKGNAME $PKGROOT$DOCDIR
fi
-if [ $BUILD32 = 1 ]; then
+if [ $UNIVERSAL = 1 -a "$BUILDDIR32" != "" ]; then
if [ ! -d $BUILDDIR32 ]; then
echo ERROR: 32-bit build directory $BUILDDIR32 does not exist
exit 1
@@ -126,27 +117,18 @@
pushd $BUILDDIR32
make install DESTDIR=$TMPDIR/dist.x86
popd
- if [ ! -h $TMPDIR/dist.x86/$LIBDIR/libjpeg.@SO_MAJOR_VERSION@.dylib -a \
- ! -h $PKGROOT/$LIBDIR/libjpeg.@SO_MAJOR_VERSION@.dylib ]; then
- $LIPO -create \
- -arch i386 $TMPDIR/dist.x86/$LIBDIR/libjpeg.@SO_MAJOR_VERSION@.dylib \
- -arch x86_64 $PKGROOT/$LIBDIR/libjpeg.@SO_MAJOR_VERSION@.dylib \
- -output $PKGROOT/$LIBDIR/libjpeg.@SO_MAJOR_VERSION@.dylib
- elif [ ! -h $TMPDIR/dist.x86/$LIBDIR/libjpeg.@SO_MAJOR_VERSION@.0.@SO_MINOR_VERSION@.dylib -a \
- ! -h $PKGROOT/$LIBDIR/libjpeg.@SO_MAJOR_VERSION@.0.@SO_MINOR_VERSION@.dylib ]; then
- $LIPO -create \
- -arch i386 $TMPDIR/dist.x86/$LIBDIR/libjpeg.@SO_MAJOR_VERSION@.0.@SO_MINOR_VERSION@.dylib \
- -arch x86_64 $PKGROOT/$LIBDIR/libjpeg.@SO_MAJOR_VERSION@.0.@SO_MINOR_VERSION@.dylib \
- -output $PKGROOT/$LIBDIR/libjpeg.@SO_MAJOR_VERSION@.0.@SO_MINOR_VERSION@.dylib
- fi
+ $LIPO -create \
+ -arch i386 $TMPDIR/dist.x86/$LIBDIR/$LIBJPEG_DSO_NAME \
+ -arch x86_64 $PKGROOT/$LIBDIR/$LIBJPEG_DSO_NAME \
+ -output $PKGROOT/$LIBDIR/$LIBJPEG_DSO_NAME
$LIPO -create \
-arch i386 $TMPDIR/dist.x86/$LIBDIR/libjpeg.a \
-arch x86_64 $PKGROOT/$LIBDIR/libjpeg.a \
-output $PKGROOT/$LIBDIR/libjpeg.a
$LIPO -create \
- -arch i386 $TMPDIR/dist.x86/$LIBDIR/libturbojpeg.0.dylib \
- -arch x86_64 $PKGROOT/$LIBDIR/libturbojpeg.0.dylib \
- -output $PKGROOT/$LIBDIR/libturbojpeg.0.dylib
+ -arch i386 $TMPDIR/dist.x86/$LIBDIR/$TURBOJPEG_DSO_NAME \
+ -arch x86_64 $PKGROOT/$LIBDIR/$TURBOJPEG_DSO_NAME \
+ -output $PKGROOT/$LIBDIR/$TURBOJPEG_DSO_NAME
$LIPO -create \
-arch i386 $TMPDIR/dist.x86/$LIBDIR/libturbojpeg.a \
-arch x86_64 $PKGROOT/$LIBDIR/libturbojpeg.a \
@@ -175,272 +157,88 @@
-arch i386 $TMPDIR/dist.x86/$BINDIR/wrjpgcom \
-arch x86_64 $PKGROOT/$BINDIR/wrjpgcom \
-output $PKGROOT/$BINDIR/wrjpgcom
-
fi
-if [ $BUILDARMV6 = 1 ]; then
- if [ ! -d $BUILDDIRARMV6 ]; then
- echo ERROR: ARMv6 build directory $BUILDDIRARMV6 does not exist
+install_ios()
+{
+ BUILDDIR=$1
+ ARCHNAME=$2
+ DIRNAME=$3
+ LIPOARCH=$4
+
+ if [ ! -d $BUILDDIR ]; then
+ echo ERROR: $ARCHNAME build directory $BUILDDIR does not exist
exit 1
fi
- if [ ! -f $BUILDDIRARMV6/Makefile ]; then
- echo ERROR: ARMv6 build directory $BUILDDIRARMV6 is not configured
+ if [ ! -f $BUILDDIR/Makefile ]; then
+ echo ERROR: $ARCHNAME build directory $BUILDDIR is not configured
exit 1
fi
- mkdir -p $TMPDIR/dist.armv6
- pushd $BUILDDIRARMV6
- make install DESTDIR=$TMPDIR/dist.armv6
+ mkdir -p $TMPDIR/dist.$DIRNAME
+ pushd $BUILDDIR
+ make install DESTDIR=$TMPDIR/dist.$DIRNAME
popd
- if [ ! -h $TMPDIR/dist.armv6/$LIBDIR/libjpeg.@SO_MAJOR_VERSION@.dylib -a \
- ! -h $PKGROOT/$LIBDIR/libjpeg.@SO_MAJOR_VERSION@.dylib ]; then
- $LIPO -create \
- $PKGROOT/$LIBDIR/libjpeg.@SO_MAJOR_VERSION@.dylib \
- -arch arm $TMPDIR/dist.armv6/$LIBDIR/libjpeg.@SO_MAJOR_VERSION@.dylib \
- -output $PKGROOT/$LIBDIR/libjpeg.@SO_MAJOR_VERSION@.dylib
- elif [ ! -h $TMPDIR/dist.armv6/$LIBDIR/libjpeg.@SO_MAJOR_VERSION@.0.@SO_MINOR_VERSION@.dylib -a \
- ! -h $PKGROOT/$LIBDIR/libjpeg.@SO_MAJOR_VERSION@.0.@SO_MINOR_VERSION@.dylib ]; then
- $LIPO -create \
- $PKGROOT/$LIBDIR/libjpeg.@SO_MAJOR_VERSION@.0.@SO_MINOR_VERSION@.dylib \
- -arch arm $TMPDIR/dist.armv6/$LIBDIR/libjpeg.@SO_MAJOR_VERSION@.0.@SO_MINOR_VERSION@.dylib \
- -output $PKGROOT/$LIBDIR/libjpeg.@SO_MAJOR_VERSION@.0.@SO_MINOR_VERSION@.dylib
- fi
+ $LIPO -create \
+ $PKGROOT/$LIBDIR/$LIBJPEG_DSO_NAME \
+ -arch $LIPOARCH $TMPDIR/dist.$DIRNAME/$LIBDIR/$LIBJPEG_DSO_NAME \
+ -output $PKGROOT/$LIBDIR/$LIBJPEG_DSO_NAME
$LIPO -create \
$PKGROOT/$LIBDIR/libjpeg.a \
- -arch arm $TMPDIR/dist.armv6/$LIBDIR/libjpeg.a \
+ -arch $LIPOARCH $TMPDIR/dist.$DIRNAME/$LIBDIR/libjpeg.a \
-output $PKGROOT/$LIBDIR/libjpeg.a
$LIPO -create \
- $PKGROOT/$LIBDIR/libturbojpeg.0.dylib \
- -arch arm $TMPDIR/dist.armv6/$LIBDIR/libturbojpeg.0.dylib \
- -output $PKGROOT/$LIBDIR/libturbojpeg.0.dylib
+ $PKGROOT/$LIBDIR/$TURBOJPEG_DSO_NAME \
+ -arch $LIPOARCH $TMPDIR/dist.$DIRNAME/$LIBDIR/$TURBOJPEG_DSO_NAME \
+ -output $PKGROOT/$LIBDIR/$TURBOJPEG_DSO_NAME
$LIPO -create \
$PKGROOT/$LIBDIR/libturbojpeg.a \
- -arch arm $TMPDIR/dist.armv6/$LIBDIR/libturbojpeg.a \
+ -arch $LIPOARCH $TMPDIR/dist.$DIRNAME/$LIBDIR/libturbojpeg.a \
-output $PKGROOT/$LIBDIR/libturbojpeg.a
$LIPO -create \
$PKGROOT/$BINDIR/cjpeg \
- -arch arm $TMPDIR/dist.armv6/$BINDIR/cjpeg \
+ -arch $LIPOARCH $TMPDIR/dist.$DIRNAME/$BINDIR/cjpeg \
-output $PKGROOT/$BINDIR/cjpeg
$LIPO -create \
$PKGROOT/$BINDIR/djpeg \
- -arch arm $TMPDIR/dist.armv6/$BINDIR/djpeg \
+ -arch $LIPOARCH $TMPDIR/dist.$DIRNAME/$BINDIR/djpeg \
-output $PKGROOT/$BINDIR/djpeg
$LIPO -create \
$PKGROOT/$BINDIR/jpegtran \
- -arch arm $TMPDIR/dist.armv6/$BINDIR/jpegtran \
+ -arch $LIPOARCH $TMPDIR/dist.$DIRNAME/$BINDIR/jpegtran \
-output $PKGROOT/$BINDIR/jpegtran
$LIPO -create \
$PKGROOT/$BINDIR/tjbench \
- -arch arm $TMPDIR/dist.armv6/$BINDIR/tjbench \
+ -arch $LIPOARCH $TMPDIR/dist.$DIRNAME/$BINDIR/tjbench \
-output $PKGROOT/$BINDIR/tjbench
$LIPO -create \
$PKGROOT/$BINDIR/rdjpgcom \
- -arch arm $TMPDIR/dist.armv6/$BINDIR/rdjpgcom \
+ -arch $LIPOARCH $TMPDIR/dist.$DIRNAME/$BINDIR/rdjpgcom \
-output $PKGROOT/$BINDIR/rdjpgcom
$LIPO -create \
$PKGROOT/$BINDIR/wrjpgcom \
- -arch arm $TMPDIR/dist.armv6/$BINDIR/wrjpgcom \
+ -arch $LIPOARCH $TMPDIR/dist.$DIRNAME/$BINDIR/wrjpgcom \
-output $PKGROOT/$BINDIR/wrjpgcom
+}
+
+if [ $UNIVERSAL = 1 -a "$BUILDDIRARMV7" != "" ]; then
+ install_ios $BUILDDIRARMV7 ARMv7 armv7 arm
fi
-if [ $BUILDARMV7 = 1 ]; then
- if [ ! -d $BUILDDIRARMV7 ]; then
- echo ERROR: ARMv7 build directory $BUILDDIRARMV7 does not exist
- exit 1
- fi
- if [ ! -f $BUILDDIRARMV7/Makefile ]; then
- echo ERROR: ARMv7 build directory $BUILDDIRARMV7 is not configured
- exit 1
- fi
- mkdir -p $TMPDIR/dist.armv7
- pushd $BUILDDIRARMV7
- make install DESTDIR=$TMPDIR/dist.armv7
- popd
- if [ ! -h $TMPDIR/dist.armv7/$LIBDIR/libjpeg.@SO_MAJOR_VERSION@.dylib -a \
- ! -h $PKGROOT/$LIBDIR/libjpeg.@SO_MAJOR_VERSION@.dylib ]; then
- $LIPO -create \
- $PKGROOT/$LIBDIR/libjpeg.@SO_MAJOR_VERSION@.dylib \
- -arch arm $TMPDIR/dist.armv7/$LIBDIR/libjpeg.@SO_MAJOR_VERSION@.dylib \
- -output $PKGROOT/$LIBDIR/libjpeg.@SO_MAJOR_VERSION@.dylib
- elif [ ! -h $TMPDIR/dist.armv7/$LIBDIR/libjpeg.@SO_MAJOR_VERSION@.0.@SO_MINOR_VERSION@.dylib -a \
- ! -h $PKGROOT/$LIBDIR/libjpeg.@SO_MAJOR_VERSION@.0.@SO_MINOR_VERSION@.dylib ]; then
- $LIPO -create \
- $PKGROOT/$LIBDIR/libjpeg.@SO_MAJOR_VERSION@.0.@SO_MINOR_VERSION@.dylib \
- -arch arm $TMPDIR/dist.armv7/$LIBDIR/libjpeg.@SO_MAJOR_VERSION@.0.@SO_MINOR_VERSION@.dylib \
- -output $PKGROOT/$LIBDIR/libjpeg.@SO_MAJOR_VERSION@.0.@SO_MINOR_VERSION@.dylib
- fi
- $LIPO -create \
- $PKGROOT/$LIBDIR/libjpeg.a \
- -arch arm $TMPDIR/dist.armv7/$LIBDIR/libjpeg.a \
- -output $PKGROOT/$LIBDIR/libjpeg.a
- $LIPO -create \
- $PKGROOT/$LIBDIR/libturbojpeg.0.dylib \
- -arch arm $TMPDIR/dist.armv7/$LIBDIR/libturbojpeg.0.dylib \
- -output $PKGROOT/$LIBDIR/libturbojpeg.0.dylib
- $LIPO -create \
- $PKGROOT/$LIBDIR/libturbojpeg.a \
- -arch arm $TMPDIR/dist.armv7/$LIBDIR/libturbojpeg.a \
- -output $PKGROOT/$LIBDIR/libturbojpeg.a
- $LIPO -create \
- $PKGROOT/$BINDIR/cjpeg \
- -arch arm $TMPDIR/dist.armv7/$BINDIR/cjpeg \
- -output $PKGROOT/$BINDIR/cjpeg
- $LIPO -create \
- $PKGROOT/$BINDIR/djpeg \
- -arch arm $TMPDIR/dist.armv7/$BINDIR/djpeg \
- -output $PKGROOT/$BINDIR/djpeg
- $LIPO -create \
- $PKGROOT/$BINDIR/jpegtran \
- -arch arm $TMPDIR/dist.armv7/$BINDIR/jpegtran \
- -output $PKGROOT/$BINDIR/jpegtran
- $LIPO -create \
- $PKGROOT/$BINDIR/tjbench \
- -arch arm $TMPDIR/dist.armv7/$BINDIR/tjbench \
- -output $PKGROOT/$BINDIR/tjbench
- $LIPO -create \
- $PKGROOT/$BINDIR/rdjpgcom \
- -arch arm $TMPDIR/dist.armv7/$BINDIR/rdjpgcom \
- -output $PKGROOT/$BINDIR/rdjpgcom
- $LIPO -create \
- $PKGROOT/$BINDIR/wrjpgcom \
- -arch arm $TMPDIR/dist.armv7/$BINDIR/wrjpgcom \
- -output $PKGROOT/$BINDIR/wrjpgcom
+if [ $UNIVERSAL = 1 -a "$BUILDDIRARMV7S" != "" ]; then
+ install_ios $BUILDDIRARMV7S ARMv7s armv7s arm
fi
-if [ $BUILDARMV7S = 1 ]; then
- if [ ! -d $BUILDDIRARMV7S ]; then
- echo ERROR: ARMv7s build directory $BUILDDIRARMV7S does not exist
- exit 1
- fi
- if [ ! -f $BUILDDIRARMV7S/Makefile ]; then
- echo ERROR: ARMv7s build directory $BUILDDIRARMV7S is not configured
- exit 1
- fi
- mkdir -p $TMPDIR/dist.armv7s
- pushd $BUILDDIRARMV7S
- make install DESTDIR=$TMPDIR/dist.armv7s
- popd
- if [ ! -h $TMPDIR/dist.armv7s/$LIBDIR/libjpeg.@SO_MAJOR_VERSION@.dylib -a \
- ! -h $PKGROOT/$LIBDIR/libjpeg.@SO_MAJOR_VERSION@.dylib ]; then
- $LIPO -create \
- $PKGROOT/$LIBDIR/libjpeg.@SO_MAJOR_VERSION@.dylib \
- -arch arm $TMPDIR/dist.armv7s/$LIBDIR/libjpeg.@SO_MAJOR_VERSION@.dylib \
- -output $PKGROOT/$LIBDIR/libjpeg.@SO_MAJOR_VERSION@.dylib
- elif [ ! -h $TMPDIR/dist.armv7s/$LIBDIR/libjpeg.@SO_MAJOR_VERSION@.0.@SO_MINOR_VERSION@.dylib -a \
- ! -h $PKGROOT/$LIBDIR/libjpeg.@SO_MAJOR_VERSION@.0.@SO_MINOR_VERSION@.dylib ]; then
- $LIPO -create \
- $PKGROOT/$LIBDIR/libjpeg.@SO_MAJOR_VERSION@.0.@SO_MINOR_VERSION@.dylib \
- -arch arm $TMPDIR/dist.armv7s/$LIBDIR/libjpeg.@SO_MAJOR_VERSION@.0.@SO_MINOR_VERSION@.dylib \
- -output $PKGROOT/$LIBDIR/libjpeg.@SO_MAJOR_VERSION@.0.@SO_MINOR_VERSION@.dylib
- fi
- $LIPO -create \
- $PKGROOT/$LIBDIR/libjpeg.a \
- -arch arm $TMPDIR/dist.armv7s/$LIBDIR/libjpeg.a \
- -output $PKGROOT/$LIBDIR/libjpeg.a
- $LIPO -create \
- $PKGROOT/$LIBDIR/libturbojpeg.0.dylib \
- -arch arm $TMPDIR/dist.armv7s/$LIBDIR/libturbojpeg.0.dylib \
- -output $PKGROOT/$LIBDIR/libturbojpeg.0.dylib
- $LIPO -create \
- $PKGROOT/$LIBDIR/libturbojpeg.a \
- -arch arm $TMPDIR/dist.armv7s/$LIBDIR/libturbojpeg.a \
- -output $PKGROOT/$LIBDIR/libturbojpeg.a
- $LIPO -create \
- $PKGROOT/$BINDIR/cjpeg \
- -arch arm $TMPDIR/dist.armv7s/$BINDIR/cjpeg \
- -output $PKGROOT/$BINDIR/cjpeg
- $LIPO -create \
- $PKGROOT/$BINDIR/djpeg \
- -arch arm $TMPDIR/dist.armv7s/$BINDIR/djpeg \
- -output $PKGROOT/$BINDIR/djpeg
- $LIPO -create \
- $PKGROOT/$BINDIR/jpegtran \
- -arch arm $TMPDIR/dist.armv7s/$BINDIR/jpegtran \
- -output $PKGROOT/$BINDIR/jpegtran
- $LIPO -create \
- $PKGROOT/$BINDIR/tjbench \
- -arch arm $TMPDIR/dist.armv7s/$BINDIR/tjbench \
- -output $PKGROOT/$BINDIR/tjbench
- $LIPO -create \
- $PKGROOT/$BINDIR/rdjpgcom \
- -arch arm $TMPDIR/dist.armv7s/$BINDIR/rdjpgcom \
- -output $PKGROOT/$BINDIR/rdjpgcom
- $LIPO -create \
- $PKGROOT/$BINDIR/wrjpgcom \
- -arch arm $TMPDIR/dist.armv7s/$BINDIR/wrjpgcom \
- -output $PKGROOT/$BINDIR/wrjpgcom
+if [ $UNIVERSAL = 1 -a "BUILDDIRARMV8" != "" ]; then
+ install_ios $BUILDDIRARMV8 ARMv8 armv8 arm64
fi
-if [ $BUILDARMV8 = 1 ]; then
- if [ ! -d $BUILDDIRARMV8 ]; then
- echo ERROR: ARMv8 build directory $BUILDDIRARMV8 does not exist
- exit 1
- fi
- if [ ! -f $BUILDDIRARMV8/Makefile ]; then
- echo ERROR: ARMv8 build directory $BUILDDIRARMV8 is not configured
- exit 1
- fi
- mkdir -p $TMPDIR/dist.armv8
- pushd $BUILDDIRARMV8
- make install DESTDIR=$TMPDIR/dist.armv8
- popd
- if [ ! -h $TMPDIR/dist.armv8/$LIBDIR/libjpeg.@SO_MAJOR_VERSION@.dylib -a \
- ! -h $PKGROOT/$LIBDIR/libjpeg.@SO_MAJOR_VERSION@.dylib ]; then
- $LIPO -create \
- $PKGROOT/$LIBDIR/libjpeg.@SO_MAJOR_VERSION@.dylib \
- -arch arm64 $TMPDIR/dist.armv8/$LIBDIR/libjpeg.@SO_MAJOR_VERSION@.dylib \
- -output $PKGROOT/$LIBDIR/libjpeg.@SO_MAJOR_VERSION@.dylib
- elif [ ! -h $TMPDIR/dist.armv8/$LIBDIR/libjpeg.@SO_MAJOR_VERSION@.0.@SO_MINOR_VERSION@.dylib -a \
- ! -h $PKGROOT/$LIBDIR/libjpeg.@SO_MAJOR_VERSION@.0.@SO_MINOR_VERSION@.dylib ]; then
- $LIPO -create \
- $PKGROOT/$LIBDIR/libjpeg.@SO_MAJOR_VERSION@.0.@SO_MINOR_VERSION@.dylib \
- -arch arm64 $TMPDIR/dist.armv8/$LIBDIR/libjpeg.@SO_MAJOR_VERSION@.0.@SO_MINOR_VERSION@.dylib \
- -output $PKGROOT/$LIBDIR/libjpeg.@SO_MAJOR_VERSION@.0.@SO_MINOR_VERSION@.dylib
- fi
- $LIPO -create \
- $PKGROOT/$LIBDIR/libjpeg.a \
- -arch arm64 $TMPDIR/dist.armv8/$LIBDIR/libjpeg.a \
- -output $PKGROOT/$LIBDIR/libjpeg.a
- $LIPO -create \
- $PKGROOT/$LIBDIR/libturbojpeg.0.dylib \
- -arch arm64 $TMPDIR/dist.armv8/$LIBDIR/libturbojpeg.0.dylib \
- -output $PKGROOT/$LIBDIR/libturbojpeg.0.dylib
- $LIPO -create \
- $PKGROOT/$LIBDIR/libturbojpeg.a \
- -arch arm64 $TMPDIR/dist.armv8/$LIBDIR/libturbojpeg.a \
- -output $PKGROOT/$LIBDIR/libturbojpeg.a
- $LIPO -create \
- $PKGROOT/$BINDIR/cjpeg \
- -arch arm64 $TMPDIR/dist.armv8/$BINDIR/cjpeg \
- -output $PKGROOT/$BINDIR/cjpeg
- $LIPO -create \
- $PKGROOT/$BINDIR/djpeg \
- -arch arm64 $TMPDIR/dist.armv8/$BINDIR/djpeg \
- -output $PKGROOT/$BINDIR/djpeg
- $LIPO -create \
- $PKGROOT/$BINDIR/jpegtran \
- -arch arm64 $TMPDIR/dist.armv8/$BINDIR/jpegtran \
- -output $PKGROOT/$BINDIR/jpegtran
- $LIPO -create \
- $PKGROOT/$BINDIR/tjbench \
- -arch arm64 $TMPDIR/dist.armv8/$BINDIR/tjbench \
- -output $PKGROOT/$BINDIR/tjbench
- $LIPO -create \
- $PKGROOT/$BINDIR/rdjpgcom \
- -arch arm64 $TMPDIR/dist.armv8/$BINDIR/rdjpgcom \
- -output $PKGROOT/$BINDIR/rdjpgcom
- $LIPO -create \
- $PKGROOT/$BINDIR/wrjpgcom \
- -arch arm64 $TMPDIR/dist.armv8/$BINDIR/wrjpgcom \
- -output $PKGROOT/$BINDIR/wrjpgcom
-fi
-
-install_name_tool -id $LIBDIR/libjpeg.@SO_MAJOR_VERSION@.dylib $PKGROOT/$LIBDIR/libjpeg.@SO_MAJOR_VERSION@.dylib
-install_name_tool -id $LIBDIR/libturbojpeg.0.dylib $PKGROOT/$LIBDIR/libturbojpeg.0.dylib
+install_name_tool -id $LIBDIR/$LIBJPEG_DSO_NAME $PKGROOT/$LIBDIR/$LIBJPEG_DSO_NAME
+install_name_tool -id $LIBDIR/$TURBOJPEG_DSO_NAME $PKGROOT/$LIBDIR/$TURBOJPEG_DSO_NAME
if [ $WITH_JAVA = 1 ]; then
- ln -fs libturbojpeg.0.dylib $PKGROOT/$LIBDIR/libturbojpeg.jnilib
+ ln -fs $TURBOJPEG_DSO_NAME $PKGROOT/$LIBDIR/libturbojpeg.jnilib
fi
-if [ "$PREFIX" = "/opt/libjpeg-turbo" -a "$LIBDIR" = "/opt/libjpeg-turbo/lib" ]; then
+if [ "$PREFIX" = "@CMAKE_INSTALL_DEFAULT_PREFIX@" -a "$LIBDIR" = "@CMAKE_INSTALL_DEFAULT_PREFIX@/lib" ]; then
if [ ! -h $PKGROOT/$PREFIX/lib32 ]; then
ln -fs lib $PKGROOT/$PREFIX/lib32
fi
@@ -458,13 +256,13 @@
cp $SRCDIR/release/License.rtf $SRCDIR/release/Welcome.rtf $SRCDIR/release/ReadMe.txt $TMPDIR/pkg/
mkdir $TMPDIR/dmg
-pkgbuild --root $PKGROOT --version $VERSION.$BUILD \
- --identifier com.libjpeg-turbo.libjpeg-turbo $TMPDIR/pkg/$PACKAGE_NAME.pkg
-productbuild --distribution $SRCDIR/release/Distribution.xml \
+pkgbuild --root $PKGROOT --version $VERSION.$BUILD --identifier @PKGID@ \
+ $TMPDIR/pkg/$PKGNAME.pkg
+productbuild --distribution pkgscripts/Distribution.xml \
--package-path $TMPDIR/pkg/ --resources $TMPDIR/pkg/ \
- $TMPDIR/dmg/$PACKAGE_NAME.pkg
-hdiutil create -fs HFS+ -volname $PACKAGE_NAME-$VERSION \
- -srcfolder "$TMPDIR/dmg" $TMPDIR/$PACKAGE_NAME-$VERSION.dmg
-cp $TMPDIR/$PACKAGE_NAME-$VERSION.dmg .
+ $TMPDIR/dmg/$PKGNAME.pkg
+hdiutil create -fs HFS+ -volname $PKGNAME-$VERSION \
+ -srcfolder "$TMPDIR/dmg" $TMPDIR/$PKGNAME-$VERSION.dmg
+cp $TMPDIR/$PKGNAME-$VERSION.dmg .
exit
diff --git a/release/makerpm.in b/release/makerpm.in
new file mode 100644
index 0000000..fc3b1d4
--- /dev/null
+++ b/release/makerpm.in
@@ -0,0 +1,30 @@
+#!/bin/sh
+
+set -u
+set -e
+trap onexit INT
+trap onexit TERM
+trap onexit EXIT
+
+TMPDIR=
+
+onexit()
+{
+ if [ ! "$TMPDIR" = "" ]; then
+ rm -rf $TMPDIR
+ fi
+}
+
+if [ -f @PKGNAME@-@VERSION@.@RPMARCH@.rpm ]; then
+ rm -f @PKGNAME@-@VERSION@.@RPMARCH@.rpm
+fi
+
+umask 022
+TMPDIR=`mktemp -d /tmp/@CMAKE_PROJECT_NAME@-build.XXXXXX`
+
+mkdir -p $TMPDIR/RPMS
+ln -fs `pwd` $TMPDIR/BUILD
+rpmbuild -bb --define "_blddir $TMPDIR/buildroot" --define "_topdir $TMPDIR" \
+ --target @RPMARCH@ pkgscripts/rpm.spec; \
+cp $TMPDIR/RPMS/@RPMARCH@/@PKGNAME@-@VERSION@-@BUILD@.@RPMARCH@.rpm \
+ @PKGNAME@-@VERSION@.@RPMARCH@.rpm
diff --git a/release/makesrpm.in b/release/makesrpm.in
new file mode 100644
index 0000000..84c39d4
--- /dev/null
+++ b/release/makesrpm.in
@@ -0,0 +1,48 @@
+#!/bin/sh
+
+set -u
+set -e
+trap onexit INT
+trap onexit TERM
+trap onexit EXIT
+
+TMPDIR=
+
+onexit()
+{
+ if [ ! "$TMPDIR" = "" ]; then
+ rm -rf $TMPDIR
+ fi
+}
+
+PKGNAME=@PKGNAME@
+PROJECT=@CMAKE_PROJECT_NAME@
+VERSION=@VERSION@
+BUILD=@BUILD@
+
+if [ -f $PKGNAME-$VERSION.src.rpm ]; then
+ rm -f $PKGNAME-$VERSION.src.rpm
+fi
+
+umask 022
+TMPDIR=`mktemp -d /tmp/$PKGNAME-build.XXXXXX`
+
+mkdir -p $TMPDIR/RPMS
+mkdir -p $TMPDIR/SRPMS
+mkdir -p $TMPDIR/BUILD
+mkdir -p $TMPDIR/SOURCES
+mkdir -p $TMPDIR/SPECS
+
+if [ ! -f $PROJECT-$VERSION.tar.gz ]; then
+ echo "ERROR: $PROJECT-$VERSION.tar.gz does not exist."
+fi
+
+cp $PROJECT-$VERSION.tar.gz $TMPDIR/SOURCES/$PROJECT-$VERSION.tar.gz
+
+cat pkgscripts/rpm.spec | sed s/%{_blddir}/%{_tmppath}/g \
+ | sed s/#--\>//g > $TMPDIR/SPECS/$PKGNAME.spec
+
+rpmbuild -bs --define "_topdir $TMPDIR" $TMPDIR/SPECS/$PKGNAME.spec
+mv $TMPDIR/SRPMS/$PKGNAME-$VERSION-$BUILD.src.rpm $PKGNAME-$VERSION.src.rpm
+
+exit
diff --git a/release/maketarball.in b/release/maketarball.in
new file mode 100644
index 0000000..00a9c7e
--- /dev/null
+++ b/release/maketarball.in
@@ -0,0 +1,51 @@
+#!/bin/sh
+
+set -u
+set -e
+trap onexit INT
+trap onexit TERM
+trap onexit EXIT
+
+TMPDIR=
+SUDO=
+
+onexit()
+{
+ if [ ! "$TMPDIR" = "" ]; then
+ rm -rf $TMPDIR
+ fi
+}
+
+uid()
+{
+ id | cut -f2 -d = | cut -f1 -d \(;
+}
+
+PKGNAME=@PKGNAME@
+VERSION=@VERSION@
+ARCH=@CPU_TYPE@
+OS=@CMAKE_SYSTEM_NAME@
+PREFIX=@CMAKE_INSTALL_PREFIX@
+
+umask 022
+rm -f $PKGNAME-$VERSION-$OS-$ARCH.tar.bz2
+TMPDIR=`mktemp -d /tmp/$PKGNAME-build.XXXXXX`
+mkdir -p $TMPDIR/install
+
+make install DESTDIR=$TMPDIR/install
+echo tartest >$TMPDIR/tartest
+GNUTAR=0
+BSDTAR=0
+tar cf $TMPDIR/tartest.tar --owner=root --group=root -C $TMPDIR tartest >/dev/null 2>&1 && GNUTAR=1
+if [ "$GNUTAR" = "1" ]; then
+ tar cf - --owner=root --group=root -C $TMPDIR/install .$PREFIX | bzip2 -c >$PKGNAME-$VERSION-$OS-$ARCH.tar.bz2
+else
+ tar cf $TMPDIR/tartest.tar --uid 0 --gid 0 -C $TMPDIR tartest >/dev/null 2>&1 && BSDTAR=1
+ if [ "$BSDTAR" = "1" ]; then
+ tar cf - --uid=0 --gid=0 -C $TMPDIR/install .$PREFIX | bzip2 -c >$PKGNAME-$VERSION-$OS-$ARCH.tar.bz2
+ else
+ tar cf - -C $TMPDIR/install .$PREFIX | bzip2 -c >$PKGNAME-$VERSION-$OS-$ARCH.tar.bz2
+ fi
+fi
+
+exit
diff --git a/release/rpm.spec.in b/release/rpm.spec.in
new file mode 100644
index 0000000..36b0c51
--- /dev/null
+++ b/release/rpm.spec.in
@@ -0,0 +1,198 @@
+%define _prefix @CMAKE_INSTALL_PREFIX@
+%define _bindir @CMAKE_INSTALL_FULL_BINDIR@
+%define _datarootdir @CMAKE_INSTALL_FULL_DATAROOTDIR@
+%define _docdir %{_defaultdocdir}/%{name}-%{version}
+%define _includedir @CMAKE_INSTALL_FULL_INCLUDEDIR@
+%define _javadir @CMAKE_INSTALL_FULL_JAVADIR@
+%define _libdir @CMAKE_INSTALL_FULL_LIBDIR@
+%define _mandir @CMAKE_INSTALL_FULL_MANDIR@
+%define _enable_static @ENABLE_STATIC@
+%define _enable_shared @ENABLE_SHARED@
+%define _with_turbojpeg @WITH_TURBOJPEG@
+%define _with_java @WITH_JAVA@
+
+Summary: A SIMD-accelerated JPEG codec that provides both the libjpeg and TurboJPEG APIs
+Name: @PKGNAME@
+Version: @VERSION@
+Vendor: @PKGVENDOR@
+URL: @PKGURL@
+Group: System Environment/Libraries
+#-->Source0: http://prdownloads.sourceforge.net/@CMAKE_PROJECT_NAME@/@CMAKE_PROJECT_NAME@-%{version}.tar.gz
+Release: @BUILD@
+License: BSD-style
+BuildRoot: %{_blddir}/%{name}-buildroot-%{version}-%{release}
+Prereq: /sbin/ldconfig
+%ifarch x86_64
+Provides: %{name} = %{version}-%{release}, @CMAKE_PROJECT_NAME@ = %{version}-%{release}, libturbojpeg.so()(64bit)
+%else
+Provides: %{name} = %{version}-%{release}, @CMAKE_PROJECT_NAME@ = %{version}-%{release}, libturbojpeg.so
+%endif
+
+%description
+libjpeg-turbo is a JPEG image codec that uses SIMD instructions (MMX, SSE2,
+AVX2, NEON, AltiVec) to accelerate baseline JPEG compression and decompression
+on x86, x86-64, ARM, and PowerPC systems. On such systems, libjpeg-turbo is
+generally 2-6x as fast as libjpeg, all else being equal. On other types of
+systems, libjpeg-turbo can still outperform libjpeg by a significant amount, by
+virtue of its highly-optimized Huffman coding routines. In many cases, the
+performance of libjpeg-turbo rivals that of proprietary high-speed JPEG codecs.
+
+libjpeg-turbo implements both the traditional libjpeg API as well as the less
+powerful but more straightforward TurboJPEG API. libjpeg-turbo also features
+colorspace extensions that allow it to compress from/decompress to 32-bit and
+big-endian pixel buffers (RGBX, XBGR, etc.), as well as a full-featured Java
+interface.
+
+libjpeg-turbo was originally based on libjpeg/SIMD, an MMX-accelerated
+derivative of libjpeg v6b developed by Miyasaka Masaru. The TigerVNC and
+VirtualGL projects made numerous enhancements to the codec in 2009, and in
+early 2010, libjpeg-turbo spun off into an independent project, with the goal
+of making high-speed JPEG compression/decompression technology available to a
+broader range of users and developers.
+
+#-->%prep
+#-->%setup -q -n @CMAKE_PROJECT_NAME@-%{version}
+
+#-->%build
+#-->cmake -G"Unix Makefiles" -DCMAKE_BUILD_TYPE=@CMAKE_BUILD_TYPE@ \
+#--> -DBUILD=%{release} \
+#--> -DCMAKE_INSTALL_BINDIR=%{_bindir} \
+#--> -DCMAKE_INSTALL_DATAROOTDIR=%{_datarootdir} \
+#--> -DCMAKE_INSTALL_DOCDIR=%{_docdir} \
+#--> -DCMAKE_INSTALL_INCLUDEDIR=%{_includedir} \
+#--> -DCMAKE_INSTALL_JAVADIR=%{_javadir} \
+#--> -DCMAKE_INSTALL_LIBDIR=%{_libdir} \
+#--> -DCMAKE_INSTALL_MANDIR=%{_mandir} \
+#--> -DCMAKE_INSTALL_PREFIX=%{_prefix} \
+#--> -DCMAKE_POSITION_INDEPENDENT_CODE=@CMAKE_POSITION_INDEPENDENT_CODE@ \
+#--> -DENABLE_SHARED=@ENABLE_SHARED@ -DENABLE_STATIC=@ENABLE_STATIC@ \
+#--> -DSO_MAJOR_VERSION=@SO_MAJOR_VERSION@ \
+#--> -DSO_MINOR_VERSION=@SO_MINOR_VERSION@ \
+#--> -DJPEG_LIB_VERSION=@JPEG_LIB_VERSION@ \
+#--> -DREQUIRE_SIMD=@REQUIRE_SIMD@ \
+#--> -DWITH_12BIT=@WITH_12BIT@ -DWITH_ARITH_DEC=@WITH_ARITH_DEC@ \
+#--> -DWITH_ARITH_ENC=@WITH_ARITH_ENC@ -DWITH_JAVA=@WITH_JAVA@ \
+#--> -DWITH_JPEG7=@WITH_JPEG7@ -DWITH_JPEG8=@WITH_JPEG8@ \
+#--> -DWITH_MEM_SRCDST=@WITH_MEM_SRCDST@ -DWITH_SIMD=@WITH_SIMD@ \
+#--> -DWITH_TURBOJPEG=@WITH_TURBOJPEG@ .
+#-->make DESTDIR=$RPM_BUILD_ROOT
+
+%install
+
+rm -rf $RPM_BUILD_ROOT
+make install DESTDIR=$RPM_BUILD_ROOT
+/sbin/ldconfig -n $RPM_BUILD_ROOT%{_libdir}
+
+#-->%if 0
+
+# This is only needed to support in-tree RPM generation via 'make rpm'. When
+# building from a SRPM, we control where things are installed via CMake
+# variables.
+
+safedirmove ()
+{
+ if [ "$1" = "$2" ]; then
+ return 0
+ fi
+ if [ "$1" = "" -o ! -d "$1" ]; then
+ echo safedirmove: source dir $1 is not valid
+ return 1
+ fi
+ if [ "$2" = "" -o -e "$2" ]; then
+ echo safedirmove: dest dir $2 is not valid
+ return 1
+ fi
+ if [ "$3" = "" -o -e "$3" ]; then
+ echo safedirmove: tmp dir $3 is not valid
+ return 1
+ fi
+ mkdir -p $3
+ mv $1/* $3/
+ rmdir $1
+ mkdir -p $2
+ mv $3/* $2/
+ rmdir $3
+ return 0
+}
+
+LJT_DOCDIR=@CMAKE_INSTALL_FULL_DOCDIR@
+if [ ! "$LJT_DOCDIR" = "%{_docdir}" ]; then
+ safedirmove $RPM_BUILD_ROOT/$LJT_DOCDIR $RPM_BUILD_ROOT/%{_docdir} $RPM_BUILD_ROOT/__tmpdoc
+fi
+
+#-->%endif
+
+LJT_DOCDIR=@CMAKE_INSTALL_FULL_DOCDIR@
+if [ "%{_prefix}" = "@CMAKE_INSTALL_DEFAULT_PREFIX@" -a "$LJT_DOCDIR" = "@CMAKE_INSTALL_DEFAULT_PREFIX@/doc" ]; then
+ ln -fs %{_docdir} $RPM_BUILD_ROOT/$LJT_DOCDIR
+fi
+
+%post -p /sbin/ldconfig
+
+%postun -p /sbin/ldconfig
+
+%clean
+rm -rf $RPM_BUILD_ROOT
+
+%files
+%defattr(-,root,root)
+%dir %{_docdir}
+%doc %{_docdir}/*
+%dir %{_prefix}
+%if "%{_prefix}" == "@CMAKE_INSTALL_DEFAULT_PREFIX@" && "%{_docdir}" != "%{_prefix}/doc"
+ %{_prefix}/doc
+%endif
+%dir %{_bindir}
+%{_bindir}/cjpeg
+%{_bindir}/djpeg
+%{_bindir}/jpegtran
+%if "%{_with_turbojpeg}" == "1"
+ %{_bindir}/tjbench
+%endif
+%{_bindir}/rdjpgcom
+%{_bindir}/wrjpgcom
+%dir %{_libdir}
+%if "%{_enable_shared}" == "1"
+ %{_libdir}/libjpeg.so.@SO_MAJOR_VERSION@.@SO_AGE@.@SO_MINOR_VERSION@
+ %{_libdir}/libjpeg.so.@SO_MAJOR_VERSION@
+ %{_libdir}/libjpeg.so
+%endif
+%if "%{_enable_static}" == "1"
+ %{_libdir}/libjpeg.a
+%endif
+%{_libdir}/pkgconfig
+%{_libdir}/pkgconfig/libjpeg.pc
+%if "%{_with_turbojpeg}" == "1"
+ %if "%{_enable_shared}" == "1" || "%{_with_java}" == "1"
+ %{_libdir}/libturbojpeg.so.@TURBOJPEG_SO_VERSION@
+ %{_libdir}/libturbojpeg.so.@TURBOJPEG_SO_MAJOR_VERSION@
+ %{_libdir}/libturbojpeg.so
+ %endif
+ %if "%{_enable_static}" == "1"
+ %{_libdir}/libturbojpeg.a
+ %endif
+ %{_libdir}/pkgconfig/libturbojpeg.pc
+%endif
+%dir %{_includedir}
+%{_includedir}/jconfig.h
+%{_includedir}/jerror.h
+%{_includedir}/jmorecfg.h
+%{_includedir}/jpeglib.h
+%if "%{_with_turbojpeg}" == "1"
+ %{_includedir}/turbojpeg.h
+%endif
+%dir %{_mandir}
+%dir %{_mandir}/man1
+%{_mandir}/man1/cjpeg.1*
+%{_mandir}/man1/djpeg.1*
+%{_mandir}/man1/jpegtran.1*
+%{_mandir}/man1/rdjpgcom.1*
+%{_mandir}/man1/wrjpgcom.1*
+%if "%{_prefix}" != "%{_datarootdir}"
+ %dir %{_datarootdir}
+%endif
+%if "%{_with_java}" == "1"
+ %dir %{_javadir}
+ %{_javadir}/turbojpeg.jar
+%endif
+%changelog
diff --git a/release/uninstall.in b/release/uninstall.in
index 6cd1f86..cf1ba77 100644
--- a/release/uninstall.in
+++ b/release/uninstall.in
@@ -31,15 +31,15 @@
exit -1
fi
-PACKAGE=@PKGNAME@
-MACPACKAGE=com.$PACKAGE.$PACKAGE
-RECEIPT=/Library/Receipts/$PACKAGE.pkg
+PKGNAME=@PKGNAME@
+PKGID=@PKGID@
+RECEIPT=/Library/Receipts/$PKGNAME.pkg
LSBOM=
if [ -d $RECEIPT ]; then
LSBOM='lsbom -s -f -l '$RECEIPT'/Contents/Archive.bom'
else
- LSBOM='pkgutil --files '$MACPACKAGE
+ LSBOM='pkgutil --files '$PKGID
fi
mylsbom()
@@ -56,12 +56,13 @@
popd
echo Removing package directories ...
-PREFIX=%{__prefix}
-BINDIR=%{__bindir}
-DATADIR=%{__datadir}
-INCLUDEDIR=%{__includedir}
-LIBDIR=%{__libdir}
-MANDIR=%{__mandir}
+PREFIX=@CMAKE_INSTALL_PREFIX@
+BINDIR=@CMAKE_INSTALL_FULL_BINDIR@
+DATAROOTDIR=@CMAKE_INSTALL_FULL_DATAROOTDIR@
+INCLUDEDIR=@CMAKE_INSTALL_FULL_INCLUDEDIR@
+JAVADIR=@CMAKE_INSTALL_FULL_JAVADIR@
+LIBDIR=@CMAKE_INSTALL_FULL_LIBDIR@
+MANDIR=@CMAKE_INSTALL_FULL_MANDIR@
if [ -d $BINDIR ]; then
rmdir $BINDIR 2>&1 || EXITSTATUS=-1
@@ -75,7 +76,7 @@
if [ -d $INCLUDEDIR ]; then
rmdir $INCLUDEDIR 2>&1 || EXITSTATUS=-1
fi
-if [ "$PREFIX" = "/opt/libjpeg-turbo" -a "$LIBDIR" = "/opt/libjpeg-turbo/lib" ]; then
+if [ "$PREFIX" = "@CMAKE_INSTALL_DEFAULT_PREFIX@" -a "$LIBDIR" = "@CMAKE_INSTALL_DEFAULT_PREFIX@/lib" ]; then
if [ -h $LIBDIR\32 ]; then
rm $LIBDIR\32 2>&1 || EXITSTATUS=-1
fi
@@ -89,24 +90,24 @@
if [ -d $MANDIR ]; then
rmdir $MANDIR 2>&1 || EXITSTATUS=-1
fi
-if [ -d $DATADIR/classes ]; then
- rmdir $DATADIR/classes 2>&1 || EXITSTATUS=-1
+if [ -d $JAVADIR ]; then
+ rmdir $JAVADIR 2>&1 || EXITSTATUS=-1
fi
-if [ -d $DATADIR -a "$DATADIR" != "$PREFIX" ]; then
- rmdir $DATADIR 2>&1 || EXITSTATUS=-1
+if [ -d $DATAROOTDIR -a "$DATAROOTDIR" != "$PREFIX" ]; then
+ rmdir $DATAROOTDIR 2>&1 || EXITSTATUS=-1
fi
-if [ "$PREFIX" = "/opt/libjpeg-turbo" -a -h "$PREFIX/doc" ]; then
+if [ "$PREFIX" = "@CMAKE_INSTALL_DEFAULT_PREFIX@" -a -h "$PREFIX/doc" ]; then
rm $PREFIX/doc 2>&1 || EXITSTATUS=-1
fi
rmdir $PREFIX 2>&1 || EXITSTATUS=-1
-rmdir /Library/Documentation/$PACKAGE 2>&1 || EXITSTATUS=-1
+rmdir /Library/Documentation/$PKGNAME 2>&1 || EXITSTATUS=-1
if [ -d $RECEIPT ]; then
echo Removing package receipt ...
rm -r $RECEIPT 2>&1 || EXITSTATUS=-1
else
- echo Forgetting package $MACPACKAGE ...
- pkgutil --forget $MACPACKAGE
+ echo Forgetting package $PKGID ...
+ pkgutil --forget $PKGID
fi
exit $EXITSTATUS
diff --git a/sharedlib/CMakeLists.txt b/sharedlib/CMakeLists.txt
index d423cce..ef7e046 100755
--- a/sharedlib/CMakeLists.txt
+++ b/sharedlib/CMakeLists.txt
@@ -5,6 +5,10 @@
# better yet, provide a friendly way of configuring a Windows target to use the
# static C library.
+set(CMAKE_RUNTIME_OUTPUT_DIRECTORY ${CMAKE_CURRENT_BINARY_DIR}/..)
+set(CMAKE_LIBRARY_OUTPUT_DIRECTORY ${CMAKE_CURRENT_BINARY_DIR}/..)
+set(CMAKE_ARCHIVE_OUTPUT_DIRECTORY ${CMAKE_CURRENT_BINARY_DIR}/..)
+
if(MSVC)
# Build all configurations against shared C library
foreach(var CMAKE_C_FLAGS CMAKE_C_FLAGS_DEBUG CMAKE_C_FLAGS_RELEASE
@@ -16,36 +20,48 @@
endif()
foreach(src ${JPEG_SOURCES})
- set(JPEG_SRCS ${JPEG_SRCS} ${CMAKE_SOURCE_DIR}/${src})
+ set(JPEG_SRCS ${JPEG_SRCS} ../${src})
endforeach()
-if(WITH_SIMD)
+if(WITH_SIMD AND MSVC_IDE)
# This tells CMake that the "source" files haven't been generated yet
set_source_files_properties(${SIMD_OBJS} PROPERTIES GENERATED 1)
endif()
-if(WITH_MEM_SRCDST AND NOT WITH_JPEG8)
- add_library(jpeg SHARED ${JPEG_SRCS} ${SIMD_OBJS}
- ${CMAKE_SOURCE_DIR}/win/jpeg${DLL_VERSION}-memsrcdst.def)
-else()
- add_library(jpeg SHARED ${JPEG_SRCS} ${SIMD_OBJS}
- ${CMAKE_SOURCE_DIR}/win/jpeg${DLL_VERSION}.def)
+if(WIN32)
+ if(WITH_MEM_SRCDST)
+ set(DEFFILE ../win/jpeg${SO_MAJOR_VERSION}-memsrcdst.def)
+ else()
+ set(DEFFILE ../win/jpeg${SO_MAJOR_VERSION}.def)
+ endif()
endif()
-set_target_properties(jpeg PROPERTIES SOVERSION ${DLL_VERSION}
- VERSION ${FULLVERSION})
+add_library(jpeg SHARED ${JPEG_SRCS} ${DEFFILE} $<TARGET_OBJECTS:simd>
+ ${SIMD_OBJS})
+
+set_target_properties(jpeg PROPERTIES SOVERSION ${SO_MAJOR_VERSION}
+ VERSION ${SO_MAJOR_VERSION}.${SO_AGE}.${SO_MINOR_VERSION})
+if(APPLE)
+ set_target_properties(jpeg PROPERTIES MACOSX_RPATH 1)
+endif()
+if(MAPFLAG)
+ set_target_properties(jpeg PROPERTIES
+ LINK_FLAGS "${MAPFLAG}${CMAKE_CURRENT_BINARY_DIR}/../libjpeg.map")
+endif()
if(MSVC)
- set_target_properties(jpeg PROPERTIES SUFFIX ${DLL_VERSION}.dll)
-elseif(MINGW OR CYGWIN)
- set_target_properties(jpeg PROPERTIES SUFFIX -${DLL_VERSION}.dll)
-endif(MSVC)
-if(WITH_SIMD)
- add_dependencies(jpeg simd)
+ set_target_properties(jpeg PROPERTIES SUFFIX ${SO_MAJOR_VERSION}.dll)
+ # The jsimd_*.c file is built using /MT, so this prevents a linker warning.
+ set_target_properties(jpeg PROPERTIES LINK_FLAGS "/NODEFAULTLIB:LIBCMT /NODEFAULTLIB:LIBCMTD")
+elseif(MINGW)
+ set_target_properties(jpeg PROPERTIES SUFFIX -${SO_MAJOR_VERSION}.dll)
endif()
+if(WIN32)
+ set(USE_SETMODE "-DUSE_SETMODE")
+endif()
if(WITH_12BIT)
- set(COMPILE_FLAGS "-DGIF_SUPPORTED -DPPM_SUPPORTED -DUSE_SETMODE")
+ set(COMPILE_FLAGS "-DGIF_SUPPORTED -DPPM_SUPPORTED ${USE_SETMODE}")
else()
- set(COMPILE_FLAGS "-DBMP_SUPPORTED -DGIF_SUPPORTED -DPPM_SUPPORTED -DTARGA_SUPPORTED -DUSE_SETMODE")
+ set(COMPILE_FLAGS "-DBMP_SUPPORTED -DGIF_SUPPORTED -DPPM_SUPPORTED -DTARGA_SUPPORTED ${USE_SETMODE}")
set(CJPEG_BMP_SOURCES ../rdbmp.c ../rdtarga.c)
set(DJPEG_BMP_SOURCES ../wrbmp.c ../wrtarga.c)
endif()
@@ -62,12 +78,12 @@
add_executable(jpegtran ../jpegtran.c ../cdjpeg.c ../rdswitch.c ../transupp.c)
target_link_libraries(jpegtran jpeg)
-set_property(TARGET jpegtran PROPERTY COMPILE_FLAGS "-DUSE_SETMODE")
+set_property(TARGET jpegtran PROPERTY COMPILE_FLAGS "${USE_SETMODE}")
add_executable(jcstest ../jcstest.c)
target_link_libraries(jcstest jpeg)
install(TARGETS jpeg cjpeg djpeg jpegtran
- ARCHIVE DESTINATION lib
- LIBRARY DESTINATION lib
- RUNTIME DESTINATION bin)
+ ARCHIVE DESTINATION ${CMAKE_INSTALL_LIBDIR}
+ LIBRARY DESTINATION ${CMAKE_INSTALL_LIBDIR}
+ RUNTIME DESTINATION ${CMAKE_INSTALL_BINDIR})
diff --git a/simd/CMakeLists.txt b/simd/CMakeLists.txt
index 6e898d8..cfc24c1 100755
--- a/simd/CMakeLists.txt
+++ b/simd/CMakeLists.txt
@@ -1,81 +1,330 @@
-if(NOT DEFINED NASM)
- find_program(NASM NAMES nasm yasm DOC "Path to NASM/YASM executable")
-endif()
-message(STATUS "NASM = ${NASM}")
-
-if(SIMD_X86_64)
- set(NAFLAGS -fwin64 -DWIN64 -D__x86_64__)
-else()
- if(BORLAND)
- set(NAFLAGS -fobj -DOBJ32)
+macro(simd_fail message)
+ if(REQUIRE_SIMD)
+ message(FATAL_ERROR "${message}.")
else()
- set(NAFLAGS -fwin32 -DWIN32)
+ message(WARNING "${message}. Performance will suffer.")
+ set(WITH_SIMD 0 PARENT_SCOPE)
+ endif()
+endmacro()
+
+
+###############################################################################
+# x86[-64] (NASM)
+###############################################################################
+
+if(CPU_TYPE STREQUAL "x86_64" OR CPU_TYPE STREQUAL "i386")
+
+set(CMAKE_ASM_NASM_FLAGS_DEBUG_INIT "-g")
+set(CMAKE_ASM_NASM_FLAGS_RELWITHDEBINFO_INIT "-g")
+
+# Allow the location of the NASM executable to be specified using the ASM_NASM
+# environment variable. This should happen automatically, but unfortunately
+# enable_language(ASM_NASM) doesn't parse the ASM_NASM environment variable
+# until after CMAKE_ASM_NASM_COMPILER has been populated with the results of
+# searching for NASM or YASM in the PATH.
+if(NOT DEFINED CMAKE_ASM_NASM_COMPILER AND DEFINED ENV{ASM_NASM})
+ set(CMAKE_ASM_NASM_COMPILER $ENV{ASM_NASM})
+endif()
+
+if(CPU_TYPE STREQUAL "x86_64")
+ if(CYGWIN)
+ set(CMAKE_ASM_NASM_OBJECT_FORMAT win64)
+ endif()
+elseif(CPU_TYPE STREQUAL "i386")
+ if(BORLAND)
+ set(CMAKE_ASM_NASM_OBJECT_FORMAT obj)
+ elseif(CYGWIN)
+ set(CMAKE_ASM_NASM_OBJECT_FORMAT win32)
endif()
endif()
-set(NAFLAGS ${NAFLAGS} -I${CMAKE_SOURCE_DIR}/win/ -I${CMAKE_CURRENT_SOURCE_DIR}/)
-# This only works if building from the command line. There is currently no way
-# to set a variable's value based on the build type when using the MSVC IDE.
-if(CMAKE_BUILD_TYPE STREQUAL "Debug"
- OR CMAKE_BUILD_TYPE STREQUAL "RelWithDebInfo")
- set(NAFLAGS ${NAFLAGS} -g)
+enable_language(ASM_NASM)
+message(STATUS "CMAKE_ASM_NASM_COMPILER = ${CMAKE_ASM_NASM_COMPILER}")
+
+if(CMAKE_ASM_NASM_OBJECT_FORMAT MATCHES "macho*")
+ set(CMAKE_ASM_NASM_FLAGS "${CMAKE_ASM_NASM_FLAGS} -DMACHO")
+elseif(CMAKE_ASM_NASM_OBJECT_FORMAT MATCHES "elf*")
+ set(CMAKE_ASM_NASM_FLAGS "${CMAKE_ASM_NASM_FLAGS} -DELF")
+ set(CMAKE_ASM_NASM_DEBUG_FORMAT "dwarf2")
+endif()
+if(CPU_TYPE STREQUAL "x86_64")
+ if(WIN32 OR CYGWIN)
+ set(CMAKE_ASM_NASM_FLAGS "${CMAKE_ASM_NASM_FLAGS} -DWIN64")
+ endif()
+ set(CMAKE_ASM_NASM_FLAGS "${CMAKE_ASM_NASM_FLAGS} -D__x86_64__")
+elseif(CPU_TYPE STREQUAL "i386")
+ if(BORLAND)
+ set(CMAKE_ASM_NASM_FLAGS "${CMAKE_ASM_NASM_FLAGS} -DOBJ32")
+ elseif(WIN32 OR CYGWIN)
+ set(CMAKE_ASM_NASM_FLAGS "${CMAKE_ASM_NASM_FLAGS} -DWIN32")
+ endif()
endif()
-if(SIMD_X86_64)
- set(SIMD_BASENAMES jfdctflt-sse-64 jccolor-sse2-64 jcgray-sse2-64
- jchuff-sse2-64 jcsample-sse2-64 jdcolor-sse2-64 jdmerge-sse2-64
- jdsample-sse2-64 jfdctfst-sse2-64 jfdctint-sse2-64 jidctflt-sse2-64
- jidctfst-sse2-64 jidctint-sse2-64 jidctred-sse2-64 jquantf-sse2-64
- jquanti-sse2-64)
- message(STATUS "Building x86_64 SIMD extensions")
+message(STATUS "CMAKE_ASM_NASM_OBJECT_FORMAT = ${CMAKE_ASM_NASM_OBJECT_FORMAT}")
+
+if(NOT CMAKE_ASM_NASM_OBJECT_FORMAT)
+ simd_fail("SIMD extensions disabled: could not determine NASM object format")
+ return()
+endif()
+
+get_filename_component(CMAKE_ASM_NASM_COMPILER_TYPE
+ "${CMAKE_ASM_NASM_COMPILER}" NAME_WE)
+if(CMAKE_ASM_NASM_COMPILER_TYPE MATCHES "yasm")
+ foreach(var CMAKE_ASM_NASM_FLAGS_DEBUG CMAKE_ASM_NASM_FLAGS_RELWITHDEBINFO)
+ if(${var} STREQUAL "-g")
+ if(CMAKE_ASM_NASM_DEBUG_FORMAT)
+ set_property(CACHE ${var} PROPERTY VALUE "-g ${CMAKE_ASM_NASM_DEBUG_FORMAT}")
+ else()
+ set_property(CACHE ${var} PROPERTY VALUE "")
+ endif()
+ endif()
+ endforeach()
+endif()
+
+if(NOT WIN32 AND (CMAKE_POSITION_INDEPENDENT_CODE OR ENABLE_SHARED))
+ set(CMAKE_ASM_NASM_FLAGS "${CMAKE_ASM_NASM_FLAGS} -DPIC")
+endif()
+
+string(TOUPPER ${CMAKE_BUILD_TYPE} CMAKE_BUILD_TYPE_UC)
+set(EFFECTIVE_ASM_NASM_FLAGS "${CMAKE_ASM_NASM_FLAGS} ${CMAKE_ASM_NASM_FLAGS_${CMAKE_BUILD_TYPE_UC}}")
+message(STATUS "CMAKE_ASM_NASM_FLAGS = ${EFFECTIVE_ASM_NASM_FLAGS}")
+
+set(CMAKE_ASM_NASM_FLAGS "${CMAKE_ASM_NASM_FLAGS} -I\"${CMAKE_CURRENT_SOURCE_DIR}/nasm/\" -I\"${CMAKE_CURRENT_SOURCE_DIR}/${CPU_TYPE}/\"")
+
+if(WIN32)
+ set(CMAKE_ASM_NASM_FLAGS "${CMAKE_ASM_NASM_FLAGS} -I\"${CMAKE_CURRENT_SOURCE_DIR}/../win/\"")
+ set(JSIMDCFG_INC ${CMAKE_CURRENT_SOURCE_DIR}/../win/jsimdcfg.inc)
else()
- set(SIMD_BASENAMES jsimdcpu jfdctflt-3dn jidctflt-3dn jquant-3dn jccolor-mmx
- jcgray-mmx jcsample-mmx jdcolor-mmx jdmerge-mmx jdsample-mmx jfdctfst-mmx
- jfdctint-mmx jidctfst-mmx jidctint-mmx jidctred-mmx jquant-mmx jfdctflt-sse
- jidctflt-sse jquant-sse jccolor-sse2 jcgray-sse2 jchuff-sse2 jcsample-sse2
- jdcolor-sse2 jdmerge-sse2 jdsample-sse2 jfdctfst-sse2 jfdctint-sse2
- jidctflt-sse2 jidctfst-sse2 jidctint-sse2 jidctred-sse2 jquantf-sse2
- jquanti-sse2)
- message(STATUS "Building i386 SIMD extensions")
+ set(GREP grep)
+ if(CMAKE_SYSTEM_NAME STREQUAL "SunOS")
+ set(GREP ggrep)
+ endif()
+ add_custom_command(OUTPUT jsimdcfg.inc
+ COMMAND ${CMAKE_C_COMPILER} -E -I${CMAKE_BINARY_DIR} -I${CMAKE_CURRENT_BINARY_DIR} ${CMAKE_CURRENT_SOURCE_DIR}/jsimdcfg.inc.h | ${GREP} -E '^[\;%]|^\ %' | sed 's%_cpp_protection_%%' | sed 's@% define@%define@g' >jsimdcfg.inc)
+ set(JSIMDCFG_INC ${CMAKE_CURRENT_BINARY_DIR}/jsimdcfg.inc)
+ set(CMAKE_ASM_NASM_FLAGS "${CMAKE_ASM_NASM_FLAGS} -I\"${CMAKE_CURRENT_BINARY_DIR}/\"")
+endif()
+
+if(CPU_TYPE STREQUAL "x86_64")
+ set(SIMD_SOURCES x86_64/jsimdcpu.asm x86_64/jfdctflt-sse.asm
+ x86_64/jccolor-sse2.asm x86_64/jcgray-sse2.asm x86_64/jchuff-sse2.asm
+ x86_64/jcsample-sse2.asm x86_64/jdcolor-sse2.asm x86_64/jdmerge-sse2.asm
+ x86_64/jdsample-sse2.asm x86_64/jfdctfst-sse2.asm x86_64/jfdctint-sse2.asm
+ x86_64/jidctflt-sse2.asm x86_64/jidctfst-sse2.asm x86_64/jidctint-sse2.asm
+ x86_64/jidctred-sse2.asm x86_64/jquantf-sse2.asm x86_64/jquanti-sse2.asm
+ x86_64/jccolor-avx2.asm x86_64/jcgray-avx2.asm x86_64/jcsample-avx2.asm
+ x86_64/jdcolor-avx2.asm x86_64/jdmerge-avx2.asm x86_64/jdsample-avx2.asm
+ x86_64/jquanti-avx2.asm)
+else()
+ set(SIMD_SOURCES i386/jsimdcpu.asm i386/jfdctflt-3dn.asm
+ i386/jidctflt-3dn.asm i386/jquant-3dn.asm
+ i386/jccolor-mmx.asm i386/jcgray-mmx.asm i386/jcsample-mmx.asm
+ i386/jdcolor-mmx.asm i386/jdmerge-mmx.asm i386/jdsample-mmx.asm
+ i386/jfdctfst-mmx.asm i386/jfdctint-mmx.asm i386/jidctfst-mmx.asm
+ i386/jidctint-mmx.asm i386/jidctred-mmx.asm i386/jquant-mmx.asm
+ i386/jfdctflt-sse.asm i386/jidctflt-sse.asm i386/jquant-sse.asm
+ i386/jccolor-sse2.asm i386/jcgray-sse2.asm i386/jchuff-sse2.asm
+ i386/jcsample-sse2.asm i386/jdcolor-sse2.asm i386/jdmerge-sse2.asm
+ i386/jdsample-sse2.asm i386/jfdctfst-sse2.asm i386/jfdctint-sse2.asm
+ i386/jidctflt-sse2.asm i386/jidctfst-sse2.asm i386/jidctint-sse2.asm
+ i386/jidctred-sse2.asm i386/jquantf-sse2.asm i386/jquanti-sse2.asm
+ i386/jccolor-avx2.asm i386/jcgray-avx2.asm i386/jcsample-avx2.asm
+ i386/jdcolor-avx2.asm i386/jdmerge-avx2.asm i386/jdsample-avx2.asm
+ i386/jquanti-avx2.asm)
endif()
if(MSVC_IDE)
set(OBJDIR "${CMAKE_CURRENT_BINARY_DIR}/${CMAKE_CFG_INTDIR}")
-else()
- set(OBJDIR ${CMAKE_CURRENT_BINARY_DIR})
+ string(REGEX REPLACE " " ";" CMAKE_ASM_NASM_FLAGS "${CMAKE_ASM_NASM_FLAGS}")
endif()
-file(GLOB INC_FILES *.inc)
+file(GLOB INC_FILES nasm/*.inc)
-foreach(file ${SIMD_BASENAMES})
- set(DEPFILE "")
- set(SIMD_SRC ${CMAKE_CURRENT_SOURCE_DIR}/${file}.asm)
+foreach(file ${SIMD_SOURCES})
+ set(OBJECT_DEPENDS "")
if(${file} MATCHES jccolor)
- set(DEPFILE ${file})
- string(REGEX REPLACE "jccolor" "jccolext" DEPFILE ${DEPFILE})
- set(DEPFILE ${CMAKE_CURRENT_SOURCE_DIR}/${DEPFILE}.asm)
+ string(REGEX REPLACE "jccolor" "jccolext" DEPFILE ${file})
+ set(OBJECT_DEPENDS ${OBJECT_DEPENDS}
+ ${CMAKE_CURRENT_SOURCE_DIR}/${DEPFILE})
endif()
if(${file} MATCHES jcgray)
- set(DEPFILE ${file})
- string(REGEX REPLACE "jcgray" "jcgryext" DEPFILE ${DEPFILE})
- set(DEPFILE ${CMAKE_CURRENT_SOURCE_DIR}/${DEPFILE}.asm)
+ string(REGEX REPLACE "jcgray" "jcgryext" DEPFILE ${file})
+ set(OBJECT_DEPENDS ${OBJECT_DEPENDS}
+ ${CMAKE_CURRENT_SOURCE_DIR}/${DEPFILE})
endif()
if(${file} MATCHES jdcolor)
- set(DEPFILE ${file})
- string(REGEX REPLACE "jdcolor" "jdcolext" DEPFILE ${DEPFILE})
- set(DEPFILE ${CMAKE_CURRENT_SOURCE_DIR}/${DEPFILE}.asm)
+ string(REGEX REPLACE "jdcolor" "jdcolext" DEPFILE ${file})
+ set(OBJECT_DEPENDS ${OBJECT_DEPENDS}
+ ${CMAKE_CURRENT_SOURCE_DIR}/${DEPFILE})
endif()
if(${file} MATCHES jdmerge)
- set(DEPFILE ${file})
- string(REGEX REPLACE "jdmerge" "jdmrgext" DEPFILE ${DEPFILE})
- set(DEPFILE ${CMAKE_CURRENT_SOURCE_DIR}/${DEPFILE}.asm)
+ string(REGEX REPLACE "jdmerge" "jdmrgext" DEPFILE ${file})
+ set(OBJECT_DEPENDS ${OBJECT_DEPENDS}
+ ${CMAKE_CURRENT_SOURCE_DIR}/${DEPFILE})
endif()
- set(SIMD_OBJ ${OBJDIR}/${file}.obj)
- add_custom_command(OUTPUT ${SIMD_OBJ}
- DEPENDS ${SIMD_SRC} ${DEPFILE} ${INC_FILES}
- COMMAND ${NASM} ${NAFLAGS} ${SIMD_SRC} -o${SIMD_OBJ})
- set(SIMD_OBJS ${SIMD_OBJS} ${SIMD_OBJ})
+ set(OBJECT_DEPENDS ${OBJECT_DEPENDS} ${INC_FILES} ${JSIMDCFG_INC})
+ if(MSVC_IDE)
+ # The CMake Visual Studio generators do not work properly with the ASM_NASM
+ # language, so we have to go rogue here and use a custom command like we
+ # did in prior versions of libjpeg-turbo. (This is why we can't have nice
+ # things.)
+ string(REGEX REPLACE "${CPU_TYPE}/" "" filename ${file})
+ set(SIMD_OBJ ${OBJDIR}/${filename}.obj)
+ add_custom_command(OUTPUT ${SIMD_OBJ} DEPENDS ${file} ${OBJECT_DEPENDS}
+ COMMAND ${CMAKE_ASM_NASM_COMPILER} -f${CMAKE_ASM_NASM_OBJECT_FORMAT}
+ ${CMAKE_ASM_NASM_FLAGS} ${CMAKE_CURRENT_SOURCE_DIR}/${file}
+ -o${SIMD_OBJ})
+ set(SIMD_OBJS ${SIMD_OBJS} ${SIMD_OBJ})
+ else()
+ set_source_files_properties(${file} PROPERTIES OBJECT_DEPENDS
+ "${OBJECT_DEPENDS}")
+ endif()
endforeach()
-set(SIMD_OBJS ${SIMD_OBJS} PARENT_SCOPE)
-add_custom_target(simd DEPENDS ${SIMD_OBJS})
+if(MSVC_IDE)
+ set(SIMD_OBJS ${SIMD_OBJS} PARENT_SCOPE)
+ add_library(simd OBJECT ${CPU_TYPE}/jsimd.c)
+ add_custom_target(simd-objs DEPENDS ${SIMD_OBJS})
+ add_dependencies(simd simd-objs)
+else()
+ add_library(simd OBJECT ${SIMD_SOURCES} ${CPU_TYPE}/jsimd.c)
+endif()
+if(NOT WIN32 AND (CMAKE_POSITION_INDEPENDENT_CODE OR ENABLE_SHARED))
+ set_target_properties(simd PROPERTIES POSITION_INDEPENDENT_CODE 1)
+endif()
+
+
+###############################################################################
+# ARM (GAS)
+###############################################################################
+
+elseif(CPU_TYPE STREQUAL "arm64" OR CPU_TYPE STREQUAL "arm")
+
+enable_language(ASM)
+
+set(CMAKE_ASM_FLAGS ${CMAKE_C_FLAGS}${CMAKE_ASM_FLAGS})
+
+string(TOUPPER ${CMAKE_BUILD_TYPE} CMAKE_BUILD_TYPE_UC)
+set(EFFECTIVE_ASM_FLAGS "${CMAKE_ASM_FLAGS} ${CMAKE_ASM_FLAGS_${CMAKE_BUILD_TYPE_UC}}")
+message(STATUS "CMAKE_ASM_FLAGS = ${EFFECTIVE_ASM_FLAGS}")
+
+# Test whether we need gas-preprocessor.pl
+if(CPU_TYPE STREQUAL "arm")
+ file(WRITE ${CMAKE_CURRENT_BINARY_DIR}/gastest.S "
+ .text
+ .fpu neon
+ .arch armv7a
+ .object_arch armv4
+ .arm
+ pld [r0]
+ vmovn.u16 d0, q0")
+else()
+ file(WRITE ${CMAKE_CURRENT_BINARY_DIR}/gastest.S "
+ .text
+ MYVAR .req x0
+ movi v0.16b, #100
+ mov MYVAR, #100
+ .unreq MYVAR")
+endif()
+
+separate_arguments(CMAKE_ASM_FLAGS_SEP UNIX_COMMAND "${CMAKE_ASM_FLAGS}")
+
+execute_process(COMMAND ${CMAKE_ASM_COMPILER} ${CMAKE_ASM_FLAGS_SEP}
+ -x assembler-with-cpp -c ${CMAKE_CURRENT_BINARY_DIR}/gastest.S
+ RESULT_VARIABLE RESULT OUTPUT_VARIABLE OUTPUT ERROR_VARIABLE ERROR)
+if(NOT RESULT EQUAL 0)
+ message(STATUS "GAS appears to be broken. Trying gas-preprocessor.pl ...")
+ execute_process(COMMAND gas-preprocessor.pl ${CMAKE_ASM_COMPILER}
+ ${CMAKE_ASM_FLAGS_SEP} -x assembler-with-cpp -c
+ ${CMAKE_CURRENT_BINARY_DIR}/gastest.S
+ RESULT_VARIABLE RESULT OUTPUT_VARIABLE OUTPUT ERROR_VARIABLE ERROR)
+ if(NOT RESULT EQUAL 0)
+ simd_fail("SIMD extensions disabled: GAS is not working properly")
+ return()
+ else()
+ message(STATUS "Using gas-preprocessor.pl")
+ configure_file(gas-preprocessor.in gas-preprocessor @ONLY)
+ set(CMAKE_ASM_COMPILER ${CMAKE_CURRENT_BINARY_DIR}/gas-preprocessor)
+ endif()
+else()
+ message(STATUS "GAS is working properly")
+endif()
+
+file(REMOVE ${CMAKE_CURRENT_BINARY_DIR}/gastest.S)
+
+add_library(simd OBJECT ${CPU_TYPE}/jsimd_neon.S ${CPU_TYPE}/jsimd.c)
+
+if(CMAKE_POSITION_INDEPENDENT_CODE OR ENABLE_SHARED)
+ set_target_properties(simd PROPERTIES POSITION_INDEPENDENT_CODE 1)
+endif()
+
+
+###############################################################################
+# MIPS (GAS)
+###############################################################################
+
+elseif(CPU_TYPE STREQUAL "mips")
+
+enable_language(ASM)
+
+string(TOUPPER ${CMAKE_BUILD_TYPE} CMAKE_BUILD_TYPE_UC)
+set(EFFECTIVE_ASM_FLAGS "${CMAKE_ASM_FLAGS} ${CMAKE_ASM_FLAGS_${CMAKE_BUILD_TYPE_UC}}")
+message(STATUS "CMAKE_ASM_FLAGS = ${EFFECTIVE_ASM_FLAGS}")
+
+add_library(simd OBJECT ${CPU_TYPE}/jsimd_dspr2.S ${CPU_TYPE}/jsimd.c)
+
+if(CMAKE_POSITION_INDEPENDENT_CODE OR ENABLE_SHARED)
+ set_target_properties(simd PROPERTIES POSITION_INDEPENDENT_CODE 1)
+endif()
+
+
+###############################################################################
+# PowerPC (Intrinsics)
+###############################################################################
+
+elseif(CPU_TYPE STREQUAL "powerpc")
+
+set(CMAKE_REQUIRED_FLAGS -maltivec)
+
+check_c_source_compiles("
+ #include <altivec.h>
+ int main(void) {
+ __vector int vi = { 0, 0, 0, 0 };
+ int i[4];
+ vec_st(vi, 0, i);
+ return i[0];
+ }" HAVE_ALTIVEC)
+
+unset(CMAKE_REQUIRED_FLAGS)
+
+if(NOT HAVE_ALTIVEC)
+ simd_fail("SIMD extensions not available for this CPU (PowerPC SPE)")
+ return()
+endif()
+
+set(SIMD_SOURCES powerpc/jccolor-altivec.c powerpc/jcgray-altivec.c
+ powerpc/jcsample-altivec.c powerpc/jdcolor-altivec.c
+ powerpc/jdmerge-altivec.c powerpc/jdsample-altivec.c
+ powerpc/jfdctfst-altivec.c powerpc/jfdctint-altivec.c
+ powerpc/jidctfst-altivec.c powerpc/jidctint-altivec.c
+ powerpc/jquanti-altivec.c)
+
+set_source_files_properties(${SIMD_SOURCES} PROPERTIES
+ COMPILE_FLAGS -maltivec)
+
+add_library(simd OBJECT ${SIMD_SOURCES} powerpc/jsimd.c)
+
+if(CMAKE_POSITION_INDEPENDENT_CODE OR ENABLE_SHARED)
+ set_target_properties(simd PROPERTIES POSITION_INDEPENDENT_CODE 1)
+endif()
+
+
+###############################################################################
+# None
+###############################################################################
+
+else()
+
+simd_fail("SIMD extensions not available for this CPU (${CMAKE_SYSTEM_PROCESSOR})")
+
+endif() # CPU_TYPE
diff --git a/simd/Makefile.am b/simd/Makefile.am
deleted file mode 100644
index b8660d1..0000000
--- a/simd/Makefile.am
+++ /dev/null
@@ -1,102 +0,0 @@
-noinst_LTLIBRARIES = libsimd.la
-
-BUILT_SOURCES = jsimdcfg.inc
-
-EXTRA_DIST = nasm_lt.sh CMakeLists.txt \
- jccolext-mmx.asm jcgryext-mmx.asm jdcolext-mmx.asm jdmrgext-mmx.asm \
- jccolext-sse2.asm jcgryext-sse2.asm jdcolext-sse2.asm jdmrgext-sse2.asm \
- jccolext-sse2-64.asm jcgryext-sse2-64.asm jdcolext-sse2-64.asm \
- jdmrgext-sse2-64.asm jccolext-altivec.c jcgryext-altivec.c \
- jdcolext-altivec.c jdmrgext-altivec.c
-
-if SIMD_X86_64
-
-libsimd_la_SOURCES = jsimd_x86_64.c jsimd.h jsimdcfg.inc.h jsimdext.inc \
- jcolsamp.inc jdct.inc jpeg_nbits_table.inc jfdctflt-sse-64.asm \
- jccolor-sse2-64.asm jcgray-sse2-64.asm jchuff-sse2-64.asm \
- jcsample-sse2-64.asm jdcolor-sse2-64.asm jdmerge-sse2-64.asm \
- jdsample-sse2-64.asm jfdctfst-sse2-64.asm jfdctint-sse2-64.asm \
- jidctflt-sse2-64.asm jidctfst-sse2-64.asm jidctint-sse2-64.asm \
- jidctred-sse2-64.asm jquantf-sse2-64.asm jquanti-sse2-64.asm
-
-jccolor-sse2-64.lo: jccolext-sse2-64.asm
-jcgray-sse2-64.lo: jcgryext-sse2-64.asm
-jdcolor-sse2-64.lo: jdcolext-sse2-64.asm
-jdmerge-sse2-64.lo: jdmrgext-sse2-64.asm
-
-endif
-
-if SIMD_I386
-
-libsimd_la_SOURCES = jsimd_i386.c jsimd.h jsimdcfg.inc.h jsimdext.inc \
- jcolsamp.inc jdct.inc jpeg_nbits_table.inc jsimdcpu.asm \
- jfdctflt-3dn.asm jidctflt-3dn.asm jquant-3dn.asm \
- jccolor-mmx.asm jcgray-mmx.asm jcsample-mmx.asm \
- jdcolor-mmx.asm jdmerge-mmx.asm jdsample-mmx.asm \
- jfdctfst-mmx.asm jfdctint-mmx.asm jidctfst-mmx.asm \
- jidctint-mmx.asm jidctred-mmx.asm jquant-mmx.asm \
- jfdctflt-sse.asm jidctflt-sse.asm jquant-sse.asm \
- jccolor-sse2.asm jcgray-sse2.asm jchuff-sse2.asm \
- jcsample-sse2.asm jdcolor-sse2.asm jdmerge-sse2.asm \
- jdsample-sse2.asm jfdctfst-sse2.asm jfdctint-sse2.asm \
- jidctflt-sse2.asm jidctfst-sse2.asm jidctint-sse2.asm \
- jidctred-sse2.asm jquantf-sse2.asm jquanti-sse2.asm
-
-jccolor-mmx.lo: jccolext-mmx.asm
-jcgray.-mmx.lo: jcgryext-mmx.asm
-jdcolor-mmx.lo: jdcolext-mmx.asm
-jdmerge-mmx.lo: jdmrgext-mmx.asm
-jccolor-sse2.lo: jccolext-sse2.asm
-jcgray-sse2.lo: jcgryext-sse2.asm
-jdcolor-sse2.lo: jdcolext-sse2.asm
-jdmerge-sse2.lo: jdmrgext-sse2.asm
-
-endif
-
-if SIMD_ARM
-
-libsimd_la_SOURCES = jsimd_arm.c jsimd_arm_neon.S
-
-endif
-
-if SIMD_ARM_64
-
-libsimd_la_SOURCES = jsimd_arm64.c jsimd_arm64_neon.S
-
-endif
-
-if SIMD_MIPS
-
-libsimd_la_SOURCES = jsimd_mips.c jsimd_mips_dspr2_asm.h jsimd_mips_dspr2.S
-
-endif
-
-if SIMD_POWERPC
-
-noinst_LTLIBRARIES += libsimd_altivec.la
-
-libsimd_altivec_la_SOURCES = \
- jccolor-altivec.c jcgray-altivec.c jcsample-altivec.c \
- jdcolor-altivec.c jdmerge-altivec.c jdsample-altivec.c \
- jfdctfst-altivec.c jfdctint-altivec.c \
- jidctfst-altivec.c jidctint-altivec.c \
- jquanti-altivec.c
-libsimd_altivec_la_CFLAGS = -maltivec
-
-jccolor-altivec.lo: jccolext-altivec.c
-jcgray-altivec.lo: jcgryext-altivec.c
-jdcolor-altivec.lo: jdcolext-altivec.c
-jdmerge-altivec.lo: jdmrgext-altivec.c
-
-libsimd_la_SOURCES = jsimd_powerpc.c jsimd_altivec.h jcsample.h
-libsimd_la_LIBADD = libsimd_altivec.la
-
-endif
-
-AM_CPPFLAGS = -I$(top_srcdir)
-
-.asm.lo:
- $(AM_V_GEN) $(LIBTOOL) $(AM_V_lt) --mode=compile --tag NASM $(srcdir)/nasm_lt.sh $(AM_V_lt) $(NASM) $(NAFLAGS) -I$(srcdir) -I. $< -o $@
-
-jsimdcfg.inc: $(srcdir)/jsimdcfg.inc.h ../jpeglib.h ../jconfig.h ../jmorecfg.h
- $(AM_V_GEN) $(CPP) -I$(top_builddir) -I$(top_builddir)/simd $(srcdir)/jsimdcfg.inc.h | $(EGREP) "^[\;%]|^\ %" | sed 's%_cpp_protection_%%' | sed 's@% define@%define@g' > $@
diff --git a/simd/jsimd_arm.c b/simd/arm/jsimd.c
similarity index 98%
rename from simd/jsimd_arm.c
rename to simd/arm/jsimd.c
index 0b955cd..1963349 100644
--- a/simd/jsimd_arm.c
+++ b/simd/arm/jsimd.c
@@ -16,12 +16,12 @@
*/
#define JPEG_INTERNALS
-#include "../jinclude.h"
-#include "../jpeglib.h"
+#include "../../jinclude.h"
+#include "../../jpeglib.h"
+#include "../../jsimd.h"
+#include "../../jdct.h"
+#include "../../jsimddct.h"
#include "../jsimd.h"
-#include "../jdct.h"
-#include "../jsimddct.h"
-#include "jsimd.h"
#include <stdio.h>
#include <string.h>
diff --git a/simd/jsimd_arm_neon.S b/simd/arm/jsimd_neon.S
similarity index 100%
rename from simd/jsimd_arm_neon.S
rename to simd/arm/jsimd_neon.S
diff --git a/simd/jsimd_arm64.c b/simd/arm64/jsimd.c
similarity index 98%
rename from simd/jsimd_arm64.c
rename to simd/arm64/jsimd.c
index f6e9736..12a7822 100644
--- a/simd/jsimd_arm64.c
+++ b/simd/arm64/jsimd.c
@@ -16,12 +16,12 @@
*/
#define JPEG_INTERNALS
-#include "../jinclude.h"
-#include "../jpeglib.h"
+#include "../../jinclude.h"
+#include "../../jpeglib.h"
+#include "../../jsimd.h"
+#include "../../jdct.h"
+#include "../../jsimddct.h"
#include "../jsimd.h"
-#include "../jdct.h"
-#include "../jsimddct.h"
-#include "jsimd.h"
#include <stdio.h>
#include <string.h>
diff --git a/simd/jsimd_arm64_neon.S b/simd/arm64/jsimd_neon.S
similarity index 100%
rename from simd/jsimd_arm64_neon.S
rename to simd/arm64/jsimd_neon.S
diff --git a/simd/gas-preprocessor.in b/simd/gas-preprocessor.in
new file mode 100755
index 0000000..560f788
--- /dev/null
+++ b/simd/gas-preprocessor.in
@@ -0,0 +1 @@
+gas-preprocessor.pl @CMAKE_ASM_COMPILER@ ${1+"$@"}
diff --git a/simd/i386/jccolext-avx2.asm b/simd/i386/jccolext-avx2.asm
new file mode 100644
index 0000000..19e05c0
--- /dev/null
+++ b/simd/i386/jccolext-avx2.asm
@@ -0,0 +1,581 @@
+;
+; jccolext.asm - colorspace conversion (AVX2)
+;
+; Copyright (C) 2015, Intel Corporation.
+; Copyright (C) 2016, D. R. Commander.
+;
+; Based on the x86 SIMD extension for IJG JPEG library
+; Copyright (C) 1999-2006, MIYASAKA Masaru.
+; For conditions of distribution and use, see copyright notice in jsimdext.inc
+;
+; This file should be assembled with NASM (Netwide Assembler),
+; can *not* be assembled with Microsoft's MASM or any compatible
+; assembler (including Borland's Turbo Assembler).
+; NASM is available from http://nasm.sourceforge.net/ or
+; http://sourceforge.net/project/showfiles.php?group_id=6208
+;
+; [TAB8]
+
+%include "jcolsamp.inc"
+
+; --------------------------------------------------------------------------
+;
+; Convert some rows of samples to the output colorspace.
+;
+; GLOBAL(void)
+; jsimd_rgb_ycc_convert_avx2 (JDIMENSION img_width,
+; JSAMPARRAY input_buf, JSAMPIMAGE output_buf,
+; JDIMENSION output_row, int num_rows);
+;
+
+%define img_width(b) (b)+8 ; JDIMENSION img_width
+%define input_buf(b) (b)+12 ; JSAMPARRAY input_buf
+%define output_buf(b) (b)+16 ; JSAMPIMAGE output_buf
+%define output_row(b) (b)+20 ; JDIMENSION output_row
+%define num_rows(b) (b)+24 ; int num_rows
+
+%define original_ebp ebp+0
+%define wk(i) ebp-(WK_NUM-(i))*SIZEOF_YMMWORD ; ymmword wk[WK_NUM]
+%define WK_NUM 8
+%define gotptr wk(0)-SIZEOF_POINTER ; void * gotptr
+
+ align 32
+
+ global EXTN(jsimd_rgb_ycc_convert_avx2)
+
+EXTN(jsimd_rgb_ycc_convert_avx2):
+ push ebp
+ mov eax, esp ; eax = original ebp
+ sub esp, byte 4
+ and esp, byte (-SIZEOF_YMMWORD) ; align to 256 bits
+ mov [esp], eax
+ mov ebp, esp ; ebp = aligned ebp
+ lea esp, [wk(0)]
+ pushpic eax ; make a room for GOT address
+ push ebx
+; push ecx ; need not be preserved
+; push edx ; need not be preserved
+ push esi
+ push edi
+
+ get_GOT ebx ; get GOT address
+ movpic POINTER [gotptr], ebx ; save GOT address
+
+ mov ecx, JDIMENSION [img_width(eax)]
+ test ecx, ecx
+ jz near .return
+
+ push ecx
+
+ mov esi, JSAMPIMAGE [output_buf(eax)]
+ mov ecx, JDIMENSION [output_row(eax)]
+ mov edi, JSAMPARRAY [esi+0*SIZEOF_JSAMPARRAY]
+ mov ebx, JSAMPARRAY [esi+1*SIZEOF_JSAMPARRAY]
+ mov edx, JSAMPARRAY [esi+2*SIZEOF_JSAMPARRAY]
+ lea edi, [edi+ecx*SIZEOF_JSAMPROW]
+ lea ebx, [ebx+ecx*SIZEOF_JSAMPROW]
+ lea edx, [edx+ecx*SIZEOF_JSAMPROW]
+
+ pop ecx
+
+ mov esi, JSAMPARRAY [input_buf(eax)]
+ mov eax, INT [num_rows(eax)]
+ test eax, eax
+ jle near .return
+ alignx 16, 7
+.rowloop:
+ pushpic eax
+ push edx
+ push ebx
+ push edi
+ push esi
+ push ecx ; col
+
+ mov esi, JSAMPROW [esi] ; inptr
+ mov edi, JSAMPROW [edi] ; outptr0
+ mov ebx, JSAMPROW [ebx] ; outptr1
+ mov edx, JSAMPROW [edx] ; outptr2
+ movpic eax, POINTER [gotptr] ; load GOT address (eax)
+
+ cmp ecx, byte SIZEOF_YMMWORD
+ jae near .columnloop
+ alignx 16, 7
+
+%if RGB_PIXELSIZE == 3 ; ---------------
+
+.column_ld1:
+ push eax
+ push edx
+ lea ecx, [ecx+ecx*2] ; imul ecx,RGB_PIXELSIZE
+ test cl, SIZEOF_BYTE
+ jz short .column_ld2
+ sub ecx, byte SIZEOF_BYTE
+ movzx eax, BYTE [esi+ecx]
+.column_ld2:
+ test cl, SIZEOF_WORD
+ jz short .column_ld4
+ sub ecx, byte SIZEOF_WORD
+ movzx edx, WORD [esi+ecx]
+ shl eax, WORD_BIT
+ or eax, edx
+.column_ld4:
+ vmovd xmmA, eax
+ pop edx
+ pop eax
+ test cl, SIZEOF_DWORD
+ jz short .column_ld8
+ sub ecx, byte SIZEOF_DWORD
+ vmovd xmmF, XMM_DWORD [esi+ecx]
+ vpslldq xmmA, xmmA, SIZEOF_DWORD
+ vpor xmmA, xmmA, xmmF
+.column_ld8:
+ test cl, SIZEOF_MMWORD
+ jz short .column_ld16
+ sub ecx, byte SIZEOF_MMWORD
+ vmovq xmmB, XMM_MMWORD [esi+ecx]
+ vpslldq xmmA, xmmA, SIZEOF_MMWORD
+ vpor xmmA, xmmA, xmmB
+.column_ld16:
+ test cl, SIZEOF_XMMWORD
+ jz short .column_ld32
+ sub ecx, byte SIZEOF_XMMWORD
+ vmovdqu xmmB, XMM_MMWORD [esi+ecx]
+ vperm2i128 ymmA, ymmA, ymmA, 1
+ vpor ymmA, ymmB
+.column_ld32:
+ test cl, SIZEOF_YMMWORD
+ jz short .column_ld64
+ sub ecx, byte SIZEOF_YMMWORD
+ vmovdqa ymmF, ymmA
+ vmovdqu ymmA, YMMWORD [esi+0*SIZEOF_YMMWORD]
+.column_ld64:
+ test cl, 2*SIZEOF_YMMWORD
+ mov ecx, SIZEOF_YMMWORD
+ jz short .rgb_ycc_cnv
+ vmovdqa ymmB, ymmA
+ vmovdqu ymmA, YMMWORD [esi+0*SIZEOF_YMMWORD]
+ vmovdqu ymmF, YMMWORD [esi+1*SIZEOF_YMMWORD]
+ jmp short .rgb_ycc_cnv
+ alignx 16, 7
+
+.columnloop:
+ vmovdqu ymmA, YMMWORD [esi+0*SIZEOF_YMMWORD]
+ vmovdqu ymmF, YMMWORD [esi+1*SIZEOF_YMMWORD]
+ vmovdqu ymmB, YMMWORD [esi+2*SIZEOF_YMMWORD]
+
+.rgb_ycc_cnv:
+ ; ymmA=(00 10 20 01 11 21 02 12 22 03 13 23 04 14 24 05
+ ; 15 25 06 16 26 07 17 27 08 18 28 09 19 29 0A 1A)
+ ; ymmF=(2A 0B 1B 2B 0C 1C 2C 0D 1D 2D 0E 1E 2E 0F 1F 2F
+ ; 0G 1G 2G 0H 1H 2H 0I 1I 2I 0J 1J 2J 0K 1K 2K 0L)
+ ; ymmB=(1L 2L 0M 1M 2M 0N 1N 2N 0O 1O 2O 0P 1P 2P 0Q 1Q
+ ; 2Q 0R 1R 2R 0S 1S 2S 0T 1T 2T 0U 1U 2U 0V 1V 2V)
+
+ vmovdqu ymmC, ymmA
+ vinserti128 ymmA, ymmF, xmmA, 0 ; ymmA=(00 10 20 01 11 21 02 12 22 03 13 23 04 14 24 05
+ ; 0G 1G 2G 0H 1H 2H 0I 1I 2I 0J 1J 2J 0K 1K 2K 0L)
+ vinserti128 ymmC, ymmC, xmmB, 0 ; ymmC=(1L 2L 0M 1M 2M 0N 1N 2N 0O 1O 2O 0P 1P 2P 0Q 1Q
+ ; 15 25 06 16 26 07 17 27 08 18 28 09 19 29 0A 1A)
+ vinserti128 ymmB, ymmB, xmmF, 0 ; ymmB=(2A 0B 1B 2B 0C 1C 2C 0D 1D 2D 0E 1E 2E 0F 1F 2F
+ ; 2Q 0R 1R 2R 0S 1S 2S 0T 1T 2T 0U 1U 2U 0V 1V 2V)
+ vperm2i128 ymmF, ymmC, ymmC, 1 ; ymmF=(15 25 06 16 26 07 17 27 08 18 28 09 19 29 0A 1A
+ ; 1L 2L 0M 1M 2M 0N 1N 2N 0O 1O 2O 0P 1P 2P 0Q 1Q)
+
+ vmovdqa ymmG, ymmA
+ vpslldq ymmA, ymmA, 8 ; ymmA=(-- -- -- -- -- -- -- -- 00 10 20 01 11 21 02 12
+ ; 22 03 13 23 04 14 24 05 0G 1G 2G 0H 1H 2H 0I 1I)
+ vpsrldq ymmG, ymmG, 8 ; ymmG=(22 03 13 23 04 14 24 05 0G 1G 2G 0H 1H 2H 0I 1I
+ ; 2I 0J 1J 2J 0K 1K 2K 0L -- -- -- -- -- -- -- --)
+
+ vpunpckhbw ymmA, ymmA, ymmF ; ymmA=(00 08 10 18 20 28 01 09 11 19 21 29 02 0A 12 1A
+ ; 0G 0O 1G 1O 2G 2O 0H 0P 1H 1P 2H 2P 0I 0Q 1I 1Q)
+ vpslldq ymmF, ymmF, 8 ; ymmF=(-- -- -- -- -- -- -- -- 15 25 06 16 26 07 17 27
+ ; 08 18 28 09 19 29 0A 1A 1L 2L 0M 1M 2M 0N 1N 2N)
+
+ vpunpcklbw ymmG, ymmG, ymmB ; ymmG=(22 2A 03 0B 13 1B 23 2B 04 0C 14 1C 24 2C 05 0D
+ ; 2I 2Q 0J 0R 1J 1R 2J 2R 0K 0S 1K 1S 2K 2S 0L 0T)
+ vpunpckhbw ymmF, ymmF, ymmB ; ymmF=(15 1D 25 2D 06 0E 16 1E 26 2E 07 0F 17 1F 27 2F
+ ; 1L 1T 2L 2T 0M 0U 1M 1U 2M 2U 0N 0V 1N 1V 2N 2V)
+
+ vmovdqa ymmD, ymmA
+ vpslldq ymmA, ymmA, 8 ; ymmA=(-- -- -- -- -- -- -- -- 00 08 10 18 20 28 01 09
+ ; 11 19 21 29 02 0A 12 1A 0G 0O 1G 1O 2G 2O 0H 0P)
+ vpsrldq ymmD, ymmD, 8 ; ymmD=(11 19 21 29 02 0A 12 1A 0G 0O 1G 1O 2G 2O 0H 0P
+ ; 1H 1P 2H 2P 0I 0Q 1I 1Q -- -- -- -- -- -- -- --)
+
+ vpunpckhbw ymmA, ymmA, ymmG ; ymmA=(00 04 08 0C 10 14 18 1C 20 24 28 2C 01 05 09 0D
+ ; 0G 0K 0O 0S 1G 1K 1O 1S 2G 2K 2O 2S 0H 0L 0P 0T)
+ vpslldq ymmG, ymmG, 8 ; ymmG=(-- -- -- -- -- -- -- -- 22 2A 03 0B 13 1B 23 2B
+ ; 04 0C 14 1C 24 2C 05 0D 2I 2Q 0J 0R 1J 1R 2J 2R)
+
+ vpunpcklbw ymmD, ymmD, ymmF ; ymmD=(11 15 19 1D 21 25 29 2D 02 06 0A 0E 12 16 1A 1E
+ ; 1H 1L 1P 1T 2H 2L 2P 2T 0I 0M 0Q 0U 1I 1M 1Q 1U)
+ vpunpckhbw ymmG, ymmG, ymmF ; ymmG=(22 26 2A 2E 03 07 0B 0F 13 17 1B 1F 23 27 2B 2F
+ ; 2I 2M 2Q 2U 0J 0N 0R 0V 1J 1N 1R 1V 2J 2N 2R 2V)
+
+ vmovdqa ymmE, ymmA
+ vpslldq ymmA, ymmA, 8 ; ymmA=(-- -- -- -- -- -- -- -- 00 04 08 0C 10 14 18 1C
+ ; 20 24 28 2C 01 05 09 0D 0G 0K 0O 0S 1G 1K 1O 1S)
+ vpsrldq ymmE, ymmE, 8 ; ymmE=(20 24 28 2C 01 05 09 0D 0G 0K 0O 0S 1G 1K 1O 1S
+ ; 2G 2K 2O 2S 0H 0L 0P 0T -- -- -- -- -- -- -- --)
+
+ vpunpckhbw ymmA, ymmA, ymmD ; ymmA=(00 02 04 06 08 0A 0C 0E 10 12 14 16 18 1A 1C 1E
+ ; 0G 0I 0K 0M 0O 0Q 0S 0U 1G 1I 1K 1M 1O 1Q 1S 1U)
+ vpslldq ymmD, ymmD, 8 ; ymmD=(-- -- -- -- -- -- -- -- 11 15 19 1D 21 25 29 2D
+ ; 02 06 0A 0E 12 16 1A 1E 1H 1L 1P 1T 2H 2L 2P 2T)
+
+ vpunpcklbw ymmE, ymmE, ymmG ; ymmE=(20 22 24 26 28 2A 2C 2E 01 03 05 07 09 0B 0D 0F
+ ; 2G 2I 2K 2M 2O 2Q 2S 2U 0H 0J 0L 0N 0P 0R 0T 0V)
+ vpunpckhbw ymmD, ymmD, ymmG ; ymmD=(11 13 15 17 19 1B 1D 1F 21 23 25 27 29 2B 2D 2F
+ ; 1H 1J 1L 1N 1P 1R 1T 1V 2H 2J 2L 2N 2P 2R 2T 2V)
+
+ vpxor ymmH, ymmH, ymmH
+
+ vmovdqa ymmC, ymmA
+ vpunpcklbw ymmA, ymmA, ymmH ; ymmA=(00 02 04 06 08 0A 0C 0E 0G 0I 0K 0M 0O 0Q 0S 0U)
+ vpunpckhbw ymmC, ymmC, ymmH ; ymmC=(10 12 14 16 18 1A 1C 1E 1G 1I 1K 1M 1O 1Q 1S 1U)
+
+ vmovdqa ymmB, ymmE
+ vpunpcklbw ymmE, ymmE, ymmH ; ymmE=(20 22 24 26 28 2A 2C 2E 2G 2I 2K 2M 2O 2Q 2S 2U)
+ vpunpckhbw ymmB, ymmB, ymmH ; ymmB=(01 03 05 07 09 0B 0D 0F 0H 0J 0L 0N 0P 0R 0T 0V)
+
+ vmovdqa ymmF, ymmD
+ vpunpcklbw ymmD, ymmD, ymmH ; ymmD=(11 13 15 17 19 1B 1D 1F 1H 1J 1L 1N 1P 1R 1T 1V)
+ vpunpckhbw ymmF, ymmF, ymmH ; ymmF=(21 23 25 27 29 2B 2D 2F 2H 2J 2L 2N 2P 2R 2T 2V)
+
+%else ; RGB_PIXELSIZE == 4 ; -----------
+
+.column_ld1:
+ test cl, SIZEOF_XMMWORD/16
+ jz short .column_ld2
+ sub ecx, byte SIZEOF_XMMWORD/16
+ vmovd xmmA, XMM_DWORD [esi+ecx*RGB_PIXELSIZE]
+.column_ld2:
+ test cl, SIZEOF_XMMWORD/8
+ jz short .column_ld4
+ sub ecx, byte SIZEOF_XMMWORD/8
+ vmovq xmmF, XMM_MMWORD [esi+ecx*RGB_PIXELSIZE]
+ vpslldq xmmA, xmmA, SIZEOF_MMWORD
+ vpor xmmA, xmmA, xmmF
+.column_ld4:
+ test cl, SIZEOF_XMMWORD/4
+ jz short .column_ld8
+ sub ecx, byte SIZEOF_XMMWORD/4
+ vmovdqa xmmF, xmmA
+ vperm2i128 ymmF, ymmF, ymmF, 1
+ vmovdqu xmmA, XMMWORD [esi+ecx*RGB_PIXELSIZE]
+ vpor ymmA, ymmA, ymmF
+.column_ld8:
+ test cl, SIZEOF_XMMWORD/2
+ jz short .column_ld16
+ sub ecx, byte SIZEOF_XMMWORD/2
+ vmovdqa ymmF, ymmA
+ vmovdqu ymmA, YMMWORD [esi+ecx*RGB_PIXELSIZE]
+.column_ld16:
+ test cl, SIZEOF_XMMWORD
+ mov ecx, SIZEOF_YMMWORD
+ jz short .rgb_ycc_cnv
+ vmovdqa ymmE, ymmA
+ vmovdqa ymmH, ymmF
+ vmovdqu ymmA, YMMWORD [esi+0*SIZEOF_YMMWORD]
+ vmovdqu ymmF, YMMWORD [esi+1*SIZEOF_YMMWORD]
+ jmp short .rgb_ycc_cnv
+ alignx 16, 7
+
+.columnloop:
+ vmovdqu ymmA, YMMWORD [esi+0*SIZEOF_YMMWORD]
+ vmovdqu ymmF, YMMWORD [esi+1*SIZEOF_YMMWORD]
+ vmovdqu ymmE, YMMWORD [esi+2*SIZEOF_YMMWORD]
+ vmovdqu ymmH, YMMWORD [esi+3*SIZEOF_YMMWORD]
+
+.rgb_ycc_cnv:
+ ; ymmA=(00 10 20 30 01 11 21 31 02 12 22 32 03 13 23 33
+ ; 04 14 24 34 05 15 25 35 06 16 26 36 07 17 27 37)
+ ; ymmF=(08 18 28 38 09 19 29 39 0A 1A 2A 3A 0B 1B 2B 3B
+ ; 0C 1C 2C 3C 0D 1D 2D 3D 0E 1E 2E 3E 0F 1F 2F 3F)
+ ; ymmE=(0G 1G 2G 3G 0H 1H 2H 3H 0I 1I 2I 3I 0J 1J 2J 3J
+ ; 0K 1K 2K 3K 0L 1L 2L 3L 0M 1M 2M 3M 0N 1N 2N 3N)
+ ; ymmH=(0O 1O 2O 3O 0P 1P 2P 3P 0Q 1Q 2Q 3Q 0R 1R 2R 3R
+ ; 0S 1S 2S 3S 0T 1T 2T 3T 0U 1U 2U 3U 0V 1V 2V 3V)
+
+ vmovdqa ymmB, ymmA
+ vinserti128 ymmA, ymmA, xmmE, 1 ; ymmA=(00 10 20 30 01 11 21 31 02 12 22 32 03 13 23 33
+ ; 0G 1G 2G 3G 0H 1H 2H 3H 0I 1I 2I 3I 0J 1J 2J 3J)
+ vperm2i128 ymmE, ymmB, ymmE, 0x31 ; ymmE=(04 14 24 34 05 15 25 35 06 16 26 36 07 17 27 37
+ ; 0K 1K 2K 3K 0L 1L 2L 3L 0M 1M 2M 3M 0N 1N 2N 3N)
+
+ vmovdqa ymmB, ymmF
+ vinserti128 ymmF, ymmF, xmmH, 1 ; ymmF=(08 18 28 38 09 19 29 39 0A 1A 2A 3A 0B 1B 2B 3B
+ ; 0O 1O 2O 3O 0P 1P 2P 3P 0Q 1Q 2Q 3Q 0R 1R 2R 3R)
+ vperm2i128 ymmH, ymmB, ymmH, 0x31 ; ymmH=(0C 1C 2C 3C 0D 1D 2D 3D 0E 1E 2E 3E 0F 1F 2F 3F
+ ; 0S 1S 2S 3S 0T 1T 2T 3T 0U 1U 2U 3U 0V 1V 2V 3V)
+
+ vmovdqa ymmD, ymmA
+ vpunpcklbw ymmA, ymmA, ymmE ; ymmA=(00 04 10 14 20 24 30 34 01 05 11 15 21 25 31 35
+ ; 0G 0K 1G 1K 2G 2K 3G 3K 0H 0L 1H 1L 2H 2L 3H 3L)
+ vpunpckhbw ymmD, ymmD, ymmE ; ymmD=(02 06 12 16 22 26 32 36 03 07 13 17 23 27 33 37
+ ; 0I 0M 1I 1M 2I 2M 3I 3M 0J 0N 1J 1N 2J 2N 3J 3N)
+
+ vmovdqa ymmC, ymmF
+ vpunpcklbw ymmF, ymmF, ymmH ; ymmF=(08 0C 18 1C 28 2C 38 3C 09 0D 19 1D 29 2D 39 3D
+ ; 0O 0S 1O 1S 2O 2S 3O 3S 0P 0T 1P 1T 2P 2T 3P 3T)
+ vpunpckhbw ymmC, ymmC, ymmH ; ymmC=(0A 0E 1A 1E 2A 2E 3A 3E 0B 0F 1B 1F 2B 2F 3B 3F
+ ; 0Q 0U 1Q 1U 2Q 2U 3Q 3U 0R 0V 1R 1V 2R 2V 3R 3V)
+
+ vmovdqa ymmB, ymmA
+ vpunpcklwd ymmA, ymmA, ymmF ; ymmA=(00 04 08 0C 10 14 18 1C 20 24 28 2C 30 34 38 3C
+ ; 0G 0K 0O 0S 1G 1K 1O 1S 2G 2K 2O 2S 3G 3K 3O 3S)
+ vpunpckhwd ymmB, ymmB, ymmF ; ymmB=(01 05 09 0D 11 15 19 1D 21 25 29 2D 31 35 39 3D
+ ; 0H 0L 0P 0T 1H 1L 1P 1T 2H 2L 2P 2T 3H 3L 3P 3T)
+
+ vmovdqa ymmG, ymmD
+ vpunpcklwd ymmD, ymmD, ymmC ; ymmD=(02 06 0A 0E 12 16 1A 1E 22 26 2A 2E 32 36 3A 3E
+ ; 0I 0M 0Q 0U 1I 1M 1Q 1U 2I 2M 2Q 2U 3I 3M 3Q 3U)
+ vpunpckhwd ymmG, ymmG, ymmC ; ymmG=(03 07 0B 0F 13 17 1B 1F 23 27 2B 2F 33 37 3B 3F
+ ; 0J 0N 0R 0V 1J 1N 1R 1V 2J 2N 2R 2V 3J 3N 3R 3V)
+
+ vmovdqa ymmE, ymmA
+ vpunpcklbw ymmA, ymmA, ymmD ; ymmA=(00 02 04 06 08 0A 0C 0E 10 12 14 16 18 1A 1C 1E
+ ; 0G 0I 0K 0M 0O 0Q 0S 0U 1G 1I 1K 1M 1O 1Q 1S 1U)
+ vpunpckhbw ymmE, ymmE, ymmD ; ymmE=(20 22 24 26 28 2A 2C 2E 30 32 34 36 38 3A 3C 3E
+ ; 2G 2I 2K 2M 2O 2Q 2S 2U 3G 3I 3K 3M 3O 3Q 3S 3U)
+
+ vmovdqa ymmH, ymmB
+ vpunpcklbw ymmB, ymmB, ymmG ; ymmB=(01 03 05 07 09 0B 0D 0F 11 13 15 17 19 1B 1D 1F
+ ; 0H 0J 0L 0N 0P 0R 0T 0V 1H 1J 1L 1N 1P 1R 1T 1V)
+ vpunpckhbw ymmH, ymmH, ymmG ; ymmH=(21 23 25 27 29 2B 2D 2F 31 33 35 37 39 3B 3D 3F
+ ; 2H 2J 2L 2N 2P 2R 2T 2V 3H 3J 3L 3N 3P 3R 3T 3V)
+
+ vpxor ymmF, ymmF, ymmF
+
+ vmovdqa ymmC, ymmA
+ vpunpcklbw ymmA, ymmA, ymmF ; ymmA=(00 02 04 06 08 0A 0C 0E 0G 0I 0K 0M 0O 0Q 0S 0U)
+ vpunpckhbw ymmC, ymmC, ymmF ; ymmC=(10 12 14 16 18 1A 1C 1E 1G 1I 1K 1M 1O 1Q 1S 1U)
+
+ vmovdqa ymmD, ymmB
+ vpunpcklbw ymmB, ymmB, ymmF ; ymmB=(01 03 05 07 09 0B 0D 0F 0H 0J 0L 0N 0P 0R 0T 0V)
+ vpunpckhbw ymmD, ymmD, ymmF ; ymmD=(11 13 15 17 19 1B 1D 1F 1H 1J 1L 1N 1P 1R 1T 1V)
+
+ vmovdqa ymmG, ymmE
+ vpunpcklbw ymmE, ymmE, ymmF ; ymmE=(20 22 24 26 28 2A 2C 2E 2G 2I 2K 2M 2O 2Q 2S 2U)
+ vpunpckhbw ymmG, ymmG, ymmF ; ymmG=(30 32 34 36 38 3A 3C 3E 3G 3I 3K 3M 3O 3Q 3S 3U)
+
+ vpunpcklbw ymmF, ymmF, ymmH
+ vpunpckhbw ymmH, ymmH, ymmH
+ vpsrlw ymmF, ymmF, BYTE_BIT ; ymmF=(21 23 25 27 29 2B 2D 2F 2H 2J 2L 2N 2P 2R 2T 2V)
+ vpsrlw ymmH, ymmH, BYTE_BIT ; ymmH=(31 33 35 37 39 3B 3D 3F 3H 3J 3L 3N 3P 3R 3T 3V)
+
+%endif ; RGB_PIXELSIZE ; ---------------
+
+ ; ymm0=R(02468ACEGIKMOQSU)=RE, ymm2=G(02468ACEGIKMOQSU)=GE, ymm4=B(02468ACEGIKMOQSU)=BE
+ ; ymm1=R(13579BDFHJLNPRTV)=RO, ymm3=G(13579BDFHJLNPRTV)=GO, ymm5=B(13579BDFHJLNPRTV)=BO
+
+ ; (Original)
+ ; Y = 0.29900 * R + 0.58700 * G + 0.11400 * B
+ ; Cb = -0.16874 * R - 0.33126 * G + 0.50000 * B + CENTERJSAMPLE
+ ; Cr = 0.50000 * R - 0.41869 * G - 0.08131 * B + CENTERJSAMPLE
+ ;
+ ; (This implementation)
+ ; Y = 0.29900 * R + 0.33700 * G + 0.11400 * B + 0.25000 * G
+ ; Cb = -0.16874 * R - 0.33126 * G + 0.50000 * B + CENTERJSAMPLE
+ ; Cr = 0.50000 * R - 0.41869 * G - 0.08131 * B + CENTERJSAMPLE
+
+ vmovdqa YMMWORD [wk(0)], ymm0 ; wk(0)=RE
+ vmovdqa YMMWORD [wk(1)], ymm1 ; wk(1)=RO
+ vmovdqa YMMWORD [wk(2)], ymm4 ; wk(2)=BE
+ vmovdqa YMMWORD [wk(3)], ymm5 ; wk(3)=BO
+
+ vmovdqa ymm6, ymm1
+ vpunpcklwd ymm1, ymm1, ymm3
+ vpunpckhwd ymm6, ymm6, ymm3
+ vmovdqa ymm7, ymm1
+ vmovdqa ymm4, ymm6
+ vpmaddwd ymm1, ymm1, [GOTOFF(eax,PW_F0299_F0337)] ; ymm1=ROL*FIX(0.299)+GOL*FIX(0.337)
+ vpmaddwd ymm6, ymm6, [GOTOFF(eax,PW_F0299_F0337)] ; ymm6=ROH*FIX(0.299)+GOH*FIX(0.337)
+ vpmaddwd ymm7, ymm7, [GOTOFF(eax,PW_MF016_MF033)] ; ymm7=ROL*-FIX(0.168)+GOL*-FIX(0.331)
+ vpmaddwd ymm4, ymm4, [GOTOFF(eax,PW_MF016_MF033)] ; ymm4=ROH*-FIX(0.168)+GOH*-FIX(0.331)
+
+ vmovdqa YMMWORD [wk(4)], ymm1 ; wk(4)=ROL*FIX(0.299)+GOL*FIX(0.337)
+ vmovdqa YMMWORD [wk(5)], ymm6 ; wk(5)=ROH*FIX(0.299)+GOH*FIX(0.337)
+
+ vpxor ymm1, ymm1, ymm1
+ vpxor ymm6, ymm6, ymm6
+ vpunpcklwd ymm1, ymm1, ymm5 ; ymm1=BOL
+ vpunpckhwd ymm6, ymm6, ymm5 ; ymm6=BOH
+ vpsrld ymm1, ymm1, 1 ; ymm1=BOL*FIX(0.500)
+ vpsrld ymm6, ymm6, 1 ; ymm6=BOH*FIX(0.500)
+
+ vmovdqa ymm5, [GOTOFF(eax,PD_ONEHALFM1_CJ)] ; ymm5=[PD_ONEHALFM1_CJ]
+
+ vpaddd ymm7, ymm7, ymm1
+ vpaddd ymm4, ymm4, ymm6
+ vpaddd ymm7, ymm7, ymm5
+ vpaddd ymm4, ymm4, ymm5
+ vpsrld ymm7, ymm7, SCALEBITS ; ymm7=CbOL
+ vpsrld ymm4, ymm4, SCALEBITS ; ymm4=CbOH
+ vpackssdw ymm7, ymm7, ymm4 ; ymm7=CbO
+
+ vmovdqa ymm1, YMMWORD [wk(2)] ; ymm1=BE
+
+ vmovdqa ymm6, ymm0
+ vpunpcklwd ymm0, ymm0, ymm2
+ vpunpckhwd ymm6, ymm6, ymm2
+ vmovdqa ymm5, ymm0
+ vmovdqa ymm4, ymm6
+ vpmaddwd ymm0, ymm0, [GOTOFF(eax,PW_F0299_F0337)] ; ymm0=REL*FIX(0.299)+GEL*FIX(0.337)
+ vpmaddwd ymm6, ymm6, [GOTOFF(eax,PW_F0299_F0337)] ; ymm6=REH*FIX(0.299)+GEH*FIX(0.337)
+ vpmaddwd ymm5, ymm5, [GOTOFF(eax,PW_MF016_MF033)] ; ymm5=REL*-FIX(0.168)+GEL*-FIX(0.331)
+ vpmaddwd ymm4, ymm4, [GOTOFF(eax,PW_MF016_MF033)] ; ymm4=REH*-FIX(0.168)+GEH*-FIX(0.331)
+
+ vmovdqa YMMWORD [wk(6)], ymm0 ; wk(6)=REL*FIX(0.299)+GEL*FIX(0.337)
+ vmovdqa YMMWORD [wk(7)], ymm6 ; wk(7)=REH*FIX(0.299)+GEH*FIX(0.337)
+
+ vpxor ymm0, ymm0, ymm0
+ vpxor ymm6, ymm6, ymm6
+ vpunpcklwd ymm0, ymm0, ymm1 ; ymm0=BEL
+ vpunpckhwd ymm6, ymm6, ymm1 ; ymm6=BEH
+ vpsrld ymm0, ymm0, 1 ; ymm0=BEL*FIX(0.500)
+ vpsrld ymm6, ymm6, 1 ; ymm6=BEH*FIX(0.500)
+
+
+ vmovdqa ymm1, [GOTOFF(eax,PD_ONEHALFM1_CJ)] ; ymm1=[PD_ONEHALFM1_CJ]
+
+ vpaddd ymm5, ymm5, ymm0
+ vpaddd ymm4, ymm4, ymm6
+ vpaddd ymm5, ymm5, ymm1
+ vpaddd ymm4, ymm4, ymm1
+ vpsrld ymm5, ymm5, SCALEBITS ; ymm5=CbEL
+ vpsrld ymm4, ymm4, SCALEBITS ; ymm4=CbEH
+ vpackssdw ymm5, ymm5, ymm4 ; ymm5=CbE
+
+ vpsllw ymm7, ymm7, BYTE_BIT
+ vpor ymm5, ymm5, ymm7 ; ymm5=Cb
+ vmovdqu YMMWORD [ebx], ymm5 ; Save Cb
+
+ vmovdqa ymm0, YMMWORD [wk(3)] ; ymm0=BO
+ vmovdqa ymm6, YMMWORD [wk(2)] ; ymm6=BE
+ vmovdqa ymm1, YMMWORD [wk(1)] ; ymm1=RO
+
+ vmovdqa ymm4, ymm0
+ vpunpcklwd ymm0, ymm0, ymm3
+ vpunpckhwd ymm4, ymm4, ymm3
+ vmovdqa ymm7, ymm0
+ vmovdqa ymm5, ymm4
+ vpmaddwd ymm0, ymm0, [GOTOFF(eax,PW_F0114_F0250)] ; ymm0=BOL*FIX(0.114)+GOL*FIX(0.250)
+ vpmaddwd ymm4, ymm4, [GOTOFF(eax,PW_F0114_F0250)] ; ymm4=BOH*FIX(0.114)+GOH*FIX(0.250)
+ vpmaddwd ymm7, ymm7, [GOTOFF(eax,PW_MF008_MF041)] ; ymm7=BOL*-FIX(0.081)+GOL*-FIX(0.418)
+ vpmaddwd ymm5, ymm5, [GOTOFF(eax,PW_MF008_MF041)] ; ymm5=BOH*-FIX(0.081)+GOH*-FIX(0.418)
+
+ vmovdqa ymm3, [GOTOFF(eax,PD_ONEHALF)] ; ymm3=[PD_ONEHALF]
+
+ vpaddd ymm0, ymm0, YMMWORD [wk(4)]
+ vpaddd ymm4, ymm4, YMMWORD [wk(5)]
+ vpaddd ymm0, ymm0, ymm3
+ vpaddd ymm4, ymm4, ymm3
+ vpsrld ymm0, ymm0, SCALEBITS ; ymm0=YOL
+ vpsrld ymm4, ymm4, SCALEBITS ; ymm4=YOH
+ vpackssdw ymm0, ymm0, ymm4 ; ymm0=YO
+
+ vpxor ymm3, ymm3, ymm3
+ vpxor ymm4, ymm4, ymm4
+ vpunpcklwd ymm3, ymm3, ymm1 ; ymm3=ROL
+ vpunpckhwd ymm4, ymm4, ymm1 ; ymm4=ROH
+ vpsrld ymm3, ymm3, 1 ; ymm3=ROL*FIX(0.500)
+ vpsrld ymm4, ymm4, 1 ; ymm4=ROH*FIX(0.500)
+
+ vmovdqa ymm1, [GOTOFF(eax,PD_ONEHALFM1_CJ)] ; ymm1=[PD_ONEHALFM1_CJ]
+
+ vpaddd ymm7, ymm7, ymm3
+ vpaddd ymm5, ymm5, ymm4
+ vpaddd ymm7, ymm7, ymm1
+ vpaddd ymm5, ymm5, ymm1
+ vpsrld ymm7, ymm7, SCALEBITS ; ymm7=CrOL
+ vpsrld ymm5, ymm5, SCALEBITS ; ymm5=CrOH
+ vpackssdw ymm7, ymm7, ymm5 ; ymm7=CrO
+
+ vmovdqa ymm3, YMMWORD [wk(0)] ; ymm3=RE
+
+ vmovdqa ymm4, ymm6
+ vpunpcklwd ymm6, ymm6, ymm2
+ vpunpckhwd ymm4, ymm4, ymm2
+ vmovdqa ymm1, ymm6
+ vmovdqa ymm5, ymm4
+ vpmaddwd ymm6, ymm6, [GOTOFF(eax,PW_F0114_F0250)] ; ymm6=BEL*FIX(0.114)+GEL*FIX(0.250)
+ vpmaddwd ymm4, ymm4, [GOTOFF(eax,PW_F0114_F0250)] ; ymm4=BEH*FIX(0.114)+GEH*FIX(0.250)
+ vpmaddwd ymm1, ymm1, [GOTOFF(eax,PW_MF008_MF041)] ; ymm1=BEL*-FIX(0.081)+GEL*-FIX(0.418)
+ vpmaddwd ymm5, ymm5, [GOTOFF(eax,PW_MF008_MF041)] ; ymm5=BEH*-FIX(0.081)+GEH*-FIX(0.418)
+
+ vmovdqa ymm2, [GOTOFF(eax,PD_ONEHALF)] ; ymm2=[PD_ONEHALF]
+
+ vpaddd ymm6, ymm6, YMMWORD [wk(6)]
+ vpaddd ymm4, ymm4, YMMWORD [wk(7)]
+ vpaddd ymm6, ymm6, ymm2
+ vpaddd ymm4, ymm4, ymm2
+ vpsrld ymm6, ymm6, SCALEBITS ; ymm6=YEL
+ vpsrld ymm4, ymm4, SCALEBITS ; ymm4=YEH
+ vpackssdw ymm6, ymm6, ymm4 ; ymm6=YE
+
+ vpsllw ymm0, ymm0, BYTE_BIT
+ vpor ymm6, ymm6, ymm0 ; ymm6=Y
+ vmovdqu YMMWORD [edi], ymm6 ; Save Y
+
+ vpxor ymm2, ymm2, ymm2
+ vpxor ymm4, ymm4, ymm4
+ vpunpcklwd ymm2, ymm2, ymm3 ; ymm2=REL
+ vpunpckhwd ymm4, ymm4, ymm3 ; ymm4=REH
+ vpsrld ymm2, ymm2, 1 ; ymm2=REL*FIX(0.500)
+ vpsrld ymm4, ymm4, 1 ; ymm4=REH*FIX(0.500)
+
+ vmovdqa ymm0, [GOTOFF(eax,PD_ONEHALFM1_CJ)] ; ymm0=[PD_ONEHALFM1_CJ]
+
+ vpaddd ymm1, ymm1, ymm2
+ vpaddd ymm5, ymm5, ymm4
+ vpaddd ymm1, ymm1, ymm0
+ vpaddd ymm5, ymm5, ymm0
+ vpsrld ymm1, ymm1, SCALEBITS ; ymm1=CrEL
+ vpsrld ymm5, ymm5, SCALEBITS ; ymm5=CrEH
+ vpackssdw ymm1, ymm1, ymm5 ; ymm1=CrE
+
+ vpsllw ymm7, ymm7, BYTE_BIT
+ vpor ymm1, ymm1, ymm7 ; ymm1=Cr
+ vmovdqu YMMWORD [edx], ymm1 ; Save Cr
+
+ sub ecx, byte SIZEOF_YMMWORD
+ add esi, RGB_PIXELSIZE*SIZEOF_YMMWORD ; inptr
+ add edi, byte SIZEOF_YMMWORD ; outptr0
+ add ebx, byte SIZEOF_YMMWORD ; outptr1
+ add edx, byte SIZEOF_YMMWORD ; outptr2
+ cmp ecx, byte SIZEOF_YMMWORD
+ jae near .columnloop
+ test ecx, ecx
+ jnz near .column_ld1
+
+ pop ecx ; col
+ pop esi
+ pop edi
+ pop ebx
+ pop edx
+ poppic eax
+
+ add esi, byte SIZEOF_JSAMPROW ; input_buf
+ add edi, byte SIZEOF_JSAMPROW
+ add ebx, byte SIZEOF_JSAMPROW
+ add edx, byte SIZEOF_JSAMPROW
+ dec eax ; num_rows
+ jg near .rowloop
+
+.return:
+ vzeroupper
+ pop edi
+ pop esi
+; pop edx ; need not be preserved
+; pop ecx ; need not be preserved
+ pop ebx
+ mov esp, ebp ; esp <- aligned ebp
+ pop esp ; esp <- original ebp
+ pop ebp
+ ret
+
+; For some reason, the OS X linker does not honor the request to align the
+; segment unless we do this.
+ align 32
diff --git a/simd/jccolext-mmx.asm b/simd/i386/jccolext-mmx.asm
similarity index 99%
rename from simd/jccolext-mmx.asm
rename to simd/i386/jccolext-mmx.asm
index 96a0372..4748b20 100644
--- a/simd/jccolext-mmx.asm
+++ b/simd/i386/jccolext-mmx.asm
@@ -2,6 +2,7 @@
; jccolext.asm - colorspace conversion (MMX)
;
; Copyright 2009 Pierre Ossman <ossman@cendio.se> for Cendio AB
+; Copyright (C) 2016, D. R. Commander.
;
; Based on the x86 SIMD extension for IJG JPEG library
; Copyright (C) 1999-2006, MIYASAKA Masaru.
@@ -38,7 +39,7 @@
%define WK_NUM 8
%define gotptr wk(0)-SIZEOF_POINTER ; void * gotptr
- align 16
+ align 32
global EXTN(jsimd_rgb_ycc_convert_mmx)
EXTN(jsimd_rgb_ycc_convert_mmx):
@@ -473,4 +474,4 @@
; For some reason, the OS X linker does not honor the request to align the
; segment unless we do this.
- align 16
+ align 32
diff --git a/simd/i386/jccolext-sse2.asm b/simd/i386/jccolext-sse2.asm
new file mode 100644
index 0000000..611d804
--- /dev/null
+++ b/simd/i386/jccolext-sse2.asm
@@ -0,0 +1,505 @@
+;
+; jccolext.asm - colorspace conversion (SSE2)
+;
+; Copyright (C) 2016, D. R. Commander.
+;
+; Based on the x86 SIMD extension for IJG JPEG library
+; Copyright (C) 1999-2006, MIYASAKA Masaru.
+; For conditions of distribution and use, see copyright notice in jsimdext.inc
+;
+; This file should be assembled with NASM (Netwide Assembler),
+; can *not* be assembled with Microsoft's MASM or any compatible
+; assembler (including Borland's Turbo Assembler).
+; NASM is available from http://nasm.sourceforge.net/ or
+; http://sourceforge.net/project/showfiles.php?group_id=6208
+;
+; [TAB8]
+
+%include "jcolsamp.inc"
+
+; --------------------------------------------------------------------------
+;
+; Convert some rows of samples to the output colorspace.
+;
+; GLOBAL(void)
+; jsimd_rgb_ycc_convert_sse2 (JDIMENSION img_width,
+; JSAMPARRAY input_buf, JSAMPIMAGE output_buf,
+; JDIMENSION output_row, int num_rows);
+;
+
+%define img_width(b) (b)+8 ; JDIMENSION img_width
+%define input_buf(b) (b)+12 ; JSAMPARRAY input_buf
+%define output_buf(b) (b)+16 ; JSAMPIMAGE output_buf
+%define output_row(b) (b)+20 ; JDIMENSION output_row
+%define num_rows(b) (b)+24 ; int num_rows
+
+%define original_ebp ebp+0
+%define wk(i) ebp-(WK_NUM-(i))*SIZEOF_XMMWORD ; xmmword wk[WK_NUM]
+%define WK_NUM 8
+%define gotptr wk(0)-SIZEOF_POINTER ; void * gotptr
+
+ align 32
+
+ global EXTN(jsimd_rgb_ycc_convert_sse2)
+
+EXTN(jsimd_rgb_ycc_convert_sse2):
+ push ebp
+ mov eax, esp ; eax = original ebp
+ sub esp, byte 4
+ and esp, byte (-SIZEOF_XMMWORD) ; align to 128 bits
+ mov [esp], eax
+ mov ebp, esp ; ebp = aligned ebp
+ lea esp, [wk(0)]
+ pushpic eax ; make a room for GOT address
+ push ebx
+; push ecx ; need not be preserved
+; push edx ; need not be preserved
+ push esi
+ push edi
+
+ get_GOT ebx ; get GOT address
+ movpic POINTER [gotptr], ebx ; save GOT address
+
+ mov ecx, JDIMENSION [img_width(eax)]
+ test ecx, ecx
+ jz near .return
+
+ push ecx
+
+ mov esi, JSAMPIMAGE [output_buf(eax)]
+ mov ecx, JDIMENSION [output_row(eax)]
+ mov edi, JSAMPARRAY [esi+0*SIZEOF_JSAMPARRAY]
+ mov ebx, JSAMPARRAY [esi+1*SIZEOF_JSAMPARRAY]
+ mov edx, JSAMPARRAY [esi+2*SIZEOF_JSAMPARRAY]
+ lea edi, [edi+ecx*SIZEOF_JSAMPROW]
+ lea ebx, [ebx+ecx*SIZEOF_JSAMPROW]
+ lea edx, [edx+ecx*SIZEOF_JSAMPROW]
+
+ pop ecx
+
+ mov esi, JSAMPARRAY [input_buf(eax)]
+ mov eax, INT [num_rows(eax)]
+ test eax, eax
+ jle near .return
+ alignx 16, 7
+.rowloop:
+ pushpic eax
+ push edx
+ push ebx
+ push edi
+ push esi
+ push ecx ; col
+
+ mov esi, JSAMPROW [esi] ; inptr
+ mov edi, JSAMPROW [edi] ; outptr0
+ mov ebx, JSAMPROW [ebx] ; outptr1
+ mov edx, JSAMPROW [edx] ; outptr2
+ movpic eax, POINTER [gotptr] ; load GOT address (eax)
+
+ cmp ecx, byte SIZEOF_XMMWORD
+ jae near .columnloop
+ alignx 16, 7
+
+%if RGB_PIXELSIZE == 3 ; ---------------
+
+.column_ld1:
+ push eax
+ push edx
+ lea ecx, [ecx+ecx*2] ; imul ecx,RGB_PIXELSIZE
+ test cl, SIZEOF_BYTE
+ jz short .column_ld2
+ sub ecx, byte SIZEOF_BYTE
+ movzx eax, BYTE [esi+ecx]
+.column_ld2:
+ test cl, SIZEOF_WORD
+ jz short .column_ld4
+ sub ecx, byte SIZEOF_WORD
+ movzx edx, WORD [esi+ecx]
+ shl eax, WORD_BIT
+ or eax, edx
+.column_ld4:
+ movd xmmA, eax
+ pop edx
+ pop eax
+ test cl, SIZEOF_DWORD
+ jz short .column_ld8
+ sub ecx, byte SIZEOF_DWORD
+ movd xmmF, XMM_DWORD [esi+ecx]
+ pslldq xmmA, SIZEOF_DWORD
+ por xmmA, xmmF
+.column_ld8:
+ test cl, SIZEOF_MMWORD
+ jz short .column_ld16
+ sub ecx, byte SIZEOF_MMWORD
+ movq xmmB, XMM_MMWORD [esi+ecx]
+ pslldq xmmA, SIZEOF_MMWORD
+ por xmmA, xmmB
+.column_ld16:
+ test cl, SIZEOF_XMMWORD
+ jz short .column_ld32
+ movdqa xmmF, xmmA
+ movdqu xmmA, XMMWORD [esi+0*SIZEOF_XMMWORD]
+ mov ecx, SIZEOF_XMMWORD
+ jmp short .rgb_ycc_cnv
+.column_ld32:
+ test cl, 2*SIZEOF_XMMWORD
+ mov ecx, SIZEOF_XMMWORD
+ jz short .rgb_ycc_cnv
+ movdqa xmmB, xmmA
+ movdqu xmmA, XMMWORD [esi+0*SIZEOF_XMMWORD]
+ movdqu xmmF, XMMWORD [esi+1*SIZEOF_XMMWORD]
+ jmp short .rgb_ycc_cnv
+ alignx 16, 7
+
+.columnloop:
+ movdqu xmmA, XMMWORD [esi+0*SIZEOF_XMMWORD]
+ movdqu xmmF, XMMWORD [esi+1*SIZEOF_XMMWORD]
+ movdqu xmmB, XMMWORD [esi+2*SIZEOF_XMMWORD]
+
+.rgb_ycc_cnv:
+ ; xmmA=(00 10 20 01 11 21 02 12 22 03 13 23 04 14 24 05)
+ ; xmmF=(15 25 06 16 26 07 17 27 08 18 28 09 19 29 0A 1A)
+ ; xmmB=(2A 0B 1B 2B 0C 1C 2C 0D 1D 2D 0E 1E 2E 0F 1F 2F)
+
+ movdqa xmmG, xmmA
+ pslldq xmmA, 8 ; xmmA=(-- -- -- -- -- -- -- -- 00 10 20 01 11 21 02 12)
+ psrldq xmmG, 8 ; xmmG=(22 03 13 23 04 14 24 05 -- -- -- -- -- -- -- --)
+
+ punpckhbw xmmA, xmmF ; xmmA=(00 08 10 18 20 28 01 09 11 19 21 29 02 0A 12 1A)
+ pslldq xmmF, 8 ; xmmF=(-- -- -- -- -- -- -- -- 15 25 06 16 26 07 17 27)
+
+ punpcklbw xmmG, xmmB ; xmmG=(22 2A 03 0B 13 1B 23 2B 04 0C 14 1C 24 2C 05 0D)
+ punpckhbw xmmF, xmmB ; xmmF=(15 1D 25 2D 06 0E 16 1E 26 2E 07 0F 17 1F 27 2F)
+
+ movdqa xmmD, xmmA
+ pslldq xmmA, 8 ; xmmA=(-- -- -- -- -- -- -- -- 00 08 10 18 20 28 01 09)
+ psrldq xmmD, 8 ; xmmD=(11 19 21 29 02 0A 12 1A -- -- -- -- -- -- -- --)
+
+ punpckhbw xmmA, xmmG ; xmmA=(00 04 08 0C 10 14 18 1C 20 24 28 2C 01 05 09 0D)
+ pslldq xmmG, 8 ; xmmG=(-- -- -- -- -- -- -- -- 22 2A 03 0B 13 1B 23 2B)
+
+ punpcklbw xmmD, xmmF ; xmmD=(11 15 19 1D 21 25 29 2D 02 06 0A 0E 12 16 1A 1E)
+ punpckhbw xmmG, xmmF ; xmmG=(22 26 2A 2E 03 07 0B 0F 13 17 1B 1F 23 27 2B 2F)
+
+ movdqa xmmE, xmmA
+ pslldq xmmA, 8 ; xmmA=(-- -- -- -- -- -- -- -- 00 04 08 0C 10 14 18 1C)
+ psrldq xmmE, 8 ; xmmE=(20 24 28 2C 01 05 09 0D -- -- -- -- -- -- -- --)
+
+ punpckhbw xmmA, xmmD ; xmmA=(00 02 04 06 08 0A 0C 0E 10 12 14 16 18 1A 1C 1E)
+ pslldq xmmD, 8 ; xmmD=(-- -- -- -- -- -- -- -- 11 15 19 1D 21 25 29 2D)
+
+ punpcklbw xmmE, xmmG ; xmmE=(20 22 24 26 28 2A 2C 2E 01 03 05 07 09 0B 0D 0F)
+ punpckhbw xmmD, xmmG ; xmmD=(11 13 15 17 19 1B 1D 1F 21 23 25 27 29 2B 2D 2F)
+
+ pxor xmmH, xmmH
+
+ movdqa xmmC, xmmA
+ punpcklbw xmmA, xmmH ; xmmA=(00 02 04 06 08 0A 0C 0E)
+ punpckhbw xmmC, xmmH ; xmmC=(10 12 14 16 18 1A 1C 1E)
+
+ movdqa xmmB, xmmE
+ punpcklbw xmmE, xmmH ; xmmE=(20 22 24 26 28 2A 2C 2E)
+ punpckhbw xmmB, xmmH ; xmmB=(01 03 05 07 09 0B 0D 0F)
+
+ movdqa xmmF, xmmD
+ punpcklbw xmmD, xmmH ; xmmD=(11 13 15 17 19 1B 1D 1F)
+ punpckhbw xmmF, xmmH ; xmmF=(21 23 25 27 29 2B 2D 2F)
+
+%else ; RGB_PIXELSIZE == 4 ; -----------
+
+.column_ld1:
+ test cl, SIZEOF_XMMWORD/16
+ jz short .column_ld2
+ sub ecx, byte SIZEOF_XMMWORD/16
+ movd xmmA, XMM_DWORD [esi+ecx*RGB_PIXELSIZE]
+.column_ld2:
+ test cl, SIZEOF_XMMWORD/8
+ jz short .column_ld4
+ sub ecx, byte SIZEOF_XMMWORD/8
+ movq xmmE, XMM_MMWORD [esi+ecx*RGB_PIXELSIZE]
+ pslldq xmmA, SIZEOF_MMWORD
+ por xmmA, xmmE
+.column_ld4:
+ test cl, SIZEOF_XMMWORD/4
+ jz short .column_ld8
+ sub ecx, byte SIZEOF_XMMWORD/4
+ movdqa xmmE, xmmA
+ movdqu xmmA, XMMWORD [esi+ecx*RGB_PIXELSIZE]
+.column_ld8:
+ test cl, SIZEOF_XMMWORD/2
+ mov ecx, SIZEOF_XMMWORD
+ jz short .rgb_ycc_cnv
+ movdqa xmmF, xmmA
+ movdqa xmmH, xmmE
+ movdqu xmmA, XMMWORD [esi+0*SIZEOF_XMMWORD]
+ movdqu xmmE, XMMWORD [esi+1*SIZEOF_XMMWORD]
+ jmp short .rgb_ycc_cnv
+ alignx 16, 7
+
+.columnloop:
+ movdqu xmmA, XMMWORD [esi+0*SIZEOF_XMMWORD]
+ movdqu xmmE, XMMWORD [esi+1*SIZEOF_XMMWORD]
+ movdqu xmmF, XMMWORD [esi+2*SIZEOF_XMMWORD]
+ movdqu xmmH, XMMWORD [esi+3*SIZEOF_XMMWORD]
+
+.rgb_ycc_cnv:
+ ; xmmA=(00 10 20 30 01 11 21 31 02 12 22 32 03 13 23 33)
+ ; xmmE=(04 14 24 34 05 15 25 35 06 16 26 36 07 17 27 37)
+ ; xmmF=(08 18 28 38 09 19 29 39 0A 1A 2A 3A 0B 1B 2B 3B)
+ ; xmmH=(0C 1C 2C 3C 0D 1D 2D 3D 0E 1E 2E 3E 0F 1F 2F 3F)
+
+ movdqa xmmD, xmmA
+ punpcklbw xmmA, xmmE ; xmmA=(00 04 10 14 20 24 30 34 01 05 11 15 21 25 31 35)
+ punpckhbw xmmD, xmmE ; xmmD=(02 06 12 16 22 26 32 36 03 07 13 17 23 27 33 37)
+
+ movdqa xmmC, xmmF
+ punpcklbw xmmF, xmmH ; xmmF=(08 0C 18 1C 28 2C 38 3C 09 0D 19 1D 29 2D 39 3D)
+ punpckhbw xmmC, xmmH ; xmmC=(0A 0E 1A 1E 2A 2E 3A 3E 0B 0F 1B 1F 2B 2F 3B 3F)
+
+ movdqa xmmB, xmmA
+ punpcklwd xmmA, xmmF ; xmmA=(00 04 08 0C 10 14 18 1C 20 24 28 2C 30 34 38 3C)
+ punpckhwd xmmB, xmmF ; xmmB=(01 05 09 0D 11 15 19 1D 21 25 29 2D 31 35 39 3D)
+
+ movdqa xmmG, xmmD
+ punpcklwd xmmD, xmmC ; xmmD=(02 06 0A 0E 12 16 1A 1E 22 26 2A 2E 32 36 3A 3E)
+ punpckhwd xmmG, xmmC ; xmmG=(03 07 0B 0F 13 17 1B 1F 23 27 2B 2F 33 37 3B 3F)
+
+ movdqa xmmE, xmmA
+ punpcklbw xmmA, xmmD ; xmmA=(00 02 04 06 08 0A 0C 0E 10 12 14 16 18 1A 1C 1E)
+ punpckhbw xmmE, xmmD ; xmmE=(20 22 24 26 28 2A 2C 2E 30 32 34 36 38 3A 3C 3E)
+
+ movdqa xmmH, xmmB
+ punpcklbw xmmB, xmmG ; xmmB=(01 03 05 07 09 0B 0D 0F 11 13 15 17 19 1B 1D 1F)
+ punpckhbw xmmH, xmmG ; xmmH=(21 23 25 27 29 2B 2D 2F 31 33 35 37 39 3B 3D 3F)
+
+ pxor xmmF, xmmF
+
+ movdqa xmmC, xmmA
+ punpcklbw xmmA, xmmF ; xmmA=(00 02 04 06 08 0A 0C 0E)
+ punpckhbw xmmC, xmmF ; xmmC=(10 12 14 16 18 1A 1C 1E)
+
+ movdqa xmmD, xmmB
+ punpcklbw xmmB, xmmF ; xmmB=(01 03 05 07 09 0B 0D 0F)
+ punpckhbw xmmD, xmmF ; xmmD=(11 13 15 17 19 1B 1D 1F)
+
+ movdqa xmmG, xmmE
+ punpcklbw xmmE, xmmF ; xmmE=(20 22 24 26 28 2A 2C 2E)
+ punpckhbw xmmG, xmmF ; xmmG=(30 32 34 36 38 3A 3C 3E)
+
+ punpcklbw xmmF, xmmH
+ punpckhbw xmmH, xmmH
+ psrlw xmmF, BYTE_BIT ; xmmF=(21 23 25 27 29 2B 2D 2F)
+ psrlw xmmH, BYTE_BIT ; xmmH=(31 33 35 37 39 3B 3D 3F)
+
+%endif ; RGB_PIXELSIZE ; ---------------
+
+ ; xmm0=R(02468ACE)=RE, xmm2=G(02468ACE)=GE, xmm4=B(02468ACE)=BE
+ ; xmm1=R(13579BDF)=RO, xmm3=G(13579BDF)=GO, xmm5=B(13579BDF)=BO
+
+ ; (Original)
+ ; Y = 0.29900 * R + 0.58700 * G + 0.11400 * B
+ ; Cb = -0.16874 * R - 0.33126 * G + 0.50000 * B + CENTERJSAMPLE
+ ; Cr = 0.50000 * R - 0.41869 * G - 0.08131 * B + CENTERJSAMPLE
+ ;
+ ; (This implementation)
+ ; Y = 0.29900 * R + 0.33700 * G + 0.11400 * B + 0.25000 * G
+ ; Cb = -0.16874 * R - 0.33126 * G + 0.50000 * B + CENTERJSAMPLE
+ ; Cr = 0.50000 * R - 0.41869 * G - 0.08131 * B + CENTERJSAMPLE
+
+ movdqa XMMWORD [wk(0)], xmm0 ; wk(0)=RE
+ movdqa XMMWORD [wk(1)], xmm1 ; wk(1)=RO
+ movdqa XMMWORD [wk(2)], xmm4 ; wk(2)=BE
+ movdqa XMMWORD [wk(3)], xmm5 ; wk(3)=BO
+
+ movdqa xmm6, xmm1
+ punpcklwd xmm1, xmm3
+ punpckhwd xmm6, xmm3
+ movdqa xmm7, xmm1
+ movdqa xmm4, xmm6
+ pmaddwd xmm1, [GOTOFF(eax,PW_F0299_F0337)] ; xmm1=ROL*FIX(0.299)+GOL*FIX(0.337)
+ pmaddwd xmm6, [GOTOFF(eax,PW_F0299_F0337)] ; xmm6=ROH*FIX(0.299)+GOH*FIX(0.337)
+ pmaddwd xmm7, [GOTOFF(eax,PW_MF016_MF033)] ; xmm7=ROL*-FIX(0.168)+GOL*-FIX(0.331)
+ pmaddwd xmm4, [GOTOFF(eax,PW_MF016_MF033)] ; xmm4=ROH*-FIX(0.168)+GOH*-FIX(0.331)
+
+ movdqa XMMWORD [wk(4)], xmm1 ; wk(4)=ROL*FIX(0.299)+GOL*FIX(0.337)
+ movdqa XMMWORD [wk(5)], xmm6 ; wk(5)=ROH*FIX(0.299)+GOH*FIX(0.337)
+
+ pxor xmm1, xmm1
+ pxor xmm6, xmm6
+ punpcklwd xmm1, xmm5 ; xmm1=BOL
+ punpckhwd xmm6, xmm5 ; xmm6=BOH
+ psrld xmm1, 1 ; xmm1=BOL*FIX(0.500)
+ psrld xmm6, 1 ; xmm6=BOH*FIX(0.500)
+
+ movdqa xmm5, [GOTOFF(eax,PD_ONEHALFM1_CJ)] ; xmm5=[PD_ONEHALFM1_CJ]
+
+ paddd xmm7, xmm1
+ paddd xmm4, xmm6
+ paddd xmm7, xmm5
+ paddd xmm4, xmm5
+ psrld xmm7, SCALEBITS ; xmm7=CbOL
+ psrld xmm4, SCALEBITS ; xmm4=CbOH
+ packssdw xmm7, xmm4 ; xmm7=CbO
+
+ movdqa xmm1, XMMWORD [wk(2)] ; xmm1=BE
+
+ movdqa xmm6, xmm0
+ punpcklwd xmm0, xmm2
+ punpckhwd xmm6, xmm2
+ movdqa xmm5, xmm0
+ movdqa xmm4, xmm6
+ pmaddwd xmm0, [GOTOFF(eax,PW_F0299_F0337)] ; xmm0=REL*FIX(0.299)+GEL*FIX(0.337)
+ pmaddwd xmm6, [GOTOFF(eax,PW_F0299_F0337)] ; xmm6=REH*FIX(0.299)+GEH*FIX(0.337)
+ pmaddwd xmm5, [GOTOFF(eax,PW_MF016_MF033)] ; xmm5=REL*-FIX(0.168)+GEL*-FIX(0.331)
+ pmaddwd xmm4, [GOTOFF(eax,PW_MF016_MF033)] ; xmm4=REH*-FIX(0.168)+GEH*-FIX(0.331)
+
+ movdqa XMMWORD [wk(6)], xmm0 ; wk(6)=REL*FIX(0.299)+GEL*FIX(0.337)
+ movdqa XMMWORD [wk(7)], xmm6 ; wk(7)=REH*FIX(0.299)+GEH*FIX(0.337)
+
+ pxor xmm0, xmm0
+ pxor xmm6, xmm6
+ punpcklwd xmm0, xmm1 ; xmm0=BEL
+ punpckhwd xmm6, xmm1 ; xmm6=BEH
+ psrld xmm0, 1 ; xmm0=BEL*FIX(0.500)
+ psrld xmm6, 1 ; xmm6=BEH*FIX(0.500)
+
+ movdqa xmm1, [GOTOFF(eax,PD_ONEHALFM1_CJ)] ; xmm1=[PD_ONEHALFM1_CJ]
+
+ paddd xmm5, xmm0
+ paddd xmm4, xmm6
+ paddd xmm5, xmm1
+ paddd xmm4, xmm1
+ psrld xmm5, SCALEBITS ; xmm5=CbEL
+ psrld xmm4, SCALEBITS ; xmm4=CbEH
+ packssdw xmm5, xmm4 ; xmm5=CbE
+
+ psllw xmm7, BYTE_BIT
+ por xmm5, xmm7 ; xmm5=Cb
+ movdqa XMMWORD [ebx], xmm5 ; Save Cb
+
+ movdqa xmm0, XMMWORD [wk(3)] ; xmm0=BO
+ movdqa xmm6, XMMWORD [wk(2)] ; xmm6=BE
+ movdqa xmm1, XMMWORD [wk(1)] ; xmm1=RO
+
+ movdqa xmm4, xmm0
+ punpcklwd xmm0, xmm3
+ punpckhwd xmm4, xmm3
+ movdqa xmm7, xmm0
+ movdqa xmm5, xmm4
+ pmaddwd xmm0, [GOTOFF(eax,PW_F0114_F0250)] ; xmm0=BOL*FIX(0.114)+GOL*FIX(0.250)
+ pmaddwd xmm4, [GOTOFF(eax,PW_F0114_F0250)] ; xmm4=BOH*FIX(0.114)+GOH*FIX(0.250)
+ pmaddwd xmm7, [GOTOFF(eax,PW_MF008_MF041)] ; xmm7=BOL*-FIX(0.081)+GOL*-FIX(0.418)
+ pmaddwd xmm5, [GOTOFF(eax,PW_MF008_MF041)] ; xmm5=BOH*-FIX(0.081)+GOH*-FIX(0.418)
+
+ movdqa xmm3, [GOTOFF(eax,PD_ONEHALF)] ; xmm3=[PD_ONEHALF]
+
+ paddd xmm0, XMMWORD [wk(4)]
+ paddd xmm4, XMMWORD [wk(5)]
+ paddd xmm0, xmm3
+ paddd xmm4, xmm3
+ psrld xmm0, SCALEBITS ; xmm0=YOL
+ psrld xmm4, SCALEBITS ; xmm4=YOH
+ packssdw xmm0, xmm4 ; xmm0=YO
+
+ pxor xmm3, xmm3
+ pxor xmm4, xmm4
+ punpcklwd xmm3, xmm1 ; xmm3=ROL
+ punpckhwd xmm4, xmm1 ; xmm4=ROH
+ psrld xmm3, 1 ; xmm3=ROL*FIX(0.500)
+ psrld xmm4, 1 ; xmm4=ROH*FIX(0.500)
+
+ movdqa xmm1, [GOTOFF(eax,PD_ONEHALFM1_CJ)] ; xmm1=[PD_ONEHALFM1_CJ]
+
+ paddd xmm7, xmm3
+ paddd xmm5, xmm4
+ paddd xmm7, xmm1
+ paddd xmm5, xmm1
+ psrld xmm7, SCALEBITS ; xmm7=CrOL
+ psrld xmm5, SCALEBITS ; xmm5=CrOH
+ packssdw xmm7, xmm5 ; xmm7=CrO
+
+ movdqa xmm3, XMMWORD [wk(0)] ; xmm3=RE
+
+ movdqa xmm4, xmm6
+ punpcklwd xmm6, xmm2
+ punpckhwd xmm4, xmm2
+ movdqa xmm1, xmm6
+ movdqa xmm5, xmm4
+ pmaddwd xmm6, [GOTOFF(eax,PW_F0114_F0250)] ; xmm6=BEL*FIX(0.114)+GEL*FIX(0.250)
+ pmaddwd xmm4, [GOTOFF(eax,PW_F0114_F0250)] ; xmm4=BEH*FIX(0.114)+GEH*FIX(0.250)
+ pmaddwd xmm1, [GOTOFF(eax,PW_MF008_MF041)] ; xmm1=BEL*-FIX(0.081)+GEL*-FIX(0.418)
+ pmaddwd xmm5, [GOTOFF(eax,PW_MF008_MF041)] ; xmm5=BEH*-FIX(0.081)+GEH*-FIX(0.418)
+
+ movdqa xmm2, [GOTOFF(eax,PD_ONEHALF)] ; xmm2=[PD_ONEHALF]
+
+ paddd xmm6, XMMWORD [wk(6)]
+ paddd xmm4, XMMWORD [wk(7)]
+ paddd xmm6, xmm2
+ paddd xmm4, xmm2
+ psrld xmm6, SCALEBITS ; xmm6=YEL
+ psrld xmm4, SCALEBITS ; xmm4=YEH
+ packssdw xmm6, xmm4 ; xmm6=YE
+
+ psllw xmm0, BYTE_BIT
+ por xmm6, xmm0 ; xmm6=Y
+ movdqa XMMWORD [edi], xmm6 ; Save Y
+
+ pxor xmm2, xmm2
+ pxor xmm4, xmm4
+ punpcklwd xmm2, xmm3 ; xmm2=REL
+ punpckhwd xmm4, xmm3 ; xmm4=REH
+ psrld xmm2, 1 ; xmm2=REL*FIX(0.500)
+ psrld xmm4, 1 ; xmm4=REH*FIX(0.500)
+
+ movdqa xmm0, [GOTOFF(eax,PD_ONEHALFM1_CJ)] ; xmm0=[PD_ONEHALFM1_CJ]
+
+ paddd xmm1, xmm2
+ paddd xmm5, xmm4
+ paddd xmm1, xmm0
+ paddd xmm5, xmm0
+ psrld xmm1, SCALEBITS ; xmm1=CrEL
+ psrld xmm5, SCALEBITS ; xmm5=CrEH
+ packssdw xmm1, xmm5 ; xmm1=CrE
+
+ psllw xmm7, BYTE_BIT
+ por xmm1, xmm7 ; xmm1=Cr
+ movdqa XMMWORD [edx], xmm1 ; Save Cr
+
+ sub ecx, byte SIZEOF_XMMWORD
+ add esi, byte RGB_PIXELSIZE*SIZEOF_XMMWORD ; inptr
+ add edi, byte SIZEOF_XMMWORD ; outptr0
+ add ebx, byte SIZEOF_XMMWORD ; outptr1
+ add edx, byte SIZEOF_XMMWORD ; outptr2
+ cmp ecx, byte SIZEOF_XMMWORD
+ jae near .columnloop
+ test ecx, ecx
+ jnz near .column_ld1
+
+ pop ecx ; col
+ pop esi
+ pop edi
+ pop ebx
+ pop edx
+ poppic eax
+
+ add esi, byte SIZEOF_JSAMPROW ; input_buf
+ add edi, byte SIZEOF_JSAMPROW
+ add ebx, byte SIZEOF_JSAMPROW
+ add edx, byte SIZEOF_JSAMPROW
+ dec eax ; num_rows
+ jg near .rowloop
+
+.return:
+ pop edi
+ pop esi
+; pop edx ; need not be preserved
+; pop ecx ; need not be preserved
+ pop ebx
+ mov esp, ebp ; esp <- aligned ebp
+ pop esp ; esp <- original ebp
+ pop ebp
+ ret
+
+; For some reason, the OS X linker does not honor the request to align the
+; segment unless we do this.
+ align 32
diff --git a/simd/i386/jccolor-avx2.asm b/simd/i386/jccolor-avx2.asm
new file mode 100644
index 0000000..9d8c56a
--- /dev/null
+++ b/simd/i386/jccolor-avx2.asm
@@ -0,0 +1,122 @@
+;
+; jccolor.asm - colorspace conversion (AVX2)
+;
+; Copyright (C) 2009, 2016, D. R. Commander.
+; Copyright (C) 2015, Intel Corporation.
+;
+; Based on the x86 SIMD extension for IJG JPEG library
+; Copyright (C) 1999-2006, MIYASAKA Masaru.
+; For conditions of distribution and use, see copyright notice in jsimdext.inc
+;
+; This file should be assembled with NASM (Netwide Assembler),
+; can *not* be assembled with Microsoft's MASM or any compatible
+; assembler (including Borland's Turbo Assembler).
+; NASM is available from http://nasm.sourceforge.net/ or
+; http://sourceforge.net/project/showfiles.php?group_id=6208
+;
+; [TAB8]
+
+%include "jsimdext.inc"
+
+; --------------------------------------------------------------------------
+
+%define SCALEBITS 16
+
+F_0_081 equ 5329 ; FIX(0.08131)
+F_0_114 equ 7471 ; FIX(0.11400)
+F_0_168 equ 11059 ; FIX(0.16874)
+F_0_250 equ 16384 ; FIX(0.25000)
+F_0_299 equ 19595 ; FIX(0.29900)
+F_0_331 equ 21709 ; FIX(0.33126)
+F_0_418 equ 27439 ; FIX(0.41869)
+F_0_587 equ 38470 ; FIX(0.58700)
+F_0_337 equ (F_0_587 - F_0_250) ; FIX(0.58700) - FIX(0.25000)
+
+; --------------------------------------------------------------------------
+ SECTION SEG_CONST
+
+ alignz 32
+ global EXTN(jconst_rgb_ycc_convert_avx2)
+
+EXTN(jconst_rgb_ycc_convert_avx2):
+
+PW_F0299_F0337 times 8 dw F_0_299, F_0_337
+PW_F0114_F0250 times 8 dw F_0_114, F_0_250
+PW_MF016_MF033 times 8 dw -F_0_168,-F_0_331
+PW_MF008_MF041 times 8 dw -F_0_081,-F_0_418
+PD_ONEHALFM1_CJ times 8 dd (1 << (SCALEBITS-1)) - 1 + (CENTERJSAMPLE << SCALEBITS)
+PD_ONEHALF times 8 dd (1 << (SCALEBITS-1))
+
+ alignz 32
+
+; --------------------------------------------------------------------------
+ SECTION SEG_TEXT
+ BITS 32
+
+%include "jccolext-avx2.asm"
+
+%undef RGB_RED
+%undef RGB_GREEN
+%undef RGB_BLUE
+%undef RGB_PIXELSIZE
+%define RGB_RED EXT_RGB_RED
+%define RGB_GREEN EXT_RGB_GREEN
+%define RGB_BLUE EXT_RGB_BLUE
+%define RGB_PIXELSIZE EXT_RGB_PIXELSIZE
+%define jsimd_rgb_ycc_convert_avx2 jsimd_extrgb_ycc_convert_avx2
+%include "jccolext-avx2.asm"
+
+%undef RGB_RED
+%undef RGB_GREEN
+%undef RGB_BLUE
+%undef RGB_PIXELSIZE
+%define RGB_RED EXT_RGBX_RED
+%define RGB_GREEN EXT_RGBX_GREEN
+%define RGB_BLUE EXT_RGBX_BLUE
+%define RGB_PIXELSIZE EXT_RGBX_PIXELSIZE
+%define jsimd_rgb_ycc_convert_avx2 jsimd_extrgbx_ycc_convert_avx2
+%include "jccolext-avx2.asm"
+
+%undef RGB_RED
+%undef RGB_GREEN
+%undef RGB_BLUE
+%undef RGB_PIXELSIZE
+%define RGB_RED EXT_BGR_RED
+%define RGB_GREEN EXT_BGR_GREEN
+%define RGB_BLUE EXT_BGR_BLUE
+%define RGB_PIXELSIZE EXT_BGR_PIXELSIZE
+%define jsimd_rgb_ycc_convert_avx2 jsimd_extbgr_ycc_convert_avx2
+%include "jccolext-avx2.asm"
+
+%undef RGB_RED
+%undef RGB_GREEN
+%undef RGB_BLUE
+%undef RGB_PIXELSIZE
+%define RGB_RED EXT_BGRX_RED
+%define RGB_GREEN EXT_BGRX_GREEN
+%define RGB_BLUE EXT_BGRX_BLUE
+%define RGB_PIXELSIZE EXT_BGRX_PIXELSIZE
+%define jsimd_rgb_ycc_convert_avx2 jsimd_extbgrx_ycc_convert_avx2
+%include "jccolext-avx2.asm"
+
+%undef RGB_RED
+%undef RGB_GREEN
+%undef RGB_BLUE
+%undef RGB_PIXELSIZE
+%define RGB_RED EXT_XBGR_RED
+%define RGB_GREEN EXT_XBGR_GREEN
+%define RGB_BLUE EXT_XBGR_BLUE
+%define RGB_PIXELSIZE EXT_XBGR_PIXELSIZE
+%define jsimd_rgb_ycc_convert_avx2 jsimd_extxbgr_ycc_convert_avx2
+%include "jccolext-avx2.asm"
+
+%undef RGB_RED
+%undef RGB_GREEN
+%undef RGB_BLUE
+%undef RGB_PIXELSIZE
+%define RGB_RED EXT_XRGB_RED
+%define RGB_GREEN EXT_XRGB_GREEN
+%define RGB_BLUE EXT_XRGB_BLUE
+%define RGB_PIXELSIZE EXT_XRGB_PIXELSIZE
+%define jsimd_rgb_ycc_convert_avx2 jsimd_extxrgb_ycc_convert_avx2
+%include "jccolext-avx2.asm"
diff --git a/simd/jccolor-mmx.asm b/simd/i386/jccolor-mmx.asm
similarity index 97%
rename from simd/jccolor-mmx.asm
rename to simd/i386/jccolor-mmx.asm
index c4e6d88..aecdeed 100644
--- a/simd/jccolor-mmx.asm
+++ b/simd/i386/jccolor-mmx.asm
@@ -2,7 +2,7 @@
; jccolor.asm - colorspace conversion (MMX)
;
; Copyright 2009 Pierre Ossman <ossman@cendio.se> for Cendio AB
-; Copyright (C) 2009, D. R. Commander.
+; Copyright (C) 2009, 2016, D. R. Commander.
;
; Based on the x86 SIMD extension for IJG JPEG library
; Copyright (C) 1999-2006, MIYASAKA Masaru.
@@ -35,7 +35,7 @@
; --------------------------------------------------------------------------
SECTION SEG_CONST
- alignz 16
+ alignz 32
global EXTN(jconst_rgb_ycc_convert_mmx)
EXTN(jconst_rgb_ycc_convert_mmx):
@@ -47,7 +47,7 @@
PD_ONEHALFM1_CJ times 2 dd (1 << (SCALEBITS-1)) - 1 + (CENTERJSAMPLE << SCALEBITS)
PD_ONEHALF times 2 dd (1 << (SCALEBITS-1))
- alignz 16
+ alignz 32
; --------------------------------------------------------------------------
SECTION SEG_TEXT
diff --git a/simd/jccolor-sse2.asm b/simd/i386/jccolor-sse2.asm
similarity index 80%
rename from simd/jccolor-sse2.asm
rename to simd/i386/jccolor-sse2.asm
index 13124d1..f9b4c91 100644
--- a/simd/jccolor-sse2.asm
+++ b/simd/i386/jccolor-sse2.asm
@@ -1,7 +1,7 @@
;
; jccolor.asm - colorspace conversion (SSE2)
;
-; Copyright (C) 2009, D. R. Commander.
+; Copyright (C) 2009, 2016, D. R. Commander.
;
; Based on the x86 SIMD extension for IJG JPEG library
; Copyright (C) 1999-2006, MIYASAKA Masaru.
@@ -19,23 +19,23 @@
; --------------------------------------------------------------------------
-%define SCALEBITS 16
+%define SCALEBITS 16
-F_0_081 equ 5329 ; FIX(0.08131)
-F_0_114 equ 7471 ; FIX(0.11400)
-F_0_168 equ 11059 ; FIX(0.16874)
-F_0_250 equ 16384 ; FIX(0.25000)
-F_0_299 equ 19595 ; FIX(0.29900)
-F_0_331 equ 21709 ; FIX(0.33126)
-F_0_418 equ 27439 ; FIX(0.41869)
-F_0_587 equ 38470 ; FIX(0.58700)
-F_0_337 equ (F_0_587 - F_0_250) ; FIX(0.58700) - FIX(0.25000)
+F_0_081 equ 5329 ; FIX(0.08131)
+F_0_114 equ 7471 ; FIX(0.11400)
+F_0_168 equ 11059 ; FIX(0.16874)
+F_0_250 equ 16384 ; FIX(0.25000)
+F_0_299 equ 19595 ; FIX(0.29900)
+F_0_331 equ 21709 ; FIX(0.33126)
+F_0_418 equ 27439 ; FIX(0.41869)
+F_0_587 equ 38470 ; FIX(0.58700)
+F_0_337 equ (F_0_587 - F_0_250) ; FIX(0.58700) - FIX(0.25000)
; --------------------------------------------------------------------------
- SECTION SEG_CONST
+ SECTION SEG_CONST
- alignz 16
- global EXTN(jconst_rgb_ycc_convert_sse2)
+ alignz 32
+ global EXTN(jconst_rgb_ycc_convert_sse2)
EXTN(jconst_rgb_ycc_convert_sse2):
@@ -46,11 +46,11 @@
PD_ONEHALFM1_CJ times 4 dd (1 << (SCALEBITS-1)) - 1 + (CENTERJSAMPLE << SCALEBITS)
PD_ONEHALF times 4 dd (1 << (SCALEBITS-1))
- alignz 16
+ alignz 32
; --------------------------------------------------------------------------
- SECTION SEG_TEXT
- BITS 32
+ SECTION SEG_TEXT
+ BITS 32
%include "jccolext-sse2.asm"
diff --git a/simd/i386/jcgray-avx2.asm b/simd/i386/jcgray-avx2.asm
new file mode 100644
index 0000000..5d78e63
--- /dev/null
+++ b/simd/i386/jcgray-avx2.asm
@@ -0,0 +1,115 @@
+;
+; jcgray.asm - grayscale colorspace conversion (AVX2)
+;
+; Copyright (C) 2011, 2016, D. R. Commander.
+; Copyright (C) 2015, Intel Corporation.
+;
+; Based on the x86 SIMD extension for IJG JPEG library
+; Copyright (C) 1999-2006, MIYASAKA Masaru.
+; For conditions of distribution and use, see copyright notice in jsimdext.inc
+;
+; This file should be assembled with NASM (Netwide Assembler),
+; can *not* be assembled with Microsoft's MASM or any compatible
+; assembler (including Borland's Turbo Assembler).
+; NASM is available from http://nasm.sourceforge.net/ or
+; http://sourceforge.net/project/showfiles.php?group_id=6208
+;
+; [TAB8]
+
+%include "jsimdext.inc"
+
+; --------------------------------------------------------------------------
+
+%define SCALEBITS 16
+
+F_0_114 equ 7471 ; FIX(0.11400)
+F_0_250 equ 16384 ; FIX(0.25000)
+F_0_299 equ 19595 ; FIX(0.29900)
+F_0_587 equ 38470 ; FIX(0.58700)
+F_0_337 equ (F_0_587 - F_0_250) ; FIX(0.58700) - FIX(0.25000)
+
+; --------------------------------------------------------------------------
+ SECTION SEG_CONST
+
+ alignz 32
+ global EXTN(jconst_rgb_gray_convert_avx2)
+
+EXTN(jconst_rgb_gray_convert_avx2):
+
+PW_F0299_F0337 times 8 dw F_0_299, F_0_337
+PW_F0114_F0250 times 8 dw F_0_114, F_0_250
+PD_ONEHALF times 8 dd (1 << (SCALEBITS-1))
+
+ alignz 32
+
+; --------------------------------------------------------------------------
+ SECTION SEG_TEXT
+ BITS 32
+
+%include "jcgryext-avx2.asm"
+
+%undef RGB_RED
+%undef RGB_GREEN
+%undef RGB_BLUE
+%undef RGB_PIXELSIZE
+%define RGB_RED EXT_RGB_RED
+%define RGB_GREEN EXT_RGB_GREEN
+%define RGB_BLUE EXT_RGB_BLUE
+%define RGB_PIXELSIZE EXT_RGB_PIXELSIZE
+%define jsimd_rgb_gray_convert_avx2 jsimd_extrgb_gray_convert_avx2
+%include "jcgryext-avx2.asm"
+
+%undef RGB_RED
+%undef RGB_GREEN
+%undef RGB_BLUE
+%undef RGB_PIXELSIZE
+%define RGB_RED EXT_RGBX_RED
+%define RGB_GREEN EXT_RGBX_GREEN
+%define RGB_BLUE EXT_RGBX_BLUE
+%define RGB_PIXELSIZE EXT_RGBX_PIXELSIZE
+%define jsimd_rgb_gray_convert_avx2 jsimd_extrgbx_gray_convert_avx2
+%include "jcgryext-avx2.asm"
+
+%undef RGB_RED
+%undef RGB_GREEN
+%undef RGB_BLUE
+%undef RGB_PIXELSIZE
+%define RGB_RED EXT_BGR_RED
+%define RGB_GREEN EXT_BGR_GREEN
+%define RGB_BLUE EXT_BGR_BLUE
+%define RGB_PIXELSIZE EXT_BGR_PIXELSIZE
+%define jsimd_rgb_gray_convert_avx2 jsimd_extbgr_gray_convert_avx2
+%include "jcgryext-avx2.asm"
+
+%undef RGB_RED
+%undef RGB_GREEN
+%undef RGB_BLUE
+%undef RGB_PIXELSIZE
+%define RGB_RED EXT_BGRX_RED
+%define RGB_GREEN EXT_BGRX_GREEN
+%define RGB_BLUE EXT_BGRX_BLUE
+%define RGB_PIXELSIZE EXT_BGRX_PIXELSIZE
+%define jsimd_rgb_gray_convert_avx2 jsimd_extbgrx_gray_convert_avx2
+%include "jcgryext-avx2.asm"
+
+%undef RGB_RED
+%undef RGB_GREEN
+%undef RGB_BLUE
+%undef RGB_PIXELSIZE
+%define RGB_RED EXT_XBGR_RED
+%define RGB_GREEN EXT_XBGR_GREEN
+%define RGB_BLUE EXT_XBGR_BLUE
+%define RGB_PIXELSIZE EXT_XBGR_PIXELSIZE
+%define jsimd_rgb_gray_convert_avx2 jsimd_extxbgr_gray_convert_avx2
+%include "jcgryext-avx2.asm"
+
+%undef RGB_RED
+%undef RGB_GREEN
+%undef RGB_BLUE
+%undef RGB_PIXELSIZE
+%define RGB_RED EXT_XRGB_RED
+%define RGB_GREEN EXT_XRGB_GREEN
+%define RGB_BLUE EXT_XRGB_BLUE
+%define RGB_PIXELSIZE EXT_XRGB_PIXELSIZE
+%define jsimd_rgb_gray_convert_avx2 jsimd_extxrgb_gray_convert_avx2
+%include "jcgryext-avx2.asm"
diff --git a/simd/jcgray-mmx.asm b/simd/i386/jcgray-mmx.asm
similarity index 97%
rename from simd/jcgray-mmx.asm
rename to simd/i386/jcgray-mmx.asm
index 0819b6c..48decc8 100644
--- a/simd/jcgray-mmx.asm
+++ b/simd/i386/jcgray-mmx.asm
@@ -2,7 +2,7 @@
; jcgray.asm - grayscale colorspace conversion (MMX)
;
; Copyright 2009 Pierre Ossman <ossman@cendio.se> for Cendio AB
-; Copyright (C) 2011, D. R. Commander.
+; Copyright (C) 2011, 2016, D. R. Commander.
;
; Based on the x86 SIMD extension for IJG JPEG library
; Copyright (C) 1999-2006, MIYASAKA Masaru.
@@ -31,7 +31,7 @@
; --------------------------------------------------------------------------
SECTION SEG_CONST
- alignz 16
+ alignz 32
global EXTN(jconst_rgb_gray_convert_mmx)
EXTN(jconst_rgb_gray_convert_mmx):
@@ -40,7 +40,7 @@
PW_F0114_F0250 times 2 dw F_0_114, F_0_250
PD_ONEHALF times 2 dd (1 << (SCALEBITS-1))
- alignz 16
+ alignz 32
; --------------------------------------------------------------------------
SECTION SEG_TEXT
diff --git a/simd/jcgray-sse2.asm b/simd/i386/jcgray-sse2.asm
similarity index 80%
rename from simd/jcgray-sse2.asm
rename to simd/i386/jcgray-sse2.asm
index 5b0b466..dedfc8a 100644
--- a/simd/jcgray-sse2.asm
+++ b/simd/i386/jcgray-sse2.asm
@@ -1,7 +1,7 @@
;
; jcgray.asm - grayscale colorspace conversion (SSE2)
;
-; Copyright (C) 2011, D. R. Commander.
+; Copyright (C) 2011, 2016, D. R. Commander.
;
; Based on the x86 SIMD extension for IJG JPEG library
; Copyright (C) 1999-2006, MIYASAKA Masaru.
@@ -19,31 +19,31 @@
; --------------------------------------------------------------------------
-%define SCALEBITS 16
+%define SCALEBITS 16
-F_0_114 equ 7471 ; FIX(0.11400)
-F_0_250 equ 16384 ; FIX(0.25000)
-F_0_299 equ 19595 ; FIX(0.29900)
-F_0_587 equ 38470 ; FIX(0.58700)
-F_0_337 equ (F_0_587 - F_0_250) ; FIX(0.58700) - FIX(0.25000)
+F_0_114 equ 7471 ; FIX(0.11400)
+F_0_250 equ 16384 ; FIX(0.25000)
+F_0_299 equ 19595 ; FIX(0.29900)
+F_0_587 equ 38470 ; FIX(0.58700)
+F_0_337 equ (F_0_587 - F_0_250) ; FIX(0.58700) - FIX(0.25000)
; --------------------------------------------------------------------------
- SECTION SEG_CONST
+ SECTION SEG_CONST
- alignz 16
- global EXTN(jconst_rgb_gray_convert_sse2)
+ alignz 32
+ global EXTN(jconst_rgb_gray_convert_sse2)
EXTN(jconst_rgb_gray_convert_sse2):
-PW_F0299_F0337 times 4 dw F_0_299, F_0_337
-PW_F0114_F0250 times 4 dw F_0_114, F_0_250
-PD_ONEHALF times 4 dd (1 << (SCALEBITS-1))
+PW_F0299_F0337 times 4 dw F_0_299, F_0_337
+PW_F0114_F0250 times 4 dw F_0_114, F_0_250
+PD_ONEHALF times 4 dd (1 << (SCALEBITS-1))
- alignz 16
+ alignz 32
; --------------------------------------------------------------------------
- SECTION SEG_TEXT
- BITS 32
+ SECTION SEG_TEXT
+ BITS 32
%include "jcgryext-sse2.asm"
diff --git a/simd/i386/jcgryext-avx2.asm b/simd/i386/jcgryext-avx2.asm
new file mode 100644
index 0000000..4f81c20
--- /dev/null
+++ b/simd/i386/jcgryext-avx2.asm
@@ -0,0 +1,459 @@
+;
+; jcgryext.asm - grayscale colorspace conversion (AVX2)
+;
+; Copyright (C) 2011, 2016, D. R. Commander.
+; Copyright (C) 2015, Intel Corporation.
+;
+; Based on the x86 SIMD extension for IJG JPEG library
+; Copyright (C) 1999-2006, MIYASAKA Masaru.
+; For conditions of distribution and use, see copyright notice in jsimdext.inc
+;
+; This file should be assembled with NASM (Netwide Assembler),
+; can *not* be assembled with Microsoft's MASM or any compatible
+; assembler (including Borland's Turbo Assembler).
+; NASM is available from http://nasm.sourceforge.net/ or
+; http://sourceforge.net/project/showfiles.php?group_id=6208
+;
+; [TAB8]
+
+%include "jcolsamp.inc"
+
+; --------------------------------------------------------------------------
+;
+; Convert some rows of samples to the output colorspace.
+;
+; GLOBAL(void)
+; jsimd_rgb_gray_convert_avx2 (JDIMENSION img_width,
+; JSAMPARRAY input_buf, JSAMPIMAGE output_buf,
+; JDIMENSION output_row, int num_rows);
+;
+
+%define img_width(b) (b)+8 ; JDIMENSION img_width
+%define input_buf(b) (b)+12 ; JSAMPARRAY input_buf
+%define output_buf(b) (b)+16 ; JSAMPIMAGE output_buf
+%define output_row(b) (b)+20 ; JDIMENSION output_row
+%define num_rows(b) (b)+24 ; int num_rows
+
+%define original_ebp ebp+0
+%define wk(i) ebp-(WK_NUM-(i))*SIZEOF_YMMWORD ; ymmword wk[WK_NUM]
+%define WK_NUM 2
+%define gotptr wk(0)-SIZEOF_POINTER ; void * gotptr
+
+ align 32
+
+ global EXTN(jsimd_rgb_gray_convert_avx2)
+
+EXTN(jsimd_rgb_gray_convert_avx2):
+ push ebp
+ mov eax, esp ; eax = original ebp
+ sub esp, byte 4
+ and esp, byte (-SIZEOF_YMMWORD) ; align to 256 bits
+ mov [esp], eax
+ mov ebp, esp ; ebp = aligned ebp
+ lea esp, [wk(0)]
+ pushpic eax ; make a room for GOT address
+ push ebx
+; push ecx ; need not be preserved
+; push edx ; need not be preserved
+ push esi
+ push edi
+
+ get_GOT ebx ; get GOT address
+ movpic POINTER [gotptr], ebx ; save GOT address
+
+ mov ecx, JDIMENSION [img_width(eax)]
+ test ecx, ecx
+ jz near .return
+
+ push ecx
+
+ mov esi, JSAMPIMAGE [output_buf(eax)]
+ mov ecx, JDIMENSION [output_row(eax)]
+ mov edi, JSAMPARRAY [esi+0*SIZEOF_JSAMPARRAY]
+ lea edi, [edi+ecx*SIZEOF_JSAMPROW]
+
+ pop ecx
+
+ mov esi, JSAMPARRAY [input_buf(eax)]
+ mov eax, INT [num_rows(eax)]
+ test eax, eax
+ jle near .return
+ alignx 16, 7
+.rowloop:
+ pushpic eax
+ push edi
+ push esi
+ push ecx ; col
+
+ mov esi, JSAMPROW [esi] ; inptr
+ mov edi, JSAMPROW [edi] ; outptr0
+ movpic eax, POINTER [gotptr] ; load GOT address (eax)
+
+ cmp ecx, byte SIZEOF_YMMWORD
+ jae near .columnloop
+ alignx 16, 7
+
+%if RGB_PIXELSIZE == 3 ; ---------------
+
+.column_ld1:
+ push eax
+ push edx
+ lea ecx, [ecx+ecx*2] ; imul ecx,RGB_PIXELSIZE
+ test cl, SIZEOF_BYTE
+ jz short .column_ld2
+ sub ecx, byte SIZEOF_BYTE
+ movzx eax, BYTE [esi+ecx]
+.column_ld2:
+ test cl, SIZEOF_WORD
+ jz short .column_ld4
+ sub ecx, byte SIZEOF_WORD
+ movzx edx, WORD [esi+ecx]
+ shl eax, WORD_BIT
+ or eax, edx
+.column_ld4:
+ vmovd xmmA, eax
+ pop edx
+ pop eax
+ test cl, SIZEOF_DWORD
+ jz short .column_ld8
+ sub ecx, byte SIZEOF_DWORD
+ vmovd xmmF, XMM_DWORD [esi+ecx]
+ vpslldq xmmA, xmmA, SIZEOF_DWORD
+ vpor xmmA, xmmA, xmmF
+.column_ld8:
+ test cl, SIZEOF_MMWORD
+ jz short .column_ld16
+ sub ecx, byte SIZEOF_MMWORD
+ vmovq xmmB, XMM_MMWORD [esi+ecx]
+ vpslldq xmmA, xmmA, SIZEOF_MMWORD
+ vpor xmmA, xmmA, xmmB
+.column_ld16:
+ test cl, SIZEOF_XMMWORD
+ jz short .column_ld32
+ sub ecx, byte SIZEOF_XMMWORD
+ vmovdqu xmmB, XMM_MMWORD [esi+ecx]
+ vperm2i128 ymmA, ymmA, ymmA, 1
+ vpor ymmA, ymmB
+.column_ld32:
+ test cl, SIZEOF_YMMWORD
+ jz short .column_ld64
+ sub ecx, byte SIZEOF_YMMWORD
+ vmovdqa ymmF, ymmA
+ vmovdqu ymmA, YMMWORD [esi+0*SIZEOF_YMMWORD]
+.column_ld64:
+ test cl, 2*SIZEOF_YMMWORD
+ mov ecx, SIZEOF_YMMWORD
+ jz short .rgb_gray_cnv
+ vmovdqa ymmB, ymmA
+ vmovdqu ymmA, YMMWORD [esi+0*SIZEOF_YMMWORD]
+ vmovdqu ymmF, YMMWORD [esi+1*SIZEOF_YMMWORD]
+ jmp short .rgb_gray_cnv
+ alignx 16, 7
+
+.columnloop:
+ vmovdqu ymmA, YMMWORD [esi+0*SIZEOF_YMMWORD]
+ vmovdqu ymmF, YMMWORD [esi+1*SIZEOF_YMMWORD]
+ vmovdqu ymmB, YMMWORD [esi+2*SIZEOF_YMMWORD]
+
+.rgb_gray_cnv:
+ ; ymmA=(00 10 20 01 11 21 02 12 22 03 13 23 04 14 24 05
+ ; 15 25 06 16 26 07 17 27 08 18 28 09 19 29 0A 1A)
+ ; ymmF=(2A 0B 1B 2B 0C 1C 2C 0D 1D 2D 0E 1E 2E 0F 1F 2F
+ ; 0G 1G 2G 0H 1H 2H 0I 1I 2I 0J 1J 2J 0K 1K 2K 0L)
+ ; ymmB=(1L 2L 0M 1M 2M 0N 1N 2N 0O 1O 2O 0P 1P 2P 0Q 1Q
+ ; 2Q 0R 1R 2R 0S 1S 2S 0T 1T 2T 0U 1U 2U 0V 1V 2V)
+
+ vmovdqu ymmC, ymmA
+ vinserti128 ymmA, ymmF, xmmA, 0 ; ymmA=(00 10 20 01 11 21 02 12 22 03 13 23 04 14 24 05
+ ; 0G 1G 2G 0H 1H 2H 0I 1I 2I 0J 1J 2J 0K 1K 2K 0L)
+ vinserti128 ymmC, ymmC, xmmB, 0 ; ymmC=(1L 2L 0M 1M 2M 0N 1N 2N 0O 1O 2O 0P 1P 2P 0Q 1Q
+ ; 15 25 06 16 26 07 17 27 08 18 28 09 19 29 0A 1A)
+ vinserti128 ymmB, ymmB, xmmF, 0 ; ymmB=(2A 0B 1B 2B 0C 1C 2C 0D 1D 2D 0E 1E 2E 0F 1F 2F
+ ; 2Q 0R 1R 2R 0S 1S 2S 0T 1T 2T 0U 1U 2U 0V 1V 2V)
+ vperm2i128 ymmF, ymmC, ymmC, 1 ; ymmF=(15 25 06 16 26 07 17 27 08 18 28 09 19 29 0A 1A
+ ; 1L 2L 0M 1M 2M 0N 1N 2N 0O 1O 2O 0P 1P 2P 0Q 1Q)
+
+ vmovdqa ymmG, ymmA
+ vpslldq ymmA, ymmA, 8 ; ymmA=(-- -- -- -- -- -- -- -- 00 10 20 01 11 21 02 12
+ ; 22 03 13 23 04 14 24 05 0G 1G 2G 0H 1H 2H 0I 1I)
+ vpsrldq ymmG, ymmG, 8 ; ymmG=(22 03 13 23 04 14 24 05 0G 1G 2G 0H 1H 2H 0I 1I
+ ; 2I 0J 1J 2J 0K 1K 2K 0L -- -- -- -- -- -- -- --)
+
+ vpunpckhbw ymmA, ymmA, ymmF ; ymmA=(00 08 10 18 20 28 01 09 11 19 21 29 02 0A 12 1A
+ ; 0G 0O 1G 1O 2G 2O 0H 0P 1H 1P 2H 2P 0I 0Q 1I 1Q)
+ vpslldq ymmF, ymmF, 8 ; ymmF=(-- -- -- -- -- -- -- -- 15 25 06 16 26 07 17 27
+ ; 08 18 28 09 19 29 0A 1A 1L 2L 0M 1M 2M 0N 1N 2N)
+
+ vpunpcklbw ymmG, ymmG, ymmB ; ymmG=(22 2A 03 0B 13 1B 23 2B 04 0C 14 1C 24 2C 05 0D
+ ; 2I 2Q 0J 0R 1J 1R 2J 2R 0K 0S 1K 1S 2K 2S 0L 0T)
+ vpunpckhbw ymmF, ymmF, ymmB ; ymmF=(15 1D 25 2D 06 0E 16 1E 26 2E 07 0F 17 1F 27 2F
+ ; 1L 1T 2L 2T 0M 0U 1M 1U 2M 2U 0N 0V 1N 1V 2N 2V)
+
+ vmovdqa ymmD, ymmA
+ vpslldq ymmA, ymmA, 8 ; ymmA=(-- -- -- -- -- -- -- -- 00 08 10 18 20 28 01 09
+ ; 11 19 21 29 02 0A 12 1A 0G 0O 1G 1O 2G 2O 0H 0P)
+ vpsrldq ymmD, ymmD, 8 ; ymmD=(11 19 21 29 02 0A 12 1A 0G 0O 1G 1O 2G 2O 0H 0P
+ ; 1H 1P 2H 2P 0I 0Q 1I 1Q -- -- -- -- -- -- -- --)
+
+ vpunpckhbw ymmA, ymmA, ymmG ; ymmA=(00 04 08 0C 10 14 18 1C 20 24 28 2C 01 05 09 0D
+ ; 0G 0K 0O 0S 1G 1K 1O 1S 2G 2K 2O 2S 0H 0L 0P 0T)
+ vpslldq ymmG, ymmG, 8 ; ymmG=(-- -- -- -- -- -- -- -- 22 2A 03 0B 13 1B 23 2B
+ ; 04 0C 14 1C 24 2C 05 0D 2I 2Q 0J 0R 1J 1R 2J 2R)
+
+ vpunpcklbw ymmD, ymmD, ymmF ; ymmD=(11 15 19 1D 21 25 29 2D 02 06 0A 0E 12 16 1A 1E
+ ; 1H 1L 1P 1T 2H 2L 2P 2T 0I 0M 0Q 0U 1I 1M 1Q 1U)
+ vpunpckhbw ymmG, ymmG, ymmF ; ymmG=(22 26 2A 2E 03 07 0B 0F 13 17 1B 1F 23 27 2B 2F
+ ; 2I 2M 2Q 2U 0J 0N 0R 0V 1J 1N 1R 1V 2J 2N 2R 2V)
+
+ vmovdqa ymmE, ymmA
+ vpslldq ymmA, ymmA, 8 ; ymmA=(-- -- -- -- -- -- -- -- 00 04 08 0C 10 14 18 1C
+ ; 20 24 28 2C 01 05 09 0D 0G 0K 0O 0S 1G 1K 1O 1S)
+ vpsrldq ymmE, ymmE, 8 ; ymmE=(20 24 28 2C 01 05 09 0D 0G 0K 0O 0S 1G 1K 1O 1S
+ ; 2G 2K 2O 2S 0H 0L 0P 0T -- -- -- -- -- -- -- --)
+
+ vpunpckhbw ymmA, ymmA, ymmD ; ymmA=(00 02 04 06 08 0A 0C 0E 10 12 14 16 18 1A 1C 1E
+ ; 0G 0I 0K 0M 0O 0Q 0S 0U 1G 1I 1K 1M 1O 1Q 1S 1U)
+ vpslldq ymmD, ymmD, 8 ; ymmD=(-- -- -- -- -- -- -- -- 11 15 19 1D 21 25 29 2D
+ ; 02 06 0A 0E 12 16 1A 1E 1H 1L 1P 1T 2H 2L 2P 2T)
+
+ vpunpcklbw ymmE, ymmE, ymmG ; ymmE=(20 22 24 26 28 2A 2C 2E 01 03 05 07 09 0B 0D 0F
+ ; 2G 2I 2K 2M 2O 2Q 2S 2U 0H 0J 0L 0N 0P 0R 0T 0V)
+ vpunpckhbw ymmD, ymmD, ymmG ; ymmD=(11 13 15 17 19 1B 1D 1F 21 23 25 27 29 2B 2D 2F
+ ; 1H 1J 1L 1N 1P 1R 1T 1V 2H 2J 2L 2N 2P 2R 2T 2V)
+
+ vpxor ymmH, ymmH, ymmH
+
+ vmovdqa ymmC, ymmA
+ vpunpcklbw ymmA, ymmA, ymmH ; ymmA=(00 02 04 06 08 0A 0C 0E 0G 0I 0K 0M 0O 0Q 0S 0U)
+ vpunpckhbw ymmC, ymmC, ymmH ; ymmC=(10 12 14 16 18 1A 1C 1E 1G 1I 1K 1M 1O 1Q 1S 1U)
+
+ vmovdqa ymmB, ymmE
+ vpunpcklbw ymmE, ymmE, ymmH ; ymmE=(20 22 24 26 28 2A 2C 2E 2G 2I 2K 2M 2O 2Q 2S 2U)
+ vpunpckhbw ymmB, ymmB, ymmH ; ymmB=(01 03 05 07 09 0B 0D 0F 0H 0J 0L 0N 0P 0R 0T 0V)
+
+ vmovdqa ymmF, ymmD
+ vpunpcklbw ymmD, ymmD, ymmH ; ymmD=(11 13 15 17 19 1B 1D 1F 1H 1J 1L 1N 1P 1R 1T 1V)
+ vpunpckhbw ymmF, ymmF, ymmH ; ymmF=(21 23 25 27 29 2B 2D 2F 2H 2J 2L 2N 2P 2R 2T 2V)
+
+%else ; RGB_PIXELSIZE == 4 ; -----------
+
+.column_ld1:
+ test cl, SIZEOF_XMMWORD/16
+ jz short .column_ld2
+ sub ecx, byte SIZEOF_XMMWORD/16
+ vmovd xmmA, XMM_DWORD [esi+ecx*RGB_PIXELSIZE]
+.column_ld2:
+ test cl, SIZEOF_XMMWORD/8
+ jz short .column_ld4
+ sub ecx, byte SIZEOF_XMMWORD/8
+ vmovq xmmF, XMM_MMWORD [esi+ecx*RGB_PIXELSIZE]
+ vpslldq xmmA, xmmA, SIZEOF_MMWORD
+ vpor xmmA, xmmA, xmmF
+.column_ld4:
+ test cl, SIZEOF_XMMWORD/4
+ jz short .column_ld8
+ sub ecx, byte SIZEOF_XMMWORD/4
+ vmovdqa xmmF, xmmA
+ vperm2i128 ymmF, ymmF, ymmF, 1
+ vmovdqu xmmA, XMMWORD [esi+ecx*RGB_PIXELSIZE]
+ vpor ymmA, ymmA, ymmF
+.column_ld8:
+ test cl, SIZEOF_XMMWORD/2
+ jz short .column_ld16
+ sub ecx, byte SIZEOF_XMMWORD/2
+ vmovdqa ymmF, ymmA
+ vmovdqu ymmA, YMMWORD [esi+ecx*RGB_PIXELSIZE]
+.column_ld16:
+ test cl, SIZEOF_XMMWORD
+ mov ecx, SIZEOF_YMMWORD
+ jz short .rgb_gray_cnv
+ vmovdqa ymmE, ymmA
+ vmovdqa ymmH, ymmF
+ vmovdqu ymmA, YMMWORD [esi+0*SIZEOF_YMMWORD]
+ vmovdqu ymmF, YMMWORD [esi+1*SIZEOF_YMMWORD]
+ jmp short .rgb_gray_cnv
+ alignx 16, 7
+
+.columnloop:
+ vmovdqu ymmA, YMMWORD [esi+0*SIZEOF_YMMWORD]
+ vmovdqu ymmF, YMMWORD [esi+1*SIZEOF_YMMWORD]
+ vmovdqu ymmE, YMMWORD [esi+2*SIZEOF_YMMWORD]
+ vmovdqu ymmH, YMMWORD [esi+3*SIZEOF_YMMWORD]
+
+.rgb_gray_cnv:
+ ; ymmA=(00 10 20 30 01 11 21 31 02 12 22 32 03 13 23 33
+ ; 04 14 24 34 05 15 25 35 06 16 26 36 07 17 27 37)
+ ; ymmF=(08 18 28 38 09 19 29 39 0A 1A 2A 3A 0B 1B 2B 3B
+ ; 0C 1C 2C 3C 0D 1D 2D 3D 0E 1E 2E 3E 0F 1F 2F 3F)
+ ; ymmE=(0G 1G 2G 3G 0H 1H 2H 3H 0I 1I 2I 3I 0J 1J 2J 3J
+ ; 0K 1K 2K 3K 0L 1L 2L 3L 0M 1M 2M 3M 0N 1N 2N 3N)
+ ; ymmH=(0O 1O 2O 3O 0P 1P 2P 3P 0Q 1Q 2Q 3Q 0R 1R 2R 3R
+ ; 0S 1S 2S 3S 0T 1T 2T 3T 0U 1U 2U 3U 0V 1V 2V 3V)
+
+ vmovdqa ymmB, ymmA
+ vinserti128 ymmA, ymmA, xmmE, 1 ; ymmA=(00 10 20 30 01 11 21 31 02 12 22 32 03 13 23 33
+ ; 0G 1G 2G 3G 0H 1H 2H 3H 0I 1I 2I 3I 0J 1J 2J 3J)
+ vperm2i128 ymmE, ymmB, ymmE, 0x31 ; ymmE=(04 14 24 34 05 15 25 35 06 16 26 36 07 17 27 37
+ ; 0K 1K 2K 3K 0L 1L 2L 3L 0M 1M 2M 3M 0N 1N 2N 3N)
+
+ vmovdqa ymmB, ymmF
+ vinserti128 ymmF, ymmF, xmmH, 1 ; ymmF=(08 18 28 38 09 19 29 39 0A 1A 2A 3A 0B 1B 2B 3B
+ ; 0O 1O 2O 3O 0P 1P 2P 3P 0Q 1Q 2Q 3Q 0R 1R 2R 3R)
+ vperm2i128 ymmH, ymmB, ymmH, 0x31 ; ymmH=(0C 1C 2C 3C 0D 1D 2D 3D 0E 1E 2E 3E 0F 1F 2F 3F
+ ; 0S 1S 2S 3S 0T 1T 2T 3T 0U 1U 2U 3U 0V 1V 2V 3V)
+
+ vmovdqa ymmD, ymmA
+ vpunpcklbw ymmA, ymmA, ymmE ; ymmA=(00 04 10 14 20 24 30 34 01 05 11 15 21 25 31 35
+ ; 0G 0K 1G 1K 2G 2K 3G 3K 0H 0L 1H 1L 2H 2L 3H 3L)
+ vpunpckhbw ymmD, ymmD, ymmE ; ymmD=(02 06 12 16 22 26 32 36 03 07 13 17 23 27 33 37
+ ; 0I 0M 1I 1M 2I 2M 3I 3M 0J 0N 1J 1N 2J 2N 3J 3N)
+
+ vmovdqa ymmC, ymmF
+ vpunpcklbw ymmF, ymmF, ymmH ; ymmF=(08 0C 18 1C 28 2C 38 3C 09 0D 19 1D 29 2D 39 3D
+ ; 0O 0S 1O 1S 2O 2S 3O 3S 0P 0T 1P 1T 2P 2T 3P 3T)
+ vpunpckhbw ymmC, ymmC, ymmH ; ymmC=(0A 0E 1A 1E 2A 2E 3A 3E 0B 0F 1B 1F 2B 2F 3B 3F
+ ; 0Q 0U 1Q 1U 2Q 2U 3Q 3U 0R 0V 1R 1V 2R 2V 3R 3V)
+
+ vmovdqa ymmB, ymmA
+ vpunpcklwd ymmA, ymmA, ymmF ; ymmA=(00 04 08 0C 10 14 18 1C 20 24 28 2C 30 34 38 3C
+ ; 0G 0K 0O 0S 1G 1K 1O 1S 2G 2K 2O 2S 3G 3K 3O 3S)
+ vpunpckhwd ymmB, ymmB, ymmF ; ymmB=(01 05 09 0D 11 15 19 1D 21 25 29 2D 31 35 39 3D
+ ; 0H 0L 0P 0T 1H 1L 1P 1T 2H 2L 2P 2T 3H 3L 3P 3T)
+
+ vmovdqa ymmG, ymmD
+ vpunpcklwd ymmD, ymmD, ymmC ; ymmD=(02 06 0A 0E 12 16 1A 1E 22 26 2A 2E 32 36 3A 3E
+ ; 0I 0M 0Q 0U 1I 1M 1Q 1U 2I 2M 2Q 2U 3I 3M 3Q 3U)
+ vpunpckhwd ymmG, ymmG, ymmC ; ymmG=(03 07 0B 0F 13 17 1B 1F 23 27 2B 2F 33 37 3B 3F
+ ; 0J 0N 0R 0V 1J 1N 1R 1V 2J 2N 2R 2V 3J 3N 3R 3V)
+
+ vmovdqa ymmE, ymmA
+ vpunpcklbw ymmA, ymmA, ymmD ; ymmA=(00 02 04 06 08 0A 0C 0E 10 12 14 16 18 1A 1C 1E
+ ; 0G 0I 0K 0M 0O 0Q 0S 0U 1G 1I 1K 1M 1O 1Q 1S 1U)
+ vpunpckhbw ymmE, ymmE, ymmD ; ymmE=(20 22 24 26 28 2A 2C 2E 30 32 34 36 38 3A 3C 3E
+ ; 2G 2I 2K 2M 2O 2Q 2S 2U 3G 3I 3K 3M 3O 3Q 3S 3U)
+
+ vmovdqa ymmH, ymmB
+ vpunpcklbw ymmB, ymmB, ymmG ; ymmB=(01 03 05 07 09 0B 0D 0F 11 13 15 17 19 1B 1D 1F
+ ; 0H 0J 0L 0N 0P 0R 0T 0V 1H 1J 1L 1N 1P 1R 1T 1V)
+ vpunpckhbw ymmH, ymmH, ymmG ; ymmH=(21 23 25 27 29 2B 2D 2F 31 33 35 37 39 3B 3D 3F
+ ; 2H 2J 2L 2N 2P 2R 2T 2V 3H 3J 3L 3N 3P 3R 3T 3V)
+
+ vpxor ymmF, ymmF, ymmF
+
+ vmovdqa ymmC, ymmA
+ vpunpcklbw ymmA, ymmA, ymmF ; ymmA=(00 02 04 06 08 0A 0C 0E 0G 0I 0K 0M 0O 0Q 0S 0U)
+ vpunpckhbw ymmC, ymmC, ymmF ; ymmC=(10 12 14 16 18 1A 1C 1E 1G 1I 1K 1M 1O 1Q 1S 1U)
+
+ vmovdqa ymmD, ymmB
+ vpunpcklbw ymmB, ymmB, ymmF ; ymmB=(01 03 05 07 09 0B 0D 0F 0H 0J 0L 0N 0P 0R 0T 0V)
+ vpunpckhbw ymmD, ymmD, ymmF ; ymmD=(11 13 15 17 19 1B 1D 1F 1H 1J 1L 1N 1P 1R 1T 1V)
+
+ vmovdqa ymmG, ymmE
+ vpunpcklbw ymmE, ymmE, ymmF ; ymmE=(20 22 24 26 28 2A 2C 2E 2G 2I 2K 2M 2O 2Q 2S 2U)
+ vpunpckhbw ymmG, ymmG, ymmF ; ymmG=(30 32 34 36 38 3A 3C 3E 3G 3I 3K 3M 3O 3Q 3S 3U)
+
+ vpunpcklbw ymmF, ymmF, ymmH
+ vpunpckhbw ymmH, ymmH, ymmH
+ vpsrlw ymmF, ymmF, BYTE_BIT ; ymmF=(21 23 25 27 29 2B 2D 2F 2H 2J 2L 2N 2P 2R 2T 2V)
+ vpsrlw ymmH, ymmH, BYTE_BIT ; ymmH=(31 33 35 37 39 3B 3D 3F 3H 3J 3L 3N 3P 3R 3T 3V)
+
+%endif ; RGB_PIXELSIZE ; ---------------
+
+ ; ymm0=R(02468ACEGIKMOQSU)=RE, ymm2=G(02468ACEGIKMOQSU)=GE, ymm4=B(02468ACEGIKMOQSU)=BE
+ ; ymm1=R(13579BDFHJLNPRTV)=RO, ymm3=G(13579BDFHJLNPRTV)=GO, ymm5=B(13579BDFHJLNPRTV)=BO
+
+ ; (Original)
+ ; Y = 0.29900 * R + 0.58700 * G + 0.11400 * B
+ ;
+ ; (This implementation)
+ ; Y = 0.29900 * R + 0.33700 * G + 0.11400 * B + 0.25000 * G
+
+ vmovdqa ymm6, ymm1
+ vpunpcklwd ymm1, ymm1, ymm3
+ vpunpckhwd ymm6, ymm6, ymm3
+ vpmaddwd ymm1, ymm1, [GOTOFF(eax,PW_F0299_F0337)] ; ymm1=ROL*FIX(0.299)+GOL*FIX(0.337)
+ vpmaddwd ymm6, ymm6, [GOTOFF(eax,PW_F0299_F0337)] ; ymm6=ROH*FIX(0.299)+GOH*FIX(0.337)
+
+ vmovdqa ymm7, ymm6 ; ymm7=ROH*FIX(0.299)+GOH*FIX(0.337)
+
+ vmovdqa ymm6, ymm0
+ vpunpcklwd ymm0, ymm0, ymm2
+ vpunpckhwd ymm6, ymm6, ymm2
+ vpmaddwd ymm0, ymm0, [GOTOFF(eax,PW_F0299_F0337)] ; ymm0=REL*FIX(0.299)+GEL*FIX(0.337)
+ vpmaddwd ymm6, ymm6, [GOTOFF(eax,PW_F0299_F0337)] ; ymm6=REH*FIX(0.299)+GEH*FIX(0.337)
+
+ vmovdqa YMMWORD [wk(0)], ymm0 ; wk(0)=REL*FIX(0.299)+GEL*FIX(0.337)
+ vmovdqa YMMWORD [wk(1)], ymm6 ; wk(1)=REH*FIX(0.299)+GEH*FIX(0.337)
+
+ vmovdqa ymm0, ymm5 ; ymm0=BO
+ vmovdqa ymm6, ymm4 ; ymm6=BE
+
+ vmovdqa ymm4, ymm0
+ vpunpcklwd ymm0, ymm0, ymm3
+ vpunpckhwd ymm4, ymm4, ymm3
+ vpmaddwd ymm0, ymm0, [GOTOFF(eax,PW_F0114_F0250)] ; ymm0=BOL*FIX(0.114)+GOL*FIX(0.250)
+ vpmaddwd ymm4, ymm4, [GOTOFF(eax,PW_F0114_F0250)] ; ymm4=BOH*FIX(0.114)+GOH*FIX(0.250)
+
+ vmovdqa ymm3, [GOTOFF(eax,PD_ONEHALF)] ; ymm3=[PD_ONEHALF]
+
+ vpaddd ymm0, ymm0, ymm1
+ vpaddd ymm4, ymm4, ymm7
+ vpaddd ymm0, ymm0, ymm3
+ vpaddd ymm4, ymm4, ymm3
+ vpsrld ymm0, ymm0, SCALEBITS ; ymm0=YOL
+ vpsrld ymm4, ymm4, SCALEBITS ; ymm4=YOH
+ vpackssdw ymm0, ymm0, ymm4 ; ymm0=YO
+
+ vmovdqa ymm4, ymm6
+ vpunpcklwd ymm6, ymm6, ymm2
+ vpunpckhwd ymm4, ymm4, ymm2
+ vpmaddwd ymm6, ymm6, [GOTOFF(eax,PW_F0114_F0250)] ; ymm6=BEL*FIX(0.114)+GEL*FIX(0.250)
+ vpmaddwd ymm4, ymm4, [GOTOFF(eax,PW_F0114_F0250)] ; ymm4=BEH*FIX(0.114)+GEH*FIX(0.250)
+
+ vmovdqa ymm2, [GOTOFF(eax,PD_ONEHALF)] ; ymm2=[PD_ONEHALF]
+
+ vpaddd ymm6, ymm6, YMMWORD [wk(0)]
+ vpaddd ymm4, ymm4, YMMWORD [wk(1)]
+ vpaddd ymm6, ymm6, ymm2
+ vpaddd ymm4, ymm4, ymm2
+ vpsrld ymm6, ymm6, SCALEBITS ; ymm6=YEL
+ vpsrld ymm4, ymm4, SCALEBITS ; ymm4=YEH
+ vpackssdw ymm6, ymm6, ymm4 ; ymm6=YE
+
+ vpsllw ymm0, ymm0, BYTE_BIT
+ vpor ymm6, ymm6, ymm0 ; ymm6=Y
+ vmovdqu YMMWORD [edi], ymm6 ; Save Y
+
+ sub ecx, byte SIZEOF_YMMWORD
+ add esi, RGB_PIXELSIZE*SIZEOF_YMMWORD ; inptr
+ add edi, byte SIZEOF_YMMWORD ; outptr0
+ cmp ecx, byte SIZEOF_YMMWORD
+ jae near .columnloop
+ test ecx, ecx
+ jnz near .column_ld1
+
+ pop ecx ; col
+ pop esi
+ pop edi
+ poppic eax
+
+ add esi, byte SIZEOF_JSAMPROW ; input_buf
+ add edi, byte SIZEOF_JSAMPROW
+ dec eax ; num_rows
+ jg near .rowloop
+
+.return:
+ vzeroupper
+ pop edi
+ pop esi
+; pop edx ; need not be preserved
+; pop ecx ; need not be preserved
+ pop ebx
+ mov esp, ebp ; esp <- aligned ebp
+ pop esp ; esp <- original ebp
+ pop ebp
+ ret
+
+; For some reason, the OS X linker does not honor the request to align the
+; segment unless we do this.
+ align 32
diff --git a/simd/jcgryext-mmx.asm b/simd/i386/jcgryext-mmx.asm
similarity index 98%
rename from simd/jcgryext-mmx.asm
rename to simd/i386/jcgryext-mmx.asm
index 1c1b8d8..841eb0b 100644
--- a/simd/jcgryext-mmx.asm
+++ b/simd/i386/jcgryext-mmx.asm
@@ -2,7 +2,7 @@
; jcgryext.asm - grayscale colorspace conversion (MMX)
;
; Copyright 2009 Pierre Ossman <ossman@cendio.se> for Cendio AB
-; Copyright (C) 2011, D. R. Commander.
+; Copyright (C) 2011, 2016, D. R. Commander.
;
; Based on the x86 SIMD extension for IJG JPEG library
; Copyright (C) 1999-2006, MIYASAKA Masaru.
@@ -39,7 +39,7 @@
%define WK_NUM 2
%define gotptr wk(0)-SIZEOF_POINTER ; void * gotptr
- align 16
+ align 32
global EXTN(jsimd_rgb_gray_convert_mmx)
EXTN(jsimd_rgb_gray_convert_mmx):
@@ -353,4 +353,4 @@
; For some reason, the OS X linker does not honor the request to align the
; segment unless we do this.
- align 16
+ align 32
diff --git a/simd/i386/jcgryext-sse2.asm b/simd/i386/jcgryext-sse2.asm
new file mode 100644
index 0000000..40505a5
--- /dev/null
+++ b/simd/i386/jcgryext-sse2.asm
@@ -0,0 +1,384 @@
+;
+; jcgryext.asm - grayscale colorspace conversion (SSE2)
+;
+; Copyright (C) 2011, 2016, D. R. Commander.
+;
+; Based on the x86 SIMD extension for IJG JPEG library
+; Copyright (C) 1999-2006, MIYASAKA Masaru.
+; For conditions of distribution and use, see copyright notice in jsimdext.inc
+;
+; This file should be assembled with NASM (Netwide Assembler),
+; can *not* be assembled with Microsoft's MASM or any compatible
+; assembler (including Borland's Turbo Assembler).
+; NASM is available from http://nasm.sourceforge.net/ or
+; http://sourceforge.net/project/showfiles.php?group_id=6208
+;
+; [TAB8]
+
+%include "jcolsamp.inc"
+
+; --------------------------------------------------------------------------
+;
+; Convert some rows of samples to the output colorspace.
+;
+; GLOBAL(void)
+; jsimd_rgb_gray_convert_sse2 (JDIMENSION img_width,
+; JSAMPARRAY input_buf, JSAMPIMAGE output_buf,
+; JDIMENSION output_row, int num_rows);
+;
+
+%define img_width(b) (b)+8 ; JDIMENSION img_width
+%define input_buf(b) (b)+12 ; JSAMPARRAY input_buf
+%define output_buf(b) (b)+16 ; JSAMPIMAGE output_buf
+%define output_row(b) (b)+20 ; JDIMENSION output_row
+%define num_rows(b) (b)+24 ; int num_rows
+
+%define original_ebp ebp+0
+%define wk(i) ebp-(WK_NUM-(i))*SIZEOF_XMMWORD ; xmmword wk[WK_NUM]
+%define WK_NUM 2
+%define gotptr wk(0)-SIZEOF_POINTER ; void * gotptr
+
+ align 32
+
+ global EXTN(jsimd_rgb_gray_convert_sse2)
+
+EXTN(jsimd_rgb_gray_convert_sse2):
+ push ebp
+ mov eax, esp ; eax = original ebp
+ sub esp, byte 4
+ and esp, byte (-SIZEOF_XMMWORD) ; align to 128 bits
+ mov [esp], eax
+ mov ebp, esp ; ebp = aligned ebp
+ lea esp, [wk(0)]
+ pushpic eax ; make a room for GOT address
+ push ebx
+; push ecx ; need not be preserved
+; push edx ; need not be preserved
+ push esi
+ push edi
+
+ get_GOT ebx ; get GOT address
+ movpic POINTER [gotptr], ebx ; save GOT address
+
+ mov ecx, JDIMENSION [img_width(eax)]
+ test ecx, ecx
+ jz near .return
+
+ push ecx
+
+ mov esi, JSAMPIMAGE [output_buf(eax)]
+ mov ecx, JDIMENSION [output_row(eax)]
+ mov edi, JSAMPARRAY [esi+0*SIZEOF_JSAMPARRAY]
+ lea edi, [edi+ecx*SIZEOF_JSAMPROW]
+
+ pop ecx
+
+ mov esi, JSAMPARRAY [input_buf(eax)]
+ mov eax, INT [num_rows(eax)]
+ test eax, eax
+ jle near .return
+ alignx 16, 7
+.rowloop:
+ pushpic eax
+ push edi
+ push esi
+ push ecx ; col
+
+ mov esi, JSAMPROW [esi] ; inptr
+ mov edi, JSAMPROW [edi] ; outptr0
+ movpic eax, POINTER [gotptr] ; load GOT address (eax)
+
+ cmp ecx, byte SIZEOF_XMMWORD
+ jae near .columnloop
+ alignx 16, 7
+
+%if RGB_PIXELSIZE == 3 ; ---------------
+
+.column_ld1:
+ push eax
+ push edx
+ lea ecx, [ecx+ecx*2] ; imul ecx,RGB_PIXELSIZE
+ test cl, SIZEOF_BYTE
+ jz short .column_ld2
+ sub ecx, byte SIZEOF_BYTE
+ movzx eax, BYTE [esi+ecx]
+.column_ld2:
+ test cl, SIZEOF_WORD
+ jz short .column_ld4
+ sub ecx, byte SIZEOF_WORD
+ movzx edx, WORD [esi+ecx]
+ shl eax, WORD_BIT
+ or eax, edx
+.column_ld4:
+ movd xmmA, eax
+ pop edx
+ pop eax
+ test cl, SIZEOF_DWORD
+ jz short .column_ld8
+ sub ecx, byte SIZEOF_DWORD
+ movd xmmF, XMM_DWORD [esi+ecx]
+ pslldq xmmA, SIZEOF_DWORD
+ por xmmA, xmmF
+.column_ld8:
+ test cl, SIZEOF_MMWORD
+ jz short .column_ld16
+ sub ecx, byte SIZEOF_MMWORD
+ movq xmmB, XMM_MMWORD [esi+ecx]
+ pslldq xmmA, SIZEOF_MMWORD
+ por xmmA, xmmB
+.column_ld16:
+ test cl, SIZEOF_XMMWORD
+ jz short .column_ld32
+ movdqa xmmF, xmmA
+ movdqu xmmA, XMMWORD [esi+0*SIZEOF_XMMWORD]
+ mov ecx, SIZEOF_XMMWORD
+ jmp short .rgb_gray_cnv
+.column_ld32:
+ test cl, 2*SIZEOF_XMMWORD
+ mov ecx, SIZEOF_XMMWORD
+ jz short .rgb_gray_cnv
+ movdqa xmmB, xmmA
+ movdqu xmmA, XMMWORD [esi+0*SIZEOF_XMMWORD]
+ movdqu xmmF, XMMWORD [esi+1*SIZEOF_XMMWORD]
+ jmp short .rgb_gray_cnv
+ alignx 16, 7
+
+.columnloop:
+ movdqu xmmA, XMMWORD [esi+0*SIZEOF_XMMWORD]
+ movdqu xmmF, XMMWORD [esi+1*SIZEOF_XMMWORD]
+ movdqu xmmB, XMMWORD [esi+2*SIZEOF_XMMWORD]
+
+.rgb_gray_cnv:
+ ; xmmA=(00 10 20 01 11 21 02 12 22 03 13 23 04 14 24 05)
+ ; xmmF=(15 25 06 16 26 07 17 27 08 18 28 09 19 29 0A 1A)
+ ; xmmB=(2A 0B 1B 2B 0C 1C 2C 0D 1D 2D 0E 1E 2E 0F 1F 2F)
+
+ movdqa xmmG, xmmA
+ pslldq xmmA, 8 ; xmmA=(-- -- -- -- -- -- -- -- 00 10 20 01 11 21 02 12)
+ psrldq xmmG, 8 ; xmmG=(22 03 13 23 04 14 24 05 -- -- -- -- -- -- -- --)
+
+ punpckhbw xmmA, xmmF ; xmmA=(00 08 10 18 20 28 01 09 11 19 21 29 02 0A 12 1A)
+ pslldq xmmF, 8 ; xmmF=(-- -- -- -- -- -- -- -- 15 25 06 16 26 07 17 27)
+
+ punpcklbw xmmG, xmmB ; xmmG=(22 2A 03 0B 13 1B 23 2B 04 0C 14 1C 24 2C 05 0D)
+ punpckhbw xmmF, xmmB ; xmmF=(15 1D 25 2D 06 0E 16 1E 26 2E 07 0F 17 1F 27 2F)
+
+ movdqa xmmD, xmmA
+ pslldq xmmA, 8 ; xmmA=(-- -- -- -- -- -- -- -- 00 08 10 18 20 28 01 09)
+ psrldq xmmD, 8 ; xmmD=(11 19 21 29 02 0A 12 1A -- -- -- -- -- -- -- --)
+
+ punpckhbw xmmA, xmmG ; xmmA=(00 04 08 0C 10 14 18 1C 20 24 28 2C 01 05 09 0D)
+ pslldq xmmG, 8 ; xmmG=(-- -- -- -- -- -- -- -- 22 2A 03 0B 13 1B 23 2B)
+
+ punpcklbw xmmD, xmmF ; xmmD=(11 15 19 1D 21 25 29 2D 02 06 0A 0E 12 16 1A 1E)
+ punpckhbw xmmG, xmmF ; xmmG=(22 26 2A 2E 03 07 0B 0F 13 17 1B 1F 23 27 2B 2F)
+
+ movdqa xmmE, xmmA
+ pslldq xmmA, 8 ; xmmA=(-- -- -- -- -- -- -- -- 00 04 08 0C 10 14 18 1C)
+ psrldq xmmE, 8 ; xmmE=(20 24 28 2C 01 05 09 0D -- -- -- -- -- -- -- --)
+
+ punpckhbw xmmA, xmmD ; xmmA=(00 02 04 06 08 0A 0C 0E 10 12 14 16 18 1A 1C 1E)
+ pslldq xmmD, 8 ; xmmD=(-- -- -- -- -- -- -- -- 11 15 19 1D 21 25 29 2D)
+
+ punpcklbw xmmE, xmmG ; xmmE=(20 22 24 26 28 2A 2C 2E 01 03 05 07 09 0B 0D 0F)
+ punpckhbw xmmD, xmmG ; xmmD=(11 13 15 17 19 1B 1D 1F 21 23 25 27 29 2B 2D 2F)
+
+ pxor xmmH, xmmH
+
+ movdqa xmmC, xmmA
+ punpcklbw xmmA, xmmH ; xmmA=(00 02 04 06 08 0A 0C 0E)
+ punpckhbw xmmC, xmmH ; xmmC=(10 12 14 16 18 1A 1C 1E)
+
+ movdqa xmmB, xmmE
+ punpcklbw xmmE, xmmH ; xmmE=(20 22 24 26 28 2A 2C 2E)
+ punpckhbw xmmB, xmmH ; xmmB=(01 03 05 07 09 0B 0D 0F)
+
+ movdqa xmmF, xmmD
+ punpcklbw xmmD, xmmH ; xmmD=(11 13 15 17 19 1B 1D 1F)
+ punpckhbw xmmF, xmmH ; xmmF=(21 23 25 27 29 2B 2D 2F)
+
+%else ; RGB_PIXELSIZE == 4 ; -----------
+
+.column_ld1:
+ test cl, SIZEOF_XMMWORD/16
+ jz short .column_ld2
+ sub ecx, byte SIZEOF_XMMWORD/16
+ movd xmmA, XMM_DWORD [esi+ecx*RGB_PIXELSIZE]
+.column_ld2:
+ test cl, SIZEOF_XMMWORD/8
+ jz short .column_ld4
+ sub ecx, byte SIZEOF_XMMWORD/8
+ movq xmmE, XMM_MMWORD [esi+ecx*RGB_PIXELSIZE]
+ pslldq xmmA, SIZEOF_MMWORD
+ por xmmA, xmmE
+.column_ld4:
+ test cl, SIZEOF_XMMWORD/4
+ jz short .column_ld8
+ sub ecx, byte SIZEOF_XMMWORD/4
+ movdqa xmmE, xmmA
+ movdqu xmmA, XMMWORD [esi+ecx*RGB_PIXELSIZE]
+.column_ld8:
+ test cl, SIZEOF_XMMWORD/2
+ mov ecx, SIZEOF_XMMWORD
+ jz short .rgb_gray_cnv
+ movdqa xmmF, xmmA
+ movdqa xmmH, xmmE
+ movdqu xmmA, XMMWORD [esi+0*SIZEOF_XMMWORD]
+ movdqu xmmE, XMMWORD [esi+1*SIZEOF_XMMWORD]
+ jmp short .rgb_gray_cnv
+ alignx 16, 7
+
+.columnloop:
+ movdqu xmmA, XMMWORD [esi+0*SIZEOF_XMMWORD]
+ movdqu xmmE, XMMWORD [esi+1*SIZEOF_XMMWORD]
+ movdqu xmmF, XMMWORD [esi+2*SIZEOF_XMMWORD]
+ movdqu xmmH, XMMWORD [esi+3*SIZEOF_XMMWORD]
+
+.rgb_gray_cnv:
+ ; xmmA=(00 10 20 30 01 11 21 31 02 12 22 32 03 13 23 33)
+ ; xmmE=(04 14 24 34 05 15 25 35 06 16 26 36 07 17 27 37)
+ ; xmmF=(08 18 28 38 09 19 29 39 0A 1A 2A 3A 0B 1B 2B 3B)
+ ; xmmH=(0C 1C 2C 3C 0D 1D 2D 3D 0E 1E 2E 3E 0F 1F 2F 3F)
+
+ movdqa xmmD, xmmA
+ punpcklbw xmmA, xmmE ; xmmA=(00 04 10 14 20 24 30 34 01 05 11 15 21 25 31 35)
+ punpckhbw xmmD, xmmE ; xmmD=(02 06 12 16 22 26 32 36 03 07 13 17 23 27 33 37)
+
+ movdqa xmmC, xmmF
+ punpcklbw xmmF, xmmH ; xmmF=(08 0C 18 1C 28 2C 38 3C 09 0D 19 1D 29 2D 39 3D)
+ punpckhbw xmmC, xmmH ; xmmC=(0A 0E 1A 1E 2A 2E 3A 3E 0B 0F 1B 1F 2B 2F 3B 3F)
+
+ movdqa xmmB, xmmA
+ punpcklwd xmmA, xmmF ; xmmA=(00 04 08 0C 10 14 18 1C 20 24 28 2C 30 34 38 3C)
+ punpckhwd xmmB, xmmF ; xmmB=(01 05 09 0D 11 15 19 1D 21 25 29 2D 31 35 39 3D)
+
+ movdqa xmmG, xmmD
+ punpcklwd xmmD, xmmC ; xmmD=(02 06 0A 0E 12 16 1A 1E 22 26 2A 2E 32 36 3A 3E)
+ punpckhwd xmmG, xmmC ; xmmG=(03 07 0B 0F 13 17 1B 1F 23 27 2B 2F 33 37 3B 3F)
+
+ movdqa xmmE, xmmA
+ punpcklbw xmmA, xmmD ; xmmA=(00 02 04 06 08 0A 0C 0E 10 12 14 16 18 1A 1C 1E)
+ punpckhbw xmmE, xmmD ; xmmE=(20 22 24 26 28 2A 2C 2E 30 32 34 36 38 3A 3C 3E)
+
+ movdqa xmmH, xmmB
+ punpcklbw xmmB, xmmG ; xmmB=(01 03 05 07 09 0B 0D 0F 11 13 15 17 19 1B 1D 1F)
+ punpckhbw xmmH, xmmG ; xmmH=(21 23 25 27 29 2B 2D 2F 31 33 35 37 39 3B 3D 3F)
+
+ pxor xmmF, xmmF
+
+ movdqa xmmC, xmmA
+ punpcklbw xmmA, xmmF ; xmmA=(00 02 04 06 08 0A 0C 0E)
+ punpckhbw xmmC, xmmF ; xmmC=(10 12 14 16 18 1A 1C 1E)
+
+ movdqa xmmD, xmmB
+ punpcklbw xmmB, xmmF ; xmmB=(01 03 05 07 09 0B 0D 0F)
+ punpckhbw xmmD, xmmF ; xmmD=(11 13 15 17 19 1B 1D 1F)
+
+ movdqa xmmG, xmmE
+ punpcklbw xmmE, xmmF ; xmmE=(20 22 24 26 28 2A 2C 2E)
+ punpckhbw xmmG, xmmF ; xmmG=(30 32 34 36 38 3A 3C 3E)
+
+ punpcklbw xmmF, xmmH
+ punpckhbw xmmH, xmmH
+ psrlw xmmF, BYTE_BIT ; xmmF=(21 23 25 27 29 2B 2D 2F)
+ psrlw xmmH, BYTE_BIT ; xmmH=(31 33 35 37 39 3B 3D 3F)
+
+%endif ; RGB_PIXELSIZE ; ---------------
+
+ ; xmm0=R(02468ACE)=RE, xmm2=G(02468ACE)=GE, xmm4=B(02468ACE)=BE
+ ; xmm1=R(13579BDF)=RO, xmm3=G(13579BDF)=GO, xmm5=B(13579BDF)=BO
+
+ ; (Original)
+ ; Y = 0.29900 * R + 0.58700 * G + 0.11400 * B
+ ;
+ ; (This implementation)
+ ; Y = 0.29900 * R + 0.33700 * G + 0.11400 * B + 0.25000 * G
+
+ movdqa xmm6, xmm1
+ punpcklwd xmm1, xmm3
+ punpckhwd xmm6, xmm3
+ pmaddwd xmm1, [GOTOFF(eax,PW_F0299_F0337)] ; xmm1=ROL*FIX(0.299)+GOL*FIX(0.337)
+ pmaddwd xmm6, [GOTOFF(eax,PW_F0299_F0337)] ; xmm6=ROH*FIX(0.299)+GOH*FIX(0.337)
+
+ movdqa xmm7, xmm6 ; xmm7=ROH*FIX(0.299)+GOH*FIX(0.337)
+
+ movdqa xmm6, xmm0
+ punpcklwd xmm0, xmm2
+ punpckhwd xmm6, xmm2
+ pmaddwd xmm0, [GOTOFF(eax,PW_F0299_F0337)] ; xmm0=REL*FIX(0.299)+GEL*FIX(0.337)
+ pmaddwd xmm6, [GOTOFF(eax,PW_F0299_F0337)] ; xmm6=REH*FIX(0.299)+GEH*FIX(0.337)
+
+ movdqa XMMWORD [wk(0)], xmm0 ; wk(0)=REL*FIX(0.299)+GEL*FIX(0.337)
+ movdqa XMMWORD [wk(1)], xmm6 ; wk(1)=REH*FIX(0.299)+GEH*FIX(0.337)
+
+ movdqa xmm0, xmm5 ; xmm0=BO
+ movdqa xmm6, xmm4 ; xmm6=BE
+
+ movdqa xmm4, xmm0
+ punpcklwd xmm0, xmm3
+ punpckhwd xmm4, xmm3
+ pmaddwd xmm0, [GOTOFF(eax,PW_F0114_F0250)] ; xmm0=BOL*FIX(0.114)+GOL*FIX(0.250)
+ pmaddwd xmm4, [GOTOFF(eax,PW_F0114_F0250)] ; xmm4=BOH*FIX(0.114)+GOH*FIX(0.250)
+
+ movdqa xmm3, [GOTOFF(eax,PD_ONEHALF)] ; xmm3=[PD_ONEHALF]
+
+ paddd xmm0, xmm1
+ paddd xmm4, xmm7
+ paddd xmm0, xmm3
+ paddd xmm4, xmm3
+ psrld xmm0, SCALEBITS ; xmm0=YOL
+ psrld xmm4, SCALEBITS ; xmm4=YOH
+ packssdw xmm0, xmm4 ; xmm0=YO
+
+ movdqa xmm4, xmm6
+ punpcklwd xmm6, xmm2
+ punpckhwd xmm4, xmm2
+ pmaddwd xmm6, [GOTOFF(eax,PW_F0114_F0250)] ; xmm6=BEL*FIX(0.114)+GEL*FIX(0.250)
+ pmaddwd xmm4, [GOTOFF(eax,PW_F0114_F0250)] ; xmm4=BEH*FIX(0.114)+GEH*FIX(0.250)
+
+ movdqa xmm2, [GOTOFF(eax,PD_ONEHALF)] ; xmm2=[PD_ONEHALF]
+
+ paddd xmm6, XMMWORD [wk(0)]
+ paddd xmm4, XMMWORD [wk(1)]
+ paddd xmm6, xmm2
+ paddd xmm4, xmm2
+ psrld xmm6, SCALEBITS ; xmm6=YEL
+ psrld xmm4, SCALEBITS ; xmm4=YEH
+ packssdw xmm6, xmm4 ; xmm6=YE
+
+ psllw xmm0, BYTE_BIT
+ por xmm6, xmm0 ; xmm6=Y
+ movdqa XMMWORD [edi], xmm6 ; Save Y
+
+ sub ecx, byte SIZEOF_XMMWORD
+ add esi, byte RGB_PIXELSIZE*SIZEOF_XMMWORD ; inptr
+ add edi, byte SIZEOF_XMMWORD ; outptr0
+ cmp ecx, byte SIZEOF_XMMWORD
+ jae near .columnloop
+ test ecx, ecx
+ jnz near .column_ld1
+
+ pop ecx ; col
+ pop esi
+ pop edi
+ poppic eax
+
+ add esi, byte SIZEOF_JSAMPROW ; input_buf
+ add edi, byte SIZEOF_JSAMPROW
+ dec eax ; num_rows
+ jg near .rowloop
+
+.return:
+ pop edi
+ pop esi
+; pop edx ; need not be preserved
+; pop ecx ; need not be preserved
+ pop ebx
+ mov esp, ebp ; esp <- aligned ebp
+ pop esp ; esp <- original ebp
+ pop ebp
+ ret
+
+; For some reason, the OS X linker does not honor the request to align the
+; segment unless we do this.
+ align 32
diff --git a/simd/i386/jchuff-sse2.asm b/simd/i386/jchuff-sse2.asm
new file mode 100644
index 0000000..85b748a
--- /dev/null
+++ b/simd/i386/jchuff-sse2.asm
@@ -0,0 +1,426 @@
+;
+; jchuff-sse2.asm - Huffman entropy encoding (SSE2)
+;
+; Copyright (C) 2009-2011, 2014-2017, D. R. Commander.
+; Copyright (C) 2015, Matthieu Darbois.
+;
+; Based on the x86 SIMD extension for IJG JPEG library
+; Copyright (C) 1999-2006, MIYASAKA Masaru.
+; For conditions of distribution and use, see copyright notice in jsimdext.inc
+;
+; This file should be assembled with NASM (Netwide Assembler),
+; can *not* be assembled with Microsoft's MASM or any compatible
+; assembler (including Borland's Turbo Assembler).
+; NASM is available from http://nasm.sourceforge.net/ or
+; http://sourceforge.net/project/showfiles.php?group_id=6208
+;
+; This file contains an SSE2 implementation for Huffman coding of one block.
+; The following code is based directly on jchuff.c; see jchuff.c for more
+; details.
+;
+; [TAB8]
+
+%include "jsimdext.inc"
+
+; --------------------------------------------------------------------------
+ SECTION SEG_CONST
+
+ alignz 32
+ global EXTN(jconst_huff_encode_one_block)
+
+EXTN(jconst_huff_encode_one_block):
+
+%include "jpeg_nbits_table.inc"
+
+ alignz 32
+
+; --------------------------------------------------------------------------
+ SECTION SEG_TEXT
+ BITS 32
+
+; These macros perform the same task as the emit_bits() function in the
+; original libjpeg code. In addition to reducing overhead by explicitly
+; inlining the code, additional performance is achieved by taking into
+; account the size of the bit buffer and waiting until it is almost full
+; before emptying it. This mostly benefits 64-bit platforms, since 6
+; bytes can be stored in a 64-bit bit buffer before it has to be emptied.
+
+%macro EMIT_BYTE 0
+ sub put_bits, 8 ; put_bits -= 8;
+ mov edx, put_buffer
+ mov ecx, put_bits
+ shr edx, cl ; c = (JOCTET)GETJOCTET(put_buffer >> put_bits);
+ mov byte [eax], dl ; *buffer++ = c;
+ add eax, 1
+ cmp dl, 0xFF ; need to stuff a zero byte?
+ jne %%.EMIT_BYTE_END
+ mov byte [eax], 0 ; *buffer++ = 0;
+ add eax, 1
+%%.EMIT_BYTE_END:
+%endmacro
+
+%macro PUT_BITS 1
+ add put_bits, ecx ; put_bits += size;
+ shl put_buffer, cl ; put_buffer = (put_buffer << size);
+ or put_buffer, %1
+%endmacro
+
+%macro CHECKBUF15 0
+ cmp put_bits, 16 ; if (put_bits > 31) {
+ jl %%.CHECKBUF15_END
+ mov eax, POINTER [esp+buffer]
+ EMIT_BYTE
+ EMIT_BYTE
+ mov POINTER [esp+buffer], eax
+%%.CHECKBUF15_END:
+%endmacro
+
+%macro EMIT_BITS 1
+ PUT_BITS %1
+ CHECKBUF15
+%endmacro
+
+%macro kloop_prepare 37 ;(ko, jno0, ..., jno31, xmm0, xmm1, xmm2, xmm3)
+ pxor xmm4, xmm4 ; __m128i neg = _mm_setzero_si128();
+ pxor xmm5, xmm5 ; __m128i neg = _mm_setzero_si128();
+ pxor xmm6, xmm6 ; __m128i neg = _mm_setzero_si128();
+ pxor xmm7, xmm7 ; __m128i neg = _mm_setzero_si128();
+ pinsrw %34, word [esi + %2 * SIZEOF_WORD], 0 ; xmm_shadow[0] = block[jno0];
+ pinsrw %35, word [esi + %10 * SIZEOF_WORD], 0 ; xmm_shadow[8] = block[jno8];
+ pinsrw %36, word [esi + %18 * SIZEOF_WORD], 0 ; xmm_shadow[16] = block[jno16];
+ pinsrw %37, word [esi + %26 * SIZEOF_WORD], 0 ; xmm_shadow[24] = block[jno24];
+ pinsrw %34, word [esi + %3 * SIZEOF_WORD], 1 ; xmm_shadow[1] = block[jno1];
+ pinsrw %35, word [esi + %11 * SIZEOF_WORD], 1 ; xmm_shadow[9] = block[jno9];
+ pinsrw %36, word [esi + %19 * SIZEOF_WORD], 1 ; xmm_shadow[17] = block[jno17];
+ pinsrw %37, word [esi + %27 * SIZEOF_WORD], 1 ; xmm_shadow[25] = block[jno25];
+ pinsrw %34, word [esi + %4 * SIZEOF_WORD], 2 ; xmm_shadow[2] = block[jno2];
+ pinsrw %35, word [esi + %12 * SIZEOF_WORD], 2 ; xmm_shadow[10] = block[jno10];
+ pinsrw %36, word [esi + %20 * SIZEOF_WORD], 2 ; xmm_shadow[18] = block[jno18];
+ pinsrw %37, word [esi + %28 * SIZEOF_WORD], 2 ; xmm_shadow[26] = block[jno26];
+ pinsrw %34, word [esi + %5 * SIZEOF_WORD], 3 ; xmm_shadow[3] = block[jno3];
+ pinsrw %35, word [esi + %13 * SIZEOF_WORD], 3 ; xmm_shadow[11] = block[jno11];
+ pinsrw %36, word [esi + %21 * SIZEOF_WORD], 3 ; xmm_shadow[19] = block[jno19];
+ pinsrw %37, word [esi + %29 * SIZEOF_WORD], 3 ; xmm_shadow[27] = block[jno27];
+ pinsrw %34, word [esi + %6 * SIZEOF_WORD], 4 ; xmm_shadow[4] = block[jno4];
+ pinsrw %35, word [esi + %14 * SIZEOF_WORD], 4 ; xmm_shadow[12] = block[jno12];
+ pinsrw %36, word [esi + %22 * SIZEOF_WORD], 4 ; xmm_shadow[20] = block[jno20];
+ pinsrw %37, word [esi + %30 * SIZEOF_WORD], 4 ; xmm_shadow[28] = block[jno28];
+ pinsrw %34, word [esi + %7 * SIZEOF_WORD], 5 ; xmm_shadow[5] = block[jno5];
+ pinsrw %35, word [esi + %15 * SIZEOF_WORD], 5 ; xmm_shadow[13] = block[jno13];
+ pinsrw %36, word [esi + %23 * SIZEOF_WORD], 5 ; xmm_shadow[21] = block[jno21];
+ pinsrw %37, word [esi + %31 * SIZEOF_WORD], 5 ; xmm_shadow[29] = block[jno29];
+ pinsrw %34, word [esi + %8 * SIZEOF_WORD], 6 ; xmm_shadow[6] = block[jno6];
+ pinsrw %35, word [esi + %16 * SIZEOF_WORD], 6 ; xmm_shadow[14] = block[jno14];
+ pinsrw %36, word [esi + %24 * SIZEOF_WORD], 6 ; xmm_shadow[22] = block[jno22];
+ pinsrw %37, word [esi + %32 * SIZEOF_WORD], 6 ; xmm_shadow[30] = block[jno30];
+ pinsrw %34, word [esi + %9 * SIZEOF_WORD], 7 ; xmm_shadow[7] = block[jno7];
+ pinsrw %35, word [esi + %17 * SIZEOF_WORD], 7 ; xmm_shadow[15] = block[jno15];
+ pinsrw %36, word [esi + %25 * SIZEOF_WORD], 7 ; xmm_shadow[23] = block[jno23];
+%if %1 != 32
+ pinsrw %37, word [esi + %33 * SIZEOF_WORD], 7 ; xmm_shadow[31] = block[jno31];
+%else
+ pinsrw %37, ecx, 7 ; xmm_shadow[31] = block[jno31];
+%endif
+ pcmpgtw xmm4, %34 ; neg = _mm_cmpgt_epi16(neg, x1);
+ pcmpgtw xmm5, %35 ; neg = _mm_cmpgt_epi16(neg, x1);
+ pcmpgtw xmm6, %36 ; neg = _mm_cmpgt_epi16(neg, x1);
+ pcmpgtw xmm7, %37 ; neg = _mm_cmpgt_epi16(neg, x1);
+ paddw %34, xmm4 ; x1 = _mm_add_epi16(x1, neg);
+ paddw %35, xmm5 ; x1 = _mm_add_epi16(x1, neg);
+ paddw %36, xmm6 ; x1 = _mm_add_epi16(x1, neg);
+ paddw %37, xmm7 ; x1 = _mm_add_epi16(x1, neg);
+ pxor %34, xmm4 ; x1 = _mm_xor_si128(x1, neg);
+ pxor %35, xmm5 ; x1 = _mm_xor_si128(x1, neg);
+ pxor %36, xmm6 ; x1 = _mm_xor_si128(x1, neg);
+ pxor %37, xmm7 ; x1 = _mm_xor_si128(x1, neg);
+ pxor xmm4, %34 ; neg = _mm_xor_si128(neg, x1);
+ pxor xmm5, %35 ; neg = _mm_xor_si128(neg, x1);
+ pxor xmm6, %36 ; neg = _mm_xor_si128(neg, x1);
+ pxor xmm7, %37 ; neg = _mm_xor_si128(neg, x1);
+ movdqa XMMWORD [esp + t1 + %1 * SIZEOF_WORD], %34 ; _mm_storeu_si128((__m128i *)(t1 + ko), x1);
+ movdqa XMMWORD [esp + t1 + (%1 + 8) * SIZEOF_WORD], %35 ; _mm_storeu_si128((__m128i *)(t1 + ko + 8), x1);
+ movdqa XMMWORD [esp + t1 + (%1 + 16) * SIZEOF_WORD], %36 ; _mm_storeu_si128((__m128i *)(t1 + ko + 16), x1);
+ movdqa XMMWORD [esp + t1 + (%1 + 24) * SIZEOF_WORD], %37 ; _mm_storeu_si128((__m128i *)(t1 + ko + 24), x1);
+ movdqa XMMWORD [esp + t2 + %1 * SIZEOF_WORD], xmm4 ; _mm_storeu_si128((__m128i *)(t2 + ko), neg);
+ movdqa XMMWORD [esp + t2 + (%1 + 8) * SIZEOF_WORD], xmm5 ; _mm_storeu_si128((__m128i *)(t2 + ko + 8), neg);
+ movdqa XMMWORD [esp + t2 + (%1 + 16) * SIZEOF_WORD], xmm6 ; _mm_storeu_si128((__m128i *)(t2 + ko + 16), neg);
+ movdqa XMMWORD [esp + t2 + (%1 + 24) * SIZEOF_WORD], xmm7 ; _mm_storeu_si128((__m128i *)(t2 + ko + 24), neg);
+%endmacro
+
+;
+; Encode a single block's worth of coefficients.
+;
+; GLOBAL(JOCTET*)
+; jsimd_huff_encode_one_block_sse2 (working_state *state, JOCTET *buffer,
+; JCOEFPTR block, int last_dc_val,
+; c_derived_tbl *dctbl, c_derived_tbl *actbl)
+;
+
+; eax + 8 = working_state *state
+; eax + 12 = JOCTET *buffer
+; eax + 16 = JCOEFPTR block
+; eax + 20 = int last_dc_val
+; eax + 24 = c_derived_tbl *dctbl
+; eax + 28 = c_derived_tbl *actbl
+
+%define pad 6*SIZEOF_DWORD ; Align to 16 bytes
+%define t1 pad
+%define t2 t1+(DCTSIZE2*SIZEOF_WORD)
+%define block t2+(DCTSIZE2*SIZEOF_WORD)
+%define actbl block+SIZEOF_DWORD
+%define buffer actbl+SIZEOF_DWORD
+%define temp buffer+SIZEOF_DWORD
+%define temp2 temp+SIZEOF_DWORD
+%define temp3 temp2+SIZEOF_DWORD
+%define temp4 temp3+SIZEOF_DWORD
+%define temp5 temp4+SIZEOF_DWORD
+%define gotptr temp5+SIZEOF_DWORD ; void *gotptr
+%define put_buffer ebx
+%define put_bits edi
+
+ align 32
+ global EXTN(jsimd_huff_encode_one_block_sse2)
+
+EXTN(jsimd_huff_encode_one_block_sse2):
+ push ebp
+ mov eax,esp ; eax = original ebp
+ sub esp, byte 4
+ and esp, byte (-SIZEOF_XMMWORD) ; align to 128 bits
+ mov [esp],eax
+ mov ebp,esp ; ebp = aligned ebp
+ sub esp, temp5+9*SIZEOF_DWORD-pad
+ push ebx
+ push ecx
+; push edx ; need not be preserved
+ push esi
+ push edi
+ push ebp
+
+ mov esi, POINTER [eax+8] ; (working_state *state)
+ mov put_buffer, DWORD [esi+8] ; put_buffer = state->cur.put_buffer;
+ mov put_bits, DWORD [esi+12] ; put_bits = state->cur.put_bits;
+ push esi ; esi is now scratch
+
+ get_GOT edx ; get GOT address
+ movpic POINTER [esp+gotptr], edx ; save GOT address
+
+ mov ecx, POINTER [eax+28]
+ mov edx, POINTER [eax+16]
+ mov esi, POINTER [eax+12]
+ mov POINTER [esp+actbl], ecx
+ mov POINTER [esp+block], edx
+ mov POINTER [esp+buffer], esi
+
+ ; Encode the DC coefficient difference per section F.1.2.1
+ mov esi, POINTER [esp+block] ; block
+ movsx ecx, word [esi] ; temp = temp2 = block[0] - last_dc_val;
+ sub ecx, DWORD [eax+20]
+ mov esi, ecx
+
+ ; This is a well-known technique for obtaining the absolute value
+ ; with out a branch. It is derived from an assembly language technique
+ ; presented in "How to Optimize for the Pentium Processors",
+ ; Copyright (c) 1996, 1997 by Agner Fog.
+ mov edx, ecx
+ sar edx, 31 ; temp3 = temp >> (CHAR_BIT * sizeof(int) - 1);
+ xor ecx, edx ; temp ^= temp3;
+ sub ecx, edx ; temp -= temp3;
+
+ ; For a negative input, want temp2 = bitwise complement of abs(input)
+ ; This code assumes we are on a two's complement machine
+ add esi, edx ; temp2 += temp3;
+ mov DWORD [esp+temp], esi ; backup temp2 in temp
+
+ ; Find the number of bits needed for the magnitude of the coefficient
+ movpic ebp, POINTER [esp+gotptr] ; load GOT address (ebp)
+ movzx edx, byte [GOTOFF(ebp, jpeg_nbits_table + ecx)] ; nbits = JPEG_NBITS(temp);
+ mov DWORD [esp+temp2], edx ; backup nbits in temp2
+
+ ; Emit the Huffman-coded symbol for the number of bits
+ mov ebp, POINTER [eax+24] ; After this point, arguments are not accessible anymore
+ mov eax, INT [ebp + edx * 4] ; code = dctbl->ehufco[nbits];
+ movzx ecx, byte [ebp + edx + 1024] ; size = dctbl->ehufsi[nbits];
+ EMIT_BITS eax ; EMIT_BITS(code, size)
+
+ mov ecx, DWORD [esp+temp2] ; restore nbits
+
+ ; Mask off any extra bits in code
+ mov eax, 1
+ shl eax, cl
+ dec eax
+ and eax, DWORD [esp+temp] ; temp2 &= (((JLONG) 1)<<nbits) - 1;
+
+ ; Emit that number of bits of the value, if positive,
+ ; or the complement of its magnitude, if negative.
+ EMIT_BITS eax ; EMIT_BITS(temp2, nbits)
+
+ ; Prepare data
+ xor ecx, ecx
+ mov esi, POINTER [esp+block]
+ kloop_prepare 0, 1, 8, 16, 9, 2, 3, 10, 17, 24, 32, 25, \
+ 18, 11, 4, 5, 12, 19, 26, 33, 40, 48, 41, 34, \
+ 27, 20, 13, 6, 7, 14, 21, 28, 35, \
+ xmm0, xmm1, xmm2, xmm3
+ kloop_prepare 32, 42, 49, 56, 57, 50, 43, 36, 29, 22, 15, 23, \
+ 30, 37, 44, 51, 58, 59, 52, 45, 38, 31, 39, 46, \
+ 53, 60, 61, 54, 47, 55, 62, 63, 63, \
+ xmm0, xmm1, xmm2, xmm3
+
+ pxor xmm7, xmm7
+ movdqa xmm0, XMMWORD [esp + t1 + 0 * SIZEOF_WORD] ; __m128i tmp0 = _mm_loadu_si128((__m128i *)(t1 + 0));
+ movdqa xmm1, XMMWORD [esp + t1 + 8 * SIZEOF_WORD] ; __m128i tmp1 = _mm_loadu_si128((__m128i *)(t1 + 8));
+ movdqa xmm2, XMMWORD [esp + t1 + 16 * SIZEOF_WORD] ; __m128i tmp2 = _mm_loadu_si128((__m128i *)(t1 + 16));
+ movdqa xmm3, XMMWORD [esp + t1 + 24 * SIZEOF_WORD] ; __m128i tmp3 = _mm_loadu_si128((__m128i *)(t1 + 24));
+ pcmpeqw xmm0, xmm7 ; tmp0 = _mm_cmpeq_epi16(tmp0, zero);
+ pcmpeqw xmm1, xmm7 ; tmp1 = _mm_cmpeq_epi16(tmp1, zero);
+ pcmpeqw xmm2, xmm7 ; tmp2 = _mm_cmpeq_epi16(tmp2, zero);
+ pcmpeqw xmm3, xmm7 ; tmp3 = _mm_cmpeq_epi16(tmp3, zero);
+ packsswb xmm0, xmm1 ; tmp0 = _mm_packs_epi16(tmp0, tmp1);
+ packsswb xmm2, xmm3 ; tmp2 = _mm_packs_epi16(tmp2, tmp3);
+ pmovmskb edx, xmm0 ; index = ((uint64_t)_mm_movemask_epi8(tmp0)) << 0;
+ pmovmskb ecx, xmm2 ; index = ((uint64_t)_mm_movemask_epi8(tmp2)) << 16;
+ shl ecx, 16
+ or edx, ecx
+ not edx ; index = ~index;
+
+ lea esi, [esp+t1]
+ mov ebp, POINTER [esp+actbl] ; ebp = actbl
+
+.BLOOP:
+ bsf ecx, edx ; r = __builtin_ctzl(index);
+ jz near .ELOOP
+ lea esi, [esi+ecx*2] ; k += r;
+ shr edx, cl ; index >>= r;
+ mov DWORD [esp+temp3], edx
+.BRLOOP:
+ cmp ecx, 16 ; while (r > 15) {
+ jl near .ERLOOP
+ sub ecx, 16 ; r -= 16;
+ mov DWORD [esp+temp], ecx
+ mov eax, INT [ebp + 240 * 4] ; code_0xf0 = actbl->ehufco[0xf0];
+ movzx ecx, byte [ebp + 1024 + 240] ; size_0xf0 = actbl->ehufsi[0xf0];
+ EMIT_BITS eax ; EMIT_BITS(code_0xf0, size_0xf0)
+ mov ecx, DWORD [esp+temp]
+ jmp .BRLOOP
+.ERLOOP:
+ movsx eax, word [esi] ; temp = t1[k];
+ movpic edx, POINTER [esp+gotptr] ; load GOT address (edx)
+ movzx eax, byte [GOTOFF(edx, jpeg_nbits_table + eax)] ; nbits = JPEG_NBITS(temp);
+ mov DWORD [esp+temp2], eax
+ ; Emit Huffman symbol for run length / number of bits
+ shl ecx, 4 ; temp3 = (r << 4) + nbits;
+ add ecx, eax
+ mov eax, INT [ebp + ecx * 4] ; code = actbl->ehufco[temp3];
+ movzx ecx, byte [ebp + ecx + 1024] ; size = actbl->ehufsi[temp3];
+ EMIT_BITS eax
+
+ movsx edx, word [esi+DCTSIZE2*2] ; temp2 = t2[k];
+ ; Mask off any extra bits in code
+ mov ecx, DWORD [esp+temp2]
+ mov eax, 1
+ shl eax, cl
+ dec eax
+ and eax, edx ; temp2 &= (((JLONG) 1)<<nbits) - 1;
+ EMIT_BITS eax ; PUT_BITS(temp2, nbits)
+ mov edx, DWORD [esp+temp3]
+ add esi, 2 ; ++k;
+ shr edx, 1 ; index >>= 1;
+
+ jmp .BLOOP
+.ELOOP:
+ movdqa xmm0, XMMWORD [esp + t1 + 32 * SIZEOF_WORD] ; __m128i tmp0 = _mm_loadu_si128((__m128i *)(t1 + 0));
+ movdqa xmm1, XMMWORD [esp + t1 + 40 * SIZEOF_WORD] ; __m128i tmp1 = _mm_loadu_si128((__m128i *)(t1 + 8));
+ movdqa xmm2, XMMWORD [esp + t1 + 48 * SIZEOF_WORD] ; __m128i tmp2 = _mm_loadu_si128((__m128i *)(t1 + 16));
+ movdqa xmm3, XMMWORD [esp + t1 + 56 * SIZEOF_WORD] ; __m128i tmp3 = _mm_loadu_si128((__m128i *)(t1 + 24));
+ pcmpeqw xmm0, xmm7 ; tmp0 = _mm_cmpeq_epi16(tmp0, zero);
+ pcmpeqw xmm1, xmm7 ; tmp1 = _mm_cmpeq_epi16(tmp1, zero);
+ pcmpeqw xmm2, xmm7 ; tmp2 = _mm_cmpeq_epi16(tmp2, zero);
+ pcmpeqw xmm3, xmm7 ; tmp3 = _mm_cmpeq_epi16(tmp3, zero);
+ packsswb xmm0, xmm1 ; tmp0 = _mm_packs_epi16(tmp0, tmp1);
+ packsswb xmm2, xmm3 ; tmp2 = _mm_packs_epi16(tmp2, tmp3);
+ pmovmskb edx, xmm0 ; index = ((uint64_t)_mm_movemask_epi8(tmp0)) << 0;
+ pmovmskb ecx, xmm2 ; index = ((uint64_t)_mm_movemask_epi8(tmp2)) << 16;
+ shl ecx, 16
+ or edx, ecx
+ not edx ; index = ~index;
+
+ lea eax, [esp + t1 + (DCTSIZE2/2) * 2]
+ sub eax, esi
+ shr eax, 1
+ bsf ecx, edx ; r = __builtin_ctzl(index);
+ jz near .ELOOP2
+ shr edx, cl ; index >>= r;
+ add ecx, eax
+ lea esi, [esi+ecx*2] ; k += r;
+ mov DWORD [esp+temp3], edx
+ jmp .BRLOOP2
+.BLOOP2:
+ bsf ecx, edx ; r = __builtin_ctzl(index);
+ jz near .ELOOP2
+ lea esi, [esi+ecx*2] ; k += r;
+ shr edx, cl ; index >>= r;
+ mov DWORD [esp+temp3], edx
+.BRLOOP2:
+ cmp ecx, 16 ; while (r > 15) {
+ jl near .ERLOOP2
+ sub ecx, 16 ; r -= 16;
+ mov DWORD [esp+temp], ecx
+ mov eax, INT [ebp + 240 * 4] ; code_0xf0 = actbl->ehufco[0xf0];
+ movzx ecx, byte [ebp + 1024 + 240] ; size_0xf0 = actbl->ehufsi[0xf0];
+ EMIT_BITS eax ; EMIT_BITS(code_0xf0, size_0xf0)
+ mov ecx, DWORD [esp+temp]
+ jmp .BRLOOP2
+.ERLOOP2:
+ movsx eax, word [esi] ; temp = t1[k];
+ bsr eax, eax ; nbits = 32 - __builtin_clz(temp);
+ inc eax
+ mov DWORD [esp+temp2], eax
+ ; Emit Huffman symbol for run length / number of bits
+ shl ecx, 4 ; temp3 = (r << 4) + nbits;
+ add ecx, eax
+ mov eax, INT [ebp + ecx * 4] ; code = actbl->ehufco[temp3];
+ movzx ecx, byte [ebp + ecx + 1024] ; size = actbl->ehufsi[temp3];
+ EMIT_BITS eax
+
+ movsx edx, word [esi+DCTSIZE2*2] ; temp2 = t2[k];
+ ; Mask off any extra bits in code
+ mov ecx, DWORD [esp+temp2]
+ mov eax, 1
+ shl eax, cl
+ dec eax
+ and eax, edx ; temp2 &= (((JLONG) 1)<<nbits) - 1;
+ EMIT_BITS eax ; PUT_BITS(temp2, nbits)
+ mov edx, DWORD [esp+temp3]
+ add esi, 2 ; ++k;
+ shr edx, 1 ; index >>= 1;
+
+ jmp .BLOOP2
+.ELOOP2:
+ ; If the last coef(s) were zero, emit an end-of-block code
+ lea edx, [esp + t1 + (DCTSIZE2-1) * 2] ; r = DCTSIZE2-1-k;
+ cmp edx, esi ; if (r > 0) {
+ je .EFN
+ mov eax, INT [ebp] ; code = actbl->ehufco[0];
+ movzx ecx, byte [ebp + 1024] ; size = actbl->ehufsi[0];
+ EMIT_BITS eax
+.EFN:
+ mov eax, [esp+buffer]
+ pop esi
+ ; Save put_buffer & put_bits
+ mov DWORD [esi+8], put_buffer ; state->cur.put_buffer = put_buffer;
+ mov DWORD [esi+12], put_bits ; state->cur.put_bits = put_bits;
+
+ pop ebp
+ pop edi
+ pop esi
+; pop edx ; need not be preserved
+ pop ecx
+ pop ebx
+ mov esp,ebp ; esp <- aligned ebp
+ pop esp ; esp <- original ebp
+ pop ebp
+ ret
+
+; For some reason, the OS X linker does not honor the request to align the
+; segment unless we do this.
+ align 32
diff --git a/simd/i386/jcsample-avx2.asm b/simd/i386/jcsample-avx2.asm
new file mode 100644
index 0000000..9efd415
--- /dev/null
+++ b/simd/i386/jcsample-avx2.asm
@@ -0,0 +1,388 @@
+;
+; jcsample.asm - downsampling (AVX2)
+;
+; Copyright 2009 Pierre Ossman <ossman@cendio.se> for Cendio AB
+; Copyright (C) 2015, Intel Corporation.
+; Copyright (C) 2016, D. R. Commander.
+;
+; Based on the x86 SIMD extension for IJG JPEG library
+; Copyright (C) 1999-2006, MIYASAKA Masaru.
+; For conditions of distribution and use, see copyright notice in jsimdext.inc
+;
+; This file should be assembled with NASM (Netwide Assembler),
+; can *not* be assembled with Microsoft's MASM or any compatible
+; assembler (including Borland's Turbo Assembler).
+; NASM is available from http://nasm.sourceforge.net/ or
+; http://sourceforge.net/project/showfiles.php?group_id=6208
+;
+; [TAB8]
+
+%include "jsimdext.inc"
+
+; --------------------------------------------------------------------------
+ SECTION SEG_TEXT
+ BITS 32
+;
+; Downsample pixel values of a single component.
+; This version handles the common case of 2:1 horizontal and 1:1 vertical,
+; without smoothing.
+;
+; GLOBAL(void)
+; jsimd_h2v1_downsample_avx2 (JDIMENSION image_width, int max_v_samp_factor,
+; JDIMENSION v_samp_factor, JDIMENSION width_blocks,
+; JSAMPARRAY input_data, JSAMPARRAY output_data);
+;
+
+%define img_width(b) (b)+8 ; JDIMENSION image_width
+%define max_v_samp(b) (b)+12 ; int max_v_samp_factor
+%define v_samp(b) (b)+16 ; JDIMENSION v_samp_factor
+%define width_blks(b) (b)+20 ; JDIMENSION width_blocks
+%define input_data(b) (b)+24 ; JSAMPARRAY input_data
+%define output_data(b) (b)+28 ; JSAMPARRAY output_data
+
+ align 32
+ global EXTN(jsimd_h2v1_downsample_avx2)
+
+EXTN(jsimd_h2v1_downsample_avx2):
+ push ebp
+ mov ebp, esp
+; push ebx ; unused
+; push ecx ; need not be preserved
+; push edx ; need not be preserved
+ push esi
+ push edi
+
+ mov ecx, JDIMENSION [width_blks(ebp)]
+ shl ecx, 3 ; imul ecx,DCTSIZE (ecx = output_cols)
+ jz near .return
+
+ mov edx, JDIMENSION [img_width(ebp)]
+
+ ; -- expand_right_edge
+
+ push ecx
+ shl ecx, 1 ; output_cols * 2
+ sub ecx, edx
+ jle short .expand_end
+
+ mov eax, INT [max_v_samp(ebp)]
+ test eax, eax
+ jle short .expand_end
+
+ cld
+ mov esi, JSAMPARRAY [input_data(ebp)] ; input_data
+ alignx 16, 7
+.expandloop:
+ push eax
+ push ecx
+
+ mov edi, JSAMPROW [esi]
+ add edi, edx
+ mov al, JSAMPLE [edi-1]
+
+ rep stosb
+
+ pop ecx
+ pop eax
+
+ add esi, byte SIZEOF_JSAMPROW
+ dec eax
+ jg short .expandloop
+
+.expand_end:
+ pop ecx ; output_cols
+
+ ; -- h2v1_downsample
+
+ mov eax, JDIMENSION [v_samp(ebp)] ; rowctr
+ test eax, eax
+ jle near .return
+
+ mov edx, 0x00010000 ; bias pattern
+ vmovd xmm7, edx
+ vpshufd xmm7, xmm7, 0x00 ; xmm7={0, 1, 0, 1, 0, 1, 0, 1}
+ vperm2i128 ymm7, ymm7, ymm7, 0 ; ymm7={xmm7, xmm7}
+ vpcmpeqw ymm6, ymm6, ymm6
+ vpsrlw ymm6, ymm6, BYTE_BIT ; ymm6={0xFF 0x00 0xFF 0x00 ..}
+
+ mov esi, JSAMPARRAY [input_data(ebp)] ; input_data
+ mov edi, JSAMPARRAY [output_data(ebp)] ; output_data
+ alignx 16, 7
+.rowloop:
+ push ecx
+ push edi
+ push esi
+
+ mov esi, JSAMPROW [esi] ; inptr
+ mov edi, JSAMPROW [edi] ; outptr
+
+ cmp ecx, byte SIZEOF_YMMWORD
+ jae short .columnloop
+ alignx 16, 7
+
+.columnloop_r24:
+ ; ecx can possibly be 8, 16, 24
+ cmp ecx, 24
+ jne .columnloop_r16
+ vmovdqu ymm0, YMMWORD [esi+0*SIZEOF_YMMWORD]
+ vmovdqu xmm1, XMMWORD [esi+1*SIZEOF_YMMWORD]
+ mov ecx, SIZEOF_YMMWORD
+ jmp short .downsample
+
+.columnloop_r16:
+ cmp ecx, 16
+ jne .columnloop_r8
+ vmovdqu ymm0, YMMWORD [esi+0*SIZEOF_YMMWORD]
+ vpxor ymm1, ymm1, ymm1
+ mov ecx, SIZEOF_YMMWORD
+ jmp short .downsample
+
+.columnloop_r8:
+ vmovdqu xmm0, XMMWORD[esi+0*SIZEOF_YMMWORD]
+ vpxor ymm1, ymm1, ymm1
+ mov ecx, SIZEOF_YMMWORD
+ jmp short .downsample
+ alignx 16, 7
+
+.columnloop:
+ vmovdqu ymm0, YMMWORD [esi+0*SIZEOF_YMMWORD]
+ vmovdqu ymm1, YMMWORD [esi+1*SIZEOF_YMMWORD]
+
+.downsample:
+ vpsrlw ymm2, ymm0, BYTE_BIT
+ vpand ymm0, ymm0, ymm6
+ vpsrlw ymm3, ymm1, BYTE_BIT
+ vpand ymm1, ymm1, ymm6
+
+ vpaddw ymm0, ymm0, ymm2
+ vpaddw ymm1, ymm1, ymm3
+ vpaddw ymm0, ymm0, ymm7
+ vpaddw ymm1, ymm1, ymm7
+ vpsrlw ymm0, ymm0, 1
+ vpsrlw ymm1, ymm1, 1
+
+ vpackuswb ymm0, ymm0, ymm1
+ vpermq ymm0, ymm0, 0xd8
+
+ vmovdqu YMMWORD [edi+0*SIZEOF_YMMWORD], ymm0
+
+ sub ecx, byte SIZEOF_YMMWORD ; outcol
+ add esi, byte 2*SIZEOF_YMMWORD ; inptr
+ add edi, byte 1*SIZEOF_YMMWORD ; outptr
+ cmp ecx, byte SIZEOF_YMMWORD
+ jae short .columnloop
+ test ecx, ecx
+ jnz near .columnloop_r24
+
+ pop esi
+ pop edi
+ pop ecx
+
+ add esi, byte SIZEOF_JSAMPROW ; input_data
+ add edi, byte SIZEOF_JSAMPROW ; output_data
+ dec eax ; rowctr
+ jg near .rowloop
+
+.return:
+ vzeroupper
+ pop edi
+ pop esi
+; pop edx ; need not be preserved
+; pop ecx ; need not be preserved
+; pop ebx ; unused
+ pop ebp
+ ret
+
+; --------------------------------------------------------------------------
+;
+; Downsample pixel values of a single component.
+; This version handles the standard case of 2:1 horizontal and 2:1 vertical,
+; without smoothing.
+;
+; GLOBAL(void)
+; jsimd_h2v2_downsample_avx2 (JDIMENSION image_width, int max_v_samp_factor,
+; JDIMENSION v_samp_factor, JDIMENSION width_blocks,
+; JSAMPARRAY input_data, JSAMPARRAY output_data);
+;
+
+%define img_width(b) (b)+8 ; JDIMENSION image_width
+%define max_v_samp(b) (b)+12 ; int max_v_samp_factor
+%define v_samp(b) (b)+16 ; JDIMENSION v_samp_factor
+%define width_blks(b) (b)+20 ; JDIMENSION width_blocks
+%define input_data(b) (b)+24 ; JSAMPARRAY input_data
+%define output_data(b) (b)+28 ; JSAMPARRAY output_data
+
+ align 32
+ global EXTN(jsimd_h2v2_downsample_avx2)
+
+EXTN(jsimd_h2v2_downsample_avx2):
+ push ebp
+ mov ebp, esp
+; push ebx ; unused
+; push ecx ; need not be preserved
+; push edx ; need not be preserved
+ push esi
+ push edi
+
+ mov ecx, JDIMENSION [width_blks(ebp)]
+ shl ecx, 3 ; imul ecx,DCTSIZE (ecx = output_cols)
+ jz near .return
+
+ mov edx, JDIMENSION [img_width(ebp)]
+
+ ; -- expand_right_edge
+
+ push ecx
+ shl ecx, 1 ; output_cols * 2
+ sub ecx, edx
+ jle short .expand_end
+
+ mov eax, INT [max_v_samp(ebp)]
+ test eax, eax
+ jle short .expand_end
+
+ cld
+ mov esi, JSAMPARRAY [input_data(ebp)] ; input_data
+ alignx 16, 7
+.expandloop:
+ push eax
+ push ecx
+
+ mov edi, JSAMPROW [esi]
+ add edi, edx
+ mov al, JSAMPLE [edi-1]
+
+ rep stosb
+
+ pop ecx
+ pop eax
+
+ add esi, byte SIZEOF_JSAMPROW
+ dec eax
+ jg short .expandloop
+
+.expand_end:
+ pop ecx ; output_cols
+
+ ; -- h2v2_downsample
+
+ mov eax, JDIMENSION [v_samp(ebp)] ; rowctr
+ test eax, eax
+ jle near .return
+
+ mov edx, 0x00020001 ; bias pattern
+ vmovd xmm7, edx
+ vpcmpeqw ymm6, ymm6, ymm6
+ vpshufd xmm7, xmm7, 0x00 ; ymm7={1, 2, 1, 2, 1, 2, 1, 2}
+ vperm2i128 ymm7, ymm7, ymm7, 0
+ vpsrlw ymm6, ymm6, BYTE_BIT ; ymm6={0xFF 0x00 0xFF 0x00 ..}
+
+ mov esi, JSAMPARRAY [input_data(ebp)] ; input_data
+ mov edi, JSAMPARRAY [output_data(ebp)] ; output_data
+ alignx 16, 7
+.rowloop:
+ push ecx
+ push edi
+ push esi
+
+ mov edx, JSAMPROW [esi+0*SIZEOF_JSAMPROW] ; inptr0
+ mov esi, JSAMPROW [esi+1*SIZEOF_JSAMPROW] ; inptr1
+ mov edi, JSAMPROW [edi] ; outptr
+
+ cmp ecx, byte SIZEOF_YMMWORD
+ jae short .columnloop
+ alignx 16, 7
+
+.columnloop_r24:
+ cmp ecx, 24
+ jne .columnloop_r16
+ vmovdqu ymm0, YMMWORD [edx+0*SIZEOF_YMMWORD]
+ vmovdqu ymm1, YMMWORD [esi+0*SIZEOF_YMMWORD]
+ vmovdqu xmm2, XMMWORD [edx+1*SIZEOF_YMMWORD]
+ vmovdqu xmm3, XMMWORD [esi+1*SIZEOF_YMMWORD]
+ mov ecx, SIZEOF_YMMWORD
+ jmp short .downsample
+
+.columnloop_r16:
+ cmp ecx, 16
+ jne .columnloop_r8
+ vmovdqu ymm0, YMMWORD [edx+0*SIZEOF_YMMWORD]
+ vmovdqu ymm1, YMMWORD [esi+0*SIZEOF_YMMWORD]
+ vpxor ymm2, ymm2, ymm2
+ vpxor ymm3, ymm3, ymm3
+ mov ecx, SIZEOF_YMMWORD
+ jmp short .downsample
+
+.columnloop_r8:
+ vmovdqu xmm0, XMMWORD [edx+0*SIZEOF_XMMWORD]
+ vmovdqu xmm1, XMMWORD [esi+0*SIZEOF_XMMWORD]
+ vpxor ymm2, ymm2, ymm2
+ vpxor ymm3, ymm3, ymm3
+ mov ecx, SIZEOF_YMMWORD
+ jmp short .downsample
+ alignx 16, 7
+
+.columnloop:
+ vmovdqu ymm0, YMMWORD [edx+0*SIZEOF_YMMWORD]
+ vmovdqu ymm1, YMMWORD [esi+0*SIZEOF_YMMWORD]
+ vmovdqu ymm2, YMMWORD [edx+1*SIZEOF_YMMWORD]
+ vmovdqu ymm3, YMMWORD [esi+1*SIZEOF_YMMWORD]
+
+.downsample:
+ vpand ymm4, ymm0, ymm6
+ vpsrlw ymm0, ymm0, BYTE_BIT
+ vpand ymm5, ymm1, ymm6
+ vpsrlw ymm1, ymm1, BYTE_BIT
+ vpaddw ymm0, ymm0, ymm4
+ vpaddw ymm1, ymm1, ymm5
+
+ vpand ymm4, ymm2, ymm6
+ vpsrlw ymm2, ymm2, BYTE_BIT
+ vpand ymm5, ymm3, ymm6
+ vpsrlw ymm3, ymm3, BYTE_BIT
+ vpaddw ymm2, ymm2, ymm4
+ vpaddw ymm3, ymm3, ymm5
+
+ vpaddw ymm0, ymm0, ymm1
+ vpaddw ymm2, ymm2, ymm3
+ vpaddw ymm0, ymm0, ymm7
+ vpaddw ymm2, ymm2, ymm7
+ vpsrlw ymm0, ymm0, 2
+ vpsrlw ymm2, ymm2, 2
+
+ vpackuswb ymm0, ymm0, ymm2
+ vpermq ymm0, ymm0, 0xd8
+
+ vmovdqu YMMWORD [edi+0*SIZEOF_YMMWORD], ymm0
+
+ sub ecx, byte SIZEOF_YMMWORD ; outcol
+ add edx, byte 2*SIZEOF_YMMWORD ; inptr0
+ add esi, byte 2*SIZEOF_YMMWORD ; inptr1
+ add edi, byte 1*SIZEOF_YMMWORD ; outptr
+ cmp ecx, byte SIZEOF_YMMWORD
+ jae near .columnloop
+ test ecx, ecx
+ jnz near .columnloop_r24
+
+ pop esi
+ pop edi
+ pop ecx
+
+ add esi, byte 2*SIZEOF_JSAMPROW ; input_data
+ add edi, byte 1*SIZEOF_JSAMPROW ; output_data
+ dec eax ; rowctr
+ jg near .rowloop
+
+.return:
+ vzeroupper
+ pop edi
+ pop esi
+; pop edx ; need not be preserved
+; pop ecx ; need not be preserved
+; pop ebx ; unused
+ pop ebp
+ ret
+
+; For some reason, the OS X linker does not honor the request to align the
+; segment unless we do this.
+ align 32
diff --git a/simd/jcsample-mmx.asm b/simd/i386/jcsample-mmx.asm
similarity index 98%
rename from simd/jcsample-mmx.asm
rename to simd/i386/jcsample-mmx.asm
index 6cd544e..2fa6a1a 100644
--- a/simd/jcsample-mmx.asm
+++ b/simd/i386/jcsample-mmx.asm
@@ -2,6 +2,7 @@
; jcsample.asm - downsampling (MMX)
;
; Copyright 2009 Pierre Ossman <ossman@cendio.se> for Cendio AB
+; Copyright (C) 2016, D. R. Commander.
;
; Based on the x86 SIMD extension for IJG JPEG library
; Copyright (C) 1999-2006, MIYASAKA Masaru.
@@ -38,7 +39,7 @@
%define input_data(b) (b)+24 ; JSAMPARRAY input_data
%define output_data(b) (b)+28 ; JSAMPARRAY output_data
- align 16
+ align 32
global EXTN(jsimd_h2v1_downsample_mmx)
EXTN(jsimd_h2v1_downsample_mmx):
@@ -180,7 +181,7 @@
%define input_data(b) (b)+24 ; JSAMPARRAY input_data
%define output_data(b) (b)+28 ; JSAMPARRAY output_data
- align 16
+ align 32
global EXTN(jsimd_h2v2_downsample_mmx)
EXTN(jsimd_h2v2_downsample_mmx):
@@ -320,4 +321,4 @@
; For some reason, the OS X linker does not honor the request to align the
; segment unless we do this.
- align 16
+ align 32
diff --git a/simd/i386/jcsample-sse2.asm b/simd/i386/jcsample-sse2.asm
new file mode 100644
index 0000000..3d34ab9
--- /dev/null
+++ b/simd/i386/jcsample-sse2.asm
@@ -0,0 +1,351 @@
+;
+; jcsample.asm - downsampling (SSE2)
+;
+; Copyright 2009 Pierre Ossman <ossman@cendio.se> for Cendio AB
+; Copyright (C) 2016, D. R. Commander.
+;
+; Based on the x86 SIMD extension for IJG JPEG library
+; Copyright (C) 1999-2006, MIYASAKA Masaru.
+; For conditions of distribution and use, see copyright notice in jsimdext.inc
+;
+; This file should be assembled with NASM (Netwide Assembler),
+; can *not* be assembled with Microsoft's MASM or any compatible
+; assembler (including Borland's Turbo Assembler).
+; NASM is available from http://nasm.sourceforge.net/ or
+; http://sourceforge.net/project/showfiles.php?group_id=6208
+;
+; [TAB8]
+
+%include "jsimdext.inc"
+
+; --------------------------------------------------------------------------
+ SECTION SEG_TEXT
+ BITS 32
+;
+; Downsample pixel values of a single component.
+; This version handles the common case of 2:1 horizontal and 1:1 vertical,
+; without smoothing.
+;
+; GLOBAL(void)
+; jsimd_h2v1_downsample_sse2 (JDIMENSION image_width, int max_v_samp_factor,
+; JDIMENSION v_samp_factor, JDIMENSION width_blocks,
+; JSAMPARRAY input_data, JSAMPARRAY output_data);
+;
+
+%define img_width(b) (b)+8 ; JDIMENSION image_width
+%define max_v_samp(b) (b)+12 ; int max_v_samp_factor
+%define v_samp(b) (b)+16 ; JDIMENSION v_samp_factor
+%define width_blks(b) (b)+20 ; JDIMENSION width_blocks
+%define input_data(b) (b)+24 ; JSAMPARRAY input_data
+%define output_data(b) (b)+28 ; JSAMPARRAY output_data
+
+ align 32
+ global EXTN(jsimd_h2v1_downsample_sse2)
+
+EXTN(jsimd_h2v1_downsample_sse2):
+ push ebp
+ mov ebp, esp
+; push ebx ; unused
+; push ecx ; need not be preserved
+; push edx ; need not be preserved
+ push esi
+ push edi
+
+ mov ecx, JDIMENSION [width_blks(ebp)]
+ shl ecx, 3 ; imul ecx,DCTSIZE (ecx = output_cols)
+ jz near .return
+
+ mov edx, JDIMENSION [img_width(ebp)]
+
+ ; -- expand_right_edge
+
+ push ecx
+ shl ecx, 1 ; output_cols * 2
+ sub ecx, edx
+ jle short .expand_end
+
+ mov eax, INT [max_v_samp(ebp)]
+ test eax, eax
+ jle short .expand_end
+
+ cld
+ mov esi, JSAMPARRAY [input_data(ebp)] ; input_data
+ alignx 16, 7
+.expandloop:
+ push eax
+ push ecx
+
+ mov edi, JSAMPROW [esi]
+ add edi, edx
+ mov al, JSAMPLE [edi-1]
+
+ rep stosb
+
+ pop ecx
+ pop eax
+
+ add esi, byte SIZEOF_JSAMPROW
+ dec eax
+ jg short .expandloop
+
+.expand_end:
+ pop ecx ; output_cols
+
+ ; -- h2v1_downsample
+
+ mov eax, JDIMENSION [v_samp(ebp)] ; rowctr
+ test eax, eax
+ jle near .return
+
+ mov edx, 0x00010000 ; bias pattern
+ movd xmm7, edx
+ pcmpeqw xmm6, xmm6
+ pshufd xmm7, xmm7, 0x00 ; xmm7={0, 1, 0, 1, 0, 1, 0, 1}
+ psrlw xmm6, BYTE_BIT ; xmm6={0xFF 0x00 0xFF 0x00 ..}
+
+ mov esi, JSAMPARRAY [input_data(ebp)] ; input_data
+ mov edi, JSAMPARRAY [output_data(ebp)] ; output_data
+ alignx 16, 7
+.rowloop:
+ push ecx
+ push edi
+ push esi
+
+ mov esi, JSAMPROW [esi] ; inptr
+ mov edi, JSAMPROW [edi] ; outptr
+
+ cmp ecx, byte SIZEOF_XMMWORD
+ jae short .columnloop
+ alignx 16, 7
+
+.columnloop_r8:
+ movdqa xmm0, XMMWORD [esi+0*SIZEOF_XMMWORD]
+ pxor xmm1, xmm1
+ mov ecx, SIZEOF_XMMWORD
+ jmp short .downsample
+ alignx 16, 7
+
+.columnloop:
+ movdqa xmm0, XMMWORD [esi+0*SIZEOF_XMMWORD]
+ movdqa xmm1, XMMWORD [esi+1*SIZEOF_XMMWORD]
+
+.downsample:
+ movdqa xmm2, xmm0
+ movdqa xmm3, xmm1
+
+ pand xmm0, xmm6
+ psrlw xmm2, BYTE_BIT
+ pand xmm1, xmm6
+ psrlw xmm3, BYTE_BIT
+
+ paddw xmm0, xmm2
+ paddw xmm1, xmm3
+ paddw xmm0, xmm7
+ paddw xmm1, xmm7
+ psrlw xmm0, 1
+ psrlw xmm1, 1
+
+ packuswb xmm0, xmm1
+
+ movdqa XMMWORD [edi+0*SIZEOF_XMMWORD], xmm0
+
+ sub ecx, byte SIZEOF_XMMWORD ; outcol
+ add esi, byte 2*SIZEOF_XMMWORD ; inptr
+ add edi, byte 1*SIZEOF_XMMWORD ; outptr
+ cmp ecx, byte SIZEOF_XMMWORD
+ jae short .columnloop
+ test ecx, ecx
+ jnz short .columnloop_r8
+
+ pop esi
+ pop edi
+ pop ecx
+
+ add esi, byte SIZEOF_JSAMPROW ; input_data
+ add edi, byte SIZEOF_JSAMPROW ; output_data
+ dec eax ; rowctr
+ jg near .rowloop
+
+.return:
+ pop edi
+ pop esi
+; pop edx ; need not be preserved
+; pop ecx ; need not be preserved
+; pop ebx ; unused
+ pop ebp
+ ret
+
+; --------------------------------------------------------------------------
+;
+; Downsample pixel values of a single component.
+; This version handles the standard case of 2:1 horizontal and 2:1 vertical,
+; without smoothing.
+;
+; GLOBAL(void)
+; jsimd_h2v2_downsample_sse2 (JDIMENSION image_width, int max_v_samp_factor,
+; JDIMENSION v_samp_factor, JDIMENSION width_blocks,
+; JSAMPARRAY input_data, JSAMPARRAY output_data);
+;
+
+%define img_width(b) (b)+8 ; JDIMENSION image_width
+%define max_v_samp(b) (b)+12 ; int max_v_samp_factor
+%define v_samp(b) (b)+16 ; JDIMENSION v_samp_factor
+%define width_blks(b) (b)+20 ; JDIMENSION width_blocks
+%define input_data(b) (b)+24 ; JSAMPARRAY input_data
+%define output_data(b) (b)+28 ; JSAMPARRAY output_data
+
+ align 32
+ global EXTN(jsimd_h2v2_downsample_sse2)
+
+EXTN(jsimd_h2v2_downsample_sse2):
+ push ebp
+ mov ebp, esp
+; push ebx ; unused
+; push ecx ; need not be preserved
+; push edx ; need not be preserved
+ push esi
+ push edi
+
+ mov ecx, JDIMENSION [width_blks(ebp)]
+ shl ecx, 3 ; imul ecx,DCTSIZE (ecx = output_cols)
+ jz near .return
+
+ mov edx, JDIMENSION [img_width(ebp)]
+
+ ; -- expand_right_edge
+
+ push ecx
+ shl ecx, 1 ; output_cols * 2
+ sub ecx, edx
+ jle short .expand_end
+
+ mov eax, INT [max_v_samp(ebp)]
+ test eax, eax
+ jle short .expand_end
+
+ cld
+ mov esi, JSAMPARRAY [input_data(ebp)] ; input_data
+ alignx 16, 7
+.expandloop:
+ push eax
+ push ecx
+
+ mov edi, JSAMPROW [esi]
+ add edi, edx
+ mov al, JSAMPLE [edi-1]
+
+ rep stosb
+
+ pop ecx
+ pop eax
+
+ add esi, byte SIZEOF_JSAMPROW
+ dec eax
+ jg short .expandloop
+
+.expand_end:
+ pop ecx ; output_cols
+
+ ; -- h2v2_downsample
+
+ mov eax, JDIMENSION [v_samp(ebp)] ; rowctr
+ test eax, eax
+ jle near .return
+
+ mov edx, 0x00020001 ; bias pattern
+ movd xmm7, edx
+ pcmpeqw xmm6, xmm6
+ pshufd xmm7, xmm7, 0x00 ; xmm7={1, 2, 1, 2, 1, 2, 1, 2}
+ psrlw xmm6, BYTE_BIT ; xmm6={0xFF 0x00 0xFF 0x00 ..}
+
+ mov esi, JSAMPARRAY [input_data(ebp)] ; input_data
+ mov edi, JSAMPARRAY [output_data(ebp)] ; output_data
+ alignx 16, 7
+.rowloop:
+ push ecx
+ push edi
+ push esi
+
+ mov edx, JSAMPROW [esi+0*SIZEOF_JSAMPROW] ; inptr0
+ mov esi, JSAMPROW [esi+1*SIZEOF_JSAMPROW] ; inptr1
+ mov edi, JSAMPROW [edi] ; outptr
+
+ cmp ecx, byte SIZEOF_XMMWORD
+ jae short .columnloop
+ alignx 16, 7
+
+.columnloop_r8:
+ movdqa xmm0, XMMWORD [edx+0*SIZEOF_XMMWORD]
+ movdqa xmm1, XMMWORD [esi+0*SIZEOF_XMMWORD]
+ pxor xmm2, xmm2
+ pxor xmm3, xmm3
+ mov ecx, SIZEOF_XMMWORD
+ jmp short .downsample
+ alignx 16, 7
+
+.columnloop:
+ movdqa xmm0, XMMWORD [edx+0*SIZEOF_XMMWORD]
+ movdqa xmm1, XMMWORD [esi+0*SIZEOF_XMMWORD]
+ movdqa xmm2, XMMWORD [edx+1*SIZEOF_XMMWORD]
+ movdqa xmm3, XMMWORD [esi+1*SIZEOF_XMMWORD]
+
+.downsample:
+ movdqa xmm4, xmm0
+ movdqa xmm5, xmm1
+ pand xmm0, xmm6
+ psrlw xmm4, BYTE_BIT
+ pand xmm1, xmm6
+ psrlw xmm5, BYTE_BIT
+ paddw xmm0, xmm4
+ paddw xmm1, xmm5
+
+ movdqa xmm4, xmm2
+ movdqa xmm5, xmm3
+ pand xmm2, xmm6
+ psrlw xmm4, BYTE_BIT
+ pand xmm3, xmm6
+ psrlw xmm5, BYTE_BIT
+ paddw xmm2, xmm4
+ paddw xmm3, xmm5
+
+ paddw xmm0, xmm1
+ paddw xmm2, xmm3
+ paddw xmm0, xmm7
+ paddw xmm2, xmm7
+ psrlw xmm0, 2
+ psrlw xmm2, 2
+
+ packuswb xmm0, xmm2
+
+ movdqa XMMWORD [edi+0*SIZEOF_XMMWORD], xmm0
+
+ sub ecx, byte SIZEOF_XMMWORD ; outcol
+ add edx, byte 2*SIZEOF_XMMWORD ; inptr0
+ add esi, byte 2*SIZEOF_XMMWORD ; inptr1
+ add edi, byte 1*SIZEOF_XMMWORD ; outptr
+ cmp ecx, byte SIZEOF_XMMWORD
+ jae near .columnloop
+ test ecx, ecx
+ jnz near .columnloop_r8
+
+ pop esi
+ pop edi
+ pop ecx
+
+ add esi, byte 2*SIZEOF_JSAMPROW ; input_data
+ add edi, byte 1*SIZEOF_JSAMPROW ; output_data
+ dec eax ; rowctr
+ jg near .rowloop
+
+.return:
+ pop edi
+ pop esi
+; pop edx ; need not be preserved
+; pop ecx ; need not be preserved
+; pop ebx ; unused
+ pop ebp
+ ret
+
+; For some reason, the OS X linker does not honor the request to align the
+; segment unless we do this.
+ align 32
diff --git a/simd/i386/jdcolext-avx2.asm b/simd/i386/jdcolext-avx2.asm
new file mode 100644
index 0000000..1222b77
--- /dev/null
+++ b/simd/i386/jdcolext-avx2.asm
@@ -0,0 +1,516 @@
+;
+; jdcolext.asm - colorspace conversion (AVX2)
+;
+; Copyright 2009, 2012 Pierre Ossman <ossman@cendio.se> for Cendio AB
+; Copyright (C) 2012, 2016, D. R. Commander.
+; Copyright (C) 2015, Intel Corporation.
+;
+; Based on the x86 SIMD extension for IJG JPEG library
+; Copyright (C) 1999-2006, MIYASAKA Masaru.
+; For conditions of distribution and use, see copyright notice in jsimdext.inc
+;
+; This file should be assembled with NASM (Netwide Assembler),
+; can *not* be assembled with Microsoft's MASM or any compatible
+; assembler (including Borland's Turbo Assembler).
+; NASM is available from http://nasm.sourceforge.net/ or
+; http://sourceforge.net/project/showfiles.php?group_id=6208
+;
+; [TAB8]
+
+%include "jcolsamp.inc"
+
+; --------------------------------------------------------------------------
+;
+; Convert some rows of samples to the output colorspace.
+;
+; GLOBAL(void)
+; jsimd_ycc_rgb_convert_avx2 (JDIMENSION out_width,
+; JSAMPIMAGE input_buf, JDIMENSION input_row,
+; JSAMPARRAY output_buf, int num_rows)
+;
+
+%define out_width(b) (b)+8 ; JDIMENSION out_width
+%define input_buf(b) (b)+12 ; JSAMPIMAGE input_buf
+%define input_row(b) (b)+16 ; JDIMENSION input_row
+%define output_buf(b) (b)+20 ; JSAMPARRAY output_buf
+%define num_rows(b) (b)+24 ; int num_rows
+
+%define original_ebp ebp+0
+%define wk(i) ebp-(WK_NUM-(i))*SIZEOF_YMMWORD ; ymmword wk[WK_NUM]
+%define WK_NUM 2
+%define gotptr wk(0)-SIZEOF_POINTER ; void * gotptr
+
+ align 32
+ global EXTN(jsimd_ycc_rgb_convert_avx2)
+
+EXTN(jsimd_ycc_rgb_convert_avx2):
+ push ebp
+ mov eax, esp ; eax = original ebp
+ sub esp, byte 4
+ and esp, byte (-SIZEOF_YMMWORD) ; align to 256 bits
+ mov [esp], eax
+ mov ebp, esp ; ebp = aligned ebp
+ lea esp, [wk(0)]
+ pushpic eax ; make a room for GOT address
+ push ebx
+; push ecx ; need not be preserved
+; push edx ; need not be preserved
+ push esi
+ push edi
+
+ get_GOT ebx ; get GOT address
+ movpic POINTER [gotptr], ebx ; save GOT address
+
+ mov ecx, JDIMENSION [out_width(eax)] ; num_cols
+ test ecx, ecx
+ jz near .return
+
+ push ecx
+
+ mov edi, JSAMPIMAGE [input_buf(eax)]
+ mov ecx, JDIMENSION [input_row(eax)]
+ mov esi, JSAMPARRAY [edi+0*SIZEOF_JSAMPARRAY]
+ mov ebx, JSAMPARRAY [edi+1*SIZEOF_JSAMPARRAY]
+ mov edx, JSAMPARRAY [edi+2*SIZEOF_JSAMPARRAY]
+ lea esi, [esi+ecx*SIZEOF_JSAMPROW]
+ lea ebx, [ebx+ecx*SIZEOF_JSAMPROW]
+ lea edx, [edx+ecx*SIZEOF_JSAMPROW]
+
+ pop ecx
+
+ mov edi, JSAMPARRAY [output_buf(eax)]
+ mov eax, INT [num_rows(eax)]
+ test eax, eax
+ jle near .return
+ alignx 16, 7
+.rowloop:
+ push eax
+ push edi
+ push edx
+ push ebx
+ push esi
+ push ecx ; col
+
+ mov esi, JSAMPROW [esi] ; inptr0
+ mov ebx, JSAMPROW [ebx] ; inptr1
+ mov edx, JSAMPROW [edx] ; inptr2
+ mov edi, JSAMPROW [edi] ; outptr
+ movpic eax, POINTER [gotptr] ; load GOT address (eax)
+ alignx 16, 7
+.columnloop:
+
+ vmovdqu ymm5, YMMWORD [ebx] ; ymm5=Cb(0123456789ABCDEFGHIJKLMNOPQRSTUV)
+ vmovdqu ymm1, YMMWORD [edx] ; ymm1=Cr(0123456789ABCDEFGHIJKLMNOPQRSTUV)
+
+ vpcmpeqw ymm0, ymm0, ymm0
+ vpcmpeqw ymm7, ymm7, ymm7
+ vpsrlw ymm0, ymm0, BYTE_BIT ; ymm0={0xFF 0x00 0xFF 0x00 ..}
+ vpsllw ymm7, ymm7, 7 ; ymm7={0xFF80 0xFF80 0xFF80 0xFF80 ..}
+
+ vpand ymm4, ymm0, ymm5 ; ymm4=Cb(02468ACEGIKMOQSU)=CbE
+ vpsrlw ymm5, ymm5, BYTE_BIT ; ymm5=Cb(13579BDFHJLNPRTV)=CbO
+ vpand ymm0, ymm0, ymm1 ; ymm0=Cr(02468ACEGIKMOQSU)=CrE
+ vpsrlw ymm1, ymm1, BYTE_BIT ; ymm1=Cr(13579BDFHJLNPRTV)=CrO
+
+ vpaddw ymm2, ymm4, ymm7
+ vpaddw ymm3, ymm5, ymm7
+ vpaddw ymm6, ymm0, ymm7
+ vpaddw ymm7, ymm1, ymm7
+
+ ; (Original)
+ ; R = Y + 1.40200 * Cr
+ ; G = Y - 0.34414 * Cb - 0.71414 * Cr
+ ; B = Y + 1.77200 * Cb
+ ;
+ ; (This implementation)
+ ; R = Y + 0.40200 * Cr + Cr
+ ; G = Y - 0.34414 * Cb + 0.28586 * Cr - Cr
+ ; B = Y - 0.22800 * Cb + Cb + Cb
+
+ vpaddw ymm4, ymm2, ymm2 ; ymm4=2*CbE
+ vpaddw ymm5, ymm3, ymm3 ; ymm5=2*CbO
+ vpaddw ymm0, ymm6, ymm6 ; ymm0=2*CrE
+ vpaddw ymm1, ymm7, ymm7 ; ymm1=2*CrO
+
+ vpmulhw ymm4, ymm4, [GOTOFF(eax,PW_MF0228)] ; ymm4=(2*CbE * -FIX(0.22800))
+ vpmulhw ymm5, ymm5, [GOTOFF(eax,PW_MF0228)] ; ymm5=(2*CbO * -FIX(0.22800))
+ vpmulhw ymm0, ymm0, [GOTOFF(eax,PW_F0402)] ; ymm0=(2*CrE * FIX(0.40200))
+ vpmulhw ymm1, ymm1, [GOTOFF(eax,PW_F0402)] ; ymm1=(2*CrO * FIX(0.40200))
+
+ vpaddw ymm4, ymm4, [GOTOFF(eax,PW_ONE)]
+ vpaddw ymm5, ymm5, [GOTOFF(eax,PW_ONE)]
+ vpsraw ymm4, ymm4, 1 ; ymm4=(CbE * -FIX(0.22800))
+ vpsraw ymm5, ymm5, 1 ; ymm5=(CbO * -FIX(0.22800))
+ vpaddw ymm0, ymm0, [GOTOFF(eax,PW_ONE)]
+ vpaddw ymm1, ymm1, [GOTOFF(eax,PW_ONE)]
+ vpsraw ymm0, ymm0, 1 ; ymm0=(CrE * FIX(0.40200))
+ vpsraw ymm1, ymm1, 1 ; ymm1=(CrO * FIX(0.40200))
+
+ vpaddw ymm4, ymm4, ymm2
+ vpaddw ymm5, ymm5, ymm3
+ vpaddw ymm4, ymm4, ymm2 ; ymm4=(CbE * FIX(1.77200))=(B-Y)E
+ vpaddw ymm5, ymm5, ymm3 ; ymm5=(CbO * FIX(1.77200))=(B-Y)O
+ vpaddw ymm0, ymm0, ymm6 ; ymm0=(CrE * FIX(1.40200))=(R-Y)E
+ vpaddw ymm1, ymm1, ymm7 ; ymm1=(CrO * FIX(1.40200))=(R-Y)O
+
+ vmovdqa YMMWORD [wk(0)], ymm4 ; wk(0)=(B-Y)E
+ vmovdqa YMMWORD [wk(1)], ymm5 ; wk(1)=(B-Y)O
+
+ vpunpckhwd ymm4, ymm2, ymm6
+ vpunpcklwd ymm2, ymm2, ymm6
+ vpmaddwd ymm2, ymm2, [GOTOFF(eax,PW_MF0344_F0285)]
+ vpmaddwd ymm4, ymm4, [GOTOFF(eax,PW_MF0344_F0285)]
+ vpunpckhwd ymm5, ymm3, ymm7
+ vpunpcklwd ymm3, ymm3, ymm7
+ vpmaddwd ymm3, ymm3, [GOTOFF(eax,PW_MF0344_F0285)]
+ vpmaddwd ymm5, ymm5, [GOTOFF(eax,PW_MF0344_F0285)]
+
+ vpaddd ymm2, ymm2, [GOTOFF(eax,PD_ONEHALF)]
+ vpaddd ymm4, ymm4, [GOTOFF(eax,PD_ONEHALF)]
+ vpsrad ymm2, ymm2, SCALEBITS
+ vpsrad ymm4, ymm4, SCALEBITS
+ vpaddd ymm3, ymm3, [GOTOFF(eax,PD_ONEHALF)]
+ vpaddd ymm5, ymm5, [GOTOFF(eax,PD_ONEHALF)]
+ vpsrad ymm3, ymm3, SCALEBITS
+ vpsrad ymm5, ymm5, SCALEBITS
+
+ vpackssdw ymm2, ymm2, ymm4 ; ymm2=CbE*-FIX(0.344)+CrE*FIX(0.285)
+ vpackssdw ymm3, ymm3, ymm5 ; ymm3=CbO*-FIX(0.344)+CrO*FIX(0.285)
+ vpsubw ymm2, ymm2, ymm6 ; ymm2=CbE*-FIX(0.344)+CrE*-FIX(0.714)=(G-Y)E
+ vpsubw ymm3, ymm3, ymm7 ; ymm3=CbO*-FIX(0.344)+CrO*-FIX(0.714)=(G-Y)O
+
+ vmovdqu ymm5, YMMWORD [esi] ; ymm5=Y(0123456789ABCDEFGHIJKLMNOPQRSTUV)
+
+ vpcmpeqw ymm4, ymm4, ymm4
+ vpsrlw ymm4, ymm4, BYTE_BIT ; ymm4={0xFF 0x00 0xFF 0x00 ..}
+ vpand ymm4, ymm4, ymm5 ; ymm4=Y(02468ACEGIKMOQSU)=YE
+ vpsrlw ymm5, ymm5, BYTE_BIT ; ymm5=Y(13579BDFHJLNPRTV)=YO
+
+ vpaddw ymm0, ymm0, ymm4 ; ymm0=((R-Y)E+YE)=RE=R(02468ACEGIKMOQSU)
+ vpaddw ymm1, ymm1, ymm5 ; ymm1=((R-Y)O+YO)=RO=R(13579BDFHJLNPRTV)
+ vpackuswb ymm0, ymm0, ymm0 ; ymm0=R(02468ACE********GIKMOQSU********)
+ vpackuswb ymm1, ymm1, ymm1 ; ymm1=R(13579BDF********HJLNPRTV********)
+
+ vpaddw ymm2, ymm2, ymm4 ; ymm2=((G-Y)E+YE)=GE=G(02468ACEGIKMOQSU)
+ vpaddw ymm3, ymm3, ymm5 ; ymm3=((G-Y)O+YO)=GO=G(13579BDFHJLNPRTV)
+ vpackuswb ymm2, ymm2, ymm2 ; ymm2=G(02468ACE********GIKMOQSU********)
+ vpackuswb ymm3, ymm3, ymm3 ; ymm3=G(13579BDF********HJLNPRTV********)
+
+ vpaddw ymm4, ymm4, YMMWORD [wk(0)] ; ymm4=(YE+(B-Y)E)=BE=B(02468ACEGIKMOQSU)
+ vpaddw ymm5, ymm5, YMMWORD [wk(1)] ; ymm5=(YO+(B-Y)O)=BO=B(13579BDFHJLNPRTV)
+ vpackuswb ymm4, ymm4, ymm4 ; ymm4=B(02468ACE********GIKMOQSU********)
+ vpackuswb ymm5, ymm5, ymm5 ; ymm5=B(13579BDF********HJLNPRTV********)
+
+%if RGB_PIXELSIZE == 3 ; ---------------
+
+ ; ymmA=(00 02 04 06 08 0A 0C 0E ** 0G 0I 0K 0M 0O 0Q 0S 0U **)
+ ; ymmB=(01 03 05 07 09 0B 0D 0F ** 0H 0J 0L 0N 0P 0R 0T 0V **)
+ ; ymmC=(10 12 14 16 18 1A 1C 1E ** 1G 1I 1K 1M 1O 1Q 1S 1U **)
+ ; ymmD=(11 13 15 17 19 1B 1D 1F ** 1H 1J 1L 1N 1P 1R 1T 1V **)
+ ; ymmE=(20 22 24 26 28 2A 2C 2E ** 2G 2I 2K 2M 2O 2Q 2S 2U **)
+ ; ymmF=(21 23 25 27 29 2B 2D 2F ** 2H 2J 2L 2N 2P 2R 2T 2V **)
+ ; ymmG=(** ** ** ** ** ** ** ** ** ** ** ** ** ** ** ** ** **)
+ ; ymmH=(** ** ** ** ** ** ** ** ** ** ** ** ** ** ** ** ** **)
+
+ vpunpcklbw ymmA, ymmA, ymmC ; ymmA=(00 10 02 12 04 14 06 16 08 18 0A 1A 0C 1C 0E 1E
+ ; 0G 1G 0I 1I 0K 1K 0M 1M 0O 1O 0Q 1Q 0S 1S 0U 1U)
+ vpunpcklbw ymmE, ymmE, ymmB ; ymmE=(20 01 22 03 24 05 26 07 28 09 2A 0B 2C 0D 2E 0F
+ ; 2G 0H 2I 0J 2K 0L 2M 0N 2O 0P 2Q 0R 2S 0T 2U 0V)
+ vpunpcklbw ymmD, ymmD, ymmF ; ymmD=(11 21 13 23 15 25 17 27 19 29 1B 2B 1D 2D 1F 2F
+ ; 1H 2H 1J 2J 1L 2L 1N 2N 1P 2P 1R 2R 1T 2T 1V 2V)
+
+ vpsrldq ymmH, ymmA, 2 ; ymmH=(02 12 04 14 06 16 08 18 0A 1A 0C 1C 0E 1E 0G 1G
+ ; 0I 1I 0K 1K 0M 1M 0O 1O 0Q 1Q 0S 1S 0U 1U -- --)
+ vpunpckhwd ymmG, ymmA, ymmE ; ymmG=(08 18 28 09 0A 1A 2A 0B 0C 1C 2C 0D 0E 1E 2E 0F
+ ; 0O 1O 2O 0P 0Q 1Q 2Q 0R 0S 1S 2S 0T 0U 1U 2U 0V)
+ vpunpcklwd ymmA, ymmA, ymmE ; ymmA=(00 10 20 01 02 12 22 03 04 14 24 05 06 16 26 07
+ ; 0G 1G 2G 0H 0I 1I 2I 0J 0K 1K 2K 0L 0M 1M 2M 0N)
+
+ vpsrldq ymmE, ymmE, 2 ; ymmE=(22 03 24 05 26 07 28 09 2A 0B 2C 0D 2E 0F 2G 0H
+ ; 2I 0J 2K 0L 2M 0N 2O 0P 2Q 0R 2S 0T 2U 0V -- --)
+
+ vpsrldq ymmB, ymmD, 2 ; ymmB=(13 23 15 25 17 27 19 29 1B 2B 1D 2D 1F 2F 1H 2H
+ ; 1J 2J 1L 2L 1N 2N 1P 2P 1R 2R 1T 2T 1V 2V -- --)
+ vpunpckhwd ymmC, ymmD, ymmH ; ymmC=(19 29 0A 1A 1B 2B 0C 1C 1D 2D 0E 1E 1F 2F 0G 1G
+ ; 1P 2P 0Q 1Q 1R 2R 0S 1S 1T 2T 0U 1U 1V 2V -- --)
+ vpunpcklwd ymmD, ymmD, ymmH ; ymmD=(11 21 02 12 13 23 04 14 15 25 06 16 17 27 08 18
+ ; 1H 2H 0I 1I 1J 2J 0K 1K 1L 2L 0M 1M 1N 2N 0O 1O)
+
+ vpunpckhwd ymmF, ymmE, ymmB ; ymmF=(2A 0B 1B 2B 2C 0D 1D 2D 2E 0F 1F 2F 2G 0H 1H 2H
+ ; 2Q 0R 1R 2R 2S 0T 1T 2T 2U 0V 1V 2V -- -- -- --)
+ vpunpcklwd ymmE, ymmE, ymmB ; ymmE=(22 03 13 23 24 05 15 25 26 07 17 27 28 09 19 29
+ ; 2I 0J 1J 2J 2K 0L 1L 2L 2M 0N 1N 2N 2O 0P 1P 2P)
+
+ vpshufd ymmH, ymmA, 0x4E ; ymmH=(04 14 24 05 06 16 26 07 00 10 20 01 02 12 22 03
+ ; 0K 1K 2K 0L 0M 1M 2M 0N 0G 1G 2G 0H 0I 1I 2I 0J)
+ vpunpckldq ymmA, ymmA, ymmD ; ymmA=(00 10 20 01 11 21 02 12 02 12 22 03 13 23 04 14
+ ; 0G 1G 2G 0H 1H 2H 0I 1I 0I 1I 2I 0J 1J 2J 0K 1K)
+ vpunpckhdq ymmD, ymmD, ymmE ; ymmD=(15 25 06 16 26 07 17 27 17 27 08 18 28 09 19 29
+ ; 1L 2L 0M 1M 2M 0N 1N 2N 1N 2N 0O 1O 2O 0P 1P 2P)
+ vpunpckldq ymmE, ymmE, ymmH ; ymmE=(22 03 13 23 04 14 24 05 24 05 15 25 06 16 26 07
+ ; 2I 0J 1J 2J 0K 1K 2K 0L 2K 0L 1L 2L 0M 1M 2M 0N)
+
+ vpshufd ymmH, ymmG, 0x4E ; ymmH=(0C 1C 2C 0D 0E 1E 2E 0F 08 18 28 09 0A 1A 2A 0B
+ ; 0S 1S 2S 0T 0U 1U 2U 0V 0O 1O 2O 0P 0Q 1Q 2Q 0R)
+ vpunpckldq ymmG, ymmG, ymmC ; ymmG=(08 18 28 09 19 29 0A 1A 0A 1A 2A 0B 1B 2B 0C 1C
+ ; 0O 1O 2O 0P 1P 2P 0Q 1Q 0Q 1Q 2Q 0R 1R 2R 0S 1S)
+ vpunpckhdq ymmC, ymmC, ymmF ; ymmC=(1D 2D 0E 1E 2E 0F 1F 2F 1F 2F 0G 1G 2G 0H 1H 2H
+ ; 1T 2T 0U 1U 2U 0V 1V 2V 1V 2V -- -- -- -- -- --)
+ vpunpckldq ymmF, ymmF, ymmH ; ymmF=(2A 0B 1B 2B 0C 1C 2C 0D 2C 0D 1D 2D 0E 1E 2E 0F
+ ; 2Q 0R 1R 2R 0S 1S 2S 0T 2S 0T 1T 2T 0U 1U 2U 0V)
+
+ vpunpcklqdq ymmH, ymmA, ymmE ; ymmH=(00 10 20 01 11 21 02 12 22 03 13 23 04 14 24 05
+ ; 0G 1G 2G 0H 1H 2H 0I 1I 2I 0J 1J 2J 0K 1K 2K 0L)
+ vpunpcklqdq ymmG, ymmD, ymmG ; ymmG=(15 25 06 16 26 07 17 27 08 18 28 09 19 29 0A 1A
+ ; 1L 2L 0M 1M 2M 0N 1N 2N 0O 1O 2O 0P 1P 2P 0Q 1Q)
+ vpunpcklqdq ymmC, ymmF, ymmC ; ymmC=(2A 0B 1B 2B 0C 1C 2C 0D 1D 2D 0E 1E 2E 0F 1F 2F
+ ; 2Q 0R 1R 2R 0S 1S 2S 0T 1T 2T 0U 1U 2U 0V 1V 2V)
+
+ vperm2i128 ymmA, ymmH, ymmG, 0x20 ; ymmA=(00 10 20 01 11 21 02 12 22 03 13 23 04 14 24 05
+ ; 15 25 06 16 26 07 17 27 08 18 28 09 19 29 0A 1A)
+ vperm2i128 ymmD, ymmC, ymmH, 0x30 ; ymmD=(2A 0B 1B 2B 0C 1C 2C 0D 1D 2D 0E 1E 2E 0F 1F 2F
+ ; 0G 1G 2G 0H 1H 2H 0I 1I 2I 0J 1J 2J 0K 1K 2K 0L)
+ vperm2i128 ymmF, ymmG, ymmC, 0x31 ; ymmF=(1L 2L 0M 1M 2M 0N 1N 2N 0O 1O 2O 0P 1P 2P 0Q 1Q
+ ; 2Q 0R 1R 2R 0S 1S 2S 0T 1T 2T 0U 1U 2U 0V 1V 2V)
+
+ cmp ecx, byte SIZEOF_YMMWORD
+ jb short .column_st64
+
+ test edi, SIZEOF_YMMWORD-1
+ jnz short .out1
+ ; --(aligned)-------------------
+ vmovntdq YMMWORD [edi+0*SIZEOF_YMMWORD], ymmA
+ vmovntdq YMMWORD [edi+1*SIZEOF_YMMWORD], ymmD
+ vmovntdq YMMWORD [edi+2*SIZEOF_YMMWORD], ymmF
+ jmp short .out0
+.out1: ; --(unaligned)-----------------
+ vmovdqu YMMWORD [edi+0*SIZEOF_YMMWORD], ymmA
+ vmovdqu YMMWORD [edi+1*SIZEOF_YMMWORD], ymmD
+ vmovdqu YMMWORD [edi+2*SIZEOF_YMMWORD], ymmF
+.out0:
+ add edi, byte RGB_PIXELSIZE*SIZEOF_YMMWORD ; outptr
+ sub ecx, byte SIZEOF_YMMWORD
+ jz near .nextrow
+
+ add esi, byte SIZEOF_YMMWORD ; inptr0
+ add ebx, byte SIZEOF_YMMWORD ; inptr1
+ add edx, byte SIZEOF_YMMWORD ; inptr2
+ jmp near .columnloop
+ alignx 16, 7
+
+.column_st64:
+ lea ecx, [ecx+ecx*2] ; imul ecx, RGB_PIXELSIZE
+ cmp ecx, byte 2*SIZEOF_YMMWORD
+ jb short .column_st32
+ vmovdqu YMMWORD [edi+0*SIZEOF_YMMWORD], ymmA
+ vmovdqu YMMWORD [edi+1*SIZEOF_YMMWORD], ymmD
+ add edi, byte 2*SIZEOF_YMMWORD ; outptr
+ vmovdqa ymmA, ymmF
+ sub ecx, byte 2*SIZEOF_YMMWORD
+ jmp short .column_st31
+.column_st32:
+ cmp ecx, byte SIZEOF_YMMWORD
+ jb short .column_st31
+ vmovdqu YMMWORD [edi+0*SIZEOF_YMMWORD], ymmA
+ add edi, byte SIZEOF_YMMWORD ; outptr
+ vmovdqa ymmA, ymmD
+ sub ecx, byte SIZEOF_YMMWORD
+ jmp short .column_st31
+.column_st31:
+ cmp ecx, byte SIZEOF_XMMWORD
+ jb short .column_st15
+ vmovdqu XMMWORD [edi+0*SIZEOF_XMMWORD], xmmA
+ add edi, byte SIZEOF_XMMWORD ; outptr
+ vperm2i128 ymmA, ymmA, ymmA, 1
+ sub ecx, byte SIZEOF_XMMWORD
+.column_st15:
+ ; Store the lower 8 bytes of xmmA to the output when it has enough
+ ; space.
+ cmp ecx, byte SIZEOF_MMWORD
+ jb short .column_st7
+ vmovq XMM_MMWORD [edi], xmmA
+ add edi, byte SIZEOF_MMWORD
+ sub ecx, byte SIZEOF_MMWORD
+ vpsrldq xmmA, xmmA, SIZEOF_MMWORD
+.column_st7:
+ ; Store the lower 4 bytes of xmmA to the output when it has enough
+ ; space.
+ cmp ecx, byte SIZEOF_DWORD
+ jb short .column_st3
+ vmovd XMM_DWORD [edi], xmmA
+ add edi, byte SIZEOF_DWORD
+ sub ecx, byte SIZEOF_DWORD
+ vpsrldq xmmA, xmmA, SIZEOF_DWORD
+.column_st3:
+ ; Store the lower 2 bytes of eax to the output when it has enough
+ ; space.
+ vmovd eax, xmmA
+ cmp ecx, byte SIZEOF_WORD
+ jb short .column_st1
+ mov WORD [edi], ax
+ add edi, byte SIZEOF_WORD
+ sub ecx, byte SIZEOF_WORD
+ shr eax, 16
+.column_st1:
+ ; Store the lower 1 byte of eax to the output when it has enough
+ ; space.
+ test ecx, ecx
+ jz short .nextrow
+ mov BYTE [edi], al
+
+%else ; RGB_PIXELSIZE == 4 ; -----------
+
+%ifdef RGBX_FILLER_0XFF
+ vpcmpeqb ymm6, ymm6, ymm6 ; ymm6=XE=X(02468ACE********GIKMOQSU********)
+ vpcmpeqb ymm7, ymm7, ymm7 ; ymm7=XO=X(13579BDF********HJLNPRTV********)
+%else
+ vpxor ymm6, ymm6, ymm6 ; ymm6=XE=X(02468ACE********GIKMOQSU********)
+ vpxor ymm7, ymm7, ymm7 ; ymm7=XO=X(13579BDF********HJLNPRTV********)
+%endif
+ ; ymmA=(00 02 04 06 08 0A 0C 0E ** 0G 0I 0K 0M 0O 0Q 0S 0U **)
+ ; ymmB=(01 03 05 07 09 0B 0D 0F ** 0H 0J 0L 0N 0P 0R 0T 0V **)
+ ; ymmC=(10 12 14 16 18 1A 1C 1E ** 1G 1I 1K 1M 1O 1Q 1S 1U **)
+ ; ymmD=(11 13 15 17 19 1B 1D 1F ** 1H 1J 1L 1N 1P 1R 1T 1V **)
+ ; ymmE=(20 22 24 26 28 2A 2C 2E ** 2G 2I 2K 2M 2O 2Q 2S 2U **)
+ ; ymmF=(21 23 25 27 29 2B 2D 2F ** 2H 2J 2L 2N 2P 2R 2T 2V **)
+ ; ymmG=(30 32 34 36 38 3A 3C 3E ** 3G 3I 3K 3M 3O 3Q 3S 3U **)
+ ; ymmH=(31 33 35 37 39 3B 3D 3F ** 3H 3J 3L 3N 3P 3R 3T 3V **)
+
+ vpunpcklbw ymmA, ymmA, ymmC ; ymmA=(00 10 02 12 04 14 06 16 08 18 0A 1A 0C 1C 0E 1E
+ ; 0G 1G 0I 1I 0K 1K 0M 1M 0O 1O 0Q 1Q 0S 1S 0U 1U)
+ vpunpcklbw ymmE, ymmE, ymmG ; ymmE=(20 30 22 32 24 34 26 36 28 38 2A 3A 2C 3C 2E 3E
+ ; 2G 3G 2I 3I 2K 3K 2M 3M 2O 3O 2Q 3Q 2S 3S 2U 3U)
+ vpunpcklbw ymmB, ymmB, ymmD ; ymmB=(01 11 03 13 05 15 07 17 09 19 0B 1B 0D 1D 0F 1F
+ ; 0H 1H 0J 1J 0L 1L 0N 1N 0P 1P 0R 1R 0T 1T 0V 1V)
+ vpunpcklbw ymmF, ymmF, ymmH ; ymmF=(21 31 23 33 25 35 27 37 29 39 2B 3B 2D 3D 2F 3F
+ ; 2H 3H 2J 3J 2L 3L 2N 3N 2P 3P 2R 3R 2T 3T 2V 3V)
+
+ vpunpckhwd ymmC, ymmA, ymmE ; ymmC=(08 18 28 38 0A 1A 2A 3A 0C 1C 2C 3C 0E 1E 2E 3E
+ ; 0O 1O 2O 3O 0Q 1Q 2Q 3Q 0S 1S 2S 3S 0U 1U 2U 3U)
+ vpunpcklwd ymmA, ymmA, ymmE ; ymmA=(00 10 20 30 02 12 22 32 04 14 24 34 06 16 26 36
+ ; 0G 1G 2G 3G 0I 1I 2I 3I 0K 1K 2K 3K 0M 1M 2M 3M)
+ vpunpckhwd ymmG, ymmB, ymmF ; ymmG=(09 19 29 39 0B 1B 2B 3B 0D 1D 2D 3D 0F 1F 2F 3F
+ ; 0P 1P 2P 3P 0R 1R 2R 3R 0T 1T 2T 3T 0V 1V 2V 3V)
+ vpunpcklwd ymmB, ymmB, ymmF ; ymmB=(01 11 21 31 03 13 23 33 05 15 25 35 07 17 27 37
+ ; 0H 1H 2H 3H 0J 1J 2J 3J 0L 1L 2L 3L 0N 1N 2N 3N)
+
+ vpunpckhdq ymmE, ymmA, ymmB ; ymmE=(04 14 24 34 05 15 25 35 06 16 26 36 07 17 27 37
+ ; 0K 1K 2K 3K 0L 1L 2L 3L 0M 1M 2M 3M 0N 1N 2N 3N)
+ vpunpckldq ymmB, ymmA, ymmB ; ymmB=(00 10 20 30 01 11 21 31 02 12 22 32 03 13 23 33
+ ; 0G 1G 2G 3G 0H 1H 2H 3H 0I 1I 2I 3I 0J 1J 2J 3J)
+ vpunpckhdq ymmF, ymmC, ymmG ; ymmF=(0C 1C 2C 3C 0D 1D 2D 3D 0E 1E 2E 3E 0F 1F 2F 3F
+ ; 0S 1S 2S 3S 0T 1T 2T 3T 0U 1U 2U 3U 0V 1V 2V 3V)
+ vpunpckldq ymmG, ymmC, ymmG ; ymmG=(08 18 28 38 09 19 29 39 0A 1A 2A 3A 0B 1B 2B 3B
+ ; 0O 1O 2O 3O 0P 1P 2P 3P 0Q 1Q 2Q 3Q 0R 1R 2R 3R)
+
+ vperm2i128 ymmA, ymmB, ymmE, 0x20 ; ymmA=(00 10 20 30 01 11 21 31 02 12 22 32 03 13 23 33
+ ; 04 14 24 34 05 15 25 35 06 16 26 36 07 17 27 37)
+ vperm2i128 ymmD, ymmG, ymmF, 0x20 ; ymmD=(08 18 28 38 09 19 29 39 0A 1A 2A 3A 0B 1B 2B 3B
+ ; 0C 1C 2C 3C 0D 1D 2D 3D 0E 1E 2E 3E 0F 1F 2F 3F)
+ vperm2i128 ymmC, ymmB, ymmE, 0x31 ; ymmC=(0G 1G 2G 3G 0H 1H 2H 3H 0I 1I 2I 3I 0J 1J 2J 3J
+ ; 0K 1K 2K 3K 0L 1L 2L 3L 0M 1M 2M 3M 0N 1N 2N 3N)
+ vperm2i128 ymmH, ymmG, ymmF, 0x31 ; ymmH=(0O 1O 2O 3O 0P 1P 2P 3P 0Q 1Q 2Q 3Q 0R 1R 2R 3R
+ ; 0S 1S 2S 3S 0T 1T 2T 3T 0U 1U 2U 3U 0V 1V 2V 3V)
+
+ cmp ecx, byte SIZEOF_YMMWORD
+ jb short .column_st64
+
+ test edi, SIZEOF_YMMWORD-1
+ jnz short .out1
+ ; --(aligned)-------------------
+ vmovntdq YMMWORD [edi+0*SIZEOF_YMMWORD], ymmA
+ vmovntdq YMMWORD [edi+1*SIZEOF_YMMWORD], ymmD
+ vmovntdq YMMWORD [edi+2*SIZEOF_YMMWORD], ymmC
+ vmovntdq YMMWORD [edi+3*SIZEOF_YMMWORD], ymmH
+ jmp short .out0
+.out1: ; --(unaligned)-----------------
+ vmovdqu YMMWORD [edi+0*SIZEOF_YMMWORD], ymmA
+ vmovdqu YMMWORD [edi+1*SIZEOF_YMMWORD], ymmD
+ vmovdqu YMMWORD [edi+2*SIZEOF_YMMWORD], ymmC
+ vmovdqu YMMWORD [edi+3*SIZEOF_YMMWORD], ymmH
+.out0:
+ add edi, RGB_PIXELSIZE*SIZEOF_YMMWORD ; outptr
+ sub ecx, byte SIZEOF_YMMWORD
+ jz near .nextrow
+
+ add esi, byte SIZEOF_YMMWORD ; inptr0
+ add ebx, byte SIZEOF_YMMWORD ; inptr1
+ add edx, byte SIZEOF_YMMWORD ; inptr2
+ jmp near .columnloop
+ alignx 16, 7
+
+.column_st64:
+ cmp ecx, byte SIZEOF_YMMWORD/2
+ jb short .column_st32
+ vmovdqu YMMWORD [edi+0*SIZEOF_YMMWORD], ymmA
+ vmovdqu YMMWORD [edi+1*SIZEOF_YMMWORD], ymmD
+ add edi, byte 2*SIZEOF_YMMWORD ; outptr
+ vmovdqa ymmA, ymmC
+ vmovdqa ymmD, ymmH
+ sub ecx, byte SIZEOF_YMMWORD/2
+.column_st32:
+ cmp ecx, byte SIZEOF_YMMWORD/4
+ jb short .column_st16
+ vmovdqu YMMWORD [edi+0*SIZEOF_YMMWORD], ymmA
+ add edi, byte SIZEOF_YMMWORD ; outptr
+ vmovdqa ymmA, ymmD
+ sub ecx, byte SIZEOF_YMMWORD/4
+.column_st16:
+ cmp ecx, byte SIZEOF_YMMWORD/8
+ jb short .column_st15
+ vmovdqu XMMWORD [edi+0*SIZEOF_XMMWORD], xmmA
+ vperm2i128 ymmA, ymmA, ymmA, 1
+ add edi, byte SIZEOF_XMMWORD ; outptr
+ sub ecx, byte SIZEOF_YMMWORD/8
+.column_st15:
+ ; Store two pixels (8 bytes) of ymmA to the output when it has enough
+ ; space.
+ cmp ecx, byte SIZEOF_YMMWORD/16
+ jb short .column_st7
+ vmovq MMWORD [edi], xmmA
+ add edi, byte SIZEOF_YMMWORD/16*4
+ sub ecx, byte SIZEOF_YMMWORD/16
+ vpsrldq xmmA, SIZEOF_YMMWORD/16*4
+.column_st7:
+ ; Store one pixel (4 bytes) of ymmA to the output when it has enough
+ ; space.
+ test ecx, ecx
+ jz short .nextrow
+ vmovd XMM_DWORD [edi], xmmA
+
+%endif ; RGB_PIXELSIZE ; ---------------
+
+ alignx 16, 7
+
+.nextrow:
+ pop ecx
+ pop esi
+ pop ebx
+ pop edx
+ pop edi
+ pop eax
+
+ add esi, byte SIZEOF_JSAMPROW
+ add ebx, byte SIZEOF_JSAMPROW
+ add edx, byte SIZEOF_JSAMPROW
+ add edi, byte SIZEOF_JSAMPROW ; output_buf
+ dec eax ; num_rows
+ jg near .rowloop
+
+ sfence ; flush the write buffer
+
+.return:
+ vzeroupper
+ pop edi
+ pop esi
+; pop edx ; need not be preserved
+; pop ecx ; need not be preserved
+ pop ebx
+ mov esp, ebp ; esp <- aligned ebp
+ pop esp ; esp <- original ebp
+ pop ebp
+ ret
+
+; For some reason, the OS X linker does not honor the request to align the
+; segment unless we do this.
+ align 32
diff --git a/simd/jdcolext-mmx.asm b/simd/i386/jdcolext-mmx.asm
similarity index 99%
rename from simd/jdcolext-mmx.asm
rename to simd/i386/jdcolext-mmx.asm
index 21e34f6..1f0614e 100644
--- a/simd/jdcolext-mmx.asm
+++ b/simd/i386/jdcolext-mmx.asm
@@ -2,6 +2,7 @@
; jdcolext.asm - colorspace conversion (MMX)
;
; Copyright 2009 Pierre Ossman <ossman@cendio.se> for Cendio AB
+; Copyright (C) 2016, D. R. Commander.
;
; Based on the x86 SIMD extension for IJG JPEG library
; Copyright (C) 1999-2006, MIYASAKA Masaru.
@@ -38,7 +39,7 @@
%define WK_NUM 2
%define gotptr wk(0)-SIZEOF_POINTER ; void * gotptr
- align 16
+ align 32
global EXTN(jsimd_ycc_rgb_convert_mmx)
EXTN(jsimd_ycc_rgb_convert_mmx):
@@ -401,4 +402,4 @@
; For some reason, the OS X linker does not honor the request to align the
; segment unless we do this.
- align 16
+ align 32
diff --git a/simd/i386/jdcolext-sse2.asm b/simd/i386/jdcolext-sse2.asm
new file mode 100644
index 0000000..69462ef
--- /dev/null
+++ b/simd/i386/jdcolext-sse2.asm
@@ -0,0 +1,459 @@
+;
+; jdcolext.asm - colorspace conversion (SSE2)
+;
+; Copyright 2009, 2012 Pierre Ossman <ossman@cendio.se> for Cendio AB
+; Copyright (C) 2012, 2016, D. R. Commander.
+;
+; Based on the x86 SIMD extension for IJG JPEG library
+; Copyright (C) 1999-2006, MIYASAKA Masaru.
+; For conditions of distribution and use, see copyright notice in jsimdext.inc
+;
+; This file should be assembled with NASM (Netwide Assembler),
+; can *not* be assembled with Microsoft's MASM or any compatible
+; assembler (including Borland's Turbo Assembler).
+; NASM is available from http://nasm.sourceforge.net/ or
+; http://sourceforge.net/project/showfiles.php?group_id=6208
+;
+; [TAB8]
+
+%include "jcolsamp.inc"
+
+; --------------------------------------------------------------------------
+;
+; Convert some rows of samples to the output colorspace.
+;
+; GLOBAL(void)
+; jsimd_ycc_rgb_convert_sse2 (JDIMENSION out_width,
+; JSAMPIMAGE input_buf, JDIMENSION input_row,
+; JSAMPARRAY output_buf, int num_rows)
+;
+
+%define out_width(b) (b)+8 ; JDIMENSION out_width
+%define input_buf(b) (b)+12 ; JSAMPIMAGE input_buf
+%define input_row(b) (b)+16 ; JDIMENSION input_row
+%define output_buf(b) (b)+20 ; JSAMPARRAY output_buf
+%define num_rows(b) (b)+24 ; int num_rows
+
+%define original_ebp ebp+0
+%define wk(i) ebp-(WK_NUM-(i))*SIZEOF_XMMWORD ; xmmword wk[WK_NUM]
+%define WK_NUM 2
+%define gotptr wk(0)-SIZEOF_POINTER ; void * gotptr
+
+ align 32
+ global EXTN(jsimd_ycc_rgb_convert_sse2)
+
+EXTN(jsimd_ycc_rgb_convert_sse2):
+ push ebp
+ mov eax, esp ; eax = original ebp
+ sub esp, byte 4
+ and esp, byte (-SIZEOF_XMMWORD) ; align to 128 bits
+ mov [esp], eax
+ mov ebp, esp ; ebp = aligned ebp
+ lea esp, [wk(0)]
+ pushpic eax ; make a room for GOT address
+ push ebx
+; push ecx ; need not be preserved
+; push edx ; need not be preserved
+ push esi
+ push edi
+
+ get_GOT ebx ; get GOT address
+ movpic POINTER [gotptr], ebx ; save GOT address
+
+ mov ecx, JDIMENSION [out_width(eax)] ; num_cols
+ test ecx, ecx
+ jz near .return
+
+ push ecx
+
+ mov edi, JSAMPIMAGE [input_buf(eax)]
+ mov ecx, JDIMENSION [input_row(eax)]
+ mov esi, JSAMPARRAY [edi+0*SIZEOF_JSAMPARRAY]
+ mov ebx, JSAMPARRAY [edi+1*SIZEOF_JSAMPARRAY]
+ mov edx, JSAMPARRAY [edi+2*SIZEOF_JSAMPARRAY]
+ lea esi, [esi+ecx*SIZEOF_JSAMPROW]
+ lea ebx, [ebx+ecx*SIZEOF_JSAMPROW]
+ lea edx, [edx+ecx*SIZEOF_JSAMPROW]
+
+ pop ecx
+
+ mov edi, JSAMPARRAY [output_buf(eax)]
+ mov eax, INT [num_rows(eax)]
+ test eax, eax
+ jle near .return
+ alignx 16, 7
+.rowloop:
+ push eax
+ push edi
+ push edx
+ push ebx
+ push esi
+ push ecx ; col
+
+ mov esi, JSAMPROW [esi] ; inptr0
+ mov ebx, JSAMPROW [ebx] ; inptr1
+ mov edx, JSAMPROW [edx] ; inptr2
+ mov edi, JSAMPROW [edi] ; outptr
+ movpic eax, POINTER [gotptr] ; load GOT address (eax)
+ alignx 16, 7
+.columnloop:
+
+ movdqa xmm5, XMMWORD [ebx] ; xmm5=Cb(0123456789ABCDEF)
+ movdqa xmm1, XMMWORD [edx] ; xmm1=Cr(0123456789ABCDEF)
+
+ pcmpeqw xmm4, xmm4
+ pcmpeqw xmm7, xmm7
+ psrlw xmm4, BYTE_BIT
+ psllw xmm7, 7 ; xmm7={0xFF80 0xFF80 0xFF80 0xFF80 ..}
+ movdqa xmm0, xmm4 ; xmm0=xmm4={0xFF 0x00 0xFF 0x00 ..}
+
+ pand xmm4, xmm5 ; xmm4=Cb(02468ACE)=CbE
+ psrlw xmm5, BYTE_BIT ; xmm5=Cb(13579BDF)=CbO
+ pand xmm0, xmm1 ; xmm0=Cr(02468ACE)=CrE
+ psrlw xmm1, BYTE_BIT ; xmm1=Cr(13579BDF)=CrO
+
+ paddw xmm4, xmm7
+ paddw xmm5, xmm7
+ paddw xmm0, xmm7
+ paddw xmm1, xmm7
+
+ ; (Original)
+ ; R = Y + 1.40200 * Cr
+ ; G = Y - 0.34414 * Cb - 0.71414 * Cr
+ ; B = Y + 1.77200 * Cb
+ ;
+ ; (This implementation)
+ ; R = Y + 0.40200 * Cr + Cr
+ ; G = Y - 0.34414 * Cb + 0.28586 * Cr - Cr
+ ; B = Y - 0.22800 * Cb + Cb + Cb
+
+ movdqa xmm2, xmm4 ; xmm2=CbE
+ movdqa xmm3, xmm5 ; xmm3=CbO
+ paddw xmm4, xmm4 ; xmm4=2*CbE
+ paddw xmm5, xmm5 ; xmm5=2*CbO
+ movdqa xmm6, xmm0 ; xmm6=CrE
+ movdqa xmm7, xmm1 ; xmm7=CrO
+ paddw xmm0, xmm0 ; xmm0=2*CrE
+ paddw xmm1, xmm1 ; xmm1=2*CrO
+
+ pmulhw xmm4, [GOTOFF(eax,PW_MF0228)] ; xmm4=(2*CbE * -FIX(0.22800))
+ pmulhw xmm5, [GOTOFF(eax,PW_MF0228)] ; xmm5=(2*CbO * -FIX(0.22800))
+ pmulhw xmm0, [GOTOFF(eax,PW_F0402)] ; xmm0=(2*CrE * FIX(0.40200))
+ pmulhw xmm1, [GOTOFF(eax,PW_F0402)] ; xmm1=(2*CrO * FIX(0.40200))
+
+ paddw xmm4, [GOTOFF(eax,PW_ONE)]
+ paddw xmm5, [GOTOFF(eax,PW_ONE)]
+ psraw xmm4, 1 ; xmm4=(CbE * -FIX(0.22800))
+ psraw xmm5, 1 ; xmm5=(CbO * -FIX(0.22800))
+ paddw xmm0, [GOTOFF(eax,PW_ONE)]
+ paddw xmm1, [GOTOFF(eax,PW_ONE)]
+ psraw xmm0, 1 ; xmm0=(CrE * FIX(0.40200))
+ psraw xmm1, 1 ; xmm1=(CrO * FIX(0.40200))
+
+ paddw xmm4, xmm2
+ paddw xmm5, xmm3
+ paddw xmm4, xmm2 ; xmm4=(CbE * FIX(1.77200))=(B-Y)E
+ paddw xmm5, xmm3 ; xmm5=(CbO * FIX(1.77200))=(B-Y)O
+ paddw xmm0, xmm6 ; xmm0=(CrE * FIX(1.40200))=(R-Y)E
+ paddw xmm1, xmm7 ; xmm1=(CrO * FIX(1.40200))=(R-Y)O
+
+ movdqa XMMWORD [wk(0)], xmm4 ; wk(0)=(B-Y)E
+ movdqa XMMWORD [wk(1)], xmm5 ; wk(1)=(B-Y)O
+
+ movdqa xmm4, xmm2
+ movdqa xmm5, xmm3
+ punpcklwd xmm2, xmm6
+ punpckhwd xmm4, xmm6
+ pmaddwd xmm2, [GOTOFF(eax,PW_MF0344_F0285)]
+ pmaddwd xmm4, [GOTOFF(eax,PW_MF0344_F0285)]
+ punpcklwd xmm3, xmm7
+ punpckhwd xmm5, xmm7
+ pmaddwd xmm3, [GOTOFF(eax,PW_MF0344_F0285)]
+ pmaddwd xmm5, [GOTOFF(eax,PW_MF0344_F0285)]
+
+ paddd xmm2, [GOTOFF(eax,PD_ONEHALF)]
+ paddd xmm4, [GOTOFF(eax,PD_ONEHALF)]
+ psrad xmm2, SCALEBITS
+ psrad xmm4, SCALEBITS
+ paddd xmm3, [GOTOFF(eax,PD_ONEHALF)]
+ paddd xmm5, [GOTOFF(eax,PD_ONEHALF)]
+ psrad xmm3, SCALEBITS
+ psrad xmm5, SCALEBITS
+
+ packssdw xmm2, xmm4 ; xmm2=CbE*-FIX(0.344)+CrE*FIX(0.285)
+ packssdw xmm3, xmm5 ; xmm3=CbO*-FIX(0.344)+CrO*FIX(0.285)
+ psubw xmm2, xmm6 ; xmm2=CbE*-FIX(0.344)+CrE*-FIX(0.714)=(G-Y)E
+ psubw xmm3, xmm7 ; xmm3=CbO*-FIX(0.344)+CrO*-FIX(0.714)=(G-Y)O
+
+ movdqa xmm5, XMMWORD [esi] ; xmm5=Y(0123456789ABCDEF)
+
+ pcmpeqw xmm4, xmm4
+ psrlw xmm4, BYTE_BIT ; xmm4={0xFF 0x00 0xFF 0x00 ..}
+ pand xmm4, xmm5 ; xmm4=Y(02468ACE)=YE
+ psrlw xmm5, BYTE_BIT ; xmm5=Y(13579BDF)=YO
+
+ paddw xmm0, xmm4 ; xmm0=((R-Y)E+YE)=RE=R(02468ACE)
+ paddw xmm1, xmm5 ; xmm1=((R-Y)O+YO)=RO=R(13579BDF)
+ packuswb xmm0, xmm0 ; xmm0=R(02468ACE********)
+ packuswb xmm1, xmm1 ; xmm1=R(13579BDF********)
+
+ paddw xmm2, xmm4 ; xmm2=((G-Y)E+YE)=GE=G(02468ACE)
+ paddw xmm3, xmm5 ; xmm3=((G-Y)O+YO)=GO=G(13579BDF)
+ packuswb xmm2, xmm2 ; xmm2=G(02468ACE********)
+ packuswb xmm3, xmm3 ; xmm3=G(13579BDF********)
+
+ paddw xmm4, XMMWORD [wk(0)] ; xmm4=(YE+(B-Y)E)=BE=B(02468ACE)
+ paddw xmm5, XMMWORD [wk(1)] ; xmm5=(YO+(B-Y)O)=BO=B(13579BDF)
+ packuswb xmm4, xmm4 ; xmm4=B(02468ACE********)
+ packuswb xmm5, xmm5 ; xmm5=B(13579BDF********)
+
+%if RGB_PIXELSIZE == 3 ; ---------------
+
+ ; xmmA=(00 02 04 06 08 0A 0C 0E **), xmmB=(01 03 05 07 09 0B 0D 0F **)
+ ; xmmC=(10 12 14 16 18 1A 1C 1E **), xmmD=(11 13 15 17 19 1B 1D 1F **)
+ ; xmmE=(20 22 24 26 28 2A 2C 2E **), xmmF=(21 23 25 27 29 2B 2D 2F **)
+ ; xmmG=(** ** ** ** ** ** ** ** **), xmmH=(** ** ** ** ** ** ** ** **)
+
+ punpcklbw xmmA, xmmC ; xmmA=(00 10 02 12 04 14 06 16 08 18 0A 1A 0C 1C 0E 1E)
+ punpcklbw xmmE, xmmB ; xmmE=(20 01 22 03 24 05 26 07 28 09 2A 0B 2C 0D 2E 0F)
+ punpcklbw xmmD, xmmF ; xmmD=(11 21 13 23 15 25 17 27 19 29 1B 2B 1D 2D 1F 2F)
+
+ movdqa xmmG, xmmA
+ movdqa xmmH, xmmA
+ punpcklwd xmmA, xmmE ; xmmA=(00 10 20 01 02 12 22 03 04 14 24 05 06 16 26 07)
+ punpckhwd xmmG, xmmE ; xmmG=(08 18 28 09 0A 1A 2A 0B 0C 1C 2C 0D 0E 1E 2E 0F)
+
+ psrldq xmmH, 2 ; xmmH=(02 12 04 14 06 16 08 18 0A 1A 0C 1C 0E 1E -- --)
+ psrldq xmmE, 2 ; xmmE=(22 03 24 05 26 07 28 09 2A 0B 2C 0D 2E 0F -- --)
+
+ movdqa xmmC, xmmD
+ movdqa xmmB, xmmD
+ punpcklwd xmmD, xmmH ; xmmD=(11 21 02 12 13 23 04 14 15 25 06 16 17 27 08 18)
+ punpckhwd xmmC, xmmH ; xmmC=(19 29 0A 1A 1B 2B 0C 1C 1D 2D 0E 1E 1F 2F -- --)
+
+ psrldq xmmB, 2 ; xmmB=(13 23 15 25 17 27 19 29 1B 2B 1D 2D 1F 2F -- --)
+
+ movdqa xmmF, xmmE
+ punpcklwd xmmE, xmmB ; xmmE=(22 03 13 23 24 05 15 25 26 07 17 27 28 09 19 29)
+ punpckhwd xmmF, xmmB ; xmmF=(2A 0B 1B 2B 2C 0D 1D 2D 2E 0F 1F 2F -- -- -- --)
+
+ pshufd xmmH, xmmA, 0x4E ; xmmH=(04 14 24 05 06 16 26 07 00 10 20 01 02 12 22 03)
+ movdqa xmmB, xmmE
+ punpckldq xmmA, xmmD ; xmmA=(00 10 20 01 11 21 02 12 02 12 22 03 13 23 04 14)
+ punpckldq xmmE, xmmH ; xmmE=(22 03 13 23 04 14 24 05 24 05 15 25 06 16 26 07)
+ punpckhdq xmmD, xmmB ; xmmD=(15 25 06 16 26 07 17 27 17 27 08 18 28 09 19 29)
+
+ pshufd xmmH, xmmG, 0x4E ; xmmH=(0C 1C 2C 0D 0E 1E 2E 0F 08 18 28 09 0A 1A 2A 0B)
+ movdqa xmmB, xmmF
+ punpckldq xmmG, xmmC ; xmmG=(08 18 28 09 19 29 0A 1A 0A 1A 2A 0B 1B 2B 0C 1C)
+ punpckldq xmmF, xmmH ; xmmF=(2A 0B 1B 2B 0C 1C 2C 0D 2C 0D 1D 2D 0E 1E 2E 0F)
+ punpckhdq xmmC, xmmB ; xmmC=(1D 2D 0E 1E 2E 0F 1F 2F 1F 2F -- -- -- -- -- --)
+
+ punpcklqdq xmmA, xmmE ; xmmA=(00 10 20 01 11 21 02 12 22 03 13 23 04 14 24 05)
+ punpcklqdq xmmD, xmmG ; xmmD=(15 25 06 16 26 07 17 27 08 18 28 09 19 29 0A 1A)
+ punpcklqdq xmmF, xmmC ; xmmF=(2A 0B 1B 2B 0C 1C 2C 0D 1D 2D 0E 1E 2E 0F 1F 2F)
+
+ cmp ecx, byte SIZEOF_XMMWORD
+ jb short .column_st32
+
+ test edi, SIZEOF_XMMWORD-1
+ jnz short .out1
+ ; --(aligned)-------------------
+ movntdq XMMWORD [edi+0*SIZEOF_XMMWORD], xmmA
+ movntdq XMMWORD [edi+1*SIZEOF_XMMWORD], xmmD
+ movntdq XMMWORD [edi+2*SIZEOF_XMMWORD], xmmF
+ jmp short .out0
+.out1: ; --(unaligned)-----------------
+ movdqu XMMWORD [edi+0*SIZEOF_XMMWORD], xmmA
+ movdqu XMMWORD [edi+1*SIZEOF_XMMWORD], xmmD
+ movdqu XMMWORD [edi+2*SIZEOF_XMMWORD], xmmF
+.out0:
+ add edi, byte RGB_PIXELSIZE*SIZEOF_XMMWORD ; outptr
+ sub ecx, byte SIZEOF_XMMWORD
+ jz near .nextrow
+
+ add esi, byte SIZEOF_XMMWORD ; inptr0
+ add ebx, byte SIZEOF_XMMWORD ; inptr1
+ add edx, byte SIZEOF_XMMWORD ; inptr2
+ jmp near .columnloop
+ alignx 16, 7
+
+.column_st32:
+ lea ecx, [ecx+ecx*2] ; imul ecx, RGB_PIXELSIZE
+ cmp ecx, byte 2*SIZEOF_XMMWORD
+ jb short .column_st16
+ movdqu XMMWORD [edi+0*SIZEOF_XMMWORD], xmmA
+ movdqu XMMWORD [edi+1*SIZEOF_XMMWORD], xmmD
+ add edi, byte 2*SIZEOF_XMMWORD ; outptr
+ movdqa xmmA, xmmF
+ sub ecx, byte 2*SIZEOF_XMMWORD
+ jmp short .column_st15
+.column_st16:
+ cmp ecx, byte SIZEOF_XMMWORD
+ jb short .column_st15
+ movdqu XMMWORD [edi+0*SIZEOF_XMMWORD], xmmA
+ add edi, byte SIZEOF_XMMWORD ; outptr
+ movdqa xmmA, xmmD
+ sub ecx, byte SIZEOF_XMMWORD
+.column_st15:
+ ; Store the lower 8 bytes of xmmA to the output when it has enough
+ ; space.
+ cmp ecx, byte SIZEOF_MMWORD
+ jb short .column_st7
+ movq XMM_MMWORD [edi], xmmA
+ add edi, byte SIZEOF_MMWORD
+ sub ecx, byte SIZEOF_MMWORD
+ psrldq xmmA, SIZEOF_MMWORD
+.column_st7:
+ ; Store the lower 4 bytes of xmmA to the output when it has enough
+ ; space.
+ cmp ecx, byte SIZEOF_DWORD
+ jb short .column_st3
+ movd XMM_DWORD [edi], xmmA
+ add edi, byte SIZEOF_DWORD
+ sub ecx, byte SIZEOF_DWORD
+ psrldq xmmA, SIZEOF_DWORD
+.column_st3:
+ ; Store the lower 2 bytes of eax to the output when it has enough
+ ; space.
+ movd eax, xmmA
+ cmp ecx, byte SIZEOF_WORD
+ jb short .column_st1
+ mov WORD [edi], ax
+ add edi, byte SIZEOF_WORD
+ sub ecx, byte SIZEOF_WORD
+ shr eax, 16
+.column_st1:
+ ; Store the lower 1 byte of eax to the output when it has enough
+ ; space.
+ test ecx, ecx
+ jz short .nextrow
+ mov BYTE [edi], al
+
+%else ; RGB_PIXELSIZE == 4 ; -----------
+
+%ifdef RGBX_FILLER_0XFF
+ pcmpeqb xmm6, xmm6 ; xmm6=XE=X(02468ACE********)
+ pcmpeqb xmm7, xmm7 ; xmm7=XO=X(13579BDF********)
+%else
+ pxor xmm6, xmm6 ; xmm6=XE=X(02468ACE********)
+ pxor xmm7, xmm7 ; xmm7=XO=X(13579BDF********)
+%endif
+ ; xmmA=(00 02 04 06 08 0A 0C 0E **), xmmB=(01 03 05 07 09 0B 0D 0F **)
+ ; xmmC=(10 12 14 16 18 1A 1C 1E **), xmmD=(11 13 15 17 19 1B 1D 1F **)
+ ; xmmE=(20 22 24 26 28 2A 2C 2E **), xmmF=(21 23 25 27 29 2B 2D 2F **)
+ ; xmmG=(30 32 34 36 38 3A 3C 3E **), xmmH=(31 33 35 37 39 3B 3D 3F **)
+
+ punpcklbw xmmA, xmmC ; xmmA=(00 10 02 12 04 14 06 16 08 18 0A 1A 0C 1C 0E 1E)
+ punpcklbw xmmE, xmmG ; xmmE=(20 30 22 32 24 34 26 36 28 38 2A 3A 2C 3C 2E 3E)
+ punpcklbw xmmB, xmmD ; xmmB=(01 11 03 13 05 15 07 17 09 19 0B 1B 0D 1D 0F 1F)
+ punpcklbw xmmF, xmmH ; xmmF=(21 31 23 33 25 35 27 37 29 39 2B 3B 2D 3D 2F 3F)
+
+ movdqa xmmC, xmmA
+ punpcklwd xmmA, xmmE ; xmmA=(00 10 20 30 02 12 22 32 04 14 24 34 06 16 26 36)
+ punpckhwd xmmC, xmmE ; xmmC=(08 18 28 38 0A 1A 2A 3A 0C 1C 2C 3C 0E 1E 2E 3E)
+ movdqa xmmG, xmmB
+ punpcklwd xmmB, xmmF ; xmmB=(01 11 21 31 03 13 23 33 05 15 25 35 07 17 27 37)
+ punpckhwd xmmG, xmmF ; xmmG=(09 19 29 39 0B 1B 2B 3B 0D 1D 2D 3D 0F 1F 2F 3F)
+
+ movdqa xmmD, xmmA
+ punpckldq xmmA, xmmB ; xmmA=(00 10 20 30 01 11 21 31 02 12 22 32 03 13 23 33)
+ punpckhdq xmmD, xmmB ; xmmD=(04 14 24 34 05 15 25 35 06 16 26 36 07 17 27 37)
+ movdqa xmmH, xmmC
+ punpckldq xmmC, xmmG ; xmmC=(08 18 28 38 09 19 29 39 0A 1A 2A 3A 0B 1B 2B 3B)
+ punpckhdq xmmH, xmmG ; xmmH=(0C 1C 2C 3C 0D 1D 2D 3D 0E 1E 2E 3E 0F 1F 2F 3F)
+
+ cmp ecx, byte SIZEOF_XMMWORD
+ jb short .column_st32
+
+ test edi, SIZEOF_XMMWORD-1
+ jnz short .out1
+ ; --(aligned)-------------------
+ movntdq XMMWORD [edi+0*SIZEOF_XMMWORD], xmmA
+ movntdq XMMWORD [edi+1*SIZEOF_XMMWORD], xmmD
+ movntdq XMMWORD [edi+2*SIZEOF_XMMWORD], xmmC
+ movntdq XMMWORD [edi+3*SIZEOF_XMMWORD], xmmH
+ jmp short .out0
+.out1: ; --(unaligned)-----------------
+ movdqu XMMWORD [edi+0*SIZEOF_XMMWORD], xmmA
+ movdqu XMMWORD [edi+1*SIZEOF_XMMWORD], xmmD
+ movdqu XMMWORD [edi+2*SIZEOF_XMMWORD], xmmC
+ movdqu XMMWORD [edi+3*SIZEOF_XMMWORD], xmmH
+.out0:
+ add edi, byte RGB_PIXELSIZE*SIZEOF_XMMWORD ; outptr
+ sub ecx, byte SIZEOF_XMMWORD
+ jz near .nextrow
+
+ add esi, byte SIZEOF_XMMWORD ; inptr0
+ add ebx, byte SIZEOF_XMMWORD ; inptr1
+ add edx, byte SIZEOF_XMMWORD ; inptr2
+ jmp near .columnloop
+ alignx 16, 7
+
+.column_st32:
+ cmp ecx, byte SIZEOF_XMMWORD/2
+ jb short .column_st16
+ movdqu XMMWORD [edi+0*SIZEOF_XMMWORD], xmmA
+ movdqu XMMWORD [edi+1*SIZEOF_XMMWORD], xmmD
+ add edi, byte 2*SIZEOF_XMMWORD ; outptr
+ movdqa xmmA, xmmC
+ movdqa xmmD, xmmH
+ sub ecx, byte SIZEOF_XMMWORD/2
+.column_st16:
+ cmp ecx, byte SIZEOF_XMMWORD/4
+ jb short .column_st15
+ movdqu XMMWORD [edi+0*SIZEOF_XMMWORD], xmmA
+ add edi, byte SIZEOF_XMMWORD ; outptr
+ movdqa xmmA, xmmD
+ sub ecx, byte SIZEOF_XMMWORD/4
+.column_st15:
+ ; Store two pixels (8 bytes) of xmmA to the output when it has enough
+ ; space.
+ cmp ecx, byte SIZEOF_XMMWORD/8
+ jb short .column_st7
+ movq XMM_MMWORD [edi], xmmA
+ add edi, byte SIZEOF_XMMWORD/8*4
+ sub ecx, byte SIZEOF_XMMWORD/8
+ psrldq xmmA, SIZEOF_XMMWORD/8*4
+.column_st7:
+ ; Store one pixel (4 bytes) of xmmA to the output when it has enough
+ ; space.
+ test ecx, ecx
+ jz short .nextrow
+ movd XMM_DWORD [edi], xmmA
+
+%endif ; RGB_PIXELSIZE ; ---------------
+
+ alignx 16, 7
+
+.nextrow:
+ pop ecx
+ pop esi
+ pop ebx
+ pop edx
+ pop edi
+ pop eax
+
+ add esi, byte SIZEOF_JSAMPROW
+ add ebx, byte SIZEOF_JSAMPROW
+ add edx, byte SIZEOF_JSAMPROW
+ add edi, byte SIZEOF_JSAMPROW ; output_buf
+ dec eax ; num_rows
+ jg near .rowloop
+
+ sfence ; flush the write buffer
+
+.return:
+ pop edi
+ pop esi
+; pop edx ; need not be preserved
+; pop ecx ; need not be preserved
+ pop ebx
+ mov esp, ebp ; esp <- aligned ebp
+ pop esp ; esp <- original ebp
+ pop ebp
+ ret
+
+; For some reason, the OS X linker does not honor the request to align the
+; segment unless we do this.
+ align 32
diff --git a/simd/i386/jdcolor-avx2.asm b/simd/i386/jdcolor-avx2.asm
new file mode 100644
index 0000000..f312064
--- /dev/null
+++ b/simd/i386/jdcolor-avx2.asm
@@ -0,0 +1,120 @@
+;
+; jdcolor.asm - colorspace conversion (AVX2)
+;
+; Copyright 2009 Pierre Ossman <ossman@cendio.se> for Cendio AB
+; Copyright (C) 2015, Intel Corporation.
+; Copyright (C) 2016, D. R. Commander.
+;
+; Based on the x86 SIMD extension for IJG JPEG library
+; Copyright (C) 1999-2006, MIYASAKA Masaru.
+; For conditions of distribution and use, see copyright notice in jsimdext.inc
+;
+; This file should be assembled with NASM (Netwide Assembler),
+; can *not* be assembled with Microsoft's MASM or any compatible
+; assembler (including Borland's Turbo Assembler).
+; NASM is available from http://nasm.sourceforge.net/ or
+; http://sourceforge.net/project/showfiles.php?group_id=6208
+;
+; [TAB8]
+
+%include "jsimdext.inc"
+
+; --------------------------------------------------------------------------
+
+%define SCALEBITS 16
+
+F_0_344 equ 22554 ; FIX(0.34414)
+F_0_714 equ 46802 ; FIX(0.71414)
+F_1_402 equ 91881 ; FIX(1.40200)
+F_1_772 equ 116130 ; FIX(1.77200)
+F_0_402 equ (F_1_402 - 65536) ; FIX(1.40200) - FIX(1)
+F_0_285 equ ( 65536 - F_0_714) ; FIX(1) - FIX(0.71414)
+F_0_228 equ (131072 - F_1_772) ; FIX(2) - FIX(1.77200)
+
+; --------------------------------------------------------------------------
+ SECTION SEG_CONST
+
+ alignz 32
+ global EXTN(jconst_ycc_rgb_convert_avx2)
+
+EXTN(jconst_ycc_rgb_convert_avx2):
+
+PW_F0402 times 16 dw F_0_402
+PW_MF0228 times 16 dw -F_0_228
+PW_MF0344_F0285 times 8 dw -F_0_344, F_0_285
+PW_ONE times 16 dw 1
+PD_ONEHALF times 8 dd 1 << (SCALEBITS-1)
+
+ alignz 32
+
+; --------------------------------------------------------------------------
+ SECTION SEG_TEXT
+ BITS 32
+
+%include "jdcolext-avx2.asm"
+
+%undef RGB_RED
+%undef RGB_GREEN
+%undef RGB_BLUE
+%undef RGB_PIXELSIZE
+%define RGB_RED EXT_RGB_RED
+%define RGB_GREEN EXT_RGB_GREEN
+%define RGB_BLUE EXT_RGB_BLUE
+%define RGB_PIXELSIZE EXT_RGB_PIXELSIZE
+%define jsimd_ycc_rgb_convert_avx2 jsimd_ycc_extrgb_convert_avx2
+%include "jdcolext-avx2.asm"
+
+%undef RGB_RED
+%undef RGB_GREEN
+%undef RGB_BLUE
+%undef RGB_PIXELSIZE
+%define RGB_RED EXT_RGBX_RED
+%define RGB_GREEN EXT_RGBX_GREEN
+%define RGB_BLUE EXT_RGBX_BLUE
+%define RGB_PIXELSIZE EXT_RGBX_PIXELSIZE
+%define jsimd_ycc_rgb_convert_avx2 jsimd_ycc_extrgbx_convert_avx2
+%include "jdcolext-avx2.asm"
+
+%undef RGB_RED
+%undef RGB_GREEN
+%undef RGB_BLUE
+%undef RGB_PIXELSIZE
+%define RGB_RED EXT_BGR_RED
+%define RGB_GREEN EXT_BGR_GREEN
+%define RGB_BLUE EXT_BGR_BLUE
+%define RGB_PIXELSIZE EXT_BGR_PIXELSIZE
+%define jsimd_ycc_rgb_convert_avx2 jsimd_ycc_extbgr_convert_avx2
+%include "jdcolext-avx2.asm"
+
+%undef RGB_RED
+%undef RGB_GREEN
+%undef RGB_BLUE
+%undef RGB_PIXELSIZE
+%define RGB_RED EXT_BGRX_RED
+%define RGB_GREEN EXT_BGRX_GREEN
+%define RGB_BLUE EXT_BGRX_BLUE
+%define RGB_PIXELSIZE EXT_BGRX_PIXELSIZE
+%define jsimd_ycc_rgb_convert_avx2 jsimd_ycc_extbgrx_convert_avx2
+%include "jdcolext-avx2.asm"
+
+%undef RGB_RED
+%undef RGB_GREEN
+%undef RGB_BLUE
+%undef RGB_PIXELSIZE
+%define RGB_RED EXT_XBGR_RED
+%define RGB_GREEN EXT_XBGR_GREEN
+%define RGB_BLUE EXT_XBGR_BLUE
+%define RGB_PIXELSIZE EXT_XBGR_PIXELSIZE
+%define jsimd_ycc_rgb_convert_avx2 jsimd_ycc_extxbgr_convert_avx2
+%include "jdcolext-avx2.asm"
+
+%undef RGB_RED
+%undef RGB_GREEN
+%undef RGB_BLUE
+%undef RGB_PIXELSIZE
+%define RGB_RED EXT_XRGB_RED
+%define RGB_GREEN EXT_XRGB_GREEN
+%define RGB_BLUE EXT_XRGB_BLUE
+%define RGB_PIXELSIZE EXT_XRGB_PIXELSIZE
+%define jsimd_ycc_rgb_convert_avx2 jsimd_ycc_extxrgb_convert_avx2
+%include "jdcolext-avx2.asm"
diff --git a/simd/jdcolor-mmx.asm b/simd/i386/jdcolor-mmx.asm
similarity index 97%
rename from simd/jdcolor-mmx.asm
rename to simd/i386/jdcolor-mmx.asm
index 4e58031..5136fe6 100644
--- a/simd/jdcolor-mmx.asm
+++ b/simd/i386/jdcolor-mmx.asm
@@ -2,7 +2,7 @@
; jdcolor.asm - colorspace conversion (MMX)
;
; Copyright 2009 Pierre Ossman <ossman@cendio.se> for Cendio AB
-; Copyright (C) 2009, D. R. Commander.
+; Copyright (C) 2009, 2016, D. R. Commander.
;
; Based on the x86 SIMD extension for IJG JPEG library
; Copyright (C) 1999-2006, MIYASAKA Masaru.
@@ -33,7 +33,7 @@
; --------------------------------------------------------------------------
SECTION SEG_CONST
- alignz 16
+ alignz 32
global EXTN(jconst_ycc_rgb_convert_mmx)
EXTN(jconst_ycc_rgb_convert_mmx):
@@ -44,7 +44,7 @@
PW_ONE times 4 dw 1
PD_ONEHALF times 2 dd 1 << (SCALEBITS-1)
- alignz 16
+ alignz 32
; --------------------------------------------------------------------------
SECTION SEG_TEXT
diff --git a/simd/jdcolor-sse2.asm b/simd/i386/jdcolor-sse2.asm
similarity index 81%
rename from simd/jdcolor-sse2.asm
rename to simd/i386/jdcolor-sse2.asm
index 7ff5d05..5552dca 100644
--- a/simd/jdcolor-sse2.asm
+++ b/simd/i386/jdcolor-sse2.asm
@@ -2,7 +2,7 @@
; jdcolor.asm - colorspace conversion (SSE2)
;
; Copyright 2009 Pierre Ossman <ossman@cendio.se> for Cendio AB
-; Copyright (C) 2009, D. R. Commander.
+; Copyright (C) 2009, 2016, D. R. Commander.
;
; Based on the x86 SIMD extension for IJG JPEG library
; Copyright (C) 1999-2006, MIYASAKA Masaru.
@@ -20,21 +20,21 @@
; --------------------------------------------------------------------------
-%define SCALEBITS 16
+%define SCALEBITS 16
-F_0_344 equ 22554 ; FIX(0.34414)
-F_0_714 equ 46802 ; FIX(0.71414)
-F_1_402 equ 91881 ; FIX(1.40200)
-F_1_772 equ 116130 ; FIX(1.77200)
-F_0_402 equ (F_1_402 - 65536) ; FIX(1.40200) - FIX(1)
-F_0_285 equ ( 65536 - F_0_714) ; FIX(1) - FIX(0.71414)
-F_0_228 equ (131072 - F_1_772) ; FIX(2) - FIX(1.77200)
+F_0_344 equ 22554 ; FIX(0.34414)
+F_0_714 equ 46802 ; FIX(0.71414)
+F_1_402 equ 91881 ; FIX(1.40200)
+F_1_772 equ 116130 ; FIX(1.77200)
+F_0_402 equ (F_1_402 - 65536) ; FIX(1.40200) - FIX(1)
+F_0_285 equ ( 65536 - F_0_714) ; FIX(1) - FIX(0.71414)
+F_0_228 equ (131072 - F_1_772) ; FIX(2) - FIX(1.77200)
; --------------------------------------------------------------------------
- SECTION SEG_CONST
+ SECTION SEG_CONST
- alignz 16
- global EXTN(jconst_ycc_rgb_convert_sse2)
+ alignz 32
+ global EXTN(jconst_ycc_rgb_convert_sse2)
EXTN(jconst_ycc_rgb_convert_sse2):
@@ -44,11 +44,11 @@
PW_ONE times 8 dw 1
PD_ONEHALF times 4 dd 1 << (SCALEBITS-1)
- alignz 16
+ alignz 32
; --------------------------------------------------------------------------
- SECTION SEG_TEXT
- BITS 32
+ SECTION SEG_TEXT
+ BITS 32
%include "jdcolext-sse2.asm"
diff --git a/simd/i386/jdmerge-avx2.asm b/simd/i386/jdmerge-avx2.asm
new file mode 100644
index 0000000..b194bb7
--- /dev/null
+++ b/simd/i386/jdmerge-avx2.asm
@@ -0,0 +1,126 @@
+;
+; jdmerge.asm - merged upsampling/color conversion (AVX2)
+;
+; Copyright 2009 Pierre Ossman <ossman@cendio.se> for Cendio AB
+; Copyright (C) 2009, 2016, D. R. Commander.
+; Copyright (C) 2015, Intel Corporation.
+;
+; Based on the x86 SIMD extension for IJG JPEG library
+; Copyright (C) 1999-2006, MIYASAKA Masaru.
+; For conditions of distribution and use, see copyright notice in jsimdext.inc
+;
+; This file should be assembled with NASM (Netwide Assembler),
+; can *not* be assembled with Microsoft's MASM or any compatible
+; assembler (including Borland's Turbo Assembler).
+; NASM is available from http://nasm.sourceforge.net/ or
+; http://sourceforge.net/project/showfiles.php?group_id=6208
+;
+; [TAB8]
+
+%include "jsimdext.inc"
+
+; --------------------------------------------------------------------------
+
+%define SCALEBITS 16
+
+F_0_344 equ 22554 ; FIX(0.34414)
+F_0_714 equ 46802 ; FIX(0.71414)
+F_1_402 equ 91881 ; FIX(1.40200)
+F_1_772 equ 116130 ; FIX(1.77200)
+F_0_402 equ (F_1_402 - 65536) ; FIX(1.40200) - FIX(1)
+F_0_285 equ ( 65536 - F_0_714) ; FIX(1) - FIX(0.71414)
+F_0_228 equ (131072 - F_1_772) ; FIX(2) - FIX(1.77200)
+
+; --------------------------------------------------------------------------
+ SECTION SEG_CONST
+
+ alignz 32
+ global EXTN(jconst_merged_upsample_avx2)
+
+EXTN(jconst_merged_upsample_avx2):
+
+PW_F0402 times 16 dw F_0_402
+PW_MF0228 times 16 dw -F_0_228
+PW_MF0344_F0285 times 8 dw -F_0_344, F_0_285
+PW_ONE times 16 dw 1
+PD_ONEHALF times 8 dd 1 << (SCALEBITS-1)
+
+ alignz 32
+
+; --------------------------------------------------------------------------
+ SECTION SEG_TEXT
+ BITS 32
+
+%include "jdmrgext-avx2.asm"
+
+%undef RGB_RED
+%undef RGB_GREEN
+%undef RGB_BLUE
+%undef RGB_PIXELSIZE
+%define RGB_RED EXT_RGB_RED
+%define RGB_GREEN EXT_RGB_GREEN
+%define RGB_BLUE EXT_RGB_BLUE
+%define RGB_PIXELSIZE EXT_RGB_PIXELSIZE
+%define jsimd_h2v1_merged_upsample_avx2 jsimd_h2v1_extrgb_merged_upsample_avx2
+%define jsimd_h2v2_merged_upsample_avx2 jsimd_h2v2_extrgb_merged_upsample_avx2
+%include "jdmrgext-avx2.asm"
+
+%undef RGB_RED
+%undef RGB_GREEN
+%undef RGB_BLUE
+%undef RGB_PIXELSIZE
+%define RGB_RED EXT_RGBX_RED
+%define RGB_GREEN EXT_RGBX_GREEN
+%define RGB_BLUE EXT_RGBX_BLUE
+%define RGB_PIXELSIZE EXT_RGBX_PIXELSIZE
+%define jsimd_h2v1_merged_upsample_avx2 jsimd_h2v1_extrgbx_merged_upsample_avx2
+%define jsimd_h2v2_merged_upsample_avx2 jsimd_h2v2_extrgbx_merged_upsample_avx2
+%include "jdmrgext-avx2.asm"
+
+%undef RGB_RED
+%undef RGB_GREEN
+%undef RGB_BLUE
+%undef RGB_PIXELSIZE
+%define RGB_RED EXT_BGR_RED
+%define RGB_GREEN EXT_BGR_GREEN
+%define RGB_BLUE EXT_BGR_BLUE
+%define RGB_PIXELSIZE EXT_BGR_PIXELSIZE
+%define jsimd_h2v1_merged_upsample_avx2 jsimd_h2v1_extbgr_merged_upsample_avx2
+%define jsimd_h2v2_merged_upsample_avx2 jsimd_h2v2_extbgr_merged_upsample_avx2
+%include "jdmrgext-avx2.asm"
+
+%undef RGB_RED
+%undef RGB_GREEN
+%undef RGB_BLUE
+%undef RGB_PIXELSIZE
+%define RGB_RED EXT_BGRX_RED
+%define RGB_GREEN EXT_BGRX_GREEN
+%define RGB_BLUE EXT_BGRX_BLUE
+%define RGB_PIXELSIZE EXT_BGRX_PIXELSIZE
+%define jsimd_h2v1_merged_upsample_avx2 jsimd_h2v1_extbgrx_merged_upsample_avx2
+%define jsimd_h2v2_merged_upsample_avx2 jsimd_h2v2_extbgrx_merged_upsample_avx2
+%include "jdmrgext-avx2.asm"
+
+%undef RGB_RED
+%undef RGB_GREEN
+%undef RGB_BLUE
+%undef RGB_PIXELSIZE
+%define RGB_RED EXT_XBGR_RED
+%define RGB_GREEN EXT_XBGR_GREEN
+%define RGB_BLUE EXT_XBGR_BLUE
+%define RGB_PIXELSIZE EXT_XBGR_PIXELSIZE
+%define jsimd_h2v1_merged_upsample_avx2 jsimd_h2v1_extxbgr_merged_upsample_avx2
+%define jsimd_h2v2_merged_upsample_avx2 jsimd_h2v2_extxbgr_merged_upsample_avx2
+%include "jdmrgext-avx2.asm"
+
+%undef RGB_RED
+%undef RGB_GREEN
+%undef RGB_BLUE
+%undef RGB_PIXELSIZE
+%define RGB_RED EXT_XRGB_RED
+%define RGB_GREEN EXT_XRGB_GREEN
+%define RGB_BLUE EXT_XRGB_BLUE
+%define RGB_PIXELSIZE EXT_XRGB_PIXELSIZE
+%define jsimd_h2v1_merged_upsample_avx2 jsimd_h2v1_extxrgb_merged_upsample_avx2
+%define jsimd_h2v2_merged_upsample_avx2 jsimd_h2v2_extxrgb_merged_upsample_avx2
+%include "jdmrgext-avx2.asm"
diff --git a/simd/jdmerge-mmx.asm b/simd/i386/jdmerge-mmx.asm
similarity index 97%
rename from simd/jdmerge-mmx.asm
rename to simd/i386/jdmerge-mmx.asm
index ee58bff..7ce1d86 100644
--- a/simd/jdmerge-mmx.asm
+++ b/simd/i386/jdmerge-mmx.asm
@@ -2,7 +2,7 @@
; jdmerge.asm - merged upsampling/color conversion (MMX)
;
; Copyright 2009 Pierre Ossman <ossman@cendio.se> for Cendio AB
-; Copyright (C) 2009, D. R. Commander.
+; Copyright (C) 2009, 2016, D. R. Commander.
;
; Based on the x86 SIMD extension for IJG JPEG library
; Copyright (C) 1999-2006, MIYASAKA Masaru.
@@ -33,7 +33,7 @@
; --------------------------------------------------------------------------
SECTION SEG_CONST
- alignz 16
+ alignz 32
global EXTN(jconst_merged_upsample_mmx)
EXTN(jconst_merged_upsample_mmx):
@@ -44,7 +44,7 @@
PW_ONE times 4 dw 1
PD_ONEHALF times 2 dd 1 << (SCALEBITS-1)
- alignz 16
+ alignz 32
; --------------------------------------------------------------------------
SECTION SEG_TEXT
diff --git a/simd/jdmerge-sse2.asm b/simd/i386/jdmerge-sse2.asm
similarity index 84%
rename from simd/jdmerge-sse2.asm
rename to simd/i386/jdmerge-sse2.asm
index 236de5a..fc1d6ad 100644
--- a/simd/jdmerge-sse2.asm
+++ b/simd/i386/jdmerge-sse2.asm
@@ -2,7 +2,7 @@
; jdmerge.asm - merged upsampling/color conversion (SSE2)
;
; Copyright 2009 Pierre Ossman <ossman@cendio.se> for Cendio AB
-; Copyright (C) 2009, D. R. Commander.
+; Copyright (C) 2009, 2016, D. R. Commander.
;
; Based on the x86 SIMD extension for IJG JPEG library
; Copyright (C) 1999-2006, MIYASAKA Masaru.
@@ -20,21 +20,21 @@
; --------------------------------------------------------------------------
-%define SCALEBITS 16
+%define SCALEBITS 16
-F_0_344 equ 22554 ; FIX(0.34414)
-F_0_714 equ 46802 ; FIX(0.71414)
-F_1_402 equ 91881 ; FIX(1.40200)
-F_1_772 equ 116130 ; FIX(1.77200)
-F_0_402 equ (F_1_402 - 65536) ; FIX(1.40200) - FIX(1)
-F_0_285 equ ( 65536 - F_0_714) ; FIX(1) - FIX(0.71414)
-F_0_228 equ (131072 - F_1_772) ; FIX(2) - FIX(1.77200)
+F_0_344 equ 22554 ; FIX(0.34414)
+F_0_714 equ 46802 ; FIX(0.71414)
+F_1_402 equ 91881 ; FIX(1.40200)
+F_1_772 equ 116130 ; FIX(1.77200)
+F_0_402 equ (F_1_402 - 65536) ; FIX(1.40200) - FIX(1)
+F_0_285 equ ( 65536 - F_0_714) ; FIX(1) - FIX(0.71414)
+F_0_228 equ (131072 - F_1_772) ; FIX(2) - FIX(1.77200)
; --------------------------------------------------------------------------
- SECTION SEG_CONST
+ SECTION SEG_CONST
- alignz 16
- global EXTN(jconst_merged_upsample_sse2)
+ alignz 32
+ global EXTN(jconst_merged_upsample_sse2)
EXTN(jconst_merged_upsample_sse2):
@@ -44,11 +44,11 @@
PW_ONE times 8 dw 1
PD_ONEHALF times 4 dd 1 << (SCALEBITS-1)
- alignz 16
+ alignz 32
; --------------------------------------------------------------------------
- SECTION SEG_TEXT
- BITS 32
+ SECTION SEG_TEXT
+ BITS 32
%include "jdmrgext-sse2.asm"
diff --git a/simd/i386/jdmrgext-avx2.asm b/simd/i386/jdmrgext-avx2.asm
new file mode 100644
index 0000000..5f2b567
--- /dev/null
+++ b/simd/i386/jdmrgext-avx2.asm
@@ -0,0 +1,576 @@
+;
+; jdmrgext.asm - merged upsampling/color conversion (AVX2)
+;
+; Copyright 2009, 2012 Pierre Ossman <ossman@cendio.se> for Cendio AB
+; Copyright (C) 2012, 2016, D. R. Commander.
+; Copyright (C) 2015, Intel Corporation.
+;
+; Based on the x86 SIMD extension for IJG JPEG library
+; Copyright (C) 1999-2006, MIYASAKA Masaru.
+; For conditions of distribution and use, see copyright notice in jsimdext.inc
+;
+; This file should be assembled with NASM (Netwide Assembler),
+; can *not* be assembled with Microsoft's MASM or any compatible
+; assembler (including Borland's Turbo Assembler).
+; NASM is available from http://nasm.sourceforge.net/ or
+; http://sourceforge.net/project/showfiles.php?group_id=6208
+;
+; [TAB8]
+
+%include "jcolsamp.inc"
+
+; --------------------------------------------------------------------------
+;
+; Upsample and color convert for the case of 2:1 horizontal and 1:1 vertical.
+;
+; GLOBAL(void)
+; jsimd_h2v1_merged_upsample_avx2 (JDIMENSION output_width,
+; JSAMPIMAGE input_buf,
+; JDIMENSION in_row_group_ctr,
+; JSAMPARRAY output_buf);
+;
+
+%define output_width(b) (b)+8 ; JDIMENSION output_width
+%define input_buf(b) (b)+12 ; JSAMPIMAGE input_buf
+%define in_row_group_ctr(b) (b)+16 ; JDIMENSION in_row_group_ctr
+%define output_buf(b) (b)+20 ; JSAMPARRAY output_buf
+
+%define original_ebp ebp+0
+%define wk(i) ebp-(WK_NUM-(i))*SIZEOF_YMMWORD ; ymmword wk[WK_NUM]
+%define WK_NUM 3
+%define gotptr wk(0)-SIZEOF_POINTER ; void * gotptr
+
+ align 32
+ global EXTN(jsimd_h2v1_merged_upsample_avx2)
+
+EXTN(jsimd_h2v1_merged_upsample_avx2):
+ push ebp
+ mov eax, esp ; eax = original ebp
+ sub esp, byte 4
+ and esp, byte (-SIZEOF_YMMWORD) ; align to 256 bits
+ mov [esp], eax
+ mov ebp, esp ; ebp = aligned ebp
+ lea esp, [wk(0)]
+ pushpic eax ; make a room for GOT address
+ push ebx
+; push ecx ; need not be preserved
+; push edx ; need not be preserved
+ push esi
+ push edi
+
+ get_GOT ebx ; get GOT address
+ movpic POINTER [gotptr], ebx ; save GOT address
+
+ mov ecx, JDIMENSION [output_width(eax)] ; col
+ test ecx, ecx
+ jz near .return
+
+ push ecx
+
+ mov edi, JSAMPIMAGE [input_buf(eax)]
+ mov ecx, JDIMENSION [in_row_group_ctr(eax)]
+ mov esi, JSAMPARRAY [edi+0*SIZEOF_JSAMPARRAY]
+ mov ebx, JSAMPARRAY [edi+1*SIZEOF_JSAMPARRAY]
+ mov edx, JSAMPARRAY [edi+2*SIZEOF_JSAMPARRAY]
+ mov edi, JSAMPARRAY [output_buf(eax)]
+ mov esi, JSAMPROW [esi+ecx*SIZEOF_JSAMPROW] ; inptr0
+ mov ebx, JSAMPROW [ebx+ecx*SIZEOF_JSAMPROW] ; inptr1
+ mov edx, JSAMPROW [edx+ecx*SIZEOF_JSAMPROW] ; inptr2
+ mov edi, JSAMPROW [edi] ; outptr
+
+ pop ecx ; col
+
+ alignx 16, 7
+.columnloop:
+ movpic eax, POINTER [gotptr] ; load GOT address (eax)
+
+ vmovdqu ymm6, YMMWORD [ebx] ; ymm6=Cb(0123456789ABCDEFGHIJKLMNOPQRSTUV)
+ vmovdqu ymm7, YMMWORD [edx] ; ymm7=Cr(0123456789ABCDEFGHIJKLMNOPQRSTUV)
+
+ vpxor ymm1, ymm1, ymm1 ; ymm1=(all 0's)
+ vpcmpeqw ymm3, ymm3, ymm3
+ vpsllw ymm3, ymm3, 7 ; ymm3={0xFF80 0xFF80 0xFF80 0xFF80 ..}
+
+ vpermq ymm6, ymm6, 0xd8 ; ymm6=Cb(01234567GHIJKLMN89ABCDEFOPQRSTUV)
+ vpermq ymm7, ymm7, 0xd8 ; ymm7=Cr(01234567GHIJKLMN89ABCDEFOPQRSTUV)
+ vpunpcklbw ymm4, ymm6, ymm1 ; ymm4=Cb(0123456789ABCDEF)=CbL
+ vpunpckhbw ymm6, ymm6, ymm1 ; ymm6=Cb(GHIJKLMNOPQRSTUV)=CbH
+ vpunpcklbw ymm0, ymm7, ymm1 ; ymm0=Cr(0123456789ABCDEF)=CrL
+ vpunpckhbw ymm7, ymm7, ymm1 ; ymm7=Cr(GHIJKLMNOPQRSTUV)=CrH
+
+ vpaddw ymm5, ymm6, ymm3
+ vpaddw ymm2, ymm4, ymm3
+ vpaddw ymm1, ymm7, ymm3
+ vpaddw ymm3, ymm0, ymm3
+
+ ; (Original)
+ ; R = Y + 1.40200 * Cr
+ ; G = Y - 0.34414 * Cb - 0.71414 * Cr
+ ; B = Y + 1.77200 * Cb
+ ;
+ ; (This implementation)
+ ; R = Y + 0.40200 * Cr + Cr
+ ; G = Y - 0.34414 * Cb + 0.28586 * Cr - Cr
+ ; B = Y - 0.22800 * Cb + Cb + Cb
+
+ vpaddw ymm6, ymm5, ymm5 ; ymm6=2*CbH
+ vpaddw ymm4, ymm2, ymm2 ; ymm4=2*CbL
+ vpaddw ymm7, ymm1, ymm1 ; ymm7=2*CrH
+ vpaddw ymm0, ymm3, ymm3 ; ymm0=2*CrL
+
+ vpmulhw ymm6, ymm6, [GOTOFF(eax,PW_MF0228)] ; ymm6=(2*CbH * -FIX(0.22800))
+ vpmulhw ymm4, ymm4, [GOTOFF(eax,PW_MF0228)] ; ymm4=(2*CbL * -FIX(0.22800))
+ vpmulhw ymm7, ymm7, [GOTOFF(eax,PW_F0402)] ; ymm7=(2*CrH * FIX(0.40200))
+ vpmulhw ymm0, ymm0, [GOTOFF(eax,PW_F0402)] ; ymm0=(2*CrL * FIX(0.40200))
+
+ vpaddw ymm6, ymm6, [GOTOFF(eax,PW_ONE)]
+ vpaddw ymm4, ymm4, [GOTOFF(eax,PW_ONE)]
+ vpsraw ymm6, ymm6, 1 ; ymm6=(CbH * -FIX(0.22800))
+ vpsraw ymm4, ymm4, 1 ; ymm4=(CbL * -FIX(0.22800))
+ vpaddw ymm7, ymm7, [GOTOFF(eax,PW_ONE)]
+ vpaddw ymm0, ymm0, [GOTOFF(eax,PW_ONE)]
+ vpsraw ymm7, ymm7, 1 ; ymm7=(CrH * FIX(0.40200))
+ vpsraw ymm0, ymm0, 1 ; ymm0=(CrL * FIX(0.40200))
+
+ vpaddw ymm6, ymm6, ymm5
+ vpaddw ymm4, ymm4, ymm2
+ vpaddw ymm6, ymm6, ymm5 ; ymm6=(CbH * FIX(1.77200))=(B-Y)H
+ vpaddw ymm4, ymm4, ymm2 ; ymm4=(CbL * FIX(1.77200))=(B-Y)L
+ vpaddw ymm7, ymm7, ymm1 ; ymm7=(CrH * FIX(1.40200))=(R-Y)H
+ vpaddw ymm0, ymm0, ymm3 ; ymm0=(CrL * FIX(1.40200))=(R-Y)L
+
+ vmovdqa YMMWORD [wk(0)], ymm6 ; wk(0)=(B-Y)H
+ vmovdqa YMMWORD [wk(1)], ymm7 ; wk(1)=(R-Y)H
+
+ vpunpckhwd ymm6, ymm5, ymm1
+ vpunpcklwd ymm5, ymm5, ymm1
+ vpmaddwd ymm5, ymm5, [GOTOFF(eax,PW_MF0344_F0285)]
+ vpmaddwd ymm6, ymm6, [GOTOFF(eax,PW_MF0344_F0285)]
+ vpunpckhwd ymm7, ymm2, ymm3
+ vpunpcklwd ymm2, ymm2, ymm3
+ vpmaddwd ymm2, ymm2, [GOTOFF(eax,PW_MF0344_F0285)]
+ vpmaddwd ymm7, ymm7, [GOTOFF(eax,PW_MF0344_F0285)]
+
+ vpaddd ymm5, ymm5, [GOTOFF(eax,PD_ONEHALF)]
+ vpaddd ymm6, ymm6, [GOTOFF(eax,PD_ONEHALF)]
+ vpsrad ymm5, ymm5, SCALEBITS
+ vpsrad ymm6, ymm6, SCALEBITS
+ vpaddd ymm2, ymm2, [GOTOFF(eax,PD_ONEHALF)]
+ vpaddd ymm7, ymm7, [GOTOFF(eax,PD_ONEHALF)]
+ vpsrad ymm2, ymm2, SCALEBITS
+ vpsrad ymm7, ymm7, SCALEBITS
+
+ vpackssdw ymm5, ymm5, ymm6 ; ymm5=CbH*-FIX(0.344)+CrH*FIX(0.285)
+ vpackssdw ymm2, ymm2, ymm7 ; ymm2=CbL*-FIX(0.344)+CrL*FIX(0.285)
+ vpsubw ymm5, ymm5, ymm1 ; ymm5=CbH*-FIX(0.344)+CrH*-FIX(0.714)=(G-Y)H
+ vpsubw ymm2, ymm2, ymm3 ; ymm2=CbL*-FIX(0.344)+CrL*-FIX(0.714)=(G-Y)L
+
+ vmovdqa YMMWORD [wk(2)], ymm5 ; wk(2)=(G-Y)H
+
+ mov al, 2 ; Yctr
+ jmp short .Yloop_1st
+ alignx 16, 7
+
+.Yloop_2nd:
+ vmovdqa ymm0, YMMWORD [wk(1)] ; ymm0=(R-Y)H
+ vmovdqa ymm2, YMMWORD [wk(2)] ; ymm2=(G-Y)H
+ vmovdqa ymm4, YMMWORD [wk(0)] ; ymm4=(B-Y)H
+ alignx 16, 7
+
+.Yloop_1st:
+ vmovdqu ymm7, YMMWORD [esi] ; ymm7=Y(0123456789ABCDEFGHIJKLMNOPQRSTUV)
+
+ vpcmpeqw ymm6, ymm6, ymm6
+ vpsrlw ymm6, ymm6, BYTE_BIT ; ymm6={0xFF 0x00 0xFF 0x00 ..}
+ vpand ymm6, ymm6, ymm7 ; ymm6=Y(02468ACEGIKMOQSU)=YE
+ vpsrlw ymm7, ymm7, BYTE_BIT ; ymm7=Y(13579BDFHJLNPRTV)=YO
+
+ vmovdqa ymm1, ymm0 ; ymm1=ymm0=(R-Y)(L/H)
+ vmovdqa ymm3, ymm2 ; ymm3=ymm2=(G-Y)(L/H)
+ vmovdqa ymm5, ymm4 ; ymm5=ymm4=(B-Y)(L/H)
+
+ vpaddw ymm0, ymm0, ymm6 ; ymm0=((R-Y)+YE)=RE=R(02468ACEGIKMOQSU)
+ vpaddw ymm1, ymm1, ymm7 ; ymm1=((R-Y)+YO)=RO=R(13579BDFHJLNPRTV)
+ vpackuswb ymm0, ymm0, ymm0 ; ymm0=R(02468ACE********GIKMOQSU********)
+ vpackuswb ymm1, ymm1, ymm1 ; ymm1=R(13579BDF********HJLNPRTV********)
+
+ vpaddw ymm2, ymm2, ymm6 ; ymm2=((G-Y)+YE)=GE=G(02468ACEGIKMOQSU)
+ vpaddw ymm3, ymm3, ymm7 ; ymm3=((G-Y)+YO)=GO=G(13579BDFHJLNPRTV)
+ vpackuswb ymm2, ymm2, ymm2 ; ymm2=G(02468ACE********GIKMOQSU********)
+ vpackuswb ymm3, ymm3, ymm3 ; ymm3=G(13579BDF********HJLNPRTV********)
+
+ vpaddw ymm4, ymm4, ymm6 ; ymm4=((B-Y)+YE)=BE=B(02468ACEGIKMOQSU)
+ vpaddw ymm5, ymm5, ymm7 ; ymm5=((B-Y)+YO)=BO=B(13579BDFHJLNPRTV)
+ vpackuswb ymm4, ymm4, ymm4 ; ymm4=B(02468ACE********GIKMOQSU********)
+ vpackuswb ymm5, ymm5, ymm5 ; ymm5=B(13579BDF********HJLNPRTV********)
+
+%if RGB_PIXELSIZE == 3 ; ---------------
+
+ ; ymmA=(00 02 04 06 08 0A 0C 0E ** 0G 0I 0K 0M 0O 0Q 0S 0U **)
+ ; ymmB=(01 03 05 07 09 0B 0D 0F ** 0H 0J 0L 0N 0P 0R 0T 0V **)
+ ; ymmC=(10 12 14 16 18 1A 1C 1E ** 1G 1I 1K 1M 1O 1Q 1S 1U **)
+ ; ymmD=(11 13 15 17 19 1B 1D 1F ** 1H 1J 1L 1N 1P 1R 1T 1V **)
+ ; ymmE=(20 22 24 26 28 2A 2C 2E ** 2G 2I 2K 2M 2O 2Q 2S 2U **)
+ ; ymmF=(21 23 25 27 29 2B 2D 2F ** 2H 2J 2L 2N 2P 2R 2T 2V **)
+ ; ymmG=(** ** ** ** ** ** ** ** ** ** ** ** ** ** ** ** ** **)
+ ; ymmH=(** ** ** ** ** ** ** ** ** ** ** ** ** ** ** ** ** **)
+
+ vpunpcklbw ymmA, ymmA, ymmC ; ymmA=(00 10 02 12 04 14 06 16 08 18 0A 1A 0C 1C 0E 1E
+ ; 0G 1G 0I 1I 0K 1K 0M 1M 0O 1O 0Q 1Q 0S 1S 0U 1U)
+ vpunpcklbw ymmE, ymmE, ymmB ; ymmE=(20 01 22 03 24 05 26 07 28 09 2A 0B 2C 0D 2E 0F
+ ; 2G 0H 2I 0J 2K 0L 2M 0N 2O 0P 2Q 0R 2S 0T 2U 0V)
+ vpunpcklbw ymmD, ymmD, ymmF ; ymmD=(11 21 13 23 15 25 17 27 19 29 1B 2B 1D 2D 1F 2F
+ ; 1H 2H 1J 2J 1L 2L 1N 2N 1P 2P 1R 2R 1T 2T 1V 2V)
+
+ vpsrldq ymmH, ymmA, 2 ; ymmH=(02 12 04 14 06 16 08 18 0A 1A 0C 1C 0E 1E 0G 1G
+ ; 0I 1I 0K 1K 0M 1M 0O 1O 0Q 1Q 0S 1S 0U 1U -- --)
+ vpunpckhwd ymmG, ymmA, ymmE ; ymmG=(08 18 28 09 0A 1A 2A 0B 0C 1C 2C 0D 0E 1E 2E 0F
+ ; 0O 1O 2O 0P 0Q 1Q 2Q 0R 0S 1S 2S 0T 0U 1U 2U 0V)
+ vpunpcklwd ymmA, ymmA, ymmE ; ymmA=(00 10 20 01 02 12 22 03 04 14 24 05 06 16 26 07
+ ; 0G 1G 2G 0H 0I 1I 2I 0J 0K 1K 2K 0L 0M 1M 2M 0N)
+
+ vpsrldq ymmE, ymmE, 2 ; ymmE=(22 03 24 05 26 07 28 09 2A 0B 2C 0D 2E 0F 2G 0H
+ ; 2I 0J 2K 0L 2M 0N 2O 0P 2Q 0R 2S 0T 2U 0V -- --)
+
+ vpsrldq ymmB, ymmD, 2 ; ymmB=(13 23 15 25 17 27 19 29 1B 2B 1D 2D 1F 2F 1H 2H
+ ; 1J 2J 1L 2L 1N 2N 1P 2P 1R 2R 1T 2T 1V 2V -- --)
+ vpunpckhwd ymmC, ymmD, ymmH ; ymmC=(19 29 0A 1A 1B 2B 0C 1C 1D 2D 0E 1E 1F 2F 0G 1G
+ ; 1P 2P 0Q 1Q 1R 2R 0S 1S 1T 2T 0U 1U 1V 2V -- --)
+ vpunpcklwd ymmD, ymmD, ymmH ; ymmD=(11 21 02 12 13 23 04 14 15 25 06 16 17 27 08 18
+ ; 1H 2H 0I 1I 1J 2J 0K 1K 1L 2L 0M 1M 1N 2N 0O 1O)
+
+ vpunpckhwd ymmF, ymmE, ymmB ; ymmF=(2A 0B 1B 2B 2C 0D 1D 2D 2E 0F 1F 2F 2G 0H 1H 2H
+ ; 2Q 0R 1R 2R 2S 0T 1T 2T 2U 0V 1V 2V -- -- -- --)
+ vpunpcklwd ymmE, ymmE, ymmB ; ymmE=(22 03 13 23 24 05 15 25 26 07 17 27 28 09 19 29
+ ; 2I 0J 1J 2J 2K 0L 1L 2L 2M 0N 1N 2N 2O 0P 1P 2P)
+
+ vpshufd ymmH, ymmA, 0x4E ; ymmH=(04 14 24 05 06 16 26 07 00 10 20 01 02 12 22 03
+ ; 0K 1K 2K 0L 0M 1M 2M 0N 0G 1G 2G 0H 0I 1I 2I 0J)
+ vpunpckldq ymmA, ymmA, ymmD ; ymmA=(00 10 20 01 11 21 02 12 02 12 22 03 13 23 04 14
+ ; 0G 1G 2G 0H 1H 2H 0I 1I 0I 1I 2I 0J 1J 2J 0K 1K)
+ vpunpckhdq ymmD, ymmD, ymmE ; ymmD=(15 25 06 16 26 07 17 27 17 27 08 18 28 09 19 29
+ ; 1L 2L 0M 1M 2M 0N 1N 2N 1N 2N 0O 1O 2O 0P 1P 2P)
+ vpunpckldq ymmE, ymmE, ymmH ; ymmE=(22 03 13 23 04 14 24 05 24 05 15 25 06 16 26 07
+ ; 2I 0J 1J 2J 0K 1K 2K 0L 2K 0L 1L 2L 0M 1M 2M 0N)
+
+ vpshufd ymmH, ymmG, 0x4E ; ymmH=(0C 1C 2C 0D 0E 1E 2E 0F 08 18 28 09 0A 1A 2A 0B
+ ; 0S 1S 2S 0T 0U 1U 2U 0V 0O 1O 2O 0P 0Q 1Q 2Q 0R)
+ vpunpckldq ymmG, ymmG, ymmC ; ymmG=(08 18 28 09 19 29 0A 1A 0A 1A 2A 0B 1B 2B 0C 1C
+ ; 0O 1O 2O 0P 1P 2P 0Q 1Q 0Q 1Q 2Q 0R 1R 2R 0S 1S)
+ vpunpckhdq ymmC, ymmC, ymmF ; ymmC=(1D 2D 0E 1E 2E 0F 1F 2F 1F 2F 0G 1G 2G 0H 1H 2H
+ ; 1T 2T 0U 1U 2U 0V 1V 2V 1V 2V -- -- -- -- -- --)
+ vpunpckldq ymmF, ymmF, ymmH ; ymmF=(2A 0B 1B 2B 0C 1C 2C 0D 2C 0D 1D 2D 0E 1E 2E 0F
+ ; 2Q 0R 1R 2R 0S 1S 2S 0T 2S 0T 1T 2T 0U 1U 2U 0V)
+
+ vpunpcklqdq ymmH, ymmA, ymmE ; ymmH=(00 10 20 01 11 21 02 12 22 03 13 23 04 14 24 05
+ ; 0G 1G 2G 0H 1H 2H 0I 1I 2I 0J 1J 2J 0K 1K 2K 0L)
+ vpunpcklqdq ymmG, ymmD, ymmG ; ymmG=(15 25 06 16 26 07 17 27 08 18 28 09 19 29 0A 1A
+ ; 1L 2L 0M 1M 2M 0N 1N 2N 0O 1O 2O 0P 1P 2P 0Q 1Q)
+ vpunpcklqdq ymmC, ymmF, ymmC ; ymmC=(2A 0B 1B 2B 0C 1C 2C 0D 1D 2D 0E 1E 2E 0F 1F 2F
+ ; 2Q 0R 1R 2R 0S 1S 2S 0T 1T 2T 0U 1U 2U 0V 1V 2V)
+
+ vperm2i128 ymmA, ymmH, ymmG, 0x20 ; ymmA=(00 10 20 01 11 21 02 12 22 03 13 23 04 14 24 05
+ ; 15 25 06 16 26 07 17 27 08 18 28 09 19 29 0A 1A)
+ vperm2i128 ymmD, ymmC, ymmH, 0x30 ; ymmD=(2A 0B 1B 2B 0C 1C 2C 0D 1D 2D 0E 1E 2E 0F 1F 2F
+ ; 0G 1G 2G 0H 1H 2H 0I 1I 2I 0J 1J 2J 0K 1K 2K 0L)
+ vperm2i128 ymmF, ymmG, ymmC, 0x31 ; ymmF=(1L 2L 0M 1M 2M 0N 1N 2N 0O 1O 2O 0P 1P 2P 0Q 1Q
+ ; 2Q 0R 1R 2R 0S 1S 2S 0T 1T 2T 0U 1U 2U 0V 1V 2V)
+
+ cmp ecx, byte SIZEOF_YMMWORD
+ jb short .column_st64
+
+ test edi, SIZEOF_YMMWORD-1
+ jnz short .out1
+ ; --(aligned)-------------------
+ vmovntdq YMMWORD [edi+0*SIZEOF_YMMWORD], ymmA
+ vmovntdq YMMWORD [edi+1*SIZEOF_YMMWORD], ymmD
+ vmovntdq YMMWORD [edi+2*SIZEOF_YMMWORD], ymmF
+ jmp short .out0
+.out1: ; --(unaligned)-----------------
+ vmovdqu YMMWORD [edi+0*SIZEOF_YMMWORD], ymmA
+ vmovdqu YMMWORD [edi+1*SIZEOF_YMMWORD], ymmD
+ vmovdqu YMMWORD [edi+2*SIZEOF_YMMWORD], ymmF
+.out0:
+ add edi, byte RGB_PIXELSIZE*SIZEOF_YMMWORD ; outptr
+ sub ecx, byte SIZEOF_YMMWORD
+ jz near .endcolumn
+
+ add esi, byte SIZEOF_YMMWORD ; inptr0
+ dec al ; Yctr
+ jnz near .Yloop_2nd
+
+ add ebx, byte SIZEOF_YMMWORD ; inptr1
+ add edx, byte SIZEOF_YMMWORD ; inptr2
+ jmp near .columnloop
+ alignx 16, 7
+
+.column_st64:
+ lea ecx, [ecx+ecx*2] ; imul ecx, RGB_PIXELSIZE
+ cmp ecx, byte 2*SIZEOF_YMMWORD
+ jb short .column_st32
+ vmovdqu YMMWORD [edi+0*SIZEOF_YMMWORD], ymmA
+ vmovdqu YMMWORD [edi+1*SIZEOF_YMMWORD], ymmD
+ add edi, byte 2*SIZEOF_YMMWORD ; outptr
+ vmovdqa ymmA, ymmF
+ sub ecx, byte 2*SIZEOF_YMMWORD
+ jmp short .column_st31
+.column_st32:
+ cmp ecx, byte SIZEOF_YMMWORD
+ jb short .column_st31
+ vmovdqu YMMWORD [edi+0*SIZEOF_YMMWORD], ymmA
+ add edi, byte SIZEOF_YMMWORD ; outptr
+ vmovdqa ymmA, ymmD
+ sub ecx, byte SIZEOF_YMMWORD
+ jmp short .column_st31
+.column_st31:
+ cmp ecx, byte SIZEOF_XMMWORD
+ jb short .column_st15
+ vmovdqu XMMWORD [edi+0*SIZEOF_XMMWORD], xmmA
+ add edi, byte SIZEOF_XMMWORD ; outptr
+ vperm2i128 ymmA, ymmA, ymmA, 1
+ sub ecx, byte SIZEOF_XMMWORD
+.column_st15:
+ ; Store the lower 8 bytes of xmmA to the output when it has enough
+ ; space.
+ cmp ecx, byte SIZEOF_MMWORD
+ jb short .column_st7
+ vmovq XMM_MMWORD [edi], xmmA
+ add edi, byte SIZEOF_MMWORD
+ sub ecx, byte SIZEOF_MMWORD
+ vpsrldq xmmA, xmmA, SIZEOF_MMWORD
+.column_st7:
+ ; Store the lower 4 bytes of xmmA to the output when it has enough
+ ; space.
+ cmp ecx, byte SIZEOF_DWORD
+ jb short .column_st3
+ vmovd XMM_DWORD [edi], xmmA
+ add edi, byte SIZEOF_DWORD
+ sub ecx, byte SIZEOF_DWORD
+ vpsrldq xmmA, xmmA, SIZEOF_DWORD
+.column_st3:
+ ; Store the lower 2 bytes of eax to the output when it has enough
+ ; space.
+ vmovd eax, xmmA
+ cmp ecx, byte SIZEOF_WORD
+ jb short .column_st1
+ mov WORD [edi], ax
+ add edi, byte SIZEOF_WORD
+ sub ecx, byte SIZEOF_WORD
+ shr eax, 16
+.column_st1:
+ ; Store the lower 1 byte of eax to the output when it has enough
+ ; space.
+ test ecx, ecx
+ jz short .endcolumn
+ mov BYTE [edi], al
+
+%else ; RGB_PIXELSIZE == 4 ; -----------
+
+%ifdef RGBX_FILLER_0XFF
+ vpcmpeqb ymm6, ymm6, ymm6 ; ymm6=XE=X(02468ACE********GIKMOQSU********)
+ vpcmpeqb ymm7, ymm7, ymm7 ; ymm7=XO=X(13579BDF********HJLNPRTV********)
+%else
+ vpxor ymm6, ymm6, ymm6 ; ymm6=XE=X(02468ACE********GIKMOQSU********)
+ vpxor ymm7, ymm7, ymm7 ; ymm7=XO=X(13579BDF********HJLNPRTV********)
+%endif
+ ; ymmA=(00 02 04 06 08 0A 0C 0E ** 0G 0I 0K 0M 0O 0Q 0S 0U **)
+ ; ymmB=(01 03 05 07 09 0B 0D 0F ** 0H 0J 0L 0N 0P 0R 0T 0V **)
+ ; ymmC=(10 12 14 16 18 1A 1C 1E ** 1G 1I 1K 1M 1O 1Q 1S 1U **)
+ ; ymmD=(11 13 15 17 19 1B 1D 1F ** 1H 1J 1L 1N 1P 1R 1T 1V **)
+ ; ymmE=(20 22 24 26 28 2A 2C 2E ** 2G 2I 2K 2M 2O 2Q 2S 2U **)
+ ; ymmF=(21 23 25 27 29 2B 2D 2F ** 2H 2J 2L 2N 2P 2R 2T 2V **)
+ ; ymmG=(30 32 34 36 38 3A 3C 3E ** 3G 3I 3K 3M 3O 3Q 3S 3U **)
+ ; ymmH=(31 33 35 37 39 3B 3D 3F ** 3H 3J 3L 3N 3P 3R 3T 3V **)
+
+ vpunpcklbw ymmA, ymmA, ymmC ; ymmA=(00 10 02 12 04 14 06 16 08 18 0A 1A 0C 1C 0E 1E
+ ; 0G 1G 0I 1I 0K 1K 0M 1M 0O 1O 0Q 1Q 0S 1S 0U 1U)
+ vpunpcklbw ymmE, ymmE, ymmG ; ymmE=(20 30 22 32 24 34 26 36 28 38 2A 3A 2C 3C 2E 3E
+ ; 2G 3G 2I 3I 2K 3K 2M 3M 2O 3O 2Q 3Q 2S 3S 2U 3U)
+ vpunpcklbw ymmB, ymmB, ymmD ; ymmB=(01 11 03 13 05 15 07 17 09 19 0B 1B 0D 1D 0F 1F
+ ; 0H 1H 0J 1J 0L 1L 0N 1N 0P 1P 0R 1R 0T 1T 0V 1V)
+ vpunpcklbw ymmF, ymmF, ymmH ; ymmF=(21 31 23 33 25 35 27 37 29 39 2B 3B 2D 3D 2F 3F
+ ; 2H 3H 2J 3J 2L 3L 2N 3N 2P 3P 2R 3R 2T 3T 2V 3V)
+
+ vpunpckhwd ymmC, ymmA, ymmE ; ymmC=(08 18 28 38 0A 1A 2A 3A 0C 1C 2C 3C 0E 1E 2E 3E
+ ; 0O 1O 2O 3O 0Q 1Q 2Q 3Q 0S 1S 2S 3S 0U 1U 2U 3U)
+ vpunpcklwd ymmA, ymmA, ymmE ; ymmA=(00 10 20 30 02 12 22 32 04 14 24 34 06 16 26 36
+ ; 0G 1G 2G 3G 0I 1I 2I 3I 0K 1K 2K 3K 0M 1M 2M 3M)
+ vpunpckhwd ymmG, ymmB, ymmF ; ymmG=(09 19 29 39 0B 1B 2B 3B 0D 1D 2D 3D 0F 1F 2F 3F
+ ; 0P 1P 2P 3P 0R 1R 2R 3R 0T 1T 2T 3T 0V 1V 2V 3V)
+ vpunpcklwd ymmB, ymmB, ymmF ; ymmB=(01 11 21 31 03 13 23 33 05 15 25 35 07 17 27 37
+ ; 0H 1H 2H 3H 0J 1J 2J 3J 0L 1L 2L 3L 0N 1N 2N 3N)
+
+ vpunpckhdq ymmE, ymmA, ymmB ; ymmE=(04 14 24 34 05 15 25 35 06 16 26 36 07 17 27 37
+ ; 0K 1K 2K 3K 0L 1L 2L 3L 0M 1M 2M 3M 0N 1N 2N 3N)
+ vpunpckldq ymmB, ymmA, ymmB ; ymmB=(00 10 20 30 01 11 21 31 02 12 22 32 03 13 23 33
+ ; 0G 1G 2G 3G 0H 1H 2H 3H 0I 1I 2I 3I 0J 1J 2J 3J)
+ vpunpckhdq ymmF, ymmC, ymmG ; ymmF=(0C 1C 2C 3C 0D 1D 2D 3D 0E 1E 2E 3E 0F 1F 2F 3F
+ ; 0S 1S 2S 3S 0T 1T 2T 3T 0U 1U 2U 3U 0V 1V 2V 3V)
+ vpunpckldq ymmG, ymmC, ymmG ; ymmG=(08 18 28 38 09 19 29 39 0A 1A 2A 3A 0B 1B 2B 3B
+ ; 0O 1O 2O 3O 0P 1P 2P 3P 0Q 1Q 2Q 3Q 0R 1R 2R 3R)
+
+ vperm2i128 ymmA, ymmB, ymmE, 0x20 ; ymmA=(00 10 20 30 01 11 21 31 02 12 22 32 03 13 23 33
+ ; 04 14 24 34 05 15 25 35 06 16 26 36 07 17 27 37)
+ vperm2i128 ymmD, ymmG, ymmF, 0x20 ; ymmD=(08 18 28 38 09 19 29 39 0A 1A 2A 3A 0B 1B 2B 3B
+ ; 0C 1C 2C 3C 0D 1D 2D 3D 0E 1E 2E 3E 0F 1F 2F 3F)
+ vperm2i128 ymmC, ymmB, ymmE, 0x31 ; ymmC=(0G 1G 2G 3G 0H 1H 2H 3H 0I 1I 2I 3I 0J 1J 2J 3J
+ ; 0K 1K 2K 3K 0L 1L 2L 3L 0M 1M 2M 3M 0N 1N 2N 3N)
+ vperm2i128 ymmH, ymmG, ymmF, 0x31 ; ymmH=(0O 1O 2O 3O 0P 1P 2P 3P 0Q 1Q 2Q 3Q 0R 1R 2R 3R
+ ; 0S 1S 2S 3S 0T 1T 2T 3T 0U 1U 2U 3U 0V 1V 2V 3V)
+
+ cmp ecx, byte SIZEOF_YMMWORD
+ jb short .column_st64
+
+ test edi, SIZEOF_YMMWORD-1
+ jnz short .out1
+ ; --(aligned)-------------------
+ vmovntdq YMMWORD [edi+0*SIZEOF_YMMWORD], ymmA
+ vmovntdq YMMWORD [edi+1*SIZEOF_YMMWORD], ymmD
+ vmovntdq YMMWORD [edi+2*SIZEOF_YMMWORD], ymmC
+ vmovntdq YMMWORD [edi+3*SIZEOF_YMMWORD], ymmH
+ jmp short .out0
+.out1: ; --(unaligned)-----------------
+ vmovdqu YMMWORD [edi+0*SIZEOF_YMMWORD], ymmA
+ vmovdqu YMMWORD [edi+1*SIZEOF_YMMWORD], ymmD
+ vmovdqu YMMWORD [edi+2*SIZEOF_YMMWORD], ymmC
+ vmovdqu YMMWORD [edi+3*SIZEOF_YMMWORD], ymmH
+.out0:
+ add edi, RGB_PIXELSIZE*SIZEOF_YMMWORD ; outptr
+ sub ecx, byte SIZEOF_YMMWORD
+ jz near .endcolumn
+
+ add esi, byte SIZEOF_YMMWORD ; inptr0
+ dec al
+ jnz near .Yloop_2nd
+
+ add ebx, byte SIZEOF_YMMWORD ; inptr1
+ add edx, byte SIZEOF_YMMWORD ; inptr2
+ jmp near .columnloop
+ alignx 16, 7
+
+.column_st64:
+ cmp ecx, byte SIZEOF_YMMWORD/2
+ jb short .column_st32
+ vmovdqu YMMWORD [edi+0*SIZEOF_YMMWORD], ymmA
+ vmovdqu YMMWORD [edi+1*SIZEOF_YMMWORD], ymmD
+ add edi, byte 2*SIZEOF_YMMWORD ; outptr
+ vmovdqa ymmA, ymmC
+ vmovdqa ymmD, ymmH
+ sub ecx, byte SIZEOF_YMMWORD/2
+.column_st32:
+ cmp ecx, byte SIZEOF_YMMWORD/4
+ jb short .column_st16
+ vmovdqu YMMWORD [edi+0*SIZEOF_YMMWORD], ymmA
+ add edi, byte SIZEOF_YMMWORD ; outptr
+ vmovdqa ymmA, ymmD
+ sub ecx, byte SIZEOF_YMMWORD/4
+.column_st16:
+ cmp ecx, byte SIZEOF_YMMWORD/8
+ jb short .column_st15
+ vmovdqu XMMWORD [edi+0*SIZEOF_XMMWORD], xmmA
+ add edi, byte SIZEOF_XMMWORD ; outptr
+ vperm2i128 ymmA, ymmA, ymmA, 1
+ sub ecx, byte SIZEOF_YMMWORD/8
+.column_st15:
+ ; Store two pixels (8 bytes) of ymmA to the output when it has enough
+ ; space.
+ cmp ecx, byte SIZEOF_YMMWORD/16
+ jb short .column_st7
+ vmovq MMWORD [edi], xmmA
+ add edi, byte SIZEOF_YMMWORD/16*4
+ sub ecx, byte SIZEOF_YMMWORD/16
+ vpsrldq xmmA, SIZEOF_YMMWORD/16*4
+.column_st7:
+ ; Store one pixel (4 bytes) of ymmA to the output when it has enough
+ ; space.
+ test ecx, ecx
+ jz short .endcolumn
+ vmovd XMM_DWORD [edi], xmmA
+
+%endif ; RGB_PIXELSIZE ; ---------------
+
+.endcolumn:
+ sfence ; flush the write buffer
+
+.return:
+ vzeroupper
+ pop edi
+ pop esi
+; pop edx ; need not be preserved
+; pop ecx ; need not be preserved
+ pop ebx
+ mov esp, ebp ; esp <- aligned ebp
+ pop esp ; esp <- original ebp
+ pop ebp
+ ret
+
+; --------------------------------------------------------------------------
+;
+; Upsample and color convert for the case of 2:1 horizontal and 2:1 vertical.
+;
+; GLOBAL(void)
+; jsimd_h2v2_merged_upsample_avx2 (JDIMENSION output_width,
+; JSAMPIMAGE input_buf,
+; JDIMENSION in_row_group_ctr,
+; JSAMPARRAY output_buf);
+;
+
+%define output_width(b) (b)+8 ; JDIMENSION output_width
+%define input_buf(b) (b)+12 ; JSAMPIMAGE input_buf
+%define in_row_group_ctr(b) (b)+16 ; JDIMENSION in_row_group_ctr
+%define output_buf(b) (b)+20 ; JSAMPARRAY output_buf
+
+ align 32
+ global EXTN(jsimd_h2v2_merged_upsample_avx2)
+
+EXTN(jsimd_h2v2_merged_upsample_avx2):
+ push ebp
+ mov ebp, esp
+ push ebx
+; push ecx ; need not be preserved
+; push edx ; need not be preserved
+ push esi
+ push edi
+
+ mov eax, POINTER [output_width(ebp)]
+
+ mov edi, JSAMPIMAGE [input_buf(ebp)]
+ mov ecx, JDIMENSION [in_row_group_ctr(ebp)]
+ mov esi, JSAMPARRAY [edi+0*SIZEOF_JSAMPARRAY]
+ mov ebx, JSAMPARRAY [edi+1*SIZEOF_JSAMPARRAY]
+ mov edx, JSAMPARRAY [edi+2*SIZEOF_JSAMPARRAY]
+ mov edi, JSAMPARRAY [output_buf(ebp)]
+ lea esi, [esi+ecx*SIZEOF_JSAMPROW]
+
+ push edx ; inptr2
+ push ebx ; inptr1
+ push esi ; inptr00
+ mov ebx, esp
+
+ push edi ; output_buf (outptr0)
+ push ecx ; in_row_group_ctr
+ push ebx ; input_buf
+ push eax ; output_width
+
+ call near EXTN(jsimd_h2v1_merged_upsample_avx2)
+
+ add esi, byte SIZEOF_JSAMPROW ; inptr01
+ add edi, byte SIZEOF_JSAMPROW ; outptr1
+ mov POINTER [ebx+0*SIZEOF_POINTER], esi
+ mov POINTER [ebx-1*SIZEOF_POINTER], edi
+
+ call near EXTN(jsimd_h2v1_merged_upsample_avx2)
+
+ add esp, byte 7*SIZEOF_DWORD
+
+ pop edi
+ pop esi
+; pop edx ; need not be preserved
+; pop ecx ; need not be preserved
+ pop ebx
+ pop ebp
+ ret
+
+; For some reason, the OS X linker does not honor the request to align the
+; segment unless we do this.
+ align 32
diff --git a/simd/jdmrgext-mmx.asm b/simd/i386/jdmrgext-mmx.asm
similarity index 99%
rename from simd/jdmrgext-mmx.asm
rename to simd/i386/jdmrgext-mmx.asm
index 63f45cf..0444632 100644
--- a/simd/jdmrgext-mmx.asm
+++ b/simd/i386/jdmrgext-mmx.asm
@@ -2,6 +2,7 @@
; jdmrgext.asm - merged upsampling/color conversion (MMX)
;
; Copyright 2009 Pierre Ossman <ossman@cendio.se> for Cendio AB
+; Copyright (C) 2016, D. R. Commander.
;
; Based on the x86 SIMD extension for IJG JPEG library
; Copyright (C) 1999-2006, MIYASAKA Masaru.
@@ -38,7 +39,7 @@
%define WK_NUM 3
%define gotptr wk(0)-SIZEOF_POINTER ; void * gotptr
- align 16
+ align 32
global EXTN(jsimd_h2v1_merged_upsample_mmx)
EXTN(jsimd_h2v1_merged_upsample_mmx):
@@ -407,7 +408,7 @@
%define in_row_group_ctr(b) (b)+16 ; JDIMENSION in_row_group_ctr
%define output_buf(b) (b)+20 ; JSAMPARRAY output_buf
- align 16
+ align 32
global EXTN(jsimd_h2v2_merged_upsample_mmx)
EXTN(jsimd_h2v2_merged_upsample_mmx):
@@ -460,4 +461,4 @@
; For some reason, the OS X linker does not honor the request to align the
; segment unless we do this.
- align 16
+ align 32
diff --git a/simd/i386/jdmrgext-sse2.asm b/simd/i386/jdmrgext-sse2.asm
new file mode 100644
index 0000000..50702c6
--- /dev/null
+++ b/simd/i386/jdmrgext-sse2.asm
@@ -0,0 +1,518 @@
+;
+; jdmrgext.asm - merged upsampling/color conversion (SSE2)
+;
+; Copyright 2009, 2012 Pierre Ossman <ossman@cendio.se> for Cendio AB
+; Copyright (C) 2012, 2016, D. R. Commander.
+;
+; Based on the x86 SIMD extension for IJG JPEG library
+; Copyright (C) 1999-2006, MIYASAKA Masaru.
+; For conditions of distribution and use, see copyright notice in jsimdext.inc
+;
+; This file should be assembled with NASM (Netwide Assembler),
+; can *not* be assembled with Microsoft's MASM or any compatible
+; assembler (including Borland's Turbo Assembler).
+; NASM is available from http://nasm.sourceforge.net/ or
+; http://sourceforge.net/project/showfiles.php?group_id=6208
+;
+; [TAB8]
+
+%include "jcolsamp.inc"
+
+; --------------------------------------------------------------------------
+;
+; Upsample and color convert for the case of 2:1 horizontal and 1:1 vertical.
+;
+; GLOBAL(void)
+; jsimd_h2v1_merged_upsample_sse2 (JDIMENSION output_width,
+; JSAMPIMAGE input_buf,
+; JDIMENSION in_row_group_ctr,
+; JSAMPARRAY output_buf);
+;
+
+%define output_width(b) (b)+8 ; JDIMENSION output_width
+%define input_buf(b) (b)+12 ; JSAMPIMAGE input_buf
+%define in_row_group_ctr(b) (b)+16 ; JDIMENSION in_row_group_ctr
+%define output_buf(b) (b)+20 ; JSAMPARRAY output_buf
+
+%define original_ebp ebp+0
+%define wk(i) ebp-(WK_NUM-(i))*SIZEOF_XMMWORD ; xmmword wk[WK_NUM]
+%define WK_NUM 3
+%define gotptr wk(0)-SIZEOF_POINTER ; void * gotptr
+
+ align 32
+ global EXTN(jsimd_h2v1_merged_upsample_sse2)
+
+EXTN(jsimd_h2v1_merged_upsample_sse2):
+ push ebp
+ mov eax, esp ; eax = original ebp
+ sub esp, byte 4
+ and esp, byte (-SIZEOF_XMMWORD) ; align to 128 bits
+ mov [esp], eax
+ mov ebp, esp ; ebp = aligned ebp
+ lea esp, [wk(0)]
+ pushpic eax ; make a room for GOT address
+ push ebx
+; push ecx ; need not be preserved
+; push edx ; need not be preserved
+ push esi
+ push edi
+
+ get_GOT ebx ; get GOT address
+ movpic POINTER [gotptr], ebx ; save GOT address
+
+ mov ecx, JDIMENSION [output_width(eax)] ; col
+ test ecx, ecx
+ jz near .return
+
+ push ecx
+
+ mov edi, JSAMPIMAGE [input_buf(eax)]
+ mov ecx, JDIMENSION [in_row_group_ctr(eax)]
+ mov esi, JSAMPARRAY [edi+0*SIZEOF_JSAMPARRAY]
+ mov ebx, JSAMPARRAY [edi+1*SIZEOF_JSAMPARRAY]
+ mov edx, JSAMPARRAY [edi+2*SIZEOF_JSAMPARRAY]
+ mov edi, JSAMPARRAY [output_buf(eax)]
+ mov esi, JSAMPROW [esi+ecx*SIZEOF_JSAMPROW] ; inptr0
+ mov ebx, JSAMPROW [ebx+ecx*SIZEOF_JSAMPROW] ; inptr1
+ mov edx, JSAMPROW [edx+ecx*SIZEOF_JSAMPROW] ; inptr2
+ mov edi, JSAMPROW [edi] ; outptr
+
+ pop ecx ; col
+
+ alignx 16, 7
+.columnloop:
+ movpic eax, POINTER [gotptr] ; load GOT address (eax)
+
+ movdqa xmm6, XMMWORD [ebx] ; xmm6=Cb(0123456789ABCDEF)
+ movdqa xmm7, XMMWORD [edx] ; xmm7=Cr(0123456789ABCDEF)
+
+ pxor xmm1, xmm1 ; xmm1=(all 0's)
+ pcmpeqw xmm3, xmm3
+ psllw xmm3, 7 ; xmm3={0xFF80 0xFF80 0xFF80 0xFF80 ..}
+
+ movdqa xmm4, xmm6
+ punpckhbw xmm6, xmm1 ; xmm6=Cb(89ABCDEF)=CbH
+ punpcklbw xmm4, xmm1 ; xmm4=Cb(01234567)=CbL
+ movdqa xmm0, xmm7
+ punpckhbw xmm7, xmm1 ; xmm7=Cr(89ABCDEF)=CrH
+ punpcklbw xmm0, xmm1 ; xmm0=Cr(01234567)=CrL
+
+ paddw xmm6, xmm3
+ paddw xmm4, xmm3
+ paddw xmm7, xmm3
+ paddw xmm0, xmm3
+
+ ; (Original)
+ ; R = Y + 1.40200 * Cr
+ ; G = Y - 0.34414 * Cb - 0.71414 * Cr
+ ; B = Y + 1.77200 * Cb
+ ;
+ ; (This implementation)
+ ; R = Y + 0.40200 * Cr + Cr
+ ; G = Y - 0.34414 * Cb + 0.28586 * Cr - Cr
+ ; B = Y - 0.22800 * Cb + Cb + Cb
+
+ movdqa xmm5, xmm6 ; xmm5=CbH
+ movdqa xmm2, xmm4 ; xmm2=CbL
+ paddw xmm6, xmm6 ; xmm6=2*CbH
+ paddw xmm4, xmm4 ; xmm4=2*CbL
+ movdqa xmm1, xmm7 ; xmm1=CrH
+ movdqa xmm3, xmm0 ; xmm3=CrL
+ paddw xmm7, xmm7 ; xmm7=2*CrH
+ paddw xmm0, xmm0 ; xmm0=2*CrL
+
+ pmulhw xmm6, [GOTOFF(eax,PW_MF0228)] ; xmm6=(2*CbH * -FIX(0.22800))
+ pmulhw xmm4, [GOTOFF(eax,PW_MF0228)] ; xmm4=(2*CbL * -FIX(0.22800))
+ pmulhw xmm7, [GOTOFF(eax,PW_F0402)] ; xmm7=(2*CrH * FIX(0.40200))
+ pmulhw xmm0, [GOTOFF(eax,PW_F0402)] ; xmm0=(2*CrL * FIX(0.40200))
+
+ paddw xmm6, [GOTOFF(eax,PW_ONE)]
+ paddw xmm4, [GOTOFF(eax,PW_ONE)]
+ psraw xmm6, 1 ; xmm6=(CbH * -FIX(0.22800))
+ psraw xmm4, 1 ; xmm4=(CbL * -FIX(0.22800))
+ paddw xmm7, [GOTOFF(eax,PW_ONE)]
+ paddw xmm0, [GOTOFF(eax,PW_ONE)]
+ psraw xmm7, 1 ; xmm7=(CrH * FIX(0.40200))
+ psraw xmm0, 1 ; xmm0=(CrL * FIX(0.40200))
+
+ paddw xmm6, xmm5
+ paddw xmm4, xmm2
+ paddw xmm6, xmm5 ; xmm6=(CbH * FIX(1.77200))=(B-Y)H
+ paddw xmm4, xmm2 ; xmm4=(CbL * FIX(1.77200))=(B-Y)L
+ paddw xmm7, xmm1 ; xmm7=(CrH * FIX(1.40200))=(R-Y)H
+ paddw xmm0, xmm3 ; xmm0=(CrL * FIX(1.40200))=(R-Y)L
+
+ movdqa XMMWORD [wk(0)], xmm6 ; wk(0)=(B-Y)H
+ movdqa XMMWORD [wk(1)], xmm7 ; wk(1)=(R-Y)H
+
+ movdqa xmm6, xmm5
+ movdqa xmm7, xmm2
+ punpcklwd xmm5, xmm1
+ punpckhwd xmm6, xmm1
+ pmaddwd xmm5, [GOTOFF(eax,PW_MF0344_F0285)]
+ pmaddwd xmm6, [GOTOFF(eax,PW_MF0344_F0285)]
+ punpcklwd xmm2, xmm3
+ punpckhwd xmm7, xmm3
+ pmaddwd xmm2, [GOTOFF(eax,PW_MF0344_F0285)]
+ pmaddwd xmm7, [GOTOFF(eax,PW_MF0344_F0285)]
+
+ paddd xmm5, [GOTOFF(eax,PD_ONEHALF)]
+ paddd xmm6, [GOTOFF(eax,PD_ONEHALF)]
+ psrad xmm5, SCALEBITS
+ psrad xmm6, SCALEBITS
+ paddd xmm2, [GOTOFF(eax,PD_ONEHALF)]
+ paddd xmm7, [GOTOFF(eax,PD_ONEHALF)]
+ psrad xmm2, SCALEBITS
+ psrad xmm7, SCALEBITS
+
+ packssdw xmm5, xmm6 ; xmm5=CbH*-FIX(0.344)+CrH*FIX(0.285)
+ packssdw xmm2, xmm7 ; xmm2=CbL*-FIX(0.344)+CrL*FIX(0.285)
+ psubw xmm5, xmm1 ; xmm5=CbH*-FIX(0.344)+CrH*-FIX(0.714)=(G-Y)H
+ psubw xmm2, xmm3 ; xmm2=CbL*-FIX(0.344)+CrL*-FIX(0.714)=(G-Y)L
+
+ movdqa XMMWORD [wk(2)], xmm5 ; wk(2)=(G-Y)H
+
+ mov al, 2 ; Yctr
+ jmp short .Yloop_1st
+ alignx 16, 7
+
+.Yloop_2nd:
+ movdqa xmm0, XMMWORD [wk(1)] ; xmm0=(R-Y)H
+ movdqa xmm2, XMMWORD [wk(2)] ; xmm2=(G-Y)H
+ movdqa xmm4, XMMWORD [wk(0)] ; xmm4=(B-Y)H
+ alignx 16, 7
+
+.Yloop_1st:
+ movdqa xmm7, XMMWORD [esi] ; xmm7=Y(0123456789ABCDEF)
+
+ pcmpeqw xmm6, xmm6
+ psrlw xmm6, BYTE_BIT ; xmm6={0xFF 0x00 0xFF 0x00 ..}
+ pand xmm6, xmm7 ; xmm6=Y(02468ACE)=YE
+ psrlw xmm7, BYTE_BIT ; xmm7=Y(13579BDF)=YO
+
+ movdqa xmm1, xmm0 ; xmm1=xmm0=(R-Y)(L/H)
+ movdqa xmm3, xmm2 ; xmm3=xmm2=(G-Y)(L/H)
+ movdqa xmm5, xmm4 ; xmm5=xmm4=(B-Y)(L/H)
+
+ paddw xmm0, xmm6 ; xmm0=((R-Y)+YE)=RE=R(02468ACE)
+ paddw xmm1, xmm7 ; xmm1=((R-Y)+YO)=RO=R(13579BDF)
+ packuswb xmm0, xmm0 ; xmm0=R(02468ACE********)
+ packuswb xmm1, xmm1 ; xmm1=R(13579BDF********)
+
+ paddw xmm2, xmm6 ; xmm2=((G-Y)+YE)=GE=G(02468ACE)
+ paddw xmm3, xmm7 ; xmm3=((G-Y)+YO)=GO=G(13579BDF)
+ packuswb xmm2, xmm2 ; xmm2=G(02468ACE********)
+ packuswb xmm3, xmm3 ; xmm3=G(13579BDF********)
+
+ paddw xmm4, xmm6 ; xmm4=((B-Y)+YE)=BE=B(02468ACE)
+ paddw xmm5, xmm7 ; xmm5=((B-Y)+YO)=BO=B(13579BDF)
+ packuswb xmm4, xmm4 ; xmm4=B(02468ACE********)
+ packuswb xmm5, xmm5 ; xmm5=B(13579BDF********)
+
+%if RGB_PIXELSIZE == 3 ; ---------------
+
+ ; xmmA=(00 02 04 06 08 0A 0C 0E **), xmmB=(01 03 05 07 09 0B 0D 0F **)
+ ; xmmC=(10 12 14 16 18 1A 1C 1E **), xmmD=(11 13 15 17 19 1B 1D 1F **)
+ ; xmmE=(20 22 24 26 28 2A 2C 2E **), xmmF=(21 23 25 27 29 2B 2D 2F **)
+ ; xmmG=(** ** ** ** ** ** ** ** **), xmmH=(** ** ** ** ** ** ** ** **)
+
+ punpcklbw xmmA, xmmC ; xmmA=(00 10 02 12 04 14 06 16 08 18 0A 1A 0C 1C 0E 1E)
+ punpcklbw xmmE, xmmB ; xmmE=(20 01 22 03 24 05 26 07 28 09 2A 0B 2C 0D 2E 0F)
+ punpcklbw xmmD, xmmF ; xmmD=(11 21 13 23 15 25 17 27 19 29 1B 2B 1D 2D 1F 2F)
+
+ movdqa xmmG, xmmA
+ movdqa xmmH, xmmA
+ punpcklwd xmmA, xmmE ; xmmA=(00 10 20 01 02 12 22 03 04 14 24 05 06 16 26 07)
+ punpckhwd xmmG, xmmE ; xmmG=(08 18 28 09 0A 1A 2A 0B 0C 1C 2C 0D 0E 1E 2E 0F)
+
+ psrldq xmmH, 2 ; xmmH=(02 12 04 14 06 16 08 18 0A 1A 0C 1C 0E 1E -- --)
+ psrldq xmmE, 2 ; xmmE=(22 03 24 05 26 07 28 09 2A 0B 2C 0D 2E 0F -- --)
+
+ movdqa xmmC, xmmD
+ movdqa xmmB, xmmD
+ punpcklwd xmmD, xmmH ; xmmD=(11 21 02 12 13 23 04 14 15 25 06 16 17 27 08 18)
+ punpckhwd xmmC, xmmH ; xmmC=(19 29 0A 1A 1B 2B 0C 1C 1D 2D 0E 1E 1F 2F -- --)
+
+ psrldq xmmB, 2 ; xmmB=(13 23 15 25 17 27 19 29 1B 2B 1D 2D 1F 2F -- --)
+
+ movdqa xmmF, xmmE
+ punpcklwd xmmE, xmmB ; xmmE=(22 03 13 23 24 05 15 25 26 07 17 27 28 09 19 29)
+ punpckhwd xmmF, xmmB ; xmmF=(2A 0B 1B 2B 2C 0D 1D 2D 2E 0F 1F 2F -- -- -- --)
+
+ pshufd xmmH, xmmA, 0x4E ; xmmH=(04 14 24 05 06 16 26 07 00 10 20 01 02 12 22 03)
+ movdqa xmmB, xmmE
+ punpckldq xmmA, xmmD ; xmmA=(00 10 20 01 11 21 02 12 02 12 22 03 13 23 04 14)
+ punpckldq xmmE, xmmH ; xmmE=(22 03 13 23 04 14 24 05 24 05 15 25 06 16 26 07)
+ punpckhdq xmmD, xmmB ; xmmD=(15 25 06 16 26 07 17 27 17 27 08 18 28 09 19 29)
+
+ pshufd xmmH, xmmG, 0x4E ; xmmH=(0C 1C 2C 0D 0E 1E 2E 0F 08 18 28 09 0A 1A 2A 0B)
+ movdqa xmmB, xmmF
+ punpckldq xmmG, xmmC ; xmmG=(08 18 28 09 19 29 0A 1A 0A 1A 2A 0B 1B 2B 0C 1C)
+ punpckldq xmmF, xmmH ; xmmF=(2A 0B 1B 2B 0C 1C 2C 0D 2C 0D 1D 2D 0E 1E 2E 0F)
+ punpckhdq xmmC, xmmB ; xmmC=(1D 2D 0E 1E 2E 0F 1F 2F 1F 2F -- -- -- -- -- --)
+
+ punpcklqdq xmmA, xmmE ; xmmA=(00 10 20 01 11 21 02 12 22 03 13 23 04 14 24 05)
+ punpcklqdq xmmD, xmmG ; xmmD=(15 25 06 16 26 07 17 27 08 18 28 09 19 29 0A 1A)
+ punpcklqdq xmmF, xmmC ; xmmF=(2A 0B 1B 2B 0C 1C 2C 0D 1D 2D 0E 1E 2E 0F 1F 2F)
+
+ cmp ecx, byte SIZEOF_XMMWORD
+ jb short .column_st32
+
+ test edi, SIZEOF_XMMWORD-1
+ jnz short .out1
+ ; --(aligned)-------------------
+ movntdq XMMWORD [edi+0*SIZEOF_XMMWORD], xmmA
+ movntdq XMMWORD [edi+1*SIZEOF_XMMWORD], xmmD
+ movntdq XMMWORD [edi+2*SIZEOF_XMMWORD], xmmF
+ jmp short .out0
+.out1: ; --(unaligned)-----------------
+ movdqu XMMWORD [edi+0*SIZEOF_XMMWORD], xmmA
+ movdqu XMMWORD [edi+1*SIZEOF_XMMWORD], xmmD
+ movdqu XMMWORD [edi+2*SIZEOF_XMMWORD], xmmF
+.out0:
+ add edi, byte RGB_PIXELSIZE*SIZEOF_XMMWORD ; outptr
+ sub ecx, byte SIZEOF_XMMWORD
+ jz near .endcolumn
+
+ add esi, byte SIZEOF_XMMWORD ; inptr0
+ dec al ; Yctr
+ jnz near .Yloop_2nd
+
+ add ebx, byte SIZEOF_XMMWORD ; inptr1
+ add edx, byte SIZEOF_XMMWORD ; inptr2
+ jmp near .columnloop
+ alignx 16, 7
+
+.column_st32:
+ lea ecx, [ecx+ecx*2] ; imul ecx, RGB_PIXELSIZE
+ cmp ecx, byte 2*SIZEOF_XMMWORD
+ jb short .column_st16
+ movdqu XMMWORD [edi+0*SIZEOF_XMMWORD], xmmA
+ movdqu XMMWORD [edi+1*SIZEOF_XMMWORD], xmmD
+ add edi, byte 2*SIZEOF_XMMWORD ; outptr
+ movdqa xmmA, xmmF
+ sub ecx, byte 2*SIZEOF_XMMWORD
+ jmp short .column_st15
+.column_st16:
+ cmp ecx, byte SIZEOF_XMMWORD
+ jb short .column_st15
+ movdqu XMMWORD [edi+0*SIZEOF_XMMWORD], xmmA
+ add edi, byte SIZEOF_XMMWORD ; outptr
+ movdqa xmmA, xmmD
+ sub ecx, byte SIZEOF_XMMWORD
+.column_st15:
+ ; Store the lower 8 bytes of xmmA to the output when it has enough
+ ; space.
+ cmp ecx, byte SIZEOF_MMWORD
+ jb short .column_st7
+ movq XMM_MMWORD [edi], xmmA
+ add edi, byte SIZEOF_MMWORD
+ sub ecx, byte SIZEOF_MMWORD
+ psrldq xmmA, SIZEOF_MMWORD
+.column_st7:
+ ; Store the lower 4 bytes of xmmA to the output when it has enough
+ ; space.
+ cmp ecx, byte SIZEOF_DWORD
+ jb short .column_st3
+ movd XMM_DWORD [edi], xmmA
+ add edi, byte SIZEOF_DWORD
+ sub ecx, byte SIZEOF_DWORD
+ psrldq xmmA, SIZEOF_DWORD
+.column_st3:
+ ; Store the lower 2 bytes of eax to the output when it has enough
+ ; space.
+ movd eax, xmmA
+ cmp ecx, byte SIZEOF_WORD
+ jb short .column_st1
+ mov WORD [edi], ax
+ add edi, byte SIZEOF_WORD
+ sub ecx, byte SIZEOF_WORD
+ shr eax, 16
+.column_st1:
+ ; Store the lower 1 byte of eax to the output when it has enough
+ ; space.
+ test ecx, ecx
+ jz short .endcolumn
+ mov BYTE [edi], al
+
+%else ; RGB_PIXELSIZE == 4 ; -----------
+
+%ifdef RGBX_FILLER_0XFF
+ pcmpeqb xmm6, xmm6 ; xmm6=XE=X(02468ACE********)
+ pcmpeqb xmm7, xmm7 ; xmm7=XO=X(13579BDF********)
+%else
+ pxor xmm6, xmm6 ; xmm6=XE=X(02468ACE********)
+ pxor xmm7, xmm7 ; xmm7=XO=X(13579BDF********)
+%endif
+ ; xmmA=(00 02 04 06 08 0A 0C 0E **), xmmB=(01 03 05 07 09 0B 0D 0F **)
+ ; xmmC=(10 12 14 16 18 1A 1C 1E **), xmmD=(11 13 15 17 19 1B 1D 1F **)
+ ; xmmE=(20 22 24 26 28 2A 2C 2E **), xmmF=(21 23 25 27 29 2B 2D 2F **)
+ ; xmmG=(30 32 34 36 38 3A 3C 3E **), xmmH=(31 33 35 37 39 3B 3D 3F **)
+
+ punpcklbw xmmA, xmmC ; xmmA=(00 10 02 12 04 14 06 16 08 18 0A 1A 0C 1C 0E 1E)
+ punpcklbw xmmE, xmmG ; xmmE=(20 30 22 32 24 34 26 36 28 38 2A 3A 2C 3C 2E 3E)
+ punpcklbw xmmB, xmmD ; xmmB=(01 11 03 13 05 15 07 17 09 19 0B 1B 0D 1D 0F 1F)
+ punpcklbw xmmF, xmmH ; xmmF=(21 31 23 33 25 35 27 37 29 39 2B 3B 2D 3D 2F 3F)
+
+ movdqa xmmC, xmmA
+ punpcklwd xmmA, xmmE ; xmmA=(00 10 20 30 02 12 22 32 04 14 24 34 06 16 26 36)
+ punpckhwd xmmC, xmmE ; xmmC=(08 18 28 38 0A 1A 2A 3A 0C 1C 2C 3C 0E 1E 2E 3E)
+ movdqa xmmG, xmmB
+ punpcklwd xmmB, xmmF ; xmmB=(01 11 21 31 03 13 23 33 05 15 25 35 07 17 27 37)
+ punpckhwd xmmG, xmmF ; xmmG=(09 19 29 39 0B 1B 2B 3B 0D 1D 2D 3D 0F 1F 2F 3F)
+
+ movdqa xmmD, xmmA
+ punpckldq xmmA, xmmB ; xmmA=(00 10 20 30 01 11 21 31 02 12 22 32 03 13 23 33)
+ punpckhdq xmmD, xmmB ; xmmD=(04 14 24 34 05 15 25 35 06 16 26 36 07 17 27 37)
+ movdqa xmmH, xmmC
+ punpckldq xmmC, xmmG ; xmmC=(08 18 28 38 09 19 29 39 0A 1A 2A 3A 0B 1B 2B 3B)
+ punpckhdq xmmH, xmmG ; xmmH=(0C 1C 2C 3C 0D 1D 2D 3D 0E 1E 2E 3E 0F 1F 2F 3F)
+
+ cmp ecx, byte SIZEOF_XMMWORD
+ jb short .column_st32
+
+ test edi, SIZEOF_XMMWORD-1
+ jnz short .out1
+ ; --(aligned)-------------------
+ movntdq XMMWORD [edi+0*SIZEOF_XMMWORD], xmmA
+ movntdq XMMWORD [edi+1*SIZEOF_XMMWORD], xmmD
+ movntdq XMMWORD [edi+2*SIZEOF_XMMWORD], xmmC
+ movntdq XMMWORD [edi+3*SIZEOF_XMMWORD], xmmH
+ jmp short .out0
+.out1: ; --(unaligned)-----------------
+ movdqu XMMWORD [edi+0*SIZEOF_XMMWORD], xmmA
+ movdqu XMMWORD [edi+1*SIZEOF_XMMWORD], xmmD
+ movdqu XMMWORD [edi+2*SIZEOF_XMMWORD], xmmC
+ movdqu XMMWORD [edi+3*SIZEOF_XMMWORD], xmmH
+.out0:
+ add edi, byte RGB_PIXELSIZE*SIZEOF_XMMWORD ; outptr
+ sub ecx, byte SIZEOF_XMMWORD
+ jz near .endcolumn
+
+ add esi, byte SIZEOF_XMMWORD ; inptr0
+ dec al ; Yctr
+ jnz near .Yloop_2nd
+
+ add ebx, byte SIZEOF_XMMWORD ; inptr1
+ add edx, byte SIZEOF_XMMWORD ; inptr2
+ jmp near .columnloop
+ alignx 16, 7
+
+.column_st32:
+ cmp ecx, byte SIZEOF_XMMWORD/2
+ jb short .column_st16
+ movdqu XMMWORD [edi+0*SIZEOF_XMMWORD], xmmA
+ movdqu XMMWORD [edi+1*SIZEOF_XMMWORD], xmmD
+ add edi, byte 2*SIZEOF_XMMWORD ; outptr
+ movdqa xmmA, xmmC
+ movdqa xmmD, xmmH
+ sub ecx, byte SIZEOF_XMMWORD/2
+.column_st16:
+ cmp ecx, byte SIZEOF_XMMWORD/4
+ jb short .column_st15
+ movdqu XMMWORD [edi+0*SIZEOF_XMMWORD], xmmA
+ add edi, byte SIZEOF_XMMWORD ; outptr
+ movdqa xmmA, xmmD
+ sub ecx, byte SIZEOF_XMMWORD/4
+.column_st15:
+ ; Store two pixels (8 bytes) of xmmA to the output when it has enough
+ ; space.
+ cmp ecx, byte SIZEOF_XMMWORD/8
+ jb short .column_st7
+ movq XMM_MMWORD [edi], xmmA
+ add edi, byte SIZEOF_XMMWORD/8*4
+ sub ecx, byte SIZEOF_XMMWORD/8
+ psrldq xmmA, SIZEOF_XMMWORD/8*4
+.column_st7:
+ ; Store one pixel (4 bytes) of xmmA to the output when it has enough
+ ; space.
+ test ecx, ecx
+ jz short .endcolumn
+ movd XMM_DWORD [edi], xmmA
+
+%endif ; RGB_PIXELSIZE ; ---------------
+
+.endcolumn:
+ sfence ; flush the write buffer
+
+.return:
+ pop edi
+ pop esi
+; pop edx ; need not be preserved
+; pop ecx ; need not be preserved
+ pop ebx
+ mov esp, ebp ; esp <- aligned ebp
+ pop esp ; esp <- original ebp
+ pop ebp
+ ret
+
+; --------------------------------------------------------------------------
+;
+; Upsample and color convert for the case of 2:1 horizontal and 2:1 vertical.
+;
+; GLOBAL(void)
+; jsimd_h2v2_merged_upsample_sse2 (JDIMENSION output_width,
+; JSAMPIMAGE input_buf,
+; JDIMENSION in_row_group_ctr,
+; JSAMPARRAY output_buf);
+;
+
+%define output_width(b) (b)+8 ; JDIMENSION output_width
+%define input_buf(b) (b)+12 ; JSAMPIMAGE input_buf
+%define in_row_group_ctr(b) (b)+16 ; JDIMENSION in_row_group_ctr
+%define output_buf(b) (b)+20 ; JSAMPARRAY output_buf
+
+ align 32
+ global EXTN(jsimd_h2v2_merged_upsample_sse2)
+
+EXTN(jsimd_h2v2_merged_upsample_sse2):
+ push ebp
+ mov ebp, esp
+ push ebx
+; push ecx ; need not be preserved
+; push edx ; need not be preserved
+ push esi
+ push edi
+
+ mov eax, POINTER [output_width(ebp)]
+
+ mov edi, JSAMPIMAGE [input_buf(ebp)]
+ mov ecx, JDIMENSION [in_row_group_ctr(ebp)]
+ mov esi, JSAMPARRAY [edi+0*SIZEOF_JSAMPARRAY]
+ mov ebx, JSAMPARRAY [edi+1*SIZEOF_JSAMPARRAY]
+ mov edx, JSAMPARRAY [edi+2*SIZEOF_JSAMPARRAY]
+ mov edi, JSAMPARRAY [output_buf(ebp)]
+ lea esi, [esi+ecx*SIZEOF_JSAMPROW]
+
+ push edx ; inptr2
+ push ebx ; inptr1
+ push esi ; inptr00
+ mov ebx, esp
+
+ push edi ; output_buf (outptr0)
+ push ecx ; in_row_group_ctr
+ push ebx ; input_buf
+ push eax ; output_width
+
+ call near EXTN(jsimd_h2v1_merged_upsample_sse2)
+
+ add esi, byte SIZEOF_JSAMPROW ; inptr01
+ add edi, byte SIZEOF_JSAMPROW ; outptr1
+ mov POINTER [ebx+0*SIZEOF_POINTER], esi
+ mov POINTER [ebx-1*SIZEOF_POINTER], edi
+
+ call near EXTN(jsimd_h2v1_merged_upsample_sse2)
+
+ add esp, byte 7*SIZEOF_DWORD
+
+ pop edi
+ pop esi
+; pop edx ; need not be preserved
+; pop ecx ; need not be preserved
+ pop ebx
+ pop ebp
+ ret
+
+; For some reason, the OS X linker does not honor the request to align the
+; segment unless we do this.
+ align 32
diff --git a/simd/i386/jdsample-avx2.asm b/simd/i386/jdsample-avx2.asm
new file mode 100644
index 0000000..d087e3d
--- /dev/null
+++ b/simd/i386/jdsample-avx2.asm
@@ -0,0 +1,765 @@
+;
+; jdsample.asm - upsampling (AVX2)
+;
+; Copyright 2009 Pierre Ossman <ossman@cendio.se> for Cendio AB
+; Copyright (C) 2015, Intel Corporation.
+; Copyright (C) 2016, D. R. Commander.
+;
+; Based on the x86 SIMD extension for IJG JPEG library
+; Copyright (C) 1999-2006, MIYASAKA Masaru.
+; For conditions of distribution and use, see copyright notice in jsimdext.inc
+;
+; This file should be assembled with NASM (Netwide Assembler),
+; can *not* be assembled with Microsoft's MASM or any compatible
+; assembler (including Borland's Turbo Assembler).
+; NASM is available from http://nasm.sourceforge.net/ or
+; http://sourceforge.net/project/showfiles.php?group_id=6208
+;
+; [TAB8]
+
+%include "jsimdext.inc"
+
+; --------------------------------------------------------------------------
+ SECTION SEG_CONST
+
+ alignz 32
+ global EXTN(jconst_fancy_upsample_avx2)
+
+EXTN(jconst_fancy_upsample_avx2):
+
+PW_ONE times 16 dw 1
+PW_TWO times 16 dw 2
+PW_THREE times 16 dw 3
+PW_SEVEN times 16 dw 7
+PW_EIGHT times 16 dw 8
+
+ alignz 32
+
+; --------------------------------------------------------------------------
+ SECTION SEG_TEXT
+ BITS 32
+;
+; Fancy processing for the common case of 2:1 horizontal and 1:1 vertical.
+;
+; The upsampling algorithm is linear interpolation between pixel centers,
+; also known as a "triangle filter". This is a good compromise between
+; speed and visual quality. The centers of the output pixels are 1/4 and 3/4
+; of the way between input pixel centers.
+;
+; GLOBAL(void)
+; jsimd_h2v1_fancy_upsample_avx2 (int max_v_samp_factor,
+; JDIMENSION downsampled_width,
+; JSAMPARRAY input_data,
+; JSAMPARRAY *output_data_ptr);
+;
+
+%define max_v_samp(b) (b)+8 ; int max_v_samp_factor
+%define downsamp_width(b) (b)+12 ; JDIMENSION downsampled_width
+%define input_data(b) (b)+16 ; JSAMPARRAY input_data
+%define output_data_ptr(b) (b)+20 ; JSAMPARRAY *output_data_ptr
+
+ align 32
+ global EXTN(jsimd_h2v1_fancy_upsample_avx2)
+
+EXTN(jsimd_h2v1_fancy_upsample_avx2):
+ push ebp
+ mov ebp, esp
+ pushpic ebx
+; push ecx ; need not be preserved
+; push edx ; need not be preserved
+ push esi
+ push edi
+
+ get_GOT ebx ; get GOT address
+
+ mov eax, JDIMENSION [downsamp_width(ebp)] ; colctr
+ test eax, eax
+ jz near .return
+
+ mov ecx, INT [max_v_samp(ebp)] ; rowctr
+ test ecx, ecx
+ jz near .return
+
+ mov esi, JSAMPARRAY [input_data(ebp)] ; input_data
+ mov edi, POINTER [output_data_ptr(ebp)]
+ mov edi, JSAMPARRAY [edi] ; output_data
+ alignx 16, 7
+.rowloop:
+ push eax ; colctr
+ push edi
+ push esi
+
+ mov esi, JSAMPROW [esi] ; inptr
+ mov edi, JSAMPROW [edi] ; outptr
+
+ test eax, SIZEOF_YMMWORD-1
+ jz short .skip
+ mov dl, JSAMPLE [esi+(eax-1)*SIZEOF_JSAMPLE]
+ mov JSAMPLE [esi+eax*SIZEOF_JSAMPLE], dl ; insert a dummy sample
+.skip:
+ vpxor ymm0, ymm0, ymm0 ; ymm0=(all 0's)
+ vpcmpeqb xmm7, xmm7, xmm7
+ vpsrldq xmm7, xmm7, (SIZEOF_XMMWORD-1) ; (ff -- -- -- ... -- --) LSB is ff
+ vpand ymm7, ymm7, YMMWORD [esi+0*SIZEOF_YMMWORD]
+
+ add eax, byte SIZEOF_YMMWORD-1
+ and eax, byte -SIZEOF_YMMWORD
+ cmp eax, byte SIZEOF_YMMWORD
+ ja short .columnloop
+ alignx 16, 7
+
+.columnloop_last:
+ vpcmpeqb xmm6, xmm6, xmm6
+ vpslldq xmm6, xmm6, (SIZEOF_XMMWORD-1)
+ vperm2i128 ymm6, ymm6, ymm6, 1 ; (---- ---- ... ---- ---- ff) MSB is ff
+ vpand ymm6, ymm6, YMMWORD [esi+0*SIZEOF_YMMWORD]
+ jmp short .upsample
+ alignx 16, 7
+
+.columnloop:
+ vmovdqu ymm6, YMMWORD [esi+1*SIZEOF_YMMWORD]
+ vperm2i128 ymm6, ymm0, ymm6, 0x20
+ vpslldq ymm6, ymm6, 15
+
+.upsample:
+ vmovdqu ymm1, YMMWORD [esi+0*SIZEOF_YMMWORD] ; ymm1=( 0 1 2 ... 29 30 31)
+
+ vperm2i128 ymm2, ymm0, ymm1, 0x20
+ vpalignr ymm2, ymm1, ymm2, 15 ; ymm2=(-- 0 1 ... 28 29 30)
+ vperm2i128 ymm4, ymm0, ymm1, 0x03
+ vpalignr ymm3, ymm4, ymm1, 1 ; ymm3=( 1 2 3 ... 30 31 --)
+
+ vpor ymm2, ymm2, ymm7 ; ymm2=(-1 0 1 ... 28 29 30)
+ vpor ymm3, ymm3, ymm6 ; ymm3=( 1 2 3 ... 30 31 32)
+
+ vpsrldq ymm7, ymm4, (SIZEOF_XMMWORD-1) ; ymm7=(31 -- -- ... -- -- --)
+
+ vpunpckhbw ymm4, ymm1, ymm0 ; ymm4=( 8 9 10 11 12 13 14 15 24 25 26 27 28 29 30 31)
+ vpunpcklbw ymm5, ymm1, ymm0 ; ymm5=( 0 1 2 3 4 5 6 7 16 17 18 19 20 21 22 23)
+ vperm2i128 ymm1, ymm5, ymm4, 0x20 ; ymm1=( 0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15)
+ vperm2i128 ymm4, ymm5, ymm4, 0x31 ; ymm4=(16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31)
+
+ vpunpckhbw ymm5, ymm2, ymm0 ; ymm5=( 7 8 9 10 11 12 13 14 23 24 25 26 27 28 29 30)
+ vpunpcklbw ymm6, ymm2, ymm0 ; ymm6=(-1 0 1 2 3 4 5 6 15 16 17 18 19 20 21 22)
+ vperm2i128 ymm2, ymm6, ymm5, 0x20 ; ymm2=(-1 0 1 2 3 4 5 6 7 8 9 10 11 12 13 14)
+ vperm2i128 ymm5, ymm6, ymm5, 0x31 ; ymm5=(15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30)
+
+ vpunpckhbw ymm6, ymm3, ymm0 ; ymm6=( 1 2 3 4 5 6 7 8 17 18 19 20 21 22 23 24)
+ vpunpcklbw ymm0, ymm3, ymm0 ; ymm0=( 9 10 11 12 13 14 15 16 25 26 27 28 29 30 31 32)
+ vperm2i128 ymm3, ymm0, ymm6, 0x20 ; ymm3=( 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16)
+ vperm2i128 ymm6, ymm0, ymm6, 0x31 ; ymm6=(17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32)
+
+ vpxor ymm0, ymm0, ymm0 ; ymm0=(all 0's)
+
+ vpmullw ymm1, ymm1, [GOTOFF(ebx,PW_THREE)]
+ vpmullw ymm4, ymm4, [GOTOFF(ebx,PW_THREE)]
+ vpaddw ymm2, ymm2, [GOTOFF(ebx,PW_ONE)]
+ vpaddw ymm5, ymm5, [GOTOFF(ebx,PW_ONE)]
+ vpaddw ymm3, ymm3, [GOTOFF(ebx,PW_TWO)]
+ vpaddw ymm6, ymm6, [GOTOFF(ebx,PW_TWO)]
+
+ vpaddw ymm2, ymm2, ymm1
+ vpaddw ymm5, ymm5, ymm4
+ vpsrlw ymm2, ymm2, 2 ; ymm2=OutLE=( 0 2 4 6 8 10 12 14 16 18 20 22 24 26 28 30)
+ vpsrlw ymm5, ymm5, 2 ; ymm5=OutHE=(32 34 36 38 40 42 44 46 48 50 52 54 56 58 60 62)
+ vpaddw ymm3, ymm3, ymm1
+ vpaddw ymm6, ymm6, ymm4
+ vpsrlw ymm3, ymm3, 2 ; ymm3=OutLO=( 1 3 5 7 9 11 13 15 17 19 21 23 25 27 29 31)
+ vpsrlw ymm6, ymm6, 2 ; ymm6=OutHO=(33 35 37 39 41 43 45 47 49 51 53 55 57 59 61 63)
+
+ vpsllw ymm3, ymm3, BYTE_BIT
+ vpsllw ymm6, ymm6, BYTE_BIT
+ vpor ymm2, ymm2, ymm3 ; ymm2=OutL=( 0 1 2 ... 29 30 31)
+ vpor ymm5, ymm5, ymm6 ; ymm5=OutH=(32 33 34 ... 61 62 63)
+
+ vmovdqu YMMWORD [edi+0*SIZEOF_YMMWORD], ymm2
+ vmovdqu YMMWORD [edi+1*SIZEOF_YMMWORD], ymm5
+
+ sub eax, byte SIZEOF_YMMWORD
+ add esi, byte 1*SIZEOF_YMMWORD ; inptr
+ add edi, byte 2*SIZEOF_YMMWORD ; outptr
+ cmp eax, byte SIZEOF_YMMWORD
+ ja near .columnloop
+ test eax, eax
+ jnz near .columnloop_last
+
+ pop esi
+ pop edi
+ pop eax
+
+ add esi, byte SIZEOF_JSAMPROW ; input_data
+ add edi, byte SIZEOF_JSAMPROW ; output_data
+ dec ecx ; rowctr
+ jg near .rowloop
+
+.return:
+ vzeroupper
+ pop edi
+ pop esi
+; pop edx ; need not be preserved
+; pop ecx ; need not be preserved
+ poppic ebx
+ pop ebp
+ ret
+
+; --------------------------------------------------------------------------
+;
+; Fancy processing for the common case of 2:1 horizontal and 2:1 vertical.
+; Again a triangle filter; see comments for h2v1 case, above.
+;
+; GLOBAL(void)
+; jsimd_h2v2_fancy_upsample_avx2 (int max_v_samp_factor,
+; JDIMENSION downsampled_width,
+; JSAMPARRAY input_data,
+; JSAMPARRAY *output_data_ptr);
+;
+
+%define max_v_samp(b) (b)+8 ; int max_v_samp_factor
+%define downsamp_width(b) (b)+12 ; JDIMENSION downsampled_width
+%define input_data(b) (b)+16 ; JSAMPARRAY input_data
+%define output_data_ptr(b) (b)+20 ; JSAMPARRAY *output_data_ptr
+
+%define original_ebp ebp+0
+%define wk(i) ebp-(WK_NUM-(i))*SIZEOF_YMMWORD ; ymmword wk[WK_NUM]
+%define WK_NUM 4
+%define gotptr wk(0)-SIZEOF_POINTER ; void *gotptr
+
+ align 32
+ global EXTN(jsimd_h2v2_fancy_upsample_avx2)
+
+EXTN(jsimd_h2v2_fancy_upsample_avx2):
+ push ebp
+ mov eax, esp ; eax = original ebp
+ sub esp, byte 4
+ and esp, byte (-SIZEOF_YMMWORD) ; align to 256 bits
+ mov [esp], eax
+ mov ebp, esp ; ebp = aligned ebp
+ lea esp, [wk(0)]
+ pushpic eax ; make a room for GOT address
+ push ebx
+; push ecx ; need not be preserved
+; push edx ; need not be preserved
+ push esi
+ push edi
+
+ get_GOT ebx ; get GOT address
+ movpic POINTER [gotptr], ebx ; save GOT address
+
+ mov edx, eax ; edx = original ebp
+ mov eax, JDIMENSION [downsamp_width(edx)] ; colctr
+ test eax, eax
+ jz near .return
+
+ mov ecx, INT [max_v_samp(edx)] ; rowctr
+ test ecx, ecx
+ jz near .return
+
+ mov esi, JSAMPARRAY [input_data(edx)] ; input_data
+ mov edi, POINTER [output_data_ptr(edx)]
+ mov edi, JSAMPARRAY [edi] ; output_data
+ alignx 16, 7
+.rowloop:
+ push eax ; colctr
+ push ecx
+ push edi
+ push esi
+
+ mov ecx, JSAMPROW [esi-1*SIZEOF_JSAMPROW] ; inptr1(above)
+ mov ebx, JSAMPROW [esi+0*SIZEOF_JSAMPROW] ; inptr0
+ mov esi, JSAMPROW [esi+1*SIZEOF_JSAMPROW] ; inptr1(below)
+ mov edx, JSAMPROW [edi+0*SIZEOF_JSAMPROW] ; outptr0
+ mov edi, JSAMPROW [edi+1*SIZEOF_JSAMPROW] ; outptr1
+
+ test eax, SIZEOF_YMMWORD-1
+ jz short .skip
+ push edx
+ mov dl, JSAMPLE [ecx+(eax-1)*SIZEOF_JSAMPLE]
+ mov JSAMPLE [ecx+eax*SIZEOF_JSAMPLE], dl
+ mov dl, JSAMPLE [ebx+(eax-1)*SIZEOF_JSAMPLE]
+ mov JSAMPLE [ebx+eax*SIZEOF_JSAMPLE], dl
+ mov dl, JSAMPLE [esi+(eax-1)*SIZEOF_JSAMPLE]
+ mov JSAMPLE [esi+eax*SIZEOF_JSAMPLE], dl ; insert a dummy sample
+ pop edx
+.skip:
+ ; -- process the first column block
+
+ vmovdqu ymm0, YMMWORD [ebx+0*SIZEOF_YMMWORD] ; ymm0=row[ 0][0]
+ vmovdqu ymm1, YMMWORD [ecx+0*SIZEOF_YMMWORD] ; ymm1=row[-1][0]
+ vmovdqu ymm2, YMMWORD [esi+0*SIZEOF_YMMWORD] ; ymm2=row[+1][0]
+
+ pushpic ebx
+ movpic ebx, POINTER [gotptr] ; load GOT address
+
+ vpxor ymm3, ymm3, ymm3 ; ymm3=(all 0's)
+
+ vpunpckhbw ymm4, ymm0, ymm3 ; ymm4=row[ 0]( 8 9 10 11 12 13 14 15 24 25 26 27 28 29 30 31)
+ vpunpcklbw ymm5, ymm0, ymm3 ; ymm5=row[ 0]( 0 1 2 3 4 5 6 7 16 17 18 19 20 21 22 23)
+ vperm2i128 ymm0, ymm5, ymm4, 0x20 ; ymm0=row[ 0]( 0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15)
+ vperm2i128 ymm4, ymm5, ymm4, 0x31 ; ymm4=row[ 0](16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31)
+
+ vpunpckhbw ymm5, ymm1, ymm3 ; ymm5=row[-1]( 8 9 10 11 12 13 14 15 24 25 26 27 28 29 30 31)
+ vpunpcklbw ymm6, ymm1, ymm3 ; ymm6=row[-1]( 0 1 2 3 4 5 6 7 16 17 18 19 20 21 22 23)
+ vperm2i128 ymm1, ymm6, ymm5, 0x20 ; ymm1=row[-1]( 0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15)
+ vperm2i128 ymm5, ymm6, ymm5, 0x31 ; ymm5=row[-1](16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31)
+
+ vpunpckhbw ymm6, ymm2, ymm3 ; ymm6=row[+1]( 8 9 10 11 12 13 14 15 24 25 26 27 28 29 30 31)
+ vpunpcklbw ymm3, ymm2, ymm3 ; ymm3=row[+1]( 0 1 2 3 4 5 6 7 16 17 18 19 20 21 22 23)
+ vperm2i128 ymm2, ymm3, ymm6, 0x20 ; ymm2=row[+1]( 0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15)
+ vperm2i128 ymm6, ymm3, ymm6, 0x31 ; ymm6=row[+1](16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31)
+
+ vpmullw ymm0, ymm0, [GOTOFF(ebx,PW_THREE)]
+ vpmullw ymm4, ymm4, [GOTOFF(ebx,PW_THREE)]
+
+ vpcmpeqb xmm7, xmm7, xmm7
+ vpsrldq xmm7, xmm7, (SIZEOF_XMMWORD-2) ; (ffff ---- ---- ... ---- ----) LSB is ffff
+
+ vpaddw ymm1, ymm1, ymm0 ; ymm1=Int0L=( 0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15)
+ vpaddw ymm5, ymm5, ymm4 ; ymm5=Int0H=(16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31)
+ vpaddw ymm2, ymm2, ymm0 ; ymm2=Int1L=( 0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15)
+ vpaddw ymm6, ymm6, ymm4 ; ymm6=Int1H=(16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31)
+
+ vmovdqu YMMWORD [edx+0*SIZEOF_YMMWORD], ymm1 ; temporarily save
+ vmovdqu YMMWORD [edx+1*SIZEOF_YMMWORD], ymm5 ; the intermediate data
+ vmovdqu YMMWORD [edi+0*SIZEOF_YMMWORD], ymm2
+ vmovdqu YMMWORD [edi+1*SIZEOF_YMMWORD], ymm6
+
+ vpand ymm1, ymm1, ymm7 ; ymm1=( 0 -- -- -- -- -- -- -- -- -- -- -- -- -- -- --)
+ vpand ymm2, ymm2, ymm7 ; ymm2=( 0 -- -- -- -- -- -- -- -- -- -- -- -- -- -- --)
+
+ vmovdqa YMMWORD [wk(0)], ymm1
+ vmovdqa YMMWORD [wk(1)], ymm2
+
+ poppic ebx
+
+ add eax, byte SIZEOF_YMMWORD-1
+ and eax, byte -SIZEOF_YMMWORD
+ cmp eax, byte SIZEOF_YMMWORD
+ ja short .columnloop
+ alignx 16, 7
+
+.columnloop_last:
+ ; -- process the last column block
+
+ pushpic ebx
+ movpic ebx, POINTER [gotptr] ; load GOT address
+
+ vpcmpeqb xmm1, xmm1, xmm1
+ vpslldq xmm1, xmm1, (SIZEOF_XMMWORD-2)
+ vperm2i128 ymm1, ymm1, ymm1, 1 ; (---- ---- ... ---- ---- ffff) MSB is ffff
+
+ vpand ymm2, ymm1, YMMWORD [edi+1*SIZEOF_YMMWORD]
+ vpand ymm1, ymm1, YMMWORD [edx+1*SIZEOF_YMMWORD]
+
+ vmovdqa YMMWORD [wk(2)], ymm1 ; ymm1=(-- -- -- -- -- -- -- -- -- -- -- -- -- -- -- 31)
+ vmovdqa YMMWORD [wk(3)], ymm2 ; ymm2=(-- -- -- -- -- -- -- -- -- -- -- -- -- -- -- 31)
+
+ jmp near .upsample
+ alignx 16, 7
+
+.columnloop:
+ ; -- process the next column block
+
+ vmovdqu ymm0, YMMWORD [ebx+1*SIZEOF_YMMWORD] ; ymm0=row[ 0][1]
+ vmovdqu ymm1, YMMWORD [ecx+1*SIZEOF_YMMWORD] ; ymm1=row[-1][1]
+ vmovdqu ymm2, YMMWORD [esi+1*SIZEOF_YMMWORD] ; ymm2=row[+1][1]
+
+ pushpic ebx
+ movpic ebx, POINTER [gotptr] ; load GOT address
+
+ vpxor ymm3, ymm3, ymm3 ; ymm3=(all 0's)
+
+ vpunpckhbw ymm4, ymm0, ymm3 ; ymm4=row[ 0]( 8 9 10 11 12 13 14 15 24 25 26 27 28 29 30 31)
+ vpunpcklbw ymm5, ymm0, ymm3 ; ymm5=row[ 0]( 0 1 2 3 4 5 6 7 16 17 18 19 20 21 22 23)
+ vperm2i128 ymm0, ymm5, ymm4, 0x20 ; ymm0=row[ 0]( 0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15)
+ vperm2i128 ymm4, ymm5, ymm4, 0x31 ; ymm4=row[ 0](16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31)
+
+ vpunpckhbw ymm5, ymm1, ymm3 ; ymm5=row[-1]( 8 9 10 11 12 13 14 15 24 25 26 27 28 29 30 31)
+ vpunpcklbw ymm6, ymm1, ymm3 ; ymm6=row[-1]( 0 1 2 3 4 5 6 7 16 17 18 19 20 21 22 23)
+ vperm2i128 ymm1, ymm6, ymm5, 0x20 ; ymm1=row[-1]( 0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15)
+ vperm2i128 ymm5, ymm6, ymm5, 0x31 ; ymm5=row[-1](16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31)
+
+ vpunpckhbw ymm6, ymm2, ymm3 ; ymm6=row[+1]( 8 9 10 11 12 13 14 15 24 25 26 27 28 29 30 31)
+ vpunpcklbw ymm7, ymm2, ymm3 ; ymm7=row[+1]( 0 1 2 3 4 5 6 7 16 17 18 19 20 21 22 23)
+ vperm2i128 ymm2, ymm7, ymm6, 0x20 ; ymm2=row[+1]( 0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15)
+ vperm2i128 ymm6, ymm7, ymm6, 0x31 ; ymm6=row[+1](16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31)
+
+ vpmullw ymm0, ymm0, [GOTOFF(ebx,PW_THREE)]
+ vpmullw ymm4, ymm4, [GOTOFF(ebx,PW_THREE)]
+
+ vpaddw ymm1, ymm1, ymm0 ; ymm1=Int0L=( 0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15)
+ vpaddw ymm5, ymm5, ymm4 ; ymm5=Int0H=(16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31)
+ vpaddw ymm2, ymm2, ymm0 ; ymm2=Int1L=( 0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15)
+ vpaddw ymm6, ymm6, ymm4 ; ymm6=Int1H=(16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31)
+
+ vmovdqu YMMWORD [edx+2*SIZEOF_YMMWORD], ymm1 ; temporarily save
+ vmovdqu YMMWORD [edx+3*SIZEOF_YMMWORD], ymm5 ; the intermediate data
+ vmovdqu YMMWORD [edi+2*SIZEOF_YMMWORD], ymm2
+ vmovdqu YMMWORD [edi+3*SIZEOF_YMMWORD], ymm6
+
+ vperm2i128 ymm1, ymm3, ymm1, 0x20
+ vpslldq ymm1, ymm1, 14 ; ymm1=(-- -- -- -- -- -- -- -- -- -- -- -- -- -- -- 0)
+ vperm2i128 ymm2, ymm3, ymm2, 0x20
+ vpslldq ymm2, ymm2, 14 ; ymm2=(-- -- -- -- -- -- -- -- -- -- -- -- -- -- -- 0)
+
+ vmovdqa YMMWORD [wk(2)], ymm1
+ vmovdqa YMMWORD [wk(3)], ymm2
+
+.upsample:
+ ; -- process the upper row
+
+ vmovdqu ymm7, YMMWORD [edx+0*SIZEOF_YMMWORD] ; ymm7=Int0L=( 0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15)
+ vmovdqu ymm3, YMMWORD [edx+1*SIZEOF_YMMWORD] ; ymm3=Int0H=(16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31)
+
+ vpxor ymm1, ymm1, ymm1 ; ymm1=(all 0's)
+
+ vperm2i128 ymm0, ymm1, ymm7, 0x03
+ vpalignr ymm0, ymm0, ymm7, 2 ; ymm0=( 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 --)
+ vperm2i128 ymm4, ymm1, ymm3, 0x20
+ vpslldq ymm4, ymm4, 14 ; ymm4=(-- -- -- -- -- -- -- -- -- -- -- -- -- -- -- 16)
+
+ vperm2i128 ymm5, ymm1, ymm7, 0x03
+ vpsrldq ymm5, ymm5, 14 ; ymm5=(15 -- -- -- -- -- -- -- -- -- -- -- -- -- -- --)
+ vperm2i128 ymm6, ymm1, ymm3, 0x20
+ vpalignr ymm6, ymm3, ymm6, 14 ; ymm6=(-- 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30)
+
+ vpor ymm0, ymm0, ymm4 ; ymm0=( 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16)
+ vpor ymm5, ymm5, ymm6 ; ymm5=(15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30)
+
+ vperm2i128 ymm2, ymm1, ymm3, 0x03
+ vpalignr ymm2, ymm2, ymm3, 2 ; ymm2=(17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 --)
+ vperm2i128 ymm4, ymm1, ymm3, 0x03
+ vpsrldq ymm4, ymm4, 14 ; ymm4=(31 -- -- -- -- -- -- -- -- -- -- -- -- -- -- --)
+ vperm2i128 ymm1, ymm1, ymm7, 0x20
+ vpalignr ymm1, ymm7, ymm1, 14 ; ymm1=(-- 0 1 2 3 4 5 6 7 8 9 10 11 12 13 14)
+
+ vpor ymm1, ymm1, YMMWORD [wk(0)] ; ymm1=(-1 0 1 2 3 4 5 6 7 8 9 10 11 12 13 14)
+ vpor ymm2, ymm2, YMMWORD [wk(2)] ; ymm2=(17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32)
+
+ vmovdqa YMMWORD [wk(0)], ymm4
+
+ vpmullw ymm7, ymm7, [GOTOFF(ebx,PW_THREE)]
+ vpmullw ymm3, ymm3, [GOTOFF(ebx,PW_THREE)]
+ vpaddw ymm1, ymm1, [GOTOFF(ebx,PW_EIGHT)]
+ vpaddw ymm5, ymm5, [GOTOFF(ebx,PW_EIGHT)]
+ vpaddw ymm0, ymm0, [GOTOFF(ebx,PW_SEVEN)]
+ vpaddw ymm2, [GOTOFF(ebx,PW_SEVEN)]
+
+ vpaddw ymm1, ymm1, ymm7
+ vpaddw ymm5, ymm5, ymm3
+ vpsrlw ymm1, ymm1, 4 ; ymm1=Out0LE=( 0 2 4 6 8 10 12 14 16 18 20 22 24 26 28 30)
+ vpsrlw ymm5, ymm5, 4 ; ymm5=Out0HE=(32 34 36 38 40 42 44 46 48 50 52 54 56 58 60 62)
+ vpaddw ymm0, ymm0, ymm7
+ vpaddw ymm2, ymm2, ymm3
+ vpsrlw ymm0, ymm0, 4 ; ymm0=Out0LO=( 1 3 5 7 9 11 13 15 17 19 21 23 25 27 29 31)
+ vpsrlw ymm2, ymm2, 4 ; ymm2=Out0HO=(33 35 37 39 41 43 45 47 49 51 53 55 57 59 61 63)
+
+ vpsllw ymm0, ymm0, BYTE_BIT
+ vpsllw ymm2, ymm2, BYTE_BIT
+ vpor ymm1, ymm1, ymm0 ; ymm1=Out0L=( 0 1 2 ... 29 30 31)
+ vpor ymm5, ymm5, ymm2 ; ymm5=Out0H=(32 33 34 ... 61 62 63)
+
+ vmovdqu YMMWORD [edx+0*SIZEOF_YMMWORD], ymm1
+ vmovdqu YMMWORD [edx+1*SIZEOF_YMMWORD], ymm5
+
+ ; -- process the lower row
+
+ vmovdqu ymm6, YMMWORD [edi+0*SIZEOF_YMMWORD] ; ymm6=Int1L=( 0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15)
+ vmovdqu ymm4, YMMWORD [edi+1*SIZEOF_YMMWORD] ; ymm4=Int1H=(16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31)
+
+ vpxor ymm1, ymm1, ymm1 ; ymm1=(all 0's)
+
+ vperm2i128 ymm7, ymm1, ymm6, 0x03
+ vpalignr ymm7, ymm7, ymm6, 2 ; ymm7=( 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 --)
+ vperm2i128 ymm3, ymm1, ymm4, 0x20
+ vpslldq ymm3, ymm3, 14 ; ymm3=(-- -- -- -- -- -- -- -- -- -- -- -- -- -- -- 16)
+
+ vperm2i128 ymm0, ymm1, ymm6, 0x03
+ vpsrldq ymm0, ymm0, 14 ; ymm0=(15 -- -- -- -- -- -- -- -- -- -- -- -- -- -- --)
+ vperm2i128 ymm2, ymm1, ymm4, 0x20
+ vpalignr ymm2, ymm4, ymm2, 14 ; ymm2=(-- 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30)
+
+ vpor ymm7, ymm7, ymm3 ; ymm7=( 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16)
+ vpor ymm0, ymm0, ymm2 ; ymm0=(15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30)
+
+ vperm2i128 ymm5, ymm1, ymm4, 0x03
+ vpalignr ymm5, ymm5, ymm4, 2 ; ymm5=(17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 --)
+ vperm2i128 ymm3, ymm1, ymm4, 0x03
+ vpsrldq ymm3, ymm3, 14 ; ymm3=(31 -- -- -- -- -- -- -- -- -- -- -- -- -- -- --)
+ vperm2i128 ymm1, ymm1, ymm6, 0x20
+ vpalignr ymm1, ymm6, ymm1, 14 ; ymm1=(-- 0 1 2 3 4 5 6 7 8 9 10 11 12 13 14)
+
+ vpor ymm1, ymm1, YMMWORD [wk(1)] ; ymm1=(-1 0 1 2 3 4 5 6 7 8 9 10 11 12 13 14)
+ vpor ymm5, ymm5, YMMWORD [wk(3)] ; ymm5=(17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32)
+
+ vmovdqa YMMWORD [wk(1)], ymm3
+
+ vpmullw ymm6, ymm6, [GOTOFF(ebx,PW_THREE)]
+ vpmullw ymm4, ymm4, [GOTOFF(ebx,PW_THREE)]
+ vpaddw ymm1, ymm1, [GOTOFF(ebx,PW_EIGHT)]
+ vpaddw ymm0, ymm0, [GOTOFF(ebx,PW_EIGHT)]
+ vpaddw ymm7, ymm7, [GOTOFF(ebx,PW_SEVEN)]
+ vpaddw ymm5, ymm5, [GOTOFF(ebx,PW_SEVEN)]
+
+ vpaddw ymm1, ymm1, ymm6
+ vpaddw ymm0, ymm0, ymm4
+ vpsrlw ymm1, ymm1, 4 ; ymm1=Out1LE=( 0 2 4 6 8 10 12 14 16 18 20 22 24 26 28 30)
+ vpsrlw ymm0, ymm0, 4 ; ymm0=Out1HE=(32 34 36 38 40 42 44 46 48 50 52 54 56 58 60 62)
+ vpaddw ymm7, ymm7, ymm6
+ vpaddw ymm5, ymm5, ymm4
+ vpsrlw ymm7, ymm7, 4 ; ymm7=Out1LO=( 1 3 5 7 9 11 13 15 17 19 21 23 25 27 29 31)
+ vpsrlw ymm5, ymm5, 4 ; ymm5=Out1HO=(33 35 37 39 41 43 45 47 49 51 53 55 57 59 61 63)
+
+ vpsllw ymm7, ymm7, BYTE_BIT
+ vpsllw ymm5, ymm5, BYTE_BIT
+ vpor ymm1, ymm1, ymm7 ; ymm1=Out1L=( 0 1 2 ... 29 30 31)
+ vpor ymm0, ymm0, ymm5 ; ymm0=Out1H=(32 33 34 ... 61 62 63)
+
+ vmovdqu YMMWORD [edi+0*SIZEOF_YMMWORD], ymm1
+ vmovdqu YMMWORD [edi+1*SIZEOF_YMMWORD], ymm0
+
+ poppic ebx
+
+ sub eax, byte SIZEOF_YMMWORD
+ add ecx, byte 1*SIZEOF_YMMWORD ; inptr1(above)
+ add ebx, byte 1*SIZEOF_YMMWORD ; inptr0
+ add esi, byte 1*SIZEOF_YMMWORD ; inptr1(below)
+ add edx, byte 2*SIZEOF_YMMWORD ; outptr0
+ add edi, byte 2*SIZEOF_YMMWORD ; outptr1
+ cmp eax, byte SIZEOF_YMMWORD
+ ja near .columnloop
+ test eax, eax
+ jnz near .columnloop_last
+
+ pop esi
+ pop edi
+ pop ecx
+ pop eax
+
+ add esi, byte 1*SIZEOF_JSAMPROW ; input_data
+ add edi, byte 2*SIZEOF_JSAMPROW ; output_data
+ sub ecx, byte 2 ; rowctr
+ jg near .rowloop
+
+.return:
+ vzeroupper
+ pop edi
+ pop esi
+; pop edx ; need not be preserved
+; pop ecx ; need not be preserved
+ pop ebx
+ mov esp, ebp ; esp <- aligned ebp
+ pop esp ; esp <- original ebp
+ pop ebp
+ ret
+
+; --------------------------------------------------------------------------
+;
+; Fast processing for the common case of 2:1 horizontal and 1:1 vertical.
+; It's still a box filter.
+;
+; GLOBAL(void)
+; jsimd_h2v1_upsample_avx2 (int max_v_samp_factor,
+; JDIMENSION output_width,
+; JSAMPARRAY input_data,
+; JSAMPARRAY *output_data_ptr);
+;
+
+%define max_v_samp(b) (b)+8 ; int max_v_samp_factor
+%define output_width(b) (b)+12 ; JDIMENSION output_width
+%define input_data(b) (b)+16 ; JSAMPARRAY input_data
+%define output_data_ptr(b) (b)+20 ; JSAMPARRAY *output_data_ptr
+
+ align 32
+ global EXTN(jsimd_h2v1_upsample_avx2)
+
+EXTN(jsimd_h2v1_upsample_avx2):
+ push ebp
+ mov ebp, esp
+; push ebx ; unused
+; push ecx ; need not be preserved
+; push edx ; need not be preserved
+ push esi
+ push edi
+
+ mov edx, JDIMENSION [output_width(ebp)]
+ add edx, byte (SIZEOF_YMMWORD-1)
+ and edx, -SIZEOF_YMMWORD
+ jz short .return
+
+ mov ecx, INT [max_v_samp(ebp)] ; rowctr
+ test ecx, ecx
+ jz short .return
+
+ mov esi, JSAMPARRAY [input_data(ebp)] ; input_data
+ mov edi, POINTER [output_data_ptr(ebp)]
+ mov edi, JSAMPARRAY [edi] ; output_data
+ alignx 16, 7
+.rowloop:
+ push edi
+ push esi
+
+ mov esi, JSAMPROW [esi] ; inptr
+ mov edi, JSAMPROW [edi] ; outptr
+ mov eax, edx ; colctr
+ alignx 16, 7
+.columnloop:
+
+ cmp eax, byte SIZEOF_YMMWORD
+ ja near .above_16
+
+ vmovdqu xmm0, XMMWORD [esi+0*SIZEOF_YMMWORD]
+ vpunpckhbw xmm1, xmm0, xmm0
+ vpunpcklbw xmm0, xmm0, xmm0
+
+ vmovdqu XMMWORD [edi+0*SIZEOF_XMMWORD], xmm0
+ vmovdqu XMMWORD [edi+1*SIZEOF_XMMWORD], xmm1
+
+ jmp short .nextrow
+
+.above_16:
+ vmovdqu ymm0, YMMWORD [esi+0*SIZEOF_YMMWORD]
+
+ vpermq ymm0, ymm0, 0xd8
+ vpunpckhbw ymm1, ymm0, ymm0
+ vpunpcklbw ymm0, ymm0, ymm0
+
+ vmovdqu YMMWORD [edi+0*SIZEOF_YMMWORD], ymm0
+ vmovdqu YMMWORD [edi+1*SIZEOF_YMMWORD], ymm1
+
+ sub eax, byte 2*SIZEOF_YMMWORD
+ jz short .nextrow
+
+ add esi, byte SIZEOF_YMMWORD ; inptr
+ add edi, byte 2*SIZEOF_YMMWORD ; outptr
+ jmp short .columnloop
+ alignx 16, 7
+
+.nextrow:
+ pop esi
+ pop edi
+
+ add esi, byte SIZEOF_JSAMPROW ; input_data
+ add edi, byte SIZEOF_JSAMPROW ; output_data
+ dec ecx ; rowctr
+ jg short .rowloop
+
+.return:
+ vzeroupper
+ pop edi
+ pop esi
+; pop edx ; need not be preserved
+; pop ecx ; need not be preserved
+; pop ebx ; unused
+ pop ebp
+ ret
+
+; --------------------------------------------------------------------------
+;
+; Fast processing for the common case of 2:1 horizontal and 2:1 vertical.
+; It's still a box filter.
+;
+; GLOBAL(void)
+; jsimd_h2v2_upsample_avx2 (int max_v_samp_factor,
+; JDIMENSION output_width,
+; JSAMPARRAY input_data,
+; JSAMPARRAY *output_data_ptr);
+;
+
+%define max_v_samp(b) (b)+8 ; int max_v_samp_factor
+%define output_width(b) (b)+12 ; JDIMENSION output_width
+%define input_data(b) (b)+16 ; JSAMPARRAY input_data
+%define output_data_ptr(b) (b)+20 ; JSAMPARRAY *output_data_ptr
+
+ align 32
+ global EXTN(jsimd_h2v2_upsample_avx2)
+
+EXTN(jsimd_h2v2_upsample_avx2):
+ push ebp
+ mov ebp, esp
+ push ebx
+; push ecx ; need not be preserved
+; push edx ; need not be preserved
+ push esi
+ push edi
+
+ mov edx, JDIMENSION [output_width(ebp)]
+ add edx, byte (SIZEOF_YMMWORD-1)
+ and edx, -SIZEOF_YMMWORD
+ jz near .return
+
+ mov ecx, INT [max_v_samp(ebp)] ; rowctr
+ test ecx, ecx
+ jz near .return
+
+ mov esi, JSAMPARRAY [input_data(ebp)] ; input_data
+ mov edi, POINTER [output_data_ptr(ebp)]
+ mov edi, JSAMPARRAY [edi] ; output_data
+ alignx 16, 7
+.rowloop:
+ push edi
+ push esi
+
+ mov esi, JSAMPROW [esi] ; inptr
+ mov ebx, JSAMPROW [edi+0*SIZEOF_JSAMPROW] ; outptr0
+ mov edi, JSAMPROW [edi+1*SIZEOF_JSAMPROW] ; outptr1
+ mov eax, edx ; colctr
+ alignx 16, 7
+.columnloop:
+
+ cmp eax, byte SIZEOF_YMMWORD
+ ja short .above_16
+
+ vmovdqu xmm0, XMMWORD [esi+0*SIZEOF_XMMWORD]
+ vpunpckhbw xmm1, xmm0, xmm0
+ vpunpcklbw xmm0, xmm0, xmm0
+
+ vmovdqu XMMWORD [ebx+0*SIZEOF_XMMWORD], xmm0
+ vmovdqu XMMWORD [ebx+1*SIZEOF_XMMWORD], xmm1
+ vmovdqu XMMWORD [edi+0*SIZEOF_XMMWORD], xmm0
+ vmovdqu XMMWORD [edi+1*SIZEOF_XMMWORD], xmm1
+
+ jmp near .nextrow
+
+.above_16:
+ vmovdqu ymm0, YMMWORD [esi+0*SIZEOF_YMMWORD]
+
+ vpermq ymm0, ymm0, 0xd8
+ vpunpckhbw ymm1, ymm0, ymm0
+ vpunpcklbw ymm0, ymm0, ymm0
+
+ vmovdqu YMMWORD [ebx+0*SIZEOF_YMMWORD], ymm0
+ vmovdqu YMMWORD [ebx+1*SIZEOF_YMMWORD], ymm1
+ vmovdqu YMMWORD [edi+0*SIZEOF_YMMWORD], ymm0
+ vmovdqu YMMWORD [edi+1*SIZEOF_YMMWORD], ymm1
+
+ sub eax, byte 2*SIZEOF_YMMWORD
+ jz short .nextrow
+
+ add esi, byte SIZEOF_YMMWORD ; inptr
+ add ebx, 2*SIZEOF_YMMWORD ; outptr0
+ add edi, 2*SIZEOF_YMMWORD ; outptr1
+ jmp short .columnloop
+ alignx 16, 7
+
+.nextrow:
+ pop esi
+ pop edi
+
+ add esi, byte 1*SIZEOF_JSAMPROW ; input_data
+ add edi, byte 2*SIZEOF_JSAMPROW ; output_data
+ sub ecx, byte 2 ; rowctr
+ jg near .rowloop
+
+.return:
+ vzeroupper
+ pop edi
+ pop esi
+; pop edx ; need not be preserved
+; pop ecx ; need not be preserved
+ pop ebx
+ pop ebp
+ ret
+
+; For some reason, the OS X linker does not honor the request to align the
+; segment unless we do this.
+ align 32
diff --git a/simd/jdsample-mmx.asm b/simd/i386/jdsample-mmx.asm
similarity index 98%
rename from simd/jdsample-mmx.asm
rename to simd/i386/jdsample-mmx.asm
index 5e4fa7a..266ee7a 100644
--- a/simd/jdsample-mmx.asm
+++ b/simd/i386/jdsample-mmx.asm
@@ -2,6 +2,7 @@
; jdsample.asm - upsampling (MMX)
;
; Copyright 2009 Pierre Ossman <ossman@cendio.se> for Cendio AB
+; Copyright (C) 2016, D. R. Commander.
;
; Based on the x86 SIMD extension for IJG JPEG library
; Copyright (C) 1999-2006, MIYASAKA Masaru.
@@ -20,7 +21,7 @@
; --------------------------------------------------------------------------
SECTION SEG_CONST
- alignz 16
+ alignz 32
global EXTN(jconst_fancy_upsample_mmx)
EXTN(jconst_fancy_upsample_mmx):
@@ -31,7 +32,7 @@
PW_SEVEN times 4 dw 7
PW_EIGHT times 4 dw 8
- alignz 16
+ alignz 32
; --------------------------------------------------------------------------
SECTION SEG_TEXT
@@ -56,7 +57,7 @@
%define input_data(b) (b)+16 ; JSAMPARRAY input_data
%define output_data_ptr(b) (b)+20 ; JSAMPARRAY *output_data_ptr
- align 16
+ align 32
global EXTN(jsimd_h2v1_fancy_upsample_mmx)
EXTN(jsimd_h2v1_fancy_upsample_mmx):
@@ -214,7 +215,7 @@
%define WK_NUM 4
%define gotptr wk(0)-SIZEOF_POINTER ; void *gotptr
- align 16
+ align 32
global EXTN(jsimd_h2v2_fancy_upsample_mmx)
EXTN(jsimd_h2v2_fancy_upsample_mmx):
@@ -540,7 +541,7 @@
%define input_data(b) (b)+16 ; JSAMPARRAY input_data
%define output_data_ptr(b) (b)+20 ; JSAMPARRAY *output_data_ptr
- align 16
+ align 32
global EXTN(jsimd_h2v1_upsample_mmx)
EXTN(jsimd_h2v1_upsample_mmx):
@@ -641,7 +642,7 @@
%define input_data(b) (b)+16 ; JSAMPARRAY input_data
%define output_data_ptr(b) (b)+20 ; JSAMPARRAY *output_data_ptr
- align 16
+ align 32
global EXTN(jsimd_h2v2_upsample_mmx)
EXTN(jsimd_h2v2_upsample_mmx):
@@ -733,4 +734,4 @@
; For some reason, the OS X linker does not honor the request to align the
; segment unless we do this.
- align 16
+ align 32
diff --git a/simd/i386/jdsample-sse2.asm b/simd/i386/jdsample-sse2.asm
new file mode 100644
index 0000000..01d96b7
--- /dev/null
+++ b/simd/i386/jdsample-sse2.asm
@@ -0,0 +1,729 @@
+;
+; jdsample.asm - upsampling (SSE2)
+;
+; Copyright 2009 Pierre Ossman <ossman@cendio.se> for Cendio AB
+; Copyright (C) 2016, D. R. Commander.
+;
+; Based on the x86 SIMD extension for IJG JPEG library
+; Copyright (C) 1999-2006, MIYASAKA Masaru.
+; For conditions of distribution and use, see copyright notice in jsimdext.inc
+;
+; This file should be assembled with NASM (Netwide Assembler),
+; can *not* be assembled with Microsoft's MASM or any compatible
+; assembler (including Borland's Turbo Assembler).
+; NASM is available from http://nasm.sourceforge.net/ or
+; http://sourceforge.net/project/showfiles.php?group_id=6208
+;
+; [TAB8]
+
+%include "jsimdext.inc"
+
+; --------------------------------------------------------------------------
+ SECTION SEG_CONST
+
+ alignz 32
+ global EXTN(jconst_fancy_upsample_sse2)
+
+EXTN(jconst_fancy_upsample_sse2):
+
+PW_ONE times 8 dw 1
+PW_TWO times 8 dw 2
+PW_THREE times 8 dw 3
+PW_SEVEN times 8 dw 7
+PW_EIGHT times 8 dw 8
+
+ alignz 32
+
+; --------------------------------------------------------------------------
+ SECTION SEG_TEXT
+ BITS 32
+;
+; Fancy processing for the common case of 2:1 horizontal and 1:1 vertical.
+;
+; The upsampling algorithm is linear interpolation between pixel centers,
+; also known as a "triangle filter". This is a good compromise between
+; speed and visual quality. The centers of the output pixels are 1/4 and 3/4
+; of the way between input pixel centers.
+;
+; GLOBAL(void)
+; jsimd_h2v1_fancy_upsample_sse2 (int max_v_samp_factor,
+; JDIMENSION downsampled_width,
+; JSAMPARRAY input_data,
+; JSAMPARRAY *output_data_ptr);
+;
+
+%define max_v_samp(b) (b)+8 ; int max_v_samp_factor
+%define downsamp_width(b) (b)+12 ; JDIMENSION downsampled_width
+%define input_data(b) (b)+16 ; JSAMPARRAY input_data
+%define output_data_ptr(b) (b)+20 ; JSAMPARRAY *output_data_ptr
+
+ align 32
+ global EXTN(jsimd_h2v1_fancy_upsample_sse2)
+
+EXTN(jsimd_h2v1_fancy_upsample_sse2):
+ push ebp
+ mov ebp, esp
+ pushpic ebx
+; push ecx ; need not be preserved
+; push edx ; need not be preserved
+ push esi
+ push edi
+
+ get_GOT ebx ; get GOT address
+
+ mov eax, JDIMENSION [downsamp_width(ebp)] ; colctr
+ test eax, eax
+ jz near .return
+
+ mov ecx, INT [max_v_samp(ebp)] ; rowctr
+ test ecx, ecx
+ jz near .return
+
+ mov esi, JSAMPARRAY [input_data(ebp)] ; input_data
+ mov edi, POINTER [output_data_ptr(ebp)]
+ mov edi, JSAMPARRAY [edi] ; output_data
+ alignx 16, 7
+.rowloop:
+ push eax ; colctr
+ push edi
+ push esi
+
+ mov esi, JSAMPROW [esi] ; inptr
+ mov edi, JSAMPROW [edi] ; outptr
+
+ test eax, SIZEOF_XMMWORD-1
+ jz short .skip
+ mov dl, JSAMPLE [esi+(eax-1)*SIZEOF_JSAMPLE]
+ mov JSAMPLE [esi+eax*SIZEOF_JSAMPLE], dl ; insert a dummy sample
+.skip:
+ pxor xmm0, xmm0 ; xmm0=(all 0's)
+ pcmpeqb xmm7, xmm7
+ psrldq xmm7, (SIZEOF_XMMWORD-1)
+ pand xmm7, XMMWORD [esi+0*SIZEOF_XMMWORD]
+
+ add eax, byte SIZEOF_XMMWORD-1
+ and eax, byte -SIZEOF_XMMWORD
+ cmp eax, byte SIZEOF_XMMWORD
+ ja short .columnloop
+ alignx 16, 7
+
+.columnloop_last:
+ pcmpeqb xmm6, xmm6
+ pslldq xmm6, (SIZEOF_XMMWORD-1)
+ pand xmm6, XMMWORD [esi+0*SIZEOF_XMMWORD]
+ jmp short .upsample
+ alignx 16, 7
+
+.columnloop:
+ movdqa xmm6, XMMWORD [esi+1*SIZEOF_XMMWORD]
+ pslldq xmm6, (SIZEOF_XMMWORD-1)
+
+.upsample:
+ movdqa xmm1, XMMWORD [esi+0*SIZEOF_XMMWORD]
+ movdqa xmm2, xmm1
+ movdqa xmm3, xmm1 ; xmm1=( 0 1 2 ... 13 14 15)
+ pslldq xmm2, 1 ; xmm2=(-- 0 1 ... 12 13 14)
+ psrldq xmm3, 1 ; xmm3=( 1 2 3 ... 14 15 --)
+
+ por xmm2, xmm7 ; xmm2=(-1 0 1 ... 12 13 14)
+ por xmm3, xmm6 ; xmm3=( 1 2 3 ... 14 15 16)
+
+ movdqa xmm7, xmm1
+ psrldq xmm7, (SIZEOF_XMMWORD-1) ; xmm7=(15 -- -- ... -- -- --)
+
+ movdqa xmm4, xmm1
+ punpcklbw xmm1, xmm0 ; xmm1=( 0 1 2 3 4 5 6 7)
+ punpckhbw xmm4, xmm0 ; xmm4=( 8 9 10 11 12 13 14 15)
+ movdqa xmm5, xmm2
+ punpcklbw xmm2, xmm0 ; xmm2=(-1 0 1 2 3 4 5 6)
+ punpckhbw xmm5, xmm0 ; xmm5=( 7 8 9 10 11 12 13 14)
+ movdqa xmm6, xmm3
+ punpcklbw xmm3, xmm0 ; xmm3=( 1 2 3 4 5 6 7 8)
+ punpckhbw xmm6, xmm0 ; xmm6=( 9 10 11 12 13 14 15 16)
+
+ pmullw xmm1, [GOTOFF(ebx,PW_THREE)]
+ pmullw xmm4, [GOTOFF(ebx,PW_THREE)]
+ paddw xmm2, [GOTOFF(ebx,PW_ONE)]
+ paddw xmm5, [GOTOFF(ebx,PW_ONE)]
+ paddw xmm3, [GOTOFF(ebx,PW_TWO)]
+ paddw xmm6, [GOTOFF(ebx,PW_TWO)]
+
+ paddw xmm2, xmm1
+ paddw xmm5, xmm4
+ psrlw xmm2, 2 ; xmm2=OutLE=( 0 2 4 6 8 10 12 14)
+ psrlw xmm5, 2 ; xmm5=OutHE=(16 18 20 22 24 26 28 30)
+ paddw xmm3, xmm1
+ paddw xmm6, xmm4
+ psrlw xmm3, 2 ; xmm3=OutLO=( 1 3 5 7 9 11 13 15)
+ psrlw xmm6, 2 ; xmm6=OutHO=(17 19 21 23 25 27 29 31)
+
+ psllw xmm3, BYTE_BIT
+ psllw xmm6, BYTE_BIT
+ por xmm2, xmm3 ; xmm2=OutL=( 0 1 2 ... 13 14 15)
+ por xmm5, xmm6 ; xmm5=OutH=(16 17 18 ... 29 30 31)
+
+ movdqa XMMWORD [edi+0*SIZEOF_XMMWORD], xmm2
+ movdqa XMMWORD [edi+1*SIZEOF_XMMWORD], xmm5
+
+ sub eax, byte SIZEOF_XMMWORD
+ add esi, byte 1*SIZEOF_XMMWORD ; inptr
+ add edi, byte 2*SIZEOF_XMMWORD ; outptr
+ cmp eax, byte SIZEOF_XMMWORD
+ ja near .columnloop
+ test eax, eax
+ jnz near .columnloop_last
+
+ pop esi
+ pop edi
+ pop eax
+
+ add esi, byte SIZEOF_JSAMPROW ; input_data
+ add edi, byte SIZEOF_JSAMPROW ; output_data
+ dec ecx ; rowctr
+ jg near .rowloop
+
+.return:
+ pop edi
+ pop esi
+; pop edx ; need not be preserved
+; pop ecx ; need not be preserved
+ poppic ebx
+ pop ebp
+ ret
+
+; --------------------------------------------------------------------------
+;
+; Fancy processing for the common case of 2:1 horizontal and 2:1 vertical.
+; Again a triangle filter; see comments for h2v1 case, above.
+;
+; GLOBAL(void)
+; jsimd_h2v2_fancy_upsample_sse2 (int max_v_samp_factor,
+; JDIMENSION downsampled_width,
+; JSAMPARRAY input_data,
+; JSAMPARRAY *output_data_ptr);
+;
+
+%define max_v_samp(b) (b)+8 ; int max_v_samp_factor
+%define downsamp_width(b) (b)+12 ; JDIMENSION downsampled_width
+%define input_data(b) (b)+16 ; JSAMPARRAY input_data
+%define output_data_ptr(b) (b)+20 ; JSAMPARRAY *output_data_ptr
+
+%define original_ebp ebp+0
+%define wk(i) ebp-(WK_NUM-(i))*SIZEOF_XMMWORD ; xmmword wk[WK_NUM]
+%define WK_NUM 4
+%define gotptr wk(0)-SIZEOF_POINTER ; void *gotptr
+
+ align 32
+ global EXTN(jsimd_h2v2_fancy_upsample_sse2)
+
+EXTN(jsimd_h2v2_fancy_upsample_sse2):
+ push ebp
+ mov eax, esp ; eax = original ebp
+ sub esp, byte 4
+ and esp, byte (-SIZEOF_XMMWORD) ; align to 128 bits
+ mov [esp], eax
+ mov ebp, esp ; ebp = aligned ebp
+ lea esp, [wk(0)]
+ pushpic eax ; make a room for GOT address
+ push ebx
+; push ecx ; need not be preserved
+; push edx ; need not be preserved
+ push esi
+ push edi
+
+ get_GOT ebx ; get GOT address
+ movpic POINTER [gotptr], ebx ; save GOT address
+
+ mov edx, eax ; edx = original ebp
+ mov eax, JDIMENSION [downsamp_width(edx)] ; colctr
+ test eax, eax
+ jz near .return
+
+ mov ecx, INT [max_v_samp(edx)] ; rowctr
+ test ecx, ecx
+ jz near .return
+
+ mov esi, JSAMPARRAY [input_data(edx)] ; input_data
+ mov edi, POINTER [output_data_ptr(edx)]
+ mov edi, JSAMPARRAY [edi] ; output_data
+ alignx 16, 7
+.rowloop:
+ push eax ; colctr
+ push ecx
+ push edi
+ push esi
+
+ mov ecx, JSAMPROW [esi-1*SIZEOF_JSAMPROW] ; inptr1(above)
+ mov ebx, JSAMPROW [esi+0*SIZEOF_JSAMPROW] ; inptr0
+ mov esi, JSAMPROW [esi+1*SIZEOF_JSAMPROW] ; inptr1(below)
+ mov edx, JSAMPROW [edi+0*SIZEOF_JSAMPROW] ; outptr0
+ mov edi, JSAMPROW [edi+1*SIZEOF_JSAMPROW] ; outptr1
+
+ test eax, SIZEOF_XMMWORD-1
+ jz short .skip
+ push edx
+ mov dl, JSAMPLE [ecx+(eax-1)*SIZEOF_JSAMPLE]
+ mov JSAMPLE [ecx+eax*SIZEOF_JSAMPLE], dl
+ mov dl, JSAMPLE [ebx+(eax-1)*SIZEOF_JSAMPLE]
+ mov JSAMPLE [ebx+eax*SIZEOF_JSAMPLE], dl
+ mov dl, JSAMPLE [esi+(eax-1)*SIZEOF_JSAMPLE]
+ mov JSAMPLE [esi+eax*SIZEOF_JSAMPLE], dl ; insert a dummy sample
+ pop edx
+.skip:
+ ; -- process the first column block
+
+ movdqa xmm0, XMMWORD [ebx+0*SIZEOF_XMMWORD] ; xmm0=row[ 0][0]
+ movdqa xmm1, XMMWORD [ecx+0*SIZEOF_XMMWORD] ; xmm1=row[-1][0]
+ movdqa xmm2, XMMWORD [esi+0*SIZEOF_XMMWORD] ; xmm2=row[+1][0]
+
+ pushpic ebx
+ movpic ebx, POINTER [gotptr] ; load GOT address
+
+ pxor xmm3, xmm3 ; xmm3=(all 0's)
+ movdqa xmm4, xmm0
+ punpcklbw xmm0, xmm3 ; xmm0=row[ 0]( 0 1 2 3 4 5 6 7)
+ punpckhbw xmm4, xmm3 ; xmm4=row[ 0]( 8 9 10 11 12 13 14 15)
+ movdqa xmm5, xmm1
+ punpcklbw xmm1, xmm3 ; xmm1=row[-1]( 0 1 2 3 4 5 6 7)
+ punpckhbw xmm5, xmm3 ; xmm5=row[-1]( 8 9 10 11 12 13 14 15)
+ movdqa xmm6, xmm2
+ punpcklbw xmm2, xmm3 ; xmm2=row[+1]( 0 1 2 3 4 5 6 7)
+ punpckhbw xmm6, xmm3 ; xmm6=row[+1]( 8 9 10 11 12 13 14 15)
+
+ pmullw xmm0, [GOTOFF(ebx,PW_THREE)]
+ pmullw xmm4, [GOTOFF(ebx,PW_THREE)]
+
+ pcmpeqb xmm7, xmm7
+ psrldq xmm7, (SIZEOF_XMMWORD-2)
+
+ paddw xmm1, xmm0 ; xmm1=Int0L=( 0 1 2 3 4 5 6 7)
+ paddw xmm5, xmm4 ; xmm5=Int0H=( 8 9 10 11 12 13 14 15)
+ paddw xmm2, xmm0 ; xmm2=Int1L=( 0 1 2 3 4 5 6 7)
+ paddw xmm6, xmm4 ; xmm6=Int1H=( 8 9 10 11 12 13 14 15)
+
+ movdqa XMMWORD [edx+0*SIZEOF_XMMWORD], xmm1 ; temporarily save
+ movdqa XMMWORD [edx+1*SIZEOF_XMMWORD], xmm5 ; the intermediate data
+ movdqa XMMWORD [edi+0*SIZEOF_XMMWORD], xmm2
+ movdqa XMMWORD [edi+1*SIZEOF_XMMWORD], xmm6
+
+ pand xmm1, xmm7 ; xmm1=( 0 -- -- -- -- -- -- --)
+ pand xmm2, xmm7 ; xmm2=( 0 -- -- -- -- -- -- --)
+
+ movdqa XMMWORD [wk(0)], xmm1
+ movdqa XMMWORD [wk(1)], xmm2
+
+ poppic ebx
+
+ add eax, byte SIZEOF_XMMWORD-1
+ and eax, byte -SIZEOF_XMMWORD
+ cmp eax, byte SIZEOF_XMMWORD
+ ja short .columnloop
+ alignx 16, 7
+
+.columnloop_last:
+ ; -- process the last column block
+
+ pushpic ebx
+ movpic ebx, POINTER [gotptr] ; load GOT address
+
+ pcmpeqb xmm1, xmm1
+ pslldq xmm1, (SIZEOF_XMMWORD-2)
+ movdqa xmm2, xmm1
+
+ pand xmm1, XMMWORD [edx+1*SIZEOF_XMMWORD]
+ pand xmm2, XMMWORD [edi+1*SIZEOF_XMMWORD]
+
+ movdqa XMMWORD [wk(2)], xmm1 ; xmm1=(-- -- -- -- -- -- -- 15)
+ movdqa XMMWORD [wk(3)], xmm2 ; xmm2=(-- -- -- -- -- -- -- 15)
+
+ jmp near .upsample
+ alignx 16, 7
+
+.columnloop:
+ ; -- process the next column block
+
+ movdqa xmm0, XMMWORD [ebx+1*SIZEOF_XMMWORD] ; xmm0=row[ 0][1]
+ movdqa xmm1, XMMWORD [ecx+1*SIZEOF_XMMWORD] ; xmm1=row[-1][1]
+ movdqa xmm2, XMMWORD [esi+1*SIZEOF_XMMWORD] ; xmm2=row[+1][1]
+
+ pushpic ebx
+ movpic ebx, POINTER [gotptr] ; load GOT address
+
+ pxor xmm3, xmm3 ; xmm3=(all 0's)
+ movdqa xmm4, xmm0
+ punpcklbw xmm0, xmm3 ; xmm0=row[ 0]( 0 1 2 3 4 5 6 7)
+ punpckhbw xmm4, xmm3 ; xmm4=row[ 0]( 8 9 10 11 12 13 14 15)
+ movdqa xmm5, xmm1
+ punpcklbw xmm1, xmm3 ; xmm1=row[-1]( 0 1 2 3 4 5 6 7)
+ punpckhbw xmm5, xmm3 ; xmm5=row[-1]( 8 9 10 11 12 13 14 15)
+ movdqa xmm6, xmm2
+ punpcklbw xmm2, xmm3 ; xmm2=row[+1]( 0 1 2 3 4 5 6 7)
+ punpckhbw xmm6, xmm3 ; xmm6=row[+1]( 8 9 10 11 12 13 14 15)
+
+ pmullw xmm0, [GOTOFF(ebx,PW_THREE)]
+ pmullw xmm4, [GOTOFF(ebx,PW_THREE)]
+
+ paddw xmm1, xmm0 ; xmm1=Int0L=( 0 1 2 3 4 5 6 7)
+ paddw xmm5, xmm4 ; xmm5=Int0H=( 8 9 10 11 12 13 14 15)
+ paddw xmm2, xmm0 ; xmm2=Int1L=( 0 1 2 3 4 5 6 7)
+ paddw xmm6, xmm4 ; xmm6=Int1H=( 8 9 10 11 12 13 14 15)
+
+ movdqa XMMWORD [edx+2*SIZEOF_XMMWORD], xmm1 ; temporarily save
+ movdqa XMMWORD [edx+3*SIZEOF_XMMWORD], xmm5 ; the intermediate data
+ movdqa XMMWORD [edi+2*SIZEOF_XMMWORD], xmm2
+ movdqa XMMWORD [edi+3*SIZEOF_XMMWORD], xmm6
+
+ pslldq xmm1, (SIZEOF_XMMWORD-2) ; xmm1=(-- -- -- -- -- -- -- 0)
+ pslldq xmm2, (SIZEOF_XMMWORD-2) ; xmm2=(-- -- -- -- -- -- -- 0)
+
+ movdqa XMMWORD [wk(2)], xmm1
+ movdqa XMMWORD [wk(3)], xmm2
+
+.upsample:
+ ; -- process the upper row
+
+ movdqa xmm7, XMMWORD [edx+0*SIZEOF_XMMWORD]
+ movdqa xmm3, XMMWORD [edx+1*SIZEOF_XMMWORD]
+
+ movdqa xmm0, xmm7 ; xmm7=Int0L=( 0 1 2 3 4 5 6 7)
+ movdqa xmm4, xmm3 ; xmm3=Int0H=( 8 9 10 11 12 13 14 15)
+ psrldq xmm0, 2 ; xmm0=( 1 2 3 4 5 6 7 --)
+ pslldq xmm4, (SIZEOF_XMMWORD-2) ; xmm4=(-- -- -- -- -- -- -- 8)
+ movdqa xmm5, xmm7
+ movdqa xmm6, xmm3
+ psrldq xmm5, (SIZEOF_XMMWORD-2) ; xmm5=( 7 -- -- -- -- -- -- --)
+ pslldq xmm6, 2 ; xmm6=(-- 8 9 10 11 12 13 14)
+
+ por xmm0, xmm4 ; xmm0=( 1 2 3 4 5 6 7 8)
+ por xmm5, xmm6 ; xmm5=( 7 8 9 10 11 12 13 14)
+
+ movdqa xmm1, xmm7
+ movdqa xmm2, xmm3
+ pslldq xmm1, 2 ; xmm1=(-- 0 1 2 3 4 5 6)
+ psrldq xmm2, 2 ; xmm2=( 9 10 11 12 13 14 15 --)
+ movdqa xmm4, xmm3
+ psrldq xmm4, (SIZEOF_XMMWORD-2) ; xmm4=(15 -- -- -- -- -- -- --)
+
+ por xmm1, XMMWORD [wk(0)] ; xmm1=(-1 0 1 2 3 4 5 6)
+ por xmm2, XMMWORD [wk(2)] ; xmm2=( 9 10 11 12 13 14 15 16)
+
+ movdqa XMMWORD [wk(0)], xmm4
+
+ pmullw xmm7, [GOTOFF(ebx,PW_THREE)]
+ pmullw xmm3, [GOTOFF(ebx,PW_THREE)]
+ paddw xmm1, [GOTOFF(ebx,PW_EIGHT)]
+ paddw xmm5, [GOTOFF(ebx,PW_EIGHT)]
+ paddw xmm0, [GOTOFF(ebx,PW_SEVEN)]
+ paddw xmm2, [GOTOFF(ebx,PW_SEVEN)]
+
+ paddw xmm1, xmm7
+ paddw xmm5, xmm3
+ psrlw xmm1, 4 ; xmm1=Out0LE=( 0 2 4 6 8 10 12 14)
+ psrlw xmm5, 4 ; xmm5=Out0HE=(16 18 20 22 24 26 28 30)
+ paddw xmm0, xmm7
+ paddw xmm2, xmm3
+ psrlw xmm0, 4 ; xmm0=Out0LO=( 1 3 5 7 9 11 13 15)
+ psrlw xmm2, 4 ; xmm2=Out0HO=(17 19 21 23 25 27 29 31)
+
+ psllw xmm0, BYTE_BIT
+ psllw xmm2, BYTE_BIT
+ por xmm1, xmm0 ; xmm1=Out0L=( 0 1 2 ... 13 14 15)
+ por xmm5, xmm2 ; xmm5=Out0H=(16 17 18 ... 29 30 31)
+
+ movdqa XMMWORD [edx+0*SIZEOF_XMMWORD], xmm1
+ movdqa XMMWORD [edx+1*SIZEOF_XMMWORD], xmm5
+
+ ; -- process the lower row
+
+ movdqa xmm6, XMMWORD [edi+0*SIZEOF_XMMWORD]
+ movdqa xmm4, XMMWORD [edi+1*SIZEOF_XMMWORD]
+
+ movdqa xmm7, xmm6 ; xmm6=Int1L=( 0 1 2 3 4 5 6 7)
+ movdqa xmm3, xmm4 ; xmm4=Int1H=( 8 9 10 11 12 13 14 15)
+ psrldq xmm7, 2 ; xmm7=( 1 2 3 4 5 6 7 --)
+ pslldq xmm3, (SIZEOF_XMMWORD-2) ; xmm3=(-- -- -- -- -- -- -- 8)
+ movdqa xmm0, xmm6
+ movdqa xmm2, xmm4
+ psrldq xmm0, (SIZEOF_XMMWORD-2) ; xmm0=( 7 -- -- -- -- -- -- --)
+ pslldq xmm2, 2 ; xmm2=(-- 8 9 10 11 12 13 14)
+
+ por xmm7, xmm3 ; xmm7=( 1 2 3 4 5 6 7 8)
+ por xmm0, xmm2 ; xmm0=( 7 8 9 10 11 12 13 14)
+
+ movdqa xmm1, xmm6
+ movdqa xmm5, xmm4
+ pslldq xmm1, 2 ; xmm1=(-- 0 1 2 3 4 5 6)
+ psrldq xmm5, 2 ; xmm5=( 9 10 11 12 13 14 15 --)
+ movdqa xmm3, xmm4
+ psrldq xmm3, (SIZEOF_XMMWORD-2) ; xmm3=(15 -- -- -- -- -- -- --)
+
+ por xmm1, XMMWORD [wk(1)] ; xmm1=(-1 0 1 2 3 4 5 6)
+ por xmm5, XMMWORD [wk(3)] ; xmm5=( 9 10 11 12 13 14 15 16)
+
+ movdqa XMMWORD [wk(1)], xmm3
+
+ pmullw xmm6, [GOTOFF(ebx,PW_THREE)]
+ pmullw xmm4, [GOTOFF(ebx,PW_THREE)]
+ paddw xmm1, [GOTOFF(ebx,PW_EIGHT)]
+ paddw xmm0, [GOTOFF(ebx,PW_EIGHT)]
+ paddw xmm7, [GOTOFF(ebx,PW_SEVEN)]
+ paddw xmm5, [GOTOFF(ebx,PW_SEVEN)]
+
+ paddw xmm1, xmm6
+ paddw xmm0, xmm4
+ psrlw xmm1, 4 ; xmm1=Out1LE=( 0 2 4 6 8 10 12 14)
+ psrlw xmm0, 4 ; xmm0=Out1HE=(16 18 20 22 24 26 28 30)
+ paddw xmm7, xmm6
+ paddw xmm5, xmm4
+ psrlw xmm7, 4 ; xmm7=Out1LO=( 1 3 5 7 9 11 13 15)
+ psrlw xmm5, 4 ; xmm5=Out1HO=(17 19 21 23 25 27 29 31)
+
+ psllw xmm7, BYTE_BIT
+ psllw xmm5, BYTE_BIT
+ por xmm1, xmm7 ; xmm1=Out1L=( 0 1 2 ... 13 14 15)
+ por xmm0, xmm5 ; xmm0=Out1H=(16 17 18 ... 29 30 31)
+
+ movdqa XMMWORD [edi+0*SIZEOF_XMMWORD], xmm1
+ movdqa XMMWORD [edi+1*SIZEOF_XMMWORD], xmm0
+
+ poppic ebx
+
+ sub eax, byte SIZEOF_XMMWORD
+ add ecx, byte 1*SIZEOF_XMMWORD ; inptr1(above)
+ add ebx, byte 1*SIZEOF_XMMWORD ; inptr0
+ add esi, byte 1*SIZEOF_XMMWORD ; inptr1(below)
+ add edx, byte 2*SIZEOF_XMMWORD ; outptr0
+ add edi, byte 2*SIZEOF_XMMWORD ; outptr1
+ cmp eax, byte SIZEOF_XMMWORD
+ ja near .columnloop
+ test eax, eax
+ jnz near .columnloop_last
+
+ pop esi
+ pop edi
+ pop ecx
+ pop eax
+
+ add esi, byte 1*SIZEOF_JSAMPROW ; input_data
+ add edi, byte 2*SIZEOF_JSAMPROW ; output_data
+ sub ecx, byte 2 ; rowctr
+ jg near .rowloop
+
+.return:
+ pop edi
+ pop esi
+; pop edx ; need not be preserved
+; pop ecx ; need not be preserved
+ pop ebx
+ mov esp, ebp ; esp <- aligned ebp
+ pop esp ; esp <- original ebp
+ pop ebp
+ ret
+
+; --------------------------------------------------------------------------
+;
+; Fast processing for the common case of 2:1 horizontal and 1:1 vertical.
+; It's still a box filter.
+;
+; GLOBAL(void)
+; jsimd_h2v1_upsample_sse2 (int max_v_samp_factor,
+; JDIMENSION output_width,
+; JSAMPARRAY input_data,
+; JSAMPARRAY *output_data_ptr);
+;
+
+%define max_v_samp(b) (b)+8 ; int max_v_samp_factor
+%define output_width(b) (b)+12 ; JDIMENSION output_width
+%define input_data(b) (b)+16 ; JSAMPARRAY input_data
+%define output_data_ptr(b) (b)+20 ; JSAMPARRAY *output_data_ptr
+
+ align 32
+ global EXTN(jsimd_h2v1_upsample_sse2)
+
+EXTN(jsimd_h2v1_upsample_sse2):
+ push ebp
+ mov ebp, esp
+; push ebx ; unused
+; push ecx ; need not be preserved
+; push edx ; need not be preserved
+ push esi
+ push edi
+
+ mov edx, JDIMENSION [output_width(ebp)]
+ add edx, byte (2*SIZEOF_XMMWORD)-1
+ and edx, byte -(2*SIZEOF_XMMWORD)
+ jz short .return
+
+ mov ecx, INT [max_v_samp(ebp)] ; rowctr
+ test ecx, ecx
+ jz short .return
+
+ mov esi, JSAMPARRAY [input_data(ebp)] ; input_data
+ mov edi, POINTER [output_data_ptr(ebp)]
+ mov edi, JSAMPARRAY [edi] ; output_data
+ alignx 16, 7
+.rowloop:
+ push edi
+ push esi
+
+ mov esi, JSAMPROW [esi] ; inptr
+ mov edi, JSAMPROW [edi] ; outptr
+ mov eax, edx ; colctr
+ alignx 16, 7
+.columnloop:
+
+ movdqa xmm0, XMMWORD [esi+0*SIZEOF_XMMWORD]
+
+ movdqa xmm1, xmm0
+ punpcklbw xmm0, xmm0
+ punpckhbw xmm1, xmm1
+
+ movdqa XMMWORD [edi+0*SIZEOF_XMMWORD], xmm0
+ movdqa XMMWORD [edi+1*SIZEOF_XMMWORD], xmm1
+
+ sub eax, byte 2*SIZEOF_XMMWORD
+ jz short .nextrow
+
+ movdqa xmm2, XMMWORD [esi+1*SIZEOF_XMMWORD]
+
+ movdqa xmm3, xmm2
+ punpcklbw xmm2, xmm2
+ punpckhbw xmm3, xmm3
+
+ movdqa XMMWORD [edi+2*SIZEOF_XMMWORD], xmm2
+ movdqa XMMWORD [edi+3*SIZEOF_XMMWORD], xmm3
+
+ sub eax, byte 2*SIZEOF_XMMWORD
+ jz short .nextrow
+
+ add esi, byte 2*SIZEOF_XMMWORD ; inptr
+ add edi, byte 4*SIZEOF_XMMWORD ; outptr
+ jmp short .columnloop
+ alignx 16, 7
+
+.nextrow:
+ pop esi
+ pop edi
+
+ add esi, byte SIZEOF_JSAMPROW ; input_data
+ add edi, byte SIZEOF_JSAMPROW ; output_data
+ dec ecx ; rowctr
+ jg short .rowloop
+
+.return:
+ pop edi
+ pop esi
+; pop edx ; need not be preserved
+; pop ecx ; need not be preserved
+; pop ebx ; unused
+ pop ebp
+ ret
+
+; --------------------------------------------------------------------------
+;
+; Fast processing for the common case of 2:1 horizontal and 2:1 vertical.
+; It's still a box filter.
+;
+; GLOBAL(void)
+; jsimd_h2v2_upsample_sse2 (nt max_v_samp_factor,
+; JDIMENSION output_width,
+; JSAMPARRAY input_data,
+; JSAMPARRAY *output_data_ptr);
+;
+
+%define max_v_samp(b) (b)+8 ; int max_v_samp_factor
+%define output_width(b) (b)+12 ; JDIMENSION output_width
+%define input_data(b) (b)+16 ; JSAMPARRAY input_data
+%define output_data_ptr(b) (b)+20 ; JSAMPARRAY *output_data_ptr
+
+ align 32
+ global EXTN(jsimd_h2v2_upsample_sse2)
+
+EXTN(jsimd_h2v2_upsample_sse2):
+ push ebp
+ mov ebp, esp
+ push ebx
+; push ecx ; need not be preserved
+; push edx ; need not be preserved
+ push esi
+ push edi
+
+ mov edx, JDIMENSION [output_width(ebp)]
+ add edx, byte (2*SIZEOF_XMMWORD)-1
+ and edx, byte -(2*SIZEOF_XMMWORD)
+ jz near .return
+
+ mov ecx, INT [max_v_samp(ebp)] ; rowctr
+ test ecx, ecx
+ jz near .return
+
+ mov esi, JSAMPARRAY [input_data(ebp)] ; input_data
+ mov edi, POINTER [output_data_ptr(ebp)]
+ mov edi, JSAMPARRAY [edi] ; output_data
+ alignx 16, 7
+.rowloop:
+ push edi
+ push esi
+
+ mov esi, JSAMPROW [esi] ; inptr
+ mov ebx, JSAMPROW [edi+0*SIZEOF_JSAMPROW] ; outptr0
+ mov edi, JSAMPROW [edi+1*SIZEOF_JSAMPROW] ; outptr1
+ mov eax, edx ; colctr
+ alignx 16, 7
+.columnloop:
+
+ movdqa xmm0, XMMWORD [esi+0*SIZEOF_XMMWORD]
+
+ movdqa xmm1, xmm0
+ punpcklbw xmm0, xmm0
+ punpckhbw xmm1, xmm1
+
+ movdqa XMMWORD [ebx+0*SIZEOF_XMMWORD], xmm0
+ movdqa XMMWORD [ebx+1*SIZEOF_XMMWORD], xmm1
+ movdqa XMMWORD [edi+0*SIZEOF_XMMWORD], xmm0
+ movdqa XMMWORD [edi+1*SIZEOF_XMMWORD], xmm1
+
+ sub eax, byte 2*SIZEOF_XMMWORD
+ jz short .nextrow
+
+ movdqa xmm2, XMMWORD [esi+1*SIZEOF_XMMWORD]
+
+ movdqa xmm3, xmm2
+ punpcklbw xmm2, xmm2
+ punpckhbw xmm3, xmm3
+
+ movdqa XMMWORD [ebx+2*SIZEOF_XMMWORD], xmm2
+ movdqa XMMWORD [ebx+3*SIZEOF_XMMWORD], xmm3
+ movdqa XMMWORD [edi+2*SIZEOF_XMMWORD], xmm2
+ movdqa XMMWORD [edi+3*SIZEOF_XMMWORD], xmm3
+
+ sub eax, byte 2*SIZEOF_XMMWORD
+ jz short .nextrow
+
+ add esi, byte 2*SIZEOF_XMMWORD ; inptr
+ add ebx, byte 4*SIZEOF_XMMWORD ; outptr0
+ add edi, byte 4*SIZEOF_XMMWORD ; outptr1
+ jmp short .columnloop
+ alignx 16, 7
+
+.nextrow:
+ pop esi
+ pop edi
+
+ add esi, byte 1*SIZEOF_JSAMPROW ; input_data
+ add edi, byte 2*SIZEOF_JSAMPROW ; output_data
+ sub ecx, byte 2 ; rowctr
+ jg short .rowloop
+
+.return:
+ pop edi
+ pop esi
+; pop edx ; need not be preserved
+; pop ecx ; need not be preserved
+ pop ebx
+ pop ebp
+ ret
+
+; For some reason, the OS X linker does not honor the request to align the
+; segment unless we do this.
+ align 32
diff --git a/simd/jfdctflt-3dn.asm b/simd/i386/jfdctflt-3dn.asm
similarity index 98%
rename from simd/jfdctflt-3dn.asm
rename to simd/i386/jfdctflt-3dn.asm
index 2191618..c3d8fba 100644
--- a/simd/jfdctflt-3dn.asm
+++ b/simd/i386/jfdctflt-3dn.asm
@@ -2,6 +2,7 @@
; jfdctflt.asm - floating-point FDCT (3DNow!)
;
; Copyright 2009 Pierre Ossman <ossman@cendio.se> for Cendio AB
+; Copyright (C) 2016, D. R. Commander.
;
; Based on the x86 SIMD extension for IJG JPEG library
; Copyright (C) 1999-2006, MIYASAKA Masaru.
@@ -25,7 +26,7 @@
; --------------------------------------------------------------------------
SECTION SEG_CONST
- alignz 16
+ alignz 32
global EXTN(jconst_fdct_float_3dnow)
EXTN(jconst_fdct_float_3dnow):
@@ -35,7 +36,7 @@
PD_0_541 times 2 dd 0.541196100146196984399723
PD_1_306 times 2 dd 1.306562964876376527856643
- alignz 16
+ alignz 32
; --------------------------------------------------------------------------
SECTION SEG_TEXT
@@ -53,7 +54,7 @@
%define wk(i) ebp-(WK_NUM-(i))*SIZEOF_MMWORD ; mmword wk[WK_NUM]
%define WK_NUM 2
- align 16
+ align 32
global EXTN(jsimd_fdct_float_3dnow)
EXTN(jsimd_fdct_float_3dnow):
@@ -316,4 +317,4 @@
; For some reason, the OS X linker does not honor the request to align the
; segment unless we do this.
- align 16
+ align 32
diff --git a/simd/i386/jfdctflt-sse.asm b/simd/i386/jfdctflt-sse.asm
new file mode 100644
index 0000000..b15eb5c
--- /dev/null
+++ b/simd/i386/jfdctflt-sse.asm
@@ -0,0 +1,370 @@
+;
+; jfdctflt.asm - floating-point FDCT (SSE)
+;
+; Copyright 2009 Pierre Ossman <ossman@cendio.se> for Cendio AB
+; Copyright (C) 2016, D. R. Commander.
+;
+; Based on the x86 SIMD extension for IJG JPEG library
+; Copyright (C) 1999-2006, MIYASAKA Masaru.
+; For conditions of distribution and use, see copyright notice in jsimdext.inc
+;
+; This file should be assembled with NASM (Netwide Assembler),
+; can *not* be assembled with Microsoft's MASM or any compatible
+; assembler (including Borland's Turbo Assembler).
+; NASM is available from http://nasm.sourceforge.net/ or
+; http://sourceforge.net/project/showfiles.php?group_id=6208
+;
+; This file contains a floating-point implementation of the forward DCT
+; (Discrete Cosine Transform). The following code is based directly on
+; the IJG's original jfdctflt.c; see the jfdctflt.c for more details.
+;
+; [TAB8]
+
+%include "jsimdext.inc"
+%include "jdct.inc"
+
+; --------------------------------------------------------------------------
+
+%macro unpcklps2 2 ; %1=(0 1 2 3) / %2=(4 5 6 7) => %1=(0 1 4 5)
+ shufps %1, %2, 0x44
+%endmacro
+
+%macro unpckhps2 2 ; %1=(0 1 2 3) / %2=(4 5 6 7) => %1=(2 3 6 7)
+ shufps %1, %2, 0xEE
+%endmacro
+
+; --------------------------------------------------------------------------
+ SECTION SEG_CONST
+
+ alignz 32
+ global EXTN(jconst_fdct_float_sse)
+
+EXTN(jconst_fdct_float_sse):
+
+PD_0_382 times 4 dd 0.382683432365089771728460
+PD_0_707 times 4 dd 0.707106781186547524400844
+PD_0_541 times 4 dd 0.541196100146196984399723
+PD_1_306 times 4 dd 1.306562964876376527856643
+
+ alignz 32
+
+; --------------------------------------------------------------------------
+ SECTION SEG_TEXT
+ BITS 32
+;
+; Perform the forward DCT on one block of samples.
+;
+; GLOBAL(void)
+; jsimd_fdct_float_sse (FAST_FLOAT *data)
+;
+
+%define data(b) (b)+8 ; FAST_FLOAT *data
+
+%define original_ebp ebp+0
+%define wk(i) ebp-(WK_NUM-(i))*SIZEOF_XMMWORD ; xmmword wk[WK_NUM]
+%define WK_NUM 2
+
+ align 32
+ global EXTN(jsimd_fdct_float_sse)
+
+EXTN(jsimd_fdct_float_sse):
+ push ebp
+ mov eax, esp ; eax = original ebp
+ sub esp, byte 4
+ and esp, byte (-SIZEOF_XMMWORD) ; align to 128 bits
+ mov [esp], eax
+ mov ebp, esp ; ebp = aligned ebp
+ lea esp, [wk(0)]
+ pushpic ebx
+; push ecx ; need not be preserved
+; push edx ; need not be preserved
+; push esi ; unused
+; push edi ; unused
+
+ get_GOT ebx ; get GOT address
+
+ ; ---- Pass 1: process rows.
+
+ mov edx, POINTER [data(eax)] ; (FAST_FLOAT *)
+ mov ecx, DCTSIZE/4
+ alignx 16, 7
+.rowloop:
+
+ movaps xmm0, XMMWORD [XMMBLOCK(2,0,edx,SIZEOF_FAST_FLOAT)]
+ movaps xmm1, XMMWORD [XMMBLOCK(3,0,edx,SIZEOF_FAST_FLOAT)]
+ movaps xmm2, XMMWORD [XMMBLOCK(2,1,edx,SIZEOF_FAST_FLOAT)]
+ movaps xmm3, XMMWORD [XMMBLOCK(3,1,edx,SIZEOF_FAST_FLOAT)]
+
+ ; xmm0=(20 21 22 23), xmm2=(24 25 26 27)
+ ; xmm1=(30 31 32 33), xmm3=(34 35 36 37)
+
+ movaps xmm4, xmm0 ; transpose coefficients(phase 1)
+ unpcklps xmm0, xmm1 ; xmm0=(20 30 21 31)
+ unpckhps xmm4, xmm1 ; xmm4=(22 32 23 33)
+ movaps xmm5, xmm2 ; transpose coefficients(phase 1)
+ unpcklps xmm2, xmm3 ; xmm2=(24 34 25 35)
+ unpckhps xmm5, xmm3 ; xmm5=(26 36 27 37)
+
+ movaps xmm6, XMMWORD [XMMBLOCK(0,0,edx,SIZEOF_FAST_FLOAT)]
+ movaps xmm7, XMMWORD [XMMBLOCK(1,0,edx,SIZEOF_FAST_FLOAT)]
+ movaps xmm1, XMMWORD [XMMBLOCK(0,1,edx,SIZEOF_FAST_FLOAT)]
+ movaps xmm3, XMMWORD [XMMBLOCK(1,1,edx,SIZEOF_FAST_FLOAT)]
+
+ ; xmm6=(00 01 02 03), xmm1=(04 05 06 07)
+ ; xmm7=(10 11 12 13), xmm3=(14 15 16 17)
+
+ movaps XMMWORD [wk(0)], xmm4 ; wk(0)=(22 32 23 33)
+ movaps XMMWORD [wk(1)], xmm2 ; wk(1)=(24 34 25 35)
+
+ movaps xmm4, xmm6 ; transpose coefficients(phase 1)
+ unpcklps xmm6, xmm7 ; xmm6=(00 10 01 11)
+ unpckhps xmm4, xmm7 ; xmm4=(02 12 03 13)
+ movaps xmm2, xmm1 ; transpose coefficients(phase 1)
+ unpcklps xmm1, xmm3 ; xmm1=(04 14 05 15)
+ unpckhps xmm2, xmm3 ; xmm2=(06 16 07 17)
+
+ movaps xmm7, xmm6 ; transpose coefficients(phase 2)
+ unpcklps2 xmm6, xmm0 ; xmm6=(00 10 20 30)=data0
+ unpckhps2 xmm7, xmm0 ; xmm7=(01 11 21 31)=data1
+ movaps xmm3, xmm2 ; transpose coefficients(phase 2)
+ unpcklps2 xmm2, xmm5 ; xmm2=(06 16 26 36)=data6
+ unpckhps2 xmm3, xmm5 ; xmm3=(07 17 27 37)=data7
+
+ movaps xmm0, xmm7
+ movaps xmm5, xmm6
+ subps xmm7, xmm2 ; xmm7=data1-data6=tmp6
+ subps xmm6, xmm3 ; xmm6=data0-data7=tmp7
+ addps xmm0, xmm2 ; xmm0=data1+data6=tmp1
+ addps xmm5, xmm3 ; xmm5=data0+data7=tmp0
+
+ movaps xmm2, XMMWORD [wk(0)] ; xmm2=(22 32 23 33)
+ movaps xmm3, XMMWORD [wk(1)] ; xmm3=(24 34 25 35)
+ movaps XMMWORD [wk(0)], xmm7 ; wk(0)=tmp6
+ movaps XMMWORD [wk(1)], xmm6 ; wk(1)=tmp7
+
+ movaps xmm7, xmm4 ; transpose coefficients(phase 2)
+ unpcklps2 xmm4, xmm2 ; xmm4=(02 12 22 32)=data2
+ unpckhps2 xmm7, xmm2 ; xmm7=(03 13 23 33)=data3
+ movaps xmm6, xmm1 ; transpose coefficients(phase 2)
+ unpcklps2 xmm1, xmm3 ; xmm1=(04 14 24 34)=data4
+ unpckhps2 xmm6, xmm3 ; xmm6=(05 15 25 35)=data5
+
+ movaps xmm2, xmm7
+ movaps xmm3, xmm4
+ addps xmm7, xmm1 ; xmm7=data3+data4=tmp3
+ addps xmm4, xmm6 ; xmm4=data2+data5=tmp2
+ subps xmm2, xmm1 ; xmm2=data3-data4=tmp4
+ subps xmm3, xmm6 ; xmm3=data2-data5=tmp5
+
+ ; -- Even part
+
+ movaps xmm1, xmm5
+ movaps xmm6, xmm0
+ subps xmm5, xmm7 ; xmm5=tmp13
+ subps xmm0, xmm4 ; xmm0=tmp12
+ addps xmm1, xmm7 ; xmm1=tmp10
+ addps xmm6, xmm4 ; xmm6=tmp11
+
+ addps xmm0, xmm5
+ mulps xmm0, [GOTOFF(ebx,PD_0_707)] ; xmm0=z1
+
+ movaps xmm7, xmm1
+ movaps xmm4, xmm5
+ subps xmm1, xmm6 ; xmm1=data4
+ subps xmm5, xmm0 ; xmm5=data6
+ addps xmm7, xmm6 ; xmm7=data0
+ addps xmm4, xmm0 ; xmm4=data2
+
+ movaps XMMWORD [XMMBLOCK(0,1,edx,SIZEOF_FAST_FLOAT)], xmm1
+ movaps XMMWORD [XMMBLOCK(2,1,edx,SIZEOF_FAST_FLOAT)], xmm5
+ movaps XMMWORD [XMMBLOCK(0,0,edx,SIZEOF_FAST_FLOAT)], xmm7
+ movaps XMMWORD [XMMBLOCK(2,0,edx,SIZEOF_FAST_FLOAT)], xmm4
+
+ ; -- Odd part
+
+ movaps xmm6, XMMWORD [wk(0)] ; xmm6=tmp6
+ movaps xmm0, XMMWORD [wk(1)] ; xmm0=tmp7
+
+ addps xmm2, xmm3 ; xmm2=tmp10
+ addps xmm3, xmm6 ; xmm3=tmp11
+ addps xmm6, xmm0 ; xmm6=tmp12, xmm0=tmp7
+
+ mulps xmm3, [GOTOFF(ebx,PD_0_707)] ; xmm3=z3
+
+ movaps xmm1, xmm2 ; xmm1=tmp10
+ subps xmm2, xmm6
+ mulps xmm2, [GOTOFF(ebx,PD_0_382)] ; xmm2=z5
+ mulps xmm1, [GOTOFF(ebx,PD_0_541)] ; xmm1=MULTIPLY(tmp10,FIX_0_541196)
+ mulps xmm6, [GOTOFF(ebx,PD_1_306)] ; xmm6=MULTIPLY(tmp12,FIX_1_306562)
+ addps xmm1, xmm2 ; xmm1=z2
+ addps xmm6, xmm2 ; xmm6=z4
+
+ movaps xmm5, xmm0
+ subps xmm0, xmm3 ; xmm0=z13
+ addps xmm5, xmm3 ; xmm5=z11
+
+ movaps xmm7, xmm0
+ movaps xmm4, xmm5
+ subps xmm0, xmm1 ; xmm0=data3
+ subps xmm5, xmm6 ; xmm5=data7
+ addps xmm7, xmm1 ; xmm7=data5
+ addps xmm4, xmm6 ; xmm4=data1
+
+ movaps XMMWORD [XMMBLOCK(3,0,edx,SIZEOF_FAST_FLOAT)], xmm0
+ movaps XMMWORD [XMMBLOCK(3,1,edx,SIZEOF_FAST_FLOAT)], xmm5
+ movaps XMMWORD [XMMBLOCK(1,1,edx,SIZEOF_FAST_FLOAT)], xmm7
+ movaps XMMWORD [XMMBLOCK(1,0,edx,SIZEOF_FAST_FLOAT)], xmm4
+
+ add edx, 4*DCTSIZE*SIZEOF_FAST_FLOAT
+ dec ecx
+ jnz near .rowloop
+
+ ; ---- Pass 2: process columns.
+
+ mov edx, POINTER [data(eax)] ; (FAST_FLOAT *)
+ mov ecx, DCTSIZE/4
+ alignx 16, 7
+.columnloop:
+
+ movaps xmm0, XMMWORD [XMMBLOCK(2,0,edx,SIZEOF_FAST_FLOAT)]
+ movaps xmm1, XMMWORD [XMMBLOCK(3,0,edx,SIZEOF_FAST_FLOAT)]
+ movaps xmm2, XMMWORD [XMMBLOCK(6,0,edx,SIZEOF_FAST_FLOAT)]
+ movaps xmm3, XMMWORD [XMMBLOCK(7,0,edx,SIZEOF_FAST_FLOAT)]
+
+ ; xmm0=(02 12 22 32), xmm2=(42 52 62 72)
+ ; xmm1=(03 13 23 33), xmm3=(43 53 63 73)
+
+ movaps xmm4, xmm0 ; transpose coefficients(phase 1)
+ unpcklps xmm0, xmm1 ; xmm0=(02 03 12 13)
+ unpckhps xmm4, xmm1 ; xmm4=(22 23 32 33)
+ movaps xmm5, xmm2 ; transpose coefficients(phase 1)
+ unpcklps xmm2, xmm3 ; xmm2=(42 43 52 53)
+ unpckhps xmm5, xmm3 ; xmm5=(62 63 72 73)
+
+ movaps xmm6, XMMWORD [XMMBLOCK(0,0,edx,SIZEOF_FAST_FLOAT)]
+ movaps xmm7, XMMWORD [XMMBLOCK(1,0,edx,SIZEOF_FAST_FLOAT)]
+ movaps xmm1, XMMWORD [XMMBLOCK(4,0,edx,SIZEOF_FAST_FLOAT)]
+ movaps xmm3, XMMWORD [XMMBLOCK(5,0,edx,SIZEOF_FAST_FLOAT)]
+
+ ; xmm6=(00 10 20 30), xmm1=(40 50 60 70)
+ ; xmm7=(01 11 21 31), xmm3=(41 51 61 71)
+
+ movaps XMMWORD [wk(0)], xmm4 ; wk(0)=(22 23 32 33)
+ movaps XMMWORD [wk(1)], xmm2 ; wk(1)=(42 43 52 53)
+
+ movaps xmm4, xmm6 ; transpose coefficients(phase 1)
+ unpcklps xmm6, xmm7 ; xmm6=(00 01 10 11)
+ unpckhps xmm4, xmm7 ; xmm4=(20 21 30 31)
+ movaps xmm2, xmm1 ; transpose coefficients(phase 1)
+ unpcklps xmm1, xmm3 ; xmm1=(40 41 50 51)
+ unpckhps xmm2, xmm3 ; xmm2=(60 61 70 71)
+
+ movaps xmm7, xmm6 ; transpose coefficients(phase 2)
+ unpcklps2 xmm6, xmm0 ; xmm6=(00 01 02 03)=data0
+ unpckhps2 xmm7, xmm0 ; xmm7=(10 11 12 13)=data1
+ movaps xmm3, xmm2 ; transpose coefficients(phase 2)
+ unpcklps2 xmm2, xmm5 ; xmm2=(60 61 62 63)=data6
+ unpckhps2 xmm3, xmm5 ; xmm3=(70 71 72 73)=data7
+
+ movaps xmm0, xmm7
+ movaps xmm5, xmm6
+ subps xmm7, xmm2 ; xmm7=data1-data6=tmp6
+ subps xmm6, xmm3 ; xmm6=data0-data7=tmp7
+ addps xmm0, xmm2 ; xmm0=data1+data6=tmp1
+ addps xmm5, xmm3 ; xmm5=data0+data7=tmp0
+
+ movaps xmm2, XMMWORD [wk(0)] ; xmm2=(22 23 32 33)
+ movaps xmm3, XMMWORD [wk(1)] ; xmm3=(42 43 52 53)
+ movaps XMMWORD [wk(0)], xmm7 ; wk(0)=tmp6
+ movaps XMMWORD [wk(1)], xmm6 ; wk(1)=tmp7
+
+ movaps xmm7, xmm4 ; transpose coefficients(phase 2)
+ unpcklps2 xmm4, xmm2 ; xmm4=(20 21 22 23)=data2
+ unpckhps2 xmm7, xmm2 ; xmm7=(30 31 32 33)=data3
+ movaps xmm6, xmm1 ; transpose coefficients(phase 2)
+ unpcklps2 xmm1, xmm3 ; xmm1=(40 41 42 43)=data4
+ unpckhps2 xmm6, xmm3 ; xmm6=(50 51 52 53)=data5
+
+ movaps xmm2, xmm7
+ movaps xmm3, xmm4
+ addps xmm7, xmm1 ; xmm7=data3+data4=tmp3
+ addps xmm4, xmm6 ; xmm4=data2+data5=tmp2
+ subps xmm2, xmm1 ; xmm2=data3-data4=tmp4
+ subps xmm3, xmm6 ; xmm3=data2-data5=tmp5
+
+ ; -- Even part
+
+ movaps xmm1, xmm5
+ movaps xmm6, xmm0
+ subps xmm5, xmm7 ; xmm5=tmp13
+ subps xmm0, xmm4 ; xmm0=tmp12
+ addps xmm1, xmm7 ; xmm1=tmp10
+ addps xmm6, xmm4 ; xmm6=tmp11
+
+ addps xmm0, xmm5
+ mulps xmm0, [GOTOFF(ebx,PD_0_707)] ; xmm0=z1
+
+ movaps xmm7, xmm1
+ movaps xmm4, xmm5
+ subps xmm1, xmm6 ; xmm1=data4
+ subps xmm5, xmm0 ; xmm5=data6
+ addps xmm7, xmm6 ; xmm7=data0
+ addps xmm4, xmm0 ; xmm4=data2
+
+ movaps XMMWORD [XMMBLOCK(4,0,edx,SIZEOF_FAST_FLOAT)], xmm1
+ movaps XMMWORD [XMMBLOCK(6,0,edx,SIZEOF_FAST_FLOAT)], xmm5
+ movaps XMMWORD [XMMBLOCK(0,0,edx,SIZEOF_FAST_FLOAT)], xmm7
+ movaps XMMWORD [XMMBLOCK(2,0,edx,SIZEOF_FAST_FLOAT)], xmm4
+
+ ; -- Odd part
+
+ movaps xmm6, XMMWORD [wk(0)] ; xmm6=tmp6
+ movaps xmm0, XMMWORD [wk(1)] ; xmm0=tmp7
+
+ addps xmm2, xmm3 ; xmm2=tmp10
+ addps xmm3, xmm6 ; xmm3=tmp11
+ addps xmm6, xmm0 ; xmm6=tmp12, xmm0=tmp7
+
+ mulps xmm3, [GOTOFF(ebx,PD_0_707)] ; xmm3=z3
+
+ movaps xmm1, xmm2 ; xmm1=tmp10
+ subps xmm2, xmm6
+ mulps xmm2, [GOTOFF(ebx,PD_0_382)] ; xmm2=z5
+ mulps xmm1, [GOTOFF(ebx,PD_0_541)] ; xmm1=MULTIPLY(tmp10,FIX_0_541196)
+ mulps xmm6, [GOTOFF(ebx,PD_1_306)] ; xmm6=MULTIPLY(tmp12,FIX_1_306562)
+ addps xmm1, xmm2 ; xmm1=z2
+ addps xmm6, xmm2 ; xmm6=z4
+
+ movaps xmm5, xmm0
+ subps xmm0, xmm3 ; xmm0=z13
+ addps xmm5, xmm3 ; xmm5=z11
+
+ movaps xmm7, xmm0
+ movaps xmm4, xmm5
+ subps xmm0, xmm1 ; xmm0=data3
+ subps xmm5, xmm6 ; xmm5=data7
+ addps xmm7, xmm1 ; xmm7=data5
+ addps xmm4, xmm6 ; xmm4=data1
+
+ movaps XMMWORD [XMMBLOCK(3,0,edx,SIZEOF_FAST_FLOAT)], xmm0
+ movaps XMMWORD [XMMBLOCK(7,0,edx,SIZEOF_FAST_FLOAT)], xmm5
+ movaps XMMWORD [XMMBLOCK(5,0,edx,SIZEOF_FAST_FLOAT)], xmm7
+ movaps XMMWORD [XMMBLOCK(1,0,edx,SIZEOF_FAST_FLOAT)], xmm4
+
+ add edx, byte 4*SIZEOF_FAST_FLOAT
+ dec ecx
+ jnz near .columnloop
+
+; pop edi ; unused
+; pop esi ; unused
+; pop edx ; need not be preserved
+; pop ecx ; need not be preserved
+ poppic ebx
+ mov esp, ebp ; esp <- aligned ebp
+ pop esp ; esp <- original ebp
+ pop ebp
+ ret
+
+; For some reason, the OS X linker does not honor the request to align the
+; segment unless we do this.
+ align 32
diff --git a/simd/jfdctfst-mmx.asm b/simd/i386/jfdctfst-mmx.asm
similarity index 98%
rename from simd/jfdctfst-mmx.asm
rename to simd/i386/jfdctfst-mmx.asm
index eb2eb9c..f274924 100644
--- a/simd/jfdctfst-mmx.asm
+++ b/simd/i386/jfdctfst-mmx.asm
@@ -2,6 +2,7 @@
; jfdctfst.asm - fast integer FDCT (MMX)
;
; Copyright 2009 Pierre Ossman <ossman@cendio.se> for Cendio AB
+; Copyright (C) 2016, D. R. Commander.
;
; Based on the x86 SIMD extension for IJG JPEG library
; Copyright (C) 1999-2006, MIYASAKA Masaru.
@@ -50,7 +51,7 @@
%define PRE_MULTIPLY_SCALE_BITS 2
%define CONST_SHIFT (16 - PRE_MULTIPLY_SCALE_BITS - CONST_BITS)
- alignz 16
+ alignz 32
global EXTN(jconst_fdct_ifast_mmx)
EXTN(jconst_fdct_ifast_mmx):
@@ -60,7 +61,7 @@
PW_F0541 times 4 dw F_0_541 << CONST_SHIFT
PW_F1306 times 4 dw F_1_306 << CONST_SHIFT
- alignz 16
+ alignz 32
; --------------------------------------------------------------------------
SECTION SEG_TEXT
@@ -78,7 +79,7 @@
%define wk(i) ebp-(WK_NUM-(i))*SIZEOF_MMWORD ; mmword wk[WK_NUM]
%define WK_NUM 2
- align 16
+ align 32
global EXTN(jsimd_fdct_ifast_mmx)
EXTN(jsimd_fdct_ifast_mmx):
@@ -393,4 +394,4 @@
; For some reason, the OS X linker does not honor the request to align the
; segment unless we do this.
- align 16
+ align 32
diff --git a/simd/i386/jfdctfst-sse2.asm b/simd/i386/jfdctfst-sse2.asm
new file mode 100644
index 0000000..fb140fa
--- /dev/null
+++ b/simd/i386/jfdctfst-sse2.asm
@@ -0,0 +1,404 @@
+;
+; jfdctfst.asm - fast integer FDCT (SSE2)
+;
+; Copyright 2009 Pierre Ossman <ossman@cendio.se> for Cendio AB
+; Copyright (C) 2016, D. R. Commander.
+;
+; Based on the x86 SIMD extension for IJG JPEG library
+; Copyright (C) 1999-2006, MIYASAKA Masaru.
+; For conditions of distribution and use, see copyright notice in jsimdext.inc
+;
+; This file should be assembled with NASM (Netwide Assembler),
+; can *not* be assembled with Microsoft's MASM or any compatible
+; assembler (including Borland's Turbo Assembler).
+; NASM is available from http://nasm.sourceforge.net/ or
+; http://sourceforge.net/project/showfiles.php?group_id=6208
+;
+; This file contains a fast, not so accurate integer implementation of
+; the forward DCT (Discrete Cosine Transform). The following code is
+; based directly on the IJG's original jfdctfst.c; see the jfdctfst.c
+; for more details.
+;
+; [TAB8]
+
+%include "jsimdext.inc"
+%include "jdct.inc"
+
+; --------------------------------------------------------------------------
+
+%define CONST_BITS 8 ; 14 is also OK.
+
+%if CONST_BITS == 8
+F_0_382 equ 98 ; FIX(0.382683433)
+F_0_541 equ 139 ; FIX(0.541196100)
+F_0_707 equ 181 ; FIX(0.707106781)
+F_1_306 equ 334 ; FIX(1.306562965)
+%else
+; NASM cannot do compile-time arithmetic on floating-point constants.
+%define DESCALE(x,n) (((x)+(1<<((n)-1)))>>(n))
+F_0_382 equ DESCALE( 410903207, 30-CONST_BITS) ; FIX(0.382683433)
+F_0_541 equ DESCALE( 581104887, 30-CONST_BITS) ; FIX(0.541196100)
+F_0_707 equ DESCALE( 759250124, 30-CONST_BITS) ; FIX(0.707106781)
+F_1_306 equ DESCALE(1402911301, 30-CONST_BITS) ; FIX(1.306562965)
+%endif
+
+; --------------------------------------------------------------------------
+ SECTION SEG_CONST
+
+; PRE_MULTIPLY_SCALE_BITS <= 2 (to avoid overflow)
+; CONST_BITS + CONST_SHIFT + PRE_MULTIPLY_SCALE_BITS == 16 (for pmulhw)
+
+%define PRE_MULTIPLY_SCALE_BITS 2
+%define CONST_SHIFT (16 - PRE_MULTIPLY_SCALE_BITS - CONST_BITS)
+
+ alignz 32
+ global EXTN(jconst_fdct_ifast_sse2)
+
+EXTN(jconst_fdct_ifast_sse2):
+
+PW_F0707 times 8 dw F_0_707 << CONST_SHIFT
+PW_F0382 times 8 dw F_0_382 << CONST_SHIFT
+PW_F0541 times 8 dw F_0_541 << CONST_SHIFT
+PW_F1306 times 8 dw F_1_306 << CONST_SHIFT
+
+ alignz 32
+
+; --------------------------------------------------------------------------
+ SECTION SEG_TEXT
+ BITS 32
+;
+; Perform the forward DCT on one block of samples.
+;
+; GLOBAL(void)
+; jsimd_fdct_ifast_sse2 (DCTELEM *data)
+;
+
+%define data(b) (b)+8 ; DCTELEM *data
+
+%define original_ebp ebp+0
+%define wk(i) ebp-(WK_NUM-(i))*SIZEOF_XMMWORD ; xmmword wk[WK_NUM]
+%define WK_NUM 2
+
+ align 32
+ global EXTN(jsimd_fdct_ifast_sse2)
+
+EXTN(jsimd_fdct_ifast_sse2):
+ push ebp
+ mov eax, esp ; eax = original ebp
+ sub esp, byte 4
+ and esp, byte (-SIZEOF_XMMWORD) ; align to 128 bits
+ mov [esp], eax
+ mov ebp, esp ; ebp = aligned ebp
+ lea esp, [wk(0)]
+ pushpic ebx
+; push ecx ; unused
+; push edx ; need not be preserved
+; push esi ; unused
+; push edi ; unused
+
+ get_GOT ebx ; get GOT address
+
+ ; ---- Pass 1: process rows.
+
+ mov edx, POINTER [data(eax)] ; (DCTELEM *)
+
+ movdqa xmm0, XMMWORD [XMMBLOCK(0,0,edx,SIZEOF_DCTELEM)]
+ movdqa xmm1, XMMWORD [XMMBLOCK(1,0,edx,SIZEOF_DCTELEM)]
+ movdqa xmm2, XMMWORD [XMMBLOCK(2,0,edx,SIZEOF_DCTELEM)]
+ movdqa xmm3, XMMWORD [XMMBLOCK(3,0,edx,SIZEOF_DCTELEM)]
+
+ ; xmm0=(00 01 02 03 04 05 06 07), xmm2=(20 21 22 23 24 25 26 27)
+ ; xmm1=(10 11 12 13 14 15 16 17), xmm3=(30 31 32 33 34 35 36 37)
+
+ movdqa xmm4, xmm0 ; transpose coefficients(phase 1)
+ punpcklwd xmm0, xmm1 ; xmm0=(00 10 01 11 02 12 03 13)
+ punpckhwd xmm4, xmm1 ; xmm4=(04 14 05 15 06 16 07 17)
+ movdqa xmm5, xmm2 ; transpose coefficients(phase 1)
+ punpcklwd xmm2, xmm3 ; xmm2=(20 30 21 31 22 32 23 33)
+ punpckhwd xmm5, xmm3 ; xmm5=(24 34 25 35 26 36 27 37)
+
+ movdqa xmm6, XMMWORD [XMMBLOCK(4,0,edx,SIZEOF_DCTELEM)]
+ movdqa xmm7, XMMWORD [XMMBLOCK(5,0,edx,SIZEOF_DCTELEM)]
+ movdqa xmm1, XMMWORD [XMMBLOCK(6,0,edx,SIZEOF_DCTELEM)]
+ movdqa xmm3, XMMWORD [XMMBLOCK(7,0,edx,SIZEOF_DCTELEM)]
+
+ ; xmm6=( 4 12 20 28 36 44 52 60), xmm1=( 6 14 22 30 38 46 54 62)
+ ; xmm7=( 5 13 21 29 37 45 53 61), xmm3=( 7 15 23 31 39 47 55 63)
+
+ movdqa XMMWORD [wk(0)], xmm2 ; wk(0)=(20 30 21 31 22 32 23 33)
+ movdqa XMMWORD [wk(1)], xmm5 ; wk(1)=(24 34 25 35 26 36 27 37)
+
+ movdqa xmm2, xmm6 ; transpose coefficients(phase 1)
+ punpcklwd xmm6, xmm7 ; xmm6=(40 50 41 51 42 52 43 53)
+ punpckhwd xmm2, xmm7 ; xmm2=(44 54 45 55 46 56 47 57)
+ movdqa xmm5, xmm1 ; transpose coefficients(phase 1)
+ punpcklwd xmm1, xmm3 ; xmm1=(60 70 61 71 62 72 63 73)
+ punpckhwd xmm5, xmm3 ; xmm5=(64 74 65 75 66 76 67 77)
+
+ movdqa xmm7, xmm6 ; transpose coefficients(phase 2)
+ punpckldq xmm6, xmm1 ; xmm6=(40 50 60 70 41 51 61 71)
+ punpckhdq xmm7, xmm1 ; xmm7=(42 52 62 72 43 53 63 73)
+ movdqa xmm3, xmm2 ; transpose coefficients(phase 2)
+ punpckldq xmm2, xmm5 ; xmm2=(44 54 64 74 45 55 65 75)
+ punpckhdq xmm3, xmm5 ; xmm3=(46 56 66 76 47 57 67 77)
+
+ movdqa xmm1, XMMWORD [wk(0)] ; xmm1=(20 30 21 31 22 32 23 33)
+ movdqa xmm5, XMMWORD [wk(1)] ; xmm5=(24 34 25 35 26 36 27 37)
+ movdqa XMMWORD [wk(0)], xmm7 ; wk(0)=(42 52 62 72 43 53 63 73)
+ movdqa XMMWORD [wk(1)], xmm2 ; wk(1)=(44 54 64 74 45 55 65 75)
+
+ movdqa xmm7, xmm0 ; transpose coefficients(phase 2)
+ punpckldq xmm0, xmm1 ; xmm0=(00 10 20 30 01 11 21 31)
+ punpckhdq xmm7, xmm1 ; xmm7=(02 12 22 32 03 13 23 33)
+ movdqa xmm2, xmm4 ; transpose coefficients(phase 2)
+ punpckldq xmm4, xmm5 ; xmm4=(04 14 24 34 05 15 25 35)
+ punpckhdq xmm2, xmm5 ; xmm2=(06 16 26 36 07 17 27 37)
+
+ movdqa xmm1, xmm0 ; transpose coefficients(phase 3)
+ punpcklqdq xmm0, xmm6 ; xmm0=(00 10 20 30 40 50 60 70)=data0
+ punpckhqdq xmm1, xmm6 ; xmm1=(01 11 21 31 41 51 61 71)=data1
+ movdqa xmm5, xmm2 ; transpose coefficients(phase 3)
+ punpcklqdq xmm2, xmm3 ; xmm2=(06 16 26 36 46 56 66 76)=data6
+ punpckhqdq xmm5, xmm3 ; xmm5=(07 17 27 37 47 57 67 77)=data7
+
+ movdqa xmm6, xmm1
+ movdqa xmm3, xmm0
+ psubw xmm1, xmm2 ; xmm1=data1-data6=tmp6
+ psubw xmm0, xmm5 ; xmm0=data0-data7=tmp7
+ paddw xmm6, xmm2 ; xmm6=data1+data6=tmp1
+ paddw xmm3, xmm5 ; xmm3=data0+data7=tmp0
+
+ movdqa xmm2, XMMWORD [wk(0)] ; xmm2=(42 52 62 72 43 53 63 73)
+ movdqa xmm5, XMMWORD [wk(1)] ; xmm5=(44 54 64 74 45 55 65 75)
+ movdqa XMMWORD [wk(0)], xmm1 ; wk(0)=tmp6
+ movdqa XMMWORD [wk(1)], xmm0 ; wk(1)=tmp7
+
+ movdqa xmm1, xmm7 ; transpose coefficients(phase 3)
+ punpcklqdq xmm7, xmm2 ; xmm7=(02 12 22 32 42 52 62 72)=data2
+ punpckhqdq xmm1, xmm2 ; xmm1=(03 13 23 33 43 53 63 73)=data3
+ movdqa xmm0, xmm4 ; transpose coefficients(phase 3)
+ punpcklqdq xmm4, xmm5 ; xmm4=(04 14 24 34 44 54 64 74)=data4
+ punpckhqdq xmm0, xmm5 ; xmm0=(05 15 25 35 45 55 65 75)=data5
+
+ movdqa xmm2, xmm1
+ movdqa xmm5, xmm7
+ paddw xmm1, xmm4 ; xmm1=data3+data4=tmp3
+ paddw xmm7, xmm0 ; xmm7=data2+data5=tmp2
+ psubw xmm2, xmm4 ; xmm2=data3-data4=tmp4
+ psubw xmm5, xmm0 ; xmm5=data2-data5=tmp5
+
+ ; -- Even part
+
+ movdqa xmm4, xmm3
+ movdqa xmm0, xmm6
+ psubw xmm3, xmm1 ; xmm3=tmp13
+ psubw xmm6, xmm7 ; xmm6=tmp12
+ paddw xmm4, xmm1 ; xmm4=tmp10
+ paddw xmm0, xmm7 ; xmm0=tmp11
+
+ paddw xmm6, xmm3
+ psllw xmm6, PRE_MULTIPLY_SCALE_BITS
+ pmulhw xmm6, [GOTOFF(ebx,PW_F0707)] ; xmm6=z1
+
+ movdqa xmm1, xmm4
+ movdqa xmm7, xmm3
+ psubw xmm4, xmm0 ; xmm4=data4
+ psubw xmm3, xmm6 ; xmm3=data6
+ paddw xmm1, xmm0 ; xmm1=data0
+ paddw xmm7, xmm6 ; xmm7=data2
+
+ movdqa xmm0, XMMWORD [wk(0)] ; xmm0=tmp6
+ movdqa xmm6, XMMWORD [wk(1)] ; xmm6=tmp7
+ movdqa XMMWORD [wk(0)], xmm4 ; wk(0)=data4
+ movdqa XMMWORD [wk(1)], xmm3 ; wk(1)=data6
+
+ ; -- Odd part
+
+ paddw xmm2, xmm5 ; xmm2=tmp10
+ paddw xmm5, xmm0 ; xmm5=tmp11
+ paddw xmm0, xmm6 ; xmm0=tmp12, xmm6=tmp7
+
+ psllw xmm2, PRE_MULTIPLY_SCALE_BITS
+ psllw xmm0, PRE_MULTIPLY_SCALE_BITS
+
+ psllw xmm5, PRE_MULTIPLY_SCALE_BITS
+ pmulhw xmm5, [GOTOFF(ebx,PW_F0707)] ; xmm5=z3
+
+ movdqa xmm4, xmm2 ; xmm4=tmp10
+ psubw xmm2, xmm0
+ pmulhw xmm2, [GOTOFF(ebx,PW_F0382)] ; xmm2=z5
+ pmulhw xmm4, [GOTOFF(ebx,PW_F0541)] ; xmm4=MULTIPLY(tmp10,FIX_0_541196)
+ pmulhw xmm0, [GOTOFF(ebx,PW_F1306)] ; xmm0=MULTIPLY(tmp12,FIX_1_306562)
+ paddw xmm4, xmm2 ; xmm4=z2
+ paddw xmm0, xmm2 ; xmm0=z4
+
+ movdqa xmm3, xmm6
+ psubw xmm6, xmm5 ; xmm6=z13
+ paddw xmm3, xmm5 ; xmm3=z11
+
+ movdqa xmm2, xmm6
+ movdqa xmm5, xmm3
+ psubw xmm6, xmm4 ; xmm6=data3
+ psubw xmm3, xmm0 ; xmm3=data7
+ paddw xmm2, xmm4 ; xmm2=data5
+ paddw xmm5, xmm0 ; xmm5=data1
+
+ ; ---- Pass 2: process columns.
+
+; mov edx, POINTER [data(eax)] ; (DCTELEM *)
+
+ ; xmm1=(00 10 20 30 40 50 60 70), xmm7=(02 12 22 32 42 52 62 72)
+ ; xmm5=(01 11 21 31 41 51 61 71), xmm6=(03 13 23 33 43 53 63 73)
+
+ movdqa xmm4, xmm1 ; transpose coefficients(phase 1)
+ punpcklwd xmm1, xmm5 ; xmm1=(00 01 10 11 20 21 30 31)
+ punpckhwd xmm4, xmm5 ; xmm4=(40 41 50 51 60 61 70 71)
+ movdqa xmm0, xmm7 ; transpose coefficients(phase 1)
+ punpcklwd xmm7, xmm6 ; xmm7=(02 03 12 13 22 23 32 33)
+ punpckhwd xmm0, xmm6 ; xmm0=(42 43 52 53 62 63 72 73)
+
+ movdqa xmm5, XMMWORD [wk(0)] ; xmm5=col4
+ movdqa xmm6, XMMWORD [wk(1)] ; xmm6=col6
+
+ ; xmm5=(04 14 24 34 44 54 64 74), xmm6=(06 16 26 36 46 56 66 76)
+ ; xmm2=(05 15 25 35 45 55 65 75), xmm3=(07 17 27 37 47 57 67 77)
+
+ movdqa XMMWORD [wk(0)], xmm7 ; wk(0)=(02 03 12 13 22 23 32 33)
+ movdqa XMMWORD [wk(1)], xmm0 ; wk(1)=(42 43 52 53 62 63 72 73)
+
+ movdqa xmm7, xmm5 ; transpose coefficients(phase 1)
+ punpcklwd xmm5, xmm2 ; xmm5=(04 05 14 15 24 25 34 35)
+ punpckhwd xmm7, xmm2 ; xmm7=(44 45 54 55 64 65 74 75)
+ movdqa xmm0, xmm6 ; transpose coefficients(phase 1)
+ punpcklwd xmm6, xmm3 ; xmm6=(06 07 16 17 26 27 36 37)
+ punpckhwd xmm0, xmm3 ; xmm0=(46 47 56 57 66 67 76 77)
+
+ movdqa xmm2, xmm5 ; transpose coefficients(phase 2)
+ punpckldq xmm5, xmm6 ; xmm5=(04 05 06 07 14 15 16 17)
+ punpckhdq xmm2, xmm6 ; xmm2=(24 25 26 27 34 35 36 37)
+ movdqa xmm3, xmm7 ; transpose coefficients(phase 2)
+ punpckldq xmm7, xmm0 ; xmm7=(44 45 46 47 54 55 56 57)
+ punpckhdq xmm3, xmm0 ; xmm3=(64 65 66 67 74 75 76 77)
+
+ movdqa xmm6, XMMWORD [wk(0)] ; xmm6=(02 03 12 13 22 23 32 33)
+ movdqa xmm0, XMMWORD [wk(1)] ; xmm0=(42 43 52 53 62 63 72 73)
+ movdqa XMMWORD [wk(0)], xmm2 ; wk(0)=(24 25 26 27 34 35 36 37)
+ movdqa XMMWORD [wk(1)], xmm7 ; wk(1)=(44 45 46 47 54 55 56 57)
+
+ movdqa xmm2, xmm1 ; transpose coefficients(phase 2)
+ punpckldq xmm1, xmm6 ; xmm1=(00 01 02 03 10 11 12 13)
+ punpckhdq xmm2, xmm6 ; xmm2=(20 21 22 23 30 31 32 33)
+ movdqa xmm7, xmm4 ; transpose coefficients(phase 2)
+ punpckldq xmm4, xmm0 ; xmm4=(40 41 42 43 50 51 52 53)
+ punpckhdq xmm7, xmm0 ; xmm7=(60 61 62 63 70 71 72 73)
+
+ movdqa xmm6, xmm1 ; transpose coefficients(phase 3)
+ punpcklqdq xmm1, xmm5 ; xmm1=(00 01 02 03 04 05 06 07)=data0
+ punpckhqdq xmm6, xmm5 ; xmm6=(10 11 12 13 14 15 16 17)=data1
+ movdqa xmm0, xmm7 ; transpose coefficients(phase 3)
+ punpcklqdq xmm7, xmm3 ; xmm7=(60 61 62 63 64 65 66 67)=data6
+ punpckhqdq xmm0, xmm3 ; xmm0=(70 71 72 73 74 75 76 77)=data7
+
+ movdqa xmm5, xmm6
+ movdqa xmm3, xmm1
+ psubw xmm6, xmm7 ; xmm6=data1-data6=tmp6
+ psubw xmm1, xmm0 ; xmm1=data0-data7=tmp7
+ paddw xmm5, xmm7 ; xmm5=data1+data6=tmp1
+ paddw xmm3, xmm0 ; xmm3=data0+data7=tmp0
+
+ movdqa xmm7, XMMWORD [wk(0)] ; xmm7=(24 25 26 27 34 35 36 37)
+ movdqa xmm0, XMMWORD [wk(1)] ; xmm0=(44 45 46 47 54 55 56 57)
+ movdqa XMMWORD [wk(0)], xmm6 ; wk(0)=tmp6
+ movdqa XMMWORD [wk(1)], xmm1 ; wk(1)=tmp7
+
+ movdqa xmm6, xmm2 ; transpose coefficients(phase 3)
+ punpcklqdq xmm2, xmm7 ; xmm2=(20 21 22 23 24 25 26 27)=data2
+ punpckhqdq xmm6, xmm7 ; xmm6=(30 31 32 33 34 35 36 37)=data3
+ movdqa xmm1, xmm4 ; transpose coefficients(phase 3)
+ punpcklqdq xmm4, xmm0 ; xmm4=(40 41 42 43 44 45 46 47)=data4
+ punpckhqdq xmm1, xmm0 ; xmm1=(50 51 52 53 54 55 56 57)=data5
+
+ movdqa xmm7, xmm6
+ movdqa xmm0, xmm2
+ paddw xmm6, xmm4 ; xmm6=data3+data4=tmp3
+ paddw xmm2, xmm1 ; xmm2=data2+data5=tmp2
+ psubw xmm7, xmm4 ; xmm7=data3-data4=tmp4
+ psubw xmm0, xmm1 ; xmm0=data2-data5=tmp5
+
+ ; -- Even part
+
+ movdqa xmm4, xmm3
+ movdqa xmm1, xmm5
+ psubw xmm3, xmm6 ; xmm3=tmp13
+ psubw xmm5, xmm2 ; xmm5=tmp12
+ paddw xmm4, xmm6 ; xmm4=tmp10
+ paddw xmm1, xmm2 ; xmm1=tmp11
+
+ paddw xmm5, xmm3
+ psllw xmm5, PRE_MULTIPLY_SCALE_BITS
+ pmulhw xmm5, [GOTOFF(ebx,PW_F0707)] ; xmm5=z1
+
+ movdqa xmm6, xmm4
+ movdqa xmm2, xmm3
+ psubw xmm4, xmm1 ; xmm4=data4
+ psubw xmm3, xmm5 ; xmm3=data6
+ paddw xmm6, xmm1 ; xmm6=data0
+ paddw xmm2, xmm5 ; xmm2=data2
+
+ movdqa XMMWORD [XMMBLOCK(4,0,edx,SIZEOF_DCTELEM)], xmm4
+ movdqa XMMWORD [XMMBLOCK(6,0,edx,SIZEOF_DCTELEM)], xmm3
+ movdqa XMMWORD [XMMBLOCK(0,0,edx,SIZEOF_DCTELEM)], xmm6
+ movdqa XMMWORD [XMMBLOCK(2,0,edx,SIZEOF_DCTELEM)], xmm2
+
+ ; -- Odd part
+
+ movdqa xmm1, XMMWORD [wk(0)] ; xmm1=tmp6
+ movdqa xmm5, XMMWORD [wk(1)] ; xmm5=tmp7
+
+ paddw xmm7, xmm0 ; xmm7=tmp10
+ paddw xmm0, xmm1 ; xmm0=tmp11
+ paddw xmm1, xmm5 ; xmm1=tmp12, xmm5=tmp7
+
+ psllw xmm7, PRE_MULTIPLY_SCALE_BITS
+ psllw xmm1, PRE_MULTIPLY_SCALE_BITS
+
+ psllw xmm0, PRE_MULTIPLY_SCALE_BITS
+ pmulhw xmm0, [GOTOFF(ebx,PW_F0707)] ; xmm0=z3
+
+ movdqa xmm4, xmm7 ; xmm4=tmp10
+ psubw xmm7, xmm1
+ pmulhw xmm7, [GOTOFF(ebx,PW_F0382)] ; xmm7=z5
+ pmulhw xmm4, [GOTOFF(ebx,PW_F0541)] ; xmm4=MULTIPLY(tmp10,FIX_0_541196)
+ pmulhw xmm1, [GOTOFF(ebx,PW_F1306)] ; xmm1=MULTIPLY(tmp12,FIX_1_306562)
+ paddw xmm4, xmm7 ; xmm4=z2
+ paddw xmm1, xmm7 ; xmm1=z4
+
+ movdqa xmm3, xmm5
+ psubw xmm5, xmm0 ; xmm5=z13
+ paddw xmm3, xmm0 ; xmm3=z11
+
+ movdqa xmm6, xmm5
+ movdqa xmm2, xmm3
+ psubw xmm5, xmm4 ; xmm5=data3
+ psubw xmm3, xmm1 ; xmm3=data7
+ paddw xmm6, xmm4 ; xmm6=data5
+ paddw xmm2, xmm1 ; xmm2=data1
+
+ movdqa XMMWORD [XMMBLOCK(3,0,edx,SIZEOF_DCTELEM)], xmm5
+ movdqa XMMWORD [XMMBLOCK(7,0,edx,SIZEOF_DCTELEM)], xmm3
+ movdqa XMMWORD [XMMBLOCK(5,0,edx,SIZEOF_DCTELEM)], xmm6
+ movdqa XMMWORD [XMMBLOCK(1,0,edx,SIZEOF_DCTELEM)], xmm2
+
+; pop edi ; unused
+; pop esi ; unused
+; pop edx ; need not be preserved
+; pop ecx ; unused
+ poppic ebx
+ mov esp,ebp ; esp <- aligned ebp
+ pop esp ; esp <- original ebp
+ pop ebp
+ ret
+
+; For some reason, the OS X linker does not honor the request to align the
+; segment unless we do this.
+ align 32
diff --git a/simd/jfdctint-mmx.asm b/simd/i386/jfdctint-mmx.asm
similarity index 99%
rename from simd/jfdctint-mmx.asm
rename to simd/i386/jfdctint-mmx.asm
index 9142ad8..c035b0c 100644
--- a/simd/jfdctint-mmx.asm
+++ b/simd/i386/jfdctint-mmx.asm
@@ -2,6 +2,7 @@
; jfdctint.asm - accurate integer FDCT (MMX)
;
; Copyright 2009 Pierre Ossman <ossman@cendio.se> for Cendio AB
+; Copyright (C) 2016, D. R. Commander.
;
; Based on the x86 SIMD extension for IJG JPEG library
; Copyright (C) 1999-2006, MIYASAKA Masaru.
@@ -64,7 +65,7 @@
; --------------------------------------------------------------------------
SECTION SEG_CONST
- alignz 16
+ alignz 32
global EXTN(jconst_fdct_islow_mmx)
EXTN(jconst_fdct_islow_mmx):
@@ -81,7 +82,7 @@
PD_DESCALE_P2 times 2 dd 1 << (DESCALE_P2-1)
PW_DESCALE_P2X times 4 dw 1 << (PASS1_BITS-1)
- alignz 16
+ alignz 32
; --------------------------------------------------------------------------
SECTION SEG_TEXT
@@ -99,7 +100,7 @@
%define wk(i) ebp-(WK_NUM-(i))*SIZEOF_MMWORD ; mmword wk[WK_NUM]
%define WK_NUM 2
- align 16
+ align 32
global EXTN(jsimd_fdct_islow_mmx)
EXTN(jsimd_fdct_islow_mmx):
@@ -618,4 +619,4 @@
; For some reason, the OS X linker does not honor the request to align the
; segment unless we do this.
- align 16
+ align 32
diff --git a/simd/i386/jfdctint-sse2.asm b/simd/i386/jfdctint-sse2.asm
new file mode 100644
index 0000000..66f2cc6
--- /dev/null
+++ b/simd/i386/jfdctint-sse2.asm
@@ -0,0 +1,634 @@
+;
+; jfdctint.asm - accurate integer FDCT (SSE2)
+;
+; Copyright 2009 Pierre Ossman <ossman@cendio.se> for Cendio AB
+; Copyright (C) 2016, D. R. Commander.
+;
+; Based on the x86 SIMD extension for IJG JPEG library
+; Copyright (C) 1999-2006, MIYASAKA Masaru.
+; For conditions of distribution and use, see copyright notice in jsimdext.inc
+;
+; This file should be assembled with NASM (Netwide Assembler),
+; can *not* be assembled with Microsoft's MASM or any compatible
+; assembler (including Borland's Turbo Assembler).
+; NASM is available from http://nasm.sourceforge.net/ or
+; http://sourceforge.net/project/showfiles.php?group_id=6208
+;
+; This file contains a slow-but-accurate integer implementation of the
+; forward DCT (Discrete Cosine Transform). The following code is based
+; directly on the IJG's original jfdctint.c; see the jfdctint.c for
+; more details.
+;
+; [TAB8]
+
+%include "jsimdext.inc"
+%include "jdct.inc"
+
+; --------------------------------------------------------------------------
+
+%define CONST_BITS 13
+%define PASS1_BITS 2
+
+%define DESCALE_P1 (CONST_BITS-PASS1_BITS)
+%define DESCALE_P2 (CONST_BITS+PASS1_BITS)
+
+%if CONST_BITS == 13
+F_0_298 equ 2446 ; FIX(0.298631336)
+F_0_390 equ 3196 ; FIX(0.390180644)
+F_0_541 equ 4433 ; FIX(0.541196100)
+F_0_765 equ 6270 ; FIX(0.765366865)
+F_0_899 equ 7373 ; FIX(0.899976223)
+F_1_175 equ 9633 ; FIX(1.175875602)
+F_1_501 equ 12299 ; FIX(1.501321110)
+F_1_847 equ 15137 ; FIX(1.847759065)
+F_1_961 equ 16069 ; FIX(1.961570560)
+F_2_053 equ 16819 ; FIX(2.053119869)
+F_2_562 equ 20995 ; FIX(2.562915447)
+F_3_072 equ 25172 ; FIX(3.072711026)
+%else
+; NASM cannot do compile-time arithmetic on floating-point constants.
+%define DESCALE(x,n) (((x)+(1<<((n)-1)))>>(n))
+F_0_298 equ DESCALE( 320652955, 30-CONST_BITS) ; FIX(0.298631336)
+F_0_390 equ DESCALE( 418953276, 30-CONST_BITS) ; FIX(0.390180644)
+F_0_541 equ DESCALE( 581104887, 30-CONST_BITS) ; FIX(0.541196100)
+F_0_765 equ DESCALE( 821806413, 30-CONST_BITS) ; FIX(0.765366865)
+F_0_899 equ DESCALE( 966342111, 30-CONST_BITS) ; FIX(0.899976223)
+F_1_175 equ DESCALE(1262586813, 30-CONST_BITS) ; FIX(1.175875602)
+F_1_501 equ DESCALE(1612031267, 30-CONST_BITS) ; FIX(1.501321110)
+F_1_847 equ DESCALE(1984016188, 30-CONST_BITS) ; FIX(1.847759065)
+F_1_961 equ DESCALE(2106220350, 30-CONST_BITS) ; FIX(1.961570560)
+F_2_053 equ DESCALE(2204520673, 30-CONST_BITS) ; FIX(2.053119869)
+F_2_562 equ DESCALE(2751909506, 30-CONST_BITS) ; FIX(2.562915447)
+F_3_072 equ DESCALE(3299298341, 30-CONST_BITS) ; FIX(3.072711026)
+%endif
+
+; --------------------------------------------------------------------------
+ SECTION SEG_CONST
+
+ alignz 32
+ global EXTN(jconst_fdct_islow_sse2)
+
+EXTN(jconst_fdct_islow_sse2):
+
+PW_F130_F054 times 4 dw (F_0_541+F_0_765), F_0_541
+PW_F054_MF130 times 4 dw F_0_541, (F_0_541-F_1_847)
+PW_MF078_F117 times 4 dw (F_1_175-F_1_961), F_1_175
+PW_F117_F078 times 4 dw F_1_175, (F_1_175-F_0_390)
+PW_MF060_MF089 times 4 dw (F_0_298-F_0_899),-F_0_899
+PW_MF089_F060 times 4 dw -F_0_899, (F_1_501-F_0_899)
+PW_MF050_MF256 times 4 dw (F_2_053-F_2_562),-F_2_562
+PW_MF256_F050 times 4 dw -F_2_562, (F_3_072-F_2_562)
+PD_DESCALE_P1 times 4 dd 1 << (DESCALE_P1-1)
+PD_DESCALE_P2 times 4 dd 1 << (DESCALE_P2-1)
+PW_DESCALE_P2X times 8 dw 1 << (PASS1_BITS-1)
+
+ alignz 32
+
+; --------------------------------------------------------------------------
+ SECTION SEG_TEXT
+ BITS 32
+;
+; Perform the forward DCT on one block of samples.
+;
+; GLOBAL(void)
+; jsimd_fdct_islow_sse2 (DCTELEM *data)
+;
+
+%define data(b) (b)+8 ; DCTELEM *data
+
+%define original_ebp ebp+0
+%define wk(i) ebp-(WK_NUM-(i))*SIZEOF_XMMWORD ; xmmword wk[WK_NUM]
+%define WK_NUM 6
+
+ align 32
+ global EXTN(jsimd_fdct_islow_sse2)
+
+EXTN(jsimd_fdct_islow_sse2):
+ push ebp
+ mov eax, esp ; eax = original ebp
+ sub esp, byte 4
+ and esp, byte (-SIZEOF_XMMWORD) ; align to 128 bits
+ mov [esp], eax
+ mov ebp, esp ; ebp = aligned ebp
+ lea esp, [wk(0)]
+ pushpic ebx
+; push ecx ; unused
+; push edx ; need not be preserved
+; push esi ; unused
+; push edi ; unused
+
+ get_GOT ebx ; get GOT address
+
+ ; ---- Pass 1: process rows.
+
+ mov edx, POINTER [data(eax)] ; (DCTELEM *)
+
+ movdqa xmm0, XMMWORD [XMMBLOCK(0,0,edx,SIZEOF_DCTELEM)]
+ movdqa xmm1, XMMWORD [XMMBLOCK(1,0,edx,SIZEOF_DCTELEM)]
+ movdqa xmm2, XMMWORD [XMMBLOCK(2,0,edx,SIZEOF_DCTELEM)]
+ movdqa xmm3, XMMWORD [XMMBLOCK(3,0,edx,SIZEOF_DCTELEM)]
+
+ ; xmm0=(00 01 02 03 04 05 06 07), xmm2=(20 21 22 23 24 25 26 27)
+ ; xmm1=(10 11 12 13 14 15 16 17), xmm3=(30 31 32 33 34 35 36 37)
+
+ movdqa xmm4, xmm0 ; transpose coefficients(phase 1)
+ punpcklwd xmm0, xmm1 ; xmm0=(00 10 01 11 02 12 03 13)
+ punpckhwd xmm4, xmm1 ; xmm4=(04 14 05 15 06 16 07 17)
+ movdqa xmm5, xmm2 ; transpose coefficients(phase 1)
+ punpcklwd xmm2, xmm3 ; xmm2=(20 30 21 31 22 32 23 33)
+ punpckhwd xmm5, xmm3 ; xmm5=(24 34 25 35 26 36 27 37)
+
+ movdqa xmm6, XMMWORD [XMMBLOCK(4,0,edx,SIZEOF_DCTELEM)]
+ movdqa xmm7, XMMWORD [XMMBLOCK(5,0,edx,SIZEOF_DCTELEM)]
+ movdqa xmm1, XMMWORD [XMMBLOCK(6,0,edx,SIZEOF_DCTELEM)]
+ movdqa xmm3, XMMWORD [XMMBLOCK(7,0,edx,SIZEOF_DCTELEM)]
+
+ ; xmm6=( 4 12 20 28 36 44 52 60), xmm1=( 6 14 22 30 38 46 54 62)
+ ; xmm7=( 5 13 21 29 37 45 53 61), xmm3=( 7 15 23 31 39 47 55 63)
+
+ movdqa XMMWORD [wk(0)], xmm2 ; wk(0)=(20 30 21 31 22 32 23 33)
+ movdqa XMMWORD [wk(1)], xmm5 ; wk(1)=(24 34 25 35 26 36 27 37)
+
+ movdqa xmm2, xmm6 ; transpose coefficients(phase 1)
+ punpcklwd xmm6, xmm7 ; xmm6=(40 50 41 51 42 52 43 53)
+ punpckhwd xmm2, xmm7 ; xmm2=(44 54 45 55 46 56 47 57)
+ movdqa xmm5, xmm1 ; transpose coefficients(phase 1)
+ punpcklwd xmm1, xmm3 ; xmm1=(60 70 61 71 62 72 63 73)
+ punpckhwd xmm5, xmm3 ; xmm5=(64 74 65 75 66 76 67 77)
+
+ movdqa xmm7, xmm6 ; transpose coefficients(phase 2)
+ punpckldq xmm6, xmm1 ; xmm6=(40 50 60 70 41 51 61 71)
+ punpckhdq xmm7, xmm1 ; xmm7=(42 52 62 72 43 53 63 73)
+ movdqa xmm3, xmm2 ; transpose coefficients(phase 2)
+ punpckldq xmm2, xmm5 ; xmm2=(44 54 64 74 45 55 65 75)
+ punpckhdq xmm3, xmm5 ; xmm3=(46 56 66 76 47 57 67 77)
+
+ movdqa xmm1, XMMWORD [wk(0)] ; xmm1=(20 30 21 31 22 32 23 33)
+ movdqa xmm5, XMMWORD [wk(1)] ; xmm5=(24 34 25 35 26 36 27 37)
+ movdqa XMMWORD [wk(2)], xmm7 ; wk(2)=(42 52 62 72 43 53 63 73)
+ movdqa XMMWORD [wk(3)], xmm2 ; wk(3)=(44 54 64 74 45 55 65 75)
+
+ movdqa xmm7, xmm0 ; transpose coefficients(phase 2)
+ punpckldq xmm0, xmm1 ; xmm0=(00 10 20 30 01 11 21 31)
+ punpckhdq xmm7, xmm1 ; xmm7=(02 12 22 32 03 13 23 33)
+ movdqa xmm2, xmm4 ; transpose coefficients(phase 2)
+ punpckldq xmm4, xmm5 ; xmm4=(04 14 24 34 05 15 25 35)
+ punpckhdq xmm2, xmm5 ; xmm2=(06 16 26 36 07 17 27 37)
+
+ movdqa xmm1, xmm0 ; transpose coefficients(phase 3)
+ punpcklqdq xmm0, xmm6 ; xmm0=(00 10 20 30 40 50 60 70)=data0
+ punpckhqdq xmm1, xmm6 ; xmm1=(01 11 21 31 41 51 61 71)=data1
+ movdqa xmm5, xmm2 ; transpose coefficients(phase 3)
+ punpcklqdq xmm2, xmm3 ; xmm2=(06 16 26 36 46 56 66 76)=data6
+ punpckhqdq xmm5, xmm3 ; xmm5=(07 17 27 37 47 57 67 77)=data7
+
+ movdqa xmm6, xmm1
+ movdqa xmm3, xmm0
+ psubw xmm1, xmm2 ; xmm1=data1-data6=tmp6
+ psubw xmm0, xmm5 ; xmm0=data0-data7=tmp7
+ paddw xmm6, xmm2 ; xmm6=data1+data6=tmp1
+ paddw xmm3, xmm5 ; xmm3=data0+data7=tmp0
+
+ movdqa xmm2, XMMWORD [wk(2)] ; xmm2=(42 52 62 72 43 53 63 73)
+ movdqa xmm5, XMMWORD [wk(3)] ; xmm5=(44 54 64 74 45 55 65 75)
+ movdqa XMMWORD [wk(0)], xmm1 ; wk(0)=tmp6
+ movdqa XMMWORD [wk(1)], xmm0 ; wk(1)=tmp7
+
+ movdqa xmm1, xmm7 ; transpose coefficients(phase 3)
+ punpcklqdq xmm7, xmm2 ; xmm7=(02 12 22 32 42 52 62 72)=data2
+ punpckhqdq xmm1, xmm2 ; xmm1=(03 13 23 33 43 53 63 73)=data3
+ movdqa xmm0, xmm4 ; transpose coefficients(phase 3)
+ punpcklqdq xmm4, xmm5 ; xmm4=(04 14 24 34 44 54 64 74)=data4
+ punpckhqdq xmm0, xmm5 ; xmm0=(05 15 25 35 45 55 65 75)=data5
+
+ movdqa xmm2, xmm1
+ movdqa xmm5, xmm7
+ paddw xmm1, xmm4 ; xmm1=data3+data4=tmp3
+ paddw xmm7, xmm0 ; xmm7=data2+data5=tmp2
+ psubw xmm2, xmm4 ; xmm2=data3-data4=tmp4
+ psubw xmm5, xmm0 ; xmm5=data2-data5=tmp5
+
+ ; -- Even part
+
+ movdqa xmm4, xmm3
+ movdqa xmm0, xmm6
+ paddw xmm3, xmm1 ; xmm3=tmp10
+ paddw xmm6, xmm7 ; xmm6=tmp11
+ psubw xmm4, xmm1 ; xmm4=tmp13
+ psubw xmm0, xmm7 ; xmm0=tmp12
+
+ movdqa xmm1, xmm3
+ paddw xmm3, xmm6 ; xmm3=tmp10+tmp11
+ psubw xmm1, xmm6 ; xmm1=tmp10-tmp11
+
+ psllw xmm3, PASS1_BITS ; xmm3=data0
+ psllw xmm1, PASS1_BITS ; xmm1=data4
+
+ movdqa XMMWORD [wk(2)], xmm3 ; wk(2)=data0
+ movdqa XMMWORD [wk(3)], xmm1 ; wk(3)=data4
+
+ ; (Original)
+ ; z1 = (tmp12 + tmp13) * 0.541196100;
+ ; data2 = z1 + tmp13 * 0.765366865;
+ ; data6 = z1 + tmp12 * -1.847759065;
+ ;
+ ; (This implementation)
+ ; data2 = tmp13 * (0.541196100 + 0.765366865) + tmp12 * 0.541196100;
+ ; data6 = tmp13 * 0.541196100 + tmp12 * (0.541196100 - 1.847759065);
+
+ movdqa xmm7, xmm4 ; xmm4=tmp13
+ movdqa xmm6, xmm4
+ punpcklwd xmm7, xmm0 ; xmm0=tmp12
+ punpckhwd xmm6, xmm0
+ movdqa xmm4, xmm7
+ movdqa xmm0, xmm6
+ pmaddwd xmm7, [GOTOFF(ebx,PW_F130_F054)] ; xmm7=data2L
+ pmaddwd xmm6, [GOTOFF(ebx,PW_F130_F054)] ; xmm6=data2H
+ pmaddwd xmm4, [GOTOFF(ebx,PW_F054_MF130)] ; xmm4=data6L
+ pmaddwd xmm0, [GOTOFF(ebx,PW_F054_MF130)] ; xmm0=data6H
+
+ paddd xmm7, [GOTOFF(ebx,PD_DESCALE_P1)]
+ paddd xmm6, [GOTOFF(ebx,PD_DESCALE_P1)]
+ psrad xmm7, DESCALE_P1
+ psrad xmm6, DESCALE_P1
+ paddd xmm4, [GOTOFF(ebx,PD_DESCALE_P1)]
+ paddd xmm0, [GOTOFF(ebx,PD_DESCALE_P1)]
+ psrad xmm4, DESCALE_P1
+ psrad xmm0, DESCALE_P1
+
+ packssdw xmm7, xmm6 ; xmm7=data2
+ packssdw xmm4, xmm0 ; xmm4=data6
+
+ movdqa XMMWORD [wk(4)], xmm7 ; wk(4)=data2
+ movdqa XMMWORD [wk(5)], xmm4 ; wk(5)=data6
+
+ ; -- Odd part
+
+ movdqa xmm3, XMMWORD [wk(0)] ; xmm3=tmp6
+ movdqa xmm1, XMMWORD [wk(1)] ; xmm1=tmp7
+
+ movdqa xmm6, xmm2 ; xmm2=tmp4
+ movdqa xmm0, xmm5 ; xmm5=tmp5
+ paddw xmm6, xmm3 ; xmm6=z3
+ paddw xmm0, xmm1 ; xmm0=z4
+
+ ; (Original)
+ ; z5 = (z3 + z4) * 1.175875602;
+ ; z3 = z3 * -1.961570560; z4 = z4 * -0.390180644;
+ ; z3 += z5; z4 += z5;
+ ;
+ ; (This implementation)
+ ; z3 = z3 * (1.175875602 - 1.961570560) + z4 * 1.175875602;
+ ; z4 = z3 * 1.175875602 + z4 * (1.175875602 - 0.390180644);
+
+ movdqa xmm7, xmm6
+ movdqa xmm4, xmm6
+ punpcklwd xmm7, xmm0
+ punpckhwd xmm4, xmm0
+ movdqa xmm6, xmm7
+ movdqa xmm0, xmm4
+ pmaddwd xmm7, [GOTOFF(ebx,PW_MF078_F117)] ; xmm7=z3L
+ pmaddwd xmm4, [GOTOFF(ebx,PW_MF078_F117)] ; xmm4=z3H
+ pmaddwd xmm6, [GOTOFF(ebx,PW_F117_F078)] ; xmm6=z4L
+ pmaddwd xmm0, [GOTOFF(ebx,PW_F117_F078)] ; xmm0=z4H
+
+ movdqa XMMWORD [wk(0)], xmm7 ; wk(0)=z3L
+ movdqa XMMWORD [wk(1)], xmm4 ; wk(1)=z3H
+
+ ; (Original)
+ ; z1 = tmp4 + tmp7; z2 = tmp5 + tmp6;
+ ; tmp4 = tmp4 * 0.298631336; tmp5 = tmp5 * 2.053119869;
+ ; tmp6 = tmp6 * 3.072711026; tmp7 = tmp7 * 1.501321110;
+ ; z1 = z1 * -0.899976223; z2 = z2 * -2.562915447;
+ ; data7 = tmp4 + z1 + z3; data5 = tmp5 + z2 + z4;
+ ; data3 = tmp6 + z2 + z3; data1 = tmp7 + z1 + z4;
+ ;
+ ; (This implementation)
+ ; tmp4 = tmp4 * (0.298631336 - 0.899976223) + tmp7 * -0.899976223;
+ ; tmp5 = tmp5 * (2.053119869 - 2.562915447) + tmp6 * -2.562915447;
+ ; tmp6 = tmp5 * -2.562915447 + tmp6 * (3.072711026 - 2.562915447);
+ ; tmp7 = tmp4 * -0.899976223 + tmp7 * (1.501321110 - 0.899976223);
+ ; data7 = tmp4 + z3; data5 = tmp5 + z4;
+ ; data3 = tmp6 + z3; data1 = tmp7 + z4;
+
+ movdqa xmm7, xmm2
+ movdqa xmm4, xmm2
+ punpcklwd xmm7, xmm1
+ punpckhwd xmm4, xmm1
+ movdqa xmm2, xmm7
+ movdqa xmm1, xmm4
+ pmaddwd xmm7, [GOTOFF(ebx,PW_MF060_MF089)] ; xmm7=tmp4L
+ pmaddwd xmm4, [GOTOFF(ebx,PW_MF060_MF089)] ; xmm4=tmp4H
+ pmaddwd xmm2, [GOTOFF(ebx,PW_MF089_F060)] ; xmm2=tmp7L
+ pmaddwd xmm1, [GOTOFF(ebx,PW_MF089_F060)] ; xmm1=tmp7H
+
+ paddd xmm7, XMMWORD [wk(0)] ; xmm7=data7L
+ paddd xmm4, XMMWORD [wk(1)] ; xmm4=data7H
+ paddd xmm2, xmm6 ; xmm2=data1L
+ paddd xmm1, xmm0 ; xmm1=data1H
+
+ paddd xmm7, [GOTOFF(ebx,PD_DESCALE_P1)]
+ paddd xmm4, [GOTOFF(ebx,PD_DESCALE_P1)]
+ psrad xmm7, DESCALE_P1
+ psrad xmm4, DESCALE_P1
+ paddd xmm2, [GOTOFF(ebx,PD_DESCALE_P1)]
+ paddd xmm1, [GOTOFF(ebx,PD_DESCALE_P1)]
+ psrad xmm2, DESCALE_P1
+ psrad xmm1, DESCALE_P1
+
+ packssdw xmm7, xmm4 ; xmm7=data7
+ packssdw xmm2, xmm1 ; xmm2=data1
+
+ movdqa xmm4, xmm5
+ movdqa xmm1, xmm5
+ punpcklwd xmm4, xmm3
+ punpckhwd xmm1, xmm3
+ movdqa xmm5, xmm4
+ movdqa xmm3, xmm1
+ pmaddwd xmm4, [GOTOFF(ebx,PW_MF050_MF256)] ; xmm4=tmp5L
+ pmaddwd xmm1, [GOTOFF(ebx,PW_MF050_MF256)] ; xmm1=tmp5H
+ pmaddwd xmm5, [GOTOFF(ebx,PW_MF256_F050)] ; xmm5=tmp6L
+ pmaddwd xmm3, [GOTOFF(ebx,PW_MF256_F050)] ; xmm3=tmp6H
+
+ paddd xmm4, xmm6 ; xmm4=data5L
+ paddd xmm1, xmm0 ; xmm1=data5H
+ paddd xmm5, XMMWORD [wk(0)] ; xmm5=data3L
+ paddd xmm3, XMMWORD [wk(1)] ; xmm3=data3H
+
+ paddd xmm4, [GOTOFF(ebx,PD_DESCALE_P1)]
+ paddd xmm1, [GOTOFF(ebx,PD_DESCALE_P1)]
+ psrad xmm4, DESCALE_P1
+ psrad xmm1, DESCALE_P1
+ paddd xmm5, [GOTOFF(ebx,PD_DESCALE_P1)]
+ paddd xmm3, [GOTOFF(ebx,PD_DESCALE_P1)]
+ psrad xmm5, DESCALE_P1
+ psrad xmm3, DESCALE_P1
+
+ packssdw xmm4, xmm1 ; xmm4=data5
+ packssdw xmm5, xmm3 ; xmm5=data3
+
+ ; ---- Pass 2: process columns.
+
+; mov edx, POINTER [data(eax)] ; (DCTELEM *)
+
+ movdqa xmm6, XMMWORD [wk(2)] ; xmm6=col0
+ movdqa xmm0, XMMWORD [wk(4)] ; xmm0=col2
+
+ ; xmm6=(00 10 20 30 40 50 60 70), xmm0=(02 12 22 32 42 52 62 72)
+ ; xmm2=(01 11 21 31 41 51 61 71), xmm5=(03 13 23 33 43 53 63 73)
+
+ movdqa xmm1, xmm6 ; transpose coefficients(phase 1)
+ punpcklwd xmm6, xmm2 ; xmm6=(00 01 10 11 20 21 30 31)
+ punpckhwd xmm1, xmm2 ; xmm1=(40 41 50 51 60 61 70 71)
+ movdqa xmm3, xmm0 ; transpose coefficients(phase 1)
+ punpcklwd xmm0, xmm5 ; xmm0=(02 03 12 13 22 23 32 33)
+ punpckhwd xmm3, xmm5 ; xmm3=(42 43 52 53 62 63 72 73)
+
+ movdqa xmm2, XMMWORD [wk(3)] ; xmm2=col4
+ movdqa xmm5, XMMWORD [wk(5)] ; xmm5=col6
+
+ ; xmm2=(04 14 24 34 44 54 64 74), xmm5=(06 16 26 36 46 56 66 76)
+ ; xmm4=(05 15 25 35 45 55 65 75), xmm7=(07 17 27 37 47 57 67 77)
+
+ movdqa XMMWORD [wk(0)], xmm0 ; wk(0)=(02 03 12 13 22 23 32 33)
+ movdqa XMMWORD [wk(1)], xmm3 ; wk(1)=(42 43 52 53 62 63 72 73)
+
+ movdqa xmm0, xmm2 ; transpose coefficients(phase 1)
+ punpcklwd xmm2, xmm4 ; xmm2=(04 05 14 15 24 25 34 35)
+ punpckhwd xmm0, xmm4 ; xmm0=(44 45 54 55 64 65 74 75)
+ movdqa xmm3, xmm5 ; transpose coefficients(phase 1)
+ punpcklwd xmm5, xmm7 ; xmm5=(06 07 16 17 26 27 36 37)
+ punpckhwd xmm3, xmm7 ; xmm3=(46 47 56 57 66 67 76 77)
+
+ movdqa xmm4, xmm2 ; transpose coefficients(phase 2)
+ punpckldq xmm2, xmm5 ; xmm2=(04 05 06 07 14 15 16 17)
+ punpckhdq xmm4, xmm5 ; xmm4=(24 25 26 27 34 35 36 37)
+ movdqa xmm7, xmm0 ; transpose coefficients(phase 2)
+ punpckldq xmm0, xmm3 ; xmm0=(44 45 46 47 54 55 56 57)
+ punpckhdq xmm7, xmm3 ; xmm7=(64 65 66 67 74 75 76 77)
+
+ movdqa xmm5, XMMWORD [wk(0)] ; xmm5=(02 03 12 13 22 23 32 33)
+ movdqa xmm3, XMMWORD [wk(1)] ; xmm3=(42 43 52 53 62 63 72 73)
+ movdqa XMMWORD [wk(2)], xmm4 ; wk(2)=(24 25 26 27 34 35 36 37)
+ movdqa XMMWORD [wk(3)], xmm0 ; wk(3)=(44 45 46 47 54 55 56 57)
+
+ movdqa xmm4, xmm6 ; transpose coefficients(phase 2)
+ punpckldq xmm6, xmm5 ; xmm6=(00 01 02 03 10 11 12 13)
+ punpckhdq xmm4, xmm5 ; xmm4=(20 21 22 23 30 31 32 33)
+ movdqa xmm0, xmm1 ; transpose coefficients(phase 2)
+ punpckldq xmm1, xmm3 ; xmm1=(40 41 42 43 50 51 52 53)
+ punpckhdq xmm0, xmm3 ; xmm0=(60 61 62 63 70 71 72 73)
+
+ movdqa xmm5, xmm6 ; transpose coefficients(phase 3)
+ punpcklqdq xmm6, xmm2 ; xmm6=(00 01 02 03 04 05 06 07)=data0
+ punpckhqdq xmm5, xmm2 ; xmm5=(10 11 12 13 14 15 16 17)=data1
+ movdqa xmm3, xmm0 ; transpose coefficients(phase 3)
+ punpcklqdq xmm0, xmm7 ; xmm0=(60 61 62 63 64 65 66 67)=data6
+ punpckhqdq xmm3, xmm7 ; xmm3=(70 71 72 73 74 75 76 77)=data7
+
+ movdqa xmm2, xmm5
+ movdqa xmm7, xmm6
+ psubw xmm5, xmm0 ; xmm5=data1-data6=tmp6
+ psubw xmm6, xmm3 ; xmm6=data0-data7=tmp7
+ paddw xmm2, xmm0 ; xmm2=data1+data6=tmp1
+ paddw xmm7, xmm3 ; xmm7=data0+data7=tmp0
+
+ movdqa xmm0, XMMWORD [wk(2)] ; xmm0=(24 25 26 27 34 35 36 37)
+ movdqa xmm3, XMMWORD [wk(3)] ; xmm3=(44 45 46 47 54 55 56 57)
+ movdqa XMMWORD [wk(0)], xmm5 ; wk(0)=tmp6
+ movdqa XMMWORD [wk(1)], xmm6 ; wk(1)=tmp7
+
+ movdqa xmm5, xmm4 ; transpose coefficients(phase 3)
+ punpcklqdq xmm4, xmm0 ; xmm4=(20 21 22 23 24 25 26 27)=data2
+ punpckhqdq xmm5, xmm0 ; xmm5=(30 31 32 33 34 35 36 37)=data3
+ movdqa xmm6, xmm1 ; transpose coefficients(phase 3)
+ punpcklqdq xmm1, xmm3 ; xmm1=(40 41 42 43 44 45 46 47)=data4
+ punpckhqdq xmm6, xmm3 ; xmm6=(50 51 52 53 54 55 56 57)=data5
+
+ movdqa xmm0, xmm5
+ movdqa xmm3, xmm4
+ paddw xmm5, xmm1 ; xmm5=data3+data4=tmp3
+ paddw xmm4, xmm6 ; xmm4=data2+data5=tmp2
+ psubw xmm0, xmm1 ; xmm0=data3-data4=tmp4
+ psubw xmm3, xmm6 ; xmm3=data2-data5=tmp5
+
+ ; -- Even part
+
+ movdqa xmm1, xmm7
+ movdqa xmm6, xmm2
+ paddw xmm7, xmm5 ; xmm7=tmp10
+ paddw xmm2, xmm4 ; xmm2=tmp11
+ psubw xmm1, xmm5 ; xmm1=tmp13
+ psubw xmm6, xmm4 ; xmm6=tmp12
+
+ movdqa xmm5, xmm7
+ paddw xmm7, xmm2 ; xmm7=tmp10+tmp11
+ psubw xmm5, xmm2 ; xmm5=tmp10-tmp11
+
+ paddw xmm7, [GOTOFF(ebx,PW_DESCALE_P2X)]
+ paddw xmm5, [GOTOFF(ebx,PW_DESCALE_P2X)]
+ psraw xmm7, PASS1_BITS ; xmm7=data0
+ psraw xmm5, PASS1_BITS ; xmm5=data4
+
+ movdqa XMMWORD [XMMBLOCK(0,0,edx,SIZEOF_DCTELEM)], xmm7
+ movdqa XMMWORD [XMMBLOCK(4,0,edx,SIZEOF_DCTELEM)], xmm5
+
+ ; (Original)
+ ; z1 = (tmp12 + tmp13) * 0.541196100;
+ ; data2 = z1 + tmp13 * 0.765366865;
+ ; data6 = z1 + tmp12 * -1.847759065;
+ ;
+ ; (This implementation)
+ ; data2 = tmp13 * (0.541196100 + 0.765366865) + tmp12 * 0.541196100;
+ ; data6 = tmp13 * 0.541196100 + tmp12 * (0.541196100 - 1.847759065);
+
+ movdqa xmm4, xmm1 ; xmm1=tmp13
+ movdqa xmm2, xmm1
+ punpcklwd xmm4, xmm6 ; xmm6=tmp12
+ punpckhwd xmm2, xmm6
+ movdqa xmm1, xmm4
+ movdqa xmm6, xmm2
+ pmaddwd xmm4, [GOTOFF(ebx,PW_F130_F054)] ; xmm4=data2L
+ pmaddwd xmm2, [GOTOFF(ebx,PW_F130_F054)] ; xmm2=data2H
+ pmaddwd xmm1, [GOTOFF(ebx,PW_F054_MF130)] ; xmm1=data6L
+ pmaddwd xmm6, [GOTOFF(ebx,PW_F054_MF130)] ; xmm6=data6H
+
+ paddd xmm4, [GOTOFF(ebx,PD_DESCALE_P2)]
+ paddd xmm2, [GOTOFF(ebx,PD_DESCALE_P2)]
+ psrad xmm4, DESCALE_P2
+ psrad xmm2, DESCALE_P2
+ paddd xmm1, [GOTOFF(ebx,PD_DESCALE_P2)]
+ paddd xmm6, [GOTOFF(ebx,PD_DESCALE_P2)]
+ psrad xmm1, DESCALE_P2
+ psrad xmm6, DESCALE_P2
+
+ packssdw xmm4, xmm2 ; xmm4=data2
+ packssdw xmm1, xmm6 ; xmm1=data6
+
+ movdqa XMMWORD [XMMBLOCK(2,0,edx,SIZEOF_DCTELEM)], xmm4
+ movdqa XMMWORD [XMMBLOCK(6,0,edx,SIZEOF_DCTELEM)], xmm1
+
+ ; -- Odd part
+
+ movdqa xmm7, XMMWORD [wk(0)] ; xmm7=tmp6
+ movdqa xmm5, XMMWORD [wk(1)] ; xmm5=tmp7
+
+ movdqa xmm2, xmm0 ; xmm0=tmp4
+ movdqa xmm6, xmm3 ; xmm3=tmp5
+ paddw xmm2, xmm7 ; xmm2=z3
+ paddw xmm6, xmm5 ; xmm6=z4
+
+ ; (Original)
+ ; z5 = (z3 + z4) * 1.175875602;
+ ; z3 = z3 * -1.961570560; z4 = z4 * -0.390180644;
+ ; z3 += z5; z4 += z5;
+ ;
+ ; (This implementation)
+ ; z3 = z3 * (1.175875602 - 1.961570560) + z4 * 1.175875602;
+ ; z4 = z3 * 1.175875602 + z4 * (1.175875602 - 0.390180644);
+
+ movdqa xmm4, xmm2
+ movdqa xmm1, xmm2
+ punpcklwd xmm4, xmm6
+ punpckhwd xmm1, xmm6
+ movdqa xmm2, xmm4
+ movdqa xmm6, xmm1
+ pmaddwd xmm4, [GOTOFF(ebx,PW_MF078_F117)] ; xmm4=z3L
+ pmaddwd xmm1, [GOTOFF(ebx,PW_MF078_F117)] ; xmm1=z3H
+ pmaddwd xmm2, [GOTOFF(ebx,PW_F117_F078)] ; xmm2=z4L
+ pmaddwd xmm6, [GOTOFF(ebx,PW_F117_F078)] ; xmm6=z4H
+
+ movdqa XMMWORD [wk(0)], xmm4 ; wk(0)=z3L
+ movdqa XMMWORD [wk(1)], xmm1 ; wk(1)=z3H
+
+ ; (Original)
+ ; z1 = tmp4 + tmp7; z2 = tmp5 + tmp6;
+ ; tmp4 = tmp4 * 0.298631336; tmp5 = tmp5 * 2.053119869;
+ ; tmp6 = tmp6 * 3.072711026; tmp7 = tmp7 * 1.501321110;
+ ; z1 = z1 * -0.899976223; z2 = z2 * -2.562915447;
+ ; data7 = tmp4 + z1 + z3; data5 = tmp5 + z2 + z4;
+ ; data3 = tmp6 + z2 + z3; data1 = tmp7 + z1 + z4;
+ ;
+ ; (This implementation)
+ ; tmp4 = tmp4 * (0.298631336 - 0.899976223) + tmp7 * -0.899976223;
+ ; tmp5 = tmp5 * (2.053119869 - 2.562915447) + tmp6 * -2.562915447;
+ ; tmp6 = tmp5 * -2.562915447 + tmp6 * (3.072711026 - 2.562915447);
+ ; tmp7 = tmp4 * -0.899976223 + tmp7 * (1.501321110 - 0.899976223);
+ ; data7 = tmp4 + z3; data5 = tmp5 + z4;
+ ; data3 = tmp6 + z3; data1 = tmp7 + z4;
+
+ movdqa xmm4, xmm0
+ movdqa xmm1, xmm0
+ punpcklwd xmm4, xmm5
+ punpckhwd xmm1, xmm5
+ movdqa xmm0, xmm4
+ movdqa xmm5, xmm1
+ pmaddwd xmm4, [GOTOFF(ebx,PW_MF060_MF089)] ; xmm4=tmp4L
+ pmaddwd xmm1, [GOTOFF(ebx,PW_MF060_MF089)] ; xmm1=tmp4H
+ pmaddwd xmm0, [GOTOFF(ebx,PW_MF089_F060)] ; xmm0=tmp7L
+ pmaddwd xmm5, [GOTOFF(ebx,PW_MF089_F060)] ; xmm5=tmp7H
+
+ paddd xmm4, XMMWORD [wk(0)] ; xmm4=data7L
+ paddd xmm1, XMMWORD [wk(1)] ; xmm1=data7H
+ paddd xmm0, xmm2 ; xmm0=data1L
+ paddd xmm5, xmm6 ; xmm5=data1H
+
+ paddd xmm4, [GOTOFF(ebx,PD_DESCALE_P2)]
+ paddd xmm1, [GOTOFF(ebx,PD_DESCALE_P2)]
+ psrad xmm4, DESCALE_P2
+ psrad xmm1, DESCALE_P2
+ paddd xmm0, [GOTOFF(ebx,PD_DESCALE_P2)]
+ paddd xmm5, [GOTOFF(ebx,PD_DESCALE_P2)]
+ psrad xmm0, DESCALE_P2
+ psrad xmm5, DESCALE_P2
+
+ packssdw xmm4, xmm1 ; xmm4=data7
+ packssdw xmm0, xmm5 ; xmm0=data1
+
+ movdqa XMMWORD [XMMBLOCK(7,0,edx,SIZEOF_DCTELEM)], xmm4
+ movdqa XMMWORD [XMMBLOCK(1,0,edx,SIZEOF_DCTELEM)], xmm0
+
+ movdqa xmm1, xmm3
+ movdqa xmm5, xmm3
+ punpcklwd xmm1, xmm7
+ punpckhwd xmm5, xmm7
+ movdqa xmm3, xmm1
+ movdqa xmm7, xmm5
+ pmaddwd xmm1, [GOTOFF(ebx,PW_MF050_MF256)] ; xmm1=tmp5L
+ pmaddwd xmm5, [GOTOFF(ebx,PW_MF050_MF256)] ; xmm5=tmp5H
+ pmaddwd xmm3, [GOTOFF(ebx,PW_MF256_F050)] ; xmm3=tmp6L
+ pmaddwd xmm7, [GOTOFF(ebx,PW_MF256_F050)] ; xmm7=tmp6H
+
+ paddd xmm1, xmm2 ; xmm1=data5L
+ paddd xmm5, xmm6 ; xmm5=data5H
+ paddd xmm3, XMMWORD [wk(0)] ; xmm3=data3L
+ paddd xmm7, XMMWORD [wk(1)] ; xmm7=data3H
+
+ paddd xmm1, [GOTOFF(ebx,PD_DESCALE_P2)]
+ paddd xmm5, [GOTOFF(ebx,PD_DESCALE_P2)]
+ psrad xmm1, DESCALE_P2
+ psrad xmm5, DESCALE_P2
+ paddd xmm3, [GOTOFF(ebx,PD_DESCALE_P2)]
+ paddd xmm7, [GOTOFF(ebx,PD_DESCALE_P2)]
+ psrad xmm3, DESCALE_P2
+ psrad xmm7, DESCALE_P2
+
+ packssdw xmm1, xmm5 ; xmm1=data5
+ packssdw xmm3, xmm7 ; xmm3=data3
+
+ movdqa XMMWORD [XMMBLOCK(5,0,edx,SIZEOF_DCTELEM)], xmm1
+ movdqa XMMWORD [XMMBLOCK(3,0,edx,SIZEOF_DCTELEM)], xmm3
+
+; pop edi ; unused
+; pop esi ; unused
+; pop edx ; need not be preserved
+; pop ecx ; unused
+ poppic ebx
+ mov esp, ebp ; esp <- aligned ebp
+ pop esp ; esp <- original ebp
+ pop ebp
+ ret
+
+; For some reason, the OS X linker does not honor the request to align the
+; segment unless we do this.
+ align 32
diff --git a/simd/jidctflt-3dn.asm b/simd/i386/jidctflt-3dn.asm
similarity index 98%
rename from simd/jidctflt-3dn.asm
rename to simd/i386/jidctflt-3dn.asm
index 99356f2..1e47cf1 100644
--- a/simd/jidctflt-3dn.asm
+++ b/simd/i386/jidctflt-3dn.asm
@@ -2,6 +2,7 @@
; jidctflt.asm - floating-point IDCT (3DNow! & MMX)
;
; Copyright 2009 Pierre Ossman <ossman@cendio.se> for Cendio AB
+; Copyright (C) 2016, D. R. Commander.
;
; Based on the x86 SIMD extension for IJG JPEG library
; Copyright (C) 1999-2006, MIYASAKA Masaru.
@@ -25,7 +26,7 @@
; --------------------------------------------------------------------------
SECTION SEG_CONST
- alignz 16
+ alignz 32
global EXTN(jconst_idct_float_3dnow)
EXTN(jconst_idct_float_3dnow):
@@ -37,7 +38,7 @@
PD_RNDINT_MAGIC times 2 dd 100663296.0 ; (float)(0x00C00000 << 3)
PB_CENTERJSAMP times 8 db CENTERJSAMPLE
- alignz 16
+ alignz 32
; --------------------------------------------------------------------------
SECTION SEG_TEXT
@@ -61,7 +62,7 @@
%define workspace wk(0)-DCTSIZE2*SIZEOF_FAST_FLOAT
; FAST_FLOAT workspace[DCTSIZE2]
- align 16
+ align 32
global EXTN(jsimd_idct_float_3dnow)
EXTN(jsimd_idct_float_3dnow):
@@ -448,4 +449,4 @@
; For some reason, the OS X linker does not honor the request to align the
; segment unless we do this.
- align 16
+ align 32
diff --git a/simd/jidctflt-sse.asm b/simd/i386/jidctflt-sse.asm
similarity index 99%
rename from simd/jidctflt-sse.asm
rename to simd/i386/jidctflt-sse.asm
index 4d4af2f..a1c99e9 100644
--- a/simd/jidctflt-sse.asm
+++ b/simd/i386/jidctflt-sse.asm
@@ -2,6 +2,7 @@
; jidctflt.asm - floating-point IDCT (SSE & MMX)
;
; Copyright 2009 Pierre Ossman <ossman@cendio.se> for Cendio AB
+; Copyright (C) 2016, D. R. Commander.
;
; Based on the x86 SIMD extension for IJG JPEG library
; Copyright (C) 1999-2006, MIYASAKA Masaru.
@@ -35,7 +36,7 @@
; --------------------------------------------------------------------------
SECTION SEG_CONST
- alignz 16
+ alignz 32
global EXTN(jconst_idct_float_sse)
EXTN(jconst_idct_float_sse):
@@ -47,7 +48,7 @@
PD_0_125 times 4 dd 0.125 ; 1/8
PB_CENTERJSAMP times 8 db CENTERJSAMPLE
- alignz 16
+ alignz 32
; --------------------------------------------------------------------------
SECTION SEG_TEXT
@@ -71,7 +72,7 @@
%define workspace wk(0)-DCTSIZE2*SIZEOF_FAST_FLOAT
; FAST_FLOAT workspace[DCTSIZE2]
- align 16
+ align 32
global EXTN(jsimd_idct_float_sse)
EXTN(jsimd_idct_float_sse):
@@ -568,4 +569,4 @@
; For some reason, the OS X linker does not honor the request to align the
; segment unless we do this.
- align 16
+ align 32
diff --git a/simd/i386/jidctflt-sse2.asm b/simd/i386/jidctflt-sse2.asm
new file mode 100644
index 0000000..e8be844
--- /dev/null
+++ b/simd/i386/jidctflt-sse2.asm
@@ -0,0 +1,498 @@
+;
+; jidctflt.asm - floating-point IDCT (SSE & SSE2)
+;
+; Copyright 2009 Pierre Ossman <ossman@cendio.se> for Cendio AB
+; Copyright (C) 2016, D. R. Commander.
+;
+; Based on the x86 SIMD extension for IJG JPEG library
+; Copyright (C) 1999-2006, MIYASAKA Masaru.
+; For conditions of distribution and use, see copyright notice in jsimdext.inc
+;
+; This file should be assembled with NASM (Netwide Assembler),
+; can *not* be assembled with Microsoft's MASM or any compatible
+; assembler (including Borland's Turbo Assembler).
+; NASM is available from http://nasm.sourceforge.net/ or
+; http://sourceforge.net/project/showfiles.php?group_id=6208
+;
+; This file contains a floating-point implementation of the inverse DCT
+; (Discrete Cosine Transform). The following code is based directly on
+; the IJG's original jidctflt.c; see the jidctflt.c for more details.
+;
+; [TAB8]
+
+%include "jsimdext.inc"
+%include "jdct.inc"
+
+; --------------------------------------------------------------------------
+
+%macro unpcklps2 2 ; %1=(0 1 2 3) / %2=(4 5 6 7) => %1=(0 1 4 5)
+ shufps %1, %2, 0x44
+%endmacro
+
+%macro unpckhps2 2 ; %1=(0 1 2 3) / %2=(4 5 6 7) => %1=(2 3 6 7)
+ shufps %1, %2, 0xEE
+%endmacro
+
+; --------------------------------------------------------------------------
+ SECTION SEG_CONST
+
+ alignz 32
+ global EXTN(jconst_idct_float_sse2)
+
+EXTN(jconst_idct_float_sse2):
+
+PD_1_414 times 4 dd 1.414213562373095048801689
+PD_1_847 times 4 dd 1.847759065022573512256366
+PD_1_082 times 4 dd 1.082392200292393968799446
+PD_M2_613 times 4 dd -2.613125929752753055713286
+PD_RNDINT_MAGIC times 4 dd 100663296.0 ; (float)(0x00C00000 << 3)
+PB_CENTERJSAMP times 16 db CENTERJSAMPLE
+
+ alignz 32
+
+; --------------------------------------------------------------------------
+ SECTION SEG_TEXT
+ BITS 32
+;
+; Perform dequantization and inverse DCT on one block of coefficients.
+;
+; GLOBAL(void)
+; jsimd_idct_float_sse2 (void *dct_table, JCOEFPTR coef_block,
+; JSAMPARRAY output_buf, JDIMENSION output_col)
+;
+
+%define dct_table(b) (b)+8 ; void *dct_table
+%define coef_block(b) (b)+12 ; JCOEFPTR coef_block
+%define output_buf(b) (b)+16 ; JSAMPARRAY output_buf
+%define output_col(b) (b)+20 ; JDIMENSION output_col
+
+%define original_ebp ebp+0
+%define wk(i) ebp-(WK_NUM-(i))*SIZEOF_XMMWORD ; xmmword wk[WK_NUM]
+%define WK_NUM 2
+%define workspace wk(0)-DCTSIZE2*SIZEOF_FAST_FLOAT
+ ; FAST_FLOAT workspace[DCTSIZE2]
+
+ align 32
+ global EXTN(jsimd_idct_float_sse2)
+
+EXTN(jsimd_idct_float_sse2):
+ push ebp
+ mov eax, esp ; eax = original ebp
+ sub esp, byte 4
+ and esp, byte (-SIZEOF_XMMWORD) ; align to 128 bits
+ mov [esp], eax
+ mov ebp, esp ; ebp = aligned ebp
+ lea esp, [workspace]
+ push ebx
+; push ecx ; need not be preserved
+; push edx ; need not be preserved
+ push esi
+ push edi
+
+ get_GOT ebx ; get GOT address
+
+ ; ---- Pass 1: process columns from input, store into work array.
+
+; mov eax, [original_ebp]
+ mov edx, POINTER [dct_table(eax)] ; quantptr
+ mov esi, JCOEFPTR [coef_block(eax)] ; inptr
+ lea edi, [workspace] ; FAST_FLOAT *wsptr
+ mov ecx, DCTSIZE/4 ; ctr
+ alignx 16, 7
+.columnloop:
+%ifndef NO_ZERO_COLUMN_TEST_FLOAT_SSE
+ mov eax, DWORD [DWBLOCK(1,0,esi,SIZEOF_JCOEF)]
+ or eax, DWORD [DWBLOCK(2,0,esi,SIZEOF_JCOEF)]
+ jnz near .columnDCT
+
+ movq xmm1, XMM_MMWORD [MMBLOCK(1,0,esi,SIZEOF_JCOEF)]
+ movq xmm2, XMM_MMWORD [MMBLOCK(2,0,esi,SIZEOF_JCOEF)]
+ movq xmm3, XMM_MMWORD [MMBLOCK(3,0,esi,SIZEOF_JCOEF)]
+ movq xmm4, XMM_MMWORD [MMBLOCK(4,0,esi,SIZEOF_JCOEF)]
+ movq xmm5, XMM_MMWORD [MMBLOCK(5,0,esi,SIZEOF_JCOEF)]
+ movq xmm6, XMM_MMWORD [MMBLOCK(6,0,esi,SIZEOF_JCOEF)]
+ movq xmm7, XMM_MMWORD [MMBLOCK(7,0,esi,SIZEOF_JCOEF)]
+ por xmm1, xmm2
+ por xmm3, xmm4
+ por xmm5, xmm6
+ por xmm1, xmm3
+ por xmm5, xmm7
+ por xmm1, xmm5
+ packsswb xmm1, xmm1
+ movd eax, xmm1
+ test eax, eax
+ jnz short .columnDCT
+
+ ; -- AC terms all zero
+
+ movq xmm0, XMM_MMWORD [MMBLOCK(0,0,esi,SIZEOF_JCOEF)]
+
+ punpcklwd xmm0, xmm0 ; xmm0=(00 00 01 01 02 02 03 03)
+ psrad xmm0, (DWORD_BIT-WORD_BIT) ; xmm0=in0=(00 01 02 03)
+ cvtdq2ps xmm0, xmm0 ; xmm0=in0=(00 01 02 03)
+
+ mulps xmm0, XMMWORD [XMMBLOCK(0,0,edx,SIZEOF_FLOAT_MULT_TYPE)]
+
+ movaps xmm1, xmm0
+ movaps xmm2, xmm0
+ movaps xmm3, xmm0
+
+ shufps xmm0, xmm0, 0x00 ; xmm0=(00 00 00 00)
+ shufps xmm1, xmm1, 0x55 ; xmm1=(01 01 01 01)
+ shufps xmm2, xmm2, 0xAA ; xmm2=(02 02 02 02)
+ shufps xmm3, xmm3, 0xFF ; xmm3=(03 03 03 03)
+
+ movaps XMMWORD [XMMBLOCK(0,0,edi,SIZEOF_FAST_FLOAT)], xmm0
+ movaps XMMWORD [XMMBLOCK(0,1,edi,SIZEOF_FAST_FLOAT)], xmm0
+ movaps XMMWORD [XMMBLOCK(1,0,edi,SIZEOF_FAST_FLOAT)], xmm1
+ movaps XMMWORD [XMMBLOCK(1,1,edi,SIZEOF_FAST_FLOAT)], xmm1
+ movaps XMMWORD [XMMBLOCK(2,0,edi,SIZEOF_FAST_FLOAT)], xmm2
+ movaps XMMWORD [XMMBLOCK(2,1,edi,SIZEOF_FAST_FLOAT)], xmm2
+ movaps XMMWORD [XMMBLOCK(3,0,edi,SIZEOF_FAST_FLOAT)], xmm3
+ movaps XMMWORD [XMMBLOCK(3,1,edi,SIZEOF_FAST_FLOAT)], xmm3
+ jmp near .nextcolumn
+ alignx 16, 7
+%endif
+.columnDCT:
+
+ ; -- Even part
+
+ movq xmm0, XMM_MMWORD [MMBLOCK(0,0,esi,SIZEOF_JCOEF)]
+ movq xmm1, XMM_MMWORD [MMBLOCK(2,0,esi,SIZEOF_JCOEF)]
+ movq xmm2, XMM_MMWORD [MMBLOCK(4,0,esi,SIZEOF_JCOEF)]
+ movq xmm3, XMM_MMWORD [MMBLOCK(6,0,esi,SIZEOF_JCOEF)]
+
+ punpcklwd xmm0, xmm0 ; xmm0=(00 00 01 01 02 02 03 03)
+ punpcklwd xmm1, xmm1 ; xmm1=(20 20 21 21 22 22 23 23)
+ psrad xmm0, (DWORD_BIT-WORD_BIT) ; xmm0=in0=(00 01 02 03)
+ psrad xmm1, (DWORD_BIT-WORD_BIT) ; xmm1=in2=(20 21 22 23)
+ cvtdq2ps xmm0, xmm0 ; xmm0=in0=(00 01 02 03)
+ cvtdq2ps xmm1, xmm1 ; xmm1=in2=(20 21 22 23)
+
+ punpcklwd xmm2, xmm2 ; xmm2=(40 40 41 41 42 42 43 43)
+ punpcklwd xmm3, xmm3 ; xmm3=(60 60 61 61 62 62 63 63)
+ psrad xmm2, (DWORD_BIT-WORD_BIT) ; xmm2=in4=(40 41 42 43)
+ psrad xmm3, (DWORD_BIT-WORD_BIT) ; xmm3=in6=(60 61 62 63)
+ cvtdq2ps xmm2, xmm2 ; xmm2=in4=(40 41 42 43)
+ cvtdq2ps xmm3, xmm3 ; xmm3=in6=(60 61 62 63)
+
+ mulps xmm0, XMMWORD [XMMBLOCK(0,0,edx,SIZEOF_FLOAT_MULT_TYPE)]
+ mulps xmm1, XMMWORD [XMMBLOCK(2,0,edx,SIZEOF_FLOAT_MULT_TYPE)]
+ mulps xmm2, XMMWORD [XMMBLOCK(4,0,edx,SIZEOF_FLOAT_MULT_TYPE)]
+ mulps xmm3, XMMWORD [XMMBLOCK(6,0,edx,SIZEOF_FLOAT_MULT_TYPE)]
+
+ movaps xmm4, xmm0
+ movaps xmm5, xmm1
+ subps xmm0, xmm2 ; xmm0=tmp11
+ subps xmm1, xmm3
+ addps xmm4, xmm2 ; xmm4=tmp10
+ addps xmm5, xmm3 ; xmm5=tmp13
+
+ mulps xmm1, [GOTOFF(ebx,PD_1_414)]
+ subps xmm1, xmm5 ; xmm1=tmp12
+
+ movaps xmm6, xmm4
+ movaps xmm7, xmm0
+ subps xmm4, xmm5 ; xmm4=tmp3
+ subps xmm0, xmm1 ; xmm0=tmp2
+ addps xmm6, xmm5 ; xmm6=tmp0
+ addps xmm7, xmm1 ; xmm7=tmp1
+
+ movaps XMMWORD [wk(1)], xmm4 ; tmp3
+ movaps XMMWORD [wk(0)], xmm0 ; tmp2
+
+ ; -- Odd part
+
+ movq xmm2, XMM_MMWORD [MMBLOCK(1,0,esi,SIZEOF_JCOEF)]
+ movq xmm3, XMM_MMWORD [MMBLOCK(3,0,esi,SIZEOF_JCOEF)]
+ movq xmm5, XMM_MMWORD [MMBLOCK(5,0,esi,SIZEOF_JCOEF)]
+ movq xmm1, XMM_MMWORD [MMBLOCK(7,0,esi,SIZEOF_JCOEF)]
+
+ punpcklwd xmm2, xmm2 ; xmm2=(10 10 11 11 12 12 13 13)
+ punpcklwd xmm3, xmm3 ; xmm3=(30 30 31 31 32 32 33 33)
+ psrad xmm2, (DWORD_BIT-WORD_BIT) ; xmm2=in1=(10 11 12 13)
+ psrad xmm3, (DWORD_BIT-WORD_BIT) ; xmm3=in3=(30 31 32 33)
+ cvtdq2ps xmm2, xmm2 ; xmm2=in1=(10 11 12 13)
+ cvtdq2ps xmm3, xmm3 ; xmm3=in3=(30 31 32 33)
+
+ punpcklwd xmm5, xmm5 ; xmm5=(50 50 51 51 52 52 53 53)
+ punpcklwd xmm1, xmm1 ; xmm1=(70 70 71 71 72 72 73 73)
+ psrad xmm5, (DWORD_BIT-WORD_BIT) ; xmm5=in5=(50 51 52 53)
+ psrad xmm1, (DWORD_BIT-WORD_BIT) ; xmm1=in7=(70 71 72 73)
+ cvtdq2ps xmm5, xmm5 ; xmm5=in5=(50 51 52 53)
+ cvtdq2ps xmm1, xmm1 ; xmm1=in7=(70 71 72 73)
+
+ mulps xmm2, XMMWORD [XMMBLOCK(1,0,edx,SIZEOF_FLOAT_MULT_TYPE)]
+ mulps xmm3, XMMWORD [XMMBLOCK(3,0,edx,SIZEOF_FLOAT_MULT_TYPE)]
+ mulps xmm5, XMMWORD [XMMBLOCK(5,0,edx,SIZEOF_FLOAT_MULT_TYPE)]
+ mulps xmm1, XMMWORD [XMMBLOCK(7,0,edx,SIZEOF_FLOAT_MULT_TYPE)]
+
+ movaps xmm4, xmm2
+ movaps xmm0, xmm5
+ addps xmm2, xmm1 ; xmm2=z11
+ addps xmm5, xmm3 ; xmm5=z13
+ subps xmm4, xmm1 ; xmm4=z12
+ subps xmm0, xmm3 ; xmm0=z10
+
+ movaps xmm1, xmm2
+ subps xmm2, xmm5
+ addps xmm1, xmm5 ; xmm1=tmp7
+
+ mulps xmm2, [GOTOFF(ebx,PD_1_414)] ; xmm2=tmp11
+
+ movaps xmm3, xmm0
+ addps xmm0, xmm4
+ mulps xmm0, [GOTOFF(ebx,PD_1_847)] ; xmm0=z5
+ mulps xmm3, [GOTOFF(ebx,PD_M2_613)] ; xmm3=(z10 * -2.613125930)
+ mulps xmm4, [GOTOFF(ebx,PD_1_082)] ; xmm4=(z12 * 1.082392200)
+ addps xmm3, xmm0 ; xmm3=tmp12
+ subps xmm4, xmm0 ; xmm4=tmp10
+
+ ; -- Final output stage
+
+ subps xmm3, xmm1 ; xmm3=tmp6
+ movaps xmm5, xmm6
+ movaps xmm0, xmm7
+ addps xmm6, xmm1 ; xmm6=data0=(00 01 02 03)
+ addps xmm7, xmm3 ; xmm7=data1=(10 11 12 13)
+ subps xmm5, xmm1 ; xmm5=data7=(70 71 72 73)
+ subps xmm0, xmm3 ; xmm0=data6=(60 61 62 63)
+ subps xmm2, xmm3 ; xmm2=tmp5
+
+ movaps xmm1, xmm6 ; transpose coefficients(phase 1)
+ unpcklps xmm6, xmm7 ; xmm6=(00 10 01 11)
+ unpckhps xmm1, xmm7 ; xmm1=(02 12 03 13)
+ movaps xmm3, xmm0 ; transpose coefficients(phase 1)
+ unpcklps xmm0, xmm5 ; xmm0=(60 70 61 71)
+ unpckhps xmm3, xmm5 ; xmm3=(62 72 63 73)
+
+ movaps xmm7, XMMWORD [wk(0)] ; xmm7=tmp2
+ movaps xmm5, XMMWORD [wk(1)] ; xmm5=tmp3
+
+ movaps XMMWORD [wk(0)], xmm0 ; wk(0)=(60 70 61 71)
+ movaps XMMWORD [wk(1)], xmm3 ; wk(1)=(62 72 63 73)
+
+ addps xmm4, xmm2 ; xmm4=tmp4
+ movaps xmm0, xmm7
+ movaps xmm3, xmm5
+ addps xmm7, xmm2 ; xmm7=data2=(20 21 22 23)
+ addps xmm5, xmm4 ; xmm5=data4=(40 41 42 43)
+ subps xmm0, xmm2 ; xmm0=data5=(50 51 52 53)
+ subps xmm3, xmm4 ; xmm3=data3=(30 31 32 33)
+
+ movaps xmm2, xmm7 ; transpose coefficients(phase 1)
+ unpcklps xmm7, xmm3 ; xmm7=(20 30 21 31)
+ unpckhps xmm2, xmm3 ; xmm2=(22 32 23 33)
+ movaps xmm4, xmm5 ; transpose coefficients(phase 1)
+ unpcklps xmm5, xmm0 ; xmm5=(40 50 41 51)
+ unpckhps xmm4, xmm0 ; xmm4=(42 52 43 53)
+
+ movaps xmm3, xmm6 ; transpose coefficients(phase 2)
+ unpcklps2 xmm6, xmm7 ; xmm6=(00 10 20 30)
+ unpckhps2 xmm3, xmm7 ; xmm3=(01 11 21 31)
+ movaps xmm0, xmm1 ; transpose coefficients(phase 2)
+ unpcklps2 xmm1, xmm2 ; xmm1=(02 12 22 32)
+ unpckhps2 xmm0, xmm2 ; xmm0=(03 13 23 33)
+
+ movaps xmm7, XMMWORD [wk(0)] ; xmm7=(60 70 61 71)
+ movaps xmm2, XMMWORD [wk(1)] ; xmm2=(62 72 63 73)
+
+ movaps XMMWORD [XMMBLOCK(0,0,edi,SIZEOF_FAST_FLOAT)], xmm6
+ movaps XMMWORD [XMMBLOCK(1,0,edi,SIZEOF_FAST_FLOAT)], xmm3
+ movaps XMMWORD [XMMBLOCK(2,0,edi,SIZEOF_FAST_FLOAT)], xmm1
+ movaps XMMWORD [XMMBLOCK(3,0,edi,SIZEOF_FAST_FLOAT)], xmm0
+
+ movaps xmm6, xmm5 ; transpose coefficients(phase 2)
+ unpcklps2 xmm5, xmm7 ; xmm5=(40 50 60 70)
+ unpckhps2 xmm6, xmm7 ; xmm6=(41 51 61 71)
+ movaps xmm3, xmm4 ; transpose coefficients(phase 2)
+ unpcklps2 xmm4, xmm2 ; xmm4=(42 52 62 72)
+ unpckhps2 xmm3, xmm2 ; xmm3=(43 53 63 73)
+
+ movaps XMMWORD [XMMBLOCK(0,1,edi,SIZEOF_FAST_FLOAT)], xmm5
+ movaps XMMWORD [XMMBLOCK(1,1,edi,SIZEOF_FAST_FLOAT)], xmm6
+ movaps XMMWORD [XMMBLOCK(2,1,edi,SIZEOF_FAST_FLOAT)], xmm4
+ movaps XMMWORD [XMMBLOCK(3,1,edi,SIZEOF_FAST_FLOAT)], xmm3
+
+.nextcolumn:
+ add esi, byte 4*SIZEOF_JCOEF ; coef_block
+ add edx, byte 4*SIZEOF_FLOAT_MULT_TYPE ; quantptr
+ add edi, 4*DCTSIZE*SIZEOF_FAST_FLOAT ; wsptr
+ dec ecx ; ctr
+ jnz near .columnloop
+
+ ; -- Prefetch the next coefficient block
+
+ prefetchnta [esi + (DCTSIZE2-8)*SIZEOF_JCOEF + 0*32]
+ prefetchnta [esi + (DCTSIZE2-8)*SIZEOF_JCOEF + 1*32]
+ prefetchnta [esi + (DCTSIZE2-8)*SIZEOF_JCOEF + 2*32]
+ prefetchnta [esi + (DCTSIZE2-8)*SIZEOF_JCOEF + 3*32]
+
+ ; ---- Pass 2: process rows from work array, store into output array.
+
+ mov eax, [original_ebp]
+ lea esi, [workspace] ; FAST_FLOAT *wsptr
+ mov edi, JSAMPARRAY [output_buf(eax)] ; (JSAMPROW *)
+ mov eax, JDIMENSION [output_col(eax)]
+ mov ecx, DCTSIZE/4 ; ctr
+ alignx 16, 7
+.rowloop:
+
+ ; -- Even part
+
+ movaps xmm0, XMMWORD [XMMBLOCK(0,0,esi,SIZEOF_FAST_FLOAT)]
+ movaps xmm1, XMMWORD [XMMBLOCK(2,0,esi,SIZEOF_FAST_FLOAT)]
+ movaps xmm2, XMMWORD [XMMBLOCK(4,0,esi,SIZEOF_FAST_FLOAT)]
+ movaps xmm3, XMMWORD [XMMBLOCK(6,0,esi,SIZEOF_FAST_FLOAT)]
+
+ movaps xmm4, xmm0
+ movaps xmm5, xmm1
+ subps xmm0, xmm2 ; xmm0=tmp11
+ subps xmm1, xmm3
+ addps xmm4, xmm2 ; xmm4=tmp10
+ addps xmm5, xmm3 ; xmm5=tmp13
+
+ mulps xmm1, [GOTOFF(ebx,PD_1_414)]
+ subps xmm1, xmm5 ; xmm1=tmp12
+
+ movaps xmm6, xmm4
+ movaps xmm7, xmm0
+ subps xmm4, xmm5 ; xmm4=tmp3
+ subps xmm0, xmm1 ; xmm0=tmp2
+ addps xmm6, xmm5 ; xmm6=tmp0
+ addps xmm7, xmm1 ; xmm7=tmp1
+
+ movaps XMMWORD [wk(1)], xmm4 ; tmp3
+ movaps XMMWORD [wk(0)], xmm0 ; tmp2
+
+ ; -- Odd part
+
+ movaps xmm2, XMMWORD [XMMBLOCK(1,0,esi,SIZEOF_FAST_FLOAT)]
+ movaps xmm3, XMMWORD [XMMBLOCK(3,0,esi,SIZEOF_FAST_FLOAT)]
+ movaps xmm5, XMMWORD [XMMBLOCK(5,0,esi,SIZEOF_FAST_FLOAT)]
+ movaps xmm1, XMMWORD [XMMBLOCK(7,0,esi,SIZEOF_FAST_FLOAT)]
+
+ movaps xmm4, xmm2
+ movaps xmm0, xmm5
+ addps xmm2, xmm1 ; xmm2=z11
+ addps xmm5, xmm3 ; xmm5=z13
+ subps xmm4, xmm1 ; xmm4=z12
+ subps xmm0, xmm3 ; xmm0=z10
+
+ movaps xmm1, xmm2
+ subps xmm2, xmm5
+ addps xmm1, xmm5 ; xmm1=tmp7
+
+ mulps xmm2, [GOTOFF(ebx,PD_1_414)] ; xmm2=tmp11
+
+ movaps xmm3, xmm0
+ addps xmm0, xmm4
+ mulps xmm0, [GOTOFF(ebx,PD_1_847)] ; xmm0=z5
+ mulps xmm3, [GOTOFF(ebx,PD_M2_613)] ; xmm3=(z10 * -2.613125930)
+ mulps xmm4, [GOTOFF(ebx,PD_1_082)] ; xmm4=(z12 * 1.082392200)
+ addps xmm3, xmm0 ; xmm3=tmp12
+ subps xmm4, xmm0 ; xmm4=tmp10
+
+ ; -- Final output stage
+
+ subps xmm3, xmm1 ; xmm3=tmp6
+ movaps xmm5, xmm6
+ movaps xmm0, xmm7
+ addps xmm6, xmm1 ; xmm6=data0=(00 10 20 30)
+ addps xmm7, xmm3 ; xmm7=data1=(01 11 21 31)
+ subps xmm5, xmm1 ; xmm5=data7=(07 17 27 37)
+ subps xmm0, xmm3 ; xmm0=data6=(06 16 26 36)
+ subps xmm2, xmm3 ; xmm2=tmp5
+
+ movaps xmm1, [GOTOFF(ebx,PD_RNDINT_MAGIC)] ; xmm1=[PD_RNDINT_MAGIC]
+ pcmpeqd xmm3, xmm3
+ psrld xmm3, WORD_BIT ; xmm3={0xFFFF 0x0000 0xFFFF 0x0000 ..}
+
+ addps xmm6, xmm1 ; xmm6=roundint(data0/8)=(00 ** 10 ** 20 ** 30 **)
+ addps xmm7, xmm1 ; xmm7=roundint(data1/8)=(01 ** 11 ** 21 ** 31 **)
+ addps xmm0, xmm1 ; xmm0=roundint(data6/8)=(06 ** 16 ** 26 ** 36 **)
+ addps xmm5, xmm1 ; xmm5=roundint(data7/8)=(07 ** 17 ** 27 ** 37 **)
+
+ pand xmm6, xmm3 ; xmm6=(00 -- 10 -- 20 -- 30 --)
+ pslld xmm7, WORD_BIT ; xmm7=(-- 01 -- 11 -- 21 -- 31)
+ pand xmm0, xmm3 ; xmm0=(06 -- 16 -- 26 -- 36 --)
+ pslld xmm5, WORD_BIT ; xmm5=(-- 07 -- 17 -- 27 -- 37)
+ por xmm6, xmm7 ; xmm6=(00 01 10 11 20 21 30 31)
+ por xmm0, xmm5 ; xmm0=(06 07 16 17 26 27 36 37)
+
+ movaps xmm1, XMMWORD [wk(0)] ; xmm1=tmp2
+ movaps xmm3, XMMWORD [wk(1)] ; xmm3=tmp3
+
+ addps xmm4, xmm2 ; xmm4=tmp4
+ movaps xmm7, xmm1
+ movaps xmm5, xmm3
+ addps xmm1, xmm2 ; xmm1=data2=(02 12 22 32)
+ addps xmm3, xmm4 ; xmm3=data4=(04 14 24 34)
+ subps xmm7, xmm2 ; xmm7=data5=(05 15 25 35)
+ subps xmm5, xmm4 ; xmm5=data3=(03 13 23 33)
+
+ movaps xmm2, [GOTOFF(ebx,PD_RNDINT_MAGIC)] ; xmm2=[PD_RNDINT_MAGIC]
+ pcmpeqd xmm4, xmm4
+ psrld xmm4, WORD_BIT ; xmm4={0xFFFF 0x0000 0xFFFF 0x0000 ..}
+
+ addps xmm3, xmm2 ; xmm3=roundint(data4/8)=(04 ** 14 ** 24 ** 34 **)
+ addps xmm7, xmm2 ; xmm7=roundint(data5/8)=(05 ** 15 ** 25 ** 35 **)
+ addps xmm1, xmm2 ; xmm1=roundint(data2/8)=(02 ** 12 ** 22 ** 32 **)
+ addps xmm5, xmm2 ; xmm5=roundint(data3/8)=(03 ** 13 ** 23 ** 33 **)
+
+ pand xmm3, xmm4 ; xmm3=(04 -- 14 -- 24 -- 34 --)
+ pslld xmm7, WORD_BIT ; xmm7=(-- 05 -- 15 -- 25 -- 35)
+ pand xmm1, xmm4 ; xmm1=(02 -- 12 -- 22 -- 32 --)
+ pslld xmm5, WORD_BIT ; xmm5=(-- 03 -- 13 -- 23 -- 33)
+ por xmm3, xmm7 ; xmm3=(04 05 14 15 24 25 34 35)
+ por xmm1, xmm5 ; xmm1=(02 03 12 13 22 23 32 33)
+
+ movdqa xmm2, [GOTOFF(ebx,PB_CENTERJSAMP)] ; xmm2=[PB_CENTERJSAMP]
+
+ packsswb xmm6, xmm3 ; xmm6=(00 01 10 11 20 21 30 31 04 05 14 15 24 25 34 35)
+ packsswb xmm1, xmm0 ; xmm1=(02 03 12 13 22 23 32 33 06 07 16 17 26 27 36 37)
+ paddb xmm6, xmm2
+ paddb xmm1, xmm2
+
+ movdqa xmm4, xmm6 ; transpose coefficients(phase 2)
+ punpcklwd xmm6, xmm1 ; xmm6=(00 01 02 03 10 11 12 13 20 21 22 23 30 31 32 33)
+ punpckhwd xmm4, xmm1 ; xmm4=(04 05 06 07 14 15 16 17 24 25 26 27 34 35 36 37)
+
+ movdqa xmm7, xmm6 ; transpose coefficients(phase 3)
+ punpckldq xmm6, xmm4 ; xmm6=(00 01 02 03 04 05 06 07 10 11 12 13 14 15 16 17)
+ punpckhdq xmm7, xmm4 ; xmm7=(20 21 22 23 24 25 26 27 30 31 32 33 34 35 36 37)
+
+ pshufd xmm5, xmm6, 0x4E ; xmm5=(10 11 12 13 14 15 16 17 00 01 02 03 04 05 06 07)
+ pshufd xmm3, xmm7, 0x4E ; xmm3=(30 31 32 33 34 35 36 37 20 21 22 23 24 25 26 27)
+
+ pushpic ebx ; save GOT address
+
+ mov edx, JSAMPROW [edi+0*SIZEOF_JSAMPROW]
+ mov ebx, JSAMPROW [edi+2*SIZEOF_JSAMPROW]
+ movq XMM_MMWORD [edx+eax*SIZEOF_JSAMPLE], xmm6
+ movq XMM_MMWORD [ebx+eax*SIZEOF_JSAMPLE], xmm7
+ mov edx, JSAMPROW [edi+1*SIZEOF_JSAMPROW]
+ mov ebx, JSAMPROW [edi+3*SIZEOF_JSAMPROW]
+ movq XMM_MMWORD [edx+eax*SIZEOF_JSAMPLE], xmm5
+ movq XMM_MMWORD [ebx+eax*SIZEOF_JSAMPLE], xmm3
+
+ poppic ebx ; restore GOT address
+
+ add esi, byte 4*SIZEOF_FAST_FLOAT ; wsptr
+ add edi, byte 4*SIZEOF_JSAMPROW
+ dec ecx ; ctr
+ jnz near .rowloop
+
+ pop edi
+ pop esi
+; pop edx ; need not be preserved
+; pop ecx ; need not be preserved
+ pop ebx
+ mov esp, ebp ; esp <- aligned ebp
+ pop esp ; esp <- original ebp
+ pop ebp
+ ret
+
+; For some reason, the OS X linker does not honor the request to align the
+; segment unless we do this.
+ align 32
diff --git a/simd/jidctfst-mmx.asm b/simd/i386/jidctfst-mmx.asm
similarity index 99%
rename from simd/jidctfst-mmx.asm
rename to simd/i386/jidctfst-mmx.asm
index 6e95bfb..7b5c2f2 100644
--- a/simd/jidctfst-mmx.asm
+++ b/simd/i386/jidctfst-mmx.asm
@@ -2,6 +2,7 @@
; jidctfst.asm - fast integer IDCT (MMX)
;
; Copyright 2009 Pierre Ossman <ossman@cendio.se> for Cendio AB
+; Copyright (C) 2016, D. R. Commander.
;
; Based on the x86 SIMD extension for IJG JPEG library
; Copyright (C) 1999-2006, MIYASAKA Masaru.
@@ -57,7 +58,7 @@
%define PRE_MULTIPLY_SCALE_BITS 2
%define CONST_SHIFT (16 - PRE_MULTIPLY_SCALE_BITS - CONST_BITS)
- alignz 16
+ alignz 32
global EXTN(jconst_idct_ifast_mmx)
EXTN(jconst_idct_ifast_mmx):
@@ -68,7 +69,7 @@
PW_F1082 times 4 dw F_1_082 << CONST_SHIFT
PB_CENTERJSAMP times 8 db CENTERJSAMPLE
- alignz 16
+ alignz 32
; --------------------------------------------------------------------------
SECTION SEG_TEXT
@@ -92,7 +93,7 @@
%define workspace wk(0)-DCTSIZE2*SIZEOF_JCOEF
; JCOEF workspace[DCTSIZE2]
- align 16
+ align 32
global EXTN(jsimd_idct_ifast_mmx)
EXTN(jsimd_idct_ifast_mmx):
@@ -496,4 +497,4 @@
; For some reason, the OS X linker does not honor the request to align the
; segment unless we do this.
- align 16
+ align 32
diff --git a/simd/i386/jidctfst-sse2.asm b/simd/i386/jidctfst-sse2.asm
new file mode 100644
index 0000000..51e3317
--- /dev/null
+++ b/simd/i386/jidctfst-sse2.asm
@@ -0,0 +1,502 @@
+;
+; jidctfst.asm - fast integer IDCT (SSE2)
+;
+; Copyright 2009 Pierre Ossman <ossman@cendio.se> for Cendio AB
+; Copyright (C) 2016, D. R. Commander.
+;
+; Based on the x86 SIMD extension for IJG JPEG library
+; Copyright (C) 1999-2006, MIYASAKA Masaru.
+; For conditions of distribution and use, see copyright notice in jsimdext.inc
+;
+; This file should be assembled with NASM (Netwide Assembler),
+; can *not* be assembled with Microsoft's MASM or any compatible
+; assembler (including Borland's Turbo Assembler).
+; NASM is available from http://nasm.sourceforge.net/ or
+; http://sourceforge.net/project/showfiles.php?group_id=6208
+;
+; This file contains a fast, not so accurate integer implementation of
+; the inverse DCT (Discrete Cosine Transform). The following code is
+; based directly on the IJG's original jidctfst.c; see the jidctfst.c
+; for more details.
+;
+; [TAB8]
+
+%include "jsimdext.inc"
+%include "jdct.inc"
+
+; --------------------------------------------------------------------------
+
+%define CONST_BITS 8 ; 14 is also OK.
+%define PASS1_BITS 2
+
+%if IFAST_SCALE_BITS != PASS1_BITS
+%error "'IFAST_SCALE_BITS' must be equal to 'PASS1_BITS'."
+%endif
+
+%if CONST_BITS == 8
+F_1_082 equ 277 ; FIX(1.082392200)
+F_1_414 equ 362 ; FIX(1.414213562)
+F_1_847 equ 473 ; FIX(1.847759065)
+F_2_613 equ 669 ; FIX(2.613125930)
+F_1_613 equ (F_2_613 - 256) ; FIX(2.613125930) - FIX(1)
+%else
+; NASM cannot do compile-time arithmetic on floating-point constants.
+%define DESCALE(x,n) (((x)+(1<<((n)-1)))>>(n))
+F_1_082 equ DESCALE(1162209775, 30-CONST_BITS) ; FIX(1.082392200)
+F_1_414 equ DESCALE(1518500249, 30-CONST_BITS) ; FIX(1.414213562)
+F_1_847 equ DESCALE(1984016188, 30-CONST_BITS) ; FIX(1.847759065)
+F_2_613 equ DESCALE(2805822602, 30-CONST_BITS) ; FIX(2.613125930)
+F_1_613 equ (F_2_613 - (1 << CONST_BITS)) ; FIX(2.613125930) - FIX(1)
+%endif
+
+; --------------------------------------------------------------------------
+ SECTION SEG_CONST
+
+; PRE_MULTIPLY_SCALE_BITS <= 2 (to avoid overflow)
+; CONST_BITS + CONST_SHIFT + PRE_MULTIPLY_SCALE_BITS == 16 (for pmulhw)
+
+%define PRE_MULTIPLY_SCALE_BITS 2
+%define CONST_SHIFT (16 - PRE_MULTIPLY_SCALE_BITS - CONST_BITS)
+
+ alignz 32
+ global EXTN(jconst_idct_ifast_sse2)
+
+EXTN(jconst_idct_ifast_sse2):
+
+PW_F1414 times 8 dw F_1_414 << CONST_SHIFT
+PW_F1847 times 8 dw F_1_847 << CONST_SHIFT
+PW_MF1613 times 8 dw -F_1_613 << CONST_SHIFT
+PW_F1082 times 8 dw F_1_082 << CONST_SHIFT
+PB_CENTERJSAMP times 16 db CENTERJSAMPLE
+
+ alignz 32
+
+; --------------------------------------------------------------------------
+ SECTION SEG_TEXT
+ BITS 32
+;
+; Perform dequantization and inverse DCT on one block of coefficients.
+;
+; GLOBAL(void)
+; jsimd_idct_ifast_sse2 (void *dct_table, JCOEFPTR coef_block,
+; JSAMPARRAY output_buf, JDIMENSION output_col)
+;
+
+%define dct_table(b) (b)+8 ; jpeg_component_info *compptr
+%define coef_block(b) (b)+12 ; JCOEFPTR coef_block
+%define output_buf(b) (b)+16 ; JSAMPARRAY output_buf
+%define output_col(b) (b)+20 ; JDIMENSION output_col
+
+%define original_ebp ebp+0
+%define wk(i) ebp-(WK_NUM-(i))*SIZEOF_XMMWORD ; xmmword wk[WK_NUM]
+%define WK_NUM 2
+
+ align 32
+ global EXTN(jsimd_idct_ifast_sse2)
+
+EXTN(jsimd_idct_ifast_sse2):
+ push ebp
+ mov eax, esp ; eax = original ebp
+ sub esp, byte 4
+ and esp, byte (-SIZEOF_XMMWORD) ; align to 128 bits
+ mov [esp], eax
+ mov ebp, esp ; ebp = aligned ebp
+ lea esp, [wk(0)]
+ pushpic ebx
+; push ecx ; unused
+; push edx ; need not be preserved
+ push esi
+ push edi
+
+ get_GOT ebx ; get GOT address
+
+ ; ---- Pass 1: process columns from input.
+
+; mov eax, [original_ebp]
+ mov edx, POINTER [dct_table(eax)] ; quantptr
+ mov esi, JCOEFPTR [coef_block(eax)] ; inptr
+
+%ifndef NO_ZERO_COLUMN_TEST_IFAST_SSE2
+ mov eax, DWORD [DWBLOCK(1,0,esi,SIZEOF_JCOEF)]
+ or eax, DWORD [DWBLOCK(2,0,esi,SIZEOF_JCOEF)]
+ jnz near .columnDCT
+
+ movdqa xmm0, XMMWORD [XMMBLOCK(1,0,esi,SIZEOF_JCOEF)]
+ movdqa xmm1, XMMWORD [XMMBLOCK(2,0,esi,SIZEOF_JCOEF)]
+ por xmm0, XMMWORD [XMMBLOCK(3,0,esi,SIZEOF_JCOEF)]
+ por xmm1, XMMWORD [XMMBLOCK(4,0,esi,SIZEOF_JCOEF)]
+ por xmm0, XMMWORD [XMMBLOCK(5,0,esi,SIZEOF_JCOEF)]
+ por xmm1, XMMWORD [XMMBLOCK(6,0,esi,SIZEOF_JCOEF)]
+ por xmm0, XMMWORD [XMMBLOCK(7,0,esi,SIZEOF_JCOEF)]
+ por xmm1, xmm0
+ packsswb xmm1, xmm1
+ packsswb xmm1, xmm1
+ movd eax, xmm1
+ test eax, eax
+ jnz short .columnDCT
+
+ ; -- AC terms all zero
+
+ movdqa xmm0, XMMWORD [XMMBLOCK(0,0,esi,SIZEOF_JCOEF)]
+ pmullw xmm0, XMMWORD [XMMBLOCK(0,0,edx,SIZEOF_ISLOW_MULT_TYPE)]
+
+ movdqa xmm7, xmm0 ; xmm0=in0=(00 01 02 03 04 05 06 07)
+ punpcklwd xmm0, xmm0 ; xmm0=(00 00 01 01 02 02 03 03)
+ punpckhwd xmm7, xmm7 ; xmm7=(04 04 05 05 06 06 07 07)
+
+ pshufd xmm6, xmm0, 0x00 ; xmm6=col0=(00 00 00 00 00 00 00 00)
+ pshufd xmm2, xmm0, 0x55 ; xmm2=col1=(01 01 01 01 01 01 01 01)
+ pshufd xmm5, xmm0, 0xAA ; xmm5=col2=(02 02 02 02 02 02 02 02)
+ pshufd xmm0, xmm0, 0xFF ; xmm0=col3=(03 03 03 03 03 03 03 03)
+ pshufd xmm1, xmm7, 0x00 ; xmm1=col4=(04 04 04 04 04 04 04 04)
+ pshufd xmm4, xmm7, 0x55 ; xmm4=col5=(05 05 05 05 05 05 05 05)
+ pshufd xmm3, xmm7, 0xAA ; xmm3=col6=(06 06 06 06 06 06 06 06)
+ pshufd xmm7, xmm7, 0xFF ; xmm7=col7=(07 07 07 07 07 07 07 07)
+
+ movdqa XMMWORD [wk(0)], xmm2 ; wk(0)=col1
+ movdqa XMMWORD [wk(1)], xmm0 ; wk(1)=col3
+ jmp near .column_end
+ alignx 16, 7
+%endif
+.columnDCT:
+
+ ; -- Even part
+
+ movdqa xmm0, XMMWORD [XMMBLOCK(0,0,esi,SIZEOF_JCOEF)]
+ movdqa xmm1, XMMWORD [XMMBLOCK(2,0,esi,SIZEOF_JCOEF)]
+ pmullw xmm0, XMMWORD [XMMBLOCK(0,0,edx,SIZEOF_IFAST_MULT_TYPE)]
+ pmullw xmm1, XMMWORD [XMMBLOCK(2,0,edx,SIZEOF_IFAST_MULT_TYPE)]
+ movdqa xmm2, XMMWORD [XMMBLOCK(4,0,esi,SIZEOF_JCOEF)]
+ movdqa xmm3, XMMWORD [XMMBLOCK(6,0,esi,SIZEOF_JCOEF)]
+ pmullw xmm2, XMMWORD [XMMBLOCK(4,0,edx,SIZEOF_IFAST_MULT_TYPE)]
+ pmullw xmm3, XMMWORD [XMMBLOCK(6,0,edx,SIZEOF_IFAST_MULT_TYPE)]
+
+ movdqa xmm4, xmm0
+ movdqa xmm5, xmm1
+ psubw xmm0, xmm2 ; xmm0=tmp11
+ psubw xmm1, xmm3
+ paddw xmm4, xmm2 ; xmm4=tmp10
+ paddw xmm5, xmm3 ; xmm5=tmp13
+
+ psllw xmm1, PRE_MULTIPLY_SCALE_BITS
+ pmulhw xmm1, [GOTOFF(ebx,PW_F1414)]
+ psubw xmm1, xmm5 ; xmm1=tmp12
+
+ movdqa xmm6, xmm4
+ movdqa xmm7, xmm0
+ psubw xmm4, xmm5 ; xmm4=tmp3
+ psubw xmm0, xmm1 ; xmm0=tmp2
+ paddw xmm6, xmm5 ; xmm6=tmp0
+ paddw xmm7, xmm1 ; xmm7=tmp1
+
+ movdqa XMMWORD [wk(1)], xmm4 ; wk(1)=tmp3
+ movdqa XMMWORD [wk(0)], xmm0 ; wk(0)=tmp2
+
+ ; -- Odd part
+
+ movdqa xmm2, XMMWORD [XMMBLOCK(1,0,esi,SIZEOF_JCOEF)]
+ movdqa xmm3, XMMWORD [XMMBLOCK(3,0,esi,SIZEOF_JCOEF)]
+ pmullw xmm2, XMMWORD [XMMBLOCK(1,0,edx,SIZEOF_IFAST_MULT_TYPE)]
+ pmullw xmm3, XMMWORD [XMMBLOCK(3,0,edx,SIZEOF_IFAST_MULT_TYPE)]
+ movdqa xmm5, XMMWORD [XMMBLOCK(5,0,esi,SIZEOF_JCOEF)]
+ movdqa xmm1, XMMWORD [XMMBLOCK(7,0,esi,SIZEOF_JCOEF)]
+ pmullw xmm5, XMMWORD [XMMBLOCK(5,0,edx,SIZEOF_IFAST_MULT_TYPE)]
+ pmullw xmm1, XMMWORD [XMMBLOCK(7,0,edx,SIZEOF_IFAST_MULT_TYPE)]
+
+ movdqa xmm4, xmm2
+ movdqa xmm0, xmm5
+ psubw xmm2, xmm1 ; xmm2=z12
+ psubw xmm5, xmm3 ; xmm5=z10
+ paddw xmm4, xmm1 ; xmm4=z11
+ paddw xmm0, xmm3 ; xmm0=z13
+
+ movdqa xmm1, xmm5 ; xmm1=z10(unscaled)
+ psllw xmm2, PRE_MULTIPLY_SCALE_BITS
+ psllw xmm5, PRE_MULTIPLY_SCALE_BITS
+
+ movdqa xmm3, xmm4
+ psubw xmm4, xmm0
+ paddw xmm3, xmm0 ; xmm3=tmp7
+
+ psllw xmm4, PRE_MULTIPLY_SCALE_BITS
+ pmulhw xmm4, [GOTOFF(ebx,PW_F1414)] ; xmm4=tmp11
+
+ ; To avoid overflow...
+ ;
+ ; (Original)
+ ; tmp12 = -2.613125930 * z10 + z5;
+ ;
+ ; (This implementation)
+ ; tmp12 = (-1.613125930 - 1) * z10 + z5;
+ ; = -1.613125930 * z10 - z10 + z5;
+
+ movdqa xmm0, xmm5
+ paddw xmm5, xmm2
+ pmulhw xmm5, [GOTOFF(ebx,PW_F1847)] ; xmm5=z5
+ pmulhw xmm0, [GOTOFF(ebx,PW_MF1613)]
+ pmulhw xmm2, [GOTOFF(ebx,PW_F1082)]
+ psubw xmm0, xmm1
+ psubw xmm2, xmm5 ; xmm2=tmp10
+ paddw xmm0, xmm5 ; xmm0=tmp12
+
+ ; -- Final output stage
+
+ psubw xmm0, xmm3 ; xmm0=tmp6
+ movdqa xmm1, xmm6
+ movdqa xmm5, xmm7
+ paddw xmm6, xmm3 ; xmm6=data0=(00 01 02 03 04 05 06 07)
+ paddw xmm7, xmm0 ; xmm7=data1=(10 11 12 13 14 15 16 17)
+ psubw xmm1, xmm3 ; xmm1=data7=(70 71 72 73 74 75 76 77)
+ psubw xmm5, xmm0 ; xmm5=data6=(60 61 62 63 64 65 66 67)
+ psubw xmm4, xmm0 ; xmm4=tmp5
+
+ movdqa xmm3, xmm6 ; transpose coefficients(phase 1)
+ punpcklwd xmm6, xmm7 ; xmm6=(00 10 01 11 02 12 03 13)
+ punpckhwd xmm3, xmm7 ; xmm3=(04 14 05 15 06 16 07 17)
+ movdqa xmm0, xmm5 ; transpose coefficients(phase 1)
+ punpcklwd xmm5, xmm1 ; xmm5=(60 70 61 71 62 72 63 73)
+ punpckhwd xmm0, xmm1 ; xmm0=(64 74 65 75 66 76 67 77)
+
+ movdqa xmm7, XMMWORD [wk(0)] ; xmm7=tmp2
+ movdqa xmm1, XMMWORD [wk(1)] ; xmm1=tmp3
+
+ movdqa XMMWORD [wk(0)], xmm5 ; wk(0)=(60 70 61 71 62 72 63 73)
+ movdqa XMMWORD [wk(1)], xmm0 ; wk(1)=(64 74 65 75 66 76 67 77)
+
+ paddw xmm2, xmm4 ; xmm2=tmp4
+ movdqa xmm5, xmm7
+ movdqa xmm0, xmm1
+ paddw xmm7, xmm4 ; xmm7=data2=(20 21 22 23 24 25 26 27)
+ paddw xmm1, xmm2 ; xmm1=data4=(40 41 42 43 44 45 46 47)
+ psubw xmm5, xmm4 ; xmm5=data5=(50 51 52 53 54 55 56 57)
+ psubw xmm0, xmm2 ; xmm0=data3=(30 31 32 33 34 35 36 37)
+
+ movdqa xmm4, xmm7 ; transpose coefficients(phase 1)
+ punpcklwd xmm7, xmm0 ; xmm7=(20 30 21 31 22 32 23 33)
+ punpckhwd xmm4, xmm0 ; xmm4=(24 34 25 35 26 36 27 37)
+ movdqa xmm2, xmm1 ; transpose coefficients(phase 1)
+ punpcklwd xmm1, xmm5 ; xmm1=(40 50 41 51 42 52 43 53)
+ punpckhwd xmm2, xmm5 ; xmm2=(44 54 45 55 46 56 47 57)
+
+ movdqa xmm0, xmm3 ; transpose coefficients(phase 2)
+ punpckldq xmm3, xmm4 ; xmm3=(04 14 24 34 05 15 25 35)
+ punpckhdq xmm0, xmm4 ; xmm0=(06 16 26 36 07 17 27 37)
+ movdqa xmm5, xmm6 ; transpose coefficients(phase 2)
+ punpckldq xmm6, xmm7 ; xmm6=(00 10 20 30 01 11 21 31)
+ punpckhdq xmm5, xmm7 ; xmm5=(02 12 22 32 03 13 23 33)
+
+ movdqa xmm4, XMMWORD [wk(0)] ; xmm4=(60 70 61 71 62 72 63 73)
+ movdqa xmm7, XMMWORD [wk(1)] ; xmm7=(64 74 65 75 66 76 67 77)
+
+ movdqa XMMWORD [wk(0)], xmm3 ; wk(0)=(04 14 24 34 05 15 25 35)
+ movdqa XMMWORD [wk(1)], xmm0 ; wk(1)=(06 16 26 36 07 17 27 37)
+
+ movdqa xmm3, xmm1 ; transpose coefficients(phase 2)
+ punpckldq xmm1, xmm4 ; xmm1=(40 50 60 70 41 51 61 71)
+ punpckhdq xmm3, xmm4 ; xmm3=(42 52 62 72 43 53 63 73)
+ movdqa xmm0, xmm2 ; transpose coefficients(phase 2)
+ punpckldq xmm2, xmm7 ; xmm2=(44 54 64 74 45 55 65 75)
+ punpckhdq xmm0, xmm7 ; xmm0=(46 56 66 76 47 57 67 77)
+
+ movdqa xmm4, xmm6 ; transpose coefficients(phase 3)
+ punpcklqdq xmm6, xmm1 ; xmm6=col0=(00 10 20 30 40 50 60 70)
+ punpckhqdq xmm4, xmm1 ; xmm4=col1=(01 11 21 31 41 51 61 71)
+ movdqa xmm7, xmm5 ; transpose coefficients(phase 3)
+ punpcklqdq xmm5, xmm3 ; xmm5=col2=(02 12 22 32 42 52 62 72)
+ punpckhqdq xmm7, xmm3 ; xmm7=col3=(03 13 23 33 43 53 63 73)
+
+ movdqa xmm1, XMMWORD [wk(0)] ; xmm1=(04 14 24 34 05 15 25 35)
+ movdqa xmm3, XMMWORD [wk(1)] ; xmm3=(06 16 26 36 07 17 27 37)
+
+ movdqa XMMWORD [wk(0)], xmm4 ; wk(0)=col1
+ movdqa XMMWORD [wk(1)], xmm7 ; wk(1)=col3
+
+ movdqa xmm4, xmm1 ; transpose coefficients(phase 3)
+ punpcklqdq xmm1, xmm2 ; xmm1=col4=(04 14 24 34 44 54 64 74)
+ punpckhqdq xmm4, xmm2 ; xmm4=col5=(05 15 25 35 45 55 65 75)
+ movdqa xmm7, xmm3 ; transpose coefficients(phase 3)
+ punpcklqdq xmm3, xmm0 ; xmm3=col6=(06 16 26 36 46 56 66 76)
+ punpckhqdq xmm7, xmm0 ; xmm7=col7=(07 17 27 37 47 57 67 77)
+.column_end:
+
+ ; -- Prefetch the next coefficient block
+
+ prefetchnta [esi + DCTSIZE2*SIZEOF_JCOEF + 0*32]
+ prefetchnta [esi + DCTSIZE2*SIZEOF_JCOEF + 1*32]
+ prefetchnta [esi + DCTSIZE2*SIZEOF_JCOEF + 2*32]
+ prefetchnta [esi + DCTSIZE2*SIZEOF_JCOEF + 3*32]
+
+ ; ---- Pass 2: process rows from work array, store into output array.
+
+ mov eax, [original_ebp]
+ mov edi, JSAMPARRAY [output_buf(eax)] ; (JSAMPROW *)
+ mov eax, JDIMENSION [output_col(eax)]
+
+ ; -- Even part
+
+ ; xmm6=col0, xmm5=col2, xmm1=col4, xmm3=col6
+
+ movdqa xmm2, xmm6
+ movdqa xmm0, xmm5
+ psubw xmm6, xmm1 ; xmm6=tmp11
+ psubw xmm5, xmm3
+ paddw xmm2, xmm1 ; xmm2=tmp10
+ paddw xmm0, xmm3 ; xmm0=tmp13
+
+ psllw xmm5, PRE_MULTIPLY_SCALE_BITS
+ pmulhw xmm5, [GOTOFF(ebx,PW_F1414)]
+ psubw xmm5, xmm0 ; xmm5=tmp12
+
+ movdqa xmm1, xmm2
+ movdqa xmm3, xmm6
+ psubw xmm2, xmm0 ; xmm2=tmp3
+ psubw xmm6, xmm5 ; xmm6=tmp2
+ paddw xmm1, xmm0 ; xmm1=tmp0
+ paddw xmm3, xmm5 ; xmm3=tmp1
+
+ movdqa xmm0, XMMWORD [wk(0)] ; xmm0=col1
+ movdqa xmm5, XMMWORD [wk(1)] ; xmm5=col3
+
+ movdqa XMMWORD [wk(0)], xmm2 ; wk(0)=tmp3
+ movdqa XMMWORD [wk(1)], xmm6 ; wk(1)=tmp2
+
+ ; -- Odd part
+
+ ; xmm0=col1, xmm5=col3, xmm4=col5, xmm7=col7
+
+ movdqa xmm2, xmm0
+ movdqa xmm6, xmm4
+ psubw xmm0, xmm7 ; xmm0=z12
+ psubw xmm4, xmm5 ; xmm4=z10
+ paddw xmm2, xmm7 ; xmm2=z11
+ paddw xmm6, xmm5 ; xmm6=z13
+
+ movdqa xmm7, xmm4 ; xmm7=z10(unscaled)
+ psllw xmm0, PRE_MULTIPLY_SCALE_BITS
+ psllw xmm4, PRE_MULTIPLY_SCALE_BITS
+
+ movdqa xmm5, xmm2
+ psubw xmm2, xmm6
+ paddw xmm5, xmm6 ; xmm5=tmp7
+
+ psllw xmm2, PRE_MULTIPLY_SCALE_BITS
+ pmulhw xmm2, [GOTOFF(ebx,PW_F1414)] ; xmm2=tmp11
+
+ ; To avoid overflow...
+ ;
+ ; (Original)
+ ; tmp12 = -2.613125930 * z10 + z5;
+ ;
+ ; (This implementation)
+ ; tmp12 = (-1.613125930 - 1) * z10 + z5;
+ ; = -1.613125930 * z10 - z10 + z5;
+
+ movdqa xmm6, xmm4
+ paddw xmm4, xmm0
+ pmulhw xmm4, [GOTOFF(ebx,PW_F1847)] ; xmm4=z5
+ pmulhw xmm6, [GOTOFF(ebx,PW_MF1613)]
+ pmulhw xmm0, [GOTOFF(ebx,PW_F1082)]
+ psubw xmm6, xmm7
+ psubw xmm0, xmm4 ; xmm0=tmp10
+ paddw xmm6, xmm4 ; xmm6=tmp12
+
+ ; -- Final output stage
+
+ psubw xmm6, xmm5 ; xmm6=tmp6
+ movdqa xmm7, xmm1
+ movdqa xmm4, xmm3
+ paddw xmm1, xmm5 ; xmm1=data0=(00 10 20 30 40 50 60 70)
+ paddw xmm3, xmm6 ; xmm3=data1=(01 11 21 31 41 51 61 71)
+ psraw xmm1, (PASS1_BITS+3) ; descale
+ psraw xmm3, (PASS1_BITS+3) ; descale
+ psubw xmm7, xmm5 ; xmm7=data7=(07 17 27 37 47 57 67 77)
+ psubw xmm4, xmm6 ; xmm4=data6=(06 16 26 36 46 56 66 76)
+ psraw xmm7, (PASS1_BITS+3) ; descale
+ psraw xmm4, (PASS1_BITS+3) ; descale
+ psubw xmm2, xmm6 ; xmm2=tmp5
+
+ packsswb xmm1, xmm4 ; xmm1=(00 10 20 30 40 50 60 70 06 16 26 36 46 56 66 76)
+ packsswb xmm3, xmm7 ; xmm3=(01 11 21 31 41 51 61 71 07 17 27 37 47 57 67 77)
+
+ movdqa xmm5, XMMWORD [wk(1)] ; xmm5=tmp2
+ movdqa xmm6, XMMWORD [wk(0)] ; xmm6=tmp3
+
+ paddw xmm0, xmm2 ; xmm0=tmp4
+ movdqa xmm4, xmm5
+ movdqa xmm7, xmm6
+ paddw xmm5, xmm2 ; xmm5=data2=(02 12 22 32 42 52 62 72)
+ paddw xmm6, xmm0 ; xmm6=data4=(04 14 24 34 44 54 64 74)
+ psraw xmm5, (PASS1_BITS+3) ; descale
+ psraw xmm6, (PASS1_BITS+3) ; descale
+ psubw xmm4, xmm2 ; xmm4=data5=(05 15 25 35 45 55 65 75)
+ psubw xmm7, xmm0 ; xmm7=data3=(03 13 23 33 43 53 63 73)
+ psraw xmm4, (PASS1_BITS+3) ; descale
+ psraw xmm7, (PASS1_BITS+3) ; descale
+
+ movdqa xmm2, [GOTOFF(ebx,PB_CENTERJSAMP)] ; xmm2=[PB_CENTERJSAMP]
+
+ packsswb xmm5, xmm6 ; xmm5=(02 12 22 32 42 52 62 72 04 14 24 34 44 54 64 74)
+ packsswb xmm7, xmm4 ; xmm7=(03 13 23 33 43 53 63 73 05 15 25 35 45 55 65 75)
+
+ paddb xmm1, xmm2
+ paddb xmm3, xmm2
+ paddb xmm5, xmm2
+ paddb xmm7, xmm2
+
+ movdqa xmm0, xmm1 ; transpose coefficients(phase 1)
+ punpcklbw xmm1, xmm3 ; xmm1=(00 01 10 11 20 21 30 31 40 41 50 51 60 61 70 71)
+ punpckhbw xmm0, xmm3 ; xmm0=(06 07 16 17 26 27 36 37 46 47 56 57 66 67 76 77)
+ movdqa xmm6, xmm5 ; transpose coefficients(phase 1)
+ punpcklbw xmm5, xmm7 ; xmm5=(02 03 12 13 22 23 32 33 42 43 52 53 62 63 72 73)
+ punpckhbw xmm6, xmm7 ; xmm6=(04 05 14 15 24 25 34 35 44 45 54 55 64 65 74 75)
+
+ movdqa xmm4, xmm1 ; transpose coefficients(phase 2)
+ punpcklwd xmm1, xmm5 ; xmm1=(00 01 02 03 10 11 12 13 20 21 22 23 30 31 32 33)
+ punpckhwd xmm4, xmm5 ; xmm4=(40 41 42 43 50 51 52 53 60 61 62 63 70 71 72 73)
+ movdqa xmm2, xmm6 ; transpose coefficients(phase 2)
+ punpcklwd xmm6, xmm0 ; xmm6=(04 05 06 07 14 15 16 17 24 25 26 27 34 35 36 37)
+ punpckhwd xmm2, xmm0 ; xmm2=(44 45 46 47 54 55 56 57 64 65 66 67 74 75 76 77)
+
+ movdqa xmm3, xmm1 ; transpose coefficients(phase 3)
+ punpckldq xmm1, xmm6 ; xmm1=(00 01 02 03 04 05 06 07 10 11 12 13 14 15 16 17)
+ punpckhdq xmm3, xmm6 ; xmm3=(20 21 22 23 24 25 26 27 30 31 32 33 34 35 36 37)
+ movdqa xmm7, xmm4 ; transpose coefficients(phase 3)
+ punpckldq xmm4, xmm2 ; xmm4=(40 41 42 43 44 45 46 47 50 51 52 53 54 55 56 57)
+ punpckhdq xmm7, xmm2 ; xmm7=(60 61 62 63 64 65 66 67 70 71 72 73 74 75 76 77)
+
+ pshufd xmm5, xmm1, 0x4E ; xmm5=(10 11 12 13 14 15 16 17 00 01 02 03 04 05 06 07)
+ pshufd xmm0, xmm3, 0x4E ; xmm0=(30 31 32 33 34 35 36 37 20 21 22 23 24 25 26 27)
+ pshufd xmm6, xmm4, 0x4E ; xmm6=(50 51 52 53 54 55 56 57 40 41 42 43 44 45 46 47)
+ pshufd xmm2, xmm7, 0x4E ; xmm2=(70 71 72 73 74 75 76 77 60 61 62 63 64 65 66 67)
+
+ mov edx, JSAMPROW [edi+0*SIZEOF_JSAMPROW]
+ mov esi, JSAMPROW [edi+2*SIZEOF_JSAMPROW]
+ movq XMM_MMWORD [edx+eax*SIZEOF_JSAMPLE], xmm1
+ movq XMM_MMWORD [esi+eax*SIZEOF_JSAMPLE], xmm3
+ mov edx, JSAMPROW [edi+4*SIZEOF_JSAMPROW]
+ mov esi, JSAMPROW [edi+6*SIZEOF_JSAMPROW]
+ movq XMM_MMWORD [edx+eax*SIZEOF_JSAMPLE], xmm4
+ movq XMM_MMWORD [esi+eax*SIZEOF_JSAMPLE], xmm7
+
+ mov edx, JSAMPROW [edi+1*SIZEOF_JSAMPROW]
+ mov esi, JSAMPROW [edi+3*SIZEOF_JSAMPROW]
+ movq XMM_MMWORD [edx+eax*SIZEOF_JSAMPLE], xmm5
+ movq XMM_MMWORD [esi+eax*SIZEOF_JSAMPLE], xmm0
+ mov edx, JSAMPROW [edi+5*SIZEOF_JSAMPROW]
+ mov esi, JSAMPROW [edi+7*SIZEOF_JSAMPROW]
+ movq XMM_MMWORD [edx+eax*SIZEOF_JSAMPLE], xmm6
+ movq XMM_MMWORD [esi+eax*SIZEOF_JSAMPLE], xmm2
+
+ pop edi
+ pop esi
+; pop edx ; need not be preserved
+; pop ecx ; unused
+ poppic ebx
+ mov esp, ebp ; esp <- aligned ebp
+ pop esp ; esp <- original ebp
+ pop ebp
+ ret
+
+; For some reason, the OS X linker does not honor the request to align the
+; segment unless we do this.
+ align 32
diff --git a/simd/jidctint-mmx.asm b/simd/i386/jidctint-mmx.asm
similarity index 99%
rename from simd/jidctint-mmx.asm
rename to simd/i386/jidctint-mmx.asm
index 5bd1981..6ef7ff7 100644
--- a/simd/jidctint-mmx.asm
+++ b/simd/i386/jidctint-mmx.asm
@@ -2,6 +2,7 @@
; jidctint.asm - accurate integer IDCT (MMX)
;
; Copyright 2009 Pierre Ossman <ossman@cendio.se> for Cendio AB
+; Copyright (C) 2016, D. R. Commander.
;
; Based on the x86 SIMD extension for IJG JPEG library
; Copyright (C) 1999-2006, MIYASAKA Masaru.
@@ -64,7 +65,7 @@
; --------------------------------------------------------------------------
SECTION SEG_CONST
- alignz 16
+ alignz 32
global EXTN(jconst_idct_islow_mmx)
EXTN(jconst_idct_islow_mmx):
@@ -81,7 +82,7 @@
PD_DESCALE_P2 times 2 dd 1 << (DESCALE_P2-1)
PB_CENTERJSAMP times 8 db CENTERJSAMPLE
- alignz 16
+ alignz 32
; --------------------------------------------------------------------------
SECTION SEG_TEXT
@@ -105,7 +106,7 @@
%define workspace wk(0)-DCTSIZE2*SIZEOF_JCOEF
; JCOEF workspace[DCTSIZE2]
- align 16
+ align 32
global EXTN(jsimd_idct_islow_mmx)
EXTN(jsimd_idct_islow_mmx):
@@ -848,4 +849,4 @@
; For some reason, the OS X linker does not honor the request to align the
; segment unless we do this.
- align 16
+ align 32
diff --git a/simd/i386/jidctint-sse2.asm b/simd/i386/jidctint-sse2.asm
new file mode 100644
index 0000000..441da70
--- /dev/null
+++ b/simd/i386/jidctint-sse2.asm
@@ -0,0 +1,859 @@
+;
+; jidctint.asm - accurate integer IDCT (SSE2)
+;
+; Copyright 2009 Pierre Ossman <ossman@cendio.se> for Cendio AB
+; Copyright (C) 2016, D. R. Commander.
+;
+; Based on the x86 SIMD extension for IJG JPEG library
+; Copyright (C) 1999-2006, MIYASAKA Masaru.
+; For conditions of distribution and use, see copyright notice in jsimdext.inc
+;
+; This file should be assembled with NASM (Netwide Assembler),
+; can *not* be assembled with Microsoft's MASM or any compatible
+; assembler (including Borland's Turbo Assembler).
+; NASM is available from http://nasm.sourceforge.net/ or
+; http://sourceforge.net/project/showfiles.php?group_id=6208
+;
+; This file contains a slow-but-accurate integer implementation of the
+; inverse DCT (Discrete Cosine Transform). The following code is based
+; directly on the IJG's original jidctint.c; see the jidctint.c for
+; more details.
+;
+; [TAB8]
+
+%include "jsimdext.inc"
+%include "jdct.inc"
+
+; --------------------------------------------------------------------------
+
+%define CONST_BITS 13
+%define PASS1_BITS 2
+
+%define DESCALE_P1 (CONST_BITS-PASS1_BITS)
+%define DESCALE_P2 (CONST_BITS+PASS1_BITS+3)
+
+%if CONST_BITS == 13
+F_0_298 equ 2446 ; FIX(0.298631336)
+F_0_390 equ 3196 ; FIX(0.390180644)
+F_0_541 equ 4433 ; FIX(0.541196100)
+F_0_765 equ 6270 ; FIX(0.765366865)
+F_0_899 equ 7373 ; FIX(0.899976223)
+F_1_175 equ 9633 ; FIX(1.175875602)
+F_1_501 equ 12299 ; FIX(1.501321110)
+F_1_847 equ 15137 ; FIX(1.847759065)
+F_1_961 equ 16069 ; FIX(1.961570560)
+F_2_053 equ 16819 ; FIX(2.053119869)
+F_2_562 equ 20995 ; FIX(2.562915447)
+F_3_072 equ 25172 ; FIX(3.072711026)
+%else
+; NASM cannot do compile-time arithmetic on floating-point constants.
+%define DESCALE(x,n) (((x)+(1<<((n)-1)))>>(n))
+F_0_298 equ DESCALE( 320652955, 30-CONST_BITS) ; FIX(0.298631336)
+F_0_390 equ DESCALE( 418953276, 30-CONST_BITS) ; FIX(0.390180644)
+F_0_541 equ DESCALE( 581104887, 30-CONST_BITS) ; FIX(0.541196100)
+F_0_765 equ DESCALE( 821806413, 30-CONST_BITS) ; FIX(0.765366865)
+F_0_899 equ DESCALE( 966342111, 30-CONST_BITS) ; FIX(0.899976223)
+F_1_175 equ DESCALE(1262586813, 30-CONST_BITS) ; FIX(1.175875602)
+F_1_501 equ DESCALE(1612031267, 30-CONST_BITS) ; FIX(1.501321110)
+F_1_847 equ DESCALE(1984016188, 30-CONST_BITS) ; FIX(1.847759065)
+F_1_961 equ DESCALE(2106220350, 30-CONST_BITS) ; FIX(1.961570560)
+F_2_053 equ DESCALE(2204520673, 30-CONST_BITS) ; FIX(2.053119869)
+F_2_562 equ DESCALE(2751909506, 30-CONST_BITS) ; FIX(2.562915447)
+F_3_072 equ DESCALE(3299298341, 30-CONST_BITS) ; FIX(3.072711026)
+%endif
+
+; --------------------------------------------------------------------------
+ SECTION SEG_CONST
+
+ alignz 32
+ global EXTN(jconst_idct_islow_sse2)
+
+EXTN(jconst_idct_islow_sse2):
+
+PW_F130_F054 times 4 dw (F_0_541+F_0_765), F_0_541
+PW_F054_MF130 times 4 dw F_0_541, (F_0_541-F_1_847)
+PW_MF078_F117 times 4 dw (F_1_175-F_1_961), F_1_175
+PW_F117_F078 times 4 dw F_1_175, (F_1_175-F_0_390)
+PW_MF060_MF089 times 4 dw (F_0_298-F_0_899),-F_0_899
+PW_MF089_F060 times 4 dw -F_0_899, (F_1_501-F_0_899)
+PW_MF050_MF256 times 4 dw (F_2_053-F_2_562),-F_2_562
+PW_MF256_F050 times 4 dw -F_2_562, (F_3_072-F_2_562)
+PD_DESCALE_P1 times 4 dd 1 << (DESCALE_P1-1)
+PD_DESCALE_P2 times 4 dd 1 << (DESCALE_P2-1)
+PB_CENTERJSAMP times 16 db CENTERJSAMPLE
+
+ alignz 32
+
+; --------------------------------------------------------------------------
+ SECTION SEG_TEXT
+ BITS 32
+;
+; Perform dequantization and inverse DCT on one block of coefficients.
+;
+; GLOBAL(void)
+; jsimd_idct_islow_sse2 (void *dct_table, JCOEFPTR coef_block,
+; JSAMPARRAY output_buf, JDIMENSION output_col)
+;
+
+%define dct_table(b) (b)+8 ; jpeg_component_info *compptr
+%define coef_block(b) (b)+12 ; JCOEFPTR coef_block
+%define output_buf(b) (b)+16 ; JSAMPARRAY output_buf
+%define output_col(b) (b)+20 ; JDIMENSION output_col
+
+%define original_ebp ebp+0
+%define wk(i) ebp-(WK_NUM-(i))*SIZEOF_XMMWORD ; xmmword wk[WK_NUM]
+%define WK_NUM 12
+
+ align 32
+ global EXTN(jsimd_idct_islow_sse2)
+
+EXTN(jsimd_idct_islow_sse2):
+ push ebp
+ mov eax, esp ; eax = original ebp
+ sub esp, byte 4
+ and esp, byte (-SIZEOF_XMMWORD) ; align to 128 bits
+ mov [esp], eax
+ mov ebp, esp ; ebp = aligned ebp
+ lea esp, [wk(0)]
+ pushpic ebx
+; push ecx ; unused
+; push edx ; need not be preserved
+ push esi
+ push edi
+
+ get_GOT ebx ; get GOT address
+
+ ; ---- Pass 1: process columns from input.
+
+; mov eax, [original_ebp]
+ mov edx, POINTER [dct_table(eax)] ; quantptr
+ mov esi, JCOEFPTR [coef_block(eax)] ; inptr
+
+%ifndef NO_ZERO_COLUMN_TEST_ISLOW_SSE2
+ mov eax, DWORD [DWBLOCK(1,0,esi,SIZEOF_JCOEF)]
+ or eax, DWORD [DWBLOCK(2,0,esi,SIZEOF_JCOEF)]
+ jnz near .columnDCT
+
+ movdqa xmm0, XMMWORD [XMMBLOCK(1,0,esi,SIZEOF_JCOEF)]
+ movdqa xmm1, XMMWORD [XMMBLOCK(2,0,esi,SIZEOF_JCOEF)]
+ por xmm0, XMMWORD [XMMBLOCK(3,0,esi,SIZEOF_JCOEF)]
+ por xmm1, XMMWORD [XMMBLOCK(4,0,esi,SIZEOF_JCOEF)]
+ por xmm0, XMMWORD [XMMBLOCK(5,0,esi,SIZEOF_JCOEF)]
+ por xmm1, XMMWORD [XMMBLOCK(6,0,esi,SIZEOF_JCOEF)]
+ por xmm0, XMMWORD [XMMBLOCK(7,0,esi,SIZEOF_JCOEF)]
+ por xmm1, xmm0
+ packsswb xmm1, xmm1
+ packsswb xmm1, xmm1
+ movd eax, xmm1
+ test eax, eax
+ jnz short .columnDCT
+
+ ; -- AC terms all zero
+
+ movdqa xmm5, XMMWORD [XMMBLOCK(0,0,esi,SIZEOF_JCOEF)]
+ pmullw xmm5, XMMWORD [XMMBLOCK(0,0,edx,SIZEOF_ISLOW_MULT_TYPE)]
+
+ psllw xmm5, PASS1_BITS
+
+ movdqa xmm4, xmm5 ; xmm5=in0=(00 01 02 03 04 05 06 07)
+ punpcklwd xmm5, xmm5 ; xmm5=(00 00 01 01 02 02 03 03)
+ punpckhwd xmm4, xmm4 ; xmm4=(04 04 05 05 06 06 07 07)
+
+ pshufd xmm7, xmm5, 0x00 ; xmm7=col0=(00 00 00 00 00 00 00 00)
+ pshufd xmm6, xmm5, 0x55 ; xmm6=col1=(01 01 01 01 01 01 01 01)
+ pshufd xmm1, xmm5, 0xAA ; xmm1=col2=(02 02 02 02 02 02 02 02)
+ pshufd xmm5, xmm5, 0xFF ; xmm5=col3=(03 03 03 03 03 03 03 03)
+ pshufd xmm0, xmm4, 0x00 ; xmm0=col4=(04 04 04 04 04 04 04 04)
+ pshufd xmm3, xmm4, 0x55 ; xmm3=col5=(05 05 05 05 05 05 05 05)
+ pshufd xmm2, xmm4, 0xAA ; xmm2=col6=(06 06 06 06 06 06 06 06)
+ pshufd xmm4, xmm4, 0xFF ; xmm4=col7=(07 07 07 07 07 07 07 07)
+
+ movdqa XMMWORD [wk(8)], xmm6 ; wk(8)=col1
+ movdqa XMMWORD [wk(9)], xmm5 ; wk(9)=col3
+ movdqa XMMWORD [wk(10)], xmm3 ; wk(10)=col5
+ movdqa XMMWORD [wk(11)], xmm4 ; wk(11)=col7
+ jmp near .column_end
+ alignx 16, 7
+%endif
+.columnDCT:
+
+ ; -- Even part
+
+ movdqa xmm0, XMMWORD [XMMBLOCK(0,0,esi,SIZEOF_JCOEF)]
+ movdqa xmm1, XMMWORD [XMMBLOCK(2,0,esi,SIZEOF_JCOEF)]
+ pmullw xmm0, XMMWORD [XMMBLOCK(0,0,edx,SIZEOF_ISLOW_MULT_TYPE)]
+ pmullw xmm1, XMMWORD [XMMBLOCK(2,0,edx,SIZEOF_ISLOW_MULT_TYPE)]
+ movdqa xmm2, XMMWORD [XMMBLOCK(4,0,esi,SIZEOF_JCOEF)]
+ movdqa xmm3, XMMWORD [XMMBLOCK(6,0,esi,SIZEOF_JCOEF)]
+ pmullw xmm2, XMMWORD [XMMBLOCK(4,0,edx,SIZEOF_ISLOW_MULT_TYPE)]
+ pmullw xmm3, XMMWORD [XMMBLOCK(6,0,edx,SIZEOF_ISLOW_MULT_TYPE)]
+
+ ; (Original)
+ ; z1 = (z2 + z3) * 0.541196100;
+ ; tmp2 = z1 + z3 * -1.847759065;
+ ; tmp3 = z1 + z2 * 0.765366865;
+ ;
+ ; (This implementation)
+ ; tmp2 = z2 * 0.541196100 + z3 * (0.541196100 - 1.847759065);
+ ; tmp3 = z2 * (0.541196100 + 0.765366865) + z3 * 0.541196100;
+
+ movdqa xmm4, xmm1 ; xmm1=in2=z2
+ movdqa xmm5, xmm1
+ punpcklwd xmm4, xmm3 ; xmm3=in6=z3
+ punpckhwd xmm5, xmm3
+ movdqa xmm1, xmm4
+ movdqa xmm3, xmm5
+ pmaddwd xmm4, [GOTOFF(ebx,PW_F130_F054)] ; xmm4=tmp3L
+ pmaddwd xmm5, [GOTOFF(ebx,PW_F130_F054)] ; xmm5=tmp3H
+ pmaddwd xmm1, [GOTOFF(ebx,PW_F054_MF130)] ; xmm1=tmp2L
+ pmaddwd xmm3, [GOTOFF(ebx,PW_F054_MF130)] ; xmm3=tmp2H
+
+ movdqa xmm6, xmm0
+ paddw xmm0, xmm2 ; xmm0=in0+in4
+ psubw xmm6, xmm2 ; xmm6=in0-in4
+
+ pxor xmm7, xmm7
+ pxor xmm2, xmm2
+ punpcklwd xmm7, xmm0 ; xmm7=tmp0L
+ punpckhwd xmm2, xmm0 ; xmm2=tmp0H
+ psrad xmm7, (16-CONST_BITS) ; psrad xmm7,16 & pslld xmm7,CONST_BITS
+ psrad xmm2, (16-CONST_BITS) ; psrad xmm2,16 & pslld xmm2,CONST_BITS
+
+ movdqa xmm0, xmm7
+ paddd xmm7, xmm4 ; xmm7=tmp10L
+ psubd xmm0, xmm4 ; xmm0=tmp13L
+ movdqa xmm4, xmm2
+ paddd xmm2, xmm5 ; xmm2=tmp10H
+ psubd xmm4, xmm5 ; xmm4=tmp13H
+
+ movdqa XMMWORD [wk(0)], xmm7 ; wk(0)=tmp10L
+ movdqa XMMWORD [wk(1)], xmm2 ; wk(1)=tmp10H
+ movdqa XMMWORD [wk(2)], xmm0 ; wk(2)=tmp13L
+ movdqa XMMWORD [wk(3)], xmm4 ; wk(3)=tmp13H
+
+ pxor xmm5, xmm5
+ pxor xmm7, xmm7
+ punpcklwd xmm5, xmm6 ; xmm5=tmp1L
+ punpckhwd xmm7, xmm6 ; xmm7=tmp1H
+ psrad xmm5, (16-CONST_BITS) ; psrad xmm5,16 & pslld xmm5,CONST_BITS
+ psrad xmm7, (16-CONST_BITS) ; psrad xmm7,16 & pslld xmm7,CONST_BITS
+
+ movdqa xmm2, xmm5
+ paddd xmm5, xmm1 ; xmm5=tmp11L
+ psubd xmm2, xmm1 ; xmm2=tmp12L
+ movdqa xmm0, xmm7
+ paddd xmm7, xmm3 ; xmm7=tmp11H
+ psubd xmm0, xmm3 ; xmm0=tmp12H
+
+ movdqa XMMWORD [wk(4)], xmm5 ; wk(4)=tmp11L
+ movdqa XMMWORD [wk(5)], xmm7 ; wk(5)=tmp11H
+ movdqa XMMWORD [wk(6)], xmm2 ; wk(6)=tmp12L
+ movdqa XMMWORD [wk(7)], xmm0 ; wk(7)=tmp12H
+
+ ; -- Odd part
+
+ movdqa xmm4, XMMWORD [XMMBLOCK(1,0,esi,SIZEOF_JCOEF)]
+ movdqa xmm6, XMMWORD [XMMBLOCK(3,0,esi,SIZEOF_JCOEF)]
+ pmullw xmm4, XMMWORD [XMMBLOCK(1,0,edx,SIZEOF_ISLOW_MULT_TYPE)]
+ pmullw xmm6, XMMWORD [XMMBLOCK(3,0,edx,SIZEOF_ISLOW_MULT_TYPE)]
+ movdqa xmm1, XMMWORD [XMMBLOCK(5,0,esi,SIZEOF_JCOEF)]
+ movdqa xmm3, XMMWORD [XMMBLOCK(7,0,esi,SIZEOF_JCOEF)]
+ pmullw xmm1, XMMWORD [XMMBLOCK(5,0,edx,SIZEOF_ISLOW_MULT_TYPE)]
+ pmullw xmm3, XMMWORD [XMMBLOCK(7,0,edx,SIZEOF_ISLOW_MULT_TYPE)]
+
+ movdqa xmm5, xmm6
+ movdqa xmm7, xmm4
+ paddw xmm5, xmm3 ; xmm5=z3
+ paddw xmm7, xmm1 ; xmm7=z4
+
+ ; (Original)
+ ; z5 = (z3 + z4) * 1.175875602;
+ ; z3 = z3 * -1.961570560; z4 = z4 * -0.390180644;
+ ; z3 += z5; z4 += z5;
+ ;
+ ; (This implementation)
+ ; z3 = z3 * (1.175875602 - 1.961570560) + z4 * 1.175875602;
+ ; z4 = z3 * 1.175875602 + z4 * (1.175875602 - 0.390180644);
+
+ movdqa xmm2, xmm5
+ movdqa xmm0, xmm5
+ punpcklwd xmm2, xmm7
+ punpckhwd xmm0, xmm7
+ movdqa xmm5, xmm2
+ movdqa xmm7, xmm0
+ pmaddwd xmm2, [GOTOFF(ebx,PW_MF078_F117)] ; xmm2=z3L
+ pmaddwd xmm0, [GOTOFF(ebx,PW_MF078_F117)] ; xmm0=z3H
+ pmaddwd xmm5, [GOTOFF(ebx,PW_F117_F078)] ; xmm5=z4L
+ pmaddwd xmm7, [GOTOFF(ebx,PW_F117_F078)] ; xmm7=z4H
+
+ movdqa XMMWORD [wk(10)], xmm2 ; wk(10)=z3L
+ movdqa XMMWORD [wk(11)], xmm0 ; wk(11)=z3H
+
+ ; (Original)
+ ; z1 = tmp0 + tmp3; z2 = tmp1 + tmp2;
+ ; tmp0 = tmp0 * 0.298631336; tmp1 = tmp1 * 2.053119869;
+ ; tmp2 = tmp2 * 3.072711026; tmp3 = tmp3 * 1.501321110;
+ ; z1 = z1 * -0.899976223; z2 = z2 * -2.562915447;
+ ; tmp0 += z1 + z3; tmp1 += z2 + z4;
+ ; tmp2 += z2 + z3; tmp3 += z1 + z4;
+ ;
+ ; (This implementation)
+ ; tmp0 = tmp0 * (0.298631336 - 0.899976223) + tmp3 * -0.899976223;
+ ; tmp1 = tmp1 * (2.053119869 - 2.562915447) + tmp2 * -2.562915447;
+ ; tmp2 = tmp1 * -2.562915447 + tmp2 * (3.072711026 - 2.562915447);
+ ; tmp3 = tmp0 * -0.899976223 + tmp3 * (1.501321110 - 0.899976223);
+ ; tmp0 += z3; tmp1 += z4;
+ ; tmp2 += z3; tmp3 += z4;
+
+ movdqa xmm2, xmm3
+ movdqa xmm0, xmm3
+ punpcklwd xmm2, xmm4
+ punpckhwd xmm0, xmm4
+ movdqa xmm3, xmm2
+ movdqa xmm4, xmm0
+ pmaddwd xmm2, [GOTOFF(ebx,PW_MF060_MF089)] ; xmm2=tmp0L
+ pmaddwd xmm0, [GOTOFF(ebx,PW_MF060_MF089)] ; xmm0=tmp0H
+ pmaddwd xmm3, [GOTOFF(ebx,PW_MF089_F060)] ; xmm3=tmp3L
+ pmaddwd xmm4, [GOTOFF(ebx,PW_MF089_F060)] ; xmm4=tmp3H
+
+ paddd xmm2, XMMWORD [wk(10)] ; xmm2=tmp0L
+ paddd xmm0, XMMWORD [wk(11)] ; xmm0=tmp0H
+ paddd xmm3, xmm5 ; xmm3=tmp3L
+ paddd xmm4, xmm7 ; xmm4=tmp3H
+
+ movdqa XMMWORD [wk(8)], xmm2 ; wk(8)=tmp0L
+ movdqa XMMWORD [wk(9)], xmm0 ; wk(9)=tmp0H
+
+ movdqa xmm2, xmm1
+ movdqa xmm0, xmm1
+ punpcklwd xmm2, xmm6
+ punpckhwd xmm0, xmm6
+ movdqa xmm1, xmm2
+ movdqa xmm6, xmm0
+ pmaddwd xmm2, [GOTOFF(ebx,PW_MF050_MF256)] ; xmm2=tmp1L
+ pmaddwd xmm0, [GOTOFF(ebx,PW_MF050_MF256)] ; xmm0=tmp1H
+ pmaddwd xmm1, [GOTOFF(ebx,PW_MF256_F050)] ; xmm1=tmp2L
+ pmaddwd xmm6, [GOTOFF(ebx,PW_MF256_F050)] ; xmm6=tmp2H
+
+ paddd xmm2, xmm5 ; xmm2=tmp1L
+ paddd xmm0, xmm7 ; xmm0=tmp1H
+ paddd xmm1, XMMWORD [wk(10)] ; xmm1=tmp2L
+ paddd xmm6, XMMWORD [wk(11)] ; xmm6=tmp2H
+
+ movdqa XMMWORD [wk(10)], xmm2 ; wk(10)=tmp1L
+ movdqa XMMWORD [wk(11)], xmm0 ; wk(11)=tmp1H
+
+ ; -- Final output stage
+
+ movdqa xmm5, XMMWORD [wk(0)] ; xmm5=tmp10L
+ movdqa xmm7, XMMWORD [wk(1)] ; xmm7=tmp10H
+
+ movdqa xmm2, xmm5
+ movdqa xmm0, xmm7
+ paddd xmm5, xmm3 ; xmm5=data0L
+ paddd xmm7, xmm4 ; xmm7=data0H
+ psubd xmm2, xmm3 ; xmm2=data7L
+ psubd xmm0, xmm4 ; xmm0=data7H
+
+ movdqa xmm3, [GOTOFF(ebx,PD_DESCALE_P1)] ; xmm3=[PD_DESCALE_P1]
+
+ paddd xmm5, xmm3
+ paddd xmm7, xmm3
+ psrad xmm5, DESCALE_P1
+ psrad xmm7, DESCALE_P1
+ paddd xmm2, xmm3
+ paddd xmm0, xmm3
+ psrad xmm2, DESCALE_P1
+ psrad xmm0, DESCALE_P1
+
+ packssdw xmm5, xmm7 ; xmm5=data0=(00 01 02 03 04 05 06 07)
+ packssdw xmm2, xmm0 ; xmm2=data7=(70 71 72 73 74 75 76 77)
+
+ movdqa xmm4, XMMWORD [wk(4)] ; xmm4=tmp11L
+ movdqa xmm3, XMMWORD [wk(5)] ; xmm3=tmp11H
+
+ movdqa xmm7, xmm4
+ movdqa xmm0, xmm3
+ paddd xmm4, xmm1 ; xmm4=data1L
+ paddd xmm3, xmm6 ; xmm3=data1H
+ psubd xmm7, xmm1 ; xmm7=data6L
+ psubd xmm0, xmm6 ; xmm0=data6H
+
+ movdqa xmm1, [GOTOFF(ebx,PD_DESCALE_P1)] ; xmm1=[PD_DESCALE_P1]
+
+ paddd xmm4, xmm1
+ paddd xmm3, xmm1
+ psrad xmm4, DESCALE_P1
+ psrad xmm3, DESCALE_P1
+ paddd xmm7, xmm1
+ paddd xmm0, xmm1
+ psrad xmm7, DESCALE_P1
+ psrad xmm0, DESCALE_P1
+
+ packssdw xmm4, xmm3 ; xmm4=data1=(10 11 12 13 14 15 16 17)
+ packssdw xmm7, xmm0 ; xmm7=data6=(60 61 62 63 64 65 66 67)
+
+ movdqa xmm6, xmm5 ; transpose coefficients(phase 1)
+ punpcklwd xmm5, xmm4 ; xmm5=(00 10 01 11 02 12 03 13)
+ punpckhwd xmm6, xmm4 ; xmm6=(04 14 05 15 06 16 07 17)
+ movdqa xmm1, xmm7 ; transpose coefficients(phase 1)
+ punpcklwd xmm7, xmm2 ; xmm7=(60 70 61 71 62 72 63 73)
+ punpckhwd xmm1, xmm2 ; xmm1=(64 74 65 75 66 76 67 77)
+
+ movdqa xmm3, XMMWORD [wk(6)] ; xmm3=tmp12L
+ movdqa xmm0, XMMWORD [wk(7)] ; xmm0=tmp12H
+ movdqa xmm4, XMMWORD [wk(10)] ; xmm4=tmp1L
+ movdqa xmm2, XMMWORD [wk(11)] ; xmm2=tmp1H
+
+ movdqa XMMWORD [wk(0)], xmm5 ; wk(0)=(00 10 01 11 02 12 03 13)
+ movdqa XMMWORD [wk(1)], xmm6 ; wk(1)=(04 14 05 15 06 16 07 17)
+ movdqa XMMWORD [wk(4)], xmm7 ; wk(4)=(60 70 61 71 62 72 63 73)
+ movdqa XMMWORD [wk(5)], xmm1 ; wk(5)=(64 74 65 75 66 76 67 77)
+
+ movdqa xmm5, xmm3
+ movdqa xmm6, xmm0
+ paddd xmm3, xmm4 ; xmm3=data2L
+ paddd xmm0, xmm2 ; xmm0=data2H
+ psubd xmm5, xmm4 ; xmm5=data5L
+ psubd xmm6, xmm2 ; xmm6=data5H
+
+ movdqa xmm7, [GOTOFF(ebx,PD_DESCALE_P1)] ; xmm7=[PD_DESCALE_P1]
+
+ paddd xmm3, xmm7
+ paddd xmm0, xmm7
+ psrad xmm3, DESCALE_P1
+ psrad xmm0, DESCALE_P1
+ paddd xmm5, xmm7
+ paddd xmm6, xmm7
+ psrad xmm5, DESCALE_P1
+ psrad xmm6, DESCALE_P1
+
+ packssdw xmm3, xmm0 ; xmm3=data2=(20 21 22 23 24 25 26 27)
+ packssdw xmm5, xmm6 ; xmm5=data5=(50 51 52 53 54 55 56 57)
+
+ movdqa xmm1, XMMWORD [wk(2)] ; xmm1=tmp13L
+ movdqa xmm4, XMMWORD [wk(3)] ; xmm4=tmp13H
+ movdqa xmm2, XMMWORD [wk(8)] ; xmm2=tmp0L
+ movdqa xmm7, XMMWORD [wk(9)] ; xmm7=tmp0H
+
+ movdqa xmm0, xmm1
+ movdqa xmm6, xmm4
+ paddd xmm1, xmm2 ; xmm1=data3L
+ paddd xmm4, xmm7 ; xmm4=data3H
+ psubd xmm0, xmm2 ; xmm0=data4L
+ psubd xmm6, xmm7 ; xmm6=data4H
+
+ movdqa xmm2, [GOTOFF(ebx,PD_DESCALE_P1)] ; xmm2=[PD_DESCALE_P1]
+
+ paddd xmm1, xmm2
+ paddd xmm4, xmm2
+ psrad xmm1, DESCALE_P1
+ psrad xmm4, DESCALE_P1
+ paddd xmm0, xmm2
+ paddd xmm6, xmm2
+ psrad xmm0, DESCALE_P1
+ psrad xmm6, DESCALE_P1
+
+ packssdw xmm1, xmm4 ; xmm1=data3=(30 31 32 33 34 35 36 37)
+ packssdw xmm0, xmm6 ; xmm0=data4=(40 41 42 43 44 45 46 47)
+
+ movdqa xmm7, XMMWORD [wk(0)] ; xmm7=(00 10 01 11 02 12 03 13)
+ movdqa xmm2, XMMWORD [wk(1)] ; xmm2=(04 14 05 15 06 16 07 17)
+
+ movdqa xmm4, xmm3 ; transpose coefficients(phase 1)
+ punpcklwd xmm3, xmm1 ; xmm3=(20 30 21 31 22 32 23 33)
+ punpckhwd xmm4, xmm1 ; xmm4=(24 34 25 35 26 36 27 37)
+ movdqa xmm6, xmm0 ; transpose coefficients(phase 1)
+ punpcklwd xmm0, xmm5 ; xmm0=(40 50 41 51 42 52 43 53)
+ punpckhwd xmm6, xmm5 ; xmm6=(44 54 45 55 46 56 47 57)
+
+ movdqa xmm1, xmm7 ; transpose coefficients(phase 2)
+ punpckldq xmm7, xmm3 ; xmm7=(00 10 20 30 01 11 21 31)
+ punpckhdq xmm1, xmm3 ; xmm1=(02 12 22 32 03 13 23 33)
+ movdqa xmm5, xmm2 ; transpose coefficients(phase 2)
+ punpckldq xmm2, xmm4 ; xmm2=(04 14 24 34 05 15 25 35)
+ punpckhdq xmm5, xmm4 ; xmm5=(06 16 26 36 07 17 27 37)
+
+ movdqa xmm3, XMMWORD [wk(4)] ; xmm3=(60 70 61 71 62 72 63 73)
+ movdqa xmm4, XMMWORD [wk(5)] ; xmm4=(64 74 65 75 66 76 67 77)
+
+ movdqa XMMWORD [wk(6)], xmm2 ; wk(6)=(04 14 24 34 05 15 25 35)
+ movdqa XMMWORD [wk(7)], xmm5 ; wk(7)=(06 16 26 36 07 17 27 37)
+
+ movdqa xmm2, xmm0 ; transpose coefficients(phase 2)
+ punpckldq xmm0, xmm3 ; xmm0=(40 50 60 70 41 51 61 71)
+ punpckhdq xmm2, xmm3 ; xmm2=(42 52 62 72 43 53 63 73)
+ movdqa xmm5, xmm6 ; transpose coefficients(phase 2)
+ punpckldq xmm6, xmm4 ; xmm6=(44 54 64 74 45 55 65 75)
+ punpckhdq xmm5, xmm4 ; xmm5=(46 56 66 76 47 57 67 77)
+
+ movdqa xmm3, xmm7 ; transpose coefficients(phase 3)
+ punpcklqdq xmm7, xmm0 ; xmm7=col0=(00 10 20 30 40 50 60 70)
+ punpckhqdq xmm3, xmm0 ; xmm3=col1=(01 11 21 31 41 51 61 71)
+ movdqa xmm4, xmm1 ; transpose coefficients(phase 3)
+ punpcklqdq xmm1, xmm2 ; xmm1=col2=(02 12 22 32 42 52 62 72)
+ punpckhqdq xmm4, xmm2 ; xmm4=col3=(03 13 23 33 43 53 63 73)
+
+ movdqa xmm0, XMMWORD [wk(6)] ; xmm0=(04 14 24 34 05 15 25 35)
+ movdqa xmm2, XMMWORD [wk(7)] ; xmm2=(06 16 26 36 07 17 27 37)
+
+ movdqa XMMWORD [wk(8)], xmm3 ; wk(8)=col1
+ movdqa XMMWORD [wk(9)], xmm4 ; wk(9)=col3
+
+ movdqa xmm3, xmm0 ; transpose coefficients(phase 3)
+ punpcklqdq xmm0, xmm6 ; xmm0=col4=(04 14 24 34 44 54 64 74)
+ punpckhqdq xmm3, xmm6 ; xmm3=col5=(05 15 25 35 45 55 65 75)
+ movdqa xmm4, xmm2 ; transpose coefficients(phase 3)
+ punpcklqdq xmm2, xmm5 ; xmm2=col6=(06 16 26 36 46 56 66 76)
+ punpckhqdq xmm4, xmm5 ; xmm4=col7=(07 17 27 37 47 57 67 77)
+
+ movdqa XMMWORD [wk(10)], xmm3 ; wk(10)=col5
+ movdqa XMMWORD [wk(11)], xmm4 ; wk(11)=col7
+.column_end:
+
+ ; -- Prefetch the next coefficient block
+
+ prefetchnta [esi + DCTSIZE2*SIZEOF_JCOEF + 0*32]
+ prefetchnta [esi + DCTSIZE2*SIZEOF_JCOEF + 1*32]
+ prefetchnta [esi + DCTSIZE2*SIZEOF_JCOEF + 2*32]
+ prefetchnta [esi + DCTSIZE2*SIZEOF_JCOEF + 3*32]
+
+ ; ---- Pass 2: process rows from work array, store into output array.
+
+ mov eax, [original_ebp]
+ mov edi, JSAMPARRAY [output_buf(eax)] ; (JSAMPROW *)
+ mov eax, JDIMENSION [output_col(eax)]
+
+ ; -- Even part
+
+ ; xmm7=col0, xmm1=col2, xmm0=col4, xmm2=col6
+
+ ; (Original)
+ ; z1 = (z2 + z3) * 0.541196100;
+ ; tmp2 = z1 + z3 * -1.847759065;
+ ; tmp3 = z1 + z2 * 0.765366865;
+ ;
+ ; (This implementation)
+ ; tmp2 = z2 * 0.541196100 + z3 * (0.541196100 - 1.847759065);
+ ; tmp3 = z2 * (0.541196100 + 0.765366865) + z3 * 0.541196100;
+
+ movdqa xmm6, xmm1 ; xmm1=in2=z2
+ movdqa xmm5, xmm1
+ punpcklwd xmm6, xmm2 ; xmm2=in6=z3
+ punpckhwd xmm5, xmm2
+ movdqa xmm1, xmm6
+ movdqa xmm2, xmm5
+ pmaddwd xmm6, [GOTOFF(ebx,PW_F130_F054)] ; xmm6=tmp3L
+ pmaddwd xmm5, [GOTOFF(ebx,PW_F130_F054)] ; xmm5=tmp3H
+ pmaddwd xmm1, [GOTOFF(ebx,PW_F054_MF130)] ; xmm1=tmp2L
+ pmaddwd xmm2, [GOTOFF(ebx,PW_F054_MF130)] ; xmm2=tmp2H
+
+ movdqa xmm3, xmm7
+ paddw xmm7, xmm0 ; xmm7=in0+in4
+ psubw xmm3, xmm0 ; xmm3=in0-in4
+
+ pxor xmm4, xmm4
+ pxor xmm0, xmm0
+ punpcklwd xmm4, xmm7 ; xmm4=tmp0L
+ punpckhwd xmm0, xmm7 ; xmm0=tmp0H
+ psrad xmm4, (16-CONST_BITS) ; psrad xmm4,16 & pslld xmm4,CONST_BITS
+ psrad xmm0, (16-CONST_BITS) ; psrad xmm0,16 & pslld xmm0,CONST_BITS
+
+ movdqa xmm7, xmm4
+ paddd xmm4, xmm6 ; xmm4=tmp10L
+ psubd xmm7, xmm6 ; xmm7=tmp13L
+ movdqa xmm6, xmm0
+ paddd xmm0, xmm5 ; xmm0=tmp10H
+ psubd xmm6, xmm5 ; xmm6=tmp13H
+
+ movdqa XMMWORD [wk(0)], xmm4 ; wk(0)=tmp10L
+ movdqa XMMWORD [wk(1)], xmm0 ; wk(1)=tmp10H
+ movdqa XMMWORD [wk(2)], xmm7 ; wk(2)=tmp13L
+ movdqa XMMWORD [wk(3)], xmm6 ; wk(3)=tmp13H
+
+ pxor xmm5, xmm5
+ pxor xmm4, xmm4
+ punpcklwd xmm5, xmm3 ; xmm5=tmp1L
+ punpckhwd xmm4, xmm3 ; xmm4=tmp1H
+ psrad xmm5, (16-CONST_BITS) ; psrad xmm5,16 & pslld xmm5,CONST_BITS
+ psrad xmm4, (16-CONST_BITS) ; psrad xmm4,16 & pslld xmm4,CONST_BITS
+
+ movdqa xmm0, xmm5
+ paddd xmm5, xmm1 ; xmm5=tmp11L
+ psubd xmm0, xmm1 ; xmm0=tmp12L
+ movdqa xmm7, xmm4
+ paddd xmm4, xmm2 ; xmm4=tmp11H
+ psubd xmm7, xmm2 ; xmm7=tmp12H
+
+ movdqa XMMWORD [wk(4)], xmm5 ; wk(4)=tmp11L
+ movdqa XMMWORD [wk(5)], xmm4 ; wk(5)=tmp11H
+ movdqa XMMWORD [wk(6)], xmm0 ; wk(6)=tmp12L
+ movdqa XMMWORD [wk(7)], xmm7 ; wk(7)=tmp12H
+
+ ; -- Odd part
+
+ movdqa xmm6, XMMWORD [wk(9)] ; xmm6=col3
+ movdqa xmm3, XMMWORD [wk(8)] ; xmm3=col1
+ movdqa xmm1, XMMWORD [wk(11)] ; xmm1=col7
+ movdqa xmm2, XMMWORD [wk(10)] ; xmm2=col5
+
+ movdqa xmm5, xmm6
+ movdqa xmm4, xmm3
+ paddw xmm5, xmm1 ; xmm5=z3
+ paddw xmm4, xmm2 ; xmm4=z4
+
+ ; (Original)
+ ; z5 = (z3 + z4) * 1.175875602;
+ ; z3 = z3 * -1.961570560; z4 = z4 * -0.390180644;
+ ; z3 += z5; z4 += z5;
+ ;
+ ; (This implementation)
+ ; z3 = z3 * (1.175875602 - 1.961570560) + z4 * 1.175875602;
+ ; z4 = z3 * 1.175875602 + z4 * (1.175875602 - 0.390180644);
+
+ movdqa xmm0, xmm5
+ movdqa xmm7, xmm5
+ punpcklwd xmm0, xmm4
+ punpckhwd xmm7, xmm4
+ movdqa xmm5, xmm0
+ movdqa xmm4, xmm7
+ pmaddwd xmm0, [GOTOFF(ebx,PW_MF078_F117)] ; xmm0=z3L
+ pmaddwd xmm7, [GOTOFF(ebx,PW_MF078_F117)] ; xmm7=z3H
+ pmaddwd xmm5, [GOTOFF(ebx,PW_F117_F078)] ; xmm5=z4L
+ pmaddwd xmm4, [GOTOFF(ebx,PW_F117_F078)] ; xmm4=z4H
+
+ movdqa XMMWORD [wk(10)], xmm0 ; wk(10)=z3L
+ movdqa XMMWORD [wk(11)], xmm7 ; wk(11)=z3H
+
+ ; (Original)
+ ; z1 = tmp0 + tmp3; z2 = tmp1 + tmp2;
+ ; tmp0 = tmp0 * 0.298631336; tmp1 = tmp1 * 2.053119869;
+ ; tmp2 = tmp2 * 3.072711026; tmp3 = tmp3 * 1.501321110;
+ ; z1 = z1 * -0.899976223; z2 = z2 * -2.562915447;
+ ; tmp0 += z1 + z3; tmp1 += z2 + z4;
+ ; tmp2 += z2 + z3; tmp3 += z1 + z4;
+ ;
+ ; (This implementation)
+ ; tmp0 = tmp0 * (0.298631336 - 0.899976223) + tmp3 * -0.899976223;
+ ; tmp1 = tmp1 * (2.053119869 - 2.562915447) + tmp2 * -2.562915447;
+ ; tmp2 = tmp1 * -2.562915447 + tmp2 * (3.072711026 - 2.562915447);
+ ; tmp3 = tmp0 * -0.899976223 + tmp3 * (1.501321110 - 0.899976223);
+ ; tmp0 += z3; tmp1 += z4;
+ ; tmp2 += z3; tmp3 += z4;
+
+ movdqa xmm0, xmm1
+ movdqa xmm7, xmm1
+ punpcklwd xmm0, xmm3
+ punpckhwd xmm7, xmm3
+ movdqa xmm1, xmm0
+ movdqa xmm3, xmm7
+ pmaddwd xmm0, [GOTOFF(ebx,PW_MF060_MF089)] ; xmm0=tmp0L
+ pmaddwd xmm7, [GOTOFF(ebx,PW_MF060_MF089)] ; xmm7=tmp0H
+ pmaddwd xmm1, [GOTOFF(ebx,PW_MF089_F060)] ; xmm1=tmp3L
+ pmaddwd xmm3, [GOTOFF(ebx,PW_MF089_F060)] ; xmm3=tmp3H
+
+ paddd xmm0, XMMWORD [wk(10)] ; xmm0=tmp0L
+ paddd xmm7, XMMWORD [wk(11)] ; xmm7=tmp0H
+ paddd xmm1, xmm5 ; xmm1=tmp3L
+ paddd xmm3, xmm4 ; xmm3=tmp3H
+
+ movdqa XMMWORD [wk(8)], xmm0 ; wk(8)=tmp0L
+ movdqa XMMWORD [wk(9)], xmm7 ; wk(9)=tmp0H
+
+ movdqa xmm0, xmm2
+ movdqa xmm7, xmm2
+ punpcklwd xmm0, xmm6
+ punpckhwd xmm7, xmm6
+ movdqa xmm2, xmm0
+ movdqa xmm6, xmm7
+ pmaddwd xmm0, [GOTOFF(ebx,PW_MF050_MF256)] ; xmm0=tmp1L
+ pmaddwd xmm7, [GOTOFF(ebx,PW_MF050_MF256)] ; xmm7=tmp1H
+ pmaddwd xmm2, [GOTOFF(ebx,PW_MF256_F050)] ; xmm2=tmp2L
+ pmaddwd xmm6, [GOTOFF(ebx,PW_MF256_F050)] ; xmm6=tmp2H
+
+ paddd xmm0, xmm5 ; xmm0=tmp1L
+ paddd xmm7, xmm4 ; xmm7=tmp1H
+ paddd xmm2, XMMWORD [wk(10)] ; xmm2=tmp2L
+ paddd xmm6, XMMWORD [wk(11)] ; xmm6=tmp2H
+
+ movdqa XMMWORD [wk(10)], xmm0 ; wk(10)=tmp1L
+ movdqa XMMWORD [wk(11)], xmm7 ; wk(11)=tmp1H
+
+ ; -- Final output stage
+
+ movdqa xmm5, XMMWORD [wk(0)] ; xmm5=tmp10L
+ movdqa xmm4, XMMWORD [wk(1)] ; xmm4=tmp10H
+
+ movdqa xmm0, xmm5
+ movdqa xmm7, xmm4
+ paddd xmm5, xmm1 ; xmm5=data0L
+ paddd xmm4, xmm3 ; xmm4=data0H
+ psubd xmm0, xmm1 ; xmm0=data7L
+ psubd xmm7, xmm3 ; xmm7=data7H
+
+ movdqa xmm1, [GOTOFF(ebx,PD_DESCALE_P2)] ; xmm1=[PD_DESCALE_P2]
+
+ paddd xmm5, xmm1
+ paddd xmm4, xmm1
+ psrad xmm5, DESCALE_P2
+ psrad xmm4, DESCALE_P2
+ paddd xmm0, xmm1
+ paddd xmm7, xmm1
+ psrad xmm0, DESCALE_P2
+ psrad xmm7, DESCALE_P2
+
+ packssdw xmm5, xmm4 ; xmm5=data0=(00 10 20 30 40 50 60 70)
+ packssdw xmm0, xmm7 ; xmm0=data7=(07 17 27 37 47 57 67 77)
+
+ movdqa xmm3, XMMWORD [wk(4)] ; xmm3=tmp11L
+ movdqa xmm1, XMMWORD [wk(5)] ; xmm1=tmp11H
+
+ movdqa xmm4, xmm3
+ movdqa xmm7, xmm1
+ paddd xmm3, xmm2 ; xmm3=data1L
+ paddd xmm1, xmm6 ; xmm1=data1H
+ psubd xmm4, xmm2 ; xmm4=data6L
+ psubd xmm7, xmm6 ; xmm7=data6H
+
+ movdqa xmm2, [GOTOFF(ebx,PD_DESCALE_P2)] ; xmm2=[PD_DESCALE_P2]
+
+ paddd xmm3, xmm2
+ paddd xmm1, xmm2
+ psrad xmm3, DESCALE_P2
+ psrad xmm1, DESCALE_P2
+ paddd xmm4, xmm2
+ paddd xmm7, xmm2
+ psrad xmm4, DESCALE_P2
+ psrad xmm7, DESCALE_P2
+
+ packssdw xmm3, xmm1 ; xmm3=data1=(01 11 21 31 41 51 61 71)
+ packssdw xmm4, xmm7 ; xmm4=data6=(06 16 26 36 46 56 66 76)
+
+ packsswb xmm5, xmm4 ; xmm5=(00 10 20 30 40 50 60 70 06 16 26 36 46 56 66 76)
+ packsswb xmm3, xmm0 ; xmm3=(01 11 21 31 41 51 61 71 07 17 27 37 47 57 67 77)
+
+ movdqa xmm6, XMMWORD [wk(6)] ; xmm6=tmp12L
+ movdqa xmm2, XMMWORD [wk(7)] ; xmm2=tmp12H
+ movdqa xmm1, XMMWORD [wk(10)] ; xmm1=tmp1L
+ movdqa xmm7, XMMWORD [wk(11)] ; xmm7=tmp1H
+
+ movdqa XMMWORD [wk(0)], xmm5 ; wk(0)=(00 10 20 30 40 50 60 70 06 16 26 36 46 56 66 76)
+ movdqa XMMWORD [wk(1)], xmm3 ; wk(1)=(01 11 21 31 41 51 61 71 07 17 27 37 47 57 67 77)
+
+ movdqa xmm4, xmm6
+ movdqa xmm0, xmm2
+ paddd xmm6, xmm1 ; xmm6=data2L
+ paddd xmm2, xmm7 ; xmm2=data2H
+ psubd xmm4, xmm1 ; xmm4=data5L
+ psubd xmm0, xmm7 ; xmm0=data5H
+
+ movdqa xmm5, [GOTOFF(ebx,PD_DESCALE_P2)] ; xmm5=[PD_DESCALE_P2]
+
+ paddd xmm6, xmm5
+ paddd xmm2, xmm5
+ psrad xmm6, DESCALE_P2
+ psrad xmm2, DESCALE_P2
+ paddd xmm4, xmm5
+ paddd xmm0, xmm5
+ psrad xmm4, DESCALE_P2
+ psrad xmm0, DESCALE_P2
+
+ packssdw xmm6, xmm2 ; xmm6=data2=(02 12 22 32 42 52 62 72)
+ packssdw xmm4, xmm0 ; xmm4=data5=(05 15 25 35 45 55 65 75)
+
+ movdqa xmm3, XMMWORD [wk(2)] ; xmm3=tmp13L
+ movdqa xmm1, XMMWORD [wk(3)] ; xmm1=tmp13H
+ movdqa xmm7, XMMWORD [wk(8)] ; xmm7=tmp0L
+ movdqa xmm5, XMMWORD [wk(9)] ; xmm5=tmp0H
+
+ movdqa xmm2, xmm3
+ movdqa xmm0, xmm1
+ paddd xmm3, xmm7 ; xmm3=data3L
+ paddd xmm1, xmm5 ; xmm1=data3H
+ psubd xmm2, xmm7 ; xmm2=data4L
+ psubd xmm0, xmm5 ; xmm0=data4H
+
+ movdqa xmm7, [GOTOFF(ebx,PD_DESCALE_P2)] ; xmm7=[PD_DESCALE_P2]
+
+ paddd xmm3, xmm7
+ paddd xmm1, xmm7
+ psrad xmm3, DESCALE_P2
+ psrad xmm1, DESCALE_P2
+ paddd xmm2, xmm7
+ paddd xmm0, xmm7
+ psrad xmm2, DESCALE_P2
+ psrad xmm0, DESCALE_P2
+
+ movdqa xmm5, [GOTOFF(ebx,PB_CENTERJSAMP)] ; xmm5=[PB_CENTERJSAMP]
+
+ packssdw xmm3, xmm1 ; xmm3=data3=(03 13 23 33 43 53 63 73)
+ packssdw xmm2, xmm0 ; xmm2=data4=(04 14 24 34 44 54 64 74)
+
+ movdqa xmm7, XMMWORD [wk(0)] ; xmm7=(00 10 20 30 40 50 60 70 06 16 26 36 46 56 66 76)
+ movdqa xmm1, XMMWORD [wk(1)] ; xmm1=(01 11 21 31 41 51 61 71 07 17 27 37 47 57 67 77)
+
+ packsswb xmm6, xmm2 ; xmm6=(02 12 22 32 42 52 62 72 04 14 24 34 44 54 64 74)
+ packsswb xmm3, xmm4 ; xmm3=(03 13 23 33 43 53 63 73 05 15 25 35 45 55 65 75)
+
+ paddb xmm7, xmm5
+ paddb xmm1, xmm5
+ paddb xmm6, xmm5
+ paddb xmm3, xmm5
+
+ movdqa xmm0, xmm7 ; transpose coefficients(phase 1)
+ punpcklbw xmm7, xmm1 ; xmm7=(00 01 10 11 20 21 30 31 40 41 50 51 60 61 70 71)
+ punpckhbw xmm0, xmm1 ; xmm0=(06 07 16 17 26 27 36 37 46 47 56 57 66 67 76 77)
+ movdqa xmm2, xmm6 ; transpose coefficients(phase 1)
+ punpcklbw xmm6, xmm3 ; xmm6=(02 03 12 13 22 23 32 33 42 43 52 53 62 63 72 73)
+ punpckhbw xmm2, xmm3 ; xmm2=(04 05 14 15 24 25 34 35 44 45 54 55 64 65 74 75)
+
+ movdqa xmm4, xmm7 ; transpose coefficients(phase 2)
+ punpcklwd xmm7, xmm6 ; xmm7=(00 01 02 03 10 11 12 13 20 21 22 23 30 31 32 33)
+ punpckhwd xmm4, xmm6 ; xmm4=(40 41 42 43 50 51 52 53 60 61 62 63 70 71 72 73)
+ movdqa xmm5, xmm2 ; transpose coefficients(phase 2)
+ punpcklwd xmm2, xmm0 ; xmm2=(04 05 06 07 14 15 16 17 24 25 26 27 34 35 36 37)
+ punpckhwd xmm5, xmm0 ; xmm5=(44 45 46 47 54 55 56 57 64 65 66 67 74 75 76 77)
+
+ movdqa xmm1, xmm7 ; transpose coefficients(phase 3)
+ punpckldq xmm7, xmm2 ; xmm7=(00 01 02 03 04 05 06 07 10 11 12 13 14 15 16 17)
+ punpckhdq xmm1, xmm2 ; xmm1=(20 21 22 23 24 25 26 27 30 31 32 33 34 35 36 37)
+ movdqa xmm3, xmm4 ; transpose coefficients(phase 3)
+ punpckldq xmm4, xmm5 ; xmm4=(40 41 42 43 44 45 46 47 50 51 52 53 54 55 56 57)
+ punpckhdq xmm3, xmm5 ; xmm3=(60 61 62 63 64 65 66 67 70 71 72 73 74 75 76 77)
+
+ pshufd xmm6, xmm7, 0x4E ; xmm6=(10 11 12 13 14 15 16 17 00 01 02 03 04 05 06 07)
+ pshufd xmm0, xmm1, 0x4E ; xmm0=(30 31 32 33 34 35 36 37 20 21 22 23 24 25 26 27)
+ pshufd xmm2, xmm4, 0x4E ; xmm2=(50 51 52 53 54 55 56 57 40 41 42 43 44 45 46 47)
+ pshufd xmm5, xmm3, 0x4E ; xmm5=(70 71 72 73 74 75 76 77 60 61 62 63 64 65 66 67)
+
+ mov edx, JSAMPROW [edi+0*SIZEOF_JSAMPROW]
+ mov esi, JSAMPROW [edi+2*SIZEOF_JSAMPROW]
+ movq XMM_MMWORD [edx+eax*SIZEOF_JSAMPLE], xmm7
+ movq XMM_MMWORD [esi+eax*SIZEOF_JSAMPLE], xmm1
+ mov edx, JSAMPROW [edi+4*SIZEOF_JSAMPROW]
+ mov esi, JSAMPROW [edi+6*SIZEOF_JSAMPROW]
+ movq XMM_MMWORD [edx+eax*SIZEOF_JSAMPLE], xmm4
+ movq XMM_MMWORD [esi+eax*SIZEOF_JSAMPLE], xmm3
+
+ mov edx, JSAMPROW [edi+1*SIZEOF_JSAMPROW]
+ mov esi, JSAMPROW [edi+3*SIZEOF_JSAMPROW]
+ movq XMM_MMWORD [edx+eax*SIZEOF_JSAMPLE], xmm6
+ movq XMM_MMWORD [esi+eax*SIZEOF_JSAMPLE], xmm0
+ mov edx, JSAMPROW [edi+5*SIZEOF_JSAMPROW]
+ mov esi, JSAMPROW [edi+7*SIZEOF_JSAMPROW]
+ movq XMM_MMWORD [edx+eax*SIZEOF_JSAMPLE], xmm2
+ movq XMM_MMWORD [esi+eax*SIZEOF_JSAMPLE], xmm5
+
+ pop edi
+ pop esi
+; pop edx ; need not be preserved
+; pop ecx ; unused
+ poppic ebx
+ mov esp, ebp ; esp <- aligned ebp
+ pop esp ; esp <- original ebp
+ pop ebp
+ ret
+
+; For some reason, the OS X linker does not honor the request to align the
+; segment unless we do this.
+ align 32
diff --git a/simd/jidctred-mmx.asm b/simd/i386/jidctred-mmx.asm
similarity index 99%
rename from simd/jidctred-mmx.asm
rename to simd/i386/jidctred-mmx.asm
index ba054e3..d5b7e2e 100644
--- a/simd/jidctred-mmx.asm
+++ b/simd/i386/jidctred-mmx.asm
@@ -2,6 +2,7 @@
; jidctred.asm - reduced-size IDCT (MMX)
;
; Copyright 2009 Pierre Ossman <ossman@cendio.se> for Cendio AB
+; Copyright (C) 2016, D. R. Commander.
;
; Based on the x86 SIMD extension for IJG JPEG library
; Copyright (C) 1999-2006, MIYASAKA Masaru.
@@ -70,7 +71,7 @@
; --------------------------------------------------------------------------
SECTION SEG_CONST
- alignz 16
+ alignz 32
global EXTN(jconst_idct_red_mmx)
EXTN(jconst_idct_red_mmx):
@@ -88,7 +89,7 @@
PD_DESCALE_P2_2 times 2 dd 1 << (DESCALE_P2_2-1)
PB_CENTERJSAMP times 8 db CENTERJSAMPLE
- alignz 16
+ alignz 32
; --------------------------------------------------------------------------
SECTION SEG_TEXT
@@ -113,7 +114,7 @@
%define workspace wk(0)-DCTSIZE2*SIZEOF_JCOEF
; JCOEF workspace[DCTSIZE2]
- align 16
+ align 32
global EXTN(jsimd_idct_4x4_mmx)
EXTN(jsimd_idct_4x4_mmx):
@@ -501,7 +502,7 @@
%define output_buf(b) (b)+16 ; JSAMPARRAY output_buf
%define output_col(b) (b)+20 ; JDIMENSION output_col
- align 16
+ align 32
global EXTN(jsimd_idct_2x2_mmx)
EXTN(jsimd_idct_2x2_mmx):
@@ -702,4 +703,4 @@
; For some reason, the OS X linker does not honor the request to align the
; segment unless we do this.
- align 16
+ align 32
diff --git a/simd/i386/jidctred-sse2.asm b/simd/i386/jidctred-sse2.asm
new file mode 100644
index 0000000..d0609ac
--- /dev/null
+++ b/simd/i386/jidctred-sse2.asm
@@ -0,0 +1,594 @@
+;
+; jidctred.asm - reduced-size IDCT (SSE2)
+;
+; Copyright 2009 Pierre Ossman <ossman@cendio.se> for Cendio AB
+; Copyright (C) 2016, D. R. Commander.
+;
+; Based on the x86 SIMD extension for IJG JPEG library
+; Copyright (C) 1999-2006, MIYASAKA Masaru.
+; For conditions of distribution and use, see copyright notice in jsimdext.inc
+;
+; This file should be assembled with NASM (Netwide Assembler),
+; can *not* be assembled with Microsoft's MASM or any compatible
+; assembler (including Borland's Turbo Assembler).
+; NASM is available from http://nasm.sourceforge.net/ or
+; http://sourceforge.net/project/showfiles.php?group_id=6208
+;
+; This file contains inverse-DCT routines that produce reduced-size
+; output: either 4x4 or 2x2 pixels from an 8x8 DCT block.
+; The following code is based directly on the IJG's original jidctred.c;
+; see the jidctred.c for more details.
+;
+; [TAB8]
+
+%include "jsimdext.inc"
+%include "jdct.inc"
+
+; --------------------------------------------------------------------------
+
+%define CONST_BITS 13
+%define PASS1_BITS 2
+
+%define DESCALE_P1_4 (CONST_BITS-PASS1_BITS+1)
+%define DESCALE_P2_4 (CONST_BITS+PASS1_BITS+3+1)
+%define DESCALE_P1_2 (CONST_BITS-PASS1_BITS+2)
+%define DESCALE_P2_2 (CONST_BITS+PASS1_BITS+3+2)
+
+%if CONST_BITS == 13
+F_0_211 equ 1730 ; FIX(0.211164243)
+F_0_509 equ 4176 ; FIX(0.509795579)
+F_0_601 equ 4926 ; FIX(0.601344887)
+F_0_720 equ 5906 ; FIX(0.720959822)
+F_0_765 equ 6270 ; FIX(0.765366865)
+F_0_850 equ 6967 ; FIX(0.850430095)
+F_0_899 equ 7373 ; FIX(0.899976223)
+F_1_061 equ 8697 ; FIX(1.061594337)
+F_1_272 equ 10426 ; FIX(1.272758580)
+F_1_451 equ 11893 ; FIX(1.451774981)
+F_1_847 equ 15137 ; FIX(1.847759065)
+F_2_172 equ 17799 ; FIX(2.172734803)
+F_2_562 equ 20995 ; FIX(2.562915447)
+F_3_624 equ 29692 ; FIX(3.624509785)
+%else
+; NASM cannot do compile-time arithmetic on floating-point constants.
+%define DESCALE(x,n) (((x)+(1<<((n)-1)))>>(n))
+F_0_211 equ DESCALE( 226735879, 30-CONST_BITS) ; FIX(0.211164243)
+F_0_509 equ DESCALE( 547388834, 30-CONST_BITS) ; FIX(0.509795579)
+F_0_601 equ DESCALE( 645689155, 30-CONST_BITS) ; FIX(0.601344887)
+F_0_720 equ DESCALE( 774124714, 30-CONST_BITS) ; FIX(0.720959822)
+F_0_765 equ DESCALE( 821806413, 30-CONST_BITS) ; FIX(0.765366865)
+F_0_850 equ DESCALE( 913142361, 30-CONST_BITS) ; FIX(0.850430095)
+F_0_899 equ DESCALE( 966342111, 30-CONST_BITS) ; FIX(0.899976223)
+F_1_061 equ DESCALE(1139878239, 30-CONST_BITS) ; FIX(1.061594337)
+F_1_272 equ DESCALE(1366614119, 30-CONST_BITS) ; FIX(1.272758580)
+F_1_451 equ DESCALE(1558831516, 30-CONST_BITS) ; FIX(1.451774981)
+F_1_847 equ DESCALE(1984016188, 30-CONST_BITS) ; FIX(1.847759065)
+F_2_172 equ DESCALE(2332956230, 30-CONST_BITS) ; FIX(2.172734803)
+F_2_562 equ DESCALE(2751909506, 30-CONST_BITS) ; FIX(2.562915447)
+F_3_624 equ DESCALE(3891787747, 30-CONST_BITS) ; FIX(3.624509785)
+%endif
+
+; --------------------------------------------------------------------------
+ SECTION SEG_CONST
+
+ alignz 32
+ global EXTN(jconst_idct_red_sse2)
+
+EXTN(jconst_idct_red_sse2):
+
+PW_F184_MF076 times 4 dw F_1_847,-F_0_765
+PW_F256_F089 times 4 dw F_2_562, F_0_899
+PW_F106_MF217 times 4 dw F_1_061,-F_2_172
+PW_MF060_MF050 times 4 dw -F_0_601,-F_0_509
+PW_F145_MF021 times 4 dw F_1_451,-F_0_211
+PW_F362_MF127 times 4 dw F_3_624,-F_1_272
+PW_F085_MF072 times 4 dw F_0_850,-F_0_720
+PD_DESCALE_P1_4 times 4 dd 1 << (DESCALE_P1_4-1)
+PD_DESCALE_P2_4 times 4 dd 1 << (DESCALE_P2_4-1)
+PD_DESCALE_P1_2 times 4 dd 1 << (DESCALE_P1_2-1)
+PD_DESCALE_P2_2 times 4 dd 1 << (DESCALE_P2_2-1)
+PB_CENTERJSAMP times 16 db CENTERJSAMPLE
+
+ alignz 32
+
+; --------------------------------------------------------------------------
+ SECTION SEG_TEXT
+ BITS 32
+;
+; Perform dequantization and inverse DCT on one block of coefficients,
+; producing a reduced-size 4x4 output block.
+;
+; GLOBAL(void)
+; jsimd_idct_4x4_sse2 (void *dct_table, JCOEFPTR coef_block,
+; JSAMPARRAY output_buf, JDIMENSION output_col)
+;
+
+%define dct_table(b) (b)+8 ; void *dct_table
+%define coef_block(b) (b)+12 ; JCOEFPTR coef_block
+%define output_buf(b) (b)+16 ; JSAMPARRAY output_buf
+%define output_col(b) (b)+20 ; JDIMENSION output_col
+
+%define original_ebp ebp+0
+%define wk(i) ebp-(WK_NUM-(i))*SIZEOF_XMMWORD ; xmmword wk[WK_NUM]
+%define WK_NUM 2
+
+ align 32
+ global EXTN(jsimd_idct_4x4_sse2)
+
+EXTN(jsimd_idct_4x4_sse2):
+ push ebp
+ mov eax, esp ; eax = original ebp
+ sub esp, byte 4
+ and esp, byte (-SIZEOF_XMMWORD) ; align to 128 bits
+ mov [esp], eax
+ mov ebp, esp ; ebp = aligned ebp
+ lea esp, [wk(0)]
+ pushpic ebx
+; push ecx ; unused
+; push edx ; need not be preserved
+ push esi
+ push edi
+
+ get_GOT ebx ; get GOT address
+
+ ; ---- Pass 1: process columns from input.
+
+; mov eax, [original_ebp]
+ mov edx, POINTER [dct_table(eax)] ; quantptr
+ mov esi, JCOEFPTR [coef_block(eax)] ; inptr
+
+%ifndef NO_ZERO_COLUMN_TEST_4X4_SSE2
+ mov eax, DWORD [DWBLOCK(1,0,esi,SIZEOF_JCOEF)]
+ or eax, DWORD [DWBLOCK(2,0,esi,SIZEOF_JCOEF)]
+ jnz short .columnDCT
+
+ movdqa xmm0, XMMWORD [XMMBLOCK(1,0,esi,SIZEOF_JCOEF)]
+ movdqa xmm1, XMMWORD [XMMBLOCK(2,0,esi,SIZEOF_JCOEF)]
+ por xmm0, XMMWORD [XMMBLOCK(3,0,esi,SIZEOF_JCOEF)]
+ por xmm1, XMMWORD [XMMBLOCK(5,0,esi,SIZEOF_JCOEF)]
+ por xmm0, XMMWORD [XMMBLOCK(6,0,esi,SIZEOF_JCOEF)]
+ por xmm1, XMMWORD [XMMBLOCK(7,0,esi,SIZEOF_JCOEF)]
+ por xmm0, xmm1
+ packsswb xmm0, xmm0
+ packsswb xmm0, xmm0
+ movd eax, xmm0
+ test eax, eax
+ jnz short .columnDCT
+
+ ; -- AC terms all zero
+
+ movdqa xmm0, XMMWORD [XMMBLOCK(0,0,esi,SIZEOF_JCOEF)]
+ pmullw xmm0, XMMWORD [XMMBLOCK(0,0,edx,SIZEOF_ISLOW_MULT_TYPE)]
+
+ psllw xmm0, PASS1_BITS
+
+ movdqa xmm3, xmm0 ; xmm0=in0=(00 01 02 03 04 05 06 07)
+ punpcklwd xmm0, xmm0 ; xmm0=(00 00 01 01 02 02 03 03)
+ punpckhwd xmm3, xmm3 ; xmm3=(04 04 05 05 06 06 07 07)
+
+ pshufd xmm1, xmm0, 0x50 ; xmm1=[col0 col1]=(00 00 00 00 01 01 01 01)
+ pshufd xmm0, xmm0, 0xFA ; xmm0=[col2 col3]=(02 02 02 02 03 03 03 03)
+ pshufd xmm6, xmm3, 0x50 ; xmm6=[col4 col5]=(04 04 04 04 05 05 05 05)
+ pshufd xmm3, xmm3, 0xFA ; xmm3=[col6 col7]=(06 06 06 06 07 07 07 07)
+
+ jmp near .column_end
+ alignx 16, 7
+%endif
+.columnDCT:
+
+ ; -- Odd part
+
+ movdqa xmm0, XMMWORD [XMMBLOCK(1,0,esi,SIZEOF_JCOEF)]
+ movdqa xmm1, XMMWORD [XMMBLOCK(3,0,esi,SIZEOF_JCOEF)]
+ pmullw xmm0, XMMWORD [XMMBLOCK(1,0,edx,SIZEOF_ISLOW_MULT_TYPE)]
+ pmullw xmm1, XMMWORD [XMMBLOCK(3,0,edx,SIZEOF_ISLOW_MULT_TYPE)]
+ movdqa xmm2, XMMWORD [XMMBLOCK(5,0,esi,SIZEOF_JCOEF)]
+ movdqa xmm3, XMMWORD [XMMBLOCK(7,0,esi,SIZEOF_JCOEF)]
+ pmullw xmm2, XMMWORD [XMMBLOCK(5,0,edx,SIZEOF_ISLOW_MULT_TYPE)]
+ pmullw xmm3, XMMWORD [XMMBLOCK(7,0,edx,SIZEOF_ISLOW_MULT_TYPE)]
+
+ movdqa xmm4, xmm0
+ movdqa xmm5, xmm0
+ punpcklwd xmm4, xmm1
+ punpckhwd xmm5, xmm1
+ movdqa xmm0, xmm4
+ movdqa xmm1, xmm5
+ pmaddwd xmm4, [GOTOFF(ebx,PW_F256_F089)] ; xmm4=(tmp2L)
+ pmaddwd xmm5, [GOTOFF(ebx,PW_F256_F089)] ; xmm5=(tmp2H)
+ pmaddwd xmm0, [GOTOFF(ebx,PW_F106_MF217)] ; xmm0=(tmp0L)
+ pmaddwd xmm1, [GOTOFF(ebx,PW_F106_MF217)] ; xmm1=(tmp0H)
+
+ movdqa xmm6, xmm2
+ movdqa xmm7, xmm2
+ punpcklwd xmm6, xmm3
+ punpckhwd xmm7, xmm3
+ movdqa xmm2, xmm6
+ movdqa xmm3, xmm7
+ pmaddwd xmm6, [GOTOFF(ebx,PW_MF060_MF050)] ; xmm6=(tmp2L)
+ pmaddwd xmm7, [GOTOFF(ebx,PW_MF060_MF050)] ; xmm7=(tmp2H)
+ pmaddwd xmm2, [GOTOFF(ebx,PW_F145_MF021)] ; xmm2=(tmp0L)
+ pmaddwd xmm3, [GOTOFF(ebx,PW_F145_MF021)] ; xmm3=(tmp0H)
+
+ paddd xmm6, xmm4 ; xmm6=tmp2L
+ paddd xmm7, xmm5 ; xmm7=tmp2H
+ paddd xmm2, xmm0 ; xmm2=tmp0L
+ paddd xmm3, xmm1 ; xmm3=tmp0H
+
+ movdqa XMMWORD [wk(0)], xmm2 ; wk(0)=tmp0L
+ movdqa XMMWORD [wk(1)], xmm3 ; wk(1)=tmp0H
+
+ ; -- Even part
+
+ movdqa xmm4, XMMWORD [XMMBLOCK(0,0,esi,SIZEOF_JCOEF)]
+ movdqa xmm5, XMMWORD [XMMBLOCK(2,0,esi,SIZEOF_JCOEF)]
+ movdqa xmm0, XMMWORD [XMMBLOCK(6,0,esi,SIZEOF_JCOEF)]
+ pmullw xmm4, XMMWORD [XMMBLOCK(0,0,edx,SIZEOF_ISLOW_MULT_TYPE)]
+ pmullw xmm5, XMMWORD [XMMBLOCK(2,0,edx,SIZEOF_ISLOW_MULT_TYPE)]
+ pmullw xmm0, XMMWORD [XMMBLOCK(6,0,edx,SIZEOF_ISLOW_MULT_TYPE)]
+
+ pxor xmm1, xmm1
+ pxor xmm2, xmm2
+ punpcklwd xmm1, xmm4 ; xmm1=tmp0L
+ punpckhwd xmm2, xmm4 ; xmm2=tmp0H
+ psrad xmm1, (16-CONST_BITS-1) ; psrad xmm1,16 & pslld xmm1,CONST_BITS+1
+ psrad xmm2, (16-CONST_BITS-1) ; psrad xmm2,16 & pslld xmm2,CONST_BITS+1
+
+ movdqa xmm3, xmm5 ; xmm5=in2=z2
+ punpcklwd xmm5, xmm0 ; xmm0=in6=z3
+ punpckhwd xmm3, xmm0
+ pmaddwd xmm5, [GOTOFF(ebx,PW_F184_MF076)] ; xmm5=tmp2L
+ pmaddwd xmm3, [GOTOFF(ebx,PW_F184_MF076)] ; xmm3=tmp2H
+
+ movdqa xmm4, xmm1
+ movdqa xmm0, xmm2
+ paddd xmm1, xmm5 ; xmm1=tmp10L
+ paddd xmm2, xmm3 ; xmm2=tmp10H
+ psubd xmm4, xmm5 ; xmm4=tmp12L
+ psubd xmm0, xmm3 ; xmm0=tmp12H
+
+ ; -- Final output stage
+
+ movdqa xmm5, xmm1
+ movdqa xmm3, xmm2
+ paddd xmm1, xmm6 ; xmm1=data0L
+ paddd xmm2, xmm7 ; xmm2=data0H
+ psubd xmm5, xmm6 ; xmm5=data3L
+ psubd xmm3, xmm7 ; xmm3=data3H
+
+ movdqa xmm6, [GOTOFF(ebx,PD_DESCALE_P1_4)] ; xmm6=[PD_DESCALE_P1_4]
+
+ paddd xmm1, xmm6
+ paddd xmm2, xmm6
+ psrad xmm1, DESCALE_P1_4
+ psrad xmm2, DESCALE_P1_4
+ paddd xmm5, xmm6
+ paddd xmm3, xmm6
+ psrad xmm5, DESCALE_P1_4
+ psrad xmm3, DESCALE_P1_4
+
+ packssdw xmm1, xmm2 ; xmm1=data0=(00 01 02 03 04 05 06 07)
+ packssdw xmm5, xmm3 ; xmm5=data3=(30 31 32 33 34 35 36 37)
+
+ movdqa xmm7, XMMWORD [wk(0)] ; xmm7=tmp0L
+ movdqa xmm6, XMMWORD [wk(1)] ; xmm6=tmp0H
+
+ movdqa xmm2, xmm4
+ movdqa xmm3, xmm0
+ paddd xmm4, xmm7 ; xmm4=data1L
+ paddd xmm0, xmm6 ; xmm0=data1H
+ psubd xmm2, xmm7 ; xmm2=data2L
+ psubd xmm3, xmm6 ; xmm3=data2H
+
+ movdqa xmm7, [GOTOFF(ebx,PD_DESCALE_P1_4)] ; xmm7=[PD_DESCALE_P1_4]
+
+ paddd xmm4, xmm7
+ paddd xmm0, xmm7
+ psrad xmm4, DESCALE_P1_4
+ psrad xmm0, DESCALE_P1_4
+ paddd xmm2, xmm7
+ paddd xmm3, xmm7
+ psrad xmm2, DESCALE_P1_4
+ psrad xmm3, DESCALE_P1_4
+
+ packssdw xmm4, xmm0 ; xmm4=data1=(10 11 12 13 14 15 16 17)
+ packssdw xmm2, xmm3 ; xmm2=data2=(20 21 22 23 24 25 26 27)
+
+ movdqa xmm6, xmm1 ; transpose coefficients(phase 1)
+ punpcklwd xmm1, xmm4 ; xmm1=(00 10 01 11 02 12 03 13)
+ punpckhwd xmm6, xmm4 ; xmm6=(04 14 05 15 06 16 07 17)
+ movdqa xmm7, xmm2 ; transpose coefficients(phase 1)
+ punpcklwd xmm2, xmm5 ; xmm2=(20 30 21 31 22 32 23 33)
+ punpckhwd xmm7, xmm5 ; xmm7=(24 34 25 35 26 36 27 37)
+
+ movdqa xmm0, xmm1 ; transpose coefficients(phase 2)
+ punpckldq xmm1, xmm2 ; xmm1=[col0 col1]=(00 10 20 30 01 11 21 31)
+ punpckhdq xmm0, xmm2 ; xmm0=[col2 col3]=(02 12 22 32 03 13 23 33)
+ movdqa xmm3, xmm6 ; transpose coefficients(phase 2)
+ punpckldq xmm6, xmm7 ; xmm6=[col4 col5]=(04 14 24 34 05 15 25 35)
+ punpckhdq xmm3, xmm7 ; xmm3=[col6 col7]=(06 16 26 36 07 17 27 37)
+.column_end:
+
+ ; -- Prefetch the next coefficient block
+
+ prefetchnta [esi + DCTSIZE2*SIZEOF_JCOEF + 0*32]
+ prefetchnta [esi + DCTSIZE2*SIZEOF_JCOEF + 1*32]
+ prefetchnta [esi + DCTSIZE2*SIZEOF_JCOEF + 2*32]
+ prefetchnta [esi + DCTSIZE2*SIZEOF_JCOEF + 3*32]
+
+ ; ---- Pass 2: process rows, store into output array.
+
+ mov eax, [original_ebp]
+ mov edi, JSAMPARRAY [output_buf(eax)] ; (JSAMPROW *)
+ mov eax, JDIMENSION [output_col(eax)]
+
+ ; -- Even part
+
+ pxor xmm4, xmm4
+ punpcklwd xmm4, xmm1 ; xmm4=tmp0
+ psrad xmm4, (16-CONST_BITS-1) ; psrad xmm4,16 & pslld xmm4,CONST_BITS+1
+
+ ; -- Odd part
+
+ punpckhwd xmm1, xmm0
+ punpckhwd xmm6, xmm3
+ movdqa xmm5, xmm1
+ movdqa xmm2, xmm6
+ pmaddwd xmm1, [GOTOFF(ebx,PW_F256_F089)] ; xmm1=(tmp2)
+ pmaddwd xmm6, [GOTOFF(ebx,PW_MF060_MF050)] ; xmm6=(tmp2)
+ pmaddwd xmm5, [GOTOFF(ebx,PW_F106_MF217)] ; xmm5=(tmp0)
+ pmaddwd xmm2, [GOTOFF(ebx,PW_F145_MF021)] ; xmm2=(tmp0)
+
+ paddd xmm6, xmm1 ; xmm6=tmp2
+ paddd xmm2, xmm5 ; xmm2=tmp0
+
+ ; -- Even part
+
+ punpcklwd xmm0, xmm3
+ pmaddwd xmm0, [GOTOFF(ebx,PW_F184_MF076)] ; xmm0=tmp2
+
+ movdqa xmm7, xmm4
+ paddd xmm4, xmm0 ; xmm4=tmp10
+ psubd xmm7, xmm0 ; xmm7=tmp12
+
+ ; -- Final output stage
+
+ movdqa xmm1, [GOTOFF(ebx,PD_DESCALE_P2_4)] ; xmm1=[PD_DESCALE_P2_4]
+
+ movdqa xmm5, xmm4
+ movdqa xmm3, xmm7
+ paddd xmm4, xmm6 ; xmm4=data0=(00 10 20 30)
+ paddd xmm7, xmm2 ; xmm7=data1=(01 11 21 31)
+ psubd xmm5, xmm6 ; xmm5=data3=(03 13 23 33)
+ psubd xmm3, xmm2 ; xmm3=data2=(02 12 22 32)
+
+ paddd xmm4, xmm1
+ paddd xmm7, xmm1
+ psrad xmm4, DESCALE_P2_4
+ psrad xmm7, DESCALE_P2_4
+ paddd xmm5, xmm1
+ paddd xmm3, xmm1
+ psrad xmm5, DESCALE_P2_4
+ psrad xmm3, DESCALE_P2_4
+
+ packssdw xmm4, xmm3 ; xmm4=(00 10 20 30 02 12 22 32)
+ packssdw xmm7, xmm5 ; xmm7=(01 11 21 31 03 13 23 33)
+
+ movdqa xmm0, xmm4 ; transpose coefficients(phase 1)
+ punpcklwd xmm4, xmm7 ; xmm4=(00 01 10 11 20 21 30 31)
+ punpckhwd xmm0, xmm7 ; xmm0=(02 03 12 13 22 23 32 33)
+
+ movdqa xmm6, xmm4 ; transpose coefficients(phase 2)
+ punpckldq xmm4, xmm0 ; xmm4=(00 01 02 03 10 11 12 13)
+ punpckhdq xmm6, xmm0 ; xmm6=(20 21 22 23 30 31 32 33)
+
+ packsswb xmm4, xmm6 ; xmm4=(00 01 02 03 10 11 12 13 20 ..)
+ paddb xmm4, [GOTOFF(ebx,PB_CENTERJSAMP)]
+
+ pshufd xmm2, xmm4, 0x39 ; xmm2=(10 11 12 13 20 21 22 23 30 ..)
+ pshufd xmm1, xmm4, 0x4E ; xmm1=(20 21 22 23 30 31 32 33 00 ..)
+ pshufd xmm3, xmm4, 0x93 ; xmm3=(30 31 32 33 00 01 02 03 10 ..)
+
+ mov edx, JSAMPROW [edi+0*SIZEOF_JSAMPROW]
+ mov esi, JSAMPROW [edi+1*SIZEOF_JSAMPROW]
+ movd XMM_DWORD [edx+eax*SIZEOF_JSAMPLE], xmm4
+ movd XMM_DWORD [esi+eax*SIZEOF_JSAMPLE], xmm2
+ mov edx, JSAMPROW [edi+2*SIZEOF_JSAMPROW]
+ mov esi, JSAMPROW [edi+3*SIZEOF_JSAMPROW]
+ movd XMM_DWORD [edx+eax*SIZEOF_JSAMPLE], xmm1
+ movd XMM_DWORD [esi+eax*SIZEOF_JSAMPLE], xmm3
+
+ pop edi
+ pop esi
+; pop edx ; need not be preserved
+; pop ecx ; unused
+ poppic ebx
+ mov esp, ebp ; esp <- aligned ebp
+ pop esp ; esp <- original ebp
+ pop ebp
+ ret
+
+
+; --------------------------------------------------------------------------
+;
+; Perform dequantization and inverse DCT on one block of coefficients,
+; producing a reduced-size 2x2 output block.
+;
+; GLOBAL(void)
+; jsimd_idct_2x2_sse2 (void *dct_table, JCOEFPTR coef_block,
+; JSAMPARRAY output_buf, JDIMENSION output_col)
+;
+
+%define dct_table(b) (b)+8 ; void *dct_table
+%define coef_block(b) (b)+12 ; JCOEFPTR coef_block
+%define output_buf(b) (b)+16 ; JSAMPARRAY output_buf
+%define output_col(b) (b)+20 ; JDIMENSION output_col
+
+ align 32
+ global EXTN(jsimd_idct_2x2_sse2)
+
+EXTN(jsimd_idct_2x2_sse2):
+ push ebp
+ mov ebp, esp
+ push ebx
+; push ecx ; need not be preserved
+; push edx ; need not be preserved
+ push esi
+ push edi
+
+ get_GOT ebx ; get GOT address
+
+ ; ---- Pass 1: process columns from input.
+
+ mov edx, POINTER [dct_table(ebp)] ; quantptr
+ mov esi, JCOEFPTR [coef_block(ebp)] ; inptr
+
+ ; | input: | result: |
+ ; | 00 01 ** 03 ** 05 ** 07 | |
+ ; | 10 11 ** 13 ** 15 ** 17 | |
+ ; | ** ** ** ** ** ** ** ** | |
+ ; | 30 31 ** 33 ** 35 ** 37 | A0 A1 A3 A5 A7 |
+ ; | ** ** ** ** ** ** ** ** | B0 B1 B3 B5 B7 |
+ ; | 50 51 ** 53 ** 55 ** 57 | |
+ ; | ** ** ** ** ** ** ** ** | |
+ ; | 70 71 ** 73 ** 75 ** 77 | |
+
+ ; -- Odd part
+
+ movdqa xmm0, XMMWORD [XMMBLOCK(1,0,esi,SIZEOF_JCOEF)]
+ movdqa xmm1, XMMWORD [XMMBLOCK(3,0,esi,SIZEOF_JCOEF)]
+ pmullw xmm0, XMMWORD [XMMBLOCK(1,0,edx,SIZEOF_ISLOW_MULT_TYPE)]
+ pmullw xmm1, XMMWORD [XMMBLOCK(3,0,edx,SIZEOF_ISLOW_MULT_TYPE)]
+ movdqa xmm2, XMMWORD [XMMBLOCK(5,0,esi,SIZEOF_JCOEF)]
+ movdqa xmm3, XMMWORD [XMMBLOCK(7,0,esi,SIZEOF_JCOEF)]
+ pmullw xmm2, XMMWORD [XMMBLOCK(5,0,edx,SIZEOF_ISLOW_MULT_TYPE)]
+ pmullw xmm3, XMMWORD [XMMBLOCK(7,0,edx,SIZEOF_ISLOW_MULT_TYPE)]
+
+ ; xmm0=(10 11 ** 13 ** 15 ** 17), xmm1=(30 31 ** 33 ** 35 ** 37)
+ ; xmm2=(50 51 ** 53 ** 55 ** 57), xmm3=(70 71 ** 73 ** 75 ** 77)
+
+ pcmpeqd xmm7, xmm7
+ pslld xmm7, WORD_BIT ; xmm7={0x0000 0xFFFF 0x0000 0xFFFF ..}
+
+ movdqa xmm4, xmm0 ; xmm4=(10 11 ** 13 ** 15 ** 17)
+ movdqa xmm5, xmm2 ; xmm5=(50 51 ** 53 ** 55 ** 57)
+ punpcklwd xmm4, xmm1 ; xmm4=(10 30 11 31 ** ** 13 33)
+ punpcklwd xmm5, xmm3 ; xmm5=(50 70 51 71 ** ** 53 73)
+ pmaddwd xmm4, [GOTOFF(ebx,PW_F362_MF127)]
+ pmaddwd xmm5, [GOTOFF(ebx,PW_F085_MF072)]
+
+ psrld xmm0, WORD_BIT ; xmm0=(11 -- 13 -- 15 -- 17 --)
+ pand xmm1, xmm7 ; xmm1=(-- 31 -- 33 -- 35 -- 37)
+ psrld xmm2, WORD_BIT ; xmm2=(51 -- 53 -- 55 -- 57 --)
+ pand xmm3, xmm7 ; xmm3=(-- 71 -- 73 -- 75 -- 77)
+ por xmm0, xmm1 ; xmm0=(11 31 13 33 15 35 17 37)
+ por xmm2, xmm3 ; xmm2=(51 71 53 73 55 75 57 77)
+ pmaddwd xmm0, [GOTOFF(ebx,PW_F362_MF127)]
+ pmaddwd xmm2, [GOTOFF(ebx,PW_F085_MF072)]
+
+ paddd xmm4, xmm5 ; xmm4=tmp0[col0 col1 **** col3]
+ paddd xmm0, xmm2 ; xmm0=tmp0[col1 col3 col5 col7]
+
+ ; -- Even part
+
+ movdqa xmm6, XMMWORD [XMMBLOCK(0,0,esi,SIZEOF_JCOEF)]
+ pmullw xmm6, XMMWORD [XMMBLOCK(0,0,edx,SIZEOF_ISLOW_MULT_TYPE)]
+
+ ; xmm6=(00 01 ** 03 ** 05 ** 07)
+
+ movdqa xmm1, xmm6 ; xmm1=(00 01 ** 03 ** 05 ** 07)
+ pslld xmm6, WORD_BIT ; xmm6=(-- 00 -- ** -- ** -- **)
+ pand xmm1, xmm7 ; xmm1=(-- 01 -- 03 -- 05 -- 07)
+ psrad xmm6, (WORD_BIT-CONST_BITS-2) ; xmm6=tmp10[col0 **** **** ****]
+ psrad xmm1, (WORD_BIT-CONST_BITS-2) ; xmm1=tmp10[col1 col3 col5 col7]
+
+ ; -- Final output stage
+
+ movdqa xmm3, xmm6
+ movdqa xmm5, xmm1
+ paddd xmm6, xmm4 ; xmm6=data0[col0 **** **** ****]=(A0 ** ** **)
+ paddd xmm1, xmm0 ; xmm1=data0[col1 col3 col5 col7]=(A1 A3 A5 A7)
+ psubd xmm3, xmm4 ; xmm3=data1[col0 **** **** ****]=(B0 ** ** **)
+ psubd xmm5, xmm0 ; xmm5=data1[col1 col3 col5 col7]=(B1 B3 B5 B7)
+
+ movdqa xmm2, [GOTOFF(ebx,PD_DESCALE_P1_2)] ; xmm2=[PD_DESCALE_P1_2]
+
+ punpckldq xmm6, xmm3 ; xmm6=(A0 B0 ** **)
+
+ movdqa xmm7, xmm1
+ punpcklqdq xmm1, xmm5 ; xmm1=(A1 A3 B1 B3)
+ punpckhqdq xmm7, xmm5 ; xmm7=(A5 A7 B5 B7)
+
+ paddd xmm6, xmm2
+ psrad xmm6, DESCALE_P1_2
+
+ paddd xmm1, xmm2
+ paddd xmm7, xmm2
+ psrad xmm1, DESCALE_P1_2
+ psrad xmm7, DESCALE_P1_2
+
+ ; -- Prefetch the next coefficient block
+
+ prefetchnta [esi + DCTSIZE2*SIZEOF_JCOEF + 0*32]
+ prefetchnta [esi + DCTSIZE2*SIZEOF_JCOEF + 1*32]
+ prefetchnta [esi + DCTSIZE2*SIZEOF_JCOEF + 2*32]
+ prefetchnta [esi + DCTSIZE2*SIZEOF_JCOEF + 3*32]
+
+ ; ---- Pass 2: process rows, store into output array.
+
+ mov edi, JSAMPARRAY [output_buf(ebp)] ; (JSAMPROW *)
+ mov eax, JDIMENSION [output_col(ebp)]
+
+ ; | input:| result:|
+ ; | A0 B0 | |
+ ; | A1 B1 | C0 C1 |
+ ; | A3 B3 | D0 D1 |
+ ; | A5 B5 | |
+ ; | A7 B7 | |
+
+ ; -- Odd part
+
+ packssdw xmm1, xmm1 ; xmm1=(A1 A3 B1 B3 A1 A3 B1 B3)
+ packssdw xmm7, xmm7 ; xmm7=(A5 A7 B5 B7 A5 A7 B5 B7)
+ pmaddwd xmm1, [GOTOFF(ebx,PW_F362_MF127)]
+ pmaddwd xmm7, [GOTOFF(ebx,PW_F085_MF072)]
+
+ paddd xmm1, xmm7 ; xmm1=tmp0[row0 row1 row0 row1]
+
+ ; -- Even part
+
+ pslld xmm6, (CONST_BITS+2) ; xmm6=tmp10[row0 row1 **** ****]
+
+ ; -- Final output stage
+
+ movdqa xmm4, xmm6
+ paddd xmm6, xmm1 ; xmm6=data0[row0 row1 **** ****]=(C0 C1 ** **)
+ psubd xmm4, xmm1 ; xmm4=data1[row0 row1 **** ****]=(D0 D1 ** **)
+
+ punpckldq xmm6, xmm4 ; xmm6=(C0 D0 C1 D1)
+
+ paddd xmm6, [GOTOFF(ebx,PD_DESCALE_P2_2)]
+ psrad xmm6, DESCALE_P2_2
+
+ packssdw xmm6, xmm6 ; xmm6=(C0 D0 C1 D1 C0 D0 C1 D1)
+ packsswb xmm6, xmm6 ; xmm6=(C0 D0 C1 D1 C0 D0 C1 D1 ..)
+ paddb xmm6, [GOTOFF(ebx,PB_CENTERJSAMP)]
+
+ pextrw ebx, xmm6, 0x00 ; ebx=(C0 D0 -- --)
+ pextrw ecx, xmm6, 0x01 ; ecx=(C1 D1 -- --)
+
+ mov edx, JSAMPROW [edi+0*SIZEOF_JSAMPROW]
+ mov esi, JSAMPROW [edi+1*SIZEOF_JSAMPROW]
+ mov WORD [edx+eax*SIZEOF_JSAMPLE], bx
+ mov WORD [esi+eax*SIZEOF_JSAMPLE], cx
+
+ pop edi
+ pop esi
+; pop edx ; need not be preserved
+; pop ecx ; need not be preserved
+ pop ebx
+ pop ebp
+ ret
+
+; For some reason, the OS X linker does not honor the request to align the
+; segment unless we do this.
+ align 32
diff --git a/simd/jquant-3dn.asm b/simd/i386/jquant-3dn.asm
similarity index 98%
rename from simd/jquant-3dn.asm
rename to simd/i386/jquant-3dn.asm
index 0b4164b..73c02cd 100644
--- a/simd/jquant-3dn.asm
+++ b/simd/i386/jquant-3dn.asm
@@ -2,6 +2,7 @@
; jquant.asm - sample data conversion and quantization (3DNow! & MMX)
;
; Copyright 2009 Pierre Ossman <ossman@cendio.se> for Cendio AB
+; Copyright (C) 2016, D. R. Commander.
;
; Based on the x86 SIMD extension for IJG JPEG library
; Copyright (C) 1999-2006, MIYASAKA Masaru.
@@ -33,7 +34,7 @@
%define start_col ebp+12 ; JDIMENSION start_col
%define workspace ebp+16 ; FAST_FLOAT *workspace
- align 16
+ align 32
global EXTN(jsimd_convsamp_float_3dnow)
EXTN(jsimd_convsamp_float_3dnow):
@@ -136,7 +137,7 @@
%define divisors ebp+12 ; FAST_FLOAT *divisors
%define workspace ebp+16 ; FAST_FLOAT *workspace
- align 16
+ align 32
global EXTN(jsimd_quantize_float_3dnow)
EXTN(jsimd_quantize_float_3dnow):
@@ -229,4 +230,4 @@
; For some reason, the OS X linker does not honor the request to align the
; segment unless we do this.
- align 16
+ align 32
diff --git a/simd/jquant-mmx.asm b/simd/i386/jquant-mmx.asm
similarity index 98%
rename from simd/jquant-mmx.asm
rename to simd/i386/jquant-mmx.asm
index aed6071..93dea31 100644
--- a/simd/jquant-mmx.asm
+++ b/simd/i386/jquant-mmx.asm
@@ -2,6 +2,7 @@
; jquant.asm - sample data conversion and quantization (MMX)
;
; Copyright 2009 Pierre Ossman <ossman@cendio.se> for Cendio AB
+; Copyright (C) 2016, D. R. Commander.
;
; Based on the x86 SIMD extension for IJG JPEG library
; Copyright (C) 1999-2006, MIYASAKA Masaru.
@@ -33,7 +34,7 @@
%define start_col ebp+12 ; JDIMENSION start_col
%define workspace ebp+16 ; DCTELEM *workspace
- align 16
+ align 32
global EXTN(jsimd_convsamp_mmx)
EXTN(jsimd_convsamp_mmx):
@@ -138,7 +139,7 @@
%define divisors ebp+12 ; DCTELEM *divisors
%define workspace ebp+16 ; DCTELEM *workspace
- align 16
+ align 32
global EXTN(jsimd_quantize_mmx)
EXTN(jsimd_quantize_mmx):
@@ -270,4 +271,4 @@
; For some reason, the OS X linker does not honor the request to align the
; segment unless we do this.
- align 16
+ align 32
diff --git a/simd/jquant-sse.asm b/simd/i386/jquant-sse.asm
similarity index 98%
rename from simd/jquant-sse.asm
rename to simd/i386/jquant-sse.asm
index 1baf88f..d6ab943 100644
--- a/simd/jquant-sse.asm
+++ b/simd/i386/jquant-sse.asm
@@ -2,6 +2,7 @@
; jquant.asm - sample data conversion and quantization (SSE & MMX)
;
; Copyright 2009 Pierre Ossman <ossman@cendio.se> for Cendio AB
+; Copyright (C) 2016, D. R. Commander.
;
; Based on the x86 SIMD extension for IJG JPEG library
; Copyright (C) 1999-2006, MIYASAKA Masaru.
@@ -33,7 +34,7 @@
%define start_col ebp+12 ; JDIMENSION start_col
%define workspace ebp+16 ; FAST_FLOAT *workspace
- align 16
+ align 32
global EXTN(jsimd_convsamp_float_sse)
EXTN(jsimd_convsamp_float_sse):
@@ -136,7 +137,7 @@
%define divisors ebp+12 ; FAST_FLOAT *divisors
%define workspace ebp+16 ; FAST_FLOAT *workspace
- align 16
+ align 32
global EXTN(jsimd_quantize_float_sse)
EXTN(jsimd_quantize_float_sse):
@@ -207,4 +208,4 @@
; For some reason, the OS X linker does not honor the request to align the
; segment unless we do this.
- align 16
+ align 32
diff --git a/simd/i386/jquantf-sse2.asm b/simd/i386/jquantf-sse2.asm
new file mode 100644
index 0000000..3c0000e
--- /dev/null
+++ b/simd/i386/jquantf-sse2.asm
@@ -0,0 +1,171 @@
+;
+; jquantf.asm - sample data conversion and quantization (SSE & SSE2)
+;
+; Copyright 2009 Pierre Ossman <ossman@cendio.se> for Cendio AB
+; Copyright (C) 2016, D. R. Commander.
+;
+; Based on the x86 SIMD extension for IJG JPEG library
+; Copyright (C) 1999-2006, MIYASAKA Masaru.
+; For conditions of distribution and use, see copyright notice in jsimdext.inc
+;
+; This file should be assembled with NASM (Netwide Assembler),
+; can *not* be assembled with Microsoft's MASM or any compatible
+; assembler (including Borland's Turbo Assembler).
+; NASM is available from http://nasm.sourceforge.net/ or
+; http://sourceforge.net/project/showfiles.php?group_id=6208
+;
+; [TAB8]
+
+%include "jsimdext.inc"
+%include "jdct.inc"
+
+; --------------------------------------------------------------------------
+ SECTION SEG_TEXT
+ BITS 32
+;
+; Load data into workspace, applying unsigned->signed conversion
+;
+; GLOBAL(void)
+; jsimd_convsamp_float_sse2 (JSAMPARRAY sample_data, JDIMENSION start_col,
+; FAST_FLOAT *workspace);
+;
+
+%define sample_data ebp+8 ; JSAMPARRAY sample_data
+%define start_col ebp+12 ; JDIMENSION start_col
+%define workspace ebp+16 ; FAST_FLOAT *workspace
+
+ align 32
+ global EXTN(jsimd_convsamp_float_sse2)
+
+EXTN(jsimd_convsamp_float_sse2):
+ push ebp
+ mov ebp, esp
+ push ebx
+; push ecx ; need not be preserved
+; push edx ; need not be preserved
+ push esi
+ push edi
+
+ pcmpeqw xmm7, xmm7
+ psllw xmm7, 7
+ packsswb xmm7, xmm7 ; xmm7 = PB_CENTERJSAMPLE (0x808080..)
+
+ mov esi, JSAMPARRAY [sample_data] ; (JSAMPROW *)
+ mov eax, JDIMENSION [start_col]
+ mov edi, POINTER [workspace] ; (DCTELEM *)
+ mov ecx, DCTSIZE/2
+ alignx 16, 7
+.convloop:
+ mov ebx, JSAMPROW [esi+0*SIZEOF_JSAMPROW] ; (JSAMPLE *)
+ mov edx, JSAMPROW [esi+1*SIZEOF_JSAMPROW] ; (JSAMPLE *)
+
+ movq xmm0, XMM_MMWORD [ebx+eax*SIZEOF_JSAMPLE]
+ movq xmm1, XMM_MMWORD [edx+eax*SIZEOF_JSAMPLE]
+
+ psubb xmm0, xmm7 ; xmm0=(01234567)
+ psubb xmm1, xmm7 ; xmm1=(89ABCDEF)
+
+ punpcklbw xmm0, xmm0 ; xmm0=(*0*1*2*3*4*5*6*7)
+ punpcklbw xmm1, xmm1 ; xmm1=(*8*9*A*B*C*D*E*F)
+
+ punpcklwd xmm2, xmm0 ; xmm2=(***0***1***2***3)
+ punpckhwd xmm0, xmm0 ; xmm0=(***4***5***6***7)
+ punpcklwd xmm3, xmm1 ; xmm3=(***8***9***A***B)
+ punpckhwd xmm1, xmm1 ; xmm1=(***C***D***E***F)
+
+ psrad xmm2, (DWORD_BIT-BYTE_BIT) ; xmm2=(0123)
+ psrad xmm0, (DWORD_BIT-BYTE_BIT) ; xmm0=(4567)
+ cvtdq2ps xmm2, xmm2 ; xmm2=(0123)
+ cvtdq2ps xmm0, xmm0 ; xmm0=(4567)
+ psrad xmm3, (DWORD_BIT-BYTE_BIT) ; xmm3=(89AB)
+ psrad xmm1, (DWORD_BIT-BYTE_BIT) ; xmm1=(CDEF)
+ cvtdq2ps xmm3, xmm3 ; xmm3=(89AB)
+ cvtdq2ps xmm1, xmm1 ; xmm1=(CDEF)
+
+ movaps XMMWORD [XMMBLOCK(0,0,edi,SIZEOF_FAST_FLOAT)], xmm2
+ movaps XMMWORD [XMMBLOCK(0,1,edi,SIZEOF_FAST_FLOAT)], xmm0
+ movaps XMMWORD [XMMBLOCK(1,0,edi,SIZEOF_FAST_FLOAT)], xmm3
+ movaps XMMWORD [XMMBLOCK(1,1,edi,SIZEOF_FAST_FLOAT)], xmm1
+
+ add esi, byte 2*SIZEOF_JSAMPROW
+ add edi, byte 2*DCTSIZE*SIZEOF_FAST_FLOAT
+ dec ecx
+ jnz short .convloop
+
+ pop edi
+ pop esi
+; pop edx ; need not be preserved
+; pop ecx ; need not be preserved
+ pop ebx
+ pop ebp
+ ret
+
+
+; --------------------------------------------------------------------------
+;
+; Quantize/descale the coefficients, and store into coef_block
+;
+; GLOBAL(void)
+; jsimd_quantize_float_sse2 (JCOEFPTR coef_block, FAST_FLOAT *divisors,
+; FAST_FLOAT *workspace);
+;
+
+%define coef_block ebp+8 ; JCOEFPTR coef_block
+%define divisors ebp+12 ; FAST_FLOAT *divisors
+%define workspace ebp+16 ; FAST_FLOAT *workspace
+
+ align 32
+ global EXTN(jsimd_quantize_float_sse2)
+
+EXTN(jsimd_quantize_float_sse2):
+ push ebp
+ mov ebp, esp
+; push ebx ; unused
+; push ecx ; unused
+; push edx ; need not be preserved
+ push esi
+ push edi
+
+ mov esi, POINTER [workspace]
+ mov edx, POINTER [divisors]
+ mov edi, JCOEFPTR [coef_block]
+ mov eax, DCTSIZE2/16
+ alignx 16, 7
+.quantloop:
+ movaps xmm0, XMMWORD [XMMBLOCK(0,0,esi,SIZEOF_FAST_FLOAT)]
+ movaps xmm1, XMMWORD [XMMBLOCK(0,1,esi,SIZEOF_FAST_FLOAT)]
+ mulps xmm0, XMMWORD [XMMBLOCK(0,0,edx,SIZEOF_FAST_FLOAT)]
+ mulps xmm1, XMMWORD [XMMBLOCK(0,1,edx,SIZEOF_FAST_FLOAT)]
+ movaps xmm2, XMMWORD [XMMBLOCK(1,0,esi,SIZEOF_FAST_FLOAT)]
+ movaps xmm3, XMMWORD [XMMBLOCK(1,1,esi,SIZEOF_FAST_FLOAT)]
+ mulps xmm2, XMMWORD [XMMBLOCK(1,0,edx,SIZEOF_FAST_FLOAT)]
+ mulps xmm3, XMMWORD [XMMBLOCK(1,1,edx,SIZEOF_FAST_FLOAT)]
+
+ cvtps2dq xmm0, xmm0
+ cvtps2dq xmm1, xmm1
+ cvtps2dq xmm2, xmm2
+ cvtps2dq xmm3, xmm3
+
+ packssdw xmm0, xmm1
+ packssdw xmm2, xmm3
+
+ movdqa XMMWORD [XMMBLOCK(0,0,edi,SIZEOF_JCOEF)], xmm0
+ movdqa XMMWORD [XMMBLOCK(1,0,edi,SIZEOF_JCOEF)], xmm2
+
+ add esi, byte 16*SIZEOF_FAST_FLOAT
+ add edx, byte 16*SIZEOF_FAST_FLOAT
+ add edi, byte 16*SIZEOF_JCOEF
+ dec eax
+ jnz short .quantloop
+
+ pop edi
+ pop esi
+; pop edx ; need not be preserved
+; pop ecx ; unused
+; pop ebx ; unused
+ pop ebp
+ ret
+
+; For some reason, the OS X linker does not honor the request to align the
+; segment unless we do this.
+ align 32
diff --git a/simd/i386/jquanti-avx2.asm b/simd/i386/jquanti-avx2.asm
new file mode 100644
index 0000000..0356185
--- /dev/null
+++ b/simd/i386/jquanti-avx2.asm
@@ -0,0 +1,107 @@
+;
+; jquanti.asm - sample quantization (AVX2)
+;
+; Copyright 2009 Pierre Ossman <ossman@cendio.se> for Cendio AB
+; Copyright (C) 2016, D. R. Commander.
+; Copyright (C) 2016, Matthieu Darbois.
+;
+; Based on the x86 SIMD extension for IJG JPEG library
+; Copyright (C) 1999-2006, MIYASAKA Masaru.
+; For conditions of distribution and use, see copyright notice in jsimdext.inc
+;
+; This file should be assembled with NASM (Netwide Assembler),
+; can *not* be assembled with Microsoft's MASM or any compatible
+; assembler (including Borland's Turbo Assembler).
+; NASM is available from http://nasm.sourceforge.net/ or
+; http://sourceforge.net/project/showfiles.php?group_id=6208
+;
+; [TAB8]
+
+%include "jsimdext.inc"
+%include "jdct.inc"
+
+; --------------------------------------------------------------------------
+ SECTION SEG_TEXT
+ BITS 32
+
+; --------------------------------------------------------------------------
+;
+; Quantize/descale the coefficients, and store into coef_block
+;
+; This implementation is based on an algorithm described in
+; "How to optimize for the Pentium family of microprocessors"
+; (http://www.agner.org/assem/).
+;
+; GLOBAL(void)
+; jsimd_quantize_avx2 (JCOEFPTR coef_block, DCTELEM *divisors,
+; DCTELEM *workspace);
+;
+
+%define RECIPROCAL(m,n,b) XMMBLOCK(DCTSIZE*0+(m),(n),(b),SIZEOF_DCTELEM)
+%define CORRECTION(m,n,b) XMMBLOCK(DCTSIZE*1+(m),(n),(b),SIZEOF_DCTELEM)
+%define SCALE(m,n,b) XMMBLOCK(DCTSIZE*2+(m),(n),(b),SIZEOF_DCTELEM)
+
+%define coef_block ebp+8 ; JCOEFPTR coef_block
+%define divisors ebp+12 ; DCTELEM *divisors
+%define workspace ebp+16 ; DCTELEM *workspace
+
+ align 32
+ global EXTN(jsimd_quantize_avx2)
+
+EXTN(jsimd_quantize_avx2):
+ push ebp
+ mov ebp, esp
+; push ebx ; unused
+; push ecx ; unused
+; push edx ; need not be preserved
+ push esi
+ push edi
+
+ mov esi, POINTER [workspace]
+ mov edx, POINTER [divisors]
+ mov edi, JCOEFPTR [coef_block]
+
+ vmovdqu ymm4, [XMMBLOCK(0,0,esi,SIZEOF_DCTELEM)]
+ vmovdqu ymm5, [XMMBLOCK(2,0,esi,SIZEOF_DCTELEM)]
+ vmovdqu ymm6, [XMMBLOCK(4,0,esi,SIZEOF_DCTELEM)]
+ vmovdqu ymm7, [XMMBLOCK(6,0,esi,SIZEOF_DCTELEM)]
+ vpabsw ymm0, ymm4
+ vpabsw ymm1, ymm5
+ vpabsw ymm2, ymm6
+ vpabsw ymm3, ymm7
+
+ vpaddw ymm0, YMMWORD [CORRECTION(0,0,edx)] ; correction + roundfactor
+ vpaddw ymm1, YMMWORD [CORRECTION(2,0,edx)]
+ vpaddw ymm2, YMMWORD [CORRECTION(4,0,edx)]
+ vpaddw ymm3, YMMWORD [CORRECTION(6,0,edx)]
+ vpmulhuw ymm0, YMMWORD [RECIPROCAL(0,0,edx)] ; reciprocal
+ vpmulhuw ymm1, YMMWORD [RECIPROCAL(2,0,edx)]
+ vpmulhuw ymm2, YMMWORD [RECIPROCAL(4,0,edx)]
+ vpmulhuw ymm3, YMMWORD [RECIPROCAL(6,0,edx)]
+ vpmulhuw ymm0, YMMWORD [SCALE(0,0,edx)] ; scale
+ vpmulhuw ymm1, YMMWORD [SCALE(2,0,edx)]
+ vpmulhuw ymm2, YMMWORD [SCALE(4,0,edx)]
+ vpmulhuw ymm3, YMMWORD [SCALE(6,0,edx)]
+
+ vpsignw ymm0, ymm0, ymm4
+ vpsignw ymm1, ymm1, ymm5
+ vpsignw ymm2, ymm2, ymm6
+ vpsignw ymm3, ymm3, ymm7
+
+ vmovdqu [XMMBLOCK(0,0,edi,SIZEOF_DCTELEM)], ymm0
+ vmovdqu [XMMBLOCK(2,0,edi,SIZEOF_DCTELEM)], ymm1
+ vmovdqu [XMMBLOCK(4,0,edi,SIZEOF_DCTELEM)], ymm2
+ vmovdqu [XMMBLOCK(6,0,edi,SIZEOF_DCTELEM)], ymm3
+
+ vzeroupper
+ pop edi
+ pop esi
+; pop edx ; need not be preserved
+; pop ecx ; unused
+; pop ebx ; unused
+ pop ebp
+ ret
+
+; For some reason, the OS X linker does not honor the request to align the
+; segment unless we do this.
+ align 32
diff --git a/simd/i386/jquanti-sse2.asm b/simd/i386/jquanti-sse2.asm
new file mode 100644
index 0000000..6f42947
--- /dev/null
+++ b/simd/i386/jquanti-sse2.asm
@@ -0,0 +1,200 @@
+;
+; jquanti.asm - sample data conversion and quantization (SSE2)
+;
+; Copyright 2009 Pierre Ossman <ossman@cendio.se> for Cendio AB
+; Copyright (C) 2016, D. R. Commander.
+;
+; Based on the x86 SIMD extension for IJG JPEG library
+; Copyright (C) 1999-2006, MIYASAKA Masaru.
+; For conditions of distribution and use, see copyright notice in jsimdext.inc
+;
+; This file should be assembled with NASM (Netwide Assembler),
+; can *not* be assembled with Microsoft's MASM or any compatible
+; assembler (including Borland's Turbo Assembler).
+; NASM is available from http://nasm.sourceforge.net/ or
+; http://sourceforge.net/project/showfiles.php?group_id=6208
+;
+; [TAB8]
+
+%include "jsimdext.inc"
+%include "jdct.inc"
+
+; --------------------------------------------------------------------------
+ SECTION SEG_TEXT
+ BITS 32
+;
+; Load data into workspace, applying unsigned->signed conversion
+;
+; GLOBAL(void)
+; jsimd_convsamp_sse2 (JSAMPARRAY sample_data, JDIMENSION start_col,
+; DCTELEM *workspace);
+;
+
+%define sample_data ebp+8 ; JSAMPARRAY sample_data
+%define start_col ebp+12 ; JDIMENSION start_col
+%define workspace ebp+16 ; DCTELEM *workspace
+
+ align 32
+ global EXTN(jsimd_convsamp_sse2)
+
+EXTN(jsimd_convsamp_sse2):
+ push ebp
+ mov ebp, esp
+ push ebx
+; push ecx ; need not be preserved
+; push edx ; need not be preserved
+ push esi
+ push edi
+
+ pxor xmm6, xmm6 ; xmm6=(all 0's)
+ pcmpeqw xmm7, xmm7
+ psllw xmm7, 7 ; xmm7={0xFF80 0xFF80 0xFF80 0xFF80 ..}
+
+ mov esi, JSAMPARRAY [sample_data] ; (JSAMPROW *)
+ mov eax, JDIMENSION [start_col]
+ mov edi, POINTER [workspace] ; (DCTELEM *)
+ mov ecx, DCTSIZE/4
+ alignx 16, 7
+.convloop:
+ mov ebx, JSAMPROW [esi+0*SIZEOF_JSAMPROW] ; (JSAMPLE *)
+ mov edx, JSAMPROW [esi+1*SIZEOF_JSAMPROW] ; (JSAMPLE *)
+
+ movq xmm0, XMM_MMWORD [ebx+eax*SIZEOF_JSAMPLE] ; xmm0=(01234567)
+ movq xmm1, XMM_MMWORD [edx+eax*SIZEOF_JSAMPLE] ; xmm1=(89ABCDEF)
+
+ mov ebx, JSAMPROW [esi+2*SIZEOF_JSAMPROW] ; (JSAMPLE *)
+ mov edx, JSAMPROW [esi+3*SIZEOF_JSAMPROW] ; (JSAMPLE *)
+
+ movq xmm2, XMM_MMWORD [ebx+eax*SIZEOF_JSAMPLE] ; xmm2=(GHIJKLMN)
+ movq xmm3, XMM_MMWORD [edx+eax*SIZEOF_JSAMPLE] ; xmm3=(OPQRSTUV)
+
+ punpcklbw xmm0, xmm6 ; xmm0=(01234567)
+ punpcklbw xmm1, xmm6 ; xmm1=(89ABCDEF)
+ paddw xmm0, xmm7
+ paddw xmm1, xmm7
+ punpcklbw xmm2, xmm6 ; xmm2=(GHIJKLMN)
+ punpcklbw xmm3, xmm6 ; xmm3=(OPQRSTUV)
+ paddw xmm2, xmm7
+ paddw xmm3, xmm7
+
+ movdqa XMMWORD [XMMBLOCK(0,0,edi,SIZEOF_DCTELEM)], xmm0
+ movdqa XMMWORD [XMMBLOCK(1,0,edi,SIZEOF_DCTELEM)], xmm1
+ movdqa XMMWORD [XMMBLOCK(2,0,edi,SIZEOF_DCTELEM)], xmm2
+ movdqa XMMWORD [XMMBLOCK(3,0,edi,SIZEOF_DCTELEM)], xmm3
+
+ add esi, byte 4*SIZEOF_JSAMPROW
+ add edi, byte 4*DCTSIZE*SIZEOF_DCTELEM
+ dec ecx
+ jnz short .convloop
+
+ pop edi
+ pop esi
+; pop edx ; need not be preserved
+; pop ecx ; need not be preserved
+ pop ebx
+ pop ebp
+ ret
+
+; --------------------------------------------------------------------------
+;
+; Quantize/descale the coefficients, and store into coef_block
+;
+; This implementation is based on an algorithm described in
+; "How to optimize for the Pentium family of microprocessors"
+; (http://www.agner.org/assem/).
+;
+; GLOBAL(void)
+; jsimd_quantize_sse2 (JCOEFPTR coef_block, DCTELEM *divisors,
+; DCTELEM *workspace);
+;
+
+%define RECIPROCAL(m,n,b) XMMBLOCK(DCTSIZE*0+(m),(n),(b),SIZEOF_DCTELEM)
+%define CORRECTION(m,n,b) XMMBLOCK(DCTSIZE*1+(m),(n),(b),SIZEOF_DCTELEM)
+%define SCALE(m,n,b) XMMBLOCK(DCTSIZE*2+(m),(n),(b),SIZEOF_DCTELEM)
+
+%define coef_block ebp+8 ; JCOEFPTR coef_block
+%define divisors ebp+12 ; DCTELEM *divisors
+%define workspace ebp+16 ; DCTELEM *workspace
+
+ align 32
+ global EXTN(jsimd_quantize_sse2)
+
+EXTN(jsimd_quantize_sse2):
+ push ebp
+ mov ebp, esp
+; push ebx ; unused
+; push ecx ; unused
+; push edx ; need not be preserved
+ push esi
+ push edi
+
+ mov esi, POINTER [workspace]
+ mov edx, POINTER [divisors]
+ mov edi, JCOEFPTR [coef_block]
+ mov eax, DCTSIZE2/32
+ alignx 16, 7
+.quantloop:
+ movdqa xmm4, XMMWORD [XMMBLOCK(0,0,esi,SIZEOF_DCTELEM)]
+ movdqa xmm5, XMMWORD [XMMBLOCK(1,0,esi,SIZEOF_DCTELEM)]
+ movdqa xmm6, XMMWORD [XMMBLOCK(2,0,esi,SIZEOF_DCTELEM)]
+ movdqa xmm7, XMMWORD [XMMBLOCK(3,0,esi,SIZEOF_DCTELEM)]
+ movdqa xmm0, xmm4
+ movdqa xmm1, xmm5
+ movdqa xmm2, xmm6
+ movdqa xmm3, xmm7
+ psraw xmm4, (WORD_BIT-1)
+ psraw xmm5, (WORD_BIT-1)
+ psraw xmm6, (WORD_BIT-1)
+ psraw xmm7, (WORD_BIT-1)
+ pxor xmm0, xmm4
+ pxor xmm1, xmm5
+ pxor xmm2, xmm6
+ pxor xmm3, xmm7
+ psubw xmm0, xmm4 ; if (xmm0 < 0) xmm0 = -xmm0;
+ psubw xmm1, xmm5 ; if (xmm1 < 0) xmm1 = -xmm1;
+ psubw xmm2, xmm6 ; if (xmm2 < 0) xmm2 = -xmm2;
+ psubw xmm3, xmm7 ; if (xmm3 < 0) xmm3 = -xmm3;
+
+ paddw xmm0, XMMWORD [CORRECTION(0,0,edx)] ; correction + roundfactor
+ paddw xmm1, XMMWORD [CORRECTION(1,0,edx)]
+ paddw xmm2, XMMWORD [CORRECTION(2,0,edx)]
+ paddw xmm3, XMMWORD [CORRECTION(3,0,edx)]
+ pmulhuw xmm0, XMMWORD [RECIPROCAL(0,0,edx)] ; reciprocal
+ pmulhuw xmm1, XMMWORD [RECIPROCAL(1,0,edx)]
+ pmulhuw xmm2, XMMWORD [RECIPROCAL(2,0,edx)]
+ pmulhuw xmm3, XMMWORD [RECIPROCAL(3,0,edx)]
+ pmulhuw xmm0, XMMWORD [SCALE(0,0,edx)] ; scale
+ pmulhuw xmm1, XMMWORD [SCALE(1,0,edx)]
+ pmulhuw xmm2, XMMWORD [SCALE(2,0,edx)]
+ pmulhuw xmm3, XMMWORD [SCALE(3,0,edx)]
+
+ pxor xmm0, xmm4
+ pxor xmm1, xmm5
+ pxor xmm2, xmm6
+ pxor xmm3, xmm7
+ psubw xmm0, xmm4
+ psubw xmm1, xmm5
+ psubw xmm2, xmm6
+ psubw xmm3, xmm7
+ movdqa XMMWORD [XMMBLOCK(0,0,edi,SIZEOF_DCTELEM)], xmm0
+ movdqa XMMWORD [XMMBLOCK(1,0,edi,SIZEOF_DCTELEM)], xmm1
+ movdqa XMMWORD [XMMBLOCK(2,0,edi,SIZEOF_DCTELEM)], xmm2
+ movdqa XMMWORD [XMMBLOCK(3,0,edi,SIZEOF_DCTELEM)], xmm3
+
+ add esi, byte 32*SIZEOF_DCTELEM
+ add edx, byte 32*SIZEOF_DCTELEM
+ add edi, byte 32*SIZEOF_JCOEF
+ dec eax
+ jnz near .quantloop
+
+ pop edi
+ pop esi
+; pop edx ; need not be preserved
+; pop ecx ; unused
+; pop ebx ; unused
+ pop ebp
+ ret
+
+; For some reason, the OS X linker does not honor the request to align the
+; segment unless we do this.
+ align 32
diff --git a/simd/jsimd_i386.c b/simd/i386/jsimd.c
similarity index 82%
rename from simd/jsimd_i386.c
rename to simd/i386/jsimd.c
index 6da8bd8..4135123 100644
--- a/simd/jsimd_i386.c
+++ b/simd/i386/jsimd.c
@@ -15,12 +15,12 @@
*/
#define JPEG_INTERNALS
-#include "../jinclude.h"
-#include "../jpeglib.h"
+#include "../../jinclude.h"
+#include "../../jpeglib.h"
+#include "../../jsimd.h"
+#include "../../jdct.h"
+#include "../../jsimddct.h"
#include "../jsimd.h"
-#include "../jdct.h"
-#include "../jsimddct.h"
-#include "jsimd.h"
/*
* In the PIC cases, we have no guarantee that constants will keep
@@ -29,6 +29,7 @@
#define IS_ALIGNED(ptr, order) (((unsigned)ptr & ((1 << order) - 1)) == 0)
#define IS_ALIGNED_SSE(ptr) (IS_ALIGNED(ptr, 4)) /* 16 byte alignment */
+#define IS_ALIGNED_AVX(ptr) (IS_ALIGNED(ptr, 5)) /* 32 byte alignment */
static unsigned int simd_support = ~0;
static unsigned int simd_huffman = 1;
@@ -61,6 +62,9 @@
env = getenv("JSIMD_FORCESSE2");
if ((env != NULL) && (strcmp(env, "1") == 0))
simd_support &= JSIMD_SSE2;
+ env = getenv("JSIMD_FORCEAVX2");
+ if ((env != NULL) && (strcmp(env, "1") == 0))
+ simd_support &= JSIMD_AVX2;
env = getenv("JSIMD_FORCENONE");
if ((env != NULL) && (strcmp(env, "1") == 0))
simd_support = 0;
@@ -82,6 +86,9 @@
if ((RGB_PIXELSIZE != 3) && (RGB_PIXELSIZE != 4))
return 0;
+ if ((simd_support & JSIMD_AVX2) &&
+ IS_ALIGNED_AVX(jconst_rgb_ycc_convert_avx2))
+ return 1;
if ((simd_support & JSIMD_SSE2) &&
IS_ALIGNED_SSE(jconst_rgb_ycc_convert_sse2))
return 1;
@@ -104,6 +111,9 @@
if ((RGB_PIXELSIZE != 3) && (RGB_PIXELSIZE != 4))
return 0;
+ if ((simd_support & JSIMD_AVX2) &&
+ IS_ALIGNED_AVX(jconst_rgb_gray_convert_avx2))
+ return 1;
if ((simd_support & JSIMD_SSE2) &&
IS_ALIGNED_SSE(jconst_rgb_gray_convert_sse2))
return 1;
@@ -126,6 +136,9 @@
if ((RGB_PIXELSIZE != 3) && (RGB_PIXELSIZE != 4))
return 0;
+ if ((simd_support & JSIMD_AVX2) &&
+ IS_ALIGNED_AVX(jconst_ycc_rgb_convert_avx2))
+ return 1;
if ((simd_support & JSIMD_SSE2) &&
IS_ALIGNED_SSE(jconst_ycc_rgb_convert_sse2))
return 1;
@@ -146,46 +159,55 @@
JSAMPARRAY input_buf, JSAMPIMAGE output_buf,
JDIMENSION output_row, int num_rows)
{
+ void (*avx2fct)(JDIMENSION, JSAMPARRAY, JSAMPIMAGE, JDIMENSION, int);
void (*sse2fct)(JDIMENSION, JSAMPARRAY, JSAMPIMAGE, JDIMENSION, int);
void (*mmxfct)(JDIMENSION, JSAMPARRAY, JSAMPIMAGE, JDIMENSION, int);
switch(cinfo->in_color_space) {
case JCS_EXT_RGB:
+ avx2fct=jsimd_extrgb_ycc_convert_avx2;
sse2fct=jsimd_extrgb_ycc_convert_sse2;
mmxfct=jsimd_extrgb_ycc_convert_mmx;
break;
case JCS_EXT_RGBX:
case JCS_EXT_RGBA:
+ avx2fct=jsimd_extrgbx_ycc_convert_avx2;
sse2fct=jsimd_extrgbx_ycc_convert_sse2;
mmxfct=jsimd_extrgbx_ycc_convert_mmx;
break;
case JCS_EXT_BGR:
+ avx2fct=jsimd_extbgr_ycc_convert_avx2;
sse2fct=jsimd_extbgr_ycc_convert_sse2;
mmxfct=jsimd_extbgr_ycc_convert_mmx;
break;
case JCS_EXT_BGRX:
case JCS_EXT_BGRA:
+ avx2fct=jsimd_extbgrx_ycc_convert_avx2;
sse2fct=jsimd_extbgrx_ycc_convert_sse2;
mmxfct=jsimd_extbgrx_ycc_convert_mmx;
break;
case JCS_EXT_XBGR:
case JCS_EXT_ABGR:
+ avx2fct=jsimd_extxbgr_ycc_convert_avx2;
sse2fct=jsimd_extxbgr_ycc_convert_sse2;
mmxfct=jsimd_extxbgr_ycc_convert_mmx;
break;
case JCS_EXT_XRGB:
case JCS_EXT_ARGB:
+ avx2fct=jsimd_extxrgb_ycc_convert_avx2;
sse2fct=jsimd_extxrgb_ycc_convert_sse2;
mmxfct=jsimd_extxrgb_ycc_convert_mmx;
break;
default:
+ avx2fct=jsimd_rgb_ycc_convert_avx2;
sse2fct=jsimd_rgb_ycc_convert_sse2;
mmxfct=jsimd_rgb_ycc_convert_mmx;
break;
}
- if ((simd_support & JSIMD_SSE2) &&
- IS_ALIGNED_SSE(jconst_rgb_ycc_convert_sse2))
+ if (simd_support & JSIMD_AVX2)
+ avx2fct(cinfo->image_width, input_buf, output_buf, output_row, num_rows);
+ else if (simd_support & JSIMD_SSE2)
sse2fct(cinfo->image_width, input_buf, output_buf, output_row, num_rows);
else if (simd_support & JSIMD_MMX)
mmxfct(cinfo->image_width, input_buf, output_buf, output_row, num_rows);
@@ -196,46 +218,55 @@
JSAMPARRAY input_buf, JSAMPIMAGE output_buf,
JDIMENSION output_row, int num_rows)
{
+ void (*avx2fct)(JDIMENSION, JSAMPARRAY, JSAMPIMAGE, JDIMENSION, int);
void (*sse2fct)(JDIMENSION, JSAMPARRAY, JSAMPIMAGE, JDIMENSION, int);
void (*mmxfct)(JDIMENSION, JSAMPARRAY, JSAMPIMAGE, JDIMENSION, int);
switch(cinfo->in_color_space) {
case JCS_EXT_RGB:
+ avx2fct=jsimd_extrgb_gray_convert_avx2;
sse2fct=jsimd_extrgb_gray_convert_sse2;
mmxfct=jsimd_extrgb_gray_convert_mmx;
break;
case JCS_EXT_RGBX:
case JCS_EXT_RGBA:
+ avx2fct=jsimd_extrgbx_gray_convert_avx2;
sse2fct=jsimd_extrgbx_gray_convert_sse2;
mmxfct=jsimd_extrgbx_gray_convert_mmx;
break;
case JCS_EXT_BGR:
+ avx2fct=jsimd_extbgr_gray_convert_avx2;
sse2fct=jsimd_extbgr_gray_convert_sse2;
mmxfct=jsimd_extbgr_gray_convert_mmx;
break;
case JCS_EXT_BGRX:
case JCS_EXT_BGRA:
+ avx2fct=jsimd_extbgrx_gray_convert_avx2;
sse2fct=jsimd_extbgrx_gray_convert_sse2;
mmxfct=jsimd_extbgrx_gray_convert_mmx;
break;
case JCS_EXT_XBGR:
case JCS_EXT_ABGR:
+ avx2fct=jsimd_extxbgr_gray_convert_avx2;
sse2fct=jsimd_extxbgr_gray_convert_sse2;
mmxfct=jsimd_extxbgr_gray_convert_mmx;
break;
case JCS_EXT_XRGB:
case JCS_EXT_ARGB:
+ avx2fct=jsimd_extxrgb_gray_convert_avx2;
sse2fct=jsimd_extxrgb_gray_convert_sse2;
mmxfct=jsimd_extxrgb_gray_convert_mmx;
break;
default:
+ avx2fct=jsimd_rgb_gray_convert_avx2;
sse2fct=jsimd_rgb_gray_convert_sse2;
mmxfct=jsimd_rgb_gray_convert_mmx;
break;
}
- if ((simd_support & JSIMD_SSE2) &&
- IS_ALIGNED_SSE(jconst_rgb_gray_convert_sse2))
+ if (simd_support & JSIMD_AVX2)
+ avx2fct(cinfo->image_width, input_buf, output_buf, output_row, num_rows);
+ else if (simd_support & JSIMD_SSE2)
sse2fct(cinfo->image_width, input_buf, output_buf, output_row, num_rows);
else if (simd_support & JSIMD_MMX)
mmxfct(cinfo->image_width, input_buf, output_buf, output_row, num_rows);
@@ -246,46 +277,55 @@
JSAMPIMAGE input_buf, JDIMENSION input_row,
JSAMPARRAY output_buf, int num_rows)
{
+ void (*avx2fct)(JDIMENSION, JSAMPIMAGE, JDIMENSION, JSAMPARRAY, int);
void (*sse2fct)(JDIMENSION, JSAMPIMAGE, JDIMENSION, JSAMPARRAY, int);
void (*mmxfct)(JDIMENSION, JSAMPIMAGE, JDIMENSION, JSAMPARRAY, int);
switch(cinfo->out_color_space) {
case JCS_EXT_RGB:
+ avx2fct=jsimd_ycc_extrgb_convert_avx2;
sse2fct=jsimd_ycc_extrgb_convert_sse2;
mmxfct=jsimd_ycc_extrgb_convert_mmx;
break;
case JCS_EXT_RGBX:
case JCS_EXT_RGBA:
+ avx2fct=jsimd_ycc_extrgbx_convert_avx2;
sse2fct=jsimd_ycc_extrgbx_convert_sse2;
mmxfct=jsimd_ycc_extrgbx_convert_mmx;
break;
case JCS_EXT_BGR:
+ avx2fct=jsimd_ycc_extbgr_convert_avx2;
sse2fct=jsimd_ycc_extbgr_convert_sse2;
mmxfct=jsimd_ycc_extbgr_convert_mmx;
break;
case JCS_EXT_BGRX:
case JCS_EXT_BGRA:
+ avx2fct=jsimd_ycc_extbgrx_convert_avx2;
sse2fct=jsimd_ycc_extbgrx_convert_sse2;
mmxfct=jsimd_ycc_extbgrx_convert_mmx;
break;
case JCS_EXT_XBGR:
case JCS_EXT_ABGR:
+ avx2fct=jsimd_ycc_extxbgr_convert_avx2;
sse2fct=jsimd_ycc_extxbgr_convert_sse2;
mmxfct=jsimd_ycc_extxbgr_convert_mmx;
break;
case JCS_EXT_XRGB:
case JCS_EXT_ARGB:
+ avx2fct=jsimd_ycc_extxrgb_convert_avx2;
sse2fct=jsimd_ycc_extxrgb_convert_sse2;
mmxfct=jsimd_ycc_extxrgb_convert_mmx;
break;
default:
+ avx2fct=jsimd_ycc_rgb_convert_avx2;
sse2fct=jsimd_ycc_rgb_convert_sse2;
mmxfct=jsimd_ycc_rgb_convert_mmx;
break;
}
- if ((simd_support & JSIMD_SSE2) &&
- IS_ALIGNED_SSE(jconst_ycc_rgb_convert_sse2))
+ if (simd_support & JSIMD_AVX2)
+ avx2fct(cinfo->output_width, input_buf, input_row, output_buf, num_rows);
+ else if (simd_support & JSIMD_SSE2)
sse2fct(cinfo->output_width, input_buf, input_row, output_buf, num_rows);
else if (simd_support & JSIMD_MMX)
mmxfct(cinfo->output_width, input_buf, input_row, output_buf, num_rows);
@@ -309,6 +349,8 @@
if (sizeof(JDIMENSION) != 4)
return 0;
+ if (simd_support & JSIMD_AVX2)
+ return 1;
if (simd_support & JSIMD_SSE2)
return 1;
if (simd_support & JSIMD_MMX)
@@ -328,6 +370,8 @@
if (sizeof(JDIMENSION) != 4)
return 0;
+ if (simd_support & JSIMD_AVX2)
+ return 1;
if (simd_support & JSIMD_SSE2)
return 1;
if (simd_support & JSIMD_MMX)
@@ -340,7 +384,12 @@
jsimd_h2v2_downsample (j_compress_ptr cinfo, jpeg_component_info *compptr,
JSAMPARRAY input_data, JSAMPARRAY output_data)
{
- if (simd_support & JSIMD_SSE2)
+ if (simd_support & JSIMD_AVX2)
+ jsimd_h2v2_downsample_avx2(cinfo->image_width, cinfo->max_v_samp_factor,
+ compptr->v_samp_factor,
+ compptr->width_in_blocks, input_data,
+ output_data);
+ else if (simd_support & JSIMD_SSE2)
jsimd_h2v2_downsample_sse2(cinfo->image_width, cinfo->max_v_samp_factor,
compptr->v_samp_factor,
compptr->width_in_blocks, input_data,
@@ -355,7 +404,12 @@
jsimd_h2v1_downsample (j_compress_ptr cinfo, jpeg_component_info *compptr,
JSAMPARRAY input_data, JSAMPARRAY output_data)
{
- if (simd_support & JSIMD_SSE2)
+ if (simd_support & JSIMD_AVX2)
+ jsimd_h2v1_downsample_avx2(cinfo->image_width, cinfo->max_v_samp_factor,
+ compptr->v_samp_factor,
+ compptr->width_in_blocks, input_data,
+ output_data);
+ else if (simd_support & JSIMD_SSE2)
jsimd_h2v1_downsample_sse2(cinfo->image_width, cinfo->max_v_samp_factor,
compptr->v_samp_factor,
compptr->width_in_blocks, input_data,
@@ -377,6 +431,8 @@
if (sizeof(JDIMENSION) != 4)
return 0;
+ if (simd_support & JSIMD_AVX2)
+ return 1;
if (simd_support & JSIMD_SSE2)
return 1;
if (simd_support & JSIMD_MMX)
@@ -396,6 +452,8 @@
if (sizeof(JDIMENSION) != 4)
return 0;
+ if (simd_support & JSIMD_AVX2)
+ return 1;
if (simd_support & JSIMD_SSE2)
return 1;
if (simd_support & JSIMD_MMX)
@@ -410,7 +468,10 @@
JSAMPARRAY input_data,
JSAMPARRAY *output_data_ptr)
{
- if (simd_support & JSIMD_SSE2)
+ if (simd_support & JSIMD_AVX2)
+ jsimd_h2v2_upsample_avx2(cinfo->max_v_samp_factor, cinfo->output_width,
+ input_data, output_data_ptr);
+ else if (simd_support & JSIMD_SSE2)
jsimd_h2v2_upsample_sse2(cinfo->max_v_samp_factor, cinfo->output_width,
input_data, output_data_ptr);
else if (simd_support & JSIMD_MMX)
@@ -424,7 +485,10 @@
JSAMPARRAY input_data,
JSAMPARRAY *output_data_ptr)
{
- if (simd_support & JSIMD_SSE2)
+ if (simd_support & JSIMD_AVX2)
+ jsimd_h2v1_upsample_avx2(cinfo->max_v_samp_factor, cinfo->output_width,
+ input_data, output_data_ptr);
+ else if (simd_support & JSIMD_SSE2)
jsimd_h2v1_upsample_sse2(cinfo->max_v_samp_factor, cinfo->output_width,
input_data, output_data_ptr);
else if (simd_support & JSIMD_MMX)
@@ -443,6 +507,9 @@
if (sizeof(JDIMENSION) != 4)
return 0;
+ if ((simd_support & JSIMD_AVX2) &&
+ IS_ALIGNED_AVX(jconst_fancy_upsample_avx2))
+ return 1;
if ((simd_support & JSIMD_SSE2) &&
IS_ALIGNED_SSE(jconst_fancy_upsample_sse2))
return 1;
@@ -463,6 +530,9 @@
if (sizeof(JDIMENSION) != 4)
return 0;
+ if ((simd_support & JSIMD_AVX2) &&
+ IS_ALIGNED_AVX(jconst_fancy_upsample_avx2))
+ return 1;
if ((simd_support & JSIMD_SSE2) &&
IS_ALIGNED_SSE(jconst_fancy_upsample_sse2))
return 1;
@@ -478,8 +548,11 @@
JSAMPARRAY input_data,
JSAMPARRAY *output_data_ptr)
{
- if ((simd_support & JSIMD_SSE2) &&
- IS_ALIGNED_SSE(jconst_fancy_upsample_sse2))
+ if (simd_support & JSIMD_AVX2)
+ jsimd_h2v2_fancy_upsample_avx2(cinfo->max_v_samp_factor,
+ compptr->downsampled_width, input_data,
+ output_data_ptr);
+ else if (simd_support & JSIMD_SSE2)
jsimd_h2v2_fancy_upsample_sse2(cinfo->max_v_samp_factor,
compptr->downsampled_width, input_data,
output_data_ptr);
@@ -495,8 +568,11 @@
JSAMPARRAY input_data,
JSAMPARRAY *output_data_ptr)
{
- if ((simd_support & JSIMD_SSE2) &&
- IS_ALIGNED_SSE(jconst_fancy_upsample_sse2))
+ if (simd_support & JSIMD_AVX2)
+ jsimd_h2v1_fancy_upsample_avx2(cinfo->max_v_samp_factor,
+ compptr->downsampled_width, input_data,
+ output_data_ptr);
+ else if (simd_support & JSIMD_SSE2)
jsimd_h2v1_fancy_upsample_sse2(cinfo->max_v_samp_factor,
compptr->downsampled_width, input_data,
output_data_ptr);
@@ -517,6 +593,9 @@
if (sizeof(JDIMENSION) != 4)
return 0;
+ if ((simd_support & JSIMD_AVX2) &&
+ IS_ALIGNED_AVX(jconst_merged_upsample_avx2))
+ return 1;
if ((simd_support & JSIMD_SSE2) &&
IS_ALIGNED_SSE(jconst_merged_upsample_sse2))
return 1;
@@ -537,6 +616,9 @@
if (sizeof(JDIMENSION) != 4)
return 0;
+ if ((simd_support & JSIMD_AVX2) &&
+ IS_ALIGNED_AVX(jconst_merged_upsample_avx2))
+ return 1;
if ((simd_support & JSIMD_SSE2) &&
IS_ALIGNED_SSE(jconst_merged_upsample_sse2))
return 1;
@@ -552,46 +634,55 @@
JDIMENSION in_row_group_ctr,
JSAMPARRAY output_buf)
{
+ void (*avx2fct)(JDIMENSION, JSAMPIMAGE, JDIMENSION, JSAMPARRAY);
void (*sse2fct)(JDIMENSION, JSAMPIMAGE, JDIMENSION, JSAMPARRAY);
void (*mmxfct)(JDIMENSION, JSAMPIMAGE, JDIMENSION, JSAMPARRAY);
switch(cinfo->out_color_space) {
case JCS_EXT_RGB:
+ avx2fct=jsimd_h2v2_extrgb_merged_upsample_avx2;
sse2fct=jsimd_h2v2_extrgb_merged_upsample_sse2;
mmxfct=jsimd_h2v2_extrgb_merged_upsample_mmx;
break;
case JCS_EXT_RGBX:
case JCS_EXT_RGBA:
+ avx2fct=jsimd_h2v2_extrgbx_merged_upsample_avx2;
sse2fct=jsimd_h2v2_extrgbx_merged_upsample_sse2;
mmxfct=jsimd_h2v2_extrgbx_merged_upsample_mmx;
break;
case JCS_EXT_BGR:
+ avx2fct=jsimd_h2v2_extbgr_merged_upsample_avx2;
sse2fct=jsimd_h2v2_extbgr_merged_upsample_sse2;
mmxfct=jsimd_h2v2_extbgr_merged_upsample_mmx;
break;
case JCS_EXT_BGRX:
case JCS_EXT_BGRA:
+ avx2fct=jsimd_h2v2_extbgrx_merged_upsample_avx2;
sse2fct=jsimd_h2v2_extbgrx_merged_upsample_sse2;
mmxfct=jsimd_h2v2_extbgrx_merged_upsample_mmx;
break;
case JCS_EXT_XBGR:
case JCS_EXT_ABGR:
+ avx2fct=jsimd_h2v2_extxbgr_merged_upsample_avx2;
sse2fct=jsimd_h2v2_extxbgr_merged_upsample_sse2;
mmxfct=jsimd_h2v2_extxbgr_merged_upsample_mmx;
break;
case JCS_EXT_XRGB:
case JCS_EXT_ARGB:
+ avx2fct=jsimd_h2v2_extxrgb_merged_upsample_avx2;
sse2fct=jsimd_h2v2_extxrgb_merged_upsample_sse2;
mmxfct=jsimd_h2v2_extxrgb_merged_upsample_mmx;
break;
default:
+ avx2fct=jsimd_h2v2_merged_upsample_avx2;
sse2fct=jsimd_h2v2_merged_upsample_sse2;
mmxfct=jsimd_h2v2_merged_upsample_mmx;
break;
}
- if ((simd_support & JSIMD_SSE2) &&
- IS_ALIGNED_SSE(jconst_merged_upsample_sse2))
+ if (simd_support & JSIMD_AVX2)
+ avx2fct(cinfo->output_width, input_buf, in_row_group_ctr, output_buf);
+ else if (simd_support & JSIMD_SSE2)
sse2fct(cinfo->output_width, input_buf, in_row_group_ctr, output_buf);
else if (simd_support & JSIMD_MMX)
mmxfct(cinfo->output_width, input_buf, in_row_group_ctr, output_buf);
@@ -603,46 +694,55 @@
JDIMENSION in_row_group_ctr,
JSAMPARRAY output_buf)
{
+ void (*avx2fct)(JDIMENSION, JSAMPIMAGE, JDIMENSION, JSAMPARRAY);
void (*sse2fct)(JDIMENSION, JSAMPIMAGE, JDIMENSION, JSAMPARRAY);
void (*mmxfct)(JDIMENSION, JSAMPIMAGE, JDIMENSION, JSAMPARRAY);
switch(cinfo->out_color_space) {
case JCS_EXT_RGB:
+ avx2fct=jsimd_h2v1_extrgb_merged_upsample_avx2;
sse2fct=jsimd_h2v1_extrgb_merged_upsample_sse2;
mmxfct=jsimd_h2v1_extrgb_merged_upsample_mmx;
break;
case JCS_EXT_RGBX:
case JCS_EXT_RGBA:
+ avx2fct=jsimd_h2v1_extrgbx_merged_upsample_avx2;
sse2fct=jsimd_h2v1_extrgbx_merged_upsample_sse2;
mmxfct=jsimd_h2v1_extrgbx_merged_upsample_mmx;
break;
case JCS_EXT_BGR:
+ avx2fct=jsimd_h2v1_extbgr_merged_upsample_avx2;
sse2fct=jsimd_h2v1_extbgr_merged_upsample_sse2;
mmxfct=jsimd_h2v1_extbgr_merged_upsample_mmx;
break;
case JCS_EXT_BGRX:
case JCS_EXT_BGRA:
+ avx2fct=jsimd_h2v1_extbgrx_merged_upsample_avx2;
sse2fct=jsimd_h2v1_extbgrx_merged_upsample_sse2;
mmxfct=jsimd_h2v1_extbgrx_merged_upsample_mmx;
break;
case JCS_EXT_XBGR:
case JCS_EXT_ABGR:
+ avx2fct=jsimd_h2v1_extxbgr_merged_upsample_avx2;
sse2fct=jsimd_h2v1_extxbgr_merged_upsample_sse2;
mmxfct=jsimd_h2v1_extxbgr_merged_upsample_mmx;
break;
case JCS_EXT_XRGB:
case JCS_EXT_ARGB:
+ avx2fct=jsimd_h2v1_extxrgb_merged_upsample_avx2;
sse2fct=jsimd_h2v1_extxrgb_merged_upsample_sse2;
mmxfct=jsimd_h2v1_extxrgb_merged_upsample_mmx;
break;
default:
+ avx2fct=jsimd_h2v1_merged_upsample_avx2;
sse2fct=jsimd_h2v1_merged_upsample_sse2;
mmxfct=jsimd_h2v1_merged_upsample_mmx;
break;
}
- if ((simd_support & JSIMD_SSE2) &&
- IS_ALIGNED_SSE(jconst_merged_upsample_sse2))
+ if (simd_support & JSIMD_AVX2)
+ avx2fct(cinfo->output_width, input_buf, in_row_group_ctr, output_buf);
+ else if (simd_support & JSIMD_SSE2)
sse2fct(cinfo->output_width, input_buf, in_row_group_ctr, output_buf);
else if (simd_support & JSIMD_MMX)
mmxfct(cinfo->output_width, input_buf, in_row_group_ctr, output_buf);
@@ -815,6 +915,8 @@
if (sizeof(DCTELEM) != 2)
return 0;
+ if (simd_support & JSIMD_AVX2)
+ return 1;
if (simd_support & JSIMD_SSE2)
return 1;
if (simd_support & JSIMD_MMX)
@@ -850,7 +952,9 @@
jsimd_quantize (JCOEFPTR coef_block, DCTELEM *divisors,
DCTELEM *workspace)
{
- if (simd_support & JSIMD_SSE2)
+ if (simd_support & JSIMD_AVX2)
+ jsimd_quantize_avx2(coef_block, divisors, workspace);
+ else if (simd_support & JSIMD_SSE2)
jsimd_quantize_sse2(coef_block, divisors, workspace);
else if (simd_support & JSIMD_MMX)
jsimd_quantize_mmx(coef_block, divisors, workspace);
diff --git a/simd/i386/jsimdcpu.asm b/simd/i386/jsimdcpu.asm
new file mode 100644
index 0000000..827aefe
--- /dev/null
+++ b/simd/i386/jsimdcpu.asm
@@ -0,0 +1,131 @@
+;
+; jsimdcpu.asm - SIMD instruction support check
+;
+; Copyright 2009 Pierre Ossman <ossman@cendio.se> for Cendio AB
+; Copyright (C) 2016, D. R. Commander.
+;
+; Based on the x86 SIMD extension for IJG JPEG library
+; Copyright (C) 1999-2006, MIYASAKA Masaru.
+; For conditions of distribution and use, see copyright notice in jsimdext.inc
+;
+; This file should be assembled with NASM (Netwide Assembler),
+; can *not* be assembled with Microsoft's MASM or any compatible
+; assembler (including Borland's Turbo Assembler).
+; NASM is available from http://nasm.sourceforge.net/ or
+; http://sourceforge.net/project/showfiles.php?group_id=6208
+;
+; [TAB8]
+
+%include "jsimdext.inc"
+
+; --------------------------------------------------------------------------
+ SECTION SEG_TEXT
+ BITS 32
+;
+; Check if the CPU supports SIMD instructions
+;
+; GLOBAL(unsigned int)
+; jpeg_simd_cpu_support (void)
+;
+
+ align 32
+ global EXTN(jpeg_simd_cpu_support)
+
+EXTN(jpeg_simd_cpu_support):
+ push ebx
+; push ecx ; need not be preserved
+; push edx ; need not be preserved
+; push esi ; unused
+ push edi
+
+ xor edi, edi ; simd support flag
+
+ pushfd
+ pop eax
+ mov edx, eax
+ xor eax, 1<<21 ; flip ID bit in EFLAGS
+ push eax
+ popfd
+ pushfd
+ pop eax
+ xor eax, edx
+ jz near .return ; CPUID is not supported
+
+ ; Check for MMX instruction support
+ xor eax, eax
+ cpuid
+ test eax, eax
+ jz near .return
+
+ xor eax, eax
+ inc eax
+ cpuid
+ mov eax, edx ; eax = Standard feature flags
+
+ test eax, 1<<23 ; bit23:MMX
+ jz short .no_mmx
+ or edi, byte JSIMD_MMX
+.no_mmx:
+ test eax, 1<<25 ; bit25:SSE
+ jz short .no_sse
+ or edi, byte JSIMD_SSE
+.no_sse:
+ test eax, 1<<26 ; bit26:SSE2
+ jz short .no_sse2
+ or edi, byte JSIMD_SSE2
+.no_sse2:
+
+ ; Check for AVX2 instruction support
+ mov eax, 7
+ xor ecx, ecx
+ cpuid
+ mov eax, ebx
+ test eax, 1<<5 ; bit5:AVX2
+ jz short .no_avx2
+
+ ; Check for AVX2 O/S support
+ mov eax, 1
+ xor ecx, ecx
+ cpuid
+ test ecx, 1<<27
+ jz short .no_avx2 ; O/S does not support XSAVE
+ test ecx, 1<<28
+ jz short .no_avx2 ; CPU does not support AVX2
+
+ xor ecx, ecx
+ xgetbv
+ test eax, 6 ; O/S does not manage XMM/YMM state
+ ; using XSAVE
+ jz short .no_avx2
+
+ or edi, JSIMD_AVX2
+.no_avx2:
+
+ ; Check for 3DNow! instruction support
+ mov eax, 0x80000000
+ cpuid
+ cmp eax, 0x80000000
+ jbe short .return
+
+ mov eax, 0x80000001
+ cpuid
+ mov eax, edx ; eax = Extended feature flags
+
+ test eax, 1<<31 ; bit31:3DNow!(vendor independent)
+ jz short .no_3dnow
+ or edi, byte JSIMD_3DNOW
+.no_3dnow:
+
+.return:
+ mov eax, edi
+
+ pop edi
+; pop esi ; unused
+; pop edx ; need not be preserved
+; pop ecx ; need not be preserved
+ pop ebx
+ ret
+
+; For some reason, the OS X linker does not honor the request to align the
+; segment unless we do this.
+ align 32
diff --git a/simd/jccolext-sse2-64.asm b/simd/jccolext-sse2-64.asm
deleted file mode 100644
index 8e4642d..0000000
--- a/simd/jccolext-sse2-64.asm
+++ /dev/null
@@ -1,486 +0,0 @@
-;
-; jccolext.asm - colorspace conversion (64-bit SSE2)
-;
-; Copyright (C) 2009, D. R. Commander.
-;
-; Based on the x86 SIMD extension for IJG JPEG library
-; Copyright (C) 1999-2006, MIYASAKA Masaru.
-; For conditions of distribution and use, see copyright notice in jsimdext.inc
-;
-; This file should be assembled with NASM (Netwide Assembler),
-; can *not* be assembled with Microsoft's MASM or any compatible
-; assembler (including Borland's Turbo Assembler).
-; NASM is available from http://nasm.sourceforge.net/ or
-; http://sourceforge.net/project/showfiles.php?group_id=6208
-;
-; [TAB8]
-
-%include "jcolsamp.inc"
-
-; --------------------------------------------------------------------------
-;
-; Convert some rows of samples to the output colorspace.
-;
-; GLOBAL(void)
-; jsimd_rgb_ycc_convert_sse2 (JDIMENSION img_width,
-; JSAMPARRAY input_buf, JSAMPIMAGE output_buf,
-; JDIMENSION output_row, int num_rows);
-;
-
-; r10 = JDIMENSION img_width
-; r11 = JSAMPARRAY input_buf
-; r12 = JSAMPIMAGE output_buf
-; r13 = JDIMENSION output_row
-; r14 = int num_rows
-
-%define wk(i) rbp-(WK_NUM-(i))*SIZEOF_XMMWORD ; xmmword wk[WK_NUM]
-%define WK_NUM 8
-
- align 16
-
- global EXTN(jsimd_rgb_ycc_convert_sse2)
-
-EXTN(jsimd_rgb_ycc_convert_sse2):
- push rbp
- mov rax,rsp ; rax = original rbp
- sub rsp, byte 4
- and rsp, byte (-SIZEOF_XMMWORD) ; align to 128 bits
- mov [rsp],rax
- mov rbp,rsp ; rbp = aligned rbp
- lea rsp, [wk(0)]
- collect_args
- push rbx
-
- mov ecx, r10d
- test rcx,rcx
- jz near .return
-
- push rcx
-
- mov rsi, r12
- mov ecx, r13d
- mov rdi, JSAMPARRAY [rsi+0*SIZEOF_JSAMPARRAY]
- mov rbx, JSAMPARRAY [rsi+1*SIZEOF_JSAMPARRAY]
- mov rdx, JSAMPARRAY [rsi+2*SIZEOF_JSAMPARRAY]
- lea rdi, [rdi+rcx*SIZEOF_JSAMPROW]
- lea rbx, [rbx+rcx*SIZEOF_JSAMPROW]
- lea rdx, [rdx+rcx*SIZEOF_JSAMPROW]
-
- pop rcx
-
- mov rsi, r11
- mov eax, r14d
- test rax,rax
- jle near .return
-.rowloop:
- push rdx
- push rbx
- push rdi
- push rsi
- push rcx ; col
-
- mov rsi, JSAMPROW [rsi] ; inptr
- mov rdi, JSAMPROW [rdi] ; outptr0
- mov rbx, JSAMPROW [rbx] ; outptr1
- mov rdx, JSAMPROW [rdx] ; outptr2
-
- cmp rcx, byte SIZEOF_XMMWORD
- jae near .columnloop
-
-%if RGB_PIXELSIZE == 3 ; ---------------
-
-.column_ld1:
- push rax
- push rdx
- lea rcx,[rcx+rcx*2] ; imul ecx,RGB_PIXELSIZE
- test cl, SIZEOF_BYTE
- jz short .column_ld2
- sub rcx, byte SIZEOF_BYTE
- movzx rax, BYTE [rsi+rcx]
-.column_ld2:
- test cl, SIZEOF_WORD
- jz short .column_ld4
- sub rcx, byte SIZEOF_WORD
- movzx rdx, WORD [rsi+rcx]
- shl rax, WORD_BIT
- or rax,rdx
-.column_ld4:
- movd xmmA,eax
- pop rdx
- pop rax
- test cl, SIZEOF_DWORD
- jz short .column_ld8
- sub rcx, byte SIZEOF_DWORD
- movd xmmF, XMM_DWORD [rsi+rcx]
- pslldq xmmA, SIZEOF_DWORD
- por xmmA,xmmF
-.column_ld8:
- test cl, SIZEOF_MMWORD
- jz short .column_ld16
- sub rcx, byte SIZEOF_MMWORD
- movq xmmB, XMM_MMWORD [rsi+rcx]
- pslldq xmmA, SIZEOF_MMWORD
- por xmmA,xmmB
-.column_ld16:
- test cl, SIZEOF_XMMWORD
- jz short .column_ld32
- movdqa xmmF,xmmA
- movdqu xmmA, XMMWORD [rsi+0*SIZEOF_XMMWORD]
- mov rcx, SIZEOF_XMMWORD
- jmp short .rgb_ycc_cnv
-.column_ld32:
- test cl, 2*SIZEOF_XMMWORD
- mov rcx, SIZEOF_XMMWORD
- jz short .rgb_ycc_cnv
- movdqa xmmB,xmmA
- movdqu xmmA, XMMWORD [rsi+0*SIZEOF_XMMWORD]
- movdqu xmmF, XMMWORD [rsi+1*SIZEOF_XMMWORD]
- jmp short .rgb_ycc_cnv
-
-.columnloop:
- movdqu xmmA, XMMWORD [rsi+0*SIZEOF_XMMWORD]
- movdqu xmmF, XMMWORD [rsi+1*SIZEOF_XMMWORD]
- movdqu xmmB, XMMWORD [rsi+2*SIZEOF_XMMWORD]
-
-.rgb_ycc_cnv:
- ; xmmA=(00 10 20 01 11 21 02 12 22 03 13 23 04 14 24 05)
- ; xmmF=(15 25 06 16 26 07 17 27 08 18 28 09 19 29 0A 1A)
- ; xmmB=(2A 0B 1B 2B 0C 1C 2C 0D 1D 2D 0E 1E 2E 0F 1F 2F)
-
- movdqa xmmG,xmmA
- pslldq xmmA,8 ; xmmA=(-- -- -- -- -- -- -- -- 00 10 20 01 11 21 02 12)
- psrldq xmmG,8 ; xmmG=(22 03 13 23 04 14 24 05 -- -- -- -- -- -- -- --)
-
- punpckhbw xmmA,xmmF ; xmmA=(00 08 10 18 20 28 01 09 11 19 21 29 02 0A 12 1A)
- pslldq xmmF,8 ; xmmF=(-- -- -- -- -- -- -- -- 15 25 06 16 26 07 17 27)
-
- punpcklbw xmmG,xmmB ; xmmG=(22 2A 03 0B 13 1B 23 2B 04 0C 14 1C 24 2C 05 0D)
- punpckhbw xmmF,xmmB ; xmmF=(15 1D 25 2D 06 0E 16 1E 26 2E 07 0F 17 1F 27 2F)
-
- movdqa xmmD,xmmA
- pslldq xmmA,8 ; xmmA=(-- -- -- -- -- -- -- -- 00 08 10 18 20 28 01 09)
- psrldq xmmD,8 ; xmmD=(11 19 21 29 02 0A 12 1A -- -- -- -- -- -- -- --)
-
- punpckhbw xmmA,xmmG ; xmmA=(00 04 08 0C 10 14 18 1C 20 24 28 2C 01 05 09 0D)
- pslldq xmmG,8 ; xmmG=(-- -- -- -- -- -- -- -- 22 2A 03 0B 13 1B 23 2B)
-
- punpcklbw xmmD,xmmF ; xmmD=(11 15 19 1D 21 25 29 2D 02 06 0A 0E 12 16 1A 1E)
- punpckhbw xmmG,xmmF ; xmmG=(22 26 2A 2E 03 07 0B 0F 13 17 1B 1F 23 27 2B 2F)
-
- movdqa xmmE,xmmA
- pslldq xmmA,8 ; xmmA=(-- -- -- -- -- -- -- -- 00 04 08 0C 10 14 18 1C)
- psrldq xmmE,8 ; xmmE=(20 24 28 2C 01 05 09 0D -- -- -- -- -- -- -- --)
-
- punpckhbw xmmA,xmmD ; xmmA=(00 02 04 06 08 0A 0C 0E 10 12 14 16 18 1A 1C 1E)
- pslldq xmmD,8 ; xmmD=(-- -- -- -- -- -- -- -- 11 15 19 1D 21 25 29 2D)
-
- punpcklbw xmmE,xmmG ; xmmE=(20 22 24 26 28 2A 2C 2E 01 03 05 07 09 0B 0D 0F)
- punpckhbw xmmD,xmmG ; xmmD=(11 13 15 17 19 1B 1D 1F 21 23 25 27 29 2B 2D 2F)
-
- pxor xmmH,xmmH
-
- movdqa xmmC,xmmA
- punpcklbw xmmA,xmmH ; xmmA=(00 02 04 06 08 0A 0C 0E)
- punpckhbw xmmC,xmmH ; xmmC=(10 12 14 16 18 1A 1C 1E)
-
- movdqa xmmB,xmmE
- punpcklbw xmmE,xmmH ; xmmE=(20 22 24 26 28 2A 2C 2E)
- punpckhbw xmmB,xmmH ; xmmB=(01 03 05 07 09 0B 0D 0F)
-
- movdqa xmmF,xmmD
- punpcklbw xmmD,xmmH ; xmmD=(11 13 15 17 19 1B 1D 1F)
- punpckhbw xmmF,xmmH ; xmmF=(21 23 25 27 29 2B 2D 2F)
-
-%else ; RGB_PIXELSIZE == 4 ; -----------
-
-.column_ld1:
- test cl, SIZEOF_XMMWORD/16
- jz short .column_ld2
- sub rcx, byte SIZEOF_XMMWORD/16
- movd xmmA, XMM_DWORD [rsi+rcx*RGB_PIXELSIZE]
-.column_ld2:
- test cl, SIZEOF_XMMWORD/8
- jz short .column_ld4
- sub rcx, byte SIZEOF_XMMWORD/8
- movq xmmE, XMM_MMWORD [rsi+rcx*RGB_PIXELSIZE]
- pslldq xmmA, SIZEOF_MMWORD
- por xmmA,xmmE
-.column_ld4:
- test cl, SIZEOF_XMMWORD/4
- jz short .column_ld8
- sub rcx, byte SIZEOF_XMMWORD/4
- movdqa xmmE,xmmA
- movdqu xmmA, XMMWORD [rsi+rcx*RGB_PIXELSIZE]
-.column_ld8:
- test cl, SIZEOF_XMMWORD/2
- mov rcx, SIZEOF_XMMWORD
- jz short .rgb_ycc_cnv
- movdqa xmmF,xmmA
- movdqa xmmH,xmmE
- movdqu xmmA, XMMWORD [rsi+0*SIZEOF_XMMWORD]
- movdqu xmmE, XMMWORD [rsi+1*SIZEOF_XMMWORD]
- jmp short .rgb_ycc_cnv
-
-.columnloop:
- movdqu xmmA, XMMWORD [rsi+0*SIZEOF_XMMWORD]
- movdqu xmmE, XMMWORD [rsi+1*SIZEOF_XMMWORD]
- movdqu xmmF, XMMWORD [rsi+2*SIZEOF_XMMWORD]
- movdqu xmmH, XMMWORD [rsi+3*SIZEOF_XMMWORD]
-
-.rgb_ycc_cnv:
- ; xmmA=(00 10 20 30 01 11 21 31 02 12 22 32 03 13 23 33)
- ; xmmE=(04 14 24 34 05 15 25 35 06 16 26 36 07 17 27 37)
- ; xmmF=(08 18 28 38 09 19 29 39 0A 1A 2A 3A 0B 1B 2B 3B)
- ; xmmH=(0C 1C 2C 3C 0D 1D 2D 3D 0E 1E 2E 3E 0F 1F 2F 3F)
-
- movdqa xmmD,xmmA
- punpcklbw xmmA,xmmE ; xmmA=(00 04 10 14 20 24 30 34 01 05 11 15 21 25 31 35)
- punpckhbw xmmD,xmmE ; xmmD=(02 06 12 16 22 26 32 36 03 07 13 17 23 27 33 37)
-
- movdqa xmmC,xmmF
- punpcklbw xmmF,xmmH ; xmmF=(08 0C 18 1C 28 2C 38 3C 09 0D 19 1D 29 2D 39 3D)
- punpckhbw xmmC,xmmH ; xmmC=(0A 0E 1A 1E 2A 2E 3A 3E 0B 0F 1B 1F 2B 2F 3B 3F)
-
- movdqa xmmB,xmmA
- punpcklwd xmmA,xmmF ; xmmA=(00 04 08 0C 10 14 18 1C 20 24 28 2C 30 34 38 3C)
- punpckhwd xmmB,xmmF ; xmmB=(01 05 09 0D 11 15 19 1D 21 25 29 2D 31 35 39 3D)
-
- movdqa xmmG,xmmD
- punpcklwd xmmD,xmmC ; xmmD=(02 06 0A 0E 12 16 1A 1E 22 26 2A 2E 32 36 3A 3E)
- punpckhwd xmmG,xmmC ; xmmG=(03 07 0B 0F 13 17 1B 1F 23 27 2B 2F 33 37 3B 3F)
-
- movdqa xmmE,xmmA
- punpcklbw xmmA,xmmD ; xmmA=(00 02 04 06 08 0A 0C 0E 10 12 14 16 18 1A 1C 1E)
- punpckhbw xmmE,xmmD ; xmmE=(20 22 24 26 28 2A 2C 2E 30 32 34 36 38 3A 3C 3E)
-
- movdqa xmmH,xmmB
- punpcklbw xmmB,xmmG ; xmmB=(01 03 05 07 09 0B 0D 0F 11 13 15 17 19 1B 1D 1F)
- punpckhbw xmmH,xmmG ; xmmH=(21 23 25 27 29 2B 2D 2F 31 33 35 37 39 3B 3D 3F)
-
- pxor xmmF,xmmF
-
- movdqa xmmC,xmmA
- punpcklbw xmmA,xmmF ; xmmA=(00 02 04 06 08 0A 0C 0E)
- punpckhbw xmmC,xmmF ; xmmC=(10 12 14 16 18 1A 1C 1E)
-
- movdqa xmmD,xmmB
- punpcklbw xmmB,xmmF ; xmmB=(01 03 05 07 09 0B 0D 0F)
- punpckhbw xmmD,xmmF ; xmmD=(11 13 15 17 19 1B 1D 1F)
-
- movdqa xmmG,xmmE
- punpcklbw xmmE,xmmF ; xmmE=(20 22 24 26 28 2A 2C 2E)
- punpckhbw xmmG,xmmF ; xmmG=(30 32 34 36 38 3A 3C 3E)
-
- punpcklbw xmmF,xmmH
- punpckhbw xmmH,xmmH
- psrlw xmmF,BYTE_BIT ; xmmF=(21 23 25 27 29 2B 2D 2F)
- psrlw xmmH,BYTE_BIT ; xmmH=(31 33 35 37 39 3B 3D 3F)
-
-%endif ; RGB_PIXELSIZE ; ---------------
-
- ; xmm0=R(02468ACE)=RE, xmm2=G(02468ACE)=GE, xmm4=B(02468ACE)=BE
- ; xmm1=R(13579BDF)=RO, xmm3=G(13579BDF)=GO, xmm5=B(13579BDF)=BO
-
- ; (Original)
- ; Y = 0.29900 * R + 0.58700 * G + 0.11400 * B
- ; Cb = -0.16874 * R - 0.33126 * G + 0.50000 * B + CENTERJSAMPLE
- ; Cr = 0.50000 * R - 0.41869 * G - 0.08131 * B + CENTERJSAMPLE
- ;
- ; (This implementation)
- ; Y = 0.29900 * R + 0.33700 * G + 0.11400 * B + 0.25000 * G
- ; Cb = -0.16874 * R - 0.33126 * G + 0.50000 * B + CENTERJSAMPLE
- ; Cr = 0.50000 * R - 0.41869 * G - 0.08131 * B + CENTERJSAMPLE
-
- movdqa XMMWORD [wk(0)], xmm0 ; wk(0)=RE
- movdqa XMMWORD [wk(1)], xmm1 ; wk(1)=RO
- movdqa XMMWORD [wk(2)], xmm4 ; wk(2)=BE
- movdqa XMMWORD [wk(3)], xmm5 ; wk(3)=BO
-
- movdqa xmm6,xmm1
- punpcklwd xmm1,xmm3
- punpckhwd xmm6,xmm3
- movdqa xmm7,xmm1
- movdqa xmm4,xmm6
- pmaddwd xmm1,[rel PW_F0299_F0337] ; xmm1=ROL*FIX(0.299)+GOL*FIX(0.337)
- pmaddwd xmm6,[rel PW_F0299_F0337] ; xmm6=ROH*FIX(0.299)+GOH*FIX(0.337)
- pmaddwd xmm7,[rel PW_MF016_MF033] ; xmm7=ROL*-FIX(0.168)+GOL*-FIX(0.331)
- pmaddwd xmm4,[rel PW_MF016_MF033] ; xmm4=ROH*-FIX(0.168)+GOH*-FIX(0.331)
-
- movdqa XMMWORD [wk(4)], xmm1 ; wk(4)=ROL*FIX(0.299)+GOL*FIX(0.337)
- movdqa XMMWORD [wk(5)], xmm6 ; wk(5)=ROH*FIX(0.299)+GOH*FIX(0.337)
-
- pxor xmm1,xmm1
- pxor xmm6,xmm6
- punpcklwd xmm1,xmm5 ; xmm1=BOL
- punpckhwd xmm6,xmm5 ; xmm6=BOH
- psrld xmm1,1 ; xmm1=BOL*FIX(0.500)
- psrld xmm6,1 ; xmm6=BOH*FIX(0.500)
-
- movdqa xmm5,[rel PD_ONEHALFM1_CJ] ; xmm5=[PD_ONEHALFM1_CJ]
-
- paddd xmm7,xmm1
- paddd xmm4,xmm6
- paddd xmm7,xmm5
- paddd xmm4,xmm5
- psrld xmm7,SCALEBITS ; xmm7=CbOL
- psrld xmm4,SCALEBITS ; xmm4=CbOH
- packssdw xmm7,xmm4 ; xmm7=CbO
-
- movdqa xmm1, XMMWORD [wk(2)] ; xmm1=BE
-
- movdqa xmm6,xmm0
- punpcklwd xmm0,xmm2
- punpckhwd xmm6,xmm2
- movdqa xmm5,xmm0
- movdqa xmm4,xmm6
- pmaddwd xmm0,[rel PW_F0299_F0337] ; xmm0=REL*FIX(0.299)+GEL*FIX(0.337)
- pmaddwd xmm6,[rel PW_F0299_F0337] ; xmm6=REH*FIX(0.299)+GEH*FIX(0.337)
- pmaddwd xmm5,[rel PW_MF016_MF033] ; xmm5=REL*-FIX(0.168)+GEL*-FIX(0.331)
- pmaddwd xmm4,[rel PW_MF016_MF033] ; xmm4=REH*-FIX(0.168)+GEH*-FIX(0.331)
-
- movdqa XMMWORD [wk(6)], xmm0 ; wk(6)=REL*FIX(0.299)+GEL*FIX(0.337)
- movdqa XMMWORD [wk(7)], xmm6 ; wk(7)=REH*FIX(0.299)+GEH*FIX(0.337)
-
- pxor xmm0,xmm0
- pxor xmm6,xmm6
- punpcklwd xmm0,xmm1 ; xmm0=BEL
- punpckhwd xmm6,xmm1 ; xmm6=BEH
- psrld xmm0,1 ; xmm0=BEL*FIX(0.500)
- psrld xmm6,1 ; xmm6=BEH*FIX(0.500)
-
- movdqa xmm1,[rel PD_ONEHALFM1_CJ] ; xmm1=[PD_ONEHALFM1_CJ]
-
- paddd xmm5,xmm0
- paddd xmm4,xmm6
- paddd xmm5,xmm1
- paddd xmm4,xmm1
- psrld xmm5,SCALEBITS ; xmm5=CbEL
- psrld xmm4,SCALEBITS ; xmm4=CbEH
- packssdw xmm5,xmm4 ; xmm5=CbE
-
- psllw xmm7,BYTE_BIT
- por xmm5,xmm7 ; xmm5=Cb
- movdqa XMMWORD [rbx], xmm5 ; Save Cb
-
- movdqa xmm0, XMMWORD [wk(3)] ; xmm0=BO
- movdqa xmm6, XMMWORD [wk(2)] ; xmm6=BE
- movdqa xmm1, XMMWORD [wk(1)] ; xmm1=RO
-
- movdqa xmm4,xmm0
- punpcklwd xmm0,xmm3
- punpckhwd xmm4,xmm3
- movdqa xmm7,xmm0
- movdqa xmm5,xmm4
- pmaddwd xmm0,[rel PW_F0114_F0250] ; xmm0=BOL*FIX(0.114)+GOL*FIX(0.250)
- pmaddwd xmm4,[rel PW_F0114_F0250] ; xmm4=BOH*FIX(0.114)+GOH*FIX(0.250)
- pmaddwd xmm7,[rel PW_MF008_MF041] ; xmm7=BOL*-FIX(0.081)+GOL*-FIX(0.418)
- pmaddwd xmm5,[rel PW_MF008_MF041] ; xmm5=BOH*-FIX(0.081)+GOH*-FIX(0.418)
-
- movdqa xmm3,[rel PD_ONEHALF] ; xmm3=[PD_ONEHALF]
-
- paddd xmm0, XMMWORD [wk(4)]
- paddd xmm4, XMMWORD [wk(5)]
- paddd xmm0,xmm3
- paddd xmm4,xmm3
- psrld xmm0,SCALEBITS ; xmm0=YOL
- psrld xmm4,SCALEBITS ; xmm4=YOH
- packssdw xmm0,xmm4 ; xmm0=YO
-
- pxor xmm3,xmm3
- pxor xmm4,xmm4
- punpcklwd xmm3,xmm1 ; xmm3=ROL
- punpckhwd xmm4,xmm1 ; xmm4=ROH
- psrld xmm3,1 ; xmm3=ROL*FIX(0.500)
- psrld xmm4,1 ; xmm4=ROH*FIX(0.500)
-
- movdqa xmm1,[rel PD_ONEHALFM1_CJ] ; xmm1=[PD_ONEHALFM1_CJ]
-
- paddd xmm7,xmm3
- paddd xmm5,xmm4
- paddd xmm7,xmm1
- paddd xmm5,xmm1
- psrld xmm7,SCALEBITS ; xmm7=CrOL
- psrld xmm5,SCALEBITS ; xmm5=CrOH
- packssdw xmm7,xmm5 ; xmm7=CrO
-
- movdqa xmm3, XMMWORD [wk(0)] ; xmm3=RE
-
- movdqa xmm4,xmm6
- punpcklwd xmm6,xmm2
- punpckhwd xmm4,xmm2
- movdqa xmm1,xmm6
- movdqa xmm5,xmm4
- pmaddwd xmm6,[rel PW_F0114_F0250] ; xmm6=BEL*FIX(0.114)+GEL*FIX(0.250)
- pmaddwd xmm4,[rel PW_F0114_F0250] ; xmm4=BEH*FIX(0.114)+GEH*FIX(0.250)
- pmaddwd xmm1,[rel PW_MF008_MF041] ; xmm1=BEL*-FIX(0.081)+GEL*-FIX(0.418)
- pmaddwd xmm5,[rel PW_MF008_MF041] ; xmm5=BEH*-FIX(0.081)+GEH*-FIX(0.418)
-
- movdqa xmm2,[rel PD_ONEHALF] ; xmm2=[PD_ONEHALF]
-
- paddd xmm6, XMMWORD [wk(6)]
- paddd xmm4, XMMWORD [wk(7)]
- paddd xmm6,xmm2
- paddd xmm4,xmm2
- psrld xmm6,SCALEBITS ; xmm6=YEL
- psrld xmm4,SCALEBITS ; xmm4=YEH
- packssdw xmm6,xmm4 ; xmm6=YE
-
- psllw xmm0,BYTE_BIT
- por xmm6,xmm0 ; xmm6=Y
- movdqa XMMWORD [rdi], xmm6 ; Save Y
-
- pxor xmm2,xmm2
- pxor xmm4,xmm4
- punpcklwd xmm2,xmm3 ; xmm2=REL
- punpckhwd xmm4,xmm3 ; xmm4=REH
- psrld xmm2,1 ; xmm2=REL*FIX(0.500)
- psrld xmm4,1 ; xmm4=REH*FIX(0.500)
-
- movdqa xmm0,[rel PD_ONEHALFM1_CJ] ; xmm0=[PD_ONEHALFM1_CJ]
-
- paddd xmm1,xmm2
- paddd xmm5,xmm4
- paddd xmm1,xmm0
- paddd xmm5,xmm0
- psrld xmm1,SCALEBITS ; xmm1=CrEL
- psrld xmm5,SCALEBITS ; xmm5=CrEH
- packssdw xmm1,xmm5 ; xmm1=CrE
-
- psllw xmm7,BYTE_BIT
- por xmm1,xmm7 ; xmm1=Cr
- movdqa XMMWORD [rdx], xmm1 ; Save Cr
-
- sub rcx, byte SIZEOF_XMMWORD
- add rsi, byte RGB_PIXELSIZE*SIZEOF_XMMWORD ; inptr
- add rdi, byte SIZEOF_XMMWORD ; outptr0
- add rbx, byte SIZEOF_XMMWORD ; outptr1
- add rdx, byte SIZEOF_XMMWORD ; outptr2
- cmp rcx, byte SIZEOF_XMMWORD
- jae near .columnloop
- test rcx,rcx
- jnz near .column_ld1
-
- pop rcx ; col
- pop rsi
- pop rdi
- pop rbx
- pop rdx
-
- add rsi, byte SIZEOF_JSAMPROW ; input_buf
- add rdi, byte SIZEOF_JSAMPROW
- add rbx, byte SIZEOF_JSAMPROW
- add rdx, byte SIZEOF_JSAMPROW
- dec rax ; num_rows
- jg near .rowloop
-
-.return:
- pop rbx
- uncollect_args
- mov rsp,rbp ; rsp <- aligned rbp
- pop rsp ; rsp <- original rbp
- pop rbp
- ret
-
-; For some reason, the OS X linker does not honor the request to align the
-; segment unless we do this.
- align 16
diff --git a/simd/jccolext-sse2.asm b/simd/jccolext-sse2.asm
deleted file mode 100644
index cc38e98..0000000
--- a/simd/jccolext-sse2.asm
+++ /dev/null
@@ -1,503 +0,0 @@
-;
-; jccolext.asm - colorspace conversion (SSE2)
-;
-; x86 SIMD extension for IJG JPEG library
-; Copyright (C) 1999-2006, MIYASAKA Masaru.
-; For conditions of distribution and use, see copyright notice in jsimdext.inc
-;
-; This file should be assembled with NASM (Netwide Assembler),
-; can *not* be assembled with Microsoft's MASM or any compatible
-; assembler (including Borland's Turbo Assembler).
-; NASM is available from http://nasm.sourceforge.net/ or
-; http://sourceforge.net/project/showfiles.php?group_id=6208
-;
-; [TAB8]
-
-%include "jcolsamp.inc"
-
-; --------------------------------------------------------------------------
-;
-; Convert some rows of samples to the output colorspace.
-;
-; GLOBAL(void)
-; jsimd_rgb_ycc_convert_sse2 (JDIMENSION img_width,
-; JSAMPARRAY input_buf, JSAMPIMAGE output_buf,
-; JDIMENSION output_row, int num_rows);
-;
-
-%define img_width(b) (b)+8 ; JDIMENSION img_width
-%define input_buf(b) (b)+12 ; JSAMPARRAY input_buf
-%define output_buf(b) (b)+16 ; JSAMPIMAGE output_buf
-%define output_row(b) (b)+20 ; JDIMENSION output_row
-%define num_rows(b) (b)+24 ; int num_rows
-
-%define original_ebp ebp+0
-%define wk(i) ebp-(WK_NUM-(i))*SIZEOF_XMMWORD ; xmmword wk[WK_NUM]
-%define WK_NUM 8
-%define gotptr wk(0)-SIZEOF_POINTER ; void * gotptr
-
- align 16
-
- global EXTN(jsimd_rgb_ycc_convert_sse2)
-
-EXTN(jsimd_rgb_ycc_convert_sse2):
- push ebp
- mov eax,esp ; eax = original ebp
- sub esp, byte 4
- and esp, byte (-SIZEOF_XMMWORD) ; align to 128 bits
- mov [esp],eax
- mov ebp,esp ; ebp = aligned ebp
- lea esp, [wk(0)]
- pushpic eax ; make a room for GOT address
- push ebx
-; push ecx ; need not be preserved
-; push edx ; need not be preserved
- push esi
- push edi
-
- get_GOT ebx ; get GOT address
- movpic POINTER [gotptr], ebx ; save GOT address
-
- mov ecx, JDIMENSION [img_width(eax)]
- test ecx,ecx
- jz near .return
-
- push ecx
-
- mov esi, JSAMPIMAGE [output_buf(eax)]
- mov ecx, JDIMENSION [output_row(eax)]
- mov edi, JSAMPARRAY [esi+0*SIZEOF_JSAMPARRAY]
- mov ebx, JSAMPARRAY [esi+1*SIZEOF_JSAMPARRAY]
- mov edx, JSAMPARRAY [esi+2*SIZEOF_JSAMPARRAY]
- lea edi, [edi+ecx*SIZEOF_JSAMPROW]
- lea ebx, [ebx+ecx*SIZEOF_JSAMPROW]
- lea edx, [edx+ecx*SIZEOF_JSAMPROW]
-
- pop ecx
-
- mov esi, JSAMPARRAY [input_buf(eax)]
- mov eax, INT [num_rows(eax)]
- test eax,eax
- jle near .return
- alignx 16,7
-.rowloop:
- pushpic eax
- push edx
- push ebx
- push edi
- push esi
- push ecx ; col
-
- mov esi, JSAMPROW [esi] ; inptr
- mov edi, JSAMPROW [edi] ; outptr0
- mov ebx, JSAMPROW [ebx] ; outptr1
- mov edx, JSAMPROW [edx] ; outptr2
- movpic eax, POINTER [gotptr] ; load GOT address (eax)
-
- cmp ecx, byte SIZEOF_XMMWORD
- jae near .columnloop
- alignx 16,7
-
-%if RGB_PIXELSIZE == 3 ; ---------------
-
-.column_ld1:
- push eax
- push edx
- lea ecx,[ecx+ecx*2] ; imul ecx,RGB_PIXELSIZE
- test cl, SIZEOF_BYTE
- jz short .column_ld2
- sub ecx, byte SIZEOF_BYTE
- movzx eax, BYTE [esi+ecx]
-.column_ld2:
- test cl, SIZEOF_WORD
- jz short .column_ld4
- sub ecx, byte SIZEOF_WORD
- movzx edx, WORD [esi+ecx]
- shl eax, WORD_BIT
- or eax,edx
-.column_ld4:
- movd xmmA,eax
- pop edx
- pop eax
- test cl, SIZEOF_DWORD
- jz short .column_ld8
- sub ecx, byte SIZEOF_DWORD
- movd xmmF, XMM_DWORD [esi+ecx]
- pslldq xmmA, SIZEOF_DWORD
- por xmmA,xmmF
-.column_ld8:
- test cl, SIZEOF_MMWORD
- jz short .column_ld16
- sub ecx, byte SIZEOF_MMWORD
- movq xmmB, XMM_MMWORD [esi+ecx]
- pslldq xmmA, SIZEOF_MMWORD
- por xmmA,xmmB
-.column_ld16:
- test cl, SIZEOF_XMMWORD
- jz short .column_ld32
- movdqa xmmF,xmmA
- movdqu xmmA, XMMWORD [esi+0*SIZEOF_XMMWORD]
- mov ecx, SIZEOF_XMMWORD
- jmp short .rgb_ycc_cnv
-.column_ld32:
- test cl, 2*SIZEOF_XMMWORD
- mov ecx, SIZEOF_XMMWORD
- jz short .rgb_ycc_cnv
- movdqa xmmB,xmmA
- movdqu xmmA, XMMWORD [esi+0*SIZEOF_XMMWORD]
- movdqu xmmF, XMMWORD [esi+1*SIZEOF_XMMWORD]
- jmp short .rgb_ycc_cnv
- alignx 16,7
-
-.columnloop:
- movdqu xmmA, XMMWORD [esi+0*SIZEOF_XMMWORD]
- movdqu xmmF, XMMWORD [esi+1*SIZEOF_XMMWORD]
- movdqu xmmB, XMMWORD [esi+2*SIZEOF_XMMWORD]
-
-.rgb_ycc_cnv:
- ; xmmA=(00 10 20 01 11 21 02 12 22 03 13 23 04 14 24 05)
- ; xmmF=(15 25 06 16 26 07 17 27 08 18 28 09 19 29 0A 1A)
- ; xmmB=(2A 0B 1B 2B 0C 1C 2C 0D 1D 2D 0E 1E 2E 0F 1F 2F)
-
- movdqa xmmG,xmmA
- pslldq xmmA,8 ; xmmA=(-- -- -- -- -- -- -- -- 00 10 20 01 11 21 02 12)
- psrldq xmmG,8 ; xmmG=(22 03 13 23 04 14 24 05 -- -- -- -- -- -- -- --)
-
- punpckhbw xmmA,xmmF ; xmmA=(00 08 10 18 20 28 01 09 11 19 21 29 02 0A 12 1A)
- pslldq xmmF,8 ; xmmF=(-- -- -- -- -- -- -- -- 15 25 06 16 26 07 17 27)
-
- punpcklbw xmmG,xmmB ; xmmG=(22 2A 03 0B 13 1B 23 2B 04 0C 14 1C 24 2C 05 0D)
- punpckhbw xmmF,xmmB ; xmmF=(15 1D 25 2D 06 0E 16 1E 26 2E 07 0F 17 1F 27 2F)
-
- movdqa xmmD,xmmA
- pslldq xmmA,8 ; xmmA=(-- -- -- -- -- -- -- -- 00 08 10 18 20 28 01 09)
- psrldq xmmD,8 ; xmmD=(11 19 21 29 02 0A 12 1A -- -- -- -- -- -- -- --)
-
- punpckhbw xmmA,xmmG ; xmmA=(00 04 08 0C 10 14 18 1C 20 24 28 2C 01 05 09 0D)
- pslldq xmmG,8 ; xmmG=(-- -- -- -- -- -- -- -- 22 2A 03 0B 13 1B 23 2B)
-
- punpcklbw xmmD,xmmF ; xmmD=(11 15 19 1D 21 25 29 2D 02 06 0A 0E 12 16 1A 1E)
- punpckhbw xmmG,xmmF ; xmmG=(22 26 2A 2E 03 07 0B 0F 13 17 1B 1F 23 27 2B 2F)
-
- movdqa xmmE,xmmA
- pslldq xmmA,8 ; xmmA=(-- -- -- -- -- -- -- -- 00 04 08 0C 10 14 18 1C)
- psrldq xmmE,8 ; xmmE=(20 24 28 2C 01 05 09 0D -- -- -- -- -- -- -- --)
-
- punpckhbw xmmA,xmmD ; xmmA=(00 02 04 06 08 0A 0C 0E 10 12 14 16 18 1A 1C 1E)
- pslldq xmmD,8 ; xmmD=(-- -- -- -- -- -- -- -- 11 15 19 1D 21 25 29 2D)
-
- punpcklbw xmmE,xmmG ; xmmE=(20 22 24 26 28 2A 2C 2E 01 03 05 07 09 0B 0D 0F)
- punpckhbw xmmD,xmmG ; xmmD=(11 13 15 17 19 1B 1D 1F 21 23 25 27 29 2B 2D 2F)
-
- pxor xmmH,xmmH
-
- movdqa xmmC,xmmA
- punpcklbw xmmA,xmmH ; xmmA=(00 02 04 06 08 0A 0C 0E)
- punpckhbw xmmC,xmmH ; xmmC=(10 12 14 16 18 1A 1C 1E)
-
- movdqa xmmB,xmmE
- punpcklbw xmmE,xmmH ; xmmE=(20 22 24 26 28 2A 2C 2E)
- punpckhbw xmmB,xmmH ; xmmB=(01 03 05 07 09 0B 0D 0F)
-
- movdqa xmmF,xmmD
- punpcklbw xmmD,xmmH ; xmmD=(11 13 15 17 19 1B 1D 1F)
- punpckhbw xmmF,xmmH ; xmmF=(21 23 25 27 29 2B 2D 2F)
-
-%else ; RGB_PIXELSIZE == 4 ; -----------
-
-.column_ld1:
- test cl, SIZEOF_XMMWORD/16
- jz short .column_ld2
- sub ecx, byte SIZEOF_XMMWORD/16
- movd xmmA, XMM_DWORD [esi+ecx*RGB_PIXELSIZE]
-.column_ld2:
- test cl, SIZEOF_XMMWORD/8
- jz short .column_ld4
- sub ecx, byte SIZEOF_XMMWORD/8
- movq xmmE, XMM_MMWORD [esi+ecx*RGB_PIXELSIZE]
- pslldq xmmA, SIZEOF_MMWORD
- por xmmA,xmmE
-.column_ld4:
- test cl, SIZEOF_XMMWORD/4
- jz short .column_ld8
- sub ecx, byte SIZEOF_XMMWORD/4
- movdqa xmmE,xmmA
- movdqu xmmA, XMMWORD [esi+ecx*RGB_PIXELSIZE]
-.column_ld8:
- test cl, SIZEOF_XMMWORD/2
- mov ecx, SIZEOF_XMMWORD
- jz short .rgb_ycc_cnv
- movdqa xmmF,xmmA
- movdqa xmmH,xmmE
- movdqu xmmA, XMMWORD [esi+0*SIZEOF_XMMWORD]
- movdqu xmmE, XMMWORD [esi+1*SIZEOF_XMMWORD]
- jmp short .rgb_ycc_cnv
- alignx 16,7
-
-.columnloop:
- movdqu xmmA, XMMWORD [esi+0*SIZEOF_XMMWORD]
- movdqu xmmE, XMMWORD [esi+1*SIZEOF_XMMWORD]
- movdqu xmmF, XMMWORD [esi+2*SIZEOF_XMMWORD]
- movdqu xmmH, XMMWORD [esi+3*SIZEOF_XMMWORD]
-
-.rgb_ycc_cnv:
- ; xmmA=(00 10 20 30 01 11 21 31 02 12 22 32 03 13 23 33)
- ; xmmE=(04 14 24 34 05 15 25 35 06 16 26 36 07 17 27 37)
- ; xmmF=(08 18 28 38 09 19 29 39 0A 1A 2A 3A 0B 1B 2B 3B)
- ; xmmH=(0C 1C 2C 3C 0D 1D 2D 3D 0E 1E 2E 3E 0F 1F 2F 3F)
-
- movdqa xmmD,xmmA
- punpcklbw xmmA,xmmE ; xmmA=(00 04 10 14 20 24 30 34 01 05 11 15 21 25 31 35)
- punpckhbw xmmD,xmmE ; xmmD=(02 06 12 16 22 26 32 36 03 07 13 17 23 27 33 37)
-
- movdqa xmmC,xmmF
- punpcklbw xmmF,xmmH ; xmmF=(08 0C 18 1C 28 2C 38 3C 09 0D 19 1D 29 2D 39 3D)
- punpckhbw xmmC,xmmH ; xmmC=(0A 0E 1A 1E 2A 2E 3A 3E 0B 0F 1B 1F 2B 2F 3B 3F)
-
- movdqa xmmB,xmmA
- punpcklwd xmmA,xmmF ; xmmA=(00 04 08 0C 10 14 18 1C 20 24 28 2C 30 34 38 3C)
- punpckhwd xmmB,xmmF ; xmmB=(01 05 09 0D 11 15 19 1D 21 25 29 2D 31 35 39 3D)
-
- movdqa xmmG,xmmD
- punpcklwd xmmD,xmmC ; xmmD=(02 06 0A 0E 12 16 1A 1E 22 26 2A 2E 32 36 3A 3E)
- punpckhwd xmmG,xmmC ; xmmG=(03 07 0B 0F 13 17 1B 1F 23 27 2B 2F 33 37 3B 3F)
-
- movdqa xmmE,xmmA
- punpcklbw xmmA,xmmD ; xmmA=(00 02 04 06 08 0A 0C 0E 10 12 14 16 18 1A 1C 1E)
- punpckhbw xmmE,xmmD ; xmmE=(20 22 24 26 28 2A 2C 2E 30 32 34 36 38 3A 3C 3E)
-
- movdqa xmmH,xmmB
- punpcklbw xmmB,xmmG ; xmmB=(01 03 05 07 09 0B 0D 0F 11 13 15 17 19 1B 1D 1F)
- punpckhbw xmmH,xmmG ; xmmH=(21 23 25 27 29 2B 2D 2F 31 33 35 37 39 3B 3D 3F)
-
- pxor xmmF,xmmF
-
- movdqa xmmC,xmmA
- punpcklbw xmmA,xmmF ; xmmA=(00 02 04 06 08 0A 0C 0E)
- punpckhbw xmmC,xmmF ; xmmC=(10 12 14 16 18 1A 1C 1E)
-
- movdqa xmmD,xmmB
- punpcklbw xmmB,xmmF ; xmmB=(01 03 05 07 09 0B 0D 0F)
- punpckhbw xmmD,xmmF ; xmmD=(11 13 15 17 19 1B 1D 1F)
-
- movdqa xmmG,xmmE
- punpcklbw xmmE,xmmF ; xmmE=(20 22 24 26 28 2A 2C 2E)
- punpckhbw xmmG,xmmF ; xmmG=(30 32 34 36 38 3A 3C 3E)
-
- punpcklbw xmmF,xmmH
- punpckhbw xmmH,xmmH
- psrlw xmmF,BYTE_BIT ; xmmF=(21 23 25 27 29 2B 2D 2F)
- psrlw xmmH,BYTE_BIT ; xmmH=(31 33 35 37 39 3B 3D 3F)
-
-%endif ; RGB_PIXELSIZE ; ---------------
-
- ; xmm0=R(02468ACE)=RE, xmm2=G(02468ACE)=GE, xmm4=B(02468ACE)=BE
- ; xmm1=R(13579BDF)=RO, xmm3=G(13579BDF)=GO, xmm5=B(13579BDF)=BO
-
- ; (Original)
- ; Y = 0.29900 * R + 0.58700 * G + 0.11400 * B
- ; Cb = -0.16874 * R - 0.33126 * G + 0.50000 * B + CENTERJSAMPLE
- ; Cr = 0.50000 * R - 0.41869 * G - 0.08131 * B + CENTERJSAMPLE
- ;
- ; (This implementation)
- ; Y = 0.29900 * R + 0.33700 * G + 0.11400 * B + 0.25000 * G
- ; Cb = -0.16874 * R - 0.33126 * G + 0.50000 * B + CENTERJSAMPLE
- ; Cr = 0.50000 * R - 0.41869 * G - 0.08131 * B + CENTERJSAMPLE
-
- movdqa XMMWORD [wk(0)], xmm0 ; wk(0)=RE
- movdqa XMMWORD [wk(1)], xmm1 ; wk(1)=RO
- movdqa XMMWORD [wk(2)], xmm4 ; wk(2)=BE
- movdqa XMMWORD [wk(3)], xmm5 ; wk(3)=BO
-
- movdqa xmm6,xmm1
- punpcklwd xmm1,xmm3
- punpckhwd xmm6,xmm3
- movdqa xmm7,xmm1
- movdqa xmm4,xmm6
- pmaddwd xmm1,[GOTOFF(eax,PW_F0299_F0337)] ; xmm1=ROL*FIX(0.299)+GOL*FIX(0.337)
- pmaddwd xmm6,[GOTOFF(eax,PW_F0299_F0337)] ; xmm6=ROH*FIX(0.299)+GOH*FIX(0.337)
- pmaddwd xmm7,[GOTOFF(eax,PW_MF016_MF033)] ; xmm7=ROL*-FIX(0.168)+GOL*-FIX(0.331)
- pmaddwd xmm4,[GOTOFF(eax,PW_MF016_MF033)] ; xmm4=ROH*-FIX(0.168)+GOH*-FIX(0.331)
-
- movdqa XMMWORD [wk(4)], xmm1 ; wk(4)=ROL*FIX(0.299)+GOL*FIX(0.337)
- movdqa XMMWORD [wk(5)], xmm6 ; wk(5)=ROH*FIX(0.299)+GOH*FIX(0.337)
-
- pxor xmm1,xmm1
- pxor xmm6,xmm6
- punpcklwd xmm1,xmm5 ; xmm1=BOL
- punpckhwd xmm6,xmm5 ; xmm6=BOH
- psrld xmm1,1 ; xmm1=BOL*FIX(0.500)
- psrld xmm6,1 ; xmm6=BOH*FIX(0.500)
-
- movdqa xmm5,[GOTOFF(eax,PD_ONEHALFM1_CJ)] ; xmm5=[PD_ONEHALFM1_CJ]
-
- paddd xmm7,xmm1
- paddd xmm4,xmm6
- paddd xmm7,xmm5
- paddd xmm4,xmm5
- psrld xmm7,SCALEBITS ; xmm7=CbOL
- psrld xmm4,SCALEBITS ; xmm4=CbOH
- packssdw xmm7,xmm4 ; xmm7=CbO
-
- movdqa xmm1, XMMWORD [wk(2)] ; xmm1=BE
-
- movdqa xmm6,xmm0
- punpcklwd xmm0,xmm2
- punpckhwd xmm6,xmm2
- movdqa xmm5,xmm0
- movdqa xmm4,xmm6
- pmaddwd xmm0,[GOTOFF(eax,PW_F0299_F0337)] ; xmm0=REL*FIX(0.299)+GEL*FIX(0.337)
- pmaddwd xmm6,[GOTOFF(eax,PW_F0299_F0337)] ; xmm6=REH*FIX(0.299)+GEH*FIX(0.337)
- pmaddwd xmm5,[GOTOFF(eax,PW_MF016_MF033)] ; xmm5=REL*-FIX(0.168)+GEL*-FIX(0.331)
- pmaddwd xmm4,[GOTOFF(eax,PW_MF016_MF033)] ; xmm4=REH*-FIX(0.168)+GEH*-FIX(0.331)
-
- movdqa XMMWORD [wk(6)], xmm0 ; wk(6)=REL*FIX(0.299)+GEL*FIX(0.337)
- movdqa XMMWORD [wk(7)], xmm6 ; wk(7)=REH*FIX(0.299)+GEH*FIX(0.337)
-
- pxor xmm0,xmm0
- pxor xmm6,xmm6
- punpcklwd xmm0,xmm1 ; xmm0=BEL
- punpckhwd xmm6,xmm1 ; xmm6=BEH
- psrld xmm0,1 ; xmm0=BEL*FIX(0.500)
- psrld xmm6,1 ; xmm6=BEH*FIX(0.500)
-
- movdqa xmm1,[GOTOFF(eax,PD_ONEHALFM1_CJ)] ; xmm1=[PD_ONEHALFM1_CJ]
-
- paddd xmm5,xmm0
- paddd xmm4,xmm6
- paddd xmm5,xmm1
- paddd xmm4,xmm1
- psrld xmm5,SCALEBITS ; xmm5=CbEL
- psrld xmm4,SCALEBITS ; xmm4=CbEH
- packssdw xmm5,xmm4 ; xmm5=CbE
-
- psllw xmm7,BYTE_BIT
- por xmm5,xmm7 ; xmm5=Cb
- movdqa XMMWORD [ebx], xmm5 ; Save Cb
-
- movdqa xmm0, XMMWORD [wk(3)] ; xmm0=BO
- movdqa xmm6, XMMWORD [wk(2)] ; xmm6=BE
- movdqa xmm1, XMMWORD [wk(1)] ; xmm1=RO
-
- movdqa xmm4,xmm0
- punpcklwd xmm0,xmm3
- punpckhwd xmm4,xmm3
- movdqa xmm7,xmm0
- movdqa xmm5,xmm4
- pmaddwd xmm0,[GOTOFF(eax,PW_F0114_F0250)] ; xmm0=BOL*FIX(0.114)+GOL*FIX(0.250)
- pmaddwd xmm4,[GOTOFF(eax,PW_F0114_F0250)] ; xmm4=BOH*FIX(0.114)+GOH*FIX(0.250)
- pmaddwd xmm7,[GOTOFF(eax,PW_MF008_MF041)] ; xmm7=BOL*-FIX(0.081)+GOL*-FIX(0.418)
- pmaddwd xmm5,[GOTOFF(eax,PW_MF008_MF041)] ; xmm5=BOH*-FIX(0.081)+GOH*-FIX(0.418)
-
- movdqa xmm3,[GOTOFF(eax,PD_ONEHALF)] ; xmm3=[PD_ONEHALF]
-
- paddd xmm0, XMMWORD [wk(4)]
- paddd xmm4, XMMWORD [wk(5)]
- paddd xmm0,xmm3
- paddd xmm4,xmm3
- psrld xmm0,SCALEBITS ; xmm0=YOL
- psrld xmm4,SCALEBITS ; xmm4=YOH
- packssdw xmm0,xmm4 ; xmm0=YO
-
- pxor xmm3,xmm3
- pxor xmm4,xmm4
- punpcklwd xmm3,xmm1 ; xmm3=ROL
- punpckhwd xmm4,xmm1 ; xmm4=ROH
- psrld xmm3,1 ; xmm3=ROL*FIX(0.500)
- psrld xmm4,1 ; xmm4=ROH*FIX(0.500)
-
- movdqa xmm1,[GOTOFF(eax,PD_ONEHALFM1_CJ)] ; xmm1=[PD_ONEHALFM1_CJ]
-
- paddd xmm7,xmm3
- paddd xmm5,xmm4
- paddd xmm7,xmm1
- paddd xmm5,xmm1
- psrld xmm7,SCALEBITS ; xmm7=CrOL
- psrld xmm5,SCALEBITS ; xmm5=CrOH
- packssdw xmm7,xmm5 ; xmm7=CrO
-
- movdqa xmm3, XMMWORD [wk(0)] ; xmm3=RE
-
- movdqa xmm4,xmm6
- punpcklwd xmm6,xmm2
- punpckhwd xmm4,xmm2
- movdqa xmm1,xmm6
- movdqa xmm5,xmm4
- pmaddwd xmm6,[GOTOFF(eax,PW_F0114_F0250)] ; xmm6=BEL*FIX(0.114)+GEL*FIX(0.250)
- pmaddwd xmm4,[GOTOFF(eax,PW_F0114_F0250)] ; xmm4=BEH*FIX(0.114)+GEH*FIX(0.250)
- pmaddwd xmm1,[GOTOFF(eax,PW_MF008_MF041)] ; xmm1=BEL*-FIX(0.081)+GEL*-FIX(0.418)
- pmaddwd xmm5,[GOTOFF(eax,PW_MF008_MF041)] ; xmm5=BEH*-FIX(0.081)+GEH*-FIX(0.418)
-
- movdqa xmm2,[GOTOFF(eax,PD_ONEHALF)] ; xmm2=[PD_ONEHALF]
-
- paddd xmm6, XMMWORD [wk(6)]
- paddd xmm4, XMMWORD [wk(7)]
- paddd xmm6,xmm2
- paddd xmm4,xmm2
- psrld xmm6,SCALEBITS ; xmm6=YEL
- psrld xmm4,SCALEBITS ; xmm4=YEH
- packssdw xmm6,xmm4 ; xmm6=YE
-
- psllw xmm0,BYTE_BIT
- por xmm6,xmm0 ; xmm6=Y
- movdqa XMMWORD [edi], xmm6 ; Save Y
-
- pxor xmm2,xmm2
- pxor xmm4,xmm4
- punpcklwd xmm2,xmm3 ; xmm2=REL
- punpckhwd xmm4,xmm3 ; xmm4=REH
- psrld xmm2,1 ; xmm2=REL*FIX(0.500)
- psrld xmm4,1 ; xmm4=REH*FIX(0.500)
-
- movdqa xmm0,[GOTOFF(eax,PD_ONEHALFM1_CJ)] ; xmm0=[PD_ONEHALFM1_CJ]
-
- paddd xmm1,xmm2
- paddd xmm5,xmm4
- paddd xmm1,xmm0
- paddd xmm5,xmm0
- psrld xmm1,SCALEBITS ; xmm1=CrEL
- psrld xmm5,SCALEBITS ; xmm5=CrEH
- packssdw xmm1,xmm5 ; xmm1=CrE
-
- psllw xmm7,BYTE_BIT
- por xmm1,xmm7 ; xmm1=Cr
- movdqa XMMWORD [edx], xmm1 ; Save Cr
-
- sub ecx, byte SIZEOF_XMMWORD
- add esi, byte RGB_PIXELSIZE*SIZEOF_XMMWORD ; inptr
- add edi, byte SIZEOF_XMMWORD ; outptr0
- add ebx, byte SIZEOF_XMMWORD ; outptr1
- add edx, byte SIZEOF_XMMWORD ; outptr2
- cmp ecx, byte SIZEOF_XMMWORD
- jae near .columnloop
- test ecx,ecx
- jnz near .column_ld1
-
- pop ecx ; col
- pop esi
- pop edi
- pop ebx
- pop edx
- poppic eax
-
- add esi, byte SIZEOF_JSAMPROW ; input_buf
- add edi, byte SIZEOF_JSAMPROW
- add ebx, byte SIZEOF_JSAMPROW
- add edx, byte SIZEOF_JSAMPROW
- dec eax ; num_rows
- jg near .rowloop
-
-.return:
- pop edi
- pop esi
-; pop edx ; need not be preserved
-; pop ecx ; need not be preserved
- pop ebx
- mov esp,ebp ; esp <- aligned ebp
- pop esp ; esp <- original ebp
- pop ebp
- ret
-
-; For some reason, the OS X linker does not honor the request to align the
-; segment unless we do this.
- align 16
diff --git a/simd/jccolor-sse2-64.asm b/simd/jccolor-sse2-64.asm
deleted file mode 100644
index bd2188b..0000000
--- a/simd/jccolor-sse2-64.asm
+++ /dev/null
@@ -1,121 +0,0 @@
-;
-; jccolor.asm - colorspace conversion (64-bit SSE2)
-;
-; Copyright (C) 2009, D. R. Commander.
-;
-; Based on the x86 SIMD extension for IJG JPEG library
-; Copyright (C) 1999-2006, MIYASAKA Masaru.
-; For conditions of distribution and use, see copyright notice in jsimdext.inc
-;
-; This file should be assembled with NASM (Netwide Assembler),
-; can *not* be assembled with Microsoft's MASM or any compatible
-; assembler (including Borland's Turbo Assembler).
-; NASM is available from http://nasm.sourceforge.net/ or
-; http://sourceforge.net/project/showfiles.php?group_id=6208
-;
-; [TAB8]
-
-%include "jsimdext.inc"
-
-; --------------------------------------------------------------------------
-
-%define SCALEBITS 16
-
-F_0_081 equ 5329 ; FIX(0.08131)
-F_0_114 equ 7471 ; FIX(0.11400)
-F_0_168 equ 11059 ; FIX(0.16874)
-F_0_250 equ 16384 ; FIX(0.25000)
-F_0_299 equ 19595 ; FIX(0.29900)
-F_0_331 equ 21709 ; FIX(0.33126)
-F_0_418 equ 27439 ; FIX(0.41869)
-F_0_587 equ 38470 ; FIX(0.58700)
-F_0_337 equ (F_0_587 - F_0_250) ; FIX(0.58700) - FIX(0.25000)
-
-; --------------------------------------------------------------------------
- SECTION SEG_CONST
-
- alignz 16
- global EXTN(jconst_rgb_ycc_convert_sse2)
-
-EXTN(jconst_rgb_ycc_convert_sse2):
-
-PW_F0299_F0337 times 4 dw F_0_299, F_0_337
-PW_F0114_F0250 times 4 dw F_0_114, F_0_250
-PW_MF016_MF033 times 4 dw -F_0_168,-F_0_331
-PW_MF008_MF041 times 4 dw -F_0_081,-F_0_418
-PD_ONEHALFM1_CJ times 4 dd (1 << (SCALEBITS-1)) - 1 + (CENTERJSAMPLE << SCALEBITS)
-PD_ONEHALF times 4 dd (1 << (SCALEBITS-1))
-
- alignz 16
-
-; --------------------------------------------------------------------------
- SECTION SEG_TEXT
- BITS 64
-
-%include "jccolext-sse2-64.asm"
-
-%undef RGB_RED
-%undef RGB_GREEN
-%undef RGB_BLUE
-%undef RGB_PIXELSIZE
-%define RGB_RED EXT_RGB_RED
-%define RGB_GREEN EXT_RGB_GREEN
-%define RGB_BLUE EXT_RGB_BLUE
-%define RGB_PIXELSIZE EXT_RGB_PIXELSIZE
-%define jsimd_rgb_ycc_convert_sse2 jsimd_extrgb_ycc_convert_sse2
-%include "jccolext-sse2-64.asm"
-
-%undef RGB_RED
-%undef RGB_GREEN
-%undef RGB_BLUE
-%undef RGB_PIXELSIZE
-%define RGB_RED EXT_RGBX_RED
-%define RGB_GREEN EXT_RGBX_GREEN
-%define RGB_BLUE EXT_RGBX_BLUE
-%define RGB_PIXELSIZE EXT_RGBX_PIXELSIZE
-%define jsimd_rgb_ycc_convert_sse2 jsimd_extrgbx_ycc_convert_sse2
-%include "jccolext-sse2-64.asm"
-
-%undef RGB_RED
-%undef RGB_GREEN
-%undef RGB_BLUE
-%undef RGB_PIXELSIZE
-%define RGB_RED EXT_BGR_RED
-%define RGB_GREEN EXT_BGR_GREEN
-%define RGB_BLUE EXT_BGR_BLUE
-%define RGB_PIXELSIZE EXT_BGR_PIXELSIZE
-%define jsimd_rgb_ycc_convert_sse2 jsimd_extbgr_ycc_convert_sse2
-%include "jccolext-sse2-64.asm"
-
-%undef RGB_RED
-%undef RGB_GREEN
-%undef RGB_BLUE
-%undef RGB_PIXELSIZE
-%define RGB_RED EXT_BGRX_RED
-%define RGB_GREEN EXT_BGRX_GREEN
-%define RGB_BLUE EXT_BGRX_BLUE
-%define RGB_PIXELSIZE EXT_BGRX_PIXELSIZE
-%define jsimd_rgb_ycc_convert_sse2 jsimd_extbgrx_ycc_convert_sse2
-%include "jccolext-sse2-64.asm"
-
-%undef RGB_RED
-%undef RGB_GREEN
-%undef RGB_BLUE
-%undef RGB_PIXELSIZE
-%define RGB_RED EXT_XBGR_RED
-%define RGB_GREEN EXT_XBGR_GREEN
-%define RGB_BLUE EXT_XBGR_BLUE
-%define RGB_PIXELSIZE EXT_XBGR_PIXELSIZE
-%define jsimd_rgb_ycc_convert_sse2 jsimd_extxbgr_ycc_convert_sse2
-%include "jccolext-sse2-64.asm"
-
-%undef RGB_RED
-%undef RGB_GREEN
-%undef RGB_BLUE
-%undef RGB_PIXELSIZE
-%define RGB_RED EXT_XRGB_RED
-%define RGB_GREEN EXT_XRGB_GREEN
-%define RGB_BLUE EXT_XRGB_BLUE
-%define RGB_PIXELSIZE EXT_XRGB_PIXELSIZE
-%define jsimd_rgb_ycc_convert_sse2 jsimd_extxrgb_ycc_convert_sse2
-%include "jccolext-sse2-64.asm"
diff --git a/simd/jcgray-sse2-64.asm b/simd/jcgray-sse2-64.asm
deleted file mode 100644
index bafd302..0000000
--- a/simd/jcgray-sse2-64.asm
+++ /dev/null
@@ -1,114 +0,0 @@
-;
-; jcgray.asm - grayscale colorspace conversion (64-bit SSE2)
-;
-; Copyright (C) 2011, D. R. Commander.
-;
-; Based on the x86 SIMD extension for IJG JPEG library
-; Copyright (C) 1999-2006, MIYASAKA Masaru.
-; For conditions of distribution and use, see copyright notice in jsimdext.inc
-;
-; This file should be assembled with NASM (Netwide Assembler),
-; can *not* be assembled with Microsoft's MASM or any compatible
-; assembler (including Borland's Turbo Assembler).
-; NASM is available from http://nasm.sourceforge.net/ or
-; http://sourceforge.net/project/showfiles.php?group_id=6208
-;
-; [TAB8]
-
-%include "jsimdext.inc"
-
-; --------------------------------------------------------------------------
-
-%define SCALEBITS 16
-
-F_0_114 equ 7471 ; FIX(0.11400)
-F_0_250 equ 16384 ; FIX(0.25000)
-F_0_299 equ 19595 ; FIX(0.29900)
-F_0_587 equ 38470 ; FIX(0.58700)
-F_0_337 equ (F_0_587 - F_0_250) ; FIX(0.58700) - FIX(0.25000)
-
-; --------------------------------------------------------------------------
- SECTION SEG_CONST
-
- alignz 16
- global EXTN(jconst_rgb_gray_convert_sse2)
-
-EXTN(jconst_rgb_gray_convert_sse2):
-
-PW_F0299_F0337 times 4 dw F_0_299, F_0_337
-PW_F0114_F0250 times 4 dw F_0_114, F_0_250
-PD_ONEHALF times 4 dd (1 << (SCALEBITS-1))
-
- alignz 16
-
-; --------------------------------------------------------------------------
- SECTION SEG_TEXT
- BITS 64
-
-%include "jcgryext-sse2-64.asm"
-
-%undef RGB_RED
-%undef RGB_GREEN
-%undef RGB_BLUE
-%undef RGB_PIXELSIZE
-%define RGB_RED EXT_RGB_RED
-%define RGB_GREEN EXT_RGB_GREEN
-%define RGB_BLUE EXT_RGB_BLUE
-%define RGB_PIXELSIZE EXT_RGB_PIXELSIZE
-%define jsimd_rgb_gray_convert_sse2 jsimd_extrgb_gray_convert_sse2
-%include "jcgryext-sse2-64.asm"
-
-%undef RGB_RED
-%undef RGB_GREEN
-%undef RGB_BLUE
-%undef RGB_PIXELSIZE
-%define RGB_RED EXT_RGBX_RED
-%define RGB_GREEN EXT_RGBX_GREEN
-%define RGB_BLUE EXT_RGBX_BLUE
-%define RGB_PIXELSIZE EXT_RGBX_PIXELSIZE
-%define jsimd_rgb_gray_convert_sse2 jsimd_extrgbx_gray_convert_sse2
-%include "jcgryext-sse2-64.asm"
-
-%undef RGB_RED
-%undef RGB_GREEN
-%undef RGB_BLUE
-%undef RGB_PIXELSIZE
-%define RGB_RED EXT_BGR_RED
-%define RGB_GREEN EXT_BGR_GREEN
-%define RGB_BLUE EXT_BGR_BLUE
-%define RGB_PIXELSIZE EXT_BGR_PIXELSIZE
-%define jsimd_rgb_gray_convert_sse2 jsimd_extbgr_gray_convert_sse2
-%include "jcgryext-sse2-64.asm"
-
-%undef RGB_RED
-%undef RGB_GREEN
-%undef RGB_BLUE
-%undef RGB_PIXELSIZE
-%define RGB_RED EXT_BGRX_RED
-%define RGB_GREEN EXT_BGRX_GREEN
-%define RGB_BLUE EXT_BGRX_BLUE
-%define RGB_PIXELSIZE EXT_BGRX_PIXELSIZE
-%define jsimd_rgb_gray_convert_sse2 jsimd_extbgrx_gray_convert_sse2
-%include "jcgryext-sse2-64.asm"
-
-%undef RGB_RED
-%undef RGB_GREEN
-%undef RGB_BLUE
-%undef RGB_PIXELSIZE
-%define RGB_RED EXT_XBGR_RED
-%define RGB_GREEN EXT_XBGR_GREEN
-%define RGB_BLUE EXT_XBGR_BLUE
-%define RGB_PIXELSIZE EXT_XBGR_PIXELSIZE
-%define jsimd_rgb_gray_convert_sse2 jsimd_extxbgr_gray_convert_sse2
-%include "jcgryext-sse2-64.asm"
-
-%undef RGB_RED
-%undef RGB_GREEN
-%undef RGB_BLUE
-%undef RGB_PIXELSIZE
-%define RGB_RED EXT_XRGB_RED
-%define RGB_GREEN EXT_XRGB_GREEN
-%define RGB_BLUE EXT_XRGB_BLUE
-%define RGB_PIXELSIZE EXT_XRGB_PIXELSIZE
-%define jsimd_rgb_gray_convert_sse2 jsimd_extxrgb_gray_convert_sse2
-%include "jcgryext-sse2-64.asm"
diff --git a/simd/jcgryext-sse2-64.asm b/simd/jcgryext-sse2-64.asm
deleted file mode 100644
index 541355a..0000000
--- a/simd/jcgryext-sse2-64.asm
+++ /dev/null
@@ -1,365 +0,0 @@
-;
-; jcgryext.asm - grayscale colorspace conversion (64-bit SSE2)
-;
-; Copyright (C) 2011, D. R. Commander.
-;
-; Based on the x86 SIMD extension for IJG JPEG library
-; Copyright (C) 1999-2006, MIYASAKA Masaru.
-; For conditions of distribution and use, see copyright notice in jsimdext.inc
-;
-; This file should be assembled with NASM (Netwide Assembler),
-; can *not* be assembled with Microsoft's MASM or any compatible
-; assembler (including Borland's Turbo Assembler).
-; NASM is available from http://nasm.sourceforge.net/ or
-; http://sourceforge.net/project/showfiles.php?group_id=6208
-;
-; [TAB8]
-
-%include "jcolsamp.inc"
-
-; --------------------------------------------------------------------------
-;
-; Convert some rows of samples to the output colorspace.
-;
-; GLOBAL(void)
-; jsimd_rgb_gray_convert_sse2 (JDIMENSION img_width,
-; JSAMPARRAY input_buf, JSAMPIMAGE output_buf,
-; JDIMENSION output_row, int num_rows);
-;
-
-; r10 = JDIMENSION img_width
-; r11 = JSAMPARRAY input_buf
-; r12 = JSAMPIMAGE output_buf
-; r13 = JDIMENSION output_row
-; r14 = int num_rows
-
-%define wk(i) rbp-(WK_NUM-(i))*SIZEOF_XMMWORD ; xmmword wk[WK_NUM]
-%define WK_NUM 2
-
- align 16
-
- global EXTN(jsimd_rgb_gray_convert_sse2)
-
-EXTN(jsimd_rgb_gray_convert_sse2):
- push rbp
- mov rax,rsp ; rax = original rbp
- sub rsp, byte 4
- and rsp, byte (-SIZEOF_XMMWORD) ; align to 128 bits
- mov [rsp],rax
- mov rbp,rsp ; rbp = aligned rbp
- lea rsp, [wk(0)]
- collect_args
- push rbx
-
- mov ecx, r10d
- test rcx,rcx
- jz near .return
-
- push rcx
-
- mov rsi, r12
- mov ecx, r13d
- mov rdi, JSAMPARRAY [rsi+0*SIZEOF_JSAMPARRAY]
- lea rdi, [rdi+rcx*SIZEOF_JSAMPROW]
-
- pop rcx
-
- mov rsi, r11
- mov eax, r14d
- test rax,rax
- jle near .return
-.rowloop:
- push rdi
- push rsi
- push rcx ; col
-
- mov rsi, JSAMPROW [rsi] ; inptr
- mov rdi, JSAMPROW [rdi] ; outptr0
-
- cmp rcx, byte SIZEOF_XMMWORD
- jae near .columnloop
-
-%if RGB_PIXELSIZE == 3 ; ---------------
-
-.column_ld1:
- push rax
- push rdx
- lea rcx,[rcx+rcx*2] ; imul ecx,RGB_PIXELSIZE
- test cl, SIZEOF_BYTE
- jz short .column_ld2
- sub rcx, byte SIZEOF_BYTE
- movzx rax, BYTE [rsi+rcx]
-.column_ld2:
- test cl, SIZEOF_WORD
- jz short .column_ld4
- sub rcx, byte SIZEOF_WORD
- movzx rdx, WORD [rsi+rcx]
- shl rax, WORD_BIT
- or rax,rdx
-.column_ld4:
- movd xmmA,eax
- pop rdx
- pop rax
- test cl, SIZEOF_DWORD
- jz short .column_ld8
- sub rcx, byte SIZEOF_DWORD
- movd xmmF, XMM_DWORD [rsi+rcx]
- pslldq xmmA, SIZEOF_DWORD
- por xmmA,xmmF
-.column_ld8:
- test cl, SIZEOF_MMWORD
- jz short .column_ld16
- sub rcx, byte SIZEOF_MMWORD
- movq xmmB, XMM_MMWORD [rsi+rcx]
- pslldq xmmA, SIZEOF_MMWORD
- por xmmA,xmmB
-.column_ld16:
- test cl, SIZEOF_XMMWORD
- jz short .column_ld32
- movdqa xmmF,xmmA
- movdqu xmmA, XMMWORD [rsi+0*SIZEOF_XMMWORD]
- mov rcx, SIZEOF_XMMWORD
- jmp short .rgb_gray_cnv
-.column_ld32:
- test cl, 2*SIZEOF_XMMWORD
- mov rcx, SIZEOF_XMMWORD
- jz short .rgb_gray_cnv
- movdqa xmmB,xmmA
- movdqu xmmA, XMMWORD [rsi+0*SIZEOF_XMMWORD]
- movdqu xmmF, XMMWORD [rsi+1*SIZEOF_XMMWORD]
- jmp short .rgb_gray_cnv
-
-.columnloop:
- movdqu xmmA, XMMWORD [rsi+0*SIZEOF_XMMWORD]
- movdqu xmmF, XMMWORD [rsi+1*SIZEOF_XMMWORD]
- movdqu xmmB, XMMWORD [rsi+2*SIZEOF_XMMWORD]
-
-.rgb_gray_cnv:
- ; xmmA=(00 10 20 01 11 21 02 12 22 03 13 23 04 14 24 05)
- ; xmmF=(15 25 06 16 26 07 17 27 08 18 28 09 19 29 0A 1A)
- ; xmmB=(2A 0B 1B 2B 0C 1C 2C 0D 1D 2D 0E 1E 2E 0F 1F 2F)
-
- movdqa xmmG,xmmA
- pslldq xmmA,8 ; xmmA=(-- -- -- -- -- -- -- -- 00 10 20 01 11 21 02 12)
- psrldq xmmG,8 ; xmmG=(22 03 13 23 04 14 24 05 -- -- -- -- -- -- -- --)
-
- punpckhbw xmmA,xmmF ; xmmA=(00 08 10 18 20 28 01 09 11 19 21 29 02 0A 12 1A)
- pslldq xmmF,8 ; xmmF=(-- -- -- -- -- -- -- -- 15 25 06 16 26 07 17 27)
-
- punpcklbw xmmG,xmmB ; xmmG=(22 2A 03 0B 13 1B 23 2B 04 0C 14 1C 24 2C 05 0D)
- punpckhbw xmmF,xmmB ; xmmF=(15 1D 25 2D 06 0E 16 1E 26 2E 07 0F 17 1F 27 2F)
-
- movdqa xmmD,xmmA
- pslldq xmmA,8 ; xmmA=(-- -- -- -- -- -- -- -- 00 08 10 18 20 28 01 09)
- psrldq xmmD,8 ; xmmD=(11 19 21 29 02 0A 12 1A -- -- -- -- -- -- -- --)
-
- punpckhbw xmmA,xmmG ; xmmA=(00 04 08 0C 10 14 18 1C 20 24 28 2C 01 05 09 0D)
- pslldq xmmG,8 ; xmmG=(-- -- -- -- -- -- -- -- 22 2A 03 0B 13 1B 23 2B)
-
- punpcklbw xmmD,xmmF ; xmmD=(11 15 19 1D 21 25 29 2D 02 06 0A 0E 12 16 1A 1E)
- punpckhbw xmmG,xmmF ; xmmG=(22 26 2A 2E 03 07 0B 0F 13 17 1B 1F 23 27 2B 2F)
-
- movdqa xmmE,xmmA
- pslldq xmmA,8 ; xmmA=(-- -- -- -- -- -- -- -- 00 04 08 0C 10 14 18 1C)
- psrldq xmmE,8 ; xmmE=(20 24 28 2C 01 05 09 0D -- -- -- -- -- -- -- --)
-
- punpckhbw xmmA,xmmD ; xmmA=(00 02 04 06 08 0A 0C 0E 10 12 14 16 18 1A 1C 1E)
- pslldq xmmD,8 ; xmmD=(-- -- -- -- -- -- -- -- 11 15 19 1D 21 25 29 2D)
-
- punpcklbw xmmE,xmmG ; xmmE=(20 22 24 26 28 2A 2C 2E 01 03 05 07 09 0B 0D 0F)
- punpckhbw xmmD,xmmG ; xmmD=(11 13 15 17 19 1B 1D 1F 21 23 25 27 29 2B 2D 2F)
-
- pxor xmmH,xmmH
-
- movdqa xmmC,xmmA
- punpcklbw xmmA,xmmH ; xmmA=(00 02 04 06 08 0A 0C 0E)
- punpckhbw xmmC,xmmH ; xmmC=(10 12 14 16 18 1A 1C 1E)
-
- movdqa xmmB,xmmE
- punpcklbw xmmE,xmmH ; xmmE=(20 22 24 26 28 2A 2C 2E)
- punpckhbw xmmB,xmmH ; xmmB=(01 03 05 07 09 0B 0D 0F)
-
- movdqa xmmF,xmmD
- punpcklbw xmmD,xmmH ; xmmD=(11 13 15 17 19 1B 1D 1F)
- punpckhbw xmmF,xmmH ; xmmF=(21 23 25 27 29 2B 2D 2F)
-
-%else ; RGB_PIXELSIZE == 4 ; -----------
-
-.column_ld1:
- test cl, SIZEOF_XMMWORD/16
- jz short .column_ld2
- sub rcx, byte SIZEOF_XMMWORD/16
- movd xmmA, XMM_DWORD [rsi+rcx*RGB_PIXELSIZE]
-.column_ld2:
- test cl, SIZEOF_XMMWORD/8
- jz short .column_ld4
- sub rcx, byte SIZEOF_XMMWORD/8
- movq xmmE, XMM_MMWORD [rsi+rcx*RGB_PIXELSIZE]
- pslldq xmmA, SIZEOF_MMWORD
- por xmmA,xmmE
-.column_ld4:
- test cl, SIZEOF_XMMWORD/4
- jz short .column_ld8
- sub rcx, byte SIZEOF_XMMWORD/4
- movdqa xmmE,xmmA
- movdqu xmmA, XMMWORD [rsi+rcx*RGB_PIXELSIZE]
-.column_ld8:
- test cl, SIZEOF_XMMWORD/2
- mov rcx, SIZEOF_XMMWORD
- jz short .rgb_gray_cnv
- movdqa xmmF,xmmA
- movdqa xmmH,xmmE
- movdqu xmmA, XMMWORD [rsi+0*SIZEOF_XMMWORD]
- movdqu xmmE, XMMWORD [rsi+1*SIZEOF_XMMWORD]
- jmp short .rgb_gray_cnv
-
-.columnloop:
- movdqu xmmA, XMMWORD [rsi+0*SIZEOF_XMMWORD]
- movdqu xmmE, XMMWORD [rsi+1*SIZEOF_XMMWORD]
- movdqu xmmF, XMMWORD [rsi+2*SIZEOF_XMMWORD]
- movdqu xmmH, XMMWORD [rsi+3*SIZEOF_XMMWORD]
-
-.rgb_gray_cnv:
- ; xmmA=(00 10 20 30 01 11 21 31 02 12 22 32 03 13 23 33)
- ; xmmE=(04 14 24 34 05 15 25 35 06 16 26 36 07 17 27 37)
- ; xmmF=(08 18 28 38 09 19 29 39 0A 1A 2A 3A 0B 1B 2B 3B)
- ; xmmH=(0C 1C 2C 3C 0D 1D 2D 3D 0E 1E 2E 3E 0F 1F 2F 3F)
-
- movdqa xmmD,xmmA
- punpcklbw xmmA,xmmE ; xmmA=(00 04 10 14 20 24 30 34 01 05 11 15 21 25 31 35)
- punpckhbw xmmD,xmmE ; xmmD=(02 06 12 16 22 26 32 36 03 07 13 17 23 27 33 37)
-
- movdqa xmmC,xmmF
- punpcklbw xmmF,xmmH ; xmmF=(08 0C 18 1C 28 2C 38 3C 09 0D 19 1D 29 2D 39 3D)
- punpckhbw xmmC,xmmH ; xmmC=(0A 0E 1A 1E 2A 2E 3A 3E 0B 0F 1B 1F 2B 2F 3B 3F)
-
- movdqa xmmB,xmmA
- punpcklwd xmmA,xmmF ; xmmA=(00 04 08 0C 10 14 18 1C 20 24 28 2C 30 34 38 3C)
- punpckhwd xmmB,xmmF ; xmmB=(01 05 09 0D 11 15 19 1D 21 25 29 2D 31 35 39 3D)
-
- movdqa xmmG,xmmD
- punpcklwd xmmD,xmmC ; xmmD=(02 06 0A 0E 12 16 1A 1E 22 26 2A 2E 32 36 3A 3E)
- punpckhwd xmmG,xmmC ; xmmG=(03 07 0B 0F 13 17 1B 1F 23 27 2B 2F 33 37 3B 3F)
-
- movdqa xmmE,xmmA
- punpcklbw xmmA,xmmD ; xmmA=(00 02 04 06 08 0A 0C 0E 10 12 14 16 18 1A 1C 1E)
- punpckhbw xmmE,xmmD ; xmmE=(20 22 24 26 28 2A 2C 2E 30 32 34 36 38 3A 3C 3E)
-
- movdqa xmmH,xmmB
- punpcklbw xmmB,xmmG ; xmmB=(01 03 05 07 09 0B 0D 0F 11 13 15 17 19 1B 1D 1F)
- punpckhbw xmmH,xmmG ; xmmH=(21 23 25 27 29 2B 2D 2F 31 33 35 37 39 3B 3D 3F)
-
- pxor xmmF,xmmF
-
- movdqa xmmC,xmmA
- punpcklbw xmmA,xmmF ; xmmA=(00 02 04 06 08 0A 0C 0E)
- punpckhbw xmmC,xmmF ; xmmC=(10 12 14 16 18 1A 1C 1E)
-
- movdqa xmmD,xmmB
- punpcklbw xmmB,xmmF ; xmmB=(01 03 05 07 09 0B 0D 0F)
- punpckhbw xmmD,xmmF ; xmmD=(11 13 15 17 19 1B 1D 1F)
-
- movdqa xmmG,xmmE
- punpcklbw xmmE,xmmF ; xmmE=(20 22 24 26 28 2A 2C 2E)
- punpckhbw xmmG,xmmF ; xmmG=(30 32 34 36 38 3A 3C 3E)
-
- punpcklbw xmmF,xmmH
- punpckhbw xmmH,xmmH
- psrlw xmmF,BYTE_BIT ; xmmF=(21 23 25 27 29 2B 2D 2F)
- psrlw xmmH,BYTE_BIT ; xmmH=(31 33 35 37 39 3B 3D 3F)
-
-%endif ; RGB_PIXELSIZE ; ---------------
-
- ; xmm0=R(02468ACE)=RE, xmm2=G(02468ACE)=GE, xmm4=B(02468ACE)=BE
- ; xmm1=R(13579BDF)=RO, xmm3=G(13579BDF)=GO, xmm5=B(13579BDF)=BO
-
- ; (Original)
- ; Y = 0.29900 * R + 0.58700 * G + 0.11400 * B
- ;
- ; (This implementation)
- ; Y = 0.29900 * R + 0.33700 * G + 0.11400 * B + 0.25000 * G
-
- movdqa xmm6,xmm1
- punpcklwd xmm1,xmm3
- punpckhwd xmm6,xmm3
- pmaddwd xmm1,[rel PW_F0299_F0337] ; xmm1=ROL*FIX(0.299)+GOL*FIX(0.337)
- pmaddwd xmm6,[rel PW_F0299_F0337] ; xmm6=ROH*FIX(0.299)+GOH*FIX(0.337)
-
- movdqa xmm7, xmm6 ; xmm7=ROH*FIX(0.299)+GOH*FIX(0.337)
-
- movdqa xmm6,xmm0
- punpcklwd xmm0,xmm2
- punpckhwd xmm6,xmm2
- pmaddwd xmm0,[rel PW_F0299_F0337] ; xmm0=REL*FIX(0.299)+GEL*FIX(0.337)
- pmaddwd xmm6,[rel PW_F0299_F0337] ; xmm6=REH*FIX(0.299)+GEH*FIX(0.337)
-
- movdqa XMMWORD [wk(0)], xmm0 ; wk(0)=REL*FIX(0.299)+GEL*FIX(0.337)
- movdqa XMMWORD [wk(1)], xmm6 ; wk(1)=REH*FIX(0.299)+GEH*FIX(0.337)
-
- movdqa xmm0, xmm5 ; xmm0=BO
- movdqa xmm6, xmm4 ; xmm6=BE
-
- movdqa xmm4,xmm0
- punpcklwd xmm0,xmm3
- punpckhwd xmm4,xmm3
- pmaddwd xmm0,[rel PW_F0114_F0250] ; xmm0=BOL*FIX(0.114)+GOL*FIX(0.250)
- pmaddwd xmm4,[rel PW_F0114_F0250] ; xmm4=BOH*FIX(0.114)+GOH*FIX(0.250)
-
- movdqa xmm3,[rel PD_ONEHALF] ; xmm3=[PD_ONEHALF]
-
- paddd xmm0, xmm1
- paddd xmm4, xmm7
- paddd xmm0,xmm3
- paddd xmm4,xmm3
- psrld xmm0,SCALEBITS ; xmm0=YOL
- psrld xmm4,SCALEBITS ; xmm4=YOH
- packssdw xmm0,xmm4 ; xmm0=YO
-
- movdqa xmm4,xmm6
- punpcklwd xmm6,xmm2
- punpckhwd xmm4,xmm2
- pmaddwd xmm6,[rel PW_F0114_F0250] ; xmm6=BEL*FIX(0.114)+GEL*FIX(0.250)
- pmaddwd xmm4,[rel PW_F0114_F0250] ; xmm4=BEH*FIX(0.114)+GEH*FIX(0.250)
-
- movdqa xmm2,[rel PD_ONEHALF] ; xmm2=[PD_ONEHALF]
-
- paddd xmm6, XMMWORD [wk(0)]
- paddd xmm4, XMMWORD [wk(1)]
- paddd xmm6,xmm2
- paddd xmm4,xmm2
- psrld xmm6,SCALEBITS ; xmm6=YEL
- psrld xmm4,SCALEBITS ; xmm4=YEH
- packssdw xmm6,xmm4 ; xmm6=YE
-
- psllw xmm0,BYTE_BIT
- por xmm6,xmm0 ; xmm6=Y
- movdqa XMMWORD [rdi], xmm6 ; Save Y
-
- sub rcx, byte SIZEOF_XMMWORD
- add rsi, byte RGB_PIXELSIZE*SIZEOF_XMMWORD ; inptr
- add rdi, byte SIZEOF_XMMWORD ; outptr0
- cmp rcx, byte SIZEOF_XMMWORD
- jae near .columnloop
- test rcx,rcx
- jnz near .column_ld1
-
- pop rcx ; col
- pop rsi
- pop rdi
-
- add rsi, byte SIZEOF_JSAMPROW ; input_buf
- add rdi, byte SIZEOF_JSAMPROW
- dec rax ; num_rows
- jg near .rowloop
-
-.return:
- pop rbx
- uncollect_args
- mov rsp,rbp ; rsp <- aligned rbp
- pop rsp ; rsp <- original rbp
- pop rbp
- ret
-
-; For some reason, the OS X linker does not honor the request to align the
-; segment unless we do this.
- align 16
diff --git a/simd/jcgryext-sse2.asm b/simd/jcgryext-sse2.asm
deleted file mode 100644
index cd16dd1..0000000
--- a/simd/jcgryext-sse2.asm
+++ /dev/null
@@ -1,384 +0,0 @@
-;
-; jcgryext.asm - grayscale colorspace conversion (SSE2)
-;
-; Copyright (C) 2011, D. R. Commander.
-;
-; Based on the x86 SIMD extension for IJG JPEG library
-; Copyright (C) 1999-2006, MIYASAKA Masaru.
-; For conditions of distribution and use, see copyright notice in jsimdext.inc
-;
-; This file should be assembled with NASM (Netwide Assembler),
-; can *not* be assembled with Microsoft's MASM or any compatible
-; assembler (including Borland's Turbo Assembler).
-; NASM is available from http://nasm.sourceforge.net/ or
-; http://sourceforge.net/project/showfiles.php?group_id=6208
-;
-; [TAB8]
-
-%include "jcolsamp.inc"
-
-; --------------------------------------------------------------------------
-;
-; Convert some rows of samples to the output colorspace.
-;
-; GLOBAL(void)
-; jsimd_rgb_gray_convert_sse2 (JDIMENSION img_width,
-; JSAMPARRAY input_buf, JSAMPIMAGE output_buf,
-; JDIMENSION output_row, int num_rows);
-;
-
-%define img_width(b) (b)+8 ; JDIMENSION img_width
-%define input_buf(b) (b)+12 ; JSAMPARRAY input_buf
-%define output_buf(b) (b)+16 ; JSAMPIMAGE output_buf
-%define output_row(b) (b)+20 ; JDIMENSION output_row
-%define num_rows(b) (b)+24 ; int num_rows
-
-%define original_ebp ebp+0
-%define wk(i) ebp-(WK_NUM-(i))*SIZEOF_XMMWORD ; xmmword wk[WK_NUM]
-%define WK_NUM 2
-%define gotptr wk(0)-SIZEOF_POINTER ; void * gotptr
-
- align 16
-
- global EXTN(jsimd_rgb_gray_convert_sse2)
-
-EXTN(jsimd_rgb_gray_convert_sse2):
- push ebp
- mov eax,esp ; eax = original ebp
- sub esp, byte 4
- and esp, byte (-SIZEOF_XMMWORD) ; align to 128 bits
- mov [esp],eax
- mov ebp,esp ; ebp = aligned ebp
- lea esp, [wk(0)]
- pushpic eax ; make a room for GOT address
- push ebx
-; push ecx ; need not be preserved
-; push edx ; need not be preserved
- push esi
- push edi
-
- get_GOT ebx ; get GOT address
- movpic POINTER [gotptr], ebx ; save GOT address
-
- mov ecx, JDIMENSION [img_width(eax)]
- test ecx,ecx
- jz near .return
-
- push ecx
-
- mov esi, JSAMPIMAGE [output_buf(eax)]
- mov ecx, JDIMENSION [output_row(eax)]
- mov edi, JSAMPARRAY [esi+0*SIZEOF_JSAMPARRAY]
- lea edi, [edi+ecx*SIZEOF_JSAMPROW]
-
- pop ecx
-
- mov esi, JSAMPARRAY [input_buf(eax)]
- mov eax, INT [num_rows(eax)]
- test eax,eax
- jle near .return
- alignx 16,7
-.rowloop:
- pushpic eax
- push edi
- push esi
- push ecx ; col
-
- mov esi, JSAMPROW [esi] ; inptr
- mov edi, JSAMPROW [edi] ; outptr0
- movpic eax, POINTER [gotptr] ; load GOT address (eax)
-
- cmp ecx, byte SIZEOF_XMMWORD
- jae near .columnloop
- alignx 16,7
-
-%if RGB_PIXELSIZE == 3 ; ---------------
-
-.column_ld1:
- push eax
- push edx
- lea ecx,[ecx+ecx*2] ; imul ecx,RGB_PIXELSIZE
- test cl, SIZEOF_BYTE
- jz short .column_ld2
- sub ecx, byte SIZEOF_BYTE
- movzx eax, BYTE [esi+ecx]
-.column_ld2:
- test cl, SIZEOF_WORD
- jz short .column_ld4
- sub ecx, byte SIZEOF_WORD
- movzx edx, WORD [esi+ecx]
- shl eax, WORD_BIT
- or eax,edx
-.column_ld4:
- movd xmmA,eax
- pop edx
- pop eax
- test cl, SIZEOF_DWORD
- jz short .column_ld8
- sub ecx, byte SIZEOF_DWORD
- movd xmmF, XMM_DWORD [esi+ecx]
- pslldq xmmA, SIZEOF_DWORD
- por xmmA,xmmF
-.column_ld8:
- test cl, SIZEOF_MMWORD
- jz short .column_ld16
- sub ecx, byte SIZEOF_MMWORD
- movq xmmB, XMM_MMWORD [esi+ecx]
- pslldq xmmA, SIZEOF_MMWORD
- por xmmA,xmmB
-.column_ld16:
- test cl, SIZEOF_XMMWORD
- jz short .column_ld32
- movdqa xmmF,xmmA
- movdqu xmmA, XMMWORD [esi+0*SIZEOF_XMMWORD]
- mov ecx, SIZEOF_XMMWORD
- jmp short .rgb_gray_cnv
-.column_ld32:
- test cl, 2*SIZEOF_XMMWORD
- mov ecx, SIZEOF_XMMWORD
- jz short .rgb_gray_cnv
- movdqa xmmB,xmmA
- movdqu xmmA, XMMWORD [esi+0*SIZEOF_XMMWORD]
- movdqu xmmF, XMMWORD [esi+1*SIZEOF_XMMWORD]
- jmp short .rgb_gray_cnv
- alignx 16,7
-
-.columnloop:
- movdqu xmmA, XMMWORD [esi+0*SIZEOF_XMMWORD]
- movdqu xmmF, XMMWORD [esi+1*SIZEOF_XMMWORD]
- movdqu xmmB, XMMWORD [esi+2*SIZEOF_XMMWORD]
-
-.rgb_gray_cnv:
- ; xmmA=(00 10 20 01 11 21 02 12 22 03 13 23 04 14 24 05)
- ; xmmF=(15 25 06 16 26 07 17 27 08 18 28 09 19 29 0A 1A)
- ; xmmB=(2A 0B 1B 2B 0C 1C 2C 0D 1D 2D 0E 1E 2E 0F 1F 2F)
-
- movdqa xmmG,xmmA
- pslldq xmmA,8 ; xmmA=(-- -- -- -- -- -- -- -- 00 10 20 01 11 21 02 12)
- psrldq xmmG,8 ; xmmG=(22 03 13 23 04 14 24 05 -- -- -- -- -- -- -- --)
-
- punpckhbw xmmA,xmmF ; xmmA=(00 08 10 18 20 28 01 09 11 19 21 29 02 0A 12 1A)
- pslldq xmmF,8 ; xmmF=(-- -- -- -- -- -- -- -- 15 25 06 16 26 07 17 27)
-
- punpcklbw xmmG,xmmB ; xmmG=(22 2A 03 0B 13 1B 23 2B 04 0C 14 1C 24 2C 05 0D)
- punpckhbw xmmF,xmmB ; xmmF=(15 1D 25 2D 06 0E 16 1E 26 2E 07 0F 17 1F 27 2F)
-
- movdqa xmmD,xmmA
- pslldq xmmA,8 ; xmmA=(-- -- -- -- -- -- -- -- 00 08 10 18 20 28 01 09)
- psrldq xmmD,8 ; xmmD=(11 19 21 29 02 0A 12 1A -- -- -- -- -- -- -- --)
-
- punpckhbw xmmA,xmmG ; xmmA=(00 04 08 0C 10 14 18 1C 20 24 28 2C 01 05 09 0D)
- pslldq xmmG,8 ; xmmG=(-- -- -- -- -- -- -- -- 22 2A 03 0B 13 1B 23 2B)
-
- punpcklbw xmmD,xmmF ; xmmD=(11 15 19 1D 21 25 29 2D 02 06 0A 0E 12 16 1A 1E)
- punpckhbw xmmG,xmmF ; xmmG=(22 26 2A 2E 03 07 0B 0F 13 17 1B 1F 23 27 2B 2F)
-
- movdqa xmmE,xmmA
- pslldq xmmA,8 ; xmmA=(-- -- -- -- -- -- -- -- 00 04 08 0C 10 14 18 1C)
- psrldq xmmE,8 ; xmmE=(20 24 28 2C 01 05 09 0D -- -- -- -- -- -- -- --)
-
- punpckhbw xmmA,xmmD ; xmmA=(00 02 04 06 08 0A 0C 0E 10 12 14 16 18 1A 1C 1E)
- pslldq xmmD,8 ; xmmD=(-- -- -- -- -- -- -- -- 11 15 19 1D 21 25 29 2D)
-
- punpcklbw xmmE,xmmG ; xmmE=(20 22 24 26 28 2A 2C 2E 01 03 05 07 09 0B 0D 0F)
- punpckhbw xmmD,xmmG ; xmmD=(11 13 15 17 19 1B 1D 1F 21 23 25 27 29 2B 2D 2F)
-
- pxor xmmH,xmmH
-
- movdqa xmmC,xmmA
- punpcklbw xmmA,xmmH ; xmmA=(00 02 04 06 08 0A 0C 0E)
- punpckhbw xmmC,xmmH ; xmmC=(10 12 14 16 18 1A 1C 1E)
-
- movdqa xmmB,xmmE
- punpcklbw xmmE,xmmH ; xmmE=(20 22 24 26 28 2A 2C 2E)
- punpckhbw xmmB,xmmH ; xmmB=(01 03 05 07 09 0B 0D 0F)
-
- movdqa xmmF,xmmD
- punpcklbw xmmD,xmmH ; xmmD=(11 13 15 17 19 1B 1D 1F)
- punpckhbw xmmF,xmmH ; xmmF=(21 23 25 27 29 2B 2D 2F)
-
-%else ; RGB_PIXELSIZE == 4 ; -----------
-
-.column_ld1:
- test cl, SIZEOF_XMMWORD/16
- jz short .column_ld2
- sub ecx, byte SIZEOF_XMMWORD/16
- movd xmmA, XMM_DWORD [esi+ecx*RGB_PIXELSIZE]
-.column_ld2:
- test cl, SIZEOF_XMMWORD/8
- jz short .column_ld4
- sub ecx, byte SIZEOF_XMMWORD/8
- movq xmmE, XMM_MMWORD [esi+ecx*RGB_PIXELSIZE]
- pslldq xmmA, SIZEOF_MMWORD
- por xmmA,xmmE
-.column_ld4:
- test cl, SIZEOF_XMMWORD/4
- jz short .column_ld8
- sub ecx, byte SIZEOF_XMMWORD/4
- movdqa xmmE,xmmA
- movdqu xmmA, XMMWORD [esi+ecx*RGB_PIXELSIZE]
-.column_ld8:
- test cl, SIZEOF_XMMWORD/2
- mov ecx, SIZEOF_XMMWORD
- jz short .rgb_gray_cnv
- movdqa xmmF,xmmA
- movdqa xmmH,xmmE
- movdqu xmmA, XMMWORD [esi+0*SIZEOF_XMMWORD]
- movdqu xmmE, XMMWORD [esi+1*SIZEOF_XMMWORD]
- jmp short .rgb_gray_cnv
- alignx 16,7
-
-.columnloop:
- movdqu xmmA, XMMWORD [esi+0*SIZEOF_XMMWORD]
- movdqu xmmE, XMMWORD [esi+1*SIZEOF_XMMWORD]
- movdqu xmmF, XMMWORD [esi+2*SIZEOF_XMMWORD]
- movdqu xmmH, XMMWORD [esi+3*SIZEOF_XMMWORD]
-
-.rgb_gray_cnv:
- ; xmmA=(00 10 20 30 01 11 21 31 02 12 22 32 03 13 23 33)
- ; xmmE=(04 14 24 34 05 15 25 35 06 16 26 36 07 17 27 37)
- ; xmmF=(08 18 28 38 09 19 29 39 0A 1A 2A 3A 0B 1B 2B 3B)
- ; xmmH=(0C 1C 2C 3C 0D 1D 2D 3D 0E 1E 2E 3E 0F 1F 2F 3F)
-
- movdqa xmmD,xmmA
- punpcklbw xmmA,xmmE ; xmmA=(00 04 10 14 20 24 30 34 01 05 11 15 21 25 31 35)
- punpckhbw xmmD,xmmE ; xmmD=(02 06 12 16 22 26 32 36 03 07 13 17 23 27 33 37)
-
- movdqa xmmC,xmmF
- punpcklbw xmmF,xmmH ; xmmF=(08 0C 18 1C 28 2C 38 3C 09 0D 19 1D 29 2D 39 3D)
- punpckhbw xmmC,xmmH ; xmmC=(0A 0E 1A 1E 2A 2E 3A 3E 0B 0F 1B 1F 2B 2F 3B 3F)
-
- movdqa xmmB,xmmA
- punpcklwd xmmA,xmmF ; xmmA=(00 04 08 0C 10 14 18 1C 20 24 28 2C 30 34 38 3C)
- punpckhwd xmmB,xmmF ; xmmB=(01 05 09 0D 11 15 19 1D 21 25 29 2D 31 35 39 3D)
-
- movdqa xmmG,xmmD
- punpcklwd xmmD,xmmC ; xmmD=(02 06 0A 0E 12 16 1A 1E 22 26 2A 2E 32 36 3A 3E)
- punpckhwd xmmG,xmmC ; xmmG=(03 07 0B 0F 13 17 1B 1F 23 27 2B 2F 33 37 3B 3F)
-
- movdqa xmmE,xmmA
- punpcklbw xmmA,xmmD ; xmmA=(00 02 04 06 08 0A 0C 0E 10 12 14 16 18 1A 1C 1E)
- punpckhbw xmmE,xmmD ; xmmE=(20 22 24 26 28 2A 2C 2E 30 32 34 36 38 3A 3C 3E)
-
- movdqa xmmH,xmmB
- punpcklbw xmmB,xmmG ; xmmB=(01 03 05 07 09 0B 0D 0F 11 13 15 17 19 1B 1D 1F)
- punpckhbw xmmH,xmmG ; xmmH=(21 23 25 27 29 2B 2D 2F 31 33 35 37 39 3B 3D 3F)
-
- pxor xmmF,xmmF
-
- movdqa xmmC,xmmA
- punpcklbw xmmA,xmmF ; xmmA=(00 02 04 06 08 0A 0C 0E)
- punpckhbw xmmC,xmmF ; xmmC=(10 12 14 16 18 1A 1C 1E)
-
- movdqa xmmD,xmmB
- punpcklbw xmmB,xmmF ; xmmB=(01 03 05 07 09 0B 0D 0F)
- punpckhbw xmmD,xmmF ; xmmD=(11 13 15 17 19 1B 1D 1F)
-
- movdqa xmmG,xmmE
- punpcklbw xmmE,xmmF ; xmmE=(20 22 24 26 28 2A 2C 2E)
- punpckhbw xmmG,xmmF ; xmmG=(30 32 34 36 38 3A 3C 3E)
-
- punpcklbw xmmF,xmmH
- punpckhbw xmmH,xmmH
- psrlw xmmF,BYTE_BIT ; xmmF=(21 23 25 27 29 2B 2D 2F)
- psrlw xmmH,BYTE_BIT ; xmmH=(31 33 35 37 39 3B 3D 3F)
-
-%endif ; RGB_PIXELSIZE ; ---------------
-
- ; xmm0=R(02468ACE)=RE, xmm2=G(02468ACE)=GE, xmm4=B(02468ACE)=BE
- ; xmm1=R(13579BDF)=RO, xmm3=G(13579BDF)=GO, xmm5=B(13579BDF)=BO
-
- ; (Original)
- ; Y = 0.29900 * R + 0.58700 * G + 0.11400 * B
- ;
- ; (This implementation)
- ; Y = 0.29900 * R + 0.33700 * G + 0.11400 * B + 0.25000 * G
-
- movdqa xmm6,xmm1
- punpcklwd xmm1,xmm3
- punpckhwd xmm6,xmm3
- pmaddwd xmm1,[GOTOFF(eax,PW_F0299_F0337)] ; xmm1=ROL*FIX(0.299)+GOL*FIX(0.337)
- pmaddwd xmm6,[GOTOFF(eax,PW_F0299_F0337)] ; xmm6=ROH*FIX(0.299)+GOH*FIX(0.337)
-
- movdqa xmm7, xmm6 ; xmm7=ROH*FIX(0.299)+GOH*FIX(0.337)
-
- movdqa xmm6,xmm0
- punpcklwd xmm0,xmm2
- punpckhwd xmm6,xmm2
- pmaddwd xmm0,[GOTOFF(eax,PW_F0299_F0337)] ; xmm0=REL*FIX(0.299)+GEL*FIX(0.337)
- pmaddwd xmm6,[GOTOFF(eax,PW_F0299_F0337)] ; xmm6=REH*FIX(0.299)+GEH*FIX(0.337)
-
- movdqa XMMWORD [wk(0)], xmm0 ; wk(0)=REL*FIX(0.299)+GEL*FIX(0.337)
- movdqa XMMWORD [wk(1)], xmm6 ; wk(1)=REH*FIX(0.299)+GEH*FIX(0.337)
-
- movdqa xmm0, xmm5 ; xmm0=BO
- movdqa xmm6, xmm4 ; xmm6=BE
-
- movdqa xmm4,xmm0
- punpcklwd xmm0,xmm3
- punpckhwd xmm4,xmm3
- pmaddwd xmm0,[GOTOFF(eax,PW_F0114_F0250)] ; xmm0=BOL*FIX(0.114)+GOL*FIX(0.250)
- pmaddwd xmm4,[GOTOFF(eax,PW_F0114_F0250)] ; xmm4=BOH*FIX(0.114)+GOH*FIX(0.250)
-
- movdqa xmm3,[GOTOFF(eax,PD_ONEHALF)] ; xmm3=[PD_ONEHALF]
-
- paddd xmm0, xmm1
- paddd xmm4, xmm7
- paddd xmm0,xmm3
- paddd xmm4,xmm3
- psrld xmm0,SCALEBITS ; xmm0=YOL
- psrld xmm4,SCALEBITS ; xmm4=YOH
- packssdw xmm0,xmm4 ; xmm0=YO
-
- movdqa xmm4,xmm6
- punpcklwd xmm6,xmm2
- punpckhwd xmm4,xmm2
- pmaddwd xmm6,[GOTOFF(eax,PW_F0114_F0250)] ; xmm6=BEL*FIX(0.114)+GEL*FIX(0.250)
- pmaddwd xmm4,[GOTOFF(eax,PW_F0114_F0250)] ; xmm4=BEH*FIX(0.114)+GEH*FIX(0.250)
-
- movdqa xmm2,[GOTOFF(eax,PD_ONEHALF)] ; xmm2=[PD_ONEHALF]
-
- paddd xmm6, XMMWORD [wk(0)]
- paddd xmm4, XMMWORD [wk(1)]
- paddd xmm6,xmm2
- paddd xmm4,xmm2
- psrld xmm6,SCALEBITS ; xmm6=YEL
- psrld xmm4,SCALEBITS ; xmm4=YEH
- packssdw xmm6,xmm4 ; xmm6=YE
-
- psllw xmm0,BYTE_BIT
- por xmm6,xmm0 ; xmm6=Y
- movdqa XMMWORD [edi], xmm6 ; Save Y
-
- sub ecx, byte SIZEOF_XMMWORD
- add esi, byte RGB_PIXELSIZE*SIZEOF_XMMWORD ; inptr
- add edi, byte SIZEOF_XMMWORD ; outptr0
- cmp ecx, byte SIZEOF_XMMWORD
- jae near .columnloop
- test ecx,ecx
- jnz near .column_ld1
-
- pop ecx ; col
- pop esi
- pop edi
- poppic eax
-
- add esi, byte SIZEOF_JSAMPROW ; input_buf
- add edi, byte SIZEOF_JSAMPROW
- dec eax ; num_rows
- jg near .rowloop
-
-.return:
- pop edi
- pop esi
-; pop edx ; need not be preserved
-; pop ecx ; need not be preserved
- pop ebx
- mov esp,ebp ; esp <- aligned ebp
- pop esp ; esp <- original ebp
- pop ebp
- ret
-
-; For some reason, the OS X linker does not honor the request to align the
-; segment unless we do this.
- align 16
diff --git a/simd/jchuff-sse2-64.asm b/simd/jchuff-sse2-64.asm
deleted file mode 100644
index b1144d1..0000000
--- a/simd/jchuff-sse2-64.asm
+++ /dev/null
@@ -1,360 +0,0 @@
-;
-; jchuff-sse2-64.asm - Huffman entropy encoding (64-bit SSE2)
-;
-; Copyright (C) 2009-2011, 2014-2016, D. R. Commander.
-; Copyright (C) 2015, Matthieu Darbois.
-;
-; Based on the x86 SIMD extension for IJG JPEG library
-; Copyright (C) 1999-2006, MIYASAKA Masaru.
-; For conditions of distribution and use, see copyright notice in jsimdext.inc
-;
-; This file should be assembled with NASM (Netwide Assembler),
-; can *not* be assembled with Microsoft's MASM or any compatible
-; assembler (including Borland's Turbo Assembler).
-; NASM is available from http://nasm.sourceforge.net/ or
-; http://sourceforge.net/project/showfiles.php?group_id=6208
-;
-; This file contains an SSE2 implementation for Huffman coding of one block.
-; The following code is based directly on jchuff.c; see jchuff.c for more
-; details.
-;
-; [TAB8]
-
-%include "jsimdext.inc"
-
-; --------------------------------------------------------------------------
- SECTION SEG_CONST
-
- alignz 16
- global EXTN(jconst_huff_encode_one_block)
-
-EXTN(jconst_huff_encode_one_block):
-
-%include "jpeg_nbits_table.inc"
-
- alignz 16
-
-; --------------------------------------------------------------------------
- SECTION SEG_TEXT
- BITS 64
-
-; These macros perform the same task as the emit_bits() function in the
-; original libjpeg code. In addition to reducing overhead by explicitly
-; inlining the code, additional performance is achieved by taking into
-; account the size of the bit buffer and waiting until it is almost full
-; before emptying it. This mostly benefits 64-bit platforms, since 6
-; bytes can be stored in a 64-bit bit buffer before it has to be emptied.
-
-%macro EMIT_BYTE 0
- sub put_bits, 8 ; put_bits -= 8;
- mov rdx, put_buffer
- mov ecx, put_bits
- shr rdx, cl ; c = (JOCTET)GETJOCTET(put_buffer >> put_bits);
- mov byte [buffer], dl ; *buffer++ = c;
- add buffer, 1
- cmp dl, 0xFF ; need to stuff a zero byte?
- jne %%.EMIT_BYTE_END
- mov byte [buffer], 0 ; *buffer++ = 0;
- add buffer, 1
-%%.EMIT_BYTE_END:
-%endmacro
-
-%macro PUT_BITS 1
- add put_bits, ecx ; put_bits += size;
- shl put_buffer, cl ; put_buffer = (put_buffer << size);
- or put_buffer, %1
-%endmacro
-
-%macro CHECKBUF31 0
- cmp put_bits, 32 ; if (put_bits > 31) {
- jl %%.CHECKBUF31_END
- EMIT_BYTE
- EMIT_BYTE
- EMIT_BYTE
- EMIT_BYTE
-%%.CHECKBUF31_END:
-%endmacro
-
-%macro CHECKBUF47 0
- cmp put_bits, 48 ; if (put_bits > 47) {
- jl %%.CHECKBUF47_END
- EMIT_BYTE
- EMIT_BYTE
- EMIT_BYTE
- EMIT_BYTE
- EMIT_BYTE
- EMIT_BYTE
-%%.CHECKBUF47_END:
-%endmacro
-
-%macro EMIT_BITS 2
- CHECKBUF47
- mov ecx, %2
- PUT_BITS %1
-%endmacro
-
-%macro kloop_prepare 37 ;(ko, jno0, ..., jno31, xmm0, xmm1, xmm2, xmm3)
- pxor xmm8, xmm8 ; __m128i neg = _mm_setzero_si128();
- pxor xmm9, xmm9 ; __m128i neg = _mm_setzero_si128();
- pxor xmm10, xmm10 ; __m128i neg = _mm_setzero_si128();
- pxor xmm11, xmm11 ; __m128i neg = _mm_setzero_si128();
- pinsrw %34, word [r12 + %2 * SIZEOF_WORD], 0 ; xmm_shadow[0] = block[jno0];
- pinsrw %35, word [r12 + %10 * SIZEOF_WORD], 0 ; xmm_shadow[8] = block[jno8];
- pinsrw %36, word [r12 + %18 * SIZEOF_WORD], 0 ; xmm_shadow[16] = block[jno16];
- pinsrw %37, word [r12 + %26 * SIZEOF_WORD], 0 ; xmm_shadow[24] = block[jno24];
- pinsrw %34, word [r12 + %3 * SIZEOF_WORD], 1 ; xmm_shadow[1] = block[jno1];
- pinsrw %35, word [r12 + %11 * SIZEOF_WORD], 1 ; xmm_shadow[9] = block[jno9];
- pinsrw %36, word [r12 + %19 * SIZEOF_WORD], 1 ; xmm_shadow[17] = block[jno17];
- pinsrw %37, word [r12 + %27 * SIZEOF_WORD], 1 ; xmm_shadow[25] = block[jno25];
- pinsrw %34, word [r12 + %4 * SIZEOF_WORD], 2 ; xmm_shadow[2] = block[jno2];
- pinsrw %35, word [r12 + %12 * SIZEOF_WORD], 2 ; xmm_shadow[10] = block[jno10];
- pinsrw %36, word [r12 + %20 * SIZEOF_WORD], 2 ; xmm_shadow[18] = block[jno18];
- pinsrw %37, word [r12 + %28 * SIZEOF_WORD], 2 ; xmm_shadow[26] = block[jno26];
- pinsrw %34, word [r12 + %5 * SIZEOF_WORD], 3 ; xmm_shadow[3] = block[jno3];
- pinsrw %35, word [r12 + %13 * SIZEOF_WORD], 3 ; xmm_shadow[11] = block[jno11];
- pinsrw %36, word [r12 + %21 * SIZEOF_WORD], 3 ; xmm_shadow[19] = block[jno19];
- pinsrw %37, word [r12 + %29 * SIZEOF_WORD], 3 ; xmm_shadow[27] = block[jno27];
- pinsrw %34, word [r12 + %6 * SIZEOF_WORD], 4 ; xmm_shadow[4] = block[jno4];
- pinsrw %35, word [r12 + %14 * SIZEOF_WORD], 4 ; xmm_shadow[12] = block[jno12];
- pinsrw %36, word [r12 + %22 * SIZEOF_WORD], 4 ; xmm_shadow[20] = block[jno20];
- pinsrw %37, word [r12 + %30 * SIZEOF_WORD], 4 ; xmm_shadow[28] = block[jno28];
- pinsrw %34, word [r12 + %7 * SIZEOF_WORD], 5 ; xmm_shadow[5] = block[jno5];
- pinsrw %35, word [r12 + %15 * SIZEOF_WORD], 5 ; xmm_shadow[13] = block[jno13];
- pinsrw %36, word [r12 + %23 * SIZEOF_WORD], 5 ; xmm_shadow[21] = block[jno21];
- pinsrw %37, word [r12 + %31 * SIZEOF_WORD], 5 ; xmm_shadow[29] = block[jno29];
- pinsrw %34, word [r12 + %8 * SIZEOF_WORD], 6 ; xmm_shadow[6] = block[jno6];
- pinsrw %35, word [r12 + %16 * SIZEOF_WORD], 6 ; xmm_shadow[14] = block[jno14];
- pinsrw %36, word [r12 + %24 * SIZEOF_WORD], 6 ; xmm_shadow[22] = block[jno22];
- pinsrw %37, word [r12 + %32 * SIZEOF_WORD], 6 ; xmm_shadow[30] = block[jno30];
- pinsrw %34, word [r12 + %9 * SIZEOF_WORD], 7 ; xmm_shadow[7] = block[jno7];
- pinsrw %35, word [r12 + %17 * SIZEOF_WORD], 7 ; xmm_shadow[15] = block[jno15];
- pinsrw %36, word [r12 + %25 * SIZEOF_WORD], 7 ; xmm_shadow[23] = block[jno23];
-%if %1 != 32
- pinsrw %37, word [r12 + %33 * SIZEOF_WORD], 7 ; xmm_shadow[31] = block[jno31];
-%else
- pinsrw %37, ebx, 7 ; xmm_shadow[31] = block[jno31];
-%endif
- pcmpgtw xmm8, %34 ; neg = _mm_cmpgt_epi16(neg, x1);
- pcmpgtw xmm9, %35 ; neg = _mm_cmpgt_epi16(neg, x1);
- pcmpgtw xmm10, %36 ; neg = _mm_cmpgt_epi16(neg, x1);
- pcmpgtw xmm11, %37 ; neg = _mm_cmpgt_epi16(neg, x1);
- paddw %34, xmm8 ; x1 = _mm_add_epi16(x1, neg);
- paddw %35, xmm9 ; x1 = _mm_add_epi16(x1, neg);
- paddw %36, xmm10 ; x1 = _mm_add_epi16(x1, neg);
- paddw %37, xmm11 ; x1 = _mm_add_epi16(x1, neg);
- pxor %34, xmm8 ; x1 = _mm_xor_si128(x1, neg);
- pxor %35, xmm9 ; x1 = _mm_xor_si128(x1, neg);
- pxor %36, xmm10 ; x1 = _mm_xor_si128(x1, neg);
- pxor %37, xmm11 ; x1 = _mm_xor_si128(x1, neg);
- pxor xmm8, %34 ; neg = _mm_xor_si128(neg, x1);
- pxor xmm9, %35 ; neg = _mm_xor_si128(neg, x1);
- pxor xmm10, %36 ; neg = _mm_xor_si128(neg, x1);
- pxor xmm11, %37 ; neg = _mm_xor_si128(neg, x1);
- movdqa XMMWORD [t1 + %1 * SIZEOF_WORD], %34 ; _mm_storeu_si128((__m128i *)(t1 + ko), x1);
- movdqa XMMWORD [t1 + (%1 + 8) * SIZEOF_WORD], %35 ; _mm_storeu_si128((__m128i *)(t1 + ko + 8), x1);
- movdqa XMMWORD [t1 + (%1 + 16) * SIZEOF_WORD], %36 ; _mm_storeu_si128((__m128i *)(t1 + ko + 16), x1);
- movdqa XMMWORD [t1 + (%1 + 24) * SIZEOF_WORD], %37 ; _mm_storeu_si128((__m128i *)(t1 + ko + 24), x1);
- movdqa XMMWORD [t2 + %1 * SIZEOF_WORD], xmm8 ; _mm_storeu_si128((__m128i *)(t2 + ko), neg);
- movdqa XMMWORD [t2 + (%1 + 8) * SIZEOF_WORD], xmm9 ; _mm_storeu_si128((__m128i *)(t2 + ko + 8), neg);
- movdqa XMMWORD [t2 + (%1 + 16) * SIZEOF_WORD], xmm10 ; _mm_storeu_si128((__m128i *)(t2 + ko + 16), neg);
- movdqa XMMWORD [t2 + (%1 + 24) * SIZEOF_WORD], xmm11 ; _mm_storeu_si128((__m128i *)(t2 + ko + 24), neg);
-%endmacro
-
-;
-; Encode a single block's worth of coefficients.
-;
-; GLOBAL(JOCTET*)
-; jsimd_huff_encode_one_block_sse2 (working_state *state, JOCTET *buffer,
-; JCOEFPTR block, int last_dc_val,
-; c_derived_tbl *dctbl, c_derived_tbl *actbl)
-;
-
-; r10 = working_state *state
-; r11 = JOCTET *buffer
-; r12 = JCOEFPTR block
-; r13 = int last_dc_val
-; r14 = c_derived_tbl *dctbl
-; r15 = c_derived_tbl *actbl
-
-%define t1 rbp-(DCTSIZE2*SIZEOF_WORD)
-%define t2 t1-(DCTSIZE2*SIZEOF_WORD)
-%define put_buffer r8
-%define put_bits r9d
-%define buffer rax
-
- align 16
- global EXTN(jsimd_huff_encode_one_block_sse2)
-
-EXTN(jsimd_huff_encode_one_block_sse2):
- push rbp
- mov rax,rsp ; rax = original rbp
- sub rsp, byte 4
- and rsp, byte (-SIZEOF_XMMWORD) ; align to 128 bits
- mov [rsp],rax
- mov rbp,rsp ; rbp = aligned rbp
- lea rsp, [t2]
- collect_args
-%ifdef WIN64
- movaps XMMWORD [rsp-1*SIZEOF_XMMWORD], xmm8
- movaps XMMWORD [rsp-2*SIZEOF_XMMWORD], xmm9
- movaps XMMWORD [rsp-3*SIZEOF_XMMWORD], xmm10
- movaps XMMWORD [rsp-4*SIZEOF_XMMWORD], xmm11
- sub rsp, 4*SIZEOF_XMMWORD
-%endif
- push rbx
-
- mov buffer, r11 ; r11 is now sratch
-
- mov put_buffer, MMWORD [r10+16] ; put_buffer = state->cur.put_buffer;
- mov put_bits, DWORD [r10+24] ; put_bits = state->cur.put_bits;
- push r10 ; r10 is now scratch
-
- ; Encode the DC coefficient difference per section F.1.2.1
- movsx edi, word [r12] ; temp = temp2 = block[0] - last_dc_val;
- sub edi, r13d ; r13 is not used anymore
- mov ebx, edi
-
- ; This is a well-known technique for obtaining the absolute value
- ; without a branch. It is derived from an assembly language technique
- ; presented in "How to Optimize for the Pentium Processors",
- ; Copyright (c) 1996, 1997 by Agner Fog.
- mov esi, edi
- sar esi, 31 ; temp3 = temp >> (CHAR_BIT * sizeof(int) - 1);
- xor edi, esi ; temp ^= temp3;
- sub edi, esi ; temp -= temp3;
-
- ; For a negative input, want temp2 = bitwise complement of abs(input)
- ; This code assumes we are on a two's complement machine
- add ebx, esi ; temp2 += temp3;
-
- ; Find the number of bits needed for the magnitude of the coefficient
- lea r11, [rel jpeg_nbits_table]
- movzx rdi, byte [r11 + rdi] ; nbits = JPEG_NBITS(temp);
- ; Emit the Huffman-coded symbol for the number of bits
- mov r11d, INT [r14 + rdi * 4] ; code = dctbl->ehufco[nbits];
- movzx esi, byte [r14 + rdi + 1024] ; size = dctbl->ehufsi[nbits];
- EMIT_BITS r11, esi ; EMIT_BITS(code, size)
-
- ; Mask off any extra bits in code
- mov esi, 1
- mov ecx, edi
- shl esi, cl
- dec esi
- and ebx, esi ; temp2 &= (((JLONG) 1)<<nbits) - 1;
-
- ; Emit that number of bits of the value, if positive,
- ; or the complement of its magnitude, if negative.
- EMIT_BITS rbx, edi ; EMIT_BITS(temp2, nbits)
-
- ; Prepare data
- xor ebx, ebx
- kloop_prepare 0, 1, 8, 16, 9, 2, 3, 10, 17, 24, 32, 25, \
- 18, 11, 4, 5, 12, 19, 26, 33, 40, 48, 41, 34, \
- 27, 20, 13, 6, 7, 14, 21, 28, 35, \
- xmm0, xmm1, xmm2, xmm3
- kloop_prepare 32, 42, 49, 56, 57, 50, 43, 36, 29, 22, 15, 23, \
- 30, 37, 44, 51, 58, 59, 52, 45, 38, 31, 39, 46, \
- 53, 60, 61, 54, 47, 55, 62, 63, 63, \
- xmm4, xmm5, xmm6, xmm7
-
- pxor xmm8, xmm8
- pcmpeqw xmm0, xmm8 ; tmp0 = _mm_cmpeq_epi16(tmp0, zero);
- pcmpeqw xmm1, xmm8 ; tmp1 = _mm_cmpeq_epi16(tmp1, zero);
- pcmpeqw xmm2, xmm8 ; tmp2 = _mm_cmpeq_epi16(tmp2, zero);
- pcmpeqw xmm3, xmm8 ; tmp3 = _mm_cmpeq_epi16(tmp3, zero);
- pcmpeqw xmm4, xmm8 ; tmp4 = _mm_cmpeq_epi16(tmp4, zero);
- pcmpeqw xmm5, xmm8 ; tmp5 = _mm_cmpeq_epi16(tmp5, zero);
- pcmpeqw xmm6, xmm8 ; tmp6 = _mm_cmpeq_epi16(tmp6, zero);
- pcmpeqw xmm7, xmm8 ; tmp7 = _mm_cmpeq_epi16(tmp7, zero);
- packsswb xmm0, xmm1 ; tmp0 = _mm_packs_epi16(tmp0, tmp1);
- packsswb xmm2, xmm3 ; tmp2 = _mm_packs_epi16(tmp2, tmp3);
- packsswb xmm4, xmm5 ; tmp4 = _mm_packs_epi16(tmp4, tmp5);
- packsswb xmm6, xmm7 ; tmp6 = _mm_packs_epi16(tmp6, tmp7);
- pmovmskb r11d, xmm0 ; index = ((uint64_t)_mm_movemask_epi8(tmp0)) << 0;
- pmovmskb r12d, xmm2 ; index = ((uint64_t)_mm_movemask_epi8(tmp2)) << 16;
- pmovmskb r13d, xmm4 ; index = ((uint64_t)_mm_movemask_epi8(tmp4)) << 32;
- pmovmskb r14d, xmm6 ; index = ((uint64_t)_mm_movemask_epi8(tmp6)) << 48;
- shl r12, 16
- shl r14, 16
- or r11, r12
- or r13, r14
- shl r13, 32
- or r11, r13
- not r11 ; index = ~index;
-
- ;mov MMWORD [ t1 + DCTSIZE2 * SIZEOF_WORD ], r11
- ;jmp .EFN
-
- mov r13d, INT [r15 + 240 * 4] ; code_0xf0 = actbl->ehufco[0xf0];
- movzx r14d, byte [r15 + 1024 + 240] ; size_0xf0 = actbl->ehufsi[0xf0];
- lea rsi, [t1]
-.BLOOP:
- bsf r12, r11 ; r = __builtin_ctzl(index);
- jz .ELOOP
- mov rcx, r12
- lea rsi, [rsi+r12*2] ; k += r;
- shr r11, cl ; index >>= r;
- movzx rdi, word [rsi] ; temp = t1[k];
- lea rbx, [rel jpeg_nbits_table]
- movzx rdi, byte [rbx + rdi] ; nbits = JPEG_NBITS(temp);
-.BRLOOP:
- cmp r12, 16 ; while (r > 15) {
- jl .ERLOOP
- EMIT_BITS r13, r14d ; EMIT_BITS(code_0xf0, size_0xf0)
- sub r12, 16 ; r -= 16;
- jmp .BRLOOP
-.ERLOOP:
- ; Emit Huffman symbol for run length / number of bits
- CHECKBUF31 ; uses rcx, rdx
-
- shl r12, 4 ; temp3 = (r << 4) + nbits;
- add r12, rdi
- mov ebx, INT [r15 + r12 * 4] ; code = actbl->ehufco[temp3];
- movzx ecx, byte [r15 + r12 + 1024] ; size = actbl->ehufsi[temp3];
- PUT_BITS rbx
-
- ;EMIT_CODE(code, size)
-
- movsx ebx, word [rsi-DCTSIZE2*2] ; temp2 = t2[k];
- ; Mask off any extra bits in code
- mov rcx, rdi
- mov rdx, 1
- shl rdx, cl
- dec rdx
- and rbx, rdx ; temp2 &= (((JLONG) 1)<<nbits) - 1;
- PUT_BITS rbx ; PUT_BITS(temp2, nbits)
-
- shr r11, 1 ; index >>= 1;
- add rsi, 2 ; ++k;
- jmp .BLOOP
-.ELOOP:
- ; If the last coef(s) were zero, emit an end-of-block code
- lea rdi, [t1 + (DCTSIZE2-1) * 2] ; r = DCTSIZE2-1-k;
- cmp rdi, rsi ; if (r > 0) {
- je .EFN
- mov ebx, INT [r15] ; code = actbl->ehufco[0];
- movzx r12d, byte [r15 + 1024] ; size = actbl->ehufsi[0];
- EMIT_BITS rbx, r12d
-.EFN:
- pop r10
- ; Save put_buffer & put_bits
- mov MMWORD [r10+16], put_buffer ; state->cur.put_buffer = put_buffer;
- mov DWORD [r10+24], put_bits ; state->cur.put_bits = put_bits;
-
- pop rbx
-%ifdef WIN64
- movaps xmm11, XMMWORD [rsp+0*SIZEOF_XMMWORD]
- movaps xmm10, XMMWORD [rsp+1*SIZEOF_XMMWORD]
- movaps xmm9, XMMWORD [rsp+2*SIZEOF_XMMWORD]
- movaps xmm8, XMMWORD [rsp+3*SIZEOF_XMMWORD]
- add rsp, 4*SIZEOF_XMMWORD
-%endif
- uncollect_args
- mov rsp,rbp ; rsp <- aligned rbp
- pop rsp ; rsp <- original rbp
- pop rbp
- ret
-
-; For some reason, the OS X linker does not honor the request to align the
-; segment unless we do this.
- align 16
diff --git a/simd/jchuff-sse2.asm b/simd/jchuff-sse2.asm
deleted file mode 100644
index b81db75..0000000
--- a/simd/jchuff-sse2.asm
+++ /dev/null
@@ -1,426 +0,0 @@
-;
-; jchuff-sse2.asm - Huffman entropy encoding (SSE2)
-;
-; Copyright (C) 2009-2011, 2014-2017, D. R. Commander.
-; Copyright (C) 2015, Matthieu Darbois.
-;
-; Based on the x86 SIMD extension for IJG JPEG library
-; Copyright (C) 1999-2006, MIYASAKA Masaru.
-; For conditions of distribution and use, see copyright notice in jsimdext.inc
-;
-; This file should be assembled with NASM (Netwide Assembler),
-; can *not* be assembled with Microsoft's MASM or any compatible
-; assembler (including Borland's Turbo Assembler).
-; NASM is available from http://nasm.sourceforge.net/ or
-; http://sourceforge.net/project/showfiles.php?group_id=6208
-;
-; This file contains an SSE2 implementation for Huffman coding of one block.
-; The following code is based directly on jchuff.c; see jchuff.c for more
-; details.
-;
-; [TAB8]
-
-%include "jsimdext.inc"
-
-; --------------------------------------------------------------------------
- SECTION SEG_CONST
-
- alignz 16
- global EXTN(jconst_huff_encode_one_block)
-
-EXTN(jconst_huff_encode_one_block):
-
-%include "jpeg_nbits_table.inc"
-
- alignz 16
-
-; --------------------------------------------------------------------------
- SECTION SEG_TEXT
- BITS 32
-
-; These macros perform the same task as the emit_bits() function in the
-; original libjpeg code. In addition to reducing overhead by explicitly
-; inlining the code, additional performance is achieved by taking into
-; account the size of the bit buffer and waiting until it is almost full
-; before emptying it. This mostly benefits 64-bit platforms, since 6
-; bytes can be stored in a 64-bit bit buffer before it has to be emptied.
-
-%macro EMIT_BYTE 0
- sub put_bits, 8 ; put_bits -= 8;
- mov edx, put_buffer
- mov ecx, put_bits
- shr edx, cl ; c = (JOCTET)GETJOCTET(put_buffer >> put_bits);
- mov byte [eax], dl ; *buffer++ = c;
- add eax, 1
- cmp dl, 0xFF ; need to stuff a zero byte?
- jne %%.EMIT_BYTE_END
- mov byte [eax], 0 ; *buffer++ = 0;
- add eax, 1
-%%.EMIT_BYTE_END:
-%endmacro
-
-%macro PUT_BITS 1
- add put_bits, ecx ; put_bits += size;
- shl put_buffer, cl ; put_buffer = (put_buffer << size);
- or put_buffer, %1
-%endmacro
-
-%macro CHECKBUF15 0
- cmp put_bits, 16 ; if (put_bits > 31) {
- jl %%.CHECKBUF15_END
- mov eax, POINTER [esp+buffer]
- EMIT_BYTE
- EMIT_BYTE
- mov POINTER [esp+buffer], eax
-%%.CHECKBUF15_END:
-%endmacro
-
-%macro EMIT_BITS 1
- PUT_BITS %1
- CHECKBUF15
-%endmacro
-
-%macro kloop_prepare 37 ;(ko, jno0, ..., jno31, xmm0, xmm1, xmm2, xmm3)
- pxor xmm4, xmm4 ; __m128i neg = _mm_setzero_si128();
- pxor xmm5, xmm5 ; __m128i neg = _mm_setzero_si128();
- pxor xmm6, xmm6 ; __m128i neg = _mm_setzero_si128();
- pxor xmm7, xmm7 ; __m128i neg = _mm_setzero_si128();
- pinsrw %34, word [esi + %2 * SIZEOF_WORD], 0 ; xmm_shadow[0] = block[jno0];
- pinsrw %35, word [esi + %10 * SIZEOF_WORD], 0 ; xmm_shadow[8] = block[jno8];
- pinsrw %36, word [esi + %18 * SIZEOF_WORD], 0 ; xmm_shadow[16] = block[jno16];
- pinsrw %37, word [esi + %26 * SIZEOF_WORD], 0 ; xmm_shadow[24] = block[jno24];
- pinsrw %34, word [esi + %3 * SIZEOF_WORD], 1 ; xmm_shadow[1] = block[jno1];
- pinsrw %35, word [esi + %11 * SIZEOF_WORD], 1 ; xmm_shadow[9] = block[jno9];
- pinsrw %36, word [esi + %19 * SIZEOF_WORD], 1 ; xmm_shadow[17] = block[jno17];
- pinsrw %37, word [esi + %27 * SIZEOF_WORD], 1 ; xmm_shadow[25] = block[jno25];
- pinsrw %34, word [esi + %4 * SIZEOF_WORD], 2 ; xmm_shadow[2] = block[jno2];
- pinsrw %35, word [esi + %12 * SIZEOF_WORD], 2 ; xmm_shadow[10] = block[jno10];
- pinsrw %36, word [esi + %20 * SIZEOF_WORD], 2 ; xmm_shadow[18] = block[jno18];
- pinsrw %37, word [esi + %28 * SIZEOF_WORD], 2 ; xmm_shadow[26] = block[jno26];
- pinsrw %34, word [esi + %5 * SIZEOF_WORD], 3 ; xmm_shadow[3] = block[jno3];
- pinsrw %35, word [esi + %13 * SIZEOF_WORD], 3 ; xmm_shadow[11] = block[jno11];
- pinsrw %36, word [esi + %21 * SIZEOF_WORD], 3 ; xmm_shadow[19] = block[jno19];
- pinsrw %37, word [esi + %29 * SIZEOF_WORD], 3 ; xmm_shadow[27] = block[jno27];
- pinsrw %34, word [esi + %6 * SIZEOF_WORD], 4 ; xmm_shadow[4] = block[jno4];
- pinsrw %35, word [esi + %14 * SIZEOF_WORD], 4 ; xmm_shadow[12] = block[jno12];
- pinsrw %36, word [esi + %22 * SIZEOF_WORD], 4 ; xmm_shadow[20] = block[jno20];
- pinsrw %37, word [esi + %30 * SIZEOF_WORD], 4 ; xmm_shadow[28] = block[jno28];
- pinsrw %34, word [esi + %7 * SIZEOF_WORD], 5 ; xmm_shadow[5] = block[jno5];
- pinsrw %35, word [esi + %15 * SIZEOF_WORD], 5 ; xmm_shadow[13] = block[jno13];
- pinsrw %36, word [esi + %23 * SIZEOF_WORD], 5 ; xmm_shadow[21] = block[jno21];
- pinsrw %37, word [esi + %31 * SIZEOF_WORD], 5 ; xmm_shadow[29] = block[jno29];
- pinsrw %34, word [esi + %8 * SIZEOF_WORD], 6 ; xmm_shadow[6] = block[jno6];
- pinsrw %35, word [esi + %16 * SIZEOF_WORD], 6 ; xmm_shadow[14] = block[jno14];
- pinsrw %36, word [esi + %24 * SIZEOF_WORD], 6 ; xmm_shadow[22] = block[jno22];
- pinsrw %37, word [esi + %32 * SIZEOF_WORD], 6 ; xmm_shadow[30] = block[jno30];
- pinsrw %34, word [esi + %9 * SIZEOF_WORD], 7 ; xmm_shadow[7] = block[jno7];
- pinsrw %35, word [esi + %17 * SIZEOF_WORD], 7 ; xmm_shadow[15] = block[jno15];
- pinsrw %36, word [esi + %25 * SIZEOF_WORD], 7 ; xmm_shadow[23] = block[jno23];
-%if %1 != 32
- pinsrw %37, word [esi + %33 * SIZEOF_WORD], 7 ; xmm_shadow[31] = block[jno31];
-%else
- pinsrw %37, ecx, 7 ; xmm_shadow[31] = block[jno31];
-%endif
- pcmpgtw xmm4, %34 ; neg = _mm_cmpgt_epi16(neg, x1);
- pcmpgtw xmm5, %35 ; neg = _mm_cmpgt_epi16(neg, x1);
- pcmpgtw xmm6, %36 ; neg = _mm_cmpgt_epi16(neg, x1);
- pcmpgtw xmm7, %37 ; neg = _mm_cmpgt_epi16(neg, x1);
- paddw %34, xmm4 ; x1 = _mm_add_epi16(x1, neg);
- paddw %35, xmm5 ; x1 = _mm_add_epi16(x1, neg);
- paddw %36, xmm6 ; x1 = _mm_add_epi16(x1, neg);
- paddw %37, xmm7 ; x1 = _mm_add_epi16(x1, neg);
- pxor %34, xmm4 ; x1 = _mm_xor_si128(x1, neg);
- pxor %35, xmm5 ; x1 = _mm_xor_si128(x1, neg);
- pxor %36, xmm6 ; x1 = _mm_xor_si128(x1, neg);
- pxor %37, xmm7 ; x1 = _mm_xor_si128(x1, neg);
- pxor xmm4, %34 ; neg = _mm_xor_si128(neg, x1);
- pxor xmm5, %35 ; neg = _mm_xor_si128(neg, x1);
- pxor xmm6, %36 ; neg = _mm_xor_si128(neg, x1);
- pxor xmm7, %37 ; neg = _mm_xor_si128(neg, x1);
- movdqa XMMWORD [esp + t1 + %1 * SIZEOF_WORD], %34 ; _mm_storeu_si128((__m128i *)(t1 + ko), x1);
- movdqa XMMWORD [esp + t1 + (%1 + 8) * SIZEOF_WORD], %35 ; _mm_storeu_si128((__m128i *)(t1 + ko + 8), x1);
- movdqa XMMWORD [esp + t1 + (%1 + 16) * SIZEOF_WORD], %36 ; _mm_storeu_si128((__m128i *)(t1 + ko + 16), x1);
- movdqa XMMWORD [esp + t1 + (%1 + 24) * SIZEOF_WORD], %37 ; _mm_storeu_si128((__m128i *)(t1 + ko + 24), x1);
- movdqa XMMWORD [esp + t2 + %1 * SIZEOF_WORD], xmm4 ; _mm_storeu_si128((__m128i *)(t2 + ko), neg);
- movdqa XMMWORD [esp + t2 + (%1 + 8) * SIZEOF_WORD], xmm5 ; _mm_storeu_si128((__m128i *)(t2 + ko + 8), neg);
- movdqa XMMWORD [esp + t2 + (%1 + 16) * SIZEOF_WORD], xmm6 ; _mm_storeu_si128((__m128i *)(t2 + ko + 16), neg);
- movdqa XMMWORD [esp + t2 + (%1 + 24) * SIZEOF_WORD], xmm7 ; _mm_storeu_si128((__m128i *)(t2 + ko + 24), neg);
-%endmacro
-
-;
-; Encode a single block's worth of coefficients.
-;
-; GLOBAL(JOCTET*)
-; jsimd_huff_encode_one_block_sse2 (working_state *state, JOCTET *buffer,
-; JCOEFPTR block, int last_dc_val,
-; c_derived_tbl *dctbl, c_derived_tbl *actbl)
-;
-
-; eax + 8 = working_state *state
-; eax + 12 = JOCTET *buffer
-; eax + 16 = JCOEFPTR block
-; eax + 20 = int last_dc_val
-; eax + 24 = c_derived_tbl *dctbl
-; eax + 28 = c_derived_tbl *actbl
-
-%define pad 6*SIZEOF_DWORD ; Align to 16 bytes
-%define t1 pad
-%define t2 t1+(DCTSIZE2*SIZEOF_WORD)
-%define block t2+(DCTSIZE2*SIZEOF_WORD)
-%define actbl block+SIZEOF_DWORD
-%define buffer actbl+SIZEOF_DWORD
-%define temp buffer+SIZEOF_DWORD
-%define temp2 temp+SIZEOF_DWORD
-%define temp3 temp2+SIZEOF_DWORD
-%define temp4 temp3+SIZEOF_DWORD
-%define temp5 temp4+SIZEOF_DWORD
-%define gotptr temp5+SIZEOF_DWORD ; void *gotptr
-%define put_buffer ebx
-%define put_bits edi
-
- align 16
- global EXTN(jsimd_huff_encode_one_block_sse2)
-
-EXTN(jsimd_huff_encode_one_block_sse2):
- push ebp
- mov eax,esp ; eax = original ebp
- sub esp, byte 4
- and esp, byte (-SIZEOF_XMMWORD) ; align to 128 bits
- mov [esp],eax
- mov ebp,esp ; ebp = aligned ebp
- sub esp, temp5+9*SIZEOF_DWORD-pad
- push ebx
- push ecx
-; push edx ; need not be preserved
- push esi
- push edi
- push ebp
-
- mov esi, POINTER [eax+8] ; (working_state *state)
- mov put_buffer, DWORD [esi+8] ; put_buffer = state->cur.put_buffer;
- mov put_bits, DWORD [esi+12] ; put_bits = state->cur.put_bits;
- push esi ; esi is now scratch
-
- get_GOT edx ; get GOT address
- movpic POINTER [esp+gotptr], edx ; save GOT address
-
- mov ecx, POINTER [eax+28]
- mov edx, POINTER [eax+16]
- mov esi, POINTER [eax+12]
- mov POINTER [esp+actbl], ecx
- mov POINTER [esp+block], edx
- mov POINTER [esp+buffer], esi
-
- ; Encode the DC coefficient difference per section F.1.2.1
- mov esi, POINTER [esp+block] ; block
- movsx ecx, word [esi] ; temp = temp2 = block[0] - last_dc_val;
- sub ecx, DWORD [eax+20]
- mov esi, ecx
-
- ; This is a well-known technique for obtaining the absolute value
- ; without a branch. It is derived from an assembly language technique
- ; presented in "How to Optimize for the Pentium Processors",
- ; Copyright (c) 1996, 1997 by Agner Fog.
- mov edx, ecx
- sar edx, 31 ; temp3 = temp >> (CHAR_BIT * sizeof(int) - 1);
- xor ecx, edx ; temp ^= temp3;
- sub ecx, edx ; temp -= temp3;
-
- ; For a negative input, want temp2 = bitwise complement of abs(input)
- ; This code assumes we are on a two's complement machine
- add esi, edx ; temp2 += temp3;
- mov DWORD [esp+temp], esi ; backup temp2 in temp
-
- ; Find the number of bits needed for the magnitude of the coefficient
- movpic ebp, POINTER [esp+gotptr] ; load GOT address (ebp)
- movzx edx, byte [GOTOFF(ebp, jpeg_nbits_table + ecx)] ; nbits = JPEG_NBITS(temp);
- mov DWORD [esp+temp2], edx ; backup nbits in temp2
-
- ; Emit the Huffman-coded symbol for the number of bits
- mov ebp, POINTER [eax+24] ; After this point, arguments are not accessible anymore
- mov eax, INT [ebp + edx * 4] ; code = dctbl->ehufco[nbits];
- movzx ecx, byte [ebp + edx + 1024] ; size = dctbl->ehufsi[nbits];
- EMIT_BITS eax ; EMIT_BITS(code, size)
-
- mov ecx, DWORD [esp+temp2] ; restore nbits
-
- ; Mask off any extra bits in code
- mov eax, 1
- shl eax, cl
- dec eax
- and eax, DWORD [esp+temp] ; temp2 &= (((JLONG) 1)<<nbits) - 1;
-
- ; Emit that number of bits of the value, if positive,
- ; or the complement of its magnitude, if negative.
- EMIT_BITS eax ; EMIT_BITS(temp2, nbits)
-
- ; Prepare data
- xor ecx, ecx
- mov esi, POINTER [esp+block]
- kloop_prepare 0, 1, 8, 16, 9, 2, 3, 10, 17, 24, 32, 25, \
- 18, 11, 4, 5, 12, 19, 26, 33, 40, 48, 41, 34, \
- 27, 20, 13, 6, 7, 14, 21, 28, 35, \
- xmm0, xmm1, xmm2, xmm3
- kloop_prepare 32, 42, 49, 56, 57, 50, 43, 36, 29, 22, 15, 23, \
- 30, 37, 44, 51, 58, 59, 52, 45, 38, 31, 39, 46, \
- 53, 60, 61, 54, 47, 55, 62, 63, 63, \
- xmm0, xmm1, xmm2, xmm3
-
- pxor xmm7, xmm7
- movdqa xmm0, XMMWORD [esp + t1 + 0 * SIZEOF_WORD] ; __m128i tmp0 = _mm_loadu_si128((__m128i *)(t1 + 0));
- movdqa xmm1, XMMWORD [esp + t1 + 8 * SIZEOF_WORD] ; __m128i tmp1 = _mm_loadu_si128((__m128i *)(t1 + 8));
- movdqa xmm2, XMMWORD [esp + t1 + 16 * SIZEOF_WORD] ; __m128i tmp2 = _mm_loadu_si128((__m128i *)(t1 + 16));
- movdqa xmm3, XMMWORD [esp + t1 + 24 * SIZEOF_WORD] ; __m128i tmp3 = _mm_loadu_si128((__m128i *)(t1 + 24));
- pcmpeqw xmm0, xmm7 ; tmp0 = _mm_cmpeq_epi16(tmp0, zero);
- pcmpeqw xmm1, xmm7 ; tmp1 = _mm_cmpeq_epi16(tmp1, zero);
- pcmpeqw xmm2, xmm7 ; tmp2 = _mm_cmpeq_epi16(tmp2, zero);
- pcmpeqw xmm3, xmm7 ; tmp3 = _mm_cmpeq_epi16(tmp3, zero);
- packsswb xmm0, xmm1 ; tmp0 = _mm_packs_epi16(tmp0, tmp1);
- packsswb xmm2, xmm3 ; tmp2 = _mm_packs_epi16(tmp2, tmp3);
- pmovmskb edx, xmm0 ; index = ((uint64_t)_mm_movemask_epi8(tmp0)) << 0;
- pmovmskb ecx, xmm2 ; index = ((uint64_t)_mm_movemask_epi8(tmp2)) << 16;
- shl ecx, 16
- or edx, ecx
- not edx ; index = ~index;
-
- lea esi, [esp+t1]
- mov ebp, POINTER [esp+actbl] ; ebp = actbl
-
-.BLOOP:
- bsf ecx, edx ; r = __builtin_ctzl(index);
- jz near .ELOOP
- lea esi, [esi+ecx*2] ; k += r;
- shr edx, cl ; index >>= r;
- mov DWORD [esp+temp3], edx
-.BRLOOP:
- cmp ecx, 16 ; while (r > 15) {
- jl near .ERLOOP
- sub ecx, 16 ; r -= 16;
- mov DWORD [esp+temp], ecx
- mov eax, INT [ebp + 240 * 4] ; code_0xf0 = actbl->ehufco[0xf0];
- movzx ecx, byte [ebp + 1024 + 240] ; size_0xf0 = actbl->ehufsi[0xf0];
- EMIT_BITS eax ; EMIT_BITS(code_0xf0, size_0xf0)
- mov ecx, DWORD [esp+temp]
- jmp .BRLOOP
-.ERLOOP:
- movsx eax, word [esi] ; temp = t1[k];
- movpic edx, POINTER [esp+gotptr] ; load GOT address (edx)
- movzx eax, byte [GOTOFF(edx, jpeg_nbits_table + eax)] ; nbits = JPEG_NBITS(temp);
- mov DWORD [esp+temp2], eax
- ; Emit Huffman symbol for run length / number of bits
- shl ecx, 4 ; temp3 = (r << 4) + nbits;
- add ecx, eax
- mov eax, INT [ebp + ecx * 4] ; code = actbl->ehufco[temp3];
- movzx ecx, byte [ebp + ecx + 1024] ; size = actbl->ehufsi[temp3];
- EMIT_BITS eax
-
- movsx edx, word [esi+DCTSIZE2*2] ; temp2 = t2[k];
- ; Mask off any extra bits in code
- mov ecx, DWORD [esp+temp2]
- mov eax, 1
- shl eax, cl
- dec eax
- and eax, edx ; temp2 &= (((JLONG) 1)<<nbits) - 1;
- EMIT_BITS eax ; PUT_BITS(temp2, nbits)
- mov edx, DWORD [esp+temp3]
- add esi, 2 ; ++k;
- shr edx, 1 ; index >>= 1;
-
- jmp .BLOOP
-.ELOOP:
- movdqa xmm0, XMMWORD [esp + t1 + 32 * SIZEOF_WORD] ; __m128i tmp0 = _mm_loadu_si128((__m128i *)(t1 + 0));
- movdqa xmm1, XMMWORD [esp + t1 + 40 * SIZEOF_WORD] ; __m128i tmp1 = _mm_loadu_si128((__m128i *)(t1 + 8));
- movdqa xmm2, XMMWORD [esp + t1 + 48 * SIZEOF_WORD] ; __m128i tmp2 = _mm_loadu_si128((__m128i *)(t1 + 16));
- movdqa xmm3, XMMWORD [esp + t1 + 56 * SIZEOF_WORD] ; __m128i tmp3 = _mm_loadu_si128((__m128i *)(t1 + 24));
- pcmpeqw xmm0, xmm7 ; tmp0 = _mm_cmpeq_epi16(tmp0, zero);
- pcmpeqw xmm1, xmm7 ; tmp1 = _mm_cmpeq_epi16(tmp1, zero);
- pcmpeqw xmm2, xmm7 ; tmp2 = _mm_cmpeq_epi16(tmp2, zero);
- pcmpeqw xmm3, xmm7 ; tmp3 = _mm_cmpeq_epi16(tmp3, zero);
- packsswb xmm0, xmm1 ; tmp0 = _mm_packs_epi16(tmp0, tmp1);
- packsswb xmm2, xmm3 ; tmp2 = _mm_packs_epi16(tmp2, tmp3);
- pmovmskb edx, xmm0 ; index = ((uint64_t)_mm_movemask_epi8(tmp0)) << 0;
- pmovmskb ecx, xmm2 ; index = ((uint64_t)_mm_movemask_epi8(tmp2)) << 16;
- shl ecx, 16
- or edx, ecx
- not edx ; index = ~index;
-
- lea eax, [esp + t1 + (DCTSIZE2/2) * 2]
- sub eax, esi
- shr eax, 1
- bsf ecx, edx ; r = __builtin_ctzl(index);
- jz near .ELOOP2
- shr edx, cl ; index >>= r;
- add ecx, eax
- lea esi, [esi+ecx*2] ; k += r;
- mov DWORD [esp+temp3], edx
- jmp .BRLOOP2
-.BLOOP2:
- bsf ecx, edx ; r = __builtin_ctzl(index);
- jz near .ELOOP2
- lea esi, [esi+ecx*2] ; k += r;
- shr edx, cl ; index >>= r;
- mov DWORD [esp+temp3], edx
-.BRLOOP2:
- cmp ecx, 16 ; while (r > 15) {
- jl near .ERLOOP2
- sub ecx, 16 ; r -= 16;
- mov DWORD [esp+temp], ecx
- mov eax, INT [ebp + 240 * 4] ; code_0xf0 = actbl->ehufco[0xf0];
- movzx ecx, byte [ebp + 1024 + 240] ; size_0xf0 = actbl->ehufsi[0xf0];
- EMIT_BITS eax ; EMIT_BITS(code_0xf0, size_0xf0)
- mov ecx, DWORD [esp+temp]
- jmp .BRLOOP2
-.ERLOOP2:
- movsx eax, word [esi] ; temp = t1[k];
- bsr eax, eax ; nbits = 32 - __builtin_clz(temp);
- inc eax
- mov DWORD [esp+temp2], eax
- ; Emit Huffman symbol for run length / number of bits
- shl ecx, 4 ; temp3 = (r << 4) + nbits;
- add ecx, eax
- mov eax, INT [ebp + ecx * 4] ; code = actbl->ehufco[temp3];
- movzx ecx, byte [ebp + ecx + 1024] ; size = actbl->ehufsi[temp3];
- EMIT_BITS eax
-
- movsx edx, word [esi+DCTSIZE2*2] ; temp2 = t2[k];
- ; Mask off any extra bits in code
- mov ecx, DWORD [esp+temp2]
- mov eax, 1
- shl eax, cl
- dec eax
- and eax, edx ; temp2 &= (((JLONG) 1)<<nbits) - 1;
- EMIT_BITS eax ; PUT_BITS(temp2, nbits)
- mov edx, DWORD [esp+temp3]
- add esi, 2 ; ++k;
- shr edx, 1 ; index >>= 1;
-
- jmp .BLOOP2
-.ELOOP2:
- ; If the last coef(s) were zero, emit an end-of-block code
- lea edx, [esp + t1 + (DCTSIZE2-1) * 2] ; r = DCTSIZE2-1-k;
- cmp edx, esi ; if (r > 0) {
- je .EFN
- mov eax, INT [ebp] ; code = actbl->ehufco[0];
- movzx ecx, byte [ebp + 1024] ; size = actbl->ehufsi[0];
- EMIT_BITS eax
-.EFN:
- mov eax, [esp+buffer]
- pop esi
- ; Save put_buffer & put_bits
- mov DWORD [esi+8], put_buffer ; state->cur.put_buffer = put_buffer;
- mov DWORD [esi+12], put_bits ; state->cur.put_bits = put_bits;
-
- pop ebp
- pop edi
- pop esi
-; pop edx ; need not be preserved
- pop ecx
- pop ebx
- mov esp,ebp ; esp <- aligned ebp
- pop esp ; esp <- original ebp
- pop ebp
- ret
-
-; For some reason, the OS X linker does not honor the request to align the
-; segment unless we do this.
- align 16
diff --git a/simd/jcsample-sse2-64.asm b/simd/jcsample-sse2-64.asm
deleted file mode 100644
index 40ee15f..0000000
--- a/simd/jcsample-sse2-64.asm
+++ /dev/null
@@ -1,329 +0,0 @@
-;
-; jcsample.asm - downsampling (64-bit SSE2)
-;
-; Copyright 2009 Pierre Ossman <ossman@cendio.se> for Cendio AB
-; Copyright (C) 2009, D. R. Commander.
-;
-; Based on the x86 SIMD extension for IJG JPEG library
-; Copyright (C) 1999-2006, MIYASAKA Masaru.
-; For conditions of distribution and use, see copyright notice in jsimdext.inc
-;
-; This file should be assembled with NASM (Netwide Assembler),
-; can *not* be assembled with Microsoft's MASM or any compatible
-; assembler (including Borland's Turbo Assembler).
-; NASM is available from http://nasm.sourceforge.net/ or
-; http://sourceforge.net/project/showfiles.php?group_id=6208
-;
-; [TAB8]
-
-%include "jsimdext.inc"
-
-; --------------------------------------------------------------------------
- SECTION SEG_TEXT
- BITS 64
-;
-; Downsample pixel values of a single component.
-; This version handles the common case of 2:1 horizontal and 1:1 vertical,
-; without smoothing.
-;
-; GLOBAL(void)
-; jsimd_h2v1_downsample_sse2 (JDIMENSION image_width, int max_v_samp_factor,
-; JDIMENSION v_samp_factor, JDIMENSION width_blocks,
-; JSAMPARRAY input_data, JSAMPARRAY output_data);
-;
-
-; r10 = JDIMENSION image_width
-; r11 = int max_v_samp_factor
-; r12 = JDIMENSION v_samp_factor
-; r13 = JDIMENSION width_blocks
-; r14 = JSAMPARRAY input_data
-; r15 = JSAMPARRAY output_data
-
- align 16
- global EXTN(jsimd_h2v1_downsample_sse2)
-
-EXTN(jsimd_h2v1_downsample_sse2):
- push rbp
- mov rax,rsp
- mov rbp,rsp
- collect_args
-
- mov ecx, r13d
- shl rcx,3 ; imul rcx,DCTSIZE (rcx = output_cols)
- jz near .return
-
- mov edx, r10d
-
- ; -- expand_right_edge
-
- push rcx
- shl rcx,1 ; output_cols * 2
- sub rcx,rdx
- jle short .expand_end
-
- mov rax, r11
- test rax,rax
- jle short .expand_end
-
- cld
- mov rsi, r14 ; input_data
-.expandloop:
- push rax
- push rcx
-
- mov rdi, JSAMPROW [rsi]
- add rdi,rdx
- mov al, JSAMPLE [rdi-1]
-
- rep stosb
-
- pop rcx
- pop rax
-
- add rsi, byte SIZEOF_JSAMPROW
- dec rax
- jg short .expandloop
-
-.expand_end:
- pop rcx ; output_cols
-
- ; -- h2v1_downsample
-
- mov eax, r12d ; rowctr
- test eax,eax
- jle near .return
-
- mov rdx, 0x00010000 ; bias pattern
- movd xmm7,edx
- pcmpeqw xmm6,xmm6
- pshufd xmm7,xmm7,0x00 ; xmm7={0, 1, 0, 1, 0, 1, 0, 1}
- psrlw xmm6,BYTE_BIT ; xmm6={0xFF 0x00 0xFF 0x00 ..}
-
- mov rsi, r14 ; input_data
- mov rdi, r15 ; output_data
-.rowloop:
- push rcx
- push rdi
- push rsi
-
- mov rsi, JSAMPROW [rsi] ; inptr
- mov rdi, JSAMPROW [rdi] ; outptr
-
- cmp rcx, byte SIZEOF_XMMWORD
- jae short .columnloop
-
-.columnloop_r8:
- movdqa xmm0, XMMWORD [rsi+0*SIZEOF_XMMWORD]
- pxor xmm1,xmm1
- mov rcx, SIZEOF_XMMWORD
- jmp short .downsample
-
-.columnloop:
- movdqa xmm0, XMMWORD [rsi+0*SIZEOF_XMMWORD]
- movdqa xmm1, XMMWORD [rsi+1*SIZEOF_XMMWORD]
-
-.downsample:
- movdqa xmm2,xmm0
- movdqa xmm3,xmm1
-
- pand xmm0,xmm6
- psrlw xmm2,BYTE_BIT
- pand xmm1,xmm6
- psrlw xmm3,BYTE_BIT
-
- paddw xmm0,xmm2
- paddw xmm1,xmm3
- paddw xmm0,xmm7
- paddw xmm1,xmm7
- psrlw xmm0,1
- psrlw xmm1,1
-
- packuswb xmm0,xmm1
-
- movdqa XMMWORD [rdi+0*SIZEOF_XMMWORD], xmm0
-
- sub rcx, byte SIZEOF_XMMWORD ; outcol
- add rsi, byte 2*SIZEOF_XMMWORD ; inptr
- add rdi, byte 1*SIZEOF_XMMWORD ; outptr
- cmp rcx, byte SIZEOF_XMMWORD
- jae short .columnloop
- test rcx,rcx
- jnz short .columnloop_r8
-
- pop rsi
- pop rdi
- pop rcx
-
- add rsi, byte SIZEOF_JSAMPROW ; input_data
- add rdi, byte SIZEOF_JSAMPROW ; output_data
- dec rax ; rowctr
- jg near .rowloop
-
-.return:
- uncollect_args
- pop rbp
- ret
-
-; --------------------------------------------------------------------------
-;
-; Downsample pixel values of a single component.
-; This version handles the standard case of 2:1 horizontal and 2:1 vertical,
-; without smoothing.
-;
-; GLOBAL(void)
-; jsimd_h2v2_downsample_sse2 (JDIMENSION image_width, int max_v_samp_factor,
-; JDIMENSION v_samp_factor, JDIMENSION width_blocks,
-; JSAMPARRAY input_data, JSAMPARRAY output_data);
-;
-
-; r10 = JDIMENSION image_width
-; r11 = int max_v_samp_factor
-; r12 = JDIMENSION v_samp_factor
-; r13 = JDIMENSION width_blocks
-; r14 = JSAMPARRAY input_data
-; r15 = JSAMPARRAY output_data
-
- align 16
- global EXTN(jsimd_h2v2_downsample_sse2)
-
-EXTN(jsimd_h2v2_downsample_sse2):
- push rbp
- mov rax,rsp
- mov rbp,rsp
- collect_args
-
- mov ecx, r13d
- shl rcx,3 ; imul rcx,DCTSIZE (rcx = output_cols)
- jz near .return
-
- mov edx, r10d
-
- ; -- expand_right_edge
-
- push rcx
- shl rcx,1 ; output_cols * 2
- sub rcx,rdx
- jle short .expand_end
-
- mov rax, r11
- test rax,rax
- jle short .expand_end
-
- cld
- mov rsi, r14 ; input_data
-.expandloop:
- push rax
- push rcx
-
- mov rdi, JSAMPROW [rsi]
- add rdi,rdx
- mov al, JSAMPLE [rdi-1]
-
- rep stosb
-
- pop rcx
- pop rax
-
- add rsi, byte SIZEOF_JSAMPROW
- dec rax
- jg short .expandloop
-
-.expand_end:
- pop rcx ; output_cols
-
- ; -- h2v2_downsample
-
- mov eax, r12d ; rowctr
- test rax,rax
- jle near .return
-
- mov rdx, 0x00020001 ; bias pattern
- movd xmm7,edx
- pcmpeqw xmm6,xmm6
- pshufd xmm7,xmm7,0x00 ; xmm7={1, 2, 1, 2, 1, 2, 1, 2}
- psrlw xmm6,BYTE_BIT ; xmm6={0xFF 0x00 0xFF 0x00 ..}
-
- mov rsi, r14 ; input_data
- mov rdi, r15 ; output_data
-.rowloop:
- push rcx
- push rdi
- push rsi
-
- mov rdx, JSAMPROW [rsi+0*SIZEOF_JSAMPROW] ; inptr0
- mov rsi, JSAMPROW [rsi+1*SIZEOF_JSAMPROW] ; inptr1
- mov rdi, JSAMPROW [rdi] ; outptr
-
- cmp rcx, byte SIZEOF_XMMWORD
- jae short .columnloop
-
-.columnloop_r8:
- movdqa xmm0, XMMWORD [rdx+0*SIZEOF_XMMWORD]
- movdqa xmm1, XMMWORD [rsi+0*SIZEOF_XMMWORD]
- pxor xmm2,xmm2
- pxor xmm3,xmm3
- mov rcx, SIZEOF_XMMWORD
- jmp short .downsample
-
-.columnloop:
- movdqa xmm0, XMMWORD [rdx+0*SIZEOF_XMMWORD]
- movdqa xmm1, XMMWORD [rsi+0*SIZEOF_XMMWORD]
- movdqa xmm2, XMMWORD [rdx+1*SIZEOF_XMMWORD]
- movdqa xmm3, XMMWORD [rsi+1*SIZEOF_XMMWORD]
-
-.downsample:
- movdqa xmm4,xmm0
- movdqa xmm5,xmm1
- pand xmm0,xmm6
- psrlw xmm4,BYTE_BIT
- pand xmm1,xmm6
- psrlw xmm5,BYTE_BIT
- paddw xmm0,xmm4
- paddw xmm1,xmm5
-
- movdqa xmm4,xmm2
- movdqa xmm5,xmm3
- pand xmm2,xmm6
- psrlw xmm4,BYTE_BIT
- pand xmm3,xmm6
- psrlw xmm5,BYTE_BIT
- paddw xmm2,xmm4
- paddw xmm3,xmm5
-
- paddw xmm0,xmm1
- paddw xmm2,xmm3
- paddw xmm0,xmm7
- paddw xmm2,xmm7
- psrlw xmm0,2
- psrlw xmm2,2
-
- packuswb xmm0,xmm2
-
- movdqa XMMWORD [rdi+0*SIZEOF_XMMWORD], xmm0
-
- sub rcx, byte SIZEOF_XMMWORD ; outcol
- add rdx, byte 2*SIZEOF_XMMWORD ; inptr0
- add rsi, byte 2*SIZEOF_XMMWORD ; inptr1
- add rdi, byte 1*SIZEOF_XMMWORD ; outptr
- cmp rcx, byte SIZEOF_XMMWORD
- jae near .columnloop
- test rcx,rcx
- jnz near .columnloop_r8
-
- pop rsi
- pop rdi
- pop rcx
-
- add rsi, byte 2*SIZEOF_JSAMPROW ; input_data
- add rdi, byte 1*SIZEOF_JSAMPROW ; output_data
- dec rax ; rowctr
- jg near .rowloop
-
-.return:
- uncollect_args
- pop rbp
- ret
-
-; For some reason, the OS X linker does not honor the request to align the
-; segment unless we do this.
- align 16
diff --git a/simd/jcsample-sse2.asm b/simd/jcsample-sse2.asm
deleted file mode 100644
index 83c9d15..0000000
--- a/simd/jcsample-sse2.asm
+++ /dev/null
@@ -1,350 +0,0 @@
-;
-; jcsample.asm - downsampling (SSE2)
-;
-; Copyright 2009 Pierre Ossman <ossman@cendio.se> for Cendio AB
-;
-; Based on the x86 SIMD extension for IJG JPEG library
-; Copyright (C) 1999-2006, MIYASAKA Masaru.
-; For conditions of distribution and use, see copyright notice in jsimdext.inc
-;
-; This file should be assembled with NASM (Netwide Assembler),
-; can *not* be assembled with Microsoft's MASM or any compatible
-; assembler (including Borland's Turbo Assembler).
-; NASM is available from http://nasm.sourceforge.net/ or
-; http://sourceforge.net/project/showfiles.php?group_id=6208
-;
-; [TAB8]
-
-%include "jsimdext.inc"
-
-; --------------------------------------------------------------------------
- SECTION SEG_TEXT
- BITS 32
-;
-; Downsample pixel values of a single component.
-; This version handles the common case of 2:1 horizontal and 1:1 vertical,
-; without smoothing.
-;
-; GLOBAL(void)
-; jsimd_h2v1_downsample_sse2 (JDIMENSION image_width, int max_v_samp_factor,
-; JDIMENSION v_samp_factor, JDIMENSION width_blocks,
-; JSAMPARRAY input_data, JSAMPARRAY output_data);
-;
-
-%define img_width(b) (b)+8 ; JDIMENSION image_width
-%define max_v_samp(b) (b)+12 ; int max_v_samp_factor
-%define v_samp(b) (b)+16 ; JDIMENSION v_samp_factor
-%define width_blks(b) (b)+20 ; JDIMENSION width_blocks
-%define input_data(b) (b)+24 ; JSAMPARRAY input_data
-%define output_data(b) (b)+28 ; JSAMPARRAY output_data
-
- align 16
- global EXTN(jsimd_h2v1_downsample_sse2)
-
-EXTN(jsimd_h2v1_downsample_sse2):
- push ebp
- mov ebp,esp
-; push ebx ; unused
-; push ecx ; need not be preserved
-; push edx ; need not be preserved
- push esi
- push edi
-
- mov ecx, JDIMENSION [width_blks(ebp)]
- shl ecx,3 ; imul ecx,DCTSIZE (ecx = output_cols)
- jz near .return
-
- mov edx, JDIMENSION [img_width(ebp)]
-
- ; -- expand_right_edge
-
- push ecx
- shl ecx,1 ; output_cols * 2
- sub ecx,edx
- jle short .expand_end
-
- mov eax, INT [max_v_samp(ebp)]
- test eax,eax
- jle short .expand_end
-
- cld
- mov esi, JSAMPARRAY [input_data(ebp)] ; input_data
- alignx 16,7
-.expandloop:
- push eax
- push ecx
-
- mov edi, JSAMPROW [esi]
- add edi,edx
- mov al, JSAMPLE [edi-1]
-
- rep stosb
-
- pop ecx
- pop eax
-
- add esi, byte SIZEOF_JSAMPROW
- dec eax
- jg short .expandloop
-
-.expand_end:
- pop ecx ; output_cols
-
- ; -- h2v1_downsample
-
- mov eax, JDIMENSION [v_samp(ebp)] ; rowctr
- test eax,eax
- jle near .return
-
- mov edx, 0x00010000 ; bias pattern
- movd xmm7,edx
- pcmpeqw xmm6,xmm6
- pshufd xmm7,xmm7,0x00 ; xmm7={0, 1, 0, 1, 0, 1, 0, 1}
- psrlw xmm6,BYTE_BIT ; xmm6={0xFF 0x00 0xFF 0x00 ..}
-
- mov esi, JSAMPARRAY [input_data(ebp)] ; input_data
- mov edi, JSAMPARRAY [output_data(ebp)] ; output_data
- alignx 16,7
-.rowloop:
- push ecx
- push edi
- push esi
-
- mov esi, JSAMPROW [esi] ; inptr
- mov edi, JSAMPROW [edi] ; outptr
-
- cmp ecx, byte SIZEOF_XMMWORD
- jae short .columnloop
- alignx 16,7
-
-.columnloop_r8:
- movdqa xmm0, XMMWORD [esi+0*SIZEOF_XMMWORD]
- pxor xmm1,xmm1
- mov ecx, SIZEOF_XMMWORD
- jmp short .downsample
- alignx 16,7
-
-.columnloop:
- movdqa xmm0, XMMWORD [esi+0*SIZEOF_XMMWORD]
- movdqa xmm1, XMMWORD [esi+1*SIZEOF_XMMWORD]
-
-.downsample:
- movdqa xmm2,xmm0
- movdqa xmm3,xmm1
-
- pand xmm0,xmm6
- psrlw xmm2,BYTE_BIT
- pand xmm1,xmm6
- psrlw xmm3,BYTE_BIT
-
- paddw xmm0,xmm2
- paddw xmm1,xmm3
- paddw xmm0,xmm7
- paddw xmm1,xmm7
- psrlw xmm0,1
- psrlw xmm1,1
-
- packuswb xmm0,xmm1
-
- movdqa XMMWORD [edi+0*SIZEOF_XMMWORD], xmm0
-
- sub ecx, byte SIZEOF_XMMWORD ; outcol
- add esi, byte 2*SIZEOF_XMMWORD ; inptr
- add edi, byte 1*SIZEOF_XMMWORD ; outptr
- cmp ecx, byte SIZEOF_XMMWORD
- jae short .columnloop
- test ecx,ecx
- jnz short .columnloop_r8
-
- pop esi
- pop edi
- pop ecx
-
- add esi, byte SIZEOF_JSAMPROW ; input_data
- add edi, byte SIZEOF_JSAMPROW ; output_data
- dec eax ; rowctr
- jg near .rowloop
-
-.return:
- pop edi
- pop esi
-; pop edx ; need not be preserved
-; pop ecx ; need not be preserved
-; pop ebx ; unused
- pop ebp
- ret
-
-; --------------------------------------------------------------------------
-;
-; Downsample pixel values of a single component.
-; This version handles the standard case of 2:1 horizontal and 2:1 vertical,
-; without smoothing.
-;
-; GLOBAL(void)
-; jsimd_h2v2_downsample_sse2 (JDIMENSION image_width, int max_v_samp_factor,
-; JDIMENSION v_samp_factor, JDIMENSION width_blocks,
-; JSAMPARRAY input_data, JSAMPARRAY output_data);
-;
-
-%define img_width(b) (b)+8 ; JDIMENSION image_width
-%define max_v_samp(b) (b)+12 ; int max_v_samp_factor
-%define v_samp(b) (b)+16 ; JDIMENSION v_samp_factor
-%define width_blks(b) (b)+20 ; JDIMENSION width_blocks
-%define input_data(b) (b)+24 ; JSAMPARRAY input_data
-%define output_data(b) (b)+28 ; JSAMPARRAY output_data
-
- align 16
- global EXTN(jsimd_h2v2_downsample_sse2)
-
-EXTN(jsimd_h2v2_downsample_sse2):
- push ebp
- mov ebp,esp
-; push ebx ; unused
-; push ecx ; need not be preserved
-; push edx ; need not be preserved
- push esi
- push edi
-
- mov ecx, JDIMENSION [width_blks(ebp)]
- shl ecx,3 ; imul ecx,DCTSIZE (ecx = output_cols)
- jz near .return
-
- mov edx, JDIMENSION [img_width(ebp)]
-
- ; -- expand_right_edge
-
- push ecx
- shl ecx,1 ; output_cols * 2
- sub ecx,edx
- jle short .expand_end
-
- mov eax, INT [max_v_samp(ebp)]
- test eax,eax
- jle short .expand_end
-
- cld
- mov esi, JSAMPARRAY [input_data(ebp)] ; input_data
- alignx 16,7
-.expandloop:
- push eax
- push ecx
-
- mov edi, JSAMPROW [esi]
- add edi,edx
- mov al, JSAMPLE [edi-1]
-
- rep stosb
-
- pop ecx
- pop eax
-
- add esi, byte SIZEOF_JSAMPROW
- dec eax
- jg short .expandloop
-
-.expand_end:
- pop ecx ; output_cols
-
- ; -- h2v2_downsample
-
- mov eax, JDIMENSION [v_samp(ebp)] ; rowctr
- test eax,eax
- jle near .return
-
- mov edx, 0x00020001 ; bias pattern
- movd xmm7,edx
- pcmpeqw xmm6,xmm6
- pshufd xmm7,xmm7,0x00 ; xmm7={1, 2, 1, 2, 1, 2, 1, 2}
- psrlw xmm6,BYTE_BIT ; xmm6={0xFF 0x00 0xFF 0x00 ..}
-
- mov esi, JSAMPARRAY [input_data(ebp)] ; input_data
- mov edi, JSAMPARRAY [output_data(ebp)] ; output_data
- alignx 16,7
-.rowloop:
- push ecx
- push edi
- push esi
-
- mov edx, JSAMPROW [esi+0*SIZEOF_JSAMPROW] ; inptr0
- mov esi, JSAMPROW [esi+1*SIZEOF_JSAMPROW] ; inptr1
- mov edi, JSAMPROW [edi] ; outptr
-
- cmp ecx, byte SIZEOF_XMMWORD
- jae short .columnloop
- alignx 16,7
-
-.columnloop_r8:
- movdqa xmm0, XMMWORD [edx+0*SIZEOF_XMMWORD]
- movdqa xmm1, XMMWORD [esi+0*SIZEOF_XMMWORD]
- pxor xmm2,xmm2
- pxor xmm3,xmm3
- mov ecx, SIZEOF_XMMWORD
- jmp short .downsample
- alignx 16,7
-
-.columnloop:
- movdqa xmm0, XMMWORD [edx+0*SIZEOF_XMMWORD]
- movdqa xmm1, XMMWORD [esi+0*SIZEOF_XMMWORD]
- movdqa xmm2, XMMWORD [edx+1*SIZEOF_XMMWORD]
- movdqa xmm3, XMMWORD [esi+1*SIZEOF_XMMWORD]
-
-.downsample:
- movdqa xmm4,xmm0
- movdqa xmm5,xmm1
- pand xmm0,xmm6
- psrlw xmm4,BYTE_BIT
- pand xmm1,xmm6
- psrlw xmm5,BYTE_BIT
- paddw xmm0,xmm4
- paddw xmm1,xmm5
-
- movdqa xmm4,xmm2
- movdqa xmm5,xmm3
- pand xmm2,xmm6
- psrlw xmm4,BYTE_BIT
- pand xmm3,xmm6
- psrlw xmm5,BYTE_BIT
- paddw xmm2,xmm4
- paddw xmm3,xmm5
-
- paddw xmm0,xmm1
- paddw xmm2,xmm3
- paddw xmm0,xmm7
- paddw xmm2,xmm7
- psrlw xmm0,2
- psrlw xmm2,2
-
- packuswb xmm0,xmm2
-
- movdqa XMMWORD [edi+0*SIZEOF_XMMWORD], xmm0
-
- sub ecx, byte SIZEOF_XMMWORD ; outcol
- add edx, byte 2*SIZEOF_XMMWORD ; inptr0
- add esi, byte 2*SIZEOF_XMMWORD ; inptr1
- add edi, byte 1*SIZEOF_XMMWORD ; outptr
- cmp ecx, byte SIZEOF_XMMWORD
- jae near .columnloop
- test ecx,ecx
- jnz near .columnloop_r8
-
- pop esi
- pop edi
- pop ecx
-
- add esi, byte 2*SIZEOF_JSAMPROW ; input_data
- add edi, byte 1*SIZEOF_JSAMPROW ; output_data
- dec eax ; rowctr
- jg near .rowloop
-
-.return:
- pop edi
- pop esi
-; pop edx ; need not be preserved
-; pop ecx ; need not be preserved
-; pop ebx ; unused
- pop ebp
- ret
-
-; For some reason, the OS X linker does not honor the request to align the
-; segment unless we do this.
- align 16
diff --git a/simd/jdcolext-sse2-64.asm b/simd/jdcolext-sse2-64.asm
deleted file mode 100644
index 4634066..0000000
--- a/simd/jdcolext-sse2-64.asm
+++ /dev/null
@@ -1,440 +0,0 @@
-;
-; jdcolext.asm - colorspace conversion (64-bit SSE2)
-;
-; Copyright 2009, 2012 Pierre Ossman <ossman@cendio.se> for Cendio AB
-; Copyright (C) 2009, 2012, D. R. Commander.
-;
-; Based on the x86 SIMD extension for IJG JPEG library
-; Copyright (C) 1999-2006, MIYASAKA Masaru.
-; For conditions of distribution and use, see copyright notice in jsimdext.inc
-;
-; This file should be assembled with NASM (Netwide Assembler),
-; can *not* be assembled with Microsoft's MASM or any compatible
-; assembler (including Borland's Turbo Assembler).
-; NASM is available from http://nasm.sourceforge.net/ or
-; http://sourceforge.net/project/showfiles.php?group_id=6208
-;
-; [TAB8]
-
-%include "jcolsamp.inc"
-
-; --------------------------------------------------------------------------
-;
-; Convert some rows of samples to the output colorspace.
-;
-; GLOBAL(void)
-; jsimd_ycc_rgb_convert_sse2 (JDIMENSION out_width,
-; JSAMPIMAGE input_buf, JDIMENSION input_row,
-; JSAMPARRAY output_buf, int num_rows)
-;
-
-; r10 = JDIMENSION out_width
-; r11 = JSAMPIMAGE input_buf
-; r12 = JDIMENSION input_row
-; r13 = JSAMPARRAY output_buf
-; r14 = int num_rows
-
-%define wk(i) rbp-(WK_NUM-(i))*SIZEOF_XMMWORD ; xmmword wk[WK_NUM]
-%define WK_NUM 2
-
- align 16
- global EXTN(jsimd_ycc_rgb_convert_sse2)
-
-EXTN(jsimd_ycc_rgb_convert_sse2):
- push rbp
- mov rax,rsp ; rax = original rbp
- sub rsp, byte 4
- and rsp, byte (-SIZEOF_XMMWORD) ; align to 128 bits
- mov [rsp],rax
- mov rbp,rsp ; rbp = aligned rbp
- lea rsp, [wk(0)]
- collect_args
- push rbx
-
- mov ecx, r10d ; num_cols
- test rcx,rcx
- jz near .return
-
- push rcx
-
- mov rdi, r11
- mov ecx, r12d
- mov rsi, JSAMPARRAY [rdi+0*SIZEOF_JSAMPARRAY]
- mov rbx, JSAMPARRAY [rdi+1*SIZEOF_JSAMPARRAY]
- mov rdx, JSAMPARRAY [rdi+2*SIZEOF_JSAMPARRAY]
- lea rsi, [rsi+rcx*SIZEOF_JSAMPROW]
- lea rbx, [rbx+rcx*SIZEOF_JSAMPROW]
- lea rdx, [rdx+rcx*SIZEOF_JSAMPROW]
-
- pop rcx
-
- mov rdi, r13
- mov eax, r14d
- test rax,rax
- jle near .return
-.rowloop:
- push rax
- push rdi
- push rdx
- push rbx
- push rsi
- push rcx ; col
-
- mov rsi, JSAMPROW [rsi] ; inptr0
- mov rbx, JSAMPROW [rbx] ; inptr1
- mov rdx, JSAMPROW [rdx] ; inptr2
- mov rdi, JSAMPROW [rdi] ; outptr
-.columnloop:
-
- movdqa xmm5, XMMWORD [rbx] ; xmm5=Cb(0123456789ABCDEF)
- movdqa xmm1, XMMWORD [rdx] ; xmm1=Cr(0123456789ABCDEF)
-
- pcmpeqw xmm4,xmm4
- pcmpeqw xmm7,xmm7
- psrlw xmm4,BYTE_BIT
- psllw xmm7,7 ; xmm7={0xFF80 0xFF80 0xFF80 0xFF80 ..}
- movdqa xmm0,xmm4 ; xmm0=xmm4={0xFF 0x00 0xFF 0x00 ..}
-
- pand xmm4,xmm5 ; xmm4=Cb(02468ACE)=CbE
- psrlw xmm5,BYTE_BIT ; xmm5=Cb(13579BDF)=CbO
- pand xmm0,xmm1 ; xmm0=Cr(02468ACE)=CrE
- psrlw xmm1,BYTE_BIT ; xmm1=Cr(13579BDF)=CrO
-
- paddw xmm4,xmm7
- paddw xmm5,xmm7
- paddw xmm0,xmm7
- paddw xmm1,xmm7
-
- ; (Original)
- ; R = Y + 1.40200 * Cr
- ; G = Y - 0.34414 * Cb - 0.71414 * Cr
- ; B = Y + 1.77200 * Cb
- ;
- ; (This implementation)
- ; R = Y + 0.40200 * Cr + Cr
- ; G = Y - 0.34414 * Cb + 0.28586 * Cr - Cr
- ; B = Y - 0.22800 * Cb + Cb + Cb
-
- movdqa xmm2,xmm4 ; xmm2=CbE
- movdqa xmm3,xmm5 ; xmm3=CbO
- paddw xmm4,xmm4 ; xmm4=2*CbE
- paddw xmm5,xmm5 ; xmm5=2*CbO
- movdqa xmm6,xmm0 ; xmm6=CrE
- movdqa xmm7,xmm1 ; xmm7=CrO
- paddw xmm0,xmm0 ; xmm0=2*CrE
- paddw xmm1,xmm1 ; xmm1=2*CrO
-
- pmulhw xmm4,[rel PW_MF0228] ; xmm4=(2*CbE * -FIX(0.22800))
- pmulhw xmm5,[rel PW_MF0228] ; xmm5=(2*CbO * -FIX(0.22800))
- pmulhw xmm0,[rel PW_F0402] ; xmm0=(2*CrE * FIX(0.40200))
- pmulhw xmm1,[rel PW_F0402] ; xmm1=(2*CrO * FIX(0.40200))
-
- paddw xmm4,[rel PW_ONE]
- paddw xmm5,[rel PW_ONE]
- psraw xmm4,1 ; xmm4=(CbE * -FIX(0.22800))
- psraw xmm5,1 ; xmm5=(CbO * -FIX(0.22800))
- paddw xmm0,[rel PW_ONE]
- paddw xmm1,[rel PW_ONE]
- psraw xmm0,1 ; xmm0=(CrE * FIX(0.40200))
- psraw xmm1,1 ; xmm1=(CrO * FIX(0.40200))
-
- paddw xmm4,xmm2
- paddw xmm5,xmm3
- paddw xmm4,xmm2 ; xmm4=(CbE * FIX(1.77200))=(B-Y)E
- paddw xmm5,xmm3 ; xmm5=(CbO * FIX(1.77200))=(B-Y)O
- paddw xmm0,xmm6 ; xmm0=(CrE * FIX(1.40200))=(R-Y)E
- paddw xmm1,xmm7 ; xmm1=(CrO * FIX(1.40200))=(R-Y)O
-
- movdqa XMMWORD [wk(0)], xmm4 ; wk(0)=(B-Y)E
- movdqa XMMWORD [wk(1)], xmm5 ; wk(1)=(B-Y)O
-
- movdqa xmm4,xmm2
- movdqa xmm5,xmm3
- punpcklwd xmm2,xmm6
- punpckhwd xmm4,xmm6
- pmaddwd xmm2,[rel PW_MF0344_F0285]
- pmaddwd xmm4,[rel PW_MF0344_F0285]
- punpcklwd xmm3,xmm7
- punpckhwd xmm5,xmm7
- pmaddwd xmm3,[rel PW_MF0344_F0285]
- pmaddwd xmm5,[rel PW_MF0344_F0285]
-
- paddd xmm2,[rel PD_ONEHALF]
- paddd xmm4,[rel PD_ONEHALF]
- psrad xmm2,SCALEBITS
- psrad xmm4,SCALEBITS
- paddd xmm3,[rel PD_ONEHALF]
- paddd xmm5,[rel PD_ONEHALF]
- psrad xmm3,SCALEBITS
- psrad xmm5,SCALEBITS
-
- packssdw xmm2,xmm4 ; xmm2=CbE*-FIX(0.344)+CrE*FIX(0.285)
- packssdw xmm3,xmm5 ; xmm3=CbO*-FIX(0.344)+CrO*FIX(0.285)
- psubw xmm2,xmm6 ; xmm2=CbE*-FIX(0.344)+CrE*-FIX(0.714)=(G-Y)E
- psubw xmm3,xmm7 ; xmm3=CbO*-FIX(0.344)+CrO*-FIX(0.714)=(G-Y)O
-
- movdqa xmm5, XMMWORD [rsi] ; xmm5=Y(0123456789ABCDEF)
-
- pcmpeqw xmm4,xmm4
- psrlw xmm4,BYTE_BIT ; xmm4={0xFF 0x00 0xFF 0x00 ..}
- pand xmm4,xmm5 ; xmm4=Y(02468ACE)=YE
- psrlw xmm5,BYTE_BIT ; xmm5=Y(13579BDF)=YO
-
- paddw xmm0,xmm4 ; xmm0=((R-Y)E+YE)=RE=R(02468ACE)
- paddw xmm1,xmm5 ; xmm1=((R-Y)O+YO)=RO=R(13579BDF)
- packuswb xmm0,xmm0 ; xmm0=R(02468ACE********)
- packuswb xmm1,xmm1 ; xmm1=R(13579BDF********)
-
- paddw xmm2,xmm4 ; xmm2=((G-Y)E+YE)=GE=G(02468ACE)
- paddw xmm3,xmm5 ; xmm3=((G-Y)O+YO)=GO=G(13579BDF)
- packuswb xmm2,xmm2 ; xmm2=G(02468ACE********)
- packuswb xmm3,xmm3 ; xmm3=G(13579BDF********)
-
- paddw xmm4, XMMWORD [wk(0)] ; xmm4=(YE+(B-Y)E)=BE=B(02468ACE)
- paddw xmm5, XMMWORD [wk(1)] ; xmm5=(YO+(B-Y)O)=BO=B(13579BDF)
- packuswb xmm4,xmm4 ; xmm4=B(02468ACE********)
- packuswb xmm5,xmm5 ; xmm5=B(13579BDF********)
-
-%if RGB_PIXELSIZE == 3 ; ---------------
-
- ; xmmA=(00 02 04 06 08 0A 0C 0E **), xmmB=(01 03 05 07 09 0B 0D 0F **)
- ; xmmC=(10 12 14 16 18 1A 1C 1E **), xmmD=(11 13 15 17 19 1B 1D 1F **)
- ; xmmE=(20 22 24 26 28 2A 2C 2E **), xmmF=(21 23 25 27 29 2B 2D 2F **)
- ; xmmG=(** ** ** ** ** ** ** ** **), xmmH=(** ** ** ** ** ** ** ** **)
-
- punpcklbw xmmA,xmmC ; xmmA=(00 10 02 12 04 14 06 16 08 18 0A 1A 0C 1C 0E 1E)
- punpcklbw xmmE,xmmB ; xmmE=(20 01 22 03 24 05 26 07 28 09 2A 0B 2C 0D 2E 0F)
- punpcklbw xmmD,xmmF ; xmmD=(11 21 13 23 15 25 17 27 19 29 1B 2B 1D 2D 1F 2F)
-
- movdqa xmmG,xmmA
- movdqa xmmH,xmmA
- punpcklwd xmmA,xmmE ; xmmA=(00 10 20 01 02 12 22 03 04 14 24 05 06 16 26 07)
- punpckhwd xmmG,xmmE ; xmmG=(08 18 28 09 0A 1A 2A 0B 0C 1C 2C 0D 0E 1E 2E 0F)
-
- psrldq xmmH,2 ; xmmH=(02 12 04 14 06 16 08 18 0A 1A 0C 1C 0E 1E -- --)
- psrldq xmmE,2 ; xmmE=(22 03 24 05 26 07 28 09 2A 0B 2C 0D 2E 0F -- --)
-
- movdqa xmmC,xmmD
- movdqa xmmB,xmmD
- punpcklwd xmmD,xmmH ; xmmD=(11 21 02 12 13 23 04 14 15 25 06 16 17 27 08 18)
- punpckhwd xmmC,xmmH ; xmmC=(19 29 0A 1A 1B 2B 0C 1C 1D 2D 0E 1E 1F 2F -- --)
-
- psrldq xmmB,2 ; xmmB=(13 23 15 25 17 27 19 29 1B 2B 1D 2D 1F 2F -- --)
-
- movdqa xmmF,xmmE
- punpcklwd xmmE,xmmB ; xmmE=(22 03 13 23 24 05 15 25 26 07 17 27 28 09 19 29)
- punpckhwd xmmF,xmmB ; xmmF=(2A 0B 1B 2B 2C 0D 1D 2D 2E 0F 1F 2F -- -- -- --)
-
- pshufd xmmH,xmmA,0x4E; xmmH=(04 14 24 05 06 16 26 07 00 10 20 01 02 12 22 03)
- movdqa xmmB,xmmE
- punpckldq xmmA,xmmD ; xmmA=(00 10 20 01 11 21 02 12 02 12 22 03 13 23 04 14)
- punpckldq xmmE,xmmH ; xmmE=(22 03 13 23 04 14 24 05 24 05 15 25 06 16 26 07)
- punpckhdq xmmD,xmmB ; xmmD=(15 25 06 16 26 07 17 27 17 27 08 18 28 09 19 29)
-
- pshufd xmmH,xmmG,0x4E; xmmH=(0C 1C 2C 0D 0E 1E 2E 0F 08 18 28 09 0A 1A 2A 0B)
- movdqa xmmB,xmmF
- punpckldq xmmG,xmmC ; xmmG=(08 18 28 09 19 29 0A 1A 0A 1A 2A 0B 1B 2B 0C 1C)
- punpckldq xmmF,xmmH ; xmmF=(2A 0B 1B 2B 0C 1C 2C 0D 2C 0D 1D 2D 0E 1E 2E 0F)
- punpckhdq xmmC,xmmB ; xmmC=(1D 2D 0E 1E 2E 0F 1F 2F 1F 2F -- -- -- -- -- --)
-
- punpcklqdq xmmA,xmmE ; xmmA=(00 10 20 01 11 21 02 12 22 03 13 23 04 14 24 05)
- punpcklqdq xmmD,xmmG ; xmmD=(15 25 06 16 26 07 17 27 08 18 28 09 19 29 0A 1A)
- punpcklqdq xmmF,xmmC ; xmmF=(2A 0B 1B 2B 0C 1C 2C 0D 1D 2D 0E 1E 2E 0F 1F 2F)
-
- cmp rcx, byte SIZEOF_XMMWORD
- jb short .column_st32
-
- test rdi, SIZEOF_XMMWORD-1
- jnz short .out1
- ; --(aligned)-------------------
- movntdq XMMWORD [rdi+0*SIZEOF_XMMWORD], xmmA
- movntdq XMMWORD [rdi+1*SIZEOF_XMMWORD], xmmD
- movntdq XMMWORD [rdi+2*SIZEOF_XMMWORD], xmmF
- jmp short .out0
-.out1: ; --(unaligned)-----------------
- movdqu XMMWORD [rdi+0*SIZEOF_XMMWORD], xmmA
- movdqu XMMWORD [rdi+1*SIZEOF_XMMWORD], xmmD
- movdqu XMMWORD [rdi+2*SIZEOF_XMMWORD], xmmF
-.out0:
- add rdi, byte RGB_PIXELSIZE*SIZEOF_XMMWORD ; outptr
- sub rcx, byte SIZEOF_XMMWORD
- jz near .nextrow
-
- add rsi, byte SIZEOF_XMMWORD ; inptr0
- add rbx, byte SIZEOF_XMMWORD ; inptr1
- add rdx, byte SIZEOF_XMMWORD ; inptr2
- jmp near .columnloop
-
-.column_st32:
- lea rcx, [rcx+rcx*2] ; imul ecx, RGB_PIXELSIZE
- cmp rcx, byte 2*SIZEOF_XMMWORD
- jb short .column_st16
- movdqu XMMWORD [rdi+0*SIZEOF_XMMWORD], xmmA
- movdqu XMMWORD [rdi+1*SIZEOF_XMMWORD], xmmD
- add rdi, byte 2*SIZEOF_XMMWORD ; outptr
- movdqa xmmA,xmmF
- sub rcx, byte 2*SIZEOF_XMMWORD
- jmp short .column_st15
-.column_st16:
- cmp rcx, byte SIZEOF_XMMWORD
- jb short .column_st15
- movdqu XMMWORD [rdi+0*SIZEOF_XMMWORD], xmmA
- add rdi, byte SIZEOF_XMMWORD ; outptr
- movdqa xmmA,xmmD
- sub rcx, byte SIZEOF_XMMWORD
-.column_st15:
- ; Store the lower 8 bytes of xmmA to the output when it has enough
- ; space.
- cmp rcx, byte SIZEOF_MMWORD
- jb short .column_st7
- movq XMM_MMWORD [rdi], xmmA
- add rdi, byte SIZEOF_MMWORD
- sub rcx, byte SIZEOF_MMWORD
- psrldq xmmA, SIZEOF_MMWORD
-.column_st7:
- ; Store the lower 4 bytes of xmmA to the output when it has enough
- ; space.
- cmp rcx, byte SIZEOF_DWORD
- jb short .column_st3
- movd XMM_DWORD [rdi], xmmA
- add rdi, byte SIZEOF_DWORD
- sub rcx, byte SIZEOF_DWORD
- psrldq xmmA, SIZEOF_DWORD
-.column_st3:
- ; Store the lower 2 bytes of rax to the output when it has enough
- ; space.
- movd eax, xmmA
- cmp rcx, byte SIZEOF_WORD
- jb short .column_st1
- mov WORD [rdi], ax
- add rdi, byte SIZEOF_WORD
- sub rcx, byte SIZEOF_WORD
- shr rax, 16
-.column_st1:
- ; Store the lower 1 byte of rax to the output when it has enough
- ; space.
- test rcx, rcx
- jz short .nextrow
- mov BYTE [rdi], al
-
-%else ; RGB_PIXELSIZE == 4 ; -----------
-
-%ifdef RGBX_FILLER_0XFF
- pcmpeqb xmm6,xmm6 ; xmm6=XE=X(02468ACE********)
- pcmpeqb xmm7,xmm7 ; xmm7=XO=X(13579BDF********)
-%else
- pxor xmm6,xmm6 ; xmm6=XE=X(02468ACE********)
- pxor xmm7,xmm7 ; xmm7=XO=X(13579BDF********)
-%endif
- ; xmmA=(00 02 04 06 08 0A 0C 0E **), xmmB=(01 03 05 07 09 0B 0D 0F **)
- ; xmmC=(10 12 14 16 18 1A 1C 1E **), xmmD=(11 13 15 17 19 1B 1D 1F **)
- ; xmmE=(20 22 24 26 28 2A 2C 2E **), xmmF=(21 23 25 27 29 2B 2D 2F **)
- ; xmmG=(30 32 34 36 38 3A 3C 3E **), xmmH=(31 33 35 37 39 3B 3D 3F **)
-
- punpcklbw xmmA,xmmC ; xmmA=(00 10 02 12 04 14 06 16 08 18 0A 1A 0C 1C 0E 1E)
- punpcklbw xmmE,xmmG ; xmmE=(20 30 22 32 24 34 26 36 28 38 2A 3A 2C 3C 2E 3E)
- punpcklbw xmmB,xmmD ; xmmB=(01 11 03 13 05 15 07 17 09 19 0B 1B 0D 1D 0F 1F)
- punpcklbw xmmF,xmmH ; xmmF=(21 31 23 33 25 35 27 37 29 39 2B 3B 2D 3D 2F 3F)
-
- movdqa xmmC,xmmA
- punpcklwd xmmA,xmmE ; xmmA=(00 10 20 30 02 12 22 32 04 14 24 34 06 16 26 36)
- punpckhwd xmmC,xmmE ; xmmC=(08 18 28 38 0A 1A 2A 3A 0C 1C 2C 3C 0E 1E 2E 3E)
- movdqa xmmG,xmmB
- punpcklwd xmmB,xmmF ; xmmB=(01 11 21 31 03 13 23 33 05 15 25 35 07 17 27 37)
- punpckhwd xmmG,xmmF ; xmmG=(09 19 29 39 0B 1B 2B 3B 0D 1D 2D 3D 0F 1F 2F 3F)
-
- movdqa xmmD,xmmA
- punpckldq xmmA,xmmB ; xmmA=(00 10 20 30 01 11 21 31 02 12 22 32 03 13 23 33)
- punpckhdq xmmD,xmmB ; xmmD=(04 14 24 34 05 15 25 35 06 16 26 36 07 17 27 37)
- movdqa xmmH,xmmC
- punpckldq xmmC,xmmG ; xmmC=(08 18 28 38 09 19 29 39 0A 1A 2A 3A 0B 1B 2B 3B)
- punpckhdq xmmH,xmmG ; xmmH=(0C 1C 2C 3C 0D 1D 2D 3D 0E 1E 2E 3E 0F 1F 2F 3F)
-
- cmp rcx, byte SIZEOF_XMMWORD
- jb short .column_st32
-
- test rdi, SIZEOF_XMMWORD-1
- jnz short .out1
- ; --(aligned)-------------------
- movntdq XMMWORD [rdi+0*SIZEOF_XMMWORD], xmmA
- movntdq XMMWORD [rdi+1*SIZEOF_XMMWORD], xmmD
- movntdq XMMWORD [rdi+2*SIZEOF_XMMWORD], xmmC
- movntdq XMMWORD [rdi+3*SIZEOF_XMMWORD], xmmH
- jmp short .out0
-.out1: ; --(unaligned)-----------------
- movdqu XMMWORD [rdi+0*SIZEOF_XMMWORD], xmmA
- movdqu XMMWORD [rdi+1*SIZEOF_XMMWORD], xmmD
- movdqu XMMWORD [rdi+2*SIZEOF_XMMWORD], xmmC
- movdqu XMMWORD [rdi+3*SIZEOF_XMMWORD], xmmH
-.out0:
- add rdi, byte RGB_PIXELSIZE*SIZEOF_XMMWORD ; outptr
- sub rcx, byte SIZEOF_XMMWORD
- jz near .nextrow
-
- add rsi, byte SIZEOF_XMMWORD ; inptr0
- add rbx, byte SIZEOF_XMMWORD ; inptr1
- add rdx, byte SIZEOF_XMMWORD ; inptr2
- jmp near .columnloop
-
-.column_st32:
- cmp rcx, byte SIZEOF_XMMWORD/2
- jb short .column_st16
- movdqu XMMWORD [rdi+0*SIZEOF_XMMWORD], xmmA
- movdqu XMMWORD [rdi+1*SIZEOF_XMMWORD], xmmD
- add rdi, byte 2*SIZEOF_XMMWORD ; outptr
- movdqa xmmA,xmmC
- movdqa xmmD,xmmH
- sub rcx, byte SIZEOF_XMMWORD/2
-.column_st16:
- cmp rcx, byte SIZEOF_XMMWORD/4
- jb short .column_st15
- movdqu XMMWORD [rdi+0*SIZEOF_XMMWORD], xmmA
- add rdi, byte SIZEOF_XMMWORD ; outptr
- movdqa xmmA,xmmD
- sub rcx, byte SIZEOF_XMMWORD/4
-.column_st15:
- ; Store two pixels (8 bytes) of xmmA to the output when it has enough
- ; space.
- cmp rcx, byte SIZEOF_XMMWORD/8
- jb short .column_st7
- movq MMWORD [rdi], xmmA
- add rdi, byte SIZEOF_XMMWORD/8*4
- sub rcx, byte SIZEOF_XMMWORD/8
- psrldq xmmA, SIZEOF_XMMWORD/8*4
-.column_st7:
- ; Store one pixel (4 bytes) of xmmA to the output when it has enough
- ; space.
- test rcx, rcx
- jz short .nextrow
- movd XMM_DWORD [rdi], xmmA
-
-%endif ; RGB_PIXELSIZE ; ---------------
-
-.nextrow:
- pop rcx
- pop rsi
- pop rbx
- pop rdx
- pop rdi
- pop rax
-
- add rsi, byte SIZEOF_JSAMPROW
- add rbx, byte SIZEOF_JSAMPROW
- add rdx, byte SIZEOF_JSAMPROW
- add rdi, byte SIZEOF_JSAMPROW ; output_buf
- dec rax ; num_rows
- jg near .rowloop
-
- sfence ; flush the write buffer
-
-.return:
- pop rbx
- uncollect_args
- mov rsp,rbp ; rsp <- aligned rbp
- pop rsp ; rsp <- original rbp
- pop rbp
- ret
-
-; For some reason, the OS X linker does not honor the request to align the
-; segment unless we do this.
- align 16
diff --git a/simd/jdcolext-sse2.asm b/simd/jdcolext-sse2.asm
deleted file mode 100644
index 682aef3..0000000
--- a/simd/jdcolext-sse2.asm
+++ /dev/null
@@ -1,459 +0,0 @@
-;
-; jdcolext.asm - colorspace conversion (SSE2)
-;
-; Copyright 2009, 2012 Pierre Ossman <ossman@cendio.se> for Cendio AB
-; Copyright (C) 2012, D. R. Commander.
-;
-; Based on the x86 SIMD extension for IJG JPEG library
-; Copyright (C) 1999-2006, MIYASAKA Masaru.
-; For conditions of distribution and use, see copyright notice in jsimdext.inc
-;
-; This file should be assembled with NASM (Netwide Assembler),
-; can *not* be assembled with Microsoft's MASM or any compatible
-; assembler (including Borland's Turbo Assembler).
-; NASM is available from http://nasm.sourceforge.net/ or
-; http://sourceforge.net/project/showfiles.php?group_id=6208
-;
-; [TAB8]
-
-%include "jcolsamp.inc"
-
-; --------------------------------------------------------------------------
-;
-; Convert some rows of samples to the output colorspace.
-;
-; GLOBAL(void)
-; jsimd_ycc_rgb_convert_sse2 (JDIMENSION out_width,
-; JSAMPIMAGE input_buf, JDIMENSION input_row,
-; JSAMPARRAY output_buf, int num_rows)
-;
-
-%define out_width(b) (b)+8 ; JDIMENSION out_width
-%define input_buf(b) (b)+12 ; JSAMPIMAGE input_buf
-%define input_row(b) (b)+16 ; JDIMENSION input_row
-%define output_buf(b) (b)+20 ; JSAMPARRAY output_buf
-%define num_rows(b) (b)+24 ; int num_rows
-
-%define original_ebp ebp+0
-%define wk(i) ebp-(WK_NUM-(i))*SIZEOF_XMMWORD ; xmmword wk[WK_NUM]
-%define WK_NUM 2
-%define gotptr wk(0)-SIZEOF_POINTER ; void * gotptr
-
- align 16
- global EXTN(jsimd_ycc_rgb_convert_sse2)
-
-EXTN(jsimd_ycc_rgb_convert_sse2):
- push ebp
- mov eax,esp ; eax = original ebp
- sub esp, byte 4
- and esp, byte (-SIZEOF_XMMWORD) ; align to 128 bits
- mov [esp],eax
- mov ebp,esp ; ebp = aligned ebp
- lea esp, [wk(0)]
- pushpic eax ; make a room for GOT address
- push ebx
-; push ecx ; need not be preserved
-; push edx ; need not be preserved
- push esi
- push edi
-
- get_GOT ebx ; get GOT address
- movpic POINTER [gotptr], ebx ; save GOT address
-
- mov ecx, JDIMENSION [out_width(eax)] ; num_cols
- test ecx,ecx
- jz near .return
-
- push ecx
-
- mov edi, JSAMPIMAGE [input_buf(eax)]
- mov ecx, JDIMENSION [input_row(eax)]
- mov esi, JSAMPARRAY [edi+0*SIZEOF_JSAMPARRAY]
- mov ebx, JSAMPARRAY [edi+1*SIZEOF_JSAMPARRAY]
- mov edx, JSAMPARRAY [edi+2*SIZEOF_JSAMPARRAY]
- lea esi, [esi+ecx*SIZEOF_JSAMPROW]
- lea ebx, [ebx+ecx*SIZEOF_JSAMPROW]
- lea edx, [edx+ecx*SIZEOF_JSAMPROW]
-
- pop ecx
-
- mov edi, JSAMPARRAY [output_buf(eax)]
- mov eax, INT [num_rows(eax)]
- test eax,eax
- jle near .return
- alignx 16,7
-.rowloop:
- push eax
- push edi
- push edx
- push ebx
- push esi
- push ecx ; col
-
- mov esi, JSAMPROW [esi] ; inptr0
- mov ebx, JSAMPROW [ebx] ; inptr1
- mov edx, JSAMPROW [edx] ; inptr2
- mov edi, JSAMPROW [edi] ; outptr
- movpic eax, POINTER [gotptr] ; load GOT address (eax)
- alignx 16,7
-.columnloop:
-
- movdqa xmm5, XMMWORD [ebx] ; xmm5=Cb(0123456789ABCDEF)
- movdqa xmm1, XMMWORD [edx] ; xmm1=Cr(0123456789ABCDEF)
-
- pcmpeqw xmm4,xmm4
- pcmpeqw xmm7,xmm7
- psrlw xmm4,BYTE_BIT
- psllw xmm7,7 ; xmm7={0xFF80 0xFF80 0xFF80 0xFF80 ..}
- movdqa xmm0,xmm4 ; xmm0=xmm4={0xFF 0x00 0xFF 0x00 ..}
-
- pand xmm4,xmm5 ; xmm4=Cb(02468ACE)=CbE
- psrlw xmm5,BYTE_BIT ; xmm5=Cb(13579BDF)=CbO
- pand xmm0,xmm1 ; xmm0=Cr(02468ACE)=CrE
- psrlw xmm1,BYTE_BIT ; xmm1=Cr(13579BDF)=CrO
-
- paddw xmm4,xmm7
- paddw xmm5,xmm7
- paddw xmm0,xmm7
- paddw xmm1,xmm7
-
- ; (Original)
- ; R = Y + 1.40200 * Cr
- ; G = Y - 0.34414 * Cb - 0.71414 * Cr
- ; B = Y + 1.77200 * Cb
- ;
- ; (This implementation)
- ; R = Y + 0.40200 * Cr + Cr
- ; G = Y - 0.34414 * Cb + 0.28586 * Cr - Cr
- ; B = Y - 0.22800 * Cb + Cb + Cb
-
- movdqa xmm2,xmm4 ; xmm2=CbE
- movdqa xmm3,xmm5 ; xmm3=CbO
- paddw xmm4,xmm4 ; xmm4=2*CbE
- paddw xmm5,xmm5 ; xmm5=2*CbO
- movdqa xmm6,xmm0 ; xmm6=CrE
- movdqa xmm7,xmm1 ; xmm7=CrO
- paddw xmm0,xmm0 ; xmm0=2*CrE
- paddw xmm1,xmm1 ; xmm1=2*CrO
-
- pmulhw xmm4,[GOTOFF(eax,PW_MF0228)] ; xmm4=(2*CbE * -FIX(0.22800))
- pmulhw xmm5,[GOTOFF(eax,PW_MF0228)] ; xmm5=(2*CbO * -FIX(0.22800))
- pmulhw xmm0,[GOTOFF(eax,PW_F0402)] ; xmm0=(2*CrE * FIX(0.40200))
- pmulhw xmm1,[GOTOFF(eax,PW_F0402)] ; xmm1=(2*CrO * FIX(0.40200))
-
- paddw xmm4,[GOTOFF(eax,PW_ONE)]
- paddw xmm5,[GOTOFF(eax,PW_ONE)]
- psraw xmm4,1 ; xmm4=(CbE * -FIX(0.22800))
- psraw xmm5,1 ; xmm5=(CbO * -FIX(0.22800))
- paddw xmm0,[GOTOFF(eax,PW_ONE)]
- paddw xmm1,[GOTOFF(eax,PW_ONE)]
- psraw xmm0,1 ; xmm0=(CrE * FIX(0.40200))
- psraw xmm1,1 ; xmm1=(CrO * FIX(0.40200))
-
- paddw xmm4,xmm2
- paddw xmm5,xmm3
- paddw xmm4,xmm2 ; xmm4=(CbE * FIX(1.77200))=(B-Y)E
- paddw xmm5,xmm3 ; xmm5=(CbO * FIX(1.77200))=(B-Y)O
- paddw xmm0,xmm6 ; xmm0=(CrE * FIX(1.40200))=(R-Y)E
- paddw xmm1,xmm7 ; xmm1=(CrO * FIX(1.40200))=(R-Y)O
-
- movdqa XMMWORD [wk(0)], xmm4 ; wk(0)=(B-Y)E
- movdqa XMMWORD [wk(1)], xmm5 ; wk(1)=(B-Y)O
-
- movdqa xmm4,xmm2
- movdqa xmm5,xmm3
- punpcklwd xmm2,xmm6
- punpckhwd xmm4,xmm6
- pmaddwd xmm2,[GOTOFF(eax,PW_MF0344_F0285)]
- pmaddwd xmm4,[GOTOFF(eax,PW_MF0344_F0285)]
- punpcklwd xmm3,xmm7
- punpckhwd xmm5,xmm7
- pmaddwd xmm3,[GOTOFF(eax,PW_MF0344_F0285)]
- pmaddwd xmm5,[GOTOFF(eax,PW_MF0344_F0285)]
-
- paddd xmm2,[GOTOFF(eax,PD_ONEHALF)]
- paddd xmm4,[GOTOFF(eax,PD_ONEHALF)]
- psrad xmm2,SCALEBITS
- psrad xmm4,SCALEBITS
- paddd xmm3,[GOTOFF(eax,PD_ONEHALF)]
- paddd xmm5,[GOTOFF(eax,PD_ONEHALF)]
- psrad xmm3,SCALEBITS
- psrad xmm5,SCALEBITS
-
- packssdw xmm2,xmm4 ; xmm2=CbE*-FIX(0.344)+CrE*FIX(0.285)
- packssdw xmm3,xmm5 ; xmm3=CbO*-FIX(0.344)+CrO*FIX(0.285)
- psubw xmm2,xmm6 ; xmm2=CbE*-FIX(0.344)+CrE*-FIX(0.714)=(G-Y)E
- psubw xmm3,xmm7 ; xmm3=CbO*-FIX(0.344)+CrO*-FIX(0.714)=(G-Y)O
-
- movdqa xmm5, XMMWORD [esi] ; xmm5=Y(0123456789ABCDEF)
-
- pcmpeqw xmm4,xmm4
- psrlw xmm4,BYTE_BIT ; xmm4={0xFF 0x00 0xFF 0x00 ..}
- pand xmm4,xmm5 ; xmm4=Y(02468ACE)=YE
- psrlw xmm5,BYTE_BIT ; xmm5=Y(13579BDF)=YO
-
- paddw xmm0,xmm4 ; xmm0=((R-Y)E+YE)=RE=R(02468ACE)
- paddw xmm1,xmm5 ; xmm1=((R-Y)O+YO)=RO=R(13579BDF)
- packuswb xmm0,xmm0 ; xmm0=R(02468ACE********)
- packuswb xmm1,xmm1 ; xmm1=R(13579BDF********)
-
- paddw xmm2,xmm4 ; xmm2=((G-Y)E+YE)=GE=G(02468ACE)
- paddw xmm3,xmm5 ; xmm3=((G-Y)O+YO)=GO=G(13579BDF)
- packuswb xmm2,xmm2 ; xmm2=G(02468ACE********)
- packuswb xmm3,xmm3 ; xmm3=G(13579BDF********)
-
- paddw xmm4, XMMWORD [wk(0)] ; xmm4=(YE+(B-Y)E)=BE=B(02468ACE)
- paddw xmm5, XMMWORD [wk(1)] ; xmm5=(YO+(B-Y)O)=BO=B(13579BDF)
- packuswb xmm4,xmm4 ; xmm4=B(02468ACE********)
- packuswb xmm5,xmm5 ; xmm5=B(13579BDF********)
-
-%if RGB_PIXELSIZE == 3 ; ---------------
-
- ; xmmA=(00 02 04 06 08 0A 0C 0E **), xmmB=(01 03 05 07 09 0B 0D 0F **)
- ; xmmC=(10 12 14 16 18 1A 1C 1E **), xmmD=(11 13 15 17 19 1B 1D 1F **)
- ; xmmE=(20 22 24 26 28 2A 2C 2E **), xmmF=(21 23 25 27 29 2B 2D 2F **)
- ; xmmG=(** ** ** ** ** ** ** ** **), xmmH=(** ** ** ** ** ** ** ** **)
-
- punpcklbw xmmA,xmmC ; xmmA=(00 10 02 12 04 14 06 16 08 18 0A 1A 0C 1C 0E 1E)
- punpcklbw xmmE,xmmB ; xmmE=(20 01 22 03 24 05 26 07 28 09 2A 0B 2C 0D 2E 0F)
- punpcklbw xmmD,xmmF ; xmmD=(11 21 13 23 15 25 17 27 19 29 1B 2B 1D 2D 1F 2F)
-
- movdqa xmmG,xmmA
- movdqa xmmH,xmmA
- punpcklwd xmmA,xmmE ; xmmA=(00 10 20 01 02 12 22 03 04 14 24 05 06 16 26 07)
- punpckhwd xmmG,xmmE ; xmmG=(08 18 28 09 0A 1A 2A 0B 0C 1C 2C 0D 0E 1E 2E 0F)
-
- psrldq xmmH,2 ; xmmH=(02 12 04 14 06 16 08 18 0A 1A 0C 1C 0E 1E -- --)
- psrldq xmmE,2 ; xmmE=(22 03 24 05 26 07 28 09 2A 0B 2C 0D 2E 0F -- --)
-
- movdqa xmmC,xmmD
- movdqa xmmB,xmmD
- punpcklwd xmmD,xmmH ; xmmD=(11 21 02 12 13 23 04 14 15 25 06 16 17 27 08 18)
- punpckhwd xmmC,xmmH ; xmmC=(19 29 0A 1A 1B 2B 0C 1C 1D 2D 0E 1E 1F 2F -- --)
-
- psrldq xmmB,2 ; xmmB=(13 23 15 25 17 27 19 29 1B 2B 1D 2D 1F 2F -- --)
-
- movdqa xmmF,xmmE
- punpcklwd xmmE,xmmB ; xmmE=(22 03 13 23 24 05 15 25 26 07 17 27 28 09 19 29)
- punpckhwd xmmF,xmmB ; xmmF=(2A 0B 1B 2B 2C 0D 1D 2D 2E 0F 1F 2F -- -- -- --)
-
- pshufd xmmH,xmmA,0x4E; xmmH=(04 14 24 05 06 16 26 07 00 10 20 01 02 12 22 03)
- movdqa xmmB,xmmE
- punpckldq xmmA,xmmD ; xmmA=(00 10 20 01 11 21 02 12 02 12 22 03 13 23 04 14)
- punpckldq xmmE,xmmH ; xmmE=(22 03 13 23 04 14 24 05 24 05 15 25 06 16 26 07)
- punpckhdq xmmD,xmmB ; xmmD=(15 25 06 16 26 07 17 27 17 27 08 18 28 09 19 29)
-
- pshufd xmmH,xmmG,0x4E; xmmH=(0C 1C 2C 0D 0E 1E 2E 0F 08 18 28 09 0A 1A 2A 0B)
- movdqa xmmB,xmmF
- punpckldq xmmG,xmmC ; xmmG=(08 18 28 09 19 29 0A 1A 0A 1A 2A 0B 1B 2B 0C 1C)
- punpckldq xmmF,xmmH ; xmmF=(2A 0B 1B 2B 0C 1C 2C 0D 2C 0D 1D 2D 0E 1E 2E 0F)
- punpckhdq xmmC,xmmB ; xmmC=(1D 2D 0E 1E 2E 0F 1F 2F 1F 2F -- -- -- -- -- --)
-
- punpcklqdq xmmA,xmmE ; xmmA=(00 10 20 01 11 21 02 12 22 03 13 23 04 14 24 05)
- punpcklqdq xmmD,xmmG ; xmmD=(15 25 06 16 26 07 17 27 08 18 28 09 19 29 0A 1A)
- punpcklqdq xmmF,xmmC ; xmmF=(2A 0B 1B 2B 0C 1C 2C 0D 1D 2D 0E 1E 2E 0F 1F 2F)
-
- cmp ecx, byte SIZEOF_XMMWORD
- jb short .column_st32
-
- test edi, SIZEOF_XMMWORD-1
- jnz short .out1
- ; --(aligned)-------------------
- movntdq XMMWORD [edi+0*SIZEOF_XMMWORD], xmmA
- movntdq XMMWORD [edi+1*SIZEOF_XMMWORD], xmmD
- movntdq XMMWORD [edi+2*SIZEOF_XMMWORD], xmmF
- jmp short .out0
-.out1: ; --(unaligned)-----------------
- movdqu XMMWORD [edi+0*SIZEOF_XMMWORD], xmmA
- movdqu XMMWORD [edi+1*SIZEOF_XMMWORD], xmmD
- movdqu XMMWORD [edi+2*SIZEOF_XMMWORD], xmmF
-.out0:
- add edi, byte RGB_PIXELSIZE*SIZEOF_XMMWORD ; outptr
- sub ecx, byte SIZEOF_XMMWORD
- jz near .nextrow
-
- add esi, byte SIZEOF_XMMWORD ; inptr0
- add ebx, byte SIZEOF_XMMWORD ; inptr1
- add edx, byte SIZEOF_XMMWORD ; inptr2
- jmp near .columnloop
- alignx 16,7
-
-.column_st32:
- lea ecx, [ecx+ecx*2] ; imul ecx, RGB_PIXELSIZE
- cmp ecx, byte 2*SIZEOF_XMMWORD
- jb short .column_st16
- movdqu XMMWORD [edi+0*SIZEOF_XMMWORD], xmmA
- movdqu XMMWORD [edi+1*SIZEOF_XMMWORD], xmmD
- add edi, byte 2*SIZEOF_XMMWORD ; outptr
- movdqa xmmA,xmmF
- sub ecx, byte 2*SIZEOF_XMMWORD
- jmp short .column_st15
-.column_st16:
- cmp ecx, byte SIZEOF_XMMWORD
- jb short .column_st15
- movdqu XMMWORD [edi+0*SIZEOF_XMMWORD], xmmA
- add edi, byte SIZEOF_XMMWORD ; outptr
- movdqa xmmA,xmmD
- sub ecx, byte SIZEOF_XMMWORD
-.column_st15:
- ; Store the lower 8 bytes of xmmA to the output when it has enough
- ; space.
- cmp ecx, byte SIZEOF_MMWORD
- jb short .column_st7
- movq XMM_MMWORD [edi], xmmA
- add edi, byte SIZEOF_MMWORD
- sub ecx, byte SIZEOF_MMWORD
- psrldq xmmA, SIZEOF_MMWORD
-.column_st7:
- ; Store the lower 4 bytes of xmmA to the output when it has enough
- ; space.
- cmp ecx, byte SIZEOF_DWORD
- jb short .column_st3
- movd XMM_DWORD [edi], xmmA
- add edi, byte SIZEOF_DWORD
- sub ecx, byte SIZEOF_DWORD
- psrldq xmmA, SIZEOF_DWORD
-.column_st3:
- ; Store the lower 2 bytes of eax to the output when it has enough
- ; space.
- movd eax, xmmA
- cmp ecx, byte SIZEOF_WORD
- jb short .column_st1
- mov WORD [edi], ax
- add edi, byte SIZEOF_WORD
- sub ecx, byte SIZEOF_WORD
- shr eax, 16
-.column_st1:
- ; Store the lower 1 byte of eax to the output when it has enough
- ; space.
- test ecx, ecx
- jz short .nextrow
- mov BYTE [edi], al
-
-%else ; RGB_PIXELSIZE == 4 ; -----------
-
-%ifdef RGBX_FILLER_0XFF
- pcmpeqb xmm6,xmm6 ; xmm6=XE=X(02468ACE********)
- pcmpeqb xmm7,xmm7 ; xmm7=XO=X(13579BDF********)
-%else
- pxor xmm6,xmm6 ; xmm6=XE=X(02468ACE********)
- pxor xmm7,xmm7 ; xmm7=XO=X(13579BDF********)
-%endif
- ; xmmA=(00 02 04 06 08 0A 0C 0E **), xmmB=(01 03 05 07 09 0B 0D 0F **)
- ; xmmC=(10 12 14 16 18 1A 1C 1E **), xmmD=(11 13 15 17 19 1B 1D 1F **)
- ; xmmE=(20 22 24 26 28 2A 2C 2E **), xmmF=(21 23 25 27 29 2B 2D 2F **)
- ; xmmG=(30 32 34 36 38 3A 3C 3E **), xmmH=(31 33 35 37 39 3B 3D 3F **)
-
- punpcklbw xmmA,xmmC ; xmmA=(00 10 02 12 04 14 06 16 08 18 0A 1A 0C 1C 0E 1E)
- punpcklbw xmmE,xmmG ; xmmE=(20 30 22 32 24 34 26 36 28 38 2A 3A 2C 3C 2E 3E)
- punpcklbw xmmB,xmmD ; xmmB=(01 11 03 13 05 15 07 17 09 19 0B 1B 0D 1D 0F 1F)
- punpcklbw xmmF,xmmH ; xmmF=(21 31 23 33 25 35 27 37 29 39 2B 3B 2D 3D 2F 3F)
-
- movdqa xmmC,xmmA
- punpcklwd xmmA,xmmE ; xmmA=(00 10 20 30 02 12 22 32 04 14 24 34 06 16 26 36)
- punpckhwd xmmC,xmmE ; xmmC=(08 18 28 38 0A 1A 2A 3A 0C 1C 2C 3C 0E 1E 2E 3E)
- movdqa xmmG,xmmB
- punpcklwd xmmB,xmmF ; xmmB=(01 11 21 31 03 13 23 33 05 15 25 35 07 17 27 37)
- punpckhwd xmmG,xmmF ; xmmG=(09 19 29 39 0B 1B 2B 3B 0D 1D 2D 3D 0F 1F 2F 3F)
-
- movdqa xmmD,xmmA
- punpckldq xmmA,xmmB ; xmmA=(00 10 20 30 01 11 21 31 02 12 22 32 03 13 23 33)
- punpckhdq xmmD,xmmB ; xmmD=(04 14 24 34 05 15 25 35 06 16 26 36 07 17 27 37)
- movdqa xmmH,xmmC
- punpckldq xmmC,xmmG ; xmmC=(08 18 28 38 09 19 29 39 0A 1A 2A 3A 0B 1B 2B 3B)
- punpckhdq xmmH,xmmG ; xmmH=(0C 1C 2C 3C 0D 1D 2D 3D 0E 1E 2E 3E 0F 1F 2F 3F)
-
- cmp ecx, byte SIZEOF_XMMWORD
- jb short .column_st32
-
- test edi, SIZEOF_XMMWORD-1
- jnz short .out1
- ; --(aligned)-------------------
- movntdq XMMWORD [edi+0*SIZEOF_XMMWORD], xmmA
- movntdq XMMWORD [edi+1*SIZEOF_XMMWORD], xmmD
- movntdq XMMWORD [edi+2*SIZEOF_XMMWORD], xmmC
- movntdq XMMWORD [edi+3*SIZEOF_XMMWORD], xmmH
- jmp short .out0
-.out1: ; --(unaligned)-----------------
- movdqu XMMWORD [edi+0*SIZEOF_XMMWORD], xmmA
- movdqu XMMWORD [edi+1*SIZEOF_XMMWORD], xmmD
- movdqu XMMWORD [edi+2*SIZEOF_XMMWORD], xmmC
- movdqu XMMWORD [edi+3*SIZEOF_XMMWORD], xmmH
-.out0:
- add edi, byte RGB_PIXELSIZE*SIZEOF_XMMWORD ; outptr
- sub ecx, byte SIZEOF_XMMWORD
- jz near .nextrow
-
- add esi, byte SIZEOF_XMMWORD ; inptr0
- add ebx, byte SIZEOF_XMMWORD ; inptr1
- add edx, byte SIZEOF_XMMWORD ; inptr2
- jmp near .columnloop
- alignx 16,7
-
-.column_st32:
- cmp ecx, byte SIZEOF_XMMWORD/2
- jb short .column_st16
- movdqu XMMWORD [edi+0*SIZEOF_XMMWORD], xmmA
- movdqu XMMWORD [edi+1*SIZEOF_XMMWORD], xmmD
- add edi, byte 2*SIZEOF_XMMWORD ; outptr
- movdqa xmmA,xmmC
- movdqa xmmD,xmmH
- sub ecx, byte SIZEOF_XMMWORD/2
-.column_st16:
- cmp ecx, byte SIZEOF_XMMWORD/4
- jb short .column_st15
- movdqu XMMWORD [edi+0*SIZEOF_XMMWORD], xmmA
- add edi, byte SIZEOF_XMMWORD ; outptr
- movdqa xmmA,xmmD
- sub ecx, byte SIZEOF_XMMWORD/4
-.column_st15:
- ; Store two pixels (8 bytes) of xmmA to the output when it has enough
- ; space.
- cmp ecx, byte SIZEOF_XMMWORD/8
- jb short .column_st7
- movq XMM_MMWORD [edi], xmmA
- add edi, byte SIZEOF_XMMWORD/8*4
- sub ecx, byte SIZEOF_XMMWORD/8
- psrldq xmmA, SIZEOF_XMMWORD/8*4
-.column_st7:
- ; Store one pixel (4 bytes) of xmmA to the output when it has enough
- ; space.
- test ecx, ecx
- jz short .nextrow
- movd XMM_DWORD [edi], xmmA
-
-%endif ; RGB_PIXELSIZE ; ---------------
-
- alignx 16,7
-
-.nextrow:
- pop ecx
- pop esi
- pop ebx
- pop edx
- pop edi
- pop eax
-
- add esi, byte SIZEOF_JSAMPROW
- add ebx, byte SIZEOF_JSAMPROW
- add edx, byte SIZEOF_JSAMPROW
- add edi, byte SIZEOF_JSAMPROW ; output_buf
- dec eax ; num_rows
- jg near .rowloop
-
- sfence ; flush the write buffer
-
-.return:
- pop edi
- pop esi
-; pop edx ; need not be preserved
-; pop ecx ; need not be preserved
- pop ebx
- mov esp,ebp ; esp <- aligned ebp
- pop esp ; esp <- original ebp
- pop ebp
- ret
-
-; For some reason, the OS X linker does not honor the request to align the
-; segment unless we do this.
- align 16
diff --git a/simd/jdcolor-sse2-64.asm b/simd/jdcolor-sse2-64.asm
deleted file mode 100644
index d2bf210..0000000
--- a/simd/jdcolor-sse2-64.asm
+++ /dev/null
@@ -1,119 +0,0 @@
-;
-; jdcolor.asm - colorspace conversion (64-bit SSE2)
-;
-; Copyright 2009 Pierre Ossman <ossman@cendio.se> for Cendio AB
-; Copyright (C) 2009, D. R. Commander.
-;
-; Based on the x86 SIMD extension for IJG JPEG library
-; Copyright (C) 1999-2006, MIYASAKA Masaru.
-; For conditions of distribution and use, see copyright notice in jsimdext.inc
-;
-; This file should be assembled with NASM (Netwide Assembler),
-; can *not* be assembled with Microsoft's MASM or any compatible
-; assembler (including Borland's Turbo Assembler).
-; NASM is available from http://nasm.sourceforge.net/ or
-; http://sourceforge.net/project/showfiles.php?group_id=6208
-;
-; [TAB8]
-
-%include "jsimdext.inc"
-
-; --------------------------------------------------------------------------
-
-%define SCALEBITS 16
-
-F_0_344 equ 22554 ; FIX(0.34414)
-F_0_714 equ 46802 ; FIX(0.71414)
-F_1_402 equ 91881 ; FIX(1.40200)
-F_1_772 equ 116130 ; FIX(1.77200)
-F_0_402 equ (F_1_402 - 65536) ; FIX(1.40200) - FIX(1)
-F_0_285 equ ( 65536 - F_0_714) ; FIX(1) - FIX(0.71414)
-F_0_228 equ (131072 - F_1_772) ; FIX(2) - FIX(1.77200)
-
-; --------------------------------------------------------------------------
- SECTION SEG_CONST
-
- alignz 16
- global EXTN(jconst_ycc_rgb_convert_sse2)
-
-EXTN(jconst_ycc_rgb_convert_sse2):
-
-PW_F0402 times 8 dw F_0_402
-PW_MF0228 times 8 dw -F_0_228
-PW_MF0344_F0285 times 4 dw -F_0_344, F_0_285
-PW_ONE times 8 dw 1
-PD_ONEHALF times 4 dd 1 << (SCALEBITS-1)
-
- alignz 16
-
-; --------------------------------------------------------------------------
- SECTION SEG_TEXT
- BITS 64
-
-%include "jdcolext-sse2-64.asm"
-
-%undef RGB_RED
-%undef RGB_GREEN
-%undef RGB_BLUE
-%undef RGB_PIXELSIZE
-%define RGB_RED EXT_RGB_RED
-%define RGB_GREEN EXT_RGB_GREEN
-%define RGB_BLUE EXT_RGB_BLUE
-%define RGB_PIXELSIZE EXT_RGB_PIXELSIZE
-%define jsimd_ycc_rgb_convert_sse2 jsimd_ycc_extrgb_convert_sse2
-%include "jdcolext-sse2-64.asm"
-
-%undef RGB_RED
-%undef RGB_GREEN
-%undef RGB_BLUE
-%undef RGB_PIXELSIZE
-%define RGB_RED EXT_RGBX_RED
-%define RGB_GREEN EXT_RGBX_GREEN
-%define RGB_BLUE EXT_RGBX_BLUE
-%define RGB_PIXELSIZE EXT_RGBX_PIXELSIZE
-%define jsimd_ycc_rgb_convert_sse2 jsimd_ycc_extrgbx_convert_sse2
-%include "jdcolext-sse2-64.asm"
-
-%undef RGB_RED
-%undef RGB_GREEN
-%undef RGB_BLUE
-%undef RGB_PIXELSIZE
-%define RGB_RED EXT_BGR_RED
-%define RGB_GREEN EXT_BGR_GREEN
-%define RGB_BLUE EXT_BGR_BLUE
-%define RGB_PIXELSIZE EXT_BGR_PIXELSIZE
-%define jsimd_ycc_rgb_convert_sse2 jsimd_ycc_extbgr_convert_sse2
-%include "jdcolext-sse2-64.asm"
-
-%undef RGB_RED
-%undef RGB_GREEN
-%undef RGB_BLUE
-%undef RGB_PIXELSIZE
-%define RGB_RED EXT_BGRX_RED
-%define RGB_GREEN EXT_BGRX_GREEN
-%define RGB_BLUE EXT_BGRX_BLUE
-%define RGB_PIXELSIZE EXT_BGRX_PIXELSIZE
-%define jsimd_ycc_rgb_convert_sse2 jsimd_ycc_extbgrx_convert_sse2
-%include "jdcolext-sse2-64.asm"
-
-%undef RGB_RED
-%undef RGB_GREEN
-%undef RGB_BLUE
-%undef RGB_PIXELSIZE
-%define RGB_RED EXT_XBGR_RED
-%define RGB_GREEN EXT_XBGR_GREEN
-%define RGB_BLUE EXT_XBGR_BLUE
-%define RGB_PIXELSIZE EXT_XBGR_PIXELSIZE
-%define jsimd_ycc_rgb_convert_sse2 jsimd_ycc_extxbgr_convert_sse2
-%include "jdcolext-sse2-64.asm"
-
-%undef RGB_RED
-%undef RGB_GREEN
-%undef RGB_BLUE
-%undef RGB_PIXELSIZE
-%define RGB_RED EXT_XRGB_RED
-%define RGB_GREEN EXT_XRGB_GREEN
-%define RGB_BLUE EXT_XRGB_BLUE
-%define RGB_PIXELSIZE EXT_XRGB_PIXELSIZE
-%define jsimd_ycc_rgb_convert_sse2 jsimd_ycc_extxrgb_convert_sse2
-%include "jdcolext-sse2-64.asm"
diff --git a/simd/jdmerge-sse2-64.asm b/simd/jdmerge-sse2-64.asm
deleted file mode 100644
index 244bd40..0000000
--- a/simd/jdmerge-sse2-64.asm
+++ /dev/null
@@ -1,125 +0,0 @@
-;
-; jdmerge.asm - merged upsampling/color conversion (64-bit SSE2)
-;
-; Copyright 2009 Pierre Ossman <ossman@cendio.se> for Cendio AB
-; Copyright (C) 2009, D. R. Commander.
-;
-; Based on the x86 SIMD extension for IJG JPEG library
-; Copyright (C) 1999-2006, MIYASAKA Masaru.
-; For conditions of distribution and use, see copyright notice in jsimdext.inc
-;
-; This file should be assembled with NASM (Netwide Assembler),
-; can *not* be assembled with Microsoft's MASM or any compatible
-; assembler (including Borland's Turbo Assembler).
-; NASM is available from http://nasm.sourceforge.net/ or
-; http://sourceforge.net/project/showfiles.php?group_id=6208
-;
-; [TAB8]
-
-%include "jsimdext.inc"
-
-; --------------------------------------------------------------------------
-
-%define SCALEBITS 16
-
-F_0_344 equ 22554 ; FIX(0.34414)
-F_0_714 equ 46802 ; FIX(0.71414)
-F_1_402 equ 91881 ; FIX(1.40200)
-F_1_772 equ 116130 ; FIX(1.77200)
-F_0_402 equ (F_1_402 - 65536) ; FIX(1.40200) - FIX(1)
-F_0_285 equ ( 65536 - F_0_714) ; FIX(1) - FIX(0.71414)
-F_0_228 equ (131072 - F_1_772) ; FIX(2) - FIX(1.77200)
-
-; --------------------------------------------------------------------------
- SECTION SEG_CONST
-
- alignz 16
- global EXTN(jconst_merged_upsample_sse2)
-
-EXTN(jconst_merged_upsample_sse2):
-
-PW_F0402 times 8 dw F_0_402
-PW_MF0228 times 8 dw -F_0_228
-PW_MF0344_F0285 times 4 dw -F_0_344, F_0_285
-PW_ONE times 8 dw 1
-PD_ONEHALF times 4 dd 1 << (SCALEBITS-1)
-
- alignz 16
-
-; --------------------------------------------------------------------------
- SECTION SEG_TEXT
- BITS 64
-
-%include "jdmrgext-sse2-64.asm"
-
-%undef RGB_RED
-%undef RGB_GREEN
-%undef RGB_BLUE
-%undef RGB_PIXELSIZE
-%define RGB_RED EXT_RGB_RED
-%define RGB_GREEN EXT_RGB_GREEN
-%define RGB_BLUE EXT_RGB_BLUE
-%define RGB_PIXELSIZE EXT_RGB_PIXELSIZE
-%define jsimd_h2v1_merged_upsample_sse2 jsimd_h2v1_extrgb_merged_upsample_sse2
-%define jsimd_h2v2_merged_upsample_sse2 jsimd_h2v2_extrgb_merged_upsample_sse2
-%include "jdmrgext-sse2-64.asm"
-
-%undef RGB_RED
-%undef RGB_GREEN
-%undef RGB_BLUE
-%undef RGB_PIXELSIZE
-%define RGB_RED EXT_RGBX_RED
-%define RGB_GREEN EXT_RGBX_GREEN
-%define RGB_BLUE EXT_RGBX_BLUE
-%define RGB_PIXELSIZE EXT_RGBX_PIXELSIZE
-%define jsimd_h2v1_merged_upsample_sse2 jsimd_h2v1_extrgbx_merged_upsample_sse2
-%define jsimd_h2v2_merged_upsample_sse2 jsimd_h2v2_extrgbx_merged_upsample_sse2
-%include "jdmrgext-sse2-64.asm"
-
-%undef RGB_RED
-%undef RGB_GREEN
-%undef RGB_BLUE
-%undef RGB_PIXELSIZE
-%define RGB_RED EXT_BGR_RED
-%define RGB_GREEN EXT_BGR_GREEN
-%define RGB_BLUE EXT_BGR_BLUE
-%define RGB_PIXELSIZE EXT_BGR_PIXELSIZE
-%define jsimd_h2v1_merged_upsample_sse2 jsimd_h2v1_extbgr_merged_upsample_sse2
-%define jsimd_h2v2_merged_upsample_sse2 jsimd_h2v2_extbgr_merged_upsample_sse2
-%include "jdmrgext-sse2-64.asm"
-
-%undef RGB_RED
-%undef RGB_GREEN
-%undef RGB_BLUE
-%undef RGB_PIXELSIZE
-%define RGB_RED EXT_BGRX_RED
-%define RGB_GREEN EXT_BGRX_GREEN
-%define RGB_BLUE EXT_BGRX_BLUE
-%define RGB_PIXELSIZE EXT_BGRX_PIXELSIZE
-%define jsimd_h2v1_merged_upsample_sse2 jsimd_h2v1_extbgrx_merged_upsample_sse2
-%define jsimd_h2v2_merged_upsample_sse2 jsimd_h2v2_extbgrx_merged_upsample_sse2
-%include "jdmrgext-sse2-64.asm"
-
-%undef RGB_RED
-%undef RGB_GREEN
-%undef RGB_BLUE
-%undef RGB_PIXELSIZE
-%define RGB_RED EXT_XBGR_RED
-%define RGB_GREEN EXT_XBGR_GREEN
-%define RGB_BLUE EXT_XBGR_BLUE
-%define RGB_PIXELSIZE EXT_XBGR_PIXELSIZE
-%define jsimd_h2v1_merged_upsample_sse2 jsimd_h2v1_extxbgr_merged_upsample_sse2
-%define jsimd_h2v2_merged_upsample_sse2 jsimd_h2v2_extxbgr_merged_upsample_sse2
-%include "jdmrgext-sse2-64.asm"
-
-%undef RGB_RED
-%undef RGB_GREEN
-%undef RGB_BLUE
-%undef RGB_PIXELSIZE
-%define RGB_RED EXT_XRGB_RED
-%define RGB_GREEN EXT_XRGB_GREEN
-%define RGB_BLUE EXT_XRGB_BLUE
-%define RGB_PIXELSIZE EXT_XRGB_PIXELSIZE
-%define jsimd_h2v1_merged_upsample_sse2 jsimd_h2v1_extxrgb_merged_upsample_sse2
-%define jsimd_h2v2_merged_upsample_sse2 jsimd_h2v2_extxrgb_merged_upsample_sse2
-%include "jdmrgext-sse2-64.asm"
diff --git a/simd/jdmrgext-sse2-64.asm b/simd/jdmrgext-sse2-64.asm
deleted file mode 100644
index ad74c5f..0000000
--- a/simd/jdmrgext-sse2-64.asm
+++ /dev/null
@@ -1,537 +0,0 @@
-;
-; jdmrgext.asm - merged upsampling/color conversion (64-bit SSE2)
-;
-; Copyright 2009, 2012 Pierre Ossman <ossman@cendio.se> for Cendio AB
-; Copyright (C) 2009, 2012, D. R. Commander.
-;
-; Based on the x86 SIMD extension for IJG JPEG library
-; Copyright (C) 1999-2006, MIYASAKA Masaru.
-; For conditions of distribution and use, see copyright notice in jsimdext.inc
-;
-; This file should be assembled with NASM (Netwide Assembler),
-; can *not* be assembled with Microsoft's MASM or any compatible
-; assembler (including Borland's Turbo Assembler).
-; NASM is available from http://nasm.sourceforge.net/ or
-; http://sourceforge.net/project/showfiles.php?group_id=6208
-;
-; [TAB8]
-
-%include "jcolsamp.inc"
-
-; --------------------------------------------------------------------------
-;
-; Upsample and color convert for the case of 2:1 horizontal and 1:1 vertical.
-;
-; GLOBAL(void)
-; jsimd_h2v1_merged_upsample_sse2 (JDIMENSION output_width,
-; JSAMPIMAGE input_buf,
-; JDIMENSION in_row_group_ctr,
-; JSAMPARRAY output_buf);
-;
-
-; r10 = JDIMENSION output_width
-; r11 = JSAMPIMAGE input_buf
-; r12 = JDIMENSION in_row_group_ctr
-; r13 = JSAMPARRAY output_buf
-
-%define wk(i) rbp-(WK_NUM-(i))*SIZEOF_XMMWORD ; xmmword wk[WK_NUM]
-%define WK_NUM 3
-
- align 16
- global EXTN(jsimd_h2v1_merged_upsample_sse2)
-
-EXTN(jsimd_h2v1_merged_upsample_sse2):
- push rbp
- mov rax,rsp ; rax = original rbp
- sub rsp, byte 4
- and rsp, byte (-SIZEOF_XMMWORD) ; align to 128 bits
- mov [rsp],rax
- mov rbp,rsp ; rbp = aligned rbp
- lea rsp, [wk(0)]
- collect_args
- push rbx
-
- mov ecx, r10d ; col
- test rcx,rcx
- jz near .return
-
- push rcx
-
- mov rdi, r11
- mov ecx, r12d
- mov rsi, JSAMPARRAY [rdi+0*SIZEOF_JSAMPARRAY]
- mov rbx, JSAMPARRAY [rdi+1*SIZEOF_JSAMPARRAY]
- mov rdx, JSAMPARRAY [rdi+2*SIZEOF_JSAMPARRAY]
- mov rdi, r13
- mov rsi, JSAMPROW [rsi+rcx*SIZEOF_JSAMPROW] ; inptr0
- mov rbx, JSAMPROW [rbx+rcx*SIZEOF_JSAMPROW] ; inptr1
- mov rdx, JSAMPROW [rdx+rcx*SIZEOF_JSAMPROW] ; inptr2
- mov rdi, JSAMPROW [rdi] ; outptr
-
- pop rcx ; col
-
-.columnloop:
-
- movdqa xmm6, XMMWORD [rbx] ; xmm6=Cb(0123456789ABCDEF)
- movdqa xmm7, XMMWORD [rdx] ; xmm7=Cr(0123456789ABCDEF)
-
- pxor xmm1,xmm1 ; xmm1=(all 0's)
- pcmpeqw xmm3,xmm3
- psllw xmm3,7 ; xmm3={0xFF80 0xFF80 0xFF80 0xFF80 ..}
-
- movdqa xmm4,xmm6
- punpckhbw xmm6,xmm1 ; xmm6=Cb(89ABCDEF)=CbH
- punpcklbw xmm4,xmm1 ; xmm4=Cb(01234567)=CbL
- movdqa xmm0,xmm7
- punpckhbw xmm7,xmm1 ; xmm7=Cr(89ABCDEF)=CrH
- punpcklbw xmm0,xmm1 ; xmm0=Cr(01234567)=CrL
-
- paddw xmm6,xmm3
- paddw xmm4,xmm3
- paddw xmm7,xmm3
- paddw xmm0,xmm3
-
- ; (Original)
- ; R = Y + 1.40200 * Cr
- ; G = Y - 0.34414 * Cb - 0.71414 * Cr
- ; B = Y + 1.77200 * Cb
- ;
- ; (This implementation)
- ; R = Y + 0.40200 * Cr + Cr
- ; G = Y - 0.34414 * Cb + 0.28586 * Cr - Cr
- ; B = Y - 0.22800 * Cb + Cb + Cb
-
- movdqa xmm5,xmm6 ; xmm5=CbH
- movdqa xmm2,xmm4 ; xmm2=CbL
- paddw xmm6,xmm6 ; xmm6=2*CbH
- paddw xmm4,xmm4 ; xmm4=2*CbL
- movdqa xmm1,xmm7 ; xmm1=CrH
- movdqa xmm3,xmm0 ; xmm3=CrL
- paddw xmm7,xmm7 ; xmm7=2*CrH
- paddw xmm0,xmm0 ; xmm0=2*CrL
-
- pmulhw xmm6,[rel PW_MF0228] ; xmm6=(2*CbH * -FIX(0.22800))
- pmulhw xmm4,[rel PW_MF0228] ; xmm4=(2*CbL * -FIX(0.22800))
- pmulhw xmm7,[rel PW_F0402] ; xmm7=(2*CrH * FIX(0.40200))
- pmulhw xmm0,[rel PW_F0402] ; xmm0=(2*CrL * FIX(0.40200))
-
- paddw xmm6,[rel PW_ONE]
- paddw xmm4,[rel PW_ONE]
- psraw xmm6,1 ; xmm6=(CbH * -FIX(0.22800))
- psraw xmm4,1 ; xmm4=(CbL * -FIX(0.22800))
- paddw xmm7,[rel PW_ONE]
- paddw xmm0,[rel PW_ONE]
- psraw xmm7,1 ; xmm7=(CrH * FIX(0.40200))
- psraw xmm0,1 ; xmm0=(CrL * FIX(0.40200))
-
- paddw xmm6,xmm5
- paddw xmm4,xmm2
- paddw xmm6,xmm5 ; xmm6=(CbH * FIX(1.77200))=(B-Y)H
- paddw xmm4,xmm2 ; xmm4=(CbL * FIX(1.77200))=(B-Y)L
- paddw xmm7,xmm1 ; xmm7=(CrH * FIX(1.40200))=(R-Y)H
- paddw xmm0,xmm3 ; xmm0=(CrL * FIX(1.40200))=(R-Y)L
-
- movdqa XMMWORD [wk(0)], xmm6 ; wk(0)=(B-Y)H
- movdqa XMMWORD [wk(1)], xmm7 ; wk(1)=(R-Y)H
-
- movdqa xmm6,xmm5
- movdqa xmm7,xmm2
- punpcklwd xmm5,xmm1
- punpckhwd xmm6,xmm1
- pmaddwd xmm5,[rel PW_MF0344_F0285]
- pmaddwd xmm6,[rel PW_MF0344_F0285]
- punpcklwd xmm2,xmm3
- punpckhwd xmm7,xmm3
- pmaddwd xmm2,[rel PW_MF0344_F0285]
- pmaddwd xmm7,[rel PW_MF0344_F0285]
-
- paddd xmm5,[rel PD_ONEHALF]
- paddd xmm6,[rel PD_ONEHALF]
- psrad xmm5,SCALEBITS
- psrad xmm6,SCALEBITS
- paddd xmm2,[rel PD_ONEHALF]
- paddd xmm7,[rel PD_ONEHALF]
- psrad xmm2,SCALEBITS
- psrad xmm7,SCALEBITS
-
- packssdw xmm5,xmm6 ; xmm5=CbH*-FIX(0.344)+CrH*FIX(0.285)
- packssdw xmm2,xmm7 ; xmm2=CbL*-FIX(0.344)+CrL*FIX(0.285)
- psubw xmm5,xmm1 ; xmm5=CbH*-FIX(0.344)+CrH*-FIX(0.714)=(G-Y)H
- psubw xmm2,xmm3 ; xmm2=CbL*-FIX(0.344)+CrL*-FIX(0.714)=(G-Y)L
-
- movdqa XMMWORD [wk(2)], xmm5 ; wk(2)=(G-Y)H
-
- mov al,2 ; Yctr
- jmp short .Yloop_1st
-
-.Yloop_2nd:
- movdqa xmm0, XMMWORD [wk(1)] ; xmm0=(R-Y)H
- movdqa xmm2, XMMWORD [wk(2)] ; xmm2=(G-Y)H
- movdqa xmm4, XMMWORD [wk(0)] ; xmm4=(B-Y)H
-
-.Yloop_1st:
- movdqa xmm7, XMMWORD [rsi] ; xmm7=Y(0123456789ABCDEF)
-
- pcmpeqw xmm6,xmm6
- psrlw xmm6,BYTE_BIT ; xmm6={0xFF 0x00 0xFF 0x00 ..}
- pand xmm6,xmm7 ; xmm6=Y(02468ACE)=YE
- psrlw xmm7,BYTE_BIT ; xmm7=Y(13579BDF)=YO
-
- movdqa xmm1,xmm0 ; xmm1=xmm0=(R-Y)(L/H)
- movdqa xmm3,xmm2 ; xmm3=xmm2=(G-Y)(L/H)
- movdqa xmm5,xmm4 ; xmm5=xmm4=(B-Y)(L/H)
-
- paddw xmm0,xmm6 ; xmm0=((R-Y)+YE)=RE=R(02468ACE)
- paddw xmm1,xmm7 ; xmm1=((R-Y)+YO)=RO=R(13579BDF)
- packuswb xmm0,xmm0 ; xmm0=R(02468ACE********)
- packuswb xmm1,xmm1 ; xmm1=R(13579BDF********)
-
- paddw xmm2,xmm6 ; xmm2=((G-Y)+YE)=GE=G(02468ACE)
- paddw xmm3,xmm7 ; xmm3=((G-Y)+YO)=GO=G(13579BDF)
- packuswb xmm2,xmm2 ; xmm2=G(02468ACE********)
- packuswb xmm3,xmm3 ; xmm3=G(13579BDF********)
-
- paddw xmm4,xmm6 ; xmm4=((B-Y)+YE)=BE=B(02468ACE)
- paddw xmm5,xmm7 ; xmm5=((B-Y)+YO)=BO=B(13579BDF)
- packuswb xmm4,xmm4 ; xmm4=B(02468ACE********)
- packuswb xmm5,xmm5 ; xmm5=B(13579BDF********)
-
-%if RGB_PIXELSIZE == 3 ; ---------------
-
- ; xmmA=(00 02 04 06 08 0A 0C 0E **), xmmB=(01 03 05 07 09 0B 0D 0F **)
- ; xmmC=(10 12 14 16 18 1A 1C 1E **), xmmD=(11 13 15 17 19 1B 1D 1F **)
- ; xmmE=(20 22 24 26 28 2A 2C 2E **), xmmF=(21 23 25 27 29 2B 2D 2F **)
- ; xmmG=(** ** ** ** ** ** ** ** **), xmmH=(** ** ** ** ** ** ** ** **)
-
- punpcklbw xmmA,xmmC ; xmmA=(00 10 02 12 04 14 06 16 08 18 0A 1A 0C 1C 0E 1E)
- punpcklbw xmmE,xmmB ; xmmE=(20 01 22 03 24 05 26 07 28 09 2A 0B 2C 0D 2E 0F)
- punpcklbw xmmD,xmmF ; xmmD=(11 21 13 23 15 25 17 27 19 29 1B 2B 1D 2D 1F 2F)
-
- movdqa xmmG,xmmA
- movdqa xmmH,xmmA
- punpcklwd xmmA,xmmE ; xmmA=(00 10 20 01 02 12 22 03 04 14 24 05 06 16 26 07)
- punpckhwd xmmG,xmmE ; xmmG=(08 18 28 09 0A 1A 2A 0B 0C 1C 2C 0D 0E 1E 2E 0F)
-
- psrldq xmmH,2 ; xmmH=(02 12 04 14 06 16 08 18 0A 1A 0C 1C 0E 1E -- --)
- psrldq xmmE,2 ; xmmE=(22 03 24 05 26 07 28 09 2A 0B 2C 0D 2E 0F -- --)
-
- movdqa xmmC,xmmD
- movdqa xmmB,xmmD
- punpcklwd xmmD,xmmH ; xmmD=(11 21 02 12 13 23 04 14 15 25 06 16 17 27 08 18)
- punpckhwd xmmC,xmmH ; xmmC=(19 29 0A 1A 1B 2B 0C 1C 1D 2D 0E 1E 1F 2F -- --)
-
- psrldq xmmB,2 ; xmmB=(13 23 15 25 17 27 19 29 1B 2B 1D 2D 1F 2F -- --)
-
- movdqa xmmF,xmmE
- punpcklwd xmmE,xmmB ; xmmE=(22 03 13 23 24 05 15 25 26 07 17 27 28 09 19 29)
- punpckhwd xmmF,xmmB ; xmmF=(2A 0B 1B 2B 2C 0D 1D 2D 2E 0F 1F 2F -- -- -- --)
-
- pshufd xmmH,xmmA,0x4E; xmmH=(04 14 24 05 06 16 26 07 00 10 20 01 02 12 22 03)
- movdqa xmmB,xmmE
- punpckldq xmmA,xmmD ; xmmA=(00 10 20 01 11 21 02 12 02 12 22 03 13 23 04 14)
- punpckldq xmmE,xmmH ; xmmE=(22 03 13 23 04 14 24 05 24 05 15 25 06 16 26 07)
- punpckhdq xmmD,xmmB ; xmmD=(15 25 06 16 26 07 17 27 17 27 08 18 28 09 19 29)
-
- pshufd xmmH,xmmG,0x4E; xmmH=(0C 1C 2C 0D 0E 1E 2E 0F 08 18 28 09 0A 1A 2A 0B)
- movdqa xmmB,xmmF
- punpckldq xmmG,xmmC ; xmmG=(08 18 28 09 19 29 0A 1A 0A 1A 2A 0B 1B 2B 0C 1C)
- punpckldq xmmF,xmmH ; xmmF=(2A 0B 1B 2B 0C 1C 2C 0D 2C 0D 1D 2D 0E 1E 2E 0F)
- punpckhdq xmmC,xmmB ; xmmC=(1D 2D 0E 1E 2E 0F 1F 2F 1F 2F -- -- -- -- -- --)
-
- punpcklqdq xmmA,xmmE ; xmmA=(00 10 20 01 11 21 02 12 22 03 13 23 04 14 24 05)
- punpcklqdq xmmD,xmmG ; xmmD=(15 25 06 16 26 07 17 27 08 18 28 09 19 29 0A 1A)
- punpcklqdq xmmF,xmmC ; xmmF=(2A 0B 1B 2B 0C 1C 2C 0D 1D 2D 0E 1E 2E 0F 1F 2F)
-
- cmp rcx, byte SIZEOF_XMMWORD
- jb short .column_st32
-
- test rdi, SIZEOF_XMMWORD-1
- jnz short .out1
- ; --(aligned)-------------------
- movntdq XMMWORD [rdi+0*SIZEOF_XMMWORD], xmmA
- movntdq XMMWORD [rdi+1*SIZEOF_XMMWORD], xmmD
- movntdq XMMWORD [rdi+2*SIZEOF_XMMWORD], xmmF
- jmp short .out0
-.out1: ; --(unaligned)-----------------
- movdqu XMMWORD [rdi+0*SIZEOF_XMMWORD], xmmA
- movdqu XMMWORD [rdi+1*SIZEOF_XMMWORD], xmmD
- movdqu XMMWORD [rdi+2*SIZEOF_XMMWORD], xmmF
-.out0:
- add rdi, byte RGB_PIXELSIZE*SIZEOF_XMMWORD ; outptr
- sub rcx, byte SIZEOF_XMMWORD
- jz near .endcolumn
-
- add rsi, byte SIZEOF_XMMWORD ; inptr0
- dec al ; Yctr
- jnz near .Yloop_2nd
-
- add rbx, byte SIZEOF_XMMWORD ; inptr1
- add rdx, byte SIZEOF_XMMWORD ; inptr2
- jmp near .columnloop
-
-.column_st32:
- lea rcx, [rcx+rcx*2] ; imul ecx, RGB_PIXELSIZE
- cmp rcx, byte 2*SIZEOF_XMMWORD
- jb short .column_st16
- movdqu XMMWORD [rdi+0*SIZEOF_XMMWORD], xmmA
- movdqu XMMWORD [rdi+1*SIZEOF_XMMWORD], xmmD
- add rdi, byte 2*SIZEOF_XMMWORD ; outptr
- movdqa xmmA,xmmF
- sub rcx, byte 2*SIZEOF_XMMWORD
- jmp short .column_st15
-.column_st16:
- cmp rcx, byte SIZEOF_XMMWORD
- jb short .column_st15
- movdqu XMMWORD [rdi+0*SIZEOF_XMMWORD], xmmA
- add rdi, byte SIZEOF_XMMWORD ; outptr
- movdqa xmmA,xmmD
- sub rcx, byte SIZEOF_XMMWORD
-.column_st15:
- ; Store the lower 8 bytes of xmmA to the output when it has enough
- ; space.
- cmp rcx, byte SIZEOF_MMWORD
- jb short .column_st7
- movq XMM_MMWORD [rdi], xmmA
- add rdi, byte SIZEOF_MMWORD
- sub rcx, byte SIZEOF_MMWORD
- psrldq xmmA, SIZEOF_MMWORD
-.column_st7:
- ; Store the lower 4 bytes of xmmA to the output when it has enough
- ; space.
- cmp rcx, byte SIZEOF_DWORD
- jb short .column_st3
- movd XMM_DWORD [rdi], xmmA
- add rdi, byte SIZEOF_DWORD
- sub rcx, byte SIZEOF_DWORD
- psrldq xmmA, SIZEOF_DWORD
-.column_st3:
- ; Store the lower 2 bytes of rax to the output when it has enough
- ; space.
- movd eax, xmmA
- cmp rcx, byte SIZEOF_WORD
- jb short .column_st1
- mov WORD [rdi], ax
- add rdi, byte SIZEOF_WORD
- sub rcx, byte SIZEOF_WORD
- shr rax, 16
-.column_st1:
- ; Store the lower 1 byte of rax to the output when it has enough
- ; space.
- test rcx, rcx
- jz short .endcolumn
- mov BYTE [rdi], al
-
-%else ; RGB_PIXELSIZE == 4 ; -----------
-
-%ifdef RGBX_FILLER_0XFF
- pcmpeqb xmm6,xmm6 ; xmm6=XE=X(02468ACE********)
- pcmpeqb xmm7,xmm7 ; xmm7=XO=X(13579BDF********)
-%else
- pxor xmm6,xmm6 ; xmm6=XE=X(02468ACE********)
- pxor xmm7,xmm7 ; xmm7=XO=X(13579BDF********)
-%endif
- ; xmmA=(00 02 04 06 08 0A 0C 0E **), xmmB=(01 03 05 07 09 0B 0D 0F **)
- ; xmmC=(10 12 14 16 18 1A 1C 1E **), xmmD=(11 13 15 17 19 1B 1D 1F **)
- ; xmmE=(20 22 24 26 28 2A 2C 2E **), xmmF=(21 23 25 27 29 2B 2D 2F **)
- ; xmmG=(30 32 34 36 38 3A 3C 3E **), xmmH=(31 33 35 37 39 3B 3D 3F **)
-
- punpcklbw xmmA,xmmC ; xmmA=(00 10 02 12 04 14 06 16 08 18 0A 1A 0C 1C 0E 1E)
- punpcklbw xmmE,xmmG ; xmmE=(20 30 22 32 24 34 26 36 28 38 2A 3A 2C 3C 2E 3E)
- punpcklbw xmmB,xmmD ; xmmB=(01 11 03 13 05 15 07 17 09 19 0B 1B 0D 1D 0F 1F)
- punpcklbw xmmF,xmmH ; xmmF=(21 31 23 33 25 35 27 37 29 39 2B 3B 2D 3D 2F 3F)
-
- movdqa xmmC,xmmA
- punpcklwd xmmA,xmmE ; xmmA=(00 10 20 30 02 12 22 32 04 14 24 34 06 16 26 36)
- punpckhwd xmmC,xmmE ; xmmC=(08 18 28 38 0A 1A 2A 3A 0C 1C 2C 3C 0E 1E 2E 3E)
- movdqa xmmG,xmmB
- punpcklwd xmmB,xmmF ; xmmB=(01 11 21 31 03 13 23 33 05 15 25 35 07 17 27 37)
- punpckhwd xmmG,xmmF ; xmmG=(09 19 29 39 0B 1B 2B 3B 0D 1D 2D 3D 0F 1F 2F 3F)
-
- movdqa xmmD,xmmA
- punpckldq xmmA,xmmB ; xmmA=(00 10 20 30 01 11 21 31 02 12 22 32 03 13 23 33)
- punpckhdq xmmD,xmmB ; xmmD=(04 14 24 34 05 15 25 35 06 16 26 36 07 17 27 37)
- movdqa xmmH,xmmC
- punpckldq xmmC,xmmG ; xmmC=(08 18 28 38 09 19 29 39 0A 1A 2A 3A 0B 1B 2B 3B)
- punpckhdq xmmH,xmmG ; xmmH=(0C 1C 2C 3C 0D 1D 2D 3D 0E 1E 2E 3E 0F 1F 2F 3F)
-
- cmp rcx, byte SIZEOF_XMMWORD
- jb short .column_st32
-
- test rdi, SIZEOF_XMMWORD-1
- jnz short .out1
- ; --(aligned)-------------------
- movntdq XMMWORD [rdi+0*SIZEOF_XMMWORD], xmmA
- movntdq XMMWORD [rdi+1*SIZEOF_XMMWORD], xmmD
- movntdq XMMWORD [rdi+2*SIZEOF_XMMWORD], xmmC
- movntdq XMMWORD [rdi+3*SIZEOF_XMMWORD], xmmH
- jmp short .out0
-.out1: ; --(unaligned)-----------------
- movdqu XMMWORD [rdi+0*SIZEOF_XMMWORD], xmmA
- movdqu XMMWORD [rdi+1*SIZEOF_XMMWORD], xmmD
- movdqu XMMWORD [rdi+2*SIZEOF_XMMWORD], xmmC
- movdqu XMMWORD [rdi+3*SIZEOF_XMMWORD], xmmH
-.out0:
- add rdi, byte RGB_PIXELSIZE*SIZEOF_XMMWORD ; outptr
- sub rcx, byte SIZEOF_XMMWORD
- jz near .endcolumn
-
- add rsi, byte SIZEOF_XMMWORD ; inptr0
- dec al ; Yctr
- jnz near .Yloop_2nd
-
- add rbx, byte SIZEOF_XMMWORD ; inptr1
- add rdx, byte SIZEOF_XMMWORD ; inptr2
- jmp near .columnloop
-
-.column_st32:
- cmp rcx, byte SIZEOF_XMMWORD/2
- jb short .column_st16
- movdqu XMMWORD [rdi+0*SIZEOF_XMMWORD], xmmA
- movdqu XMMWORD [rdi+1*SIZEOF_XMMWORD], xmmD
- add rdi, byte 2*SIZEOF_XMMWORD ; outptr
- movdqa xmmA,xmmC
- movdqa xmmD,xmmH
- sub rcx, byte SIZEOF_XMMWORD/2
-.column_st16:
- cmp rcx, byte SIZEOF_XMMWORD/4
- jb short .column_st15
- movdqu XMMWORD [rdi+0*SIZEOF_XMMWORD], xmmA
- add rdi, byte SIZEOF_XMMWORD ; outptr
- movdqa xmmA,xmmD
- sub rcx, byte SIZEOF_XMMWORD/4
-.column_st15:
- ; Store two pixels (8 bytes) of xmmA to the output when it has enough
- ; space.
- cmp rcx, byte SIZEOF_XMMWORD/8
- jb short .column_st7
- movq XMM_MMWORD [rdi], xmmA
- add rdi, byte SIZEOF_XMMWORD/8*4
- sub rcx, byte SIZEOF_XMMWORD/8
- psrldq xmmA, SIZEOF_XMMWORD/8*4
-.column_st7:
- ; Store one pixel (4 bytes) of xmmA to the output when it has enough
- ; space.
- test rcx, rcx
- jz short .endcolumn
- movd XMM_DWORD [rdi], xmmA
-
-%endif ; RGB_PIXELSIZE ; ---------------
-
-.endcolumn:
- sfence ; flush the write buffer
-
-.return:
- pop rbx
- uncollect_args
- mov rsp,rbp ; rsp <- aligned rbp
- pop rsp ; rsp <- original rbp
- pop rbp
- ret
-
-; --------------------------------------------------------------------------
-;
-; Upsample and color convert for the case of 2:1 horizontal and 2:1 vertical.
-;
-; GLOBAL(void)
-; jsimd_h2v2_merged_upsample_sse2 (JDIMENSION output_width,
-; JSAMPIMAGE input_buf,
-; JDIMENSION in_row_group_ctr,
-; JSAMPARRAY output_buf);
-;
-
-; r10 = JDIMENSION output_width
-; r11 = JSAMPIMAGE input_buf
-; r12 = JDIMENSION in_row_group_ctr
-; r13 = JSAMPARRAY output_buf
-
- align 16
- global EXTN(jsimd_h2v2_merged_upsample_sse2)
-
-EXTN(jsimd_h2v2_merged_upsample_sse2):
- push rbp
- mov rax,rsp
- mov rbp,rsp
- collect_args
- push rbx
-
- mov eax, r10d
-
- mov rdi, r11
- mov ecx, r12d
- mov rsi, JSAMPARRAY [rdi+0*SIZEOF_JSAMPARRAY]
- mov rbx, JSAMPARRAY [rdi+1*SIZEOF_JSAMPARRAY]
- mov rdx, JSAMPARRAY [rdi+2*SIZEOF_JSAMPARRAY]
- mov rdi, r13
- lea rsi, [rsi+rcx*SIZEOF_JSAMPROW]
-
- push rdx ; inptr2
- push rbx ; inptr1
- push rsi ; inptr00
- mov rbx,rsp
-
- push rdi
- push rcx
- push rax
-
- %ifdef WIN64
- mov r8, rcx
- mov r9, rdi
- mov rcx, rax
- mov rdx, rbx
- %else
- mov rdx, rcx
- mov rcx, rdi
- mov rdi, rax
- mov rsi, rbx
- %endif
-
- call EXTN(jsimd_h2v1_merged_upsample_sse2)
-
- pop rax
- pop rcx
- pop rdi
- pop rsi
- pop rbx
- pop rdx
-
- add rdi, byte SIZEOF_JSAMPROW ; outptr1
- add rsi, byte SIZEOF_JSAMPROW ; inptr01
-
- push rdx ; inptr2
- push rbx ; inptr1
- push rsi ; inptr00
- mov rbx,rsp
-
- push rdi
- push rcx
- push rax
-
- %ifdef WIN64
- mov r8, rcx
- mov r9, rdi
- mov rcx, rax
- mov rdx, rbx
- %else
- mov rdx, rcx
- mov rcx, rdi
- mov rdi, rax
- mov rsi, rbx
- %endif
-
- call EXTN(jsimd_h2v1_merged_upsample_sse2)
-
- pop rax
- pop rcx
- pop rdi
- pop rsi
- pop rbx
- pop rdx
-
- pop rbx
- uncollect_args
- pop rbp
- ret
-
-; For some reason, the OS X linker does not honor the request to align the
-; segment unless we do this.
- align 16
diff --git a/simd/jdmrgext-sse2.asm b/simd/jdmrgext-sse2.asm
deleted file mode 100644
index b50f698..0000000
--- a/simd/jdmrgext-sse2.asm
+++ /dev/null
@@ -1,518 +0,0 @@
-;
-; jdmrgext.asm - merged upsampling/color conversion (SSE2)
-;
-; Copyright 2009, 2012 Pierre Ossman <ossman@cendio.se> for Cendio AB
-; Copyright (C) 2012, D. R. Commander.
-;
-; Based on the x86 SIMD extension for IJG JPEG library
-; Copyright (C) 1999-2006, MIYASAKA Masaru.
-; For conditions of distribution and use, see copyright notice in jsimdext.inc
-;
-; This file should be assembled with NASM (Netwide Assembler),
-; can *not* be assembled with Microsoft's MASM or any compatible
-; assembler (including Borland's Turbo Assembler).
-; NASM is available from http://nasm.sourceforge.net/ or
-; http://sourceforge.net/project/showfiles.php?group_id=6208
-;
-; [TAB8]
-
-%include "jcolsamp.inc"
-
-; --------------------------------------------------------------------------
-;
-; Upsample and color convert for the case of 2:1 horizontal and 1:1 vertical.
-;
-; GLOBAL(void)
-; jsimd_h2v1_merged_upsample_sse2 (JDIMENSION output_width,
-; JSAMPIMAGE input_buf,
-; JDIMENSION in_row_group_ctr,
-; JSAMPARRAY output_buf);
-;
-
-%define output_width(b) (b)+8 ; JDIMENSION output_width
-%define input_buf(b) (b)+12 ; JSAMPIMAGE input_buf
-%define in_row_group_ctr(b) (b)+16 ; JDIMENSION in_row_group_ctr
-%define output_buf(b) (b)+20 ; JSAMPARRAY output_buf
-
-%define original_ebp ebp+0
-%define wk(i) ebp-(WK_NUM-(i))*SIZEOF_XMMWORD ; xmmword wk[WK_NUM]
-%define WK_NUM 3
-%define gotptr wk(0)-SIZEOF_POINTER ; void * gotptr
-
- align 16
- global EXTN(jsimd_h2v1_merged_upsample_sse2)
-
-EXTN(jsimd_h2v1_merged_upsample_sse2):
- push ebp
- mov eax,esp ; eax = original ebp
- sub esp, byte 4
- and esp, byte (-SIZEOF_XMMWORD) ; align to 128 bits
- mov [esp],eax
- mov ebp,esp ; ebp = aligned ebp
- lea esp, [wk(0)]
- pushpic eax ; make a room for GOT address
- push ebx
-; push ecx ; need not be preserved
-; push edx ; need not be preserved
- push esi
- push edi
-
- get_GOT ebx ; get GOT address
- movpic POINTER [gotptr], ebx ; save GOT address
-
- mov ecx, JDIMENSION [output_width(eax)] ; col
- test ecx,ecx
- jz near .return
-
- push ecx
-
- mov edi, JSAMPIMAGE [input_buf(eax)]
- mov ecx, JDIMENSION [in_row_group_ctr(eax)]
- mov esi, JSAMPARRAY [edi+0*SIZEOF_JSAMPARRAY]
- mov ebx, JSAMPARRAY [edi+1*SIZEOF_JSAMPARRAY]
- mov edx, JSAMPARRAY [edi+2*SIZEOF_JSAMPARRAY]
- mov edi, JSAMPARRAY [output_buf(eax)]
- mov esi, JSAMPROW [esi+ecx*SIZEOF_JSAMPROW] ; inptr0
- mov ebx, JSAMPROW [ebx+ecx*SIZEOF_JSAMPROW] ; inptr1
- mov edx, JSAMPROW [edx+ecx*SIZEOF_JSAMPROW] ; inptr2
- mov edi, JSAMPROW [edi] ; outptr
-
- pop ecx ; col
-
- alignx 16,7
-.columnloop:
- movpic eax, POINTER [gotptr] ; load GOT address (eax)
-
- movdqa xmm6, XMMWORD [ebx] ; xmm6=Cb(0123456789ABCDEF)
- movdqa xmm7, XMMWORD [edx] ; xmm7=Cr(0123456789ABCDEF)
-
- pxor xmm1,xmm1 ; xmm1=(all 0's)
- pcmpeqw xmm3,xmm3
- psllw xmm3,7 ; xmm3={0xFF80 0xFF80 0xFF80 0xFF80 ..}
-
- movdqa xmm4,xmm6
- punpckhbw xmm6,xmm1 ; xmm6=Cb(89ABCDEF)=CbH
- punpcklbw xmm4,xmm1 ; xmm4=Cb(01234567)=CbL
- movdqa xmm0,xmm7
- punpckhbw xmm7,xmm1 ; xmm7=Cr(89ABCDEF)=CrH
- punpcklbw xmm0,xmm1 ; xmm0=Cr(01234567)=CrL
-
- paddw xmm6,xmm3
- paddw xmm4,xmm3
- paddw xmm7,xmm3
- paddw xmm0,xmm3
-
- ; (Original)
- ; R = Y + 1.40200 * Cr
- ; G = Y - 0.34414 * Cb - 0.71414 * Cr
- ; B = Y + 1.77200 * Cb
- ;
- ; (This implementation)
- ; R = Y + 0.40200 * Cr + Cr
- ; G = Y - 0.34414 * Cb + 0.28586 * Cr - Cr
- ; B = Y - 0.22800 * Cb + Cb + Cb
-
- movdqa xmm5,xmm6 ; xmm5=CbH
- movdqa xmm2,xmm4 ; xmm2=CbL
- paddw xmm6,xmm6 ; xmm6=2*CbH
- paddw xmm4,xmm4 ; xmm4=2*CbL
- movdqa xmm1,xmm7 ; xmm1=CrH
- movdqa xmm3,xmm0 ; xmm3=CrL
- paddw xmm7,xmm7 ; xmm7=2*CrH
- paddw xmm0,xmm0 ; xmm0=2*CrL
-
- pmulhw xmm6,[GOTOFF(eax,PW_MF0228)] ; xmm6=(2*CbH * -FIX(0.22800))
- pmulhw xmm4,[GOTOFF(eax,PW_MF0228)] ; xmm4=(2*CbL * -FIX(0.22800))
- pmulhw xmm7,[GOTOFF(eax,PW_F0402)] ; xmm7=(2*CrH * FIX(0.40200))
- pmulhw xmm0,[GOTOFF(eax,PW_F0402)] ; xmm0=(2*CrL * FIX(0.40200))
-
- paddw xmm6,[GOTOFF(eax,PW_ONE)]
- paddw xmm4,[GOTOFF(eax,PW_ONE)]
- psraw xmm6,1 ; xmm6=(CbH * -FIX(0.22800))
- psraw xmm4,1 ; xmm4=(CbL * -FIX(0.22800))
- paddw xmm7,[GOTOFF(eax,PW_ONE)]
- paddw xmm0,[GOTOFF(eax,PW_ONE)]
- psraw xmm7,1 ; xmm7=(CrH * FIX(0.40200))
- psraw xmm0,1 ; xmm0=(CrL * FIX(0.40200))
-
- paddw xmm6,xmm5
- paddw xmm4,xmm2
- paddw xmm6,xmm5 ; xmm6=(CbH * FIX(1.77200))=(B-Y)H
- paddw xmm4,xmm2 ; xmm4=(CbL * FIX(1.77200))=(B-Y)L
- paddw xmm7,xmm1 ; xmm7=(CrH * FIX(1.40200))=(R-Y)H
- paddw xmm0,xmm3 ; xmm0=(CrL * FIX(1.40200))=(R-Y)L
-
- movdqa XMMWORD [wk(0)], xmm6 ; wk(0)=(B-Y)H
- movdqa XMMWORD [wk(1)], xmm7 ; wk(1)=(R-Y)H
-
- movdqa xmm6,xmm5
- movdqa xmm7,xmm2
- punpcklwd xmm5,xmm1
- punpckhwd xmm6,xmm1
- pmaddwd xmm5,[GOTOFF(eax,PW_MF0344_F0285)]
- pmaddwd xmm6,[GOTOFF(eax,PW_MF0344_F0285)]
- punpcklwd xmm2,xmm3
- punpckhwd xmm7,xmm3
- pmaddwd xmm2,[GOTOFF(eax,PW_MF0344_F0285)]
- pmaddwd xmm7,[GOTOFF(eax,PW_MF0344_F0285)]
-
- paddd xmm5,[GOTOFF(eax,PD_ONEHALF)]
- paddd xmm6,[GOTOFF(eax,PD_ONEHALF)]
- psrad xmm5,SCALEBITS
- psrad xmm6,SCALEBITS
- paddd xmm2,[GOTOFF(eax,PD_ONEHALF)]
- paddd xmm7,[GOTOFF(eax,PD_ONEHALF)]
- psrad xmm2,SCALEBITS
- psrad xmm7,SCALEBITS
-
- packssdw xmm5,xmm6 ; xmm5=CbH*-FIX(0.344)+CrH*FIX(0.285)
- packssdw xmm2,xmm7 ; xmm2=CbL*-FIX(0.344)+CrL*FIX(0.285)
- psubw xmm5,xmm1 ; xmm5=CbH*-FIX(0.344)+CrH*-FIX(0.714)=(G-Y)H
- psubw xmm2,xmm3 ; xmm2=CbL*-FIX(0.344)+CrL*-FIX(0.714)=(G-Y)L
-
- movdqa XMMWORD [wk(2)], xmm5 ; wk(2)=(G-Y)H
-
- mov al,2 ; Yctr
- jmp short .Yloop_1st
- alignx 16,7
-
-.Yloop_2nd:
- movdqa xmm0, XMMWORD [wk(1)] ; xmm0=(R-Y)H
- movdqa xmm2, XMMWORD [wk(2)] ; xmm2=(G-Y)H
- movdqa xmm4, XMMWORD [wk(0)] ; xmm4=(B-Y)H
- alignx 16,7
-
-.Yloop_1st:
- movdqa xmm7, XMMWORD [esi] ; xmm7=Y(0123456789ABCDEF)
-
- pcmpeqw xmm6,xmm6
- psrlw xmm6,BYTE_BIT ; xmm6={0xFF 0x00 0xFF 0x00 ..}
- pand xmm6,xmm7 ; xmm6=Y(02468ACE)=YE
- psrlw xmm7,BYTE_BIT ; xmm7=Y(13579BDF)=YO
-
- movdqa xmm1,xmm0 ; xmm1=xmm0=(R-Y)(L/H)
- movdqa xmm3,xmm2 ; xmm3=xmm2=(G-Y)(L/H)
- movdqa xmm5,xmm4 ; xmm5=xmm4=(B-Y)(L/H)
-
- paddw xmm0,xmm6 ; xmm0=((R-Y)+YE)=RE=R(02468ACE)
- paddw xmm1,xmm7 ; xmm1=((R-Y)+YO)=RO=R(13579BDF)
- packuswb xmm0,xmm0 ; xmm0=R(02468ACE********)
- packuswb xmm1,xmm1 ; xmm1=R(13579BDF********)
-
- paddw xmm2,xmm6 ; xmm2=((G-Y)+YE)=GE=G(02468ACE)
- paddw xmm3,xmm7 ; xmm3=((G-Y)+YO)=GO=G(13579BDF)
- packuswb xmm2,xmm2 ; xmm2=G(02468ACE********)
- packuswb xmm3,xmm3 ; xmm3=G(13579BDF********)
-
- paddw xmm4,xmm6 ; xmm4=((B-Y)+YE)=BE=B(02468ACE)
- paddw xmm5,xmm7 ; xmm5=((B-Y)+YO)=BO=B(13579BDF)
- packuswb xmm4,xmm4 ; xmm4=B(02468ACE********)
- packuswb xmm5,xmm5 ; xmm5=B(13579BDF********)
-
-%if RGB_PIXELSIZE == 3 ; ---------------
-
- ; xmmA=(00 02 04 06 08 0A 0C 0E **), xmmB=(01 03 05 07 09 0B 0D 0F **)
- ; xmmC=(10 12 14 16 18 1A 1C 1E **), xmmD=(11 13 15 17 19 1B 1D 1F **)
- ; xmmE=(20 22 24 26 28 2A 2C 2E **), xmmF=(21 23 25 27 29 2B 2D 2F **)
- ; xmmG=(** ** ** ** ** ** ** ** **), xmmH=(** ** ** ** ** ** ** ** **)
-
- punpcklbw xmmA,xmmC ; xmmA=(00 10 02 12 04 14 06 16 08 18 0A 1A 0C 1C 0E 1E)
- punpcklbw xmmE,xmmB ; xmmE=(20 01 22 03 24 05 26 07 28 09 2A 0B 2C 0D 2E 0F)
- punpcklbw xmmD,xmmF ; xmmD=(11 21 13 23 15 25 17 27 19 29 1B 2B 1D 2D 1F 2F)
-
- movdqa xmmG,xmmA
- movdqa xmmH,xmmA
- punpcklwd xmmA,xmmE ; xmmA=(00 10 20 01 02 12 22 03 04 14 24 05 06 16 26 07)
- punpckhwd xmmG,xmmE ; xmmG=(08 18 28 09 0A 1A 2A 0B 0C 1C 2C 0D 0E 1E 2E 0F)
-
- psrldq xmmH,2 ; xmmH=(02 12 04 14 06 16 08 18 0A 1A 0C 1C 0E 1E -- --)
- psrldq xmmE,2 ; xmmE=(22 03 24 05 26 07 28 09 2A 0B 2C 0D 2E 0F -- --)
-
- movdqa xmmC,xmmD
- movdqa xmmB,xmmD
- punpcklwd xmmD,xmmH ; xmmD=(11 21 02 12 13 23 04 14 15 25 06 16 17 27 08 18)
- punpckhwd xmmC,xmmH ; xmmC=(19 29 0A 1A 1B 2B 0C 1C 1D 2D 0E 1E 1F 2F -- --)
-
- psrldq xmmB,2 ; xmmB=(13 23 15 25 17 27 19 29 1B 2B 1D 2D 1F 2F -- --)
-
- movdqa xmmF,xmmE
- punpcklwd xmmE,xmmB ; xmmE=(22 03 13 23 24 05 15 25 26 07 17 27 28 09 19 29)
- punpckhwd xmmF,xmmB ; xmmF=(2A 0B 1B 2B 2C 0D 1D 2D 2E 0F 1F 2F -- -- -- --)
-
- pshufd xmmH,xmmA,0x4E; xmmH=(04 14 24 05 06 16 26 07 00 10 20 01 02 12 22 03)
- movdqa xmmB,xmmE
- punpckldq xmmA,xmmD ; xmmA=(00 10 20 01 11 21 02 12 02 12 22 03 13 23 04 14)
- punpckldq xmmE,xmmH ; xmmE=(22 03 13 23 04 14 24 05 24 05 15 25 06 16 26 07)
- punpckhdq xmmD,xmmB ; xmmD=(15 25 06 16 26 07 17 27 17 27 08 18 28 09 19 29)
-
- pshufd xmmH,xmmG,0x4E; xmmH=(0C 1C 2C 0D 0E 1E 2E 0F 08 18 28 09 0A 1A 2A 0B)
- movdqa xmmB,xmmF
- punpckldq xmmG,xmmC ; xmmG=(08 18 28 09 19 29 0A 1A 0A 1A 2A 0B 1B 2B 0C 1C)
- punpckldq xmmF,xmmH ; xmmF=(2A 0B 1B 2B 0C 1C 2C 0D 2C 0D 1D 2D 0E 1E 2E 0F)
- punpckhdq xmmC,xmmB ; xmmC=(1D 2D 0E 1E 2E 0F 1F 2F 1F 2F -- -- -- -- -- --)
-
- punpcklqdq xmmA,xmmE ; xmmA=(00 10 20 01 11 21 02 12 22 03 13 23 04 14 24 05)
- punpcklqdq xmmD,xmmG ; xmmD=(15 25 06 16 26 07 17 27 08 18 28 09 19 29 0A 1A)
- punpcklqdq xmmF,xmmC ; xmmF=(2A 0B 1B 2B 0C 1C 2C 0D 1D 2D 0E 1E 2E 0F 1F 2F)
-
- cmp ecx, byte SIZEOF_XMMWORD
- jb short .column_st32
-
- test edi, SIZEOF_XMMWORD-1
- jnz short .out1
- ; --(aligned)-------------------
- movntdq XMMWORD [edi+0*SIZEOF_XMMWORD], xmmA
- movntdq XMMWORD [edi+1*SIZEOF_XMMWORD], xmmD
- movntdq XMMWORD [edi+2*SIZEOF_XMMWORD], xmmF
- jmp short .out0
-.out1: ; --(unaligned)-----------------
- movdqu XMMWORD [edi+0*SIZEOF_XMMWORD], xmmA
- movdqu XMMWORD [edi+1*SIZEOF_XMMWORD], xmmD
- movdqu XMMWORD [edi+2*SIZEOF_XMMWORD], xmmF
-.out0:
- add edi, byte RGB_PIXELSIZE*SIZEOF_XMMWORD ; outptr
- sub ecx, byte SIZEOF_XMMWORD
- jz near .endcolumn
-
- add esi, byte SIZEOF_XMMWORD ; inptr0
- dec al ; Yctr
- jnz near .Yloop_2nd
-
- add ebx, byte SIZEOF_XMMWORD ; inptr1
- add edx, byte SIZEOF_XMMWORD ; inptr2
- jmp near .columnloop
- alignx 16,7
-
-.column_st32:
- lea ecx, [ecx+ecx*2] ; imul ecx, RGB_PIXELSIZE
- cmp ecx, byte 2*SIZEOF_XMMWORD
- jb short .column_st16
- movdqu XMMWORD [edi+0*SIZEOF_XMMWORD], xmmA
- movdqu XMMWORD [edi+1*SIZEOF_XMMWORD], xmmD
- add edi, byte 2*SIZEOF_XMMWORD ; outptr
- movdqa xmmA,xmmF
- sub ecx, byte 2*SIZEOF_XMMWORD
- jmp short .column_st15
-.column_st16:
- cmp ecx, byte SIZEOF_XMMWORD
- jb short .column_st15
- movdqu XMMWORD [edi+0*SIZEOF_XMMWORD], xmmA
- add edi, byte SIZEOF_XMMWORD ; outptr
- movdqa xmmA,xmmD
- sub ecx, byte SIZEOF_XMMWORD
-.column_st15:
- ; Store the lower 8 bytes of xmmA to the output when it has enough
- ; space.
- cmp ecx, byte SIZEOF_MMWORD
- jb short .column_st7
- movq XMM_MMWORD [edi], xmmA
- add edi, byte SIZEOF_MMWORD
- sub ecx, byte SIZEOF_MMWORD
- psrldq xmmA, SIZEOF_MMWORD
-.column_st7:
- ; Store the lower 4 bytes of xmmA to the output when it has enough
- ; space.
- cmp ecx, byte SIZEOF_DWORD
- jb short .column_st3
- movd XMM_DWORD [edi], xmmA
- add edi, byte SIZEOF_DWORD
- sub ecx, byte SIZEOF_DWORD
- psrldq xmmA, SIZEOF_DWORD
-.column_st3:
- ; Store the lower 2 bytes of eax to the output when it has enough
- ; space.
- movd eax, xmmA
- cmp ecx, byte SIZEOF_WORD
- jb short .column_st1
- mov WORD [edi], ax
- add edi, byte SIZEOF_WORD
- sub ecx, byte SIZEOF_WORD
- shr eax, 16
-.column_st1:
- ; Store the lower 1 byte of eax to the output when it has enough
- ; space.
- test ecx, ecx
- jz short .endcolumn
- mov BYTE [edi], al
-
-%else ; RGB_PIXELSIZE == 4 ; -----------
-
-%ifdef RGBX_FILLER_0XFF
- pcmpeqb xmm6,xmm6 ; xmm6=XE=X(02468ACE********)
- pcmpeqb xmm7,xmm7 ; xmm7=XO=X(13579BDF********)
-%else
- pxor xmm6,xmm6 ; xmm6=XE=X(02468ACE********)
- pxor xmm7,xmm7 ; xmm7=XO=X(13579BDF********)
-%endif
- ; xmmA=(00 02 04 06 08 0A 0C 0E **), xmmB=(01 03 05 07 09 0B 0D 0F **)
- ; xmmC=(10 12 14 16 18 1A 1C 1E **), xmmD=(11 13 15 17 19 1B 1D 1F **)
- ; xmmE=(20 22 24 26 28 2A 2C 2E **), xmmF=(21 23 25 27 29 2B 2D 2F **)
- ; xmmG=(30 32 34 36 38 3A 3C 3E **), xmmH=(31 33 35 37 39 3B 3D 3F **)
-
- punpcklbw xmmA,xmmC ; xmmA=(00 10 02 12 04 14 06 16 08 18 0A 1A 0C 1C 0E 1E)
- punpcklbw xmmE,xmmG ; xmmE=(20 30 22 32 24 34 26 36 28 38 2A 3A 2C 3C 2E 3E)
- punpcklbw xmmB,xmmD ; xmmB=(01 11 03 13 05 15 07 17 09 19 0B 1B 0D 1D 0F 1F)
- punpcklbw xmmF,xmmH ; xmmF=(21 31 23 33 25 35 27 37 29 39 2B 3B 2D 3D 2F 3F)
-
- movdqa xmmC,xmmA
- punpcklwd xmmA,xmmE ; xmmA=(00 10 20 30 02 12 22 32 04 14 24 34 06 16 26 36)
- punpckhwd xmmC,xmmE ; xmmC=(08 18 28 38 0A 1A 2A 3A 0C 1C 2C 3C 0E 1E 2E 3E)
- movdqa xmmG,xmmB
- punpcklwd xmmB,xmmF ; xmmB=(01 11 21 31 03 13 23 33 05 15 25 35 07 17 27 37)
- punpckhwd xmmG,xmmF ; xmmG=(09 19 29 39 0B 1B 2B 3B 0D 1D 2D 3D 0F 1F 2F 3F)
-
- movdqa xmmD,xmmA
- punpckldq xmmA,xmmB ; xmmA=(00 10 20 30 01 11 21 31 02 12 22 32 03 13 23 33)
- punpckhdq xmmD,xmmB ; xmmD=(04 14 24 34 05 15 25 35 06 16 26 36 07 17 27 37)
- movdqa xmmH,xmmC
- punpckldq xmmC,xmmG ; xmmC=(08 18 28 38 09 19 29 39 0A 1A 2A 3A 0B 1B 2B 3B)
- punpckhdq xmmH,xmmG ; xmmH=(0C 1C 2C 3C 0D 1D 2D 3D 0E 1E 2E 3E 0F 1F 2F 3F)
-
- cmp ecx, byte SIZEOF_XMMWORD
- jb short .column_st32
-
- test edi, SIZEOF_XMMWORD-1
- jnz short .out1
- ; --(aligned)-------------------
- movntdq XMMWORD [edi+0*SIZEOF_XMMWORD], xmmA
- movntdq XMMWORD [edi+1*SIZEOF_XMMWORD], xmmD
- movntdq XMMWORD [edi+2*SIZEOF_XMMWORD], xmmC
- movntdq XMMWORD [edi+3*SIZEOF_XMMWORD], xmmH
- jmp short .out0
-.out1: ; --(unaligned)-----------------
- movdqu XMMWORD [edi+0*SIZEOF_XMMWORD], xmmA
- movdqu XMMWORD [edi+1*SIZEOF_XMMWORD], xmmD
- movdqu XMMWORD [edi+2*SIZEOF_XMMWORD], xmmC
- movdqu XMMWORD [edi+3*SIZEOF_XMMWORD], xmmH
-.out0:
- add edi, byte RGB_PIXELSIZE*SIZEOF_XMMWORD ; outptr
- sub ecx, byte SIZEOF_XMMWORD
- jz near .endcolumn
-
- add esi, byte SIZEOF_XMMWORD ; inptr0
- dec al ; Yctr
- jnz near .Yloop_2nd
-
- add ebx, byte SIZEOF_XMMWORD ; inptr1
- add edx, byte SIZEOF_XMMWORD ; inptr2
- jmp near .columnloop
- alignx 16,7
-
-.column_st32:
- cmp ecx, byte SIZEOF_XMMWORD/2
- jb short .column_st16
- movdqu XMMWORD [edi+0*SIZEOF_XMMWORD], xmmA
- movdqu XMMWORD [edi+1*SIZEOF_XMMWORD], xmmD
- add edi, byte 2*SIZEOF_XMMWORD ; outptr
- movdqa xmmA,xmmC
- movdqa xmmD,xmmH
- sub ecx, byte SIZEOF_XMMWORD/2
-.column_st16:
- cmp ecx, byte SIZEOF_XMMWORD/4
- jb short .column_st15
- movdqu XMMWORD [edi+0*SIZEOF_XMMWORD], xmmA
- add edi, byte SIZEOF_XMMWORD ; outptr
- movdqa xmmA,xmmD
- sub ecx, byte SIZEOF_XMMWORD/4
-.column_st15:
- ; Store two pixels (8 bytes) of xmmA to the output when it has enough
- ; space.
- cmp ecx, byte SIZEOF_XMMWORD/8
- jb short .column_st7
- movq XMM_MMWORD [edi], xmmA
- add edi, byte SIZEOF_XMMWORD/8*4
- sub ecx, byte SIZEOF_XMMWORD/8
- psrldq xmmA, SIZEOF_XMMWORD/8*4
-.column_st7:
- ; Store one pixel (4 bytes) of xmmA to the output when it has enough
- ; space.
- test ecx, ecx
- jz short .endcolumn
- movd XMM_DWORD [edi], xmmA
-
-%endif ; RGB_PIXELSIZE ; ---------------
-
-.endcolumn:
- sfence ; flush the write buffer
-
-.return:
- pop edi
- pop esi
-; pop edx ; need not be preserved
-; pop ecx ; need not be preserved
- pop ebx
- mov esp,ebp ; esp <- aligned ebp
- pop esp ; esp <- original ebp
- pop ebp
- ret
-
-; --------------------------------------------------------------------------
-;
-; Upsample and color convert for the case of 2:1 horizontal and 2:1 vertical.
-;
-; GLOBAL(void)
-; jsimd_h2v2_merged_upsample_sse2 (JDIMENSION output_width,
-; JSAMPIMAGE input_buf,
-; JDIMENSION in_row_group_ctr,
-; JSAMPARRAY output_buf);
-;
-
-%define output_width(b) (b)+8 ; JDIMENSION output_width
-%define input_buf(b) (b)+12 ; JSAMPIMAGE input_buf
-%define in_row_group_ctr(b) (b)+16 ; JDIMENSION in_row_group_ctr
-%define output_buf(b) (b)+20 ; JSAMPARRAY output_buf
-
- align 16
- global EXTN(jsimd_h2v2_merged_upsample_sse2)
-
-EXTN(jsimd_h2v2_merged_upsample_sse2):
- push ebp
- mov ebp,esp
- push ebx
-; push ecx ; need not be preserved
-; push edx ; need not be preserved
- push esi
- push edi
-
- mov eax, POINTER [output_width(ebp)]
-
- mov edi, JSAMPIMAGE [input_buf(ebp)]
- mov ecx, JDIMENSION [in_row_group_ctr(ebp)]
- mov esi, JSAMPARRAY [edi+0*SIZEOF_JSAMPARRAY]
- mov ebx, JSAMPARRAY [edi+1*SIZEOF_JSAMPARRAY]
- mov edx, JSAMPARRAY [edi+2*SIZEOF_JSAMPARRAY]
- mov edi, JSAMPARRAY [output_buf(ebp)]
- lea esi, [esi+ecx*SIZEOF_JSAMPROW]
-
- push edx ; inptr2
- push ebx ; inptr1
- push esi ; inptr00
- mov ebx,esp
-
- push edi ; output_buf (outptr0)
- push ecx ; in_row_group_ctr
- push ebx ; input_buf
- push eax ; output_width
-
- call near EXTN(jsimd_h2v1_merged_upsample_sse2)
-
- add esi, byte SIZEOF_JSAMPROW ; inptr01
- add edi, byte SIZEOF_JSAMPROW ; outptr1
- mov POINTER [ebx+0*SIZEOF_POINTER], esi
- mov POINTER [ebx-1*SIZEOF_POINTER], edi
-
- call near EXTN(jsimd_h2v1_merged_upsample_sse2)
-
- add esp, byte 7*SIZEOF_DWORD
-
- pop edi
- pop esi
-; pop edx ; need not be preserved
-; pop ecx ; need not be preserved
- pop ebx
- pop ebp
- ret
-
-; For some reason, the OS X linker does not honor the request to align the
-; segment unless we do this.
- align 16
diff --git a/simd/jdsample-sse2-64.asm b/simd/jdsample-sse2-64.asm
deleted file mode 100644
index 1faaed6..0000000
--- a/simd/jdsample-sse2-64.asm
+++ /dev/null
@@ -1,670 +0,0 @@
-;
-; jdsample.asm - upsampling (64-bit SSE2)
-;
-; Copyright 2009 Pierre Ossman <ossman@cendio.se> for Cendio AB
-; Copyright (C) 2009, D. R. Commander.
-;
-; Based on the x86 SIMD extension for IJG JPEG library
-; Copyright (C) 1999-2006, MIYASAKA Masaru.
-; For conditions of distribution and use, see copyright notice in jsimdext.inc
-;
-; This file should be assembled with NASM (Netwide Assembler),
-; can *not* be assembled with Microsoft's MASM or any compatible
-; assembler (including Borland's Turbo Assembler).
-; NASM is available from http://nasm.sourceforge.net/ or
-; http://sourceforge.net/project/showfiles.php?group_id=6208
-;
-; [TAB8]
-
-%include "jsimdext.inc"
-
-; --------------------------------------------------------------------------
- SECTION SEG_CONST
-
- alignz 16
- global EXTN(jconst_fancy_upsample_sse2)
-
-EXTN(jconst_fancy_upsample_sse2):
-
-PW_ONE times 8 dw 1
-PW_TWO times 8 dw 2
-PW_THREE times 8 dw 3
-PW_SEVEN times 8 dw 7
-PW_EIGHT times 8 dw 8
-
- alignz 16
-
-; --------------------------------------------------------------------------
- SECTION SEG_TEXT
- BITS 64
-;
-; Fancy processing for the common case of 2:1 horizontal and 1:1 vertical.
-;
-; The upsampling algorithm is linear interpolation between pixel centers,
-; also known as a "triangle filter". This is a good compromise between
-; speed and visual quality. The centers of the output pixels are 1/4 and 3/4
-; of the way between input pixel centers.
-;
-; GLOBAL(void)
-; jsimd_h2v1_fancy_upsample_sse2 (int max_v_samp_factor,
-; JDIMENSION downsampled_width,
-; JSAMPARRAY input_data,
-; JSAMPARRAY *output_data_ptr);
-;
-
-; r10 = int max_v_samp_factor
-; r11 = JDIMENSION downsampled_width
-; r12 = JSAMPARRAY input_data
-; r13 = JSAMPARRAY *output_data_ptr
-
- align 16
- global EXTN(jsimd_h2v1_fancy_upsample_sse2)
-
-EXTN(jsimd_h2v1_fancy_upsample_sse2):
- push rbp
- mov rax,rsp
- mov rbp,rsp
- collect_args
-
- mov eax, r11d ; colctr
- test rax,rax
- jz near .return
-
- mov rcx, r10 ; rowctr
- test rcx,rcx
- jz near .return
-
- mov rsi, r12 ; input_data
- mov rdi, r13
- mov rdi, JSAMPARRAY [rdi] ; output_data
-.rowloop:
- push rax ; colctr
- push rdi
- push rsi
-
- mov rsi, JSAMPROW [rsi] ; inptr
- mov rdi, JSAMPROW [rdi] ; outptr
-
- test rax, SIZEOF_XMMWORD-1
- jz short .skip
- mov dl, JSAMPLE [rsi+(rax-1)*SIZEOF_JSAMPLE]
- mov JSAMPLE [rsi+rax*SIZEOF_JSAMPLE], dl ; insert a dummy sample
-.skip:
- pxor xmm0,xmm0 ; xmm0=(all 0's)
- pcmpeqb xmm7,xmm7
- psrldq xmm7,(SIZEOF_XMMWORD-1)
- pand xmm7, XMMWORD [rsi+0*SIZEOF_XMMWORD]
-
- add rax, byte SIZEOF_XMMWORD-1
- and rax, byte -SIZEOF_XMMWORD
- cmp rax, byte SIZEOF_XMMWORD
- ja short .columnloop
-
-.columnloop_last:
- pcmpeqb xmm6,xmm6
- pslldq xmm6,(SIZEOF_XMMWORD-1)
- pand xmm6, XMMWORD [rsi+0*SIZEOF_XMMWORD]
- jmp short .upsample
-
-.columnloop:
- movdqa xmm6, XMMWORD [rsi+1*SIZEOF_XMMWORD]
- pslldq xmm6,(SIZEOF_XMMWORD-1)
-
-.upsample:
- movdqa xmm1, XMMWORD [rsi+0*SIZEOF_XMMWORD]
- movdqa xmm2,xmm1
- movdqa xmm3,xmm1 ; xmm1=( 0 1 2 ... 13 14 15)
- pslldq xmm2,1 ; xmm2=(-- 0 1 ... 12 13 14)
- psrldq xmm3,1 ; xmm3=( 1 2 3 ... 14 15 --)
-
- por xmm2,xmm7 ; xmm2=(-1 0 1 ... 12 13 14)
- por xmm3,xmm6 ; xmm3=( 1 2 3 ... 14 15 16)
-
- movdqa xmm7,xmm1
- psrldq xmm7,(SIZEOF_XMMWORD-1) ; xmm7=(15 -- -- ... -- -- --)
-
- movdqa xmm4,xmm1
- punpcklbw xmm1,xmm0 ; xmm1=( 0 1 2 3 4 5 6 7)
- punpckhbw xmm4,xmm0 ; xmm4=( 8 9 10 11 12 13 14 15)
- movdqa xmm5,xmm2
- punpcklbw xmm2,xmm0 ; xmm2=(-1 0 1 2 3 4 5 6)
- punpckhbw xmm5,xmm0 ; xmm5=( 7 8 9 10 11 12 13 14)
- movdqa xmm6,xmm3
- punpcklbw xmm3,xmm0 ; xmm3=( 1 2 3 4 5 6 7 8)
- punpckhbw xmm6,xmm0 ; xmm6=( 9 10 11 12 13 14 15 16)
-
- pmullw xmm1,[rel PW_THREE]
- pmullw xmm4,[rel PW_THREE]
- paddw xmm2,[rel PW_ONE]
- paddw xmm5,[rel PW_ONE]
- paddw xmm3,[rel PW_TWO]
- paddw xmm6,[rel PW_TWO]
-
- paddw xmm2,xmm1
- paddw xmm5,xmm4
- psrlw xmm2,2 ; xmm2=OutLE=( 0 2 4 6 8 10 12 14)
- psrlw xmm5,2 ; xmm5=OutHE=(16 18 20 22 24 26 28 30)
- paddw xmm3,xmm1
- paddw xmm6,xmm4
- psrlw xmm3,2 ; xmm3=OutLO=( 1 3 5 7 9 11 13 15)
- psrlw xmm6,2 ; xmm6=OutHO=(17 19 21 23 25 27 29 31)
-
- psllw xmm3,BYTE_BIT
- psllw xmm6,BYTE_BIT
- por xmm2,xmm3 ; xmm2=OutL=( 0 1 2 ... 13 14 15)
- por xmm5,xmm6 ; xmm5=OutH=(16 17 18 ... 29 30 31)
-
- movdqa XMMWORD [rdi+0*SIZEOF_XMMWORD], xmm2
- movdqa XMMWORD [rdi+1*SIZEOF_XMMWORD], xmm5
-
- sub rax, byte SIZEOF_XMMWORD
- add rsi, byte 1*SIZEOF_XMMWORD ; inptr
- add rdi, byte 2*SIZEOF_XMMWORD ; outptr
- cmp rax, byte SIZEOF_XMMWORD
- ja near .columnloop
- test eax,eax
- jnz near .columnloop_last
-
- pop rsi
- pop rdi
- pop rax
-
- add rsi, byte SIZEOF_JSAMPROW ; input_data
- add rdi, byte SIZEOF_JSAMPROW ; output_data
- dec rcx ; rowctr
- jg near .rowloop
-
-.return:
- uncollect_args
- pop rbp
- ret
-
-; --------------------------------------------------------------------------
-;
-; Fancy processing for the common case of 2:1 horizontal and 2:1 vertical.
-; Again a triangle filter; see comments for h2v1 case, above.
-;
-; GLOBAL(void)
-; jsimd_h2v2_fancy_upsample_sse2 (int max_v_samp_factor,
-; JDIMENSION downsampled_width,
-; JSAMPARRAY input_data,
-; JSAMPARRAY *output_data_ptr);
-;
-
-; r10 = int max_v_samp_factor
-; r11 = JDIMENSION downsampled_width
-; r12 = JSAMPARRAY input_data
-; r13 = JSAMPARRAY *output_data_ptr
-
-%define wk(i) rbp-(WK_NUM-(i))*SIZEOF_XMMWORD ; xmmword wk[WK_NUM]
-%define WK_NUM 4
-
- align 16
- global EXTN(jsimd_h2v2_fancy_upsample_sse2)
-
-EXTN(jsimd_h2v2_fancy_upsample_sse2):
- push rbp
- mov rax,rsp ; rax = original rbp
- sub rsp, byte 4
- and rsp, byte (-SIZEOF_XMMWORD) ; align to 128 bits
- mov [rsp],rax
- mov rbp,rsp ; rbp = aligned rbp
- lea rsp, [wk(0)]
- collect_args
- push rbx
-
- mov eax, r11d ; colctr
- test rax,rax
- jz near .return
-
- mov rcx, r10 ; rowctr
- test rcx,rcx
- jz near .return
-
- mov rsi, r12 ; input_data
- mov rdi, r13
- mov rdi, JSAMPARRAY [rdi] ; output_data
-.rowloop:
- push rax ; colctr
- push rcx
- push rdi
- push rsi
-
- mov rcx, JSAMPROW [rsi-1*SIZEOF_JSAMPROW] ; inptr1(above)
- mov rbx, JSAMPROW [rsi+0*SIZEOF_JSAMPROW] ; inptr0
- mov rsi, JSAMPROW [rsi+1*SIZEOF_JSAMPROW] ; inptr1(below)
- mov rdx, JSAMPROW [rdi+0*SIZEOF_JSAMPROW] ; outptr0
- mov rdi, JSAMPROW [rdi+1*SIZEOF_JSAMPROW] ; outptr1
-
- test rax, SIZEOF_XMMWORD-1
- jz short .skip
- push rdx
- mov dl, JSAMPLE [rcx+(rax-1)*SIZEOF_JSAMPLE]
- mov JSAMPLE [rcx+rax*SIZEOF_JSAMPLE], dl
- mov dl, JSAMPLE [rbx+(rax-1)*SIZEOF_JSAMPLE]
- mov JSAMPLE [rbx+rax*SIZEOF_JSAMPLE], dl
- mov dl, JSAMPLE [rsi+(rax-1)*SIZEOF_JSAMPLE]
- mov JSAMPLE [rsi+rax*SIZEOF_JSAMPLE], dl ; insert a dummy sample
- pop rdx
-.skip:
- ; -- process the first column block
-
- movdqa xmm0, XMMWORD [rbx+0*SIZEOF_XMMWORD] ; xmm0=row[ 0][0]
- movdqa xmm1, XMMWORD [rcx+0*SIZEOF_XMMWORD] ; xmm1=row[-1][0]
- movdqa xmm2, XMMWORD [rsi+0*SIZEOF_XMMWORD] ; xmm2=row[+1][0]
-
- pxor xmm3,xmm3 ; xmm3=(all 0's)
- movdqa xmm4,xmm0
- punpcklbw xmm0,xmm3 ; xmm0=row[ 0]( 0 1 2 3 4 5 6 7)
- punpckhbw xmm4,xmm3 ; xmm4=row[ 0]( 8 9 10 11 12 13 14 15)
- movdqa xmm5,xmm1
- punpcklbw xmm1,xmm3 ; xmm1=row[-1]( 0 1 2 3 4 5 6 7)
- punpckhbw xmm5,xmm3 ; xmm5=row[-1]( 8 9 10 11 12 13 14 15)
- movdqa xmm6,xmm2
- punpcklbw xmm2,xmm3 ; xmm2=row[+1]( 0 1 2 3 4 5 6 7)
- punpckhbw xmm6,xmm3 ; xmm6=row[+1]( 8 9 10 11 12 13 14 15)
-
- pmullw xmm0,[rel PW_THREE]
- pmullw xmm4,[rel PW_THREE]
-
- pcmpeqb xmm7,xmm7
- psrldq xmm7,(SIZEOF_XMMWORD-2)
-
- paddw xmm1,xmm0 ; xmm1=Int0L=( 0 1 2 3 4 5 6 7)
- paddw xmm5,xmm4 ; xmm5=Int0H=( 8 9 10 11 12 13 14 15)
- paddw xmm2,xmm0 ; xmm2=Int1L=( 0 1 2 3 4 5 6 7)
- paddw xmm6,xmm4 ; xmm6=Int1H=( 8 9 10 11 12 13 14 15)
-
- movdqa XMMWORD [rdx+0*SIZEOF_XMMWORD], xmm1 ; temporarily save
- movdqa XMMWORD [rdx+1*SIZEOF_XMMWORD], xmm5 ; the intermediate data
- movdqa XMMWORD [rdi+0*SIZEOF_XMMWORD], xmm2
- movdqa XMMWORD [rdi+1*SIZEOF_XMMWORD], xmm6
-
- pand xmm1,xmm7 ; xmm1=( 0 -- -- -- -- -- -- --)
- pand xmm2,xmm7 ; xmm2=( 0 -- -- -- -- -- -- --)
-
- movdqa XMMWORD [wk(0)], xmm1
- movdqa XMMWORD [wk(1)], xmm2
-
- add rax, byte SIZEOF_XMMWORD-1
- and rax, byte -SIZEOF_XMMWORD
- cmp rax, byte SIZEOF_XMMWORD
- ja short .columnloop
-
-.columnloop_last:
- ; -- process the last column block
-
- pcmpeqb xmm1,xmm1
- pslldq xmm1,(SIZEOF_XMMWORD-2)
- movdqa xmm2,xmm1
-
- pand xmm1, XMMWORD [rdx+1*SIZEOF_XMMWORD]
- pand xmm2, XMMWORD [rdi+1*SIZEOF_XMMWORD]
-
- movdqa XMMWORD [wk(2)], xmm1 ; xmm1=(-- -- -- -- -- -- -- 15)
- movdqa XMMWORD [wk(3)], xmm2 ; xmm2=(-- -- -- -- -- -- -- 15)
-
- jmp near .upsample
-
-.columnloop:
- ; -- process the next column block
-
- movdqa xmm0, XMMWORD [rbx+1*SIZEOF_XMMWORD] ; xmm0=row[ 0][1]
- movdqa xmm1, XMMWORD [rcx+1*SIZEOF_XMMWORD] ; xmm1=row[-1][1]
- movdqa xmm2, XMMWORD [rsi+1*SIZEOF_XMMWORD] ; xmm2=row[+1][1]
-
- pxor xmm3,xmm3 ; xmm3=(all 0's)
- movdqa xmm4,xmm0
- punpcklbw xmm0,xmm3 ; xmm0=row[ 0]( 0 1 2 3 4 5 6 7)
- punpckhbw xmm4,xmm3 ; xmm4=row[ 0]( 8 9 10 11 12 13 14 15)
- movdqa xmm5,xmm1
- punpcklbw xmm1,xmm3 ; xmm1=row[-1]( 0 1 2 3 4 5 6 7)
- punpckhbw xmm5,xmm3 ; xmm5=row[-1]( 8 9 10 11 12 13 14 15)
- movdqa xmm6,xmm2
- punpcklbw xmm2,xmm3 ; xmm2=row[+1]( 0 1 2 3 4 5 6 7)
- punpckhbw xmm6,xmm3 ; xmm6=row[+1]( 8 9 10 11 12 13 14 15)
-
- pmullw xmm0,[rel PW_THREE]
- pmullw xmm4,[rel PW_THREE]
-
- paddw xmm1,xmm0 ; xmm1=Int0L=( 0 1 2 3 4 5 6 7)
- paddw xmm5,xmm4 ; xmm5=Int0H=( 8 9 10 11 12 13 14 15)
- paddw xmm2,xmm0 ; xmm2=Int1L=( 0 1 2 3 4 5 6 7)
- paddw xmm6,xmm4 ; xmm6=Int1H=( 8 9 10 11 12 13 14 15)
-
- movdqa XMMWORD [rdx+2*SIZEOF_XMMWORD], xmm1 ; temporarily save
- movdqa XMMWORD [rdx+3*SIZEOF_XMMWORD], xmm5 ; the intermediate data
- movdqa XMMWORD [rdi+2*SIZEOF_XMMWORD], xmm2
- movdqa XMMWORD [rdi+3*SIZEOF_XMMWORD], xmm6
-
- pslldq xmm1,(SIZEOF_XMMWORD-2) ; xmm1=(-- -- -- -- -- -- -- 0)
- pslldq xmm2,(SIZEOF_XMMWORD-2) ; xmm2=(-- -- -- -- -- -- -- 0)
-
- movdqa XMMWORD [wk(2)], xmm1
- movdqa XMMWORD [wk(3)], xmm2
-
-.upsample:
- ; -- process the upper row
-
- movdqa xmm7, XMMWORD [rdx+0*SIZEOF_XMMWORD]
- movdqa xmm3, XMMWORD [rdx+1*SIZEOF_XMMWORD]
-
- movdqa xmm0,xmm7 ; xmm7=Int0L=( 0 1 2 3 4 5 6 7)
- movdqa xmm4,xmm3 ; xmm3=Int0H=( 8 9 10 11 12 13 14 15)
- psrldq xmm0,2 ; xmm0=( 1 2 3 4 5 6 7 --)
- pslldq xmm4,(SIZEOF_XMMWORD-2) ; xmm4=(-- -- -- -- -- -- -- 8)
- movdqa xmm5,xmm7
- movdqa xmm6,xmm3
- psrldq xmm5,(SIZEOF_XMMWORD-2) ; xmm5=( 7 -- -- -- -- -- -- --)
- pslldq xmm6,2 ; xmm6=(-- 8 9 10 11 12 13 14)
-
- por xmm0,xmm4 ; xmm0=( 1 2 3 4 5 6 7 8)
- por xmm5,xmm6 ; xmm5=( 7 8 9 10 11 12 13 14)
-
- movdqa xmm1,xmm7
- movdqa xmm2,xmm3
- pslldq xmm1,2 ; xmm1=(-- 0 1 2 3 4 5 6)
- psrldq xmm2,2 ; xmm2=( 9 10 11 12 13 14 15 --)
- movdqa xmm4,xmm3
- psrldq xmm4,(SIZEOF_XMMWORD-2) ; xmm4=(15 -- -- -- -- -- -- --)
-
- por xmm1, XMMWORD [wk(0)] ; xmm1=(-1 0 1 2 3 4 5 6)
- por xmm2, XMMWORD [wk(2)] ; xmm2=( 9 10 11 12 13 14 15 16)
-
- movdqa XMMWORD [wk(0)], xmm4
-
- pmullw xmm7,[rel PW_THREE]
- pmullw xmm3,[rel PW_THREE]
- paddw xmm1,[rel PW_EIGHT]
- paddw xmm5,[rel PW_EIGHT]
- paddw xmm0,[rel PW_SEVEN]
- paddw xmm2,[rel PW_SEVEN]
-
- paddw xmm1,xmm7
- paddw xmm5,xmm3
- psrlw xmm1,4 ; xmm1=Out0LE=( 0 2 4 6 8 10 12 14)
- psrlw xmm5,4 ; xmm5=Out0HE=(16 18 20 22 24 26 28 30)
- paddw xmm0,xmm7
- paddw xmm2,xmm3
- psrlw xmm0,4 ; xmm0=Out0LO=( 1 3 5 7 9 11 13 15)
- psrlw xmm2,4 ; xmm2=Out0HO=(17 19 21 23 25 27 29 31)
-
- psllw xmm0,BYTE_BIT
- psllw xmm2,BYTE_BIT
- por xmm1,xmm0 ; xmm1=Out0L=( 0 1 2 ... 13 14 15)
- por xmm5,xmm2 ; xmm5=Out0H=(16 17 18 ... 29 30 31)
-
- movdqa XMMWORD [rdx+0*SIZEOF_XMMWORD], xmm1
- movdqa XMMWORD [rdx+1*SIZEOF_XMMWORD], xmm5
-
- ; -- process the lower row
-
- movdqa xmm6, XMMWORD [rdi+0*SIZEOF_XMMWORD]
- movdqa xmm4, XMMWORD [rdi+1*SIZEOF_XMMWORD]
-
- movdqa xmm7,xmm6 ; xmm6=Int1L=( 0 1 2 3 4 5 6 7)
- movdqa xmm3,xmm4 ; xmm4=Int1H=( 8 9 10 11 12 13 14 15)
- psrldq xmm7,2 ; xmm7=( 1 2 3 4 5 6 7 --)
- pslldq xmm3,(SIZEOF_XMMWORD-2) ; xmm3=(-- -- -- -- -- -- -- 8)
- movdqa xmm0,xmm6
- movdqa xmm2,xmm4
- psrldq xmm0,(SIZEOF_XMMWORD-2) ; xmm0=( 7 -- -- -- -- -- -- --)
- pslldq xmm2,2 ; xmm2=(-- 8 9 10 11 12 13 14)
-
- por xmm7,xmm3 ; xmm7=( 1 2 3 4 5 6 7 8)
- por xmm0,xmm2 ; xmm0=( 7 8 9 10 11 12 13 14)
-
- movdqa xmm1,xmm6
- movdqa xmm5,xmm4
- pslldq xmm1,2 ; xmm1=(-- 0 1 2 3 4 5 6)
- psrldq xmm5,2 ; xmm5=( 9 10 11 12 13 14 15 --)
- movdqa xmm3,xmm4
- psrldq xmm3,(SIZEOF_XMMWORD-2) ; xmm3=(15 -- -- -- -- -- -- --)
-
- por xmm1, XMMWORD [wk(1)] ; xmm1=(-1 0 1 2 3 4 5 6)
- por xmm5, XMMWORD [wk(3)] ; xmm5=( 9 10 11 12 13 14 15 16)
-
- movdqa XMMWORD [wk(1)], xmm3
-
- pmullw xmm6,[rel PW_THREE]
- pmullw xmm4,[rel PW_THREE]
- paddw xmm1,[rel PW_EIGHT]
- paddw xmm0,[rel PW_EIGHT]
- paddw xmm7,[rel PW_SEVEN]
- paddw xmm5,[rel PW_SEVEN]
-
- paddw xmm1,xmm6
- paddw xmm0,xmm4
- psrlw xmm1,4 ; xmm1=Out1LE=( 0 2 4 6 8 10 12 14)
- psrlw xmm0,4 ; xmm0=Out1HE=(16 18 20 22 24 26 28 30)
- paddw xmm7,xmm6
- paddw xmm5,xmm4
- psrlw xmm7,4 ; xmm7=Out1LO=( 1 3 5 7 9 11 13 15)
- psrlw xmm5,4 ; xmm5=Out1HO=(17 19 21 23 25 27 29 31)
-
- psllw xmm7,BYTE_BIT
- psllw xmm5,BYTE_BIT
- por xmm1,xmm7 ; xmm1=Out1L=( 0 1 2 ... 13 14 15)
- por xmm0,xmm5 ; xmm0=Out1H=(16 17 18 ... 29 30 31)
-
- movdqa XMMWORD [rdi+0*SIZEOF_XMMWORD], xmm1
- movdqa XMMWORD [rdi+1*SIZEOF_XMMWORD], xmm0
-
- sub rax, byte SIZEOF_XMMWORD
- add rcx, byte 1*SIZEOF_XMMWORD ; inptr1(above)
- add rbx, byte 1*SIZEOF_XMMWORD ; inptr0
- add rsi, byte 1*SIZEOF_XMMWORD ; inptr1(below)
- add rdx, byte 2*SIZEOF_XMMWORD ; outptr0
- add rdi, byte 2*SIZEOF_XMMWORD ; outptr1
- cmp rax, byte SIZEOF_XMMWORD
- ja near .columnloop
- test rax,rax
- jnz near .columnloop_last
-
- pop rsi
- pop rdi
- pop rcx
- pop rax
-
- add rsi, byte 1*SIZEOF_JSAMPROW ; input_data
- add rdi, byte 2*SIZEOF_JSAMPROW ; output_data
- sub rcx, byte 2 ; rowctr
- jg near .rowloop
-
-.return:
- pop rbx
- uncollect_args
- mov rsp,rbp ; rsp <- aligned rbp
- pop rsp ; rsp <- original rbp
- pop rbp
- ret
-
-; --------------------------------------------------------------------------
-;
-; Fast processing for the common case of 2:1 horizontal and 1:1 vertical.
-; It's still a box filter.
-;
-; GLOBAL(void)
-; jsimd_h2v1_upsample_sse2 (int max_v_samp_factor,
-; JDIMENSION output_width,
-; JSAMPARRAY input_data,
-; JSAMPARRAY *output_data_ptr);
-;
-
-; r10 = int max_v_samp_factor
-; r11 = JDIMENSION output_width
-; r12 = JSAMPARRAY input_data
-; r13 = JSAMPARRAY *output_data_ptr
-
- align 16
- global EXTN(jsimd_h2v1_upsample_sse2)
-
-EXTN(jsimd_h2v1_upsample_sse2):
- push rbp
- mov rax,rsp
- mov rbp,rsp
- collect_args
-
- mov edx, r11d
- add rdx, byte (2*SIZEOF_XMMWORD)-1
- and rdx, byte -(2*SIZEOF_XMMWORD)
- jz near .return
-
- mov rcx, r10 ; rowctr
- test rcx,rcx
- jz short .return
-
- mov rsi, r12 ; input_data
- mov rdi, r13
- mov rdi, JSAMPARRAY [rdi] ; output_data
-.rowloop:
- push rdi
- push rsi
-
- mov rsi, JSAMPROW [rsi] ; inptr
- mov rdi, JSAMPROW [rdi] ; outptr
- mov rax,rdx ; colctr
-.columnloop:
-
- movdqa xmm0, XMMWORD [rsi+0*SIZEOF_XMMWORD]
-
- movdqa xmm1,xmm0
- punpcklbw xmm0,xmm0
- punpckhbw xmm1,xmm1
-
- movdqa XMMWORD [rdi+0*SIZEOF_XMMWORD], xmm0
- movdqa XMMWORD [rdi+1*SIZEOF_XMMWORD], xmm1
-
- sub rax, byte 2*SIZEOF_XMMWORD
- jz short .nextrow
-
- movdqa xmm2, XMMWORD [rsi+1*SIZEOF_XMMWORD]
-
- movdqa xmm3,xmm2
- punpcklbw xmm2,xmm2
- punpckhbw xmm3,xmm3
-
- movdqa XMMWORD [rdi+2*SIZEOF_XMMWORD], xmm2
- movdqa XMMWORD [rdi+3*SIZEOF_XMMWORD], xmm3
-
- sub rax, byte 2*SIZEOF_XMMWORD
- jz short .nextrow
-
- add rsi, byte 2*SIZEOF_XMMWORD ; inptr
- add rdi, byte 4*SIZEOF_XMMWORD ; outptr
- jmp short .columnloop
-
-.nextrow:
- pop rsi
- pop rdi
-
- add rsi, byte SIZEOF_JSAMPROW ; input_data
- add rdi, byte SIZEOF_JSAMPROW ; output_data
- dec rcx ; rowctr
- jg short .rowloop
-
-.return:
- uncollect_args
- pop rbp
- ret
-
-; --------------------------------------------------------------------------
-;
-; Fast processing for the common case of 2:1 horizontal and 2:1 vertical.
-; It's still a box filter.
-;
-; GLOBAL(void)
-; jsimd_h2v2_upsample_sse2 (nt max_v_samp_factor,
-; JDIMENSION output_width,
-; JSAMPARRAY input_data,
-; JSAMPARRAY *output_data_ptr);
-;
-
-; r10 = int max_v_samp_factor
-; r11 = JDIMENSION output_width
-; r12 = JSAMPARRAY input_data
-; r13 = JSAMPARRAY *output_data_ptr
-
- align 16
- global EXTN(jsimd_h2v2_upsample_sse2)
-
-EXTN(jsimd_h2v2_upsample_sse2):
- push rbp
- mov rax,rsp
- mov rbp,rsp
- collect_args
- push rbx
-
- mov edx, r11d
- add rdx, byte (2*SIZEOF_XMMWORD)-1
- and rdx, byte -(2*SIZEOF_XMMWORD)
- jz near .return
-
- mov rcx, r10 ; rowctr
- test rcx,rcx
- jz near .return
-
- mov rsi, r12 ; input_data
- mov rdi, r13
- mov rdi, JSAMPARRAY [rdi] ; output_data
-.rowloop:
- push rdi
- push rsi
-
- mov rsi, JSAMPROW [rsi] ; inptr
- mov rbx, JSAMPROW [rdi+0*SIZEOF_JSAMPROW] ; outptr0
- mov rdi, JSAMPROW [rdi+1*SIZEOF_JSAMPROW] ; outptr1
- mov rax,rdx ; colctr
-.columnloop:
-
- movdqa xmm0, XMMWORD [rsi+0*SIZEOF_XMMWORD]
-
- movdqa xmm1,xmm0
- punpcklbw xmm0,xmm0
- punpckhbw xmm1,xmm1
-
- movdqa XMMWORD [rbx+0*SIZEOF_XMMWORD], xmm0
- movdqa XMMWORD [rbx+1*SIZEOF_XMMWORD], xmm1
- movdqa XMMWORD [rdi+0*SIZEOF_XMMWORD], xmm0
- movdqa XMMWORD [rdi+1*SIZEOF_XMMWORD], xmm1
-
- sub rax, byte 2*SIZEOF_XMMWORD
- jz short .nextrow
-
- movdqa xmm2, XMMWORD [rsi+1*SIZEOF_XMMWORD]
-
- movdqa xmm3,xmm2
- punpcklbw xmm2,xmm2
- punpckhbw xmm3,xmm3
-
- movdqa XMMWORD [rbx+2*SIZEOF_XMMWORD], xmm2
- movdqa XMMWORD [rbx+3*SIZEOF_XMMWORD], xmm3
- movdqa XMMWORD [rdi+2*SIZEOF_XMMWORD], xmm2
- movdqa XMMWORD [rdi+3*SIZEOF_XMMWORD], xmm3
-
- sub rax, byte 2*SIZEOF_XMMWORD
- jz short .nextrow
-
- add rsi, byte 2*SIZEOF_XMMWORD ; inptr
- add rbx, byte 4*SIZEOF_XMMWORD ; outptr0
- add rdi, byte 4*SIZEOF_XMMWORD ; outptr1
- jmp short .columnloop
-
-.nextrow:
- pop rsi
- pop rdi
-
- add rsi, byte 1*SIZEOF_JSAMPROW ; input_data
- add rdi, byte 2*SIZEOF_JSAMPROW ; output_data
- sub rcx, byte 2 ; rowctr
- jg near .rowloop
-
-.return:
- pop rbx
- uncollect_args
- pop rbp
- ret
-
-; For some reason, the OS X linker does not honor the request to align the
-; segment unless we do this.
- align 16
diff --git a/simd/jdsample-sse2.asm b/simd/jdsample-sse2.asm
deleted file mode 100644
index 1d0059e..0000000
--- a/simd/jdsample-sse2.asm
+++ /dev/null
@@ -1,728 +0,0 @@
-;
-; jdsample.asm - upsampling (SSE2)
-;
-; Copyright 2009 Pierre Ossman <ossman@cendio.se> for Cendio AB
-;
-; Based on the x86 SIMD extension for IJG JPEG library
-; Copyright (C) 1999-2006, MIYASAKA Masaru.
-; For conditions of distribution and use, see copyright notice in jsimdext.inc
-;
-; This file should be assembled with NASM (Netwide Assembler),
-; can *not* be assembled with Microsoft's MASM or any compatible
-; assembler (including Borland's Turbo Assembler).
-; NASM is available from http://nasm.sourceforge.net/ or
-; http://sourceforge.net/project/showfiles.php?group_id=6208
-;
-; [TAB8]
-
-%include "jsimdext.inc"
-
-; --------------------------------------------------------------------------
- SECTION SEG_CONST
-
- alignz 16
- global EXTN(jconst_fancy_upsample_sse2)
-
-EXTN(jconst_fancy_upsample_sse2):
-
-PW_ONE times 8 dw 1
-PW_TWO times 8 dw 2
-PW_THREE times 8 dw 3
-PW_SEVEN times 8 dw 7
-PW_EIGHT times 8 dw 8
-
- alignz 16
-
-; --------------------------------------------------------------------------
- SECTION SEG_TEXT
- BITS 32
-;
-; Fancy processing for the common case of 2:1 horizontal and 1:1 vertical.
-;
-; The upsampling algorithm is linear interpolation between pixel centers,
-; also known as a "triangle filter". This is a good compromise between
-; speed and visual quality. The centers of the output pixels are 1/4 and 3/4
-; of the way between input pixel centers.
-;
-; GLOBAL(void)
-; jsimd_h2v1_fancy_upsample_sse2 (int max_v_samp_factor,
-; JDIMENSION downsampled_width,
-; JSAMPARRAY input_data,
-; JSAMPARRAY *output_data_ptr);
-;
-
-%define max_v_samp(b) (b)+8 ; int max_v_samp_factor
-%define downsamp_width(b) (b)+12 ; JDIMENSION downsampled_width
-%define input_data(b) (b)+16 ; JSAMPARRAY input_data
-%define output_data_ptr(b) (b)+20 ; JSAMPARRAY *output_data_ptr
-
- align 16
- global EXTN(jsimd_h2v1_fancy_upsample_sse2)
-
-EXTN(jsimd_h2v1_fancy_upsample_sse2):
- push ebp
- mov ebp,esp
- pushpic ebx
-; push ecx ; need not be preserved
-; push edx ; need not be preserved
- push esi
- push edi
-
- get_GOT ebx ; get GOT address
-
- mov eax, JDIMENSION [downsamp_width(ebp)] ; colctr
- test eax,eax
- jz near .return
-
- mov ecx, INT [max_v_samp(ebp)] ; rowctr
- test ecx,ecx
- jz near .return
-
- mov esi, JSAMPARRAY [input_data(ebp)] ; input_data
- mov edi, POINTER [output_data_ptr(ebp)]
- mov edi, JSAMPARRAY [edi] ; output_data
- alignx 16,7
-.rowloop:
- push eax ; colctr
- push edi
- push esi
-
- mov esi, JSAMPROW [esi] ; inptr
- mov edi, JSAMPROW [edi] ; outptr
-
- test eax, SIZEOF_XMMWORD-1
- jz short .skip
- mov dl, JSAMPLE [esi+(eax-1)*SIZEOF_JSAMPLE]
- mov JSAMPLE [esi+eax*SIZEOF_JSAMPLE], dl ; insert a dummy sample
-.skip:
- pxor xmm0,xmm0 ; xmm0=(all 0's)
- pcmpeqb xmm7,xmm7
- psrldq xmm7,(SIZEOF_XMMWORD-1)
- pand xmm7, XMMWORD [esi+0*SIZEOF_XMMWORD]
-
- add eax, byte SIZEOF_XMMWORD-1
- and eax, byte -SIZEOF_XMMWORD
- cmp eax, byte SIZEOF_XMMWORD
- ja short .columnloop
- alignx 16,7
-
-.columnloop_last:
- pcmpeqb xmm6,xmm6
- pslldq xmm6,(SIZEOF_XMMWORD-1)
- pand xmm6, XMMWORD [esi+0*SIZEOF_XMMWORD]
- jmp short .upsample
- alignx 16,7
-
-.columnloop:
- movdqa xmm6, XMMWORD [esi+1*SIZEOF_XMMWORD]
- pslldq xmm6,(SIZEOF_XMMWORD-1)
-
-.upsample:
- movdqa xmm1, XMMWORD [esi+0*SIZEOF_XMMWORD]
- movdqa xmm2,xmm1
- movdqa xmm3,xmm1 ; xmm1=( 0 1 2 ... 13 14 15)
- pslldq xmm2,1 ; xmm2=(-- 0 1 ... 12 13 14)
- psrldq xmm3,1 ; xmm3=( 1 2 3 ... 14 15 --)
-
- por xmm2,xmm7 ; xmm2=(-1 0 1 ... 12 13 14)
- por xmm3,xmm6 ; xmm3=( 1 2 3 ... 14 15 16)
-
- movdqa xmm7,xmm1
- psrldq xmm7,(SIZEOF_XMMWORD-1) ; xmm7=(15 -- -- ... -- -- --)
-
- movdqa xmm4,xmm1
- punpcklbw xmm1,xmm0 ; xmm1=( 0 1 2 3 4 5 6 7)
- punpckhbw xmm4,xmm0 ; xmm4=( 8 9 10 11 12 13 14 15)
- movdqa xmm5,xmm2
- punpcklbw xmm2,xmm0 ; xmm2=(-1 0 1 2 3 4 5 6)
- punpckhbw xmm5,xmm0 ; xmm5=( 7 8 9 10 11 12 13 14)
- movdqa xmm6,xmm3
- punpcklbw xmm3,xmm0 ; xmm3=( 1 2 3 4 5 6 7 8)
- punpckhbw xmm6,xmm0 ; xmm6=( 9 10 11 12 13 14 15 16)
-
- pmullw xmm1,[GOTOFF(ebx,PW_THREE)]
- pmullw xmm4,[GOTOFF(ebx,PW_THREE)]
- paddw xmm2,[GOTOFF(ebx,PW_ONE)]
- paddw xmm5,[GOTOFF(ebx,PW_ONE)]
- paddw xmm3,[GOTOFF(ebx,PW_TWO)]
- paddw xmm6,[GOTOFF(ebx,PW_TWO)]
-
- paddw xmm2,xmm1
- paddw xmm5,xmm4
- psrlw xmm2,2 ; xmm2=OutLE=( 0 2 4 6 8 10 12 14)
- psrlw xmm5,2 ; xmm5=OutHE=(16 18 20 22 24 26 28 30)
- paddw xmm3,xmm1
- paddw xmm6,xmm4
- psrlw xmm3,2 ; xmm3=OutLO=( 1 3 5 7 9 11 13 15)
- psrlw xmm6,2 ; xmm6=OutHO=(17 19 21 23 25 27 29 31)
-
- psllw xmm3,BYTE_BIT
- psllw xmm6,BYTE_BIT
- por xmm2,xmm3 ; xmm2=OutL=( 0 1 2 ... 13 14 15)
- por xmm5,xmm6 ; xmm5=OutH=(16 17 18 ... 29 30 31)
-
- movdqa XMMWORD [edi+0*SIZEOF_XMMWORD], xmm2
- movdqa XMMWORD [edi+1*SIZEOF_XMMWORD], xmm5
-
- sub eax, byte SIZEOF_XMMWORD
- add esi, byte 1*SIZEOF_XMMWORD ; inptr
- add edi, byte 2*SIZEOF_XMMWORD ; outptr
- cmp eax, byte SIZEOF_XMMWORD
- ja near .columnloop
- test eax,eax
- jnz near .columnloop_last
-
- pop esi
- pop edi
- pop eax
-
- add esi, byte SIZEOF_JSAMPROW ; input_data
- add edi, byte SIZEOF_JSAMPROW ; output_data
- dec ecx ; rowctr
- jg near .rowloop
-
-.return:
- pop edi
- pop esi
-; pop edx ; need not be preserved
-; pop ecx ; need not be preserved
- poppic ebx
- pop ebp
- ret
-
-; --------------------------------------------------------------------------
-;
-; Fancy processing for the common case of 2:1 horizontal and 2:1 vertical.
-; Again a triangle filter; see comments for h2v1 case, above.
-;
-; GLOBAL(void)
-; jsimd_h2v2_fancy_upsample_sse2 (int max_v_samp_factor,
-; JDIMENSION downsampled_width,
-; JSAMPARRAY input_data,
-; JSAMPARRAY *output_data_ptr);
-;
-
-%define max_v_samp(b) (b)+8 ; int max_v_samp_factor
-%define downsamp_width(b) (b)+12 ; JDIMENSION downsampled_width
-%define input_data(b) (b)+16 ; JSAMPARRAY input_data
-%define output_data_ptr(b) (b)+20 ; JSAMPARRAY *output_data_ptr
-
-%define original_ebp ebp+0
-%define wk(i) ebp-(WK_NUM-(i))*SIZEOF_XMMWORD ; xmmword wk[WK_NUM]
-%define WK_NUM 4
-%define gotptr wk(0)-SIZEOF_POINTER ; void *gotptr
-
- align 16
- global EXTN(jsimd_h2v2_fancy_upsample_sse2)
-
-EXTN(jsimd_h2v2_fancy_upsample_sse2):
- push ebp
- mov eax,esp ; eax = original ebp
- sub esp, byte 4
- and esp, byte (-SIZEOF_XMMWORD) ; align to 128 bits
- mov [esp],eax
- mov ebp,esp ; ebp = aligned ebp
- lea esp, [wk(0)]
- pushpic eax ; make a room for GOT address
- push ebx
-; push ecx ; need not be preserved
-; push edx ; need not be preserved
- push esi
- push edi
-
- get_GOT ebx ; get GOT address
- movpic POINTER [gotptr], ebx ; save GOT address
-
- mov edx,eax ; edx = original ebp
- mov eax, JDIMENSION [downsamp_width(edx)] ; colctr
- test eax,eax
- jz near .return
-
- mov ecx, INT [max_v_samp(edx)] ; rowctr
- test ecx,ecx
- jz near .return
-
- mov esi, JSAMPARRAY [input_data(edx)] ; input_data
- mov edi, POINTER [output_data_ptr(edx)]
- mov edi, JSAMPARRAY [edi] ; output_data
- alignx 16,7
-.rowloop:
- push eax ; colctr
- push ecx
- push edi
- push esi
-
- mov ecx, JSAMPROW [esi-1*SIZEOF_JSAMPROW] ; inptr1(above)
- mov ebx, JSAMPROW [esi+0*SIZEOF_JSAMPROW] ; inptr0
- mov esi, JSAMPROW [esi+1*SIZEOF_JSAMPROW] ; inptr1(below)
- mov edx, JSAMPROW [edi+0*SIZEOF_JSAMPROW] ; outptr0
- mov edi, JSAMPROW [edi+1*SIZEOF_JSAMPROW] ; outptr1
-
- test eax, SIZEOF_XMMWORD-1
- jz short .skip
- push edx
- mov dl, JSAMPLE [ecx+(eax-1)*SIZEOF_JSAMPLE]
- mov JSAMPLE [ecx+eax*SIZEOF_JSAMPLE], dl
- mov dl, JSAMPLE [ebx+(eax-1)*SIZEOF_JSAMPLE]
- mov JSAMPLE [ebx+eax*SIZEOF_JSAMPLE], dl
- mov dl, JSAMPLE [esi+(eax-1)*SIZEOF_JSAMPLE]
- mov JSAMPLE [esi+eax*SIZEOF_JSAMPLE], dl ; insert a dummy sample
- pop edx
-.skip:
- ; -- process the first column block
-
- movdqa xmm0, XMMWORD [ebx+0*SIZEOF_XMMWORD] ; xmm0=row[ 0][0]
- movdqa xmm1, XMMWORD [ecx+0*SIZEOF_XMMWORD] ; xmm1=row[-1][0]
- movdqa xmm2, XMMWORD [esi+0*SIZEOF_XMMWORD] ; xmm2=row[+1][0]
-
- pushpic ebx
- movpic ebx, POINTER [gotptr] ; load GOT address
-
- pxor xmm3,xmm3 ; xmm3=(all 0's)
- movdqa xmm4,xmm0
- punpcklbw xmm0,xmm3 ; xmm0=row[ 0]( 0 1 2 3 4 5 6 7)
- punpckhbw xmm4,xmm3 ; xmm4=row[ 0]( 8 9 10 11 12 13 14 15)
- movdqa xmm5,xmm1
- punpcklbw xmm1,xmm3 ; xmm1=row[-1]( 0 1 2 3 4 5 6 7)
- punpckhbw xmm5,xmm3 ; xmm5=row[-1]( 8 9 10 11 12 13 14 15)
- movdqa xmm6,xmm2
- punpcklbw xmm2,xmm3 ; xmm2=row[+1]( 0 1 2 3 4 5 6 7)
- punpckhbw xmm6,xmm3 ; xmm6=row[+1]( 8 9 10 11 12 13 14 15)
-
- pmullw xmm0,[GOTOFF(ebx,PW_THREE)]
- pmullw xmm4,[GOTOFF(ebx,PW_THREE)]
-
- pcmpeqb xmm7,xmm7
- psrldq xmm7,(SIZEOF_XMMWORD-2)
-
- paddw xmm1,xmm0 ; xmm1=Int0L=( 0 1 2 3 4 5 6 7)
- paddw xmm5,xmm4 ; xmm5=Int0H=( 8 9 10 11 12 13 14 15)
- paddw xmm2,xmm0 ; xmm2=Int1L=( 0 1 2 3 4 5 6 7)
- paddw xmm6,xmm4 ; xmm6=Int1H=( 8 9 10 11 12 13 14 15)
-
- movdqa XMMWORD [edx+0*SIZEOF_XMMWORD], xmm1 ; temporarily save
- movdqa XMMWORD [edx+1*SIZEOF_XMMWORD], xmm5 ; the intermediate data
- movdqa XMMWORD [edi+0*SIZEOF_XMMWORD], xmm2
- movdqa XMMWORD [edi+1*SIZEOF_XMMWORD], xmm6
-
- pand xmm1,xmm7 ; xmm1=( 0 -- -- -- -- -- -- --)
- pand xmm2,xmm7 ; xmm2=( 0 -- -- -- -- -- -- --)
-
- movdqa XMMWORD [wk(0)], xmm1
- movdqa XMMWORD [wk(1)], xmm2
-
- poppic ebx
-
- add eax, byte SIZEOF_XMMWORD-1
- and eax, byte -SIZEOF_XMMWORD
- cmp eax, byte SIZEOF_XMMWORD
- ja short .columnloop
- alignx 16,7
-
-.columnloop_last:
- ; -- process the last column block
-
- pushpic ebx
- movpic ebx, POINTER [gotptr] ; load GOT address
-
- pcmpeqb xmm1,xmm1
- pslldq xmm1,(SIZEOF_XMMWORD-2)
- movdqa xmm2,xmm1
-
- pand xmm1, XMMWORD [edx+1*SIZEOF_XMMWORD]
- pand xmm2, XMMWORD [edi+1*SIZEOF_XMMWORD]
-
- movdqa XMMWORD [wk(2)], xmm1 ; xmm1=(-- -- -- -- -- -- -- 15)
- movdqa XMMWORD [wk(3)], xmm2 ; xmm2=(-- -- -- -- -- -- -- 15)
-
- jmp near .upsample
- alignx 16,7
-
-.columnloop:
- ; -- process the next column block
-
- movdqa xmm0, XMMWORD [ebx+1*SIZEOF_XMMWORD] ; xmm0=row[ 0][1]
- movdqa xmm1, XMMWORD [ecx+1*SIZEOF_XMMWORD] ; xmm1=row[-1][1]
- movdqa xmm2, XMMWORD [esi+1*SIZEOF_XMMWORD] ; xmm2=row[+1][1]
-
- pushpic ebx
- movpic ebx, POINTER [gotptr] ; load GOT address
-
- pxor xmm3,xmm3 ; xmm3=(all 0's)
- movdqa xmm4,xmm0
- punpcklbw xmm0,xmm3 ; xmm0=row[ 0]( 0 1 2 3 4 5 6 7)
- punpckhbw xmm4,xmm3 ; xmm4=row[ 0]( 8 9 10 11 12 13 14 15)
- movdqa xmm5,xmm1
- punpcklbw xmm1,xmm3 ; xmm1=row[-1]( 0 1 2 3 4 5 6 7)
- punpckhbw xmm5,xmm3 ; xmm5=row[-1]( 8 9 10 11 12 13 14 15)
- movdqa xmm6,xmm2
- punpcklbw xmm2,xmm3 ; xmm2=row[+1]( 0 1 2 3 4 5 6 7)
- punpckhbw xmm6,xmm3 ; xmm6=row[+1]( 8 9 10 11 12 13 14 15)
-
- pmullw xmm0,[GOTOFF(ebx,PW_THREE)]
- pmullw xmm4,[GOTOFF(ebx,PW_THREE)]
-
- paddw xmm1,xmm0 ; xmm1=Int0L=( 0 1 2 3 4 5 6 7)
- paddw xmm5,xmm4 ; xmm5=Int0H=( 8 9 10 11 12 13 14 15)
- paddw xmm2,xmm0 ; xmm2=Int1L=( 0 1 2 3 4 5 6 7)
- paddw xmm6,xmm4 ; xmm6=Int1H=( 8 9 10 11 12 13 14 15)
-
- movdqa XMMWORD [edx+2*SIZEOF_XMMWORD], xmm1 ; temporarily save
- movdqa XMMWORD [edx+3*SIZEOF_XMMWORD], xmm5 ; the intermediate data
- movdqa XMMWORD [edi+2*SIZEOF_XMMWORD], xmm2
- movdqa XMMWORD [edi+3*SIZEOF_XMMWORD], xmm6
-
- pslldq xmm1,(SIZEOF_XMMWORD-2) ; xmm1=(-- -- -- -- -- -- -- 0)
- pslldq xmm2,(SIZEOF_XMMWORD-2) ; xmm2=(-- -- -- -- -- -- -- 0)
-
- movdqa XMMWORD [wk(2)], xmm1
- movdqa XMMWORD [wk(3)], xmm2
-
-.upsample:
- ; -- process the upper row
-
- movdqa xmm7, XMMWORD [edx+0*SIZEOF_XMMWORD]
- movdqa xmm3, XMMWORD [edx+1*SIZEOF_XMMWORD]
-
- movdqa xmm0,xmm7 ; xmm7=Int0L=( 0 1 2 3 4 5 6 7)
- movdqa xmm4,xmm3 ; xmm3=Int0H=( 8 9 10 11 12 13 14 15)
- psrldq xmm0,2 ; xmm0=( 1 2 3 4 5 6 7 --)
- pslldq xmm4,(SIZEOF_XMMWORD-2) ; xmm4=(-- -- -- -- -- -- -- 8)
- movdqa xmm5,xmm7
- movdqa xmm6,xmm3
- psrldq xmm5,(SIZEOF_XMMWORD-2) ; xmm5=( 7 -- -- -- -- -- -- --)
- pslldq xmm6,2 ; xmm6=(-- 8 9 10 11 12 13 14)
-
- por xmm0,xmm4 ; xmm0=( 1 2 3 4 5 6 7 8)
- por xmm5,xmm6 ; xmm5=( 7 8 9 10 11 12 13 14)
-
- movdqa xmm1,xmm7
- movdqa xmm2,xmm3
- pslldq xmm1,2 ; xmm1=(-- 0 1 2 3 4 5 6)
- psrldq xmm2,2 ; xmm2=( 9 10 11 12 13 14 15 --)
- movdqa xmm4,xmm3
- psrldq xmm4,(SIZEOF_XMMWORD-2) ; xmm4=(15 -- -- -- -- -- -- --)
-
- por xmm1, XMMWORD [wk(0)] ; xmm1=(-1 0 1 2 3 4 5 6)
- por xmm2, XMMWORD [wk(2)] ; xmm2=( 9 10 11 12 13 14 15 16)
-
- movdqa XMMWORD [wk(0)], xmm4
-
- pmullw xmm7,[GOTOFF(ebx,PW_THREE)]
- pmullw xmm3,[GOTOFF(ebx,PW_THREE)]
- paddw xmm1,[GOTOFF(ebx,PW_EIGHT)]
- paddw xmm5,[GOTOFF(ebx,PW_EIGHT)]
- paddw xmm0,[GOTOFF(ebx,PW_SEVEN)]
- paddw xmm2,[GOTOFF(ebx,PW_SEVEN)]
-
- paddw xmm1,xmm7
- paddw xmm5,xmm3
- psrlw xmm1,4 ; xmm1=Out0LE=( 0 2 4 6 8 10 12 14)
- psrlw xmm5,4 ; xmm5=Out0HE=(16 18 20 22 24 26 28 30)
- paddw xmm0,xmm7
- paddw xmm2,xmm3
- psrlw xmm0,4 ; xmm0=Out0LO=( 1 3 5 7 9 11 13 15)
- psrlw xmm2,4 ; xmm2=Out0HO=(17 19 21 23 25 27 29 31)
-
- psllw xmm0,BYTE_BIT
- psllw xmm2,BYTE_BIT
- por xmm1,xmm0 ; xmm1=Out0L=( 0 1 2 ... 13 14 15)
- por xmm5,xmm2 ; xmm5=Out0H=(16 17 18 ... 29 30 31)
-
- movdqa XMMWORD [edx+0*SIZEOF_XMMWORD], xmm1
- movdqa XMMWORD [edx+1*SIZEOF_XMMWORD], xmm5
-
- ; -- process the lower row
-
- movdqa xmm6, XMMWORD [edi+0*SIZEOF_XMMWORD]
- movdqa xmm4, XMMWORD [edi+1*SIZEOF_XMMWORD]
-
- movdqa xmm7,xmm6 ; xmm6=Int1L=( 0 1 2 3 4 5 6 7)
- movdqa xmm3,xmm4 ; xmm4=Int1H=( 8 9 10 11 12 13 14 15)
- psrldq xmm7,2 ; xmm7=( 1 2 3 4 5 6 7 --)
- pslldq xmm3,(SIZEOF_XMMWORD-2) ; xmm3=(-- -- -- -- -- -- -- 8)
- movdqa xmm0,xmm6
- movdqa xmm2,xmm4
- psrldq xmm0,(SIZEOF_XMMWORD-2) ; xmm0=( 7 -- -- -- -- -- -- --)
- pslldq xmm2,2 ; xmm2=(-- 8 9 10 11 12 13 14)
-
- por xmm7,xmm3 ; xmm7=( 1 2 3 4 5 6 7 8)
- por xmm0,xmm2 ; xmm0=( 7 8 9 10 11 12 13 14)
-
- movdqa xmm1,xmm6
- movdqa xmm5,xmm4
- pslldq xmm1,2 ; xmm1=(-- 0 1 2 3 4 5 6)
- psrldq xmm5,2 ; xmm5=( 9 10 11 12 13 14 15 --)
- movdqa xmm3,xmm4
- psrldq xmm3,(SIZEOF_XMMWORD-2) ; xmm3=(15 -- -- -- -- -- -- --)
-
- por xmm1, XMMWORD [wk(1)] ; xmm1=(-1 0 1 2 3 4 5 6)
- por xmm5, XMMWORD [wk(3)] ; xmm5=( 9 10 11 12 13 14 15 16)
-
- movdqa XMMWORD [wk(1)], xmm3
-
- pmullw xmm6,[GOTOFF(ebx,PW_THREE)]
- pmullw xmm4,[GOTOFF(ebx,PW_THREE)]
- paddw xmm1,[GOTOFF(ebx,PW_EIGHT)]
- paddw xmm0,[GOTOFF(ebx,PW_EIGHT)]
- paddw xmm7,[GOTOFF(ebx,PW_SEVEN)]
- paddw xmm5,[GOTOFF(ebx,PW_SEVEN)]
-
- paddw xmm1,xmm6
- paddw xmm0,xmm4
- psrlw xmm1,4 ; xmm1=Out1LE=( 0 2 4 6 8 10 12 14)
- psrlw xmm0,4 ; xmm0=Out1HE=(16 18 20 22 24 26 28 30)
- paddw xmm7,xmm6
- paddw xmm5,xmm4
- psrlw xmm7,4 ; xmm7=Out1LO=( 1 3 5 7 9 11 13 15)
- psrlw xmm5,4 ; xmm5=Out1HO=(17 19 21 23 25 27 29 31)
-
- psllw xmm7,BYTE_BIT
- psllw xmm5,BYTE_BIT
- por xmm1,xmm7 ; xmm1=Out1L=( 0 1 2 ... 13 14 15)
- por xmm0,xmm5 ; xmm0=Out1H=(16 17 18 ... 29 30 31)
-
- movdqa XMMWORD [edi+0*SIZEOF_XMMWORD], xmm1
- movdqa XMMWORD [edi+1*SIZEOF_XMMWORD], xmm0
-
- poppic ebx
-
- sub eax, byte SIZEOF_XMMWORD
- add ecx, byte 1*SIZEOF_XMMWORD ; inptr1(above)
- add ebx, byte 1*SIZEOF_XMMWORD ; inptr0
- add esi, byte 1*SIZEOF_XMMWORD ; inptr1(below)
- add edx, byte 2*SIZEOF_XMMWORD ; outptr0
- add edi, byte 2*SIZEOF_XMMWORD ; outptr1
- cmp eax, byte SIZEOF_XMMWORD
- ja near .columnloop
- test eax,eax
- jnz near .columnloop_last
-
- pop esi
- pop edi
- pop ecx
- pop eax
-
- add esi, byte 1*SIZEOF_JSAMPROW ; input_data
- add edi, byte 2*SIZEOF_JSAMPROW ; output_data
- sub ecx, byte 2 ; rowctr
- jg near .rowloop
-
-.return:
- pop edi
- pop esi
-; pop edx ; need not be preserved
-; pop ecx ; need not be preserved
- pop ebx
- mov esp,ebp ; esp <- aligned ebp
- pop esp ; esp <- original ebp
- pop ebp
- ret
-
-; --------------------------------------------------------------------------
-;
-; Fast processing for the common case of 2:1 horizontal and 1:1 vertical.
-; It's still a box filter.
-;
-; GLOBAL(void)
-; jsimd_h2v1_upsample_sse2 (int max_v_samp_factor,
-; JDIMENSION output_width,
-; JSAMPARRAY input_data,
-; JSAMPARRAY *output_data_ptr);
-;
-
-%define max_v_samp(b) (b)+8 ; int max_v_samp_factor
-%define output_width(b) (b)+12 ; JDIMENSION output_width
-%define input_data(b) (b)+16 ; JSAMPARRAY input_data
-%define output_data_ptr(b) (b)+20 ; JSAMPARRAY *output_data_ptr
-
- align 16
- global EXTN(jsimd_h2v1_upsample_sse2)
-
-EXTN(jsimd_h2v1_upsample_sse2):
- push ebp
- mov ebp,esp
-; push ebx ; unused
-; push ecx ; need not be preserved
-; push edx ; need not be preserved
- push esi
- push edi
-
- mov edx, JDIMENSION [output_width(ebp)]
- add edx, byte (2*SIZEOF_XMMWORD)-1
- and edx, byte -(2*SIZEOF_XMMWORD)
- jz short .return
-
- mov ecx, INT [max_v_samp(ebp)] ; rowctr
- test ecx,ecx
- jz short .return
-
- mov esi, JSAMPARRAY [input_data(ebp)] ; input_data
- mov edi, POINTER [output_data_ptr(ebp)]
- mov edi, JSAMPARRAY [edi] ; output_data
- alignx 16,7
-.rowloop:
- push edi
- push esi
-
- mov esi, JSAMPROW [esi] ; inptr
- mov edi, JSAMPROW [edi] ; outptr
- mov eax,edx ; colctr
- alignx 16,7
-.columnloop:
-
- movdqa xmm0, XMMWORD [esi+0*SIZEOF_XMMWORD]
-
- movdqa xmm1,xmm0
- punpcklbw xmm0,xmm0
- punpckhbw xmm1,xmm1
-
- movdqa XMMWORD [edi+0*SIZEOF_XMMWORD], xmm0
- movdqa XMMWORD [edi+1*SIZEOF_XMMWORD], xmm1
-
- sub eax, byte 2*SIZEOF_XMMWORD
- jz short .nextrow
-
- movdqa xmm2, XMMWORD [esi+1*SIZEOF_XMMWORD]
-
- movdqa xmm3,xmm2
- punpcklbw xmm2,xmm2
- punpckhbw xmm3,xmm3
-
- movdqa XMMWORD [edi+2*SIZEOF_XMMWORD], xmm2
- movdqa XMMWORD [edi+3*SIZEOF_XMMWORD], xmm3
-
- sub eax, byte 2*SIZEOF_XMMWORD
- jz short .nextrow
-
- add esi, byte 2*SIZEOF_XMMWORD ; inptr
- add edi, byte 4*SIZEOF_XMMWORD ; outptr
- jmp short .columnloop
- alignx 16,7
-
-.nextrow:
- pop esi
- pop edi
-
- add esi, byte SIZEOF_JSAMPROW ; input_data
- add edi, byte SIZEOF_JSAMPROW ; output_data
- dec ecx ; rowctr
- jg short .rowloop
-
-.return:
- pop edi
- pop esi
-; pop edx ; need not be preserved
-; pop ecx ; need not be preserved
-; pop ebx ; unused
- pop ebp
- ret
-
-; --------------------------------------------------------------------------
-;
-; Fast processing for the common case of 2:1 horizontal and 2:1 vertical.
-; It's still a box filter.
-;
-; GLOBAL(void)
-; jsimd_h2v2_upsample_sse2 (nt max_v_samp_factor,
-; JDIMENSION output_width,
-; JSAMPARRAY input_data,
-; JSAMPARRAY *output_data_ptr);
-;
-
-%define max_v_samp(b) (b)+8 ; int max_v_samp_factor
-%define output_width(b) (b)+12 ; JDIMENSION output_width
-%define input_data(b) (b)+16 ; JSAMPARRAY input_data
-%define output_data_ptr(b) (b)+20 ; JSAMPARRAY *output_data_ptr
-
- align 16
- global EXTN(jsimd_h2v2_upsample_sse2)
-
-EXTN(jsimd_h2v2_upsample_sse2):
- push ebp
- mov ebp,esp
- push ebx
-; push ecx ; need not be preserved
-; push edx ; need not be preserved
- push esi
- push edi
-
- mov edx, JDIMENSION [output_width(ebp)]
- add edx, byte (2*SIZEOF_XMMWORD)-1
- and edx, byte -(2*SIZEOF_XMMWORD)
- jz near .return
-
- mov ecx, INT [max_v_samp(ebp)] ; rowctr
- test ecx,ecx
- jz near .return
-
- mov esi, JSAMPARRAY [input_data(ebp)] ; input_data
- mov edi, POINTER [output_data_ptr(ebp)]
- mov edi, JSAMPARRAY [edi] ; output_data
- alignx 16,7
-.rowloop:
- push edi
- push esi
-
- mov esi, JSAMPROW [esi] ; inptr
- mov ebx, JSAMPROW [edi+0*SIZEOF_JSAMPROW] ; outptr0
- mov edi, JSAMPROW [edi+1*SIZEOF_JSAMPROW] ; outptr1
- mov eax,edx ; colctr
- alignx 16,7
-.columnloop:
-
- movdqa xmm0, XMMWORD [esi+0*SIZEOF_XMMWORD]
-
- movdqa xmm1,xmm0
- punpcklbw xmm0,xmm0
- punpckhbw xmm1,xmm1
-
- movdqa XMMWORD [ebx+0*SIZEOF_XMMWORD], xmm0
- movdqa XMMWORD [ebx+1*SIZEOF_XMMWORD], xmm1
- movdqa XMMWORD [edi+0*SIZEOF_XMMWORD], xmm0
- movdqa XMMWORD [edi+1*SIZEOF_XMMWORD], xmm1
-
- sub eax, byte 2*SIZEOF_XMMWORD
- jz short .nextrow
-
- movdqa xmm2, XMMWORD [esi+1*SIZEOF_XMMWORD]
-
- movdqa xmm3,xmm2
- punpcklbw xmm2,xmm2
- punpckhbw xmm3,xmm3
-
- movdqa XMMWORD [ebx+2*SIZEOF_XMMWORD], xmm2
- movdqa XMMWORD [ebx+3*SIZEOF_XMMWORD], xmm3
- movdqa XMMWORD [edi+2*SIZEOF_XMMWORD], xmm2
- movdqa XMMWORD [edi+3*SIZEOF_XMMWORD], xmm3
-
- sub eax, byte 2*SIZEOF_XMMWORD
- jz short .nextrow
-
- add esi, byte 2*SIZEOF_XMMWORD ; inptr
- add ebx, byte 4*SIZEOF_XMMWORD ; outptr0
- add edi, byte 4*SIZEOF_XMMWORD ; outptr1
- jmp short .columnloop
- alignx 16,7
-
-.nextrow:
- pop esi
- pop edi
-
- add esi, byte 1*SIZEOF_JSAMPROW ; input_data
- add edi, byte 2*SIZEOF_JSAMPROW ; output_data
- sub ecx, byte 2 ; rowctr
- jg short .rowloop
-
-.return:
- pop edi
- pop esi
-; pop edx ; need not be preserved
-; pop ecx ; need not be preserved
- pop ebx
- pop ebp
- ret
-
-; For some reason, the OS X linker does not honor the request to align the
-; segment unless we do this.
- align 16
diff --git a/simd/jfdctflt-sse-64.asm b/simd/jfdctflt-sse-64.asm
deleted file mode 100644
index 4b64ea4..0000000
--- a/simd/jfdctflt-sse-64.asm
+++ /dev/null
@@ -1,357 +0,0 @@
-;
-; jfdctflt.asm - floating-point FDCT (64-bit SSE)
-;
-; Copyright 2009 Pierre Ossman <ossman@cendio.se> for Cendio AB
-; Copyright (C) 2009, D. R. Commander.
-;
-; Based on the x86 SIMD extension for IJG JPEG library
-; Copyright (C) 1999-2006, MIYASAKA Masaru.
-; For conditions of distribution and use, see copyright notice in jsimdext.inc
-;
-; This file should be assembled with NASM (Netwide Assembler),
-; can *not* be assembled with Microsoft's MASM or any compatible
-; assembler (including Borland's Turbo Assembler).
-; NASM is available from http://nasm.sourceforge.net/ or
-; http://sourceforge.net/project/showfiles.php?group_id=6208
-;
-; This file contains a floating-point implementation of the forward DCT
-; (Discrete Cosine Transform). The following code is based directly on
-; the IJG's original jfdctflt.c; see the jfdctflt.c for more details.
-;
-; [TAB8]
-
-%include "jsimdext.inc"
-%include "jdct.inc"
-
-; --------------------------------------------------------------------------
-
-%macro unpcklps2 2 ; %1=(0 1 2 3) / %2=(4 5 6 7) => %1=(0 1 4 5)
- shufps %1,%2,0x44
-%endmacro
-
-%macro unpckhps2 2 ; %1=(0 1 2 3) / %2=(4 5 6 7) => %1=(2 3 6 7)
- shufps %1,%2,0xEE
-%endmacro
-
-; --------------------------------------------------------------------------
- SECTION SEG_CONST
-
- alignz 16
- global EXTN(jconst_fdct_float_sse)
-
-EXTN(jconst_fdct_float_sse):
-
-PD_0_382 times 4 dd 0.382683432365089771728460
-PD_0_707 times 4 dd 0.707106781186547524400844
-PD_0_541 times 4 dd 0.541196100146196984399723
-PD_1_306 times 4 dd 1.306562964876376527856643
-
- alignz 16
-
-; --------------------------------------------------------------------------
- SECTION SEG_TEXT
- BITS 64
-;
-; Perform the forward DCT on one block of samples.
-;
-; GLOBAL(void)
-; jsimd_fdct_float_sse (FAST_FLOAT *data)
-;
-
-; r10 = FAST_FLOAT *data
-
-%define wk(i) rbp-(WK_NUM-(i))*SIZEOF_XMMWORD ; xmmword wk[WK_NUM]
-%define WK_NUM 2
-
- align 16
- global EXTN(jsimd_fdct_float_sse)
-
-EXTN(jsimd_fdct_float_sse):
- push rbp
- mov rax,rsp ; rax = original rbp
- sub rsp, byte 4
- and rsp, byte (-SIZEOF_XMMWORD) ; align to 128 bits
- mov [rsp],rax
- mov rbp,rsp ; rbp = aligned rbp
- lea rsp, [wk(0)]
- collect_args
-
- ; ---- Pass 1: process rows.
-
- mov rdx, r10 ; (FAST_FLOAT *)
- mov rcx, DCTSIZE/4
-.rowloop:
-
- movaps xmm0, XMMWORD [XMMBLOCK(2,0,rdx,SIZEOF_FAST_FLOAT)]
- movaps xmm1, XMMWORD [XMMBLOCK(3,0,rdx,SIZEOF_FAST_FLOAT)]
- movaps xmm2, XMMWORD [XMMBLOCK(2,1,rdx,SIZEOF_FAST_FLOAT)]
- movaps xmm3, XMMWORD [XMMBLOCK(3,1,rdx,SIZEOF_FAST_FLOAT)]
-
- ; xmm0=(20 21 22 23), xmm2=(24 25 26 27)
- ; xmm1=(30 31 32 33), xmm3=(34 35 36 37)
-
- movaps xmm4,xmm0 ; transpose coefficients(phase 1)
- unpcklps xmm0,xmm1 ; xmm0=(20 30 21 31)
- unpckhps xmm4,xmm1 ; xmm4=(22 32 23 33)
- movaps xmm5,xmm2 ; transpose coefficients(phase 1)
- unpcklps xmm2,xmm3 ; xmm2=(24 34 25 35)
- unpckhps xmm5,xmm3 ; xmm5=(26 36 27 37)
-
- movaps xmm6, XMMWORD [XMMBLOCK(0,0,rdx,SIZEOF_FAST_FLOAT)]
- movaps xmm7, XMMWORD [XMMBLOCK(1,0,rdx,SIZEOF_FAST_FLOAT)]
- movaps xmm1, XMMWORD [XMMBLOCK(0,1,rdx,SIZEOF_FAST_FLOAT)]
- movaps xmm3, XMMWORD [XMMBLOCK(1,1,rdx,SIZEOF_FAST_FLOAT)]
-
- ; xmm6=(00 01 02 03), xmm1=(04 05 06 07)
- ; xmm7=(10 11 12 13), xmm3=(14 15 16 17)
-
- movaps XMMWORD [wk(0)], xmm4 ; wk(0)=(22 32 23 33)
- movaps XMMWORD [wk(1)], xmm2 ; wk(1)=(24 34 25 35)
-
- movaps xmm4,xmm6 ; transpose coefficients(phase 1)
- unpcklps xmm6,xmm7 ; xmm6=(00 10 01 11)
- unpckhps xmm4,xmm7 ; xmm4=(02 12 03 13)
- movaps xmm2,xmm1 ; transpose coefficients(phase 1)
- unpcklps xmm1,xmm3 ; xmm1=(04 14 05 15)
- unpckhps xmm2,xmm3 ; xmm2=(06 16 07 17)
-
- movaps xmm7,xmm6 ; transpose coefficients(phase 2)
- unpcklps2 xmm6,xmm0 ; xmm6=(00 10 20 30)=data0
- unpckhps2 xmm7,xmm0 ; xmm7=(01 11 21 31)=data1
- movaps xmm3,xmm2 ; transpose coefficients(phase 2)
- unpcklps2 xmm2,xmm5 ; xmm2=(06 16 26 36)=data6
- unpckhps2 xmm3,xmm5 ; xmm3=(07 17 27 37)=data7
-
- movaps xmm0,xmm7
- movaps xmm5,xmm6
- subps xmm7,xmm2 ; xmm7=data1-data6=tmp6
- subps xmm6,xmm3 ; xmm6=data0-data7=tmp7
- addps xmm0,xmm2 ; xmm0=data1+data6=tmp1
- addps xmm5,xmm3 ; xmm5=data0+data7=tmp0
-
- movaps xmm2, XMMWORD [wk(0)] ; xmm2=(22 32 23 33)
- movaps xmm3, XMMWORD [wk(1)] ; xmm3=(24 34 25 35)
- movaps XMMWORD [wk(0)], xmm7 ; wk(0)=tmp6
- movaps XMMWORD [wk(1)], xmm6 ; wk(1)=tmp7
-
- movaps xmm7,xmm4 ; transpose coefficients(phase 2)
- unpcklps2 xmm4,xmm2 ; xmm4=(02 12 22 32)=data2
- unpckhps2 xmm7,xmm2 ; xmm7=(03 13 23 33)=data3
- movaps xmm6,xmm1 ; transpose coefficients(phase 2)
- unpcklps2 xmm1,xmm3 ; xmm1=(04 14 24 34)=data4
- unpckhps2 xmm6,xmm3 ; xmm6=(05 15 25 35)=data5
-
- movaps xmm2,xmm7
- movaps xmm3,xmm4
- addps xmm7,xmm1 ; xmm7=data3+data4=tmp3
- addps xmm4,xmm6 ; xmm4=data2+data5=tmp2
- subps xmm2,xmm1 ; xmm2=data3-data4=tmp4
- subps xmm3,xmm6 ; xmm3=data2-data5=tmp5
-
- ; -- Even part
-
- movaps xmm1,xmm5
- movaps xmm6,xmm0
- subps xmm5,xmm7 ; xmm5=tmp13
- subps xmm0,xmm4 ; xmm0=tmp12
- addps xmm1,xmm7 ; xmm1=tmp10
- addps xmm6,xmm4 ; xmm6=tmp11
-
- addps xmm0,xmm5
- mulps xmm0,[rel PD_0_707] ; xmm0=z1
-
- movaps xmm7,xmm1
- movaps xmm4,xmm5
- subps xmm1,xmm6 ; xmm1=data4
- subps xmm5,xmm0 ; xmm5=data6
- addps xmm7,xmm6 ; xmm7=data0
- addps xmm4,xmm0 ; xmm4=data2
-
- movaps XMMWORD [XMMBLOCK(0,1,rdx,SIZEOF_FAST_FLOAT)], xmm1
- movaps XMMWORD [XMMBLOCK(2,1,rdx,SIZEOF_FAST_FLOAT)], xmm5
- movaps XMMWORD [XMMBLOCK(0,0,rdx,SIZEOF_FAST_FLOAT)], xmm7
- movaps XMMWORD [XMMBLOCK(2,0,rdx,SIZEOF_FAST_FLOAT)], xmm4
-
- ; -- Odd part
-
- movaps xmm6, XMMWORD [wk(0)] ; xmm6=tmp6
- movaps xmm0, XMMWORD [wk(1)] ; xmm0=tmp7
-
- addps xmm2,xmm3 ; xmm2=tmp10
- addps xmm3,xmm6 ; xmm3=tmp11
- addps xmm6,xmm0 ; xmm6=tmp12, xmm0=tmp7
-
- mulps xmm3,[rel PD_0_707] ; xmm3=z3
-
- movaps xmm1,xmm2 ; xmm1=tmp10
- subps xmm2,xmm6
- mulps xmm2,[rel PD_0_382] ; xmm2=z5
- mulps xmm1,[rel PD_0_541] ; xmm1=MULTIPLY(tmp10,FIX_0_541196)
- mulps xmm6,[rel PD_1_306] ; xmm6=MULTIPLY(tmp12,FIX_1_306562)
- addps xmm1,xmm2 ; xmm1=z2
- addps xmm6,xmm2 ; xmm6=z4
-
- movaps xmm5,xmm0
- subps xmm0,xmm3 ; xmm0=z13
- addps xmm5,xmm3 ; xmm5=z11
-
- movaps xmm7,xmm0
- movaps xmm4,xmm5
- subps xmm0,xmm1 ; xmm0=data3
- subps xmm5,xmm6 ; xmm5=data7
- addps xmm7,xmm1 ; xmm7=data5
- addps xmm4,xmm6 ; xmm4=data1
-
- movaps XMMWORD [XMMBLOCK(3,0,rdx,SIZEOF_FAST_FLOAT)], xmm0
- movaps XMMWORD [XMMBLOCK(3,1,rdx,SIZEOF_FAST_FLOAT)], xmm5
- movaps XMMWORD [XMMBLOCK(1,1,rdx,SIZEOF_FAST_FLOAT)], xmm7
- movaps XMMWORD [XMMBLOCK(1,0,rdx,SIZEOF_FAST_FLOAT)], xmm4
-
- add rdx, 4*DCTSIZE*SIZEOF_FAST_FLOAT
- dec rcx
- jnz near .rowloop
-
- ; ---- Pass 2: process columns.
-
- mov rdx, r10 ; (FAST_FLOAT *)
- mov rcx, DCTSIZE/4
-.columnloop:
-
- movaps xmm0, XMMWORD [XMMBLOCK(2,0,rdx,SIZEOF_FAST_FLOAT)]
- movaps xmm1, XMMWORD [XMMBLOCK(3,0,rdx,SIZEOF_FAST_FLOAT)]
- movaps xmm2, XMMWORD [XMMBLOCK(6,0,rdx,SIZEOF_FAST_FLOAT)]
- movaps xmm3, XMMWORD [XMMBLOCK(7,0,rdx,SIZEOF_FAST_FLOAT)]
-
- ; xmm0=(02 12 22 32), xmm2=(42 52 62 72)
- ; xmm1=(03 13 23 33), xmm3=(43 53 63 73)
-
- movaps xmm4,xmm0 ; transpose coefficients(phase 1)
- unpcklps xmm0,xmm1 ; xmm0=(02 03 12 13)
- unpckhps xmm4,xmm1 ; xmm4=(22 23 32 33)
- movaps xmm5,xmm2 ; transpose coefficients(phase 1)
- unpcklps xmm2,xmm3 ; xmm2=(42 43 52 53)
- unpckhps xmm5,xmm3 ; xmm5=(62 63 72 73)
-
- movaps xmm6, XMMWORD [XMMBLOCK(0,0,rdx,SIZEOF_FAST_FLOAT)]
- movaps xmm7, XMMWORD [XMMBLOCK(1,0,rdx,SIZEOF_FAST_FLOAT)]
- movaps xmm1, XMMWORD [XMMBLOCK(4,0,rdx,SIZEOF_FAST_FLOAT)]
- movaps xmm3, XMMWORD [XMMBLOCK(5,0,rdx,SIZEOF_FAST_FLOAT)]
-
- ; xmm6=(00 10 20 30), xmm1=(40 50 60 70)
- ; xmm7=(01 11 21 31), xmm3=(41 51 61 71)
-
- movaps XMMWORD [wk(0)], xmm4 ; wk(0)=(22 23 32 33)
- movaps XMMWORD [wk(1)], xmm2 ; wk(1)=(42 43 52 53)
-
- movaps xmm4,xmm6 ; transpose coefficients(phase 1)
- unpcklps xmm6,xmm7 ; xmm6=(00 01 10 11)
- unpckhps xmm4,xmm7 ; xmm4=(20 21 30 31)
- movaps xmm2,xmm1 ; transpose coefficients(phase 1)
- unpcklps xmm1,xmm3 ; xmm1=(40 41 50 51)
- unpckhps xmm2,xmm3 ; xmm2=(60 61 70 71)
-
- movaps xmm7,xmm6 ; transpose coefficients(phase 2)
- unpcklps2 xmm6,xmm0 ; xmm6=(00 01 02 03)=data0
- unpckhps2 xmm7,xmm0 ; xmm7=(10 11 12 13)=data1
- movaps xmm3,xmm2 ; transpose coefficients(phase 2)
- unpcklps2 xmm2,xmm5 ; xmm2=(60 61 62 63)=data6
- unpckhps2 xmm3,xmm5 ; xmm3=(70 71 72 73)=data7
-
- movaps xmm0,xmm7
- movaps xmm5,xmm6
- subps xmm7,xmm2 ; xmm7=data1-data6=tmp6
- subps xmm6,xmm3 ; xmm6=data0-data7=tmp7
- addps xmm0,xmm2 ; xmm0=data1+data6=tmp1
- addps xmm5,xmm3 ; xmm5=data0+data7=tmp0
-
- movaps xmm2, XMMWORD [wk(0)] ; xmm2=(22 23 32 33)
- movaps xmm3, XMMWORD [wk(1)] ; xmm3=(42 43 52 53)
- movaps XMMWORD [wk(0)], xmm7 ; wk(0)=tmp6
- movaps XMMWORD [wk(1)], xmm6 ; wk(1)=tmp7
-
- movaps xmm7,xmm4 ; transpose coefficients(phase 2)
- unpcklps2 xmm4,xmm2 ; xmm4=(20 21 22 23)=data2
- unpckhps2 xmm7,xmm2 ; xmm7=(30 31 32 33)=data3
- movaps xmm6,xmm1 ; transpose coefficients(phase 2)
- unpcklps2 xmm1,xmm3 ; xmm1=(40 41 42 43)=data4
- unpckhps2 xmm6,xmm3 ; xmm6=(50 51 52 53)=data5
-
- movaps xmm2,xmm7
- movaps xmm3,xmm4
- addps xmm7,xmm1 ; xmm7=data3+data4=tmp3
- addps xmm4,xmm6 ; xmm4=data2+data5=tmp2
- subps xmm2,xmm1 ; xmm2=data3-data4=tmp4
- subps xmm3,xmm6 ; xmm3=data2-data5=tmp5
-
- ; -- Even part
-
- movaps xmm1,xmm5
- movaps xmm6,xmm0
- subps xmm5,xmm7 ; xmm5=tmp13
- subps xmm0,xmm4 ; xmm0=tmp12
- addps xmm1,xmm7 ; xmm1=tmp10
- addps xmm6,xmm4 ; xmm6=tmp11
-
- addps xmm0,xmm5
- mulps xmm0,[rel PD_0_707] ; xmm0=z1
-
- movaps xmm7,xmm1
- movaps xmm4,xmm5
- subps xmm1,xmm6 ; xmm1=data4
- subps xmm5,xmm0 ; xmm5=data6
- addps xmm7,xmm6 ; xmm7=data0
- addps xmm4,xmm0 ; xmm4=data2
-
- movaps XMMWORD [XMMBLOCK(4,0,rdx,SIZEOF_FAST_FLOAT)], xmm1
- movaps XMMWORD [XMMBLOCK(6,0,rdx,SIZEOF_FAST_FLOAT)], xmm5
- movaps XMMWORD [XMMBLOCK(0,0,rdx,SIZEOF_FAST_FLOAT)], xmm7
- movaps XMMWORD [XMMBLOCK(2,0,rdx,SIZEOF_FAST_FLOAT)], xmm4
-
- ; -- Odd part
-
- movaps xmm6, XMMWORD [wk(0)] ; xmm6=tmp6
- movaps xmm0, XMMWORD [wk(1)] ; xmm0=tmp7
-
- addps xmm2,xmm3 ; xmm2=tmp10
- addps xmm3,xmm6 ; xmm3=tmp11
- addps xmm6,xmm0 ; xmm6=tmp12, xmm0=tmp7
-
- mulps xmm3,[rel PD_0_707] ; xmm3=z3
-
- movaps xmm1,xmm2 ; xmm1=tmp10
- subps xmm2,xmm6
- mulps xmm2,[rel PD_0_382] ; xmm2=z5
- mulps xmm1,[rel PD_0_541] ; xmm1=MULTIPLY(tmp10,FIX_0_541196)
- mulps xmm6,[rel PD_1_306] ; xmm6=MULTIPLY(tmp12,FIX_1_306562)
- addps xmm1,xmm2 ; xmm1=z2
- addps xmm6,xmm2 ; xmm6=z4
-
- movaps xmm5,xmm0
- subps xmm0,xmm3 ; xmm0=z13
- addps xmm5,xmm3 ; xmm5=z11
-
- movaps xmm7,xmm0
- movaps xmm4,xmm5
- subps xmm0,xmm1 ; xmm0=data3
- subps xmm5,xmm6 ; xmm5=data7
- addps xmm7,xmm1 ; xmm7=data5
- addps xmm4,xmm6 ; xmm4=data1
-
- movaps XMMWORD [XMMBLOCK(3,0,rdx,SIZEOF_FAST_FLOAT)], xmm0
- movaps XMMWORD [XMMBLOCK(7,0,rdx,SIZEOF_FAST_FLOAT)], xmm5
- movaps XMMWORD [XMMBLOCK(5,0,rdx,SIZEOF_FAST_FLOAT)], xmm7
- movaps XMMWORD [XMMBLOCK(1,0,rdx,SIZEOF_FAST_FLOAT)], xmm4
-
- add rdx, byte 4*SIZEOF_FAST_FLOAT
- dec rcx
- jnz near .columnloop
-
- uncollect_args
- mov rsp,rbp ; rsp <- aligned rbp
- pop rsp ; rsp <- original rbp
- pop rbp
- ret
-
-; For some reason, the OS X linker does not honor the request to align the
-; segment unless we do this.
- align 16
diff --git a/simd/jfdctflt-sse.asm b/simd/jfdctflt-sse.asm
deleted file mode 100644
index e7ede26..0000000
--- a/simd/jfdctflt-sse.asm
+++ /dev/null
@@ -1,369 +0,0 @@
-;
-; jfdctflt.asm - floating-point FDCT (SSE)
-;
-; Copyright 2009 Pierre Ossman <ossman@cendio.se> for Cendio AB
-;
-; Based on the x86 SIMD extension for IJG JPEG library
-; Copyright (C) 1999-2006, MIYASAKA Masaru.
-; For conditions of distribution and use, see copyright notice in jsimdext.inc
-;
-; This file should be assembled with NASM (Netwide Assembler),
-; can *not* be assembled with Microsoft's MASM or any compatible
-; assembler (including Borland's Turbo Assembler).
-; NASM is available from http://nasm.sourceforge.net/ or
-; http://sourceforge.net/project/showfiles.php?group_id=6208
-;
-; This file contains a floating-point implementation of the forward DCT
-; (Discrete Cosine Transform). The following code is based directly on
-; the IJG's original jfdctflt.c; see the jfdctflt.c for more details.
-;
-; [TAB8]
-
-%include "jsimdext.inc"
-%include "jdct.inc"
-
-; --------------------------------------------------------------------------
-
-%macro unpcklps2 2 ; %1=(0 1 2 3) / %2=(4 5 6 7) => %1=(0 1 4 5)
- shufps %1,%2,0x44
-%endmacro
-
-%macro unpckhps2 2 ; %1=(0 1 2 3) / %2=(4 5 6 7) => %1=(2 3 6 7)
- shufps %1,%2,0xEE
-%endmacro
-
-; --------------------------------------------------------------------------
- SECTION SEG_CONST
-
- alignz 16
- global EXTN(jconst_fdct_float_sse)
-
-EXTN(jconst_fdct_float_sse):
-
-PD_0_382 times 4 dd 0.382683432365089771728460
-PD_0_707 times 4 dd 0.707106781186547524400844
-PD_0_541 times 4 dd 0.541196100146196984399723
-PD_1_306 times 4 dd 1.306562964876376527856643
-
- alignz 16
-
-; --------------------------------------------------------------------------
- SECTION SEG_TEXT
- BITS 32
-;
-; Perform the forward DCT on one block of samples.
-;
-; GLOBAL(void)
-; jsimd_fdct_float_sse (FAST_FLOAT *data)
-;
-
-%define data(b) (b)+8 ; FAST_FLOAT *data
-
-%define original_ebp ebp+0
-%define wk(i) ebp-(WK_NUM-(i))*SIZEOF_XMMWORD ; xmmword wk[WK_NUM]
-%define WK_NUM 2
-
- align 16
- global EXTN(jsimd_fdct_float_sse)
-
-EXTN(jsimd_fdct_float_sse):
- push ebp
- mov eax,esp ; eax = original ebp
- sub esp, byte 4
- and esp, byte (-SIZEOF_XMMWORD) ; align to 128 bits
- mov [esp],eax
- mov ebp,esp ; ebp = aligned ebp
- lea esp, [wk(0)]
- pushpic ebx
-; push ecx ; need not be preserved
-; push edx ; need not be preserved
-; push esi ; unused
-; push edi ; unused
-
- get_GOT ebx ; get GOT address
-
- ; ---- Pass 1: process rows.
-
- mov edx, POINTER [data(eax)] ; (FAST_FLOAT *)
- mov ecx, DCTSIZE/4
- alignx 16,7
-.rowloop:
-
- movaps xmm0, XMMWORD [XMMBLOCK(2,0,edx,SIZEOF_FAST_FLOAT)]
- movaps xmm1, XMMWORD [XMMBLOCK(3,0,edx,SIZEOF_FAST_FLOAT)]
- movaps xmm2, XMMWORD [XMMBLOCK(2,1,edx,SIZEOF_FAST_FLOAT)]
- movaps xmm3, XMMWORD [XMMBLOCK(3,1,edx,SIZEOF_FAST_FLOAT)]
-
- ; xmm0=(20 21 22 23), xmm2=(24 25 26 27)
- ; xmm1=(30 31 32 33), xmm3=(34 35 36 37)
-
- movaps xmm4,xmm0 ; transpose coefficients(phase 1)
- unpcklps xmm0,xmm1 ; xmm0=(20 30 21 31)
- unpckhps xmm4,xmm1 ; xmm4=(22 32 23 33)
- movaps xmm5,xmm2 ; transpose coefficients(phase 1)
- unpcklps xmm2,xmm3 ; xmm2=(24 34 25 35)
- unpckhps xmm5,xmm3 ; xmm5=(26 36 27 37)
-
- movaps xmm6, XMMWORD [XMMBLOCK(0,0,edx,SIZEOF_FAST_FLOAT)]
- movaps xmm7, XMMWORD [XMMBLOCK(1,0,edx,SIZEOF_FAST_FLOAT)]
- movaps xmm1, XMMWORD [XMMBLOCK(0,1,edx,SIZEOF_FAST_FLOAT)]
- movaps xmm3, XMMWORD [XMMBLOCK(1,1,edx,SIZEOF_FAST_FLOAT)]
-
- ; xmm6=(00 01 02 03), xmm1=(04 05 06 07)
- ; xmm7=(10 11 12 13), xmm3=(14 15 16 17)
-
- movaps XMMWORD [wk(0)], xmm4 ; wk(0)=(22 32 23 33)
- movaps XMMWORD [wk(1)], xmm2 ; wk(1)=(24 34 25 35)
-
- movaps xmm4,xmm6 ; transpose coefficients(phase 1)
- unpcklps xmm6,xmm7 ; xmm6=(00 10 01 11)
- unpckhps xmm4,xmm7 ; xmm4=(02 12 03 13)
- movaps xmm2,xmm1 ; transpose coefficients(phase 1)
- unpcklps xmm1,xmm3 ; xmm1=(04 14 05 15)
- unpckhps xmm2,xmm3 ; xmm2=(06 16 07 17)
-
- movaps xmm7,xmm6 ; transpose coefficients(phase 2)
- unpcklps2 xmm6,xmm0 ; xmm6=(00 10 20 30)=data0
- unpckhps2 xmm7,xmm0 ; xmm7=(01 11 21 31)=data1
- movaps xmm3,xmm2 ; transpose coefficients(phase 2)
- unpcklps2 xmm2,xmm5 ; xmm2=(06 16 26 36)=data6
- unpckhps2 xmm3,xmm5 ; xmm3=(07 17 27 37)=data7
-
- movaps xmm0,xmm7
- movaps xmm5,xmm6
- subps xmm7,xmm2 ; xmm7=data1-data6=tmp6
- subps xmm6,xmm3 ; xmm6=data0-data7=tmp7
- addps xmm0,xmm2 ; xmm0=data1+data6=tmp1
- addps xmm5,xmm3 ; xmm5=data0+data7=tmp0
-
- movaps xmm2, XMMWORD [wk(0)] ; xmm2=(22 32 23 33)
- movaps xmm3, XMMWORD [wk(1)] ; xmm3=(24 34 25 35)
- movaps XMMWORD [wk(0)], xmm7 ; wk(0)=tmp6
- movaps XMMWORD [wk(1)], xmm6 ; wk(1)=tmp7
-
- movaps xmm7,xmm4 ; transpose coefficients(phase 2)
- unpcklps2 xmm4,xmm2 ; xmm4=(02 12 22 32)=data2
- unpckhps2 xmm7,xmm2 ; xmm7=(03 13 23 33)=data3
- movaps xmm6,xmm1 ; transpose coefficients(phase 2)
- unpcklps2 xmm1,xmm3 ; xmm1=(04 14 24 34)=data4
- unpckhps2 xmm6,xmm3 ; xmm6=(05 15 25 35)=data5
-
- movaps xmm2,xmm7
- movaps xmm3,xmm4
- addps xmm7,xmm1 ; xmm7=data3+data4=tmp3
- addps xmm4,xmm6 ; xmm4=data2+data5=tmp2
- subps xmm2,xmm1 ; xmm2=data3-data4=tmp4
- subps xmm3,xmm6 ; xmm3=data2-data5=tmp5
-
- ; -- Even part
-
- movaps xmm1,xmm5
- movaps xmm6,xmm0
- subps xmm5,xmm7 ; xmm5=tmp13
- subps xmm0,xmm4 ; xmm0=tmp12
- addps xmm1,xmm7 ; xmm1=tmp10
- addps xmm6,xmm4 ; xmm6=tmp11
-
- addps xmm0,xmm5
- mulps xmm0,[GOTOFF(ebx,PD_0_707)] ; xmm0=z1
-
- movaps xmm7,xmm1
- movaps xmm4,xmm5
- subps xmm1,xmm6 ; xmm1=data4
- subps xmm5,xmm0 ; xmm5=data6
- addps xmm7,xmm6 ; xmm7=data0
- addps xmm4,xmm0 ; xmm4=data2
-
- movaps XMMWORD [XMMBLOCK(0,1,edx,SIZEOF_FAST_FLOAT)], xmm1
- movaps XMMWORD [XMMBLOCK(2,1,edx,SIZEOF_FAST_FLOAT)], xmm5
- movaps XMMWORD [XMMBLOCK(0,0,edx,SIZEOF_FAST_FLOAT)], xmm7
- movaps XMMWORD [XMMBLOCK(2,0,edx,SIZEOF_FAST_FLOAT)], xmm4
-
- ; -- Odd part
-
- movaps xmm6, XMMWORD [wk(0)] ; xmm6=tmp6
- movaps xmm0, XMMWORD [wk(1)] ; xmm0=tmp7
-
- addps xmm2,xmm3 ; xmm2=tmp10
- addps xmm3,xmm6 ; xmm3=tmp11
- addps xmm6,xmm0 ; xmm6=tmp12, xmm0=tmp7
-
- mulps xmm3,[GOTOFF(ebx,PD_0_707)] ; xmm3=z3
-
- movaps xmm1,xmm2 ; xmm1=tmp10
- subps xmm2,xmm6
- mulps xmm2,[GOTOFF(ebx,PD_0_382)] ; xmm2=z5
- mulps xmm1,[GOTOFF(ebx,PD_0_541)] ; xmm1=MULTIPLY(tmp10,FIX_0_541196)
- mulps xmm6,[GOTOFF(ebx,PD_1_306)] ; xmm6=MULTIPLY(tmp12,FIX_1_306562)
- addps xmm1,xmm2 ; xmm1=z2
- addps xmm6,xmm2 ; xmm6=z4
-
- movaps xmm5,xmm0
- subps xmm0,xmm3 ; xmm0=z13
- addps xmm5,xmm3 ; xmm5=z11
-
- movaps xmm7,xmm0
- movaps xmm4,xmm5
- subps xmm0,xmm1 ; xmm0=data3
- subps xmm5,xmm6 ; xmm5=data7
- addps xmm7,xmm1 ; xmm7=data5
- addps xmm4,xmm6 ; xmm4=data1
-
- movaps XMMWORD [XMMBLOCK(3,0,edx,SIZEOF_FAST_FLOAT)], xmm0
- movaps XMMWORD [XMMBLOCK(3,1,edx,SIZEOF_FAST_FLOAT)], xmm5
- movaps XMMWORD [XMMBLOCK(1,1,edx,SIZEOF_FAST_FLOAT)], xmm7
- movaps XMMWORD [XMMBLOCK(1,0,edx,SIZEOF_FAST_FLOAT)], xmm4
-
- add edx, 4*DCTSIZE*SIZEOF_FAST_FLOAT
- dec ecx
- jnz near .rowloop
-
- ; ---- Pass 2: process columns.
-
- mov edx, POINTER [data(eax)] ; (FAST_FLOAT *)
- mov ecx, DCTSIZE/4
- alignx 16,7
-.columnloop:
-
- movaps xmm0, XMMWORD [XMMBLOCK(2,0,edx,SIZEOF_FAST_FLOAT)]
- movaps xmm1, XMMWORD [XMMBLOCK(3,0,edx,SIZEOF_FAST_FLOAT)]
- movaps xmm2, XMMWORD [XMMBLOCK(6,0,edx,SIZEOF_FAST_FLOAT)]
- movaps xmm3, XMMWORD [XMMBLOCK(7,0,edx,SIZEOF_FAST_FLOAT)]
-
- ; xmm0=(02 12 22 32), xmm2=(42 52 62 72)
- ; xmm1=(03 13 23 33), xmm3=(43 53 63 73)
-
- movaps xmm4,xmm0 ; transpose coefficients(phase 1)
- unpcklps xmm0,xmm1 ; xmm0=(02 03 12 13)
- unpckhps xmm4,xmm1 ; xmm4=(22 23 32 33)
- movaps xmm5,xmm2 ; transpose coefficients(phase 1)
- unpcklps xmm2,xmm3 ; xmm2=(42 43 52 53)
- unpckhps xmm5,xmm3 ; xmm5=(62 63 72 73)
-
- movaps xmm6, XMMWORD [XMMBLOCK(0,0,edx,SIZEOF_FAST_FLOAT)]
- movaps xmm7, XMMWORD [XMMBLOCK(1,0,edx,SIZEOF_FAST_FLOAT)]
- movaps xmm1, XMMWORD [XMMBLOCK(4,0,edx,SIZEOF_FAST_FLOAT)]
- movaps xmm3, XMMWORD [XMMBLOCK(5,0,edx,SIZEOF_FAST_FLOAT)]
-
- ; xmm6=(00 10 20 30), xmm1=(40 50 60 70)
- ; xmm7=(01 11 21 31), xmm3=(41 51 61 71)
-
- movaps XMMWORD [wk(0)], xmm4 ; wk(0)=(22 23 32 33)
- movaps XMMWORD [wk(1)], xmm2 ; wk(1)=(42 43 52 53)
-
- movaps xmm4,xmm6 ; transpose coefficients(phase 1)
- unpcklps xmm6,xmm7 ; xmm6=(00 01 10 11)
- unpckhps xmm4,xmm7 ; xmm4=(20 21 30 31)
- movaps xmm2,xmm1 ; transpose coefficients(phase 1)
- unpcklps xmm1,xmm3 ; xmm1=(40 41 50 51)
- unpckhps xmm2,xmm3 ; xmm2=(60 61 70 71)
-
- movaps xmm7,xmm6 ; transpose coefficients(phase 2)
- unpcklps2 xmm6,xmm0 ; xmm6=(00 01 02 03)=data0
- unpckhps2 xmm7,xmm0 ; xmm7=(10 11 12 13)=data1
- movaps xmm3,xmm2 ; transpose coefficients(phase 2)
- unpcklps2 xmm2,xmm5 ; xmm2=(60 61 62 63)=data6
- unpckhps2 xmm3,xmm5 ; xmm3=(70 71 72 73)=data7
-
- movaps xmm0,xmm7
- movaps xmm5,xmm6
- subps xmm7,xmm2 ; xmm7=data1-data6=tmp6
- subps xmm6,xmm3 ; xmm6=data0-data7=tmp7
- addps xmm0,xmm2 ; xmm0=data1+data6=tmp1
- addps xmm5,xmm3 ; xmm5=data0+data7=tmp0
-
- movaps xmm2, XMMWORD [wk(0)] ; xmm2=(22 23 32 33)
- movaps xmm3, XMMWORD [wk(1)] ; xmm3=(42 43 52 53)
- movaps XMMWORD [wk(0)], xmm7 ; wk(0)=tmp6
- movaps XMMWORD [wk(1)], xmm6 ; wk(1)=tmp7
-
- movaps xmm7,xmm4 ; transpose coefficients(phase 2)
- unpcklps2 xmm4,xmm2 ; xmm4=(20 21 22 23)=data2
- unpckhps2 xmm7,xmm2 ; xmm7=(30 31 32 33)=data3
- movaps xmm6,xmm1 ; transpose coefficients(phase 2)
- unpcklps2 xmm1,xmm3 ; xmm1=(40 41 42 43)=data4
- unpckhps2 xmm6,xmm3 ; xmm6=(50 51 52 53)=data5
-
- movaps xmm2,xmm7
- movaps xmm3,xmm4
- addps xmm7,xmm1 ; xmm7=data3+data4=tmp3
- addps xmm4,xmm6 ; xmm4=data2+data5=tmp2
- subps xmm2,xmm1 ; xmm2=data3-data4=tmp4
- subps xmm3,xmm6 ; xmm3=data2-data5=tmp5
-
- ; -- Even part
-
- movaps xmm1,xmm5
- movaps xmm6,xmm0
- subps xmm5,xmm7 ; xmm5=tmp13
- subps xmm0,xmm4 ; xmm0=tmp12
- addps xmm1,xmm7 ; xmm1=tmp10
- addps xmm6,xmm4 ; xmm6=tmp11
-
- addps xmm0,xmm5
- mulps xmm0,[GOTOFF(ebx,PD_0_707)] ; xmm0=z1
-
- movaps xmm7,xmm1
- movaps xmm4,xmm5
- subps xmm1,xmm6 ; xmm1=data4
- subps xmm5,xmm0 ; xmm5=data6
- addps xmm7,xmm6 ; xmm7=data0
- addps xmm4,xmm0 ; xmm4=data2
-
- movaps XMMWORD [XMMBLOCK(4,0,edx,SIZEOF_FAST_FLOAT)], xmm1
- movaps XMMWORD [XMMBLOCK(6,0,edx,SIZEOF_FAST_FLOAT)], xmm5
- movaps XMMWORD [XMMBLOCK(0,0,edx,SIZEOF_FAST_FLOAT)], xmm7
- movaps XMMWORD [XMMBLOCK(2,0,edx,SIZEOF_FAST_FLOAT)], xmm4
-
- ; -- Odd part
-
- movaps xmm6, XMMWORD [wk(0)] ; xmm6=tmp6
- movaps xmm0, XMMWORD [wk(1)] ; xmm0=tmp7
-
- addps xmm2,xmm3 ; xmm2=tmp10
- addps xmm3,xmm6 ; xmm3=tmp11
- addps xmm6,xmm0 ; xmm6=tmp12, xmm0=tmp7
-
- mulps xmm3,[GOTOFF(ebx,PD_0_707)] ; xmm3=z3
-
- movaps xmm1,xmm2 ; xmm1=tmp10
- subps xmm2,xmm6
- mulps xmm2,[GOTOFF(ebx,PD_0_382)] ; xmm2=z5
- mulps xmm1,[GOTOFF(ebx,PD_0_541)] ; xmm1=MULTIPLY(tmp10,FIX_0_541196)
- mulps xmm6,[GOTOFF(ebx,PD_1_306)] ; xmm6=MULTIPLY(tmp12,FIX_1_306562)
- addps xmm1,xmm2 ; xmm1=z2
- addps xmm6,xmm2 ; xmm6=z4
-
- movaps xmm5,xmm0
- subps xmm0,xmm3 ; xmm0=z13
- addps xmm5,xmm3 ; xmm5=z11
-
- movaps xmm7,xmm0
- movaps xmm4,xmm5
- subps xmm0,xmm1 ; xmm0=data3
- subps xmm5,xmm6 ; xmm5=data7
- addps xmm7,xmm1 ; xmm7=data5
- addps xmm4,xmm6 ; xmm4=data1
-
- movaps XMMWORD [XMMBLOCK(3,0,edx,SIZEOF_FAST_FLOAT)], xmm0
- movaps XMMWORD [XMMBLOCK(7,0,edx,SIZEOF_FAST_FLOAT)], xmm5
- movaps XMMWORD [XMMBLOCK(5,0,edx,SIZEOF_FAST_FLOAT)], xmm7
- movaps XMMWORD [XMMBLOCK(1,0,edx,SIZEOF_FAST_FLOAT)], xmm4
-
- add edx, byte 4*SIZEOF_FAST_FLOAT
- dec ecx
- jnz near .columnloop
-
-; pop edi ; unused
-; pop esi ; unused
-; pop edx ; need not be preserved
-; pop ecx ; need not be preserved
- poppic ebx
- mov esp,ebp ; esp <- aligned ebp
- pop esp ; esp <- original ebp
- pop ebp
- ret
-
-; For some reason, the OS X linker does not honor the request to align the
-; segment unless we do this.
- align 16
diff --git a/simd/jfdctfst-sse2-64.asm b/simd/jfdctfst-sse2-64.asm
deleted file mode 100644
index 4c96685..0000000
--- a/simd/jfdctfst-sse2-64.asm
+++ /dev/null
@@ -1,391 +0,0 @@
-;
-; jfdctfst.asm - fast integer FDCT (64-bit SSE2)
-;
-; Copyright 2009 Pierre Ossman <ossman@cendio.se> for Cendio AB
-; Copyright (C) 2009, D. R. Commander.
-;
-; Based on the x86 SIMD extension for IJG JPEG library
-; Copyright (C) 1999-2006, MIYASAKA Masaru.
-; For conditions of distribution and use, see copyright notice in jsimdext.inc
-;
-; This file should be assembled with NASM (Netwide Assembler),
-; can *not* be assembled with Microsoft's MASM or any compatible
-; assembler (including Borland's Turbo Assembler).
-; NASM is available from http://nasm.sourceforge.net/ or
-; http://sourceforge.net/project/showfiles.php?group_id=6208
-;
-; This file contains a fast, not so accurate integer implementation of
-; the forward DCT (Discrete Cosine Transform). The following code is
-; based directly on the IJG's original jfdctfst.c; see the jfdctfst.c
-; for more details.
-;
-; [TAB8]
-
-%include "jsimdext.inc"
-%include "jdct.inc"
-
-; --------------------------------------------------------------------------
-
-%define CONST_BITS 8 ; 14 is also OK.
-
-%if CONST_BITS == 8
-F_0_382 equ 98 ; FIX(0.382683433)
-F_0_541 equ 139 ; FIX(0.541196100)
-F_0_707 equ 181 ; FIX(0.707106781)
-F_1_306 equ 334 ; FIX(1.306562965)
-%else
-; NASM cannot do compile-time arithmetic on floating-point constants.
-%define DESCALE(x,n) (((x)+(1<<((n)-1)))>>(n))
-F_0_382 equ DESCALE( 410903207,30-CONST_BITS) ; FIX(0.382683433)
-F_0_541 equ DESCALE( 581104887,30-CONST_BITS) ; FIX(0.541196100)
-F_0_707 equ DESCALE( 759250124,30-CONST_BITS) ; FIX(0.707106781)
-F_1_306 equ DESCALE(1402911301,30-CONST_BITS) ; FIX(1.306562965)
-%endif
-
-; --------------------------------------------------------------------------
- SECTION SEG_CONST
-
-; PRE_MULTIPLY_SCALE_BITS <= 2 (to avoid overflow)
-; CONST_BITS + CONST_SHIFT + PRE_MULTIPLY_SCALE_BITS == 16 (for pmulhw)
-
-%define PRE_MULTIPLY_SCALE_BITS 2
-%define CONST_SHIFT (16 - PRE_MULTIPLY_SCALE_BITS - CONST_BITS)
-
- alignz 16
- global EXTN(jconst_fdct_ifast_sse2)
-
-EXTN(jconst_fdct_ifast_sse2):
-
-PW_F0707 times 8 dw F_0_707 << CONST_SHIFT
-PW_F0382 times 8 dw F_0_382 << CONST_SHIFT
-PW_F0541 times 8 dw F_0_541 << CONST_SHIFT
-PW_F1306 times 8 dw F_1_306 << CONST_SHIFT
-
- alignz 16
-
-; --------------------------------------------------------------------------
- SECTION SEG_TEXT
- BITS 64
-;
-; Perform the forward DCT on one block of samples.
-;
-; GLOBAL(void)
-; jsimd_fdct_ifast_sse2 (DCTELEM *data)
-;
-
-; r10 = DCTELEM *data
-
-%define wk(i) rbp-(WK_NUM-(i))*SIZEOF_XMMWORD ; xmmword wk[WK_NUM]
-%define WK_NUM 2
-
- align 16
- global EXTN(jsimd_fdct_ifast_sse2)
-
-EXTN(jsimd_fdct_ifast_sse2):
- push rbp
- mov rax,rsp ; rax = original rbp
- sub rsp, byte 4
- and rsp, byte (-SIZEOF_XMMWORD) ; align to 128 bits
- mov [rsp],rax
- mov rbp,rsp ; rbp = aligned rbp
- lea rsp, [wk(0)]
- collect_args
-
- ; ---- Pass 1: process rows.
-
- mov rdx, r10 ; (DCTELEM *)
-
- movdqa xmm0, XMMWORD [XMMBLOCK(0,0,rdx,SIZEOF_DCTELEM)]
- movdqa xmm1, XMMWORD [XMMBLOCK(1,0,rdx,SIZEOF_DCTELEM)]
- movdqa xmm2, XMMWORD [XMMBLOCK(2,0,rdx,SIZEOF_DCTELEM)]
- movdqa xmm3, XMMWORD [XMMBLOCK(3,0,rdx,SIZEOF_DCTELEM)]
-
- ; xmm0=(00 01 02 03 04 05 06 07), xmm2=(20 21 22 23 24 25 26 27)
- ; xmm1=(10 11 12 13 14 15 16 17), xmm3=(30 31 32 33 34 35 36 37)
-
- movdqa xmm4,xmm0 ; transpose coefficients(phase 1)
- punpcklwd xmm0,xmm1 ; xmm0=(00 10 01 11 02 12 03 13)
- punpckhwd xmm4,xmm1 ; xmm4=(04 14 05 15 06 16 07 17)
- movdqa xmm5,xmm2 ; transpose coefficients(phase 1)
- punpcklwd xmm2,xmm3 ; xmm2=(20 30 21 31 22 32 23 33)
- punpckhwd xmm5,xmm3 ; xmm5=(24 34 25 35 26 36 27 37)
-
- movdqa xmm6, XMMWORD [XMMBLOCK(4,0,rdx,SIZEOF_DCTELEM)]
- movdqa xmm7, XMMWORD [XMMBLOCK(5,0,rdx,SIZEOF_DCTELEM)]
- movdqa xmm1, XMMWORD [XMMBLOCK(6,0,rdx,SIZEOF_DCTELEM)]
- movdqa xmm3, XMMWORD [XMMBLOCK(7,0,rdx,SIZEOF_DCTELEM)]
-
- ; xmm6=( 4 12 20 28 36 44 52 60), xmm1=( 6 14 22 30 38 46 54 62)
- ; xmm7=( 5 13 21 29 37 45 53 61), xmm3=( 7 15 23 31 39 47 55 63)
-
- movdqa XMMWORD [wk(0)], xmm2 ; wk(0)=(20 30 21 31 22 32 23 33)
- movdqa XMMWORD [wk(1)], xmm5 ; wk(1)=(24 34 25 35 26 36 27 37)
-
- movdqa xmm2,xmm6 ; transpose coefficients(phase 1)
- punpcklwd xmm6,xmm7 ; xmm6=(40 50 41 51 42 52 43 53)
- punpckhwd xmm2,xmm7 ; xmm2=(44 54 45 55 46 56 47 57)
- movdqa xmm5,xmm1 ; transpose coefficients(phase 1)
- punpcklwd xmm1,xmm3 ; xmm1=(60 70 61 71 62 72 63 73)
- punpckhwd xmm5,xmm3 ; xmm5=(64 74 65 75 66 76 67 77)
-
- movdqa xmm7,xmm6 ; transpose coefficients(phase 2)
- punpckldq xmm6,xmm1 ; xmm6=(40 50 60 70 41 51 61 71)
- punpckhdq xmm7,xmm1 ; xmm7=(42 52 62 72 43 53 63 73)
- movdqa xmm3,xmm2 ; transpose coefficients(phase 2)
- punpckldq xmm2,xmm5 ; xmm2=(44 54 64 74 45 55 65 75)
- punpckhdq xmm3,xmm5 ; xmm3=(46 56 66 76 47 57 67 77)
-
- movdqa xmm1, XMMWORD [wk(0)] ; xmm1=(20 30 21 31 22 32 23 33)
- movdqa xmm5, XMMWORD [wk(1)] ; xmm5=(24 34 25 35 26 36 27 37)
- movdqa XMMWORD [wk(0)], xmm7 ; wk(0)=(42 52 62 72 43 53 63 73)
- movdqa XMMWORD [wk(1)], xmm2 ; wk(1)=(44 54 64 74 45 55 65 75)
-
- movdqa xmm7,xmm0 ; transpose coefficients(phase 2)
- punpckldq xmm0,xmm1 ; xmm0=(00 10 20 30 01 11 21 31)
- punpckhdq xmm7,xmm1 ; xmm7=(02 12 22 32 03 13 23 33)
- movdqa xmm2,xmm4 ; transpose coefficients(phase 2)
- punpckldq xmm4,xmm5 ; xmm4=(04 14 24 34 05 15 25 35)
- punpckhdq xmm2,xmm5 ; xmm2=(06 16 26 36 07 17 27 37)
-
- movdqa xmm1,xmm0 ; transpose coefficients(phase 3)
- punpcklqdq xmm0,xmm6 ; xmm0=(00 10 20 30 40 50 60 70)=data0
- punpckhqdq xmm1,xmm6 ; xmm1=(01 11 21 31 41 51 61 71)=data1
- movdqa xmm5,xmm2 ; transpose coefficients(phase 3)
- punpcklqdq xmm2,xmm3 ; xmm2=(06 16 26 36 46 56 66 76)=data6
- punpckhqdq xmm5,xmm3 ; xmm5=(07 17 27 37 47 57 67 77)=data7
-
- movdqa xmm6,xmm1
- movdqa xmm3,xmm0
- psubw xmm1,xmm2 ; xmm1=data1-data6=tmp6
- psubw xmm0,xmm5 ; xmm0=data0-data7=tmp7
- paddw xmm6,xmm2 ; xmm6=data1+data6=tmp1
- paddw xmm3,xmm5 ; xmm3=data0+data7=tmp0
-
- movdqa xmm2, XMMWORD [wk(0)] ; xmm2=(42 52 62 72 43 53 63 73)
- movdqa xmm5, XMMWORD [wk(1)] ; xmm5=(44 54 64 74 45 55 65 75)
- movdqa XMMWORD [wk(0)], xmm1 ; wk(0)=tmp6
- movdqa XMMWORD [wk(1)], xmm0 ; wk(1)=tmp7
-
- movdqa xmm1,xmm7 ; transpose coefficients(phase 3)
- punpcklqdq xmm7,xmm2 ; xmm7=(02 12 22 32 42 52 62 72)=data2
- punpckhqdq xmm1,xmm2 ; xmm1=(03 13 23 33 43 53 63 73)=data3
- movdqa xmm0,xmm4 ; transpose coefficients(phase 3)
- punpcklqdq xmm4,xmm5 ; xmm4=(04 14 24 34 44 54 64 74)=data4
- punpckhqdq xmm0,xmm5 ; xmm0=(05 15 25 35 45 55 65 75)=data5
-
- movdqa xmm2,xmm1
- movdqa xmm5,xmm7
- paddw xmm1,xmm4 ; xmm1=data3+data4=tmp3
- paddw xmm7,xmm0 ; xmm7=data2+data5=tmp2
- psubw xmm2,xmm4 ; xmm2=data3-data4=tmp4
- psubw xmm5,xmm0 ; xmm5=data2-data5=tmp5
-
- ; -- Even part
-
- movdqa xmm4,xmm3
- movdqa xmm0,xmm6
- psubw xmm3,xmm1 ; xmm3=tmp13
- psubw xmm6,xmm7 ; xmm6=tmp12
- paddw xmm4,xmm1 ; xmm4=tmp10
- paddw xmm0,xmm7 ; xmm0=tmp11
-
- paddw xmm6,xmm3
- psllw xmm6,PRE_MULTIPLY_SCALE_BITS
- pmulhw xmm6,[rel PW_F0707] ; xmm6=z1
-
- movdqa xmm1,xmm4
- movdqa xmm7,xmm3
- psubw xmm4,xmm0 ; xmm4=data4
- psubw xmm3,xmm6 ; xmm3=data6
- paddw xmm1,xmm0 ; xmm1=data0
- paddw xmm7,xmm6 ; xmm7=data2
-
- movdqa xmm0, XMMWORD [wk(0)] ; xmm0=tmp6
- movdqa xmm6, XMMWORD [wk(1)] ; xmm6=tmp7
- movdqa XMMWORD [wk(0)], xmm4 ; wk(0)=data4
- movdqa XMMWORD [wk(1)], xmm3 ; wk(1)=data6
-
- ; -- Odd part
-
- paddw xmm2,xmm5 ; xmm2=tmp10
- paddw xmm5,xmm0 ; xmm5=tmp11
- paddw xmm0,xmm6 ; xmm0=tmp12, xmm6=tmp7
-
- psllw xmm2,PRE_MULTIPLY_SCALE_BITS
- psllw xmm0,PRE_MULTIPLY_SCALE_BITS
-
- psllw xmm5,PRE_MULTIPLY_SCALE_BITS
- pmulhw xmm5,[rel PW_F0707] ; xmm5=z3
-
- movdqa xmm4,xmm2 ; xmm4=tmp10
- psubw xmm2,xmm0
- pmulhw xmm2,[rel PW_F0382] ; xmm2=z5
- pmulhw xmm4,[rel PW_F0541] ; xmm4=MULTIPLY(tmp10,FIX_0_541196)
- pmulhw xmm0,[rel PW_F1306] ; xmm0=MULTIPLY(tmp12,FIX_1_306562)
- paddw xmm4,xmm2 ; xmm4=z2
- paddw xmm0,xmm2 ; xmm0=z4
-
- movdqa xmm3,xmm6
- psubw xmm6,xmm5 ; xmm6=z13
- paddw xmm3,xmm5 ; xmm3=z11
-
- movdqa xmm2,xmm6
- movdqa xmm5,xmm3
- psubw xmm6,xmm4 ; xmm6=data3
- psubw xmm3,xmm0 ; xmm3=data7
- paddw xmm2,xmm4 ; xmm2=data5
- paddw xmm5,xmm0 ; xmm5=data1
-
- ; ---- Pass 2: process columns.
-
- ; xmm1=(00 10 20 30 40 50 60 70), xmm7=(02 12 22 32 42 52 62 72)
- ; xmm5=(01 11 21 31 41 51 61 71), xmm6=(03 13 23 33 43 53 63 73)
-
- movdqa xmm4,xmm1 ; transpose coefficients(phase 1)
- punpcklwd xmm1,xmm5 ; xmm1=(00 01 10 11 20 21 30 31)
- punpckhwd xmm4,xmm5 ; xmm4=(40 41 50 51 60 61 70 71)
- movdqa xmm0,xmm7 ; transpose coefficients(phase 1)
- punpcklwd xmm7,xmm6 ; xmm7=(02 03 12 13 22 23 32 33)
- punpckhwd xmm0,xmm6 ; xmm0=(42 43 52 53 62 63 72 73)
-
- movdqa xmm5, XMMWORD [wk(0)] ; xmm5=col4
- movdqa xmm6, XMMWORD [wk(1)] ; xmm6=col6
-
- ; xmm5=(04 14 24 34 44 54 64 74), xmm6=(06 16 26 36 46 56 66 76)
- ; xmm2=(05 15 25 35 45 55 65 75), xmm3=(07 17 27 37 47 57 67 77)
-
- movdqa XMMWORD [wk(0)], xmm7 ; wk(0)=(02 03 12 13 22 23 32 33)
- movdqa XMMWORD [wk(1)], xmm0 ; wk(1)=(42 43 52 53 62 63 72 73)
-
- movdqa xmm7,xmm5 ; transpose coefficients(phase 1)
- punpcklwd xmm5,xmm2 ; xmm5=(04 05 14 15 24 25 34 35)
- punpckhwd xmm7,xmm2 ; xmm7=(44 45 54 55 64 65 74 75)
- movdqa xmm0,xmm6 ; transpose coefficients(phase 1)
- punpcklwd xmm6,xmm3 ; xmm6=(06 07 16 17 26 27 36 37)
- punpckhwd xmm0,xmm3 ; xmm0=(46 47 56 57 66 67 76 77)
-
- movdqa xmm2,xmm5 ; transpose coefficients(phase 2)
- punpckldq xmm5,xmm6 ; xmm5=(04 05 06 07 14 15 16 17)
- punpckhdq xmm2,xmm6 ; xmm2=(24 25 26 27 34 35 36 37)
- movdqa xmm3,xmm7 ; transpose coefficients(phase 2)
- punpckldq xmm7,xmm0 ; xmm7=(44 45 46 47 54 55 56 57)
- punpckhdq xmm3,xmm0 ; xmm3=(64 65 66 67 74 75 76 77)
-
- movdqa xmm6, XMMWORD [wk(0)] ; xmm6=(02 03 12 13 22 23 32 33)
- movdqa xmm0, XMMWORD [wk(1)] ; xmm0=(42 43 52 53 62 63 72 73)
- movdqa XMMWORD [wk(0)], xmm2 ; wk(0)=(24 25 26 27 34 35 36 37)
- movdqa XMMWORD [wk(1)], xmm7 ; wk(1)=(44 45 46 47 54 55 56 57)
-
- movdqa xmm2,xmm1 ; transpose coefficients(phase 2)
- punpckldq xmm1,xmm6 ; xmm1=(00 01 02 03 10 11 12 13)
- punpckhdq xmm2,xmm6 ; xmm2=(20 21 22 23 30 31 32 33)
- movdqa xmm7,xmm4 ; transpose coefficients(phase 2)
- punpckldq xmm4,xmm0 ; xmm4=(40 41 42 43 50 51 52 53)
- punpckhdq xmm7,xmm0 ; xmm7=(60 61 62 63 70 71 72 73)
-
- movdqa xmm6,xmm1 ; transpose coefficients(phase 3)
- punpcklqdq xmm1,xmm5 ; xmm1=(00 01 02 03 04 05 06 07)=data0
- punpckhqdq xmm6,xmm5 ; xmm6=(10 11 12 13 14 15 16 17)=data1
- movdqa xmm0,xmm7 ; transpose coefficients(phase 3)
- punpcklqdq xmm7,xmm3 ; xmm7=(60 61 62 63 64 65 66 67)=data6
- punpckhqdq xmm0,xmm3 ; xmm0=(70 71 72 73 74 75 76 77)=data7
-
- movdqa xmm5,xmm6
- movdqa xmm3,xmm1
- psubw xmm6,xmm7 ; xmm6=data1-data6=tmp6
- psubw xmm1,xmm0 ; xmm1=data0-data7=tmp7
- paddw xmm5,xmm7 ; xmm5=data1+data6=tmp1
- paddw xmm3,xmm0 ; xmm3=data0+data7=tmp0
-
- movdqa xmm7, XMMWORD [wk(0)] ; xmm7=(24 25 26 27 34 35 36 37)
- movdqa xmm0, XMMWORD [wk(1)] ; xmm0=(44 45 46 47 54 55 56 57)
- movdqa XMMWORD [wk(0)], xmm6 ; wk(0)=tmp6
- movdqa XMMWORD [wk(1)], xmm1 ; wk(1)=tmp7
-
- movdqa xmm6,xmm2 ; transpose coefficients(phase 3)
- punpcklqdq xmm2,xmm7 ; xmm2=(20 21 22 23 24 25 26 27)=data2
- punpckhqdq xmm6,xmm7 ; xmm6=(30 31 32 33 34 35 36 37)=data3
- movdqa xmm1,xmm4 ; transpose coefficients(phase 3)
- punpcklqdq xmm4,xmm0 ; xmm4=(40 41 42 43 44 45 46 47)=data4
- punpckhqdq xmm1,xmm0 ; xmm1=(50 51 52 53 54 55 56 57)=data5
-
- movdqa xmm7,xmm6
- movdqa xmm0,xmm2
- paddw xmm6,xmm4 ; xmm6=data3+data4=tmp3
- paddw xmm2,xmm1 ; xmm2=data2+data5=tmp2
- psubw xmm7,xmm4 ; xmm7=data3-data4=tmp4
- psubw xmm0,xmm1 ; xmm0=data2-data5=tmp5
-
- ; -- Even part
-
- movdqa xmm4,xmm3
- movdqa xmm1,xmm5
- psubw xmm3,xmm6 ; xmm3=tmp13
- psubw xmm5,xmm2 ; xmm5=tmp12
- paddw xmm4,xmm6 ; xmm4=tmp10
- paddw xmm1,xmm2 ; xmm1=tmp11
-
- paddw xmm5,xmm3
- psllw xmm5,PRE_MULTIPLY_SCALE_BITS
- pmulhw xmm5,[rel PW_F0707] ; xmm5=z1
-
- movdqa xmm6,xmm4
- movdqa xmm2,xmm3
- psubw xmm4,xmm1 ; xmm4=data4
- psubw xmm3,xmm5 ; xmm3=data6
- paddw xmm6,xmm1 ; xmm6=data0
- paddw xmm2,xmm5 ; xmm2=data2
-
- movdqa XMMWORD [XMMBLOCK(4,0,rdx,SIZEOF_DCTELEM)], xmm4
- movdqa XMMWORD [XMMBLOCK(6,0,rdx,SIZEOF_DCTELEM)], xmm3
- movdqa XMMWORD [XMMBLOCK(0,0,rdx,SIZEOF_DCTELEM)], xmm6
- movdqa XMMWORD [XMMBLOCK(2,0,rdx,SIZEOF_DCTELEM)], xmm2
-
- ; -- Odd part
-
- movdqa xmm1, XMMWORD [wk(0)] ; xmm1=tmp6
- movdqa xmm5, XMMWORD [wk(1)] ; xmm5=tmp7
-
- paddw xmm7,xmm0 ; xmm7=tmp10
- paddw xmm0,xmm1 ; xmm0=tmp11
- paddw xmm1,xmm5 ; xmm1=tmp12, xmm5=tmp7
-
- psllw xmm7,PRE_MULTIPLY_SCALE_BITS
- psllw xmm1,PRE_MULTIPLY_SCALE_BITS
-
- psllw xmm0,PRE_MULTIPLY_SCALE_BITS
- pmulhw xmm0,[rel PW_F0707] ; xmm0=z3
-
- movdqa xmm4,xmm7 ; xmm4=tmp10
- psubw xmm7,xmm1
- pmulhw xmm7,[rel PW_F0382] ; xmm7=z5
- pmulhw xmm4,[rel PW_F0541] ; xmm4=MULTIPLY(tmp10,FIX_0_541196)
- pmulhw xmm1,[rel PW_F1306] ; xmm1=MULTIPLY(tmp12,FIX_1_306562)
- paddw xmm4,xmm7 ; xmm4=z2
- paddw xmm1,xmm7 ; xmm1=z4
-
- movdqa xmm3,xmm5
- psubw xmm5,xmm0 ; xmm5=z13
- paddw xmm3,xmm0 ; xmm3=z11
-
- movdqa xmm6,xmm5
- movdqa xmm2,xmm3
- psubw xmm5,xmm4 ; xmm5=data3
- psubw xmm3,xmm1 ; xmm3=data7
- paddw xmm6,xmm4 ; xmm6=data5
- paddw xmm2,xmm1 ; xmm2=data1
-
- movdqa XMMWORD [XMMBLOCK(3,0,rdx,SIZEOF_DCTELEM)], xmm5
- movdqa XMMWORD [XMMBLOCK(7,0,rdx,SIZEOF_DCTELEM)], xmm3
- movdqa XMMWORD [XMMBLOCK(5,0,rdx,SIZEOF_DCTELEM)], xmm6
- movdqa XMMWORD [XMMBLOCK(1,0,rdx,SIZEOF_DCTELEM)], xmm2
-
- uncollect_args
- mov rsp,rbp ; rsp <- aligned rbp
- pop rsp ; rsp <- original rbp
- pop rbp
- ret
-
-; For some reason, the OS X linker does not honor the request to align the
-; segment unless we do this.
- align 16
diff --git a/simd/jfdctfst-sse2.asm b/simd/jfdctfst-sse2.asm
deleted file mode 100644
index 54856a2..0000000
--- a/simd/jfdctfst-sse2.asm
+++ /dev/null
@@ -1,403 +0,0 @@
-;
-; jfdctfst.asm - fast integer FDCT (SSE2)
-;
-; Copyright 2009 Pierre Ossman <ossman@cendio.se> for Cendio AB
-;
-; Based on the x86 SIMD extension for IJG JPEG library
-; Copyright (C) 1999-2006, MIYASAKA Masaru.
-; For conditions of distribution and use, see copyright notice in jsimdext.inc
-;
-; This file should be assembled with NASM (Netwide Assembler),
-; can *not* be assembled with Microsoft's MASM or any compatible
-; assembler (including Borland's Turbo Assembler).
-; NASM is available from http://nasm.sourceforge.net/ or
-; http://sourceforge.net/project/showfiles.php?group_id=6208
-;
-; This file contains a fast, not so accurate integer implementation of
-; the forward DCT (Discrete Cosine Transform). The following code is
-; based directly on the IJG's original jfdctfst.c; see the jfdctfst.c
-; for more details.
-;
-; [TAB8]
-
-%include "jsimdext.inc"
-%include "jdct.inc"
-
-; --------------------------------------------------------------------------
-
-%define CONST_BITS 8 ; 14 is also OK.
-
-%if CONST_BITS == 8
-F_0_382 equ 98 ; FIX(0.382683433)
-F_0_541 equ 139 ; FIX(0.541196100)
-F_0_707 equ 181 ; FIX(0.707106781)
-F_1_306 equ 334 ; FIX(1.306562965)
-%else
-; NASM cannot do compile-time arithmetic on floating-point constants.
-%define DESCALE(x,n) (((x)+(1<<((n)-1)))>>(n))
-F_0_382 equ DESCALE( 410903207,30-CONST_BITS) ; FIX(0.382683433)
-F_0_541 equ DESCALE( 581104887,30-CONST_BITS) ; FIX(0.541196100)
-F_0_707 equ DESCALE( 759250124,30-CONST_BITS) ; FIX(0.707106781)
-F_1_306 equ DESCALE(1402911301,30-CONST_BITS) ; FIX(1.306562965)
-%endif
-
-; --------------------------------------------------------------------------
- SECTION SEG_CONST
-
-; PRE_MULTIPLY_SCALE_BITS <= 2 (to avoid overflow)
-; CONST_BITS + CONST_SHIFT + PRE_MULTIPLY_SCALE_BITS == 16 (for pmulhw)
-
-%define PRE_MULTIPLY_SCALE_BITS 2
-%define CONST_SHIFT (16 - PRE_MULTIPLY_SCALE_BITS - CONST_BITS)
-
- alignz 16
- global EXTN(jconst_fdct_ifast_sse2)
-
-EXTN(jconst_fdct_ifast_sse2):
-
-PW_F0707 times 8 dw F_0_707 << CONST_SHIFT
-PW_F0382 times 8 dw F_0_382 << CONST_SHIFT
-PW_F0541 times 8 dw F_0_541 << CONST_SHIFT
-PW_F1306 times 8 dw F_1_306 << CONST_SHIFT
-
- alignz 16
-
-; --------------------------------------------------------------------------
- SECTION SEG_TEXT
- BITS 32
-;
-; Perform the forward DCT on one block of samples.
-;
-; GLOBAL(void)
-; jsimd_fdct_ifast_sse2 (DCTELEM *data)
-;
-
-%define data(b) (b)+8 ; DCTELEM *data
-
-%define original_ebp ebp+0
-%define wk(i) ebp-(WK_NUM-(i))*SIZEOF_XMMWORD ; xmmword wk[WK_NUM]
-%define WK_NUM 2
-
- align 16
- global EXTN(jsimd_fdct_ifast_sse2)
-
-EXTN(jsimd_fdct_ifast_sse2):
- push ebp
- mov eax,esp ; eax = original ebp
- sub esp, byte 4
- and esp, byte (-SIZEOF_XMMWORD) ; align to 128 bits
- mov [esp],eax
- mov ebp,esp ; ebp = aligned ebp
- lea esp, [wk(0)]
- pushpic ebx
-; push ecx ; unused
-; push edx ; need not be preserved
-; push esi ; unused
-; push edi ; unused
-
- get_GOT ebx ; get GOT address
-
- ; ---- Pass 1: process rows.
-
- mov edx, POINTER [data(eax)] ; (DCTELEM *)
-
- movdqa xmm0, XMMWORD [XMMBLOCK(0,0,edx,SIZEOF_DCTELEM)]
- movdqa xmm1, XMMWORD [XMMBLOCK(1,0,edx,SIZEOF_DCTELEM)]
- movdqa xmm2, XMMWORD [XMMBLOCK(2,0,edx,SIZEOF_DCTELEM)]
- movdqa xmm3, XMMWORD [XMMBLOCK(3,0,edx,SIZEOF_DCTELEM)]
-
- ; xmm0=(00 01 02 03 04 05 06 07), xmm2=(20 21 22 23 24 25 26 27)
- ; xmm1=(10 11 12 13 14 15 16 17), xmm3=(30 31 32 33 34 35 36 37)
-
- movdqa xmm4,xmm0 ; transpose coefficients(phase 1)
- punpcklwd xmm0,xmm1 ; xmm0=(00 10 01 11 02 12 03 13)
- punpckhwd xmm4,xmm1 ; xmm4=(04 14 05 15 06 16 07 17)
- movdqa xmm5,xmm2 ; transpose coefficients(phase 1)
- punpcklwd xmm2,xmm3 ; xmm2=(20 30 21 31 22 32 23 33)
- punpckhwd xmm5,xmm3 ; xmm5=(24 34 25 35 26 36 27 37)
-
- movdqa xmm6, XMMWORD [XMMBLOCK(4,0,edx,SIZEOF_DCTELEM)]
- movdqa xmm7, XMMWORD [XMMBLOCK(5,0,edx,SIZEOF_DCTELEM)]
- movdqa xmm1, XMMWORD [XMMBLOCK(6,0,edx,SIZEOF_DCTELEM)]
- movdqa xmm3, XMMWORD [XMMBLOCK(7,0,edx,SIZEOF_DCTELEM)]
-
- ; xmm6=( 4 12 20 28 36 44 52 60), xmm1=( 6 14 22 30 38 46 54 62)
- ; xmm7=( 5 13 21 29 37 45 53 61), xmm3=( 7 15 23 31 39 47 55 63)
-
- movdqa XMMWORD [wk(0)], xmm2 ; wk(0)=(20 30 21 31 22 32 23 33)
- movdqa XMMWORD [wk(1)], xmm5 ; wk(1)=(24 34 25 35 26 36 27 37)
-
- movdqa xmm2,xmm6 ; transpose coefficients(phase 1)
- punpcklwd xmm6,xmm7 ; xmm6=(40 50 41 51 42 52 43 53)
- punpckhwd xmm2,xmm7 ; xmm2=(44 54 45 55 46 56 47 57)
- movdqa xmm5,xmm1 ; transpose coefficients(phase 1)
- punpcklwd xmm1,xmm3 ; xmm1=(60 70 61 71 62 72 63 73)
- punpckhwd xmm5,xmm3 ; xmm5=(64 74 65 75 66 76 67 77)
-
- movdqa xmm7,xmm6 ; transpose coefficients(phase 2)
- punpckldq xmm6,xmm1 ; xmm6=(40 50 60 70 41 51 61 71)
- punpckhdq xmm7,xmm1 ; xmm7=(42 52 62 72 43 53 63 73)
- movdqa xmm3,xmm2 ; transpose coefficients(phase 2)
- punpckldq xmm2,xmm5 ; xmm2=(44 54 64 74 45 55 65 75)
- punpckhdq xmm3,xmm5 ; xmm3=(46 56 66 76 47 57 67 77)
-
- movdqa xmm1, XMMWORD [wk(0)] ; xmm1=(20 30 21 31 22 32 23 33)
- movdqa xmm5, XMMWORD [wk(1)] ; xmm5=(24 34 25 35 26 36 27 37)
- movdqa XMMWORD [wk(0)], xmm7 ; wk(0)=(42 52 62 72 43 53 63 73)
- movdqa XMMWORD [wk(1)], xmm2 ; wk(1)=(44 54 64 74 45 55 65 75)
-
- movdqa xmm7,xmm0 ; transpose coefficients(phase 2)
- punpckldq xmm0,xmm1 ; xmm0=(00 10 20 30 01 11 21 31)
- punpckhdq xmm7,xmm1 ; xmm7=(02 12 22 32 03 13 23 33)
- movdqa xmm2,xmm4 ; transpose coefficients(phase 2)
- punpckldq xmm4,xmm5 ; xmm4=(04 14 24 34 05 15 25 35)
- punpckhdq xmm2,xmm5 ; xmm2=(06 16 26 36 07 17 27 37)
-
- movdqa xmm1,xmm0 ; transpose coefficients(phase 3)
- punpcklqdq xmm0,xmm6 ; xmm0=(00 10 20 30 40 50 60 70)=data0
- punpckhqdq xmm1,xmm6 ; xmm1=(01 11 21 31 41 51 61 71)=data1
- movdqa xmm5,xmm2 ; transpose coefficients(phase 3)
- punpcklqdq xmm2,xmm3 ; xmm2=(06 16 26 36 46 56 66 76)=data6
- punpckhqdq xmm5,xmm3 ; xmm5=(07 17 27 37 47 57 67 77)=data7
-
- movdqa xmm6,xmm1
- movdqa xmm3,xmm0
- psubw xmm1,xmm2 ; xmm1=data1-data6=tmp6
- psubw xmm0,xmm5 ; xmm0=data0-data7=tmp7
- paddw xmm6,xmm2 ; xmm6=data1+data6=tmp1
- paddw xmm3,xmm5 ; xmm3=data0+data7=tmp0
-
- movdqa xmm2, XMMWORD [wk(0)] ; xmm2=(42 52 62 72 43 53 63 73)
- movdqa xmm5, XMMWORD [wk(1)] ; xmm5=(44 54 64 74 45 55 65 75)
- movdqa XMMWORD [wk(0)], xmm1 ; wk(0)=tmp6
- movdqa XMMWORD [wk(1)], xmm0 ; wk(1)=tmp7
-
- movdqa xmm1,xmm7 ; transpose coefficients(phase 3)
- punpcklqdq xmm7,xmm2 ; xmm7=(02 12 22 32 42 52 62 72)=data2
- punpckhqdq xmm1,xmm2 ; xmm1=(03 13 23 33 43 53 63 73)=data3
- movdqa xmm0,xmm4 ; transpose coefficients(phase 3)
- punpcklqdq xmm4,xmm5 ; xmm4=(04 14 24 34 44 54 64 74)=data4
- punpckhqdq xmm0,xmm5 ; xmm0=(05 15 25 35 45 55 65 75)=data5
-
- movdqa xmm2,xmm1
- movdqa xmm5,xmm7
- paddw xmm1,xmm4 ; xmm1=data3+data4=tmp3
- paddw xmm7,xmm0 ; xmm7=data2+data5=tmp2
- psubw xmm2,xmm4 ; xmm2=data3-data4=tmp4
- psubw xmm5,xmm0 ; xmm5=data2-data5=tmp5
-
- ; -- Even part
-
- movdqa xmm4,xmm3
- movdqa xmm0,xmm6
- psubw xmm3,xmm1 ; xmm3=tmp13
- psubw xmm6,xmm7 ; xmm6=tmp12
- paddw xmm4,xmm1 ; xmm4=tmp10
- paddw xmm0,xmm7 ; xmm0=tmp11
-
- paddw xmm6,xmm3
- psllw xmm6,PRE_MULTIPLY_SCALE_BITS
- pmulhw xmm6,[GOTOFF(ebx,PW_F0707)] ; xmm6=z1
-
- movdqa xmm1,xmm4
- movdqa xmm7,xmm3
- psubw xmm4,xmm0 ; xmm4=data4
- psubw xmm3,xmm6 ; xmm3=data6
- paddw xmm1,xmm0 ; xmm1=data0
- paddw xmm7,xmm6 ; xmm7=data2
-
- movdqa xmm0, XMMWORD [wk(0)] ; xmm0=tmp6
- movdqa xmm6, XMMWORD [wk(1)] ; xmm6=tmp7
- movdqa XMMWORD [wk(0)], xmm4 ; wk(0)=data4
- movdqa XMMWORD [wk(1)], xmm3 ; wk(1)=data6
-
- ; -- Odd part
-
- paddw xmm2,xmm5 ; xmm2=tmp10
- paddw xmm5,xmm0 ; xmm5=tmp11
- paddw xmm0,xmm6 ; xmm0=tmp12, xmm6=tmp7
-
- psllw xmm2,PRE_MULTIPLY_SCALE_BITS
- psllw xmm0,PRE_MULTIPLY_SCALE_BITS
-
- psllw xmm5,PRE_MULTIPLY_SCALE_BITS
- pmulhw xmm5,[GOTOFF(ebx,PW_F0707)] ; xmm5=z3
-
- movdqa xmm4,xmm2 ; xmm4=tmp10
- psubw xmm2,xmm0
- pmulhw xmm2,[GOTOFF(ebx,PW_F0382)] ; xmm2=z5
- pmulhw xmm4,[GOTOFF(ebx,PW_F0541)] ; xmm4=MULTIPLY(tmp10,FIX_0_541196)
- pmulhw xmm0,[GOTOFF(ebx,PW_F1306)] ; xmm0=MULTIPLY(tmp12,FIX_1_306562)
- paddw xmm4,xmm2 ; xmm4=z2
- paddw xmm0,xmm2 ; xmm0=z4
-
- movdqa xmm3,xmm6
- psubw xmm6,xmm5 ; xmm6=z13
- paddw xmm3,xmm5 ; xmm3=z11
-
- movdqa xmm2,xmm6
- movdqa xmm5,xmm3
- psubw xmm6,xmm4 ; xmm6=data3
- psubw xmm3,xmm0 ; xmm3=data7
- paddw xmm2,xmm4 ; xmm2=data5
- paddw xmm5,xmm0 ; xmm5=data1
-
- ; ---- Pass 2: process columns.
-
-; mov edx, POINTER [data(eax)] ; (DCTELEM *)
-
- ; xmm1=(00 10 20 30 40 50 60 70), xmm7=(02 12 22 32 42 52 62 72)
- ; xmm5=(01 11 21 31 41 51 61 71), xmm6=(03 13 23 33 43 53 63 73)
-
- movdqa xmm4,xmm1 ; transpose coefficients(phase 1)
- punpcklwd xmm1,xmm5 ; xmm1=(00 01 10 11 20 21 30 31)
- punpckhwd xmm4,xmm5 ; xmm4=(40 41 50 51 60 61 70 71)
- movdqa xmm0,xmm7 ; transpose coefficients(phase 1)
- punpcklwd xmm7,xmm6 ; xmm7=(02 03 12 13 22 23 32 33)
- punpckhwd xmm0,xmm6 ; xmm0=(42 43 52 53 62 63 72 73)
-
- movdqa xmm5, XMMWORD [wk(0)] ; xmm5=col4
- movdqa xmm6, XMMWORD [wk(1)] ; xmm6=col6
-
- ; xmm5=(04 14 24 34 44 54 64 74), xmm6=(06 16 26 36 46 56 66 76)
- ; xmm2=(05 15 25 35 45 55 65 75), xmm3=(07 17 27 37 47 57 67 77)
-
- movdqa XMMWORD [wk(0)], xmm7 ; wk(0)=(02 03 12 13 22 23 32 33)
- movdqa XMMWORD [wk(1)], xmm0 ; wk(1)=(42 43 52 53 62 63 72 73)
-
- movdqa xmm7,xmm5 ; transpose coefficients(phase 1)
- punpcklwd xmm5,xmm2 ; xmm5=(04 05 14 15 24 25 34 35)
- punpckhwd xmm7,xmm2 ; xmm7=(44 45 54 55 64 65 74 75)
- movdqa xmm0,xmm6 ; transpose coefficients(phase 1)
- punpcklwd xmm6,xmm3 ; xmm6=(06 07 16 17 26 27 36 37)
- punpckhwd xmm0,xmm3 ; xmm0=(46 47 56 57 66 67 76 77)
-
- movdqa xmm2,xmm5 ; transpose coefficients(phase 2)
- punpckldq xmm5,xmm6 ; xmm5=(04 05 06 07 14 15 16 17)
- punpckhdq xmm2,xmm6 ; xmm2=(24 25 26 27 34 35 36 37)
- movdqa xmm3,xmm7 ; transpose coefficients(phase 2)
- punpckldq xmm7,xmm0 ; xmm7=(44 45 46 47 54 55 56 57)
- punpckhdq xmm3,xmm0 ; xmm3=(64 65 66 67 74 75 76 77)
-
- movdqa xmm6, XMMWORD [wk(0)] ; xmm6=(02 03 12 13 22 23 32 33)
- movdqa xmm0, XMMWORD [wk(1)] ; xmm0=(42 43 52 53 62 63 72 73)
- movdqa XMMWORD [wk(0)], xmm2 ; wk(0)=(24 25 26 27 34 35 36 37)
- movdqa XMMWORD [wk(1)], xmm7 ; wk(1)=(44 45 46 47 54 55 56 57)
-
- movdqa xmm2,xmm1 ; transpose coefficients(phase 2)
- punpckldq xmm1,xmm6 ; xmm1=(00 01 02 03 10 11 12 13)
- punpckhdq xmm2,xmm6 ; xmm2=(20 21 22 23 30 31 32 33)
- movdqa xmm7,xmm4 ; transpose coefficients(phase 2)
- punpckldq xmm4,xmm0 ; xmm4=(40 41 42 43 50 51 52 53)
- punpckhdq xmm7,xmm0 ; xmm7=(60 61 62 63 70 71 72 73)
-
- movdqa xmm6,xmm1 ; transpose coefficients(phase 3)
- punpcklqdq xmm1,xmm5 ; xmm1=(00 01 02 03 04 05 06 07)=data0
- punpckhqdq xmm6,xmm5 ; xmm6=(10 11 12 13 14 15 16 17)=data1
- movdqa xmm0,xmm7 ; transpose coefficients(phase 3)
- punpcklqdq xmm7,xmm3 ; xmm7=(60 61 62 63 64 65 66 67)=data6
- punpckhqdq xmm0,xmm3 ; xmm0=(70 71 72 73 74 75 76 77)=data7
-
- movdqa xmm5,xmm6
- movdqa xmm3,xmm1
- psubw xmm6,xmm7 ; xmm6=data1-data6=tmp6
- psubw xmm1,xmm0 ; xmm1=data0-data7=tmp7
- paddw xmm5,xmm7 ; xmm5=data1+data6=tmp1
- paddw xmm3,xmm0 ; xmm3=data0+data7=tmp0
-
- movdqa xmm7, XMMWORD [wk(0)] ; xmm7=(24 25 26 27 34 35 36 37)
- movdqa xmm0, XMMWORD [wk(1)] ; xmm0=(44 45 46 47 54 55 56 57)
- movdqa XMMWORD [wk(0)], xmm6 ; wk(0)=tmp6
- movdqa XMMWORD [wk(1)], xmm1 ; wk(1)=tmp7
-
- movdqa xmm6,xmm2 ; transpose coefficients(phase 3)
- punpcklqdq xmm2,xmm7 ; xmm2=(20 21 22 23 24 25 26 27)=data2
- punpckhqdq xmm6,xmm7 ; xmm6=(30 31 32 33 34 35 36 37)=data3
- movdqa xmm1,xmm4 ; transpose coefficients(phase 3)
- punpcklqdq xmm4,xmm0 ; xmm4=(40 41 42 43 44 45 46 47)=data4
- punpckhqdq xmm1,xmm0 ; xmm1=(50 51 52 53 54 55 56 57)=data5
-
- movdqa xmm7,xmm6
- movdqa xmm0,xmm2
- paddw xmm6,xmm4 ; xmm6=data3+data4=tmp3
- paddw xmm2,xmm1 ; xmm2=data2+data5=tmp2
- psubw xmm7,xmm4 ; xmm7=data3-data4=tmp4
- psubw xmm0,xmm1 ; xmm0=data2-data5=tmp5
-
- ; -- Even part
-
- movdqa xmm4,xmm3
- movdqa xmm1,xmm5
- psubw xmm3,xmm6 ; xmm3=tmp13
- psubw xmm5,xmm2 ; xmm5=tmp12
- paddw xmm4,xmm6 ; xmm4=tmp10
- paddw xmm1,xmm2 ; xmm1=tmp11
-
- paddw xmm5,xmm3
- psllw xmm5,PRE_MULTIPLY_SCALE_BITS
- pmulhw xmm5,[GOTOFF(ebx,PW_F0707)] ; xmm5=z1
-
- movdqa xmm6,xmm4
- movdqa xmm2,xmm3
- psubw xmm4,xmm1 ; xmm4=data4
- psubw xmm3,xmm5 ; xmm3=data6
- paddw xmm6,xmm1 ; xmm6=data0
- paddw xmm2,xmm5 ; xmm2=data2
-
- movdqa XMMWORD [XMMBLOCK(4,0,edx,SIZEOF_DCTELEM)], xmm4
- movdqa XMMWORD [XMMBLOCK(6,0,edx,SIZEOF_DCTELEM)], xmm3
- movdqa XMMWORD [XMMBLOCK(0,0,edx,SIZEOF_DCTELEM)], xmm6
- movdqa XMMWORD [XMMBLOCK(2,0,edx,SIZEOF_DCTELEM)], xmm2
-
- ; -- Odd part
-
- movdqa xmm1, XMMWORD [wk(0)] ; xmm1=tmp6
- movdqa xmm5, XMMWORD [wk(1)] ; xmm5=tmp7
-
- paddw xmm7,xmm0 ; xmm7=tmp10
- paddw xmm0,xmm1 ; xmm0=tmp11
- paddw xmm1,xmm5 ; xmm1=tmp12, xmm5=tmp7
-
- psllw xmm7,PRE_MULTIPLY_SCALE_BITS
- psllw xmm1,PRE_MULTIPLY_SCALE_BITS
-
- psllw xmm0,PRE_MULTIPLY_SCALE_BITS
- pmulhw xmm0,[GOTOFF(ebx,PW_F0707)] ; xmm0=z3
-
- movdqa xmm4,xmm7 ; xmm4=tmp10
- psubw xmm7,xmm1
- pmulhw xmm7,[GOTOFF(ebx,PW_F0382)] ; xmm7=z5
- pmulhw xmm4,[GOTOFF(ebx,PW_F0541)] ; xmm4=MULTIPLY(tmp10,FIX_0_541196)
- pmulhw xmm1,[GOTOFF(ebx,PW_F1306)] ; xmm1=MULTIPLY(tmp12,FIX_1_306562)
- paddw xmm4,xmm7 ; xmm4=z2
- paddw xmm1,xmm7 ; xmm1=z4
-
- movdqa xmm3,xmm5
- psubw xmm5,xmm0 ; xmm5=z13
- paddw xmm3,xmm0 ; xmm3=z11
-
- movdqa xmm6,xmm5
- movdqa xmm2,xmm3
- psubw xmm5,xmm4 ; xmm5=data3
- psubw xmm3,xmm1 ; xmm3=data7
- paddw xmm6,xmm4 ; xmm6=data5
- paddw xmm2,xmm1 ; xmm2=data1
-
- movdqa XMMWORD [XMMBLOCK(3,0,edx,SIZEOF_DCTELEM)], xmm5
- movdqa XMMWORD [XMMBLOCK(7,0,edx,SIZEOF_DCTELEM)], xmm3
- movdqa XMMWORD [XMMBLOCK(5,0,edx,SIZEOF_DCTELEM)], xmm6
- movdqa XMMWORD [XMMBLOCK(1,0,edx,SIZEOF_DCTELEM)], xmm2
-
-; pop edi ; unused
-; pop esi ; unused
-; pop edx ; need not be preserved
-; pop ecx ; unused
- poppic ebx
- mov esp,ebp ; esp <- aligned ebp
- pop esp ; esp <- original ebp
- pop ebp
- ret
-
-; For some reason, the OS X linker does not honor the request to align the
-; segment unless we do this.
- align 16
diff --git a/simd/jfdctint-sse2-64.asm b/simd/jfdctint-sse2-64.asm
deleted file mode 100644
index 9a0ca0f..0000000
--- a/simd/jfdctint-sse2-64.asm
+++ /dev/null
@@ -1,621 +0,0 @@
-;
-; jfdctint.asm - accurate integer FDCT (64-bit SSE2)
-;
-; Copyright 2009 Pierre Ossman <ossman@cendio.se> for Cendio AB
-; Copyright (C) 2009, D. R. Commander.
-;
-; Based on the x86 SIMD extension for IJG JPEG library
-; Copyright (C) 1999-2006, MIYASAKA Masaru.
-; For conditions of distribution and use, see copyright notice in jsimdext.inc
-;
-; This file should be assembled with NASM (Netwide Assembler),
-; can *not* be assembled with Microsoft's MASM or any compatible
-; assembler (including Borland's Turbo Assembler).
-; NASM is available from http://nasm.sourceforge.net/ or
-; http://sourceforge.net/project/showfiles.php?group_id=6208
-;
-; This file contains a slow-but-accurate integer implementation of the
-; forward DCT (Discrete Cosine Transform). The following code is based
-; directly on the IJG's original jfdctint.c; see the jfdctint.c for
-; more details.
-;
-; [TAB8]
-
-%include "jsimdext.inc"
-%include "jdct.inc"
-
-; --------------------------------------------------------------------------
-
-%define CONST_BITS 13
-%define PASS1_BITS 2
-
-%define DESCALE_P1 (CONST_BITS-PASS1_BITS)
-%define DESCALE_P2 (CONST_BITS+PASS1_BITS)
-
-%if CONST_BITS == 13
-F_0_298 equ 2446 ; FIX(0.298631336)
-F_0_390 equ 3196 ; FIX(0.390180644)
-F_0_541 equ 4433 ; FIX(0.541196100)
-F_0_765 equ 6270 ; FIX(0.765366865)
-F_0_899 equ 7373 ; FIX(0.899976223)
-F_1_175 equ 9633 ; FIX(1.175875602)
-F_1_501 equ 12299 ; FIX(1.501321110)
-F_1_847 equ 15137 ; FIX(1.847759065)
-F_1_961 equ 16069 ; FIX(1.961570560)
-F_2_053 equ 16819 ; FIX(2.053119869)
-F_2_562 equ 20995 ; FIX(2.562915447)
-F_3_072 equ 25172 ; FIX(3.072711026)
-%else
-; NASM cannot do compile-time arithmetic on floating-point constants.
-%define DESCALE(x,n) (((x)+(1<<((n)-1)))>>(n))
-F_0_298 equ DESCALE( 320652955,30-CONST_BITS) ; FIX(0.298631336)
-F_0_390 equ DESCALE( 418953276,30-CONST_BITS) ; FIX(0.390180644)
-F_0_541 equ DESCALE( 581104887,30-CONST_BITS) ; FIX(0.541196100)
-F_0_765 equ DESCALE( 821806413,30-CONST_BITS) ; FIX(0.765366865)
-F_0_899 equ DESCALE( 966342111,30-CONST_BITS) ; FIX(0.899976223)
-F_1_175 equ DESCALE(1262586813,30-CONST_BITS) ; FIX(1.175875602)
-F_1_501 equ DESCALE(1612031267,30-CONST_BITS) ; FIX(1.501321110)
-F_1_847 equ DESCALE(1984016188,30-CONST_BITS) ; FIX(1.847759065)
-F_1_961 equ DESCALE(2106220350,30-CONST_BITS) ; FIX(1.961570560)
-F_2_053 equ DESCALE(2204520673,30-CONST_BITS) ; FIX(2.053119869)
-F_2_562 equ DESCALE(2751909506,30-CONST_BITS) ; FIX(2.562915447)
-F_3_072 equ DESCALE(3299298341,30-CONST_BITS) ; FIX(3.072711026)
-%endif
-
-; --------------------------------------------------------------------------
- SECTION SEG_CONST
-
- alignz 16
- global EXTN(jconst_fdct_islow_sse2)
-
-EXTN(jconst_fdct_islow_sse2):
-
-PW_F130_F054 times 4 dw (F_0_541+F_0_765), F_0_541
-PW_F054_MF130 times 4 dw F_0_541, (F_0_541-F_1_847)
-PW_MF078_F117 times 4 dw (F_1_175-F_1_961), F_1_175
-PW_F117_F078 times 4 dw F_1_175, (F_1_175-F_0_390)
-PW_MF060_MF089 times 4 dw (F_0_298-F_0_899),-F_0_899
-PW_MF089_F060 times 4 dw -F_0_899, (F_1_501-F_0_899)
-PW_MF050_MF256 times 4 dw (F_2_053-F_2_562),-F_2_562
-PW_MF256_F050 times 4 dw -F_2_562, (F_3_072-F_2_562)
-PD_DESCALE_P1 times 4 dd 1 << (DESCALE_P1-1)
-PD_DESCALE_P2 times 4 dd 1 << (DESCALE_P2-1)
-PW_DESCALE_P2X times 8 dw 1 << (PASS1_BITS-1)
-
- alignz 16
-
-; --------------------------------------------------------------------------
- SECTION SEG_TEXT
- BITS 64
-;
-; Perform the forward DCT on one block of samples.
-;
-; GLOBAL(void)
-; jsimd_fdct_islow_sse2 (DCTELEM *data)
-;
-
-; r10 = DCTELEM *data
-
-%define wk(i) rbp-(WK_NUM-(i))*SIZEOF_XMMWORD ; xmmword wk[WK_NUM]
-%define WK_NUM 6
-
- align 16
- global EXTN(jsimd_fdct_islow_sse2)
-
-EXTN(jsimd_fdct_islow_sse2):
- push rbp
- mov rax,rsp ; rax = original rbp
- sub rsp, byte 4
- and rsp, byte (-SIZEOF_XMMWORD) ; align to 128 bits
- mov [rsp],rax
- mov rbp,rsp ; rbp = aligned rbp
- lea rsp, [wk(0)]
- collect_args
-
- ; ---- Pass 1: process rows.
-
- mov rdx, r10 ; (DCTELEM *)
-
- movdqa xmm0, XMMWORD [XMMBLOCK(0,0,rdx,SIZEOF_DCTELEM)]
- movdqa xmm1, XMMWORD [XMMBLOCK(1,0,rdx,SIZEOF_DCTELEM)]
- movdqa xmm2, XMMWORD [XMMBLOCK(2,0,rdx,SIZEOF_DCTELEM)]
- movdqa xmm3, XMMWORD [XMMBLOCK(3,0,rdx,SIZEOF_DCTELEM)]
-
- ; xmm0=(00 01 02 03 04 05 06 07), xmm2=(20 21 22 23 24 25 26 27)
- ; xmm1=(10 11 12 13 14 15 16 17), xmm3=(30 31 32 33 34 35 36 37)
-
- movdqa xmm4,xmm0 ; transpose coefficients(phase 1)
- punpcklwd xmm0,xmm1 ; xmm0=(00 10 01 11 02 12 03 13)
- punpckhwd xmm4,xmm1 ; xmm4=(04 14 05 15 06 16 07 17)
- movdqa xmm5,xmm2 ; transpose coefficients(phase 1)
- punpcklwd xmm2,xmm3 ; xmm2=(20 30 21 31 22 32 23 33)
- punpckhwd xmm5,xmm3 ; xmm5=(24 34 25 35 26 36 27 37)
-
- movdqa xmm6, XMMWORD [XMMBLOCK(4,0,rdx,SIZEOF_DCTELEM)]
- movdqa xmm7, XMMWORD [XMMBLOCK(5,0,rdx,SIZEOF_DCTELEM)]
- movdqa xmm1, XMMWORD [XMMBLOCK(6,0,rdx,SIZEOF_DCTELEM)]
- movdqa xmm3, XMMWORD [XMMBLOCK(7,0,rdx,SIZEOF_DCTELEM)]
-
- ; xmm6=( 4 12 20 28 36 44 52 60), xmm1=( 6 14 22 30 38 46 54 62)
- ; xmm7=( 5 13 21 29 37 45 53 61), xmm3=( 7 15 23 31 39 47 55 63)
-
- movdqa XMMWORD [wk(0)], xmm2 ; wk(0)=(20 30 21 31 22 32 23 33)
- movdqa XMMWORD [wk(1)], xmm5 ; wk(1)=(24 34 25 35 26 36 27 37)
-
- movdqa xmm2,xmm6 ; transpose coefficients(phase 1)
- punpcklwd xmm6,xmm7 ; xmm6=(40 50 41 51 42 52 43 53)
- punpckhwd xmm2,xmm7 ; xmm2=(44 54 45 55 46 56 47 57)
- movdqa xmm5,xmm1 ; transpose coefficients(phase 1)
- punpcklwd xmm1,xmm3 ; xmm1=(60 70 61 71 62 72 63 73)
- punpckhwd xmm5,xmm3 ; xmm5=(64 74 65 75 66 76 67 77)
-
- movdqa xmm7,xmm6 ; transpose coefficients(phase 2)
- punpckldq xmm6,xmm1 ; xmm6=(40 50 60 70 41 51 61 71)
- punpckhdq xmm7,xmm1 ; xmm7=(42 52 62 72 43 53 63 73)
- movdqa xmm3,xmm2 ; transpose coefficients(phase 2)
- punpckldq xmm2,xmm5 ; xmm2=(44 54 64 74 45 55 65 75)
- punpckhdq xmm3,xmm5 ; xmm3=(46 56 66 76 47 57 67 77)
-
- movdqa xmm1, XMMWORD [wk(0)] ; xmm1=(20 30 21 31 22 32 23 33)
- movdqa xmm5, XMMWORD [wk(1)] ; xmm5=(24 34 25 35 26 36 27 37)
- movdqa XMMWORD [wk(2)], xmm7 ; wk(2)=(42 52 62 72 43 53 63 73)
- movdqa XMMWORD [wk(3)], xmm2 ; wk(3)=(44 54 64 74 45 55 65 75)
-
- movdqa xmm7,xmm0 ; transpose coefficients(phase 2)
- punpckldq xmm0,xmm1 ; xmm0=(00 10 20 30 01 11 21 31)
- punpckhdq xmm7,xmm1 ; xmm7=(02 12 22 32 03 13 23 33)
- movdqa xmm2,xmm4 ; transpose coefficients(phase 2)
- punpckldq xmm4,xmm5 ; xmm4=(04 14 24 34 05 15 25 35)
- punpckhdq xmm2,xmm5 ; xmm2=(06 16 26 36 07 17 27 37)
-
- movdqa xmm1,xmm0 ; transpose coefficients(phase 3)
- punpcklqdq xmm0,xmm6 ; xmm0=(00 10 20 30 40 50 60 70)=data0
- punpckhqdq xmm1,xmm6 ; xmm1=(01 11 21 31 41 51 61 71)=data1
- movdqa xmm5,xmm2 ; transpose coefficients(phase 3)
- punpcklqdq xmm2,xmm3 ; xmm2=(06 16 26 36 46 56 66 76)=data6
- punpckhqdq xmm5,xmm3 ; xmm5=(07 17 27 37 47 57 67 77)=data7
-
- movdqa xmm6,xmm1
- movdqa xmm3,xmm0
- psubw xmm1,xmm2 ; xmm1=data1-data6=tmp6
- psubw xmm0,xmm5 ; xmm0=data0-data7=tmp7
- paddw xmm6,xmm2 ; xmm6=data1+data6=tmp1
- paddw xmm3,xmm5 ; xmm3=data0+data7=tmp0
-
- movdqa xmm2, XMMWORD [wk(2)] ; xmm2=(42 52 62 72 43 53 63 73)
- movdqa xmm5, XMMWORD [wk(3)] ; xmm5=(44 54 64 74 45 55 65 75)
- movdqa XMMWORD [wk(0)], xmm1 ; wk(0)=tmp6
- movdqa XMMWORD [wk(1)], xmm0 ; wk(1)=tmp7
-
- movdqa xmm1,xmm7 ; transpose coefficients(phase 3)
- punpcklqdq xmm7,xmm2 ; xmm7=(02 12 22 32 42 52 62 72)=data2
- punpckhqdq xmm1,xmm2 ; xmm1=(03 13 23 33 43 53 63 73)=data3
- movdqa xmm0,xmm4 ; transpose coefficients(phase 3)
- punpcklqdq xmm4,xmm5 ; xmm4=(04 14 24 34 44 54 64 74)=data4
- punpckhqdq xmm0,xmm5 ; xmm0=(05 15 25 35 45 55 65 75)=data5
-
- movdqa xmm2,xmm1
- movdqa xmm5,xmm7
- paddw xmm1,xmm4 ; xmm1=data3+data4=tmp3
- paddw xmm7,xmm0 ; xmm7=data2+data5=tmp2
- psubw xmm2,xmm4 ; xmm2=data3-data4=tmp4
- psubw xmm5,xmm0 ; xmm5=data2-data5=tmp5
-
- ; -- Even part
-
- movdqa xmm4,xmm3
- movdqa xmm0,xmm6
- paddw xmm3,xmm1 ; xmm3=tmp10
- paddw xmm6,xmm7 ; xmm6=tmp11
- psubw xmm4,xmm1 ; xmm4=tmp13
- psubw xmm0,xmm7 ; xmm0=tmp12
-
- movdqa xmm1,xmm3
- paddw xmm3,xmm6 ; xmm3=tmp10+tmp11
- psubw xmm1,xmm6 ; xmm1=tmp10-tmp11
-
- psllw xmm3,PASS1_BITS ; xmm3=data0
- psllw xmm1,PASS1_BITS ; xmm1=data4
-
- movdqa XMMWORD [wk(2)], xmm3 ; wk(2)=data0
- movdqa XMMWORD [wk(3)], xmm1 ; wk(3)=data4
-
- ; (Original)
- ; z1 = (tmp12 + tmp13) * 0.541196100;
- ; data2 = z1 + tmp13 * 0.765366865;
- ; data6 = z1 + tmp12 * -1.847759065;
- ;
- ; (This implementation)
- ; data2 = tmp13 * (0.541196100 + 0.765366865) + tmp12 * 0.541196100;
- ; data6 = tmp13 * 0.541196100 + tmp12 * (0.541196100 - 1.847759065);
-
- movdqa xmm7,xmm4 ; xmm4=tmp13
- movdqa xmm6,xmm4
- punpcklwd xmm7,xmm0 ; xmm0=tmp12
- punpckhwd xmm6,xmm0
- movdqa xmm4,xmm7
- movdqa xmm0,xmm6
- pmaddwd xmm7,[rel PW_F130_F054] ; xmm7=data2L
- pmaddwd xmm6,[rel PW_F130_F054] ; xmm6=data2H
- pmaddwd xmm4,[rel PW_F054_MF130] ; xmm4=data6L
- pmaddwd xmm0,[rel PW_F054_MF130] ; xmm0=data6H
-
- paddd xmm7,[rel PD_DESCALE_P1]
- paddd xmm6,[rel PD_DESCALE_P1]
- psrad xmm7,DESCALE_P1
- psrad xmm6,DESCALE_P1
- paddd xmm4,[rel PD_DESCALE_P1]
- paddd xmm0,[rel PD_DESCALE_P1]
- psrad xmm4,DESCALE_P1
- psrad xmm0,DESCALE_P1
-
- packssdw xmm7,xmm6 ; xmm7=data2
- packssdw xmm4,xmm0 ; xmm4=data6
-
- movdqa XMMWORD [wk(4)], xmm7 ; wk(4)=data2
- movdqa XMMWORD [wk(5)], xmm4 ; wk(5)=data6
-
- ; -- Odd part
-
- movdqa xmm3, XMMWORD [wk(0)] ; xmm3=tmp6
- movdqa xmm1, XMMWORD [wk(1)] ; xmm1=tmp7
-
- movdqa xmm6,xmm2 ; xmm2=tmp4
- movdqa xmm0,xmm5 ; xmm5=tmp5
- paddw xmm6,xmm3 ; xmm6=z3
- paddw xmm0,xmm1 ; xmm0=z4
-
- ; (Original)
- ; z5 = (z3 + z4) * 1.175875602;
- ; z3 = z3 * -1.961570560; z4 = z4 * -0.390180644;
- ; z3 += z5; z4 += z5;
- ;
- ; (This implementation)
- ; z3 = z3 * (1.175875602 - 1.961570560) + z4 * 1.175875602;
- ; z4 = z3 * 1.175875602 + z4 * (1.175875602 - 0.390180644);
-
- movdqa xmm7,xmm6
- movdqa xmm4,xmm6
- punpcklwd xmm7,xmm0
- punpckhwd xmm4,xmm0
- movdqa xmm6,xmm7
- movdqa xmm0,xmm4
- pmaddwd xmm7,[rel PW_MF078_F117] ; xmm7=z3L
- pmaddwd xmm4,[rel PW_MF078_F117] ; xmm4=z3H
- pmaddwd xmm6,[rel PW_F117_F078] ; xmm6=z4L
- pmaddwd xmm0,[rel PW_F117_F078] ; xmm0=z4H
-
- movdqa XMMWORD [wk(0)], xmm7 ; wk(0)=z3L
- movdqa XMMWORD [wk(1)], xmm4 ; wk(1)=z3H
-
- ; (Original)
- ; z1 = tmp4 + tmp7; z2 = tmp5 + tmp6;
- ; tmp4 = tmp4 * 0.298631336; tmp5 = tmp5 * 2.053119869;
- ; tmp6 = tmp6 * 3.072711026; tmp7 = tmp7 * 1.501321110;
- ; z1 = z1 * -0.899976223; z2 = z2 * -2.562915447;
- ; data7 = tmp4 + z1 + z3; data5 = tmp5 + z2 + z4;
- ; data3 = tmp6 + z2 + z3; data1 = tmp7 + z1 + z4;
- ;
- ; (This implementation)
- ; tmp4 = tmp4 * (0.298631336 - 0.899976223) + tmp7 * -0.899976223;
- ; tmp5 = tmp5 * (2.053119869 - 2.562915447) + tmp6 * -2.562915447;
- ; tmp6 = tmp5 * -2.562915447 + tmp6 * (3.072711026 - 2.562915447);
- ; tmp7 = tmp4 * -0.899976223 + tmp7 * (1.501321110 - 0.899976223);
- ; data7 = tmp4 + z3; data5 = tmp5 + z4;
- ; data3 = tmp6 + z3; data1 = tmp7 + z4;
-
- movdqa xmm7,xmm2
- movdqa xmm4,xmm2
- punpcklwd xmm7,xmm1
- punpckhwd xmm4,xmm1
- movdqa xmm2,xmm7
- movdqa xmm1,xmm4
- pmaddwd xmm7,[rel PW_MF060_MF089] ; xmm7=tmp4L
- pmaddwd xmm4,[rel PW_MF060_MF089] ; xmm4=tmp4H
- pmaddwd xmm2,[rel PW_MF089_F060] ; xmm2=tmp7L
- pmaddwd xmm1,[rel PW_MF089_F060] ; xmm1=tmp7H
-
- paddd xmm7, XMMWORD [wk(0)] ; xmm7=data7L
- paddd xmm4, XMMWORD [wk(1)] ; xmm4=data7H
- paddd xmm2,xmm6 ; xmm2=data1L
- paddd xmm1,xmm0 ; xmm1=data1H
-
- paddd xmm7,[rel PD_DESCALE_P1]
- paddd xmm4,[rel PD_DESCALE_P1]
- psrad xmm7,DESCALE_P1
- psrad xmm4,DESCALE_P1
- paddd xmm2,[rel PD_DESCALE_P1]
- paddd xmm1,[rel PD_DESCALE_P1]
- psrad xmm2,DESCALE_P1
- psrad xmm1,DESCALE_P1
-
- packssdw xmm7,xmm4 ; xmm7=data7
- packssdw xmm2,xmm1 ; xmm2=data1
-
- movdqa xmm4,xmm5
- movdqa xmm1,xmm5
- punpcklwd xmm4,xmm3
- punpckhwd xmm1,xmm3
- movdqa xmm5,xmm4
- movdqa xmm3,xmm1
- pmaddwd xmm4,[rel PW_MF050_MF256] ; xmm4=tmp5L
- pmaddwd xmm1,[rel PW_MF050_MF256] ; xmm1=tmp5H
- pmaddwd xmm5,[rel PW_MF256_F050] ; xmm5=tmp6L
- pmaddwd xmm3,[rel PW_MF256_F050] ; xmm3=tmp6H
-
- paddd xmm4,xmm6 ; xmm4=data5L
- paddd xmm1,xmm0 ; xmm1=data5H
- paddd xmm5, XMMWORD [wk(0)] ; xmm5=data3L
- paddd xmm3, XMMWORD [wk(1)] ; xmm3=data3H
-
- paddd xmm4,[rel PD_DESCALE_P1]
- paddd xmm1,[rel PD_DESCALE_P1]
- psrad xmm4,DESCALE_P1
- psrad xmm1,DESCALE_P1
- paddd xmm5,[rel PD_DESCALE_P1]
- paddd xmm3,[rel PD_DESCALE_P1]
- psrad xmm5,DESCALE_P1
- psrad xmm3,DESCALE_P1
-
- packssdw xmm4,xmm1 ; xmm4=data5
- packssdw xmm5,xmm3 ; xmm5=data3
-
- ; ---- Pass 2: process columns.
-
- movdqa xmm6, XMMWORD [wk(2)] ; xmm6=col0
- movdqa xmm0, XMMWORD [wk(4)] ; xmm0=col2
-
- ; xmm6=(00 10 20 30 40 50 60 70), xmm0=(02 12 22 32 42 52 62 72)
- ; xmm2=(01 11 21 31 41 51 61 71), xmm5=(03 13 23 33 43 53 63 73)
-
- movdqa xmm1,xmm6 ; transpose coefficients(phase 1)
- punpcklwd xmm6,xmm2 ; xmm6=(00 01 10 11 20 21 30 31)
- punpckhwd xmm1,xmm2 ; xmm1=(40 41 50 51 60 61 70 71)
- movdqa xmm3,xmm0 ; transpose coefficients(phase 1)
- punpcklwd xmm0,xmm5 ; xmm0=(02 03 12 13 22 23 32 33)
- punpckhwd xmm3,xmm5 ; xmm3=(42 43 52 53 62 63 72 73)
-
- movdqa xmm2, XMMWORD [wk(3)] ; xmm2=col4
- movdqa xmm5, XMMWORD [wk(5)] ; xmm5=col6
-
- ; xmm2=(04 14 24 34 44 54 64 74), xmm5=(06 16 26 36 46 56 66 76)
- ; xmm4=(05 15 25 35 45 55 65 75), xmm7=(07 17 27 37 47 57 67 77)
-
- movdqa XMMWORD [wk(0)], xmm0 ; wk(0)=(02 03 12 13 22 23 32 33)
- movdqa XMMWORD [wk(1)], xmm3 ; wk(1)=(42 43 52 53 62 63 72 73)
-
- movdqa xmm0,xmm2 ; transpose coefficients(phase 1)
- punpcklwd xmm2,xmm4 ; xmm2=(04 05 14 15 24 25 34 35)
- punpckhwd xmm0,xmm4 ; xmm0=(44 45 54 55 64 65 74 75)
- movdqa xmm3,xmm5 ; transpose coefficients(phase 1)
- punpcklwd xmm5,xmm7 ; xmm5=(06 07 16 17 26 27 36 37)
- punpckhwd xmm3,xmm7 ; xmm3=(46 47 56 57 66 67 76 77)
-
- movdqa xmm4,xmm2 ; transpose coefficients(phase 2)
- punpckldq xmm2,xmm5 ; xmm2=(04 05 06 07 14 15 16 17)
- punpckhdq xmm4,xmm5 ; xmm4=(24 25 26 27 34 35 36 37)
- movdqa xmm7,xmm0 ; transpose coefficients(phase 2)
- punpckldq xmm0,xmm3 ; xmm0=(44 45 46 47 54 55 56 57)
- punpckhdq xmm7,xmm3 ; xmm7=(64 65 66 67 74 75 76 77)
-
- movdqa xmm5, XMMWORD [wk(0)] ; xmm5=(02 03 12 13 22 23 32 33)
- movdqa xmm3, XMMWORD [wk(1)] ; xmm3=(42 43 52 53 62 63 72 73)
- movdqa XMMWORD [wk(2)], xmm4 ; wk(2)=(24 25 26 27 34 35 36 37)
- movdqa XMMWORD [wk(3)], xmm0 ; wk(3)=(44 45 46 47 54 55 56 57)
-
- movdqa xmm4,xmm6 ; transpose coefficients(phase 2)
- punpckldq xmm6,xmm5 ; xmm6=(00 01 02 03 10 11 12 13)
- punpckhdq xmm4,xmm5 ; xmm4=(20 21 22 23 30 31 32 33)
- movdqa xmm0,xmm1 ; transpose coefficients(phase 2)
- punpckldq xmm1,xmm3 ; xmm1=(40 41 42 43 50 51 52 53)
- punpckhdq xmm0,xmm3 ; xmm0=(60 61 62 63 70 71 72 73)
-
- movdqa xmm5,xmm6 ; transpose coefficients(phase 3)
- punpcklqdq xmm6,xmm2 ; xmm6=(00 01 02 03 04 05 06 07)=data0
- punpckhqdq xmm5,xmm2 ; xmm5=(10 11 12 13 14 15 16 17)=data1
- movdqa xmm3,xmm0 ; transpose coefficients(phase 3)
- punpcklqdq xmm0,xmm7 ; xmm0=(60 61 62 63 64 65 66 67)=data6
- punpckhqdq xmm3,xmm7 ; xmm3=(70 71 72 73 74 75 76 77)=data7
-
- movdqa xmm2,xmm5
- movdqa xmm7,xmm6
- psubw xmm5,xmm0 ; xmm5=data1-data6=tmp6
- psubw xmm6,xmm3 ; xmm6=data0-data7=tmp7
- paddw xmm2,xmm0 ; xmm2=data1+data6=tmp1
- paddw xmm7,xmm3 ; xmm7=data0+data7=tmp0
-
- movdqa xmm0, XMMWORD [wk(2)] ; xmm0=(24 25 26 27 34 35 36 37)
- movdqa xmm3, XMMWORD [wk(3)] ; xmm3=(44 45 46 47 54 55 56 57)
- movdqa XMMWORD [wk(0)], xmm5 ; wk(0)=tmp6
- movdqa XMMWORD [wk(1)], xmm6 ; wk(1)=tmp7
-
- movdqa xmm5,xmm4 ; transpose coefficients(phase 3)
- punpcklqdq xmm4,xmm0 ; xmm4=(20 21 22 23 24 25 26 27)=data2
- punpckhqdq xmm5,xmm0 ; xmm5=(30 31 32 33 34 35 36 37)=data3
- movdqa xmm6,xmm1 ; transpose coefficients(phase 3)
- punpcklqdq xmm1,xmm3 ; xmm1=(40 41 42 43 44 45 46 47)=data4
- punpckhqdq xmm6,xmm3 ; xmm6=(50 51 52 53 54 55 56 57)=data5
-
- movdqa xmm0,xmm5
- movdqa xmm3,xmm4
- paddw xmm5,xmm1 ; xmm5=data3+data4=tmp3
- paddw xmm4,xmm6 ; xmm4=data2+data5=tmp2
- psubw xmm0,xmm1 ; xmm0=data3-data4=tmp4
- psubw xmm3,xmm6 ; xmm3=data2-data5=tmp5
-
- ; -- Even part
-
- movdqa xmm1,xmm7
- movdqa xmm6,xmm2
- paddw xmm7,xmm5 ; xmm7=tmp10
- paddw xmm2,xmm4 ; xmm2=tmp11
- psubw xmm1,xmm5 ; xmm1=tmp13
- psubw xmm6,xmm4 ; xmm6=tmp12
-
- movdqa xmm5,xmm7
- paddw xmm7,xmm2 ; xmm7=tmp10+tmp11
- psubw xmm5,xmm2 ; xmm5=tmp10-tmp11
-
- paddw xmm7,[rel PW_DESCALE_P2X]
- paddw xmm5,[rel PW_DESCALE_P2X]
- psraw xmm7,PASS1_BITS ; xmm7=data0
- psraw xmm5,PASS1_BITS ; xmm5=data4
-
- movdqa XMMWORD [XMMBLOCK(0,0,rdx,SIZEOF_DCTELEM)], xmm7
- movdqa XMMWORD [XMMBLOCK(4,0,rdx,SIZEOF_DCTELEM)], xmm5
-
- ; (Original)
- ; z1 = (tmp12 + tmp13) * 0.541196100;
- ; data2 = z1 + tmp13 * 0.765366865;
- ; data6 = z1 + tmp12 * -1.847759065;
- ;
- ; (This implementation)
- ; data2 = tmp13 * (0.541196100 + 0.765366865) + tmp12 * 0.541196100;
- ; data6 = tmp13 * 0.541196100 + tmp12 * (0.541196100 - 1.847759065);
-
- movdqa xmm4,xmm1 ; xmm1=tmp13
- movdqa xmm2,xmm1
- punpcklwd xmm4,xmm6 ; xmm6=tmp12
- punpckhwd xmm2,xmm6
- movdqa xmm1,xmm4
- movdqa xmm6,xmm2
- pmaddwd xmm4,[rel PW_F130_F054] ; xmm4=data2L
- pmaddwd xmm2,[rel PW_F130_F054] ; xmm2=data2H
- pmaddwd xmm1,[rel PW_F054_MF130] ; xmm1=data6L
- pmaddwd xmm6,[rel PW_F054_MF130] ; xmm6=data6H
-
- paddd xmm4,[rel PD_DESCALE_P2]
- paddd xmm2,[rel PD_DESCALE_P2]
- psrad xmm4,DESCALE_P2
- psrad xmm2,DESCALE_P2
- paddd xmm1,[rel PD_DESCALE_P2]
- paddd xmm6,[rel PD_DESCALE_P2]
- psrad xmm1,DESCALE_P2
- psrad xmm6,DESCALE_P2
-
- packssdw xmm4,xmm2 ; xmm4=data2
- packssdw xmm1,xmm6 ; xmm1=data6
-
- movdqa XMMWORD [XMMBLOCK(2,0,rdx,SIZEOF_DCTELEM)], xmm4
- movdqa XMMWORD [XMMBLOCK(6,0,rdx,SIZEOF_DCTELEM)], xmm1
-
- ; -- Odd part
-
- movdqa xmm7, XMMWORD [wk(0)] ; xmm7=tmp6
- movdqa xmm5, XMMWORD [wk(1)] ; xmm5=tmp7
-
- movdqa xmm2,xmm0 ; xmm0=tmp4
- movdqa xmm6,xmm3 ; xmm3=tmp5
- paddw xmm2,xmm7 ; xmm2=z3
- paddw xmm6,xmm5 ; xmm6=z4
-
- ; (Original)
- ; z5 = (z3 + z4) * 1.175875602;
- ; z3 = z3 * -1.961570560; z4 = z4 * -0.390180644;
- ; z3 += z5; z4 += z5;
- ;
- ; (This implementation)
- ; z3 = z3 * (1.175875602 - 1.961570560) + z4 * 1.175875602;
- ; z4 = z3 * 1.175875602 + z4 * (1.175875602 - 0.390180644);
-
- movdqa xmm4,xmm2
- movdqa xmm1,xmm2
- punpcklwd xmm4,xmm6
- punpckhwd xmm1,xmm6
- movdqa xmm2,xmm4
- movdqa xmm6,xmm1
- pmaddwd xmm4,[rel PW_MF078_F117] ; xmm4=z3L
- pmaddwd xmm1,[rel PW_MF078_F117] ; xmm1=z3H
- pmaddwd xmm2,[rel PW_F117_F078] ; xmm2=z4L
- pmaddwd xmm6,[rel PW_F117_F078] ; xmm6=z4H
-
- movdqa XMMWORD [wk(0)], xmm4 ; wk(0)=z3L
- movdqa XMMWORD [wk(1)], xmm1 ; wk(1)=z3H
-
- ; (Original)
- ; z1 = tmp4 + tmp7; z2 = tmp5 + tmp6;
- ; tmp4 = tmp4 * 0.298631336; tmp5 = tmp5 * 2.053119869;
- ; tmp6 = tmp6 * 3.072711026; tmp7 = tmp7 * 1.501321110;
- ; z1 = z1 * -0.899976223; z2 = z2 * -2.562915447;
- ; data7 = tmp4 + z1 + z3; data5 = tmp5 + z2 + z4;
- ; data3 = tmp6 + z2 + z3; data1 = tmp7 + z1 + z4;
- ;
- ; (This implementation)
- ; tmp4 = tmp4 * (0.298631336 - 0.899976223) + tmp7 * -0.899976223;
- ; tmp5 = tmp5 * (2.053119869 - 2.562915447) + tmp6 * -2.562915447;
- ; tmp6 = tmp5 * -2.562915447 + tmp6 * (3.072711026 - 2.562915447);
- ; tmp7 = tmp4 * -0.899976223 + tmp7 * (1.501321110 - 0.899976223);
- ; data7 = tmp4 + z3; data5 = tmp5 + z4;
- ; data3 = tmp6 + z3; data1 = tmp7 + z4;
-
- movdqa xmm4,xmm0
- movdqa xmm1,xmm0
- punpcklwd xmm4,xmm5
- punpckhwd xmm1,xmm5
- movdqa xmm0,xmm4
- movdqa xmm5,xmm1
- pmaddwd xmm4,[rel PW_MF060_MF089] ; xmm4=tmp4L
- pmaddwd xmm1,[rel PW_MF060_MF089] ; xmm1=tmp4H
- pmaddwd xmm0,[rel PW_MF089_F060] ; xmm0=tmp7L
- pmaddwd xmm5,[rel PW_MF089_F060] ; xmm5=tmp7H
-
- paddd xmm4, XMMWORD [wk(0)] ; xmm4=data7L
- paddd xmm1, XMMWORD [wk(1)] ; xmm1=data7H
- paddd xmm0,xmm2 ; xmm0=data1L
- paddd xmm5,xmm6 ; xmm5=data1H
-
- paddd xmm4,[rel PD_DESCALE_P2]
- paddd xmm1,[rel PD_DESCALE_P2]
- psrad xmm4,DESCALE_P2
- psrad xmm1,DESCALE_P2
- paddd xmm0,[rel PD_DESCALE_P2]
- paddd xmm5,[rel PD_DESCALE_P2]
- psrad xmm0,DESCALE_P2
- psrad xmm5,DESCALE_P2
-
- packssdw xmm4,xmm1 ; xmm4=data7
- packssdw xmm0,xmm5 ; xmm0=data1
-
- movdqa XMMWORD [XMMBLOCK(7,0,rdx,SIZEOF_DCTELEM)], xmm4
- movdqa XMMWORD [XMMBLOCK(1,0,rdx,SIZEOF_DCTELEM)], xmm0
-
- movdqa xmm1,xmm3
- movdqa xmm5,xmm3
- punpcklwd xmm1,xmm7
- punpckhwd xmm5,xmm7
- movdqa xmm3,xmm1
- movdqa xmm7,xmm5
- pmaddwd xmm1,[rel PW_MF050_MF256] ; xmm1=tmp5L
- pmaddwd xmm5,[rel PW_MF050_MF256] ; xmm5=tmp5H
- pmaddwd xmm3,[rel PW_MF256_F050] ; xmm3=tmp6L
- pmaddwd xmm7,[rel PW_MF256_F050] ; xmm7=tmp6H
-
- paddd xmm1,xmm2 ; xmm1=data5L
- paddd xmm5,xmm6 ; xmm5=data5H
- paddd xmm3, XMMWORD [wk(0)] ; xmm3=data3L
- paddd xmm7, XMMWORD [wk(1)] ; xmm7=data3H
-
- paddd xmm1,[rel PD_DESCALE_P2]
- paddd xmm5,[rel PD_DESCALE_P2]
- psrad xmm1,DESCALE_P2
- psrad xmm5,DESCALE_P2
- paddd xmm3,[rel PD_DESCALE_P2]
- paddd xmm7,[rel PD_DESCALE_P2]
- psrad xmm3,DESCALE_P2
- psrad xmm7,DESCALE_P2
-
- packssdw xmm1,xmm5 ; xmm1=data5
- packssdw xmm3,xmm7 ; xmm3=data3
-
- movdqa XMMWORD [XMMBLOCK(5,0,rdx,SIZEOF_DCTELEM)], xmm1
- movdqa XMMWORD [XMMBLOCK(3,0,rdx,SIZEOF_DCTELEM)], xmm3
-
- uncollect_args
- mov rsp,rbp ; rsp <- aligned rbp
- pop rsp ; rsp <- original rbp
- pop rbp
- ret
-
-; For some reason, the OS X linker does not honor the request to align the
-; segment unless we do this.
- align 16
diff --git a/simd/jfdctint-sse2.asm b/simd/jfdctint-sse2.asm
deleted file mode 100644
index db9d0bb..0000000
--- a/simd/jfdctint-sse2.asm
+++ /dev/null
@@ -1,633 +0,0 @@
-;
-; jfdctint.asm - accurate integer FDCT (SSE2)
-;
-; Copyright 2009 Pierre Ossman <ossman@cendio.se> for Cendio AB
-;
-; Based on the x86 SIMD extension for IJG JPEG library
-; Copyright (C) 1999-2006, MIYASAKA Masaru.
-; For conditions of distribution and use, see copyright notice in jsimdext.inc
-;
-; This file should be assembled with NASM (Netwide Assembler),
-; can *not* be assembled with Microsoft's MASM or any compatible
-; assembler (including Borland's Turbo Assembler).
-; NASM is available from http://nasm.sourceforge.net/ or
-; http://sourceforge.net/project/showfiles.php?group_id=6208
-;
-; This file contains a slow-but-accurate integer implementation of the
-; forward DCT (Discrete Cosine Transform). The following code is based
-; directly on the IJG's original jfdctint.c; see the jfdctint.c for
-; more details.
-;
-; [TAB8]
-
-%include "jsimdext.inc"
-%include "jdct.inc"
-
-; --------------------------------------------------------------------------
-
-%define CONST_BITS 13
-%define PASS1_BITS 2
-
-%define DESCALE_P1 (CONST_BITS-PASS1_BITS)
-%define DESCALE_P2 (CONST_BITS+PASS1_BITS)
-
-%if CONST_BITS == 13
-F_0_298 equ 2446 ; FIX(0.298631336)
-F_0_390 equ 3196 ; FIX(0.390180644)
-F_0_541 equ 4433 ; FIX(0.541196100)
-F_0_765 equ 6270 ; FIX(0.765366865)
-F_0_899 equ 7373 ; FIX(0.899976223)
-F_1_175 equ 9633 ; FIX(1.175875602)
-F_1_501 equ 12299 ; FIX(1.501321110)
-F_1_847 equ 15137 ; FIX(1.847759065)
-F_1_961 equ 16069 ; FIX(1.961570560)
-F_2_053 equ 16819 ; FIX(2.053119869)
-F_2_562 equ 20995 ; FIX(2.562915447)
-F_3_072 equ 25172 ; FIX(3.072711026)
-%else
-; NASM cannot do compile-time arithmetic on floating-point constants.
-%define DESCALE(x,n) (((x)+(1<<((n)-1)))>>(n))
-F_0_298 equ DESCALE( 320652955,30-CONST_BITS) ; FIX(0.298631336)
-F_0_390 equ DESCALE( 418953276,30-CONST_BITS) ; FIX(0.390180644)
-F_0_541 equ DESCALE( 581104887,30-CONST_BITS) ; FIX(0.541196100)
-F_0_765 equ DESCALE( 821806413,30-CONST_BITS) ; FIX(0.765366865)
-F_0_899 equ DESCALE( 966342111,30-CONST_BITS) ; FIX(0.899976223)
-F_1_175 equ DESCALE(1262586813,30-CONST_BITS) ; FIX(1.175875602)
-F_1_501 equ DESCALE(1612031267,30-CONST_BITS) ; FIX(1.501321110)
-F_1_847 equ DESCALE(1984016188,30-CONST_BITS) ; FIX(1.847759065)
-F_1_961 equ DESCALE(2106220350,30-CONST_BITS) ; FIX(1.961570560)
-F_2_053 equ DESCALE(2204520673,30-CONST_BITS) ; FIX(2.053119869)
-F_2_562 equ DESCALE(2751909506,30-CONST_BITS) ; FIX(2.562915447)
-F_3_072 equ DESCALE(3299298341,30-CONST_BITS) ; FIX(3.072711026)
-%endif
-
-; --------------------------------------------------------------------------
- SECTION SEG_CONST
-
- alignz 16
- global EXTN(jconst_fdct_islow_sse2)
-
-EXTN(jconst_fdct_islow_sse2):
-
-PW_F130_F054 times 4 dw (F_0_541+F_0_765), F_0_541
-PW_F054_MF130 times 4 dw F_0_541, (F_0_541-F_1_847)
-PW_MF078_F117 times 4 dw (F_1_175-F_1_961), F_1_175
-PW_F117_F078 times 4 dw F_1_175, (F_1_175-F_0_390)
-PW_MF060_MF089 times 4 dw (F_0_298-F_0_899),-F_0_899
-PW_MF089_F060 times 4 dw -F_0_899, (F_1_501-F_0_899)
-PW_MF050_MF256 times 4 dw (F_2_053-F_2_562),-F_2_562
-PW_MF256_F050 times 4 dw -F_2_562, (F_3_072-F_2_562)
-PD_DESCALE_P1 times 4 dd 1 << (DESCALE_P1-1)
-PD_DESCALE_P2 times 4 dd 1 << (DESCALE_P2-1)
-PW_DESCALE_P2X times 8 dw 1 << (PASS1_BITS-1)
-
- alignz 16
-
-; --------------------------------------------------------------------------
- SECTION SEG_TEXT
- BITS 32
-;
-; Perform the forward DCT on one block of samples.
-;
-; GLOBAL(void)
-; jsimd_fdct_islow_sse2 (DCTELEM *data)
-;
-
-%define data(b) (b)+8 ; DCTELEM *data
-
-%define original_ebp ebp+0
-%define wk(i) ebp-(WK_NUM-(i))*SIZEOF_XMMWORD ; xmmword wk[WK_NUM]
-%define WK_NUM 6
-
- align 16
- global EXTN(jsimd_fdct_islow_sse2)
-
-EXTN(jsimd_fdct_islow_sse2):
- push ebp
- mov eax,esp ; eax = original ebp
- sub esp, byte 4
- and esp, byte (-SIZEOF_XMMWORD) ; align to 128 bits
- mov [esp],eax
- mov ebp,esp ; ebp = aligned ebp
- lea esp, [wk(0)]
- pushpic ebx
-; push ecx ; unused
-; push edx ; need not be preserved
-; push esi ; unused
-; push edi ; unused
-
- get_GOT ebx ; get GOT address
-
- ; ---- Pass 1: process rows.
-
- mov edx, POINTER [data(eax)] ; (DCTELEM *)
-
- movdqa xmm0, XMMWORD [XMMBLOCK(0,0,edx,SIZEOF_DCTELEM)]
- movdqa xmm1, XMMWORD [XMMBLOCK(1,0,edx,SIZEOF_DCTELEM)]
- movdqa xmm2, XMMWORD [XMMBLOCK(2,0,edx,SIZEOF_DCTELEM)]
- movdqa xmm3, XMMWORD [XMMBLOCK(3,0,edx,SIZEOF_DCTELEM)]
-
- ; xmm0=(00 01 02 03 04 05 06 07), xmm2=(20 21 22 23 24 25 26 27)
- ; xmm1=(10 11 12 13 14 15 16 17), xmm3=(30 31 32 33 34 35 36 37)
-
- movdqa xmm4,xmm0 ; transpose coefficients(phase 1)
- punpcklwd xmm0,xmm1 ; xmm0=(00 10 01 11 02 12 03 13)
- punpckhwd xmm4,xmm1 ; xmm4=(04 14 05 15 06 16 07 17)
- movdqa xmm5,xmm2 ; transpose coefficients(phase 1)
- punpcklwd xmm2,xmm3 ; xmm2=(20 30 21 31 22 32 23 33)
- punpckhwd xmm5,xmm3 ; xmm5=(24 34 25 35 26 36 27 37)
-
- movdqa xmm6, XMMWORD [XMMBLOCK(4,0,edx,SIZEOF_DCTELEM)]
- movdqa xmm7, XMMWORD [XMMBLOCK(5,0,edx,SIZEOF_DCTELEM)]
- movdqa xmm1, XMMWORD [XMMBLOCK(6,0,edx,SIZEOF_DCTELEM)]
- movdqa xmm3, XMMWORD [XMMBLOCK(7,0,edx,SIZEOF_DCTELEM)]
-
- ; xmm6=( 4 12 20 28 36 44 52 60), xmm1=( 6 14 22 30 38 46 54 62)
- ; xmm7=( 5 13 21 29 37 45 53 61), xmm3=( 7 15 23 31 39 47 55 63)
-
- movdqa XMMWORD [wk(0)], xmm2 ; wk(0)=(20 30 21 31 22 32 23 33)
- movdqa XMMWORD [wk(1)], xmm5 ; wk(1)=(24 34 25 35 26 36 27 37)
-
- movdqa xmm2,xmm6 ; transpose coefficients(phase 1)
- punpcklwd xmm6,xmm7 ; xmm6=(40 50 41 51 42 52 43 53)
- punpckhwd xmm2,xmm7 ; xmm2=(44 54 45 55 46 56 47 57)
- movdqa xmm5,xmm1 ; transpose coefficients(phase 1)
- punpcklwd xmm1,xmm3 ; xmm1=(60 70 61 71 62 72 63 73)
- punpckhwd xmm5,xmm3 ; xmm5=(64 74 65 75 66 76 67 77)
-
- movdqa xmm7,xmm6 ; transpose coefficients(phase 2)
- punpckldq xmm6,xmm1 ; xmm6=(40 50 60 70 41 51 61 71)
- punpckhdq xmm7,xmm1 ; xmm7=(42 52 62 72 43 53 63 73)
- movdqa xmm3,xmm2 ; transpose coefficients(phase 2)
- punpckldq xmm2,xmm5 ; xmm2=(44 54 64 74 45 55 65 75)
- punpckhdq xmm3,xmm5 ; xmm3=(46 56 66 76 47 57 67 77)
-
- movdqa xmm1, XMMWORD [wk(0)] ; xmm1=(20 30 21 31 22 32 23 33)
- movdqa xmm5, XMMWORD [wk(1)] ; xmm5=(24 34 25 35 26 36 27 37)
- movdqa XMMWORD [wk(2)], xmm7 ; wk(2)=(42 52 62 72 43 53 63 73)
- movdqa XMMWORD [wk(3)], xmm2 ; wk(3)=(44 54 64 74 45 55 65 75)
-
- movdqa xmm7,xmm0 ; transpose coefficients(phase 2)
- punpckldq xmm0,xmm1 ; xmm0=(00 10 20 30 01 11 21 31)
- punpckhdq xmm7,xmm1 ; xmm7=(02 12 22 32 03 13 23 33)
- movdqa xmm2,xmm4 ; transpose coefficients(phase 2)
- punpckldq xmm4,xmm5 ; xmm4=(04 14 24 34 05 15 25 35)
- punpckhdq xmm2,xmm5 ; xmm2=(06 16 26 36 07 17 27 37)
-
- movdqa xmm1,xmm0 ; transpose coefficients(phase 3)
- punpcklqdq xmm0,xmm6 ; xmm0=(00 10 20 30 40 50 60 70)=data0
- punpckhqdq xmm1,xmm6 ; xmm1=(01 11 21 31 41 51 61 71)=data1
- movdqa xmm5,xmm2 ; transpose coefficients(phase 3)
- punpcklqdq xmm2,xmm3 ; xmm2=(06 16 26 36 46 56 66 76)=data6
- punpckhqdq xmm5,xmm3 ; xmm5=(07 17 27 37 47 57 67 77)=data7
-
- movdqa xmm6,xmm1
- movdqa xmm3,xmm0
- psubw xmm1,xmm2 ; xmm1=data1-data6=tmp6
- psubw xmm0,xmm5 ; xmm0=data0-data7=tmp7
- paddw xmm6,xmm2 ; xmm6=data1+data6=tmp1
- paddw xmm3,xmm5 ; xmm3=data0+data7=tmp0
-
- movdqa xmm2, XMMWORD [wk(2)] ; xmm2=(42 52 62 72 43 53 63 73)
- movdqa xmm5, XMMWORD [wk(3)] ; xmm5=(44 54 64 74 45 55 65 75)
- movdqa XMMWORD [wk(0)], xmm1 ; wk(0)=tmp6
- movdqa XMMWORD [wk(1)], xmm0 ; wk(1)=tmp7
-
- movdqa xmm1,xmm7 ; transpose coefficients(phase 3)
- punpcklqdq xmm7,xmm2 ; xmm7=(02 12 22 32 42 52 62 72)=data2
- punpckhqdq xmm1,xmm2 ; xmm1=(03 13 23 33 43 53 63 73)=data3
- movdqa xmm0,xmm4 ; transpose coefficients(phase 3)
- punpcklqdq xmm4,xmm5 ; xmm4=(04 14 24 34 44 54 64 74)=data4
- punpckhqdq xmm0,xmm5 ; xmm0=(05 15 25 35 45 55 65 75)=data5
-
- movdqa xmm2,xmm1
- movdqa xmm5,xmm7
- paddw xmm1,xmm4 ; xmm1=data3+data4=tmp3
- paddw xmm7,xmm0 ; xmm7=data2+data5=tmp2
- psubw xmm2,xmm4 ; xmm2=data3-data4=tmp4
- psubw xmm5,xmm0 ; xmm5=data2-data5=tmp5
-
- ; -- Even part
-
- movdqa xmm4,xmm3
- movdqa xmm0,xmm6
- paddw xmm3,xmm1 ; xmm3=tmp10
- paddw xmm6,xmm7 ; xmm6=tmp11
- psubw xmm4,xmm1 ; xmm4=tmp13
- psubw xmm0,xmm7 ; xmm0=tmp12
-
- movdqa xmm1,xmm3
- paddw xmm3,xmm6 ; xmm3=tmp10+tmp11
- psubw xmm1,xmm6 ; xmm1=tmp10-tmp11
-
- psllw xmm3,PASS1_BITS ; xmm3=data0
- psllw xmm1,PASS1_BITS ; xmm1=data4
-
- movdqa XMMWORD [wk(2)], xmm3 ; wk(2)=data0
- movdqa XMMWORD [wk(3)], xmm1 ; wk(3)=data4
-
- ; (Original)
- ; z1 = (tmp12 + tmp13) * 0.541196100;
- ; data2 = z1 + tmp13 * 0.765366865;
- ; data6 = z1 + tmp12 * -1.847759065;
- ;
- ; (This implementation)
- ; data2 = tmp13 * (0.541196100 + 0.765366865) + tmp12 * 0.541196100;
- ; data6 = tmp13 * 0.541196100 + tmp12 * (0.541196100 - 1.847759065);
-
- movdqa xmm7,xmm4 ; xmm4=tmp13
- movdqa xmm6,xmm4
- punpcklwd xmm7,xmm0 ; xmm0=tmp12
- punpckhwd xmm6,xmm0
- movdqa xmm4,xmm7
- movdqa xmm0,xmm6
- pmaddwd xmm7,[GOTOFF(ebx,PW_F130_F054)] ; xmm7=data2L
- pmaddwd xmm6,[GOTOFF(ebx,PW_F130_F054)] ; xmm6=data2H
- pmaddwd xmm4,[GOTOFF(ebx,PW_F054_MF130)] ; xmm4=data6L
- pmaddwd xmm0,[GOTOFF(ebx,PW_F054_MF130)] ; xmm0=data6H
-
- paddd xmm7,[GOTOFF(ebx,PD_DESCALE_P1)]
- paddd xmm6,[GOTOFF(ebx,PD_DESCALE_P1)]
- psrad xmm7,DESCALE_P1
- psrad xmm6,DESCALE_P1
- paddd xmm4,[GOTOFF(ebx,PD_DESCALE_P1)]
- paddd xmm0,[GOTOFF(ebx,PD_DESCALE_P1)]
- psrad xmm4,DESCALE_P1
- psrad xmm0,DESCALE_P1
-
- packssdw xmm7,xmm6 ; xmm7=data2
- packssdw xmm4,xmm0 ; xmm4=data6
-
- movdqa XMMWORD [wk(4)], xmm7 ; wk(4)=data2
- movdqa XMMWORD [wk(5)], xmm4 ; wk(5)=data6
-
- ; -- Odd part
-
- movdqa xmm3, XMMWORD [wk(0)] ; xmm3=tmp6
- movdqa xmm1, XMMWORD [wk(1)] ; xmm1=tmp7
-
- movdqa xmm6,xmm2 ; xmm2=tmp4
- movdqa xmm0,xmm5 ; xmm5=tmp5
- paddw xmm6,xmm3 ; xmm6=z3
- paddw xmm0,xmm1 ; xmm0=z4
-
- ; (Original)
- ; z5 = (z3 + z4) * 1.175875602;
- ; z3 = z3 * -1.961570560; z4 = z4 * -0.390180644;
- ; z3 += z5; z4 += z5;
- ;
- ; (This implementation)
- ; z3 = z3 * (1.175875602 - 1.961570560) + z4 * 1.175875602;
- ; z4 = z3 * 1.175875602 + z4 * (1.175875602 - 0.390180644);
-
- movdqa xmm7,xmm6
- movdqa xmm4,xmm6
- punpcklwd xmm7,xmm0
- punpckhwd xmm4,xmm0
- movdqa xmm6,xmm7
- movdqa xmm0,xmm4
- pmaddwd xmm7,[GOTOFF(ebx,PW_MF078_F117)] ; xmm7=z3L
- pmaddwd xmm4,[GOTOFF(ebx,PW_MF078_F117)] ; xmm4=z3H
- pmaddwd xmm6,[GOTOFF(ebx,PW_F117_F078)] ; xmm6=z4L
- pmaddwd xmm0,[GOTOFF(ebx,PW_F117_F078)] ; xmm0=z4H
-
- movdqa XMMWORD [wk(0)], xmm7 ; wk(0)=z3L
- movdqa XMMWORD [wk(1)], xmm4 ; wk(1)=z3H
-
- ; (Original)
- ; z1 = tmp4 + tmp7; z2 = tmp5 + tmp6;
- ; tmp4 = tmp4 * 0.298631336; tmp5 = tmp5 * 2.053119869;
- ; tmp6 = tmp6 * 3.072711026; tmp7 = tmp7 * 1.501321110;
- ; z1 = z1 * -0.899976223; z2 = z2 * -2.562915447;
- ; data7 = tmp4 + z1 + z3; data5 = tmp5 + z2 + z4;
- ; data3 = tmp6 + z2 + z3; data1 = tmp7 + z1 + z4;
- ;
- ; (This implementation)
- ; tmp4 = tmp4 * (0.298631336 - 0.899976223) + tmp7 * -0.899976223;
- ; tmp5 = tmp5 * (2.053119869 - 2.562915447) + tmp6 * -2.562915447;
- ; tmp6 = tmp5 * -2.562915447 + tmp6 * (3.072711026 - 2.562915447);
- ; tmp7 = tmp4 * -0.899976223 + tmp7 * (1.501321110 - 0.899976223);
- ; data7 = tmp4 + z3; data5 = tmp5 + z4;
- ; data3 = tmp6 + z3; data1 = tmp7 + z4;
-
- movdqa xmm7,xmm2
- movdqa xmm4,xmm2
- punpcklwd xmm7,xmm1
- punpckhwd xmm4,xmm1
- movdqa xmm2,xmm7
- movdqa xmm1,xmm4
- pmaddwd xmm7,[GOTOFF(ebx,PW_MF060_MF089)] ; xmm7=tmp4L
- pmaddwd xmm4,[GOTOFF(ebx,PW_MF060_MF089)] ; xmm4=tmp4H
- pmaddwd xmm2,[GOTOFF(ebx,PW_MF089_F060)] ; xmm2=tmp7L
- pmaddwd xmm1,[GOTOFF(ebx,PW_MF089_F060)] ; xmm1=tmp7H
-
- paddd xmm7, XMMWORD [wk(0)] ; xmm7=data7L
- paddd xmm4, XMMWORD [wk(1)] ; xmm4=data7H
- paddd xmm2,xmm6 ; xmm2=data1L
- paddd xmm1,xmm0 ; xmm1=data1H
-
- paddd xmm7,[GOTOFF(ebx,PD_DESCALE_P1)]
- paddd xmm4,[GOTOFF(ebx,PD_DESCALE_P1)]
- psrad xmm7,DESCALE_P1
- psrad xmm4,DESCALE_P1
- paddd xmm2,[GOTOFF(ebx,PD_DESCALE_P1)]
- paddd xmm1,[GOTOFF(ebx,PD_DESCALE_P1)]
- psrad xmm2,DESCALE_P1
- psrad xmm1,DESCALE_P1
-
- packssdw xmm7,xmm4 ; xmm7=data7
- packssdw xmm2,xmm1 ; xmm2=data1
-
- movdqa xmm4,xmm5
- movdqa xmm1,xmm5
- punpcklwd xmm4,xmm3
- punpckhwd xmm1,xmm3
- movdqa xmm5,xmm4
- movdqa xmm3,xmm1
- pmaddwd xmm4,[GOTOFF(ebx,PW_MF050_MF256)] ; xmm4=tmp5L
- pmaddwd xmm1,[GOTOFF(ebx,PW_MF050_MF256)] ; xmm1=tmp5H
- pmaddwd xmm5,[GOTOFF(ebx,PW_MF256_F050)] ; xmm5=tmp6L
- pmaddwd xmm3,[GOTOFF(ebx,PW_MF256_F050)] ; xmm3=tmp6H
-
- paddd xmm4,xmm6 ; xmm4=data5L
- paddd xmm1,xmm0 ; xmm1=data5H
- paddd xmm5, XMMWORD [wk(0)] ; xmm5=data3L
- paddd xmm3, XMMWORD [wk(1)] ; xmm3=data3H
-
- paddd xmm4,[GOTOFF(ebx,PD_DESCALE_P1)]
- paddd xmm1,[GOTOFF(ebx,PD_DESCALE_P1)]
- psrad xmm4,DESCALE_P1
- psrad xmm1,DESCALE_P1
- paddd xmm5,[GOTOFF(ebx,PD_DESCALE_P1)]
- paddd xmm3,[GOTOFF(ebx,PD_DESCALE_P1)]
- psrad xmm5,DESCALE_P1
- psrad xmm3,DESCALE_P1
-
- packssdw xmm4,xmm1 ; xmm4=data5
- packssdw xmm5,xmm3 ; xmm5=data3
-
- ; ---- Pass 2: process columns.
-
-; mov edx, POINTER [data(eax)] ; (DCTELEM *)
-
- movdqa xmm6, XMMWORD [wk(2)] ; xmm6=col0
- movdqa xmm0, XMMWORD [wk(4)] ; xmm0=col2
-
- ; xmm6=(00 10 20 30 40 50 60 70), xmm0=(02 12 22 32 42 52 62 72)
- ; xmm2=(01 11 21 31 41 51 61 71), xmm5=(03 13 23 33 43 53 63 73)
-
- movdqa xmm1,xmm6 ; transpose coefficients(phase 1)
- punpcklwd xmm6,xmm2 ; xmm6=(00 01 10 11 20 21 30 31)
- punpckhwd xmm1,xmm2 ; xmm1=(40 41 50 51 60 61 70 71)
- movdqa xmm3,xmm0 ; transpose coefficients(phase 1)
- punpcklwd xmm0,xmm5 ; xmm0=(02 03 12 13 22 23 32 33)
- punpckhwd xmm3,xmm5 ; xmm3=(42 43 52 53 62 63 72 73)
-
- movdqa xmm2, XMMWORD [wk(3)] ; xmm2=col4
- movdqa xmm5, XMMWORD [wk(5)] ; xmm5=col6
-
- ; xmm2=(04 14 24 34 44 54 64 74), xmm5=(06 16 26 36 46 56 66 76)
- ; xmm4=(05 15 25 35 45 55 65 75), xmm7=(07 17 27 37 47 57 67 77)
-
- movdqa XMMWORD [wk(0)], xmm0 ; wk(0)=(02 03 12 13 22 23 32 33)
- movdqa XMMWORD [wk(1)], xmm3 ; wk(1)=(42 43 52 53 62 63 72 73)
-
- movdqa xmm0,xmm2 ; transpose coefficients(phase 1)
- punpcklwd xmm2,xmm4 ; xmm2=(04 05 14 15 24 25 34 35)
- punpckhwd xmm0,xmm4 ; xmm0=(44 45 54 55 64 65 74 75)
- movdqa xmm3,xmm5 ; transpose coefficients(phase 1)
- punpcklwd xmm5,xmm7 ; xmm5=(06 07 16 17 26 27 36 37)
- punpckhwd xmm3,xmm7 ; xmm3=(46 47 56 57 66 67 76 77)
-
- movdqa xmm4,xmm2 ; transpose coefficients(phase 2)
- punpckldq xmm2,xmm5 ; xmm2=(04 05 06 07 14 15 16 17)
- punpckhdq xmm4,xmm5 ; xmm4=(24 25 26 27 34 35 36 37)
- movdqa xmm7,xmm0 ; transpose coefficients(phase 2)
- punpckldq xmm0,xmm3 ; xmm0=(44 45 46 47 54 55 56 57)
- punpckhdq xmm7,xmm3 ; xmm7=(64 65 66 67 74 75 76 77)
-
- movdqa xmm5, XMMWORD [wk(0)] ; xmm5=(02 03 12 13 22 23 32 33)
- movdqa xmm3, XMMWORD [wk(1)] ; xmm3=(42 43 52 53 62 63 72 73)
- movdqa XMMWORD [wk(2)], xmm4 ; wk(2)=(24 25 26 27 34 35 36 37)
- movdqa XMMWORD [wk(3)], xmm0 ; wk(3)=(44 45 46 47 54 55 56 57)
-
- movdqa xmm4,xmm6 ; transpose coefficients(phase 2)
- punpckldq xmm6,xmm5 ; xmm6=(00 01 02 03 10 11 12 13)
- punpckhdq xmm4,xmm5 ; xmm4=(20 21 22 23 30 31 32 33)
- movdqa xmm0,xmm1 ; transpose coefficients(phase 2)
- punpckldq xmm1,xmm3 ; xmm1=(40 41 42 43 50 51 52 53)
- punpckhdq xmm0,xmm3 ; xmm0=(60 61 62 63 70 71 72 73)
-
- movdqa xmm5,xmm6 ; transpose coefficients(phase 3)
- punpcklqdq xmm6,xmm2 ; xmm6=(00 01 02 03 04 05 06 07)=data0
- punpckhqdq xmm5,xmm2 ; xmm5=(10 11 12 13 14 15 16 17)=data1
- movdqa xmm3,xmm0 ; transpose coefficients(phase 3)
- punpcklqdq xmm0,xmm7 ; xmm0=(60 61 62 63 64 65 66 67)=data6
- punpckhqdq xmm3,xmm7 ; xmm3=(70 71 72 73 74 75 76 77)=data7
-
- movdqa xmm2,xmm5
- movdqa xmm7,xmm6
- psubw xmm5,xmm0 ; xmm5=data1-data6=tmp6
- psubw xmm6,xmm3 ; xmm6=data0-data7=tmp7
- paddw xmm2,xmm0 ; xmm2=data1+data6=tmp1
- paddw xmm7,xmm3 ; xmm7=data0+data7=tmp0
-
- movdqa xmm0, XMMWORD [wk(2)] ; xmm0=(24 25 26 27 34 35 36 37)
- movdqa xmm3, XMMWORD [wk(3)] ; xmm3=(44 45 46 47 54 55 56 57)
- movdqa XMMWORD [wk(0)], xmm5 ; wk(0)=tmp6
- movdqa XMMWORD [wk(1)], xmm6 ; wk(1)=tmp7
-
- movdqa xmm5,xmm4 ; transpose coefficients(phase 3)
- punpcklqdq xmm4,xmm0 ; xmm4=(20 21 22 23 24 25 26 27)=data2
- punpckhqdq xmm5,xmm0 ; xmm5=(30 31 32 33 34 35 36 37)=data3
- movdqa xmm6,xmm1 ; transpose coefficients(phase 3)
- punpcklqdq xmm1,xmm3 ; xmm1=(40 41 42 43 44 45 46 47)=data4
- punpckhqdq xmm6,xmm3 ; xmm6=(50 51 52 53 54 55 56 57)=data5
-
- movdqa xmm0,xmm5
- movdqa xmm3,xmm4
- paddw xmm5,xmm1 ; xmm5=data3+data4=tmp3
- paddw xmm4,xmm6 ; xmm4=data2+data5=tmp2
- psubw xmm0,xmm1 ; xmm0=data3-data4=tmp4
- psubw xmm3,xmm6 ; xmm3=data2-data5=tmp5
-
- ; -- Even part
-
- movdqa xmm1,xmm7
- movdqa xmm6,xmm2
- paddw xmm7,xmm5 ; xmm7=tmp10
- paddw xmm2,xmm4 ; xmm2=tmp11
- psubw xmm1,xmm5 ; xmm1=tmp13
- psubw xmm6,xmm4 ; xmm6=tmp12
-
- movdqa xmm5,xmm7
- paddw xmm7,xmm2 ; xmm7=tmp10+tmp11
- psubw xmm5,xmm2 ; xmm5=tmp10-tmp11
-
- paddw xmm7,[GOTOFF(ebx,PW_DESCALE_P2X)]
- paddw xmm5,[GOTOFF(ebx,PW_DESCALE_P2X)]
- psraw xmm7,PASS1_BITS ; xmm7=data0
- psraw xmm5,PASS1_BITS ; xmm5=data4
-
- movdqa XMMWORD [XMMBLOCK(0,0,edx,SIZEOF_DCTELEM)], xmm7
- movdqa XMMWORD [XMMBLOCK(4,0,edx,SIZEOF_DCTELEM)], xmm5
-
- ; (Original)
- ; z1 = (tmp12 + tmp13) * 0.541196100;
- ; data2 = z1 + tmp13 * 0.765366865;
- ; data6 = z1 + tmp12 * -1.847759065;
- ;
- ; (This implementation)
- ; data2 = tmp13 * (0.541196100 + 0.765366865) + tmp12 * 0.541196100;
- ; data6 = tmp13 * 0.541196100 + tmp12 * (0.541196100 - 1.847759065);
-
- movdqa xmm4,xmm1 ; xmm1=tmp13
- movdqa xmm2,xmm1
- punpcklwd xmm4,xmm6 ; xmm6=tmp12
- punpckhwd xmm2,xmm6
- movdqa xmm1,xmm4
- movdqa xmm6,xmm2
- pmaddwd xmm4,[GOTOFF(ebx,PW_F130_F054)] ; xmm4=data2L
- pmaddwd xmm2,[GOTOFF(ebx,PW_F130_F054)] ; xmm2=data2H
- pmaddwd xmm1,[GOTOFF(ebx,PW_F054_MF130)] ; xmm1=data6L
- pmaddwd xmm6,[GOTOFF(ebx,PW_F054_MF130)] ; xmm6=data6H
-
- paddd xmm4,[GOTOFF(ebx,PD_DESCALE_P2)]
- paddd xmm2,[GOTOFF(ebx,PD_DESCALE_P2)]
- psrad xmm4,DESCALE_P2
- psrad xmm2,DESCALE_P2
- paddd xmm1,[GOTOFF(ebx,PD_DESCALE_P2)]
- paddd xmm6,[GOTOFF(ebx,PD_DESCALE_P2)]
- psrad xmm1,DESCALE_P2
- psrad xmm6,DESCALE_P2
-
- packssdw xmm4,xmm2 ; xmm4=data2
- packssdw xmm1,xmm6 ; xmm1=data6
-
- movdqa XMMWORD [XMMBLOCK(2,0,edx,SIZEOF_DCTELEM)], xmm4
- movdqa XMMWORD [XMMBLOCK(6,0,edx,SIZEOF_DCTELEM)], xmm1
-
- ; -- Odd part
-
- movdqa xmm7, XMMWORD [wk(0)] ; xmm7=tmp6
- movdqa xmm5, XMMWORD [wk(1)] ; xmm5=tmp7
-
- movdqa xmm2,xmm0 ; xmm0=tmp4
- movdqa xmm6,xmm3 ; xmm3=tmp5
- paddw xmm2,xmm7 ; xmm2=z3
- paddw xmm6,xmm5 ; xmm6=z4
-
- ; (Original)
- ; z5 = (z3 + z4) * 1.175875602;
- ; z3 = z3 * -1.961570560; z4 = z4 * -0.390180644;
- ; z3 += z5; z4 += z5;
- ;
- ; (This implementation)
- ; z3 = z3 * (1.175875602 - 1.961570560) + z4 * 1.175875602;
- ; z4 = z3 * 1.175875602 + z4 * (1.175875602 - 0.390180644);
-
- movdqa xmm4,xmm2
- movdqa xmm1,xmm2
- punpcklwd xmm4,xmm6
- punpckhwd xmm1,xmm6
- movdqa xmm2,xmm4
- movdqa xmm6,xmm1
- pmaddwd xmm4,[GOTOFF(ebx,PW_MF078_F117)] ; xmm4=z3L
- pmaddwd xmm1,[GOTOFF(ebx,PW_MF078_F117)] ; xmm1=z3H
- pmaddwd xmm2,[GOTOFF(ebx,PW_F117_F078)] ; xmm2=z4L
- pmaddwd xmm6,[GOTOFF(ebx,PW_F117_F078)] ; xmm6=z4H
-
- movdqa XMMWORD [wk(0)], xmm4 ; wk(0)=z3L
- movdqa XMMWORD [wk(1)], xmm1 ; wk(1)=z3H
-
- ; (Original)
- ; z1 = tmp4 + tmp7; z2 = tmp5 + tmp6;
- ; tmp4 = tmp4 * 0.298631336; tmp5 = tmp5 * 2.053119869;
- ; tmp6 = tmp6 * 3.072711026; tmp7 = tmp7 * 1.501321110;
- ; z1 = z1 * -0.899976223; z2 = z2 * -2.562915447;
- ; data7 = tmp4 + z1 + z3; data5 = tmp5 + z2 + z4;
- ; data3 = tmp6 + z2 + z3; data1 = tmp7 + z1 + z4;
- ;
- ; (This implementation)
- ; tmp4 = tmp4 * (0.298631336 - 0.899976223) + tmp7 * -0.899976223;
- ; tmp5 = tmp5 * (2.053119869 - 2.562915447) + tmp6 * -2.562915447;
- ; tmp6 = tmp5 * -2.562915447 + tmp6 * (3.072711026 - 2.562915447);
- ; tmp7 = tmp4 * -0.899976223 + tmp7 * (1.501321110 - 0.899976223);
- ; data7 = tmp4 + z3; data5 = tmp5 + z4;
- ; data3 = tmp6 + z3; data1 = tmp7 + z4;
-
- movdqa xmm4,xmm0
- movdqa xmm1,xmm0
- punpcklwd xmm4,xmm5
- punpckhwd xmm1,xmm5
- movdqa xmm0,xmm4
- movdqa xmm5,xmm1
- pmaddwd xmm4,[GOTOFF(ebx,PW_MF060_MF089)] ; xmm4=tmp4L
- pmaddwd xmm1,[GOTOFF(ebx,PW_MF060_MF089)] ; xmm1=tmp4H
- pmaddwd xmm0,[GOTOFF(ebx,PW_MF089_F060)] ; xmm0=tmp7L
- pmaddwd xmm5,[GOTOFF(ebx,PW_MF089_F060)] ; xmm5=tmp7H
-
- paddd xmm4, XMMWORD [wk(0)] ; xmm4=data7L
- paddd xmm1, XMMWORD [wk(1)] ; xmm1=data7H
- paddd xmm0,xmm2 ; xmm0=data1L
- paddd xmm5,xmm6 ; xmm5=data1H
-
- paddd xmm4,[GOTOFF(ebx,PD_DESCALE_P2)]
- paddd xmm1,[GOTOFF(ebx,PD_DESCALE_P2)]
- psrad xmm4,DESCALE_P2
- psrad xmm1,DESCALE_P2
- paddd xmm0,[GOTOFF(ebx,PD_DESCALE_P2)]
- paddd xmm5,[GOTOFF(ebx,PD_DESCALE_P2)]
- psrad xmm0,DESCALE_P2
- psrad xmm5,DESCALE_P2
-
- packssdw xmm4,xmm1 ; xmm4=data7
- packssdw xmm0,xmm5 ; xmm0=data1
-
- movdqa XMMWORD [XMMBLOCK(7,0,edx,SIZEOF_DCTELEM)], xmm4
- movdqa XMMWORD [XMMBLOCK(1,0,edx,SIZEOF_DCTELEM)], xmm0
-
- movdqa xmm1,xmm3
- movdqa xmm5,xmm3
- punpcklwd xmm1,xmm7
- punpckhwd xmm5,xmm7
- movdqa xmm3,xmm1
- movdqa xmm7,xmm5
- pmaddwd xmm1,[GOTOFF(ebx,PW_MF050_MF256)] ; xmm1=tmp5L
- pmaddwd xmm5,[GOTOFF(ebx,PW_MF050_MF256)] ; xmm5=tmp5H
- pmaddwd xmm3,[GOTOFF(ebx,PW_MF256_F050)] ; xmm3=tmp6L
- pmaddwd xmm7,[GOTOFF(ebx,PW_MF256_F050)] ; xmm7=tmp6H
-
- paddd xmm1,xmm2 ; xmm1=data5L
- paddd xmm5,xmm6 ; xmm5=data5H
- paddd xmm3, XMMWORD [wk(0)] ; xmm3=data3L
- paddd xmm7, XMMWORD [wk(1)] ; xmm7=data3H
-
- paddd xmm1,[GOTOFF(ebx,PD_DESCALE_P2)]
- paddd xmm5,[GOTOFF(ebx,PD_DESCALE_P2)]
- psrad xmm1,DESCALE_P2
- psrad xmm5,DESCALE_P2
- paddd xmm3,[GOTOFF(ebx,PD_DESCALE_P2)]
- paddd xmm7,[GOTOFF(ebx,PD_DESCALE_P2)]
- psrad xmm3,DESCALE_P2
- psrad xmm7,DESCALE_P2
-
- packssdw xmm1,xmm5 ; xmm1=data5
- packssdw xmm3,xmm7 ; xmm3=data3
-
- movdqa XMMWORD [XMMBLOCK(5,0,edx,SIZEOF_DCTELEM)], xmm1
- movdqa XMMWORD [XMMBLOCK(3,0,edx,SIZEOF_DCTELEM)], xmm3
-
-; pop edi ; unused
-; pop esi ; unused
-; pop edx ; need not be preserved
-; pop ecx ; unused
- poppic ebx
- mov esp,ebp ; esp <- aligned ebp
- pop esp ; esp <- original ebp
- pop ebp
- ret
-
-; For some reason, the OS X linker does not honor the request to align the
-; segment unless we do this.
- align 16
diff --git a/simd/jidctflt-sse2-64.asm b/simd/jidctflt-sse2-64.asm
deleted file mode 100644
index bdda05d..0000000
--- a/simd/jidctflt-sse2-64.asm
+++ /dev/null
@@ -1,482 +0,0 @@
-;
-; jidctflt.asm - floating-point IDCT (64-bit SSE & SSE2)
-;
-; Copyright 2009 Pierre Ossman <ossman@cendio.se> for Cendio AB
-; Copyright (C) 2009, D. R. Commander.
-;
-; Based on the x86 SIMD extension for IJG JPEG library
-; Copyright (C) 1999-2006, MIYASAKA Masaru.
-; For conditions of distribution and use, see copyright notice in jsimdext.inc
-;
-; This file should be assembled with NASM (Netwide Assembler),
-; can *not* be assembled with Microsoft's MASM or any compatible
-; assembler (including Borland's Turbo Assembler).
-; NASM is available from http://nasm.sourceforge.net/ or
-; http://sourceforge.net/project/showfiles.php?group_id=6208
-;
-; This file contains a floating-point implementation of the inverse DCT
-; (Discrete Cosine Transform). The following code is based directly on
-; the IJG's original jidctflt.c; see the jidctflt.c for more details.
-;
-; [TAB8]
-
-%include "jsimdext.inc"
-%include "jdct.inc"
-
-; --------------------------------------------------------------------------
-
-%macro unpcklps2 2 ; %1=(0 1 2 3) / %2=(4 5 6 7) => %1=(0 1 4 5)
- shufps %1,%2,0x44
-%endmacro
-
-%macro unpckhps2 2 ; %1=(0 1 2 3) / %2=(4 5 6 7) => %1=(2 3 6 7)
- shufps %1,%2,0xEE
-%endmacro
-
-; --------------------------------------------------------------------------
- SECTION SEG_CONST
-
- alignz 16
- global EXTN(jconst_idct_float_sse2)
-
-EXTN(jconst_idct_float_sse2):
-
-PD_1_414 times 4 dd 1.414213562373095048801689
-PD_1_847 times 4 dd 1.847759065022573512256366
-PD_1_082 times 4 dd 1.082392200292393968799446
-PD_M2_613 times 4 dd -2.613125929752753055713286
-PD_RNDINT_MAGIC times 4 dd 100663296.0 ; (float)(0x00C00000 << 3)
-PB_CENTERJSAMP times 16 db CENTERJSAMPLE
-
- alignz 16
-
-; --------------------------------------------------------------------------
- SECTION SEG_TEXT
- BITS 64
-;
-; Perform dequantization and inverse DCT on one block of coefficients.
-;
-; GLOBAL(void)
-; jsimd_idct_float_sse2 (void *dct_table, JCOEFPTR coef_block,
-; JSAMPARRAY output_buf, JDIMENSION output_col)
-;
-
-; r10 = void *dct_table
-; r11 = JCOEFPTR coef_block
-; r12 = JSAMPARRAY output_buf
-; r13 = JDIMENSION output_col
-
-%define original_rbp rbp+0
-%define wk(i) rbp-(WK_NUM-(i))*SIZEOF_XMMWORD ; xmmword wk[WK_NUM]
-%define WK_NUM 2
-%define workspace wk(0)-DCTSIZE2*SIZEOF_FAST_FLOAT
- ; FAST_FLOAT workspace[DCTSIZE2]
-
- align 16
- global EXTN(jsimd_idct_float_sse2)
-
-EXTN(jsimd_idct_float_sse2):
- push rbp
- mov rax,rsp ; rax = original rbp
- sub rsp, byte 4
- and rsp, byte (-SIZEOF_XMMWORD) ; align to 128 bits
- mov [rsp],rax
- mov rbp,rsp ; rbp = aligned rbp
- lea rsp, [workspace]
- collect_args
- push rbx
-
- ; ---- Pass 1: process columns from input, store into work array.
-
- mov rdx, r10 ; quantptr
- mov rsi, r11 ; inptr
- lea rdi, [workspace] ; FAST_FLOAT *wsptr
- mov rcx, DCTSIZE/4 ; ctr
-.columnloop:
-%ifndef NO_ZERO_COLUMN_TEST_FLOAT_SSE
- mov eax, DWORD [DWBLOCK(1,0,rsi,SIZEOF_JCOEF)]
- or eax, DWORD [DWBLOCK(2,0,rsi,SIZEOF_JCOEF)]
- jnz near .columnDCT
-
- movq xmm1, XMM_MMWORD [MMBLOCK(1,0,rsi,SIZEOF_JCOEF)]
- movq xmm2, XMM_MMWORD [MMBLOCK(2,0,rsi,SIZEOF_JCOEF)]
- movq xmm3, XMM_MMWORD [MMBLOCK(3,0,rsi,SIZEOF_JCOEF)]
- movq xmm4, XMM_MMWORD [MMBLOCK(4,0,rsi,SIZEOF_JCOEF)]
- movq xmm5, XMM_MMWORD [MMBLOCK(5,0,rsi,SIZEOF_JCOEF)]
- movq xmm6, XMM_MMWORD [MMBLOCK(6,0,rsi,SIZEOF_JCOEF)]
- movq xmm7, XMM_MMWORD [MMBLOCK(7,0,rsi,SIZEOF_JCOEF)]
- por xmm1,xmm2
- por xmm3,xmm4
- por xmm5,xmm6
- por xmm1,xmm3
- por xmm5,xmm7
- por xmm1,xmm5
- packsswb xmm1,xmm1
- movd eax,xmm1
- test rax,rax
- jnz short .columnDCT
-
- ; -- AC terms all zero
-
- movq xmm0, XMM_MMWORD [MMBLOCK(0,0,rsi,SIZEOF_JCOEF)]
-
- punpcklwd xmm0,xmm0 ; xmm0=(00 00 01 01 02 02 03 03)
- psrad xmm0,(DWORD_BIT-WORD_BIT) ; xmm0=in0=(00 01 02 03)
- cvtdq2ps xmm0,xmm0 ; xmm0=in0=(00 01 02 03)
-
- mulps xmm0, XMMWORD [XMMBLOCK(0,0,rdx,SIZEOF_FLOAT_MULT_TYPE)]
-
- movaps xmm1,xmm0
- movaps xmm2,xmm0
- movaps xmm3,xmm0
-
- shufps xmm0,xmm0,0x00 ; xmm0=(00 00 00 00)
- shufps xmm1,xmm1,0x55 ; xmm1=(01 01 01 01)
- shufps xmm2,xmm2,0xAA ; xmm2=(02 02 02 02)
- shufps xmm3,xmm3,0xFF ; xmm3=(03 03 03 03)
-
- movaps XMMWORD [XMMBLOCK(0,0,rdi,SIZEOF_FAST_FLOAT)], xmm0
- movaps XMMWORD [XMMBLOCK(0,1,rdi,SIZEOF_FAST_FLOAT)], xmm0
- movaps XMMWORD [XMMBLOCK(1,0,rdi,SIZEOF_FAST_FLOAT)], xmm1
- movaps XMMWORD [XMMBLOCK(1,1,rdi,SIZEOF_FAST_FLOAT)], xmm1
- movaps XMMWORD [XMMBLOCK(2,0,rdi,SIZEOF_FAST_FLOAT)], xmm2
- movaps XMMWORD [XMMBLOCK(2,1,rdi,SIZEOF_FAST_FLOAT)], xmm2
- movaps XMMWORD [XMMBLOCK(3,0,rdi,SIZEOF_FAST_FLOAT)], xmm3
- movaps XMMWORD [XMMBLOCK(3,1,rdi,SIZEOF_FAST_FLOAT)], xmm3
- jmp near .nextcolumn
-%endif
-.columnDCT:
-
- ; -- Even part
-
- movq xmm0, XMM_MMWORD [MMBLOCK(0,0,rsi,SIZEOF_JCOEF)]
- movq xmm1, XMM_MMWORD [MMBLOCK(2,0,rsi,SIZEOF_JCOEF)]
- movq xmm2, XMM_MMWORD [MMBLOCK(4,0,rsi,SIZEOF_JCOEF)]
- movq xmm3, XMM_MMWORD [MMBLOCK(6,0,rsi,SIZEOF_JCOEF)]
-
- punpcklwd xmm0,xmm0 ; xmm0=(00 00 01 01 02 02 03 03)
- punpcklwd xmm1,xmm1 ; xmm1=(20 20 21 21 22 22 23 23)
- psrad xmm0,(DWORD_BIT-WORD_BIT) ; xmm0=in0=(00 01 02 03)
- psrad xmm1,(DWORD_BIT-WORD_BIT) ; xmm1=in2=(20 21 22 23)
- cvtdq2ps xmm0,xmm0 ; xmm0=in0=(00 01 02 03)
- cvtdq2ps xmm1,xmm1 ; xmm1=in2=(20 21 22 23)
-
- punpcklwd xmm2,xmm2 ; xmm2=(40 40 41 41 42 42 43 43)
- punpcklwd xmm3,xmm3 ; xmm3=(60 60 61 61 62 62 63 63)
- psrad xmm2,(DWORD_BIT-WORD_BIT) ; xmm2=in4=(40 41 42 43)
- psrad xmm3,(DWORD_BIT-WORD_BIT) ; xmm3=in6=(60 61 62 63)
- cvtdq2ps xmm2,xmm2 ; xmm2=in4=(40 41 42 43)
- cvtdq2ps xmm3,xmm3 ; xmm3=in6=(60 61 62 63)
-
- mulps xmm0, XMMWORD [XMMBLOCK(0,0,rdx,SIZEOF_FLOAT_MULT_TYPE)]
- mulps xmm1, XMMWORD [XMMBLOCK(2,0,rdx,SIZEOF_FLOAT_MULT_TYPE)]
- mulps xmm2, XMMWORD [XMMBLOCK(4,0,rdx,SIZEOF_FLOAT_MULT_TYPE)]
- mulps xmm3, XMMWORD [XMMBLOCK(6,0,rdx,SIZEOF_FLOAT_MULT_TYPE)]
-
- movaps xmm4,xmm0
- movaps xmm5,xmm1
- subps xmm0,xmm2 ; xmm0=tmp11
- subps xmm1,xmm3
- addps xmm4,xmm2 ; xmm4=tmp10
- addps xmm5,xmm3 ; xmm5=tmp13
-
- mulps xmm1,[rel PD_1_414]
- subps xmm1,xmm5 ; xmm1=tmp12
-
- movaps xmm6,xmm4
- movaps xmm7,xmm0
- subps xmm4,xmm5 ; xmm4=tmp3
- subps xmm0,xmm1 ; xmm0=tmp2
- addps xmm6,xmm5 ; xmm6=tmp0
- addps xmm7,xmm1 ; xmm7=tmp1
-
- movaps XMMWORD [wk(1)], xmm4 ; tmp3
- movaps XMMWORD [wk(0)], xmm0 ; tmp2
-
- ; -- Odd part
-
- movq xmm2, XMM_MMWORD [MMBLOCK(1,0,rsi,SIZEOF_JCOEF)]
- movq xmm3, XMM_MMWORD [MMBLOCK(3,0,rsi,SIZEOF_JCOEF)]
- movq xmm5, XMM_MMWORD [MMBLOCK(5,0,rsi,SIZEOF_JCOEF)]
- movq xmm1, XMM_MMWORD [MMBLOCK(7,0,rsi,SIZEOF_JCOEF)]
-
- punpcklwd xmm2,xmm2 ; xmm2=(10 10 11 11 12 12 13 13)
- punpcklwd xmm3,xmm3 ; xmm3=(30 30 31 31 32 32 33 33)
- psrad xmm2,(DWORD_BIT-WORD_BIT) ; xmm2=in1=(10 11 12 13)
- psrad xmm3,(DWORD_BIT-WORD_BIT) ; xmm3=in3=(30 31 32 33)
- cvtdq2ps xmm2,xmm2 ; xmm2=in1=(10 11 12 13)
- cvtdq2ps xmm3,xmm3 ; xmm3=in3=(30 31 32 33)
-
- punpcklwd xmm5,xmm5 ; xmm5=(50 50 51 51 52 52 53 53)
- punpcklwd xmm1,xmm1 ; xmm1=(70 70 71 71 72 72 73 73)
- psrad xmm5,(DWORD_BIT-WORD_BIT) ; xmm5=in5=(50 51 52 53)
- psrad xmm1,(DWORD_BIT-WORD_BIT) ; xmm1=in7=(70 71 72 73)
- cvtdq2ps xmm5,xmm5 ; xmm5=in5=(50 51 52 53)
- cvtdq2ps xmm1,xmm1 ; xmm1=in7=(70 71 72 73)
-
- mulps xmm2, XMMWORD [XMMBLOCK(1,0,rdx,SIZEOF_FLOAT_MULT_TYPE)]
- mulps xmm3, XMMWORD [XMMBLOCK(3,0,rdx,SIZEOF_FLOAT_MULT_TYPE)]
- mulps xmm5, XMMWORD [XMMBLOCK(5,0,rdx,SIZEOF_FLOAT_MULT_TYPE)]
- mulps xmm1, XMMWORD [XMMBLOCK(7,0,rdx,SIZEOF_FLOAT_MULT_TYPE)]
-
- movaps xmm4,xmm2
- movaps xmm0,xmm5
- addps xmm2,xmm1 ; xmm2=z11
- addps xmm5,xmm3 ; xmm5=z13
- subps xmm4,xmm1 ; xmm4=z12
- subps xmm0,xmm3 ; xmm0=z10
-
- movaps xmm1,xmm2
- subps xmm2,xmm5
- addps xmm1,xmm5 ; xmm1=tmp7
-
- mulps xmm2,[rel PD_1_414] ; xmm2=tmp11
-
- movaps xmm3,xmm0
- addps xmm0,xmm4
- mulps xmm0,[rel PD_1_847] ; xmm0=z5
- mulps xmm3,[rel PD_M2_613] ; xmm3=(z10 * -2.613125930)
- mulps xmm4,[rel PD_1_082] ; xmm4=(z12 * 1.082392200)
- addps xmm3,xmm0 ; xmm3=tmp12
- subps xmm4,xmm0 ; xmm4=tmp10
-
- ; -- Final output stage
-
- subps xmm3,xmm1 ; xmm3=tmp6
- movaps xmm5,xmm6
- movaps xmm0,xmm7
- addps xmm6,xmm1 ; xmm6=data0=(00 01 02 03)
- addps xmm7,xmm3 ; xmm7=data1=(10 11 12 13)
- subps xmm5,xmm1 ; xmm5=data7=(70 71 72 73)
- subps xmm0,xmm3 ; xmm0=data6=(60 61 62 63)
- subps xmm2,xmm3 ; xmm2=tmp5
-
- movaps xmm1,xmm6 ; transpose coefficients(phase 1)
- unpcklps xmm6,xmm7 ; xmm6=(00 10 01 11)
- unpckhps xmm1,xmm7 ; xmm1=(02 12 03 13)
- movaps xmm3,xmm0 ; transpose coefficients(phase 1)
- unpcklps xmm0,xmm5 ; xmm0=(60 70 61 71)
- unpckhps xmm3,xmm5 ; xmm3=(62 72 63 73)
-
- movaps xmm7, XMMWORD [wk(0)] ; xmm7=tmp2
- movaps xmm5, XMMWORD [wk(1)] ; xmm5=tmp3
-
- movaps XMMWORD [wk(0)], xmm0 ; wk(0)=(60 70 61 71)
- movaps XMMWORD [wk(1)], xmm3 ; wk(1)=(62 72 63 73)
-
- addps xmm4,xmm2 ; xmm4=tmp4
- movaps xmm0,xmm7
- movaps xmm3,xmm5
- addps xmm7,xmm2 ; xmm7=data2=(20 21 22 23)
- addps xmm5,xmm4 ; xmm5=data4=(40 41 42 43)
- subps xmm0,xmm2 ; xmm0=data5=(50 51 52 53)
- subps xmm3,xmm4 ; xmm3=data3=(30 31 32 33)
-
- movaps xmm2,xmm7 ; transpose coefficients(phase 1)
- unpcklps xmm7,xmm3 ; xmm7=(20 30 21 31)
- unpckhps xmm2,xmm3 ; xmm2=(22 32 23 33)
- movaps xmm4,xmm5 ; transpose coefficients(phase 1)
- unpcklps xmm5,xmm0 ; xmm5=(40 50 41 51)
- unpckhps xmm4,xmm0 ; xmm4=(42 52 43 53)
-
- movaps xmm3,xmm6 ; transpose coefficients(phase 2)
- unpcklps2 xmm6,xmm7 ; xmm6=(00 10 20 30)
- unpckhps2 xmm3,xmm7 ; xmm3=(01 11 21 31)
- movaps xmm0,xmm1 ; transpose coefficients(phase 2)
- unpcklps2 xmm1,xmm2 ; xmm1=(02 12 22 32)
- unpckhps2 xmm0,xmm2 ; xmm0=(03 13 23 33)
-
- movaps xmm7, XMMWORD [wk(0)] ; xmm7=(60 70 61 71)
- movaps xmm2, XMMWORD [wk(1)] ; xmm2=(62 72 63 73)
-
- movaps XMMWORD [XMMBLOCK(0,0,rdi,SIZEOF_FAST_FLOAT)], xmm6
- movaps XMMWORD [XMMBLOCK(1,0,rdi,SIZEOF_FAST_FLOAT)], xmm3
- movaps XMMWORD [XMMBLOCK(2,0,rdi,SIZEOF_FAST_FLOAT)], xmm1
- movaps XMMWORD [XMMBLOCK(3,0,rdi,SIZEOF_FAST_FLOAT)], xmm0
-
- movaps xmm6,xmm5 ; transpose coefficients(phase 2)
- unpcklps2 xmm5,xmm7 ; xmm5=(40 50 60 70)
- unpckhps2 xmm6,xmm7 ; xmm6=(41 51 61 71)
- movaps xmm3,xmm4 ; transpose coefficients(phase 2)
- unpcklps2 xmm4,xmm2 ; xmm4=(42 52 62 72)
- unpckhps2 xmm3,xmm2 ; xmm3=(43 53 63 73)
-
- movaps XMMWORD [XMMBLOCK(0,1,rdi,SIZEOF_FAST_FLOAT)], xmm5
- movaps XMMWORD [XMMBLOCK(1,1,rdi,SIZEOF_FAST_FLOAT)], xmm6
- movaps XMMWORD [XMMBLOCK(2,1,rdi,SIZEOF_FAST_FLOAT)], xmm4
- movaps XMMWORD [XMMBLOCK(3,1,rdi,SIZEOF_FAST_FLOAT)], xmm3
-
-.nextcolumn:
- add rsi, byte 4*SIZEOF_JCOEF ; coef_block
- add rdx, byte 4*SIZEOF_FLOAT_MULT_TYPE ; quantptr
- add rdi, 4*DCTSIZE*SIZEOF_FAST_FLOAT ; wsptr
- dec rcx ; ctr
- jnz near .columnloop
-
- ; -- Prefetch the next coefficient block
-
- prefetchnta [rsi + (DCTSIZE2-8)*SIZEOF_JCOEF + 0*32]
- prefetchnta [rsi + (DCTSIZE2-8)*SIZEOF_JCOEF + 1*32]
- prefetchnta [rsi + (DCTSIZE2-8)*SIZEOF_JCOEF + 2*32]
- prefetchnta [rsi + (DCTSIZE2-8)*SIZEOF_JCOEF + 3*32]
-
- ; ---- Pass 2: process rows from work array, store into output array.
-
- mov rax, [original_rbp]
- lea rsi, [workspace] ; FAST_FLOAT *wsptr
- mov rdi, r12 ; (JSAMPROW *)
- mov eax, r13d
- mov rcx, DCTSIZE/4 ; ctr
-.rowloop:
-
- ; -- Even part
-
- movaps xmm0, XMMWORD [XMMBLOCK(0,0,rsi,SIZEOF_FAST_FLOAT)]
- movaps xmm1, XMMWORD [XMMBLOCK(2,0,rsi,SIZEOF_FAST_FLOAT)]
- movaps xmm2, XMMWORD [XMMBLOCK(4,0,rsi,SIZEOF_FAST_FLOAT)]
- movaps xmm3, XMMWORD [XMMBLOCK(6,0,rsi,SIZEOF_FAST_FLOAT)]
-
- movaps xmm4,xmm0
- movaps xmm5,xmm1
- subps xmm0,xmm2 ; xmm0=tmp11
- subps xmm1,xmm3
- addps xmm4,xmm2 ; xmm4=tmp10
- addps xmm5,xmm3 ; xmm5=tmp13
-
- mulps xmm1,[rel PD_1_414]
- subps xmm1,xmm5 ; xmm1=tmp12
-
- movaps xmm6,xmm4
- movaps xmm7,xmm0
- subps xmm4,xmm5 ; xmm4=tmp3
- subps xmm0,xmm1 ; xmm0=tmp2
- addps xmm6,xmm5 ; xmm6=tmp0
- addps xmm7,xmm1 ; xmm7=tmp1
-
- movaps XMMWORD [wk(1)], xmm4 ; tmp3
- movaps XMMWORD [wk(0)], xmm0 ; tmp2
-
- ; -- Odd part
-
- movaps xmm2, XMMWORD [XMMBLOCK(1,0,rsi,SIZEOF_FAST_FLOAT)]
- movaps xmm3, XMMWORD [XMMBLOCK(3,0,rsi,SIZEOF_FAST_FLOAT)]
- movaps xmm5, XMMWORD [XMMBLOCK(5,0,rsi,SIZEOF_FAST_FLOAT)]
- movaps xmm1, XMMWORD [XMMBLOCK(7,0,rsi,SIZEOF_FAST_FLOAT)]
-
- movaps xmm4,xmm2
- movaps xmm0,xmm5
- addps xmm2,xmm1 ; xmm2=z11
- addps xmm5,xmm3 ; xmm5=z13
- subps xmm4,xmm1 ; xmm4=z12
- subps xmm0,xmm3 ; xmm0=z10
-
- movaps xmm1,xmm2
- subps xmm2,xmm5
- addps xmm1,xmm5 ; xmm1=tmp7
-
- mulps xmm2,[rel PD_1_414] ; xmm2=tmp11
-
- movaps xmm3,xmm0
- addps xmm0,xmm4
- mulps xmm0,[rel PD_1_847] ; xmm0=z5
- mulps xmm3,[rel PD_M2_613] ; xmm3=(z10 * -2.613125930)
- mulps xmm4,[rel PD_1_082] ; xmm4=(z12 * 1.082392200)
- addps xmm3,xmm0 ; xmm3=tmp12
- subps xmm4,xmm0 ; xmm4=tmp10
-
- ; -- Final output stage
-
- subps xmm3,xmm1 ; xmm3=tmp6
- movaps xmm5,xmm6
- movaps xmm0,xmm7
- addps xmm6,xmm1 ; xmm6=data0=(00 10 20 30)
- addps xmm7,xmm3 ; xmm7=data1=(01 11 21 31)
- subps xmm5,xmm1 ; xmm5=data7=(07 17 27 37)
- subps xmm0,xmm3 ; xmm0=data6=(06 16 26 36)
- subps xmm2,xmm3 ; xmm2=tmp5
-
- movaps xmm1,[rel PD_RNDINT_MAGIC] ; xmm1=[rel PD_RNDINT_MAGIC]
- pcmpeqd xmm3,xmm3
- psrld xmm3,WORD_BIT ; xmm3={0xFFFF 0x0000 0xFFFF 0x0000 ..}
-
- addps xmm6,xmm1 ; xmm6=roundint(data0/8)=(00 ** 10 ** 20 ** 30 **)
- addps xmm7,xmm1 ; xmm7=roundint(data1/8)=(01 ** 11 ** 21 ** 31 **)
- addps xmm0,xmm1 ; xmm0=roundint(data6/8)=(06 ** 16 ** 26 ** 36 **)
- addps xmm5,xmm1 ; xmm5=roundint(data7/8)=(07 ** 17 ** 27 ** 37 **)
-
- pand xmm6,xmm3 ; xmm6=(00 -- 10 -- 20 -- 30 --)
- pslld xmm7,WORD_BIT ; xmm7=(-- 01 -- 11 -- 21 -- 31)
- pand xmm0,xmm3 ; xmm0=(06 -- 16 -- 26 -- 36 --)
- pslld xmm5,WORD_BIT ; xmm5=(-- 07 -- 17 -- 27 -- 37)
- por xmm6,xmm7 ; xmm6=(00 01 10 11 20 21 30 31)
- por xmm0,xmm5 ; xmm0=(06 07 16 17 26 27 36 37)
-
- movaps xmm1, XMMWORD [wk(0)] ; xmm1=tmp2
- movaps xmm3, XMMWORD [wk(1)] ; xmm3=tmp3
-
- addps xmm4,xmm2 ; xmm4=tmp4
- movaps xmm7,xmm1
- movaps xmm5,xmm3
- addps xmm1,xmm2 ; xmm1=data2=(02 12 22 32)
- addps xmm3,xmm4 ; xmm3=data4=(04 14 24 34)
- subps xmm7,xmm2 ; xmm7=data5=(05 15 25 35)
- subps xmm5,xmm4 ; xmm5=data3=(03 13 23 33)
-
- movaps xmm2,[rel PD_RNDINT_MAGIC] ; xmm2=[rel PD_RNDINT_MAGIC]
- pcmpeqd xmm4,xmm4
- psrld xmm4,WORD_BIT ; xmm4={0xFFFF 0x0000 0xFFFF 0x0000 ..}
-
- addps xmm3,xmm2 ; xmm3=roundint(data4/8)=(04 ** 14 ** 24 ** 34 **)
- addps xmm7,xmm2 ; xmm7=roundint(data5/8)=(05 ** 15 ** 25 ** 35 **)
- addps xmm1,xmm2 ; xmm1=roundint(data2/8)=(02 ** 12 ** 22 ** 32 **)
- addps xmm5,xmm2 ; xmm5=roundint(data3/8)=(03 ** 13 ** 23 ** 33 **)
-
- pand xmm3,xmm4 ; xmm3=(04 -- 14 -- 24 -- 34 --)
- pslld xmm7,WORD_BIT ; xmm7=(-- 05 -- 15 -- 25 -- 35)
- pand xmm1,xmm4 ; xmm1=(02 -- 12 -- 22 -- 32 --)
- pslld xmm5,WORD_BIT ; xmm5=(-- 03 -- 13 -- 23 -- 33)
- por xmm3,xmm7 ; xmm3=(04 05 14 15 24 25 34 35)
- por xmm1,xmm5 ; xmm1=(02 03 12 13 22 23 32 33)
-
- movdqa xmm2,[rel PB_CENTERJSAMP] ; xmm2=[rel PB_CENTERJSAMP]
-
- packsswb xmm6,xmm3 ; xmm6=(00 01 10 11 20 21 30 31 04 05 14 15 24 25 34 35)
- packsswb xmm1,xmm0 ; xmm1=(02 03 12 13 22 23 32 33 06 07 16 17 26 27 36 37)
- paddb xmm6,xmm2
- paddb xmm1,xmm2
-
- movdqa xmm4,xmm6 ; transpose coefficients(phase 2)
- punpcklwd xmm6,xmm1 ; xmm6=(00 01 02 03 10 11 12 13 20 21 22 23 30 31 32 33)
- punpckhwd xmm4,xmm1 ; xmm4=(04 05 06 07 14 15 16 17 24 25 26 27 34 35 36 37)
-
- movdqa xmm7,xmm6 ; transpose coefficients(phase 3)
- punpckldq xmm6,xmm4 ; xmm6=(00 01 02 03 04 05 06 07 10 11 12 13 14 15 16 17)
- punpckhdq xmm7,xmm4 ; xmm7=(20 21 22 23 24 25 26 27 30 31 32 33 34 35 36 37)
-
- pshufd xmm5,xmm6,0x4E ; xmm5=(10 11 12 13 14 15 16 17 00 01 02 03 04 05 06 07)
- pshufd xmm3,xmm7,0x4E ; xmm3=(30 31 32 33 34 35 36 37 20 21 22 23 24 25 26 27)
-
- mov rdx, JSAMPROW [rdi+0*SIZEOF_JSAMPROW]
- mov rbx, JSAMPROW [rdi+2*SIZEOF_JSAMPROW]
- movq XMM_MMWORD [rdx+rax*SIZEOF_JSAMPLE], xmm6
- movq XMM_MMWORD [rbx+rax*SIZEOF_JSAMPLE], xmm7
- mov rdx, JSAMPROW [rdi+1*SIZEOF_JSAMPROW]
- mov rbx, JSAMPROW [rdi+3*SIZEOF_JSAMPROW]
- movq XMM_MMWORD [rdx+rax*SIZEOF_JSAMPLE], xmm5
- movq XMM_MMWORD [rbx+rax*SIZEOF_JSAMPLE], xmm3
-
- add rsi, byte 4*SIZEOF_FAST_FLOAT ; wsptr
- add rdi, byte 4*SIZEOF_JSAMPROW
- dec rcx ; ctr
- jnz near .rowloop
-
- pop rbx
- uncollect_args
- mov rsp,rbp ; rsp <- aligned rbp
- pop rsp ; rsp <- original rbp
- pop rbp
- ret
-
-; For some reason, the OS X linker does not honor the request to align the
-; segment unless we do this.
- align 16
diff --git a/simd/jidctflt-sse2.asm b/simd/jidctflt-sse2.asm
deleted file mode 100644
index a15a9c1..0000000
--- a/simd/jidctflt-sse2.asm
+++ /dev/null
@@ -1,497 +0,0 @@
-;
-; jidctflt.asm - floating-point IDCT (SSE & SSE2)
-;
-; Copyright 2009 Pierre Ossman <ossman@cendio.se> for Cendio AB
-;
-; Based on the x86 SIMD extension for IJG JPEG library
-; Copyright (C) 1999-2006, MIYASAKA Masaru.
-; For conditions of distribution and use, see copyright notice in jsimdext.inc
-;
-; This file should be assembled with NASM (Netwide Assembler),
-; can *not* be assembled with Microsoft's MASM or any compatible
-; assembler (including Borland's Turbo Assembler).
-; NASM is available from http://nasm.sourceforge.net/ or
-; http://sourceforge.net/project/showfiles.php?group_id=6208
-;
-; This file contains a floating-point implementation of the inverse DCT
-; (Discrete Cosine Transform). The following code is based directly on
-; the IJG's original jidctflt.c; see the jidctflt.c for more details.
-;
-; [TAB8]
-
-%include "jsimdext.inc"
-%include "jdct.inc"
-
-; --------------------------------------------------------------------------
-
-%macro unpcklps2 2 ; %1=(0 1 2 3) / %2=(4 5 6 7) => %1=(0 1 4 5)
- shufps %1,%2,0x44
-%endmacro
-
-%macro unpckhps2 2 ; %1=(0 1 2 3) / %2=(4 5 6 7) => %1=(2 3 6 7)
- shufps %1,%2,0xEE
-%endmacro
-
-; --------------------------------------------------------------------------
- SECTION SEG_CONST
-
- alignz 16
- global EXTN(jconst_idct_float_sse2)
-
-EXTN(jconst_idct_float_sse2):
-
-PD_1_414 times 4 dd 1.414213562373095048801689
-PD_1_847 times 4 dd 1.847759065022573512256366
-PD_1_082 times 4 dd 1.082392200292393968799446
-PD_M2_613 times 4 dd -2.613125929752753055713286
-PD_RNDINT_MAGIC times 4 dd 100663296.0 ; (float)(0x00C00000 << 3)
-PB_CENTERJSAMP times 16 db CENTERJSAMPLE
-
- alignz 16
-
-; --------------------------------------------------------------------------
- SECTION SEG_TEXT
- BITS 32
-;
-; Perform dequantization and inverse DCT on one block of coefficients.
-;
-; GLOBAL(void)
-; jsimd_idct_float_sse2 (void *dct_table, JCOEFPTR coef_block,
-; JSAMPARRAY output_buf, JDIMENSION output_col)
-;
-
-%define dct_table(b) (b)+8 ; void *dct_table
-%define coef_block(b) (b)+12 ; JCOEFPTR coef_block
-%define output_buf(b) (b)+16 ; JSAMPARRAY output_buf
-%define output_col(b) (b)+20 ; JDIMENSION output_col
-
-%define original_ebp ebp+0
-%define wk(i) ebp-(WK_NUM-(i))*SIZEOF_XMMWORD ; xmmword wk[WK_NUM]
-%define WK_NUM 2
-%define workspace wk(0)-DCTSIZE2*SIZEOF_FAST_FLOAT
- ; FAST_FLOAT workspace[DCTSIZE2]
-
- align 16
- global EXTN(jsimd_idct_float_sse2)
-
-EXTN(jsimd_idct_float_sse2):
- push ebp
- mov eax,esp ; eax = original ebp
- sub esp, byte 4
- and esp, byte (-SIZEOF_XMMWORD) ; align to 128 bits
- mov [esp],eax
- mov ebp,esp ; ebp = aligned ebp
- lea esp, [workspace]
- push ebx
-; push ecx ; need not be preserved
-; push edx ; need not be preserved
- push esi
- push edi
-
- get_GOT ebx ; get GOT address
-
- ; ---- Pass 1: process columns from input, store into work array.
-
-; mov eax, [original_ebp]
- mov edx, POINTER [dct_table(eax)] ; quantptr
- mov esi, JCOEFPTR [coef_block(eax)] ; inptr
- lea edi, [workspace] ; FAST_FLOAT *wsptr
- mov ecx, DCTSIZE/4 ; ctr
- alignx 16,7
-.columnloop:
-%ifndef NO_ZERO_COLUMN_TEST_FLOAT_SSE
- mov eax, DWORD [DWBLOCK(1,0,esi,SIZEOF_JCOEF)]
- or eax, DWORD [DWBLOCK(2,0,esi,SIZEOF_JCOEF)]
- jnz near .columnDCT
-
- movq xmm1, XMM_MMWORD [MMBLOCK(1,0,esi,SIZEOF_JCOEF)]
- movq xmm2, XMM_MMWORD [MMBLOCK(2,0,esi,SIZEOF_JCOEF)]
- movq xmm3, XMM_MMWORD [MMBLOCK(3,0,esi,SIZEOF_JCOEF)]
- movq xmm4, XMM_MMWORD [MMBLOCK(4,0,esi,SIZEOF_JCOEF)]
- movq xmm5, XMM_MMWORD [MMBLOCK(5,0,esi,SIZEOF_JCOEF)]
- movq xmm6, XMM_MMWORD [MMBLOCK(6,0,esi,SIZEOF_JCOEF)]
- movq xmm7, XMM_MMWORD [MMBLOCK(7,0,esi,SIZEOF_JCOEF)]
- por xmm1,xmm2
- por xmm3,xmm4
- por xmm5,xmm6
- por xmm1,xmm3
- por xmm5,xmm7
- por xmm1,xmm5
- packsswb xmm1,xmm1
- movd eax,xmm1
- test eax,eax
- jnz short .columnDCT
-
- ; -- AC terms all zero
-
- movq xmm0, XMM_MMWORD [MMBLOCK(0,0,esi,SIZEOF_JCOEF)]
-
- punpcklwd xmm0,xmm0 ; xmm0=(00 00 01 01 02 02 03 03)
- psrad xmm0,(DWORD_BIT-WORD_BIT) ; xmm0=in0=(00 01 02 03)
- cvtdq2ps xmm0,xmm0 ; xmm0=in0=(00 01 02 03)
-
- mulps xmm0, XMMWORD [XMMBLOCK(0,0,edx,SIZEOF_FLOAT_MULT_TYPE)]
-
- movaps xmm1,xmm0
- movaps xmm2,xmm0
- movaps xmm3,xmm0
-
- shufps xmm0,xmm0,0x00 ; xmm0=(00 00 00 00)
- shufps xmm1,xmm1,0x55 ; xmm1=(01 01 01 01)
- shufps xmm2,xmm2,0xAA ; xmm2=(02 02 02 02)
- shufps xmm3,xmm3,0xFF ; xmm3=(03 03 03 03)
-
- movaps XMMWORD [XMMBLOCK(0,0,edi,SIZEOF_FAST_FLOAT)], xmm0
- movaps XMMWORD [XMMBLOCK(0,1,edi,SIZEOF_FAST_FLOAT)], xmm0
- movaps XMMWORD [XMMBLOCK(1,0,edi,SIZEOF_FAST_FLOAT)], xmm1
- movaps XMMWORD [XMMBLOCK(1,1,edi,SIZEOF_FAST_FLOAT)], xmm1
- movaps XMMWORD [XMMBLOCK(2,0,edi,SIZEOF_FAST_FLOAT)], xmm2
- movaps XMMWORD [XMMBLOCK(2,1,edi,SIZEOF_FAST_FLOAT)], xmm2
- movaps XMMWORD [XMMBLOCK(3,0,edi,SIZEOF_FAST_FLOAT)], xmm3
- movaps XMMWORD [XMMBLOCK(3,1,edi,SIZEOF_FAST_FLOAT)], xmm3
- jmp near .nextcolumn
- alignx 16,7
-%endif
-.columnDCT:
-
- ; -- Even part
-
- movq xmm0, XMM_MMWORD [MMBLOCK(0,0,esi,SIZEOF_JCOEF)]
- movq xmm1, XMM_MMWORD [MMBLOCK(2,0,esi,SIZEOF_JCOEF)]
- movq xmm2, XMM_MMWORD [MMBLOCK(4,0,esi,SIZEOF_JCOEF)]
- movq xmm3, XMM_MMWORD [MMBLOCK(6,0,esi,SIZEOF_JCOEF)]
-
- punpcklwd xmm0,xmm0 ; xmm0=(00 00 01 01 02 02 03 03)
- punpcklwd xmm1,xmm1 ; xmm1=(20 20 21 21 22 22 23 23)
- psrad xmm0,(DWORD_BIT-WORD_BIT) ; xmm0=in0=(00 01 02 03)
- psrad xmm1,(DWORD_BIT-WORD_BIT) ; xmm1=in2=(20 21 22 23)
- cvtdq2ps xmm0,xmm0 ; xmm0=in0=(00 01 02 03)
- cvtdq2ps xmm1,xmm1 ; xmm1=in2=(20 21 22 23)
-
- punpcklwd xmm2,xmm2 ; xmm2=(40 40 41 41 42 42 43 43)
- punpcklwd xmm3,xmm3 ; xmm3=(60 60 61 61 62 62 63 63)
- psrad xmm2,(DWORD_BIT-WORD_BIT) ; xmm2=in4=(40 41 42 43)
- psrad xmm3,(DWORD_BIT-WORD_BIT) ; xmm3=in6=(60 61 62 63)
- cvtdq2ps xmm2,xmm2 ; xmm2=in4=(40 41 42 43)
- cvtdq2ps xmm3,xmm3 ; xmm3=in6=(60 61 62 63)
-
- mulps xmm0, XMMWORD [XMMBLOCK(0,0,edx,SIZEOF_FLOAT_MULT_TYPE)]
- mulps xmm1, XMMWORD [XMMBLOCK(2,0,edx,SIZEOF_FLOAT_MULT_TYPE)]
- mulps xmm2, XMMWORD [XMMBLOCK(4,0,edx,SIZEOF_FLOAT_MULT_TYPE)]
- mulps xmm3, XMMWORD [XMMBLOCK(6,0,edx,SIZEOF_FLOAT_MULT_TYPE)]
-
- movaps xmm4,xmm0
- movaps xmm5,xmm1
- subps xmm0,xmm2 ; xmm0=tmp11
- subps xmm1,xmm3
- addps xmm4,xmm2 ; xmm4=tmp10
- addps xmm5,xmm3 ; xmm5=tmp13
-
- mulps xmm1,[GOTOFF(ebx,PD_1_414)]
- subps xmm1,xmm5 ; xmm1=tmp12
-
- movaps xmm6,xmm4
- movaps xmm7,xmm0
- subps xmm4,xmm5 ; xmm4=tmp3
- subps xmm0,xmm1 ; xmm0=tmp2
- addps xmm6,xmm5 ; xmm6=tmp0
- addps xmm7,xmm1 ; xmm7=tmp1
-
- movaps XMMWORD [wk(1)], xmm4 ; tmp3
- movaps XMMWORD [wk(0)], xmm0 ; tmp2
-
- ; -- Odd part
-
- movq xmm2, XMM_MMWORD [MMBLOCK(1,0,esi,SIZEOF_JCOEF)]
- movq xmm3, XMM_MMWORD [MMBLOCK(3,0,esi,SIZEOF_JCOEF)]
- movq xmm5, XMM_MMWORD [MMBLOCK(5,0,esi,SIZEOF_JCOEF)]
- movq xmm1, XMM_MMWORD [MMBLOCK(7,0,esi,SIZEOF_JCOEF)]
-
- punpcklwd xmm2,xmm2 ; xmm2=(10 10 11 11 12 12 13 13)
- punpcklwd xmm3,xmm3 ; xmm3=(30 30 31 31 32 32 33 33)
- psrad xmm2,(DWORD_BIT-WORD_BIT) ; xmm2=in1=(10 11 12 13)
- psrad xmm3,(DWORD_BIT-WORD_BIT) ; xmm3=in3=(30 31 32 33)
- cvtdq2ps xmm2,xmm2 ; xmm2=in1=(10 11 12 13)
- cvtdq2ps xmm3,xmm3 ; xmm3=in3=(30 31 32 33)
-
- punpcklwd xmm5,xmm5 ; xmm5=(50 50 51 51 52 52 53 53)
- punpcklwd xmm1,xmm1 ; xmm1=(70 70 71 71 72 72 73 73)
- psrad xmm5,(DWORD_BIT-WORD_BIT) ; xmm5=in5=(50 51 52 53)
- psrad xmm1,(DWORD_BIT-WORD_BIT) ; xmm1=in7=(70 71 72 73)
- cvtdq2ps xmm5,xmm5 ; xmm5=in5=(50 51 52 53)
- cvtdq2ps xmm1,xmm1 ; xmm1=in7=(70 71 72 73)
-
- mulps xmm2, XMMWORD [XMMBLOCK(1,0,edx,SIZEOF_FLOAT_MULT_TYPE)]
- mulps xmm3, XMMWORD [XMMBLOCK(3,0,edx,SIZEOF_FLOAT_MULT_TYPE)]
- mulps xmm5, XMMWORD [XMMBLOCK(5,0,edx,SIZEOF_FLOAT_MULT_TYPE)]
- mulps xmm1, XMMWORD [XMMBLOCK(7,0,edx,SIZEOF_FLOAT_MULT_TYPE)]
-
- movaps xmm4,xmm2
- movaps xmm0,xmm5
- addps xmm2,xmm1 ; xmm2=z11
- addps xmm5,xmm3 ; xmm5=z13
- subps xmm4,xmm1 ; xmm4=z12
- subps xmm0,xmm3 ; xmm0=z10
-
- movaps xmm1,xmm2
- subps xmm2,xmm5
- addps xmm1,xmm5 ; xmm1=tmp7
-
- mulps xmm2,[GOTOFF(ebx,PD_1_414)] ; xmm2=tmp11
-
- movaps xmm3,xmm0
- addps xmm0,xmm4
- mulps xmm0,[GOTOFF(ebx,PD_1_847)] ; xmm0=z5
- mulps xmm3,[GOTOFF(ebx,PD_M2_613)] ; xmm3=(z10 * -2.613125930)
- mulps xmm4,[GOTOFF(ebx,PD_1_082)] ; xmm4=(z12 * 1.082392200)
- addps xmm3,xmm0 ; xmm3=tmp12
- subps xmm4,xmm0 ; xmm4=tmp10
-
- ; -- Final output stage
-
- subps xmm3,xmm1 ; xmm3=tmp6
- movaps xmm5,xmm6
- movaps xmm0,xmm7
- addps xmm6,xmm1 ; xmm6=data0=(00 01 02 03)
- addps xmm7,xmm3 ; xmm7=data1=(10 11 12 13)
- subps xmm5,xmm1 ; xmm5=data7=(70 71 72 73)
- subps xmm0,xmm3 ; xmm0=data6=(60 61 62 63)
- subps xmm2,xmm3 ; xmm2=tmp5
-
- movaps xmm1,xmm6 ; transpose coefficients(phase 1)
- unpcklps xmm6,xmm7 ; xmm6=(00 10 01 11)
- unpckhps xmm1,xmm7 ; xmm1=(02 12 03 13)
- movaps xmm3,xmm0 ; transpose coefficients(phase 1)
- unpcklps xmm0,xmm5 ; xmm0=(60 70 61 71)
- unpckhps xmm3,xmm5 ; xmm3=(62 72 63 73)
-
- movaps xmm7, XMMWORD [wk(0)] ; xmm7=tmp2
- movaps xmm5, XMMWORD [wk(1)] ; xmm5=tmp3
-
- movaps XMMWORD [wk(0)], xmm0 ; wk(0)=(60 70 61 71)
- movaps XMMWORD [wk(1)], xmm3 ; wk(1)=(62 72 63 73)
-
- addps xmm4,xmm2 ; xmm4=tmp4
- movaps xmm0,xmm7
- movaps xmm3,xmm5
- addps xmm7,xmm2 ; xmm7=data2=(20 21 22 23)
- addps xmm5,xmm4 ; xmm5=data4=(40 41 42 43)
- subps xmm0,xmm2 ; xmm0=data5=(50 51 52 53)
- subps xmm3,xmm4 ; xmm3=data3=(30 31 32 33)
-
- movaps xmm2,xmm7 ; transpose coefficients(phase 1)
- unpcklps xmm7,xmm3 ; xmm7=(20 30 21 31)
- unpckhps xmm2,xmm3 ; xmm2=(22 32 23 33)
- movaps xmm4,xmm5 ; transpose coefficients(phase 1)
- unpcklps xmm5,xmm0 ; xmm5=(40 50 41 51)
- unpckhps xmm4,xmm0 ; xmm4=(42 52 43 53)
-
- movaps xmm3,xmm6 ; transpose coefficients(phase 2)
- unpcklps2 xmm6,xmm7 ; xmm6=(00 10 20 30)
- unpckhps2 xmm3,xmm7 ; xmm3=(01 11 21 31)
- movaps xmm0,xmm1 ; transpose coefficients(phase 2)
- unpcklps2 xmm1,xmm2 ; xmm1=(02 12 22 32)
- unpckhps2 xmm0,xmm2 ; xmm0=(03 13 23 33)
-
- movaps xmm7, XMMWORD [wk(0)] ; xmm7=(60 70 61 71)
- movaps xmm2, XMMWORD [wk(1)] ; xmm2=(62 72 63 73)
-
- movaps XMMWORD [XMMBLOCK(0,0,edi,SIZEOF_FAST_FLOAT)], xmm6
- movaps XMMWORD [XMMBLOCK(1,0,edi,SIZEOF_FAST_FLOAT)], xmm3
- movaps XMMWORD [XMMBLOCK(2,0,edi,SIZEOF_FAST_FLOAT)], xmm1
- movaps XMMWORD [XMMBLOCK(3,0,edi,SIZEOF_FAST_FLOAT)], xmm0
-
- movaps xmm6,xmm5 ; transpose coefficients(phase 2)
- unpcklps2 xmm5,xmm7 ; xmm5=(40 50 60 70)
- unpckhps2 xmm6,xmm7 ; xmm6=(41 51 61 71)
- movaps xmm3,xmm4 ; transpose coefficients(phase 2)
- unpcklps2 xmm4,xmm2 ; xmm4=(42 52 62 72)
- unpckhps2 xmm3,xmm2 ; xmm3=(43 53 63 73)
-
- movaps XMMWORD [XMMBLOCK(0,1,edi,SIZEOF_FAST_FLOAT)], xmm5
- movaps XMMWORD [XMMBLOCK(1,1,edi,SIZEOF_FAST_FLOAT)], xmm6
- movaps XMMWORD [XMMBLOCK(2,1,edi,SIZEOF_FAST_FLOAT)], xmm4
- movaps XMMWORD [XMMBLOCK(3,1,edi,SIZEOF_FAST_FLOAT)], xmm3
-
-.nextcolumn:
- add esi, byte 4*SIZEOF_JCOEF ; coef_block
- add edx, byte 4*SIZEOF_FLOAT_MULT_TYPE ; quantptr
- add edi, 4*DCTSIZE*SIZEOF_FAST_FLOAT ; wsptr
- dec ecx ; ctr
- jnz near .columnloop
-
- ; -- Prefetch the next coefficient block
-
- prefetchnta [esi + (DCTSIZE2-8)*SIZEOF_JCOEF + 0*32]
- prefetchnta [esi + (DCTSIZE2-8)*SIZEOF_JCOEF + 1*32]
- prefetchnta [esi + (DCTSIZE2-8)*SIZEOF_JCOEF + 2*32]
- prefetchnta [esi + (DCTSIZE2-8)*SIZEOF_JCOEF + 3*32]
-
- ; ---- Pass 2: process rows from work array, store into output array.
-
- mov eax, [original_ebp]
- lea esi, [workspace] ; FAST_FLOAT *wsptr
- mov edi, JSAMPARRAY [output_buf(eax)] ; (JSAMPROW *)
- mov eax, JDIMENSION [output_col(eax)]
- mov ecx, DCTSIZE/4 ; ctr
- alignx 16,7
-.rowloop:
-
- ; -- Even part
-
- movaps xmm0, XMMWORD [XMMBLOCK(0,0,esi,SIZEOF_FAST_FLOAT)]
- movaps xmm1, XMMWORD [XMMBLOCK(2,0,esi,SIZEOF_FAST_FLOAT)]
- movaps xmm2, XMMWORD [XMMBLOCK(4,0,esi,SIZEOF_FAST_FLOAT)]
- movaps xmm3, XMMWORD [XMMBLOCK(6,0,esi,SIZEOF_FAST_FLOAT)]
-
- movaps xmm4,xmm0
- movaps xmm5,xmm1
- subps xmm0,xmm2 ; xmm0=tmp11
- subps xmm1,xmm3
- addps xmm4,xmm2 ; xmm4=tmp10
- addps xmm5,xmm3 ; xmm5=tmp13
-
- mulps xmm1,[GOTOFF(ebx,PD_1_414)]
- subps xmm1,xmm5 ; xmm1=tmp12
-
- movaps xmm6,xmm4
- movaps xmm7,xmm0
- subps xmm4,xmm5 ; xmm4=tmp3
- subps xmm0,xmm1 ; xmm0=tmp2
- addps xmm6,xmm5 ; xmm6=tmp0
- addps xmm7,xmm1 ; xmm7=tmp1
-
- movaps XMMWORD [wk(1)], xmm4 ; tmp3
- movaps XMMWORD [wk(0)], xmm0 ; tmp2
-
- ; -- Odd part
-
- movaps xmm2, XMMWORD [XMMBLOCK(1,0,esi,SIZEOF_FAST_FLOAT)]
- movaps xmm3, XMMWORD [XMMBLOCK(3,0,esi,SIZEOF_FAST_FLOAT)]
- movaps xmm5, XMMWORD [XMMBLOCK(5,0,esi,SIZEOF_FAST_FLOAT)]
- movaps xmm1, XMMWORD [XMMBLOCK(7,0,esi,SIZEOF_FAST_FLOAT)]
-
- movaps xmm4,xmm2
- movaps xmm0,xmm5
- addps xmm2,xmm1 ; xmm2=z11
- addps xmm5,xmm3 ; xmm5=z13
- subps xmm4,xmm1 ; xmm4=z12
- subps xmm0,xmm3 ; xmm0=z10
-
- movaps xmm1,xmm2
- subps xmm2,xmm5
- addps xmm1,xmm5 ; xmm1=tmp7
-
- mulps xmm2,[GOTOFF(ebx,PD_1_414)] ; xmm2=tmp11
-
- movaps xmm3,xmm0
- addps xmm0,xmm4
- mulps xmm0,[GOTOFF(ebx,PD_1_847)] ; xmm0=z5
- mulps xmm3,[GOTOFF(ebx,PD_M2_613)] ; xmm3=(z10 * -2.613125930)
- mulps xmm4,[GOTOFF(ebx,PD_1_082)] ; xmm4=(z12 * 1.082392200)
- addps xmm3,xmm0 ; xmm3=tmp12
- subps xmm4,xmm0 ; xmm4=tmp10
-
- ; -- Final output stage
-
- subps xmm3,xmm1 ; xmm3=tmp6
- movaps xmm5,xmm6
- movaps xmm0,xmm7
- addps xmm6,xmm1 ; xmm6=data0=(00 10 20 30)
- addps xmm7,xmm3 ; xmm7=data1=(01 11 21 31)
- subps xmm5,xmm1 ; xmm5=data7=(07 17 27 37)
- subps xmm0,xmm3 ; xmm0=data6=(06 16 26 36)
- subps xmm2,xmm3 ; xmm2=tmp5
-
- movaps xmm1,[GOTOFF(ebx,PD_RNDINT_MAGIC)] ; xmm1=[PD_RNDINT_MAGIC]
- pcmpeqd xmm3,xmm3
- psrld xmm3,WORD_BIT ; xmm3={0xFFFF 0x0000 0xFFFF 0x0000 ..}
-
- addps xmm6,xmm1 ; xmm6=roundint(data0/8)=(00 ** 10 ** 20 ** 30 **)
- addps xmm7,xmm1 ; xmm7=roundint(data1/8)=(01 ** 11 ** 21 ** 31 **)
- addps xmm0,xmm1 ; xmm0=roundint(data6/8)=(06 ** 16 ** 26 ** 36 **)
- addps xmm5,xmm1 ; xmm5=roundint(data7/8)=(07 ** 17 ** 27 ** 37 **)
-
- pand xmm6,xmm3 ; xmm6=(00 -- 10 -- 20 -- 30 --)
- pslld xmm7,WORD_BIT ; xmm7=(-- 01 -- 11 -- 21 -- 31)
- pand xmm0,xmm3 ; xmm0=(06 -- 16 -- 26 -- 36 --)
- pslld xmm5,WORD_BIT ; xmm5=(-- 07 -- 17 -- 27 -- 37)
- por xmm6,xmm7 ; xmm6=(00 01 10 11 20 21 30 31)
- por xmm0,xmm5 ; xmm0=(06 07 16 17 26 27 36 37)
-
- movaps xmm1, XMMWORD [wk(0)] ; xmm1=tmp2
- movaps xmm3, XMMWORD [wk(1)] ; xmm3=tmp3
-
- addps xmm4,xmm2 ; xmm4=tmp4
- movaps xmm7,xmm1
- movaps xmm5,xmm3
- addps xmm1,xmm2 ; xmm1=data2=(02 12 22 32)
- addps xmm3,xmm4 ; xmm3=data4=(04 14 24 34)
- subps xmm7,xmm2 ; xmm7=data5=(05 15 25 35)
- subps xmm5,xmm4 ; xmm5=data3=(03 13 23 33)
-
- movaps xmm2,[GOTOFF(ebx,PD_RNDINT_MAGIC)] ; xmm2=[PD_RNDINT_MAGIC]
- pcmpeqd xmm4,xmm4
- psrld xmm4,WORD_BIT ; xmm4={0xFFFF 0x0000 0xFFFF 0x0000 ..}
-
- addps xmm3,xmm2 ; xmm3=roundint(data4/8)=(04 ** 14 ** 24 ** 34 **)
- addps xmm7,xmm2 ; xmm7=roundint(data5/8)=(05 ** 15 ** 25 ** 35 **)
- addps xmm1,xmm2 ; xmm1=roundint(data2/8)=(02 ** 12 ** 22 ** 32 **)
- addps xmm5,xmm2 ; xmm5=roundint(data3/8)=(03 ** 13 ** 23 ** 33 **)
-
- pand xmm3,xmm4 ; xmm3=(04 -- 14 -- 24 -- 34 --)
- pslld xmm7,WORD_BIT ; xmm7=(-- 05 -- 15 -- 25 -- 35)
- pand xmm1,xmm4 ; xmm1=(02 -- 12 -- 22 -- 32 --)
- pslld xmm5,WORD_BIT ; xmm5=(-- 03 -- 13 -- 23 -- 33)
- por xmm3,xmm7 ; xmm3=(04 05 14 15 24 25 34 35)
- por xmm1,xmm5 ; xmm1=(02 03 12 13 22 23 32 33)
-
- movdqa xmm2,[GOTOFF(ebx,PB_CENTERJSAMP)] ; xmm2=[PB_CENTERJSAMP]
-
- packsswb xmm6,xmm3 ; xmm6=(00 01 10 11 20 21 30 31 04 05 14 15 24 25 34 35)
- packsswb xmm1,xmm0 ; xmm1=(02 03 12 13 22 23 32 33 06 07 16 17 26 27 36 37)
- paddb xmm6,xmm2
- paddb xmm1,xmm2
-
- movdqa xmm4,xmm6 ; transpose coefficients(phase 2)
- punpcklwd xmm6,xmm1 ; xmm6=(00 01 02 03 10 11 12 13 20 21 22 23 30 31 32 33)
- punpckhwd xmm4,xmm1 ; xmm4=(04 05 06 07 14 15 16 17 24 25 26 27 34 35 36 37)
-
- movdqa xmm7,xmm6 ; transpose coefficients(phase 3)
- punpckldq xmm6,xmm4 ; xmm6=(00 01 02 03 04 05 06 07 10 11 12 13 14 15 16 17)
- punpckhdq xmm7,xmm4 ; xmm7=(20 21 22 23 24 25 26 27 30 31 32 33 34 35 36 37)
-
- pshufd xmm5,xmm6,0x4E ; xmm5=(10 11 12 13 14 15 16 17 00 01 02 03 04 05 06 07)
- pshufd xmm3,xmm7,0x4E ; xmm3=(30 31 32 33 34 35 36 37 20 21 22 23 24 25 26 27)
-
- pushpic ebx ; save GOT address
-
- mov edx, JSAMPROW [edi+0*SIZEOF_JSAMPROW]
- mov ebx, JSAMPROW [edi+2*SIZEOF_JSAMPROW]
- movq XMM_MMWORD [edx+eax*SIZEOF_JSAMPLE], xmm6
- movq XMM_MMWORD [ebx+eax*SIZEOF_JSAMPLE], xmm7
- mov edx, JSAMPROW [edi+1*SIZEOF_JSAMPROW]
- mov ebx, JSAMPROW [edi+3*SIZEOF_JSAMPROW]
- movq XMM_MMWORD [edx+eax*SIZEOF_JSAMPLE], xmm5
- movq XMM_MMWORD [ebx+eax*SIZEOF_JSAMPLE], xmm3
-
- poppic ebx ; restore GOT address
-
- add esi, byte 4*SIZEOF_FAST_FLOAT ; wsptr
- add edi, byte 4*SIZEOF_JSAMPROW
- dec ecx ; ctr
- jnz near .rowloop
-
- pop edi
- pop esi
-; pop edx ; need not be preserved
-; pop ecx ; need not be preserved
- pop ebx
- mov esp,ebp ; esp <- aligned ebp
- pop esp ; esp <- original ebp
- pop ebp
- ret
-
-; For some reason, the OS X linker does not honor the request to align the
-; segment unless we do this.
- align 16
diff --git a/simd/jidctfst-sse2-64.asm b/simd/jidctfst-sse2-64.asm
deleted file mode 100644
index 4884642..0000000
--- a/simd/jidctfst-sse2-64.asm
+++ /dev/null
@@ -1,491 +0,0 @@
-;
-; jidctfst.asm - fast integer IDCT (64-bit SSE2)
-;
-; Copyright 2009 Pierre Ossman <ossman@cendio.se> for Cendio AB
-; Copyright (C) 2009, D. R. Commander.
-;
-; Based on the x86 SIMD extension for IJG JPEG library
-; Copyright (C) 1999-2006, MIYASAKA Masaru.
-; For conditions of distribution and use, see copyright notice in jsimdext.inc
-;
-; This file should be assembled with NASM (Netwide Assembler),
-; can *not* be assembled with Microsoft's MASM or any compatible
-; assembler (including Borland's Turbo Assembler).
-; NASM is available from http://nasm.sourceforge.net/ or
-; http://sourceforge.net/project/showfiles.php?group_id=6208
-;
-; This file contains a fast, not so accurate integer implementation of
-; the inverse DCT (Discrete Cosine Transform). The following code is
-; based directly on the IJG's original jidctfst.c; see the jidctfst.c
-; for more details.
-;
-; [TAB8]
-
-%include "jsimdext.inc"
-%include "jdct.inc"
-
-; --------------------------------------------------------------------------
-
-%define CONST_BITS 8 ; 14 is also OK.
-%define PASS1_BITS 2
-
-%if IFAST_SCALE_BITS != PASS1_BITS
-%error "'IFAST_SCALE_BITS' must be equal to 'PASS1_BITS'."
-%endif
-
-%if CONST_BITS == 8
-F_1_082 equ 277 ; FIX(1.082392200)
-F_1_414 equ 362 ; FIX(1.414213562)
-F_1_847 equ 473 ; FIX(1.847759065)
-F_2_613 equ 669 ; FIX(2.613125930)
-F_1_613 equ (F_2_613 - 256) ; FIX(2.613125930) - FIX(1)
-%else
-; NASM cannot do compile-time arithmetic on floating-point constants.
-%define DESCALE(x,n) (((x)+(1<<((n)-1)))>>(n))
-F_1_082 equ DESCALE(1162209775,30-CONST_BITS) ; FIX(1.082392200)
-F_1_414 equ DESCALE(1518500249,30-CONST_BITS) ; FIX(1.414213562)
-F_1_847 equ DESCALE(1984016188,30-CONST_BITS) ; FIX(1.847759065)
-F_2_613 equ DESCALE(2805822602,30-CONST_BITS) ; FIX(2.613125930)
-F_1_613 equ (F_2_613 - (1 << CONST_BITS)) ; FIX(2.613125930) - FIX(1)
-%endif
-
-; --------------------------------------------------------------------------
- SECTION SEG_CONST
-
-; PRE_MULTIPLY_SCALE_BITS <= 2 (to avoid overflow)
-; CONST_BITS + CONST_SHIFT + PRE_MULTIPLY_SCALE_BITS == 16 (for pmulhw)
-
-%define PRE_MULTIPLY_SCALE_BITS 2
-%define CONST_SHIFT (16 - PRE_MULTIPLY_SCALE_BITS - CONST_BITS)
-
- alignz 16
- global EXTN(jconst_idct_ifast_sse2)
-
-EXTN(jconst_idct_ifast_sse2):
-
-PW_F1414 times 8 dw F_1_414 << CONST_SHIFT
-PW_F1847 times 8 dw F_1_847 << CONST_SHIFT
-PW_MF1613 times 8 dw -F_1_613 << CONST_SHIFT
-PW_F1082 times 8 dw F_1_082 << CONST_SHIFT
-PB_CENTERJSAMP times 16 db CENTERJSAMPLE
-
- alignz 16
-
-; --------------------------------------------------------------------------
- SECTION SEG_TEXT
- BITS 64
-;
-; Perform dequantization and inverse DCT on one block of coefficients.
-;
-; GLOBAL(void)
-; jsimd_idct_ifast_sse2 (void *dct_table, JCOEFPTR coef_block,
-; JSAMPARRAY output_buf, JDIMENSION output_col)
-;
-
-; r10 = jpeg_component_info *compptr
-; r11 = JCOEFPTR coef_block
-; r12 = JSAMPARRAY output_buf
-; r13 = JDIMENSION output_col
-
-%define original_rbp rbp+0
-%define wk(i) rbp-(WK_NUM-(i))*SIZEOF_XMMWORD ; xmmword wk[WK_NUM]
-%define WK_NUM 2
-
- align 16
- global EXTN(jsimd_idct_ifast_sse2)
-
-EXTN(jsimd_idct_ifast_sse2):
- push rbp
- mov rax,rsp ; rax = original rbp
- sub rsp, byte 4
- and rsp, byte (-SIZEOF_XMMWORD) ; align to 128 bits
- mov [rsp],rax
- mov rbp,rsp ; rbp = aligned rbp
- lea rsp, [wk(0)]
- collect_args
-
- ; ---- Pass 1: process columns from input.
-
- mov rdx, r10 ; quantptr
- mov rsi, r11 ; inptr
-
-%ifndef NO_ZERO_COLUMN_TEST_IFAST_SSE2
- mov eax, DWORD [DWBLOCK(1,0,rsi,SIZEOF_JCOEF)]
- or eax, DWORD [DWBLOCK(2,0,rsi,SIZEOF_JCOEF)]
- jnz near .columnDCT
-
- movdqa xmm0, XMMWORD [XMMBLOCK(1,0,rsi,SIZEOF_JCOEF)]
- movdqa xmm1, XMMWORD [XMMBLOCK(2,0,rsi,SIZEOF_JCOEF)]
- por xmm0, XMMWORD [XMMBLOCK(3,0,rsi,SIZEOF_JCOEF)]
- por xmm1, XMMWORD [XMMBLOCK(4,0,rsi,SIZEOF_JCOEF)]
- por xmm0, XMMWORD [XMMBLOCK(5,0,rsi,SIZEOF_JCOEF)]
- por xmm1, XMMWORD [XMMBLOCK(6,0,rsi,SIZEOF_JCOEF)]
- por xmm0, XMMWORD [XMMBLOCK(7,0,rsi,SIZEOF_JCOEF)]
- por xmm1,xmm0
- packsswb xmm1,xmm1
- packsswb xmm1,xmm1
- movd eax,xmm1
- test rax,rax
- jnz short .columnDCT
-
- ; -- AC terms all zero
-
- movdqa xmm0, XMMWORD [XMMBLOCK(0,0,rsi,SIZEOF_JCOEF)]
- pmullw xmm0, XMMWORD [XMMBLOCK(0,0,rdx,SIZEOF_ISLOW_MULT_TYPE)]
-
- movdqa xmm7,xmm0 ; xmm0=in0=(00 01 02 03 04 05 06 07)
- punpcklwd xmm0,xmm0 ; xmm0=(00 00 01 01 02 02 03 03)
- punpckhwd xmm7,xmm7 ; xmm7=(04 04 05 05 06 06 07 07)
-
- pshufd xmm6,xmm0,0x00 ; xmm6=col0=(00 00 00 00 00 00 00 00)
- pshufd xmm2,xmm0,0x55 ; xmm2=col1=(01 01 01 01 01 01 01 01)
- pshufd xmm5,xmm0,0xAA ; xmm5=col2=(02 02 02 02 02 02 02 02)
- pshufd xmm0,xmm0,0xFF ; xmm0=col3=(03 03 03 03 03 03 03 03)
- pshufd xmm1,xmm7,0x00 ; xmm1=col4=(04 04 04 04 04 04 04 04)
- pshufd xmm4,xmm7,0x55 ; xmm4=col5=(05 05 05 05 05 05 05 05)
- pshufd xmm3,xmm7,0xAA ; xmm3=col6=(06 06 06 06 06 06 06 06)
- pshufd xmm7,xmm7,0xFF ; xmm7=col7=(07 07 07 07 07 07 07 07)
-
- movdqa XMMWORD [wk(0)], xmm2 ; wk(0)=col1
- movdqa XMMWORD [wk(1)], xmm0 ; wk(1)=col3
- jmp near .column_end
-%endif
-.columnDCT:
-
- ; -- Even part
-
- movdqa xmm0, XMMWORD [XMMBLOCK(0,0,rsi,SIZEOF_JCOEF)]
- movdqa xmm1, XMMWORD [XMMBLOCK(2,0,rsi,SIZEOF_JCOEF)]
- pmullw xmm0, XMMWORD [XMMBLOCK(0,0,rdx,SIZEOF_IFAST_MULT_TYPE)]
- pmullw xmm1, XMMWORD [XMMBLOCK(2,0,rdx,SIZEOF_IFAST_MULT_TYPE)]
- movdqa xmm2, XMMWORD [XMMBLOCK(4,0,rsi,SIZEOF_JCOEF)]
- movdqa xmm3, XMMWORD [XMMBLOCK(6,0,rsi,SIZEOF_JCOEF)]
- pmullw xmm2, XMMWORD [XMMBLOCK(4,0,rdx,SIZEOF_IFAST_MULT_TYPE)]
- pmullw xmm3, XMMWORD [XMMBLOCK(6,0,rdx,SIZEOF_IFAST_MULT_TYPE)]
-
- movdqa xmm4,xmm0
- movdqa xmm5,xmm1
- psubw xmm0,xmm2 ; xmm0=tmp11
- psubw xmm1,xmm3
- paddw xmm4,xmm2 ; xmm4=tmp10
- paddw xmm5,xmm3 ; xmm5=tmp13
-
- psllw xmm1,PRE_MULTIPLY_SCALE_BITS
- pmulhw xmm1,[rel PW_F1414]
- psubw xmm1,xmm5 ; xmm1=tmp12
-
- movdqa xmm6,xmm4
- movdqa xmm7,xmm0
- psubw xmm4,xmm5 ; xmm4=tmp3
- psubw xmm0,xmm1 ; xmm0=tmp2
- paddw xmm6,xmm5 ; xmm6=tmp0
- paddw xmm7,xmm1 ; xmm7=tmp1
-
- movdqa XMMWORD [wk(1)], xmm4 ; wk(1)=tmp3
- movdqa XMMWORD [wk(0)], xmm0 ; wk(0)=tmp2
-
- ; -- Odd part
-
- movdqa xmm2, XMMWORD [XMMBLOCK(1,0,rsi,SIZEOF_JCOEF)]
- movdqa xmm3, XMMWORD [XMMBLOCK(3,0,rsi,SIZEOF_JCOEF)]
- pmullw xmm2, XMMWORD [XMMBLOCK(1,0,rdx,SIZEOF_IFAST_MULT_TYPE)]
- pmullw xmm3, XMMWORD [XMMBLOCK(3,0,rdx,SIZEOF_IFAST_MULT_TYPE)]
- movdqa xmm5, XMMWORD [XMMBLOCK(5,0,rsi,SIZEOF_JCOEF)]
- movdqa xmm1, XMMWORD [XMMBLOCK(7,0,rsi,SIZEOF_JCOEF)]
- pmullw xmm5, XMMWORD [XMMBLOCK(5,0,rdx,SIZEOF_IFAST_MULT_TYPE)]
- pmullw xmm1, XMMWORD [XMMBLOCK(7,0,rdx,SIZEOF_IFAST_MULT_TYPE)]
-
- movdqa xmm4,xmm2
- movdqa xmm0,xmm5
- psubw xmm2,xmm1 ; xmm2=z12
- psubw xmm5,xmm3 ; xmm5=z10
- paddw xmm4,xmm1 ; xmm4=z11
- paddw xmm0,xmm3 ; xmm0=z13
-
- movdqa xmm1,xmm5 ; xmm1=z10(unscaled)
- psllw xmm2,PRE_MULTIPLY_SCALE_BITS
- psllw xmm5,PRE_MULTIPLY_SCALE_BITS
-
- movdqa xmm3,xmm4
- psubw xmm4,xmm0
- paddw xmm3,xmm0 ; xmm3=tmp7
-
- psllw xmm4,PRE_MULTIPLY_SCALE_BITS
- pmulhw xmm4,[rel PW_F1414] ; xmm4=tmp11
-
- ; To avoid overflow...
- ;
- ; (Original)
- ; tmp12 = -2.613125930 * z10 + z5;
- ;
- ; (This implementation)
- ; tmp12 = (-1.613125930 - 1) * z10 + z5;
- ; = -1.613125930 * z10 - z10 + z5;
-
- movdqa xmm0,xmm5
- paddw xmm5,xmm2
- pmulhw xmm5,[rel PW_F1847] ; xmm5=z5
- pmulhw xmm0,[rel PW_MF1613]
- pmulhw xmm2,[rel PW_F1082]
- psubw xmm0,xmm1
- psubw xmm2,xmm5 ; xmm2=tmp10
- paddw xmm0,xmm5 ; xmm0=tmp12
-
- ; -- Final output stage
-
- psubw xmm0,xmm3 ; xmm0=tmp6
- movdqa xmm1,xmm6
- movdqa xmm5,xmm7
- paddw xmm6,xmm3 ; xmm6=data0=(00 01 02 03 04 05 06 07)
- paddw xmm7,xmm0 ; xmm7=data1=(10 11 12 13 14 15 16 17)
- psubw xmm1,xmm3 ; xmm1=data7=(70 71 72 73 74 75 76 77)
- psubw xmm5,xmm0 ; xmm5=data6=(60 61 62 63 64 65 66 67)
- psubw xmm4,xmm0 ; xmm4=tmp5
-
- movdqa xmm3,xmm6 ; transpose coefficients(phase 1)
- punpcklwd xmm6,xmm7 ; xmm6=(00 10 01 11 02 12 03 13)
- punpckhwd xmm3,xmm7 ; xmm3=(04 14 05 15 06 16 07 17)
- movdqa xmm0,xmm5 ; transpose coefficients(phase 1)
- punpcklwd xmm5,xmm1 ; xmm5=(60 70 61 71 62 72 63 73)
- punpckhwd xmm0,xmm1 ; xmm0=(64 74 65 75 66 76 67 77)
-
- movdqa xmm7, XMMWORD [wk(0)] ; xmm7=tmp2
- movdqa xmm1, XMMWORD [wk(1)] ; xmm1=tmp3
-
- movdqa XMMWORD [wk(0)], xmm5 ; wk(0)=(60 70 61 71 62 72 63 73)
- movdqa XMMWORD [wk(1)], xmm0 ; wk(1)=(64 74 65 75 66 76 67 77)
-
- paddw xmm2,xmm4 ; xmm2=tmp4
- movdqa xmm5,xmm7
- movdqa xmm0,xmm1
- paddw xmm7,xmm4 ; xmm7=data2=(20 21 22 23 24 25 26 27)
- paddw xmm1,xmm2 ; xmm1=data4=(40 41 42 43 44 45 46 47)
- psubw xmm5,xmm4 ; xmm5=data5=(50 51 52 53 54 55 56 57)
- psubw xmm0,xmm2 ; xmm0=data3=(30 31 32 33 34 35 36 37)
-
- movdqa xmm4,xmm7 ; transpose coefficients(phase 1)
- punpcklwd xmm7,xmm0 ; xmm7=(20 30 21 31 22 32 23 33)
- punpckhwd xmm4,xmm0 ; xmm4=(24 34 25 35 26 36 27 37)
- movdqa xmm2,xmm1 ; transpose coefficients(phase 1)
- punpcklwd xmm1,xmm5 ; xmm1=(40 50 41 51 42 52 43 53)
- punpckhwd xmm2,xmm5 ; xmm2=(44 54 45 55 46 56 47 57)
-
- movdqa xmm0,xmm3 ; transpose coefficients(phase 2)
- punpckldq xmm3,xmm4 ; xmm3=(04 14 24 34 05 15 25 35)
- punpckhdq xmm0,xmm4 ; xmm0=(06 16 26 36 07 17 27 37)
- movdqa xmm5,xmm6 ; transpose coefficients(phase 2)
- punpckldq xmm6,xmm7 ; xmm6=(00 10 20 30 01 11 21 31)
- punpckhdq xmm5,xmm7 ; xmm5=(02 12 22 32 03 13 23 33)
-
- movdqa xmm4, XMMWORD [wk(0)] ; xmm4=(60 70 61 71 62 72 63 73)
- movdqa xmm7, XMMWORD [wk(1)] ; xmm7=(64 74 65 75 66 76 67 77)
-
- movdqa XMMWORD [wk(0)], xmm3 ; wk(0)=(04 14 24 34 05 15 25 35)
- movdqa XMMWORD [wk(1)], xmm0 ; wk(1)=(06 16 26 36 07 17 27 37)
-
- movdqa xmm3,xmm1 ; transpose coefficients(phase 2)
- punpckldq xmm1,xmm4 ; xmm1=(40 50 60 70 41 51 61 71)
- punpckhdq xmm3,xmm4 ; xmm3=(42 52 62 72 43 53 63 73)
- movdqa xmm0,xmm2 ; transpose coefficients(phase 2)
- punpckldq xmm2,xmm7 ; xmm2=(44 54 64 74 45 55 65 75)
- punpckhdq xmm0,xmm7 ; xmm0=(46 56 66 76 47 57 67 77)
-
- movdqa xmm4,xmm6 ; transpose coefficients(phase 3)
- punpcklqdq xmm6,xmm1 ; xmm6=col0=(00 10 20 30 40 50 60 70)
- punpckhqdq xmm4,xmm1 ; xmm4=col1=(01 11 21 31 41 51 61 71)
- movdqa xmm7,xmm5 ; transpose coefficients(phase 3)
- punpcklqdq xmm5,xmm3 ; xmm5=col2=(02 12 22 32 42 52 62 72)
- punpckhqdq xmm7,xmm3 ; xmm7=col3=(03 13 23 33 43 53 63 73)
-
- movdqa xmm1, XMMWORD [wk(0)] ; xmm1=(04 14 24 34 05 15 25 35)
- movdqa xmm3, XMMWORD [wk(1)] ; xmm3=(06 16 26 36 07 17 27 37)
-
- movdqa XMMWORD [wk(0)], xmm4 ; wk(0)=col1
- movdqa XMMWORD [wk(1)], xmm7 ; wk(1)=col3
-
- movdqa xmm4,xmm1 ; transpose coefficients(phase 3)
- punpcklqdq xmm1,xmm2 ; xmm1=col4=(04 14 24 34 44 54 64 74)
- punpckhqdq xmm4,xmm2 ; xmm4=col5=(05 15 25 35 45 55 65 75)
- movdqa xmm7,xmm3 ; transpose coefficients(phase 3)
- punpcklqdq xmm3,xmm0 ; xmm3=col6=(06 16 26 36 46 56 66 76)
- punpckhqdq xmm7,xmm0 ; xmm7=col7=(07 17 27 37 47 57 67 77)
-.column_end:
-
- ; -- Prefetch the next coefficient block
-
- prefetchnta [rsi + DCTSIZE2*SIZEOF_JCOEF + 0*32]
- prefetchnta [rsi + DCTSIZE2*SIZEOF_JCOEF + 1*32]
- prefetchnta [rsi + DCTSIZE2*SIZEOF_JCOEF + 2*32]
- prefetchnta [rsi + DCTSIZE2*SIZEOF_JCOEF + 3*32]
-
- ; ---- Pass 2: process rows from work array, store into output array.
-
- mov rax, [original_rbp]
- mov rdi, r12 ; (JSAMPROW *)
- mov eax, r13d
-
- ; -- Even part
-
- ; xmm6=col0, xmm5=col2, xmm1=col4, xmm3=col6
-
- movdqa xmm2,xmm6
- movdqa xmm0,xmm5
- psubw xmm6,xmm1 ; xmm6=tmp11
- psubw xmm5,xmm3
- paddw xmm2,xmm1 ; xmm2=tmp10
- paddw xmm0,xmm3 ; xmm0=tmp13
-
- psllw xmm5,PRE_MULTIPLY_SCALE_BITS
- pmulhw xmm5,[rel PW_F1414]
- psubw xmm5,xmm0 ; xmm5=tmp12
-
- movdqa xmm1,xmm2
- movdqa xmm3,xmm6
- psubw xmm2,xmm0 ; xmm2=tmp3
- psubw xmm6,xmm5 ; xmm6=tmp2
- paddw xmm1,xmm0 ; xmm1=tmp0
- paddw xmm3,xmm5 ; xmm3=tmp1
-
- movdqa xmm0, XMMWORD [wk(0)] ; xmm0=col1
- movdqa xmm5, XMMWORD [wk(1)] ; xmm5=col3
-
- movdqa XMMWORD [wk(0)], xmm2 ; wk(0)=tmp3
- movdqa XMMWORD [wk(1)], xmm6 ; wk(1)=tmp2
-
- ; -- Odd part
-
- ; xmm0=col1, xmm5=col3, xmm4=col5, xmm7=col7
-
- movdqa xmm2,xmm0
- movdqa xmm6,xmm4
- psubw xmm0,xmm7 ; xmm0=z12
- psubw xmm4,xmm5 ; xmm4=z10
- paddw xmm2,xmm7 ; xmm2=z11
- paddw xmm6,xmm5 ; xmm6=z13
-
- movdqa xmm7,xmm4 ; xmm7=z10(unscaled)
- psllw xmm0,PRE_MULTIPLY_SCALE_BITS
- psllw xmm4,PRE_MULTIPLY_SCALE_BITS
-
- movdqa xmm5,xmm2
- psubw xmm2,xmm6
- paddw xmm5,xmm6 ; xmm5=tmp7
-
- psllw xmm2,PRE_MULTIPLY_SCALE_BITS
- pmulhw xmm2,[rel PW_F1414] ; xmm2=tmp11
-
- ; To avoid overflow...
- ;
- ; (Original)
- ; tmp12 = -2.613125930 * z10 + z5;
- ;
- ; (This implementation)
- ; tmp12 = (-1.613125930 - 1) * z10 + z5;
- ; = -1.613125930 * z10 - z10 + z5;
-
- movdqa xmm6,xmm4
- paddw xmm4,xmm0
- pmulhw xmm4,[rel PW_F1847] ; xmm4=z5
- pmulhw xmm6,[rel PW_MF1613]
- pmulhw xmm0,[rel PW_F1082]
- psubw xmm6,xmm7
- psubw xmm0,xmm4 ; xmm0=tmp10
- paddw xmm6,xmm4 ; xmm6=tmp12
-
- ; -- Final output stage
-
- psubw xmm6,xmm5 ; xmm6=tmp6
- movdqa xmm7,xmm1
- movdqa xmm4,xmm3
- paddw xmm1,xmm5 ; xmm1=data0=(00 10 20 30 40 50 60 70)
- paddw xmm3,xmm6 ; xmm3=data1=(01 11 21 31 41 51 61 71)
- psraw xmm1,(PASS1_BITS+3) ; descale
- psraw xmm3,(PASS1_BITS+3) ; descale
- psubw xmm7,xmm5 ; xmm7=data7=(07 17 27 37 47 57 67 77)
- psubw xmm4,xmm6 ; xmm4=data6=(06 16 26 36 46 56 66 76)
- psraw xmm7,(PASS1_BITS+3) ; descale
- psraw xmm4,(PASS1_BITS+3) ; descale
- psubw xmm2,xmm6 ; xmm2=tmp5
-
- packsswb xmm1,xmm4 ; xmm1=(00 10 20 30 40 50 60 70 06 16 26 36 46 56 66 76)
- packsswb xmm3,xmm7 ; xmm3=(01 11 21 31 41 51 61 71 07 17 27 37 47 57 67 77)
-
- movdqa xmm5, XMMWORD [wk(1)] ; xmm5=tmp2
- movdqa xmm6, XMMWORD [wk(0)] ; xmm6=tmp3
-
- paddw xmm0,xmm2 ; xmm0=tmp4
- movdqa xmm4,xmm5
- movdqa xmm7,xmm6
- paddw xmm5,xmm2 ; xmm5=data2=(02 12 22 32 42 52 62 72)
- paddw xmm6,xmm0 ; xmm6=data4=(04 14 24 34 44 54 64 74)
- psraw xmm5,(PASS1_BITS+3) ; descale
- psraw xmm6,(PASS1_BITS+3) ; descale
- psubw xmm4,xmm2 ; xmm4=data5=(05 15 25 35 45 55 65 75)
- psubw xmm7,xmm0 ; xmm7=data3=(03 13 23 33 43 53 63 73)
- psraw xmm4,(PASS1_BITS+3) ; descale
- psraw xmm7,(PASS1_BITS+3) ; descale
-
- movdqa xmm2,[rel PB_CENTERJSAMP] ; xmm2=[rel PB_CENTERJSAMP]
-
- packsswb xmm5,xmm6 ; xmm5=(02 12 22 32 42 52 62 72 04 14 24 34 44 54 64 74)
- packsswb xmm7,xmm4 ; xmm7=(03 13 23 33 43 53 63 73 05 15 25 35 45 55 65 75)
-
- paddb xmm1,xmm2
- paddb xmm3,xmm2
- paddb xmm5,xmm2
- paddb xmm7,xmm2
-
- movdqa xmm0,xmm1 ; transpose coefficients(phase 1)
- punpcklbw xmm1,xmm3 ; xmm1=(00 01 10 11 20 21 30 31 40 41 50 51 60 61 70 71)
- punpckhbw xmm0,xmm3 ; xmm0=(06 07 16 17 26 27 36 37 46 47 56 57 66 67 76 77)
- movdqa xmm6,xmm5 ; transpose coefficients(phase 1)
- punpcklbw xmm5,xmm7 ; xmm5=(02 03 12 13 22 23 32 33 42 43 52 53 62 63 72 73)
- punpckhbw xmm6,xmm7 ; xmm6=(04 05 14 15 24 25 34 35 44 45 54 55 64 65 74 75)
-
- movdqa xmm4,xmm1 ; transpose coefficients(phase 2)
- punpcklwd xmm1,xmm5 ; xmm1=(00 01 02 03 10 11 12 13 20 21 22 23 30 31 32 33)
- punpckhwd xmm4,xmm5 ; xmm4=(40 41 42 43 50 51 52 53 60 61 62 63 70 71 72 73)
- movdqa xmm2,xmm6 ; transpose coefficients(phase 2)
- punpcklwd xmm6,xmm0 ; xmm6=(04 05 06 07 14 15 16 17 24 25 26 27 34 35 36 37)
- punpckhwd xmm2,xmm0 ; xmm2=(44 45 46 47 54 55 56 57 64 65 66 67 74 75 76 77)
-
- movdqa xmm3,xmm1 ; transpose coefficients(phase 3)
- punpckldq xmm1,xmm6 ; xmm1=(00 01 02 03 04 05 06 07 10 11 12 13 14 15 16 17)
- punpckhdq xmm3,xmm6 ; xmm3=(20 21 22 23 24 25 26 27 30 31 32 33 34 35 36 37)
- movdqa xmm7,xmm4 ; transpose coefficients(phase 3)
- punpckldq xmm4,xmm2 ; xmm4=(40 41 42 43 44 45 46 47 50 51 52 53 54 55 56 57)
- punpckhdq xmm7,xmm2 ; xmm7=(60 61 62 63 64 65 66 67 70 71 72 73 74 75 76 77)
-
- pshufd xmm5,xmm1,0x4E ; xmm5=(10 11 12 13 14 15 16 17 00 01 02 03 04 05 06 07)
- pshufd xmm0,xmm3,0x4E ; xmm0=(30 31 32 33 34 35 36 37 20 21 22 23 24 25 26 27)
- pshufd xmm6,xmm4,0x4E ; xmm6=(50 51 52 53 54 55 56 57 40 41 42 43 44 45 46 47)
- pshufd xmm2,xmm7,0x4E ; xmm2=(70 71 72 73 74 75 76 77 60 61 62 63 64 65 66 67)
-
- mov rdx, JSAMPROW [rdi+0*SIZEOF_JSAMPROW]
- mov rsi, JSAMPROW [rdi+2*SIZEOF_JSAMPROW]
- movq XMM_MMWORD [rdx+rax*SIZEOF_JSAMPLE], xmm1
- movq XMM_MMWORD [rsi+rax*SIZEOF_JSAMPLE], xmm3
- mov rdx, JSAMPROW [rdi+4*SIZEOF_JSAMPROW]
- mov rsi, JSAMPROW [rdi+6*SIZEOF_JSAMPROW]
- movq XMM_MMWORD [rdx+rax*SIZEOF_JSAMPLE], xmm4
- movq XMM_MMWORD [rsi+rax*SIZEOF_JSAMPLE], xmm7
-
- mov rdx, JSAMPROW [rdi+1*SIZEOF_JSAMPROW]
- mov rsi, JSAMPROW [rdi+3*SIZEOF_JSAMPROW]
- movq XMM_MMWORD [rdx+rax*SIZEOF_JSAMPLE], xmm5
- movq XMM_MMWORD [rsi+rax*SIZEOF_JSAMPLE], xmm0
- mov rdx, JSAMPROW [rdi+5*SIZEOF_JSAMPROW]
- mov rsi, JSAMPROW [rdi+7*SIZEOF_JSAMPROW]
- movq XMM_MMWORD [rdx+rax*SIZEOF_JSAMPLE], xmm6
- movq XMM_MMWORD [rsi+rax*SIZEOF_JSAMPLE], xmm2
-
- uncollect_args
- mov rsp,rbp ; rsp <- aligned rbp
- pop rsp ; rsp <- original rbp
- pop rbp
- ret
- ret
-
-; For some reason, the OS X linker does not honor the request to align the
-; segment unless we do this.
- align 16
diff --git a/simd/jidctfst-sse2.asm b/simd/jidctfst-sse2.asm
deleted file mode 100644
index f591e55..0000000
--- a/simd/jidctfst-sse2.asm
+++ /dev/null
@@ -1,501 +0,0 @@
-;
-; jidctfst.asm - fast integer IDCT (SSE2)
-;
-; Copyright 2009 Pierre Ossman <ossman@cendio.se> for Cendio AB
-;
-; Based on the x86 SIMD extension for IJG JPEG library
-; Copyright (C) 1999-2006, MIYASAKA Masaru.
-; For conditions of distribution and use, see copyright notice in jsimdext.inc
-;
-; This file should be assembled with NASM (Netwide Assembler),
-; can *not* be assembled with Microsoft's MASM or any compatible
-; assembler (including Borland's Turbo Assembler).
-; NASM is available from http://nasm.sourceforge.net/ or
-; http://sourceforge.net/project/showfiles.php?group_id=6208
-;
-; This file contains a fast, not so accurate integer implementation of
-; the inverse DCT (Discrete Cosine Transform). The following code is
-; based directly on the IJG's original jidctfst.c; see the jidctfst.c
-; for more details.
-;
-; [TAB8]
-
-%include "jsimdext.inc"
-%include "jdct.inc"
-
-; --------------------------------------------------------------------------
-
-%define CONST_BITS 8 ; 14 is also OK.
-%define PASS1_BITS 2
-
-%if IFAST_SCALE_BITS != PASS1_BITS
-%error "'IFAST_SCALE_BITS' must be equal to 'PASS1_BITS'."
-%endif
-
-%if CONST_BITS == 8
-F_1_082 equ 277 ; FIX(1.082392200)
-F_1_414 equ 362 ; FIX(1.414213562)
-F_1_847 equ 473 ; FIX(1.847759065)
-F_2_613 equ 669 ; FIX(2.613125930)
-F_1_613 equ (F_2_613 - 256) ; FIX(2.613125930) - FIX(1)
-%else
-; NASM cannot do compile-time arithmetic on floating-point constants.
-%define DESCALE(x,n) (((x)+(1<<((n)-1)))>>(n))
-F_1_082 equ DESCALE(1162209775,30-CONST_BITS) ; FIX(1.082392200)
-F_1_414 equ DESCALE(1518500249,30-CONST_BITS) ; FIX(1.414213562)
-F_1_847 equ DESCALE(1984016188,30-CONST_BITS) ; FIX(1.847759065)
-F_2_613 equ DESCALE(2805822602,30-CONST_BITS) ; FIX(2.613125930)
-F_1_613 equ (F_2_613 - (1 << CONST_BITS)) ; FIX(2.613125930) - FIX(1)
-%endif
-
-; --------------------------------------------------------------------------
- SECTION SEG_CONST
-
-; PRE_MULTIPLY_SCALE_BITS <= 2 (to avoid overflow)
-; CONST_BITS + CONST_SHIFT + PRE_MULTIPLY_SCALE_BITS == 16 (for pmulhw)
-
-%define PRE_MULTIPLY_SCALE_BITS 2
-%define CONST_SHIFT (16 - PRE_MULTIPLY_SCALE_BITS - CONST_BITS)
-
- alignz 16
- global EXTN(jconst_idct_ifast_sse2)
-
-EXTN(jconst_idct_ifast_sse2):
-
-PW_F1414 times 8 dw F_1_414 << CONST_SHIFT
-PW_F1847 times 8 dw F_1_847 << CONST_SHIFT
-PW_MF1613 times 8 dw -F_1_613 << CONST_SHIFT
-PW_F1082 times 8 dw F_1_082 << CONST_SHIFT
-PB_CENTERJSAMP times 16 db CENTERJSAMPLE
-
- alignz 16
-
-; --------------------------------------------------------------------------
- SECTION SEG_TEXT
- BITS 32
-;
-; Perform dequantization and inverse DCT on one block of coefficients.
-;
-; GLOBAL(void)
-; jsimd_idct_ifast_sse2 (void *dct_table, JCOEFPTR coef_block,
-; JSAMPARRAY output_buf, JDIMENSION output_col)
-;
-
-%define dct_table(b) (b)+8 ; jpeg_component_info *compptr
-%define coef_block(b) (b)+12 ; JCOEFPTR coef_block
-%define output_buf(b) (b)+16 ; JSAMPARRAY output_buf
-%define output_col(b) (b)+20 ; JDIMENSION output_col
-
-%define original_ebp ebp+0
-%define wk(i) ebp-(WK_NUM-(i))*SIZEOF_XMMWORD ; xmmword wk[WK_NUM]
-%define WK_NUM 2
-
- align 16
- global EXTN(jsimd_idct_ifast_sse2)
-
-EXTN(jsimd_idct_ifast_sse2):
- push ebp
- mov eax,esp ; eax = original ebp
- sub esp, byte 4
- and esp, byte (-SIZEOF_XMMWORD) ; align to 128 bits
- mov [esp],eax
- mov ebp,esp ; ebp = aligned ebp
- lea esp, [wk(0)]
- pushpic ebx
-; push ecx ; unused
-; push edx ; need not be preserved
- push esi
- push edi
-
- get_GOT ebx ; get GOT address
-
- ; ---- Pass 1: process columns from input.
-
-; mov eax, [original_ebp]
- mov edx, POINTER [dct_table(eax)] ; quantptr
- mov esi, JCOEFPTR [coef_block(eax)] ; inptr
-
-%ifndef NO_ZERO_COLUMN_TEST_IFAST_SSE2
- mov eax, DWORD [DWBLOCK(1,0,esi,SIZEOF_JCOEF)]
- or eax, DWORD [DWBLOCK(2,0,esi,SIZEOF_JCOEF)]
- jnz near .columnDCT
-
- movdqa xmm0, XMMWORD [XMMBLOCK(1,0,esi,SIZEOF_JCOEF)]
- movdqa xmm1, XMMWORD [XMMBLOCK(2,0,esi,SIZEOF_JCOEF)]
- por xmm0, XMMWORD [XMMBLOCK(3,0,esi,SIZEOF_JCOEF)]
- por xmm1, XMMWORD [XMMBLOCK(4,0,esi,SIZEOF_JCOEF)]
- por xmm0, XMMWORD [XMMBLOCK(5,0,esi,SIZEOF_JCOEF)]
- por xmm1, XMMWORD [XMMBLOCK(6,0,esi,SIZEOF_JCOEF)]
- por xmm0, XMMWORD [XMMBLOCK(7,0,esi,SIZEOF_JCOEF)]
- por xmm1,xmm0
- packsswb xmm1,xmm1
- packsswb xmm1,xmm1
- movd eax,xmm1
- test eax,eax
- jnz short .columnDCT
-
- ; -- AC terms all zero
-
- movdqa xmm0, XMMWORD [XMMBLOCK(0,0,esi,SIZEOF_JCOEF)]
- pmullw xmm0, XMMWORD [XMMBLOCK(0,0,edx,SIZEOF_ISLOW_MULT_TYPE)]
-
- movdqa xmm7,xmm0 ; xmm0=in0=(00 01 02 03 04 05 06 07)
- punpcklwd xmm0,xmm0 ; xmm0=(00 00 01 01 02 02 03 03)
- punpckhwd xmm7,xmm7 ; xmm7=(04 04 05 05 06 06 07 07)
-
- pshufd xmm6,xmm0,0x00 ; xmm6=col0=(00 00 00 00 00 00 00 00)
- pshufd xmm2,xmm0,0x55 ; xmm2=col1=(01 01 01 01 01 01 01 01)
- pshufd xmm5,xmm0,0xAA ; xmm5=col2=(02 02 02 02 02 02 02 02)
- pshufd xmm0,xmm0,0xFF ; xmm0=col3=(03 03 03 03 03 03 03 03)
- pshufd xmm1,xmm7,0x00 ; xmm1=col4=(04 04 04 04 04 04 04 04)
- pshufd xmm4,xmm7,0x55 ; xmm4=col5=(05 05 05 05 05 05 05 05)
- pshufd xmm3,xmm7,0xAA ; xmm3=col6=(06 06 06 06 06 06 06 06)
- pshufd xmm7,xmm7,0xFF ; xmm7=col7=(07 07 07 07 07 07 07 07)
-
- movdqa XMMWORD [wk(0)], xmm2 ; wk(0)=col1
- movdqa XMMWORD [wk(1)], xmm0 ; wk(1)=col3
- jmp near .column_end
- alignx 16,7
-%endif
-.columnDCT:
-
- ; -- Even part
-
- movdqa xmm0, XMMWORD [XMMBLOCK(0,0,esi,SIZEOF_JCOEF)]
- movdqa xmm1, XMMWORD [XMMBLOCK(2,0,esi,SIZEOF_JCOEF)]
- pmullw xmm0, XMMWORD [XMMBLOCK(0,0,edx,SIZEOF_IFAST_MULT_TYPE)]
- pmullw xmm1, XMMWORD [XMMBLOCK(2,0,edx,SIZEOF_IFAST_MULT_TYPE)]
- movdqa xmm2, XMMWORD [XMMBLOCK(4,0,esi,SIZEOF_JCOEF)]
- movdqa xmm3, XMMWORD [XMMBLOCK(6,0,esi,SIZEOF_JCOEF)]
- pmullw xmm2, XMMWORD [XMMBLOCK(4,0,edx,SIZEOF_IFAST_MULT_TYPE)]
- pmullw xmm3, XMMWORD [XMMBLOCK(6,0,edx,SIZEOF_IFAST_MULT_TYPE)]
-
- movdqa xmm4,xmm0
- movdqa xmm5,xmm1
- psubw xmm0,xmm2 ; xmm0=tmp11
- psubw xmm1,xmm3
- paddw xmm4,xmm2 ; xmm4=tmp10
- paddw xmm5,xmm3 ; xmm5=tmp13
-
- psllw xmm1,PRE_MULTIPLY_SCALE_BITS
- pmulhw xmm1,[GOTOFF(ebx,PW_F1414)]
- psubw xmm1,xmm5 ; xmm1=tmp12
-
- movdqa xmm6,xmm4
- movdqa xmm7,xmm0
- psubw xmm4,xmm5 ; xmm4=tmp3
- psubw xmm0,xmm1 ; xmm0=tmp2
- paddw xmm6,xmm5 ; xmm6=tmp0
- paddw xmm7,xmm1 ; xmm7=tmp1
-
- movdqa XMMWORD [wk(1)], xmm4 ; wk(1)=tmp3
- movdqa XMMWORD [wk(0)], xmm0 ; wk(0)=tmp2
-
- ; -- Odd part
-
- movdqa xmm2, XMMWORD [XMMBLOCK(1,0,esi,SIZEOF_JCOEF)]
- movdqa xmm3, XMMWORD [XMMBLOCK(3,0,esi,SIZEOF_JCOEF)]
- pmullw xmm2, XMMWORD [XMMBLOCK(1,0,edx,SIZEOF_IFAST_MULT_TYPE)]
- pmullw xmm3, XMMWORD [XMMBLOCK(3,0,edx,SIZEOF_IFAST_MULT_TYPE)]
- movdqa xmm5, XMMWORD [XMMBLOCK(5,0,esi,SIZEOF_JCOEF)]
- movdqa xmm1, XMMWORD [XMMBLOCK(7,0,esi,SIZEOF_JCOEF)]
- pmullw xmm5, XMMWORD [XMMBLOCK(5,0,edx,SIZEOF_IFAST_MULT_TYPE)]
- pmullw xmm1, XMMWORD [XMMBLOCK(7,0,edx,SIZEOF_IFAST_MULT_TYPE)]
-
- movdqa xmm4,xmm2
- movdqa xmm0,xmm5
- psubw xmm2,xmm1 ; xmm2=z12
- psubw xmm5,xmm3 ; xmm5=z10
- paddw xmm4,xmm1 ; xmm4=z11
- paddw xmm0,xmm3 ; xmm0=z13
-
- movdqa xmm1,xmm5 ; xmm1=z10(unscaled)
- psllw xmm2,PRE_MULTIPLY_SCALE_BITS
- psllw xmm5,PRE_MULTIPLY_SCALE_BITS
-
- movdqa xmm3,xmm4
- psubw xmm4,xmm0
- paddw xmm3,xmm0 ; xmm3=tmp7
-
- psllw xmm4,PRE_MULTIPLY_SCALE_BITS
- pmulhw xmm4,[GOTOFF(ebx,PW_F1414)] ; xmm4=tmp11
-
- ; To avoid overflow...
- ;
- ; (Original)
- ; tmp12 = -2.613125930 * z10 + z5;
- ;
- ; (This implementation)
- ; tmp12 = (-1.613125930 - 1) * z10 + z5;
- ; = -1.613125930 * z10 - z10 + z5;
-
- movdqa xmm0,xmm5
- paddw xmm5,xmm2
- pmulhw xmm5,[GOTOFF(ebx,PW_F1847)] ; xmm5=z5
- pmulhw xmm0,[GOTOFF(ebx,PW_MF1613)]
- pmulhw xmm2,[GOTOFF(ebx,PW_F1082)]
- psubw xmm0,xmm1
- psubw xmm2,xmm5 ; xmm2=tmp10
- paddw xmm0,xmm5 ; xmm0=tmp12
-
- ; -- Final output stage
-
- psubw xmm0,xmm3 ; xmm0=tmp6
- movdqa xmm1,xmm6
- movdqa xmm5,xmm7
- paddw xmm6,xmm3 ; xmm6=data0=(00 01 02 03 04 05 06 07)
- paddw xmm7,xmm0 ; xmm7=data1=(10 11 12 13 14 15 16 17)
- psubw xmm1,xmm3 ; xmm1=data7=(70 71 72 73 74 75 76 77)
- psubw xmm5,xmm0 ; xmm5=data6=(60 61 62 63 64 65 66 67)
- psubw xmm4,xmm0 ; xmm4=tmp5
-
- movdqa xmm3,xmm6 ; transpose coefficients(phase 1)
- punpcklwd xmm6,xmm7 ; xmm6=(00 10 01 11 02 12 03 13)
- punpckhwd xmm3,xmm7 ; xmm3=(04 14 05 15 06 16 07 17)
- movdqa xmm0,xmm5 ; transpose coefficients(phase 1)
- punpcklwd xmm5,xmm1 ; xmm5=(60 70 61 71 62 72 63 73)
- punpckhwd xmm0,xmm1 ; xmm0=(64 74 65 75 66 76 67 77)
-
- movdqa xmm7, XMMWORD [wk(0)] ; xmm7=tmp2
- movdqa xmm1, XMMWORD [wk(1)] ; xmm1=tmp3
-
- movdqa XMMWORD [wk(0)], xmm5 ; wk(0)=(60 70 61 71 62 72 63 73)
- movdqa XMMWORD [wk(1)], xmm0 ; wk(1)=(64 74 65 75 66 76 67 77)
-
- paddw xmm2,xmm4 ; xmm2=tmp4
- movdqa xmm5,xmm7
- movdqa xmm0,xmm1
- paddw xmm7,xmm4 ; xmm7=data2=(20 21 22 23 24 25 26 27)
- paddw xmm1,xmm2 ; xmm1=data4=(40 41 42 43 44 45 46 47)
- psubw xmm5,xmm4 ; xmm5=data5=(50 51 52 53 54 55 56 57)
- psubw xmm0,xmm2 ; xmm0=data3=(30 31 32 33 34 35 36 37)
-
- movdqa xmm4,xmm7 ; transpose coefficients(phase 1)
- punpcklwd xmm7,xmm0 ; xmm7=(20 30 21 31 22 32 23 33)
- punpckhwd xmm4,xmm0 ; xmm4=(24 34 25 35 26 36 27 37)
- movdqa xmm2,xmm1 ; transpose coefficients(phase 1)
- punpcklwd xmm1,xmm5 ; xmm1=(40 50 41 51 42 52 43 53)
- punpckhwd xmm2,xmm5 ; xmm2=(44 54 45 55 46 56 47 57)
-
- movdqa xmm0,xmm3 ; transpose coefficients(phase 2)
- punpckldq xmm3,xmm4 ; xmm3=(04 14 24 34 05 15 25 35)
- punpckhdq xmm0,xmm4 ; xmm0=(06 16 26 36 07 17 27 37)
- movdqa xmm5,xmm6 ; transpose coefficients(phase 2)
- punpckldq xmm6,xmm7 ; xmm6=(00 10 20 30 01 11 21 31)
- punpckhdq xmm5,xmm7 ; xmm5=(02 12 22 32 03 13 23 33)
-
- movdqa xmm4, XMMWORD [wk(0)] ; xmm4=(60 70 61 71 62 72 63 73)
- movdqa xmm7, XMMWORD [wk(1)] ; xmm7=(64 74 65 75 66 76 67 77)
-
- movdqa XMMWORD [wk(0)], xmm3 ; wk(0)=(04 14 24 34 05 15 25 35)
- movdqa XMMWORD [wk(1)], xmm0 ; wk(1)=(06 16 26 36 07 17 27 37)
-
- movdqa xmm3,xmm1 ; transpose coefficients(phase 2)
- punpckldq xmm1,xmm4 ; xmm1=(40 50 60 70 41 51 61 71)
- punpckhdq xmm3,xmm4 ; xmm3=(42 52 62 72 43 53 63 73)
- movdqa xmm0,xmm2 ; transpose coefficients(phase 2)
- punpckldq xmm2,xmm7 ; xmm2=(44 54 64 74 45 55 65 75)
- punpckhdq xmm0,xmm7 ; xmm0=(46 56 66 76 47 57 67 77)
-
- movdqa xmm4,xmm6 ; transpose coefficients(phase 3)
- punpcklqdq xmm6,xmm1 ; xmm6=col0=(00 10 20 30 40 50 60 70)
- punpckhqdq xmm4,xmm1 ; xmm4=col1=(01 11 21 31 41 51 61 71)
- movdqa xmm7,xmm5 ; transpose coefficients(phase 3)
- punpcklqdq xmm5,xmm3 ; xmm5=col2=(02 12 22 32 42 52 62 72)
- punpckhqdq xmm7,xmm3 ; xmm7=col3=(03 13 23 33 43 53 63 73)
-
- movdqa xmm1, XMMWORD [wk(0)] ; xmm1=(04 14 24 34 05 15 25 35)
- movdqa xmm3, XMMWORD [wk(1)] ; xmm3=(06 16 26 36 07 17 27 37)
-
- movdqa XMMWORD [wk(0)], xmm4 ; wk(0)=col1
- movdqa XMMWORD [wk(1)], xmm7 ; wk(1)=col3
-
- movdqa xmm4,xmm1 ; transpose coefficients(phase 3)
- punpcklqdq xmm1,xmm2 ; xmm1=col4=(04 14 24 34 44 54 64 74)
- punpckhqdq xmm4,xmm2 ; xmm4=col5=(05 15 25 35 45 55 65 75)
- movdqa xmm7,xmm3 ; transpose coefficients(phase 3)
- punpcklqdq xmm3,xmm0 ; xmm3=col6=(06 16 26 36 46 56 66 76)
- punpckhqdq xmm7,xmm0 ; xmm7=col7=(07 17 27 37 47 57 67 77)
-.column_end:
-
- ; -- Prefetch the next coefficient block
-
- prefetchnta [esi + DCTSIZE2*SIZEOF_JCOEF + 0*32]
- prefetchnta [esi + DCTSIZE2*SIZEOF_JCOEF + 1*32]
- prefetchnta [esi + DCTSIZE2*SIZEOF_JCOEF + 2*32]
- prefetchnta [esi + DCTSIZE2*SIZEOF_JCOEF + 3*32]
-
- ; ---- Pass 2: process rows from work array, store into output array.
-
- mov eax, [original_ebp]
- mov edi, JSAMPARRAY [output_buf(eax)] ; (JSAMPROW *)
- mov eax, JDIMENSION [output_col(eax)]
-
- ; -- Even part
-
- ; xmm6=col0, xmm5=col2, xmm1=col4, xmm3=col6
-
- movdqa xmm2,xmm6
- movdqa xmm0,xmm5
- psubw xmm6,xmm1 ; xmm6=tmp11
- psubw xmm5,xmm3
- paddw xmm2,xmm1 ; xmm2=tmp10
- paddw xmm0,xmm3 ; xmm0=tmp13
-
- psllw xmm5,PRE_MULTIPLY_SCALE_BITS
- pmulhw xmm5,[GOTOFF(ebx,PW_F1414)]
- psubw xmm5,xmm0 ; xmm5=tmp12
-
- movdqa xmm1,xmm2
- movdqa xmm3,xmm6
- psubw xmm2,xmm0 ; xmm2=tmp3
- psubw xmm6,xmm5 ; xmm6=tmp2
- paddw xmm1,xmm0 ; xmm1=tmp0
- paddw xmm3,xmm5 ; xmm3=tmp1
-
- movdqa xmm0, XMMWORD [wk(0)] ; xmm0=col1
- movdqa xmm5, XMMWORD [wk(1)] ; xmm5=col3
-
- movdqa XMMWORD [wk(0)], xmm2 ; wk(0)=tmp3
- movdqa XMMWORD [wk(1)], xmm6 ; wk(1)=tmp2
-
- ; -- Odd part
-
- ; xmm0=col1, xmm5=col3, xmm4=col5, xmm7=col7
-
- movdqa xmm2,xmm0
- movdqa xmm6,xmm4
- psubw xmm0,xmm7 ; xmm0=z12
- psubw xmm4,xmm5 ; xmm4=z10
- paddw xmm2,xmm7 ; xmm2=z11
- paddw xmm6,xmm5 ; xmm6=z13
-
- movdqa xmm7,xmm4 ; xmm7=z10(unscaled)
- psllw xmm0,PRE_MULTIPLY_SCALE_BITS
- psllw xmm4,PRE_MULTIPLY_SCALE_BITS
-
- movdqa xmm5,xmm2
- psubw xmm2,xmm6
- paddw xmm5,xmm6 ; xmm5=tmp7
-
- psllw xmm2,PRE_MULTIPLY_SCALE_BITS
- pmulhw xmm2,[GOTOFF(ebx,PW_F1414)] ; xmm2=tmp11
-
- ; To avoid overflow...
- ;
- ; (Original)
- ; tmp12 = -2.613125930 * z10 + z5;
- ;
- ; (This implementation)
- ; tmp12 = (-1.613125930 - 1) * z10 + z5;
- ; = -1.613125930 * z10 - z10 + z5;
-
- movdqa xmm6,xmm4
- paddw xmm4,xmm0
- pmulhw xmm4,[GOTOFF(ebx,PW_F1847)] ; xmm4=z5
- pmulhw xmm6,[GOTOFF(ebx,PW_MF1613)]
- pmulhw xmm0,[GOTOFF(ebx,PW_F1082)]
- psubw xmm6,xmm7
- psubw xmm0,xmm4 ; xmm0=tmp10
- paddw xmm6,xmm4 ; xmm6=tmp12
-
- ; -- Final output stage
-
- psubw xmm6,xmm5 ; xmm6=tmp6
- movdqa xmm7,xmm1
- movdqa xmm4,xmm3
- paddw xmm1,xmm5 ; xmm1=data0=(00 10 20 30 40 50 60 70)
- paddw xmm3,xmm6 ; xmm3=data1=(01 11 21 31 41 51 61 71)
- psraw xmm1,(PASS1_BITS+3) ; descale
- psraw xmm3,(PASS1_BITS+3) ; descale
- psubw xmm7,xmm5 ; xmm7=data7=(07 17 27 37 47 57 67 77)
- psubw xmm4,xmm6 ; xmm4=data6=(06 16 26 36 46 56 66 76)
- psraw xmm7,(PASS1_BITS+3) ; descale
- psraw xmm4,(PASS1_BITS+3) ; descale
- psubw xmm2,xmm6 ; xmm2=tmp5
-
- packsswb xmm1,xmm4 ; xmm1=(00 10 20 30 40 50 60 70 06 16 26 36 46 56 66 76)
- packsswb xmm3,xmm7 ; xmm3=(01 11 21 31 41 51 61 71 07 17 27 37 47 57 67 77)
-
- movdqa xmm5, XMMWORD [wk(1)] ; xmm5=tmp2
- movdqa xmm6, XMMWORD [wk(0)] ; xmm6=tmp3
-
- paddw xmm0,xmm2 ; xmm0=tmp4
- movdqa xmm4,xmm5
- movdqa xmm7,xmm6
- paddw xmm5,xmm2 ; xmm5=data2=(02 12 22 32 42 52 62 72)
- paddw xmm6,xmm0 ; xmm6=data4=(04 14 24 34 44 54 64 74)
- psraw xmm5,(PASS1_BITS+3) ; descale
- psraw xmm6,(PASS1_BITS+3) ; descale
- psubw xmm4,xmm2 ; xmm4=data5=(05 15 25 35 45 55 65 75)
- psubw xmm7,xmm0 ; xmm7=data3=(03 13 23 33 43 53 63 73)
- psraw xmm4,(PASS1_BITS+3) ; descale
- psraw xmm7,(PASS1_BITS+3) ; descale
-
- movdqa xmm2,[GOTOFF(ebx,PB_CENTERJSAMP)] ; xmm2=[PB_CENTERJSAMP]
-
- packsswb xmm5,xmm6 ; xmm5=(02 12 22 32 42 52 62 72 04 14 24 34 44 54 64 74)
- packsswb xmm7,xmm4 ; xmm7=(03 13 23 33 43 53 63 73 05 15 25 35 45 55 65 75)
-
- paddb xmm1,xmm2
- paddb xmm3,xmm2
- paddb xmm5,xmm2
- paddb xmm7,xmm2
-
- movdqa xmm0,xmm1 ; transpose coefficients(phase 1)
- punpcklbw xmm1,xmm3 ; xmm1=(00 01 10 11 20 21 30 31 40 41 50 51 60 61 70 71)
- punpckhbw xmm0,xmm3 ; xmm0=(06 07 16 17 26 27 36 37 46 47 56 57 66 67 76 77)
- movdqa xmm6,xmm5 ; transpose coefficients(phase 1)
- punpcklbw xmm5,xmm7 ; xmm5=(02 03 12 13 22 23 32 33 42 43 52 53 62 63 72 73)
- punpckhbw xmm6,xmm7 ; xmm6=(04 05 14 15 24 25 34 35 44 45 54 55 64 65 74 75)
-
- movdqa xmm4,xmm1 ; transpose coefficients(phase 2)
- punpcklwd xmm1,xmm5 ; xmm1=(00 01 02 03 10 11 12 13 20 21 22 23 30 31 32 33)
- punpckhwd xmm4,xmm5 ; xmm4=(40 41 42 43 50 51 52 53 60 61 62 63 70 71 72 73)
- movdqa xmm2,xmm6 ; transpose coefficients(phase 2)
- punpcklwd xmm6,xmm0 ; xmm6=(04 05 06 07 14 15 16 17 24 25 26 27 34 35 36 37)
- punpckhwd xmm2,xmm0 ; xmm2=(44 45 46 47 54 55 56 57 64 65 66 67 74 75 76 77)
-
- movdqa xmm3,xmm1 ; transpose coefficients(phase 3)
- punpckldq xmm1,xmm6 ; xmm1=(00 01 02 03 04 05 06 07 10 11 12 13 14 15 16 17)
- punpckhdq xmm3,xmm6 ; xmm3=(20 21 22 23 24 25 26 27 30 31 32 33 34 35 36 37)
- movdqa xmm7,xmm4 ; transpose coefficients(phase 3)
- punpckldq xmm4,xmm2 ; xmm4=(40 41 42 43 44 45 46 47 50 51 52 53 54 55 56 57)
- punpckhdq xmm7,xmm2 ; xmm7=(60 61 62 63 64 65 66 67 70 71 72 73 74 75 76 77)
-
- pshufd xmm5,xmm1,0x4E ; xmm5=(10 11 12 13 14 15 16 17 00 01 02 03 04 05 06 07)
- pshufd xmm0,xmm3,0x4E ; xmm0=(30 31 32 33 34 35 36 37 20 21 22 23 24 25 26 27)
- pshufd xmm6,xmm4,0x4E ; xmm6=(50 51 52 53 54 55 56 57 40 41 42 43 44 45 46 47)
- pshufd xmm2,xmm7,0x4E ; xmm2=(70 71 72 73 74 75 76 77 60 61 62 63 64 65 66 67)
-
- mov edx, JSAMPROW [edi+0*SIZEOF_JSAMPROW]
- mov esi, JSAMPROW [edi+2*SIZEOF_JSAMPROW]
- movq XMM_MMWORD [edx+eax*SIZEOF_JSAMPLE], xmm1
- movq XMM_MMWORD [esi+eax*SIZEOF_JSAMPLE], xmm3
- mov edx, JSAMPROW [edi+4*SIZEOF_JSAMPROW]
- mov esi, JSAMPROW [edi+6*SIZEOF_JSAMPROW]
- movq XMM_MMWORD [edx+eax*SIZEOF_JSAMPLE], xmm4
- movq XMM_MMWORD [esi+eax*SIZEOF_JSAMPLE], xmm7
-
- mov edx, JSAMPROW [edi+1*SIZEOF_JSAMPROW]
- mov esi, JSAMPROW [edi+3*SIZEOF_JSAMPROW]
- movq XMM_MMWORD [edx+eax*SIZEOF_JSAMPLE], xmm5
- movq XMM_MMWORD [esi+eax*SIZEOF_JSAMPLE], xmm0
- mov edx, JSAMPROW [edi+5*SIZEOF_JSAMPROW]
- mov esi, JSAMPROW [edi+7*SIZEOF_JSAMPROW]
- movq XMM_MMWORD [edx+eax*SIZEOF_JSAMPLE], xmm6
- movq XMM_MMWORD [esi+eax*SIZEOF_JSAMPLE], xmm2
-
- pop edi
- pop esi
-; pop edx ; need not be preserved
-; pop ecx ; unused
- poppic ebx
- mov esp,ebp ; esp <- aligned ebp
- pop esp ; esp <- original ebp
- pop ebp
- ret
-
-; For some reason, the OS X linker does not honor the request to align the
-; segment unless we do this.
- align 16
diff --git a/simd/jidctint-sse2-64.asm b/simd/jidctint-sse2-64.asm
deleted file mode 100644
index afe1d6a..0000000
--- a/simd/jidctint-sse2-64.asm
+++ /dev/null
@@ -1,847 +0,0 @@
-;
-; jidctint.asm - accurate integer IDCT (64-bit SSE2)
-;
-; Copyright 2009 Pierre Ossman <ossman@cendio.se> for Cendio AB
-; Copyright (C) 2009, D. R. Commander.
-;
-; Based on the x86 SIMD extension for IJG JPEG library
-; Copyright (C) 1999-2006, MIYASAKA Masaru.
-; For conditions of distribution and use, see copyright notice in jsimdext.inc
-;
-; This file should be assembled with NASM (Netwide Assembler),
-; can *not* be assembled with Microsoft's MASM or any compatible
-; assembler (including Borland's Turbo Assembler).
-; NASM is available from http://nasm.sourceforge.net/ or
-; http://sourceforge.net/project/showfiles.php?group_id=6208
-;
-; This file contains a slow-but-accurate integer implementation of the
-; inverse DCT (Discrete Cosine Transform). The following code is based
-; directly on the IJG's original jidctint.c; see the jidctint.c for
-; more details.
-;
-; [TAB8]
-
-%include "jsimdext.inc"
-%include "jdct.inc"
-
-; --------------------------------------------------------------------------
-
-%define CONST_BITS 13
-%define PASS1_BITS 2
-
-%define DESCALE_P1 (CONST_BITS-PASS1_BITS)
-%define DESCALE_P2 (CONST_BITS+PASS1_BITS+3)
-
-%if CONST_BITS == 13
-F_0_298 equ 2446 ; FIX(0.298631336)
-F_0_390 equ 3196 ; FIX(0.390180644)
-F_0_541 equ 4433 ; FIX(0.541196100)
-F_0_765 equ 6270 ; FIX(0.765366865)
-F_0_899 equ 7373 ; FIX(0.899976223)
-F_1_175 equ 9633 ; FIX(1.175875602)
-F_1_501 equ 12299 ; FIX(1.501321110)
-F_1_847 equ 15137 ; FIX(1.847759065)
-F_1_961 equ 16069 ; FIX(1.961570560)
-F_2_053 equ 16819 ; FIX(2.053119869)
-F_2_562 equ 20995 ; FIX(2.562915447)
-F_3_072 equ 25172 ; FIX(3.072711026)
-%else
-; NASM cannot do compile-time arithmetic on floating-point constants.
-%define DESCALE(x,n) (((x)+(1<<((n)-1)))>>(n))
-F_0_298 equ DESCALE( 320652955,30-CONST_BITS) ; FIX(0.298631336)
-F_0_390 equ DESCALE( 418953276,30-CONST_BITS) ; FIX(0.390180644)
-F_0_541 equ DESCALE( 581104887,30-CONST_BITS) ; FIX(0.541196100)
-F_0_765 equ DESCALE( 821806413,30-CONST_BITS) ; FIX(0.765366865)
-F_0_899 equ DESCALE( 966342111,30-CONST_BITS) ; FIX(0.899976223)
-F_1_175 equ DESCALE(1262586813,30-CONST_BITS) ; FIX(1.175875602)
-F_1_501 equ DESCALE(1612031267,30-CONST_BITS) ; FIX(1.501321110)
-F_1_847 equ DESCALE(1984016188,30-CONST_BITS) ; FIX(1.847759065)
-F_1_961 equ DESCALE(2106220350,30-CONST_BITS) ; FIX(1.961570560)
-F_2_053 equ DESCALE(2204520673,30-CONST_BITS) ; FIX(2.053119869)
-F_2_562 equ DESCALE(2751909506,30-CONST_BITS) ; FIX(2.562915447)
-F_3_072 equ DESCALE(3299298341,30-CONST_BITS) ; FIX(3.072711026)
-%endif
-
-; --------------------------------------------------------------------------
- SECTION SEG_CONST
-
- alignz 16
- global EXTN(jconst_idct_islow_sse2)
-
-EXTN(jconst_idct_islow_sse2):
-
-PW_F130_F054 times 4 dw (F_0_541+F_0_765), F_0_541
-PW_F054_MF130 times 4 dw F_0_541, (F_0_541-F_1_847)
-PW_MF078_F117 times 4 dw (F_1_175-F_1_961), F_1_175
-PW_F117_F078 times 4 dw F_1_175, (F_1_175-F_0_390)
-PW_MF060_MF089 times 4 dw (F_0_298-F_0_899),-F_0_899
-PW_MF089_F060 times 4 dw -F_0_899, (F_1_501-F_0_899)
-PW_MF050_MF256 times 4 dw (F_2_053-F_2_562),-F_2_562
-PW_MF256_F050 times 4 dw -F_2_562, (F_3_072-F_2_562)
-PD_DESCALE_P1 times 4 dd 1 << (DESCALE_P1-1)
-PD_DESCALE_P2 times 4 dd 1 << (DESCALE_P2-1)
-PB_CENTERJSAMP times 16 db CENTERJSAMPLE
-
- alignz 16
-
-; --------------------------------------------------------------------------
- SECTION SEG_TEXT
- BITS 64
-;
-; Perform dequantization and inverse DCT on one block of coefficients.
-;
-; GLOBAL(void)
-; jsimd_idct_islow_sse2 (void *dct_table, JCOEFPTR coef_block,
-; JSAMPARRAY output_buf, JDIMENSION output_col)
-;
-
-; r10 = jpeg_component_info *compptr
-; r11 = JCOEFPTR coef_block
-; r12 = JSAMPARRAY output_buf
-; r13 = JDIMENSION output_col
-
-%define original_rbp rbp+0
-%define wk(i) rbp-(WK_NUM-(i))*SIZEOF_XMMWORD ; xmmword wk[WK_NUM]
-%define WK_NUM 12
-
- align 16
- global EXTN(jsimd_idct_islow_sse2)
-
-EXTN(jsimd_idct_islow_sse2):
- push rbp
- mov rax,rsp ; rax = original rbp
- sub rsp, byte 4
- and rsp, byte (-SIZEOF_XMMWORD) ; align to 128 bits
- mov [rsp],rax
- mov rbp,rsp ; rbp = aligned rbp
- lea rsp, [wk(0)]
- collect_args
-
- ; ---- Pass 1: process columns from input.
-
- mov rdx, r10 ; quantptr
- mov rsi, r11 ; inptr
-
-%ifndef NO_ZERO_COLUMN_TEST_ISLOW_SSE2
- mov eax, DWORD [DWBLOCK(1,0,rsi,SIZEOF_JCOEF)]
- or eax, DWORD [DWBLOCK(2,0,rsi,SIZEOF_JCOEF)]
- jnz near .columnDCT
-
- movdqa xmm0, XMMWORD [XMMBLOCK(1,0,rsi,SIZEOF_JCOEF)]
- movdqa xmm1, XMMWORD [XMMBLOCK(2,0,rsi,SIZEOF_JCOEF)]
- por xmm0, XMMWORD [XMMBLOCK(3,0,rsi,SIZEOF_JCOEF)]
- por xmm1, XMMWORD [XMMBLOCK(4,0,rsi,SIZEOF_JCOEF)]
- por xmm0, XMMWORD [XMMBLOCK(5,0,rsi,SIZEOF_JCOEF)]
- por xmm1, XMMWORD [XMMBLOCK(6,0,rsi,SIZEOF_JCOEF)]
- por xmm0, XMMWORD [XMMBLOCK(7,0,rsi,SIZEOF_JCOEF)]
- por xmm1,xmm0
- packsswb xmm1,xmm1
- packsswb xmm1,xmm1
- movd eax,xmm1
- test rax,rax
- jnz short .columnDCT
-
- ; -- AC terms all zero
-
- movdqa xmm5, XMMWORD [XMMBLOCK(0,0,rsi,SIZEOF_JCOEF)]
- pmullw xmm5, XMMWORD [XMMBLOCK(0,0,rdx,SIZEOF_ISLOW_MULT_TYPE)]
-
- psllw xmm5,PASS1_BITS
-
- movdqa xmm4,xmm5 ; xmm5=in0=(00 01 02 03 04 05 06 07)
- punpcklwd xmm5,xmm5 ; xmm5=(00 00 01 01 02 02 03 03)
- punpckhwd xmm4,xmm4 ; xmm4=(04 04 05 05 06 06 07 07)
-
- pshufd xmm7,xmm5,0x00 ; xmm7=col0=(00 00 00 00 00 00 00 00)
- pshufd xmm6,xmm5,0x55 ; xmm6=col1=(01 01 01 01 01 01 01 01)
- pshufd xmm1,xmm5,0xAA ; xmm1=col2=(02 02 02 02 02 02 02 02)
- pshufd xmm5,xmm5,0xFF ; xmm5=col3=(03 03 03 03 03 03 03 03)
- pshufd xmm0,xmm4,0x00 ; xmm0=col4=(04 04 04 04 04 04 04 04)
- pshufd xmm3,xmm4,0x55 ; xmm3=col5=(05 05 05 05 05 05 05 05)
- pshufd xmm2,xmm4,0xAA ; xmm2=col6=(06 06 06 06 06 06 06 06)
- pshufd xmm4,xmm4,0xFF ; xmm4=col7=(07 07 07 07 07 07 07 07)
-
- movdqa XMMWORD [wk(8)], xmm6 ; wk(8)=col1
- movdqa XMMWORD [wk(9)], xmm5 ; wk(9)=col3
- movdqa XMMWORD [wk(10)], xmm3 ; wk(10)=col5
- movdqa XMMWORD [wk(11)], xmm4 ; wk(11)=col7
- jmp near .column_end
-%endif
-.columnDCT:
-
- ; -- Even part
-
- movdqa xmm0, XMMWORD [XMMBLOCK(0,0,rsi,SIZEOF_JCOEF)]
- movdqa xmm1, XMMWORD [XMMBLOCK(2,0,rsi,SIZEOF_JCOEF)]
- pmullw xmm0, XMMWORD [XMMBLOCK(0,0,rdx,SIZEOF_ISLOW_MULT_TYPE)]
- pmullw xmm1, XMMWORD [XMMBLOCK(2,0,rdx,SIZEOF_ISLOW_MULT_TYPE)]
- movdqa xmm2, XMMWORD [XMMBLOCK(4,0,rsi,SIZEOF_JCOEF)]
- movdqa xmm3, XMMWORD [XMMBLOCK(6,0,rsi,SIZEOF_JCOEF)]
- pmullw xmm2, XMMWORD [XMMBLOCK(4,0,rdx,SIZEOF_ISLOW_MULT_TYPE)]
- pmullw xmm3, XMMWORD [XMMBLOCK(6,0,rdx,SIZEOF_ISLOW_MULT_TYPE)]
-
- ; (Original)
- ; z1 = (z2 + z3) * 0.541196100;
- ; tmp2 = z1 + z3 * -1.847759065;
- ; tmp3 = z1 + z2 * 0.765366865;
- ;
- ; (This implementation)
- ; tmp2 = z2 * 0.541196100 + z3 * (0.541196100 - 1.847759065);
- ; tmp3 = z2 * (0.541196100 + 0.765366865) + z3 * 0.541196100;
-
- movdqa xmm4,xmm1 ; xmm1=in2=z2
- movdqa xmm5,xmm1
- punpcklwd xmm4,xmm3 ; xmm3=in6=z3
- punpckhwd xmm5,xmm3
- movdqa xmm1,xmm4
- movdqa xmm3,xmm5
- pmaddwd xmm4,[rel PW_F130_F054] ; xmm4=tmp3L
- pmaddwd xmm5,[rel PW_F130_F054] ; xmm5=tmp3H
- pmaddwd xmm1,[rel PW_F054_MF130] ; xmm1=tmp2L
- pmaddwd xmm3,[rel PW_F054_MF130] ; xmm3=tmp2H
-
- movdqa xmm6,xmm0
- paddw xmm0,xmm2 ; xmm0=in0+in4
- psubw xmm6,xmm2 ; xmm6=in0-in4
-
- pxor xmm7,xmm7
- pxor xmm2,xmm2
- punpcklwd xmm7,xmm0 ; xmm7=tmp0L
- punpckhwd xmm2,xmm0 ; xmm2=tmp0H
- psrad xmm7,(16-CONST_BITS) ; psrad xmm7,16 & pslld xmm7,CONST_BITS
- psrad xmm2,(16-CONST_BITS) ; psrad xmm2,16 & pslld xmm2,CONST_BITS
-
- movdqa xmm0,xmm7
- paddd xmm7,xmm4 ; xmm7=tmp10L
- psubd xmm0,xmm4 ; xmm0=tmp13L
- movdqa xmm4,xmm2
- paddd xmm2,xmm5 ; xmm2=tmp10H
- psubd xmm4,xmm5 ; xmm4=tmp13H
-
- movdqa XMMWORD [wk(0)], xmm7 ; wk(0)=tmp10L
- movdqa XMMWORD [wk(1)], xmm2 ; wk(1)=tmp10H
- movdqa XMMWORD [wk(2)], xmm0 ; wk(2)=tmp13L
- movdqa XMMWORD [wk(3)], xmm4 ; wk(3)=tmp13H
-
- pxor xmm5,xmm5
- pxor xmm7,xmm7
- punpcklwd xmm5,xmm6 ; xmm5=tmp1L
- punpckhwd xmm7,xmm6 ; xmm7=tmp1H
- psrad xmm5,(16-CONST_BITS) ; psrad xmm5,16 & pslld xmm5,CONST_BITS
- psrad xmm7,(16-CONST_BITS) ; psrad xmm7,16 & pslld xmm7,CONST_BITS
-
- movdqa xmm2,xmm5
- paddd xmm5,xmm1 ; xmm5=tmp11L
- psubd xmm2,xmm1 ; xmm2=tmp12L
- movdqa xmm0,xmm7
- paddd xmm7,xmm3 ; xmm7=tmp11H
- psubd xmm0,xmm3 ; xmm0=tmp12H
-
- movdqa XMMWORD [wk(4)], xmm5 ; wk(4)=tmp11L
- movdqa XMMWORD [wk(5)], xmm7 ; wk(5)=tmp11H
- movdqa XMMWORD [wk(6)], xmm2 ; wk(6)=tmp12L
- movdqa XMMWORD [wk(7)], xmm0 ; wk(7)=tmp12H
-
- ; -- Odd part
-
- movdqa xmm4, XMMWORD [XMMBLOCK(1,0,rsi,SIZEOF_JCOEF)]
- movdqa xmm6, XMMWORD [XMMBLOCK(3,0,rsi,SIZEOF_JCOEF)]
- pmullw xmm4, XMMWORD [XMMBLOCK(1,0,rdx,SIZEOF_ISLOW_MULT_TYPE)]
- pmullw xmm6, XMMWORD [XMMBLOCK(3,0,rdx,SIZEOF_ISLOW_MULT_TYPE)]
- movdqa xmm1, XMMWORD [XMMBLOCK(5,0,rsi,SIZEOF_JCOEF)]
- movdqa xmm3, XMMWORD [XMMBLOCK(7,0,rsi,SIZEOF_JCOEF)]
- pmullw xmm1, XMMWORD [XMMBLOCK(5,0,rdx,SIZEOF_ISLOW_MULT_TYPE)]
- pmullw xmm3, XMMWORD [XMMBLOCK(7,0,rdx,SIZEOF_ISLOW_MULT_TYPE)]
-
- movdqa xmm5,xmm6
- movdqa xmm7,xmm4
- paddw xmm5,xmm3 ; xmm5=z3
- paddw xmm7,xmm1 ; xmm7=z4
-
- ; (Original)
- ; z5 = (z3 + z4) * 1.175875602;
- ; z3 = z3 * -1.961570560; z4 = z4 * -0.390180644;
- ; z3 += z5; z4 += z5;
- ;
- ; (This implementation)
- ; z3 = z3 * (1.175875602 - 1.961570560) + z4 * 1.175875602;
- ; z4 = z3 * 1.175875602 + z4 * (1.175875602 - 0.390180644);
-
- movdqa xmm2,xmm5
- movdqa xmm0,xmm5
- punpcklwd xmm2,xmm7
- punpckhwd xmm0,xmm7
- movdqa xmm5,xmm2
- movdqa xmm7,xmm0
- pmaddwd xmm2,[rel PW_MF078_F117] ; xmm2=z3L
- pmaddwd xmm0,[rel PW_MF078_F117] ; xmm0=z3H
- pmaddwd xmm5,[rel PW_F117_F078] ; xmm5=z4L
- pmaddwd xmm7,[rel PW_F117_F078] ; xmm7=z4H
-
- movdqa XMMWORD [wk(10)], xmm2 ; wk(10)=z3L
- movdqa XMMWORD [wk(11)], xmm0 ; wk(11)=z3H
-
- ; (Original)
- ; z1 = tmp0 + tmp3; z2 = tmp1 + tmp2;
- ; tmp0 = tmp0 * 0.298631336; tmp1 = tmp1 * 2.053119869;
- ; tmp2 = tmp2 * 3.072711026; tmp3 = tmp3 * 1.501321110;
- ; z1 = z1 * -0.899976223; z2 = z2 * -2.562915447;
- ; tmp0 += z1 + z3; tmp1 += z2 + z4;
- ; tmp2 += z2 + z3; tmp3 += z1 + z4;
- ;
- ; (This implementation)
- ; tmp0 = tmp0 * (0.298631336 - 0.899976223) + tmp3 * -0.899976223;
- ; tmp1 = tmp1 * (2.053119869 - 2.562915447) + tmp2 * -2.562915447;
- ; tmp2 = tmp1 * -2.562915447 + tmp2 * (3.072711026 - 2.562915447);
- ; tmp3 = tmp0 * -0.899976223 + tmp3 * (1.501321110 - 0.899976223);
- ; tmp0 += z3; tmp1 += z4;
- ; tmp2 += z3; tmp3 += z4;
-
- movdqa xmm2,xmm3
- movdqa xmm0,xmm3
- punpcklwd xmm2,xmm4
- punpckhwd xmm0,xmm4
- movdqa xmm3,xmm2
- movdqa xmm4,xmm0
- pmaddwd xmm2,[rel PW_MF060_MF089] ; xmm2=tmp0L
- pmaddwd xmm0,[rel PW_MF060_MF089] ; xmm0=tmp0H
- pmaddwd xmm3,[rel PW_MF089_F060] ; xmm3=tmp3L
- pmaddwd xmm4,[rel PW_MF089_F060] ; xmm4=tmp3H
-
- paddd xmm2, XMMWORD [wk(10)] ; xmm2=tmp0L
- paddd xmm0, XMMWORD [wk(11)] ; xmm0=tmp0H
- paddd xmm3,xmm5 ; xmm3=tmp3L
- paddd xmm4,xmm7 ; xmm4=tmp3H
-
- movdqa XMMWORD [wk(8)], xmm2 ; wk(8)=tmp0L
- movdqa XMMWORD [wk(9)], xmm0 ; wk(9)=tmp0H
-
- movdqa xmm2,xmm1
- movdqa xmm0,xmm1
- punpcklwd xmm2,xmm6
- punpckhwd xmm0,xmm6
- movdqa xmm1,xmm2
- movdqa xmm6,xmm0
- pmaddwd xmm2,[rel PW_MF050_MF256] ; xmm2=tmp1L
- pmaddwd xmm0,[rel PW_MF050_MF256] ; xmm0=tmp1H
- pmaddwd xmm1,[rel PW_MF256_F050] ; xmm1=tmp2L
- pmaddwd xmm6,[rel PW_MF256_F050] ; xmm6=tmp2H
-
- paddd xmm2,xmm5 ; xmm2=tmp1L
- paddd xmm0,xmm7 ; xmm0=tmp1H
- paddd xmm1, XMMWORD [wk(10)] ; xmm1=tmp2L
- paddd xmm6, XMMWORD [wk(11)] ; xmm6=tmp2H
-
- movdqa XMMWORD [wk(10)], xmm2 ; wk(10)=tmp1L
- movdqa XMMWORD [wk(11)], xmm0 ; wk(11)=tmp1H
-
- ; -- Final output stage
-
- movdqa xmm5, XMMWORD [wk(0)] ; xmm5=tmp10L
- movdqa xmm7, XMMWORD [wk(1)] ; xmm7=tmp10H
-
- movdqa xmm2,xmm5
- movdqa xmm0,xmm7
- paddd xmm5,xmm3 ; xmm5=data0L
- paddd xmm7,xmm4 ; xmm7=data0H
- psubd xmm2,xmm3 ; xmm2=data7L
- psubd xmm0,xmm4 ; xmm0=data7H
-
- movdqa xmm3,[rel PD_DESCALE_P1] ; xmm3=[rel PD_DESCALE_P1]
-
- paddd xmm5,xmm3
- paddd xmm7,xmm3
- psrad xmm5,DESCALE_P1
- psrad xmm7,DESCALE_P1
- paddd xmm2,xmm3
- paddd xmm0,xmm3
- psrad xmm2,DESCALE_P1
- psrad xmm0,DESCALE_P1
-
- packssdw xmm5,xmm7 ; xmm5=data0=(00 01 02 03 04 05 06 07)
- packssdw xmm2,xmm0 ; xmm2=data7=(70 71 72 73 74 75 76 77)
-
- movdqa xmm4, XMMWORD [wk(4)] ; xmm4=tmp11L
- movdqa xmm3, XMMWORD [wk(5)] ; xmm3=tmp11H
-
- movdqa xmm7,xmm4
- movdqa xmm0,xmm3
- paddd xmm4,xmm1 ; xmm4=data1L
- paddd xmm3,xmm6 ; xmm3=data1H
- psubd xmm7,xmm1 ; xmm7=data6L
- psubd xmm0,xmm6 ; xmm0=data6H
-
- movdqa xmm1,[rel PD_DESCALE_P1] ; xmm1=[rel PD_DESCALE_P1]
-
- paddd xmm4,xmm1
- paddd xmm3,xmm1
- psrad xmm4,DESCALE_P1
- psrad xmm3,DESCALE_P1
- paddd xmm7,xmm1
- paddd xmm0,xmm1
- psrad xmm7,DESCALE_P1
- psrad xmm0,DESCALE_P1
-
- packssdw xmm4,xmm3 ; xmm4=data1=(10 11 12 13 14 15 16 17)
- packssdw xmm7,xmm0 ; xmm7=data6=(60 61 62 63 64 65 66 67)
-
- movdqa xmm6,xmm5 ; transpose coefficients(phase 1)
- punpcklwd xmm5,xmm4 ; xmm5=(00 10 01 11 02 12 03 13)
- punpckhwd xmm6,xmm4 ; xmm6=(04 14 05 15 06 16 07 17)
- movdqa xmm1,xmm7 ; transpose coefficients(phase 1)
- punpcklwd xmm7,xmm2 ; xmm7=(60 70 61 71 62 72 63 73)
- punpckhwd xmm1,xmm2 ; xmm1=(64 74 65 75 66 76 67 77)
-
- movdqa xmm3, XMMWORD [wk(6)] ; xmm3=tmp12L
- movdqa xmm0, XMMWORD [wk(7)] ; xmm0=tmp12H
- movdqa xmm4, XMMWORD [wk(10)] ; xmm4=tmp1L
- movdqa xmm2, XMMWORD [wk(11)] ; xmm2=tmp1H
-
- movdqa XMMWORD [wk(0)], xmm5 ; wk(0)=(00 10 01 11 02 12 03 13)
- movdqa XMMWORD [wk(1)], xmm6 ; wk(1)=(04 14 05 15 06 16 07 17)
- movdqa XMMWORD [wk(4)], xmm7 ; wk(4)=(60 70 61 71 62 72 63 73)
- movdqa XMMWORD [wk(5)], xmm1 ; wk(5)=(64 74 65 75 66 76 67 77)
-
- movdqa xmm5,xmm3
- movdqa xmm6,xmm0
- paddd xmm3,xmm4 ; xmm3=data2L
- paddd xmm0,xmm2 ; xmm0=data2H
- psubd xmm5,xmm4 ; xmm5=data5L
- psubd xmm6,xmm2 ; xmm6=data5H
-
- movdqa xmm7,[rel PD_DESCALE_P1] ; xmm7=[rel PD_DESCALE_P1]
-
- paddd xmm3,xmm7
- paddd xmm0,xmm7
- psrad xmm3,DESCALE_P1
- psrad xmm0,DESCALE_P1
- paddd xmm5,xmm7
- paddd xmm6,xmm7
- psrad xmm5,DESCALE_P1
- psrad xmm6,DESCALE_P1
-
- packssdw xmm3,xmm0 ; xmm3=data2=(20 21 22 23 24 25 26 27)
- packssdw xmm5,xmm6 ; xmm5=data5=(50 51 52 53 54 55 56 57)
-
- movdqa xmm1, XMMWORD [wk(2)] ; xmm1=tmp13L
- movdqa xmm4, XMMWORD [wk(3)] ; xmm4=tmp13H
- movdqa xmm2, XMMWORD [wk(8)] ; xmm2=tmp0L
- movdqa xmm7, XMMWORD [wk(9)] ; xmm7=tmp0H
-
- movdqa xmm0,xmm1
- movdqa xmm6,xmm4
- paddd xmm1,xmm2 ; xmm1=data3L
- paddd xmm4,xmm7 ; xmm4=data3H
- psubd xmm0,xmm2 ; xmm0=data4L
- psubd xmm6,xmm7 ; xmm6=data4H
-
- movdqa xmm2,[rel PD_DESCALE_P1] ; xmm2=[rel PD_DESCALE_P1]
-
- paddd xmm1,xmm2
- paddd xmm4,xmm2
- psrad xmm1,DESCALE_P1
- psrad xmm4,DESCALE_P1
- paddd xmm0,xmm2
- paddd xmm6,xmm2
- psrad xmm0,DESCALE_P1
- psrad xmm6,DESCALE_P1
-
- packssdw xmm1,xmm4 ; xmm1=data3=(30 31 32 33 34 35 36 37)
- packssdw xmm0,xmm6 ; xmm0=data4=(40 41 42 43 44 45 46 47)
-
- movdqa xmm7, XMMWORD [wk(0)] ; xmm7=(00 10 01 11 02 12 03 13)
- movdqa xmm2, XMMWORD [wk(1)] ; xmm2=(04 14 05 15 06 16 07 17)
-
- movdqa xmm4,xmm3 ; transpose coefficients(phase 1)
- punpcklwd xmm3,xmm1 ; xmm3=(20 30 21 31 22 32 23 33)
- punpckhwd xmm4,xmm1 ; xmm4=(24 34 25 35 26 36 27 37)
- movdqa xmm6,xmm0 ; transpose coefficients(phase 1)
- punpcklwd xmm0,xmm5 ; xmm0=(40 50 41 51 42 52 43 53)
- punpckhwd xmm6,xmm5 ; xmm6=(44 54 45 55 46 56 47 57)
-
- movdqa xmm1,xmm7 ; transpose coefficients(phase 2)
- punpckldq xmm7,xmm3 ; xmm7=(00 10 20 30 01 11 21 31)
- punpckhdq xmm1,xmm3 ; xmm1=(02 12 22 32 03 13 23 33)
- movdqa xmm5,xmm2 ; transpose coefficients(phase 2)
- punpckldq xmm2,xmm4 ; xmm2=(04 14 24 34 05 15 25 35)
- punpckhdq xmm5,xmm4 ; xmm5=(06 16 26 36 07 17 27 37)
-
- movdqa xmm3, XMMWORD [wk(4)] ; xmm3=(60 70 61 71 62 72 63 73)
- movdqa xmm4, XMMWORD [wk(5)] ; xmm4=(64 74 65 75 66 76 67 77)
-
- movdqa XMMWORD [wk(6)], xmm2 ; wk(6)=(04 14 24 34 05 15 25 35)
- movdqa XMMWORD [wk(7)], xmm5 ; wk(7)=(06 16 26 36 07 17 27 37)
-
- movdqa xmm2,xmm0 ; transpose coefficients(phase 2)
- punpckldq xmm0,xmm3 ; xmm0=(40 50 60 70 41 51 61 71)
- punpckhdq xmm2,xmm3 ; xmm2=(42 52 62 72 43 53 63 73)
- movdqa xmm5,xmm6 ; transpose coefficients(phase 2)
- punpckldq xmm6,xmm4 ; xmm6=(44 54 64 74 45 55 65 75)
- punpckhdq xmm5,xmm4 ; xmm5=(46 56 66 76 47 57 67 77)
-
- movdqa xmm3,xmm7 ; transpose coefficients(phase 3)
- punpcklqdq xmm7,xmm0 ; xmm7=col0=(00 10 20 30 40 50 60 70)
- punpckhqdq xmm3,xmm0 ; xmm3=col1=(01 11 21 31 41 51 61 71)
- movdqa xmm4,xmm1 ; transpose coefficients(phase 3)
- punpcklqdq xmm1,xmm2 ; xmm1=col2=(02 12 22 32 42 52 62 72)
- punpckhqdq xmm4,xmm2 ; xmm4=col3=(03 13 23 33 43 53 63 73)
-
- movdqa xmm0, XMMWORD [wk(6)] ; xmm0=(04 14 24 34 05 15 25 35)
- movdqa xmm2, XMMWORD [wk(7)] ; xmm2=(06 16 26 36 07 17 27 37)
-
- movdqa XMMWORD [wk(8)], xmm3 ; wk(8)=col1
- movdqa XMMWORD [wk(9)], xmm4 ; wk(9)=col3
-
- movdqa xmm3,xmm0 ; transpose coefficients(phase 3)
- punpcklqdq xmm0,xmm6 ; xmm0=col4=(04 14 24 34 44 54 64 74)
- punpckhqdq xmm3,xmm6 ; xmm3=col5=(05 15 25 35 45 55 65 75)
- movdqa xmm4,xmm2 ; transpose coefficients(phase 3)
- punpcklqdq xmm2,xmm5 ; xmm2=col6=(06 16 26 36 46 56 66 76)
- punpckhqdq xmm4,xmm5 ; xmm4=col7=(07 17 27 37 47 57 67 77)
-
- movdqa XMMWORD [wk(10)], xmm3 ; wk(10)=col5
- movdqa XMMWORD [wk(11)], xmm4 ; wk(11)=col7
-.column_end:
-
- ; -- Prefetch the next coefficient block
-
- prefetchnta [rsi + DCTSIZE2*SIZEOF_JCOEF + 0*32]
- prefetchnta [rsi + DCTSIZE2*SIZEOF_JCOEF + 1*32]
- prefetchnta [rsi + DCTSIZE2*SIZEOF_JCOEF + 2*32]
- prefetchnta [rsi + DCTSIZE2*SIZEOF_JCOEF + 3*32]
-
- ; ---- Pass 2: process rows from work array, store into output array.
-
- mov rax, [original_rbp]
- mov rdi, r12 ; (JSAMPROW *)
- mov eax, r13d
-
- ; -- Even part
-
- ; xmm7=col0, xmm1=col2, xmm0=col4, xmm2=col6
-
- ; (Original)
- ; z1 = (z2 + z3) * 0.541196100;
- ; tmp2 = z1 + z3 * -1.847759065;
- ; tmp3 = z1 + z2 * 0.765366865;
- ;
- ; (This implementation)
- ; tmp2 = z2 * 0.541196100 + z3 * (0.541196100 - 1.847759065);
- ; tmp3 = z2 * (0.541196100 + 0.765366865) + z3 * 0.541196100;
-
- movdqa xmm6,xmm1 ; xmm1=in2=z2
- movdqa xmm5,xmm1
- punpcklwd xmm6,xmm2 ; xmm2=in6=z3
- punpckhwd xmm5,xmm2
- movdqa xmm1,xmm6
- movdqa xmm2,xmm5
- pmaddwd xmm6,[rel PW_F130_F054] ; xmm6=tmp3L
- pmaddwd xmm5,[rel PW_F130_F054] ; xmm5=tmp3H
- pmaddwd xmm1,[rel PW_F054_MF130] ; xmm1=tmp2L
- pmaddwd xmm2,[rel PW_F054_MF130] ; xmm2=tmp2H
-
- movdqa xmm3,xmm7
- paddw xmm7,xmm0 ; xmm7=in0+in4
- psubw xmm3,xmm0 ; xmm3=in0-in4
-
- pxor xmm4,xmm4
- pxor xmm0,xmm0
- punpcklwd xmm4,xmm7 ; xmm4=tmp0L
- punpckhwd xmm0,xmm7 ; xmm0=tmp0H
- psrad xmm4,(16-CONST_BITS) ; psrad xmm4,16 & pslld xmm4,CONST_BITS
- psrad xmm0,(16-CONST_BITS) ; psrad xmm0,16 & pslld xmm0,CONST_BITS
-
- movdqa xmm7,xmm4
- paddd xmm4,xmm6 ; xmm4=tmp10L
- psubd xmm7,xmm6 ; xmm7=tmp13L
- movdqa xmm6,xmm0
- paddd xmm0,xmm5 ; xmm0=tmp10H
- psubd xmm6,xmm5 ; xmm6=tmp13H
-
- movdqa XMMWORD [wk(0)], xmm4 ; wk(0)=tmp10L
- movdqa XMMWORD [wk(1)], xmm0 ; wk(1)=tmp10H
- movdqa XMMWORD [wk(2)], xmm7 ; wk(2)=tmp13L
- movdqa XMMWORD [wk(3)], xmm6 ; wk(3)=tmp13H
-
- pxor xmm5,xmm5
- pxor xmm4,xmm4
- punpcklwd xmm5,xmm3 ; xmm5=tmp1L
- punpckhwd xmm4,xmm3 ; xmm4=tmp1H
- psrad xmm5,(16-CONST_BITS) ; psrad xmm5,16 & pslld xmm5,CONST_BITS
- psrad xmm4,(16-CONST_BITS) ; psrad xmm4,16 & pslld xmm4,CONST_BITS
-
- movdqa xmm0,xmm5
- paddd xmm5,xmm1 ; xmm5=tmp11L
- psubd xmm0,xmm1 ; xmm0=tmp12L
- movdqa xmm7,xmm4
- paddd xmm4,xmm2 ; xmm4=tmp11H
- psubd xmm7,xmm2 ; xmm7=tmp12H
-
- movdqa XMMWORD [wk(4)], xmm5 ; wk(4)=tmp11L
- movdqa XMMWORD [wk(5)], xmm4 ; wk(5)=tmp11H
- movdqa XMMWORD [wk(6)], xmm0 ; wk(6)=tmp12L
- movdqa XMMWORD [wk(7)], xmm7 ; wk(7)=tmp12H
-
- ; -- Odd part
-
- movdqa xmm6, XMMWORD [wk(9)] ; xmm6=col3
- movdqa xmm3, XMMWORD [wk(8)] ; xmm3=col1
- movdqa xmm1, XMMWORD [wk(11)] ; xmm1=col7
- movdqa xmm2, XMMWORD [wk(10)] ; xmm2=col5
-
- movdqa xmm5,xmm6
- movdqa xmm4,xmm3
- paddw xmm5,xmm1 ; xmm5=z3
- paddw xmm4,xmm2 ; xmm4=z4
-
- ; (Original)
- ; z5 = (z3 + z4) * 1.175875602;
- ; z3 = z3 * -1.961570560; z4 = z4 * -0.390180644;
- ; z3 += z5; z4 += z5;
- ;
- ; (This implementation)
- ; z3 = z3 * (1.175875602 - 1.961570560) + z4 * 1.175875602;
- ; z4 = z3 * 1.175875602 + z4 * (1.175875602 - 0.390180644);
-
- movdqa xmm0,xmm5
- movdqa xmm7,xmm5
- punpcklwd xmm0,xmm4
- punpckhwd xmm7,xmm4
- movdqa xmm5,xmm0
- movdqa xmm4,xmm7
- pmaddwd xmm0,[rel PW_MF078_F117] ; xmm0=z3L
- pmaddwd xmm7,[rel PW_MF078_F117] ; xmm7=z3H
- pmaddwd xmm5,[rel PW_F117_F078] ; xmm5=z4L
- pmaddwd xmm4,[rel PW_F117_F078] ; xmm4=z4H
-
- movdqa XMMWORD [wk(10)], xmm0 ; wk(10)=z3L
- movdqa XMMWORD [wk(11)], xmm7 ; wk(11)=z3H
-
- ; (Original)
- ; z1 = tmp0 + tmp3; z2 = tmp1 + tmp2;
- ; tmp0 = tmp0 * 0.298631336; tmp1 = tmp1 * 2.053119869;
- ; tmp2 = tmp2 * 3.072711026; tmp3 = tmp3 * 1.501321110;
- ; z1 = z1 * -0.899976223; z2 = z2 * -2.562915447;
- ; tmp0 += z1 + z3; tmp1 += z2 + z4;
- ; tmp2 += z2 + z3; tmp3 += z1 + z4;
- ;
- ; (This implementation)
- ; tmp0 = tmp0 * (0.298631336 - 0.899976223) + tmp3 * -0.899976223;
- ; tmp1 = tmp1 * (2.053119869 - 2.562915447) + tmp2 * -2.562915447;
- ; tmp2 = tmp1 * -2.562915447 + tmp2 * (3.072711026 - 2.562915447);
- ; tmp3 = tmp0 * -0.899976223 + tmp3 * (1.501321110 - 0.899976223);
- ; tmp0 += z3; tmp1 += z4;
- ; tmp2 += z3; tmp3 += z4;
-
- movdqa xmm0,xmm1
- movdqa xmm7,xmm1
- punpcklwd xmm0,xmm3
- punpckhwd xmm7,xmm3
- movdqa xmm1,xmm0
- movdqa xmm3,xmm7
- pmaddwd xmm0,[rel PW_MF060_MF089] ; xmm0=tmp0L
- pmaddwd xmm7,[rel PW_MF060_MF089] ; xmm7=tmp0H
- pmaddwd xmm1,[rel PW_MF089_F060] ; xmm1=tmp3L
- pmaddwd xmm3,[rel PW_MF089_F060] ; xmm3=tmp3H
-
- paddd xmm0, XMMWORD [wk(10)] ; xmm0=tmp0L
- paddd xmm7, XMMWORD [wk(11)] ; xmm7=tmp0H
- paddd xmm1,xmm5 ; xmm1=tmp3L
- paddd xmm3,xmm4 ; xmm3=tmp3H
-
- movdqa XMMWORD [wk(8)], xmm0 ; wk(8)=tmp0L
- movdqa XMMWORD [wk(9)], xmm7 ; wk(9)=tmp0H
-
- movdqa xmm0,xmm2
- movdqa xmm7,xmm2
- punpcklwd xmm0,xmm6
- punpckhwd xmm7,xmm6
- movdqa xmm2,xmm0
- movdqa xmm6,xmm7
- pmaddwd xmm0,[rel PW_MF050_MF256] ; xmm0=tmp1L
- pmaddwd xmm7,[rel PW_MF050_MF256] ; xmm7=tmp1H
- pmaddwd xmm2,[rel PW_MF256_F050] ; xmm2=tmp2L
- pmaddwd xmm6,[rel PW_MF256_F050] ; xmm6=tmp2H
-
- paddd xmm0,xmm5 ; xmm0=tmp1L
- paddd xmm7,xmm4 ; xmm7=tmp1H
- paddd xmm2, XMMWORD [wk(10)] ; xmm2=tmp2L
- paddd xmm6, XMMWORD [wk(11)] ; xmm6=tmp2H
-
- movdqa XMMWORD [wk(10)], xmm0 ; wk(10)=tmp1L
- movdqa XMMWORD [wk(11)], xmm7 ; wk(11)=tmp1H
-
- ; -- Final output stage
-
- movdqa xmm5, XMMWORD [wk(0)] ; xmm5=tmp10L
- movdqa xmm4, XMMWORD [wk(1)] ; xmm4=tmp10H
-
- movdqa xmm0,xmm5
- movdqa xmm7,xmm4
- paddd xmm5,xmm1 ; xmm5=data0L
- paddd xmm4,xmm3 ; xmm4=data0H
- psubd xmm0,xmm1 ; xmm0=data7L
- psubd xmm7,xmm3 ; xmm7=data7H
-
- movdqa xmm1,[rel PD_DESCALE_P2] ; xmm1=[rel PD_DESCALE_P2]
-
- paddd xmm5,xmm1
- paddd xmm4,xmm1
- psrad xmm5,DESCALE_P2
- psrad xmm4,DESCALE_P2
- paddd xmm0,xmm1
- paddd xmm7,xmm1
- psrad xmm0,DESCALE_P2
- psrad xmm7,DESCALE_P2
-
- packssdw xmm5,xmm4 ; xmm5=data0=(00 10 20 30 40 50 60 70)
- packssdw xmm0,xmm7 ; xmm0=data7=(07 17 27 37 47 57 67 77)
-
- movdqa xmm3, XMMWORD [wk(4)] ; xmm3=tmp11L
- movdqa xmm1, XMMWORD [wk(5)] ; xmm1=tmp11H
-
- movdqa xmm4,xmm3
- movdqa xmm7,xmm1
- paddd xmm3,xmm2 ; xmm3=data1L
- paddd xmm1,xmm6 ; xmm1=data1H
- psubd xmm4,xmm2 ; xmm4=data6L
- psubd xmm7,xmm6 ; xmm7=data6H
-
- movdqa xmm2,[rel PD_DESCALE_P2] ; xmm2=[rel PD_DESCALE_P2]
-
- paddd xmm3,xmm2
- paddd xmm1,xmm2
- psrad xmm3,DESCALE_P2
- psrad xmm1,DESCALE_P2
- paddd xmm4,xmm2
- paddd xmm7,xmm2
- psrad xmm4,DESCALE_P2
- psrad xmm7,DESCALE_P2
-
- packssdw xmm3,xmm1 ; xmm3=data1=(01 11 21 31 41 51 61 71)
- packssdw xmm4,xmm7 ; xmm4=data6=(06 16 26 36 46 56 66 76)
-
- packsswb xmm5,xmm4 ; xmm5=(00 10 20 30 40 50 60 70 06 16 26 36 46 56 66 76)
- packsswb xmm3,xmm0 ; xmm3=(01 11 21 31 41 51 61 71 07 17 27 37 47 57 67 77)
-
- movdqa xmm6, XMMWORD [wk(6)] ; xmm6=tmp12L
- movdqa xmm2, XMMWORD [wk(7)] ; xmm2=tmp12H
- movdqa xmm1, XMMWORD [wk(10)] ; xmm1=tmp1L
- movdqa xmm7, XMMWORD [wk(11)] ; xmm7=tmp1H
-
- movdqa XMMWORD [wk(0)], xmm5 ; wk(0)=(00 10 20 30 40 50 60 70 06 16 26 36 46 56 66 76)
- movdqa XMMWORD [wk(1)], xmm3 ; wk(1)=(01 11 21 31 41 51 61 71 07 17 27 37 47 57 67 77)
-
- movdqa xmm4,xmm6
- movdqa xmm0,xmm2
- paddd xmm6,xmm1 ; xmm6=data2L
- paddd xmm2,xmm7 ; xmm2=data2H
- psubd xmm4,xmm1 ; xmm4=data5L
- psubd xmm0,xmm7 ; xmm0=data5H
-
- movdqa xmm5,[rel PD_DESCALE_P2] ; xmm5=[rel PD_DESCALE_P2]
-
- paddd xmm6,xmm5
- paddd xmm2,xmm5
- psrad xmm6,DESCALE_P2
- psrad xmm2,DESCALE_P2
- paddd xmm4,xmm5
- paddd xmm0,xmm5
- psrad xmm4,DESCALE_P2
- psrad xmm0,DESCALE_P2
-
- packssdw xmm6,xmm2 ; xmm6=data2=(02 12 22 32 42 52 62 72)
- packssdw xmm4,xmm0 ; xmm4=data5=(05 15 25 35 45 55 65 75)
-
- movdqa xmm3, XMMWORD [wk(2)] ; xmm3=tmp13L
- movdqa xmm1, XMMWORD [wk(3)] ; xmm1=tmp13H
- movdqa xmm7, XMMWORD [wk(8)] ; xmm7=tmp0L
- movdqa xmm5, XMMWORD [wk(9)] ; xmm5=tmp0H
-
- movdqa xmm2,xmm3
- movdqa xmm0,xmm1
- paddd xmm3,xmm7 ; xmm3=data3L
- paddd xmm1,xmm5 ; xmm1=data3H
- psubd xmm2,xmm7 ; xmm2=data4L
- psubd xmm0,xmm5 ; xmm0=data4H
-
- movdqa xmm7,[rel PD_DESCALE_P2] ; xmm7=[rel PD_DESCALE_P2]
-
- paddd xmm3,xmm7
- paddd xmm1,xmm7
- psrad xmm3,DESCALE_P2
- psrad xmm1,DESCALE_P2
- paddd xmm2,xmm7
- paddd xmm0,xmm7
- psrad xmm2,DESCALE_P2
- psrad xmm0,DESCALE_P2
-
- movdqa xmm5,[rel PB_CENTERJSAMP] ; xmm5=[rel PB_CENTERJSAMP]
-
- packssdw xmm3,xmm1 ; xmm3=data3=(03 13 23 33 43 53 63 73)
- packssdw xmm2,xmm0 ; xmm2=data4=(04 14 24 34 44 54 64 74)
-
- movdqa xmm7, XMMWORD [wk(0)] ; xmm7=(00 10 20 30 40 50 60 70 06 16 26 36 46 56 66 76)
- movdqa xmm1, XMMWORD [wk(1)] ; xmm1=(01 11 21 31 41 51 61 71 07 17 27 37 47 57 67 77)
-
- packsswb xmm6,xmm2 ; xmm6=(02 12 22 32 42 52 62 72 04 14 24 34 44 54 64 74)
- packsswb xmm3,xmm4 ; xmm3=(03 13 23 33 43 53 63 73 05 15 25 35 45 55 65 75)
-
- paddb xmm7,xmm5
- paddb xmm1,xmm5
- paddb xmm6,xmm5
- paddb xmm3,xmm5
-
- movdqa xmm0,xmm7 ; transpose coefficients(phase 1)
- punpcklbw xmm7,xmm1 ; xmm7=(00 01 10 11 20 21 30 31 40 41 50 51 60 61 70 71)
- punpckhbw xmm0,xmm1 ; xmm0=(06 07 16 17 26 27 36 37 46 47 56 57 66 67 76 77)
- movdqa xmm2,xmm6 ; transpose coefficients(phase 1)
- punpcklbw xmm6,xmm3 ; xmm6=(02 03 12 13 22 23 32 33 42 43 52 53 62 63 72 73)
- punpckhbw xmm2,xmm3 ; xmm2=(04 05 14 15 24 25 34 35 44 45 54 55 64 65 74 75)
-
- movdqa xmm4,xmm7 ; transpose coefficients(phase 2)
- punpcklwd xmm7,xmm6 ; xmm7=(00 01 02 03 10 11 12 13 20 21 22 23 30 31 32 33)
- punpckhwd xmm4,xmm6 ; xmm4=(40 41 42 43 50 51 52 53 60 61 62 63 70 71 72 73)
- movdqa xmm5,xmm2 ; transpose coefficients(phase 2)
- punpcklwd xmm2,xmm0 ; xmm2=(04 05 06 07 14 15 16 17 24 25 26 27 34 35 36 37)
- punpckhwd xmm5,xmm0 ; xmm5=(44 45 46 47 54 55 56 57 64 65 66 67 74 75 76 77)
-
- movdqa xmm1,xmm7 ; transpose coefficients(phase 3)
- punpckldq xmm7,xmm2 ; xmm7=(00 01 02 03 04 05 06 07 10 11 12 13 14 15 16 17)
- punpckhdq xmm1,xmm2 ; xmm1=(20 21 22 23 24 25 26 27 30 31 32 33 34 35 36 37)
- movdqa xmm3,xmm4 ; transpose coefficients(phase 3)
- punpckldq xmm4,xmm5 ; xmm4=(40 41 42 43 44 45 46 47 50 51 52 53 54 55 56 57)
- punpckhdq xmm3,xmm5 ; xmm3=(60 61 62 63 64 65 66 67 70 71 72 73 74 75 76 77)
-
- pshufd xmm6,xmm7,0x4E ; xmm6=(10 11 12 13 14 15 16 17 00 01 02 03 04 05 06 07)
- pshufd xmm0,xmm1,0x4E ; xmm0=(30 31 32 33 34 35 36 37 20 21 22 23 24 25 26 27)
- pshufd xmm2,xmm4,0x4E ; xmm2=(50 51 52 53 54 55 56 57 40 41 42 43 44 45 46 47)
- pshufd xmm5,xmm3,0x4E ; xmm5=(70 71 72 73 74 75 76 77 60 61 62 63 64 65 66 67)
-
- mov rdx, JSAMPROW [rdi+0*SIZEOF_JSAMPROW]
- mov rsi, JSAMPROW [rdi+2*SIZEOF_JSAMPROW]
- movq XMM_MMWORD [rdx+rax*SIZEOF_JSAMPLE], xmm7
- movq XMM_MMWORD [rsi+rax*SIZEOF_JSAMPLE], xmm1
- mov rdx, JSAMPROW [rdi+4*SIZEOF_JSAMPROW]
- mov rsi, JSAMPROW [rdi+6*SIZEOF_JSAMPROW]
- movq XMM_MMWORD [rdx+rax*SIZEOF_JSAMPLE], xmm4
- movq XMM_MMWORD [rsi+rax*SIZEOF_JSAMPLE], xmm3
-
- mov rdx, JSAMPROW [rdi+1*SIZEOF_JSAMPROW]
- mov rsi, JSAMPROW [rdi+3*SIZEOF_JSAMPROW]
- movq XMM_MMWORD [rdx+rax*SIZEOF_JSAMPLE], xmm6
- movq XMM_MMWORD [rsi+rax*SIZEOF_JSAMPLE], xmm0
- mov rdx, JSAMPROW [rdi+5*SIZEOF_JSAMPROW]
- mov rsi, JSAMPROW [rdi+7*SIZEOF_JSAMPROW]
- movq XMM_MMWORD [rdx+rax*SIZEOF_JSAMPLE], xmm2
- movq XMM_MMWORD [rsi+rax*SIZEOF_JSAMPLE], xmm5
-
- uncollect_args
- mov rsp,rbp ; rsp <- aligned rbp
- pop rsp ; rsp <- original rbp
- pop rbp
- ret
-
-; For some reason, the OS X linker does not honor the request to align the
-; segment unless we do this.
- align 16
diff --git a/simd/jidctint-sse2.asm b/simd/jidctint-sse2.asm
deleted file mode 100644
index 6c7e7d9..0000000
--- a/simd/jidctint-sse2.asm
+++ /dev/null
@@ -1,858 +0,0 @@
-;
-; jidctint.asm - accurate integer IDCT (SSE2)
-;
-; Copyright 2009 Pierre Ossman <ossman@cendio.se> for Cendio AB
-;
-; Based on the x86 SIMD extension for IJG JPEG library
-; Copyright (C) 1999-2006, MIYASAKA Masaru.
-; For conditions of distribution and use, see copyright notice in jsimdext.inc
-;
-; This file should be assembled with NASM (Netwide Assembler),
-; can *not* be assembled with Microsoft's MASM or any compatible
-; assembler (including Borland's Turbo Assembler).
-; NASM is available from http://nasm.sourceforge.net/ or
-; http://sourceforge.net/project/showfiles.php?group_id=6208
-;
-; This file contains a slow-but-accurate integer implementation of the
-; inverse DCT (Discrete Cosine Transform). The following code is based
-; directly on the IJG's original jidctint.c; see the jidctint.c for
-; more details.
-;
-; [TAB8]
-
-%include "jsimdext.inc"
-%include "jdct.inc"
-
-; --------------------------------------------------------------------------
-
-%define CONST_BITS 13
-%define PASS1_BITS 2
-
-%define DESCALE_P1 (CONST_BITS-PASS1_BITS)
-%define DESCALE_P2 (CONST_BITS+PASS1_BITS+3)
-
-%if CONST_BITS == 13
-F_0_298 equ 2446 ; FIX(0.298631336)
-F_0_390 equ 3196 ; FIX(0.390180644)
-F_0_541 equ 4433 ; FIX(0.541196100)
-F_0_765 equ 6270 ; FIX(0.765366865)
-F_0_899 equ 7373 ; FIX(0.899976223)
-F_1_175 equ 9633 ; FIX(1.175875602)
-F_1_501 equ 12299 ; FIX(1.501321110)
-F_1_847 equ 15137 ; FIX(1.847759065)
-F_1_961 equ 16069 ; FIX(1.961570560)
-F_2_053 equ 16819 ; FIX(2.053119869)
-F_2_562 equ 20995 ; FIX(2.562915447)
-F_3_072 equ 25172 ; FIX(3.072711026)
-%else
-; NASM cannot do compile-time arithmetic on floating-point constants.
-%define DESCALE(x,n) (((x)+(1<<((n)-1)))>>(n))
-F_0_298 equ DESCALE( 320652955,30-CONST_BITS) ; FIX(0.298631336)
-F_0_390 equ DESCALE( 418953276,30-CONST_BITS) ; FIX(0.390180644)
-F_0_541 equ DESCALE( 581104887,30-CONST_BITS) ; FIX(0.541196100)
-F_0_765 equ DESCALE( 821806413,30-CONST_BITS) ; FIX(0.765366865)
-F_0_899 equ DESCALE( 966342111,30-CONST_BITS) ; FIX(0.899976223)
-F_1_175 equ DESCALE(1262586813,30-CONST_BITS) ; FIX(1.175875602)
-F_1_501 equ DESCALE(1612031267,30-CONST_BITS) ; FIX(1.501321110)
-F_1_847 equ DESCALE(1984016188,30-CONST_BITS) ; FIX(1.847759065)
-F_1_961 equ DESCALE(2106220350,30-CONST_BITS) ; FIX(1.961570560)
-F_2_053 equ DESCALE(2204520673,30-CONST_BITS) ; FIX(2.053119869)
-F_2_562 equ DESCALE(2751909506,30-CONST_BITS) ; FIX(2.562915447)
-F_3_072 equ DESCALE(3299298341,30-CONST_BITS) ; FIX(3.072711026)
-%endif
-
-; --------------------------------------------------------------------------
- SECTION SEG_CONST
-
- alignz 16
- global EXTN(jconst_idct_islow_sse2)
-
-EXTN(jconst_idct_islow_sse2):
-
-PW_F130_F054 times 4 dw (F_0_541+F_0_765), F_0_541
-PW_F054_MF130 times 4 dw F_0_541, (F_0_541-F_1_847)
-PW_MF078_F117 times 4 dw (F_1_175-F_1_961), F_1_175
-PW_F117_F078 times 4 dw F_1_175, (F_1_175-F_0_390)
-PW_MF060_MF089 times 4 dw (F_0_298-F_0_899),-F_0_899
-PW_MF089_F060 times 4 dw -F_0_899, (F_1_501-F_0_899)
-PW_MF050_MF256 times 4 dw (F_2_053-F_2_562),-F_2_562
-PW_MF256_F050 times 4 dw -F_2_562, (F_3_072-F_2_562)
-PD_DESCALE_P1 times 4 dd 1 << (DESCALE_P1-1)
-PD_DESCALE_P2 times 4 dd 1 << (DESCALE_P2-1)
-PB_CENTERJSAMP times 16 db CENTERJSAMPLE
-
- alignz 16
-
-; --------------------------------------------------------------------------
- SECTION SEG_TEXT
- BITS 32
-;
-; Perform dequantization and inverse DCT on one block of coefficients.
-;
-; GLOBAL(void)
-; jsimd_idct_islow_sse2 (void *dct_table, JCOEFPTR coef_block,
-; JSAMPARRAY output_buf, JDIMENSION output_col)
-;
-
-%define dct_table(b) (b)+8 ; jpeg_component_info *compptr
-%define coef_block(b) (b)+12 ; JCOEFPTR coef_block
-%define output_buf(b) (b)+16 ; JSAMPARRAY output_buf
-%define output_col(b) (b)+20 ; JDIMENSION output_col
-
-%define original_ebp ebp+0
-%define wk(i) ebp-(WK_NUM-(i))*SIZEOF_XMMWORD ; xmmword wk[WK_NUM]
-%define WK_NUM 12
-
- align 16
- global EXTN(jsimd_idct_islow_sse2)
-
-EXTN(jsimd_idct_islow_sse2):
- push ebp
- mov eax,esp ; eax = original ebp
- sub esp, byte 4
- and esp, byte (-SIZEOF_XMMWORD) ; align to 128 bits
- mov [esp],eax
- mov ebp,esp ; ebp = aligned ebp
- lea esp, [wk(0)]
- pushpic ebx
-; push ecx ; unused
-; push edx ; need not be preserved
- push esi
- push edi
-
- get_GOT ebx ; get GOT address
-
- ; ---- Pass 1: process columns from input.
-
-; mov eax, [original_ebp]
- mov edx, POINTER [dct_table(eax)] ; quantptr
- mov esi, JCOEFPTR [coef_block(eax)] ; inptr
-
-%ifndef NO_ZERO_COLUMN_TEST_ISLOW_SSE2
- mov eax, DWORD [DWBLOCK(1,0,esi,SIZEOF_JCOEF)]
- or eax, DWORD [DWBLOCK(2,0,esi,SIZEOF_JCOEF)]
- jnz near .columnDCT
-
- movdqa xmm0, XMMWORD [XMMBLOCK(1,0,esi,SIZEOF_JCOEF)]
- movdqa xmm1, XMMWORD [XMMBLOCK(2,0,esi,SIZEOF_JCOEF)]
- por xmm0, XMMWORD [XMMBLOCK(3,0,esi,SIZEOF_JCOEF)]
- por xmm1, XMMWORD [XMMBLOCK(4,0,esi,SIZEOF_JCOEF)]
- por xmm0, XMMWORD [XMMBLOCK(5,0,esi,SIZEOF_JCOEF)]
- por xmm1, XMMWORD [XMMBLOCK(6,0,esi,SIZEOF_JCOEF)]
- por xmm0, XMMWORD [XMMBLOCK(7,0,esi,SIZEOF_JCOEF)]
- por xmm1,xmm0
- packsswb xmm1,xmm1
- packsswb xmm1,xmm1
- movd eax,xmm1
- test eax,eax
- jnz short .columnDCT
-
- ; -- AC terms all zero
-
- movdqa xmm5, XMMWORD [XMMBLOCK(0,0,esi,SIZEOF_JCOEF)]
- pmullw xmm5, XMMWORD [XMMBLOCK(0,0,edx,SIZEOF_ISLOW_MULT_TYPE)]
-
- psllw xmm5,PASS1_BITS
-
- movdqa xmm4,xmm5 ; xmm5=in0=(00 01 02 03 04 05 06 07)
- punpcklwd xmm5,xmm5 ; xmm5=(00 00 01 01 02 02 03 03)
- punpckhwd xmm4,xmm4 ; xmm4=(04 04 05 05 06 06 07 07)
-
- pshufd xmm7,xmm5,0x00 ; xmm7=col0=(00 00 00 00 00 00 00 00)
- pshufd xmm6,xmm5,0x55 ; xmm6=col1=(01 01 01 01 01 01 01 01)
- pshufd xmm1,xmm5,0xAA ; xmm1=col2=(02 02 02 02 02 02 02 02)
- pshufd xmm5,xmm5,0xFF ; xmm5=col3=(03 03 03 03 03 03 03 03)
- pshufd xmm0,xmm4,0x00 ; xmm0=col4=(04 04 04 04 04 04 04 04)
- pshufd xmm3,xmm4,0x55 ; xmm3=col5=(05 05 05 05 05 05 05 05)
- pshufd xmm2,xmm4,0xAA ; xmm2=col6=(06 06 06 06 06 06 06 06)
- pshufd xmm4,xmm4,0xFF ; xmm4=col7=(07 07 07 07 07 07 07 07)
-
- movdqa XMMWORD [wk(8)], xmm6 ; wk(8)=col1
- movdqa XMMWORD [wk(9)], xmm5 ; wk(9)=col3
- movdqa XMMWORD [wk(10)], xmm3 ; wk(10)=col5
- movdqa XMMWORD [wk(11)], xmm4 ; wk(11)=col7
- jmp near .column_end
- alignx 16,7
-%endif
-.columnDCT:
-
- ; -- Even part
-
- movdqa xmm0, XMMWORD [XMMBLOCK(0,0,esi,SIZEOF_JCOEF)]
- movdqa xmm1, XMMWORD [XMMBLOCK(2,0,esi,SIZEOF_JCOEF)]
- pmullw xmm0, XMMWORD [XMMBLOCK(0,0,edx,SIZEOF_ISLOW_MULT_TYPE)]
- pmullw xmm1, XMMWORD [XMMBLOCK(2,0,edx,SIZEOF_ISLOW_MULT_TYPE)]
- movdqa xmm2, XMMWORD [XMMBLOCK(4,0,esi,SIZEOF_JCOEF)]
- movdqa xmm3, XMMWORD [XMMBLOCK(6,0,esi,SIZEOF_JCOEF)]
- pmullw xmm2, XMMWORD [XMMBLOCK(4,0,edx,SIZEOF_ISLOW_MULT_TYPE)]
- pmullw xmm3, XMMWORD [XMMBLOCK(6,0,edx,SIZEOF_ISLOW_MULT_TYPE)]
-
- ; (Original)
- ; z1 = (z2 + z3) * 0.541196100;
- ; tmp2 = z1 + z3 * -1.847759065;
- ; tmp3 = z1 + z2 * 0.765366865;
- ;
- ; (This implementation)
- ; tmp2 = z2 * 0.541196100 + z3 * (0.541196100 - 1.847759065);
- ; tmp3 = z2 * (0.541196100 + 0.765366865) + z3 * 0.541196100;
-
- movdqa xmm4,xmm1 ; xmm1=in2=z2
- movdqa xmm5,xmm1
- punpcklwd xmm4,xmm3 ; xmm3=in6=z3
- punpckhwd xmm5,xmm3
- movdqa xmm1,xmm4
- movdqa xmm3,xmm5
- pmaddwd xmm4,[GOTOFF(ebx,PW_F130_F054)] ; xmm4=tmp3L
- pmaddwd xmm5,[GOTOFF(ebx,PW_F130_F054)] ; xmm5=tmp3H
- pmaddwd xmm1,[GOTOFF(ebx,PW_F054_MF130)] ; xmm1=tmp2L
- pmaddwd xmm3,[GOTOFF(ebx,PW_F054_MF130)] ; xmm3=tmp2H
-
- movdqa xmm6,xmm0
- paddw xmm0,xmm2 ; xmm0=in0+in4
- psubw xmm6,xmm2 ; xmm6=in0-in4
-
- pxor xmm7,xmm7
- pxor xmm2,xmm2
- punpcklwd xmm7,xmm0 ; xmm7=tmp0L
- punpckhwd xmm2,xmm0 ; xmm2=tmp0H
- psrad xmm7,(16-CONST_BITS) ; psrad xmm7,16 & pslld xmm7,CONST_BITS
- psrad xmm2,(16-CONST_BITS) ; psrad xmm2,16 & pslld xmm2,CONST_BITS
-
- movdqa xmm0,xmm7
- paddd xmm7,xmm4 ; xmm7=tmp10L
- psubd xmm0,xmm4 ; xmm0=tmp13L
- movdqa xmm4,xmm2
- paddd xmm2,xmm5 ; xmm2=tmp10H
- psubd xmm4,xmm5 ; xmm4=tmp13H
-
- movdqa XMMWORD [wk(0)], xmm7 ; wk(0)=tmp10L
- movdqa XMMWORD [wk(1)], xmm2 ; wk(1)=tmp10H
- movdqa XMMWORD [wk(2)], xmm0 ; wk(2)=tmp13L
- movdqa XMMWORD [wk(3)], xmm4 ; wk(3)=tmp13H
-
- pxor xmm5,xmm5
- pxor xmm7,xmm7
- punpcklwd xmm5,xmm6 ; xmm5=tmp1L
- punpckhwd xmm7,xmm6 ; xmm7=tmp1H
- psrad xmm5,(16-CONST_BITS) ; psrad xmm5,16 & pslld xmm5,CONST_BITS
- psrad xmm7,(16-CONST_BITS) ; psrad xmm7,16 & pslld xmm7,CONST_BITS
-
- movdqa xmm2,xmm5
- paddd xmm5,xmm1 ; xmm5=tmp11L
- psubd xmm2,xmm1 ; xmm2=tmp12L
- movdqa xmm0,xmm7
- paddd xmm7,xmm3 ; xmm7=tmp11H
- psubd xmm0,xmm3 ; xmm0=tmp12H
-
- movdqa XMMWORD [wk(4)], xmm5 ; wk(4)=tmp11L
- movdqa XMMWORD [wk(5)], xmm7 ; wk(5)=tmp11H
- movdqa XMMWORD [wk(6)], xmm2 ; wk(6)=tmp12L
- movdqa XMMWORD [wk(7)], xmm0 ; wk(7)=tmp12H
-
- ; -- Odd part
-
- movdqa xmm4, XMMWORD [XMMBLOCK(1,0,esi,SIZEOF_JCOEF)]
- movdqa xmm6, XMMWORD [XMMBLOCK(3,0,esi,SIZEOF_JCOEF)]
- pmullw xmm4, XMMWORD [XMMBLOCK(1,0,edx,SIZEOF_ISLOW_MULT_TYPE)]
- pmullw xmm6, XMMWORD [XMMBLOCK(3,0,edx,SIZEOF_ISLOW_MULT_TYPE)]
- movdqa xmm1, XMMWORD [XMMBLOCK(5,0,esi,SIZEOF_JCOEF)]
- movdqa xmm3, XMMWORD [XMMBLOCK(7,0,esi,SIZEOF_JCOEF)]
- pmullw xmm1, XMMWORD [XMMBLOCK(5,0,edx,SIZEOF_ISLOW_MULT_TYPE)]
- pmullw xmm3, XMMWORD [XMMBLOCK(7,0,edx,SIZEOF_ISLOW_MULT_TYPE)]
-
- movdqa xmm5,xmm6
- movdqa xmm7,xmm4
- paddw xmm5,xmm3 ; xmm5=z3
- paddw xmm7,xmm1 ; xmm7=z4
-
- ; (Original)
- ; z5 = (z3 + z4) * 1.175875602;
- ; z3 = z3 * -1.961570560; z4 = z4 * -0.390180644;
- ; z3 += z5; z4 += z5;
- ;
- ; (This implementation)
- ; z3 = z3 * (1.175875602 - 1.961570560) + z4 * 1.175875602;
- ; z4 = z3 * 1.175875602 + z4 * (1.175875602 - 0.390180644);
-
- movdqa xmm2,xmm5
- movdqa xmm0,xmm5
- punpcklwd xmm2,xmm7
- punpckhwd xmm0,xmm7
- movdqa xmm5,xmm2
- movdqa xmm7,xmm0
- pmaddwd xmm2,[GOTOFF(ebx,PW_MF078_F117)] ; xmm2=z3L
- pmaddwd xmm0,[GOTOFF(ebx,PW_MF078_F117)] ; xmm0=z3H
- pmaddwd xmm5,[GOTOFF(ebx,PW_F117_F078)] ; xmm5=z4L
- pmaddwd xmm7,[GOTOFF(ebx,PW_F117_F078)] ; xmm7=z4H
-
- movdqa XMMWORD [wk(10)], xmm2 ; wk(10)=z3L
- movdqa XMMWORD [wk(11)], xmm0 ; wk(11)=z3H
-
- ; (Original)
- ; z1 = tmp0 + tmp3; z2 = tmp1 + tmp2;
- ; tmp0 = tmp0 * 0.298631336; tmp1 = tmp1 * 2.053119869;
- ; tmp2 = tmp2 * 3.072711026; tmp3 = tmp3 * 1.501321110;
- ; z1 = z1 * -0.899976223; z2 = z2 * -2.562915447;
- ; tmp0 += z1 + z3; tmp1 += z2 + z4;
- ; tmp2 += z2 + z3; tmp3 += z1 + z4;
- ;
- ; (This implementation)
- ; tmp0 = tmp0 * (0.298631336 - 0.899976223) + tmp3 * -0.899976223;
- ; tmp1 = tmp1 * (2.053119869 - 2.562915447) + tmp2 * -2.562915447;
- ; tmp2 = tmp1 * -2.562915447 + tmp2 * (3.072711026 - 2.562915447);
- ; tmp3 = tmp0 * -0.899976223 + tmp3 * (1.501321110 - 0.899976223);
- ; tmp0 += z3; tmp1 += z4;
- ; tmp2 += z3; tmp3 += z4;
-
- movdqa xmm2,xmm3
- movdqa xmm0,xmm3
- punpcklwd xmm2,xmm4
- punpckhwd xmm0,xmm4
- movdqa xmm3,xmm2
- movdqa xmm4,xmm0
- pmaddwd xmm2,[GOTOFF(ebx,PW_MF060_MF089)] ; xmm2=tmp0L
- pmaddwd xmm0,[GOTOFF(ebx,PW_MF060_MF089)] ; xmm0=tmp0H
- pmaddwd xmm3,[GOTOFF(ebx,PW_MF089_F060)] ; xmm3=tmp3L
- pmaddwd xmm4,[GOTOFF(ebx,PW_MF089_F060)] ; xmm4=tmp3H
-
- paddd xmm2, XMMWORD [wk(10)] ; xmm2=tmp0L
- paddd xmm0, XMMWORD [wk(11)] ; xmm0=tmp0H
- paddd xmm3,xmm5 ; xmm3=tmp3L
- paddd xmm4,xmm7 ; xmm4=tmp3H
-
- movdqa XMMWORD [wk(8)], xmm2 ; wk(8)=tmp0L
- movdqa XMMWORD [wk(9)], xmm0 ; wk(9)=tmp0H
-
- movdqa xmm2,xmm1
- movdqa xmm0,xmm1
- punpcklwd xmm2,xmm6
- punpckhwd xmm0,xmm6
- movdqa xmm1,xmm2
- movdqa xmm6,xmm0
- pmaddwd xmm2,[GOTOFF(ebx,PW_MF050_MF256)] ; xmm2=tmp1L
- pmaddwd xmm0,[GOTOFF(ebx,PW_MF050_MF256)] ; xmm0=tmp1H
- pmaddwd xmm1,[GOTOFF(ebx,PW_MF256_F050)] ; xmm1=tmp2L
- pmaddwd xmm6,[GOTOFF(ebx,PW_MF256_F050)] ; xmm6=tmp2H
-
- paddd xmm2,xmm5 ; xmm2=tmp1L
- paddd xmm0,xmm7 ; xmm0=tmp1H
- paddd xmm1, XMMWORD [wk(10)] ; xmm1=tmp2L
- paddd xmm6, XMMWORD [wk(11)] ; xmm6=tmp2H
-
- movdqa XMMWORD [wk(10)], xmm2 ; wk(10)=tmp1L
- movdqa XMMWORD [wk(11)], xmm0 ; wk(11)=tmp1H
-
- ; -- Final output stage
-
- movdqa xmm5, XMMWORD [wk(0)] ; xmm5=tmp10L
- movdqa xmm7, XMMWORD [wk(1)] ; xmm7=tmp10H
-
- movdqa xmm2,xmm5
- movdqa xmm0,xmm7
- paddd xmm5,xmm3 ; xmm5=data0L
- paddd xmm7,xmm4 ; xmm7=data0H
- psubd xmm2,xmm3 ; xmm2=data7L
- psubd xmm0,xmm4 ; xmm0=data7H
-
- movdqa xmm3,[GOTOFF(ebx,PD_DESCALE_P1)] ; xmm3=[PD_DESCALE_P1]
-
- paddd xmm5,xmm3
- paddd xmm7,xmm3
- psrad xmm5,DESCALE_P1
- psrad xmm7,DESCALE_P1
- paddd xmm2,xmm3
- paddd xmm0,xmm3
- psrad xmm2,DESCALE_P1
- psrad xmm0,DESCALE_P1
-
- packssdw xmm5,xmm7 ; xmm5=data0=(00 01 02 03 04 05 06 07)
- packssdw xmm2,xmm0 ; xmm2=data7=(70 71 72 73 74 75 76 77)
-
- movdqa xmm4, XMMWORD [wk(4)] ; xmm4=tmp11L
- movdqa xmm3, XMMWORD [wk(5)] ; xmm3=tmp11H
-
- movdqa xmm7,xmm4
- movdqa xmm0,xmm3
- paddd xmm4,xmm1 ; xmm4=data1L
- paddd xmm3,xmm6 ; xmm3=data1H
- psubd xmm7,xmm1 ; xmm7=data6L
- psubd xmm0,xmm6 ; xmm0=data6H
-
- movdqa xmm1,[GOTOFF(ebx,PD_DESCALE_P1)] ; xmm1=[PD_DESCALE_P1]
-
- paddd xmm4,xmm1
- paddd xmm3,xmm1
- psrad xmm4,DESCALE_P1
- psrad xmm3,DESCALE_P1
- paddd xmm7,xmm1
- paddd xmm0,xmm1
- psrad xmm7,DESCALE_P1
- psrad xmm0,DESCALE_P1
-
- packssdw xmm4,xmm3 ; xmm4=data1=(10 11 12 13 14 15 16 17)
- packssdw xmm7,xmm0 ; xmm7=data6=(60 61 62 63 64 65 66 67)
-
- movdqa xmm6,xmm5 ; transpose coefficients(phase 1)
- punpcklwd xmm5,xmm4 ; xmm5=(00 10 01 11 02 12 03 13)
- punpckhwd xmm6,xmm4 ; xmm6=(04 14 05 15 06 16 07 17)
- movdqa xmm1,xmm7 ; transpose coefficients(phase 1)
- punpcklwd xmm7,xmm2 ; xmm7=(60 70 61 71 62 72 63 73)
- punpckhwd xmm1,xmm2 ; xmm1=(64 74 65 75 66 76 67 77)
-
- movdqa xmm3, XMMWORD [wk(6)] ; xmm3=tmp12L
- movdqa xmm0, XMMWORD [wk(7)] ; xmm0=tmp12H
- movdqa xmm4, XMMWORD [wk(10)] ; xmm4=tmp1L
- movdqa xmm2, XMMWORD [wk(11)] ; xmm2=tmp1H
-
- movdqa XMMWORD [wk(0)], xmm5 ; wk(0)=(00 10 01 11 02 12 03 13)
- movdqa XMMWORD [wk(1)], xmm6 ; wk(1)=(04 14 05 15 06 16 07 17)
- movdqa XMMWORD [wk(4)], xmm7 ; wk(4)=(60 70 61 71 62 72 63 73)
- movdqa XMMWORD [wk(5)], xmm1 ; wk(5)=(64 74 65 75 66 76 67 77)
-
- movdqa xmm5,xmm3
- movdqa xmm6,xmm0
- paddd xmm3,xmm4 ; xmm3=data2L
- paddd xmm0,xmm2 ; xmm0=data2H
- psubd xmm5,xmm4 ; xmm5=data5L
- psubd xmm6,xmm2 ; xmm6=data5H
-
- movdqa xmm7,[GOTOFF(ebx,PD_DESCALE_P1)] ; xmm7=[PD_DESCALE_P1]
-
- paddd xmm3,xmm7
- paddd xmm0,xmm7
- psrad xmm3,DESCALE_P1
- psrad xmm0,DESCALE_P1
- paddd xmm5,xmm7
- paddd xmm6,xmm7
- psrad xmm5,DESCALE_P1
- psrad xmm6,DESCALE_P1
-
- packssdw xmm3,xmm0 ; xmm3=data2=(20 21 22 23 24 25 26 27)
- packssdw xmm5,xmm6 ; xmm5=data5=(50 51 52 53 54 55 56 57)
-
- movdqa xmm1, XMMWORD [wk(2)] ; xmm1=tmp13L
- movdqa xmm4, XMMWORD [wk(3)] ; xmm4=tmp13H
- movdqa xmm2, XMMWORD [wk(8)] ; xmm2=tmp0L
- movdqa xmm7, XMMWORD [wk(9)] ; xmm7=tmp0H
-
- movdqa xmm0,xmm1
- movdqa xmm6,xmm4
- paddd xmm1,xmm2 ; xmm1=data3L
- paddd xmm4,xmm7 ; xmm4=data3H
- psubd xmm0,xmm2 ; xmm0=data4L
- psubd xmm6,xmm7 ; xmm6=data4H
-
- movdqa xmm2,[GOTOFF(ebx,PD_DESCALE_P1)] ; xmm2=[PD_DESCALE_P1]
-
- paddd xmm1,xmm2
- paddd xmm4,xmm2
- psrad xmm1,DESCALE_P1
- psrad xmm4,DESCALE_P1
- paddd xmm0,xmm2
- paddd xmm6,xmm2
- psrad xmm0,DESCALE_P1
- psrad xmm6,DESCALE_P1
-
- packssdw xmm1,xmm4 ; xmm1=data3=(30 31 32 33 34 35 36 37)
- packssdw xmm0,xmm6 ; xmm0=data4=(40 41 42 43 44 45 46 47)
-
- movdqa xmm7, XMMWORD [wk(0)] ; xmm7=(00 10 01 11 02 12 03 13)
- movdqa xmm2, XMMWORD [wk(1)] ; xmm2=(04 14 05 15 06 16 07 17)
-
- movdqa xmm4,xmm3 ; transpose coefficients(phase 1)
- punpcklwd xmm3,xmm1 ; xmm3=(20 30 21 31 22 32 23 33)
- punpckhwd xmm4,xmm1 ; xmm4=(24 34 25 35 26 36 27 37)
- movdqa xmm6,xmm0 ; transpose coefficients(phase 1)
- punpcklwd xmm0,xmm5 ; xmm0=(40 50 41 51 42 52 43 53)
- punpckhwd xmm6,xmm5 ; xmm6=(44 54 45 55 46 56 47 57)
-
- movdqa xmm1,xmm7 ; transpose coefficients(phase 2)
- punpckldq xmm7,xmm3 ; xmm7=(00 10 20 30 01 11 21 31)
- punpckhdq xmm1,xmm3 ; xmm1=(02 12 22 32 03 13 23 33)
- movdqa xmm5,xmm2 ; transpose coefficients(phase 2)
- punpckldq xmm2,xmm4 ; xmm2=(04 14 24 34 05 15 25 35)
- punpckhdq xmm5,xmm4 ; xmm5=(06 16 26 36 07 17 27 37)
-
- movdqa xmm3, XMMWORD [wk(4)] ; xmm3=(60 70 61 71 62 72 63 73)
- movdqa xmm4, XMMWORD [wk(5)] ; xmm4=(64 74 65 75 66 76 67 77)
-
- movdqa XMMWORD [wk(6)], xmm2 ; wk(6)=(04 14 24 34 05 15 25 35)
- movdqa XMMWORD [wk(7)], xmm5 ; wk(7)=(06 16 26 36 07 17 27 37)
-
- movdqa xmm2,xmm0 ; transpose coefficients(phase 2)
- punpckldq xmm0,xmm3 ; xmm0=(40 50 60 70 41 51 61 71)
- punpckhdq xmm2,xmm3 ; xmm2=(42 52 62 72 43 53 63 73)
- movdqa xmm5,xmm6 ; transpose coefficients(phase 2)
- punpckldq xmm6,xmm4 ; xmm6=(44 54 64 74 45 55 65 75)
- punpckhdq xmm5,xmm4 ; xmm5=(46 56 66 76 47 57 67 77)
-
- movdqa xmm3,xmm7 ; transpose coefficients(phase 3)
- punpcklqdq xmm7,xmm0 ; xmm7=col0=(00 10 20 30 40 50 60 70)
- punpckhqdq xmm3,xmm0 ; xmm3=col1=(01 11 21 31 41 51 61 71)
- movdqa xmm4,xmm1 ; transpose coefficients(phase 3)
- punpcklqdq xmm1,xmm2 ; xmm1=col2=(02 12 22 32 42 52 62 72)
- punpckhqdq xmm4,xmm2 ; xmm4=col3=(03 13 23 33 43 53 63 73)
-
- movdqa xmm0, XMMWORD [wk(6)] ; xmm0=(04 14 24 34 05 15 25 35)
- movdqa xmm2, XMMWORD [wk(7)] ; xmm2=(06 16 26 36 07 17 27 37)
-
- movdqa XMMWORD [wk(8)], xmm3 ; wk(8)=col1
- movdqa XMMWORD [wk(9)], xmm4 ; wk(9)=col3
-
- movdqa xmm3,xmm0 ; transpose coefficients(phase 3)
- punpcklqdq xmm0,xmm6 ; xmm0=col4=(04 14 24 34 44 54 64 74)
- punpckhqdq xmm3,xmm6 ; xmm3=col5=(05 15 25 35 45 55 65 75)
- movdqa xmm4,xmm2 ; transpose coefficients(phase 3)
- punpcklqdq xmm2,xmm5 ; xmm2=col6=(06 16 26 36 46 56 66 76)
- punpckhqdq xmm4,xmm5 ; xmm4=col7=(07 17 27 37 47 57 67 77)
-
- movdqa XMMWORD [wk(10)], xmm3 ; wk(10)=col5
- movdqa XMMWORD [wk(11)], xmm4 ; wk(11)=col7
-.column_end:
-
- ; -- Prefetch the next coefficient block
-
- prefetchnta [esi + DCTSIZE2*SIZEOF_JCOEF + 0*32]
- prefetchnta [esi + DCTSIZE2*SIZEOF_JCOEF + 1*32]
- prefetchnta [esi + DCTSIZE2*SIZEOF_JCOEF + 2*32]
- prefetchnta [esi + DCTSIZE2*SIZEOF_JCOEF + 3*32]
-
- ; ---- Pass 2: process rows from work array, store into output array.
-
- mov eax, [original_ebp]
- mov edi, JSAMPARRAY [output_buf(eax)] ; (JSAMPROW *)
- mov eax, JDIMENSION [output_col(eax)]
-
- ; -- Even part
-
- ; xmm7=col0, xmm1=col2, xmm0=col4, xmm2=col6
-
- ; (Original)
- ; z1 = (z2 + z3) * 0.541196100;
- ; tmp2 = z1 + z3 * -1.847759065;
- ; tmp3 = z1 + z2 * 0.765366865;
- ;
- ; (This implementation)
- ; tmp2 = z2 * 0.541196100 + z3 * (0.541196100 - 1.847759065);
- ; tmp3 = z2 * (0.541196100 + 0.765366865) + z3 * 0.541196100;
-
- movdqa xmm6,xmm1 ; xmm1=in2=z2
- movdqa xmm5,xmm1
- punpcklwd xmm6,xmm2 ; xmm2=in6=z3
- punpckhwd xmm5,xmm2
- movdqa xmm1,xmm6
- movdqa xmm2,xmm5
- pmaddwd xmm6,[GOTOFF(ebx,PW_F130_F054)] ; xmm6=tmp3L
- pmaddwd xmm5,[GOTOFF(ebx,PW_F130_F054)] ; xmm5=tmp3H
- pmaddwd xmm1,[GOTOFF(ebx,PW_F054_MF130)] ; xmm1=tmp2L
- pmaddwd xmm2,[GOTOFF(ebx,PW_F054_MF130)] ; xmm2=tmp2H
-
- movdqa xmm3,xmm7
- paddw xmm7,xmm0 ; xmm7=in0+in4
- psubw xmm3,xmm0 ; xmm3=in0-in4
-
- pxor xmm4,xmm4
- pxor xmm0,xmm0
- punpcklwd xmm4,xmm7 ; xmm4=tmp0L
- punpckhwd xmm0,xmm7 ; xmm0=tmp0H
- psrad xmm4,(16-CONST_BITS) ; psrad xmm4,16 & pslld xmm4,CONST_BITS
- psrad xmm0,(16-CONST_BITS) ; psrad xmm0,16 & pslld xmm0,CONST_BITS
-
- movdqa xmm7,xmm4
- paddd xmm4,xmm6 ; xmm4=tmp10L
- psubd xmm7,xmm6 ; xmm7=tmp13L
- movdqa xmm6,xmm0
- paddd xmm0,xmm5 ; xmm0=tmp10H
- psubd xmm6,xmm5 ; xmm6=tmp13H
-
- movdqa XMMWORD [wk(0)], xmm4 ; wk(0)=tmp10L
- movdqa XMMWORD [wk(1)], xmm0 ; wk(1)=tmp10H
- movdqa XMMWORD [wk(2)], xmm7 ; wk(2)=tmp13L
- movdqa XMMWORD [wk(3)], xmm6 ; wk(3)=tmp13H
-
- pxor xmm5,xmm5
- pxor xmm4,xmm4
- punpcklwd xmm5,xmm3 ; xmm5=tmp1L
- punpckhwd xmm4,xmm3 ; xmm4=tmp1H
- psrad xmm5,(16-CONST_BITS) ; psrad xmm5,16 & pslld xmm5,CONST_BITS
- psrad xmm4,(16-CONST_BITS) ; psrad xmm4,16 & pslld xmm4,CONST_BITS
-
- movdqa xmm0,xmm5
- paddd xmm5,xmm1 ; xmm5=tmp11L
- psubd xmm0,xmm1 ; xmm0=tmp12L
- movdqa xmm7,xmm4
- paddd xmm4,xmm2 ; xmm4=tmp11H
- psubd xmm7,xmm2 ; xmm7=tmp12H
-
- movdqa XMMWORD [wk(4)], xmm5 ; wk(4)=tmp11L
- movdqa XMMWORD [wk(5)], xmm4 ; wk(5)=tmp11H
- movdqa XMMWORD [wk(6)], xmm0 ; wk(6)=tmp12L
- movdqa XMMWORD [wk(7)], xmm7 ; wk(7)=tmp12H
-
- ; -- Odd part
-
- movdqa xmm6, XMMWORD [wk(9)] ; xmm6=col3
- movdqa xmm3, XMMWORD [wk(8)] ; xmm3=col1
- movdqa xmm1, XMMWORD [wk(11)] ; xmm1=col7
- movdqa xmm2, XMMWORD [wk(10)] ; xmm2=col5
-
- movdqa xmm5,xmm6
- movdqa xmm4,xmm3
- paddw xmm5,xmm1 ; xmm5=z3
- paddw xmm4,xmm2 ; xmm4=z4
-
- ; (Original)
- ; z5 = (z3 + z4) * 1.175875602;
- ; z3 = z3 * -1.961570560; z4 = z4 * -0.390180644;
- ; z3 += z5; z4 += z5;
- ;
- ; (This implementation)
- ; z3 = z3 * (1.175875602 - 1.961570560) + z4 * 1.175875602;
- ; z4 = z3 * 1.175875602 + z4 * (1.175875602 - 0.390180644);
-
- movdqa xmm0,xmm5
- movdqa xmm7,xmm5
- punpcklwd xmm0,xmm4
- punpckhwd xmm7,xmm4
- movdqa xmm5,xmm0
- movdqa xmm4,xmm7
- pmaddwd xmm0,[GOTOFF(ebx,PW_MF078_F117)] ; xmm0=z3L
- pmaddwd xmm7,[GOTOFF(ebx,PW_MF078_F117)] ; xmm7=z3H
- pmaddwd xmm5,[GOTOFF(ebx,PW_F117_F078)] ; xmm5=z4L
- pmaddwd xmm4,[GOTOFF(ebx,PW_F117_F078)] ; xmm4=z4H
-
- movdqa XMMWORD [wk(10)], xmm0 ; wk(10)=z3L
- movdqa XMMWORD [wk(11)], xmm7 ; wk(11)=z3H
-
- ; (Original)
- ; z1 = tmp0 + tmp3; z2 = tmp1 + tmp2;
- ; tmp0 = tmp0 * 0.298631336; tmp1 = tmp1 * 2.053119869;
- ; tmp2 = tmp2 * 3.072711026; tmp3 = tmp3 * 1.501321110;
- ; z1 = z1 * -0.899976223; z2 = z2 * -2.562915447;
- ; tmp0 += z1 + z3; tmp1 += z2 + z4;
- ; tmp2 += z2 + z3; tmp3 += z1 + z4;
- ;
- ; (This implementation)
- ; tmp0 = tmp0 * (0.298631336 - 0.899976223) + tmp3 * -0.899976223;
- ; tmp1 = tmp1 * (2.053119869 - 2.562915447) + tmp2 * -2.562915447;
- ; tmp2 = tmp1 * -2.562915447 + tmp2 * (3.072711026 - 2.562915447);
- ; tmp3 = tmp0 * -0.899976223 + tmp3 * (1.501321110 - 0.899976223);
- ; tmp0 += z3; tmp1 += z4;
- ; tmp2 += z3; tmp3 += z4;
-
- movdqa xmm0,xmm1
- movdqa xmm7,xmm1
- punpcklwd xmm0,xmm3
- punpckhwd xmm7,xmm3
- movdqa xmm1,xmm0
- movdqa xmm3,xmm7
- pmaddwd xmm0,[GOTOFF(ebx,PW_MF060_MF089)] ; xmm0=tmp0L
- pmaddwd xmm7,[GOTOFF(ebx,PW_MF060_MF089)] ; xmm7=tmp0H
- pmaddwd xmm1,[GOTOFF(ebx,PW_MF089_F060)] ; xmm1=tmp3L
- pmaddwd xmm3,[GOTOFF(ebx,PW_MF089_F060)] ; xmm3=tmp3H
-
- paddd xmm0, XMMWORD [wk(10)] ; xmm0=tmp0L
- paddd xmm7, XMMWORD [wk(11)] ; xmm7=tmp0H
- paddd xmm1,xmm5 ; xmm1=tmp3L
- paddd xmm3,xmm4 ; xmm3=tmp3H
-
- movdqa XMMWORD [wk(8)], xmm0 ; wk(8)=tmp0L
- movdqa XMMWORD [wk(9)], xmm7 ; wk(9)=tmp0H
-
- movdqa xmm0,xmm2
- movdqa xmm7,xmm2
- punpcklwd xmm0,xmm6
- punpckhwd xmm7,xmm6
- movdqa xmm2,xmm0
- movdqa xmm6,xmm7
- pmaddwd xmm0,[GOTOFF(ebx,PW_MF050_MF256)] ; xmm0=tmp1L
- pmaddwd xmm7,[GOTOFF(ebx,PW_MF050_MF256)] ; xmm7=tmp1H
- pmaddwd xmm2,[GOTOFF(ebx,PW_MF256_F050)] ; xmm2=tmp2L
- pmaddwd xmm6,[GOTOFF(ebx,PW_MF256_F050)] ; xmm6=tmp2H
-
- paddd xmm0,xmm5 ; xmm0=tmp1L
- paddd xmm7,xmm4 ; xmm7=tmp1H
- paddd xmm2, XMMWORD [wk(10)] ; xmm2=tmp2L
- paddd xmm6, XMMWORD [wk(11)] ; xmm6=tmp2H
-
- movdqa XMMWORD [wk(10)], xmm0 ; wk(10)=tmp1L
- movdqa XMMWORD [wk(11)], xmm7 ; wk(11)=tmp1H
-
- ; -- Final output stage
-
- movdqa xmm5, XMMWORD [wk(0)] ; xmm5=tmp10L
- movdqa xmm4, XMMWORD [wk(1)] ; xmm4=tmp10H
-
- movdqa xmm0,xmm5
- movdqa xmm7,xmm4
- paddd xmm5,xmm1 ; xmm5=data0L
- paddd xmm4,xmm3 ; xmm4=data0H
- psubd xmm0,xmm1 ; xmm0=data7L
- psubd xmm7,xmm3 ; xmm7=data7H
-
- movdqa xmm1,[GOTOFF(ebx,PD_DESCALE_P2)] ; xmm1=[PD_DESCALE_P2]
-
- paddd xmm5,xmm1
- paddd xmm4,xmm1
- psrad xmm5,DESCALE_P2
- psrad xmm4,DESCALE_P2
- paddd xmm0,xmm1
- paddd xmm7,xmm1
- psrad xmm0,DESCALE_P2
- psrad xmm7,DESCALE_P2
-
- packssdw xmm5,xmm4 ; xmm5=data0=(00 10 20 30 40 50 60 70)
- packssdw xmm0,xmm7 ; xmm0=data7=(07 17 27 37 47 57 67 77)
-
- movdqa xmm3, XMMWORD [wk(4)] ; xmm3=tmp11L
- movdqa xmm1, XMMWORD [wk(5)] ; xmm1=tmp11H
-
- movdqa xmm4,xmm3
- movdqa xmm7,xmm1
- paddd xmm3,xmm2 ; xmm3=data1L
- paddd xmm1,xmm6 ; xmm1=data1H
- psubd xmm4,xmm2 ; xmm4=data6L
- psubd xmm7,xmm6 ; xmm7=data6H
-
- movdqa xmm2,[GOTOFF(ebx,PD_DESCALE_P2)] ; xmm2=[PD_DESCALE_P2]
-
- paddd xmm3,xmm2
- paddd xmm1,xmm2
- psrad xmm3,DESCALE_P2
- psrad xmm1,DESCALE_P2
- paddd xmm4,xmm2
- paddd xmm7,xmm2
- psrad xmm4,DESCALE_P2
- psrad xmm7,DESCALE_P2
-
- packssdw xmm3,xmm1 ; xmm3=data1=(01 11 21 31 41 51 61 71)
- packssdw xmm4,xmm7 ; xmm4=data6=(06 16 26 36 46 56 66 76)
-
- packsswb xmm5,xmm4 ; xmm5=(00 10 20 30 40 50 60 70 06 16 26 36 46 56 66 76)
- packsswb xmm3,xmm0 ; xmm3=(01 11 21 31 41 51 61 71 07 17 27 37 47 57 67 77)
-
- movdqa xmm6, XMMWORD [wk(6)] ; xmm6=tmp12L
- movdqa xmm2, XMMWORD [wk(7)] ; xmm2=tmp12H
- movdqa xmm1, XMMWORD [wk(10)] ; xmm1=tmp1L
- movdqa xmm7, XMMWORD [wk(11)] ; xmm7=tmp1H
-
- movdqa XMMWORD [wk(0)], xmm5 ; wk(0)=(00 10 20 30 40 50 60 70 06 16 26 36 46 56 66 76)
- movdqa XMMWORD [wk(1)], xmm3 ; wk(1)=(01 11 21 31 41 51 61 71 07 17 27 37 47 57 67 77)
-
- movdqa xmm4,xmm6
- movdqa xmm0,xmm2
- paddd xmm6,xmm1 ; xmm6=data2L
- paddd xmm2,xmm7 ; xmm2=data2H
- psubd xmm4,xmm1 ; xmm4=data5L
- psubd xmm0,xmm7 ; xmm0=data5H
-
- movdqa xmm5,[GOTOFF(ebx,PD_DESCALE_P2)] ; xmm5=[PD_DESCALE_P2]
-
- paddd xmm6,xmm5
- paddd xmm2,xmm5
- psrad xmm6,DESCALE_P2
- psrad xmm2,DESCALE_P2
- paddd xmm4,xmm5
- paddd xmm0,xmm5
- psrad xmm4,DESCALE_P2
- psrad xmm0,DESCALE_P2
-
- packssdw xmm6,xmm2 ; xmm6=data2=(02 12 22 32 42 52 62 72)
- packssdw xmm4,xmm0 ; xmm4=data5=(05 15 25 35 45 55 65 75)
-
- movdqa xmm3, XMMWORD [wk(2)] ; xmm3=tmp13L
- movdqa xmm1, XMMWORD [wk(3)] ; xmm1=tmp13H
- movdqa xmm7, XMMWORD [wk(8)] ; xmm7=tmp0L
- movdqa xmm5, XMMWORD [wk(9)] ; xmm5=tmp0H
-
- movdqa xmm2,xmm3
- movdqa xmm0,xmm1
- paddd xmm3,xmm7 ; xmm3=data3L
- paddd xmm1,xmm5 ; xmm1=data3H
- psubd xmm2,xmm7 ; xmm2=data4L
- psubd xmm0,xmm5 ; xmm0=data4H
-
- movdqa xmm7,[GOTOFF(ebx,PD_DESCALE_P2)] ; xmm7=[PD_DESCALE_P2]
-
- paddd xmm3,xmm7
- paddd xmm1,xmm7
- psrad xmm3,DESCALE_P2
- psrad xmm1,DESCALE_P2
- paddd xmm2,xmm7
- paddd xmm0,xmm7
- psrad xmm2,DESCALE_P2
- psrad xmm0,DESCALE_P2
-
- movdqa xmm5,[GOTOFF(ebx,PB_CENTERJSAMP)] ; xmm5=[PB_CENTERJSAMP]
-
- packssdw xmm3,xmm1 ; xmm3=data3=(03 13 23 33 43 53 63 73)
- packssdw xmm2,xmm0 ; xmm2=data4=(04 14 24 34 44 54 64 74)
-
- movdqa xmm7, XMMWORD [wk(0)] ; xmm7=(00 10 20 30 40 50 60 70 06 16 26 36 46 56 66 76)
- movdqa xmm1, XMMWORD [wk(1)] ; xmm1=(01 11 21 31 41 51 61 71 07 17 27 37 47 57 67 77)
-
- packsswb xmm6,xmm2 ; xmm6=(02 12 22 32 42 52 62 72 04 14 24 34 44 54 64 74)
- packsswb xmm3,xmm4 ; xmm3=(03 13 23 33 43 53 63 73 05 15 25 35 45 55 65 75)
-
- paddb xmm7,xmm5
- paddb xmm1,xmm5
- paddb xmm6,xmm5
- paddb xmm3,xmm5
-
- movdqa xmm0,xmm7 ; transpose coefficients(phase 1)
- punpcklbw xmm7,xmm1 ; xmm7=(00 01 10 11 20 21 30 31 40 41 50 51 60 61 70 71)
- punpckhbw xmm0,xmm1 ; xmm0=(06 07 16 17 26 27 36 37 46 47 56 57 66 67 76 77)
- movdqa xmm2,xmm6 ; transpose coefficients(phase 1)
- punpcklbw xmm6,xmm3 ; xmm6=(02 03 12 13 22 23 32 33 42 43 52 53 62 63 72 73)
- punpckhbw xmm2,xmm3 ; xmm2=(04 05 14 15 24 25 34 35 44 45 54 55 64 65 74 75)
-
- movdqa xmm4,xmm7 ; transpose coefficients(phase 2)
- punpcklwd xmm7,xmm6 ; xmm7=(00 01 02 03 10 11 12 13 20 21 22 23 30 31 32 33)
- punpckhwd xmm4,xmm6 ; xmm4=(40 41 42 43 50 51 52 53 60 61 62 63 70 71 72 73)
- movdqa xmm5,xmm2 ; transpose coefficients(phase 2)
- punpcklwd xmm2,xmm0 ; xmm2=(04 05 06 07 14 15 16 17 24 25 26 27 34 35 36 37)
- punpckhwd xmm5,xmm0 ; xmm5=(44 45 46 47 54 55 56 57 64 65 66 67 74 75 76 77)
-
- movdqa xmm1,xmm7 ; transpose coefficients(phase 3)
- punpckldq xmm7,xmm2 ; xmm7=(00 01 02 03 04 05 06 07 10 11 12 13 14 15 16 17)
- punpckhdq xmm1,xmm2 ; xmm1=(20 21 22 23 24 25 26 27 30 31 32 33 34 35 36 37)
- movdqa xmm3,xmm4 ; transpose coefficients(phase 3)
- punpckldq xmm4,xmm5 ; xmm4=(40 41 42 43 44 45 46 47 50 51 52 53 54 55 56 57)
- punpckhdq xmm3,xmm5 ; xmm3=(60 61 62 63 64 65 66 67 70 71 72 73 74 75 76 77)
-
- pshufd xmm6,xmm7,0x4E ; xmm6=(10 11 12 13 14 15 16 17 00 01 02 03 04 05 06 07)
- pshufd xmm0,xmm1,0x4E ; xmm0=(30 31 32 33 34 35 36 37 20 21 22 23 24 25 26 27)
- pshufd xmm2,xmm4,0x4E ; xmm2=(50 51 52 53 54 55 56 57 40 41 42 43 44 45 46 47)
- pshufd xmm5,xmm3,0x4E ; xmm5=(70 71 72 73 74 75 76 77 60 61 62 63 64 65 66 67)
-
- mov edx, JSAMPROW [edi+0*SIZEOF_JSAMPROW]
- mov esi, JSAMPROW [edi+2*SIZEOF_JSAMPROW]
- movq XMM_MMWORD [edx+eax*SIZEOF_JSAMPLE], xmm7
- movq XMM_MMWORD [esi+eax*SIZEOF_JSAMPLE], xmm1
- mov edx, JSAMPROW [edi+4*SIZEOF_JSAMPROW]
- mov esi, JSAMPROW [edi+6*SIZEOF_JSAMPROW]
- movq XMM_MMWORD [edx+eax*SIZEOF_JSAMPLE], xmm4
- movq XMM_MMWORD [esi+eax*SIZEOF_JSAMPLE], xmm3
-
- mov edx, JSAMPROW [edi+1*SIZEOF_JSAMPROW]
- mov esi, JSAMPROW [edi+3*SIZEOF_JSAMPROW]
- movq XMM_MMWORD [edx+eax*SIZEOF_JSAMPLE], xmm6
- movq XMM_MMWORD [esi+eax*SIZEOF_JSAMPLE], xmm0
- mov edx, JSAMPROW [edi+5*SIZEOF_JSAMPROW]
- mov esi, JSAMPROW [edi+7*SIZEOF_JSAMPROW]
- movq XMM_MMWORD [edx+eax*SIZEOF_JSAMPLE], xmm2
- movq XMM_MMWORD [esi+eax*SIZEOF_JSAMPLE], xmm5
-
- pop edi
- pop esi
-; pop edx ; need not be preserved
-; pop ecx ; unused
- poppic ebx
- mov esp,ebp ; esp <- aligned ebp
- pop esp ; esp <- original ebp
- pop ebp
- ret
-
-; For some reason, the OS X linker does not honor the request to align the
-; segment unless we do this.
- align 16
diff --git a/simd/jidctred-sse2-64.asm b/simd/jidctred-sse2-64.asm
deleted file mode 100644
index a54bbe2..0000000
--- a/simd/jidctred-sse2-64.asm
+++ /dev/null
@@ -1,575 +0,0 @@
-;
-; jidctred.asm - reduced-size IDCT (64-bit SSE2)
-;
-; Copyright 2009 Pierre Ossman <ossman@cendio.se> for Cendio AB
-; Copyright (C) 2009, D. R. Commander.
-;
-; Based on the x86 SIMD extension for IJG JPEG library
-; Copyright (C) 1999-2006, MIYASAKA Masaru.
-; For conditions of distribution and use, see copyright notice in jsimdext.inc
-;
-; This file should be assembled with NASM (Netwide Assembler),
-; can *not* be assembled with Microsoft's MASM or any compatible
-; assembler (including Borland's Turbo Assembler).
-; NASM is available from http://nasm.sourceforge.net/ or
-; http://sourceforge.net/project/showfiles.php?group_id=6208
-;
-; This file contains inverse-DCT routines that produce reduced-size
-; output: either 4x4 or 2x2 pixels from an 8x8 DCT block.
-; The following code is based directly on the IJG's original jidctred.c;
-; see the jidctred.c for more details.
-;
-; [TAB8]
-
-%include "jsimdext.inc"
-%include "jdct.inc"
-
-; --------------------------------------------------------------------------
-
-%define CONST_BITS 13
-%define PASS1_BITS 2
-
-%define DESCALE_P1_4 (CONST_BITS-PASS1_BITS+1)
-%define DESCALE_P2_4 (CONST_BITS+PASS1_BITS+3+1)
-%define DESCALE_P1_2 (CONST_BITS-PASS1_BITS+2)
-%define DESCALE_P2_2 (CONST_BITS+PASS1_BITS+3+2)
-
-%if CONST_BITS == 13
-F_0_211 equ 1730 ; FIX(0.211164243)
-F_0_509 equ 4176 ; FIX(0.509795579)
-F_0_601 equ 4926 ; FIX(0.601344887)
-F_0_720 equ 5906 ; FIX(0.720959822)
-F_0_765 equ 6270 ; FIX(0.765366865)
-F_0_850 equ 6967 ; FIX(0.850430095)
-F_0_899 equ 7373 ; FIX(0.899976223)
-F_1_061 equ 8697 ; FIX(1.061594337)
-F_1_272 equ 10426 ; FIX(1.272758580)
-F_1_451 equ 11893 ; FIX(1.451774981)
-F_1_847 equ 15137 ; FIX(1.847759065)
-F_2_172 equ 17799 ; FIX(2.172734803)
-F_2_562 equ 20995 ; FIX(2.562915447)
-F_3_624 equ 29692 ; FIX(3.624509785)
-%else
-; NASM cannot do compile-time arithmetic on floating-point constants.
-%define DESCALE(x,n) (((x)+(1<<((n)-1)))>>(n))
-F_0_211 equ DESCALE( 226735879,30-CONST_BITS) ; FIX(0.211164243)
-F_0_509 equ DESCALE( 547388834,30-CONST_BITS) ; FIX(0.509795579)
-F_0_601 equ DESCALE( 645689155,30-CONST_BITS) ; FIX(0.601344887)
-F_0_720 equ DESCALE( 774124714,30-CONST_BITS) ; FIX(0.720959822)
-F_0_765 equ DESCALE( 821806413,30-CONST_BITS) ; FIX(0.765366865)
-F_0_850 equ DESCALE( 913142361,30-CONST_BITS) ; FIX(0.850430095)
-F_0_899 equ DESCALE( 966342111,30-CONST_BITS) ; FIX(0.899976223)
-F_1_061 equ DESCALE(1139878239,30-CONST_BITS) ; FIX(1.061594337)
-F_1_272 equ DESCALE(1366614119,30-CONST_BITS) ; FIX(1.272758580)
-F_1_451 equ DESCALE(1558831516,30-CONST_BITS) ; FIX(1.451774981)
-F_1_847 equ DESCALE(1984016188,30-CONST_BITS) ; FIX(1.847759065)
-F_2_172 equ DESCALE(2332956230,30-CONST_BITS) ; FIX(2.172734803)
-F_2_562 equ DESCALE(2751909506,30-CONST_BITS) ; FIX(2.562915447)
-F_3_624 equ DESCALE(3891787747,30-CONST_BITS) ; FIX(3.624509785)
-%endif
-
-; --------------------------------------------------------------------------
- SECTION SEG_CONST
-
- alignz 16
- global EXTN(jconst_idct_red_sse2)
-
-EXTN(jconst_idct_red_sse2):
-
-PW_F184_MF076 times 4 dw F_1_847,-F_0_765
-PW_F256_F089 times 4 dw F_2_562, F_0_899
-PW_F106_MF217 times 4 dw F_1_061,-F_2_172
-PW_MF060_MF050 times 4 dw -F_0_601,-F_0_509
-PW_F145_MF021 times 4 dw F_1_451,-F_0_211
-PW_F362_MF127 times 4 dw F_3_624,-F_1_272
-PW_F085_MF072 times 4 dw F_0_850,-F_0_720
-PD_DESCALE_P1_4 times 4 dd 1 << (DESCALE_P1_4-1)
-PD_DESCALE_P2_4 times 4 dd 1 << (DESCALE_P2_4-1)
-PD_DESCALE_P1_2 times 4 dd 1 << (DESCALE_P1_2-1)
-PD_DESCALE_P2_2 times 4 dd 1 << (DESCALE_P2_2-1)
-PB_CENTERJSAMP times 16 db CENTERJSAMPLE
-
- alignz 16
-
-; --------------------------------------------------------------------------
- SECTION SEG_TEXT
- BITS 64
-;
-; Perform dequantization and inverse DCT on one block of coefficients,
-; producing a reduced-size 4x4 output block.
-;
-; GLOBAL(void)
-; jsimd_idct_4x4_sse2 (void *dct_table, JCOEFPTR coef_block,
-; JSAMPARRAY output_buf, JDIMENSION output_col)
-;
-
-; r10 = void *dct_table
-; r11 = JCOEFPTR coef_block
-; r12 = JSAMPARRAY output_buf
-; r13 = JDIMENSION output_col
-
-%define original_rbp rbp+0
-%define wk(i) rbp-(WK_NUM-(i))*SIZEOF_XMMWORD ; xmmword wk[WK_NUM]
-%define WK_NUM 2
-
- align 16
- global EXTN(jsimd_idct_4x4_sse2)
-
-EXTN(jsimd_idct_4x4_sse2):
- push rbp
- mov rax,rsp ; rax = original rbp
- sub rsp, byte 4
- and rsp, byte (-SIZEOF_XMMWORD) ; align to 128 bits
- mov [rsp],rax
- mov rbp,rsp ; rbp = aligned rbp
- lea rsp, [wk(0)]
- collect_args
-
- ; ---- Pass 1: process columns from input.
-
- mov rdx, r10 ; quantptr
- mov rsi, r11 ; inptr
-
-%ifndef NO_ZERO_COLUMN_TEST_4X4_SSE2
- mov eax, DWORD [DWBLOCK(1,0,rsi,SIZEOF_JCOEF)]
- or eax, DWORD [DWBLOCK(2,0,rsi,SIZEOF_JCOEF)]
- jnz short .columnDCT
-
- movdqa xmm0, XMMWORD [XMMBLOCK(1,0,rsi,SIZEOF_JCOEF)]
- movdqa xmm1, XMMWORD [XMMBLOCK(2,0,rsi,SIZEOF_JCOEF)]
- por xmm0, XMMWORD [XMMBLOCK(3,0,rsi,SIZEOF_JCOEF)]
- por xmm1, XMMWORD [XMMBLOCK(5,0,rsi,SIZEOF_JCOEF)]
- por xmm0, XMMWORD [XMMBLOCK(6,0,rsi,SIZEOF_JCOEF)]
- por xmm1, XMMWORD [XMMBLOCK(7,0,rsi,SIZEOF_JCOEF)]
- por xmm0,xmm1
- packsswb xmm0,xmm0
- packsswb xmm0,xmm0
- movd eax,xmm0
- test rax,rax
- jnz short .columnDCT
-
- ; -- AC terms all zero
-
- movdqa xmm0, XMMWORD [XMMBLOCK(0,0,rsi,SIZEOF_JCOEF)]
- pmullw xmm0, XMMWORD [XMMBLOCK(0,0,rdx,SIZEOF_ISLOW_MULT_TYPE)]
-
- psllw xmm0,PASS1_BITS
-
- movdqa xmm3,xmm0 ; xmm0=in0=(00 01 02 03 04 05 06 07)
- punpcklwd xmm0,xmm0 ; xmm0=(00 00 01 01 02 02 03 03)
- punpckhwd xmm3,xmm3 ; xmm3=(04 04 05 05 06 06 07 07)
-
- pshufd xmm1,xmm0,0x50 ; xmm1=[col0 col1]=(00 00 00 00 01 01 01 01)
- pshufd xmm0,xmm0,0xFA ; xmm0=[col2 col3]=(02 02 02 02 03 03 03 03)
- pshufd xmm6,xmm3,0x50 ; xmm6=[col4 col5]=(04 04 04 04 05 05 05 05)
- pshufd xmm3,xmm3,0xFA ; xmm3=[col6 col7]=(06 06 06 06 07 07 07 07)
-
- jmp near .column_end
-%endif
-.columnDCT:
-
- ; -- Odd part
-
- movdqa xmm0, XMMWORD [XMMBLOCK(1,0,rsi,SIZEOF_JCOEF)]
- movdqa xmm1, XMMWORD [XMMBLOCK(3,0,rsi,SIZEOF_JCOEF)]
- pmullw xmm0, XMMWORD [XMMBLOCK(1,0,rdx,SIZEOF_ISLOW_MULT_TYPE)]
- pmullw xmm1, XMMWORD [XMMBLOCK(3,0,rdx,SIZEOF_ISLOW_MULT_TYPE)]
- movdqa xmm2, XMMWORD [XMMBLOCK(5,0,rsi,SIZEOF_JCOEF)]
- movdqa xmm3, XMMWORD [XMMBLOCK(7,0,rsi,SIZEOF_JCOEF)]
- pmullw xmm2, XMMWORD [XMMBLOCK(5,0,rdx,SIZEOF_ISLOW_MULT_TYPE)]
- pmullw xmm3, XMMWORD [XMMBLOCK(7,0,rdx,SIZEOF_ISLOW_MULT_TYPE)]
-
- movdqa xmm4,xmm0
- movdqa xmm5,xmm0
- punpcklwd xmm4,xmm1
- punpckhwd xmm5,xmm1
- movdqa xmm0,xmm4
- movdqa xmm1,xmm5
- pmaddwd xmm4,[rel PW_F256_F089] ; xmm4=(tmp2L)
- pmaddwd xmm5,[rel PW_F256_F089] ; xmm5=(tmp2H)
- pmaddwd xmm0,[rel PW_F106_MF217] ; xmm0=(tmp0L)
- pmaddwd xmm1,[rel PW_F106_MF217] ; xmm1=(tmp0H)
-
- movdqa xmm6,xmm2
- movdqa xmm7,xmm2
- punpcklwd xmm6,xmm3
- punpckhwd xmm7,xmm3
- movdqa xmm2,xmm6
- movdqa xmm3,xmm7
- pmaddwd xmm6,[rel PW_MF060_MF050] ; xmm6=(tmp2L)
- pmaddwd xmm7,[rel PW_MF060_MF050] ; xmm7=(tmp2H)
- pmaddwd xmm2,[rel PW_F145_MF021] ; xmm2=(tmp0L)
- pmaddwd xmm3,[rel PW_F145_MF021] ; xmm3=(tmp0H)
-
- paddd xmm6,xmm4 ; xmm6=tmp2L
- paddd xmm7,xmm5 ; xmm7=tmp2H
- paddd xmm2,xmm0 ; xmm2=tmp0L
- paddd xmm3,xmm1 ; xmm3=tmp0H
-
- movdqa XMMWORD [wk(0)], xmm2 ; wk(0)=tmp0L
- movdqa XMMWORD [wk(1)], xmm3 ; wk(1)=tmp0H
-
- ; -- Even part
-
- movdqa xmm4, XMMWORD [XMMBLOCK(0,0,rsi,SIZEOF_JCOEF)]
- movdqa xmm5, XMMWORD [XMMBLOCK(2,0,rsi,SIZEOF_JCOEF)]
- movdqa xmm0, XMMWORD [XMMBLOCK(6,0,rsi,SIZEOF_JCOEF)]
- pmullw xmm4, XMMWORD [XMMBLOCK(0,0,rdx,SIZEOF_ISLOW_MULT_TYPE)]
- pmullw xmm5, XMMWORD [XMMBLOCK(2,0,rdx,SIZEOF_ISLOW_MULT_TYPE)]
- pmullw xmm0, XMMWORD [XMMBLOCK(6,0,rdx,SIZEOF_ISLOW_MULT_TYPE)]
-
- pxor xmm1,xmm1
- pxor xmm2,xmm2
- punpcklwd xmm1,xmm4 ; xmm1=tmp0L
- punpckhwd xmm2,xmm4 ; xmm2=tmp0H
- psrad xmm1,(16-CONST_BITS-1) ; psrad xmm1,16 & pslld xmm1,CONST_BITS+1
- psrad xmm2,(16-CONST_BITS-1) ; psrad xmm2,16 & pslld xmm2,CONST_BITS+1
-
- movdqa xmm3,xmm5 ; xmm5=in2=z2
- punpcklwd xmm5,xmm0 ; xmm0=in6=z3
- punpckhwd xmm3,xmm0
- pmaddwd xmm5,[rel PW_F184_MF076] ; xmm5=tmp2L
- pmaddwd xmm3,[rel PW_F184_MF076] ; xmm3=tmp2H
-
- movdqa xmm4,xmm1
- movdqa xmm0,xmm2
- paddd xmm1,xmm5 ; xmm1=tmp10L
- paddd xmm2,xmm3 ; xmm2=tmp10H
- psubd xmm4,xmm5 ; xmm4=tmp12L
- psubd xmm0,xmm3 ; xmm0=tmp12H
-
- ; -- Final output stage
-
- movdqa xmm5,xmm1
- movdqa xmm3,xmm2
- paddd xmm1,xmm6 ; xmm1=data0L
- paddd xmm2,xmm7 ; xmm2=data0H
- psubd xmm5,xmm6 ; xmm5=data3L
- psubd xmm3,xmm7 ; xmm3=data3H
-
- movdqa xmm6,[rel PD_DESCALE_P1_4] ; xmm6=[rel PD_DESCALE_P1_4]
-
- paddd xmm1,xmm6
- paddd xmm2,xmm6
- psrad xmm1,DESCALE_P1_4
- psrad xmm2,DESCALE_P1_4
- paddd xmm5,xmm6
- paddd xmm3,xmm6
- psrad xmm5,DESCALE_P1_4
- psrad xmm3,DESCALE_P1_4
-
- packssdw xmm1,xmm2 ; xmm1=data0=(00 01 02 03 04 05 06 07)
- packssdw xmm5,xmm3 ; xmm5=data3=(30 31 32 33 34 35 36 37)
-
- movdqa xmm7, XMMWORD [wk(0)] ; xmm7=tmp0L
- movdqa xmm6, XMMWORD [wk(1)] ; xmm6=tmp0H
-
- movdqa xmm2,xmm4
- movdqa xmm3,xmm0
- paddd xmm4,xmm7 ; xmm4=data1L
- paddd xmm0,xmm6 ; xmm0=data1H
- psubd xmm2,xmm7 ; xmm2=data2L
- psubd xmm3,xmm6 ; xmm3=data2H
-
- movdqa xmm7,[rel PD_DESCALE_P1_4] ; xmm7=[rel PD_DESCALE_P1_4]
-
- paddd xmm4,xmm7
- paddd xmm0,xmm7
- psrad xmm4,DESCALE_P1_4
- psrad xmm0,DESCALE_P1_4
- paddd xmm2,xmm7
- paddd xmm3,xmm7
- psrad xmm2,DESCALE_P1_4
- psrad xmm3,DESCALE_P1_4
-
- packssdw xmm4,xmm0 ; xmm4=data1=(10 11 12 13 14 15 16 17)
- packssdw xmm2,xmm3 ; xmm2=data2=(20 21 22 23 24 25 26 27)
-
- movdqa xmm6,xmm1 ; transpose coefficients(phase 1)
- punpcklwd xmm1,xmm4 ; xmm1=(00 10 01 11 02 12 03 13)
- punpckhwd xmm6,xmm4 ; xmm6=(04 14 05 15 06 16 07 17)
- movdqa xmm7,xmm2 ; transpose coefficients(phase 1)
- punpcklwd xmm2,xmm5 ; xmm2=(20 30 21 31 22 32 23 33)
- punpckhwd xmm7,xmm5 ; xmm7=(24 34 25 35 26 36 27 37)
-
- movdqa xmm0,xmm1 ; transpose coefficients(phase 2)
- punpckldq xmm1,xmm2 ; xmm1=[col0 col1]=(00 10 20 30 01 11 21 31)
- punpckhdq xmm0,xmm2 ; xmm0=[col2 col3]=(02 12 22 32 03 13 23 33)
- movdqa xmm3,xmm6 ; transpose coefficients(phase 2)
- punpckldq xmm6,xmm7 ; xmm6=[col4 col5]=(04 14 24 34 05 15 25 35)
- punpckhdq xmm3,xmm7 ; xmm3=[col6 col7]=(06 16 26 36 07 17 27 37)
-.column_end:
-
- ; -- Prefetch the next coefficient block
-
- prefetchnta [rsi + DCTSIZE2*SIZEOF_JCOEF + 0*32]
- prefetchnta [rsi + DCTSIZE2*SIZEOF_JCOEF + 1*32]
- prefetchnta [rsi + DCTSIZE2*SIZEOF_JCOEF + 2*32]
- prefetchnta [rsi + DCTSIZE2*SIZEOF_JCOEF + 3*32]
-
- ; ---- Pass 2: process rows, store into output array.
-
- mov rax, [original_rbp]
- mov rdi, r12 ; (JSAMPROW *)
- mov eax, r13d
-
- ; -- Even part
-
- pxor xmm4,xmm4
- punpcklwd xmm4,xmm1 ; xmm4=tmp0
- psrad xmm4,(16-CONST_BITS-1) ; psrad xmm4,16 & pslld xmm4,CONST_BITS+1
-
- ; -- Odd part
-
- punpckhwd xmm1,xmm0
- punpckhwd xmm6,xmm3
- movdqa xmm5,xmm1
- movdqa xmm2,xmm6
- pmaddwd xmm1,[rel PW_F256_F089] ; xmm1=(tmp2)
- pmaddwd xmm6,[rel PW_MF060_MF050] ; xmm6=(tmp2)
- pmaddwd xmm5,[rel PW_F106_MF217] ; xmm5=(tmp0)
- pmaddwd xmm2,[rel PW_F145_MF021] ; xmm2=(tmp0)
-
- paddd xmm6,xmm1 ; xmm6=tmp2
- paddd xmm2,xmm5 ; xmm2=tmp0
-
- ; -- Even part
-
- punpcklwd xmm0,xmm3
- pmaddwd xmm0,[rel PW_F184_MF076] ; xmm0=tmp2
-
- movdqa xmm7,xmm4
- paddd xmm4,xmm0 ; xmm4=tmp10
- psubd xmm7,xmm0 ; xmm7=tmp12
-
- ; -- Final output stage
-
- movdqa xmm1,[rel PD_DESCALE_P2_4] ; xmm1=[rel PD_DESCALE_P2_4]
-
- movdqa xmm5,xmm4
- movdqa xmm3,xmm7
- paddd xmm4,xmm6 ; xmm4=data0=(00 10 20 30)
- paddd xmm7,xmm2 ; xmm7=data1=(01 11 21 31)
- psubd xmm5,xmm6 ; xmm5=data3=(03 13 23 33)
- psubd xmm3,xmm2 ; xmm3=data2=(02 12 22 32)
-
- paddd xmm4,xmm1
- paddd xmm7,xmm1
- psrad xmm4,DESCALE_P2_4
- psrad xmm7,DESCALE_P2_4
- paddd xmm5,xmm1
- paddd xmm3,xmm1
- psrad xmm5,DESCALE_P2_4
- psrad xmm3,DESCALE_P2_4
-
- packssdw xmm4,xmm3 ; xmm4=(00 10 20 30 02 12 22 32)
- packssdw xmm7,xmm5 ; xmm7=(01 11 21 31 03 13 23 33)
-
- movdqa xmm0,xmm4 ; transpose coefficients(phase 1)
- punpcklwd xmm4,xmm7 ; xmm4=(00 01 10 11 20 21 30 31)
- punpckhwd xmm0,xmm7 ; xmm0=(02 03 12 13 22 23 32 33)
-
- movdqa xmm6,xmm4 ; transpose coefficients(phase 2)
- punpckldq xmm4,xmm0 ; xmm4=(00 01 02 03 10 11 12 13)
- punpckhdq xmm6,xmm0 ; xmm6=(20 21 22 23 30 31 32 33)
-
- packsswb xmm4,xmm6 ; xmm4=(00 01 02 03 10 11 12 13 20 ..)
- paddb xmm4,[rel PB_CENTERJSAMP]
-
- pshufd xmm2,xmm4,0x39 ; xmm2=(10 11 12 13 20 21 22 23 30 ..)
- pshufd xmm1,xmm4,0x4E ; xmm1=(20 21 22 23 30 31 32 33 00 ..)
- pshufd xmm3,xmm4,0x93 ; xmm3=(30 31 32 33 00 01 02 03 10 ..)
-
- mov rdx, JSAMPROW [rdi+0*SIZEOF_JSAMPROW]
- mov rsi, JSAMPROW [rdi+1*SIZEOF_JSAMPROW]
- movd XMM_DWORD [rdx+rax*SIZEOF_JSAMPLE], xmm4
- movd XMM_DWORD [rsi+rax*SIZEOF_JSAMPLE], xmm2
- mov rdx, JSAMPROW [rdi+2*SIZEOF_JSAMPROW]
- mov rsi, JSAMPROW [rdi+3*SIZEOF_JSAMPROW]
- movd XMM_DWORD [rdx+rax*SIZEOF_JSAMPLE], xmm1
- movd XMM_DWORD [rsi+rax*SIZEOF_JSAMPLE], xmm3
-
- uncollect_args
- mov rsp,rbp ; rsp <- aligned rbp
- pop rsp ; rsp <- original rbp
- pop rbp
- ret
-
-
-; --------------------------------------------------------------------------
-;
-; Perform dequantization and inverse DCT on one block of coefficients,
-; producing a reduced-size 2x2 output block.
-;
-; GLOBAL(void)
-; jsimd_idct_2x2_sse2 (void *dct_table, JCOEFPTR coef_block,
-; JSAMPARRAY output_buf, JDIMENSION output_col)
-;
-
-; r10 = void *dct_table
-; r11 = JCOEFPTR coef_block
-; r12 = JSAMPARRAY output_buf
-; r13 = JDIMENSION output_col
-
- align 16
- global EXTN(jsimd_idct_2x2_sse2)
-
-EXTN(jsimd_idct_2x2_sse2):
- push rbp
- mov rax,rsp
- mov rbp,rsp
- collect_args
- push rbx
-
- ; ---- Pass 1: process columns from input.
-
- mov rdx, r10 ; quantptr
- mov rsi, r11 ; inptr
-
- ; | input: | result: |
- ; | 00 01 ** 03 ** 05 ** 07 | |
- ; | 10 11 ** 13 ** 15 ** 17 | |
- ; | ** ** ** ** ** ** ** ** | |
- ; | 30 31 ** 33 ** 35 ** 37 | A0 A1 A3 A5 A7 |
- ; | ** ** ** ** ** ** ** ** | B0 B1 B3 B5 B7 |
- ; | 50 51 ** 53 ** 55 ** 57 | |
- ; | ** ** ** ** ** ** ** ** | |
- ; | 70 71 ** 73 ** 75 ** 77 | |
-
- ; -- Odd part
-
- movdqa xmm0, XMMWORD [XMMBLOCK(1,0,rsi,SIZEOF_JCOEF)]
- movdqa xmm1, XMMWORD [XMMBLOCK(3,0,rsi,SIZEOF_JCOEF)]
- pmullw xmm0, XMMWORD [XMMBLOCK(1,0,rdx,SIZEOF_ISLOW_MULT_TYPE)]
- pmullw xmm1, XMMWORD [XMMBLOCK(3,0,rdx,SIZEOF_ISLOW_MULT_TYPE)]
- movdqa xmm2, XMMWORD [XMMBLOCK(5,0,rsi,SIZEOF_JCOEF)]
- movdqa xmm3, XMMWORD [XMMBLOCK(7,0,rsi,SIZEOF_JCOEF)]
- pmullw xmm2, XMMWORD [XMMBLOCK(5,0,rdx,SIZEOF_ISLOW_MULT_TYPE)]
- pmullw xmm3, XMMWORD [XMMBLOCK(7,0,rdx,SIZEOF_ISLOW_MULT_TYPE)]
-
- ; xmm0=(10 11 ** 13 ** 15 ** 17), xmm1=(30 31 ** 33 ** 35 ** 37)
- ; xmm2=(50 51 ** 53 ** 55 ** 57), xmm3=(70 71 ** 73 ** 75 ** 77)
-
- pcmpeqd xmm7,xmm7
- pslld xmm7,WORD_BIT ; xmm7={0x0000 0xFFFF 0x0000 0xFFFF ..}
-
- movdqa xmm4,xmm0 ; xmm4=(10 11 ** 13 ** 15 ** 17)
- movdqa xmm5,xmm2 ; xmm5=(50 51 ** 53 ** 55 ** 57)
- punpcklwd xmm4,xmm1 ; xmm4=(10 30 11 31 ** ** 13 33)
- punpcklwd xmm5,xmm3 ; xmm5=(50 70 51 71 ** ** 53 73)
- pmaddwd xmm4,[rel PW_F362_MF127]
- pmaddwd xmm5,[rel PW_F085_MF072]
-
- psrld xmm0,WORD_BIT ; xmm0=(11 -- 13 -- 15 -- 17 --)
- pand xmm1,xmm7 ; xmm1=(-- 31 -- 33 -- 35 -- 37)
- psrld xmm2,WORD_BIT ; xmm2=(51 -- 53 -- 55 -- 57 --)
- pand xmm3,xmm7 ; xmm3=(-- 71 -- 73 -- 75 -- 77)
- por xmm0,xmm1 ; xmm0=(11 31 13 33 15 35 17 37)
- por xmm2,xmm3 ; xmm2=(51 71 53 73 55 75 57 77)
- pmaddwd xmm0,[rel PW_F362_MF127]
- pmaddwd xmm2,[rel PW_F085_MF072]
-
- paddd xmm4,xmm5 ; xmm4=tmp0[col0 col1 **** col3]
- paddd xmm0,xmm2 ; xmm0=tmp0[col1 col3 col5 col7]
-
- ; -- Even part
-
- movdqa xmm6, XMMWORD [XMMBLOCK(0,0,rsi,SIZEOF_JCOEF)]
- pmullw xmm6, XMMWORD [XMMBLOCK(0,0,rdx,SIZEOF_ISLOW_MULT_TYPE)]
-
- ; xmm6=(00 01 ** 03 ** 05 ** 07)
-
- movdqa xmm1,xmm6 ; xmm1=(00 01 ** 03 ** 05 ** 07)
- pslld xmm6,WORD_BIT ; xmm6=(-- 00 -- ** -- ** -- **)
- pand xmm1,xmm7 ; xmm1=(-- 01 -- 03 -- 05 -- 07)
- psrad xmm6,(WORD_BIT-CONST_BITS-2) ; xmm6=tmp10[col0 **** **** ****]
- psrad xmm1,(WORD_BIT-CONST_BITS-2) ; xmm1=tmp10[col1 col3 col5 col7]
-
- ; -- Final output stage
-
- movdqa xmm3,xmm6
- movdqa xmm5,xmm1
- paddd xmm6,xmm4 ; xmm6=data0[col0 **** **** ****]=(A0 ** ** **)
- paddd xmm1,xmm0 ; xmm1=data0[col1 col3 col5 col7]=(A1 A3 A5 A7)
- psubd xmm3,xmm4 ; xmm3=data1[col0 **** **** ****]=(B0 ** ** **)
- psubd xmm5,xmm0 ; xmm5=data1[col1 col3 col5 col7]=(B1 B3 B5 B7)
-
- movdqa xmm2,[rel PD_DESCALE_P1_2] ; xmm2=[rel PD_DESCALE_P1_2]
-
- punpckldq xmm6,xmm3 ; xmm6=(A0 B0 ** **)
-
- movdqa xmm7,xmm1
- punpcklqdq xmm1,xmm5 ; xmm1=(A1 A3 B1 B3)
- punpckhqdq xmm7,xmm5 ; xmm7=(A5 A7 B5 B7)
-
- paddd xmm6,xmm2
- psrad xmm6,DESCALE_P1_2
-
- paddd xmm1,xmm2
- paddd xmm7,xmm2
- psrad xmm1,DESCALE_P1_2
- psrad xmm7,DESCALE_P1_2
-
- ; -- Prefetch the next coefficient block
-
- prefetchnta [rsi + DCTSIZE2*SIZEOF_JCOEF + 0*32]
- prefetchnta [rsi + DCTSIZE2*SIZEOF_JCOEF + 1*32]
- prefetchnta [rsi + DCTSIZE2*SIZEOF_JCOEF + 2*32]
- prefetchnta [rsi + DCTSIZE2*SIZEOF_JCOEF + 3*32]
-
- ; ---- Pass 2: process rows, store into output array.
-
- mov rdi, r12 ; (JSAMPROW *)
- mov eax, r13d
-
- ; | input:| result:|
- ; | A0 B0 | |
- ; | A1 B1 | C0 C1 |
- ; | A3 B3 | D0 D1 |
- ; | A5 B5 | |
- ; | A7 B7 | |
-
- ; -- Odd part
-
- packssdw xmm1,xmm1 ; xmm1=(A1 A3 B1 B3 A1 A3 B1 B3)
- packssdw xmm7,xmm7 ; xmm7=(A5 A7 B5 B7 A5 A7 B5 B7)
- pmaddwd xmm1,[rel PW_F362_MF127]
- pmaddwd xmm7,[rel PW_F085_MF072]
-
- paddd xmm1,xmm7 ; xmm1=tmp0[row0 row1 row0 row1]
-
- ; -- Even part
-
- pslld xmm6,(CONST_BITS+2) ; xmm6=tmp10[row0 row1 **** ****]
-
- ; -- Final output stage
-
- movdqa xmm4,xmm6
- paddd xmm6,xmm1 ; xmm6=data0[row0 row1 **** ****]=(C0 C1 ** **)
- psubd xmm4,xmm1 ; xmm4=data1[row0 row1 **** ****]=(D0 D1 ** **)
-
- punpckldq xmm6,xmm4 ; xmm6=(C0 D0 C1 D1)
-
- paddd xmm6,[rel PD_DESCALE_P2_2]
- psrad xmm6,DESCALE_P2_2
-
- packssdw xmm6,xmm6 ; xmm6=(C0 D0 C1 D1 C0 D0 C1 D1)
- packsswb xmm6,xmm6 ; xmm6=(C0 D0 C1 D1 C0 D0 C1 D1 ..)
- paddb xmm6,[rel PB_CENTERJSAMP]
-
- pextrw ebx,xmm6,0x00 ; ebx=(C0 D0 -- --)
- pextrw ecx,xmm6,0x01 ; ecx=(C1 D1 -- --)
-
- mov rdx, JSAMPROW [rdi+0*SIZEOF_JSAMPROW]
- mov rsi, JSAMPROW [rdi+1*SIZEOF_JSAMPROW]
- mov WORD [rdx+rax*SIZEOF_JSAMPLE], bx
- mov WORD [rsi+rax*SIZEOF_JSAMPLE], cx
-
- pop rbx
- uncollect_args
- pop rbp
- ret
-
-; For some reason, the OS X linker does not honor the request to align the
-; segment unless we do this.
- align 16
diff --git a/simd/jidctred-sse2.asm b/simd/jidctred-sse2.asm
deleted file mode 100644
index 232d983..0000000
--- a/simd/jidctred-sse2.asm
+++ /dev/null
@@ -1,593 +0,0 @@
-;
-; jidctred.asm - reduced-size IDCT (SSE2)
-;
-; Copyright 2009 Pierre Ossman <ossman@cendio.se> for Cendio AB
-;
-; Based on the x86 SIMD extension for IJG JPEG library
-; Copyright (C) 1999-2006, MIYASAKA Masaru.
-; For conditions of distribution and use, see copyright notice in jsimdext.inc
-;
-; This file should be assembled with NASM (Netwide Assembler),
-; can *not* be assembled with Microsoft's MASM or any compatible
-; assembler (including Borland's Turbo Assembler).
-; NASM is available from http://nasm.sourceforge.net/ or
-; http://sourceforge.net/project/showfiles.php?group_id=6208
-;
-; This file contains inverse-DCT routines that produce reduced-size
-; output: either 4x4 or 2x2 pixels from an 8x8 DCT block.
-; The following code is based directly on the IJG's original jidctred.c;
-; see the jidctred.c for more details.
-;
-; [TAB8]
-
-%include "jsimdext.inc"
-%include "jdct.inc"
-
-; --------------------------------------------------------------------------
-
-%define CONST_BITS 13
-%define PASS1_BITS 2
-
-%define DESCALE_P1_4 (CONST_BITS-PASS1_BITS+1)
-%define DESCALE_P2_4 (CONST_BITS+PASS1_BITS+3+1)
-%define DESCALE_P1_2 (CONST_BITS-PASS1_BITS+2)
-%define DESCALE_P2_2 (CONST_BITS+PASS1_BITS+3+2)
-
-%if CONST_BITS == 13
-F_0_211 equ 1730 ; FIX(0.211164243)
-F_0_509 equ 4176 ; FIX(0.509795579)
-F_0_601 equ 4926 ; FIX(0.601344887)
-F_0_720 equ 5906 ; FIX(0.720959822)
-F_0_765 equ 6270 ; FIX(0.765366865)
-F_0_850 equ 6967 ; FIX(0.850430095)
-F_0_899 equ 7373 ; FIX(0.899976223)
-F_1_061 equ 8697 ; FIX(1.061594337)
-F_1_272 equ 10426 ; FIX(1.272758580)
-F_1_451 equ 11893 ; FIX(1.451774981)
-F_1_847 equ 15137 ; FIX(1.847759065)
-F_2_172 equ 17799 ; FIX(2.172734803)
-F_2_562 equ 20995 ; FIX(2.562915447)
-F_3_624 equ 29692 ; FIX(3.624509785)
-%else
-; NASM cannot do compile-time arithmetic on floating-point constants.
-%define DESCALE(x,n) (((x)+(1<<((n)-1)))>>(n))
-F_0_211 equ DESCALE( 226735879,30-CONST_BITS) ; FIX(0.211164243)
-F_0_509 equ DESCALE( 547388834,30-CONST_BITS) ; FIX(0.509795579)
-F_0_601 equ DESCALE( 645689155,30-CONST_BITS) ; FIX(0.601344887)
-F_0_720 equ DESCALE( 774124714,30-CONST_BITS) ; FIX(0.720959822)
-F_0_765 equ DESCALE( 821806413,30-CONST_BITS) ; FIX(0.765366865)
-F_0_850 equ DESCALE( 913142361,30-CONST_BITS) ; FIX(0.850430095)
-F_0_899 equ DESCALE( 966342111,30-CONST_BITS) ; FIX(0.899976223)
-F_1_061 equ DESCALE(1139878239,30-CONST_BITS) ; FIX(1.061594337)
-F_1_272 equ DESCALE(1366614119,30-CONST_BITS) ; FIX(1.272758580)
-F_1_451 equ DESCALE(1558831516,30-CONST_BITS) ; FIX(1.451774981)
-F_1_847 equ DESCALE(1984016188,30-CONST_BITS) ; FIX(1.847759065)
-F_2_172 equ DESCALE(2332956230,30-CONST_BITS) ; FIX(2.172734803)
-F_2_562 equ DESCALE(2751909506,30-CONST_BITS) ; FIX(2.562915447)
-F_3_624 equ DESCALE(3891787747,30-CONST_BITS) ; FIX(3.624509785)
-%endif
-
-; --------------------------------------------------------------------------
- SECTION SEG_CONST
-
- alignz 16
- global EXTN(jconst_idct_red_sse2)
-
-EXTN(jconst_idct_red_sse2):
-
-PW_F184_MF076 times 4 dw F_1_847,-F_0_765
-PW_F256_F089 times 4 dw F_2_562, F_0_899
-PW_F106_MF217 times 4 dw F_1_061,-F_2_172
-PW_MF060_MF050 times 4 dw -F_0_601,-F_0_509
-PW_F145_MF021 times 4 dw F_1_451,-F_0_211
-PW_F362_MF127 times 4 dw F_3_624,-F_1_272
-PW_F085_MF072 times 4 dw F_0_850,-F_0_720
-PD_DESCALE_P1_4 times 4 dd 1 << (DESCALE_P1_4-1)
-PD_DESCALE_P2_4 times 4 dd 1 << (DESCALE_P2_4-1)
-PD_DESCALE_P1_2 times 4 dd 1 << (DESCALE_P1_2-1)
-PD_DESCALE_P2_2 times 4 dd 1 << (DESCALE_P2_2-1)
-PB_CENTERJSAMP times 16 db CENTERJSAMPLE
-
- alignz 16
-
-; --------------------------------------------------------------------------
- SECTION SEG_TEXT
- BITS 32
-;
-; Perform dequantization and inverse DCT on one block of coefficients,
-; producing a reduced-size 4x4 output block.
-;
-; GLOBAL(void)
-; jsimd_idct_4x4_sse2 (void *dct_table, JCOEFPTR coef_block,
-; JSAMPARRAY output_buf, JDIMENSION output_col)
-;
-
-%define dct_table(b) (b)+8 ; void *dct_table
-%define coef_block(b) (b)+12 ; JCOEFPTR coef_block
-%define output_buf(b) (b)+16 ; JSAMPARRAY output_buf
-%define output_col(b) (b)+20 ; JDIMENSION output_col
-
-%define original_ebp ebp+0
-%define wk(i) ebp-(WK_NUM-(i))*SIZEOF_XMMWORD ; xmmword wk[WK_NUM]
-%define WK_NUM 2
-
- align 16
- global EXTN(jsimd_idct_4x4_sse2)
-
-EXTN(jsimd_idct_4x4_sse2):
- push ebp
- mov eax,esp ; eax = original ebp
- sub esp, byte 4
- and esp, byte (-SIZEOF_XMMWORD) ; align to 128 bits
- mov [esp],eax
- mov ebp,esp ; ebp = aligned ebp
- lea esp, [wk(0)]
- pushpic ebx
-; push ecx ; unused
-; push edx ; need not be preserved
- push esi
- push edi
-
- get_GOT ebx ; get GOT address
-
- ; ---- Pass 1: process columns from input.
-
-; mov eax, [original_ebp]
- mov edx, POINTER [dct_table(eax)] ; quantptr
- mov esi, JCOEFPTR [coef_block(eax)] ; inptr
-
-%ifndef NO_ZERO_COLUMN_TEST_4X4_SSE2
- mov eax, DWORD [DWBLOCK(1,0,esi,SIZEOF_JCOEF)]
- or eax, DWORD [DWBLOCK(2,0,esi,SIZEOF_JCOEF)]
- jnz short .columnDCT
-
- movdqa xmm0, XMMWORD [XMMBLOCK(1,0,esi,SIZEOF_JCOEF)]
- movdqa xmm1, XMMWORD [XMMBLOCK(2,0,esi,SIZEOF_JCOEF)]
- por xmm0, XMMWORD [XMMBLOCK(3,0,esi,SIZEOF_JCOEF)]
- por xmm1, XMMWORD [XMMBLOCK(5,0,esi,SIZEOF_JCOEF)]
- por xmm0, XMMWORD [XMMBLOCK(6,0,esi,SIZEOF_JCOEF)]
- por xmm1, XMMWORD [XMMBLOCK(7,0,esi,SIZEOF_JCOEF)]
- por xmm0,xmm1
- packsswb xmm0,xmm0
- packsswb xmm0,xmm0
- movd eax,xmm0
- test eax,eax
- jnz short .columnDCT
-
- ; -- AC terms all zero
-
- movdqa xmm0, XMMWORD [XMMBLOCK(0,0,esi,SIZEOF_JCOEF)]
- pmullw xmm0, XMMWORD [XMMBLOCK(0,0,edx,SIZEOF_ISLOW_MULT_TYPE)]
-
- psllw xmm0,PASS1_BITS
-
- movdqa xmm3,xmm0 ; xmm0=in0=(00 01 02 03 04 05 06 07)
- punpcklwd xmm0,xmm0 ; xmm0=(00 00 01 01 02 02 03 03)
- punpckhwd xmm3,xmm3 ; xmm3=(04 04 05 05 06 06 07 07)
-
- pshufd xmm1,xmm0,0x50 ; xmm1=[col0 col1]=(00 00 00 00 01 01 01 01)
- pshufd xmm0,xmm0,0xFA ; xmm0=[col2 col3]=(02 02 02 02 03 03 03 03)
- pshufd xmm6,xmm3,0x50 ; xmm6=[col4 col5]=(04 04 04 04 05 05 05 05)
- pshufd xmm3,xmm3,0xFA ; xmm3=[col6 col7]=(06 06 06 06 07 07 07 07)
-
- jmp near .column_end
- alignx 16,7
-%endif
-.columnDCT:
-
- ; -- Odd part
-
- movdqa xmm0, XMMWORD [XMMBLOCK(1,0,esi,SIZEOF_JCOEF)]
- movdqa xmm1, XMMWORD [XMMBLOCK(3,0,esi,SIZEOF_JCOEF)]
- pmullw xmm0, XMMWORD [XMMBLOCK(1,0,edx,SIZEOF_ISLOW_MULT_TYPE)]
- pmullw xmm1, XMMWORD [XMMBLOCK(3,0,edx,SIZEOF_ISLOW_MULT_TYPE)]
- movdqa xmm2, XMMWORD [XMMBLOCK(5,0,esi,SIZEOF_JCOEF)]
- movdqa xmm3, XMMWORD [XMMBLOCK(7,0,esi,SIZEOF_JCOEF)]
- pmullw xmm2, XMMWORD [XMMBLOCK(5,0,edx,SIZEOF_ISLOW_MULT_TYPE)]
- pmullw xmm3, XMMWORD [XMMBLOCK(7,0,edx,SIZEOF_ISLOW_MULT_TYPE)]
-
- movdqa xmm4,xmm0
- movdqa xmm5,xmm0
- punpcklwd xmm4,xmm1
- punpckhwd xmm5,xmm1
- movdqa xmm0,xmm4
- movdqa xmm1,xmm5
- pmaddwd xmm4,[GOTOFF(ebx,PW_F256_F089)] ; xmm4=(tmp2L)
- pmaddwd xmm5,[GOTOFF(ebx,PW_F256_F089)] ; xmm5=(tmp2H)
- pmaddwd xmm0,[GOTOFF(ebx,PW_F106_MF217)] ; xmm0=(tmp0L)
- pmaddwd xmm1,[GOTOFF(ebx,PW_F106_MF217)] ; xmm1=(tmp0H)
-
- movdqa xmm6,xmm2
- movdqa xmm7,xmm2
- punpcklwd xmm6,xmm3
- punpckhwd xmm7,xmm3
- movdqa xmm2,xmm6
- movdqa xmm3,xmm7
- pmaddwd xmm6,[GOTOFF(ebx,PW_MF060_MF050)] ; xmm6=(tmp2L)
- pmaddwd xmm7,[GOTOFF(ebx,PW_MF060_MF050)] ; xmm7=(tmp2H)
- pmaddwd xmm2,[GOTOFF(ebx,PW_F145_MF021)] ; xmm2=(tmp0L)
- pmaddwd xmm3,[GOTOFF(ebx,PW_F145_MF021)] ; xmm3=(tmp0H)
-
- paddd xmm6,xmm4 ; xmm6=tmp2L
- paddd xmm7,xmm5 ; xmm7=tmp2H
- paddd xmm2,xmm0 ; xmm2=tmp0L
- paddd xmm3,xmm1 ; xmm3=tmp0H
-
- movdqa XMMWORD [wk(0)], xmm2 ; wk(0)=tmp0L
- movdqa XMMWORD [wk(1)], xmm3 ; wk(1)=tmp0H
-
- ; -- Even part
-
- movdqa xmm4, XMMWORD [XMMBLOCK(0,0,esi,SIZEOF_JCOEF)]
- movdqa xmm5, XMMWORD [XMMBLOCK(2,0,esi,SIZEOF_JCOEF)]
- movdqa xmm0, XMMWORD [XMMBLOCK(6,0,esi,SIZEOF_JCOEF)]
- pmullw xmm4, XMMWORD [XMMBLOCK(0,0,edx,SIZEOF_ISLOW_MULT_TYPE)]
- pmullw xmm5, XMMWORD [XMMBLOCK(2,0,edx,SIZEOF_ISLOW_MULT_TYPE)]
- pmullw xmm0, XMMWORD [XMMBLOCK(6,0,edx,SIZEOF_ISLOW_MULT_TYPE)]
-
- pxor xmm1,xmm1
- pxor xmm2,xmm2
- punpcklwd xmm1,xmm4 ; xmm1=tmp0L
- punpckhwd xmm2,xmm4 ; xmm2=tmp0H
- psrad xmm1,(16-CONST_BITS-1) ; psrad xmm1,16 & pslld xmm1,CONST_BITS+1
- psrad xmm2,(16-CONST_BITS-1) ; psrad xmm2,16 & pslld xmm2,CONST_BITS+1
-
- movdqa xmm3,xmm5 ; xmm5=in2=z2
- punpcklwd xmm5,xmm0 ; xmm0=in6=z3
- punpckhwd xmm3,xmm0
- pmaddwd xmm5,[GOTOFF(ebx,PW_F184_MF076)] ; xmm5=tmp2L
- pmaddwd xmm3,[GOTOFF(ebx,PW_F184_MF076)] ; xmm3=tmp2H
-
- movdqa xmm4,xmm1
- movdqa xmm0,xmm2
- paddd xmm1,xmm5 ; xmm1=tmp10L
- paddd xmm2,xmm3 ; xmm2=tmp10H
- psubd xmm4,xmm5 ; xmm4=tmp12L
- psubd xmm0,xmm3 ; xmm0=tmp12H
-
- ; -- Final output stage
-
- movdqa xmm5,xmm1
- movdqa xmm3,xmm2
- paddd xmm1,xmm6 ; xmm1=data0L
- paddd xmm2,xmm7 ; xmm2=data0H
- psubd xmm5,xmm6 ; xmm5=data3L
- psubd xmm3,xmm7 ; xmm3=data3H
-
- movdqa xmm6,[GOTOFF(ebx,PD_DESCALE_P1_4)] ; xmm6=[PD_DESCALE_P1_4]
-
- paddd xmm1,xmm6
- paddd xmm2,xmm6
- psrad xmm1,DESCALE_P1_4
- psrad xmm2,DESCALE_P1_4
- paddd xmm5,xmm6
- paddd xmm3,xmm6
- psrad xmm5,DESCALE_P1_4
- psrad xmm3,DESCALE_P1_4
-
- packssdw xmm1,xmm2 ; xmm1=data0=(00 01 02 03 04 05 06 07)
- packssdw xmm5,xmm3 ; xmm5=data3=(30 31 32 33 34 35 36 37)
-
- movdqa xmm7, XMMWORD [wk(0)] ; xmm7=tmp0L
- movdqa xmm6, XMMWORD [wk(1)] ; xmm6=tmp0H
-
- movdqa xmm2,xmm4
- movdqa xmm3,xmm0
- paddd xmm4,xmm7 ; xmm4=data1L
- paddd xmm0,xmm6 ; xmm0=data1H
- psubd xmm2,xmm7 ; xmm2=data2L
- psubd xmm3,xmm6 ; xmm3=data2H
-
- movdqa xmm7,[GOTOFF(ebx,PD_DESCALE_P1_4)] ; xmm7=[PD_DESCALE_P1_4]
-
- paddd xmm4,xmm7
- paddd xmm0,xmm7
- psrad xmm4,DESCALE_P1_4
- psrad xmm0,DESCALE_P1_4
- paddd xmm2,xmm7
- paddd xmm3,xmm7
- psrad xmm2,DESCALE_P1_4
- psrad xmm3,DESCALE_P1_4
-
- packssdw xmm4,xmm0 ; xmm4=data1=(10 11 12 13 14 15 16 17)
- packssdw xmm2,xmm3 ; xmm2=data2=(20 21 22 23 24 25 26 27)
-
- movdqa xmm6,xmm1 ; transpose coefficients(phase 1)
- punpcklwd xmm1,xmm4 ; xmm1=(00 10 01 11 02 12 03 13)
- punpckhwd xmm6,xmm4 ; xmm6=(04 14 05 15 06 16 07 17)
- movdqa xmm7,xmm2 ; transpose coefficients(phase 1)
- punpcklwd xmm2,xmm5 ; xmm2=(20 30 21 31 22 32 23 33)
- punpckhwd xmm7,xmm5 ; xmm7=(24 34 25 35 26 36 27 37)
-
- movdqa xmm0,xmm1 ; transpose coefficients(phase 2)
- punpckldq xmm1,xmm2 ; xmm1=[col0 col1]=(00 10 20 30 01 11 21 31)
- punpckhdq xmm0,xmm2 ; xmm0=[col2 col3]=(02 12 22 32 03 13 23 33)
- movdqa xmm3,xmm6 ; transpose coefficients(phase 2)
- punpckldq xmm6,xmm7 ; xmm6=[col4 col5]=(04 14 24 34 05 15 25 35)
- punpckhdq xmm3,xmm7 ; xmm3=[col6 col7]=(06 16 26 36 07 17 27 37)
-.column_end:
-
- ; -- Prefetch the next coefficient block
-
- prefetchnta [esi + DCTSIZE2*SIZEOF_JCOEF + 0*32]
- prefetchnta [esi + DCTSIZE2*SIZEOF_JCOEF + 1*32]
- prefetchnta [esi + DCTSIZE2*SIZEOF_JCOEF + 2*32]
- prefetchnta [esi + DCTSIZE2*SIZEOF_JCOEF + 3*32]
-
- ; ---- Pass 2: process rows, store into output array.
-
- mov eax, [original_ebp]
- mov edi, JSAMPARRAY [output_buf(eax)] ; (JSAMPROW *)
- mov eax, JDIMENSION [output_col(eax)]
-
- ; -- Even part
-
- pxor xmm4,xmm4
- punpcklwd xmm4,xmm1 ; xmm4=tmp0
- psrad xmm4,(16-CONST_BITS-1) ; psrad xmm4,16 & pslld xmm4,CONST_BITS+1
-
- ; -- Odd part
-
- punpckhwd xmm1,xmm0
- punpckhwd xmm6,xmm3
- movdqa xmm5,xmm1
- movdqa xmm2,xmm6
- pmaddwd xmm1,[GOTOFF(ebx,PW_F256_F089)] ; xmm1=(tmp2)
- pmaddwd xmm6,[GOTOFF(ebx,PW_MF060_MF050)] ; xmm6=(tmp2)
- pmaddwd xmm5,[GOTOFF(ebx,PW_F106_MF217)] ; xmm5=(tmp0)
- pmaddwd xmm2,[GOTOFF(ebx,PW_F145_MF021)] ; xmm2=(tmp0)
-
- paddd xmm6,xmm1 ; xmm6=tmp2
- paddd xmm2,xmm5 ; xmm2=tmp0
-
- ; -- Even part
-
- punpcklwd xmm0,xmm3
- pmaddwd xmm0,[GOTOFF(ebx,PW_F184_MF076)] ; xmm0=tmp2
-
- movdqa xmm7,xmm4
- paddd xmm4,xmm0 ; xmm4=tmp10
- psubd xmm7,xmm0 ; xmm7=tmp12
-
- ; -- Final output stage
-
- movdqa xmm1,[GOTOFF(ebx,PD_DESCALE_P2_4)] ; xmm1=[PD_DESCALE_P2_4]
-
- movdqa xmm5,xmm4
- movdqa xmm3,xmm7
- paddd xmm4,xmm6 ; xmm4=data0=(00 10 20 30)
- paddd xmm7,xmm2 ; xmm7=data1=(01 11 21 31)
- psubd xmm5,xmm6 ; xmm5=data3=(03 13 23 33)
- psubd xmm3,xmm2 ; xmm3=data2=(02 12 22 32)
-
- paddd xmm4,xmm1
- paddd xmm7,xmm1
- psrad xmm4,DESCALE_P2_4
- psrad xmm7,DESCALE_P2_4
- paddd xmm5,xmm1
- paddd xmm3,xmm1
- psrad xmm5,DESCALE_P2_4
- psrad xmm3,DESCALE_P2_4
-
- packssdw xmm4,xmm3 ; xmm4=(00 10 20 30 02 12 22 32)
- packssdw xmm7,xmm5 ; xmm7=(01 11 21 31 03 13 23 33)
-
- movdqa xmm0,xmm4 ; transpose coefficients(phase 1)
- punpcklwd xmm4,xmm7 ; xmm4=(00 01 10 11 20 21 30 31)
- punpckhwd xmm0,xmm7 ; xmm0=(02 03 12 13 22 23 32 33)
-
- movdqa xmm6,xmm4 ; transpose coefficients(phase 2)
- punpckldq xmm4,xmm0 ; xmm4=(00 01 02 03 10 11 12 13)
- punpckhdq xmm6,xmm0 ; xmm6=(20 21 22 23 30 31 32 33)
-
- packsswb xmm4,xmm6 ; xmm4=(00 01 02 03 10 11 12 13 20 ..)
- paddb xmm4,[GOTOFF(ebx,PB_CENTERJSAMP)]
-
- pshufd xmm2,xmm4,0x39 ; xmm2=(10 11 12 13 20 21 22 23 30 ..)
- pshufd xmm1,xmm4,0x4E ; xmm1=(20 21 22 23 30 31 32 33 00 ..)
- pshufd xmm3,xmm4,0x93 ; xmm3=(30 31 32 33 00 01 02 03 10 ..)
-
- mov edx, JSAMPROW [edi+0*SIZEOF_JSAMPROW]
- mov esi, JSAMPROW [edi+1*SIZEOF_JSAMPROW]
- movd XMM_DWORD [edx+eax*SIZEOF_JSAMPLE], xmm4
- movd XMM_DWORD [esi+eax*SIZEOF_JSAMPLE], xmm2
- mov edx, JSAMPROW [edi+2*SIZEOF_JSAMPROW]
- mov esi, JSAMPROW [edi+3*SIZEOF_JSAMPROW]
- movd XMM_DWORD [edx+eax*SIZEOF_JSAMPLE], xmm1
- movd XMM_DWORD [esi+eax*SIZEOF_JSAMPLE], xmm3
-
- pop edi
- pop esi
-; pop edx ; need not be preserved
-; pop ecx ; unused
- poppic ebx
- mov esp,ebp ; esp <- aligned ebp
- pop esp ; esp <- original ebp
- pop ebp
- ret
-
-
-; --------------------------------------------------------------------------
-;
-; Perform dequantization and inverse DCT on one block of coefficients,
-; producing a reduced-size 2x2 output block.
-;
-; GLOBAL(void)
-; jsimd_idct_2x2_sse2 (void *dct_table, JCOEFPTR coef_block,
-; JSAMPARRAY output_buf, JDIMENSION output_col)
-;
-
-%define dct_table(b) (b)+8 ; void *dct_table
-%define coef_block(b) (b)+12 ; JCOEFPTR coef_block
-%define output_buf(b) (b)+16 ; JSAMPARRAY output_buf
-%define output_col(b) (b)+20 ; JDIMENSION output_col
-
- align 16
- global EXTN(jsimd_idct_2x2_sse2)
-
-EXTN(jsimd_idct_2x2_sse2):
- push ebp
- mov ebp,esp
- push ebx
-; push ecx ; need not be preserved
-; push edx ; need not be preserved
- push esi
- push edi
-
- get_GOT ebx ; get GOT address
-
- ; ---- Pass 1: process columns from input.
-
- mov edx, POINTER [dct_table(ebp)] ; quantptr
- mov esi, JCOEFPTR [coef_block(ebp)] ; inptr
-
- ; | input: | result: |
- ; | 00 01 ** 03 ** 05 ** 07 | |
- ; | 10 11 ** 13 ** 15 ** 17 | |
- ; | ** ** ** ** ** ** ** ** | |
- ; | 30 31 ** 33 ** 35 ** 37 | A0 A1 A3 A5 A7 |
- ; | ** ** ** ** ** ** ** ** | B0 B1 B3 B5 B7 |
- ; | 50 51 ** 53 ** 55 ** 57 | |
- ; | ** ** ** ** ** ** ** ** | |
- ; | 70 71 ** 73 ** 75 ** 77 | |
-
- ; -- Odd part
-
- movdqa xmm0, XMMWORD [XMMBLOCK(1,0,esi,SIZEOF_JCOEF)]
- movdqa xmm1, XMMWORD [XMMBLOCK(3,0,esi,SIZEOF_JCOEF)]
- pmullw xmm0, XMMWORD [XMMBLOCK(1,0,edx,SIZEOF_ISLOW_MULT_TYPE)]
- pmullw xmm1, XMMWORD [XMMBLOCK(3,0,edx,SIZEOF_ISLOW_MULT_TYPE)]
- movdqa xmm2, XMMWORD [XMMBLOCK(5,0,esi,SIZEOF_JCOEF)]
- movdqa xmm3, XMMWORD [XMMBLOCK(7,0,esi,SIZEOF_JCOEF)]
- pmullw xmm2, XMMWORD [XMMBLOCK(5,0,edx,SIZEOF_ISLOW_MULT_TYPE)]
- pmullw xmm3, XMMWORD [XMMBLOCK(7,0,edx,SIZEOF_ISLOW_MULT_TYPE)]
-
- ; xmm0=(10 11 ** 13 ** 15 ** 17), xmm1=(30 31 ** 33 ** 35 ** 37)
- ; xmm2=(50 51 ** 53 ** 55 ** 57), xmm3=(70 71 ** 73 ** 75 ** 77)
-
- pcmpeqd xmm7,xmm7
- pslld xmm7,WORD_BIT ; xmm7={0x0000 0xFFFF 0x0000 0xFFFF ..}
-
- movdqa xmm4,xmm0 ; xmm4=(10 11 ** 13 ** 15 ** 17)
- movdqa xmm5,xmm2 ; xmm5=(50 51 ** 53 ** 55 ** 57)
- punpcklwd xmm4,xmm1 ; xmm4=(10 30 11 31 ** ** 13 33)
- punpcklwd xmm5,xmm3 ; xmm5=(50 70 51 71 ** ** 53 73)
- pmaddwd xmm4,[GOTOFF(ebx,PW_F362_MF127)]
- pmaddwd xmm5,[GOTOFF(ebx,PW_F085_MF072)]
-
- psrld xmm0,WORD_BIT ; xmm0=(11 -- 13 -- 15 -- 17 --)
- pand xmm1,xmm7 ; xmm1=(-- 31 -- 33 -- 35 -- 37)
- psrld xmm2,WORD_BIT ; xmm2=(51 -- 53 -- 55 -- 57 --)
- pand xmm3,xmm7 ; xmm3=(-- 71 -- 73 -- 75 -- 77)
- por xmm0,xmm1 ; xmm0=(11 31 13 33 15 35 17 37)
- por xmm2,xmm3 ; xmm2=(51 71 53 73 55 75 57 77)
- pmaddwd xmm0,[GOTOFF(ebx,PW_F362_MF127)]
- pmaddwd xmm2,[GOTOFF(ebx,PW_F085_MF072)]
-
- paddd xmm4,xmm5 ; xmm4=tmp0[col0 col1 **** col3]
- paddd xmm0,xmm2 ; xmm0=tmp0[col1 col3 col5 col7]
-
- ; -- Even part
-
- movdqa xmm6, XMMWORD [XMMBLOCK(0,0,esi,SIZEOF_JCOEF)]
- pmullw xmm6, XMMWORD [XMMBLOCK(0,0,edx,SIZEOF_ISLOW_MULT_TYPE)]
-
- ; xmm6=(00 01 ** 03 ** 05 ** 07)
-
- movdqa xmm1,xmm6 ; xmm1=(00 01 ** 03 ** 05 ** 07)
- pslld xmm6,WORD_BIT ; xmm6=(-- 00 -- ** -- ** -- **)
- pand xmm1,xmm7 ; xmm1=(-- 01 -- 03 -- 05 -- 07)
- psrad xmm6,(WORD_BIT-CONST_BITS-2) ; xmm6=tmp10[col0 **** **** ****]
- psrad xmm1,(WORD_BIT-CONST_BITS-2) ; xmm1=tmp10[col1 col3 col5 col7]
-
- ; -- Final output stage
-
- movdqa xmm3,xmm6
- movdqa xmm5,xmm1
- paddd xmm6,xmm4 ; xmm6=data0[col0 **** **** ****]=(A0 ** ** **)
- paddd xmm1,xmm0 ; xmm1=data0[col1 col3 col5 col7]=(A1 A3 A5 A7)
- psubd xmm3,xmm4 ; xmm3=data1[col0 **** **** ****]=(B0 ** ** **)
- psubd xmm5,xmm0 ; xmm5=data1[col1 col3 col5 col7]=(B1 B3 B5 B7)
-
- movdqa xmm2,[GOTOFF(ebx,PD_DESCALE_P1_2)] ; xmm2=[PD_DESCALE_P1_2]
-
- punpckldq xmm6,xmm3 ; xmm6=(A0 B0 ** **)
-
- movdqa xmm7,xmm1
- punpcklqdq xmm1,xmm5 ; xmm1=(A1 A3 B1 B3)
- punpckhqdq xmm7,xmm5 ; xmm7=(A5 A7 B5 B7)
-
- paddd xmm6,xmm2
- psrad xmm6,DESCALE_P1_2
-
- paddd xmm1,xmm2
- paddd xmm7,xmm2
- psrad xmm1,DESCALE_P1_2
- psrad xmm7,DESCALE_P1_2
-
- ; -- Prefetch the next coefficient block
-
- prefetchnta [esi + DCTSIZE2*SIZEOF_JCOEF + 0*32]
- prefetchnta [esi + DCTSIZE2*SIZEOF_JCOEF + 1*32]
- prefetchnta [esi + DCTSIZE2*SIZEOF_JCOEF + 2*32]
- prefetchnta [esi + DCTSIZE2*SIZEOF_JCOEF + 3*32]
-
- ; ---- Pass 2: process rows, store into output array.
-
- mov edi, JSAMPARRAY [output_buf(ebp)] ; (JSAMPROW *)
- mov eax, JDIMENSION [output_col(ebp)]
-
- ; | input:| result:|
- ; | A0 B0 | |
- ; | A1 B1 | C0 C1 |
- ; | A3 B3 | D0 D1 |
- ; | A5 B5 | |
- ; | A7 B7 | |
-
- ; -- Odd part
-
- packssdw xmm1,xmm1 ; xmm1=(A1 A3 B1 B3 A1 A3 B1 B3)
- packssdw xmm7,xmm7 ; xmm7=(A5 A7 B5 B7 A5 A7 B5 B7)
- pmaddwd xmm1,[GOTOFF(ebx,PW_F362_MF127)]
- pmaddwd xmm7,[GOTOFF(ebx,PW_F085_MF072)]
-
- paddd xmm1,xmm7 ; xmm1=tmp0[row0 row1 row0 row1]
-
- ; -- Even part
-
- pslld xmm6,(CONST_BITS+2) ; xmm6=tmp10[row0 row1 **** ****]
-
- ; -- Final output stage
-
- movdqa xmm4,xmm6
- paddd xmm6,xmm1 ; xmm6=data0[row0 row1 **** ****]=(C0 C1 ** **)
- psubd xmm4,xmm1 ; xmm4=data1[row0 row1 **** ****]=(D0 D1 ** **)
-
- punpckldq xmm6,xmm4 ; xmm6=(C0 D0 C1 D1)
-
- paddd xmm6,[GOTOFF(ebx,PD_DESCALE_P2_2)]
- psrad xmm6,DESCALE_P2_2
-
- packssdw xmm6,xmm6 ; xmm6=(C0 D0 C1 D1 C0 D0 C1 D1)
- packsswb xmm6,xmm6 ; xmm6=(C0 D0 C1 D1 C0 D0 C1 D1 ..)
- paddb xmm6,[GOTOFF(ebx,PB_CENTERJSAMP)]
-
- pextrw ebx,xmm6,0x00 ; ebx=(C0 D0 -- --)
- pextrw ecx,xmm6,0x01 ; ecx=(C1 D1 -- --)
-
- mov edx, JSAMPROW [edi+0*SIZEOF_JSAMPROW]
- mov esi, JSAMPROW [edi+1*SIZEOF_JSAMPROW]
- mov WORD [edx+eax*SIZEOF_JSAMPLE], bx
- mov WORD [esi+eax*SIZEOF_JSAMPLE], cx
-
- pop edi
- pop esi
-; pop edx ; need not be preserved
-; pop ecx ; need not be preserved
- pop ebx
- pop ebp
- ret
-
-; For some reason, the OS X linker does not honor the request to align the
-; segment unless we do this.
- align 16
diff --git a/simd/jquantf-sse2-64.asm b/simd/jquantf-sse2-64.asm
deleted file mode 100644
index ef5c1f9..0000000
--- a/simd/jquantf-sse2-64.asm
+++ /dev/null
@@ -1,157 +0,0 @@
-;
-; jquantf.asm - sample data conversion and quantization (64-bit SSE & SSE2)
-;
-; Copyright 2009 Pierre Ossman <ossman@cendio.se> for Cendio AB
-; Copyright (C) 2009, D. R. Commander.
-;
-; Based on the x86 SIMD extension for IJG JPEG library
-; Copyright (C) 1999-2006, MIYASAKA Masaru.
-; For conditions of distribution and use, see copyright notice in jsimdext.inc
-;
-; This file should be assembled with NASM (Netwide Assembler),
-; can *not* be assembled with Microsoft's MASM or any compatible
-; assembler (including Borland's Turbo Assembler).
-; NASM is available from http://nasm.sourceforge.net/ or
-; http://sourceforge.net/project/showfiles.php?group_id=6208
-;
-; [TAB8]
-
-%include "jsimdext.inc"
-%include "jdct.inc"
-
-; --------------------------------------------------------------------------
- SECTION SEG_TEXT
- BITS 64
-;
-; Load data into workspace, applying unsigned->signed conversion
-;
-; GLOBAL(void)
-; jsimd_convsamp_float_sse2 (JSAMPARRAY sample_data, JDIMENSION start_col,
-; FAST_FLOAT *workspace);
-;
-
-; r10 = JSAMPARRAY sample_data
-; r11 = JDIMENSION start_col
-; r12 = FAST_FLOAT *workspace
-
- align 16
- global EXTN(jsimd_convsamp_float_sse2)
-
-EXTN(jsimd_convsamp_float_sse2):
- push rbp
- mov rax,rsp
- mov rbp,rsp
- collect_args
- push rbx
-
- pcmpeqw xmm7,xmm7
- psllw xmm7,7
- packsswb xmm7,xmm7 ; xmm7 = PB_CENTERJSAMPLE (0x808080..)
-
- mov rsi, r10
- mov eax, r11d
- mov rdi, r12
- mov rcx, DCTSIZE/2
-.convloop:
- mov rbx, JSAMPROW [rsi+0*SIZEOF_JSAMPROW] ; (JSAMPLE *)
- mov rdx, JSAMPROW [rsi+1*SIZEOF_JSAMPROW] ; (JSAMPLE *)
-
- movq xmm0, XMM_MMWORD [rbx+rax*SIZEOF_JSAMPLE]
- movq xmm1, XMM_MMWORD [rdx+rax*SIZEOF_JSAMPLE]
-
- psubb xmm0,xmm7 ; xmm0=(01234567)
- psubb xmm1,xmm7 ; xmm1=(89ABCDEF)
-
- punpcklbw xmm0,xmm0 ; xmm0=(*0*1*2*3*4*5*6*7)
- punpcklbw xmm1,xmm1 ; xmm1=(*8*9*A*B*C*D*E*F)
-
- punpcklwd xmm2,xmm0 ; xmm2=(***0***1***2***3)
- punpckhwd xmm0,xmm0 ; xmm0=(***4***5***6***7)
- punpcklwd xmm3,xmm1 ; xmm3=(***8***9***A***B)
- punpckhwd xmm1,xmm1 ; xmm1=(***C***D***E***F)
-
- psrad xmm2,(DWORD_BIT-BYTE_BIT) ; xmm2=(0123)
- psrad xmm0,(DWORD_BIT-BYTE_BIT) ; xmm0=(4567)
- cvtdq2ps xmm2,xmm2 ; xmm2=(0123)
- cvtdq2ps xmm0,xmm0 ; xmm0=(4567)
- psrad xmm3,(DWORD_BIT-BYTE_BIT) ; xmm3=(89AB)
- psrad xmm1,(DWORD_BIT-BYTE_BIT) ; xmm1=(CDEF)
- cvtdq2ps xmm3,xmm3 ; xmm3=(89AB)
- cvtdq2ps xmm1,xmm1 ; xmm1=(CDEF)
-
- movaps XMMWORD [XMMBLOCK(0,0,rdi,SIZEOF_FAST_FLOAT)], xmm2
- movaps XMMWORD [XMMBLOCK(0,1,rdi,SIZEOF_FAST_FLOAT)], xmm0
- movaps XMMWORD [XMMBLOCK(1,0,rdi,SIZEOF_FAST_FLOAT)], xmm3
- movaps XMMWORD [XMMBLOCK(1,1,rdi,SIZEOF_FAST_FLOAT)], xmm1
-
- add rsi, byte 2*SIZEOF_JSAMPROW
- add rdi, byte 2*DCTSIZE*SIZEOF_FAST_FLOAT
- dec rcx
- jnz short .convloop
-
- pop rbx
- uncollect_args
- pop rbp
- ret
-
-
-; --------------------------------------------------------------------------
-;
-; Quantize/descale the coefficients, and store into coef_block
-;
-; GLOBAL(void)
-; jsimd_quantize_float_sse2 (JCOEFPTR coef_block, FAST_FLOAT *divisors,
-; FAST_FLOAT *workspace);
-;
-
-; r10 = JCOEFPTR coef_block
-; r11 = FAST_FLOAT *divisors
-; r12 = FAST_FLOAT *workspace
-
- align 16
- global EXTN(jsimd_quantize_float_sse2)
-
-EXTN(jsimd_quantize_float_sse2):
- push rbp
- mov rax,rsp
- mov rbp,rsp
- collect_args
-
- mov rsi, r12
- mov rdx, r11
- mov rdi, r10
- mov rax, DCTSIZE2/16
-.quantloop:
- movaps xmm0, XMMWORD [XMMBLOCK(0,0,rsi,SIZEOF_FAST_FLOAT)]
- movaps xmm1, XMMWORD [XMMBLOCK(0,1,rsi,SIZEOF_FAST_FLOAT)]
- mulps xmm0, XMMWORD [XMMBLOCK(0,0,rdx,SIZEOF_FAST_FLOAT)]
- mulps xmm1, XMMWORD [XMMBLOCK(0,1,rdx,SIZEOF_FAST_FLOAT)]
- movaps xmm2, XMMWORD [XMMBLOCK(1,0,rsi,SIZEOF_FAST_FLOAT)]
- movaps xmm3, XMMWORD [XMMBLOCK(1,1,rsi,SIZEOF_FAST_FLOAT)]
- mulps xmm2, XMMWORD [XMMBLOCK(1,0,rdx,SIZEOF_FAST_FLOAT)]
- mulps xmm3, XMMWORD [XMMBLOCK(1,1,rdx,SIZEOF_FAST_FLOAT)]
-
- cvtps2dq xmm0,xmm0
- cvtps2dq xmm1,xmm1
- cvtps2dq xmm2,xmm2
- cvtps2dq xmm3,xmm3
-
- packssdw xmm0,xmm1
- packssdw xmm2,xmm3
-
- movdqa XMMWORD [XMMBLOCK(0,0,rdi,SIZEOF_JCOEF)], xmm0
- movdqa XMMWORD [XMMBLOCK(1,0,rdi,SIZEOF_JCOEF)], xmm2
-
- add rsi, byte 16*SIZEOF_FAST_FLOAT
- add rdx, byte 16*SIZEOF_FAST_FLOAT
- add rdi, byte 16*SIZEOF_JCOEF
- dec rax
- jnz short .quantloop
-
- uncollect_args
- pop rbp
- ret
-
-; For some reason, the OS X linker does not honor the request to align the
-; segment unless we do this.
- align 16
diff --git a/simd/jquantf-sse2.asm b/simd/jquantf-sse2.asm
deleted file mode 100644
index 1cbc267..0000000
--- a/simd/jquantf-sse2.asm
+++ /dev/null
@@ -1,170 +0,0 @@
-;
-; jquantf.asm - sample data conversion and quantization (SSE & SSE2)
-;
-; Copyright 2009 Pierre Ossman <ossman@cendio.se> for Cendio AB
-;
-; Based on the x86 SIMD extension for IJG JPEG library
-; Copyright (C) 1999-2006, MIYASAKA Masaru.
-; For conditions of distribution and use, see copyright notice in jsimdext.inc
-;
-; This file should be assembled with NASM (Netwide Assembler),
-; can *not* be assembled with Microsoft's MASM or any compatible
-; assembler (including Borland's Turbo Assembler).
-; NASM is available from http://nasm.sourceforge.net/ or
-; http://sourceforge.net/project/showfiles.php?group_id=6208
-;
-; [TAB8]
-
-%include "jsimdext.inc"
-%include "jdct.inc"
-
-; --------------------------------------------------------------------------
- SECTION SEG_TEXT
- BITS 32
-;
-; Load data into workspace, applying unsigned->signed conversion
-;
-; GLOBAL(void)
-; jsimd_convsamp_float_sse2 (JSAMPARRAY sample_data, JDIMENSION start_col,
-; FAST_FLOAT *workspace);
-;
-
-%define sample_data ebp+8 ; JSAMPARRAY sample_data
-%define start_col ebp+12 ; JDIMENSION start_col
-%define workspace ebp+16 ; FAST_FLOAT *workspace
-
- align 16
- global EXTN(jsimd_convsamp_float_sse2)
-
-EXTN(jsimd_convsamp_float_sse2):
- push ebp
- mov ebp,esp
- push ebx
-; push ecx ; need not be preserved
-; push edx ; need not be preserved
- push esi
- push edi
-
- pcmpeqw xmm7,xmm7
- psllw xmm7,7
- packsswb xmm7,xmm7 ; xmm7 = PB_CENTERJSAMPLE (0x808080..)
-
- mov esi, JSAMPARRAY [sample_data] ; (JSAMPROW *)
- mov eax, JDIMENSION [start_col]
- mov edi, POINTER [workspace] ; (DCTELEM *)
- mov ecx, DCTSIZE/2
- alignx 16,7
-.convloop:
- mov ebx, JSAMPROW [esi+0*SIZEOF_JSAMPROW] ; (JSAMPLE *)
- mov edx, JSAMPROW [esi+1*SIZEOF_JSAMPROW] ; (JSAMPLE *)
-
- movq xmm0, XMM_MMWORD [ebx+eax*SIZEOF_JSAMPLE]
- movq xmm1, XMM_MMWORD [edx+eax*SIZEOF_JSAMPLE]
-
- psubb xmm0,xmm7 ; xmm0=(01234567)
- psubb xmm1,xmm7 ; xmm1=(89ABCDEF)
-
- punpcklbw xmm0,xmm0 ; xmm0=(*0*1*2*3*4*5*6*7)
- punpcklbw xmm1,xmm1 ; xmm1=(*8*9*A*B*C*D*E*F)
-
- punpcklwd xmm2,xmm0 ; xmm2=(***0***1***2***3)
- punpckhwd xmm0,xmm0 ; xmm0=(***4***5***6***7)
- punpcklwd xmm3,xmm1 ; xmm3=(***8***9***A***B)
- punpckhwd xmm1,xmm1 ; xmm1=(***C***D***E***F)
-
- psrad xmm2,(DWORD_BIT-BYTE_BIT) ; xmm2=(0123)
- psrad xmm0,(DWORD_BIT-BYTE_BIT) ; xmm0=(4567)
- cvtdq2ps xmm2,xmm2 ; xmm2=(0123)
- cvtdq2ps xmm0,xmm0 ; xmm0=(4567)
- psrad xmm3,(DWORD_BIT-BYTE_BIT) ; xmm3=(89AB)
- psrad xmm1,(DWORD_BIT-BYTE_BIT) ; xmm1=(CDEF)
- cvtdq2ps xmm3,xmm3 ; xmm3=(89AB)
- cvtdq2ps xmm1,xmm1 ; xmm1=(CDEF)
-
- movaps XMMWORD [XMMBLOCK(0,0,edi,SIZEOF_FAST_FLOAT)], xmm2
- movaps XMMWORD [XMMBLOCK(0,1,edi,SIZEOF_FAST_FLOAT)], xmm0
- movaps XMMWORD [XMMBLOCK(1,0,edi,SIZEOF_FAST_FLOAT)], xmm3
- movaps XMMWORD [XMMBLOCK(1,1,edi,SIZEOF_FAST_FLOAT)], xmm1
-
- add esi, byte 2*SIZEOF_JSAMPROW
- add edi, byte 2*DCTSIZE*SIZEOF_FAST_FLOAT
- dec ecx
- jnz short .convloop
-
- pop edi
- pop esi
-; pop edx ; need not be preserved
-; pop ecx ; need not be preserved
- pop ebx
- pop ebp
- ret
-
-
-; --------------------------------------------------------------------------
-;
-; Quantize/descale the coefficients, and store into coef_block
-;
-; GLOBAL(void)
-; jsimd_quantize_float_sse2 (JCOEFPTR coef_block, FAST_FLOAT *divisors,
-; FAST_FLOAT *workspace);
-;
-
-%define coef_block ebp+8 ; JCOEFPTR coef_block
-%define divisors ebp+12 ; FAST_FLOAT *divisors
-%define workspace ebp+16 ; FAST_FLOAT *workspace
-
- align 16
- global EXTN(jsimd_quantize_float_sse2)
-
-EXTN(jsimd_quantize_float_sse2):
- push ebp
- mov ebp,esp
-; push ebx ; unused
-; push ecx ; unused
-; push edx ; need not be preserved
- push esi
- push edi
-
- mov esi, POINTER [workspace]
- mov edx, POINTER [divisors]
- mov edi, JCOEFPTR [coef_block]
- mov eax, DCTSIZE2/16
- alignx 16,7
-.quantloop:
- movaps xmm0, XMMWORD [XMMBLOCK(0,0,esi,SIZEOF_FAST_FLOAT)]
- movaps xmm1, XMMWORD [XMMBLOCK(0,1,esi,SIZEOF_FAST_FLOAT)]
- mulps xmm0, XMMWORD [XMMBLOCK(0,0,edx,SIZEOF_FAST_FLOAT)]
- mulps xmm1, XMMWORD [XMMBLOCK(0,1,edx,SIZEOF_FAST_FLOAT)]
- movaps xmm2, XMMWORD [XMMBLOCK(1,0,esi,SIZEOF_FAST_FLOAT)]
- movaps xmm3, XMMWORD [XMMBLOCK(1,1,esi,SIZEOF_FAST_FLOAT)]
- mulps xmm2, XMMWORD [XMMBLOCK(1,0,edx,SIZEOF_FAST_FLOAT)]
- mulps xmm3, XMMWORD [XMMBLOCK(1,1,edx,SIZEOF_FAST_FLOAT)]
-
- cvtps2dq xmm0,xmm0
- cvtps2dq xmm1,xmm1
- cvtps2dq xmm2,xmm2
- cvtps2dq xmm3,xmm3
-
- packssdw xmm0,xmm1
- packssdw xmm2,xmm3
-
- movdqa XMMWORD [XMMBLOCK(0,0,edi,SIZEOF_JCOEF)], xmm0
- movdqa XMMWORD [XMMBLOCK(1,0,edi,SIZEOF_JCOEF)], xmm2
-
- add esi, byte 16*SIZEOF_FAST_FLOAT
- add edx, byte 16*SIZEOF_FAST_FLOAT
- add edi, byte 16*SIZEOF_JCOEF
- dec eax
- jnz short .quantloop
-
- pop edi
- pop esi
-; pop edx ; need not be preserved
-; pop ecx ; unused
-; pop ebx ; unused
- pop ebp
- ret
-
-; For some reason, the OS X linker does not honor the request to align the
-; segment unless we do this.
- align 16
diff --git a/simd/jquanti-sse2-64.asm b/simd/jquanti-sse2-64.asm
deleted file mode 100644
index 66c4e51..0000000
--- a/simd/jquanti-sse2-64.asm
+++ /dev/null
@@ -1,186 +0,0 @@
-;
-; jquanti.asm - sample data conversion and quantization (64-bit SSE2)
-;
-; Copyright 2009 Pierre Ossman <ossman@cendio.se> for Cendio AB
-; Copyright (C) 2009, D. R. Commander.
-;
-; Based on the x86 SIMD extension for IJG JPEG library
-; Copyright (C) 1999-2006, MIYASAKA Masaru.
-; For conditions of distribution and use, see copyright notice in jsimdext.inc
-;
-; This file should be assembled with NASM (Netwide Assembler),
-; can *not* be assembled with Microsoft's MASM or any compatible
-; assembler (including Borland's Turbo Assembler).
-; NASM is available from http://nasm.sourceforge.net/ or
-; http://sourceforge.net/project/showfiles.php?group_id=6208
-;
-; [TAB8]
-
-%include "jsimdext.inc"
-%include "jdct.inc"
-
-; --------------------------------------------------------------------------
- SECTION SEG_TEXT
- BITS 64
-;
-; Load data into workspace, applying unsigned->signed conversion
-;
-; GLOBAL(void)
-; jsimd_convsamp_sse2 (JSAMPARRAY sample_data, JDIMENSION start_col,
-; DCTELEM *workspace);
-;
-
-; r10 = JSAMPARRAY sample_data
-; r11 = JDIMENSION start_col
-; r12 = DCTELEM *workspace
-
- align 16
- global EXTN(jsimd_convsamp_sse2)
-
-EXTN(jsimd_convsamp_sse2):
- push rbp
- mov rax,rsp
- mov rbp,rsp
- collect_args
- push rbx
-
- pxor xmm6,xmm6 ; xmm6=(all 0's)
- pcmpeqw xmm7,xmm7
- psllw xmm7,7 ; xmm7={0xFF80 0xFF80 0xFF80 0xFF80 ..}
-
- mov rsi, r10
- mov eax, r11d
- mov rdi, r12
- mov rcx, DCTSIZE/4
-.convloop:
- mov rbx, JSAMPROW [rsi+0*SIZEOF_JSAMPROW] ; (JSAMPLE *)
- mov rdx, JSAMPROW [rsi+1*SIZEOF_JSAMPROW] ; (JSAMPLE *)
-
- movq xmm0, XMM_MMWORD [rbx+rax*SIZEOF_JSAMPLE] ; xmm0=(01234567)
- movq xmm1, XMM_MMWORD [rdx+rax*SIZEOF_JSAMPLE] ; xmm1=(89ABCDEF)
-
- mov rbx, JSAMPROW [rsi+2*SIZEOF_JSAMPROW] ; (JSAMPLE *)
- mov rdx, JSAMPROW [rsi+3*SIZEOF_JSAMPROW] ; (JSAMPLE *)
-
- movq xmm2, XMM_MMWORD [rbx+rax*SIZEOF_JSAMPLE] ; xmm2=(GHIJKLMN)
- movq xmm3, XMM_MMWORD [rdx+rax*SIZEOF_JSAMPLE] ; xmm3=(OPQRSTUV)
-
- punpcklbw xmm0,xmm6 ; xmm0=(01234567)
- punpcklbw xmm1,xmm6 ; xmm1=(89ABCDEF)
- paddw xmm0,xmm7
- paddw xmm1,xmm7
- punpcklbw xmm2,xmm6 ; xmm2=(GHIJKLMN)
- punpcklbw xmm3,xmm6 ; xmm3=(OPQRSTUV)
- paddw xmm2,xmm7
- paddw xmm3,xmm7
-
- movdqa XMMWORD [XMMBLOCK(0,0,rdi,SIZEOF_DCTELEM)], xmm0
- movdqa XMMWORD [XMMBLOCK(1,0,rdi,SIZEOF_DCTELEM)], xmm1
- movdqa XMMWORD [XMMBLOCK(2,0,rdi,SIZEOF_DCTELEM)], xmm2
- movdqa XMMWORD [XMMBLOCK(3,0,rdi,SIZEOF_DCTELEM)], xmm3
-
- add rsi, byte 4*SIZEOF_JSAMPROW
- add rdi, byte 4*DCTSIZE*SIZEOF_DCTELEM
- dec rcx
- jnz short .convloop
-
- pop rbx
- uncollect_args
- pop rbp
- ret
-
-; --------------------------------------------------------------------------
-;
-; Quantize/descale the coefficients, and store into coef_block
-;
-; This implementation is based on an algorithm described in
-; "How to optimize for the Pentium family of microprocessors"
-; (http://www.agner.org/assem/).
-;
-; GLOBAL(void)
-; jsimd_quantize_sse2 (JCOEFPTR coef_block, DCTELEM *divisors,
-; DCTELEM *workspace);
-;
-
-%define RECIPROCAL(m,n,b) XMMBLOCK(DCTSIZE*0+(m),(n),(b),SIZEOF_DCTELEM)
-%define CORRECTION(m,n,b) XMMBLOCK(DCTSIZE*1+(m),(n),(b),SIZEOF_DCTELEM)
-%define SCALE(m,n,b) XMMBLOCK(DCTSIZE*2+(m),(n),(b),SIZEOF_DCTELEM)
-
-; r10 = JCOEFPTR coef_block
-; r11 = DCTELEM *divisors
-; r12 = DCTELEM *workspace
-
- align 16
- global EXTN(jsimd_quantize_sse2)
-
-EXTN(jsimd_quantize_sse2):
- push rbp
- mov rax,rsp
- mov rbp,rsp
- collect_args
-
- mov rsi, r12
- mov rdx, r11
- mov rdi, r10
- mov rax, DCTSIZE2/32
-.quantloop:
- movdqa xmm4, XMMWORD [XMMBLOCK(0,0,rsi,SIZEOF_DCTELEM)]
- movdqa xmm5, XMMWORD [XMMBLOCK(1,0,rsi,SIZEOF_DCTELEM)]
- movdqa xmm6, XMMWORD [XMMBLOCK(2,0,rsi,SIZEOF_DCTELEM)]
- movdqa xmm7, XMMWORD [XMMBLOCK(3,0,rsi,SIZEOF_DCTELEM)]
- movdqa xmm0,xmm4
- movdqa xmm1,xmm5
- movdqa xmm2,xmm6
- movdqa xmm3,xmm7
- psraw xmm4,(WORD_BIT-1)
- psraw xmm5,(WORD_BIT-1)
- psraw xmm6,(WORD_BIT-1)
- psraw xmm7,(WORD_BIT-1)
- pxor xmm0,xmm4
- pxor xmm1,xmm5
- pxor xmm2,xmm6
- pxor xmm3,xmm7
- psubw xmm0,xmm4 ; if (xmm0 < 0) xmm0 = -xmm0;
- psubw xmm1,xmm5 ; if (xmm1 < 0) xmm1 = -xmm1;
- psubw xmm2,xmm6 ; if (xmm2 < 0) xmm2 = -xmm2;
- psubw xmm3,xmm7 ; if (xmm3 < 0) xmm3 = -xmm3;
-
- paddw xmm0, XMMWORD [CORRECTION(0,0,rdx)] ; correction + roundfactor
- paddw xmm1, XMMWORD [CORRECTION(1,0,rdx)]
- paddw xmm2, XMMWORD [CORRECTION(2,0,rdx)]
- paddw xmm3, XMMWORD [CORRECTION(3,0,rdx)]
- pmulhuw xmm0, XMMWORD [RECIPROCAL(0,0,rdx)] ; reciprocal
- pmulhuw xmm1, XMMWORD [RECIPROCAL(1,0,rdx)]
- pmulhuw xmm2, XMMWORD [RECIPROCAL(2,0,rdx)]
- pmulhuw xmm3, XMMWORD [RECIPROCAL(3,0,rdx)]
- pmulhuw xmm0, XMMWORD [SCALE(0,0,rdx)] ; scale
- pmulhuw xmm1, XMMWORD [SCALE(1,0,rdx)]
- pmulhuw xmm2, XMMWORD [SCALE(2,0,rdx)]
- pmulhuw xmm3, XMMWORD [SCALE(3,0,rdx)]
-
- pxor xmm0,xmm4
- pxor xmm1,xmm5
- pxor xmm2,xmm6
- pxor xmm3,xmm7
- psubw xmm0,xmm4
- psubw xmm1,xmm5
- psubw xmm2,xmm6
- psubw xmm3,xmm7
- movdqa XMMWORD [XMMBLOCK(0,0,rdi,SIZEOF_DCTELEM)], xmm0
- movdqa XMMWORD [XMMBLOCK(1,0,rdi,SIZEOF_DCTELEM)], xmm1
- movdqa XMMWORD [XMMBLOCK(2,0,rdi,SIZEOF_DCTELEM)], xmm2
- movdqa XMMWORD [XMMBLOCK(3,0,rdi,SIZEOF_DCTELEM)], xmm3
-
- add rsi, byte 32*SIZEOF_DCTELEM
- add rdx, byte 32*SIZEOF_DCTELEM
- add rdi, byte 32*SIZEOF_JCOEF
- dec rax
- jnz near .quantloop
-
- uncollect_args
- pop rbp
- ret
-
-; For some reason, the OS X linker does not honor the request to align the
-; segment unless we do this.
- align 16
diff --git a/simd/jquanti-sse2.asm b/simd/jquanti-sse2.asm
deleted file mode 100644
index aea8604..0000000
--- a/simd/jquanti-sse2.asm
+++ /dev/null
@@ -1,199 +0,0 @@
-;
-; jquanti.asm - sample data conversion and quantization (SSE2)
-;
-; Copyright 2009 Pierre Ossman <ossman@cendio.se> for Cendio AB
-;
-; Based on the x86 SIMD extension for IJG JPEG library
-; Copyright (C) 1999-2006, MIYASAKA Masaru.
-; For conditions of distribution and use, see copyright notice in jsimdext.inc
-;
-; This file should be assembled with NASM (Netwide Assembler),
-; can *not* be assembled with Microsoft's MASM or any compatible
-; assembler (including Borland's Turbo Assembler).
-; NASM is available from http://nasm.sourceforge.net/ or
-; http://sourceforge.net/project/showfiles.php?group_id=6208
-;
-; [TAB8]
-
-%include "jsimdext.inc"
-%include "jdct.inc"
-
-; --------------------------------------------------------------------------
- SECTION SEG_TEXT
- BITS 32
-;
-; Load data into workspace, applying unsigned->signed conversion
-;
-; GLOBAL(void)
-; jsimd_convsamp_sse2 (JSAMPARRAY sample_data, JDIMENSION start_col,
-; DCTELEM *workspace);
-;
-
-%define sample_data ebp+8 ; JSAMPARRAY sample_data
-%define start_col ebp+12 ; JDIMENSION start_col
-%define workspace ebp+16 ; DCTELEM *workspace
-
- align 16
- global EXTN(jsimd_convsamp_sse2)
-
-EXTN(jsimd_convsamp_sse2):
- push ebp
- mov ebp,esp
- push ebx
-; push ecx ; need not be preserved
-; push edx ; need not be preserved
- push esi
- push edi
-
- pxor xmm6,xmm6 ; xmm6=(all 0's)
- pcmpeqw xmm7,xmm7
- psllw xmm7,7 ; xmm7={0xFF80 0xFF80 0xFF80 0xFF80 ..}
-
- mov esi, JSAMPARRAY [sample_data] ; (JSAMPROW *)
- mov eax, JDIMENSION [start_col]
- mov edi, POINTER [workspace] ; (DCTELEM *)
- mov ecx, DCTSIZE/4
- alignx 16,7
-.convloop:
- mov ebx, JSAMPROW [esi+0*SIZEOF_JSAMPROW] ; (JSAMPLE *)
- mov edx, JSAMPROW [esi+1*SIZEOF_JSAMPROW] ; (JSAMPLE *)
-
- movq xmm0, XMM_MMWORD [ebx+eax*SIZEOF_JSAMPLE] ; xmm0=(01234567)
- movq xmm1, XMM_MMWORD [edx+eax*SIZEOF_JSAMPLE] ; xmm1=(89ABCDEF)
-
- mov ebx, JSAMPROW [esi+2*SIZEOF_JSAMPROW] ; (JSAMPLE *)
- mov edx, JSAMPROW [esi+3*SIZEOF_JSAMPROW] ; (JSAMPLE *)
-
- movq xmm2, XMM_MMWORD [ebx+eax*SIZEOF_JSAMPLE] ; xmm2=(GHIJKLMN)
- movq xmm3, XMM_MMWORD [edx+eax*SIZEOF_JSAMPLE] ; xmm3=(OPQRSTUV)
-
- punpcklbw xmm0,xmm6 ; xmm0=(01234567)
- punpcklbw xmm1,xmm6 ; xmm1=(89ABCDEF)
- paddw xmm0,xmm7
- paddw xmm1,xmm7
- punpcklbw xmm2,xmm6 ; xmm2=(GHIJKLMN)
- punpcklbw xmm3,xmm6 ; xmm3=(OPQRSTUV)
- paddw xmm2,xmm7
- paddw xmm3,xmm7
-
- movdqa XMMWORD [XMMBLOCK(0,0,edi,SIZEOF_DCTELEM)], xmm0
- movdqa XMMWORD [XMMBLOCK(1,0,edi,SIZEOF_DCTELEM)], xmm1
- movdqa XMMWORD [XMMBLOCK(2,0,edi,SIZEOF_DCTELEM)], xmm2
- movdqa XMMWORD [XMMBLOCK(3,0,edi,SIZEOF_DCTELEM)], xmm3
-
- add esi, byte 4*SIZEOF_JSAMPROW
- add edi, byte 4*DCTSIZE*SIZEOF_DCTELEM
- dec ecx
- jnz short .convloop
-
- pop edi
- pop esi
-; pop edx ; need not be preserved
-; pop ecx ; need not be preserved
- pop ebx
- pop ebp
- ret
-
-; --------------------------------------------------------------------------
-;
-; Quantize/descale the coefficients, and store into coef_block
-;
-; This implementation is based on an algorithm described in
-; "How to optimize for the Pentium family of microprocessors"
-; (http://www.agner.org/assem/).
-;
-; GLOBAL(void)
-; jsimd_quantize_sse2 (JCOEFPTR coef_block, DCTELEM *divisors,
-; DCTELEM *workspace);
-;
-
-%define RECIPROCAL(m,n,b) XMMBLOCK(DCTSIZE*0+(m),(n),(b),SIZEOF_DCTELEM)
-%define CORRECTION(m,n,b) XMMBLOCK(DCTSIZE*1+(m),(n),(b),SIZEOF_DCTELEM)
-%define SCALE(m,n,b) XMMBLOCK(DCTSIZE*2+(m),(n),(b),SIZEOF_DCTELEM)
-
-%define coef_block ebp+8 ; JCOEFPTR coef_block
-%define divisors ebp+12 ; DCTELEM *divisors
-%define workspace ebp+16 ; DCTELEM *workspace
-
- align 16
- global EXTN(jsimd_quantize_sse2)
-
-EXTN(jsimd_quantize_sse2):
- push ebp
- mov ebp,esp
-; push ebx ; unused
-; push ecx ; unused
-; push edx ; need not be preserved
- push esi
- push edi
-
- mov esi, POINTER [workspace]
- mov edx, POINTER [divisors]
- mov edi, JCOEFPTR [coef_block]
- mov eax, DCTSIZE2/32
- alignx 16,7
-.quantloop:
- movdqa xmm4, XMMWORD [XMMBLOCK(0,0,esi,SIZEOF_DCTELEM)]
- movdqa xmm5, XMMWORD [XMMBLOCK(1,0,esi,SIZEOF_DCTELEM)]
- movdqa xmm6, XMMWORD [XMMBLOCK(2,0,esi,SIZEOF_DCTELEM)]
- movdqa xmm7, XMMWORD [XMMBLOCK(3,0,esi,SIZEOF_DCTELEM)]
- movdqa xmm0,xmm4
- movdqa xmm1,xmm5
- movdqa xmm2,xmm6
- movdqa xmm3,xmm7
- psraw xmm4,(WORD_BIT-1)
- psraw xmm5,(WORD_BIT-1)
- psraw xmm6,(WORD_BIT-1)
- psraw xmm7,(WORD_BIT-1)
- pxor xmm0,xmm4
- pxor xmm1,xmm5
- pxor xmm2,xmm6
- pxor xmm3,xmm7
- psubw xmm0,xmm4 ; if (xmm0 < 0) xmm0 = -xmm0;
- psubw xmm1,xmm5 ; if (xmm1 < 0) xmm1 = -xmm1;
- psubw xmm2,xmm6 ; if (xmm2 < 0) xmm2 = -xmm2;
- psubw xmm3,xmm7 ; if (xmm3 < 0) xmm3 = -xmm3;
-
- paddw xmm0, XMMWORD [CORRECTION(0,0,edx)] ; correction + roundfactor
- paddw xmm1, XMMWORD [CORRECTION(1,0,edx)]
- paddw xmm2, XMMWORD [CORRECTION(2,0,edx)]
- paddw xmm3, XMMWORD [CORRECTION(3,0,edx)]
- pmulhuw xmm0, XMMWORD [RECIPROCAL(0,0,edx)] ; reciprocal
- pmulhuw xmm1, XMMWORD [RECIPROCAL(1,0,edx)]
- pmulhuw xmm2, XMMWORD [RECIPROCAL(2,0,edx)]
- pmulhuw xmm3, XMMWORD [RECIPROCAL(3,0,edx)]
- pmulhuw xmm0, XMMWORD [SCALE(0,0,edx)] ; scale
- pmulhuw xmm1, XMMWORD [SCALE(1,0,edx)]
- pmulhuw xmm2, XMMWORD [SCALE(2,0,edx)]
- pmulhuw xmm3, XMMWORD [SCALE(3,0,edx)]
-
- pxor xmm0,xmm4
- pxor xmm1,xmm5
- pxor xmm2,xmm6
- pxor xmm3,xmm7
- psubw xmm0,xmm4
- psubw xmm1,xmm5
- psubw xmm2,xmm6
- psubw xmm3,xmm7
- movdqa XMMWORD [XMMBLOCK(0,0,edi,SIZEOF_DCTELEM)], xmm0
- movdqa XMMWORD [XMMBLOCK(1,0,edi,SIZEOF_DCTELEM)], xmm1
- movdqa XMMWORD [XMMBLOCK(2,0,edi,SIZEOF_DCTELEM)], xmm2
- movdqa XMMWORD [XMMBLOCK(3,0,edi,SIZEOF_DCTELEM)], xmm3
-
- add esi, byte 32*SIZEOF_DCTELEM
- add edx, byte 32*SIZEOF_DCTELEM
- add edi, byte 32*SIZEOF_JCOEF
- dec eax
- jnz near .quantloop
-
- pop edi
- pop esi
-; pop edx ; need not be preserved
-; pop ecx ; unused
-; pop ebx ; unused
- pop ebp
- ret
-
-; For some reason, the OS X linker does not honor the request to align the
-; segment unless we do this.
- align 16
diff --git a/simd/jsimd.h b/simd/jsimd.h
index dc6ec43..f3ead87 100644
--- a/simd/jsimd.h
+++ b/simd/jsimd.h
@@ -23,6 +23,7 @@
#define JSIMD_ARM_NEON 0x10
#define JSIMD_MIPS_DSPR2 0x20
#define JSIMD_ALTIVEC 0x40
+#define JSIMD_AVX2 0x80
/* SIMD Ext: retrieve SIMD/CPU information */
EXTERN(unsigned int) jpeg_simd_cpu_support (void);
@@ -73,6 +74,29 @@
(JDIMENSION img_width, JSAMPARRAY input_buf, JSAMPIMAGE output_buf,
JDIMENSION output_row, int num_rows);
+extern const int jconst_rgb_ycc_convert_avx2[];
+EXTERN(void) jsimd_rgb_ycc_convert_avx2
+ (JDIMENSION img_width, JSAMPARRAY input_buf, JSAMPIMAGE output_buf,
+ JDIMENSION output_row, int num_rows);
+EXTERN(void) jsimd_extrgb_ycc_convert_avx2
+ (JDIMENSION img_width, JSAMPARRAY input_buf, JSAMPIMAGE output_buf,
+ JDIMENSION output_row, int num_rows);
+EXTERN(void) jsimd_extrgbx_ycc_convert_avx2
+ (JDIMENSION img_width, JSAMPARRAY input_buf, JSAMPIMAGE output_buf,
+ JDIMENSION output_row, int num_rows);
+EXTERN(void) jsimd_extbgr_ycc_convert_avx2
+ (JDIMENSION img_width, JSAMPARRAY input_buf, JSAMPIMAGE output_buf,
+ JDIMENSION output_row, int num_rows);
+EXTERN(void) jsimd_extbgrx_ycc_convert_avx2
+ (JDIMENSION img_width, JSAMPARRAY input_buf, JSAMPIMAGE output_buf,
+ JDIMENSION output_row, int num_rows);
+EXTERN(void) jsimd_extxbgr_ycc_convert_avx2
+ (JDIMENSION img_width, JSAMPARRAY input_buf, JSAMPIMAGE output_buf,
+ JDIMENSION output_row, int num_rows);
+EXTERN(void) jsimd_extxrgb_ycc_convert_avx2
+ (JDIMENSION img_width, JSAMPARRAY input_buf, JSAMPIMAGE output_buf,
+ JDIMENSION output_row, int num_rows);
+
EXTERN(void) jsimd_rgb_ycc_convert_neon
(JDIMENSION img_width, JSAMPARRAY input_buf, JSAMPIMAGE output_buf,
JDIMENSION output_row, int num_rows);
@@ -192,6 +216,29 @@
(JDIMENSION img_width, JSAMPARRAY input_buf, JSAMPIMAGE output_buf,
JDIMENSION output_row, int num_rows);
+extern const int jconst_rgb_gray_convert_avx2[];
+EXTERN(void) jsimd_rgb_gray_convert_avx2
+ (JDIMENSION img_width, JSAMPARRAY input_buf, JSAMPIMAGE output_buf,
+ JDIMENSION output_row, int num_rows);
+EXTERN(void) jsimd_extrgb_gray_convert_avx2
+ (JDIMENSION img_width, JSAMPARRAY input_buf, JSAMPIMAGE output_buf,
+ JDIMENSION output_row, int num_rows);
+EXTERN(void) jsimd_extrgbx_gray_convert_avx2
+ (JDIMENSION img_width, JSAMPARRAY input_buf, JSAMPIMAGE output_buf,
+ JDIMENSION output_row, int num_rows);
+EXTERN(void) jsimd_extbgr_gray_convert_avx2
+ (JDIMENSION img_width, JSAMPARRAY input_buf, JSAMPIMAGE output_buf,
+ JDIMENSION output_row, int num_rows);
+EXTERN(void) jsimd_extbgrx_gray_convert_avx2
+ (JDIMENSION img_width, JSAMPARRAY input_buf, JSAMPIMAGE output_buf,
+ JDIMENSION output_row, int num_rows);
+EXTERN(void) jsimd_extxbgr_gray_convert_avx2
+ (JDIMENSION img_width, JSAMPARRAY input_buf, JSAMPIMAGE output_buf,
+ JDIMENSION output_row, int num_rows);
+EXTERN(void) jsimd_extxrgb_gray_convert_avx2
+ (JDIMENSION img_width, JSAMPARRAY input_buf, JSAMPIMAGE output_buf,
+ JDIMENSION output_row, int num_rows);
+
EXTERN(void) jsimd_rgb_gray_convert_mips_dspr2
(JDIMENSION img_width, JSAMPARRAY input_buf, JSAMPIMAGE output_buf,
JDIMENSION output_row, int num_rows);
@@ -282,6 +329,29 @@
(JDIMENSION out_width, JSAMPIMAGE input_buf, JDIMENSION input_row,
JSAMPARRAY output_buf, int num_rows);
+extern const int jconst_ycc_rgb_convert_avx2[];
+EXTERN(void) jsimd_ycc_rgb_convert_avx2
+ (JDIMENSION out_width, JSAMPIMAGE input_buf, JDIMENSION input_row,
+ JSAMPARRAY output_buf, int num_rows);
+EXTERN(void) jsimd_ycc_extrgb_convert_avx2
+ (JDIMENSION out_width, JSAMPIMAGE input_buf, JDIMENSION input_row,
+ JSAMPARRAY output_buf, int num_rows);
+EXTERN(void) jsimd_ycc_extrgbx_convert_avx2
+ (JDIMENSION out_width, JSAMPIMAGE input_buf, JDIMENSION input_row,
+ JSAMPARRAY output_buf, int num_rows);
+EXTERN(void) jsimd_ycc_extbgr_convert_avx2
+ (JDIMENSION out_width, JSAMPIMAGE input_buf, JDIMENSION input_row,
+ JSAMPARRAY output_buf, int num_rows);
+EXTERN(void) jsimd_ycc_extbgrx_convert_avx2
+ (JDIMENSION out_width, JSAMPIMAGE input_buf, JDIMENSION input_row,
+ JSAMPARRAY output_buf, int num_rows);
+EXTERN(void) jsimd_ycc_extxbgr_convert_avx2
+ (JDIMENSION out_width, JSAMPIMAGE input_buf, JDIMENSION input_row,
+ JSAMPARRAY output_buf, int num_rows);
+EXTERN(void) jsimd_ycc_extxrgb_convert_avx2
+ (JDIMENSION out_width, JSAMPIMAGE input_buf, JDIMENSION input_row,
+ JSAMPARRAY output_buf, int num_rows);
+
EXTERN(void) jsimd_ycc_rgb_convert_neon
(JDIMENSION out_width, JSAMPIMAGE input_buf, JDIMENSION input_row,
JSAMPARRAY output_buf, int num_rows);
@@ -374,6 +444,11 @@
JDIMENSION v_samp_factor, JDIMENSION width_blocks,
JSAMPARRAY input_data, JSAMPARRAY output_data);
+EXTERN(void) jsimd_h2v1_downsample_avx2
+ (JDIMENSION image_width, int max_v_samp_factor,
+ JDIMENSION v_samp_factor, JDIMENSION width_blocks,
+ JSAMPARRAY input_data, JSAMPARRAY output_data);
+
EXTERN(void) jsimd_h2v1_downsample_neon
(JDIMENSION image_width, int max_v_samp_factor,
JDIMENSION v_samp_factor, JDIMENSION width_blocks,
@@ -400,6 +475,11 @@
JDIMENSION v_samp_factor, JDIMENSION width_blocks,
JSAMPARRAY input_data, JSAMPARRAY output_data);
+EXTERN(void) jsimd_h2v2_downsample_avx2
+ (JDIMENSION image_width, int max_v_samp_factor,
+ JDIMENSION v_samp_factor, JDIMENSION width_blocks,
+ JSAMPARRAY input_data, JSAMPARRAY output_data);
+
EXTERN(void) jsimd_h2v2_downsample_neon
(JDIMENSION image_width, int max_v_samp_factor,
JDIMENSION v_samp_factor, JDIMENSION width_blocks,
@@ -438,6 +518,13 @@
(int max_v_samp_factor, JDIMENSION output_width, JSAMPARRAY input_data,
JSAMPARRAY *output_data_ptr);
+EXTERN(void) jsimd_h2v1_upsample_avx2
+ (int max_v_samp_factor, JDIMENSION output_width, JSAMPARRAY input_data,
+ JSAMPARRAY *output_data_ptr);
+EXTERN(void) jsimd_h2v2_upsample_avx2
+ (int max_v_samp_factor, JDIMENSION output_width, JSAMPARRAY input_data,
+ JSAMPARRAY *output_data_ptr);
+
EXTERN(void) jsimd_h2v1_upsample_mips_dspr2
(int max_v_samp_factor, JDIMENSION output_width, JSAMPARRAY input_data,
JSAMPARRAY *output_data_ptr);
@@ -473,6 +560,14 @@
(int max_v_samp_factor, JDIMENSION downsampled_width,
JSAMPARRAY input_data, JSAMPARRAY *output_data_ptr);
+extern const int jconst_fancy_upsample_avx2[];
+EXTERN(void) jsimd_h2v1_fancy_upsample_avx2
+ (int max_v_samp_factor, JDIMENSION downsampled_width,
+ JSAMPARRAY input_data, JSAMPARRAY *output_data_ptr);
+EXTERN(void) jsimd_h2v2_fancy_upsample_avx2
+ (int max_v_samp_factor, JDIMENSION downsampled_width,
+ JSAMPARRAY input_data, JSAMPARRAY *output_data_ptr);
+
EXTERN(void) jsimd_h2v1_fancy_upsample_neon
(int max_v_samp_factor, JDIMENSION downsampled_width,
JSAMPARRAY input_data, JSAMPARRAY *output_data_ptr);
@@ -581,6 +676,51 @@
(JDIMENSION output_width, JSAMPIMAGE input_buf,
JDIMENSION in_row_group_ctr, JSAMPARRAY output_buf);
+extern const int jconst_merged_upsample_avx2[];
+EXTERN(void) jsimd_h2v1_merged_upsample_avx2
+ (JDIMENSION output_width, JSAMPIMAGE input_buf,
+ JDIMENSION in_row_group_ctr, JSAMPARRAY output_buf);
+EXTERN(void) jsimd_h2v1_extrgb_merged_upsample_avx2
+ (JDIMENSION output_width, JSAMPIMAGE input_buf,
+ JDIMENSION in_row_group_ctr, JSAMPARRAY output_buf);
+EXTERN(void) jsimd_h2v1_extrgbx_merged_upsample_avx2
+ (JDIMENSION output_width, JSAMPIMAGE input_buf,
+ JDIMENSION in_row_group_ctr, JSAMPARRAY output_buf);
+EXTERN(void) jsimd_h2v1_extbgr_merged_upsample_avx2
+ (JDIMENSION output_width, JSAMPIMAGE input_buf,
+ JDIMENSION in_row_group_ctr, JSAMPARRAY output_buf);
+EXTERN(void) jsimd_h2v1_extbgrx_merged_upsample_avx2
+ (JDIMENSION output_width, JSAMPIMAGE input_buf,
+ JDIMENSION in_row_group_ctr, JSAMPARRAY output_buf);
+EXTERN(void) jsimd_h2v1_extxbgr_merged_upsample_avx2
+ (JDIMENSION output_width, JSAMPIMAGE input_buf,
+ JDIMENSION in_row_group_ctr, JSAMPARRAY output_buf);
+EXTERN(void) jsimd_h2v1_extxrgb_merged_upsample_avx2
+ (JDIMENSION output_width, JSAMPIMAGE input_buf,
+ JDIMENSION in_row_group_ctr, JSAMPARRAY output_buf);
+
+EXTERN(void) jsimd_h2v2_merged_upsample_avx2
+ (JDIMENSION output_width, JSAMPIMAGE input_buf,
+ JDIMENSION in_row_group_ctr, JSAMPARRAY output_buf);
+EXTERN(void) jsimd_h2v2_extrgb_merged_upsample_avx2
+ (JDIMENSION output_width, JSAMPIMAGE input_buf,
+ JDIMENSION in_row_group_ctr, JSAMPARRAY output_buf);
+EXTERN(void) jsimd_h2v2_extrgbx_merged_upsample_avx2
+ (JDIMENSION output_width, JSAMPIMAGE input_buf,
+ JDIMENSION in_row_group_ctr, JSAMPARRAY output_buf);
+EXTERN(void) jsimd_h2v2_extbgr_merged_upsample_avx2
+ (JDIMENSION output_width, JSAMPIMAGE input_buf,
+ JDIMENSION in_row_group_ctr, JSAMPARRAY output_buf);
+EXTERN(void) jsimd_h2v2_extbgrx_merged_upsample_avx2
+ (JDIMENSION output_width, JSAMPIMAGE input_buf,
+ JDIMENSION in_row_group_ctr, JSAMPARRAY output_buf);
+EXTERN(void) jsimd_h2v2_extxbgr_merged_upsample_avx2
+ (JDIMENSION output_width, JSAMPIMAGE input_buf,
+ JDIMENSION in_row_group_ctr, JSAMPARRAY output_buf);
+EXTERN(void) jsimd_h2v2_extxrgb_merged_upsample_avx2
+ (JDIMENSION output_width, JSAMPIMAGE input_buf,
+ JDIMENSION in_row_group_ctr, JSAMPARRAY output_buf);
+
EXTERN(void) jsimd_h2v1_merged_upsample_mips_dspr2
(JDIMENSION output_width, JSAMPIMAGE input_buf,
JDIMENSION in_row_group_ctr, JSAMPARRAY output_buf, JSAMPLE* range);
@@ -735,6 +875,9 @@
EXTERN(void) jsimd_quantize_sse2
(JCOEFPTR coef_block, DCTELEM *divisors, DCTELEM *workspace);
+EXTERN(void) jsimd_quantize_avx2
+ (JCOEFPTR coef_block, DCTELEM *divisors, DCTELEM *workspace);
+
EXTERN(void) jsimd_quantize_neon
(JCOEFPTR coef_block, DCTELEM *divisors, DCTELEM *workspace);
diff --git a/simd/jsimdcfg.inc.h b/simd/jsimdcfg.inc.h
index d2b499f..7ff7e29 100644
--- a/simd/jsimdcfg.inc.h
+++ b/simd/jsimdcfg.inc.h
@@ -19,79 +19,79 @@
; -- jpeglib.h
;
-%define _cpp_protection_DCTSIZE DCTSIZE
-%define _cpp_protection_DCTSIZE2 DCTSIZE2
+%define _cpp_protection_DCTSIZE DCTSIZE
+%define _cpp_protection_DCTSIZE2 DCTSIZE2
;
; -- jmorecfg.h
;
-%define _cpp_protection_RGB_RED RGB_RED
-%define _cpp_protection_RGB_GREEN RGB_GREEN
-%define _cpp_protection_RGB_BLUE RGB_BLUE
-%define _cpp_protection_RGB_PIXELSIZE RGB_PIXELSIZE
+%define _cpp_protection_RGB_RED RGB_RED
+%define _cpp_protection_RGB_GREEN RGB_GREEN
+%define _cpp_protection_RGB_BLUE RGB_BLUE
+%define _cpp_protection_RGB_PIXELSIZE RGB_PIXELSIZE
-%define _cpp_protection_EXT_RGB_RED EXT_RGB_RED
-%define _cpp_protection_EXT_RGB_GREEN EXT_RGB_GREEN
-%define _cpp_protection_EXT_RGB_BLUE EXT_RGB_BLUE
-%define _cpp_protection_EXT_RGB_PIXELSIZE EXT_RGB_PIXELSIZE
+%define _cpp_protection_EXT_RGB_RED EXT_RGB_RED
+%define _cpp_protection_EXT_RGB_GREEN EXT_RGB_GREEN
+%define _cpp_protection_EXT_RGB_BLUE EXT_RGB_BLUE
+%define _cpp_protection_EXT_RGB_PIXELSIZE EXT_RGB_PIXELSIZE
-%define _cpp_protection_EXT_RGBX_RED EXT_RGBX_RED
-%define _cpp_protection_EXT_RGBX_GREEN EXT_RGBX_GREEN
-%define _cpp_protection_EXT_RGBX_BLUE EXT_RGBX_BLUE
-%define _cpp_protection_EXT_RGBX_PIXELSIZE EXT_RGBX_PIXELSIZE
+%define _cpp_protection_EXT_RGBX_RED EXT_RGBX_RED
+%define _cpp_protection_EXT_RGBX_GREEN EXT_RGBX_GREEN
+%define _cpp_protection_EXT_RGBX_BLUE EXT_RGBX_BLUE
+%define _cpp_protection_EXT_RGBX_PIXELSIZE EXT_RGBX_PIXELSIZE
-%define _cpp_protection_EXT_BGR_RED EXT_BGR_RED
-%define _cpp_protection_EXT_BGR_GREEN EXT_BGR_GREEN
-%define _cpp_protection_EXT_BGR_BLUE EXT_BGR_BLUE
-%define _cpp_protection_EXT_BGR_PIXELSIZE EXT_BGR_PIXELSIZE
+%define _cpp_protection_EXT_BGR_RED EXT_BGR_RED
+%define _cpp_protection_EXT_BGR_GREEN EXT_BGR_GREEN
+%define _cpp_protection_EXT_BGR_BLUE EXT_BGR_BLUE
+%define _cpp_protection_EXT_BGR_PIXELSIZE EXT_BGR_PIXELSIZE
-%define _cpp_protection_EXT_BGRX_RED EXT_BGRX_RED
-%define _cpp_protection_EXT_BGRX_GREEN EXT_BGRX_GREEN
-%define _cpp_protection_EXT_BGRX_BLUE EXT_BGRX_BLUE
-%define _cpp_protection_EXT_BGRX_PIXELSIZE EXT_BGRX_PIXELSIZE
+%define _cpp_protection_EXT_BGRX_RED EXT_BGRX_RED
+%define _cpp_protection_EXT_BGRX_GREEN EXT_BGRX_GREEN
+%define _cpp_protection_EXT_BGRX_BLUE EXT_BGRX_BLUE
+%define _cpp_protection_EXT_BGRX_PIXELSIZE EXT_BGRX_PIXELSIZE
-%define _cpp_protection_EXT_XBGR_RED EXT_XBGR_RED
-%define _cpp_protection_EXT_XBGR_GREEN EXT_XBGR_GREEN
-%define _cpp_protection_EXT_XBGR_BLUE EXT_XBGR_BLUE
-%define _cpp_protection_EXT_XBGR_PIXELSIZE EXT_XBGR_PIXELSIZE
+%define _cpp_protection_EXT_XBGR_RED EXT_XBGR_RED
+%define _cpp_protection_EXT_XBGR_GREEN EXT_XBGR_GREEN
+%define _cpp_protection_EXT_XBGR_BLUE EXT_XBGR_BLUE
+%define _cpp_protection_EXT_XBGR_PIXELSIZE EXT_XBGR_PIXELSIZE
-%define _cpp_protection_EXT_XRGB_RED EXT_XRGB_RED
-%define _cpp_protection_EXT_XRGB_GREEN EXT_XRGB_GREEN
-%define _cpp_protection_EXT_XRGB_BLUE EXT_XRGB_BLUE
-%define _cpp_protection_EXT_XRGB_PIXELSIZE EXT_XRGB_PIXELSIZE
+%define _cpp_protection_EXT_XRGB_RED EXT_XRGB_RED
+%define _cpp_protection_EXT_XRGB_GREEN EXT_XRGB_GREEN
+%define _cpp_protection_EXT_XRGB_BLUE EXT_XRGB_BLUE
+%define _cpp_protection_EXT_XRGB_PIXELSIZE EXT_XRGB_PIXELSIZE
-%define RGBX_FILLER_0XFF 1
+%define RGBX_FILLER_0XFF 1
; Representation of a single sample (pixel element value).
; On this SIMD implementation, this must be 'unsigned char'.
;
-%define JSAMPLE byte ; unsigned char
-%define SIZEOF_JSAMPLE SIZEOF_BYTE ; sizeof(JSAMPLE)
+%define JSAMPLE byte ; unsigned char
+%define SIZEOF_JSAMPLE SIZEOF_BYTE ; sizeof(JSAMPLE)
-%define _cpp_protection_CENTERJSAMPLE CENTERJSAMPLE
+%define _cpp_protection_CENTERJSAMPLE CENTERJSAMPLE
; Representation of a DCT frequency coefficient.
; On this SIMD implementation, this must be 'short'.
;
-%define JCOEF word ; short
-%define SIZEOF_JCOEF SIZEOF_WORD ; sizeof(JCOEF)
+%define JCOEF word ; short
+%define SIZEOF_JCOEF SIZEOF_WORD ; sizeof(JCOEF)
; Datatype used for image dimensions.
; On this SIMD implementation, this must be 'unsigned int'.
;
-%define JDIMENSION dword ; unsigned int
-%define SIZEOF_JDIMENSION SIZEOF_DWORD ; sizeof(JDIMENSION)
+%define JDIMENSION dword ; unsigned int
+%define SIZEOF_JDIMENSION SIZEOF_DWORD ; sizeof(JDIMENSION)
-%define JSAMPROW POINTER ; JSAMPLE * (jpeglib.h)
-%define JSAMPARRAY POINTER ; JSAMPROW * (jpeglib.h)
-%define JSAMPIMAGE POINTER ; JSAMPARRAY * (jpeglib.h)
-%define JCOEFPTR POINTER ; JCOEF * (jpeglib.h)
-%define SIZEOF_JSAMPROW SIZEOF_POINTER ; sizeof(JSAMPROW)
-%define SIZEOF_JSAMPARRAY SIZEOF_POINTER ; sizeof(JSAMPARRAY)
-%define SIZEOF_JSAMPIMAGE SIZEOF_POINTER ; sizeof(JSAMPIMAGE)
-%define SIZEOF_JCOEFPTR SIZEOF_POINTER ; sizeof(JCOEFPTR)
+%define JSAMPROW POINTER ; JSAMPLE * (jpeglib.h)
+%define JSAMPARRAY POINTER ; JSAMPROW * (jpeglib.h)
+%define JSAMPIMAGE POINTER ; JSAMPARRAY * (jpeglib.h)
+%define JCOEFPTR POINTER ; JCOEF * (jpeglib.h)
+%define SIZEOF_JSAMPROW SIZEOF_POINTER ; sizeof(JSAMPROW)
+%define SIZEOF_JSAMPARRAY SIZEOF_POINTER ; sizeof(JSAMPARRAY)
+%define SIZEOF_JSAMPIMAGE SIZEOF_POINTER ; sizeof(JSAMPIMAGE)
+%define SIZEOF_JCOEFPTR SIZEOF_POINTER ; sizeof(JCOEFPTR)
;
; -- jdct.h
@@ -101,30 +101,31 @@
; the DCT is to be performed in-place in that buffer.
; To maximize parallelism, Type DCTELEM is changed to short (originally, int).
;
-%define DCTELEM word ; short
-%define SIZEOF_DCTELEM SIZEOF_WORD ; sizeof(DCTELEM)
+%define DCTELEM word ; short
+%define SIZEOF_DCTELEM SIZEOF_WORD ; sizeof(DCTELEM)
-%define FAST_FLOAT FP32 ; float
-%define SIZEOF_FAST_FLOAT SIZEOF_FP32 ; sizeof(FAST_FLOAT)
+%define FAST_FLOAT FP32 ; float
+%define SIZEOF_FAST_FLOAT SIZEOF_FP32 ; sizeof(FAST_FLOAT)
; To maximize parallelism, Type MULTIPLIER is changed to short.
;
-%define ISLOW_MULT_TYPE word ; must be short
-%define SIZEOF_ISLOW_MULT_TYPE SIZEOF_WORD ; sizeof(ISLOW_MULT_TYPE)
+%define ISLOW_MULT_TYPE word ; must be short
+%define SIZEOF_ISLOW_MULT_TYPE SIZEOF_WORD ; sizeof(ISLOW_MULT_TYPE)
-%define IFAST_MULT_TYPE word ; must be short
-%define SIZEOF_IFAST_MULT_TYPE SIZEOF_WORD ; sizeof(IFAST_MULT_TYPE)
-%define IFAST_SCALE_BITS 2 ; fractional bits in scale factors
+%define IFAST_MULT_TYPE word ; must be short
+%define SIZEOF_IFAST_MULT_TYPE SIZEOF_WORD ; sizeof(IFAST_MULT_TYPE)
+%define IFAST_SCALE_BITS 2 ; fractional bits in scale factors
-%define FLOAT_MULT_TYPE FP32 ; must be float
-%define SIZEOF_FLOAT_MULT_TYPE SIZEOF_FP32 ; sizeof(FLOAT_MULT_TYPE)
+%define FLOAT_MULT_TYPE FP32 ; must be float
+%define SIZEOF_FLOAT_MULT_TYPE SIZEOF_FP32 ; sizeof(FLOAT_MULT_TYPE)
;
; -- jsimd.h
;
-%define _cpp_protection_JSIMD_NONE JSIMD_NONE
-%define _cpp_protection_JSIMD_MMX JSIMD_MMX
-%define _cpp_protection_JSIMD_3DNOW JSIMD_3DNOW
-%define _cpp_protection_JSIMD_SSE JSIMD_SSE
-%define _cpp_protection_JSIMD_SSE2 JSIMD_SSE2
+%define _cpp_protection_JSIMD_NONE JSIMD_NONE
+%define _cpp_protection_JSIMD_MMX JSIMD_MMX
+%define _cpp_protection_JSIMD_3DNOW JSIMD_3DNOW
+%define _cpp_protection_JSIMD_SSE JSIMD_SSE
+%define _cpp_protection_JSIMD_SSE2 JSIMD_SSE2
+%define _cpp_protection_JSIMD_AVX2 JSIMD_AVX2
diff --git a/simd/jsimdcpu.asm b/simd/jsimdcpu.asm
deleted file mode 100644
index 599083b..0000000
--- a/simd/jsimdcpu.asm
+++ /dev/null
@@ -1,104 +0,0 @@
-;
-; jsimdcpu.asm - SIMD instruction support check
-;
-; Copyright 2009 Pierre Ossman <ossman@cendio.se> for Cendio AB
-;
-; Based on the x86 SIMD extension for IJG JPEG library
-; Copyright (C) 1999-2006, MIYASAKA Masaru.
-; For conditions of distribution and use, see copyright notice in jsimdext.inc
-;
-; This file should be assembled with NASM (Netwide Assembler),
-; can *not* be assembled with Microsoft's MASM or any compatible
-; assembler (including Borland's Turbo Assembler).
-; NASM is available from http://nasm.sourceforge.net/ or
-; http://sourceforge.net/project/showfiles.php?group_id=6208
-;
-; [TAB8]
-
-%include "jsimdext.inc"
-
-; --------------------------------------------------------------------------
- SECTION SEG_TEXT
- BITS 32
-;
-; Check if the CPU supports SIMD instructions
-;
-; GLOBAL(unsigned int)
-; jpeg_simd_cpu_support (void)
-;
-
- align 16
- global EXTN(jpeg_simd_cpu_support)
-
-EXTN(jpeg_simd_cpu_support):
- push ebx
-; push ecx ; need not be preserved
-; push edx ; need not be preserved
-; push esi ; unused
- push edi
-
- xor edi,edi ; simd support flag
-
- pushfd
- pop eax
- mov edx,eax
- xor eax, 1<<21 ; flip ID bit in EFLAGS
- push eax
- popfd
- pushfd
- pop eax
- xor eax,edx
- jz short .return ; CPUID is not supported
-
- ; Check for MMX instruction support
- xor eax,eax
- cpuid
- test eax,eax
- jz short .return
-
- xor eax,eax
- inc eax
- cpuid
- mov eax,edx ; eax = Standard feature flags
-
- test eax, 1<<23 ; bit23:MMX
- jz short .no_mmx
- or edi, byte JSIMD_MMX
-.no_mmx:
- test eax, 1<<25 ; bit25:SSE
- jz short .no_sse
- or edi, byte JSIMD_SSE
-.no_sse:
- test eax, 1<<26 ; bit26:SSE2
- jz short .no_sse2
- or edi, byte JSIMD_SSE2
-.no_sse2:
-
- ; Check for 3DNow! instruction support
- mov eax, 0x80000000
- cpuid
- cmp eax, 0x80000000
- jbe short .return
-
- mov eax, 0x80000001
- cpuid
- mov eax,edx ; eax = Extended feature flags
-
- test eax, 1<<31 ; bit31:3DNow!(vendor independent)
- jz short .no_3dnow
- or edi, byte JSIMD_3DNOW
-.no_3dnow:
-
-.return:
- mov eax,edi
-
- pop edi
-; pop esi ; unused
-; pop edx ; need not be preserved
-; pop ecx ; need not be preserved
- pop ebx
- ret
-
-; For some reason, the OS X linker does not honor the request to align the
-; segment unless we do this.
- align 16
diff --git a/simd/jsimdext.inc b/simd/jsimdext.inc
deleted file mode 100644
index f28db60..0000000
--- a/simd/jsimdext.inc
+++ /dev/null
@@ -1,375 +0,0 @@
-;
-; jsimdext.inc - common declarations
-;
-; Copyright 2009 Pierre Ossman <ossman@cendio.se> for Cendio AB
-; Copyright (C) 2010, D. R. Commander.
-;
-; Based on the x86 SIMD extension for IJG JPEG library - version 1.02
-;
-; Copyright (C) 1999-2006, MIYASAKA Masaru.
-;
-; This software is provided 'as-is', without any express or implied
-; warranty. In no event will the authors be held liable for any damages
-; arising from the use of this software.
-;
-; Permission is granted to anyone to use this software for any purpose,
-; including commercial applications, and to alter it and redistribute it
-; freely, subject to the following restrictions:
-;
-; 1. The origin of this software must not be misrepresented; you must not
-; claim that you wrote the original software. If you use this software
-; in a product, an acknowledgment in the product documentation would be
-; appreciated but is not required.
-; 2. Altered source versions must be plainly marked as such, and must not be
-; misrepresented as being the original software.
-; 3. This notice may not be removed or altered from any source distribution.
-;
-; [TAB8]
-
-; ==========================================================================
-; System-dependent configurations
-
-%ifdef WIN32 ; ----(nasm -fwin32 -DWIN32 ...)--------
-; * Microsoft Visual C++
-; * MinGW (Minimalist GNU for Windows)
-; * CygWin
-; * LCC-Win32
-
-; -- segment definition --
-;
-%ifdef __YASM_VER__
-%define SEG_TEXT .text align=16
-%define SEG_CONST .rdata align=16
-%else
-%define SEG_TEXT .text align=16 public use32 class=CODE
-%define SEG_CONST .rdata align=16 public use32 class=CONST
-%endif
-
-%elifdef WIN64 ; ----(nasm -fwin64 -DWIN64 ...)--------
-; * Microsoft Visual C++
-
-; -- segment definition --
-;
-%ifdef __YASM_VER__
-%define SEG_TEXT .text align=16
-%define SEG_CONST .rdata align=16
-%else
-%define SEG_TEXT .text align=16 public use64 class=CODE
-%define SEG_CONST .rdata align=16 public use64 class=CONST
-%endif
-%define EXTN(name) name ; foo() -> foo
-
-%elifdef OBJ32 ; ----(nasm -fobj -DOBJ32 ...)----------
-; * Borland C++ (Win32)
-
-; -- segment definition --
-;
-%define SEG_TEXT _text align=16 public use32 class=CODE
-%define SEG_CONST _data align=16 public use32 class=DATA
-
-%elifdef ELF ; ----(nasm -felf[64] -DELF ...)------------
-; * Linux
-; * *BSD family Unix using elf format
-; * Unix System V, including Solaris x86, UnixWare and SCO Unix
-
-; mark stack as non-executable
-section .note.GNU-stack noalloc noexec nowrite progbits
-
-; -- segment definition --
-;
-%ifdef __x86_64__
-%define SEG_TEXT .text progbits align=16
-%define SEG_CONST .rodata progbits align=16
-%else
-%define SEG_TEXT .text progbits alloc exec nowrite align=16
-%define SEG_CONST .rodata progbits alloc noexec nowrite align=16
-%endif
-
-; To make the code position-independent, append -DPIC to the commandline
-;
-%define GOT_SYMBOL _GLOBAL_OFFSET_TABLE_ ; ELF supports PIC
-%define EXTN(name) name ; foo() -> foo
-
-%elifdef AOUT ; ----(nasm -faoutb/aout -DAOUT ...)----
-; * Older Linux using a.out format (nasm -f aout -DAOUT ...)
-; * *BSD family Unix using a.out format (nasm -f aoutb -DAOUT ...)
-
-; -- segment definition --
-;
-%define SEG_TEXT .text
-%define SEG_CONST .data
-
-; To make the code position-independent, append -DPIC to the commandline
-;
-%define GOT_SYMBOL __GLOBAL_OFFSET_TABLE_ ; BSD-style a.out supports PIC
-
-%elifdef MACHO ; ----(nasm -fmacho -DMACHO ...)--------
-; * NeXTstep/OpenStep/Rhapsody/Darwin/MacOS X (Mach-O format)
-
-; -- segment definition --
-;
-%define SEG_TEXT .text ;align=16 ; nasm doesn't accept align=16. why?
-%define SEG_CONST .rodata align=16
-
-; The generation of position-independent code (PIC) is the default on Darwin.
-;
-%define PIC
-%define GOT_SYMBOL _MACHO_PIC_ ; Mach-O style code-relative addressing
-
-%else ; ----(Other case)----------------------
-
-; -- segment definition --
-;
-%define SEG_TEXT .text
-%define SEG_CONST .data
-
-%endif ; ----------------------------------------------
-
-; ==========================================================================
-
-; --------------------------------------------------------------------------
-; Common types
-;
-%ifdef __x86_64__
-%define POINTER qword ; general pointer type
-%define SIZEOF_POINTER SIZEOF_QWORD ; sizeof(POINTER)
-%define POINTER_BIT QWORD_BIT ; sizeof(POINTER)*BYTE_BIT
-%else
-%define POINTER dword ; general pointer type
-%define SIZEOF_POINTER SIZEOF_DWORD ; sizeof(POINTER)
-%define POINTER_BIT DWORD_BIT ; sizeof(POINTER)*BYTE_BIT
-%endif
-
-%define INT dword ; signed integer type
-%define SIZEOF_INT SIZEOF_DWORD ; sizeof(INT)
-%define INT_BIT DWORD_BIT ; sizeof(INT)*BYTE_BIT
-
-%define FP32 dword ; IEEE754 single
-%define SIZEOF_FP32 SIZEOF_DWORD ; sizeof(FP32)
-%define FP32_BIT DWORD_BIT ; sizeof(FP32)*BYTE_BIT
-
-%define MMWORD qword ; int64 (MMX register)
-%define SIZEOF_MMWORD SIZEOF_QWORD ; sizeof(MMWORD)
-%define MMWORD_BIT QWORD_BIT ; sizeof(MMWORD)*BYTE_BIT
-
-; NASM is buggy and doesn't properly handle operand sizes for SSE
-; instructions, so for now we have to define XMMWORD as blank.
-%define XMMWORD ; int128 (SSE register)
-%define SIZEOF_XMMWORD SIZEOF_OWORD ; sizeof(XMMWORD)
-%define XMMWORD_BIT OWORD_BIT ; sizeof(XMMWORD)*BYTE_BIT
-
-; Similar hacks for when we load a dword or MMWORD into an xmm# register
-%define XMM_DWORD
-%define XMM_MMWORD
-
-%define SIZEOF_BYTE 1 ; sizeof(BYTE)
-%define SIZEOF_WORD 2 ; sizeof(WORD)
-%define SIZEOF_DWORD 4 ; sizeof(DWORD)
-%define SIZEOF_QWORD 8 ; sizeof(QWORD)
-%define SIZEOF_OWORD 16 ; sizeof(OWORD)
-
-%define BYTE_BIT 8 ; CHAR_BIT in C
-%define WORD_BIT 16 ; sizeof(WORD)*BYTE_BIT
-%define DWORD_BIT 32 ; sizeof(DWORD)*BYTE_BIT
-%define QWORD_BIT 64 ; sizeof(QWORD)*BYTE_BIT
-%define OWORD_BIT 128 ; sizeof(OWORD)*BYTE_BIT
-
-; --------------------------------------------------------------------------
-; External Symbol Name
-;
-%ifndef EXTN
-%define EXTN(name) _ %+ name ; foo() -> _foo
-%endif
-
-; --------------------------------------------------------------------------
-; Macros for position-independent code (PIC) support
-;
-%ifndef GOT_SYMBOL
-%undef PIC
-%endif
-
-%ifdef PIC ; -------------------------------------------
-
-%ifidn GOT_SYMBOL,_MACHO_PIC_ ; --------------------
-
-; At present, nasm doesn't seem to support PIC generation for Mach-O.
-; The PIC support code below is a little tricky.
-
- SECTION SEG_CONST
-const_base:
-
-%define GOTOFF(got,sym) (got) + (sym) - const_base
-
-%imacro get_GOT 1
- ; NOTE: this macro destroys ecx resister.
- call %%geteip
- add ecx, byte (%%ref - $)
- jmp short %%adjust
-%%geteip:
- mov ecx, POINTER [esp]
- ret
-%%adjust:
- push ebp
- xor ebp,ebp ; ebp = 0
-%ifidni %1,ebx ; (%1 == ebx)
- ; db 0x8D,0x9C + jmp near const_base =
- ; lea ebx, [ecx+ebp*8+(const_base-%%ref)] ; 8D,9C,E9,(offset32)
- db 0x8D,0x9C ; 8D,9C
- jmp near const_base ; E9,(const_base-%%ref)
-%%ref:
-%else ; (%1 != ebx)
- ; db 0x8D,0x8C + jmp near const_base =
- ; lea ecx, [ecx+ebp*8+(const_base-%%ref)] ; 8D,8C,E9,(offset32)
- db 0x8D,0x8C ; 8D,8C
- jmp near const_base ; E9,(const_base-%%ref)
-%%ref: mov %1, ecx
-%endif ; (%1 == ebx)
- pop ebp
-%endmacro
-
-%else ; GOT_SYMBOL != _MACHO_PIC_ ----------------
-
-%define GOTOFF(got,sym) (got) + (sym) wrt ..gotoff
-
-%imacro get_GOT 1
- extern GOT_SYMBOL
- call %%geteip
- add %1, GOT_SYMBOL + $$ - $ wrt ..gotpc
- jmp short %%done
-%%geteip:
- mov %1, POINTER [esp]
- ret
-%%done:
-%endmacro
-
-%endif ; GOT_SYMBOL == _MACHO_PIC_ ----------------
-
-%imacro pushpic 1.nolist
- push %1
-%endmacro
-%imacro poppic 1.nolist
- pop %1
-%endmacro
-%imacro movpic 2.nolist
- mov %1,%2
-%endmacro
-
-%else ; !PIC -----------------------------------------
-
-%define GOTOFF(got,sym) (sym)
-
-%imacro get_GOT 1.nolist
-%endmacro
-%imacro pushpic 1.nolist
-%endmacro
-%imacro poppic 1.nolist
-%endmacro
-%imacro movpic 2.nolist
-%endmacro
-
-%endif ; PIC -----------------------------------------
-
-; --------------------------------------------------------------------------
-; Align the next instruction on {2,4,8,16,..}-byte boundary.
-; ".balign n,,m" in GNU as
-;
-%define MSKLE(x,y) (~(((y) & 0xFFFF) - ((x) & 0xFFFF)) >> 16)
-%define FILLB(b,n) (($$-(b)) & ((n)-1))
-
-%imacro alignx 1-2.nolist 0xFFFF
-%%bs: times MSKLE(FILLB(%%bs,%1),%2) & MSKLE(16,FILLB($,%1)) & FILLB($,%1) \
- db 0x90 ; nop
- times MSKLE(FILLB(%%bs,%1),%2) & FILLB($,%1)/9 \
- db 0x8D,0x9C,0x23,0x00,0x00,0x00,0x00 ; lea ebx,[ebx+0x00000000]
- times MSKLE(FILLB(%%bs,%1),%2) & FILLB($,%1)/7 \
- db 0x8D,0xAC,0x25,0x00,0x00,0x00,0x00 ; lea ebp,[ebp+0x00000000]
- times MSKLE(FILLB(%%bs,%1),%2) & FILLB($,%1)/6 \
- db 0x8D,0xAD,0x00,0x00,0x00,0x00 ; lea ebp,[ebp+0x00000000]
- times MSKLE(FILLB(%%bs,%1),%2) & FILLB($,%1)/4 \
- db 0x8D,0x6C,0x25,0x00 ; lea ebp,[ebp+0x00]
- times MSKLE(FILLB(%%bs,%1),%2) & FILLB($,%1)/3 \
- db 0x8D,0x6D,0x00 ; lea ebp,[ebp+0x00]
- times MSKLE(FILLB(%%bs,%1),%2) & FILLB($,%1)/2 \
- db 0x8B,0xED ; mov ebp,ebp
- times MSKLE(FILLB(%%bs,%1),%2) & FILLB($,%1)/1 \
- db 0x90 ; nop
-%endmacro
-
-; Align the next data on {2,4,8,16,..}-byte boundary.
-;
-%imacro alignz 1.nolist
- align %1, db 0 ; filling zeros
-%endmacro
-
-%ifdef __x86_64__
-
-%ifdef WIN64
-
-%imacro collect_args 0
- push r12
- push r13
- push r14
- push r15
- mov r10, rcx
- mov r11, rdx
- mov r12, r8
- mov r13, r9
- mov r14, [rax+48]
- mov r15, [rax+56]
- push rsi
- push rdi
- sub rsp, SIZEOF_XMMWORD
- movaps XMMWORD [rsp], xmm6
- sub rsp, SIZEOF_XMMWORD
- movaps XMMWORD [rsp], xmm7
-%endmacro
-
-%imacro uncollect_args 0
- movaps xmm7, XMMWORD [rsp]
- add rsp, SIZEOF_XMMWORD
- movaps xmm6, XMMWORD [rsp]
- add rsp, SIZEOF_XMMWORD
- pop rdi
- pop rsi
- pop r15
- pop r14
- pop r13
- pop r12
-%endmacro
-
-%else
-
-%imacro collect_args 0
- push r10
- push r11
- push r12
- push r13
- push r14
- push r15
- mov r10, rdi
- mov r11, rsi
- mov r12, rdx
- mov r13, rcx
- mov r14, r8
- mov r15, r9
-%endmacro
-
-%imacro uncollect_args 0
- pop r15
- pop r14
- pop r13
- pop r12
- pop r11
- pop r10
-%endmacro
-
-%endif
-
-%endif
-
-; --------------------------------------------------------------------------
-; Defines picked up from the C headers
-;
-%include "jsimdcfg.inc"
-
-; --------------------------------------------------------------------------
diff --git a/simd/jsimd_mips.c b/simd/mips/jsimd.c
similarity index 98%
rename from simd/jsimd_mips.c
rename to simd/mips/jsimd.c
index 02e90cd..05ab13a 100644
--- a/simd/jsimd_mips.c
+++ b/simd/mips/jsimd.c
@@ -16,12 +16,12 @@
*/
#define JPEG_INTERNALS
-#include "../jinclude.h"
-#include "../jpeglib.h"
+#include "../../jinclude.h"
+#include "../../jpeglib.h"
+#include "../../jsimd.h"
+#include "../../jdct.h"
+#include "../../jsimddct.h"
#include "../jsimd.h"
-#include "../jdct.h"
-#include "../jsimddct.h"
-#include "jsimd.h"
#include <stdio.h>
#include <string.h>
diff --git a/simd/jsimd_mips_dspr2.S b/simd/mips/jsimd_dspr2.S
similarity index 99%
rename from simd/jsimd_mips_dspr2.S
rename to simd/mips/jsimd_dspr2.S
index c8c286c..60ff6f9 100644
--- a/simd/jsimd_mips_dspr2.S
+++ b/simd/mips/jsimd_dspr2.S
@@ -23,7 +23,7 @@
* 3. This notice may not be removed or altered from any source distribution.
*/
-#include "jsimd_mips_dspr2_asm.h"
+#include "jsimd_dspr2_asm.h"
/*****************************************************************************/
LEAF_MIPS_DSPR2(jsimd_c_null_convert_mips_dspr2)
@@ -4484,4 +4484,3 @@
END(jsimd_convsamp_float_mips_dspr2)
/*****************************************************************************/
-
diff --git a/simd/jsimd_mips_dspr2_asm.h b/simd/mips/jsimd_dspr2_asm.h
similarity index 99%
rename from simd/jsimd_mips_dspr2_asm.h
rename to simd/mips/jsimd_dspr2_asm.h
index 64f9880..499e34b 100644
--- a/simd/jsimd_mips_dspr2_asm.h
+++ b/simd/mips/jsimd_dspr2_asm.h
@@ -281,5 +281,3 @@
addiu sp, sp, \stack_offset
.endif
.endm
-
-
diff --git a/simd/jcolsamp.inc b/simd/nasm/jcolsamp.inc
similarity index 76%
rename from simd/jcolsamp.inc
rename to simd/nasm/jcolsamp.inc
index 3be446e..3fa6214 100644
--- a/simd/jcolsamp.inc
+++ b/simd/nasm/jcolsamp.inc
@@ -2,6 +2,7 @@
; jcolsamp.inc - private declarations for color conversion & up/downsampling
;
; Copyright 2009 Pierre Ossman <ossman@cendio.se> for Cendio AB
+; Copyright (C) 2015, Intel Corporation.
;
; Based on the x86 SIMD extension for IJG JPEG library
; Copyright (C) 1999-2006, MIYASAKA Masaru.
@@ -18,21 +19,29 @@
%define mmB mm1
%define xmmA xmm0
%define xmmB xmm1
+%define ymmA ymm0
+%define ymmB ymm1
%elif RGB_GREEN == 0
%define mmA mm2
%define mmB mm3
%define xmmA xmm2
%define xmmB xmm3
+%define ymmA ymm2
+%define ymmB ymm3
%elif RGB_BLUE == 0
%define mmA mm4
%define mmB mm5
%define xmmA xmm4
%define xmmB xmm5
+%define ymmA ymm4
+%define ymmB ymm5
%else
%define mmA mm6
%define mmB mm7
%define xmmA xmm6
%define xmmB xmm7
+%define ymmA ymm6
+%define ymmB ymm7
%endif
%if RGB_RED == 1
@@ -40,21 +49,29 @@
%define mmD mm1
%define xmmC xmm0
%define xmmD xmm1
+%define ymmC ymm0
+%define ymmD ymm1
%elif RGB_GREEN == 1
%define mmC mm2
%define mmD mm3
%define xmmC xmm2
%define xmmD xmm3
+%define ymmC ymm2
+%define ymmD ymm3
%elif RGB_BLUE == 1
%define mmC mm4
%define mmD mm5
%define xmmC xmm4
%define xmmD xmm5
+%define ymmC ymm4
+%define ymmD ymm5
%else
%define mmC mm6
%define mmD mm7
%define xmmC xmm6
%define xmmD xmm7
+%define ymmC ymm6
+%define ymmD ymm7
%endif
%if RGB_RED == 2
@@ -62,21 +79,29 @@
%define mmF mm1
%define xmmE xmm0
%define xmmF xmm1
+%define ymmE ymm0
+%define ymmF ymm1
%elif RGB_GREEN == 2
%define mmE mm2
%define mmF mm3
%define xmmE xmm2
%define xmmF xmm3
+%define ymmE ymm2
+%define ymmF ymm3
%elif RGB_BLUE == 2
%define mmE mm4
%define mmF mm5
%define xmmE xmm4
%define xmmF xmm5
+%define ymmE ymm4
+%define ymmF ymm5
%else
%define mmE mm6
%define mmF mm7
%define xmmE xmm6
%define xmmF xmm7
+%define ymmE ymm6
+%define ymmF ymm7
%endif
%if RGB_RED == 3
@@ -84,21 +109,29 @@
%define mmH mm1
%define xmmG xmm0
%define xmmH xmm1
+%define ymmG ymm0
+%define ymmH ymm1
%elif RGB_GREEN == 3
%define mmG mm2
%define mmH mm3
%define xmmG xmm2
%define xmmH xmm3
+%define ymmG ymm2
+%define ymmH ymm3
%elif RGB_BLUE == 3
%define mmG mm4
%define mmH mm5
%define xmmG xmm4
%define xmmH xmm5
+%define ymmG ymm4
+%define ymmH ymm5
%else
%define mmG mm6
%define mmH mm7
%define xmmG xmm6
%define xmmH xmm7
+%define ymmG ymm6
+%define ymmH ymm7
%endif
; --------------------------------------------------------------------------
diff --git a/simd/jdct.inc b/simd/nasm/jdct.inc
similarity index 72%
rename from simd/jdct.inc
rename to simd/nasm/jdct.inc
index b976107..7ae2ca4 100644
--- a/simd/jdct.inc
+++ b/simd/nasm/jdct.inc
@@ -17,11 +17,11 @@
;
%define RANGE_MASK (MAXJSAMPLE * 4 + 3) ; 2 bits wider than legal samples
-%define ROW(n,b,s) ((b)+(n)*(s))
-%define COL(n,b,s) ((b)+(n)*(s)*DCTSIZE)
+%define ROW(n,b,s) ((b)+(n)*(s))
+%define COL(n,b,s) ((b)+(n)*(s)*DCTSIZE)
-%define DWBLOCK(m,n,b,s) ((b)+(m)*DCTSIZE*(s)+(n)*SIZEOF_DWORD)
-%define MMBLOCK(m,n,b,s) ((b)+(m)*DCTSIZE*(s)+(n)*SIZEOF_MMWORD)
-%define XMMBLOCK(m,n,b,s) ((b)+(m)*DCTSIZE*(s)+(n)*SIZEOF_XMMWORD)
+%define DWBLOCK(m,n,b,s) ((b)+(m)*DCTSIZE*(s)+(n)*SIZEOF_DWORD)
+%define MMBLOCK(m,n,b,s) ((b)+(m)*DCTSIZE*(s)+(n)*SIZEOF_MMWORD)
+%define XMMBLOCK(m,n,b,s) ((b)+(m)*DCTSIZE*(s)+(n)*SIZEOF_XMMWORD)
; --------------------------------------------------------------------------
diff --git a/simd/jpeg_nbits_table.inc b/simd/nasm/jpeg_nbits_table.inc
similarity index 100%
rename from simd/jpeg_nbits_table.inc
rename to simd/nasm/jpeg_nbits_table.inc
diff --git a/simd/nasm/jsimdext.inc b/simd/nasm/jsimdext.inc
new file mode 100644
index 0000000..3485d6d
--- /dev/null
+++ b/simd/nasm/jsimdext.inc
@@ -0,0 +1,455 @@
+;
+; jsimdext.inc - common declarations
+;
+; Copyright 2009 Pierre Ossman <ossman@cendio.se> for Cendio AB
+; Copyright (C) 2010, 2016, D. R. Commander.
+;
+; Based on the x86 SIMD extension for IJG JPEG library - version 1.02
+;
+; Copyright (C) 1999-2006, MIYASAKA Masaru.
+;
+; This software is provided 'as-is', without any express or implied
+; warranty. In no event will the authors be held liable for any damages
+; arising from the use of this software.
+;
+; Permission is granted to anyone to use this software for any purpose,
+; including commercial applications, and to alter it and redistribute it
+; freely, subject to the following restrictions:
+;
+; 1. The origin of this software must not be misrepresented; you must not
+; claim that you wrote the original software. If you use this software
+; in a product, an acknowledgment in the product documentation would be
+; appreciated but is not required.
+; 2. Altered source versions must be plainly marked as such, and must not be
+; misrepresented as being the original software.
+; 3. This notice may not be removed or altered from any source distribution.
+;
+; [TAB8]
+
+; ==========================================================================
+; System-dependent configurations
+
+%ifdef WIN32 ; ----(nasm -fwin32 -DWIN32 ...)--------
+; * Microsoft Visual C++
+; * MinGW (Minimalist GNU for Windows)
+; * CygWin
+; * LCC-Win32
+
+; -- segment definition --
+;
+%ifdef __YASM_VER__
+%define SEG_TEXT .text align=32
+%define SEG_CONST .rdata align=32
+%else
+%define SEG_TEXT .text align=32 public use32 class=CODE
+%define SEG_CONST .rdata align=32 public use32 class=CONST
+%endif
+
+%elifdef WIN64 ; ----(nasm -fwin64 -DWIN64 ...)--------
+; * Microsoft Visual C++
+
+; -- segment definition --
+;
+%ifdef __YASM_VER__
+%define SEG_TEXT .text align=32
+%define SEG_CONST .rdata align=32
+%else
+%define SEG_TEXT .text align=32 public use64 class=CODE
+%define SEG_CONST .rdata align=32 public use64 class=CONST
+%endif
+%define EXTN(name) name ; foo() -> foo
+
+%elifdef OBJ32 ; ----(nasm -fobj -DOBJ32 ...)----------
+; * Borland C++ (Win32)
+
+; -- segment definition --
+;
+%define SEG_TEXT _text align=32 public use32 class=CODE
+%define SEG_CONST _data align=32 public use32 class=DATA
+
+%elifdef ELF ; ----(nasm -felf[64] -DELF ...)------------
+; * Linux
+; * *BSD family Unix using elf format
+; * Unix System V, including Solaris x86, UnixWare and SCO Unix
+
+; mark stack as non-executable
+section .note.GNU-stack noalloc noexec nowrite progbits
+
+; -- segment definition --
+;
+%ifdef __x86_64__
+%define SEG_TEXT .text progbits align=32
+%define SEG_CONST .rodata progbits align=32
+%else
+%define SEG_TEXT .text progbits alloc exec nowrite align=32
+%define SEG_CONST .rodata progbits alloc noexec nowrite align=32
+%endif
+
+; To make the code position-independent, append -DPIC to the commandline
+;
+%define GOT_SYMBOL _GLOBAL_OFFSET_TABLE_ ; ELF supports PIC
+%define EXTN(name) name ; foo() -> foo
+
+%elifdef AOUT ; ----(nasm -faoutb/aout -DAOUT ...)----
+; * Older Linux using a.out format (nasm -f aout -DAOUT ...)
+; * *BSD family Unix using a.out format (nasm -f aoutb -DAOUT ...)
+
+; -- segment definition --
+;
+%define SEG_TEXT .text
+%define SEG_CONST .data
+
+; To make the code position-independent, append -DPIC to the commandline
+;
+%define GOT_SYMBOL __GLOBAL_OFFSET_TABLE_ ; BSD-style a.out supports PIC
+
+%elifdef MACHO ; ----(nasm -fmacho -DMACHO ...)--------
+; * NeXTstep/OpenStep/Rhapsody/Darwin/MacOS X (Mach-O format)
+
+; -- segment definition --
+;
+%define SEG_TEXT .text ;align=32 ; nasm doesn't accept align=32. why?
+%define SEG_CONST .rodata align=32
+
+; The generation of position-independent code (PIC) is the default on Darwin.
+;
+%define PIC
+%define GOT_SYMBOL _MACHO_PIC_ ; Mach-O style code-relative addressing
+
+%else ; ----(Other case)----------------------
+
+; -- segment definition --
+;
+%define SEG_TEXT .text
+%define SEG_CONST .data
+
+%endif ; ----------------------------------------------
+
+; ==========================================================================
+
+; --------------------------------------------------------------------------
+; Common types
+;
+%ifdef __x86_64__
+%define POINTER qword ; general pointer type
+%define SIZEOF_POINTER SIZEOF_QWORD ; sizeof(POINTER)
+%define POINTER_BIT QWORD_BIT ; sizeof(POINTER)*BYTE_BIT
+%else
+%define POINTER dword ; general pointer type
+%define SIZEOF_POINTER SIZEOF_DWORD ; sizeof(POINTER)
+%define POINTER_BIT DWORD_BIT ; sizeof(POINTER)*BYTE_BIT
+%endif
+
+%define INT dword ; signed integer type
+%define SIZEOF_INT SIZEOF_DWORD ; sizeof(INT)
+%define INT_BIT DWORD_BIT ; sizeof(INT)*BYTE_BIT
+
+%define FP32 dword ; IEEE754 single
+%define SIZEOF_FP32 SIZEOF_DWORD ; sizeof(FP32)
+%define FP32_BIT DWORD_BIT ; sizeof(FP32)*BYTE_BIT
+
+%define MMWORD qword ; int64 (MMX register)
+%define SIZEOF_MMWORD SIZEOF_QWORD ; sizeof(MMWORD)
+%define MMWORD_BIT QWORD_BIT ; sizeof(MMWORD)*BYTE_BIT
+
+; NASM is buggy and doesn't properly handle operand sizes for SSE
+; instructions, so for now we have to define XMMWORD as blank.
+%define XMMWORD ; int128 (SSE register)
+%define SIZEOF_XMMWORD SIZEOF_OWORD ; sizeof(XMMWORD)
+%define XMMWORD_BIT OWORD_BIT ; sizeof(XMMWORD)*BYTE_BIT
+
+%define YMMWORD ; int256 (AVX register)
+%define SIZEOF_YMMWORD SIZEOF_YWORD ; sizeof(YMMWORD)
+%define YMMWORD_BIT YWORD_BIT ; sizeof(YMMWORD)*BYTE_BIT
+
+; Similar hacks for when we load a dword or MMWORD into an xmm# register
+%define XMM_DWORD
+%define XMM_MMWORD
+
+%define SIZEOF_BYTE 1 ; sizeof(BYTE)
+%define SIZEOF_WORD 2 ; sizeof(WORD)
+%define SIZEOF_DWORD 4 ; sizeof(DWORD)
+%define SIZEOF_QWORD 8 ; sizeof(QWORD)
+%define SIZEOF_OWORD 16 ; sizeof(OWORD)
+%define SIZEOF_YWORD 32 ; sizeof(YWORD)
+
+%define BYTE_BIT 8 ; CHAR_BIT in C
+%define WORD_BIT 16 ; sizeof(WORD)*BYTE_BIT
+%define DWORD_BIT 32 ; sizeof(DWORD)*BYTE_BIT
+%define QWORD_BIT 64 ; sizeof(QWORD)*BYTE_BIT
+%define OWORD_BIT 128 ; sizeof(OWORD)*BYTE_BIT
+%define YWORD_BIT 256 ; sizeof(YWORD)*BYTE_BIT
+
+; --------------------------------------------------------------------------
+; External Symbol Name
+;
+%ifndef EXTN
+%define EXTN(name) _ %+ name ; foo() -> _foo
+%endif
+
+; --------------------------------------------------------------------------
+; Macros for position-independent code (PIC) support
+;
+%ifndef GOT_SYMBOL
+%undef PIC
+%endif
+
+%ifdef PIC ; -------------------------------------------
+
+%ifidn GOT_SYMBOL, _MACHO_PIC_ ; --------------------
+
+; At present, nasm doesn't seem to support PIC generation for Mach-O.
+; The PIC support code below is a little tricky.
+
+ SECTION SEG_CONST
+const_base:
+
+%define GOTOFF(got,sym) (got) + (sym) - const_base
+
+%imacro get_GOT 1
+ ; NOTE: this macro destroys ecx resister.
+ call %%geteip
+ add ecx, byte (%%ref - $)
+ jmp short %%adjust
+%%geteip:
+ mov ecx, POINTER [esp]
+ ret
+%%adjust:
+ push ebp
+ xor ebp, ebp ; ebp = 0
+%ifidni %1, ebx ; (%1 == ebx)
+ ; db 0x8D,0x9C + jmp near const_base =
+ ; lea ebx, [ecx+ebp*8+(const_base-%%ref)] ; 8D,9C,E9,(offset32)
+ db 0x8D, 0x9C ; 8D,9C
+ jmp near const_base ; E9,(const_base-%%ref)
+%%ref:
+%else ; (%1 != ebx)
+ ; db 0x8D,0x8C + jmp near const_base =
+ ; lea ecx, [ecx+ebp*8+(const_base-%%ref)] ; 8D,8C,E9,(offset32)
+ db 0x8D, 0x8C ; 8D,8C
+ jmp near const_base ; E9,(const_base-%%ref)
+%%ref:
+ mov %1, ecx
+%endif ; (%1 == ebx)
+ pop ebp
+%endmacro
+
+%else ; GOT_SYMBOL != _MACHO_PIC_ ----------------
+
+%define GOTOFF(got,sym) (got) + (sym) wrt ..gotoff
+
+%imacro get_GOT 1
+ extern GOT_SYMBOL
+ call %%geteip
+ add %1, GOT_SYMBOL + $$ - $ wrt ..gotpc
+ jmp short %%done
+%%geteip:
+ mov %1, POINTER [esp]
+ ret
+%%done:
+%endmacro
+
+%endif ; GOT_SYMBOL == _MACHO_PIC_ ----------------
+
+%imacro pushpic 1.nolist
+ push %1
+%endmacro
+%imacro poppic 1.nolist
+ pop %1
+%endmacro
+%imacro movpic 2.nolist
+ mov %1, %2
+%endmacro
+
+%else ; !PIC -----------------------------------------
+
+%define GOTOFF(got,sym) (sym)
+
+%imacro get_GOT 1.nolist
+%endmacro
+%imacro pushpic 1.nolist
+%endmacro
+%imacro poppic 1.nolist
+%endmacro
+%imacro movpic 2.nolist
+%endmacro
+
+%endif ; PIC -----------------------------------------
+
+; --------------------------------------------------------------------------
+; Align the next instruction on {2,4,8,16,..}-byte boundary.
+; ".balign n,,m" in GNU as
+;
+%define MSKLE(x,y) (~(((y) & 0xFFFF) - ((x) & 0xFFFF)) >> 16)
+%define FILLB(b,n) (($$-(b)) & ((n)-1))
+
+%imacro alignx 1-2.nolist 0xFFFF
+%%bs: \
+ times MSKLE(FILLB(%%bs,%1),%2) & MSKLE(16,FILLB($,%1)) & FILLB($,%1) \
+ db 0x90 ; nop
+ times MSKLE(FILLB(%%bs,%1),%2) & FILLB($,%1)/9 \
+ db 0x8D,0x9C,0x23,0x00,0x00,0x00,0x00 ; lea ebx,[ebx+0x00000000]
+ times MSKLE(FILLB(%%bs,%1),%2) & FILLB($,%1)/7 \
+ db 0x8D,0xAC,0x25,0x00,0x00,0x00,0x00 ; lea ebp,[ebp+0x00000000]
+ times MSKLE(FILLB(%%bs,%1),%2) & FILLB($,%1)/6 \
+ db 0x8D,0xAD,0x00,0x00,0x00,0x00 ; lea ebp,[ebp+0x00000000]
+ times MSKLE(FILLB(%%bs,%1),%2) & FILLB($,%1)/4 \
+ db 0x8D,0x6C,0x25,0x00 ; lea ebp,[ebp+0x00]
+ times MSKLE(FILLB(%%bs,%1),%2) & FILLB($,%1)/3 \
+ db 0x8D,0x6D,0x00 ; lea ebp,[ebp+0x00]
+ times MSKLE(FILLB(%%bs,%1),%2) & FILLB($,%1)/2 \
+ db 0x8B,0xED ; mov ebp,ebp
+ times MSKLE(FILLB(%%bs,%1),%2) & FILLB($,%1)/1 \
+ db 0x90 ; nop
+%endmacro
+
+; Align the next data on {2,4,8,16,..}-byte boundary.
+;
+%imacro alignz 1.nolist
+ align %1, db 0 ; filling zeros
+%endmacro
+
+%ifdef __x86_64__
+
+%ifdef WIN64
+
+%imacro collect_args 1
+ sub rsp, SIZEOF_XMMWORD
+ movaps XMMWORD [rsp], xmm6
+ sub rsp, SIZEOF_XMMWORD
+ movaps XMMWORD [rsp], xmm7
+ mov r10, rcx
+%if %1 > 1
+ mov r11, rdx
+%endif
+%if %1 > 2
+ push r12
+ mov r12, r8
+%endif
+%if %1 > 3
+ push r13
+ mov r13, r9
+%endif
+%if %1 > 4
+ push r14
+ mov r14, [rax+48]
+%endif
+%if %1 > 5
+ push r15
+ mov r15, [rax+56]
+%endif
+ push rsi
+ push rdi
+%endmacro
+
+%imacro uncollect_args 1
+ pop rdi
+ pop rsi
+%if %1 > 5
+ pop r15
+%endif
+%if %1 > 4
+ pop r14
+%endif
+%if %1 > 3
+ pop r13
+%endif
+%if %1 > 2
+ pop r12
+%endif
+ movaps xmm7, XMMWORD [rsp]
+ add rsp, SIZEOF_XMMWORD
+ movaps xmm6, XMMWORD [rsp]
+ add rsp, SIZEOF_XMMWORD
+%endmacro
+
+%imacro push_xmm 1
+ movaps XMMWORD [rsp-1*SIZEOF_XMMWORD], xmm8
+%if %1 > 1
+ movaps XMMWORD [rsp-2*SIZEOF_XMMWORD], xmm9
+%endif
+%if %1 > 2
+ movaps XMMWORD [rsp-3*SIZEOF_XMMWORD], xmm10
+%endif
+%if %1 > 3
+ movaps XMMWORD [rsp-4*SIZEOF_XMMWORD], xmm11
+%endif
+ sub rsp, %1 * SIZEOF_XMMWORD
+%endmacro
+
+%imacro pop_xmm 1
+%if %1 > 3
+ movaps xmm11, XMMWORD [rsp+0*SIZEOF_XMMWORD]
+%endif
+%if %1 > 2
+ movaps xmm10, XMMWORD [rsp+1*SIZEOF_XMMWORD]
+%endif
+%if %1 > 1
+ movaps xmm9, XMMWORD [rsp+2*SIZEOF_XMMWORD]
+%endif
+ movaps xmm8, XMMWORD [rsp+3*SIZEOF_XMMWORD]
+ add rsp, %1 * SIZEOF_XMMWORD
+%endmacro
+
+%else
+
+%imacro collect_args 1
+ push r10
+ mov r10, rdi
+%if %1 > 1
+ push r11
+ mov r11, rsi
+%endif
+%if %1 > 2
+ push r12
+ mov r12, rdx
+%endif
+%if %1 > 3
+ push r13
+ mov r13, rcx
+%endif
+%if %1 > 4
+ push r14
+ mov r14, r8
+%endif
+%if %1 > 5
+ push r15
+ mov r15, r9
+%endif
+%endmacro
+
+%imacro uncollect_args 1
+%if %1 > 5
+ pop r15
+%endif
+%if %1 > 4
+ pop r14
+%endif
+%if %1 > 3
+ pop r13
+%endif
+%if %1 > 2
+ pop r12
+%endif
+%if %1 > 1
+ pop r11
+%endif
+ pop r10
+%endmacro
+
+%imacro push_xmm 1
+%endmacro
+
+%imacro pop_xmm 1
+%endmacro
+
+%endif
+
+%endif
+
+; --------------------------------------------------------------------------
+; Defines picked up from the C headers
+;
+%include "jsimdcfg.inc"
+
+; --------------------------------------------------------------------------
diff --git a/simd/nasm_lt.sh b/simd/nasm_lt.sh
deleted file mode 100755
index 817be16..0000000
--- a/simd/nasm_lt.sh
+++ /dev/null
@@ -1,60 +0,0 @@
-#! /bin/sh
-command=""
-infile=""
-o_opt=no
-pic=no
-while [ $# -gt 0 ]; do
- case "$1" in
- --silent)
- exec > /dev/null
- ;;
- -DPIC|-fPIC|-fpic|-Kpic|-KPIC)
- if [ "$pic" != "yes" ] ; then
- command="$command -DPIC"
- pic=yes
- fi
- ;;
- -f|-fbin|-faout|-faoutb|-fcoff|-felf|-felf64|-fas86| \
- -fobj|-fwin32|-fwin64|-frdf|-fieee|-fmacho|-fmacho64)
- # it's a file format specifier for nasm.
- command="$command $1"
- ;;
- -f*)
- # maybe a code-generation flag for gcc.
- ;;
- -[Ii]*)
- incdir=`echo "$1" | sed 's/^-[Ii]//'`
- if [ "x$incdir" = x -a "x$2" != x ] ; then
- case "$2" in
- -*) ;;
- *) incdir="$2"; shift;;
- esac
- fi
- if [ "x$incdir" != x ] ; then
- # In the case of NASM, the trailing slash is necessary.
- incdir=`echo "$incdir" | sed 's%/*$%/%'`
- command="$command -I$incdir"
- fi
- ;;
- -o*)
- o_opt=yes
- command="$command $1"
- ;;
- *.asm)
- infile=$1
- command="$command $1"
- ;;
- *)
- command="$command $1"
- ;;
- esac
- shift
-done
-if [ "$o_opt" != yes ] ; then
- # By default, NASM creates an output file
- # in the same directory as the input file.
- outfile="-o `echo $infile | sed -e 's%^.*/%%' -e 's%\.[^.]*$%%'`.o"
- command="$command $outfile"
-fi
-echo $command
-exec $command
diff --git a/simd/jccolext-altivec.c b/simd/powerpc/jccolext-altivec.c
similarity index 100%
rename from simd/jccolext-altivec.c
rename to simd/powerpc/jccolext-altivec.c
diff --git a/simd/jccolor-altivec.c b/simd/powerpc/jccolor-altivec.c
similarity index 100%
rename from simd/jccolor-altivec.c
rename to simd/powerpc/jccolor-altivec.c
diff --git a/simd/jcgray-altivec.c b/simd/powerpc/jcgray-altivec.c
similarity index 100%
rename from simd/jcgray-altivec.c
rename to simd/powerpc/jcgray-altivec.c
diff --git a/simd/jcgryext-altivec.c b/simd/powerpc/jcgryext-altivec.c
similarity index 100%
rename from simd/jcgryext-altivec.c
rename to simd/powerpc/jcgryext-altivec.c
diff --git a/simd/jcsample-altivec.c b/simd/powerpc/jcsample-altivec.c
similarity index 100%
rename from simd/jcsample-altivec.c
rename to simd/powerpc/jcsample-altivec.c
diff --git a/simd/jcsample.h b/simd/powerpc/jcsample.h
similarity index 100%
rename from simd/jcsample.h
rename to simd/powerpc/jcsample.h
diff --git a/simd/jdcolext-altivec.c b/simd/powerpc/jdcolext-altivec.c
similarity index 100%
rename from simd/jdcolext-altivec.c
rename to simd/powerpc/jdcolext-altivec.c
diff --git a/simd/jdcolor-altivec.c b/simd/powerpc/jdcolor-altivec.c
similarity index 100%
rename from simd/jdcolor-altivec.c
rename to simd/powerpc/jdcolor-altivec.c
diff --git a/simd/jdmerge-altivec.c b/simd/powerpc/jdmerge-altivec.c
similarity index 100%
rename from simd/jdmerge-altivec.c
rename to simd/powerpc/jdmerge-altivec.c
diff --git a/simd/jdmrgext-altivec.c b/simd/powerpc/jdmrgext-altivec.c
similarity index 100%
rename from simd/jdmrgext-altivec.c
rename to simd/powerpc/jdmrgext-altivec.c
diff --git a/simd/jdsample-altivec.c b/simd/powerpc/jdsample-altivec.c
similarity index 100%
rename from simd/jdsample-altivec.c
rename to simd/powerpc/jdsample-altivec.c
diff --git a/simd/jfdctfst-altivec.c b/simd/powerpc/jfdctfst-altivec.c
similarity index 100%
rename from simd/jfdctfst-altivec.c
rename to simd/powerpc/jfdctfst-altivec.c
diff --git a/simd/jfdctint-altivec.c b/simd/powerpc/jfdctint-altivec.c
similarity index 100%
rename from simd/jfdctint-altivec.c
rename to simd/powerpc/jfdctint-altivec.c
diff --git a/simd/jidctfst-altivec.c b/simd/powerpc/jidctfst-altivec.c
similarity index 100%
rename from simd/jidctfst-altivec.c
rename to simd/powerpc/jidctfst-altivec.c
diff --git a/simd/jidctint-altivec.c b/simd/powerpc/jidctint-altivec.c
similarity index 100%
rename from simd/jidctint-altivec.c
rename to simd/powerpc/jidctint-altivec.c
diff --git a/simd/jquanti-altivec.c b/simd/powerpc/jquanti-altivec.c
similarity index 100%
rename from simd/jquanti-altivec.c
rename to simd/powerpc/jquanti-altivec.c
diff --git a/simd/jsimd_powerpc.c b/simd/powerpc/jsimd.c
similarity index 98%
rename from simd/jsimd_powerpc.c
rename to simd/powerpc/jsimd.c
index 47dd746..56e4d22 100644
--- a/simd/jsimd_powerpc.c
+++ b/simd/powerpc/jsimd.c
@@ -20,12 +20,12 @@
#endif
#define JPEG_INTERNALS
-#include "../jinclude.h"
-#include "../jpeglib.h"
+#include "../../jinclude.h"
+#include "../../jpeglib.h"
+#include "../../jsimd.h"
+#include "../../jdct.h"
+#include "../../jsimddct.h"
#include "../jsimd.h"
-#include "../jdct.h"
-#include "../jsimddct.h"
-#include "jsimd.h"
#include <stdio.h>
#include <string.h>
diff --git a/simd/jsimd_altivec.h b/simd/powerpc/jsimd_altivec.h
similarity index 96%
rename from simd/jsimd_altivec.h
rename to simd/powerpc/jsimd_altivec.h
index 62dbc5c..e204997 100644
--- a/simd/jsimd_altivec.h
+++ b/simd/powerpc/jsimd_altivec.h
@@ -21,12 +21,12 @@
*/
#define JPEG_INTERNALS
-#include "../jinclude.h"
-#include "../jpeglib.h"
+#include "../../jinclude.h"
+#include "../../jpeglib.h"
+#include "../../jsimd.h"
+#include "../../jdct.h"
+#include "../../jsimddct.h"
#include "../jsimd.h"
-#include "../jdct.h"
-#include "../jsimddct.h"
-#include "jsimd.h"
#include <altivec.h>
diff --git a/simd/x86_64/jccolext-avx2.asm b/simd/x86_64/jccolext-avx2.asm
new file mode 100644
index 0000000..a7e977a
--- /dev/null
+++ b/simd/x86_64/jccolext-avx2.asm
@@ -0,0 +1,562 @@
+;
+; jccolext.asm - colorspace conversion (64-bit AVX2)
+;
+; Copyright (C) 2009, 2016, D. R. Commander.
+; Copyright (C) 2015, Intel Corporation.
+;
+; Based on the x86 SIMD extension for IJG JPEG library
+; Copyright (C) 1999-2006, MIYASAKA Masaru.
+; For conditions of distribution and use, see copyright notice in jsimdext.inc
+;
+; This file should be assembled with NASM (Netwide Assembler),
+; can *not* be assembled with Microsoft's MASM or any compatible
+; assembler (including Borland's Turbo Assembler).
+; NASM is available from http://nasm.sourceforge.net/ or
+; http://sourceforge.net/project/showfiles.php?group_id=6208
+;
+; [TAB8]
+
+%include "jcolsamp.inc"
+
+; --------------------------------------------------------------------------
+;
+; Convert some rows of samples to the output colorspace.
+;
+; GLOBAL(void)
+; jsimd_rgb_ycc_convert_avx2 (JDIMENSION img_width,
+; JSAMPARRAY input_buf, JSAMPIMAGE output_buf,
+; JDIMENSION output_row, int num_rows);
+;
+
+; r10d = JDIMENSION img_width
+; r11 = JSAMPARRAY input_buf
+; r12 = JSAMPIMAGE output_buf
+; r13d = JDIMENSION output_row
+; r14d = int num_rows
+
+%define wk(i) rbp-(WK_NUM-(i))*SIZEOF_YMMWORD ; ymmword wk[WK_NUM]
+%define WK_NUM 8
+
+ align 32
+
+ global EXTN(jsimd_rgb_ycc_convert_avx2)
+
+EXTN(jsimd_rgb_ycc_convert_avx2):
+ push rbp
+ mov rax, rsp ; rax = original rbp
+ sub rsp, byte 4
+ and rsp, byte (-SIZEOF_YMMWORD) ; align to 256 bits
+ mov [rsp], rax
+ mov rbp, rsp ; rbp = aligned rbp
+ lea rsp, [wk(0)]
+ collect_args 5
+ push rbx
+
+ mov ecx, r10d
+ test rcx, rcx
+ jz near .return
+
+ push rcx
+
+ mov rsi, r12
+ mov ecx, r13d
+ mov rdi, JSAMPARRAY [rsi+0*SIZEOF_JSAMPARRAY]
+ mov rbx, JSAMPARRAY [rsi+1*SIZEOF_JSAMPARRAY]
+ mov rdx, JSAMPARRAY [rsi+2*SIZEOF_JSAMPARRAY]
+ lea rdi, [rdi+rcx*SIZEOF_JSAMPROW]
+ lea rbx, [rbx+rcx*SIZEOF_JSAMPROW]
+ lea rdx, [rdx+rcx*SIZEOF_JSAMPROW]
+
+ pop rcx
+
+ mov rsi, r11
+ mov eax, r14d
+ test rax, rax
+ jle near .return
+.rowloop:
+ push rdx
+ push rbx
+ push rdi
+ push rsi
+ push rcx ; col
+
+ mov rsi, JSAMPROW [rsi] ; inptr
+ mov rdi, JSAMPROW [rdi] ; outptr0
+ mov rbx, JSAMPROW [rbx] ; outptr1
+ mov rdx, JSAMPROW [rdx] ; outptr2
+
+ cmp rcx, byte SIZEOF_YMMWORD
+ jae near .columnloop
+
+%if RGB_PIXELSIZE == 3 ; ---------------
+
+.column_ld1:
+ push rax
+ push rdx
+ lea rcx, [rcx+rcx*2] ; imul ecx,RGB_PIXELSIZE
+ test cl, SIZEOF_BYTE
+ jz short .column_ld2
+ sub rcx, byte SIZEOF_BYTE
+ movzx rax, BYTE [rsi+rcx]
+.column_ld2:
+ test cl, SIZEOF_WORD
+ jz short .column_ld4
+ sub rcx, byte SIZEOF_WORD
+ movzx rdx, WORD [rsi+rcx]
+ shl rax, WORD_BIT
+ or rax, rdx
+.column_ld4:
+ vmovd xmmA, eax
+ pop rdx
+ pop rax
+ test cl, SIZEOF_DWORD
+ jz short .column_ld8
+ sub rcx, byte SIZEOF_DWORD
+ vmovd xmmF, XMM_DWORD [rsi+rcx]
+ vpslldq xmmA, xmmA, SIZEOF_DWORD
+ vpor xmmA, xmmA, xmmF
+.column_ld8:
+ test cl, SIZEOF_MMWORD
+ jz short .column_ld16
+ sub rcx, byte SIZEOF_MMWORD
+ vmovq xmmB, XMM_MMWORD [rsi+rcx]
+ vpslldq xmmA, xmmA, SIZEOF_MMWORD
+ vpor xmmA, xmmA, xmmB
+.column_ld16:
+ test cl, SIZEOF_XMMWORD
+ jz short .column_ld32
+ sub rcx, byte SIZEOF_XMMWORD
+ vmovdqu xmmB, XMM_MMWORD [rsi+rcx]
+ vperm2i128 ymmA, ymmA, ymmA, 1
+ vpor ymmA, ymmB
+.column_ld32:
+ test cl, SIZEOF_YMMWORD
+ jz short .column_ld64
+ sub rcx, byte SIZEOF_YMMWORD
+ vmovdqa ymmF, ymmA
+ vmovdqu ymmA, YMMWORD [rsi+0*SIZEOF_YMMWORD]
+.column_ld64:
+ test cl, 2*SIZEOF_YMMWORD
+ mov rcx, SIZEOF_YMMWORD
+ jz short .rgb_ycc_cnv
+ vmovdqa ymmB, ymmA
+ vmovdqu ymmA, YMMWORD [rsi+0*SIZEOF_YMMWORD]
+ vmovdqu ymmF, YMMWORD [rsi+1*SIZEOF_YMMWORD]
+ jmp short .rgb_ycc_cnv
+
+.columnloop:
+ vmovdqu ymmA, YMMWORD [rsi+0*SIZEOF_YMMWORD]
+ vmovdqu ymmF, YMMWORD [rsi+1*SIZEOF_YMMWORD]
+ vmovdqu ymmB, YMMWORD [rsi+2*SIZEOF_YMMWORD]
+
+.rgb_ycc_cnv:
+ ; ymmA=(00 10 20 01 11 21 02 12 22 03 13 23 04 14 24 05
+ ; 15 25 06 16 26 07 17 27 08 18 28 09 19 29 0A 1A)
+ ; ymmF=(2A 0B 1B 2B 0C 1C 2C 0D 1D 2D 0E 1E 2E 0F 1F 2F
+ ; 0G 1G 2G 0H 1H 2H 0I 1I 2I 0J 1J 2J 0K 1K 2K 0L)
+ ; ymmB=(1L 2L 0M 1M 2M 0N 1N 2N 0O 1O 2O 0P 1P 2P 0Q 1Q
+ ; 2Q 0R 1R 2R 0S 1S 2S 0T 1T 2T 0U 1U 2U 0V 1V 2V)
+
+ vmovdqu ymmC, ymmA
+ vinserti128 ymmA, ymmF, xmmA, 0 ; ymmA=(00 10 20 01 11 21 02 12 22 03 13 23 04 14 24 05
+ ; 0G 1G 2G 0H 1H 2H 0I 1I 2I 0J 1J 2J 0K 1K 2K 0L)
+ vinserti128 ymmC, ymmC, xmmB, 0 ; ymmC=(1L 2L 0M 1M 2M 0N 1N 2N 0O 1O 2O 0P 1P 2P 0Q 1Q
+ ; 15 25 06 16 26 07 17 27 08 18 28 09 19 29 0A 1A)
+ vinserti128 ymmB, ymmB, xmmF, 0 ; ymmB=(2A 0B 1B 2B 0C 1C 2C 0D 1D 2D 0E 1E 2E 0F 1F 2F
+ ; 2Q 0R 1R 2R 0S 1S 2S 0T 1T 2T 0U 1U 2U 0V 1V 2V)
+ vperm2i128 ymmF, ymmC, ymmC, 1 ; ymmF=(15 25 06 16 26 07 17 27 08 18 28 09 19 29 0A 1A
+ ; 1L 2L 0M 1M 2M 0N 1N 2N 0O 1O 2O 0P 1P 2P 0Q 1Q)
+
+ vmovdqa ymmG, ymmA
+ vpslldq ymmA, ymmA, 8 ; ymmA=(-- -- -- -- -- -- -- -- 00 10 20 01 11 21 02 12
+ ; 22 03 13 23 04 14 24 05 0G 1G 2G 0H 1H 2H 0I 1I)
+ vpsrldq ymmG, ymmG, 8 ; ymmG=(22 03 13 23 04 14 24 05 0G 1G 2G 0H 1H 2H 0I 1I
+ ; 2I 0J 1J 2J 0K 1K 2K 0L -- -- -- -- -- -- -- --)
+
+ vpunpckhbw ymmA, ymmA, ymmF ; ymmA=(00 08 10 18 20 28 01 09 11 19 21 29 02 0A 12 1A
+ ; 0G 0O 1G 1O 2G 2O 0H 0P 1H 1P 2H 2P 0I 0Q 1I 1Q)
+ vpslldq ymmF, ymmF, 8 ; ymmF=(-- -- -- -- -- -- -- -- 15 25 06 16 26 07 17 27
+ ; 08 18 28 09 19 29 0A 1A 1L 2L 0M 1M 2M 0N 1N 2N)
+
+ vpunpcklbw ymmG, ymmG, ymmB ; ymmG=(22 2A 03 0B 13 1B 23 2B 04 0C 14 1C 24 2C 05 0D
+ ; 2I 2Q 0J 0R 1J 1R 2J 2R 0K 0S 1K 1S 2K 2S 0L 0T)
+ vpunpckhbw ymmF, ymmF, ymmB ; ymmF=(15 1D 25 2D 06 0E 16 1E 26 2E 07 0F 17 1F 27 2F
+ ; 1L 1T 2L 2T 0M 0U 1M 1U 2M 2U 0N 0V 1N 1V 2N 2V)
+
+ vmovdqa ymmD, ymmA
+ vpslldq ymmA, ymmA, 8 ; ymmA=(-- -- -- -- -- -- -- -- 00 08 10 18 20 28 01 09
+ ; 11 19 21 29 02 0A 12 1A 0G 0O 1G 1O 2G 2O 0H 0P)
+ vpsrldq ymmD, ymmD, 8 ; ymmD=(11 19 21 29 02 0A 12 1A 0G 0O 1G 1O 2G 2O 0H 0P
+ ; 1H 1P 2H 2P 0I 0Q 1I 1Q -- -- -- -- -- -- -- --)
+
+ vpunpckhbw ymmA, ymmA, ymmG ; ymmA=(00 04 08 0C 10 14 18 1C 20 24 28 2C 01 05 09 0D
+ ; 0G 0K 0O 0S 1G 1K 1O 1S 2G 2K 2O 2S 0H 0L 0P 0T)
+ vpslldq ymmG, ymmG, 8 ; ymmG=(-- -- -- -- -- -- -- -- 22 2A 03 0B 13 1B 23 2B
+ ; 04 0C 14 1C 24 2C 05 0D 2I 2Q 0J 0R 1J 1R 2J 2R)
+
+ vpunpcklbw ymmD, ymmD, ymmF ; ymmD=(11 15 19 1D 21 25 29 2D 02 06 0A 0E 12 16 1A 1E
+ ; 1H 1L 1P 1T 2H 2L 2P 2T 0I 0M 0Q 0U 1I 1M 1Q 1U)
+ vpunpckhbw ymmG, ymmG, ymmF ; ymmG=(22 26 2A 2E 03 07 0B 0F 13 17 1B 1F 23 27 2B 2F
+ ; 2I 2M 2Q 2U 0J 0N 0R 0V 1J 1N 1R 1V 2J 2N 2R 2V)
+
+ vmovdqa ymmE, ymmA
+ vpslldq ymmA, ymmA, 8 ; ymmA=(-- -- -- -- -- -- -- -- 00 04 08 0C 10 14 18 1C
+ ; 20 24 28 2C 01 05 09 0D 0G 0K 0O 0S 1G 1K 1O 1S)
+ vpsrldq ymmE, ymmE, 8 ; ymmE=(20 24 28 2C 01 05 09 0D 0G 0K 0O 0S 1G 1K 1O 1S
+ ; 2G 2K 2O 2S 0H 0L 0P 0T -- -- -- -- -- -- -- --)
+
+ vpunpckhbw ymmA, ymmA, ymmD ; ymmA=(00 02 04 06 08 0A 0C 0E 10 12 14 16 18 1A 1C 1E
+ ; 0G 0I 0K 0M 0O 0Q 0S 0U 1G 1I 1K 1M 1O 1Q 1S 1U)
+ vpslldq ymmD, ymmD, 8 ; ymmD=(-- -- -- -- -- -- -- -- 11 15 19 1D 21 25 29 2D
+ ; 02 06 0A 0E 12 16 1A 1E 1H 1L 1P 1T 2H 2L 2P 2T)
+
+ vpunpcklbw ymmE, ymmE, ymmG ; ymmE=(20 22 24 26 28 2A 2C 2E 01 03 05 07 09 0B 0D 0F
+ ; 2G 2I 2K 2M 2O 2Q 2S 2U 0H 0J 0L 0N 0P 0R 0T 0V)
+ vpunpckhbw ymmD, ymmD, ymmG ; ymmD=(11 13 15 17 19 1B 1D 1F 21 23 25 27 29 2B 2D 2F
+ ; 1H 1J 1L 1N 1P 1R 1T 1V 2H 2J 2L 2N 2P 2R 2T 2V)
+
+ vpxor ymmH, ymmH, ymmH
+
+ vmovdqa ymmC, ymmA
+ vpunpcklbw ymmA, ymmA, ymmH ; ymmA=(00 02 04 06 08 0A 0C 0E 0G 0I 0K 0M 0O 0Q 0S 0U)
+ vpunpckhbw ymmC, ymmC, ymmH ; ymmC=(10 12 14 16 18 1A 1C 1E 1G 1I 1K 1M 1O 1Q 1S 1U)
+
+ vmovdqa ymmB, ymmE
+ vpunpcklbw ymmE, ymmE, ymmH ; ymmE=(20 22 24 26 28 2A 2C 2E 2G 2I 2K 2M 2O 2Q 2S 2U)
+ vpunpckhbw ymmB, ymmB, ymmH ; ymmB=(01 03 05 07 09 0B 0D 0F 0H 0J 0L 0N 0P 0R 0T 0V)
+
+ vmovdqa ymmF, ymmD
+ vpunpcklbw ymmD, ymmD, ymmH ; ymmD=(11 13 15 17 19 1B 1D 1F 1H 1J 1L 1N 1P 1R 1T 1V)
+ vpunpckhbw ymmF, ymmF, ymmH ; ymmF=(21 23 25 27 29 2B 2D 2F 2H 2J 2L 2N 2P 2R 2T 2V)
+
+%else ; RGB_PIXELSIZE == 4 ; -----------
+
+.column_ld1:
+ test cl, SIZEOF_XMMWORD/16
+ jz short .column_ld2
+ sub rcx, byte SIZEOF_XMMWORD/16
+ vmovd xmmA, XMM_DWORD [rsi+rcx*RGB_PIXELSIZE]
+.column_ld2:
+ test cl, SIZEOF_XMMWORD/8
+ jz short .column_ld4
+ sub rcx, byte SIZEOF_XMMWORD/8
+ vmovq xmmF, XMM_MMWORD [rsi+rcx*RGB_PIXELSIZE]
+ vpslldq xmmA, xmmA, SIZEOF_MMWORD
+ vpor xmmA, xmmA, xmmF
+.column_ld4:
+ test cl, SIZEOF_XMMWORD/4
+ jz short .column_ld8
+ sub rcx, byte SIZEOF_XMMWORD/4
+ vmovdqa xmmF, xmmA
+ vperm2i128 ymmF, ymmF, ymmF, 1
+ vmovdqu xmmA, XMMWORD [rsi+rcx*RGB_PIXELSIZE]
+ vpor ymmA, ymmA, ymmF
+.column_ld8:
+ test cl, SIZEOF_XMMWORD/2
+ jz short .column_ld16
+ sub rcx, byte SIZEOF_XMMWORD/2
+ vmovdqa ymmF, ymmA
+ vmovdqu ymmA, YMMWORD [rsi+rcx*RGB_PIXELSIZE]
+.column_ld16:
+ test cl, SIZEOF_XMMWORD
+ mov rcx, SIZEOF_YMMWORD
+ jz short .rgb_ycc_cnv
+ vmovdqa ymmE, ymmA
+ vmovdqa ymmH, ymmF
+ vmovdqu ymmA, YMMWORD [rsi+0*SIZEOF_YMMWORD]
+ vmovdqu ymmF, YMMWORD [rsi+1*SIZEOF_YMMWORD]
+ jmp short .rgb_ycc_cnv
+
+.columnloop:
+ vmovdqu ymmA, YMMWORD [rsi+0*SIZEOF_YMMWORD]
+ vmovdqu ymmF, YMMWORD [rsi+1*SIZEOF_YMMWORD]
+ vmovdqu ymmE, YMMWORD [rsi+2*SIZEOF_YMMWORD]
+ vmovdqu ymmH, YMMWORD [rsi+3*SIZEOF_YMMWORD]
+
+.rgb_ycc_cnv:
+ ; ymmA=(00 10 20 30 01 11 21 31 02 12 22 32 03 13 23 33
+ ; 04 14 24 34 05 15 25 35 06 16 26 36 07 17 27 37)
+ ; ymmF=(08 18 28 38 09 19 29 39 0A 1A 2A 3A 0B 1B 2B 3B
+ ; 0C 1C 2C 3C 0D 1D 2D 3D 0E 1E 2E 3E 0F 1F 2F 3F)
+ ; ymmE=(0G 1G 2G 3G 0H 1H 2H 3H 0I 1I 2I 3I 0J 1J 2J 3J
+ ; 0K 1K 2K 3K 0L 1L 2L 3L 0M 1M 2M 3M 0N 1N 2N 3N)
+ ; ymmH=(0O 1O 2O 3O 0P 1P 2P 3P 0Q 1Q 2Q 3Q 0R 1R 2R 3R
+ ; 0S 1S 2S 3S 0T 1T 2T 3T 0U 1U 2U 3U 0V 1V 2V 3V)
+
+ vmovdqa ymmB, ymmA
+ vinserti128 ymmA, ymmA, xmmE, 1 ; ymmA=(00 10 20 30 01 11 21 31 02 12 22 32 03 13 23 33
+ ; 0G 1G 2G 3G 0H 1H 2H 3H 0I 1I 2I 3I 0J 1J 2J 3J)
+ vperm2i128 ymmE, ymmB, ymmE, 0x31 ; ymmE=(04 14 24 34 05 15 25 35 06 16 26 36 07 17 27 37
+ ; 0K 1K 2K 3K 0L 1L 2L 3L 0M 1M 2M 3M 0N 1N 2N 3N)
+
+ vmovdqa ymmB, ymmF
+ vinserti128 ymmF, ymmF, xmmH, 1 ; ymmF=(08 18 28 38 09 19 29 39 0A 1A 2A 3A 0B 1B 2B 3B
+ ; 0O 1O 2O 3O 0P 1P 2P 3P 0Q 1Q 2Q 3Q 0R 1R 2R 3R)
+ vperm2i128 ymmH, ymmB, ymmH, 0x31 ; ymmH=(0C 1C 2C 3C 0D 1D 2D 3D 0E 1E 2E 3E 0F 1F 2F 3F
+ ; 0S 1S 2S 3S 0T 1T 2T 3T 0U 1U 2U 3U 0V 1V 2V 3V)
+
+ vmovdqa ymmD, ymmA
+ vpunpcklbw ymmA, ymmA, ymmE ; ymmA=(00 04 10 14 20 24 30 34 01 05 11 15 21 25 31 35
+ ; 0G 0K 1G 1K 2G 2K 3G 3K 0H 0L 1H 1L 2H 2L 3H 3L)
+ vpunpckhbw ymmD, ymmD, ymmE ; ymmD=(02 06 12 16 22 26 32 36 03 07 13 17 23 27 33 37
+ ; 0I 0M 1I 1M 2I 2M 3I 3M 0J 0N 1J 1N 2J 2N 3J 3N)
+
+ vmovdqa ymmC, ymmF
+ vpunpcklbw ymmF, ymmF, ymmH ; ymmF=(08 0C 18 1C 28 2C 38 3C 09 0D 19 1D 29 2D 39 3D
+ ; 0O 0S 1O 1S 2O 2S 3O 3S 0P 0T 1P 1T 2P 2T 3P 3T)
+ vpunpckhbw ymmC, ymmC, ymmH ; ymmC=(0A 0E 1A 1E 2A 2E 3A 3E 0B 0F 1B 1F 2B 2F 3B 3F
+ ; 0Q 0U 1Q 1U 2Q 2U 3Q 3U 0R 0V 1R 1V 2R 2V 3R 3V)
+
+ vmovdqa ymmB, ymmA
+ vpunpcklwd ymmA, ymmA, ymmF ; ymmA=(00 04 08 0C 10 14 18 1C 20 24 28 2C 30 34 38 3C
+ ; 0G 0K 0O 0S 1G 1K 1O 1S 2G 2K 2O 2S 3G 3K 3O 3S)
+ vpunpckhwd ymmB, ymmB, ymmF ; ymmB=(01 05 09 0D 11 15 19 1D 21 25 29 2D 31 35 39 3D
+ ; 0H 0L 0P 0T 1H 1L 1P 1T 2H 2L 2P 2T 3H 3L 3P 3T)
+
+ vmovdqa ymmG, ymmD
+ vpunpcklwd ymmD, ymmD, ymmC ; ymmD=(02 06 0A 0E 12 16 1A 1E 22 26 2A 2E 32 36 3A 3E
+ ; 0I 0M 0Q 0U 1I 1M 1Q 1U 2I 2M 2Q 2U 3I 3M 3Q 3U)
+ vpunpckhwd ymmG, ymmG, ymmC ; ymmG=(03 07 0B 0F 13 17 1B 1F 23 27 2B 2F 33 37 3B 3F
+ ; 0J 0N 0R 0V 1J 1N 1R 1V 2J 2N 2R 2V 3J 3N 3R 3V)
+
+ vmovdqa ymmE, ymmA
+ vpunpcklbw ymmA, ymmA, ymmD ; ymmA=(00 02 04 06 08 0A 0C 0E 10 12 14 16 18 1A 1C 1E
+ ; 0G 0I 0K 0M 0O 0Q 0S 0U 1G 1I 1K 1M 1O 1Q 1S 1U)
+ vpunpckhbw ymmE, ymmE, ymmD ; ymmE=(20 22 24 26 28 2A 2C 2E 30 32 34 36 38 3A 3C 3E
+ ; 2G 2I 2K 2M 2O 2Q 2S 2U 3G 3I 3K 3M 3O 3Q 3S 3U)
+
+ vmovdqa ymmH, ymmB
+ vpunpcklbw ymmB, ymmB, ymmG ; ymmB=(01 03 05 07 09 0B 0D 0F 11 13 15 17 19 1B 1D 1F
+ ; 0H 0J 0L 0N 0P 0R 0T 0V 1H 1J 1L 1N 1P 1R 1T 1V)
+ vpunpckhbw ymmH, ymmH, ymmG ; ymmH=(21 23 25 27 29 2B 2D 2F 31 33 35 37 39 3B 3D 3F
+ ; 2H 2J 2L 2N 2P 2R 2T 2V 3H 3J 3L 3N 3P 3R 3T 3V)
+
+ vpxor ymmF, ymmF, ymmF
+
+ vmovdqa ymmC, ymmA
+ vpunpcklbw ymmA, ymmA, ymmF ; ymmA=(00 02 04 06 08 0A 0C 0E 0G 0I 0K 0M 0O 0Q 0S 0U)
+ vpunpckhbw ymmC, ymmC, ymmF ; ymmC=(10 12 14 16 18 1A 1C 1E 1G 1I 1K 1M 1O 1Q 1S 1U)
+
+ vmovdqa ymmD, ymmB
+ vpunpcklbw ymmB, ymmB, ymmF ; ymmB=(01 03 05 07 09 0B 0D 0F 0H 0J 0L 0N 0P 0R 0T 0V)
+ vpunpckhbw ymmD, ymmD, ymmF ; ymmD=(11 13 15 17 19 1B 1D 1F 1H 1J 1L 1N 1P 1R 1T 1V)
+
+ vmovdqa ymmG, ymmE
+ vpunpcklbw ymmE, ymmE, ymmF ; ymmE=(20 22 24 26 28 2A 2C 2E 2G 2I 2K 2M 2O 2Q 2S 2U)
+ vpunpckhbw ymmG, ymmG, ymmF ; ymmG=(30 32 34 36 38 3A 3C 3E 3G 3I 3K 3M 3O 3Q 3S 3U)
+
+ vpunpcklbw ymmF, ymmF, ymmH
+ vpunpckhbw ymmH, ymmH, ymmH
+ vpsrlw ymmF, ymmF, BYTE_BIT ; ymmF=(21 23 25 27 29 2B 2D 2F 2H 2J 2L 2N 2P 2R 2T 2V)
+ vpsrlw ymmH, ymmH, BYTE_BIT ; ymmH=(31 33 35 37 39 3B 3D 3F 3H 3J 3L 3N 3P 3R 3T 3V)
+
+%endif ; RGB_PIXELSIZE ; ---------------
+
+ ; ymm0=R(02468ACEGIKMOQSU)=RE, ymm2=G(02468ACEGIKMOQSU)=GE, ymm4=B(02468ACEGIKMOQSU)=BE
+ ; ymm1=R(13579BDFHJLNPRTV)=RO, ymm3=G(13579BDFHJLNPRTV)=GO, ymm5=B(13579BDFHJLNPRTV)=BO
+
+ ; (Original)
+ ; Y = 0.29900 * R + 0.58700 * G + 0.11400 * B
+ ; Cb = -0.16874 * R - 0.33126 * G + 0.50000 * B + CENTERJSAMPLE
+ ; Cr = 0.50000 * R - 0.41869 * G - 0.08131 * B + CENTERJSAMPLE
+ ;
+ ; (This implementation)
+ ; Y = 0.29900 * R + 0.33700 * G + 0.11400 * B + 0.25000 * G
+ ; Cb = -0.16874 * R - 0.33126 * G + 0.50000 * B + CENTERJSAMPLE
+ ; Cr = 0.50000 * R - 0.41869 * G - 0.08131 * B + CENTERJSAMPLE
+
+ vmovdqa YMMWORD [wk(0)], ymm0 ; wk(0)=RE
+ vmovdqa YMMWORD [wk(1)], ymm1 ; wk(1)=RO
+ vmovdqa YMMWORD [wk(2)], ymm4 ; wk(2)=BE
+ vmovdqa YMMWORD [wk(3)], ymm5 ; wk(3)=BO
+
+ vmovdqa ymm6, ymm1
+ vpunpcklwd ymm1, ymm1, ymm3
+ vpunpckhwd ymm6, ymm6, ymm3
+ vmovdqa ymm7, ymm1
+ vmovdqa ymm4, ymm6
+ vpmaddwd ymm1, ymm1, [rel PW_F0299_F0337] ; ymm1=ROL*FIX(0.299)+GOL*FIX(0.337)
+ vpmaddwd ymm6, ymm6, [rel PW_F0299_F0337] ; ymm6=ROH*FIX(0.299)+GOH*FIX(0.337)
+ vpmaddwd ymm7, ymm7, [rel PW_MF016_MF033] ; ymm7=ROL*-FIX(0.168)+GOL*-FIX(0.331)
+ vpmaddwd ymm4, ymm4, [rel PW_MF016_MF033] ; ymm4=ROH*-FIX(0.168)+GOH*-FIX(0.331)
+
+ vmovdqa YMMWORD [wk(4)], ymm1 ; wk(4)=ROL*FIX(0.299)+GOL*FIX(0.337)
+ vmovdqa YMMWORD [wk(5)], ymm6 ; wk(5)=ROH*FIX(0.299)+GOH*FIX(0.337)
+
+ vpxor ymm1, ymm1, ymm1
+ vpxor ymm6, ymm6, ymm6
+ vpunpcklwd ymm1, ymm1, ymm5 ; ymm1=BOL
+ vpunpckhwd ymm6, ymm6, ymm5 ; ymm6=BOH
+ vpsrld ymm1, ymm1, 1 ; ymm1=BOL*FIX(0.500)
+ vpsrld ymm6, ymm6, 1 ; ymm6=BOH*FIX(0.500)
+
+ vmovdqa ymm5, [rel PD_ONEHALFM1_CJ] ; ymm5=[PD_ONEHALFM1_CJ]
+
+ vpaddd ymm7, ymm7, ymm1
+ vpaddd ymm4, ymm4, ymm6
+ vpaddd ymm7, ymm7, ymm5
+ vpaddd ymm4, ymm4, ymm5
+ vpsrld ymm7, ymm7, SCALEBITS ; ymm7=CbOL
+ vpsrld ymm4, ymm4, SCALEBITS ; ymm4=CbOH
+ vpackssdw ymm7, ymm7, ymm4 ; ymm7=CbO
+
+ vmovdqa ymm1, YMMWORD [wk(2)] ; ymm1=BE
+
+ vmovdqa ymm6, ymm0
+ vpunpcklwd ymm0, ymm0, ymm2
+ vpunpckhwd ymm6, ymm6, ymm2
+ vmovdqa ymm5, ymm0
+ vmovdqa ymm4, ymm6
+ vpmaddwd ymm0, ymm0, [rel PW_F0299_F0337] ; ymm0=REL*FIX(0.299)+GEL*FIX(0.337)
+ vpmaddwd ymm6, ymm6, [rel PW_F0299_F0337] ; ymm6=REH*FIX(0.299)+GEH*FIX(0.337)
+ vpmaddwd ymm5, ymm5, [rel PW_MF016_MF033] ; ymm5=REL*-FIX(0.168)+GEL*-FIX(0.331)
+ vpmaddwd ymm4, ymm4, [rel PW_MF016_MF033] ; ymm4=REH*-FIX(0.168)+GEH*-FIX(0.331)
+
+ vmovdqa YMMWORD [wk(6)], ymm0 ; wk(6)=REL*FIX(0.299)+GEL*FIX(0.337)
+ vmovdqa YMMWORD [wk(7)], ymm6 ; wk(7)=REH*FIX(0.299)+GEH*FIX(0.337)
+
+ vpxor ymm0, ymm0, ymm0
+ vpxor ymm6, ymm6, ymm6
+ vpunpcklwd ymm0, ymm0, ymm1 ; ymm0=BEL
+ vpunpckhwd ymm6, ymm6, ymm1 ; ymm6=BEH
+ vpsrld ymm0, ymm0, 1 ; ymm0=BEL*FIX(0.500)
+ vpsrld ymm6, ymm6, 1 ; ymm6=BEH*FIX(0.500)
+
+
+ vmovdqa ymm1, [rel PD_ONEHALFM1_CJ] ; ymm1=[PD_ONEHALFM1_CJ]
+
+ vpaddd ymm5, ymm5, ymm0
+ vpaddd ymm4, ymm4, ymm6
+ vpaddd ymm5, ymm5, ymm1
+ vpaddd ymm4, ymm4, ymm1
+ vpsrld ymm5, ymm5, SCALEBITS ; ymm5=CbEL
+ vpsrld ymm4, ymm4, SCALEBITS ; ymm4=CbEH
+ vpackssdw ymm5, ymm5, ymm4 ; ymm5=CbE
+
+ vpsllw ymm7, ymm7, BYTE_BIT
+ vpor ymm5, ymm5, ymm7 ; ymm5=Cb
+ vmovdqu YMMWORD [rbx], ymm5 ; Save Cb
+
+ vmovdqa ymm0, YMMWORD [wk(3)] ; ymm0=BO
+ vmovdqa ymm6, YMMWORD [wk(2)] ; ymm6=BE
+ vmovdqa ymm1, YMMWORD [wk(1)] ; ymm1=RO
+
+ vmovdqa ymm4, ymm0
+ vpunpcklwd ymm0, ymm0, ymm3
+ vpunpckhwd ymm4, ymm4, ymm3
+ vmovdqa ymm7, ymm0
+ vmovdqa ymm5, ymm4
+ vpmaddwd ymm0, ymm0, [rel PW_F0114_F0250] ; ymm0=BOL*FIX(0.114)+GOL*FIX(0.250)
+ vpmaddwd ymm4, ymm4, [rel PW_F0114_F0250] ; ymm4=BOH*FIX(0.114)+GOH*FIX(0.250)
+ vpmaddwd ymm7, ymm7, [rel PW_MF008_MF041] ; ymm7=BOL*-FIX(0.081)+GOL*-FIX(0.418)
+ vpmaddwd ymm5, ymm5, [rel PW_MF008_MF041] ; ymm5=BOH*-FIX(0.081)+GOH*-FIX(0.418)
+
+ vmovdqa ymm3, [rel PD_ONEHALF] ; ymm3=[PD_ONEHALF]
+
+ vpaddd ymm0, ymm0, YMMWORD [wk(4)]
+ vpaddd ymm4, ymm4, YMMWORD [wk(5)]
+ vpaddd ymm0, ymm0, ymm3
+ vpaddd ymm4, ymm4, ymm3
+ vpsrld ymm0, ymm0, SCALEBITS ; ymm0=YOL
+ vpsrld ymm4, ymm4, SCALEBITS ; ymm4=YOH
+ vpackssdw ymm0, ymm0, ymm4 ; ymm0=YO
+
+ vpxor ymm3, ymm3, ymm3
+ vpxor ymm4, ymm4, ymm4
+ vpunpcklwd ymm3, ymm3, ymm1 ; ymm3=ROL
+ vpunpckhwd ymm4, ymm4, ymm1 ; ymm4=ROH
+ vpsrld ymm3, ymm3, 1 ; ymm3=ROL*FIX(0.500)
+ vpsrld ymm4, ymm4, 1 ; ymm4=ROH*FIX(0.500)
+
+ vmovdqa ymm1, [rel PD_ONEHALFM1_CJ] ; ymm1=[PD_ONEHALFM1_CJ]
+
+ vpaddd ymm7, ymm7, ymm3
+ vpaddd ymm5, ymm5, ymm4
+ vpaddd ymm7, ymm7, ymm1
+ vpaddd ymm5, ymm5, ymm1
+ vpsrld ymm7, ymm7, SCALEBITS ; ymm7=CrOL
+ vpsrld ymm5, ymm5, SCALEBITS ; ymm5=CrOH
+ vpackssdw ymm7, ymm7, ymm5 ; ymm7=CrO
+
+ vmovdqa ymm3, YMMWORD [wk(0)] ; ymm3=RE
+
+ vmovdqa ymm4, ymm6
+ vpunpcklwd ymm6, ymm6, ymm2
+ vpunpckhwd ymm4, ymm4, ymm2
+ vmovdqa ymm1, ymm6
+ vmovdqa ymm5, ymm4
+ vpmaddwd ymm6, ymm6, [rel PW_F0114_F0250] ; ymm6=BEL*FIX(0.114)+GEL*FIX(0.250)
+ vpmaddwd ymm4, ymm4, [rel PW_F0114_F0250] ; ymm4=BEH*FIX(0.114)+GEH*FIX(0.250)
+ vpmaddwd ymm1, ymm1, [rel PW_MF008_MF041] ; ymm1=BEL*-FIX(0.081)+GEL*-FIX(0.418)
+ vpmaddwd ymm5, ymm5, [rel PW_MF008_MF041] ; ymm5=BEH*-FIX(0.081)+GEH*-FIX(0.418)
+
+ vmovdqa ymm2, [rel PD_ONEHALF] ; ymm2=[PD_ONEHALF]
+
+ vpaddd ymm6, ymm6, YMMWORD [wk(6)]
+ vpaddd ymm4, ymm4, YMMWORD [wk(7)]
+ vpaddd ymm6, ymm6, ymm2
+ vpaddd ymm4, ymm4, ymm2
+ vpsrld ymm6, ymm6, SCALEBITS ; ymm6=YEL
+ vpsrld ymm4, ymm4, SCALEBITS ; ymm4=YEH
+ vpackssdw ymm6, ymm6, ymm4 ; ymm6=YE
+
+ vpsllw ymm0, ymm0, BYTE_BIT
+ vpor ymm6, ymm6, ymm0 ; ymm6=Y
+ vmovdqu YMMWORD [rdi], ymm6 ; Save Y
+
+ vpxor ymm2, ymm2, ymm2
+ vpxor ymm4, ymm4, ymm4
+ vpunpcklwd ymm2, ymm2, ymm3 ; ymm2=REL
+ vpunpckhwd ymm4, ymm4, ymm3 ; ymm4=REH
+ vpsrld ymm2, ymm2, 1 ; ymm2=REL*FIX(0.500)
+ vpsrld ymm4, ymm4, 1 ; ymm4=REH*FIX(0.500)
+
+ vmovdqa ymm0, [rel PD_ONEHALFM1_CJ] ; ymm0=[PD_ONEHALFM1_CJ]
+
+ vpaddd ymm1, ymm1, ymm2
+ vpaddd ymm5, ymm5, ymm4
+ vpaddd ymm1, ymm1, ymm0
+ vpaddd ymm5, ymm5, ymm0
+ vpsrld ymm1, ymm1, SCALEBITS ; ymm1=CrEL
+ vpsrld ymm5, ymm5, SCALEBITS ; ymm5=CrEH
+ vpackssdw ymm1, ymm1, ymm5 ; ymm1=CrE
+
+ vpsllw ymm7, ymm7, BYTE_BIT
+ vpor ymm1, ymm1, ymm7 ; ymm1=Cr
+ vmovdqu YMMWORD [rdx], ymm1 ; Save Cr
+
+ sub rcx, byte SIZEOF_YMMWORD
+ add rsi, RGB_PIXELSIZE*SIZEOF_YMMWORD ; inptr
+ add rdi, byte SIZEOF_YMMWORD ; outptr0
+ add rbx, byte SIZEOF_YMMWORD ; outptr1
+ add rdx, byte SIZEOF_YMMWORD ; outptr2
+ cmp rcx, byte SIZEOF_YMMWORD
+ jae near .columnloop
+ test rcx, rcx
+ jnz near .column_ld1
+
+ pop rcx ; col
+ pop rsi
+ pop rdi
+ pop rbx
+ pop rdx
+
+ add rsi, byte SIZEOF_JSAMPROW ; input_buf
+ add rdi, byte SIZEOF_JSAMPROW
+ add rbx, byte SIZEOF_JSAMPROW
+ add rdx, byte SIZEOF_JSAMPROW
+ dec rax ; num_rows
+ jg near .rowloop
+
+.return:
+ pop rbx
+ vzeroupper
+ uncollect_args 5
+ mov rsp, rbp ; rsp <- aligned rbp
+ pop rsp ; rsp <- original rbp
+ pop rbp
+ ret
+
+; For some reason, the OS X linker does not honor the request to align the
+; segment unless we do this.
+ align 32
diff --git a/simd/x86_64/jccolext-sse2.asm b/simd/x86_64/jccolext-sse2.asm
new file mode 100644
index 0000000..8fe072d
--- /dev/null
+++ b/simd/x86_64/jccolext-sse2.asm
@@ -0,0 +1,486 @@
+;
+; jccolext.asm - colorspace conversion (64-bit SSE2)
+;
+; Copyright (C) 2009, 2016, D. R. Commander.
+;
+; Based on the x86 SIMD extension for IJG JPEG library
+; Copyright (C) 1999-2006, MIYASAKA Masaru.
+; For conditions of distribution and use, see copyright notice in jsimdext.inc
+;
+; This file should be assembled with NASM (Netwide Assembler),
+; can *not* be assembled with Microsoft's MASM or any compatible
+; assembler (including Borland's Turbo Assembler).
+; NASM is available from http://nasm.sourceforge.net/ or
+; http://sourceforge.net/project/showfiles.php?group_id=6208
+;
+; [TAB8]
+
+%include "jcolsamp.inc"
+
+; --------------------------------------------------------------------------
+;
+; Convert some rows of samples to the output colorspace.
+;
+; GLOBAL(void)
+; jsimd_rgb_ycc_convert_sse2 (JDIMENSION img_width,
+; JSAMPARRAY input_buf, JSAMPIMAGE output_buf,
+; JDIMENSION output_row, int num_rows);
+;
+
+; r10d = JDIMENSION img_width
+; r11 = JSAMPARRAY input_buf
+; r12 = JSAMPIMAGE output_buf
+; r13d = JDIMENSION output_row
+; r14d = int num_rows
+
+%define wk(i) rbp-(WK_NUM-(i))*SIZEOF_XMMWORD ; xmmword wk[WK_NUM]
+%define WK_NUM 8
+
+ align 32
+
+ global EXTN(jsimd_rgb_ycc_convert_sse2)
+
+EXTN(jsimd_rgb_ycc_convert_sse2):
+ push rbp
+ mov rax, rsp ; rax = original rbp
+ sub rsp, byte 4
+ and rsp, byte (-SIZEOF_XMMWORD) ; align to 128 bits
+ mov [rsp], rax
+ mov rbp, rsp ; rbp = aligned rbp
+ lea rsp, [wk(0)]
+ collect_args 5
+ push rbx
+
+ mov ecx, r10d
+ test rcx, rcx
+ jz near .return
+
+ push rcx
+
+ mov rsi, r12
+ mov ecx, r13d
+ mov rdi, JSAMPARRAY [rsi+0*SIZEOF_JSAMPARRAY]
+ mov rbx, JSAMPARRAY [rsi+1*SIZEOF_JSAMPARRAY]
+ mov rdx, JSAMPARRAY [rsi+2*SIZEOF_JSAMPARRAY]
+ lea rdi, [rdi+rcx*SIZEOF_JSAMPROW]
+ lea rbx, [rbx+rcx*SIZEOF_JSAMPROW]
+ lea rdx, [rdx+rcx*SIZEOF_JSAMPROW]
+
+ pop rcx
+
+ mov rsi, r11
+ mov eax, r14d
+ test rax, rax
+ jle near .return
+.rowloop:
+ push rdx
+ push rbx
+ push rdi
+ push rsi
+ push rcx ; col
+
+ mov rsi, JSAMPROW [rsi] ; inptr
+ mov rdi, JSAMPROW [rdi] ; outptr0
+ mov rbx, JSAMPROW [rbx] ; outptr1
+ mov rdx, JSAMPROW [rdx] ; outptr2
+
+ cmp rcx, byte SIZEOF_XMMWORD
+ jae near .columnloop
+
+%if RGB_PIXELSIZE == 3 ; ---------------
+
+.column_ld1:
+ push rax
+ push rdx
+ lea rcx, [rcx+rcx*2] ; imul ecx,RGB_PIXELSIZE
+ test cl, SIZEOF_BYTE
+ jz short .column_ld2
+ sub rcx, byte SIZEOF_BYTE
+ movzx rax, BYTE [rsi+rcx]
+.column_ld2:
+ test cl, SIZEOF_WORD
+ jz short .column_ld4
+ sub rcx, byte SIZEOF_WORD
+ movzx rdx, WORD [rsi+rcx]
+ shl rax, WORD_BIT
+ or rax, rdx
+.column_ld4:
+ movd xmmA, eax
+ pop rdx
+ pop rax
+ test cl, SIZEOF_DWORD
+ jz short .column_ld8
+ sub rcx, byte SIZEOF_DWORD
+ movd xmmF, XMM_DWORD [rsi+rcx]
+ pslldq xmmA, SIZEOF_DWORD
+ por xmmA, xmmF
+.column_ld8:
+ test cl, SIZEOF_MMWORD
+ jz short .column_ld16
+ sub rcx, byte SIZEOF_MMWORD
+ movq xmmB, XMM_MMWORD [rsi+rcx]
+ pslldq xmmA, SIZEOF_MMWORD
+ por xmmA, xmmB
+.column_ld16:
+ test cl, SIZEOF_XMMWORD
+ jz short .column_ld32
+ movdqa xmmF, xmmA
+ movdqu xmmA, XMMWORD [rsi+0*SIZEOF_XMMWORD]
+ mov rcx, SIZEOF_XMMWORD
+ jmp short .rgb_ycc_cnv
+.column_ld32:
+ test cl, 2*SIZEOF_XMMWORD
+ mov rcx, SIZEOF_XMMWORD
+ jz short .rgb_ycc_cnv
+ movdqa xmmB, xmmA
+ movdqu xmmA, XMMWORD [rsi+0*SIZEOF_XMMWORD]
+ movdqu xmmF, XMMWORD [rsi+1*SIZEOF_XMMWORD]
+ jmp short .rgb_ycc_cnv
+
+.columnloop:
+ movdqu xmmA, XMMWORD [rsi+0*SIZEOF_XMMWORD]
+ movdqu xmmF, XMMWORD [rsi+1*SIZEOF_XMMWORD]
+ movdqu xmmB, XMMWORD [rsi+2*SIZEOF_XMMWORD]
+
+.rgb_ycc_cnv:
+ ; xmmA=(00 10 20 01 11 21 02 12 22 03 13 23 04 14 24 05)
+ ; xmmF=(15 25 06 16 26 07 17 27 08 18 28 09 19 29 0A 1A)
+ ; xmmB=(2A 0B 1B 2B 0C 1C 2C 0D 1D 2D 0E 1E 2E 0F 1F 2F)
+
+ movdqa xmmG, xmmA
+ pslldq xmmA, 8 ; xmmA=(-- -- -- -- -- -- -- -- 00 10 20 01 11 21 02 12)
+ psrldq xmmG, 8 ; xmmG=(22 03 13 23 04 14 24 05 -- -- -- -- -- -- -- --)
+
+ punpckhbw xmmA, xmmF ; xmmA=(00 08 10 18 20 28 01 09 11 19 21 29 02 0A 12 1A)
+ pslldq xmmF, 8 ; xmmF=(-- -- -- -- -- -- -- -- 15 25 06 16 26 07 17 27)
+
+ punpcklbw xmmG, xmmB ; xmmG=(22 2A 03 0B 13 1B 23 2B 04 0C 14 1C 24 2C 05 0D)
+ punpckhbw xmmF, xmmB ; xmmF=(15 1D 25 2D 06 0E 16 1E 26 2E 07 0F 17 1F 27 2F)
+
+ movdqa xmmD, xmmA
+ pslldq xmmA, 8 ; xmmA=(-- -- -- -- -- -- -- -- 00 08 10 18 20 28 01 09)
+ psrldq xmmD, 8 ; xmmD=(11 19 21 29 02 0A 12 1A -- -- -- -- -- -- -- --)
+
+ punpckhbw xmmA, xmmG ; xmmA=(00 04 08 0C 10 14 18 1C 20 24 28 2C 01 05 09 0D)
+ pslldq xmmG, 8 ; xmmG=(-- -- -- -- -- -- -- -- 22 2A 03 0B 13 1B 23 2B)
+
+ punpcklbw xmmD, xmmF ; xmmD=(11 15 19 1D 21 25 29 2D 02 06 0A 0E 12 16 1A 1E)
+ punpckhbw xmmG, xmmF ; xmmG=(22 26 2A 2E 03 07 0B 0F 13 17 1B 1F 23 27 2B 2F)
+
+ movdqa xmmE, xmmA
+ pslldq xmmA, 8 ; xmmA=(-- -- -- -- -- -- -- -- 00 04 08 0C 10 14 18 1C)
+ psrldq xmmE, 8 ; xmmE=(20 24 28 2C 01 05 09 0D -- -- -- -- -- -- -- --)
+
+ punpckhbw xmmA, xmmD ; xmmA=(00 02 04 06 08 0A 0C 0E 10 12 14 16 18 1A 1C 1E)
+ pslldq xmmD, 8 ; xmmD=(-- -- -- -- -- -- -- -- 11 15 19 1D 21 25 29 2D)
+
+ punpcklbw xmmE, xmmG ; xmmE=(20 22 24 26 28 2A 2C 2E 01 03 05 07 09 0B 0D 0F)
+ punpckhbw xmmD, xmmG ; xmmD=(11 13 15 17 19 1B 1D 1F 21 23 25 27 29 2B 2D 2F)
+
+ pxor xmmH, xmmH
+
+ movdqa xmmC, xmmA
+ punpcklbw xmmA, xmmH ; xmmA=(00 02 04 06 08 0A 0C 0E)
+ punpckhbw xmmC, xmmH ; xmmC=(10 12 14 16 18 1A 1C 1E)
+
+ movdqa xmmB, xmmE
+ punpcklbw xmmE, xmmH ; xmmE=(20 22 24 26 28 2A 2C 2E)
+ punpckhbw xmmB, xmmH ; xmmB=(01 03 05 07 09 0B 0D 0F)
+
+ movdqa xmmF, xmmD
+ punpcklbw xmmD, xmmH ; xmmD=(11 13 15 17 19 1B 1D 1F)
+ punpckhbw xmmF, xmmH ; xmmF=(21 23 25 27 29 2B 2D 2F)
+
+%else ; RGB_PIXELSIZE == 4 ; -----------
+
+.column_ld1:
+ test cl, SIZEOF_XMMWORD/16
+ jz short .column_ld2
+ sub rcx, byte SIZEOF_XMMWORD/16
+ movd xmmA, XMM_DWORD [rsi+rcx*RGB_PIXELSIZE]
+.column_ld2:
+ test cl, SIZEOF_XMMWORD/8
+ jz short .column_ld4
+ sub rcx, byte SIZEOF_XMMWORD/8
+ movq xmmE, XMM_MMWORD [rsi+rcx*RGB_PIXELSIZE]
+ pslldq xmmA, SIZEOF_MMWORD
+ por xmmA, xmmE
+.column_ld4:
+ test cl, SIZEOF_XMMWORD/4
+ jz short .column_ld8
+ sub rcx, byte SIZEOF_XMMWORD/4
+ movdqa xmmE, xmmA
+ movdqu xmmA, XMMWORD [rsi+rcx*RGB_PIXELSIZE]
+.column_ld8:
+ test cl, SIZEOF_XMMWORD/2
+ mov rcx, SIZEOF_XMMWORD
+ jz short .rgb_ycc_cnv
+ movdqa xmmF, xmmA
+ movdqa xmmH, xmmE
+ movdqu xmmA, XMMWORD [rsi+0*SIZEOF_XMMWORD]
+ movdqu xmmE, XMMWORD [rsi+1*SIZEOF_XMMWORD]
+ jmp short .rgb_ycc_cnv
+
+.columnloop:
+ movdqu xmmA, XMMWORD [rsi+0*SIZEOF_XMMWORD]
+ movdqu xmmE, XMMWORD [rsi+1*SIZEOF_XMMWORD]
+ movdqu xmmF, XMMWORD [rsi+2*SIZEOF_XMMWORD]
+ movdqu xmmH, XMMWORD [rsi+3*SIZEOF_XMMWORD]
+
+.rgb_ycc_cnv:
+ ; xmmA=(00 10 20 30 01 11 21 31 02 12 22 32 03 13 23 33)
+ ; xmmE=(04 14 24 34 05 15 25 35 06 16 26 36 07 17 27 37)
+ ; xmmF=(08 18 28 38 09 19 29 39 0A 1A 2A 3A 0B 1B 2B 3B)
+ ; xmmH=(0C 1C 2C 3C 0D 1D 2D 3D 0E 1E 2E 3E 0F 1F 2F 3F)
+
+ movdqa xmmD, xmmA
+ punpcklbw xmmA, xmmE ; xmmA=(00 04 10 14 20 24 30 34 01 05 11 15 21 25 31 35)
+ punpckhbw xmmD, xmmE ; xmmD=(02 06 12 16 22 26 32 36 03 07 13 17 23 27 33 37)
+
+ movdqa xmmC, xmmF
+ punpcklbw xmmF, xmmH ; xmmF=(08 0C 18 1C 28 2C 38 3C 09 0D 19 1D 29 2D 39 3D)
+ punpckhbw xmmC, xmmH ; xmmC=(0A 0E 1A 1E 2A 2E 3A 3E 0B 0F 1B 1F 2B 2F 3B 3F)
+
+ movdqa xmmB, xmmA
+ punpcklwd xmmA, xmmF ; xmmA=(00 04 08 0C 10 14 18 1C 20 24 28 2C 30 34 38 3C)
+ punpckhwd xmmB, xmmF ; xmmB=(01 05 09 0D 11 15 19 1D 21 25 29 2D 31 35 39 3D)
+
+ movdqa xmmG, xmmD
+ punpcklwd xmmD, xmmC ; xmmD=(02 06 0A 0E 12 16 1A 1E 22 26 2A 2E 32 36 3A 3E)
+ punpckhwd xmmG, xmmC ; xmmG=(03 07 0B 0F 13 17 1B 1F 23 27 2B 2F 33 37 3B 3F)
+
+ movdqa xmmE, xmmA
+ punpcklbw xmmA, xmmD ; xmmA=(00 02 04 06 08 0A 0C 0E 10 12 14 16 18 1A 1C 1E)
+ punpckhbw xmmE, xmmD ; xmmE=(20 22 24 26 28 2A 2C 2E 30 32 34 36 38 3A 3C 3E)
+
+ movdqa xmmH, xmmB
+ punpcklbw xmmB, xmmG ; xmmB=(01 03 05 07 09 0B 0D 0F 11 13 15 17 19 1B 1D 1F)
+ punpckhbw xmmH, xmmG ; xmmH=(21 23 25 27 29 2B 2D 2F 31 33 35 37 39 3B 3D 3F)
+
+ pxor xmmF, xmmF
+
+ movdqa xmmC, xmmA
+ punpcklbw xmmA, xmmF ; xmmA=(00 02 04 06 08 0A 0C 0E)
+ punpckhbw xmmC, xmmF ; xmmC=(10 12 14 16 18 1A 1C 1E)
+
+ movdqa xmmD, xmmB
+ punpcklbw xmmB, xmmF ; xmmB=(01 03 05 07 09 0B 0D 0F)
+ punpckhbw xmmD, xmmF ; xmmD=(11 13 15 17 19 1B 1D 1F)
+
+ movdqa xmmG, xmmE
+ punpcklbw xmmE, xmmF ; xmmE=(20 22 24 26 28 2A 2C 2E)
+ punpckhbw xmmG, xmmF ; xmmG=(30 32 34 36 38 3A 3C 3E)
+
+ punpcklbw xmmF, xmmH
+ punpckhbw xmmH, xmmH
+ psrlw xmmF, BYTE_BIT ; xmmF=(21 23 25 27 29 2B 2D 2F)
+ psrlw xmmH, BYTE_BIT ; xmmH=(31 33 35 37 39 3B 3D 3F)
+
+%endif ; RGB_PIXELSIZE ; ---------------
+
+ ; xmm0=R(02468ACE)=RE, xmm2=G(02468ACE)=GE, xmm4=B(02468ACE)=BE
+ ; xmm1=R(13579BDF)=RO, xmm3=G(13579BDF)=GO, xmm5=B(13579BDF)=BO
+
+ ; (Original)
+ ; Y = 0.29900 * R + 0.58700 * G + 0.11400 * B
+ ; Cb = -0.16874 * R - 0.33126 * G + 0.50000 * B + CENTERJSAMPLE
+ ; Cr = 0.50000 * R - 0.41869 * G - 0.08131 * B + CENTERJSAMPLE
+ ;
+ ; (This implementation)
+ ; Y = 0.29900 * R + 0.33700 * G + 0.11400 * B + 0.25000 * G
+ ; Cb = -0.16874 * R - 0.33126 * G + 0.50000 * B + CENTERJSAMPLE
+ ; Cr = 0.50000 * R - 0.41869 * G - 0.08131 * B + CENTERJSAMPLE
+
+ movdqa XMMWORD [wk(0)], xmm0 ; wk(0)=RE
+ movdqa XMMWORD [wk(1)], xmm1 ; wk(1)=RO
+ movdqa XMMWORD [wk(2)], xmm4 ; wk(2)=BE
+ movdqa XMMWORD [wk(3)], xmm5 ; wk(3)=BO
+
+ movdqa xmm6, xmm1
+ punpcklwd xmm1, xmm3
+ punpckhwd xmm6, xmm3
+ movdqa xmm7, xmm1
+ movdqa xmm4, xmm6
+ pmaddwd xmm1, [rel PW_F0299_F0337] ; xmm1=ROL*FIX(0.299)+GOL*FIX(0.337)
+ pmaddwd xmm6, [rel PW_F0299_F0337] ; xmm6=ROH*FIX(0.299)+GOH*FIX(0.337)
+ pmaddwd xmm7, [rel PW_MF016_MF033] ; xmm7=ROL*-FIX(0.168)+GOL*-FIX(0.331)
+ pmaddwd xmm4, [rel PW_MF016_MF033] ; xmm4=ROH*-FIX(0.168)+GOH*-FIX(0.331)
+
+ movdqa XMMWORD [wk(4)], xmm1 ; wk(4)=ROL*FIX(0.299)+GOL*FIX(0.337)
+ movdqa XMMWORD [wk(5)], xmm6 ; wk(5)=ROH*FIX(0.299)+GOH*FIX(0.337)
+
+ pxor xmm1, xmm1
+ pxor xmm6, xmm6
+ punpcklwd xmm1, xmm5 ; xmm1=BOL
+ punpckhwd xmm6, xmm5 ; xmm6=BOH
+ psrld xmm1, 1 ; xmm1=BOL*FIX(0.500)
+ psrld xmm6, 1 ; xmm6=BOH*FIX(0.500)
+
+ movdqa xmm5, [rel PD_ONEHALFM1_CJ] ; xmm5=[PD_ONEHALFM1_CJ]
+
+ paddd xmm7, xmm1
+ paddd xmm4, xmm6
+ paddd xmm7, xmm5
+ paddd xmm4, xmm5
+ psrld xmm7, SCALEBITS ; xmm7=CbOL
+ psrld xmm4, SCALEBITS ; xmm4=CbOH
+ packssdw xmm7, xmm4 ; xmm7=CbO
+
+ movdqa xmm1, XMMWORD [wk(2)] ; xmm1=BE
+
+ movdqa xmm6, xmm0
+ punpcklwd xmm0, xmm2
+ punpckhwd xmm6, xmm2
+ movdqa xmm5, xmm0
+ movdqa xmm4, xmm6
+ pmaddwd xmm0, [rel PW_F0299_F0337] ; xmm0=REL*FIX(0.299)+GEL*FIX(0.337)
+ pmaddwd xmm6, [rel PW_F0299_F0337] ; xmm6=REH*FIX(0.299)+GEH*FIX(0.337)
+ pmaddwd xmm5, [rel PW_MF016_MF033] ; xmm5=REL*-FIX(0.168)+GEL*-FIX(0.331)
+ pmaddwd xmm4, [rel PW_MF016_MF033] ; xmm4=REH*-FIX(0.168)+GEH*-FIX(0.331)
+
+ movdqa XMMWORD [wk(6)], xmm0 ; wk(6)=REL*FIX(0.299)+GEL*FIX(0.337)
+ movdqa XMMWORD [wk(7)], xmm6 ; wk(7)=REH*FIX(0.299)+GEH*FIX(0.337)
+
+ pxor xmm0, xmm0
+ pxor xmm6, xmm6
+ punpcklwd xmm0, xmm1 ; xmm0=BEL
+ punpckhwd xmm6, xmm1 ; xmm6=BEH
+ psrld xmm0, 1 ; xmm0=BEL*FIX(0.500)
+ psrld xmm6, 1 ; xmm6=BEH*FIX(0.500)
+
+ movdqa xmm1, [rel PD_ONEHALFM1_CJ] ; xmm1=[PD_ONEHALFM1_CJ]
+
+ paddd xmm5, xmm0
+ paddd xmm4, xmm6
+ paddd xmm5, xmm1
+ paddd xmm4, xmm1
+ psrld xmm5, SCALEBITS ; xmm5=CbEL
+ psrld xmm4, SCALEBITS ; xmm4=CbEH
+ packssdw xmm5, xmm4 ; xmm5=CbE
+
+ psllw xmm7, BYTE_BIT
+ por xmm5, xmm7 ; xmm5=Cb
+ movdqa XMMWORD [rbx], xmm5 ; Save Cb
+
+ movdqa xmm0, XMMWORD [wk(3)] ; xmm0=BO
+ movdqa xmm6, XMMWORD [wk(2)] ; xmm6=BE
+ movdqa xmm1, XMMWORD [wk(1)] ; xmm1=RO
+
+ movdqa xmm4, xmm0
+ punpcklwd xmm0, xmm3
+ punpckhwd xmm4, xmm3
+ movdqa xmm7, xmm0
+ movdqa xmm5, xmm4
+ pmaddwd xmm0, [rel PW_F0114_F0250] ; xmm0=BOL*FIX(0.114)+GOL*FIX(0.250)
+ pmaddwd xmm4, [rel PW_F0114_F0250] ; xmm4=BOH*FIX(0.114)+GOH*FIX(0.250)
+ pmaddwd xmm7, [rel PW_MF008_MF041] ; xmm7=BOL*-FIX(0.081)+GOL*-FIX(0.418)
+ pmaddwd xmm5, [rel PW_MF008_MF041] ; xmm5=BOH*-FIX(0.081)+GOH*-FIX(0.418)
+
+ movdqa xmm3, [rel PD_ONEHALF] ; xmm3=[PD_ONEHALF]
+
+ paddd xmm0, XMMWORD [wk(4)]
+ paddd xmm4, XMMWORD [wk(5)]
+ paddd xmm0, xmm3
+ paddd xmm4, xmm3
+ psrld xmm0, SCALEBITS ; xmm0=YOL
+ psrld xmm4, SCALEBITS ; xmm4=YOH
+ packssdw xmm0, xmm4 ; xmm0=YO
+
+ pxor xmm3, xmm3
+ pxor xmm4, xmm4
+ punpcklwd xmm3, xmm1 ; xmm3=ROL
+ punpckhwd xmm4, xmm1 ; xmm4=ROH
+ psrld xmm3, 1 ; xmm3=ROL*FIX(0.500)
+ psrld xmm4, 1 ; xmm4=ROH*FIX(0.500)
+
+ movdqa xmm1, [rel PD_ONEHALFM1_CJ] ; xmm1=[PD_ONEHALFM1_CJ]
+
+ paddd xmm7, xmm3
+ paddd xmm5, xmm4
+ paddd xmm7, xmm1
+ paddd xmm5, xmm1
+ psrld xmm7, SCALEBITS ; xmm7=CrOL
+ psrld xmm5, SCALEBITS ; xmm5=CrOH
+ packssdw xmm7, xmm5 ; xmm7=CrO
+
+ movdqa xmm3, XMMWORD [wk(0)] ; xmm3=RE
+
+ movdqa xmm4, xmm6
+ punpcklwd xmm6, xmm2
+ punpckhwd xmm4, xmm2
+ movdqa xmm1, xmm6
+ movdqa xmm5, xmm4
+ pmaddwd xmm6, [rel PW_F0114_F0250] ; xmm6=BEL*FIX(0.114)+GEL*FIX(0.250)
+ pmaddwd xmm4, [rel PW_F0114_F0250] ; xmm4=BEH*FIX(0.114)+GEH*FIX(0.250)
+ pmaddwd xmm1, [rel PW_MF008_MF041] ; xmm1=BEL*-FIX(0.081)+GEL*-FIX(0.418)
+ pmaddwd xmm5, [rel PW_MF008_MF041] ; xmm5=BEH*-FIX(0.081)+GEH*-FIX(0.418)
+
+ movdqa xmm2, [rel PD_ONEHALF] ; xmm2=[PD_ONEHALF]
+
+ paddd xmm6, XMMWORD [wk(6)]
+ paddd xmm4, XMMWORD [wk(7)]
+ paddd xmm6, xmm2
+ paddd xmm4, xmm2
+ psrld xmm6, SCALEBITS ; xmm6=YEL
+ psrld xmm4, SCALEBITS ; xmm4=YEH
+ packssdw xmm6, xmm4 ; xmm6=YE
+
+ psllw xmm0, BYTE_BIT
+ por xmm6, xmm0 ; xmm6=Y
+ movdqa XMMWORD [rdi], xmm6 ; Save Y
+
+ pxor xmm2, xmm2
+ pxor xmm4, xmm4
+ punpcklwd xmm2, xmm3 ; xmm2=REL
+ punpckhwd xmm4, xmm3 ; xmm4=REH
+ psrld xmm2, 1 ; xmm2=REL*FIX(0.500)
+ psrld xmm4, 1 ; xmm4=REH*FIX(0.500)
+
+ movdqa xmm0, [rel PD_ONEHALFM1_CJ] ; xmm0=[PD_ONEHALFM1_CJ]
+
+ paddd xmm1, xmm2
+ paddd xmm5, xmm4
+ paddd xmm1, xmm0
+ paddd xmm5, xmm0
+ psrld xmm1, SCALEBITS ; xmm1=CrEL
+ psrld xmm5, SCALEBITS ; xmm5=CrEH
+ packssdw xmm1, xmm5 ; xmm1=CrE
+
+ psllw xmm7, BYTE_BIT
+ por xmm1, xmm7 ; xmm1=Cr
+ movdqa XMMWORD [rdx], xmm1 ; Save Cr
+
+ sub rcx, byte SIZEOF_XMMWORD
+ add rsi, byte RGB_PIXELSIZE*SIZEOF_XMMWORD ; inptr
+ add rdi, byte SIZEOF_XMMWORD ; outptr0
+ add rbx, byte SIZEOF_XMMWORD ; outptr1
+ add rdx, byte SIZEOF_XMMWORD ; outptr2
+ cmp rcx, byte SIZEOF_XMMWORD
+ jae near .columnloop
+ test rcx, rcx
+ jnz near .column_ld1
+
+ pop rcx ; col
+ pop rsi
+ pop rdi
+ pop rbx
+ pop rdx
+
+ add rsi, byte SIZEOF_JSAMPROW ; input_buf
+ add rdi, byte SIZEOF_JSAMPROW
+ add rbx, byte SIZEOF_JSAMPROW
+ add rdx, byte SIZEOF_JSAMPROW
+ dec rax ; num_rows
+ jg near .rowloop
+
+.return:
+ pop rbx
+ uncollect_args 5
+ mov rsp, rbp ; rsp <- aligned rbp
+ pop rsp ; rsp <- original rbp
+ pop rbp
+ ret
+
+; For some reason, the OS X linker does not honor the request to align the
+; segment unless we do this.
+ align 32
diff --git a/simd/x86_64/jccolor-avx2.asm b/simd/x86_64/jccolor-avx2.asm
new file mode 100644
index 0000000..805c03d
--- /dev/null
+++ b/simd/x86_64/jccolor-avx2.asm
@@ -0,0 +1,122 @@
+;
+; jccolor.asm - colorspace conversion (64-bit AVX2)
+;
+; Copyright (C) 2009, 2016, D. R. Commander.
+; Copyright (C) 2015, Intel Corporation.
+;
+; Based on the x86 SIMD extension for IJG JPEG library
+; Copyright (C) 1999-2006, MIYASAKA Masaru.
+; For conditions of distribution and use, see copyright notice in jsimdext.inc
+;
+; This file should be assembled with NASM (Netwide Assembler),
+; can *not* be assembled with Microsoft's MASM or any compatible
+; assembler (including Borland's Turbo Assembler).
+; NASM is available from http://nasm.sourceforge.net/ or
+; http://sourceforge.net/project/showfiles.php?group_id=6208
+;
+; [TAB8]
+
+%include "jsimdext.inc"
+
+; --------------------------------------------------------------------------
+
+%define SCALEBITS 16
+
+F_0_081 equ 5329 ; FIX(0.08131)
+F_0_114 equ 7471 ; FIX(0.11400)
+F_0_168 equ 11059 ; FIX(0.16874)
+F_0_250 equ 16384 ; FIX(0.25000)
+F_0_299 equ 19595 ; FIX(0.29900)
+F_0_331 equ 21709 ; FIX(0.33126)
+F_0_418 equ 27439 ; FIX(0.41869)
+F_0_587 equ 38470 ; FIX(0.58700)
+F_0_337 equ (F_0_587 - F_0_250) ; FIX(0.58700) - FIX(0.25000)
+
+; --------------------------------------------------------------------------
+ SECTION SEG_CONST
+
+ alignz 32
+ global EXTN(jconst_rgb_ycc_convert_avx2)
+
+EXTN(jconst_rgb_ycc_convert_avx2):
+
+PW_F0299_F0337 times 8 dw F_0_299, F_0_337
+PW_F0114_F0250 times 8 dw F_0_114, F_0_250
+PW_MF016_MF033 times 8 dw -F_0_168,-F_0_331
+PW_MF008_MF041 times 8 dw -F_0_081,-F_0_418
+PD_ONEHALFM1_CJ times 8 dd (1 << (SCALEBITS-1)) - 1 + (CENTERJSAMPLE << SCALEBITS)
+PD_ONEHALF times 8 dd (1 << (SCALEBITS-1))
+
+ alignz 32
+
+; --------------------------------------------------------------------------
+ SECTION SEG_TEXT
+ BITS 64
+
+%include "jccolext-avx2.asm"
+
+%undef RGB_RED
+%undef RGB_GREEN
+%undef RGB_BLUE
+%undef RGB_PIXELSIZE
+%define RGB_RED EXT_RGB_RED
+%define RGB_GREEN EXT_RGB_GREEN
+%define RGB_BLUE EXT_RGB_BLUE
+%define RGB_PIXELSIZE EXT_RGB_PIXELSIZE
+%define jsimd_rgb_ycc_convert_avx2 jsimd_extrgb_ycc_convert_avx2
+%include "jccolext-avx2.asm"
+
+%undef RGB_RED
+%undef RGB_GREEN
+%undef RGB_BLUE
+%undef RGB_PIXELSIZE
+%define RGB_RED EXT_RGBX_RED
+%define RGB_GREEN EXT_RGBX_GREEN
+%define RGB_BLUE EXT_RGBX_BLUE
+%define RGB_PIXELSIZE EXT_RGBX_PIXELSIZE
+%define jsimd_rgb_ycc_convert_avx2 jsimd_extrgbx_ycc_convert_avx2
+%include "jccolext-avx2.asm"
+
+%undef RGB_RED
+%undef RGB_GREEN
+%undef RGB_BLUE
+%undef RGB_PIXELSIZE
+%define RGB_RED EXT_BGR_RED
+%define RGB_GREEN EXT_BGR_GREEN
+%define RGB_BLUE EXT_BGR_BLUE
+%define RGB_PIXELSIZE EXT_BGR_PIXELSIZE
+%define jsimd_rgb_ycc_convert_avx2 jsimd_extbgr_ycc_convert_avx2
+%include "jccolext-avx2.asm"
+
+%undef RGB_RED
+%undef RGB_GREEN
+%undef RGB_BLUE
+%undef RGB_PIXELSIZE
+%define RGB_RED EXT_BGRX_RED
+%define RGB_GREEN EXT_BGRX_GREEN
+%define RGB_BLUE EXT_BGRX_BLUE
+%define RGB_PIXELSIZE EXT_BGRX_PIXELSIZE
+%define jsimd_rgb_ycc_convert_avx2 jsimd_extbgrx_ycc_convert_avx2
+%include "jccolext-avx2.asm"
+
+%undef RGB_RED
+%undef RGB_GREEN
+%undef RGB_BLUE
+%undef RGB_PIXELSIZE
+%define RGB_RED EXT_XBGR_RED
+%define RGB_GREEN EXT_XBGR_GREEN
+%define RGB_BLUE EXT_XBGR_BLUE
+%define RGB_PIXELSIZE EXT_XBGR_PIXELSIZE
+%define jsimd_rgb_ycc_convert_avx2 jsimd_extxbgr_ycc_convert_avx2
+%include "jccolext-avx2.asm"
+
+%undef RGB_RED
+%undef RGB_GREEN
+%undef RGB_BLUE
+%undef RGB_PIXELSIZE
+%define RGB_RED EXT_XRGB_RED
+%define RGB_GREEN EXT_XRGB_GREEN
+%define RGB_BLUE EXT_XRGB_BLUE
+%define RGB_PIXELSIZE EXT_XRGB_PIXELSIZE
+%define jsimd_rgb_ycc_convert_avx2 jsimd_extxrgb_ycc_convert_avx2
+%include "jccolext-avx2.asm"
diff --git a/simd/jccolor-sse2.asm b/simd/x86_64/jccolor-sse2.asm
similarity index 78%
copy from simd/jccolor-sse2.asm
copy to simd/x86_64/jccolor-sse2.asm
index 13124d1..c2179db 100644
--- a/simd/jccolor-sse2.asm
+++ b/simd/x86_64/jccolor-sse2.asm
@@ -1,7 +1,7 @@
;
-; jccolor.asm - colorspace conversion (SSE2)
+; jccolor.asm - colorspace conversion (64-bit SSE2)
;
-; Copyright (C) 2009, D. R. Commander.
+; Copyright (C) 2009, 2016, D. R. Commander.
;
; Based on the x86 SIMD extension for IJG JPEG library
; Copyright (C) 1999-2006, MIYASAKA Masaru.
@@ -19,23 +19,23 @@
; --------------------------------------------------------------------------
-%define SCALEBITS 16
+%define SCALEBITS 16
-F_0_081 equ 5329 ; FIX(0.08131)
-F_0_114 equ 7471 ; FIX(0.11400)
-F_0_168 equ 11059 ; FIX(0.16874)
-F_0_250 equ 16384 ; FIX(0.25000)
-F_0_299 equ 19595 ; FIX(0.29900)
-F_0_331 equ 21709 ; FIX(0.33126)
-F_0_418 equ 27439 ; FIX(0.41869)
-F_0_587 equ 38470 ; FIX(0.58700)
-F_0_337 equ (F_0_587 - F_0_250) ; FIX(0.58700) - FIX(0.25000)
+F_0_081 equ 5329 ; FIX(0.08131)
+F_0_114 equ 7471 ; FIX(0.11400)
+F_0_168 equ 11059 ; FIX(0.16874)
+F_0_250 equ 16384 ; FIX(0.25000)
+F_0_299 equ 19595 ; FIX(0.29900)
+F_0_331 equ 21709 ; FIX(0.33126)
+F_0_418 equ 27439 ; FIX(0.41869)
+F_0_587 equ 38470 ; FIX(0.58700)
+F_0_337 equ (F_0_587 - F_0_250) ; FIX(0.58700) - FIX(0.25000)
; --------------------------------------------------------------------------
- SECTION SEG_CONST
+ SECTION SEG_CONST
- alignz 16
- global EXTN(jconst_rgb_ycc_convert_sse2)
+ alignz 32
+ global EXTN(jconst_rgb_ycc_convert_sse2)
EXTN(jconst_rgb_ycc_convert_sse2):
@@ -46,11 +46,11 @@
PD_ONEHALFM1_CJ times 4 dd (1 << (SCALEBITS-1)) - 1 + (CENTERJSAMPLE << SCALEBITS)
PD_ONEHALF times 4 dd (1 << (SCALEBITS-1))
- alignz 16
+ alignz 32
; --------------------------------------------------------------------------
- SECTION SEG_TEXT
- BITS 32
+ SECTION SEG_TEXT
+ BITS 64
%include "jccolext-sse2.asm"
diff --git a/simd/x86_64/jcgray-avx2.asm b/simd/x86_64/jcgray-avx2.asm
new file mode 100644
index 0000000..e6ea21e
--- /dev/null
+++ b/simd/x86_64/jcgray-avx2.asm
@@ -0,0 +1,115 @@
+;
+; jcgray.asm - grayscale colorspace conversion (64-bit AVX2)
+;
+; Copyright (C) 2011, 2016, D. R. Commander.
+; Copyright (C) 2015, Intel Corporation.
+;
+; Based on the x86 SIMD extension for IJG JPEG library
+; Copyright (C) 1999-2006, MIYASAKA Masaru.
+; For conditions of distribution and use, see copyright notice in jsimdext.inc
+;
+; This file should be assembled with NASM (Netwide Assembler),
+; can *not* be assembled with Microsoft's MASM or any compatible
+; assembler (including Borland's Turbo Assembler).
+; NASM is available from http://nasm.sourceforge.net/ or
+; http://sourceforge.net/project/showfiles.php?group_id=6208
+;
+; [TAB8]
+
+%include "jsimdext.inc"
+
+; --------------------------------------------------------------------------
+
+%define SCALEBITS 16
+
+F_0_114 equ 7471 ; FIX(0.11400)
+F_0_250 equ 16384 ; FIX(0.25000)
+F_0_299 equ 19595 ; FIX(0.29900)
+F_0_587 equ 38470 ; FIX(0.58700)
+F_0_337 equ (F_0_587 - F_0_250) ; FIX(0.58700) - FIX(0.25000)
+
+; --------------------------------------------------------------------------
+ SECTION SEG_CONST
+
+ alignz 32
+ global EXTN(jconst_rgb_gray_convert_avx2)
+
+EXTN(jconst_rgb_gray_convert_avx2):
+
+PW_F0299_F0337 times 8 dw F_0_299, F_0_337
+PW_F0114_F0250 times 8 dw F_0_114, F_0_250
+PD_ONEHALF times 8 dd (1 << (SCALEBITS-1))
+
+ alignz 32
+
+; --------------------------------------------------------------------------
+ SECTION SEG_TEXT
+ BITS 64
+
+%include "jcgryext-avx2.asm"
+
+%undef RGB_RED
+%undef RGB_GREEN
+%undef RGB_BLUE
+%undef RGB_PIXELSIZE
+%define RGB_RED EXT_RGB_RED
+%define RGB_GREEN EXT_RGB_GREEN
+%define RGB_BLUE EXT_RGB_BLUE
+%define RGB_PIXELSIZE EXT_RGB_PIXELSIZE
+%define jsimd_rgb_gray_convert_avx2 jsimd_extrgb_gray_convert_avx2
+%include "jcgryext-avx2.asm"
+
+%undef RGB_RED
+%undef RGB_GREEN
+%undef RGB_BLUE
+%undef RGB_PIXELSIZE
+%define RGB_RED EXT_RGBX_RED
+%define RGB_GREEN EXT_RGBX_GREEN
+%define RGB_BLUE EXT_RGBX_BLUE
+%define RGB_PIXELSIZE EXT_RGBX_PIXELSIZE
+%define jsimd_rgb_gray_convert_avx2 jsimd_extrgbx_gray_convert_avx2
+%include "jcgryext-avx2.asm"
+
+%undef RGB_RED
+%undef RGB_GREEN
+%undef RGB_BLUE
+%undef RGB_PIXELSIZE
+%define RGB_RED EXT_BGR_RED
+%define RGB_GREEN EXT_BGR_GREEN
+%define RGB_BLUE EXT_BGR_BLUE
+%define RGB_PIXELSIZE EXT_BGR_PIXELSIZE
+%define jsimd_rgb_gray_convert_avx2 jsimd_extbgr_gray_convert_avx2
+%include "jcgryext-avx2.asm"
+
+%undef RGB_RED
+%undef RGB_GREEN
+%undef RGB_BLUE
+%undef RGB_PIXELSIZE
+%define RGB_RED EXT_BGRX_RED
+%define RGB_GREEN EXT_BGRX_GREEN
+%define RGB_BLUE EXT_BGRX_BLUE
+%define RGB_PIXELSIZE EXT_BGRX_PIXELSIZE
+%define jsimd_rgb_gray_convert_avx2 jsimd_extbgrx_gray_convert_avx2
+%include "jcgryext-avx2.asm"
+
+%undef RGB_RED
+%undef RGB_GREEN
+%undef RGB_BLUE
+%undef RGB_PIXELSIZE
+%define RGB_RED EXT_XBGR_RED
+%define RGB_GREEN EXT_XBGR_GREEN
+%define RGB_BLUE EXT_XBGR_BLUE
+%define RGB_PIXELSIZE EXT_XBGR_PIXELSIZE
+%define jsimd_rgb_gray_convert_avx2 jsimd_extxbgr_gray_convert_avx2
+%include "jcgryext-avx2.asm"
+
+%undef RGB_RED
+%undef RGB_GREEN
+%undef RGB_BLUE
+%undef RGB_PIXELSIZE
+%define RGB_RED EXT_XRGB_RED
+%define RGB_GREEN EXT_XRGB_GREEN
+%define RGB_BLUE EXT_XRGB_BLUE
+%define RGB_PIXELSIZE EXT_XRGB_PIXELSIZE
+%define jsimd_rgb_gray_convert_avx2 jsimd_extxrgb_gray_convert_avx2
+%include "jcgryext-avx2.asm"
diff --git a/simd/jcgray-sse2.asm b/simd/x86_64/jcgray-sse2.asm
similarity index 78%
copy from simd/jcgray-sse2.asm
copy to simd/x86_64/jcgray-sse2.asm
index 5b0b466..7cb9819 100644
--- a/simd/jcgray-sse2.asm
+++ b/simd/x86_64/jcgray-sse2.asm
@@ -1,7 +1,7 @@
;
-; jcgray.asm - grayscale colorspace conversion (SSE2)
+; jcgray.asm - grayscale colorspace conversion (64-bit SSE2)
;
-; Copyright (C) 2011, D. R. Commander.
+; Copyright (C) 2011, 2016, D. R. Commander.
;
; Based on the x86 SIMD extension for IJG JPEG library
; Copyright (C) 1999-2006, MIYASAKA Masaru.
@@ -19,31 +19,31 @@
; --------------------------------------------------------------------------
-%define SCALEBITS 16
+%define SCALEBITS 16
-F_0_114 equ 7471 ; FIX(0.11400)
-F_0_250 equ 16384 ; FIX(0.25000)
-F_0_299 equ 19595 ; FIX(0.29900)
-F_0_587 equ 38470 ; FIX(0.58700)
-F_0_337 equ (F_0_587 - F_0_250) ; FIX(0.58700) - FIX(0.25000)
+F_0_114 equ 7471 ; FIX(0.11400)
+F_0_250 equ 16384 ; FIX(0.25000)
+F_0_299 equ 19595 ; FIX(0.29900)
+F_0_587 equ 38470 ; FIX(0.58700)
+F_0_337 equ (F_0_587 - F_0_250) ; FIX(0.58700) - FIX(0.25000)
; --------------------------------------------------------------------------
- SECTION SEG_CONST
+ SECTION SEG_CONST
- alignz 16
- global EXTN(jconst_rgb_gray_convert_sse2)
+ alignz 32
+ global EXTN(jconst_rgb_gray_convert_sse2)
EXTN(jconst_rgb_gray_convert_sse2):
-PW_F0299_F0337 times 4 dw F_0_299, F_0_337
-PW_F0114_F0250 times 4 dw F_0_114, F_0_250
-PD_ONEHALF times 4 dd (1 << (SCALEBITS-1))
+PW_F0299_F0337 times 4 dw F_0_299, F_0_337
+PW_F0114_F0250 times 4 dw F_0_114, F_0_250
+PD_ONEHALF times 4 dd (1 << (SCALEBITS-1))
- alignz 16
+ alignz 32
; --------------------------------------------------------------------------
- SECTION SEG_TEXT
- BITS 32
+ SECTION SEG_TEXT
+ BITS 64
%include "jcgryext-sse2.asm"
diff --git a/simd/x86_64/jcgryext-avx2.asm b/simd/x86_64/jcgryext-avx2.asm
new file mode 100644
index 0000000..8824a45
--- /dev/null
+++ b/simd/x86_64/jcgryext-avx2.asm
@@ -0,0 +1,440 @@
+;
+; jcgryext.asm - grayscale colorspace conversion (64-bit AVX2)
+;
+; Copyright (C) 2011, 2016, D. R. Commander.
+; Copyright (C) 2015, Intel Corporation.
+;
+; Based on the x86 SIMD extension for IJG JPEG library
+; Copyright (C) 1999-2006, MIYASAKA Masaru.
+; For conditions of distribution and use, see copyright notice in jsimdext.inc
+;
+; This file should be assembled with NASM (Netwide Assembler),
+; can *not* be assembled with Microsoft's MASM or any compatible
+; assembler (including Borland's Turbo Assembler).
+; NASM is available from http://nasm.sourceforge.net/ or
+; http://sourceforge.net/project/showfiles.php?group_id=6208
+;
+; [TAB8]
+
+%include "jcolsamp.inc"
+
+; --------------------------------------------------------------------------
+;
+; Convert some rows of samples to the output colorspace.
+;
+; GLOBAL(void)
+; jsimd_rgb_gray_convert_avx2 (JDIMENSION img_width,
+; JSAMPARRAY input_buf, JSAMPIMAGE output_buf,
+; JDIMENSION output_row, int num_rows);
+;
+
+; r10d = JDIMENSION img_width
+; r11 = JSAMPARRAY input_buf
+; r12 = JSAMPIMAGE output_buf
+; r13d = JDIMENSION output_row
+; r14d = int num_rows
+
+%define wk(i) rbp-(WK_NUM-(i))*SIZEOF_YMMWORD ; ymmword wk[WK_NUM]
+%define WK_NUM 2
+
+ align 32
+
+ global EXTN(jsimd_rgb_gray_convert_avx2)
+
+EXTN(jsimd_rgb_gray_convert_avx2):
+ push rbp
+ mov rax, rsp ; rax = original rbp
+ sub rsp, byte 4
+ and rsp, byte (-SIZEOF_YMMWORD) ; align to 256 bits
+ mov [rsp], rax
+ mov rbp, rsp ; rbp = aligned rbp
+ lea rsp, [wk(0)]
+ collect_args 5
+ push rbx
+
+ mov ecx, r10d
+ test rcx, rcx
+ jz near .return
+
+ push rcx
+
+ mov rsi, r12
+ mov ecx, r13d
+ mov rdi, JSAMPARRAY [rsi+0*SIZEOF_JSAMPARRAY]
+ lea rdi, [rdi+rcx*SIZEOF_JSAMPROW]
+
+ pop rcx
+
+ mov rsi, r11
+ mov eax, r14d
+ test rax, rax
+ jle near .return
+.rowloop:
+ push rdi
+ push rsi
+ push rcx ; col
+
+ mov rsi, JSAMPROW [rsi] ; inptr
+ mov rdi, JSAMPROW [rdi] ; outptr0
+
+ cmp rcx, byte SIZEOF_YMMWORD
+ jae near .columnloop
+
+%if RGB_PIXELSIZE == 3 ; ---------------
+
+.column_ld1:
+ push rax
+ push rdx
+ lea rcx, [rcx+rcx*2] ; imul ecx,RGB_PIXELSIZE
+ test cl, SIZEOF_BYTE
+ jz short .column_ld2
+ sub rcx, byte SIZEOF_BYTE
+ movzx rax, BYTE [rsi+rcx]
+.column_ld2:
+ test cl, SIZEOF_WORD
+ jz short .column_ld4
+ sub rcx, byte SIZEOF_WORD
+ movzx rdx, WORD [rsi+rcx]
+ shl rax, WORD_BIT
+ or rax, rdx
+.column_ld4:
+ vmovd xmmA, eax
+ pop rdx
+ pop rax
+ test cl, SIZEOF_DWORD
+ jz short .column_ld8
+ sub rcx, byte SIZEOF_DWORD
+ vmovd xmmF, XMM_DWORD [rsi+rcx]
+ vpslldq xmmA, xmmA, SIZEOF_DWORD
+ vpor xmmA, xmmA, xmmF
+.column_ld8:
+ test cl, SIZEOF_MMWORD
+ jz short .column_ld16
+ sub rcx, byte SIZEOF_MMWORD
+ vmovq xmmB, XMM_MMWORD [rsi+rcx]
+ vpslldq xmmA, xmmA, SIZEOF_MMWORD
+ vpor xmmA, xmmA, xmmB
+.column_ld16:
+ test cl, SIZEOF_XMMWORD
+ jz short .column_ld32
+ sub rcx, byte SIZEOF_XMMWORD
+ vmovdqu xmmB, XMM_MMWORD [rsi+rcx]
+ vperm2i128 ymmA, ymmA, ymmA, 1
+ vpor ymmA, ymmB
+.column_ld32:
+ test cl, SIZEOF_YMMWORD
+ jz short .column_ld64
+ sub rcx, byte SIZEOF_YMMWORD
+ vmovdqa ymmF, ymmA
+ vmovdqu ymmA, YMMWORD [rsi+0*SIZEOF_YMMWORD]
+.column_ld64:
+ test cl, 2*SIZEOF_YMMWORD
+ mov rcx, SIZEOF_YMMWORD
+ jz short .rgb_gray_cnv
+ vmovdqa ymmB, ymmA
+ vmovdqu ymmA, YMMWORD [rsi+0*SIZEOF_YMMWORD]
+ vmovdqu ymmF, YMMWORD [rsi+1*SIZEOF_YMMWORD]
+ jmp short .rgb_gray_cnv
+
+.columnloop:
+ vmovdqu ymmA, YMMWORD [rsi+0*SIZEOF_YMMWORD]
+ vmovdqu ymmF, YMMWORD [rsi+1*SIZEOF_YMMWORD]
+ vmovdqu ymmB, YMMWORD [rsi+2*SIZEOF_YMMWORD]
+
+.rgb_gray_cnv:
+ ; ymmA=(00 10 20 01 11 21 02 12 22 03 13 23 04 14 24 05
+ ; 15 25 06 16 26 07 17 27 08 18 28 09 19 29 0A 1A)
+ ; ymmF=(2A 0B 1B 2B 0C 1C 2C 0D 1D 2D 0E 1E 2E 0F 1F 2F
+ ; 0G 1G 2G 0H 1H 2H 0I 1I 2I 0J 1J 2J 0K 1K 2K 0L)
+ ; ymmB=(1L 2L 0M 1M 2M 0N 1N 2N 0O 1O 2O 0P 1P 2P 0Q 1Q
+ ; 2Q 0R 1R 2R 0S 1S 2S 0T 1T 2T 0U 1U 2U 0V 1V 2V)
+
+ vmovdqu ymmC, ymmA
+ vinserti128 ymmA, ymmF, xmmA, 0 ; ymmA=(00 10 20 01 11 21 02 12 22 03 13 23 04 14 24 05
+ ; 0G 1G 2G 0H 1H 2H 0I 1I 2I 0J 1J 2J 0K 1K 2K 0L)
+ vinserti128 ymmC, ymmC, xmmB, 0 ; ymmC=(1L 2L 0M 1M 2M 0N 1N 2N 0O 1O 2O 0P 1P 2P 0Q 1Q
+ ; 15 25 06 16 26 07 17 27 08 18 28 09 19 29 0A 1A)
+ vinserti128 ymmB, ymmB, xmmF, 0 ; ymmB=(2A 0B 1B 2B 0C 1C 2C 0D 1D 2D 0E 1E 2E 0F 1F 2F
+ ; 2Q 0R 1R 2R 0S 1S 2S 0T 1T 2T 0U 1U 2U 0V 1V 2V)
+ vperm2i128 ymmF, ymmC, ymmC, 1 ; ymmF=(15 25 06 16 26 07 17 27 08 18 28 09 19 29 0A 1A
+ ; 1L 2L 0M 1M 2M 0N 1N 2N 0O 1O 2O 0P 1P 2P 0Q 1Q)
+
+ vmovdqa ymmG, ymmA
+ vpslldq ymmA, ymmA, 8 ; ymmA=(-- -- -- -- -- -- -- -- 00 10 20 01 11 21 02 12
+ ; 22 03 13 23 04 14 24 05 0G 1G 2G 0H 1H 2H 0I 1I)
+ vpsrldq ymmG, ymmG, 8 ; ymmG=(22 03 13 23 04 14 24 05 0G 1G 2G 0H 1H 2H 0I 1I
+ ; 2I 0J 1J 2J 0K 1K 2K 0L -- -- -- -- -- -- -- --)
+
+ vpunpckhbw ymmA, ymmA, ymmF ; ymmA=(00 08 10 18 20 28 01 09 11 19 21 29 02 0A 12 1A
+ ; 0G 0O 1G 1O 2G 2O 0H 0P 1H 1P 2H 2P 0I 0Q 1I 1Q)
+ vpslldq ymmF, ymmF, 8 ; ymmF=(-- -- -- -- -- -- -- -- 15 25 06 16 26 07 17 27
+ ; 08 18 28 09 19 29 0A 1A 1L 2L 0M 1M 2M 0N 1N 2N)
+
+ vpunpcklbw ymmG, ymmG, ymmB ; ymmG=(22 2A 03 0B 13 1B 23 2B 04 0C 14 1C 24 2C 05 0D
+ ; 2I 2Q 0J 0R 1J 1R 2J 2R 0K 0S 1K 1S 2K 2S 0L 0T)
+ vpunpckhbw ymmF, ymmF, ymmB ; ymmF=(15 1D 25 2D 06 0E 16 1E 26 2E 07 0F 17 1F 27 2F
+ ; 1L 1T 2L 2T 0M 0U 1M 1U 2M 2U 0N 0V 1N 1V 2N 2V)
+
+ vmovdqa ymmD, ymmA
+ vpslldq ymmA, ymmA, 8 ; ymmA=(-- -- -- -- -- -- -- -- 00 08 10 18 20 28 01 09
+ ; 11 19 21 29 02 0A 12 1A 0G 0O 1G 1O 2G 2O 0H 0P)
+ vpsrldq ymmD, ymmD, 8 ; ymmD=(11 19 21 29 02 0A 12 1A 0G 0O 1G 1O 2G 2O 0H 0P
+ ; 1H 1P 2H 2P 0I 0Q 1I 1Q -- -- -- -- -- -- -- --)
+
+ vpunpckhbw ymmA, ymmA, ymmG ; ymmA=(00 04 08 0C 10 14 18 1C 20 24 28 2C 01 05 09 0D
+ ; 0G 0K 0O 0S 1G 1K 1O 1S 2G 2K 2O 2S 0H 0L 0P 0T)
+ vpslldq ymmG, ymmG, 8 ; ymmG=(-- -- -- -- -- -- -- -- 22 2A 03 0B 13 1B 23 2B
+ ; 04 0C 14 1C 24 2C 05 0D 2I 2Q 0J 0R 1J 1R 2J 2R)
+
+ vpunpcklbw ymmD, ymmD, ymmF ; ymmD=(11 15 19 1D 21 25 29 2D 02 06 0A 0E 12 16 1A 1E
+ ; 1H 1L 1P 1T 2H 2L 2P 2T 0I 0M 0Q 0U 1I 1M 1Q 1U)
+ vpunpckhbw ymmG, ymmG, ymmF ; ymmG=(22 26 2A 2E 03 07 0B 0F 13 17 1B 1F 23 27 2B 2F
+ ; 2I 2M 2Q 2U 0J 0N 0R 0V 1J 1N 1R 1V 2J 2N 2R 2V)
+
+ vmovdqa ymmE, ymmA
+ vpslldq ymmA, ymmA, 8 ; ymmA=(-- -- -- -- -- -- -- -- 00 04 08 0C 10 14 18 1C
+ ; 20 24 28 2C 01 05 09 0D 0G 0K 0O 0S 1G 1K 1O 1S)
+ vpsrldq ymmE, ymmE, 8 ; ymmE=(20 24 28 2C 01 05 09 0D 0G 0K 0O 0S 1G 1K 1O 1S
+ ; 2G 2K 2O 2S 0H 0L 0P 0T -- -- -- -- -- -- -- --)
+
+ vpunpckhbw ymmA, ymmA, ymmD ; ymmA=(00 02 04 06 08 0A 0C 0E 10 12 14 16 18 1A 1C 1E
+ ; 0G 0I 0K 0M 0O 0Q 0S 0U 1G 1I 1K 1M 1O 1Q 1S 1U)
+ vpslldq ymmD, ymmD, 8 ; ymmD=(-- -- -- -- -- -- -- -- 11 15 19 1D 21 25 29 2D
+ ; 02 06 0A 0E 12 16 1A 1E 1H 1L 1P 1T 2H 2L 2P 2T)
+
+ vpunpcklbw ymmE, ymmE, ymmG ; ymmE=(20 22 24 26 28 2A 2C 2E 01 03 05 07 09 0B 0D 0F
+ ; 2G 2I 2K 2M 2O 2Q 2S 2U 0H 0J 0L 0N 0P 0R 0T 0V)
+ vpunpckhbw ymmD, ymmD, ymmG ; ymmD=(11 13 15 17 19 1B 1D 1F 21 23 25 27 29 2B 2D 2F
+ ; 1H 1J 1L 1N 1P 1R 1T 1V 2H 2J 2L 2N 2P 2R 2T 2V)
+
+ vpxor ymmH, ymmH, ymmH
+
+ vmovdqa ymmC, ymmA
+ vpunpcklbw ymmA, ymmA, ymmH ; ymmA=(00 02 04 06 08 0A 0C 0E 0G 0I 0K 0M 0O 0Q 0S 0U)
+ vpunpckhbw ymmC, ymmC, ymmH ; ymmC=(10 12 14 16 18 1A 1C 1E 1G 1I 1K 1M 1O 1Q 1S 1U)
+
+ vmovdqa ymmB, ymmE
+ vpunpcklbw ymmE, ymmE, ymmH ; ymmE=(20 22 24 26 28 2A 2C 2E 2G 2I 2K 2M 2O 2Q 2S 2U)
+ vpunpckhbw ymmB, ymmB, ymmH ; ymmB=(01 03 05 07 09 0B 0D 0F 0H 0J 0L 0N 0P 0R 0T 0V)
+
+ vmovdqa ymmF, ymmD
+ vpunpcklbw ymmD, ymmD, ymmH ; ymmD=(11 13 15 17 19 1B 1D 1F 1H 1J 1L 1N 1P 1R 1T 1V)
+ vpunpckhbw ymmF, ymmF, ymmH ; ymmF=(21 23 25 27 29 2B 2D 2F 2H 2J 2L 2N 2P 2R 2T 2V)
+
+%else ; RGB_PIXELSIZE == 4 ; -----------
+
+.column_ld1:
+ test cl, SIZEOF_XMMWORD/16
+ jz short .column_ld2
+ sub rcx, byte SIZEOF_XMMWORD/16
+ vmovd xmmA, XMM_DWORD [rsi+rcx*RGB_PIXELSIZE]
+.column_ld2:
+ test cl, SIZEOF_XMMWORD/8
+ jz short .column_ld4
+ sub rcx, byte SIZEOF_XMMWORD/8
+ vmovq xmmF, XMM_MMWORD [rsi+rcx*RGB_PIXELSIZE]
+ vpslldq xmmA, xmmA, SIZEOF_MMWORD
+ vpor xmmA, xmmA, xmmF
+.column_ld4:
+ test cl, SIZEOF_XMMWORD/4
+ jz short .column_ld8
+ sub rcx, byte SIZEOF_XMMWORD/4
+ vmovdqa xmmF, xmmA
+ vperm2i128 ymmF, ymmF, ymmF, 1
+ vmovdqu xmmA, XMMWORD [rsi+rcx*RGB_PIXELSIZE]
+ vpor ymmA, ymmA, ymmF
+.column_ld8:
+ test cl, SIZEOF_XMMWORD/2
+ jz short .column_ld16
+ sub rcx, byte SIZEOF_XMMWORD/2
+ vmovdqa ymmF, ymmA
+ vmovdqu ymmA, YMMWORD [rsi+rcx*RGB_PIXELSIZE]
+.column_ld16:
+ test cl, SIZEOF_XMMWORD
+ mov rcx, SIZEOF_YMMWORD
+ jz short .rgb_gray_cnv
+ vmovdqa ymmE, ymmA
+ vmovdqa ymmH, ymmF
+ vmovdqu ymmA, YMMWORD [rsi+0*SIZEOF_YMMWORD]
+ vmovdqu ymmF, YMMWORD [rsi+1*SIZEOF_YMMWORD]
+ jmp short .rgb_gray_cnv
+
+.columnloop:
+ vmovdqu ymmA, YMMWORD [rsi+0*SIZEOF_YMMWORD]
+ vmovdqu ymmF, YMMWORD [rsi+1*SIZEOF_YMMWORD]
+ vmovdqu ymmE, YMMWORD [rsi+2*SIZEOF_YMMWORD]
+ vmovdqu ymmH, YMMWORD [rsi+3*SIZEOF_YMMWORD]
+
+.rgb_gray_cnv:
+ ; ymmA=(00 10 20 30 01 11 21 31 02 12 22 32 03 13 23 33
+ ; 04 14 24 34 05 15 25 35 06 16 26 36 07 17 27 37)
+ ; ymmF=(08 18 28 38 09 19 29 39 0A 1A 2A 3A 0B 1B 2B 3B
+ ; 0C 1C 2C 3C 0D 1D 2D 3D 0E 1E 2E 3E 0F 1F 2F 3F)
+ ; ymmE=(0G 1G 2G 3G 0H 1H 2H 3H 0I 1I 2I 3I 0J 1J 2J 3J
+ ; 0K 1K 2K 3K 0L 1L 2L 3L 0M 1M 2M 3M 0N 1N 2N 3N)
+ ; ymmH=(0O 1O 2O 3O 0P 1P 2P 3P 0Q 1Q 2Q 3Q 0R 1R 2R 3R
+ ; 0S 1S 2S 3S 0T 1T 2T 3T 0U 1U 2U 3U 0V 1V 2V 3V)
+
+ vmovdqa ymmB, ymmA
+ vinserti128 ymmA, ymmA, xmmE, 1 ; ymmA=(00 10 20 30 01 11 21 31 02 12 22 32 03 13 23 33
+ ; 0G 1G 2G 3G 0H 1H 2H 3H 0I 1I 2I 3I 0J 1J 2J 3J)
+ vperm2i128 ymmE, ymmB, ymmE, 0x31 ; ymmE=(04 14 24 34 05 15 25 35 06 16 26 36 07 17 27 37
+ ; 0K 1K 2K 3K 0L 1L 2L 3L 0M 1M 2M 3M 0N 1N 2N 3N)
+
+ vmovdqa ymmB, ymmF
+ vinserti128 ymmF, ymmF, xmmH, 1 ; ymmF=(08 18 28 38 09 19 29 39 0A 1A 2A 3A 0B 1B 2B 3B
+ ; 0O 1O 2O 3O 0P 1P 2P 3P 0Q 1Q 2Q 3Q 0R 1R 2R 3R)
+ vperm2i128 ymmH, ymmB, ymmH, 0x31 ; ymmH=(0C 1C 2C 3C 0D 1D 2D 3D 0E 1E 2E 3E 0F 1F 2F 3F
+ ; 0S 1S 2S 3S 0T 1T 2T 3T 0U 1U 2U 3U 0V 1V 2V 3V)
+
+ vmovdqa ymmD, ymmA
+ vpunpcklbw ymmA, ymmA, ymmE ; ymmA=(00 04 10 14 20 24 30 34 01 05 11 15 21 25 31 35
+ ; 0G 0K 1G 1K 2G 2K 3G 3K 0H 0L 1H 1L 2H 2L 3H 3L)
+ vpunpckhbw ymmD, ymmD, ymmE ; ymmD=(02 06 12 16 22 26 32 36 03 07 13 17 23 27 33 37
+ ; 0I 0M 1I 1M 2I 2M 3I 3M 0J 0N 1J 1N 2J 2N 3J 3N)
+
+ vmovdqa ymmC, ymmF
+ vpunpcklbw ymmF, ymmF, ymmH ; ymmF=(08 0C 18 1C 28 2C 38 3C 09 0D 19 1D 29 2D 39 3D
+ ; 0O 0S 1O 1S 2O 2S 3O 3S 0P 0T 1P 1T 2P 2T 3P 3T)
+ vpunpckhbw ymmC, ymmC, ymmH ; ymmC=(0A 0E 1A 1E 2A 2E 3A 3E 0B 0F 1B 1F 2B 2F 3B 3F
+ ; 0Q 0U 1Q 1U 2Q 2U 3Q 3U 0R 0V 1R 1V 2R 2V 3R 3V)
+
+ vmovdqa ymmB, ymmA
+ vpunpcklwd ymmA, ymmA, ymmF ; ymmA=(00 04 08 0C 10 14 18 1C 20 24 28 2C 30 34 38 3C
+ ; 0G 0K 0O 0S 1G 1K 1O 1S 2G 2K 2O 2S 3G 3K 3O 3S)
+ vpunpckhwd ymmB, ymmB, ymmF ; ymmB=(01 05 09 0D 11 15 19 1D 21 25 29 2D 31 35 39 3D
+ ; 0H 0L 0P 0T 1H 1L 1P 1T 2H 2L 2P 2T 3H 3L 3P 3T)
+
+ vmovdqa ymmG, ymmD
+ vpunpcklwd ymmD, ymmD, ymmC ; ymmD=(02 06 0A 0E 12 16 1A 1E 22 26 2A 2E 32 36 3A 3E
+ ; 0I 0M 0Q 0U 1I 1M 1Q 1U 2I 2M 2Q 2U 3I 3M 3Q 3U)
+ vpunpckhwd ymmG, ymmG, ymmC ; ymmG=(03 07 0B 0F 13 17 1B 1F 23 27 2B 2F 33 37 3B 3F
+ ; 0J 0N 0R 0V 1J 1N 1R 1V 2J 2N 2R 2V 3J 3N 3R 3V)
+
+ vmovdqa ymmE, ymmA
+ vpunpcklbw ymmA, ymmA, ymmD ; ymmA=(00 02 04 06 08 0A 0C 0E 10 12 14 16 18 1A 1C 1E
+ ; 0G 0I 0K 0M 0O 0Q 0S 0U 1G 1I 1K 1M 1O 1Q 1S 1U)
+ vpunpckhbw ymmE, ymmE, ymmD ; ymmE=(20 22 24 26 28 2A 2C 2E 30 32 34 36 38 3A 3C 3E
+ ; 2G 2I 2K 2M 2O 2Q 2S 2U 3G 3I 3K 3M 3O 3Q 3S 3U)
+
+ vmovdqa ymmH, ymmB
+ vpunpcklbw ymmB, ymmB, ymmG ; ymmB=(01 03 05 07 09 0B 0D 0F 11 13 15 17 19 1B 1D 1F
+ ; 0H 0J 0L 0N 0P 0R 0T 0V 1H 1J 1L 1N 1P 1R 1T 1V)
+ vpunpckhbw ymmH, ymmH, ymmG ; ymmH=(21 23 25 27 29 2B 2D 2F 31 33 35 37 39 3B 3D 3F
+ ; 2H 2J 2L 2N 2P 2R 2T 2V 3H 3J 3L 3N 3P 3R 3T 3V)
+
+ vpxor ymmF, ymmF, ymmF
+
+ vmovdqa ymmC, ymmA
+ vpunpcklbw ymmA, ymmA, ymmF ; ymmA=(00 02 04 06 08 0A 0C 0E 0G 0I 0K 0M 0O 0Q 0S 0U)
+ vpunpckhbw ymmC, ymmC, ymmF ; ymmC=(10 12 14 16 18 1A 1C 1E 1G 1I 1K 1M 1O 1Q 1S 1U)
+
+ vmovdqa ymmD, ymmB
+ vpunpcklbw ymmB, ymmB, ymmF ; ymmB=(01 03 05 07 09 0B 0D 0F 0H 0J 0L 0N 0P 0R 0T 0V)
+ vpunpckhbw ymmD, ymmD, ymmF ; ymmD=(11 13 15 17 19 1B 1D 1F 1H 1J 1L 1N 1P 1R 1T 1V)
+
+ vmovdqa ymmG, ymmE
+ vpunpcklbw ymmE, ymmE, ymmF ; ymmE=(20 22 24 26 28 2A 2C 2E 2G 2I 2K 2M 2O 2Q 2S 2U)
+ vpunpckhbw ymmG, ymmG, ymmF ; ymmG=(30 32 34 36 38 3A 3C 3E 3G 3I 3K 3M 3O 3Q 3S 3U)
+
+ vpunpcklbw ymmF, ymmF, ymmH
+ vpunpckhbw ymmH, ymmH, ymmH
+ vpsrlw ymmF, ymmF, BYTE_BIT ; ymmF=(21 23 25 27 29 2B 2D 2F 2H 2J 2L 2N 2P 2R 2T 2V)
+ vpsrlw ymmH, ymmH, BYTE_BIT ; ymmH=(31 33 35 37 39 3B 3D 3F 3H 3J 3L 3N 3P 3R 3T 3V)
+
+%endif ; RGB_PIXELSIZE ; ---------------
+
+ ; ymm0=R(02468ACEGIKMOQSU)=RE, ymm2=G(02468ACEGIKMOQSU)=GE, ymm4=B(02468ACEGIKMOQSU)=BE
+ ; ymm1=R(13579BDFHJLNPRTV)=RO, ymm3=G(13579BDFHJLNPRTV)=GO, ymm5=B(13579BDFHJLNPRTV)=BO
+
+ ; (Original)
+ ; Y = 0.29900 * R + 0.58700 * G + 0.11400 * B
+ ;
+ ; (This implementation)
+ ; Y = 0.29900 * R + 0.33700 * G + 0.11400 * B + 0.25000 * G
+
+ vmovdqa ymm6, ymm1
+ vpunpcklwd ymm1, ymm1, ymm3
+ vpunpckhwd ymm6, ymm6, ymm3
+ vpmaddwd ymm1, ymm1, [rel PW_F0299_F0337] ; ymm1=ROL*FIX(0.299)+GOL*FIX(0.337)
+ vpmaddwd ymm6, ymm6, [rel PW_F0299_F0337] ; ymm6=ROH*FIX(0.299)+GOH*FIX(0.337)
+
+ vmovdqa ymm7, ymm6 ; ymm7=ROH*FIX(0.299)+GOH*FIX(0.337)
+
+ vmovdqa ymm6, ymm0
+ vpunpcklwd ymm0, ymm0, ymm2
+ vpunpckhwd ymm6, ymm6, ymm2
+ vpmaddwd ymm0, ymm0, [rel PW_F0299_F0337] ; ymm0=REL*FIX(0.299)+GEL*FIX(0.337)
+ vpmaddwd ymm6, ymm6, [rel PW_F0299_F0337] ; ymm6=REH*FIX(0.299)+GEH*FIX(0.337)
+
+ vmovdqa YMMWORD [wk(0)], ymm0 ; wk(0)=REL*FIX(0.299)+GEL*FIX(0.337)
+ vmovdqa YMMWORD [wk(1)], ymm6 ; wk(1)=REH*FIX(0.299)+GEH*FIX(0.337)
+
+ vmovdqa ymm0, ymm5 ; ymm0=BO
+ vmovdqa ymm6, ymm4 ; ymm6=BE
+
+ vmovdqa ymm4, ymm0
+ vpunpcklwd ymm0, ymm0, ymm3
+ vpunpckhwd ymm4, ymm4, ymm3
+ vpmaddwd ymm0, ymm0, [rel PW_F0114_F0250] ; ymm0=BOL*FIX(0.114)+GOL*FIX(0.250)
+ vpmaddwd ymm4, ymm4, [rel PW_F0114_F0250] ; ymm4=BOH*FIX(0.114)+GOH*FIX(0.250)
+
+ vmovdqa ymm3, [rel PD_ONEHALF] ; ymm3=[PD_ONEHALF]
+
+ vpaddd ymm0, ymm0, ymm1
+ vpaddd ymm4, ymm4, ymm7
+ vpaddd ymm0, ymm0, ymm3
+ vpaddd ymm4, ymm4, ymm3
+ vpsrld ymm0, ymm0, SCALEBITS ; ymm0=YOL
+ vpsrld ymm4, ymm4, SCALEBITS ; ymm4=YOH
+ vpackssdw ymm0, ymm0, ymm4 ; ymm0=YO
+
+ vmovdqa ymm4, ymm6
+ vpunpcklwd ymm6, ymm6, ymm2
+ vpunpckhwd ymm4, ymm4, ymm2
+ vpmaddwd ymm6, ymm6, [rel PW_F0114_F0250] ; ymm6=BEL*FIX(0.114)+GEL*FIX(0.250)
+ vpmaddwd ymm4, ymm4, [rel PW_F0114_F0250] ; ymm4=BEH*FIX(0.114)+GEH*FIX(0.250)
+
+ vmovdqa ymm2, [rel PD_ONEHALF] ; ymm2=[PD_ONEHALF]
+
+ vpaddd ymm6, ymm6, YMMWORD [wk(0)]
+ vpaddd ymm4, ymm4, YMMWORD [wk(1)]
+ vpaddd ymm6, ymm6, ymm2
+ vpaddd ymm4, ymm4, ymm2
+ vpsrld ymm6, ymm6, SCALEBITS ; ymm6=YEL
+ vpsrld ymm4, ymm4, SCALEBITS ; ymm4=YEH
+ vpackssdw ymm6, ymm6, ymm4 ; ymm6=YE
+
+ vpsllw ymm0, ymm0, BYTE_BIT
+ vpor ymm6, ymm6, ymm0 ; ymm6=Y
+ vmovdqu YMMWORD [rdi], ymm6 ; Save Y
+
+ sub rcx, byte SIZEOF_YMMWORD
+ add rsi, RGB_PIXELSIZE*SIZEOF_YMMWORD ; inptr
+ add rdi, byte SIZEOF_YMMWORD ; outptr0
+ cmp rcx, byte SIZEOF_YMMWORD
+ jae near .columnloop
+ test rcx, rcx
+ jnz near .column_ld1
+
+ pop rcx ; col
+ pop rsi
+ pop rdi
+
+ add rsi, byte SIZEOF_JSAMPROW ; input_buf
+ add rdi, byte SIZEOF_JSAMPROW
+ dec rax ; num_rows
+ jg near .rowloop
+
+.return:
+ pop rbx
+ vzeroupper
+ uncollect_args 5
+ mov rsp, rbp ; rsp <- aligned rbp
+ pop rsp ; rsp <- original rbp
+ pop rbp
+ ret
+
+; For some reason, the OS X linker does not honor the request to align the
+; segment unless we do this.
+ align 32
diff --git a/simd/x86_64/jcgryext-sse2.asm b/simd/x86_64/jcgryext-sse2.asm
new file mode 100644
index 0000000..d26ada3
--- /dev/null
+++ b/simd/x86_64/jcgryext-sse2.asm
@@ -0,0 +1,365 @@
+;
+; jcgryext.asm - grayscale colorspace conversion (64-bit SSE2)
+;
+; Copyright (C) 2011, 2016, D. R. Commander.
+;
+; Based on the x86 SIMD extension for IJG JPEG library
+; Copyright (C) 1999-2006, MIYASAKA Masaru.
+; For conditions of distribution and use, see copyright notice in jsimdext.inc
+;
+; This file should be assembled with NASM (Netwide Assembler),
+; can *not* be assembled with Microsoft's MASM or any compatible
+; assembler (including Borland's Turbo Assembler).
+; NASM is available from http://nasm.sourceforge.net/ or
+; http://sourceforge.net/project/showfiles.php?group_id=6208
+;
+; [TAB8]
+
+%include "jcolsamp.inc"
+
+; --------------------------------------------------------------------------
+;
+; Convert some rows of samples to the output colorspace.
+;
+; GLOBAL(void)
+; jsimd_rgb_gray_convert_sse2 (JDIMENSION img_width,
+; JSAMPARRAY input_buf, JSAMPIMAGE output_buf,
+; JDIMENSION output_row, int num_rows);
+;
+
+; r10d = JDIMENSION img_width
+; r11 = JSAMPARRAY input_buf
+; r12 = JSAMPIMAGE output_buf
+; r13d = JDIMENSION output_row
+; r14d = int num_rows
+
+%define wk(i) rbp-(WK_NUM-(i))*SIZEOF_XMMWORD ; xmmword wk[WK_NUM]
+%define WK_NUM 2
+
+ align 32
+
+ global EXTN(jsimd_rgb_gray_convert_sse2)
+
+EXTN(jsimd_rgb_gray_convert_sse2):
+ push rbp
+ mov rax, rsp ; rax = original rbp
+ sub rsp, byte 4
+ and rsp, byte (-SIZEOF_XMMWORD) ; align to 128 bits
+ mov [rsp], rax
+ mov rbp, rsp ; rbp = aligned rbp
+ lea rsp, [wk(0)]
+ collect_args 5
+ push rbx
+
+ mov ecx, r10d
+ test rcx, rcx
+ jz near .return
+
+ push rcx
+
+ mov rsi, r12
+ mov ecx, r13d
+ mov rdi, JSAMPARRAY [rsi+0*SIZEOF_JSAMPARRAY]
+ lea rdi, [rdi+rcx*SIZEOF_JSAMPROW]
+
+ pop rcx
+
+ mov rsi, r11
+ mov eax, r14d
+ test rax, rax
+ jle near .return
+.rowloop:
+ push rdi
+ push rsi
+ push rcx ; col
+
+ mov rsi, JSAMPROW [rsi] ; inptr
+ mov rdi, JSAMPROW [rdi] ; outptr0
+
+ cmp rcx, byte SIZEOF_XMMWORD
+ jae near .columnloop
+
+%if RGB_PIXELSIZE == 3 ; ---------------
+
+.column_ld1:
+ push rax
+ push rdx
+ lea rcx, [rcx+rcx*2] ; imul ecx,RGB_PIXELSIZE
+ test cl, SIZEOF_BYTE
+ jz short .column_ld2
+ sub rcx, byte SIZEOF_BYTE
+ movzx rax, BYTE [rsi+rcx]
+.column_ld2:
+ test cl, SIZEOF_WORD
+ jz short .column_ld4
+ sub rcx, byte SIZEOF_WORD
+ movzx rdx, WORD [rsi+rcx]
+ shl rax, WORD_BIT
+ or rax, rdx
+.column_ld4:
+ movd xmmA, eax
+ pop rdx
+ pop rax
+ test cl, SIZEOF_DWORD
+ jz short .column_ld8
+ sub rcx, byte SIZEOF_DWORD
+ movd xmmF, XMM_DWORD [rsi+rcx]
+ pslldq xmmA, SIZEOF_DWORD
+ por xmmA, xmmF
+.column_ld8:
+ test cl, SIZEOF_MMWORD
+ jz short .column_ld16
+ sub rcx, byte SIZEOF_MMWORD
+ movq xmmB, XMM_MMWORD [rsi+rcx]
+ pslldq xmmA, SIZEOF_MMWORD
+ por xmmA, xmmB
+.column_ld16:
+ test cl, SIZEOF_XMMWORD
+ jz short .column_ld32
+ movdqa xmmF, xmmA
+ movdqu xmmA, XMMWORD [rsi+0*SIZEOF_XMMWORD]
+ mov rcx, SIZEOF_XMMWORD
+ jmp short .rgb_gray_cnv
+.column_ld32:
+ test cl, 2*SIZEOF_XMMWORD
+ mov rcx, SIZEOF_XMMWORD
+ jz short .rgb_gray_cnv
+ movdqa xmmB, xmmA
+ movdqu xmmA, XMMWORD [rsi+0*SIZEOF_XMMWORD]
+ movdqu xmmF, XMMWORD [rsi+1*SIZEOF_XMMWORD]
+ jmp short .rgb_gray_cnv
+
+.columnloop:
+ movdqu xmmA, XMMWORD [rsi+0*SIZEOF_XMMWORD]
+ movdqu xmmF, XMMWORD [rsi+1*SIZEOF_XMMWORD]
+ movdqu xmmB, XMMWORD [rsi+2*SIZEOF_XMMWORD]
+
+.rgb_gray_cnv:
+ ; xmmA=(00 10 20 01 11 21 02 12 22 03 13 23 04 14 24 05)
+ ; xmmF=(15 25 06 16 26 07 17 27 08 18 28 09 19 29 0A 1A)
+ ; xmmB=(2A 0B 1B 2B 0C 1C 2C 0D 1D 2D 0E 1E 2E 0F 1F 2F)
+
+ movdqa xmmG, xmmA
+ pslldq xmmA, 8 ; xmmA=(-- -- -- -- -- -- -- -- 00 10 20 01 11 21 02 12)
+ psrldq xmmG, 8 ; xmmG=(22 03 13 23 04 14 24 05 -- -- -- -- -- -- -- --)
+
+ punpckhbw xmmA, xmmF ; xmmA=(00 08 10 18 20 28 01 09 11 19 21 29 02 0A 12 1A)
+ pslldq xmmF, 8 ; xmmF=(-- -- -- -- -- -- -- -- 15 25 06 16 26 07 17 27)
+
+ punpcklbw xmmG, xmmB ; xmmG=(22 2A 03 0B 13 1B 23 2B 04 0C 14 1C 24 2C 05 0D)
+ punpckhbw xmmF, xmmB ; xmmF=(15 1D 25 2D 06 0E 16 1E 26 2E 07 0F 17 1F 27 2F)
+
+ movdqa xmmD, xmmA
+ pslldq xmmA, 8 ; xmmA=(-- -- -- -- -- -- -- -- 00 08 10 18 20 28 01 09)
+ psrldq xmmD, 8 ; xmmD=(11 19 21 29 02 0A 12 1A -- -- -- -- -- -- -- --)
+
+ punpckhbw xmmA, xmmG ; xmmA=(00 04 08 0C 10 14 18 1C 20 24 28 2C 01 05 09 0D)
+ pslldq xmmG, 8 ; xmmG=(-- -- -- -- -- -- -- -- 22 2A 03 0B 13 1B 23 2B)
+
+ punpcklbw xmmD, xmmF ; xmmD=(11 15 19 1D 21 25 29 2D 02 06 0A 0E 12 16 1A 1E)
+ punpckhbw xmmG, xmmF ; xmmG=(22 26 2A 2E 03 07 0B 0F 13 17 1B 1F 23 27 2B 2F)
+
+ movdqa xmmE, xmmA
+ pslldq xmmA, 8 ; xmmA=(-- -- -- -- -- -- -- -- 00 04 08 0C 10 14 18 1C)
+ psrldq xmmE, 8 ; xmmE=(20 24 28 2C 01 05 09 0D -- -- -- -- -- -- -- --)
+
+ punpckhbw xmmA, xmmD ; xmmA=(00 02 04 06 08 0A 0C 0E 10 12 14 16 18 1A 1C 1E)
+ pslldq xmmD, 8 ; xmmD=(-- -- -- -- -- -- -- -- 11 15 19 1D 21 25 29 2D)
+
+ punpcklbw xmmE, xmmG ; xmmE=(20 22 24 26 28 2A 2C 2E 01 03 05 07 09 0B 0D 0F)
+ punpckhbw xmmD, xmmG ; xmmD=(11 13 15 17 19 1B 1D 1F 21 23 25 27 29 2B 2D 2F)
+
+ pxor xmmH, xmmH
+
+ movdqa xmmC, xmmA
+ punpcklbw xmmA, xmmH ; xmmA=(00 02 04 06 08 0A 0C 0E)
+ punpckhbw xmmC, xmmH ; xmmC=(10 12 14 16 18 1A 1C 1E)
+
+ movdqa xmmB, xmmE
+ punpcklbw xmmE, xmmH ; xmmE=(20 22 24 26 28 2A 2C 2E)
+ punpckhbw xmmB, xmmH ; xmmB=(01 03 05 07 09 0B 0D 0F)
+
+ movdqa xmmF, xmmD
+ punpcklbw xmmD, xmmH ; xmmD=(11 13 15 17 19 1B 1D 1F)
+ punpckhbw xmmF, xmmH ; xmmF=(21 23 25 27 29 2B 2D 2F)
+
+%else ; RGB_PIXELSIZE == 4 ; -----------
+
+.column_ld1:
+ test cl, SIZEOF_XMMWORD/16
+ jz short .column_ld2
+ sub rcx, byte SIZEOF_XMMWORD/16
+ movd xmmA, XMM_DWORD [rsi+rcx*RGB_PIXELSIZE]
+.column_ld2:
+ test cl, SIZEOF_XMMWORD/8
+ jz short .column_ld4
+ sub rcx, byte SIZEOF_XMMWORD/8
+ movq xmmE, XMM_MMWORD [rsi+rcx*RGB_PIXELSIZE]
+ pslldq xmmA, SIZEOF_MMWORD
+ por xmmA, xmmE
+.column_ld4:
+ test cl, SIZEOF_XMMWORD/4
+ jz short .column_ld8
+ sub rcx, byte SIZEOF_XMMWORD/4
+ movdqa xmmE, xmmA
+ movdqu xmmA, XMMWORD [rsi+rcx*RGB_PIXELSIZE]
+.column_ld8:
+ test cl, SIZEOF_XMMWORD/2
+ mov rcx, SIZEOF_XMMWORD
+ jz short .rgb_gray_cnv
+ movdqa xmmF, xmmA
+ movdqa xmmH, xmmE
+ movdqu xmmA, XMMWORD [rsi+0*SIZEOF_XMMWORD]
+ movdqu xmmE, XMMWORD [rsi+1*SIZEOF_XMMWORD]
+ jmp short .rgb_gray_cnv
+
+.columnloop:
+ movdqu xmmA, XMMWORD [rsi+0*SIZEOF_XMMWORD]
+ movdqu xmmE, XMMWORD [rsi+1*SIZEOF_XMMWORD]
+ movdqu xmmF, XMMWORD [rsi+2*SIZEOF_XMMWORD]
+ movdqu xmmH, XMMWORD [rsi+3*SIZEOF_XMMWORD]
+
+.rgb_gray_cnv:
+ ; xmmA=(00 10 20 30 01 11 21 31 02 12 22 32 03 13 23 33)
+ ; xmmE=(04 14 24 34 05 15 25 35 06 16 26 36 07 17 27 37)
+ ; xmmF=(08 18 28 38 09 19 29 39 0A 1A 2A 3A 0B 1B 2B 3B)
+ ; xmmH=(0C 1C 2C 3C 0D 1D 2D 3D 0E 1E 2E 3E 0F 1F 2F 3F)
+
+ movdqa xmmD, xmmA
+ punpcklbw xmmA, xmmE ; xmmA=(00 04 10 14 20 24 30 34 01 05 11 15 21 25 31 35)
+ punpckhbw xmmD, xmmE ; xmmD=(02 06 12 16 22 26 32 36 03 07 13 17 23 27 33 37)
+
+ movdqa xmmC, xmmF
+ punpcklbw xmmF, xmmH ; xmmF=(08 0C 18 1C 28 2C 38 3C 09 0D 19 1D 29 2D 39 3D)
+ punpckhbw xmmC, xmmH ; xmmC=(0A 0E 1A 1E 2A 2E 3A 3E 0B 0F 1B 1F 2B 2F 3B 3F)
+
+ movdqa xmmB, xmmA
+ punpcklwd xmmA, xmmF ; xmmA=(00 04 08 0C 10 14 18 1C 20 24 28 2C 30 34 38 3C)
+ punpckhwd xmmB, xmmF ; xmmB=(01 05 09 0D 11 15 19 1D 21 25 29 2D 31 35 39 3D)
+
+ movdqa xmmG, xmmD
+ punpcklwd xmmD, xmmC ; xmmD=(02 06 0A 0E 12 16 1A 1E 22 26 2A 2E 32 36 3A 3E)
+ punpckhwd xmmG, xmmC ; xmmG=(03 07 0B 0F 13 17 1B 1F 23 27 2B 2F 33 37 3B 3F)
+
+ movdqa xmmE, xmmA
+ punpcklbw xmmA, xmmD ; xmmA=(00 02 04 06 08 0A 0C 0E 10 12 14 16 18 1A 1C 1E)
+ punpckhbw xmmE, xmmD ; xmmE=(20 22 24 26 28 2A 2C 2E 30 32 34 36 38 3A 3C 3E)
+
+ movdqa xmmH, xmmB
+ punpcklbw xmmB, xmmG ; xmmB=(01 03 05 07 09 0B 0D 0F 11 13 15 17 19 1B 1D 1F)
+ punpckhbw xmmH, xmmG ; xmmH=(21 23 25 27 29 2B 2D 2F 31 33 35 37 39 3B 3D 3F)
+
+ pxor xmmF, xmmF
+
+ movdqa xmmC, xmmA
+ punpcklbw xmmA, xmmF ; xmmA=(00 02 04 06 08 0A 0C 0E)
+ punpckhbw xmmC, xmmF ; xmmC=(10 12 14 16 18 1A 1C 1E)
+
+ movdqa xmmD, xmmB
+ punpcklbw xmmB, xmmF ; xmmB=(01 03 05 07 09 0B 0D 0F)
+ punpckhbw xmmD, xmmF ; xmmD=(11 13 15 17 19 1B 1D 1F)
+
+ movdqa xmmG, xmmE
+ punpcklbw xmmE, xmmF ; xmmE=(20 22 24 26 28 2A 2C 2E)
+ punpckhbw xmmG, xmmF ; xmmG=(30 32 34 36 38 3A 3C 3E)
+
+ punpcklbw xmmF, xmmH
+ punpckhbw xmmH, xmmH
+ psrlw xmmF, BYTE_BIT ; xmmF=(21 23 25 27 29 2B 2D 2F)
+ psrlw xmmH, BYTE_BIT ; xmmH=(31 33 35 37 39 3B 3D 3F)
+
+%endif ; RGB_PIXELSIZE ; ---------------
+
+ ; xmm0=R(02468ACE)=RE, xmm2=G(02468ACE)=GE, xmm4=B(02468ACE)=BE
+ ; xmm1=R(13579BDF)=RO, xmm3=G(13579BDF)=GO, xmm5=B(13579BDF)=BO
+
+ ; (Original)
+ ; Y = 0.29900 * R + 0.58700 * G + 0.11400 * B
+ ;
+ ; (This implementation)
+ ; Y = 0.29900 * R + 0.33700 * G + 0.11400 * B + 0.25000 * G
+
+ movdqa xmm6, xmm1
+ punpcklwd xmm1, xmm3
+ punpckhwd xmm6, xmm3
+ pmaddwd xmm1, [rel PW_F0299_F0337] ; xmm1=ROL*FIX(0.299)+GOL*FIX(0.337)
+ pmaddwd xmm6, [rel PW_F0299_F0337] ; xmm6=ROH*FIX(0.299)+GOH*FIX(0.337)
+
+ movdqa xmm7, xmm6 ; xmm7=ROH*FIX(0.299)+GOH*FIX(0.337)
+
+ movdqa xmm6, xmm0
+ punpcklwd xmm0, xmm2
+ punpckhwd xmm6, xmm2
+ pmaddwd xmm0, [rel PW_F0299_F0337] ; xmm0=REL*FIX(0.299)+GEL*FIX(0.337)
+ pmaddwd xmm6, [rel PW_F0299_F0337] ; xmm6=REH*FIX(0.299)+GEH*FIX(0.337)
+
+ movdqa XMMWORD [wk(0)], xmm0 ; wk(0)=REL*FIX(0.299)+GEL*FIX(0.337)
+ movdqa XMMWORD [wk(1)], xmm6 ; wk(1)=REH*FIX(0.299)+GEH*FIX(0.337)
+
+ movdqa xmm0, xmm5 ; xmm0=BO
+ movdqa xmm6, xmm4 ; xmm6=BE
+
+ movdqa xmm4, xmm0
+ punpcklwd xmm0, xmm3
+ punpckhwd xmm4, xmm3
+ pmaddwd xmm0, [rel PW_F0114_F0250] ; xmm0=BOL*FIX(0.114)+GOL*FIX(0.250)
+ pmaddwd xmm4, [rel PW_F0114_F0250] ; xmm4=BOH*FIX(0.114)+GOH*FIX(0.250)
+
+ movdqa xmm3, [rel PD_ONEHALF] ; xmm3=[PD_ONEHALF]
+
+ paddd xmm0, xmm1
+ paddd xmm4, xmm7
+ paddd xmm0, xmm3
+ paddd xmm4, xmm3
+ psrld xmm0, SCALEBITS ; xmm0=YOL
+ psrld xmm4, SCALEBITS ; xmm4=YOH
+ packssdw xmm0, xmm4 ; xmm0=YO
+
+ movdqa xmm4, xmm6
+ punpcklwd xmm6, xmm2
+ punpckhwd xmm4, xmm2
+ pmaddwd xmm6, [rel PW_F0114_F0250] ; xmm6=BEL*FIX(0.114)+GEL*FIX(0.250)
+ pmaddwd xmm4, [rel PW_F0114_F0250] ; xmm4=BEH*FIX(0.114)+GEH*FIX(0.250)
+
+ movdqa xmm2, [rel PD_ONEHALF] ; xmm2=[PD_ONEHALF]
+
+ paddd xmm6, XMMWORD [wk(0)]
+ paddd xmm4, XMMWORD [wk(1)]
+ paddd xmm6, xmm2
+ paddd xmm4, xmm2
+ psrld xmm6, SCALEBITS ; xmm6=YEL
+ psrld xmm4, SCALEBITS ; xmm4=YEH
+ packssdw xmm6, xmm4 ; xmm6=YE
+
+ psllw xmm0, BYTE_BIT
+ por xmm6, xmm0 ; xmm6=Y
+ movdqa XMMWORD [rdi], xmm6 ; Save Y
+
+ sub rcx, byte SIZEOF_XMMWORD
+ add rsi, byte RGB_PIXELSIZE*SIZEOF_XMMWORD ; inptr
+ add rdi, byte SIZEOF_XMMWORD ; outptr0
+ cmp rcx, byte SIZEOF_XMMWORD
+ jae near .columnloop
+ test rcx, rcx
+ jnz near .column_ld1
+
+ pop rcx ; col
+ pop rsi
+ pop rdi
+
+ add rsi, byte SIZEOF_JSAMPROW ; input_buf
+ add rdi, byte SIZEOF_JSAMPROW
+ dec rax ; num_rows
+ jg near .rowloop
+
+.return:
+ pop rbx
+ uncollect_args 5
+ mov rsp, rbp ; rsp <- aligned rbp
+ pop rsp ; rsp <- original rbp
+ pop rbp
+ ret
+
+; For some reason, the OS X linker does not honor the request to align the
+; segment unless we do this.
+ align 32
diff --git a/simd/x86_64/jchuff-sse2.asm b/simd/x86_64/jchuff-sse2.asm
new file mode 100644
index 0000000..bedc639
--- /dev/null
+++ b/simd/x86_64/jchuff-sse2.asm
@@ -0,0 +1,348 @@
+;
+; jchuff-sse2.asm - Huffman entropy encoding (64-bit SSE2)
+;
+; Copyright (C) 2009-2011, 2014-2016, D. R. Commander.
+; Copyright (C) 2015, Matthieu Darbois.
+;
+; Based on the x86 SIMD extension for IJG JPEG library
+; Copyright (C) 1999-2006, MIYASAKA Masaru.
+; For conditions of distribution and use, see copyright notice in jsimdext.inc
+;
+; This file should be assembled with NASM (Netwide Assembler),
+; can *not* be assembled with Microsoft's MASM or any compatible
+; assembler (including Borland's Turbo Assembler).
+; NASM is available from http://nasm.sourceforge.net/ or
+; http://sourceforge.net/project/showfiles.php?group_id=6208
+;
+; This file contains an SSE2 implementation for Huffman coding of one block.
+; The following code is based directly on jchuff.c; see jchuff.c for more
+; details.
+;
+; [TAB8]
+
+%include "jsimdext.inc"
+
+; --------------------------------------------------------------------------
+ SECTION SEG_CONST
+
+ alignz 32
+ global EXTN(jconst_huff_encode_one_block)
+
+EXTN(jconst_huff_encode_one_block):
+
+%include "jpeg_nbits_table.inc"
+
+ alignz 32
+
+; --------------------------------------------------------------------------
+ SECTION SEG_TEXT
+ BITS 64
+
+; These macros perform the same task as the emit_bits() function in the
+; original libjpeg code. In addition to reducing overhead by explicitly
+; inlining the code, additional performance is achieved by taking into
+; account the size of the bit buffer and waiting until it is almost full
+; before emptying it. This mostly benefits 64-bit platforms, since 6
+; bytes can be stored in a 64-bit bit buffer before it has to be emptied.
+
+%macro EMIT_BYTE 0
+ sub put_bits, 8 ; put_bits -= 8;
+ mov rdx, put_buffer
+ mov ecx, put_bits
+ shr rdx, cl ; c = (JOCTET)GETJOCTET(put_buffer >> put_bits);
+ mov byte [buffer], dl ; *buffer++ = c;
+ add buffer, 1
+ cmp dl, 0xFF ; need to stuff a zero byte?
+ jne %%.EMIT_BYTE_END
+ mov byte [buffer], 0 ; *buffer++ = 0;
+ add buffer, 1
+%%.EMIT_BYTE_END:
+%endmacro
+
+%macro PUT_BITS 1
+ add put_bits, ecx ; put_bits += size;
+ shl put_buffer, cl ; put_buffer = (put_buffer << size);
+ or put_buffer, %1
+%endmacro
+
+%macro CHECKBUF31 0
+ cmp put_bits, 32 ; if (put_bits > 31) {
+ jl %%.CHECKBUF31_END
+ EMIT_BYTE
+ EMIT_BYTE
+ EMIT_BYTE
+ EMIT_BYTE
+%%.CHECKBUF31_END:
+%endmacro
+
+%macro CHECKBUF47 0
+ cmp put_bits, 48 ; if (put_bits > 47) {
+ jl %%.CHECKBUF47_END
+ EMIT_BYTE
+ EMIT_BYTE
+ EMIT_BYTE
+ EMIT_BYTE
+ EMIT_BYTE
+ EMIT_BYTE
+%%.CHECKBUF47_END:
+%endmacro
+
+%macro EMIT_BITS 2
+ CHECKBUF47
+ mov ecx, %2
+ PUT_BITS %1
+%endmacro
+
+%macro kloop_prepare 37 ;(ko, jno0, ..., jno31, xmm0, xmm1, xmm2, xmm3)
+ pxor xmm8, xmm8 ; __m128i neg = _mm_setzero_si128();
+ pxor xmm9, xmm9 ; __m128i neg = _mm_setzero_si128();
+ pxor xmm10, xmm10 ; __m128i neg = _mm_setzero_si128();
+ pxor xmm11, xmm11 ; __m128i neg = _mm_setzero_si128();
+ pinsrw %34, word [r12 + %2 * SIZEOF_WORD], 0 ; xmm_shadow[0] = block[jno0];
+ pinsrw %35, word [r12 + %10 * SIZEOF_WORD], 0 ; xmm_shadow[8] = block[jno8];
+ pinsrw %36, word [r12 + %18 * SIZEOF_WORD], 0 ; xmm_shadow[16] = block[jno16];
+ pinsrw %37, word [r12 + %26 * SIZEOF_WORD], 0 ; xmm_shadow[24] = block[jno24];
+ pinsrw %34, word [r12 + %3 * SIZEOF_WORD], 1 ; xmm_shadow[1] = block[jno1];
+ pinsrw %35, word [r12 + %11 * SIZEOF_WORD], 1 ; xmm_shadow[9] = block[jno9];
+ pinsrw %36, word [r12 + %19 * SIZEOF_WORD], 1 ; xmm_shadow[17] = block[jno17];
+ pinsrw %37, word [r12 + %27 * SIZEOF_WORD], 1 ; xmm_shadow[25] = block[jno25];
+ pinsrw %34, word [r12 + %4 * SIZEOF_WORD], 2 ; xmm_shadow[2] = block[jno2];
+ pinsrw %35, word [r12 + %12 * SIZEOF_WORD], 2 ; xmm_shadow[10] = block[jno10];
+ pinsrw %36, word [r12 + %20 * SIZEOF_WORD], 2 ; xmm_shadow[18] = block[jno18];
+ pinsrw %37, word [r12 + %28 * SIZEOF_WORD], 2 ; xmm_shadow[26] = block[jno26];
+ pinsrw %34, word [r12 + %5 * SIZEOF_WORD], 3 ; xmm_shadow[3] = block[jno3];
+ pinsrw %35, word [r12 + %13 * SIZEOF_WORD], 3 ; xmm_shadow[11] = block[jno11];
+ pinsrw %36, word [r12 + %21 * SIZEOF_WORD], 3 ; xmm_shadow[19] = block[jno19];
+ pinsrw %37, word [r12 + %29 * SIZEOF_WORD], 3 ; xmm_shadow[27] = block[jno27];
+ pinsrw %34, word [r12 + %6 * SIZEOF_WORD], 4 ; xmm_shadow[4] = block[jno4];
+ pinsrw %35, word [r12 + %14 * SIZEOF_WORD], 4 ; xmm_shadow[12] = block[jno12];
+ pinsrw %36, word [r12 + %22 * SIZEOF_WORD], 4 ; xmm_shadow[20] = block[jno20];
+ pinsrw %37, word [r12 + %30 * SIZEOF_WORD], 4 ; xmm_shadow[28] = block[jno28];
+ pinsrw %34, word [r12 + %7 * SIZEOF_WORD], 5 ; xmm_shadow[5] = block[jno5];
+ pinsrw %35, word [r12 + %15 * SIZEOF_WORD], 5 ; xmm_shadow[13] = block[jno13];
+ pinsrw %36, word [r12 + %23 * SIZEOF_WORD], 5 ; xmm_shadow[21] = block[jno21];
+ pinsrw %37, word [r12 + %31 * SIZEOF_WORD], 5 ; xmm_shadow[29] = block[jno29];
+ pinsrw %34, word [r12 + %8 * SIZEOF_WORD], 6 ; xmm_shadow[6] = block[jno6];
+ pinsrw %35, word [r12 + %16 * SIZEOF_WORD], 6 ; xmm_shadow[14] = block[jno14];
+ pinsrw %36, word [r12 + %24 * SIZEOF_WORD], 6 ; xmm_shadow[22] = block[jno22];
+ pinsrw %37, word [r12 + %32 * SIZEOF_WORD], 6 ; xmm_shadow[30] = block[jno30];
+ pinsrw %34, word [r12 + %9 * SIZEOF_WORD], 7 ; xmm_shadow[7] = block[jno7];
+ pinsrw %35, word [r12 + %17 * SIZEOF_WORD], 7 ; xmm_shadow[15] = block[jno15];
+ pinsrw %36, word [r12 + %25 * SIZEOF_WORD], 7 ; xmm_shadow[23] = block[jno23];
+%if %1 != 32
+ pinsrw %37, word [r12 + %33 * SIZEOF_WORD], 7 ; xmm_shadow[31] = block[jno31];
+%else
+ pinsrw %37, ebx, 7 ; xmm_shadow[31] = block[jno31];
+%endif
+ pcmpgtw xmm8, %34 ; neg = _mm_cmpgt_epi16(neg, x1);
+ pcmpgtw xmm9, %35 ; neg = _mm_cmpgt_epi16(neg, x1);
+ pcmpgtw xmm10, %36 ; neg = _mm_cmpgt_epi16(neg, x1);
+ pcmpgtw xmm11, %37 ; neg = _mm_cmpgt_epi16(neg, x1);
+ paddw %34, xmm8 ; x1 = _mm_add_epi16(x1, neg);
+ paddw %35, xmm9 ; x1 = _mm_add_epi16(x1, neg);
+ paddw %36, xmm10 ; x1 = _mm_add_epi16(x1, neg);
+ paddw %37, xmm11 ; x1 = _mm_add_epi16(x1, neg);
+ pxor %34, xmm8 ; x1 = _mm_xor_si128(x1, neg);
+ pxor %35, xmm9 ; x1 = _mm_xor_si128(x1, neg);
+ pxor %36, xmm10 ; x1 = _mm_xor_si128(x1, neg);
+ pxor %37, xmm11 ; x1 = _mm_xor_si128(x1, neg);
+ pxor xmm8, %34 ; neg = _mm_xor_si128(neg, x1);
+ pxor xmm9, %35 ; neg = _mm_xor_si128(neg, x1);
+ pxor xmm10, %36 ; neg = _mm_xor_si128(neg, x1);
+ pxor xmm11, %37 ; neg = _mm_xor_si128(neg, x1);
+ movdqa XMMWORD [t1 + %1 * SIZEOF_WORD], %34 ; _mm_storeu_si128((__m128i *)(t1 + ko), x1);
+ movdqa XMMWORD [t1 + (%1 + 8) * SIZEOF_WORD], %35 ; _mm_storeu_si128((__m128i *)(t1 + ko + 8), x1);
+ movdqa XMMWORD [t1 + (%1 + 16) * SIZEOF_WORD], %36 ; _mm_storeu_si128((__m128i *)(t1 + ko + 16), x1);
+ movdqa XMMWORD [t1 + (%1 + 24) * SIZEOF_WORD], %37 ; _mm_storeu_si128((__m128i *)(t1 + ko + 24), x1);
+ movdqa XMMWORD [t2 + %1 * SIZEOF_WORD], xmm8 ; _mm_storeu_si128((__m128i *)(t2 + ko), neg);
+ movdqa XMMWORD [t2 + (%1 + 8) * SIZEOF_WORD], xmm9 ; _mm_storeu_si128((__m128i *)(t2 + ko + 8), neg);
+ movdqa XMMWORD [t2 + (%1 + 16) * SIZEOF_WORD], xmm10 ; _mm_storeu_si128((__m128i *)(t2 + ko + 16), neg);
+ movdqa XMMWORD [t2 + (%1 + 24) * SIZEOF_WORD], xmm11 ; _mm_storeu_si128((__m128i *)(t2 + ko + 24), neg);
+%endmacro
+
+;
+; Encode a single block's worth of coefficients.
+;
+; GLOBAL(JOCTET*)
+; jsimd_huff_encode_one_block_sse2 (working_state *state, JOCTET *buffer,
+; JCOEFPTR block, int last_dc_val,
+; c_derived_tbl *dctbl, c_derived_tbl *actbl)
+;
+
+; r10 = working_state *state
+; r11 = JOCTET *buffer
+; r12 = JCOEFPTR block
+; r13d = int last_dc_val
+; r14 = c_derived_tbl *dctbl
+; r15 = c_derived_tbl *actbl
+
+%define t1 rbp-(DCTSIZE2*SIZEOF_WORD)
+%define t2 t1-(DCTSIZE2*SIZEOF_WORD)
+%define put_buffer r8
+%define put_bits r9d
+%define buffer rax
+
+ align 32
+ global EXTN(jsimd_huff_encode_one_block_sse2)
+
+EXTN(jsimd_huff_encode_one_block_sse2):
+ push rbp
+ mov rax, rsp ; rax = original rbp
+ sub rsp, byte 4
+ and rsp, byte (-SIZEOF_XMMWORD) ; align to 128 bits
+ mov [rsp], rax
+ mov rbp,rsp ; rbp = aligned rbp
+ lea rsp, [t2]
+ push_xmm 4
+ collect_args 6
+ push rbx
+
+ mov buffer, r11 ; r11 is now sratch
+
+ mov put_buffer, MMWORD [r10+16] ; put_buffer = state->cur.put_buffer;
+ mov put_bits, DWORD [r10+24] ; put_bits = state->cur.put_bits;
+ push r10 ; r10 is now scratch
+
+ ; Encode the DC coefficient difference per section F.1.2.1
+ movsx edi, word [r12] ; temp = temp2 = block[0] - last_dc_val;
+ sub edi, r13d ; r13 is not used anymore
+ mov ebx, edi
+
+ ; This is a well-known technique for obtaining the absolute value
+ ; without a branch. It is derived from an assembly language technique
+ ; presented in "How to Optimize for the Pentium Processors",
+ ; Copyright (c) 1996, 1997 by Agner Fog.
+ mov esi, edi
+ sar esi, 31 ; temp3 = temp >> (CHAR_BIT * sizeof(int) - 1);
+ xor edi, esi ; temp ^= temp3;
+ sub edi, esi ; temp -= temp3;
+
+ ; For a negative input, want temp2 = bitwise complement of abs(input)
+ ; This code assumes we are on a two's complement machine
+ add ebx, esi ; temp2 += temp3;
+
+ ; Find the number of bits needed for the magnitude of the coefficient
+ lea r11, [rel jpeg_nbits_table]
+ movzx rdi, byte [r11 + rdi] ; nbits = JPEG_NBITS(temp);
+ ; Emit the Huffman-coded symbol for the number of bits
+ mov r11d, INT [r14 + rdi * 4] ; code = dctbl->ehufco[nbits];
+ movzx esi, byte [r14 + rdi + 1024] ; size = dctbl->ehufsi[nbits];
+ EMIT_BITS r11, esi ; EMIT_BITS(code, size)
+
+ ; Mask off any extra bits in code
+ mov esi, 1
+ mov ecx, edi
+ shl esi, cl
+ dec esi
+ and ebx, esi ; temp2 &= (((JLONG) 1)<<nbits) - 1;
+
+ ; Emit that number of bits of the value, if positive,
+ ; or the complement of its magnitude, if negative.
+ EMIT_BITS rbx, edi ; EMIT_BITS(temp2, nbits)
+
+ ; Prepare data
+ xor ebx, ebx
+ kloop_prepare 0, 1, 8, 16, 9, 2, 3, 10, 17, 24, 32, 25, \
+ 18, 11, 4, 5, 12, 19, 26, 33, 40, 48, 41, 34, \
+ 27, 20, 13, 6, 7, 14, 21, 28, 35, \
+ xmm0, xmm1, xmm2, xmm3
+ kloop_prepare 32, 42, 49, 56, 57, 50, 43, 36, 29, 22, 15, 23, \
+ 30, 37, 44, 51, 58, 59, 52, 45, 38, 31, 39, 46, \
+ 53, 60, 61, 54, 47, 55, 62, 63, 63, \
+ xmm4, xmm5, xmm6, xmm7
+
+ pxor xmm8, xmm8
+ pcmpeqw xmm0, xmm8 ; tmp0 = _mm_cmpeq_epi16(tmp0, zero);
+ pcmpeqw xmm1, xmm8 ; tmp1 = _mm_cmpeq_epi16(tmp1, zero);
+ pcmpeqw xmm2, xmm8 ; tmp2 = _mm_cmpeq_epi16(tmp2, zero);
+ pcmpeqw xmm3, xmm8 ; tmp3 = _mm_cmpeq_epi16(tmp3, zero);
+ pcmpeqw xmm4, xmm8 ; tmp4 = _mm_cmpeq_epi16(tmp4, zero);
+ pcmpeqw xmm5, xmm8 ; tmp5 = _mm_cmpeq_epi16(tmp5, zero);
+ pcmpeqw xmm6, xmm8 ; tmp6 = _mm_cmpeq_epi16(tmp6, zero);
+ pcmpeqw xmm7, xmm8 ; tmp7 = _mm_cmpeq_epi16(tmp7, zero);
+ packsswb xmm0, xmm1 ; tmp0 = _mm_packs_epi16(tmp0, tmp1);
+ packsswb xmm2, xmm3 ; tmp2 = _mm_packs_epi16(tmp2, tmp3);
+ packsswb xmm4, xmm5 ; tmp4 = _mm_packs_epi16(tmp4, tmp5);
+ packsswb xmm6, xmm7 ; tmp6 = _mm_packs_epi16(tmp6, tmp7);
+ pmovmskb r11d, xmm0 ; index = ((uint64_t)_mm_movemask_epi8(tmp0)) << 0;
+ pmovmskb r12d, xmm2 ; index = ((uint64_t)_mm_movemask_epi8(tmp2)) << 16;
+ pmovmskb r13d, xmm4 ; index = ((uint64_t)_mm_movemask_epi8(tmp4)) << 32;
+ pmovmskb r14d, xmm6 ; index = ((uint64_t)_mm_movemask_epi8(tmp6)) << 48;
+ shl r12, 16
+ shl r14, 16
+ or r11, r12
+ or r13, r14
+ shl r13, 32
+ or r11, r13
+ not r11 ; index = ~index;
+
+ ;mov MMWORD [ t1 + DCTSIZE2 * SIZEOF_WORD ], r11
+ ;jmp .EFN
+
+ mov r13d, INT [r15 + 240 * 4] ; code_0xf0 = actbl->ehufco[0xf0];
+ movzx r14d, byte [r15 + 1024 + 240] ; size_0xf0 = actbl->ehufsi[0xf0];
+ lea rsi, [t1]
+.BLOOP:
+ bsf r12, r11 ; r = __builtin_ctzl(index);
+ jz .ELOOP
+ mov rcx, r12
+ lea rsi, [rsi+r12*2] ; k += r;
+ shr r11, cl ; index >>= r;
+ movzx rdi, word [rsi] ; temp = t1[k];
+ lea rbx, [rel jpeg_nbits_table]
+ movzx rdi, byte [rbx + rdi] ; nbits = JPEG_NBITS(temp);
+.BRLOOP:
+ cmp r12, 16 ; while (r > 15) {
+ jl .ERLOOP
+ EMIT_BITS r13, r14d ; EMIT_BITS(code_0xf0, size_0xf0)
+ sub r12, 16 ; r -= 16;
+ jmp .BRLOOP
+.ERLOOP:
+ ; Emit Huffman symbol for run length / number of bits
+ CHECKBUF31 ; uses rcx, rdx
+
+ shl r12, 4 ; temp3 = (r << 4) + nbits;
+ add r12, rdi
+ mov ebx, INT [r15 + r12 * 4] ; code = actbl->ehufco[temp3];
+ movzx ecx, byte [r15 + r12 + 1024] ; size = actbl->ehufsi[temp3];
+ PUT_BITS rbx
+
+ ;EMIT_CODE(code, size)
+
+ movsx ebx, word [rsi-DCTSIZE2*2] ; temp2 = t2[k];
+ ; Mask off any extra bits in code
+ mov rcx, rdi
+ mov rdx, 1
+ shl rdx, cl
+ dec rdx
+ and rbx, rdx ; temp2 &= (((JLONG) 1)<<nbits) - 1;
+ PUT_BITS rbx ; PUT_BITS(temp2, nbits)
+
+ shr r11, 1 ; index >>= 1;
+ add rsi, 2 ; ++k;
+ jmp .BLOOP
+.ELOOP:
+ ; If the last coef(s) were zero, emit an end-of-block code
+ lea rdi, [t1 + (DCTSIZE2-1) * 2] ; r = DCTSIZE2-1-k;
+ cmp rdi, rsi ; if (r > 0) {
+ je .EFN
+ mov ebx, INT [r15] ; code = actbl->ehufco[0];
+ movzx r12d, byte [r15 + 1024] ; size = actbl->ehufsi[0];
+ EMIT_BITS rbx, r12d
+.EFN:
+ pop r10
+ ; Save put_buffer & put_bits
+ mov MMWORD [r10+16], put_buffer ; state->cur.put_buffer = put_buffer;
+ mov DWORD [r10+24], put_bits ; state->cur.put_bits = put_bits;
+
+ pop rbx
+ uncollect_args 6
+ pop_xmm 4
+ mov rsp, rbp ; rsp <- aligned rbp
+ pop rsp ; rsp <- original rbp
+ pop rbp
+ ret
+
+; For some reason, the OS X linker does not honor the request to align the
+; segment unless we do this.
+ align 32
diff --git a/simd/x86_64/jcsample-avx2.asm b/simd/x86_64/jcsample-avx2.asm
new file mode 100644
index 0000000..d7877ea
--- /dev/null
+++ b/simd/x86_64/jcsample-avx2.asm
@@ -0,0 +1,366 @@
+;
+; jcsample.asm - downsampling (64-bit AVX2)
+;
+; Copyright 2009 Pierre Ossman <ossman@cendio.se> for Cendio AB
+; Copyright (C) 2009, 2016, D. R. Commander.
+; Copyright (C) 2015, Intel Corporation.
+;
+; Based on the x86 SIMD extension for IJG JPEG library
+; Copyright (C) 1999-2006, MIYASAKA Masaru.
+; For conditions of distribution and use, see copyright notice in jsimdext.inc
+;
+; This file should be assembled with NASM (Netwide Assembler),
+; can *not* be assembled with Microsoft's MASM or any compatible
+; assembler (including Borland's Turbo Assembler).
+; NASM is available from http://nasm.sourceforge.net/ or
+; http://sourceforge.net/project/showfiles.php?group_id=6208
+;
+; [TAB8]
+
+%include "jsimdext.inc"
+
+; --------------------------------------------------------------------------
+ SECTION SEG_TEXT
+ BITS 64
+;
+; Downsample pixel values of a single component.
+; This version handles the common case of 2:1 horizontal and 1:1 vertical,
+; without smoothing.
+;
+; GLOBAL(void)
+; jsimd_h2v1_downsample_avx2 (JDIMENSION image_width, int max_v_samp_factor,
+; JDIMENSION v_samp_factor, JDIMENSION width_blocks,
+; JSAMPARRAY input_data, JSAMPARRAY output_data);
+;
+
+; r10d = JDIMENSION image_width
+; r11 = int max_v_samp_factor
+; r12d = JDIMENSION v_samp_factor
+; r13d = JDIMENSION width_blocks
+; r14 = JSAMPARRAY input_data
+; r15 = JSAMPARRAY output_data
+
+ align 32
+ global EXTN(jsimd_h2v1_downsample_avx2)
+
+EXTN(jsimd_h2v1_downsample_avx2):
+ push rbp
+ mov rax, rsp
+ mov rbp, rsp
+ collect_args 6
+
+ mov ecx, r13d
+ shl rcx, 3 ; imul rcx,DCTSIZE (rcx = output_cols)
+ jz near .return
+
+ mov edx, r10d
+
+ ; -- expand_right_edge
+
+ push rcx
+ shl rcx, 1 ; output_cols * 2
+ sub rcx, rdx
+ jle short .expand_end
+
+ mov rax, r11
+ test rax, rax
+ jle short .expand_end
+
+ cld
+ mov rsi, r14 ; input_data
+.expandloop:
+ push rax
+ push rcx
+
+ mov rdi, JSAMPROW [rsi]
+ add rdi, rdx
+ mov al, JSAMPLE [rdi-1]
+
+ rep stosb
+
+ pop rcx
+ pop rax
+
+ add rsi, byte SIZEOF_JSAMPROW
+ dec rax
+ jg short .expandloop
+
+.expand_end:
+ pop rcx ; output_cols
+
+ ; -- h2v1_downsample
+
+ mov eax, r12d ; rowctr
+ test eax, eax
+ jle near .return
+
+ mov rdx, 0x00010000 ; bias pattern
+ vmovd xmm7, edx
+ vpshufd xmm7, xmm7, 0x00 ; xmm7={0, 1, 0, 1, 0, 1, 0, 1}
+ vperm2i128 ymm7, ymm7, ymm7, 0 ; ymm7={xmm7, xmm7}
+ vpcmpeqw ymm6, ymm6, ymm6
+ vpsrlw ymm6, ymm6, BYTE_BIT ; ymm6={0xFF 0x00 0xFF 0x00 ..}
+
+ mov rsi, r14 ; input_data
+ mov rdi, r15 ; output_data
+.rowloop:
+ push rcx
+ push rdi
+ push rsi
+
+ mov rsi, JSAMPROW [rsi] ; inptr
+ mov rdi, JSAMPROW [rdi] ; outptr
+
+ cmp rcx, byte SIZEOF_YMMWORD
+ jae short .columnloop
+
+.columnloop_r24:
+ ; rcx can possibly be 8, 16, 24
+ cmp rcx, 24
+ jne .columnloop_r16
+ vmovdqu ymm0, YMMWORD [rsi+0*SIZEOF_YMMWORD]
+ vmovdqu xmm1, XMMWORD [rsi+1*SIZEOF_YMMWORD]
+ mov rcx, SIZEOF_YMMWORD
+ jmp short .downsample
+
+.columnloop_r16:
+ cmp rcx, 16
+ jne .columnloop_r8
+ vmovdqu ymm0, YMMWORD [rsi+0*SIZEOF_YMMWORD]
+ vpxor ymm1, ymm1, ymm1
+ mov rcx, SIZEOF_YMMWORD
+ jmp short .downsample
+
+.columnloop_r8:
+ vmovdqu xmm0, XMMWORD[rsi+0*SIZEOF_YMMWORD]
+ vpxor ymm1, ymm1, ymm1
+ mov rcx, SIZEOF_YMMWORD
+ jmp short .downsample
+
+.columnloop:
+ vmovdqu ymm0, YMMWORD [rsi+0*SIZEOF_YMMWORD]
+ vmovdqu ymm1, YMMWORD [rsi+1*SIZEOF_YMMWORD]
+
+.downsample:
+ vpsrlw ymm2, ymm0, BYTE_BIT
+ vpand ymm0, ymm0, ymm6
+ vpsrlw ymm3, ymm1, BYTE_BIT
+ vpand ymm1, ymm1, ymm6
+
+ vpaddw ymm0, ymm0, ymm2
+ vpaddw ymm1, ymm1, ymm3
+ vpaddw ymm0, ymm0, ymm7
+ vpaddw ymm1, ymm1, ymm7
+ vpsrlw ymm0, ymm0, 1
+ vpsrlw ymm1, ymm1, 1
+
+ vpackuswb ymm0, ymm0, ymm1
+ vpermq ymm0, ymm0, 0xd8
+
+ vmovdqu YMMWORD [rdi+0*SIZEOF_YMMWORD], ymm0
+
+ sub rcx, byte SIZEOF_YMMWORD ; outcol
+ add rsi, byte 2*SIZEOF_YMMWORD ; inptr
+ add rdi, byte 1*SIZEOF_YMMWORD ; outptr
+ cmp rcx, byte SIZEOF_YMMWORD
+ jae short .columnloop
+ test rcx, rcx
+ jnz near .columnloop_r24
+
+ pop rsi
+ pop rdi
+ pop rcx
+
+ add rsi, byte SIZEOF_JSAMPROW ; input_data
+ add rdi, byte SIZEOF_JSAMPROW ; output_data
+ dec rax ; rowctr
+ jg near .rowloop
+
+.return:
+ vzeroupper
+ uncollect_args 6
+ pop rbp
+ ret
+
+; --------------------------------------------------------------------------
+;
+; Downsample pixel values of a single component.
+; This version handles the standard case of 2:1 horizontal and 2:1 vertical,
+; without smoothing.
+;
+; GLOBAL(void)
+; jsimd_h2v2_downsample_avx2 (JDIMENSION image_width, int max_v_samp_factor,
+; JDIMENSION v_samp_factor, JDIMENSION width_blocks,
+; JSAMPARRAY input_data, JSAMPARRAY output_data);
+;
+
+; r10d = JDIMENSION image_width
+; r11 = int max_v_samp_factor
+; r12d = JDIMENSION v_samp_factor
+; r13d = JDIMENSION width_blocks
+; r14 = JSAMPARRAY input_data
+; r15 = JSAMPARRAY output_data
+
+ align 32
+ global EXTN(jsimd_h2v2_downsample_avx2)
+
+EXTN(jsimd_h2v2_downsample_avx2):
+ push rbp
+ mov rax, rsp
+ mov rbp, rsp
+ collect_args 6
+
+ mov ecx, r13d
+ shl rcx, 3 ; imul rcx,DCTSIZE (rcx = output_cols)
+ jz near .return
+
+ mov edx, r10d
+
+ ; -- expand_right_edge
+
+ push rcx
+ shl rcx, 1 ; output_cols * 2
+ sub rcx, rdx
+ jle short .expand_end
+
+ mov rax, r11
+ test rax, rax
+ jle short .expand_end
+
+ cld
+ mov rsi, r14 ; input_data
+.expandloop:
+ push rax
+ push rcx
+
+ mov rdi, JSAMPROW [rsi]
+ add rdi, rdx
+ mov al, JSAMPLE [rdi-1]
+
+ rep stosb
+
+ pop rcx
+ pop rax
+
+ add rsi, byte SIZEOF_JSAMPROW
+ dec rax
+ jg short .expandloop
+
+.expand_end:
+ pop rcx ; output_cols
+
+ ; -- h2v2_downsample
+
+ mov eax, r12d ; rowctr
+ test rax, rax
+ jle near .return
+
+ mov rdx, 0x00020001 ; bias pattern
+ vmovd xmm7, edx
+ vpcmpeqw ymm6, ymm6, ymm6
+ vpshufd xmm7, xmm7, 0x00 ; ymm7={1, 2, 1, 2, 1, 2, 1, 2}
+ vperm2i128 ymm7, ymm7, ymm7, 0
+ vpsrlw ymm6, ymm6, BYTE_BIT ; ymm6={0xFF 0x00 0xFF 0x00 ..}
+
+ mov rsi, r14 ; input_data
+ mov rdi, r15 ; output_data
+.rowloop:
+ push rcx
+ push rdi
+ push rsi
+
+ mov rdx, JSAMPROW [rsi+0*SIZEOF_JSAMPROW] ; inptr0
+ mov rsi, JSAMPROW [rsi+1*SIZEOF_JSAMPROW] ; inptr1
+ mov rdi, JSAMPROW [rdi] ; outptr
+
+ cmp rcx, byte SIZEOF_YMMWORD
+ jae short .columnloop
+
+.columnloop_r24:
+ cmp rcx, 24
+ jne .columnloop_r16
+ vmovdqu ymm0, YMMWORD [rdx+0*SIZEOF_YMMWORD]
+ vmovdqu ymm1, YMMWORD [rsi+0*SIZEOF_YMMWORD]
+ vmovdqu xmm2, XMMWORD [rdx+1*SIZEOF_YMMWORD]
+ vmovdqu xmm3, XMMWORD [rsi+1*SIZEOF_YMMWORD]
+ mov rcx, SIZEOF_YMMWORD
+ jmp short .downsample
+
+.columnloop_r16:
+ cmp rcx, 16
+ jne .columnloop_r8
+ vmovdqu ymm0, YMMWORD [rdx+0*SIZEOF_YMMWORD]
+ vmovdqu ymm1, YMMWORD [rsi+0*SIZEOF_YMMWORD]
+ vpxor ymm2, ymm2, ymm2
+ vpxor ymm3, ymm3, ymm3
+ mov rcx, SIZEOF_YMMWORD
+ jmp short .downsample
+
+.columnloop_r8:
+ vmovdqu xmm0, XMMWORD [rdx+0*SIZEOF_XMMWORD]
+ vmovdqu xmm1, XMMWORD [rsi+0*SIZEOF_XMMWORD]
+ vpxor ymm2, ymm2, ymm2
+ vpxor ymm3, ymm3, ymm3
+ mov rcx, SIZEOF_YMMWORD
+ jmp short .downsample
+
+.columnloop:
+ vmovdqu ymm0, YMMWORD [rdx+0*SIZEOF_YMMWORD]
+ vmovdqu ymm1, YMMWORD [rsi+0*SIZEOF_YMMWORD]
+ vmovdqu ymm2, YMMWORD [rdx+1*SIZEOF_YMMWORD]
+ vmovdqu ymm3, YMMWORD [rsi+1*SIZEOF_YMMWORD]
+
+.downsample:
+ vpand ymm4, ymm0, ymm6
+ vpsrlw ymm0, ymm0, BYTE_BIT
+ vpand ymm5, ymm1, ymm6
+ vpsrlw ymm1, ymm1, BYTE_BIT
+ vpaddw ymm0, ymm0, ymm4
+ vpaddw ymm1, ymm1, ymm5
+
+ vpand ymm4, ymm2, ymm6
+ vpsrlw ymm2, ymm2, BYTE_BIT
+ vpand ymm5, ymm3, ymm6
+ vpsrlw ymm3, ymm3, BYTE_BIT
+ vpaddw ymm2, ymm2, ymm4
+ vpaddw ymm3, ymm3, ymm5
+
+ vpaddw ymm0, ymm0, ymm1
+ vpaddw ymm2, ymm2, ymm3
+ vpaddw ymm0, ymm0, ymm7
+ vpaddw ymm2, ymm2, ymm7
+ vpsrlw ymm0, ymm0, 2
+ vpsrlw ymm2, ymm2, 2
+
+ vpackuswb ymm0, ymm0, ymm2
+ vpermq ymm0, ymm0, 0xd8
+
+ vmovdqu YMMWORD [rdi+0*SIZEOF_YMMWORD], ymm0
+
+ sub rcx, byte SIZEOF_YMMWORD ; outcol
+ add rdx, byte 2*SIZEOF_YMMWORD ; inptr0
+ add rsi, byte 2*SIZEOF_YMMWORD ; inptr1
+ add rdi, byte 1*SIZEOF_YMMWORD ; outptr
+ cmp rcx, byte SIZEOF_YMMWORD
+ jae near .columnloop
+ test rcx, rcx
+ jnz near .columnloop_r24
+
+ pop rsi
+ pop rdi
+ pop rcx
+
+ add rsi, byte 2*SIZEOF_JSAMPROW ; input_data
+ add rdi, byte 1*SIZEOF_JSAMPROW ; output_data
+ dec rax ; rowctr
+ jg near .rowloop
+
+.return:
+ vzeroupper
+ uncollect_args 6
+ pop rbp
+ ret
+
+; For some reason, the OS X linker does not honor the request to align the
+; segment unless we do this.
+ align 32
diff --git a/simd/x86_64/jcsample-sse2.asm b/simd/x86_64/jcsample-sse2.asm
new file mode 100644
index 0000000..2e5d3f5
--- /dev/null
+++ b/simd/x86_64/jcsample-sse2.asm
@@ -0,0 +1,329 @@
+;
+; jcsample.asm - downsampling (64-bit SSE2)
+;
+; Copyright 2009 Pierre Ossman <ossman@cendio.se> for Cendio AB
+; Copyright (C) 2009, 2016, D. R. Commander.
+;
+; Based on the x86 SIMD extension for IJG JPEG library
+; Copyright (C) 1999-2006, MIYASAKA Masaru.
+; For conditions of distribution and use, see copyright notice in jsimdext.inc
+;
+; This file should be assembled with NASM (Netwide Assembler),
+; can *not* be assembled with Microsoft's MASM or any compatible
+; assembler (including Borland's Turbo Assembler).
+; NASM is available from http://nasm.sourceforge.net/ or
+; http://sourceforge.net/project/showfiles.php?group_id=6208
+;
+; [TAB8]
+
+%include "jsimdext.inc"
+
+; --------------------------------------------------------------------------
+ SECTION SEG_TEXT
+ BITS 64
+;
+; Downsample pixel values of a single component.
+; This version handles the common case of 2:1 horizontal and 1:1 vertical,
+; without smoothing.
+;
+; GLOBAL(void)
+; jsimd_h2v1_downsample_sse2 (JDIMENSION image_width, int max_v_samp_factor,
+; JDIMENSION v_samp_factor, JDIMENSION width_blocks,
+; JSAMPARRAY input_data, JSAMPARRAY output_data);
+;
+
+; r10d = JDIMENSION image_width
+; r11 = int max_v_samp_factor
+; r12d = JDIMENSION v_samp_factor
+; r13d = JDIMENSION width_blocks
+; r14 = JSAMPARRAY input_data
+; r15 = JSAMPARRAY output_data
+
+ align 32
+ global EXTN(jsimd_h2v1_downsample_sse2)
+
+EXTN(jsimd_h2v1_downsample_sse2):
+ push rbp
+ mov rax, rsp
+ mov rbp, rsp
+ collect_args 6
+
+ mov ecx, r13d
+ shl rcx, 3 ; imul rcx,DCTSIZE (rcx = output_cols)
+ jz near .return
+
+ mov edx, r10d
+
+ ; -- expand_right_edge
+
+ push rcx
+ shl rcx, 1 ; output_cols * 2
+ sub rcx, rdx
+ jle short .expand_end
+
+ mov rax, r11
+ test rax, rax
+ jle short .expand_end
+
+ cld
+ mov rsi, r14 ; input_data
+.expandloop:
+ push rax
+ push rcx
+
+ mov rdi, JSAMPROW [rsi]
+ add rdi, rdx
+ mov al, JSAMPLE [rdi-1]
+
+ rep stosb
+
+ pop rcx
+ pop rax
+
+ add rsi, byte SIZEOF_JSAMPROW
+ dec rax
+ jg short .expandloop
+
+.expand_end:
+ pop rcx ; output_cols
+
+ ; -- h2v1_downsample
+
+ mov eax, r12d ; rowctr
+ test eax, eax
+ jle near .return
+
+ mov rdx, 0x00010000 ; bias pattern
+ movd xmm7, edx
+ pcmpeqw xmm6, xmm6
+ pshufd xmm7, xmm7, 0x00 ; xmm7={0, 1, 0, 1, 0, 1, 0, 1}
+ psrlw xmm6, BYTE_BIT ; xmm6={0xFF 0x00 0xFF 0x00 ..}
+
+ mov rsi, r14 ; input_data
+ mov rdi, r15 ; output_data
+.rowloop:
+ push rcx
+ push rdi
+ push rsi
+
+ mov rsi, JSAMPROW [rsi] ; inptr
+ mov rdi, JSAMPROW [rdi] ; outptr
+
+ cmp rcx, byte SIZEOF_XMMWORD
+ jae short .columnloop
+
+.columnloop_r8:
+ movdqa xmm0, XMMWORD [rsi+0*SIZEOF_XMMWORD]
+ pxor xmm1, xmm1
+ mov rcx, SIZEOF_XMMWORD
+ jmp short .downsample
+
+.columnloop:
+ movdqa xmm0, XMMWORD [rsi+0*SIZEOF_XMMWORD]
+ movdqa xmm1, XMMWORD [rsi+1*SIZEOF_XMMWORD]
+
+.downsample:
+ movdqa xmm2, xmm0
+ movdqa xmm3, xmm1
+
+ pand xmm0, xmm6
+ psrlw xmm2, BYTE_BIT
+ pand xmm1, xmm6
+ psrlw xmm3, BYTE_BIT
+
+ paddw xmm0, xmm2
+ paddw xmm1, xmm3
+ paddw xmm0, xmm7
+ paddw xmm1, xmm7
+ psrlw xmm0, 1
+ psrlw xmm1, 1
+
+ packuswb xmm0, xmm1
+
+ movdqa XMMWORD [rdi+0*SIZEOF_XMMWORD], xmm0
+
+ sub rcx, byte SIZEOF_XMMWORD ; outcol
+ add rsi, byte 2*SIZEOF_XMMWORD ; inptr
+ add rdi, byte 1*SIZEOF_XMMWORD ; outptr
+ cmp rcx, byte SIZEOF_XMMWORD
+ jae short .columnloop
+ test rcx, rcx
+ jnz short .columnloop_r8
+
+ pop rsi
+ pop rdi
+ pop rcx
+
+ add rsi, byte SIZEOF_JSAMPROW ; input_data
+ add rdi, byte SIZEOF_JSAMPROW ; output_data
+ dec rax ; rowctr
+ jg near .rowloop
+
+.return:
+ uncollect_args 6
+ pop rbp
+ ret
+
+; --------------------------------------------------------------------------
+;
+; Downsample pixel values of a single component.
+; This version handles the standard case of 2:1 horizontal and 2:1 vertical,
+; without smoothing.
+;
+; GLOBAL(void)
+; jsimd_h2v2_downsample_sse2 (JDIMENSION image_width, int max_v_samp_factor,
+; JDIMENSION v_samp_factor, JDIMENSION width_blocks,
+; JSAMPARRAY input_data, JSAMPARRAY output_data);
+;
+
+; r10d = JDIMENSION image_width
+; r11 = int max_v_samp_factor
+; r12d = JDIMENSION v_samp_factor
+; r13d = JDIMENSION width_blocks
+; r14 = JSAMPARRAY input_data
+; r15 = JSAMPARRAY output_data
+
+ align 32
+ global EXTN(jsimd_h2v2_downsample_sse2)
+
+EXTN(jsimd_h2v2_downsample_sse2):
+ push rbp
+ mov rax, rsp
+ mov rbp, rsp
+ collect_args 6
+
+ mov ecx, r13d
+ shl rcx, 3 ; imul rcx,DCTSIZE (rcx = output_cols)
+ jz near .return
+
+ mov edx, r10d
+
+ ; -- expand_right_edge
+
+ push rcx
+ shl rcx, 1 ; output_cols * 2
+ sub rcx, rdx
+ jle short .expand_end
+
+ mov rax, r11
+ test rax, rax
+ jle short .expand_end
+
+ cld
+ mov rsi, r14 ; input_data
+.expandloop:
+ push rax
+ push rcx
+
+ mov rdi, JSAMPROW [rsi]
+ add rdi, rdx
+ mov al, JSAMPLE [rdi-1]
+
+ rep stosb
+
+ pop rcx
+ pop rax
+
+ add rsi, byte SIZEOF_JSAMPROW
+ dec rax
+ jg short .expandloop
+
+.expand_end:
+ pop rcx ; output_cols
+
+ ; -- h2v2_downsample
+
+ mov eax, r12d ; rowctr
+ test rax, rax
+ jle near .return
+
+ mov rdx, 0x00020001 ; bias pattern
+ movd xmm7, edx
+ pcmpeqw xmm6, xmm6
+ pshufd xmm7, xmm7, 0x00 ; xmm7={1, 2, 1, 2, 1, 2, 1, 2}
+ psrlw xmm6, BYTE_BIT ; xmm6={0xFF 0x00 0xFF 0x00 ..}
+
+ mov rsi, r14 ; input_data
+ mov rdi, r15 ; output_data
+.rowloop:
+ push rcx
+ push rdi
+ push rsi
+
+ mov rdx, JSAMPROW [rsi+0*SIZEOF_JSAMPROW] ; inptr0
+ mov rsi, JSAMPROW [rsi+1*SIZEOF_JSAMPROW] ; inptr1
+ mov rdi, JSAMPROW [rdi] ; outptr
+
+ cmp rcx, byte SIZEOF_XMMWORD
+ jae short .columnloop
+
+.columnloop_r8:
+ movdqa xmm0, XMMWORD [rdx+0*SIZEOF_XMMWORD]
+ movdqa xmm1, XMMWORD [rsi+0*SIZEOF_XMMWORD]
+ pxor xmm2, xmm2
+ pxor xmm3, xmm3
+ mov rcx, SIZEOF_XMMWORD
+ jmp short .downsample
+
+.columnloop:
+ movdqa xmm0, XMMWORD [rdx+0*SIZEOF_XMMWORD]
+ movdqa xmm1, XMMWORD [rsi+0*SIZEOF_XMMWORD]
+ movdqa xmm2, XMMWORD [rdx+1*SIZEOF_XMMWORD]
+ movdqa xmm3, XMMWORD [rsi+1*SIZEOF_XMMWORD]
+
+.downsample:
+ movdqa xmm4, xmm0
+ movdqa xmm5, xmm1
+ pand xmm0, xmm6
+ psrlw xmm4, BYTE_BIT
+ pand xmm1, xmm6
+ psrlw xmm5, BYTE_BIT
+ paddw xmm0, xmm4
+ paddw xmm1, xmm5
+
+ movdqa xmm4, xmm2
+ movdqa xmm5, xmm3
+ pand xmm2, xmm6
+ psrlw xmm4, BYTE_BIT
+ pand xmm3, xmm6
+ psrlw xmm5, BYTE_BIT
+ paddw xmm2, xmm4
+ paddw xmm3, xmm5
+
+ paddw xmm0, xmm1
+ paddw xmm2, xmm3
+ paddw xmm0, xmm7
+ paddw xmm2, xmm7
+ psrlw xmm0, 2
+ psrlw xmm2, 2
+
+ packuswb xmm0, xmm2
+
+ movdqa XMMWORD [rdi+0*SIZEOF_XMMWORD], xmm0
+
+ sub rcx, byte SIZEOF_XMMWORD ; outcol
+ add rdx, byte 2*SIZEOF_XMMWORD ; inptr0
+ add rsi, byte 2*SIZEOF_XMMWORD ; inptr1
+ add rdi, byte 1*SIZEOF_XMMWORD ; outptr
+ cmp rcx, byte SIZEOF_XMMWORD
+ jae near .columnloop
+ test rcx, rcx
+ jnz near .columnloop_r8
+
+ pop rsi
+ pop rdi
+ pop rcx
+
+ add rsi, byte 2*SIZEOF_JSAMPROW ; input_data
+ add rdi, byte 1*SIZEOF_JSAMPROW ; output_data
+ dec rax ; rowctr
+ jg near .rowloop
+
+.return:
+ uncollect_args 6
+ pop rbp
+ ret
+
+; For some reason, the OS X linker does not honor the request to align the
+; segment unless we do this.
+ align 32
diff --git a/simd/x86_64/jdcolext-avx2.asm b/simd/x86_64/jdcolext-avx2.asm
new file mode 100644
index 0000000..ca81ad8
--- /dev/null
+++ b/simd/x86_64/jdcolext-avx2.asm
@@ -0,0 +1,497 @@
+;
+; jdcolext.asm - colorspace conversion (64-bit AVX2)
+;
+; Copyright 2009, 2012 Pierre Ossman <ossman@cendio.se> for Cendio AB
+; Copyright (C) 2009, 2012, 2016, D. R. Commander.
+; Copyright (C) 2015, Intel Corporation.
+;
+; Based on the x86 SIMD extension for IJG JPEG library
+; Copyright (C) 1999-2006, MIYASAKA Masaru.
+; For conditions of distribution and use, see copyright notice in jsimdext.inc
+;
+; This file should be assembled with NASM (Netwide Assembler),
+; can *not* be assembled with Microsoft's MASM or any compatible
+; assembler (including Borland's Turbo Assembler).
+; NASM is available from http://nasm.sourceforge.net/ or
+; http://sourceforge.net/project/showfiles.php?group_id=6208
+;
+; [TAB8]
+
+%include "jcolsamp.inc"
+
+; --------------------------------------------------------------------------
+;
+; Convert some rows of samples to the output colorspace.
+;
+; GLOBAL(void)
+; jsimd_ycc_rgb_convert_avx2 (JDIMENSION out_width,
+; JSAMPIMAGE input_buf, JDIMENSION input_row,
+; JSAMPARRAY output_buf, int num_rows)
+;
+
+; r10d = JDIMENSION out_width
+; r11 = JSAMPIMAGE input_buf
+; r12d = JDIMENSION input_row
+; r13 = JSAMPARRAY output_buf
+; r14d = int num_rows
+
+%define wk(i) rbp-(WK_NUM-(i))*SIZEOF_YMMWORD ; ymmword wk[WK_NUM]
+%define WK_NUM 2
+
+ align 32
+ global EXTN(jsimd_ycc_rgb_convert_avx2)
+
+EXTN(jsimd_ycc_rgb_convert_avx2):
+ push rbp
+ mov rax, rsp ; rax = original rbp
+ sub rsp, byte 4
+ and rsp, byte (-SIZEOF_YMMWORD) ; align to 256 bits
+ mov [rsp], rax
+ mov rbp, rsp ; rbp = aligned rbp
+ lea rsp, [wk(0)]
+ collect_args 5
+ push rbx
+
+ mov ecx, r10d ; num_cols
+ test rcx, rcx
+ jz near .return
+
+ push rcx
+
+ mov rdi, r11
+ mov ecx, r12d
+ mov rsi, JSAMPARRAY [rdi+0*SIZEOF_JSAMPARRAY]
+ mov rbx, JSAMPARRAY [rdi+1*SIZEOF_JSAMPARRAY]
+ mov rdx, JSAMPARRAY [rdi+2*SIZEOF_JSAMPARRAY]
+ lea rsi, [rsi+rcx*SIZEOF_JSAMPROW]
+ lea rbx, [rbx+rcx*SIZEOF_JSAMPROW]
+ lea rdx, [rdx+rcx*SIZEOF_JSAMPROW]
+
+ pop rcx
+
+ mov rdi, r13
+ mov eax, r14d
+ test rax, rax
+ jle near .return
+.rowloop:
+ push rax
+ push rdi
+ push rdx
+ push rbx
+ push rsi
+ push rcx ; col
+
+ mov rsi, JSAMPROW [rsi] ; inptr0
+ mov rbx, JSAMPROW [rbx] ; inptr1
+ mov rdx, JSAMPROW [rdx] ; inptr2
+ mov rdi, JSAMPROW [rdi] ; outptr
+.columnloop:
+
+ vmovdqu ymm5, YMMWORD [rbx] ; ymm5=Cb(0123456789ABCDEFGHIJKLMNOPQRSTUV)
+ vmovdqu ymm1, YMMWORD [rdx] ; ymm1=Cr(0123456789ABCDEFGHIJKLMNOPQRSTUV)
+
+ vpcmpeqw ymm0, ymm0, ymm0
+ vpcmpeqw ymm7, ymm7, ymm7
+ vpsrlw ymm0, ymm0, BYTE_BIT ; ymm0={0xFF 0x00 0xFF 0x00 ..}
+ vpsllw ymm7, ymm7, 7 ; ymm7={0xFF80 0xFF80 0xFF80 0xFF80 ..}
+
+ vpand ymm4, ymm0, ymm5 ; ymm4=Cb(02468ACEGIKMOQSU)=CbE
+ vpsrlw ymm5, ymm5, BYTE_BIT ; ymm5=Cb(13579BDFHJLNPRTV)=CbO
+ vpand ymm0, ymm0, ymm1 ; ymm0=Cr(02468ACEGIKMOQSU)=CrE
+ vpsrlw ymm1, ymm1, BYTE_BIT ; ymm1=Cr(13579BDFHJLNPRTV)=CrO
+
+ vpaddw ymm2, ymm4, ymm7
+ vpaddw ymm3, ymm5, ymm7
+ vpaddw ymm6, ymm0, ymm7
+ vpaddw ymm7, ymm1, ymm7
+
+ ; (Original)
+ ; R = Y + 1.40200 * Cr
+ ; G = Y - 0.34414 * Cb - 0.71414 * Cr
+ ; B = Y + 1.77200 * Cb
+ ;
+ ; (This implementation)
+ ; R = Y + 0.40200 * Cr + Cr
+ ; G = Y - 0.34414 * Cb + 0.28586 * Cr - Cr
+ ; B = Y - 0.22800 * Cb + Cb + Cb
+
+ vpaddw ymm4, ymm2, ymm2 ; ymm4=2*CbE
+ vpaddw ymm5, ymm3, ymm3 ; ymm5=2*CbO
+ vpaddw ymm0, ymm6, ymm6 ; ymm0=2*CrE
+ vpaddw ymm1, ymm7, ymm7 ; ymm1=2*CrO
+
+ vpmulhw ymm4, ymm4, [rel PW_MF0228] ; ymm4=(2*CbE * -FIX(0.22800))
+ vpmulhw ymm5, ymm5, [rel PW_MF0228] ; ymm5=(2*CbO * -FIX(0.22800))
+ vpmulhw ymm0, ymm0, [rel PW_F0402] ; ymm0=(2*CrE * FIX(0.40200))
+ vpmulhw ymm1, ymm1, [rel PW_F0402] ; ymm1=(2*CrO * FIX(0.40200))
+
+ vpaddw ymm4, ymm4, [rel PW_ONE]
+ vpaddw ymm5, ymm5, [rel PW_ONE]
+ vpsraw ymm4, ymm4, 1 ; ymm4=(CbE * -FIX(0.22800))
+ vpsraw ymm5, ymm5, 1 ; ymm5=(CbO * -FIX(0.22800))
+ vpaddw ymm0, ymm0, [rel PW_ONE]
+ vpaddw ymm1, ymm1, [rel PW_ONE]
+ vpsraw ymm0, ymm0, 1 ; ymm0=(CrE * FIX(0.40200))
+ vpsraw ymm1, ymm1, 1 ; ymm1=(CrO * FIX(0.40200))
+
+ vpaddw ymm4, ymm4, ymm2
+ vpaddw ymm5, ymm5, ymm3
+ vpaddw ymm4, ymm4, ymm2 ; ymm4=(CbE * FIX(1.77200))=(B-Y)E
+ vpaddw ymm5, ymm5, ymm3 ; ymm5=(CbO * FIX(1.77200))=(B-Y)O
+ vpaddw ymm0, ymm0, ymm6 ; ymm0=(CrE * FIX(1.40200))=(R-Y)E
+ vpaddw ymm1, ymm1, ymm7 ; ymm1=(CrO * FIX(1.40200))=(R-Y)O
+
+ vmovdqa YMMWORD [wk(0)], ymm4 ; wk(0)=(B-Y)E
+ vmovdqa YMMWORD [wk(1)], ymm5 ; wk(1)=(B-Y)O
+
+ vpunpckhwd ymm4, ymm2, ymm6
+ vpunpcklwd ymm2, ymm2, ymm6
+ vpmaddwd ymm2, ymm2, [rel PW_MF0344_F0285]
+ vpmaddwd ymm4, ymm4, [rel PW_MF0344_F0285]
+ vpunpckhwd ymm5, ymm3, ymm7
+ vpunpcklwd ymm3, ymm3, ymm7
+ vpmaddwd ymm3, ymm3, [rel PW_MF0344_F0285]
+ vpmaddwd ymm5, ymm5, [rel PW_MF0344_F0285]
+
+ vpaddd ymm2, ymm2, [rel PD_ONEHALF]
+ vpaddd ymm4, ymm4, [rel PD_ONEHALF]
+ vpsrad ymm2, ymm2, SCALEBITS
+ vpsrad ymm4, ymm4, SCALEBITS
+ vpaddd ymm3, ymm3, [rel PD_ONEHALF]
+ vpaddd ymm5, ymm5, [rel PD_ONEHALF]
+ vpsrad ymm3, ymm3, SCALEBITS
+ vpsrad ymm5, ymm5, SCALEBITS
+
+ vpackssdw ymm2, ymm2, ymm4 ; ymm2=CbE*-FIX(0.344)+CrE*FIX(0.285)
+ vpackssdw ymm3, ymm3, ymm5 ; ymm3=CbO*-FIX(0.344)+CrO*FIX(0.285)
+ vpsubw ymm2, ymm2, ymm6 ; ymm2=CbE*-FIX(0.344)+CrE*-FIX(0.714)=(G-Y)E
+ vpsubw ymm3, ymm3, ymm7 ; ymm3=CbO*-FIX(0.344)+CrO*-FIX(0.714)=(G-Y)O
+
+ vmovdqu ymm5, YMMWORD [rsi] ; ymm5=Y(0123456789ABCDEFGHIJKLMNOPQRSTUV)
+
+ vpcmpeqw ymm4, ymm4, ymm4
+ vpsrlw ymm4, ymm4, BYTE_BIT ; ymm4={0xFF 0x00 0xFF 0x00 ..}
+ vpand ymm4, ymm4, ymm5 ; ymm4=Y(02468ACEGIKMOQSU)=YE
+ vpsrlw ymm5, ymm5, BYTE_BIT ; ymm5=Y(13579BDFHJLNPRTV)=YO
+
+ vpaddw ymm0, ymm0, ymm4 ; ymm0=((R-Y)E+YE)=RE=R(02468ACEGIKMOQSU)
+ vpaddw ymm1, ymm1, ymm5 ; ymm1=((R-Y)O+YO)=RO=R(13579BDFHJLNPRTV)
+ vpackuswb ymm0, ymm0, ymm0 ; ymm0=R(02468ACE********GIKMOQSU********)
+ vpackuswb ymm1, ymm1, ymm1 ; ymm1=R(13579BDF********HJLNPRTV********)
+
+ vpaddw ymm2, ymm2, ymm4 ; ymm2=((G-Y)E+YE)=GE=G(02468ACEGIKMOQSU)
+ vpaddw ymm3, ymm3, ymm5 ; ymm3=((G-Y)O+YO)=GO=G(13579BDFHJLNPRTV)
+ vpackuswb ymm2, ymm2, ymm2 ; ymm2=G(02468ACE********GIKMOQSU********)
+ vpackuswb ymm3, ymm3, ymm3 ; ymm3=G(13579BDF********HJLNPRTV********)
+
+ vpaddw ymm4, ymm4, YMMWORD [wk(0)] ; ymm4=(YE+(B-Y)E)=BE=B(02468ACEGIKMOQSU)
+ vpaddw ymm5, ymm5, YMMWORD [wk(1)] ; ymm5=(YO+(B-Y)O)=BO=B(13579BDFHJLNPRTV)
+ vpackuswb ymm4, ymm4, ymm4 ; ymm4=B(02468ACE********GIKMOQSU********)
+ vpackuswb ymm5, ymm5, ymm5 ; ymm5=B(13579BDF********HJLNPRTV********)
+
+%if RGB_PIXELSIZE == 3 ; ---------------
+
+ ; ymmA=(00 02 04 06 08 0A 0C 0E ** 0G 0I 0K 0M 0O 0Q 0S 0U **)
+ ; ymmB=(01 03 05 07 09 0B 0D 0F ** 0H 0J 0L 0N 0P 0R 0T 0V **)
+ ; ymmC=(10 12 14 16 18 1A 1C 1E ** 1G 1I 1K 1M 1O 1Q 1S 1U **)
+ ; ymmD=(11 13 15 17 19 1B 1D 1F ** 1H 1J 1L 1N 1P 1R 1T 1V **)
+ ; ymmE=(20 22 24 26 28 2A 2C 2E ** 2G 2I 2K 2M 2O 2Q 2S 2U **)
+ ; ymmF=(21 23 25 27 29 2B 2D 2F ** 2H 2J 2L 2N 2P 2R 2T 2V **)
+ ; ymmG=(** ** ** ** ** ** ** ** ** ** ** ** ** ** ** ** ** **)
+ ; ymmH=(** ** ** ** ** ** ** ** ** ** ** ** ** ** ** ** ** **)
+
+ vpunpcklbw ymmA, ymmA, ymmC ; ymmA=(00 10 02 12 04 14 06 16 08 18 0A 1A 0C 1C 0E 1E
+ ; 0G 1G 0I 1I 0K 1K 0M 1M 0O 1O 0Q 1Q 0S 1S 0U 1U)
+ vpunpcklbw ymmE, ymmE, ymmB ; ymmE=(20 01 22 03 24 05 26 07 28 09 2A 0B 2C 0D 2E 0F
+ ; 2G 0H 2I 0J 2K 0L 2M 0N 2O 0P 2Q 0R 2S 0T 2U 0V)
+ vpunpcklbw ymmD, ymmD, ymmF ; ymmD=(11 21 13 23 15 25 17 27 19 29 1B 2B 1D 2D 1F 2F
+ ; 1H 2H 1J 2J 1L 2L 1N 2N 1P 2P 1R 2R 1T 2T 1V 2V)
+
+ vpsrldq ymmH, ymmA, 2 ; ymmH=(02 12 04 14 06 16 08 18 0A 1A 0C 1C 0E 1E 0G 1G
+ ; 0I 1I 0K 1K 0M 1M 0O 1O 0Q 1Q 0S 1S 0U 1U -- --)
+ vpunpckhwd ymmG, ymmA, ymmE ; ymmG=(08 18 28 09 0A 1A 2A 0B 0C 1C 2C 0D 0E 1E 2E 0F
+ ; 0O 1O 2O 0P 0Q 1Q 2Q 0R 0S 1S 2S 0T 0U 1U 2U 0V)
+ vpunpcklwd ymmA, ymmA, ymmE ; ymmA=(00 10 20 01 02 12 22 03 04 14 24 05 06 16 26 07
+ ; 0G 1G 2G 0H 0I 1I 2I 0J 0K 1K 2K 0L 0M 1M 2M 0N)
+
+ vpsrldq ymmE, ymmE, 2 ; ymmE=(22 03 24 05 26 07 28 09 2A 0B 2C 0D 2E 0F 2G 0H
+ ; 2I 0J 2K 0L 2M 0N 2O 0P 2Q 0R 2S 0T 2U 0V -- --)
+
+ vpsrldq ymmB, ymmD, 2 ; ymmB=(13 23 15 25 17 27 19 29 1B 2B 1D 2D 1F 2F 1H 2H
+ ; 1J 2J 1L 2L 1N 2N 1P 2P 1R 2R 1T 2T 1V 2V -- --)
+ vpunpckhwd ymmC, ymmD, ymmH ; ymmC=(19 29 0A 1A 1B 2B 0C 1C 1D 2D 0E 1E 1F 2F 0G 1G
+ ; 1P 2P 0Q 1Q 1R 2R 0S 1S 1T 2T 0U 1U 1V 2V -- --)
+ vpunpcklwd ymmD, ymmD, ymmH ; ymmD=(11 21 02 12 13 23 04 14 15 25 06 16 17 27 08 18
+ ; 1H 2H 0I 1I 1J 2J 0K 1K 1L 2L 0M 1M 1N 2N 0O 1O)
+
+ vpunpckhwd ymmF, ymmE, ymmB ; ymmF=(2A 0B 1B 2B 2C 0D 1D 2D 2E 0F 1F 2F 2G 0H 1H 2H
+ ; 2Q 0R 1R 2R 2S 0T 1T 2T 2U 0V 1V 2V -- -- -- --)
+ vpunpcklwd ymmE, ymmE, ymmB ; ymmE=(22 03 13 23 24 05 15 25 26 07 17 27 28 09 19 29
+ ; 2I 0J 1J 2J 2K 0L 1L 2L 2M 0N 1N 2N 2O 0P 1P 2P)
+
+ vpshufd ymmH, ymmA, 0x4E ; ymmH=(04 14 24 05 06 16 26 07 00 10 20 01 02 12 22 03
+ ; 0K 1K 2K 0L 0M 1M 2M 0N 0G 1G 2G 0H 0I 1I 2I 0J)
+ vpunpckldq ymmA, ymmA, ymmD ; ymmA=(00 10 20 01 11 21 02 12 02 12 22 03 13 23 04 14
+ ; 0G 1G 2G 0H 1H 2H 0I 1I 0I 1I 2I 0J 1J 2J 0K 1K)
+ vpunpckhdq ymmD, ymmD, ymmE ; ymmD=(15 25 06 16 26 07 17 27 17 27 08 18 28 09 19 29
+ ; 1L 2L 0M 1M 2M 0N 1N 2N 1N 2N 0O 1O 2O 0P 1P 2P)
+ vpunpckldq ymmE, ymmE, ymmH ; ymmE=(22 03 13 23 04 14 24 05 24 05 15 25 06 16 26 07
+ ; 2I 0J 1J 2J 0K 1K 2K 0L 2K 0L 1L 2L 0M 1M 2M 0N)
+
+ vpshufd ymmH, ymmG, 0x4E ; ymmH=(0C 1C 2C 0D 0E 1E 2E 0F 08 18 28 09 0A 1A 2A 0B
+ ; 0S 1S 2S 0T 0U 1U 2U 0V 0O 1O 2O 0P 0Q 1Q 2Q 0R)
+ vpunpckldq ymmG, ymmG, ymmC ; ymmG=(08 18 28 09 19 29 0A 1A 0A 1A 2A 0B 1B 2B 0C 1C
+ ; 0O 1O 2O 0P 1P 2P 0Q 1Q 0Q 1Q 2Q 0R 1R 2R 0S 1S)
+ vpunpckhdq ymmC, ymmC, ymmF ; ymmC=(1D 2D 0E 1E 2E 0F 1F 2F 1F 2F 0G 1G 2G 0H 1H 2H
+ ; 1T 2T 0U 1U 2U 0V 1V 2V 1V 2V -- -- -- -- -- --)
+ vpunpckldq ymmF, ymmF, ymmH ; ymmF=(2A 0B 1B 2B 0C 1C 2C 0D 2C 0D 1D 2D 0E 1E 2E 0F
+ ; 2Q 0R 1R 2R 0S 1S 2S 0T 2S 0T 1T 2T 0U 1U 2U 0V)
+
+ vpunpcklqdq ymmH, ymmA, ymmE ; ymmH=(00 10 20 01 11 21 02 12 22 03 13 23 04 14 24 05
+ ; 0G 1G 2G 0H 1H 2H 0I 1I 2I 0J 1J 2J 0K 1K 2K 0L)
+ vpunpcklqdq ymmG, ymmD, ymmG ; ymmG=(15 25 06 16 26 07 17 27 08 18 28 09 19 29 0A 1A
+ ; 1L 2L 0M 1M 2M 0N 1N 2N 0O 1O 2O 0P 1P 2P 0Q 1Q)
+ vpunpcklqdq ymmC, ymmF, ymmC ; ymmC=(2A 0B 1B 2B 0C 1C 2C 0D 1D 2D 0E 1E 2E 0F 1F 2F
+ ; 2Q 0R 1R 2R 0S 1S 2S 0T 1T 2T 0U 1U 2U 0V 1V 2V)
+
+ vperm2i128 ymmA, ymmH, ymmG, 0x20 ; ymmA=(00 10 20 01 11 21 02 12 22 03 13 23 04 14 24 05
+ ; 15 25 06 16 26 07 17 27 08 18 28 09 19 29 0A 1A)
+ vperm2i128 ymmD, ymmC, ymmH, 0x30 ; ymmD=(2A 0B 1B 2B 0C 1C 2C 0D 1D 2D 0E 1E 2E 0F 1F 2F
+ ; 0G 1G 2G 0H 1H 2H 0I 1I 2I 0J 1J 2J 0K 1K 2K 0L)
+ vperm2i128 ymmF, ymmG, ymmC, 0x31 ; ymmF=(1L 2L 0M 1M 2M 0N 1N 2N 0O 1O 2O 0P 1P 2P 0Q 1Q
+ ; 2Q 0R 1R 2R 0S 1S 2S 0T 1T 2T 0U 1U 2U 0V 1V 2V)
+
+ cmp rcx, byte SIZEOF_YMMWORD
+ jb short .column_st64
+
+ test rdi, SIZEOF_YMMWORD-1
+ jnz short .out1
+ ; --(aligned)-------------------
+ vmovntdq YMMWORD [rdi+0*SIZEOF_YMMWORD], ymmA
+ vmovntdq YMMWORD [rdi+1*SIZEOF_YMMWORD], ymmD
+ vmovntdq YMMWORD [rdi+2*SIZEOF_YMMWORD], ymmF
+ jmp short .out0
+.out1: ; --(unaligned)-----------------
+ vmovdqu YMMWORD [rdi+0*SIZEOF_YMMWORD], ymmA
+ vmovdqu YMMWORD [rdi+1*SIZEOF_YMMWORD], ymmD
+ vmovdqu YMMWORD [rdi+2*SIZEOF_YMMWORD], ymmF
+.out0:
+ add rdi, byte RGB_PIXELSIZE*SIZEOF_YMMWORD ; outptr
+ sub rcx, byte SIZEOF_YMMWORD
+ jz near .nextrow
+
+ add rsi, byte SIZEOF_YMMWORD ; inptr0
+ add rbx, byte SIZEOF_YMMWORD ; inptr1
+ add rdx, byte SIZEOF_YMMWORD ; inptr2
+ jmp near .columnloop
+
+.column_st64:
+ lea rcx, [rcx+rcx*2] ; imul ecx, RGB_PIXELSIZE
+ cmp rcx, byte 2*SIZEOF_YMMWORD
+ jb short .column_st32
+ vmovdqu YMMWORD [rdi+0*SIZEOF_YMMWORD], ymmA
+ vmovdqu YMMWORD [rdi+1*SIZEOF_YMMWORD], ymmD
+ add rdi, byte 2*SIZEOF_YMMWORD ; outptr
+ vmovdqa ymmA, ymmF
+ sub rcx, byte 2*SIZEOF_YMMWORD
+ jmp short .column_st31
+.column_st32:
+ cmp rcx, byte SIZEOF_YMMWORD
+ jb short .column_st31
+ vmovdqu YMMWORD [rdi+0*SIZEOF_YMMWORD], ymmA
+ add rdi, byte SIZEOF_YMMWORD ; outptr
+ vmovdqa ymmA, ymmD
+ sub rcx, byte SIZEOF_YMMWORD
+ jmp short .column_st31
+.column_st31:
+ cmp rcx, byte SIZEOF_XMMWORD
+ jb short .column_st15
+ vmovdqu XMMWORD [rdi+0*SIZEOF_XMMWORD], xmmA
+ add rdi, byte SIZEOF_XMMWORD ; outptr
+ vperm2i128 ymmA, ymmA, ymmA, 1
+ sub rcx, byte SIZEOF_XMMWORD
+.column_st15:
+ ; Store the lower 8 bytes of xmmA to the output when it has enough
+ ; space.
+ cmp rcx, byte SIZEOF_MMWORD
+ jb short .column_st7
+ vmovq XMM_MMWORD [rdi], xmmA
+ add rdi, byte SIZEOF_MMWORD
+ sub rcx, byte SIZEOF_MMWORD
+ vpsrldq xmmA, xmmA, SIZEOF_MMWORD
+.column_st7:
+ ; Store the lower 4 bytes of xmmA to the output when it has enough
+ ; space.
+ cmp rcx, byte SIZEOF_DWORD
+ jb short .column_st3
+ vmovd XMM_DWORD [rdi], xmmA
+ add rdi, byte SIZEOF_DWORD
+ sub rcx, byte SIZEOF_DWORD
+ vpsrldq xmmA, xmmA, SIZEOF_DWORD
+.column_st3:
+ ; Store the lower 2 bytes of rax to the output when it has enough
+ ; space.
+ vmovd eax, xmmA
+ cmp rcx, byte SIZEOF_WORD
+ jb short .column_st1
+ mov WORD [rdi], ax
+ add rdi, byte SIZEOF_WORD
+ sub rcx, byte SIZEOF_WORD
+ shr rax, 16
+.column_st1:
+ ; Store the lower 1 byte of rax to the output when it has enough
+ ; space.
+ test rcx, rcx
+ jz short .nextrow
+ mov BYTE [rdi], al
+
+%else ; RGB_PIXELSIZE == 4 ; -----------
+
+%ifdef RGBX_FILLER_0XFF
+ vpcmpeqb ymm6, ymm6, ymm6 ; ymm6=XE=X(02468ACE********GIKMOQSU********)
+ vpcmpeqb ymm7, ymm7, ymm7 ; ymm7=XO=X(13579BDF********HJLNPRTV********)
+%else
+ vpxor ymm6, ymm6, ymm6 ; ymm6=XE=X(02468ACE********GIKMOQSU********)
+ vpxor ymm7, ymm7, ymm7 ; ymm7=XO=X(13579BDF********HJLNPRTV********)
+%endif
+ ; ymmA=(00 02 04 06 08 0A 0C 0E ** 0G 0I 0K 0M 0O 0Q 0S 0U **)
+ ; ymmB=(01 03 05 07 09 0B 0D 0F ** 0H 0J 0L 0N 0P 0R 0T 0V **)
+ ; ymmC=(10 12 14 16 18 1A 1C 1E ** 1G 1I 1K 1M 1O 1Q 1S 1U **)
+ ; ymmD=(11 13 15 17 19 1B 1D 1F ** 1H 1J 1L 1N 1P 1R 1T 1V **)
+ ; ymmE=(20 22 24 26 28 2A 2C 2E ** 2G 2I 2K 2M 2O 2Q 2S 2U **)
+ ; ymmF=(21 23 25 27 29 2B 2D 2F ** 2H 2J 2L 2N 2P 2R 2T 2V **)
+ ; ymmG=(30 32 34 36 38 3A 3C 3E ** 3G 3I 3K 3M 3O 3Q 3S 3U **)
+ ; ymmH=(31 33 35 37 39 3B 3D 3F ** 3H 3J 3L 3N 3P 3R 3T 3V **)
+
+ vpunpcklbw ymmA, ymmA, ymmC ; ymmA=(00 10 02 12 04 14 06 16 08 18 0A 1A 0C 1C 0E 1E
+ ; 0G 1G 0I 1I 0K 1K 0M 1M 0O 1O 0Q 1Q 0S 1S 0U 1U)
+ vpunpcklbw ymmE, ymmE, ymmG ; ymmE=(20 30 22 32 24 34 26 36 28 38 2A 3A 2C 3C 2E 3E
+ ; 2G 3G 2I 3I 2K 3K 2M 3M 2O 3O 2Q 3Q 2S 3S 2U 3U)
+ vpunpcklbw ymmB, ymmB, ymmD ; ymmB=(01 11 03 13 05 15 07 17 09 19 0B 1B 0D 1D 0F 1F
+ ; 0H 1H 0J 1J 0L 1L 0N 1N 0P 1P 0R 1R 0T 1T 0V 1V)
+ vpunpcklbw ymmF, ymmF, ymmH ; ymmF=(21 31 23 33 25 35 27 37 29 39 2B 3B 2D 3D 2F 3F
+ ; 2H 3H 2J 3J 2L 3L 2N 3N 2P 3P 2R 3R 2T 3T 2V 3V)
+
+ vpunpckhwd ymmC, ymmA, ymmE ; ymmC=(08 18 28 38 0A 1A 2A 3A 0C 1C 2C 3C 0E 1E 2E 3E
+ ; 0O 1O 2O 3O 0Q 1Q 2Q 3Q 0S 1S 2S 3S 0U 1U 2U 3U)
+ vpunpcklwd ymmA, ymmA, ymmE ; ymmA=(00 10 20 30 02 12 22 32 04 14 24 34 06 16 26 36
+ ; 0G 1G 2G 3G 0I 1I 2I 3I 0K 1K 2K 3K 0M 1M 2M 3M)
+ vpunpckhwd ymmG, ymmB, ymmF ; ymmG=(09 19 29 39 0B 1B 2B 3B 0D 1D 2D 3D 0F 1F 2F 3F
+ ; 0P 1P 2P 3P 0R 1R 2R 3R 0T 1T 2T 3T 0V 1V 2V 3V)
+ vpunpcklwd ymmB, ymmB, ymmF ; ymmB=(01 11 21 31 03 13 23 33 05 15 25 35 07 17 27 37
+ ; 0H 1H 2H 3H 0J 1J 2J 3J 0L 1L 2L 3L 0N 1N 2N 3N)
+
+ vpunpckhdq ymmE, ymmA, ymmB ; ymmE=(04 14 24 34 05 15 25 35 06 16 26 36 07 17 27 37
+ ; 0K 1K 2K 3K 0L 1L 2L 3L 0M 1M 2M 3M 0N 1N 2N 3N)
+ vpunpckldq ymmB, ymmA, ymmB ; ymmB=(00 10 20 30 01 11 21 31 02 12 22 32 03 13 23 33
+ ; 0G 1G 2G 3G 0H 1H 2H 3H 0I 1I 2I 3I 0J 1J 2J 3J)
+ vpunpckhdq ymmF, ymmC, ymmG ; ymmF=(0C 1C 2C 3C 0D 1D 2D 3D 0E 1E 2E 3E 0F 1F 2F 3F
+ ; 0S 1S 2S 3S 0T 1T 2T 3T 0U 1U 2U 3U 0V 1V 2V 3V)
+ vpunpckldq ymmG, ymmC, ymmG ; ymmG=(08 18 28 38 09 19 29 39 0A 1A 2A 3A 0B 1B 2B 3B
+ ; 0O 1O 2O 3O 0P 1P 2P 3P 0Q 1Q 2Q 3Q 0R 1R 2R 3R)
+
+ vperm2i128 ymmA, ymmB, ymmE, 0x20 ; ymmA=(00 10 20 30 01 11 21 31 02 12 22 32 03 13 23 33
+ ; 04 14 24 34 05 15 25 35 06 16 26 36 07 17 27 37)
+ vperm2i128 ymmD, ymmG, ymmF, 0x20 ; ymmD=(08 18 28 38 09 19 29 39 0A 1A 2A 3A 0B 1B 2B 3B
+ ; 0C 1C 2C 3C 0D 1D 2D 3D 0E 1E 2E 3E 0F 1F 2F 3F)
+ vperm2i128 ymmC, ymmB, ymmE, 0x31 ; ymmC=(0G 1G 2G 3G 0H 1H 2H 3H 0I 1I 2I 3I 0J 1J 2J 3J
+ ; 0K 1K 2K 3K 0L 1L 2L 3L 0M 1M 2M 3M 0N 1N 2N 3N)
+ vperm2i128 ymmH, ymmG, ymmF, 0x31 ; ymmH=(0O 1O 2O 3O 0P 1P 2P 3P 0Q 1Q 2Q 3Q 0R 1R 2R 3R
+ ; 0S 1S 2S 3S 0T 1T 2T 3T 0U 1U 2U 3U 0V 1V 2V 3V)
+
+ cmp rcx, byte SIZEOF_YMMWORD
+ jb short .column_st64
+
+ test rdi, SIZEOF_YMMWORD-1
+ jnz short .out1
+ ; --(aligned)-------------------
+ vmovntdq YMMWORD [rdi+0*SIZEOF_YMMWORD], ymmA
+ vmovntdq YMMWORD [rdi+1*SIZEOF_YMMWORD], ymmD
+ vmovntdq YMMWORD [rdi+2*SIZEOF_YMMWORD], ymmC
+ vmovntdq YMMWORD [rdi+3*SIZEOF_YMMWORD], ymmH
+ jmp short .out0
+.out1: ; --(unaligned)-----------------
+ vmovdqu YMMWORD [rdi+0*SIZEOF_YMMWORD], ymmA
+ vmovdqu YMMWORD [rdi+1*SIZEOF_YMMWORD], ymmD
+ vmovdqu YMMWORD [rdi+2*SIZEOF_YMMWORD], ymmC
+ vmovdqu YMMWORD [rdi+3*SIZEOF_YMMWORD], ymmH
+.out0:
+ add rdi, RGB_PIXELSIZE*SIZEOF_YMMWORD ; outptr
+ sub rcx, byte SIZEOF_YMMWORD
+ jz near .nextrow
+
+ add rsi, byte SIZEOF_YMMWORD ; inptr0
+ add rbx, byte SIZEOF_YMMWORD ; inptr1
+ add rdx, byte SIZEOF_YMMWORD ; inptr2
+ jmp near .columnloop
+
+.column_st64:
+ cmp rcx, byte SIZEOF_YMMWORD/2
+ jb short .column_st32
+ vmovdqu YMMWORD [rdi+0*SIZEOF_YMMWORD], ymmA
+ vmovdqu YMMWORD [rdi+1*SIZEOF_YMMWORD], ymmD
+ add rdi, byte 2*SIZEOF_YMMWORD ; outptr
+ vmovdqa ymmA, ymmC
+ vmovdqa ymmD, ymmH
+ sub rcx, byte SIZEOF_YMMWORD/2
+.column_st32:
+ cmp rcx, byte SIZEOF_YMMWORD/4
+ jb short .column_st16
+ vmovdqu YMMWORD [rdi+0*SIZEOF_YMMWORD], ymmA
+ add rdi, byte SIZEOF_YMMWORD ; outptr
+ vmovdqa ymmA, ymmD
+ sub rcx, byte SIZEOF_YMMWORD/4
+.column_st16:
+ cmp rcx, byte SIZEOF_YMMWORD/8
+ jb short .column_st15
+ vmovdqu XMMWORD [rdi+0*SIZEOF_XMMWORD], xmmA
+ vperm2i128 ymmA, ymmA, ymmA, 1
+ add rdi, byte SIZEOF_XMMWORD ; outptr
+ sub rcx, byte SIZEOF_YMMWORD/8
+.column_st15:
+ ; Store two pixels (8 bytes) of ymmA to the output when it has enough
+ ; space.
+ cmp rcx, byte SIZEOF_YMMWORD/16
+ jb short .column_st7
+ vmovq MMWORD [rdi], xmmA
+ add rdi, byte SIZEOF_YMMWORD/16*4
+ sub rcx, byte SIZEOF_YMMWORD/16
+ vpsrldq xmmA, SIZEOF_YMMWORD/16*4
+.column_st7:
+ ; Store one pixel (4 bytes) of ymmA to the output when it has enough
+ ; space.
+ test rcx, rcx
+ jz short .nextrow
+ vmovd XMM_DWORD [rdi], xmmA
+
+%endif ; RGB_PIXELSIZE ; ---------------
+
+.nextrow:
+ pop rcx
+ pop rsi
+ pop rbx
+ pop rdx
+ pop rdi
+ pop rax
+
+ add rsi, byte SIZEOF_JSAMPROW
+ add rbx, byte SIZEOF_JSAMPROW
+ add rdx, byte SIZEOF_JSAMPROW
+ add rdi, byte SIZEOF_JSAMPROW ; output_buf
+ dec rax ; num_rows
+ jg near .rowloop
+
+ sfence ; flush the write buffer
+
+.return:
+ pop rbx
+ vzeroupper
+ uncollect_args 5
+ mov rsp, rbp ; rsp <- aligned rbp
+ pop rsp ; rsp <- original rbp
+ pop rbp
+ ret
+
+; For some reason, the OS X linker does not honor the request to align the
+; segment unless we do this.
+ align 32
diff --git a/simd/x86_64/jdcolext-sse2.asm b/simd/x86_64/jdcolext-sse2.asm
new file mode 100644
index 0000000..f49709f
--- /dev/null
+++ b/simd/x86_64/jdcolext-sse2.asm
@@ -0,0 +1,440 @@
+;
+; jdcolext.asm - colorspace conversion (64-bit SSE2)
+;
+; Copyright 2009, 2012 Pierre Ossman <ossman@cendio.se> for Cendio AB
+; Copyright (C) 2009, 2012, 2016, D. R. Commander.
+;
+; Based on the x86 SIMD extension for IJG JPEG library
+; Copyright (C) 1999-2006, MIYASAKA Masaru.
+; For conditions of distribution and use, see copyright notice in jsimdext.inc
+;
+; This file should be assembled with NASM (Netwide Assembler),
+; can *not* be assembled with Microsoft's MASM or any compatible
+; assembler (including Borland's Turbo Assembler).
+; NASM is available from http://nasm.sourceforge.net/ or
+; http://sourceforge.net/project/showfiles.php?group_id=6208
+;
+; [TAB8]
+
+%include "jcolsamp.inc"
+
+; --------------------------------------------------------------------------
+;
+; Convert some rows of samples to the output colorspace.
+;
+; GLOBAL(void)
+; jsimd_ycc_rgb_convert_sse2 (JDIMENSION out_width,
+; JSAMPIMAGE input_buf, JDIMENSION input_row,
+; JSAMPARRAY output_buf, int num_rows)
+;
+
+; r10d = JDIMENSION out_width
+; r11 = JSAMPIMAGE input_buf
+; r12d = JDIMENSION input_row
+; r13 = JSAMPARRAY output_buf
+; r14d = int num_rows
+
+%define wk(i) rbp-(WK_NUM-(i))*SIZEOF_XMMWORD ; xmmword wk[WK_NUM]
+%define WK_NUM 2
+
+ align 32
+ global EXTN(jsimd_ycc_rgb_convert_sse2)
+
+EXTN(jsimd_ycc_rgb_convert_sse2):
+ push rbp
+ mov rax, rsp ; rax = original rbp
+ sub rsp, byte 4
+ and rsp, byte (-SIZEOF_XMMWORD) ; align to 128 bits
+ mov [rsp], rax
+ mov rbp, rsp ; rbp = aligned rbp
+ lea rsp, [wk(0)]
+ collect_args 5
+ push rbx
+
+ mov ecx, r10d ; num_cols
+ test rcx, rcx
+ jz near .return
+
+ push rcx
+
+ mov rdi, r11
+ mov ecx, r12d
+ mov rsi, JSAMPARRAY [rdi+0*SIZEOF_JSAMPARRAY]
+ mov rbx, JSAMPARRAY [rdi+1*SIZEOF_JSAMPARRAY]
+ mov rdx, JSAMPARRAY [rdi+2*SIZEOF_JSAMPARRAY]
+ lea rsi, [rsi+rcx*SIZEOF_JSAMPROW]
+ lea rbx, [rbx+rcx*SIZEOF_JSAMPROW]
+ lea rdx, [rdx+rcx*SIZEOF_JSAMPROW]
+
+ pop rcx
+
+ mov rdi, r13
+ mov eax, r14d
+ test rax, rax
+ jle near .return
+.rowloop:
+ push rax
+ push rdi
+ push rdx
+ push rbx
+ push rsi
+ push rcx ; col
+
+ mov rsi, JSAMPROW [rsi] ; inptr0
+ mov rbx, JSAMPROW [rbx] ; inptr1
+ mov rdx, JSAMPROW [rdx] ; inptr2
+ mov rdi, JSAMPROW [rdi] ; outptr
+.columnloop:
+
+ movdqa xmm5, XMMWORD [rbx] ; xmm5=Cb(0123456789ABCDEF)
+ movdqa xmm1, XMMWORD [rdx] ; xmm1=Cr(0123456789ABCDEF)
+
+ pcmpeqw xmm4, xmm4
+ pcmpeqw xmm7, xmm7
+ psrlw xmm4, BYTE_BIT
+ psllw xmm7, 7 ; xmm7={0xFF80 0xFF80 0xFF80 0xFF80 ..}
+ movdqa xmm0, xmm4 ; xmm0=xmm4={0xFF 0x00 0xFF 0x00 ..}
+
+ pand xmm4, xmm5 ; xmm4=Cb(02468ACE)=CbE
+ psrlw xmm5, BYTE_BIT ; xmm5=Cb(13579BDF)=CbO
+ pand xmm0, xmm1 ; xmm0=Cr(02468ACE)=CrE
+ psrlw xmm1, BYTE_BIT ; xmm1=Cr(13579BDF)=CrO
+
+ paddw xmm4, xmm7
+ paddw xmm5, xmm7
+ paddw xmm0, xmm7
+ paddw xmm1, xmm7
+
+ ; (Original)
+ ; R = Y + 1.40200 * Cr
+ ; G = Y - 0.34414 * Cb - 0.71414 * Cr
+ ; B = Y + 1.77200 * Cb
+ ;
+ ; (This implementation)
+ ; R = Y + 0.40200 * Cr + Cr
+ ; G = Y - 0.34414 * Cb + 0.28586 * Cr - Cr
+ ; B = Y - 0.22800 * Cb + Cb + Cb
+
+ movdqa xmm2, xmm4 ; xmm2=CbE
+ movdqa xmm3, xmm5 ; xmm3=CbO
+ paddw xmm4, xmm4 ; xmm4=2*CbE
+ paddw xmm5, xmm5 ; xmm5=2*CbO
+ movdqa xmm6, xmm0 ; xmm6=CrE
+ movdqa xmm7, xmm1 ; xmm7=CrO
+ paddw xmm0, xmm0 ; xmm0=2*CrE
+ paddw xmm1, xmm1 ; xmm1=2*CrO
+
+ pmulhw xmm4, [rel PW_MF0228] ; xmm4=(2*CbE * -FIX(0.22800))
+ pmulhw xmm5, [rel PW_MF0228] ; xmm5=(2*CbO * -FIX(0.22800))
+ pmulhw xmm0, [rel PW_F0402] ; xmm0=(2*CrE * FIX(0.40200))
+ pmulhw xmm1, [rel PW_F0402] ; xmm1=(2*CrO * FIX(0.40200))
+
+ paddw xmm4, [rel PW_ONE]
+ paddw xmm5, [rel PW_ONE]
+ psraw xmm4, 1 ; xmm4=(CbE * -FIX(0.22800))
+ psraw xmm5, 1 ; xmm5=(CbO * -FIX(0.22800))
+ paddw xmm0, [rel PW_ONE]
+ paddw xmm1, [rel PW_ONE]
+ psraw xmm0, 1 ; xmm0=(CrE * FIX(0.40200))
+ psraw xmm1, 1 ; xmm1=(CrO * FIX(0.40200))
+
+ paddw xmm4, xmm2
+ paddw xmm5, xmm3
+ paddw xmm4, xmm2 ; xmm4=(CbE * FIX(1.77200))=(B-Y)E
+ paddw xmm5, xmm3 ; xmm5=(CbO * FIX(1.77200))=(B-Y)O
+ paddw xmm0, xmm6 ; xmm0=(CrE * FIX(1.40200))=(R-Y)E
+ paddw xmm1, xmm7 ; xmm1=(CrO * FIX(1.40200))=(R-Y)O
+
+ movdqa XMMWORD [wk(0)], xmm4 ; wk(0)=(B-Y)E
+ movdqa XMMWORD [wk(1)], xmm5 ; wk(1)=(B-Y)O
+
+ movdqa xmm4, xmm2
+ movdqa xmm5, xmm3
+ punpcklwd xmm2, xmm6
+ punpckhwd xmm4, xmm6
+ pmaddwd xmm2, [rel PW_MF0344_F0285]
+ pmaddwd xmm4, [rel PW_MF0344_F0285]
+ punpcklwd xmm3, xmm7
+ punpckhwd xmm5, xmm7
+ pmaddwd xmm3, [rel PW_MF0344_F0285]
+ pmaddwd xmm5, [rel PW_MF0344_F0285]
+
+ paddd xmm2, [rel PD_ONEHALF]
+ paddd xmm4, [rel PD_ONEHALF]
+ psrad xmm2, SCALEBITS
+ psrad xmm4, SCALEBITS
+ paddd xmm3, [rel PD_ONEHALF]
+ paddd xmm5, [rel PD_ONEHALF]
+ psrad xmm3, SCALEBITS
+ psrad xmm5, SCALEBITS
+
+ packssdw xmm2, xmm4 ; xmm2=CbE*-FIX(0.344)+CrE*FIX(0.285)
+ packssdw xmm3, xmm5 ; xmm3=CbO*-FIX(0.344)+CrO*FIX(0.285)
+ psubw xmm2, xmm6 ; xmm2=CbE*-FIX(0.344)+CrE*-FIX(0.714)=(G-Y)E
+ psubw xmm3, xmm7 ; xmm3=CbO*-FIX(0.344)+CrO*-FIX(0.714)=(G-Y)O
+
+ movdqa xmm5, XMMWORD [rsi] ; xmm5=Y(0123456789ABCDEF)
+
+ pcmpeqw xmm4, xmm4
+ psrlw xmm4, BYTE_BIT ; xmm4={0xFF 0x00 0xFF 0x00 ..}
+ pand xmm4, xmm5 ; xmm4=Y(02468ACE)=YE
+ psrlw xmm5, BYTE_BIT ; xmm5=Y(13579BDF)=YO
+
+ paddw xmm0, xmm4 ; xmm0=((R-Y)E+YE)=RE=R(02468ACE)
+ paddw xmm1, xmm5 ; xmm1=((R-Y)O+YO)=RO=R(13579BDF)
+ packuswb xmm0, xmm0 ; xmm0=R(02468ACE********)
+ packuswb xmm1, xmm1 ; xmm1=R(13579BDF********)
+
+ paddw xmm2, xmm4 ; xmm2=((G-Y)E+YE)=GE=G(02468ACE)
+ paddw xmm3, xmm5 ; xmm3=((G-Y)O+YO)=GO=G(13579BDF)
+ packuswb xmm2, xmm2 ; xmm2=G(02468ACE********)
+ packuswb xmm3, xmm3 ; xmm3=G(13579BDF********)
+
+ paddw xmm4, XMMWORD [wk(0)] ; xmm4=(YE+(B-Y)E)=BE=B(02468ACE)
+ paddw xmm5, XMMWORD [wk(1)] ; xmm5=(YO+(B-Y)O)=BO=B(13579BDF)
+ packuswb xmm4, xmm4 ; xmm4=B(02468ACE********)
+ packuswb xmm5, xmm5 ; xmm5=B(13579BDF********)
+
+%if RGB_PIXELSIZE == 3 ; ---------------
+
+ ; xmmA=(00 02 04 06 08 0A 0C 0E **), xmmB=(01 03 05 07 09 0B 0D 0F **)
+ ; xmmC=(10 12 14 16 18 1A 1C 1E **), xmmD=(11 13 15 17 19 1B 1D 1F **)
+ ; xmmE=(20 22 24 26 28 2A 2C 2E **), xmmF=(21 23 25 27 29 2B 2D 2F **)
+ ; xmmG=(** ** ** ** ** ** ** ** **), xmmH=(** ** ** ** ** ** ** ** **)
+
+ punpcklbw xmmA, xmmC ; xmmA=(00 10 02 12 04 14 06 16 08 18 0A 1A 0C 1C 0E 1E)
+ punpcklbw xmmE, xmmB ; xmmE=(20 01 22 03 24 05 26 07 28 09 2A 0B 2C 0D 2E 0F)
+ punpcklbw xmmD, xmmF ; xmmD=(11 21 13 23 15 25 17 27 19 29 1B 2B 1D 2D 1F 2F)
+
+ movdqa xmmG, xmmA
+ movdqa xmmH, xmmA
+ punpcklwd xmmA, xmmE ; xmmA=(00 10 20 01 02 12 22 03 04 14 24 05 06 16 26 07)
+ punpckhwd xmmG, xmmE ; xmmG=(08 18 28 09 0A 1A 2A 0B 0C 1C 2C 0D 0E 1E 2E 0F)
+
+ psrldq xmmH, 2 ; xmmH=(02 12 04 14 06 16 08 18 0A 1A 0C 1C 0E 1E -- --)
+ psrldq xmmE, 2 ; xmmE=(22 03 24 05 26 07 28 09 2A 0B 2C 0D 2E 0F -- --)
+
+ movdqa xmmC, xmmD
+ movdqa xmmB, xmmD
+ punpcklwd xmmD, xmmH ; xmmD=(11 21 02 12 13 23 04 14 15 25 06 16 17 27 08 18)
+ punpckhwd xmmC, xmmH ; xmmC=(19 29 0A 1A 1B 2B 0C 1C 1D 2D 0E 1E 1F 2F -- --)
+
+ psrldq xmmB, 2 ; xmmB=(13 23 15 25 17 27 19 29 1B 2B 1D 2D 1F 2F -- --)
+
+ movdqa xmmF, xmmE
+ punpcklwd xmmE, xmmB ; xmmE=(22 03 13 23 24 05 15 25 26 07 17 27 28 09 19 29)
+ punpckhwd xmmF, xmmB ; xmmF=(2A 0B 1B 2B 2C 0D 1D 2D 2E 0F 1F 2F -- -- -- --)
+
+ pshufd xmmH, xmmA, 0x4E ; xmmH=(04 14 24 05 06 16 26 07 00 10 20 01 02 12 22 03)
+ movdqa xmmB, xmmE
+ punpckldq xmmA, xmmD ; xmmA=(00 10 20 01 11 21 02 12 02 12 22 03 13 23 04 14)
+ punpckldq xmmE, xmmH ; xmmE=(22 03 13 23 04 14 24 05 24 05 15 25 06 16 26 07)
+ punpckhdq xmmD, xmmB ; xmmD=(15 25 06 16 26 07 17 27 17 27 08 18 28 09 19 29)
+
+ pshufd xmmH, xmmG, 0x4E ; xmmH=(0C 1C 2C 0D 0E 1E 2E 0F 08 18 28 09 0A 1A 2A 0B)
+ movdqa xmmB, xmmF
+ punpckldq xmmG, xmmC ; xmmG=(08 18 28 09 19 29 0A 1A 0A 1A 2A 0B 1B 2B 0C 1C)
+ punpckldq xmmF, xmmH ; xmmF=(2A 0B 1B 2B 0C 1C 2C 0D 2C 0D 1D 2D 0E 1E 2E 0F)
+ punpckhdq xmmC, xmmB ; xmmC=(1D 2D 0E 1E 2E 0F 1F 2F 1F 2F -- -- -- -- -- --)
+
+ punpcklqdq xmmA, xmmE ; xmmA=(00 10 20 01 11 21 02 12 22 03 13 23 04 14 24 05)
+ punpcklqdq xmmD, xmmG ; xmmD=(15 25 06 16 26 07 17 27 08 18 28 09 19 29 0A 1A)
+ punpcklqdq xmmF, xmmC ; xmmF=(2A 0B 1B 2B 0C 1C 2C 0D 1D 2D 0E 1E 2E 0F 1F 2F)
+
+ cmp rcx, byte SIZEOF_XMMWORD
+ jb short .column_st32
+
+ test rdi, SIZEOF_XMMWORD-1
+ jnz short .out1
+ ; --(aligned)-------------------
+ movntdq XMMWORD [rdi+0*SIZEOF_XMMWORD], xmmA
+ movntdq XMMWORD [rdi+1*SIZEOF_XMMWORD], xmmD
+ movntdq XMMWORD [rdi+2*SIZEOF_XMMWORD], xmmF
+ jmp short .out0
+.out1: ; --(unaligned)-----------------
+ movdqu XMMWORD [rdi+0*SIZEOF_XMMWORD], xmmA
+ movdqu XMMWORD [rdi+1*SIZEOF_XMMWORD], xmmD
+ movdqu XMMWORD [rdi+2*SIZEOF_XMMWORD], xmmF
+.out0:
+ add rdi, byte RGB_PIXELSIZE*SIZEOF_XMMWORD ; outptr
+ sub rcx, byte SIZEOF_XMMWORD
+ jz near .nextrow
+
+ add rsi, byte SIZEOF_XMMWORD ; inptr0
+ add rbx, byte SIZEOF_XMMWORD ; inptr1
+ add rdx, byte SIZEOF_XMMWORD ; inptr2
+ jmp near .columnloop
+
+.column_st32:
+ lea rcx, [rcx+rcx*2] ; imul ecx, RGB_PIXELSIZE
+ cmp rcx, byte 2*SIZEOF_XMMWORD
+ jb short .column_st16
+ movdqu XMMWORD [rdi+0*SIZEOF_XMMWORD], xmmA
+ movdqu XMMWORD [rdi+1*SIZEOF_XMMWORD], xmmD
+ add rdi, byte 2*SIZEOF_XMMWORD ; outptr
+ movdqa xmmA, xmmF
+ sub rcx, byte 2*SIZEOF_XMMWORD
+ jmp short .column_st15
+.column_st16:
+ cmp rcx, byte SIZEOF_XMMWORD
+ jb short .column_st15
+ movdqu XMMWORD [rdi+0*SIZEOF_XMMWORD], xmmA
+ add rdi, byte SIZEOF_XMMWORD ; outptr
+ movdqa xmmA, xmmD
+ sub rcx, byte SIZEOF_XMMWORD
+.column_st15:
+ ; Store the lower 8 bytes of xmmA to the output when it has enough
+ ; space.
+ cmp rcx, byte SIZEOF_MMWORD
+ jb short .column_st7
+ movq XMM_MMWORD [rdi], xmmA
+ add rdi, byte SIZEOF_MMWORD
+ sub rcx, byte SIZEOF_MMWORD
+ psrldq xmmA, SIZEOF_MMWORD
+.column_st7:
+ ; Store the lower 4 bytes of xmmA to the output when it has enough
+ ; space.
+ cmp rcx, byte SIZEOF_DWORD
+ jb short .column_st3
+ movd XMM_DWORD [rdi], xmmA
+ add rdi, byte SIZEOF_DWORD
+ sub rcx, byte SIZEOF_DWORD
+ psrldq xmmA, SIZEOF_DWORD
+.column_st3:
+ ; Store the lower 2 bytes of rax to the output when it has enough
+ ; space.
+ movd eax, xmmA
+ cmp rcx, byte SIZEOF_WORD
+ jb short .column_st1
+ mov WORD [rdi], ax
+ add rdi, byte SIZEOF_WORD
+ sub rcx, byte SIZEOF_WORD
+ shr rax, 16
+.column_st1:
+ ; Store the lower 1 byte of rax to the output when it has enough
+ ; space.
+ test rcx, rcx
+ jz short .nextrow
+ mov BYTE [rdi], al
+
+%else ; RGB_PIXELSIZE == 4 ; -----------
+
+%ifdef RGBX_FILLER_0XFF
+ pcmpeqb xmm6, xmm6 ; xmm6=XE=X(02468ACE********)
+ pcmpeqb xmm7, xmm7 ; xmm7=XO=X(13579BDF********)
+%else
+ pxor xmm6, xmm6 ; xmm6=XE=X(02468ACE********)
+ pxor xmm7, xmm7 ; xmm7=XO=X(13579BDF********)
+%endif
+ ; xmmA=(00 02 04 06 08 0A 0C 0E **), xmmB=(01 03 05 07 09 0B 0D 0F **)
+ ; xmmC=(10 12 14 16 18 1A 1C 1E **), xmmD=(11 13 15 17 19 1B 1D 1F **)
+ ; xmmE=(20 22 24 26 28 2A 2C 2E **), xmmF=(21 23 25 27 29 2B 2D 2F **)
+ ; xmmG=(30 32 34 36 38 3A 3C 3E **), xmmH=(31 33 35 37 39 3B 3D 3F **)
+
+ punpcklbw xmmA, xmmC ; xmmA=(00 10 02 12 04 14 06 16 08 18 0A 1A 0C 1C 0E 1E)
+ punpcklbw xmmE, xmmG ; xmmE=(20 30 22 32 24 34 26 36 28 38 2A 3A 2C 3C 2E 3E)
+ punpcklbw xmmB, xmmD ; xmmB=(01 11 03 13 05 15 07 17 09 19 0B 1B 0D 1D 0F 1F)
+ punpcklbw xmmF, xmmH ; xmmF=(21 31 23 33 25 35 27 37 29 39 2B 3B 2D 3D 2F 3F)
+
+ movdqa xmmC, xmmA
+ punpcklwd xmmA, xmmE ; xmmA=(00 10 20 30 02 12 22 32 04 14 24 34 06 16 26 36)
+ punpckhwd xmmC, xmmE ; xmmC=(08 18 28 38 0A 1A 2A 3A 0C 1C 2C 3C 0E 1E 2E 3E)
+ movdqa xmmG, xmmB
+ punpcklwd xmmB, xmmF ; xmmB=(01 11 21 31 03 13 23 33 05 15 25 35 07 17 27 37)
+ punpckhwd xmmG, xmmF ; xmmG=(09 19 29 39 0B 1B 2B 3B 0D 1D 2D 3D 0F 1F 2F 3F)
+
+ movdqa xmmD, xmmA
+ punpckldq xmmA, xmmB ; xmmA=(00 10 20 30 01 11 21 31 02 12 22 32 03 13 23 33)
+ punpckhdq xmmD, xmmB ; xmmD=(04 14 24 34 05 15 25 35 06 16 26 36 07 17 27 37)
+ movdqa xmmH, xmmC
+ punpckldq xmmC, xmmG ; xmmC=(08 18 28 38 09 19 29 39 0A 1A 2A 3A 0B 1B 2B 3B)
+ punpckhdq xmmH, xmmG ; xmmH=(0C 1C 2C 3C 0D 1D 2D 3D 0E 1E 2E 3E 0F 1F 2F 3F)
+
+ cmp rcx, byte SIZEOF_XMMWORD
+ jb short .column_st32
+
+ test rdi, SIZEOF_XMMWORD-1
+ jnz short .out1
+ ; --(aligned)-------------------
+ movntdq XMMWORD [rdi+0*SIZEOF_XMMWORD], xmmA
+ movntdq XMMWORD [rdi+1*SIZEOF_XMMWORD], xmmD
+ movntdq XMMWORD [rdi+2*SIZEOF_XMMWORD], xmmC
+ movntdq XMMWORD [rdi+3*SIZEOF_XMMWORD], xmmH
+ jmp short .out0
+.out1: ; --(unaligned)-----------------
+ movdqu XMMWORD [rdi+0*SIZEOF_XMMWORD], xmmA
+ movdqu XMMWORD [rdi+1*SIZEOF_XMMWORD], xmmD
+ movdqu XMMWORD [rdi+2*SIZEOF_XMMWORD], xmmC
+ movdqu XMMWORD [rdi+3*SIZEOF_XMMWORD], xmmH
+.out0:
+ add rdi, byte RGB_PIXELSIZE*SIZEOF_XMMWORD ; outptr
+ sub rcx, byte SIZEOF_XMMWORD
+ jz near .nextrow
+
+ add rsi, byte SIZEOF_XMMWORD ; inptr0
+ add rbx, byte SIZEOF_XMMWORD ; inptr1
+ add rdx, byte SIZEOF_XMMWORD ; inptr2
+ jmp near .columnloop
+
+.column_st32:
+ cmp rcx, byte SIZEOF_XMMWORD/2
+ jb short .column_st16
+ movdqu XMMWORD [rdi+0*SIZEOF_XMMWORD], xmmA
+ movdqu XMMWORD [rdi+1*SIZEOF_XMMWORD], xmmD
+ add rdi, byte 2*SIZEOF_XMMWORD ; outptr
+ movdqa xmmA, xmmC
+ movdqa xmmD, xmmH
+ sub rcx, byte SIZEOF_XMMWORD/2
+.column_st16:
+ cmp rcx, byte SIZEOF_XMMWORD/4
+ jb short .column_st15
+ movdqu XMMWORD [rdi+0*SIZEOF_XMMWORD], xmmA
+ add rdi, byte SIZEOF_XMMWORD ; outptr
+ movdqa xmmA, xmmD
+ sub rcx, byte SIZEOF_XMMWORD/4
+.column_st15:
+ ; Store two pixels (8 bytes) of xmmA to the output when it has enough
+ ; space.
+ cmp rcx, byte SIZEOF_XMMWORD/8
+ jb short .column_st7
+ movq MMWORD [rdi], xmmA
+ add rdi, byte SIZEOF_XMMWORD/8*4
+ sub rcx, byte SIZEOF_XMMWORD/8
+ psrldq xmmA, SIZEOF_XMMWORD/8*4
+.column_st7:
+ ; Store one pixel (4 bytes) of xmmA to the output when it has enough
+ ; space.
+ test rcx, rcx
+ jz short .nextrow
+ movd XMM_DWORD [rdi], xmmA
+
+%endif ; RGB_PIXELSIZE ; ---------------
+
+.nextrow:
+ pop rcx
+ pop rsi
+ pop rbx
+ pop rdx
+ pop rdi
+ pop rax
+
+ add rsi, byte SIZEOF_JSAMPROW
+ add rbx, byte SIZEOF_JSAMPROW
+ add rdx, byte SIZEOF_JSAMPROW
+ add rdi, byte SIZEOF_JSAMPROW ; output_buf
+ dec rax ; num_rows
+ jg near .rowloop
+
+ sfence ; flush the write buffer
+
+.return:
+ pop rbx
+ uncollect_args 5
+ mov rsp, rbp ; rsp <- aligned rbp
+ pop rsp ; rsp <- original rbp
+ pop rbp
+ ret
+
+; For some reason, the OS X linker does not honor the request to align the
+; segment unless we do this.
+ align 32
diff --git a/simd/x86_64/jdcolor-avx2.asm b/simd/x86_64/jdcolor-avx2.asm
new file mode 100644
index 0000000..dd878a9
--- /dev/null
+++ b/simd/x86_64/jdcolor-avx2.asm
@@ -0,0 +1,120 @@
+;
+; jdcolor.asm - colorspace conversion (64-bit AVX2)
+;
+; Copyright 2009 Pierre Ossman <ossman@cendio.se> for Cendio AB
+; Copyright (C) 2009, 2016, D. R. Commander.
+; Copyright (C) 2015, Intel Corporation.
+;
+; Based on the x86 SIMD extension for IJG JPEG library
+; Copyright (C) 1999-2006, MIYASAKA Masaru.
+; For conditions of distribution and use, see copyright notice in jsimdext.inc
+;
+; This file should be assembled with NASM (Netwide Assembler),
+; can *not* be assembled with Microsoft's MASM or any compatible
+; assembler (including Borland's Turbo Assembler).
+; NASM is available from http://nasm.sourceforge.net/ or
+; http://sourceforge.net/project/showfiles.php?group_id=6208
+;
+; [TAB8]
+
+%include "jsimdext.inc"
+
+; --------------------------------------------------------------------------
+
+%define SCALEBITS 16
+
+F_0_344 equ 22554 ; FIX(0.34414)
+F_0_714 equ 46802 ; FIX(0.71414)
+F_1_402 equ 91881 ; FIX(1.40200)
+F_1_772 equ 116130 ; FIX(1.77200)
+F_0_402 equ (F_1_402 - 65536) ; FIX(1.40200) - FIX(1)
+F_0_285 equ ( 65536 - F_0_714) ; FIX(1) - FIX(0.71414)
+F_0_228 equ (131072 - F_1_772) ; FIX(2) - FIX(1.77200)
+
+; --------------------------------------------------------------------------
+ SECTION SEG_CONST
+
+ alignz 32
+ global EXTN(jconst_ycc_rgb_convert_avx2)
+
+EXTN(jconst_ycc_rgb_convert_avx2):
+
+PW_F0402 times 16 dw F_0_402
+PW_MF0228 times 16 dw -F_0_228
+PW_MF0344_F0285 times 8 dw -F_0_344, F_0_285
+PW_ONE times 16 dw 1
+PD_ONEHALF times 8 dd 1 << (SCALEBITS-1)
+
+ alignz 32
+
+; --------------------------------------------------------------------------
+ SECTION SEG_TEXT
+ BITS 64
+
+%include "jdcolext-avx2.asm"
+
+%undef RGB_RED
+%undef RGB_GREEN
+%undef RGB_BLUE
+%undef RGB_PIXELSIZE
+%define RGB_RED EXT_RGB_RED
+%define RGB_GREEN EXT_RGB_GREEN
+%define RGB_BLUE EXT_RGB_BLUE
+%define RGB_PIXELSIZE EXT_RGB_PIXELSIZE
+%define jsimd_ycc_rgb_convert_avx2 jsimd_ycc_extrgb_convert_avx2
+%include "jdcolext-avx2.asm"
+
+%undef RGB_RED
+%undef RGB_GREEN
+%undef RGB_BLUE
+%undef RGB_PIXELSIZE
+%define RGB_RED EXT_RGBX_RED
+%define RGB_GREEN EXT_RGBX_GREEN
+%define RGB_BLUE EXT_RGBX_BLUE
+%define RGB_PIXELSIZE EXT_RGBX_PIXELSIZE
+%define jsimd_ycc_rgb_convert_avx2 jsimd_ycc_extrgbx_convert_avx2
+%include "jdcolext-avx2.asm"
+
+%undef RGB_RED
+%undef RGB_GREEN
+%undef RGB_BLUE
+%undef RGB_PIXELSIZE
+%define RGB_RED EXT_BGR_RED
+%define RGB_GREEN EXT_BGR_GREEN
+%define RGB_BLUE EXT_BGR_BLUE
+%define RGB_PIXELSIZE EXT_BGR_PIXELSIZE
+%define jsimd_ycc_rgb_convert_avx2 jsimd_ycc_extbgr_convert_avx2
+%include "jdcolext-avx2.asm"
+
+%undef RGB_RED
+%undef RGB_GREEN
+%undef RGB_BLUE
+%undef RGB_PIXELSIZE
+%define RGB_RED EXT_BGRX_RED
+%define RGB_GREEN EXT_BGRX_GREEN
+%define RGB_BLUE EXT_BGRX_BLUE
+%define RGB_PIXELSIZE EXT_BGRX_PIXELSIZE
+%define jsimd_ycc_rgb_convert_avx2 jsimd_ycc_extbgrx_convert_avx2
+%include "jdcolext-avx2.asm"
+
+%undef RGB_RED
+%undef RGB_GREEN
+%undef RGB_BLUE
+%undef RGB_PIXELSIZE
+%define RGB_RED EXT_XBGR_RED
+%define RGB_GREEN EXT_XBGR_GREEN
+%define RGB_BLUE EXT_XBGR_BLUE
+%define RGB_PIXELSIZE EXT_XBGR_PIXELSIZE
+%define jsimd_ycc_rgb_convert_avx2 jsimd_ycc_extxbgr_convert_avx2
+%include "jdcolext-avx2.asm"
+
+%undef RGB_RED
+%undef RGB_GREEN
+%undef RGB_BLUE
+%undef RGB_PIXELSIZE
+%define RGB_RED EXT_XRGB_RED
+%define RGB_GREEN EXT_XRGB_GREEN
+%define RGB_BLUE EXT_XRGB_BLUE
+%define RGB_PIXELSIZE EXT_XRGB_PIXELSIZE
+%define jsimd_ycc_rgb_convert_avx2 jsimd_ycc_extxrgb_convert_avx2
+%include "jdcolext-avx2.asm"
diff --git a/simd/jdcolor-sse2.asm b/simd/x86_64/jdcolor-sse2.asm
similarity index 80%
copy from simd/jdcolor-sse2.asm
copy to simd/x86_64/jdcolor-sse2.asm
index 7ff5d05..55e10a4 100644
--- a/simd/jdcolor-sse2.asm
+++ b/simd/x86_64/jdcolor-sse2.asm
@@ -1,8 +1,8 @@
;
-; jdcolor.asm - colorspace conversion (SSE2)
+; jdcolor.asm - colorspace conversion (64-bit SSE2)
;
; Copyright 2009 Pierre Ossman <ossman@cendio.se> for Cendio AB
-; Copyright (C) 2009, D. R. Commander.
+; Copyright (C) 2009, 2016, D. R. Commander.
;
; Based on the x86 SIMD extension for IJG JPEG library
; Copyright (C) 1999-2006, MIYASAKA Masaru.
@@ -20,21 +20,21 @@
; --------------------------------------------------------------------------
-%define SCALEBITS 16
+%define SCALEBITS 16
-F_0_344 equ 22554 ; FIX(0.34414)
-F_0_714 equ 46802 ; FIX(0.71414)
-F_1_402 equ 91881 ; FIX(1.40200)
-F_1_772 equ 116130 ; FIX(1.77200)
-F_0_402 equ (F_1_402 - 65536) ; FIX(1.40200) - FIX(1)
-F_0_285 equ ( 65536 - F_0_714) ; FIX(1) - FIX(0.71414)
-F_0_228 equ (131072 - F_1_772) ; FIX(2) - FIX(1.77200)
+F_0_344 equ 22554 ; FIX(0.34414)
+F_0_714 equ 46802 ; FIX(0.71414)
+F_1_402 equ 91881 ; FIX(1.40200)
+F_1_772 equ 116130 ; FIX(1.77200)
+F_0_402 equ (F_1_402 - 65536) ; FIX(1.40200) - FIX(1)
+F_0_285 equ ( 65536 - F_0_714) ; FIX(1) - FIX(0.71414)
+F_0_228 equ (131072 - F_1_772) ; FIX(2) - FIX(1.77200)
; --------------------------------------------------------------------------
- SECTION SEG_CONST
+ SECTION SEG_CONST
- alignz 16
- global EXTN(jconst_ycc_rgb_convert_sse2)
+ alignz 32
+ global EXTN(jconst_ycc_rgb_convert_sse2)
EXTN(jconst_ycc_rgb_convert_sse2):
@@ -44,11 +44,11 @@
PW_ONE times 8 dw 1
PD_ONEHALF times 4 dd 1 << (SCALEBITS-1)
- alignz 16
+ alignz 32
; --------------------------------------------------------------------------
- SECTION SEG_TEXT
- BITS 32
+ SECTION SEG_TEXT
+ BITS 64
%include "jdcolext-sse2.asm"
diff --git a/simd/x86_64/jdmerge-avx2.asm b/simd/x86_64/jdmerge-avx2.asm
new file mode 100644
index 0000000..fa3a321
--- /dev/null
+++ b/simd/x86_64/jdmerge-avx2.asm
@@ -0,0 +1,126 @@
+;
+; jdmerge.asm - merged upsampling/color conversion (64-bit AVX2)
+;
+; Copyright 2009 Pierre Ossman <ossman@cendio.se> for Cendio AB
+; Copyright (C) 2009, 2016, D. R. Commander.
+; Copyright (C) 2015, Intel Corporation.
+;
+; Based on the x86 SIMD extension for IJG JPEG library
+; Copyright (C) 1999-2006, MIYASAKA Masaru.
+; For conditions of distribution and use, see copyright notice in jsimdext.inc
+;
+; This file should be assembled with NASM (Netwide Assembler),
+; can *not* be assembled with Microsoft's MASM or any compatible
+; assembler (including Borland's Turbo Assembler).
+; NASM is available from http://nasm.sourceforge.net/ or
+; http://sourceforge.net/project/showfiles.php?group_id=6208
+;
+; [TAB8]
+
+%include "jsimdext.inc"
+
+; --------------------------------------------------------------------------
+
+%define SCALEBITS 16
+
+F_0_344 equ 22554 ; FIX(0.34414)
+F_0_714 equ 46802 ; FIX(0.71414)
+F_1_402 equ 91881 ; FIX(1.40200)
+F_1_772 equ 116130 ; FIX(1.77200)
+F_0_402 equ (F_1_402 - 65536) ; FIX(1.40200) - FIX(1)
+F_0_285 equ ( 65536 - F_0_714) ; FIX(1) - FIX(0.71414)
+F_0_228 equ (131072 - F_1_772) ; FIX(2) - FIX(1.77200)
+
+; --------------------------------------------------------------------------
+ SECTION SEG_CONST
+
+ alignz 32
+ global EXTN(jconst_merged_upsample_avx2)
+
+EXTN(jconst_merged_upsample_avx2):
+
+PW_F0402 times 16 dw F_0_402
+PW_MF0228 times 16 dw -F_0_228
+PW_MF0344_F0285 times 8 dw -F_0_344, F_0_285
+PW_ONE times 16 dw 1
+PD_ONEHALF times 8 dd 1 << (SCALEBITS-1)
+
+ alignz 32
+
+; --------------------------------------------------------------------------
+ SECTION SEG_TEXT
+ BITS 64
+
+%include "jdmrgext-avx2.asm"
+
+%undef RGB_RED
+%undef RGB_GREEN
+%undef RGB_BLUE
+%undef RGB_PIXELSIZE
+%define RGB_RED EXT_RGB_RED
+%define RGB_GREEN EXT_RGB_GREEN
+%define RGB_BLUE EXT_RGB_BLUE
+%define RGB_PIXELSIZE EXT_RGB_PIXELSIZE
+%define jsimd_h2v1_merged_upsample_avx2 jsimd_h2v1_extrgb_merged_upsample_avx2
+%define jsimd_h2v2_merged_upsample_avx2 jsimd_h2v2_extrgb_merged_upsample_avx2
+%include "jdmrgext-avx2.asm"
+
+%undef RGB_RED
+%undef RGB_GREEN
+%undef RGB_BLUE
+%undef RGB_PIXELSIZE
+%define RGB_RED EXT_RGBX_RED
+%define RGB_GREEN EXT_RGBX_GREEN
+%define RGB_BLUE EXT_RGBX_BLUE
+%define RGB_PIXELSIZE EXT_RGBX_PIXELSIZE
+%define jsimd_h2v1_merged_upsample_avx2 jsimd_h2v1_extrgbx_merged_upsample_avx2
+%define jsimd_h2v2_merged_upsample_avx2 jsimd_h2v2_extrgbx_merged_upsample_avx2
+%include "jdmrgext-avx2.asm"
+
+%undef RGB_RED
+%undef RGB_GREEN
+%undef RGB_BLUE
+%undef RGB_PIXELSIZE
+%define RGB_RED EXT_BGR_RED
+%define RGB_GREEN EXT_BGR_GREEN
+%define RGB_BLUE EXT_BGR_BLUE
+%define RGB_PIXELSIZE EXT_BGR_PIXELSIZE
+%define jsimd_h2v1_merged_upsample_avx2 jsimd_h2v1_extbgr_merged_upsample_avx2
+%define jsimd_h2v2_merged_upsample_avx2 jsimd_h2v2_extbgr_merged_upsample_avx2
+%include "jdmrgext-avx2.asm"
+
+%undef RGB_RED
+%undef RGB_GREEN
+%undef RGB_BLUE
+%undef RGB_PIXELSIZE
+%define RGB_RED EXT_BGRX_RED
+%define RGB_GREEN EXT_BGRX_GREEN
+%define RGB_BLUE EXT_BGRX_BLUE
+%define RGB_PIXELSIZE EXT_BGRX_PIXELSIZE
+%define jsimd_h2v1_merged_upsample_avx2 jsimd_h2v1_extbgrx_merged_upsample_avx2
+%define jsimd_h2v2_merged_upsample_avx2 jsimd_h2v2_extbgrx_merged_upsample_avx2
+%include "jdmrgext-avx2.asm"
+
+%undef RGB_RED
+%undef RGB_GREEN
+%undef RGB_BLUE
+%undef RGB_PIXELSIZE
+%define RGB_RED EXT_XBGR_RED
+%define RGB_GREEN EXT_XBGR_GREEN
+%define RGB_BLUE EXT_XBGR_BLUE
+%define RGB_PIXELSIZE EXT_XBGR_PIXELSIZE
+%define jsimd_h2v1_merged_upsample_avx2 jsimd_h2v1_extxbgr_merged_upsample_avx2
+%define jsimd_h2v2_merged_upsample_avx2 jsimd_h2v2_extxbgr_merged_upsample_avx2
+%include "jdmrgext-avx2.asm"
+
+%undef RGB_RED
+%undef RGB_GREEN
+%undef RGB_BLUE
+%undef RGB_PIXELSIZE
+%define RGB_RED EXT_XRGB_RED
+%define RGB_GREEN EXT_XRGB_GREEN
+%define RGB_BLUE EXT_XRGB_BLUE
+%define RGB_PIXELSIZE EXT_XRGB_PIXELSIZE
+%define jsimd_h2v1_merged_upsample_avx2 jsimd_h2v1_extxrgb_merged_upsample_avx2
+%define jsimd_h2v2_merged_upsample_avx2 jsimd_h2v2_extxrgb_merged_upsample_avx2
+%include "jdmrgext-avx2.asm"
diff --git a/simd/jdmerge-sse2.asm b/simd/x86_64/jdmerge-sse2.asm
similarity index 82%
copy from simd/jdmerge-sse2.asm
copy to simd/x86_64/jdmerge-sse2.asm
index 236de5a..a488916 100644
--- a/simd/jdmerge-sse2.asm
+++ b/simd/x86_64/jdmerge-sse2.asm
@@ -1,8 +1,8 @@
;
-; jdmerge.asm - merged upsampling/color conversion (SSE2)
+; jdmerge.asm - merged upsampling/color conversion (64-bit SSE2)
;
; Copyright 2009 Pierre Ossman <ossman@cendio.se> for Cendio AB
-; Copyright (C) 2009, D. R. Commander.
+; Copyright (C) 2009, 2016, D. R. Commander.
;
; Based on the x86 SIMD extension for IJG JPEG library
; Copyright (C) 1999-2006, MIYASAKA Masaru.
@@ -20,21 +20,21 @@
; --------------------------------------------------------------------------
-%define SCALEBITS 16
+%define SCALEBITS 16
-F_0_344 equ 22554 ; FIX(0.34414)
-F_0_714 equ 46802 ; FIX(0.71414)
-F_1_402 equ 91881 ; FIX(1.40200)
-F_1_772 equ 116130 ; FIX(1.77200)
-F_0_402 equ (F_1_402 - 65536) ; FIX(1.40200) - FIX(1)
-F_0_285 equ ( 65536 - F_0_714) ; FIX(1) - FIX(0.71414)
-F_0_228 equ (131072 - F_1_772) ; FIX(2) - FIX(1.77200)
+F_0_344 equ 22554 ; FIX(0.34414)
+F_0_714 equ 46802 ; FIX(0.71414)
+F_1_402 equ 91881 ; FIX(1.40200)
+F_1_772 equ 116130 ; FIX(1.77200)
+F_0_402 equ (F_1_402 - 65536) ; FIX(1.40200) - FIX(1)
+F_0_285 equ ( 65536 - F_0_714) ; FIX(1) - FIX(0.71414)
+F_0_228 equ (131072 - F_1_772) ; FIX(2) - FIX(1.77200)
; --------------------------------------------------------------------------
- SECTION SEG_CONST
+ SECTION SEG_CONST
- alignz 16
- global EXTN(jconst_merged_upsample_sse2)
+ alignz 32
+ global EXTN(jconst_merged_upsample_sse2)
EXTN(jconst_merged_upsample_sse2):
@@ -44,11 +44,11 @@
PW_ONE times 8 dw 1
PD_ONEHALF times 4 dd 1 << (SCALEBITS-1)
- alignz 16
+ alignz 32
; --------------------------------------------------------------------------
- SECTION SEG_TEXT
- BITS 32
+ SECTION SEG_TEXT
+ BITS 64
%include "jdmrgext-sse2.asm"
diff --git a/simd/x86_64/jdmrgext-avx2.asm b/simd/x86_64/jdmrgext-avx2.asm
new file mode 100644
index 0000000..546a9e5
--- /dev/null
+++ b/simd/x86_64/jdmrgext-avx2.asm
@@ -0,0 +1,595 @@
+;
+; jdmrgext.asm - merged upsampling/color conversion (64-bit AVX2)
+;
+; Copyright 2009, 2012 Pierre Ossman <ossman@cendio.se> for Cendio AB
+; Copyright (C) 2009, 2012, 2016, D. R. Commander.
+; Copyright (C) 2015, Intel Corporation.
+;
+; Based on the x86 SIMD extension for IJG JPEG library
+; Copyright (C) 1999-2006, MIYASAKA Masaru.
+; For conditions of distribution and use, see copyright notice in jsimdext.inc
+;
+; This file should be assembled with NASM (Netwide Assembler),
+; can *not* be assembled with Microsoft's MASM or any compatible
+; assembler (including Borland's Turbo Assembler).
+; NASM is available from http://nasm.sourceforge.net/ or
+; http://sourceforge.net/project/showfiles.php?group_id=6208
+;
+; [TAB8]
+
+%include "jcolsamp.inc"
+
+; --------------------------------------------------------------------------
+;
+; Upsample and color convert for the case of 2:1 horizontal and 1:1 vertical.
+;
+; GLOBAL(void)
+; jsimd_h2v1_merged_upsample_avx2 (JDIMENSION output_width,
+; JSAMPIMAGE input_buf,
+; JDIMENSION in_row_group_ctr,
+; JSAMPARRAY output_buf);
+;
+
+; r10d = JDIMENSION output_width
+; r11 = JSAMPIMAGE input_buf
+; r12d = JDIMENSION in_row_group_ctr
+; r13 = JSAMPARRAY output_buf
+
+%define wk(i) rbp-(WK_NUM-(i))*SIZEOF_YMMWORD ; ymmword wk[WK_NUM]
+%define WK_NUM 3
+
+ align 32
+ global EXTN(jsimd_h2v1_merged_upsample_avx2)
+
+EXTN(jsimd_h2v1_merged_upsample_avx2):
+ push rbp
+ mov rax, rsp ; rax = original rbp
+ sub rsp, byte 4
+ and rsp, byte (-SIZEOF_YMMWORD) ; align to 256 bits
+ mov [rsp], rax
+ mov rbp, rsp ; rbp = aligned rbp
+ lea rsp, [wk(0)]
+ collect_args 4
+ push rbx
+
+ mov ecx, r10d ; col
+ test rcx, rcx
+ jz near .return
+
+ push rcx
+
+ mov rdi, r11
+ mov ecx, r12d
+ mov rsi, JSAMPARRAY [rdi+0*SIZEOF_JSAMPARRAY]
+ mov rbx, JSAMPARRAY [rdi+1*SIZEOF_JSAMPARRAY]
+ mov rdx, JSAMPARRAY [rdi+2*SIZEOF_JSAMPARRAY]
+ mov rdi, r13
+ mov rsi, JSAMPROW [rsi+rcx*SIZEOF_JSAMPROW] ; inptr0
+ mov rbx, JSAMPROW [rbx+rcx*SIZEOF_JSAMPROW] ; inptr1
+ mov rdx, JSAMPROW [rdx+rcx*SIZEOF_JSAMPROW] ; inptr2
+ mov rdi, JSAMPROW [rdi] ; outptr
+
+ pop rcx ; col
+
+.columnloop:
+
+ vmovdqu ymm6, YMMWORD [rbx] ; ymm6=Cb(0123456789ABCDEFGHIJKLMNOPQRSTUV)
+ vmovdqu ymm7, YMMWORD [rdx] ; ymm7=Cr(0123456789ABCDEFGHIJKLMNOPQRSTUV)
+
+ vpxor ymm1, ymm1, ymm1 ; ymm1=(all 0's)
+ vpcmpeqw ymm3, ymm3, ymm3
+ vpsllw ymm3, ymm3, 7 ; ymm3={0xFF80 0xFF80 0xFF80 0xFF80 ..}
+
+ vpermq ymm6, ymm6, 0xd8 ; ymm6=Cb(01234567GHIJKLMN89ABCDEFOPQRSTUV)
+ vpermq ymm7, ymm7, 0xd8 ; ymm7=Cr(01234567GHIJKLMN89ABCDEFOPQRSTUV)
+ vpunpcklbw ymm4, ymm6, ymm1 ; ymm4=Cb(0123456789ABCDEF)=CbL
+ vpunpckhbw ymm6, ymm6, ymm1 ; ymm6=Cb(GHIJKLMNOPQRSTUV)=CbH
+ vpunpcklbw ymm0, ymm7, ymm1 ; ymm0=Cr(0123456789ABCDEF)=CrL
+ vpunpckhbw ymm7, ymm7, ymm1 ; ymm7=Cr(GHIJKLMNOPQRSTUV)=CrH
+
+ vpaddw ymm5, ymm6, ymm3
+ vpaddw ymm2, ymm4, ymm3
+ vpaddw ymm1, ymm7, ymm3
+ vpaddw ymm3, ymm0, ymm3
+
+ ; (Original)
+ ; R = Y + 1.40200 * Cr
+ ; G = Y - 0.34414 * Cb - 0.71414 * Cr
+ ; B = Y + 1.77200 * Cb
+ ;
+ ; (This implementation)
+ ; R = Y + 0.40200 * Cr + Cr
+ ; G = Y - 0.34414 * Cb + 0.28586 * Cr - Cr
+ ; B = Y - 0.22800 * Cb + Cb + Cb
+
+ vpaddw ymm6, ymm5, ymm5 ; ymm6=2*CbH
+ vpaddw ymm4, ymm2, ymm2 ; ymm4=2*CbL
+ vpaddw ymm7, ymm1, ymm1 ; ymm7=2*CrH
+ vpaddw ymm0, ymm3, ymm3 ; ymm0=2*CrL
+
+ vpmulhw ymm6, ymm6, [rel PW_MF0228] ; ymm6=(2*CbH * -FIX(0.22800))
+ vpmulhw ymm4, ymm4, [rel PW_MF0228] ; ymm4=(2*CbL * -FIX(0.22800))
+ vpmulhw ymm7, ymm7, [rel PW_F0402] ; ymm7=(2*CrH * FIX(0.40200))
+ vpmulhw ymm0, ymm0, [rel PW_F0402] ; ymm0=(2*CrL * FIX(0.40200))
+
+ vpaddw ymm6, ymm6, [rel PW_ONE]
+ vpaddw ymm4, ymm4, [rel PW_ONE]
+ vpsraw ymm6, ymm6, 1 ; ymm6=(CbH * -FIX(0.22800))
+ vpsraw ymm4, ymm4, 1 ; ymm4=(CbL * -FIX(0.22800))
+ vpaddw ymm7, ymm7, [rel PW_ONE]
+ vpaddw ymm0, ymm0, [rel PW_ONE]
+ vpsraw ymm7, ymm7, 1 ; ymm7=(CrH * FIX(0.40200))
+ vpsraw ymm0, ymm0, 1 ; ymm0=(CrL * FIX(0.40200))
+
+ vpaddw ymm6, ymm6, ymm5
+ vpaddw ymm4, ymm4, ymm2
+ vpaddw ymm6, ymm6, ymm5 ; ymm6=(CbH * FIX(1.77200))=(B-Y)H
+ vpaddw ymm4, ymm4, ymm2 ; ymm4=(CbL * FIX(1.77200))=(B-Y)L
+ vpaddw ymm7, ymm7, ymm1 ; ymm7=(CrH * FIX(1.40200))=(R-Y)H
+ vpaddw ymm0, ymm0, ymm3 ; ymm0=(CrL * FIX(1.40200))=(R-Y)L
+
+ vmovdqa YMMWORD [wk(0)], ymm6 ; wk(0)=(B-Y)H
+ vmovdqa YMMWORD [wk(1)], ymm7 ; wk(1)=(R-Y)H
+
+ vpunpckhwd ymm6, ymm5, ymm1
+ vpunpcklwd ymm5, ymm5, ymm1
+ vpmaddwd ymm5, ymm5, [rel PW_MF0344_F0285]
+ vpmaddwd ymm6, ymm6, [rel PW_MF0344_F0285]
+ vpunpckhwd ymm7, ymm2, ymm3
+ vpunpcklwd ymm2, ymm2, ymm3
+ vpmaddwd ymm2, ymm2, [rel PW_MF0344_F0285]
+ vpmaddwd ymm7, ymm7, [rel PW_MF0344_F0285]
+
+ vpaddd ymm5, ymm5, [rel PD_ONEHALF]
+ vpaddd ymm6, ymm6, [rel PD_ONEHALF]
+ vpsrad ymm5, ymm5, SCALEBITS
+ vpsrad ymm6, ymm6, SCALEBITS
+ vpaddd ymm2, ymm2, [rel PD_ONEHALF]
+ vpaddd ymm7, ymm7, [rel PD_ONEHALF]
+ vpsrad ymm2, ymm2, SCALEBITS
+ vpsrad ymm7, ymm7, SCALEBITS
+
+ vpackssdw ymm5, ymm5, ymm6 ; ymm5=CbH*-FIX(0.344)+CrH*FIX(0.285)
+ vpackssdw ymm2, ymm2, ymm7 ; ymm2=CbL*-FIX(0.344)+CrL*FIX(0.285)
+ vpsubw ymm5, ymm5, ymm1 ; ymm5=CbH*-FIX(0.344)+CrH*-FIX(0.714)=(G-Y)H
+ vpsubw ymm2, ymm2, ymm3 ; ymm2=CbL*-FIX(0.344)+CrL*-FIX(0.714)=(G-Y)L
+
+ vmovdqa YMMWORD [wk(2)], ymm5 ; wk(2)=(G-Y)H
+
+ mov al, 2 ; Yctr
+ jmp short .Yloop_1st
+
+.Yloop_2nd:
+ vmovdqa ymm0, YMMWORD [wk(1)] ; ymm0=(R-Y)H
+ vmovdqa ymm2, YMMWORD [wk(2)] ; ymm2=(G-Y)H
+ vmovdqa ymm4, YMMWORD [wk(0)] ; ymm4=(B-Y)H
+
+.Yloop_1st:
+ vmovdqu ymm7, YMMWORD [rsi] ; ymm7=Y(0123456789ABCDEFGHIJKLMNOPQRSTUV)
+
+ vpcmpeqw ymm6, ymm6, ymm6
+ vpsrlw ymm6, ymm6, BYTE_BIT ; ymm6={0xFF 0x00 0xFF 0x00 ..}
+ vpand ymm6, ymm6, ymm7 ; ymm6=Y(02468ACEGIKMOQSU)=YE
+ vpsrlw ymm7, ymm7, BYTE_BIT ; ymm7=Y(13579BDFHJLNPRTV)=YO
+
+ vmovdqa ymm1, ymm0 ; ymm1=ymm0=(R-Y)(L/H)
+ vmovdqa ymm3, ymm2 ; ymm3=ymm2=(G-Y)(L/H)
+ vmovdqa ymm5, ymm4 ; ymm5=ymm4=(B-Y)(L/H)
+
+ vpaddw ymm0, ymm0, ymm6 ; ymm0=((R-Y)+YE)=RE=R(02468ACEGIKMOQSU)
+ vpaddw ymm1, ymm1, ymm7 ; ymm1=((R-Y)+YO)=RO=R(13579BDFHJLNPRTV)
+ vpackuswb ymm0, ymm0, ymm0 ; ymm0=R(02468ACE********GIKMOQSU********)
+ vpackuswb ymm1, ymm1, ymm1 ; ymm1=R(13579BDF********HJLNPRTV********)
+
+ vpaddw ymm2, ymm2, ymm6 ; ymm2=((G-Y)+YE)=GE=G(02468ACEGIKMOQSU)
+ vpaddw ymm3, ymm3, ymm7 ; ymm3=((G-Y)+YO)=GO=G(13579BDFHJLNPRTV)
+ vpackuswb ymm2, ymm2, ymm2 ; ymm2=G(02468ACE********GIKMOQSU********)
+ vpackuswb ymm3, ymm3, ymm3 ; ymm3=G(13579BDF********HJLNPRTV********)
+
+ vpaddw ymm4, ymm4, ymm6 ; ymm4=((B-Y)+YE)=BE=B(02468ACEGIKMOQSU)
+ vpaddw ymm5, ymm5, ymm7 ; ymm5=((B-Y)+YO)=BO=B(13579BDFHJLNPRTV)
+ vpackuswb ymm4, ymm4, ymm4 ; ymm4=B(02468ACE********GIKMOQSU********)
+ vpackuswb ymm5, ymm5, ymm5 ; ymm5=B(13579BDF********HJLNPRTV********)
+
+%if RGB_PIXELSIZE == 3 ; ---------------
+
+ ; ymmA=(00 02 04 06 08 0A 0C 0E ** 0G 0I 0K 0M 0O 0Q 0S 0U **)
+ ; ymmB=(01 03 05 07 09 0B 0D 0F ** 0H 0J 0L 0N 0P 0R 0T 0V **)
+ ; ymmC=(10 12 14 16 18 1A 1C 1E ** 1G 1I 1K 1M 1O 1Q 1S 1U **)
+ ; ymmD=(11 13 15 17 19 1B 1D 1F ** 1H 1J 1L 1N 1P 1R 1T 1V **)
+ ; ymmE=(20 22 24 26 28 2A 2C 2E ** 2G 2I 2K 2M 2O 2Q 2S 2U **)
+ ; ymmF=(21 23 25 27 29 2B 2D 2F ** 2H 2J 2L 2N 2P 2R 2T 2V **)
+ ; ymmG=(** ** ** ** ** ** ** ** ** ** ** ** ** ** ** ** ** **)
+ ; ymmH=(** ** ** ** ** ** ** ** ** ** ** ** ** ** ** ** ** **)
+
+ vpunpcklbw ymmA, ymmA, ymmC ; ymmA=(00 10 02 12 04 14 06 16 08 18 0A 1A 0C 1C 0E 1E
+ ; 0G 1G 0I 1I 0K 1K 0M 1M 0O 1O 0Q 1Q 0S 1S 0U 1U)
+ vpunpcklbw ymmE, ymmE, ymmB ; ymmE=(20 01 22 03 24 05 26 07 28 09 2A 0B 2C 0D 2E 0F
+ ; 2G 0H 2I 0J 2K 0L 2M 0N 2O 0P 2Q 0R 2S 0T 2U 0V)
+ vpunpcklbw ymmD, ymmD, ymmF ; ymmD=(11 21 13 23 15 25 17 27 19 29 1B 2B 1D 2D 1F 2F
+ ; 1H 2H 1J 2J 1L 2L 1N 2N 1P 2P 1R 2R 1T 2T 1V 2V)
+
+ vpsrldq ymmH, ymmA, 2 ; ymmH=(02 12 04 14 06 16 08 18 0A 1A 0C 1C 0E 1E 0G 1G
+ ; 0I 1I 0K 1K 0M 1M 0O 1O 0Q 1Q 0S 1S 0U 1U -- --)
+ vpunpckhwd ymmG, ymmA, ymmE ; ymmG=(08 18 28 09 0A 1A 2A 0B 0C 1C 2C 0D 0E 1E 2E 0F
+ ; 0O 1O 2O 0P 0Q 1Q 2Q 0R 0S 1S 2S 0T 0U 1U 2U 0V)
+ vpunpcklwd ymmA, ymmA, ymmE ; ymmA=(00 10 20 01 02 12 22 03 04 14 24 05 06 16 26 07
+ ; 0G 1G 2G 0H 0I 1I 2I 0J 0K 1K 2K 0L 0M 1M 2M 0N)
+
+ vpsrldq ymmE, ymmE, 2 ; ymmE=(22 03 24 05 26 07 28 09 2A 0B 2C 0D 2E 0F 2G 0H
+ ; 2I 0J 2K 0L 2M 0N 2O 0P 2Q 0R 2S 0T 2U 0V -- --)
+
+ vpsrldq ymmB, ymmD, 2 ; ymmB=(13 23 15 25 17 27 19 29 1B 2B 1D 2D 1F 2F 1H 2H
+ ; 1J 2J 1L 2L 1N 2N 1P 2P 1R 2R 1T 2T 1V 2V -- --)
+ vpunpckhwd ymmC, ymmD, ymmH ; ymmC=(19 29 0A 1A 1B 2B 0C 1C 1D 2D 0E 1E 1F 2F 0G 1G
+ ; 1P 2P 0Q 1Q 1R 2R 0S 1S 1T 2T 0U 1U 1V 2V -- --)
+ vpunpcklwd ymmD, ymmD, ymmH ; ymmD=(11 21 02 12 13 23 04 14 15 25 06 16 17 27 08 18
+ ; 1H 2H 0I 1I 1J 2J 0K 1K 1L 2L 0M 1M 1N 2N 0O 1O)
+
+ vpunpckhwd ymmF, ymmE, ymmB ; ymmF=(2A 0B 1B 2B 2C 0D 1D 2D 2E 0F 1F 2F 2G 0H 1H 2H
+ ; 2Q 0R 1R 2R 2S 0T 1T 2T 2U 0V 1V 2V -- -- -- --)
+ vpunpcklwd ymmE, ymmE, ymmB ; ymmE=(22 03 13 23 24 05 15 25 26 07 17 27 28 09 19 29
+ ; 2I 0J 1J 2J 2K 0L 1L 2L 2M 0N 1N 2N 2O 0P 1P 2P)
+
+ vpshufd ymmH, ymmA, 0x4E ; ymmH=(04 14 24 05 06 16 26 07 00 10 20 01 02 12 22 03
+ ; 0K 1K 2K 0L 0M 1M 2M 0N 0G 1G 2G 0H 0I 1I 2I 0J)
+ vpunpckldq ymmA, ymmA, ymmD ; ymmA=(00 10 20 01 11 21 02 12 02 12 22 03 13 23 04 14
+ ; 0G 1G 2G 0H 1H 2H 0I 1I 0I 1I 2I 0J 1J 2J 0K 1K)
+ vpunpckhdq ymmD, ymmD, ymmE ; ymmD=(15 25 06 16 26 07 17 27 17 27 08 18 28 09 19 29
+ ; 1L 2L 0M 1M 2M 0N 1N 2N 1N 2N 0O 1O 2O 0P 1P 2P)
+ vpunpckldq ymmE, ymmE, ymmH ; ymmE=(22 03 13 23 04 14 24 05 24 05 15 25 06 16 26 07
+ ; 2I 0J 1J 2J 0K 1K 2K 0L 2K 0L 1L 2L 0M 1M 2M 0N)
+
+ vpshufd ymmH, ymmG, 0x4E ; ymmH=(0C 1C 2C 0D 0E 1E 2E 0F 08 18 28 09 0A 1A 2A 0B
+ ; 0S 1S 2S 0T 0U 1U 2U 0V 0O 1O 2O 0P 0Q 1Q 2Q 0R)
+ vpunpckldq ymmG, ymmG, ymmC ; ymmG=(08 18 28 09 19 29 0A 1A 0A 1A 2A 0B 1B 2B 0C 1C
+ ; 0O 1O 2O 0P 1P 2P 0Q 1Q 0Q 1Q 2Q 0R 1R 2R 0S 1S)
+ vpunpckhdq ymmC, ymmC, ymmF ; ymmC=(1D 2D 0E 1E 2E 0F 1F 2F 1F 2F 0G 1G 2G 0H 1H 2H
+ ; 1T 2T 0U 1U 2U 0V 1V 2V 1V 2V -- -- -- -- -- --)
+ vpunpckldq ymmF, ymmF, ymmH ; ymmF=(2A 0B 1B 2B 0C 1C 2C 0D 2C 0D 1D 2D 0E 1E 2E 0F
+ ; 2Q 0R 1R 2R 0S 1S 2S 0T 2S 0T 1T 2T 0U 1U 2U 0V)
+
+ vpunpcklqdq ymmH, ymmA, ymmE ; ymmH=(00 10 20 01 11 21 02 12 22 03 13 23 04 14 24 05
+ ; 0G 1G 2G 0H 1H 2H 0I 1I 2I 0J 1J 2J 0K 1K 2K 0L)
+ vpunpcklqdq ymmG, ymmD, ymmG ; ymmG=(15 25 06 16 26 07 17 27 08 18 28 09 19 29 0A 1A
+ ; 1L 2L 0M 1M 2M 0N 1N 2N 0O 1O 2O 0P 1P 2P 0Q 1Q)
+ vpunpcklqdq ymmC, ymmF, ymmC ; ymmC=(2A 0B 1B 2B 0C 1C 2C 0D 1D 2D 0E 1E 2E 0F 1F 2F
+ ; 2Q 0R 1R 2R 0S 1S 2S 0T 1T 2T 0U 1U 2U 0V 1V 2V)
+
+ vperm2i128 ymmA, ymmH, ymmG, 0x20 ; ymmA=(00 10 20 01 11 21 02 12 22 03 13 23 04 14 24 05
+ ; 15 25 06 16 26 07 17 27 08 18 28 09 19 29 0A 1A)
+ vperm2i128 ymmD, ymmC, ymmH, 0x30 ; ymmD=(2A 0B 1B 2B 0C 1C 2C 0D 1D 2D 0E 1E 2E 0F 1F 2F
+ ; 0G 1G 2G 0H 1H 2H 0I 1I 2I 0J 1J 2J 0K 1K 2K 0L)
+ vperm2i128 ymmF, ymmG, ymmC, 0x31 ; ymmF=(1L 2L 0M 1M 2M 0N 1N 2N 0O 1O 2O 0P 1P 2P 0Q 1Q
+ ; 2Q 0R 1R 2R 0S 1S 2S 0T 1T 2T 0U 1U 2U 0V 1V 2V)
+
+ cmp rcx, byte SIZEOF_YMMWORD
+ jb short .column_st64
+
+ test rdi, SIZEOF_YMMWORD-1
+ jnz short .out1
+ ; --(aligned)-------------------
+ vmovntdq YMMWORD [rdi+0*SIZEOF_YMMWORD], ymmA
+ vmovntdq YMMWORD [rdi+1*SIZEOF_YMMWORD], ymmD
+ vmovntdq YMMWORD [rdi+2*SIZEOF_YMMWORD], ymmF
+ jmp short .out0
+.out1: ; --(unaligned)-----------------
+ vmovdqu YMMWORD [rdi+0*SIZEOF_YMMWORD], ymmA
+ vmovdqu YMMWORD [rdi+1*SIZEOF_YMMWORD], ymmD
+ vmovdqu YMMWORD [rdi+2*SIZEOF_YMMWORD], ymmF
+.out0:
+ add rdi, byte RGB_PIXELSIZE*SIZEOF_YMMWORD ; outptr
+ sub rcx, byte SIZEOF_YMMWORD
+ jz near .endcolumn
+
+ add rsi, byte SIZEOF_YMMWORD ; inptr0
+ dec al ; Yctr
+ jnz near .Yloop_2nd
+
+ add rbx, byte SIZEOF_YMMWORD ; inptr1
+ add rdx, byte SIZEOF_YMMWORD ; inptr2
+ jmp near .columnloop
+
+.column_st64:
+ lea rcx, [rcx+rcx*2] ; imul ecx, RGB_PIXELSIZE
+ cmp rcx, byte 2*SIZEOF_YMMWORD
+ jb short .column_st32
+ vmovdqu YMMWORD [rdi+0*SIZEOF_YMMWORD], ymmA
+ vmovdqu YMMWORD [rdi+1*SIZEOF_YMMWORD], ymmD
+ add rdi, byte 2*SIZEOF_YMMWORD ; outptr
+ vmovdqa ymmA, ymmF
+ sub rcx, byte 2*SIZEOF_YMMWORD
+ jmp short .column_st31
+.column_st32:
+ cmp rcx, byte SIZEOF_YMMWORD
+ jb short .column_st31
+ vmovdqu YMMWORD [rdi+0*SIZEOF_YMMWORD], ymmA
+ add rdi, byte SIZEOF_YMMWORD ; outptr
+ vmovdqa ymmA, ymmD
+ sub rcx, byte SIZEOF_YMMWORD
+ jmp short .column_st31
+.column_st31:
+ cmp rcx, byte SIZEOF_XMMWORD
+ jb short .column_st15
+ vmovdqu XMMWORD [rdi+0*SIZEOF_XMMWORD], xmmA
+ add rdi, byte SIZEOF_XMMWORD ; outptr
+ vperm2i128 ymmA, ymmA, ymmA, 1
+ sub rcx, byte SIZEOF_XMMWORD
+.column_st15:
+ ; Store the lower 8 bytes of xmmA to the output when it has enough
+ ; space.
+ cmp rcx, byte SIZEOF_MMWORD
+ jb short .column_st7
+ vmovq XMM_MMWORD [rdi], xmmA
+ add rdi, byte SIZEOF_MMWORD
+ sub rcx, byte SIZEOF_MMWORD
+ vpsrldq xmmA, xmmA, SIZEOF_MMWORD
+.column_st7:
+ ; Store the lower 4 bytes of xmmA to the output when it has enough
+ ; space.
+ cmp rcx, byte SIZEOF_DWORD
+ jb short .column_st3
+ vmovd XMM_DWORD [rdi], xmmA
+ add rdi, byte SIZEOF_DWORD
+ sub rcx, byte SIZEOF_DWORD
+ vpsrldq xmmA, xmmA, SIZEOF_DWORD
+.column_st3:
+ ; Store the lower 2 bytes of rax to the output when it has enough
+ ; space.
+ vmovd eax, xmmA
+ cmp rcx, byte SIZEOF_WORD
+ jb short .column_st1
+ mov WORD [rdi], ax
+ add rdi, byte SIZEOF_WORD
+ sub rcx, byte SIZEOF_WORD
+ shr rax, 16
+.column_st1:
+ ; Store the lower 1 byte of rax to the output when it has enough
+ ; space.
+ test rcx, rcx
+ jz short .endcolumn
+ mov BYTE [rdi], al
+
+%else ; RGB_PIXELSIZE == 4 ; -----------
+
+%ifdef RGBX_FILLER_0XFF
+ vpcmpeqb ymm6, ymm6, ymm6 ; ymm6=XE=X(02468ACE********GIKMOQSU********)
+ vpcmpeqb ymm7, ymm7, ymm7 ; ymm7=XO=X(13579BDF********HJLNPRTV********)
+%else
+ vpxor ymm6, ymm6, ymm6 ; ymm6=XE=X(02468ACE********GIKMOQSU********)
+ vpxor ymm7, ymm7, ymm7 ; ymm7=XO=X(13579BDF********HJLNPRTV********)
+%endif
+ ; ymmA=(00 02 04 06 08 0A 0C 0E ** 0G 0I 0K 0M 0O 0Q 0S 0U **)
+ ; ymmB=(01 03 05 07 09 0B 0D 0F ** 0H 0J 0L 0N 0P 0R 0T 0V **)
+ ; ymmC=(10 12 14 16 18 1A 1C 1E ** 1G 1I 1K 1M 1O 1Q 1S 1U **)
+ ; ymmD=(11 13 15 17 19 1B 1D 1F ** 1H 1J 1L 1N 1P 1R 1T 1V **)
+ ; ymmE=(20 22 24 26 28 2A 2C 2E ** 2G 2I 2K 2M 2O 2Q 2S 2U **)
+ ; ymmF=(21 23 25 27 29 2B 2D 2F ** 2H 2J 2L 2N 2P 2R 2T 2V **)
+ ; ymmG=(30 32 34 36 38 3A 3C 3E ** 3G 3I 3K 3M 3O 3Q 3S 3U **)
+ ; ymmH=(31 33 35 37 39 3B 3D 3F ** 3H 3J 3L 3N 3P 3R 3T 3V **)
+
+ vpunpcklbw ymmA, ymmA, ymmC ; ymmA=(00 10 02 12 04 14 06 16 08 18 0A 1A 0C 1C 0E 1E
+ ; 0G 1G 0I 1I 0K 1K 0M 1M 0O 1O 0Q 1Q 0S 1S 0U 1U)
+ vpunpcklbw ymmE, ymmE, ymmG ; ymmE=(20 30 22 32 24 34 26 36 28 38 2A 3A 2C 3C 2E 3E
+ ; 2G 3G 2I 3I 2K 3K 2M 3M 2O 3O 2Q 3Q 2S 3S 2U 3U)
+ vpunpcklbw ymmB, ymmB, ymmD ; ymmB=(01 11 03 13 05 15 07 17 09 19 0B 1B 0D 1D 0F 1F
+ ; 0H 1H 0J 1J 0L 1L 0N 1N 0P 1P 0R 1R 0T 1T 0V 1V)
+ vpunpcklbw ymmF, ymmF, ymmH ; ymmF=(21 31 23 33 25 35 27 37 29 39 2B 3B 2D 3D 2F 3F
+ ; 2H 3H 2J 3J 2L 3L 2N 3N 2P 3P 2R 3R 2T 3T 2V 3V)
+
+ vpunpckhwd ymmC, ymmA, ymmE ; ymmC=(08 18 28 38 0A 1A 2A 3A 0C 1C 2C 3C 0E 1E 2E 3E
+ ; 0O 1O 2O 3O 0Q 1Q 2Q 3Q 0S 1S 2S 3S 0U 1U 2U 3U)
+ vpunpcklwd ymmA, ymmA, ymmE ; ymmA=(00 10 20 30 02 12 22 32 04 14 24 34 06 16 26 36
+ ; 0G 1G 2G 3G 0I 1I 2I 3I 0K 1K 2K 3K 0M 1M 2M 3M)
+ vpunpckhwd ymmG, ymmB, ymmF ; ymmG=(09 19 29 39 0B 1B 2B 3B 0D 1D 2D 3D 0F 1F 2F 3F
+ ; 0P 1P 2P 3P 0R 1R 2R 3R 0T 1T 2T 3T 0V 1V 2V 3V)
+ vpunpcklwd ymmB, ymmB, ymmF ; ymmB=(01 11 21 31 03 13 23 33 05 15 25 35 07 17 27 37
+ ; 0H 1H 2H 3H 0J 1J 2J 3J 0L 1L 2L 3L 0N 1N 2N 3N)
+
+ vpunpckhdq ymmE, ymmA, ymmB ; ymmE=(04 14 24 34 05 15 25 35 06 16 26 36 07 17 27 37
+ ; 0K 1K 2K 3K 0L 1L 2L 3L 0M 1M 2M 3M 0N 1N 2N 3N)
+ vpunpckldq ymmB, ymmA, ymmB ; ymmB=(00 10 20 30 01 11 21 31 02 12 22 32 03 13 23 33
+ ; 0G 1G 2G 3G 0H 1H 2H 3H 0I 1I 2I 3I 0J 1J 2J 3J)
+ vpunpckhdq ymmF, ymmC, ymmG ; ymmF=(0C 1C 2C 3C 0D 1D 2D 3D 0E 1E 2E 3E 0F 1F 2F 3F
+ ; 0S 1S 2S 3S 0T 1T 2T 3T 0U 1U 2U 3U 0V 1V 2V 3V)
+ vpunpckldq ymmG, ymmC, ymmG ; ymmG=(08 18 28 38 09 19 29 39 0A 1A 2A 3A 0B 1B 2B 3B
+ ; 0O 1O 2O 3O 0P 1P 2P 3P 0Q 1Q 2Q 3Q 0R 1R 2R 3R)
+
+ vperm2i128 ymmA, ymmB, ymmE, 0x20 ; ymmA=(00 10 20 30 01 11 21 31 02 12 22 32 03 13 23 33
+ ; 04 14 24 34 05 15 25 35 06 16 26 36 07 17 27 37)
+ vperm2i128 ymmD, ymmG, ymmF, 0x20 ; ymmD=(08 18 28 38 09 19 29 39 0A 1A 2A 3A 0B 1B 2B 3B
+ ; 0C 1C 2C 3C 0D 1D 2D 3D 0E 1E 2E 3E 0F 1F 2F 3F)
+ vperm2i128 ymmC, ymmB, ymmE, 0x31 ; ymmC=(0G 1G 2G 3G 0H 1H 2H 3H 0I 1I 2I 3I 0J 1J 2J 3J
+ ; 0K 1K 2K 3K 0L 1L 2L 3L 0M 1M 2M 3M 0N 1N 2N 3N)
+ vperm2i128 ymmH, ymmG, ymmF, 0x31 ; ymmH=(0O 1O 2O 3O 0P 1P 2P 3P 0Q 1Q 2Q 3Q 0R 1R 2R 3R
+ ; 0S 1S 2S 3S 0T 1T 2T 3T 0U 1U 2U 3U 0V 1V 2V 3V)
+
+ cmp rcx, byte SIZEOF_YMMWORD
+ jb short .column_st64
+
+ test rdi, SIZEOF_YMMWORD-1
+ jnz short .out1
+ ; --(aligned)-------------------
+ vmovntdq YMMWORD [rdi+0*SIZEOF_YMMWORD], ymmA
+ vmovntdq YMMWORD [rdi+1*SIZEOF_YMMWORD], ymmD
+ vmovntdq YMMWORD [rdi+2*SIZEOF_YMMWORD], ymmC
+ vmovntdq YMMWORD [rdi+3*SIZEOF_YMMWORD], ymmH
+ jmp short .out0
+.out1: ; --(unaligned)-----------------
+ vmovdqu YMMWORD [rdi+0*SIZEOF_YMMWORD], ymmA
+ vmovdqu YMMWORD [rdi+1*SIZEOF_YMMWORD], ymmD
+ vmovdqu YMMWORD [rdi+2*SIZEOF_YMMWORD], ymmC
+ vmovdqu YMMWORD [rdi+3*SIZEOF_YMMWORD], ymmH
+.out0:
+ add rdi, RGB_PIXELSIZE*SIZEOF_YMMWORD ; outptr
+ sub rcx, byte SIZEOF_YMMWORD
+ jz near .endcolumn
+
+ add rsi, byte SIZEOF_YMMWORD ; inptr0
+ dec al
+ jnz near .Yloop_2nd
+
+ add rbx, byte SIZEOF_YMMWORD ; inptr1
+ add rdx, byte SIZEOF_YMMWORD ; inptr2
+ jmp near .columnloop
+
+.column_st64:
+ cmp rcx, byte SIZEOF_YMMWORD/2
+ jb short .column_st32
+ vmovdqu YMMWORD [rdi+0*SIZEOF_YMMWORD], ymmA
+ vmovdqu YMMWORD [rdi+1*SIZEOF_YMMWORD], ymmD
+ add rdi, byte 2*SIZEOF_YMMWORD ; outptr
+ vmovdqa ymmA, ymmC
+ vmovdqa ymmD, ymmH
+ sub rcx, byte SIZEOF_YMMWORD/2
+.column_st32:
+ cmp rcx, byte SIZEOF_YMMWORD/4
+ jb short .column_st16
+ vmovdqu YMMWORD [rdi+0*SIZEOF_YMMWORD], ymmA
+ add rdi, byte SIZEOF_YMMWORD ; outptr
+ vmovdqa ymmA, ymmD
+ sub rcx, byte SIZEOF_YMMWORD/4
+.column_st16:
+ cmp rcx, byte SIZEOF_YMMWORD/8
+ jb short .column_st15
+ vmovdqu XMMWORD [rdi+0*SIZEOF_XMMWORD], xmmA
+ add rdi, byte SIZEOF_XMMWORD ; outptr
+ vperm2i128 ymmA, ymmA, ymmA, 1
+ sub rcx, byte SIZEOF_YMMWORD/8
+.column_st15:
+ ; Store two pixels (8 bytes) of ymmA to the output when it has enough
+ ; space.
+ cmp rcx, byte SIZEOF_YMMWORD/16
+ jb short .column_st7
+ vmovq MMWORD [rdi], xmmA
+ add rdi, byte SIZEOF_YMMWORD/16*4
+ sub rcx, byte SIZEOF_YMMWORD/16
+ vpsrldq xmmA, SIZEOF_YMMWORD/16*4
+.column_st7:
+ ; Store one pixel (4 bytes) of ymmA to the output when it has enough
+ ; space.
+ test rcx, rcx
+ jz short .endcolumn
+ vmovd XMM_DWORD [rdi], xmmA
+
+%endif ; RGB_PIXELSIZE ; ---------------
+
+.endcolumn:
+ sfence ; flush the write buffer
+
+.return:
+ pop rbx
+ vzeroupper
+ uncollect_args 4
+ mov rsp, rbp ; rsp <- aligned rbp
+ pop rsp ; rsp <- original rbp
+ pop rbp
+ ret
+
+; --------------------------------------------------------------------------
+;
+; Upsample and color convert for the case of 2:1 horizontal and 2:1 vertical.
+;
+; GLOBAL(void)
+; jsimd_h2v2_merged_upsample_avx2 (JDIMENSION output_width,
+; JSAMPIMAGE input_buf,
+; JDIMENSION in_row_group_ctr,
+; JSAMPARRAY output_buf);
+;
+
+; r10d = JDIMENSION output_width
+; r11 = JSAMPIMAGE input_buf
+; r12d = JDIMENSION in_row_group_ctr
+; r13 = JSAMPARRAY output_buf
+
+ align 32
+ global EXTN(jsimd_h2v2_merged_upsample_avx2)
+
+EXTN(jsimd_h2v2_merged_upsample_avx2):
+ push rbp
+ mov rax, rsp
+ mov rbp, rsp
+ collect_args 4
+ push rbx
+
+ mov eax, r10d
+
+ mov rdi, r11
+ mov ecx, r12d
+ mov rsi, JSAMPARRAY [rdi+0*SIZEOF_JSAMPARRAY]
+ mov rbx, JSAMPARRAY [rdi+1*SIZEOF_JSAMPARRAY]
+ mov rdx, JSAMPARRAY [rdi+2*SIZEOF_JSAMPARRAY]
+ mov rdi, r13
+ lea rsi, [rsi+rcx*SIZEOF_JSAMPROW]
+
+ push rdx ; inptr2
+ push rbx ; inptr1
+ push rsi ; inptr00
+ mov rbx, rsp
+
+ push rdi
+ push rcx
+ push rax
+
+ %ifdef WIN64
+ mov r8, rcx
+ mov r9, rdi
+ mov rcx, rax
+ mov rdx, rbx
+ %else
+ mov rdx, rcx
+ mov rcx, rdi
+ mov rdi, rax
+ mov rsi, rbx
+ %endif
+
+ call EXTN(jsimd_h2v1_merged_upsample_avx2)
+
+ pop rax
+ pop rcx
+ pop rdi
+ pop rsi
+ pop rbx
+ pop rdx
+
+ add rdi, byte SIZEOF_JSAMPROW ; outptr1
+ add rsi, byte SIZEOF_JSAMPROW ; inptr01
+
+ push rdx ; inptr2
+ push rbx ; inptr1
+ push rsi ; inptr00
+ mov rbx, rsp
+
+ push rdi
+ push rcx
+ push rax
+
+ %ifdef WIN64
+ mov r8, rcx
+ mov r9, rdi
+ mov rcx, rax
+ mov rdx, rbx
+ %else
+ mov rdx, rcx
+ mov rcx, rdi
+ mov rdi, rax
+ mov rsi, rbx
+ %endif
+
+ call EXTN(jsimd_h2v1_merged_upsample_avx2)
+
+ pop rax
+ pop rcx
+ pop rdi
+ pop rsi
+ pop rbx
+ pop rdx
+
+ pop rbx
+ uncollect_args 4
+ pop rbp
+ ret
+
+; For some reason, the OS X linker does not honor the request to align the
+; segment unless we do this.
+ align 32
diff --git a/simd/x86_64/jdmrgext-sse2.asm b/simd/x86_64/jdmrgext-sse2.asm
new file mode 100644
index 0000000..4c2bb04
--- /dev/null
+++ b/simd/x86_64/jdmrgext-sse2.asm
@@ -0,0 +1,537 @@
+;
+; jdmrgext.asm - merged upsampling/color conversion (64-bit SSE2)
+;
+; Copyright 2009, 2012 Pierre Ossman <ossman@cendio.se> for Cendio AB
+; Copyright (C) 2009, 2012, 2016, D. R. Commander.
+;
+; Based on the x86 SIMD extension for IJG JPEG library
+; Copyright (C) 1999-2006, MIYASAKA Masaru.
+; For conditions of distribution and use, see copyright notice in jsimdext.inc
+;
+; This file should be assembled with NASM (Netwide Assembler),
+; can *not* be assembled with Microsoft's MASM or any compatible
+; assembler (including Borland's Turbo Assembler).
+; NASM is available from http://nasm.sourceforge.net/ or
+; http://sourceforge.net/project/showfiles.php?group_id=6208
+;
+; [TAB8]
+
+%include "jcolsamp.inc"
+
+; --------------------------------------------------------------------------
+;
+; Upsample and color convert for the case of 2:1 horizontal and 1:1 vertical.
+;
+; GLOBAL(void)
+; jsimd_h2v1_merged_upsample_sse2 (JDIMENSION output_width,
+; JSAMPIMAGE input_buf,
+; JDIMENSION in_row_group_ctr,
+; JSAMPARRAY output_buf);
+;
+
+; r10d = JDIMENSION output_width
+; r11 = JSAMPIMAGE input_buf
+; r12d = JDIMENSION in_row_group_ctr
+; r13 = JSAMPARRAY output_buf
+
+%define wk(i) rbp-(WK_NUM-(i))*SIZEOF_XMMWORD ; xmmword wk[WK_NUM]
+%define WK_NUM 3
+
+ align 32
+ global EXTN(jsimd_h2v1_merged_upsample_sse2)
+
+EXTN(jsimd_h2v1_merged_upsample_sse2):
+ push rbp
+ mov rax, rsp ; rax = original rbp
+ sub rsp, byte 4
+ and rsp, byte (-SIZEOF_XMMWORD) ; align to 128 bits
+ mov [rsp], rax
+ mov rbp, rsp ; rbp = aligned rbp
+ lea rsp, [wk(0)]
+ collect_args 4
+ push rbx
+
+ mov ecx, r10d ; col
+ test rcx, rcx
+ jz near .return
+
+ push rcx
+
+ mov rdi, r11
+ mov ecx, r12d
+ mov rsi, JSAMPARRAY [rdi+0*SIZEOF_JSAMPARRAY]
+ mov rbx, JSAMPARRAY [rdi+1*SIZEOF_JSAMPARRAY]
+ mov rdx, JSAMPARRAY [rdi+2*SIZEOF_JSAMPARRAY]
+ mov rdi, r13
+ mov rsi, JSAMPROW [rsi+rcx*SIZEOF_JSAMPROW] ; inptr0
+ mov rbx, JSAMPROW [rbx+rcx*SIZEOF_JSAMPROW] ; inptr1
+ mov rdx, JSAMPROW [rdx+rcx*SIZEOF_JSAMPROW] ; inptr2
+ mov rdi, JSAMPROW [rdi] ; outptr
+
+ pop rcx ; col
+
+.columnloop:
+
+ movdqa xmm6, XMMWORD [rbx] ; xmm6=Cb(0123456789ABCDEF)
+ movdqa xmm7, XMMWORD [rdx] ; xmm7=Cr(0123456789ABCDEF)
+
+ pxor xmm1, xmm1 ; xmm1=(all 0's)
+ pcmpeqw xmm3, xmm3
+ psllw xmm3, 7 ; xmm3={0xFF80 0xFF80 0xFF80 0xFF80 ..}
+
+ movdqa xmm4, xmm6
+ punpckhbw xmm6, xmm1 ; xmm6=Cb(89ABCDEF)=CbH
+ punpcklbw xmm4, xmm1 ; xmm4=Cb(01234567)=CbL
+ movdqa xmm0, xmm7
+ punpckhbw xmm7, xmm1 ; xmm7=Cr(89ABCDEF)=CrH
+ punpcklbw xmm0, xmm1 ; xmm0=Cr(01234567)=CrL
+
+ paddw xmm6, xmm3
+ paddw xmm4, xmm3
+ paddw xmm7, xmm3
+ paddw xmm0, xmm3
+
+ ; (Original)
+ ; R = Y + 1.40200 * Cr
+ ; G = Y - 0.34414 * Cb - 0.71414 * Cr
+ ; B = Y + 1.77200 * Cb
+ ;
+ ; (This implementation)
+ ; R = Y + 0.40200 * Cr + Cr
+ ; G = Y - 0.34414 * Cb + 0.28586 * Cr - Cr
+ ; B = Y - 0.22800 * Cb + Cb + Cb
+
+ movdqa xmm5, xmm6 ; xmm5=CbH
+ movdqa xmm2, xmm4 ; xmm2=CbL
+ paddw xmm6, xmm6 ; xmm6=2*CbH
+ paddw xmm4, xmm4 ; xmm4=2*CbL
+ movdqa xmm1, xmm7 ; xmm1=CrH
+ movdqa xmm3, xmm0 ; xmm3=CrL
+ paddw xmm7, xmm7 ; xmm7=2*CrH
+ paddw xmm0, xmm0 ; xmm0=2*CrL
+
+ pmulhw xmm6, [rel PW_MF0228] ; xmm6=(2*CbH * -FIX(0.22800))
+ pmulhw xmm4, [rel PW_MF0228] ; xmm4=(2*CbL * -FIX(0.22800))
+ pmulhw xmm7, [rel PW_F0402] ; xmm7=(2*CrH * FIX(0.40200))
+ pmulhw xmm0, [rel PW_F0402] ; xmm0=(2*CrL * FIX(0.40200))
+
+ paddw xmm6, [rel PW_ONE]
+ paddw xmm4, [rel PW_ONE]
+ psraw xmm6, 1 ; xmm6=(CbH * -FIX(0.22800))
+ psraw xmm4, 1 ; xmm4=(CbL * -FIX(0.22800))
+ paddw xmm7, [rel PW_ONE]
+ paddw xmm0, [rel PW_ONE]
+ psraw xmm7, 1 ; xmm7=(CrH * FIX(0.40200))
+ psraw xmm0, 1 ; xmm0=(CrL * FIX(0.40200))
+
+ paddw xmm6, xmm5
+ paddw xmm4, xmm2
+ paddw xmm6, xmm5 ; xmm6=(CbH * FIX(1.77200))=(B-Y)H
+ paddw xmm4, xmm2 ; xmm4=(CbL * FIX(1.77200))=(B-Y)L
+ paddw xmm7, xmm1 ; xmm7=(CrH * FIX(1.40200))=(R-Y)H
+ paddw xmm0, xmm3 ; xmm0=(CrL * FIX(1.40200))=(R-Y)L
+
+ movdqa XMMWORD [wk(0)], xmm6 ; wk(0)=(B-Y)H
+ movdqa XMMWORD [wk(1)], xmm7 ; wk(1)=(R-Y)H
+
+ movdqa xmm6, xmm5
+ movdqa xmm7, xmm2
+ punpcklwd xmm5, xmm1
+ punpckhwd xmm6, xmm1
+ pmaddwd xmm5, [rel PW_MF0344_F0285]
+ pmaddwd xmm6, [rel PW_MF0344_F0285]
+ punpcklwd xmm2, xmm3
+ punpckhwd xmm7, xmm3
+ pmaddwd xmm2, [rel PW_MF0344_F0285]
+ pmaddwd xmm7, [rel PW_MF0344_F0285]
+
+ paddd xmm5, [rel PD_ONEHALF]
+ paddd xmm6, [rel PD_ONEHALF]
+ psrad xmm5, SCALEBITS
+ psrad xmm6, SCALEBITS
+ paddd xmm2, [rel PD_ONEHALF]
+ paddd xmm7, [rel PD_ONEHALF]
+ psrad xmm2, SCALEBITS
+ psrad xmm7, SCALEBITS
+
+ packssdw xmm5, xmm6 ; xmm5=CbH*-FIX(0.344)+CrH*FIX(0.285)
+ packssdw xmm2, xmm7 ; xmm2=CbL*-FIX(0.344)+CrL*FIX(0.285)
+ psubw xmm5, xmm1 ; xmm5=CbH*-FIX(0.344)+CrH*-FIX(0.714)=(G-Y)H
+ psubw xmm2, xmm3 ; xmm2=CbL*-FIX(0.344)+CrL*-FIX(0.714)=(G-Y)L
+
+ movdqa XMMWORD [wk(2)], xmm5 ; wk(2)=(G-Y)H
+
+ mov al, 2 ; Yctr
+ jmp short .Yloop_1st
+
+.Yloop_2nd:
+ movdqa xmm0, XMMWORD [wk(1)] ; xmm0=(R-Y)H
+ movdqa xmm2, XMMWORD [wk(2)] ; xmm2=(G-Y)H
+ movdqa xmm4, XMMWORD [wk(0)] ; xmm4=(B-Y)H
+
+.Yloop_1st:
+ movdqa xmm7, XMMWORD [rsi] ; xmm7=Y(0123456789ABCDEF)
+
+ pcmpeqw xmm6, xmm6
+ psrlw xmm6, BYTE_BIT ; xmm6={0xFF 0x00 0xFF 0x00 ..}
+ pand xmm6, xmm7 ; xmm6=Y(02468ACE)=YE
+ psrlw xmm7, BYTE_BIT ; xmm7=Y(13579BDF)=YO
+
+ movdqa xmm1, xmm0 ; xmm1=xmm0=(R-Y)(L/H)
+ movdqa xmm3, xmm2 ; xmm3=xmm2=(G-Y)(L/H)
+ movdqa xmm5, xmm4 ; xmm5=xmm4=(B-Y)(L/H)
+
+ paddw xmm0, xmm6 ; xmm0=((R-Y)+YE)=RE=R(02468ACE)
+ paddw xmm1, xmm7 ; xmm1=((R-Y)+YO)=RO=R(13579BDF)
+ packuswb xmm0, xmm0 ; xmm0=R(02468ACE********)
+ packuswb xmm1, xmm1 ; xmm1=R(13579BDF********)
+
+ paddw xmm2, xmm6 ; xmm2=((G-Y)+YE)=GE=G(02468ACE)
+ paddw xmm3, xmm7 ; xmm3=((G-Y)+YO)=GO=G(13579BDF)
+ packuswb xmm2, xmm2 ; xmm2=G(02468ACE********)
+ packuswb xmm3, xmm3 ; xmm3=G(13579BDF********)
+
+ paddw xmm4, xmm6 ; xmm4=((B-Y)+YE)=BE=B(02468ACE)
+ paddw xmm5, xmm7 ; xmm5=((B-Y)+YO)=BO=B(13579BDF)
+ packuswb xmm4, xmm4 ; xmm4=B(02468ACE********)
+ packuswb xmm5, xmm5 ; xmm5=B(13579BDF********)
+
+%if RGB_PIXELSIZE == 3 ; ---------------
+
+ ; xmmA=(00 02 04 06 08 0A 0C 0E **), xmmB=(01 03 05 07 09 0B 0D 0F **)
+ ; xmmC=(10 12 14 16 18 1A 1C 1E **), xmmD=(11 13 15 17 19 1B 1D 1F **)
+ ; xmmE=(20 22 24 26 28 2A 2C 2E **), xmmF=(21 23 25 27 29 2B 2D 2F **)
+ ; xmmG=(** ** ** ** ** ** ** ** **), xmmH=(** ** ** ** ** ** ** ** **)
+
+ punpcklbw xmmA, xmmC ; xmmA=(00 10 02 12 04 14 06 16 08 18 0A 1A 0C 1C 0E 1E)
+ punpcklbw xmmE, xmmB ; xmmE=(20 01 22 03 24 05 26 07 28 09 2A 0B 2C 0D 2E 0F)
+ punpcklbw xmmD, xmmF ; xmmD=(11 21 13 23 15 25 17 27 19 29 1B 2B 1D 2D 1F 2F)
+
+ movdqa xmmG, xmmA
+ movdqa xmmH, xmmA
+ punpcklwd xmmA, xmmE ; xmmA=(00 10 20 01 02 12 22 03 04 14 24 05 06 16 26 07)
+ punpckhwd xmmG, xmmE ; xmmG=(08 18 28 09 0A 1A 2A 0B 0C 1C 2C 0D 0E 1E 2E 0F)
+
+ psrldq xmmH, 2 ; xmmH=(02 12 04 14 06 16 08 18 0A 1A 0C 1C 0E 1E -- --)
+ psrldq xmmE, 2 ; xmmE=(22 03 24 05 26 07 28 09 2A 0B 2C 0D 2E 0F -- --)
+
+ movdqa xmmC, xmmD
+ movdqa xmmB, xmmD
+ punpcklwd xmmD, xmmH ; xmmD=(11 21 02 12 13 23 04 14 15 25 06 16 17 27 08 18)
+ punpckhwd xmmC, xmmH ; xmmC=(19 29 0A 1A 1B 2B 0C 1C 1D 2D 0E 1E 1F 2F -- --)
+
+ psrldq xmmB, 2 ; xmmB=(13 23 15 25 17 27 19 29 1B 2B 1D 2D 1F 2F -- --)
+
+ movdqa xmmF, xmmE
+ punpcklwd xmmE, xmmB ; xmmE=(22 03 13 23 24 05 15 25 26 07 17 27 28 09 19 29)
+ punpckhwd xmmF, xmmB ; xmmF=(2A 0B 1B 2B 2C 0D 1D 2D 2E 0F 1F 2F -- -- -- --)
+
+ pshufd xmmH, xmmA, 0x4E ; xmmH=(04 14 24 05 06 16 26 07 00 10 20 01 02 12 22 03)
+ movdqa xmmB, xmmE
+ punpckldq xmmA, xmmD ; xmmA=(00 10 20 01 11 21 02 12 02 12 22 03 13 23 04 14)
+ punpckldq xmmE, xmmH ; xmmE=(22 03 13 23 04 14 24 05 24 05 15 25 06 16 26 07)
+ punpckhdq xmmD, xmmB ; xmmD=(15 25 06 16 26 07 17 27 17 27 08 18 28 09 19 29)
+
+ pshufd xmmH, xmmG, 0x4E ; xmmH=(0C 1C 2C 0D 0E 1E 2E 0F 08 18 28 09 0A 1A 2A 0B)
+ movdqa xmmB, xmmF
+ punpckldq xmmG, xmmC ; xmmG=(08 18 28 09 19 29 0A 1A 0A 1A 2A 0B 1B 2B 0C 1C)
+ punpckldq xmmF, xmmH ; xmmF=(2A 0B 1B 2B 0C 1C 2C 0D 2C 0D 1D 2D 0E 1E 2E 0F)
+ punpckhdq xmmC, xmmB ; xmmC=(1D 2D 0E 1E 2E 0F 1F 2F 1F 2F -- -- -- -- -- --)
+
+ punpcklqdq xmmA, xmmE ; xmmA=(00 10 20 01 11 21 02 12 22 03 13 23 04 14 24 05)
+ punpcklqdq xmmD, xmmG ; xmmD=(15 25 06 16 26 07 17 27 08 18 28 09 19 29 0A 1A)
+ punpcklqdq xmmF, xmmC ; xmmF=(2A 0B 1B 2B 0C 1C 2C 0D 1D 2D 0E 1E 2E 0F 1F 2F)
+
+ cmp rcx, byte SIZEOF_XMMWORD
+ jb short .column_st32
+
+ test rdi, SIZEOF_XMMWORD-1
+ jnz short .out1
+ ; --(aligned)-------------------
+ movntdq XMMWORD [rdi+0*SIZEOF_XMMWORD], xmmA
+ movntdq XMMWORD [rdi+1*SIZEOF_XMMWORD], xmmD
+ movntdq XMMWORD [rdi+2*SIZEOF_XMMWORD], xmmF
+ jmp short .out0
+.out1: ; --(unaligned)-----------------
+ movdqu XMMWORD [rdi+0*SIZEOF_XMMWORD], xmmA
+ movdqu XMMWORD [rdi+1*SIZEOF_XMMWORD], xmmD
+ movdqu XMMWORD [rdi+2*SIZEOF_XMMWORD], xmmF
+.out0:
+ add rdi, byte RGB_PIXELSIZE*SIZEOF_XMMWORD ; outptr
+ sub rcx, byte SIZEOF_XMMWORD
+ jz near .endcolumn
+
+ add rsi, byte SIZEOF_XMMWORD ; inptr0
+ dec al ; Yctr
+ jnz near .Yloop_2nd
+
+ add rbx, byte SIZEOF_XMMWORD ; inptr1
+ add rdx, byte SIZEOF_XMMWORD ; inptr2
+ jmp near .columnloop
+
+.column_st32:
+ lea rcx, [rcx+rcx*2] ; imul ecx, RGB_PIXELSIZE
+ cmp rcx, byte 2*SIZEOF_XMMWORD
+ jb short .column_st16
+ movdqu XMMWORD [rdi+0*SIZEOF_XMMWORD], xmmA
+ movdqu XMMWORD [rdi+1*SIZEOF_XMMWORD], xmmD
+ add rdi, byte 2*SIZEOF_XMMWORD ; outptr
+ movdqa xmmA, xmmF
+ sub rcx, byte 2*SIZEOF_XMMWORD
+ jmp short .column_st15
+.column_st16:
+ cmp rcx, byte SIZEOF_XMMWORD
+ jb short .column_st15
+ movdqu XMMWORD [rdi+0*SIZEOF_XMMWORD], xmmA
+ add rdi, byte SIZEOF_XMMWORD ; outptr
+ movdqa xmmA, xmmD
+ sub rcx, byte SIZEOF_XMMWORD
+.column_st15:
+ ; Store the lower 8 bytes of xmmA to the output when it has enough
+ ; space.
+ cmp rcx, byte SIZEOF_MMWORD
+ jb short .column_st7
+ movq XMM_MMWORD [rdi], xmmA
+ add rdi, byte SIZEOF_MMWORD
+ sub rcx, byte SIZEOF_MMWORD
+ psrldq xmmA, SIZEOF_MMWORD
+.column_st7:
+ ; Store the lower 4 bytes of xmmA to the output when it has enough
+ ; space.
+ cmp rcx, byte SIZEOF_DWORD
+ jb short .column_st3
+ movd XMM_DWORD [rdi], xmmA
+ add rdi, byte SIZEOF_DWORD
+ sub rcx, byte SIZEOF_DWORD
+ psrldq xmmA, SIZEOF_DWORD
+.column_st3:
+ ; Store the lower 2 bytes of rax to the output when it has enough
+ ; space.
+ movd eax, xmmA
+ cmp rcx, byte SIZEOF_WORD
+ jb short .column_st1
+ mov WORD [rdi], ax
+ add rdi, byte SIZEOF_WORD
+ sub rcx, byte SIZEOF_WORD
+ shr rax, 16
+.column_st1:
+ ; Store the lower 1 byte of rax to the output when it has enough
+ ; space.
+ test rcx, rcx
+ jz short .endcolumn
+ mov BYTE [rdi], al
+
+%else ; RGB_PIXELSIZE == 4 ; -----------
+
+%ifdef RGBX_FILLER_0XFF
+ pcmpeqb xmm6, xmm6 ; xmm6=XE=X(02468ACE********)
+ pcmpeqb xmm7, xmm7 ; xmm7=XO=X(13579BDF********)
+%else
+ pxor xmm6, xmm6 ; xmm6=XE=X(02468ACE********)
+ pxor xmm7, xmm7 ; xmm7=XO=X(13579BDF********)
+%endif
+ ; xmmA=(00 02 04 06 08 0A 0C 0E **), xmmB=(01 03 05 07 09 0B 0D 0F **)
+ ; xmmC=(10 12 14 16 18 1A 1C 1E **), xmmD=(11 13 15 17 19 1B 1D 1F **)
+ ; xmmE=(20 22 24 26 28 2A 2C 2E **), xmmF=(21 23 25 27 29 2B 2D 2F **)
+ ; xmmG=(30 32 34 36 38 3A 3C 3E **), xmmH=(31 33 35 37 39 3B 3D 3F **)
+
+ punpcklbw xmmA, xmmC ; xmmA=(00 10 02 12 04 14 06 16 08 18 0A 1A 0C 1C 0E 1E)
+ punpcklbw xmmE, xmmG ; xmmE=(20 30 22 32 24 34 26 36 28 38 2A 3A 2C 3C 2E 3E)
+ punpcklbw xmmB, xmmD ; xmmB=(01 11 03 13 05 15 07 17 09 19 0B 1B 0D 1D 0F 1F)
+ punpcklbw xmmF, xmmH ; xmmF=(21 31 23 33 25 35 27 37 29 39 2B 3B 2D 3D 2F 3F)
+
+ movdqa xmmC, xmmA
+ punpcklwd xmmA, xmmE ; xmmA=(00 10 20 30 02 12 22 32 04 14 24 34 06 16 26 36)
+ punpckhwd xmmC, xmmE ; xmmC=(08 18 28 38 0A 1A 2A 3A 0C 1C 2C 3C 0E 1E 2E 3E)
+ movdqa xmmG, xmmB
+ punpcklwd xmmB, xmmF ; xmmB=(01 11 21 31 03 13 23 33 05 15 25 35 07 17 27 37)
+ punpckhwd xmmG, xmmF ; xmmG=(09 19 29 39 0B 1B 2B 3B 0D 1D 2D 3D 0F 1F 2F 3F)
+
+ movdqa xmmD, xmmA
+ punpckldq xmmA, xmmB ; xmmA=(00 10 20 30 01 11 21 31 02 12 22 32 03 13 23 33)
+ punpckhdq xmmD, xmmB ; xmmD=(04 14 24 34 05 15 25 35 06 16 26 36 07 17 27 37)
+ movdqa xmmH, xmmC
+ punpckldq xmmC, xmmG ; xmmC=(08 18 28 38 09 19 29 39 0A 1A 2A 3A 0B 1B 2B 3B)
+ punpckhdq xmmH, xmmG ; xmmH=(0C 1C 2C 3C 0D 1D 2D 3D 0E 1E 2E 3E 0F 1F 2F 3F)
+
+ cmp rcx, byte SIZEOF_XMMWORD
+ jb short .column_st32
+
+ test rdi, SIZEOF_XMMWORD-1
+ jnz short .out1
+ ; --(aligned)-------------------
+ movntdq XMMWORD [rdi+0*SIZEOF_XMMWORD], xmmA
+ movntdq XMMWORD [rdi+1*SIZEOF_XMMWORD], xmmD
+ movntdq XMMWORD [rdi+2*SIZEOF_XMMWORD], xmmC
+ movntdq XMMWORD [rdi+3*SIZEOF_XMMWORD], xmmH
+ jmp short .out0
+.out1: ; --(unaligned)-----------------
+ movdqu XMMWORD [rdi+0*SIZEOF_XMMWORD], xmmA
+ movdqu XMMWORD [rdi+1*SIZEOF_XMMWORD], xmmD
+ movdqu XMMWORD [rdi+2*SIZEOF_XMMWORD], xmmC
+ movdqu XMMWORD [rdi+3*SIZEOF_XMMWORD], xmmH
+.out0:
+ add rdi, byte RGB_PIXELSIZE*SIZEOF_XMMWORD ; outptr
+ sub rcx, byte SIZEOF_XMMWORD
+ jz near .endcolumn
+
+ add rsi, byte SIZEOF_XMMWORD ; inptr0
+ dec al ; Yctr
+ jnz near .Yloop_2nd
+
+ add rbx, byte SIZEOF_XMMWORD ; inptr1
+ add rdx, byte SIZEOF_XMMWORD ; inptr2
+ jmp near .columnloop
+
+.column_st32:
+ cmp rcx, byte SIZEOF_XMMWORD/2
+ jb short .column_st16
+ movdqu XMMWORD [rdi+0*SIZEOF_XMMWORD], xmmA
+ movdqu XMMWORD [rdi+1*SIZEOF_XMMWORD], xmmD
+ add rdi, byte 2*SIZEOF_XMMWORD ; outptr
+ movdqa xmmA, xmmC
+ movdqa xmmD, xmmH
+ sub rcx, byte SIZEOF_XMMWORD/2
+.column_st16:
+ cmp rcx, byte SIZEOF_XMMWORD/4
+ jb short .column_st15
+ movdqu XMMWORD [rdi+0*SIZEOF_XMMWORD], xmmA
+ add rdi, byte SIZEOF_XMMWORD ; outptr
+ movdqa xmmA, xmmD
+ sub rcx, byte SIZEOF_XMMWORD/4
+.column_st15:
+ ; Store two pixels (8 bytes) of xmmA to the output when it has enough
+ ; space.
+ cmp rcx, byte SIZEOF_XMMWORD/8
+ jb short .column_st7
+ movq XMM_MMWORD [rdi], xmmA
+ add rdi, byte SIZEOF_XMMWORD/8*4
+ sub rcx, byte SIZEOF_XMMWORD/8
+ psrldq xmmA, SIZEOF_XMMWORD/8*4
+.column_st7:
+ ; Store one pixel (4 bytes) of xmmA to the output when it has enough
+ ; space.
+ test rcx, rcx
+ jz short .endcolumn
+ movd XMM_DWORD [rdi], xmmA
+
+%endif ; RGB_PIXELSIZE ; ---------------
+
+.endcolumn:
+ sfence ; flush the write buffer
+
+.return:
+ pop rbx
+ uncollect_args 4
+ mov rsp, rbp ; rsp <- aligned rbp
+ pop rsp ; rsp <- original rbp
+ pop rbp
+ ret
+
+; --------------------------------------------------------------------------
+;
+; Upsample and color convert for the case of 2:1 horizontal and 2:1 vertical.
+;
+; GLOBAL(void)
+; jsimd_h2v2_merged_upsample_sse2 (JDIMENSION output_width,
+; JSAMPIMAGE input_buf,
+; JDIMENSION in_row_group_ctr,
+; JSAMPARRAY output_buf);
+;
+
+; r10d = JDIMENSION output_width
+; r11 = JSAMPIMAGE input_buf
+; r12d = JDIMENSION in_row_group_ctr
+; r13 = JSAMPARRAY output_buf
+
+ align 32
+ global EXTN(jsimd_h2v2_merged_upsample_sse2)
+
+EXTN(jsimd_h2v2_merged_upsample_sse2):
+ push rbp
+ mov rax, rsp
+ mov rbp, rsp
+ collect_args 4
+ push rbx
+
+ mov eax, r10d
+
+ mov rdi, r11
+ mov ecx, r12d
+ mov rsi, JSAMPARRAY [rdi+0*SIZEOF_JSAMPARRAY]
+ mov rbx, JSAMPARRAY [rdi+1*SIZEOF_JSAMPARRAY]
+ mov rdx, JSAMPARRAY [rdi+2*SIZEOF_JSAMPARRAY]
+ mov rdi, r13
+ lea rsi, [rsi+rcx*SIZEOF_JSAMPROW]
+
+ push rdx ; inptr2
+ push rbx ; inptr1
+ push rsi ; inptr00
+ mov rbx, rsp
+
+ push rdi
+ push rcx
+ push rax
+
+ %ifdef WIN64
+ mov r8, rcx
+ mov r9, rdi
+ mov rcx, rax
+ mov rdx, rbx
+ %else
+ mov rdx, rcx
+ mov rcx, rdi
+ mov rdi, rax
+ mov rsi, rbx
+ %endif
+
+ call EXTN(jsimd_h2v1_merged_upsample_sse2)
+
+ pop rax
+ pop rcx
+ pop rdi
+ pop rsi
+ pop rbx
+ pop rdx
+
+ add rdi, byte SIZEOF_JSAMPROW ; outptr1
+ add rsi, byte SIZEOF_JSAMPROW ; inptr01
+
+ push rdx ; inptr2
+ push rbx ; inptr1
+ push rsi ; inptr00
+ mov rbx, rsp
+
+ push rdi
+ push rcx
+ push rax
+
+ %ifdef WIN64
+ mov r8, rcx
+ mov r9, rdi
+ mov rcx, rax
+ mov rdx, rbx
+ %else
+ mov rdx, rcx
+ mov rcx, rdi
+ mov rdi, rax
+ mov rsi, rbx
+ %endif
+
+ call EXTN(jsimd_h2v1_merged_upsample_sse2)
+
+ pop rax
+ pop rcx
+ pop rdi
+ pop rsi
+ pop rbx
+ pop rdx
+
+ pop rbx
+ uncollect_args 4
+ pop rbp
+ ret
+
+; For some reason, the OS X linker does not honor the request to align the
+; segment unless we do this.
+ align 32
diff --git a/simd/x86_64/jdsample-avx2.asm b/simd/x86_64/jdsample-avx2.asm
new file mode 100644
index 0000000..f3384ca
--- /dev/null
+++ b/simd/x86_64/jdsample-avx2.asm
@@ -0,0 +1,701 @@
+;
+; jdsample.asm - upsampling (64-bit AVX2)
+;
+; Copyright 2009 Pierre Ossman <ossman@cendio.se> for Cendio AB
+; Copyright (C) 2009, 2016, D. R. Commander.
+; Copyright (C) 2015, Intel Corporation.
+;
+; Based on the x86 SIMD extension for IJG JPEG library
+; Copyright (C) 1999-2006, MIYASAKA Masaru.
+; For conditions of distribution and use, see copyright notice in jsimdext.inc
+;
+; This file should be assembled with NASM (Netwide Assembler),
+; can *not* be assembled with Microsoft's MASM or any compatible
+; assembler (including Borland's Turbo Assembler).
+; NASM is available from http://nasm.sourceforge.net/ or
+; http://sourceforge.net/project/showfiles.php?group_id=6208
+;
+; [TAB8]
+
+%include "jsimdext.inc"
+
+; --------------------------------------------------------------------------
+ SECTION SEG_CONST
+
+ alignz 32
+ global EXTN(jconst_fancy_upsample_avx2)
+
+EXTN(jconst_fancy_upsample_avx2):
+
+PW_ONE times 16 dw 1
+PW_TWO times 16 dw 2
+PW_THREE times 16 dw 3
+PW_SEVEN times 16 dw 7
+PW_EIGHT times 16 dw 8
+
+ alignz 32
+
+; --------------------------------------------------------------------------
+ SECTION SEG_TEXT
+ BITS 64
+;
+; Fancy processing for the common case of 2:1 horizontal and 1:1 vertical.
+;
+; The upsampling algorithm is linear interpolation between pixel centers,
+; also known as a "triangle filter". This is a good compromise between
+; speed and visual quality. The centers of the output pixels are 1/4 and 3/4
+; of the way between input pixel centers.
+;
+; GLOBAL(void)
+; jsimd_h2v1_fancy_upsample_avx2 (int max_v_samp_factor,
+; JDIMENSION downsampled_width,
+; JSAMPARRAY input_data,
+; JSAMPARRAY *output_data_ptr);
+;
+
+; r10 = int max_v_samp_factor
+; r11d = JDIMENSION downsampled_width
+; r12 = JSAMPARRAY input_data
+; r13 = JSAMPARRAY *output_data_ptr
+
+ align 32
+ global EXTN(jsimd_h2v1_fancy_upsample_avx2)
+
+EXTN(jsimd_h2v1_fancy_upsample_avx2):
+ push rbp
+ mov rax, rsp
+ mov rbp, rsp
+ push_xmm 3
+ collect_args 4
+
+ mov eax, r11d ; colctr
+ test rax, rax
+ jz near .return
+
+ mov rcx, r10 ; rowctr
+ test rcx, rcx
+ jz near .return
+
+ mov rsi, r12 ; input_data
+ mov rdi, r13
+ mov rdi, JSAMPARRAY [rdi] ; output_data
+
+ vpxor ymm0, ymm0, ymm0 ; ymm0=(all 0's)
+ vpcmpeqb xmm9, xmm9, xmm9
+ vpsrldq xmm10, xmm9, (SIZEOF_XMMWORD-1) ; (ff -- -- -- ... -- --) LSB is ff
+
+ vpslldq xmm9, xmm9, (SIZEOF_XMMWORD-1)
+ vperm2i128 ymm9, ymm9, ymm9, 1 ; (---- ---- ... ---- ---- ff) MSB is ff
+
+.rowloop:
+ push rax ; colctr
+ push rdi
+ push rsi
+
+ mov rsi, JSAMPROW [rsi] ; inptr
+ mov rdi, JSAMPROW [rdi] ; outptr
+
+ test rax, SIZEOF_YMMWORD-1
+ jz short .skip
+ mov dl, JSAMPLE [rsi+(rax-1)*SIZEOF_JSAMPLE]
+ mov JSAMPLE [rsi+rax*SIZEOF_JSAMPLE], dl ; insert a dummy sample
+.skip:
+ vpand ymm7, ymm10, YMMWORD [rsi+0*SIZEOF_YMMWORD]
+
+ add rax, byte SIZEOF_YMMWORD-1
+ and rax, byte -SIZEOF_YMMWORD
+ cmp rax, byte SIZEOF_YMMWORD
+ ja short .columnloop
+
+.columnloop_last:
+ vpand ymm6, ymm9, YMMWORD [rsi+0*SIZEOF_YMMWORD]
+ jmp short .upsample
+
+.columnloop:
+ vmovdqu ymm6, YMMWORD [rsi+1*SIZEOF_YMMWORD]
+ vperm2i128 ymm6, ymm0, ymm6, 0x20
+ vpslldq ymm6, ymm6, 15
+
+.upsample:
+ vmovdqu ymm1, YMMWORD [rsi+0*SIZEOF_YMMWORD] ; ymm1=( 0 1 2 ... 29 30 31)
+
+ vperm2i128 ymm2, ymm0, ymm1, 0x20
+ vpalignr ymm2, ymm1, ymm2, 15 ; ymm2=(-- 0 1 ... 28 29 30)
+ vperm2i128 ymm4, ymm0, ymm1, 0x03
+ vpalignr ymm3, ymm4, ymm1, 1 ; ymm3=( 1 2 3 ... 30 31 --)
+
+ vpor ymm2, ymm2, ymm7 ; ymm2=(-1 0 1 ... 28 29 30)
+ vpor ymm3, ymm3, ymm6 ; ymm3=( 1 2 3 ... 30 31 32)
+
+ vpsrldq ymm7, ymm4, (SIZEOF_XMMWORD-1) ; ymm7=(31 -- -- ... -- -- --)
+
+ vpunpckhbw ymm4, ymm1, ymm0 ; ymm4=( 8 9 10 11 12 13 14 15 24 25 26 27 28 29 30 31)
+ vpunpcklbw ymm5, ymm1, ymm0 ; ymm5=( 0 1 2 3 4 5 6 7 16 17 18 19 20 21 22 23)
+ vperm2i128 ymm1, ymm5, ymm4, 0x20 ; ymm1=( 0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15)
+ vperm2i128 ymm4, ymm5, ymm4, 0x31 ; ymm4=(16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31)
+
+ vpunpckhbw ymm5, ymm2, ymm0 ; ymm5=( 7 8 9 10 11 12 13 14 23 24 25 26 27 28 29 30)
+ vpunpcklbw ymm6, ymm2, ymm0 ; ymm6=(-1 0 1 2 3 4 5 6 15 16 17 18 19 20 21 22)
+ vperm2i128 ymm2, ymm6, ymm5, 0x20 ; ymm2=(-1 0 1 2 3 4 5 6 7 8 9 10 11 12 13 14)
+ vperm2i128 ymm5, ymm6, ymm5, 0x31 ; ymm5=(15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30)
+
+ vpunpckhbw ymm6, ymm3, ymm0 ; ymm6=( 1 2 3 4 5 6 7 8 17 18 19 20 21 22 23 24)
+ vpunpcklbw ymm8, ymm3, ymm0 ; ymm8=( 9 10 11 12 13 14 15 16 25 26 27 28 29 30 31 32)
+ vperm2i128 ymm3, ymm8, ymm6, 0x20 ; ymm3=( 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16)
+ vperm2i128 ymm6, ymm8, ymm6, 0x31 ; ymm6=(17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32)
+
+ vpmullw ymm1, ymm1, [rel PW_THREE]
+ vpmullw ymm4, ymm4, [rel PW_THREE]
+ vpaddw ymm2, ymm2, [rel PW_ONE]
+ vpaddw ymm5, ymm5, [rel PW_ONE]
+ vpaddw ymm3, ymm3, [rel PW_TWO]
+ vpaddw ymm6, ymm6, [rel PW_TWO]
+
+ vpaddw ymm2, ymm2, ymm1
+ vpaddw ymm5, ymm5, ymm4
+ vpsrlw ymm2, ymm2, 2 ; ymm2=OutLE=( 0 2 4 6 8 10 12 14 16 18 20 22 24 26 28 30)
+ vpsrlw ymm5, ymm5, 2 ; ymm5=OutHE=(32 34 36 38 40 42 44 46 48 50 52 54 56 58 60 62)
+ vpaddw ymm3, ymm3, ymm1
+ vpaddw ymm6, ymm6, ymm4
+ vpsrlw ymm3, ymm3, 2 ; ymm3=OutLO=( 1 3 5 7 9 11 13 15 17 19 21 23 25 27 29 31)
+ vpsrlw ymm6, ymm6, 2 ; ymm6=OutHO=(33 35 37 39 41 43 45 47 49 51 53 55 57 59 61 63)
+
+ vpsllw ymm3, ymm3, BYTE_BIT
+ vpsllw ymm6, ymm6, BYTE_BIT
+ vpor ymm2, ymm2, ymm3 ; ymm2=OutL=( 0 1 2 ... 29 30 31)
+ vpor ymm5, ymm5, ymm6 ; ymm5=OutH=(32 33 34 ... 61 62 63)
+
+ vmovdqu YMMWORD [rdi+0*SIZEOF_YMMWORD], ymm2
+ vmovdqu YMMWORD [rdi+1*SIZEOF_YMMWORD], ymm5
+
+ sub rax, byte SIZEOF_YMMWORD
+ add rsi, byte 1*SIZEOF_YMMWORD ; inptr
+ add rdi, byte 2*SIZEOF_YMMWORD ; outptr
+ cmp rax, byte SIZEOF_YMMWORD
+ ja near .columnloop
+ test eax, eax
+ jnz near .columnloop_last
+
+ pop rsi
+ pop rdi
+ pop rax
+
+ add rsi, byte SIZEOF_JSAMPROW ; input_data
+ add rdi, byte SIZEOF_JSAMPROW ; output_data
+ dec rcx ; rowctr
+ jg near .rowloop
+
+.return:
+ vzeroupper
+ uncollect_args 4
+ pop_xmm 3
+ pop rbp
+ ret
+
+; --------------------------------------------------------------------------
+;
+; Fancy processing for the common case of 2:1 horizontal and 2:1 vertical.
+; Again a triangle filter; see comments for h2v1 case, above.
+;
+; GLOBAL(void)
+; jsimd_h2v2_fancy_upsample_avx2 (int max_v_samp_factor,
+; JDIMENSION downsampled_width,
+; JSAMPARRAY input_data,
+; JSAMPARRAY *output_data_ptr);
+;
+
+; r10 = int max_v_samp_factor
+; r11d = JDIMENSION downsampled_width
+; r12 = JSAMPARRAY input_data
+; r13 = JSAMPARRAY *output_data_ptr
+
+%define wk(i) rbp-(WK_NUM-(i))*SIZEOF_YMMWORD ; ymmword wk[WK_NUM]
+%define WK_NUM 4
+
+ align 32
+ global EXTN(jsimd_h2v2_fancy_upsample_avx2)
+
+EXTN(jsimd_h2v2_fancy_upsample_avx2):
+ push rbp
+ mov rax, rsp ; rax = original rbp
+ sub rsp, byte 4
+ and rsp, byte (-SIZEOF_YMMWORD) ; align to 256 bits
+ mov [rsp], rax
+ mov rbp, rsp ; rbp = aligned rbp
+ lea rsp, [wk(0)]
+ push_xmm 3
+ collect_args 4
+ push rbx
+
+ mov eax, r11d ; colctr
+ test rax, rax
+ jz near .return
+
+ mov rcx, r10 ; rowctr
+ test rcx, rcx
+ jz near .return
+
+ mov rsi, r12 ; input_data
+ mov rdi, r13
+ mov rdi, JSAMPARRAY [rdi] ; output_data
+.rowloop:
+ push rax ; colctr
+ push rcx
+ push rdi
+ push rsi
+
+ mov rcx, JSAMPROW [rsi-1*SIZEOF_JSAMPROW] ; inptr1(above)
+ mov rbx, JSAMPROW [rsi+0*SIZEOF_JSAMPROW] ; inptr0
+ mov rsi, JSAMPROW [rsi+1*SIZEOF_JSAMPROW] ; inptr1(below)
+ mov rdx, JSAMPROW [rdi+0*SIZEOF_JSAMPROW] ; outptr0
+ mov rdi, JSAMPROW [rdi+1*SIZEOF_JSAMPROW] ; outptr1
+
+ vpxor ymm8, ymm8, ymm8 ; ymm8=(all 0's)
+ vpcmpeqb xmm9, xmm9, xmm9
+ vpsrldq xmm10, xmm9, (SIZEOF_XMMWORD-2) ; (ffff ---- ---- ... ---- ----) LSB is ffff
+ vpslldq xmm9, xmm9, (SIZEOF_XMMWORD-2)
+ vperm2i128 ymm9, ymm9, ymm9, 1 ; (---- ---- ... ---- ---- ffff) MSB is ffff
+
+ test rax, SIZEOF_YMMWORD-1
+ jz short .skip
+ push rdx
+ mov dl, JSAMPLE [rcx+(rax-1)*SIZEOF_JSAMPLE]
+ mov JSAMPLE [rcx+rax*SIZEOF_JSAMPLE], dl
+ mov dl, JSAMPLE [rbx+(rax-1)*SIZEOF_JSAMPLE]
+ mov JSAMPLE [rbx+rax*SIZEOF_JSAMPLE], dl
+ mov dl, JSAMPLE [rsi+(rax-1)*SIZEOF_JSAMPLE]
+ mov JSAMPLE [rsi+rax*SIZEOF_JSAMPLE], dl ; insert a dummy sample
+ pop rdx
+.skip:
+ ; -- process the first column block
+
+ vmovdqu ymm0, YMMWORD [rbx+0*SIZEOF_YMMWORD] ; ymm0=row[ 0][0]
+ vmovdqu ymm1, YMMWORD [rcx+0*SIZEOF_YMMWORD] ; ymm1=row[-1][0]
+ vmovdqu ymm2, YMMWORD [rsi+0*SIZEOF_YMMWORD] ; ymm2=row[+1][0]
+
+ vpunpckhbw ymm4, ymm0, ymm8 ; ymm4=row[ 0]( 8 9 10 11 12 13 14 15 24 25 26 27 28 29 30 31)
+ vpunpcklbw ymm5, ymm0, ymm8 ; ymm5=row[ 0]( 0 1 2 3 4 5 6 7 16 17 18 19 20 21 22 23)
+ vperm2i128 ymm0, ymm5, ymm4, 0x20 ; ymm0=row[ 0]( 0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15)
+ vperm2i128 ymm4, ymm5, ymm4, 0x31 ; ymm4=row[ 0](16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31)
+
+ vpunpckhbw ymm5, ymm1, ymm8 ; ymm5=row[-1]( 8 9 10 11 12 13 14 15 24 25 26 27 28 29 30 31)
+ vpunpcklbw ymm6, ymm1, ymm8 ; ymm6=row[-1]( 0 1 2 3 4 5 6 7 16 17 18 19 20 21 22 23)
+ vperm2i128 ymm1, ymm6, ymm5, 0x20 ; ymm1=row[-1]( 0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15)
+ vperm2i128 ymm5, ymm6, ymm5, 0x31 ; ymm5=row[-1](16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31)
+
+ vpunpckhbw ymm6, ymm2, ymm8 ; ymm6=row[+1]( 8 9 10 11 12 13 14 15 24 25 26 27 28 29 30 31)
+ vpunpcklbw ymm3, ymm2, ymm8 ; ymm3=row[+1]( 0 1 2 3 4 5 6 7 16 17 18 19 20 21 22 23)
+ vperm2i128 ymm2, ymm3, ymm6, 0x20 ; ymm2=row[+1]( 0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15)
+ vperm2i128 ymm6, ymm3, ymm6, 0x31 ; ymm6=row[+1](16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31)
+
+ vpmullw ymm0, ymm0, [rel PW_THREE]
+ vpmullw ymm4, ymm4, [rel PW_THREE]
+
+ vpaddw ymm1, ymm1, ymm0 ; ymm1=Int0L=( 0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15)
+ vpaddw ymm5, ymm5, ymm4 ; ymm5=Int0H=(16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31)
+ vpaddw ymm2, ymm2, ymm0 ; ymm2=Int1L=( 0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15)
+ vpaddw ymm6, ymm6, ymm4 ; ymm6=Int1H=(16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31)
+
+ vmovdqu YMMWORD [rdx+0*SIZEOF_YMMWORD], ymm1 ; temporarily save
+ vmovdqu YMMWORD [rdx+1*SIZEOF_YMMWORD], ymm5 ; the intermediate data
+ vmovdqu YMMWORD [rdi+0*SIZEOF_YMMWORD], ymm2
+ vmovdqu YMMWORD [rdi+1*SIZEOF_YMMWORD], ymm6
+
+ vpand ymm1, ymm1, ymm10 ; ymm1=( 0 -- -- -- -- -- -- -- -- -- -- -- -- -- -- --)
+ vpand ymm2, ymm2, ymm10 ; ymm2=( 0 -- -- -- -- -- -- -- -- -- -- -- -- -- -- --)
+
+ vmovdqa YMMWORD [wk(0)], ymm1
+ vmovdqa YMMWORD [wk(1)], ymm2
+
+ add rax, byte SIZEOF_YMMWORD-1
+ and rax, byte -SIZEOF_YMMWORD
+ cmp rax, byte SIZEOF_YMMWORD
+ ja short .columnloop
+
+.columnloop_last:
+ ; -- process the last column block
+
+ vpand ymm1, ymm9, YMMWORD [rdx+1*SIZEOF_YMMWORD]
+ vpand ymm2, ymm9, YMMWORD [rdi+1*SIZEOF_YMMWORD]
+
+ vmovdqa YMMWORD [wk(2)], ymm1 ; ymm1=(-- -- -- -- -- -- -- -- -- -- -- -- -- -- -- 31)
+ vmovdqa YMMWORD [wk(3)], ymm2 ; ymm2=(-- -- -- -- -- -- -- -- -- -- -- -- -- -- -- 31)
+
+ jmp near .upsample
+
+.columnloop:
+ ; -- process the next column block
+
+ vmovdqu ymm0, YMMWORD [rbx+1*SIZEOF_YMMWORD] ; ymm0=row[ 0][1]
+ vmovdqu ymm1, YMMWORD [rcx+1*SIZEOF_YMMWORD] ; ymm1=row[-1][1]
+ vmovdqu ymm2, YMMWORD [rsi+1*SIZEOF_YMMWORD] ; ymm2=row[+1][1]
+
+ vpunpckhbw ymm4, ymm0, ymm8 ; ymm4=row[ 0]( 8 9 10 11 12 13 14 15 24 25 26 27 28 29 30 31)
+ vpunpcklbw ymm5, ymm0, ymm8 ; ymm5=row[ 0]( 0 1 2 3 4 5 6 7 16 17 18 19 20 21 22 23)
+ vperm2i128 ymm0, ymm5, ymm4, 0x20 ; ymm0=row[ 0]( 0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15)
+ vperm2i128 ymm4, ymm5, ymm4, 0x31 ; ymm4=row[ 0](16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31)
+
+ vpunpckhbw ymm5, ymm1, ymm8 ; ymm5=row[-1]( 8 9 10 11 12 13 14 15 24 25 26 27 28 29 30 31)
+ vpunpcklbw ymm6, ymm1, ymm8 ; ymm6=row[-1]( 0 1 2 3 4 5 6 7 16 17 18 19 20 21 22 23)
+ vperm2i128 ymm1, ymm6, ymm5, 0x20 ; ymm1=row[-1]( 0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15)
+ vperm2i128 ymm5, ymm6, ymm5, 0x31 ; ymm5=row[-1](16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31)
+
+ vpunpckhbw ymm6, ymm2, ymm8 ; ymm6=row[+1]( 8 9 10 11 12 13 14 15 24 25 26 27 28 29 30 31)
+ vpunpcklbw ymm7, ymm2, ymm8 ; ymm7=row[+1]( 0 1 2 3 4 5 6 7 16 17 18 19 20 21 22 23)
+ vperm2i128 ymm2, ymm7, ymm6, 0x20 ; ymm2=row[+1]( 0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15)
+ vperm2i128 ymm6, ymm7, ymm6, 0x31 ; ymm6=row[+1](16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31)
+
+ vpmullw ymm0, ymm0, [rel PW_THREE]
+ vpmullw ymm4, ymm4, [rel PW_THREE]
+
+ vpaddw ymm1, ymm1, ymm0 ; ymm1=Int0L=( 0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15)
+ vpaddw ymm5, ymm5, ymm4 ; ymm5=Int0H=(16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31)
+ vpaddw ymm2, ymm2, ymm0 ; ymm2=Int1L=( 0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15)
+ vpaddw ymm6, ymm6, ymm4 ; ymm6=Int1H=(16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31)
+
+ vmovdqu YMMWORD [rdx+2*SIZEOF_YMMWORD], ymm1 ; temporarily save
+ vmovdqu YMMWORD [rdx+3*SIZEOF_YMMWORD], ymm5 ; the intermediate data
+ vmovdqu YMMWORD [rdi+2*SIZEOF_YMMWORD], ymm2
+ vmovdqu YMMWORD [rdi+3*SIZEOF_YMMWORD], ymm6
+
+ vperm2i128 ymm1, ymm8, ymm1, 0x20
+ vpslldq ymm1, ymm1, 14 ; ymm1=(-- -- -- -- -- -- -- -- -- -- -- -- -- -- -- 0)
+ vperm2i128 ymm2, ymm8, ymm2, 0x20
+ vpslldq ymm2, ymm2, 14 ; ymm2=(-- -- -- -- -- -- -- -- -- -- -- -- -- -- -- 0)
+
+ vmovdqa YMMWORD [wk(2)], ymm1
+ vmovdqa YMMWORD [wk(3)], ymm2
+
+.upsample:
+ ; -- process the upper row
+
+ vmovdqu ymm7, YMMWORD [rdx+0*SIZEOF_YMMWORD] ; ymm7=Int0L=( 0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15)
+ vmovdqu ymm3, YMMWORD [rdx+1*SIZEOF_YMMWORD] ; ymm3=Int0H=(16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31)
+
+ vperm2i128 ymm0, ymm8, ymm7, 0x03
+ vpalignr ymm0, ymm0, ymm7, 2 ; ymm0=( 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 --)
+ vperm2i128 ymm4, ymm8, ymm3, 0x20
+ vpslldq ymm4, ymm4, 14 ; ymm4=(-- -- -- -- -- -- -- -- -- -- -- -- -- -- -- 16)
+
+ vperm2i128 ymm5, ymm8, ymm7, 0x03
+ vpsrldq ymm5, ymm5, 14 ; ymm5=(15 -- -- -- -- -- -- -- -- -- -- -- -- -- -- --)
+ vperm2i128 ymm6, ymm8, ymm3, 0x20
+ vpalignr ymm6, ymm3, ymm6, 14 ; ymm6=(-- 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30)
+
+ vpor ymm0, ymm0, ymm4 ; ymm0=( 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16)
+ vpor ymm5, ymm5, ymm6 ; ymm5=(15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30)
+
+ vperm2i128 ymm2, ymm8, ymm3, 0x03
+ vpalignr ymm2, ymm2, ymm3, 2 ; ymm2=(17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 --)
+ vperm2i128 ymm4, ymm8, ymm3, 0x03
+ vpsrldq ymm4, ymm4, 14 ; ymm4=(31 -- -- -- -- -- -- -- -- -- -- -- -- -- -- --)
+ vperm2i128 ymm1, ymm8, ymm7, 0x20
+ vpalignr ymm1, ymm7, ymm1, 14 ; ymm1=(-- 0 1 2 3 4 5 6 7 8 9 10 11 12 13 14)
+
+ vpor ymm1, ymm1, YMMWORD [wk(0)] ; ymm1=(-1 0 1 2 3 4 5 6 7 8 9 10 11 12 13 14)
+ vpor ymm2, ymm2, YMMWORD [wk(2)] ; ymm2=(17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32)
+
+ vmovdqa YMMWORD [wk(0)], ymm4
+
+ vpmullw ymm7, ymm7, [rel PW_THREE]
+ vpmullw ymm3, ymm3, [rel PW_THREE]
+ vpaddw ymm1, ymm1, [rel PW_EIGHT]
+ vpaddw ymm5, ymm5, [rel PW_EIGHT]
+ vpaddw ymm0, ymm0, [rel PW_SEVEN]
+ vpaddw ymm2, [rel PW_SEVEN]
+
+ vpaddw ymm1, ymm1, ymm7
+ vpaddw ymm5, ymm5, ymm3
+ vpsrlw ymm1, ymm1, 4 ; ymm1=Out0LE=( 0 2 4 6 8 10 12 14 16 18 20 22 24 26 28 30)
+ vpsrlw ymm5, ymm5, 4 ; ymm5=Out0HE=(32 34 36 38 40 42 44 46 48 50 52 54 56 58 60 62)
+ vpaddw ymm0, ymm0, ymm7
+ vpaddw ymm2, ymm2, ymm3
+ vpsrlw ymm0, ymm0, 4 ; ymm0=Out0LO=( 1 3 5 7 9 11 13 15 17 19 21 23 25 27 29 31)
+ vpsrlw ymm2, ymm2, 4 ; ymm2=Out0HO=(33 35 37 39 41 43 45 47 49 51 53 55 57 59 61 63)
+
+ vpsllw ymm0, ymm0, BYTE_BIT
+ vpsllw ymm2, ymm2, BYTE_BIT
+ vpor ymm1, ymm1, ymm0 ; ymm1=Out0L=( 0 1 2 ... 29 30 31)
+ vpor ymm5, ymm5, ymm2 ; ymm5=Out0H=(32 33 34 ... 61 62 63)
+
+ vmovdqu YMMWORD [rdx+0*SIZEOF_YMMWORD], ymm1
+ vmovdqu YMMWORD [rdx+1*SIZEOF_YMMWORD], ymm5
+
+ ; -- process the lower row
+
+ vmovdqu ymm6, YMMWORD [rdi+0*SIZEOF_YMMWORD] ; ymm6=Int1L=( 0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15)
+ vmovdqu ymm4, YMMWORD [rdi+1*SIZEOF_YMMWORD] ; ymm4=Int1H=(16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31)
+
+ vperm2i128 ymm7, ymm8, ymm6, 0x03
+ vpalignr ymm7, ymm7, ymm6, 2 ; ymm7=( 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 --)
+ vperm2i128 ymm3, ymm8, ymm4, 0x20
+ vpslldq ymm3, ymm3, 14 ; ymm3=(-- -- -- -- -- -- -- -- -- -- -- -- -- -- -- 16)
+
+ vperm2i128 ymm0, ymm8, ymm6, 0x03
+ vpsrldq ymm0, ymm0, 14 ; ymm0=(15 -- -- -- -- -- -- -- -- -- -- -- -- -- -- --)
+ vperm2i128 ymm2, ymm8, ymm4, 0x20
+ vpalignr ymm2, ymm4, ymm2, 14 ; ymm2=(-- 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30)
+
+ vpor ymm7, ymm7, ymm3 ; ymm7=( 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16)
+ vpor ymm0, ymm0, ymm2 ; ymm0=(15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30)
+
+ vperm2i128 ymm5, ymm8, ymm4, 0x03
+ vpalignr ymm5, ymm5, ymm4, 2 ; ymm5=(17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 --)
+ vperm2i128 ymm3, ymm8, ymm4, 0x03
+ vpsrldq ymm3, ymm3, 14 ; ymm3=(31 -- -- -- -- -- -- -- -- -- -- -- -- -- -- --)
+ vperm2i128 ymm1, ymm8, ymm6, 0x20
+ vpalignr ymm1, ymm6, ymm1, 14 ; ymm1=(-- 0 1 2 3 4 5 6 7 8 9 10 11 12 13 14)
+
+ vpor ymm1, ymm1, YMMWORD [wk(1)] ; ymm1=(-1 0 1 2 3 4 5 6 7 8 9 10 11 12 13 14)
+ vpor ymm5, ymm5, YMMWORD [wk(3)] ; ymm5=(17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32)
+
+ vmovdqa YMMWORD [wk(1)], ymm3
+
+ vpmullw ymm6, ymm6, [rel PW_THREE]
+ vpmullw ymm4, ymm4, [rel PW_THREE]
+ vpaddw ymm1, ymm1, [rel PW_EIGHT]
+ vpaddw ymm0, ymm0, [rel PW_EIGHT]
+ vpaddw ymm7, ymm7, [rel PW_SEVEN]
+ vpaddw ymm5, ymm5, [rel PW_SEVEN]
+
+ vpaddw ymm1, ymm1, ymm6
+ vpaddw ymm0, ymm0, ymm4
+ vpsrlw ymm1, ymm1, 4 ; ymm1=Out1LE=( 0 2 4 6 8 10 12 14 16 18 20 22 24 26 28 30)
+ vpsrlw ymm0, ymm0, 4 ; ymm0=Out1HE=(32 34 36 38 40 42 44 46 48 50 52 54 56 58 60 62)
+ vpaddw ymm7, ymm7, ymm6
+ vpaddw ymm5, ymm5, ymm4
+ vpsrlw ymm7, ymm7, 4 ; ymm7=Out1LO=( 1 3 5 7 9 11 13 15 17 19 21 23 25 27 29 31)
+ vpsrlw ymm5, ymm5, 4 ; ymm5=Out1HO=(33 35 37 39 41 43 45 47 49 51 53 55 57 59 61 63)
+
+ vpsllw ymm7, ymm7, BYTE_BIT
+ vpsllw ymm5, ymm5, BYTE_BIT
+ vpor ymm1, ymm1, ymm7 ; ymm1=Out1L=( 0 1 2 ... 29 30 31)
+ vpor ymm0, ymm0, ymm5 ; ymm0=Out1H=(32 33 34 ... 61 62 63)
+
+ vmovdqu YMMWORD [rdi+0*SIZEOF_YMMWORD], ymm1
+ vmovdqu YMMWORD [rdi+1*SIZEOF_YMMWORD], ymm0
+
+ sub rax, byte SIZEOF_YMMWORD
+ add rcx, byte 1*SIZEOF_YMMWORD ; inptr1(above)
+ add rbx, byte 1*SIZEOF_YMMWORD ; inptr0
+ add rsi, byte 1*SIZEOF_YMMWORD ; inptr1(below)
+ add rdx, byte 2*SIZEOF_YMMWORD ; outptr0
+ add rdi, byte 2*SIZEOF_YMMWORD ; outptr1
+ cmp rax, byte SIZEOF_YMMWORD
+ ja near .columnloop
+ test rax, rax
+ jnz near .columnloop_last
+
+ pop rsi
+ pop rdi
+ pop rcx
+ pop rax
+
+ add rsi, byte 1*SIZEOF_JSAMPROW ; input_data
+ add rdi, byte 2*SIZEOF_JSAMPROW ; output_data
+ sub rcx, byte 2 ; rowctr
+ jg near .rowloop
+
+.return:
+ pop rbx
+ vzeroupper
+ uncollect_args 4
+ pop_xmm 3
+ mov rsp, rbp ; rsp <- aligned rbp
+ pop rsp ; rsp <- original rbp
+ pop rbp
+ ret
+
+; --------------------------------------------------------------------------
+;
+; Fast processing for the common case of 2:1 horizontal and 1:1 vertical.
+; It's still a box filter.
+;
+; GLOBAL(void)
+; jsimd_h2v1_upsample_avx2 (int max_v_samp_factor,
+; JDIMENSION output_width,
+; JSAMPARRAY input_data,
+; JSAMPARRAY *output_data_ptr);
+;
+
+; r10 = int max_v_samp_factor
+; r11d = JDIMENSION output_width
+; r12 = JSAMPARRAY input_data
+; r13 = JSAMPARRAY *output_data_ptr
+
+ align 32
+ global EXTN(jsimd_h2v1_upsample_avx2)
+
+EXTN(jsimd_h2v1_upsample_avx2):
+ push rbp
+ mov rax, rsp
+ mov rbp, rsp
+ collect_args 4
+
+ mov edx, r11d
+ add rdx, byte (SIZEOF_YMMWORD-1)
+ and rdx, -SIZEOF_YMMWORD
+ jz near .return
+
+ mov rcx, r10 ; rowctr
+ test rcx, rcx
+ jz short .return
+
+ mov rsi, r12 ; input_data
+ mov rdi, r13
+ mov rdi, JSAMPARRAY [rdi] ; output_data
+.rowloop:
+ push rdi
+ push rsi
+
+ mov rsi, JSAMPROW [rsi] ; inptr
+ mov rdi, JSAMPROW [rdi] ; outptr
+ mov rax, rdx ; colctr
+.columnloop:
+
+ cmp rax, byte SIZEOF_YMMWORD
+ ja near .above_16
+
+ vmovdqu xmm0, XMMWORD [rsi+0*SIZEOF_YMMWORD]
+ vpunpckhbw xmm1, xmm0, xmm0
+ vpunpcklbw xmm0, xmm0, xmm0
+
+ vmovdqu XMMWORD [rdi+0*SIZEOF_XMMWORD], xmm0
+ vmovdqu XMMWORD [rdi+1*SIZEOF_XMMWORD], xmm1
+
+ jmp short .nextrow
+
+.above_16:
+ vmovdqu ymm0, YMMWORD [rsi+0*SIZEOF_YMMWORD]
+
+ vpermq ymm0, ymm0, 0xd8
+ vpunpckhbw ymm1, ymm0, ymm0
+ vpunpcklbw ymm0, ymm0, ymm0
+
+ vmovdqu YMMWORD [rdi+0*SIZEOF_YMMWORD], ymm0
+ vmovdqu YMMWORD [rdi+1*SIZEOF_YMMWORD], ymm1
+
+ sub rax, byte 2*SIZEOF_YMMWORD
+ jz short .nextrow
+
+ add rsi, byte SIZEOF_YMMWORD ; inptr
+ add rdi, byte 2*SIZEOF_YMMWORD ; outptr
+ jmp short .columnloop
+
+.nextrow:
+ pop rsi
+ pop rdi
+
+ add rsi, byte SIZEOF_JSAMPROW ; input_data
+ add rdi, byte SIZEOF_JSAMPROW ; output_data
+ dec rcx ; rowctr
+ jg short .rowloop
+
+.return:
+ vzeroupper
+ uncollect_args 4
+ pop rbp
+ ret
+
+; --------------------------------------------------------------------------
+;
+; Fast processing for the common case of 2:1 horizontal and 2:1 vertical.
+; It's still a box filter.
+;
+; GLOBAL(void)
+; jsimd_h2v2_upsample_avx2 (int max_v_samp_factor,
+; JDIMENSION output_width,
+; JSAMPARRAY input_data,
+; JSAMPARRAY *output_data_ptr);
+;
+
+; r10 = int max_v_samp_factor
+; r11d = JDIMENSION output_width
+; r12 = JSAMPARRAY input_data
+; r13 = JSAMPARRAY *output_data_ptr
+
+ align 32
+ global EXTN(jsimd_h2v2_upsample_avx2)
+
+EXTN(jsimd_h2v2_upsample_avx2):
+ push rbp
+ mov rax, rsp
+ mov rbp, rsp
+ collect_args 4
+ push rbx
+
+ mov edx, r11d
+ add rdx, byte (SIZEOF_YMMWORD-1)
+ and rdx, -SIZEOF_YMMWORD
+ jz near .return
+
+ mov rcx, r10 ; rowctr
+ test rcx, rcx
+ jz near .return
+
+ mov rsi, r12 ; input_data
+ mov rdi, r13
+ mov rdi, JSAMPARRAY [rdi] ; output_data
+.rowloop:
+ push rdi
+ push rsi
+
+ mov rsi, JSAMPROW [rsi] ; inptr
+ mov rbx, JSAMPROW [rdi+0*SIZEOF_JSAMPROW] ; outptr0
+ mov rdi, JSAMPROW [rdi+1*SIZEOF_JSAMPROW] ; outptr1
+ mov rax, rdx ; colctr
+.columnloop:
+
+ cmp rax, byte SIZEOF_YMMWORD
+ ja short .above_16
+
+ vmovdqu xmm0, XMMWORD [rsi+0*SIZEOF_XMMWORD]
+ vpunpckhbw xmm1, xmm0, xmm0
+ vpunpcklbw xmm0, xmm0, xmm0
+
+ vmovdqu XMMWORD [rbx+0*SIZEOF_XMMWORD], xmm0
+ vmovdqu XMMWORD [rbx+1*SIZEOF_XMMWORD], xmm1
+ vmovdqu XMMWORD [rdi+0*SIZEOF_XMMWORD], xmm0
+ vmovdqu XMMWORD [rdi+1*SIZEOF_XMMWORD], xmm1
+
+ jmp near .nextrow
+
+.above_16:
+ vmovdqu ymm0, YMMWORD [rsi+0*SIZEOF_YMMWORD]
+
+ vpermq ymm0, ymm0, 0xd8
+ vpunpckhbw ymm1, ymm0, ymm0
+ vpunpcklbw ymm0, ymm0, ymm0
+
+ vmovdqu YMMWORD [rbx+0*SIZEOF_YMMWORD], ymm0
+ vmovdqu YMMWORD [rbx+1*SIZEOF_YMMWORD], ymm1
+ vmovdqu YMMWORD [rdi+0*SIZEOF_YMMWORD], ymm0
+ vmovdqu YMMWORD [rdi+1*SIZEOF_YMMWORD], ymm1
+
+ sub rax, byte 2*SIZEOF_YMMWORD
+ jz short .nextrow
+
+ add rsi, byte SIZEOF_YMMWORD ; inptr
+ add rbx, 2*SIZEOF_YMMWORD ; outptr0
+ add rdi, 2*SIZEOF_YMMWORD ; outptr1
+ jmp short .columnloop
+
+.nextrow:
+ pop rsi
+ pop rdi
+
+ add rsi, byte 1*SIZEOF_JSAMPROW ; input_data
+ add rdi, byte 2*SIZEOF_JSAMPROW ; output_data
+ sub rcx, byte 2 ; rowctr
+ jg near .rowloop
+
+.return:
+ pop rbx
+ vzeroupper
+ uncollect_args 4
+ pop rbp
+ ret
+
+; For some reason, the OS X linker does not honor the request to align the
+; segment unless we do this.
+ align 32
diff --git a/simd/x86_64/jdsample-sse2.asm b/simd/x86_64/jdsample-sse2.asm
new file mode 100644
index 0000000..42a8e78
--- /dev/null
+++ b/simd/x86_64/jdsample-sse2.asm
@@ -0,0 +1,670 @@
+;
+; jdsample.asm - upsampling (64-bit SSE2)
+;
+; Copyright 2009 Pierre Ossman <ossman@cendio.se> for Cendio AB
+; Copyright (C) 2009, 2016, D. R. Commander.
+;
+; Based on the x86 SIMD extension for IJG JPEG library
+; Copyright (C) 1999-2006, MIYASAKA Masaru.
+; For conditions of distribution and use, see copyright notice in jsimdext.inc
+;
+; This file should be assembled with NASM (Netwide Assembler),
+; can *not* be assembled with Microsoft's MASM or any compatible
+; assembler (including Borland's Turbo Assembler).
+; NASM is available from http://nasm.sourceforge.net/ or
+; http://sourceforge.net/project/showfiles.php?group_id=6208
+;
+; [TAB8]
+
+%include "jsimdext.inc"
+
+; --------------------------------------------------------------------------
+ SECTION SEG_CONST
+
+ alignz 32
+ global EXTN(jconst_fancy_upsample_sse2)
+
+EXTN(jconst_fancy_upsample_sse2):
+
+PW_ONE times 8 dw 1
+PW_TWO times 8 dw 2
+PW_THREE times 8 dw 3
+PW_SEVEN times 8 dw 7
+PW_EIGHT times 8 dw 8
+
+ alignz 32
+
+; --------------------------------------------------------------------------
+ SECTION SEG_TEXT
+ BITS 64
+;
+; Fancy processing for the common case of 2:1 horizontal and 1:1 vertical.
+;
+; The upsampling algorithm is linear interpolation between pixel centers,
+; also known as a "triangle filter". This is a good compromise between
+; speed and visual quality. The centers of the output pixels are 1/4 and 3/4
+; of the way between input pixel centers.
+;
+; GLOBAL(void)
+; jsimd_h2v1_fancy_upsample_sse2 (int max_v_samp_factor,
+; JDIMENSION downsampled_width,
+; JSAMPARRAY input_data,
+; JSAMPARRAY *output_data_ptr);
+;
+
+; r10 = int max_v_samp_factor
+; r11d = JDIMENSION downsampled_width
+; r12 = JSAMPARRAY input_data
+; r13 = JSAMPARRAY *output_data_ptr
+
+ align 32
+ global EXTN(jsimd_h2v1_fancy_upsample_sse2)
+
+EXTN(jsimd_h2v1_fancy_upsample_sse2):
+ push rbp
+ mov rax, rsp
+ mov rbp, rsp
+ collect_args 4
+
+ mov eax, r11d ; colctr
+ test rax, rax
+ jz near .return
+
+ mov rcx, r10 ; rowctr
+ test rcx, rcx
+ jz near .return
+
+ mov rsi, r12 ; input_data
+ mov rdi, r13
+ mov rdi, JSAMPARRAY [rdi] ; output_data
+.rowloop:
+ push rax ; colctr
+ push rdi
+ push rsi
+
+ mov rsi, JSAMPROW [rsi] ; inptr
+ mov rdi, JSAMPROW [rdi] ; outptr
+
+ test rax, SIZEOF_XMMWORD-1
+ jz short .skip
+ mov dl, JSAMPLE [rsi+(rax-1)*SIZEOF_JSAMPLE]
+ mov JSAMPLE [rsi+rax*SIZEOF_JSAMPLE], dl ; insert a dummy sample
+.skip:
+ pxor xmm0, xmm0 ; xmm0=(all 0's)
+ pcmpeqb xmm7, xmm7
+ psrldq xmm7, (SIZEOF_XMMWORD-1)
+ pand xmm7, XMMWORD [rsi+0*SIZEOF_XMMWORD]
+
+ add rax, byte SIZEOF_XMMWORD-1
+ and rax, byte -SIZEOF_XMMWORD
+ cmp rax, byte SIZEOF_XMMWORD
+ ja short .columnloop
+
+.columnloop_last:
+ pcmpeqb xmm6, xmm6
+ pslldq xmm6, (SIZEOF_XMMWORD-1)
+ pand xmm6, XMMWORD [rsi+0*SIZEOF_XMMWORD]
+ jmp short .upsample
+
+.columnloop:
+ movdqa xmm6, XMMWORD [rsi+1*SIZEOF_XMMWORD]
+ pslldq xmm6, (SIZEOF_XMMWORD-1)
+
+.upsample:
+ movdqa xmm1, XMMWORD [rsi+0*SIZEOF_XMMWORD]
+ movdqa xmm2, xmm1
+ movdqa xmm3, xmm1 ; xmm1=( 0 1 2 ... 13 14 15)
+ pslldq xmm2, 1 ; xmm2=(-- 0 1 ... 12 13 14)
+ psrldq xmm3, 1 ; xmm3=( 1 2 3 ... 14 15 --)
+
+ por xmm2, xmm7 ; xmm2=(-1 0 1 ... 12 13 14)
+ por xmm3, xmm6 ; xmm3=( 1 2 3 ... 14 15 16)
+
+ movdqa xmm7, xmm1
+ psrldq xmm7, (SIZEOF_XMMWORD-1) ; xmm7=(15 -- -- ... -- -- --)
+
+ movdqa xmm4, xmm1
+ punpcklbw xmm1, xmm0 ; xmm1=( 0 1 2 3 4 5 6 7)
+ punpckhbw xmm4, xmm0 ; xmm4=( 8 9 10 11 12 13 14 15)
+ movdqa xmm5, xmm2
+ punpcklbw xmm2, xmm0 ; xmm2=(-1 0 1 2 3 4 5 6)
+ punpckhbw xmm5, xmm0 ; xmm5=( 7 8 9 10 11 12 13 14)
+ movdqa xmm6, xmm3
+ punpcklbw xmm3, xmm0 ; xmm3=( 1 2 3 4 5 6 7 8)
+ punpckhbw xmm6, xmm0 ; xmm6=( 9 10 11 12 13 14 15 16)
+
+ pmullw xmm1, [rel PW_THREE]
+ pmullw xmm4, [rel PW_THREE]
+ paddw xmm2, [rel PW_ONE]
+ paddw xmm5, [rel PW_ONE]
+ paddw xmm3, [rel PW_TWO]
+ paddw xmm6, [rel PW_TWO]
+
+ paddw xmm2, xmm1
+ paddw xmm5, xmm4
+ psrlw xmm2, 2 ; xmm2=OutLE=( 0 2 4 6 8 10 12 14)
+ psrlw xmm5, 2 ; xmm5=OutHE=(16 18 20 22 24 26 28 30)
+ paddw xmm3, xmm1
+ paddw xmm6, xmm4
+ psrlw xmm3, 2 ; xmm3=OutLO=( 1 3 5 7 9 11 13 15)
+ psrlw xmm6, 2 ; xmm6=OutHO=(17 19 21 23 25 27 29 31)
+
+ psllw xmm3, BYTE_BIT
+ psllw xmm6, BYTE_BIT
+ por xmm2, xmm3 ; xmm2=OutL=( 0 1 2 ... 13 14 15)
+ por xmm5, xmm6 ; xmm5=OutH=(16 17 18 ... 29 30 31)
+
+ movdqa XMMWORD [rdi+0*SIZEOF_XMMWORD], xmm2
+ movdqa XMMWORD [rdi+1*SIZEOF_XMMWORD], xmm5
+
+ sub rax, byte SIZEOF_XMMWORD
+ add rsi, byte 1*SIZEOF_XMMWORD ; inptr
+ add rdi, byte 2*SIZEOF_XMMWORD ; outptr
+ cmp rax, byte SIZEOF_XMMWORD
+ ja near .columnloop
+ test eax, eax
+ jnz near .columnloop_last
+
+ pop rsi
+ pop rdi
+ pop rax
+
+ add rsi, byte SIZEOF_JSAMPROW ; input_data
+ add rdi, byte SIZEOF_JSAMPROW ; output_data
+ dec rcx ; rowctr
+ jg near .rowloop
+
+.return:
+ uncollect_args 4
+ pop rbp
+ ret
+
+; --------------------------------------------------------------------------
+;
+; Fancy processing for the common case of 2:1 horizontal and 2:1 vertical.
+; Again a triangle filter; see comments for h2v1 case, above.
+;
+; GLOBAL(void)
+; jsimd_h2v2_fancy_upsample_sse2 (int max_v_samp_factor,
+; JDIMENSION downsampled_width,
+; JSAMPARRAY input_data,
+; JSAMPARRAY *output_data_ptr);
+;
+
+; r10 = int max_v_samp_factor
+; r11d = JDIMENSION downsampled_width
+; r12 = JSAMPARRAY input_data
+; r13 = JSAMPARRAY *output_data_ptr
+
+%define wk(i) rbp-(WK_NUM-(i))*SIZEOF_XMMWORD ; xmmword wk[WK_NUM]
+%define WK_NUM 4
+
+ align 32
+ global EXTN(jsimd_h2v2_fancy_upsample_sse2)
+
+EXTN(jsimd_h2v2_fancy_upsample_sse2):
+ push rbp
+ mov rax, rsp ; rax = original rbp
+ sub rsp, byte 4
+ and rsp, byte (-SIZEOF_XMMWORD) ; align to 128 bits
+ mov [rsp], rax
+ mov rbp, rsp ; rbp = aligned rbp
+ lea rsp, [wk(0)]
+ collect_args 4
+ push rbx
+
+ mov eax, r11d ; colctr
+ test rax, rax
+ jz near .return
+
+ mov rcx, r10 ; rowctr
+ test rcx, rcx
+ jz near .return
+
+ mov rsi, r12 ; input_data
+ mov rdi, r13
+ mov rdi, JSAMPARRAY [rdi] ; output_data
+.rowloop:
+ push rax ; colctr
+ push rcx
+ push rdi
+ push rsi
+
+ mov rcx, JSAMPROW [rsi-1*SIZEOF_JSAMPROW] ; inptr1(above)
+ mov rbx, JSAMPROW [rsi+0*SIZEOF_JSAMPROW] ; inptr0
+ mov rsi, JSAMPROW [rsi+1*SIZEOF_JSAMPROW] ; inptr1(below)
+ mov rdx, JSAMPROW [rdi+0*SIZEOF_JSAMPROW] ; outptr0
+ mov rdi, JSAMPROW [rdi+1*SIZEOF_JSAMPROW] ; outptr1
+
+ test rax, SIZEOF_XMMWORD-1
+ jz short .skip
+ push rdx
+ mov dl, JSAMPLE [rcx+(rax-1)*SIZEOF_JSAMPLE]
+ mov JSAMPLE [rcx+rax*SIZEOF_JSAMPLE], dl
+ mov dl, JSAMPLE [rbx+(rax-1)*SIZEOF_JSAMPLE]
+ mov JSAMPLE [rbx+rax*SIZEOF_JSAMPLE], dl
+ mov dl, JSAMPLE [rsi+(rax-1)*SIZEOF_JSAMPLE]
+ mov JSAMPLE [rsi+rax*SIZEOF_JSAMPLE], dl ; insert a dummy sample
+ pop rdx
+.skip:
+ ; -- process the first column block
+
+ movdqa xmm0, XMMWORD [rbx+0*SIZEOF_XMMWORD] ; xmm0=row[ 0][0]
+ movdqa xmm1, XMMWORD [rcx+0*SIZEOF_XMMWORD] ; xmm1=row[-1][0]
+ movdqa xmm2, XMMWORD [rsi+0*SIZEOF_XMMWORD] ; xmm2=row[+1][0]
+
+ pxor xmm3, xmm3 ; xmm3=(all 0's)
+ movdqa xmm4, xmm0
+ punpcklbw xmm0, xmm3 ; xmm0=row[ 0]( 0 1 2 3 4 5 6 7)
+ punpckhbw xmm4, xmm3 ; xmm4=row[ 0]( 8 9 10 11 12 13 14 15)
+ movdqa xmm5, xmm1
+ punpcklbw xmm1, xmm3 ; xmm1=row[-1]( 0 1 2 3 4 5 6 7)
+ punpckhbw xmm5, xmm3 ; xmm5=row[-1]( 8 9 10 11 12 13 14 15)
+ movdqa xmm6, xmm2
+ punpcklbw xmm2, xmm3 ; xmm2=row[+1]( 0 1 2 3 4 5 6 7)
+ punpckhbw xmm6, xmm3 ; xmm6=row[+1]( 8 9 10 11 12 13 14 15)
+
+ pmullw xmm0, [rel PW_THREE]
+ pmullw xmm4, [rel PW_THREE]
+
+ pcmpeqb xmm7, xmm7
+ psrldq xmm7, (SIZEOF_XMMWORD-2)
+
+ paddw xmm1, xmm0 ; xmm1=Int0L=( 0 1 2 3 4 5 6 7)
+ paddw xmm5, xmm4 ; xmm5=Int0H=( 8 9 10 11 12 13 14 15)
+ paddw xmm2, xmm0 ; xmm2=Int1L=( 0 1 2 3 4 5 6 7)
+ paddw xmm6, xmm4 ; xmm6=Int1H=( 8 9 10 11 12 13 14 15)
+
+ movdqa XMMWORD [rdx+0*SIZEOF_XMMWORD], xmm1 ; temporarily save
+ movdqa XMMWORD [rdx+1*SIZEOF_XMMWORD], xmm5 ; the intermediate data
+ movdqa XMMWORD [rdi+0*SIZEOF_XMMWORD], xmm2
+ movdqa XMMWORD [rdi+1*SIZEOF_XMMWORD], xmm6
+
+ pand xmm1, xmm7 ; xmm1=( 0 -- -- -- -- -- -- --)
+ pand xmm2, xmm7 ; xmm2=( 0 -- -- -- -- -- -- --)
+
+ movdqa XMMWORD [wk(0)], xmm1
+ movdqa XMMWORD [wk(1)], xmm2
+
+ add rax, byte SIZEOF_XMMWORD-1
+ and rax, byte -SIZEOF_XMMWORD
+ cmp rax, byte SIZEOF_XMMWORD
+ ja short .columnloop
+
+.columnloop_last:
+ ; -- process the last column block
+
+ pcmpeqb xmm1, xmm1
+ pslldq xmm1, (SIZEOF_XMMWORD-2)
+ movdqa xmm2, xmm1
+
+ pand xmm1, XMMWORD [rdx+1*SIZEOF_XMMWORD]
+ pand xmm2, XMMWORD [rdi+1*SIZEOF_XMMWORD]
+
+ movdqa XMMWORD [wk(2)], xmm1 ; xmm1=(-- -- -- -- -- -- -- 15)
+ movdqa XMMWORD [wk(3)], xmm2 ; xmm2=(-- -- -- -- -- -- -- 15)
+
+ jmp near .upsample
+
+.columnloop:
+ ; -- process the next column block
+
+ movdqa xmm0, XMMWORD [rbx+1*SIZEOF_XMMWORD] ; xmm0=row[ 0][1]
+ movdqa xmm1, XMMWORD [rcx+1*SIZEOF_XMMWORD] ; xmm1=row[-1][1]
+ movdqa xmm2, XMMWORD [rsi+1*SIZEOF_XMMWORD] ; xmm2=row[+1][1]
+
+ pxor xmm3, xmm3 ; xmm3=(all 0's)
+ movdqa xmm4, xmm0
+ punpcklbw xmm0, xmm3 ; xmm0=row[ 0]( 0 1 2 3 4 5 6 7)
+ punpckhbw xmm4, xmm3 ; xmm4=row[ 0]( 8 9 10 11 12 13 14 15)
+ movdqa xmm5, xmm1
+ punpcklbw xmm1, xmm3 ; xmm1=row[-1]( 0 1 2 3 4 5 6 7)
+ punpckhbw xmm5, xmm3 ; xmm5=row[-1]( 8 9 10 11 12 13 14 15)
+ movdqa xmm6, xmm2
+ punpcklbw xmm2, xmm3 ; xmm2=row[+1]( 0 1 2 3 4 5 6 7)
+ punpckhbw xmm6, xmm3 ; xmm6=row[+1]( 8 9 10 11 12 13 14 15)
+
+ pmullw xmm0, [rel PW_THREE]
+ pmullw xmm4, [rel PW_THREE]
+
+ paddw xmm1, xmm0 ; xmm1=Int0L=( 0 1 2 3 4 5 6 7)
+ paddw xmm5, xmm4 ; xmm5=Int0H=( 8 9 10 11 12 13 14 15)
+ paddw xmm2, xmm0 ; xmm2=Int1L=( 0 1 2 3 4 5 6 7)
+ paddw xmm6, xmm4 ; xmm6=Int1H=( 8 9 10 11 12 13 14 15)
+
+ movdqa XMMWORD [rdx+2*SIZEOF_XMMWORD], xmm1 ; temporarily save
+ movdqa XMMWORD [rdx+3*SIZEOF_XMMWORD], xmm5 ; the intermediate data
+ movdqa XMMWORD [rdi+2*SIZEOF_XMMWORD], xmm2
+ movdqa XMMWORD [rdi+3*SIZEOF_XMMWORD], xmm6
+
+ pslldq xmm1, (SIZEOF_XMMWORD-2) ; xmm1=(-- -- -- -- -- -- -- 0)
+ pslldq xmm2, (SIZEOF_XMMWORD-2) ; xmm2=(-- -- -- -- -- -- -- 0)
+
+ movdqa XMMWORD [wk(2)], xmm1
+ movdqa XMMWORD [wk(3)], xmm2
+
+.upsample:
+ ; -- process the upper row
+
+ movdqa xmm7, XMMWORD [rdx+0*SIZEOF_XMMWORD]
+ movdqa xmm3, XMMWORD [rdx+1*SIZEOF_XMMWORD]
+
+ movdqa xmm0, xmm7 ; xmm7=Int0L=( 0 1 2 3 4 5 6 7)
+ movdqa xmm4, xmm3 ; xmm3=Int0H=( 8 9 10 11 12 13 14 15)
+ psrldq xmm0, 2 ; xmm0=( 1 2 3 4 5 6 7 --)
+ pslldq xmm4, (SIZEOF_XMMWORD-2) ; xmm4=(-- -- -- -- -- -- -- 8)
+ movdqa xmm5, xmm7
+ movdqa xmm6, xmm3
+ psrldq xmm5, (SIZEOF_XMMWORD-2) ; xmm5=( 7 -- -- -- -- -- -- --)
+ pslldq xmm6, 2 ; xmm6=(-- 8 9 10 11 12 13 14)
+
+ por xmm0, xmm4 ; xmm0=( 1 2 3 4 5 6 7 8)
+ por xmm5, xmm6 ; xmm5=( 7 8 9 10 11 12 13 14)
+
+ movdqa xmm1, xmm7
+ movdqa xmm2, xmm3
+ pslldq xmm1, 2 ; xmm1=(-- 0 1 2 3 4 5 6)
+ psrldq xmm2, 2 ; xmm2=( 9 10 11 12 13 14 15 --)
+ movdqa xmm4, xmm3
+ psrldq xmm4, (SIZEOF_XMMWORD-2) ; xmm4=(15 -- -- -- -- -- -- --)
+
+ por xmm1, XMMWORD [wk(0)] ; xmm1=(-1 0 1 2 3 4 5 6)
+ por xmm2, XMMWORD [wk(2)] ; xmm2=( 9 10 11 12 13 14 15 16)
+
+ movdqa XMMWORD [wk(0)], xmm4
+
+ pmullw xmm7, [rel PW_THREE]
+ pmullw xmm3, [rel PW_THREE]
+ paddw xmm1, [rel PW_EIGHT]
+ paddw xmm5, [rel PW_EIGHT]
+ paddw xmm0, [rel PW_SEVEN]
+ paddw xmm2, [rel PW_SEVEN]
+
+ paddw xmm1, xmm7
+ paddw xmm5, xmm3
+ psrlw xmm1, 4 ; xmm1=Out0LE=( 0 2 4 6 8 10 12 14)
+ psrlw xmm5, 4 ; xmm5=Out0HE=(16 18 20 22 24 26 28 30)
+ paddw xmm0, xmm7
+ paddw xmm2, xmm3
+ psrlw xmm0, 4 ; xmm0=Out0LO=( 1 3 5 7 9 11 13 15)
+ psrlw xmm2, 4 ; xmm2=Out0HO=(17 19 21 23 25 27 29 31)
+
+ psllw xmm0, BYTE_BIT
+ psllw xmm2, BYTE_BIT
+ por xmm1, xmm0 ; xmm1=Out0L=( 0 1 2 ... 13 14 15)
+ por xmm5, xmm2 ; xmm5=Out0H=(16 17 18 ... 29 30 31)
+
+ movdqa XMMWORD [rdx+0*SIZEOF_XMMWORD], xmm1
+ movdqa XMMWORD [rdx+1*SIZEOF_XMMWORD], xmm5
+
+ ; -- process the lower row
+
+ movdqa xmm6, XMMWORD [rdi+0*SIZEOF_XMMWORD]
+ movdqa xmm4, XMMWORD [rdi+1*SIZEOF_XMMWORD]
+
+ movdqa xmm7, xmm6 ; xmm6=Int1L=( 0 1 2 3 4 5 6 7)
+ movdqa xmm3, xmm4 ; xmm4=Int1H=( 8 9 10 11 12 13 14 15)
+ psrldq xmm7, 2 ; xmm7=( 1 2 3 4 5 6 7 --)
+ pslldq xmm3, (SIZEOF_XMMWORD-2) ; xmm3=(-- -- -- -- -- -- -- 8)
+ movdqa xmm0, xmm6
+ movdqa xmm2, xmm4
+ psrldq xmm0, (SIZEOF_XMMWORD-2) ; xmm0=( 7 -- -- -- -- -- -- --)
+ pslldq xmm2, 2 ; xmm2=(-- 8 9 10 11 12 13 14)
+
+ por xmm7, xmm3 ; xmm7=( 1 2 3 4 5 6 7 8)
+ por xmm0, xmm2 ; xmm0=( 7 8 9 10 11 12 13 14)
+
+ movdqa xmm1, xmm6
+ movdqa xmm5, xmm4
+ pslldq xmm1, 2 ; xmm1=(-- 0 1 2 3 4 5 6)
+ psrldq xmm5, 2 ; xmm5=( 9 10 11 12 13 14 15 --)
+ movdqa xmm3, xmm4
+ psrldq xmm3, (SIZEOF_XMMWORD-2) ; xmm3=(15 -- -- -- -- -- -- --)
+
+ por xmm1, XMMWORD [wk(1)] ; xmm1=(-1 0 1 2 3 4 5 6)
+ por xmm5, XMMWORD [wk(3)] ; xmm5=( 9 10 11 12 13 14 15 16)
+
+ movdqa XMMWORD [wk(1)], xmm3
+
+ pmullw xmm6, [rel PW_THREE]
+ pmullw xmm4, [rel PW_THREE]
+ paddw xmm1, [rel PW_EIGHT]
+ paddw xmm0, [rel PW_EIGHT]
+ paddw xmm7, [rel PW_SEVEN]
+ paddw xmm5, [rel PW_SEVEN]
+
+ paddw xmm1, xmm6
+ paddw xmm0, xmm4
+ psrlw xmm1, 4 ; xmm1=Out1LE=( 0 2 4 6 8 10 12 14)
+ psrlw xmm0, 4 ; xmm0=Out1HE=(16 18 20 22 24 26 28 30)
+ paddw xmm7, xmm6
+ paddw xmm5, xmm4
+ psrlw xmm7, 4 ; xmm7=Out1LO=( 1 3 5 7 9 11 13 15)
+ psrlw xmm5, 4 ; xmm5=Out1HO=(17 19 21 23 25 27 29 31)
+
+ psllw xmm7, BYTE_BIT
+ psllw xmm5, BYTE_BIT
+ por xmm1, xmm7 ; xmm1=Out1L=( 0 1 2 ... 13 14 15)
+ por xmm0, xmm5 ; xmm0=Out1H=(16 17 18 ... 29 30 31)
+
+ movdqa XMMWORD [rdi+0*SIZEOF_XMMWORD], xmm1
+ movdqa XMMWORD [rdi+1*SIZEOF_XMMWORD], xmm0
+
+ sub rax, byte SIZEOF_XMMWORD
+ add rcx, byte 1*SIZEOF_XMMWORD ; inptr1(above)
+ add rbx, byte 1*SIZEOF_XMMWORD ; inptr0
+ add rsi, byte 1*SIZEOF_XMMWORD ; inptr1(below)
+ add rdx, byte 2*SIZEOF_XMMWORD ; outptr0
+ add rdi, byte 2*SIZEOF_XMMWORD ; outptr1
+ cmp rax, byte SIZEOF_XMMWORD
+ ja near .columnloop
+ test rax, rax
+ jnz near .columnloop_last
+
+ pop rsi
+ pop rdi
+ pop rcx
+ pop rax
+
+ add rsi, byte 1*SIZEOF_JSAMPROW ; input_data
+ add rdi, byte 2*SIZEOF_JSAMPROW ; output_data
+ sub rcx, byte 2 ; rowctr
+ jg near .rowloop
+
+.return:
+ pop rbx
+ uncollect_args 4
+ mov rsp, rbp ; rsp <- aligned rbp
+ pop rsp ; rsp <- original rbp
+ pop rbp
+ ret
+
+; --------------------------------------------------------------------------
+;
+; Fast processing for the common case of 2:1 horizontal and 1:1 vertical.
+; It's still a box filter.
+;
+; GLOBAL(void)
+; jsimd_h2v1_upsample_sse2 (int max_v_samp_factor,
+; JDIMENSION output_width,
+; JSAMPARRAY input_data,
+; JSAMPARRAY *output_data_ptr);
+;
+
+; r10 = int max_v_samp_factor
+; r11d = JDIMENSION output_width
+; r12 = JSAMPARRAY input_data
+; r13 = JSAMPARRAY *output_data_ptr
+
+ align 32
+ global EXTN(jsimd_h2v1_upsample_sse2)
+
+EXTN(jsimd_h2v1_upsample_sse2):
+ push rbp
+ mov rax, rsp
+ mov rbp, rsp
+ collect_args 4
+
+ mov edx, r11d
+ add rdx, byte (2*SIZEOF_XMMWORD)-1
+ and rdx, byte -(2*SIZEOF_XMMWORD)
+ jz near .return
+
+ mov rcx, r10 ; rowctr
+ test rcx, rcx
+ jz short .return
+
+ mov rsi, r12 ; input_data
+ mov rdi, r13
+ mov rdi, JSAMPARRAY [rdi] ; output_data
+.rowloop:
+ push rdi
+ push rsi
+
+ mov rsi, JSAMPROW [rsi] ; inptr
+ mov rdi, JSAMPROW [rdi] ; outptr
+ mov rax, rdx ; colctr
+.columnloop:
+
+ movdqa xmm0, XMMWORD [rsi+0*SIZEOF_XMMWORD]
+
+ movdqa xmm1, xmm0
+ punpcklbw xmm0, xmm0
+ punpckhbw xmm1, xmm1
+
+ movdqa XMMWORD [rdi+0*SIZEOF_XMMWORD], xmm0
+ movdqa XMMWORD [rdi+1*SIZEOF_XMMWORD], xmm1
+
+ sub rax, byte 2*SIZEOF_XMMWORD
+ jz short .nextrow
+
+ movdqa xmm2, XMMWORD [rsi+1*SIZEOF_XMMWORD]
+
+ movdqa xmm3, xmm2
+ punpcklbw xmm2, xmm2
+ punpckhbw xmm3, xmm3
+
+ movdqa XMMWORD [rdi+2*SIZEOF_XMMWORD], xmm2
+ movdqa XMMWORD [rdi+3*SIZEOF_XMMWORD], xmm3
+
+ sub rax, byte 2*SIZEOF_XMMWORD
+ jz short .nextrow
+
+ add rsi, byte 2*SIZEOF_XMMWORD ; inptr
+ add rdi, byte 4*SIZEOF_XMMWORD ; outptr
+ jmp short .columnloop
+
+.nextrow:
+ pop rsi
+ pop rdi
+
+ add rsi, byte SIZEOF_JSAMPROW ; input_data
+ add rdi, byte SIZEOF_JSAMPROW ; output_data
+ dec rcx ; rowctr
+ jg short .rowloop
+
+.return:
+ uncollect_args 4
+ pop rbp
+ ret
+
+; --------------------------------------------------------------------------
+;
+; Fast processing for the common case of 2:1 horizontal and 2:1 vertical.
+; It's still a box filter.
+;
+; GLOBAL(void)
+; jsimd_h2v2_upsample_sse2 (int max_v_samp_factor,
+; JDIMENSION output_width,
+; JSAMPARRAY input_data,
+; JSAMPARRAY *output_data_ptr);
+;
+
+; r10 = int max_v_samp_factor
+; r11d = JDIMENSION output_width
+; r12 = JSAMPARRAY input_data
+; r13 = JSAMPARRAY *output_data_ptr
+
+ align 32
+ global EXTN(jsimd_h2v2_upsample_sse2)
+
+EXTN(jsimd_h2v2_upsample_sse2):
+ push rbp
+ mov rax, rsp
+ mov rbp, rsp
+ collect_args 4
+ push rbx
+
+ mov edx, r11d
+ add rdx, byte (2*SIZEOF_XMMWORD)-1
+ and rdx, byte -(2*SIZEOF_XMMWORD)
+ jz near .return
+
+ mov rcx, r10 ; rowctr
+ test rcx, rcx
+ jz near .return
+
+ mov rsi, r12 ; input_data
+ mov rdi, r13
+ mov rdi, JSAMPARRAY [rdi] ; output_data
+.rowloop:
+ push rdi
+ push rsi
+
+ mov rsi, JSAMPROW [rsi] ; inptr
+ mov rbx, JSAMPROW [rdi+0*SIZEOF_JSAMPROW] ; outptr0
+ mov rdi, JSAMPROW [rdi+1*SIZEOF_JSAMPROW] ; outptr1
+ mov rax, rdx ; colctr
+.columnloop:
+
+ movdqa xmm0, XMMWORD [rsi+0*SIZEOF_XMMWORD]
+
+ movdqa xmm1, xmm0
+ punpcklbw xmm0, xmm0
+ punpckhbw xmm1, xmm1
+
+ movdqa XMMWORD [rbx+0*SIZEOF_XMMWORD], xmm0
+ movdqa XMMWORD [rbx+1*SIZEOF_XMMWORD], xmm1
+ movdqa XMMWORD [rdi+0*SIZEOF_XMMWORD], xmm0
+ movdqa XMMWORD [rdi+1*SIZEOF_XMMWORD], xmm1
+
+ sub rax, byte 2*SIZEOF_XMMWORD
+ jz short .nextrow
+
+ movdqa xmm2, XMMWORD [rsi+1*SIZEOF_XMMWORD]
+
+ movdqa xmm3, xmm2
+ punpcklbw xmm2, xmm2
+ punpckhbw xmm3, xmm3
+
+ movdqa XMMWORD [rbx+2*SIZEOF_XMMWORD], xmm2
+ movdqa XMMWORD [rbx+3*SIZEOF_XMMWORD], xmm3
+ movdqa XMMWORD [rdi+2*SIZEOF_XMMWORD], xmm2
+ movdqa XMMWORD [rdi+3*SIZEOF_XMMWORD], xmm3
+
+ sub rax, byte 2*SIZEOF_XMMWORD
+ jz short .nextrow
+
+ add rsi, byte 2*SIZEOF_XMMWORD ; inptr
+ add rbx, byte 4*SIZEOF_XMMWORD ; outptr0
+ add rdi, byte 4*SIZEOF_XMMWORD ; outptr1
+ jmp short .columnloop
+
+.nextrow:
+ pop rsi
+ pop rdi
+
+ add rsi, byte 1*SIZEOF_JSAMPROW ; input_data
+ add rdi, byte 2*SIZEOF_JSAMPROW ; output_data
+ sub rcx, byte 2 ; rowctr
+ jg near .rowloop
+
+.return:
+ pop rbx
+ uncollect_args 4
+ pop rbp
+ ret
+
+; For some reason, the OS X linker does not honor the request to align the
+; segment unless we do this.
+ align 32
diff --git a/simd/x86_64/jfdctflt-sse.asm b/simd/x86_64/jfdctflt-sse.asm
new file mode 100644
index 0000000..127ef28
--- /dev/null
+++ b/simd/x86_64/jfdctflt-sse.asm
@@ -0,0 +1,357 @@
+;
+; jfdctflt.asm - floating-point FDCT (64-bit SSE)
+;
+; Copyright 2009 Pierre Ossman <ossman@cendio.se> for Cendio AB
+; Copyright (C) 2009, 2016, D. R. Commander.
+;
+; Based on the x86 SIMD extension for IJG JPEG library
+; Copyright (C) 1999-2006, MIYASAKA Masaru.
+; For conditions of distribution and use, see copyright notice in jsimdext.inc
+;
+; This file should be assembled with NASM (Netwide Assembler),
+; can *not* be assembled with Microsoft's MASM or any compatible
+; assembler (including Borland's Turbo Assembler).
+; NASM is available from http://nasm.sourceforge.net/ or
+; http://sourceforge.net/project/showfiles.php?group_id=6208
+;
+; This file contains a floating-point implementation of the forward DCT
+; (Discrete Cosine Transform). The following code is based directly on
+; the IJG's original jfdctflt.c; see the jfdctflt.c for more details.
+;
+; [TAB8]
+
+%include "jsimdext.inc"
+%include "jdct.inc"
+
+; --------------------------------------------------------------------------
+
+%macro unpcklps2 2 ; %1=(0 1 2 3) / %2=(4 5 6 7) => %1=(0 1 4 5)
+ shufps %1, %2, 0x44
+%endmacro
+
+%macro unpckhps2 2 ; %1=(0 1 2 3) / %2=(4 5 6 7) => %1=(2 3 6 7)
+ shufps %1, %2, 0xEE
+%endmacro
+
+; --------------------------------------------------------------------------
+ SECTION SEG_CONST
+
+ alignz 32
+ global EXTN(jconst_fdct_float_sse)
+
+EXTN(jconst_fdct_float_sse):
+
+PD_0_382 times 4 dd 0.382683432365089771728460
+PD_0_707 times 4 dd 0.707106781186547524400844
+PD_0_541 times 4 dd 0.541196100146196984399723
+PD_1_306 times 4 dd 1.306562964876376527856643
+
+ alignz 32
+
+; --------------------------------------------------------------------------
+ SECTION SEG_TEXT
+ BITS 64
+;
+; Perform the forward DCT on one block of samples.
+;
+; GLOBAL(void)
+; jsimd_fdct_float_sse (FAST_FLOAT *data)
+;
+
+; r10 = FAST_FLOAT *data
+
+%define wk(i) rbp-(WK_NUM-(i))*SIZEOF_XMMWORD ; xmmword wk[WK_NUM]
+%define WK_NUM 2
+
+ align 32
+ global EXTN(jsimd_fdct_float_sse)
+
+EXTN(jsimd_fdct_float_sse):
+ push rbp
+ mov rax, rsp ; rax = original rbp
+ sub rsp, byte 4
+ and rsp, byte (-SIZEOF_XMMWORD) ; align to 128 bits
+ mov [rsp], rax
+ mov rbp, rsp ; rbp = aligned rbp
+ lea rsp, [wk(0)]
+ collect_args 1
+
+ ; ---- Pass 1: process rows.
+
+ mov rdx, r10 ; (FAST_FLOAT *)
+ mov rcx, DCTSIZE/4
+.rowloop:
+
+ movaps xmm0, XMMWORD [XMMBLOCK(2,0,rdx,SIZEOF_FAST_FLOAT)]
+ movaps xmm1, XMMWORD [XMMBLOCK(3,0,rdx,SIZEOF_FAST_FLOAT)]
+ movaps xmm2, XMMWORD [XMMBLOCK(2,1,rdx,SIZEOF_FAST_FLOAT)]
+ movaps xmm3, XMMWORD [XMMBLOCK(3,1,rdx,SIZEOF_FAST_FLOAT)]
+
+ ; xmm0=(20 21 22 23), xmm2=(24 25 26 27)
+ ; xmm1=(30 31 32 33), xmm3=(34 35 36 37)
+
+ movaps xmm4, xmm0 ; transpose coefficients(phase 1)
+ unpcklps xmm0, xmm1 ; xmm0=(20 30 21 31)
+ unpckhps xmm4, xmm1 ; xmm4=(22 32 23 33)
+ movaps xmm5, xmm2 ; transpose coefficients(phase 1)
+ unpcklps xmm2, xmm3 ; xmm2=(24 34 25 35)
+ unpckhps xmm5, xmm3 ; xmm5=(26 36 27 37)
+
+ movaps xmm6, XMMWORD [XMMBLOCK(0,0,rdx,SIZEOF_FAST_FLOAT)]
+ movaps xmm7, XMMWORD [XMMBLOCK(1,0,rdx,SIZEOF_FAST_FLOAT)]
+ movaps xmm1, XMMWORD [XMMBLOCK(0,1,rdx,SIZEOF_FAST_FLOAT)]
+ movaps xmm3, XMMWORD [XMMBLOCK(1,1,rdx,SIZEOF_FAST_FLOAT)]
+
+ ; xmm6=(00 01 02 03), xmm1=(04 05 06 07)
+ ; xmm7=(10 11 12 13), xmm3=(14 15 16 17)
+
+ movaps XMMWORD [wk(0)], xmm4 ; wk(0)=(22 32 23 33)
+ movaps XMMWORD [wk(1)], xmm2 ; wk(1)=(24 34 25 35)
+
+ movaps xmm4, xmm6 ; transpose coefficients(phase 1)
+ unpcklps xmm6, xmm7 ; xmm6=(00 10 01 11)
+ unpckhps xmm4, xmm7 ; xmm4=(02 12 03 13)
+ movaps xmm2, xmm1 ; transpose coefficients(phase 1)
+ unpcklps xmm1, xmm3 ; xmm1=(04 14 05 15)
+ unpckhps xmm2, xmm3 ; xmm2=(06 16 07 17)
+
+ movaps xmm7, xmm6 ; transpose coefficients(phase 2)
+ unpcklps2 xmm6, xmm0 ; xmm6=(00 10 20 30)=data0
+ unpckhps2 xmm7, xmm0 ; xmm7=(01 11 21 31)=data1
+ movaps xmm3, xmm2 ; transpose coefficients(phase 2)
+ unpcklps2 xmm2, xmm5 ; xmm2=(06 16 26 36)=data6
+ unpckhps2 xmm3, xmm5 ; xmm3=(07 17 27 37)=data7
+
+ movaps xmm0, xmm7
+ movaps xmm5, xmm6
+ subps xmm7, xmm2 ; xmm7=data1-data6=tmp6
+ subps xmm6, xmm3 ; xmm6=data0-data7=tmp7
+ addps xmm0, xmm2 ; xmm0=data1+data6=tmp1
+ addps xmm5, xmm3 ; xmm5=data0+data7=tmp0
+
+ movaps xmm2, XMMWORD [wk(0)] ; xmm2=(22 32 23 33)
+ movaps xmm3, XMMWORD [wk(1)] ; xmm3=(24 34 25 35)
+ movaps XMMWORD [wk(0)], xmm7 ; wk(0)=tmp6
+ movaps XMMWORD [wk(1)], xmm6 ; wk(1)=tmp7
+
+ movaps xmm7, xmm4 ; transpose coefficients(phase 2)
+ unpcklps2 xmm4, xmm2 ; xmm4=(02 12 22 32)=data2
+ unpckhps2 xmm7, xmm2 ; xmm7=(03 13 23 33)=data3
+ movaps xmm6, xmm1 ; transpose coefficients(phase 2)
+ unpcklps2 xmm1, xmm3 ; xmm1=(04 14 24 34)=data4
+ unpckhps2 xmm6, xmm3 ; xmm6=(05 15 25 35)=data5
+
+ movaps xmm2, xmm7
+ movaps xmm3, xmm4
+ addps xmm7, xmm1 ; xmm7=data3+data4=tmp3
+ addps xmm4, xmm6 ; xmm4=data2+data5=tmp2
+ subps xmm2, xmm1 ; xmm2=data3-data4=tmp4
+ subps xmm3, xmm6 ; xmm3=data2-data5=tmp5
+
+ ; -- Even part
+
+ movaps xmm1, xmm5
+ movaps xmm6, xmm0
+ subps xmm5, xmm7 ; xmm5=tmp13
+ subps xmm0, xmm4 ; xmm0=tmp12
+ addps xmm1, xmm7 ; xmm1=tmp10
+ addps xmm6, xmm4 ; xmm6=tmp11
+
+ addps xmm0, xmm5
+ mulps xmm0, [rel PD_0_707] ; xmm0=z1
+
+ movaps xmm7, xmm1
+ movaps xmm4, xmm5
+ subps xmm1, xmm6 ; xmm1=data4
+ subps xmm5, xmm0 ; xmm5=data6
+ addps xmm7, xmm6 ; xmm7=data0
+ addps xmm4, xmm0 ; xmm4=data2
+
+ movaps XMMWORD [XMMBLOCK(0,1,rdx,SIZEOF_FAST_FLOAT)], xmm1
+ movaps XMMWORD [XMMBLOCK(2,1,rdx,SIZEOF_FAST_FLOAT)], xmm5
+ movaps XMMWORD [XMMBLOCK(0,0,rdx,SIZEOF_FAST_FLOAT)], xmm7
+ movaps XMMWORD [XMMBLOCK(2,0,rdx,SIZEOF_FAST_FLOAT)], xmm4
+
+ ; -- Odd part
+
+ movaps xmm6, XMMWORD [wk(0)] ; xmm6=tmp6
+ movaps xmm0, XMMWORD [wk(1)] ; xmm0=tmp7
+
+ addps xmm2, xmm3 ; xmm2=tmp10
+ addps xmm3, xmm6 ; xmm3=tmp11
+ addps xmm6, xmm0 ; xmm6=tmp12, xmm0=tmp7
+
+ mulps xmm3, [rel PD_0_707] ; xmm3=z3
+
+ movaps xmm1, xmm2 ; xmm1=tmp10
+ subps xmm2, xmm6
+ mulps xmm2, [rel PD_0_382] ; xmm2=z5
+ mulps xmm1, [rel PD_0_541] ; xmm1=MULTIPLY(tmp10,FIX_0_541196)
+ mulps xmm6, [rel PD_1_306] ; xmm6=MULTIPLY(tmp12,FIX_1_306562)
+ addps xmm1, xmm2 ; xmm1=z2
+ addps xmm6, xmm2 ; xmm6=z4
+
+ movaps xmm5, xmm0
+ subps xmm0, xmm3 ; xmm0=z13
+ addps xmm5, xmm3 ; xmm5=z11
+
+ movaps xmm7, xmm0
+ movaps xmm4, xmm5
+ subps xmm0, xmm1 ; xmm0=data3
+ subps xmm5, xmm6 ; xmm5=data7
+ addps xmm7, xmm1 ; xmm7=data5
+ addps xmm4, xmm6 ; xmm4=data1
+
+ movaps XMMWORD [XMMBLOCK(3,0,rdx,SIZEOF_FAST_FLOAT)], xmm0
+ movaps XMMWORD [XMMBLOCK(3,1,rdx,SIZEOF_FAST_FLOAT)], xmm5
+ movaps XMMWORD [XMMBLOCK(1,1,rdx,SIZEOF_FAST_FLOAT)], xmm7
+ movaps XMMWORD [XMMBLOCK(1,0,rdx,SIZEOF_FAST_FLOAT)], xmm4
+
+ add rdx, 4*DCTSIZE*SIZEOF_FAST_FLOAT
+ dec rcx
+ jnz near .rowloop
+
+ ; ---- Pass 2: process columns.
+
+ mov rdx, r10 ; (FAST_FLOAT *)
+ mov rcx, DCTSIZE/4
+.columnloop:
+
+ movaps xmm0, XMMWORD [XMMBLOCK(2,0,rdx,SIZEOF_FAST_FLOAT)]
+ movaps xmm1, XMMWORD [XMMBLOCK(3,0,rdx,SIZEOF_FAST_FLOAT)]
+ movaps xmm2, XMMWORD [XMMBLOCK(6,0,rdx,SIZEOF_FAST_FLOAT)]
+ movaps xmm3, XMMWORD [XMMBLOCK(7,0,rdx,SIZEOF_FAST_FLOAT)]
+
+ ; xmm0=(02 12 22 32), xmm2=(42 52 62 72)
+ ; xmm1=(03 13 23 33), xmm3=(43 53 63 73)
+
+ movaps xmm4, xmm0 ; transpose coefficients(phase 1)
+ unpcklps xmm0, xmm1 ; xmm0=(02 03 12 13)
+ unpckhps xmm4, xmm1 ; xmm4=(22 23 32 33)
+ movaps xmm5, xmm2 ; transpose coefficients(phase 1)
+ unpcklps xmm2, xmm3 ; xmm2=(42 43 52 53)
+ unpckhps xmm5, xmm3 ; xmm5=(62 63 72 73)
+
+ movaps xmm6, XMMWORD [XMMBLOCK(0,0,rdx,SIZEOF_FAST_FLOAT)]
+ movaps xmm7, XMMWORD [XMMBLOCK(1,0,rdx,SIZEOF_FAST_FLOAT)]
+ movaps xmm1, XMMWORD [XMMBLOCK(4,0,rdx,SIZEOF_FAST_FLOAT)]
+ movaps xmm3, XMMWORD [XMMBLOCK(5,0,rdx,SIZEOF_FAST_FLOAT)]
+
+ ; xmm6=(00 10 20 30), xmm1=(40 50 60 70)
+ ; xmm7=(01 11 21 31), xmm3=(41 51 61 71)
+
+ movaps XMMWORD [wk(0)], xmm4 ; wk(0)=(22 23 32 33)
+ movaps XMMWORD [wk(1)], xmm2 ; wk(1)=(42 43 52 53)
+
+ movaps xmm4, xmm6 ; transpose coefficients(phase 1)
+ unpcklps xmm6, xmm7 ; xmm6=(00 01 10 11)
+ unpckhps xmm4, xmm7 ; xmm4=(20 21 30 31)
+ movaps xmm2, xmm1 ; transpose coefficients(phase 1)
+ unpcklps xmm1, xmm3 ; xmm1=(40 41 50 51)
+ unpckhps xmm2, xmm3 ; xmm2=(60 61 70 71)
+
+ movaps xmm7, xmm6 ; transpose coefficients(phase 2)
+ unpcklps2 xmm6, xmm0 ; xmm6=(00 01 02 03)=data0
+ unpckhps2 xmm7, xmm0 ; xmm7=(10 11 12 13)=data1
+ movaps xmm3, xmm2 ; transpose coefficients(phase 2)
+ unpcklps2 xmm2, xmm5 ; xmm2=(60 61 62 63)=data6
+ unpckhps2 xmm3, xmm5 ; xmm3=(70 71 72 73)=data7
+
+ movaps xmm0, xmm7
+ movaps xmm5, xmm6
+ subps xmm7, xmm2 ; xmm7=data1-data6=tmp6
+ subps xmm6, xmm3 ; xmm6=data0-data7=tmp7
+ addps xmm0, xmm2 ; xmm0=data1+data6=tmp1
+ addps xmm5, xmm3 ; xmm5=data0+data7=tmp0
+
+ movaps xmm2, XMMWORD [wk(0)] ; xmm2=(22 23 32 33)
+ movaps xmm3, XMMWORD [wk(1)] ; xmm3=(42 43 52 53)
+ movaps XMMWORD [wk(0)], xmm7 ; wk(0)=tmp6
+ movaps XMMWORD [wk(1)], xmm6 ; wk(1)=tmp7
+
+ movaps xmm7, xmm4 ; transpose coefficients(phase 2)
+ unpcklps2 xmm4, xmm2 ; xmm4=(20 21 22 23)=data2
+ unpckhps2 xmm7, xmm2 ; xmm7=(30 31 32 33)=data3
+ movaps xmm6, xmm1 ; transpose coefficients(phase 2)
+ unpcklps2 xmm1, xmm3 ; xmm1=(40 41 42 43)=data4
+ unpckhps2 xmm6, xmm3 ; xmm6=(50 51 52 53)=data5
+
+ movaps xmm2, xmm7
+ movaps xmm3, xmm4
+ addps xmm7, xmm1 ; xmm7=data3+data4=tmp3
+ addps xmm4, xmm6 ; xmm4=data2+data5=tmp2
+ subps xmm2, xmm1 ; xmm2=data3-data4=tmp4
+ subps xmm3, xmm6 ; xmm3=data2-data5=tmp5
+
+ ; -- Even part
+
+ movaps xmm1, xmm5
+ movaps xmm6, xmm0
+ subps xmm5, xmm7 ; xmm5=tmp13
+ subps xmm0, xmm4 ; xmm0=tmp12
+ addps xmm1, xmm7 ; xmm1=tmp10
+ addps xmm6, xmm4 ; xmm6=tmp11
+
+ addps xmm0, xmm5
+ mulps xmm0, [rel PD_0_707] ; xmm0=z1
+
+ movaps xmm7, xmm1
+ movaps xmm4, xmm5
+ subps xmm1, xmm6 ; xmm1=data4
+ subps xmm5, xmm0 ; xmm5=data6
+ addps xmm7, xmm6 ; xmm7=data0
+ addps xmm4, xmm0 ; xmm4=data2
+
+ movaps XMMWORD [XMMBLOCK(4,0,rdx,SIZEOF_FAST_FLOAT)], xmm1
+ movaps XMMWORD [XMMBLOCK(6,0,rdx,SIZEOF_FAST_FLOAT)], xmm5
+ movaps XMMWORD [XMMBLOCK(0,0,rdx,SIZEOF_FAST_FLOAT)], xmm7
+ movaps XMMWORD [XMMBLOCK(2,0,rdx,SIZEOF_FAST_FLOAT)], xmm4
+
+ ; -- Odd part
+
+ movaps xmm6, XMMWORD [wk(0)] ; xmm6=tmp6
+ movaps xmm0, XMMWORD [wk(1)] ; xmm0=tmp7
+
+ addps xmm2, xmm3 ; xmm2=tmp10
+ addps xmm3, xmm6 ; xmm3=tmp11
+ addps xmm6, xmm0 ; xmm6=tmp12, xmm0=tmp7
+
+ mulps xmm3, [rel PD_0_707] ; xmm3=z3
+
+ movaps xmm1, xmm2 ; xmm1=tmp10
+ subps xmm2, xmm6
+ mulps xmm2, [rel PD_0_382] ; xmm2=z5
+ mulps xmm1, [rel PD_0_541] ; xmm1=MULTIPLY(tmp10,FIX_0_541196)
+ mulps xmm6, [rel PD_1_306] ; xmm6=MULTIPLY(tmp12,FIX_1_306562)
+ addps xmm1, xmm2 ; xmm1=z2
+ addps xmm6, xmm2 ; xmm6=z4
+
+ movaps xmm5, xmm0
+ subps xmm0, xmm3 ; xmm0=z13
+ addps xmm5, xmm3 ; xmm5=z11
+
+ movaps xmm7, xmm0
+ movaps xmm4, xmm5
+ subps xmm0, xmm1 ; xmm0=data3
+ subps xmm5, xmm6 ; xmm5=data7
+ addps xmm7, xmm1 ; xmm7=data5
+ addps xmm4, xmm6 ; xmm4=data1
+
+ movaps XMMWORD [XMMBLOCK(3,0,rdx,SIZEOF_FAST_FLOAT)], xmm0
+ movaps XMMWORD [XMMBLOCK(7,0,rdx,SIZEOF_FAST_FLOAT)], xmm5
+ movaps XMMWORD [XMMBLOCK(5,0,rdx,SIZEOF_FAST_FLOAT)], xmm7
+ movaps XMMWORD [XMMBLOCK(1,0,rdx,SIZEOF_FAST_FLOAT)], xmm4
+
+ add rdx, byte 4*SIZEOF_FAST_FLOAT
+ dec rcx
+ jnz near .columnloop
+
+ uncollect_args 1
+ mov rsp, rbp ; rsp <- aligned rbp
+ pop rsp ; rsp <- original rbp
+ pop rbp
+ ret
+
+; For some reason, the OS X linker does not honor the request to align the
+; segment unless we do this.
+ align 32
diff --git a/simd/x86_64/jfdctfst-sse2.asm b/simd/x86_64/jfdctfst-sse2.asm
new file mode 100644
index 0000000..479e5da
--- /dev/null
+++ b/simd/x86_64/jfdctfst-sse2.asm
@@ -0,0 +1,391 @@
+;
+; jfdctfst.asm - fast integer FDCT (64-bit SSE2)
+;
+; Copyright 2009 Pierre Ossman <ossman@cendio.se> for Cendio AB
+; Copyright (C) 2009, 2016, D. R. Commander.
+;
+; Based on the x86 SIMD extension for IJG JPEG library
+; Copyright (C) 1999-2006, MIYASAKA Masaru.
+; For conditions of distribution and use, see copyright notice in jsimdext.inc
+;
+; This file should be assembled with NASM (Netwide Assembler),
+; can *not* be assembled with Microsoft's MASM or any compatible
+; assembler (including Borland's Turbo Assembler).
+; NASM is available from http://nasm.sourceforge.net/ or
+; http://sourceforge.net/project/showfiles.php?group_id=6208
+;
+; This file contains a fast, not so accurate integer implementation of
+; the forward DCT (Discrete Cosine Transform). The following code is
+; based directly on the IJG's original jfdctfst.c; see the jfdctfst.c
+; for more details.
+;
+; [TAB8]
+
+%include "jsimdext.inc"
+%include "jdct.inc"
+
+; --------------------------------------------------------------------------
+
+%define CONST_BITS 8 ; 14 is also OK.
+
+%if CONST_BITS == 8
+F_0_382 equ 98 ; FIX(0.382683433)
+F_0_541 equ 139 ; FIX(0.541196100)
+F_0_707 equ 181 ; FIX(0.707106781)
+F_1_306 equ 334 ; FIX(1.306562965)
+%else
+; NASM cannot do compile-time arithmetic on floating-point constants.
+%define DESCALE(x,n) (((x)+(1<<((n)-1)))>>(n))
+F_0_382 equ DESCALE( 410903207, 30-CONST_BITS) ; FIX(0.382683433)
+F_0_541 equ DESCALE( 581104887, 30-CONST_BITS) ; FIX(0.541196100)
+F_0_707 equ DESCALE( 759250124, 30-CONST_BITS) ; FIX(0.707106781)
+F_1_306 equ DESCALE(1402911301, 30-CONST_BITS) ; FIX(1.306562965)
+%endif
+
+; --------------------------------------------------------------------------
+ SECTION SEG_CONST
+
+; PRE_MULTIPLY_SCALE_BITS <= 2 (to avoid overflow)
+; CONST_BITS + CONST_SHIFT + PRE_MULTIPLY_SCALE_BITS == 16 (for pmulhw)
+
+%define PRE_MULTIPLY_SCALE_BITS 2
+%define CONST_SHIFT (16 - PRE_MULTIPLY_SCALE_BITS - CONST_BITS)
+
+ alignz 32
+ global EXTN(jconst_fdct_ifast_sse2)
+
+EXTN(jconst_fdct_ifast_sse2):
+
+PW_F0707 times 8 dw F_0_707 << CONST_SHIFT
+PW_F0382 times 8 dw F_0_382 << CONST_SHIFT
+PW_F0541 times 8 dw F_0_541 << CONST_SHIFT
+PW_F1306 times 8 dw F_1_306 << CONST_SHIFT
+
+ alignz 32
+
+; --------------------------------------------------------------------------
+ SECTION SEG_TEXT
+ BITS 64
+;
+; Perform the forward DCT on one block of samples.
+;
+; GLOBAL(void)
+; jsimd_fdct_ifast_sse2 (DCTELEM *data)
+;
+
+; r10 = DCTELEM *data
+
+%define wk(i) rbp-(WK_NUM-(i))*SIZEOF_XMMWORD ; xmmword wk[WK_NUM]
+%define WK_NUM 2
+
+ align 32
+ global EXTN(jsimd_fdct_ifast_sse2)
+
+EXTN(jsimd_fdct_ifast_sse2):
+ push rbp
+ mov rax, rsp ; rax = original rbp
+ sub rsp, byte 4
+ and rsp, byte (-SIZEOF_XMMWORD) ; align to 128 bits
+ mov [rsp], rax
+ mov rbp, rsp ; rbp = aligned rbp
+ lea rsp, [wk(0)]
+ collect_args 1
+
+ ; ---- Pass 1: process rows.
+
+ mov rdx, r10 ; (DCTELEM *)
+
+ movdqa xmm0, XMMWORD [XMMBLOCK(0,0,rdx,SIZEOF_DCTELEM)]
+ movdqa xmm1, XMMWORD [XMMBLOCK(1,0,rdx,SIZEOF_DCTELEM)]
+ movdqa xmm2, XMMWORD [XMMBLOCK(2,0,rdx,SIZEOF_DCTELEM)]
+ movdqa xmm3, XMMWORD [XMMBLOCK(3,0,rdx,SIZEOF_DCTELEM)]
+
+ ; xmm0=(00 01 02 03 04 05 06 07), xmm2=(20 21 22 23 24 25 26 27)
+ ; xmm1=(10 11 12 13 14 15 16 17), xmm3=(30 31 32 33 34 35 36 37)
+
+ movdqa xmm4, xmm0 ; transpose coefficients(phase 1)
+ punpcklwd xmm0, xmm1 ; xmm0=(00 10 01 11 02 12 03 13)
+ punpckhwd xmm4, xmm1 ; xmm4=(04 14 05 15 06 16 07 17)
+ movdqa xmm5, xmm2 ; transpose coefficients(phase 1)
+ punpcklwd xmm2, xmm3 ; xmm2=(20 30 21 31 22 32 23 33)
+ punpckhwd xmm5, xmm3 ; xmm5=(24 34 25 35 26 36 27 37)
+
+ movdqa xmm6, XMMWORD [XMMBLOCK(4,0,rdx,SIZEOF_DCTELEM)]
+ movdqa xmm7, XMMWORD [XMMBLOCK(5,0,rdx,SIZEOF_DCTELEM)]
+ movdqa xmm1, XMMWORD [XMMBLOCK(6,0,rdx,SIZEOF_DCTELEM)]
+ movdqa xmm3, XMMWORD [XMMBLOCK(7,0,rdx,SIZEOF_DCTELEM)]
+
+ ; xmm6=( 4 12 20 28 36 44 52 60), xmm1=( 6 14 22 30 38 46 54 62)
+ ; xmm7=( 5 13 21 29 37 45 53 61), xmm3=( 7 15 23 31 39 47 55 63)
+
+ movdqa XMMWORD [wk(0)], xmm2 ; wk(0)=(20 30 21 31 22 32 23 33)
+ movdqa XMMWORD [wk(1)], xmm5 ; wk(1)=(24 34 25 35 26 36 27 37)
+
+ movdqa xmm2, xmm6 ; transpose coefficients(phase 1)
+ punpcklwd xmm6, xmm7 ; xmm6=(40 50 41 51 42 52 43 53)
+ punpckhwd xmm2, xmm7 ; xmm2=(44 54 45 55 46 56 47 57)
+ movdqa xmm5, xmm1 ; transpose coefficients(phase 1)
+ punpcklwd xmm1, xmm3 ; xmm1=(60 70 61 71 62 72 63 73)
+ punpckhwd xmm5, xmm3 ; xmm5=(64 74 65 75 66 76 67 77)
+
+ movdqa xmm7, xmm6 ; transpose coefficients(phase 2)
+ punpckldq xmm6, xmm1 ; xmm6=(40 50 60 70 41 51 61 71)
+ punpckhdq xmm7, xmm1 ; xmm7=(42 52 62 72 43 53 63 73)
+ movdqa xmm3, xmm2 ; transpose coefficients(phase 2)
+ punpckldq xmm2, xmm5 ; xmm2=(44 54 64 74 45 55 65 75)
+ punpckhdq xmm3, xmm5 ; xmm3=(46 56 66 76 47 57 67 77)
+
+ movdqa xmm1, XMMWORD [wk(0)] ; xmm1=(20 30 21 31 22 32 23 33)
+ movdqa xmm5, XMMWORD [wk(1)] ; xmm5=(24 34 25 35 26 36 27 37)
+ movdqa XMMWORD [wk(0)], xmm7 ; wk(0)=(42 52 62 72 43 53 63 73)
+ movdqa XMMWORD [wk(1)], xmm2 ; wk(1)=(44 54 64 74 45 55 65 75)
+
+ movdqa xmm7, xmm0 ; transpose coefficients(phase 2)
+ punpckldq xmm0, xmm1 ; xmm0=(00 10 20 30 01 11 21 31)
+ punpckhdq xmm7, xmm1 ; xmm7=(02 12 22 32 03 13 23 33)
+ movdqa xmm2, xmm4 ; transpose coefficients(phase 2)
+ punpckldq xmm4, xmm5 ; xmm4=(04 14 24 34 05 15 25 35)
+ punpckhdq xmm2, xmm5 ; xmm2=(06 16 26 36 07 17 27 37)
+
+ movdqa xmm1, xmm0 ; transpose coefficients(phase 3)
+ punpcklqdq xmm0, xmm6 ; xmm0=(00 10 20 30 40 50 60 70)=data0
+ punpckhqdq xmm1, xmm6 ; xmm1=(01 11 21 31 41 51 61 71)=data1
+ movdqa xmm5, xmm2 ; transpose coefficients(phase 3)
+ punpcklqdq xmm2, xmm3 ; xmm2=(06 16 26 36 46 56 66 76)=data6
+ punpckhqdq xmm5, xmm3 ; xmm5=(07 17 27 37 47 57 67 77)=data7
+
+ movdqa xmm6, xmm1
+ movdqa xmm3, xmm0
+ psubw xmm1, xmm2 ; xmm1=data1-data6=tmp6
+ psubw xmm0, xmm5 ; xmm0=data0-data7=tmp7
+ paddw xmm6, xmm2 ; xmm6=data1+data6=tmp1
+ paddw xmm3, xmm5 ; xmm3=data0+data7=tmp0
+
+ movdqa xmm2, XMMWORD [wk(0)] ; xmm2=(42 52 62 72 43 53 63 73)
+ movdqa xmm5, XMMWORD [wk(1)] ; xmm5=(44 54 64 74 45 55 65 75)
+ movdqa XMMWORD [wk(0)], xmm1 ; wk(0)=tmp6
+ movdqa XMMWORD [wk(1)], xmm0 ; wk(1)=tmp7
+
+ movdqa xmm1, xmm7 ; transpose coefficients(phase 3)
+ punpcklqdq xmm7, xmm2 ; xmm7=(02 12 22 32 42 52 62 72)=data2
+ punpckhqdq xmm1, xmm2 ; xmm1=(03 13 23 33 43 53 63 73)=data3
+ movdqa xmm0, xmm4 ; transpose coefficients(phase 3)
+ punpcklqdq xmm4, xmm5 ; xmm4=(04 14 24 34 44 54 64 74)=data4
+ punpckhqdq xmm0, xmm5 ; xmm0=(05 15 25 35 45 55 65 75)=data5
+
+ movdqa xmm2, xmm1
+ movdqa xmm5, xmm7
+ paddw xmm1, xmm4 ; xmm1=data3+data4=tmp3
+ paddw xmm7, xmm0 ; xmm7=data2+data5=tmp2
+ psubw xmm2, xmm4 ; xmm2=data3-data4=tmp4
+ psubw xmm5, xmm0 ; xmm5=data2-data5=tmp5
+
+ ; -- Even part
+
+ movdqa xmm4, xmm3
+ movdqa xmm0, xmm6
+ psubw xmm3, xmm1 ; xmm3=tmp13
+ psubw xmm6, xmm7 ; xmm6=tmp12
+ paddw xmm4, xmm1 ; xmm4=tmp10
+ paddw xmm0, xmm7 ; xmm0=tmp11
+
+ paddw xmm6, xmm3
+ psllw xmm6, PRE_MULTIPLY_SCALE_BITS
+ pmulhw xmm6, [rel PW_F0707] ; xmm6=z1
+
+ movdqa xmm1, xmm4
+ movdqa xmm7, xmm3
+ psubw xmm4, xmm0 ; xmm4=data4
+ psubw xmm3, xmm6 ; xmm3=data6
+ paddw xmm1, xmm0 ; xmm1=data0
+ paddw xmm7, xmm6 ; xmm7=data2
+
+ movdqa xmm0, XMMWORD [wk(0)] ; xmm0=tmp6
+ movdqa xmm6, XMMWORD [wk(1)] ; xmm6=tmp7
+ movdqa XMMWORD [wk(0)], xmm4 ; wk(0)=data4
+ movdqa XMMWORD [wk(1)], xmm3 ; wk(1)=data6
+
+ ; -- Odd part
+
+ paddw xmm2, xmm5 ; xmm2=tmp10
+ paddw xmm5, xmm0 ; xmm5=tmp11
+ paddw xmm0, xmm6 ; xmm0=tmp12, xmm6=tmp7
+
+ psllw xmm2, PRE_MULTIPLY_SCALE_BITS
+ psllw xmm0, PRE_MULTIPLY_SCALE_BITS
+
+ psllw xmm5, PRE_MULTIPLY_SCALE_BITS
+ pmulhw xmm5, [rel PW_F0707] ; xmm5=z3
+
+ movdqa xmm4, xmm2 ; xmm4=tmp10
+ psubw xmm2, xmm0
+ pmulhw xmm2, [rel PW_F0382] ; xmm2=z5
+ pmulhw xmm4, [rel PW_F0541] ; xmm4=MULTIPLY(tmp10,FIX_0_541196)
+ pmulhw xmm0, [rel PW_F1306] ; xmm0=MULTIPLY(tmp12,FIX_1_306562)
+ paddw xmm4, xmm2 ; xmm4=z2
+ paddw xmm0, xmm2 ; xmm0=z4
+
+ movdqa xmm3, xmm6
+ psubw xmm6, xmm5 ; xmm6=z13
+ paddw xmm3, xmm5 ; xmm3=z11
+
+ movdqa xmm2, xmm6
+ movdqa xmm5, xmm3
+ psubw xmm6, xmm4 ; xmm6=data3
+ psubw xmm3, xmm0 ; xmm3=data7
+ paddw xmm2, xmm4 ; xmm2=data5
+ paddw xmm5, xmm0 ; xmm5=data1
+
+ ; ---- Pass 2: process columns.
+
+ ; xmm1=(00 10 20 30 40 50 60 70), xmm7=(02 12 22 32 42 52 62 72)
+ ; xmm5=(01 11 21 31 41 51 61 71), xmm6=(03 13 23 33 43 53 63 73)
+
+ movdqa xmm4, xmm1 ; transpose coefficients(phase 1)
+ punpcklwd xmm1, xmm5 ; xmm1=(00 01 10 11 20 21 30 31)
+ punpckhwd xmm4, xmm5 ; xmm4=(40 41 50 51 60 61 70 71)
+ movdqa xmm0, xmm7 ; transpose coefficients(phase 1)
+ punpcklwd xmm7, xmm6 ; xmm7=(02 03 12 13 22 23 32 33)
+ punpckhwd xmm0, xmm6 ; xmm0=(42 43 52 53 62 63 72 73)
+
+ movdqa xmm5, XMMWORD [wk(0)] ; xmm5=col4
+ movdqa xmm6, XMMWORD [wk(1)] ; xmm6=col6
+
+ ; xmm5=(04 14 24 34 44 54 64 74), xmm6=(06 16 26 36 46 56 66 76)
+ ; xmm2=(05 15 25 35 45 55 65 75), xmm3=(07 17 27 37 47 57 67 77)
+
+ movdqa XMMWORD [wk(0)], xmm7 ; wk(0)=(02 03 12 13 22 23 32 33)
+ movdqa XMMWORD [wk(1)], xmm0 ; wk(1)=(42 43 52 53 62 63 72 73)
+
+ movdqa xmm7, xmm5 ; transpose coefficients(phase 1)
+ punpcklwd xmm5, xmm2 ; xmm5=(04 05 14 15 24 25 34 35)
+ punpckhwd xmm7, xmm2 ; xmm7=(44 45 54 55 64 65 74 75)
+ movdqa xmm0, xmm6 ; transpose coefficients(phase 1)
+ punpcklwd xmm6, xmm3 ; xmm6=(06 07 16 17 26 27 36 37)
+ punpckhwd xmm0, xmm3 ; xmm0=(46 47 56 57 66 67 76 77)
+
+ movdqa xmm2, xmm5 ; transpose coefficients(phase 2)
+ punpckldq xmm5, xmm6 ; xmm5=(04 05 06 07 14 15 16 17)
+ punpckhdq xmm2, xmm6 ; xmm2=(24 25 26 27 34 35 36 37)
+ movdqa xmm3, xmm7 ; transpose coefficients(phase 2)
+ punpckldq xmm7, xmm0 ; xmm7=(44 45 46 47 54 55 56 57)
+ punpckhdq xmm3, xmm0 ; xmm3=(64 65 66 67 74 75 76 77)
+
+ movdqa xmm6, XMMWORD [wk(0)] ; xmm6=(02 03 12 13 22 23 32 33)
+ movdqa xmm0, XMMWORD [wk(1)] ; xmm0=(42 43 52 53 62 63 72 73)
+ movdqa XMMWORD [wk(0)], xmm2 ; wk(0)=(24 25 26 27 34 35 36 37)
+ movdqa XMMWORD [wk(1)], xmm7 ; wk(1)=(44 45 46 47 54 55 56 57)
+
+ movdqa xmm2, xmm1 ; transpose coefficients(phase 2)
+ punpckldq xmm1, xmm6 ; xmm1=(00 01 02 03 10 11 12 13)
+ punpckhdq xmm2, xmm6 ; xmm2=(20 21 22 23 30 31 32 33)
+ movdqa xmm7, xmm4 ; transpose coefficients(phase 2)
+ punpckldq xmm4, xmm0 ; xmm4=(40 41 42 43 50 51 52 53)
+ punpckhdq xmm7, xmm0 ; xmm7=(60 61 62 63 70 71 72 73)
+
+ movdqa xmm6, xmm1 ; transpose coefficients(phase 3)
+ punpcklqdq xmm1, xmm5 ; xmm1=(00 01 02 03 04 05 06 07)=data0
+ punpckhqdq xmm6, xmm5 ; xmm6=(10 11 12 13 14 15 16 17)=data1
+ movdqa xmm0, xmm7 ; transpose coefficients(phase 3)
+ punpcklqdq xmm7, xmm3 ; xmm7=(60 61 62 63 64 65 66 67)=data6
+ punpckhqdq xmm0, xmm3 ; xmm0=(70 71 72 73 74 75 76 77)=data7
+
+ movdqa xmm5, xmm6
+ movdqa xmm3, xmm1
+ psubw xmm6, xmm7 ; xmm6=data1-data6=tmp6
+ psubw xmm1, xmm0 ; xmm1=data0-data7=tmp7
+ paddw xmm5, xmm7 ; xmm5=data1+data6=tmp1
+ paddw xmm3, xmm0 ; xmm3=data0+data7=tmp0
+
+ movdqa xmm7, XMMWORD [wk(0)] ; xmm7=(24 25 26 27 34 35 36 37)
+ movdqa xmm0, XMMWORD [wk(1)] ; xmm0=(44 45 46 47 54 55 56 57)
+ movdqa XMMWORD [wk(0)], xmm6 ; wk(0)=tmp6
+ movdqa XMMWORD [wk(1)], xmm1 ; wk(1)=tmp7
+
+ movdqa xmm6, xmm2 ; transpose coefficients(phase 3)
+ punpcklqdq xmm2, xmm7 ; xmm2=(20 21 22 23 24 25 26 27)=data2
+ punpckhqdq xmm6, xmm7 ; xmm6=(30 31 32 33 34 35 36 37)=data3
+ movdqa xmm1, xmm4 ; transpose coefficients(phase 3)
+ punpcklqdq xmm4, xmm0 ; xmm4=(40 41 42 43 44 45 46 47)=data4
+ punpckhqdq xmm1, xmm0 ; xmm1=(50 51 52 53 54 55 56 57)=data5
+
+ movdqa xmm7, xmm6
+ movdqa xmm0, xmm2
+ paddw xmm6, xmm4 ; xmm6=data3+data4=tmp3
+ paddw xmm2, xmm1 ; xmm2=data2+data5=tmp2
+ psubw xmm7, xmm4 ; xmm7=data3-data4=tmp4
+ psubw xmm0, xmm1 ; xmm0=data2-data5=tmp5
+
+ ; -- Even part
+
+ movdqa xmm4, xmm3
+ movdqa xmm1, xmm5
+ psubw xmm3, xmm6 ; xmm3=tmp13
+ psubw xmm5, xmm2 ; xmm5=tmp12
+ paddw xmm4, xmm6 ; xmm4=tmp10
+ paddw xmm1, xmm2 ; xmm1=tmp11
+
+ paddw xmm5, xmm3
+ psllw xmm5, PRE_MULTIPLY_SCALE_BITS
+ pmulhw xmm5, [rel PW_F0707] ; xmm5=z1
+
+ movdqa xmm6, xmm4
+ movdqa xmm2, xmm3
+ psubw xmm4, xmm1 ; xmm4=data4
+ psubw xmm3, xmm5 ; xmm3=data6
+ paddw xmm6, xmm1 ; xmm6=data0
+ paddw xmm2, xmm5 ; xmm2=data2
+
+ movdqa XMMWORD [XMMBLOCK(4,0,rdx,SIZEOF_DCTELEM)], xmm4
+ movdqa XMMWORD [XMMBLOCK(6,0,rdx,SIZEOF_DCTELEM)], xmm3
+ movdqa XMMWORD [XMMBLOCK(0,0,rdx,SIZEOF_DCTELEM)], xmm6
+ movdqa XMMWORD [XMMBLOCK(2,0,rdx,SIZEOF_DCTELEM)], xmm2
+
+ ; -- Odd part
+
+ movdqa xmm1, XMMWORD [wk(0)] ; xmm1=tmp6
+ movdqa xmm5, XMMWORD [wk(1)] ; xmm5=tmp7
+
+ paddw xmm7, xmm0 ; xmm7=tmp10
+ paddw xmm0, xmm1 ; xmm0=tmp11
+ paddw xmm1, xmm5 ; xmm1=tmp12, xmm5=tmp7
+
+ psllw xmm7, PRE_MULTIPLY_SCALE_BITS
+ psllw xmm1, PRE_MULTIPLY_SCALE_BITS
+
+ psllw xmm0, PRE_MULTIPLY_SCALE_BITS
+ pmulhw xmm0, [rel PW_F0707] ; xmm0=z3
+
+ movdqa xmm4, xmm7 ; xmm4=tmp10
+ psubw xmm7, xmm1
+ pmulhw xmm7, [rel PW_F0382] ; xmm7=z5
+ pmulhw xmm4, [rel PW_F0541] ; xmm4=MULTIPLY(tmp10,FIX_0_541196)
+ pmulhw xmm1, [rel PW_F1306] ; xmm1=MULTIPLY(tmp12,FIX_1_306562)
+ paddw xmm4, xmm7 ; xmm4=z2
+ paddw xmm1, xmm7 ; xmm1=z4
+
+ movdqa xmm3, xmm5
+ psubw xmm5, xmm0 ; xmm5=z13
+ paddw xmm3, xmm0 ; xmm3=z11
+
+ movdqa xmm6, xmm5
+ movdqa xmm2, xmm3
+ psubw xmm5, xmm4 ; xmm5=data3
+ psubw xmm3, xmm1 ; xmm3=data7
+ paddw xmm6, xmm4 ; xmm6=data5
+ paddw xmm2, xmm1 ; xmm2=data1
+
+ movdqa XMMWORD [XMMBLOCK(3,0,rdx,SIZEOF_DCTELEM)], xmm5
+ movdqa XMMWORD [XMMBLOCK(7,0,rdx,SIZEOF_DCTELEM)], xmm3
+ movdqa XMMWORD [XMMBLOCK(5,0,rdx,SIZEOF_DCTELEM)], xmm6
+ movdqa XMMWORD [XMMBLOCK(1,0,rdx,SIZEOF_DCTELEM)], xmm2
+
+ uncollect_args 1
+ mov rsp, rbp ; rsp <- aligned rbp
+ pop rsp ; rsp <- original rbp
+ pop rbp
+ ret
+
+; For some reason, the OS X linker does not honor the request to align the
+; segment unless we do this.
+ align 32
diff --git a/simd/x86_64/jfdctint-sse2.asm b/simd/x86_64/jfdctint-sse2.asm
new file mode 100644
index 0000000..cdaa56f
--- /dev/null
+++ b/simd/x86_64/jfdctint-sse2.asm
@@ -0,0 +1,621 @@
+;
+; jfdctint.asm - accurate integer FDCT (64-bit SSE2)
+;
+; Copyright 2009 Pierre Ossman <ossman@cendio.se> for Cendio AB
+; Copyright (C) 2009, 2016, D. R. Commander.
+;
+; Based on the x86 SIMD extension for IJG JPEG library
+; Copyright (C) 1999-2006, MIYASAKA Masaru.
+; For conditions of distribution and use, see copyright notice in jsimdext.inc
+;
+; This file should be assembled with NASM (Netwide Assembler),
+; can *not* be assembled with Microsoft's MASM or any compatible
+; assembler (including Borland's Turbo Assembler).
+; NASM is available from http://nasm.sourceforge.net/ or
+; http://sourceforge.net/project/showfiles.php?group_id=6208
+;
+; This file contains a slow-but-accurate integer implementation of the
+; forward DCT (Discrete Cosine Transform). The following code is based
+; directly on the IJG's original jfdctint.c; see the jfdctint.c for
+; more details.
+;
+; [TAB8]
+
+%include "jsimdext.inc"
+%include "jdct.inc"
+
+; --------------------------------------------------------------------------
+
+%define CONST_BITS 13
+%define PASS1_BITS 2
+
+%define DESCALE_P1 (CONST_BITS-PASS1_BITS)
+%define DESCALE_P2 (CONST_BITS+PASS1_BITS)
+
+%if CONST_BITS == 13
+F_0_298 equ 2446 ; FIX(0.298631336)
+F_0_390 equ 3196 ; FIX(0.390180644)
+F_0_541 equ 4433 ; FIX(0.541196100)
+F_0_765 equ 6270 ; FIX(0.765366865)
+F_0_899 equ 7373 ; FIX(0.899976223)
+F_1_175 equ 9633 ; FIX(1.175875602)
+F_1_501 equ 12299 ; FIX(1.501321110)
+F_1_847 equ 15137 ; FIX(1.847759065)
+F_1_961 equ 16069 ; FIX(1.961570560)
+F_2_053 equ 16819 ; FIX(2.053119869)
+F_2_562 equ 20995 ; FIX(2.562915447)
+F_3_072 equ 25172 ; FIX(3.072711026)
+%else
+; NASM cannot do compile-time arithmetic on floating-point constants.
+%define DESCALE(x,n) (((x)+(1<<((n)-1)))>>(n))
+F_0_298 equ DESCALE( 320652955, 30-CONST_BITS) ; FIX(0.298631336)
+F_0_390 equ DESCALE( 418953276, 30-CONST_BITS) ; FIX(0.390180644)
+F_0_541 equ DESCALE( 581104887, 30-CONST_BITS) ; FIX(0.541196100)
+F_0_765 equ DESCALE( 821806413, 30-CONST_BITS) ; FIX(0.765366865)
+F_0_899 equ DESCALE( 966342111, 30-CONST_BITS) ; FIX(0.899976223)
+F_1_175 equ DESCALE(1262586813, 30-CONST_BITS) ; FIX(1.175875602)
+F_1_501 equ DESCALE(1612031267, 30-CONST_BITS) ; FIX(1.501321110)
+F_1_847 equ DESCALE(1984016188, 30-CONST_BITS) ; FIX(1.847759065)
+F_1_961 equ DESCALE(2106220350, 30-CONST_BITS) ; FIX(1.961570560)
+F_2_053 equ DESCALE(2204520673, 30-CONST_BITS) ; FIX(2.053119869)
+F_2_562 equ DESCALE(2751909506, 30-CONST_BITS) ; FIX(2.562915447)
+F_3_072 equ DESCALE(3299298341, 30-CONST_BITS) ; FIX(3.072711026)
+%endif
+
+; --------------------------------------------------------------------------
+ SECTION SEG_CONST
+
+ alignz 32
+ global EXTN(jconst_fdct_islow_sse2)
+
+EXTN(jconst_fdct_islow_sse2):
+
+PW_F130_F054 times 4 dw (F_0_541+F_0_765), F_0_541
+PW_F054_MF130 times 4 dw F_0_541, (F_0_541-F_1_847)
+PW_MF078_F117 times 4 dw (F_1_175-F_1_961), F_1_175
+PW_F117_F078 times 4 dw F_1_175, (F_1_175-F_0_390)
+PW_MF060_MF089 times 4 dw (F_0_298-F_0_899),-F_0_899
+PW_MF089_F060 times 4 dw -F_0_899, (F_1_501-F_0_899)
+PW_MF050_MF256 times 4 dw (F_2_053-F_2_562),-F_2_562
+PW_MF256_F050 times 4 dw -F_2_562, (F_3_072-F_2_562)
+PD_DESCALE_P1 times 4 dd 1 << (DESCALE_P1-1)
+PD_DESCALE_P2 times 4 dd 1 << (DESCALE_P2-1)
+PW_DESCALE_P2X times 8 dw 1 << (PASS1_BITS-1)
+
+ alignz 32
+
+; --------------------------------------------------------------------------
+ SECTION SEG_TEXT
+ BITS 64
+;
+; Perform the forward DCT on one block of samples.
+;
+; GLOBAL(void)
+; jsimd_fdct_islow_sse2 (DCTELEM *data)
+;
+
+; r10 = DCTELEM *data
+
+%define wk(i) rbp-(WK_NUM-(i))*SIZEOF_XMMWORD ; xmmword wk[WK_NUM]
+%define WK_NUM 6
+
+ align 32
+ global EXTN(jsimd_fdct_islow_sse2)
+
+EXTN(jsimd_fdct_islow_sse2):
+ push rbp
+ mov rax, rsp ; rax = original rbp
+ sub rsp, byte 4
+ and rsp, byte (-SIZEOF_XMMWORD) ; align to 128 bits
+ mov [rsp], rax
+ mov rbp, rsp ; rbp = aligned rbp
+ lea rsp, [wk(0)]
+ collect_args 1
+
+ ; ---- Pass 1: process rows.
+
+ mov rdx, r10 ; (DCTELEM *)
+
+ movdqa xmm0, XMMWORD [XMMBLOCK(0,0,rdx,SIZEOF_DCTELEM)]
+ movdqa xmm1, XMMWORD [XMMBLOCK(1,0,rdx,SIZEOF_DCTELEM)]
+ movdqa xmm2, XMMWORD [XMMBLOCK(2,0,rdx,SIZEOF_DCTELEM)]
+ movdqa xmm3, XMMWORD [XMMBLOCK(3,0,rdx,SIZEOF_DCTELEM)]
+
+ ; xmm0=(00 01 02 03 04 05 06 07), xmm2=(20 21 22 23 24 25 26 27)
+ ; xmm1=(10 11 12 13 14 15 16 17), xmm3=(30 31 32 33 34 35 36 37)
+
+ movdqa xmm4, xmm0 ; transpose coefficients(phase 1)
+ punpcklwd xmm0, xmm1 ; xmm0=(00 10 01 11 02 12 03 13)
+ punpckhwd xmm4, xmm1 ; xmm4=(04 14 05 15 06 16 07 17)
+ movdqa xmm5, xmm2 ; transpose coefficients(phase 1)
+ punpcklwd xmm2, xmm3 ; xmm2=(20 30 21 31 22 32 23 33)
+ punpckhwd xmm5, xmm3 ; xmm5=(24 34 25 35 26 36 27 37)
+
+ movdqa xmm6, XMMWORD [XMMBLOCK(4,0,rdx,SIZEOF_DCTELEM)]
+ movdqa xmm7, XMMWORD [XMMBLOCK(5,0,rdx,SIZEOF_DCTELEM)]
+ movdqa xmm1, XMMWORD [XMMBLOCK(6,0,rdx,SIZEOF_DCTELEM)]
+ movdqa xmm3, XMMWORD [XMMBLOCK(7,0,rdx,SIZEOF_DCTELEM)]
+
+ ; xmm6=( 4 12 20 28 36 44 52 60), xmm1=( 6 14 22 30 38 46 54 62)
+ ; xmm7=( 5 13 21 29 37 45 53 61), xmm3=( 7 15 23 31 39 47 55 63)
+
+ movdqa XMMWORD [wk(0)], xmm2 ; wk(0)=(20 30 21 31 22 32 23 33)
+ movdqa XMMWORD [wk(1)], xmm5 ; wk(1)=(24 34 25 35 26 36 27 37)
+
+ movdqa xmm2, xmm6 ; transpose coefficients(phase 1)
+ punpcklwd xmm6, xmm7 ; xmm6=(40 50 41 51 42 52 43 53)
+ punpckhwd xmm2, xmm7 ; xmm2=(44 54 45 55 46 56 47 57)
+ movdqa xmm5, xmm1 ; transpose coefficients(phase 1)
+ punpcklwd xmm1, xmm3 ; xmm1=(60 70 61 71 62 72 63 73)
+ punpckhwd xmm5, xmm3 ; xmm5=(64 74 65 75 66 76 67 77)
+
+ movdqa xmm7, xmm6 ; transpose coefficients(phase 2)
+ punpckldq xmm6, xmm1 ; xmm6=(40 50 60 70 41 51 61 71)
+ punpckhdq xmm7, xmm1 ; xmm7=(42 52 62 72 43 53 63 73)
+ movdqa xmm3, xmm2 ; transpose coefficients(phase 2)
+ punpckldq xmm2, xmm5 ; xmm2=(44 54 64 74 45 55 65 75)
+ punpckhdq xmm3, xmm5 ; xmm3=(46 56 66 76 47 57 67 77)
+
+ movdqa xmm1, XMMWORD [wk(0)] ; xmm1=(20 30 21 31 22 32 23 33)
+ movdqa xmm5, XMMWORD [wk(1)] ; xmm5=(24 34 25 35 26 36 27 37)
+ movdqa XMMWORD [wk(2)], xmm7 ; wk(2)=(42 52 62 72 43 53 63 73)
+ movdqa XMMWORD [wk(3)], xmm2 ; wk(3)=(44 54 64 74 45 55 65 75)
+
+ movdqa xmm7, xmm0 ; transpose coefficients(phase 2)
+ punpckldq xmm0, xmm1 ; xmm0=(00 10 20 30 01 11 21 31)
+ punpckhdq xmm7, xmm1 ; xmm7=(02 12 22 32 03 13 23 33)
+ movdqa xmm2, xmm4 ; transpose coefficients(phase 2)
+ punpckldq xmm4, xmm5 ; xmm4=(04 14 24 34 05 15 25 35)
+ punpckhdq xmm2, xmm5 ; xmm2=(06 16 26 36 07 17 27 37)
+
+ movdqa xmm1, xmm0 ; transpose coefficients(phase 3)
+ punpcklqdq xmm0, xmm6 ; xmm0=(00 10 20 30 40 50 60 70)=data0
+ punpckhqdq xmm1, xmm6 ; xmm1=(01 11 21 31 41 51 61 71)=data1
+ movdqa xmm5, xmm2 ; transpose coefficients(phase 3)
+ punpcklqdq xmm2, xmm3 ; xmm2=(06 16 26 36 46 56 66 76)=data6
+ punpckhqdq xmm5, xmm3 ; xmm5=(07 17 27 37 47 57 67 77)=data7
+
+ movdqa xmm6, xmm1
+ movdqa xmm3, xmm0
+ psubw xmm1, xmm2 ; xmm1=data1-data6=tmp6
+ psubw xmm0, xmm5 ; xmm0=data0-data7=tmp7
+ paddw xmm6, xmm2 ; xmm6=data1+data6=tmp1
+ paddw xmm3, xmm5 ; xmm3=data0+data7=tmp0
+
+ movdqa xmm2, XMMWORD [wk(2)] ; xmm2=(42 52 62 72 43 53 63 73)
+ movdqa xmm5, XMMWORD [wk(3)] ; xmm5=(44 54 64 74 45 55 65 75)
+ movdqa XMMWORD [wk(0)], xmm1 ; wk(0)=tmp6
+ movdqa XMMWORD [wk(1)], xmm0 ; wk(1)=tmp7
+
+ movdqa xmm1, xmm7 ; transpose coefficients(phase 3)
+ punpcklqdq xmm7, xmm2 ; xmm7=(02 12 22 32 42 52 62 72)=data2
+ punpckhqdq xmm1, xmm2 ; xmm1=(03 13 23 33 43 53 63 73)=data3
+ movdqa xmm0, xmm4 ; transpose coefficients(phase 3)
+ punpcklqdq xmm4, xmm5 ; xmm4=(04 14 24 34 44 54 64 74)=data4
+ punpckhqdq xmm0, xmm5 ; xmm0=(05 15 25 35 45 55 65 75)=data5
+
+ movdqa xmm2, xmm1
+ movdqa xmm5, xmm7
+ paddw xmm1, xmm4 ; xmm1=data3+data4=tmp3
+ paddw xmm7, xmm0 ; xmm7=data2+data5=tmp2
+ psubw xmm2, xmm4 ; xmm2=data3-data4=tmp4
+ psubw xmm5, xmm0 ; xmm5=data2-data5=tmp5
+
+ ; -- Even part
+
+ movdqa xmm4, xmm3
+ movdqa xmm0, xmm6
+ paddw xmm3, xmm1 ; xmm3=tmp10
+ paddw xmm6, xmm7 ; xmm6=tmp11
+ psubw xmm4, xmm1 ; xmm4=tmp13
+ psubw xmm0, xmm7 ; xmm0=tmp12
+
+ movdqa xmm1, xmm3
+ paddw xmm3, xmm6 ; xmm3=tmp10+tmp11
+ psubw xmm1, xmm6 ; xmm1=tmp10-tmp11
+
+ psllw xmm3, PASS1_BITS ; xmm3=data0
+ psllw xmm1, PASS1_BITS ; xmm1=data4
+
+ movdqa XMMWORD [wk(2)], xmm3 ; wk(2)=data0
+ movdqa XMMWORD [wk(3)], xmm1 ; wk(3)=data4
+
+ ; (Original)
+ ; z1 = (tmp12 + tmp13) * 0.541196100;
+ ; data2 = z1 + tmp13 * 0.765366865;
+ ; data6 = z1 + tmp12 * -1.847759065;
+ ;
+ ; (This implementation)
+ ; data2 = tmp13 * (0.541196100 + 0.765366865) + tmp12 * 0.541196100;
+ ; data6 = tmp13 * 0.541196100 + tmp12 * (0.541196100 - 1.847759065);
+
+ movdqa xmm7, xmm4 ; xmm4=tmp13
+ movdqa xmm6, xmm4
+ punpcklwd xmm7, xmm0 ; xmm0=tmp12
+ punpckhwd xmm6, xmm0
+ movdqa xmm4, xmm7
+ movdqa xmm0, xmm6
+ pmaddwd xmm7, [rel PW_F130_F054] ; xmm7=data2L
+ pmaddwd xmm6, [rel PW_F130_F054] ; xmm6=data2H
+ pmaddwd xmm4, [rel PW_F054_MF130] ; xmm4=data6L
+ pmaddwd xmm0, [rel PW_F054_MF130] ; xmm0=data6H
+
+ paddd xmm7, [rel PD_DESCALE_P1]
+ paddd xmm6, [rel PD_DESCALE_P1]
+ psrad xmm7, DESCALE_P1
+ psrad xmm6, DESCALE_P1
+ paddd xmm4, [rel PD_DESCALE_P1]
+ paddd xmm0, [rel PD_DESCALE_P1]
+ psrad xmm4, DESCALE_P1
+ psrad xmm0, DESCALE_P1
+
+ packssdw xmm7, xmm6 ; xmm7=data2
+ packssdw xmm4, xmm0 ; xmm4=data6
+
+ movdqa XMMWORD [wk(4)], xmm7 ; wk(4)=data2
+ movdqa XMMWORD [wk(5)], xmm4 ; wk(5)=data6
+
+ ; -- Odd part
+
+ movdqa xmm3, XMMWORD [wk(0)] ; xmm3=tmp6
+ movdqa xmm1, XMMWORD [wk(1)] ; xmm1=tmp7
+
+ movdqa xmm6, xmm2 ; xmm2=tmp4
+ movdqa xmm0, xmm5 ; xmm5=tmp5
+ paddw xmm6, xmm3 ; xmm6=z3
+ paddw xmm0, xmm1 ; xmm0=z4
+
+ ; (Original)
+ ; z5 = (z3 + z4) * 1.175875602;
+ ; z3 = z3 * -1.961570560; z4 = z4 * -0.390180644;
+ ; z3 += z5; z4 += z5;
+ ;
+ ; (This implementation)
+ ; z3 = z3 * (1.175875602 - 1.961570560) + z4 * 1.175875602;
+ ; z4 = z3 * 1.175875602 + z4 * (1.175875602 - 0.390180644);
+
+ movdqa xmm7, xmm6
+ movdqa xmm4, xmm6
+ punpcklwd xmm7, xmm0
+ punpckhwd xmm4, xmm0
+ movdqa xmm6, xmm7
+ movdqa xmm0, xmm4
+ pmaddwd xmm7, [rel PW_MF078_F117] ; xmm7=z3L
+ pmaddwd xmm4, [rel PW_MF078_F117] ; xmm4=z3H
+ pmaddwd xmm6, [rel PW_F117_F078] ; xmm6=z4L
+ pmaddwd xmm0, [rel PW_F117_F078] ; xmm0=z4H
+
+ movdqa XMMWORD [wk(0)], xmm7 ; wk(0)=z3L
+ movdqa XMMWORD [wk(1)], xmm4 ; wk(1)=z3H
+
+ ; (Original)
+ ; z1 = tmp4 + tmp7; z2 = tmp5 + tmp6;
+ ; tmp4 = tmp4 * 0.298631336; tmp5 = tmp5 * 2.053119869;
+ ; tmp6 = tmp6 * 3.072711026; tmp7 = tmp7 * 1.501321110;
+ ; z1 = z1 * -0.899976223; z2 = z2 * -2.562915447;
+ ; data7 = tmp4 + z1 + z3; data5 = tmp5 + z2 + z4;
+ ; data3 = tmp6 + z2 + z3; data1 = tmp7 + z1 + z4;
+ ;
+ ; (This implementation)
+ ; tmp4 = tmp4 * (0.298631336 - 0.899976223) + tmp7 * -0.899976223;
+ ; tmp5 = tmp5 * (2.053119869 - 2.562915447) + tmp6 * -2.562915447;
+ ; tmp6 = tmp5 * -2.562915447 + tmp6 * (3.072711026 - 2.562915447);
+ ; tmp7 = tmp4 * -0.899976223 + tmp7 * (1.501321110 - 0.899976223);
+ ; data7 = tmp4 + z3; data5 = tmp5 + z4;
+ ; data3 = tmp6 + z3; data1 = tmp7 + z4;
+
+ movdqa xmm7, xmm2
+ movdqa xmm4, xmm2
+ punpcklwd xmm7, xmm1
+ punpckhwd xmm4, xmm1
+ movdqa xmm2, xmm7
+ movdqa xmm1, xmm4
+ pmaddwd xmm7, [rel PW_MF060_MF089] ; xmm7=tmp4L
+ pmaddwd xmm4, [rel PW_MF060_MF089] ; xmm4=tmp4H
+ pmaddwd xmm2, [rel PW_MF089_F060] ; xmm2=tmp7L
+ pmaddwd xmm1, [rel PW_MF089_F060] ; xmm1=tmp7H
+
+ paddd xmm7, XMMWORD [wk(0)] ; xmm7=data7L
+ paddd xmm4, XMMWORD [wk(1)] ; xmm4=data7H
+ paddd xmm2, xmm6 ; xmm2=data1L
+ paddd xmm1, xmm0 ; xmm1=data1H
+
+ paddd xmm7, [rel PD_DESCALE_P1]
+ paddd xmm4, [rel PD_DESCALE_P1]
+ psrad xmm7, DESCALE_P1
+ psrad xmm4, DESCALE_P1
+ paddd xmm2, [rel PD_DESCALE_P1]
+ paddd xmm1, [rel PD_DESCALE_P1]
+ psrad xmm2, DESCALE_P1
+ psrad xmm1, DESCALE_P1
+
+ packssdw xmm7, xmm4 ; xmm7=data7
+ packssdw xmm2, xmm1 ; xmm2=data1
+
+ movdqa xmm4, xmm5
+ movdqa xmm1, xmm5
+ punpcklwd xmm4, xmm3
+ punpckhwd xmm1, xmm3
+ movdqa xmm5, xmm4
+ movdqa xmm3, xmm1
+ pmaddwd xmm4, [rel PW_MF050_MF256] ; xmm4=tmp5L
+ pmaddwd xmm1, [rel PW_MF050_MF256] ; xmm1=tmp5H
+ pmaddwd xmm5, [rel PW_MF256_F050] ; xmm5=tmp6L
+ pmaddwd xmm3, [rel PW_MF256_F050] ; xmm3=tmp6H
+
+ paddd xmm4, xmm6 ; xmm4=data5L
+ paddd xmm1, xmm0 ; xmm1=data5H
+ paddd xmm5, XMMWORD [wk(0)] ; xmm5=data3L
+ paddd xmm3, XMMWORD [wk(1)] ; xmm3=data3H
+
+ paddd xmm4, [rel PD_DESCALE_P1]
+ paddd xmm1, [rel PD_DESCALE_P1]
+ psrad xmm4, DESCALE_P1
+ psrad xmm1, DESCALE_P1
+ paddd xmm5, [rel PD_DESCALE_P1]
+ paddd xmm3, [rel PD_DESCALE_P1]
+ psrad xmm5, DESCALE_P1
+ psrad xmm3, DESCALE_P1
+
+ packssdw xmm4, xmm1 ; xmm4=data5
+ packssdw xmm5, xmm3 ; xmm5=data3
+
+ ; ---- Pass 2: process columns.
+
+ movdqa xmm6, XMMWORD [wk(2)] ; xmm6=col0
+ movdqa xmm0, XMMWORD [wk(4)] ; xmm0=col2
+
+ ; xmm6=(00 10 20 30 40 50 60 70), xmm0=(02 12 22 32 42 52 62 72)
+ ; xmm2=(01 11 21 31 41 51 61 71), xmm5=(03 13 23 33 43 53 63 73)
+
+ movdqa xmm1, xmm6 ; transpose coefficients(phase 1)
+ punpcklwd xmm6, xmm2 ; xmm6=(00 01 10 11 20 21 30 31)
+ punpckhwd xmm1, xmm2 ; xmm1=(40 41 50 51 60 61 70 71)
+ movdqa xmm3, xmm0 ; transpose coefficients(phase 1)
+ punpcklwd xmm0, xmm5 ; xmm0=(02 03 12 13 22 23 32 33)
+ punpckhwd xmm3, xmm5 ; xmm3=(42 43 52 53 62 63 72 73)
+
+ movdqa xmm2, XMMWORD [wk(3)] ; xmm2=col4
+ movdqa xmm5, XMMWORD [wk(5)] ; xmm5=col6
+
+ ; xmm2=(04 14 24 34 44 54 64 74), xmm5=(06 16 26 36 46 56 66 76)
+ ; xmm4=(05 15 25 35 45 55 65 75), xmm7=(07 17 27 37 47 57 67 77)
+
+ movdqa XMMWORD [wk(0)], xmm0 ; wk(0)=(02 03 12 13 22 23 32 33)
+ movdqa XMMWORD [wk(1)], xmm3 ; wk(1)=(42 43 52 53 62 63 72 73)
+
+ movdqa xmm0, xmm2 ; transpose coefficients(phase 1)
+ punpcklwd xmm2, xmm4 ; xmm2=(04 05 14 15 24 25 34 35)
+ punpckhwd xmm0, xmm4 ; xmm0=(44 45 54 55 64 65 74 75)
+ movdqa xmm3, xmm5 ; transpose coefficients(phase 1)
+ punpcklwd xmm5, xmm7 ; xmm5=(06 07 16 17 26 27 36 37)
+ punpckhwd xmm3, xmm7 ; xmm3=(46 47 56 57 66 67 76 77)
+
+ movdqa xmm4, xmm2 ; transpose coefficients(phase 2)
+ punpckldq xmm2, xmm5 ; xmm2=(04 05 06 07 14 15 16 17)
+ punpckhdq xmm4, xmm5 ; xmm4=(24 25 26 27 34 35 36 37)
+ movdqa xmm7, xmm0 ; transpose coefficients(phase 2)
+ punpckldq xmm0, xmm3 ; xmm0=(44 45 46 47 54 55 56 57)
+ punpckhdq xmm7, xmm3 ; xmm7=(64 65 66 67 74 75 76 77)
+
+ movdqa xmm5, XMMWORD [wk(0)] ; xmm5=(02 03 12 13 22 23 32 33)
+ movdqa xmm3, XMMWORD [wk(1)] ; xmm3=(42 43 52 53 62 63 72 73)
+ movdqa XMMWORD [wk(2)], xmm4 ; wk(2)=(24 25 26 27 34 35 36 37)
+ movdqa XMMWORD [wk(3)], xmm0 ; wk(3)=(44 45 46 47 54 55 56 57)
+
+ movdqa xmm4, xmm6 ; transpose coefficients(phase 2)
+ punpckldq xmm6, xmm5 ; xmm6=(00 01 02 03 10 11 12 13)
+ punpckhdq xmm4, xmm5 ; xmm4=(20 21 22 23 30 31 32 33)
+ movdqa xmm0, xmm1 ; transpose coefficients(phase 2)
+ punpckldq xmm1, xmm3 ; xmm1=(40 41 42 43 50 51 52 53)
+ punpckhdq xmm0, xmm3 ; xmm0=(60 61 62 63 70 71 72 73)
+
+ movdqa xmm5, xmm6 ; transpose coefficients(phase 3)
+ punpcklqdq xmm6, xmm2 ; xmm6=(00 01 02 03 04 05 06 07)=data0
+ punpckhqdq xmm5, xmm2 ; xmm5=(10 11 12 13 14 15 16 17)=data1
+ movdqa xmm3, xmm0 ; transpose coefficients(phase 3)
+ punpcklqdq xmm0, xmm7 ; xmm0=(60 61 62 63 64 65 66 67)=data6
+ punpckhqdq xmm3, xmm7 ; xmm3=(70 71 72 73 74 75 76 77)=data7
+
+ movdqa xmm2, xmm5
+ movdqa xmm7, xmm6
+ psubw xmm5, xmm0 ; xmm5=data1-data6=tmp6
+ psubw xmm6, xmm3 ; xmm6=data0-data7=tmp7
+ paddw xmm2, xmm0 ; xmm2=data1+data6=tmp1
+ paddw xmm7, xmm3 ; xmm7=data0+data7=tmp0
+
+ movdqa xmm0, XMMWORD [wk(2)] ; xmm0=(24 25 26 27 34 35 36 37)
+ movdqa xmm3, XMMWORD [wk(3)] ; xmm3=(44 45 46 47 54 55 56 57)
+ movdqa XMMWORD [wk(0)], xmm5 ; wk(0)=tmp6
+ movdqa XMMWORD [wk(1)], xmm6 ; wk(1)=tmp7
+
+ movdqa xmm5, xmm4 ; transpose coefficients(phase 3)
+ punpcklqdq xmm4, xmm0 ; xmm4=(20 21 22 23 24 25 26 27)=data2
+ punpckhqdq xmm5, xmm0 ; xmm5=(30 31 32 33 34 35 36 37)=data3
+ movdqa xmm6, xmm1 ; transpose coefficients(phase 3)
+ punpcklqdq xmm1, xmm3 ; xmm1=(40 41 42 43 44 45 46 47)=data4
+ punpckhqdq xmm6, xmm3 ; xmm6=(50 51 52 53 54 55 56 57)=data5
+
+ movdqa xmm0, xmm5
+ movdqa xmm3, xmm4
+ paddw xmm5, xmm1 ; xmm5=data3+data4=tmp3
+ paddw xmm4, xmm6 ; xmm4=data2+data5=tmp2
+ psubw xmm0, xmm1 ; xmm0=data3-data4=tmp4
+ psubw xmm3, xmm6 ; xmm3=data2-data5=tmp5
+
+ ; -- Even part
+
+ movdqa xmm1, xmm7
+ movdqa xmm6, xmm2
+ paddw xmm7, xmm5 ; xmm7=tmp10
+ paddw xmm2, xmm4 ; xmm2=tmp11
+ psubw xmm1, xmm5 ; xmm1=tmp13
+ psubw xmm6, xmm4 ; xmm6=tmp12
+
+ movdqa xmm5, xmm7
+ paddw xmm7, xmm2 ; xmm7=tmp10+tmp11
+ psubw xmm5, xmm2 ; xmm5=tmp10-tmp11
+
+ paddw xmm7, [rel PW_DESCALE_P2X]
+ paddw xmm5, [rel PW_DESCALE_P2X]
+ psraw xmm7, PASS1_BITS ; xmm7=data0
+ psraw xmm5, PASS1_BITS ; xmm5=data4
+
+ movdqa XMMWORD [XMMBLOCK(0,0,rdx,SIZEOF_DCTELEM)], xmm7
+ movdqa XMMWORD [XMMBLOCK(4,0,rdx,SIZEOF_DCTELEM)], xmm5
+
+ ; (Original)
+ ; z1 = (tmp12 + tmp13) * 0.541196100;
+ ; data2 = z1 + tmp13 * 0.765366865;
+ ; data6 = z1 + tmp12 * -1.847759065;
+ ;
+ ; (This implementation)
+ ; data2 = tmp13 * (0.541196100 + 0.765366865) + tmp12 * 0.541196100;
+ ; data6 = tmp13 * 0.541196100 + tmp12 * (0.541196100 - 1.847759065);
+
+ movdqa xmm4, xmm1 ; xmm1=tmp13
+ movdqa xmm2, xmm1
+ punpcklwd xmm4, xmm6 ; xmm6=tmp12
+ punpckhwd xmm2, xmm6
+ movdqa xmm1, xmm4
+ movdqa xmm6, xmm2
+ pmaddwd xmm4, [rel PW_F130_F054] ; xmm4=data2L
+ pmaddwd xmm2, [rel PW_F130_F054] ; xmm2=data2H
+ pmaddwd xmm1, [rel PW_F054_MF130] ; xmm1=data6L
+ pmaddwd xmm6, [rel PW_F054_MF130] ; xmm6=data6H
+
+ paddd xmm4, [rel PD_DESCALE_P2]
+ paddd xmm2, [rel PD_DESCALE_P2]
+ psrad xmm4, DESCALE_P2
+ psrad xmm2, DESCALE_P2
+ paddd xmm1, [rel PD_DESCALE_P2]
+ paddd xmm6, [rel PD_DESCALE_P2]
+ psrad xmm1, DESCALE_P2
+ psrad xmm6, DESCALE_P2
+
+ packssdw xmm4, xmm2 ; xmm4=data2
+ packssdw xmm1, xmm6 ; xmm1=data6
+
+ movdqa XMMWORD [XMMBLOCK(2,0,rdx,SIZEOF_DCTELEM)], xmm4
+ movdqa XMMWORD [XMMBLOCK(6,0,rdx,SIZEOF_DCTELEM)], xmm1
+
+ ; -- Odd part
+
+ movdqa xmm7, XMMWORD [wk(0)] ; xmm7=tmp6
+ movdqa xmm5, XMMWORD [wk(1)] ; xmm5=tmp7
+
+ movdqa xmm2, xmm0 ; xmm0=tmp4
+ movdqa xmm6, xmm3 ; xmm3=tmp5
+ paddw xmm2, xmm7 ; xmm2=z3
+ paddw xmm6, xmm5 ; xmm6=z4
+
+ ; (Original)
+ ; z5 = (z3 + z4) * 1.175875602;
+ ; z3 = z3 * -1.961570560; z4 = z4 * -0.390180644;
+ ; z3 += z5; z4 += z5;
+ ;
+ ; (This implementation)
+ ; z3 = z3 * (1.175875602 - 1.961570560) + z4 * 1.175875602;
+ ; z4 = z3 * 1.175875602 + z4 * (1.175875602 - 0.390180644);
+
+ movdqa xmm4, xmm2
+ movdqa xmm1, xmm2
+ punpcklwd xmm4, xmm6
+ punpckhwd xmm1, xmm6
+ movdqa xmm2, xmm4
+ movdqa xmm6, xmm1
+ pmaddwd xmm4, [rel PW_MF078_F117] ; xmm4=z3L
+ pmaddwd xmm1, [rel PW_MF078_F117] ; xmm1=z3H
+ pmaddwd xmm2, [rel PW_F117_F078] ; xmm2=z4L
+ pmaddwd xmm6, [rel PW_F117_F078] ; xmm6=z4H
+
+ movdqa XMMWORD [wk(0)], xmm4 ; wk(0)=z3L
+ movdqa XMMWORD [wk(1)], xmm1 ; wk(1)=z3H
+
+ ; (Original)
+ ; z1 = tmp4 + tmp7; z2 = tmp5 + tmp6;
+ ; tmp4 = tmp4 * 0.298631336; tmp5 = tmp5 * 2.053119869;
+ ; tmp6 = tmp6 * 3.072711026; tmp7 = tmp7 * 1.501321110;
+ ; z1 = z1 * -0.899976223; z2 = z2 * -2.562915447;
+ ; data7 = tmp4 + z1 + z3; data5 = tmp5 + z2 + z4;
+ ; data3 = tmp6 + z2 + z3; data1 = tmp7 + z1 + z4;
+ ;
+ ; (This implementation)
+ ; tmp4 = tmp4 * (0.298631336 - 0.899976223) + tmp7 * -0.899976223;
+ ; tmp5 = tmp5 * (2.053119869 - 2.562915447) + tmp6 * -2.562915447;
+ ; tmp6 = tmp5 * -2.562915447 + tmp6 * (3.072711026 - 2.562915447);
+ ; tmp7 = tmp4 * -0.899976223 + tmp7 * (1.501321110 - 0.899976223);
+ ; data7 = tmp4 + z3; data5 = tmp5 + z4;
+ ; data3 = tmp6 + z3; data1 = tmp7 + z4;
+
+ movdqa xmm4, xmm0
+ movdqa xmm1, xmm0
+ punpcklwd xmm4, xmm5
+ punpckhwd xmm1, xmm5
+ movdqa xmm0, xmm4
+ movdqa xmm5, xmm1
+ pmaddwd xmm4, [rel PW_MF060_MF089] ; xmm4=tmp4L
+ pmaddwd xmm1, [rel PW_MF060_MF089] ; xmm1=tmp4H
+ pmaddwd xmm0, [rel PW_MF089_F060] ; xmm0=tmp7L
+ pmaddwd xmm5, [rel PW_MF089_F060] ; xmm5=tmp7H
+
+ paddd xmm4, XMMWORD [wk(0)] ; xmm4=data7L
+ paddd xmm1, XMMWORD [wk(1)] ; xmm1=data7H
+ paddd xmm0, xmm2 ; xmm0=data1L
+ paddd xmm5, xmm6 ; xmm5=data1H
+
+ paddd xmm4, [rel PD_DESCALE_P2]
+ paddd xmm1, [rel PD_DESCALE_P2]
+ psrad xmm4, DESCALE_P2
+ psrad xmm1, DESCALE_P2
+ paddd xmm0, [rel PD_DESCALE_P2]
+ paddd xmm5, [rel PD_DESCALE_P2]
+ psrad xmm0, DESCALE_P2
+ psrad xmm5, DESCALE_P2
+
+ packssdw xmm4, xmm1 ; xmm4=data7
+ packssdw xmm0, xmm5 ; xmm0=data1
+
+ movdqa XMMWORD [XMMBLOCK(7,0,rdx,SIZEOF_DCTELEM)], xmm4
+ movdqa XMMWORD [XMMBLOCK(1,0,rdx,SIZEOF_DCTELEM)], xmm0
+
+ movdqa xmm1, xmm3
+ movdqa xmm5, xmm3
+ punpcklwd xmm1, xmm7
+ punpckhwd xmm5, xmm7
+ movdqa xmm3, xmm1
+ movdqa xmm7, xmm5
+ pmaddwd xmm1, [rel PW_MF050_MF256] ; xmm1=tmp5L
+ pmaddwd xmm5, [rel PW_MF050_MF256] ; xmm5=tmp5H
+ pmaddwd xmm3, [rel PW_MF256_F050] ; xmm3=tmp6L
+ pmaddwd xmm7, [rel PW_MF256_F050] ; xmm7=tmp6H
+
+ paddd xmm1, xmm2 ; xmm1=data5L
+ paddd xmm5, xmm6 ; xmm5=data5H
+ paddd xmm3, XMMWORD [wk(0)] ; xmm3=data3L
+ paddd xmm7, XMMWORD [wk(1)] ; xmm7=data3H
+
+ paddd xmm1, [rel PD_DESCALE_P2]
+ paddd xmm5, [rel PD_DESCALE_P2]
+ psrad xmm1, DESCALE_P2
+ psrad xmm5, DESCALE_P2
+ paddd xmm3, [rel PD_DESCALE_P2]
+ paddd xmm7, [rel PD_DESCALE_P2]
+ psrad xmm3, DESCALE_P2
+ psrad xmm7, DESCALE_P2
+
+ packssdw xmm1, xmm5 ; xmm1=data5
+ packssdw xmm3, xmm7 ; xmm3=data3
+
+ movdqa XMMWORD [XMMBLOCK(5,0,rdx,SIZEOF_DCTELEM)], xmm1
+ movdqa XMMWORD [XMMBLOCK(3,0,rdx,SIZEOF_DCTELEM)], xmm3
+
+ uncollect_args 1
+ mov rsp, rbp ; rsp <- aligned rbp
+ pop rsp ; rsp <- original rbp
+ pop rbp
+ ret
+
+; For some reason, the OS X linker does not honor the request to align the
+; segment unless we do this.
+ align 32
diff --git a/simd/x86_64/jidctflt-sse2.asm b/simd/x86_64/jidctflt-sse2.asm
new file mode 100644
index 0000000..6f22339
--- /dev/null
+++ b/simd/x86_64/jidctflt-sse2.asm
@@ -0,0 +1,482 @@
+;
+; jidctflt.asm - floating-point IDCT (64-bit SSE & SSE2)
+;
+; Copyright 2009 Pierre Ossman <ossman@cendio.se> for Cendio AB
+; Copyright (C) 2009, 2016, D. R. Commander.
+;
+; Based on the x86 SIMD extension for IJG JPEG library
+; Copyright (C) 1999-2006, MIYASAKA Masaru.
+; For conditions of distribution and use, see copyright notice in jsimdext.inc
+;
+; This file should be assembled with NASM (Netwide Assembler),
+; can *not* be assembled with Microsoft's MASM or any compatible
+; assembler (including Borland's Turbo Assembler).
+; NASM is available from http://nasm.sourceforge.net/ or
+; http://sourceforge.net/project/showfiles.php?group_id=6208
+;
+; This file contains a floating-point implementation of the inverse DCT
+; (Discrete Cosine Transform). The following code is based directly on
+; the IJG's original jidctflt.c; see the jidctflt.c for more details.
+;
+; [TAB8]
+
+%include "jsimdext.inc"
+%include "jdct.inc"
+
+; --------------------------------------------------------------------------
+
+%macro unpcklps2 2 ; %1=(0 1 2 3) / %2=(4 5 6 7) => %1=(0 1 4 5)
+ shufps %1, %2, 0x44
+%endmacro
+
+%macro unpckhps2 2 ; %1=(0 1 2 3) / %2=(4 5 6 7) => %1=(2 3 6 7)
+ shufps %1, %2, 0xEE
+%endmacro
+
+; --------------------------------------------------------------------------
+ SECTION SEG_CONST
+
+ alignz 32
+ global EXTN(jconst_idct_float_sse2)
+
+EXTN(jconst_idct_float_sse2):
+
+PD_1_414 times 4 dd 1.414213562373095048801689
+PD_1_847 times 4 dd 1.847759065022573512256366
+PD_1_082 times 4 dd 1.082392200292393968799446
+PD_M2_613 times 4 dd -2.613125929752753055713286
+PD_RNDINT_MAGIC times 4 dd 100663296.0 ; (float)(0x00C00000 << 3)
+PB_CENTERJSAMP times 16 db CENTERJSAMPLE
+
+ alignz 32
+
+; --------------------------------------------------------------------------
+ SECTION SEG_TEXT
+ BITS 64
+;
+; Perform dequantization and inverse DCT on one block of coefficients.
+;
+; GLOBAL(void)
+; jsimd_idct_float_sse2 (void *dct_table, JCOEFPTR coef_block,
+; JSAMPARRAY output_buf, JDIMENSION output_col)
+;
+
+; r10 = void *dct_table
+; r11 = JCOEFPTR coef_block
+; r12 = JSAMPARRAY output_buf
+; r13d = JDIMENSION output_col
+
+%define original_rbp rbp+0
+%define wk(i) rbp-(WK_NUM-(i))*SIZEOF_XMMWORD ; xmmword wk[WK_NUM]
+%define WK_NUM 2
+%define workspace wk(0)-DCTSIZE2*SIZEOF_FAST_FLOAT
+ ; FAST_FLOAT workspace[DCTSIZE2]
+
+ align 32
+ global EXTN(jsimd_idct_float_sse2)
+
+EXTN(jsimd_idct_float_sse2):
+ push rbp
+ mov rax, rsp ; rax = original rbp
+ sub rsp, byte 4
+ and rsp, byte (-SIZEOF_XMMWORD) ; align to 128 bits
+ mov [rsp], rax
+ mov rbp, rsp ; rbp = aligned rbp
+ lea rsp, [workspace]
+ collect_args 4
+ push rbx
+
+ ; ---- Pass 1: process columns from input, store into work array.
+
+ mov rdx, r10 ; quantptr
+ mov rsi, r11 ; inptr
+ lea rdi, [workspace] ; FAST_FLOAT *wsptr
+ mov rcx, DCTSIZE/4 ; ctr
+.columnloop:
+%ifndef NO_ZERO_COLUMN_TEST_FLOAT_SSE
+ mov eax, DWORD [DWBLOCK(1,0,rsi,SIZEOF_JCOEF)]
+ or eax, DWORD [DWBLOCK(2,0,rsi,SIZEOF_JCOEF)]
+ jnz near .columnDCT
+
+ movq xmm1, XMM_MMWORD [MMBLOCK(1,0,rsi,SIZEOF_JCOEF)]
+ movq xmm2, XMM_MMWORD [MMBLOCK(2,0,rsi,SIZEOF_JCOEF)]
+ movq xmm3, XMM_MMWORD [MMBLOCK(3,0,rsi,SIZEOF_JCOEF)]
+ movq xmm4, XMM_MMWORD [MMBLOCK(4,0,rsi,SIZEOF_JCOEF)]
+ movq xmm5, XMM_MMWORD [MMBLOCK(5,0,rsi,SIZEOF_JCOEF)]
+ movq xmm6, XMM_MMWORD [MMBLOCK(6,0,rsi,SIZEOF_JCOEF)]
+ movq xmm7, XMM_MMWORD [MMBLOCK(7,0,rsi,SIZEOF_JCOEF)]
+ por xmm1, xmm2
+ por xmm3, xmm4
+ por xmm5, xmm6
+ por xmm1, xmm3
+ por xmm5, xmm7
+ por xmm1, xmm5
+ packsswb xmm1, xmm1
+ movd eax, xmm1
+ test rax, rax
+ jnz short .columnDCT
+
+ ; -- AC terms all zero
+
+ movq xmm0, XMM_MMWORD [MMBLOCK(0,0,rsi,SIZEOF_JCOEF)]
+
+ punpcklwd xmm0, xmm0 ; xmm0=(00 00 01 01 02 02 03 03)
+ psrad xmm0, (DWORD_BIT-WORD_BIT) ; xmm0=in0=(00 01 02 03)
+ cvtdq2ps xmm0, xmm0 ; xmm0=in0=(00 01 02 03)
+
+ mulps xmm0, XMMWORD [XMMBLOCK(0,0,rdx,SIZEOF_FLOAT_MULT_TYPE)]
+
+ movaps xmm1, xmm0
+ movaps xmm2, xmm0
+ movaps xmm3, xmm0
+
+ shufps xmm0, xmm0, 0x00 ; xmm0=(00 00 00 00)
+ shufps xmm1, xmm1, 0x55 ; xmm1=(01 01 01 01)
+ shufps xmm2, xmm2, 0xAA ; xmm2=(02 02 02 02)
+ shufps xmm3, xmm3, 0xFF ; xmm3=(03 03 03 03)
+
+ movaps XMMWORD [XMMBLOCK(0,0,rdi,SIZEOF_FAST_FLOAT)], xmm0
+ movaps XMMWORD [XMMBLOCK(0,1,rdi,SIZEOF_FAST_FLOAT)], xmm0
+ movaps XMMWORD [XMMBLOCK(1,0,rdi,SIZEOF_FAST_FLOAT)], xmm1
+ movaps XMMWORD [XMMBLOCK(1,1,rdi,SIZEOF_FAST_FLOAT)], xmm1
+ movaps XMMWORD [XMMBLOCK(2,0,rdi,SIZEOF_FAST_FLOAT)], xmm2
+ movaps XMMWORD [XMMBLOCK(2,1,rdi,SIZEOF_FAST_FLOAT)], xmm2
+ movaps XMMWORD [XMMBLOCK(3,0,rdi,SIZEOF_FAST_FLOAT)], xmm3
+ movaps XMMWORD [XMMBLOCK(3,1,rdi,SIZEOF_FAST_FLOAT)], xmm3
+ jmp near .nextcolumn
+%endif
+.columnDCT:
+
+ ; -- Even part
+
+ movq xmm0, XMM_MMWORD [MMBLOCK(0,0,rsi,SIZEOF_JCOEF)]
+ movq xmm1, XMM_MMWORD [MMBLOCK(2,0,rsi,SIZEOF_JCOEF)]
+ movq xmm2, XMM_MMWORD [MMBLOCK(4,0,rsi,SIZEOF_JCOEF)]
+ movq xmm3, XMM_MMWORD [MMBLOCK(6,0,rsi,SIZEOF_JCOEF)]
+
+ punpcklwd xmm0, xmm0 ; xmm0=(00 00 01 01 02 02 03 03)
+ punpcklwd xmm1, xmm1 ; xmm1=(20 20 21 21 22 22 23 23)
+ psrad xmm0, (DWORD_BIT-WORD_BIT) ; xmm0=in0=(00 01 02 03)
+ psrad xmm1, (DWORD_BIT-WORD_BIT) ; xmm1=in2=(20 21 22 23)
+ cvtdq2ps xmm0, xmm0 ; xmm0=in0=(00 01 02 03)
+ cvtdq2ps xmm1, xmm1 ; xmm1=in2=(20 21 22 23)
+
+ punpcklwd xmm2, xmm2 ; xmm2=(40 40 41 41 42 42 43 43)
+ punpcklwd xmm3, xmm3 ; xmm3=(60 60 61 61 62 62 63 63)
+ psrad xmm2, (DWORD_BIT-WORD_BIT) ; xmm2=in4=(40 41 42 43)
+ psrad xmm3, (DWORD_BIT-WORD_BIT) ; xmm3=in6=(60 61 62 63)
+ cvtdq2ps xmm2, xmm2 ; xmm2=in4=(40 41 42 43)
+ cvtdq2ps xmm3, xmm3 ; xmm3=in6=(60 61 62 63)
+
+ mulps xmm0, XMMWORD [XMMBLOCK(0,0,rdx,SIZEOF_FLOAT_MULT_TYPE)]
+ mulps xmm1, XMMWORD [XMMBLOCK(2,0,rdx,SIZEOF_FLOAT_MULT_TYPE)]
+ mulps xmm2, XMMWORD [XMMBLOCK(4,0,rdx,SIZEOF_FLOAT_MULT_TYPE)]
+ mulps xmm3, XMMWORD [XMMBLOCK(6,0,rdx,SIZEOF_FLOAT_MULT_TYPE)]
+
+ movaps xmm4, xmm0
+ movaps xmm5, xmm1
+ subps xmm0, xmm2 ; xmm0=tmp11
+ subps xmm1, xmm3
+ addps xmm4, xmm2 ; xmm4=tmp10
+ addps xmm5, xmm3 ; xmm5=tmp13
+
+ mulps xmm1, [rel PD_1_414]
+ subps xmm1, xmm5 ; xmm1=tmp12
+
+ movaps xmm6, xmm4
+ movaps xmm7, xmm0
+ subps xmm4, xmm5 ; xmm4=tmp3
+ subps xmm0, xmm1 ; xmm0=tmp2
+ addps xmm6, xmm5 ; xmm6=tmp0
+ addps xmm7, xmm1 ; xmm7=tmp1
+
+ movaps XMMWORD [wk(1)], xmm4 ; tmp3
+ movaps XMMWORD [wk(0)], xmm0 ; tmp2
+
+ ; -- Odd part
+
+ movq xmm2, XMM_MMWORD [MMBLOCK(1,0,rsi,SIZEOF_JCOEF)]
+ movq xmm3, XMM_MMWORD [MMBLOCK(3,0,rsi,SIZEOF_JCOEF)]
+ movq xmm5, XMM_MMWORD [MMBLOCK(5,0,rsi,SIZEOF_JCOEF)]
+ movq xmm1, XMM_MMWORD [MMBLOCK(7,0,rsi,SIZEOF_JCOEF)]
+
+ punpcklwd xmm2, xmm2 ; xmm2=(10 10 11 11 12 12 13 13)
+ punpcklwd xmm3, xmm3 ; xmm3=(30 30 31 31 32 32 33 33)
+ psrad xmm2, (DWORD_BIT-WORD_BIT) ; xmm2=in1=(10 11 12 13)
+ psrad xmm3, (DWORD_BIT-WORD_BIT) ; xmm3=in3=(30 31 32 33)
+ cvtdq2ps xmm2, xmm2 ; xmm2=in1=(10 11 12 13)
+ cvtdq2ps xmm3, xmm3 ; xmm3=in3=(30 31 32 33)
+
+ punpcklwd xmm5, xmm5 ; xmm5=(50 50 51 51 52 52 53 53)
+ punpcklwd xmm1, xmm1 ; xmm1=(70 70 71 71 72 72 73 73)
+ psrad xmm5, (DWORD_BIT-WORD_BIT) ; xmm5=in5=(50 51 52 53)
+ psrad xmm1, (DWORD_BIT-WORD_BIT) ; xmm1=in7=(70 71 72 73)
+ cvtdq2ps xmm5, xmm5 ; xmm5=in5=(50 51 52 53)
+ cvtdq2ps xmm1, xmm1 ; xmm1=in7=(70 71 72 73)
+
+ mulps xmm2, XMMWORD [XMMBLOCK(1,0,rdx,SIZEOF_FLOAT_MULT_TYPE)]
+ mulps xmm3, XMMWORD [XMMBLOCK(3,0,rdx,SIZEOF_FLOAT_MULT_TYPE)]
+ mulps xmm5, XMMWORD [XMMBLOCK(5,0,rdx,SIZEOF_FLOAT_MULT_TYPE)]
+ mulps xmm1, XMMWORD [XMMBLOCK(7,0,rdx,SIZEOF_FLOAT_MULT_TYPE)]
+
+ movaps xmm4, xmm2
+ movaps xmm0, xmm5
+ addps xmm2, xmm1 ; xmm2=z11
+ addps xmm5, xmm3 ; xmm5=z13
+ subps xmm4, xmm1 ; xmm4=z12
+ subps xmm0, xmm3 ; xmm0=z10
+
+ movaps xmm1, xmm2
+ subps xmm2, xmm5
+ addps xmm1, xmm5 ; xmm1=tmp7
+
+ mulps xmm2, [rel PD_1_414] ; xmm2=tmp11
+
+ movaps xmm3, xmm0
+ addps xmm0, xmm4
+ mulps xmm0, [rel PD_1_847] ; xmm0=z5
+ mulps xmm3, [rel PD_M2_613] ; xmm3=(z10 * -2.613125930)
+ mulps xmm4, [rel PD_1_082] ; xmm4=(z12 * 1.082392200)
+ addps xmm3, xmm0 ; xmm3=tmp12
+ subps xmm4, xmm0 ; xmm4=tmp10
+
+ ; -- Final output stage
+
+ subps xmm3, xmm1 ; xmm3=tmp6
+ movaps xmm5, xmm6
+ movaps xmm0, xmm7
+ addps xmm6, xmm1 ; xmm6=data0=(00 01 02 03)
+ addps xmm7, xmm3 ; xmm7=data1=(10 11 12 13)
+ subps xmm5, xmm1 ; xmm5=data7=(70 71 72 73)
+ subps xmm0, xmm3 ; xmm0=data6=(60 61 62 63)
+ subps xmm2, xmm3 ; xmm2=tmp5
+
+ movaps xmm1, xmm6 ; transpose coefficients(phase 1)
+ unpcklps xmm6, xmm7 ; xmm6=(00 10 01 11)
+ unpckhps xmm1, xmm7 ; xmm1=(02 12 03 13)
+ movaps xmm3, xmm0 ; transpose coefficients(phase 1)
+ unpcklps xmm0, xmm5 ; xmm0=(60 70 61 71)
+ unpckhps xmm3, xmm5 ; xmm3=(62 72 63 73)
+
+ movaps xmm7, XMMWORD [wk(0)] ; xmm7=tmp2
+ movaps xmm5, XMMWORD [wk(1)] ; xmm5=tmp3
+
+ movaps XMMWORD [wk(0)], xmm0 ; wk(0)=(60 70 61 71)
+ movaps XMMWORD [wk(1)], xmm3 ; wk(1)=(62 72 63 73)
+
+ addps xmm4, xmm2 ; xmm4=tmp4
+ movaps xmm0, xmm7
+ movaps xmm3, xmm5
+ addps xmm7, xmm2 ; xmm7=data2=(20 21 22 23)
+ addps xmm5, xmm4 ; xmm5=data4=(40 41 42 43)
+ subps xmm0, xmm2 ; xmm0=data5=(50 51 52 53)
+ subps xmm3, xmm4 ; xmm3=data3=(30 31 32 33)
+
+ movaps xmm2, xmm7 ; transpose coefficients(phase 1)
+ unpcklps xmm7, xmm3 ; xmm7=(20 30 21 31)
+ unpckhps xmm2, xmm3 ; xmm2=(22 32 23 33)
+ movaps xmm4, xmm5 ; transpose coefficients(phase 1)
+ unpcklps xmm5, xmm0 ; xmm5=(40 50 41 51)
+ unpckhps xmm4, xmm0 ; xmm4=(42 52 43 53)
+
+ movaps xmm3, xmm6 ; transpose coefficients(phase 2)
+ unpcklps2 xmm6, xmm7 ; xmm6=(00 10 20 30)
+ unpckhps2 xmm3, xmm7 ; xmm3=(01 11 21 31)
+ movaps xmm0, xmm1 ; transpose coefficients(phase 2)
+ unpcklps2 xmm1, xmm2 ; xmm1=(02 12 22 32)
+ unpckhps2 xmm0, xmm2 ; xmm0=(03 13 23 33)
+
+ movaps xmm7, XMMWORD [wk(0)] ; xmm7=(60 70 61 71)
+ movaps xmm2, XMMWORD [wk(1)] ; xmm2=(62 72 63 73)
+
+ movaps XMMWORD [XMMBLOCK(0,0,rdi,SIZEOF_FAST_FLOAT)], xmm6
+ movaps XMMWORD [XMMBLOCK(1,0,rdi,SIZEOF_FAST_FLOAT)], xmm3
+ movaps XMMWORD [XMMBLOCK(2,0,rdi,SIZEOF_FAST_FLOAT)], xmm1
+ movaps XMMWORD [XMMBLOCK(3,0,rdi,SIZEOF_FAST_FLOAT)], xmm0
+
+ movaps xmm6, xmm5 ; transpose coefficients(phase 2)
+ unpcklps2 xmm5, xmm7 ; xmm5=(40 50 60 70)
+ unpckhps2 xmm6, xmm7 ; xmm6=(41 51 61 71)
+ movaps xmm3, xmm4 ; transpose coefficients(phase 2)
+ unpcklps2 xmm4, xmm2 ; xmm4=(42 52 62 72)
+ unpckhps2 xmm3, xmm2 ; xmm3=(43 53 63 73)
+
+ movaps XMMWORD [XMMBLOCK(0,1,rdi,SIZEOF_FAST_FLOAT)], xmm5
+ movaps XMMWORD [XMMBLOCK(1,1,rdi,SIZEOF_FAST_FLOAT)], xmm6
+ movaps XMMWORD [XMMBLOCK(2,1,rdi,SIZEOF_FAST_FLOAT)], xmm4
+ movaps XMMWORD [XMMBLOCK(3,1,rdi,SIZEOF_FAST_FLOAT)], xmm3
+
+.nextcolumn:
+ add rsi, byte 4*SIZEOF_JCOEF ; coef_block
+ add rdx, byte 4*SIZEOF_FLOAT_MULT_TYPE ; quantptr
+ add rdi, 4*DCTSIZE*SIZEOF_FAST_FLOAT ; wsptr
+ dec rcx ; ctr
+ jnz near .columnloop
+
+ ; -- Prefetch the next coefficient block
+
+ prefetchnta [rsi + (DCTSIZE2-8)*SIZEOF_JCOEF + 0*32]
+ prefetchnta [rsi + (DCTSIZE2-8)*SIZEOF_JCOEF + 1*32]
+ prefetchnta [rsi + (DCTSIZE2-8)*SIZEOF_JCOEF + 2*32]
+ prefetchnta [rsi + (DCTSIZE2-8)*SIZEOF_JCOEF + 3*32]
+
+ ; ---- Pass 2: process rows from work array, store into output array.
+
+ mov rax, [original_rbp]
+ lea rsi, [workspace] ; FAST_FLOAT *wsptr
+ mov rdi, r12 ; (JSAMPROW *)
+ mov eax, r13d
+ mov rcx, DCTSIZE/4 ; ctr
+.rowloop:
+
+ ; -- Even part
+
+ movaps xmm0, XMMWORD [XMMBLOCK(0,0,rsi,SIZEOF_FAST_FLOAT)]
+ movaps xmm1, XMMWORD [XMMBLOCK(2,0,rsi,SIZEOF_FAST_FLOAT)]
+ movaps xmm2, XMMWORD [XMMBLOCK(4,0,rsi,SIZEOF_FAST_FLOAT)]
+ movaps xmm3, XMMWORD [XMMBLOCK(6,0,rsi,SIZEOF_FAST_FLOAT)]
+
+ movaps xmm4, xmm0
+ movaps xmm5, xmm1
+ subps xmm0, xmm2 ; xmm0=tmp11
+ subps xmm1, xmm3
+ addps xmm4, xmm2 ; xmm4=tmp10
+ addps xmm5, xmm3 ; xmm5=tmp13
+
+ mulps xmm1, [rel PD_1_414]
+ subps xmm1, xmm5 ; xmm1=tmp12
+
+ movaps xmm6, xmm4
+ movaps xmm7, xmm0
+ subps xmm4, xmm5 ; xmm4=tmp3
+ subps xmm0, xmm1 ; xmm0=tmp2
+ addps xmm6, xmm5 ; xmm6=tmp0
+ addps xmm7, xmm1 ; xmm7=tmp1
+
+ movaps XMMWORD [wk(1)], xmm4 ; tmp3
+ movaps XMMWORD [wk(0)], xmm0 ; tmp2
+
+ ; -- Odd part
+
+ movaps xmm2, XMMWORD [XMMBLOCK(1,0,rsi,SIZEOF_FAST_FLOAT)]
+ movaps xmm3, XMMWORD [XMMBLOCK(3,0,rsi,SIZEOF_FAST_FLOAT)]
+ movaps xmm5, XMMWORD [XMMBLOCK(5,0,rsi,SIZEOF_FAST_FLOAT)]
+ movaps xmm1, XMMWORD [XMMBLOCK(7,0,rsi,SIZEOF_FAST_FLOAT)]
+
+ movaps xmm4, xmm2
+ movaps xmm0, xmm5
+ addps xmm2, xmm1 ; xmm2=z11
+ addps xmm5, xmm3 ; xmm5=z13
+ subps xmm4, xmm1 ; xmm4=z12
+ subps xmm0, xmm3 ; xmm0=z10
+
+ movaps xmm1, xmm2
+ subps xmm2, xmm5
+ addps xmm1, xmm5 ; xmm1=tmp7
+
+ mulps xmm2, [rel PD_1_414] ; xmm2=tmp11
+
+ movaps xmm3, xmm0
+ addps xmm0, xmm4
+ mulps xmm0, [rel PD_1_847] ; xmm0=z5
+ mulps xmm3, [rel PD_M2_613] ; xmm3=(z10 * -2.613125930)
+ mulps xmm4, [rel PD_1_082] ; xmm4=(z12 * 1.082392200)
+ addps xmm3, xmm0 ; xmm3=tmp12
+ subps xmm4, xmm0 ; xmm4=tmp10
+
+ ; -- Final output stage
+
+ subps xmm3, xmm1 ; xmm3=tmp6
+ movaps xmm5, xmm6
+ movaps xmm0, xmm7
+ addps xmm6, xmm1 ; xmm6=data0=(00 10 20 30)
+ addps xmm7, xmm3 ; xmm7=data1=(01 11 21 31)
+ subps xmm5, xmm1 ; xmm5=data7=(07 17 27 37)
+ subps xmm0, xmm3 ; xmm0=data6=(06 16 26 36)
+ subps xmm2, xmm3 ; xmm2=tmp5
+
+ movaps xmm1, [rel PD_RNDINT_MAGIC] ; xmm1=[rel PD_RNDINT_MAGIC]
+ pcmpeqd xmm3, xmm3
+ psrld xmm3, WORD_BIT ; xmm3={0xFFFF 0x0000 0xFFFF 0x0000 ..}
+
+ addps xmm6, xmm1 ; xmm6=roundint(data0/8)=(00 ** 10 ** 20 ** 30 **)
+ addps xmm7, xmm1 ; xmm7=roundint(data1/8)=(01 ** 11 ** 21 ** 31 **)
+ addps xmm0, xmm1 ; xmm0=roundint(data6/8)=(06 ** 16 ** 26 ** 36 **)
+ addps xmm5, xmm1 ; xmm5=roundint(data7/8)=(07 ** 17 ** 27 ** 37 **)
+
+ pand xmm6, xmm3 ; xmm6=(00 -- 10 -- 20 -- 30 --)
+ pslld xmm7, WORD_BIT ; xmm7=(-- 01 -- 11 -- 21 -- 31)
+ pand xmm0, xmm3 ; xmm0=(06 -- 16 -- 26 -- 36 --)
+ pslld xmm5, WORD_BIT ; xmm5=(-- 07 -- 17 -- 27 -- 37)
+ por xmm6, xmm7 ; xmm6=(00 01 10 11 20 21 30 31)
+ por xmm0, xmm5 ; xmm0=(06 07 16 17 26 27 36 37)
+
+ movaps xmm1, XMMWORD [wk(0)] ; xmm1=tmp2
+ movaps xmm3, XMMWORD [wk(1)] ; xmm3=tmp3
+
+ addps xmm4, xmm2 ; xmm4=tmp4
+ movaps xmm7, xmm1
+ movaps xmm5, xmm3
+ addps xmm1, xmm2 ; xmm1=data2=(02 12 22 32)
+ addps xmm3, xmm4 ; xmm3=data4=(04 14 24 34)
+ subps xmm7, xmm2 ; xmm7=data5=(05 15 25 35)
+ subps xmm5, xmm4 ; xmm5=data3=(03 13 23 33)
+
+ movaps xmm2, [rel PD_RNDINT_MAGIC] ; xmm2=[rel PD_RNDINT_MAGIC]
+ pcmpeqd xmm4, xmm4
+ psrld xmm4, WORD_BIT ; xmm4={0xFFFF 0x0000 0xFFFF 0x0000 ..}
+
+ addps xmm3, xmm2 ; xmm3=roundint(data4/8)=(04 ** 14 ** 24 ** 34 **)
+ addps xmm7, xmm2 ; xmm7=roundint(data5/8)=(05 ** 15 ** 25 ** 35 **)
+ addps xmm1, xmm2 ; xmm1=roundint(data2/8)=(02 ** 12 ** 22 ** 32 **)
+ addps xmm5, xmm2 ; xmm5=roundint(data3/8)=(03 ** 13 ** 23 ** 33 **)
+
+ pand xmm3, xmm4 ; xmm3=(04 -- 14 -- 24 -- 34 --)
+ pslld xmm7, WORD_BIT ; xmm7=(-- 05 -- 15 -- 25 -- 35)
+ pand xmm1, xmm4 ; xmm1=(02 -- 12 -- 22 -- 32 --)
+ pslld xmm5, WORD_BIT ; xmm5=(-- 03 -- 13 -- 23 -- 33)
+ por xmm3, xmm7 ; xmm3=(04 05 14 15 24 25 34 35)
+ por xmm1, xmm5 ; xmm1=(02 03 12 13 22 23 32 33)
+
+ movdqa xmm2, [rel PB_CENTERJSAMP] ; xmm2=[rel PB_CENTERJSAMP]
+
+ packsswb xmm6, xmm3 ; xmm6=(00 01 10 11 20 21 30 31 04 05 14 15 24 25 34 35)
+ packsswb xmm1, xmm0 ; xmm1=(02 03 12 13 22 23 32 33 06 07 16 17 26 27 36 37)
+ paddb xmm6, xmm2
+ paddb xmm1, xmm2
+
+ movdqa xmm4, xmm6 ; transpose coefficients(phase 2)
+ punpcklwd xmm6, xmm1 ; xmm6=(00 01 02 03 10 11 12 13 20 21 22 23 30 31 32 33)
+ punpckhwd xmm4, xmm1 ; xmm4=(04 05 06 07 14 15 16 17 24 25 26 27 34 35 36 37)
+
+ movdqa xmm7, xmm6 ; transpose coefficients(phase 3)
+ punpckldq xmm6, xmm4 ; xmm6=(00 01 02 03 04 05 06 07 10 11 12 13 14 15 16 17)
+ punpckhdq xmm7, xmm4 ; xmm7=(20 21 22 23 24 25 26 27 30 31 32 33 34 35 36 37)
+
+ pshufd xmm5, xmm6, 0x4E ; xmm5=(10 11 12 13 14 15 16 17 00 01 02 03 04 05 06 07)
+ pshufd xmm3, xmm7, 0x4E ; xmm3=(30 31 32 33 34 35 36 37 20 21 22 23 24 25 26 27)
+
+ mov rdx, JSAMPROW [rdi+0*SIZEOF_JSAMPROW]
+ mov rbx, JSAMPROW [rdi+2*SIZEOF_JSAMPROW]
+ movq XMM_MMWORD [rdx+rax*SIZEOF_JSAMPLE], xmm6
+ movq XMM_MMWORD [rbx+rax*SIZEOF_JSAMPLE], xmm7
+ mov rdx, JSAMPROW [rdi+1*SIZEOF_JSAMPROW]
+ mov rbx, JSAMPROW [rdi+3*SIZEOF_JSAMPROW]
+ movq XMM_MMWORD [rdx+rax*SIZEOF_JSAMPLE], xmm5
+ movq XMM_MMWORD [rbx+rax*SIZEOF_JSAMPLE], xmm3
+
+ add rsi, byte 4*SIZEOF_FAST_FLOAT ; wsptr
+ add rdi, byte 4*SIZEOF_JSAMPROW
+ dec rcx ; ctr
+ jnz near .rowloop
+
+ pop rbx
+ uncollect_args 4
+ mov rsp, rbp ; rsp <- aligned rbp
+ pop rsp ; rsp <- original rbp
+ pop rbp
+ ret
+
+; For some reason, the OS X linker does not honor the request to align the
+; segment unless we do this.
+ align 32
diff --git a/simd/x86_64/jidctfst-sse2.asm b/simd/x86_64/jidctfst-sse2.asm
new file mode 100644
index 0000000..422ad1b
--- /dev/null
+++ b/simd/x86_64/jidctfst-sse2.asm
@@ -0,0 +1,491 @@
+;
+; jidctfst.asm - fast integer IDCT (64-bit SSE2)
+;
+; Copyright 2009 Pierre Ossman <ossman@cendio.se> for Cendio AB
+; Copyright (C) 2009, 2016, D. R. Commander.
+;
+; Based on the x86 SIMD extension for IJG JPEG library
+; Copyright (C) 1999-2006, MIYASAKA Masaru.
+; For conditions of distribution and use, see copyright notice in jsimdext.inc
+;
+; This file should be assembled with NASM (Netwide Assembler),
+; can *not* be assembled with Microsoft's MASM or any compatible
+; assembler (including Borland's Turbo Assembler).
+; NASM is available from http://nasm.sourceforge.net/ or
+; http://sourceforge.net/project/showfiles.php?group_id=6208
+;
+; This file contains a fast, not so accurate integer implementation of
+; the inverse DCT (Discrete Cosine Transform). The following code is
+; based directly on the IJG's original jidctfst.c; see the jidctfst.c
+; for more details.
+;
+; [TAB8]
+
+%include "jsimdext.inc"
+%include "jdct.inc"
+
+; --------------------------------------------------------------------------
+
+%define CONST_BITS 8 ; 14 is also OK.
+%define PASS1_BITS 2
+
+%if IFAST_SCALE_BITS != PASS1_BITS
+%error "'IFAST_SCALE_BITS' must be equal to 'PASS1_BITS'."
+%endif
+
+%if CONST_BITS == 8
+F_1_082 equ 277 ; FIX(1.082392200)
+F_1_414 equ 362 ; FIX(1.414213562)
+F_1_847 equ 473 ; FIX(1.847759065)
+F_2_613 equ 669 ; FIX(2.613125930)
+F_1_613 equ (F_2_613 - 256) ; FIX(2.613125930) - FIX(1)
+%else
+; NASM cannot do compile-time arithmetic on floating-point constants.
+%define DESCALE(x,n) (((x)+(1<<((n)-1)))>>(n))
+F_1_082 equ DESCALE(1162209775, 30-CONST_BITS) ; FIX(1.082392200)
+F_1_414 equ DESCALE(1518500249, 30-CONST_BITS) ; FIX(1.414213562)
+F_1_847 equ DESCALE(1984016188, 30-CONST_BITS) ; FIX(1.847759065)
+F_2_613 equ DESCALE(2805822602, 30-CONST_BITS) ; FIX(2.613125930)
+F_1_613 equ (F_2_613 - (1 << CONST_BITS)) ; FIX(2.613125930) - FIX(1)
+%endif
+
+; --------------------------------------------------------------------------
+ SECTION SEG_CONST
+
+; PRE_MULTIPLY_SCALE_BITS <= 2 (to avoid overflow)
+; CONST_BITS + CONST_SHIFT + PRE_MULTIPLY_SCALE_BITS == 16 (for pmulhw)
+
+%define PRE_MULTIPLY_SCALE_BITS 2
+%define CONST_SHIFT (16 - PRE_MULTIPLY_SCALE_BITS - CONST_BITS)
+
+ alignz 32
+ global EXTN(jconst_idct_ifast_sse2)
+
+EXTN(jconst_idct_ifast_sse2):
+
+PW_F1414 times 8 dw F_1_414 << CONST_SHIFT
+PW_F1847 times 8 dw F_1_847 << CONST_SHIFT
+PW_MF1613 times 8 dw -F_1_613 << CONST_SHIFT
+PW_F1082 times 8 dw F_1_082 << CONST_SHIFT
+PB_CENTERJSAMP times 16 db CENTERJSAMPLE
+
+ alignz 32
+
+; --------------------------------------------------------------------------
+ SECTION SEG_TEXT
+ BITS 64
+;
+; Perform dequantization and inverse DCT on one block of coefficients.
+;
+; GLOBAL(void)
+; jsimd_idct_ifast_sse2 (void *dct_table, JCOEFPTR coef_block,
+; JSAMPARRAY output_buf, JDIMENSION output_col)
+;
+
+; r10 = jpeg_component_info *compptr
+; r11 = JCOEFPTR coef_block
+; r12 = JSAMPARRAY output_buf
+; r13d = JDIMENSION output_col
+
+%define original_rbp rbp+0
+%define wk(i) rbp-(WK_NUM-(i))*SIZEOF_XMMWORD ; xmmword wk[WK_NUM]
+%define WK_NUM 2
+
+ align 32
+ global EXTN(jsimd_idct_ifast_sse2)
+
+EXTN(jsimd_idct_ifast_sse2):
+ push rbp
+ mov rax, rsp ; rax = original rbp
+ sub rsp, byte 4
+ and rsp, byte (-SIZEOF_XMMWORD) ; align to 128 bits
+ mov [rsp], rax
+ mov rbp, rsp ; rbp = aligned rbp
+ lea rsp, [wk(0)]
+ collect_args 4
+
+ ; ---- Pass 1: process columns from input.
+
+ mov rdx, r10 ; quantptr
+ mov rsi, r11 ; inptr
+
+%ifndef NO_ZERO_COLUMN_TEST_IFAST_SSE2
+ mov eax, DWORD [DWBLOCK(1,0,rsi,SIZEOF_JCOEF)]
+ or eax, DWORD [DWBLOCK(2,0,rsi,SIZEOF_JCOEF)]
+ jnz near .columnDCT
+
+ movdqa xmm0, XMMWORD [XMMBLOCK(1,0,rsi,SIZEOF_JCOEF)]
+ movdqa xmm1, XMMWORD [XMMBLOCK(2,0,rsi,SIZEOF_JCOEF)]
+ por xmm0, XMMWORD [XMMBLOCK(3,0,rsi,SIZEOF_JCOEF)]
+ por xmm1, XMMWORD [XMMBLOCK(4,0,rsi,SIZEOF_JCOEF)]
+ por xmm0, XMMWORD [XMMBLOCK(5,0,rsi,SIZEOF_JCOEF)]
+ por xmm1, XMMWORD [XMMBLOCK(6,0,rsi,SIZEOF_JCOEF)]
+ por xmm0, XMMWORD [XMMBLOCK(7,0,rsi,SIZEOF_JCOEF)]
+ por xmm1, xmm0
+ packsswb xmm1, xmm1
+ packsswb xmm1, xmm1
+ movd eax, xmm1
+ test rax, rax
+ jnz short .columnDCT
+
+ ; -- AC terms all zero
+
+ movdqa xmm0, XMMWORD [XMMBLOCK(0,0,rsi,SIZEOF_JCOEF)]
+ pmullw xmm0, XMMWORD [XMMBLOCK(0,0,rdx,SIZEOF_ISLOW_MULT_TYPE)]
+
+ movdqa xmm7, xmm0 ; xmm0=in0=(00 01 02 03 04 05 06 07)
+ punpcklwd xmm0, xmm0 ; xmm0=(00 00 01 01 02 02 03 03)
+ punpckhwd xmm7, xmm7 ; xmm7=(04 04 05 05 06 06 07 07)
+
+ pshufd xmm6, xmm0, 0x00 ; xmm6=col0=(00 00 00 00 00 00 00 00)
+ pshufd xmm2, xmm0, 0x55 ; xmm2=col1=(01 01 01 01 01 01 01 01)
+ pshufd xmm5, xmm0, 0xAA ; xmm5=col2=(02 02 02 02 02 02 02 02)
+ pshufd xmm0, xmm0, 0xFF ; xmm0=col3=(03 03 03 03 03 03 03 03)
+ pshufd xmm1, xmm7, 0x00 ; xmm1=col4=(04 04 04 04 04 04 04 04)
+ pshufd xmm4, xmm7, 0x55 ; xmm4=col5=(05 05 05 05 05 05 05 05)
+ pshufd xmm3, xmm7, 0xAA ; xmm3=col6=(06 06 06 06 06 06 06 06)
+ pshufd xmm7, xmm7, 0xFF ; xmm7=col7=(07 07 07 07 07 07 07 07)
+
+ movdqa XMMWORD [wk(0)], xmm2 ; wk(0)=col1
+ movdqa XMMWORD [wk(1)], xmm0 ; wk(1)=col3
+ jmp near .column_end
+%endif
+.columnDCT:
+
+ ; -- Even part
+
+ movdqa xmm0, XMMWORD [XMMBLOCK(0,0,rsi,SIZEOF_JCOEF)]
+ movdqa xmm1, XMMWORD [XMMBLOCK(2,0,rsi,SIZEOF_JCOEF)]
+ pmullw xmm0, XMMWORD [XMMBLOCK(0,0,rdx,SIZEOF_IFAST_MULT_TYPE)]
+ pmullw xmm1, XMMWORD [XMMBLOCK(2,0,rdx,SIZEOF_IFAST_MULT_TYPE)]
+ movdqa xmm2, XMMWORD [XMMBLOCK(4,0,rsi,SIZEOF_JCOEF)]
+ movdqa xmm3, XMMWORD [XMMBLOCK(6,0,rsi,SIZEOF_JCOEF)]
+ pmullw xmm2, XMMWORD [XMMBLOCK(4,0,rdx,SIZEOF_IFAST_MULT_TYPE)]
+ pmullw xmm3, XMMWORD [XMMBLOCK(6,0,rdx,SIZEOF_IFAST_MULT_TYPE)]
+
+ movdqa xmm4, xmm0
+ movdqa xmm5, xmm1
+ psubw xmm0, xmm2 ; xmm0=tmp11
+ psubw xmm1, xmm3
+ paddw xmm4, xmm2 ; xmm4=tmp10
+ paddw xmm5, xmm3 ; xmm5=tmp13
+
+ psllw xmm1, PRE_MULTIPLY_SCALE_BITS
+ pmulhw xmm1, [rel PW_F1414]
+ psubw xmm1, xmm5 ; xmm1=tmp12
+
+ movdqa xmm6, xmm4
+ movdqa xmm7, xmm0
+ psubw xmm4, xmm5 ; xmm4=tmp3
+ psubw xmm0, xmm1 ; xmm0=tmp2
+ paddw xmm6, xmm5 ; xmm6=tmp0
+ paddw xmm7, xmm1 ; xmm7=tmp1
+
+ movdqa XMMWORD [wk(1)], xmm4 ; wk(1)=tmp3
+ movdqa XMMWORD [wk(0)], xmm0 ; wk(0)=tmp2
+
+ ; -- Odd part
+
+ movdqa xmm2, XMMWORD [XMMBLOCK(1,0,rsi,SIZEOF_JCOEF)]
+ movdqa xmm3, XMMWORD [XMMBLOCK(3,0,rsi,SIZEOF_JCOEF)]
+ pmullw xmm2, XMMWORD [XMMBLOCK(1,0,rdx,SIZEOF_IFAST_MULT_TYPE)]
+ pmullw xmm3, XMMWORD [XMMBLOCK(3,0,rdx,SIZEOF_IFAST_MULT_TYPE)]
+ movdqa xmm5, XMMWORD [XMMBLOCK(5,0,rsi,SIZEOF_JCOEF)]
+ movdqa xmm1, XMMWORD [XMMBLOCK(7,0,rsi,SIZEOF_JCOEF)]
+ pmullw xmm5, XMMWORD [XMMBLOCK(5,0,rdx,SIZEOF_IFAST_MULT_TYPE)]
+ pmullw xmm1, XMMWORD [XMMBLOCK(7,0,rdx,SIZEOF_IFAST_MULT_TYPE)]
+
+ movdqa xmm4, xmm2
+ movdqa xmm0, xmm5
+ psubw xmm2, xmm1 ; xmm2=z12
+ psubw xmm5, xmm3 ; xmm5=z10
+ paddw xmm4, xmm1 ; xmm4=z11
+ paddw xmm0, xmm3 ; xmm0=z13
+
+ movdqa xmm1, xmm5 ; xmm1=z10(unscaled)
+ psllw xmm2, PRE_MULTIPLY_SCALE_BITS
+ psllw xmm5, PRE_MULTIPLY_SCALE_BITS
+
+ movdqa xmm3, xmm4
+ psubw xmm4, xmm0
+ paddw xmm3, xmm0 ; xmm3=tmp7
+
+ psllw xmm4, PRE_MULTIPLY_SCALE_BITS
+ pmulhw xmm4, [rel PW_F1414] ; xmm4=tmp11
+
+ ; To avoid overflow...
+ ;
+ ; (Original)
+ ; tmp12 = -2.613125930 * z10 + z5;
+ ;
+ ; (This implementation)
+ ; tmp12 = (-1.613125930 - 1) * z10 + z5;
+ ; = -1.613125930 * z10 - z10 + z5;
+
+ movdqa xmm0, xmm5
+ paddw xmm5, xmm2
+ pmulhw xmm5, [rel PW_F1847] ; xmm5=z5
+ pmulhw xmm0, [rel PW_MF1613]
+ pmulhw xmm2, [rel PW_F1082]
+ psubw xmm0, xmm1
+ psubw xmm2, xmm5 ; xmm2=tmp10
+ paddw xmm0, xmm5 ; xmm0=tmp12
+
+ ; -- Final output stage
+
+ psubw xmm0, xmm3 ; xmm0=tmp6
+ movdqa xmm1, xmm6
+ movdqa xmm5, xmm7
+ paddw xmm6, xmm3 ; xmm6=data0=(00 01 02 03 04 05 06 07)
+ paddw xmm7, xmm0 ; xmm7=data1=(10 11 12 13 14 15 16 17)
+ psubw xmm1, xmm3 ; xmm1=data7=(70 71 72 73 74 75 76 77)
+ psubw xmm5, xmm0 ; xmm5=data6=(60 61 62 63 64 65 66 67)
+ psubw xmm4, xmm0 ; xmm4=tmp5
+
+ movdqa xmm3, xmm6 ; transpose coefficients(phase 1)
+ punpcklwd xmm6, xmm7 ; xmm6=(00 10 01 11 02 12 03 13)
+ punpckhwd xmm3, xmm7 ; xmm3=(04 14 05 15 06 16 07 17)
+ movdqa xmm0, xmm5 ; transpose coefficients(phase 1)
+ punpcklwd xmm5, xmm1 ; xmm5=(60 70 61 71 62 72 63 73)
+ punpckhwd xmm0, xmm1 ; xmm0=(64 74 65 75 66 76 67 77)
+
+ movdqa xmm7, XMMWORD [wk(0)] ; xmm7=tmp2
+ movdqa xmm1, XMMWORD [wk(1)] ; xmm1=tmp3
+
+ movdqa XMMWORD [wk(0)], xmm5 ; wk(0)=(60 70 61 71 62 72 63 73)
+ movdqa XMMWORD [wk(1)], xmm0 ; wk(1)=(64 74 65 75 66 76 67 77)
+
+ paddw xmm2, xmm4 ; xmm2=tmp4
+ movdqa xmm5, xmm7
+ movdqa xmm0, xmm1
+ paddw xmm7, xmm4 ; xmm7=data2=(20 21 22 23 24 25 26 27)
+ paddw xmm1, xmm2 ; xmm1=data4=(40 41 42 43 44 45 46 47)
+ psubw xmm5, xmm4 ; xmm5=data5=(50 51 52 53 54 55 56 57)
+ psubw xmm0, xmm2 ; xmm0=data3=(30 31 32 33 34 35 36 37)
+
+ movdqa xmm4, xmm7 ; transpose coefficients(phase 1)
+ punpcklwd xmm7, xmm0 ; xmm7=(20 30 21 31 22 32 23 33)
+ punpckhwd xmm4, xmm0 ; xmm4=(24 34 25 35 26 36 27 37)
+ movdqa xmm2, xmm1 ; transpose coefficients(phase 1)
+ punpcklwd xmm1, xmm5 ; xmm1=(40 50 41 51 42 52 43 53)
+ punpckhwd xmm2, xmm5 ; xmm2=(44 54 45 55 46 56 47 57)
+
+ movdqa xmm0, xmm3 ; transpose coefficients(phase 2)
+ punpckldq xmm3, xmm4 ; xmm3=(04 14 24 34 05 15 25 35)
+ punpckhdq xmm0, xmm4 ; xmm0=(06 16 26 36 07 17 27 37)
+ movdqa xmm5, xmm6 ; transpose coefficients(phase 2)
+ punpckldq xmm6, xmm7 ; xmm6=(00 10 20 30 01 11 21 31)
+ punpckhdq xmm5, xmm7 ; xmm5=(02 12 22 32 03 13 23 33)
+
+ movdqa xmm4, XMMWORD [wk(0)] ; xmm4=(60 70 61 71 62 72 63 73)
+ movdqa xmm7, XMMWORD [wk(1)] ; xmm7=(64 74 65 75 66 76 67 77)
+
+ movdqa XMMWORD [wk(0)], xmm3 ; wk(0)=(04 14 24 34 05 15 25 35)
+ movdqa XMMWORD [wk(1)], xmm0 ; wk(1)=(06 16 26 36 07 17 27 37)
+
+ movdqa xmm3, xmm1 ; transpose coefficients(phase 2)
+ punpckldq xmm1, xmm4 ; xmm1=(40 50 60 70 41 51 61 71)
+ punpckhdq xmm3, xmm4 ; xmm3=(42 52 62 72 43 53 63 73)
+ movdqa xmm0, xmm2 ; transpose coefficients(phase 2)
+ punpckldq xmm2, xmm7 ; xmm2=(44 54 64 74 45 55 65 75)
+ punpckhdq xmm0, xmm7 ; xmm0=(46 56 66 76 47 57 67 77)
+
+ movdqa xmm4, xmm6 ; transpose coefficients(phase 3)
+ punpcklqdq xmm6, xmm1 ; xmm6=col0=(00 10 20 30 40 50 60 70)
+ punpckhqdq xmm4, xmm1 ; xmm4=col1=(01 11 21 31 41 51 61 71)
+ movdqa xmm7, xmm5 ; transpose coefficients(phase 3)
+ punpcklqdq xmm5, xmm3 ; xmm5=col2=(02 12 22 32 42 52 62 72)
+ punpckhqdq xmm7, xmm3 ; xmm7=col3=(03 13 23 33 43 53 63 73)
+
+ movdqa xmm1, XMMWORD [wk(0)] ; xmm1=(04 14 24 34 05 15 25 35)
+ movdqa xmm3, XMMWORD [wk(1)] ; xmm3=(06 16 26 36 07 17 27 37)
+
+ movdqa XMMWORD [wk(0)], xmm4 ; wk(0)=col1
+ movdqa XMMWORD [wk(1)], xmm7 ; wk(1)=col3
+
+ movdqa xmm4, xmm1 ; transpose coefficients(phase 3)
+ punpcklqdq xmm1, xmm2 ; xmm1=col4=(04 14 24 34 44 54 64 74)
+ punpckhqdq xmm4, xmm2 ; xmm4=col5=(05 15 25 35 45 55 65 75)
+ movdqa xmm7, xmm3 ; transpose coefficients(phase 3)
+ punpcklqdq xmm3, xmm0 ; xmm3=col6=(06 16 26 36 46 56 66 76)
+ punpckhqdq xmm7, xmm0 ; xmm7=col7=(07 17 27 37 47 57 67 77)
+.column_end:
+
+ ; -- Prefetch the next coefficient block
+
+ prefetchnta [rsi + DCTSIZE2*SIZEOF_JCOEF + 0*32]
+ prefetchnta [rsi + DCTSIZE2*SIZEOF_JCOEF + 1*32]
+ prefetchnta [rsi + DCTSIZE2*SIZEOF_JCOEF + 2*32]
+ prefetchnta [rsi + DCTSIZE2*SIZEOF_JCOEF + 3*32]
+
+ ; ---- Pass 2: process rows from work array, store into output array.
+
+ mov rax, [original_rbp]
+ mov rdi, r12 ; (JSAMPROW *)
+ mov eax, r13d
+
+ ; -- Even part
+
+ ; xmm6=col0, xmm5=col2, xmm1=col4, xmm3=col6
+
+ movdqa xmm2, xmm6
+ movdqa xmm0, xmm5
+ psubw xmm6, xmm1 ; xmm6=tmp11
+ psubw xmm5, xmm3
+ paddw xmm2, xmm1 ; xmm2=tmp10
+ paddw xmm0, xmm3 ; xmm0=tmp13
+
+ psllw xmm5, PRE_MULTIPLY_SCALE_BITS
+ pmulhw xmm5, [rel PW_F1414]
+ psubw xmm5, xmm0 ; xmm5=tmp12
+
+ movdqa xmm1, xmm2
+ movdqa xmm3, xmm6
+ psubw xmm2, xmm0 ; xmm2=tmp3
+ psubw xmm6, xmm5 ; xmm6=tmp2
+ paddw xmm1, xmm0 ; xmm1=tmp0
+ paddw xmm3, xmm5 ; xmm3=tmp1
+
+ movdqa xmm0, XMMWORD [wk(0)] ; xmm0=col1
+ movdqa xmm5, XMMWORD [wk(1)] ; xmm5=col3
+
+ movdqa XMMWORD [wk(0)], xmm2 ; wk(0)=tmp3
+ movdqa XMMWORD [wk(1)], xmm6 ; wk(1)=tmp2
+
+ ; -- Odd part
+
+ ; xmm0=col1, xmm5=col3, xmm4=col5, xmm7=col7
+
+ movdqa xmm2, xmm0
+ movdqa xmm6, xmm4
+ psubw xmm0, xmm7 ; xmm0=z12
+ psubw xmm4, xmm5 ; xmm4=z10
+ paddw xmm2, xmm7 ; xmm2=z11
+ paddw xmm6, xmm5 ; xmm6=z13
+
+ movdqa xmm7, xmm4 ; xmm7=z10(unscaled)
+ psllw xmm0, PRE_MULTIPLY_SCALE_BITS
+ psllw xmm4, PRE_MULTIPLY_SCALE_BITS
+
+ movdqa xmm5, xmm2
+ psubw xmm2, xmm6
+ paddw xmm5, xmm6 ; xmm5=tmp7
+
+ psllw xmm2, PRE_MULTIPLY_SCALE_BITS
+ pmulhw xmm2, [rel PW_F1414] ; xmm2=tmp11
+
+ ; To avoid overflow...
+ ;
+ ; (Original)
+ ; tmp12 = -2.613125930 * z10 + z5;
+ ;
+ ; (This implementation)
+ ; tmp12 = (-1.613125930 - 1) * z10 + z5;
+ ; = -1.613125930 * z10 - z10 + z5;
+
+ movdqa xmm6, xmm4
+ paddw xmm4, xmm0
+ pmulhw xmm4, [rel PW_F1847] ; xmm4=z5
+ pmulhw xmm6, [rel PW_MF1613]
+ pmulhw xmm0, [rel PW_F1082]
+ psubw xmm6, xmm7
+ psubw xmm0, xmm4 ; xmm0=tmp10
+ paddw xmm6, xmm4 ; xmm6=tmp12
+
+ ; -- Final output stage
+
+ psubw xmm6, xmm5 ; xmm6=tmp6
+ movdqa xmm7, xmm1
+ movdqa xmm4, xmm3
+ paddw xmm1, xmm5 ; xmm1=data0=(00 10 20 30 40 50 60 70)
+ paddw xmm3, xmm6 ; xmm3=data1=(01 11 21 31 41 51 61 71)
+ psraw xmm1, (PASS1_BITS+3) ; descale
+ psraw xmm3, (PASS1_BITS+3) ; descale
+ psubw xmm7, xmm5 ; xmm7=data7=(07 17 27 37 47 57 67 77)
+ psubw xmm4, xmm6 ; xmm4=data6=(06 16 26 36 46 56 66 76)
+ psraw xmm7, (PASS1_BITS+3) ; descale
+ psraw xmm4, (PASS1_BITS+3) ; descale
+ psubw xmm2, xmm6 ; xmm2=tmp5
+
+ packsswb xmm1, xmm4 ; xmm1=(00 10 20 30 40 50 60 70 06 16 26 36 46 56 66 76)
+ packsswb xmm3, xmm7 ; xmm3=(01 11 21 31 41 51 61 71 07 17 27 37 47 57 67 77)
+
+ movdqa xmm5, XMMWORD [wk(1)] ; xmm5=tmp2
+ movdqa xmm6, XMMWORD [wk(0)] ; xmm6=tmp3
+
+ paddw xmm0, xmm2 ; xmm0=tmp4
+ movdqa xmm4, xmm5
+ movdqa xmm7, xmm6
+ paddw xmm5, xmm2 ; xmm5=data2=(02 12 22 32 42 52 62 72)
+ paddw xmm6, xmm0 ; xmm6=data4=(04 14 24 34 44 54 64 74)
+ psraw xmm5, (PASS1_BITS+3) ; descale
+ psraw xmm6, (PASS1_BITS+3) ; descale
+ psubw xmm4, xmm2 ; xmm4=data5=(05 15 25 35 45 55 65 75)
+ psubw xmm7, xmm0 ; xmm7=data3=(03 13 23 33 43 53 63 73)
+ psraw xmm4, (PASS1_BITS+3) ; descale
+ psraw xmm7, (PASS1_BITS+3) ; descale
+
+ movdqa xmm2, [rel PB_CENTERJSAMP] ; xmm2=[rel PB_CENTERJSAMP]
+
+ packsswb xmm5, xmm6 ; xmm5=(02 12 22 32 42 52 62 72 04 14 24 34 44 54 64 74)
+ packsswb xmm7, xmm4 ; xmm7=(03 13 23 33 43 53 63 73 05 15 25 35 45 55 65 75)
+
+ paddb xmm1, xmm2
+ paddb xmm3, xmm2
+ paddb xmm5, xmm2
+ paddb xmm7, xmm2
+
+ movdqa xmm0, xmm1 ; transpose coefficients(phase 1)
+ punpcklbw xmm1, xmm3 ; xmm1=(00 01 10 11 20 21 30 31 40 41 50 51 60 61 70 71)
+ punpckhbw xmm0, xmm3 ; xmm0=(06 07 16 17 26 27 36 37 46 47 56 57 66 67 76 77)
+ movdqa xmm6, xmm5 ; transpose coefficients(phase 1)
+ punpcklbw xmm5, xmm7 ; xmm5=(02 03 12 13 22 23 32 33 42 43 52 53 62 63 72 73)
+ punpckhbw xmm6, xmm7 ; xmm6=(04 05 14 15 24 25 34 35 44 45 54 55 64 65 74 75)
+
+ movdqa xmm4, xmm1 ; transpose coefficients(phase 2)
+ punpcklwd xmm1, xmm5 ; xmm1=(00 01 02 03 10 11 12 13 20 21 22 23 30 31 32 33)
+ punpckhwd xmm4, xmm5 ; xmm4=(40 41 42 43 50 51 52 53 60 61 62 63 70 71 72 73)
+ movdqa xmm2, xmm6 ; transpose coefficients(phase 2)
+ punpcklwd xmm6, xmm0 ; xmm6=(04 05 06 07 14 15 16 17 24 25 26 27 34 35 36 37)
+ punpckhwd xmm2, xmm0 ; xmm2=(44 45 46 47 54 55 56 57 64 65 66 67 74 75 76 77)
+
+ movdqa xmm3, xmm1 ; transpose coefficients(phase 3)
+ punpckldq xmm1, xmm6 ; xmm1=(00 01 02 03 04 05 06 07 10 11 12 13 14 15 16 17)
+ punpckhdq xmm3, xmm6 ; xmm3=(20 21 22 23 24 25 26 27 30 31 32 33 34 35 36 37)
+ movdqa xmm7, xmm4 ; transpose coefficients(phase 3)
+ punpckldq xmm4, xmm2 ; xmm4=(40 41 42 43 44 45 46 47 50 51 52 53 54 55 56 57)
+ punpckhdq xmm7, xmm2 ; xmm7=(60 61 62 63 64 65 66 67 70 71 72 73 74 75 76 77)
+
+ pshufd xmm5, xmm1, 0x4E ; xmm5=(10 11 12 13 14 15 16 17 00 01 02 03 04 05 06 07)
+ pshufd xmm0, xmm3, 0x4E ; xmm0=(30 31 32 33 34 35 36 37 20 21 22 23 24 25 26 27)
+ pshufd xmm6, xmm4, 0x4E ; xmm6=(50 51 52 53 54 55 56 57 40 41 42 43 44 45 46 47)
+ pshufd xmm2, xmm7, 0x4E ; xmm2=(70 71 72 73 74 75 76 77 60 61 62 63 64 65 66 67)
+
+ mov rdx, JSAMPROW [rdi+0*SIZEOF_JSAMPROW]
+ mov rsi, JSAMPROW [rdi+2*SIZEOF_JSAMPROW]
+ movq XMM_MMWORD [rdx+rax*SIZEOF_JSAMPLE], xmm1
+ movq XMM_MMWORD [rsi+rax*SIZEOF_JSAMPLE], xmm3
+ mov rdx, JSAMPROW [rdi+4*SIZEOF_JSAMPROW]
+ mov rsi, JSAMPROW [rdi+6*SIZEOF_JSAMPROW]
+ movq XMM_MMWORD [rdx+rax*SIZEOF_JSAMPLE], xmm4
+ movq XMM_MMWORD [rsi+rax*SIZEOF_JSAMPLE], xmm7
+
+ mov rdx, JSAMPROW [rdi+1*SIZEOF_JSAMPROW]
+ mov rsi, JSAMPROW [rdi+3*SIZEOF_JSAMPROW]
+ movq XMM_MMWORD [rdx+rax*SIZEOF_JSAMPLE], xmm5
+ movq XMM_MMWORD [rsi+rax*SIZEOF_JSAMPLE], xmm0
+ mov rdx, JSAMPROW [rdi+5*SIZEOF_JSAMPROW]
+ mov rsi, JSAMPROW [rdi+7*SIZEOF_JSAMPROW]
+ movq XMM_MMWORD [rdx+rax*SIZEOF_JSAMPLE], xmm6
+ movq XMM_MMWORD [rsi+rax*SIZEOF_JSAMPLE], xmm2
+
+ uncollect_args 4
+ mov rsp, rbp ; rsp <- aligned rbp
+ pop rsp ; rsp <- original rbp
+ pop rbp
+ ret
+ ret
+
+; For some reason, the OS X linker does not honor the request to align the
+; segment unless we do this.
+ align 32
diff --git a/simd/x86_64/jidctint-sse2.asm b/simd/x86_64/jidctint-sse2.asm
new file mode 100644
index 0000000..6487266
--- /dev/null
+++ b/simd/x86_64/jidctint-sse2.asm
@@ -0,0 +1,847 @@
+;
+; jidctint.asm - accurate integer IDCT (64-bit SSE2)
+;
+; Copyright 2009 Pierre Ossman <ossman@cendio.se> for Cendio AB
+; Copyright (C) 2009, 2016, D. R. Commander.
+;
+; Based on the x86 SIMD extension for IJG JPEG library
+; Copyright (C) 1999-2006, MIYASAKA Masaru.
+; For conditions of distribution and use, see copyright notice in jsimdext.inc
+;
+; This file should be assembled with NASM (Netwide Assembler),
+; can *not* be assembled with Microsoft's MASM or any compatible
+; assembler (including Borland's Turbo Assembler).
+; NASM is available from http://nasm.sourceforge.net/ or
+; http://sourceforge.net/project/showfiles.php?group_id=6208
+;
+; This file contains a slow-but-accurate integer implementation of the
+; inverse DCT (Discrete Cosine Transform). The following code is based
+; directly on the IJG's original jidctint.c; see the jidctint.c for
+; more details.
+;
+; [TAB8]
+
+%include "jsimdext.inc"
+%include "jdct.inc"
+
+; --------------------------------------------------------------------------
+
+%define CONST_BITS 13
+%define PASS1_BITS 2
+
+%define DESCALE_P1 (CONST_BITS-PASS1_BITS)
+%define DESCALE_P2 (CONST_BITS+PASS1_BITS+3)
+
+%if CONST_BITS == 13
+F_0_298 equ 2446 ; FIX(0.298631336)
+F_0_390 equ 3196 ; FIX(0.390180644)
+F_0_541 equ 4433 ; FIX(0.541196100)
+F_0_765 equ 6270 ; FIX(0.765366865)
+F_0_899 equ 7373 ; FIX(0.899976223)
+F_1_175 equ 9633 ; FIX(1.175875602)
+F_1_501 equ 12299 ; FIX(1.501321110)
+F_1_847 equ 15137 ; FIX(1.847759065)
+F_1_961 equ 16069 ; FIX(1.961570560)
+F_2_053 equ 16819 ; FIX(2.053119869)
+F_2_562 equ 20995 ; FIX(2.562915447)
+F_3_072 equ 25172 ; FIX(3.072711026)
+%else
+; NASM cannot do compile-time arithmetic on floating-point constants.
+%define DESCALE(x,n) (((x)+(1<<((n)-1)))>>(n))
+F_0_298 equ DESCALE( 320652955, 30-CONST_BITS) ; FIX(0.298631336)
+F_0_390 equ DESCALE( 418953276, 30-CONST_BITS) ; FIX(0.390180644)
+F_0_541 equ DESCALE( 581104887, 30-CONST_BITS) ; FIX(0.541196100)
+F_0_765 equ DESCALE( 821806413, 30-CONST_BITS) ; FIX(0.765366865)
+F_0_899 equ DESCALE( 966342111, 30-CONST_BITS) ; FIX(0.899976223)
+F_1_175 equ DESCALE(1262586813, 30-CONST_BITS) ; FIX(1.175875602)
+F_1_501 equ DESCALE(1612031267, 30-CONST_BITS) ; FIX(1.501321110)
+F_1_847 equ DESCALE(1984016188, 30-CONST_BITS) ; FIX(1.847759065)
+F_1_961 equ DESCALE(2106220350, 30-CONST_BITS) ; FIX(1.961570560)
+F_2_053 equ DESCALE(2204520673, 30-CONST_BITS) ; FIX(2.053119869)
+F_2_562 equ DESCALE(2751909506, 30-CONST_BITS) ; FIX(2.562915447)
+F_3_072 equ DESCALE(3299298341, 30-CONST_BITS) ; FIX(3.072711026)
+%endif
+
+; --------------------------------------------------------------------------
+ SECTION SEG_CONST
+
+ alignz 32
+ global EXTN(jconst_idct_islow_sse2)
+
+EXTN(jconst_idct_islow_sse2):
+
+PW_F130_F054 times 4 dw (F_0_541+F_0_765), F_0_541
+PW_F054_MF130 times 4 dw F_0_541, (F_0_541-F_1_847)
+PW_MF078_F117 times 4 dw (F_1_175-F_1_961), F_1_175
+PW_F117_F078 times 4 dw F_1_175, (F_1_175-F_0_390)
+PW_MF060_MF089 times 4 dw (F_0_298-F_0_899),-F_0_899
+PW_MF089_F060 times 4 dw -F_0_899, (F_1_501-F_0_899)
+PW_MF050_MF256 times 4 dw (F_2_053-F_2_562),-F_2_562
+PW_MF256_F050 times 4 dw -F_2_562, (F_3_072-F_2_562)
+PD_DESCALE_P1 times 4 dd 1 << (DESCALE_P1-1)
+PD_DESCALE_P2 times 4 dd 1 << (DESCALE_P2-1)
+PB_CENTERJSAMP times 16 db CENTERJSAMPLE
+
+ alignz 32
+
+; --------------------------------------------------------------------------
+ SECTION SEG_TEXT
+ BITS 64
+;
+; Perform dequantization and inverse DCT on one block of coefficients.
+;
+; GLOBAL(void)
+; jsimd_idct_islow_sse2 (void *dct_table, JCOEFPTR coef_block,
+; JSAMPARRAY output_buf, JDIMENSION output_col)
+;
+
+; r10 = jpeg_component_info *compptr
+; r11 = JCOEFPTR coef_block
+; r12 = JSAMPARRAY output_buf
+; r13d = JDIMENSION output_col
+
+%define original_rbp rbp+0
+%define wk(i) rbp-(WK_NUM-(i))*SIZEOF_XMMWORD ; xmmword wk[WK_NUM]
+%define WK_NUM 12
+
+ align 32
+ global EXTN(jsimd_idct_islow_sse2)
+
+EXTN(jsimd_idct_islow_sse2):
+ push rbp
+ mov rax, rsp ; rax = original rbp
+ sub rsp, byte 4
+ and rsp, byte (-SIZEOF_XMMWORD) ; align to 128 bits
+ mov [rsp], rax
+ mov rbp, rsp ; rbp = aligned rbp
+ lea rsp, [wk(0)]
+ collect_args 4
+
+ ; ---- Pass 1: process columns from input.
+
+ mov rdx, r10 ; quantptr
+ mov rsi, r11 ; inptr
+
+%ifndef NO_ZERO_COLUMN_TEST_ISLOW_SSE2
+ mov eax, DWORD [DWBLOCK(1,0,rsi,SIZEOF_JCOEF)]
+ or eax, DWORD [DWBLOCK(2,0,rsi,SIZEOF_JCOEF)]
+ jnz near .columnDCT
+
+ movdqa xmm0, XMMWORD [XMMBLOCK(1,0,rsi,SIZEOF_JCOEF)]
+ movdqa xmm1, XMMWORD [XMMBLOCK(2,0,rsi,SIZEOF_JCOEF)]
+ por xmm0, XMMWORD [XMMBLOCK(3,0,rsi,SIZEOF_JCOEF)]
+ por xmm1, XMMWORD [XMMBLOCK(4,0,rsi,SIZEOF_JCOEF)]
+ por xmm0, XMMWORD [XMMBLOCK(5,0,rsi,SIZEOF_JCOEF)]
+ por xmm1, XMMWORD [XMMBLOCK(6,0,rsi,SIZEOF_JCOEF)]
+ por xmm0, XMMWORD [XMMBLOCK(7,0,rsi,SIZEOF_JCOEF)]
+ por xmm1, xmm0
+ packsswb xmm1, xmm1
+ packsswb xmm1, xmm1
+ movd eax, xmm1
+ test rax, rax
+ jnz short .columnDCT
+
+ ; -- AC terms all zero
+
+ movdqa xmm5, XMMWORD [XMMBLOCK(0,0,rsi,SIZEOF_JCOEF)]
+ pmullw xmm5, XMMWORD [XMMBLOCK(0,0,rdx,SIZEOF_ISLOW_MULT_TYPE)]
+
+ psllw xmm5, PASS1_BITS
+
+ movdqa xmm4, xmm5 ; xmm5=in0=(00 01 02 03 04 05 06 07)
+ punpcklwd xmm5, xmm5 ; xmm5=(00 00 01 01 02 02 03 03)
+ punpckhwd xmm4, xmm4 ; xmm4=(04 04 05 05 06 06 07 07)
+
+ pshufd xmm7, xmm5, 0x00 ; xmm7=col0=(00 00 00 00 00 00 00 00)
+ pshufd xmm6, xmm5, 0x55 ; xmm6=col1=(01 01 01 01 01 01 01 01)
+ pshufd xmm1, xmm5, 0xAA ; xmm1=col2=(02 02 02 02 02 02 02 02)
+ pshufd xmm5, xmm5, 0xFF ; xmm5=col3=(03 03 03 03 03 03 03 03)
+ pshufd xmm0, xmm4, 0x00 ; xmm0=col4=(04 04 04 04 04 04 04 04)
+ pshufd xmm3, xmm4, 0x55 ; xmm3=col5=(05 05 05 05 05 05 05 05)
+ pshufd xmm2, xmm4, 0xAA ; xmm2=col6=(06 06 06 06 06 06 06 06)
+ pshufd xmm4, xmm4, 0xFF ; xmm4=col7=(07 07 07 07 07 07 07 07)
+
+ movdqa XMMWORD [wk(8)], xmm6 ; wk(8)=col1
+ movdqa XMMWORD [wk(9)], xmm5 ; wk(9)=col3
+ movdqa XMMWORD [wk(10)], xmm3 ; wk(10)=col5
+ movdqa XMMWORD [wk(11)], xmm4 ; wk(11)=col7
+ jmp near .column_end
+%endif
+.columnDCT:
+
+ ; -- Even part
+
+ movdqa xmm0, XMMWORD [XMMBLOCK(0,0,rsi,SIZEOF_JCOEF)]
+ movdqa xmm1, XMMWORD [XMMBLOCK(2,0,rsi,SIZEOF_JCOEF)]
+ pmullw xmm0, XMMWORD [XMMBLOCK(0,0,rdx,SIZEOF_ISLOW_MULT_TYPE)]
+ pmullw xmm1, XMMWORD [XMMBLOCK(2,0,rdx,SIZEOF_ISLOW_MULT_TYPE)]
+ movdqa xmm2, XMMWORD [XMMBLOCK(4,0,rsi,SIZEOF_JCOEF)]
+ movdqa xmm3, XMMWORD [XMMBLOCK(6,0,rsi,SIZEOF_JCOEF)]
+ pmullw xmm2, XMMWORD [XMMBLOCK(4,0,rdx,SIZEOF_ISLOW_MULT_TYPE)]
+ pmullw xmm3, XMMWORD [XMMBLOCK(6,0,rdx,SIZEOF_ISLOW_MULT_TYPE)]
+
+ ; (Original)
+ ; z1 = (z2 + z3) * 0.541196100;
+ ; tmp2 = z1 + z3 * -1.847759065;
+ ; tmp3 = z1 + z2 * 0.765366865;
+ ;
+ ; (This implementation)
+ ; tmp2 = z2 * 0.541196100 + z3 * (0.541196100 - 1.847759065);
+ ; tmp3 = z2 * (0.541196100 + 0.765366865) + z3 * 0.541196100;
+
+ movdqa xmm4, xmm1 ; xmm1=in2=z2
+ movdqa xmm5, xmm1
+ punpcklwd xmm4, xmm3 ; xmm3=in6=z3
+ punpckhwd xmm5, xmm3
+ movdqa xmm1, xmm4
+ movdqa xmm3, xmm5
+ pmaddwd xmm4, [rel PW_F130_F054] ; xmm4=tmp3L
+ pmaddwd xmm5, [rel PW_F130_F054] ; xmm5=tmp3H
+ pmaddwd xmm1, [rel PW_F054_MF130] ; xmm1=tmp2L
+ pmaddwd xmm3, [rel PW_F054_MF130] ; xmm3=tmp2H
+
+ movdqa xmm6, xmm0
+ paddw xmm0, xmm2 ; xmm0=in0+in4
+ psubw xmm6, xmm2 ; xmm6=in0-in4
+
+ pxor xmm7, xmm7
+ pxor xmm2, xmm2
+ punpcklwd xmm7, xmm0 ; xmm7=tmp0L
+ punpckhwd xmm2, xmm0 ; xmm2=tmp0H
+ psrad xmm7, (16-CONST_BITS) ; psrad xmm7,16 & pslld xmm7,CONST_BITS
+ psrad xmm2, (16-CONST_BITS) ; psrad xmm2,16 & pslld xmm2,CONST_BITS
+
+ movdqa xmm0, xmm7
+ paddd xmm7, xmm4 ; xmm7=tmp10L
+ psubd xmm0, xmm4 ; xmm0=tmp13L
+ movdqa xmm4, xmm2
+ paddd xmm2, xmm5 ; xmm2=tmp10H
+ psubd xmm4, xmm5 ; xmm4=tmp13H
+
+ movdqa XMMWORD [wk(0)], xmm7 ; wk(0)=tmp10L
+ movdqa XMMWORD [wk(1)], xmm2 ; wk(1)=tmp10H
+ movdqa XMMWORD [wk(2)], xmm0 ; wk(2)=tmp13L
+ movdqa XMMWORD [wk(3)], xmm4 ; wk(3)=tmp13H
+
+ pxor xmm5, xmm5
+ pxor xmm7, xmm7
+ punpcklwd xmm5, xmm6 ; xmm5=tmp1L
+ punpckhwd xmm7, xmm6 ; xmm7=tmp1H
+ psrad xmm5, (16-CONST_BITS) ; psrad xmm5,16 & pslld xmm5,CONST_BITS
+ psrad xmm7, (16-CONST_BITS) ; psrad xmm7,16 & pslld xmm7,CONST_BITS
+
+ movdqa xmm2, xmm5
+ paddd xmm5, xmm1 ; xmm5=tmp11L
+ psubd xmm2, xmm1 ; xmm2=tmp12L
+ movdqa xmm0, xmm7
+ paddd xmm7, xmm3 ; xmm7=tmp11H
+ psubd xmm0, xmm3 ; xmm0=tmp12H
+
+ movdqa XMMWORD [wk(4)], xmm5 ; wk(4)=tmp11L
+ movdqa XMMWORD [wk(5)], xmm7 ; wk(5)=tmp11H
+ movdqa XMMWORD [wk(6)], xmm2 ; wk(6)=tmp12L
+ movdqa XMMWORD [wk(7)], xmm0 ; wk(7)=tmp12H
+
+ ; -- Odd part
+
+ movdqa xmm4, XMMWORD [XMMBLOCK(1,0,rsi,SIZEOF_JCOEF)]
+ movdqa xmm6, XMMWORD [XMMBLOCK(3,0,rsi,SIZEOF_JCOEF)]
+ pmullw xmm4, XMMWORD [XMMBLOCK(1,0,rdx,SIZEOF_ISLOW_MULT_TYPE)]
+ pmullw xmm6, XMMWORD [XMMBLOCK(3,0,rdx,SIZEOF_ISLOW_MULT_TYPE)]
+ movdqa xmm1, XMMWORD [XMMBLOCK(5,0,rsi,SIZEOF_JCOEF)]
+ movdqa xmm3, XMMWORD [XMMBLOCK(7,0,rsi,SIZEOF_JCOEF)]
+ pmullw xmm1, XMMWORD [XMMBLOCK(5,0,rdx,SIZEOF_ISLOW_MULT_TYPE)]
+ pmullw xmm3, XMMWORD [XMMBLOCK(7,0,rdx,SIZEOF_ISLOW_MULT_TYPE)]
+
+ movdqa xmm5, xmm6
+ movdqa xmm7, xmm4
+ paddw xmm5, xmm3 ; xmm5=z3
+ paddw xmm7, xmm1 ; xmm7=z4
+
+ ; (Original)
+ ; z5 = (z3 + z4) * 1.175875602;
+ ; z3 = z3 * -1.961570560; z4 = z4 * -0.390180644;
+ ; z3 += z5; z4 += z5;
+ ;
+ ; (This implementation)
+ ; z3 = z3 * (1.175875602 - 1.961570560) + z4 * 1.175875602;
+ ; z4 = z3 * 1.175875602 + z4 * (1.175875602 - 0.390180644);
+
+ movdqa xmm2, xmm5
+ movdqa xmm0, xmm5
+ punpcklwd xmm2, xmm7
+ punpckhwd xmm0, xmm7
+ movdqa xmm5, xmm2
+ movdqa xmm7, xmm0
+ pmaddwd xmm2, [rel PW_MF078_F117] ; xmm2=z3L
+ pmaddwd xmm0, [rel PW_MF078_F117] ; xmm0=z3H
+ pmaddwd xmm5, [rel PW_F117_F078] ; xmm5=z4L
+ pmaddwd xmm7, [rel PW_F117_F078] ; xmm7=z4H
+
+ movdqa XMMWORD [wk(10)], xmm2 ; wk(10)=z3L
+ movdqa XMMWORD [wk(11)], xmm0 ; wk(11)=z3H
+
+ ; (Original)
+ ; z1 = tmp0 + tmp3; z2 = tmp1 + tmp2;
+ ; tmp0 = tmp0 * 0.298631336; tmp1 = tmp1 * 2.053119869;
+ ; tmp2 = tmp2 * 3.072711026; tmp3 = tmp3 * 1.501321110;
+ ; z1 = z1 * -0.899976223; z2 = z2 * -2.562915447;
+ ; tmp0 += z1 + z3; tmp1 += z2 + z4;
+ ; tmp2 += z2 + z3; tmp3 += z1 + z4;
+ ;
+ ; (This implementation)
+ ; tmp0 = tmp0 * (0.298631336 - 0.899976223) + tmp3 * -0.899976223;
+ ; tmp1 = tmp1 * (2.053119869 - 2.562915447) + tmp2 * -2.562915447;
+ ; tmp2 = tmp1 * -2.562915447 + tmp2 * (3.072711026 - 2.562915447);
+ ; tmp3 = tmp0 * -0.899976223 + tmp3 * (1.501321110 - 0.899976223);
+ ; tmp0 += z3; tmp1 += z4;
+ ; tmp2 += z3; tmp3 += z4;
+
+ movdqa xmm2, xmm3
+ movdqa xmm0, xmm3
+ punpcklwd xmm2, xmm4
+ punpckhwd xmm0, xmm4
+ movdqa xmm3, xmm2
+ movdqa xmm4, xmm0
+ pmaddwd xmm2, [rel PW_MF060_MF089] ; xmm2=tmp0L
+ pmaddwd xmm0, [rel PW_MF060_MF089] ; xmm0=tmp0H
+ pmaddwd xmm3, [rel PW_MF089_F060] ; xmm3=tmp3L
+ pmaddwd xmm4, [rel PW_MF089_F060] ; xmm4=tmp3H
+
+ paddd xmm2, XMMWORD [wk(10)] ; xmm2=tmp0L
+ paddd xmm0, XMMWORD [wk(11)] ; xmm0=tmp0H
+ paddd xmm3, xmm5 ; xmm3=tmp3L
+ paddd xmm4, xmm7 ; xmm4=tmp3H
+
+ movdqa XMMWORD [wk(8)], xmm2 ; wk(8)=tmp0L
+ movdqa XMMWORD [wk(9)], xmm0 ; wk(9)=tmp0H
+
+ movdqa xmm2, xmm1
+ movdqa xmm0, xmm1
+ punpcklwd xmm2, xmm6
+ punpckhwd xmm0, xmm6
+ movdqa xmm1, xmm2
+ movdqa xmm6, xmm0
+ pmaddwd xmm2, [rel PW_MF050_MF256] ; xmm2=tmp1L
+ pmaddwd xmm0, [rel PW_MF050_MF256] ; xmm0=tmp1H
+ pmaddwd xmm1, [rel PW_MF256_F050] ; xmm1=tmp2L
+ pmaddwd xmm6, [rel PW_MF256_F050] ; xmm6=tmp2H
+
+ paddd xmm2, xmm5 ; xmm2=tmp1L
+ paddd xmm0, xmm7 ; xmm0=tmp1H
+ paddd xmm1, XMMWORD [wk(10)] ; xmm1=tmp2L
+ paddd xmm6, XMMWORD [wk(11)] ; xmm6=tmp2H
+
+ movdqa XMMWORD [wk(10)], xmm2 ; wk(10)=tmp1L
+ movdqa XMMWORD [wk(11)], xmm0 ; wk(11)=tmp1H
+
+ ; -- Final output stage
+
+ movdqa xmm5, XMMWORD [wk(0)] ; xmm5=tmp10L
+ movdqa xmm7, XMMWORD [wk(1)] ; xmm7=tmp10H
+
+ movdqa xmm2, xmm5
+ movdqa xmm0, xmm7
+ paddd xmm5, xmm3 ; xmm5=data0L
+ paddd xmm7, xmm4 ; xmm7=data0H
+ psubd xmm2, xmm3 ; xmm2=data7L
+ psubd xmm0, xmm4 ; xmm0=data7H
+
+ movdqa xmm3, [rel PD_DESCALE_P1] ; xmm3=[rel PD_DESCALE_P1]
+
+ paddd xmm5, xmm3
+ paddd xmm7, xmm3
+ psrad xmm5, DESCALE_P1
+ psrad xmm7, DESCALE_P1
+ paddd xmm2, xmm3
+ paddd xmm0, xmm3
+ psrad xmm2, DESCALE_P1
+ psrad xmm0, DESCALE_P1
+
+ packssdw xmm5, xmm7 ; xmm5=data0=(00 01 02 03 04 05 06 07)
+ packssdw xmm2, xmm0 ; xmm2=data7=(70 71 72 73 74 75 76 77)
+
+ movdqa xmm4, XMMWORD [wk(4)] ; xmm4=tmp11L
+ movdqa xmm3, XMMWORD [wk(5)] ; xmm3=tmp11H
+
+ movdqa xmm7, xmm4
+ movdqa xmm0, xmm3
+ paddd xmm4, xmm1 ; xmm4=data1L
+ paddd xmm3, xmm6 ; xmm3=data1H
+ psubd xmm7, xmm1 ; xmm7=data6L
+ psubd xmm0, xmm6 ; xmm0=data6H
+
+ movdqa xmm1, [rel PD_DESCALE_P1] ; xmm1=[rel PD_DESCALE_P1]
+
+ paddd xmm4, xmm1
+ paddd xmm3, xmm1
+ psrad xmm4, DESCALE_P1
+ psrad xmm3, DESCALE_P1
+ paddd xmm7, xmm1
+ paddd xmm0, xmm1
+ psrad xmm7, DESCALE_P1
+ psrad xmm0, DESCALE_P1
+
+ packssdw xmm4, xmm3 ; xmm4=data1=(10 11 12 13 14 15 16 17)
+ packssdw xmm7, xmm0 ; xmm7=data6=(60 61 62 63 64 65 66 67)
+
+ movdqa xmm6, xmm5 ; transpose coefficients(phase 1)
+ punpcklwd xmm5, xmm4 ; xmm5=(00 10 01 11 02 12 03 13)
+ punpckhwd xmm6, xmm4 ; xmm6=(04 14 05 15 06 16 07 17)
+ movdqa xmm1, xmm7 ; transpose coefficients(phase 1)
+ punpcklwd xmm7, xmm2 ; xmm7=(60 70 61 71 62 72 63 73)
+ punpckhwd xmm1, xmm2 ; xmm1=(64 74 65 75 66 76 67 77)
+
+ movdqa xmm3, XMMWORD [wk(6)] ; xmm3=tmp12L
+ movdqa xmm0, XMMWORD [wk(7)] ; xmm0=tmp12H
+ movdqa xmm4, XMMWORD [wk(10)] ; xmm4=tmp1L
+ movdqa xmm2, XMMWORD [wk(11)] ; xmm2=tmp1H
+
+ movdqa XMMWORD [wk(0)], xmm5 ; wk(0)=(00 10 01 11 02 12 03 13)
+ movdqa XMMWORD [wk(1)], xmm6 ; wk(1)=(04 14 05 15 06 16 07 17)
+ movdqa XMMWORD [wk(4)], xmm7 ; wk(4)=(60 70 61 71 62 72 63 73)
+ movdqa XMMWORD [wk(5)], xmm1 ; wk(5)=(64 74 65 75 66 76 67 77)
+
+ movdqa xmm5, xmm3
+ movdqa xmm6, xmm0
+ paddd xmm3, xmm4 ; xmm3=data2L
+ paddd xmm0, xmm2 ; xmm0=data2H
+ psubd xmm5, xmm4 ; xmm5=data5L
+ psubd xmm6, xmm2 ; xmm6=data5H
+
+ movdqa xmm7, [rel PD_DESCALE_P1] ; xmm7=[rel PD_DESCALE_P1]
+
+ paddd xmm3, xmm7
+ paddd xmm0, xmm7
+ psrad xmm3, DESCALE_P1
+ psrad xmm0, DESCALE_P1
+ paddd xmm5, xmm7
+ paddd xmm6, xmm7
+ psrad xmm5, DESCALE_P1
+ psrad xmm6, DESCALE_P1
+
+ packssdw xmm3, xmm0 ; xmm3=data2=(20 21 22 23 24 25 26 27)
+ packssdw xmm5, xmm6 ; xmm5=data5=(50 51 52 53 54 55 56 57)
+
+ movdqa xmm1, XMMWORD [wk(2)] ; xmm1=tmp13L
+ movdqa xmm4, XMMWORD [wk(3)] ; xmm4=tmp13H
+ movdqa xmm2, XMMWORD [wk(8)] ; xmm2=tmp0L
+ movdqa xmm7, XMMWORD [wk(9)] ; xmm7=tmp0H
+
+ movdqa xmm0, xmm1
+ movdqa xmm6, xmm4
+ paddd xmm1, xmm2 ; xmm1=data3L
+ paddd xmm4, xmm7 ; xmm4=data3H
+ psubd xmm0, xmm2 ; xmm0=data4L
+ psubd xmm6, xmm7 ; xmm6=data4H
+
+ movdqa xmm2, [rel PD_DESCALE_P1] ; xmm2=[rel PD_DESCALE_P1]
+
+ paddd xmm1, xmm2
+ paddd xmm4, xmm2
+ psrad xmm1, DESCALE_P1
+ psrad xmm4, DESCALE_P1
+ paddd xmm0, xmm2
+ paddd xmm6, xmm2
+ psrad xmm0, DESCALE_P1
+ psrad xmm6, DESCALE_P1
+
+ packssdw xmm1, xmm4 ; xmm1=data3=(30 31 32 33 34 35 36 37)
+ packssdw xmm0, xmm6 ; xmm0=data4=(40 41 42 43 44 45 46 47)
+
+ movdqa xmm7, XMMWORD [wk(0)] ; xmm7=(00 10 01 11 02 12 03 13)
+ movdqa xmm2, XMMWORD [wk(1)] ; xmm2=(04 14 05 15 06 16 07 17)
+
+ movdqa xmm4, xmm3 ; transpose coefficients(phase 1)
+ punpcklwd xmm3, xmm1 ; xmm3=(20 30 21 31 22 32 23 33)
+ punpckhwd xmm4, xmm1 ; xmm4=(24 34 25 35 26 36 27 37)
+ movdqa xmm6, xmm0 ; transpose coefficients(phase 1)
+ punpcklwd xmm0, xmm5 ; xmm0=(40 50 41 51 42 52 43 53)
+ punpckhwd xmm6, xmm5 ; xmm6=(44 54 45 55 46 56 47 57)
+
+ movdqa xmm1, xmm7 ; transpose coefficients(phase 2)
+ punpckldq xmm7, xmm3 ; xmm7=(00 10 20 30 01 11 21 31)
+ punpckhdq xmm1, xmm3 ; xmm1=(02 12 22 32 03 13 23 33)
+ movdqa xmm5, xmm2 ; transpose coefficients(phase 2)
+ punpckldq xmm2, xmm4 ; xmm2=(04 14 24 34 05 15 25 35)
+ punpckhdq xmm5, xmm4 ; xmm5=(06 16 26 36 07 17 27 37)
+
+ movdqa xmm3, XMMWORD [wk(4)] ; xmm3=(60 70 61 71 62 72 63 73)
+ movdqa xmm4, XMMWORD [wk(5)] ; xmm4=(64 74 65 75 66 76 67 77)
+
+ movdqa XMMWORD [wk(6)], xmm2 ; wk(6)=(04 14 24 34 05 15 25 35)
+ movdqa XMMWORD [wk(7)], xmm5 ; wk(7)=(06 16 26 36 07 17 27 37)
+
+ movdqa xmm2, xmm0 ; transpose coefficients(phase 2)
+ punpckldq xmm0, xmm3 ; xmm0=(40 50 60 70 41 51 61 71)
+ punpckhdq xmm2, xmm3 ; xmm2=(42 52 62 72 43 53 63 73)
+ movdqa xmm5, xmm6 ; transpose coefficients(phase 2)
+ punpckldq xmm6, xmm4 ; xmm6=(44 54 64 74 45 55 65 75)
+ punpckhdq xmm5, xmm4 ; xmm5=(46 56 66 76 47 57 67 77)
+
+ movdqa xmm3, xmm7 ; transpose coefficients(phase 3)
+ punpcklqdq xmm7, xmm0 ; xmm7=col0=(00 10 20 30 40 50 60 70)
+ punpckhqdq xmm3, xmm0 ; xmm3=col1=(01 11 21 31 41 51 61 71)
+ movdqa xmm4, xmm1 ; transpose coefficients(phase 3)
+ punpcklqdq xmm1, xmm2 ; xmm1=col2=(02 12 22 32 42 52 62 72)
+ punpckhqdq xmm4, xmm2 ; xmm4=col3=(03 13 23 33 43 53 63 73)
+
+ movdqa xmm0, XMMWORD [wk(6)] ; xmm0=(04 14 24 34 05 15 25 35)
+ movdqa xmm2, XMMWORD [wk(7)] ; xmm2=(06 16 26 36 07 17 27 37)
+
+ movdqa XMMWORD [wk(8)], xmm3 ; wk(8)=col1
+ movdqa XMMWORD [wk(9)], xmm4 ; wk(9)=col3
+
+ movdqa xmm3, xmm0 ; transpose coefficients(phase 3)
+ punpcklqdq xmm0, xmm6 ; xmm0=col4=(04 14 24 34 44 54 64 74)
+ punpckhqdq xmm3, xmm6 ; xmm3=col5=(05 15 25 35 45 55 65 75)
+ movdqa xmm4, xmm2 ; transpose coefficients(phase 3)
+ punpcklqdq xmm2, xmm5 ; xmm2=col6=(06 16 26 36 46 56 66 76)
+ punpckhqdq xmm4, xmm5 ; xmm4=col7=(07 17 27 37 47 57 67 77)
+
+ movdqa XMMWORD [wk(10)], xmm3 ; wk(10)=col5
+ movdqa XMMWORD [wk(11)], xmm4 ; wk(11)=col7
+.column_end:
+
+ ; -- Prefetch the next coefficient block
+
+ prefetchnta [rsi + DCTSIZE2*SIZEOF_JCOEF + 0*32]
+ prefetchnta [rsi + DCTSIZE2*SIZEOF_JCOEF + 1*32]
+ prefetchnta [rsi + DCTSIZE2*SIZEOF_JCOEF + 2*32]
+ prefetchnta [rsi + DCTSIZE2*SIZEOF_JCOEF + 3*32]
+
+ ; ---- Pass 2: process rows from work array, store into output array.
+
+ mov rax, [original_rbp]
+ mov rdi, r12 ; (JSAMPROW *)
+ mov eax, r13d
+
+ ; -- Even part
+
+ ; xmm7=col0, xmm1=col2, xmm0=col4, xmm2=col6
+
+ ; (Original)
+ ; z1 = (z2 + z3) * 0.541196100;
+ ; tmp2 = z1 + z3 * -1.847759065;
+ ; tmp3 = z1 + z2 * 0.765366865;
+ ;
+ ; (This implementation)
+ ; tmp2 = z2 * 0.541196100 + z3 * (0.541196100 - 1.847759065);
+ ; tmp3 = z2 * (0.541196100 + 0.765366865) + z3 * 0.541196100;
+
+ movdqa xmm6, xmm1 ; xmm1=in2=z2
+ movdqa xmm5, xmm1
+ punpcklwd xmm6, xmm2 ; xmm2=in6=z3
+ punpckhwd xmm5, xmm2
+ movdqa xmm1, xmm6
+ movdqa xmm2, xmm5
+ pmaddwd xmm6, [rel PW_F130_F054] ; xmm6=tmp3L
+ pmaddwd xmm5, [rel PW_F130_F054] ; xmm5=tmp3H
+ pmaddwd xmm1, [rel PW_F054_MF130] ; xmm1=tmp2L
+ pmaddwd xmm2, [rel PW_F054_MF130] ; xmm2=tmp2H
+
+ movdqa xmm3, xmm7
+ paddw xmm7, xmm0 ; xmm7=in0+in4
+ psubw xmm3, xmm0 ; xmm3=in0-in4
+
+ pxor xmm4, xmm4
+ pxor xmm0, xmm0
+ punpcklwd xmm4, xmm7 ; xmm4=tmp0L
+ punpckhwd xmm0, xmm7 ; xmm0=tmp0H
+ psrad xmm4, (16-CONST_BITS) ; psrad xmm4,16 & pslld xmm4,CONST_BITS
+ psrad xmm0, (16-CONST_BITS) ; psrad xmm0,16 & pslld xmm0,CONST_BITS
+
+ movdqa xmm7, xmm4
+ paddd xmm4, xmm6 ; xmm4=tmp10L
+ psubd xmm7, xmm6 ; xmm7=tmp13L
+ movdqa xmm6, xmm0
+ paddd xmm0, xmm5 ; xmm0=tmp10H
+ psubd xmm6, xmm5 ; xmm6=tmp13H
+
+ movdqa XMMWORD [wk(0)], xmm4 ; wk(0)=tmp10L
+ movdqa XMMWORD [wk(1)], xmm0 ; wk(1)=tmp10H
+ movdqa XMMWORD [wk(2)], xmm7 ; wk(2)=tmp13L
+ movdqa XMMWORD [wk(3)], xmm6 ; wk(3)=tmp13H
+
+ pxor xmm5, xmm5
+ pxor xmm4, xmm4
+ punpcklwd xmm5, xmm3 ; xmm5=tmp1L
+ punpckhwd xmm4, xmm3 ; xmm4=tmp1H
+ psrad xmm5, (16-CONST_BITS) ; psrad xmm5,16 & pslld xmm5,CONST_BITS
+ psrad xmm4, (16-CONST_BITS) ; psrad xmm4,16 & pslld xmm4,CONST_BITS
+
+ movdqa xmm0, xmm5
+ paddd xmm5, xmm1 ; xmm5=tmp11L
+ psubd xmm0, xmm1 ; xmm0=tmp12L
+ movdqa xmm7, xmm4
+ paddd xmm4, xmm2 ; xmm4=tmp11H
+ psubd xmm7, xmm2 ; xmm7=tmp12H
+
+ movdqa XMMWORD [wk(4)], xmm5 ; wk(4)=tmp11L
+ movdqa XMMWORD [wk(5)], xmm4 ; wk(5)=tmp11H
+ movdqa XMMWORD [wk(6)], xmm0 ; wk(6)=tmp12L
+ movdqa XMMWORD [wk(7)], xmm7 ; wk(7)=tmp12H
+
+ ; -- Odd part
+
+ movdqa xmm6, XMMWORD [wk(9)] ; xmm6=col3
+ movdqa xmm3, XMMWORD [wk(8)] ; xmm3=col1
+ movdqa xmm1, XMMWORD [wk(11)] ; xmm1=col7
+ movdqa xmm2, XMMWORD [wk(10)] ; xmm2=col5
+
+ movdqa xmm5, xmm6
+ movdqa xmm4, xmm3
+ paddw xmm5, xmm1 ; xmm5=z3
+ paddw xmm4, xmm2 ; xmm4=z4
+
+ ; (Original)
+ ; z5 = (z3 + z4) * 1.175875602;
+ ; z3 = z3 * -1.961570560; z4 = z4 * -0.390180644;
+ ; z3 += z5; z4 += z5;
+ ;
+ ; (This implementation)
+ ; z3 = z3 * (1.175875602 - 1.961570560) + z4 * 1.175875602;
+ ; z4 = z3 * 1.175875602 + z4 * (1.175875602 - 0.390180644);
+
+ movdqa xmm0, xmm5
+ movdqa xmm7, xmm5
+ punpcklwd xmm0, xmm4
+ punpckhwd xmm7, xmm4
+ movdqa xmm5, xmm0
+ movdqa xmm4, xmm7
+ pmaddwd xmm0, [rel PW_MF078_F117] ; xmm0=z3L
+ pmaddwd xmm7, [rel PW_MF078_F117] ; xmm7=z3H
+ pmaddwd xmm5, [rel PW_F117_F078] ; xmm5=z4L
+ pmaddwd xmm4, [rel PW_F117_F078] ; xmm4=z4H
+
+ movdqa XMMWORD [wk(10)], xmm0 ; wk(10)=z3L
+ movdqa XMMWORD [wk(11)], xmm7 ; wk(11)=z3H
+
+ ; (Original)
+ ; z1 = tmp0 + tmp3; z2 = tmp1 + tmp2;
+ ; tmp0 = tmp0 * 0.298631336; tmp1 = tmp1 * 2.053119869;
+ ; tmp2 = tmp2 * 3.072711026; tmp3 = tmp3 * 1.501321110;
+ ; z1 = z1 * -0.899976223; z2 = z2 * -2.562915447;
+ ; tmp0 += z1 + z3; tmp1 += z2 + z4;
+ ; tmp2 += z2 + z3; tmp3 += z1 + z4;
+ ;
+ ; (This implementation)
+ ; tmp0 = tmp0 * (0.298631336 - 0.899976223) + tmp3 * -0.899976223;
+ ; tmp1 = tmp1 * (2.053119869 - 2.562915447) + tmp2 * -2.562915447;
+ ; tmp2 = tmp1 * -2.562915447 + tmp2 * (3.072711026 - 2.562915447);
+ ; tmp3 = tmp0 * -0.899976223 + tmp3 * (1.501321110 - 0.899976223);
+ ; tmp0 += z3; tmp1 += z4;
+ ; tmp2 += z3; tmp3 += z4;
+
+ movdqa xmm0, xmm1
+ movdqa xmm7, xmm1
+ punpcklwd xmm0, xmm3
+ punpckhwd xmm7, xmm3
+ movdqa xmm1, xmm0
+ movdqa xmm3, xmm7
+ pmaddwd xmm0, [rel PW_MF060_MF089] ; xmm0=tmp0L
+ pmaddwd xmm7, [rel PW_MF060_MF089] ; xmm7=tmp0H
+ pmaddwd xmm1, [rel PW_MF089_F060] ; xmm1=tmp3L
+ pmaddwd xmm3, [rel PW_MF089_F060] ; xmm3=tmp3H
+
+ paddd xmm0, XMMWORD [wk(10)] ; xmm0=tmp0L
+ paddd xmm7, XMMWORD [wk(11)] ; xmm7=tmp0H
+ paddd xmm1, xmm5 ; xmm1=tmp3L
+ paddd xmm3, xmm4 ; xmm3=tmp3H
+
+ movdqa XMMWORD [wk(8)], xmm0 ; wk(8)=tmp0L
+ movdqa XMMWORD [wk(9)], xmm7 ; wk(9)=tmp0H
+
+ movdqa xmm0, xmm2
+ movdqa xmm7, xmm2
+ punpcklwd xmm0, xmm6
+ punpckhwd xmm7, xmm6
+ movdqa xmm2, xmm0
+ movdqa xmm6, xmm7
+ pmaddwd xmm0, [rel PW_MF050_MF256] ; xmm0=tmp1L
+ pmaddwd xmm7, [rel PW_MF050_MF256] ; xmm7=tmp1H
+ pmaddwd xmm2, [rel PW_MF256_F050] ; xmm2=tmp2L
+ pmaddwd xmm6, [rel PW_MF256_F050] ; xmm6=tmp2H
+
+ paddd xmm0, xmm5 ; xmm0=tmp1L
+ paddd xmm7, xmm4 ; xmm7=tmp1H
+ paddd xmm2, XMMWORD [wk(10)] ; xmm2=tmp2L
+ paddd xmm6, XMMWORD [wk(11)] ; xmm6=tmp2H
+
+ movdqa XMMWORD [wk(10)], xmm0 ; wk(10)=tmp1L
+ movdqa XMMWORD [wk(11)], xmm7 ; wk(11)=tmp1H
+
+ ; -- Final output stage
+
+ movdqa xmm5, XMMWORD [wk(0)] ; xmm5=tmp10L
+ movdqa xmm4, XMMWORD [wk(1)] ; xmm4=tmp10H
+
+ movdqa xmm0, xmm5
+ movdqa xmm7, xmm4
+ paddd xmm5, xmm1 ; xmm5=data0L
+ paddd xmm4, xmm3 ; xmm4=data0H
+ psubd xmm0, xmm1 ; xmm0=data7L
+ psubd xmm7, xmm3 ; xmm7=data7H
+
+ movdqa xmm1, [rel PD_DESCALE_P2] ; xmm1=[rel PD_DESCALE_P2]
+
+ paddd xmm5, xmm1
+ paddd xmm4, xmm1
+ psrad xmm5, DESCALE_P2
+ psrad xmm4, DESCALE_P2
+ paddd xmm0, xmm1
+ paddd xmm7, xmm1
+ psrad xmm0, DESCALE_P2
+ psrad xmm7, DESCALE_P2
+
+ packssdw xmm5, xmm4 ; xmm5=data0=(00 10 20 30 40 50 60 70)
+ packssdw xmm0, xmm7 ; xmm0=data7=(07 17 27 37 47 57 67 77)
+
+ movdqa xmm3, XMMWORD [wk(4)] ; xmm3=tmp11L
+ movdqa xmm1, XMMWORD [wk(5)] ; xmm1=tmp11H
+
+ movdqa xmm4, xmm3
+ movdqa xmm7, xmm1
+ paddd xmm3, xmm2 ; xmm3=data1L
+ paddd xmm1, xmm6 ; xmm1=data1H
+ psubd xmm4, xmm2 ; xmm4=data6L
+ psubd xmm7, xmm6 ; xmm7=data6H
+
+ movdqa xmm2, [rel PD_DESCALE_P2] ; xmm2=[rel PD_DESCALE_P2]
+
+ paddd xmm3, xmm2
+ paddd xmm1, xmm2
+ psrad xmm3, DESCALE_P2
+ psrad xmm1, DESCALE_P2
+ paddd xmm4, xmm2
+ paddd xmm7, xmm2
+ psrad xmm4, DESCALE_P2
+ psrad xmm7, DESCALE_P2
+
+ packssdw xmm3, xmm1 ; xmm3=data1=(01 11 21 31 41 51 61 71)
+ packssdw xmm4, xmm7 ; xmm4=data6=(06 16 26 36 46 56 66 76)
+
+ packsswb xmm5, xmm4 ; xmm5=(00 10 20 30 40 50 60 70 06 16 26 36 46 56 66 76)
+ packsswb xmm3, xmm0 ; xmm3=(01 11 21 31 41 51 61 71 07 17 27 37 47 57 67 77)
+
+ movdqa xmm6, XMMWORD [wk(6)] ; xmm6=tmp12L
+ movdqa xmm2, XMMWORD [wk(7)] ; xmm2=tmp12H
+ movdqa xmm1, XMMWORD [wk(10)] ; xmm1=tmp1L
+ movdqa xmm7, XMMWORD [wk(11)] ; xmm7=tmp1H
+
+ movdqa XMMWORD [wk(0)], xmm5 ; wk(0)=(00 10 20 30 40 50 60 70 06 16 26 36 46 56 66 76)
+ movdqa XMMWORD [wk(1)], xmm3 ; wk(1)=(01 11 21 31 41 51 61 71 07 17 27 37 47 57 67 77)
+
+ movdqa xmm4, xmm6
+ movdqa xmm0, xmm2
+ paddd xmm6, xmm1 ; xmm6=data2L
+ paddd xmm2, xmm7 ; xmm2=data2H
+ psubd xmm4, xmm1 ; xmm4=data5L
+ psubd xmm0, xmm7 ; xmm0=data5H
+
+ movdqa xmm5, [rel PD_DESCALE_P2] ; xmm5=[rel PD_DESCALE_P2]
+
+ paddd xmm6, xmm5
+ paddd xmm2, xmm5
+ psrad xmm6, DESCALE_P2
+ psrad xmm2, DESCALE_P2
+ paddd xmm4, xmm5
+ paddd xmm0, xmm5
+ psrad xmm4, DESCALE_P2
+ psrad xmm0, DESCALE_P2
+
+ packssdw xmm6, xmm2 ; xmm6=data2=(02 12 22 32 42 52 62 72)
+ packssdw xmm4, xmm0 ; xmm4=data5=(05 15 25 35 45 55 65 75)
+
+ movdqa xmm3, XMMWORD [wk(2)] ; xmm3=tmp13L
+ movdqa xmm1, XMMWORD [wk(3)] ; xmm1=tmp13H
+ movdqa xmm7, XMMWORD [wk(8)] ; xmm7=tmp0L
+ movdqa xmm5, XMMWORD [wk(9)] ; xmm5=tmp0H
+
+ movdqa xmm2, xmm3
+ movdqa xmm0, xmm1
+ paddd xmm3, xmm7 ; xmm3=data3L
+ paddd xmm1, xmm5 ; xmm1=data3H
+ psubd xmm2, xmm7 ; xmm2=data4L
+ psubd xmm0, xmm5 ; xmm0=data4H
+
+ movdqa xmm7, [rel PD_DESCALE_P2] ; xmm7=[rel PD_DESCALE_P2]
+
+ paddd xmm3, xmm7
+ paddd xmm1, xmm7
+ psrad xmm3, DESCALE_P2
+ psrad xmm1, DESCALE_P2
+ paddd xmm2, xmm7
+ paddd xmm0, xmm7
+ psrad xmm2, DESCALE_P2
+ psrad xmm0, DESCALE_P2
+
+ movdqa xmm5, [rel PB_CENTERJSAMP] ; xmm5=[rel PB_CENTERJSAMP]
+
+ packssdw xmm3, xmm1 ; xmm3=data3=(03 13 23 33 43 53 63 73)
+ packssdw xmm2, xmm0 ; xmm2=data4=(04 14 24 34 44 54 64 74)
+
+ movdqa xmm7, XMMWORD [wk(0)] ; xmm7=(00 10 20 30 40 50 60 70 06 16 26 36 46 56 66 76)
+ movdqa xmm1, XMMWORD [wk(1)] ; xmm1=(01 11 21 31 41 51 61 71 07 17 27 37 47 57 67 77)
+
+ packsswb xmm6, xmm2 ; xmm6=(02 12 22 32 42 52 62 72 04 14 24 34 44 54 64 74)
+ packsswb xmm3, xmm4 ; xmm3=(03 13 23 33 43 53 63 73 05 15 25 35 45 55 65 75)
+
+ paddb xmm7, xmm5
+ paddb xmm1, xmm5
+ paddb xmm6, xmm5
+ paddb xmm3, xmm5
+
+ movdqa xmm0, xmm7 ; transpose coefficients(phase 1)
+ punpcklbw xmm7, xmm1 ; xmm7=(00 01 10 11 20 21 30 31 40 41 50 51 60 61 70 71)
+ punpckhbw xmm0, xmm1 ; xmm0=(06 07 16 17 26 27 36 37 46 47 56 57 66 67 76 77)
+ movdqa xmm2, xmm6 ; transpose coefficients(phase 1)
+ punpcklbw xmm6, xmm3 ; xmm6=(02 03 12 13 22 23 32 33 42 43 52 53 62 63 72 73)
+ punpckhbw xmm2, xmm3 ; xmm2=(04 05 14 15 24 25 34 35 44 45 54 55 64 65 74 75)
+
+ movdqa xmm4, xmm7 ; transpose coefficients(phase 2)
+ punpcklwd xmm7, xmm6 ; xmm7=(00 01 02 03 10 11 12 13 20 21 22 23 30 31 32 33)
+ punpckhwd xmm4, xmm6 ; xmm4=(40 41 42 43 50 51 52 53 60 61 62 63 70 71 72 73)
+ movdqa xmm5, xmm2 ; transpose coefficients(phase 2)
+ punpcklwd xmm2, xmm0 ; xmm2=(04 05 06 07 14 15 16 17 24 25 26 27 34 35 36 37)
+ punpckhwd xmm5, xmm0 ; xmm5=(44 45 46 47 54 55 56 57 64 65 66 67 74 75 76 77)
+
+ movdqa xmm1, xmm7 ; transpose coefficients(phase 3)
+ punpckldq xmm7, xmm2 ; xmm7=(00 01 02 03 04 05 06 07 10 11 12 13 14 15 16 17)
+ punpckhdq xmm1, xmm2 ; xmm1=(20 21 22 23 24 25 26 27 30 31 32 33 34 35 36 37)
+ movdqa xmm3, xmm4 ; transpose coefficients(phase 3)
+ punpckldq xmm4, xmm5 ; xmm4=(40 41 42 43 44 45 46 47 50 51 52 53 54 55 56 57)
+ punpckhdq xmm3, xmm5 ; xmm3=(60 61 62 63 64 65 66 67 70 71 72 73 74 75 76 77)
+
+ pshufd xmm6, xmm7, 0x4E ; xmm6=(10 11 12 13 14 15 16 17 00 01 02 03 04 05 06 07)
+ pshufd xmm0, xmm1, 0x4E ; xmm0=(30 31 32 33 34 35 36 37 20 21 22 23 24 25 26 27)
+ pshufd xmm2, xmm4, 0x4E ; xmm2=(50 51 52 53 54 55 56 57 40 41 42 43 44 45 46 47)
+ pshufd xmm5, xmm3, 0x4E ; xmm5=(70 71 72 73 74 75 76 77 60 61 62 63 64 65 66 67)
+
+ mov rdx, JSAMPROW [rdi+0*SIZEOF_JSAMPROW]
+ mov rsi, JSAMPROW [rdi+2*SIZEOF_JSAMPROW]
+ movq XMM_MMWORD [rdx+rax*SIZEOF_JSAMPLE], xmm7
+ movq XMM_MMWORD [rsi+rax*SIZEOF_JSAMPLE], xmm1
+ mov rdx, JSAMPROW [rdi+4*SIZEOF_JSAMPROW]
+ mov rsi, JSAMPROW [rdi+6*SIZEOF_JSAMPROW]
+ movq XMM_MMWORD [rdx+rax*SIZEOF_JSAMPLE], xmm4
+ movq XMM_MMWORD [rsi+rax*SIZEOF_JSAMPLE], xmm3
+
+ mov rdx, JSAMPROW [rdi+1*SIZEOF_JSAMPROW]
+ mov rsi, JSAMPROW [rdi+3*SIZEOF_JSAMPROW]
+ movq XMM_MMWORD [rdx+rax*SIZEOF_JSAMPLE], xmm6
+ movq XMM_MMWORD [rsi+rax*SIZEOF_JSAMPLE], xmm0
+ mov rdx, JSAMPROW [rdi+5*SIZEOF_JSAMPROW]
+ mov rsi, JSAMPROW [rdi+7*SIZEOF_JSAMPROW]
+ movq XMM_MMWORD [rdx+rax*SIZEOF_JSAMPLE], xmm2
+ movq XMM_MMWORD [rsi+rax*SIZEOF_JSAMPLE], xmm5
+
+ uncollect_args 4
+ mov rsp, rbp ; rsp <- aligned rbp
+ pop rsp ; rsp <- original rbp
+ pop rbp
+ ret
+
+; For some reason, the OS X linker does not honor the request to align the
+; segment unless we do this.
+ align 32
diff --git a/simd/x86_64/jidctred-sse2.asm b/simd/x86_64/jidctred-sse2.asm
new file mode 100644
index 0000000..69b71f0
--- /dev/null
+++ b/simd/x86_64/jidctred-sse2.asm
@@ -0,0 +1,575 @@
+;
+; jidctred.asm - reduced-size IDCT (64-bit SSE2)
+;
+; Copyright 2009 Pierre Ossman <ossman@cendio.se> for Cendio AB
+; Copyright (C) 2009, 2016, D. R. Commander.
+;
+; Based on the x86 SIMD extension for IJG JPEG library
+; Copyright (C) 1999-2006, MIYASAKA Masaru.
+; For conditions of distribution and use, see copyright notice in jsimdext.inc
+;
+; This file should be assembled with NASM (Netwide Assembler),
+; can *not* be assembled with Microsoft's MASM or any compatible
+; assembler (including Borland's Turbo Assembler).
+; NASM is available from http://nasm.sourceforge.net/ or
+; http://sourceforge.net/project/showfiles.php?group_id=6208
+;
+; This file contains inverse-DCT routines that produce reduced-size
+; output: either 4x4 or 2x2 pixels from an 8x8 DCT block.
+; The following code is based directly on the IJG's original jidctred.c;
+; see the jidctred.c for more details.
+;
+; [TAB8]
+
+%include "jsimdext.inc"
+%include "jdct.inc"
+
+; --------------------------------------------------------------------------
+
+%define CONST_BITS 13
+%define PASS1_BITS 2
+
+%define DESCALE_P1_4 (CONST_BITS-PASS1_BITS+1)
+%define DESCALE_P2_4 (CONST_BITS+PASS1_BITS+3+1)
+%define DESCALE_P1_2 (CONST_BITS-PASS1_BITS+2)
+%define DESCALE_P2_2 (CONST_BITS+PASS1_BITS+3+2)
+
+%if CONST_BITS == 13
+F_0_211 equ 1730 ; FIX(0.211164243)
+F_0_509 equ 4176 ; FIX(0.509795579)
+F_0_601 equ 4926 ; FIX(0.601344887)
+F_0_720 equ 5906 ; FIX(0.720959822)
+F_0_765 equ 6270 ; FIX(0.765366865)
+F_0_850 equ 6967 ; FIX(0.850430095)
+F_0_899 equ 7373 ; FIX(0.899976223)
+F_1_061 equ 8697 ; FIX(1.061594337)
+F_1_272 equ 10426 ; FIX(1.272758580)
+F_1_451 equ 11893 ; FIX(1.451774981)
+F_1_847 equ 15137 ; FIX(1.847759065)
+F_2_172 equ 17799 ; FIX(2.172734803)
+F_2_562 equ 20995 ; FIX(2.562915447)
+F_3_624 equ 29692 ; FIX(3.624509785)
+%else
+; NASM cannot do compile-time arithmetic on floating-point constants.
+%define DESCALE(x,n) (((x)+(1<<((n)-1)))>>(n))
+F_0_211 equ DESCALE( 226735879, 30-CONST_BITS) ; FIX(0.211164243)
+F_0_509 equ DESCALE( 547388834, 30-CONST_BITS) ; FIX(0.509795579)
+F_0_601 equ DESCALE( 645689155, 30-CONST_BITS) ; FIX(0.601344887)
+F_0_720 equ DESCALE( 774124714, 30-CONST_BITS) ; FIX(0.720959822)
+F_0_765 equ DESCALE( 821806413, 30-CONST_BITS) ; FIX(0.765366865)
+F_0_850 equ DESCALE( 913142361, 30-CONST_BITS) ; FIX(0.850430095)
+F_0_899 equ DESCALE( 966342111, 30-CONST_BITS) ; FIX(0.899976223)
+F_1_061 equ DESCALE(1139878239, 30-CONST_BITS) ; FIX(1.061594337)
+F_1_272 equ DESCALE(1366614119, 30-CONST_BITS) ; FIX(1.272758580)
+F_1_451 equ DESCALE(1558831516, 30-CONST_BITS) ; FIX(1.451774981)
+F_1_847 equ DESCALE(1984016188, 30-CONST_BITS) ; FIX(1.847759065)
+F_2_172 equ DESCALE(2332956230, 30-CONST_BITS) ; FIX(2.172734803)
+F_2_562 equ DESCALE(2751909506, 30-CONST_BITS) ; FIX(2.562915447)
+F_3_624 equ DESCALE(3891787747, 30-CONST_BITS) ; FIX(3.624509785)
+%endif
+
+; --------------------------------------------------------------------------
+ SECTION SEG_CONST
+
+ alignz 32
+ global EXTN(jconst_idct_red_sse2)
+
+EXTN(jconst_idct_red_sse2):
+
+PW_F184_MF076 times 4 dw F_1_847,-F_0_765
+PW_F256_F089 times 4 dw F_2_562, F_0_899
+PW_F106_MF217 times 4 dw F_1_061,-F_2_172
+PW_MF060_MF050 times 4 dw -F_0_601,-F_0_509
+PW_F145_MF021 times 4 dw F_1_451,-F_0_211
+PW_F362_MF127 times 4 dw F_3_624,-F_1_272
+PW_F085_MF072 times 4 dw F_0_850,-F_0_720
+PD_DESCALE_P1_4 times 4 dd 1 << (DESCALE_P1_4-1)
+PD_DESCALE_P2_4 times 4 dd 1 << (DESCALE_P2_4-1)
+PD_DESCALE_P1_2 times 4 dd 1 << (DESCALE_P1_2-1)
+PD_DESCALE_P2_2 times 4 dd 1 << (DESCALE_P2_2-1)
+PB_CENTERJSAMP times 16 db CENTERJSAMPLE
+
+ alignz 32
+
+; --------------------------------------------------------------------------
+ SECTION SEG_TEXT
+ BITS 64
+;
+; Perform dequantization and inverse DCT on one block of coefficients,
+; producing a reduced-size 4x4 output block.
+;
+; GLOBAL(void)
+; jsimd_idct_4x4_sse2 (void *dct_table, JCOEFPTR coef_block,
+; JSAMPARRAY output_buf, JDIMENSION output_col)
+;
+
+; r10 = void *dct_table
+; r11 = JCOEFPTR coef_block
+; r12 = JSAMPARRAY output_buf
+; r13d = JDIMENSION output_col
+
+%define original_rbp rbp+0
+%define wk(i) rbp-(WK_NUM-(i))*SIZEOF_XMMWORD ; xmmword wk[WK_NUM]
+%define WK_NUM 2
+
+ align 32
+ global EXTN(jsimd_idct_4x4_sse2)
+
+EXTN(jsimd_idct_4x4_sse2):
+ push rbp
+ mov rax, rsp ; rax = original rbp
+ sub rsp, byte 4
+ and rsp, byte (-SIZEOF_XMMWORD) ; align to 128 bits
+ mov [rsp], rax
+ mov rbp, rsp ; rbp = aligned rbp
+ lea rsp, [wk(0)]
+ collect_args 4
+
+ ; ---- Pass 1: process columns from input.
+
+ mov rdx, r10 ; quantptr
+ mov rsi, r11 ; inptr
+
+%ifndef NO_ZERO_COLUMN_TEST_4X4_SSE2
+ mov eax, DWORD [DWBLOCK(1,0,rsi,SIZEOF_JCOEF)]
+ or eax, DWORD [DWBLOCK(2,0,rsi,SIZEOF_JCOEF)]
+ jnz short .columnDCT
+
+ movdqa xmm0, XMMWORD [XMMBLOCK(1,0,rsi,SIZEOF_JCOEF)]
+ movdqa xmm1, XMMWORD [XMMBLOCK(2,0,rsi,SIZEOF_JCOEF)]
+ por xmm0, XMMWORD [XMMBLOCK(3,0,rsi,SIZEOF_JCOEF)]
+ por xmm1, XMMWORD [XMMBLOCK(5,0,rsi,SIZEOF_JCOEF)]
+ por xmm0, XMMWORD [XMMBLOCK(6,0,rsi,SIZEOF_JCOEF)]
+ por xmm1, XMMWORD [XMMBLOCK(7,0,rsi,SIZEOF_JCOEF)]
+ por xmm0, xmm1
+ packsswb xmm0, xmm0
+ packsswb xmm0, xmm0
+ movd eax, xmm0
+ test rax, rax
+ jnz short .columnDCT
+
+ ; -- AC terms all zero
+
+ movdqa xmm0, XMMWORD [XMMBLOCK(0,0,rsi,SIZEOF_JCOEF)]
+ pmullw xmm0, XMMWORD [XMMBLOCK(0,0,rdx,SIZEOF_ISLOW_MULT_TYPE)]
+
+ psllw xmm0, PASS1_BITS
+
+ movdqa xmm3, xmm0 ; xmm0=in0=(00 01 02 03 04 05 06 07)
+ punpcklwd xmm0, xmm0 ; xmm0=(00 00 01 01 02 02 03 03)
+ punpckhwd xmm3, xmm3 ; xmm3=(04 04 05 05 06 06 07 07)
+
+ pshufd xmm1, xmm0, 0x50 ; xmm1=[col0 col1]=(00 00 00 00 01 01 01 01)
+ pshufd xmm0, xmm0, 0xFA ; xmm0=[col2 col3]=(02 02 02 02 03 03 03 03)
+ pshufd xmm6, xmm3, 0x50 ; xmm6=[col4 col5]=(04 04 04 04 05 05 05 05)
+ pshufd xmm3, xmm3, 0xFA ; xmm3=[col6 col7]=(06 06 06 06 07 07 07 07)
+
+ jmp near .column_end
+%endif
+.columnDCT:
+
+ ; -- Odd part
+
+ movdqa xmm0, XMMWORD [XMMBLOCK(1,0,rsi,SIZEOF_JCOEF)]
+ movdqa xmm1, XMMWORD [XMMBLOCK(3,0,rsi,SIZEOF_JCOEF)]
+ pmullw xmm0, XMMWORD [XMMBLOCK(1,0,rdx,SIZEOF_ISLOW_MULT_TYPE)]
+ pmullw xmm1, XMMWORD [XMMBLOCK(3,0,rdx,SIZEOF_ISLOW_MULT_TYPE)]
+ movdqa xmm2, XMMWORD [XMMBLOCK(5,0,rsi,SIZEOF_JCOEF)]
+ movdqa xmm3, XMMWORD [XMMBLOCK(7,0,rsi,SIZEOF_JCOEF)]
+ pmullw xmm2, XMMWORD [XMMBLOCK(5,0,rdx,SIZEOF_ISLOW_MULT_TYPE)]
+ pmullw xmm3, XMMWORD [XMMBLOCK(7,0,rdx,SIZEOF_ISLOW_MULT_TYPE)]
+
+ movdqa xmm4, xmm0
+ movdqa xmm5, xmm0
+ punpcklwd xmm4, xmm1
+ punpckhwd xmm5, xmm1
+ movdqa xmm0, xmm4
+ movdqa xmm1, xmm5
+ pmaddwd xmm4, [rel PW_F256_F089] ; xmm4=(tmp2L)
+ pmaddwd xmm5, [rel PW_F256_F089] ; xmm5=(tmp2H)
+ pmaddwd xmm0, [rel PW_F106_MF217] ; xmm0=(tmp0L)
+ pmaddwd xmm1, [rel PW_F106_MF217] ; xmm1=(tmp0H)
+
+ movdqa xmm6, xmm2
+ movdqa xmm7, xmm2
+ punpcklwd xmm6, xmm3
+ punpckhwd xmm7, xmm3
+ movdqa xmm2, xmm6
+ movdqa xmm3, xmm7
+ pmaddwd xmm6, [rel PW_MF060_MF050] ; xmm6=(tmp2L)
+ pmaddwd xmm7, [rel PW_MF060_MF050] ; xmm7=(tmp2H)
+ pmaddwd xmm2, [rel PW_F145_MF021] ; xmm2=(tmp0L)
+ pmaddwd xmm3, [rel PW_F145_MF021] ; xmm3=(tmp0H)
+
+ paddd xmm6, xmm4 ; xmm6=tmp2L
+ paddd xmm7, xmm5 ; xmm7=tmp2H
+ paddd xmm2, xmm0 ; xmm2=tmp0L
+ paddd xmm3, xmm1 ; xmm3=tmp0H
+
+ movdqa XMMWORD [wk(0)], xmm2 ; wk(0)=tmp0L
+ movdqa XMMWORD [wk(1)], xmm3 ; wk(1)=tmp0H
+
+ ; -- Even part
+
+ movdqa xmm4, XMMWORD [XMMBLOCK(0,0,rsi,SIZEOF_JCOEF)]
+ movdqa xmm5, XMMWORD [XMMBLOCK(2,0,rsi,SIZEOF_JCOEF)]
+ movdqa xmm0, XMMWORD [XMMBLOCK(6,0,rsi,SIZEOF_JCOEF)]
+ pmullw xmm4, XMMWORD [XMMBLOCK(0,0,rdx,SIZEOF_ISLOW_MULT_TYPE)]
+ pmullw xmm5, XMMWORD [XMMBLOCK(2,0,rdx,SIZEOF_ISLOW_MULT_TYPE)]
+ pmullw xmm0, XMMWORD [XMMBLOCK(6,0,rdx,SIZEOF_ISLOW_MULT_TYPE)]
+
+ pxor xmm1, xmm1
+ pxor xmm2, xmm2
+ punpcklwd xmm1, xmm4 ; xmm1=tmp0L
+ punpckhwd xmm2, xmm4 ; xmm2=tmp0H
+ psrad xmm1, (16-CONST_BITS-1) ; psrad xmm1,16 & pslld xmm1,CONST_BITS+1
+ psrad xmm2, (16-CONST_BITS-1) ; psrad xmm2,16 & pslld xmm2,CONST_BITS+1
+
+ movdqa xmm3, xmm5 ; xmm5=in2=z2
+ punpcklwd xmm5, xmm0 ; xmm0=in6=z3
+ punpckhwd xmm3, xmm0
+ pmaddwd xmm5, [rel PW_F184_MF076] ; xmm5=tmp2L
+ pmaddwd xmm3, [rel PW_F184_MF076] ; xmm3=tmp2H
+
+ movdqa xmm4, xmm1
+ movdqa xmm0, xmm2
+ paddd xmm1, xmm5 ; xmm1=tmp10L
+ paddd xmm2, xmm3 ; xmm2=tmp10H
+ psubd xmm4, xmm5 ; xmm4=tmp12L
+ psubd xmm0, xmm3 ; xmm0=tmp12H
+
+ ; -- Final output stage
+
+ movdqa xmm5, xmm1
+ movdqa xmm3, xmm2
+ paddd xmm1, xmm6 ; xmm1=data0L
+ paddd xmm2, xmm7 ; xmm2=data0H
+ psubd xmm5, xmm6 ; xmm5=data3L
+ psubd xmm3, xmm7 ; xmm3=data3H
+
+ movdqa xmm6, [rel PD_DESCALE_P1_4] ; xmm6=[rel PD_DESCALE_P1_4]
+
+ paddd xmm1, xmm6
+ paddd xmm2, xmm6
+ psrad xmm1, DESCALE_P1_4
+ psrad xmm2, DESCALE_P1_4
+ paddd xmm5, xmm6
+ paddd xmm3, xmm6
+ psrad xmm5, DESCALE_P1_4
+ psrad xmm3, DESCALE_P1_4
+
+ packssdw xmm1, xmm2 ; xmm1=data0=(00 01 02 03 04 05 06 07)
+ packssdw xmm5, xmm3 ; xmm5=data3=(30 31 32 33 34 35 36 37)
+
+ movdqa xmm7, XMMWORD [wk(0)] ; xmm7=tmp0L
+ movdqa xmm6, XMMWORD [wk(1)] ; xmm6=tmp0H
+
+ movdqa xmm2, xmm4
+ movdqa xmm3, xmm0
+ paddd xmm4, xmm7 ; xmm4=data1L
+ paddd xmm0, xmm6 ; xmm0=data1H
+ psubd xmm2, xmm7 ; xmm2=data2L
+ psubd xmm3, xmm6 ; xmm3=data2H
+
+ movdqa xmm7, [rel PD_DESCALE_P1_4] ; xmm7=[rel PD_DESCALE_P1_4]
+
+ paddd xmm4, xmm7
+ paddd xmm0, xmm7
+ psrad xmm4, DESCALE_P1_4
+ psrad xmm0, DESCALE_P1_4
+ paddd xmm2, xmm7
+ paddd xmm3, xmm7
+ psrad xmm2, DESCALE_P1_4
+ psrad xmm3, DESCALE_P1_4
+
+ packssdw xmm4, xmm0 ; xmm4=data1=(10 11 12 13 14 15 16 17)
+ packssdw xmm2, xmm3 ; xmm2=data2=(20 21 22 23 24 25 26 27)
+
+ movdqa xmm6, xmm1 ; transpose coefficients(phase 1)
+ punpcklwd xmm1, xmm4 ; xmm1=(00 10 01 11 02 12 03 13)
+ punpckhwd xmm6, xmm4 ; xmm6=(04 14 05 15 06 16 07 17)
+ movdqa xmm7, xmm2 ; transpose coefficients(phase 1)
+ punpcklwd xmm2, xmm5 ; xmm2=(20 30 21 31 22 32 23 33)
+ punpckhwd xmm7, xmm5 ; xmm7=(24 34 25 35 26 36 27 37)
+
+ movdqa xmm0, xmm1 ; transpose coefficients(phase 2)
+ punpckldq xmm1, xmm2 ; xmm1=[col0 col1]=(00 10 20 30 01 11 21 31)
+ punpckhdq xmm0, xmm2 ; xmm0=[col2 col3]=(02 12 22 32 03 13 23 33)
+ movdqa xmm3, xmm6 ; transpose coefficients(phase 2)
+ punpckldq xmm6, xmm7 ; xmm6=[col4 col5]=(04 14 24 34 05 15 25 35)
+ punpckhdq xmm3, xmm7 ; xmm3=[col6 col7]=(06 16 26 36 07 17 27 37)
+.column_end:
+
+ ; -- Prefetch the next coefficient block
+
+ prefetchnta [rsi + DCTSIZE2*SIZEOF_JCOEF + 0*32]
+ prefetchnta [rsi + DCTSIZE2*SIZEOF_JCOEF + 1*32]
+ prefetchnta [rsi + DCTSIZE2*SIZEOF_JCOEF + 2*32]
+ prefetchnta [rsi + DCTSIZE2*SIZEOF_JCOEF + 3*32]
+
+ ; ---- Pass 2: process rows, store into output array.
+
+ mov rax, [original_rbp]
+ mov rdi, r12 ; (JSAMPROW *)
+ mov eax, r13d
+
+ ; -- Even part
+
+ pxor xmm4, xmm4
+ punpcklwd xmm4, xmm1 ; xmm4=tmp0
+ psrad xmm4, (16-CONST_BITS-1) ; psrad xmm4,16 & pslld xmm4,CONST_BITS+1
+
+ ; -- Odd part
+
+ punpckhwd xmm1, xmm0
+ punpckhwd xmm6, xmm3
+ movdqa xmm5, xmm1
+ movdqa xmm2, xmm6
+ pmaddwd xmm1, [rel PW_F256_F089] ; xmm1=(tmp2)
+ pmaddwd xmm6, [rel PW_MF060_MF050] ; xmm6=(tmp2)
+ pmaddwd xmm5, [rel PW_F106_MF217] ; xmm5=(tmp0)
+ pmaddwd xmm2, [rel PW_F145_MF021] ; xmm2=(tmp0)
+
+ paddd xmm6, xmm1 ; xmm6=tmp2
+ paddd xmm2, xmm5 ; xmm2=tmp0
+
+ ; -- Even part
+
+ punpcklwd xmm0, xmm3
+ pmaddwd xmm0, [rel PW_F184_MF076] ; xmm0=tmp2
+
+ movdqa xmm7, xmm4
+ paddd xmm4, xmm0 ; xmm4=tmp10
+ psubd xmm7, xmm0 ; xmm7=tmp12
+
+ ; -- Final output stage
+
+ movdqa xmm1, [rel PD_DESCALE_P2_4] ; xmm1=[rel PD_DESCALE_P2_4]
+
+ movdqa xmm5, xmm4
+ movdqa xmm3, xmm7
+ paddd xmm4, xmm6 ; xmm4=data0=(00 10 20 30)
+ paddd xmm7, xmm2 ; xmm7=data1=(01 11 21 31)
+ psubd xmm5, xmm6 ; xmm5=data3=(03 13 23 33)
+ psubd xmm3, xmm2 ; xmm3=data2=(02 12 22 32)
+
+ paddd xmm4, xmm1
+ paddd xmm7, xmm1
+ psrad xmm4, DESCALE_P2_4
+ psrad xmm7, DESCALE_P2_4
+ paddd xmm5, xmm1
+ paddd xmm3, xmm1
+ psrad xmm5, DESCALE_P2_4
+ psrad xmm3, DESCALE_P2_4
+
+ packssdw xmm4, xmm3 ; xmm4=(00 10 20 30 02 12 22 32)
+ packssdw xmm7, xmm5 ; xmm7=(01 11 21 31 03 13 23 33)
+
+ movdqa xmm0, xmm4 ; transpose coefficients(phase 1)
+ punpcklwd xmm4, xmm7 ; xmm4=(00 01 10 11 20 21 30 31)
+ punpckhwd xmm0, xmm7 ; xmm0=(02 03 12 13 22 23 32 33)
+
+ movdqa xmm6, xmm4 ; transpose coefficients(phase 2)
+ punpckldq xmm4, xmm0 ; xmm4=(00 01 02 03 10 11 12 13)
+ punpckhdq xmm6, xmm0 ; xmm6=(20 21 22 23 30 31 32 33)
+
+ packsswb xmm4, xmm6 ; xmm4=(00 01 02 03 10 11 12 13 20 ..)
+ paddb xmm4, [rel PB_CENTERJSAMP]
+
+ pshufd xmm2, xmm4, 0x39 ; xmm2=(10 11 12 13 20 21 22 23 30 ..)
+ pshufd xmm1, xmm4, 0x4E ; xmm1=(20 21 22 23 30 31 32 33 00 ..)
+ pshufd xmm3, xmm4, 0x93 ; xmm3=(30 31 32 33 00 01 02 03 10 ..)
+
+ mov rdx, JSAMPROW [rdi+0*SIZEOF_JSAMPROW]
+ mov rsi, JSAMPROW [rdi+1*SIZEOF_JSAMPROW]
+ movd XMM_DWORD [rdx+rax*SIZEOF_JSAMPLE], xmm4
+ movd XMM_DWORD [rsi+rax*SIZEOF_JSAMPLE], xmm2
+ mov rdx, JSAMPROW [rdi+2*SIZEOF_JSAMPROW]
+ mov rsi, JSAMPROW [rdi+3*SIZEOF_JSAMPROW]
+ movd XMM_DWORD [rdx+rax*SIZEOF_JSAMPLE], xmm1
+ movd XMM_DWORD [rsi+rax*SIZEOF_JSAMPLE], xmm3
+
+ uncollect_args 4
+ mov rsp, rbp ; rsp <- aligned rbp
+ pop rsp ; rsp <- original rbp
+ pop rbp
+ ret
+
+
+; --------------------------------------------------------------------------
+;
+; Perform dequantization and inverse DCT on one block of coefficients,
+; producing a reduced-size 2x2 output block.
+;
+; GLOBAL(void)
+; jsimd_idct_2x2_sse2 (void *dct_table, JCOEFPTR coef_block,
+; JSAMPARRAY output_buf, JDIMENSION output_col)
+;
+
+; r10 = void *dct_table
+; r11 = JCOEFPTR coef_block
+; r12 = JSAMPARRAY output_buf
+; r13d = JDIMENSION output_col
+
+ align 32
+ global EXTN(jsimd_idct_2x2_sse2)
+
+EXTN(jsimd_idct_2x2_sse2):
+ push rbp
+ mov rax, rsp
+ mov rbp, rsp
+ collect_args 4
+ push rbx
+
+ ; ---- Pass 1: process columns from input.
+
+ mov rdx, r10 ; quantptr
+ mov rsi, r11 ; inptr
+
+ ; | input: | result: |
+ ; | 00 01 ** 03 ** 05 ** 07 | |
+ ; | 10 11 ** 13 ** 15 ** 17 | |
+ ; | ** ** ** ** ** ** ** ** | |
+ ; | 30 31 ** 33 ** 35 ** 37 | A0 A1 A3 A5 A7 |
+ ; | ** ** ** ** ** ** ** ** | B0 B1 B3 B5 B7 |
+ ; | 50 51 ** 53 ** 55 ** 57 | |
+ ; | ** ** ** ** ** ** ** ** | |
+ ; | 70 71 ** 73 ** 75 ** 77 | |
+
+ ; -- Odd part
+
+ movdqa xmm0, XMMWORD [XMMBLOCK(1,0,rsi,SIZEOF_JCOEF)]
+ movdqa xmm1, XMMWORD [XMMBLOCK(3,0,rsi,SIZEOF_JCOEF)]
+ pmullw xmm0, XMMWORD [XMMBLOCK(1,0,rdx,SIZEOF_ISLOW_MULT_TYPE)]
+ pmullw xmm1, XMMWORD [XMMBLOCK(3,0,rdx,SIZEOF_ISLOW_MULT_TYPE)]
+ movdqa xmm2, XMMWORD [XMMBLOCK(5,0,rsi,SIZEOF_JCOEF)]
+ movdqa xmm3, XMMWORD [XMMBLOCK(7,0,rsi,SIZEOF_JCOEF)]
+ pmullw xmm2, XMMWORD [XMMBLOCK(5,0,rdx,SIZEOF_ISLOW_MULT_TYPE)]
+ pmullw xmm3, XMMWORD [XMMBLOCK(7,0,rdx,SIZEOF_ISLOW_MULT_TYPE)]
+
+ ; xmm0=(10 11 ** 13 ** 15 ** 17), xmm1=(30 31 ** 33 ** 35 ** 37)
+ ; xmm2=(50 51 ** 53 ** 55 ** 57), xmm3=(70 71 ** 73 ** 75 ** 77)
+
+ pcmpeqd xmm7, xmm7
+ pslld xmm7, WORD_BIT ; xmm7={0x0000 0xFFFF 0x0000 0xFFFF ..}
+
+ movdqa xmm4, xmm0 ; xmm4=(10 11 ** 13 ** 15 ** 17)
+ movdqa xmm5, xmm2 ; xmm5=(50 51 ** 53 ** 55 ** 57)
+ punpcklwd xmm4, xmm1 ; xmm4=(10 30 11 31 ** ** 13 33)
+ punpcklwd xmm5, xmm3 ; xmm5=(50 70 51 71 ** ** 53 73)
+ pmaddwd xmm4, [rel PW_F362_MF127]
+ pmaddwd xmm5, [rel PW_F085_MF072]
+
+ psrld xmm0, WORD_BIT ; xmm0=(11 -- 13 -- 15 -- 17 --)
+ pand xmm1, xmm7 ; xmm1=(-- 31 -- 33 -- 35 -- 37)
+ psrld xmm2, WORD_BIT ; xmm2=(51 -- 53 -- 55 -- 57 --)
+ pand xmm3, xmm7 ; xmm3=(-- 71 -- 73 -- 75 -- 77)
+ por xmm0, xmm1 ; xmm0=(11 31 13 33 15 35 17 37)
+ por xmm2, xmm3 ; xmm2=(51 71 53 73 55 75 57 77)
+ pmaddwd xmm0, [rel PW_F362_MF127]
+ pmaddwd xmm2, [rel PW_F085_MF072]
+
+ paddd xmm4, xmm5 ; xmm4=tmp0[col0 col1 **** col3]
+ paddd xmm0, xmm2 ; xmm0=tmp0[col1 col3 col5 col7]
+
+ ; -- Even part
+
+ movdqa xmm6, XMMWORD [XMMBLOCK(0,0,rsi,SIZEOF_JCOEF)]
+ pmullw xmm6, XMMWORD [XMMBLOCK(0,0,rdx,SIZEOF_ISLOW_MULT_TYPE)]
+
+ ; xmm6=(00 01 ** 03 ** 05 ** 07)
+
+ movdqa xmm1, xmm6 ; xmm1=(00 01 ** 03 ** 05 ** 07)
+ pslld xmm6, WORD_BIT ; xmm6=(-- 00 -- ** -- ** -- **)
+ pand xmm1, xmm7 ; xmm1=(-- 01 -- 03 -- 05 -- 07)
+ psrad xmm6, (WORD_BIT-CONST_BITS-2) ; xmm6=tmp10[col0 **** **** ****]
+ psrad xmm1, (WORD_BIT-CONST_BITS-2) ; xmm1=tmp10[col1 col3 col5 col7]
+
+ ; -- Final output stage
+
+ movdqa xmm3, xmm6
+ movdqa xmm5, xmm1
+ paddd xmm6, xmm4 ; xmm6=data0[col0 **** **** ****]=(A0 ** ** **)
+ paddd xmm1, xmm0 ; xmm1=data0[col1 col3 col5 col7]=(A1 A3 A5 A7)
+ psubd xmm3, xmm4 ; xmm3=data1[col0 **** **** ****]=(B0 ** ** **)
+ psubd xmm5, xmm0 ; xmm5=data1[col1 col3 col5 col7]=(B1 B3 B5 B7)
+
+ movdqa xmm2, [rel PD_DESCALE_P1_2] ; xmm2=[rel PD_DESCALE_P1_2]
+
+ punpckldq xmm6, xmm3 ; xmm6=(A0 B0 ** **)
+
+ movdqa xmm7, xmm1
+ punpcklqdq xmm1, xmm5 ; xmm1=(A1 A3 B1 B3)
+ punpckhqdq xmm7, xmm5 ; xmm7=(A5 A7 B5 B7)
+
+ paddd xmm6, xmm2
+ psrad xmm6, DESCALE_P1_2
+
+ paddd xmm1, xmm2
+ paddd xmm7, xmm2
+ psrad xmm1, DESCALE_P1_2
+ psrad xmm7, DESCALE_P1_2
+
+ ; -- Prefetch the next coefficient block
+
+ prefetchnta [rsi + DCTSIZE2*SIZEOF_JCOEF + 0*32]
+ prefetchnta [rsi + DCTSIZE2*SIZEOF_JCOEF + 1*32]
+ prefetchnta [rsi + DCTSIZE2*SIZEOF_JCOEF + 2*32]
+ prefetchnta [rsi + DCTSIZE2*SIZEOF_JCOEF + 3*32]
+
+ ; ---- Pass 2: process rows, store into output array.
+
+ mov rdi, r12 ; (JSAMPROW *)
+ mov eax, r13d
+
+ ; | input:| result:|
+ ; | A0 B0 | |
+ ; | A1 B1 | C0 C1 |
+ ; | A3 B3 | D0 D1 |
+ ; | A5 B5 | |
+ ; | A7 B7 | |
+
+ ; -- Odd part
+
+ packssdw xmm1, xmm1 ; xmm1=(A1 A3 B1 B3 A1 A3 B1 B3)
+ packssdw xmm7, xmm7 ; xmm7=(A5 A7 B5 B7 A5 A7 B5 B7)
+ pmaddwd xmm1, [rel PW_F362_MF127]
+ pmaddwd xmm7, [rel PW_F085_MF072]
+
+ paddd xmm1, xmm7 ; xmm1=tmp0[row0 row1 row0 row1]
+
+ ; -- Even part
+
+ pslld xmm6, (CONST_BITS+2) ; xmm6=tmp10[row0 row1 **** ****]
+
+ ; -- Final output stage
+
+ movdqa xmm4, xmm6
+ paddd xmm6, xmm1 ; xmm6=data0[row0 row1 **** ****]=(C0 C1 ** **)
+ psubd xmm4, xmm1 ; xmm4=data1[row0 row1 **** ****]=(D0 D1 ** **)
+
+ punpckldq xmm6, xmm4 ; xmm6=(C0 D0 C1 D1)
+
+ paddd xmm6, [rel PD_DESCALE_P2_2]
+ psrad xmm6, DESCALE_P2_2
+
+ packssdw xmm6, xmm6 ; xmm6=(C0 D0 C1 D1 C0 D0 C1 D1)
+ packsswb xmm6, xmm6 ; xmm6=(C0 D0 C1 D1 C0 D0 C1 D1 ..)
+ paddb xmm6, [rel PB_CENTERJSAMP]
+
+ pextrw ebx, xmm6, 0x00 ; ebx=(C0 D0 -- --)
+ pextrw ecx, xmm6, 0x01 ; ecx=(C1 D1 -- --)
+
+ mov rdx, JSAMPROW [rdi+0*SIZEOF_JSAMPROW]
+ mov rsi, JSAMPROW [rdi+1*SIZEOF_JSAMPROW]
+ mov WORD [rdx+rax*SIZEOF_JSAMPLE], bx
+ mov WORD [rsi+rax*SIZEOF_JSAMPLE], cx
+
+ pop rbx
+ uncollect_args 4
+ pop rbp
+ ret
+
+; For some reason, the OS X linker does not honor the request to align the
+; segment unless we do this.
+ align 32
diff --git a/simd/x86_64/jquantf-sse2.asm b/simd/x86_64/jquantf-sse2.asm
new file mode 100644
index 0000000..3699d44
--- /dev/null
+++ b/simd/x86_64/jquantf-sse2.asm
@@ -0,0 +1,157 @@
+;
+; jquantf.asm - sample data conversion and quantization (64-bit SSE & SSE2)
+;
+; Copyright 2009 Pierre Ossman <ossman@cendio.se> for Cendio AB
+; Copyright (C) 2009, 2016, D. R. Commander.
+;
+; Based on the x86 SIMD extension for IJG JPEG library
+; Copyright (C) 1999-2006, MIYASAKA Masaru.
+; For conditions of distribution and use, see copyright notice in jsimdext.inc
+;
+; This file should be assembled with NASM (Netwide Assembler),
+; can *not* be assembled with Microsoft's MASM or any compatible
+; assembler (including Borland's Turbo Assembler).
+; NASM is available from http://nasm.sourceforge.net/ or
+; http://sourceforge.net/project/showfiles.php?group_id=6208
+;
+; [TAB8]
+
+%include "jsimdext.inc"
+%include "jdct.inc"
+
+; --------------------------------------------------------------------------
+ SECTION SEG_TEXT
+ BITS 64
+;
+; Load data into workspace, applying unsigned->signed conversion
+;
+; GLOBAL(void)
+; jsimd_convsamp_float_sse2 (JSAMPARRAY sample_data, JDIMENSION start_col,
+; FAST_FLOAT *workspace);
+;
+
+; r10 = JSAMPARRAY sample_data
+; r11d = JDIMENSION start_col
+; r12 = FAST_FLOAT *workspace
+
+ align 32
+ global EXTN(jsimd_convsamp_float_sse2)
+
+EXTN(jsimd_convsamp_float_sse2):
+ push rbp
+ mov rax, rsp
+ mov rbp, rsp
+ collect_args 3
+ push rbx
+
+ pcmpeqw xmm7, xmm7
+ psllw xmm7, 7
+ packsswb xmm7, xmm7 ; xmm7 = PB_CENTERJSAMPLE (0x808080..)
+
+ mov rsi, r10
+ mov eax, r11d
+ mov rdi, r12
+ mov rcx, DCTSIZE/2
+.convloop:
+ mov rbx, JSAMPROW [rsi+0*SIZEOF_JSAMPROW] ; (JSAMPLE *)
+ mov rdx, JSAMPROW [rsi+1*SIZEOF_JSAMPROW] ; (JSAMPLE *)
+
+ movq xmm0, XMM_MMWORD [rbx+rax*SIZEOF_JSAMPLE]
+ movq xmm1, XMM_MMWORD [rdx+rax*SIZEOF_JSAMPLE]
+
+ psubb xmm0, xmm7 ; xmm0=(01234567)
+ psubb xmm1, xmm7 ; xmm1=(89ABCDEF)
+
+ punpcklbw xmm0, xmm0 ; xmm0=(*0*1*2*3*4*5*6*7)
+ punpcklbw xmm1, xmm1 ; xmm1=(*8*9*A*B*C*D*E*F)
+
+ punpcklwd xmm2, xmm0 ; xmm2=(***0***1***2***3)
+ punpckhwd xmm0, xmm0 ; xmm0=(***4***5***6***7)
+ punpcklwd xmm3, xmm1 ; xmm3=(***8***9***A***B)
+ punpckhwd xmm1, xmm1 ; xmm1=(***C***D***E***F)
+
+ psrad xmm2, (DWORD_BIT-BYTE_BIT) ; xmm2=(0123)
+ psrad xmm0, (DWORD_BIT-BYTE_BIT) ; xmm0=(4567)
+ cvtdq2ps xmm2, xmm2 ; xmm2=(0123)
+ cvtdq2ps xmm0, xmm0 ; xmm0=(4567)
+ psrad xmm3, (DWORD_BIT-BYTE_BIT) ; xmm3=(89AB)
+ psrad xmm1, (DWORD_BIT-BYTE_BIT) ; xmm1=(CDEF)
+ cvtdq2ps xmm3, xmm3 ; xmm3=(89AB)
+ cvtdq2ps xmm1, xmm1 ; xmm1=(CDEF)
+
+ movaps XMMWORD [XMMBLOCK(0,0,rdi,SIZEOF_FAST_FLOAT)], xmm2
+ movaps XMMWORD [XMMBLOCK(0,1,rdi,SIZEOF_FAST_FLOAT)], xmm0
+ movaps XMMWORD [XMMBLOCK(1,0,rdi,SIZEOF_FAST_FLOAT)], xmm3
+ movaps XMMWORD [XMMBLOCK(1,1,rdi,SIZEOF_FAST_FLOAT)], xmm1
+
+ add rsi, byte 2*SIZEOF_JSAMPROW
+ add rdi, byte 2*DCTSIZE*SIZEOF_FAST_FLOAT
+ dec rcx
+ jnz short .convloop
+
+ pop rbx
+ uncollect_args 3
+ pop rbp
+ ret
+
+
+; --------------------------------------------------------------------------
+;
+; Quantize/descale the coefficients, and store into coef_block
+;
+; GLOBAL(void)
+; jsimd_quantize_float_sse2 (JCOEFPTR coef_block, FAST_FLOAT *divisors,
+; FAST_FLOAT *workspace);
+;
+
+; r10 = JCOEFPTR coef_block
+; r11 = FAST_FLOAT *divisors
+; r12 = FAST_FLOAT *workspace
+
+ align 32
+ global EXTN(jsimd_quantize_float_sse2)
+
+EXTN(jsimd_quantize_float_sse2):
+ push rbp
+ mov rax, rsp
+ mov rbp, rsp
+ collect_args 3
+
+ mov rsi, r12
+ mov rdx, r11
+ mov rdi, r10
+ mov rax, DCTSIZE2/16
+.quantloop:
+ movaps xmm0, XMMWORD [XMMBLOCK(0,0,rsi,SIZEOF_FAST_FLOAT)]
+ movaps xmm1, XMMWORD [XMMBLOCK(0,1,rsi,SIZEOF_FAST_FLOAT)]
+ mulps xmm0, XMMWORD [XMMBLOCK(0,0,rdx,SIZEOF_FAST_FLOAT)]
+ mulps xmm1, XMMWORD [XMMBLOCK(0,1,rdx,SIZEOF_FAST_FLOAT)]
+ movaps xmm2, XMMWORD [XMMBLOCK(1,0,rsi,SIZEOF_FAST_FLOAT)]
+ movaps xmm3, XMMWORD [XMMBLOCK(1,1,rsi,SIZEOF_FAST_FLOAT)]
+ mulps xmm2, XMMWORD [XMMBLOCK(1,0,rdx,SIZEOF_FAST_FLOAT)]
+ mulps xmm3, XMMWORD [XMMBLOCK(1,1,rdx,SIZEOF_FAST_FLOAT)]
+
+ cvtps2dq xmm0, xmm0
+ cvtps2dq xmm1, xmm1
+ cvtps2dq xmm2, xmm2
+ cvtps2dq xmm3, xmm3
+
+ packssdw xmm0, xmm1
+ packssdw xmm2, xmm3
+
+ movdqa XMMWORD [XMMBLOCK(0,0,rdi,SIZEOF_JCOEF)], xmm0
+ movdqa XMMWORD [XMMBLOCK(1,0,rdi,SIZEOF_JCOEF)], xmm2
+
+ add rsi, byte 16*SIZEOF_FAST_FLOAT
+ add rdx, byte 16*SIZEOF_FAST_FLOAT
+ add rdi, byte 16*SIZEOF_JCOEF
+ dec rax
+ jnz short .quantloop
+
+ uncollect_args 3
+ pop rbp
+ ret
+
+; For some reason, the OS X linker does not honor the request to align the
+; segment unless we do this.
+ align 32
diff --git a/simd/x86_64/jquanti-avx2.asm b/simd/x86_64/jquanti-avx2.asm
new file mode 100644
index 0000000..60ce19f
--- /dev/null
+++ b/simd/x86_64/jquanti-avx2.asm
@@ -0,0 +1,96 @@
+;
+; jquanti.asm - sample data conversion and quantization (64-bit AVX2)
+;
+; Copyright 2009 Pierre Ossman <ossman@cendio.se> for Cendio AB
+; Copyright (C) 2009, 2016, D. R. Commander.
+; Copyright (C) 2016, Matthieu Darbois.
+;
+; Based on the x86 SIMD extension for IJG JPEG library
+; Copyright (C) 1999-2006, MIYASAKA Masaru.
+; For conditions of distribution and use, see copyright notice in jsimdext.inc
+;
+; This file should be assembled with NASM (Netwide Assembler),
+; can *not* be assembled with Microsoft's MASM or any compatible
+; assembler (including Borland's Turbo Assembler).
+; NASM is available from http://nasm.sourceforge.net/ or
+; http://sourceforge.net/project/showfiles.php?group_id=6208
+;
+; [TAB8]
+
+%include "jsimdext.inc"
+%include "jdct.inc"
+
+; --------------------------------------------------------------------------
+ SECTION SEG_TEXT
+ BITS 64
+
+; --------------------------------------------------------------------------
+;
+; Quantize/descale the coefficients, and store into coef_block
+;
+; This implementation is based on an algorithm described in
+; "How to optimize for the Pentium family of microprocessors"
+; (http://www.agner.org/assem/).
+;
+; GLOBAL(void)
+; jsimd_quantize_avx2 (JCOEFPTR coef_block, DCTELEM *divisors,
+; DCTELEM *workspace);
+;
+
+%define RECIPROCAL(m,n,b) XMMBLOCK(DCTSIZE*0+(m),(n),(b),SIZEOF_DCTELEM)
+%define CORRECTION(m,n,b) XMMBLOCK(DCTSIZE*1+(m),(n),(b),SIZEOF_DCTELEM)
+%define SCALE(m,n,b) XMMBLOCK(DCTSIZE*2+(m),(n),(b),SIZEOF_DCTELEM)
+
+; r10 = JCOEFPTR coef_block
+; r11 = DCTELEM *divisors
+; r12 = DCTELEM *workspace
+
+ align 32
+ global EXTN(jsimd_quantize_avx2)
+
+EXTN(jsimd_quantize_avx2):
+ push rbp
+ mov rax, rsp
+ mov rbp, rsp
+ collect_args 3
+
+ vmovdqu ymm4, [XMMBLOCK(0,0,r12,SIZEOF_DCTELEM)]
+ vmovdqu ymm5, [XMMBLOCK(2,0,r12,SIZEOF_DCTELEM)]
+ vmovdqu ymm6, [XMMBLOCK(4,0,r12,SIZEOF_DCTELEM)]
+ vmovdqu ymm7, [XMMBLOCK(6,0,r12,SIZEOF_DCTELEM)]
+ vpabsw ymm0, ymm4
+ vpabsw ymm1, ymm5
+ vpabsw ymm2, ymm6
+ vpabsw ymm3, ymm7
+
+ vpaddw ymm0, YMMWORD [CORRECTION(0,0,r11)] ; correction + roundfactor
+ vpaddw ymm1, YMMWORD [CORRECTION(2,0,r11)]
+ vpaddw ymm2, YMMWORD [CORRECTION(4,0,r11)]
+ vpaddw ymm3, YMMWORD [CORRECTION(6,0,r11)]
+ vpmulhuw ymm0, YMMWORD [RECIPROCAL(0,0,r11)] ; reciprocal
+ vpmulhuw ymm1, YMMWORD [RECIPROCAL(2,0,r11)]
+ vpmulhuw ymm2, YMMWORD [RECIPROCAL(4,0,r11)]
+ vpmulhuw ymm3, YMMWORD [RECIPROCAL(6,0,r11)]
+ vpmulhuw ymm0, YMMWORD [SCALE(0,0,r11)] ; scale
+ vpmulhuw ymm1, YMMWORD [SCALE(2,0,r11)]
+ vpmulhuw ymm2, YMMWORD [SCALE(4,0,r11)]
+ vpmulhuw ymm3, YMMWORD [SCALE(6,0,r11)]
+
+ vpsignw ymm0, ymm0, ymm4
+ vpsignw ymm1, ymm1, ymm5
+ vpsignw ymm2, ymm2, ymm6
+ vpsignw ymm3, ymm3, ymm7
+
+ vmovdqu [XMMBLOCK(0,0,r10,SIZEOF_DCTELEM)], ymm0
+ vmovdqu [XMMBLOCK(2,0,r10,SIZEOF_DCTELEM)], ymm1
+ vmovdqu [XMMBLOCK(4,0,r10,SIZEOF_DCTELEM)], ymm2
+ vmovdqu [XMMBLOCK(6,0,r10,SIZEOF_DCTELEM)], ymm3
+
+ vzeroupper
+ uncollect_args 3
+ pop rbp
+ ret
+
+; For some reason, the OS X linker does not honor the request to align the
+; segment unless we do this.
+ align 32
diff --git a/simd/x86_64/jquanti-sse2.asm b/simd/x86_64/jquanti-sse2.asm
new file mode 100644
index 0000000..4b0873e
--- /dev/null
+++ b/simd/x86_64/jquanti-sse2.asm
@@ -0,0 +1,186 @@
+;
+; jquanti.asm - sample data conversion and quantization (64-bit SSE2)
+;
+; Copyright 2009 Pierre Ossman <ossman@cendio.se> for Cendio AB
+; Copyright (C) 2009, 2016, D. R. Commander.
+;
+; Based on the x86 SIMD extension for IJG JPEG library
+; Copyright (C) 1999-2006, MIYASAKA Masaru.
+; For conditions of distribution and use, see copyright notice in jsimdext.inc
+;
+; This file should be assembled with NASM (Netwide Assembler),
+; can *not* be assembled with Microsoft's MASM or any compatible
+; assembler (including Borland's Turbo Assembler).
+; NASM is available from http://nasm.sourceforge.net/ or
+; http://sourceforge.net/project/showfiles.php?group_id=6208
+;
+; [TAB8]
+
+%include "jsimdext.inc"
+%include "jdct.inc"
+
+; --------------------------------------------------------------------------
+ SECTION SEG_TEXT
+ BITS 64
+;
+; Load data into workspace, applying unsigned->signed conversion
+;
+; GLOBAL(void)
+; jsimd_convsamp_sse2 (JSAMPARRAY sample_data, JDIMENSION start_col,
+; DCTELEM *workspace);
+;
+
+; r10 = JSAMPARRAY sample_data
+; r11d = JDIMENSION start_col
+; r12 = DCTELEM *workspace
+
+ align 32
+ global EXTN(jsimd_convsamp_sse2)
+
+EXTN(jsimd_convsamp_sse2):
+ push rbp
+ mov rax, rsp
+ mov rbp, rsp
+ collect_args 3
+ push rbx
+
+ pxor xmm6, xmm6 ; xmm6=(all 0's)
+ pcmpeqw xmm7, xmm7
+ psllw xmm7, 7 ; xmm7={0xFF80 0xFF80 0xFF80 0xFF80 ..}
+
+ mov rsi, r10
+ mov eax, r11d
+ mov rdi, r12
+ mov rcx, DCTSIZE/4
+.convloop:
+ mov rbx, JSAMPROW [rsi+0*SIZEOF_JSAMPROW] ; (JSAMPLE *)
+ mov rdx, JSAMPROW [rsi+1*SIZEOF_JSAMPROW] ; (JSAMPLE *)
+
+ movq xmm0, XMM_MMWORD [rbx+rax*SIZEOF_JSAMPLE] ; xmm0=(01234567)
+ movq xmm1, XMM_MMWORD [rdx+rax*SIZEOF_JSAMPLE] ; xmm1=(89ABCDEF)
+
+ mov rbx, JSAMPROW [rsi+2*SIZEOF_JSAMPROW] ; (JSAMPLE *)
+ mov rdx, JSAMPROW [rsi+3*SIZEOF_JSAMPROW] ; (JSAMPLE *)
+
+ movq xmm2, XMM_MMWORD [rbx+rax*SIZEOF_JSAMPLE] ; xmm2=(GHIJKLMN)
+ movq xmm3, XMM_MMWORD [rdx+rax*SIZEOF_JSAMPLE] ; xmm3=(OPQRSTUV)
+
+ punpcklbw xmm0, xmm6 ; xmm0=(01234567)
+ punpcklbw xmm1, xmm6 ; xmm1=(89ABCDEF)
+ paddw xmm0, xmm7
+ paddw xmm1, xmm7
+ punpcklbw xmm2, xmm6 ; xmm2=(GHIJKLMN)
+ punpcklbw xmm3, xmm6 ; xmm3=(OPQRSTUV)
+ paddw xmm2, xmm7
+ paddw xmm3, xmm7
+
+ movdqa XMMWORD [XMMBLOCK(0,0,rdi,SIZEOF_DCTELEM)], xmm0
+ movdqa XMMWORD [XMMBLOCK(1,0,rdi,SIZEOF_DCTELEM)], xmm1
+ movdqa XMMWORD [XMMBLOCK(2,0,rdi,SIZEOF_DCTELEM)], xmm2
+ movdqa XMMWORD [XMMBLOCK(3,0,rdi,SIZEOF_DCTELEM)], xmm3
+
+ add rsi, byte 4*SIZEOF_JSAMPROW
+ add rdi, byte 4*DCTSIZE*SIZEOF_DCTELEM
+ dec rcx
+ jnz short .convloop
+
+ pop rbx
+ uncollect_args 3
+ pop rbp
+ ret
+
+; --------------------------------------------------------------------------
+;
+; Quantize/descale the coefficients, and store into coef_block
+;
+; This implementation is based on an algorithm described in
+; "How to optimize for the Pentium family of microprocessors"
+; (http://www.agner.org/assem/).
+;
+; GLOBAL(void)
+; jsimd_quantize_sse2 (JCOEFPTR coef_block, DCTELEM *divisors,
+; DCTELEM *workspace);
+;
+
+%define RECIPROCAL(m,n,b) XMMBLOCK(DCTSIZE*0+(m),(n),(b),SIZEOF_DCTELEM)
+%define CORRECTION(m,n,b) XMMBLOCK(DCTSIZE*1+(m),(n),(b),SIZEOF_DCTELEM)
+%define SCALE(m,n,b) XMMBLOCK(DCTSIZE*2+(m),(n),(b),SIZEOF_DCTELEM)
+
+; r10 = JCOEFPTR coef_block
+; r11 = DCTELEM *divisors
+; r12 = DCTELEM *workspace
+
+ align 32
+ global EXTN(jsimd_quantize_sse2)
+
+EXTN(jsimd_quantize_sse2):
+ push rbp
+ mov rax, rsp
+ mov rbp, rsp
+ collect_args 3
+
+ mov rsi, r12
+ mov rdx, r11
+ mov rdi, r10
+ mov rax, DCTSIZE2/32
+.quantloop:
+ movdqa xmm4, XMMWORD [XMMBLOCK(0,0,rsi,SIZEOF_DCTELEM)]
+ movdqa xmm5, XMMWORD [XMMBLOCK(1,0,rsi,SIZEOF_DCTELEM)]
+ movdqa xmm6, XMMWORD [XMMBLOCK(2,0,rsi,SIZEOF_DCTELEM)]
+ movdqa xmm7, XMMWORD [XMMBLOCK(3,0,rsi,SIZEOF_DCTELEM)]
+ movdqa xmm0, xmm4
+ movdqa xmm1, xmm5
+ movdqa xmm2, xmm6
+ movdqa xmm3, xmm7
+ psraw xmm4, (WORD_BIT-1)
+ psraw xmm5, (WORD_BIT-1)
+ psraw xmm6, (WORD_BIT-1)
+ psraw xmm7, (WORD_BIT-1)
+ pxor xmm0, xmm4
+ pxor xmm1, xmm5
+ pxor xmm2, xmm6
+ pxor xmm3, xmm7
+ psubw xmm0, xmm4 ; if (xmm0 < 0) xmm0 = -xmm0;
+ psubw xmm1, xmm5 ; if (xmm1 < 0) xmm1 = -xmm1;
+ psubw xmm2, xmm6 ; if (xmm2 < 0) xmm2 = -xmm2;
+ psubw xmm3, xmm7 ; if (xmm3 < 0) xmm3 = -xmm3;
+
+ paddw xmm0, XMMWORD [CORRECTION(0,0,rdx)] ; correction + roundfactor
+ paddw xmm1, XMMWORD [CORRECTION(1,0,rdx)]
+ paddw xmm2, XMMWORD [CORRECTION(2,0,rdx)]
+ paddw xmm3, XMMWORD [CORRECTION(3,0,rdx)]
+ pmulhuw xmm0, XMMWORD [RECIPROCAL(0,0,rdx)] ; reciprocal
+ pmulhuw xmm1, XMMWORD [RECIPROCAL(1,0,rdx)]
+ pmulhuw xmm2, XMMWORD [RECIPROCAL(2,0,rdx)]
+ pmulhuw xmm3, XMMWORD [RECIPROCAL(3,0,rdx)]
+ pmulhuw xmm0, XMMWORD [SCALE(0,0,rdx)] ; scale
+ pmulhuw xmm1, XMMWORD [SCALE(1,0,rdx)]
+ pmulhuw xmm2, XMMWORD [SCALE(2,0,rdx)]
+ pmulhuw xmm3, XMMWORD [SCALE(3,0,rdx)]
+
+ pxor xmm0, xmm4
+ pxor xmm1, xmm5
+ pxor xmm2, xmm6
+ pxor xmm3, xmm7
+ psubw xmm0, xmm4
+ psubw xmm1, xmm5
+ psubw xmm2, xmm6
+ psubw xmm3, xmm7
+ movdqa XMMWORD [XMMBLOCK(0,0,rdi,SIZEOF_DCTELEM)], xmm0
+ movdqa XMMWORD [XMMBLOCK(1,0,rdi,SIZEOF_DCTELEM)], xmm1
+ movdqa XMMWORD [XMMBLOCK(2,0,rdi,SIZEOF_DCTELEM)], xmm2
+ movdqa XMMWORD [XMMBLOCK(3,0,rdi,SIZEOF_DCTELEM)], xmm3
+
+ add rsi, byte 32*SIZEOF_DCTELEM
+ add rdx, byte 32*SIZEOF_DCTELEM
+ add rdi, byte 32*SIZEOF_JCOEF
+ dec rax
+ jnz near .quantloop
+
+ uncollect_args 3
+ pop rbp
+ ret
+
+; For some reason, the OS X linker does not honor the request to align the
+; segment unless we do this.
+ align 32
diff --git a/simd/jsimd_x86_64.c b/simd/x86_64/jsimd.c
similarity index 71%
rename from simd/jsimd_x86_64.c
rename to simd/x86_64/jsimd.c
index a62bcdb..e40f0cf 100644
--- a/simd/jsimd_x86_64.c
+++ b/simd/x86_64/jsimd.c
@@ -15,12 +15,12 @@
*/
#define JPEG_INTERNALS
-#include "../jinclude.h"
-#include "../jpeglib.h"
+#include "../../jinclude.h"
+#include "../../jpeglib.h"
+#include "../../jsimd.h"
+#include "../../jdct.h"
+#include "../../jsimddct.h"
#include "../jsimd.h"
-#include "../jdct.h"
-#include "../jsimddct.h"
-#include "jsimd.h"
/*
* In the PIC cases, we have no guarantee that constants will keep
@@ -29,6 +29,7 @@
#define IS_ALIGNED(ptr, order) (((size_t)ptr & ((1 << order) - 1)) == 0)
#define IS_ALIGNED_SSE(ptr) (IS_ALIGNED(ptr, 4)) /* 16 byte alignment */
+#define IS_ALIGNED_AVX(ptr) (IS_ALIGNED(ptr, 5)) /* 32 byte alignment */
static unsigned int simd_support = ~0;
static unsigned int simd_huffman = 1;
@@ -46,9 +47,15 @@
if (simd_support != ~0U)
return;
- simd_support = JSIMD_SSE2 | JSIMD_SSE;
+ simd_support = jpeg_simd_cpu_support();
/* Force different settings through environment variables */
+ env = getenv("JSIMD_FORCESSE2");
+ if ((env != NULL) && (strcmp(env, "1") == 0))
+ simd_support &= JSIMD_SSE2;
+ env = getenv("JSIMD_FORCEAVX2");
+ if ((env != NULL) && (strcmp(env, "1") == 0))
+ simd_support &= JSIMD_AVX2;
env = getenv("JSIMD_FORCENONE");
if ((env != NULL) && (strcmp(env, "1") == 0))
simd_support = 0;
@@ -70,6 +77,9 @@
if ((RGB_PIXELSIZE != 3) && (RGB_PIXELSIZE != 4))
return 0;
+ if ((simd_support & JSIMD_AVX2) &&
+ IS_ALIGNED_AVX(jconst_rgb_ycc_convert_avx2))
+ return 1;
if ((simd_support & JSIMD_SSE2) &&
IS_ALIGNED_SSE(jconst_rgb_ycc_convert_sse2))
return 1;
@@ -90,6 +100,9 @@
if ((RGB_PIXELSIZE != 3) && (RGB_PIXELSIZE != 4))
return 0;
+ if ((simd_support & JSIMD_AVX2) &&
+ IS_ALIGNED_AVX(jconst_rgb_gray_convert_avx2))
+ return 1;
if ((simd_support & JSIMD_SSE2) &&
IS_ALIGNED_SSE(jconst_rgb_gray_convert_sse2))
return 1;
@@ -110,6 +123,9 @@
if ((RGB_PIXELSIZE != 3) && (RGB_PIXELSIZE != 4))
return 0;
+ if ((simd_support & JSIMD_AVX2) &&
+ IS_ALIGNED_AVX(jconst_ycc_rgb_convert_avx2))
+ return 1;
if ((simd_support & JSIMD_SSE2) &&
IS_ALIGNED_SSE(jconst_ycc_rgb_convert_sse2))
return 1;
@@ -128,37 +144,48 @@
JSAMPARRAY input_buf, JSAMPIMAGE output_buf,
JDIMENSION output_row, int num_rows)
{
+ void (*avx2fct)(JDIMENSION, JSAMPARRAY, JSAMPIMAGE, JDIMENSION, int);
void (*sse2fct)(JDIMENSION, JSAMPARRAY, JSAMPIMAGE, JDIMENSION, int);
switch(cinfo->in_color_space) {
case JCS_EXT_RGB:
+ avx2fct=jsimd_extrgb_ycc_convert_avx2;
sse2fct=jsimd_extrgb_ycc_convert_sse2;
break;
case JCS_EXT_RGBX:
case JCS_EXT_RGBA:
+ avx2fct=jsimd_extrgbx_ycc_convert_avx2;
sse2fct=jsimd_extrgbx_ycc_convert_sse2;
break;
case JCS_EXT_BGR:
+ avx2fct=jsimd_extbgr_ycc_convert_avx2;
sse2fct=jsimd_extbgr_ycc_convert_sse2;
break;
case JCS_EXT_BGRX:
case JCS_EXT_BGRA:
+ avx2fct=jsimd_extbgrx_ycc_convert_avx2;
sse2fct=jsimd_extbgrx_ycc_convert_sse2;
break;
case JCS_EXT_XBGR:
case JCS_EXT_ABGR:
+ avx2fct=jsimd_extxbgr_ycc_convert_avx2;
sse2fct=jsimd_extxbgr_ycc_convert_sse2;
break;
case JCS_EXT_XRGB:
case JCS_EXT_ARGB:
+ avx2fct=jsimd_extxrgb_ycc_convert_avx2;
sse2fct=jsimd_extxrgb_ycc_convert_sse2;
break;
default:
+ avx2fct=jsimd_rgb_ycc_convert_avx2;
sse2fct=jsimd_rgb_ycc_convert_sse2;
break;
}
- sse2fct(cinfo->image_width, input_buf, output_buf, output_row, num_rows);
+ if (simd_support & JSIMD_AVX2)
+ avx2fct(cinfo->image_width, input_buf, output_buf, output_row, num_rows);
+ else
+ sse2fct(cinfo->image_width, input_buf, output_buf, output_row, num_rows);
}
GLOBAL(void)
@@ -166,37 +193,48 @@
JSAMPARRAY input_buf, JSAMPIMAGE output_buf,
JDIMENSION output_row, int num_rows)
{
+ void (*avx2fct)(JDIMENSION, JSAMPARRAY, JSAMPIMAGE, JDIMENSION, int);
void (*sse2fct)(JDIMENSION, JSAMPARRAY, JSAMPIMAGE, JDIMENSION, int);
switch(cinfo->in_color_space) {
case JCS_EXT_RGB:
+ avx2fct=jsimd_extrgb_gray_convert_avx2;
sse2fct=jsimd_extrgb_gray_convert_sse2;
break;
case JCS_EXT_RGBX:
case JCS_EXT_RGBA:
+ avx2fct=jsimd_extrgbx_gray_convert_avx2;
sse2fct=jsimd_extrgbx_gray_convert_sse2;
break;
case JCS_EXT_BGR:
+ avx2fct=jsimd_extbgr_gray_convert_avx2;
sse2fct=jsimd_extbgr_gray_convert_sse2;
break;
case JCS_EXT_BGRX:
case JCS_EXT_BGRA:
+ avx2fct=jsimd_extbgrx_gray_convert_avx2;
sse2fct=jsimd_extbgrx_gray_convert_sse2;
break;
case JCS_EXT_XBGR:
case JCS_EXT_ABGR:
+ avx2fct=jsimd_extxbgr_gray_convert_avx2;
sse2fct=jsimd_extxbgr_gray_convert_sse2;
break;
case JCS_EXT_XRGB:
case JCS_EXT_ARGB:
+ avx2fct=jsimd_extxrgb_gray_convert_avx2;
sse2fct=jsimd_extxrgb_gray_convert_sse2;
break;
default:
+ avx2fct=jsimd_rgb_gray_convert_avx2;
sse2fct=jsimd_rgb_gray_convert_sse2;
break;
}
- sse2fct(cinfo->image_width, input_buf, output_buf, output_row, num_rows);
+ if (simd_support & JSIMD_AVX2)
+ avx2fct(cinfo->image_width, input_buf, output_buf, output_row, num_rows);
+ else
+ sse2fct(cinfo->image_width, input_buf, output_buf, output_row, num_rows);
}
GLOBAL(void)
@@ -204,37 +242,48 @@
JSAMPIMAGE input_buf, JDIMENSION input_row,
JSAMPARRAY output_buf, int num_rows)
{
+ void (*avx2fct)(JDIMENSION, JSAMPIMAGE, JDIMENSION, JSAMPARRAY, int);
void (*sse2fct)(JDIMENSION, JSAMPIMAGE, JDIMENSION, JSAMPARRAY, int);
switch(cinfo->out_color_space) {
case JCS_EXT_RGB:
+ avx2fct=jsimd_ycc_extrgb_convert_avx2;
sse2fct=jsimd_ycc_extrgb_convert_sse2;
break;
case JCS_EXT_RGBX:
case JCS_EXT_RGBA:
+ avx2fct=jsimd_ycc_extrgbx_convert_avx2;
sse2fct=jsimd_ycc_extrgbx_convert_sse2;
break;
case JCS_EXT_BGR:
+ avx2fct=jsimd_ycc_extbgr_convert_avx2;
sse2fct=jsimd_ycc_extbgr_convert_sse2;
break;
case JCS_EXT_BGRX:
case JCS_EXT_BGRA:
+ avx2fct=jsimd_ycc_extbgrx_convert_avx2;
sse2fct=jsimd_ycc_extbgrx_convert_sse2;
break;
case JCS_EXT_XBGR:
case JCS_EXT_ABGR:
+ avx2fct=jsimd_ycc_extxbgr_convert_avx2;
sse2fct=jsimd_ycc_extxbgr_convert_sse2;
break;
case JCS_EXT_XRGB:
case JCS_EXT_ARGB:
+ avx2fct=jsimd_ycc_extxrgb_convert_avx2;
sse2fct=jsimd_ycc_extxrgb_convert_sse2;
break;
default:
+ avx2fct=jsimd_ycc_rgb_convert_avx2;
sse2fct=jsimd_ycc_rgb_convert_sse2;
break;
}
- sse2fct(cinfo->output_width, input_buf, input_row, output_buf, num_rows);
+ if (simd_support & JSIMD_AVX2)
+ avx2fct(cinfo->output_width, input_buf, input_row, output_buf, num_rows);
+ else
+ sse2fct(cinfo->output_width, input_buf, input_row, output_buf, num_rows);
}
GLOBAL(void)
@@ -255,6 +304,8 @@
if (sizeof(JDIMENSION) != 4)
return 0;
+ if (simd_support & JSIMD_AVX2)
+ return 1;
if (simd_support & JSIMD_SSE2)
return 1;
@@ -272,6 +323,8 @@
if (sizeof(JDIMENSION) != 4)
return 0;
+ if (simd_support & JSIMD_AVX2)
+ return 1;
if (simd_support & JSIMD_SSE2)
return 1;
@@ -282,18 +335,32 @@
jsimd_h2v2_downsample (j_compress_ptr cinfo, jpeg_component_info *compptr,
JSAMPARRAY input_data, JSAMPARRAY output_data)
{
- jsimd_h2v2_downsample_sse2(cinfo->image_width, cinfo->max_v_samp_factor,
- compptr->v_samp_factor, compptr->width_in_blocks,
- input_data, output_data);
+ if (simd_support & JSIMD_AVX2)
+ jsimd_h2v2_downsample_avx2(cinfo->image_width, cinfo->max_v_samp_factor,
+ compptr->v_samp_factor,
+ compptr->width_in_blocks, input_data,
+ output_data);
+ else
+ jsimd_h2v2_downsample_sse2(cinfo->image_width, cinfo->max_v_samp_factor,
+ compptr->v_samp_factor,
+ compptr->width_in_blocks, input_data,
+ output_data);
}
GLOBAL(void)
jsimd_h2v1_downsample (j_compress_ptr cinfo, jpeg_component_info *compptr,
JSAMPARRAY input_data, JSAMPARRAY output_data)
{
- jsimd_h2v1_downsample_sse2(cinfo->image_width, cinfo->max_v_samp_factor,
- compptr->v_samp_factor, compptr->width_in_blocks,
- input_data, output_data);
+ if (simd_support & JSIMD_AVX2)
+ jsimd_h2v1_downsample_avx2(cinfo->image_width, cinfo->max_v_samp_factor,
+ compptr->v_samp_factor,
+ compptr->width_in_blocks, input_data,
+ output_data);
+ else
+ jsimd_h2v1_downsample_sse2(cinfo->image_width, cinfo->max_v_samp_factor,
+ compptr->v_samp_factor,
+ compptr->width_in_blocks, input_data,
+ output_data);
}
GLOBAL(int)
@@ -307,6 +374,8 @@
if (sizeof(JDIMENSION) != 4)
return 0;
+ if (simd_support & JSIMD_AVX2)
+ return 1;
if (simd_support & JSIMD_SSE2)
return 1;
@@ -324,6 +393,8 @@
if (sizeof(JDIMENSION) != 4)
return 0;
+ if (simd_support & JSIMD_AVX2)
+ return 1;
if (simd_support & JSIMD_SSE2)
return 1;
@@ -336,8 +407,12 @@
JSAMPARRAY input_data,
JSAMPARRAY *output_data_ptr)
{
- jsimd_h2v2_upsample_sse2(cinfo->max_v_samp_factor, cinfo->output_width,
- input_data, output_data_ptr);
+ if (simd_support & JSIMD_AVX2)
+ jsimd_h2v2_upsample_avx2(cinfo->max_v_samp_factor, cinfo->output_width,
+ input_data, output_data_ptr);
+ else
+ jsimd_h2v2_upsample_sse2(cinfo->max_v_samp_factor, cinfo->output_width,
+ input_data, output_data_ptr);
}
GLOBAL(void)
@@ -346,8 +421,12 @@
JSAMPARRAY input_data,
JSAMPARRAY *output_data_ptr)
{
- jsimd_h2v1_upsample_sse2(cinfo->max_v_samp_factor, cinfo->output_width,
- input_data, output_data_ptr);
+ if (simd_support & JSIMD_AVX2)
+ jsimd_h2v1_upsample_avx2(cinfo->max_v_samp_factor, cinfo->output_width,
+ input_data, output_data_ptr);
+ else
+ jsimd_h2v1_upsample_sse2(cinfo->max_v_samp_factor, cinfo->output_width,
+ input_data, output_data_ptr);
}
GLOBAL(int)
@@ -361,6 +440,9 @@
if (sizeof(JDIMENSION) != 4)
return 0;
+ if ((simd_support & JSIMD_AVX2) &&
+ IS_ALIGNED_AVX(jconst_fancy_upsample_avx2))
+ return 1;
if ((simd_support & JSIMD_SSE2) &&
IS_ALIGNED_SSE(jconst_fancy_upsample_sse2))
return 1;
@@ -379,6 +461,9 @@
if (sizeof(JDIMENSION) != 4)
return 0;
+ if ((simd_support & JSIMD_AVX2) &&
+ IS_ALIGNED_AVX(jconst_fancy_upsample_avx2))
+ return 1;
if ((simd_support & JSIMD_SSE2) &&
IS_ALIGNED_SSE(jconst_fancy_upsample_sse2))
return 1;
@@ -392,9 +477,14 @@
JSAMPARRAY input_data,
JSAMPARRAY *output_data_ptr)
{
- jsimd_h2v2_fancy_upsample_sse2(cinfo->max_v_samp_factor,
- compptr->downsampled_width, input_data,
- output_data_ptr);
+ if (simd_support & JSIMD_AVX2)
+ jsimd_h2v2_fancy_upsample_avx2(cinfo->max_v_samp_factor,
+ compptr->downsampled_width, input_data,
+ output_data_ptr);
+ else
+ jsimd_h2v2_fancy_upsample_sse2(cinfo->max_v_samp_factor,
+ compptr->downsampled_width, input_data,
+ output_data_ptr);
}
GLOBAL(void)
@@ -403,9 +493,14 @@
JSAMPARRAY input_data,
JSAMPARRAY *output_data_ptr)
{
- jsimd_h2v1_fancy_upsample_sse2(cinfo->max_v_samp_factor,
- compptr->downsampled_width, input_data,
- output_data_ptr);
+ if (simd_support & JSIMD_AVX2)
+ jsimd_h2v1_fancy_upsample_avx2(cinfo->max_v_samp_factor,
+ compptr->downsampled_width, input_data,
+ output_data_ptr);
+ else
+ jsimd_h2v1_fancy_upsample_sse2(cinfo->max_v_samp_factor,
+ compptr->downsampled_width, input_data,
+ output_data_ptr);
}
GLOBAL(int)
@@ -419,6 +514,9 @@
if (sizeof(JDIMENSION) != 4)
return 0;
+ if ((simd_support & JSIMD_AVX2) &&
+ IS_ALIGNED_AVX(jconst_merged_upsample_avx2))
+ return 1;
if ((simd_support & JSIMD_SSE2) &&
IS_ALIGNED_SSE(jconst_merged_upsample_sse2))
return 1;
@@ -437,6 +535,9 @@
if (sizeof(JDIMENSION) != 4)
return 0;
+ if ((simd_support & JSIMD_AVX2) &&
+ IS_ALIGNED_AVX(jconst_merged_upsample_avx2))
+ return 1;
if ((simd_support & JSIMD_SSE2) &&
IS_ALIGNED_SSE(jconst_merged_upsample_sse2))
return 1;
@@ -450,37 +551,48 @@
JDIMENSION in_row_group_ctr,
JSAMPARRAY output_buf)
{
+ void (*avx2fct)(JDIMENSION, JSAMPIMAGE, JDIMENSION, JSAMPARRAY);
void (*sse2fct)(JDIMENSION, JSAMPIMAGE, JDIMENSION, JSAMPARRAY);
switch(cinfo->out_color_space) {
case JCS_EXT_RGB:
+ avx2fct=jsimd_h2v2_extrgb_merged_upsample_avx2;
sse2fct=jsimd_h2v2_extrgb_merged_upsample_sse2;
break;
case JCS_EXT_RGBX:
case JCS_EXT_RGBA:
+ avx2fct=jsimd_h2v2_extrgbx_merged_upsample_avx2;
sse2fct=jsimd_h2v2_extrgbx_merged_upsample_sse2;
break;
case JCS_EXT_BGR:
+ avx2fct=jsimd_h2v2_extbgr_merged_upsample_avx2;
sse2fct=jsimd_h2v2_extbgr_merged_upsample_sse2;
break;
case JCS_EXT_BGRX:
case JCS_EXT_BGRA:
+ avx2fct=jsimd_h2v2_extbgrx_merged_upsample_avx2;
sse2fct=jsimd_h2v2_extbgrx_merged_upsample_sse2;
break;
case JCS_EXT_XBGR:
case JCS_EXT_ABGR:
+ avx2fct=jsimd_h2v2_extxbgr_merged_upsample_avx2;
sse2fct=jsimd_h2v2_extxbgr_merged_upsample_sse2;
break;
case JCS_EXT_XRGB:
case JCS_EXT_ARGB:
+ avx2fct=jsimd_h2v2_extxrgb_merged_upsample_avx2;
sse2fct=jsimd_h2v2_extxrgb_merged_upsample_sse2;
break;
default:
+ avx2fct=jsimd_h2v2_merged_upsample_avx2;
sse2fct=jsimd_h2v2_merged_upsample_sse2;
break;
}
- sse2fct(cinfo->output_width, input_buf, in_row_group_ctr, output_buf);
+ if (simd_support & JSIMD_AVX2)
+ avx2fct(cinfo->output_width, input_buf, in_row_group_ctr, output_buf);
+ else
+ sse2fct(cinfo->output_width, input_buf, in_row_group_ctr, output_buf);
}
GLOBAL(void)
@@ -489,37 +601,48 @@
JDIMENSION in_row_group_ctr,
JSAMPARRAY output_buf)
{
+ void (*avx2fct)(JDIMENSION, JSAMPIMAGE, JDIMENSION, JSAMPARRAY);
void (*sse2fct)(JDIMENSION, JSAMPIMAGE, JDIMENSION, JSAMPARRAY);
switch(cinfo->out_color_space) {
case JCS_EXT_RGB:
+ avx2fct=jsimd_h2v1_extrgb_merged_upsample_avx2;
sse2fct=jsimd_h2v1_extrgb_merged_upsample_sse2;
break;
case JCS_EXT_RGBX:
case JCS_EXT_RGBA:
+ avx2fct=jsimd_h2v1_extrgbx_merged_upsample_avx2;
sse2fct=jsimd_h2v1_extrgbx_merged_upsample_sse2;
break;
case JCS_EXT_BGR:
+ avx2fct=jsimd_h2v1_extbgr_merged_upsample_avx2;
sse2fct=jsimd_h2v1_extbgr_merged_upsample_sse2;
break;
case JCS_EXT_BGRX:
case JCS_EXT_BGRA:
+ avx2fct=jsimd_h2v1_extbgrx_merged_upsample_avx2;
sse2fct=jsimd_h2v1_extbgrx_merged_upsample_sse2;
break;
case JCS_EXT_XBGR:
case JCS_EXT_ABGR:
+ avx2fct=jsimd_h2v1_extxbgr_merged_upsample_avx2;
sse2fct=jsimd_h2v1_extxbgr_merged_upsample_sse2;
break;
case JCS_EXT_XRGB:
case JCS_EXT_ARGB:
+ avx2fct=jsimd_h2v1_extxrgb_merged_upsample_avx2;
sse2fct=jsimd_h2v1_extxrgb_merged_upsample_sse2;
break;
default:
+ avx2fct=jsimd_h2v1_merged_upsample_avx2;
sse2fct=jsimd_h2v1_merged_upsample_sse2;
break;
}
- sse2fct(cinfo->output_width, input_buf, in_row_group_ctr, output_buf);
+ if (simd_support & JSIMD_AVX2)
+ avx2fct(cinfo->output_width, input_buf, in_row_group_ctr, output_buf);
+ else
+ sse2fct(cinfo->output_width, input_buf, in_row_group_ctr, output_buf);
}
GLOBAL(int)
@@ -660,6 +783,8 @@
if (sizeof(DCTELEM) != 2)
return 0;
+ if (simd_support & JSIMD_AVX2)
+ return 1;
if (simd_support & JSIMD_SSE2)
return 1;
@@ -689,7 +814,10 @@
jsimd_quantize (JCOEFPTR coef_block, DCTELEM *divisors,
DCTELEM *workspace)
{
- jsimd_quantize_sse2(coef_block, divisors, workspace);
+ if (simd_support & JSIMD_AVX2)
+ jsimd_quantize_avx2(coef_block, divisors, workspace);
+ else
+ jsimd_quantize_sse2(coef_block, divisors, workspace);
}
GLOBAL(void)
diff --git a/simd/x86_64/jsimdcpu.asm b/simd/x86_64/jsimdcpu.asm
new file mode 100644
index 0000000..a1ebb85
--- /dev/null
+++ b/simd/x86_64/jsimdcpu.asm
@@ -0,0 +1,78 @@
+;
+; jsimdcpu.asm - SIMD instruction support check
+;
+; Copyright 2009 Pierre Ossman <ossman@cendio.se> for Cendio AB
+; Copyright (C) 2016, D. R. Commander.
+;
+; Based on
+; x86 SIMD extension for IJG JPEG library
+; Copyright (C) 1999-2006, MIYASAKA Masaru.
+; For conditions of distribution and use, see copyright notice in jsimdext.inc
+;
+; This file should be assembled with NASM (Netwide Assembler),
+; can *not* be assembled with Microsoft's MASM or any compatible
+; assembler (including Borland's Turbo Assembler).
+; NASM is available from http://nasm.sourceforge.net/ or
+; http://sourceforge.net/project/showfiles.php?group_id=6208
+;
+; [TAB8]
+
+%include "jsimdext.inc"
+
+; --------------------------------------------------------------------------
+ SECTION SEG_TEXT
+ BITS 64
+;
+; Check if the CPU supports SIMD instructions
+;
+; GLOBAL(unsigned int)
+; jpeg_simd_cpu_support (void)
+;
+
+ align 32
+ global EXTN(jpeg_simd_cpu_support)
+
+EXTN(jpeg_simd_cpu_support):
+ push rbx
+ push rdi
+
+ xor rdi, rdi ; simd support flag
+
+ ; Check for AVX2 instruction support
+ mov rax, 7
+ xor rcx, rcx
+ cpuid
+ mov rax, rbx ; rax = Extended feature flags
+
+ or rdi, JSIMD_SSE2
+ or rdi, JSIMD_SSE
+ test rax, 1<<5 ; bit5:AVX2
+ jz short .return
+
+ ; Check for AVX2 O/S support
+ mov rax, 1
+ xor rcx, rcx
+ cpuid
+ test rcx, 1<<27
+ jz short .return ; O/S does not support XSAVE
+ test rcx, 1<<28
+ jz short .return ; CPU does not support AVX2
+
+ xor rcx, rcx
+ xgetbv
+ test rax, 6 ; O/S does not manage XMM/YMM state
+ ; using XSAVE
+ jz short .return
+
+ or rdi, JSIMD_AVX2
+
+.return:
+ mov rax, rdi
+
+ pop rdi
+ pop rbx
+ ret
+
+; For some reason, the OS X linker does not honor the request to align the
+; segment unless we do this.
+ align 32
diff --git a/testimages/test1.icc b/testimages/test1.icc
new file mode 100644
index 0000000..d0245c8
--- /dev/null
+++ b/testimages/test1.icc
Binary files differ
diff --git a/testimages/test1.icc.txt b/testimages/test1.icc.txt
new file mode 100644
index 0000000..57fc52f
--- /dev/null
+++ b/testimages/test1.icc.txt
@@ -0,0 +1,20 @@
+Little CMS
+Copyright (c) 1998-2011 Marti Maria Saguer
+
+Permission is hereby granted, free of charge, to any person obtaining a copy of
+this software and associated documentation files (the "Software"), to deal in
+the Software without restriction, including without limitation the rights to
+use, copy, modify, merge, publish, distribute, sublicense, and/or sell copies
+of the Software, and to permit persons to whom the Software is furnished to do
+so, subject to the following conditions:
+
+The above copyright notice and this permission notice shall be included in all
+copies or substantial portions of the Software.
+
+THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
+AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
+LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
+OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
+SOFTWARE.
diff --git a/testimages/test2.icc b/testimages/test2.icc
new file mode 100644
index 0000000..73f1b5a
--- /dev/null
+++ b/testimages/test2.icc
Binary files differ
diff --git a/testimages/test2.icc.txt b/testimages/test2.icc.txt
new file mode 100644
index 0000000..57fc52f
--- /dev/null
+++ b/testimages/test2.icc.txt
@@ -0,0 +1,20 @@
+Little CMS
+Copyright (c) 1998-2011 Marti Maria Saguer
+
+Permission is hereby granted, free of charge, to any person obtaining a copy of
+this software and associated documentation files (the "Software"), to deal in
+the Software without restriction, including without limitation the rights to
+use, copy, modify, merge, publish, distribute, sublicense, and/or sell copies
+of the Software, and to permit persons to whom the Software is furnished to do
+so, subject to the following conditions:
+
+The above copyright notice and this permission notice shall be included in all
+copies or substantial portions of the Software.
+
+THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
+AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
+LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
+OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
+SOFTWARE.
diff --git a/tjbench.c b/tjbench.c
index 21b297c..a627510 100644
--- a/tjbench.c
+++ b/tjbench.c
@@ -38,11 +38,44 @@
#include "./turbojpeg.h"
-#define _throw(op, err) { \
+#define _throw(op, err) \
+{ \
printf("ERROR in line %d while %s:\n%s\n", __LINE__, op, err); \
- retval=-1; goto bailout;}
+ retval=-1; goto bailout; \
+}
#define _throwunix(m) _throw(m, strerror(errno))
-#define _throwtj(m) _throw(m, tjGetErrorStr())
+
+char tjErrorStr[JMSG_LENGTH_MAX]="\0", tjErrorMsg[JMSG_LENGTH_MAX]="\0";
+int tjErrorLine=-1, tjErrorCode=-1;
+
+#define _throwtj(m) \
+{ \
+ int _tjErrorCode=tjGetErrorCode(handle); \
+ char *_tjErrorStr=tjGetErrorStr2(handle); \
+ \
+ if(!(flags&TJFLAG_STOPONWARNING) && _tjErrorCode==TJERR_WARNING) \
+ { \
+ if(strncmp(tjErrorStr, _tjErrorStr, JMSG_LENGTH_MAX) || \
+ strncmp(tjErrorMsg, m, JMSG_LENGTH_MAX) || \
+ tjErrorCode!=_tjErrorCode || tjErrorLine!=__LINE__) \
+ { \
+ strncpy(tjErrorStr, _tjErrorStr, JMSG_LENGTH_MAX); \
+ strncpy(tjErrorMsg, m, JMSG_LENGTH_MAX); \
+ tjErrorCode=_tjErrorCode; \
+ tjErrorLine=__LINE__; \
+ printf("WARNING in line %d while %s:\n%s\n", __LINE__, m, \
+ _tjErrorStr); \
+ } \
+ } \
+ else \
+ { \
+ printf("%s in line %d while %s:\n%s\n", \
+ _tjErrorCode==TJERR_WARNING ? "WARNING" : "ERROR", __LINE__, m, \
+ _tjErrorStr); \
+ retval=-1; goto bailout; \
+ } \
+}
+
#define _throwbmp(m) _throw(m, bmpgeterr())
int flags=TJFLAG_NOREALLOC, componly=0, decomponly=0, doyuv=0, quiet=0,
@@ -751,6 +784,8 @@
printf(" codec\n");
printf("-accuratedct = Use the most accurate DCT/IDCT algorithms available in the\n");
printf(" underlying codec\n");
+ printf("-progressive = Use progressive entropy coding in JPEG images generated by\n");
+ printf(" compression and transform operations.\n");
printf("-subsamp <s> = When testing JPEG compression, this option specifies the level\n");
printf(" of chrominance subsampling to use (<s> = 444, 422, 440, 420, 411, or\n");
printf(" GRAY). The default is to test Grayscale, 4:2:0, 4:2:2, and 4:4:4 in\n");
@@ -779,13 +814,18 @@
printf(" decompression (these options are mutually exclusive)\n");
printf("-grayscale = Perform lossless grayscale conversion prior to decompression\n");
printf(" test (can be combined with the other transforms above)\n");
+ printf("-copynone = Do not copy any extra markers (including EXIF and ICC profile data)\n");
+ printf(" when transforming the image.\n");
printf("-benchtime <t> = Run each benchmark for at least <t> seconds (default = 5.0)\n");
printf("-warmup <t> = Run each benchmark for <t> seconds (default = 1.0) prior to\n");
printf(" starting the timer, in order to prime the caches and thus improve the\n");
printf(" consistency of the results.\n");
printf("-componly = Stop after running compression tests. Do not test decompression.\n");
printf("-nowrite = Do not write reference or output images (improves consistency of\n");
- printf(" performance measurements.)\n\n");
+ printf(" performance measurements.)\n");
+ printf("-stoponwarning = Immediately discontinue the current\n");
+ printf(" compression/decompression/transform operation if the underlying codec\n");
+ printf(" throws a warning (non-fatal error)\n\n");
printf("NOTE: If the quality is specified as a range (e.g. 90-100), a separate\n");
printf("test will be performed for all quality values in the range.\n\n");
exit(1);
@@ -799,7 +839,7 @@
int minarg=2, retval=0, subsamp=-1;
if((scalingfactors=tjGetScalingFactors(&nsf))==NULL || nsf==0)
- _throwtj("executing tjGetScalingFactors()");
+ _throw("executing tjGetScalingFactors()", tjGetErrorStr());
if(argc<minarg) usage(argv[0]);
@@ -850,6 +890,11 @@
printf("Using most accurate DCT/IDCT algorithm\n\n");
flags|=TJFLAG_ACCURATEDCT;
}
+ if(!strcasecmp(argv[i], "-progressive"))
+ {
+ printf("Using progressive entropy coding\n\n");
+ flags|=TJFLAG_PROGRESSIVE;
+ }
if(!strcasecmp(argv[i], "-rgb")) pf=TJPF_RGB;
if(!strcasecmp(argv[i], "-rgbx")) pf=TJPF_RGBX;
if(!strcasecmp(argv[i], "-bgr")) pf=TJPF_BGR;
@@ -888,6 +933,7 @@
if(!strcasecmp(argv[i], "-grayscale")) xformopt|=TJXOPT_GRAY;
if(!strcasecmp(argv[i], "-custom")) customFilter=dummyDCTFilter;
if(!strcasecmp(argv[i], "-nooutput")) xformopt|=TJXOPT_NOOUTPUT;
+ if(!strcasecmp(argv[i], "-copynone")) xformopt|=TJXOPT_COPYNONE;
if(!strcasecmp(argv[i], "-benchtime") && i<argc-1)
{
double temp=atof(argv[++i]);
@@ -933,6 +979,7 @@
}
if(!strcasecmp(argv[i], "-componly")) componly=1;
if(!strcasecmp(argv[i], "-nowrite")) dowrite=0;
+ if(!strcasecmp(argv[i], "-stoponwarning")) flags|=TJFLAG_STOPONWARNING;
}
}
diff --git a/tjbenchtest.in b/tjbenchtest.in
index ef11b24..ad4a493 100755
--- a/tjbenchtest.in
+++ b/tjbenchtest.in
@@ -21,9 +21,9 @@
EXT=bmp
IMAGES="vgl_5674_0098.${EXT} vgl_6434_0018a.${EXT} vgl_6548_0026a.${EXT} nightshot_iso_100.${EXT}"
-IMGDIR=@srcdir@/testimages
+IMGDIR=@CMAKE_CURRENT_SOURCE_DIR@/testimages
OUTDIR=`mktemp -d /tmp/__tjbenchtest_output.XXXXXX`
-EXEDIR=.
+EXEDIR=@CMAKE_CURRENT_BINARY_DIR@
BMPARG=
NSARG=
YUVARG=
@@ -164,7 +164,7 @@
# Scaled decompression
for scale in 2_1 15_8 7_4 13_8 3_2 11_8 5_4 9_8 7_8 3_4 5_8 1_2 3_8 1_4 1_8; do
- scalearg=`echo $scale | sed s@_@/@g`
+ scalearg=`echo $scale | sed 's/\_/\//g'`
for samp in GRAY 420 422 444; do
runme $EXEDIR/djpeg -rgb -scale ${scalearg} $NSARG $BMPARG -outfile $OUTDIR/${basename}_${samp}_${scale}_djpeg.${EXT} $OUTDIR/${basename}_${samp}_fast_cjpeg.jpg
runme $EXEDIR/tjbench $OUTDIR/${basename}_${samp}_Q95.jpg $BMPARG -scale ${scalearg} -quiet -benchtime 0.01 -warmup 0 $YUVARG $ALLOCARG
@@ -235,7 +235,7 @@
for xform in hflip vflip transpose transverse rot90 rot180 rot270; do
for samp in GRAY 444 422 420; do
for scale in 2_1 15_8 7_4 13_8 3_2 11_8 5_4 9_8 7_8 3_4 5_8 1_2 3_8 1_4 1_8; do
- scalearg=`echo $scale | sed s@_@/@g`
+ scalearg=`echo $scale | sed 's/\_/\//g'`
runme $EXEDIR/djpeg -rgb -scale ${scalearg} $NSARG $BMPARG -outfile $OUTDIR/${basename}_${samp}_${xform}_${scale}_jpegtran.${EXT} $OUTDIR/${basename}_${samp}_${xform}_jpegtran.jpg
runme $EXEDIR/tjbench $OUTDIR/${basename}_${samp}_Q95.jpg $BMPARG -$xform -scale ${scalearg} -quiet -benchtime 0.01 -warmup 0 $YUVARG $ALLOCARG
runme cmp $OUTDIR/${basename}_${samp}_Q95_${scale}.${EXT} $OUTDIR/${basename}_${samp}_${xform}_${scale}_jpegtran.${EXT}
diff --git a/tjbenchtest.java.in b/tjbenchtest.java.in
index acdabd0..91f2112 100755
--- a/tjbenchtest.java.in
+++ b/tjbenchtest.java.in
@@ -20,10 +20,10 @@
}
IMAGES="vgl_5674_0098.bmp vgl_6434_0018a.bmp vgl_6548_0026a.bmp nightshot_iso_100.bmp"
-IMGDIR=@srcdir@/testimages
+IMGDIR=@CMAKE_CURRENT_SOURCE_DIR@/testimages
OUTDIR=`mktemp -d /tmp/__tjbenchtest_java_output.XXXXXX`
-EXEDIR=.
-JAVA="@JAVA@ -cp java/turbojpeg.jar -Djava.library.path=.libs"
+EXEDIR=@CMAKE_CURRENT_BINARY_DIR@
+JAVA="@Java_JAVA_EXECUTABLE@ -cp $EXEDIR/java/turbojpeg.jar -Djava.library.path=$EXEDIR"
BMPARG=
NSARG=
YUVARG=
@@ -137,7 +137,7 @@
# Scaled decompression
for scale in 2_1 15_8 7_4 13_8 3_2 11_8 5_4 9_8 7_8 3_4 5_8 1_2 3_8 1_4 1_8; do
- scalearg=`echo $scale | sed s@_@/@g`
+ scalearg=`echo $scale | sed 's/\_/\//g'`
for samp in GRAY 420 422 444; do
runme $EXEDIR/djpeg -rgb -scale ${scalearg} $NSARG -bmp -outfile $OUTDIR/${basename}_${samp}_${scale}_djpeg.bmp $OUTDIR/${basename}_${samp}_fast_cjpeg.jpg
runme $JAVA TJBench $OUTDIR/${basename}_${samp}_Q95.jpg -scale ${scalearg} -quiet -benchtime 0.01 -warmup 0 $YUVARG
@@ -193,7 +193,7 @@
for xform in hflip vflip transpose transverse rot90 rot180 rot270; do
for samp in GRAY 444 422 420; do
for scale in 2_1 15_8 7_4 13_8 3_2 11_8 5_4 9_8 7_8 3_4 5_8 1_2 3_8 1_4 1_8; do
- scalearg=`echo $scale | sed s@_@/@g`
+ scalearg=`echo $scale | sed 's/\_/\//g'`
runme $EXEDIR/djpeg -rgb -scale ${scalearg} $NSARG -bmp -outfile $OUTDIR/${basename}_${samp}_${xform}_${scale}_jpegtran.bmp $OUTDIR/${basename}_${samp}_${xform}_jpegtran.jpg
runme $JAVA TJBench $OUTDIR/${basename}_${samp}_Q95.jpg -$xform -scale ${scalearg} -quiet -benchtime 0.01 -warmup 0 $YUVARG
runme cmp -i 54:54 $OUTDIR/${basename}_${samp}_Q95_${scale}.bmp $OUTDIR/${basename}_${samp}_${xform}_${scale}_jpegtran.bmp
diff --git a/tjexampletest.in b/tjexampletest.in
index 40b342e..619d04f 100755
--- a/tjexampletest.in
+++ b/tjexampletest.in
@@ -20,10 +20,10 @@
}
IMAGES="vgl_5674_0098.bmp vgl_6434_0018a.bmp vgl_6548_0026a.bmp nightshot_iso_100.bmp"
-IMGDIR=@srcdir@/testimages
+IMGDIR=@CMAKE_CURRENT_SOURCE_DIR@/testimages
OUTDIR=__tjexampletest_output
-EXEDIR=.
-JAVA="@JAVA@ -cp java/turbojpeg.jar -Djava.library.path=.libs"
+EXEDIR=@CMAKE_CURRENT_BINARY_DIR@
+JAVA="@Java_JAVA_EXECUTABLE@ -cp $EXEDIR/java/turbojpeg.jar -Djava.library.path=$EXEDIR"
if [ -d $OUTDIR ]; then
rm -rf $OUTDIR
@@ -85,7 +85,7 @@
# Scaled decompression
for scale in 2_1 15_8 7_4 13_8 3_2 11_8 5_4 9_8 7_8 3_4 5_8 1_2 3_8 1_4 1_8; do
- scalearg=`echo $scale | sed s@_@/@g`
+ scalearg=`echo $scale | sed 's/\_/\//g'`
for samp in GRAY 420 422 444; do
$EXEDIR/djpeg -rgb -bmp -scale ${scalearg} $OUTDIR/${basename}_${samp}_fast_cjpeg.jpg >$OUTDIR/${basename}_${samp}_${scale}_djpeg.bmp
runme $JAVA TJExample $OUTDIR/${basename}_${samp}_fast.jpg $OUTDIR/${basename}_${samp}_${scale}.bmp -scale ${scalearg}
@@ -136,7 +136,7 @@
for xform in hflip vflip transpose transverse rot90 rot180 rot270; do
for samp in GRAY 444 422 420; do
for scale in 2_1 15_8 7_4 13_8 3_2 11_8 5_4 9_8 7_8 3_4 5_8 1_2 3_8 1_4 1_8; do
- scalearg=`echo $scale | sed s@_@/@g`
+ scalearg=`echo $scale | sed 's/\_/\//g'`
$EXEDIR/djpeg -rgb -bmp -scale ${scalearg} $OUTDIR/${basename}_${samp}_${xform}_jpegtran.jpg >$OUTDIR/${basename}_${samp}_${xform}_${scale}_jpegtran.bmp
runme $JAVA TJExample $OUTDIR/${basename}_${samp}_fast.jpg $OUTDIR/${basename}_${samp}_${xform}_${scale}.bmp -$xform -scale ${scalearg} -crop 16,16,70x60
runme cmp -i 54:54 $OUTDIR/${basename}_${samp}_${xform}_${scale}.bmp $OUTDIR/${basename}_${samp}_${xform}_${scale}_jpegtran.bmp
diff --git a/transupp.c b/transupp.c
index b51ef39..569c509 100644
--- a/transupp.c
+++ b/transupp.c
@@ -1576,9 +1576,12 @@
jpeg_save_markers(srcinfo, JPEG_COM, 0xFFFF);
}
/* Save all types of APPn markers iff ALL option */
- if (option == JCOPYOPT_ALL) {
- for (m = 0; m < 16; m++)
+ if (option == JCOPYOPT_ALL || option == JCOPYOPT_ALL_EXCEPT_ICC) {
+ for (m = 0; m < 16; m++) {
+ if (option == JCOPYOPT_ALL_EXCEPT_ICC && m == 2)
+ continue;
jpeg_save_markers(srcinfo, JPEG_APP0 + m, 0xFFFF);
+ }
}
#endif /* SAVE_MARKERS_SUPPORTED */
}
diff --git a/transupp.h b/transupp.h
index bf3118a..ac96b98 100644
--- a/transupp.h
+++ b/transupp.h
@@ -3,8 +3,8 @@
*
* This file was part of the Independent JPEG Group's software:
* Copyright (C) 1997-2011, Thomas G. Lane, Guido Vollbeding.
- * It was modified by The libjpeg-turbo Project to include only code relevant
- * to libjpeg-turbo.
+ * libjpeg-turbo Modifications:
+ * Copyright (C) 2017, D. R. Commander.
* For conditions of distribution and use, see the accompanying README.ijg
* file.
*
@@ -193,7 +193,8 @@
typedef enum {
JCOPYOPT_NONE, /* copy no optional markers */
JCOPYOPT_COMMENTS, /* copy only comment (COM) markers */
- JCOPYOPT_ALL /* copy all optional markers */
+ JCOPYOPT_ALL, /* copy all optional markers */
+ JCOPYOPT_ALL_EXCEPT_ICC /* copy all optional markers except APP2 */
} JCOPY_OPTION;
#define JCOPYOPT_DEFAULT JCOPYOPT_COMMENTS /* recommended default */
diff --git a/turbojpeg-jni.c b/turbojpeg-jni.c
index eaba670..4f39c91 100644
--- a/turbojpeg-jni.c
+++ b/turbojpeg-jni.c
@@ -1,5 +1,5 @@
/*
- * Copyright (C)2011-2016 D. R. Commander. All Rights Reserved.
+ * Copyright (C)2011-2017 D. R. Commander. All Rights Reserved.
*
* Redistribution and use in source and binary forms, with or without
* modification, are permitted provided that the following conditions are met:
@@ -39,27 +39,41 @@
#define PAD(v, p) ((v+(p)-1)&(~((p)-1)))
+#define bailif0(f) {if(!(f) || (*env)->ExceptionCheck(env)) { \
+ goto bailout; \
+}}
+
#define _throw(msg, exceptionClass) { \
jclass _exccls=(*env)->FindClass(env, exceptionClass); \
- if(!_exccls || (*env)->ExceptionCheck(env)) goto bailout; \
+ bailif0(_exccls); \
(*env)->ThrowNew(env, _exccls, msg); \
goto bailout; \
}
-#define _throwtj() _throw(tjGetErrorStr(), "org/libjpegturbo/turbojpeg/TJException")
+#define _throwtj() { \
+ jclass _exccls; \
+ jmethodID _excid; \
+ jobject _excobj; \
+ jstring _errstr; \
+ bailif0(_errstr = (*env)->NewStringUTF(env, tjGetErrorStr2(handle))); \
+ bailif0(_exccls = (*env)->FindClass(env, \
+ "org/libjpegturbo/turbojpeg/TJException")); \
+ bailif0(_excid = (*env)->GetMethodID(env, _exccls, "<init>", \
+ "(Ljava/lang/String;I)V")); \
+ bailif0(_excobj = (*env)->NewObject(env, _exccls, _excid, _errstr, \
+ tjGetErrorCode(handle))); \
+ (*env)->Throw(env, _excobj); \
+ goto bailout; \
+}
#define _throwarg(msg) _throw(msg, "java/lang/IllegalArgumentException")
#define _throwmem() _throw("Memory allocation failure", "java/lang/OutOfMemoryError");
-#define bailif0(f) {if(!(f) || (*env)->ExceptionCheck(env)) { \
- goto bailout; \
-}}
-
#define gethandle() \
jclass _cls=(*env)->GetObjectClass(env, obj); \
jfieldID _fid; \
- if(!_cls || (*env)->ExceptionCheck(env)) goto bailout; \
+ bailif0(_cls); \
bailif0(_fid=(*env)->GetFieldID(env, _cls, "handle", "J")); \
handle=(tjhandle)(size_t)(*env)->GetLongField(env, obj, _fid); \
@@ -174,7 +188,7 @@
tjhandle handle;
if((handle=tjInitCompress())==NULL)
- _throwtj();
+ _throw(tjGetErrorStr(), "org/libjpegturbo/turbojpeg/TJException");
bailif0(cls=(*env)->GetObjectClass(env, obj));
bailif0(fid=(*env)->GetFieldID(env, cls, "handle", "J"));
@@ -551,7 +565,8 @@
jfieldID fid;
tjhandle handle;
- if((handle=tjInitDecompress())==NULL) _throwtj();
+ if((handle=tjInitDecompress())==NULL)
+ _throw(tjGetErrorStr(), "org/libjpegturbo/turbojpeg/TJException");
bailif0(cls=(*env)->GetObjectClass(env, obj));
bailif0(fid=(*env)->GetFieldID(env, cls, "handle", "J"));
@@ -958,7 +973,8 @@
jfieldID fid;
tjhandle handle;
- if((handle=tjInitTransform())==NULL) _throwtj();
+ if((handle=tjInitTransform())==NULL)
+ _throw(tjGetErrorStr(), "org/libjpegturbo/turbojpeg/TJException");
bailif0(cls=(*env)->GetObjectClass(env, obj));
bailif0(fid=(*env)->GetFieldID(env, cls, "handle", "J"));
diff --git a/turbojpeg-mapfile b/turbojpeg-mapfile
index 35d55ae..8df3687 100755
--- a/turbojpeg-mapfile
+++ b/turbojpeg-mapfile
@@ -54,3 +54,10 @@
tjPlaneSizeYUV;
tjPlaneWidth;
} TURBOJPEG_1.2;
+
+TURBOJPEG_1.6
+{
+ global:
+ tjGetErrorCode;
+ tjGetErrorStr2;
+} TURBOJPEG_1.4;
diff --git a/turbojpeg-mapfile.jni b/turbojpeg-mapfile.jni
index 9c1d25b..9395dca 100755
--- a/turbojpeg-mapfile.jni
+++ b/turbojpeg-mapfile.jni
@@ -90,3 +90,10 @@
Java_org_libjpegturbo_turbojpeg_TJ_planeSizeYUV__IIIII;
Java_org_libjpegturbo_turbojpeg_TJ_planeWidth__III;
} TURBOJPEG_1.3;
+
+TURBOJPEG_1.6
+{
+ global:
+ tjGetErrorCode;
+ tjGetErrorStr2;
+} TURBOJPEG_1.4;
diff --git a/turbojpeg.c b/turbojpeg.c
index c9bc619..89c4400 100644
--- a/turbojpeg.c
+++ b/turbojpeg.c
@@ -60,7 +60,7 @@
struct jpeg_error_mgr pub;
jmp_buf setjmp_buffer;
void (*emit_message)(j_common_ptr, int);
- boolean warning;
+ boolean warning, stopOnWarning;
};
typedef struct my_error_mgr *my_error_ptr;
@@ -82,7 +82,11 @@
{
my_error_ptr myerr=(my_error_ptr)cinfo->err;
myerr->emit_message(cinfo, msg_level);
- if(msg_level<0) myerr->warning=TRUE;
+ if(msg_level<0)
+ {
+ myerr->warning=TRUE;
+ if(myerr->stopOnWarning) longjmp(myerr->setjmp_buffer, 1);
+ }
}
@@ -96,6 +100,8 @@
struct jpeg_decompress_struct dinfo;
struct my_error_mgr jerr;
int init, headerRead;
+ char errStr[JMSG_LENGTH_MAX];
+ boolean isInstanceError;
} tjinstance;
static const int pixelsize[TJ_NUMSAMP]={3, 3, 3, 1, 3, 3};
@@ -126,26 +132,31 @@
{1, 8}
};
-#define _throw(m) {snprintf(errStr, JMSG_LENGTH_MAX, "%s", m); \
+#define _throwg(m) {snprintf(errStr, JMSG_LENGTH_MAX, "%s", m); \
retval=-1; goto bailout;}
+#define _throw(m) {snprintf(this->errStr, JMSG_LENGTH_MAX, "%s", m); \
+ this->isInstanceError=TRUE; _throwg(m);}
#define getinstance(handle) tjinstance *this=(tjinstance *)handle; \
j_compress_ptr cinfo=NULL; j_decompress_ptr dinfo=NULL; \
if(!this) {snprintf(errStr, JMSG_LENGTH_MAX, "Invalid handle"); \
return -1;} \
cinfo=&this->cinfo; dinfo=&this->dinfo; \
- this->jerr.warning=FALSE;
+ this->jerr.warning=FALSE; \
+ this->isInstanceError=FALSE;
#define getcinstance(handle) tjinstance *this=(tjinstance *)handle; \
j_compress_ptr cinfo=NULL; \
if(!this) {snprintf(errStr, JMSG_LENGTH_MAX, "Invalid handle"); \
return -1;} \
cinfo=&this->cinfo; \
- this->jerr.warning=FALSE;
+ this->jerr.warning=FALSE; \
+ this->isInstanceError=FALSE;
#define getdinstance(handle) tjinstance *this=(tjinstance *)handle; \
j_decompress_ptr dinfo=NULL; \
if(!this) {snprintf(errStr, JMSG_LENGTH_MAX, "Invalid handle"); \
return -1;} \
dinfo=&this->dinfo; \
- this->jerr.warning=FALSE;
+ this->jerr.warning=FALSE; \
+ this->isInstanceError=FALSE;
static int getPixelFormat(int pixelSize, int flags)
{
@@ -252,8 +263,10 @@
jpeg_set_colorspace(cinfo, JCS_YCCK);
else jpeg_set_colorspace(cinfo, JCS_YCbCr);
+ if(flags&TJFLAG_PROGRESSIVE)
+ jpeg_simple_progression(cinfo);
#ifndef NO_GETENV
- if((env=getenv("TJ_PROGRESSIVE"))!=NULL && strlen(env)>0
+ else if((env=getenv("TJ_PROGRESSIVE"))!=NULL && strlen(env)>0
&& !strcmp(env, "1"))
jpeg_simple_progression(cinfo);
#endif
@@ -272,37 +285,36 @@
return retval;
}
-static int setDecompDefaults(struct jpeg_decompress_struct *dinfo,
- int pixelFormat, int flags)
+static int setDecompDefaults(tjinstance *this, int pixelFormat, int flags)
{
int retval=0;
switch(pixelFormat)
{
case TJPF_GRAY:
- dinfo->out_color_space=JCS_GRAYSCALE; break;
+ this->dinfo.out_color_space=JCS_GRAYSCALE; break;
#if JCS_EXTENSIONS==1
case TJPF_RGB:
- dinfo->out_color_space=JCS_EXT_RGB; break;
+ this->dinfo.out_color_space=JCS_EXT_RGB; break;
case TJPF_BGR:
- dinfo->out_color_space=JCS_EXT_BGR; break;
+ this->dinfo.out_color_space=JCS_EXT_BGR; break;
case TJPF_RGBX:
- dinfo->out_color_space=JCS_EXT_RGBX; break;
+ this->dinfo.out_color_space=JCS_EXT_RGBX; break;
case TJPF_BGRX:
- dinfo->out_color_space=JCS_EXT_BGRX; break;
+ this->dinfo.out_color_space=JCS_EXT_BGRX; break;
case TJPF_XRGB:
- dinfo->out_color_space=JCS_EXT_XRGB; break;
+ this->dinfo.out_color_space=JCS_EXT_XRGB; break;
case TJPF_XBGR:
- dinfo->out_color_space=JCS_EXT_XBGR; break;
+ this->dinfo.out_color_space=JCS_EXT_XBGR; break;
#if JCS_ALPHA_EXTENSIONS==1
case TJPF_RGBA:
- dinfo->out_color_space=JCS_EXT_RGBA; break;
+ this->dinfo.out_color_space=JCS_EXT_RGBA; break;
case TJPF_BGRA:
- dinfo->out_color_space=JCS_EXT_BGRA; break;
+ this->dinfo.out_color_space=JCS_EXT_BGRA; break;
case TJPF_ARGB:
- dinfo->out_color_space=JCS_EXT_ARGB; break;
+ this->dinfo.out_color_space=JCS_EXT_ARGB; break;
case TJPF_ABGR:
- dinfo->out_color_space=JCS_EXT_ABGR; break;
+ this->dinfo.out_color_space=JCS_EXT_ABGR; break;
#endif
#else
case TJPF_RGB:
@@ -315,15 +327,15 @@
case TJPF_BGRA:
case TJPF_ARGB:
case TJPF_ABGR:
- dinfo->out_color_space=JCS_RGB; break;
+ this->dinfo.out_color_space=JCS_RGB; break;
#endif
case TJPF_CMYK:
- dinfo->out_color_space=JCS_CMYK; break;
+ this->dinfo.out_color_space=JCS_CMYK; break;
default:
_throw("Unsupported pixel format");
}
- if(flags&TJFLAG_FASTDCT) dinfo->dct_method=JDCT_FASTEST;
+ if(flags&TJFLAG_FASTDCT) this->dinfo.dct_method=JDCT_FASTEST;
bailout:
return retval;
@@ -542,12 +554,32 @@
/* General API functions */
+DLLEXPORT char* DLLCALL tjGetErrorStr2(tjhandle handle)
+{
+ tjinstance *this=(tjinstance *)handle;
+ if(this && this->isInstanceError)
+ {
+ this->isInstanceError=FALSE;
+ return this->errStr;
+ }
+ else return errStr;
+}
+
+
DLLEXPORT char* DLLCALL tjGetErrorStr(void)
{
return errStr;
}
+DLLEXPORT int DLLCALL tjGetErrorCode(tjhandle handle)
+{
+ tjinstance *this=(tjinstance *)handle;
+ if(this && this->jerr.warning) return TJERR_WARNING;
+ else return TJERR_FATAL;
+}
+
+
DLLEXPORT int DLLCALL tjDestroy(tjhandle handle)
{
getinstance(handle);
@@ -615,6 +647,7 @@
return NULL;
}
MEMZERO(this, sizeof(tjinstance));
+ snprintf(this->errStr, JMSG_LENGTH_MAX, "No error");
return _tjInitCompress(this);
}
@@ -624,7 +657,7 @@
{
unsigned long retval=0; int mcuw, mcuh, chromasf;
if(width<1 || height<1 || jpegSubsamp<0 || jpegSubsamp>=NUMSUBOPT)
- _throw("tjBufSize(): Invalid argument");
+ _throwg("tjBufSize(): Invalid argument");
/* This allows for rare corner cases in which a JPEG image can actually be
larger than the uncompressed input (we wouldn't mention it if it hadn't
@@ -642,7 +675,7 @@
{
unsigned long retval=0;
if(width<1 || height<1)
- _throw("TJBUFSIZE(): Invalid argument");
+ _throwg("TJBUFSIZE(): Invalid argument");
/* This allows for rare corner cases in which a JPEG image can actually be
larger than the uncompressed input (we wouldn't mention it if it hadn't
@@ -660,7 +693,7 @@
int retval=0, nc, i;
if(subsamp<0 || subsamp>=NUMSUBOPT)
- _throw("tjBufSizeYUV2(): Invalid argument");
+ _throwg("tjBufSizeYUV2(): Invalid argument");
nc=(subsamp==TJSAMP_GRAY? 1:3);
for(i=0; i<nc; i++)
@@ -694,10 +727,10 @@
int pw, nc, retval=0;
if(width<1 || subsamp<0 || subsamp>=TJ_NUMSAMP)
- _throw("tjPlaneWidth(): Invalid argument");
+ _throwg("tjPlaneWidth(): Invalid argument");
nc=(subsamp==TJSAMP_GRAY? 1:3);
if(componentID<0 || componentID>=nc)
- _throw("tjPlaneWidth(): Invalid argument");
+ _throwg("tjPlaneWidth(): Invalid argument");
pw=PAD(width, tjMCUWidth[subsamp]/8);
if(componentID==0)
@@ -715,10 +748,10 @@
int ph, nc, retval=0;
if(height<1 || subsamp<0 || subsamp>=TJ_NUMSAMP)
- _throw("tjPlaneHeight(): Invalid argument");
+ _throwg("tjPlaneHeight(): Invalid argument");
nc=(subsamp==TJSAMP_GRAY? 1:3);
if(componentID<0 || componentID>=nc)
- _throw("tjPlaneHeight(): Invalid argument");
+ _throwg("tjPlaneHeight(): Invalid argument");
ph=PAD(height, tjMCUHeight[subsamp]/8);
if(componentID==0)
@@ -738,7 +771,7 @@
int pw, ph;
if(width<1 || height<1 || subsamp<0 || subsamp>=NUMSUBOPT)
- _throw("tjPlaneSizeYUV(): Invalid argument");
+ _throwg("tjPlaneSizeYUV(): Invalid argument");
pw=tjPlaneWidth(componentID, width, subsamp);
ph=tjPlaneHeight(componentID, height, subsamp);
@@ -764,6 +797,7 @@
#endif
getcinstance(handle)
+ this->jerr.stopOnWarning=(flags & TJFLAG_STOPONWARNING) ? TRUE : FALSE;
if((this->init&COMPRESS)==0)
_throw("tjCompress2(): Instance has not been initialized for compression");
@@ -830,6 +864,7 @@
#endif
if(row_pointer) free(row_pointer);
if(this->jerr.warning) retval=-1;
+ this->jerr.stopOnWarning=FALSE;
return retval;
}
@@ -872,6 +907,7 @@
#endif
getcinstance(handle);
+ this->jerr.stopOnWarning=(flags & TJFLAG_STOPONWARNING) ? TRUE : FALSE;
for(i=0; i<MAX_COMPONENTS; i++)
{
@@ -951,29 +987,29 @@
compptr=&cinfo->comp_info[i];
_tmpbuf[i]=(JSAMPLE *)malloc(
PAD((compptr->width_in_blocks*cinfo->max_h_samp_factor*DCTSIZE)
- /compptr->h_samp_factor, 16) * cinfo->max_v_samp_factor + 16);
+ /compptr->h_samp_factor, 32) * cinfo->max_v_samp_factor + 32);
if(!_tmpbuf[i]) _throw("tjEncodeYUVPlanes(): Memory allocation failure");
tmpbuf[i]=(JSAMPROW *)malloc(sizeof(JSAMPROW)*cinfo->max_v_samp_factor);
if(!tmpbuf[i]) _throw("tjEncodeYUVPlanes(): Memory allocation failure");
for(row=0; row<cinfo->max_v_samp_factor; row++)
{
unsigned char *_tmpbuf_aligned=
- (unsigned char *)PAD((size_t)_tmpbuf[i], 16);
+ (unsigned char *)PAD((size_t)_tmpbuf[i], 32);
tmpbuf[i][row]=&_tmpbuf_aligned[
PAD((compptr->width_in_blocks*cinfo->max_h_samp_factor*DCTSIZE)
- /compptr->h_samp_factor, 16) * row];
+ /compptr->h_samp_factor, 32) * row];
}
- _tmpbuf2[i]=(JSAMPLE *)malloc(PAD(compptr->width_in_blocks*DCTSIZE, 16)
- * compptr->v_samp_factor + 16);
+ _tmpbuf2[i]=(JSAMPLE *)malloc(PAD(compptr->width_in_blocks*DCTSIZE, 32)
+ * compptr->v_samp_factor + 32);
if(!_tmpbuf2[i]) _throw("tjEncodeYUVPlanes(): Memory allocation failure");
tmpbuf2[i]=(JSAMPROW *)malloc(sizeof(JSAMPROW)*compptr->v_samp_factor);
if(!tmpbuf2[i]) _throw("tjEncodeYUVPlanes(): Memory allocation failure");
for(row=0; row<compptr->v_samp_factor; row++)
{
unsigned char *_tmpbuf2_aligned=
- (unsigned char *)PAD((size_t)_tmpbuf2[i], 16);
+ (unsigned char *)PAD((size_t)_tmpbuf2[i], 32);
tmpbuf2[i][row]=&_tmpbuf2_aligned[
- PAD(compptr->width_in_blocks*DCTSIZE, 16) * row];
+ PAD(compptr->width_in_blocks*DCTSIZE, 32) * row];
}
pw[i]=pw0*compptr->h_samp_factor/cinfo->max_h_samp_factor;
ph[i]=ph0*compptr->v_samp_factor/cinfo->max_v_samp_factor;
@@ -1022,6 +1058,7 @@
if(outbuf[i]!=NULL) free(outbuf[i]);
}
if(this->jerr.warning) retval=-1;
+ this->jerr.stopOnWarning=FALSE;
return retval;
}
@@ -1032,6 +1069,8 @@
unsigned char *dstPlanes[3];
int pw0, ph0, strides[3], retval=-1;
+ getcinstance(handle);
+
if(width<=0 || height<=0 || dstBuf==NULL || pad<0 || !isPow2(pad)
|| subsamp<0 || subsamp>=NUMSUBOPT)
_throw("tjEncodeYUV3(): Invalid argument");
@@ -1089,6 +1128,7 @@
JSAMPLE *_tmpbuf=NULL, *ptr; JSAMPROW *tmpbuf[MAX_COMPONENTS];
getcinstance(handle)
+ this->jerr.stopOnWarning=(flags & TJFLAG_STOPONWARNING) ? TRUE : FALSE;
for(i=0; i<MAX_COMPONENTS; i++)
{
@@ -1214,6 +1254,7 @@
}
if(_tmpbuf) free(_tmpbuf);
if(this->jerr.warning) retval=-1;
+ this->jerr.stopOnWarning=FALSE;
return retval;
}
@@ -1224,6 +1265,8 @@
const unsigned char *srcPlanes[3];
int pw0, ph0, strides[3], retval=-1;
+ getcinstance(handle);
+
if(srcBuf==NULL || width<=0 || pad<1 || height<=0 || subsamp<0
|| subsamp>=NUMSUBOPT)
_throw("tjCompressFromYUV(): Invalid argument");
@@ -1292,6 +1335,7 @@
return NULL;
}
MEMZERO(this, sizeof(tjinstance));
+ snprintf(this->errStr, JMSG_LENGTH_MAX, "No error");
return _tjInitDecompress(this);
}
@@ -1390,6 +1434,7 @@
#endif
getdinstance(handle);
+ this->jerr.stopOnWarning=(flags & TJFLAG_STOPONWARNING) ? TRUE : FALSE;
if((this->init&DECOMPRESS)==0)
_throw("tjDecompress2(): Instance has not been initialized for decompression");
@@ -1410,7 +1455,7 @@
jpeg_mem_src_tj(dinfo, jpegBuf, jpegSize);
jpeg_read_header(dinfo, TRUE);
- if(setDecompDefaults(dinfo, pixelFormat, flags)==-1)
+ if(setDecompDefaults(this, pixelFormat, flags)==-1)
{
retval=-1; goto bailout;
}
@@ -1483,6 +1528,7 @@
#endif
if(row_pointer) free(row_pointer);
if(this->jerr.warning) retval=-1;
+ this->jerr.stopOnWarning=FALSE;
return retval;
}
@@ -1570,6 +1616,7 @@
void (*old_reset_marker_reader)(j_decompress_ptr);
getdinstance(handle);
+ this->jerr.stopOnWarning=(flags & TJFLAG_STOPONWARNING) ? TRUE : FALSE;
for(i=0; i<MAX_COMPONENTS; i++)
{
@@ -1616,7 +1663,7 @@
dinfo->marker->read_markers=old_read_markers;
dinfo->marker->reset_marker_reader=old_reset_marker_reader;
- if(setDecompDefaults(dinfo, pixelFormat, flags)==-1)
+ if(setDecompDefaults(this, pixelFormat, flags)==-1)
{
retval=-1; goto bailout;
}
@@ -1657,17 +1704,17 @@
for(i=0; i<dinfo->num_components; i++)
{
compptr=&dinfo->comp_info[i];
- _tmpbuf[i]=(JSAMPLE *)malloc(PAD(compptr->width_in_blocks*DCTSIZE, 16)
- * compptr->v_samp_factor + 16);
+ _tmpbuf[i]=(JSAMPLE *)malloc(PAD(compptr->width_in_blocks*DCTSIZE, 32)
+ * compptr->v_samp_factor + 32);
if(!_tmpbuf[i]) _throw("tjDecodeYUVPlanes(): Memory allocation failure");
tmpbuf[i]=(JSAMPROW *)malloc(sizeof(JSAMPROW)*compptr->v_samp_factor);
if(!tmpbuf[i]) _throw("tjDecodeYUVPlanes(): Memory allocation failure");
for(row=0; row<compptr->v_samp_factor; row++)
{
unsigned char *_tmpbuf_aligned=
- (unsigned char *)PAD((size_t)_tmpbuf[i], 16);
+ (unsigned char *)PAD((size_t)_tmpbuf[i], 32);
tmpbuf[i][row]=&_tmpbuf_aligned[
- PAD(compptr->width_in_blocks*DCTSIZE, 16) * row];
+ PAD(compptr->width_in_blocks*DCTSIZE, 32) * row];
}
pw[i]=pw0*compptr->h_samp_factor/dinfo->max_h_samp_factor;
ph[i]=ph0*compptr->v_samp_factor/dinfo->max_v_samp_factor;
@@ -1718,6 +1765,7 @@
if(inbuf[i]!=NULL) free(inbuf[i]);
}
if(this->jerr.warning) retval=-1;
+ this->jerr.stopOnWarning=FALSE;
return retval;
}
@@ -1728,6 +1776,8 @@
const unsigned char *srcPlanes[3];
int pw0, ph0, strides[3], retval=-1;
+ getdinstance(handle);
+
if(srcBuf==NULL || pad<0 || !isPow2(pad) || subsamp<0 || subsamp>=NUMSUBOPT
|| width<=0 || height<=0)
_throw("tjDecodeYUV(): Invalid argument");
@@ -1769,6 +1819,7 @@
int dctsize;
getdinstance(handle);
+ this->jerr.stopOnWarning=(flags & TJFLAG_STOPONWARNING) ? TRUE : FALSE;
for(i=0; i<MAX_COMPONENTS; i++)
{
@@ -1935,6 +1986,7 @@
}
if(_tmpbuf) free(_tmpbuf);
if(this->jerr.warning) retval=-1;
+ this->jerr.stopOnWarning=FALSE;
return retval;
}
@@ -1947,6 +1999,7 @@
int i, jpegwidth, jpegheight, scaledw, scaledh;
getdinstance(handle);
+ this->jerr.stopOnWarning=(flags & TJFLAG_STOPONWARNING) ? TRUE : FALSE;
if(jpegBuf==NULL || jpegSize<=0 || dstBuf==NULL || width<0 || pad<1
|| !isPow2(pad) || height<0)
@@ -2001,6 +2054,7 @@
strides, height, flags);
bailout:
+ this->jerr.stopOnWarning=FALSE;
return retval;
}
@@ -2025,6 +2079,7 @@
return NULL;
}
MEMZERO(this, sizeof(tjinstance));
+ snprintf(this->errStr, JMSG_LENGTH_MAX, "No error");
handle=_tjInitCompress(this);
if(!handle) return NULL;
handle=_tjInitDecompress(this);
@@ -2038,9 +2093,10 @@
{
jpeg_transform_info *xinfo=NULL;
jvirt_barray_ptr *srccoefs, *dstcoefs;
- int retval=0, i, jpegSubsamp;
+ int retval=0, i, jpegSubsamp, saveMarkers=0;
getinstance(handle);
+ this->jerr.stopOnWarning=(flags & TJFLAG_STOPONWARNING) ? TRUE : FALSE;
if((this->init&COMPRESS)==0 || (this->init&DECOMPRESS)==0)
_throw("tjTransform(): Instance has not been initialized for transformation");
@@ -2091,9 +2147,10 @@
}
else xinfo[i].crop_height=JCROP_UNSET;
}
+ if(!(t[i].options&TJXOPT_COPYNONE)) saveMarkers=1;
}
- jcopy_markers_setup(dinfo, JCOPYOPT_ALL);
+ jcopy_markers_setup(dinfo, saveMarkers ? JCOPYOPT_ALL:JCOPYOPT_NONE);
jpeg_read_header(dinfo, TRUE);
jpegSubsamp=getSubsamp(dinfo);
if(jpegSubsamp<0)
@@ -2140,10 +2197,13 @@
jpeg_copy_critical_parameters(dinfo, cinfo);
dstcoefs=jtransform_adjust_parameters(dinfo, cinfo, srccoefs,
&xinfo[i]);
+ if(flags&TJFLAG_PROGRESSIVE || t[i].options&TJXOPT_PROGRESSIVE)
+ jpeg_simple_progression(cinfo);
if(!(t[i].options&TJXOPT_NOOUTPUT))
{
jpeg_write_coefficients(cinfo, dstcoefs);
- jcopy_markers_execute(dinfo, cinfo, JCOPYOPT_ALL);
+ jcopy_markers_execute(dinfo, cinfo,
+ t[i].options&TJXOPT_COPYNONE ? JCOPYOPT_NONE:JCOPYOPT_ALL);
}
else jinit_c_master_control(cinfo, TRUE);
jtransform_execute_transformation(dinfo, cinfo, srccoefs,
@@ -2183,5 +2243,6 @@
if(dinfo->global_state>DSTATE_START) jpeg_abort_decompress(dinfo);
if(xinfo) free(xinfo);
if(this->jerr.warning) retval=-1;
+ this->jerr.stopOnWarning=FALSE;
return retval;
}
diff --git a/turbojpeg.h b/turbojpeg.h
index f60caf8..c4e48cc 100644
--- a/turbojpeg.h
+++ b/turbojpeg.h
@@ -383,6 +383,42 @@
* when decompressing, because this has been shown to have a larger effect.
*/
#define TJFLAG_ACCURATEDCT 4096
+/**
+ * Immediately discontinue the current compression/decompression/transform
+ * operation if the underlying codec throws a warning (non-fatal error). The
+ * default behavior is to allow the operation to complete unless a fatal error
+ * is encountered.
+ */
+#define TJFLAG_STOPONWARNING 8192
+/**
+ * Use progressive entropy coding in JPEG images generated by the compression
+ * and transform functions. Progressive entropy coding will generally improve
+ * compression relative to baseline entropy coding (the default), but it will
+ * reduce compression and decompression performance considerably.
+ */
+#define TJFLAG_PROGRESSIVE 16384
+
+
+/**
+ * The number of error codes
+ */
+#define TJ_NUMERR 2
+
+/**
+ * Error codes
+ */
+enum TJERR
+{
+ /**
+ * The error was non-fatal and recoverable, but the image may still be
+ * corrupt.
+ */
+ TJERR_WARNING=0,
+ /**
+ * The error was fatal and non-recoverable.
+ */
+ TJERR_FATAL
+};
/**
@@ -452,29 +488,43 @@
* that cannot be transformed will be left in place, which will create
* odd-looking strips on the right or bottom edge of the image.
*/
-#define TJXOPT_PERFECT 1
+#define TJXOPT_PERFECT 1
/**
* This option will cause #tjTransform() to discard any partial MCU blocks that
* cannot be transformed.
*/
-#define TJXOPT_TRIM 2
+#define TJXOPT_TRIM 2
/**
* This option will enable lossless cropping. See #tjTransform() for more
* information.
*/
-#define TJXOPT_CROP 4
+#define TJXOPT_CROP 4
/**
* This option will discard the color data in the input image and produce
* a grayscale output image.
*/
-#define TJXOPT_GRAY 8
+#define TJXOPT_GRAY 8
/**
* This option will prevent #tjTransform() from outputting a JPEG image for
* this particular transform (this can be used in conjunction with a custom
* filter to capture the transformed DCT coefficients without transcoding
* them.)
*/
-#define TJXOPT_NOOUTPUT 16
+#define TJXOPT_NOOUTPUT 16
+/**
+ * This option will enable progressive entropy coding in the output image
+ * generated by this particular transform. Progressive entropy coding will
+ * generally improve compression relative to baseline entropy coding (the
+ * default), but it will reduce compression and decompression performance
+ * considerably.
+ */
+#define TJXOPT_PROGRESSIVE 32
+/**
+ * This option will prevent #tjTransform() from copying any extra markers
+ * (including EXIF and ICC profile data) from the source image to the output
+ * image.
+ */
+#define TJXOPT_COPYNONE 64
/**
@@ -608,7 +658,7 @@
* Create a TurboJPEG compressor instance.
*
* @return a handle to the newly-created instance, or NULL if an error
- * occurred (see #tjGetErrorStr().)
+ * occurred (see #tjGetErrorStr2().)
*/
DLLEXPORT tjhandle DLLCALL tjInitCompress(void);
@@ -670,7 +720,8 @@
* @param flags the bitwise OR of one or more of the @ref TJFLAG_BOTTOMUP
* "flags"
*
- * @return 0 if successful, or -1 if an error occurred (see #tjGetErrorStr().)
+ * @return 0 if successful, or -1 if an error occurred (see #tjGetErrorStr2()
+ * and #tjGetErrorCode().)
*/
DLLEXPORT int DLLCALL tjCompress2(tjhandle handle, const unsigned char *srcBuf,
int width, int pitch, int height, int pixelFormat, unsigned char **jpegBuf,
@@ -734,7 +785,8 @@
* @param flags the bitwise OR of one or more of the @ref TJFLAG_BOTTOMUP
* "flags"
*
- * @return 0 if successful, or -1 if an error occurred (see #tjGetErrorStr().)
+ * @return 0 if successful, or -1 if an error occurred (see #tjGetErrorStr2()
+ * and #tjGetErrorCode().)
*/
DLLEXPORT int DLLCALL tjCompressFromYUV(tjhandle handle,
const unsigned char *srcBuf, int width, int pad, int height, int subsamp,
@@ -804,7 +856,8 @@
* @param flags the bitwise OR of one or more of the @ref TJFLAG_BOTTOMUP
* "flags"
*
- * @return 0 if successful, or -1 if an error occurred (see #tjGetErrorStr().)
+ * @return 0 if successful, or -1 if an error occurred (see #tjGetErrorStr2()
+ * and #tjGetErrorCode().)
*/
DLLEXPORT int DLLCALL tjCompressFromYUVPlanes(tjhandle handle,
const unsigned char **srcPlanes, int width, const int *strides, int height,
@@ -964,7 +1017,8 @@
* @param flags the bitwise OR of one or more of the @ref TJFLAG_BOTTOMUP
* "flags"
*
- * @return 0 if successful, or -1 if an error occurred (see #tjGetErrorStr().)
+ * @return 0 if successful, or -1 if an error occurred (see #tjGetErrorStr2()
+ * and #tjGetErrorCode().)
*/
DLLEXPORT int DLLCALL tjEncodeYUV3(tjhandle handle,
const unsigned char *srcBuf, int width, int pitch, int height,
@@ -1022,7 +1076,8 @@
* @param flags the bitwise OR of one or more of the @ref TJFLAG_BOTTOMUP
* "flags"
*
- * @return 0 if successful, or -1 if an error occurred (see #tjGetErrorStr().)
+ * @return 0 if successful, or -1 if an error occurred (see #tjGetErrorStr2()
+ * and #tjGetErrorCode().)
*/
DLLEXPORT int DLLCALL tjEncodeYUVPlanes(tjhandle handle,
const unsigned char *srcBuf, int width, int pitch, int height,
@@ -1034,7 +1089,7 @@
* Create a TurboJPEG decompressor instance.
*
* @return a handle to the newly-created instance, or NULL if an error
- * occurred (see #tjGetErrorStr().)
+ * occurred (see #tjGetErrorStr2().)
*/
DLLEXPORT tjhandle DLLCALL tjInitDecompress(void);
@@ -1062,7 +1117,8 @@
* of the JPEG colorspace constants, indicating the colorspace of the JPEG
* image (see @ref TJCS "JPEG colorspaces".)
*
- * @return 0 if successful, or -1 if an error occurred (see #tjGetErrorStr().)
+ * @return 0 if successful, or -1 if an error occurred (see #tjGetErrorStr2()
+ * and #tjGetErrorCode().)
*/
DLLEXPORT int DLLCALL tjDecompressHeader3(tjhandle handle,
const unsigned char *jpegBuf, unsigned long jpegSize, int *width,
@@ -1077,7 +1133,7 @@
* the number of elements in the list
*
* @return a pointer to a list of fractional scaling factors, or NULL if an
- * error is encountered (see #tjGetErrorStr().)
+ * error is encountered (see #tjGetErrorStr2().)
*/
DLLEXPORT tjscalingfactor* DLLCALL tjGetScalingFactors(int *numscalingfactors);
@@ -1129,7 +1185,8 @@
* @param flags the bitwise OR of one or more of the @ref TJFLAG_BOTTOMUP
* "flags"
*
- * @return 0 if successful, or -1 if an error occurred (see #tjGetErrorStr().)
+ * @return 0 if successful, or -1 if an error occurred (see #tjGetErrorStr2()
+ * and #tjGetErrorCode().)
*/
DLLEXPORT int DLLCALL tjDecompress2(tjhandle handle,
const unsigned char *jpegBuf, unsigned long jpegSize, unsigned char *dstBuf,
@@ -1179,7 +1236,8 @@
* @param flags the bitwise OR of one or more of the @ref TJFLAG_BOTTOMUP
* "flags"
*
- * @return 0 if successful, or -1 if an error occurred (see #tjGetErrorStr().)
+ * @return 0 if successful, or -1 if an error occurred (see #tjGetErrorStr2()
+ * and #tjGetErrorCode().)
*/
DLLEXPORT int DLLCALL tjDecompressToYUV2(tjhandle handle,
const unsigned char *jpegBuf, unsigned long jpegSize, unsigned char *dstBuf,
@@ -1235,7 +1293,8 @@
* @param flags the bitwise OR of one or more of the @ref TJFLAG_BOTTOMUP
* "flags"
*
- * @return 0 if successful, or -1 if an error occurred (see #tjGetErrorStr().)
+ * @return 0 if successful, or -1 if an error occurred (see #tjGetErrorStr2()
+ * and #tjGetErrorCode().)
*/
DLLEXPORT int DLLCALL tjDecompressToYUVPlanes(tjhandle handle,
const unsigned char *jpegBuf, unsigned long jpegSize,
@@ -1287,7 +1346,8 @@
* @param flags the bitwise OR of one or more of the @ref TJFLAG_BOTTOMUP
* "flags"
*
- * @return 0 if successful, or -1 if an error occurred (see #tjGetErrorStr().)
+ * @return 0 if successful, or -1 if an error occurred (see #tjGetErrorStr2()
+ * and #tjGetErrorCode().)
*/
DLLEXPORT int DLLCALL tjDecodeYUV(tjhandle handle, const unsigned char *srcBuf,
int pad, int subsamp, unsigned char *dstBuf, int width, int pitch,
@@ -1344,7 +1404,8 @@
* @param flags the bitwise OR of one or more of the @ref TJFLAG_BOTTOMUP
* "flags"
*
- * @return 0 if successful, or -1 if an error occurred (see #tjGetErrorStr().)
+ * @return 0 if successful, or -1 if an error occurred (see #tjGetErrorStr2()
+ * and #tjGetErrorCode().)
*/
DLLEXPORT int DLLCALL tjDecodeYUVPlanes(tjhandle handle,
const unsigned char **srcPlanes, const int *strides, int subsamp,
@@ -1356,7 +1417,7 @@
* Create a new TurboJPEG transformer instance.
*
* @return a handle to the newly-created instance, or NULL if an error
- * occurred (see #tjGetErrorStr().)
+ * occurred (see #tjGetErrorStr2().)
*/
DLLEXPORT tjhandle DLLCALL tjInitTransform(void);
@@ -1418,7 +1479,8 @@
* @param flags the bitwise OR of one or more of the @ref TJFLAG_BOTTOMUP
* "flags"
*
- * @return 0 if successful, or -1 if an error occurred (see #tjGetErrorStr().)
+ * @return 0 if successful, or -1 if an error occurred (see #tjGetErrorStr2()
+ * and #tjGetErrorCode().)
*/
DLLEXPORT int DLLCALL tjTransform(tjhandle handle,
const unsigned char *jpegBuf, unsigned long jpegSize, int n,
@@ -1432,7 +1494,7 @@
* @param handle a handle to a TurboJPEG compressor, decompressor or
* transformer instance
*
- * @return 0 if successful, or -1 if an error occurred (see #tjGetErrorStr().)
+ * @return 0 if successful, or -1 if an error occurred (see #tjGetErrorStr2().)
*/
DLLEXPORT int DLLCALL tjDestroy(tjhandle handle);
@@ -1469,9 +1531,27 @@
/**
* Returns a descriptive error message explaining why the last command failed.
*
+ * @param handle a handle to a TurboJPEG compressor, decompressor, or
+ * transformer instance, or NULL if the error was generated by a global
+ * function (but note that retrieving the error message for a global function
+ * is not thread-safe.)
+ *
* @return a descriptive error message explaining why the last command failed.
*/
-DLLEXPORT char* DLLCALL tjGetErrorStr(void);
+DLLEXPORT char* DLLCALL tjGetErrorStr2(tjhandle handle);
+
+
+/**
+ * Returns a code indicating the severity of the last error. See
+ * @ref TJERR "Error codes".
+ *
+ * @param handle a handle to a TurboJPEG compressor, decompressor or
+ * transformer instance
+ *
+ * @return a code indicating the severity of the last error. See
+ * @ref TJERR "Error codes".
+ */
+DLLEXPORT int DLLCALL tjGetErrorCode(tjhandle handle);
/* Deprecated functions and macros */
@@ -1534,6 +1614,8 @@
unsigned char *jpegBuf, unsigned long jpegSize, unsigned char *dstBuf,
int flags);
+DLLEXPORT char* DLLCALL tjGetErrorStr(void);
+
/**
* @}
diff --git a/win/jconfig.h.in b/win/jconfig.h.in
index 9d35121..e3c3d1c 100644
--- a/win/jconfig.h.in
+++ b/win/jconfig.h.in
@@ -1,35 +1,24 @@
-/* jconfig.vc --- jconfig.h for Microsoft Visual C++ on Windows 95 or NT. */
-/* see jconfig.txt for explanations */
-
#define JPEG_LIB_VERSION @JPEG_LIB_VERSION@
#define LIBJPEG_TURBO_VERSION @VERSION@
#define LIBJPEG_TURBO_VERSION_NUMBER @LIBJPEG_TURBO_VERSION_NUMBER@
+
#cmakedefine C_ARITH_CODING_SUPPORTED
#cmakedefine D_ARITH_CODING_SUPPORTED
#cmakedefine MEM_SRCDST_SUPPORTED
-
-/*
- * Define BITS_IN_JSAMPLE as either
- * 8 for 8-bit sample values (the usual setting)
- * 12 for 12-bit sample values
- * Only 8 and 12 are legal data precisions for lossy JPEG according to the
- * JPEG standard, and the IJG code does not support anything else!
- * We do not support run-time selection of data precision, sorry.
- */
+#cmakedefine WITH_SIMD
#define BITS_IN_JSAMPLE @BITS_IN_JSAMPLE@ /* use 8 or 12 */
-#define HAVE_UNSIGNED_CHAR
-#define HAVE_UNSIGNED_SHORT
-/* #define void char */
-/* #define const */
-#undef __CHAR_UNSIGNED__
#define HAVE_STDDEF_H
#define HAVE_STDLIB_H
-#undef NEED_BSD_STRINGS
#undef NEED_SYS_TYPES_H
-#undef NEED_FAR_POINTERS /* we presume a 32-bit flat memory model */
+#undef NEED_BSD_STRINGS
+
+#define HAVE_UNSIGNED_CHAR
+#define HAVE_UNSIGNED_SHORT
#undef INCOMPLETE_TYPES_BROKEN
+#undef RIGHT_SHIFT_IS_UNSIGNED
+#undef __CHAR_UNSIGNED__
/* Define "boolean" as unsigned char, not int, per Windows custom */
#ifndef __RPCNDR_H__ /* don't conflict if rpcndr.h already read */
@@ -43,9 +32,3 @@
typedef signed int INT32;
#endif
#define XMD_H /* prevent jmorecfg.h from redefining it */
-
-#ifdef JPEG_INTERNALS
-
-#undef RIGHT_SHIFT_IS_UNSIGNED
-
-#endif /* JPEG_INTERNALS */
diff --git a/win/jconfigint.h.in b/win/jconfigint.h.in
deleted file mode 100644
index 2131bf5..0000000
--- a/win/jconfigint.h.in
+++ /dev/null
@@ -1,13 +0,0 @@
-#define VERSION "@VERSION@"
-#define BUILD "@BUILD@"
-#define PACKAGE_NAME "@CMAKE_PROJECT_NAME@"
-
-#ifndef INLINE
-#if defined(__GNUC__)
-#define INLINE inline __attribute__((always_inline))
-#elif defined(_MSC_VER)
-#define INLINE __forceinline
-#else
-#define INLINE
-#endif
-#endif
diff --git a/win/jpeg62-memsrcdst.def b/win/jpeg62-memsrcdst.def
index 6499316..6f44d28 100755
--- a/win/jpeg62-memsrcdst.def
+++ b/win/jpeg62-memsrcdst.def
@@ -104,3 +104,5 @@
jpeg_mem_src @ 103 ;
jpeg_skip_scanlines @ 104 ;
jpeg_crop_scanline @ 105 ;
+ jpeg_read_icc_profile @ 106 ;
+ jpeg_write_icc_profile @ 107 ;
diff --git a/win/jpeg62.def b/win/jpeg62.def
index 9f30b1a..8f70df6 100755
--- a/win/jpeg62.def
+++ b/win/jpeg62.def
@@ -102,3 +102,5 @@
jzero_far @ 101 ;
jpeg_skip_scanlines @ 102 ;
jpeg_crop_scanline @ 103 ;
+ jpeg_read_icc_profile @ 104 ;
+ jpeg_write_icc_profile @ 105 ;
diff --git a/win/jpeg7-memsrcdst.def b/win/jpeg7-memsrcdst.def
index 37a4777..521d957 100644
--- a/win/jpeg7-memsrcdst.def
+++ b/win/jpeg7-memsrcdst.def
@@ -106,3 +106,5 @@
jpeg_mem_src @ 105 ;
jpeg_skip_scanlines @ 106 ;
jpeg_crop_scanline @ 107 ;
+ jpeg_read_icc_profile @ 108 ;
+ jpeg_write_icc_profile @ 109 ;
diff --git a/win/jpeg7.def b/win/jpeg7.def
index 92463c5..2415bb2 100644
--- a/win/jpeg7.def
+++ b/win/jpeg7.def
@@ -104,3 +104,5 @@
jzero_far @ 103 ;
jpeg_skip_scanlines @ 104 ;
jpeg_crop_scanline @ 105 ;
+ jpeg_read_icc_profile @ 106 ;
+ jpeg_write_icc_profile @ 107 ;
diff --git a/win/jpeg8.def b/win/jpeg8.def
index 19246ac..b8d0455 100644
--- a/win/jpeg8.def
+++ b/win/jpeg8.def
@@ -107,3 +107,5 @@
jzero_far @ 106 ;
jpeg_skip_scanlines @ 107 ;
jpeg_crop_scanline @ 108 ;
+ jpeg_read_icc_profile @ 109 ;
+ jpeg_write_icc_profile @ 110 ;
diff --git a/win/jsimdcfg.inc b/win/jsimdcfg.inc
index 9d4aede..667024a 100755
--- a/win/jsimdcfg.inc
+++ b/win/jsimdcfg.inc
@@ -90,5 +90,4 @@
%define JSIMD_3DNOW 0x02
%define JSIMD_SSE 0x04
%define JSIMD_SSE2 0x08
-; Short forms of external names for systems with brain-damaged linkers.
-;
+%define JSIMD_AVX2 0x80