Merge commit 'b10a68c3' into manualmerge
Conflicts:
cpu_ref/rsCpuIntrinsicColorMatrix.cpp
Change-Id: Ibc2f1514f8858d99f08380f698bc9ae533c69212
diff --git a/Android.mk b/Android.mk
index 4c02056..bdde8a1 100644
--- a/Android.mk
+++ b/Android.mk
@@ -40,6 +40,7 @@
LOCAL_SHARED_LIBRARIES += libbcc libbcinfo libLLVM libui libgui libsync
LOCAL_C_INCLUDES += frameworks/compile/libbcc/include
+LOCAL_C_INCLUDES += frameworks/rs/cpu_ref/linkloader/include
LOCAL_CFLAGS += $(rs_base_CFLAGS)
@@ -117,12 +118,14 @@
rsAnimation.cpp \
rsComponent.cpp \
rsContext.cpp \
+ rsCppUtils.cpp \
rsDevice.cpp \
rsElement.cpp \
rsFBOCache.cpp \
rsFifoSocket.cpp \
rsFileA3D.cpp \
rsFont.cpp \
+ rsGrallocConsumer.cpp \
rsObjectBase.cpp \
rsMatrix2x2.cpp \
rsMatrix3x3.cpp \
@@ -149,8 +152,7 @@
LOCAL_SHARED_LIBRARIES += liblog libcutils libutils libEGL libGLESv1_CM libGLESv2 libbcc
LOCAL_SHARED_LIBRARIES += libui libbcinfo libLLVM libgui libsync libdl
-
-LOCAL_STATIC_LIBRARIES := libft2
+LOCAL_SHARED_LIBRARIES += libft2 libpng libz
LOCAL_C_INCLUDES += external/freetype/include
LOCAL_C_INCLUDES += frameworks/compile/libbcc/include
@@ -253,17 +255,17 @@
#-----------------------------------------------------------------------------
rsloader_SRC_FILES := \
- driver/linkloader/android/librsloader.cpp \
- driver/linkloader/lib/ELFHeader.cpp \
- driver/linkloader/lib/ELFSymbol.cpp \
- driver/linkloader/lib/ELFSectionHeader.cpp \
- driver/linkloader/lib/ELFTypes.cpp \
- driver/linkloader/lib/GOT.cpp \
- driver/linkloader/lib/MemChunk.cpp \
- driver/linkloader/lib/StubLayout.cpp \
- driver/linkloader/utils/helper.cpp \
- driver/linkloader/utils/raw_ostream.cpp \
- driver/linkloader/utils/rsl_assert.cpp
+ cpu_ref/linkloader/android/librsloader.cpp \
+ cpu_ref/linkloader/lib/ELFHeader.cpp \
+ cpu_ref/linkloader/lib/ELFSymbol.cpp \
+ cpu_ref/linkloader/lib/ELFSectionHeader.cpp \
+ cpu_ref/linkloader/lib/ELFTypes.cpp \
+ cpu_ref/linkloader/lib/GOT.cpp \
+ cpu_ref/linkloader/lib/MemChunk.cpp \
+ cpu_ref/linkloader/lib/StubLayout.cpp \
+ cpu_ref/linkloader/utils/helper.cpp \
+ cpu_ref/linkloader/utils/raw_ostream.cpp \
+ cpu_ref/linkloader/utils/rsl_assert.cpp
include $(CLEAR_VARS)
@@ -278,8 +280,8 @@
LOCAL_CFLAGS += $(rs_base_CFLAGS)
LOCAL_C_INCLUDES := \
- $(LOCAL_PATH)/driver/linkloader \
- $(LOCAL_PATH)/driver/linkloader/include \
+ $(LOCAL_PATH)/cpu_ref/linkloader \
+ $(LOCAL_PATH)/cpu_ref/linkloader/include \
$(LOCAL_C_INCLUDES)
include $(LLVM_ROOT_PATH)/llvm-device-build.mk
@@ -299,7 +301,7 @@
LOCAL_SRC_FILES := $(rsloader_SRC_FILES)
ifdef USE_MINGW
-LOCAL_SRC_FILES += driver/linkloader/lib/mmanWindows.cpp
+LOCAL_SRC_FILES += cpu_ref/linkloader/lib/mmanWindows.cpp
endif
LOCAL_ADDITIONAL_DEPENDENCIES := $(LOCAL_PATH)/Android.mk
@@ -308,149 +310,11 @@
LOCAL_CFLAGS += -D__HOST__
LOCAL_C_INCLUDES := \
- $(LOCAL_PATH)/driver/linkloader \
- $(LOCAL_PATH)/driver/linkloader/include \
+ $(LOCAL_PATH)/cpu_ref/linkloader \
+ $(LOCAL_PATH)/cpu_ref/linkloader/include \
$(LOCAL_C_INCLUDES)
include $(LLVM_ROOT_PATH)/llvm-host-build.mk
include $(BUILD_HOST_STATIC_LIBRARY)
-
-#=============================================================================
-# librsloader-test (Device)
-#-----------------------------------------------------------------------------
-
-include $(CLEAR_VARS)
-
-LOCAL_MODULE := test-librsloader
-
-LOCAL_MODULE_TAGS := tests
-
-LOCAL_SHARED_LIBRARIES := \
- libstlport
-
-LOCAL_STATIC_LIBRARIES := \
- librsloader \
- libcutils \
- liblog \
- libLLVMSupport
-
-LOCAL_SRC_FILES := \
- driver/linkloader/android/test-librsloader.c
-
-LOCAL_ADDITIONAL_DEPENDENCIES := $(LOCAL_PATH)/Android.mk
-
-LOCAL_CFLAGS += $(rs_base_CFLAGS)
-
-LOCAL_C_INCLUDES := \
- $(LOCAL_PATH)/driver/linkloader \
- $(LOCAL_PATH)/driver/linkloader/include
-
-include $(LLVM_ROOT_PATH)/llvm-device-build.mk
-include $(BUILD_EXECUTABLE)
-
-
-#=============================================================================
-# librsloader-test (Host)
-#-----------------------------------------------------------------------------
-
-include $(CLEAR_VARS)
-
-LOCAL_MODULE := test-librsloader
-
-LOCAL_MODULE_TAGS := tests
-
-LOCAL_LDLIBS := \
- -lpthread \
- -ldl
-
-LOCAL_STATIC_LIBRARIES := \
- librsloader \
- libcutils \
- liblog \
- libLLVMSupport
-
-LOCAL_SRC_FILES := \
- driver/linkloader/android/test-librsloader.c
-
-LOCAL_ADDITIONAL_DEPENDENCIES := $(LOCAL_PATH)/Android.mk
-
-LOCAL_CFLAGS += $(rs_base_CFLAGS)
-
-LOCAL_C_INCLUDES := \
- $(LOCAL_PATH)/driver/linkloader \
- $(LOCAL_PATH)/driver/linkloader/include
-
-include $(LLVM_ROOT_PATH)/llvm-host-build.mk
-include $(BUILD_HOST_EXECUTABLE)
-
-
-#=============================================================================
-# rsloader
-#-----------------------------------------------------------------------------
-
-ifdef BUILD_RSLOADER_TOOL
-include $(CLEAR_VARS)
-
-LOCAL_MODULE := rsloader
-
-LOCAL_MODULE_TAGS := tests
-
-LOCAL_SHARED_LIBRARIES := \
- libstlport
-
-LOCAL_STATIC_LIBRARIES := \
- libLLVMSupport
-
-LOCAL_SRC_FILES := \
- driver/linkloader/lib/ELFHeader.cpp \
- driver/linkloader/lib/ELFSymbol.cpp \
- driver/linkloader/lib/ELFSectionHeader.cpp \
- driver/linkloader/lib/ELFTypes.cpp \
- driver/linkloader/lib/StubLayout.cpp \
- driver/linkloader/utils/raw_ostream.cpp \
- driver/linkloader/utils/rsl_assert.cpp \
- driver/linkloader/utils/helper.cpp \
- driver/linkloader/main.cpp
-
-LOCAL_C_INCLUDES := \
- $(LOCAL_PATH)/driver/linkloader \
- $(LOCAL_PATH)/driver/linkloader/include \
- $(LOCAL_C_INCLUDES)
-
-include $(LLVM_ROOT_PATH)/llvm-device-build.mk
-include $(BUILD_EXECUTABLE)
-endif
-
-
-#=============================================================================
-# stub-layout-unit-test
-#-----------------------------------------------------------------------------
-
-ifdef BUILD_STUB_LAYOUT_TEST
-include $(CLEAR_VARS)
-
-LOCAL_MODULE := stub-layout-unit-test
-
-LOCAL_MODULE_TAGS := tests
-
-LOCAL_SHARED_LIBRARIES := \
- libstlport
-
-LOCAL_SRC_FILES := \
- driver/linkloader/lib/StubLayout.cpp \
- driver/linkloader/utils/raw_ostream.cpp \
- driver/linkloader/utils/helper.cpp \
- driver/linkloader/tests/stub-test.cpp
-
-LOCAL_C_INCLUDES := \
- $(LOCAL_PATH)/driver/linkloader \
- $(LOCAL_PATH)/driver/linkloader/include \
- $(LOCAL_C_INCLUDES)
-
-include $(LLVM_ROOT_PATH)/llvm-device-build.mk
-include $(BUILD_EXECUTABLE)
-endif
-
-
include $(call all-makefiles-under,$(LOCAL_PATH))
diff --git a/cpp/Allocation.cpp b/cpp/Allocation.cpp
index 7755f74..f05b005 100644
--- a/cpp/Allocation.cpp
+++ b/cpp/Allocation.cpp
@@ -15,7 +15,6 @@
*/
#include "RenderScript.h"
-#include <rs.h>
using namespace android;
using namespace RSC;
@@ -64,11 +63,14 @@
mType = t;
mUsage = usage;
- if (t.get() != NULL) {
+ if (t != NULL) {
updateCacheInfo(t);
}
+
}
+
+
void Allocation::validateIsInt32() {
RsDataType dt = mType->getElement()->getDataType();
if ((dt == RS_TYPE_SIGNED_32) || (dt == RS_TYPE_UNSIGNED_32)) {
@@ -121,7 +123,7 @@
void Allocation::updateFromNative() {
BaseObj::updateFromNative();
- const void *typeID = rsaAllocationGetType(mRS->getContext(), getID());
+ const void *typeID = RS::dispatch->AllocationGetType(mRS->getContext(), getID());
if(typeID != NULL) {
sp<const Type> old = mType;
sp<Type> t = new Type((void *)typeID, mRS);
@@ -141,7 +143,7 @@
default:
ALOGE("Source must be exactly one usage type.");
}
- rsAllocationSyncAll(mRS->getContext(), getIDSafe(), srcLocation);
+ RS::dispatch->AllocationSyncAll(mRS->getContext(), getIDSafe(), srcLocation);
}
void Allocation::ioSendOutput() {
@@ -149,7 +151,7 @@
if ((mUsage & RS_ALLOCATION_USAGE_IO_OUTPUT) == 0) {
ALOGE("Can only send buffer if IO_OUTPUT usage specified.");
}
- rsAllocationIoSend(mRS->getContext(), getID());
+ RS::dispatch->AllocationIoSend(mRS->getContext(), getID());
#endif
}
@@ -158,12 +160,12 @@
if ((mUsage & RS_ALLOCATION_USAGE_IO_INPUT) == 0) {
ALOGE("Can only send buffer if IO_OUTPUT usage specified.");
}
- rsAllocationIoReceive(mRS->getContext(), getID());
+ RS::dispatch->AllocationIoReceive(mRS->getContext(), getID());
#endif
}
void Allocation::generateMipmaps() {
- rsAllocationGenerateMipmaps(mRS->getContext(), getID());
+ RS::dispatch->AllocationGenerateMipmaps(mRS->getContext(), getID());
}
void Allocation::copy1DRangeFrom(uint32_t off, size_t count, const void *data) {
@@ -177,8 +179,8 @@
return;
}
- rsAllocation1DData(mRS->getContext(), getIDSafe(), off, mSelectedLOD, count, data,
- count * mType->getElement()->getSizeBytes());
+ RS::dispatch->Allocation1DData(mRS->getContext(), getIDSafe(), off, mSelectedLOD, count, data,
+ count * mType->getElement()->getSizeBytes());
}
void Allocation::copy1DRangeTo(uint32_t off, size_t count, void *data) {
@@ -191,17 +193,17 @@
return;
}
- rsAllocation1DRead(mRS->getContext(), getIDSafe(), off, mSelectedLOD, count, data,
- count * mType->getElement()->getSizeBytes());
+ RS::dispatch->Allocation1DRead(mRS->getContext(), getIDSafe(), off, mSelectedLOD, count, data,
+ count * mType->getElement()->getSizeBytes());
}
void Allocation::copy1DRangeFrom(uint32_t off, size_t count, sp<const Allocation> data,
uint32_t dataOff) {
- rsAllocationCopy2DRange(mRS->getContext(), getIDSafe(), off, 0,
- mSelectedLOD, mSelectedFace,
- count, 1, data->getIDSafe(), dataOff, 0,
- data->mSelectedLOD, data->mSelectedFace);
+ RS::dispatch->AllocationCopy2DRange(mRS->getContext(), getIDSafe(), off, 0,
+ mSelectedLOD, mSelectedFace,
+ count, 1, data->getIDSafe(), dataOff, 0,
+ data->mSelectedLOD, data->mSelectedFace);
}
void Allocation::copy1DFrom(const void* data) {
@@ -226,31 +228,31 @@
void Allocation::copy2DRangeFrom(uint32_t xoff, uint32_t yoff, uint32_t w, uint32_t h,
const void *data) {
validate2DRange(xoff, yoff, w, h);
- rsAllocation2DData(mRS->getContext(), getIDSafe(), xoff, yoff, mSelectedLOD, mSelectedFace,
- w, h, data, w * h * mType->getElement()->getSizeBytes(), w * mType->getElement()->getSizeBytes());
+ RS::dispatch->Allocation2DData(mRS->getContext(), getIDSafe(), xoff, yoff, mSelectedLOD, mSelectedFace,
+ w, h, data, w * h * mType->getElement()->getSizeBytes(), w * mType->getElement()->getSizeBytes());
}
void Allocation::copy2DRangeFrom(uint32_t xoff, uint32_t yoff, uint32_t w, uint32_t h,
sp<const Allocation> data, uint32_t dataXoff, uint32_t dataYoff) {
validate2DRange(xoff, yoff, w, h);
- rsAllocationCopy2DRange(mRS->getContext(), getIDSafe(), xoff, yoff,
- mSelectedLOD, mSelectedFace,
- w, h, data->getIDSafe(), dataXoff, dataYoff,
- data->mSelectedLOD, data->mSelectedFace);
+ RS::dispatch->AllocationCopy2DRange(mRS->getContext(), getIDSafe(), xoff, yoff,
+ mSelectedLOD, mSelectedFace,
+ w, h, data->getIDSafe(), dataXoff, dataYoff,
+ data->mSelectedLOD, data->mSelectedFace);
}
void Allocation::copy2DRangeTo(uint32_t xoff, uint32_t yoff, uint32_t w, uint32_t h,
void* data) {
validate2DRange(xoff, yoff, w, h);
- rsAllocation2DRead(mRS->getContext(), getIDSafe(), xoff, yoff, mSelectedLOD, mSelectedFace,
- w, h, data, w * h * mType->getElement()->getSizeBytes(), w * mType->getElement()->getSizeBytes());
+ RS::dispatch->Allocation2DRead(mRS->getContext(), getIDSafe(), xoff, yoff, mSelectedLOD, mSelectedFace,
+ w, h, data, w * h * mType->getElement()->getSizeBytes(), w * mType->getElement()->getSizeBytes());
}
void Allocation::copy2DStridedFrom(uint32_t xoff, uint32_t yoff, uint32_t w, uint32_t h,
const void *data, size_t stride) {
validate2DRange(xoff, yoff, w, h);
- rsAllocation2DData(mRS->getContext(), getIDSafe(), xoff, yoff, mSelectedLOD, mSelectedFace,
- w, h, data, w * h * mType->getElement()->getSizeBytes(), stride);
+ RS::dispatch->Allocation2DData(mRS->getContext(), getIDSafe(), xoff, yoff, mSelectedLOD, mSelectedFace,
+ w, h, data, w * h * mType->getElement()->getSizeBytes(), stride);
}
void Allocation::copy2DStridedFrom(const void* data, size_t stride) {
@@ -260,52 +262,17 @@
void Allocation::copy2DStridedTo(uint32_t xoff, uint32_t yoff, uint32_t w, uint32_t h,
void *data, size_t stride) {
validate2DRange(xoff, yoff, w, h);
- rsAllocation2DRead(mRS->getContext(), getIDSafe(), xoff, yoff, mSelectedLOD, mSelectedFace,
- w, h, data, w * h * mType->getElement()->getSizeBytes(), stride);
+ RS::dispatch->Allocation2DRead(mRS->getContext(), getIDSafe(), xoff, yoff, mSelectedLOD, mSelectedFace,
+ w, h, data, w * h * mType->getElement()->getSizeBytes(), stride);
}
void Allocation::copy2DStridedTo(void* data, size_t stride) {
copy2DStridedTo(0, 0, mCurrentDimX, mCurrentDimY, data, stride);
}
-
-/*
-void resize(int dimX) {
- if ((mType.getY() > 0)|| (mType.getZ() > 0) || mType.hasFaces() || mType.hasMipmaps()) {
- throw new RSInvalidStateException("Resize only support for 1D allocations at this time.");
- }
- mRS.nAllocationResize1D(getID(), dimX);
- mRS.finish(); // Necessary because resize is fifoed and update is async.
-
- int typeID = mRS.nAllocationGetType(getID());
- mType = new Type(typeID, mRS);
- mType.updateFromNative();
- updateCacheInfo(mType);
-}
-
-void resize(int dimX, int dimY) {
- if ((mType.getZ() > 0) || mType.hasFaces() || mType.hasMipmaps()) {
- throw new RSInvalidStateException(
- "Resize only support for 2D allocations at this time.");
- }
- if (mType.getY() == 0) {
- throw new RSInvalidStateException(
- "Resize only support for 2D allocations at this time.");
- }
- mRS.nAllocationResize2D(getID(), dimX, dimY);
- mRS.finish(); // Necessary because resize is fifoed and update is async.
-
- int typeID = mRS.nAllocationGetType(getID());
- mType = new Type(typeID, mRS);
- mType.updateFromNative();
- updateCacheInfo(mType);
-}
-*/
-
-
-android::sp<Allocation> Allocation::createTyped(sp<RS> rs, sp<const Type> type,
- RsAllocationMipmapControl mips, uint32_t usage) {
- void *id = rsAllocationCreateTyped(rs->getContext(), type->getID(), mips, usage, 0);
+sp<Allocation> Allocation::createTyped(sp<RS> rs, sp<const Type> type,
+ RsAllocationMipmapControl mips, uint32_t usage) {
+ void *id = RS::dispatch->AllocationCreateTyped(rs->getContext(), type->getID(), mips, usage, 0);
if (id == 0) {
ALOGE("Allocation creation failed.");
return NULL;
@@ -313,24 +280,24 @@
return new Allocation(id, rs, type, usage);
}
-android::sp<Allocation> Allocation::createTyped(sp<RS> rs, sp<const Type> type,
- RsAllocationMipmapControl mips, uint32_t usage,
- void *pointer) {
- void *id = rsAllocationCreateTyped(rs->getContext(), type->getID(), mips, usage,
- (uintptr_t)pointer);
+sp<Allocation> Allocation::createTyped(sp<RS> rs, sp<const Type> type,
+ RsAllocationMipmapControl mips, uint32_t usage,
+ void *pointer) {
+ void *id = RS::dispatch->AllocationCreateTyped(rs->getContext(), type->getID(), mips, usage,
+ (uintptr_t)pointer);
if (id == 0) {
ALOGE("Allocation creation failed.");
}
return new Allocation(id, rs, type, usage);
}
-android::sp<Allocation> Allocation::createTyped(sp<RS> rs, sp<const Type> type,
- uint32_t usage) {
+sp<Allocation> Allocation::createTyped(sp<RS> rs, sp<const Type> type,
+ uint32_t usage) {
return createTyped(rs, type, RS_ALLOCATION_MIPMAP_NONE, usage);
}
-android::sp<Allocation> Allocation::createSized(sp<RS> rs, sp<const Element> e,
- size_t count, uint32_t usage) {
+sp<Allocation> Allocation::createSized(sp<RS> rs, sp<const Element> e,
+ size_t count, uint32_t usage) {
Type::Builder b(rs, e);
b.setX(count);
sp<const Type> t = b.create();
@@ -338,8 +305,8 @@
return createTyped(rs, t, usage);
}
-android::sp<Allocation> Allocation::createSized2D(sp<RS> rs, sp<const Element> e,
- size_t x, size_t y, uint32_t usage) {
+sp<Allocation> Allocation::createSized2D(sp<RS> rs, sp<const Element> e,
+ size_t x, size_t y, uint32_t usage) {
Type::Builder b(rs, e);
b.setX(x);
b.setY(y);
diff --git a/cpp/Android.mk b/cpp/Android.mk
index 145f487..d266d14 100644
--- a/cpp/Android.mk
+++ b/cpp/Android.mk
@@ -1,3 +1,14 @@
+rs_cpp_SRC_FILES := \
+ RenderScript.cpp \
+ BaseObj.cpp \
+ Element.cpp \
+ Type.cpp \
+ Allocation.cpp \
+ Script.cpp \
+ ScriptC.cpp \
+ ScriptIntrinsics.cpp \
+ Sampler.cpp
+
LOCAL_PATH:= $(call my-dir)
include $(CLEAR_VARS)
@@ -11,37 +22,50 @@
endif
local_cflags_for_rs_cpp += -DRS_VERSION=$(RS_VERSION)
+LOCAL_SRC_FILES := $(rs_cpp_SRC_FILES)
+
LOCAL_CFLAGS += $(local_cflags_for_rs_cpp)
-LOCAL_SRC_FILES:= \
- RenderScript.cpp \
- BaseObj.cpp \
- Element.cpp \
- Type.cpp \
- Allocation.cpp \
- Script.cpp \
- ScriptC.cpp \
- ScriptIntrinsics.cpp
-
LOCAL_SHARED_LIBRARIES := \
- libRS \
libz \
libcutils \
libutils \
- liblog
+ liblog \
+ libdl \
+ libstlport
LOCAL_MODULE:= libRScpp
LOCAL_MODULE_TAGS := optional
-intermediates := $(call intermediates-dir-for,STATIC_LIBRARIES,libRS,TARGET,)
-librs_generated_headers := \
- $(intermediates)/rsgApiStructs.h \
- $(intermediates)/rsgApiFuncDecl.h
-LOCAL_GENERATED_SOURCES := $(librs_generated_headers)
-
LOCAL_C_INCLUDES += frameworks/rs
+LOCAL_C_INCLUDES += external/stlport/stlport bionic/ bionic/libstdc++/include
LOCAL_C_INCLUDES += $(intermediates)
-
include $(BUILD_SHARED_LIBRARY)
+
+
+include $(CLEAR_VARS)
+
+LOCAL_CFLAGS += $(local_cflags_for_rs_cpp)
+
+LOCAL_SRC_FILES := $(rs_cpp_SRC_FILES)
+
+LOCAL_STATIC_LIBRARIES := \
+ libz \
+ libcutils \
+ libutils \
+ liblog \
+ libstlport_static
+
+LOCAL_SHARED_LIBRARIES := libdl
+
+LOCAL_MODULE:= libRScpp_static
+
+LOCAL_MODULE_TAGS := optional
+
+LOCAL_C_INCLUDES += frameworks/rs
+LOCAL_C_INCLUDES += external/stlport/stlport bionic/ bionic/libstdc++/include
+LOCAL_C_INCLUDES += $(intermediates)
+
+include $(BUILD_STATIC_LIBRARY)
diff --git a/cpp/BaseObj.cpp b/cpp/BaseObj.cpp
index 828bd87..efcb83a 100644
--- a/cpp/BaseObj.cpp
+++ b/cpp/BaseObj.cpp
@@ -15,7 +15,6 @@
*/
#include "RenderScript.h"
-#include <rs.h>
using namespace android;
using namespace RSC;
@@ -28,7 +27,7 @@
}
void * BaseObj::getObjID(sp<const BaseObj> o) {
- return o.get() == NULL ? NULL : o->getID();
+ return o == NULL ? NULL : o->getID();
}
@@ -44,20 +43,20 @@
}
BaseObj::~BaseObj() {
- rsObjDestroy(mRS->getContext(), mID);
+ RS::dispatch->ObjDestroy(mRS->getContext(), mID);
mRS = NULL;
mID = NULL;
}
void BaseObj::updateFromNative() {
const char *name = NULL;
- rsaGetName(mRS->getContext(), mID, &name);
+ RS::dispatch->GetName(mRS->getContext(), mID, &name);
mName = name;
}
-bool BaseObj::equals(const BaseObj *obj) {
+bool BaseObj::equals(sp<const BaseObj> obj) {
// Early-out check to see if both BaseObjs are actually the same
- if (this == obj)
+ if (this == obj.get())
return true;
return mID == obj->mID;
}
diff --git a/cpp/Element.cpp b/cpp/Element.cpp
index 51a96cd..d4b2c66 100644
--- a/cpp/Element.cpp
+++ b/cpp/Element.cpp
@@ -18,12 +18,11 @@
#include <string.h>
#include "RenderScript.h"
-#include <rs.h>
using namespace android;
using namespace RSC;
-sp<const Element> Element::getSubElement(uint32_t index) {
+android::RSC::sp<const Element> Element::getSubElement(uint32_t index) {
if (!mVisibleElementMap.size()) {
mRS->throwError("Element contains no sub-elements");
}
@@ -64,9 +63,13 @@
}
-#define CREATE_USER(N, T) sp<const Element> Element::N(sp<RS> rs) { \
- return createUser(rs, RS_TYPE_##T); \
-}
+#define CREATE_USER(N, T) android::RSC::sp<const Element> Element::N(android::RSC::sp<RS> rs) { \
+ if (rs->mElements.N == NULL) { \
+ rs->mElements.N = (createUser(rs, RS_TYPE_##T)).get(); \
+ } \
+ return rs->mElements.N; \
+ }
+
CREATE_USER(BOOLEAN, BOOLEAN);
CREATE_USER(U8, UNSIGNED_8);
CREATE_USER(I8, SIGNED_8);
@@ -92,7 +95,7 @@
CREATE_USER(MATRIX_3X3, MATRIX_3X3);
CREATE_USER(MATRIX_2X2, MATRIX_2X2);
-#define CREATE_PIXEL(N, T, K) sp<const Element> Element::N(sp<RS> rs) { \
+#define CREATE_PIXEL(N, T, K) android::RSC::sp<const Element> Element::N(android::RSC::sp<RS> rs) { \
return createPixel(rs, RS_TYPE_##T, RS_KIND_##K); \
}
CREATE_PIXEL(A_8, UNSIGNED_8, PIXEL_A);
@@ -101,13 +104,13 @@
CREATE_PIXEL(RGBA_4444, UNSIGNED_4_4_4_4, PIXEL_RGBA);
CREATE_PIXEL(RGBA_8888, UNSIGNED_8, PIXEL_RGBA);
-#define CREATE_VECTOR(N, T) sp<const Element> Element::N##_2(sp<RS> rs) { \
+#define CREATE_VECTOR(N, T) android::RSC::sp<const Element> Element::N##_2(android::RSC::sp<RS> rs) { \
return createVector(rs, RS_TYPE_##T, 2); \
} \
-sp<const Element> Element::N##_3(sp<RS> rs) { \
+android::RSC::sp<const Element> Element::N##_3(android::RSC::sp<RS> rs) { \
return createVector(rs, RS_TYPE_##T, 3); \
} \
-sp<const Element> Element::N##_4(sp<RS> rs) { \
+android::RSC::sp<const Element> Element::N##_4(android::RSC::sp<RS> rs) { \
return createVector(rs, RS_TYPE_##T, 4); \
}
CREATE_VECTOR(U8, UNSIGNED_8);
@@ -140,15 +143,15 @@
// Make a map that points us at non-padding elements
for (size_t ct = 0; ct < fieldCount; ct ++) {
if (mElementNames[ct].string()[0] != '#') {
- mVisibleElementMap.push((uint32_t)ct);
+ mVisibleElementMap.push_back((uint32_t)ct);
}
}
}
-Element::Element(void *id, sp<RS> rs,
- android::Vector<sp<Element> > &elements,
- android::Vector<android::String8> &elementNames,
- android::Vector<uint32_t> &arraySizes) : BaseObj(id, rs) {
+Element::Element(void *id, android::RSC::sp<RS> rs,
+ std::vector<android::RSC::sp<Element> > &elements,
+ std::vector<android::String8> &elementNames,
+ std::vector<uint32_t> &arraySizes) : BaseObj(id, rs) {
mSizeBytes = 0;
mVectorSize = 1;
mElements = elements;
@@ -159,7 +162,7 @@
mKind = RS_KIND_USER;
for (size_t ct = 0; ct < mElements.size(); ct++ ) {
- mOffsetInBytes.push(mSizeBytes);
+ mOffsetInBytes.push_back(mSizeBytes);
mSizeBytes += mElements[ct]->mSizeBytes * mArraySizes[ct];
}
updateVisibleSubElements();
@@ -219,7 +222,7 @@
return 0;
}
-Element::Element(void *id, sp<RS> rs,
+Element::Element(void *id, android::RSC::sp<RS> rs,
RsDataType dt, RsDataKind dk, bool norm, uint32_t size) :
BaseObj(id, rs)
{
@@ -249,20 +252,20 @@
updateVisibleSubElements();
}
-sp<const Element> Element::createUser(sp<RS> rs, RsDataType dt) {
- void * id = rsElementCreate(rs->getContext(), dt, RS_KIND_USER, false, 1);
+android::RSC::sp<const Element> Element::createUser(android::RSC::sp<RS> rs, RsDataType dt) {
+ void * id = RS::dispatch->ElementCreate(rs->getContext(), dt, RS_KIND_USER, false, 1);
return new Element(id, rs, dt, RS_KIND_USER, false, 1);
}
-sp<const Element> Element::createVector(sp<RS> rs, RsDataType dt, uint32_t size) {
+android::RSC::sp<const Element> Element::createVector(android::RSC::sp<RS> rs, RsDataType dt, uint32_t size) {
if (size < 2 || size > 4) {
rs->throwError("Vector size out of range 2-4.");
}
- void *id = rsElementCreate(rs->getContext(), dt, RS_KIND_USER, false, size);
+ void *id = RS::dispatch->ElementCreate(rs->getContext(), dt, RS_KIND_USER, false, size);
return new Element(id, rs, dt, RS_KIND_USER, false, size);
}
-sp<const Element> Element::createPixel(sp<RS> rs, RsDataType dt, RsDataKind dk) {
+android::RSC::sp<const Element> Element::createPixel(android::RSC::sp<RS> rs, RsDataType dt, RsDataKind dk) {
if (!(dk == RS_KIND_PIXEL_L ||
dk == RS_KIND_PIXEL_A ||
dk == RS_KIND_PIXEL_LA ||
@@ -309,11 +312,11 @@
break;
}
- void * id = rsElementCreate(rs->getContext(), dt, dk, true, size);
+ void * id = RS::dispatch->ElementCreate(rs->getContext(), dt, dk, true, size);
return new Element(id, rs, dt, dk, true, size);
}
-bool Element::isCompatible(sp<const Element>e) {
+bool Element::isCompatible(android::RSC::sp<const Element>e) {
// Try strict BaseObj equality to start with.
if (this == e.get()) {
return true;
@@ -329,12 +332,12 @@
(mVectorSize == e->mVectorSize));
}
-Element::Builder::Builder(sp<RS> rs) {
+Element::Builder::Builder(android::RSC::sp<RS> rs) {
mRS = rs;
mSkipPadding = false;
}
-void Element::Builder::add(sp</*const*/ Element>e, android::String8 &name, uint32_t arraySize) {
+void Element::Builder::add(android::RSC::sp</*const*/ Element>e, android::String8 &name, uint32_t arraySize) {
// Skip padding fields after a vector 3 type.
if (mSkipPadding) {
const char *s1 = "#padding_";
@@ -354,12 +357,12 @@
mSkipPadding = false;
}
- mElements.add(e);
- mElementNames.add(name);
- mArraySizes.add(arraySize);
+ mElements.push_back(e);
+ mElementNames.push_back(name);
+ mArraySizes.push_back(arraySize);
}
-sp<const Element> Element::Builder::create() {
+android::RSC::sp<const Element> Element::Builder::create() {
size_t fieldCount = mElements.size();
const char ** nameArray = (const char **)calloc(fieldCount, sizeof(char *));
const Element ** elementArray = (const Element **)calloc(fieldCount, sizeof(Element *));
@@ -371,10 +374,10 @@
sizeArray[ct] = mElementNames[ct].length();
}
- void *id = rsElementCreate2(mRS->getContext(),
+ void *id = RS::dispatch->ElementCreate2(mRS->getContext(),
(RsElement *)elementArray, fieldCount,
nameArray, fieldCount * sizeof(size_t), sizeArray,
- (const uint32_t *)mArraySizes.array(), fieldCount);
+ (const uint32_t *)&mArraySizes[0], fieldCount);
free(nameArray);
diff --git a/cpp/RenderScript.cpp b/cpp/RenderScript.cpp
index 134d34b..18e7117 100644
--- a/cpp/RenderScript.cpp
+++ b/cpp/RenderScript.cpp
@@ -1,5 +1,5 @@
/*
- * Copyright (C) 2012 The Android Open Source Project
+ * Copyright (C) 2013 The Android Open Source Project
*
* Licensed under the Apache License, Version 2.0 (the "License");
* you may not use this file except in compliance with the License.
@@ -19,13 +19,25 @@
#include <pthread.h>
#include "RenderScript.h"
-#include "rs.h"
+#include "rsCppStructs.h"
+
+#include <dlfcn.h>
+
+#if !defined(RS_SERVER) && defined(HAVE_ANDROID_OS)
+#include <cutils/properties.h>
+#endif
+
+#define LOG_NDEBUG 0
+#define LOG_TAG "rsC++"
using namespace android;
using namespace RSC;
bool RS::gInitialized = false;
+bool RS::usingNative = false;
pthread_mutex_t RS::gInitMutex = PTHREAD_MUTEX_INITIALIZER;
+dispatchTable* RS::dispatch = NULL;
+static int gInitError = 0;
RS::RS() {
mDev = NULL;
@@ -33,36 +45,435 @@
mErrorFunc = NULL;
mMessageFunc = NULL;
mMessageRun = false;
+ mInit = false;
memset(&mElements, 0, sizeof(mElements));
+ memset(&mSamplers, 0, sizeof(mSamplers));
}
RS::~RS() {
- mMessageRun = false;
+ if (mInit == true) {
+ mMessageRun = false;
- rsContextDeinitToClient(mContext);
+ RS::dispatch->ContextDeinitToClient(mContext);
- void *res = NULL;
- int status = pthread_join(mMessageThreadId, &res);
+ void *res = NULL;
+ int status = pthread_join(mMessageThreadId, &res);
- rsContextDestroy(mContext);
- mContext = NULL;
- rsDeviceDestroy(mDev);
- mDev = NULL;
+ RS::dispatch->ContextDestroy(mContext);
+ mContext = NULL;
+ RS::dispatch->DeviceDestroy(mDev);
+ mDev = NULL;
+ }
}
bool RS::init(bool forceCpu, bool synchronous) {
return RS::init(RS_VERSION, forceCpu, synchronous);
}
+static bool loadSymbols(void* handle) {
+
+ RS::dispatch->AllocationGetType = (AllocationGetTypeFnPtr)dlsym(handle, "rsaAllocationGetType");
+ if (RS::dispatch->AllocationGetType == NULL) {
+ ALOGE("Couldn't initialize RS::dispatch->AllocationGetType");
+ return false;
+ }
+ RS::dispatch->TypeGetNativeData = (TypeGetNativeDataFnPtr)dlsym(handle, "rsaTypeGetNativeData");
+ if (RS::dispatch->TypeGetNativeData == NULL) {
+ ALOGE("Couldn't initialize RS::dispatch->TypeGetNativeData");
+ return false;
+ }
+ RS::dispatch->ElementGetNativeData = (ElementGetNativeDataFnPtr)dlsym(handle, "rsaElementGetNativeData");
+ if (RS::dispatch->ElementGetNativeData == NULL) {
+ ALOGE("Couldn't initialize RS::dispatch->ElementGetNativeData");
+ return false;
+ }
+ RS::dispatch->ElementGetSubElements = (ElementGetSubElementsFnPtr)dlsym(handle, "rsaElementGetSubElements");
+ if (RS::dispatch->ElementGetSubElements == NULL) {
+ ALOGE("Couldn't initialize RS::dispatch->ElementGetSubElements");
+ return false;
+ }
+ RS::dispatch->DeviceCreate = (DeviceCreateFnPtr)dlsym(handle, "rsDeviceCreate");
+ if (RS::dispatch->DeviceCreate == NULL) {
+ ALOGE("Couldn't initialize RS::dispatch->DeviceCreate");
+ return false;
+ }
+ RS::dispatch->DeviceDestroy = (DeviceDestroyFnPtr)dlsym(handle, "rsDeviceDestroy");
+ if (RS::dispatch->DeviceDestroy == NULL) {
+ ALOGE("Couldn't initialize RS::dispatch->DeviceDestroy");
+ return false;
+ }
+ RS::dispatch->DeviceSetConfig = (DeviceSetConfigFnPtr)dlsym(handle, "rsDeviceSetConfig");
+ if (RS::dispatch->DeviceSetConfig == NULL) {
+ ALOGE("Couldn't initialize RS::dispatch->DeviceSetConfig");
+ return false;
+ }
+ RS::dispatch->ContextCreate = (ContextCreateFnPtr)dlsym(handle, "rsContextCreate");;
+ if (RS::dispatch->ContextCreate == NULL) {
+ ALOGE("Couldn't initialize RS::dispatch->ContextCreate");
+ return false;
+ }
+ RS::dispatch->GetName = (GetNameFnPtr)dlsym(handle, "rsaGetName");;
+ if (RS::dispatch->GetName == NULL) {
+ ALOGE("Couldn't initialize RS::dispatch->GetName");
+ return false;
+ }
+ RS::dispatch->ContextDestroy = (ContextDestroyFnPtr)dlsym(handle, "rsContextDestroy");
+ if (RS::dispatch->ContextDestroy == NULL) {
+ ALOGE("Couldn't initialize RS::dispatch->ContextDestroy");
+ return false;
+ }
+ RS::dispatch->ContextGetMessage = (ContextGetMessageFnPtr)dlsym(handle, "rsContextGetMessage");
+ if (RS::dispatch->ContextGetMessage == NULL) {
+ ALOGE("Couldn't initialize RS::dispatch->ContextGetMessage");
+ return false;
+ }
+ RS::dispatch->ContextPeekMessage = (ContextPeekMessageFnPtr)dlsym(handle, "rsContextPeekMessage");
+ if (RS::dispatch->ContextPeekMessage == NULL) {
+ ALOGE("Couldn't initialize RS::dispatch->ContextPeekMessage");
+ return false;
+ }
+ RS::dispatch->ContextSendMessage = (ContextSendMessageFnPtr)dlsym(handle, "rsContextSendMessage");
+ if (RS::dispatch->ContextSendMessage == NULL) {
+ ALOGE("Couldn't initialize RS::dispatch->ContextSendMessage");
+ return false;
+ }
+ RS::dispatch->ContextInitToClient = (ContextInitToClientFnPtr)dlsym(handle, "rsContextInitToClient");
+ if (RS::dispatch->ContextInitToClient == NULL) {
+ ALOGE("Couldn't initialize RS::dispatch->ContextInitToClient");
+ return false;
+ }
+ RS::dispatch->ContextDeinitToClient = (ContextDeinitToClientFnPtr)dlsym(handle, "rsContextDeinitToClient");
+ if (RS::dispatch->ContextDeinitToClient == NULL) {
+ ALOGE("Couldn't initialize RS::dispatch->ContextDeinitToClient");
+ return false;
+ }
+ RS::dispatch->TypeCreate = (TypeCreateFnPtr)dlsym(handle, "rsTypeCreate");
+ if (RS::dispatch->TypeCreate == NULL) {
+ ALOGE("Couldn't initialize RS::dispatch->TypeCreate");
+ return false;
+ }
+ RS::dispatch->AllocationCreateTyped = (AllocationCreateTypedFnPtr)dlsym(handle, "rsAllocationCreateTyped");
+ if (RS::dispatch->AllocationCreateTyped == NULL) {
+ ALOGE("Couldn't initialize RS::dispatch->AllocationCreateTyped");
+ return false;
+ }
+ RS::dispatch->AllocationCreateFromBitmap = (AllocationCreateFromBitmapFnPtr)dlsym(handle, "rsAllocationCreateFromBitmap");
+ if (RS::dispatch->AllocationCreateFromBitmap == NULL) {
+ ALOGE("Couldn't initialize RS::dispatch->AllocationCreateFromBitmap");
+ return false;
+ }
+ RS::dispatch->AllocationCubeCreateFromBitmap = (AllocationCubeCreateFromBitmapFnPtr)dlsym(handle, "rsAllocationCubeCreateFromBitmap");
+ if (RS::dispatch->AllocationCubeCreateFromBitmap == NULL) {
+ ALOGE("Couldn't initialize RS::dispatch->AllocationCubeCreateFromBitmap");
+ return false;
+ }
+ RS::dispatch->AllocationGetSurface = (AllocationGetSurfaceFnPtr)dlsym(handle, "rsAllocationGetSurface");
+ if (RS::dispatch->AllocationGetSurface == NULL) {
+ ALOGE("Couldn't initialize RS::dispatch->AllocationGetSurface");
+ return false;
+ }
+ RS::dispatch->AllocationSetSurface = (AllocationSetSurfaceFnPtr)dlsym(handle, "rsAllocationSetSurface");
+ if (RS::dispatch->AllocationSetSurface == NULL) {
+ ALOGE("Couldn't initialize RS::dispatch->AllocationSetSurface");
+ return false;
+ }
+ RS::dispatch->ContextFinish = (ContextFinishFnPtr)dlsym(handle, "rsContextFinish");
+ if (RS::dispatch->ContextFinish == NULL) {
+ ALOGE("Couldn't initialize RS::dispatch->ContextFinish");
+ return false;
+ }
+ RS::dispatch->ContextDump = (ContextDumpFnPtr)dlsym(handle, "rsContextDump");
+ if (RS::dispatch->ContextDump == NULL) {
+ ALOGE("Couldn't initialize RS::dispatch->ContextDump");
+ return false;
+ }
+ RS::dispatch->ContextSetPriority = (ContextSetPriorityFnPtr)dlsym(handle, "rsContextSetPriority");
+ if (RS::dispatch->ContextSetPriority == NULL) {
+ ALOGE("Couldn't initialize RS::dispatch->ContextSetPriority");
+ return false;
+ }
+ RS::dispatch->AssignName = (AssignNameFnPtr)dlsym(handle, "rsAssignName");
+ if (RS::dispatch->AssignName == NULL) {
+ ALOGE("Couldn't initialize RS::dispatch->AssignName");
+ return false;
+ }
+ RS::dispatch->ObjDestroy = (ObjDestroyFnPtr)dlsym(handle, "rsObjDestroy");
+ if (RS::dispatch->ObjDestroy == NULL) {
+ ALOGE("Couldn't initialize RS::dispatch->ObjDestroy");
+ return false;
+ }
+ RS::dispatch->ElementCreate = (ElementCreateFnPtr)dlsym(handle, "rsElementCreate");
+ if (RS::dispatch->ElementCreate == NULL) {
+ ALOGE("Couldn't initialize RS::dispatch->ElementCreate");
+ return false;
+ }
+ RS::dispatch->ElementCreate2 = (ElementCreate2FnPtr)dlsym(handle, "rsElementCreate2");
+ if (RS::dispatch->ElementCreate2 == NULL) {
+ ALOGE("Couldn't initialize RS::dispatch->ElementCreate2");
+ return false;
+ }
+ RS::dispatch->AllocationCopyToBitmap = (AllocationCopyToBitmapFnPtr)dlsym(handle, "rsAllocationCopyToBitmap");
+ if (RS::dispatch->AllocationCopyToBitmap == NULL) {
+ ALOGE("Couldn't initialize RS::dispatch->AllocationCopyToBitmap");
+ return false;
+ }
+ RS::dispatch->Allocation1DData = (Allocation1DDataFnPtr)dlsym(handle, "rsAllocation1DData");
+ if (RS::dispatch->Allocation1DData == NULL) {
+ ALOGE("Couldn't initialize RS::dispatch->Allocation1DData");
+ return false;
+ }
+ RS::dispatch->Allocation1DElementData = (Allocation1DElementDataFnPtr)dlsym(handle, "rsAllocation1DElementData");
+ if (RS::dispatch->Allocation1DElementData == NULL) {
+ ALOGE("Couldn't initialize RS::dispatch->Allocation1DElementData");
+ return false;
+ }
+ RS::dispatch->Allocation2DData = (Allocation2DDataFnPtr)dlsym(handle, "rsAllocation2DData");
+ if (RS::dispatch->Allocation2DData == NULL) {
+ ALOGE("Couldn't initialize RS::dispatch->Allocation2DData");
+ return false;
+ }
+ RS::dispatch->Allocation3DData = (Allocation3DDataFnPtr)dlsym(handle, "rsAllocation3DData");
+ if (RS::dispatch->Allocation3DData == NULL) {
+ ALOGE("Couldn't initialize RS::dispatch->Allocation3DData");
+ return false;
+ }
+ RS::dispatch->AllocationGenerateMipmaps = (AllocationGenerateMipmapsFnPtr)dlsym(handle, "rsAllocationGenerateMipmaps");
+ if (RS::dispatch->AllocationGenerateMipmaps == NULL) {
+ ALOGE("Couldn't initialize RS::dispatch->AllocationGenerateMipmaps");
+ return false;
+ }
+ RS::dispatch->AllocationRead = (AllocationReadFnPtr)dlsym(handle, "rsAllocationRead");
+ if (RS::dispatch->AllocationRead == NULL) {
+ ALOGE("Couldn't initialize RS::dispatch->AllocationRead");
+ return false;
+ }
+ RS::dispatch->Allocation1DRead = (Allocation1DReadFnPtr)dlsym(handle, "rsAllocation1DRead");
+ if (RS::dispatch->Allocation1DRead == NULL) {
+ ALOGE("Couldn't initialize RS::dispatch->Allocation1DRead");
+ return false;
+ }
+ RS::dispatch->Allocation2DRead = (Allocation2DReadFnPtr)dlsym(handle, "rsAllocation2DRead");
+ if (RS::dispatch->Allocation2DRead == NULL) {
+ ALOGE("Couldn't initialize RS::dispatch->Allocation2DRead");
+ return false;
+ }
+ RS::dispatch->AllocationSyncAll = (AllocationSyncAllFnPtr)dlsym(handle, "rsAllocationSyncAll");
+ if (RS::dispatch->AllocationSyncAll == NULL) {
+ ALOGE("Couldn't initialize RS::dispatch->AllocationSyncAll");
+ return false;
+ }
+ RS::dispatch->AllocationResize1D = (AllocationResize1DFnPtr)dlsym(handle, "rsAllocationResize1D");
+ if (RS::dispatch->AllocationResize1D == NULL) {
+ ALOGE("Couldn't initialize RS::dispatch->AllocationResize1D");
+ return false;
+ }
+ RS::dispatch->AllocationCopy2DRange = (AllocationCopy2DRangeFnPtr)dlsym(handle, "rsAllocationCopy2DRange");
+ if (RS::dispatch->AllocationCopy2DRange == NULL) {
+ ALOGE("Couldn't initialize RS::dispatch->AllocationCopy2DRange");
+ return false;
+ }
+ RS::dispatch->AllocationCopy3DRange = (AllocationCopy3DRangeFnPtr)dlsym(handle, "rsAllocationCopy3DRange");
+ if (RS::dispatch->AllocationCopy3DRange == NULL) {
+ ALOGE("Couldn't initialize RS::dispatch->AllocationCopy3DRange");
+ return false;
+ }
+ RS::dispatch->SamplerCreate = (SamplerCreateFnPtr)dlsym(handle, "rsSamplerCreate");
+ if (RS::dispatch->SamplerCreate == NULL) {
+ ALOGE("Couldn't initialize RS::dispatch->SamplerCreate");
+ return false;
+ }
+ RS::dispatch->ScriptBindAllocation = (ScriptBindAllocationFnPtr)dlsym(handle, "rsScriptBindAllocation");
+ if (RS::dispatch->ScriptBindAllocation == NULL) {
+ ALOGE("Couldn't initialize RS::dispatch->ScriptBindAllocation");
+ return false;
+ }
+ RS::dispatch->ScriptSetTimeZone = (ScriptSetTimeZoneFnPtr)dlsym(handle, "rsScriptSetTimeZone");
+ if (RS::dispatch->ScriptSetTimeZone == NULL) {
+ ALOGE("Couldn't initialize RS::dispatch->ScriptSetTimeZone");
+ return false;
+ }
+ RS::dispatch->ScriptInvoke = (ScriptInvokeFnPtr)dlsym(handle, "rsScriptInvoke");
+ if (RS::dispatch->ScriptInvoke == NULL) {
+ ALOGE("Couldn't initialize RS::dispatch->ScriptInvoke");
+ return false;
+ }
+ RS::dispatch->ScriptInvokeV = (ScriptInvokeVFnPtr)dlsym(handle, "rsScriptInvokeV");
+ if (RS::dispatch->ScriptInvokeV == NULL) {
+ ALOGE("Couldn't initialize RS::dispatch->ScriptInvokeV");
+ return false;
+ }
+ RS::dispatch->ScriptForEach = (ScriptForEachFnPtr)dlsym(handle, "rsScriptForEach");
+ if (RS::dispatch->ScriptForEach == NULL) {
+ ALOGE("Couldn't initialize RS::dispatch->ScriptForEach");
+ return false;
+ }
+ RS::dispatch->ScriptSetVarI = (ScriptSetVarIFnPtr)dlsym(handle, "rsScriptSetVarI");
+ if (RS::dispatch->ScriptSetVarI == NULL) {
+ ALOGE("Couldn't initialize RS::dispatch->ScriptSetVarI");
+ return false;
+ }
+ RS::dispatch->ScriptSetVarObj = (ScriptSetVarObjFnPtr)dlsym(handle, "rsScriptSetVarObj");
+ if (RS::dispatch->ScriptSetVarObj == NULL) {
+ ALOGE("Couldn't initialize RS::dispatch->ScriptSetVarObj");
+ return false;
+ }
+ RS::dispatch->ScriptSetVarJ = (ScriptSetVarJFnPtr)dlsym(handle, "rsScriptSetVarJ");
+ if (RS::dispatch->ScriptSetVarJ == NULL) {
+ ALOGE("Couldn't initialize RS::dispatch->ScriptSetVarJ");
+ return false;
+ }
+ RS::dispatch->ScriptSetVarF = (ScriptSetVarFFnPtr)dlsym(handle, "rsScriptSetVarF");
+ if (RS::dispatch->ScriptSetVarF == NULL) {
+ ALOGE("Couldn't initialize RS::dispatch->ScriptSetVarF");
+ return false;
+ }
+ RS::dispatch->ScriptSetVarD = (ScriptSetVarDFnPtr)dlsym(handle, "rsScriptSetVarD");
+ if (RS::dispatch->ScriptSetVarD == NULL) {
+ ALOGE("Couldn't initialize RS::dispatch->ScriptSetVarD");
+ return false;
+ }
+ RS::dispatch->ScriptSetVarV = (ScriptSetVarVFnPtr)dlsym(handle, "rsScriptSetVarV");
+ if (RS::dispatch->ScriptSetVarV == NULL) {
+ ALOGE("Couldn't initialize RS::dispatch->ScriptSetVarV");
+ return false;
+ }
+ RS::dispatch->ScriptGetVarV = (ScriptGetVarVFnPtr)dlsym(handle, "rsScriptGetVarV");
+ if (RS::dispatch->ScriptGetVarV == NULL) {
+ ALOGE("Couldn't initialize RS::dispatch->ScriptGetVarV");
+ return false;
+ }
+ RS::dispatch->ScriptSetVarVE = (ScriptSetVarVEFnPtr)dlsym(handle, "rsScriptSetVarVE");
+ if (RS::dispatch->ScriptSetVarVE == NULL) {
+ ALOGE("Couldn't initialize RS::dispatch->ScriptSetVarVE");
+ return false;
+ }
+ RS::dispatch->ScriptCCreate = (ScriptCCreateFnPtr)dlsym(handle, "rsScriptCCreate");
+ if (RS::dispatch->ScriptCCreate == NULL) {
+ ALOGE("Couldn't initialize RS::dispatch->ScriptCCreate");
+ return false;
+ }
+ RS::dispatch->ScriptIntrinsicCreate = (ScriptIntrinsicCreateFnPtr)dlsym(handle, "rsScriptIntrinsicCreate");
+ if (RS::dispatch->ScriptIntrinsicCreate == NULL) {
+ ALOGE("Couldn't initialize RS::dispatch->ScriptIntrinsicCreate");
+ return false;
+ }
+ RS::dispatch->ScriptKernelIDCreate = (ScriptKernelIDCreateFnPtr)dlsym(handle, "rsScriptKernelIDCreate");
+ if (RS::dispatch->ScriptKernelIDCreate == NULL) {
+ ALOGE("Couldn't initialize RS::dispatch->ScriptKernelIDCreate");
+ return false;
+ }
+ RS::dispatch->ScriptFieldIDCreate = (ScriptFieldIDCreateFnPtr)dlsym(handle, "rsScriptFieldIDCreate");
+ if (RS::dispatch->ScriptFieldIDCreate == NULL) {
+ ALOGE("Couldn't initialize RS::dispatch->ScriptFieldIDCreate");
+ return false;
+ }
+ RS::dispatch->ScriptGroupCreate = (ScriptGroupCreateFnPtr)dlsym(handle, "rsScriptGroupCreate");
+ if (RS::dispatch->ScriptGroupCreate == NULL) {
+ ALOGE("Couldn't initialize RS::dispatch->ScriptGroupCreate");
+ return false;
+ }
+ RS::dispatch->ScriptGroupSetOutput = (ScriptGroupSetOutputFnPtr)dlsym(handle, "rsScriptGroupSetOutput");
+ if (RS::dispatch->ScriptGroupSetOutput == NULL) {
+ ALOGE("Couldn't initialize RS::dispatch->ScriptGroupSetOutput");
+ return false;
+ }
+ RS::dispatch->ScriptGroupSetInput = (ScriptGroupSetInputFnPtr)dlsym(handle, "rsScriptGroupSetInput");
+ if (RS::dispatch->ScriptGroupSetInput == NULL) {
+ ALOGE("Couldn't initialize RS::dispatch->ScriptGroupSetInput");
+ return false;
+ }
+ RS::dispatch->ScriptGroupExecute = (ScriptGroupExecuteFnPtr)dlsym(handle, "rsScriptGroupExecute");
+ if (RS::dispatch->ScriptGroupExecute == NULL) {
+ ALOGE("Couldn't initialize RS::dispatch->ScriptGroupExecute");
+ return false;
+ }
+ RS::dispatch->AllocationIoSend = (AllocationIoSendFnPtr)dlsym(handle, "rsAllocationIoSend");
+ if (RS::dispatch->AllocationIoSend == NULL) {
+ ALOGE("Couldn't initialize RS::dispatch->AllocationIoSend");
+ return false;
+ }
+ RS::dispatch->AllocationIoReceive = (AllocationIoReceiveFnPtr)dlsym(handle, "rsAllocationIoReceive");
+ if (RS::dispatch->AllocationIoReceive == NULL) {
+ ALOGE("Couldn't initialize RS::dispatch->AllocationIoReceive");
+ return false;
+ }
+
+ return true;
+}
+
+static bool loadSO(const char* filename) {
+ void* handle = dlopen(filename, RTLD_LAZY | RTLD_LOCAL);
+ if (handle == NULL) {
+ ALOGE("couldn't dlopen %s, %s", filename, dlerror());
+ return false;
+ }
+
+ if (loadSymbols(handle) == false) {
+ ALOGE("%s init failed!", filename);
+ return false;
+ }
+ ALOGE("Successfully loaded %s", filename);
+ return true;
+}
+
+static uint32_t getProp(const char *str) {
+#if !defined(RS_SERVER) && defined(HAVE_ANDROID_OS)
+ char buf[256];
+ property_get(str, buf, "0");
+ return atoi(buf);
+#else
+ return 0;
+#endif
+}
+
+bool RS::initDispatch(int targetApi) {
+ pthread_mutex_lock(&gInitMutex);
+ if (gInitError) {
+ goto error;
+ } else if (gInitialized) {
+ pthread_mutex_unlock(&gInitMutex);
+ return true;
+ }
+
+ RS::dispatch = new dispatchTable;
+
+ // attempt to load libRS, load libRSSupport on failure
+ // if property is set, proceed directly to libRSSupport
+ if (getProp("debug.rs.forcecompat") == 0) {
+ usingNative = loadSO("libRS.so");
+ }
+ if (usingNative == false) {
+ if (loadSO("libRSSupport.so") == false) {
+ ALOGE("Failed to load libRS.so and libRSSupport.so");
+ goto error;
+ }
+ }
+
+ gInitialized = true;
+
+ pthread_mutex_unlock(&gInitMutex);
+ return true;
+
+ error:
+ gInitError = 1;
+ pthread_mutex_unlock(&gInitMutex);
+ return false;
+}
+
bool RS::init(int targetApi, bool forceCpu, bool synchronous) {
- mDev = rsDeviceCreate();
+ if (initDispatch(targetApi) == false) {
+ ALOGE("Couldn't initialize dispatch table");
+ return false;
+ }
+
+ mDev = RS::dispatch->DeviceCreate();
if (mDev == 0) {
ALOGE("Device creation failed");
return false;
}
- mContext = rsContextCreate(mDev, 0, targetApi, RS_CONTEXT_TYPE_NORMAL, forceCpu, synchronous);
+ mContext = RS::dispatch->ContextCreate(mDev, 0, targetApi, RS_CONTEXT_TYPE_NORMAL, forceCpu, synchronous);
if (mContext == 0) {
ALOGE("Context creation failed");
return false;
@@ -80,6 +491,8 @@
usleep(1000);
}
+ mInit = true;
+
return true;
}
@@ -95,16 +508,16 @@
size_t rbuf_size = 256;
void * rbuf = malloc(rbuf_size);
- rsContextInitToClient(rs->mContext);
+ RS::dispatch->ContextInitToClient(rs->mContext);
rs->mMessageRun = true;
while (rs->mMessageRun) {
size_t receiveLen = 0;
uint32_t usrID = 0;
uint32_t subID = 0;
- RsMessageToClientType r = rsContextPeekMessage(rs->mContext,
- &receiveLen, sizeof(receiveLen),
- &usrID, sizeof(usrID));
+ RsMessageToClientType r = RS::dispatch->ContextPeekMessage(rs->mContext,
+ &receiveLen, sizeof(receiveLen),
+ &usrID, sizeof(usrID));
if (receiveLen >= rbuf_size) {
rbuf_size = receiveLen + 32;
@@ -114,7 +527,7 @@
ALOGE("RS::message handler realloc error %zu", rbuf_size);
// No clean way to recover now?
}
- rsContextGetMessage(rs->mContext, rbuf, rbuf_size, &receiveLen, sizeof(receiveLen),
+ RS::dispatch->ContextGetMessage(rs->mContext, rbuf, rbuf_size, &receiveLen, sizeof(receiveLen),
&subID, sizeof(subID));
switch(r) {
@@ -163,5 +576,5 @@
}
void RS::finish() {
- rsContextFinish(mContext);
+ RS::dispatch->ContextFinish(mContext);
}
diff --git a/cpp/RenderScript.h b/cpp/RenderScript.h
index 9a7a565..8ed8d0e 100644
--- a/cpp/RenderScript.h
+++ b/cpp/RenderScript.h
@@ -17,7 +17,6 @@
#ifndef ANDROID_RENDERSCRIPT_H
#define ANDROID_RENDERSCRIPT_H
-#include "rsDefines.h"
#include "rsCppStructs.h"
#ifdef RS_SERVER
diff --git a/cpp/Sampler.cpp b/cpp/Sampler.cpp
new file mode 100644
index 0000000..a1f4669
--- /dev/null
+++ b/cpp/Sampler.cpp
@@ -0,0 +1,73 @@
+/*
+ * Copyright (C) 2008-2012 The Android Open Source Project
+ *
+ * Licensed under the Apache License, Version 2.0 (the "License");
+ * you may not use this file except in compliance with the License.
+ * You may obtain a copy of the License at
+ *
+ * http://www.apache.org/licenses/LICENSE-2.0
+ *
+ * Unless required by applicable law or agreed to in writing, software
+ * distributed under the License is distributed on an "AS IS" BASIS,
+ * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ * See the License for the specific language governing permissions and
+ * limitations under the License.
+ */
+
+#include "RenderScript.h"
+
+using namespace android;
+using namespace RSC;
+
+Sampler::Sampler(sp<RS> rs, void* id):
+ BaseObj(id, rs)
+{
+ RsSamplerValue mMin = RS_SAMPLER_INVALID;
+ RsSamplerValue mMag = RS_SAMPLER_INVALID;
+ RsSamplerValue mWrapS = RS_SAMPLER_INVALID;;
+ RsSamplerValue mWrapT = RS_SAMPLER_INVALID;;
+ float mAniso = 0.f;
+}
+
+RsSamplerValue Sampler::getMinification() {
+ return mMin;
+}
+
+RsSamplerValue Sampler::getMagnification() {
+ return mMag;
+}
+
+RsSamplerValue Sampler::getWrapS() {
+ return mWrapS;
+}
+
+RsSamplerValue Sampler::getWrapT() {
+ return mWrapT;
+}
+
+float Sampler::getAnisotropy() {
+ return mAniso;
+}
+
+sp<Sampler> Sampler::create(sp<RS> rs, RsSamplerValue min, RsSamplerValue mag, RsSamplerValue wrapS, RsSamplerValue wrapT, float anisotropy) {
+ // we aren't supporting wrapR in C++ API atm, so always pass wrap for that
+ void* id = RS::dispatch->SamplerCreate(rs.get(), min, mag, wrapS, wrapT, RS_SAMPLER_WRAP, anisotropy);
+ return new Sampler(rs, id);
+}
+
+#define CREATE_SAMPLER(N, MIN, MAG, WRAPS, WRAPT) sp<const Sampler> Sampler::N(sp<RS> rs) { \
+ if (rs->mSamplers.N == NULL) { \
+ rs->mSamplers.N = (create(rs, MIN, MAG, WRAPS, WRAPT, 0.f)); \
+ } \
+ return rs->mSamplers.N; \
+ }
+
+CREATE_SAMPLER(CLAMP_NEAREST, RS_SAMPLER_CLAMP, RS_SAMPLER_CLAMP, RS_SAMPLER_NEAREST, RS_SAMPLER_NEAREST);
+CREATE_SAMPLER(CLAMP_LINEAR, RS_SAMPLER_CLAMP, RS_SAMPLER_CLAMP, RS_SAMPLER_LINEAR, RS_SAMPLER_LINEAR);
+CREATE_SAMPLER(CLAMP_LINEAR_MIP_LINEAR, RS_SAMPLER_CLAMP, RS_SAMPLER_CLAMP, RS_SAMPLER_LINEAR_MIP_LINEAR, RS_SAMPLER_LINEAR_MIP_LINEAR);
+CREATE_SAMPLER(WRAP_NEAREST, RS_SAMPLER_WRAP, RS_SAMPLER_WRAP, RS_SAMPLER_NEAREST, RS_SAMPLER_NEAREST);
+CREATE_SAMPLER(WRAP_LINEAR, RS_SAMPLER_WRAP, RS_SAMPLER_WRAP, RS_SAMPLER_LINEAR, RS_SAMPLER_LINEAR);
+CREATE_SAMPLER(WRAP_LINEAR_MIP_LINEAR, RS_SAMPLER_WRAP, RS_SAMPLER_WRAP, RS_SAMPLER_LINEAR_MIP_LINEAR, RS_SAMPLER_LINEAR_MIP_LINEAR);
+CREATE_SAMPLER(MIRRORED_REPEAT_NEAREST, RS_SAMPLER_MIRRORED_REPEAT, RS_SAMPLER_MIRRORED_REPEAT, RS_SAMPLER_NEAREST, RS_SAMPLER_NEAREST);
+CREATE_SAMPLER(MIRRORED_REPEAT_LINEAR, RS_SAMPLER_MIRRORED_REPEAT, RS_SAMPLER_MIRRORED_REPEAT, RS_SAMPLER_LINEAR, RS_SAMPLER_LINEAR);
+CREATE_SAMPLER(MIRRORED_REPEAT_LINEAR_MIP_LINEAR, RS_SAMPLER_MIRRORED_REPEAT, RS_SAMPLER_MIRRORED_REPEAT, RS_SAMPLER_LINEAR_MIP_LINEAR, RS_SAMPLER_LINEAR_MIP_LINEAR);
diff --git a/cpp/Script.cpp b/cpp/Script.cpp
index 54a571a..9bc55ff 100644
--- a/cpp/Script.cpp
+++ b/cpp/Script.cpp
@@ -17,13 +17,12 @@
#include <malloc.h>
#include "RenderScript.h"
-#include <rs.h>
using namespace android;
using namespace RSC;
void Script::invoke(uint32_t slot, const void *v, size_t len) const {
- rsScriptInvokeV(mRS->getContext(), getID(), slot, v, len);
+ RS::dispatch->ScriptInvokeV(mRS->getContext(), getID(), slot, v, len);
}
void Script::forEach(uint32_t slot, sp<const Allocation> ain, sp<const Allocation> aout,
@@ -33,7 +32,7 @@
}
void *in_id = BaseObj::getObjID(ain);
void *out_id = BaseObj::getObjID(aout);
- rsScriptForEach(mRS->getContext(), getID(), slot, in_id, out_id, usr, usrLen, NULL, 0);
+ RS::dispatch->ScriptForEach(mRS->getContext(), getID(), slot, in_id, out_id, usr, usrLen, NULL, 0);
}
@@ -42,16 +41,16 @@
void Script::bindAllocation(sp<Allocation> va, uint32_t slot) const {
- rsScriptBindAllocation(mRS->getContext(), getID(), BaseObj::getObjID(va), slot);
+ RS::dispatch->ScriptBindAllocation(mRS->getContext(), getID(), BaseObj::getObjID(va), slot);
}
void Script::setVar(uint32_t index, sp<const BaseObj> o) const {
- rsScriptSetVarObj(mRS->getContext(), getID(), index, (o == NULL) ? 0 : o->getID());
+ RS::dispatch->ScriptSetVarObj(mRS->getContext(), getID(), index, (o == NULL) ? 0 : o->getID());
}
void Script::setVar(uint32_t index, const void *v, size_t len) const {
- rsScriptSetVarV(mRS->getContext(), getID(), index, v, len);
+ RS::dispatch->ScriptSetVarV(mRS->getContext(), getID(), index, v, len);
}
void Script::FieldBase::init(sp<RS> rs, uint32_t dimx, uint32_t usages) {
diff --git a/cpp/ScriptC.cpp b/cpp/ScriptC.cpp
index f66e0ec..0d653bd 100644
--- a/cpp/ScriptC.cpp
+++ b/cpp/ScriptC.cpp
@@ -14,10 +14,7 @@
* limitations under the License.
*/
-#include <malloc.h>
-
#include "RenderScript.h"
-#include <rs.h>
using namespace android;
using namespace RSC;
@@ -27,7 +24,7 @@
const char *cachedName, size_t cachedNameLength,
const char *cacheDir, size_t cacheDirLength)
: Script(NULL, rs) {
- mID = rsScriptCCreate(rs->getContext(), cachedName, cachedNameLength,
- cacheDir, cacheDirLength, (const char *)codeTxt, codeLength);
+ mID = RS::dispatch->ScriptCCreate(rs->getContext(), cachedName, cachedNameLength,
+ cacheDir, cacheDirLength, (const char *)codeTxt, codeLength);
}
diff --git a/cpp/ScriptIntrinsics.cpp b/cpp/ScriptIntrinsics.cpp
index 44e8760..76328c4 100644
--- a/cpp/ScriptIntrinsics.cpp
+++ b/cpp/ScriptIntrinsics.cpp
@@ -17,20 +17,28 @@
#include <malloc.h>
#include "RenderScript.h"
-#include <rs.h>
-#include "rsDefines.h"
using namespace android;
using namespace RSC;
ScriptIntrinsic::ScriptIntrinsic(sp<RS> rs, int id, sp<const Element> e)
: Script(NULL, rs) {
- mID = rsScriptIntrinsicCreate(rs->getContext(), id, e->getID());
+ mID = RS::dispatch->ScriptIntrinsicCreate(rs->getContext(), id, e->getID());
+}
+
+ScriptIntrinsic3DLUT::ScriptIntrinsic3DLUT(sp<RS> rs, sp<const Element> e)
+ : ScriptIntrinsic(rs, RS_SCRIPT_INTRINSIC_ID_3DLUT, e) {
+
+}
+void ScriptIntrinsic3DLUT::forEach(sp<Allocation> ain, sp<Allocation> aout) {
+ Script::forEach(0, ain, aout, NULL, 0);
+}
+void ScriptIntrinsic3DLUT::setLUT(sp<Allocation> lut) {
+ Script::setVar(0, lut);
}
ScriptIntrinsicBlend::ScriptIntrinsicBlend(sp<RS> rs, sp<const Element> e)
: ScriptIntrinsic(rs, RS_SCRIPT_INTRINSIC_ID_BLEND, e) {
-
}
void ScriptIntrinsicBlend::blendClear(sp<Allocation> in, sp<Allocation> out) {
@@ -95,6 +103,9 @@
Script::forEach(35, in, out, NULL, 0);
}
+
+
+
ScriptIntrinsicBlur::ScriptIntrinsicBlur(sp<RS> rs, sp<const Element> e)
: ScriptIntrinsic(rs, RS_SCRIPT_INTRINSIC_ID_BLUR, e) {
@@ -108,3 +119,90 @@
void ScriptIntrinsicBlur::setRadius(float radius) {
Script::setVar(0, &radius, sizeof(float));
}
+
+
+
+ScriptIntrinsicColorMatrix::ScriptIntrinsicColorMatrix(sp<RS> rs, sp<const Element> e)
+ : ScriptIntrinsic(rs, RS_SCRIPT_INTRINSIC_ID_COLOR_MATRIX, e) {
+
+}
+
+void ScriptIntrinsicColorMatrix::forEach(sp<Allocation> in, sp<Allocation> out) {
+ Script::forEach(0, in, out, NULL, 0);
+}
+
+
+void ScriptIntrinsicColorMatrix::setColorMatrix3(float* m) {
+ Script::setVar(0, (void*)m, sizeof(float) * 9);
+}
+
+
+void ScriptIntrinsicColorMatrix::setColorMatrix4(float* m) {
+ Script::setVar(0, (void*)m, sizeof(float) * 16);
+}
+
+
+void ScriptIntrinsicColorMatrix::setGreyscale() {
+ float matrix[] = {0.299f, 0.587f, 0.114f, 0.299f, 0.587f, 0.114f, 0.299f, 0.587f, 0.114f};
+ setColorMatrix3(matrix);
+}
+
+
+void ScriptIntrinsicColorMatrix::setRGBtoYUV() {
+ float matrix[] = {0.299f,0.587f,0.114f,-0.14713f,-0.28886f,0.436f,0.615f,-0.51499f,-0.10001f};
+ setColorMatrix3(matrix);
+}
+
+
+void ScriptIntrinsicColorMatrix::setYUVtoRGB() {
+ float matrix[] = {1.f,0.f,1.13983f,1.f,-0.39465f,-0.5806f,1.f,2.03211f,0.f};
+ setColorMatrix3(matrix);
+}
+
+ScriptIntrinsicConvolve3x3::ScriptIntrinsicConvolve3x3(sp<RS> rs, sp<const Element> e)
+ : ScriptIntrinsic(rs, RS_SCRIPT_INTRINSIC_ID_CONVOLVE_3x3, e) {
+
+}
+
+void ScriptIntrinsicConvolve3x3::setInput(sp<Allocation> in) {
+ Script::setVar(1, in);
+}
+
+void ScriptIntrinsicConvolve3x3::forEach(sp<Allocation> out) {
+ Script::forEach(0, NULL, out, NULL, 0);
+}
+
+void ScriptIntrinsicConvolve3x3::setCoefficients(float* v) {
+ Script::setVar(0, (void*)v, sizeof(float) * 9);
+}
+
+ScriptIntrinsicConvolve5x5::ScriptIntrinsicConvolve5x5(sp<RS> rs, sp<const Element> e)
+ : ScriptIntrinsic(rs, RS_SCRIPT_INTRINSIC_ID_CONVOLVE_5x5, e) {
+
+}
+
+void ScriptIntrinsicConvolve5x5::setInput(sp<Allocation> in) {
+ Script::setVar(1, in);
+}
+
+void ScriptIntrinsicConvolve5x5::forEach(sp<Allocation> out) {
+ Script::forEach(0, NULL, out, NULL, 0);
+}
+
+void ScriptIntrinsicConvolve5x5::setCoefficients(float* v) {
+ Script::setVar(0, (void*)v, sizeof(float) * 25);
+}
+
+/*ScriptIntrinsicLUT::ScriptIntrinsicLUT(sp<RS> rs, sp<const Element> e)
+ : ScriptIntrinsic(rs, RS_SCRIPT_INTRINSIC_ID_LUT, e) {
+
+}
+
+void ScriptIntrinsicLUT::forEach(sp<Allocation> ain, sp<Allocation> aout) {
+
+}
+
+void ScriptIntrinsicLUT::setLUT(sp<Allocation> lut) {
+
+}*/
+
diff --git a/cpp/Type.cpp b/cpp/Type.cpp
index 312020a..192dc68 100644
--- a/cpp/Type.cpp
+++ b/cpp/Type.cpp
@@ -17,7 +17,6 @@
#include <malloc.h>
#include <string.h>
-#include <rs.h>
#include "RenderScript.h"
using namespace android;
@@ -93,7 +92,7 @@
}
sp<const Type> Type::create(sp<RS> rs, sp<const Element> e, uint32_t dimX, uint32_t dimY, uint32_t dimZ) {
- void * id = rsTypeCreate(rs->getContext(), e->getID(), dimX, dimY, dimZ, false, false, 0);
+ void * id = RS::dispatch->TypeCreate(rs->getContext(), e->getID(), dimX, dimY, dimZ, false, false, 0);
Type *t = new Type(id, rs);
t->mElement = e;
@@ -160,8 +159,8 @@
}
}
- void * id = rsTypeCreate(mRS->getContext(), mElement->getID(), mDimX, mDimY, mDimZ,
- mDimMipmaps, mDimFaces, 0);
+ void * id = RS::dispatch->TypeCreate(mRS->getContext(), mElement->getID(), mDimX, mDimY, mDimZ,
+ mDimMipmaps, mDimFaces, 0);
Type *t = new Type(id, mRS);
t->mElement = mElement;
t->mDimX = mDimX;
diff --git a/cpp/rsCppStructs.h b/cpp/rsCppStructs.h
index be011aa..b85499c 100644
--- a/cpp/rsCppStructs.h
+++ b/cpp/rsCppStructs.h
@@ -1,5 +1,5 @@
/*
- * Copyright (C) 2012 The Android Open Source Project
+ * Copyright (C) 2013 The Android Open Source Project
*
* Licensed under the Apache License, Version 2.0 (the "License");
* you may not use this file except in compliance with the License.
@@ -17,12 +17,12 @@
#ifndef ANDROID_RSCPPSTRUCTS_H
#define ANDROID_RSCPPSTRUCTS_H
+#include "rsDefines.h"
#include "rsCppUtils.h"
-#ifndef RS_SERVER
-#include "utils/RefBase.h"
-#else
-#include "RefBase.h"
-#endif
+#include "util/RefBase.h"
+#include "rsDispatch.h"
+
+#include <vector>
// Every row in an RS allocation is guaranteed to be aligned by this amount
// Every row in a user-backed allocation must be aligned by this amount
@@ -41,8 +41,9 @@
class Allocation;
class Script;
class ScriptC;
+class Sampler;
-class RS : public android::LightRefBase<RS> {
+ class RS : public android::RSC::LightRefBase<RS> {
public:
RS();
@@ -62,7 +63,12 @@
void finish();
+ static dispatchTable* dispatch;
+
private:
+ static bool usingNative;
+ static bool initDispatch(int targetApi);
+
bool init(int targetApi, bool forceCpu, bool synchronous);
static void * threadProc(void *);
@@ -78,86 +84,106 @@
ErrorHandlerFunc_t mErrorFunc;
MessageHandlerFunc_t mMessageFunc;
+ bool mInit;
struct {
- Element *U8;
- Element *I8;
- Element *U16;
- Element *I16;
- Element *U32;
- Element *I32;
- Element *U64;
- Element *I64;
- Element *F32;
- Element *F64;
- Element *BOOLEAN;
+ sp<const Element> U8;
+ sp<const Element> I8;
+ sp<const Element> U16;
+ sp<const Element> I16;
+ sp<const Element> U32;
+ sp<const Element> I32;
+ sp<const Element> U64;
+ sp<const Element> I64;
+ sp<const Element> F32;
+ sp<const Element> F64;
+ sp<const Element> BOOLEAN;
- Element *ELEMENT;
- Element *TYPE;
- Element *ALLOCATION;
- Element *SAMPLER;
- Element *SCRIPT;
- Element *MESH;
- Element *PROGRAM_FRAGMENT;
- Element *PROGRAM_VERTEX;
- Element *PROGRAM_RASTER;
- Element *PROGRAM_STORE;
+ sp<const Element> ELEMENT;
+ sp<const Element> TYPE;
+ sp<const Element> ALLOCATION;
+ sp<const Element> SAMPLER;
+ sp<const Element> SCRIPT;
+ sp<const Element> MESH;
+ sp<const Element> PROGRAM_FRAGMENT;
+ sp<const Element> PROGRAM_VERTEX;
+ sp<const Element> PROGRAM_RASTER;
+ sp<const Element> PROGRAM_STORE;
- Element *A_8;
- Element *RGB_565;
- Element *RGB_888;
- Element *RGBA_5551;
- Element *RGBA_4444;
- Element *RGBA_8888;
+ sp<const Element> A_8;
+ sp<const Element> RGB_565;
+ sp<const Element> RGB_888;
+ sp<const Element> RGBA_5551;
+ sp<const Element> RGBA_4444;
+ sp<const Element> RGBA_8888;
- Element *FLOAT_2;
- Element *FLOAT_3;
- Element *FLOAT_4;
+ sp<const Element> FLOAT_2;
+ sp<const Element> FLOAT_3;
+ sp<const Element> FLOAT_4;
- Element *DOUBLE_2;
- Element *DOUBLE_3;
- Element *DOUBLE_4;
+ sp<const Element> DOUBLE_2;
+ sp<const Element> DOUBLE_3;
+ sp<const Element> DOUBLE_4;
- Element *UCHAR_2;
- Element *UCHAR_3;
- Element *UCHAR_4;
+ sp<const Element> UCHAR_2;
+ sp<const Element> UCHAR_3;
+ sp<const Element> UCHAR_4;
- Element *CHAR_2;
- Element *CHAR_3;
- Element *CHAR_4;
+ sp<const Element> CHAR_2;
+ sp<const Element> CHAR_3;
+ sp<const Element> CHAR_4;
- Element *USHORT_2;
- Element *USHORT_3;
- Element *USHORT_4;
+ sp<const Element> USHORT_2;
+ sp<const Element> USHORT_3;
+ sp<const Element> USHORT_4;
- Element *SHORT_2;
- Element *SHORT_3;
- Element *SHORT_4;
+ sp<const Element> SHORT_2;
+ sp<const Element> SHORT_3;
+ sp<const Element> SHORT_4;
- Element *UINT_2;
- Element *UINT_3;
- Element *UINT_4;
+ sp<const Element> UINT_2;
+ sp<const Element> UINT_3;
+ sp<const Element> UINT_4;
- Element *INT_2;
- Element *INT_3;
- Element *INT_4;
+ sp<const Element> INT_2;
+ sp<const Element> INT_3;
+ sp<const Element> INT_4;
- Element *ULONG_2;
- Element *ULONG_3;
- Element *ULONG_4;
+ sp<const Element> ULONG_2;
+ sp<const Element> ULONG_3;
+ sp<const Element> ULONG_4;
- Element *LONG_2;
- Element *LONG_3;
- Element *LONG_4;
+ sp<const Element> LONG_2;
+ sp<const Element> LONG_3;
+ sp<const Element> LONG_4;
- Element *MATRIX_4X4;
- Element *MATRIX_3X3;
- Element *MATRIX_2X2;
+ sp<const Element> MATRIX_4X4;
+ sp<const Element> MATRIX_3X3;
+ sp<const Element> MATRIX_2X2;
} mElements;
+ struct {
+ sp<const Sampler> CLAMP_NEAREST;
+ sp<const Sampler> CLAMP_LINEAR;
+ sp<const Sampler> CLAMP_LINEAR_MIP_LINEAR;
+ sp<const Sampler> WRAP_NEAREST;
+ sp<const Sampler> WRAP_LINEAR;
+ sp<const Sampler> WRAP_LINEAR_MIP_LINEAR;
+ sp<const Sampler> MIRRORED_REPEAT_NEAREST;
+ sp<const Sampler> MIRRORED_REPEAT_LINEAR;
+ sp<const Sampler> MIRRORED_REPEAT_LINEAR_MIP_LINEAR;
+ } mSamplers;
+ friend class Sampler;
+ friend class Element;
};
-class BaseObj : public android::LightRefBase<BaseObj> {
+class BaseObj : public android::RSC::LightRefBase<BaseObj> {
+public:
+ void * getID() const;
+ virtual ~BaseObj();
+ virtual void updateFromNative();
+ virtual bool equals(sp<const BaseObj> obj);
+
protected:
void *mID;
sp<RS> mRS;
@@ -168,20 +194,14 @@
static void * getObjID(sp<const BaseObj> o);
-public:
-
- void * getID() const;
- virtual ~BaseObj();
- virtual void updateFromNative();
- virtual bool equals(const BaseObj *obj);
};
class Allocation : public BaseObj {
protected:
- android::sp<const Type> mType;
+ sp<const Type> mType;
uint32_t mUsage;
- android::sp<Allocation> mAdaptedAllocation;
+ sp<Allocation> mAdaptedAllocation;
bool mConstrainedLOD;
bool mConstrainedFace;
@@ -215,7 +235,7 @@
void validate2DRange(uint32_t xoff, uint32_t yoff, uint32_t w, uint32_t h);
public:
- android::sp<const Type> getType() {
+ sp<const Type> getType() {
return mType;
}
@@ -355,43 +375,44 @@
static sp<const Element> MATRIX_3X3(sp<RS> rs);
static sp<const Element> MATRIX_2X2(sp<RS> rs);
- Element(void *id, sp<RS> rs,
- android::Vector<sp<Element> > &elements,
- android::Vector<android::String8> &elementNames,
- android::Vector<uint32_t> &arraySizes);
- Element(void *id, sp<RS> rs, RsDataType dt, RsDataKind dk, bool norm, uint32_t size);
- Element(sp<RS> rs);
- virtual ~Element();
-
void updateFromNative();
static sp<const Element> createUser(sp<RS> rs, RsDataType dt);
static sp<const Element> createVector(sp<RS> rs, RsDataType dt, uint32_t size);
static sp<const Element> createPixel(sp<RS> rs, RsDataType dt, RsDataKind dk);
bool isCompatible(sp<const Element>e);
+ Element(void *id, sp<RS> rs,
+ std::vector<sp<Element> > &elements,
+ std::vector<android::String8> &elementNames,
+ std::vector<uint32_t> &arraySizes);
+ Element(void *id, sp<RS> rs, RsDataType dt, RsDataKind dk, bool norm, uint32_t size);
+ Element(sp<RS> rs);
+ virtual ~Element();
+
+ protected:
class Builder {
private:
sp<RS> mRS;
- android::Vector<sp<Element> > mElements;
- android::Vector<android::String8> mElementNames;
- android::Vector<uint32_t> mArraySizes;
+ std::vector<sp<Element> > mElements;
+ std::vector<android::String8> mElementNames;
+ std::vector<uint32_t> mArraySizes;
bool mSkipPadding;
public:
Builder(sp<RS> rs);
~Builder();
- void add(sp<Element>, android::String8 &name, uint32_t arraySize = 1);
+ void add(sp<Element> e, android::String8 &name, uint32_t arraySize = 1);
sp<const Element> create();
};
private:
void updateVisibleSubElements();
- android::Vector<sp</*const*/ Element> > mElements;
- android::Vector<android::String8> mElementNames;
- android::Vector<uint32_t> mArraySizes;
- android::Vector<uint32_t> mVisibleElementMap;
- android::Vector<uint32_t> mOffsetInBytes;
+ std::vector<sp<Element> > mElements;
+ std::vector<android::String8> mElementNames;
+ std::vector<uint32_t> mArraySizes;
+ std::vector<uint32_t> mVisibleElementMap;
+ std::vector<uint32_t> mOffsetInBytes;
RsDataType mType;
RsDataKind mKind;
@@ -408,10 +429,9 @@
public:
FieldPacker(size_t len)
- : mPos(0),
- mLen(len) {
- mData = new unsigned char[len];
- }
+ : mPos(0), mLen(len) {
+ mData = new unsigned char[len];
+ }
virtual ~FieldPacker() {
delete [] mData;
@@ -458,7 +478,7 @@
}
template <typename T>
- void add(T t) {
+ void add(T t) {
align(sizeof(t));
if (mPos + sizeof(t) <= mLen) {
memcpy(&mData[mPos], &t, sizeof(t));
@@ -467,26 +487,26 @@
}
/*
- void add(rs_matrix4x4 m) {
- for (size_t i = 0; i < 16; i++) {
- add(m.m[i]);
- }
- }
+ void add(rs_matrix4x4 m) {
+ for (size_t i = 0; i < 16; i++) {
+ add(m.m[i]);
+ }
+ }
- void add(rs_matrix3x3 m) {
- for (size_t i = 0; i < 9; i++) {
- add(m.m[i]);
- }
- }
+ void add(rs_matrix3x3 m) {
+ for (size_t i = 0; i < 9; i++) {
+ add(m.m[i]);
+ }
+ }
- void add(rs_matrix2x2 m) {
- for (size_t i = 0; i < 4; i++) {
- add(m.m[i]);
- }
- }
+ void add(rs_matrix2x2 m) {
+ for (size_t i = 0; i < 4; i++) {
+ add(m.m[i]);
+ }
+ }
*/
- void add(BaseObj* obj) {
+ void add(sp<BaseObj> obj) {
if (obj != NULL) {
add((uint32_t) (uintptr_t) obj->getID());
} else {
@@ -495,6 +515,7 @@
}
};
+
class Type : public BaseObj {
protected:
friend class Allocation;
@@ -641,9 +662,16 @@
ScriptIntrinsic(sp<RS> rs, int id, sp<const Element> e);
};
+class ScriptIntrinsic3DLUT : public ScriptIntrinsic {
+ public:
+ ScriptIntrinsic3DLUT(sp<RS> rs, sp<const Element> e);
+ void forEach(sp<Allocation> ain, sp<Allocation> aout);
+ void setLUT(sp<Allocation> lut);
+};
+
class ScriptIntrinsicBlend : public ScriptIntrinsic {
public:
- ScriptIntrinsicBlend(sp<RS> rs, sp <const Element> e);
+ ScriptIntrinsicBlend(sp<RS> rs, sp<const Element> e);
void blendClear(sp<Allocation> in, sp<Allocation> out);
void blendSrc(sp<Allocation> in, sp<Allocation> out);
void blendDst(sp<Allocation> in, sp<Allocation> out);
@@ -663,11 +691,81 @@
class ScriptIntrinsicBlur : public ScriptIntrinsic {
public:
- ScriptIntrinsicBlur(sp<RS> rs, sp <const Element> e);
+ ScriptIntrinsicBlur(sp<RS> rs, sp<const Element> e);
void blur(sp<Allocation> in, sp<Allocation> out);
void setRadius(float radius);
};
+class ScriptIntrinsicColorMatrix : public ScriptIntrinsic {
+ public:
+ ScriptIntrinsicColorMatrix(sp<RS> rs, sp<const Element> e);
+ void forEach(sp<Allocation> in, sp<Allocation> out);
+ void setColorMatrix3(float* m);
+ void setColorMatrix4(float* m);
+ void setGreyscale();
+ void setRGBtoYUV();
+ void setYUVtoRGB();
+};
+
+class ScriptIntrinsicConvolve3x3 : public ScriptIntrinsic {
+ public:
+ ScriptIntrinsicConvolve3x3(sp<RS> rs, sp<const Element> e);
+ void setInput(sp<Allocation> in);
+ void forEach(sp<Allocation> out);
+ void setCoefficients(float* v);
+};
+
+class ScriptIntrinsicConvolve5x5 : public ScriptIntrinsic {
+ public:
+ ScriptIntrinsicConvolve5x5(sp<RS> rs, sp<const Element> e);
+ void setInput(sp<Allocation> in);
+ void forEach(sp<Allocation> out);
+ void setCoefficients(float* v);
+};
+
+/*class ScriptIntrinsicLUT : public ScriptIntrinsic {
+ public:
+ ScriptIntrinsicLUT(sp<RS> rs, sp<const Element> e);
+ void forEach(sp<Allocation> ain, sp<Allocation> aout);
+ void setLUT(sp<Allocation> lut);
+ };*/
+/*
+class ScriptIntrinsicYuvToRGB : public ScriptIntrinsic {
+
+};
+*/
+
+ class Sampler : public BaseObj {
+ private:
+ Sampler(sp<RS> rs, void* id);
+ RsSamplerValue mMin;
+ RsSamplerValue mMag;
+ RsSamplerValue mWrapS;
+ RsSamplerValue mWrapT;
+ RsSamplerValue mWrapR;
+ float mAniso;
+
+ public:
+ static sp<Sampler> create(sp<RS> rs, RsSamplerValue min, RsSamplerValue mag, RsSamplerValue wrapS, RsSamplerValue wrapT, float anisotropy);
+
+ RsSamplerValue getMinification();
+ RsSamplerValue getMagnification();
+ RsSamplerValue getWrapS();
+ RsSamplerValue getWrapT();
+ float getAnisotropy();
+
+ sp<const Sampler> CLAMP_NEAREST(sp<RS> rs);
+ sp<const Sampler> CLAMP_LINEAR(sp<RS> rs);
+ sp<const Sampler> CLAMP_LINEAR_MIP_LINEAR(sp<RS> rs);
+ sp<const Sampler> WRAP_NEAREST(sp<RS> rs);
+ sp<const Sampler> WRAP_LINEAR(sp<RS> rs);
+ sp<const Sampler> WRAP_LINEAR_MIP_LINEAR(sp<RS> rs);
+ sp<const Sampler> MIRRORED_REPEAT_NEAREST(sp<RS> rs);
+ sp<const Sampler> MIRRORED_REPEAT_LINEAR(sp<RS> rs);
+ sp<const Sampler> MIRRORED_REPEAT_LINEAR_MIP_LINEAR(sp<RS> rs);
+
+};
+
}
}
diff --git a/cpp/rsDispatch.h b/cpp/rsDispatch.h
new file mode 100644
index 0000000..139ef52
--- /dev/null
+++ b/cpp/rsDispatch.h
@@ -0,0 +1,161 @@
+/*
+ * Copyright (C) 2012 The Android Open Source Project
+ *
+ * Licensed under the Apache License, Version 2.0 (the "License");
+ * you may not use this file except in compliance with the License.
+ * You may obtain a copy of the License at
+ *
+ * http://www.apache.org/licenses/LICENSE-2.0
+ *
+ * Unless required by applicable law or agreed to in writing, software
+ * distributed under the License is distributed on an "AS IS" BASIS,
+ * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ * See the License for the specific language governing permissions and
+ * limitations under the License.
+ */
+
+#ifndef ANDROID_RSDISPATCH_H
+#define ANDROID_RSDISPATCH_H
+
+#include "rsDefines.h"
+
+typedef const void* (*AllocationGetTypeFnPtr)(RsContext con, RsAllocation va);
+typedef void (*TypeGetNativeDataFnPtr)(RsContext, RsType, uint32_t *typeData, uint32_t typeDataSize);
+typedef void (*ElementGetNativeDataFnPtr)(RsContext, RsElement, uint32_t *elemData, uint32_t elemDataSize);
+typedef void (*ElementGetSubElementsFnPtr)(RsContext, RsElement, uint32_t *ids, const char **names, uint32_t *arraySizes, uint32_t dataSize);
+typedef RsDevice (*DeviceCreateFnPtr) ();
+typedef void (*DeviceDestroyFnPtr) (RsDevice dev);
+typedef void (*DeviceSetConfigFnPtr) (RsDevice dev, RsDeviceParam p, int32_t value);
+typedef RsContext (*ContextCreateFnPtr)(RsDevice vdev, uint32_t version, uint32_t sdkVersion, RsContextType ct, bool forceCpu, bool synchronous);
+typedef void (*GetNameFnPtr)(RsContext, void * obj, const char **name);
+
+typedef void (*ContextDestroyFnPtr) (RsContext);
+typedef RsMessageToClientType (*ContextGetMessageFnPtr) (RsContext, void*, size_t, size_t*, size_t, uint32_t*, size_t);
+typedef RsMessageToClientType (*ContextPeekMessageFnPtr) (RsContext, size_t*, size_t, uint32_t*, size_t);
+typedef void (*ContextSendMessageFnPtr) (RsContext, uint32_t, const uint8_t*, size_t);
+typedef void (*ContextInitToClientFnPtr) (RsContext);
+typedef void (*ContextDeinitToClientFnPtr) (RsContext);
+typedef RsType (*TypeCreateFnPtr) (RsContext, RsElement, uint32_t, uint32_t, uint32_t, bool, bool, uint32_t);
+typedef RsAllocation (*AllocationCreateTypedFnPtr) (RsContext, RsType, RsAllocationMipmapControl, uint32_t, uintptr_t);
+typedef RsAllocation (*AllocationCreateFromBitmapFnPtr) (RsContext, RsType, RsAllocationMipmapControl, const void*, size_t, uint32_t);
+typedef RsAllocation (*AllocationCubeCreateFromBitmapFnPtr) (RsContext, RsType, RsAllocationMipmapControl, const void*, size_t, uint32_t);
+typedef RsNativeWindow (*AllocationGetSurfaceFnPtr) (RsContext, RsAllocation);
+typedef void (*AllocationSetSurfaceFnPtr) (RsContext, RsAllocation, RsNativeWindow);
+typedef void (*ContextFinishFnPtr) (RsContext);
+typedef void (*ContextDumpFnPtr) (RsContext, int32_t);
+typedef void (*ContextSetPriorityFnPtr) (RsContext, int32_t);
+typedef void (*AssignNameFnPtr) (RsContext, RsObjectBase, const char*, size_t);
+typedef void (*ObjDestroyFnPtr) (RsContext, RsAsyncVoidPtr);
+typedef RsElement (*ElementCreateFnPtr) (RsContext, RsDataType, RsDataKind, bool, uint32_t);
+typedef RsElement (*ElementCreate2FnPtr) (RsContext, const RsElement*, size_t, const char**, size_t, const size_t*, const uint32_t*, size_t);
+typedef void (*AllocationCopyToBitmapFnPtr) (RsContext, RsAllocation, void*, size_t);
+typedef void (*Allocation1DDataFnPtr) (RsContext, RsAllocation, uint32_t, uint32_t, uint32_t, const void*, size_t);
+typedef void (*Allocation1DElementDataFnPtr) (RsContext, RsAllocation, uint32_t, uint32_t, const void*, size_t, size_t);
+typedef void (*Allocation2DDataFnPtr) (RsContext, RsAllocation, uint32_t, uint32_t, uint32_t, RsAllocationCubemapFace, uint32_t, uint32_t, const void*, size_t, size_t);
+typedef void (*Allocation3DDataFnPtr) (RsContext, RsAllocation, uint32_t, uint32_t, uint32_t, uint32_t, uint32_t, uint32_t, uint32_t, const void*, size_t, size_t);
+typedef void (*AllocationGenerateMipmapsFnPtr) (RsContext, RsAllocation);
+typedef void (*AllocationReadFnPtr) (RsContext, RsAllocation, void*, size_t);
+typedef void (*Allocation1DReadFnPtr) (RsContext, RsAllocation, uint32_t, uint32_t, uint32_t, void*, size_t);
+typedef void (*Allocation2DReadFnPtr) (RsContext, RsAllocation, uint32_t, uint32_t, uint32_t, RsAllocationCubemapFace, uint32_t, uint32_t, void*, size_t, size_t);
+typedef void (*AllocationSyncAllFnPtr) (RsContext, RsAllocation, RsAllocationUsageType);
+typedef void (*AllocationResize1DFnPtr) (RsContext, RsAllocation, uint32_t);
+typedef void (*AllocationCopy2DRangeFnPtr) (RsContext, RsAllocation, uint32_t, uint32_t, uint32_t, uint32_t, uint32_t, uint32_t, RsAllocation, uint32_t, uint32_t, uint32_t, uint32_t);
+typedef void (*AllocationCopy3DRangeFnPtr) (RsContext, RsAllocation, uint32_t, uint32_t, uint32_t, uint32_t, uint32_t, uint32_t, uint32_t, RsAllocation, uint32_t, uint32_t, uint32_t, uint32_t);
+typedef RsSampler (*SamplerCreateFnPtr) (RsContext, RsSamplerValue, RsSamplerValue, RsSamplerValue, RsSamplerValue, RsSamplerValue, float);
+typedef void (*ScriptBindAllocationFnPtr) (RsContext, RsScript, RsAllocation, uint32_t);
+typedef void (*ScriptSetTimeZoneFnPtr) (RsContext, RsScript, const char*, size_t);
+typedef void (*ScriptInvokeFnPtr) (RsContext, RsScript, uint32_t);
+typedef void (*ScriptInvokeVFnPtr) (RsContext, RsScript, uint32_t, const void*, size_t);
+typedef void (*ScriptForEachFnPtr) (RsContext, RsScript, uint32_t, RsAllocation, RsAllocation, const void*, size_t, const RsScriptCall*, size_t);
+typedef void (*ScriptSetVarIFnPtr) (RsContext, RsScript, uint32_t, int);
+typedef void (*ScriptSetVarObjFnPtr) (RsContext, RsScript, uint32_t, RsObjectBase);
+typedef void (*ScriptSetVarJFnPtr) (RsContext, RsScript, uint32_t, int64_t);
+typedef void (*ScriptSetVarFFnPtr) (RsContext, RsScript, uint32_t, float);
+typedef void (*ScriptSetVarDFnPtr) (RsContext, RsScript, uint32_t, double);
+typedef void (*ScriptSetVarVFnPtr) (RsContext, RsScript, uint32_t, const void*, size_t);
+typedef void (*ScriptGetVarVFnPtr) (RsContext, RsScript, uint32_t, void*, size_t);
+typedef void (*ScriptSetVarVEFnPtr) (RsContext, RsScript, uint32_t, const void*, size_t, RsElement, const size_t*, size_t);
+typedef RsScript (*ScriptCCreateFnPtr) (RsContext, const char*, size_t, const char*, size_t, const char*, size_t);
+typedef RsScript (*ScriptIntrinsicCreateFnPtr) (RsContext, uint32_t id, RsElement);
+typedef RsScriptKernelID (*ScriptKernelIDCreateFnPtr) (RsContext, RsScript, int, int);
+typedef RsScriptFieldID (*ScriptFieldIDCreateFnPtr) (RsContext, RsScript, int);
+typedef RsScriptGroup (*ScriptGroupCreateFnPtr) (RsContext, RsScriptKernelID*, size_t, RsScriptKernelID*, size_t, RsScriptKernelID*, size_t, RsScriptFieldID*, size_t, const RsType*, size_t);
+typedef void (*ScriptGroupSetOutputFnPtr) (RsContext, RsScriptGroup, RsScriptKernelID, RsAllocation);
+typedef void (*ScriptGroupSetInputFnPtr) (RsContext, RsScriptGroup, RsScriptKernelID, RsAllocation);
+typedef void (*ScriptGroupExecuteFnPtr) (RsContext, RsScriptGroup);
+typedef void (*AllocationIoSendFnPtr) (RsContext, RsAllocation);
+typedef void (*AllocationIoReceiveFnPtr) (RsContext, RsAllocation);
+
+typedef struct {
+ // inserted by hand from rs.h
+ AllocationGetTypeFnPtr AllocationGetType;
+ TypeGetNativeDataFnPtr TypeGetNativeData;
+ ElementGetNativeDataFnPtr ElementGetNativeData;
+ ElementGetSubElementsFnPtr ElementGetSubElements;
+
+ DeviceCreateFnPtr DeviceCreate;
+ DeviceDestroyFnPtr DeviceDestroy;
+ DeviceSetConfigFnPtr DeviceSetConfig;
+ ContextCreateFnPtr ContextCreate;
+ GetNameFnPtr GetName;
+
+ // generated from rs.spec
+ ContextDestroyFnPtr ContextDestroy;
+ ContextGetMessageFnPtr ContextGetMessage;
+ ContextPeekMessageFnPtr ContextPeekMessage;
+ ContextSendMessageFnPtr ContextSendMessage;
+ ContextInitToClientFnPtr ContextInitToClient;
+ ContextDeinitToClientFnPtr ContextDeinitToClient;
+ TypeCreateFnPtr TypeCreate;
+ AllocationCreateTypedFnPtr AllocationCreateTyped;
+ AllocationCreateFromBitmapFnPtr AllocationCreateFromBitmap;
+ AllocationCubeCreateFromBitmapFnPtr AllocationCubeCreateFromBitmap;
+ AllocationGetSurfaceFnPtr AllocationGetSurface;
+ AllocationSetSurfaceFnPtr AllocationSetSurface;
+ ContextFinishFnPtr ContextFinish;
+ ContextDumpFnPtr ContextDump;
+ ContextSetPriorityFnPtr ContextSetPriority;
+ AssignNameFnPtr AssignName;
+ ObjDestroyFnPtr ObjDestroy;
+ ElementCreateFnPtr ElementCreate;
+ ElementCreate2FnPtr ElementCreate2;
+ AllocationCopyToBitmapFnPtr AllocationCopyToBitmap;
+ Allocation1DDataFnPtr Allocation1DData;
+ Allocation1DElementDataFnPtr Allocation1DElementData;
+ Allocation2DDataFnPtr Allocation2DData;
+ Allocation3DDataFnPtr Allocation3DData;
+ AllocationGenerateMipmapsFnPtr AllocationGenerateMipmaps;
+ AllocationReadFnPtr AllocationRead;
+ Allocation1DReadFnPtr Allocation1DRead;
+ Allocation2DReadFnPtr Allocation2DRead;
+ AllocationSyncAllFnPtr AllocationSyncAll;
+ AllocationResize1DFnPtr AllocationResize1D;
+ AllocationCopy2DRangeFnPtr AllocationCopy2DRange;
+ AllocationCopy3DRangeFnPtr AllocationCopy3DRange;
+ SamplerCreateFnPtr SamplerCreate;
+ ScriptBindAllocationFnPtr ScriptBindAllocation;
+ ScriptSetTimeZoneFnPtr ScriptSetTimeZone;
+ ScriptInvokeFnPtr ScriptInvoke;
+ ScriptInvokeVFnPtr ScriptInvokeV;
+ ScriptForEachFnPtr ScriptForEach;
+ ScriptSetVarIFnPtr ScriptSetVarI;
+ ScriptSetVarObjFnPtr ScriptSetVarObj;
+ ScriptSetVarJFnPtr ScriptSetVarJ;
+ ScriptSetVarFFnPtr ScriptSetVarF;
+ ScriptSetVarDFnPtr ScriptSetVarD;
+ ScriptSetVarVFnPtr ScriptSetVarV;
+ ScriptGetVarVFnPtr ScriptGetVarV;
+ ScriptSetVarVEFnPtr ScriptSetVarVE;
+ ScriptCCreateFnPtr ScriptCCreate;
+ ScriptIntrinsicCreateFnPtr ScriptIntrinsicCreate;
+ ScriptKernelIDCreateFnPtr ScriptKernelIDCreate;
+ ScriptFieldIDCreateFnPtr ScriptFieldIDCreate;
+ ScriptGroupCreateFnPtr ScriptGroupCreate;
+ ScriptGroupSetOutputFnPtr ScriptGroupSetOutput;
+ ScriptGroupSetInputFnPtr ScriptGroupSetInput;
+ ScriptGroupExecuteFnPtr ScriptGroupExecute;
+ AllocationIoSendFnPtr AllocationIoSend;
+ AllocationIoReceiveFnPtr AllocationIoReceive;
+} dispatchTable;
+
+#endif
diff --git a/cpp/util/RefBase.h b/cpp/util/RefBase.h
new file mode 100644
index 0000000..eab26ab
--- /dev/null
+++ b/cpp/util/RefBase.h
@@ -0,0 +1,529 @@
+/*
+ * Copyright (C) 2013 The Android Open Source Project
+ *
+ * Licensed under the Apache License, Version 2.0 (the "License");
+ * you may not use this file except in compliance with the License.
+ * You may obtain a copy of the License at
+ *
+ * http://www.apache.org/licenses/LICENSE-2.0
+ *
+ * Unless required by applicable law or agreed to in writing, software
+ * distributed under the License is distributed on an "AS IS" BASIS,
+ * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ * See the License for the specific language governing permissions and
+ * limitations under the License.
+ */
+
+#ifndef RS_REF_BASE_H
+#define RS_REF_BASE_H
+
+
+#include <stdint.h>
+#include <sys/types.h>
+#include <stdlib.h>
+#include <string.h>
+
+#include "StrongPointer.h"
+#include "TypeHelpers.h"
+
+// ---------------------------------------------------------------------------
+namespace android{
+namespace RSC {
+
+class TextOutput;
+TextOutput& printWeakPointer(TextOutput& to, const void* val);
+
+// ---------------------------------------------------------------------------
+
+#define COMPARE_WEAK(_op_) \
+inline bool operator _op_ (const sp<T>& o) const { \
+ return m_ptr _op_ o.m_ptr; \
+} \
+inline bool operator _op_ (const T* o) const { \
+ return m_ptr _op_ o; \
+} \
+template<typename U> \
+inline bool operator _op_ (const sp<U>& o) const { \
+ return m_ptr _op_ o.m_ptr; \
+} \
+template<typename U> \
+inline bool operator _op_ (const U* o) const { \
+ return m_ptr _op_ o; \
+}
+
+// ---------------------------------------------------------------------------
+class ReferenceMover;
+class ReferenceConverterBase {
+public:
+ virtual size_t getReferenceTypeSize() const = 0;
+ virtual void* getReferenceBase(void const*) const = 0;
+ inline virtual ~ReferenceConverterBase() { }
+};
+
+// ---------------------------------------------------------------------------
+
+class RefBase
+{
+public:
+ void incStrong(const void* id) const;
+ void decStrong(const void* id) const;
+
+ void forceIncStrong(const void* id) const;
+
+ //! DEBUGGING ONLY: Get current strong ref count.
+ int32_t getStrongCount() const;
+
+ class weakref_type
+ {
+ public:
+ RefBase* refBase() const;
+
+ void incWeak(const void* id);
+ void decWeak(const void* id);
+
+ // acquires a strong reference if there is already one.
+ bool attemptIncStrong(const void* id);
+
+ // acquires a weak reference if there is already one.
+ // This is not always safe. see ProcessState.cpp and BpBinder.cpp
+ // for proper use.
+ bool attemptIncWeak(const void* id);
+
+ //! DEBUGGING ONLY: Get current weak ref count.
+ int32_t getWeakCount() const;
+
+ //! DEBUGGING ONLY: Print references held on object.
+ void printRefs() const;
+
+ //! DEBUGGING ONLY: Enable tracking for this object.
+ // enable -- enable/disable tracking
+ // retain -- when tracking is enable, if true, then we save a stack trace
+ // for each reference and dereference; when retain == false, we
+ // match up references and dereferences and keep only the
+ // outstanding ones.
+
+ void trackMe(bool enable, bool retain);
+ };
+
+ weakref_type* createWeak(const void* id) const;
+
+ weakref_type* getWeakRefs() const;
+
+ //! DEBUGGING ONLY: Print references held on object.
+ inline void printRefs() const { getWeakRefs()->printRefs(); }
+
+ //! DEBUGGING ONLY: Enable tracking of object.
+ inline void trackMe(bool enable, bool retain)
+ {
+ getWeakRefs()->trackMe(enable, retain);
+ }
+
+ typedef RefBase basetype;
+
+protected:
+ RefBase();
+ virtual ~RefBase();
+
+ //! Flags for extendObjectLifetime()
+ enum {
+ OBJECT_LIFETIME_STRONG = 0x0000,
+ OBJECT_LIFETIME_WEAK = 0x0001,
+ OBJECT_LIFETIME_MASK = 0x0001
+ };
+
+ void extendObjectLifetime(int32_t mode);
+
+ //! Flags for onIncStrongAttempted()
+ enum {
+ FIRST_INC_STRONG = 0x0001
+ };
+
+ virtual void onFirstRef();
+ virtual void onLastStrongRef(const void* id);
+ virtual bool onIncStrongAttempted(uint32_t flags, const void* id);
+ virtual void onLastWeakRef(const void* id);
+
+private:
+ friend class ReferenceMover;
+ static void moveReferences(void* d, void const* s, size_t n,
+ const ReferenceConverterBase& caster);
+
+private:
+ friend class weakref_type;
+ class weakref_impl;
+
+ RefBase(const RefBase& o);
+ RefBase& operator=(const RefBase& o);
+
+ weakref_impl* const mRefs;
+};
+
+// ---------------------------------------------------------------------------
+
+template <class T>
+class LightRefBase
+{
+public:
+ inline LightRefBase() : mCount(0) { }
+ inline void incStrong(__attribute__((unused)) const void* id) const {
+ __sync_fetch_and_add(&mCount, 1);
+ }
+ inline void decStrong(__attribute__((unused)) const void* id) const {
+ if (__sync_fetch_and_sub(&mCount, 1) == 1) {
+ delete static_cast<const T*>(this);
+ }
+ }
+ //! DEBUGGING ONLY: Get current strong ref count.
+ inline int32_t getStrongCount() const {
+ return mCount;
+ }
+
+ typedef LightRefBase<T> basetype;
+
+protected:
+ inline ~LightRefBase() { }
+
+private:
+ friend class ReferenceMover;
+ inline static void moveReferences(void* d, void const* s, size_t n,
+ const ReferenceConverterBase& caster) { }
+
+private:
+ mutable volatile int32_t mCount;
+};
+
+// ---------------------------------------------------------------------------
+
+template <typename T>
+class wp
+{
+public:
+ typedef typename RefBase::weakref_type weakref_type;
+
+ inline wp() : m_ptr(0) { }
+
+ wp(T* other);
+ wp(const wp<T>& other);
+ wp(const sp<T>& other);
+ template<typename U> wp(U* other);
+ template<typename U> wp(const sp<U>& other);
+ template<typename U> wp(const wp<U>& other);
+
+ ~wp();
+
+ // Assignment
+
+ wp& operator = (T* other);
+ wp& operator = (const wp<T>& other);
+ wp& operator = (const sp<T>& other);
+
+ template<typename U> wp& operator = (U* other);
+ template<typename U> wp& operator = (const wp<U>& other);
+ template<typename U> wp& operator = (const sp<U>& other);
+
+ void set_object_and_refs(T* other, weakref_type* refs);
+
+ // promotion to sp
+
+ sp<T> promote() const;
+
+ // Reset
+
+ void clear();
+
+ // Accessors
+
+ inline weakref_type* get_refs() const { return m_refs; }
+
+ inline T* unsafe_get() const { return m_ptr; }
+
+ // Operators
+
+ COMPARE_WEAK(==)
+ COMPARE_WEAK(!=)
+ COMPARE_WEAK(>)
+ COMPARE_WEAK(<)
+ COMPARE_WEAK(<=)
+ COMPARE_WEAK(>=)
+
+ inline bool operator == (const wp<T>& o) const {
+ return (m_ptr == o.m_ptr) && (m_refs == o.m_refs);
+ }
+ template<typename U>
+ inline bool operator == (const wp<U>& o) const {
+ return m_ptr == o.m_ptr;
+ }
+
+ inline bool operator > (const wp<T>& o) const {
+ return (m_ptr == o.m_ptr) ? (m_refs > o.m_refs) : (m_ptr > o.m_ptr);
+ }
+ template<typename U>
+ inline bool operator > (const wp<U>& o) const {
+ return (m_ptr == o.m_ptr) ? (m_refs > o.m_refs) : (m_ptr > o.m_ptr);
+ }
+
+ inline bool operator < (const wp<T>& o) const {
+ return (m_ptr == o.m_ptr) ? (m_refs < o.m_refs) : (m_ptr < o.m_ptr);
+ }
+ template<typename U>
+ inline bool operator < (const wp<U>& o) const {
+ return (m_ptr == o.m_ptr) ? (m_refs < o.m_refs) : (m_ptr < o.m_ptr);
+ }
+ inline bool operator != (const wp<T>& o) const { return m_refs != o.m_refs; }
+ template<typename U> inline bool operator != (const wp<U>& o) const { return !operator == (o); }
+ inline bool operator <= (const wp<T>& o) const { return !operator > (o); }
+ template<typename U> inline bool operator <= (const wp<U>& o) const { return !operator > (o); }
+ inline bool operator >= (const wp<T>& o) const { return !operator < (o); }
+ template<typename U> inline bool operator >= (const wp<U>& o) const { return !operator < (o); }
+
+private:
+ template<typename Y> friend class sp;
+ template<typename Y> friend class wp;
+
+ T* m_ptr;
+ weakref_type* m_refs;
+};
+
+template <typename T>
+TextOutput& operator<<(TextOutput& to, const wp<T>& val);
+
+#undef COMPARE_WEAK
+
+// ---------------------------------------------------------------------------
+// No user serviceable parts below here.
+
+template<typename T>
+wp<T>::wp(T* other)
+ : m_ptr(other)
+{
+ if (other) m_refs = other->createWeak(this);
+}
+
+template<typename T>
+wp<T>::wp(const wp<T>& other)
+ : m_ptr(other.m_ptr), m_refs(other.m_refs)
+{
+ if (m_ptr) m_refs->incWeak(this);
+}
+
+template<typename T>
+wp<T>::wp(const sp<T>& other)
+ : m_ptr(other.m_ptr)
+{
+ if (m_ptr) {
+ m_refs = m_ptr->createWeak(this);
+ }
+}
+
+template<typename T> template<typename U>
+wp<T>::wp(U* other)
+ : m_ptr(other)
+{
+ if (other) m_refs = other->createWeak(this);
+}
+
+template<typename T> template<typename U>
+wp<T>::wp(const wp<U>& other)
+ : m_ptr(other.m_ptr)
+{
+ if (m_ptr) {
+ m_refs = other.m_refs;
+ m_refs->incWeak(this);
+ }
+}
+
+template<typename T> template<typename U>
+wp<T>::wp(const sp<U>& other)
+ : m_ptr(other.m_ptr)
+{
+ if (m_ptr) {
+ m_refs = m_ptr->createWeak(this);
+ }
+}
+
+template<typename T>
+wp<T>::~wp()
+{
+ if (m_ptr) m_refs->decWeak(this);
+}
+
+template<typename T>
+wp<T>& wp<T>::operator = (T* other)
+{
+ weakref_type* newRefs =
+ other ? other->createWeak(this) : 0;
+ if (m_ptr) m_refs->decWeak(this);
+ m_ptr = other;
+ m_refs = newRefs;
+ return *this;
+}
+
+template<typename T>
+wp<T>& wp<T>::operator = (const wp<T>& other)
+{
+ weakref_type* otherRefs(other.m_refs);
+ T* otherPtr(other.m_ptr);
+ if (otherPtr) otherRefs->incWeak(this);
+ if (m_ptr) m_refs->decWeak(this);
+ m_ptr = otherPtr;
+ m_refs = otherRefs;
+ return *this;
+}
+
+template<typename T>
+wp<T>& wp<T>::operator = (const sp<T>& other)
+{
+ weakref_type* newRefs =
+ other != NULL ? other->createWeak(this) : 0;
+ T* otherPtr(other.m_ptr);
+ if (m_ptr) m_refs->decWeak(this);
+ m_ptr = otherPtr;
+ m_refs = newRefs;
+ return *this;
+}
+
+template<typename T> template<typename U>
+wp<T>& wp<T>::operator = (U* other)
+{
+ weakref_type* newRefs =
+ other ? other->createWeak(this) : 0;
+ if (m_ptr) m_refs->decWeak(this);
+ m_ptr = other;
+ m_refs = newRefs;
+ return *this;
+}
+
+template<typename T> template<typename U>
+wp<T>& wp<T>::operator = (const wp<U>& other)
+{
+ weakref_type* otherRefs(other.m_refs);
+ U* otherPtr(other.m_ptr);
+ if (otherPtr) otherRefs->incWeak(this);
+ if (m_ptr) m_refs->decWeak(this);
+ m_ptr = otherPtr;
+ m_refs = otherRefs;
+ return *this;
+}
+
+template<typename T> template<typename U>
+wp<T>& wp<T>::operator = (const sp<U>& other)
+{
+ weakref_type* newRefs =
+ other != NULL ? other->createWeak(this) : 0;
+ U* otherPtr(other.m_ptr);
+ if (m_ptr) m_refs->decWeak(this);
+ m_ptr = otherPtr;
+ m_refs = newRefs;
+ return *this;
+}
+
+template<typename T>
+void wp<T>::set_object_and_refs(T* other, weakref_type* refs)
+{
+ if (other) refs->incWeak(this);
+ if (m_ptr) m_refs->decWeak(this);
+ m_ptr = other;
+ m_refs = refs;
+}
+
+template<typename T>
+sp<T> wp<T>::promote() const
+{
+ sp<T> result;
+ if (m_ptr && m_refs->attemptIncStrong(&result)) {
+ result.set_pointer(m_ptr);
+ }
+ return result;
+}
+
+template<typename T>
+void wp<T>::clear()
+{
+ if (m_ptr) {
+ m_refs->decWeak(this);
+ m_ptr = 0;
+ }
+}
+
+template <typename T>
+inline TextOutput& operator<<(TextOutput& to, const wp<T>& val)
+{
+ return printWeakPointer(to, val.unsafe_get());
+}
+
+// ---------------------------------------------------------------------------
+
+// this class just serves as a namespace so TYPE::moveReferences can stay
+// private.
+
+class ReferenceMover {
+ // StrongReferenceCast and WeakReferenceCast do the impedance matching
+ // between the generic (void*) implementation in Refbase and the strongly typed
+ // template specializations below.
+
+ template <typename TYPE>
+ struct StrongReferenceCast : public ReferenceConverterBase {
+ virtual size_t getReferenceTypeSize() const { return sizeof( sp<TYPE> ); }
+ virtual void* getReferenceBase(void const* p) const {
+ sp<TYPE> const* sptr(reinterpret_cast<sp<TYPE> const*>(p));
+ return static_cast<typename TYPE::basetype *>(sptr->get());
+ }
+ };
+
+ template <typename TYPE>
+ struct WeakReferenceCast : public ReferenceConverterBase {
+ virtual size_t getReferenceTypeSize() const { return sizeof( wp<TYPE> ); }
+ virtual void* getReferenceBase(void const* p) const {
+ wp<TYPE> const* sptr(reinterpret_cast<wp<TYPE> const*>(p));
+ return static_cast<typename TYPE::basetype *>(sptr->unsafe_get());
+ }
+ };
+
+public:
+ template<typename TYPE> static inline
+ void move_references(sp<TYPE>* d, sp<TYPE> const* s, size_t n) {
+ memmove(d, s, n*sizeof(sp<TYPE>));
+ StrongReferenceCast<TYPE> caster;
+ TYPE::moveReferences(d, s, n, caster);
+ }
+ template<typename TYPE> static inline
+ void move_references(wp<TYPE>* d, wp<TYPE> const* s, size_t n) {
+ memmove(d, s, n*sizeof(wp<TYPE>));
+ WeakReferenceCast<TYPE> caster;
+ TYPE::moveReferences(d, s, n, caster);
+ }
+};
+
+// specialization for moving sp<> and wp<> types.
+// these are used by the [Sorted|Keyed]Vector<> implementations
+// sp<> and wp<> need to be handled specially, because they do not
+// have trivial copy operation in the general case (see RefBase.cpp
+// when DEBUG ops are enabled), but can be implemented very
+// efficiently in most cases.
+
+template<typename TYPE> inline
+void move_forward_type(sp<TYPE>* d, sp<TYPE> const* s, size_t n) {
+ ReferenceMover::move_references(d, s, n);
+}
+
+template<typename TYPE> inline
+void move_backward_type(sp<TYPE>* d, sp<TYPE> const* s, size_t n) {
+ ReferenceMover::move_references(d, s, n);
+}
+
+template<typename TYPE> inline
+void move_forward_type(wp<TYPE>* d, wp<TYPE> const* s, size_t n) {
+ ReferenceMover::move_references(d, s, n);
+}
+
+template<typename TYPE> inline
+void move_backward_type(wp<TYPE>* d, wp<TYPE> const* s, size_t n) {
+ ReferenceMover::move_references(d, s, n);
+}
+
+
+}; // namespace RSC
+}; // namespace android
+// ---------------------------------------------------------------------------
+
+#endif // RS_REF_BASE_H
diff --git a/cpp/util/StrongPointer.h b/cpp/util/StrongPointer.h
new file mode 100644
index 0000000..0f68615
--- /dev/null
+++ b/cpp/util/StrongPointer.h
@@ -0,0 +1,222 @@
+/*
+ * Copyright (C) 2013 The Android Open Source Project
+ *
+ * Licensed under the Apache License, Version 2.0 (the "License");
+ * you may not use this file except in compliance with the License.
+ * You may obtain a copy of the License at
+ *
+ * http://www.apache.org/licenses/LICENSE-2.0
+ *
+ * Unless required by applicable law or agreed to in writing, software
+ * distributed under the License is distributed on an "AS IS" BASIS,
+ * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ * See the License for the specific language governing permissions and
+ * limitations under the License.
+ */
+
+#ifndef RS_STRONG_POINTER_H
+#define RS_STRONG_POINTER_H
+
+//#include <cutils/atomic.h>
+
+#include <stdint.h>
+#include <sys/types.h>
+#include <stdlib.h>
+
+// ---------------------------------------------------------------------------
+namespace android {
+namespace RSC {
+
+class TextOutput;
+TextOutput& printStrongPointer(TextOutput& to, const void* val);
+
+template<typename T> class wp;
+
+// ---------------------------------------------------------------------------
+
+#define COMPARE(_op_) \
+inline bool operator _op_ (const sp<T>& o) const { \
+ return m_ptr _op_ o.m_ptr; \
+} \
+inline bool operator _op_ (const T* o) const { \
+ return m_ptr _op_ o; \
+} \
+template<typename U> \
+inline bool operator _op_ (const sp<U>& o) const { \
+ return m_ptr _op_ o.m_ptr; \
+} \
+template<typename U> \
+inline bool operator _op_ (const U* o) const { \
+ return m_ptr _op_ o; \
+} \
+inline bool operator _op_ (const wp<T>& o) const { \
+ return m_ptr _op_ o.m_ptr; \
+} \
+template<typename U> \
+inline bool operator _op_ (const wp<U>& o) const { \
+ return m_ptr _op_ o.m_ptr; \
+}
+
+// ---------------------------------------------------------------------------
+
+template <typename T>
+class sp
+{
+public:
+ inline sp() : m_ptr(0) { }
+
+ sp(T* other);
+ sp(const sp<T>& other);
+ template<typename U> sp(U* other);
+ template<typename U> sp(const sp<U>& other);
+
+ ~sp();
+
+ // Assignment
+
+ sp& operator = (T* other);
+ sp& operator = (const sp<T>& other);
+
+ template<typename U> sp& operator = (const sp<U>& other);
+ template<typename U> sp& operator = (U* other);
+
+ //! Special optimization for use by ProcessState (and nobody else).
+ void force_set(T* other);
+
+ // Reset
+
+ void clear();
+
+ // Accessors
+
+ inline T& operator* () const { return *m_ptr; }
+ inline T* operator-> () const { return m_ptr; }
+ inline T* get() const { return m_ptr; }
+
+ // Operators
+
+ COMPARE(==)
+ COMPARE(!=)
+ COMPARE(>)
+ COMPARE(<)
+ COMPARE(<=)
+ COMPARE(>=)
+
+private:
+ template<typename Y> friend class sp;
+ template<typename Y> friend class wp;
+ void set_pointer(T* ptr);
+ T* m_ptr;
+};
+
+#undef COMPARE
+
+template <typename T>
+TextOutput& operator<<(TextOutput& to, const sp<T>& val);
+
+// ---------------------------------------------------------------------------
+// No user serviceable parts below here.
+
+template<typename T>
+sp<T>::sp(T* other)
+: m_ptr(other)
+ {
+ if (other) other->incStrong(this);
+ }
+
+template<typename T>
+sp<T>::sp(const sp<T>& other)
+: m_ptr(other.m_ptr)
+ {
+ if (m_ptr) m_ptr->incStrong(this);
+ }
+
+template<typename T> template<typename U>
+sp<T>::sp(U* other) : m_ptr(other)
+{
+ if (other) ((T*)other)->incStrong(this);
+}
+
+template<typename T> template<typename U>
+sp<T>::sp(const sp<U>& other)
+: m_ptr(other.m_ptr)
+ {
+ if (m_ptr) m_ptr->incStrong(this);
+ }
+
+template<typename T>
+sp<T>::~sp()
+{
+ if (m_ptr) m_ptr->decStrong(this);
+}
+
+template<typename T>
+sp<T>& sp<T>::operator = (const sp<T>& other) {
+ T* otherPtr(other.m_ptr);
+ if (otherPtr) otherPtr->incStrong(this);
+ if (m_ptr) m_ptr->decStrong(this);
+ m_ptr = otherPtr;
+ return *this;
+}
+
+template<typename T>
+sp<T>& sp<T>::operator = (T* other)
+{
+ if (other) other->incStrong(this);
+ if (m_ptr) m_ptr->decStrong(this);
+ m_ptr = other;
+ return *this;
+}
+
+template<typename T> template<typename U>
+sp<T>& sp<T>::operator = (const sp<U>& other)
+{
+ T* otherPtr(other.m_ptr);
+ if (otherPtr) otherPtr->incStrong(this);
+ if (m_ptr) m_ptr->decStrong(this);
+ m_ptr = otherPtr;
+ return *this;
+}
+
+template<typename T> template<typename U>
+sp<T>& sp<T>::operator = (U* other)
+{
+ if (other) ((T*)other)->incStrong(this);
+ if (m_ptr) m_ptr->decStrong(this);
+ m_ptr = other;
+ return *this;
+}
+
+template<typename T>
+void sp<T>::force_set(T* other)
+{
+ other->forceIncStrong(this);
+ m_ptr = other;
+}
+
+template<typename T>
+void sp<T>::clear()
+{
+ if (m_ptr) {
+ m_ptr->decStrong(this);
+ m_ptr = 0;
+ }
+}
+
+template<typename T>
+void sp<T>::set_pointer(T* ptr) {
+ m_ptr = ptr;
+}
+
+template <typename T>
+inline TextOutput& operator<<(TextOutput& to, const sp<T>& val)
+{
+ return printStrongPointer(to, val.get());
+}
+
+}; // namespace RSC
+}; // namespace android
+
+// ---------------------------------------------------------------------------
+
+#endif // RS_STRONG_POINTER_H
diff --git a/cpp/util/TypeHelpers.h b/cpp/util/TypeHelpers.h
new file mode 100644
index 0000000..33a5201
--- /dev/null
+++ b/cpp/util/TypeHelpers.h
@@ -0,0 +1,302 @@
+/*
+ * Copyright (C) 2013 The Android Open Source Project
+ *
+ * Licensed under the Apache License, Version 2.0 (the "License");
+ * you may not use this file except in compliance with the License.
+ * You may obtain a copy of the License at
+ *
+ * http://www.apache.org/licenses/LICENSE-2.0
+ *
+ * Unless required by applicable law or agreed to in writing, software
+ * distributed under the License is distributed on an "AS IS" BASIS,
+ * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ * See the License for the specific language governing permissions and
+ * limitations under the License.
+ */
+
+#ifndef RS_TYPE_HELPERS_H
+#define RS_TYPE_HELPERS_H
+
+#include <new>
+#include <stdint.h>
+#include <string.h>
+#include <sys/types.h>
+
+// ---------------------------------------------------------------------------
+namespace android {
+namespace RSC {
+
+/*
+ * Types traits
+ */
+
+template <typename T> struct trait_trivial_ctor { enum { value = false }; };
+template <typename T> struct trait_trivial_dtor { enum { value = false }; };
+template <typename T> struct trait_trivial_copy { enum { value = false }; };
+template <typename T> struct trait_trivial_move { enum { value = false }; };
+template <typename T> struct trait_pointer { enum { value = false }; };
+template <typename T> struct trait_pointer<T*> { enum { value = true }; };
+
+template <typename TYPE>
+struct traits {
+ enum {
+ // whether this type is a pointer
+ is_pointer = trait_pointer<TYPE>::value,
+ // whether this type's constructor is a no-op
+ has_trivial_ctor = is_pointer || trait_trivial_ctor<TYPE>::value,
+ // whether this type's destructor is a no-op
+ has_trivial_dtor = is_pointer || trait_trivial_dtor<TYPE>::value,
+ // whether this type type can be copy-constructed with memcpy
+ has_trivial_copy = is_pointer || trait_trivial_copy<TYPE>::value,
+ // whether this type can be moved with memmove
+ has_trivial_move = is_pointer || trait_trivial_move<TYPE>::value
+ };
+};
+
+template <typename T, typename U>
+struct aggregate_traits {
+ enum {
+ is_pointer = false,
+ has_trivial_ctor =
+ traits<T>::has_trivial_ctor && traits<U>::has_trivial_ctor,
+ has_trivial_dtor =
+ traits<T>::has_trivial_dtor && traits<U>::has_trivial_dtor,
+ has_trivial_copy =
+ traits<T>::has_trivial_copy && traits<U>::has_trivial_copy,
+ has_trivial_move =
+ traits<T>::has_trivial_move && traits<U>::has_trivial_move
+ };
+};
+
+#define RS_TRIVIAL_CTOR_TRAIT( T ) \
+ template<> struct trait_trivial_ctor< T > { enum { value = true }; };
+
+#define RS_TRIVIAL_DTOR_TRAIT( T ) \
+ template<> struct trait_trivial_dtor< T > { enum { value = true }; };
+
+#define RS_TRIVIAL_COPY_TRAIT( T ) \
+ template<> struct trait_trivial_copy< T > { enum { value = true }; };
+
+#define RS_TRIVIAL_MOVE_TRAIT( T ) \
+ template<> struct trait_trivial_move< T > { enum { value = true }; };
+
+#define RS_BASIC_TYPES_TRAITS( T ) \
+ RS_TRIVIAL_CTOR_TRAIT( T ) \
+ RS_TRIVIAL_DTOR_TRAIT( T ) \
+ RS_TRIVIAL_COPY_TRAIT( T ) \
+ RS_TRIVIAL_MOVE_TRAIT( T )
+
+// ---------------------------------------------------------------------------
+
+/*
+ * basic types traits
+ */
+
+RS_BASIC_TYPES_TRAITS( void )
+RS_BASIC_TYPES_TRAITS( bool )
+RS_BASIC_TYPES_TRAITS( char )
+RS_BASIC_TYPES_TRAITS( unsigned char )
+RS_BASIC_TYPES_TRAITS( short )
+RS_BASIC_TYPES_TRAITS( unsigned short )
+RS_BASIC_TYPES_TRAITS( int )
+RS_BASIC_TYPES_TRAITS( unsigned int )
+RS_BASIC_TYPES_TRAITS( long )
+RS_BASIC_TYPES_TRAITS( unsigned long )
+RS_BASIC_TYPES_TRAITS( long long )
+RS_BASIC_TYPES_TRAITS( unsigned long long )
+RS_BASIC_TYPES_TRAITS( float )
+RS_BASIC_TYPES_TRAITS( double )
+
+// ---------------------------------------------------------------------------
+
+
+/*
+ * compare and order types
+ */
+
+template<typename TYPE> inline
+int strictly_order_type(const TYPE& lhs, const TYPE& rhs) {
+ return (lhs < rhs) ? 1 : 0;
+}
+
+template<typename TYPE> inline
+int compare_type(const TYPE& lhs, const TYPE& rhs) {
+ return strictly_order_type(rhs, lhs) - strictly_order_type(lhs, rhs);
+}
+
+/*
+ * create, destroy, copy and move types...
+ */
+
+template<typename TYPE> inline
+void construct_type(TYPE* p, size_t n) {
+ if (!traits<TYPE>::has_trivial_ctor) {
+ while (n--) {
+ new(p++) TYPE;
+ }
+ }
+}
+
+template<typename TYPE> inline
+void destroy_type(TYPE* p, size_t n) {
+ if (!traits<TYPE>::has_trivial_dtor) {
+ while (n--) {
+ p->~TYPE();
+ p++;
+ }
+ }
+}
+
+template<typename TYPE> inline
+void copy_type(TYPE* d, const TYPE* s, size_t n) {
+ if (!traits<TYPE>::has_trivial_copy) {
+ while (n--) {
+ new(d) TYPE(*s);
+ d++, s++;
+ }
+ } else {
+ memcpy(d,s,n*sizeof(TYPE));
+ }
+}
+
+template<typename TYPE> inline
+void splat_type(TYPE* where, const TYPE* what, size_t n) {
+ if (!traits<TYPE>::has_trivial_copy) {
+ while (n--) {
+ new(where) TYPE(*what);
+ where++;
+ }
+ } else {
+ while (n--) {
+ *where++ = *what;
+ }
+ }
+}
+
+template<typename TYPE> inline
+void move_forward_type(TYPE* d, const TYPE* s, size_t n = 1) {
+ if ((traits<TYPE>::has_trivial_dtor && traits<TYPE>::has_trivial_copy)
+ || traits<TYPE>::has_trivial_move)
+ {
+ memmove(d,s,n*sizeof(TYPE));
+ } else {
+ d += n;
+ s += n;
+ while (n--) {
+ --d, --s;
+ if (!traits<TYPE>::has_trivial_copy) {
+ new(d) TYPE(*s);
+ } else {
+ *d = *s;
+ }
+ if (!traits<TYPE>::has_trivial_dtor) {
+ s->~TYPE();
+ }
+ }
+ }
+}
+
+template<typename TYPE> inline
+void move_backward_type(TYPE* d, const TYPE* s, size_t n = 1) {
+ if ((traits<TYPE>::has_trivial_dtor && traits<TYPE>::has_trivial_copy)
+ || traits<TYPE>::has_trivial_move)
+ {
+ memmove(d,s,n*sizeof(TYPE));
+ } else {
+ while (n--) {
+ if (!traits<TYPE>::has_trivial_copy) {
+ new(d) TYPE(*s);
+ } else {
+ *d = *s;
+ }
+ if (!traits<TYPE>::has_trivial_dtor) {
+ s->~TYPE();
+ }
+ d++, s++;
+ }
+ }
+}
+
+// ---------------------------------------------------------------------------
+
+/*
+ * a key/value pair
+ */
+
+template <typename KEY, typename VALUE>
+struct key_value_pair_t {
+ typedef KEY key_t;
+ typedef VALUE value_t;
+
+ KEY key;
+ VALUE value;
+ key_value_pair_t() { }
+ key_value_pair_t(const key_value_pair_t& o) : key(o.key), value(o.value) { }
+ key_value_pair_t(const KEY& k, const VALUE& v) : key(k), value(v) { }
+ key_value_pair_t(const KEY& k) : key(k) { }
+ inline bool operator < (const key_value_pair_t& o) const {
+ return strictly_order_type(key, o.key);
+ }
+ inline const KEY& getKey() const {
+ return key;
+ }
+ inline const VALUE& getValue() const {
+ return value;
+ }
+};
+
+template <typename K, typename V>
+struct trait_trivial_ctor< key_value_pair_t<K, V> >
+{ enum { value = aggregate_traits<K,V>::has_trivial_ctor }; };
+template <typename K, typename V>
+struct trait_trivial_dtor< key_value_pair_t<K, V> >
+{ enum { value = aggregate_traits<K,V>::has_trivial_dtor }; };
+template <typename K, typename V>
+struct trait_trivial_copy< key_value_pair_t<K, V> >
+{ enum { value = aggregate_traits<K,V>::has_trivial_copy }; };
+template <typename K, typename V>
+struct trait_trivial_move< key_value_pair_t<K, V> >
+{ enum { value = aggregate_traits<K,V>::has_trivial_move }; };
+
+// ---------------------------------------------------------------------------
+
+/*
+ * Hash codes.
+ */
+typedef uint32_t hash_t;
+
+template <typename TKey>
+hash_t hash_type(const TKey& key);
+
+/* Built-in hash code specializations.
+ * Assumes pointers are 32bit. */
+#define RS_INT32_HASH(T) \
+ template <> inline hash_t hash_type(const T& value) { return hash_t(value); }
+#define RS_INT64_HASH(T) \
+ template <> inline hash_t hash_type(const T& value) { \
+ return hash_t((value >> 32) ^ value); }
+#define RS_REINTERPRET_HASH(T, R) \
+ template <> inline hash_t hash_type(const T& value) { \
+ return hash_type(*reinterpret_cast<const R*>(&value)); }
+
+RS_INT32_HASH(bool)
+RS_INT32_HASH(int8_t)
+RS_INT32_HASH(uint8_t)
+RS_INT32_HASH(int16_t)
+RS_INT32_HASH(uint16_t)
+RS_INT32_HASH(int32_t)
+RS_INT32_HASH(uint32_t)
+RS_INT64_HASH(int64_t)
+RS_INT64_HASH(uint64_t)
+RS_REINTERPRET_HASH(float, uint32_t)
+RS_REINTERPRET_HASH(double, uint64_t)
+
+template <typename T> inline hash_t hash_type(T* const & value) {
+ return hash_type(uintptr_t(value));
+}
+
+}; // namespace RSC
+}; // namespace android
+// ---------------------------------------------------------------------------
+
+#endif // RS_TYPE_HELPERS_H
diff --git a/cpu_ref/Android.mk b/cpu_ref/Android.mk
index 67e0786..b3de4f4 100644
--- a/cpu_ref/Android.mk
+++ b/cpu_ref/Android.mk
@@ -27,19 +27,25 @@
rsCpuIntrinsicColorMatrix.cpp \
rsCpuIntrinsicConvolve3x3.cpp \
rsCpuIntrinsicConvolve5x5.cpp \
+ rsCpuIntrinsicHistogram.cpp \
rsCpuIntrinsicLUT.cpp \
rsCpuIntrinsicYuvToRGB.cpp
ifeq ($(ARCH_ARM_HAVE_NEON),true)
LOCAL_CFLAGS += -DARCH_ARM_HAVE_NEON
LOCAL_SRC_FILES+= \
- rsCpuIntrinsics_neon.S
+ rsCpuIntrinsics_neon.S \
+ rsCpuIntrinsics_neon_ColorMatrix.S
endif
ifeq ($(ARCH_ARM_HAVE_VFP),true)
LOCAL_CFLAGS += -DARCH_ARM_HAVE_VFP
endif
+ifeq ($(ARCH_X86_HAVE_SSE2), true)
+ LOCAL_CFLAGS += -DARCH_X86_HAVE_SSE2
+endif
+
LOCAL_SHARED_LIBRARIES += libRS libcutils libutils liblog libsync
LOCAL_SHARED_LIBRARIES += libbcc libbcinfo
diff --git a/driver/linkloader/android/librsloader.cpp b/cpu_ref/linkloader/android/librsloader.cpp
similarity index 100%
rename from driver/linkloader/android/librsloader.cpp
rename to cpu_ref/linkloader/android/librsloader.cpp
diff --git a/driver/linkloader/android/librsloader.h b/cpu_ref/linkloader/android/librsloader.h
similarity index 100%
rename from driver/linkloader/android/librsloader.h
rename to cpu_ref/linkloader/android/librsloader.h
diff --git a/driver/linkloader/android/test-librsloader.c b/cpu_ref/linkloader/android/test-librsloader.c
similarity index 100%
rename from driver/linkloader/android/test-librsloader.c
rename to cpu_ref/linkloader/android/test-librsloader.c
diff --git a/driver/linkloader/include/ELF.h b/cpu_ref/linkloader/include/ELF.h
similarity index 100%
rename from driver/linkloader/include/ELF.h
rename to cpu_ref/linkloader/include/ELF.h
diff --git a/driver/linkloader/include/ELFHeader.h b/cpu_ref/linkloader/include/ELFHeader.h
similarity index 100%
rename from driver/linkloader/include/ELFHeader.h
rename to cpu_ref/linkloader/include/ELFHeader.h
diff --git a/driver/linkloader/include/ELFObject.h b/cpu_ref/linkloader/include/ELFObject.h
similarity index 100%
rename from driver/linkloader/include/ELFObject.h
rename to cpu_ref/linkloader/include/ELFObject.h
diff --git a/driver/linkloader/include/ELFReloc.h b/cpu_ref/linkloader/include/ELFReloc.h
similarity index 100%
rename from driver/linkloader/include/ELFReloc.h
rename to cpu_ref/linkloader/include/ELFReloc.h
diff --git a/driver/linkloader/include/ELFSection.h b/cpu_ref/linkloader/include/ELFSection.h
similarity index 100%
rename from driver/linkloader/include/ELFSection.h
rename to cpu_ref/linkloader/include/ELFSection.h
diff --git a/driver/linkloader/include/ELFSectionBits.h b/cpu_ref/linkloader/include/ELFSectionBits.h
similarity index 100%
rename from driver/linkloader/include/ELFSectionBits.h
rename to cpu_ref/linkloader/include/ELFSectionBits.h
diff --git a/driver/linkloader/include/ELFSectionHeader.h b/cpu_ref/linkloader/include/ELFSectionHeader.h
similarity index 100%
rename from driver/linkloader/include/ELFSectionHeader.h
rename to cpu_ref/linkloader/include/ELFSectionHeader.h
diff --git a/driver/linkloader/include/ELFSectionHeaderTable.h b/cpu_ref/linkloader/include/ELFSectionHeaderTable.h
similarity index 100%
rename from driver/linkloader/include/ELFSectionHeaderTable.h
rename to cpu_ref/linkloader/include/ELFSectionHeaderTable.h
diff --git a/driver/linkloader/include/ELFSectionNoBits.h b/cpu_ref/linkloader/include/ELFSectionNoBits.h
similarity index 100%
rename from driver/linkloader/include/ELFSectionNoBits.h
rename to cpu_ref/linkloader/include/ELFSectionNoBits.h
diff --git a/driver/linkloader/include/ELFSectionProgBits.h b/cpu_ref/linkloader/include/ELFSectionProgBits.h
similarity index 100%
rename from driver/linkloader/include/ELFSectionProgBits.h
rename to cpu_ref/linkloader/include/ELFSectionProgBits.h
diff --git a/driver/linkloader/include/ELFSectionRelTable.h b/cpu_ref/linkloader/include/ELFSectionRelTable.h
similarity index 100%
rename from driver/linkloader/include/ELFSectionRelTable.h
rename to cpu_ref/linkloader/include/ELFSectionRelTable.h
diff --git a/driver/linkloader/include/ELFSectionStrTab.h b/cpu_ref/linkloader/include/ELFSectionStrTab.h
similarity index 100%
rename from driver/linkloader/include/ELFSectionStrTab.h
rename to cpu_ref/linkloader/include/ELFSectionStrTab.h
diff --git a/driver/linkloader/include/ELFSectionSymTab.h b/cpu_ref/linkloader/include/ELFSectionSymTab.h
similarity index 100%
rename from driver/linkloader/include/ELFSectionSymTab.h
rename to cpu_ref/linkloader/include/ELFSectionSymTab.h
diff --git a/driver/linkloader/include/ELFSymbol.h b/cpu_ref/linkloader/include/ELFSymbol.h
similarity index 100%
rename from driver/linkloader/include/ELFSymbol.h
rename to cpu_ref/linkloader/include/ELFSymbol.h
diff --git a/driver/linkloader/include/ELFTypes.h b/cpu_ref/linkloader/include/ELFTypes.h
similarity index 100%
rename from driver/linkloader/include/ELFTypes.h
rename to cpu_ref/linkloader/include/ELFTypes.h
diff --git a/driver/linkloader/include/GOT.h b/cpu_ref/linkloader/include/GOT.h
similarity index 100%
rename from driver/linkloader/include/GOT.h
rename to cpu_ref/linkloader/include/GOT.h
diff --git a/driver/linkloader/include/MemChunk.h b/cpu_ref/linkloader/include/MemChunk.h
similarity index 73%
rename from driver/linkloader/include/MemChunk.h
rename to cpu_ref/linkloader/include/MemChunk.h
index fb853e9..c4d78c9 100644
--- a/driver/linkloader/include/MemChunk.h
+++ b/cpu_ref/linkloader/include/MemChunk.h
@@ -18,12 +18,22 @@
#define MEM_CHUNK_H
#include <stddef.h>
+#include <stdint.h>
#include <stdlib.h>
+typedef void *(*AllocFunc) (size_t, uint32_t);
+typedef void (*FreeFunc) (void *);
+
class MemChunk {
private:
unsigned char *buf;
size_t buf_size;
+ bool bVendorBuf;
+
+ static AllocFunc VendorAlloc;
+ static FreeFunc VendorFree;
+
+ bool invalidBuf() const;
public:
MemChunk();
@@ -56,6 +66,12 @@
return buf_size;
}
+ // The allocation function must return page-aligned memory or we will be
+ // unable to mprotect the region appropriately.
+ static void registerAllocFreeCallbacks(AllocFunc a, FreeFunc f) {
+ VendorAlloc = a;
+ VendorFree = f;
+ }
};
#endif // MEM_CHUNK_H
diff --git a/driver/linkloader/include/StubLayout.h b/cpu_ref/linkloader/include/StubLayout.h
similarity index 100%
rename from driver/linkloader/include/StubLayout.h
rename to cpu_ref/linkloader/include/StubLayout.h
diff --git a/driver/linkloader/include/impl/ELFHeader.hxx b/cpu_ref/linkloader/include/impl/ELFHeader.hxx
similarity index 100%
rename from driver/linkloader/include/impl/ELFHeader.hxx
rename to cpu_ref/linkloader/include/impl/ELFHeader.hxx
diff --git a/driver/linkloader/include/impl/ELFObject.hxx b/cpu_ref/linkloader/include/impl/ELFObject.hxx
similarity index 100%
rename from driver/linkloader/include/impl/ELFObject.hxx
rename to cpu_ref/linkloader/include/impl/ELFObject.hxx
diff --git a/driver/linkloader/include/impl/ELFReloc.hxx b/cpu_ref/linkloader/include/impl/ELFReloc.hxx
similarity index 100%
rename from driver/linkloader/include/impl/ELFReloc.hxx
rename to cpu_ref/linkloader/include/impl/ELFReloc.hxx
diff --git a/driver/linkloader/include/impl/ELFSection.hxx b/cpu_ref/linkloader/include/impl/ELFSection.hxx
similarity index 100%
rename from driver/linkloader/include/impl/ELFSection.hxx
rename to cpu_ref/linkloader/include/impl/ELFSection.hxx
diff --git a/driver/linkloader/include/impl/ELFSectionBits.hxx b/cpu_ref/linkloader/include/impl/ELFSectionBits.hxx
similarity index 100%
rename from driver/linkloader/include/impl/ELFSectionBits.hxx
rename to cpu_ref/linkloader/include/impl/ELFSectionBits.hxx
diff --git a/driver/linkloader/include/impl/ELFSectionHeader.hxx b/cpu_ref/linkloader/include/impl/ELFSectionHeader.hxx
similarity index 100%
rename from driver/linkloader/include/impl/ELFSectionHeader.hxx
rename to cpu_ref/linkloader/include/impl/ELFSectionHeader.hxx
diff --git a/driver/linkloader/include/impl/ELFSectionHeaderTable.hxx b/cpu_ref/linkloader/include/impl/ELFSectionHeaderTable.hxx
similarity index 100%
rename from driver/linkloader/include/impl/ELFSectionHeaderTable.hxx
rename to cpu_ref/linkloader/include/impl/ELFSectionHeaderTable.hxx
diff --git a/driver/linkloader/include/impl/ELFSectionNoBits.hxx b/cpu_ref/linkloader/include/impl/ELFSectionNoBits.hxx
similarity index 100%
rename from driver/linkloader/include/impl/ELFSectionNoBits.hxx
rename to cpu_ref/linkloader/include/impl/ELFSectionNoBits.hxx
diff --git a/driver/linkloader/include/impl/ELFSectionProgBits.hxx b/cpu_ref/linkloader/include/impl/ELFSectionProgBits.hxx
similarity index 100%
rename from driver/linkloader/include/impl/ELFSectionProgBits.hxx
rename to cpu_ref/linkloader/include/impl/ELFSectionProgBits.hxx
diff --git a/driver/linkloader/include/impl/ELFSectionRelTable.hxx b/cpu_ref/linkloader/include/impl/ELFSectionRelTable.hxx
similarity index 100%
rename from driver/linkloader/include/impl/ELFSectionRelTable.hxx
rename to cpu_ref/linkloader/include/impl/ELFSectionRelTable.hxx
diff --git a/driver/linkloader/include/impl/ELFSectionStrTab.hxx b/cpu_ref/linkloader/include/impl/ELFSectionStrTab.hxx
similarity index 100%
rename from driver/linkloader/include/impl/ELFSectionStrTab.hxx
rename to cpu_ref/linkloader/include/impl/ELFSectionStrTab.hxx
diff --git a/driver/linkloader/include/impl/ELFSectionSymTab.hxx b/cpu_ref/linkloader/include/impl/ELFSectionSymTab.hxx
similarity index 100%
rename from driver/linkloader/include/impl/ELFSectionSymTab.hxx
rename to cpu_ref/linkloader/include/impl/ELFSectionSymTab.hxx
diff --git a/driver/linkloader/include/impl/ELFSymbol.hxx b/cpu_ref/linkloader/include/impl/ELFSymbol.hxx
similarity index 100%
rename from driver/linkloader/include/impl/ELFSymbol.hxx
rename to cpu_ref/linkloader/include/impl/ELFSymbol.hxx
diff --git a/driver/linkloader/include/mmanWindows.h b/cpu_ref/linkloader/include/mmanWindows.h
similarity index 100%
rename from driver/linkloader/include/mmanWindows.h
rename to cpu_ref/linkloader/include/mmanWindows.h
diff --git a/driver/linkloader/lib/ELFHeader.cpp b/cpu_ref/linkloader/lib/ELFHeader.cpp
similarity index 100%
rename from driver/linkloader/lib/ELFHeader.cpp
rename to cpu_ref/linkloader/lib/ELFHeader.cpp
diff --git a/driver/linkloader/lib/ELFSectionHeader.cpp b/cpu_ref/linkloader/lib/ELFSectionHeader.cpp
similarity index 100%
rename from driver/linkloader/lib/ELFSectionHeader.cpp
rename to cpu_ref/linkloader/lib/ELFSectionHeader.cpp
diff --git a/driver/linkloader/lib/ELFSymbol.cpp b/cpu_ref/linkloader/lib/ELFSymbol.cpp
similarity index 100%
rename from driver/linkloader/lib/ELFSymbol.cpp
rename to cpu_ref/linkloader/lib/ELFSymbol.cpp
diff --git a/driver/linkloader/lib/ELFTypes.cpp b/cpu_ref/linkloader/lib/ELFTypes.cpp
similarity index 100%
rename from driver/linkloader/lib/ELFTypes.cpp
rename to cpu_ref/linkloader/lib/ELFTypes.cpp
diff --git a/driver/linkloader/lib/GOT.cpp b/cpu_ref/linkloader/lib/GOT.cpp
similarity index 100%
rename from driver/linkloader/lib/GOT.cpp
rename to cpu_ref/linkloader/lib/GOT.cpp
diff --git a/driver/linkloader/lib/MemChunk.cpp b/cpu_ref/linkloader/lib/MemChunk.cpp
similarity index 66%
rename from driver/linkloader/lib/MemChunk.cpp
rename to cpu_ref/linkloader/lib/MemChunk.cpp
index 6a04916..5d6c102 100644
--- a/driver/linkloader/lib/MemChunk.cpp
+++ b/cpu_ref/linkloader/lib/MemChunk.cpp
@@ -41,32 +41,49 @@
static uintptr_t StartAddr = 0x7e000000UL;
#endif
-MemChunk::MemChunk() : buf((unsigned char *)MAP_FAILED), buf_size(0) {
+AllocFunc MemChunk::VendorAlloc = NULL;
+FreeFunc MemChunk::VendorFree = NULL;
+
+MemChunk::MemChunk() : buf(NULL), buf_size(0), bVendorBuf(true) {
}
MemChunk::~MemChunk() {
- if (buf != MAP_FAILED) {
+ if (!invalidBuf() && bVendorBuf && VendorFree) {
+ (*VendorFree)(buf);
+ return;
+ }
+ if (!invalidBuf()) {
munmap(buf, buf_size);
}
}
+bool MemChunk::invalidBuf() const {
+ return (buf == 0 || buf == (unsigned char *)MAP_FAILED);
+}
+
bool MemChunk::allocate(size_t size) {
if (size == 0) {
return true;
}
+ if (VendorAlloc) {
+ buf = (unsigned char*)(*VendorAlloc)(size, 0);
+ }
+ if (invalidBuf()) {
+ bVendorBuf = false;
#if USE_FIXED_ADDR_MEM_CHUNK
- buf = (unsigned char *)mmap((void *)StartAddr, size,
- PROT_READ | PROT_WRITE,
- MAP_PRIVATE | MAP_ANON | MAP_32BIT,
- -1, 0);
+ buf = (unsigned char *)mmap((void *)StartAddr, size,
+ PROT_READ | PROT_WRITE,
+ MAP_PRIVATE | MAP_ANON | MAP_32BIT,
+ -1, 0);
#else
- buf = (unsigned char *)mmap(0, size,
- PROT_READ | PROT_WRITE,
- MAP_PRIVATE | MAP_ANON | MAP_32BIT,
- -1, 0);
+ buf = (unsigned char *)mmap(0, size,
+ PROT_READ | PROT_WRITE,
+ MAP_PRIVATE | MAP_ANON | MAP_32BIT,
+ -1, 0);
#endif
+ }
- if (buf == MAP_FAILED) {
+ if (invalidBuf()) {
return false;
}
@@ -79,7 +96,7 @@
}
void MemChunk::print() const {
- if (buf != MAP_FAILED) {
+ if (!invalidBuf()) {
dump_hex(buf, buf_size, 0, buf_size);
}
}
diff --git a/driver/linkloader/lib/StubLayout.cpp b/cpu_ref/linkloader/lib/StubLayout.cpp
similarity index 100%
rename from driver/linkloader/lib/StubLayout.cpp
rename to cpu_ref/linkloader/lib/StubLayout.cpp
diff --git a/driver/linkloader/lib/mmanWindows.cpp b/cpu_ref/linkloader/lib/mmanWindows.cpp
similarity index 100%
rename from driver/linkloader/lib/mmanWindows.cpp
rename to cpu_ref/linkloader/lib/mmanWindows.cpp
diff --git a/driver/linkloader/main.cpp b/cpu_ref/linkloader/main.cpp
similarity index 100%
rename from driver/linkloader/main.cpp
rename to cpu_ref/linkloader/main.cpp
diff --git a/driver/linkloader/utils/flush_cpu_cache.h b/cpu_ref/linkloader/utils/flush_cpu_cache.h
similarity index 100%
rename from driver/linkloader/utils/flush_cpu_cache.h
rename to cpu_ref/linkloader/utils/flush_cpu_cache.h
diff --git a/driver/linkloader/utils/helper.cpp b/cpu_ref/linkloader/utils/helper.cpp
similarity index 100%
rename from driver/linkloader/utils/helper.cpp
rename to cpu_ref/linkloader/utils/helper.cpp
diff --git a/driver/linkloader/utils/helper.h b/cpu_ref/linkloader/utils/helper.h
similarity index 100%
rename from driver/linkloader/utils/helper.h
rename to cpu_ref/linkloader/utils/helper.h
diff --git a/driver/linkloader/utils/raw_ostream.cpp b/cpu_ref/linkloader/utils/raw_ostream.cpp
similarity index 100%
rename from driver/linkloader/utils/raw_ostream.cpp
rename to cpu_ref/linkloader/utils/raw_ostream.cpp
diff --git a/driver/linkloader/utils/raw_ostream.h b/cpu_ref/linkloader/utils/raw_ostream.h
similarity index 100%
rename from driver/linkloader/utils/raw_ostream.h
rename to cpu_ref/linkloader/utils/raw_ostream.h
diff --git a/driver/linkloader/utils/rsl_assert.cpp b/cpu_ref/linkloader/utils/rsl_assert.cpp
similarity index 100%
rename from driver/linkloader/utils/rsl_assert.cpp
rename to cpu_ref/linkloader/utils/rsl_assert.cpp
diff --git a/driver/linkloader/utils/rsl_assert.h b/cpu_ref/linkloader/utils/rsl_assert.h
similarity index 100%
rename from driver/linkloader/utils/rsl_assert.h
rename to cpu_ref/linkloader/utils/rsl_assert.h
diff --git a/driver/linkloader/utils/serialize.h b/cpu_ref/linkloader/utils/serialize.h
similarity index 100%
rename from driver/linkloader/utils/serialize.h
rename to cpu_ref/linkloader/utils/serialize.h
diff --git a/driver/linkloader/utils/traits.h b/cpu_ref/linkloader/utils/traits.h
similarity index 100%
rename from driver/linkloader/utils/traits.h
rename to cpu_ref/linkloader/utils/traits.h
diff --git a/cpu_ref/rsCpuCore.cpp b/cpu_ref/rsCpuCore.cpp
index 79b908c..7475ccb 100644
--- a/cpu_ref/rsCpuCore.cpp
+++ b/cpu_ref/rsCpuCore.cpp
@@ -115,6 +115,7 @@
#ifndef RS_COMPATIBILITY_LIB
mLinkRuntimeCallback = NULL;
mSelectRTCallback = NULL;
+ mSetupCompilerCallback = NULL;
#endif
}
@@ -519,6 +520,8 @@
const Script *s, const Element *e);
extern RsdCpuScriptImpl * rsdIntrinsic_Blend(RsdCpuReferenceImpl *ctx,
const Script *s, const Element *e);
+extern RsdCpuScriptImpl * rsdIntrinsic_Histogram(RsdCpuReferenceImpl *ctx,
+ const Script *s, const Element *e);
RsdCpuReference::CpuScript * RsdCpuReferenceImpl::createIntrinsic(const Script *s,
RsScriptIntrinsicID iid, Element *e) {
@@ -549,6 +552,9 @@
case RS_SCRIPT_INTRINSIC_ID_BLEND:
i = rsdIntrinsic_Blend(this, s, e);
break;
+ case RS_SCRIPT_INTRINSIC_ID_HISTOGRAM:
+ i = rsdIntrinsic_Histogram(this, s, e);
+ break;
default:
rsAssert(0);
diff --git a/cpu_ref/rsCpuCore.h b/cpu_ref/rsCpuCore.h
index 0f30a48..22d0d28 100644
--- a/cpu_ref/rsCpuCore.h
+++ b/cpu_ref/rsCpuCore.h
@@ -126,6 +126,14 @@
RSSelectRTCallback getSelectRTCallback() {
return mSelectRTCallback;
}
+
+ virtual void setSetupCompilerCallback(
+ RSSetupCompilerCallback pSetupCompilerCallback) {
+ mSetupCompilerCallback = pSetupCompilerCallback;
+ }
+ virtual RSSetupCompilerCallback getSetupCompilerCallback() const {
+ return mSetupCompilerCallback;
+ }
#endif
virtual bool getInForEach() { return mInForEach; }
@@ -157,6 +165,7 @@
#ifndef RS_COMPATIBILITY_LIB
bcc::RSLinkRuntimeCallback mLinkRuntimeCallback;
RSSelectRTCallback mSelectRTCallback;
+ RSSetupCompilerCallback mSetupCompilerCallback;
#endif
};
diff --git a/cpu_ref/rsCpuIntrinsic.cpp b/cpu_ref/rsCpuIntrinsic.cpp
index 450ee30..cda40f0 100644
--- a/cpu_ref/rsCpuIntrinsic.cpp
+++ b/cpu_ref/rsCpuIntrinsic.cpp
@@ -73,6 +73,16 @@
}
+void RsdCpuScriptIntrinsic::preLaunch(uint32_t slot, const Allocation * ain,
+ Allocation * aout, const void * usr,
+ uint32_t usrLen, const RsScriptCall *sc) {
+}
+
+void RsdCpuScriptIntrinsic::postLaunch(uint32_t slot, const Allocation * ain,
+ Allocation * aout, const void * usr,
+ uint32_t usrLen, const RsScriptCall *sc) {
+}
+
void RsdCpuScriptIntrinsic::invokeForEach(uint32_t slot,
const Allocation * ain,
Allocation * aout,
@@ -81,6 +91,8 @@
const RsScriptCall *sc) {
MTLaunchStruct mtls;
+ preLaunch(slot, ain, aout, usr, usrLen, sc);
+
forEachMtlsSetup(ain, aout, usr, usrLen, sc, &mtls);
mtls.script = this;
mtls.fep.slot = slot;
@@ -91,6 +103,8 @@
RsdCpuScriptImpl * oldTLS = mCtx->setTLS(this);
mCtx->launchThreads(ain, aout, sc, &mtls);
mCtx->setTLS(oldTLS);
+
+ postLaunch(slot, ain, aout, usr, usrLen, sc);
}
void RsdCpuScriptIntrinsic::forEachKernelSetup(uint32_t slot, MTLaunchStruct *mtls) {
diff --git a/cpu_ref/rsCpuIntrinsic.h b/cpu_ref/rsCpuIntrinsic.h
index 35ffc69..917b235 100644
--- a/cpu_ref/rsCpuIntrinsic.h
+++ b/cpu_ref/rsCpuIntrinsic.h
@@ -40,6 +40,13 @@
virtual void invokeInit();
virtual void invokeFreeChildren();
+ virtual void preLaunch(uint32_t slot, const Allocation * ain,
+ Allocation * aout, const void * usr,
+ uint32_t usrLen, const RsScriptCall *sc);
+ virtual void postLaunch(uint32_t slot, const Allocation * ain,
+ Allocation * aout, const void * usr,
+ uint32_t usrLen, const RsScriptCall *sc);
+
virtual void setGlobalVar(uint32_t slot, const void *data, size_t dataLength);
virtual void setGlobalVarWithElemDims(uint32_t slot, const void *data, size_t dataLength,
const Element *e, const size_t *dims, size_t dimLength);
diff --git a/cpu_ref/rsCpuIntrinsicColorMatrix.cpp b/cpu_ref/rsCpuIntrinsicColorMatrix.cpp
index c6e38c0..3ad3063 100644
--- a/cpu_ref/rsCpuIntrinsicColorMatrix.cpp
+++ b/cpu_ref/rsCpuIntrinsicColorMatrix.cpp
@@ -14,9 +14,19 @@
* limitations under the License.
*/
+#include <sys/mman.h>
+#include <unistd.h>
#include "rsCpuIntrinsic.h"
#include "rsCpuIntrinsicInlines.h"
+#include "linkloader/include/MemChunk.h"
+
+#include <sys/mman.h>
+#include <stddef.h>
+#include <stdint.h>
+#include <stdlib.h>
+//#include <utils/StopWatch.h>
+
using namespace android;
using namespace android::renderscript;
@@ -34,49 +44,425 @@
virtual ~RsdCpuScriptIntrinsicColorMatrix();
RsdCpuScriptIntrinsicColorMatrix(RsdCpuReferenceImpl *ctx, const Script *s, const Element *e);
+ virtual void preLaunch(uint32_t slot, const Allocation * ain, Allocation * aout,
+ const void * usr, uint32_t usrLen, const RsScriptCall *sc);
+ virtual void postLaunch(uint32_t slot, const Allocation * ain, Allocation * aout,
+ const void * usr, uint32_t usrLen, const RsScriptCall *sc);
+
protected:
float fp[16];
+ float fpa[4];
short ip[16];
+ int ipa[4];
- static void kernel4x4(const RsForEachStubParamStruct *p,
- uint32_t xstart, uint32_t xend,
- uint32_t instep, uint32_t outstep);
- static void kernel3x3(const RsForEachStubParamStruct *p,
- uint32_t xstart, uint32_t xend,
- uint32_t instep, uint32_t outstep);
- static void kernelDot(const RsForEachStubParamStruct *p,
- uint32_t xstart, uint32_t xend,
- uint32_t instep, uint32_t outstep);
+ static void kernel(const RsForEachStubParamStruct *p,
+ uint32_t xstart, uint32_t xend,
+ uint32_t instep, uint32_t outstep);
+
+ int64_t mLastKey;
+ unsigned char *mBuf;
+ size_t mBufSize;
+
+ int64_t computeKey(const Element *ein, const Element *eout);
+
+ bool build(int64_t key);
+
+ void (*mOptKernel)(void *dst, const void *src, const short *coef, uint32_t count);
+
};
}
}
-void RsdCpuScriptIntrinsicColorMatrix::setGlobalVar(uint32_t slot, const void *data,
- size_t dataLength) {
- rsAssert(slot == 0);
- memcpy (fp, data, dataLength);
- for(int ct=0; ct < 16; ct++) {
- ip[ct] = (short)(fp[ct] * 255.f + 0.5f);
+#define CM_IN_VEC_SIZE_MASK 0x00000003
+#define CM_OUT_VEC_SIZE_MASK 0x0000000c
+#define CM_IN_TYPE_SIZE_MASK 0x000000F0
+#define CM_OUT_TYPE_SIZE_MASK 0x00000F00
+#define CM_DOT_MASK 0x00001000
+#define CM_ADD_MASK 0x00002000
+#define CM_COPY_ALPHA 0x00004000
+#define CM_MATRIX_MASK 0xFFFF0000
+
+
+int64_t RsdCpuScriptIntrinsicColorMatrix::computeKey(
+ const Element *ein, const Element *eout) {
+
+ int64_t key = 0;
+
+ // Compute a unique code key for this operation
+
+ // Add to the key the input and output types
+ key |= (ein->getVectorSize() - 1);
+ key |= (eout->getVectorSize() - 1) << 2;
+
+ bool hasFloat = false;
+ if (ein->getType() == RS_TYPE_FLOAT_32) {
+ hasFloat = true;
+ key |= 1 << 4;
+ }
+ if (eout->getType() == RS_TYPE_FLOAT_32) {
+ hasFloat = true;
+ key |= 1 << 8;
}
- mRootPtr = &kernel4x4;
- if ((ip[3] == 0) && (ip[7] == 0) && (ip[11] == 0) &&
- (ip[12] == 0) && (ip[13] == 0) && (ip[14] == 0) && (ip[15] == 255)) {
- mRootPtr = &kernel3x3;
-
- if ((ip[0] == ip[1]) && (ip[0] == ip[2]) &&
- (ip[4] == ip[5]) && (ip[4] == ip[6]) &&
- (ip[8] == ip[9]) && (ip[8] == ip[10])) {
- mRootPtr = &kernelDot;
+ // Mask in the bits indicating which coefficients in the
+ // color matrix are needed.
+ if (hasFloat) {
+ for (uint32_t i=0; i < 16; i++) {
+ if (fabs(fp[i]) != 0.f) {
+ key |= (uint32_t)(1 << (i + 16));
+ }
+ }
+ } else {
+ for (uint32_t i=0; i < 16; i++) {
+ if (ip[i] != 0) {
+ key |= (uint32_t)(1 << (i + 16));
+ }
}
}
+
+ // Look for a dot product where the r,g,b colums are the same
+ if ((ip[0] == ip[1]) && (ip[0] == ip[2]) &&
+ (ip[4] == ip[5]) && (ip[4] == ip[6]) &&
+ (ip[8] == ip[9]) && (ip[8] == ip[10]) &&
+ (ip[12] == ip[13]) && (ip[12] == ip[14])) {
+
+ key |= CM_DOT_MASK;
+ }
+
+ // Is alpha a simple copy
+ if (!(key & 0x08880000) && (ip[15] == 256)) {
+ key |= CM_COPY_ALPHA;
+ }
+
+ //ALOGE("build key %08x, %08x", (int32_t)(key >> 32), (int32_t)key);
+ return key;
}
-extern "C" void rsdIntrinsicColorMatrix4x4_K(void *dst, const void *src, const short *coef, uint32_t count);
-extern "C" void rsdIntrinsicColorMatrix3x3_K(void *dst, const void *src, const short *coef, uint32_t count);
-extern "C" void rsdIntrinsicColorMatrixDot_K(void *dst, const void *src, const short *coef, uint32_t count);
+#if defined(ARCH_ARM_HAVE_NEON)
+
+#define DEF_SYM(x) \
+ extern "C" uint32_t _N_ColorMatrix_##x; \
+ extern "C" uint32_t _N_ColorMatrix_##x##_end; \
+ extern "C" uint32_t _N_ColorMatrix_##x##_len;
+
+DEF_SYM(prefix)
+DEF_SYM(postfix1)
+DEF_SYM(postfix2)
+DEF_SYM(load_u8_4)
+DEF_SYM(load_u8_2)
+DEF_SYM(load_u8_1)
+DEF_SYM(store_u8_4)
+DEF_SYM(store_u8_2)
+DEF_SYM(store_u8_1)
+DEF_SYM(unpack_u8_4)
+DEF_SYM(unpack_u8_3)
+DEF_SYM(unpack_u8_2)
+DEF_SYM(unpack_u8_1)
+DEF_SYM(pack_u8_4)
+DEF_SYM(pack_u8_3)
+DEF_SYM(pack_u8_2)
+DEF_SYM(pack_u8_1)
+DEF_SYM(dot)
+DEF_SYM(add_0_u8)
+DEF_SYM(add_1_u8)
+DEF_SYM(add_2_u8)
+DEF_SYM(add_3_u8)
+
+#define ADD_CHUNK(x) \
+ memcpy(buf, &_N_ColorMatrix_##x, _N_ColorMatrix_##x##_len); \
+ buf += _N_ColorMatrix_##x##_len
+
+
+static uint8_t * addBranch(uint8_t *buf, const uint8_t *target, uint32_t condition) {
+ size_t off = (target - buf - 8) >> 2;
+ rsAssert(((off & 0xff000000) == 0) ||
+ ((off & 0xff000000) == 0xff000000));
+
+ uint32_t op = (condition << 28);
+ op |= 0xa << 24; // branch
+ op |= 0xffffff & off;
+ ((uint32_t *)buf)[0] = op;
+ return buf + 4;
+}
+
+static uint32_t encodeSIMDRegs(uint32_t vd, uint32_t vn, uint32_t vm, bool q, bool sz) {
+ rsAssert(vd < 32);
+ rsAssert(vm < 32);
+ rsAssert(vn < 32);
+
+ uint32_t op = ((vd & 0xf) << 12) | (((vd & 0x10) >> 4) << 22);
+ op |= (vm & 0xf) | (((vm & 0x10) >> 4) << 5);
+ op |= ((vn & 0xf) << 16) | (((vn & 0x10) >> 4) << 7);
+ if (q) op |= 1 << 6;
+ if (sz) op |= 1 << 8;
+ return op;
+}
+
+static uint8_t * addVMLAL_S16(uint8_t *buf, uint32_t dest_q, uint32_t src_d1, uint32_t src_d2, uint32_t src_d2_s) {
+ //vmlal.s16 Q#1, D#1, D#2[#]
+ uint32_t op = 0xf2900240 | encodeSIMDRegs(dest_q << 1, src_d1, src_d2 | (src_d2_s << 3), false, false);
+ ((uint32_t *)buf)[0] = op;
+ return buf + 4;
+}
+
+static uint8_t * addVMULL_S16(uint8_t *buf, uint32_t dest_q, uint32_t src_d1, uint32_t src_d2, uint32_t src_d2_s) {
+ //vmull.s16 Q#1, D#1, D#2[#]
+ uint32_t op = 0xf2900A40 | encodeSIMDRegs(dest_q << 1, src_d1, src_d2 | (src_d2_s << 3), false, false);
+ ((uint32_t *)buf)[0] = op;
+ return buf + 4;
+}
+#endif
+
+
+bool RsdCpuScriptIntrinsicColorMatrix::build(int64_t key) {
+#if defined(ARCH_ARM_HAVE_NEON)
+ mBufSize = 4096;
+ //StopWatch build_time("rs cm: build time");
+ mBuf = (uint8_t *)mmap(0, mBufSize, PROT_READ | PROT_WRITE,
+ MAP_PRIVATE | MAP_ANON, -1, 0);
+ if (!mBuf) {
+ return false;
+ }
+
+ uint8_t *buf = mBuf;
+
+ // Add the function prefix
+ // Store the address for the loop return
+ ADD_CHUNK(prefix);
+ uint8_t *buf2 = buf;
+
+ // Load the incoming r,g,b,a as needed
+ switch(key & CM_IN_VEC_SIZE_MASK) {
+ case 3:
+ ADD_CHUNK(load_u8_4);
+ if (key & CM_COPY_ALPHA) {
+ ADD_CHUNK(unpack_u8_3);
+ } else {
+ ADD_CHUNK(unpack_u8_4);
+ }
+ break;
+ case 2:
+ ADD_CHUNK(load_u8_4);
+ ADD_CHUNK(unpack_u8_3);
+ break;
+ case 1:
+ ADD_CHUNK(load_u8_2);
+ ADD_CHUNK(unpack_u8_2);
+ break;
+ case 0:
+ ADD_CHUNK(load_u8_1);
+ ADD_CHUNK(unpack_u8_1);
+ break;
+ }
+
+ // Add multiply and accumulate
+ // use MULL to init the output register,
+ // use MLAL from there
+ bool linit[4] = {false, false, false, false};
+ if (key & (1 << 16)) {
+ buf = addVMULL_S16(buf, 8, 24, 4, 0);
+ linit[0] = true;
+ }
+ if (!(key & CM_DOT_MASK)) {
+ if (key & (1 << 17)) {
+ buf = addVMULL_S16(buf, 9, 24, 4, 1);
+ linit[1] = true;
+ }
+ if (key & (1 << 18)) {
+ buf = addVMULL_S16(buf, 10, 24, 4, 2);
+ linit[2] = true;
+ }
+ }
+ if (key & (1 << 19)) {
+ buf = addVMULL_S16(buf, 11, 24, 4, 3);
+ linit[3] = true;
+ }
+
+ if (key & (1 << 20)) {
+ if (linit[0]) {
+ buf = addVMLAL_S16(buf, 8, 26, 5, 0);
+ } else {
+ buf = addVMULL_S16(buf, 8, 26, 5, 0);
+ linit[0] = true;
+ }
+ }
+ if (!(key & CM_DOT_MASK)) {
+ if (key & (1 << 21)) {
+ if (linit[1]) {
+ buf = addVMLAL_S16(buf, 9, 26, 5, 1);
+ } else {
+ buf = addVMULL_S16(buf, 9, 26, 5, 1);
+ linit[1] = true;
+ }
+ }
+ if (key & (1 << 22)) {
+ if (linit[2]) {
+ buf = addVMLAL_S16(buf, 10, 26, 5, 2);
+ } else {
+ buf = addVMULL_S16(buf, 10, 26, 5, 2);
+ linit[2] = true;
+ }
+ }
+ }
+ if (key & (1 << 23)) {
+ if (linit[3]) {
+ buf = addVMLAL_S16(buf, 11, 26, 5, 3);
+ } else {
+ buf = addVMULL_S16(buf, 11, 26, 5, 3);
+ linit[3] = true;
+ }
+ }
+
+ if (key & (1 << 24)) {
+ if (linit[0]) {
+ buf = addVMLAL_S16(buf, 8, 28, 6, 0);
+ } else {
+ buf = addVMULL_S16(buf, 8, 28, 6, 0);
+ linit[0] = true;
+ }
+ }
+ if (!(key & CM_DOT_MASK)) {
+ if (key & (1 << 25)) {
+ if (linit[1]) {
+ buf = addVMLAL_S16(buf, 9, 28, 6, 1);
+ } else {
+ buf = addVMULL_S16(buf, 9, 28, 6, 1);
+ linit[1] = true;
+ }
+ }
+ if (key & (1 << 26)) {
+ if (linit[2]) {
+ buf = addVMLAL_S16(buf, 10, 28, 6, 2);
+ } else {
+ buf = addVMULL_S16(buf, 10, 28, 6, 2);
+ linit[2] = true;
+ }
+ }
+ }
+ if (key & (1 << 27)) {
+ if (linit[3]) {
+ buf = addVMLAL_S16(buf, 11, 28, 6, 3);
+ } else {
+ buf = addVMULL_S16(buf, 11, 28, 6, 3);
+ linit[3] = true;
+ }
+ }
+
+ if (key & (1 << 28)) {
+ if (linit[0]) {
+ buf = addVMLAL_S16(buf, 8, 30, 7, 0);
+ } else {
+ buf = addVMULL_S16(buf, 8, 30, 7, 0);
+ linit[0] = true;
+ }
+ }
+ if (!(key & CM_DOT_MASK)) {
+ if (key & (1 << 29)) {
+ if (linit[1]) {
+ buf = addVMLAL_S16(buf, 9, 30, 7, 1);
+ } else {
+ buf = addVMULL_S16(buf, 9, 30, 7, 1);
+ linit[1] = true;
+ }
+ }
+ if (key & (1 << 30)) {
+ if (linit[2]) {
+ buf = addVMLAL_S16(buf, 10, 30, 7, 2);
+ } else {
+ buf = addVMULL_S16(buf, 10, 30, 7, 2);
+ linit[2] = true;
+ }
+ }
+ }
+ if (!(key & CM_COPY_ALPHA)) {
+ if (key & (1 << 31)) {
+ if (linit[3]) {
+ buf = addVMLAL_S16(buf, 11, 30, 7, 3);
+ } else {
+ buf = addVMULL_S16(buf, 11, 30, 7, 3);
+ linit[3] = true;
+ }
+ }
+ }
+
+ // If we have a dot product, perform the special pack.
+ if (key & CM_DOT_MASK) {
+ ADD_CHUNK(pack_u8_1);
+ ADD_CHUNK(dot);
+ } else {
+ switch(key & CM_IN_VEC_SIZE_MASK) {
+ case 3:
+ ADD_CHUNK(pack_u8_4);
+ break;
+ case 2:
+ ADD_CHUNK(pack_u8_3);
+ break;
+ case 1:
+ ADD_CHUNK(pack_u8_2);
+ break;
+ case 0:
+ ADD_CHUNK(pack_u8_1);
+ break;
+ }
+ }
+
+ // Write out result
+ switch(key & CM_IN_VEC_SIZE_MASK) {
+ case 3:
+ case 2:
+ ADD_CHUNK(store_u8_4);
+ break;
+ case 1:
+ ADD_CHUNK(store_u8_2);
+ break;
+ case 0:
+ ADD_CHUNK(store_u8_1);
+ break;
+ }
+
+ // Loop, branch, and cleanup
+ ADD_CHUNK(postfix1);
+ buf = addBranch(buf, buf2, 0x01);
+ ADD_CHUNK(postfix2);
+
+ int ret = mprotect(mBuf, mBufSize, PROT_READ | PROT_EXEC);
+ if (ret == -1) {
+ ALOGE("mprotect error %i", ret);
+ return false;
+ }
+
+ cacheflush((long)mBuf, (long)mBuf + mBufSize, 0);
+ return true;
+#else
+ return false;
+#endif
+}
+
+void RsdCpuScriptIntrinsicColorMatrix::setGlobalVar(uint32_t slot, const void *data,
+ size_t dataLength) {
+ switch(slot) {
+ case 0:
+ memcpy (fp, data, dataLength);
+ for(int ct=0; ct < 16; ct++) {
+ ip[ct] = (short)(fp[ct] * 256.f + 0.5f);
+ }
+ break;
+ case 1:
+ memcpy (fpa, data, dataLength);
+ for(int ct=0; ct < 4; ct++) {
+ ipa[ct] = (int)(fpa[ct] * 256.f + 0.5f);
+ }
+ break;
+ default:
+ rsAssert(0);
+ break;
+ }
+
+ mRootPtr = &kernel;
+}
+
static void One(const RsForEachStubParamStruct *p, uchar4 *out,
const uchar4 *py, const float* coeff) {
@@ -108,9 +494,9 @@
*out = convert_uchar4(sum);
}
-void RsdCpuScriptIntrinsicColorMatrix::kernel4x4(const RsForEachStubParamStruct *p,
- uint32_t xstart, uint32_t xend,
- uint32_t instep, uint32_t outstep) {
+void RsdCpuScriptIntrinsicColorMatrix::kernel(const RsForEachStubParamStruct *p,
+ uint32_t xstart, uint32_t xend,
+ uint32_t instep, uint32_t outstep) {
RsdCpuScriptIntrinsicColorMatrix *cp = (RsdCpuScriptIntrinsicColorMatrix *)p->usr;
uchar4 *out = (uchar4 *)p->out;
uchar4 *in = (uchar4 *)p->in;
@@ -118,17 +504,13 @@
uint32_t x2 = xend;
if(x2 > x1) {
-#if defined(ARCH_ARM_HAVE_VFP)
- if (gArchUseSIMD) {
- int32_t len = (x2 - x1) >> 2;
- if(len > 0) {
- rsdIntrinsicColorMatrix4x4_K(out, in, cp->ip, len);
- x1 += len << 2;
- out += len << 2;
- in += len << 2;
- }
+ int32_t len = (x2 - x1) >> 2;
+ if((cp->mOptKernel != NULL) && (len > 0)) {
+ cp->mOptKernel(out, in, cp->ip, len);
+ x1 += len << 2;
+ out += len << 2;
+ in += len << 2;
}
-#endif
while(x1 != x2) {
One(p, out++, in++, cp->fp);
@@ -137,83 +519,56 @@
}
}
-void RsdCpuScriptIntrinsicColorMatrix::kernel3x3(const RsForEachStubParamStruct *p,
- uint32_t xstart, uint32_t xend,
- uint32_t instep, uint32_t outstep) {
- RsdCpuScriptIntrinsicColorMatrix *cp = (RsdCpuScriptIntrinsicColorMatrix *)p->usr;
- uchar4 *out = (uchar4 *)p->out;
- uchar4 *in = (uchar4 *)p->in;
- uint32_t x1 = xstart;
- uint32_t x2 = xend;
+void RsdCpuScriptIntrinsicColorMatrix::preLaunch(
+ uint32_t slot, const Allocation * ain, Allocation * aout,
+ const void * usr, uint32_t usrLen, const RsScriptCall *sc) {
- if(x2 > x1) {
-#if defined(ARCH_ARM_HAVE_VFP)
- if (gArchUseSIMD) {
- int32_t len = (x2 - x1) >> 2;
- if(len > 0) {
- rsdIntrinsicColorMatrix3x3_K(out, in, cp->ip, len);
- x1 += len << 2;
- out += len << 2;
- in += len << 2;
- }
- }
-#endif
-
- while(x1 != x2) {
- One(p, out++, in++, cp->fp);
- x1++;
+ int64_t key = computeKey(ain->mHal.state.type->getElement(),
+ aout->mHal.state.type->getElement());
+ if ((mOptKernel == NULL) || (mLastKey != key)) {
+ if (mBuf) munmap(mBuf, mBufSize);
+ mBuf = NULL;
+ mOptKernel = NULL;
+ if (build(key)) {
+ mOptKernel = (void (*)(void *, const void *, const short *, uint32_t)) mBuf;
+ mLastKey = key;
}
}
}
-void RsdCpuScriptIntrinsicColorMatrix::kernelDot(const RsForEachStubParamStruct *p,
- uint32_t xstart, uint32_t xend,
- uint32_t instep, uint32_t outstep) {
- RsdCpuScriptIntrinsicColorMatrix *cp = (RsdCpuScriptIntrinsicColorMatrix *)p->usr;
- uchar4 *out = (uchar4 *)p->out;
- uchar4 *in = (uchar4 *)p->in;
- uint32_t x1 = xstart;
- uint32_t x2 = xend;
+void RsdCpuScriptIntrinsicColorMatrix::postLaunch(
+ uint32_t slot, const Allocation * ain, Allocation * aout,
+ const void * usr, uint32_t usrLen, const RsScriptCall *sc) {
- if(x2 > x1) {
-#if defined(ARCH_ARM_HAVE_VFP)
- if (gArchUseSIMD) {
- int32_t len = (x2 - x1) >> 2;
- if(len > 0) {
- rsdIntrinsicColorMatrixDot_K(out, in, cp->ip, len);
- x1 += len << 2;
- out += len << 2;
- in += len << 2;
- }
- }
-#endif
-
- while(x1 != x2) {
- One(p, out++, in++, cp->fp);
- x1++;
- }
- }
}
-
RsdCpuScriptIntrinsicColorMatrix::RsdCpuScriptIntrinsicColorMatrix(
RsdCpuReferenceImpl *ctx, const Script *s, const Element *e)
: RsdCpuScriptIntrinsic(ctx, s, e, RS_SCRIPT_INTRINSIC_ID_COLOR_MATRIX) {
+ mLastKey = 0;
+ mBuf = NULL;
+ mBufSize = 0;
+ mOptKernel = NULL;
const static float defaultMatrix[] = {
1.f, 0.f, 0.f, 0.f,
0.f, 1.f, 0.f, 0.f,
0.f, 0.f, 1.f, 0.f,
0.f, 0.f, 0.f, 1.f
};
+ const static float defaultAdd[] = {0.f, 0.f, 0.f, 0.f};
setGlobalVar(0, defaultMatrix, sizeof(defaultMatrix));
+ setGlobalVar(1, defaultAdd, sizeof(defaultAdd));
}
RsdCpuScriptIntrinsicColorMatrix::~RsdCpuScriptIntrinsicColorMatrix() {
+ if (mBuf) munmap(mBuf, mBufSize);
+ mBuf = NULL;
+ mOptKernel = NULL;
}
void RsdCpuScriptIntrinsicColorMatrix::populateScript(Script *s) {
- s->mHal.info.exportedVariableCount = 1;
+ s->mHal.info.exportedVariableCount = 2;
}
RsdCpuScriptImpl * rsdIntrinsic_ColorMatrix(RsdCpuReferenceImpl *ctx,
diff --git a/cpu_ref/rsCpuIntrinsicConvolve3x3.cpp b/cpu_ref/rsCpuIntrinsicConvolve3x3.cpp
index 82ae1a4..5862428 100644
--- a/cpu_ref/rsCpuIntrinsicConvolve3x3.cpp
+++ b/cpu_ref/rsCpuIntrinsicConvolve3x3.cpp
@@ -42,9 +42,24 @@
ObjectBaseRef<const Allocation> mAlloc;
ObjectBaseRef<const Element> mElement;
- static void kernel(const RsForEachStubParamStruct *p,
- uint32_t xstart, uint32_t xend,
- uint32_t instep, uint32_t outstep);
+ static void kernelU1(const RsForEachStubParamStruct *p,
+ uint32_t xstart, uint32_t xend,
+ uint32_t instep, uint32_t outstep);
+ static void kernelU2(const RsForEachStubParamStruct *p,
+ uint32_t xstart, uint32_t xend,
+ uint32_t instep, uint32_t outstep);
+ static void kernelU4(const RsForEachStubParamStruct *p,
+ uint32_t xstart, uint32_t xend,
+ uint32_t instep, uint32_t outstep);
+ static void kernelF1(const RsForEachStubParamStruct *p,
+ uint32_t xstart, uint32_t xend,
+ uint32_t instep, uint32_t outstep);
+ static void kernelF2(const RsForEachStubParamStruct *p,
+ uint32_t xstart, uint32_t xend,
+ uint32_t instep, uint32_t outstep);
+ static void kernelF4(const RsForEachStubParamStruct *p,
+ uint32_t xstart, uint32_t xend,
+ uint32_t instep, uint32_t outstep);
};
}
@@ -61,7 +76,11 @@
rsAssert(slot == 0);
memcpy (&mFp, data, dataLength);
for(int ct=0; ct < 9; ct++) {
- mIp[ct] = (short)(mFp[ct] * 255.f + 0.5f);
+ if (mFp[ct] >= 0) {
+ mIp[ct] = (short)(mFp[ct] * 256.f + 0.5f);
+ } else {
+ mIp[ct] = (short)(mFp[ct] * 256.f - 0.5f);
+ }
}
}
@@ -69,9 +88,9 @@
const void *y2, const short *coef, uint32_t count);
-static void ConvolveOne(const RsForEachStubParamStruct *p, uint32_t x, uchar4 *out,
- const uchar4 *py0, const uchar4 *py1, const uchar4 *py2,
- const float* coeff) {
+static void ConvolveOneU4(const RsForEachStubParamStruct *p, uint32_t x, uchar4 *out,
+ const uchar4 *py0, const uchar4 *py1, const uchar4 *py2,
+ const float* coeff) {
uint32_t x1 = rsMax((int32_t)x-1, 0);
uint32_t x2 = rsMin((int32_t)x+1, (int32_t)p->dimX-1);
@@ -91,9 +110,82 @@
*out = o;
}
-void RsdCpuScriptIntrinsicConvolve3x3::kernel(const RsForEachStubParamStruct *p,
- uint32_t xstart, uint32_t xend,
- uint32_t instep, uint32_t outstep) {
+static void ConvolveOneU2(const RsForEachStubParamStruct *p, uint32_t x, uchar2 *out,
+ const uchar2 *py0, const uchar2 *py1, const uchar2 *py2,
+ const float* coeff) {
+
+ uint32_t x1 = rsMax((int32_t)x-1, 0);
+ uint32_t x2 = rsMin((int32_t)x+1, (int32_t)p->dimX-1);
+
+ float2 px = convert_float2(py0[x1]) * coeff[0] +
+ convert_float2(py0[x]) * coeff[1] +
+ convert_float2(py0[x2]) * coeff[2] +
+ convert_float2(py1[x1]) * coeff[3] +
+ convert_float2(py1[x]) * coeff[4] +
+ convert_float2(py1[x2]) * coeff[5] +
+ convert_float2(py2[x1]) * coeff[6] +
+ convert_float2(py2[x]) * coeff[7] +
+ convert_float2(py2[x2]) * coeff[8];
+
+ px = clamp(px, 0.f, 255.f);
+ *out = convert_uchar2(px);
+}
+
+static void ConvolveOneU1(const RsForEachStubParamStruct *p, uint32_t x, uchar *out,
+ const uchar *py0, const uchar *py1, const uchar *py2,
+ const float* coeff) {
+
+ uint32_t x1 = rsMax((int32_t)x-1, 0);
+ uint32_t x2 = rsMin((int32_t)x+1, (int32_t)p->dimX-1);
+
+ float px = ((float)py0[x1]) * coeff[0] +
+ ((float)py0[x]) * coeff[1] +
+ ((float)py0[x2]) * coeff[2] +
+ ((float)py1[x1]) * coeff[3] +
+ ((float)py1[x]) * coeff[4] +
+ ((float)py1[x2]) * coeff[5] +
+ ((float)py2[x1]) * coeff[6] +
+ ((float)py2[x]) * coeff[7] +
+ ((float)py2[x2]) * coeff[8];
+ *out = clamp(px, 0.f, 255.f);
+}
+
+static void ConvolveOneF4(const RsForEachStubParamStruct *p, uint32_t x, float4 *out,
+ const float4 *py0, const float4 *py1, const float4 *py2,
+ const float* coeff) {
+
+ uint32_t x1 = rsMax((int32_t)x-1, 0);
+ uint32_t x2 = rsMin((int32_t)x+1, (int32_t)p->dimX-1);
+ *out = (py0[x1] * coeff[0]) + (py0[x] * coeff[1]) + (py0[x2] * coeff[2]) +
+ (py1[x1] * coeff[3]) + (py1[x] * coeff[4]) + (py1[x2] * coeff[5]) +
+ (py2[x1] * coeff[6]) + (py2[x] * coeff[7]) + (py2[x2] * coeff[8]);
+}
+
+static void ConvolveOneF2(const RsForEachStubParamStruct *p, uint32_t x, float2 *out,
+ const float2 *py0, const float2 *py1, const float2 *py2,
+ const float* coeff) {
+
+ uint32_t x1 = rsMax((int32_t)x-1, 0);
+ uint32_t x2 = rsMin((int32_t)x+1, (int32_t)p->dimX-1);
+ *out = (py0[x1] * coeff[0]) + (py0[x] * coeff[1]) + (py0[x2] * coeff[2]) +
+ (py1[x1] * coeff[3]) + (py1[x] * coeff[4]) + (py1[x2] * coeff[5]) +
+ (py2[x1] * coeff[6]) + (py2[x] * coeff[7]) + (py2[x2] * coeff[8]);
+}
+
+static void ConvolveOneF1(const RsForEachStubParamStruct *p, uint32_t x, float *out,
+ const float *py0, const float *py1, const float *py2,
+ const float* coeff) {
+
+ uint32_t x1 = rsMax((int32_t)x-1, 0);
+ uint32_t x2 = rsMin((int32_t)x+1, (int32_t)p->dimX-1);
+ *out = (py0[x1] * coeff[0]) + (py0[x] * coeff[1]) + (py0[x2] * coeff[2]) +
+ (py1[x1] * coeff[3]) + (py1[x] * coeff[4]) + (py1[x2] * coeff[5]) +
+ (py2[x1] * coeff[6]) + (py2[x] * coeff[7]) + (py2[x2] * coeff[8]);
+}
+
+void RsdCpuScriptIntrinsicConvolve3x3::kernelU4(const RsForEachStubParamStruct *p,
+ uint32_t xstart, uint32_t xend,
+ uint32_t instep, uint32_t outstep) {
RsdCpuScriptIntrinsicConvolve3x3 *cp = (RsdCpuScriptIntrinsicConvolve3x3 *)p->usr;
if (!cp->mAlloc.get()) {
@@ -113,7 +205,7 @@
uint32_t x1 = xstart;
uint32_t x2 = xend;
if(x1 == 0) {
- ConvolveOne(p, 0, out, py0, py1, py2, cp->mFp);
+ ConvolveOneU4(p, 0, out, py0, py1, py2, cp->mFp);
x1 ++;
out++;
}
@@ -131,7 +223,231 @@
#endif
while(x1 != x2) {
- ConvolveOne(p, x1, out, py0, py1, py2, cp->mFp);
+ ConvolveOneU4(p, x1, out, py0, py1, py2, cp->mFp);
+ out++;
+ x1++;
+ }
+ }
+}
+
+void RsdCpuScriptIntrinsicConvolve3x3::kernelU2(const RsForEachStubParamStruct *p,
+ uint32_t xstart, uint32_t xend,
+ uint32_t instep, uint32_t outstep) {
+ RsdCpuScriptIntrinsicConvolve3x3 *cp = (RsdCpuScriptIntrinsicConvolve3x3 *)p->usr;
+
+ if (!cp->mAlloc.get()) {
+ ALOGE("Convolve3x3 executed without input, skipping");
+ return;
+ }
+ const uchar *pin = (const uchar *)cp->mAlloc->mHal.drvState.lod[0].mallocPtr;
+ const size_t stride = cp->mAlloc->mHal.drvState.lod[0].stride;
+
+ uint32_t y1 = rsMin((int32_t)p->y + 1, (int32_t)(p->dimY-1));
+ uint32_t y2 = rsMax((int32_t)p->y - 1, 0);
+ const uchar2 *py0 = (const uchar2 *)(pin + stride * y2);
+ const uchar2 *py1 = (const uchar2 *)(pin + stride * p->y);
+ const uchar2 *py2 = (const uchar2 *)(pin + stride * y1);
+
+ uchar2 *out = (uchar2 *)p->out;
+ uint32_t x1 = xstart;
+ uint32_t x2 = xend;
+ if(x1 == 0) {
+ ConvolveOneU2(p, 0, out, py0, py1, py2, cp->mFp);
+ x1 ++;
+ out++;
+ }
+
+ if(x2 > x1) {
+#if 0//defined(ARCH_ARM_HAVE_NEON)
+ int32_t len = (x2 - x1 - 1) >> 1;
+ if(len > 0) {
+ rsdIntrinsicConvolve3x3_K(out, &py0[x1-1], &py1[x1-1], &py2[x1-1], cp->mIp, len);
+ x1 += len << 1;
+ out += len << 1;
+ }
+#endif
+
+ while(x1 != x2) {
+ ConvolveOneU2(p, x1, out, py0, py1, py2, cp->mFp);
+ out++;
+ x1++;
+ }
+ }
+}
+
+void RsdCpuScriptIntrinsicConvolve3x3::kernelU1(const RsForEachStubParamStruct *p,
+ uint32_t xstart, uint32_t xend,
+ uint32_t instep, uint32_t outstep) {
+ RsdCpuScriptIntrinsicConvolve3x3 *cp = (RsdCpuScriptIntrinsicConvolve3x3 *)p->usr;
+
+ if (!cp->mAlloc.get()) {
+ ALOGE("Convolve3x3 executed without input, skipping");
+ return;
+ }
+ const uchar *pin = (const uchar *)cp->mAlloc->mHal.drvState.lod[0].mallocPtr;
+ const size_t stride = cp->mAlloc->mHal.drvState.lod[0].stride;
+
+ uint32_t y1 = rsMin((int32_t)p->y + 1, (int32_t)(p->dimY-1));
+ uint32_t y2 = rsMax((int32_t)p->y - 1, 0);
+ const uchar *py0 = (const uchar *)(pin + stride * y2);
+ const uchar *py1 = (const uchar *)(pin + stride * p->y);
+ const uchar *py2 = (const uchar *)(pin + stride * y1);
+
+ uchar *out = (uchar *)p->out;
+ uint32_t x1 = xstart;
+ uint32_t x2 = xend;
+ if(x1 == 0) {
+ ConvolveOneU1(p, 0, out, py0, py1, py2, cp->mFp);
+ x1 ++;
+ out++;
+ }
+
+ if(x2 > x1) {
+#if 0//defined(ARCH_ARM_HAVE_NEON)
+ int32_t len = (x2 - x1 - 1) >> 1;
+ if(len > 0) {
+ rsdIntrinsicConvolve3x3_K(out, &py0[x1-1], &py1[x1-1], &py2[x1-1], cp->mIp, len);
+ x1 += len << 1;
+ out += len << 1;
+ }
+#endif
+
+ while(x1 != x2) {
+ ConvolveOneU1(p, x1, out, py0, py1, py2, cp->mFp);
+ out++;
+ x1++;
+ }
+ }
+}
+
+void RsdCpuScriptIntrinsicConvolve3x3::kernelF4(const RsForEachStubParamStruct *p,
+ uint32_t xstart, uint32_t xend,
+ uint32_t instep, uint32_t outstep) {
+ RsdCpuScriptIntrinsicConvolve3x3 *cp = (RsdCpuScriptIntrinsicConvolve3x3 *)p->usr;
+
+ if (!cp->mAlloc.get()) {
+ ALOGE("Convolve3x3 executed without input, skipping");
+ return;
+ }
+ const uchar *pin = (const uchar *)cp->mAlloc->mHal.drvState.lod[0].mallocPtr;
+ const size_t stride = cp->mAlloc->mHal.drvState.lod[0].stride;
+
+ uint32_t y1 = rsMin((int32_t)p->y + 1, (int32_t)(p->dimY-1));
+ uint32_t y2 = rsMax((int32_t)p->y - 1, 0);
+ const float4 *py0 = (const float4 *)(pin + stride * y2);
+ const float4 *py1 = (const float4 *)(pin + stride * p->y);
+ const float4 *py2 = (const float4 *)(pin + stride * y1);
+
+ float4 *out = (float4 *)p->out;
+ uint32_t x1 = xstart;
+ uint32_t x2 = xend;
+ if(x1 == 0) {
+ ConvolveOneF4(p, 0, out, py0, py1, py2, cp->mFp);
+ x1 ++;
+ out++;
+ }
+
+ if(x2 > x1) {
+#if 0//defined(ARCH_ARM_HAVE_NEON)
+ int32_t len = (x2 - x1 - 1) >> 1;
+ if(len > 0) {
+ rsdIntrinsicConvolve3x3_K(out, &py0[x1-1], &py1[x1-1], &py2[x1-1], cp->mIp, len);
+ x1 += len << 1;
+ out += len << 1;
+ }
+#endif
+
+ while(x1 != x2) {
+ ConvolveOneF4(p, x1, out, py0, py1, py2, cp->mFp);
+ out++;
+ x1++;
+ }
+ }
+}
+
+void RsdCpuScriptIntrinsicConvolve3x3::kernelF2(const RsForEachStubParamStruct *p,
+ uint32_t xstart, uint32_t xend,
+ uint32_t instep, uint32_t outstep) {
+ RsdCpuScriptIntrinsicConvolve3x3 *cp = (RsdCpuScriptIntrinsicConvolve3x3 *)p->usr;
+
+ if (!cp->mAlloc.get()) {
+ ALOGE("Convolve3x3 executed without input, skipping");
+ return;
+ }
+ const uchar *pin = (const uchar *)cp->mAlloc->mHal.drvState.lod[0].mallocPtr;
+ const size_t stride = cp->mAlloc->mHal.drvState.lod[0].stride;
+
+ uint32_t y1 = rsMin((int32_t)p->y + 1, (int32_t)(p->dimY-1));
+ uint32_t y2 = rsMax((int32_t)p->y - 1, 0);
+ const float2 *py0 = (const float2 *)(pin + stride * y2);
+ const float2 *py1 = (const float2 *)(pin + stride * p->y);
+ const float2 *py2 = (const float2 *)(pin + stride * y1);
+
+ float2 *out = (float2 *)p->out;
+ uint32_t x1 = xstart;
+ uint32_t x2 = xend;
+ if(x1 == 0) {
+ ConvolveOneF2(p, 0, out, py0, py1, py2, cp->mFp);
+ x1 ++;
+ out++;
+ }
+
+ if(x2 > x1) {
+#if 0//defined(ARCH_ARM_HAVE_NEON)
+ int32_t len = (x2 - x1 - 1) >> 1;
+ if(len > 0) {
+ rsdIntrinsicConvolve3x3_K(out, &py0[x1-1], &py1[x1-1], &py2[x1-1], cp->mIp, len);
+ x1 += len << 1;
+ out += len << 1;
+ }
+#endif
+
+ while(x1 != x2) {
+ ConvolveOneF2(p, x1, out, py0, py1, py2, cp->mFp);
+ out++;
+ x1++;
+ }
+ }
+}
+void RsdCpuScriptIntrinsicConvolve3x3::kernelF1(const RsForEachStubParamStruct *p,
+ uint32_t xstart, uint32_t xend,
+ uint32_t instep, uint32_t outstep) {
+ RsdCpuScriptIntrinsicConvolve3x3 *cp = (RsdCpuScriptIntrinsicConvolve3x3 *)p->usr;
+
+ if (!cp->mAlloc.get()) {
+ ALOGE("Convolve3x3 executed without input, skipping");
+ return;
+ }
+ const uchar *pin = (const uchar *)cp->mAlloc->mHal.drvState.lod[0].mallocPtr;
+ const size_t stride = cp->mAlloc->mHal.drvState.lod[0].stride;
+
+ uint32_t y1 = rsMin((int32_t)p->y + 1, (int32_t)(p->dimY-1));
+ uint32_t y2 = rsMax((int32_t)p->y - 1, 0);
+ const float *py0 = (const float *)(pin + stride * y2);
+ const float *py1 = (const float *)(pin + stride * p->y);
+ const float *py2 = (const float *)(pin + stride * y1);
+
+ float *out = (float *)p->out;
+ uint32_t x1 = xstart;
+ uint32_t x2 = xend;
+ if(x1 == 0) {
+ ConvolveOneF1(p, 0, out, py0, py1, py2, cp->mFp);
+ x1 ++;
+ out++;
+ }
+
+ if(x2 > x1) {
+#if 0//defined(ARCH_ARM_HAVE_NEON)
+ int32_t len = (x2 - x1 - 1) >> 1;
+ if(len > 0) {
+ rsdIntrinsicConvolve3x3_K(out, &py0[x1-1], &py1[x1-1], &py2[x1-1], cp->mIp, len);
+ x1 += len << 1;
+ out += len << 1;
+ }
+#endif
+
+ while(x1 != x2) {
+ ConvolveOneF1(p, x1, out, py0, py1, py2, cp->mFp);
out++;
x1++;
}
@@ -142,10 +458,36 @@
RsdCpuReferenceImpl *ctx, const Script *s, const Element *e)
: RsdCpuScriptIntrinsic(ctx, s, e, RS_SCRIPT_INTRINSIC_ID_CONVOLVE_3x3) {
- mRootPtr = &kernel;
+ if (e->getType() == RS_TYPE_FLOAT_32) {
+ switch(e->getVectorSize()) {
+ case 1:
+ mRootPtr = &kernelF1;
+ break;
+ case 2:
+ mRootPtr = &kernelF2;
+ break;
+ case 3:
+ case 4:
+ mRootPtr = &kernelF4;
+ break;
+ }
+ } else {
+ switch(e->getVectorSize()) {
+ case 1:
+ mRootPtr = &kernelU1;
+ break;
+ case 2:
+ mRootPtr = &kernelU2;
+ break;
+ case 3:
+ case 4:
+ mRootPtr = &kernelU4;
+ break;
+ }
+ }
for(int ct=0; ct < 9; ct++) {
mFp[ct] = 1.f / 9.f;
- mIp[ct] = (short)(mFp[ct] * 255.f + 0.5f);
+ mIp[ct] = (short)(mFp[ct] * 256.f + 0.5f);
}
}
diff --git a/cpu_ref/rsCpuIntrinsicConvolve5x5.cpp b/cpu_ref/rsCpuIntrinsicConvolve5x5.cpp
index b4932bd..11dda59 100644
--- a/cpu_ref/rsCpuIntrinsicConvolve5x5.cpp
+++ b/cpu_ref/rsCpuIntrinsicConvolve5x5.cpp
@@ -37,14 +37,29 @@
RsdCpuScriptIntrinsicConvolve5x5(RsdCpuReferenceImpl *ctx, const Script *s, const Element *e);
protected:
- float fp[28];
- short ip[28];
+ float mFp[28];
+ short mIp[28];
ObjectBaseRef<Allocation> alloc;
- static void kernel(const RsForEachStubParamStruct *p,
- uint32_t xstart, uint32_t xend,
- uint32_t instep, uint32_t outstep);
+ static void kernelU1(const RsForEachStubParamStruct *p,
+ uint32_t xstart, uint32_t xend,
+ uint32_t instep, uint32_t outstep);
+ static void kernelU2(const RsForEachStubParamStruct *p,
+ uint32_t xstart, uint32_t xend,
+ uint32_t instep, uint32_t outstep);
+ static void kernelU4(const RsForEachStubParamStruct *p,
+ uint32_t xstart, uint32_t xend,
+ uint32_t instep, uint32_t outstep);
+ static void kernelF1(const RsForEachStubParamStruct *p,
+ uint32_t xstart, uint32_t xend,
+ uint32_t instep, uint32_t outstep);
+ static void kernelF2(const RsForEachStubParamStruct *p,
+ uint32_t xstart, uint32_t xend,
+ uint32_t instep, uint32_t outstep);
+ static void kernelF4(const RsForEachStubParamStruct *p,
+ uint32_t xstart, uint32_t xend,
+ uint32_t instep, uint32_t outstep);
};
@@ -60,16 +75,20 @@
void RsdCpuScriptIntrinsicConvolve5x5::setGlobalVar(uint32_t slot,
const void *data, size_t dataLength) {
rsAssert(slot == 0);
- memcpy (&fp, data, dataLength);
+ memcpy (&mFp, data, dataLength);
for(int ct=0; ct < 25; ct++) {
- ip[ct] = (short)(fp[ct] * 255.f);
+ if (mFp[ct] >= 0) {
+ mIp[ct] = (short)(mFp[ct] * 256.f + 0.5f);
+ } else {
+ mIp[ct] = (short)(mFp[ct] * 256.f - 0.5f);
+ }
}
}
-static void One(const RsForEachStubParamStruct *p, uint32_t x, uchar4 *out,
- const uchar4 *py0, const uchar4 *py1, const uchar4 *py2, const uchar4 *py3, const uchar4 *py4,
- const float* coeff) {
+static void OneU4(const RsForEachStubParamStruct *p, uint32_t x, uchar4 *out,
+ const uchar4 *py0, const uchar4 *py1, const uchar4 *py2, const uchar4 *py3, const uchar4 *py4,
+ const float* coeff) {
uint32_t x0 = rsMax((int32_t)x-2, 0);
uint32_t x1 = rsMax((int32_t)x-1, 0);
@@ -106,23 +125,230 @@
convert_float4(py4[x2]) * coeff[22] +
convert_float4(py4[x3]) * coeff[23] +
convert_float4(py4[x4]) * coeff[24];
-
px = clamp(px, 0.f, 255.f);
- uchar4 o = {(uchar)px.x, (uchar)px.y, (uchar)px.z, (uchar)px.w};
- //if ((out[0].r != o.r) || (out[0].y != o.y) || (out[0].z != o.z) || (out[0].w != o.w)) {
- //ALOGE("x %i %i,%i,%i,%i %i,%i,%i,%i", x, o.x, o.y, o.z, o.w, out[0].x, out[0].y, out[0].z, out[0].w);
- //}
- //o.w = 0xff;
- out->xyzw = o.xyzw;
+ *out = convert_uchar4(px);
}
+static void OneU2(const RsForEachStubParamStruct *p, uint32_t x, uchar2 *out,
+ const uchar2 *py0, const uchar2 *py1, const uchar2 *py2, const uchar2 *py3, const uchar2 *py4,
+ const float* coeff) {
+
+ uint32_t x0 = rsMax((int32_t)x-2, 0);
+ uint32_t x1 = rsMax((int32_t)x-1, 0);
+ uint32_t x2 = x;
+ uint32_t x3 = rsMin((int32_t)x+1, (int32_t)(p->dimX-1));
+ uint32_t x4 = rsMin((int32_t)x+2, (int32_t)(p->dimX-1));
+
+ float2 px = convert_float2(py0[x0]) * coeff[0] +
+ convert_float2(py0[x1]) * coeff[1] +
+ convert_float2(py0[x2]) * coeff[2] +
+ convert_float2(py0[x3]) * coeff[3] +
+ convert_float2(py0[x4]) * coeff[4] +
+
+ convert_float2(py1[x0]) * coeff[5] +
+ convert_float2(py1[x1]) * coeff[6] +
+ convert_float2(py1[x2]) * coeff[7] +
+ convert_float2(py1[x3]) * coeff[8] +
+ convert_float2(py1[x4]) * coeff[9] +
+
+ convert_float2(py2[x0]) * coeff[10] +
+ convert_float2(py2[x1]) * coeff[11] +
+ convert_float2(py2[x2]) * coeff[12] +
+ convert_float2(py2[x3]) * coeff[13] +
+ convert_float2(py2[x4]) * coeff[14] +
+
+ convert_float2(py3[x0]) * coeff[15] +
+ convert_float2(py3[x1]) * coeff[16] +
+ convert_float2(py3[x2]) * coeff[17] +
+ convert_float2(py3[x3]) * coeff[18] +
+ convert_float2(py3[x4]) * coeff[19] +
+
+ convert_float2(py4[x0]) * coeff[20] +
+ convert_float2(py4[x1]) * coeff[21] +
+ convert_float2(py4[x2]) * coeff[22] +
+ convert_float2(py4[x3]) * coeff[23] +
+ convert_float2(py4[x4]) * coeff[24];
+ px = clamp(px, 0.f, 255.f);
+ *out = convert_uchar2(px);
+}
+
+static void OneU1(const RsForEachStubParamStruct *p, uint32_t x, uchar *out,
+ const uchar *py0, const uchar *py1, const uchar *py2, const uchar *py3, const uchar *py4,
+ const float* coeff) {
+
+ uint32_t x0 = rsMax((int32_t)x-2, 0);
+ uint32_t x1 = rsMax((int32_t)x-1, 0);
+ uint32_t x2 = x;
+ uint32_t x3 = rsMin((int32_t)x+1, (int32_t)(p->dimX-1));
+ uint32_t x4 = rsMin((int32_t)x+2, (int32_t)(p->dimX-1));
+
+ float px = (float)(py0[x0]) * coeff[0] +
+ (float)(py0[x1]) * coeff[1] +
+ (float)(py0[x2]) * coeff[2] +
+ (float)(py0[x3]) * coeff[3] +
+ (float)(py0[x4]) * coeff[4] +
+
+ (float)(py1[x0]) * coeff[5] +
+ (float)(py1[x1]) * coeff[6] +
+ (float)(py1[x2]) * coeff[7] +
+ (float)(py1[x3]) * coeff[8] +
+ (float)(py1[x4]) * coeff[9] +
+
+ (float)(py2[x0]) * coeff[10] +
+ (float)(py2[x1]) * coeff[11] +
+ (float)(py2[x2]) * coeff[12] +
+ (float)(py2[x3]) * coeff[13] +
+ (float)(py2[x4]) * coeff[14] +
+
+ (float)(py3[x0]) * coeff[15] +
+ (float)(py3[x1]) * coeff[16] +
+ (float)(py3[x2]) * coeff[17] +
+ (float)(py3[x3]) * coeff[18] +
+ (float)(py3[x4]) * coeff[19] +
+
+ (float)(py4[x0]) * coeff[20] +
+ (float)(py4[x1]) * coeff[21] +
+ (float)(py4[x2]) * coeff[22] +
+ (float)(py4[x3]) * coeff[23] +
+ (float)(py4[x4]) * coeff[24];
+ px = clamp(px, 0.f, 255.f);
+ *out = px;
+}
+
+static void OneF4(const RsForEachStubParamStruct *p, uint32_t x, float4 *out,
+ const float4 *py0, const float4 *py1, const float4 *py2, const float4 *py3, const float4 *py4,
+ const float* coeff) {
+
+ uint32_t x0 = rsMax((int32_t)x-2, 0);
+ uint32_t x1 = rsMax((int32_t)x-1, 0);
+ uint32_t x2 = x;
+ uint32_t x3 = rsMin((int32_t)x+1, (int32_t)(p->dimX-1));
+ uint32_t x4 = rsMin((int32_t)x+2, (int32_t)(p->dimX-1));
+
+ float4 px = py0[x0] * coeff[0] +
+ py0[x1] * coeff[1] +
+ py0[x2] * coeff[2] +
+ py0[x3] * coeff[3] +
+ py0[x4] * coeff[4] +
+
+ py1[x0] * coeff[5] +
+ py1[x1] * coeff[6] +
+ py1[x2] * coeff[7] +
+ py1[x3] * coeff[8] +
+ py1[x4] * coeff[9] +
+
+ py2[x0] * coeff[10] +
+ py2[x1] * coeff[11] +
+ py2[x2] * coeff[12] +
+ py2[x3] * coeff[13] +
+ py2[x4] * coeff[14] +
+
+ py3[x0] * coeff[15] +
+ py3[x1] * coeff[16] +
+ py3[x2] * coeff[17] +
+ py3[x3] * coeff[18] +
+ py3[x4] * coeff[19] +
+
+ py4[x0] * coeff[20] +
+ py4[x1] * coeff[21] +
+ py4[x2] * coeff[22] +
+ py4[x3] * coeff[23] +
+ py4[x4] * coeff[24];
+ *out = px;
+}
+
+static void OneF2(const RsForEachStubParamStruct *p, uint32_t x, float2 *out,
+ const float2 *py0, const float2 *py1, const float2 *py2, const float2 *py3, const float2 *py4,
+ const float* coeff) {
+
+ uint32_t x0 = rsMax((int32_t)x-2, 0);
+ uint32_t x1 = rsMax((int32_t)x-1, 0);
+ uint32_t x2 = x;
+ uint32_t x3 = rsMin((int32_t)x+1, (int32_t)(p->dimX-1));
+ uint32_t x4 = rsMin((int32_t)x+2, (int32_t)(p->dimX-1));
+
+ float2 px = py0[x0] * coeff[0] +
+ py0[x1] * coeff[1] +
+ py0[x2] * coeff[2] +
+ py0[x3] * coeff[3] +
+ py0[x4] * coeff[4] +
+
+ py1[x0] * coeff[5] +
+ py1[x1] * coeff[6] +
+ py1[x2] * coeff[7] +
+ py1[x3] * coeff[8] +
+ py1[x4] * coeff[9] +
+
+ py2[x0] * coeff[10] +
+ py2[x1] * coeff[11] +
+ py2[x2] * coeff[12] +
+ py2[x3] * coeff[13] +
+ py2[x4] * coeff[14] +
+
+ py3[x0] * coeff[15] +
+ py3[x1] * coeff[16] +
+ py3[x2] * coeff[17] +
+ py3[x3] * coeff[18] +
+ py3[x4] * coeff[19] +
+
+ py4[x0] * coeff[20] +
+ py4[x1] * coeff[21] +
+ py4[x2] * coeff[22] +
+ py4[x3] * coeff[23] +
+ py4[x4] * coeff[24];
+ *out = px;
+}
+
+static void OneF1(const RsForEachStubParamStruct *p, uint32_t x, float *out,
+ const float *py0, const float *py1, const float *py2, const float *py3, const float *py4,
+ const float* coeff) {
+
+ uint32_t x0 = rsMax((int32_t)x-2, 0);
+ uint32_t x1 = rsMax((int32_t)x-1, 0);
+ uint32_t x2 = x;
+ uint32_t x3 = rsMin((int32_t)x+1, (int32_t)(p->dimX-1));
+ uint32_t x4 = rsMin((int32_t)x+2, (int32_t)(p->dimX-1));
+
+ float px = py0[x0] * coeff[0] +
+ py0[x1] * coeff[1] +
+ py0[x2] * coeff[2] +
+ py0[x3] * coeff[3] +
+ py0[x4] * coeff[4] +
+
+ py1[x0] * coeff[5] +
+ py1[x1] * coeff[6] +
+ py1[x2] * coeff[7] +
+ py1[x3] * coeff[8] +
+ py1[x4] * coeff[9] +
+
+ py2[x0] * coeff[10] +
+ py2[x1] * coeff[11] +
+ py2[x2] * coeff[12] +
+ py2[x3] * coeff[13] +
+ py2[x4] * coeff[14] +
+
+ py3[x0] * coeff[15] +
+ py3[x1] * coeff[16] +
+ py3[x2] * coeff[17] +
+ py3[x3] * coeff[18] +
+ py3[x4] * coeff[19] +
+
+ py4[x0] * coeff[20] +
+ py4[x1] * coeff[21] +
+ py4[x2] * coeff[22] +
+ py4[x3] * coeff[23] +
+ py4[x4] * coeff[24];
+ *out = px;
+}
+
+
extern "C" void rsdIntrinsicConvolve5x5_K(void *dst, const void *y0, const void *y1,
const void *y2, const void *y3, const void *y4,
const short *coef, uint32_t count);
-void RsdCpuScriptIntrinsicConvolve5x5::kernel(const RsForEachStubParamStruct *p,
- uint32_t xstart, uint32_t xend,
- uint32_t instep, uint32_t outstep) {
+void RsdCpuScriptIntrinsicConvolve5x5::kernelU4(const RsForEachStubParamStruct *p,
+ uint32_t xstart, uint32_t xend,
+ uint32_t instep, uint32_t outstep) {
RsdCpuScriptIntrinsicConvolve5x5 *cp = (RsdCpuScriptIntrinsicConvolve5x5 *)p->usr;
if (!cp->alloc.get()) {
ALOGE("Convolve5x5 executed without input, skipping");
@@ -148,7 +374,7 @@
uint32_t x2 = xend;
while((x1 < x2) && (x1 < 2)) {
- One(p, x1, out, py0, py1, py2, py3, py4, cp->fp);
+ OneU4(p, x1, out, py0, py1, py2, py3, py4, cp->mFp);
out++;
x1++;
}
@@ -156,6 +382,55 @@
#if defined(ARCH_ARM_HAVE_VFP)
if(gArchUseSIMD && ((x1 + 3) < x2)) {
uint32_t len = (x2 - x1 - 3) >> 1;
+ rsdIntrinsicConvolve5x5_K(out, py0, py1, py2, py3, py4, cp->mIp, len);
+ out += len << 1;
+ x1 += len << 1;
+ }
+#endif
+
+ while(x1 < x2) {
+ OneU4(p, x1, out, py0, py1, py2, py3, py4, cp->mFp);
+ out++;
+ x1++;
+ }
+}
+
+void RsdCpuScriptIntrinsicConvolve5x5::kernelU2(const RsForEachStubParamStruct *p,
+ uint32_t xstart, uint32_t xend,
+ uint32_t instep, uint32_t outstep) {
+ RsdCpuScriptIntrinsicConvolve5x5 *cp = (RsdCpuScriptIntrinsicConvolve5x5 *)p->usr;
+ if (!cp->alloc.get()) {
+ ALOGE("Convolve5x5 executed without input, skipping");
+ return;
+ }
+ const uchar *pin = (const uchar *)cp->alloc->mHal.drvState.lod[0].mallocPtr;
+ const size_t stride = cp->alloc->mHal.drvState.lod[0].stride;
+
+ uint32_t y0 = rsMax((int32_t)p->y-2, 0);
+ uint32_t y1 = rsMax((int32_t)p->y-1, 0);
+ uint32_t y2 = p->y;
+ uint32_t y3 = rsMin((int32_t)p->y+1, (int32_t)(p->dimY-1));
+ uint32_t y4 = rsMin((int32_t)p->y+2, (int32_t)(p->dimY-1));
+
+ const uchar2 *py0 = (const uchar2 *)(pin + stride * y0);
+ const uchar2 *py1 = (const uchar2 *)(pin + stride * y1);
+ const uchar2 *py2 = (const uchar2 *)(pin + stride * y2);
+ const uchar2 *py3 = (const uchar2 *)(pin + stride * y3);
+ const uchar2 *py4 = (const uchar2 *)(pin + stride * y4);
+
+ uchar2 *out = (uchar2 *)p->out;
+ uint32_t x1 = xstart;
+ uint32_t x2 = xend;
+
+ while((x1 < x2) && (x1 < 2)) {
+ OneU2(p, x1, out, py0, py1, py2, py3, py4, cp->mFp);
+ out++;
+ x1++;
+ }
+
+#if 0//defined(ARCH_ARM_HAVE_NEON)
+ if((x1 + 3) < x2) {
+ uint32_t len = (x2 - x1 - 3) >> 1;
rsdIntrinsicConvolve5x5_K(out, py0, py1, py2, py3, py4, cp->ip, len);
out += len << 1;
x1 += len << 1;
@@ -163,21 +438,242 @@
#endif
while(x1 < x2) {
- One(p, x1, out, py0, py1, py2, py3, py4, cp->fp);
+ OneU2(p, x1, out, py0, py1, py2, py3, py4, cp->mFp);
out++;
x1++;
}
}
+void RsdCpuScriptIntrinsicConvolve5x5::kernelU1(const RsForEachStubParamStruct *p,
+ uint32_t xstart, uint32_t xend,
+ uint32_t instep, uint32_t outstep) {
+ RsdCpuScriptIntrinsicConvolve5x5 *cp = (RsdCpuScriptIntrinsicConvolve5x5 *)p->usr;
+ if (!cp->alloc.get()) {
+ ALOGE("Convolve5x5 executed without input, skipping");
+ return;
+ }
+ const uchar *pin = (const uchar *)cp->alloc->mHal.drvState.lod[0].mallocPtr;
+ const size_t stride = cp->alloc->mHal.drvState.lod[0].stride;
+
+ uint32_t y0 = rsMax((int32_t)p->y-2, 0);
+ uint32_t y1 = rsMax((int32_t)p->y-1, 0);
+ uint32_t y2 = p->y;
+ uint32_t y3 = rsMin((int32_t)p->y+1, (int32_t)(p->dimY-1));
+ uint32_t y4 = rsMin((int32_t)p->y+2, (int32_t)(p->dimY-1));
+
+ const uchar *py0 = (const uchar *)(pin + stride * y0);
+ const uchar *py1 = (const uchar *)(pin + stride * y1);
+ const uchar *py2 = (const uchar *)(pin + stride * y2);
+ const uchar *py3 = (const uchar *)(pin + stride * y3);
+ const uchar *py4 = (const uchar *)(pin + stride * y4);
+
+ uchar *out = (uchar *)p->out;
+ uint32_t x1 = xstart;
+ uint32_t x2 = xend;
+
+ while((x1 < x2) && (x1 < 2)) {
+ OneU1(p, x1, out, py0, py1, py2, py3, py4, cp->mFp);
+ out++;
+ x1++;
+ }
+
+#if 0//defined(ARCH_ARM_HAVE_NEON)
+ if((x1 + 3) < x2) {
+ uint32_t len = (x2 - x1 - 3) >> 1;
+ rsdIntrinsicConvolve5x5_K(out, py0, py1, py2, py3, py4, cp->ip, len);
+ out += len << 1;
+ x1 += len << 1;
+ }
+#endif
+
+ while(x1 < x2) {
+ OneU1(p, x1, out, py0, py1, py2, py3, py4, cp->mFp);
+ out++;
+ x1++;
+ }
+}
+
+void RsdCpuScriptIntrinsicConvolve5x5::kernelF4(const RsForEachStubParamStruct *p,
+ uint32_t xstart, uint32_t xend,
+ uint32_t instep, uint32_t outstep) {
+ RsdCpuScriptIntrinsicConvolve5x5 *cp = (RsdCpuScriptIntrinsicConvolve5x5 *)p->usr;
+ if (!cp->alloc.get()) {
+ ALOGE("Convolve5x5 executed without input, skipping");
+ return;
+ }
+ const uchar *pin = (const uchar *)cp->alloc->mHal.drvState.lod[0].mallocPtr;
+ const size_t stride = cp->alloc->mHal.drvState.lod[0].stride;
+
+ uint32_t y0 = rsMax((int32_t)p->y-2, 0);
+ uint32_t y1 = rsMax((int32_t)p->y-1, 0);
+ uint32_t y2 = p->y;
+ uint32_t y3 = rsMin((int32_t)p->y+1, (int32_t)(p->dimY-1));
+ uint32_t y4 = rsMin((int32_t)p->y+2, (int32_t)(p->dimY-1));
+
+ const float4 *py0 = (const float4 *)(pin + stride * y0);
+ const float4 *py1 = (const float4 *)(pin + stride * y1);
+ const float4 *py2 = (const float4 *)(pin + stride * y2);
+ const float4 *py3 = (const float4 *)(pin + stride * y3);
+ const float4 *py4 = (const float4 *)(pin + stride * y4);
+
+ float4 *out = (float4 *)p->out;
+ uint32_t x1 = xstart;
+ uint32_t x2 = xend;
+
+ while((x1 < x2) && (x1 < 2)) {
+ OneF4(p, x1, out, py0, py1, py2, py3, py4, cp->mFp);
+ out++;
+ x1++;
+ }
+
+#if 0//defined(ARCH_ARM_HAVE_NEON)
+ if((x1 + 3) < x2) {
+ uint32_t len = (x2 - x1 - 3) >> 1;
+ rsdIntrinsicConvolve5x5_K(out, py0, py1, py2, py3, py4, cp->ip, len);
+ out += len << 1;
+ x1 += len << 1;
+ }
+#endif
+
+ while(x1 < x2) {
+ OneF4(p, x1, out, py0, py1, py2, py3, py4, cp->mFp);
+ out++;
+ x1++;
+ }
+}
+
+void RsdCpuScriptIntrinsicConvolve5x5::kernelF2(const RsForEachStubParamStruct *p,
+ uint32_t xstart, uint32_t xend,
+ uint32_t instep, uint32_t outstep) {
+ RsdCpuScriptIntrinsicConvolve5x5 *cp = (RsdCpuScriptIntrinsicConvolve5x5 *)p->usr;
+ if (!cp->alloc.get()) {
+ ALOGE("Convolve5x5 executed without input, skipping");
+ return;
+ }
+ const uchar *pin = (const uchar *)cp->alloc->mHal.drvState.lod[0].mallocPtr;
+ const size_t stride = cp->alloc->mHal.drvState.lod[0].stride;
+
+ uint32_t y0 = rsMax((int32_t)p->y-2, 0);
+ uint32_t y1 = rsMax((int32_t)p->y-1, 0);
+ uint32_t y2 = p->y;
+ uint32_t y3 = rsMin((int32_t)p->y+1, (int32_t)(p->dimY-1));
+ uint32_t y4 = rsMin((int32_t)p->y+2, (int32_t)(p->dimY-1));
+
+ const float2 *py0 = (const float2 *)(pin + stride * y0);
+ const float2 *py1 = (const float2 *)(pin + stride * y1);
+ const float2 *py2 = (const float2 *)(pin + stride * y2);
+ const float2 *py3 = (const float2 *)(pin + stride * y3);
+ const float2 *py4 = (const float2 *)(pin + stride * y4);
+
+ float2 *out = (float2 *)p->out;
+ uint32_t x1 = xstart;
+ uint32_t x2 = xend;
+
+ while((x1 < x2) && (x1 < 2)) {
+ OneF2(p, x1, out, py0, py1, py2, py3, py4, cp->mFp);
+ out++;
+ x1++;
+ }
+
+#if 0//defined(ARCH_ARM_HAVE_NEON)
+ if((x1 + 3) < x2) {
+ uint32_t len = (x2 - x1 - 3) >> 1;
+ rsdIntrinsicConvolve5x5_K(out, py0, py1, py2, py3, py4, cp->ip, len);
+ out += len << 1;
+ x1 += len << 1;
+ }
+#endif
+
+ while(x1 < x2) {
+ OneF2(p, x1, out, py0, py1, py2, py3, py4, cp->mFp);
+ out++;
+ x1++;
+ }
+}
+
+void RsdCpuScriptIntrinsicConvolve5x5::kernelF1(const RsForEachStubParamStruct *p,
+ uint32_t xstart, uint32_t xend,
+ uint32_t instep, uint32_t outstep) {
+ RsdCpuScriptIntrinsicConvolve5x5 *cp = (RsdCpuScriptIntrinsicConvolve5x5 *)p->usr;
+ if (!cp->alloc.get()) {
+ ALOGE("Convolve5x5 executed without input, skipping");
+ return;
+ }
+ const uchar *pin = (const uchar *)cp->alloc->mHal.drvState.lod[0].mallocPtr;
+ const size_t stride = cp->alloc->mHal.drvState.lod[0].stride;
+
+ uint32_t y0 = rsMax((int32_t)p->y-2, 0);
+ uint32_t y1 = rsMax((int32_t)p->y-1, 0);
+ uint32_t y2 = p->y;
+ uint32_t y3 = rsMin((int32_t)p->y+1, (int32_t)(p->dimY-1));
+ uint32_t y4 = rsMin((int32_t)p->y+2, (int32_t)(p->dimY-1));
+
+ const float *py0 = (const float *)(pin + stride * y0);
+ const float *py1 = (const float *)(pin + stride * y1);
+ const float *py2 = (const float *)(pin + stride * y2);
+ const float *py3 = (const float *)(pin + stride * y3);
+ const float *py4 = (const float *)(pin + stride * y4);
+
+ float *out = (float *)p->out;
+ uint32_t x1 = xstart;
+ uint32_t x2 = xend;
+
+ while((x1 < x2) && (x1 < 2)) {
+ OneF1(p, x1, out, py0, py1, py2, py3, py4, cp->mFp);
+ out++;
+ x1++;
+ }
+
+#if 0//defined(ARCH_ARM_HAVE_NEON)
+ if((x1 + 3) < x2) {
+ uint32_t len = (x2 - x1 - 3) >> 1;
+ rsdIntrinsicConvolve5x5_K(out, py0, py1, py2, py3, py4, cp->ip, len);
+ out += len << 1;
+ x1 += len << 1;
+ }
+#endif
+
+ while(x1 < x2) {
+ OneF1(p, x1, out, py0, py1, py2, py3, py4, cp->mFp);
+ out++;
+ x1++;
+ }
+}
RsdCpuScriptIntrinsicConvolve5x5::RsdCpuScriptIntrinsicConvolve5x5(
RsdCpuReferenceImpl *ctx, const Script *s, const Element *e)
: RsdCpuScriptIntrinsic(ctx, s, e, RS_SCRIPT_INTRINSIC_ID_CONVOLVE_5x5) {
- mRootPtr = &kernel;
+ if (e->getType() == RS_TYPE_FLOAT_32) {
+ switch(e->getVectorSize()) {
+ case 1:
+ mRootPtr = &kernelF1;
+ break;
+ case 2:
+ mRootPtr = &kernelF2;
+ break;
+ case 3:
+ case 4:
+ mRootPtr = &kernelF4;
+ break;
+ }
+ } else {
+ switch(e->getVectorSize()) {
+ case 1:
+ mRootPtr = &kernelU1;
+ break;
+ case 2:
+ mRootPtr = &kernelU2;
+ break;
+ case 3:
+ case 4:
+ mRootPtr = &kernelU4;
+ break;
+ }
+ }
for(int ct=0; ct < 25; ct++) {
- fp[ct] = 1.f / 25.f;
- ip[ct] = (short)(fp[ct] * 255.f);
+ mFp[ct] = 1.f / 25.f;
+ mIp[ct] = (short)(mFp[ct] * 256.f);
}
}
diff --git a/cpu_ref/rsCpuIntrinsicHistogram.cpp b/cpu_ref/rsCpuIntrinsicHistogram.cpp
new file mode 100644
index 0000000..e2ac102
--- /dev/null
+++ b/cpu_ref/rsCpuIntrinsicHistogram.cpp
@@ -0,0 +1,245 @@
+/*
+ * Copyright (C) 2013 The Android Open Source Project
+ *
+ * Licensed under the Apache License, Version 2.0 (the "License");
+ * you may not use this file except in compliance with the License.
+ * You may obtain a copy of the License at
+ *
+ * http://www.apache.org/licenses/LICENSE-2.0
+ *
+ * Unless required by applicable law or agreed to in writing, software
+ * distributed under the License is distributed on an "AS IS" BASIS,
+ * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ * See the License for the specific language governing permissions and
+ * limitations under the License.
+ */
+
+#include "rsCpuIntrinsic.h"
+#include "rsCpuIntrinsicInlines.h"
+
+using namespace android;
+using namespace android::renderscript;
+
+namespace android {
+namespace renderscript {
+
+
+class RsdCpuScriptIntrinsicHistogram : public RsdCpuScriptIntrinsic {
+public:
+ virtual void populateScript(Script *);
+ virtual void invokeFreeChildren();
+
+ virtual void setGlobalVar(uint32_t slot, const void *data, size_t dataLength);
+ virtual void setGlobalObj(uint32_t slot, ObjectBase *data);
+
+ virtual ~RsdCpuScriptIntrinsicHistogram();
+ RsdCpuScriptIntrinsicHistogram(RsdCpuReferenceImpl *ctx, const Script *s, const Element *e);
+
+protected:
+ void preLaunch(uint32_t slot, const Allocation * ain,
+ Allocation * aout, const void * usr,
+ uint32_t usrLen, const RsScriptCall *sc);
+ void postLaunch(uint32_t slot, const Allocation * ain,
+ Allocation * aout, const void * usr,
+ uint32_t usrLen, const RsScriptCall *sc);
+
+
+ float mDot[4];
+ int mDotI[4];
+ int *mSums;
+ ObjectBaseRef<Allocation> mAllocOut;
+
+ static void kernelP1U4(const RsForEachStubParamStruct *p,
+ uint32_t xstart, uint32_t xend,
+ uint32_t instep, uint32_t outstep);
+ static void kernelP1U3(const RsForEachStubParamStruct *p,
+ uint32_t xstart, uint32_t xend,
+ uint32_t instep, uint32_t outstep);
+ static void kernelP1U2(const RsForEachStubParamStruct *p,
+ uint32_t xstart, uint32_t xend,
+ uint32_t instep, uint32_t outstep);
+ static void kernelP1L(const RsForEachStubParamStruct *p,
+ uint32_t xstart, uint32_t xend,
+ uint32_t instep, uint32_t outstep);
+ static void kernelP1U1(const RsForEachStubParamStruct *p,
+ uint32_t xstart, uint32_t xend,
+ uint32_t instep, uint32_t outstep);
+
+};
+
+}
+}
+
+void RsdCpuScriptIntrinsicHistogram::setGlobalObj(uint32_t slot, ObjectBase *data) {
+ rsAssert(slot == 1);
+ mAllocOut.set(static_cast<Allocation *>(data));
+}
+
+void RsdCpuScriptIntrinsicHistogram::setGlobalVar(uint32_t slot, const void *data, size_t dataLength) {
+ rsAssert(slot == 0);
+ rsAssert(dataLength == 16);
+ memcpy(mDot, data, 16);
+ mDotI[0] = (int)((mDot[0] * 256.f) + 0.5f);
+ mDotI[1] = (int)((mDot[1] * 256.f) + 0.5f);
+ mDotI[2] = (int)((mDot[2] * 256.f) + 0.5f);
+ mDotI[3] = (int)((mDot[3] * 256.f) + 0.5f);
+}
+
+
+
+void RsdCpuScriptIntrinsicHistogram::preLaunch(uint32_t slot, const Allocation * ain,
+ Allocation * aout, const void * usr,
+ uint32_t usrLen, const RsScriptCall *sc) {
+
+ const uint32_t threads = mCtx->getThreadCount();
+ const uint32_t vSize = mAllocOut->getType()->getElement()->getVectorSize();
+
+ switch (slot) {
+ case 0:
+ switch(mAllocOut->getType()->getElement()->getVectorSize()) {
+ case 1:
+ mRootPtr = &kernelP1U1;
+ break;
+ case 2:
+ mRootPtr = &kernelP1U2;
+ break;
+ case 3:
+ mRootPtr = &kernelP1U3;
+ break;
+ case 4:
+ mRootPtr = &kernelP1U4;
+ break;
+ }
+ break;
+ case 1:
+ mRootPtr = &kernelP1L;
+ break;
+ }
+ memset(mSums, 0, 256 * 4 * threads * vSize);
+}
+
+void RsdCpuScriptIntrinsicHistogram::postLaunch(uint32_t slot, const Allocation * ain,
+ Allocation * aout, const void * usr,
+ uint32_t usrLen, const RsScriptCall *sc) {
+
+ unsigned int *o = (unsigned int *)mAllocOut->mHal.drvState.lod[0].mallocPtr;
+ uint32_t threads = mCtx->getThreadCount();
+ uint32_t vSize = mAllocOut->getType()->getElement()->getVectorSize();
+
+ for (uint32_t ct=0; ct < (256 * vSize); ct++) {
+ o[ct] = mSums[ct];
+ for (uint32_t t=1; t < threads; t++) {
+ o[ct] += mSums[ct + 256 * vSize];
+ }
+ }
+}
+
+void RsdCpuScriptIntrinsicHistogram::kernelP1U4(const RsForEachStubParamStruct *p,
+ uint32_t xstart, uint32_t xend,
+ uint32_t instep, uint32_t outstep) {
+
+ RsdCpuScriptIntrinsicHistogram *cp = (RsdCpuScriptIntrinsicHistogram *)p->usr;
+ uchar *in = (uchar *)p->in;
+ int * sums = &cp->mSums[256 * 4 * p->lid];
+
+ for (uint32_t x = xstart; x < xend; x++) {
+ sums[(in[0] << 2) ] ++;
+ sums[(in[1] << 2) + 1] ++;
+ sums[(in[2] << 2) + 2] ++;
+ sums[(in[3] << 2) + 3] ++;
+ in += 4;
+ }
+}
+
+void RsdCpuScriptIntrinsicHistogram::kernelP1U3(const RsForEachStubParamStruct *p,
+ uint32_t xstart, uint32_t xend,
+ uint32_t instep, uint32_t outstep) {
+
+ RsdCpuScriptIntrinsicHistogram *cp = (RsdCpuScriptIntrinsicHistogram *)p->usr;
+ uchar *in = (uchar *)p->in;
+ int * sums = &cp->mSums[256 * 4 * p->lid];
+
+ for (uint32_t x = xstart; x < xend; x++) {
+ sums[(in[0] << 2) ] ++;
+ sums[(in[1] << 2) + 1] ++;
+ sums[(in[2] << 2) + 2] ++;
+ in += 4;
+ }
+}
+
+void RsdCpuScriptIntrinsicHistogram::kernelP1U2(const RsForEachStubParamStruct *p,
+ uint32_t xstart, uint32_t xend,
+ uint32_t instep, uint32_t outstep) {
+
+ RsdCpuScriptIntrinsicHistogram *cp = (RsdCpuScriptIntrinsicHistogram *)p->usr;
+ uchar *in = (uchar *)p->in;
+ int * sums = &cp->mSums[256 * 2 * p->lid];
+
+ for (uint32_t x = xstart; x < xend; x++) {
+ sums[(in[0] << 2) ] ++;
+ sums[(in[1] << 2) + 1] ++;
+ in += 2;
+ }
+}
+
+void RsdCpuScriptIntrinsicHistogram::kernelP1L(const RsForEachStubParamStruct *p,
+ uint32_t xstart, uint32_t xend,
+ uint32_t instep, uint32_t outstep) {
+
+ RsdCpuScriptIntrinsicHistogram *cp = (RsdCpuScriptIntrinsicHistogram *)p->usr;
+ uchar *in = (uchar *)p->in;
+ int * sums = &cp->mSums[256 * p->lid];
+
+ for (uint32_t x = xstart; x < xend; x++) {
+ int t = (cp->mDotI[0] * in[0]) +
+ (cp->mDotI[1] * in[1]) +
+ (cp->mDotI[2] * in[2]) +
+ (cp->mDotI[3] * in[3]);
+ sums[t >> 8] ++;
+ in += 4;
+ }
+}
+
+void RsdCpuScriptIntrinsicHistogram::kernelP1U1(const RsForEachStubParamStruct *p,
+ uint32_t xstart, uint32_t xend,
+ uint32_t instep, uint32_t outstep) {
+
+}
+
+
+RsdCpuScriptIntrinsicHistogram::RsdCpuScriptIntrinsicHistogram(RsdCpuReferenceImpl *ctx,
+ const Script *s, const Element *e)
+ : RsdCpuScriptIntrinsic(ctx, s, e, RS_SCRIPT_INTRINSIC_ID_BLUR) {
+
+ mRootPtr = NULL;
+ mSums = new int[256 * 4 * mCtx->getThreadCount()];
+ mDot[0] = 0.299f;
+ mDot[1] = 0.587f;
+ mDot[2] = 0.114f;
+ mDot[3] = 0;
+ mDotI[0] = (int)((mDot[0] * 256.f) + 0.5f);
+ mDotI[1] = (int)((mDot[1] * 256.f) + 0.5f);
+ mDotI[2] = (int)((mDot[2] * 256.f) + 0.5f);
+ mDotI[3] = (int)((mDot[3] * 256.f) + 0.5f);
+}
+
+RsdCpuScriptIntrinsicHistogram::~RsdCpuScriptIntrinsicHistogram() {
+ if (mSums) {
+ delete []mSums;
+ }
+}
+
+void RsdCpuScriptIntrinsicHistogram::populateScript(Script *s) {
+ s->mHal.info.exportedVariableCount = 2;
+}
+
+void RsdCpuScriptIntrinsicHistogram::invokeFreeChildren() {
+}
+
+
+RsdCpuScriptImpl * rsdIntrinsic_Histogram(RsdCpuReferenceImpl *ctx, const Script *s, const Element *e) {
+
+ return new RsdCpuScriptIntrinsicHistogram(ctx, s, e);
+}
+
+
diff --git a/cpu_ref/rsCpuIntrinsicInlines.h b/cpu_ref/rsCpuIntrinsicInlines.h
index d6644ca..fb5003a 100644
--- a/cpu_ref/rsCpuIntrinsicInlines.h
+++ b/cpu_ref/rsCpuIntrinsicInlines.h
@@ -52,55 +52,36 @@
};
-static inline int4 convert_int4(uchar4 i) {
- int4 f4 = {i.x, i.y, i.z, i.w};
- return f4;
-}
-
-static inline uint4 convert_uint4(uchar4 i) {
- uint4 f4 = {i.x, i.y, i.z, i.w};
- return f4;
-}
-
-static inline int4 convert_int4(float4 i) {
- int4 f4 = {i.x, i.y, i.z, i.w};
- return f4;
-}
-
-static inline short4 convert_short4(uchar4 i) {
- short4 f4 = {i.x, i.y, i.z, i.w};
- return f4;
-}
-
-static inline float4 convert_float4(uchar4 i) {
- float4 f4 = {i.x, i.y, i.z, i.w};
- return f4;
-}
-
-static inline float4 convert_float4(int4 i) {
- float4 f4 = {i.x, i.y, i.z, i.w};
- return f4;
-}
-
-static inline uchar4 convert_uchar4(short4 i) {
- uchar4 f4 = {(uchar)i.x, (uchar)i.y, (uchar)i.z, (uchar)i.w};
- return f4;
-}
-
-static inline uchar4 convert_uchar4(int4 i) {
- uchar4 f4 = {(uchar)i.x, (uchar)i.y, (uchar)i.z, (uchar)i.w};
- return f4;
-}
-
-static inline uchar4 convert_uchar4(uint4 i) {
- uchar4 f4 = {(uchar)i.x, (uchar)i.y, (uchar)i.z, (uchar)i.w};
- return f4;
-}
-
-static inline uchar4 convert_uchar4(float4 i) {
- uchar4 f4 = {(uchar)i.x, (uchar)i.y, (uchar)i.z, (uchar)i.w};
- return f4;
-}
+#define CVT_FUNC_2(typeout, typein) \
+static inline typeout##2 __attribute__((const, overloadable)) \
+ convert_##typeout##2(typein##2 i) { \
+ typeout##2 f = {(typeout)i.x, (typeout)i.y}; \
+ return f; \
+ } \
+static inline typeout##3 __attribute__((const, overloadable)) \
+ convert_##typeout##3(typein##3 i) { \
+ typeout##3 f = {(typeout)i.x, (typeout)i.y, (typeout)i.z}; \
+ return f; \
+ } \
+static inline typeout##4 __attribute__((const, overloadable)) \
+ convert_##typeout##4(typein##4 i) { \
+ typeout##4 f = {(typeout)i.x, (typeout)i.y, (typeout)i.z, (typeout)i.w}; \
+ return f; \
+ }
+#define CVT_FUNC(type) CVT_FUNC_2(type, uchar) \
+ CVT_FUNC_2(type, char) \
+ CVT_FUNC_2(type, ushort) \
+ CVT_FUNC_2(type, short) \
+ CVT_FUNC_2(type, uint) \
+ CVT_FUNC_2(type, int) \
+ CVT_FUNC_2(type, float)
+CVT_FUNC(char)
+CVT_FUNC(uchar)
+CVT_FUNC(short)
+CVT_FUNC(ushort)
+CVT_FUNC(int)
+CVT_FUNC(uint)
+CVT_FUNC(float)
static inline int4 clamp(int4 amount, int low, int high) {
@@ -121,4 +102,25 @@
return r;
}
+static inline int2 clamp(int2 amount, int low, int high) {
+ int2 r;
+ r.x = amount.x < low ? low : (amount.x > high ? high : amount.x);
+ r.y = amount.y < low ? low : (amount.y > high ? high : amount.y);
+ return r;
+}
+
+static inline float2 clamp(float2 amount, float low, float high) {
+ float2 r;
+ r.x = amount.x < low ? low : (amount.x > high ? high : amount.x);
+ r.y = amount.y < low ? low : (amount.y > high ? high : amount.y);
+ return r;
+}
+
+static inline int clamp(int amount, int low, int high) {
+ return amount < low ? low : (amount > high ? high : amount);
+}
+
+static inline float clamp(float amount, float low, float high) {
+ return amount < low ? low : (amount > high ? high : amount);
+}
diff --git a/cpu_ref/rsCpuIntrinsics_neon.S b/cpu_ref/rsCpuIntrinsics_neon.S
index c8dc9bf..52fd565 100644
--- a/cpu_ref/rsCpuIntrinsics_neon.S
+++ b/cpu_ref/rsCpuIntrinsics_neon.S
@@ -104,173 +104,6 @@
bx lr
END(rsdIntrinsicConvolve3x3_K)
-/*
- r0 = dst
- r1 = src
- r2 = matrix
- r3 = length
-*/
-ENTRY(rsdIntrinsicColorMatrix4x4_K)
- stmfd sp!, {r4, lr}
- vpush {q4-q7}
-
- vld1.16 {q2}, [r2]!
- vld1.16 {q3}, [r2]!
-
-1:
- vld4.8 {d0[0],d1[0],d2[0],d3[0]}, [r1]!
- vld4.8 {d0[1],d1[1],d2[1],d3[1]}, [r1]!
- vld4.8 {d0[2],d1[2],d2[2],d3[2]}, [r1]!
- vld4.8 {d0[3],d1[3],d2[3],d3[3]}, [r1]!
-
- vmovl.u8 q12, d0 /* R */
- vmovl.u8 q13, d1 /* G */
- vmovl.u8 q14, d2 /* B */
- vmovl.u8 q15, d3 /* A */
-
- vmull.s16 q8, d24, d4[0]
- vmull.s16 q9, d24, d4[1]
- vmull.s16 q10, d24, d4[2]
- vmull.s16 q11, d24, d4[3]
-
- vmlal.s16 q8, d26, d5[0]
- vmlal.s16 q9, d26, d5[1]
- vmlal.s16 q10, d26, d5[2]
- vmlal.s16 q11, d26, d5[3]
-
- vmlal.s16 q8, d28, d6[0]
- vmlal.s16 q9, d28, d6[1]
- vmlal.s16 q10, d28, d6[2]
- vmlal.s16 q11, d28, d6[3]
-
- vmlal.s16 q8, d30, d7[0]
- vmlal.s16 q9, d30, d7[1]
- vmlal.s16 q10, d30, d7[2]
- vmlal.s16 q11, d30, d7[3]
-
- vshrn.i32 d24, q8, #8
- vshrn.i32 d26, q9, #8
- vshrn.i32 d28, q10, #8
- vshrn.i32 d30, q11, #8
-
- vqmovun.s16 d0, q12
- vqmovun.s16 d1, q13
- vqmovun.s16 d2, q14
- vqmovun.s16 d3, q15
-
- vst4.8 {d0[0],d1[0],d2[0],d3[0]}, [r0]!
- vst4.8 {d0[1],d1[1],d2[1],d3[1]}, [r0]!
- vst4.8 {d0[2],d1[2],d2[2],d3[2]}, [r0]!
- vst4.8 {d0[3],d1[3],d2[3],d3[3]}, [r0]!
-
- subs r3, r3, #1
- bne 1b
-
- vpop {q4-q7}
- ldmfd sp!, {r4, lr}
- bx lr
-END(rsdIntrinsicColorMatrix4x4_K)
-
-/*
- r0 = dst
- r1 = src
- r2 = matrix
- r3 = length
-*/
-ENTRY(rsdIntrinsicColorMatrix3x3_K)
- stmfd sp!, {r4, lr}
- vpush {q4-q7}
-
- vld1.16 {q2}, [r2]!
- vld1.16 {q3}, [r2]!
-
-1:
- vld4.8 {d0[0],d1[0],d2[0],d3[0]}, [r1]!
- vld4.8 {d0[1],d1[1],d2[1],d3[1]}, [r1]!
- vld4.8 {d0[2],d1[2],d2[2],d3[2]}, [r1]!
- vld4.8 {d0[3],d1[3],d2[3],d3[3]}, [r1]!
-
- vmovl.u8 q12, d0
- vmovl.u8 q13, d1
- vmovl.u8 q14, d2
-
- vmull.s16 q8, d24, d4[0]
- vmull.s16 q9, d24, d4[1]
- vmull.s16 q10, d24, d4[2]
-
- vmlal.s16 q8, d26, d5[0]
- vmlal.s16 q9, d26, d5[1]
- vmlal.s16 q10, d26, d5[2]
-
- vmlal.s16 q8, d28, d6[0]
- vmlal.s16 q9, d28, d6[1]
- vmlal.s16 q10, d28, d6[2]
-
- vshrn.i32 d24, q8, #8
- vshrn.i32 d26, q9, #8
- vshrn.i32 d28, q10, #8
-
- vqmovun.s16 d0, q12
- vqmovun.s16 d1, q13
- vqmovun.s16 d2, q14
-
- vst4.8 {d0[0],d1[0],d2[0],d3[0]}, [r0]!
- vst4.8 {d0[1],d1[1],d2[1],d3[1]}, [r0]!
- vst4.8 {d0[2],d1[2],d2[2],d3[2]}, [r0]!
- vst4.8 {d0[3],d1[3],d2[3],d3[3]}, [r0]!
-
- subs r3, r3, #1
- bne 1b
-
- vpop {q4-q7}
- ldmfd sp!, {r4, lr}
- bx lr
-END(rsdIntrinsicColorMatrix3x3_K)
-
-/*
- r0 = dst
- r1 = src
- r2 = matrix
- r3 = length
-*/
-ENTRY(rsdIntrinsicColorMatrixDot_K)
- stmfd sp!, {r4, lr}
- vpush {q4-q7}
-
- vld1.16 {q2}, [r2]!
- vld1.16 {q3}, [r2]!
-
-1:
- vld4.8 {d0[0],d1[0],d2[0],d3[0]}, [r1]!
- vld4.8 {d0[1],d1[1],d2[1],d3[1]}, [r1]!
- vld4.8 {d0[2],d1[2],d2[2],d3[2]}, [r1]!
- vld4.8 {d0[3],d1[3],d2[3],d3[3]}, [r1]!
-
- vmovl.u8 q12, d0
- vmovl.u8 q13, d1
- vmovl.u8 q14, d2
-
- vmull.s16 q8, d24, d4[0]
- vmlal.s16 q8, d26, d5[0]
- vmlal.s16 q8, d28, d6[0]
- vshrn.i32 d24, q8, #8
- vqmovun.s16 d0, q12
- vmov.u8 d1, d0
- vmov.u8 d2, d0
-
- vst4.8 {d0[0],d1[0],d2[0],d3[0]}, [r0]!
- vst4.8 {d0[1],d1[1],d2[1],d3[1]}, [r0]!
- vst4.8 {d0[2],d1[2],d2[2],d3[2]}, [r0]!
- vst4.8 {d0[3],d1[3],d2[3],d3[3]}, [r0]!
-
- subs r3, r3, #1
- bne 1b
-
- vpop {q4-q7}
- ldmfd sp!, {r4, lr}
- bx lr
-END(rsdIntrinsicColorMatrixDot_K)
-
/*
static void OneVF(float4 *out, const uchar *ptrIn, int iStride,
diff --git a/cpu_ref/rsCpuIntrinsics_neon_ColorMatrix.S b/cpu_ref/rsCpuIntrinsics_neon_ColorMatrix.S
new file mode 100644
index 0000000..ce8c033
--- /dev/null
+++ b/cpu_ref/rsCpuIntrinsics_neon_ColorMatrix.S
@@ -0,0 +1,145 @@
+/*
+ * Copyright (C) 2013 The Android Open Source Project
+ *
+ * Licensed under the Apache License, Version 2.0 (the "License");
+ * you may not use this file except in compliance with the License.
+ * You may obtain a copy of the License at
+ *
+ * http://www.apache.org/licenses/LICENSE-2.0
+ *
+ * Unless required by applicable law or agreed to in writing, software
+ * distributed under the License is distributed on an "AS IS" BASIS,
+ * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ * See the License for the specific language governing permissions and
+ * limitations under the License.
+ */
+
+#include <machine/cpu-features.h>
+#include <machine/asm.h>
+
+#define SNIP_START(x) \
+ .globl x; x:
+
+#define SNIP_END(x) \
+ .globl x##_end; x##_end: \
+ .globl x##_len; x##_len: \
+ .word x##_end-x
+
+SNIP_START(_N_ColorMatrix_prefix)
+ stmfd sp!, {r4, lr}
+ vpush {q4-q7}
+ vld1.16 {q2}, [r2]!
+ vld1.16 {q3}, [r2]!
+ vld1.16 {d8}, [r2]!
+SNIP_END(_N_ColorMatrix_prefix)
+
+SNIP_START(_N_ColorMatrix_postfix1)
+ subs r3, r3, #1
+ #bne 1b
+SNIP_END(_N_ColorMatrix_postfix1)
+
+SNIP_START(_N_ColorMatrix_postfix2)
+
+ #mov r0, #0
+ #ldr r0, [r0]
+
+ vpop {q4-q7}
+ ldmfd sp!, {r4, lr}
+ bx lr
+SNIP_END(_N_ColorMatrix_postfix2)
+
+SNIP_START(_N_ColorMatrix_load_u8_4)
+ vld4.8 {d0[0],d1[0],d2[0],d3[0]}, [r1]!
+ vld4.8 {d0[1],d1[1],d2[1],d3[1]}, [r1]!
+ vld4.8 {d0[2],d1[2],d2[2],d3[2]}, [r1]!
+ vld4.8 {d0[3],d1[3],d2[3],d3[3]}, [r1]!
+SNIP_END(_N_ColorMatrix_load_u8_4)
+
+SNIP_START(_N_ColorMatrix_load_u8_2)
+ vld2.8 {d0[0],d1[0]}, [r1]!
+ vld2.8 {d0[1],d1[1]}, [r1]!
+ vld2.8 {d0[2],d1[2]}, [r1]!
+ vld2.8 {d0[3],d1[3]}, [r1]!
+SNIP_END(_N_ColorMatrix_load_u8_2)
+
+SNIP_START(_N_ColorMatrix_load_u8_1)
+ vld1.32 {d0}, [r1]!
+SNIP_END(_N_ColorMatrix_load_u8_1)
+
+SNIP_START(_N_ColorMatrix_store_u8_4)
+ vst4.8 {d0[0],d1[0],d2[0],d3[0]}, [r0]!
+ vst4.8 {d0[1],d1[1],d2[1],d3[1]}, [r0]!
+ vst4.8 {d0[2],d1[2],d2[2],d3[2]}, [r0]!
+ vst4.8 {d0[3],d1[3],d2[3],d3[3]}, [r0]!
+SNIP_END(_N_ColorMatrix_store_u8_4)
+
+SNIP_START(_N_ColorMatrix_store_u8_2)
+ vst2.8 {d0[0],d1[0]}, [r0]!
+ vst2.8 {d0[1],d1[1]}, [r0]!
+ vst2.8 {d0[2],d1[2]}, [r0]!
+ vst2.8 {d0[3],d1[3]}, [r0]!
+SNIP_END(_N_ColorMatrix_store_u8_2)
+
+SNIP_START(_N_ColorMatrix_store_u8_1)
+ vst1.32 {d0}, [r0]!
+SNIP_END(_N_ColorMatrix_store_u8_1)
+
+SNIP_START(_N_ColorMatrix_unpack_u8_4)
+ vmovl.u8 q12, d0 /* R */
+ vmovl.u8 q13, d1 /* G */
+ vmovl.u8 q14, d2 /* B */
+ vmovl.u8 q15, d3 /* A */
+SNIP_END(_N_ColorMatrix_unpack_u8_4)
+
+SNIP_START(_N_ColorMatrix_unpack_u8_3)
+ vmovl.u8 q12, d0 /* R */
+ vmovl.u8 q13, d1 /* G */
+ vmovl.u8 q14, d2 /* B */
+SNIP_END(_N_ColorMatrix_unpack_u8_3)
+
+SNIP_START(_N_ColorMatrix_unpack_u8_2)
+ vmovl.u8 q12, d0 /* R */
+ vmovl.u8 q13, d1 /* G */
+SNIP_END(_N_ColorMatrix_unpack_u8_2)
+
+SNIP_START(_N_ColorMatrix_unpack_u8_1)
+ vmovl.u8 q12, d0 /* R */
+SNIP_END(_N_ColorMatrix_unpack_u8_1)
+
+SNIP_START(_N_ColorMatrix_pack_u8_4)
+ vshrn.i32 d24, q8, #8
+ vshrn.i32 d26, q9, #8
+ vshrn.i32 d28, q10, #8
+ vshrn.i32 d30, q11, #8
+ vqmovun.s16 d0, q12
+ vqmovun.s16 d1, q13
+ vqmovun.s16 d2, q14
+ vqmovun.s16 d3, q15
+SNIP_END(_N_ColorMatrix_pack_u8_4)
+
+SNIP_START(_N_ColorMatrix_pack_u8_3)
+ vshrn.i32 d24, q8, #8
+ vshrn.i32 d26, q9, #8
+ vshrn.i32 d28, q10, #8
+ vqmovun.s16 d0, q12
+ vqmovun.s16 d1, q13
+ vqmovun.s16 d2, q14
+SNIP_END(_N_ColorMatrix_pack_u8_3)
+
+SNIP_START(_N_ColorMatrix_pack_u8_2)
+ vshrn.i32 d24, q8, #8
+ vshrn.i32 d26, q9, #8
+ vqmovun.s16 d0, q12
+ vqmovun.s16 d1, q13
+SNIP_END(_N_ColorMatrix_pack_u8_2)
+
+SNIP_START(_N_ColorMatrix_pack_u8_1)
+ vshrn.i32 d24, q8, #8
+ vqmovun.s16 d0, q12
+SNIP_END(_N_ColorMatrix_pack_u8_1)
+
+SNIP_START(_N_ColorMatrix_dot)
+ vmov.u8 d1, d0
+ vmov.u8 d2, d0
+SNIP_END(_N_ColorMatrix_dot)
+
diff --git a/cpu_ref/rsCpuScript.cpp b/cpu_ref/rsCpuScript.cpp
index 22e2f0e..0d34d96 100644
--- a/cpu_ref/rsCpuScript.cpp
+++ b/cpu_ref/rsCpuScript.cpp
@@ -14,23 +14,252 @@
* limitations under the License.
*/
-
-
#include "rsCpuCore.h"
-
#include "rsCpuScript.h"
#ifdef RS_COMPATIBILITY_LIB
+ #include <set>
+ #include <string>
#include <dlfcn.h>
#include <stdio.h>
+ #include <stdlib.h>
#include <string.h>
+ #include <unistd.h>
#else
#include <bcc/BCCContext.h>
#include <bcc/Renderscript/RSCompilerDriver.h>
#include <bcc/Renderscript/RSExecutable.h>
#include <bcc/Renderscript/RSInfo.h>
+ #include <bcinfo/MetadataExtractor.h>
+ #include <cutils/properties.h>
+
+ #include <sys/types.h>
+ #include <sys/wait.h>
+ #include <unistd.h>
#endif
+namespace {
+#ifdef RS_COMPATIBILITY_LIB
+
+// Create a len length string containing random characters from [A-Za-z0-9].
+static std::string getRandomString(size_t len) {
+ char buf[len + 1];
+ for (size_t i = 0; i < len; i++) {
+ uint32_t r = arc4random() & 0xffff;
+ r %= 62;
+ if (r < 26) {
+ // lowercase
+ buf[i] = 'a' + r;
+ } else if (r < 52) {
+ // uppercase
+ buf[i] = 'A' + (r - 26);
+ } else {
+ // Use a number
+ buf[i] = '0' + (r - 52);
+ }
+ }
+ buf[len] = '\0';
+ return std::string(buf);
+}
+
+// Attempt to load the shared library from origName, but then fall back to
+// creating the symlinked shared library if necessary (to ensure instancing).
+// This function returns the dlopen()-ed handle if successful.
+static void *loadSOHelper(const char *origName, const char *cacheDir,
+ const char *resName) {
+ // Keep track of which .so libraries have been loaded. Once a library is
+ // in the set (per-process granularity), we must instead make a symlink to
+ // the original shared object (randomly named .so file) and load that one
+ // instead. If we don't do this, we end up aliasing global data between
+ // the various Script instances (which are supposed to be completely
+ // independent).
+ static std::set<std::string> LoadedLibraries;
+
+ void *loaded = NULL;
+
+ // Skip everything if we don't even have the original library available.
+ if (access(origName, F_OK) != 0) {
+ return NULL;
+ }
+
+ // Common path is that we have not loaded this Script/library before.
+ if (LoadedLibraries.find(origName) == LoadedLibraries.end()) {
+ loaded = dlopen(origName, RTLD_NOW | RTLD_LOCAL);
+ if (loaded) {
+ LoadedLibraries.insert(origName);
+ }
+ return loaded;
+ }
+
+ // Construct an appropriately randomized filename for the symlink.
+ std::string newName(cacheDir);
+ newName.append("/com.android.renderscript.cache/librs.");
+ newName.append(resName);
+ newName.append("#");
+ newName.append(getRandomString(6)); // 62^6 potential filename variants.
+ newName.append(".so");
+
+ int r = symlink(origName, newName.c_str());
+ if (r != 0) {
+ ALOGE("Could not create symlink %s -> %s", newName.c_str(), origName);
+ return NULL;
+ }
+ loaded = dlopen(newName.c_str(), RTLD_NOW | RTLD_LOCAL);
+ r = unlink(newName.c_str());
+ if (r != 0) {
+ ALOGE("Could not unlink symlink %s", newName.c_str());
+ }
+ if (loaded) {
+ LoadedLibraries.insert(newName.c_str());
+ }
+
+ return loaded;
+}
+
+// Load the shared library referred to by cacheDir and resName. If we have
+// already loaded this library, we instead create a new symlink (in the
+// cache dir) and then load that. We then immediately destroy the symlink.
+// This is required behavior to implement script instancing for the support
+// library, since shared objects are loaded and de-duped by name only.
+static void *loadSharedLibrary(const char *cacheDir, const char *resName) {
+ void *loaded = NULL;
+ //arc4random_stir();
+#ifndef RS_SERVER
+ std::string scriptSOName(cacheDir);
+ size_t cutPos = scriptSOName.rfind("cache");
+ if (cutPos != std::string::npos) {
+ scriptSOName.erase(cutPos);
+ } else {
+ ALOGE("Found peculiar cacheDir (missing \"cache\"): %s", cacheDir);
+ }
+ scriptSOName.append("/lib/librs.");
+#else
+ std::string scriptSOName("lib");
+#endif
+ scriptSOName.append(resName);
+ scriptSOName.append(".so");
+
+ // We should check if we can load the library from the standard app
+ // location for shared libraries first.
+ loaded = loadSOHelper(scriptSOName.c_str(), cacheDir, resName);
+
+ if (loaded == NULL) {
+ ALOGE("Unable to open shared library (%s): %s",
+ scriptSOName.c_str(), dlerror());
+
+ // One final attempt to find the library in "/system/lib".
+ // We do this to allow bundled applications to use the compatibility
+ // library fallback path. Those applications don't have a private
+ // library path, so they need to install to the system directly.
+ // Note that this is really just a testing path.
+ android::String8 scriptSONameSystem("/system/lib/librs.");
+ scriptSONameSystem.append(resName);
+ scriptSONameSystem.append(".so");
+ loaded = loadSOHelper(scriptSONameSystem.c_str(), cacheDir,
+ resName);
+ if (loaded == NULL) {
+ ALOGE("Unable to open system shared library (%s): %s",
+ scriptSONameSystem.c_str(), dlerror());
+ }
+ }
+
+ return loaded;
+}
+
+
+#else
+static bool is_force_recompile() {
+#ifdef RS_SERVER
+ return false;
+#else
+ char buf[PROPERTY_VALUE_MAX];
+
+ // Re-compile if floating point precision has been overridden.
+ property_get("debug.rs.precision", buf, "");
+ if (buf[0] != '\0') {
+ return true;
+ }
+
+ // Re-compile if debug.rs.forcerecompile is set.
+ property_get("debug.rs.forcerecompile", buf, "0");
+ if ((::strcmp(buf, "1") == 0) || (::strcmp(buf, "true") == 0)) {
+ return true;
+ } else {
+ return false;
+ }
+#endif // RS_SERVER
+}
+
+//#define EXTERNAL_BCC_COMPILER 1
+#ifdef EXTERNAL_BCC_COMPILER
+const static char *BCC_EXE_PATH = "/system/bin/bcc";
+
+static bool compileBitcode(const char *cacheDir,
+ const char *resName,
+ const char *bitcode,
+ size_t bitcodeSize,
+ const char *core_lib) {
+ rsAssert(cacheDir && resName && bitcode && bitcodeSize && core_lib);
+
+ android::String8 bcFilename(cacheDir);
+ bcFilename.append("/");
+ bcFilename.append(resName);
+ bcFilename.append(".bc");
+ FILE *bcfile = fopen(bcFilename.string(), "w");
+ if (!bcfile) {
+ ALOGE("Could not write to %s", bcFilename.string());
+ return false;
+ }
+ size_t nwritten = fwrite(bitcode, 1, bitcodeSize, bcfile);
+ fclose(bcfile);
+ if (nwritten != bitcodeSize) {
+ ALOGE("Could not write %zu bytes to %s", bitcodeSize,
+ bcFilename.string());
+ return false;
+ }
+
+ pid_t pid = fork();
+ switch (pid) {
+ case -1: { // Error occurred (we attempt no recovery)
+ ALOGE("Couldn't fork for bcc compiler execution");
+ return false;
+ }
+ case 0: { // Child process
+ // Execute the bcc compiler.
+ execl(BCC_EXE_PATH,
+ BCC_EXE_PATH,
+ "-o", resName,
+ "-output_path", cacheDir,
+ "-bclib", core_lib,
+ bcFilename.string(),
+ (char *) NULL);
+ ALOGE("execl() failed: %s", strerror(errno));
+ abort();
+ return false;
+ }
+ default: { // Parent process (actual driver)
+ // Wait on child process to finish compiling the source.
+ int status = 0;
+ pid_t w = waitpid(pid, &status, 0);
+ if (w == -1) {
+ ALOGE("Could not wait for bcc compiler");
+ return false;
+ }
+
+ if (WIFEXITED(status) && WEXITSTATUS(status) == 0) {
+ return true;
+ }
+
+ ALOGE("bcc compiler terminated unexpectedly");
+ return false;
+ }
+ }
+}
+#endif // EXTERNAL_BCC_COMPILER
+
+#endif // !defined(RS_COMPATIBILITY_LIB)
+} // namespace
+
namespace android {
namespace renderscript {
@@ -112,7 +341,7 @@
mCtx->lockMutex();
#ifndef RS_COMPATIBILITY_LIB
- bcc::RSExecutable *exec;
+ bcc::RSExecutable *exec = NULL;
mCompilerContext = NULL;
mCompilerDriver = NULL;
@@ -135,7 +364,43 @@
mCompilerDriver->setRSRuntimeLookupFunction(lookupRuntimeStub);
mCompilerDriver->setRSRuntimeLookupContext(this);
- const char *core_lib = NULL;
+ // Run any compiler setup functions we have been provided with.
+ RSSetupCompilerCallback setupCompilerCallback =
+ mCtx->getSetupCompilerCallback();
+ if (setupCompilerCallback != NULL) {
+ setupCompilerCallback(mCompilerDriver);
+ }
+
+ const char *core_lib = bcc::RSInfo::LibCLCorePath;
+
+ bcinfo::MetadataExtractor ME((const char *) bitcode, bitcodeSize);
+ if (!ME.extract()) {
+ ALOGE("Could not extract metadata from bitcode");
+ return false;
+ }
+
+ enum bcinfo::RSFloatPrecision prec = ME.getRSFloatPrecision();
+ switch (prec) {
+ case bcinfo::RS_FP_Imprecise:
+ case bcinfo::RS_FP_Relaxed:
+#if defined(ARCH_ARM_HAVE_NEON)
+ // NEON-capable devices can use an accelerated math library for all
+ // reduced precision scripts.
+ core_lib = bcc::RSInfo::LibCLCoreNEONPath;
+#endif
+ break;
+ case bcinfo::RS_FP_Full:
+ break;
+ default:
+ ALOGE("Unknown precision for bitcode");
+ return false;
+ }
+
+#if defined(ARCH_X86_HAVE_SSE2)
+ // SSE2- or above capable devices will use an optimized library.
+ core_lib = bcc::RSInfo::LibCLCoreX86Path;
+#endif
+
RSSelectRTCallback selectRTCallback = mCtx->getSelectRTCallback();
if (selectRTCallback != NULL) {
core_lib = selectRTCallback((const char *)bitcode, bitcodeSize);
@@ -145,10 +410,29 @@
// Use the libclcore_debug.bc instead of the default library.
core_lib = bcc::RSInfo::LibCLCoreDebugPath;
mCompilerDriver->setDebugContext(true);
+ // Skip the cache lookup
+ } else if (!is_force_recompile()) {
+ // Attempt to just load the script from cache first if we can.
+ exec = mCompilerDriver->loadScript(cacheDir, resName,
+ (const char *)bitcode, bitcodeSize);
}
- exec = mCompilerDriver->build(*mCompilerContext, cacheDir, resName,
- (const char *)bitcode, bitcodeSize, core_lib,
- mCtx->getLinkRuntimeCallback());
+
+ if (exec == NULL) {
+#ifdef EXTERNAL_BCC_COMPILER
+ bool built = compileBitcode(cacheDir, resName, (const char *)bitcode,
+ bitcodeSize, core_lib);
+#else
+ bool built = mCompilerDriver->build(*mCompilerContext, cacheDir,
+ resName, (const char *)bitcode,
+ bitcodeSize, core_lib,
+ mCtx->getLinkRuntimeCallback());
+#endif // EXTERNAL_BCC_COMPILER
+ if (built) {
+ exec = mCompilerDriver->loadScript(cacheDir, resName,
+ (const char *)bitcode,
+ bitcodeSize);
+ }
+ }
if (exec == NULL) {
ALOGE("bcc: FAILS to prepare executable for '%s'", resName);
@@ -179,38 +463,7 @@
#else
-#ifndef RS_SERVER
- String8 scriptSOName(cacheDir);
- scriptSOName = scriptSOName.getPathDir();
- scriptSOName.append("/lib/librs.");
-#else
- String8 scriptSOName("lib");
-#endif
- scriptSOName.append(resName);
- scriptSOName.append(".so");
-
- //script->mHal.drv = drv;
-
- //ALOGV("Opening up shared object: %s", scriptSOName.string());
- mScriptSO = dlopen(scriptSOName.string(), RTLD_NOW | RTLD_LOCAL);
- if (mScriptSO == NULL) {
- ALOGE("Unable to open shared library (%s): %s",
- scriptSOName.string(), dlerror());
-
- // One final attempt to find the library in "/system/lib".
- // We do this to allow bundled applications to use the compatibility
- // library fallback path. Those applications don't have a private
- // library path, so they need to install to the system directly.
- String8 scriptSONameSystem("/system/lib/librs.");
- scriptSONameSystem.append(resName);
- scriptSONameSystem.append(".so");
- mScriptSO = dlopen(scriptSONameSystem.string(), RTLD_NOW | RTLD_LOCAL);
- if (mScriptSO == NULL) {
- ALOGE("Unable to open system shared library (%s): %s",
- scriptSONameSystem.string(), dlerror());
- goto error;
- }
- }
+ mScriptSO = loadSharedLibrary(cacheDir, resName);
if (mScriptSO) {
char line[MAXLINE];
@@ -403,6 +656,8 @@
if (mScriptSO == (void*)1) {
//rsdLookupRuntimeStub(script, "acos");
}
+ } else {
+ goto error;
}
#endif
@@ -433,6 +688,7 @@
// Copy info over to runtime
script->mHal.info.exportedFunctionCount = info->getExportFuncNames().size();
script->mHal.info.exportedVariableCount = info->getExportVarNames().size();
+ script->mHal.info.exportedForeachFuncList = info->getExportForeachFuncs().array();
script->mHal.info.exportedPragmaCount = info->getPragmas().size();
script->mHal.info.exportedPragmaKeyList =
const_cast<const char**>(mExecutable->getPragmaKeys().array());
@@ -473,11 +729,11 @@
// possible for this to occur if IO_OUTPUT/IO_INPUT with no bound surface
if (ain && (const uint8_t *)ain->mHal.drvState.lod[0].mallocPtr == NULL) {
- mCtx->getContext()->setError(RS_ERROR_BAD_SCRIPT, "rsForEach called with null allocations");
+ mCtx->getContext()->setError(RS_ERROR_BAD_SCRIPT, "rsForEach called with null in allocations");
return;
}
if (aout && (const uint8_t *)aout->mHal.drvState.lod[0].mallocPtr == NULL) {
- mCtx->getContext()->setError(RS_ERROR_BAD_SCRIPT, "rsForEach called with null allocations");
+ mCtx->getContext()->setError(RS_ERROR_BAD_SCRIPT, "rsForEach called with null out allocations");
return;
}
diff --git a/cpu_ref/rsd_cpu.h b/cpu_ref/rsd_cpu.h
index d5642aa..d81a145 100644
--- a/cpu_ref/rsd_cpu.h
+++ b/cpu_ref/rsd_cpu.h
@@ -28,12 +28,16 @@
namespace bcc {
+class RSCompilerDriver;
class RSScript;
-typedef llvm::Module* (*RSLinkRuntimeCallback) (bcc::RSScript *, llvm::Module *, llvm::Module *);
+typedef llvm::Module* (*RSLinkRuntimeCallback)
+ (bcc::RSScript *, llvm::Module *, llvm::Module *);
} // end namespace bcc;
typedef const char* (*RSSelectRTCallback) (const char*, size_t);
+
+typedef void (*RSSetupCompilerCallback) (bcc::RSCompilerDriver *);
#endif
namespace android {
@@ -119,6 +123,11 @@
virtual CpuScriptGroup * createScriptGroup(const ScriptGroup *sg) = 0;
virtual bool getInForEach() = 0;
+#ifndef RS_COMPATIBILITY_LIB
+ virtual void setSetupCompilerCallback(
+ RSSetupCompilerCallback pSetupCompilerCallback) = 0;
+ virtual RSSetupCompilerCallback getSetupCompilerCallback() const = 0;
+#endif
};
diff --git a/driver/Android.mk b/driver/Android.mk
new file mode 100644
index 0000000..271de29
--- /dev/null
+++ b/driver/Android.mk
@@ -0,0 +1,4 @@
+
+LOCAL_PATH:=$(call my-dir)
+
+include $(call all-makefiles-under,$(LOCAL_PATH))
diff --git a/driver/linkloader/tests/images/gen-testcases.sh b/driver/linkloader/tests/images/gen-testcases.sh
deleted file mode 100755
index 26c7cdf..0000000
--- a/driver/linkloader/tests/images/gen-testcases.sh
+++ /dev/null
@@ -1,41 +0,0 @@
-#!/bin/bash -e
-
-
-# Copyright (C) 2011-2012 The Android Open Source Project
-#
-# Licensed under the Apache License, Version 2.0 (the "License");
-# you may not use this file except in compliance with the License.
-# You may obtain a copy of the License at
-#
-# http://www.apache.org/licenses/LICENSE-2.0
-#
-# Unless required by applicable law or agreed to in writing, software
-# distributed under the License is distributed on an "AS IS" BASIS,
-# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
-# See the License for the specific language governing permissions and
-# limitations under the License.
-
-
-CYAN='\033[1;36m'
-RESET='\033[m'
-
-echo -e "${CYAN}Generating bitcode ...${RESET}"
-clang -emit-llvm -std=c89 -Wall -c test.c -o test.bc
-clang -emit-llvm -std=c89 -Wall -c simple-test.c -o simple-test.bc
-clang -emit-llvm -std=c89 -Wall -c rodata-test.c -o rodata-test.bc
-
-function gen_test_cases {
- echo -e "${CYAN}Generating for $1 ...${RESET}"
- llc -filetype=obj -relocation-model=static -mtriple $2 $3 test.bc -o test-$1.o
- llc -filetype=obj -relocation-model=static -mtriple $2 $3 simple-test.bc -o simple-test-$1.o
- llc -filetype=obj -relocation-model=static -mtriple $2 $3 rodata-test.bc -o rodata-test-$1.o
-}
-
-gen_test_cases arm armv7-none-linux-gnueabi
-gen_test_cases tegra2 armv7-none-linux-gnueabi '-mcpu=cortex-a9 -mattr=+vfp3'
-gen_test_cases thumb2 thumb-none-linux-gnueabi '-march=thumb -mattr=+thumb2'
-gen_test_cases thumb2lc thumb-none-linux-gnueabi '-mattr=+thumb2,+neonfp,+vfp3 -arm-long-calls'
-gen_test_cases thumb2lc-xoom thumb-none-linux-gnueabi '-mattr=+thumb2 -arm-long-calls'
-gen_test_cases x86_32 i686-none-linux
-gen_test_cases x86_64 x86_64-none-linux
-gen_test_cases mipsel mipsel-none-linux-gnueabi
diff --git a/driver/linkloader/tests/images/rodata-test.c b/driver/linkloader/tests/images/rodata-test.c
deleted file mode 100644
index d695d01..0000000
--- a/driver/linkloader/tests/images/rodata-test.c
+++ /dev/null
@@ -1,37 +0,0 @@
-/*
- * Copyright 2011, The Android Open Source Project
- *
- * Licensed under the Apache License, Version 2.0 (the "License");
- * you may not use this file except in compliance with the License.
- * You may obtain a copy of the License at
- *
- * http://www.apache.org/licenses/LICENSE-2.0
- *
- * Unless required by applicable law or agreed to in writing, software
- * distributed under the License is distributed on an "AS IS" BASIS,
- * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
- * See the License for the specific language governing permissions and
- * limitations under the License.
- */
-
-#include<stdio.h>
-
-static char const *const test_str[] = {
- "string 1",
- "string 2",
- "string 3",
- "long long long long long long string"
-};
-
-static size_t test_str_count = sizeof(test_str) / sizeof(char const *const);
-
-int main(){
- int i;
- printf("test_str: %p\n", &test_str);
- for (i = 0; i < test_str_count; ++i) {
- printf("%p\n", test_str[i]);
- printf("%s\n", test_str[i]);
- }
-
- return 0;
-}
diff --git a/driver/linkloader/tests/images/simple-test.c b/driver/linkloader/tests/images/simple-test.c
deleted file mode 100644
index c59201d..0000000
--- a/driver/linkloader/tests/images/simple-test.c
+++ /dev/null
@@ -1,52 +0,0 @@
-/*
- * Copyright 2011, The Android Open Source Project
- *
- * Licensed under the Apache License, Version 2.0 (the "License");
- * you may not use this file except in compliance with the License.
- * You may obtain a copy of the License at
- *
- * http://www.apache.org/licenses/LICENSE-2.0
- *
- * Unless required by applicable law or agreed to in writing, software
- * distributed under the License is distributed on an "AS IS" BASIS,
- * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
- * See the License for the specific language governing permissions and
- * limitations under the License.
- */
-
-#include <stdlib.h>
-#include <stdio.h>
-#include <time.h>
-
-int main() {
- srand(time(NULL));
-
- unsigned int ans = rand() % 100;
- unsigned int user = 100;
- unsigned int left = 0;
- unsigned int right = 99;
-
- printf("Hello, droid! Let's play a number guessing game!\n");
-
- while (user != ans) {
- printf("Please input a number [%d-%d]:\n", left, right);
-
- if (scanf("%u", &user) != 1) {
- break;
- }
-
- if (user < left || user > right) {
- /* Out of range, ignore this answer. */
- continue;
- } else if (user == ans) {
- printf("You got it!\n");
- break;
- } else if (user < ans) {
- left = user;
- } else {
- right = user;
- }
- }
-
- return 0;
-}
diff --git a/driver/linkloader/tests/images/test.c b/driver/linkloader/tests/images/test.c
deleted file mode 100644
index 01b69a1..0000000
--- a/driver/linkloader/tests/images/test.c
+++ /dev/null
@@ -1,68 +0,0 @@
-/*
- * Copyright 2011, The Android Open Source Project
- *
- * Licensed under the Apache License, Version 2.0 (the "License");
- * you may not use this file except in compliance with the License.
- * You may obtain a copy of the License at
- *
- * http://www.apache.org/licenses/LICENSE-2.0
- *
- * Unless required by applicable law or agreed to in writing, software
- * distributed under the License is distributed on an "AS IS" BASIS,
- * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
- * See the License for the specific language governing permissions and
- * limitations under the License.
- */
-
-#include<stdio.h>
-static void hello_function(const char *ptr){
- printf("%s", ptr);
-}
-int my_add(int para_x, int para_y){
- return para_x + para_y;
-}
-int global_z_i;
-double global_z_d;
-int global_big_z_i[1000];
-double global_big_z_d[1000];
-static int global_static_z_i;
-static double global_static_z_d;
-static int global_static_big_z_i[1000];
-static double global_static_big_z_d[1000];
-int global_z_i_init = 1;
-double global_z_d_init = 1.1;
-/*extern int extern_z_i; */
-/*extern double extern_z_d;*/
-int main(){
- static int local_static_z_i;
- static double local_static_z_d;
- static int local_static_z_i_init = 2;
- static double local_static_z_d_init = 2.2;
- local_static_z_i = local_static_z_i_init;
- local_static_z_d = local_static_z_d_init;
- printf("%d %f\n", local_static_z_i, local_static_z_d);
- printf("%d %f\n", local_static_z_i_init, local_static_z_d_init);
- hello_function("Hello world!1\n");
- hello_function("Hello world!2\n");
- hello_function("Hello world!3\n");
- global_z_i = my_add(1,2);
- global_z_d = 3.3;
- printf("%d %f\n", global_z_i, global_z_d);
- global_big_z_i[100] = 4;
- global_big_z_d[100] = 4.4;
- printf("%d %f\n", global_big_z_i[100], global_big_z_d[100]);
- global_static_z_i = my_add(2,1);
- global_static_z_d = 3.3;
- printf("%d %f\n", global_static_z_i, global_static_z_d);
- int local_z_i = global_static_z_i = global_z_i;
- double local_z_d = global_static_z_d = global_z_d;
- printf("%d %f\n", local_z_i, local_z_d);
- global_static_big_z_i[500] = 5;
- global_static_big_z_d[500] = 5.5;
- printf("%d %f\n", global_static_big_z_i[500], global_static_big_z_d[500]);
- global_z_i_init = 6;
- global_z_d_init = 6.6;
- printf("%d %f\n", global_z_i_init, global_z_d_init);
- /*printf("%d %f\n", extern_z_i, extern_z_d);*/
- return 0;
-}
diff --git a/driver/linkloader/tests/stubs/arm-stub.c b/driver/linkloader/tests/stubs/arm-stub.c
deleted file mode 100644
index 2138f9a..0000000
--- a/driver/linkloader/tests/stubs/arm-stub.c
+++ /dev/null
@@ -1,27 +0,0 @@
-#include <stdio.h>
-#include <stdlib.h>
-#include <stdint.h>
-
-
-// Note: The first instruction stands for ldr, which loads the data from
-// memory to the specified register. Notice that due to the pipeline design,
-// when ldr is executed, the program will be advanced by 8. So, to get our
-// address we should substract it by 4.
-
-uint32_t stub[] = {
- 0xe51ff004ul, // ldr pc, [pc, #-4]
- 0x00000000ul // address
-};
-
-int test() {
- printf("hello world!\n");
- return 5;
-}
-
-int main() {
- int (*f)() = (int (*)())stub;
- stub[1] = (uint32_t)(uintptr_t)test;
-
- printf("return = %d\n", f());
- return EXIT_SUCCESS;
-}
diff --git a/driver/linkloader/tests/stubs/mips-stub.c b/driver/linkloader/tests/stubs/mips-stub.c
deleted file mode 100644
index ff34d03..0000000
--- a/driver/linkloader/tests/stubs/mips-stub.c
+++ /dev/null
@@ -1,48 +0,0 @@
-#include <stdio.h>
-#include <stdlib.h>
-#include <stdint.h>
-
-// --------------
-// Register Usage
-// --------------
-// $0 zero
-// $1 at
-// $2-$3 function return value registers
-// $4-$7 function argument registers
-// $8-$15 temporary
-// $16-$23 saved register
-// $24-$25 temporary
-// $26-$27 os kernel
-// $28 global pointer
-// $29 stack pointer
-// $30 saved register
-// $31 return addres reigster
-
-// --------------------
-// Instruction Encoding
-// --------------------
-// lui: 0011 1100 000t tttt iiii iiii iiii iiii
-// ori: 0011 01ss ssst tttt iiii iiii iiii iiii
-// jr: 0000 00ss sss0 0000 0000 0000 0000 1000
-// nop: 0000 0000 0000 0000 0000 0000 0000 0000
-
-uint32_t stub[] = {
- 0x3c190000ul,
- 0x37390000ul,
- 0x03200008ul,
- 0x00000000ul
-};
-
-int test() {
- printf("hello world!\n");
- return 5;
-}
-
-int main() {
- int (*f)() = (int (*)())stub;
- stub[0] |= (((uint32_t)(uintptr_t)test) >> 16) & 0xffff;
- stub[1] |= (((uint32_t)(uintptr_t)test)) & 0xffff;
-
- printf("return = %d\n", f());
- return EXIT_SUCCESS;
-}
diff --git a/driver/linkloader/tests/stubs/stub-layout-test.cpp b/driver/linkloader/tests/stubs/stub-layout-test.cpp
deleted file mode 100644
index ba07cd2..0000000
--- a/driver/linkloader/tests/stubs/stub-layout-test.cpp
+++ /dev/null
@@ -1,53 +0,0 @@
-/*
- * Copyright 2011, The Android Open Source Project
- *
- * Licensed under the Apache License, Version 2.0 (the "License");
- * you may not use this file except in compliance with the License.
- * You may obtain a copy of the License at
- *
- * http://www.apache.org/licenses/LICENSE-2.0
- *
- * Unless required by applicable law or agreed to in writing, software
- * distributed under the License is distributed on an "AS IS" BASIS,
- * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
- * See the License for the specific language governing permissions and
- * limitations under the License.
- */
-
-#include <stdio.h>
-#include <stdlib.h>
-
-#include "StubLayout.h"
-
-void function1() {
- printf("hello ");
-}
-
-void function2() {
- printf("world!\n");
-}
-
-int main() {
- StubLayout stubs;
-
- void (*func1)() = (void (*)())stubs.allocateStub((void *)&function1);
- void (*func2)() = (void (*)())stubs.allocateStub((void *)&function2);
-
- if (!func1) {
- fprintf(stderr, "ERROR: Unable to allocate stub for function1\n");
- exit(EXIT_FAILURE);
- }
-
- if (!func2) {
- fprintf(stderr, "ERROR: Unable to allocate stub for function2\n");
- exit(EXIT_FAILURE);
- }
-
- function1();
- function2();
-
- func1();
- func2();
-
- return EXIT_SUCCESS;
-}
diff --git a/driver/rsdAllocation.cpp b/driver/rsdAllocation.cpp
index 2daafbe..19b080f 100644
--- a/driver/rsdAllocation.cpp
+++ b/driver/rsdAllocation.cpp
@@ -617,34 +617,6 @@
}
#ifndef RS_COMPATIBILITY_LIB
-void DrvAllocation::NewBufferListener::onFrameAvailable() {
- intptr_t ip = (intptr_t)alloc;
- rsc->sendMessageToClient(NULL, RS_MESSAGE_TO_CLIENT_NEW_BUFFER, ip, 0, true);
-}
-#endif
-
-void* rsdAllocationGetSurface(const Context *rsc, const Allocation *alloc) {
-#ifndef RS_COMPATIBILITY_LIB
- DrvAllocation *drv = (DrvAllocation *)alloc->mHal.drv;
-
- // Configure CpuConsumer to be in asynchronous mode
- drv->cpuConsumer = new CpuConsumer(2, false);
- sp<IGraphicBufferProducer> bp = drv->cpuConsumer->getProducerInterface();
- bp->incStrong(NULL);
-
- drv->mBufferListener = new DrvAllocation::NewBufferListener();
- drv->mBufferListener->rsc = rsc;
- drv->mBufferListener->alloc = alloc;
-
- drv->cpuConsumer->setFrameAvailableListener(drv->mBufferListener);
-
- return bp.get();
-#else
- return NULL;
-#endif
-}
-
-#ifndef RS_COMPATIBILITY_LIB
static bool IoGetBuffer(const Context *rsc, Allocation *alloc, ANativeWindow *nw) {
DrvAllocation *drv = (DrvAllocation *)alloc->mHal.drv;
@@ -721,24 +693,10 @@
int format = 0;
const Element *e = alloc->mHal.state.type->getElement();
- switch(e->getType()) {
- case RS_TYPE_UNSIGNED_8:
- switch (e->getVectorSize()) {
- case 1:
- rsAssert(e->getKind() == RS_KIND_PIXEL_A);
- format = PIXEL_FORMAT_A_8;
- break;
- case 4:
- rsAssert(e->getKind() == RS_KIND_PIXEL_RGBA);
- format = PIXEL_FORMAT_RGBA_8888;
- break;
- default:
- rsAssert(0);
- }
- break;
- default:
- rsAssert(0);
- }
+ rsAssert(e->getType() == RS_TYPE_UNSIGNED_8);
+ rsAssert(e->getVectorSize() == 4);
+ rsAssert(e->getKind() == RS_KIND_PIXEL_RGBA);
+ format = PIXEL_FORMAT_RGBA_8888;
r = native_window_set_buffers_format(nw, format);
if (r) {
@@ -793,33 +751,9 @@
void rsdAllocationIoReceive(const Context *rsc, Allocation *alloc) {
#ifndef RS_COMPATIBILITY_LIB
DrvAllocation *drv = (DrvAllocation *)alloc->mHal.drv;
-
- if (alloc->mHal.state.usageFlags & RS_ALLOCATION_USAGE_SCRIPT) {
- CpuConsumer::LockedBuffer lb;
- status_t ret = drv->cpuConsumer->lockNextBuffer(&lb);
- if (ret == OK) {
- if (drv->lb.data != NULL) {
- drv->cpuConsumer->unlockBuffer(drv->lb);
- }
- drv->lb = lb;
- alloc->mHal.drvState.lod[0].mallocPtr = drv->lb.data;
- alloc->mHal.drvState.lod[0].stride = drv->lb.stride *
- alloc->mHal.state.elementSizeBytes;
-
- if (alloc->mHal.state.yuv) {
- DeriveYUVLayout(alloc->mHal.state.yuv, &alloc->mHal.drvState);
- }
- } else if (ret == BAD_VALUE) {
- // No new frame, don't do anything
- } else {
- rsc->setError(RS_ERROR_DRIVER, "Error receiving IO input buffer.");
- }
-
- } else {
+ if (!(alloc->mHal.state.usageFlags & RS_ALLOCATION_USAGE_SCRIPT)) {
drv->surfaceTexture->updateTexImage();
}
-
-
#endif
}
@@ -1213,3 +1147,10 @@
}
}
}
+
+uint32_t rsdAllocationGrallocBits(const android::renderscript::Context *rsc,
+ android::renderscript::Allocation *alloc)
+{
+ return 0;
+}
+
diff --git a/driver/rsdAllocation.h b/driver/rsdAllocation.h
index 35999d3..ff47f03 100644
--- a/driver/rsdAllocation.h
+++ b/driver/rsdAllocation.h
@@ -29,7 +29,6 @@
#endif
#if !defined(RS_SERVER) && !defined(RS_COMPATIBILITY_LIB)
-#include "gui/CpuConsumer.h"
#include "gui/GLConsumer.h"
#endif
@@ -51,16 +50,6 @@
uint32_t renderTargetID;
#ifndef RS_COMPATIBILITY_LIB
- class NewBufferListener : public android::ConsumerBase::FrameAvailableListener {
- public:
- const android::renderscript::Context *rsc;
- const android::renderscript::Allocation *alloc;
-
- virtual void onFrameAvailable();
- };
- android::sp<NewBufferListener> mBufferListener;
-
-
GLenum glTarget;
GLenum glType;
GLenum glFormat;
@@ -79,11 +68,6 @@
RsdFrameBufferObj * readBackFBO;
ANativeWindow *wnd;
ANativeWindowBuffer *wndBuffer;
-
-#if !defined(RS_SERVER) && !defined(RS_COMPATIBILITY_LIB)
- android::sp< android::CpuConsumer > cpuConsumer;
- android::CpuConsumer::LockedBuffer lb;
-#endif
};
#ifndef RS_COMPATIBILITY_LIB
@@ -92,6 +76,8 @@
#endif
+uint32_t rsdAllocationGrallocBits(const android::renderscript::Context *rsc,
+ android::renderscript::Allocation *alloc);
bool rsdAllocationInit(const android::renderscript::Context *rsc,
android::renderscript::Allocation *alloc,
bool forceZero);
@@ -106,8 +92,6 @@
RsAllocationUsageType src);
void rsdAllocationMarkDirty(const android::renderscript::Context *rsc,
const android::renderscript::Allocation *alloc);
-void* rsdAllocationGetSurface(const android::renderscript::Context *rsc,
- const android::renderscript::Allocation *alloc);
void rsdAllocationSetSurface(const android::renderscript::Context *rsc,
android::renderscript::Allocation *alloc, ANativeWindow *nw);
void rsdAllocationIoSend(const android::renderscript::Context *rsc,
diff --git a/driver/rsdCore.cpp b/driver/rsdCore.cpp
index d69255b..2b473c4 100644
--- a/driver/rsdCore.cpp
+++ b/driver/rsdCore.cpp
@@ -20,6 +20,7 @@
#include "rsdAllocation.h"
#include "rsdBcc.h"
#ifndef RS_COMPATIBILITY_LIB
+ #include "MemChunk.h"
#include "rsdGL.h"
#include "rsdPath.h"
#include "rsdProgramStore.h"
@@ -88,10 +89,10 @@
{
rsdAllocationInit,
rsdAllocationDestroy,
+ rsdAllocationGrallocBits,
rsdAllocationResize,
rsdAllocationSyncAll,
rsdAllocationMarkDirty,
- NATIVE_FUNC(rsdAllocationGetSurface),
NATIVE_FUNC(rsdAllocationSetSurface),
NATIVE_FUNC(rsdAllocationIoSend),
NATIVE_FUNC(rsdAllocationIoReceive),
@@ -189,7 +190,7 @@
}
rsc->mHal.drv = dc;
- dc->mCpuRef = RsdCpuReference::create((Context *)c, version_major, version_minor,
+ dc->mCpuRef = RsdCpuReference::create(rsc, version_major, version_minor,
&rsdLookupRuntimeStub, &LookupScript);
if (!dc->mCpuRef) {
ALOGE("RsdCpuReference::create for driver hal failed.");
@@ -197,6 +198,23 @@
return false;
}
+#ifndef RS_COMPATIBILITY_LIB
+ // Set a callback for compiler setup here.
+ if (false) {
+ dc->mCpuRef->setSetupCompilerCallback(NULL);
+ }
+
+ // Set a callback for switching MemChunk's allocator here.
+ // Note that the allocation function must return page-aligned memory, so
+ // that it can be mprotected properly (i.e. code should be written and
+ // later switched to read+execute only).
+ if (false) {
+ MemChunk::registerAllocFreeCallbacks(
+ rsc->mHal.funcs.allocRuntimeMem,
+ rsc->mHal.funcs.freeRuntimeMem);
+ }
+#endif
+
return true;
}
diff --git a/driver/rsdGL.cpp b/driver/rsdGL.cpp
index 0a23a44..c421ade 100644
--- a/driver/rsdGL.cpp
+++ b/driver/rsdGL.cpp
@@ -43,7 +43,6 @@
#include "rsdFrameBufferObj.h"
#include <gui/Surface.h>
-#include <gui/DummyConsumer.h>
using namespace android;
using namespace android::renderscript;
@@ -326,18 +325,13 @@
}
gGLContextCount++;
- // Create a BufferQueue with a fake consumer
- sp<BufferQueue> bq = new BufferQueue();
- bq->consumerConnect(new DummyConsumer());
- sp<Surface> stc(new Surface(static_cast<sp<IGraphicBufferProducer> >(bq)));
-
- dc->gl.egl.surfaceDefault = eglCreateWindowSurface(dc->gl.egl.display, dc->gl.egl.config,
- static_cast<ANativeWindow*>(stc.get()),
- NULL);
-
- checkEglError("eglCreateWindowSurface");
+ EGLint pbuffer_attribs[] = { EGL_WIDTH, 1, EGL_HEIGHT, 1, EGL_NONE };
+ rsc->setWatchdogGL("eglCreatePbufferSurface", __LINE__, __FILE__);
+ dc->gl.egl.surfaceDefault = eglCreatePbufferSurface(dc->gl.egl.display, dc->gl.egl.config,
+ pbuffer_attribs);
+ checkEglError("eglCreatePbufferSurface");
if (dc->gl.egl.surfaceDefault == EGL_NO_SURFACE) {
- ALOGE("eglCreateWindowSurface returned EGL_NO_SURFACE");
+ ALOGE("eglCreatePbufferSurface returned EGL_NO_SURFACE");
rsdGLShutdown(rsc);
rsc->setWatchdogGL(NULL, 0, NULL);
return false;
diff --git a/driver/rsdRuntimeStubs.cpp b/driver/rsdRuntimeStubs.cpp
index cb3a5b4..ac80ba3 100644
--- a/driver/rsdRuntimeStubs.cpp
+++ b/driver/rsdRuntimeStubs.cpp
@@ -141,13 +141,13 @@
#ifndef RS_COMPATIBILITY_LIB
static void SC_AllocationIoSend(Allocation *alloc) {
Context *rsc = RsdCpuReference::getTlsContext();
- rsdAllocationIoSend(rsc, alloc);
+ rsrAllocationIoSend(rsc, alloc);
}
static void SC_AllocationIoReceive(Allocation *alloc) {
Context *rsc = RsdCpuReference::getTlsContext();
- rsdAllocationIoReceive(rsc, alloc);
+ rsrAllocationIoReceive(rsc, alloc);
}
diff --git a/driver/runtime/Android.mk b/driver/runtime/Android.mk
new file mode 100755
index 0000000..d2ced77
--- /dev/null
+++ b/driver/runtime/Android.mk
@@ -0,0 +1,115 @@
+#
+# Copyright (C) 2013 The Android Open Source Project
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+# http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+#
+
+LOCAL_PATH := $(call my-dir)
+
+# C/LLVM-IR source files for the library
+clcore_base_files := \
+ rs_allocation.c \
+ rs_cl.c \
+ rs_core.c \
+ rs_element.c \
+ rs_mesh.c \
+ rs_matrix.c \
+ rs_program.c \
+ rs_sample.c \
+ rs_sampler.c \
+ convert.ll \
+ allocation.ll \
+ rsClamp.ll
+
+clcore_files := \
+ $(clcore_base_files) \
+ math.ll \
+ arch/generic.c \
+ arch/sqrt.c \
+ arch/dot_length.c
+
+clcore_neon_files := \
+ $(clcore_base_files) \
+ math.ll \
+ arch/neon.ll \
+ arch/sqrt.c \
+ arch/dot_length.c \
+ arch/clamp.c
+
+ifeq ($(ARCH_X86_HAVE_SSE2), true)
+ clcore_x86_files := \
+ $(clcore_base_files) \
+ arch/x86_generic.c \
+ arch/x86_clamp.ll \
+ arch/x86_math.ll
+
+ ifeq ($(ARCH_X86_HAVE_SSE3), true)
+ clcore_x86_files += arch/x86_dot_length.ll
+ else
+ # FIXME: without SSE3, it is still able to get better code through PSHUFD. But,
+ # so far, there is no such device with SSE2 only.
+ clcore_x86_files += arch/dot_length.c
+ endif
+endif
+
+ifeq "REL" "$(PLATFORM_VERSION_CODENAME)"
+ RS_VERSION := $(PLATFORM_SDK_VERSION)
+else
+ # Increment by 1 whenever this is not a final release build, since we want to
+ # be able to see the RS version number change during development.
+ # See build/core/version_defaults.mk for more information about this.
+ RS_VERSION := "(1 + $(PLATFORM_SDK_VERSION))"
+endif
+
+# Build the base version of the library
+include $(CLEAR_VARS)
+LOCAL_MODULE := libclcore.bc
+LOCAL_MODULE_TAGS := optional
+LOCAL_MODULE_CLASS := SHARED_LIBRARIES
+LOCAL_SRC_FILES := $(clcore_files)
+
+include $(LOCAL_PATH)/build_bc_lib.mk
+
+# Build a debug version of the library
+include $(CLEAR_VARS)
+LOCAL_MODULE := libclcore_debug.bc
+LOCAL_MODULE_TAGS := optional
+LOCAL_MODULE_CLASS := SHARED_LIBRARIES
+rs_debug_runtime := 1
+LOCAL_SRC_FILES := $(clcore_files)
+
+include $(LOCAL_PATH)/build_bc_lib.mk
+
+# Build an optimized version of the library if the device is SSE2- or above
+# capable.
+ifeq ($(ARCH_X86_HAVE_SSE2),true)
+include $(CLEAR_VARS)
+LOCAL_MODULE := libclcore_x86.bc
+LOCAL_MODULE_TAGS := optional
+LOCAL_MODULE_CLASS := SHARED_LIBRARIES
+LOCAL_SRC_FILES := $(clcore_x86_files)
+
+include $(LOCAL_PATH)/build_bc_lib.mk
+endif
+
+# Build a NEON-enabled version of the library (if possible)
+ifeq ($(ARCH_ARM_HAVE_NEON),true)
+ include $(CLEAR_VARS)
+ LOCAL_MODULE := libclcore_neon.bc
+ LOCAL_MODULE_TAGS := optional
+ LOCAL_MODULE_CLASS := SHARED_LIBRARIES
+ LOCAL_SRC_FILES := $(clcore_neon_files)
+ LOCAL_CFLAGS += -DARCH_ARM_HAVE_NEON
+
+ include $(LOCAL_PATH)/build_bc_lib.mk
+endif
diff --git a/driver/runtime/allocation.ll b/driver/runtime/allocation.ll
new file mode 100644
index 0000000..e1d6c7e
--- /dev/null
+++ b/driver/runtime/allocation.ll
@@ -0,0 +1,664 @@
+target datalayout = "e-p:32:32:32-i1:8:8-i8:8:8-i16:16:16-i32:32:32-i64:64:64-f32:32:32-f64:64:64-v64:64:64-v128:64:128-a0:0:64-n32-S64"
+target triple = "armv7-none-linux-gnueabi"
+
+declare i8* @rsOffset([1 x i32] %a.coerce, i32 %sizeOf, i32 %x, i32 %y, i32 %z)
+
+; The loads and stores in this file are annotated with RenderScript-specific
+; information for the type based alias analysis, such that the TBAA analysis
+; understands that loads and stores from two allocations with different types
+; can never access the same memory element. This is different from C, where
+; a char or uchar load/store is special as it can alias with about everything.
+;
+; The TBAA tree in this file has the the node "RenderScript TBAA" as its root.
+; This means all loads/stores that share this common root can be proven to not
+; alias. However, the alias analysis still has to assume MayAlias between
+; memory accesses in this file and memory accesses annotated with the C/C++
+; TBAA metadata.
+; If we can ensure that all accesses to elements loaded from RenderScript
+; allocations are either annotated with the RenderScript TBAA information or
+; not annotated at all, but never annotated with the C/C++ metadata, we
+; can add the RenderScript TBAA tree under the C/C++ TBAA tree. This enables
+; then the TBAA to prove that an access to data from the RenderScript allocation
+; does not alias with a load/store accessing something not part of a RenderScript
+; allocation.
+
+
+!14 = metadata !{metadata !"RenderScript TBAA"}
+!15 = metadata !{metadata !"allocation", metadata !14}
+
+!21 = metadata !{metadata !"char", metadata !15}
+define void @rsSetElementAtImpl_char([1 x i32] %a.coerce, i8 signext %val, i32 %x, i32 %y, i32 %z) #2 {
+ %1 = tail call i8* @rsOffset([1 x i32] %a.coerce, i32 1, i32 %x, i32 %y, i32 %z) #10
+ store i8 %val, i8* %1, align 1, !tbaa !21
+ ret void
+}
+
+define signext i8 @rsGetElementAtImpl_char([1 x i32] %a.coerce, i32 %x, i32 %y, i32 %z) #3 {
+ %1 = tail call i8* @rsOffset([1 x i32] %a.coerce, i32 1, i32 %x, i32 %y, i32 %z) #10
+ %2 = load i8* %1, align 1, !tbaa !21
+ ret i8 %2
+}
+
+!22 = metadata !{metadata !"char2", metadata !15}
+define void @rsSetElementAtImpl_char2([1 x i32] %a.coerce, <2 x i8> %val, i32 %x, i32 %y, i32 %z) #2 {
+ %1 = tail call i8* @rsOffset([1 x i32] %a.coerce, i32 2, i32 %x, i32 %y, i32 %z) #10
+ %2 = bitcast i8* %1 to <2 x i8>*
+ store <2 x i8> %val, <2 x i8>* %2, align 2, !tbaa !22
+ ret void
+}
+
+define <2 x i8> @rsGetElementAtImpl_char2([1 x i32] %a.coerce, i32 %x, i32 %y, i32 %z) #3 {
+ %1 = tail call i8* @rsOffset([1 x i32] %a.coerce, i32 2, i32 %x, i32 %y, i32 %z) #10
+ %2 = bitcast i8* %1 to <2 x i8>*
+ %3 = load <2 x i8>* %2, align 2, !tbaa !22
+ ret <2 x i8> %3
+}
+
+!23 = metadata !{metadata !"char3", metadata !15}
+define void @rsSetElementAtImpl_char3([1 x i32] %a.coerce, <3 x i8> %val, i32 %x, i32 %y, i32 %z) #2 {
+ %1 = tail call i8* @rsOffset([1 x i32] %a.coerce, i32 4, i32 %x, i32 %y, i32 %z) #10
+ %2 = shufflevector <3 x i8> %val, <3 x i8> undef, <4 x i32> <i32 0, i32 1, i32 2, i32 undef>
+ %3 = bitcast i8* %1 to <4 x i8>*
+ store <4 x i8> %2, <4 x i8>* %3, align 4, !tbaa !23
+ ret void
+}
+
+define <3 x i8> @rsGetElementAtImpl_char3([1 x i32] %a.coerce, i32 %x, i32 %y, i32 %z) #3 {
+ %1 = tail call i8* @rsOffset([1 x i32] %a.coerce, i32 4, i32 %x, i32 %y, i32 %z) #10
+ %2 = bitcast i8* %1 to <4 x i8>*
+ %3 = load <4 x i8>* %2, align 4, !tbaa !23
+ %4 = shufflevector <4 x i8> %3, <4 x i8> undef, <3 x i32> <i32 0, i32 1, i32 2>
+ ret <3 x i8> %4
+}
+
+!24 = metadata !{metadata !"char4", metadata !15}
+define void @rsSetElementAtImpl_char4([1 x i32] %a.coerce, <4 x i8> %val, i32 %x, i32 %y, i32 %z) #2 {
+ %1 = tail call i8* @rsOffset([1 x i32] %a.coerce, i32 4, i32 %x, i32 %y, i32 %z) #10
+ %2 = bitcast i8* %1 to <4 x i8>*
+ store <4 x i8> %val, <4 x i8>* %2, align 4, !tbaa !24
+ ret void
+}
+
+define <4 x i8> @rsGetElementAtImpl_char4([1 x i32] %a.coerce, i32 %x, i32 %y, i32 %z) #3 {
+ %1 = tail call i8* @rsOffset([1 x i32] %a.coerce, i32 4, i32 %x, i32 %y, i32 %z) #10
+ %2 = bitcast i8* %1 to <4 x i8>*
+ %3 = load <4 x i8>* %2, align 4, !tbaa !24
+ ret <4 x i8> %3
+}
+
+!25 = metadata !{metadata !"uchar", metadata !15}
+define void @rsSetElementAtImpl_uchar([1 x i32] %a.coerce, i8 zeroext %val, i32 %x, i32 %y, i32 %z) #2 {
+ %1 = tail call i8* @rsOffset([1 x i32] %a.coerce, i32 1, i32 %x, i32 %y, i32 %z) #10
+ store i8 %val, i8* %1, align 1, !tbaa !25
+ ret void
+}
+
+define zeroext i8 @rsGetElementAtImpl_uchar([1 x i32] %a.coerce, i32 %x, i32 %y, i32 %z) #3 {
+ %1 = tail call i8* @rsOffset([1 x i32] %a.coerce, i32 1, i32 %x, i32 %y, i32 %z) #10
+ %2 = load i8* %1, align 1, !tbaa !25
+ ret i8 %2
+}
+
+!26 = metadata !{metadata !"uchar2", metadata !15}
+define void @rsSetElementAtImpl_uchar2([1 x i32] %a.coerce, <2 x i8> %val, i32 %x, i32 %y, i32 %z) #2 {
+ %1 = tail call i8* @rsOffset([1 x i32] %a.coerce, i32 2, i32 %x, i32 %y, i32 %z) #10
+ %2 = bitcast i8* %1 to <2 x i8>*
+ store <2 x i8> %val, <2 x i8>* %2, align 2, !tbaa !26
+ ret void
+}
+
+define <2 x i8> @rsGetElementAtImpl_uchar2([1 x i32] %a.coerce, i32 %x, i32 %y, i32 %z) #3 {
+ %1 = tail call i8* @rsOffset([1 x i32] %a.coerce, i32 2, i32 %x, i32 %y, i32 %z) #10
+ %2 = bitcast i8* %1 to <2 x i8>*
+ %3 = load <2 x i8>* %2, align 2, !tbaa !26
+ ret <2 x i8> %3
+}
+
+!27 = metadata !{metadata !"uchar3", metadata !15}
+define void @rsSetElementAtImpl_uchar3([1 x i32] %a.coerce, <3 x i8> %val, i32 %x, i32 %y, i32 %z) #2 {
+ %1 = tail call i8* @rsOffset([1 x i32] %a.coerce, i32 4, i32 %x, i32 %y, i32 %z) #10
+ %2 = shufflevector <3 x i8> %val, <3 x i8> undef, <4 x i32> <i32 0, i32 1, i32 2, i32 undef>
+ %3 = bitcast i8* %1 to <4 x i8>*
+ store <4 x i8> %2, <4 x i8>* %3, align 4, !tbaa !27
+ ret void
+}
+
+define <3 x i8> @rsGetElementAtImpl_uchar3([1 x i32] %a.coerce, i32 %x, i32 %y, i32 %z) #3 {
+ %1 = tail call i8* @rsOffset([1 x i32] %a.coerce, i32 4, i32 %x, i32 %y, i32 %z) #10
+ %2 = bitcast i8* %1 to <4 x i8>*
+ %3 = load <4 x i8>* %2, align 4, !tbaa !27
+ %4 = shufflevector <4 x i8> %3, <4 x i8> undef, <3 x i32> <i32 0, i32 1, i32 2>
+ ret <3 x i8> %4
+}
+
+!28 = metadata !{metadata !"uchar4", metadata !15}
+define void @rsSetElementAtImpl_uchar4([1 x i32] %a.coerce, <4 x i8> %val, i32 %x, i32 %y, i32 %z) #2 {
+ %1 = tail call i8* @rsOffset([1 x i32] %a.coerce, i32 4, i32 %x, i32 %y, i32 %z) #10
+ %2 = bitcast i8* %1 to <4 x i8>*
+ store <4 x i8> %val, <4 x i8>* %2, align 4, !tbaa !28
+ ret void
+}
+
+define <4 x i8> @rsGetElementAtImpl_uchar4([1 x i32] %a.coerce, i32 %x, i32 %y, i32 %z) #3 {
+ %1 = tail call i8* @rsOffset([1 x i32] %a.coerce, i32 4, i32 %x, i32 %y, i32 %z) #10
+ %2 = bitcast i8* %1 to <4 x i8>*
+ %3 = load <4 x i8>* %2, align 4, !tbaa !28
+ ret <4 x i8> %3
+}
+
+!29 = metadata !{metadata !"short", metadata !15}
+define void @rsSetElementAtImpl_short([1 x i32] %a.coerce, i16 signext %val, i32 %x, i32 %y, i32 %z) #2 {
+ %1 = tail call i8* @rsOffset([1 x i32] %a.coerce, i32 2, i32 %x, i32 %y, i32 %z) #10
+ %2 = bitcast i8* %1 to i16*
+ store i16 %val, i16* %2, align 2, !tbaa !29
+ ret void
+}
+
+define signext i16 @rsGetElementAtImpl_short([1 x i32] %a.coerce, i32 %x, i32 %y, i32 %z) #3 {
+ %1 = tail call i8* @rsOffset([1 x i32] %a.coerce, i32 2, i32 %x, i32 %y, i32 %z) #10
+ %2 = bitcast i8* %1 to i16*
+ %3 = load i16* %2, align 2, !tbaa !29
+ ret i16 %3
+}
+
+!30 = metadata !{metadata !"short2", metadata !15}
+define void @rsSetElementAtImpl_short2([1 x i32] %a.coerce, <2 x i16> %val, i32 %x, i32 %y, i32 %z) #2 {
+ %1 = tail call i8* @rsOffset([1 x i32] %a.coerce, i32 4, i32 %x, i32 %y, i32 %z) #10
+ %2 = bitcast i8* %1 to <2 x i16>*
+ store <2 x i16> %val, <2 x i16>* %2, align 4, !tbaa !30
+ ret void
+}
+
+define <2 x i16> @rsGetElementAtImpl_short2([1 x i32] %a.coerce, i32 %x, i32 %y, i32 %z) #3 {
+ %1 = tail call i8* @rsOffset([1 x i32] %a.coerce, i32 4, i32 %x, i32 %y, i32 %z) #10
+ %2 = bitcast i8* %1 to <2 x i16>*
+ %3 = load <2 x i16>* %2, align 4, !tbaa !30
+ ret <2 x i16> %3
+}
+
+!31 = metadata !{metadata !"short3", metadata !15}
+define void @rsSetElementAtImpl_short3([1 x i32] %a.coerce, <3 x i16> %val, i32 %x, i32 %y, i32 %z) #2 {
+ %1 = tail call i8* @rsOffset([1 x i32] %a.coerce, i32 8, i32 %x, i32 %y, i32 %z) #10
+ %2 = shufflevector <3 x i16> %val, <3 x i16> undef, <4 x i32> <i32 0, i32 1, i32 2, i32 undef>
+ %3 = bitcast i8* %1 to <4 x i16>*
+ store <4 x i16> %2, <4 x i16>* %3, align 8, !tbaa !31
+ ret void
+}
+
+define <3 x i16> @rsGetElementAtImpl_short3([1 x i32] %a.coerce, i32 %x, i32 %y, i32 %z) #3 {
+ %1 = tail call i8* @rsOffset([1 x i32] %a.coerce, i32 8, i32 %x, i32 %y, i32 %z) #10
+ %2 = bitcast i8* %1 to <4 x i16>*
+ %3 = load <4 x i16>* %2, align 8, !tbaa !31
+ %4 = shufflevector <4 x i16> %3, <4 x i16> undef, <3 x i32> <i32 0, i32 1, i32 2>
+ ret <3 x i16> %4
+}
+
+!32 = metadata !{metadata !"short4", metadata !15}
+define void @rsSetElementAtImpl_short4([1 x i32] %a.coerce, <4 x i16> %val, i32 %x, i32 %y, i32 %z) #2 {
+ %1 = tail call i8* @rsOffset([1 x i32] %a.coerce, i32 8, i32 %x, i32 %y, i32 %z) #10
+ %2 = bitcast i8* %1 to <4 x i16>*
+ store <4 x i16> %val, <4 x i16>* %2, align 8, !tbaa !32
+ ret void
+}
+
+define <4 x i16> @rsGetElementAtImpl_short4([1 x i32] %a.coerce, i32 %x, i32 %y, i32 %z) #3 {
+ %1 = tail call i8* @rsOffset([1 x i32] %a.coerce, i32 8, i32 %x, i32 %y, i32 %z) #10
+ %2 = bitcast i8* %1 to <4 x i16>*
+ %3 = load <4 x i16>* %2, align 8, !tbaa !32
+ ret <4 x i16> %3
+}
+
+!33 = metadata !{metadata !"ushort", metadata !15}
+define void @rsSetElementAtImpl_ushort([1 x i32] %a.coerce, i16 zeroext %val, i32 %x, i32 %y, i32 %z) #2 {
+ %1 = tail call i8* @rsOffset([1 x i32] %a.coerce, i32 2, i32 %x, i32 %y, i32 %z) #10
+ %2 = bitcast i8* %1 to i16*
+ store i16 %val, i16* %2, align 2, !tbaa !33
+ ret void
+}
+
+define zeroext i16 @rsGetElementAtImpl_ushort([1 x i32] %a.coerce, i32 %x, i32 %y, i32 %z) #3 {
+ %1 = tail call i8* @rsOffset([1 x i32] %a.coerce, i32 2, i32 %x, i32 %y, i32 %z) #10
+ %2 = bitcast i8* %1 to i16*
+ %3 = load i16* %2, align 2, !tbaa !33
+ ret i16 %3
+}
+
+!34 = metadata !{metadata !"ushort2", metadata !15}
+define void @rsSetElementAtImpl_ushort2([1 x i32] %a.coerce, <2 x i16> %val, i32 %x, i32 %y, i32 %z) #2 {
+ %1 = tail call i8* @rsOffset([1 x i32] %a.coerce, i32 4, i32 %x, i32 %y, i32 %z) #10
+ %2 = bitcast i8* %1 to <2 x i16>*
+ store <2 x i16> %val, <2 x i16>* %2, align 4, !tbaa !34
+ ret void
+}
+
+define <2 x i16> @rsGetElementAtImpl_ushort2([1 x i32] %a.coerce, i32 %x, i32 %y, i32 %z) #3 {
+ %1 = tail call i8* @rsOffset([1 x i32] %a.coerce, i32 4, i32 %x, i32 %y, i32 %z) #10
+ %2 = bitcast i8* %1 to <2 x i16>*
+ %3 = load <2 x i16>* %2, align 4, !tbaa !34
+ ret <2 x i16> %3
+}
+
+!35 = metadata !{metadata !"ushort3", metadata !15}
+define void @rsSetElementAtImpl_ushort3([1 x i32] %a.coerce, <3 x i16> %val, i32 %x, i32 %y, i32 %z) #2 {
+ %1 = tail call i8* @rsOffset([1 x i32] %a.coerce, i32 8, i32 %x, i32 %y, i32 %z) #10
+ %2 = shufflevector <3 x i16> %val, <3 x i16> undef, <4 x i32> <i32 0, i32 1, i32 2, i32 undef>
+ %3 = bitcast i8* %1 to <4 x i16>*
+ store <4 x i16> %2, <4 x i16>* %3, align 8, !tbaa !35
+ ret void
+}
+
+define <3 x i16> @rsGetElementAtImpl_ushort3([1 x i32] %a.coerce, i32 %x, i32 %y, i32 %z) #3 {
+ %1 = tail call i8* @rsOffset([1 x i32] %a.coerce, i32 8, i32 %x, i32 %y, i32 %z) #10
+ %2 = bitcast i8* %1 to <4 x i16>*
+ %3 = load <4 x i16>* %2, align 8, !tbaa !35
+ %4 = shufflevector <4 x i16> %3, <4 x i16> undef, <3 x i32> <i32 0, i32 1, i32 2>
+ ret <3 x i16> %4
+}
+
+!36 = metadata !{metadata !"ushort4", metadata !15}
+define void @rsSetElementAtImpl_ushort4([1 x i32] %a.coerce, <4 x i16> %val, i32 %x, i32 %y, i32 %z) #2 {
+ %1 = tail call i8* @rsOffset([1 x i32] %a.coerce, i32 8, i32 %x, i32 %y, i32 %z) #10
+ %2 = bitcast i8* %1 to <4 x i16>*
+ store <4 x i16> %val, <4 x i16>* %2, align 8, !tbaa !36
+ ret void
+}
+
+define <4 x i16> @rsGetElementAtImpl_ushort4([1 x i32] %a.coerce, i32 %x, i32 %y, i32 %z) #3 {
+ %1 = tail call i8* @rsOffset([1 x i32] %a.coerce, i32 8, i32 %x, i32 %y, i32 %z) #10
+ %2 = bitcast i8* %1 to <4 x i16>*
+ %3 = load <4 x i16>* %2, align 8, !tbaa !36
+ ret <4 x i16> %3
+}
+
+!37 = metadata !{metadata !"int", metadata !15}
+define void @rsSetElementAtImpl_int([1 x i32] %a.coerce, i32 %val, i32 %x, i32 %y, i32 %z) #2 {
+ %1 = tail call i8* @rsOffset([1 x i32] %a.coerce, i32 4, i32 %x, i32 %y, i32 %z) #10
+ %2 = bitcast i8* %1 to i32*
+ store i32 %val, i32* %2, align 4, !tbaa !37
+ ret void
+}
+
+define i32 @rsGetElementAtImpl_int([1 x i32] %a.coerce, i32 %x, i32 %y, i32 %z) #3 {
+ %1 = tail call i8* @rsOffset([1 x i32] %a.coerce, i32 4, i32 %x, i32 %y, i32 %z) #10
+ %2 = bitcast i8* %1 to i32*
+ %3 = load i32* %2, align 4, !tbaa !37
+ ret i32 %3
+}
+
+!38 = metadata !{metadata !"int2", metadata !15}
+define void @rsSetElementAtImpl_int2([1 x i32] %a.coerce, <2 x i32> %val, i32 %x, i32 %y, i32 %z) #2 {
+ %1 = tail call i8* @rsOffset([1 x i32] %a.coerce, i32 8, i32 %x, i32 %y, i32 %z) #10
+ %2 = bitcast i8* %1 to <2 x i32>*
+ store <2 x i32> %val, <2 x i32>* %2, align 8, !tbaa !38
+ ret void
+}
+
+define <2 x i32> @rsGetElementAtImpl_int2([1 x i32] %a.coerce, i32 %x, i32 %y, i32 %z) #3 {
+ %1 = tail call i8* @rsOffset([1 x i32] %a.coerce, i32 8, i32 %x, i32 %y, i32 %z) #10
+ %2 = bitcast i8* %1 to <2 x i32>*
+ %3 = load <2 x i32>* %2, align 8, !tbaa !38
+ ret <2 x i32> %3
+}
+
+!39 = metadata !{metadata !"int3", metadata !15}
+define void @rsSetElementAtImpl_int3([1 x i32] %a.coerce, <3 x i32> %val, i32 %x, i32 %y, i32 %z) #2 {
+ %1 = tail call i8* @rsOffset([1 x i32] %a.coerce, i32 16, i32 %x, i32 %y, i32 %z) #10
+ %2 = shufflevector <3 x i32> %val, <3 x i32> undef, <4 x i32> <i32 0, i32 1, i32 2, i32 undef>
+ %3 = bitcast i8* %1 to <4 x i32>*
+ store <4 x i32> %2, <4 x i32>* %3, align 16, !tbaa !39
+ ret void
+}
+
+define <3 x i32> @rsGetElementAtImpl_int3([1 x i32] %a.coerce, i32 %x, i32 %y, i32 %z) #3 {
+ %1 = tail call i8* @rsOffset([1 x i32] %a.coerce, i32 16, i32 %x, i32 %y, i32 %z) #10
+ %2 = bitcast i8* %1 to <4 x i32>*
+ %3 = load <4 x i32>* %2, align 8, !tbaa !39
+ %4 = shufflevector <4 x i32> %3, <4 x i32> undef, <3 x i32> <i32 0, i32 1, i32 2>
+ ret <3 x i32> %4
+}
+
+!40 = metadata !{metadata !"int4", metadata !15}
+define void @rsSetElementAtImpl_int4([1 x i32] %a.coerce, <4 x i32> %val, i32 %x, i32 %y, i32 %z) #2 {
+ %1 = tail call i8* @rsOffset([1 x i32] %a.coerce, i32 16, i32 %x, i32 %y, i32 %z) #10
+ %2 = bitcast i8* %1 to <4 x i32>*
+ store <4 x i32> %val, <4 x i32>* %2, align 16, !tbaa !40
+ ret void
+}
+
+define <4 x i32> @rsGetElementAtImpl_int4([1 x i32] %a.coerce, i32 %x, i32 %y, i32 %z) #3 {
+ %1 = tail call i8* @rsOffset([1 x i32] %a.coerce, i32 16, i32 %x, i32 %y, i32 %z) #10
+ %2 = bitcast i8* %1 to <4 x i32>*
+ %3 = load <4 x i32>* %2, align 16, !tbaa !40
+ ret <4 x i32> %3
+}
+
+!41 = metadata !{metadata !"uint", metadata !15}
+define void @rsSetElementAtImpl_uint([1 x i32] %a.coerce, i32 %val, i32 %x, i32 %y, i32 %z) #2 {
+ %1 = tail call i8* @rsOffset([1 x i32] %a.coerce, i32 4, i32 %x, i32 %y, i32 %z) #10
+ %2 = bitcast i8* %1 to i32*
+ store i32 %val, i32* %2, align 4, !tbaa !41
+ ret void
+}
+
+define i32 @rsGetElementAtImpl_uint([1 x i32] %a.coerce, i32 %x, i32 %y, i32 %z) #3 {
+ %1 = tail call i8* @rsOffset([1 x i32] %a.coerce, i32 4, i32 %x, i32 %y, i32 %z) #10
+ %2 = bitcast i8* %1 to i32*
+ %3 = load i32* %2, align 4, !tbaa !41
+ ret i32 %3
+}
+
+!42 = metadata !{metadata !"uint2", metadata !15}
+define void @rsSetElementAtImpl_uint2([1 x i32] %a.coerce, <2 x i32> %val, i32 %x, i32 %y, i32 %z) #2 {
+ %1 = tail call i8* @rsOffset([1 x i32] %a.coerce, i32 8, i32 %x, i32 %y, i32 %z) #10
+ %2 = bitcast i8* %1 to <2 x i32>*
+ store <2 x i32> %val, <2 x i32>* %2, align 8, !tbaa !42
+ ret void
+}
+
+define <2 x i32> @rsGetElementAtImpl_uint2([1 x i32] %a.coerce, i32 %x, i32 %y, i32 %z) #3 {
+ %1 = tail call i8* @rsOffset([1 x i32] %a.coerce, i32 8, i32 %x, i32 %y, i32 %z) #10
+ %2 = bitcast i8* %1 to <2 x i32>*
+ %3 = load <2 x i32>* %2, align 8, !tbaa !42
+ ret <2 x i32> %3
+}
+
+!43 = metadata !{metadata !"uint3", metadata !15}
+define void @rsSetElementAtImpl_uint3([1 x i32] %a.coerce, <3 x i32> %val, i32 %x, i32 %y, i32 %z) #2 {
+ %1 = tail call i8* @rsOffset([1 x i32] %a.coerce, i32 16, i32 %x, i32 %y, i32 %z) #10
+ %2 = shufflevector <3 x i32> %val, <3 x i32> undef, <4 x i32> <i32 0, i32 1, i32 2, i32 undef>
+ %3 = bitcast i8* %1 to <4 x i32>*
+ store <4 x i32> %2, <4 x i32>* %3, align 16, !tbaa !43
+ ret void
+}
+
+define <3 x i32> @rsGetElementAtImpl_uint3([1 x i32] %a.coerce, i32 %x, i32 %y, i32 %z) #3 {
+ %1 = tail call i8* @rsOffset([1 x i32] %a.coerce, i32 16, i32 %x, i32 %y, i32 %z) #10
+ %2 = bitcast i8* %1 to <4 x i32>*
+ %3 = load <4 x i32>* %2, align 8, !tbaa !43
+ %4 = shufflevector <4 x i32> %3, <4 x i32> undef, <3 x i32> <i32 0, i32 1, i32 2>
+ ret <3 x i32> %4
+}
+
+!44 = metadata !{metadata !"uint4", metadata !15}
+define void @rsSetElementAtImpl_uint4([1 x i32] %a.coerce, <4 x i32> %val, i32 %x, i32 %y, i32 %z) #2 {
+ %1 = tail call i8* @rsOffset([1 x i32] %a.coerce, i32 16, i32 %x, i32 %y, i32 %z) #10
+ %2 = bitcast i8* %1 to <4 x i32>*
+ store <4 x i32> %val, <4 x i32>* %2, align 16, !tbaa !44
+ ret void
+}
+
+define <4 x i32> @rsGetElementAtImpl_uint4([1 x i32] %a.coerce, i32 %x, i32 %y, i32 %z) #3 {
+ %1 = tail call i8* @rsOffset([1 x i32] %a.coerce, i32 16, i32 %x, i32 %y, i32 %z) #10
+ %2 = bitcast i8* %1 to <4 x i32>*
+ %3 = load <4 x i32>* %2, align 16, !tbaa !44
+ ret <4 x i32> %3
+}
+
+!45 = metadata !{metadata !"long", metadata !15}
+define void @rsSetElementAtImpl_long([1 x i32] %a.coerce, i64 %val, i32 %x, i32 %y, i32 %z) #2 {
+ %1 = tail call i8* @rsOffset([1 x i32] %a.coerce, i32 8, i32 %x, i32 %y, i32 %z) #10
+ %2 = bitcast i8* %1 to i64*
+ store i64 %val, i64* %2, align 8, !tbaa !45
+ ret void
+}
+
+define i64 @rsGetElementAtImpl_long([1 x i32] %a.coerce, i32 %x, i32 %y, i32 %z) #3 {
+ %1 = tail call i8* @rsOffset([1 x i32] %a.coerce, i32 8, i32 %x, i32 %y, i32 %z) #10
+ %2 = bitcast i8* %1 to i64*
+ %3 = load i64* %2, align 8, !tbaa !45
+ ret i64 %3
+}
+
+!46 = metadata !{metadata !"long2", metadata !15}
+define void @rsSetElementAtImpl_long2([1 x i32] %a.coerce, <2 x i64> %val, i32 %x, i32 %y, i32 %z) #2 {
+ %1 = tail call i8* @rsOffset([1 x i32] %a.coerce, i32 16, i32 %x, i32 %y, i32 %z) #10
+ %2 = bitcast i8* %1 to <2 x i64>*
+ store <2 x i64> %val, <2 x i64>* %2, align 16, !tbaa !46
+ ret void
+}
+
+define <2 x i64> @rsGetElementAtImpl_long2([1 x i32] %a.coerce, i32 %x, i32 %y, i32 %z) #3 {
+ %1 = tail call i8* @rsOffset([1 x i32] %a.coerce, i32 16, i32 %x, i32 %y, i32 %z) #10
+ %2 = bitcast i8* %1 to <2 x i64>*
+ %3 = load <2 x i64>* %2, align 16, !tbaa !46
+ ret <2 x i64> %3
+}
+
+!47 = metadata !{metadata !"long3", metadata !15}
+define void @rsSetElementAtImpl_long3([1 x i32] %a.coerce, <3 x i64> %val, i32 %x, i32 %y, i32 %z) #2 {
+ %1 = tail call i8* @rsOffset([1 x i32] %a.coerce, i32 32, i32 %x, i32 %y, i32 %z) #10
+ %2 = shufflevector <3 x i64> %val, <3 x i64> undef, <4 x i32> <i32 0, i32 1, i32 2, i32 undef>
+ %3 = bitcast i8* %1 to <4 x i64>*
+ store <4 x i64> %2, <4 x i64>* %3, align 32, !tbaa !47
+ ret void
+}
+
+define void @rsGetElementAtImpl_long3(<3 x i64>* noalias nocapture sret %agg.result, [1 x i32] %a.coerce, i32 %x, i32 %y, i32 %z) #2 {
+ %1 = tail call i8* @rsOffset([1 x i32] %a.coerce, i32 32, i32 %x, i32 %y, i32 %z) #10
+ %2 = bitcast i8* %1 to <4 x i64>*
+ %3 = load <4 x i64>* %2, align 32
+ %4 = bitcast <3 x i64>* %agg.result to <4 x i64>*
+ store <4 x i64> %3, <4 x i64>* %4, align 32, !tbaa !47
+ ret void
+}
+
+!48 = metadata !{metadata !"long4", metadata !15}
+define void @rsSetElementAtImpl_long4([1 x i32] %a.coerce, <4 x i64> %val, i32 %x, i32 %y, i32 %z) #2 {
+ %1 = tail call i8* @rsOffset([1 x i32] %a.coerce, i32 32, i32 %x, i32 %y, i32 %z) #10
+ %2 = bitcast i8* %1 to <4 x i64>*
+ store <4 x i64> %val, <4 x i64>* %2, align 32, !tbaa !48
+ ret void
+}
+
+define void @rsGetElementAtImpl_long4(<4 x i64>* noalias nocapture sret %agg.result, [1 x i32] %a.coerce, i32 %x, i32 %y, i32 %z) #2 {
+ %1 = tail call i8* @rsOffset([1 x i32] %a.coerce, i32 32, i32 %x, i32 %y, i32 %z) #10
+ %2 = bitcast i8* %1 to <4 x i64>*
+ %3 = load <4 x i64>* %2, align 32, !tbaa !15
+ store <4 x i64> %3, <4 x i64>* %agg.result, align 32, !tbaa !48
+ ret void
+}
+
+!49 = metadata !{metadata !"ulong", metadata !15}
+define void @rsSetElementAtImpl_ulong([1 x i32] %a.coerce, i64 %val, i32 %x, i32 %y, i32 %z) #2 {
+ %1 = tail call i8* @rsOffset([1 x i32] %a.coerce, i32 8, i32 %x, i32 %y, i32 %z) #10
+ %2 = bitcast i8* %1 to i64*
+ store i64 %val, i64* %2, align 8, !tbaa !49
+ ret void
+}
+
+define i64 @rsGetElementAtImpl_ulong([1 x i32] %a.coerce, i32 %x, i32 %y, i32 %z) #3 {
+ %1 = tail call i8* @rsOffset([1 x i32] %a.coerce, i32 8, i32 %x, i32 %y, i32 %z) #10
+ %2 = bitcast i8* %1 to i64*
+ %3 = load i64* %2, align 8, !tbaa !49
+ ret i64 %3
+}
+
+!50 = metadata !{metadata !"ulong2", metadata !15}
+define void @rsSetElementAtImpl_ulong2([1 x i32] %a.coerce, <2 x i64> %val, i32 %x, i32 %y, i32 %z) #2 {
+ %1 = tail call i8* @rsOffset([1 x i32] %a.coerce, i32 16, i32 %x, i32 %y, i32 %z) #10
+ %2 = bitcast i8* %1 to <2 x i64>*
+ store <2 x i64> %val, <2 x i64>* %2, align 16, !tbaa !50
+ ret void
+}
+
+define <2 x i64> @rsGetElementAtImpl_ulong2([1 x i32] %a.coerce, i32 %x, i32 %y, i32 %z) #3 {
+ %1 = tail call i8* @rsOffset([1 x i32] %a.coerce, i32 16, i32 %x, i32 %y, i32 %z) #10
+ %2 = bitcast i8* %1 to <2 x i64>*
+ %3 = load <2 x i64>* %2, align 16, !tbaa !50
+ ret <2 x i64> %3
+}
+
+!51 = metadata !{metadata !"ulong3", metadata !15}
+define void @rsSetElementAtImpl_ulong3([1 x i32] %a.coerce, <3 x i64> %val, i32 %x, i32 %y, i32 %z) #2 {
+ %1 = tail call i8* @rsOffset([1 x i32] %a.coerce, i32 32, i32 %x, i32 %y, i32 %z) #10
+ %2 = shufflevector <3 x i64> %val, <3 x i64> undef, <4 x i32> <i32 0, i32 1, i32 2, i32 undef>
+ %3 = bitcast i8* %1 to <4 x i64>*
+ store <4 x i64> %2, <4 x i64>* %3, align 32, !tbaa !51
+ ret void
+}
+
+define void @rsGetElementAtImpl_ulong3(<3 x i64>* noalias nocapture sret %agg.result, [1 x i32] %a.coerce, i32 %x, i32 %y, i32 %z) #2 {
+ %1 = tail call i8* @rsOffset([1 x i32] %a.coerce, i32 32, i32 %x, i32 %y, i32 %z) #10
+ %2 = bitcast i8* %1 to <4 x i64>*
+ %3 = load <4 x i64>* %2, align 32
+ %4 = bitcast <3 x i64>* %agg.result to <4 x i64>*
+ store <4 x i64> %3, <4 x i64>* %4, align 32, !tbaa !51
+ ret void
+}
+
+!52 = metadata !{metadata !"ulong4", metadata !15}
+define void @rsSetElementAtImpl_ulong4([1 x i32] %a.coerce, <4 x i64> %val, i32 %x, i32 %y, i32 %z) #2 {
+ %1 = tail call i8* @rsOffset([1 x i32] %a.coerce, i32 32, i32 %x, i32 %y, i32 %z) #10
+ %2 = bitcast i8* %1 to <4 x i64>*
+ store <4 x i64> %val, <4 x i64>* %2, align 32, !tbaa !52
+ ret void
+}
+
+define void @rsGetElementAtImpl_ulong4(<4 x i64>* noalias nocapture sret %agg.result, [1 x i32] %a.coerce, i32 %x, i32 %y, i32 %z) #2 {
+ %1 = tail call i8* @rsOffset([1 x i32] %a.coerce, i32 32, i32 %x, i32 %y, i32 %z) #10
+ %2 = bitcast i8* %1 to <4 x i64>*
+ %3 = load <4 x i64>* %2, align 32, !tbaa !15
+ store <4 x i64> %3, <4 x i64>* %agg.result, align 32, !tbaa !52
+ ret void
+}
+
+!53 = metadata !{metadata !"float", metadata !15}
+define void @rsSetElementAtImpl_float([1 x i32] %a.coerce, float %val, i32 %x, i32 %y, i32 %z) #2 {
+ %1 = tail call i8* @rsOffset([1 x i32] %a.coerce, i32 4, i32 %x, i32 %y, i32 %z) #10
+ %2 = bitcast i8* %1 to float*
+ store float %val, float* %2, align 4, !tbaa !53
+ ret void
+}
+
+define float @rsGetElementAtImpl_float([1 x i32] %a.coerce, i32 %x, i32 %y, i32 %z) #3 {
+ %1 = tail call i8* @rsOffset([1 x i32] %a.coerce, i32 4, i32 %x, i32 %y, i32 %z) #10
+ %2 = bitcast i8* %1 to float*
+ %3 = load float* %2, align 4, !tbaa !53
+ ret float %3
+}
+
+!54 = metadata !{metadata !"float2", metadata !15}
+define void @rsSetElementAtImpl_float2([1 x i32] %a.coerce, <2 x float> %val, i32 %x, i32 %y, i32 %z) #2 {
+ %1 = tail call i8* @rsOffset([1 x i32] %a.coerce, i32 8, i32 %x, i32 %y, i32 %z) #10
+ %2 = bitcast i8* %1 to <2 x float>*
+ store <2 x float> %val, <2 x float>* %2, align 8, !tbaa !54
+ ret void
+}
+
+define <2 x float> @rsGetElementAtImpl_float2([1 x i32] %a.coerce, i32 %x, i32 %y, i32 %z) #3 {
+ %1 = tail call i8* @rsOffset([1 x i32] %a.coerce, i32 8, i32 %x, i32 %y, i32 %z) #10
+ %2 = bitcast i8* %1 to <2 x float>*
+ %3 = load <2 x float>* %2, align 8, !tbaa !54
+ ret <2 x float> %3
+}
+
+!55 = metadata !{metadata !"float3", metadata !15}
+define void @rsSetElementAtImpl_float3([1 x i32] %a.coerce, <3 x float> %val, i32 %x, i32 %y, i32 %z) #2 {
+ %1 = tail call i8* @rsOffset([1 x i32] %a.coerce, i32 16, i32 %x, i32 %y, i32 %z) #10
+ %2 = shufflevector <3 x float> %val, <3 x float> undef, <4 x i32> <i32 0, i32 1, i32 2, i32 undef>
+ %3 = bitcast i8* %1 to <4 x float>*
+ store <4 x float> %2, <4 x float>* %3, align 16, !tbaa !55
+ ret void
+}
+
+define <3 x float> @rsGetElementAtImpl_float3([1 x i32] %a.coerce, i32 %x, i32 %y, i32 %z) #3 {
+ %1 = tail call i8* @rsOffset([1 x i32] %a.coerce, i32 16, i32 %x, i32 %y, i32 %z) #10
+ %2 = bitcast i8* %1 to <4 x float>*
+ %3 = load <4 x float>* %2, align 8, !tbaa !55
+ %4 = shufflevector <4 x float> %3, <4 x float> undef, <3 x i32> <i32 0, i32 1, i32 2>
+ ret <3 x float> %4
+}
+
+!56 = metadata !{metadata !"float4", metadata !15}
+define void @rsSetElementAtImpl_float4([1 x i32] %a.coerce, <4 x float> %val, i32 %x, i32 %y, i32 %z) #2 {
+ %1 = tail call i8* @rsOffset([1 x i32] %a.coerce, i32 16, i32 %x, i32 %y, i32 %z) #10
+ %2 = bitcast i8* %1 to <4 x float>*
+ store <4 x float> %val, <4 x float>* %2, align 16, !tbaa !56
+ ret void
+}
+
+define <4 x float> @rsGetElementAtImpl_float4([1 x i32] %a.coerce, i32 %x, i32 %y, i32 %z) #3 {
+ %1 = tail call i8* @rsOffset([1 x i32] %a.coerce, i32 16, i32 %x, i32 %y, i32 %z) #10
+ %2 = bitcast i8* %1 to <4 x float>*
+ %3 = load <4 x float>* %2, align 16, !tbaa !56
+ ret <4 x float> %3
+}
+
+!57 = metadata !{metadata !"double", metadata !15}
+define void @rsSetElementAtImpl_double([1 x i32] %a.coerce, double %val, i32 %x, i32 %y, i32 %z) #2 {
+ %1 = tail call i8* @rsOffset([1 x i32] %a.coerce, i32 8, i32 %x, i32 %y, i32 %z) #10
+ %2 = bitcast i8* %1 to double*
+ store double %val, double* %2, align 8, !tbaa !57
+ ret void
+}
+
+define double @rsGetElementAtImpl_double([1 x i32] %a.coerce, i32 %x, i32 %y, i32 %z) #3 {
+ %1 = tail call i8* @rsOffset([1 x i32] %a.coerce, i32 8, i32 %x, i32 %y, i32 %z) #10
+ %2 = bitcast i8* %1 to double*
+ %3 = load double* %2, align 8, !tbaa !57
+ ret double %3
+}
+
+!58 = metadata !{metadata !"double2", metadata !15}
+define void @rsSetElementAtImpl_double2([1 x i32] %a.coerce, <2 x double> %val, i32 %x, i32 %y, i32 %z) #2 {
+ %1 = tail call i8* @rsOffset([1 x i32] %a.coerce, i32 16, i32 %x, i32 %y, i32 %z) #10
+ %2 = bitcast i8* %1 to <2 x double>*
+ store <2 x double> %val, <2 x double>* %2, align 16, !tbaa !58
+ ret void
+}
+
+define <2 x double> @rsGetElementAtImpl_double2([1 x i32] %a.coerce, i32 %x, i32 %y, i32 %z) #3 {
+ %1 = tail call i8* @rsOffset([1 x i32] %a.coerce, i32 16, i32 %x, i32 %y, i32 %z) #10
+ %2 = bitcast i8* %1 to <2 x double>*
+ %3 = load <2 x double>* %2, align 16, !tbaa !58
+ ret <2 x double> %3
+}
+
+!59 = metadata !{metadata !"double3", metadata !15}
+define void @rsSetElementAtImpl_double3([1 x i32] %a.coerce, <3 x double> %val, i32 %x, i32 %y, i32 %z) #2 {
+ %1 = tail call i8* @rsOffset([1 x i32] %a.coerce, i32 32, i32 %x, i32 %y, i32 %z) #10
+ %2 = shufflevector <3 x double> %val, <3 x double> undef, <4 x i32> <i32 0, i32 1, i32 2, i32 undef>
+ %3 = bitcast i8* %1 to <4 x double>*
+ store <4 x double> %2, <4 x double>* %3, align 32, !tbaa !59
+ ret void
+}
+
+
+define void @rsGetElementAtImpl_double3(<3 x double>* noalias nocapture sret %agg.result, [1 x i32] %a.coerce, i32 %x, i32 %y, i32 %z) #2 {
+ %1 = tail call i8* @rsOffset([1 x i32] %a.coerce, i32 32, i32 %x, i32 %y, i32 %z) #10
+ %2 = bitcast i8* %1 to <4 x double>*
+ %3 = load <4 x double>* %2, align 32
+ %4 = bitcast <3 x double>* %agg.result to <4 x double>*
+ store <4 x double> %3, <4 x double>* %4, align 32, !tbaa !59
+ ret void
+}
+
+!60 = metadata !{metadata !"double4", metadata !15}
+define void @rsSetElementAtImpl_double4([1 x i32] %a.coerce, <4 x double> %val, i32 %x, i32 %y, i32 %z) #2 {
+ %1 = tail call i8* @rsOffset([1 x i32] %a.coerce, i32 32, i32 %x, i32 %y, i32 %z) #10
+ %2 = bitcast i8* %1 to <4 x double>*
+ store <4 x double> %val, <4 x double>* %2, align 32, !tbaa !60
+ ret void
+}
+define void @rsGetElementAtImpl_double4(<4 x double>* noalias nocapture sret %agg.result, [1 x i32] %a.coerce, i32 %x, i32 %y, i32 %z) #2 {
+ %1 = tail call i8* @rsOffset([1 x i32] %a.coerce, i32 32, i32 %x, i32 %y, i32 %z) #10
+ %2 = bitcast i8* %1 to <4 x double>*
+ %3 = load <4 x double>* %2, align 32, !tbaa !15
+ store <4 x double> %3, <4 x double>* %agg.result, align 32, !tbaa !60
+ ret void
+}
+
+attributes #0 = { nounwind readonly "less-precise-fpmad"="false" "no-frame-pointer-elim"="true" "no-frame-pointer-elim-non-leaf"="true" "no-infs-fp-math"="false" "no-nans-fp-math"="false" "unsafe-fp-math"="false" "use-soft-float"="false" }
+attributes #1 = { nounwind "less-precise-fpmad"="false" "no-frame-pointer-elim"="true" "no-frame-pointer-elim-non-leaf"="true" "no-infs-fp-math"="false" "no-nans-fp-math"="false" "unsafe-fp-math"="false" "use-soft-float"="false" }
+attributes #2 = { nounwind "less-precise-fpmad"="false" "no-frame-pointer-elim"="true" "no-frame-pointer-elim-non-leaf"="true" "no-infs-fp-math"="false" "no-nans-fp-math"="false" "unsafe-fp-math"="false" "use-soft-float"="false" }
+attributes #3 = { nounwind readonly "less-precise-fpmad"="false" "no-frame-pointer-elim"="true" "no-frame-pointer-elim-non-leaf"="true" "no-infs-fp-math"="false" "no-nans-fp-math"="false" "unsafe-fp-math"="false" "use-soft-float"="false" }
+attributes #4 = { "less-precise-fpmad"="false" "no-frame-pointer-elim"="true" "no-frame-pointer-elim-non-leaf"="true" "no-infs-fp-math"="false" "no-nans-fp-math"="false" "unsafe-fp-math"="false" "use-soft-float"="false" }
+attributes #5 = { nounwind readnone "less-precise-fpmad"="false" "no-frame-pointer-elim"="true" "no-frame-pointer-elim-non-leaf"="true" "no-infs-fp-math"="false" "no-nans-fp-math"="false" "unsafe-fp-math"="false" "use-soft-float"="false" }
+attributes #6 = { nounwind readnone }
+attributes #7 = { nounwind }
+attributes #8 = { alwaysinline nounwind readnone }
+attributes #9 = { nounwind readonly }
+attributes #10 = { nobuiltin }
+attributes #11 = { nobuiltin nounwind }
+attributes #12 = { nobuiltin nounwind readnone }
+
diff --git a/driver/runtime/arch/clamp.c b/driver/runtime/arch/clamp.c
new file mode 100644
index 0000000..c2c2226
--- /dev/null
+++ b/driver/runtime/arch/clamp.c
@@ -0,0 +1,102 @@
+/*
+ * Copyright (C) 2013 The Android Open Source Project
+ *
+ * Licensed under the Apache License, Version 2.0 (the "License");
+ * you may not use this file except in compliance with the License.
+ * You may obtain a copy of the License at
+ *
+ * http://www.apache.org/licenses/LICENSE-2.0
+ *
+ * Unless required by applicable law or agreed to in writing, software
+ * distributed under the License is distributed on an "AS IS" BASIS,
+ * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ * See the License for the specific language governing permissions and
+ * limitations under the License.
+ */
+
+#include "rs_types.rsh"
+
+#define S_CLAMP(T) \
+extern T __attribute__((overloadable)) clamp(T amount, T low, T high) { \
+ return amount < low ? low : (amount > high ? high : amount); \
+}
+
+//_CLAMP(float); implemented in .ll
+S_CLAMP(double);
+S_CLAMP(char);
+S_CLAMP(uchar);
+S_CLAMP(short);
+S_CLAMP(ushort);
+S_CLAMP(int);
+S_CLAMP(uint);
+S_CLAMP(long);
+S_CLAMP(ulong);
+
+#undef S_CLAMP
+
+
+ \
+#define V_CLAMP(T) \
+extern T##2 __attribute__((overloadable)) clamp(T##2 amount, T##2 low, T##2 high) { \
+ T##2 r; \
+ r.x = amount.x < low.x ? low.x : (amount.x > high.x ? high.x : amount.x); \
+ r.y = amount.y < low.y ? low.y : (amount.y > high.y ? high.y : amount.y); \
+ return r; \
+} \
+ \
+extern T##3 __attribute__((overloadable)) clamp(T##3 amount, T##3 low, T##3 high) { \
+ T##3 r; \
+ r.x = amount.x < low.x ? low.x : (amount.x > high.x ? high.x : amount.x); \
+ r.y = amount.y < low.y ? low.y : (amount.y > high.y ? high.y : amount.y); \
+ r.z = amount.z < low.z ? low.z : (amount.z > high.z ? high.z : amount.z); \
+ return r; \
+} \
+ \
+extern T##4 __attribute__((overloadable)) clamp(T##4 amount, T##4 low, T##4 high) { \
+ T##4 r; \
+ r.x = amount.x < low.x ? low.x : (amount.x > high.x ? high.x : amount.x); \
+ r.y = amount.y < low.y ? low.y : (amount.y > high.y ? high.y : amount.y); \
+ r.z = amount.z < low.z ? low.z : (amount.z > high.z ? high.z : amount.z); \
+ r.w = amount.w < low.w ? low.w : (amount.w > high.w ? high.w : amount.w); \
+ return r; \
+} \
+ \
+extern T##2 __attribute__((overloadable)) clamp(T##2 amount, T low, T high) { \
+ T##2 r; \
+ r.x = amount.x < low ? low : (amount.x > high ? high : amount.x); \
+ r.y = amount.y < low ? low : (amount.y > high ? high : amount.y); \
+ return r; \
+} \
+ \
+extern T##3 __attribute__((overloadable)) clamp(T##3 amount, T low, T high) { \
+ T##3 r; \
+ r.x = amount.x < low ? low : (amount.x > high ? high : amount.x); \
+ r.y = amount.y < low ? low : (amount.y > high ? high : amount.y); \
+ r.z = amount.z < low ? low : (amount.z > high ? high : amount.z); \
+ return r; \
+} \
+ \
+extern T##4 __attribute__((overloadable)) clamp(T##4 amount, T low, T high) { \
+ T##4 r; \
+ r.x = amount.x < low ? low : (amount.x > high ? high : amount.x); \
+ r.y = amount.y < low ? low : (amount.y > high ? high : amount.y); \
+ r.z = amount.z < low ? low : (amount.z > high ? high : amount.z); \
+ r.w = amount.w < low ? low : (amount.w > high ? high : amount.w); \
+ return r; \
+}
+
+//V_CLAMP(float); implemented in .ll
+V_CLAMP(double);
+V_CLAMP(char);
+V_CLAMP(uchar);
+V_CLAMP(short);
+V_CLAMP(ushort);
+#ifndef ARCH_ARM_HAVE_NEON
+ V_CLAMP(int); //implemented in .ll
+ V_CLAMP(uint); //implemented in .ll
+#endif
+V_CLAMP(long);
+V_CLAMP(ulong);
+
+#undef _CLAMP
+
diff --git a/driver/runtime/arch/dot_length.c b/driver/runtime/arch/dot_length.c
new file mode 100644
index 0000000..94c99b6
--- /dev/null
+++ b/driver/runtime/arch/dot_length.c
@@ -0,0 +1,47 @@
+/*
+ * Copyright (C) 2012 The Android Open Source Project
+ *
+ * Licensed under the Apache License, Version 2.0 (the "License");
+ * you may not use this file except in compliance with the License.
+ * You may obtain a copy of the License at
+ *
+ * http://www.apache.org/licenses/LICENSE-2.0
+ *
+ * Unless required by applicable law or agreed to in writing, software
+ * distributed under the License is distributed on an "AS IS" BASIS,
+ * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ * See the License for the specific language governing permissions and
+ * limitations under the License.
+ */
+
+#include "rs_types.rsh"
+
+extern float __attribute__((overloadable)) dot(float lhs, float rhs) {
+ return lhs * rhs;
+}
+extern float __attribute__((overloadable)) dot(float2 lhs, float2 rhs) {
+ return lhs.x*rhs.x + lhs.y*rhs.y;
+}
+extern float __attribute__((overloadable)) dot(float3 lhs, float3 rhs) {
+ return lhs.x*rhs.x + lhs.y*rhs.y + lhs.z*rhs.z;
+}
+extern float __attribute__((overloadable)) dot(float4 lhs, float4 rhs) {
+ return lhs.x*rhs.x + lhs.y*rhs.y + lhs.z*rhs.z + lhs.w*rhs.w;
+}
+
+extern float __attribute__((overloadable)) fabs(float);
+extern float __attribute__((overloadable)) sqrt(float);
+
+extern float __attribute__((overloadable)) length(float v) {
+ return fabs(v);
+}
+extern float __attribute__((overloadable)) length(float2 v) {
+ return sqrt(v.x*v.x + v.y*v.y);
+}
+extern float __attribute__((overloadable)) length(float3 v) {
+ return sqrt(v.x*v.x + v.y*v.y + v.z*v.z);
+}
+extern float __attribute__((overloadable)) length(float4 v) {
+ return sqrt(v.x*v.x + v.y*v.y + v.z*v.z + v.w*v.w);
+}
+
diff --git a/driver/runtime/arch/generic.c b/driver/runtime/arch/generic.c
new file mode 100644
index 0000000..da83c2a
--- /dev/null
+++ b/driver/runtime/arch/generic.c
@@ -0,0 +1,948 @@
+/*
+ * Copyright (C) 2012 The Android Open Source Project
+ *
+ * Licensed under the Apache License, Version 2.0 (the "License");
+ * you may not use this file except in compliance with the License.
+ * You may obtain a copy of the License at
+ *
+ * http://www.apache.org/licenses/LICENSE-2.0
+ *
+ * Unless required by applicable law or agreed to in writing, software
+ * distributed under the License is distributed on an "AS IS" BASIS,
+ * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ * See the License for the specific language governing permissions and
+ * limitations under the License.
+ */
+
+
+#include "rs_types.rsh"
+
+extern short __attribute__((overloadable, always_inline)) rsClamp(short amount, short low, short high);
+extern uchar4 __attribute__((overloadable)) convert_uchar4(short4);
+extern uchar4 __attribute__((overloadable)) convert_uchar4(float4);
+extern float4 __attribute__((overloadable)) convert_float4(uchar4);
+extern float __attribute__((overloadable)) sqrt(float);
+
+/*
+ * CLAMP
+ */
+#define _CLAMP(T) \
+extern T __attribute__((overloadable)) clamp(T amount, T low, T high) { \
+ return amount < low ? low : (amount > high ? high : amount); \
+} \
+ \
+extern T##2 __attribute__((overloadable)) clamp(T##2 amount, T##2 low, T##2 high) { \
+ T##2 r; \
+ r.x = amount.x < low.x ? low.x : (amount.x > high.x ? high.x : amount.x); \
+ r.y = amount.y < low.y ? low.y : (amount.y > high.y ? high.y : amount.y); \
+ return r; \
+} \
+ \
+extern T##3 __attribute__((overloadable)) clamp(T##3 amount, T##3 low, T##3 high) { \
+ T##3 r; \
+ r.x = amount.x < low.x ? low.x : (amount.x > high.x ? high.x : amount.x); \
+ r.y = amount.y < low.y ? low.y : (amount.y > high.y ? high.y : amount.y); \
+ r.z = amount.z < low.z ? low.z : (amount.z > high.z ? high.z : amount.z); \
+ return r; \
+} \
+ \
+extern T##4 __attribute__((overloadable)) clamp(T##4 amount, T##4 low, T##4 high) { \
+ T##4 r; \
+ r.x = amount.x < low.x ? low.x : (amount.x > high.x ? high.x : amount.x); \
+ r.y = amount.y < low.y ? low.y : (amount.y > high.y ? high.y : amount.y); \
+ r.z = amount.z < low.z ? low.z : (amount.z > high.z ? high.z : amount.z); \
+ r.w = amount.w < low.w ? low.w : (amount.w > high.w ? high.w : amount.w); \
+ return r; \
+} \
+ \
+extern T##2 __attribute__((overloadable)) clamp(T##2 amount, T low, T high) { \
+ T##2 r; \
+ r.x = amount.x < low ? low : (amount.x > high ? high : amount.x); \
+ r.y = amount.y < low ? low : (amount.y > high ? high : amount.y); \
+ return r; \
+} \
+ \
+extern T##3 __attribute__((overloadable)) clamp(T##3 amount, T low, T high) { \
+ T##3 r; \
+ r.x = amount.x < low ? low : (amount.x > high ? high : amount.x); \
+ r.y = amount.y < low ? low : (amount.y > high ? high : amount.y); \
+ r.z = amount.z < low ? low : (amount.z > high ? high : amount.z); \
+ return r; \
+} \
+ \
+extern T##4 __attribute__((overloadable)) clamp(T##4 amount, T low, T high) { \
+ T##4 r; \
+ r.x = amount.x < low ? low : (amount.x > high ? high : amount.x); \
+ r.y = amount.y < low ? low : (amount.y > high ? high : amount.y); \
+ r.z = amount.z < low ? low : (amount.z > high ? high : amount.z); \
+ r.w = amount.w < low ? low : (amount.w > high ? high : amount.w); \
+ return r; \
+}
+
+_CLAMP(float);
+_CLAMP(double);
+_CLAMP(char);
+_CLAMP(uchar);
+_CLAMP(short);
+_CLAMP(ushort);
+_CLAMP(int);
+_CLAMP(uint);
+_CLAMP(long);
+_CLAMP(ulong);
+
+#undef _CLAMP
+
+/*
+ * FMAX
+ */
+
+extern float __attribute__((overloadable)) fmax(float v1, float v2) {
+ return v1 > v2 ? v1 : v2;
+}
+
+extern float2 __attribute__((overloadable)) fmax(float2 v1, float2 v2) {
+ float2 r;
+ r.x = v1.x > v2.x ? v1.x : v2.x;
+ r.y = v1.y > v2.y ? v1.y : v2.y;
+ return r;
+}
+
+extern float3 __attribute__((overloadable)) fmax(float3 v1, float3 v2) {
+ float3 r;
+ r.x = v1.x > v2.x ? v1.x : v2.x;
+ r.y = v1.y > v2.y ? v1.y : v2.y;
+ r.z = v1.z > v2.z ? v1.z : v2.z;
+ return r;
+}
+
+extern float4 __attribute__((overloadable)) fmax(float4 v1, float4 v2) {
+ float4 r;
+ r.x = v1.x > v2.x ? v1.x : v2.x;
+ r.y = v1.y > v2.y ? v1.y : v2.y;
+ r.z = v1.z > v2.z ? v1.z : v2.z;
+ r.w = v1.w > v2.w ? v1.w : v2.w;
+ return r;
+}
+
+extern float2 __attribute__((overloadable)) fmax(float2 v1, float v2) {
+ float2 r;
+ r.x = v1.x > v2 ? v1.x : v2;
+ r.y = v1.y > v2 ? v1.y : v2;
+ return r;
+}
+
+extern float3 __attribute__((overloadable)) fmax(float3 v1, float v2) {
+ float3 r;
+ r.x = v1.x > v2 ? v1.x : v2;
+ r.y = v1.y > v2 ? v1.y : v2;
+ r.z = v1.z > v2 ? v1.z : v2;
+ return r;
+}
+
+extern float4 __attribute__((overloadable)) fmax(float4 v1, float v2) {
+ float4 r;
+ r.x = v1.x > v2 ? v1.x : v2;
+ r.y = v1.y > v2 ? v1.y : v2;
+ r.z = v1.z > v2 ? v1.z : v2;
+ r.w = v1.w > v2 ? v1.w : v2;
+ return r;
+}
+
+extern float __attribute__((overloadable)) fmin(float v1, float v2) {
+ return v1 < v2 ? v1 : v2;
+}
+
+
+/*
+ * FMIN
+ */
+extern float2 __attribute__((overloadable)) fmin(float2 v1, float2 v2) {
+ float2 r;
+ r.x = v1.x < v2.x ? v1.x : v2.x;
+ r.y = v1.y < v2.y ? v1.y : v2.y;
+ return r;
+}
+
+extern float3 __attribute__((overloadable)) fmin(float3 v1, float3 v2) {
+ float3 r;
+ r.x = v1.x < v2.x ? v1.x : v2.x;
+ r.y = v1.y < v2.y ? v1.y : v2.y;
+ r.z = v1.z < v2.z ? v1.z : v2.z;
+ return r;
+}
+
+extern float4 __attribute__((overloadable)) fmin(float4 v1, float4 v2) {
+ float4 r;
+ r.x = v1.x < v2.x ? v1.x : v2.x;
+ r.y = v1.y < v2.y ? v1.y : v2.y;
+ r.z = v1.z < v2.z ? v1.z : v2.z;
+ r.w = v1.w < v2.w ? v1.w : v2.w;
+ return r;
+}
+
+extern float2 __attribute__((overloadable)) fmin(float2 v1, float v2) {
+ float2 r;
+ r.x = v1.x < v2 ? v1.x : v2;
+ r.y = v1.y < v2 ? v1.y : v2;
+ return r;
+}
+
+extern float3 __attribute__((overloadable)) fmin(float3 v1, float v2) {
+ float3 r;
+ r.x = v1.x < v2 ? v1.x : v2;
+ r.y = v1.y < v2 ? v1.y : v2;
+ r.z = v1.z < v2 ? v1.z : v2;
+ return r;
+}
+
+extern float4 __attribute__((overloadable)) fmin(float4 v1, float v2) {
+ float4 r;
+ r.x = v1.x < v2 ? v1.x : v2;
+ r.y = v1.y < v2 ? v1.y : v2;
+ r.z = v1.z < v2 ? v1.z : v2;
+ r.w = v1.w < v2 ? v1.w : v2;
+ return r;
+}
+
+
+/*
+ * MAX
+ */
+
+extern char __attribute__((overloadable)) max(char v1, char v2) {
+ return v1 > v2 ? v1 : v2;
+}
+
+extern char2 __attribute__((overloadable)) max(char2 v1, char2 v2) {
+ char2 r;
+ r.x = v1.x > v2.x ? v1.x : v2.x;
+ r.y = v1.y > v2.y ? v1.y : v2.y;
+ return r;
+}
+
+extern char3 __attribute__((overloadable)) max(char3 v1, char3 v2) {
+ char3 r;
+ r.x = v1.x > v2.x ? v1.x : v2.x;
+ r.y = v1.y > v2.y ? v1.y : v2.y;
+ r.z = v1.z > v2.z ? v1.z : v2.z;
+ return r;
+}
+
+extern char4 __attribute__((overloadable)) max(char4 v1, char4 v2) {
+ char4 r;
+ r.x = v1.x > v2.x ? v1.x : v2.x;
+ r.y = v1.y > v2.y ? v1.y : v2.y;
+ r.z = v1.z > v2.z ? v1.z : v2.z;
+ r.w = v1.w > v2.w ? v1.w : v2.w;
+ return r;
+}
+
+extern short __attribute__((overloadable)) max(short v1, short v2) {
+ return v1 > v2 ? v1 : v2;
+}
+
+extern short2 __attribute__((overloadable)) max(short2 v1, short2 v2) {
+ short2 r;
+ r.x = v1.x > v2.x ? v1.x : v2.x;
+ r.y = v1.y > v2.y ? v1.y : v2.y;
+ return r;
+}
+
+extern short3 __attribute__((overloadable)) max(short3 v1, short3 v2) {
+ short3 r;
+ r.x = v1.x > v2.x ? v1.x : v2.x;
+ r.y = v1.y > v2.y ? v1.y : v2.y;
+ r.z = v1.z > v2.z ? v1.z : v2.z;
+ return r;
+}
+
+extern short4 __attribute__((overloadable)) max(short4 v1, short4 v2) {
+ short4 r;
+ r.x = v1.x > v2.x ? v1.x : v2.x;
+ r.y = v1.y > v2.y ? v1.y : v2.y;
+ r.z = v1.z > v2.z ? v1.z : v2.z;
+ r.w = v1.w > v2.w ? v1.w : v2.w;
+ return r;
+}
+
+extern int __attribute__((overloadable)) max(int v1, int v2) {
+ return v1 > v2 ? v1 : v2;
+}
+
+extern int2 __attribute__((overloadable)) max(int2 v1, int2 v2) {
+ int2 r;
+ r.x = v1.x > v2.x ? v1.x : v2.x;
+ r.y = v1.y > v2.y ? v1.y : v2.y;
+ return r;
+}
+
+extern int3 __attribute__((overloadable)) max(int3 v1, int3 v2) {
+ int3 r;
+ r.x = v1.x > v2.x ? v1.x : v2.x;
+ r.y = v1.y > v2.y ? v1.y : v2.y;
+ r.z = v1.z > v2.z ? v1.z : v2.z;
+ return r;
+}
+
+extern int4 __attribute__((overloadable)) max(int4 v1, int4 v2) {
+ int4 r;
+ r.x = v1.x > v2.x ? v1.x : v2.x;
+ r.y = v1.y > v2.y ? v1.y : v2.y;
+ r.z = v1.z > v2.z ? v1.z : v2.z;
+ r.w = v1.w > v2.w ? v1.w : v2.w;
+ return r;
+}
+
+extern int64_t __attribute__((overloadable)) max(int64_t v1, int64_t v2) {
+ return v1 > v2 ? v1 : v2;
+}
+
+extern long2 __attribute__((overloadable)) max(long2 v1, long2 v2) {
+ long2 r;
+ r.x = v1.x > v2.x ? v1.x : v2.x;
+ r.y = v1.y > v2.y ? v1.y : v2.y;
+ return r;
+}
+
+extern long3 __attribute__((overloadable)) max(long3 v1, long3 v2) {
+ long3 r;
+ r.x = v1.x > v2.x ? v1.x : v2.x;
+ r.y = v1.y > v2.y ? v1.y : v2.y;
+ r.z = v1.z > v2.z ? v1.z : v2.z;
+ return r;
+}
+
+extern long4 __attribute__((overloadable)) max(long4 v1, long4 v2) {
+ long4 r;
+ r.x = v1.x > v2.x ? v1.x : v2.x;
+ r.y = v1.y > v2.y ? v1.y : v2.y;
+ r.z = v1.z > v2.z ? v1.z : v2.z;
+ r.w = v1.w > v2.w ? v1.w : v2.w;
+ return r;
+}
+
+extern uchar __attribute__((overloadable)) max(uchar v1, uchar v2) {
+ return v1 > v2 ? v1 : v2;
+}
+
+extern uchar2 __attribute__((overloadable)) max(uchar2 v1, uchar2 v2) {
+ uchar2 r;
+ r.x = v1.x > v2.x ? v1.x : v2.x;
+ r.y = v1.y > v2.y ? v1.y : v2.y;
+ return r;
+}
+
+extern uchar3 __attribute__((overloadable)) max(uchar3 v1, uchar3 v2) {
+ uchar3 r;
+ r.x = v1.x > v2.x ? v1.x : v2.x;
+ r.y = v1.y > v2.y ? v1.y : v2.y;
+ r.z = v1.z > v2.z ? v1.z : v2.z;
+ return r;
+}
+
+extern uchar4 __attribute__((overloadable)) max(uchar4 v1, uchar4 v2) {
+ uchar4 r;
+ r.x = v1.x > v2.x ? v1.x : v2.x;
+ r.y = v1.y > v2.y ? v1.y : v2.y;
+ r.z = v1.z > v2.z ? v1.z : v2.z;
+ r.w = v1.w > v2.w ? v1.w : v2.w;
+ return r;
+}
+
+extern ushort __attribute__((overloadable)) max(ushort v1, ushort v2) {
+ return v1 > v2 ? v1 : v2;
+}
+
+extern ushort2 __attribute__((overloadable)) max(ushort2 v1, ushort2 v2) {
+ ushort2 r;
+ r.x = v1.x > v2.x ? v1.x : v2.x;
+ r.y = v1.y > v2.y ? v1.y : v2.y;
+ return r;
+}
+
+extern ushort3 __attribute__((overloadable)) max(ushort3 v1, ushort3 v2) {
+ ushort3 r;
+ r.x = v1.x > v2.x ? v1.x : v2.x;
+ r.y = v1.y > v2.y ? v1.y : v2.y;
+ r.z = v1.z > v2.z ? v1.z : v2.z;
+ return r;
+}
+
+extern ushort4 __attribute__((overloadable)) max(ushort4 v1, ushort4 v2) {
+ ushort4 r;
+ r.x = v1.x > v2.x ? v1.x : v2.x;
+ r.y = v1.y > v2.y ? v1.y : v2.y;
+ r.z = v1.z > v2.z ? v1.z : v2.z;
+ r.w = v1.w > v2.w ? v1.w : v2.w;
+ return r;
+}
+
+extern uint __attribute__((overloadable)) max(uint v1, uint v2) {
+ return v1 > v2 ? v1 : v2;
+}
+
+extern uint2 __attribute__((overloadable)) max(uint2 v1, uint2 v2) {
+ uint2 r;
+ r.x = v1.x > v2.x ? v1.x : v2.x;
+ r.y = v1.y > v2.y ? v1.y : v2.y;
+ return r;
+}
+
+extern uint3 __attribute__((overloadable)) max(uint3 v1, uint3 v2) {
+ uint3 r;
+ r.x = v1.x > v2.x ? v1.x : v2.x;
+ r.y = v1.y > v2.y ? v1.y : v2.y;
+ r.z = v1.z > v2.z ? v1.z : v2.z;
+ return r;
+}
+
+extern uint4 __attribute__((overloadable)) max(uint4 v1, uint4 v2) {
+ uint4 r;
+ r.x = v1.x > v2.x ? v1.x : v2.x;
+ r.y = v1.y > v2.y ? v1.y : v2.y;
+ r.z = v1.z > v2.z ? v1.z : v2.z;
+ r.w = v1.w > v2.w ? v1.w : v2.w;
+ return r;
+}
+
+extern ulong __attribute__((overloadable)) max(ulong v1, ulong v2) {
+ return v1 > v2 ? v1 : v2;
+}
+
+extern ulong2 __attribute__((overloadable)) max(ulong2 v1, ulong2 v2) {
+ ulong2 r;
+ r.x = v1.x > v2.x ? v1.x : v2.x;
+ r.y = v1.y > v2.y ? v1.y : v2.y;
+ return r;
+}
+
+extern ulong3 __attribute__((overloadable)) max(ulong3 v1, ulong3 v2) {
+ ulong3 r;
+ r.x = v1.x > v2.x ? v1.x : v2.x;
+ r.y = v1.y > v2.y ? v1.y : v2.y;
+ r.z = v1.z > v2.z ? v1.z : v2.z;
+ return r;
+}
+
+extern ulong4 __attribute__((overloadable)) max(ulong4 v1, ulong4 v2) {
+ ulong4 r;
+ r.x = v1.x > v2.x ? v1.x : v2.x;
+ r.y = v1.y > v2.y ? v1.y : v2.y;
+ r.z = v1.z > v2.z ? v1.z : v2.z;
+ r.w = v1.w > v2.w ? v1.w : v2.w;
+ return r;
+}
+
+extern float __attribute__((overloadable)) max(float v1, float v2) {
+ return fmax(v1, v2);
+}
+
+extern float2 __attribute__((overloadable)) max(float2 v1, float2 v2) {
+ return fmax(v1, v2);
+}
+
+extern float2 __attribute__((overloadable)) max(float2 v1, float v2) {
+ return fmax(v1, v2);
+}
+
+extern float3 __attribute__((overloadable)) max(float3 v1, float3 v2) {
+ return fmax(v1, v2);
+}
+
+extern float3 __attribute__((overloadable)) max(float3 v1, float v2) {
+ return fmax(v1, v2);
+}
+
+extern float4 __attribute__((overloadable)) max(float4 v1, float4 v2) {
+ return fmax(v1, v2);
+}
+
+extern float4 __attribute__((overloadable)) max(float4 v1, float v2) {
+ return fmax(v1, v2);
+}
+
+
+/*
+ * MIN
+ */
+
+extern int8_t __attribute__((overloadable)) min(int8_t v1, int8_t v2) {
+ return v1 < v2 ? v1 : v2;
+}
+
+extern char2 __attribute__((overloadable)) min(char2 v1, char2 v2) {
+ char2 r;
+ r.x = v1.x < v2.x ? v1.x : v2.x;
+ r.y = v1.y < v2.y ? v1.y : v2.y;
+ return r;
+}
+
+extern char3 __attribute__((overloadable)) min(char3 v1, char3 v2) {
+ char3 r;
+ r.x = v1.x < v2.x ? v1.x : v2.x;
+ r.y = v1.y < v2.y ? v1.y : v2.y;
+ r.z = v1.z < v2.z ? v1.z : v2.z;
+ return r;
+}
+
+extern char4 __attribute__((overloadable)) min(char4 v1, char4 v2) {
+ char4 r;
+ r.x = v1.x < v2.x ? v1.x : v2.x;
+ r.y = v1.y < v2.y ? v1.y : v2.y;
+ r.z = v1.z < v2.z ? v1.z : v2.z;
+ r.w = v1.w < v2.w ? v1.w : v2.w;
+ return r;
+}
+
+extern int16_t __attribute__((overloadable)) min(int16_t v1, int16_t v2) {
+ return v1 < v2 ? v1 : v2;
+}
+
+extern short2 __attribute__((overloadable)) min(short2 v1, short2 v2) {
+ short2 r;
+ r.x = v1.x < v2.x ? v1.x : v2.x;
+ r.y = v1.y < v2.y ? v1.y : v2.y;
+ return r;
+}
+
+extern short3 __attribute__((overloadable)) min(short3 v1, short3 v2) {
+ short3 r;
+ r.x = v1.x < v2.x ? v1.x : v2.x;
+ r.y = v1.y < v2.y ? v1.y : v2.y;
+ r.z = v1.z < v2.z ? v1.z : v2.z;
+ return r;
+}
+
+extern short4 __attribute__((overloadable)) min(short4 v1, short4 v2) {
+ short4 r;
+ r.x = v1.x < v2.x ? v1.x : v2.x;
+ r.y = v1.y < v2.y ? v1.y : v2.y;
+ r.z = v1.z < v2.z ? v1.z : v2.z;
+ r.w = v1.w < v2.w ? v1.w : v2.w;
+ return r;
+}
+
+extern int32_t __attribute__((overloadable)) min(int32_t v1, int32_t v2) {
+ return v1 < v2 ? v1 : v2;
+}
+
+extern int2 __attribute__((overloadable)) min(int2 v1, int2 v2) {
+ int2 r;
+ r.x = v1.x < v2.x ? v1.x : v2.x;
+ r.y = v1.y < v2.y ? v1.y : v2.y;
+ return r;
+}
+
+extern int3 __attribute__((overloadable)) min(int3 v1, int3 v2) {
+ int3 r;
+ r.x = v1.x < v2.x ? v1.x : v2.x;
+ r.y = v1.y < v2.y ? v1.y : v2.y;
+ r.z = v1.z < v2.z ? v1.z : v2.z;
+ return r;
+}
+
+extern int4 __attribute__((overloadable)) min(int4 v1, int4 v2) {
+ int4 r;
+ r.x = v1.x < v2.x ? v1.x : v2.x;
+ r.y = v1.y < v2.y ? v1.y : v2.y;
+ r.z = v1.z < v2.z ? v1.z : v2.z;
+ r.w = v1.w < v2.w ? v1.w : v2.w;
+ return r;
+}
+
+extern int64_t __attribute__((overloadable)) min(int64_t v1, int64_t v2) {
+ return v1 < v2 ? v1 : v2;
+}
+
+extern long2 __attribute__((overloadable)) min(long2 v1, long2 v2) {
+ long2 r;
+ r.x = v1.x < v2.x ? v1.x : v2.x;
+ r.y = v1.y < v2.y ? v1.y : v2.y;
+ return r;
+}
+
+extern long3 __attribute__((overloadable)) min(long3 v1, long3 v2) {
+ long3 r;
+ r.x = v1.x < v2.x ? v1.x : v2.x;
+ r.y = v1.y < v2.y ? v1.y : v2.y;
+ r.z = v1.z < v2.z ? v1.z : v2.z;
+ return r;
+}
+
+extern long4 __attribute__((overloadable)) min(long4 v1, long4 v2) {
+ long4 r;
+ r.x = v1.x < v2.x ? v1.x : v2.x;
+ r.y = v1.y < v2.y ? v1.y : v2.y;
+ r.z = v1.z < v2.z ? v1.z : v2.z;
+ r.w = v1.w < v2.w ? v1.w : v2.w;
+ return r;
+}
+
+extern uchar __attribute__((overloadable)) min(uchar v1, uchar v2) {
+ return v1 < v2 ? v1 : v2;
+}
+
+extern uchar2 __attribute__((overloadable)) min(uchar2 v1, uchar2 v2) {
+ uchar2 r;
+ r.x = v1.x < v2.x ? v1.x : v2.x;
+ r.y = v1.y < v2.y ? v1.y : v2.y;
+ return r;
+}
+
+extern uchar3 __attribute__((overloadable)) min(uchar3 v1, uchar3 v2) {
+ uchar3 r;
+ r.x = v1.x < v2.x ? v1.x : v2.x;
+ r.y = v1.y < v2.y ? v1.y : v2.y;
+ r.z = v1.z < v2.z ? v1.z : v2.z;
+ return r;
+}
+
+extern uchar4 __attribute__((overloadable)) min(uchar4 v1, uchar4 v2) {
+ uchar4 r;
+ r.x = v1.x < v2.x ? v1.x : v2.x;
+ r.y = v1.y < v2.y ? v1.y : v2.y;
+ r.z = v1.z < v2.z ? v1.z : v2.z;
+ r.w = v1.w < v2.w ? v1.w : v2.w;
+ return r;
+}
+
+extern ushort __attribute__((overloadable)) min(ushort v1, ushort v2) {
+ return v1 < v2 ? v1 : v2;
+}
+
+extern ushort2 __attribute__((overloadable)) min(ushort2 v1, ushort2 v2) {
+ ushort2 r;
+ r.x = v1.x < v2.x ? v1.x : v2.x;
+ r.y = v1.y < v2.y ? v1.y : v2.y;
+ return r;
+}
+
+extern ushort3 __attribute__((overloadable)) min(ushort3 v1, ushort3 v2) {
+ ushort3 r;
+ r.x = v1.x < v2.x ? v1.x : v2.x;
+ r.y = v1.y < v2.y ? v1.y : v2.y;
+ r.z = v1.z < v2.z ? v1.z : v2.z;
+ return r;
+}
+
+extern ushort4 __attribute__((overloadable)) min(ushort4 v1, ushort4 v2) {
+ ushort4 r;
+ r.x = v1.x < v2.x ? v1.x : v2.x;
+ r.y = v1.y < v2.y ? v1.y : v2.y;
+ r.z = v1.z < v2.z ? v1.z : v2.z;
+ r.w = v1.w < v2.w ? v1.w : v2.w;
+ return r;
+}
+
+extern uint __attribute__((overloadable)) min(uint v1, uint v2) {
+ return v1 < v2 ? v1 : v2;
+}
+
+extern uint2 __attribute__((overloadable)) min(uint2 v1, uint2 v2) {
+ uint2 r;
+ r.x = v1.x < v2.x ? v1.x : v2.x;
+ r.y = v1.y < v2.y ? v1.y : v2.y;
+ return r;
+}
+
+extern uint3 __attribute__((overloadable)) min(uint3 v1, uint3 v2) {
+ uint3 r;
+ r.x = v1.x < v2.x ? v1.x : v2.x;
+ r.y = v1.y < v2.y ? v1.y : v2.y;
+ r.z = v1.z < v2.z ? v1.z : v2.z;
+ return r;
+}
+
+extern uint4 __attribute__((overloadable)) min(uint4 v1, uint4 v2) {
+ uint4 r;
+ r.x = v1.x < v2.x ? v1.x : v2.x;
+ r.y = v1.y < v2.y ? v1.y : v2.y;
+ r.z = v1.z < v2.z ? v1.z : v2.z;
+ r.w = v1.w < v2.w ? v1.w : v2.w;
+ return r;
+}
+
+extern ulong __attribute__((overloadable)) min(ulong v1, ulong v2) {
+ return v1 < v2 ? v1 : v2;
+}
+
+extern ulong2 __attribute__((overloadable)) min(ulong2 v1, ulong2 v2) {
+ ulong2 r;
+ r.x = v1.x < v2.x ? v1.x : v2.x;
+ r.y = v1.y < v2.y ? v1.y : v2.y;
+ return r;
+}
+
+extern ulong3 __attribute__((overloadable)) min(ulong3 v1, ulong3 v2) {
+ ulong3 r;
+ r.x = v1.x < v2.x ? v1.x : v2.x;
+ r.y = v1.y < v2.y ? v1.y : v2.y;
+ r.z = v1.z < v2.z ? v1.z : v2.z;
+ return r;
+}
+
+extern ulong4 __attribute__((overloadable)) min(ulong4 v1, ulong4 v2) {
+ ulong4 r;
+ r.x = v1.x < v2.x ? v1.x : v2.x;
+ r.y = v1.y < v2.y ? v1.y : v2.y;
+ r.z = v1.z < v2.z ? v1.z : v2.z;
+ r.w = v1.w < v2.w ? v1.w : v2.w;
+ return r;
+}
+
+extern float __attribute__((overloadable)) min(float v1, float v2) {
+ return fmin(v1, v2);
+}
+
+extern float2 __attribute__((overloadable)) min(float2 v1, float2 v2) {
+ return fmin(v1, v2);
+}
+
+extern float2 __attribute__((overloadable)) min(float2 v1, float v2) {
+ return fmin(v1, v2);
+}
+
+extern float3 __attribute__((overloadable)) min(float3 v1, float3 v2) {
+ return fmin(v1, v2);
+}
+
+extern float3 __attribute__((overloadable)) min(float3 v1, float v2) {
+ return fmin(v1, v2);
+}
+
+extern float4 __attribute__((overloadable)) min(float4 v1, float4 v2) {
+ return fmin(v1, v2);
+}
+
+extern float4 __attribute__((overloadable)) min(float4 v1, float v2) {
+ return fmin(v1, v2);
+}
+
+/*
+ * YUV
+ */
+
+extern uchar4 __attribute__((overloadable)) rsYuvToRGBA_uchar4(uchar y, uchar u, uchar v) {
+ short Y = ((short)y) - 16;
+ short U = ((short)u) - 128;
+ short V = ((short)v) - 128;
+
+ short4 p;
+ p.r = (Y * 298 + V * 409 + 128) >> 8;
+ p.g = (Y * 298 - U * 100 - V * 208 + 128) >> 8;
+ p.b = (Y * 298 + U * 516 + 128) >> 8;
+ p.a = 255;
+ p.r = rsClamp(p.r, (short)0, (short)255);
+ p.g = rsClamp(p.g, (short)0, (short)255);
+ p.b = rsClamp(p.b, (short)0, (short)255);
+
+ return convert_uchar4(p);
+}
+
+static float4 yuv_U_values = {0.f, -0.392f * 0.003921569f, +2.02 * 0.003921569f, 0.f};
+static float4 yuv_V_values = {1.603f * 0.003921569f, -0.815f * 0.003921569f, 0.f, 0.f};
+
+extern float4 __attribute__((overloadable)) rsYuvToRGBA_float4(uchar y, uchar u, uchar v) {
+ float4 color = (float)y * 0.003921569f;
+ float4 fU = ((float)u) - 128.f;
+ float4 fV = ((float)v) - 128.f;
+
+ color += fU * yuv_U_values;
+ color += fV * yuv_V_values;
+ color = clamp(color, 0.f, 1.f);
+ return color;
+}
+
+
+/*
+ * half_RECIP
+ */
+
+extern float __attribute__((overloadable)) half_recip(float v) {
+ // FIXME: actual algorithm for generic approximate reciprocal
+ return 1.f / v;
+}
+
+extern float2 __attribute__((overloadable)) half_recip(float2 v) {
+ float2 r;
+ r.x = half_recip(r.x);
+ r.y = half_recip(r.y);
+ return r;
+}
+
+extern float3 __attribute__((overloadable)) half_recip(float3 v) {
+ float3 r;
+ r.x = half_recip(r.x);
+ r.y = half_recip(r.y);
+ r.z = half_recip(r.z);
+ return r;
+}
+
+extern float4 __attribute__((overloadable)) half_recip(float4 v) {
+ float4 r;
+ r.x = half_recip(r.x);
+ r.y = half_recip(r.y);
+ r.z = half_recip(r.z);
+ r.w = half_recip(r.w);
+ return r;
+}
+
+
+/*
+ * half_SQRT
+ */
+
+extern float __attribute__((overloadable)) half_sqrt(float v) {
+ return sqrt(v);
+}
+
+extern float2 __attribute__((overloadable)) half_sqrt(float2 v) {
+ float2 r;
+ r.x = half_sqrt(v.x);
+ r.y = half_sqrt(v.y);
+ return r;
+}
+
+extern float3 __attribute__((overloadable)) half_sqrt(float3 v) {
+ float3 r;
+ r.x = half_sqrt(v.x);
+ r.y = half_sqrt(v.y);
+ r.z = half_sqrt(v.z);
+ return r;
+}
+
+extern float4 __attribute__((overloadable)) half_sqrt(float4 v) {
+ float4 r;
+ r.x = half_sqrt(v.x);
+ r.y = half_sqrt(v.y);
+ r.z = half_sqrt(v.z);
+ r.w = half_sqrt(v.w);
+ return r;
+}
+
+
+/*
+ * half_rsqrt
+ */
+
+extern float __attribute__((overloadable)) half_rsqrt(float v) {
+ return 1.f / sqrt(v);
+}
+
+extern float2 __attribute__((overloadable)) half_rsqrt(float2 v) {
+ float2 r;
+ r.x = half_rsqrt(v.x);
+ r.y = half_rsqrt(v.y);
+ return r;
+}
+
+extern float3 __attribute__((overloadable)) half_rsqrt(float3 v) {
+ float3 r;
+ r.x = half_rsqrt(v.x);
+ r.y = half_rsqrt(v.y);
+ r.z = half_rsqrt(v.z);
+ return r;
+}
+
+extern float4 __attribute__((overloadable)) half_rsqrt(float4 v) {
+ float4 r;
+ r.x = half_rsqrt(v.x);
+ r.y = half_rsqrt(v.y);
+ r.z = half_rsqrt(v.z);
+ r.w = half_rsqrt(v.w);
+ return r;
+}
+
+/**
+ * matrix ops
+ */
+
+extern float4 __attribute__((overloadable))
+rsMatrixMultiply(const rs_matrix4x4 *m, float4 in) {
+ float4 ret;
+ ret.x = (m->m[0] * in.x) + (m->m[4] * in.y) + (m->m[8] * in.z) + (m->m[12] * in.w);
+ ret.y = (m->m[1] * in.x) + (m->m[5] * in.y) + (m->m[9] * in.z) + (m->m[13] * in.w);
+ ret.z = (m->m[2] * in.x) + (m->m[6] * in.y) + (m->m[10] * in.z) + (m->m[14] * in.w);
+ ret.w = (m->m[3] * in.x) + (m->m[7] * in.y) + (m->m[11] * in.z) + (m->m[15] * in.w);
+ return ret;
+}
+
+extern float4 __attribute__((overloadable))
+rsMatrixMultiply(const rs_matrix4x4 *m, float3 in) {
+ float4 ret;
+ ret.x = (m->m[0] * in.x) + (m->m[4] * in.y) + (m->m[8] * in.z) + m->m[12];
+ ret.y = (m->m[1] * in.x) + (m->m[5] * in.y) + (m->m[9] * in.z) + m->m[13];
+ ret.z = (m->m[2] * in.x) + (m->m[6] * in.y) + (m->m[10] * in.z) + m->m[14];
+ ret.w = (m->m[3] * in.x) + (m->m[7] * in.y) + (m->m[11] * in.z) + m->m[15];
+ return ret;
+}
+
+extern float4 __attribute__((overloadable))
+rsMatrixMultiply(const rs_matrix4x4 *m, float2 in) {
+ float4 ret;
+ ret.x = (m->m[0] * in.x) + (m->m[4] * in.y) + m->m[12];
+ ret.y = (m->m[1] * in.x) + (m->m[5] * in.y) + m->m[13];
+ ret.z = (m->m[2] * in.x) + (m->m[6] * in.y) + m->m[14];
+ ret.w = (m->m[3] * in.x) + (m->m[7] * in.y) + m->m[15];
+ return ret;
+}
+
+extern float3 __attribute__((overloadable))
+rsMatrixMultiply(const rs_matrix3x3 *m, float3 in) {
+ float3 ret;
+ ret.x = (m->m[0] * in.x) + (m->m[3] * in.y) + (m->m[6] * in.z);
+ ret.y = (m->m[1] * in.x) + (m->m[4] * in.y) + (m->m[7] * in.z);
+ ret.z = (m->m[2] * in.x) + (m->m[5] * in.y) + (m->m[8] * in.z);
+ return ret;
+}
+
+extern float3 __attribute__((overloadable))
+rsMatrixMultiply(const rs_matrix3x3 *m, float2 in) {
+ float3 ret;
+ ret.x = (m->m[0] * in.x) + (m->m[3] * in.y);
+ ret.y = (m->m[1] * in.x) + (m->m[4] * in.y);
+ ret.z = (m->m[2] * in.x) + (m->m[5] * in.y);
+ return ret;
+}
+
+/**
+ * Pixel Ops
+ */
+extern uchar4 __attribute__((overloadable)) rsPackColorTo8888(float r, float g, float b)
+{
+ uchar4 c;
+ c.x = (uchar)clamp((r * 255.f + 0.5f), 0.f, 255.f);
+ c.y = (uchar)clamp((g * 255.f + 0.5f), 0.f, 255.f);
+ c.z = (uchar)clamp((b * 255.f + 0.5f), 0.f, 255.f);
+ c.w = 255;
+ return c;
+}
+
+extern uchar4 __attribute__((overloadable)) rsPackColorTo8888(float r, float g, float b, float a)
+{
+ uchar4 c;
+ c.x = (uchar)clamp((r * 255.f + 0.5f), 0.f, 255.f);
+ c.y = (uchar)clamp((g * 255.f + 0.5f), 0.f, 255.f);
+ c.z = (uchar)clamp((b * 255.f + 0.5f), 0.f, 255.f);
+ c.w = (uchar)clamp((a * 255.f + 0.5f), 0.f, 255.f);
+ return c;
+}
+
+extern uchar4 __attribute__((overloadable)) rsPackColorTo8888(float3 color)
+{
+ color *= 255.f;
+ color += 0.5f;
+ color = clamp(color, 0.f, 255.f);
+ uchar4 c = {color.x, color.y, color.z, 255};
+ return c;
+}
+
+extern uchar4 __attribute__((overloadable)) rsPackColorTo8888(float4 color)
+{
+ color *= 255.f;
+ color += 0.5f;
+ color = clamp(color, 0.f, 255.f);
+ uchar4 c = {color.x, color.y, color.z, color.w};
+ return c;
+}
+
diff --git a/driver/runtime/arch/neon.ll b/driver/runtime/arch/neon.ll
new file mode 100644
index 0000000..4a1172b
--- /dev/null
+++ b/driver/runtime/arch/neon.ll
@@ -0,0 +1,1172 @@
+target datalayout = "e-p:32:32:32-i1:8:8-i8:8:8-i16:16:16-i32:32:32-i64:64:64-f32:32:32-f64:64:64-v64:64:64-v128:64:128-a0:0:64-n32-S64"
+target triple = "armv7-none-linux-gnueabi"
+
+;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;
+;;;;;;;;; INTRINSICS ;;;;;;;;;;
+;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;
+
+declare <2 x float> @llvm.arm.neon.vmaxs.v2f32(<2 x float>, <2 x float>) nounwind readnone
+declare <4 x float> @llvm.arm.neon.vmaxs.v4f32(<4 x float>, <4 x float>) nounwind readnone
+declare <2 x i32> @llvm.arm.neon.vmaxs.v2i32(<2 x i32>, <2 x i32>) nounwind readnone
+declare <4 x i32> @llvm.arm.neon.vmaxs.v4i32(<4 x i32>, <4 x i32>) nounwind readnone
+declare <2 x i32> @llvm.arm.neon.vmaxu.v2i32(<2 x i32>, <2 x i32>) nounwind readnone
+declare <4 x i32> @llvm.arm.neon.vmaxu.v4i32(<4 x i32>, <4 x i32>) nounwind readnone
+declare <4 x i16> @llvm.arm.neon.vmaxs.v4i16(<4 x i16>, <4 x i16>) nounwind readnone
+declare <4 x i16> @llvm.arm.neon.vmaxu.v4i16(<4 x i16>, <4 x i16>) nounwind readnone
+
+declare <2 x float> @llvm.arm.neon.vmins.v2f32(<2 x float>, <2 x float>) nounwind readnone
+declare <4 x float> @llvm.arm.neon.vmins.v4f32(<4 x float>, <4 x float>) nounwind readnone
+declare <2 x i32> @llvm.arm.neon.vmins.v2i32(<2 x i32>, <2 x i32>) nounwind readnone
+declare <4 x i32> @llvm.arm.neon.vmins.v4i32(<4 x i32>, <4 x i32>) nounwind readnone
+declare <2 x i32> @llvm.arm.neon.vminu.v2i32(<2 x i32>, <2 x i32>) nounwind readnone
+declare <4 x i32> @llvm.arm.neon.vminu.v4i32(<4 x i32>, <4 x i32>) nounwind readnone
+declare <4 x i16> @llvm.arm.neon.vmins.v4i16(<4 x i16>, <4 x i16>) nounwind readnone
+declare <4 x i16> @llvm.arm.neon.vminu.v4i16(<4 x i16>, <4 x i16>) nounwind readnone
+
+declare <8 x i8> @llvm.arm.neon.vqshiftns.v8i8(<8 x i16>, <8 x i16>) nounwind readnone
+declare <4 x i16> @llvm.arm.neon.vqshiftns.v4i16(<4 x i32>, <4 x i32>) nounwind readnone
+declare <2 x i32> @llvm.arm.neon.vqshiftns.v2i32(<2 x i64>, <2 x i64>) nounwind readnone
+
+declare <8 x i8> @llvm.arm.neon.vqshiftnu.v8i8(<8 x i16>, <8 x i16>) nounwind readnone
+declare <4 x i16> @llvm.arm.neon.vqshiftnu.v4i16(<4 x i32>, <4 x i32>) nounwind readnone
+declare <2 x i32> @llvm.arm.neon.vqshiftnu.v2i32(<2 x i64>, <2 x i64>) nounwind readnone
+
+declare <8 x i8> @llvm.arm.neon.vqshiftnsu.v8i8(<8 x i16>, <8 x i16>) nounwind readnone
+declare <4 x i16> @llvm.arm.neon.vqshiftnsu.v4i16(<4 x i32>, <4 x i32>) nounwind readnone
+declare <2 x i32> @llvm.arm.neon.vqshiftnsu.v2i32(<2 x i64>, <2 x i64>) nounwind readnone
+
+declare <2 x float> @llvm.arm.neon.vrecpe.v2f32(<2 x float>) nounwind readnone
+declare <4 x float> @llvm.arm.neon.vrecpe.v4f32(<4 x float>) nounwind readnone
+
+declare <2 x float> @llvm.arm.neon.vrsqrte.v2f32(<2 x float>) nounwind readnone
+declare <4 x float> @llvm.arm.neon.vrsqrte.v4f32(<4 x float>) nounwind readnone
+
+;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;
+;;;;;;;;; HELPERS ;;;;;;;;;;
+;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;
+
+define internal <4 x float> @smear_4f(float %in) nounwind readnone alwaysinline {
+ %1 = insertelement <4 x float> undef, float %in, i32 0
+ %2 = insertelement <4 x float> %1, float %in, i32 1
+ %3 = insertelement <4 x float> %2, float %in, i32 2
+ %4 = insertelement <4 x float> %3, float %in, i32 3
+ ret <4 x float> %4
+}
+
+define internal <4 x i32> @smear_4i(i32 %in) nounwind readnone alwaysinline {
+ %1 = insertelement <4 x i32> undef, i32 %in, i32 0
+ %2 = insertelement <4 x i32> %1, i32 %in, i32 1
+ %3 = insertelement <4 x i32> %2, i32 %in, i32 2
+ %4 = insertelement <4 x i32> %3, i32 %in, i32 3
+ ret <4 x i32> %4
+}
+
+define internal <4 x i16> @smear_4s(i16 %in) nounwind readnone alwaysinline {
+ %1 = insertelement <4 x i16> undef, i16 %in, i32 0
+ %2 = insertelement <4 x i16> %1, i16 %in, i32 1
+ %3 = insertelement <4 x i16> %2, i16 %in, i32 2
+ %4 = insertelement <4 x i16> %3, i16 %in, i32 3
+ ret <4 x i16> %4
+}
+
+
+
+define internal <2 x float> @smear_2f(float %in) nounwind readnone alwaysinline {
+ %1 = insertelement <2 x float> undef, float %in, i32 0
+ %2 = insertelement <2 x float> %1, float %in, i32 1
+ ret <2 x float> %2
+}
+
+define internal <2 x i32> @smear_2i(i32 %in) nounwind readnone alwaysinline {
+ %1 = insertelement <2 x i32> undef, i32 %in, i32 0
+ %2 = insertelement <2 x i32> %1, i32 %in, i32 1
+ ret <2 x i32> %2
+}
+
+define internal <2 x i16> @smear_2s(i16 %in) nounwind readnone alwaysinline {
+ %1 = insertelement <2 x i16> undef, i16 %in, i32 0
+ %2 = insertelement <2 x i16> %1, i16 %in, i32 1
+ ret <2 x i16> %2
+}
+
+
+define internal <4 x i32> @smear_4i32(i32 %in) nounwind readnone alwaysinline {
+ %1 = insertelement <4 x i32> undef, i32 %in, i32 0
+ %2 = insertelement <4 x i32> %1, i32 %in, i32 1
+ %3 = insertelement <4 x i32> %2, i32 %in, i32 2
+ %4 = insertelement <4 x i32> %3, i32 %in, i32 3
+ ret <4 x i32> %4
+}
+
+
+;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;
+;;;;;;;;; CLAMP ;;;;;;;;;;
+;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;
+
+define <4 x float> @_Z5clampDv4_fS_S_(<4 x float> %value, <4 x float> %low, <4 x float> %high) nounwind readonly {
+ %1 = tail call <4 x float> @llvm.arm.neon.vmins.v4f32(<4 x float> %value, <4 x float> %high) nounwind readnone
+ %2 = tail call <4 x float> @llvm.arm.neon.vmaxs.v4f32(<4 x float> %1, <4 x float> %low) nounwind readnone
+ ret <4 x float> %2
+}
+
+define <4 x float> @_Z5clampDv4_fff(<4 x float> %value, float %low, float %high) nounwind readonly {
+ %_high = tail call <4 x float> @smear_4f(float %high) nounwind readnone
+ %_low = tail call <4 x float> @smear_4f(float %low) nounwind readnone
+ %out = tail call <4 x float> @_Z5clampDv4_fS_S_(<4 x float> %value, <4 x float> %_low, <4 x float> %_high) nounwind readonly
+ ret <4 x float> %out
+}
+
+define <3 x float> @_Z5clampDv3_fS_S_(<3 x float> %value, <3 x float> %low, <3 x float> %high) nounwind readonly {
+ %_value = shufflevector <3 x float> %value, <3 x float> undef, <4 x i32> <i32 0, i32 1, i32 2, i32 3>
+ %_low = shufflevector <3 x float> %low, <3 x float> undef, <4 x i32> <i32 0, i32 1, i32 2, i32 3>
+ %_high = shufflevector <3 x float> %high, <3 x float> undef, <4 x i32> <i32 0, i32 1, i32 2, i32 3>
+ %a = tail call <4 x float> @llvm.arm.neon.vmins.v4f32(<4 x float> %_value, <4 x float> %_high) nounwind readnone
+ %b = tail call <4 x float> @llvm.arm.neon.vmaxs.v4f32(<4 x float> %a, <4 x float> %_low) nounwind readnone
+ %c = shufflevector <4 x float> %b, <4 x float> undef, <3 x i32> <i32 0, i32 1, i32 2>
+ ret <3 x float> %c
+}
+
+define <3 x float> @_Z5clampDv3_fff(<3 x float> %value, float %low, float %high) nounwind readonly {
+ %_value = shufflevector <3 x float> %value, <3 x float> undef, <4 x i32> <i32 0, i32 1, i32 2, i32 3>
+ %_high = tail call <4 x float> @smear_4f(float %high) nounwind readnone
+ %_low = tail call <4 x float> @smear_4f(float %low) nounwind readnone
+ %a = tail call <4 x float> @llvm.arm.neon.vmins.v4f32(<4 x float> %_value, <4 x float> %_high) nounwind readnone
+ %b = tail call <4 x float> @llvm.arm.neon.vmaxs.v4f32(<4 x float> %a, <4 x float> %_low) nounwind readnone
+ %c = shufflevector <4 x float> %b, <4 x float> undef, <3 x i32> <i32 0, i32 1, i32 2>
+ ret <3 x float> %c
+}
+
+define <2 x float> @_Z5clampDv2_fS_S_(<2 x float> %value, <2 x float> %low, <2 x float> %high) nounwind readonly {
+ %1 = tail call <2 x float> @llvm.arm.neon.vmins.v2f32(<2 x float> %value, <2 x float> %high) nounwind readnone
+ %2 = tail call <2 x float> @llvm.arm.neon.vmaxs.v2f32(<2 x float> %1, <2 x float> %low) nounwind readnone
+ ret <2 x float> %2
+}
+
+define <2 x float> @_Z5clampDv2_fff(<2 x float> %value, float %low, float %high) nounwind readonly {
+ %_high = tail call <2 x float> @smear_2f(float %high) nounwind readnone
+ %_low = tail call <2 x float> @smear_2f(float %low) nounwind readnone
+ %a = tail call <2 x float> @llvm.arm.neon.vmins.v2f32(<2 x float> %value, <2 x float> %_high) nounwind readnone
+ %b = tail call <2 x float> @llvm.arm.neon.vmaxs.v2f32(<2 x float> %a, <2 x float> %_low) nounwind readnone
+ ret <2 x float> %b
+}
+
+define float @_Z5clampfff(float %value, float %low, float %high) nounwind readonly {
+ %1 = fcmp olt float %value, %high
+ %2 = select i1 %1, float %value, float %high
+ %3 = fcmp ogt float %2, %low
+ %4 = select i1 %3, float %2, float %low
+ ret float %4
+}
+
+
+
+define <4 x i32> @_Z5clampDv4_iS_S_(<4 x i32> %value, <4 x i32> %low, <4 x i32> %high) nounwind readonly {
+ %1 = tail call <4 x i32> @llvm.arm.neon.vmins.v4i32(<4 x i32> %value, <4 x i32> %high) nounwind readnone
+ %2 = tail call <4 x i32> @llvm.arm.neon.vmaxs.v4i32(<4 x i32> %1, <4 x i32> %low) nounwind readnone
+ ret <4 x i32> %2
+}
+
+define <4 x i32> @_Z5clampDv4_iii(<4 x i32> %value, i32 %low, i32 %high) nounwind readonly {
+ %_high = tail call <4 x i32> @smear_4i(i32 %high) nounwind readnone
+ %_low = tail call <4 x i32> @smear_4i(i32 %low) nounwind readnone
+ %1 = tail call <4 x i32> @llvm.arm.neon.vmins.v4i32(<4 x i32> %value, <4 x i32> %_high) nounwind readnone
+ %2 = tail call <4 x i32> @llvm.arm.neon.vmaxs.v4i32(<4 x i32> %1, <4 x i32> %_low) nounwind readnone
+ ret <4 x i32> %2
+}
+
+define <3 x i32> @_Z5clampDv3_iS_S_(<3 x i32> %value, <3 x i32> %low, <3 x i32> %high) nounwind readonly {
+ %_value = shufflevector <3 x i32> %value, <3 x i32> undef, <4 x i32> <i32 0, i32 1, i32 2, i32 3>
+ %_low = shufflevector <3 x i32> %low, <3 x i32> undef, <4 x i32> <i32 0, i32 1, i32 2, i32 3>
+ %_high = shufflevector <3 x i32> %high, <3 x i32> undef, <4 x i32> <i32 0, i32 1, i32 2, i32 3>
+ %a = tail call <4 x i32> @llvm.arm.neon.vmins.v4i32(<4 x i32> %_value, <4 x i32> %_high) nounwind readnone
+ %b = tail call <4 x i32> @llvm.arm.neon.vmaxs.v4i32(<4 x i32> %a, <4 x i32> %_low) nounwind readnone
+ %c = shufflevector <4 x i32> %b, <4 x i32> undef, <3 x i32> <i32 0, i32 1, i32 2>
+ ret <3 x i32> %c
+}
+
+define <3 x i32> @_Z5clampDv3_iii(<3 x i32> %value, i32 %low, i32 %high) nounwind readonly {
+ %_value = shufflevector <3 x i32> %value, <3 x i32> undef, <4 x i32> <i32 0, i32 1, i32 2, i32 3>
+ %_high = tail call <4 x i32> @smear_4i(i32 %high) nounwind readnone
+ %_low = tail call <4 x i32> @smear_4i(i32 %low) nounwind readnone
+ %a = tail call <4 x i32> @llvm.arm.neon.vmins.v4i32(<4 x i32> %_value, <4 x i32> %_high) nounwind readnone
+ %b = tail call <4 x i32> @llvm.arm.neon.vmaxs.v4i32(<4 x i32> %a, <4 x i32> %_low) nounwind readnone
+ %c = shufflevector <4 x i32> %b, <4 x i32> undef, <3 x i32> <i32 0, i32 1, i32 2>
+ ret <3 x i32> %c
+}
+
+define <2 x i32> @_Z5clampDv2_iS_S_(<2 x i32> %value, <2 x i32> %low, <2 x i32> %high) nounwind readonly {
+ %1 = tail call <2 x i32> @llvm.arm.neon.vmins.v2i32(<2 x i32> %value, <2 x i32> %high) nounwind readnone
+ %2 = tail call <2 x i32> @llvm.arm.neon.vmaxs.v2i32(<2 x i32> %1, <2 x i32> %low) nounwind readnone
+ ret <2 x i32> %2
+}
+
+define <2 x i32> @_Z5clampDv2_iii(<2 x i32> %value, i32 %low, i32 %high) nounwind readonly {
+ %_high = tail call <2 x i32> @smear_2i(i32 %high) nounwind readnone
+ %_low = tail call <2 x i32> @smear_2i(i32 %low) nounwind readnone
+ %a = tail call <2 x i32> @llvm.arm.neon.vmins.v2i32(<2 x i32> %value, <2 x i32> %_high) nounwind readnone
+ %b = tail call <2 x i32> @llvm.arm.neon.vmaxs.v2i32(<2 x i32> %a, <2 x i32> %_low) nounwind readnone
+ ret <2 x i32> %b
+}
+
+
+
+define <4 x i32> @_Z5clampDv4_jS_S_(<4 x i32> %value, <4 x i32> %low, <4 x i32> %high) nounwind readonly {
+ %1 = tail call <4 x i32> @llvm.arm.neon.vminu.v4i32(<4 x i32> %value, <4 x i32> %high) nounwind readnone
+ %2 = tail call <4 x i32> @llvm.arm.neon.vmaxu.v4i32(<4 x i32> %1, <4 x i32> %low) nounwind readnone
+ ret <4 x i32> %2
+}
+
+define <4 x i32> @_Z5clampDv4_jjj(<4 x i32> %value, i32 %low, i32 %high) nounwind readonly {
+ %_high = tail call <4 x i32> @smear_4i(i32 %high) nounwind readnone
+ %_low = tail call <4 x i32> @smear_4i(i32 %low) nounwind readnone
+ %1 = tail call <4 x i32> @llvm.arm.neon.vminu.v4i32(<4 x i32> %value, <4 x i32> %_high) nounwind readnone
+ %2 = tail call <4 x i32> @llvm.arm.neon.vmaxu.v4i32(<4 x i32> %1, <4 x i32> %_low) nounwind readnone
+ ret <4 x i32> %2
+}
+
+define <3 x i32> @_Z5clampDv3_jS_S_(<3 x i32> %value, <3 x i32> %low, <3 x i32> %high) nounwind readonly {
+ %_value = shufflevector <3 x i32> %value, <3 x i32> undef, <4 x i32> <i32 0, i32 1, i32 2, i32 3>
+ %_low = shufflevector <3 x i32> %low, <3 x i32> undef, <4 x i32> <i32 0, i32 1, i32 2, i32 3>
+ %_high = shufflevector <3 x i32> %high, <3 x i32> undef, <4 x i32> <i32 0, i32 1, i32 2, i32 3>
+ %a = tail call <4 x i32> @llvm.arm.neon.vminu.v4i32(<4 x i32> %_value, <4 x i32> %_high) nounwind readnone
+ %b = tail call <4 x i32> @llvm.arm.neon.vmaxu.v4i32(<4 x i32> %a, <4 x i32> %_low) nounwind readnone
+ %c = shufflevector <4 x i32> %b, <4 x i32> undef, <3 x i32> <i32 0, i32 1, i32 2>
+ ret <3 x i32> %c
+}
+
+define <3 x i32> @_Z5clampDv3_jjj(<3 x i32> %value, i32 %low, i32 %high) nounwind readonly {
+ %_value = shufflevector <3 x i32> %value, <3 x i32> undef, <4 x i32> <i32 0, i32 1, i32 2, i32 3>
+ %_high = tail call <4 x i32> @smear_4i(i32 %high) nounwind readnone
+ %_low = tail call <4 x i32> @smear_4i(i32 %low) nounwind readnone
+ %a = tail call <4 x i32> @llvm.arm.neon.vminu.v4i32(<4 x i32> %_value, <4 x i32> %_high) nounwind readnone
+ %b = tail call <4 x i32> @llvm.arm.neon.vmaxu.v4i32(<4 x i32> %a, <4 x i32> %_low) nounwind readnone
+ %c = shufflevector <4 x i32> %b, <4 x i32> undef, <3 x i32> <i32 0, i32 1, i32 2>
+ ret <3 x i32> %c
+}
+
+define <2 x i32> @_Z5clampDv2_jS_S_(<2 x i32> %value, <2 x i32> %low, <2 x i32> %high) nounwind readonly {
+ %1 = tail call <2 x i32> @llvm.arm.neon.vminu.v2i32(<2 x i32> %value, <2 x i32> %high) nounwind readnone
+ %2 = tail call <2 x i32> @llvm.arm.neon.vmaxu.v2i32(<2 x i32> %1, <2 x i32> %low) nounwind readnone
+ ret <2 x i32> %2
+}
+
+define <2 x i32> @_Z5clampDv2_jjj(<2 x i32> %value, i32 %low, i32 %high) nounwind readonly {
+ %_high = tail call <2 x i32> @smear_2i(i32 %high) nounwind readnone
+ %_low = tail call <2 x i32> @smear_2i(i32 %low) nounwind readnone
+ %a = tail call <2 x i32> @llvm.arm.neon.vminu.v2i32(<2 x i32> %value, <2 x i32> %_high) nounwind readnone
+ %b = tail call <2 x i32> @llvm.arm.neon.vmaxu.v2i32(<2 x i32> %a, <2 x i32> %_low) nounwind readnone
+ ret <2 x i32> %b
+}
+
+
+;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;
+;;;;;;;;; FMAX ;;;;;;;;;;
+;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;
+
+define <4 x float> @_Z4fmaxDv4_fS_(<4 x float> %v1, <4 x float> %v2) nounwind readonly {
+ %1 = tail call <4 x float> @llvm.arm.neon.vmaxs.v4f32(<4 x float> %v1, <4 x float> %v2) nounwind readnone
+ ret <4 x float> %1
+}
+
+define <4 x float> @_Z4fmaxDv4_ff(<4 x float> %v1, float %v2) nounwind readonly {
+ %1 = tail call <4 x float> @smear_4f(float %v2) nounwind readnone
+ %2 = tail call <4 x float> @llvm.arm.neon.vmaxs.v4f32(<4 x float> %v1, <4 x float> %1) nounwind readnone
+ ret <4 x float> %2
+}
+
+define <3 x float> @_Z4fmaxDv3_fS_(<3 x float> %v1, <3 x float> %v2) nounwind readonly {
+ %1 = shufflevector <3 x float> %v1, <3 x float> undef, <4 x i32> <i32 0, i32 1, i32 2, i32 3>
+ %2 = shufflevector <3 x float> %v2, <3 x float> undef, <4 x i32> <i32 0, i32 1, i32 2, i32 3>
+ %3 = tail call <4 x float> @llvm.arm.neon.vmaxs.v4f32(<4 x float> %1, <4 x float> %2) nounwind readnone
+ %4 = shufflevector <4 x float> %3, <4 x float> undef, <3 x i32> <i32 0, i32 1, i32 2>
+ ret <3 x float> %4
+}
+
+define <3 x float> @_Z4fmaxDv3_ff(<3 x float> %v1, float %v2) nounwind readonly {
+ %1 = shufflevector <3 x float> %v1, <3 x float> undef, <4 x i32> <i32 0, i32 1, i32 2, i32 3>
+ %2 = tail call <4 x float> @smear_4f(float %v2) nounwind readnone
+ %3 = tail call <4 x float> @llvm.arm.neon.vmaxs.v4f32(<4 x float> %1, <4 x float> %2) nounwind readnone
+ %c = shufflevector <4 x float> %3, <4 x float> undef, <3 x i32> <i32 0, i32 1, i32 2>
+ ret <3 x float> %c
+}
+
+define <2 x float> @_Z4fmaxDv2_fS_(<2 x float> %v1, <2 x float> %v2) nounwind readonly {
+ %1 = tail call <2 x float> @llvm.arm.neon.vmaxs.v2f32(<2 x float> %v1, <2 x float> %v2) nounwind readnone
+ ret <2 x float> %1
+}
+
+define <2 x float> @_Z4fmaxDv2_ff(<2 x float> %v1, float %v2) nounwind readonly {
+ %1 = tail call <2 x float> @smear_2f(float %v2) nounwind readnone
+ %2 = tail call <2 x float> @llvm.arm.neon.vmaxs.v2f32(<2 x float> %v1, <2 x float> %1) nounwind readnone
+ ret <2 x float> %2
+}
+
+define float @_Z4fmaxff(float %v1, float %v2) nounwind readonly {
+ %1 = fcmp ogt float %v1, %v2
+ %2 = select i1 %1, float %v1, float %v2
+ ret float %2
+}
+
+
+;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;
+;;;;;;;;; FMIN ;;;;;;;;;;
+;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;
+
+define <4 x float> @_Z4fminDv4_fS_(<4 x float> %v1, <4 x float> %v2) nounwind readonly {
+ %1 = tail call <4 x float> @llvm.arm.neon.vmins.v4f32(<4 x float> %v1, <4 x float> %v2) nounwind readnone
+ ret <4 x float> %1
+}
+
+define <4 x float> @_Z4fminDv4_ff(<4 x float> %v1, float %v2) nounwind readonly {
+ %1 = tail call <4 x float> @smear_4f(float %v2) nounwind readnone
+ %2 = tail call <4 x float> @llvm.arm.neon.vmins.v4f32(<4 x float> %v1, <4 x float> %1) nounwind readnone
+ ret <4 x float> %2
+}
+
+define <3 x float> @_Z4fminDv3_fS_(<3 x float> %v1, <3 x float> %v2) nounwind readonly {
+ %1 = shufflevector <3 x float> %v1, <3 x float> undef, <4 x i32> <i32 0, i32 1, i32 2, i32 3>
+ %2 = shufflevector <3 x float> %v2, <3 x float> undef, <4 x i32> <i32 0, i32 1, i32 2, i32 3>
+ %3 = tail call <4 x float> @llvm.arm.neon.vmins.v4f32(<4 x float> %1, <4 x float> %2) nounwind readnone
+ %4 = shufflevector <4 x float> %3, <4 x float> undef, <3 x i32> <i32 0, i32 1, i32 2>
+ ret <3 x float> %4
+}
+
+define <3 x float> @_Z4fminDv3_ff(<3 x float> %v1, float %v2) nounwind readonly {
+ %1 = shufflevector <3 x float> %v1, <3 x float> undef, <4 x i32> <i32 0, i32 1, i32 2, i32 3>
+ %2 = tail call <4 x float> @smear_4f(float %v2) nounwind readnone
+ %3 = tail call <4 x float> @llvm.arm.neon.vmins.v4f32(<4 x float> %1, <4 x float> %2) nounwind readnone
+ %c = shufflevector <4 x float> %3, <4 x float> undef, <3 x i32> <i32 0, i32 1, i32 2>
+ ret <3 x float> %c
+}
+
+define <2 x float> @_Z4fminDv2_fS_(<2 x float> %v1, <2 x float> %v2) nounwind readonly {
+ %1 = tail call <2 x float> @llvm.arm.neon.vmins.v2f32(<2 x float> %v1, <2 x float> %v2) nounwind readnone
+ ret <2 x float> %1
+}
+
+define <2 x float> @_Z4fminDv2_ff(<2 x float> %v1, float %v2) nounwind readonly {
+ %1 = tail call <2 x float> @smear_2f(float %v2) nounwind readnone
+ %2 = tail call <2 x float> @llvm.arm.neon.vmins.v2f32(<2 x float> %v1, <2 x float> %1) nounwind readnone
+ ret <2 x float> %2
+}
+
+define float @_Z4fminff(float %v1, float %v2) nounwind readnone {
+ %1 = fcmp olt float %v1, %v2
+ %2 = select i1 %1, float %v1, float %v2
+ ret float %2
+}
+
+
+;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;
+;;;;;;;;; MAX ;;;;;;;;;;
+;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;
+
+define signext i8 @_Z3maxcc(i8 signext %v1, i8 signext %v2) nounwind readnone {
+ %1 = icmp sgt i8 %v1, %v2
+ %2 = select i1 %1, i8 %v1, i8 %v2
+ ret i8 %2
+}
+
+define <2 x i8> @_Z3maxDv2_cS_(<2 x i8> %v1, <2 x i8> %v2) nounwind readnone {
+ %1 = sext <2 x i8> %v1 to <2 x i32>
+ %2 = sext <2 x i8> %v2 to <2 x i32>
+ %3 = tail call <2 x i32> @llvm.arm.neon.vmaxs.v2i32(<2 x i32> %1, <2 x i32> %2) nounwind readnone
+ %4 = trunc <2 x i32> %3 to <2 x i8>
+ ret <2 x i8> %4
+}
+
+define <3 x i8> @_Z3maxDv3_cS_(<3 x i8> %v1, <3 x i8> %v2) nounwind readnone {
+ %1 = sext <3 x i8> %v1 to <3 x i32>
+ %2 = sext <3 x i8> %v2 to <3 x i32>
+ %3 = shufflevector <3 x i32> %1, <3 x i32> undef, <4 x i32> <i32 0, i32 1, i32 2, i32 3>
+ %4 = shufflevector <3 x i32> %2, <3 x i32> undef, <4 x i32> <i32 0, i32 1, i32 2, i32 3>
+ %5 = tail call <4 x i32> @llvm.arm.neon.vmaxs.v4i32(<4 x i32> %3, <4 x i32> %4) nounwind readnone
+ %6 = shufflevector <4 x i32> %5, <4 x i32> undef, <3 x i32> <i32 0, i32 1, i32 2>
+ %7 = trunc <3 x i32> %6 to <3 x i8>
+ ret <3 x i8> %7
+}
+
+define <4 x i8> @_Z3maxDv4_cS_(<4 x i8> %v1, <4 x i8> %v2) nounwind readnone {
+ %1 = sext <4 x i8> %v1 to <4 x i32>
+ %2 = sext <4 x i8> %v2 to <4 x i32>
+ %3 = tail call <4 x i32> @llvm.arm.neon.vmaxs.v4i32(<4 x i32> %1, <4 x i32> %2) nounwind readnone
+ %4 = trunc <4 x i32> %3 to <4 x i8>
+ ret <4 x i8> %4
+}
+
+define signext i16 @_Z3maxss(i16 signext %v1, i16 signext %v2) nounwind readnone {
+ %1 = icmp sgt i16 %v1, %v2
+ %2 = select i1 %1, i16 %v1, i16 %v2
+ ret i16 %2
+}
+
+define <2 x i16> @_Z3maxDv2_sS_(<2 x i16> %v1, <2 x i16> %v2) nounwind readnone {
+ %1 = sext <2 x i16> %v1 to <2 x i32>
+ %2 = sext <2 x i16> %v2 to <2 x i32>
+ %3 = tail call <2 x i32> @llvm.arm.neon.vmaxs.v2i32(<2 x i32> %1, <2 x i32> %2) nounwind readnone
+ %4 = trunc <2 x i32> %3 to <2 x i16>
+ ret <2 x i16> %4
+}
+
+define <3 x i16> @_Z3maxDv3_sS_(<3 x i16> %v1, <3 x i16> %v2) nounwind readnone {
+ %1 = sext <3 x i16> %v1 to <3 x i32>
+ %2 = sext <3 x i16> %v2 to <3 x i32>
+ %3 = shufflevector <3 x i32> %1, <3 x i32> undef, <4 x i32> <i32 0, i32 1, i32 2, i32 3>
+ %4 = shufflevector <3 x i32> %2, <3 x i32> undef, <4 x i32> <i32 0, i32 1, i32 2, i32 3>
+ %5 = tail call <4 x i32> @llvm.arm.neon.vmaxs.v4i32(<4 x i32> %3, <4 x i32> %4) nounwind readnone
+ %6 = shufflevector <4 x i32> %5, <4 x i32> undef, <3 x i32> <i32 0, i32 1, i32 2>
+ %7 = trunc <3 x i32> %6 to <3 x i16>
+ ret <3 x i16> %7
+}
+
+define <4 x i16> @_Z3maxDv4_sS_(<4 x i16> %v1, <4 x i16> %v2) nounwind readnone {
+ %1 = sext <4 x i16> %v1 to <4 x i32>
+ %2 = sext <4 x i16> %v2 to <4 x i32>
+ %3 = tail call <4 x i32> @llvm.arm.neon.vmaxs.v4i32(<4 x i32> %1, <4 x i32> %2) nounwind readnone
+ %4 = trunc <4 x i32> %3 to <4 x i16>
+ ret <4 x i16> %4
+}
+
+define i32 @_Z3maxii(i32 %v1, i32 %v2) nounwind readnone {
+ %1 = icmp sgt i32 %v1, %v2
+ %2 = select i1 %1, i32 %v1, i32 %v2
+ ret i32 %2
+}
+
+define <2 x i32> @_Z3maxDv2_iS_(<2 x i32> %v1, <2 x i32> %v2) nounwind readnone {
+ %1 = tail call <2 x i32> @llvm.arm.neon.vmaxs.v2i32(<2 x i32> %v1, <2 x i32> %v2) nounwind readnone
+ ret <2 x i32> %1
+}
+
+define <3 x i32> @_Z3maxDv3_iS_(<3 x i32> %v1, <3 x i32> %v2) nounwind readnone {
+ %1 = shufflevector <3 x i32> %v1, <3 x i32> undef, <4 x i32> <i32 0, i32 1, i32 2, i32 3>
+ %2 = shufflevector <3 x i32> %v2, <3 x i32> undef, <4 x i32> <i32 0, i32 1, i32 2, i32 3>
+ %3 = tail call <4 x i32 > @llvm.arm.neon.vmaxs.v4i32(<4 x i32> %1, <4 x i32> %2) nounwind readnone
+ %4 = shufflevector <4 x i32> %3, <4 x i32> undef, <3 x i32> <i32 0, i32 1, i32 2>
+ ret <3 x i32> %4
+}
+
+define <4 x i32> @_Z3maxDv4_iS_(<4 x i32> %v1, <4 x i32> %v2) nounwind readnone {
+ %1 = tail call <4 x i32> @llvm.arm.neon.vmaxs.v4i32(<4 x i32> %v1, <4 x i32> %v2) nounwind readnone
+ ret <4 x i32> %1
+}
+
+define i64 @_Z3maxxx(i64 %v1, i64 %v2) nounwind readnone {
+ %1 = icmp sgt i64 %v1, %v2
+ %2 = select i1 %1, i64 %v1, i64 %v2
+ ret i64 %2
+}
+
+; TODO: long vector types
+
+define zeroext i8 @_Z3maxhh(i8 zeroext %v1, i8 zeroext %v2) nounwind readnone {
+ %1 = icmp ugt i8 %v1, %v2
+ %2 = select i1 %1, i8 %v1, i8 %v2
+ ret i8 %2
+}
+
+define <2 x i8> @_Z3maxDv2_hS_(<2 x i8> %v1, <2 x i8> %v2) nounwind readnone {
+ %1 = zext <2 x i8> %v1 to <2 x i32>
+ %2 = zext <2 x i8> %v2 to <2 x i32>
+ %3 = tail call <2 x i32> @llvm.arm.neon.vmaxu.v2i32(<2 x i32> %1, <2 x i32> %2) nounwind readnone
+ %4 = trunc <2 x i32> %3 to <2 x i8>
+ ret <2 x i8> %4
+}
+
+define <3 x i8> @_Z3maxDv3_hS_(<3 x i8> %v1, <3 x i8> %v2) nounwind readnone {
+ %1 = zext <3 x i8> %v1 to <3 x i32>
+ %2 = zext <3 x i8> %v2 to <3 x i32>
+ %3 = shufflevector <3 x i32> %1, <3 x i32> undef, <4 x i32> <i32 0, i32 1, i32 2, i32 3>
+ %4 = shufflevector <3 x i32> %2, <3 x i32> undef, <4 x i32> <i32 0, i32 1, i32 2, i32 3>
+ %5 = tail call <4 x i32> @llvm.arm.neon.vmaxu.v4i32(<4 x i32> %3, <4 x i32> %4) nounwind readnone
+ %6 = shufflevector <4 x i32> %5, <4 x i32> undef, <3 x i32> <i32 0, i32 1, i32 2>
+ %7 = trunc <3 x i32> %6 to <3 x i8>
+ ret <3 x i8> %7
+}
+
+define <4 x i8> @_Z3maxDv4_hS_(<4 x i8> %v1, <4 x i8> %v2) nounwind readnone {
+ %1 = zext <4 x i8> %v1 to <4 x i32>
+ %2 = zext <4 x i8> %v2 to <4 x i32>
+ %3 = tail call <4 x i32> @llvm.arm.neon.vmaxu.v4i32(<4 x i32> %1, <4 x i32> %2) nounwind readnone
+ %4 = trunc <4 x i32> %3 to <4 x i8>
+ ret <4 x i8> %4
+}
+
+define zeroext i16 @_Z3maxtt(i16 zeroext %v1, i16 zeroext %v2) nounwind readnone {
+ %1 = icmp ugt i16 %v1, %v2
+ %2 = select i1 %1, i16 %v1, i16 %v2
+ ret i16 %2
+}
+
+define <2 x i16> @_Z3maxDv2_tS_(<2 x i16> %v1, <2 x i16> %v2) nounwind readnone {
+ %1 = zext <2 x i16> %v1 to <2 x i32>
+ %2 = zext <2 x i16> %v2 to <2 x i32>
+ %3 = tail call <2 x i32> @llvm.arm.neon.vmaxu.v2i32(<2 x i32> %1, <2 x i32> %2) nounwind readnone
+ %4 = trunc <2 x i32> %3 to <2 x i16>
+ ret <2 x i16> %4
+}
+
+define <3 x i16> @_Z3maxDv3_tS_(<3 x i16> %v1, <3 x i16> %v2) nounwind readnone {
+ %1 = zext <3 x i16> %v1 to <3 x i32>
+ %2 = zext <3 x i16> %v2 to <3 x i32>
+ %3 = shufflevector <3 x i32> %1, <3 x i32> undef, <4 x i32> <i32 0, i32 1, i32 2, i32 3>
+ %4 = shufflevector <3 x i32> %2, <3 x i32> undef, <4 x i32> <i32 0, i32 1, i32 2, i32 3>
+ %5 = tail call <4 x i32> @llvm.arm.neon.vmaxu.v4i32(<4 x i32> %3, <4 x i32> %4) nounwind readnone
+ %6 = shufflevector <4 x i32> %5, <4 x i32> undef, <3 x i32> <i32 0, i32 1, i32 2>
+ %7 = trunc <3 x i32> %6 to <3 x i16>
+ ret <3 x i16> %7
+}
+
+define <4 x i16> @_Z3maxDv4_tS_(<4 x i16> %v1, <4 x i16> %v2) nounwind readnone {
+ %1 = zext <4 x i16> %v1 to <4 x i32>
+ %2 = zext <4 x i16> %v2 to <4 x i32>
+ %3 = tail call <4 x i32> @llvm.arm.neon.vmaxu.v4i32(<4 x i32> %1, <4 x i32> %2) nounwind readnone
+ %4 = trunc <4 x i32> %3 to <4 x i16>
+ ret <4 x i16> %4
+}
+
+define i32 @_Z3maxjj(i32 %v1, i32 %v2) nounwind readnone {
+ %1 = icmp ugt i32 %v1, %v2
+ %2 = select i1 %1, i32 %v1, i32 %v2
+ ret i32 %2
+}
+
+define <2 x i32> @_Z3maxDv2_jS_(<2 x i32> %v1, <2 x i32> %v2) nounwind readnone {
+ %1 = tail call <2 x i32> @llvm.arm.neon.vmaxu.v2i32(<2 x i32> %v1, <2 x i32> %v2) nounwind readnone
+ ret <2 x i32> %1
+}
+
+define <3 x i32> @_Z3maxDv3_jS_(<3 x i32> %v1, <3 x i32> %v2) nounwind readnone {
+ %1 = shufflevector <3 x i32> %v1, <3 x i32> undef, <4 x i32> <i32 0, i32 1, i32 2, i32 3>
+ %2 = shufflevector <3 x i32> %v2, <3 x i32> undef, <4 x i32> <i32 0, i32 1, i32 2, i32 3>
+ %3 = tail call <4 x i32 > @llvm.arm.neon.vmaxu.v4i32(<4 x i32> %1, <4 x i32> %2) nounwind readnone
+ %4 = shufflevector <4 x i32> %3, <4 x i32> undef, <3 x i32> <i32 0, i32 1, i32 2>
+ ret <3 x i32> %4
+}
+
+define <4 x i32> @_Z3maxDv4_jS_(<4 x i32> %v1, <4 x i32> %v2) nounwind readnone {
+ %1 = tail call <4 x i32> @llvm.arm.neon.vmaxu.v4i32(<4 x i32> %v1, <4 x i32> %v2) nounwind readnone
+ ret <4 x i32> %1
+}
+
+define i64 @_Z3maxyy(i64 %v1, i64 %v2) nounwind readnone {
+ %1 = icmp ugt i64 %v1, %v2
+ %2 = select i1 %1, i64 %v1, i64 %v2
+ ret i64 %2
+}
+
+; TODO: long vector types
+
+define float @_Z3maxff(float %v1, float %v2) nounwind readnone {
+ %1 = tail call float @_Z4fmaxff(float %v1, float %v2)
+ ret float %1
+}
+
+define <2 x float> @_Z3maxDv2_fS_(<2 x float> %v1, <2 x float> %v2) nounwind readnone {
+ %1 = tail call <2 x float> @_Z4fmaxDv2_fS_(<2 x float> %v1, <2 x float> %v2)
+ ret <2 x float> %1
+}
+
+define <2 x float> @_Z3maxDv2_ff(<2 x float> %v1, float %v2) nounwind readnone {
+ %1 = tail call <2 x float> @_Z4fmaxDv2_ff(<2 x float> %v1, float %v2)
+ ret <2 x float> %1
+}
+
+define <3 x float> @_Z3maxDv3_fS_(<3 x float> %v1, <3 x float> %v2) nounwind readnone {
+ %1 = tail call <3 x float> @_Z4fmaxDv3_fS_(<3 x float> %v1, <3 x float> %v2)
+ ret <3 x float> %1
+}
+
+define <3 x float> @_Z3maxDv3_ff(<3 x float> %v1, float %v2) nounwind readnone {
+ %1 = tail call <3 x float> @_Z4fmaxDv3_ff(<3 x float> %v1, float %v2)
+ ret <3 x float> %1
+}
+
+define <4 x float> @_Z3maxDv4_fS_(<4 x float> %v1, <4 x float> %v2) nounwind readnone {
+ %1 = tail call <4 x float> @_Z4fmaxDv4_fS_(<4 x float> %v1, <4 x float> %v2)
+ ret <4 x float> %1
+}
+
+define <4 x float> @_Z3maxDv4_ff(<4 x float> %v1, float %v2) nounwind readnone {
+ %1 = tail call <4 x float> @_Z4fmaxDv4_ff(<4 x float> %v1, float %v2)
+ ret <4 x float> %1
+}
+
+
+;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;
+;;;;;;;;; MIN ;;;;;;;;;;
+;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;
+
+define signext i8 @_Z3mincc(i8 signext %v1, i8 signext %v2) nounwind readnone {
+ %1 = icmp slt i8 %v1, %v2
+ %2 = select i1 %1, i8 %v1, i8 %v2
+ ret i8 %2
+}
+
+define <2 x i8> @_Z3minDv2_cS_(<2 x i8> %v1, <2 x i8> %v2) nounwind readnone {
+ %1 = sext <2 x i8> %v1 to <2 x i32>
+ %2 = sext <2 x i8> %v2 to <2 x i32>
+ %3 = tail call <2 x i32> @llvm.arm.neon.vmins.v2i32(<2 x i32> %1, <2 x i32> %2) nounwind readnone
+ %4 = trunc <2 x i32> %3 to <2 x i8>
+ ret <2 x i8> %4
+}
+
+define <3 x i8> @_Z3minDv3_cS_(<3 x i8> %v1, <3 x i8> %v2) nounwind readnone {
+ %1 = sext <3 x i8> %v1 to <3 x i32>
+ %2 = sext <3 x i8> %v2 to <3 x i32>
+ %3 = shufflevector <3 x i32> %1, <3 x i32> undef, <4 x i32> <i32 0, i32 1, i32 2, i32 3>
+ %4 = shufflevector <3 x i32> %2, <3 x i32> undef, <4 x i32> <i32 0, i32 1, i32 2, i32 3>
+ %5 = tail call <4 x i32> @llvm.arm.neon.vmins.v4i32(<4 x i32> %3, <4 x i32> %4) nounwind readnone
+ %6 = shufflevector <4 x i32> %5, <4 x i32> undef, <3 x i32> <i32 0, i32 1, i32 2>
+ %7 = trunc <3 x i32> %6 to <3 x i8>
+ ret <3 x i8> %7
+}
+
+define <4 x i8> @_Z3minDv4_cS_(<4 x i8> %v1, <4 x i8> %v2) nounwind readnone {
+ %1 = sext <4 x i8> %v1 to <4 x i32>
+ %2 = sext <4 x i8> %v2 to <4 x i32>
+ %3 = tail call <4 x i32> @llvm.arm.neon.vmins.v4i32(<4 x i32> %1, <4 x i32> %2) nounwind readnone
+ %4 = trunc <4 x i32> %3 to <4 x i8>
+ ret <4 x i8> %4
+}
+
+define signext i16 @_Z3minss(i16 signext %v1, i16 signext %v2) nounwind readnone {
+ %1 = icmp slt i16 %v1, %v2
+ %2 = select i1 %1, i16 %v1, i16 %v2
+ ret i16 %2
+}
+
+define <2 x i16> @_Z3minDv2_sS_(<2 x i16> %v1, <2 x i16> %v2) nounwind readnone {
+ %1 = sext <2 x i16> %v1 to <2 x i32>
+ %2 = sext <2 x i16> %v2 to <2 x i32>
+ %3 = tail call <2 x i32> @llvm.arm.neon.vmins.v2i32(<2 x i32> %1, <2 x i32> %2) nounwind readnone
+ %4 = trunc <2 x i32> %3 to <2 x i16>
+ ret <2 x i16> %4
+}
+
+define <3 x i16> @_Z3minDv3_sS_(<3 x i16> %v1, <3 x i16> %v2) nounwind readnone {
+ %1 = sext <3 x i16> %v1 to <3 x i32>
+ %2 = sext <3 x i16> %v2 to <3 x i32>
+ %3 = shufflevector <3 x i32> %1, <3 x i32> undef, <4 x i32> <i32 0, i32 1, i32 2, i32 3>
+ %4 = shufflevector <3 x i32> %2, <3 x i32> undef, <4 x i32> <i32 0, i32 1, i32 2, i32 3>
+ %5 = tail call <4 x i32> @llvm.arm.neon.vmins.v4i32(<4 x i32> %3, <4 x i32> %4) nounwind readnone
+ %6 = shufflevector <4 x i32> %5, <4 x i32> undef, <3 x i32> <i32 0, i32 1, i32 2>
+ %7 = trunc <3 x i32> %6 to <3 x i16>
+ ret <3 x i16> %7
+}
+
+define <4 x i16> @_Z3minDv4_sS_(<4 x i16> %v1, <4 x i16> %v2) nounwind readnone {
+ %1 = sext <4 x i16> %v1 to <4 x i32>
+ %2 = sext <4 x i16> %v2 to <4 x i32>
+ %3 = tail call <4 x i32> @llvm.arm.neon.vmins.v4i32(<4 x i32> %1, <4 x i32> %2) nounwind readnone
+ %4 = trunc <4 x i32> %3 to <4 x i16>
+ ret <4 x i16> %4
+}
+
+define i32 @_Z3minii(i32 %v1, i32 %v2) nounwind readnone {
+ %1 = icmp slt i32 %v1, %v2
+ %2 = select i1 %1, i32 %v1, i32 %v2
+ ret i32 %2
+}
+
+define <2 x i32> @_Z3minDv2_iS_(<2 x i32> %v1, <2 x i32> %v2) nounwind readnone {
+ %1 = tail call <2 x i32> @llvm.arm.neon.vmins.v2i32(<2 x i32> %v1, <2 x i32> %v2) nounwind readnone
+ ret <2 x i32> %1
+}
+
+define <3 x i32> @_Z3minDv3_iS_(<3 x i32> %v1, <3 x i32> %v2) nounwind readnone {
+ %1 = shufflevector <3 x i32> %v1, <3 x i32> undef, <4 x i32> <i32 0, i32 1, i32 2, i32 3>
+ %2 = shufflevector <3 x i32> %v2, <3 x i32> undef, <4 x i32> <i32 0, i32 1, i32 2, i32 3>
+ %3 = tail call <4 x i32 > @llvm.arm.neon.vmins.v4i32(<4 x i32> %1, <4 x i32> %2) nounwind readnone
+ %4 = shufflevector <4 x i32> %3, <4 x i32> undef, <3 x i32> <i32 0, i32 1, i32 2>
+ ret <3 x i32> %4
+}
+
+define <4 x i32> @_Z3minDv4_iS_(<4 x i32> %v1, <4 x i32> %v2) nounwind readnone {
+ %1 = tail call <4 x i32> @llvm.arm.neon.vmins.v4i32(<4 x i32> %v1, <4 x i32> %v2) nounwind readnone
+ ret <4 x i32> %1
+}
+
+define i64 @_Z3minxx(i64 %v1, i64 %v2) nounwind readnone {
+ %1 = icmp slt i64 %v1, %v2
+ %2 = select i1 %1, i64 %v1, i64 %v2
+ ret i64 %2
+}
+
+; TODO: long vector types
+
+define zeroext i8 @_Z3minhh(i8 zeroext %v1, i8 zeroext %v2) nounwind readnone {
+ %1 = icmp ult i8 %v1, %v2
+ %2 = select i1 %1, i8 %v1, i8 %v2
+ ret i8 %2
+}
+
+define <2 x i8> @_Z3minDv2_hS_(<2 x i8> %v1, <2 x i8> %v2) nounwind readnone {
+ %1 = zext <2 x i8> %v1 to <2 x i32>
+ %2 = zext <2 x i8> %v2 to <2 x i32>
+ %3 = tail call <2 x i32> @llvm.arm.neon.vminu.v2i32(<2 x i32> %1, <2 x i32> %2) nounwind readnone
+ %4 = trunc <2 x i32> %3 to <2 x i8>
+ ret <2 x i8> %4
+}
+
+define <3 x i8> @_Z3minDv3_hS_(<3 x i8> %v1, <3 x i8> %v2) nounwind readnone {
+ %1 = zext <3 x i8> %v1 to <3 x i32>
+ %2 = zext <3 x i8> %v2 to <3 x i32>
+ %3 = shufflevector <3 x i32> %1, <3 x i32> undef, <4 x i32> <i32 0, i32 1, i32 2, i32 3>
+ %4 = shufflevector <3 x i32> %2, <3 x i32> undef, <4 x i32> <i32 0, i32 1, i32 2, i32 3>
+ %5 = tail call <4 x i32> @llvm.arm.neon.vminu.v4i32(<4 x i32> %3, <4 x i32> %4) nounwind readnone
+ %6 = shufflevector <4 x i32> %5, <4 x i32> undef, <3 x i32> <i32 0, i32 1, i32 2>
+ %7 = trunc <3 x i32> %6 to <3 x i8>
+ ret <3 x i8> %7
+}
+
+define <4 x i8> @_Z3minDv4_hS_(<4 x i8> %v1, <4 x i8> %v2) nounwind readnone {
+ %1 = zext <4 x i8> %v1 to <4 x i32>
+ %2 = zext <4 x i8> %v2 to <4 x i32>
+ %3 = tail call <4 x i32> @llvm.arm.neon.vminu.v4i32(<4 x i32> %1, <4 x i32> %2) nounwind readnone
+ %4 = trunc <4 x i32> %3 to <4 x i8>
+ ret <4 x i8> %4
+}
+
+define zeroext i16 @_Z3mintt(i16 zeroext %v1, i16 zeroext %v2) nounwind readnone {
+ %1 = icmp ult i16 %v1, %v2
+ %2 = select i1 %1, i16 %v1, i16 %v2
+ ret i16 %2
+}
+
+define <2 x i16> @_Z3minDv2_tS_(<2 x i16> %v1, <2 x i16> %v2) nounwind readnone {
+ %1 = zext <2 x i16> %v1 to <2 x i32>
+ %2 = zext <2 x i16> %v2 to <2 x i32>
+ %3 = tail call <2 x i32> @llvm.arm.neon.vminu.v2i32(<2 x i32> %1, <2 x i32> %2) nounwind readnone
+ %4 = trunc <2 x i32> %3 to <2 x i16>
+ ret <2 x i16> %4
+}
+
+define <3 x i16> @_Z3minDv3_tS_(<3 x i16> %v1, <3 x i16> %v2) nounwind readnone {
+ %1 = zext <3 x i16> %v1 to <3 x i32>
+ %2 = zext <3 x i16> %v2 to <3 x i32>
+ %3 = shufflevector <3 x i32> %1, <3 x i32> undef, <4 x i32> <i32 0, i32 1, i32 2, i32 3>
+ %4 = shufflevector <3 x i32> %2, <3 x i32> undef, <4 x i32> <i32 0, i32 1, i32 2, i32 3>
+ %5 = tail call <4 x i32> @llvm.arm.neon.vminu.v4i32(<4 x i32> %3, <4 x i32> %4) nounwind readnone
+ %6 = shufflevector <4 x i32> %5, <4 x i32> undef, <3 x i32> <i32 0, i32 1, i32 2>
+ %7 = trunc <3 x i32> %6 to <3 x i16>
+ ret <3 x i16> %7
+}
+
+define <4 x i16> @_Z3minDv4_tS_(<4 x i16> %v1, <4 x i16> %v2) nounwind readnone {
+ %1 = zext <4 x i16> %v1 to <4 x i32>
+ %2 = zext <4 x i16> %v2 to <4 x i32>
+ %3 = tail call <4 x i32> @llvm.arm.neon.vminu.v4i32(<4 x i32> %1, <4 x i32> %2) nounwind readnone
+ %4 = trunc <4 x i32> %3 to <4 x i16>
+ ret <4 x i16> %4
+}
+
+define i32 @_Z3minjj(i32 %v1, i32 %v2) nounwind readnone {
+ %1 = icmp ult i32 %v1, %v2
+ %2 = select i1 %1, i32 %v1, i32 %v2
+ ret i32 %2
+}
+
+define <2 x i32> @_Z3minDv2_jS_(<2 x i32> %v1, <2 x i32> %v2) nounwind readnone {
+ %1 = tail call <2 x i32> @llvm.arm.neon.vminu.v2i32(<2 x i32> %v1, <2 x i32> %v2) nounwind readnone
+ ret <2 x i32> %1
+}
+
+define <3 x i32> @_Z3minDv3_jS_(<3 x i32> %v1, <3 x i32> %v2) nounwind readnone {
+ %1 = shufflevector <3 x i32> %v1, <3 x i32> undef, <4 x i32> <i32 0, i32 1, i32 2, i32 3>
+ %2 = shufflevector <3 x i32> %v2, <3 x i32> undef, <4 x i32> <i32 0, i32 1, i32 2, i32 3>
+ %3 = tail call <4 x i32 > @llvm.arm.neon.vminu.v4i32(<4 x i32> %1, <4 x i32> %2) nounwind readnone
+ %4 = shufflevector <4 x i32> %3, <4 x i32> undef, <3 x i32> <i32 0, i32 1, i32 2>
+ ret <3 x i32> %4
+}
+
+define <4 x i32> @_Z3minDv4_jS_(<4 x i32> %v1, <4 x i32> %v2) nounwind readnone {
+ %1 = tail call <4 x i32> @llvm.arm.neon.vminu.v4i32(<4 x i32> %v1, <4 x i32> %v2) nounwind readnone
+ ret <4 x i32> %1
+}
+
+define i64 @_Z3minyy(i64 %v1, i64 %v2) nounwind readnone {
+ %1 = icmp ult i64 %v1, %v2
+ %2 = select i1 %1, i64 %v1, i64 %v2
+ ret i64 %2
+}
+
+; TODO: long vector types
+
+define float @_Z3minff(float %v1, float %v2) nounwind readnone {
+ %1 = tail call float @_Z4fminff(float %v1, float %v2)
+ ret float %1
+}
+
+define <2 x float> @_Z3minDv2_fS_(<2 x float> %v1, <2 x float> %v2) nounwind readnone {
+ %1 = tail call <2 x float> @_Z4fminDv2_fS_(<2 x float> %v1, <2 x float> %v2)
+ ret <2 x float> %1
+}
+
+define <2 x float> @_Z3minDv2_ff(<2 x float> %v1, float %v2) nounwind readnone {
+ %1 = tail call <2 x float> @_Z4fminDv2_ff(<2 x float> %v1, float %v2)
+ ret <2 x float> %1
+}
+
+define <3 x float> @_Z3minDv3_fS_(<3 x float> %v1, <3 x float> %v2) nounwind readnone {
+ %1 = tail call <3 x float> @_Z4fminDv3_fS_(<3 x float> %v1, <3 x float> %v2)
+ ret <3 x float> %1
+}
+
+define <3 x float> @_Z3minDv3_ff(<3 x float> %v1, float %v2) nounwind readnone {
+ %1 = tail call <3 x float> @_Z4fminDv3_ff(<3 x float> %v1, float %v2)
+ ret <3 x float> %1
+}
+
+define <4 x float> @_Z3minDv4_fS_(<4 x float> %v1, <4 x float> %v2) nounwind readnone {
+ %1 = tail call <4 x float> @_Z4fminDv4_fS_(<4 x float> %v1, <4 x float> %v2)
+ ret <4 x float> %1
+}
+
+define <4 x float> @_Z3minDv4_ff(<4 x float> %v1, float %v2) nounwind readnone {
+ %1 = tail call <4 x float> @_Z4fminDv4_ff(<4 x float> %v1, float %v2)
+ ret <4 x float> %1
+}
+
+
+;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;
+;;;;;;;;; YUV ;;;;;;;;;;
+;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;
+
+@yuv_U = internal constant <4 x i32> <i32 0, i32 -100, i32 516, i32 0>, align 16
+@yuv_V = internal constant <4 x i32> <i32 409, i32 -208, i32 0, i32 0>, align 16
+@yuv_0 = internal constant <4 x i32> <i32 0, i32 0, i32 0, i32 0>, align 16
+@yuv_255 = internal constant <4 x i32> <i32 65535, i32 65535, i32 65535, i32 65535>, align 16
+
+
+define <4 x i8> @_Z18rsYuvToRGBA_uchar4hhh(i8 %pY, i8 %pU, i8 %pV) nounwind readnone alwaysinline {
+ %_sy = zext i8 %pY to i32
+ %_su = zext i8 %pU to i32
+ %_sv = zext i8 %pV to i32
+
+ %_sy2 = add i32 -16, %_sy
+ %_sy3 = mul i32 298, %_sy2
+ %_su2 = add i32 -128, %_su
+ %_sv2 = add i32 -128, %_sv
+ %_y = tail call <4 x i32> @smear_4i32(i32 %_sy3) nounwind readnone
+ %_u = tail call <4 x i32> @smear_4i32(i32 %_su2) nounwind readnone
+ %_v = tail call <4 x i32> @smear_4i32(i32 %_sv2) nounwind readnone
+
+ %mu = load <4 x i32>* @yuv_U, align 8
+ %mv = load <4 x i32>* @yuv_V, align 8
+ %_u2 = mul <4 x i32> %_u, %mu
+ %_v2 = mul <4 x i32> %_v, %mv
+ %_y2 = add <4 x i32> %_y, %_u2
+ %_y3 = add <4 x i32> %_y2, %_v2
+
+ ; %r1 = tail call <4 x i16> @llvm.arm.neon.vqshiftnsu.v4i16(<4 x i32> %_y3, <4 x i32> <i32 8, i32 8, i32 8, i32 8>) nounwind readnone
+; %r2 = trunc <4 x i16> %r1 to <4 x i8>
+; ret <4 x i8> %r2
+
+ %c0 = load <4 x i32>* @yuv_0, align 8
+ %c255 = load <4 x i32>* @yuv_255, align 8
+ %r1 = tail call <4 x i32> @llvm.arm.neon.vmaxs.v4i32(<4 x i32> %_y3, <4 x i32> %c0) nounwind readnone
+ %r2 = tail call <4 x i32> @llvm.arm.neon.vmins.v4i32(<4 x i32> %r1, <4 x i32> %c255) nounwind readnone
+ %r3 = lshr <4 x i32> %r2, <i32 8, i32 8, i32 8, i32 8>
+ %r4 = trunc <4 x i32> %r3 to <4 x i8>
+ ret <4 x i8> %r4
+}
+
+;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;
+;;;;;;;;; half_RECIP ;;;;;;;;;;
+;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;
+
+define float @_Z10half_recipf(float %v) {
+ %1 = insertelement <2 x float> undef, float %v, i32 0
+ %2 = tail call <2 x float> @llvm.arm.neon.vrecpe.v2f32(<2 x float> %1) nounwind readnone
+ %3 = extractelement <2 x float> %2, i32 0
+ ret float %3
+}
+
+define <2 x float> @_Z10half_recip2Dv2_h(<2 x float> %v) nounwind readnone {
+ %1 = tail call <2 x float> @llvm.arm.neon.vrecpe.v2f32(<2 x float> %v) nounwind readnone
+ ret <2 x float> %1
+}
+
+define <3 x float> @_Z10half_recip3Dv3_h(<3 x float> %v) nounwind readnone {
+ %1 = shufflevector <3 x float> %v, <3 x float> undef, <4 x i32> <i32 0, i32 1, i32 2, i32 3>
+ %2 = tail call <4 x float> @llvm.arm.neon.vrecpe.v4f32(<4 x float> %1) nounwind readnone
+ %3 = shufflevector <4 x float> %2, <4 x float> undef, <3 x i32> <i32 0, i32 1, i32 2>
+ ret <3 x float> %3
+}
+
+define <4 x float> @_Z10half_recip4Dv4_h(<4 x float> %v) nounwind readnone {
+ %1 = tail call <4 x float> @llvm.arm.neon.vrecpe.v4f32(<4 x float> %v) nounwind readnone
+ ret <4 x float> %1
+}
+
+;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;
+;;;;;;;;; half_SQRT ;;;;;;;;;;
+;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;
+
+define float @_Z9half_sqrtf(float %v) {
+ %1 = insertelement <2 x float> undef, float %v, i32 0
+ %2 = tail call <2 x float> @llvm.arm.neon.vrsqrte.v2f32(<2 x float> %1) nounwind readnone
+ %3 = tail call <2 x float> @llvm.arm.neon.vrecpe.v2f32(<2 x float> %2) nounwind readnone
+ %4 = extractelement <2 x float> %3, i32 0
+ ret float %4
+}
+
+define <2 x float> @_Z9half_sqrt2Dv2_h(<2 x float> %v) nounwind readnone {
+ %1 = tail call <2 x float> @llvm.arm.neon.vrsqrte.v2f32(<2 x float> %v) nounwind readnone
+ %2 = tail call <2 x float> @llvm.arm.neon.vrecpe.v2f32(<2 x float> %1) nounwind readnone
+ ret <2 x float> %2
+}
+
+define <3 x float> @_Z9half_sqrt3Dv3_h(<3 x float> %v) nounwind readnone {
+ %1 = shufflevector <3 x float> %v, <3 x float> undef, <4 x i32> <i32 0, i32 1, i32 2, i32 3>
+ %2 = tail call <4 x float> @llvm.arm.neon.vrsqrte.v4f32(<4 x float> %1) nounwind readnone
+ %3 = tail call <4 x float> @llvm.arm.neon.vrecpe.v4f32(<4 x float> %2) nounwind readnone
+ %4 = shufflevector <4 x float> %3, <4 x float> undef, <3 x i32> <i32 0, i32 1, i32 2>
+ ret <3 x float> %4
+}
+
+define <4 x float> @_Z9half_sqrt4Dv4_h(<4 x float> %v) nounwind readnone {
+ %1 = tail call <4 x float> @llvm.arm.neon.vrsqrte.v4f32(<4 x float> %v) nounwind readnone
+ %2 = tail call <4 x float> @llvm.arm.neon.vrecpe.v4f32(<4 x float> %1) nounwind readnone
+ ret <4 x float> %2
+}
+
+
+;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;
+;;;;;;;;; half_RSQRT ;;;;;;;;;;
+;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;
+
+define float @_Z10half_rsqrtf(float %v) {
+ %1 = insertelement <2 x float> undef, float %v, i32 0
+ %2 = tail call <2 x float> @llvm.arm.neon.vrsqrte.v2f32(<2 x float> %1) nounwind readnone
+ %3 = extractelement <2 x float> %2, i32 0
+ ret float %3
+}
+
+define <2 x float> @_Z10half_rsqrt2Dv2_h(<2 x float> %v) nounwind readnone {
+ %1 = tail call <2 x float> @llvm.arm.neon.vrsqrte.v2f32(<2 x float> %v) nounwind readnone
+ ret <2 x float> %1
+}
+
+define <3 x float> @_Z10half_rsqrt3Dv3_h(<3 x float> %v) nounwind readnone {
+ %1 = shufflevector <3 x float> %v, <3 x float> undef, <4 x i32> <i32 0, i32 1, i32 2, i32 3>
+ %2 = tail call <4 x float> @llvm.arm.neon.vrsqrte.v4f32(<4 x float> %1) nounwind readnone
+ %3 = shufflevector <4 x float> %2, <4 x float> undef, <3 x i32> <i32 0, i32 1, i32 2>
+ ret <3 x float> %3
+}
+
+define <4 x float> @_Z10half_rsqrt4Dv4_h(<4 x float> %v) nounwind readnone {
+ %1 = tail call <4 x float> @llvm.arm.neon.vrsqrte.v4f32(<4 x float> %v) nounwind readnone
+ ret <4 x float> %1
+}
+
+;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;
+;;;;;;;;; matrix ;;;;;;;;;;
+;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;
+
+declare <4 x float> @llvm.arm.neon.vld1.v4f32(i8*, i32) nounwind readonly
+
+%struct.rs_matrix4x4 = type { [16 x float] }
+%struct.rs_matrix3x3 = type { [9 x float] }
+%struct.rs_matrix2x2 = type { [4 x float] }
+
+define internal <4 x float> @smear_f(float %in) nounwind readnone alwaysinline {
+ %1 = insertelement <4 x float> undef, float %in, i32 0
+ %2 = insertelement <4 x float> %1, float %in, i32 1
+ %3 = insertelement <4 x float> %2, float %in, i32 2
+ %4 = insertelement <4 x float> %3, float %in, i32 3
+ ret <4 x float> %4
+}
+
+
+define <3 x float> @_Z16rsMatrixMultiplyPK12rs_matrix3x3Dv3_f(%struct.rs_matrix3x3* nocapture %m, <3 x float> %in) nounwind readonly {
+ %x0 = extractelement <3 x float> %in, i32 0
+ %x = tail call <4 x float> @smear_f(float %x0) nounwind readnone
+ %y0 = extractelement <3 x float> %in, i32 1
+ %y = tail call <4 x float> @smear_f(float %y0) nounwind readnone
+ %z0 = extractelement <3 x float> %in, i32 2
+ %z = tail call <4 x float> @smear_f(float %z0) nounwind readnone
+
+ %px = getelementptr inbounds %struct.rs_matrix3x3* %m, i32 0, i32 0, i32 0
+ %px2 = bitcast float* %px to i8*
+ %xm = call <4 x float> @llvm.arm.neon.vld1.v4f32(i8* %px2, i32 4) nounwind
+
+ %py = getelementptr inbounds %struct.rs_matrix3x3* %m, i32 0, i32 0, i32 3
+ %py2 = bitcast float* %py to i8*
+ %ym = call <4 x float> @llvm.arm.neon.vld1.v4f32(i8* %py2, i32 4) nounwind
+
+ %pz = getelementptr inbounds %struct.rs_matrix3x3* %m, i32 0, i32 0, i32 5
+ %pz2 = bitcast float* %pz to i8*
+ %zm2 = call <4 x float> @llvm.arm.neon.vld1.v4f32(i8* %pz2, i32 4) nounwind
+ %zm = shufflevector <4 x float> %zm2, <4 x float> undef, <4 x i32> <i32 1, i32 2, i32 3, i32 4>
+
+ %a1 = fmul <4 x float> %x, %xm
+ %a2 = fmul <4 x float> %y, %ym
+ %a3 = fadd <4 x float> %a1, %a2
+ %a4 = fmul <4 x float> %z, %zm
+ %a5 = fadd <4 x float> %a4, %a3
+ %a6 = shufflevector <4 x float> %a5, <4 x float> undef, <3 x i32> <i32 0, i32 1, i32 2>
+ ret <3 x float> %a6
+}
+
+define <3 x float> @_Z16rsMatrixMultiplyPK12rs_matrix3x3Dv2_f(%struct.rs_matrix3x3* nocapture %m, <2 x float> %in) nounwind readonly {
+ %x0 = extractelement <2 x float> %in, i32 0
+ %x = tail call <4 x float> @smear_f(float %x0) nounwind readnone
+ %y0 = extractelement <2 x float> %in, i32 1
+ %y = tail call <4 x float> @smear_f(float %y0) nounwind readnone
+
+ %px = getelementptr inbounds %struct.rs_matrix3x3* %m, i32 0, i32 0, i32 0
+ %px2 = bitcast float* %px to <4 x float>*
+ %xm = load <4 x float>* %px2, align 4
+ %py = getelementptr inbounds %struct.rs_matrix3x3* %m, i32 0, i32 0, i32 3
+ %py2 = bitcast float* %py to <4 x float>*
+ %ym = load <4 x float>* %py2, align 4
+
+ %a1 = fmul <4 x float> %x, %xm
+ %a2 = fmul <4 x float> %y, %ym
+ %a3 = fadd <4 x float> %a1, %a2
+ %a4 = shufflevector <4 x float> %a3, <4 x float> undef, <3 x i32> <i32 0, i32 1, i32 2>
+ ret <3 x float> %a4
+}
+
+define <4 x float> @_Z16rsMatrixMultiplyPK12rs_matrix4x4Dv4_f(%struct.rs_matrix4x4* nocapture %m, <4 x float> %in) nounwind readonly {
+ %x0 = extractelement <4 x float> %in, i32 0
+ %x = tail call <4 x float> @smear_f(float %x0) nounwind readnone
+ %y0 = extractelement <4 x float> %in, i32 1
+ %y = tail call <4 x float> @smear_f(float %y0) nounwind readnone
+ %z0 = extractelement <4 x float> %in, i32 2
+ %z = tail call <4 x float> @smear_f(float %z0) nounwind readnone
+ %w0 = extractelement <4 x float> %in, i32 3
+ %w = tail call <4 x float> @smear_f(float %w0) nounwind readnone
+
+ %px = getelementptr inbounds %struct.rs_matrix4x4* %m, i32 0, i32 0, i32 0
+ %px2 = bitcast float* %px to <4 x float>*
+ %xm = load <4 x float>* %px2, align 4
+ %py = getelementptr inbounds %struct.rs_matrix4x4* %m, i32 0, i32 0, i32 4
+ %py2 = bitcast float* %py to <4 x float>*
+ %ym = load <4 x float>* %py2, align 4
+ %pz = getelementptr inbounds %struct.rs_matrix4x4* %m, i32 0, i32 0, i32 8
+ %pz2 = bitcast float* %pz to <4 x float>*
+ %zm = load <4 x float>* %pz2, align 4
+ %pw = getelementptr inbounds %struct.rs_matrix4x4* %m, i32 0, i32 0, i32 12
+ %pw2 = bitcast float* %pw to <4 x float>*
+ %wm = load <4 x float>* %pw2, align 4
+
+ %a1 = fmul <4 x float> %x, %xm
+ %a2 = fmul <4 x float> %y, %ym
+ %a3 = fadd <4 x float> %a1, %a2
+ %a4 = fmul <4 x float> %z, %zm
+ %a5 = fadd <4 x float> %a3, %a4
+ %a6 = fmul <4 x float> %w, %wm
+ %a7 = fadd <4 x float> %a5, %a6
+ ret <4 x float> %a7
+}
+
+define <4 x float> @_Z16rsMatrixMultiplyPK12rs_matrix4x4Dv3_f(%struct.rs_matrix4x4* nocapture %m, <3 x float> %in) nounwind readonly {
+ %x0 = extractelement <3 x float> %in, i32 0
+ %x = tail call <4 x float> @smear_f(float %x0) nounwind readnone
+ %y0 = extractelement <3 x float> %in, i32 1
+ %y = tail call <4 x float> @smear_f(float %y0) nounwind readnone
+ %z0 = extractelement <3 x float> %in, i32 2
+ %z = tail call <4 x float> @smear_f(float %z0) nounwind readnone
+
+ %px = getelementptr inbounds %struct.rs_matrix4x4* %m, i32 0, i32 0, i32 0
+ %px2 = bitcast float* %px to <4 x float>*
+ %xm = load <4 x float>* %px2, align 4
+ %py = getelementptr inbounds %struct.rs_matrix4x4* %m, i32 0, i32 0, i32 4
+ %py2 = bitcast float* %py to <4 x float>*
+ %ym = load <4 x float>* %py2, align 4
+ %pz = getelementptr inbounds %struct.rs_matrix4x4* %m, i32 0, i32 0, i32 8
+ %pz2 = bitcast float* %pz to <4 x float>*
+ %zm = load <4 x float>* %pz2, align 4
+ %pw = getelementptr inbounds %struct.rs_matrix4x4* %m, i32 0, i32 0, i32 12
+ %pw2 = bitcast float* %pw to <4 x float>*
+ %wm = load <4 x float>* %pw2, align 4
+
+ %a1 = fmul <4 x float> %x, %xm
+ %a2 = fadd <4 x float> %wm, %a1
+ %a3 = fmul <4 x float> %y, %ym
+ %a4 = fadd <4 x float> %a2, %a3
+ %a5 = fmul <4 x float> %z, %zm
+ %a6 = fadd <4 x float> %a4, %a5
+ ret <4 x float> %a6
+}
+
+define <4 x float> @_Z16rsMatrixMultiplyPK12rs_matrix4x4Dv2_f(%struct.rs_matrix4x4* nocapture %m, <2 x float> %in) nounwind readonly {
+ %x0 = extractelement <2 x float> %in, i32 0
+ %x = tail call <4 x float> @smear_f(float %x0) nounwind readnone
+ %y0 = extractelement <2 x float> %in, i32 1
+ %y = tail call <4 x float> @smear_f(float %y0) nounwind readnone
+
+ %px = getelementptr inbounds %struct.rs_matrix4x4* %m, i32 0, i32 0, i32 0
+ %px2 = bitcast float* %px to <4 x float>*
+ %xm = load <4 x float>* %px2, align 4
+ %py = getelementptr inbounds %struct.rs_matrix4x4* %m, i32 0, i32 0, i32 4
+ %py2 = bitcast float* %py to <4 x float>*
+ %ym = load <4 x float>* %py2, align 4
+ %pw = getelementptr inbounds %struct.rs_matrix4x4* %m, i32 0, i32 0, i32 12
+ %pw2 = bitcast float* %pw to <4 x float>*
+ %wm = load <4 x float>* %pw2, align 4
+
+ %a1 = fmul <4 x float> %x, %xm
+ %a2 = fadd <4 x float> %wm, %a1
+ %a3 = fmul <4 x float> %y, %ym
+ %a4 = fadd <4 x float> %a2, %a3
+ ret <4 x float> %a4
+}
+
+
+
+;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;
+;;;;;;;;; pixel ops ;;;;;;;;;;
+;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;
+
+
+@fc_255.0 = internal constant <4 x float> <float 255.0, float 255.0, float 255.0, float 255.0>, align 16
+@fc_0.5 = internal constant <4 x float> <float 0.5, float 0.5, float 0.5, float 0.5>, align 16
+@fc_0 = internal constant <4 x float> <float 0.0, float 0.0, float 0.0, float 0.0>, align 16
+
+declare <4 x i8> @_Z14convert_uchar4Dv4_f(<4 x float> %in) nounwind readnone
+declare <4 x float> @_Z14convert_float4Dv4_h(<4 x i8> %in) nounwind readnone
+
+; uchar4 __attribute__((overloadable)) rsPackColorTo8888(float4 color)
+define <4 x i8> @_Z17rsPackColorTo8888Dv4_f(<4 x float> %color) nounwind readnone {
+ %f255 = load <4 x float>* @fc_255.0, align 16
+ %f05 = load <4 x float>* @fc_0.5, align 16
+ %f0 = load <4 x float>* @fc_0, align 16
+ %v1 = fmul <4 x float> %f255, %color
+ %v2 = fadd <4 x float> %f05, %v1
+ %v3 = tail call <4 x float> @_Z5clampDv4_fS_S_(<4 x float> %v2, <4 x float> %f0, <4 x float> %f255) nounwind readnone
+ %v4 = tail call <4 x i8> @_Z14convert_uchar4Dv4_f(<4 x float> %v3) nounwind readnone
+ ret <4 x i8> %v4
+}
+
+; uchar4 __attribute__((overloadable)) rsPackColorTo8888(float3 color)
+define <4 x i8> @_Z17rsPackColorTo8888Dv3_f(<3 x float> %color) nounwind readnone {
+ %1 = shufflevector <3 x float> %color, <3 x float> undef, <4 x i32> <i32 0, i32 1, i32 2, i32 3>
+ %2 = insertelement <4 x float> %1, float 1.0, i32 3
+ %3 = tail call <4 x i8> @_Z17rsPackColorTo8888Dv4_f(<4 x float> %2) nounwind readnone
+ ret <4 x i8> %3
+}
+
+; uchar4 __attribute__((overloadable)) rsPackColorTo8888(float r, float g, float b)
+define <4 x i8> @_Z17rsPackColorTo8888fff(float %r, float %g, float %b) nounwind readnone {
+ %1 = insertelement <4 x float> undef, float %r, i32 0
+ %2 = insertelement <4 x float> %1, float %g, i32 1
+ %3 = insertelement <4 x float> %2, float %b, i32 2
+ %4 = insertelement <4 x float> %3, float 1.0, i32 3
+ %5 = tail call <4 x i8> @_Z17rsPackColorTo8888Dv4_f(<4 x float> %4) nounwind readnone
+ ret <4 x i8> %5
+}
+
+; uchar4 __attribute__((overloadable)) rsPackColorTo8888(float r, float g, float b, float a)
+define <4 x i8> @_Z17rsPackColorTo8888ffff(float %r, float %g, float %b, float %a) nounwind readnone {
+ %1 = insertelement <4 x float> undef, float %r, i32 0
+ %2 = insertelement <4 x float> %1, float %g, i32 1
+ %3 = insertelement <4 x float> %2, float %b, i32 2
+ %4 = insertelement <4 x float> %3, float %a, i32 3
+ %5 = tail call <4 x i8> @_Z17rsPackColorTo8888Dv4_f(<4 x float> %4) nounwind readnone
+ ret <4 x i8> %5
+}
+
diff --git a/driver/runtime/arch/sqrt.c b/driver/runtime/arch/sqrt.c
new file mode 100755
index 0000000..f1dac5f
--- /dev/null
+++ b/driver/runtime/arch/sqrt.c
@@ -0,0 +1,45 @@
+/*
+ * Copyright (C) 2012 The Android Open Source Project
+ *
+ * Licensed under the Apache License, Version 2.0 (the "License");
+ * you may not use this file except in compliance with the License.
+ * You may obtain a copy of the License at
+ *
+ * http://www.apache.org/licenses/LICENSE-2.0
+ *
+ * Unless required by applicable law or agreed to in writing, software
+ * distributed under the License is distributed on an "AS IS" BASIS,
+ * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ * See the License for the specific language governing permissions and
+ * limitations under the License.
+ */
+
+
+#include "rs_types.rsh"
+
+#define FN_FUNC_FN(fnc) \
+extern float2 __attribute__((overloadable)) fnc(float2 v) { \
+ float2 r; \
+ r.x = fnc(v.x); \
+ r.y = fnc(v.y); \
+ return r; \
+} \
+extern float3 __attribute__((overloadable)) fnc(float3 v) { \
+ float3 r; \
+ r.x = fnc(v.x); \
+ r.y = fnc(v.y); \
+ r.z = fnc(v.z); \
+ return r; \
+} \
+extern float4 __attribute__((overloadable)) fnc(float4 v) { \
+ float4 r; \
+ r.x = fnc(v.x); \
+ r.y = fnc(v.y); \
+ r.z = fnc(v.z); \
+ r.w = fnc(v.w); \
+ return r; \
+}
+
+extern float __attribute__((overloadable)) sqrt(float);
+
+FN_FUNC_FN(sqrt)
diff --git a/driver/runtime/arch/x86_clamp.ll b/driver/runtime/arch/x86_clamp.ll
new file mode 100755
index 0000000..422e9f6
--- /dev/null
+++ b/driver/runtime/arch/x86_clamp.ll
@@ -0,0 +1,74 @@
+target datalayout = "e-p:32:32:32-i1:8:8-i8:8:8-i16:16:16-i32:32:32-i64:32:64-f32:32:32-f64:32:64-v64:64:64-v128:128:128-a0:0:64-f80:32:32-n8:16:32-S128"
+target triple = "i386-unknown-linux-gnu"
+
+declare <4 x float> @llvm.x86.sse.min.ps(<4 x float>, <4 x float>)
+declare <4 x float> @llvm.x86.sse.max.ps(<4 x float>, <4 x float>)
+declare <4 x float> @llvm.x86.sse.min.ss(<4 x float>, <4 x float>)
+declare <4 x float> @llvm.x86.sse.max.ss(<4 x float>, <4 x float>)
+
+define <4 x float> @_Z5clampDv4_fS_S_(<4 x float> %in, <4 x float> %low, <4 x float> %high) nounwind readnone alwaysinline {
+ %1 = tail call <4 x float> @llvm.x86.sse.min.ps(<4 x float> %in, <4 x float> %high) nounwind readnone
+ %2 = tail call <4 x float> @llvm.x86.sse.max.ps(<4 x float> %1, <4 x float> %low) nounwind readnone
+ ret <4 x float> %2
+}
+
+define <3 x float> @_Z5clampDv3_fS_S_(<3 x float> %in, <3 x float> %low, <3 x float> %high) nounwind readnone alwaysinline {
+ %1 = shufflevector <3 x float> %in, <3 x float> undef, <4 x i32> <i32 0, i32 1, i32 2, i32 3>
+ %2 = shufflevector <3 x float> %low, <3 x float> undef, <4 x i32> <i32 0, i32 1, i32 2, i32 3>
+ %3 = shufflevector <3 x float> %high, <3 x float> undef, <4 x i32> <i32 0, i32 1, i32 2, i32 3>
+ %4 = tail call <4 x float> @_Z5clampDv4_fS_S_(<4 x float> %1, <4 x float> %2, <4 x float> %3) nounwind readnone
+ %5 = shufflevector <4 x float> %4, <4 x float> undef, <3 x i32> <i32 0, i32 1, i32 2>
+ ret <3 x float> %5
+}
+
+define <2 x float> @_Z5clampDv2_fS_S_(<2 x float> %in, <2 x float> %low, <2 x float> %high) nounwind readnone alwaysinline {
+ %1 = shufflevector <2 x float> %in, <2 x float> undef, <4 x i32> <i32 0, i32 1, i32 2, i32 3>
+ %2 = shufflevector <2 x float> %low, <2 x float> undef, <4 x i32> <i32 0, i32 1, i32 2, i32 3>
+ %3 = shufflevector <2 x float> %high, <2 x float> undef, <4 x i32> <i32 0, i32 1, i32 2, i32 3>
+ %4 = tail call <4 x float> @_Z5clampDv4_fS_S_(<4 x float> %1, <4 x float> %2, <4 x float> %3) nounwind readnone
+ %5 = shufflevector <4 x float> %4, <4 x float> undef, <2 x i32> <i32 0, i32 1>
+ ret <2 x float> %5
+}
+
+define float @_Z5clampfff(float %in, float %low, float %high) nounwind readnone alwaysinline {
+ %1 = insertelement <4 x float> undef, float %in, i32 0
+ %2 = insertelement <4 x float> undef, float %low, i32 0
+ %3 = insertelement <4 x float> undef, float %high, i32 0
+ %4 = tail call <4 x float> @llvm.x86.sse.min.ss(<4 x float> %1, <4 x float> %3) nounwind readnone
+ %5 = tail call <4 x float> @llvm.x86.sse.max.ss(<4 x float> %4, <4 x float> %2) nounwind readnone
+ %6 = extractelement <4 x float> %5, i32 0
+ ret float %6
+}
+
+define <4 x float> @_Z5clampDv4_fff(<4 x float> %in, float %low, float %high) nounwind readonly {
+ %1 = insertelement <4 x float> undef, float %low, i32 0
+ %2 = insertelement <4 x float> %1, float %low, i32 1
+ %3 = insertelement <4 x float> %2, float %low, i32 2
+ %4 = insertelement <4 x float> %3, float %low, i32 3
+ %5 = insertelement <4 x float> undef, float %high, i32 0
+ %6 = insertelement <4 x float> %5, float %high, i32 1
+ %7 = insertelement <4 x float> %6, float %high, i32 2
+ %8 = insertelement <4 x float> %7, float %high, i32 3
+ %9 = tail call <4 x float> @_Z5clampDv4_fS_S_(<4 x float> %in, <4 x float> %4, <4 x float> %8) nounwind readnone
+ ret <4 x float> %9
+}
+
+define <3 x float> @_Z5clampDv3_fff(<3 x float> %in, float %low, float %high) nounwind readonly {
+ %1 = insertelement <3 x float> undef, float %low, i32 0
+ %2 = insertelement <3 x float> %1, float %low, i32 1
+ %3 = insertelement <3 x float> %2, float %low, i32 2
+ %4 = insertelement <3 x float> undef, float %high, i32 0
+ %5 = insertelement <3 x float> %4, float %high, i32 1
+ %6 = insertelement <3 x float> %5, float %high, i32 2
+ %7 = tail call <3 x float> @_Z5clampDv3_fS_S_(<3 x float> %in, <3 x float> %3, <3 x float> %6) nounwind readnone
+ ret <3 x float> %7
+}
+
+define <2 x float> @_Z5clampDv2_fff(<2 x float> %in, float %low, float %high) nounwind readonly {
+ %1 = insertelement <2 x float> undef, float %low, i32 0
+ %2 = insertelement <2 x float> %1, float %low, i32 1
+ %3 = insertelement <2 x float> undef, float %high, i32 0
+ %4 = insertelement <2 x float> %3, float %high, i32 1
+ %5 = tail call <2 x float> @_Z5clampDv2_fS_S_(<2 x float> %in, <2 x float> %2, <2 x float> %4) nounwind readnone
+ ret <2 x float> %5
+}
diff --git a/driver/runtime/arch/x86_dot_length.ll b/driver/runtime/arch/x86_dot_length.ll
new file mode 100644
index 0000000..21f2f3e
--- /dev/null
+++ b/driver/runtime/arch/x86_dot_length.ll
@@ -0,0 +1,75 @@
+target datalayout = "e-p:32:32:32-i1:8:8-i8:8:8-i16:16:16-i32:32:32-i64:32:64-f32:32:32-f64:32:64-v64:64:64-v128:128:128-a0:0:64-f80:32:32-n8:16:32-S128"
+target triple = "i386-unknown-linux-gnu"
+
+declare <2 x i64> @llvm.x86.sse2.psll.dq(<2 x i64>, i32) nounwind readnone
+declare <4 x float> @llvm.x86.sse3.hadd.ps(<4 x float>, <4 x float>) nounwind readnone
+declare float @llvm.sqrt.f32(float) nounwind readnone
+
+define float @_Z3dotDv4_fS_(<4 x float> %lhs, <4 x float> %rhs) nounwind readnone {
+ %1 = fmul <4 x float> %lhs, %rhs
+ %2 = tail call <4 x float> @llvm.x86.sse3.hadd.ps(<4 x float> %1, <4 x float> %1) nounwind readnone
+ %3 = tail call <4 x float> @llvm.x86.sse3.hadd.ps(<4 x float> %2, <4 x float> %2) nounwind readnone
+ %4 = extractelement <4 x float> %3, i32 0
+ ret float %4
+}
+
+define float @_Z3dotDv3_fS_(<3 x float> %lhs, <3 x float> %rhs) nounwind readnone {
+ %1 = fmul <3 x float> %lhs, %rhs
+ %2 = shufflevector <3 x float> %1, <3 x float> undef, <4 x i32> <i32 0, i32 1, i32 2, i32 3>
+ %3 = bitcast <4 x float> %2 to <2 x i64>
+ %4 = tail call <2 x i64> @llvm.x86.sse2.psll.dq(<2 x i64> %3, i32 32)
+ %5 = bitcast <2 x i64> %4 to <4 x float>
+ %6 = tail call <4 x float> @llvm.x86.sse3.hadd.ps(<4 x float> %5, <4 x float> %5) nounwind readnone
+ %7 = tail call <4 x float> @llvm.x86.sse3.hadd.ps(<4 x float> %6, <4 x float> %6) nounwind readnone
+ %8 = extractelement <4 x float> %7, i32 0
+ ret float %8
+}
+
+define float @_Z3dotDv2_fS_(<2 x float> %lhs, <2 x float> %rhs) nounwind readnone {
+ %1 = fmul <2 x float> %lhs, %rhs
+ %2 = shufflevector <2 x float> %1, <2 x float> undef, <4 x i32> <i32 0, i32 1, i32 2, i32 3>
+ %3 = tail call <4 x float> @llvm.x86.sse3.hadd.ps(<4 x float> %2, <4 x float> %2) nounwind readnone
+ %4 = extractelement <4 x float> %3, i32 0
+ ret float %4
+}
+
+define float @_Z3dotff(float %lhs, float %rhs) nounwind readnone {
+ %1 = fmul float %lhs, %rhs
+ ret float %1
+}
+
+define float @_Z6lengthDv4_f(<4 x float> %in) nounwind readnone alwaysinline {
+ %1 = fmul <4 x float> %in, %in
+ %2 = tail call <4 x float> @llvm.x86.sse3.hadd.ps(<4 x float> %1, <4 x float> %1) nounwind readnone
+ %3 = tail call <4 x float> @llvm.x86.sse3.hadd.ps(<4 x float> %2, <4 x float> %2) nounwind readnone
+ %4 = extractelement <4 x float> %3, i32 0
+ %5 = tail call float @llvm.sqrt.f32(float %4) nounwind readnone
+ ret float %5
+}
+
+define float @_Z6lengthDv3_f(<3 x float> %in) nounwind readnone alwaysinline {
+ %1 = fmul <3 x float> %in, %in
+ %2 = shufflevector <3 x float> %1, <3 x float> undef, <4 x i32> <i32 0, i32 1, i32 2, i32 3>
+ %3 = bitcast <4 x float> %2 to <2 x i64>
+ %4 = tail call <2 x i64> @llvm.x86.sse2.psll.dq(<2 x i64> %3, i32 32)
+ %5 = bitcast <2 x i64> %4 to <4 x float>
+ %6 = tail call <4 x float> @llvm.x86.sse3.hadd.ps(<4 x float> %5, <4 x float> %5) nounwind readnone
+ %7 = tail call <4 x float> @llvm.x86.sse3.hadd.ps(<4 x float> %6, <4 x float> %6) nounwind readnone
+ %8 = extractelement <4 x float> %7, i32 0
+ %9 = tail call float @llvm.sqrt.f32(float %8) nounwind readnone
+ ret float %9
+}
+
+define float @_Z6lengthDv2_f(<2 x float> %in) nounwind readnone alwaysinline {
+ %1 = fmul <2 x float> %in, %in
+ %2 = shufflevector <2 x float> %1, <2 x float> undef, <4 x i32> <i32 0, i32 1, i32 2, i32 3>
+ %3 = tail call <4 x float> @llvm.x86.sse3.hadd.ps(<4 x float> %2, <4 x float> %2) nounwind readnone
+ %4 = extractelement <4 x float> %3, i32 0
+ %5 = tail call float @llvm.sqrt.f32(float %4) nounwind readnone
+ ret float %5
+}
+
+define float @_Z6lengthf(float %in) nounwind readnone alwaysinline {
+ ret float %in
+}
+
diff --git a/driver/runtime/arch/x86_generic.c b/driver/runtime/arch/x86_generic.c
new file mode 100644
index 0000000..c46c54a
--- /dev/null
+++ b/driver/runtime/arch/x86_generic.c
@@ -0,0 +1,786 @@
+/*
+ * Copyright (C) 2012 The Android Open Source Project
+ *
+ * Licensed under the Apache License, Version 2.0 (the "License");
+ * you may not use this file except in compliance with the License.
+ * You may obtain a copy of the License at
+ *
+ * http://www.apache.org/licenses/LICENSE-2.0
+ *
+ * Unless required by applicable law or agreed to in writing, software
+ * distributed under the License is distributed on an "AS IS" BASIS,
+ * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ * See the License for the specific language governing permissions and
+ * limitations under the License.
+ */
+
+
+#include "rs_types.rsh"
+
+extern short __attribute__((overloadable, always_inline)) rsClamp(short amount, short low, short high);
+extern float4 __attribute__((overloadable)) clamp(float4 amount, float4 low, float4 high);
+extern uchar4 __attribute__((overloadable)) convert_uchar4(short4);
+extern float __attribute__((overloadable)) sqrt(float);
+
+/*
+ * FMAX
+ */
+
+extern float __attribute__((overloadable)) fmax(float v1, float v2) {
+ return v1 > v2 ? v1 : v2;
+}
+
+extern float2 __attribute__((overloadable)) fmax(float2 v1, float2 v2) {
+ float2 r;
+ r.x = v1.x > v2.x ? v1.x : v2.x;
+ r.y = v1.y > v2.y ? v1.y : v2.y;
+ return r;
+}
+
+extern float3 __attribute__((overloadable)) fmax(float3 v1, float3 v2) {
+ float3 r;
+ r.x = v1.x > v2.x ? v1.x : v2.x;
+ r.y = v1.y > v2.y ? v1.y : v2.y;
+ r.z = v1.z > v2.z ? v1.z : v2.z;
+ return r;
+}
+
+extern float4 __attribute__((overloadable)) fmax(float4 v1, float4 v2) {
+ float4 r;
+ r.x = v1.x > v2.x ? v1.x : v2.x;
+ r.y = v1.y > v2.y ? v1.y : v2.y;
+ r.z = v1.z > v2.z ? v1.z : v2.z;
+ r.w = v1.w > v2.w ? v1.w : v2.w;
+ return r;
+}
+
+extern float2 __attribute__((overloadable)) fmax(float2 v1, float v2) {
+ float2 r;
+ r.x = v1.x > v2 ? v1.x : v2;
+ r.y = v1.y > v2 ? v1.y : v2;
+ return r;
+}
+
+extern float3 __attribute__((overloadable)) fmax(float3 v1, float v2) {
+ float3 r;
+ r.x = v1.x > v2 ? v1.x : v2;
+ r.y = v1.y > v2 ? v1.y : v2;
+ r.z = v1.z > v2 ? v1.z : v2;
+ return r;
+}
+
+extern float4 __attribute__((overloadable)) fmax(float4 v1, float v2) {
+ float4 r;
+ r.x = v1.x > v2 ? v1.x : v2;
+ r.y = v1.y > v2 ? v1.y : v2;
+ r.z = v1.z > v2 ? v1.z : v2;
+ r.w = v1.w > v2 ? v1.w : v2;
+ return r;
+}
+
+extern float __attribute__((overloadable)) fmin(float v1, float v2) {
+ return v1 < v2 ? v1 : v2;
+}
+
+
+/*
+ * FMIN
+ */
+extern float2 __attribute__((overloadable)) fmin(float2 v1, float2 v2) {
+ float2 r;
+ r.x = v1.x < v2.x ? v1.x : v2.x;
+ r.y = v1.y < v2.y ? v1.y : v2.y;
+ return r;
+}
+
+extern float3 __attribute__((overloadable)) fmin(float3 v1, float3 v2) {
+ float3 r;
+ r.x = v1.x < v2.x ? v1.x : v2.x;
+ r.y = v1.y < v2.y ? v1.y : v2.y;
+ r.z = v1.z < v2.z ? v1.z : v2.z;
+ return r;
+}
+
+extern float4 __attribute__((overloadable)) fmin(float4 v1, float4 v2) {
+ float4 r;
+ r.x = v1.x < v2.x ? v1.x : v2.x;
+ r.y = v1.y < v2.y ? v1.y : v2.y;
+ r.z = v1.z < v2.z ? v1.z : v2.z;
+ r.w = v1.w < v2.w ? v1.w : v2.w;
+ return r;
+}
+
+extern float2 __attribute__((overloadable)) fmin(float2 v1, float v2) {
+ float2 r;
+ r.x = v1.x < v2 ? v1.x : v2;
+ r.y = v1.y < v2 ? v1.y : v2;
+ return r;
+}
+
+extern float3 __attribute__((overloadable)) fmin(float3 v1, float v2) {
+ float3 r;
+ r.x = v1.x < v2 ? v1.x : v2;
+ r.y = v1.y < v2 ? v1.y : v2;
+ r.z = v1.z < v2 ? v1.z : v2;
+ return r;
+}
+
+extern float4 __attribute__((overloadable)) fmin(float4 v1, float v2) {
+ float4 r;
+ r.x = v1.x < v2 ? v1.x : v2;
+ r.y = v1.y < v2 ? v1.y : v2;
+ r.z = v1.z < v2 ? v1.z : v2;
+ r.w = v1.w < v2 ? v1.w : v2;
+ return r;
+}
+
+
+/*
+ * MAX
+ */
+
+extern char __attribute__((overloadable)) max(char v1, char v2) {
+ return v1 > v2 ? v1 : v2;
+}
+
+extern char2 __attribute__((overloadable)) max(char2 v1, char2 v2) {
+ char2 r;
+ r.x = v1.x > v2.x ? v1.x : v2.x;
+ r.y = v1.y > v2.y ? v1.y : v2.y;
+ return r;
+}
+
+extern char3 __attribute__((overloadable)) max(char3 v1, char3 v2) {
+ char3 r;
+ r.x = v1.x > v2.x ? v1.x : v2.x;
+ r.y = v1.y > v2.y ? v1.y : v2.y;
+ r.z = v1.z > v2.z ? v1.z : v2.z;
+ return r;
+}
+
+extern char4 __attribute__((overloadable)) max(char4 v1, char4 v2) {
+ char4 r;
+ r.x = v1.x > v2.x ? v1.x : v2.x;
+ r.y = v1.y > v2.y ? v1.y : v2.y;
+ r.z = v1.z > v2.z ? v1.z : v2.z;
+ r.w = v1.w > v2.w ? v1.w : v2.w;
+ return r;
+}
+
+extern short __attribute__((overloadable)) max(short v1, short v2) {
+ return v1 > v2 ? v1 : v2;
+}
+
+extern short2 __attribute__((overloadable)) max(short2 v1, short2 v2) {
+ short2 r;
+ r.x = v1.x > v2.x ? v1.x : v2.x;
+ r.y = v1.y > v2.y ? v1.y : v2.y;
+ return r;
+}
+
+extern short3 __attribute__((overloadable)) max(short3 v1, short3 v2) {
+ short3 r;
+ r.x = v1.x > v2.x ? v1.x : v2.x;
+ r.y = v1.y > v2.y ? v1.y : v2.y;
+ r.z = v1.z > v2.z ? v1.z : v2.z;
+ return r;
+}
+
+extern short4 __attribute__((overloadable)) max(short4 v1, short4 v2) {
+ short4 r;
+ r.x = v1.x > v2.x ? v1.x : v2.x;
+ r.y = v1.y > v2.y ? v1.y : v2.y;
+ r.z = v1.z > v2.z ? v1.z : v2.z;
+ r.w = v1.w > v2.w ? v1.w : v2.w;
+ return r;
+}
+
+extern int __attribute__((overloadable)) max(int v1, int v2) {
+ return v1 > v2 ? v1 : v2;
+}
+
+extern int2 __attribute__((overloadable)) max(int2 v1, int2 v2) {
+ int2 r;
+ r.x = v1.x > v2.x ? v1.x : v2.x;
+ r.y = v1.y > v2.y ? v1.y : v2.y;
+ return r;
+}
+
+extern int3 __attribute__((overloadable)) max(int3 v1, int3 v2) {
+ int3 r;
+ r.x = v1.x > v2.x ? v1.x : v2.x;
+ r.y = v1.y > v2.y ? v1.y : v2.y;
+ r.z = v1.z > v2.z ? v1.z : v2.z;
+ return r;
+}
+
+extern int4 __attribute__((overloadable)) max(int4 v1, int4 v2) {
+ int4 r;
+ r.x = v1.x > v2.x ? v1.x : v2.x;
+ r.y = v1.y > v2.y ? v1.y : v2.y;
+ r.z = v1.z > v2.z ? v1.z : v2.z;
+ r.w = v1.w > v2.w ? v1.w : v2.w;
+ return r;
+}
+
+extern int64_t __attribute__((overloadable)) max(int64_t v1, int64_t v2) {
+ return v1 > v2 ? v1 : v2;
+}
+
+extern long2 __attribute__((overloadable)) max(long2 v1, long2 v2) {
+ long2 r;
+ r.x = v1.x > v2.x ? v1.x : v2.x;
+ r.y = v1.y > v2.y ? v1.y : v2.y;
+ return r;
+}
+
+extern long3 __attribute__((overloadable)) max(long3 v1, long3 v2) {
+ long3 r;
+ r.x = v1.x > v2.x ? v1.x : v2.x;
+ r.y = v1.y > v2.y ? v1.y : v2.y;
+ r.z = v1.z > v2.z ? v1.z : v2.z;
+ return r;
+}
+
+extern long4 __attribute__((overloadable)) max(long4 v1, long4 v2) {
+ long4 r;
+ r.x = v1.x > v2.x ? v1.x : v2.x;
+ r.y = v1.y > v2.y ? v1.y : v2.y;
+ r.z = v1.z > v2.z ? v1.z : v2.z;
+ r.w = v1.w > v2.w ? v1.w : v2.w;
+ return r;
+}
+
+extern uchar __attribute__((overloadable)) max(uchar v1, uchar v2) {
+ return v1 > v2 ? v1 : v2;
+}
+
+extern uchar2 __attribute__((overloadable)) max(uchar2 v1, uchar2 v2) {
+ uchar2 r;
+ r.x = v1.x > v2.x ? v1.x : v2.x;
+ r.y = v1.y > v2.y ? v1.y : v2.y;
+ return r;
+}
+
+extern uchar3 __attribute__((overloadable)) max(uchar3 v1, uchar3 v2) {
+ uchar3 r;
+ r.x = v1.x > v2.x ? v1.x : v2.x;
+ r.y = v1.y > v2.y ? v1.y : v2.y;
+ r.z = v1.z > v2.z ? v1.z : v2.z;
+ return r;
+}
+
+extern uchar4 __attribute__((overloadable)) max(uchar4 v1, uchar4 v2) {
+ uchar4 r;
+ r.x = v1.x > v2.x ? v1.x : v2.x;
+ r.y = v1.y > v2.y ? v1.y : v2.y;
+ r.z = v1.z > v2.z ? v1.z : v2.z;
+ r.w = v1.w > v2.w ? v1.w : v2.w;
+ return r;
+}
+
+extern ushort __attribute__((overloadable)) max(ushort v1, ushort v2) {
+ return v1 > v2 ? v1 : v2;
+}
+
+extern ushort2 __attribute__((overloadable)) max(ushort2 v1, ushort2 v2) {
+ ushort2 r;
+ r.x = v1.x > v2.x ? v1.x : v2.x;
+ r.y = v1.y > v2.y ? v1.y : v2.y;
+ return r;
+}
+
+extern ushort3 __attribute__((overloadable)) max(ushort3 v1, ushort3 v2) {
+ ushort3 r;
+ r.x = v1.x > v2.x ? v1.x : v2.x;
+ r.y = v1.y > v2.y ? v1.y : v2.y;
+ r.z = v1.z > v2.z ? v1.z : v2.z;
+ return r;
+}
+
+extern ushort4 __attribute__((overloadable)) max(ushort4 v1, ushort4 v2) {
+ ushort4 r;
+ r.x = v1.x > v2.x ? v1.x : v2.x;
+ r.y = v1.y > v2.y ? v1.y : v2.y;
+ r.z = v1.z > v2.z ? v1.z : v2.z;
+ r.w = v1.w > v2.w ? v1.w : v2.w;
+ return r;
+}
+
+extern uint __attribute__((overloadable)) max(uint v1, uint v2) {
+ return v1 > v2 ? v1 : v2;
+}
+
+extern uint2 __attribute__((overloadable)) max(uint2 v1, uint2 v2) {
+ uint2 r;
+ r.x = v1.x > v2.x ? v1.x : v2.x;
+ r.y = v1.y > v2.y ? v1.y : v2.y;
+ return r;
+}
+
+extern uint3 __attribute__((overloadable)) max(uint3 v1, uint3 v2) {
+ uint3 r;
+ r.x = v1.x > v2.x ? v1.x : v2.x;
+ r.y = v1.y > v2.y ? v1.y : v2.y;
+ r.z = v1.z > v2.z ? v1.z : v2.z;
+ return r;
+}
+
+extern uint4 __attribute__((overloadable)) max(uint4 v1, uint4 v2) {
+ uint4 r;
+ r.x = v1.x > v2.x ? v1.x : v2.x;
+ r.y = v1.y > v2.y ? v1.y : v2.y;
+ r.z = v1.z > v2.z ? v1.z : v2.z;
+ r.w = v1.w > v2.w ? v1.w : v2.w;
+ return r;
+}
+
+extern ulong __attribute__((overloadable)) max(ulong v1, ulong v2) {
+ return v1 > v2 ? v1 : v2;
+}
+
+extern ulong2 __attribute__((overloadable)) max(ulong2 v1, ulong2 v2) {
+ ulong2 r;
+ r.x = v1.x > v2.x ? v1.x : v2.x;
+ r.y = v1.y > v2.y ? v1.y : v2.y;
+ return r;
+}
+
+extern ulong3 __attribute__((overloadable)) max(ulong3 v1, ulong3 v2) {
+ ulong3 r;
+ r.x = v1.x > v2.x ? v1.x : v2.x;
+ r.y = v1.y > v2.y ? v1.y : v2.y;
+ r.z = v1.z > v2.z ? v1.z : v2.z;
+ return r;
+}
+
+extern ulong4 __attribute__((overloadable)) max(ulong4 v1, ulong4 v2) {
+ ulong4 r;
+ r.x = v1.x > v2.x ? v1.x : v2.x;
+ r.y = v1.y > v2.y ? v1.y : v2.y;
+ r.z = v1.z > v2.z ? v1.z : v2.z;
+ r.w = v1.w > v2.w ? v1.w : v2.w;
+ return r;
+}
+
+extern float __attribute__((overloadable)) max(float v1, float v2) {
+ return fmax(v1, v2);
+}
+
+extern float2 __attribute__((overloadable)) max(float2 v1, float2 v2) {
+ return fmax(v1, v2);
+}
+
+extern float2 __attribute__((overloadable)) max(float2 v1, float v2) {
+ return fmax(v1, v2);
+}
+
+extern float3 __attribute__((overloadable)) max(float3 v1, float3 v2) {
+ return fmax(v1, v2);
+}
+
+extern float3 __attribute__((overloadable)) max(float3 v1, float v2) {
+ return fmax(v1, v2);
+}
+
+extern float4 __attribute__((overloadable)) max(float4 v1, float4 v2) {
+ return fmax(v1, v2);
+}
+
+extern float4 __attribute__((overloadable)) max(float4 v1, float v2) {
+ return fmax(v1, v2);
+}
+
+
+/*
+ * MIN
+ */
+
+extern int8_t __attribute__((overloadable)) min(int8_t v1, int8_t v2) {
+ return v1 < v2 ? v1 : v2;
+}
+
+extern char2 __attribute__((overloadable)) min(char2 v1, char2 v2) {
+ char2 r;
+ r.x = v1.x < v2.x ? v1.x : v2.x;
+ r.y = v1.y < v2.y ? v1.y : v2.y;
+ return r;
+}
+
+extern char3 __attribute__((overloadable)) min(char3 v1, char3 v2) {
+ char3 r;
+ r.x = v1.x < v2.x ? v1.x : v2.x;
+ r.y = v1.y < v2.y ? v1.y : v2.y;
+ r.z = v1.z < v2.z ? v1.z : v2.z;
+ return r;
+}
+
+extern char4 __attribute__((overloadable)) min(char4 v1, char4 v2) {
+ char4 r;
+ r.x = v1.x < v2.x ? v1.x : v2.x;
+ r.y = v1.y < v2.y ? v1.y : v2.y;
+ r.z = v1.z < v2.z ? v1.z : v2.z;
+ r.w = v1.w < v2.w ? v1.w : v2.w;
+ return r;
+}
+
+extern int16_t __attribute__((overloadable)) min(int16_t v1, int16_t v2) {
+ return v1 < v2 ? v1 : v2;
+}
+
+extern short2 __attribute__((overloadable)) min(short2 v1, short2 v2) {
+ short2 r;
+ r.x = v1.x < v2.x ? v1.x : v2.x;
+ r.y = v1.y < v2.y ? v1.y : v2.y;
+ return r;
+}
+
+extern short3 __attribute__((overloadable)) min(short3 v1, short3 v2) {
+ short3 r;
+ r.x = v1.x < v2.x ? v1.x : v2.x;
+ r.y = v1.y < v2.y ? v1.y : v2.y;
+ r.z = v1.z < v2.z ? v1.z : v2.z;
+ return r;
+}
+
+extern short4 __attribute__((overloadable)) min(short4 v1, short4 v2) {
+ short4 r;
+ r.x = v1.x < v2.x ? v1.x : v2.x;
+ r.y = v1.y < v2.y ? v1.y : v2.y;
+ r.z = v1.z < v2.z ? v1.z : v2.z;
+ r.w = v1.w < v2.w ? v1.w : v2.w;
+ return r;
+}
+
+extern int32_t __attribute__((overloadable)) min(int32_t v1, int32_t v2) {
+ return v1 < v2 ? v1 : v2;
+}
+
+extern int2 __attribute__((overloadable)) min(int2 v1, int2 v2) {
+ int2 r;
+ r.x = v1.x < v2.x ? v1.x : v2.x;
+ r.y = v1.y < v2.y ? v1.y : v2.y;
+ return r;
+}
+
+extern int3 __attribute__((overloadable)) min(int3 v1, int3 v2) {
+ int3 r;
+ r.x = v1.x < v2.x ? v1.x : v2.x;
+ r.y = v1.y < v2.y ? v1.y : v2.y;
+ r.z = v1.z < v2.z ? v1.z : v2.z;
+ return r;
+}
+
+extern int4 __attribute__((overloadable)) min(int4 v1, int4 v2) {
+ int4 r;
+ r.x = v1.x < v2.x ? v1.x : v2.x;
+ r.y = v1.y < v2.y ? v1.y : v2.y;
+ r.z = v1.z < v2.z ? v1.z : v2.z;
+ r.w = v1.w < v2.w ? v1.w : v2.w;
+ return r;
+}
+
+extern int64_t __attribute__((overloadable)) min(int64_t v1, int64_t v2) {
+ return v1 < v2 ? v1 : v2;
+}
+
+extern long2 __attribute__((overloadable)) min(long2 v1, long2 v2) {
+ long2 r;
+ r.x = v1.x < v2.x ? v1.x : v2.x;
+ r.y = v1.y < v2.y ? v1.y : v2.y;
+ return r;
+}
+
+extern long3 __attribute__((overloadable)) min(long3 v1, long3 v2) {
+ long3 r;
+ r.x = v1.x < v2.x ? v1.x : v2.x;
+ r.y = v1.y < v2.y ? v1.y : v2.y;
+ r.z = v1.z < v2.z ? v1.z : v2.z;
+ return r;
+}
+
+extern long4 __attribute__((overloadable)) min(long4 v1, long4 v2) {
+ long4 r;
+ r.x = v1.x < v2.x ? v1.x : v2.x;
+ r.y = v1.y < v2.y ? v1.y : v2.y;
+ r.z = v1.z < v2.z ? v1.z : v2.z;
+ r.w = v1.w < v2.w ? v1.w : v2.w;
+ return r;
+}
+
+extern uchar __attribute__((overloadable)) min(uchar v1, uchar v2) {
+ return v1 < v2 ? v1 : v2;
+}
+
+extern uchar2 __attribute__((overloadable)) min(uchar2 v1, uchar2 v2) {
+ uchar2 r;
+ r.x = v1.x < v2.x ? v1.x : v2.x;
+ r.y = v1.y < v2.y ? v1.y : v2.y;
+ return r;
+}
+
+extern uchar3 __attribute__((overloadable)) min(uchar3 v1, uchar3 v2) {
+ uchar3 r;
+ r.x = v1.x < v2.x ? v1.x : v2.x;
+ r.y = v1.y < v2.y ? v1.y : v2.y;
+ r.z = v1.z < v2.z ? v1.z : v2.z;
+ return r;
+}
+
+extern uchar4 __attribute__((overloadable)) min(uchar4 v1, uchar4 v2) {
+ uchar4 r;
+ r.x = v1.x < v2.x ? v1.x : v2.x;
+ r.y = v1.y < v2.y ? v1.y : v2.y;
+ r.z = v1.z < v2.z ? v1.z : v2.z;
+ r.w = v1.w < v2.w ? v1.w : v2.w;
+ return r;
+}
+
+extern ushort __attribute__((overloadable)) min(ushort v1, ushort v2) {
+ return v1 < v2 ? v1 : v2;
+}
+
+extern ushort2 __attribute__((overloadable)) min(ushort2 v1, ushort2 v2) {
+ ushort2 r;
+ r.x = v1.x < v2.x ? v1.x : v2.x;
+ r.y = v1.y < v2.y ? v1.y : v2.y;
+ return r;
+}
+
+extern ushort3 __attribute__((overloadable)) min(ushort3 v1, ushort3 v2) {
+ ushort3 r;
+ r.x = v1.x < v2.x ? v1.x : v2.x;
+ r.y = v1.y < v2.y ? v1.y : v2.y;
+ r.z = v1.z < v2.z ? v1.z : v2.z;
+ return r;
+}
+
+extern ushort4 __attribute__((overloadable)) min(ushort4 v1, ushort4 v2) {
+ ushort4 r;
+ r.x = v1.x < v2.x ? v1.x : v2.x;
+ r.y = v1.y < v2.y ? v1.y : v2.y;
+ r.z = v1.z < v2.z ? v1.z : v2.z;
+ r.w = v1.w < v2.w ? v1.w : v2.w;
+ return r;
+}
+
+extern uint __attribute__((overloadable)) min(uint v1, uint v2) {
+ return v1 < v2 ? v1 : v2;
+}
+
+extern uint2 __attribute__((overloadable)) min(uint2 v1, uint2 v2) {
+ uint2 r;
+ r.x = v1.x < v2.x ? v1.x : v2.x;
+ r.y = v1.y < v2.y ? v1.y : v2.y;
+ return r;
+}
+
+extern uint3 __attribute__((overloadable)) min(uint3 v1, uint3 v2) {
+ uint3 r;
+ r.x = v1.x < v2.x ? v1.x : v2.x;
+ r.y = v1.y < v2.y ? v1.y : v2.y;
+ r.z = v1.z < v2.z ? v1.z : v2.z;
+ return r;
+}
+
+extern uint4 __attribute__((overloadable)) min(uint4 v1, uint4 v2) {
+ uint4 r;
+ r.x = v1.x < v2.x ? v1.x : v2.x;
+ r.y = v1.y < v2.y ? v1.y : v2.y;
+ r.z = v1.z < v2.z ? v1.z : v2.z;
+ r.w = v1.w < v2.w ? v1.w : v2.w;
+ return r;
+}
+
+extern ulong __attribute__((overloadable)) min(ulong v1, ulong v2) {
+ return v1 < v2 ? v1 : v2;
+}
+
+extern ulong2 __attribute__((overloadable)) min(ulong2 v1, ulong2 v2) {
+ ulong2 r;
+ r.x = v1.x < v2.x ? v1.x : v2.x;
+ r.y = v1.y < v2.y ? v1.y : v2.y;
+ return r;
+}
+
+extern ulong3 __attribute__((overloadable)) min(ulong3 v1, ulong3 v2) {
+ ulong3 r;
+ r.x = v1.x < v2.x ? v1.x : v2.x;
+ r.y = v1.y < v2.y ? v1.y : v2.y;
+ r.z = v1.z < v2.z ? v1.z : v2.z;
+ return r;
+}
+
+extern ulong4 __attribute__((overloadable)) min(ulong4 v1, ulong4 v2) {
+ ulong4 r;
+ r.x = v1.x < v2.x ? v1.x : v2.x;
+ r.y = v1.y < v2.y ? v1.y : v2.y;
+ r.z = v1.z < v2.z ? v1.z : v2.z;
+ r.w = v1.w < v2.w ? v1.w : v2.w;
+ return r;
+}
+
+extern float __attribute__((overloadable)) min(float v1, float v2) {
+ return fmin(v1, v2);
+}
+
+extern float2 __attribute__((overloadable)) min(float2 v1, float2 v2) {
+ return fmin(v1, v2);
+}
+
+extern float2 __attribute__((overloadable)) min(float2 v1, float v2) {
+ return fmin(v1, v2);
+}
+
+extern float3 __attribute__((overloadable)) min(float3 v1, float3 v2) {
+ return fmin(v1, v2);
+}
+
+extern float3 __attribute__((overloadable)) min(float3 v1, float v2) {
+ return fmin(v1, v2);
+}
+
+extern float4 __attribute__((overloadable)) min(float4 v1, float4 v2) {
+ return fmin(v1, v2);
+}
+
+extern float4 __attribute__((overloadable)) min(float4 v1, float v2) {
+ return fmin(v1, v2);
+}
+
+
+/*
+ * YUV
+ */
+
+extern uchar4 __attribute__((overloadable)) rsYuvToRGBA_uchar4(uchar y, uchar u, uchar v) {
+ short Y = ((short)y) - 16;
+ short U = ((short)u) - 128;
+ short V = ((short)v) - 128;
+
+ short4 p;
+ p.r = (Y * 298 + V * 409 + 128) >> 8;
+ p.g = (Y * 298 - U * 100 - V * 208 + 128) >> 8;
+ p.b = (Y * 298 + U * 516 + 128) >> 8;
+ p.a = 255;
+ p.r = rsClamp(p.r, (short)0, (short)255);
+ p.g = rsClamp(p.g, (short)0, (short)255);
+ p.b = rsClamp(p.b, (short)0, (short)255);
+
+ return convert_uchar4(p);
+}
+
+static float4 yuv_U_values = {0.f, -0.392f * 0.003921569f, +2.02 * 0.003921569f, 0.f};
+static float4 yuv_V_values = {1.603f * 0.003921569f, -0.815f * 0.003921569f, 0.f, 0.f};
+
+extern float4 __attribute__((overloadable)) rsYuvToRGBA_float4(uchar y, uchar u, uchar v) {
+ float4 color = (float)y * 0.003921569f;
+ float4 fU = ((float)u) - 128.f;
+ float4 fV = ((float)v) - 128.f;
+
+ color += fU * yuv_U_values;
+ color += fV * yuv_V_values;
+ color = clamp(color, 0.f, 1.f);
+ return color;
+}
+
+
+/*
+ * half_RECIP
+ */
+
+extern float __attribute__((overloadable)) half_recip(float v) {
+ // FIXME: actual algorithm for generic approximate reciprocal
+ return 1.f / v;
+}
+
+extern float2 __attribute__((overloadable)) half_recip(float2 v) {
+ float2 r;
+ r.x = half_recip(r.x);
+ r.y = half_recip(r.y);
+ return r;
+}
+
+extern float3 __attribute__((overloadable)) half_recip(float3 v) {
+ float3 r;
+ r.x = half_recip(r.x);
+ r.y = half_recip(r.y);
+ r.z = half_recip(r.z);
+ return r;
+}
+
+extern float4 __attribute__((overloadable)) half_recip(float4 v) {
+ float4 r;
+ r.x = half_recip(r.x);
+ r.y = half_recip(r.y);
+ r.z = half_recip(r.z);
+ r.w = half_recip(r.w);
+ return r;
+}
+
+
+/*
+ * half_SQRT
+ */
+
+extern float __attribute__((overloadable)) half_sqrt(float v) {
+ return sqrt(v);
+}
+
+extern float2 __attribute__((overloadable)) half_sqrt(float2 v) {
+ float2 r;
+ r.x = half_sqrt(v.x);
+ r.y = half_sqrt(v.y);
+ return r;
+}
+
+extern float3 __attribute__((overloadable)) half_sqrt(float3 v) {
+ float3 r;
+ r.x = half_sqrt(v.x);
+ r.y = half_sqrt(v.y);
+ r.z = half_sqrt(v.z);
+ return r;
+}
+
+extern float4 __attribute__((overloadable)) half_sqrt(float4 v) {
+ float4 r;
+ r.x = half_sqrt(v.x);
+ r.y = half_sqrt(v.y);
+ r.z = half_sqrt(v.z);
+ r.w = half_sqrt(v.w);
+ return r;
+}
+
+
+/*
+ * half_rsqrt
+ */
+
+extern float __attribute__((overloadable)) half_rsqrt(float v) {
+ return 1.f / sqrt(v);
+}
+
+extern float2 __attribute__((overloadable)) half_rsqrt(float2 v) {
+ float2 r;
+ r.x = half_rsqrt(v.x);
+ r.y = half_rsqrt(v.y);
+ return r;
+}
+
+extern float3 __attribute__((overloadable)) half_rsqrt(float3 v) {
+ float3 r;
+ r.x = half_rsqrt(v.x);
+ r.y = half_rsqrt(v.y);
+ r.z = half_rsqrt(v.z);
+ return r;
+}
+
+extern float4 __attribute__((overloadable)) half_rsqrt(float4 v) {
+ float4 r;
+ r.x = half_rsqrt(v.x);
+ r.y = half_rsqrt(v.y);
+ r.z = half_rsqrt(v.z);
+ r.w = half_rsqrt(v.w);
+ return r;
+}
+
diff --git a/driver/runtime/arch/x86_math.ll b/driver/runtime/arch/x86_math.ll
new file mode 100755
index 0000000..60add80
--- /dev/null
+++ b/driver/runtime/arch/x86_math.ll
@@ -0,0 +1,40 @@
+target datalayout = "e-p:32:32:32-i1:8:8-i8:8:8-i16:16:16-i32:32:32-i64:32:64-f32:32:32-f64:32:64-v64:64:64-v128:128:128-a0:0:64-f80:32:32-n8:16:32-S128"
+target triple = "i386-unknown-linux-gnu"
+
+declare float @llvm.sqrt.f32(float) nounwind readnone
+declare <2 x float> @llvm.sqrt.v2f32(<2 x float>) nounwind readnone
+declare <3 x float> @llvm.sqrt.v3f32(<3 x float>) nounwind readnone
+declare <4 x float> @llvm.sqrt.v4f32(<4 x float>) nounwind readnone
+declare float @llvm.exp.f32(float) nounwind readonly
+declare float @llvm.pow.f32(float, float) nounwind readonly
+
+define float @_Z4sqrtf(float %in) nounwind readnone alwaysinline {
+ %1 = tail call float @llvm.sqrt.f32(float %in) nounwind readnone
+ ret float %1
+}
+
+define <2 x float> @_Z4sqrtDv2_f(<2 x float> %in) nounwind readnone alwaysinline {
+ %1 = tail call <2 x float> @llvm.sqrt.v2f32(<2 x float> %in) nounwind readnone
+ ret <2 x float> %1
+}
+
+define <3 x float> @_Z4sqrtDv3_f(<3 x float> %in) nounwind readnone alwaysinline {
+ %1 = tail call <3 x float> @llvm.sqrt.v3f32(<3 x float> %in) nounwind readnone
+ ret <3 x float> %1
+}
+
+define <4 x float> @_Z4sqrtDv4_f(<4 x float> %in) nounwind readnone alwaysinline {
+ %1 = tail call <4 x float> @llvm.sqrt.v4f32(<4 x float> %in) nounwind readnone
+ ret <4 x float> %1
+}
+
+define float @_Z3expf(float %in) nounwind readnone {
+ %1 = tail call float @llvm.exp.f32(float %in) nounwind readnone
+ ret float %1
+}
+
+define float @_Z3powff(float %v1, float %v2) nounwind readnone {
+ %1 = tail call float @llvm.pow.f32(float %v1, float %v2) nounwind readnone
+ ret float %1
+}
+
diff --git a/driver/runtime/build_bc_lib.mk b/driver/runtime/build_bc_lib.mk
new file mode 100644
index 0000000..0344983
--- /dev/null
+++ b/driver/runtime/build_bc_lib.mk
@@ -0,0 +1,75 @@
+#
+# Copyright (C) 2012 The Android Open Source Project
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+# http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+#
+
+include $(BUILD_SYSTEM)/base_rules.mk
+
+BCC_STRIP_ATTR := $(BUILD_OUT_EXECUTABLES)/bcc_strip_attr$(BUILD_EXECUTABLE_SUFFIX)
+
+# We need to pass the +long64 flag to the underlying version of Clang, since
+# we are generating a library for use with Renderscript (64-bit long type,
+# not 32-bit).
+bc_clang_cc1_cflags := -target-feature +long64
+bc_translated_clang_cc1_cflags := $(addprefix -Xclang , $(bc_clang_cc1_cflags))
+
+bc_cflags := -MD \
+ -DRS_VERSION=$(RS_VERSION) \
+ -std=c99 \
+ -c \
+ -O3 \
+ -fno-builtin \
+ -emit-llvm \
+ -target armv7-none-linux-gnueabi \
+ -fsigned-char \
+ $(LOCAL_CFLAGS) \
+ $(bc_translated_clang_cc1_cflags)
+
+ifeq ($(rs_debug_runtime),1)
+bc_cflags += -DRS_DEBUG_RUNTIME
+endif
+rs_debug_runtime:=
+
+c_sources := $(filter %.c,$(LOCAL_SRC_FILES))
+ll_sources := $(filter %.ll,$(LOCAL_SRC_FILES))
+
+c_bc_files := $(patsubst %.c,%.bc, \
+ $(addprefix $(intermediates)/, $(c_sources)))
+
+ll_bc_files := $(patsubst %.ll,%.bc, \
+ $(addprefix $(intermediates)/, $(ll_sources)))
+
+$(c_bc_files): PRIVATE_INCLUDES := \
+ frameworks/rs/scriptc \
+ external/clang/lib/Headers
+$(c_bc_files): PRIVATE_CFLAGS := $(bc_cflags)
+
+$(c_bc_files): $(intermediates)/%.bc: $(LOCAL_PATH)/%.c $(CLANG)
+ @mkdir -p $(dir $@)
+ $(hide) $(CLANG) $(addprefix -I, $(PRIVATE_INCLUDES)) $(PRIVATE_CFLAGS) $< -o $@
+
+$(ll_bc_files): $(intermediates)/%.bc: $(LOCAL_PATH)/%.ll $(LLVM_AS)
+ @mkdir -p $(dir $@)
+ $(hide) $(LLVM_AS) $< -o $@
+
+-include $(c_bc_files:%.bc=%.d)
+-include $(ll_bc_files:%.bc=%.d)
+
+$(LOCAL_BUILT_MODULE): PRIVATE_BC_FILES := $(c_bc_files) $(ll_bc_files)
+$(LOCAL_BUILT_MODULE): $(c_bc_files) $(ll_bc_files)
+$(LOCAL_BUILT_MODULE): $(LLVM_LINK) $(clcore_LLVM_LD)
+$(LOCAL_BUILT_MODULE): $(LLVM_AS) $(BCC_STRIP_ATTR)
+ @mkdir -p $(dir $@)
+ $(hide) $(LLVM_LINK) $(PRIVATE_BC_FILES) -o $@.unstripped
+ $(hide) $(BCC_STRIP_ATTR) -o $@ $@.unstripped
diff --git a/driver/runtime/convert.ll b/driver/runtime/convert.ll
new file mode 100644
index 0000000..f45850d
--- /dev/null
+++ b/driver/runtime/convert.ll
@@ -0,0 +1,731 @@
+target datalayout = "e-p:32:32:32-i1:8:8-i8:8:8-i16:16:16-i32:32:32-i64:64:64-f32:32:32-f64:64:64-v64:64:64-v128:64:128-a0:0:64-n32-S64"
+target triple = "armv7-none-linux-gnueabi"
+
+
+;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;
+;;;;;;;;; FLOAT ;;;;;;;;;;
+;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;
+
+define <2 x float> @_Z14convert_float2Dv2_h(<2 x i8> %in) nounwind readnone alwaysinline {
+ %1 = uitofp <2 x i8> %in to <2 x float>
+ ret <2 x float> %1
+}
+
+define <3 x float> @_Z14convert_float3Dv3_h(<3 x i8> %in) nounwind readnone alwaysinline {
+ %1 = uitofp <3 x i8> %in to <3 x float>
+ ret <3 x float> %1
+}
+
+define <4 x float> @_Z14convert_float4Dv4_h(<4 x i8> %in) nounwind readnone alwaysinline {
+ %1 = uitofp <4 x i8> %in to <4 x float>
+ ret <4 x float> %1
+}
+
+define <2 x float> @_Z14convert_float2Dv2_c(<2 x i8> %in) nounwind readnone alwaysinline {
+ %1 = sitofp <2 x i8> %in to <2 x float>
+ ret <2 x float> %1
+}
+
+define <3 x float> @_Z14convert_float3Dv3_c(<3 x i8> %in) nounwind readnone alwaysinline {
+ %1 = sitofp <3 x i8> %in to <3 x float>
+ ret <3 x float> %1
+}
+
+define <4 x float> @_Z14convert_float4Dv4_c(<4 x i8> %in) nounwind readnone alwaysinline {
+ %1 = sitofp <4 x i8> %in to <4 x float>
+ ret <4 x float> %1
+}
+
+define <2 x float> @_Z14convert_float2Dv2_t(<2 x i16> %in) nounwind readnone alwaysinline {
+ %1 = uitofp <2 x i16> %in to <2 x float>
+ ret <2 x float> %1
+}
+
+define <3 x float> @_Z14convert_float3Dv3_t(<3 x i16> %in) nounwind readnone alwaysinline {
+ %1 = uitofp <3 x i16> %in to <3 x float>
+ ret <3 x float> %1
+}
+
+define <4 x float> @_Z14convert_float4Dv4_t(<4 x i16> %in) nounwind readnone alwaysinline {
+ %1 = uitofp <4 x i16> %in to <4 x float>
+ ret <4 x float> %1
+}
+
+define <2 x float> @_Z14convert_float2Dv2_s(<2 x i16> %in) nounwind readnone alwaysinline {
+ %1 = sitofp <2 x i16> %in to <2 x float>
+ ret <2 x float> %1
+}
+
+define <3 x float> @_Z14convert_float3Dv3_s(<3 x i16> %in) nounwind readnone alwaysinline {
+ %1 = sitofp <3 x i16> %in to <3 x float>
+ ret <3 x float> %1
+}
+
+define <4 x float> @_Z14convert_float4Dv4_s(<4 x i16> %in) nounwind readnone alwaysinline {
+ %1 = sitofp <4 x i16> %in to <4 x float>
+ ret <4 x float> %1
+}
+
+define <2 x float> @_Z14convert_float2Dv2_j(<2 x i32> %in) nounwind readnone alwaysinline {
+ %1 = uitofp <2 x i32> %in to <2 x float>
+ ret <2 x float> %1
+}
+
+define <3 x float> @_Z14convert_float3Dv3_j(<3 x i32> %in) nounwind readnone alwaysinline {
+ %1 = uitofp <3 x i32> %in to <3 x float>
+ ret <3 x float> %1
+}
+
+define <4 x float> @_Z14convert_float4Dv4_j(<4 x i32> %in) nounwind readnone alwaysinline {
+ %1 = uitofp <4 x i32> %in to <4 x float>
+ ret <4 x float> %1
+}
+
+define <2 x float> @_Z14convert_float2Dv2_i(<2 x i32> %in) nounwind readnone alwaysinline {
+ %1 = sitofp <2 x i32> %in to <2 x float>
+ ret <2 x float> %1
+}
+
+define <3 x float> @_Z14convert_float3Dv3_i(<3 x i32> %in) nounwind readnone alwaysinline {
+ %1 = sitofp <3 x i32> %in to <3 x float>
+ ret <3 x float> %1
+}
+
+define <4 x float> @_Z14convert_float4Dv4_i(<4 x i32> %in) nounwind readnone alwaysinline {
+ %1 = sitofp <4 x i32> %in to <4 x float>
+ ret <4 x float> %1
+}
+
+define <2 x float> @_Z14convert_float2Dv2_f(<2 x float> %in) nounwind readnone alwaysinline {
+ ret <2 x float> %in
+}
+
+define <3 x float> @_Z14convert_float3Dv3_f(<3 x float> %in) nounwind readnone alwaysinline {
+ ret <3 x float> %in
+}
+
+define <4 x float> @_Z14convert_float4Dv4_f(<4 x float> %in) nounwind readnone alwaysinline {
+ ret <4 x float> %in
+}
+
+
+;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;
+;;;;;;;;; CHAR ;;;;;;;;;;
+;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;
+define <4 x i8> @_Z13convert_char4Dv4_f(<4 x float> %in) nounwind readnone alwaysinline {
+ %1 = fptosi <4 x float> %in to <4 x i8>
+ ret <4 x i8> %1
+}
+
+define <3 x i8> @_Z13convert_char3Dv3_f(<3 x float> %in) nounwind readnone alwaysinline {
+ %1 = fptosi <3 x float> %in to <3 x i8>
+ ret <3 x i8> %1
+}
+
+define <2 x i8> @_Z13convert_char2Dv2_f(<2 x float> %in) nounwind readnone alwaysinline {
+ %1 = fptosi <2 x float> %in to <2 x i8>
+ ret <2 x i8> %1
+}
+
+define <4 x i8> @_Z13convert_char4Dv4_h(<4 x i8> %in) nounwind readnone alwaysinline {
+ ret <4 x i8> %in
+}
+
+define <3 x i8> @_Z13convert_char3Dv3_h(<3 x i8> %in) nounwind readnone alwaysinline {
+ ret <3 x i8> %in
+}
+
+define <2 x i8> @_Z13convert_char2Dv2_h(<2 x i8> %in) nounwind readnone alwaysinline {
+ ret <2 x i8> %in
+}
+
+define <4 x i8> @_Z13convert_char4Dv4_c(<4 x i8> %in) nounwind readnone alwaysinline {
+ ret <4 x i8> %in
+}
+
+define <3 x i8> @_Z13convert_char3Dv3_c(<3 x i8> %in) nounwind readnone alwaysinline {
+ ret <3 x i8> %in
+}
+
+define <2 x i8> @_Z13convert_char2Dv2_c(<2 x i8> %in) nounwind readnone alwaysinline {
+ ret <2 x i8> %in
+}
+
+define <4 x i8> @_Z13convert_char4Dv4_t(<4 x i16> %in) nounwind readnone alwaysinline {
+ %1 = trunc <4 x i16> %in to <4 x i8>
+ ret <4 x i8> %1
+}
+
+define <3 x i8> @_Z13convert_char3Dv3_t(<3 x i16> %in) nounwind readnone alwaysinline {
+ %1 = trunc <3 x i16> %in to <3 x i8>
+ ret <3 x i8> %1
+}
+
+define <2 x i8> @_Z13convert_char2Dv2_t(<2 x i16> %in) nounwind readnone alwaysinline {
+ %1 = trunc <2 x i16> %in to <2 x i8>
+ ret <2 x i8> %1
+}
+
+define <4 x i8> @_Z13convert_char4Dv4_s(<4 x i16> %in) nounwind readnone alwaysinline {
+ %1 = trunc <4 x i16> %in to <4 x i8>
+ ret <4 x i8> %1
+}
+
+define <3 x i8> @_Z13convert_char3Dv3_s(<3 x i16> %in) nounwind readnone alwaysinline {
+ %1 = trunc <3 x i16> %in to <3 x i8>
+ ret <3 x i8> %1
+}
+
+define <2 x i8> @_Z13convert_char2Dv2_s(<2 x i16> %in) nounwind readnone alwaysinline {
+ %1 = trunc <2 x i16> %in to <2 x i8>
+ ret <2 x i8> %1
+}
+
+define <4 x i8> @_Z13convert_char4Dv4_j(<4 x i32> %in) nounwind readnone alwaysinline {
+ %1 = trunc <4 x i32> %in to <4 x i8>
+ ret <4 x i8> %1
+}
+
+define <3 x i8> @_Z13convert_char3Dv3_j(<3 x i32> %in) nounwind readnone alwaysinline {
+ %1 = trunc <3 x i32> %in to <3 x i8>
+ ret <3 x i8> %1
+}
+
+define <2 x i8> @_Z13convert_char2Dv2_j(<2 x i32> %in) nounwind readnone alwaysinline {
+ %1 = trunc <2 x i32> %in to <2 x i8>
+ ret <2 x i8> %1
+}
+
+define <4 x i8> @_Z13convert_char4Dv4_i(<4 x i32> %in) nounwind readnone alwaysinline {
+ %1 = trunc <4 x i32> %in to <4 x i8>
+ ret <4 x i8> %1
+}
+
+define <3 x i8> @_Z13convert_char3Dv3_i(<3 x i32> %in) nounwind readnone alwaysinline {
+ %1 = trunc <3 x i32> %in to <3 x i8>
+ ret <3 x i8> %1
+}
+
+define <2 x i8> @_Z13convert_char2Dv2_i(<2 x i32> %in) nounwind readnone alwaysinline {
+ %1 = trunc <2 x i32> %in to <2 x i8>
+ ret <2 x i8> %1
+}
+
+
+;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;
+;;;;;;;;; UCHAR ;;;;;;;;;;
+;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;
+
+define <4 x i8> @_Z14convert_uchar4Dv4_f(<4 x float> %in) nounwind readnone alwaysinline {
+ %1 = fptoui <4 x float> %in to <4 x i8>
+ ret <4 x i8> %1
+}
+
+define <3 x i8> @_Z14convert_uchar3Dv3_f(<3 x float> %in) nounwind readnone alwaysinline {
+ %1 = fptoui <3 x float> %in to <3 x i8>
+ ret <3 x i8> %1
+}
+
+define <2 x i8> @_Z14convert_uchar2Dv2_f(<2 x float> %in) nounwind readnone alwaysinline {
+ %1 = fptoui <2 x float> %in to <2 x i8>
+ ret <2 x i8> %1
+}
+
+define <4 x i8> @_Z14convert_uchar4Dv4_h(<4 x i8> %in) nounwind readnone alwaysinline {
+ ret <4 x i8> %in
+}
+
+define <3 x i8> @_Z14convert_uchar3Dv3_h(<3 x i8> %in) nounwind readnone alwaysinline {
+ ret <3 x i8> %in
+}
+
+define <2 x i8> @_Z14convert_uchar2Dv2_h(<2 x i8> %in) nounwind readnone alwaysinline {
+ ret <2 x i8> %in
+}
+
+define <4 x i8> @_Z14convert_uchar4Dv4_c(<4 x i8> %in) nounwind readnone alwaysinline {
+ ret <4 x i8> %in
+}
+
+define <3 x i8> @_Z14convert_uchar3Dv3_c(<3 x i8> %in) nounwind readnone alwaysinline {
+ ret <3 x i8> %in
+}
+
+define <2 x i8> @_Z14convert_uchar2Dv2_c(<2 x i8> %in) nounwind readnone alwaysinline {
+ ret <2 x i8> %in
+}
+
+define <4 x i8> @_Z14convert_uchar4Dv4_t(<4 x i16> %in) nounwind readnone alwaysinline {
+ %1 = trunc <4 x i16> %in to <4 x i8>
+ ret <4 x i8> %1
+}
+
+define <3 x i8> @_Z14convert_uchar3Dv3_t(<3 x i16> %in) nounwind readnone alwaysinline {
+ %1 = trunc <3 x i16> %in to <3 x i8>
+ ret <3 x i8> %1
+}
+
+define <2 x i8> @_Z14convert_uchar2Dv2_t(<2 x i16> %in) nounwind readnone alwaysinline {
+ %1 = trunc <2 x i16> %in to <2 x i8>
+ ret <2 x i8> %1
+}
+
+define <4 x i8> @_Z14convert_uchar4Dv4_s(<4 x i16> %in) nounwind readnone alwaysinline {
+ %1 = trunc <4 x i16> %in to <4 x i8>
+ ret <4 x i8> %1
+}
+
+define <3 x i8> @_Z14convert_uchar3Dv3_s(<3 x i16> %in) nounwind readnone alwaysinline {
+ %1 = trunc <3 x i16> %in to <3 x i8>
+ ret <3 x i8> %1
+}
+
+define <2 x i8> @_Z14convert_uchar2Dv2_s(<2 x i16> %in) nounwind readnone alwaysinline {
+ %1 = trunc <2 x i16> %in to <2 x i8>
+ ret <2 x i8> %1
+}
+
+define <4 x i8> @_Z14convert_uchar4Dv4_j(<4 x i32> %in) nounwind readnone alwaysinline {
+ %1 = trunc <4 x i32> %in to <4 x i8>
+ ret <4 x i8> %1
+}
+
+define <3 x i8> @_Z14convert_uchar3Dv3_j(<3 x i32> %in) nounwind readnone alwaysinline {
+ %1 = trunc <3 x i32> %in to <3 x i8>
+ ret <3 x i8> %1
+}
+
+define <2 x i8> @_Z14convert_uchar2Dv2_j(<2 x i32> %in) nounwind readnone alwaysinline {
+ %1 = trunc <2 x i32> %in to <2 x i8>
+ ret <2 x i8> %1
+}
+
+define <4 x i8> @_Z14convert_uchar4Dv4_i(<4 x i32> %in) nounwind readnone alwaysinline {
+ %1 = trunc <4 x i32> %in to <4 x i8>
+ ret <4 x i8> %1
+}
+
+define <3 x i8> @_Z14convert_uchar3Dv3_i(<3 x i32> %in) nounwind readnone alwaysinline {
+ %1 = trunc <3 x i32> %in to <3 x i8>
+ ret <3 x i8> %1
+}
+
+define <2 x i8> @_Z14convert_uchar2Dv2_i(<2 x i32> %in) nounwind readnone alwaysinline {
+ %1 = trunc <2 x i32> %in to <2 x i8>
+ ret <2 x i8> %1
+}
+
+;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;
+;;;;;;;;; SHORT ;;;;;;;;;;
+;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;
+
+define <4 x i16> @_Z14convert_short4Dv4_f(<4 x float> %in) nounwind readnone alwaysinline {
+ %1 = fptosi <4 x float> %in to <4 x i16>
+ ret <4 x i16> %1
+}
+
+define <3 x i16> @_Z14convert_short3Dv3_f(<3 x float> %in) nounwind readnone alwaysinline {
+ %1 = fptosi <3 x float> %in to <3 x i16>
+ ret <3 x i16> %1
+}
+
+define <2 x i16> @_Z14convert_short2Dv2_f(<2 x float> %in) nounwind readnone alwaysinline {
+ %1 = fptosi <2 x float> %in to <2 x i16>
+ ret <2 x i16> %1
+}
+
+define <4 x i16> @_Z14convert_short4Dv4_h(<4 x i8> %in) nounwind readnone alwaysinline {
+ %1 = zext <4 x i8> %in to <4 x i16>
+ ret <4 x i16> %1
+}
+
+define <3 x i16> @_Z14convert_short3Dv3_h(<3 x i8> %in) nounwind readnone alwaysinline {
+ %1 = zext <3 x i8> %in to <3 x i16>
+ ret <3 x i16> %1
+}
+
+define <2 x i16> @_Z14convert_short2Dv2_h(<2 x i8> %in) nounwind readnone alwaysinline {
+ %1 = zext <2 x i8> %in to <2 x i16>
+ ret <2 x i16> %1
+}
+
+define <4 x i16> @_Z14convert_short4Dv4_c(<4 x i8> %in) nounwind readnone alwaysinline {
+ %1 = sext <4 x i8> %in to <4 x i16>
+ ret <4 x i16> %1
+}
+
+define <3 x i16> @_Z14convert_short3Dv3_c(<3 x i8> %in) nounwind readnone alwaysinline {
+ %1 = sext <3 x i8> %in to <3 x i16>
+ ret <3 x i16> %1
+}
+
+define <2 x i16> @_Z14convert_short2Dv2_c(<2 x i8> %in) nounwind readnone alwaysinline {
+ %1 = sext <2 x i8> %in to <2 x i16>
+ ret <2 x i16> %1
+}
+
+define <4 x i16> @_Z14convert_short4Dv4_t(<4 x i16> %in) nounwind readnone alwaysinline {
+ ret <4 x i16> %in
+}
+
+define <3 x i16> @_Z14convert_short3Dv3_t(<3 x i16> %in) nounwind readnone alwaysinline {
+ ret <3 x i16> %in
+}
+
+define <2 x i16> @_Z14convert_short2Dv2_t(<2 x i16> %in) nounwind readnone alwaysinline {
+ ret <2 x i16> %in
+}
+
+define <4 x i16> @_Z14convert_short4Dv4_s(<4 x i16> %in) nounwind readnone alwaysinline {
+ ret <4 x i16> %in
+}
+
+define <3 x i16> @_Z14convert_short3Dv3_s(<3 x i16> %in) nounwind readnone alwaysinline {
+ ret <3 x i16> %in
+}
+
+define <2 x i16> @_Z14convert_short2Dv2_s(<2 x i16> %in) nounwind readnone alwaysinline {
+ ret <2 x i16> %in
+}
+
+define <4 x i16> @_Z14convert_short4Dv4_j(<4 x i32> %in) nounwind readnone alwaysinline {
+ %1 = trunc <4 x i32> %in to <4 x i16>
+ ret <4 x i16> %1
+}
+
+define <3 x i16> @_Z14convert_short3Dv3_j(<3 x i32> %in) nounwind readnone alwaysinline {
+ %1 = trunc <3 x i32> %in to <3 x i16>
+ ret <3 x i16> %1
+}
+
+define <2 x i16> @_Z14convert_short2Dv2_j(<2 x i32> %in) nounwind readnone alwaysinline {
+ %1 = trunc <2 x i32> %in to <2 x i16>
+ ret <2 x i16> %1
+}
+
+define <4 x i16> @_Z14convert_short4Dv4_i(<4 x i32> %in) nounwind readnone alwaysinline {
+ %1 = trunc <4 x i32> %in to <4 x i16>
+ ret <4 x i16> %1
+}
+
+define <3 x i16> @_Z14convert_short3Dv3_i(<3 x i32> %in) nounwind readnone alwaysinline {
+ %1 = trunc <3 x i32> %in to <3 x i16>
+ ret <3 x i16> %1
+}
+
+define <2 x i16> @_Z14convert_short2Dv2_i(<2 x i32> %in) nounwind readnone alwaysinline {
+ %1 = trunc <2 x i32> %in to <2 x i16>
+ ret <2 x i16> %1
+}
+
+
+;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;
+;;;;;;;;; USHORT ;;;;;;;;;;
+;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;
+
+define <4 x i16> @_Z15convert_ushort4Dv4_f(<4 x float> %in) nounwind readnone alwaysinline {
+ %1 = fptoui <4 x float> %in to <4 x i16>
+ ret <4 x i16> %1
+}
+
+define <3 x i16> @_Z15convert_ushort3Dv3_f(<3 x float> %in) nounwind readnone alwaysinline {
+ %1 = fptoui <3 x float> %in to <3 x i16>
+ ret <3 x i16> %1
+}
+
+define <2 x i16> @_Z15convert_ushort2Dv2_f(<2 x float> %in) nounwind readnone alwaysinline {
+ %1 = fptoui <2 x float> %in to <2 x i16>
+ ret <2 x i16> %1
+}
+
+define <4 x i16> @_Z15convert_ushort4Dv4_h(<4 x i8> %in) nounwind readnone alwaysinline {
+ %1 = zext <4 x i8> %in to <4 x i16>
+ ret <4 x i16> %1
+}
+
+define <3 x i16> @_Z15convert_ushort3Dv3_h(<3 x i8> %in) nounwind readnone alwaysinline {
+ %1 = zext <3 x i8> %in to <3 x i16>
+ ret <3 x i16> %1
+}
+
+define <2 x i16> @_Z15convert_ushort2Dv2_h(<2 x i8> %in) nounwind readnone alwaysinline {
+ %1 = zext <2 x i8> %in to <2 x i16>
+ ret <2 x i16> %1
+}
+
+define <4 x i16> @_Z15convert_ushort4Dv4_c(<4 x i8> %in) nounwind readnone alwaysinline {
+ %1 = zext <4 x i8> %in to <4 x i16>
+ ret <4 x i16> %1
+}
+
+define <3 x i16> @_Z15convert_ushort3Dv3_c(<3 x i8> %in) nounwind readnone alwaysinline {
+ %1 = zext <3 x i8> %in to <3 x i16>
+ ret <3 x i16> %1
+}
+
+define <2 x i16> @_Z15convert_ushort2Dv2_c(<2 x i8> %in) nounwind readnone alwaysinline {
+ %1 = zext <2 x i8> %in to <2 x i16>
+ ret <2 x i16> %1
+}
+
+define <4 x i16> @_Z15convert_ushort4Dv4_t(<4 x i16> %in) nounwind readnone alwaysinline {
+ ret <4 x i16> %in
+}
+
+define <3 x i16> @_Z15convert_ushort3Dv3_t(<3 x i16> %in) nounwind readnone alwaysinline {
+ ret <3 x i16> %in
+}
+
+define <2 x i16> @_Z15convert_ushort2Dv2_t(<2 x i16> %in) nounwind readnone alwaysinline {
+ ret <2 x i16> %in
+}
+
+define <4 x i16> @_Z15convert_ushort4Dv4_s(<4 x i16> %in) nounwind readnone alwaysinline {
+ ret <4 x i16> %in
+}
+
+define <3 x i16> @_Z15convert_ushort3Dv3_s(<3 x i16> %in) nounwind readnone alwaysinline {
+ ret <3 x i16> %in
+}
+
+define <2 x i16> @_Z15convert_ushort2Dv2_s(<2 x i16> %in) nounwind readnone alwaysinline {
+ ret <2 x i16> %in
+}
+
+define <4 x i16> @_Z15convert_ushort4Dv4_j(<4 x i32> %in) nounwind readnone alwaysinline {
+ %1 = trunc <4 x i32> %in to <4 x i16>
+ ret <4 x i16> %1
+}
+
+define <3 x i16> @_Z15convert_ushort3Dv3_j(<3 x i32> %in) nounwind readnone alwaysinline {
+ %1 = trunc <3 x i32> %in to <3 x i16>
+ ret <3 x i16> %1
+}
+
+define <2 x i16> @_Z15convert_ushort2Dv2_j(<2 x i32> %in) nounwind readnone alwaysinline {
+ %1 = trunc <2 x i32> %in to <2 x i16>
+ ret <2 x i16> %1
+}
+
+define <4 x i16> @_Z15convert_ushort4Dv4_i(<4 x i32> %in) nounwind readnone alwaysinline {
+ %1 = trunc <4 x i32> %in to <4 x i16>
+ ret <4 x i16> %1
+}
+
+define <3 x i16> @_Z15convert_ushort3Dv3_i(<3 x i32> %in) nounwind readnone alwaysinline {
+ %1 = trunc <3 x i32> %in to <3 x i16>
+ ret <3 x i16> %1
+}
+
+define <2 x i16> @_Z15convert_ushort2Dv2_i(<2 x i32> %in) nounwind readnone alwaysinline {
+ %1 = trunc <2 x i32> %in to <2 x i16>
+ ret <2 x i16> %1
+}
+
+
+;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;
+;;;;;;;;; INT ;;;;;;;;;;
+;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;
+
+define <4 x i32> @_Z12convert_int4Dv4_f(<4 x float> %in) nounwind readnone alwaysinline {
+ %1 = fptosi <4 x float> %in to <4 x i32>
+ ret <4 x i32> %1
+}
+
+define <3 x i32> @_Z12convert_int3Dv3_f(<3 x float> %in) nounwind readnone alwaysinline {
+ %1 = fptosi <3 x float> %in to <3 x i32>
+ ret <3 x i32> %1
+}
+
+define <2 x i32> @_Z12convert_int2Dv2_f(<2 x float> %in) nounwind readnone alwaysinline {
+ %1 = fptosi <2 x float> %in to <2 x i32>
+ ret <2 x i32> %1
+}
+
+define <4 x i32> @_Z12convert_int4Dv4_h(<4 x i8> %in) nounwind readnone alwaysinline {
+ %1 = zext <4 x i8> %in to <4 x i32>
+ ret <4 x i32> %1
+}
+
+define <3 x i32> @_Z12convert_int3Dv3_h(<3 x i8> %in) nounwind readnone alwaysinline {
+ %1 = zext <3 x i8> %in to <3 x i32>
+ ret <3 x i32> %1
+}
+
+define <2 x i32> @_Z12convert_int2Dv2_h(<2 x i8> %in) nounwind readnone alwaysinline {
+ %1 = zext <2 x i8> %in to <2 x i32>
+ ret <2 x i32> %1
+}
+
+define <4 x i32> @_Z12convert_int4Dv4_c(<4 x i8> %in) nounwind readnone alwaysinline {
+ %1 = sext <4 x i8> %in to <4 x i32>
+ ret <4 x i32> %1
+}
+
+define <3 x i32> @_Z12convert_int3Dv3_c(<3 x i8> %in) nounwind readnone alwaysinline {
+ %1 = sext <3 x i8> %in to <3 x i32>
+ ret <3 x i32> %1
+}
+
+define <2 x i32> @_Z12convert_int2Dv2_c(<2 x i8> %in) nounwind readnone alwaysinline {
+ %1 = sext <2 x i8> %in to <2 x i32>
+ ret <2 x i32> %1
+}
+
+define <4 x i32> @_Z12convert_int4Dv4_t(<4 x i16> %in) nounwind readnone alwaysinline {
+ %1 = zext <4 x i16> %in to <4 x i32>
+ ret <4 x i32> %1
+}
+
+define <3 x i32> @_Z12convert_int3Dv3_t(<3 x i16> %in) nounwind readnone alwaysinline {
+ %1 = zext <3 x i16> %in to <3 x i32>
+ ret <3 x i32> %1
+}
+
+define <2 x i32> @_Z12convert_int2Dv2_t(<2 x i16> %in) nounwind readnone alwaysinline {
+ %1 = zext <2 x i16> %in to <2 x i32>
+ ret <2 x i32> %1
+}
+
+define <4 x i32> @_Z12convert_int4Dv4_s(<4 x i16> %in) nounwind readnone alwaysinline {
+ %1 = sext <4 x i16> %in to <4 x i32>
+ ret <4 x i32> %1
+}
+
+define <3 x i32> @_Z12convert_int3Dv3_s(<3 x i16> %in) nounwind readnone alwaysinline {
+ %1 = sext <3 x i16> %in to <3 x i32>
+ ret <3 x i32> %1
+}
+
+define <2 x i32> @_Z12convert_int2Dv2_s(<2 x i16> %in) nounwind readnone alwaysinline {
+ %1 = sext <2 x i16> %in to <2 x i32>
+ ret <2 x i32> %1
+}
+
+define <4 x i32> @_Z12convert_int4Dv4_j(<4 x i32> %in) nounwind readnone alwaysinline {
+ ret <4 x i32> %in
+}
+
+define <3 x i32> @_Z12convert_int3Dv3_j(<3 x i32> %in) nounwind readnone alwaysinline {
+ ret <3 x i32> %in
+}
+
+define <2 x i32> @_Z12convert_int2Dv2_j(<2 x i32> %in) nounwind readnone alwaysinline {
+ ret <2 x i32> %in
+}
+
+define <4 x i32> @_Z12convert_int4Dv4_i(<4 x i32> %in) nounwind readnone alwaysinline {
+ ret <4 x i32> %in
+}
+
+define <3 x i32> @_Z12convert_int3Dv3_i(<3 x i32> %in) nounwind readnone alwaysinline {
+ ret <3 x i32> %in
+}
+
+define <2 x i32> @_Z12convert_int2Dv2_i(<2 x i32> %in) nounwind readnone alwaysinline {
+ ret <2 x i32> %in
+}
+
+
+;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;
+;;;;;;;;; UINT ;;;;;;;;;;
+;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;
+
+define <4 x i32> @_Z13convert_uint4Dv4_f(<4 x float> %in) nounwind readnone alwaysinline {
+ %1 = fptoui <4 x float> %in to <4 x i32>
+ ret <4 x i32> %1
+}
+
+define <3 x i32> @_Z13convert_uint3Dv3_f(<3 x float> %in) nounwind readnone alwaysinline {
+ %1 = fptoui <3 x float> %in to <3 x i32>
+ ret <3 x i32> %1
+}
+
+define <2 x i32> @_Z13convert_uint2Dv2_f(<2 x float> %in) nounwind readnone alwaysinline {
+ %1 = fptoui <2 x float> %in to <2 x i32>
+ ret <2 x i32> %1
+}
+
+define <4 x i32> @_Z13convert_uint4Dv4_h(<4 x i8> %in) nounwind readnone alwaysinline {
+ %1 = zext <4 x i8> %in to <4 x i32>
+ ret <4 x i32> %1
+}
+
+define <3 x i32> @_Z13convert_uint3Dv3_h(<3 x i8> %in) nounwind readnone alwaysinline {
+ %1 = zext <3 x i8> %in to <3 x i32>
+ ret <3 x i32> %1
+}
+
+define <2 x i32> @_Z13convert_uint2Dv2_h(<2 x i8> %in) nounwind readnone alwaysinline {
+ %1 = zext <2 x i8> %in to <2 x i32>
+ ret <2 x i32> %1
+}
+
+define <4 x i32> @_Z13convert_uint4Dv4_c(<4 x i8> %in) nounwind readnone alwaysinline {
+ %1 = zext <4 x i8> %in to <4 x i32>
+ ret <4 x i32> %1
+}
+
+define <3 x i32> @_Z13convert_uint3Dv3_c(<3 x i8> %in) nounwind readnone alwaysinline {
+ %1 = zext <3 x i8> %in to <3 x i32>
+ ret <3 x i32> %1
+}
+
+define <2 x i32> @_Z13convert_uint2Dv2_c(<2 x i8> %in) nounwind readnone alwaysinline {
+ %1 = zext <2 x i8> %in to <2 x i32>
+ ret <2 x i32> %1
+}
+
+define <4 x i32> @_Z13convert_uint4Dv4_t(<4 x i16> %in) nounwind readnone alwaysinline {
+ %1 = zext <4 x i16> %in to <4 x i32>
+ ret <4 x i32> %1
+}
+
+define <3 x i32> @_Z13convert_uint3Dv3_t(<3 x i16> %in) nounwind readnone alwaysinline {
+ %1 = zext <3 x i16> %in to <3 x i32>
+ ret <3 x i32> %1
+}
+
+define <2 x i32> @_Z13convert_uint2Dv2_t(<2 x i16> %in) nounwind readnone alwaysinline {
+ %1 = zext <2 x i16> %in to <2 x i32>
+ ret <2 x i32> %1
+}
+
+define <4 x i32> @_Z13convert_uint4Dv4_s(<4 x i16> %in) nounwind readnone alwaysinline {
+ %1 = zext <4 x i16> %in to <4 x i32>
+ ret <4 x i32> %1
+}
+
+define <3 x i32> @_Z13convert_uint3Dv3_s(<3 x i16> %in) nounwind readnone alwaysinline {
+ %1 = zext <3 x i16> %in to <3 x i32>
+ ret <3 x i32> %1
+}
+
+define <2 x i32> @_Z13convert_uint2Dv2_s(<2 x i16> %in) nounwind readnone alwaysinline {
+ %1 = zext <2 x i16> %in to <2 x i32>
+ ret <2 x i32> %1
+}
+
+define <4 x i32> @_Z13convert_uint4Dv4_j(<4 x i32> %in) nounwind readnone alwaysinline {
+ ret <4 x i32> %in
+}
+
+define <3 x i32> @_Z13convert_uint3Dv3_j(<3 x i32> %in) nounwind readnone alwaysinline {
+ ret <3 x i32> %in
+}
+
+define <2 x i32> @_Z13convert_uint2Dv2_j(<2 x i32> %in) nounwind readnone alwaysinline {
+ ret <2 x i32> %in
+}
+
+define <4 x i32> @_Z13convert_uint4Dv4_i(<4 x i32> %in) nounwind readnone alwaysinline {
+ ret <4 x i32> %in
+}
+
+define <3 x i32> @_Z13convert_uint3Dv3_i(<3 x i32> %in) nounwind readnone alwaysinline {
+ ret <3 x i32> %in
+}
+
+define <2 x i32> @_Z13convert_uint2Dv2_i(<2 x i32> %in) nounwind readnone alwaysinline {
+ ret <2 x i32> %in
+}
diff --git a/driver/runtime/math.ll b/driver/runtime/math.ll
new file mode 100644
index 0000000..f026d15
--- /dev/null
+++ b/driver/runtime/math.ll
@@ -0,0 +1,19 @@
+target datalayout = "e-p:32:32:32-i1:8:8-i8:8:8-i16:16:16-i32:32:32-i64:64:64-f32:32:32-f64:64:64-v64:64:64-v128:64:128-a0:0:64-n32-S64"
+target triple = "armv7-none-linux-gnueabi"
+
+declare float @llvm.sqrt.f32(float)
+declare float @llvm.pow.f32(float, float)
+declare float @llvm.fabs.f32(float)
+declare <2 x float> @llvm.fabs.v2f32(<2 x float>)
+declare <3 x float> @llvm.fabs.v3f32(<3 x float>)
+declare <4 x float> @llvm.fabs.v4f32(<4 x float>)
+
+define float @_Z4sqrtf(float %v) nounwind readnone alwaysinline {
+ %1 = tail call float @llvm.sqrt.f32(float %v)
+ ret float %1
+}
+
+define float @_Z3powf(float %v1, float %v2) nounwind readnone alwaysinline {
+ %1 = tail call float @llvm.pow.f32(float %v1, float %v2)
+ ret float %1
+}
diff --git a/driver/runtime/matrix.ll b/driver/runtime/matrix.ll
new file mode 100644
index 0000000..c56405d
--- /dev/null
+++ b/driver/runtime/matrix.ll
@@ -0,0 +1,176 @@
+target datalayout = "e-p:32:32:32-i1:8:8-i8:8:8-i16:16:16-i32:32:32-i64:64:64-f32:32:32-f64:64:64-v64:64:64-v128:64:128-a0:0:64-n32-S64"
+target triple = "armv7-none-linux-gnueabi"
+
+
+%struct.rs_matrix4x4 = type { [16 x float] }
+%struct.rs_matrix3x3 = type { [9 x float] }
+%struct.rs_matrix2x2 = type { [4 x float] }
+
+define internal <4 x float> @smear_f(float %in) nounwind readnone alwaysinline {
+ %1 = insertelement <4 x float> undef, float %in, i32 0
+ %2 = insertelement <4 x float> %1, float %in, i32 1
+ %3 = insertelement <4 x float> %2, float %in, i32 2
+ %4 = insertelement <4 x float> %3, float %in, i32 3
+ ret <4 x float> %4
+}
+
+
+define <3 x float> @_Z16rsMatrixMultiplyPK12rs_matrix3x3Dv3_f(%struct.rs_matrix3x3* nocapture %m, <3 x float> %in) nounwind readonly {
+ %x0 = extractelement <3 x float> %in, i32 0
+ %x = tail call <4 x float> @smear_f(float %x0) nounwind readnone
+ %y0 = extractelement <3 x float> %in, i32 1
+ %y = tail call <4 x float> @smear_f(float %y0) nounwind readnone
+ %z0 = extractelement <3 x float> %in, i32 2
+ %z = tail call <4 x float> @smear_f(float %z0) nounwind readnone
+
+ %px = getelementptr inbounds %struct.rs_matrix3x3* %m, i32 0, i32 0, i32 0
+ %px2 = bitcast float* %px to <4 x float>*
+ %xm = load <4 x float>* %px2, align 4
+ %py = getelementptr inbounds %struct.rs_matrix3x3* %m, i32 0, i32 0, i32 3
+ %py2 = bitcast float* %py to <4 x float>*
+ %ym = load <4 x float>* %py2, align 4
+ %pz = getelementptr inbounds %struct.rs_matrix3x3* %m, i32 0, i32 0, i32 6
+ %pz2 = bitcast float* %pz to <3 x float>*
+ %zm2 = load <3 x float>* %pz2, align 4
+ %zm = shufflevector <3 x float> %zm2, <3 x float> undef, <4 x i32> <i32 0, i32 1, i32 2, i32 3>
+
+ %a1 = fmul <4 x float> %x, %xm
+ %a2 = fmul <4 x float> %y, %ym
+ %a3 = fadd <4 x float> %a1, %a2
+ %a4 = fmul <4 x float> %z, %zm
+ %a5 = fadd <4 x float> %a4, %a3
+ %a6 = shufflevector <4 x float> %a5, <4 x float> undef, <3 x i32> <i32 0, i32 1, i32 2>
+ ret <3 x float> %a6
+}
+
+define <3 x float> @_Z16rsMatrixMultiplyP12rs_matrix3x3Dv3_f(%struct.rs_matrix3x3* nocapture %m, <3 x float> %in) nounwind readonly {
+ %r = tail call <3 x float> @_Z16rsMatrixMultiplyPK12rs_matrix3x3Dv3_f(%struct.rs_matrix3x3* nocapture %m, <3 x float> %in) nounwind
+ ret <3 x float> %r
+}
+
+define <3 x float> @_Z16rsMatrixMultiplyPK12rs_matrix3x3Dv2_f(%struct.rs_matrix3x3* nocapture %m, <2 x float> %in) nounwind readonly {
+ %x0 = extractelement <2 x float> %in, i32 0
+ %x = tail call <4 x float> @smear_f(float %x0) nounwind readnone
+ %y0 = extractelement <2 x float> %in, i32 1
+ %y = tail call <4 x float> @smear_f(float %y0) nounwind readnone
+
+ %px = getelementptr inbounds %struct.rs_matrix3x3* %m, i32 0, i32 0, i32 0
+ %px2 = bitcast float* %px to <4 x float>*
+ %xm = load <4 x float>* %px2, align 4
+ %py = getelementptr inbounds %struct.rs_matrix3x3* %m, i32 0, i32 0, i32 3
+ %py2 = bitcast float* %py to <4 x float>*
+ %ym = load <4 x float>* %py2, align 4
+
+ %a1 = fmul <4 x float> %x, %xm
+ %a2 = fmul <4 x float> %y, %ym
+ %a3 = fadd <4 x float> %a1, %a2
+ %a4 = shufflevector <4 x float> %a3, <4 x float> undef, <3 x i32> <i32 0, i32 1, i32 2>
+ ret <3 x float> %a4
+}
+
+define <3 x float> @_Z16rsMatrixMultiplyP12rs_matrix3x3Dv2_f(%struct.rs_matrix3x3* nocapture %m, <2 x float> %in) nounwind readonly {
+ %r = tail call <3 x float> @_Z16rsMatrixMultiplyPK12rs_matrix3x3Dv2_f(%struct.rs_matrix3x3* nocapture %m, <2 x float> %in) nounwind
+ ret <3 x float> %r
+}
+
+define <4 x float> @_Z16rsMatrixMultiplyPK12rs_matrix4x4Dv4_f(%struct.rs_matrix4x4* nocapture %m, <4 x float> %in) nounwind readonly {
+ %x0 = extractelement <4 x float> %in, i32 0
+ %x = tail call <4 x float> @smear_f(float %x0) nounwind readnone
+ %y0 = extractelement <4 x float> %in, i32 1
+ %y = tail call <4 x float> @smear_f(float %y0) nounwind readnone
+ %z0 = extractelement <4 x float> %in, i32 2
+ %z = tail call <4 x float> @smear_f(float %z0) nounwind readnone
+ %w0 = extractelement <4 x float> %in, i32 3
+ %w = tail call <4 x float> @smear_f(float %w0) nounwind readnone
+
+ %px = getelementptr inbounds %struct.rs_matrix4x4* %m, i32 0, i32 0, i32 0
+ %px2 = bitcast float* %px to <4 x float>*
+ %xm = load <4 x float>* %px2, align 4
+ %py = getelementptr inbounds %struct.rs_matrix4x4* %m, i32 0, i32 0, i32 4
+ %py2 = bitcast float* %py to <4 x float>*
+ %ym = load <4 x float>* %py2, align 4
+ %pz = getelementptr inbounds %struct.rs_matrix4x4* %m, i32 0, i32 0, i32 8
+ %pz2 = bitcast float* %pz to <4 x float>*
+ %zm = load <4 x float>* %pz2, align 4
+ %pw = getelementptr inbounds %struct.rs_matrix4x4* %m, i32 0, i32 0, i32 12
+ %pw2 = bitcast float* %pw to <4 x float>*
+ %wm = load <4 x float>* %pw2, align 4
+
+ %a1 = fmul <4 x float> %x, %xm
+ %a2 = fmul <4 x float> %y, %ym
+ %a3 = fadd <4 x float> %a1, %a2
+ %a4 = fmul <4 x float> %z, %zm
+ %a5 = fadd <4 x float> %a3, %a4
+ %a6 = fmul <4 x float> %w, %wm
+ %a7 = fadd <4 x float> %a5, %a6
+ ret <4 x float> %a7
+}
+
+define <4 x float> @_Z16rsMatrixMultiplyP12rs_matrix4x4Dv4_f(%struct.rs_matrix4x4* nocapture %m, <4 x float> %in) nounwind readonly {
+ %r = tail call <4 x float> @_Z16rsMatrixMultiplyPK12rs_matrix4x4Dv4_f(%struct.rs_matrix4x4* nocapture %m, <4 x float> %in) nounwind
+ ret <4 x float> %r
+}
+
+define <4 x float> @_Z16rsMatrixMultiplyPK12rs_matrix4x4Dv3_f(%struct.rs_matrix4x4* nocapture %m, <3 x float> %in) nounwind readonly {
+ %x0 = extractelement <3 x float> %in, i32 0
+ %x = tail call <4 x float> @smear_f(float %x0) nounwind readnone
+ %y0 = extractelement <3 x float> %in, i32 1
+ %y = tail call <4 x float> @smear_f(float %y0) nounwind readnone
+ %z0 = extractelement <3 x float> %in, i32 2
+ %z = tail call <4 x float> @smear_f(float %z0) nounwind readnone
+
+ %px = getelementptr inbounds %struct.rs_matrix4x4* %m, i32 0, i32 0, i32 0
+ %px2 = bitcast float* %px to <4 x float>*
+ %xm = load <4 x float>* %px2, align 4
+ %py = getelementptr inbounds %struct.rs_matrix4x4* %m, i32 0, i32 0, i32 4
+ %py2 = bitcast float* %py to <4 x float>*
+ %ym = load <4 x float>* %py2, align 4
+ %pz = getelementptr inbounds %struct.rs_matrix4x4* %m, i32 0, i32 0, i32 8
+ %pz2 = bitcast float* %pz to <4 x float>*
+ %zm = load <4 x float>* %pz2, align 4
+ %pw = getelementptr inbounds %struct.rs_matrix4x4* %m, i32 0, i32 0, i32 12
+ %pw2 = bitcast float* %pw to <4 x float>*
+ %wm = load <4 x float>* %pw2, align 4
+
+ %a1 = fmul <4 x float> %x, %xm
+ %a2 = fadd <4 x float> %wm, %a1
+ %a3 = fmul <4 x float> %y, %ym
+ %a4 = fadd <4 x float> %a2, %a3
+ %a5 = fmul <4 x float> %z, %zm
+ %a6 = fadd <4 x float> %a4, %a5
+ ret <4 x float> %a6
+}
+
+define <4 x float> @_Z16rsMatrixMultiplyP12rs_matrix4x4Dv3_f(%struct.rs_matrix4x4* nocapture %m, <3 x float> %in) nounwind readonly {
+ %r = tail call <4 x float> @_Z16rsMatrixMultiplyPK12rs_matrix4x4Dv3_f(%struct.rs_matrix4x4* nocapture %m, <3 x float> %in) nounwind
+ ret <4 x float> %r
+}
+
+define <4 x float> @_Z16rsMatrixMultiplyPK12rs_matrix4x4Dv2_f(%struct.rs_matrix4x4* nocapture %m, <2 x float> %in) nounwind readonly {
+ %x0 = extractelement <2 x float> %in, i32 0
+ %x = tail call <4 x float> @smear_f(float %x0) nounwind readnone
+ %y0 = extractelement <2 x float> %in, i32 1
+ %y = tail call <4 x float> @smear_f(float %y0) nounwind readnone
+
+ %px = getelementptr inbounds %struct.rs_matrix4x4* %m, i32 0, i32 0, i32 0
+ %px2 = bitcast float* %px to <4 x float>*
+ %xm = load <4 x float>* %px2, align 4
+ %py = getelementptr inbounds %struct.rs_matrix4x4* %m, i32 0, i32 0, i32 4
+ %py2 = bitcast float* %py to <4 x float>*
+ %ym = load <4 x float>* %py2, align 4
+ %pw = getelementptr inbounds %struct.rs_matrix4x4* %m, i32 0, i32 0, i32 12
+ %pw2 = bitcast float* %pw to <4 x float>*
+ %wm = load <4 x float>* %pw2, align 4
+
+ %a1 = fmul <4 x float> %x, %xm
+ %a2 = fadd <4 x float> %wm, %a1
+ %a3 = fmul <4 x float> %y, %ym
+ %a4 = fadd <4 x float> %a2, %a3
+ ret <4 x float> %a4
+}
+
+define <4 x float> @_Z16rsMatrixMultiplyP12rs_matrix4x4Dv2_f(%struct.rs_matrix4x4* nocapture %m, <2 x float> %in) nounwind readonly {
+ %r = tail call <4 x float> @_Z16rsMatrixMultiplyPK12rs_matrix4x4Dv2_f(%struct.rs_matrix4x4* nocapture %m, <2 x float> %in) nounwind
+ ret <4 x float> %r
+}
+
diff --git a/driver/runtime/rsClamp.ll b/driver/runtime/rsClamp.ll
new file mode 100644
index 0000000..eba678a
--- /dev/null
+++ b/driver/runtime/rsClamp.ll
@@ -0,0 +1,60 @@
+target datalayout = "e-p:32:32:32-i1:8:8-i8:8:8-i16:16:16-i32:32:32-i64:64:64-f32:32:32-f64:64:64-v64:64:64-v128:64:128-a0:0:64-n32-S64"
+target triple = "armv7-none-linux-gnueabi"
+
+
+define float @_Z7rsClampfff(float %value, float %low, float %high) nounwind readonly {
+ %1 = fcmp olt float %value, %high
+ %2 = select i1 %1, float %value, float %high
+ %3 = fcmp ogt float %2, %low
+ %4 = select i1 %3, float %2, float %low
+ ret float %4
+}
+
+define signext i8 @_Z7rsClampccc(i8 signext %value, i8 signext %low, i8 signext %high) nounwind readonly {
+ %1 = icmp slt i8 %value, %high
+ %2 = select i1 %1, i8 %value, i8 %high
+ %3 = icmp sgt i8 %2, %low
+ %4 = select i1 %3, i8 %2, i8 %low
+ ret i8 %4
+}
+
+define zeroext i8 @_Z7rsClamphhh(i8 zeroext %value, i8 zeroext %low, i8 zeroext %high) nounwind readonly {
+ %1 = icmp ult i8 %value, %high
+ %2 = select i1 %1, i8 %value, i8 %high
+ %3 = icmp ugt i8 %2, %low
+ %4 = select i1 %3, i8 %2, i8 %low
+ ret i8 %4
+}
+
+define signext i16 @_Z7rsClampsss(i16 signext %value, i16 signext %low, i16 signext %high) nounwind readonly {
+ %1 = icmp slt i16 %value, %high
+ %2 = select i1 %1, i16 %value, i16 %high
+ %3 = icmp sgt i16 %2, %low
+ %4 = select i1 %3, i16 %2, i16 %low
+ ret i16 %4
+}
+
+define zeroext i16 @_Z7rsClampttt(i16 zeroext %value, i16 zeroext %low, i16 zeroext %high) nounwind readonly {
+ %1 = icmp ult i16 %value, %high
+ %2 = select i1 %1, i16 %value, i16 %high
+ %3 = icmp ugt i16 %2, %low
+ %4 = select i1 %3, i16 %2, i16 %low
+ ret i16 %4
+}
+
+define i32 @_Z7rsClampiii(i32 %value, i32 %low, i32 %high) nounwind readonly {
+ %1 = icmp slt i32 %value, %high
+ %2 = select i1 %1, i32 %value, i32 %high
+ %3 = icmp sgt i32 %2, %low
+ %4 = select i1 %3, i32 %2, i32 %low
+ ret i32 %4
+}
+
+define i32 @_Z7rsClampjjj(i32 %value, i32 %low, i32 %high) nounwind readonly {
+ %1 = icmp ult i32 %value, %high
+ %2 = select i1 %1, i32 %value, i32 %high
+ %3 = icmp ugt i32 %2, %low
+ %4 = select i1 %3, i32 %2, i32 %low
+ ret i32 %4
+}
+
diff --git a/driver/runtime/rs_allocation.c b/driver/runtime/rs_allocation.c
new file mode 100644
index 0000000..198e9af
--- /dev/null
+++ b/driver/runtime/rs_allocation.c
@@ -0,0 +1,318 @@
+#include "rs_core.rsh"
+#include "rs_graphics.rsh"
+#include "rs_structs.h"
+
+// Opaque Allocation type operations
+extern uint32_t __attribute__((overloadable))
+ rsAllocationGetDimX(rs_allocation a) {
+ Allocation_t *alloc = (Allocation_t *)a.p;
+ return alloc->mHal.drvState.lod[0].dimX;
+}
+
+extern uint32_t __attribute__((overloadable))
+ rsAllocationGetDimY(rs_allocation a) {
+ Allocation_t *alloc = (Allocation_t *)a.p;
+ return alloc->mHal.drvState.lod[0].dimY;
+}
+
+extern uint32_t __attribute__((overloadable))
+ rsAllocationGetDimZ(rs_allocation a) {
+ Allocation_t *alloc = (Allocation_t *)a.p;
+ return alloc->mHal.drvState.lod[0].dimZ;
+}
+
+extern uint32_t __attribute__((overloadable))
+ rsAllocationGetDimLOD(rs_allocation a) {
+ Allocation_t *alloc = (Allocation_t *)a.p;
+ return alloc->mHal.state.hasMipmaps;
+}
+
+extern uint32_t __attribute__((overloadable))
+ rsAllocationGetDimFaces(rs_allocation a) {
+ Allocation_t *alloc = (Allocation_t *)a.p;
+ return alloc->mHal.state.hasFaces;
+}
+
+
+extern rs_element __attribute__((overloadable))
+ rsAllocationGetElement(rs_allocation a) {
+ Allocation_t *alloc = (Allocation_t *)a.p;
+ if (alloc == NULL) {
+ rs_element nullElem = {0};
+ return nullElem;
+ }
+ Type_t *type = (Type_t *)alloc->mHal.state.type;
+ rs_element returnElem = {type->mHal.state.element};
+ return returnElem;
+}
+
+// TODO: this needs to be optimized, obviously
+static void memcpy(void* dst, void* src, size_t size) {
+ char* dst_c = (char*) dst, *src_c = (char*) src;
+ for (; size > 0; size--) {
+ *dst_c++ = *src_c++;
+ }
+}
+
+#ifdef RS_DEBUG_RUNTIME
+#define ELEMENT_AT(T) \
+ extern void __attribute__((overloadable)) \
+ rsSetElementAt_##T(rs_allocation a, const T *val, uint32_t x); \
+ extern void __attribute__((overloadable)) \
+ rsSetElementAt_##T(rs_allocation a, const T *val, uint32_t x, uint32_t y); \
+ extern void __attribute__((overloadable)) \
+ rsSetElementAt_##T(rs_allocation a, const T *val, uint32_t x, uint32_t y, uint32_t z); \
+ extern void __attribute__((overloadable)) \
+ rsGetElementAt_##T(rs_allocation a, T *val, uint32_t x); \
+ extern void __attribute__((overloadable)) \
+ rsGetElementAt_##T(rs_allocation a, T *val, uint32_t x, uint32_t y); \
+ extern void __attribute__((overloadable)) \
+ rsGetElementAt_##T(rs_allocation a, T *val, uint32_t x, uint32_t y, uint32_t z); \
+ \
+ extern void __attribute__((overloadable)) \
+ rsSetElementAt_##T(rs_allocation a, T val, uint32_t x) { \
+ rsSetElementAt_##T(a, &val, x); \
+ } \
+ extern void __attribute__((overloadable)) \
+ rsSetElementAt_##T(rs_allocation a, T val, uint32_t x, uint32_t y) { \
+ rsSetElementAt_##T(a, &val, x, y); \
+ } \
+ extern void __attribute__((overloadable)) \
+ rsSetElementAt_##T(rs_allocation a, T val, uint32_t x, uint32_t y, uint32_t z) { \
+ rsSetElementAt_##T(a, &val, x, y, z); \
+ } \
+ extern T __attribute__((overloadable)) \
+ rsGetElementAt_##T(rs_allocation a, uint32_t x) { \
+ T tmp; \
+ rsGetElementAt_##T(a, &tmp, x); \
+ return tmp; \
+ } \
+ extern T __attribute__((overloadable)) \
+ rsGetElementAt_##T(rs_allocation a, uint32_t x, uint32_t y) { \
+ T tmp; \
+ rsGetElementAt_##T(a, &tmp, x, y); \
+ return tmp; \
+ } \
+ extern T __attribute__((overloadable)) \
+ rsGetElementAt_##T(rs_allocation a, uint32_t x, uint32_t y, uint32_t z) { \
+ T tmp; \
+ rsGetElementAt_##T(a, &tmp, x, y, z); \
+ return tmp; \
+ }
+
+#else
+
+uint8_t*
+rsOffset(rs_allocation a, uint32_t sizeOf, uint32_t x, uint32_t y,
+ uint32_t z) {
+ Allocation_t *alloc = (Allocation_t *)a.p;
+ uint8_t *p = (uint8_t *)alloc->mHal.drvState.lod[0].mallocPtr;
+ const uint32_t stride = alloc->mHal.drvState.lod[0].stride;
+ const uint32_t dimY = alloc->mHal.drvState.lod[0].dimY;
+ uint8_t *dp = &p[(sizeOf * x) + (y * stride) +
+ (z * stride * dimY)];
+ return dp;
+}
+
+#define ELEMENT_AT(T) \
+ \
+ void \
+ rsSetElementAtImpl_##T(rs_allocation a, T val, uint32_t x, \
+ uint32_t y, uint32_t z); \
+ \
+ extern void __attribute__((overloadable)) \
+ rsSetElementAt_##T(rs_allocation a, T val, uint32_t x) { \
+ rsSetElementAtImpl_##T(a, val, x, 0, 0); \
+ } \
+ \
+ extern void __attribute__((overloadable)) \
+ rsSetElementAt_##T(rs_allocation a, T val, uint32_t x, \
+ uint32_t y) { \
+ rsSetElementAtImpl_##T(a, val, x, y, 0); \
+ } \
+ \
+ extern void __attribute__((overloadable)) \
+ rsSetElementAt_##T(rs_allocation a, T val, uint32_t x, uint32_t y, \
+ uint32_t z) { \
+ rsSetElementAtImpl_##T(a, val, x, y, z); \
+ } \
+ \
+ T \
+ rsGetElementAtImpl_##T(rs_allocation a, uint32_t x, uint32_t y, \
+ uint32_t z); \
+ \
+ extern T __attribute__((overloadable)) \
+ rsGetElementAt_##T(rs_allocation a, uint32_t x) { \
+ return rsGetElementAtImpl_##T(a, x, 0, 0); \
+ } \
+ \
+ extern T __attribute__((overloadable)) \
+ rsGetElementAt_##T(rs_allocation a, uint32_t x, uint32_t y) { \
+ return rsGetElementAtImpl_##T(a, x, y, 0); \
+ } \
+ \
+ extern T __attribute__((overloadable)) \
+ rsGetElementAt_##T(rs_allocation a, uint32_t x, uint32_t y, \
+ uint32_t z) { \
+ return rsGetElementAtImpl_##T(a, x, y, z); \
+ }
+
+
+
+extern const void * __attribute__((overloadable))
+ rsGetElementAt(rs_allocation a, uint32_t x) {
+ Allocation_t *alloc = (Allocation_t *)a.p;
+ const uint8_t *p = (const uint8_t *)alloc->mHal.drvState.lod[0].mallocPtr;
+ const uint32_t eSize = alloc->mHal.state.elementSizeBytes;
+ return &p[eSize * x];
+}
+
+extern const void * __attribute__((overloadable))
+ rsGetElementAt(rs_allocation a, uint32_t x, uint32_t y) {
+ Allocation_t *alloc = (Allocation_t *)a.p;
+ const uint8_t *p = (const uint8_t *)alloc->mHal.drvState.lod[0].mallocPtr;
+ const uint32_t eSize = alloc->mHal.state.elementSizeBytes;
+ const uint32_t stride = alloc->mHal.drvState.lod[0].stride;
+ return &p[(eSize * x) + (y * stride)];
+}
+
+extern const void * __attribute__((overloadable))
+ rsGetElementAt(rs_allocation a, uint32_t x, uint32_t y, uint32_t z) {
+ Allocation_t *alloc = (Allocation_t *)a.p;
+ const uint8_t *p = (const uint8_t *)alloc->mHal.drvState.lod[0].mallocPtr;
+ const uint32_t eSize = alloc->mHal.state.elementSizeBytes;
+ const uint32_t stride = alloc->mHal.drvState.lod[0].stride;
+ const uint32_t dimY = alloc->mHal.drvState.lod[0].dimY;
+ return &p[(eSize * x) + (y * stride) + (z * stride * dimY)];
+}
+extern void __attribute__((overloadable))
+ rsSetElementAt(rs_allocation a, void* ptr, uint32_t x) {
+ Allocation_t *alloc = (Allocation_t *)a.p;
+ const uint8_t *p = (const uint8_t *)alloc->mHal.drvState.lod[0].mallocPtr;
+ const uint32_t eSize = alloc->mHal.state.elementSizeBytes;
+ memcpy((void*)&p[eSize * x], ptr, eSize);
+}
+
+extern void __attribute__((overloadable))
+ rsSetElementAt(rs_allocation a, void* ptr, uint32_t x, uint32_t y) {
+ Allocation_t *alloc = (Allocation_t *)a.p;
+ const uint8_t *p = (const uint8_t *)alloc->mHal.drvState.lod[0].mallocPtr;
+ const uint32_t eSize = alloc->mHal.state.elementSizeBytes;
+ const uint32_t stride = alloc->mHal.drvState.lod[0].stride;
+ memcpy((void*)&p[(eSize * x) + (y * stride)], ptr, eSize);
+}
+
+extern void __attribute__((overloadable))
+ rsSetElementAt(rs_allocation a, void* ptr, uint32_t x, uint32_t y, uint32_t z) {
+ Allocation_t *alloc = (Allocation_t *)a.p;
+ const uint8_t *p = (const uint8_t *)alloc->mHal.drvState.lod[0].mallocPtr;
+ const uint32_t eSize = alloc->mHal.state.elementSizeBytes;
+ const uint32_t stride = alloc->mHal.drvState.lod[0].stride;
+ const uint32_t dimY = alloc->mHal.drvState.lod[0].dimY;
+ memcpy((void*)&p[(eSize * x) + (y * stride) + (z * stride * dimY)], ptr, eSize);
+}
+#endif
+
+ELEMENT_AT(char)
+ELEMENT_AT(char2)
+ELEMENT_AT(char3)
+ELEMENT_AT(char4)
+ELEMENT_AT(uchar)
+ELEMENT_AT(uchar2)
+ELEMENT_AT(uchar3)
+ELEMENT_AT(uchar4)
+ELEMENT_AT(short)
+ELEMENT_AT(short2)
+ELEMENT_AT(short3)
+ELEMENT_AT(short4)
+ELEMENT_AT(ushort)
+ELEMENT_AT(ushort2)
+ELEMENT_AT(ushort3)
+ELEMENT_AT(ushort4)
+ELEMENT_AT(int)
+ELEMENT_AT(int2)
+ELEMENT_AT(int3)
+ELEMENT_AT(int4)
+ELEMENT_AT(uint)
+ELEMENT_AT(uint2)
+ELEMENT_AT(uint3)
+ELEMENT_AT(uint4)
+ELEMENT_AT(long)
+ELEMENT_AT(long2)
+ELEMENT_AT(long3)
+ELEMENT_AT(long4)
+ELEMENT_AT(ulong)
+ELEMENT_AT(ulong2)
+ELEMENT_AT(ulong3)
+ELEMENT_AT(ulong4)
+ELEMENT_AT(float)
+ELEMENT_AT(float2)
+ELEMENT_AT(float3)
+ELEMENT_AT(float4)
+ELEMENT_AT(double)
+ELEMENT_AT(double2)
+ELEMENT_AT(double3)
+ELEMENT_AT(double4)
+
+#undef ELEMENT_AT
+
+
+extern const uchar __attribute__((overloadable))
+ rsGetElementAtYuv_uchar_Y(rs_allocation a, uint32_t x, uint32_t y) {
+ return rsGetElementAt_uchar(a, x, y);
+}
+
+extern const uchar __attribute__((overloadable))
+ rsGetElementAtYuv_uchar_U(rs_allocation a, uint32_t x, uint32_t y) {
+
+ Allocation_t *alloc = (Allocation_t *)a.p;
+ const uint32_t yuvID = alloc->mHal.state.yuv;
+ const uint8_t *p = (const uint8_t *)alloc->mHal.drvState.lod[1].mallocPtr;
+ const uint32_t stride = alloc->mHal.drvState.lod[1].stride;
+
+ switch(yuvID) {
+ case 0x32315659: //HAL_PIXEL_FORMAT_YV12:
+ x >>= 1;
+ y >>= 1;
+ return p[x + (y * stride)];
+ case 11: //HAL_PIXEL_FORMAT_YCrCb_420_SP: // NV21
+ x >>= 1;
+ y >>= 1;
+ return p[(x<<1) + (y * stride)];
+ default:
+ break;
+ }
+
+ return 0;
+}
+
+extern const uchar __attribute__((overloadable))
+ rsGetElementAtYuv_uchar_V(rs_allocation a, uint32_t x, uint32_t y) {
+
+ Allocation_t *alloc = (Allocation_t *)a.p;
+ const uint32_t yuvID = alloc->mHal.state.yuv;
+
+ switch(yuvID) {
+ case 0x32315659: //HAL_PIXEL_FORMAT_YV12:
+ {
+ const uint8_t *p = (const uint8_t *)alloc->mHal.drvState.lod[2].mallocPtr;
+ const uint32_t stride = alloc->mHal.drvState.lod[2].stride;
+ x >>= 1;
+ y >>= 1;
+ return p[x + (y * stride)];
+ }
+ case 11: //HAL_PIXEL_FORMAT_YCrCb_420_SP: // NV21
+ {
+ const uint8_t *p = (const uint8_t *)alloc->mHal.drvState.lod[1].mallocPtr;
+ const uint32_t stride = alloc->mHal.drvState.lod[1].stride;
+ x >>= 1;
+ y >>= 1;
+ return p[(x<<1) + (y * stride) + 1];
+ }
+ default:
+ break;
+ }
+
+ return 0;
+}
+
diff --git a/driver/runtime/rs_cl.c b/driver/runtime/rs_cl.c
new file mode 100755
index 0000000..b7f9158
--- /dev/null
+++ b/driver/runtime/rs_cl.c
@@ -0,0 +1,1194 @@
+#include "rs_types.rsh"
+
+extern float2 __attribute__((overloadable)) convert_float2(int2 c);
+extern float3 __attribute__((overloadable)) convert_float3(int3 c);
+extern float4 __attribute__((overloadable)) convert_float4(int4 c);
+
+extern int2 __attribute__((overloadable)) convert_int2(float2 c);
+extern int3 __attribute__((overloadable)) convert_int3(float3 c);
+extern int4 __attribute__((overloadable)) convert_int4(float4 c);
+
+
+extern float __attribute__((overloadable)) fmin(float v, float v2);
+extern float2 __attribute__((overloadable)) fmin(float2 v, float v2);
+extern float3 __attribute__((overloadable)) fmin(float3 v, float v2);
+extern float4 __attribute__((overloadable)) fmin(float4 v, float v2);
+
+extern float __attribute__((overloadable)) fmax(float v, float v2);
+extern float2 __attribute__((overloadable)) fmax(float2 v, float v2);
+extern float3 __attribute__((overloadable)) fmax(float3 v, float v2);
+extern float4 __attribute__((overloadable)) fmax(float4 v, float v2);
+
+// Float ops, 6.11.2
+
+#define FN_FUNC_FN(fnc) \
+extern float2 __attribute__((overloadable)) fnc(float2 v) { \
+ float2 r; \
+ r.x = fnc(v.x); \
+ r.y = fnc(v.y); \
+ return r; \
+} \
+extern float3 __attribute__((overloadable)) fnc(float3 v) { \
+ float3 r; \
+ r.x = fnc(v.x); \
+ r.y = fnc(v.y); \
+ r.z = fnc(v.z); \
+ return r; \
+} \
+extern float4 __attribute__((overloadable)) fnc(float4 v) { \
+ float4 r; \
+ r.x = fnc(v.x); \
+ r.y = fnc(v.y); \
+ r.z = fnc(v.z); \
+ r.w = fnc(v.w); \
+ return r; \
+}
+
+#define IN_FUNC_FN(fnc) \
+extern int2 __attribute__((overloadable)) fnc(float2 v) { \
+ int2 r; \
+ r.x = fnc(v.x); \
+ r.y = fnc(v.y); \
+ return r; \
+} \
+extern int3 __attribute__((overloadable)) fnc(float3 v) { \
+ int3 r; \
+ r.x = fnc(v.x); \
+ r.y = fnc(v.y); \
+ r.z = fnc(v.z); \
+ return r; \
+} \
+extern int4 __attribute__((overloadable)) fnc(float4 v) { \
+ int4 r; \
+ r.x = fnc(v.x); \
+ r.y = fnc(v.y); \
+ r.z = fnc(v.z); \
+ r.w = fnc(v.w); \
+ return r; \
+}
+
+#define FN_FUNC_FN_FN(fnc) \
+extern float2 __attribute__((overloadable)) fnc(float2 v1, float2 v2) { \
+ float2 r; \
+ r.x = fnc(v1.x, v2.x); \
+ r.y = fnc(v1.y, v2.y); \
+ return r; \
+} \
+extern float3 __attribute__((overloadable)) fnc(float3 v1, float3 v2) { \
+ float3 r; \
+ r.x = fnc(v1.x, v2.x); \
+ r.y = fnc(v1.y, v2.y); \
+ r.z = fnc(v1.z, v2.z); \
+ return r; \
+} \
+extern float4 __attribute__((overloadable)) fnc(float4 v1, float4 v2) { \
+ float4 r; \
+ r.x = fnc(v1.x, v2.x); \
+ r.y = fnc(v1.y, v2.y); \
+ r.z = fnc(v1.z, v2.z); \
+ r.w = fnc(v1.w, v2.w); \
+ return r; \
+}
+
+#define FN_FUNC_FN_F(fnc) \
+extern float2 __attribute__((overloadable)) fnc(float2 v1, float v2) { \
+ float2 r; \
+ r.x = fnc(v1.x, v2); \
+ r.y = fnc(v1.y, v2); \
+ return r; \
+} \
+extern float3 __attribute__((overloadable)) fnc(float3 v1, float v2) { \
+ float3 r; \
+ r.x = fnc(v1.x, v2); \
+ r.y = fnc(v1.y, v2); \
+ r.z = fnc(v1.z, v2); \
+ return r; \
+} \
+extern float4 __attribute__((overloadable)) fnc(float4 v1, float v2) { \
+ float4 r; \
+ r.x = fnc(v1.x, v2); \
+ r.y = fnc(v1.y, v2); \
+ r.z = fnc(v1.z, v2); \
+ r.w = fnc(v1.w, v2); \
+ return r; \
+}
+
+#define FN_FUNC_FN_IN(fnc) \
+extern float2 __attribute__((overloadable)) fnc(float2 v1, int2 v2) { \
+ float2 r; \
+ r.x = fnc(v1.x, v2.x); \
+ r.y = fnc(v1.y, v2.y); \
+ return r; \
+} \
+extern float3 __attribute__((overloadable)) fnc(float3 v1, int3 v2) { \
+ float3 r; \
+ r.x = fnc(v1.x, v2.x); \
+ r.y = fnc(v1.y, v2.y); \
+ r.z = fnc(v1.z, v2.z); \
+ return r; \
+} \
+extern float4 __attribute__((overloadable)) fnc(float4 v1, int4 v2) { \
+ float4 r; \
+ r.x = fnc(v1.x, v2.x); \
+ r.y = fnc(v1.y, v2.y); \
+ r.z = fnc(v1.z, v2.z); \
+ r.w = fnc(v1.w, v2.w); \
+ return r; \
+}
+
+#define FN_FUNC_FN_I(fnc) \
+extern float2 __attribute__((overloadable)) fnc(float2 v1, int v2) { \
+ float2 r; \
+ r.x = fnc(v1.x, v2); \
+ r.y = fnc(v1.y, v2); \
+ return r; \
+} \
+extern float3 __attribute__((overloadable)) fnc(float3 v1, int v2) { \
+ float3 r; \
+ r.x = fnc(v1.x, v2); \
+ r.y = fnc(v1.y, v2); \
+ r.z = fnc(v1.z, v2); \
+ return r; \
+} \
+extern float4 __attribute__((overloadable)) fnc(float4 v1, int v2) { \
+ float4 r; \
+ r.x = fnc(v1.x, v2); \
+ r.y = fnc(v1.y, v2); \
+ r.z = fnc(v1.z, v2); \
+ r.w = fnc(v1.w, v2); \
+ return r; \
+}
+
+#define FN_FUNC_FN_PFN(fnc) \
+extern float2 __attribute__((overloadable)) \
+ fnc(float2 v1, float2 *v2) { \
+ float2 r; \
+ float t[2]; \
+ r.x = fnc(v1.x, &t[0]); \
+ r.y = fnc(v1.y, &t[1]); \
+ v2->x = t[0]; \
+ v2->y = t[1]; \
+ return r; \
+} \
+extern float3 __attribute__((overloadable)) \
+ fnc(float3 v1, float3 *v2) { \
+ float3 r; \
+ float t[3]; \
+ r.x = fnc(v1.x, &t[0]); \
+ r.y = fnc(v1.y, &t[1]); \
+ r.z = fnc(v1.z, &t[2]); \
+ v2->x = t[0]; \
+ v2->y = t[1]; \
+ v2->z = t[2]; \
+ return r; \
+} \
+extern float4 __attribute__((overloadable)) \
+ fnc(float4 v1, float4 *v2) { \
+ float4 r; \
+ float t[4]; \
+ r.x = fnc(v1.x, &t[0]); \
+ r.y = fnc(v1.y, &t[1]); \
+ r.z = fnc(v1.z, &t[2]); \
+ r.w = fnc(v1.w, &t[3]); \
+ v2->x = t[0]; \
+ v2->y = t[1]; \
+ v2->z = t[2]; \
+ v2->w = t[3]; \
+ return r; \
+}
+
+#define FN_FUNC_FN_PIN(fnc) \
+extern float2 __attribute__((overloadable)) fnc(float2 v1, int2 *v2) { \
+ float2 r; \
+ int t[2]; \
+ r.x = fnc(v1.x, &t[0]); \
+ r.y = fnc(v1.y, &t[1]); \
+ v2->x = t[0]; \
+ v2->y = t[1]; \
+ return r; \
+} \
+extern float3 __attribute__((overloadable)) fnc(float3 v1, int3 *v2) { \
+ float3 r; \
+ int t[3]; \
+ r.x = fnc(v1.x, &t[0]); \
+ r.y = fnc(v1.y, &t[1]); \
+ r.z = fnc(v1.z, &t[2]); \
+ v2->x = t[0]; \
+ v2->y = t[1]; \
+ v2->z = t[2]; \
+ return r; \
+} \
+extern float4 __attribute__((overloadable)) fnc(float4 v1, int4 *v2) { \
+ float4 r; \
+ int t[4]; \
+ r.x = fnc(v1.x, &t[0]); \
+ r.y = fnc(v1.y, &t[1]); \
+ r.z = fnc(v1.z, &t[2]); \
+ r.w = fnc(v1.w, &t[3]); \
+ v2->x = t[0]; \
+ v2->y = t[1]; \
+ v2->z = t[2]; \
+ v2->w = t[3]; \
+ return r; \
+}
+
+#define FN_FUNC_FN_FN_FN(fnc) \
+extern float2 __attribute__((overloadable)) \
+ fnc(float2 v1, float2 v2, float2 v3) { \
+ float2 r; \
+ r.x = fnc(v1.x, v2.x, v3.x); \
+ r.y = fnc(v1.y, v2.y, v3.y); \
+ return r; \
+} \
+extern float3 __attribute__((overloadable)) \
+ fnc(float3 v1, float3 v2, float3 v3) { \
+ float3 r; \
+ r.x = fnc(v1.x, v2.x, v3.x); \
+ r.y = fnc(v1.y, v2.y, v3.y); \
+ r.z = fnc(v1.z, v2.z, v3.z); \
+ return r; \
+} \
+extern float4 __attribute__((overloadable)) \
+ fnc(float4 v1, float4 v2, float4 v3) { \
+ float4 r; \
+ r.x = fnc(v1.x, v2.x, v3.x); \
+ r.y = fnc(v1.y, v2.y, v3.y); \
+ r.z = fnc(v1.z, v2.z, v3.z); \
+ r.w = fnc(v1.w, v2.w, v3.w); \
+ return r; \
+}
+
+#define FN_FUNC_FN_FN_PIN(fnc) \
+extern float2 __attribute__((overloadable)) \
+ fnc(float2 v1, float2 v2, int2 *v3) { \
+ float2 r; \
+ int t[2]; \
+ r.x = fnc(v1.x, v2.x, &t[0]); \
+ r.y = fnc(v1.y, v2.y, &t[1]); \
+ v3->x = t[0]; \
+ v3->y = t[1]; \
+ return r; \
+} \
+extern float3 __attribute__((overloadable)) \
+ fnc(float3 v1, float3 v2, int3 *v3) { \
+ float3 r; \
+ int t[3]; \
+ r.x = fnc(v1.x, v2.x, &t[0]); \
+ r.y = fnc(v1.y, v2.y, &t[1]); \
+ r.z = fnc(v1.z, v2.z, &t[2]); \
+ v3->x = t[0]; \
+ v3->y = t[1]; \
+ v3->z = t[2]; \
+ return r; \
+} \
+extern float4 __attribute__((overloadable)) \
+ fnc(float4 v1, float4 v2, int4 *v3) { \
+ float4 r; \
+ int t[4]; \
+ r.x = fnc(v1.x, v2.x, &t[0]); \
+ r.y = fnc(v1.y, v2.y, &t[1]); \
+ r.z = fnc(v1.z, v2.z, &t[2]); \
+ r.w = fnc(v1.w, v2.w, &t[3]); \
+ v3->x = t[0]; \
+ v3->y = t[1]; \
+ v3->z = t[2]; \
+ v3->w = t[3]; \
+ return r; \
+}
+
+static const int iposinf = 0x7f800000;
+static const int ineginf = 0xff800000;
+
+static const float posinf() {
+ float f = *((float*)&iposinf);
+ return f;
+}
+
+static const float neginf() {
+ float f = *((float*)&ineginf);
+ return f;
+}
+
+static bool isinf(float f) {
+ int i = *((int*)(void*)&f);
+ return (i == iposinf) || (i == ineginf);
+}
+
+static bool isnan(float f) {
+ int i = *((int*)(void*)&f);
+ return (((i & 0x7f800000) == 0x7f800000) && (i & 0x007fffff));
+}
+
+static bool isposzero(float f) {
+ int i = *((int*)(void*)&f);
+ return (i == 0x00000000);
+}
+
+static bool isnegzero(float f) {
+ int i = *((int*)(void*)&f);
+ return (i == 0x80000000);
+}
+
+static bool iszero(float f) {
+ return isposzero(f) || isnegzero(f);
+}
+
+
+extern float __attribute__((overloadable)) acos(float);
+FN_FUNC_FN(acos)
+
+extern float __attribute__((overloadable)) acosh(float);
+FN_FUNC_FN(acosh)
+
+
+extern float __attribute__((overloadable)) acospi(float v) {
+ return acos(v) / M_PI;
+}
+FN_FUNC_FN(acospi)
+
+extern float __attribute__((overloadable)) asin(float);
+FN_FUNC_FN(asin)
+
+extern float __attribute__((overloadable)) asinh(float);
+FN_FUNC_FN(asinh)
+
+extern float __attribute__((overloadable)) asinpi(float v) {
+ return asin(v) / M_PI;
+}
+FN_FUNC_FN(asinpi)
+
+extern float __attribute__((overloadable)) atan(float);
+FN_FUNC_FN(atan)
+
+extern float __attribute__((overloadable)) atan2(float, float);
+FN_FUNC_FN_FN(atan2)
+
+extern float __attribute__((overloadable)) atanh(float);
+FN_FUNC_FN(atanh)
+
+extern float __attribute__((overloadable)) atanpi(float v) {
+ return atan(v) / M_PI;
+}
+FN_FUNC_FN(atanpi)
+
+
+extern float __attribute__((overloadable)) atan2pi(float y, float x) {
+ return atan2(y, x) / M_PI;
+}
+FN_FUNC_FN_FN(atan2pi)
+
+extern float __attribute__((overloadable)) cbrt(float);
+FN_FUNC_FN(cbrt)
+
+extern float __attribute__((overloadable)) ceil(float);
+FN_FUNC_FN(ceil)
+
+extern float __attribute__((overloadable)) copysign(float, float);
+FN_FUNC_FN_FN(copysign)
+
+extern float __attribute__((overloadable)) cos(float);
+FN_FUNC_FN(cos)
+
+extern float __attribute__((overloadable)) cosh(float);
+FN_FUNC_FN(cosh)
+
+extern float __attribute__((overloadable)) cospi(float v) {
+ return cos(v * M_PI);
+}
+FN_FUNC_FN(cospi)
+
+extern float __attribute__((overloadable)) erfc(float);
+FN_FUNC_FN(erfc)
+
+extern float __attribute__((overloadable)) erf(float);
+FN_FUNC_FN(erf)
+
+extern float __attribute__((overloadable)) exp(float);
+FN_FUNC_FN(exp)
+
+extern float __attribute__((overloadable)) exp2(float);
+FN_FUNC_FN(exp2)
+
+extern float __attribute__((overloadable)) pow(float, float);
+
+extern float __attribute__((overloadable)) exp10(float v) {
+ return exp2(v * 3.321928095f);
+}
+FN_FUNC_FN(exp10)
+
+extern float __attribute__((overloadable)) expm1(float);
+FN_FUNC_FN(expm1)
+
+extern float __attribute__((overloadable)) fabs(float v) {
+ int i = *((int*)(void*)&v) & 0x7fffffff;
+ return *((float*)(void*)&i);
+}
+FN_FUNC_FN(fabs)
+
+extern float __attribute__((overloadable)) fdim(float, float);
+FN_FUNC_FN_FN(fdim)
+
+extern float __attribute__((overloadable)) floor(float);
+FN_FUNC_FN(floor)
+
+extern float __attribute__((overloadable)) fma(float, float, float);
+FN_FUNC_FN_FN_FN(fma)
+
+extern float __attribute__((overloadable)) fmin(float, float);
+
+extern float __attribute__((overloadable)) fmod(float, float);
+FN_FUNC_FN_FN(fmod)
+
+extern float __attribute__((overloadable)) fract(float v, float *iptr) {
+ int i = (int)floor(v);
+ if (iptr) {
+ iptr[0] = i;
+ }
+ return fmin(v - i, 0x1.fffffep-1f);
+}
+FN_FUNC_FN_PFN(fract)
+
+extern float __attribute__((overloadable)) frexp(float, int *);
+FN_FUNC_FN_PIN(frexp)
+
+extern float __attribute__((overloadable)) hypot(float, float);
+FN_FUNC_FN_FN(hypot)
+
+extern int __attribute__((overloadable)) ilogb(float);
+IN_FUNC_FN(ilogb)
+
+extern float __attribute__((overloadable)) ldexp(float, int);
+FN_FUNC_FN_IN(ldexp)
+FN_FUNC_FN_I(ldexp)
+
+extern float __attribute__((overloadable)) lgamma(float);
+FN_FUNC_FN(lgamma)
+extern float __attribute__((overloadable)) lgamma(float, int*);
+FN_FUNC_FN_PIN(lgamma)
+
+extern float __attribute__((overloadable)) log(float);
+FN_FUNC_FN(log)
+
+extern float __attribute__((overloadable)) log10(float);
+FN_FUNC_FN(log10)
+
+
+extern float __attribute__((overloadable)) log2(float v) {
+ return log10(v) * 3.321928095f;
+}
+FN_FUNC_FN(log2)
+
+extern float __attribute__((overloadable)) log1p(float);
+FN_FUNC_FN(log1p)
+
+extern float __attribute__((overloadable)) logb(float);
+FN_FUNC_FN(logb)
+
+extern float __attribute__((overloadable)) mad(float a, float b, float c) {
+ return a * b + c;
+}
+extern float2 __attribute__((overloadable)) mad(float2 a, float2 b, float2 c) {
+ return a * b + c;
+}
+extern float3 __attribute__((overloadable)) mad(float3 a, float3 b, float3 c) {
+ return a * b + c;
+}
+extern float4 __attribute__((overloadable)) mad(float4 a, float4 b, float4 c) {
+ return a * b + c;
+}
+
+extern float __attribute__((overloadable)) modf(float, float *);
+FN_FUNC_FN_PFN(modf);
+
+extern float __attribute__((overloadable)) nan(uint v) {
+ float f[1];
+ uint32_t *ip = (uint32_t *)f;
+ *ip = v | 0x7fc00000;
+ return f[0];
+}
+
+extern float __attribute__((overloadable)) nextafter(float, float);
+FN_FUNC_FN_FN(nextafter)
+
+FN_FUNC_FN_FN(pow)
+
+extern float __attribute__((overloadable)) pown(float v, int p) {
+ return pow(v, (float)p);
+}
+extern float2 __attribute__((overloadable)) pown(float2 v, int2 p) {
+ float2 f2 = convert_float2(p);
+ return pow(v, f2);
+}
+extern float3 __attribute__((overloadable)) pown(float3 v, int3 p) {
+ float3 f3 = convert_float3(p);
+ return pow(v, f3);
+}
+extern float4 __attribute__((overloadable)) pown(float4 v, int4 p) {
+ float4 f4 = convert_float4(p);
+ return pow(v, f4);
+}
+
+extern float __attribute__((overloadable)) powr(float v, float p) {
+ return pow(v, p);
+}
+extern float2 __attribute__((overloadable)) powr(float2 v, float2 p) {
+ return pow(v, p);
+}
+extern float3 __attribute__((overloadable)) powr(float3 v, float3 p) {
+ return pow(v, p);
+}
+extern float4 __attribute__((overloadable)) powr(float4 v, float4 p) {
+ return pow(v, p);
+}
+
+extern float __attribute__((overloadable)) remainder(float, float);
+FN_FUNC_FN_FN(remainder)
+
+extern float __attribute__((overloadable)) remquo(float, float, int *);
+FN_FUNC_FN_FN_PIN(remquo)
+
+extern float __attribute__((overloadable)) rint(float);
+FN_FUNC_FN(rint)
+
+extern float __attribute__((overloadable)) rootn(float v, int r) {
+ if (r == 0) {
+ return nan(0);
+ }
+
+ if (iszero(v)) {
+ if (r < 0) {
+ if (r & 1) {
+ return copysign(posinf(), v);
+ } else {
+ return posinf();
+ }
+ } else {
+ if (r & 1) {
+ return copysign(0.f, v);
+ } else {
+ return 0.f;
+ }
+ }
+ }
+
+ if (!isinf(v) && !isnan(v) && (v < 0.f)) {
+ if (r & 1) {
+ return (-1.f * pow(-1.f * v, 1.f / r));
+ } else {
+ return nan(0);
+ }
+ }
+
+ return pow(v, 1.f / r);
+}
+FN_FUNC_FN_IN(rootn);
+
+extern float __attribute__((overloadable)) round(float);
+FN_FUNC_FN(round)
+
+
+extern float __attribute__((overloadable)) sqrt(float);
+extern float __attribute__((overloadable)) rsqrt(float v) {
+ return 1.f / sqrt(v);
+}
+FN_FUNC_FN(rsqrt)
+
+extern float __attribute__((overloadable)) sin(float);
+FN_FUNC_FN(sin)
+
+extern float __attribute__((overloadable)) sincos(float v, float *cosptr) {
+ *cosptr = cos(v);
+ return sin(v);
+}
+extern float2 __attribute__((overloadable)) sincos(float2 v, float2 *cosptr) {
+ *cosptr = cos(v);
+ return sin(v);
+}
+extern float3 __attribute__((overloadable)) sincos(float3 v, float3 *cosptr) {
+ *cosptr = cos(v);
+ return sin(v);
+}
+extern float4 __attribute__((overloadable)) sincos(float4 v, float4 *cosptr) {
+ *cosptr = cos(v);
+ return sin(v);
+}
+
+extern float __attribute__((overloadable)) sinh(float);
+FN_FUNC_FN(sinh)
+
+extern float __attribute__((overloadable)) sinpi(float v) {
+ return sin(v * M_PI);
+}
+FN_FUNC_FN(sinpi)
+
+extern float __attribute__((overloadable)) tan(float);
+FN_FUNC_FN(tan)
+
+extern float __attribute__((overloadable)) tanh(float);
+FN_FUNC_FN(tanh)
+
+extern float __attribute__((overloadable)) tanpi(float v) {
+ return tan(v * M_PI);
+}
+FN_FUNC_FN(tanpi)
+
+
+extern float __attribute__((overloadable)) tgamma(float);
+FN_FUNC_FN(tgamma)
+
+extern float __attribute__((overloadable)) trunc(float);
+FN_FUNC_FN(trunc)
+
+// Int ops (partial), 6.11.3
+
+#define XN_FUNC_YN(typeout, fnc, typein) \
+extern typeout __attribute__((overloadable)) fnc(typein); \
+extern typeout##2 __attribute__((overloadable)) fnc(typein##2 v) { \
+ typeout##2 r; \
+ r.x = fnc(v.x); \
+ r.y = fnc(v.y); \
+ return r; \
+} \
+extern typeout##3 __attribute__((overloadable)) fnc(typein##3 v) { \
+ typeout##3 r; \
+ r.x = fnc(v.x); \
+ r.y = fnc(v.y); \
+ r.z = fnc(v.z); \
+ return r; \
+} \
+extern typeout##4 __attribute__((overloadable)) fnc(typein##4 v) { \
+ typeout##4 r; \
+ r.x = fnc(v.x); \
+ r.y = fnc(v.y); \
+ r.z = fnc(v.z); \
+ r.w = fnc(v.w); \
+ return r; \
+}
+
+
+#define UIN_FUNC_IN(fnc) \
+XN_FUNC_YN(uchar, fnc, char) \
+XN_FUNC_YN(ushort, fnc, short) \
+XN_FUNC_YN(uint, fnc, int)
+
+#define IN_FUNC_IN(fnc) \
+XN_FUNC_YN(uchar, fnc, uchar) \
+XN_FUNC_YN(char, fnc, char) \
+XN_FUNC_YN(ushort, fnc, ushort) \
+XN_FUNC_YN(short, fnc, short) \
+XN_FUNC_YN(uint, fnc, uint) \
+XN_FUNC_YN(int, fnc, int)
+
+
+#define XN_FUNC_XN_XN_BODY(type, fnc, body) \
+extern type __attribute__((overloadable)) \
+ fnc(type v1, type v2) { \
+ return body; \
+} \
+extern type##2 __attribute__((overloadable)) \
+ fnc(type##2 v1, type##2 v2) { \
+ type##2 r; \
+ r.x = fnc(v1.x, v2.x); \
+ r.y = fnc(v1.y, v2.y); \
+ return r; \
+} \
+extern type##3 __attribute__((overloadable)) \
+ fnc(type##3 v1, type##3 v2) { \
+ type##3 r; \
+ r.x = fnc(v1.x, v2.x); \
+ r.y = fnc(v1.y, v2.y); \
+ r.z = fnc(v1.z, v2.z); \
+ return r; \
+} \
+extern type##4 __attribute__((overloadable)) \
+ fnc(type##4 v1, type##4 v2) { \
+ type##4 r; \
+ r.x = fnc(v1.x, v2.x); \
+ r.y = fnc(v1.y, v2.y); \
+ r.z = fnc(v1.z, v2.z); \
+ r.w = fnc(v1.w, v2.w); \
+ return r; \
+}
+
+#define IN_FUNC_IN_IN_BODY(fnc, body) \
+XN_FUNC_XN_XN_BODY(uchar, fnc, body) \
+XN_FUNC_XN_XN_BODY(char, fnc, body) \
+XN_FUNC_XN_XN_BODY(ushort, fnc, body) \
+XN_FUNC_XN_XN_BODY(short, fnc, body) \
+XN_FUNC_XN_XN_BODY(uint, fnc, body) \
+XN_FUNC_XN_XN_BODY(int, fnc, body) \
+XN_FUNC_XN_XN_BODY(float, fnc, body)
+
+
+/**
+ * abs
+ */
+extern uint32_t __attribute__((overloadable)) abs(int32_t v) {
+ if (v < 0)
+ return -v;
+ return v;
+}
+extern uint16_t __attribute__((overloadable)) abs(int16_t v) {
+ if (v < 0)
+ return -v;
+ return v;
+}
+extern uint8_t __attribute__((overloadable)) abs(int8_t v) {
+ if (v < 0)
+ return -v;
+ return v;
+}
+
+/**
+ * clz
+ */
+extern uint32_t __attribute__((overloadable)) clz(uint32_t v) {
+ return __builtin_clz(v);
+}
+extern uint16_t __attribute__((overloadable)) clz(uint16_t v) {
+ return (uint16_t)__builtin_clz(v);
+}
+extern uint8_t __attribute__((overloadable)) clz(uint8_t v) {
+ return (uint8_t)__builtin_clz(v);
+}
+extern int32_t __attribute__((overloadable)) clz(int32_t v) {
+ return (int32_t)__builtin_clz((uint32_t)v);
+}
+extern int16_t __attribute__((overloadable)) clz(int16_t v) {
+ return (int16_t)__builtin_clz(v);
+}
+extern int8_t __attribute__((overloadable)) clz(int8_t v) {
+ return (int8_t)__builtin_clz(v);
+}
+
+
+UIN_FUNC_IN(abs)
+IN_FUNC_IN(clz)
+
+
+// 6.11.4
+
+
+extern float __attribute__((overloadable)) degrees(float radians) {
+ return radians * (180.f / M_PI);
+}
+extern float2 __attribute__((overloadable)) degrees(float2 radians) {
+ return radians * (180.f / M_PI);
+}
+extern float3 __attribute__((overloadable)) degrees(float3 radians) {
+ return radians * (180.f / M_PI);
+}
+extern float4 __attribute__((overloadable)) degrees(float4 radians) {
+ return radians * (180.f / M_PI);
+}
+
+extern float __attribute__((overloadable)) mix(float start, float stop, float amount) {
+ return start + (stop - start) * amount;
+}
+extern float2 __attribute__((overloadable)) mix(float2 start, float2 stop, float2 amount) {
+ return start + (stop - start) * amount;
+}
+extern float3 __attribute__((overloadable)) mix(float3 start, float3 stop, float3 amount) {
+ return start + (stop - start) * amount;
+}
+extern float4 __attribute__((overloadable)) mix(float4 start, float4 stop, float4 amount) {
+ return start + (stop - start) * amount;
+}
+extern float2 __attribute__((overloadable)) mix(float2 start, float2 stop, float amount) {
+ return start + (stop - start) * amount;
+}
+extern float3 __attribute__((overloadable)) mix(float3 start, float3 stop, float amount) {
+ return start + (stop - start) * amount;
+}
+extern float4 __attribute__((overloadable)) mix(float4 start, float4 stop, float amount) {
+ return start + (stop - start) * amount;
+}
+
+extern float __attribute__((overloadable)) radians(float degrees) {
+ return degrees * (M_PI / 180.f);
+}
+extern float2 __attribute__((overloadable)) radians(float2 degrees) {
+ return degrees * (M_PI / 180.f);
+}
+extern float3 __attribute__((overloadable)) radians(float3 degrees) {
+ return degrees * (M_PI / 180.f);
+}
+extern float4 __attribute__((overloadable)) radians(float4 degrees) {
+ return degrees * (M_PI / 180.f);
+}
+
+extern float __attribute__((overloadable)) step(float edge, float v) {
+ return (v < edge) ? 0.f : 1.f;
+}
+extern float2 __attribute__((overloadable)) step(float2 edge, float2 v) {
+ float2 r;
+ r.x = (v.x < edge.x) ? 0.f : 1.f;
+ r.y = (v.y < edge.y) ? 0.f : 1.f;
+ return r;
+}
+extern float3 __attribute__((overloadable)) step(float3 edge, float3 v) {
+ float3 r;
+ r.x = (v.x < edge.x) ? 0.f : 1.f;
+ r.y = (v.y < edge.y) ? 0.f : 1.f;
+ r.z = (v.z < edge.z) ? 0.f : 1.f;
+ return r;
+}
+extern float4 __attribute__((overloadable)) step(float4 edge, float4 v) {
+ float4 r;
+ r.x = (v.x < edge.x) ? 0.f : 1.f;
+ r.y = (v.y < edge.y) ? 0.f : 1.f;
+ r.z = (v.z < edge.z) ? 0.f : 1.f;
+ r.w = (v.w < edge.w) ? 0.f : 1.f;
+ return r;
+}
+extern float2 __attribute__((overloadable)) step(float2 edge, float v) {
+ float2 r;
+ r.x = (v < edge.x) ? 0.f : 1.f;
+ r.y = (v < edge.y) ? 0.f : 1.f;
+ return r;
+}
+extern float3 __attribute__((overloadable)) step(float3 edge, float v) {
+ float3 r;
+ r.x = (v < edge.x) ? 0.f : 1.f;
+ r.y = (v < edge.y) ? 0.f : 1.f;
+ r.z = (v < edge.z) ? 0.f : 1.f;
+ return r;
+}
+extern float4 __attribute__((overloadable)) step(float4 edge, float v) {
+ float4 r;
+ r.x = (v < edge.x) ? 0.f : 1.f;
+ r.y = (v < edge.y) ? 0.f : 1.f;
+ r.z = (v < edge.z) ? 0.f : 1.f;
+ r.w = (v < edge.w) ? 0.f : 1.f;
+ return r;
+}
+
+extern float __attribute__((overloadable)) smoothstep(float, float, float);
+extern float2 __attribute__((overloadable)) smoothstep(float2, float2, float2);
+extern float3 __attribute__((overloadable)) smoothstep(float3, float3, float3);
+extern float4 __attribute__((overloadable)) smoothstep(float4, float4, float4);
+extern float2 __attribute__((overloadable)) smoothstep(float, float, float2);
+extern float3 __attribute__((overloadable)) smoothstep(float, float, float3);
+extern float4 __attribute__((overloadable)) smoothstep(float, float, float4);
+
+extern float __attribute__((overloadable)) sign(float v) {
+ if (v > 0) return 1.f;
+ if (v < 0) return -1.f;
+ return v;
+}
+FN_FUNC_FN(sign)
+
+
+// 6.11.5
+extern float3 __attribute__((overloadable)) cross(float3 lhs, float3 rhs) {
+ float3 r;
+ r.x = lhs.y * rhs.z - lhs.z * rhs.y;
+ r.y = lhs.z * rhs.x - lhs.x * rhs.z;
+ r.z = lhs.x * rhs.y - lhs.y * rhs.x;
+ return r;
+}
+
+extern float4 __attribute__((overloadable)) cross(float4 lhs, float4 rhs) {
+ float4 r;
+ r.x = lhs.y * rhs.z - lhs.z * rhs.y;
+ r.y = lhs.z * rhs.x - lhs.x * rhs.z;
+ r.z = lhs.x * rhs.y - lhs.y * rhs.x;
+ r.w = 0.f;
+ return r;
+}
+
+extern float __attribute__((overloadable)) length(float v);
+extern float __attribute__((overloadable)) length(float2 v);
+extern float __attribute__((overloadable)) length(float3 v);
+extern float __attribute__((overloadable)) length(float4 v);
+
+extern float __attribute__((overloadable)) distance(float lhs, float rhs) {
+ return length(lhs - rhs);
+}
+extern float __attribute__((overloadable)) distance(float2 lhs, float2 rhs) {
+ return length(lhs - rhs);
+}
+extern float __attribute__((overloadable)) distance(float3 lhs, float3 rhs) {
+ return length(lhs - rhs);
+}
+extern float __attribute__((overloadable)) distance(float4 lhs, float4 rhs) {
+ return length(lhs - rhs);
+}
+
+extern float __attribute__((overloadable)) normalize(float v) {
+ return 1.f;
+}
+extern float2 __attribute__((overloadable)) normalize(float2 v) {
+ return v / length(v);
+}
+extern float3 __attribute__((overloadable)) normalize(float3 v) {
+ return v / length(v);
+}
+extern float4 __attribute__((overloadable)) normalize(float4 v) {
+ return v / length(v);
+}
+
+extern float __attribute__((overloadable)) half_sqrt(float);
+
+extern float __attribute__((overloadable)) fast_length(float v) {
+ return fabs(v);
+}
+extern float __attribute__((overloadable)) fast_length(float2 v) {
+ return half_sqrt(v.x*v.x + v.y*v.y);
+}
+extern float __attribute__((overloadable)) fast_length(float3 v) {
+ return half_sqrt(v.x*v.x + v.y*v.y + v.z*v.z);
+}
+extern float __attribute__((overloadable)) fast_length(float4 v) {
+ return half_sqrt(v.x*v.x + v.y*v.y + v.z*v.z + v.w*v.w);
+}
+
+extern float __attribute__((overloadable)) fast_distance(float lhs, float rhs) {
+ return fast_length(lhs - rhs);
+}
+extern float __attribute__((overloadable)) fast_distance(float2 lhs, float2 rhs) {
+ return fast_length(lhs - rhs);
+}
+extern float __attribute__((overloadable)) fast_distance(float3 lhs, float3 rhs) {
+ return fast_length(lhs - rhs);
+}
+extern float __attribute__((overloadable)) fast_distance(float4 lhs, float4 rhs) {
+ return fast_length(lhs - rhs);
+}
+
+extern float __attribute__((overloadable)) half_rsqrt(float);
+
+extern float __attribute__((overloadable)) fast_normalize(float v) {
+ return 1.f;
+}
+extern float2 __attribute__((overloadable)) fast_normalize(float2 v) {
+ return v * half_rsqrt(v.x*v.x + v.y*v.y);
+}
+extern float3 __attribute__((overloadable)) fast_normalize(float3 v) {
+ return v * half_rsqrt(v.x*v.x + v.y*v.y + v.z*v.z);
+}
+extern float4 __attribute__((overloadable)) fast_normalize(float4 v) {
+ return v * half_rsqrt(v.x*v.x + v.y*v.y + v.z*v.z + v.w*v.w);
+}
+
+extern float __attribute__((overloadable)) half_recip(float);
+
+/*
+extern float __attribute__((overloadable)) approx_atan(float x) {
+ if (x == 0.f)
+ return 0.f;
+ if (x < 0.f)
+ return -1.f * approx_atan(-1.f * x);
+ if (x > 1.f)
+ return M_PI_2 - approx_atan(approx_recip(x));
+ return x * approx_recip(1.f + 0.28f * x*x);
+}
+FN_FUNC_FN(approx_atan)
+*/
+
+typedef union
+{
+ float fv;
+ int32_t iv;
+} ieee_float_shape_type;
+
+/* Get a 32 bit int from a float. */
+
+#define GET_FLOAT_WORD(i,d) \
+do { \
+ ieee_float_shape_type gf_u; \
+ gf_u.fv = (d); \
+ (i) = gf_u.iv; \
+} while (0)
+
+/* Set a float from a 32 bit int. */
+
+#define SET_FLOAT_WORD(d,i) \
+do { \
+ ieee_float_shape_type sf_u; \
+ sf_u.iv = (i); \
+ (d) = sf_u.fv; \
+} while (0)
+
+
+
+// Valid -125 to 125
+extern float __attribute__((overloadable)) native_exp2(float v) {
+ int32_t iv = (int)v;
+ int32_t x = iv + (iv >> 31); // ~floor(v)
+ float r = (v - x);
+
+ float fo;
+ SET_FLOAT_WORD(fo, (x + 127) << 23);
+
+ r *= 0.694f; // ~ log(e) / log(2)
+ float r2 = r*r;
+ float adj = 1.f + r + (r2 * 0.5f) + (r2*r * 0.166666f) + (r2*r2 * 0.0416666f);
+ return fo * adj;
+}
+
+extern float2 __attribute__((overloadable)) native_exp2(float2 v) {
+ int2 iv = convert_int2(v);
+ int2 x = iv + (iv >> (int2)31);//floor(v);
+ float2 r = (v - convert_float2(x));
+
+ x += 127;
+
+ float2 fo = (float2)(x << (int2)23);
+
+ r *= 0.694f; // ~ log(e) / log(2)
+ float2 r2 = r*r;
+ float2 adj = 1.f + r + (r2 * 0.5f) + (r2*r * 0.166666f) + (r2*r2 * 0.0416666f);
+ return fo * adj;
+}
+
+extern float4 __attribute__((overloadable)) native_exp2(float4 v) {
+ int4 iv = convert_int4(v);
+ int4 x = iv + (iv >> (int4)31);//floor(v);
+ float4 r = (v - convert_float4(x));
+
+ x += 127;
+
+ float4 fo = (float4)(x << (int4)23);
+
+ r *= 0.694f; // ~ log(e) / log(2)
+ float4 r2 = r*r;
+ float4 adj = 1.f + r + (r2 * 0.5f) + (r2*r * 0.166666f) + (r2*r2 * 0.0416666f);
+ return fo * adj;
+}
+
+extern float3 __attribute__((overloadable)) native_exp2(float3 v) {
+ float4 t = 1.f;
+ t.xyz = v;
+ return native_exp2(t).xyz;
+}
+
+
+extern float __attribute__((overloadable)) native_exp(float v) {
+ return native_exp2(v * 1.442695041f);
+}
+extern float2 __attribute__((overloadable)) native_exp(float2 v) {
+ return native_exp2(v * 1.442695041f);
+}
+extern float3 __attribute__((overloadable)) native_exp(float3 v) {
+ return native_exp2(v * 1.442695041f);
+}
+extern float4 __attribute__((overloadable)) native_exp(float4 v) {
+ return native_exp2(v * 1.442695041f);
+}
+
+extern float __attribute__((overloadable)) native_exp10(float v) {
+ return native_exp2(v * 3.321928095f);
+}
+extern float2 __attribute__((overloadable)) native_exp10(float2 v) {
+ return native_exp2(v * 3.321928095f);
+}
+extern float3 __attribute__((overloadable)) native_exp10(float3 v) {
+ return native_exp2(v * 3.321928095f);
+}
+extern float4 __attribute__((overloadable)) native_exp10(float4 v) {
+ return native_exp2(v * 3.321928095f);
+}
+
+extern float __attribute__((overloadable)) native_log2(float v) {
+ int32_t ibits;
+ GET_FLOAT_WORD(ibits, v);
+
+ int32_t e = (ibits >> 23) & 0xff;
+
+ ibits &= 0x7fffff;
+ ibits |= 127 << 23;
+
+ float ir;
+ SET_FLOAT_WORD(ir, ibits);
+
+ ir -= 1.5f;
+ float ir2 = ir*ir;
+ float adj2 = 0.405465108f + // -0.00009f +
+ (0.666666667f * ir) -
+ (0.222222222f * ir2) +
+ (0.098765432f * ir*ir2) -
+ (0.049382716f * ir2*ir2) +
+ (0.026337449f * ir*ir2*ir2) -
+ (0.014631916f * ir2*ir2*ir2);
+ adj2 *= (1.f / 0.693147181f);
+
+ return (float)(e - 127) + adj2;
+}
+extern float2 __attribute__((overloadable)) native_log2(float2 v) {
+ float2 v2 = {native_log2(v.x), native_log2(v.y)};
+ return v2;
+}
+extern float3 __attribute__((overloadable)) native_log2(float3 v) {
+ float3 v2 = {native_log2(v.x), native_log2(v.y), native_log2(v.z)};
+ return v2;
+}
+extern float4 __attribute__((overloadable)) native_log2(float4 v) {
+ float4 v2 = {native_log2(v.x), native_log2(v.y), native_log2(v.z), native_log2(v.w)};
+ return v2;
+}
+
+extern float __attribute__((overloadable)) native_log(float v) {
+ return native_log2(v) * (1.f / 1.442695041f);
+}
+extern float2 __attribute__((overloadable)) native_log(float2 v) {
+ return native_log2(v) * (1.f / 1.442695041f);
+}
+extern float3 __attribute__((overloadable)) native_log(float3 v) {
+ return native_log2(v) * (1.f / 1.442695041f);
+}
+extern float4 __attribute__((overloadable)) native_log(float4 v) {
+ return native_log2(v) * (1.f / 1.442695041f);
+}
+
+extern float __attribute__((overloadable)) native_log10(float v) {
+ return native_log2(v) * (1.f / 3.321928095f);
+}
+extern float2 __attribute__((overloadable)) native_log10(float2 v) {
+ return native_log2(v) * (1.f / 3.321928095f);
+}
+extern float3 __attribute__((overloadable)) native_log10(float3 v) {
+ return native_log2(v) * (1.f / 3.321928095f);
+}
+extern float4 __attribute__((overloadable)) native_log10(float4 v) {
+ return native_log2(v) * (1.f / 3.321928095f);
+}
+
+
+extern float __attribute__((overloadable)) native_powr(float v, float y) {
+ float v2 = native_log2(v);
+ v2 = fmax(v2, -125.f);
+ return native_exp2(v2 * y);
+}
+extern float2 __attribute__((overloadable)) native_powr(float2 v, float2 y) {
+ float2 v2 = native_log2(v);
+ v2 = fmax(v2, -125.f);
+ return native_exp2(v2 * y);
+}
+extern float3 __attribute__((overloadable)) native_powr(float3 v, float3 y) {
+ float3 v2 = native_log2(v);
+ v2 = fmax(v2, -125.f);
+ return native_exp2(v2 * y);
+}
+extern float4 __attribute__((overloadable)) native_powr(float4 v, float4 y) {
+ float4 v2 = native_log2(v);
+ v2 = fmax(v2, -125.f);
+ return native_exp2(v2 * y);
+}
+
+
+#undef FN_FUNC_FN
+#undef IN_FUNC_FN
+#undef FN_FUNC_FN_FN
+#undef FN_FUNC_FN_F
+#undef FN_FUNC_FN_IN
+#undef FN_FUNC_FN_I
+#undef FN_FUNC_FN_PFN
+#undef FN_FUNC_FN_PIN
+#undef FN_FUNC_FN_FN_FN
+#undef FN_FUNC_FN_FN_PIN
+#undef XN_FUNC_YN
+#undef UIN_FUNC_IN
+#undef IN_FUNC_IN
+#undef XN_FUNC_XN_XN_BODY
+#undef IN_FUNC_IN_IN_BODY
diff --git a/driver/runtime/rs_core.c b/driver/runtime/rs_core.c
new file mode 100644
index 0000000..da51a9a
--- /dev/null
+++ b/driver/runtime/rs_core.c
@@ -0,0 +1,166 @@
+#include "rs_core.rsh"
+#include "rs_graphics.rsh"
+#include "rs_structs.h"
+
+/* Function declarations from libRS */
+extern float4 __attribute__((overloadable)) convert_float4(uchar4 c);
+
+/* Implementation of Core Runtime */
+
+extern float4 rsUnpackColor8888(uchar4 c)
+{
+ return convert_float4(c) * 0.003921569f;
+}
+
+
+extern int32_t __attribute__((overloadable)) rsAtomicCas(volatile int32_t *ptr, int32_t expectedValue, int32_t newValue) {
+ return __sync_val_compare_and_swap(ptr, expectedValue, newValue);
+}
+
+extern uint32_t __attribute__((overloadable)) rsAtomicCas(volatile uint32_t *ptr, uint32_t expectedValue, uint32_t newValue) {
+ return __sync_val_compare_and_swap((volatile int32_t *)ptr, (int32_t)expectedValue, (int32_t)newValue);
+}
+
+extern int32_t __attribute__((overloadable)) rsAtomicInc(volatile int32_t *ptr) {
+ return __sync_fetch_and_add(ptr, 1);
+}
+
+extern int32_t __attribute__((overloadable)) rsAtomicDec(volatile int32_t *ptr) {
+ return __sync_fetch_and_sub(ptr, 1);
+}
+
+extern int32_t __attribute__((overloadable)) rsAtomicAdd(volatile int32_t *ptr, int32_t value) {
+ return __sync_fetch_and_add(ptr, value);
+}
+
+extern int32_t __attribute__((overloadable)) rsAtomicSub(volatile int32_t *ptr, int32_t value) {
+ return __sync_fetch_and_sub(ptr, value);
+}
+
+extern int32_t __attribute__((overloadable)) rsAtomicAnd(volatile int32_t *ptr, int32_t value) {
+ return __sync_fetch_and_and(ptr, value);
+}
+
+extern int32_t __attribute__((overloadable)) rsAtomicOr(volatile int32_t *ptr, int32_t value) {
+ return __sync_fetch_and_or(ptr, value);
+}
+
+extern int32_t __attribute__((overloadable)) rsAtomicXor(volatile int32_t *ptr, int32_t value) {
+ return __sync_fetch_and_xor(ptr, value);
+}
+
+extern uint32_t __attribute__((overloadable)) min(uint32_t, uint32_t);
+extern int32_t __attribute__((overloadable)) min(int32_t, int32_t);
+extern uint32_t __attribute__((overloadable)) max(uint32_t, uint32_t);
+extern int32_t __attribute__((overloadable)) max(int32_t, int32_t);
+
+extern uint32_t __attribute__((overloadable)) rsAtomicMin(volatile uint32_t *ptr, uint32_t value) {
+ uint32_t prev, status;
+ do {
+ prev = *ptr;
+ uint32_t n = min(value, prev);
+ status = rsAtomicCas((volatile int32_t*) ptr, (int32_t) prev, (int32_t)n);
+ } while (status != prev);
+ return prev;
+}
+
+extern int32_t __attribute__((overloadable)) rsAtomicMin(volatile int32_t *ptr, int32_t value) {
+ int32_t prev, status;
+ do {
+ prev = *ptr;
+ int32_t n = min(value, prev);
+ status = rsAtomicCas(ptr, prev, n);
+ } while (status != prev);
+ return prev;
+}
+
+extern uint32_t __attribute__((overloadable)) rsAtomicMax(volatile uint32_t *ptr, uint32_t value) {
+ uint32_t prev, status;
+ do {
+ prev = *ptr;
+ uint32_t n = max(value, prev);
+ status = rsAtomicCas((volatile int32_t*) ptr, (int32_t) prev, (int32_t) n);
+ } while (status != prev);
+ return prev;
+}
+
+extern int32_t __attribute__((overloadable)) rsAtomicMax(volatile int32_t *ptr, int32_t value) {
+ int32_t prev, status;
+ do {
+ prev = *ptr;
+ int32_t n = max(value, prev);
+ status = rsAtomicCas(ptr, prev, n);
+ } while (status != prev);
+ return prev;
+}
+
+
+
+extern int32_t rand();
+#define RAND_MAX 0x7fffffff
+
+
+
+extern float __attribute__((overloadable)) rsRand(float min, float max);/* {
+ float r = (float)rand();
+ r /= RAND_MAX;
+ r = r * (max - min) + min;
+ return r;
+}
+*/
+
+extern float __attribute__((overloadable)) rsRand(float max) {
+ return rsRand(0.f, max);
+ //float r = (float)rand();
+ //r *= max;
+ //r /= RAND_MAX;
+ //return r;
+}
+
+extern int __attribute__((overloadable)) rsRand(int max) {
+ return (int)rsRand((float)max);
+}
+
+extern int __attribute__((overloadable)) rsRand(int min, int max) {
+ return (int)rsRand((float)min, (float)max);
+}
+
+#define PRIM_DEBUG(T) \
+extern void __attribute__((overloadable)) rsDebug(const char *, const T *); \
+void __attribute__((overloadable)) rsDebug(const char *txt, T val) { \
+ rsDebug(txt, &val); \
+}
+
+PRIM_DEBUG(char2)
+PRIM_DEBUG(char3)
+PRIM_DEBUG(char4)
+PRIM_DEBUG(uchar2)
+PRIM_DEBUG(uchar3)
+PRIM_DEBUG(uchar4)
+PRIM_DEBUG(short2)
+PRIM_DEBUG(short3)
+PRIM_DEBUG(short4)
+PRIM_DEBUG(ushort2)
+PRIM_DEBUG(ushort3)
+PRIM_DEBUG(ushort4)
+PRIM_DEBUG(int2)
+PRIM_DEBUG(int3)
+PRIM_DEBUG(int4)
+PRIM_DEBUG(uint2)
+PRIM_DEBUG(uint3)
+PRIM_DEBUG(uint4)
+PRIM_DEBUG(long2)
+PRIM_DEBUG(long3)
+PRIM_DEBUG(long4)
+PRIM_DEBUG(ulong2)
+PRIM_DEBUG(ulong3)
+PRIM_DEBUG(ulong4)
+PRIM_DEBUG(float2)
+PRIM_DEBUG(float3)
+PRIM_DEBUG(float4)
+PRIM_DEBUG(double2)
+PRIM_DEBUG(double3)
+PRIM_DEBUG(double4)
+
+#undef PRIM_DEBUG
+
diff --git a/driver/runtime/rs_element.c b/driver/runtime/rs_element.c
new file mode 100644
index 0000000..4db5883
--- /dev/null
+++ b/driver/runtime/rs_element.c
@@ -0,0 +1,111 @@
+#include "rs_core.rsh"
+#include "rs_graphics.rsh"
+#include "rs_structs.h"
+
+/**
+* Element
+*/
+extern uint32_t __attribute__((overloadable))
+ rsElementGetSubElementCount(rs_element e) {
+ Element_t *element = (Element_t *)e.p;
+ if (element == NULL) {
+ return 0;
+ }
+ return element->mHal.state.fieldsCount;
+}
+
+extern rs_element __attribute__((overloadable))
+ rsElementGetSubElement(rs_element e, uint32_t index) {
+ Element_t *element = (Element_t *)e.p;
+ if (element == NULL || index >= element->mHal.state.fieldsCount) {
+ rs_element nullElem = {0};
+ return nullElem;
+ }
+ rs_element returnElem = {element->mHal.state.fields[index]};
+ return returnElem;
+}
+
+extern uint32_t __attribute__((overloadable))
+ rsElementGetSubElementNameLength(rs_element e, uint32_t index) {
+ Element_t *element = (Element_t *)e.p;
+ if (element == NULL || index >= element->mHal.state.fieldsCount) {
+ return 0;
+ }
+ return element->mHal.state.fieldNameLengths[index];
+}
+
+extern uint32_t __attribute__((overloadable))
+ rsElementGetSubElementName(rs_element e, uint32_t index, char *name, uint32_t nameLength) {
+ Element_t *element = (Element_t *)e.p;
+ if (element == NULL || index >= element->mHal.state.fieldsCount ||
+ nameLength == 0 || name == 0) {
+ return 0;
+ }
+
+ uint32_t numToCopy = element->mHal.state.fieldNameLengths[index];
+ if (nameLength < numToCopy) {
+ numToCopy = nameLength;
+ }
+ // Place the null terminator manually, in case of partial string
+ numToCopy --;
+ name[numToCopy] = '\0';
+ const char *nameSource = element->mHal.state.fieldNames[index];
+ for (uint32_t i = 0; i < numToCopy; i ++) {
+ name[i] = nameSource[i];
+ }
+ return numToCopy;
+}
+
+extern uint32_t __attribute__((overloadable))
+ rsElementGetSubElementArraySize(rs_element e, uint32_t index) {
+ Element_t *element = (Element_t *)e.p;
+ if (element == NULL || index >= element->mHal.state.fieldsCount) {
+ return 0;
+ }
+ return element->mHal.state.fieldArraySizes[index];
+}
+
+extern uint32_t __attribute__((overloadable))
+ rsElementGetSubElementOffsetBytes(rs_element e, uint32_t index) {
+ Element_t *element = (Element_t *)e.p;
+ if (element == NULL || index >= element->mHal.state.fieldsCount) {
+ return 0;
+ }
+ return element->mHal.state.fieldOffsetBytes[index];
+}
+
+extern uint32_t __attribute__((overloadable))
+ rsElementGetBytesSize(rs_element e) {
+ Element_t *element = (Element_t *)e.p;
+ if (element == NULL) {
+ return 0;
+ }
+ return element->mHal.state.elementSizeBytes;
+}
+
+extern rs_data_type __attribute__((overloadable))
+ rsElementGetDataType(rs_element e) {
+ Element_t *element = (Element_t *)e.p;
+ if (element == NULL) {
+ return RS_TYPE_INVALID;
+ }
+ return element->mHal.state.dataType;
+}
+
+extern rs_data_kind __attribute__((overloadable))
+ rsElementGetDataKind(rs_element e) {
+ Element_t *element = (Element_t *)e.p;
+ if (element == NULL) {
+ return RS_KIND_INVALID;
+ }
+ return element->mHal.state.dataKind;
+}
+
+extern uint32_t __attribute__((overloadable))
+ rsElementGetVectorSize(rs_element e) {
+ Element_t *element = (Element_t *)e.p;
+ if (element == NULL) {
+ return 0;
+ }
+ return element->mHal.state.vectorSize;
+}
diff --git a/driver/runtime/rs_matrix.c b/driver/runtime/rs_matrix.c
new file mode 100644
index 0000000..3afccc1
--- /dev/null
+++ b/driver/runtime/rs_matrix.c
@@ -0,0 +1,314 @@
+#include "rs_core.rsh"
+#include "rs_graphics.rsh"
+#include "rs_structs.h"
+
+/* Function declarations from libRS */
+extern float4 __attribute__((overloadable)) convert_float4(uchar4 c);
+
+/* Implementation of Core Runtime */
+
+
+/////////////////////////////////////////////////////
+// Matrix ops
+/////////////////////////////////////////////////////
+
+
+extern void __attribute__((overloadable))
+rsMatrixLoadIdentity(rs_matrix4x4 *m) {
+ m->m[0] = 1.f;
+ m->m[1] = 0.f;
+ m->m[2] = 0.f;
+ m->m[3] = 0.f;
+ m->m[4] = 0.f;
+ m->m[5] = 1.f;
+ m->m[6] = 0.f;
+ m->m[7] = 0.f;
+ m->m[8] = 0.f;
+ m->m[9] = 0.f;
+ m->m[10] = 1.f;
+ m->m[11] = 0.f;
+ m->m[12] = 0.f;
+ m->m[13] = 0.f;
+ m->m[14] = 0.f;
+ m->m[15] = 1.f;
+}
+
+extern void __attribute__((overloadable))
+rsMatrixLoadIdentity(rs_matrix3x3 *m) {
+ m->m[0] = 1.f;
+ m->m[1] = 0.f;
+ m->m[2] = 0.f;
+ m->m[3] = 0.f;
+ m->m[4] = 1.f;
+ m->m[5] = 0.f;
+ m->m[6] = 0.f;
+ m->m[7] = 0.f;
+ m->m[8] = 1.f;
+}
+extern void __attribute__((overloadable))
+rsMatrixLoadIdentity(rs_matrix2x2 *m) {
+ m->m[0] = 1.f;
+ m->m[1] = 0.f;
+ m->m[2] = 0.f;
+ m->m[3] = 1.f;
+}
+
+extern void __attribute__((overloadable))
+rsMatrixLoad(rs_matrix4x4 *m, const float *f) {
+ m->m[0] = f[0];
+ m->m[1] = f[1];
+ m->m[2] = f[2];
+ m->m[3] = f[3];
+ m->m[4] = f[4];
+ m->m[5] = f[5];
+ m->m[6] = f[6];
+ m->m[7] = f[7];
+ m->m[8] = f[8];
+ m->m[9] = f[9];
+ m->m[10] = f[10];
+ m->m[11] = f[11];
+ m->m[12] = f[12];
+ m->m[13] = f[13];
+ m->m[14] = f[14];
+ m->m[15] = f[15];
+}
+extern void __attribute__((overloadable))
+rsMatrixLoad(rs_matrix3x3 *m, const float *f) {
+ m->m[0] = f[0];
+ m->m[1] = f[1];
+ m->m[2] = f[2];
+ m->m[3] = f[3];
+ m->m[4] = f[4];
+ m->m[5] = f[5];
+ m->m[6] = f[6];
+ m->m[7] = f[7];
+ m->m[8] = f[8];
+}
+extern void __attribute__((overloadable))
+rsMatrixLoad(rs_matrix2x2 *m, const float *f) {
+ m->m[0] = f[0];
+ m->m[1] = f[1];
+ m->m[2] = f[2];
+ m->m[3] = f[3];
+}
+
+extern void __attribute__((overloadable))
+rsMatrixLoad(rs_matrix4x4 *m, const rs_matrix4x4 *s) {
+ m->m[0] = s->m[0];
+ m->m[1] = s->m[1];
+ m->m[2] = s->m[2];
+ m->m[3] = s->m[3];
+ m->m[4] = s->m[4];
+ m->m[5] = s->m[5];
+ m->m[6] = s->m[6];
+ m->m[7] = s->m[7];
+ m->m[8] = s->m[8];
+ m->m[9] = s->m[9];
+ m->m[10] = s->m[10];
+ m->m[11] = s->m[11];
+ m->m[12] = s->m[12];
+ m->m[13] = s->m[13];
+ m->m[14] = s->m[14];
+ m->m[15] = s->m[15];
+}
+extern void __attribute__((overloadable))
+rsMatrixLoad(rs_matrix4x4 *m, const rs_matrix3x3 *v) {
+ m->m[0] = v->m[0];
+ m->m[1] = v->m[1];
+ m->m[2] = v->m[2];
+ m->m[3] = 0.f;
+ m->m[4] = v->m[3];
+ m->m[5] = v->m[4];
+ m->m[6] = v->m[5];
+ m->m[7] = 0.f;
+ m->m[8] = v->m[6];
+ m->m[9] = v->m[7];
+ m->m[10] = v->m[8];
+ m->m[11] = 0.f;
+ m->m[12] = 0.f;
+ m->m[13] = 0.f;
+ m->m[14] = 0.f;
+ m->m[15] = 1.f;
+}
+extern void __attribute__((overloadable))
+rsMatrixLoad(rs_matrix4x4 *m, const rs_matrix2x2 *v) {
+ m->m[0] = v->m[0];
+ m->m[1] = v->m[1];
+ m->m[2] = 0.f;
+ m->m[3] = 0.f;
+ m->m[4] = v->m[2];
+ m->m[5] = v->m[3];
+ m->m[6] = 0.f;
+ m->m[7] = 0.f;
+ m->m[8] = 0.f;
+ m->m[9] = 0.f;
+ m->m[10] = 1.f;
+ m->m[11] = 0.f;
+ m->m[12] = 0.f;
+ m->m[13] = 0.f;
+ m->m[14] = 0.f;
+ m->m[15] = 1.f;
+}
+extern void __attribute__((overloadable))
+rsMatrixLoad(rs_matrix3x3 *m, const rs_matrix3x3 *s) {
+ m->m[0] = s->m[0];
+ m->m[1] = s->m[1];
+ m->m[2] = s->m[2];
+ m->m[3] = s->m[3];
+ m->m[4] = s->m[4];
+ m->m[5] = s->m[5];
+ m->m[6] = s->m[6];
+ m->m[7] = s->m[7];
+ m->m[8] = s->m[8];
+}
+extern void __attribute__((overloadable))
+rsMatrixLoad(rs_matrix2x2 *m, const rs_matrix2x2 *s) {
+ m->m[0] = s->m[0];
+ m->m[1] = s->m[1];
+ m->m[2] = s->m[2];
+ m->m[3] = s->m[3];
+}
+
+
+extern void __attribute__((overloadable))
+rsMatrixSet(rs_matrix4x4 *m, uint32_t row, uint32_t col, float v) {
+ m->m[row * 4 + col] = v;
+}
+
+extern float __attribute__((overloadable))
+rsMatrixGet(const rs_matrix4x4 *m, uint32_t row, uint32_t col) {
+ return m->m[row * 4 + col];
+}
+
+extern void __attribute__((overloadable))
+rsMatrixSet(rs_matrix3x3 *m, uint32_t row, uint32_t col, float v) {
+ m->m[row * 3 + col] = v;
+}
+
+extern float __attribute__((overloadable))
+rsMatrixGet(const rs_matrix3x3 *m, uint32_t row, uint32_t col) {
+ return m->m[row * 3 + col];
+}
+
+extern void __attribute__((overloadable))
+rsMatrixSet(rs_matrix2x2 *m, uint32_t row, uint32_t col, float v) {
+ m->m[row * 2 + col] = v;
+}
+
+extern float __attribute__((overloadable))
+rsMatrixGet(const rs_matrix2x2 *m, uint32_t row, uint32_t col) {
+ return m->m[row * 2 + col];
+}
+
+extern float2 __attribute__((overloadable))
+rsMatrixMultiply(const rs_matrix2x2 *m, float2 in) {
+ float2 ret;
+ ret.x = (m->m[0] * in.x) + (m->m[2] * in.y);
+ ret.y = (m->m[1] * in.x) + (m->m[3] * in.y);
+ return ret;
+}
+extern float2 __attribute__((overloadable))
+rsMatrixMultiply(rs_matrix2x2 *m, float2 in) {
+ return rsMatrixMultiply((const rs_matrix2x2 *)m, in);
+}
+
+extern float4 __attribute__((overloadable))
+rsMatrixMultiply(rs_matrix4x4 *m, float4 in) {
+ return rsMatrixMultiply((const rs_matrix4x4 *)m, in);
+}
+
+extern float4 __attribute__((overloadable))
+rsMatrixMultiply(rs_matrix4x4 *m, float3 in) {
+ return rsMatrixMultiply((const rs_matrix4x4 *)m, in);
+}
+
+extern float4 __attribute__((overloadable))
+rsMatrixMultiply(rs_matrix4x4 *m, float2 in) {
+ return rsMatrixMultiply((const rs_matrix4x4 *)m, in);
+}
+
+extern float3 __attribute__((overloadable))
+rsMatrixMultiply(rs_matrix3x3 *m, float3 in) {
+ return rsMatrixMultiply((const rs_matrix3x3 *)m, in);
+}
+
+extern float3 __attribute__((overloadable))
+rsMatrixMultiply(rs_matrix3x3 *m, float2 in) {
+ return rsMatrixMultiply((const rs_matrix3x3 *)m, in);
+}
+
+extern void __attribute__((overloadable))
+rsMatrixLoadMultiply(rs_matrix4x4 *ret, const rs_matrix4x4 *lhs, const rs_matrix4x4 *rhs) {
+ for (int i=0 ; i<4 ; i++) {
+ float ri0 = 0;
+ float ri1 = 0;
+ float ri2 = 0;
+ float ri3 = 0;
+ for (int j=0 ; j<4 ; j++) {
+ const float rhs_ij = rsMatrixGet(rhs, i, j);
+ ri0 += rsMatrixGet(lhs, j, 0) * rhs_ij;
+ ri1 += rsMatrixGet(lhs, j, 1) * rhs_ij;
+ ri2 += rsMatrixGet(lhs, j, 2) * rhs_ij;
+ ri3 += rsMatrixGet(lhs, j, 3) * rhs_ij;
+ }
+ rsMatrixSet(ret, i, 0, ri0);
+ rsMatrixSet(ret, i, 1, ri1);
+ rsMatrixSet(ret, i, 2, ri2);
+ rsMatrixSet(ret, i, 3, ri3);
+ }
+}
+
+extern void __attribute__((overloadable))
+rsMatrixMultiply(rs_matrix4x4 *lhs, const rs_matrix4x4 *rhs) {
+ rs_matrix4x4 r;
+ rsMatrixLoadMultiply(&r, lhs, rhs);
+ rsMatrixLoad(lhs, &r);
+}
+
+extern void __attribute__((overloadable))
+rsMatrixLoadMultiply(rs_matrix3x3 *ret, const rs_matrix3x3 *lhs, const rs_matrix3x3 *rhs) {
+ for (int i=0 ; i<3 ; i++) {
+ float ri0 = 0;
+ float ri1 = 0;
+ float ri2 = 0;
+ for (int j=0 ; j<3 ; j++) {
+ const float rhs_ij = rsMatrixGet(rhs, i, j);
+ ri0 += rsMatrixGet(lhs, j, 0) * rhs_ij;
+ ri1 += rsMatrixGet(lhs, j, 1) * rhs_ij;
+ ri2 += rsMatrixGet(lhs, j, 2) * rhs_ij;
+ }
+ rsMatrixSet(ret, i, 0, ri0);
+ rsMatrixSet(ret, i, 1, ri1);
+ rsMatrixSet(ret, i, 2, ri2);
+ }
+}
+
+extern void __attribute__((overloadable))
+rsMatrixMultiply(rs_matrix3x3 *lhs, const rs_matrix3x3 *rhs) {
+ rs_matrix3x3 r;
+ rsMatrixLoadMultiply(&r, lhs, rhs);
+ rsMatrixLoad(lhs, &r);
+}
+
+extern void __attribute__((overloadable))
+rsMatrixLoadMultiply(rs_matrix2x2 *ret, const rs_matrix2x2 *lhs, const rs_matrix2x2 *rhs) {
+ for (int i=0 ; i<2 ; i++) {
+ float ri0 = 0;
+ float ri1 = 0;
+ for (int j=0 ; j<2 ; j++) {
+ const float rhs_ij = rsMatrixGet(rhs, i, j);
+ ri0 += rsMatrixGet(lhs, j, 0) * rhs_ij;
+ ri1 += rsMatrixGet(lhs, j, 1) * rhs_ij;
+ }
+ rsMatrixSet(ret, i, 0, ri0);
+ rsMatrixSet(ret, i, 1, ri1);
+ }
+}
+
+extern void __attribute__((overloadable))
+rsMatrixMultiply(rs_matrix2x2 *lhs, const rs_matrix2x2 *rhs) {
+ rs_matrix2x2 r;
+ rsMatrixLoadMultiply(&r, lhs, rhs);
+ rsMatrixLoad(lhs, &r);
+}
+
diff --git a/driver/runtime/rs_mesh.c b/driver/runtime/rs_mesh.c
new file mode 100644
index 0000000..bb533bc
--- /dev/null
+++ b/driver/runtime/rs_mesh.c
@@ -0,0 +1,55 @@
+#include "rs_core.rsh"
+#include "rs_graphics.rsh"
+#include "rs_structs.h"
+
+/**
+* Mesh
+*/
+extern uint32_t __attribute__((overloadable))
+ rsgMeshGetVertexAllocationCount(rs_mesh m) {
+ Mesh_t *mesh = (Mesh_t *)m.p;
+ if (mesh == NULL) {
+ return 0;
+ }
+ return mesh->mHal.state.vertexBuffersCount;
+}
+
+extern uint32_t __attribute__((overloadable))
+ rsgMeshGetPrimitiveCount(rs_mesh m) {
+ Mesh_t *mesh = (Mesh_t *)m.p;
+ if (mesh == NULL) {
+ return 0;
+ }
+ return mesh->mHal.state.primitivesCount;
+}
+
+extern rs_allocation __attribute__((overloadable))
+ rsgMeshGetVertexAllocation(rs_mesh m, uint32_t index) {
+ Mesh_t *mesh = (Mesh_t *)m.p;
+ if (mesh == NULL || index >= mesh->mHal.state.vertexBuffersCount) {
+ rs_allocation nullAlloc = {0};
+ return nullAlloc;
+ }
+ rs_allocation returnAlloc = {mesh->mHal.state.vertexBuffers[index]};
+ return returnAlloc;
+}
+
+extern rs_allocation __attribute__((overloadable))
+ rsgMeshGetIndexAllocation(rs_mesh m, uint32_t index) {
+ Mesh_t *mesh = (Mesh_t *)m.p;
+ if (mesh == NULL || index >= mesh->mHal.state.primitivesCount) {
+ rs_allocation nullAlloc = {0};
+ return nullAlloc;
+ }
+ rs_allocation returnAlloc = {mesh->mHal.state.indexBuffers[index]};
+ return returnAlloc;
+}
+
+extern rs_primitive __attribute__((overloadable))
+ rsgMeshGetPrimitive(rs_mesh m, uint32_t index) {
+ Mesh_t *mesh = (Mesh_t *)m.p;
+ if (mesh == NULL || index >= mesh->mHal.state.primitivesCount) {
+ return RS_PRIMITIVE_INVALID;
+ }
+ return mesh->mHal.state.primitives[index];
+}
diff --git a/driver/runtime/rs_program.c b/driver/runtime/rs_program.c
new file mode 100644
index 0000000..64c656f
--- /dev/null
+++ b/driver/runtime/rs_program.c
@@ -0,0 +1,108 @@
+#include "rs_core.rsh"
+#include "rs_graphics.rsh"
+#include "rs_structs.h"
+
+/**
+* Program Store
+*/
+extern rs_depth_func __attribute__((overloadable))
+ rsgProgramStoreGetDepthFunc(rs_program_store ps) {
+ ProgramStore_t *prog = (ProgramStore_t *)ps.p;
+ if (prog == NULL) {
+ return RS_DEPTH_FUNC_INVALID;
+ }
+ return prog->mHal.state.depthFunc;
+}
+
+extern bool __attribute__((overloadable))
+ rsgProgramStoreIsDepthMaskEnabled(rs_program_store ps) {
+ ProgramStore_t *prog = (ProgramStore_t *)ps.p;
+ if (prog == NULL) {
+ return false;
+ }
+ return prog->mHal.state.depthWriteEnable;
+}
+
+extern bool __attribute__((overloadable))
+ rsgProgramStoreIsColorMaskRedEnabled(rs_program_store ps) {
+ ProgramStore_t *prog = (ProgramStore_t *)ps.p;
+ if (prog == NULL) {
+ return false;
+ }
+ return prog->mHal.state.colorRWriteEnable;
+}
+
+extern bool __attribute__((overloadable))
+ rsgProgramStoreIsColorMaskGreenEnabled(rs_program_store ps) {
+ ProgramStore_t *prog = (ProgramStore_t *)ps.p;
+ if (prog == NULL) {
+ return false;
+ }
+ return prog->mHal.state.colorGWriteEnable;
+}
+
+extern bool __attribute__((overloadable))
+ rsgProgramStoreIsColorMaskBlueEnabled(rs_program_store ps) {
+ ProgramStore_t *prog = (ProgramStore_t *)ps.p;
+ if (prog == NULL) {
+ return false;
+ }
+ return prog->mHal.state.colorBWriteEnable;
+}
+
+extern bool __attribute__((overloadable))
+ rsgProgramStoreIsColorMaskAlphaEnabled(rs_program_store ps) {
+ ProgramStore_t *prog = (ProgramStore_t *)ps.p;
+ if (prog == NULL) {
+ return false;
+ }
+ return prog->mHal.state.colorAWriteEnable;
+}
+
+extern rs_blend_src_func __attribute__((overloadable))
+ rsgProgramStoreGetBlendSrcFunc(rs_program_store ps) {
+ ProgramStore_t *prog = (ProgramStore_t *)ps.p;
+ if (prog == NULL) {
+ return RS_BLEND_SRC_INVALID;
+ }
+ return prog->mHal.state.blendSrc;
+}
+
+extern rs_blend_dst_func __attribute__((overloadable))
+ rsgProgramStoreGetBlendDstFunc(rs_program_store ps) {
+ ProgramStore_t *prog = (ProgramStore_t *)ps.p;
+ if (prog == NULL) {
+ return RS_BLEND_DST_INVALID;
+ }
+ return prog->mHal.state.blendDst;
+}
+
+extern bool __attribute__((overloadable))
+ rsgProgramStoreIsDitherEnabled(rs_program_store ps) {
+ ProgramStore_t *prog = (ProgramStore_t *)ps.p;
+ if (prog == NULL) {
+ return false;
+ }
+ return prog->mHal.state.ditherEnable;
+}
+
+/**
+* Program Raster
+*/
+extern bool __attribute__((overloadable))
+ rsgProgramRasterIsPointSpriteEnabled(rs_program_raster pr) {
+ ProgramRaster_t *prog = (ProgramRaster_t *)pr.p;
+ if (prog == NULL) {
+ return false;
+ }
+ return prog->mHal.state.pointSprite;
+}
+
+extern rs_cull_mode __attribute__((overloadable))
+ rsgProgramRasterGetCullMode(rs_program_raster pr) {
+ ProgramRaster_t *prog = (ProgramRaster_t *)pr.p;
+ if (prog == NULL) {
+ return RS_CULL_INVALID;
+ }
+ return prog->mHal.state.cull;
+}
diff --git a/driver/runtime/rs_sample.c b/driver/runtime/rs_sample.c
new file mode 100644
index 0000000..2cd5bdc
--- /dev/null
+++ b/driver/runtime/rs_sample.c
@@ -0,0 +1,654 @@
+#include "rs_core.rsh"
+#include "rs_graphics.rsh"
+#include "rs_structs.h"
+
+
+// 565 Conversion bits taken from SkBitmap
+#define SK_R16_BITS 5
+#define SK_G16_BITS 6
+#define SK_B16_BITS 5
+
+#define SK_R16_SHIFT (SK_B16_BITS + SK_G16_BITS)
+#define SK_G16_SHIFT (SK_B16_BITS)
+#define SK_B16_SHIFT 0
+
+#define SK_R16_MASK ((1 << SK_R16_BITS) - 1)
+#define SK_G16_MASK ((1 << SK_G16_BITS) - 1)
+#define SK_B16_MASK ((1 << SK_B16_BITS) - 1)
+
+#define SkGetPackedR16(color) (((unsigned)(color) >> SK_R16_SHIFT) & SK_R16_MASK)
+#define SkGetPackedG16(color) (((unsigned)(color) >> SK_G16_SHIFT) & SK_G16_MASK)
+#define SkGetPackedB16(color) (((unsigned)(color) >> SK_B16_SHIFT) & SK_B16_MASK)
+
+static inline unsigned SkR16ToR32(unsigned r) {
+ return (r << (8 - SK_R16_BITS)) | (r >> (2 * SK_R16_BITS - 8));
+}
+
+static inline unsigned SkG16ToG32(unsigned g) {
+ return (g << (8 - SK_G16_BITS)) | (g >> (2 * SK_G16_BITS - 8));
+}
+
+static inline unsigned SkB16ToB32(unsigned b) {
+ return (b << (8 - SK_B16_BITS)) | (b >> (2 * SK_B16_BITS - 8));
+}
+
+#define SkPacked16ToR32(c) SkR16ToR32(SkGetPackedR16(c))
+#define SkPacked16ToG32(c) SkG16ToG32(SkGetPackedG16(c))
+#define SkPacked16ToB32(c) SkB16ToB32(SkGetPackedB16(c))
+
+static float3 getFrom565(uint16_t color) {
+ float3 result;
+ result.x = (float)SkPacked16ToR32(color);
+ result.y = (float)SkPacked16ToG32(color);
+ result.z = (float)SkPacked16ToB32(color);
+ return result;
+}
+
+/**
+* Allocation sampling
+*/
+static inline float __attribute__((overloadable))
+ getElementAt1(const uint8_t *p, int32_t x) {
+ float r = p[x];
+ return r;
+}
+
+static inline float2 __attribute__((overloadable))
+ getElementAt2(const uint8_t *p, int32_t x) {
+ x *= 2;
+ float2 r = {p[x], p[x+1]};
+ return r;
+}
+
+static inline float3 __attribute__((overloadable))
+ getElementAt3(const uint8_t *p, int32_t x) {
+ x *= 4;
+ float3 r = {p[x], p[x+1], p[x+2]};
+ return r;
+}
+
+static inline float4 __attribute__((overloadable))
+ getElementAt4(const uint8_t *p, int32_t x) {
+ x *= 4;
+ const uchar4 *p2 = (const uchar4 *)&p[x];
+ return convert_float4(p2[0]);
+}
+
+static inline float3 __attribute__((overloadable))
+ getElementAt565(const uint8_t *p, int32_t x) {
+ x *= 2;
+ float3 r = getFrom565(((const uint16_t *)p)[0]);
+ return r;
+}
+
+static inline float __attribute__((overloadable))
+ getElementAt1(const uint8_t *p, size_t stride, int32_t x, int32_t y) {
+ p += y * stride;
+ float r = p[x];
+ return r;
+}
+
+static inline float2 __attribute__((overloadable))
+ getElementAt2(const uint8_t *p, size_t stride, int32_t x, int32_t y) {
+ p += y * stride;
+ x *= 2;
+ float2 r = {p[x], p[x+1]};
+ return r;
+}
+
+static inline float3 __attribute__((overloadable))
+ getElementAt3(const uint8_t *p, size_t stride, int32_t x, int32_t y) {
+ p += y * stride;
+ x *= 4;
+ float3 r = {p[x], p[x+1], p[x+2]};
+ return r;
+}
+
+static inline float4 __attribute__((overloadable))
+ getElementAt4(const uint8_t *p, size_t stride, int32_t x, int32_t y) {
+ p += y * stride;
+ x *= 4;
+ float4 r = {p[x], p[x+1], p[x+2], p[x+3]};
+ return r;
+}
+
+static inline float3 __attribute__((overloadable))
+ getElementAt565(const uint8_t *p, size_t stride, int32_t x, int32_t y) {
+ p += y * stride;
+ x *= 2;
+ float3 r = getFrom565(((const uint16_t *)p)[0]);
+ return r;
+}
+
+
+
+
+
+static float4 __attribute__((overloadable))
+ getSample_A(const uint8_t *p, int32_t iPixel,
+ int32_t next, float w0, float w1) {
+ float p0 = getElementAt1(p, iPixel);
+ float p1 = getElementAt1(p, next);
+ float r = p0 * w0 + p1 * w1;
+ r *= (1.f / 255.f);
+ float4 ret = {0.f, 0.f, 0.f, r};
+ return ret;
+}
+static float4 __attribute__((overloadable))
+ getSample_L(const uint8_t *p, int32_t iPixel,
+ int32_t next, float w0, float w1) {
+ float p0 = getElementAt1(p, iPixel);
+ float p1 = getElementAt1(p, next);
+ float r = p0 * w0 + p1 * w1;
+ r *= (1.f / 255.f);
+ float4 ret = {r, r, r, 1.f};
+ return ret;
+}
+static float4 __attribute__((overloadable))
+ getSample_LA(const uint8_t *p, int32_t iPixel,
+ int32_t next, float w0, float w1) {
+ float2 p0 = getElementAt2(p, iPixel);
+ float2 p1 = getElementAt2(p, next);
+ float2 r = p0 * w0 + p1 * w1;
+ r *= (1.f / 255.f);
+ float4 ret = {r.x, r.x, r.x, r.y};
+ return ret;
+}
+static float4 __attribute__((overloadable))
+ getSample_RGB(const uint8_t *p, int32_t iPixel,
+ int32_t next, float w0, float w1) {
+ float3 p0 = getElementAt3(p, iPixel);
+ float3 p1 = getElementAt3(p, next);
+ float3 r = p0 * w0 + p1 * w1;
+ r *= (1.f / 255.f);
+ float4 ret = {r.x, r.x, r.z, 1.f};
+ return ret;
+}
+static float4 __attribute__((overloadable))
+ getSample_565(const uint8_t *p, int32_t iPixel,
+ int32_t next, float w0, float w1) {
+ float3 p0 = getElementAt565(p, iPixel);
+ float3 p1 = getElementAt565(p, next);
+ float3 r = p0 * w0 + p1 * w1;
+ r *= (1.f / 255.f);
+ float4 ret = {r.x, r.x, r.z, 1.f};
+ return ret;
+}
+static float4 __attribute__((overloadable))
+ getSample_RGBA(const uint8_t *p, int32_t iPixel,
+ int32_t next, float w0, float w1) {
+ float4 p0 = getElementAt4(p, iPixel);
+ float4 p1 = getElementAt4(p, next);
+ float4 r = p0 * w0 + p1 * w1;
+ r *= (1.f / 255.f);
+ return r;
+}
+
+
+static float4 __attribute__((overloadable))
+ getSample_A(const uint8_t *p, size_t stride,
+ int locX, int locY, int nextX, int nextY,
+ float w0, float w1, float w2, float w3) {
+ float p0 = getElementAt1(p, stride, locX, locY);
+ float p1 = getElementAt1(p, stride, nextX, locY);
+ float p2 = getElementAt1(p, stride, locX, nextY);
+ float p3 = getElementAt1(p, stride, nextX, nextY);
+ float r = p0 * w0 + p1 * w1 + p2 * w2 + p3 * w3;
+ r *= (1.f / 255.f);
+ float4 ret = {0.f, 0.f, 0.f, r};
+ return ret;
+}
+static float4 __attribute__((overloadable))
+ getSample_L(const uint8_t *p, size_t stride,
+ int locX, int locY, int nextX, int nextY,
+ float w0, float w1, float w2, float w3) {
+ float p0 = getElementAt1(p, stride, locX, locY);
+ float p1 = getElementAt1(p, stride, nextX, locY);
+ float p2 = getElementAt1(p, stride, locX, nextY);
+ float p3 = getElementAt1(p, stride, nextX, nextY);
+ float r = p0 * w0 + p1 * w1 + p2 * w2 + p3 * w3;
+ r *= (1.f / 255.f);
+ float4 ret = {r, r, r, 1.f};
+ return ret;
+}
+static float4 __attribute__((overloadable))
+ getSample_LA(const uint8_t *p, size_t stride,
+ int locX, int locY, int nextX, int nextY,
+ float w0, float w1, float w2, float w3) {
+ float2 p0 = getElementAt2(p, stride, locX, locY);
+ float2 p1 = getElementAt2(p, stride, nextX, locY);
+ float2 p2 = getElementAt2(p, stride, locX, nextY);
+ float2 p3 = getElementAt2(p, stride, nextX, nextY);
+ float2 r = p0 * w0 + p1 * w1 + p2 * w2 + p3 * w3;
+ r *= (1.f / 255.f);
+ float4 ret = {r.x, r.x, r.x, r.y};
+ return ret;
+}
+static float4 __attribute__((overloadable))
+ getSample_RGB(const uint8_t *p, size_t stride,
+ int locX, int locY, int nextX, int nextY,
+ float w0, float w1, float w2, float w3) {
+ float4 p0 = getElementAt4(p, stride, locX, locY);
+ float4 p1 = getElementAt4(p, stride, nextX, locY);
+ float4 p2 = getElementAt4(p, stride, locX, nextY);
+ float4 p3 = getElementAt4(p, stride, nextX, nextY);
+ float4 r = p0 * w0 + p1 * w1 + p2 * w2 + p3 * w3;
+ r *= (1.f / 255.f);
+ float4 ret = {r.x, r.y, r.z, 1.f};
+ return ret;
+}
+static float4 __attribute__((overloadable))
+ getSample_RGBA(const uint8_t *p, size_t stride,
+ int locX, int locY, int nextX, int nextY,
+ float w0, float w1, float w2, float w3) {
+ float4 p0 = getElementAt4(p, stride, locX, locY);
+ float4 p1 = getElementAt4(p, stride, nextX, locY);
+ float4 p2 = getElementAt4(p, stride, locX, nextY);
+ float4 p3 = getElementAt4(p, stride, nextX, nextY);
+ float4 r = p0 * w0 + p1 * w1 + p2 * w2 + p3 * w3;
+ r *= (1.f / 255.f);
+ return r;
+}
+static float4 __attribute__((overloadable))
+ getSample_565(const uint8_t *p, size_t stride,
+ int locX, int locY, int nextX, int nextY,
+ float w0, float w1, float w2, float w3) {
+ float3 p0 = getElementAt565(p, stride, locX, locY);
+ float3 p1 = getElementAt565(p, stride, nextX, locY);
+ float3 p2 = getElementAt565(p, stride, locX, nextY);
+ float3 p3 = getElementAt565(p, stride, nextX, nextY);
+ float3 r = p0 * w0 + p1 * w1 + p2 * w2 + p3 * w3;
+ r *= (1.f / 255.f);
+ float4 ret;
+ ret.rgb = r;
+ ret.w = 1.f;
+ return ret;
+}
+
+static float4 __attribute__((overloadable))
+ getBilinearSample1D(const Allocation_t *alloc, float2 weights,
+ uint32_t iPixel, uint32_t next,
+ rs_data_kind dk, rs_data_type dt, uint32_t lod) {
+
+ const uint8_t *p = (const uint8_t *)alloc->mHal.drvState.lod[lod].mallocPtr;
+
+ switch(dk) {
+ case RS_KIND_PIXEL_RGBA:
+ return getSample_RGBA(p, iPixel, next, weights.x, weights.y);
+ case RS_KIND_PIXEL_A:
+ return getSample_A(p, iPixel, next, weights.x, weights.y);
+ case RS_KIND_PIXEL_RGB:
+ if (dt == RS_TYPE_UNSIGNED_5_6_5) {
+ return getSample_565(p, iPixel, next, weights.x, weights.y);
+ }
+ return getSample_RGB(p, iPixel, next, weights.x, weights.y);
+ case RS_KIND_PIXEL_L:
+ return getSample_L(p, iPixel, next, weights.x, weights.y);
+ case RS_KIND_PIXEL_LA:
+ return getSample_LA(p, iPixel, next, weights.x, weights.y);
+
+ default:
+ //__builtin_unreachable();
+ break;
+ }
+
+ //__builtin_unreachable();
+ return 0.f;
+}
+
+static uint32_t wrapI(rs_sampler_value wrap, int32_t coord, int32_t size) {
+ if (wrap == RS_SAMPLER_WRAP) {
+ coord = coord % size;
+ if (coord < 0) {
+ coord += size;
+ }
+ }
+ if (wrap == RS_SAMPLER_MIRRORED_REPEAT) {
+ coord = coord % (size * 2);
+ if (coord < 0) {
+ coord = (size * 2) + coord;
+ }
+ if (coord >= size) {
+ coord = (size * 2) - coord;
+ }
+ }
+ return (uint32_t)max(0, min(coord, size - 1));
+}
+
+static float4 __attribute__((overloadable))
+ getBilinearSample2D(const Allocation_t *alloc, float w0, float w1, float w2, float w3,
+ int lx, int ly, int nx, int ny,
+ rs_data_kind dk, rs_data_type dt, uint32_t lod) {
+
+ const uint8_t *p = (const uint8_t *)alloc->mHal.drvState.lod[lod].mallocPtr;
+ size_t stride = alloc->mHal.drvState.lod[lod].stride;
+
+ switch(dk) {
+ case RS_KIND_PIXEL_RGBA:
+ return getSample_RGBA(p, stride, lx, ly, nx, ny, w0, w1, w2, w3);
+ case RS_KIND_PIXEL_A:
+ return getSample_A(p, stride, lx, ly, nx, ny, w0, w1, w2, w3);
+ case RS_KIND_PIXEL_LA:
+ return getSample_LA(p, stride, lx, ly, nx, ny, w0, w1, w2, w3);
+ case RS_KIND_PIXEL_RGB:
+ if (dt == RS_TYPE_UNSIGNED_5_6_5) {
+ return getSample_565(p, stride, lx, ly, nx, ny, w0, w1, w2, w3);
+ }
+ return getSample_RGB(p, stride, lx, ly, nx, ny, w0, w1, w2, w3);
+ case RS_KIND_PIXEL_L:
+ return getSample_L(p, stride, lx, ly, nx, ny, w0, w1, w2, w3);
+
+ default:
+ break;
+ }
+
+ return 0.f;
+}
+
+static float4 __attribute__((overloadable))
+ getNearestSample(const Allocation_t *alloc, uint32_t iPixel, rs_data_kind dk,
+ rs_data_type dt, uint32_t lod) {
+
+ const uint8_t *p = (const uint8_t *)alloc->mHal.drvState.lod[lod].mallocPtr;
+
+ float4 result = {0.f, 0.f, 0.f, 255.f};
+
+ switch(dk) {
+ case RS_KIND_PIXEL_RGBA:
+ result = getElementAt4(p, iPixel);
+ break;
+ case RS_KIND_PIXEL_A:
+ result.w = getElementAt1(p, iPixel);
+ break;
+ case RS_KIND_PIXEL_LA:
+ result.zw = getElementAt2(p, iPixel);
+ result.xy = result.z;
+ break;
+ case RS_KIND_PIXEL_RGB:
+ if (dt == RS_TYPE_UNSIGNED_5_6_5) {
+ result.xyz = getElementAt565(p, iPixel);
+ } else {
+ result.xyz = getElementAt3(p, iPixel);
+ }
+ break;
+ case RS_KIND_PIXEL_L:
+ result.xyz = getElementAt1(p, iPixel);
+
+ default:
+ //__builtin_unreachable();
+ break;
+ }
+
+ return result * 0.003921569f;
+}
+
+static float4 __attribute__((overloadable))
+ getNearestSample(const Allocation_t *alloc, uint2 iPixel, rs_data_kind dk,
+ rs_data_type dt, uint32_t lod) {
+
+ const uint8_t *p = (const uint8_t *)alloc->mHal.drvState.lod[lod].mallocPtr;
+ size_t stride = alloc->mHal.drvState.lod[lod].stride;
+
+ float4 result = {0.f, 0.f, 0.f, 255.f};
+
+ switch(dk) {
+ case RS_KIND_PIXEL_RGBA:
+ result = getElementAt4(p, stride, iPixel.x, iPixel.y);
+ break;
+ case RS_KIND_PIXEL_A:
+ result.w = getElementAt1(p, stride, iPixel.x, iPixel.y);
+ break;
+ case RS_KIND_PIXEL_LA:
+ result.zw = getElementAt2(p, stride, iPixel.x, iPixel.y);
+ result.xy = result.z;
+ break;
+ case RS_KIND_PIXEL_RGB:
+ if (dt == RS_TYPE_UNSIGNED_5_6_5) {
+ result.xyz = getElementAt565(p, stride, iPixel.x, iPixel.y);
+ } else {
+ result.xyz = getElementAt3(p, stride, iPixel.x, iPixel.y);
+ }
+ break;
+
+ default:
+ //__builtin_unreachable();
+ break;
+ }
+
+ return result * 0.003921569f;
+}
+
+static float4 __attribute__((overloadable))
+ sample_LOD_LinearPixel(const Allocation_t *alloc,
+ rs_data_kind dk, rs_data_type dt,
+ rs_sampler_value wrapS,
+ float uv, uint32_t lod) {
+
+ const uint8_t *p = (const uint8_t *)alloc->mHal.drvState.lod[lod].mallocPtr;
+
+ int32_t sourceW = alloc->mHal.drvState.lod[lod].dimX;
+ float pixelUV = uv * (float)(sourceW);
+ int32_t iPixel = (int32_t)(pixelUV);
+ float frac = pixelUV - (float)iPixel;
+
+ if (frac < 0.5f) {
+ iPixel -= 1;
+ frac += 0.5f;
+ } else {
+ frac -= 0.5f;
+ }
+
+ float oneMinusFrac = 1.0f - frac;
+
+ float2 weights;
+ weights.x = oneMinusFrac;
+ weights.y = frac;
+
+ uint32_t next = wrapI(wrapS, iPixel + 1, sourceW);
+ uint32_t location = wrapI(wrapS, iPixel, sourceW);
+
+ return getBilinearSample1D(alloc, weights, location, next, dk, dt, lod);
+}
+
+static float4 __attribute__((overloadable))
+ sample_LOD_NearestPixel(const Allocation_t *alloc,
+ rs_data_kind dk, rs_data_type dt,
+ rs_sampler_value wrapS,
+ float uv, uint32_t lod) {
+
+ int32_t sourceW = alloc->mHal.drvState.lod[lod].dimX;
+ int32_t iPixel = (int32_t)(uv * (float)(sourceW));
+ uint32_t location = wrapI(wrapS, iPixel, sourceW);
+
+ return getNearestSample(alloc, location, dk, dt, lod);
+}
+
+static float4 __attribute__((overloadable))
+ sample_LOD_LinearPixel(const Allocation_t *alloc,
+ rs_data_kind dk, rs_data_type dt,
+ rs_sampler_value wrapS,
+ rs_sampler_value wrapT,
+ float2 uv, uint32_t lod) {
+
+ const uint8_t *p = (const uint8_t *)alloc->mHal.drvState.lod[lod].mallocPtr;
+
+ int sourceW = alloc->mHal.drvState.lod[lod].dimX;
+ int sourceH = alloc->mHal.drvState.lod[lod].dimY;
+
+ float pixelU = uv.x * sourceW;
+ float pixelV = uv.y * sourceH;
+ int iPixelU = pixelU;
+ int iPixelV = pixelV;
+ float fracU = pixelU - iPixelU;
+ float fracV = pixelV - iPixelV;
+
+ if (fracU < 0.5f) {
+ iPixelU -= 1;
+ fracU += 0.5f;
+ } else {
+ fracU -= 0.5f;
+ }
+ if (fracV < 0.5f) {
+ iPixelV -= 1;
+ fracV += 0.5f;
+ } else {
+ fracV -= 0.5f;
+ }
+ float oneMinusFracU = 1.0f - fracU;
+ float oneMinusFracV = 1.0f - fracV;
+
+ float w0 = oneMinusFracU * oneMinusFracV;
+ float w1 = fracU * oneMinusFracV;
+ float w2 = oneMinusFracU * fracV;
+ float w3 = fracU * fracV;
+
+ int nx = wrapI(wrapS, iPixelU + 1, sourceW);
+ int ny = wrapI(wrapT, iPixelV + 1, sourceH);
+ int lx = wrapI(wrapS, iPixelU, sourceW);
+ int ly = wrapI(wrapT, iPixelV, sourceH);
+
+ return getBilinearSample2D(alloc, w0, w1, w2, w3, lx, ly, nx, ny, dk, dt, lod);
+
+}
+
+static float4 __attribute__((overloadable))
+ sample_LOD_NearestPixel(const Allocation_t *alloc,
+ rs_data_kind dk, rs_data_type dt,
+ rs_sampler_value wrapS,
+ rs_sampler_value wrapT,
+ float2 uv, uint32_t lod) {
+ int sourceW = alloc->mHal.drvState.lod[lod].dimX;
+ int sourceH = alloc->mHal.drvState.lod[lod].dimY;
+
+ float2 dimF;
+ dimF.x = (float)(sourceW);
+ dimF.y = (float)(sourceH);
+ int2 iPixel = convert_int2(uv * dimF);
+
+ uint2 location;
+ location.x = wrapI(wrapS, iPixel.x, sourceW);
+ location.y = wrapI(wrapT, iPixel.y, sourceH);
+ return getNearestSample(alloc, location, dk, dt, lod);
+}
+
+extern const float4 __attribute__((overloadable))
+ rsSample(rs_allocation a, rs_sampler s, float uv, float lod) {
+
+ const Allocation_t *alloc = (const Allocation_t *)a.p;
+ const Sampler_t *prog = (Sampler_t *)s.p;
+ const Type_t *type = (Type_t *)alloc->mHal.state.type;
+ const Element_t *elem = type->mHal.state.element;
+ rs_data_kind dk = elem->mHal.state.dataKind;
+ rs_data_type dt = elem->mHal.state.dataType;
+ rs_sampler_value sampleMin = prog->mHal.state.minFilter;
+ rs_sampler_value sampleMag = prog->mHal.state.magFilter;
+ rs_sampler_value wrapS = prog->mHal.state.wrapS;
+
+ if (!(alloc->mHal.state.usageFlags & RS_ALLOCATION_USAGE_GRAPHICS_TEXTURE)) {
+ return 0.f;
+ }
+
+ if (lod <= 0.0f) {
+ if (sampleMag == RS_SAMPLER_NEAREST) {
+ return sample_LOD_NearestPixel(alloc, dk, dt, wrapS, uv, 0);
+ }
+ return sample_LOD_LinearPixel(alloc, dk, dt, wrapS, uv, 0);
+ }
+
+ if (sampleMin == RS_SAMPLER_LINEAR_MIP_NEAREST) {
+ uint32_t maxLOD = type->mHal.state.lodCount - 1;
+ lod = min(lod, (float)maxLOD);
+ uint32_t nearestLOD = (uint32_t)round(lod);
+ return sample_LOD_LinearPixel(alloc, dk, dt, wrapS, uv, nearestLOD);
+ }
+
+ if (sampleMin == RS_SAMPLER_LINEAR_MIP_LINEAR) {
+ uint32_t lod0 = (uint32_t)floor(lod);
+ uint32_t lod1 = (uint32_t)ceil(lod);
+ uint32_t maxLOD = type->mHal.state.lodCount - 1;
+ lod0 = min(lod0, maxLOD);
+ lod1 = min(lod1, maxLOD);
+ float4 sample0 = sample_LOD_LinearPixel(alloc, dk, dt, wrapS, uv, lod0);
+ float4 sample1 = sample_LOD_LinearPixel(alloc, dk, dt, wrapS, uv, lod1);
+ float frac = lod - (float)lod0;
+ return sample0 * (1.0f - frac) + sample1 * frac;
+ }
+
+ return sample_LOD_NearestPixel(alloc, dk, dt, wrapS, uv, 0);
+}
+
+extern const float4 __attribute__((overloadable))
+ rsSample(rs_allocation a, rs_sampler s, float location) {
+ return rsSample(a, s, location, 0);
+}
+
+
+extern const float4 __attribute__((overloadable))
+ rsSample(rs_allocation a, rs_sampler s, float2 uv, float lod) {
+
+ const Allocation_t *alloc = (const Allocation_t *)a.p;
+ const Sampler_t *prog = (Sampler_t *)s.p;
+ const Type_t *type = (Type_t *)alloc->mHal.state.type;
+ const Element_t *elem = type->mHal.state.element;
+ rs_data_kind dk = elem->mHal.state.dataKind;
+ rs_data_type dt = elem->mHal.state.dataType;
+ rs_sampler_value sampleMin = prog->mHal.state.minFilter;
+ rs_sampler_value sampleMag = prog->mHal.state.magFilter;
+ rs_sampler_value wrapS = prog->mHal.state.wrapS;
+ rs_sampler_value wrapT = prog->mHal.state.wrapT;
+
+ if (!(alloc->mHal.state.usageFlags & RS_ALLOCATION_USAGE_GRAPHICS_TEXTURE)) {
+ return 0.f;
+ }
+
+ if (lod <= 0.0f) {
+ if (sampleMag == RS_SAMPLER_NEAREST) {
+ return sample_LOD_NearestPixel(alloc, dk, dt, wrapS, wrapT, uv, 0);
+ }
+ return sample_LOD_LinearPixel(alloc, dk, dt, wrapS, wrapT, uv, 0);
+ }
+
+ if (sampleMin == RS_SAMPLER_LINEAR_MIP_NEAREST) {
+ uint32_t maxLOD = type->mHal.state.lodCount - 1;
+ lod = min(lod, (float)maxLOD);
+ uint32_t nearestLOD = (uint32_t)round(lod);
+ return sample_LOD_LinearPixel(alloc, dk, dt, wrapS, wrapT, uv, nearestLOD);
+ }
+
+ if (sampleMin == RS_SAMPLER_LINEAR_MIP_LINEAR) {
+ uint32_t lod0 = (uint32_t)floor(lod);
+ uint32_t lod1 = (uint32_t)ceil(lod);
+ uint32_t maxLOD = type->mHal.state.lodCount - 1;
+ lod0 = min(lod0, maxLOD);
+ lod1 = min(lod1, maxLOD);
+ float4 sample0 = sample_LOD_LinearPixel(alloc, dk, dt, wrapS, wrapT, uv, lod0);
+ float4 sample1 = sample_LOD_LinearPixel(alloc, dk, dt, wrapS, wrapT, uv, lod1);
+ float frac = lod - (float)lod0;
+ return sample0 * (1.0f - frac) + sample1 * frac;
+ }
+
+ return sample_LOD_NearestPixel(alloc, dk, dt, wrapS, wrapT, uv, 0);
+}
+
+extern const float4 __attribute__((overloadable))
+ rsSample(rs_allocation a, rs_sampler s, float2 uv) {
+
+ const Allocation_t *alloc = (const Allocation_t *)a.p;
+ const Sampler_t *prog = (Sampler_t *)s.p;
+ const Type_t *type = (Type_t *)alloc->mHal.state.type;
+ const Element_t *elem = type->mHal.state.element;
+ rs_data_kind dk = elem->mHal.state.dataKind;
+ rs_data_type dt = elem->mHal.state.dataType;
+ rs_sampler_value wrapS = prog->mHal.state.wrapS;
+ rs_sampler_value wrapT = prog->mHal.state.wrapT;
+
+ if (!(alloc->mHal.state.usageFlags & RS_ALLOCATION_USAGE_GRAPHICS_TEXTURE)) {
+ return 0.f;
+ }
+
+ if (prog->mHal.state.magFilter == RS_SAMPLER_NEAREST) {
+ return sample_LOD_NearestPixel(alloc, dk, dt, wrapS, wrapT, uv, 0);
+ }
+ return sample_LOD_LinearPixel(alloc, dk, dt, wrapS, wrapT, uv, 0);
+}
+
diff --git a/driver/runtime/rs_sampler.c b/driver/runtime/rs_sampler.c
new file mode 100644
index 0000000..39782de
--- /dev/null
+++ b/driver/runtime/rs_sampler.c
@@ -0,0 +1,51 @@
+#include "rs_core.rsh"
+#include "rs_graphics.rsh"
+#include "rs_structs.h"
+
+/**
+* Sampler
+*/
+extern rs_sampler_value __attribute__((overloadable))
+ rsSamplerGetMinification(rs_sampler s) {
+ Sampler_t *prog = (Sampler_t *)s.p;
+ if (prog == NULL) {
+ return RS_SAMPLER_INVALID;
+ }
+ return prog->mHal.state.minFilter;
+}
+
+extern rs_sampler_value __attribute__((overloadable))
+ rsSamplerGetMagnification(rs_sampler s) {
+ Sampler_t *prog = (Sampler_t *)s.p;
+ if (prog == NULL) {
+ return RS_SAMPLER_INVALID;
+ }
+ return prog->mHal.state.magFilter;
+}
+
+extern rs_sampler_value __attribute__((overloadable))
+ rsSamplerGetWrapS(rs_sampler s) {
+ Sampler_t *prog = (Sampler_t *)s.p;
+ if (prog == NULL) {
+ return RS_SAMPLER_INVALID;
+ }
+ return prog->mHal.state.wrapS;
+}
+
+extern rs_sampler_value __attribute__((overloadable))
+ rsSamplerGetWrapT(rs_sampler s) {
+ Sampler_t *prog = (Sampler_t *)s.p;
+ if (prog == NULL) {
+ return RS_SAMPLER_INVALID;
+ }
+ return prog->mHal.state.wrapT;
+}
+
+extern float __attribute__((overloadable))
+ rsSamplerGetAnisotropy(rs_sampler s) {
+ Sampler_t *prog = (Sampler_t *)s.p;
+ if (prog == NULL) {
+ return 0.0f;
+ }
+ return prog->mHal.state.aniso;
+}
diff --git a/driver/runtime/rs_structs.h b/driver/runtime/rs_structs.h
new file mode 100644
index 0000000..204717c
--- /dev/null
+++ b/driver/runtime/rs_structs.h
@@ -0,0 +1,262 @@
+#ifndef _RS_STRUCTS_H_
+#define _RS_STRUCTS_H_
+
+/*****************************************************************************
+ * CAUTION
+ *
+ * The following structure layout provides a more efficient way to access
+ * internal members of the C++ class Allocation owned by librs. Unfortunately,
+ * since this class has virtual members, we can't simply use offsetof() or any
+ * other compiler trickery to dynamically get the appropriate values at
+ * build-time. This layout may need to be updated whenever
+ * frameworks/base/libs/rs/rsAllocation.h is modified.
+ *
+ * Having the layout information available in this file allows us to
+ * accelerate functionality like rsAllocationGetDimX(). Without this
+ * information, we would not be able to inline the bitcode, thus resulting in
+ * potential runtime performance penalties for tight loops operating on
+ * allocations.
+ *
+ *****************************************************************************/
+typedef enum {
+ RS_ALLOCATION_MIPMAP_NONE = 0,
+ RS_ALLOCATION_MIPMAP_FULL = 1,
+ RS_ALLOCATION_MIPMAP_ON_SYNC_TO_TEXTURE = 2
+} rs_allocation_mipmap_control;
+
+typedef struct Allocation {
+ char __pad[32];
+ struct {
+ void * drv;
+ struct {
+ const void *type;
+ uint32_t usageFlags;
+ rs_allocation_mipmap_control mipmapControl;
+ uint32_t yuv;
+ uint32_t elementSizeBytes;
+ bool hasMipmaps;
+ bool hasFaces;
+ bool hasReferences;
+ void * usrPtr;
+ int32_t surfaceTextureID;
+ void * nativeBuffer;
+ int64_t timestamp;
+ } state;
+
+ struct DrvState {
+ struct LodState {
+ void * mallocPtr;
+ size_t stride;
+ uint32_t dimX;
+ uint32_t dimY;
+ uint32_t dimZ;
+ } lod[16/*android::renderscript::Allocation::MAX_LOD*/];
+ size_t faceOffset;
+ uint32_t lodCount;
+ uint32_t faceCount;
+ } drvState;
+ } mHal;
+} Allocation_t;
+
+/*****************************************************************************
+ * CAUTION
+ *
+ * The following structure layout provides a more efficient way to access
+ * internal members of the C++ class ProgramStore owned by librs. Unfortunately,
+ * since this class has virtual members, we can't simply use offsetof() or any
+ * other compiler trickery to dynamically get the appropriate values at
+ * build-time. This layout may need to be updated whenever
+ * frameworks/base/libs/rs/rsProgramStore.h is modified.
+ *
+ * Having the layout information available in this file allows us to
+ * accelerate functionality like rsgProgramStoreGetDepthFunc(). Without this
+ * information, we would not be able to inline the bitcode, thus resulting in
+ * potential runtime performance penalties for tight loops operating on
+ * program store.
+ *
+ *****************************************************************************/
+typedef struct ProgramStore {
+ char __pad[40];
+ struct {
+ struct {
+ bool ditherEnable;
+ bool colorRWriteEnable;
+ bool colorGWriteEnable;
+ bool colorBWriteEnable;
+ bool colorAWriteEnable;
+ rs_blend_src_func blendSrc;
+ rs_blend_dst_func blendDst;
+ bool depthWriteEnable;
+ rs_depth_func depthFunc;
+ } state;
+ } mHal;
+} ProgramStore_t;
+
+/*****************************************************************************
+ * CAUTION
+ *
+ * The following structure layout provides a more efficient way to access
+ * internal members of the C++ class ProgramRaster owned by librs. Unfortunately,
+ * since this class has virtual members, we can't simply use offsetof() or any
+ * other compiler trickery to dynamically get the appropriate values at
+ * build-time. This layout may need to be updated whenever
+ * frameworks/base/libs/rs/rsProgramRaster.h is modified.
+ *
+ * Having the layout information available in this file allows us to
+ * accelerate functionality like rsgProgramRasterGetCullMode(). Without this
+ * information, we would not be able to inline the bitcode, thus resulting in
+ * potential runtime performance penalties for tight loops operating on
+ * program raster.
+ *
+ *****************************************************************************/
+typedef struct ProgramRaster {
+ char __pad[36];
+ struct {
+ void * drv;
+ struct {
+ bool pointSprite;
+ rs_cull_mode cull;
+ } state;
+ } mHal;
+} ProgramRaster_t;
+
+/*****************************************************************************
+ * CAUTION
+ *
+ * The following structure layout provides a more efficient way to access
+ * internal members of the C++ class Sampler owned by librs. Unfortunately,
+ * since this class has virtual members, we can't simply use offsetof() or any
+ * other compiler trickery to dynamically get the appropriate values at
+ * build-time. This layout may need to be updated whenever
+ * frameworks/base/libs/rs/rsSampler.h is modified.
+ *
+ * Having the layout information available in this file allows us to
+ * accelerate functionality like rsgProgramRasterGetMagFilter(). Without this
+ * information, we would not be able to inline the bitcode, thus resulting in
+ * potential runtime performance penalties for tight loops operating on
+ * samplers.
+ *
+ *****************************************************************************/
+typedef struct Sampler {
+ char __pad[32];
+ struct {
+ void *drv;
+ struct {
+ rs_sampler_value magFilter;
+ rs_sampler_value minFilter;
+ rs_sampler_value wrapS;
+ rs_sampler_value wrapT;
+ rs_sampler_value wrapR;
+ float aniso;
+ } state;
+ } mHal;
+} Sampler_t;
+
+/*****************************************************************************
+ * CAUTION
+ *
+ * The following structure layout provides a more efficient way to access
+ * internal members of the C++ class Element owned by librs. Unfortunately,
+ * since this class has virtual members, we can't simply use offsetof() or any
+ * other compiler trickery to dynamically get the appropriate values at
+ * build-time. This layout may need to be updated whenever
+ * frameworks/base/libs/rs/rsElement.h is modified.
+ *
+ * Having the layout information available in this file allows us to
+ * accelerate functionality like rsElementGetSubElementCount(). Without this
+ * information, we would not be able to inline the bitcode, thus resulting in
+ * potential runtime performance penalties for tight loops operating on
+ * elements.
+ *
+ *****************************************************************************/
+typedef struct Element {
+ char __pad[32];
+ struct {
+ void *drv;
+ struct {
+ rs_data_type dataType;
+ rs_data_kind dataKind;
+ uint32_t vectorSize;
+ uint32_t elementSizeBytes;
+
+ // Subelements
+ const void **fields;
+ uint32_t *fieldArraySizes;
+ const char **fieldNames;
+ uint32_t *fieldNameLengths;
+ uint32_t *fieldOffsetBytes;
+ uint32_t fieldsCount;
+ } state;
+ } mHal;
+} Element_t;
+
+/*****************************************************************************
+ * CAUTION
+ *
+ * The following structure layout provides a more efficient way to access
+ * internal members of the C++ class Type owned by librs. Unfortunately,
+ * since this class has virtual members, we can't simply use offsetof() or any
+ * other compiler trickery to dynamically get the appropriate values at
+ * build-time. This layout may need to be updated whenever
+ * frameworks/base/libs/rs/rsType.h is modified.
+ *
+ * Having the layout information available in this file allows us to
+ * accelerate functionality like rsAllocationGetElement(). Without this
+ * information, we would not be able to inline the bitcode, thus resulting in
+ * potential runtime performance penalties for tight loops operating on
+ * types.
+ *
+ *****************************************************************************/
+typedef struct Type {
+ char __pad[32];
+ struct {
+ void *drv;
+ struct {
+ const void * element;
+ uint32_t dimX;
+ uint32_t dimY;
+ uint32_t dimZ;
+ uint32_t *lodDimX;
+ uint32_t *lodDimY;
+ uint32_t *lodDimZ;
+ uint32_t *lodOffset;
+ uint32_t lodCount;
+ bool faces;
+ } state;
+ } mHal;
+} Type_t;
+
+/*****************************************************************************
+ * CAUTION
+ *
+ * The following structure layout provides a more efficient way to access
+ * internal members of the C++ class Mesh owned by librs. Unfortunately,
+ * since this class has virtual members, we can't simply use offsetof() or any
+ * other compiler trickery to dynamically get the appropriate values at
+ * build-time. This layout may need to be updated whenever
+ * frameworks/base/libs/rs/rsMesh.h is modified.
+ *
+ * Having the layout information available in this file allows us to
+ * accelerate functionality like rsMeshGetVertexAllocationCount(). Without this
+ * information, we would not be able to inline the bitcode, thus resulting in
+ * potential runtime performance penalties for tight loops operating on
+ * meshes.
+ *
+ *****************************************************************************/
+typedef struct Mesh {
+ char __pad[32];
+ struct {
+ void *drv;
+ struct {
+ void **vertexBuffers;
+ uint32_t vertexBuffersCount;
+
+ // indexBuffers[i] could be NULL, in which case only primitives[i] is used
+ void **indexBuffers;
+ uint32_t indexBuffersCount;
+ rs_primitive *primitives;
+ uint32_t primitivesCount;
+ } state;
+ } mHal;
+} Mesh_t;
+#endif // _RS_CORE_H_
diff --git a/driver/linkloader/tests/images/clean-testcases.sh b/java/tests/GenImages/Android.mk
old mode 100755
new mode 100644
similarity index 63%
rename from driver/linkloader/tests/images/clean-testcases.sh
rename to java/tests/GenImages/Android.mk
index 1c2d0be..69ef52c
--- a/driver/linkloader/tests/images/clean-testcases.sh
+++ b/java/tests/GenImages/Android.mk
@@ -1,7 +1,5 @@
-#!/bin/bash -e
-
-
-# Copyright (C) 2011 The Android Open Source Project
+#
+# Copyright (C) 2013 The Android Open Source Project
#
# Licensed under the Apache License, Version 2.0 (the "License");
# you may not use this file except in compliance with the License.
@@ -14,5 +12,16 @@
# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
# See the License for the specific language governing permissions and
# limitations under the License.
+#
-rm -rf *.o *.bc
+LOCAL_PATH := $(call my-dir)
+include $(CLEAR_VARS)
+
+LOCAL_MODULE_TAGS := tests
+
+LOCAL_SRC_FILES := $(call all-java-files-under, src) \
+ $(call all-renderscript-files-under, src)
+
+LOCAL_PACKAGE_NAME := RsGenImages
+
+include $(BUILD_PACKAGE)
diff --git a/java/tests/GenImages/AndroidManifest.xml b/java/tests/GenImages/AndroidManifest.xml
new file mode 100644
index 0000000..c7b690e
--- /dev/null
+++ b/java/tests/GenImages/AndroidManifest.xml
@@ -0,0 +1,40 @@
+<?xml version="1.0" encoding="utf-8"?>
+<!-- Copyright (C) 2013 The Android Open Source Project
+
+ Licensed under the Apache License, Version 2.0 (the "License");
+ you may not use this file except in compliance with the License.
+ You may obtain a copy of the License at
+
+ http://www.apache.org/licenses/LICENSE-2.0
+
+ Unless required by applicable law or agreed to in writing, software
+ distributed under the License is distributed on an "AS IS" BASIS,
+ WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ See the License for the specific language governing permissions and
+ limitations under the License.
+-->
+<manifest xmlns:android="http://schemas.android.com/apk/res/android"
+ package="com.android.rs.genimage"
+ android:versionCode="1"
+ android:versionName="1.0" >
+ <uses-permission android:name="android.permission.WRITE_EXTERNAL_STORAGE" />
+
+ <uses-sdk android:minSdkVersion="14" />
+ <!-- Tell the system this app requires OpenGL ES 2.0. -->
+ <uses-feature android:glEsVersion="0x00020000" android:required="true" />
+
+ <application
+ android:icon="@drawable/ic_launcher"
+ android:label="@string/app_name" >
+ <activity
+ android:name="com.android.rs.genimage.GenImageAct"
+ android:label="@string/app_name" >
+ <intent-filter>
+ <action android:name="android.intent.action.MAIN" />
+
+ <category android:name="android.intent.category.LAUNCHER" />
+ </intent-filter>
+ </activity>
+ </application>
+
+</manifest>
diff --git a/java/tests/GenImages/res/drawable-nodpi/test_pattern.png b/java/tests/GenImages/res/drawable-nodpi/test_pattern.png
new file mode 100644
index 0000000..e7d1455
--- /dev/null
+++ b/java/tests/GenImages/res/drawable-nodpi/test_pattern.png
Binary files differ
diff --git a/java/tests/GenImages/res/drawable/ic_launcher.png b/java/tests/GenImages/res/drawable/ic_launcher.png
new file mode 100644
index 0000000..359047d
--- /dev/null
+++ b/java/tests/GenImages/res/drawable/ic_launcher.png
Binary files differ
diff --git a/java/tests/GenImages/res/layout/main.xml b/java/tests/GenImages/res/layout/main.xml
new file mode 100644
index 0000000..b8db6d8
--- /dev/null
+++ b/java/tests/GenImages/res/layout/main.xml
@@ -0,0 +1,26 @@
+<?xml version="1.0" encoding="utf-8"?>
+<!-- Copyright (C) 2013 The Android Open Source Project
+
+ Licensed under the Apache License, Version 2.0 (the "License");
+ you may not use this file except in compliance with the License.
+ You may obtain a copy of the License at
+
+ http://www.apache.org/licenses/LICENSE-2.0
+
+ Unless required by applicable law or agreed to in writing, software
+ distributed under the License is distributed on an "AS IS" BASIS,
+ WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ See the License for the specific language governing permissions and
+ limitations under the License.
+-->
+<LinearLayout xmlns:android="http://schemas.android.com/apk/res/android"
+ android:layout_width="fill_parent"
+ android:layout_height="fill_parent"
+ android:orientation="vertical" >
+
+ <TextView
+ android:layout_width="fill_parent"
+ android:layout_height="wrap_content"
+ android:text="@string/hello" />
+
+</LinearLayout>
diff --git a/java/tests/GenImages/res/values/strings.xml b/java/tests/GenImages/res/values/strings.xml
new file mode 100644
index 0000000..a0cc805
--- /dev/null
+++ b/java/tests/GenImages/res/values/strings.xml
@@ -0,0 +1,21 @@
+<?xml version="1.0" encoding="utf-8"?>
+<!-- Copyright (C) 2013 The Android Open Source Project
+
+ Licensed under the Apache License, Version 2.0 (the "License");
+ you may not use this file except in compliance with the License.
+ You may obtain a copy of the License at
+
+ http://www.apache.org/licenses/LICENSE-2.0
+
+ Unless required by applicable law or agreed to in writing, software
+ distributed under the License is distributed on an "AS IS" BASIS,
+ WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ See the License for the specific language governing permissions and
+ limitations under the License.
+-->
+<resources>
+
+ <string name="hello">Hello</string>
+ <string name="app_name">RS Image Generator</string>
+
+</resources>
diff --git a/java/tests/GenImages/src/com/android/rs/genimage/GenImage.java b/java/tests/GenImages/src/com/android/rs/genimage/GenImage.java
new file mode 100644
index 0000000..f238095
--- /dev/null
+++ b/java/tests/GenImages/src/com/android/rs/genimage/GenImage.java
@@ -0,0 +1,271 @@
+/*
+ * Copyright (C) 2013 The Android Open Source Project
+ *
+ * Licensed under the Apache License, Version 2.0 (the "License");
+ * you may not use this file except in compliance with the License.
+ * You may obtain a copy of the License at
+ *
+ * http://www.apache.org/licenses/LICENSE-2.0
+ *
+ * Unless required by applicable law or agreed to in writing, software
+ * distributed under the License is distributed on an "AS IS" BASIS,
+ * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ * See the License for the specific language governing permissions and
+ * limitations under the License.
+ */
+
+package com.android.rs.genimage;
+
+import android.content.Context;
+
+import java.io.FileOutputStream;
+import java.nio.ByteBuffer;
+import java.nio.IntBuffer;
+import java.nio.ByteOrder;
+import java.nio.FloatBuffer;
+
+import android.graphics.Bitmap;
+import android.graphics.BitmapFactory;
+
+import javax.microedition.khronos.egl.EGLConfig;
+import javax.microedition.khronos.opengles.GL10;
+import android.opengl.GLUtils;
+
+import android.opengl.GLES20;
+import android.opengl.GLSurfaceView;
+
+public class GenImage implements GLSurfaceView.Renderer {
+ private Bitmap mTestImage;
+
+ private Triangle mTriangle;
+
+
+ private Bitmap loadBitmap(Context context, int resource) {
+ final BitmapFactory.Options options = new BitmapFactory.Options();
+ options.inPreferredConfig = Bitmap.Config.ARGB_8888;
+ return BitmapFactory.decodeResource(context.getResources(), resource, options);
+ }
+
+ GenImage(Context context) {
+
+ mTestImage = loadBitmap(context, R.drawable.test_pattern);
+
+ }
+
+ @Override
+ public void onSurfaceCreated(GL10 unused, EGLConfig config) {
+
+ // Set the background frame color
+ GLES20.glClearColor(0.0f, 0.0f, 0.0f, 1.0f);
+
+ mTriangle = new Triangle(mTestImage);
+ }
+
+ @Override
+ public void onDrawFrame(GL10 unused) {
+
+ // Draw background color
+ GLES20.glClear(GLES20.GL_COLOR_BUFFER_BIT);
+
+ // Draw triangle
+ mTriangle.draw();
+ }
+
+ @Override
+ public void onSurfaceChanged(GL10 unused, int width, int height) {
+ // Adjust the viewport based on geometry changes,
+ // such as screen rotation
+ GLES20.glViewport(0, 0, 512, 512);
+ }
+
+ public static int loadShader(int type, String shaderCode){
+
+ // create a vertex shader type (GLES20.GL_VERTEX_SHADER)
+ // or a fragment shader type (GLES20.GL_FRAGMENT_SHADER)
+ int shader = GLES20.glCreateShader(type);
+
+ // add the source code to the shader and compile it
+ GLES20.glShaderSource(shader, shaderCode);
+ GLES20.glCompileShader(shader);
+
+ return shader;
+ }
+
+}
+
+
+
+class Triangle {
+ int mTextureIDs[] = new int[1];
+
+ private final String vertexShaderCode =
+ "varying vec2 vTex0;" +
+ "varying vec2 vPos0;" +
+ "attribute vec4 aPosition;" +
+ "void main() {" +
+ " gl_Position = aPosition;" +
+ " vPos0 = aPosition.xy;" +
+ " vTex0 = ((aPosition.xy + 1.0) * 0.6);" +
+ //" vTex0 = (aPosition.xy * 1.7) + 0.5;" +
+ "}";
+
+ private final String fragmentShaderCode =
+ "precision mediump float;" +
+ "varying vec2 vTex0;" +
+ "varying vec2 vPos0;" +
+ "uniform sampler2D uSamp;" +
+ "void main() {" +
+ " vec2 tc = vTex0;" +
+ //" tc.x *= pow(vPos0.y + 1.0, 2.0);" +
+ //" tc.y *= pow(vPos0.x + 1.0, 2.0);" +
+ " vec4 c = texture2D(uSamp, tc);" +
+ " c.a = 1.0;" +
+ " gl_FragColor = c;" +
+ "}";
+
+ private final FloatBuffer vertexBuffer;
+ private final int mProgram;
+
+ // number of coordinates per vertex in this array
+ static float triangleCoords[] = { // in counterclockwise order:
+ -1.0f, 1.0f, 0.0f, // top left
+ -1.0f, -1.0f, 0.0f, // bottom left
+ 1.0f, -1.0f, 0.0f, // bottom right
+
+ -1.0f, 1.0f, 0.0f, // top left
+ 1.0f, -1.0f, 0.0f, // bottom right
+ 1.0f, 1.0f, 0.0f // top right
+ };
+
+ FloatBuffer createFloatBuffer(float buf[]) {
+ ByteBuffer bb = ByteBuffer.allocateDirect(buf.length * 4);
+ bb.order(ByteOrder.nativeOrder());
+ FloatBuffer fb = bb.asFloatBuffer();
+ fb.put(buf);
+ fb.position(0);
+ return fb;
+ }
+
+ public String setup(int key) {
+ String s = new String();
+ int tmp;
+
+ tmp = key % 2;
+ key /= 2;
+ if (tmp != 0) {
+ s += "N";
+ GLES20.glTexParameteri(GLES20.GL_TEXTURE_2D, GLES20.GL_TEXTURE_MIN_FILTER, GLES20.GL_NEAREST);
+ } else {
+ s += "L";
+ GLES20.glTexParameteri(GLES20.GL_TEXTURE_2D, GLES20.GL_TEXTURE_MIN_FILTER, GLES20.GL_LINEAR);
+ }
+
+ tmp = key % 2;
+ key /= 2;
+ if (tmp != 0) {
+ s += "N";
+ GLES20.glTexParameteri(GLES20.GL_TEXTURE_2D, GLES20.GL_TEXTURE_MAG_FILTER, GLES20.GL_NEAREST);
+ } else {
+ s += "L";
+ GLES20.glTexParameteri(GLES20.GL_TEXTURE_2D, GLES20.GL_TEXTURE_MAG_FILTER, GLES20.GL_LINEAR);
+ }
+
+ tmp = key % 3;
+ key /= 3;
+ switch(tmp) {
+ case 0:
+ s += "_CE";
+ GLES20.glTexParameteri(GLES20.GL_TEXTURE_2D, GLES20.GL_TEXTURE_WRAP_S, GLES20.GL_CLAMP_TO_EDGE);
+ break;
+ case 1:
+ s += "_RE";
+ GLES20.glTexParameteri(GLES20.GL_TEXTURE_2D, GLES20.GL_TEXTURE_WRAP_S, GLES20.GL_REPEAT);
+ break;
+ case 2:
+ s += "_MR";
+ GLES20.glTexParameteri(GLES20.GL_TEXTURE_2D, GLES20.GL_TEXTURE_WRAP_S, GLES20.GL_MIRRORED_REPEAT);
+ break;
+ }
+
+ tmp = key % 3;
+ key /= 3;
+ switch(tmp) {
+ case 0:
+ s += "_CE";
+ GLES20.glTexParameteri(GLES20.GL_TEXTURE_2D, GLES20.GL_TEXTURE_WRAP_T, GLES20.GL_CLAMP_TO_EDGE);
+ break;
+ case 1:
+ s += "_RE";
+ GLES20.glTexParameteri(GLES20.GL_TEXTURE_2D, GLES20.GL_TEXTURE_WRAP_T, GLES20.GL_REPEAT);
+ break;
+ case 2:
+ s += "_MR";
+ GLES20.glTexParameteri(GLES20.GL_TEXTURE_2D, GLES20.GL_TEXTURE_WRAP_T, GLES20.GL_MIRRORED_REPEAT);
+ break;
+ }
+
+ if (key > 0) done = true;
+ return s;
+ }
+
+ public Triangle(Bitmap testImage) {
+ vertexBuffer = createFloatBuffer(triangleCoords);
+
+ // prepare shaders and OpenGL program
+ int vertexShader = GenImage.loadShader(GLES20.GL_VERTEX_SHADER, vertexShaderCode);
+ int fragmentShader = GenImage.loadShader(GLES20.GL_FRAGMENT_SHADER, fragmentShaderCode);
+
+ mProgram = GLES20.glCreateProgram(); // create empty OpenGL Program
+ GLES20.glAttachShader(mProgram, vertexShader); // add the vertex shader to program
+ GLES20.glAttachShader(mProgram, fragmentShader); // add the fragment shader to program
+ GLES20.glLinkProgram(mProgram); // create OpenGL program executables
+
+ GLES20.glGenTextures(1, mTextureIDs, 0);
+
+ // Bind to the texture in OpenGL
+ GLES20.glBindTexture(GLES20.GL_TEXTURE_2D, mTextureIDs[0]);
+ GLUtils.texImage2D(GLES20.GL_TEXTURE_2D, 0, testImage, 0);
+ }
+
+ boolean done = false;
+ int key = 0;
+
+ public void draw() {
+ GLES20.glActiveTexture(GLES20.GL_TEXTURE0);
+ String ext = setup(key++);
+
+ // Add program to OpenGL environment
+ GLES20.glUseProgram(mProgram);
+
+ int posA = GLES20.glGetAttribLocation(mProgram, "aPosition");
+ GLES20.glEnableVertexAttribArray(posA);
+ GLES20.glVertexAttribPointer(posA, 3, GLES20.GL_FLOAT, false, 12, vertexBuffer);
+
+ int sampUni = GLES20.glGetUniformLocation(mProgram, "uSamp");
+ GLES20.glUniform1i(sampUni, 0);
+
+ // Draw the triangle
+ GLES20.glDrawArrays(GLES20.GL_TRIANGLES, 0, triangleCoords.length / 3);
+
+ if (!done) {
+ IntBuffer ib = IntBuffer.allocate(512*512);
+ ib.position(0);
+ GLES20.glReadPixels(0,0, 512, 512, GLES20.GL_RGBA,
+ GLES20.GL_UNSIGNED_BYTE, ib);
+
+ Bitmap bmp = Bitmap.createBitmap(512, 512, Bitmap.Config.ARGB_8888);
+ bmp.setPixels(ib.array(), 0, 512, 0, 0, 512, 512);
+
+ try {
+ String s = new String("/sdcard/imgs/RsSampImg_");
+ s += ext + ".png";
+ FileOutputStream out = new FileOutputStream(s);
+ bmp.compress(Bitmap.CompressFormat.PNG, 95, out);
+ out.close();
+ } catch (Exception e) {
+ e.printStackTrace();
+ }
+ bmp.recycle();
+ }
+ }
+}
diff --git a/java/tests/GenImages/src/com/android/rs/genimage/GenImageAct.java b/java/tests/GenImages/src/com/android/rs/genimage/GenImageAct.java
new file mode 100644
index 0000000..ea60d59
--- /dev/null
+++ b/java/tests/GenImages/src/com/android/rs/genimage/GenImageAct.java
@@ -0,0 +1,72 @@
+/*
+ * Copyright (C) 2013 The Android Open Source Project
+ *
+ * Licensed under the Apache License, Version 2.0 (the "License");
+ * you may not use this file except in compliance with the License.
+ * You may obtain a copy of the License at
+ *
+ * http://www.apache.org/licenses/LICENSE-2.0
+ *
+ * Unless required by applicable law or agreed to in writing, software
+ * distributed under the License is distributed on an "AS IS" BASIS,
+ * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ * See the License for the specific language governing permissions and
+ * limitations under the License.
+ */
+
+package com.android.rs.genimage;
+
+import android.app.Activity;
+import android.content.Context;
+import android.opengl.GLSurfaceView;
+import android.os.Bundle;
+
+public class GenImageAct extends Activity {
+
+ private GLSurfaceView mGLView;
+
+ @Override
+ public void onCreate(Bundle savedInstanceState) {
+ super.onCreate(savedInstanceState);
+
+ // Create a GLSurfaceView instance and set it
+ // as the ContentView for this Activity
+ mGLView = new MyGLSurfaceView(this);
+ setContentView(mGLView);
+ }
+
+ @Override
+ protected void onPause() {
+ super.onPause();
+ // The following call pauses the rendering thread.
+ // If your OpenGL application is memory intensive,
+ // you should consider de-allocating objects that
+ // consume significant memory here.
+ mGLView.onPause();
+ }
+
+ @Override
+ protected void onResume() {
+ super.onResume();
+ // The following call resumes a paused rendering thread.
+ // If you de-allocated graphic objects for onPause()
+ // this is a good place to re-allocate them.
+ mGLView.onResume();
+ }
+}
+
+class MyGLSurfaceView extends GLSurfaceView {
+
+ public MyGLSurfaceView(Context context) {
+ super(context);
+
+ // Create an OpenGL ES 2.0 context.
+ setEGLContextClientVersion(2);
+
+ // Set the Renderer for drawing on the GLSurfaceView
+ setRenderer(new GenImage(context));
+
+ // Render the view only when there is a change in the drawing data
+ setRenderMode(GLSurfaceView.RENDERMODE_CONTINUOUSLY);
+ }
+}
diff --git a/java/tests/ImageProcessing/src/com/android/rs/image/Artistic1.java b/java/tests/ImageProcessing/src/com/android/rs/image/Artistic1.java
new file mode 100644
index 0000000..f6288d3
--- /dev/null
+++ b/java/tests/ImageProcessing/src/com/android/rs/image/Artistic1.java
@@ -0,0 +1,44 @@
+/*
+ * Copyright (C) 2012 The Android Open Source Project
+ *
+ * Licensed under the Apache License, Version 2.0 (the "License");
+ * you may not use this file except in compliance with the License.
+ * You may obtain a copy of the License at
+ *
+ * http://www.apache.org/licenses/LICENSE-2.0
+ *
+ * Unless required by applicable law or agreed to in writing, software
+ * distributed under the License is distributed on an "AS IS" BASIS,
+ * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ * See the License for the specific language governing permissions and
+ * limitations under the License.
+ */
+
+package com.android.rs.image;
+
+import java.lang.Math;
+
+import android.renderscript.*;
+import android.util.Log;
+
+public class Artistic1 extends TestBase {
+ private ScriptC_artistic1 mScript;
+ private Allocation mBlured;
+
+ public void createTest(android.content.res.Resources res) {
+ mScript = new ScriptC_artistic1(mRS);
+ mBlured = Allocation.createTyped(mRS, mInPixelsAllocation.getType());
+ mScript.set_gBlur(mBlured);
+
+ ScriptIntrinsicBlur blur = ScriptIntrinsicBlur.create(mRS, Element.U8_4(mRS));
+ blur.setRadius(20);
+ blur.setInput(mInPixelsAllocation);
+ blur.forEach(mBlured);
+ }
+
+ public void runTest() {
+ mScript.invoke_setup();
+ mScript.forEach_process(mInPixelsAllocation, mOutPixelsAllocation);
+ }
+
+}
diff --git a/java/tests/ImageProcessing/src/com/android/rs/image/Blend.java b/java/tests/ImageProcessing/src/com/android/rs/image/Blend.java
index 2303fc3..f385bdb 100644
--- a/java/tests/ImageProcessing/src/com/android/rs/image/Blend.java
+++ b/java/tests/ImageProcessing/src/com/android/rs/image/Blend.java
@@ -25,7 +25,6 @@
import android.renderscript.RenderScript;
import android.renderscript.Script;
import android.renderscript.ScriptC;
-import android.renderscript.ScriptGroup;
import android.renderscript.ScriptIntrinsicBlend;
import android.renderscript.Type;
import android.util.Log;
diff --git a/java/tests/ImageProcessing/src/com/android/rs/image/ColorCube.java b/java/tests/ImageProcessing/src/com/android/rs/image/ColorCube.java
index f313c46..d7f58c7 100644
--- a/java/tests/ImageProcessing/src/com/android/rs/image/ColorCube.java
+++ b/java/tests/ImageProcessing/src/com/android/rs/image/ColorCube.java
@@ -24,7 +24,6 @@
import android.renderscript.RenderScript;
import android.renderscript.Script;
import android.renderscript.ScriptC;
-import android.renderscript.ScriptGroup;
import android.renderscript.ScriptIntrinsic3DLUT;
import android.renderscript.ScriptIntrinsicColorMatrix;
import android.renderscript.Type;
diff --git a/java/tests/ImageProcessing/src/com/android/rs/image/ColorMatrix.java b/java/tests/ImageProcessing/src/com/android/rs/image/ColorMatrix.java
index 2ac40a1..4e8e06c 100644
--- a/java/tests/ImageProcessing/src/com/android/rs/image/ColorMatrix.java
+++ b/java/tests/ImageProcessing/src/com/android/rs/image/ColorMatrix.java
@@ -24,7 +24,6 @@
import android.renderscript.RenderScript;
import android.renderscript.Script;
import android.renderscript.ScriptC;
-import android.renderscript.ScriptGroup;
import android.renderscript.ScriptIntrinsicColorMatrix;
import android.renderscript.Type;
import android.util.Log;
diff --git a/java/tests/ImageProcessing/src/com/android/rs/image/Convolve3x3.java b/java/tests/ImageProcessing/src/com/android/rs/image/Convolve3x3.java
index 18e9b43..76706c9 100644
--- a/java/tests/ImageProcessing/src/com/android/rs/image/Convolve3x3.java
+++ b/java/tests/ImageProcessing/src/com/android/rs/image/Convolve3x3.java
@@ -24,7 +24,6 @@
import android.renderscript.RenderScript;
import android.renderscript.Script;
import android.renderscript.ScriptC;
-import android.renderscript.ScriptGroup;
import android.renderscript.ScriptIntrinsicConvolve3x3;
import android.renderscript.Type;
import android.util.Log;
diff --git a/java/tests/ImageProcessing/src/com/android/rs/image/Convolve5x5.java b/java/tests/ImageProcessing/src/com/android/rs/image/Convolve5x5.java
index 03b3bb8..a0f4fd1 100644
--- a/java/tests/ImageProcessing/src/com/android/rs/image/Convolve5x5.java
+++ b/java/tests/ImageProcessing/src/com/android/rs/image/Convolve5x5.java
@@ -24,7 +24,6 @@
import android.renderscript.RenderScript;
import android.renderscript.Script;
import android.renderscript.ScriptC;
-import android.renderscript.ScriptGroup;
import android.renderscript.ScriptIntrinsicConvolve5x5;
import android.renderscript.Type;
import android.util.Log;
diff --git a/java/tests/ImageProcessing/src/com/android/rs/image/Histogram.java b/java/tests/ImageProcessing/src/com/android/rs/image/Histogram.java
new file mode 100644
index 0000000..5035561
--- /dev/null
+++ b/java/tests/ImageProcessing/src/com/android/rs/image/Histogram.java
@@ -0,0 +1,75 @@
+/*
+ * Copyright (C) 2013 The Android Open Source Project
+ *
+ * Licensed under the Apache License, Version 2.0 (the "License");
+ * you may not use this file except in compliance with the License.
+ * You may obtain a copy of the License at
+ *
+ * http://www.apache.org/licenses/LICENSE-2.0
+ *
+ * Unless required by applicable law or agreed to in writing, software
+ * distributed under the License is distributed on an "AS IS" BASIS,
+ * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ * See the License for the specific language governing permissions and
+ * limitations under the License.
+ */
+
+package com.android.rs.image;
+
+import java.lang.Math;
+
+import android.renderscript.*;
+import android.util.Log;
+
+public class Histogram extends TestBase {
+ private ScriptC_histogram mScript;
+ private ScriptIntrinsicHistogram mHist;
+ private Allocation mSum;
+ private Allocation mSums;
+
+ public void createTest(android.content.res.Resources res) {
+ mScript = new ScriptC_histogram(mRS);
+ mHist = ScriptIntrinsicHistogram.create(mRS, Element.U8_4(mRS));
+
+ int w = mInPixelsAllocation.getType().getX();
+ int h = mInPixelsAllocation.getType().getY();
+ int step = 8;
+ int steps = (h + step - 1) / step;
+
+ mScript.set_gWidth(w);
+ mScript.set_gHeight(h);
+ mScript.set_gStep(step);
+ mScript.set_gSteps(steps);
+
+ Type.Builder tb = new Type.Builder(mRS, Element.I32(mRS));
+ tb.setX(256).setY(steps);
+ Type t = tb.create();
+ mSums = Allocation.createTyped(mRS, t);
+ mSum = Allocation.createSized(mRS, Element.I32(mRS), 256);
+
+ mScript.set_gSums(mSums);
+ mScript.set_gSum(mSum);
+ mScript.set_gSrc(mInPixelsAllocation);
+ mScript.set_gDest(mOutPixelsAllocation);
+
+ mScript.forEach_clear(mOutPixelsAllocation);
+ }
+
+
+
+ public void runTest() {
+ Script.LaunchOptions lo = new Script.LaunchOptions();
+ lo.setX(0, 1);
+ //mScript.forEach_pass1(mSums, lo);
+ //mScript.forEach_pass2(mSum);
+
+ mHist.setOutput(mSum);
+ mHist.forEach_dot(mInPixelsAllocation);
+
+ mScript.invoke_rescale();
+
+ lo.setX(0, 1024);
+ mScript.forEach_draw(mOutPixelsAllocation, lo);
+ }
+
+}
diff --git a/java/tests/ImageProcessing/src/com/android/rs/image/ImageProcessingActivity.java b/java/tests/ImageProcessing/src/com/android/rs/image/ImageProcessingActivity.java
index 975027a..14543c9 100644
--- a/java/tests/ImageProcessing/src/com/android/rs/image/ImageProcessingActivity.java
+++ b/java/tests/ImageProcessing/src/com/android/rs/image/ImageProcessingActivity.java
@@ -98,7 +98,9 @@
WHITE_BALANCE ("White Balance"),
COLOR_CUBE ("Color Cube"),
COLOR_CUBE_3D_INTRINSIC ("Color Cube (3D LUT intrinsic)"),
- USAGE_IO ("Usage io)");
+ USAGE_IO ("Usage io"),
+ ARTISTIC_1("Artistic 1"),
+ HISTOGRAM ("Histogram");
private final String name;
@@ -359,6 +361,12 @@
case USAGE_IO:
mTest = new UsageIO();
break;
+ case ARTISTIC_1:
+ mTest = new Artistic1();
+ break;
+ case HISTOGRAM:
+ mTest = new Histogram();
+ break;
}
mTest.createBaseTest(this, mBitmapIn, mBitmapIn2, mBitmapOut);
diff --git a/java/tests/ImageProcessing/src/com/android/rs/image/LevelsV4.java b/java/tests/ImageProcessing/src/com/android/rs/image/LevelsV4.java
index 9eb5647..98c6460 100644
--- a/java/tests/ImageProcessing/src/com/android/rs/image/LevelsV4.java
+++ b/java/tests/ImageProcessing/src/com/android/rs/image/LevelsV4.java
@@ -109,7 +109,7 @@
public boolean onBar4Setup(SeekBar b, TextView t) {
b.setMax(128);
b.setProgress(128);
- t.setText("Out White");
+ t.setText("In White");
return true;
}
public boolean onBar5Setup(SeekBar b, TextView t) {
diff --git a/java/tests/ImageProcessing/src/com/android/rs/image/UsageIO.java b/java/tests/ImageProcessing/src/com/android/rs/image/UsageIO.java
index 3f86311..9c29329 100644
--- a/java/tests/ImageProcessing/src/com/android/rs/image/UsageIO.java
+++ b/java/tests/ImageProcessing/src/com/android/rs/image/UsageIO.java
@@ -26,7 +26,6 @@
import android.renderscript.ScriptIntrinsicColorMatrix;
import android.renderscript.Type;
import android.renderscript.Matrix4f;
-import android.renderscript.ScriptGroup;
import android.util.Log;
public class UsageIO extends TestBase {
diff --git a/java/tests/ImageProcessing/src/com/android/rs/image/WhiteBalance.java b/java/tests/ImageProcessing/src/com/android/rs/image/WhiteBalance.java
index a836067..e7f12d6 100644
--- a/java/tests/ImageProcessing/src/com/android/rs/image/WhiteBalance.java
+++ b/java/tests/ImageProcessing/src/com/android/rs/image/WhiteBalance.java
@@ -19,18 +19,24 @@
import java.lang.Math;
import android.renderscript.Allocation;
+import android.renderscript.Element;
+import android.renderscript.ScriptIntrinsicHistogram;
public class WhiteBalance extends TestBase {
private ScriptC_wbalance mScript;
+ private ScriptIntrinsicHistogram mHist;
+ private Allocation mSums;
public void createTest(android.content.res.Resources res) {
mScript = new ScriptC_wbalance(mRS);
+ mHist = ScriptIntrinsicHistogram.create(mRS, Element.U8_4(mRS));
+ mSums = Allocation.createSized(mRS, Element.I32_3(mRS), 256);
+ mHist.setOutput(mSums);
+ mScript.set_histogramValues(mSums);
}
public void runTest() {
- mScript.set_histogramSource(mInPixelsAllocation);
- mScript.set_histogramWidth(mInPixelsAllocation.getType().getX());
- mScript.set_histogramHeight(mInPixelsAllocation.getType().getY());
+ mHist.forEach(mInPixelsAllocation);
mScript.invoke_prepareWhiteBalance();
mScript.forEach_whiteBalanceKernel(mInPixelsAllocation, mOutPixelsAllocation);
}
diff --git a/java/tests/ImageProcessing/src/com/android/rs/image/artistic1.rs b/java/tests/ImageProcessing/src/com/android/rs/image/artistic1.rs
new file mode 100644
index 0000000..8051f29
--- /dev/null
+++ b/java/tests/ImageProcessing/src/com/android/rs/image/artistic1.rs
@@ -0,0 +1,67 @@
+/*
+ * Copyright (C) 2012 The Android Open Source Project
+ *
+ * Licensed under the Apache License, Version 2.0 (the "License");
+ * you may not use this file except in compliance with the License.
+ * You may obtain a copy of the License at
+ *
+ * http://www.apache.org/licenses/LICENSE-2.0
+ *
+ * Unless required by applicable law or agreed to in writing, software
+ * distributed under the License is distributed on an "AS IS" BASIS,
+ * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ * See the License for the specific language governing permissions and
+ * limitations under the License.
+ */
+
+#include "ip.rsh"
+#pragma rs_fp_relaxed
+
+rs_allocation gBlur;
+
+static float gOverWm1;
+static float gOverHm1;
+static uchar gLutR[256];
+static uchar gLutG[256];
+static uchar gLutB[256];
+
+void setup() {
+ int w = rsAllocationGetDimX(gBlur);
+ int h = rsAllocationGetDimY(gBlur);
+ gOverWm1 = 1.f / w;
+ gOverHm1 = 1.f / h;
+
+ for (int x=0; x < 256; x++) {
+ gLutR[x] = x;//255-x;
+ gLutG[x] = x;//255-x;
+ gLutB[x] = x;//255-x;
+ }
+}
+
+uchar4 __attribute__((kernel)) process(uchar4 in, uint32_t x, uint32_t y) {
+ float2 xyDist;
+ xyDist.x = (x * gOverWm1 - 0.5f);
+ xyDist.y = (y * gOverHm1 - 0.5f);
+
+ // color
+ float4 v1 = rsUnpackColor8888(in);
+ float4 v2 = rsUnpackColor8888(rsGetElementAt_uchar4(gBlur, x, y));
+
+ float dist = dot(xyDist, xyDist) * 1.4f;
+ float pdist = native_powr(dist, 2.7f * 0.5f);
+ //float pdist = powr(dist, 2.7f * 0.5f);
+
+ pdist = clamp(pdist, 0.f, 1.f);
+ v1 = mix(v1, v2, dist * 2.f);
+ v1 *= 1.f - pdist;
+
+ // apply curve
+ uchar4 out = rsPackColorTo8888(v1);
+
+ out.r = gLutR[out.r];
+ out.g = gLutG[out.g];
+ out.b = gLutB[out.b];
+ return out;
+}
+
+
diff --git a/java/tests/ImageProcessing/src/com/android/rs/image/colorcube.rs b/java/tests/ImageProcessing/src/com/android/rs/image/colorcube.rs
index 4f1e73e..c0d6ace 100644
--- a/java/tests/ImageProcessing/src/com/android/rs/image/colorcube.rs
+++ b/java/tests/ImageProcessing/src/com/android/rs/image/colorcube.rs
@@ -56,16 +56,16 @@
uint4 v011 = convert_uint4(rsGetElementAt_uchar4(gCube, coord1.x, coord2.y, coord2.z));
uint4 v111 = convert_uint4(rsGetElementAt_uchar4(gCube, coord2.x, coord2.y, coord2.z));
- uint4 yz00 = ((v000 * weight1.x) + (v100 * weight2.x)) >> (int4)8;
- uint4 yz10 = ((v010 * weight1.x) + (v110 * weight2.x)) >> (int4)8;
- uint4 yz01 = ((v001 * weight1.x) + (v101 * weight2.x)) >> (int4)8;
- uint4 yz11 = ((v011 * weight1.x) + (v111 * weight2.x)) >> (int4)8;
+ uint4 yz00 = ((v000 * weight1.x) + (v100 * weight2.x)) >> (uint4)8;
+ uint4 yz10 = ((v010 * weight1.x) + (v110 * weight2.x)) >> (uint4)8;
+ uint4 yz01 = ((v001 * weight1.x) + (v101 * weight2.x)) >> (uint4)8;
+ uint4 yz11 = ((v011 * weight1.x) + (v111 * weight2.x)) >> (uint4)8;
- uint4 z0 = ((yz00 * weight1.y) + (yz10 * weight2.y)) >> (int4)16;
- uint4 z1 = ((yz01 * weight1.y) + (yz11 * weight2.y)) >> (int4)16;
+ uint4 z0 = ((yz00 * weight1.y) + (yz10 * weight2.y)) >> (uint4)16;
+ uint4 z1 = ((yz01 * weight1.y) + (yz11 * weight2.y)) >> (uint4)16;
- uint4 v = ((z0 * weight1.z) + (z1 * weight2.z)) >> (int4)16;
- uint4 v2 = (v + 0x7f) >> (int4)8;
+ uint4 v = ((z0 * weight1.z) + (z1 * weight2.z)) >> (uint4)16;
+ uint4 v2 = (v + 0x7f) >> (uint4)8;
*out = convert_uchar4(v2);
out->a = 0xff;
diff --git a/java/tests/ImageProcessing/src/com/android/rs/image/contrast.rs b/java/tests/ImageProcessing/src/com/android/rs/image/contrast.rs
index 5b67252..ef6fd63 100644
--- a/java/tests/ImageProcessing/src/com/android/rs/image/contrast.rs
+++ b/java/tests/ImageProcessing/src/com/android/rs/image/contrast.rs
@@ -27,12 +27,6 @@
void contrast(const uchar4 *in, uchar4 *out)
{
-#if 0
- out->r = rsClamp((int)(brightM * in->r + brightC), 0, 255);
- out->g = rsClamp((int)(brightM * in->g + brightC), 0, 255);
- out->b = rsClamp((int)(brightM * in->b + brightC), 0, 255);
-#else
float3 v = convert_float3(in->rgb) * brightM + brightC;
out->rgb = convert_uchar3(clamp(v, 0.f, 255.f));
-#endif
}
diff --git a/java/tests/ImageProcessing/src/com/android/rs/image/convolve3x3.fs b/java/tests/ImageProcessing/src/com/android/rs/image/convolve3x3.fs
index 177e86e..4f8b4d8 100644
--- a/java/tests/ImageProcessing/src/com/android/rs/image/convolve3x3.fs
+++ b/java/tests/ImageProcessing/src/com/android/rs/image/convolve3x3.fs
@@ -28,6 +28,8 @@
uint32_t y1 = min((int32_t)y+1, gHeight-1);
uint32_t y2 = max((int32_t)y-1, 0);
+ float4 result;
+
float4 p00 = convert_float4(rsGetElementAt_uchar4(gIn, x1, y1));
float4 p01 = convert_float4(rsGetElementAt_uchar4(gIn, x, y1));
float4 p02 = convert_float4(rsGetElementAt_uchar4(gIn, x2, y1));
@@ -37,29 +39,19 @@
float4 p20 = convert_float4(rsGetElementAt_uchar4(gIn, x1, y2));
float4 p21 = convert_float4(rsGetElementAt_uchar4(gIn, x, y2));
float4 p22 = convert_float4(rsGetElementAt_uchar4(gIn, x2, y2));
- p00 *= gCoeffs[0];
- p01 *= gCoeffs[1];
- p02 *= gCoeffs[2];
- p10 *= gCoeffs[3];
- p11 *= gCoeffs[4];
- p12 *= gCoeffs[5];
- p20 *= gCoeffs[6];
- p21 *= gCoeffs[7];
- p22 *= gCoeffs[8];
- p00 += p01;
- p02 += p10;
- p11 += p12;
- p20 += p21;
+ result = p00 * gCoeffs[0];
+ result += p01 * gCoeffs[1];
+ result += p02 * gCoeffs[2];
+ result += p10 * gCoeffs[3];
+ result += p11 * gCoeffs[4];
+ result += p12 * gCoeffs[5];
+ result += p20 * gCoeffs[6];
+ result += p21 * gCoeffs[7];
+ result += p22 * gCoeffs[8];
- p22 += p00;
- p02 += p11;
-
- p20 += p22;
- p20 += p02;
-
- p20 = clamp(p20, 0.f, 255.f);
- return convert_uchar4(p20);
+ result = clamp(result, 0.f, 255.f);
+ return convert_uchar4(result);
}
diff --git a/java/tests/ImageProcessing/src/com/android/rs/image/exposure.rs b/java/tests/ImageProcessing/src/com/android/rs/image/exposure.rs
index d9a8f62..88bb1d6 100644
--- a/java/tests/ImageProcessing/src/com/android/rs/image/exposure.rs
+++ b/java/tests/ImageProcessing/src/com/android/rs/image/exposure.rs
@@ -23,10 +23,11 @@
bright = 255.f / (255.f - v);
}
-void exposure(const uchar4 *in, uchar4 *out)
+uchar4 __attribute__((kernel)) exposure(uchar4 in)
{
- out->r = rsClamp((int)(bright * in->r), 0, 255);
- out->g = rsClamp((int)(bright * in->g), 0, 255);
- out->b = rsClamp((int)(bright * in->b), 0, 255);
+ uchar4 out = {0, 0, 0, 255};
+ float3 t = convert_float3(in.rgb);
+ out.rgb = convert_uchar3(clamp(convert_int3(t * bright), 0, 255));
+ return out;
}
diff --git a/java/tests/ImageProcessing/src/com/android/rs/image/histogram.rs b/java/tests/ImageProcessing/src/com/android/rs/image/histogram.rs
new file mode 100644
index 0000000..dc0ec59
--- /dev/null
+++ b/java/tests/ImageProcessing/src/com/android/rs/image/histogram.rs
@@ -0,0 +1,84 @@
+/*
+ * Copyright (C) 2013 The Android Open Source Project
+ *
+ * Licensed under the Apache License, Version 2.0 (the "License");
+ * you may not use this file except in compliance with the License.
+ * You may obtain a copy of the License at
+ *
+ * http://www.apache.org/licenses/LICENSE-2.0
+ *
+ * Unless required by applicable law or agreed to in writing, software
+ * distributed under the License is distributed on an "AS IS" BASIS,
+ * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ * See the License for the specific language governing permissions and
+ * limitations under the License.
+ */
+
+#include "ip.rsh"
+
+rs_allocation gSrc;
+rs_allocation gDest;
+rs_allocation gSums;
+rs_allocation gSum;
+
+int gWidth;
+int gHeight;
+int gStep;
+int gSteps;
+
+void __attribute__((kernel)) pass1(int in, uint x, uint y) {
+ for (int i=0; i < (256); i++) {
+ rsSetElementAt_int(gSums, 0, i, y);
+ }
+
+ for (int i = 0; i < gStep; i++) {
+ int py = y*gStep + i;
+ if (py >= gHeight) return;
+
+ for (int px=0; px < gWidth; px++) {
+ uchar4 c = rsGetElementAt_uchar4(gSrc, px, py);
+ int lum = (77 * c.r + 150 * c.g + 29 * c.b) >> 8;
+
+ int old = rsGetElementAt_int(gSums, lum, y);
+ rsSetElementAt_int(gSums, old+1, lum, y);
+ }
+ }
+}
+
+int __attribute__((kernel)) pass2(uint x) {
+ int sum = 0;
+ for (int i=0; i < gSteps; i++) {
+ sum += rsGetElementAt_int(gSums, x, i);
+ }
+ return sum;
+}
+
+void rescale() {
+ int maxv = 0;
+
+ for (int i=0; i < 256; i++) {
+ maxv = max(maxv, rsGetElementAt_int(gSum, i));
+ }
+ float overMax = (1.f / maxv) * gHeight;
+
+ for (int i=0; i < 256; i++) {
+ int t = rsGetElementAt_int(gSum, i);
+ t = gHeight - (overMax * rsGetElementAt_int(gSum, i));
+ t = max(0, t);
+ rsSetElementAt_int(gSum, t, i);
+ }
+}
+
+static const uchar4 gClear = {0, 0, 0, 0xff};
+
+uchar4 __attribute__((kernel)) clear() {
+ return gClear;
+}
+
+uchar4 __attribute__((kernel)) draw(uint x, uint y) {
+ int l = rsGetElementAt_int(gSum, x >> 2);
+ if (y > l) {
+ return 0xff;
+ }
+ return gClear;
+}
diff --git a/java/tests/ImageProcessing/src/com/android/rs/image/vibrance.rs b/java/tests/ImageProcessing/src/com/android/rs/image/vibrance.rs
index 174c2c3..b82e1d3 100644
--- a/java/tests/ImageProcessing/src/com/android/rs/image/vibrance.rs
+++ b/java/tests/ImageProcessing/src/com/android/rs/image/vibrance.rs
@@ -23,11 +23,6 @@
static const float Gf = 0.587f;
static const float Bf = 0.114f;
-static float S = 0.f;
-static float MS = 0.f;
-static float Rt = 0.f;
-static float Gt = 0.f;
-static float Bt = 0.f;
static float Vib = 0.f;
void vibranceKernel(const uchar4 *in, uchar4 *out) {
@@ -37,14 +32,13 @@
int r = in->r;
int g = in->g;
int b = in->b;
- float red = (r-max(g, b))/256.f;
- float sx = (float)(Vib/(1+native_exp(-red*3)));
- S = sx+1;
- MS = 1.0f - S;
- Rt = Rf * MS;
- Gt = Gf * MS;
- Bt = Bf * MS;
- int t = (r + g) / 2;
+ float red = (r-max(g, b)) * (1.f / 256.f);
+ float S = (float)(Vib/(1+native_exp(-red*3)))+1;
+ float MS = 1.0f - S;
+ float Rt = Rf * MS;
+ float Gt = Gf * MS;
+ float Bt = Bf * MS;
+ int t = (r + g) >> 1;
R = r;
G = g;
B = b;
@@ -60,12 +54,5 @@
}
void prepareVibrance() {
-
Vib = vibrance/100.f;
- S = Vib + 1;
- MS = 1.0f - S;
- Rt = Rf * MS;
- Gt = Gf * MS;
- Bt = Bf * MS;
-
}
diff --git a/java/tests/ImageProcessing/src/com/android/rs/image/wbalance.rs b/java/tests/ImageProcessing/src/com/android/rs/image/wbalance.rs
index b5ab14f..461204c 100644
--- a/java/tests/ImageProcessing/src/com/android/rs/image/wbalance.rs
+++ b/java/tests/ImageProcessing/src/com/android/rs/image/wbalance.rs
@@ -17,39 +17,19 @@
#include "ip.rsh"
#pragma rs_fp_relaxed
-static int histR[256] = {0}, histG[256] = {0}, histB[256] = {0};
-
-rs_allocation histogramSource;
-uint32_t histogramHeight;
-uint32_t histogramWidth;
-
-static float scaleR;
-static float scaleG;
-static float scaleB;
+rs_allocation histogramValues;
+static float3 scale;
static uchar4 estimateWhite() {
-
- for (int i = 0; i < 256; i++) {
- histR[i] = 0; histG[i] = 0; histB[i] = 0;
- }
-
- for (uint32_t i = 0; i < histogramHeight; i++) {
- for (uint32_t j = 0; j < histogramWidth; j++) {
- uchar4 in = rsGetElementAt_uchar4(histogramSource, j, i);
- histR[in.r]++;
- histG[in.g]++;
- histB[in.b]++;
- }
- }
-
int min_r = -1, min_g = -1, min_b = -1;
int max_r = 0, max_g = 0, max_b = 0;
int sum_r = 0, sum_g = 0, sum_b = 0;
for (int i = 1; i < 255; i++) {
- int r = histR[i];
- int g = histG[i];
- int b = histB[i];
+ int4 hv = rsGetElementAt_int4(histogramValues, i);
+ int r = hv.r;
+ int g = hv.g;
+ int b = hv.b;
sum_r += r;
sum_g += g;
sum_b += b;
@@ -73,9 +53,10 @@
int tmp_r = 0, tmp_g = 0, tmp_b = 0;
for (int i = 254; i >0; i--) {
- int r = histR[i];
- int g = histG[i];
- int b = histB[i];
+ int4 hv = rsGetElementAt_int4(histogramValues, i);
+ int r = hv.r;
+ int g = hv.g;
+ int b = hv.b;
tmp_r += r;
tmp_g += g;
tmp_b += b;
@@ -115,28 +96,18 @@
int maximum = max(estimation.r, max(estimation.g, estimation.b));
float avg = (minimum + maximum) / 2.f;
- scaleR = avg/estimation.r;
- scaleG = avg/estimation.g;
- scaleB = avg/estimation.b;
-
+ scale.r = avg / estimation.r;
+ scale.g = avg / estimation.g;
+ scale.b = avg / estimation.b;
}
-static unsigned char contrastClamp(int c)
-{
- int N = 255;
- c &= ~(c >> 31);
- c -= N;
- c &= (c >> 31);
- c += N;
- return (unsigned char) c;
-}
+uchar4 __attribute__((kernel)) whiteBalanceKernel(uchar4 in) {
+ float3 t = convert_float3(in.rgb);
+ t *= scale;
+ t = min(t, 255.f);
-void whiteBalanceKernel(const uchar4 *in, uchar4 *out) {
- float Rc = in->r*scaleR;
- float Gc = in->g*scaleG;
- float Bc = in->b*scaleB;
-
- out->r = contrastClamp(Rc);
- out->g = contrastClamp(Gc);
- out->b = contrastClamp(Bc);
+ uchar4 out;
+ out.rgb = convert_uchar3(t);
+ out.a = 255;
+ return out;
}
diff --git a/java/tests/ImageProcessing2/src/com/android/rs/image/colorcube.rs b/java/tests/ImageProcessing2/src/com/android/rs/image/colorcube.rs
index 4f1e73e..c0d6ace 100644
--- a/java/tests/ImageProcessing2/src/com/android/rs/image/colorcube.rs
+++ b/java/tests/ImageProcessing2/src/com/android/rs/image/colorcube.rs
@@ -56,16 +56,16 @@
uint4 v011 = convert_uint4(rsGetElementAt_uchar4(gCube, coord1.x, coord2.y, coord2.z));
uint4 v111 = convert_uint4(rsGetElementAt_uchar4(gCube, coord2.x, coord2.y, coord2.z));
- uint4 yz00 = ((v000 * weight1.x) + (v100 * weight2.x)) >> (int4)8;
- uint4 yz10 = ((v010 * weight1.x) + (v110 * weight2.x)) >> (int4)8;
- uint4 yz01 = ((v001 * weight1.x) + (v101 * weight2.x)) >> (int4)8;
- uint4 yz11 = ((v011 * weight1.x) + (v111 * weight2.x)) >> (int4)8;
+ uint4 yz00 = ((v000 * weight1.x) + (v100 * weight2.x)) >> (uint4)8;
+ uint4 yz10 = ((v010 * weight1.x) + (v110 * weight2.x)) >> (uint4)8;
+ uint4 yz01 = ((v001 * weight1.x) + (v101 * weight2.x)) >> (uint4)8;
+ uint4 yz11 = ((v011 * weight1.x) + (v111 * weight2.x)) >> (uint4)8;
- uint4 z0 = ((yz00 * weight1.y) + (yz10 * weight2.y)) >> (int4)16;
- uint4 z1 = ((yz01 * weight1.y) + (yz11 * weight2.y)) >> (int4)16;
+ uint4 z0 = ((yz00 * weight1.y) + (yz10 * weight2.y)) >> (uint4)16;
+ uint4 z1 = ((yz01 * weight1.y) + (yz11 * weight2.y)) >> (uint4)16;
- uint4 v = ((z0 * weight1.z) + (z1 * weight2.z)) >> (int4)16;
- uint4 v2 = (v + 0x7f) >> (int4)8;
+ uint4 v = ((z0 * weight1.z) + (z1 * weight2.z)) >> (uint4)16;
+ uint4 v2 = (v + 0x7f) >> (uint4)8;
*out = convert_uchar4(v2);
out->a = 0xff;
diff --git a/java/tests/LivePreview/src/com/android/rs/livepreview/CameraPreviewActivity.java b/java/tests/LivePreview/src/com/android/rs/livepreview/CameraPreviewActivity.java
index 62dcaa8..be0af4b 100644
--- a/java/tests/LivePreview/src/com/android/rs/livepreview/CameraPreviewActivity.java
+++ b/java/tests/LivePreview/src/com/android/rs/livepreview/CameraPreviewActivity.java
@@ -228,11 +228,14 @@
// Set initial values
+ //
+ int initialSize = mPreviewSizes.size() - 1;
- mNextPreviewSize = mPreviewSizes.get(15);
- mResolutionSpinner.setSelection(15);
+ mNextPreviewSize = mPreviewSizes.get(initialSize);
+ mResolutionSpinner.setSelection(initialSize);
- if (mPreviewTexture != null) {
+ if(mPreviewTexture != null)
+ {
startPreview();
}
}
@@ -371,4 +374,4 @@
-}
\ No newline at end of file
+}
diff --git a/java/tests/RSTest_CompatLib/src/com/android/rs/test/RSTest.java b/java/tests/RSTest_CompatLib/src/com/android/rs/test/RSTest.java
index b76f21e..cc07e9c 100644
--- a/java/tests/RSTest_CompatLib/src/com/android/rs/test/RSTest.java
+++ b/java/tests/RSTest_CompatLib/src/com/android/rs/test/RSTest.java
@@ -40,7 +40,7 @@
private static final String LOG_TAG = "RSTest_Compat";
private static final boolean DEBUG = false;
- private static final boolean LOG_ENABLED = false;
+ private static final boolean LOG_ENABLED = true;
private RenderScript mRS;
private RSTestCore RSTC;
@@ -54,10 +54,6 @@
RSTC = new RSTestCore(this);
RSTC.init(mRS, getResources());
-
-
-
-
}
static void log(String message) {
diff --git a/java/tests/RSTest_CompatLib/src/com/android/rs/test/RSTestCore.java b/java/tests/RSTest_CompatLib/src/com/android/rs/test/RSTestCore.java
index 51f8a4d..2ee2a33 100644
--- a/java/tests/RSTest_CompatLib/src/com/android/rs/test/RSTestCore.java
+++ b/java/tests/RSTest_CompatLib/src/com/android/rs/test/RSTestCore.java
@@ -58,6 +58,7 @@
unitTests = new ArrayList<UnitTest>();
unitTests.add(new UT_primitives(this, mRes, mCtx));
+ unitTests.add(new UT_instance(this, mRes, mCtx));
unitTests.add(new UT_constant(this, mRes, mCtx));
unitTests.add(new UT_vector(this, mRes, mCtx));
unitTests.add(new UT_unsigned(this, mRes, mCtx));
diff --git a/java/tests/RSTest_CompatLib/src/com/android/rs/test/UT_instance.java b/java/tests/RSTest_CompatLib/src/com/android/rs/test/UT_instance.java
new file mode 100644
index 0000000..071b80e
--- /dev/null
+++ b/java/tests/RSTest_CompatLib/src/com/android/rs/test/UT_instance.java
@@ -0,0 +1,102 @@
+/*
+ * Copyright (C) 2013 The Android Open Source Project
+ *
+ * Licensed under the Apache License, Version 2.0 (the "License");
+ * you may not use this file except in compliance with the License.
+ * You may obtain a copy of the License at
+ *
+ * http://www.apache.org/licenses/LICENSE-2.0
+ *
+ * Unless required by applicable law or agreed to in writing, software
+ * distributed under the License is distributed on an "AS IS" BASIS,
+ * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ * See the License for the specific language governing permissions and
+ * limitations under the License.
+ */
+
+package com.android.rs.test_compat;
+
+import android.content.Context;
+import android.content.res.Resources;
+import android.support.v8.renderscript.*;
+
+public class UT_instance extends UnitTest {
+ private Resources mRes;
+
+ protected UT_instance(RSTestCore rstc, Resources res, Context ctx) {
+ super(rstc, "Instance", ctx);
+ mRes = res;
+ }
+
+ void assertEquals(int e, int v) {
+ if (e != v) {
+ RSTest.log("Assertion failed! Expected: <" + e + "> Got: <" + v + ">");
+ failTest();
+ }
+ }
+
+ public void run() {
+ RenderScript mRS = RenderScript.create(mCtx);
+ mRS.setMessageHandler(mRsMessage);
+
+ ScriptC_instance instance_1 = new ScriptC_instance(mRS);
+ ScriptC_instance instance_2 = new ScriptC_instance(mRS);
+ ScriptC_instance instance_3 = new ScriptC_instance(mRS);
+ ScriptC_instance instance_4 = new ScriptC_instance(mRS);
+ ScriptC_instance instance_5 = new ScriptC_instance(mRS);
+
+ Type t = new Type.Builder(mRS, Element.I32(mRS)).setX(1).create();
+ Allocation ai1 = Allocation.createTyped(mRS, t);
+ Allocation ai2 = Allocation.createTyped(mRS, t);
+ Allocation ai3 = Allocation.createTyped(mRS, t);
+ Allocation ai4 = Allocation.createTyped(mRS, t);
+ Allocation ai5 = Allocation.createTyped(mRS, t);
+
+ instance_1.set_i(1);
+ instance_2.set_i(2);
+ instance_3.set_i(3);
+ instance_4.set_i(4);
+ instance_5.set_i(5);
+ instance_1.set_ai(ai1);
+ instance_2.set_ai(ai2);
+ instance_3.set_ai(ai3);
+ instance_4.set_ai(ai4);
+ instance_5.set_ai(ai5);
+
+ // We now check to ensure that the global is not being shared across
+ // our separate script instances. Our invoke here merely sets the
+ // instanced allocation with the instanced global variable's value.
+ // If globals are being shared (i.e. not instancing scripts), then
+ // both instanced allocations will have the same resulting value
+ // (depending on the order in which the invokes complete).
+ instance_1.invoke_instance_test();
+ instance_2.invoke_instance_test();
+ instance_3.invoke_instance_test();
+ instance_4.invoke_instance_test();
+ instance_5.invoke_instance_test();
+
+ int i1[] = new int[1];
+ int i2[] = new int[1];
+ int i3[] = new int[1];
+ int i4[] = new int[1];
+ int i5[] = new int[1];
+
+ ai1.copyTo(i1);
+ ai2.copyTo(i2);
+ ai3.copyTo(i3);
+ ai4.copyTo(i4);
+ ai5.copyTo(i5);
+
+ assertEquals(1, i1[0]);
+ assertEquals(2, i2[0]);
+ assertEquals(3, i3[0]);
+ assertEquals(4, i4[0]);
+ assertEquals(5, i5[0]);
+ assertEquals(1, i1[0]);
+ assertEquals(2, i2[0]);
+ assertEquals(3, i3[0]);
+ assertEquals(4, i4[0]);
+ assertEquals(5, i5[0]);
+ passTest(); // Set to pass (as long as existing checks didn't fail).
+ }
+}
diff --git a/java/tests/RSTest_CompatLib/src/com/android/rs/test/instance.rs b/java/tests/RSTest_CompatLib/src/com/android/rs/test/instance.rs
new file mode 100644
index 0000000..65ce032
--- /dev/null
+++ b/java/tests/RSTest_CompatLib/src/com/android/rs/test/instance.rs
@@ -0,0 +1,9 @@
+#include "shared.rsh"
+
+int i;
+rs_allocation ai;
+
+void instance_test() {
+ // Set our allocation based on the global input value.
+ rsSetElementAt_int(ai, i, 0);
+}
diff --git a/java/tests/RsTest/src/com/android/rs/test/RSTestCore.java b/java/tests/RsTest/src/com/android/rs/test/RSTestCore.java
index c0eeeea..d5f56b3 100644
--- a/java/tests/RsTest/src/com/android/rs/test/RSTestCore.java
+++ b/java/tests/RsTest/src/com/android/rs/test/RSTestCore.java
@@ -82,6 +82,7 @@
unitTests.add(new UT_rstime(this, mRes, mCtx));
unitTests.add(new UT_rstypes(this, mRes, mCtx));
unitTests.add(new UT_alloc(this, mRes, mCtx));
+ unitTests.add(new UT_check_dims(this, mRes, mCtx));
unitTests.add(new UT_static_globals(this, mRes, mCtx));
unitTests.add(new UT_refcount(this, mRes, mCtx));
unitTests.add(new UT_foreach(this, mRes, mCtx));
diff --git a/java/tests/RsTest/src/com/android/rs/test/UT_check_dims.java b/java/tests/RsTest/src/com/android/rs/test/UT_check_dims.java
new file mode 100644
index 0000000..71752a3
--- /dev/null
+++ b/java/tests/RsTest/src/com/android/rs/test/UT_check_dims.java
@@ -0,0 +1,72 @@
+/*
+ * Copyright (C) 2013 The Android Open Source Project
+ *
+ * Licensed under the Apache License, Version 2.0 (the "License");
+ * you may not use this file except in compliance with the License.
+ * You may obtain a copy of the License at
+ *
+ * http://www.apache.org/licenses/LICENSE-2.0
+ *
+ * Unless required by applicable law or agreed to in writing, software
+ * distributed under the License is distributed on an "AS IS" BASIS,
+ * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ * See the License for the specific language governing permissions and
+ * limitations under the License.
+ */
+
+package com.android.rs.test;
+
+import android.content.Context;
+import android.content.res.Resources;
+import android.renderscript.*;
+
+public class UT_check_dims extends UnitTest {
+ private Resources mRes;
+ byte mFailedArr[];
+ int mData[];
+ Allocation mA;
+ static final int Pattern = 0xA5A5A5A5;
+
+ protected UT_check_dims(RSTestCore rstc, Resources res, Context ctx) {
+ super(rstc, "Check Dims", ctx);
+ mRes = res;
+ }
+
+ private void initializeGlobals(RenderScript RS, ScriptC_check_dims s) {
+ Type.Builder typeBuilder = new Type.Builder(RS, Element.U8(RS));
+ typeBuilder.setX(1);
+ Allocation AFailed = Allocation.createTyped(RS, typeBuilder.create());
+ s.set_aFailed(AFailed);
+
+ mFailedArr = new byte[1];
+ mFailedArr[0] = 0;
+ AFailed.copyFrom(mFailedArr);
+
+ typeBuilder = new Type.Builder(RS, Element.I32(RS));
+ int X = 5;
+ int Y = 7;
+ typeBuilder.setX(X).setY(Y);
+ mA = Allocation.createTyped(RS, typeBuilder.create());
+ s.set_pattern(Pattern);
+
+ mData = new int[X*Y];
+ for (int i = 0; i < X*Y; i++) {
+ mData[i] = Pattern;
+ }
+ mA.copyFrom(mData);
+
+ return;
+ }
+
+ public void run() {
+ RenderScript pRS = RenderScript.create(mCtx);
+ ScriptC_check_dims s = new ScriptC_check_dims(pRS);
+ pRS.setMessageHandler(mRsMessage);
+ initializeGlobals(pRS, s);
+ s.forEach_root(mA);
+ s.invoke_check_dims_test();
+ pRS.finish();
+ waitForMessage();
+ pRS.destroy();
+ }
+}
diff --git a/java/tests/RsTest/src/com/android/rs/test/check_dims.rs b/java/tests/RsTest/src/com/android/rs/test/check_dims.rs
new file mode 100644
index 0000000..b18796c
--- /dev/null
+++ b/java/tests/RsTest/src/com/android/rs/test/check_dims.rs
@@ -0,0 +1,26 @@
+#include "shared.rsh"
+
+int pattern;
+
+rs_allocation aFailed;
+
+// This test checks to see that we only work on the cells specified for the
+// input allocation (i.e. don't affect anything between dimX and stride for
+// each row). If we don't see the pattern that we wrote, we know that we
+// are definitely working outside our proper bounds.
+void root(const int *o, uint32_t x, uint32_t y) {
+ if (*o != pattern) {
+ rsSetElementAt_uchar(aFailed, 1, 0);
+ }
+}
+
+void check_dims_test() {
+ bool failed = rsGetElementAt_uchar(aFailed, 0);
+ if (failed) {
+ rsSendToClientBlocking(RS_MSG_TEST_FAILED);
+ }
+ else {
+ rsSendToClientBlocking(RS_MSG_TEST_PASSED);
+ }
+}
+
diff --git a/rs.h b/rs.h
index 7f592f7..8a0761a 100644
--- a/rs.h
+++ b/rs.h
@@ -22,9 +22,8 @@
#include "rsDefines.h"
-//
-// A3D loading and object update code.
-// Should only be called at object creation, not thread safe
+// Legacy graphics functions
+// Not extern C because not used from C++ API
RsObjectBase rsaFileA3DGetEntryByIndex(RsContext, uint32_t idx, RsFile);
RsFile rsaFileA3DCreateFromMemory(RsContext, const void *data, uint32_t len);
RsFile rsaFileA3DCreateFromAsset(RsContext, void *asset);
@@ -32,30 +31,32 @@
void rsaFileA3DGetNumIndexEntries(RsContext, int32_t *numEntries, RsFile);
void rsaFileA3DGetIndexEntries(RsContext, RsFileIndexEntry *fileEntries,
uint32_t numEntries, RsFile);
-void rsaGetName(RsContext, void * obj, const char **name);
// Mesh update functions
void rsaMeshGetVertexBufferCount(RsContext, RsMesh, int32_t *vtxCount);
void rsaMeshGetIndexCount(RsContext, RsMesh, int32_t *idxCount);
void rsaMeshGetVertices(RsContext, RsMesh, RsAllocation *vtxData, uint32_t vtxDataCount);
void rsaMeshGetIndices(RsContext, RsMesh, RsAllocation *va,
uint32_t *primType, uint32_t idxDataCount);
-// Allocation update
-const void* rsaAllocationGetType(RsContext con, RsAllocation va);
-// Type update
-void rsaTypeGetNativeData(RsContext, RsType, uint32_t *typeData, uint32_t typeDataSize);
-// Element update
-void rsaElementGetNativeData(RsContext, RsElement, uint32_t *elemData, uint32_t elemDataSize);
-void rsaElementGetSubElements(RsContext, RsElement, uint32_t *ids, const char **names,
- uint32_t *arraySizes, uint32_t dataSize);
-
-RsDevice rsDeviceCreate();
-void rsDeviceDestroy(RsDevice dev);
-void rsDeviceSetConfig(RsDevice dev, RsDeviceParam p, int32_t value);
-RsContext rsContextCreate(RsDevice dev, uint32_t version, uint32_t sdkVersion,
- RsContextType ct, bool forceCpu, bool synchronous);
RsContext rsContextCreateGL(RsDevice dev, uint32_t version, uint32_t sdkVersion,
RsSurfaceConfig sc, uint32_t dpi);
+extern "C" {
+ void rsaGetName(RsContext, void * obj, const char **name);
+ // Allocation update
+ const void* rsaAllocationGetType(RsContext con, RsAllocation va);
+ // Type update
+ void rsaTypeGetNativeData(RsContext, RsType, uint32_t *typeData, uint32_t typeDataSize);
+ // Element update
+ void rsaElementGetNativeData(RsContext, RsElement, uint32_t *elemData, uint32_t elemDataSize);
+ void rsaElementGetSubElements(RsContext, RsElement, uint32_t *ids, const char **names,
+ uint32_t *arraySizes, uint32_t dataSize);
+
+ RsDevice rsDeviceCreate();
+ void rsDeviceDestroy(RsDevice dev);
+ void rsDeviceSetConfig(RsDevice dev, RsDeviceParam p, int32_t value);
+ RsContext rsContextCreate(RsDevice dev, uint32_t version, uint32_t sdkVersion,
+ RsContextType ct, bool forceCpu, bool synchronous);
+}
#include "rsgApiFuncDecl.h"
#endif // RENDER_SCRIPT_H
diff --git a/rs.spec b/rs.spec
index 385622f..aab3d46 100644
--- a/rs.spec
+++ b/rs.spec
@@ -95,7 +95,7 @@
}
ContextDestroyWorker {
- sync
+ sync
}
AssignName {
@@ -171,16 +171,6 @@
param size_t stride
}
-Allocation2DElementData {
- param RsAllocation va
- param uint32_t x
- param uint32_t y
- param uint32_t lod
- param RsAllocationCubemapFace face
- param const void *data
- param size_t element_offset
- }
-
AllocationGenerateMipmaps {
param RsAllocation va
}
@@ -210,12 +200,12 @@
param size_t stride
}
+
AllocationSyncAll {
param RsAllocation va
param RsAllocationUsageType src
}
-
AllocationResize1D {
param RsAllocation va
param uint32_t dimX
@@ -401,5 +391,11 @@
param RsScriptGroup group
}
+AllocationIoSend {
+ param RsAllocation alloc
+ }
+AllocationIoReceive {
+ param RsAllocation alloc
+ }
diff --git a/rsAllocation.cpp b/rsAllocation.cpp
index 0c271c1..9bf8709 100644
--- a/rsAllocation.cpp
+++ b/rsAllocation.cpp
@@ -79,6 +79,13 @@
}
Allocation::~Allocation() {
+#if !defined(RS_SERVER) && !defined(RS_COMPATIBILITY_LIB)
+ if (mGrallocConsumer.get()) {
+ mGrallocConsumer->unlockBuffer();
+ mGrallocConsumer = NULL;
+ }
+#endif
+
freeChildrenUnlocked();
mRSC->mHal.funcs.allocation.destroy(mRSC, this);
}
@@ -236,13 +243,14 @@
void Allocation::dumpLOGV(const char *prefix) const {
ObjectBase::dumpLOGV(prefix);
+ char buf[1024];
- String8 s(prefix);
- s.append(" type ");
- if (mHal.state.type) {
- mHal.state.type->dumpLOGV(s.string());
+ if ((strlen(prefix) + 10) < sizeof(buf)) {
+ sprintf(buf, "%s type ", prefix);
+ if (mHal.state.type) {
+ mHal.state.type->dumpLOGV(buf);
+ }
}
-
ALOGV("%s allocation ptr=%p mUsageFlags=0x04%x, mMipmapControl=0x%04x",
prefix, mHal.drvState.lod[0].mallocPtr, mHal.state.usageFlags, mHal.state.mipmapControl);
}
@@ -327,9 +335,7 @@
void Allocation::serialize(Context *rsc, OStream *stream) const {
// Need to identify ourselves
stream->addU32((uint32_t)getClassId());
-
- String8 name(getName());
- stream->addString(&name);
+ stream->addString(getName());
// First thing we need to serialize is the type object since it will be needed
// to initialize the class
@@ -358,8 +364,7 @@
return NULL;
}
- String8 name;
- stream->loadString(&name);
+ const char *name = stream->loadString();
Type *type = Type::createFromStream(rsc, stream);
if (!type) {
@@ -382,8 +387,7 @@
return NULL;
}
- alloc->setName(name.string(), name.size());
-
+ alloc->assignName(name);
if (dataSize == type->getSizeBytes()) {
uint32_t count = dataSize / type->getElementSizeBytes();
// Read in all of our allocation data
@@ -454,8 +458,31 @@
ALOGE("not implemented");
}
+#ifndef RS_COMPATIBILITY_LIB
+void Allocation::NewBufferListener::onFrameAvailable() {
+ intptr_t ip = (intptr_t)alloc;
+ rsc->sendMessageToClient(NULL, RS_MESSAGE_TO_CLIENT_NEW_BUFFER, ip, 0, true);
+}
+#endif
+
void * Allocation::getSurface(const Context *rsc) {
- return rsc->mHal.funcs.allocation.getSurface(rsc, this);
+#ifndef RS_COMPATIBILITY_LIB
+ // Configure GrallocConsumer to be in asynchronous mode
+ sp<BufferQueue> bq = new BufferQueue();
+ mGrallocConsumer = new GrallocConsumer(this, bq);
+ sp<IGraphicBufferProducer> bp = bq;
+ bp->incStrong(NULL);
+
+ mBufferListener = new NewBufferListener();
+ mBufferListener->rsc = rsc;
+ mBufferListener->alloc = this;
+
+ mGrallocConsumer->setFrameAvailableListener(mBufferListener);
+ return bp.get();
+#else
+ return NULL;
+#endif
+ //return rsc->mHal.funcs.allocation.getSurface(rsc, this);
}
void Allocation::setSurface(const Context *rsc, RsNativeWindow sur) {
@@ -468,7 +495,22 @@
}
void Allocation::ioReceive(const Context *rsc) {
- rsc->mHal.funcs.allocation.ioReceive(rsc, this);
+ void *ptr = NULL;
+ size_t stride = 0;
+#ifndef RS_COMPATIBILITY_LIB
+ if (mHal.state.usageFlags & RS_ALLOCATION_USAGE_SCRIPT) {
+ status_t ret = mGrallocConsumer->lockNextBuffer();
+
+ if (ret == OK) {
+ rsc->mHal.funcs.allocation.ioReceive(rsc, this);
+ } else if (ret == BAD_VALUE) {
+ // No new frame, don't do anything
+ } else {
+ rsc->setError(RS_ERROR_DRIVER, "Error receiving IO input buffer.");
+ }
+
+ }
+#endif
}
@@ -690,7 +732,7 @@
}
}
-const void * rsaAllocationGetType(RsContext con, RsAllocation va) {
+extern "C" const void * rsaAllocationGetType(RsContext con, RsAllocation va) {
Allocation *a = static_cast<Allocation *>(va);
a->getType()->incUserRef();
diff --git a/rsAllocation.h b/rsAllocation.h
index dffa440..440246e 100644
--- a/rsAllocation.h
+++ b/rsAllocation.h
@@ -19,6 +19,14 @@
#include "rsType.h"
+#include <ui/GraphicBuffer.h>
+
+#if !defined(RS_SERVER) && !defined(RS_COMPATIBILITY_LIB)
+#include "rsGrallocConsumer.h"
+#include "gui/CpuConsumer.h"
+#include "gui/GLConsumer.h"
+#endif
+
// ---------------------------------------------------------------------------
namespace android {
@@ -58,8 +66,8 @@
bool hasReferences;
void * userProvidedPtr;
int32_t surfaceTextureID;
- void *deprecated01;
- void *deprecated02;
+ ANativeWindowBuffer *nativeBuffer;
+ int64_t timestamp;
};
State state;
@@ -157,6 +165,20 @@
mHal.state.type = t;
}
+#if !defined(RS_SERVER) && !defined(RS_COMPATIBILITY_LIB)
+ class NewBufferListener : public android::ConsumerBase::FrameAvailableListener {
+ public:
+ const android::renderscript::Context *rsc;
+ const android::renderscript::Allocation *alloc;
+
+ virtual void onFrameAvailable();
+ };
+
+ sp<NewBufferListener> mBufferListener;
+ sp< GrallocConsumer > mGrallocConsumer;
+#endif
+
+
private:
void freeChildrenUnlocked();
Allocation(Context *rsc, const Type *, uint32_t usages, RsAllocationMipmapControl mc, void *ptr);
diff --git a/rsContext.cpp b/rsContext.cpp
index 74c40c5..44f2d74 100644
--- a/rsContext.cpp
+++ b/rsContext.cpp
@@ -34,7 +34,7 @@
#include <dlfcn.h>
#include <unistd.h>
-#if !defined(RS_SERVER)
+#if !defined(RS_SERVER) && defined(HAVE_ANDROID_OS)
#include <cutils/properties.h>
#endif
@@ -205,7 +205,7 @@
#endif
static uint32_t getProp(const char *str) {
-#ifndef RS_SERVER
+#if !defined(RS_SERVER) && defined(HAVE_ANDROID_OS)
char buf[PROPERTY_VALUE_MAX];
property_get(str, buf, "0");
return atoi(buf);
@@ -868,7 +868,6 @@
//ALOGV("%p rsContextDestroy done", rsc);
}
-
RsMessageToClientType rsi_ContextPeekMessage(Context *rsc,
size_t * receiveLen, size_t receiveLen_length,
uint32_t * subID, size_t subID_length) {
@@ -898,8 +897,8 @@
}
}
-RsContext rsContextCreate(RsDevice vdev, uint32_t version, uint32_t sdkVersion,
- RsContextType ct, bool forceCpu, bool synchronous) {
+extern "C" RsContext rsContextCreate(RsDevice vdev, uint32_t version, uint32_t sdkVersion,
+ RsContextType ct, bool forceCpu, bool synchronous) {
//ALOGV("rsContextCreate dev=%p", vdev);
Device * dev = static_cast<Device *>(vdev);
Context *rsc = Context::createContext(dev, NULL, ct, forceCpu, synchronous);
diff --git a/rsContext.h b/rsContext.h
index abc9b27..a29313e 100644
--- a/rsContext.h
+++ b/rsContext.h
@@ -26,6 +26,15 @@
#include "rsScriptGroup.h"
#include "rsSampler.h"
+#if !defined(RS_SERVER) && !defined(RS_COMPATIBILITY_LIB)
+#define ATRACE_TAG ATRACE_TAG_RS
+#include "utils/Trace.h"
+#else
+#define ATRACE_ENABLED(...) false
+#define ATRACE_NAME(...)
+#define ATRACE_CALL(...)
+#endif
+
#ifndef RS_COMPATIBILITY_LIB
#include "rsFont.h"
#include "rsPath.h"
@@ -34,6 +43,7 @@
#include "rsProgramRaster.h"
#include "rsProgramVertex.h"
#include "rsFBOCache.h"
+
#endif
@@ -62,6 +72,8 @@
#define CHECK_OBJ_OR_NULL(o)
#endif
+
+
class Context {
public:
struct Hal {
diff --git a/rsCppUtils.cpp b/rsCppUtils.cpp
new file mode 100644
index 0000000..59ec9bb
--- /dev/null
+++ b/rsCppUtils.cpp
@@ -0,0 +1,36 @@
+/*
+ * Copyright (C) 2013 The Android Open Source Project
+ *
+ * Licensed under the Apache License, Version 2.0 (the "License");
+ * you may not use this file except in compliance with the License.
+ * You may obtain a copy of the License at
+ *
+ * http://www.apache.org/licenses/LICENSE-2.0
+ *
+ * Unless required by applicable law or agreed to in writing, software
+ * distributed under the License is distributed on an "AS IS" BASIS,
+ * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ * See the License for the specific language governing permissions and
+ * limitations under the License.
+ */
+
+#include "rsUtils.h"
+#include "rsCppUtils.h"
+
+namespace android {
+namespace renderscript {
+
+const char * rsuCopyString(const char *name) {
+ return rsuCopyString(name, strlen(name));
+}
+
+const char * rsuCopyString(const char *name, size_t len) {
+ char *n = new char[len+1];
+ memcpy(n, name, len);
+ n[len] = 0;
+ return n;
+}
+
+
+}
+}
diff --git a/rsCppUtils.h b/rsCppUtils.h
index ced13c1..36b52aa 100644
--- a/rsCppUtils.h
+++ b/rsCppUtils.h
@@ -38,6 +38,9 @@
#if defined(RS_SERVER) || defined(RS_COMPATIBILITY_LIB)
+#define ATRACE_TAG
+#define ATRACE_CALL(...)
+
#include <string>
#include <vector>
#include <algorithm>
@@ -185,6 +188,9 @@
namespace android {
namespace renderscript {
+const char * rsuCopyString(const char *name);
+const char * rsuCopyString(const char *name, size_t len);
+
#if 1
#define rsAssert(v) do {if(!(v)) ALOGE("rsAssert failed: %s, in %s at %i", #v, __FILE__, __LINE__);} while (0)
#else
diff --git a/rsDefines.h b/rsDefines.h
index 6e080de..0287f67 100644
--- a/rsDefines.h
+++ b/rsDefines.h
@@ -363,7 +363,8 @@
RS_SCRIPT_INTRINSIC_ID_BLUR = 5,
RS_SCRIPT_INTRINSIC_ID_YUV_TO_RGB = 6,
RS_SCRIPT_INTRINSIC_ID_BLEND = 7,
- RS_SCRIPT_INTRINSIC_ID_3DLUT = 8
+ RS_SCRIPT_INTRINSIC_ID_3DLUT = 8,
+ RS_SCRIPT_INTRINSIC_ID_HISTOGRAM = 9
};
typedef struct {
diff --git a/rsDevice.cpp b/rsDevice.cpp
index d7d03f6..2688890 100644
--- a/rsDevice.cpp
+++ b/rsDevice.cpp
@@ -40,17 +40,17 @@
}
}
-RsDevice rsDeviceCreate() {
+extern "C" RsDevice rsDeviceCreate() {
Device * d = new Device();
return d;
}
-void rsDeviceDestroy(RsDevice dev) {
+extern "C" void rsDeviceDestroy(RsDevice dev) {
Device * d = static_cast<Device *>(dev);
delete d;
}
-void rsDeviceSetConfig(RsDevice dev, RsDeviceParam p, int32_t value) {
+extern "C" void rsDeviceSetConfig(RsDevice dev, RsDeviceParam p, int32_t value) {
Device * d = static_cast<Device *>(dev);
if (p == RS_DEVICE_PARAM_FORCE_SOFTWARE_GL) {
d->mForceSW = value != 0;
diff --git a/rsElement.cpp b/rsElement.cpp
index 712f17d..a81df48 100644
--- a/rsElement.cpp
+++ b/rsElement.cpp
@@ -44,7 +44,12 @@
}
void Element::clear() {
- delete [] mFields;
+ if (mFields) {
+ for (size_t i = 0; i < mFieldCount; i++) {
+ delete[] mFields[i].name;
+ }
+ delete [] mFields;
+ }
mFields = NULL;
mFieldCount = 0;
mHasReference = false;
@@ -87,7 +92,7 @@
for (uint32_t ct = 0; ct < mFieldCount; ct++) {
ALOGV("%s Element field index: %u ------------------", prefix, ct);
ALOGV("%s name: %s, offsetBits: %u, arraySize: %u",
- prefix, mFields[ct].name.string(), mFields[ct].offsetBits, mFields[ct].arraySize);
+ prefix, mFields[ct].name, mFields[ct].offsetBits, mFields[ct].arraySize);
mFields[ct].e->dumpLOGV(prefix);
}
}
@@ -95,16 +100,14 @@
void Element::serialize(Context *rsc, OStream *stream) const {
// Need to identify ourselves
stream->addU32((uint32_t)getClassId());
-
- String8 name(getName());
- stream->addString(&name);
+ stream->addString(getName());
mComponent.serialize(stream);
// Now serialize all the fields
stream->addU32(mFieldCount);
for (uint32_t ct = 0; ct < mFieldCount; ct++) {
- stream->addString(&mFields[ct].name);
+ stream->addString(mFields[ct].name);
stream->addU32(mFields[ct].arraySize);
mFields[ct].e->serialize(rsc, stream);
}
@@ -118,8 +121,7 @@
return NULL;
}
- String8 name;
- stream->loadString(&name);
+ const char *name = stream->loadString();
Component component;
component.loadFromStream(stream);
@@ -138,13 +140,9 @@
size_t *subElemNamesLengths = new size_t[fieldCount];
uint32_t *arraySizes = new uint32_t[fieldCount];
- String8 elemName;
for (uint32_t ct = 0; ct < fieldCount; ct ++) {
- stream->loadString(&elemName);
- subElemNamesLengths[ct] = elemName.length();
- char *tmpName = new char[subElemNamesLengths[ct]];
- memcpy(tmpName, elemName.string(), subElemNamesLengths[ct]);
- subElemNames[ct] = tmpName;
+ subElemNames[ct] = stream->loadString();
+ subElemNamesLengths[ct] = strlen(subElemNames[ct]);
arraySizes[ct] = stream->loadU32();
subElems[ct] = Element::createFromStream(rsc, stream);
}
@@ -155,6 +153,7 @@
delete [] subElemNames[ct];
subElems[ct]->decUserRef();
}
+ delete[] name;
delete[] subElems;
delete[] subElemNames;
delete[] subElemNamesLengths;
@@ -179,7 +178,7 @@
uint32_t noPaddingFieldCount = 0;
for (uint32_t ct = 0; ct < mFieldCount; ct ++) {
- if (mFields[ct].name.string()[0] != '#') {
+ if (mFields[ct].name[0] != '#') {
noPaddingFieldCount ++;
}
}
@@ -203,14 +202,14 @@
mHasReference = true;
}
- if (mFields[ct].name.string()[0] == '#') {
+ if (mFields[ct].name[0] == '#') {
continue;
}
mHal.state.fields[ctNoPadding] = mFields[ct].e.get();
mHal.state.fieldArraySizes[ctNoPadding] = mFields[ct].arraySize;
- mHal.state.fieldNames[ctNoPadding] = mFields[ct].name.string();
- mHal.state.fieldNameLengths[ctNoPadding] = mFields[ct].name.length() + 1; // to include 0
+ mHal.state.fieldNames[ctNoPadding] = mFields[ct].name;
+ mHal.state.fieldNameLengths[ctNoPadding] = strlen(mFields[ct].name) + 1; // to include 0
mHal.state.fieldOffsetBytes[ctNoPadding] = mFields[ct].offsetBits >> 3;
ctNoPadding ++;
@@ -262,10 +261,21 @@
if (ee->getFieldCount() == count) {
bool match = true;
for (uint32_t i=0; i < count; i++) {
+ size_t len;
+ uint32_t asize = 1;
+ if (lengths) {
+ len = lengths[i];
+ } else {
+ len = strlen(nin[i]);
+ }
+ if (asin) {
+ asize = asin[i];
+ }
+
if ((ee->mFields[i].e.get() != ein[i]) ||
- (ee->mFields[i].name.length() != lengths[i]) ||
- (ee->mFields[i].name != nin[i]) ||
- (ee->mFields[i].arraySize != asin[i])) {
+ (strlen(ee->mFields[i].name) != len) ||
+ strcmp(ee->mFields[i].name, nin[i]) ||
+ (ee->mFields[i].arraySize != asize)) {
match = false;
break;
}
@@ -284,9 +294,20 @@
e->mFields = new ElementField_t [count];
e->mFieldCount = count;
for (size_t ct=0; ct < count; ct++) {
+ size_t len;
+ uint32_t asize = 1;
+ if (lengths) {
+ len = lengths[ct];
+ } else {
+ len = strlen(nin[ct]);
+ }
+ if (asin) {
+ asize = asin[ct];
+ }
+
e->mFields[ct].e.set(ein[ct]);
- e->mFields[ct].name.setTo(nin[ct], lengths[ct]);
- e->mFields[ct].arraySize = asin[ct];
+ e->mFields[ct].name = rsuCopyString(nin[ct], len);
+ e->mFields[ct].arraySize = asize;
}
e->compute();
@@ -341,33 +362,6 @@
}
}
-Element::Builder::Builder() {
- const uint32_t initialCapacity = 32;
- mBuilderElementRefs.setCapacity(initialCapacity);
- mBuilderElements.setCapacity(initialCapacity);
- mBuilderNameStrings.setCapacity(initialCapacity);
- mBuilderNameLengths.setCapacity(initialCapacity);
- mBuilderArrays.setCapacity(initialCapacity);
-}
-
-void Element::Builder::add(const Element *e, const char *nameStr, uint32_t arraySize) {
- mBuilderElementRefs.push(ObjectBaseRef<const Element>(e));
- mBuilderElements.push(e);
- mBuilderNameStrings.push(nameStr);
- mBuilderNameLengths.push(strlen(nameStr));
- mBuilderArrays.push(arraySize);
-
-}
-
-ObjectBaseRef<const Element> Element::Builder::create(Context *rsc) {
- return Element::createRef(rsc, mBuilderElements.size(),
- &(mBuilderElements.editArray()[0]),
- &(mBuilderNameStrings.editArray()[0]),
- mBuilderNameLengths.editArray(),
- mBuilderArrays.editArray());
-}
-
-
ElementState::ElementState() {
}
@@ -407,7 +401,7 @@
}
}
-void rsaElementGetNativeData(RsContext con, RsElement elem,
+extern "C" void rsaElementGetNativeData(RsContext con, RsElement elem,
uint32_t *elemData, uint32_t elemDataSize) {
rsAssert(elemDataSize == 5);
// we will pack mType; mKind; mNormalized; mVectorSize; NumSubElements
@@ -420,7 +414,7 @@
(*elemData++) = e->getFieldCount();
}
-void rsaElementGetSubElements(RsContext con, RsElement elem, uintptr_t *ids,
+extern "C" void rsaElementGetSubElements(RsContext con, RsElement elem, uintptr_t *ids,
const char **names, size_t *arraySizes, uint32_t dataSize) {
Element *e = static_cast<Element *>(elem);
rsAssert(e->getFieldCount() == dataSize);
diff --git a/rsElement.h b/rsElement.h
index 1eae46d..b97dfe5 100644
--- a/rsElement.h
+++ b/rsElement.h
@@ -57,18 +57,6 @@
};
Hal mHal;
- class Builder {
- public:
- Builder();
- void add(const Element *e, const char *nameStr, uint32_t arraySize);
- ObjectBaseRef<const Element> create(Context *rsc);
- private:
- Vector<ObjectBaseRef<const Element> > mBuilderElementRefs;
- Vector<const Element *> mBuilderElements;
- Vector<const char*> mBuilderNameStrings;
- Vector<size_t> mBuilderNameLengths;
- Vector<uint32_t> mBuilderArrays;
- };
uint32_t getGLType() const;
uint32_t getGLFormat() const;
@@ -95,7 +83,7 @@
uint32_t getFieldCount() const {return mFieldCount;}
const Element * getField(uint32_t idx) const {return mFields[idx].e.get();}
- const char * getFieldName(uint32_t idx) const {return mFields[idx].name.string();}
+ const char * getFieldName(uint32_t idx) const {return mFields[idx].name;}
uint32_t getFieldArraySize(uint32_t idx) const {return mFields[idx].arraySize;}
const Component & getComponent() const {return mComponent;}
@@ -133,8 +121,8 @@
static const Element* create(Context *rsc, size_t count,
const Element **ein,
const char **nin,
- const size_t * lengths,
- const uint32_t *asin) {
+ const size_t * lengths = NULL,
+ const uint32_t *asin = NULL) {
ObjectBaseRef<const Element> elem = createRef(rsc, count, ein, nin, lengths, asin);
elem->incUserRef();
return elem.get();
@@ -149,7 +137,7 @@
void clear();
typedef struct {
- String8 name;
+ const char *name;
ObjectBaseRef<const Element> e;
uint32_t offsetBits;
uint32_t offsetBitsUnpadded;
diff --git a/rsFileA3D.cpp b/rsFileA3D.cpp
index c79d008..3b963fe 100644
--- a/rsFileA3D.cpp
+++ b/rsFileA3D.cpp
@@ -73,7 +73,8 @@
uint32_t numIndexEntries = headerStream->loadU32();
for (uint32_t i = 0; i < numIndexEntries; i ++) {
A3DIndexEntry *entry = new A3DIndexEntry();
- headerStream->loadString(&entry->mObjectName);
+ entry->mObjectName = headerStream->loadString();
+
//ALOGV("Header data, entry name = %s", entry->mObjectName.string());
entry->mType = (RsA3DClassID)headerStream->loadU32();
if (mUse64BitOffsets){
@@ -214,6 +215,10 @@
return mIndex.size();
}
+FileA3D::A3DIndexEntry::~A3DIndexEntry() {
+ delete[] mObjectName;
+}
+
const FileA3D::A3DIndexEntry *FileA3D::getIndexEntry(size_t index) const {
if (index < mIndex.size()) {
return mIndex[index];
@@ -320,7 +325,7 @@
uint32_t writeIndexSize = mWriteIndex.size();
headerStream.addU32(writeIndexSize);
for (uint32_t i = 0; i < writeIndexSize; i ++) {
- headerStream.addString(&mWriteIndex[i]->mObjectName);
+ headerStream.addString(mWriteIndex[i]->mObjectName);
headerStream.addU32((uint32_t)mWriteIndex[i]->mType);
if (mUse64BitOffsets){
headerStream.addOffset(mWriteIndex[i]->mOffset);
@@ -334,8 +339,7 @@
}
// Write our magic string so we know we are reading the right file
- String8 magicString(A3D_MAGIC_KEY);
- fwrite(magicString.string(), sizeof(char), magicString.size(), writeHandle);
+ fwrite(A3D_MAGIC_KEY, sizeof(char), strlen(A3D_MAGIC_KEY), writeHandle);
// Store the size of the header to make it easier to parse when we read it
uint64_t headerSize = headerStream.getPos();
@@ -369,7 +373,7 @@
mWriteStream = new OStream(initialStreamSize, false);
}
A3DIndexEntry *indexEntry = new A3DIndexEntry();
- indexEntry->mObjectName.setTo(obj->getName());
+ indexEntry->mObjectName = rsuCopyString(obj->getName());
indexEntry->mType = obj->getClassId();
indexEntry->mOffset = mWriteStream->getPos();
indexEntry->mRsObj = obj;
@@ -420,7 +424,7 @@
for (uint32_t i = 0; i < numFileEntries; i ++) {
const FileA3D::A3DIndexEntry *entry = fa3d->getIndexEntry(i);
fileEntries[i].classID = entry->getType();
- fileEntries[i].objectName = entry->getObjectName().string();
+ fileEntries[i].objectName = rsuCopyString(entry->getObjectName());
}
}
diff --git a/rsFileA3D.h b/rsFileA3D.h
index 06b90d7..8bf36b9 100644
--- a/rsFileA3D.h
+++ b/rsFileA3D.h
@@ -19,7 +19,6 @@
#include "rsMesh.h"
-#include <utils/String8.h>
#include "rsStream.h"
#include <stdio.h>
@@ -43,19 +42,21 @@
bool mUse64BitOffsets;
class A3DIndexEntry {
- String8 mObjectName;
+ const char *mObjectName;
RsA3DClassID mType;
uint64_t mOffset;
uint64_t mLength;
ObjectBase *mRsObj;
public:
friend class FileA3D;
- const String8 &getObjectName() const {
+ const char *getObjectName() const {
return mObjectName;
}
RsA3DClassID getType() const {
return mType;
}
+
+ ~A3DIndexEntry();
};
bool load(FILE *f);
diff --git a/rsFont.cpp b/rsFont.cpp
index 3665a3d..d302182 100644
--- a/rsFont.cpp
+++ b/rsFont.cpp
@@ -20,7 +20,9 @@
#include "rsFont.h"
#include "rsProgramFragment.h"
#include "rsMesh.h"
+#ifdef HAVE_ANDROID_OS
#include <cutils/properties.h>
+#endif
#ifndef ANDROID_RS_SERIALIZE
#include <ft2build.h>
@@ -56,7 +58,7 @@
return false;
}
- mFontName = name;
+ mFontName = rsuCopyString(name);
mFontSize = fontSize;
mDpi = dpi;
@@ -337,27 +339,31 @@
mLibrary = NULL;
#endif //ANDROID_RS_SERIALIZE
+ float gamma = DEFAULT_TEXT_GAMMA;
+ int32_t blackThreshold = DEFAULT_TEXT_BLACK_GAMMA_THRESHOLD;
+ int32_t whiteThreshold = DEFAULT_TEXT_WHITE_GAMMA_THRESHOLD;
+
+#ifdef HAVE_ANDROID_OS
// Get the renderer properties
char property[PROPERTY_VALUE_MAX];
// Get the gamma
- float gamma = DEFAULT_TEXT_GAMMA;
if (property_get(PROPERTY_TEXT_GAMMA, property, NULL) > 0) {
gamma = atof(property);
}
// Get the black gamma threshold
- int32_t blackThreshold = DEFAULT_TEXT_BLACK_GAMMA_THRESHOLD;
if (property_get(PROPERTY_TEXT_BLACK_GAMMA_THRESHOLD, property, NULL) > 0) {
blackThreshold = atoi(property);
}
- mBlackThreshold = (float)(blackThreshold) / 255.0f;
// Get the white gamma threshold
- int32_t whiteThreshold = DEFAULT_TEXT_WHITE_GAMMA_THRESHOLD;
if (property_get(PROPERTY_TEXT_WHITE_GAMMA_THRESHOLD, property, NULL) > 0) {
whiteThreshold = atoi(property);
}
+#endif
+
+ mBlackThreshold = (float)(blackThreshold) / 255.0f;
mWhiteThreshold = (float)(whiteThreshold) / 255.0f;
// Compute the gamma tables
@@ -486,13 +492,13 @@
#endif //ANDROID_RS_SERIALIZE
void FontState::initRenderState() {
- String8 shaderString("varying vec2 varTex0;\n");
- shaderString.append("void main() {\n");
- shaderString.append(" lowp vec4 col = UNI_Color;\n");
- shaderString.append(" col.a = texture2D(UNI_Tex0, varTex0.xy).a;\n");
- shaderString.append(" col.a = pow(col.a, UNI_Gamma);\n");
- shaderString.append(" gl_FragColor = col;\n");
- shaderString.append("}\n");
+ const char *shaderString = "varying vec2 varTex0;\n"
+ "void main() {\n"
+ " lowp vec4 col = UNI_Color;\n"
+ " col.a = texture2D(UNI_Tex0, varTex0.xy).a;\n"
+ " col.a = pow(col.a, UNI_Gamma);\n"
+ " gl_FragColor = col;\n"
+ "}\n";
const char *textureNames[] = { "Tex0" };
const size_t textureNamesLengths[] = { 4 };
@@ -502,10 +508,10 @@
RS_KIND_USER, false, 4);
ObjectBaseRef<const Element> gammaElem = Element::createRef(mRSC, RS_TYPE_FLOAT_32,
RS_KIND_USER, false, 1);
- Element::Builder builder;
- builder.add(colorElem.get(), "Color", 1);
- builder.add(gammaElem.get(), "Gamma", 1);
- ObjectBaseRef<const Element> constInput = builder.create(mRSC);
+
+ const char *ebn1[] = { "Color", "Gamma" };
+ const Element *ebe1[] = {colorElem.get(), gammaElem.get()};
+ ObjectBaseRef<const Element> constInput = Element::create(mRSC, 2, ebe1, ebn1);
ObjectBaseRef<Type> inputType = Type::getTypeRef(mRSC, constInput.get(), 1, 0, 0, false, false, 0);
@@ -518,7 +524,7 @@
mFontShaderFConstant.set(Allocation::createAllocation(mRSC, inputType.get(),
RS_ALLOCATION_USAGE_SCRIPT |
RS_ALLOCATION_USAGE_GRAPHICS_CONSTANTS));
- ProgramFragment *pf = new ProgramFragment(mRSC, shaderString.string(), shaderString.length(),
+ ProgramFragment *pf = new ProgramFragment(mRSC, shaderString, strlen(shaderString),
textureNames, numTextures, textureNamesLengths,
tmp, 4);
mFontShaderF.set(pf);
@@ -601,10 +607,9 @@
ObjectBaseRef<const Element> posElem = Element::createRef(mRSC, RS_TYPE_FLOAT_32, RS_KIND_USER, false, 3);
ObjectBaseRef<const Element> texElem = Element::createRef(mRSC, RS_TYPE_FLOAT_32, RS_KIND_USER, false, 2);
- Element::Builder builder;
- builder.add(posElem.get(), "position", 1);
- builder.add(texElem.get(), "texture0", 1);
- ObjectBaseRef<const Element> vertexDataElem = builder.create(mRSC);
+ const char *ebn1[] = { "position", "texture0" };
+ const Element *ebe1[] = {posElem.get(), texElem.get()};
+ ObjectBaseRef<const Element> vertexDataElem = Element::create(mRSC, 2, ebe1, ebn1);
ObjectBaseRef<Type> vertexDataType = Type::getTypeRef(mRSC, vertexDataElem.get(),
mMaxNumberOfQuads * 4,
@@ -634,11 +639,9 @@
initVertexArrayBuffers();
// We store a string with letters in a rough frequency of occurrence
- mLatinPrecache = String8(" eisarntolcdugpmhbyfvkwzxjq");
- mLatinPrecache += String8("EISARNTOLCDUGPMHBYFVKWZXJQ");
- mLatinPrecache += String8(",.?!()-+@;:`'");
- mLatinPrecache += String8("0123456789");
-
+ mLatinPrecache = " eisarntolcdugpmhbyfvkwzxjq"
+ "EISARNTOLCDUGPMHBYFVKWZXJQ"
+ ",.?!()-+@;:`'0123456789";
mInitialized = true;
}
@@ -734,7 +737,8 @@
// Remaining capacity is measured in %
uint32_t remainingCapacity = getRemainingCacheCapacity();
uint32_t precacheIdx = 0;
- while (remainingCapacity > 25 && precacheIdx < mLatinPrecache.size()) {
+ const size_t l = strlen(mLatinPrecache);
+ while ((remainingCapacity > 25) && (precacheIdx < l)) {
font->getCachedUTFChar((int32_t)mLatinPrecache[precacheIdx]);
remainingCapacity = getRemainingCacheCapacity();
precacheIdx ++;
@@ -753,11 +757,12 @@
Font *currentFont = mRSC->getFont();
if (!currentFont) {
if (!mDefault.get()) {
- String8 fontsDir("/fonts/Roboto-Regular.ttf");
- String8 fullPath(getenv("ANDROID_ROOT"));
- fullPath += fontsDir;
-
- mDefault.set(Font::create(mRSC, fullPath.string(), 8, mRSC->getDPI()));
+ char fullPath[1024];
+ const char * root = getenv("ANDROID_ROOT");
+ rsAssert(strlen(root) < 256);
+ strcpy(fullPath, root);
+ strcat(fullPath, "/fonts/Roboto-Regular.ttf");
+ mDefault.set(Font::create(mRSC, fullPath, 8, mRSC->getDPI()));
}
currentFont = mDefault.get();
}
diff --git a/rsFont.h b/rsFont.h
index 8f43a2a..7bac508 100644
--- a/rsFont.h
+++ b/rsFont.h
@@ -18,7 +18,6 @@
#define ANDROID_RS_FONT_H
#include "rsStream.h"
-#include <utils/String8.h>
#include <utils/Vector.h>
#include <utils/KeyedVector.h>
@@ -113,7 +112,7 @@
int32_t mBitmapTop;
};
- String8 mFontName;
+ const char *mFontName;
float mFontSize;
uint32_t mDpi;
@@ -183,7 +182,7 @@
uint32_t getRemainingCacheCapacity();
void precacheLatin(Font *font);
- String8 mLatinPrecache;
+ const char *mLatinPrecache;
Context *mRSC;
diff --git a/rsGrallocConsumer.cpp b/rsGrallocConsumer.cpp
new file mode 100644
index 0000000..e3bd9d4
--- /dev/null
+++ b/rsGrallocConsumer.cpp
@@ -0,0 +1,187 @@
+/*
+ * Copyright (C) 2013 The Android Open Source Project
+ *
+ * Licensed under the Apache License, Version 2.0 (the "License");
+ * you may not use this file except in compliance with the License.
+ * You may obtain a copy of the License at
+ *
+ * http://www.apache.org/licenses/LICENSE-2.0
+ *
+ * Unless required by applicable law or agreed to in writing, software
+ * distributed under the License is distributed on an "AS IS" BASIS,
+ * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ * See the License for the specific language governing permissions and
+ * limitations under the License.
+ */
+
+#define ATRACE_TAG ATRACE_TAG_RS
+
+#include "rsContext.h"
+#include "rsAllocation.h"
+#include "rsAdapter.h"
+#include "rs_hal.h"
+
+#include <cutils/compiler.h>
+#include <utils/Log.h>
+#include "rsGrallocConsumer.h"
+#include <ui/GraphicBuffer.h>
+
+
+namespace android {
+namespace renderscript {
+
+GrallocConsumer::GrallocConsumer(Allocation *a, const sp<IGraphicBufferConsumer>& bq) :
+ ConsumerBase(bq, true)
+{
+ mAlloc = a;
+ mConsumer->setConsumerUsageBits(GRALLOC_USAGE_SW_READ_OFTEN);
+ mConsumer->setMaxAcquiredBufferCount(2);
+
+ uint32_t y = a->mHal.drvState.lod[0].dimY;
+ if (y < 1) y = 1;
+ mConsumer->setDefaultBufferSize(a->mHal.drvState.lod[0].dimX, y);
+
+ //mBufferQueue->setDefaultBufferFormat(defaultFormat);
+ //mBufferQueue->setConsumerName(name);
+}
+
+GrallocConsumer::~GrallocConsumer() {
+ // ConsumerBase destructor does all the work.
+}
+
+
+
+status_t GrallocConsumer::lockNextBuffer() {
+ Mutex::Autolock _l(mMutex);
+ status_t err;
+
+ if (mAcquiredBuffer.mSlot != BufferQueue::INVALID_BUFFER_SLOT) {
+ err = releaseAcquiredBufferLocked();
+ if (err) {
+ return err;
+ }
+ }
+
+ BufferQueue::BufferItem b;
+
+ err = acquireBufferLocked(&b, 0);
+ if (err != OK) {
+ if (err == BufferQueue::NO_BUFFER_AVAILABLE) {
+ return BAD_VALUE;
+ } else {
+ ALOGE("Error acquiring buffer: %s (%d)", strerror(err), err);
+ return err;
+ }
+ }
+
+ int buf = b.mBuf;
+
+ if (b.mFence.get()) {
+ err = b.mFence->waitForever("GrallocConsumer::lockNextBuffer");
+ if (err != OK) {
+ ALOGE("Failed to wait for fence of acquired buffer: %s (%d)",
+ strerror(-err), err);
+ return err;
+ }
+ }
+
+ void *bufferPointer = NULL;
+ android_ycbcr ycbcr = android_ycbcr();
+
+ if (mSlots[buf].mGraphicBuffer->getPixelFormat() ==
+ HAL_PIXEL_FORMAT_YCbCr_420_888) {
+ err = mSlots[buf].mGraphicBuffer->lockYCbCr(
+ GraphicBuffer::USAGE_SW_READ_OFTEN,
+ b.mCrop,
+ &ycbcr);
+
+ if (err != OK) {
+ ALOGE("Unable to lock YCbCr buffer for CPU reading: %s (%d)",
+ strerror(-err), err);
+ return err;
+ }
+ bufferPointer = ycbcr.y;
+ } else {
+ err = mSlots[buf].mGraphicBuffer->lock(
+ GraphicBuffer::USAGE_SW_READ_OFTEN,
+ b.mCrop,
+ &bufferPointer);
+
+ if (err != OK) {
+ ALOGE("Unable to lock buffer for CPU reading: %s (%d)",
+ strerror(-err), err);
+ return err;
+ }
+ }
+
+ size_t lockedIdx = 0;
+ assert(mAcquiredBuffer.mSlot == BufferQueue::INVALID_BUFFER_SLOT);
+
+ mAcquiredBuffer.mSlot = buf;
+ mAcquiredBuffer.mBufferPointer = bufferPointer;
+ mAcquiredBuffer.mGraphicBuffer = mSlots[buf].mGraphicBuffer;
+
+ mAlloc->mHal.drvState.lod[0].mallocPtr = reinterpret_cast<uint8_t*>(bufferPointer);
+ mAlloc->mHal.drvState.lod[0].stride = mSlots[buf].mGraphicBuffer->getStride() *
+ mAlloc->mHal.state.type->getElementSizeBytes();
+ mAlloc->mHal.state.nativeBuffer = mAcquiredBuffer.mGraphicBuffer->getNativeBuffer();
+ mAlloc->mHal.state.timestamp = b.mTimestamp;
+
+ assert(mAlloc->mHal.drvState.lod[0].dimX ==
+ mSlots[buf].mGraphicBuffer->getWidth());
+ assert(mAlloc->mHal.drvState.lod[0].dimY ==
+ mSlots[buf].mGraphicBuffer->getHeight());
+
+ //mAlloc->format = mSlots[buf].mGraphicBuffer->getPixelFormat();
+
+ //mAlloc->crop = b.mCrop;
+ //mAlloc->transform = b.mTransform;
+ //mAlloc->scalingMode = b.mScalingMode;
+ //mAlloc->frameNumber = b.mFrameNumber;
+
+ if (mAlloc->mHal.state.yuv) {
+ mAlloc->mHal.drvState.lod[1].mallocPtr = ycbcr.cb;
+ mAlloc->mHal.drvState.lod[2].mallocPtr = ycbcr.cr;
+
+ mAlloc->mHal.drvState.lod[0].stride = ycbcr.ystride;
+ mAlloc->mHal.drvState.lod[1].stride = ycbcr.cstride;
+ mAlloc->mHal.drvState.lod[2].stride = ycbcr.cstride;
+ }
+
+ return OK;
+}
+
+status_t GrallocConsumer::unlockBuffer() {
+ Mutex::Autolock _l(mMutex);
+ return releaseAcquiredBufferLocked();
+}
+
+status_t GrallocConsumer::releaseAcquiredBufferLocked() {
+ status_t err;
+
+ err = mAcquiredBuffer.mGraphicBuffer->unlock();
+ if (err != OK) {
+ ALOGE("%s: Unable to unlock graphic buffer", __FUNCTION__);
+ return err;
+ }
+ int buf = mAcquiredBuffer.mSlot;
+
+ // release the buffer if it hasn't already been freed by the BufferQueue.
+ // This can happen, for example, when the producer of this buffer
+ // disconnected after this buffer was acquired.
+ if (CC_LIKELY(mAcquiredBuffer.mGraphicBuffer ==
+ mSlots[buf].mGraphicBuffer)) {
+ releaseBufferLocked(
+ buf, mAcquiredBuffer.mGraphicBuffer,
+ EGL_NO_DISPLAY, EGL_NO_SYNC_KHR);
+ }
+
+ mAcquiredBuffer.mSlot = BufferQueue::INVALID_BUFFER_SLOT;
+ mAcquiredBuffer.mBufferPointer = NULL;
+ mAcquiredBuffer.mGraphicBuffer.clear();
+ return OK;
+}
+
+} // namespace renderscript
+} // namespace android
+
diff --git a/rsGrallocConsumer.h b/rsGrallocConsumer.h
new file mode 100644
index 0000000..9e4fc58
--- /dev/null
+++ b/rsGrallocConsumer.h
@@ -0,0 +1,78 @@
+/*
+ * Copyright (C) 2013 The Android Open Source Project
+ *
+ * Licensed under the Apache License, Version 2.0 (the "License");
+ * you may not use this file except in compliance with the License.
+ * You may obtain a copy of the License at
+ *
+ * http://www.apache.org/licenses/LICENSE-2.0
+ *
+ * Unless required by applicable law or agreed to in writing, software
+ * distributed under the License is distributed on an "AS IS" BASIS,
+ * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ * See the License for the specific language governing permissions and
+ * limitations under the License.
+ */
+
+#ifndef ANDROID_RS_GRALLOC_CONSUMER_H
+#define ANDROID_RS_GRALLOC_CONSUMER_H
+
+#include <gui/ConsumerBase.h>
+
+#include <ui/GraphicBuffer.h>
+
+#include <utils/String8.h>
+#include <utils/Vector.h>
+#include <utils/threads.h>
+
+
+// ---------------------------------------------------------------------------
+namespace android {
+namespace renderscript {
+
+class Allocation;
+
+/**
+ * CpuConsumer is a BufferQueue consumer endpoint that allows direct CPU
+ * access to the underlying gralloc buffers provided by BufferQueue. Multiple
+ * buffers may be acquired by it at once, to be used concurrently by the
+ * CpuConsumer owner. Sets gralloc usage flags to be software-read-only.
+ * This queue is synchronous by default.
+ */
+class GrallocConsumer : public ConsumerBase
+{
+ public:
+ typedef ConsumerBase::FrameAvailableListener FrameAvailableListener;
+
+ GrallocConsumer(Allocation *, const sp<IGraphicBufferConsumer>& bq);
+
+ virtual ~GrallocConsumer();
+ status_t lockNextBuffer();
+ status_t unlockBuffer();
+
+ private:
+ status_t releaseAcquiredBufferLocked();
+ Allocation *mAlloc;
+
+ // Tracking for buffers acquired by the user
+ struct AcquiredBuffer {
+ // Need to track the original mSlot index and the buffer itself because
+ // the mSlot entry may be freed/reused before the acquired buffer is
+ // released.
+ int mSlot;
+ sp<GraphicBuffer> mGraphicBuffer;
+ void *mBufferPointer;
+
+ AcquiredBuffer() :
+ mSlot(BufferQueue::INVALID_BUFFER_SLOT),
+ mBufferPointer(NULL) {
+ }
+ };
+ AcquiredBuffer mAcquiredBuffer;
+};
+
+} // namespace renderscript
+} // namespace android
+
+#endif // ANDROID_RS_GRALLOC_CONSUMER_H
+
diff --git a/rsMesh.cpp b/rsMesh.cpp
index 651a8f3..8decfc5 100644
--- a/rsMesh.cpp
+++ b/rsMesh.cpp
@@ -81,9 +81,7 @@
void Mesh::serialize(Context *rsc, OStream *stream) const {
// Need to identify ourselves
stream->addU32((uint32_t)getClassId());
-
- String8 name(getName());
- stream->addString(&name);
+ stream->addString(getName());
// Store number of vertex streams
stream->addU32(mHal.state.vertexBuffersCount);
@@ -113,8 +111,7 @@
return NULL;
}
- String8 name;
- stream->loadString(&name);
+ const char *name = stream->loadString();
uint32_t vertexBuffersCount = stream->loadU32();
ObjectBaseRef<Allocation> *vertexBuffers = NULL;
@@ -148,7 +145,7 @@
}
Mesh *mesh = new Mesh(rsc, vertexBuffersCount, primitivesCount);
- mesh->setName(name.string(), name.size());
+ mesh->assignName(name);
for (uint32_t vCount = 0; vCount < vertexBuffersCount; vCount ++) {
mesh->setVertexBuffer(vertexBuffers[vCount].get(), vCount);
}
diff --git a/rsObjectBase.cpp b/rsObjectBase.cpp
index c8d7349..bf5d3b5 100644
--- a/rsObjectBase.cpp
+++ b/rsObjectBase.cpp
@@ -284,4 +284,3 @@
asyncUnlock();
return false;
}
-
diff --git a/rsObjectBase.h b/rsObjectBase.h
index 12c89a3..85f8fd8 100644
--- a/rsObjectBase.h
+++ b/rsObjectBase.h
@@ -44,6 +44,7 @@
const char * getName() const {
return mName;
}
+ void assignName(const char *s) {mName = s;}
void setName(const char *);
void setName(const char *, uint32_t len);
diff --git a/rsProgram.cpp b/rsProgram.cpp
index 7114f29..806d1de 100644
--- a/rsProgram.cpp
+++ b/rsProgram.cpp
@@ -77,7 +77,9 @@
shaderText += internalTokenLen;
shaderLength -= internalTokenLen;
}
- mUserShader.setTo(shaderText, shaderLength);
+
+ mUserShader = rsuCopyString(shaderText, shaderLength);
+ mUserShaderLen = shaderLength;
}
Program::~Program() {
@@ -98,6 +100,12 @@
mHal.state.inputElementsCount = 0;
mHal.state.constantsCount = 0;
mHal.state.texturesCount = 0;
+
+ if (mUserShader != NULL) {
+ delete[] mUserShader;
+ mUserShader = NULL;
+ }
+ mUserShaderLen = 0;
}
bool Program::freeChildren() {
@@ -134,6 +142,9 @@
mConstants = NULL;
mIsInternal = false;
+
+ mUserShader = NULL;
+ mUserShaderLen = 0;
}
void Program::bindAllocation(Context *rsc, Allocation *alloc, uint32_t slot) {
diff --git a/rsProgram.h b/rsProgram.h
index d032930..be5631f 100644
--- a/rsProgram.h
+++ b/rsProgram.h
@@ -78,7 +78,8 @@
ObjectBaseRef<Element> *mInputElements;
bool mIsInternal;
- String8 mUserShader;
+ const char *mUserShader;
+ size_t mUserShaderLen;
void initMemberVars();
};
diff --git a/rsProgramFragment.cpp b/rsProgramFragment.cpp
index f2885a9..877b8bd 100644
--- a/rsProgramFragment.cpp
+++ b/rsProgramFragment.cpp
@@ -30,7 +30,7 @@
mConstantColor[2] = 1.f;
mConstantColor[3] = 1.f;
- mRSC->mHal.funcs.fragment.init(mRSC, this, mUserShader.string(), mUserShader.length(),
+ mRSC->mHal.funcs.fragment.init(mRSC, this, mUserShader, mUserShaderLen,
textureNames, textureNamesCount, textureNamesLength);
}
@@ -93,18 +93,20 @@
}
void ProgramFragmentState::init(Context *rsc) {
- String8 shaderString(RS_SHADER_INTERNAL);
- shaderString.append("varying lowp vec4 varColor;\n");
- shaderString.append("varying vec2 varTex0;\n");
- shaderString.append("void main() {\n");
- shaderString.append(" lowp vec4 col = UNI_Color;\n");
- shaderString.append(" gl_FragColor = col;\n");
- shaderString.append("}\n");
+ const char *shaderString =
+ RS_SHADER_INTERNAL
+ "varying lowp vec4 varColor;\n"
+ "varying vec2 varTex0;\n"
+ "void main() {\n"
+ " lowp vec4 col = UNI_Color;\n"
+ " gl_FragColor = col;\n"
+ "}\n";
ObjectBaseRef<const Element> colorElem = Element::createRef(rsc, RS_TYPE_FLOAT_32, RS_KIND_USER, false, 4);
- Element::Builder builder;
- builder.add(colorElem.get(), "Color", 1);
- ObjectBaseRef<const Element> constInput = builder.create(rsc);
+
+ const char *enames[] = { "Color" };
+ const Element *eins[] = {colorElem.get()};
+ ObjectBaseRef<const Element> constInput = Element::create(rsc, 1, eins, enames);
ObjectBaseRef<Type> inputType = Type::getTypeRef(rsc, constInput.get(), 1, 0, 0, false, false, 0);
@@ -114,7 +116,7 @@
Allocation *constAlloc = Allocation::createAllocation(rsc, inputType.get(),
RS_ALLOCATION_USAGE_SCRIPT | RS_ALLOCATION_USAGE_GRAPHICS_CONSTANTS);
- ProgramFragment *pf = new ProgramFragment(rsc, shaderString.string(), shaderString.length(),
+ ProgramFragment *pf = new ProgramFragment(rsc, shaderString, strlen(shaderString),
NULL, 0, NULL, tmp, 2);
pf->bindAllocation(rsc, constAlloc, 0);
pf->setConstantColor(rsc, 1.0f, 1.0f, 1.0f, 1.0f);
diff --git a/rsProgramVertex.cpp b/rsProgramVertex.cpp
index 5265216..549dcae 100644
--- a/rsProgramVertex.cpp
+++ b/rsProgramVertex.cpp
@@ -27,7 +27,7 @@
const uint32_t * params, size_t paramLength)
: Program(rsc, shaderText, shaderLength, params, paramLength) {
- mRSC->mHal.funcs.vertex.init(mRSC, this, mUserShader.string(), mUserShader.length(),
+ mRSC->mHal.funcs.vertex.init(mRSC, this, mUserShader, mUserShaderLen,
textureNames, textureNamesCount, textureNamesLength);
}
@@ -174,31 +174,27 @@
ObjectBaseRef<const Element> f4Elem = Element::createRef(rsc, RS_TYPE_FLOAT_32,
RS_KIND_USER, false, 4);
- Element::Builder constBuilder;
- constBuilder.add(matrixElem.get(), "MV", 1);
- constBuilder.add(matrixElem.get(), "P", 1);
- constBuilder.add(matrixElem.get(), "TexMatrix", 1);
- constBuilder.add(matrixElem.get(), "MVP", 1);
- ObjectBaseRef<const Element> constInput = constBuilder.create(rsc);
+ const char *ebn1[] = { "MV", "P", "TexMatrix", "MVP" };
+ const Element *ebe1[] = {matrixElem.get(), matrixElem.get(),
+ matrixElem.get(), matrixElem.get()};
+ ObjectBaseRef<const Element> constInput = Element::create(rsc, 4, ebe1, ebn1);
- Element::Builder inputBuilder;
- inputBuilder.add(f4Elem.get(), "position", 1);
- inputBuilder.add(f4Elem.get(), "color", 1);
- inputBuilder.add(f3Elem.get(), "normal", 1);
- inputBuilder.add(f2Elem.get(), "texture0", 1);
- ObjectBaseRef<const Element> attrElem = inputBuilder.create(rsc);
+ const char *ebn2[] = { "position", "color", "normal", "texture0" };
+ const Element *ebe2[] = {f4Elem.get(), f4Elem.get(), f3Elem.get(), f2Elem.get()};
+ ObjectBaseRef<const Element> attrElem = Element::create(rsc, 4, ebe2, ebn2);
ObjectBaseRef<Type> inputType = Type::getTypeRef(rsc, constInput.get(), 1, 0, 0, false, false, 0);
- String8 shaderString(RS_SHADER_INTERNAL);
- shaderString.append("varying vec4 varColor;\n");
- shaderString.append("varying vec2 varTex0;\n");
- shaderString.append("void main() {\n");
- shaderString.append(" gl_Position = UNI_MVP * ATTRIB_position;\n");
- shaderString.append(" gl_PointSize = 1.0;\n");
- shaderString.append(" varColor = ATTRIB_color;\n");
- shaderString.append(" varTex0 = ATTRIB_texture0;\n");
- shaderString.append("}\n");
+ const char *shaderString =
+ RS_SHADER_INTERNAL
+ "varying vec4 varColor;\n"
+ "varying vec2 varTex0;\n"
+ "void main() {\n"
+ " gl_Position = UNI_MVP * ATTRIB_position;\n"
+ " gl_PointSize = 1.0;\n"
+ " varColor = ATTRIB_color;\n"
+ " varTex0 = ATTRIB_texture0;\n"
+ "}\n";
uint32_t tmp[4];
tmp[0] = RS_PROGRAM_PARAM_CONSTANT;
@@ -206,7 +202,7 @@
tmp[2] = RS_PROGRAM_PARAM_INPUT;
tmp[3] = (uint32_t)attrElem.get();
- ProgramVertex *pv = new ProgramVertex(rsc, shaderString.string(), shaderString.length(),
+ ProgramVertex *pv = new ProgramVertex(rsc, shaderString, strlen(shaderString),
NULL, 0, NULL, tmp, 4);
Allocation *alloc = Allocation::createAllocation(rsc, inputType.get(),
RS_ALLOCATION_USAGE_SCRIPT | RS_ALLOCATION_USAGE_GRAPHICS_CONSTANTS);
diff --git a/rsRuntime.h b/rsRuntime.h
index 2939176..1d81ffb 100644
--- a/rsRuntime.h
+++ b/rsRuntime.h
@@ -108,6 +108,10 @@
int32_t *left, int32_t *right, int32_t *top, int32_t *bottom);
void rsrBindFont(Context *, Font *);
void rsrFontColor(Context *, float r, float g, float b, float a);
+
+void rsrAllocationIoSend(Context *, Allocation *);
+void rsrAllocationIoReceive(Context *, Allocation *);
+
#endif
//////////////////////////////////////////////////////////////////////////////
diff --git a/rsSampler.cpp b/rsSampler.cpp
index fededb1..2a8476a 100644
--- a/rsSampler.cpp
+++ b/rsSampler.cpp
@@ -98,7 +98,13 @@
}
ObjectBase::asyncUnlock();
- Sampler *s = new Sampler(rsc, magFilter, minFilter, wrapS, wrapT, wrapR, aniso);
+ void* allocMem = rsc->mHal.funcs.allocRuntimeMem(sizeof(Sampler), 0);
+ if (!allocMem) {
+ rsc->setError(RS_ERROR_FATAL_DRIVER, "Couldn't allocate memory for Allocation");
+ return NULL;
+ }
+
+ Sampler *s = new (allocMem) Sampler(rsc, magFilter, minFilter, wrapS, wrapT, wrapR, aniso);
returnRef.set(s);
ObjectBase::asyncLock();
@@ -108,6 +114,14 @@
return returnRef;
}
+void Sampler::operator delete(void* ptr) {
+ if (ptr) {
+ Sampler *s = (Sampler*) ptr;
+ s->getContext()->mHal.funcs.freeRuntimeMem(ptr);
+ }
+}
+
+
////////////////////////////////
namespace android {
diff --git a/rsSampler.h b/rsSampler.h
index 81220a8..2fdf707 100644
--- a/rsSampler.h
+++ b/rsSampler.h
@@ -51,6 +51,8 @@
};
Hal mHal;
+ void operator delete(void* ptr);
+
static ObjectBaseRef<Sampler> getSampler(Context *,
RsSamplerValue magFilter,
RsSamplerValue minFilter,
diff --git a/rsScript.h b/rsScript.h
index a1360b5..c868aaa 100644
--- a/rsScript.h
+++ b/rsScript.h
@@ -19,6 +19,7 @@
#include "rsAllocation.h"
+#include <utility>
// ---------------------------------------------------------------------------
namespace android {
@@ -72,6 +73,7 @@
size_t exportedPragmaCount;
char const **exportedPragmaKeyList;
char const **exportedPragmaValueList;
+ const std::pair<const char *, uint32_t> *exportedForeachFuncList;
int (* root)();
};
diff --git a/rsScriptC.cpp b/rsScriptC.cpp
index d8f356c..49c9b57 100644
--- a/rsScriptC.cpp
+++ b/rsScriptC.cpp
@@ -26,6 +26,7 @@
#if !defined(RS_SERVER) && !defined(RS_COMPATIBILITY_LIB)
#include "utils/Timers.h"
+#include "cutils/trace.h"
#endif
#include <sys/stat.h>
@@ -166,15 +167,32 @@
const void * usr,
size_t usrBytes,
const RsScriptCall *sc) {
+ // Trace this function call.
+ // To avoid overhead, we only build the string, if tracing is actually
+ // enabled.
+ String8 *AString = NULL;
+ const char *String = "";
+ if (ATRACE_ENABLED()) {
+ AString = new String8("runForEach_");
+ AString->append(mHal.info.exportedForeachFuncList[slot].first);
+ String = AString->string();
+ }
+ ATRACE_NAME(String);
+ (void)String;
Context::PushState ps(rsc);
setupGLState(rsc);
setupScript(rsc);
rsc->mHal.funcs.script.invokeForEach(rsc, this, slot, ain, aout, usr, usrBytes, sc);
+
+ if (AString)
+ delete AString;
}
void ScriptC::Invoke(Context *rsc, uint32_t slot, const void *data, size_t len) {
+ ATRACE_CALL();
+
if (slot >= mHal.info.exportedFunctionCount) {
rsc->setError(RS_ERROR_BAD_SCRIPT, "Calling invoke on bad script");
return;
@@ -229,7 +247,7 @@
const char *cacheDir,
const uint8_t *bitcode,
size_t bitcodeLen) {
-
+ ATRACE_CALL();
//ALOGE("runCompiler %p %p %p %p %p %i", rsc, this, resName, cacheDir, bitcode, bitcodeLen);
#ifndef RS_COMPATIBILITY_LIB
#ifndef ANDROID_RS_SERIALIZE
diff --git a/rsScriptC_Lib.cpp b/rsScriptC_Lib.cpp
index 123b8b3..842d8d2 100644
--- a/rsScriptC_Lib.cpp
+++ b/rsScriptC_Lib.cpp
@@ -171,6 +171,13 @@
return rsc->sendMessageToClient(data, RS_MESSAGE_TO_CLIENT_USER, cmdID, len, true);
}
+void rsrAllocationIoSend(Context *rsc, Allocation *src) {
+ src->ioSend(rsc);
+}
+
+void rsrAllocationIoReceive(Context *rsc, Allocation *src) {
+ src->ioReceive(rsc);
+}
void rsrForEach(Context *rsc,
Script *target,
diff --git a/rsStream.cpp b/rsStream.cpp
index b9df0cc..a9587c1 100644
--- a/rsStream.cpp
+++ b/rsStream.cpp
@@ -43,10 +43,11 @@
return loadU32();
}
-void IStream::loadString(String8 *s) {
+const char * IStream::loadString() {
uint32_t len = loadU32();
- s->setTo((const char *)&mData[mPos], len);
+ const char *s = rsuCopyString((const char *)&mData[mPos], len);
mPos += len;
+ return s;
}
// Output stream implementation
@@ -89,19 +90,20 @@
}
}
-void OStream::addString(String8 *s) {
- uint32_t len = s->size();
+void OStream::addString(const char *s, size_t len) {
addU32(len);
if (mPos + len*sizeof(char) >= mLength) {
growSize();
}
char *stringData = reinterpret_cast<char *>(&mData[mPos]);
- for (uint32_t i = 0; i < len; i ++) {
- stringData[i] = s->string()[i];
- }
+ memcpy(stringData, s, len);
mPos += len*sizeof(char);
}
+void OStream::addString(const char *s) {
+ addString(s, strlen(s));
+}
+
void OStream::growSize() {
uint8_t *newData = (uint8_t*)malloc(mLength*2);
memcpy(newData, mData, mLength*sizeof(uint8_t));
diff --git a/rsStream.h b/rsStream.h
index 8a192e6..d0ed62e 100644
--- a/rsStream.h
+++ b/rsStream.h
@@ -59,7 +59,7 @@
}
void loadByteArray(void *dest, size_t numBytes);
uint64_t loadOffset();
- void loadString(String8 *s);
+ const char * loadString();
uint64_t getPos() const {
return mPos;
}
@@ -132,7 +132,8 @@
}
void addByteArray(const void *src, size_t numBytes);
void addOffset(uint64_t v);
- void addString(String8 *s);
+ void addString(const char *name);
+ void addString(const char *name, size_t len);
uint64_t getPos() const {
return mPos;
}
diff --git a/rsType.cpp b/rsType.cpp
index 9719a04..15e2550 100644
--- a/rsType.cpp
+++ b/rsType.cpp
@@ -188,9 +188,7 @@
void Type::serialize(Context *rsc, OStream *stream) const {
// Need to identify ourselves
stream->addU32((uint32_t)getClassId());
-
- String8 name(getName());
- stream->addString(&name);
+ stream->addString(getName());
mElement->serialize(rsc, stream);
@@ -210,8 +208,7 @@
return NULL;
}
- String8 name;
- stream->loadString(&name);
+ const char *name = stream->loadString();
Element *elem = Element::createFromStream(rsc, stream);
if (!elem) {
@@ -225,6 +222,8 @@
uint8_t faces = stream->loadU8();
Type *type = Type::getType(rsc, elem, x, y, z, lod != 0, faces !=0, 0);
elem->decUserRef();
+
+ delete [] name;
return type;
}
@@ -346,7 +345,7 @@
}
}
-void rsaTypeGetNativeData(RsContext con, RsType type, uintptr_t *typeData, uint32_t typeDataSize) {
+extern "C" void rsaTypeGetNativeData(RsContext con, RsType type, uintptr_t *typeData, uint32_t typeDataSize) {
rsAssert(typeDataSize == 6);
// Pack the data in the follofing way mHal.state.dimX; mHal.state.dimY; mHal.state.dimZ;
// mHal.state.lodCount; mHal.state.faces; mElement; into typeData
diff --git a/rs_hal.h b/rs_hal.h
index 1ed26dc..16bd890 100644
--- a/rs_hal.h
+++ b/rs_hal.h
@@ -145,15 +145,23 @@
struct {
bool (*init)(const Context *rsc, Allocation *alloc, bool forceZero);
void (*destroy)(const Context *rsc, Allocation *alloc);
+ uint32_t (*grallocBits)(const Context *rsc, Allocation *alloc);
void (*resize)(const Context *rsc, const Allocation *alloc, const Type *newType,
bool zeroNew);
void (*syncAll)(const Context *rsc, const Allocation *alloc, RsAllocationUsageType src);
void (*markDirty)(const Context *rsc, const Allocation *alloc);
- void * (*getSurface)(const Context *rsc, const Allocation *alloc);
void (*setSurface)(const Context *rsc, Allocation *alloc, ANativeWindow *sur);
void (*ioSend)(const Context *rsc, Allocation *alloc);
+
+ /**
+ * A new gralloc buffer is in use. The pointers and strides in
+ * mHal.drvState.lod[0-2] will be updated with the new values.
+ *
+ * The new gralloc handle is provided in mHal.state.nativeBuffer
+ *
+ */
void (*ioReceive)(const Context *rsc, Allocation *alloc);
void (*data1D)(const Context *rsc, const Allocation *alloc,
diff --git a/rs_native.spec b/rs_native.spec
index fcede40..e69de29 100644
--- a/rs_native.spec
+++ b/rs_native.spec
@@ -1,9 +0,0 @@
-
-AllocationIoSend {
- param RsAllocation alloc
- }
-
-AllocationIoReceive {
- param RsAllocation alloc
- }
-
diff --git a/rsg_generator.c b/rsg_generator.c
index ca26851..fd531b6 100644
--- a/rsg_generator.c
+++ b/rsg_generator.c
@@ -123,9 +123,12 @@
fprintf(f, ")");
}
-void printFuncDecls(FILE *f, const char *prefix, int addContext) {
+void printFuncDecls(FILE *f, const char *prefix, int addContext, int externC) {
int ct;
for (ct=0; ct < apiCount; ct++) {
+ if (externC) {
+ fprintf(f, "extern \"C\" ");
+ }
printFuncDecl(f, &apis[ct], prefix, addContext, 0);
fprintf(f, ";\n");
}
@@ -400,6 +403,8 @@
int needFlush = 0;
const ApiEntry * api = &apis[ct];
+ fprintf(f, "extern \"C\" ");
+
printFuncDecl(f, api, "rs", 0, 0);
fprintf(f, "\n{\n");
fprintf(f, " ");
@@ -655,7 +660,7 @@
fprintf(f, "namespace android {\n");
fprintf(f, "namespace renderscript {\n");
printStructures(f);
- printFuncDecls(f, "rsi_", 1);
+ printFuncDecls(f, "rsi_", 1, 0);
printPlaybackFuncs(f, "rsp_");
fprintf(f, "\n\ntypedef struct RsPlaybackRemoteHeaderRec {\n");
fprintf(f, " uint32_t command;\n");
@@ -673,7 +678,7 @@
case '1': // rsgApiFuncDecl.h
{
- printFuncDecls(f, "rs", 0);
+ printFuncDecls(f, "rs", 0, 1);
}
break;
diff --git a/scriptc/rs_allocation.rsh b/scriptc/rs_allocation.rsh
index 0f5e79d..7dffa54 100644
--- a/scriptc/rs_allocation.rsh
+++ b/scriptc/rs_allocation.rsh
@@ -134,17 +134,17 @@
* Extract a single element from an allocation.
*/
extern const void * __attribute__((overloadable))
- rsGetElementAt(rs_allocation, uint32_t x);
+ rsGetElementAt(rs_allocation a, uint32_t x);
/**
* \overload
*/
extern const void * __attribute__((overloadable))
- rsGetElementAt(rs_allocation, uint32_t x, uint32_t y);
+ rsGetElementAt(rs_allocation a, uint32_t x, uint32_t y);
/**
* \overload
*/
extern const void * __attribute__((overloadable))
- rsGetElementAt(rs_allocation, uint32_t x, uint32_t y, uint32_t z);
+ rsGetElementAt(rs_allocation a, uint32_t x, uint32_t y, uint32_t z);
#if (defined(RS_VERSION) && (RS_VERSION >= 18))
diff --git a/scriptc/rs_cl.rsh b/scriptc/rs_cl.rsh
index 788aea8..2ff5d8b 100644
--- a/scriptc/rs_cl.rsh
+++ b/scriptc/rs_cl.rsh
@@ -25,11 +25,11 @@
// Conversions
#define CVT_FUNC_2(typeout, typein) \
-_RS_RUNTIME typeout##2 __attribute__((overloadable)) \
+_RS_RUNTIME typeout##2 __attribute__((const, overloadable)) \
convert_##typeout##2(typein##2 v); \
-_RS_RUNTIME typeout##3 __attribute__((overloadable)) \
+_RS_RUNTIME typeout##3 __attribute__((const, overloadable)) \
convert_##typeout##3(typein##3 v); \
-_RS_RUNTIME typeout##4 __attribute__((overloadable)) \
+_RS_RUNTIME typeout##4 __attribute__((const, overloadable)) \
convert_##typeout##4(typein##4 v);
@@ -92,98 +92,119 @@
// Float ops, 6.11.2
-#define FN_FUNC_FN(fnc) \
-_RS_RUNTIME float2 __attribute__((overloadable)) fnc(float2 v); \
-_RS_RUNTIME float3 __attribute__((overloadable)) fnc(float3 v); \
-_RS_RUNTIME float4 __attribute__((overloadable)) fnc(float4 v);
+#ifdef DOXYGEN
-#define F_FUNC_FN(fnc) \
-_RS_RUNTIME float __attribute__((overloadable)) fnc(float2 v); \
-_RS_RUNTIME float __attribute__((overloadable)) fnc(float3 v); \
-_RS_RUNTIME float __attribute__((overloadable)) fnc(float4 v);
+#define FN_FUNC_FN(fnc)
+#define F_FUNC_FN(fnc)
+#define IN_FUNC_FN(fnc)
+#define FN_FUNC_FN_FN(fnc)
+#define F_FUNC_FN_FN(fnc)
+#define FN_FUNC_FN_F(fnc)
+#define FN_FUNC_FN_IN(fnc)
+#define FN_FUNC_FN_I(fnc)
+#define FN_FUNC_FN_PFN(fnc)
+#define FN_FUNC_FN_PIN(fnc)
+#define FN_FUNC_FN_FN_FN(fnc)
+#define FN_FUNC_FN_FN_F(fnc)
+#define FN_FUNC_FN_F_F(fnc)
+#define FN_FUNC_FN_FN_PIN(fnc)
-#define IN_FUNC_FN(fnc) \
-_RS_RUNTIME int2 __attribute__((overloadable)) fnc(float2 v); \
-_RS_RUNTIME int3 __attribute__((overloadable)) fnc(float3 v); \
-_RS_RUNTIME int4 __attribute__((overloadable)) fnc(float4 v);
+#else
-#define FN_FUNC_FN_FN(fnc) \
-_RS_RUNTIME float2 __attribute__((overloadable)) fnc(float2 v1, float2 v2); \
-_RS_RUNTIME float3 __attribute__((overloadable)) fnc(float3 v1, float3 v2); \
-_RS_RUNTIME float4 __attribute__((overloadable)) fnc(float4 v1, float4 v2);
+#define FN_FUNC_FN(fnc) \
+_RS_RUNTIME float2 __attribute__((const, overloadable)) fnc(float2 v); \
+_RS_RUNTIME float3 __attribute__((const, overloadable)) fnc(float3 v); \
+_RS_RUNTIME float4 __attribute__((const, overloadable)) fnc(float4 v);
-#define F_FUNC_FN_FN(fnc) \
-_RS_RUNTIME float __attribute__((overloadable)) fnc(float2 v1, float2 v2); \
-_RS_RUNTIME float __attribute__((overloadable)) fnc(float3 v1, float3 v2); \
-_RS_RUNTIME float __attribute__((overloadable)) fnc(float4 v1, float4 v2);
+#define F_FUNC_FN(fnc) \
+_RS_RUNTIME float __attribute__((const, overloadable)) fnc(float2 v); \
+_RS_RUNTIME float __attribute__((const, overloadable)) fnc(float3 v); \
+_RS_RUNTIME float __attribute__((const, overloadable)) fnc(float4 v);
-#define FN_FUNC_FN_F(fnc) \
-_RS_RUNTIME float2 __attribute__((overloadable)) fnc(float2 v1, float v2); \
-_RS_RUNTIME float3 __attribute__((overloadable)) fnc(float3 v1, float v2); \
-_RS_RUNTIME float4 __attribute__((overloadable)) fnc(float4 v1, float v2);
+#define IN_FUNC_FN(fnc) \
+_RS_RUNTIME int2 __attribute__((const, overloadable)) fnc(float2 v); \
+_RS_RUNTIME int3 __attribute__((const, overloadable)) fnc(float3 v); \
+_RS_RUNTIME int4 __attribute__((const, overloadable)) fnc(float4 v);
-#define FN_FUNC_FN_IN(fnc) \
-_RS_RUNTIME float2 __attribute__((overloadable)) fnc(float2 v1, int2 v2); \
-_RS_RUNTIME float3 __attribute__((overloadable)) fnc(float3 v1, int3 v2); \
-_RS_RUNTIME float4 __attribute__((overloadable)) fnc(float4 v1, int4 v2); \
+#define FN_FUNC_FN_FN(fnc) \
+_RS_RUNTIME float2 __attribute__((const, overloadable)) fnc(float2 v1, float2 v2); \
+_RS_RUNTIME float3 __attribute__((const, overloadable)) fnc(float3 v1, float3 v2); \
+_RS_RUNTIME float4 __attribute__((const, overloadable)) fnc(float4 v1, float4 v2);
-#define FN_FUNC_FN_I(fnc) \
-_RS_RUNTIME float2 __attribute__((overloadable)) fnc(float2 v1, int v2); \
-_RS_RUNTIME float3 __attribute__((overloadable)) fnc(float3 v1, int v2); \
-_RS_RUNTIME float4 __attribute__((overloadable)) fnc(float4 v1, int v2);
+#define F_FUNC_FN_FN(fnc) \
+_RS_RUNTIME float __attribute__((const, overloadable)) fnc(float2 v1, float2 v2); \
+_RS_RUNTIME float __attribute__((const, overloadable)) fnc(float3 v1, float3 v2); \
+_RS_RUNTIME float __attribute__((const, overloadable)) fnc(float4 v1, float4 v2);
-#define FN_FUNC_FN_PFN(fnc) \
-_RS_RUNTIME float2 __attribute__((overloadable)) \
- fnc(float2 v1, float2 *v2); \
-_RS_RUNTIME float3 __attribute__((overloadable)) \
- fnc(float3 v1, float3 *v2); \
-_RS_RUNTIME float4 __attribute__((overloadable)) \
+#define FN_FUNC_FN_F(fnc) \
+_RS_RUNTIME float2 __attribute__((const, overloadable)) fnc(float2 v1, float v2); \
+_RS_RUNTIME float3 __attribute__((const, overloadable)) fnc(float3 v1, float v2); \
+_RS_RUNTIME float4 __attribute__((const, overloadable)) fnc(float4 v1, float v2);
+
+#define FN_FUNC_FN_IN(fnc) \
+_RS_RUNTIME float2 __attribute__((const, overloadable)) fnc(float2 v1, int2 v2); \
+_RS_RUNTIME float3 __attribute__((const, overloadable)) fnc(float3 v1, int3 v2); \
+_RS_RUNTIME float4 __attribute__((const, overloadable)) fnc(float4 v1, int4 v2);
+
+#define FN_FUNC_FN_I(fnc) \
+_RS_RUNTIME float2 __attribute__((const, overloadable)) fnc(float2 v1, int v2); \
+_RS_RUNTIME float3 __attribute__((const, overloadable)) fnc(float3 v1, int v2); \
+_RS_RUNTIME float4 __attribute__((const, overloadable)) fnc(float4 v1, int v2);
+
+#define FN_FUNC_FN_PFN(fnc) \
+_RS_RUNTIME float2 __attribute__((pure, overloadable)) \
+ fnc(float2 v1, float2 *v2); \
+_RS_RUNTIME float3 __attribute__((pure, overloadable)) \
+ fnc(float3 v1, float3 *v2); \
+_RS_RUNTIME float4 __attribute__((pure, overloadable)) \
fnc(float4 v1, float4 *v2);
-#define FN_FUNC_FN_PIN(fnc) \
-_RS_RUNTIME float2 __attribute__((overloadable)) fnc(float2 v1, int2 *v2); \
-_RS_RUNTIME float3 __attribute__((overloadable)) fnc(float3 v1, int3 *v2); \
-_RS_RUNTIME float4 __attribute__((overloadable)) fnc(float4 v1, int4 *v2);
+#define FN_FUNC_FN_PIN(fnc) \
+_RS_RUNTIME float2 __attribute__((pure, overloadable)) fnc(float2 v1, int2 *v2); \
+_RS_RUNTIME float3 __attribute__((pure, overloadable)) fnc(float3 v1, int3 *v2); \
+_RS_RUNTIME float4 __attribute__((pure, overloadable)) fnc(float4 v1, int4 *v2);
-#define FN_FUNC_FN_FN_FN(fnc) \
-_RS_RUNTIME float2 __attribute__((overloadable)) \
- fnc(float2 v1, float2 v2, float2 v3); \
-_RS_RUNTIME float3 __attribute__((overloadable)) \
- fnc(float3 v1, float3 v2, float3 v3); \
-_RS_RUNTIME float4 __attribute__((overloadable)) \
+#define FN_FUNC_FN_FN_FN(fnc) \
+_RS_RUNTIME float2 __attribute__((const, overloadable)) \
+ fnc(float2 v1, float2 v2, float2 v3); \
+_RS_RUNTIME float3 __attribute__((const, overloadable)) \
+ fnc(float3 v1, float3 v2, float3 v3); \
+_RS_RUNTIME float4 __attribute__((const, overloadable)) \
fnc(float4 v1, float4 v2, float4 v3);
-#define FN_FUNC_FN_FN_F(fnc) \
-_RS_RUNTIME float2 __attribute__((overloadable)) \
- fnc(float2 v1, float2 v2, float v3); \
-_RS_RUNTIME float3 __attribute__((overloadable)) \
- fnc(float3 v1, float3 v2, float v3); \
-_RS_RUNTIME float4 __attribute__((overloadable)) \
+#define FN_FUNC_FN_FN_F(fnc) \
+_RS_RUNTIME float2 __attribute__((const, overloadable)) \
+ fnc(float2 v1, float2 v2, float v3); \
+_RS_RUNTIME float3 __attribute__((const, overloadable)) \
+ fnc(float3 v1, float3 v2, float v3); \
+_RS_RUNTIME float4 __attribute__((const, overloadable)) \
fnc(float4 v1, float4 v2, float v3);
-#define FN_FUNC_FN_F_F(fnc) \
-_RS_RUNTIME float2 __attribute__((overloadable)) \
- fnc(float2 v1, float v2, float v3); \
-_RS_RUNTIME float3 __attribute__((overloadable)) \
- fnc(float3 v1, float v2, float v3); \
-_RS_RUNTIME float4 __attribute__((overloadable)) \
+#define FN_FUNC_FN_F_F(fnc) \
+_RS_RUNTIME float2 __attribute__((const, overloadable)) \
+ fnc(float2 v1, float v2, float v3); \
+_RS_RUNTIME float3 __attribute__((const, overloadable)) \
+ fnc(float3 v1, float v2, float v3); \
+_RS_RUNTIME float4 __attribute__((const, overloadable)) \
fnc(float4 v1, float v2, float v3);
-#define FN_FUNC_FN_FN_PIN(fnc) \
-_RS_RUNTIME float2 __attribute__((overloadable)) \
- fnc(float2 v1, float2 v2, int2 *v3); \
-_RS_RUNTIME float3 __attribute__((overloadable)) \
- fnc(float3 v1, float3 v2, int3 *v3); \
-_RS_RUNTIME float4 __attribute__((overloadable)) \
+#define FN_FUNC_FN_FN_PIN(fnc) \
+_RS_RUNTIME float2 __attribute__((pure, overloadable)) \
+ fnc(float2 v1, float2 v2, int2 *v3); \
+_RS_RUNTIME float3 __attribute__((pure, overloadable)) \
+ fnc(float3 v1, float3 v2, int3 *v3); \
+_RS_RUNTIME float4 __attribute__((pure, overloadable)) \
fnc(float4 v1, float4 v2, int4 *v3);
+#endif // DOXYGEN
+
/**
* Return the inverse cosine.
*
* Supports float, float2, float3, float4
*/
-extern float __attribute__((overloadable)) acos(float);
+extern float __attribute__((const, overloadable)) acos(float);
FN_FUNC_FN(acos)
/**
@@ -191,7 +212,7 @@
*
* Supports float, float2, float3, float4
*/
-extern float __attribute__((overloadable)) acosh(float);
+extern float __attribute__((const, overloadable)) acosh(float);
FN_FUNC_FN(acosh)
/**
@@ -199,7 +220,7 @@
*
* Supports float, float2, float3, float4
*/
-_RS_RUNTIME float __attribute__((overloadable)) acospi(float v);
+_RS_RUNTIME float __attribute__((const, overloadable)) acospi(float v);
FN_FUNC_FN(acospi)
/**
@@ -207,7 +228,7 @@
*
* Supports float, float2, float3, float4
*/
-extern float __attribute__((overloadable)) asin(float);
+extern float __attribute__((const, overloadable)) asin(float);
FN_FUNC_FN(asin)
/**
@@ -215,7 +236,7 @@
*
* Supports float, float2, float3, float4
*/
-extern float __attribute__((overloadable)) asinh(float);
+extern float __attribute__((const, overloadable)) asinh(float);
FN_FUNC_FN(asinh)
@@ -224,7 +245,7 @@
*
* Supports float, float2, float3, float4
*/
-_RS_RUNTIME float __attribute__((overloadable)) asinpi(float v);
+_RS_RUNTIME float __attribute__((const, overloadable)) asinpi(float v);
FN_FUNC_FN(asinpi)
/**
@@ -232,7 +253,7 @@
*
* Supports float, float2, float3, float4
*/
-extern float __attribute__((overloadable)) atan(float);
+extern float __attribute__((const, overloadable)) atan(float);
FN_FUNC_FN(atan)
/**
@@ -244,7 +265,7 @@
* @param y
* @param x
*/
-extern float __attribute__((overloadable)) atan2(float y, float x);
+extern float __attribute__((const, overloadable)) atan2(float y, float x);
FN_FUNC_FN_FN(atan2)
/**
@@ -252,7 +273,7 @@
*
* Supports float, float2, float3, float4
*/
-extern float __attribute__((overloadable)) atanh(float);
+extern float __attribute__((const, overloadable)) atanh(float);
FN_FUNC_FN(atanh)
/**
@@ -260,7 +281,7 @@
*
* Supports float, float2, float3, float4
*/
-_RS_RUNTIME float __attribute__((overloadable)) atanpi(float v);
+_RS_RUNTIME float __attribute__((const, overloadable)) atanpi(float v);
FN_FUNC_FN(atanpi)
/**
@@ -272,7 +293,7 @@
* @param y
* @param x
*/
-_RS_RUNTIME float __attribute__((overloadable)) atan2pi(float y, float x);
+_RS_RUNTIME float __attribute__((const, overloadable)) atan2pi(float y, float x);
FN_FUNC_FN_FN(atan2pi)
@@ -281,7 +302,7 @@
*
* Supports float, float2, float3, float4.
*/
-extern float __attribute__((overloadable)) cbrt(float);
+extern float __attribute__((const, overloadable)) cbrt(float);
FN_FUNC_FN(cbrt)
/**
@@ -289,7 +310,7 @@
*
* Supports float, float2, float3, float4.
*/
-extern float __attribute__((overloadable)) ceil(float);
+extern float __attribute__((const, overloadable)) ceil(float);
FN_FUNC_FN(ceil)
/**
@@ -301,7 +322,7 @@
* @param x
* @param y
*/
-extern float __attribute__((overloadable)) copysign(float x, float y);
+extern float __attribute__((const, overloadable)) copysign(float x, float y);
FN_FUNC_FN_FN(copysign)
/**
@@ -309,7 +330,7 @@
*
* Supports float, float2, float3, float4.
*/
-extern float __attribute__((overloadable)) cos(float);
+extern float __attribute__((const, overloadable)) cos(float);
FN_FUNC_FN(cos)
/**
@@ -317,7 +338,7 @@
*
* Supports float, float2, float3, float4.
*/
-extern float __attribute__((overloadable)) cosh(float);
+extern float __attribute__((const, overloadable)) cosh(float);
FN_FUNC_FN(cosh)
/**
@@ -325,7 +346,7 @@
*
* Supports float, float2, float3, float4.
*/
-_RS_RUNTIME float __attribute__((overloadable)) cospi(float v);
+_RS_RUNTIME float __attribute__((const, overloadable)) cospi(float v);
FN_FUNC_FN(cospi)
/**
@@ -333,7 +354,7 @@
*
* Supports float, float2, float3, float4.
*/
-extern float __attribute__((overloadable)) erfc(float);
+extern float __attribute__((const, overloadable)) erfc(float);
FN_FUNC_FN(erfc)
/**
@@ -341,7 +362,7 @@
*
* Supports float, float2, float3, float4.
*/
-extern float __attribute__((overloadable)) erf(float);
+extern float __attribute__((const, overloadable)) erf(float);
FN_FUNC_FN(erf)
/**
@@ -349,7 +370,7 @@
*
* Supports float, float2, float3, float4.
*/
-extern float __attribute__((overloadable)) exp(float);
+extern float __attribute__((const, overloadable)) exp(float);
FN_FUNC_FN(exp)
/**
@@ -357,7 +378,7 @@
*
* Supports float, float2, float3, float4.
*/
-extern float __attribute__((overloadable)) exp2(float);
+extern float __attribute__((const, overloadable)) exp2(float);
FN_FUNC_FN(exp2)
/**
@@ -366,7 +387,7 @@
* Supports float, float2, float3, float4. Both arguments must be of the same
* type.
*/
-extern float __attribute__((overloadable)) pow(float x, float y);
+extern float __attribute__((const, overloadable)) pow(float x, float y);
FN_FUNC_FN_FN(pow)
/**
@@ -374,7 +395,7 @@
*
* Supports float, float2, float3, float4.
*/
-_RS_RUNTIME float __attribute__((overloadable)) exp10(float v);
+_RS_RUNTIME float __attribute__((const, overloadable)) exp10(float v);
FN_FUNC_FN(exp10)
/**
@@ -382,7 +403,7 @@
*
* Supports float, float2, float3, float4.
*/
-extern float __attribute__((overloadable)) expm1(float);
+extern float __attribute__((const, overloadable)) expm1(float);
FN_FUNC_FN(expm1)
/**
@@ -390,7 +411,7 @@
*
* Supports float, float2, float3, float4.
*/
-extern float __attribute__((overloadable)) fabs(float);
+extern float __attribute__((const, overloadable)) fabs(float);
FN_FUNC_FN(fabs)
/**
@@ -399,7 +420,7 @@
* Supports float, float2, float3, float4. Both arguments must be of the same
* type.
*/
-extern float __attribute__((overloadable)) fdim(float, float);
+extern float __attribute__((const, overloadable)) fdim(float, float);
FN_FUNC_FN_FN(fdim)
/**
@@ -407,7 +428,7 @@
*
* Supports float, float2, float3, float4.
*/
-extern float __attribute__((overloadable)) floor(float);
+extern float __attribute__((const, overloadable)) floor(float);
FN_FUNC_FN(floor)
/**
@@ -415,7 +436,7 @@
*
* Supports float, float2, float3, float4.
*/
-extern float __attribute__((overloadable)) fma(float a, float b, float c);
+extern float __attribute__((const, overloadable)) fma(float a, float b, float c);
FN_FUNC_FN_FN_FN(fma)
/**
@@ -425,7 +446,7 @@
* @param x: may be float, float2, float3, float4
* @param y: may be float or vector. If vector must match type of x.
*/
-extern float __attribute__((overloadable)) fmax(float x, float y);
+extern float __attribute__((const, overloadable)) fmax(float x, float y);
FN_FUNC_FN_FN(fmax);
FN_FUNC_FN_F(fmax);
@@ -435,7 +456,7 @@
* @param x: may be float, float2, float3, float4
* @param y: may be float or vector. If vector must match type of x.
*/
-extern float __attribute__((overloadable)) fmin(float x, float y);
+extern float __attribute__((const, overloadable)) fmin(float x, float y);
FN_FUNC_FN_FN(fmin);
FN_FUNC_FN_F(fmin);
@@ -444,7 +465,7 @@
*
* Supports float, float2, float3, float4.
*/
-extern float __attribute__((overloadable)) fmod(float x, float y);
+extern float __attribute__((const, overloadable)) fmod(float x, float y);
FN_FUNC_FN_FN(fmod)
/**
@@ -453,7 +474,7 @@
* @param iptr iptr[0] will be set to the floor of the input value.
* Supports float, float2, float3, float4.
*/
-_RS_RUNTIME float __attribute__((overloadable)) fract(float v, float *iptr);
+_RS_RUNTIME float __attribute__((pure, overloadable)) fract(float v, float *iptr);
FN_FUNC_FN_PFN(fract)
/**
@@ -461,22 +482,22 @@
*
* Supports float, float2, float3, float4.
*/
-static inline float __attribute__((overloadable)) fract(float v) {
+static inline float __attribute__((const, overloadable)) fract(float v) {
float unused;
return fract(v, &unused);
}
-static inline float2 __attribute__((overloadable)) fract(float2 v) {
+static inline float2 __attribute__((const, overloadable)) fract(float2 v) {
float2 unused;
return fract(v, &unused);
}
-static inline float3 __attribute__((overloadable)) fract(float3 v) {
+static inline float3 __attribute__((const, overloadable)) fract(float3 v) {
float3 unused;
return fract(v, &unused);
}
-static inline float4 __attribute__((overloadable)) fract(float4 v) {
+static inline float4 __attribute__((const, overloadable)) fract(float4 v) {
float4 unused;
return fract(v, &unused);
}
@@ -487,7 +508,7 @@
* @param v Supports float, float2, float3, float4.
* @param iptr Must have the same vector size as v.
*/
-extern float __attribute__((overloadable)) frexp(float v, int *iptr);
+extern float __attribute__((pure, overloadable)) frexp(float v, int *iptr);
FN_FUNC_FN_PIN(frexp)
/**
@@ -495,7 +516,7 @@
*
* Supports float, float2, float3, float4.
*/
-extern float __attribute__((overloadable)) hypot(float x, float y);
+extern float __attribute__((const, overloadable)) hypot(float x, float y);
FN_FUNC_FN_FN(hypot)
/**
@@ -503,7 +524,7 @@
*
* Supports 1,2,3,4 components
*/
-extern int __attribute__((overloadable)) ilogb(float);
+extern int __attribute__((const, overloadable)) ilogb(float);
IN_FUNC_FN(ilogb)
/**
@@ -512,7 +533,7 @@
* @param x Supports 1,2,3,4 components
* @param y Supports single component or matching vector.
*/
-extern float __attribute__((overloadable)) ldexp(float x, int y);
+extern float __attribute__((const, overloadable)) ldexp(float x, int y);
FN_FUNC_FN_IN(ldexp)
FN_FUNC_FN_I(ldexp)
@@ -521,7 +542,7 @@
*
* Supports 1,2,3,4 components
*/
-extern float __attribute__((overloadable)) lgamma(float);
+extern float __attribute__((const, overloadable)) lgamma(float);
FN_FUNC_FN(lgamma)
/**
@@ -530,7 +551,7 @@
* @param x Supports 1,2,3,4 components
* @param y Supports matching vector.
*/
-extern float __attribute__((overloadable)) lgamma(float x, int* y);
+extern float __attribute__((pure, overloadable)) lgamma(float x, int* y);
FN_FUNC_FN_PIN(lgamma)
/**
@@ -538,7 +559,7 @@
*
* Supports 1,2,3,4 components
*/
-extern float __attribute__((overloadable)) log(float);
+extern float __attribute__((const, overloadable)) log(float);
FN_FUNC_FN(log)
/**
@@ -546,7 +567,7 @@
*
* Supports 1,2,3,4 components
*/
-extern float __attribute__((overloadable)) log10(float);
+extern float __attribute__((const, overloadable)) log10(float);
FN_FUNC_FN(log10)
/**
@@ -554,7 +575,7 @@
*
* Supports 1,2,3,4 components
*/
-_RS_RUNTIME float __attribute__((overloadable)) log2(float v);
+_RS_RUNTIME float __attribute__((const, overloadable)) log2(float v);
FN_FUNC_FN(log2)
/**
@@ -562,7 +583,7 @@
*
* Supports 1,2,3,4 components
*/
-extern float __attribute__((overloadable)) log1p(float v);
+extern float __attribute__((const, overloadable)) log1p(float v);
FN_FUNC_FN(log1p)
/**
@@ -570,7 +591,7 @@
*
* Supports 1,2,3,4 components
*/
-extern float __attribute__((overloadable)) logb(float);
+extern float __attribute__((const, overloadable)) logb(float);
FN_FUNC_FN(logb)
/**
@@ -578,7 +599,7 @@
*
* Supports 1,2,3,4 components
*/
-extern float __attribute__((overloadable)) mad(float a, float b, float c);
+extern float __attribute__((const, overloadable)) mad(float a, float b, float c);
FN_FUNC_FN_FN_FN(mad)
/**
@@ -589,17 +610,17 @@
* @param iret iret[0] will be set to the integral portion of the number.
* @return The floating point portion of the value.
*/
-extern float __attribute__((overloadable)) modf(float x, float *iret);
+extern float __attribute__((pure, overloadable)) modf(float x, float *iret);
FN_FUNC_FN_PFN(modf);
-extern float __attribute__((overloadable)) nan(uint);
+extern float __attribute__((const, overloadable)) nan(uint);
/**
* Return the next floating point number from x towards y.
*
* Supports 1,2,3,4 components
*/
-extern float __attribute__((overloadable)) nextafter(float x, float y);
+extern float __attribute__((const, overloadable)) nextafter(float x, float y);
FN_FUNC_FN_FN(nextafter)
/**
@@ -607,7 +628,7 @@
*
* Supports 1,2,3,4 components
*/
-_RS_RUNTIME float __attribute__((overloadable)) pown(float v, int p);
+_RS_RUNTIME float __attribute__((const, overloadable)) pown(float v, int p);
FN_FUNC_FN_IN(pown)
/**
@@ -616,7 +637,7 @@
*
* Supports 1,2,3,4 components
*/
-_RS_RUNTIME float __attribute__((overloadable)) powr(float v, float p);
+_RS_RUNTIME float __attribute__((const, overloadable)) powr(float v, float p);
FN_FUNC_FN_FN(powr)
/**
@@ -624,11 +645,11 @@
*
* Supports 1,2,3,4 components
*/
-extern float __attribute__((overloadable)) remainder(float x, float y);
+extern float __attribute__((const, overloadable)) remainder(float x, float y);
FN_FUNC_FN_FN(remainder)
// document once we know the precision of bionic
-extern float __attribute__((overloadable)) remquo(float, float, int *);
+extern float __attribute__((pure, overloadable)) remquo(float, float, int *);
FN_FUNC_FN_FN_PIN(remquo)
/**
@@ -636,7 +657,7 @@
*
* Supports 1,2,3,4 components
*/
-extern float __attribute__((overloadable)) rint(float);
+extern float __attribute__((const, overloadable)) rint(float);
FN_FUNC_FN(rint)
/**
@@ -644,7 +665,7 @@
*
* Supports 1,2,3,4 components
*/
-_RS_RUNTIME float __attribute__((overloadable)) rootn(float v, int n);
+_RS_RUNTIME float __attribute__((const, overloadable)) rootn(float v, int n);
FN_FUNC_FN_IN(rootn)
/**
@@ -652,7 +673,7 @@
*
* Supports 1,2,3,4 components
*/
-extern float __attribute__((overloadable)) round(float);
+extern float __attribute__((const, overloadable)) round(float);
FN_FUNC_FN(round)
/**
@@ -660,7 +681,7 @@
*
* Supports 1,2,3,4 components
*/
-extern float __attribute__((overloadable)) sqrt(float);
+extern float __attribute__((const, overloadable)) sqrt(float);
FN_FUNC_FN(sqrt)
/**
@@ -668,7 +689,7 @@
*
* Supports 1,2,3,4 components
*/
-_RS_RUNTIME float __attribute__((overloadable)) rsqrt(float v);
+_RS_RUNTIME float __attribute__((const, overloadable)) rsqrt(float v);
FN_FUNC_FN(rsqrt)
/**
@@ -677,7 +698,7 @@
* @param v The incoming value in radians
* Supports 1,2,3,4 components
*/
-extern float __attribute__((overloadable)) sin(float v);
+extern float __attribute__((const, overloadable)) sin(float v);
FN_FUNC_FN(sin)
/**
@@ -689,7 +710,7 @@
*
* Supports 1,2,3,4 components
*/
-_RS_RUNTIME float __attribute__((overloadable)) sincos(float v, float *cosptr);
+_RS_RUNTIME float __attribute__((pure, overloadable)) sincos(float v, float *cosptr);
FN_FUNC_FN_PFN(sincos);
/**
@@ -697,7 +718,7 @@
*
* Supports 1,2,3,4 components
*/
-extern float __attribute__((overloadable)) sinh(float);
+extern float __attribute__((const, overloadable)) sinh(float);
FN_FUNC_FN(sinh)
/**
@@ -705,7 +726,7 @@
*
* Supports 1,2,3,4 components
*/
-_RS_RUNTIME float __attribute__((overloadable)) sinpi(float v);
+_RS_RUNTIME float __attribute__((const, overloadable)) sinpi(float v);
FN_FUNC_FN(sinpi)
/**
@@ -714,7 +735,7 @@
* Supports 1,2,3,4 components
* @param v The incoming value in radians
*/
-extern float __attribute__((overloadable)) tan(float v);
+extern float __attribute__((const, overloadable)) tan(float v);
FN_FUNC_FN(tan)
/**
@@ -723,7 +744,7 @@
* Supports 1,2,3,4 components
* @param v The incoming value in radians
*/
-extern float __attribute__((overloadable)) tanh(float);
+extern float __attribute__((const, overloadable)) tanh(float);
FN_FUNC_FN(tanh)
/**
@@ -731,7 +752,7 @@
*
* Supports 1,2,3,4 components
*/
-_RS_RUNTIME float __attribute__((overloadable)) tanpi(float v);
+_RS_RUNTIME float __attribute__((const, overloadable)) tanpi(float v);
FN_FUNC_FN(tanpi)
/**
@@ -739,7 +760,7 @@
*
* Supports 1,2,3,4 components
*/
-extern float __attribute__((overloadable)) tgamma(float);
+extern float __attribute__((const, overloadable)) tgamma(float);
FN_FUNC_FN(tgamma)
/**
@@ -747,15 +768,37 @@
*
* Supports 1,2,3,4 components
*/
-extern float __attribute__((overloadable)) trunc(float);
+extern float __attribute__((const, overloadable)) trunc(float);
FN_FUNC_FN(trunc)
+#ifdef DOXYGEN
#define XN_FUNC_YN(typeout, fnc, typein) \
-extern typeout __attribute__((overloadable)) fnc(typein); \
-_RS_RUNTIME typeout##2 __attribute__((overloadable)) fnc(typein##2 v); \
-_RS_RUNTIME typeout##3 __attribute__((overloadable)) fnc(typein##3 v); \
-_RS_RUNTIME typeout##4 __attribute__((overloadable)) fnc(typein##4 v);
+extern typeout __attribute__((overloadable)) fnc(typein v);
+
+#define XN_FUNC_XN_XN_BODY(type, fnc, body) \
+_RS_RUNTIME type __attribute__((overloadable)) \
+ fnc(type v1, type v2);
+
+#else
+
+#define XN_FUNC_YN(typeout, fnc, typein) \
+extern typeout __attribute__((const, overloadable)) fnc(typein v); \
+_RS_RUNTIME typeout##2 __attribute__((const, overloadable)) fnc(typein##2 v); \
+_RS_RUNTIME typeout##3 __attribute__((const, overloadable)) fnc(typein##3 v); \
+_RS_RUNTIME typeout##4 __attribute__((const, overloadable)) fnc(typein##4 v);
+
+#define XN_FUNC_XN_XN_BODY(type, fnc, body) \
+_RS_RUNTIME type __attribute__((const, overloadable)) \
+ fnc(type v1, type v2); \
+_RS_RUNTIME type##2 __attribute__((const, overloadable)) \
+ fnc(type##2 v1, type##2 v2); \
+_RS_RUNTIME type##3 __attribute__((const, overloadable)) \
+ fnc(type##3 v1, type##3 v2); \
+_RS_RUNTIME type##4 __attribute__((const, overloadable)) \
+ fnc(type##4 v1, type##4 v2);
+
+#endif // DOXYGEN
#define UIN_FUNC_IN(fnc) \
XN_FUNC_YN(uchar, fnc, char) \
@@ -770,17 +813,6 @@
XN_FUNC_YN(uint, fnc, uint) \
XN_FUNC_YN(int, fnc, int)
-
-#define XN_FUNC_XN_XN_BODY(type, fnc, body) \
-_RS_RUNTIME type __attribute__((overloadable)) \
- fnc(type v1, type v2); \
-_RS_RUNTIME type##2 __attribute__((overloadable)) \
- fnc(type##2 v1, type##2 v2); \
-_RS_RUNTIME type##3 __attribute__((overloadable)) \
- fnc(type##3 v1, type##3 v2); \
-_RS_RUNTIME type##4 __attribute__((overloadable)) \
- fnc(type##4 v1, type##4 v2);
-
#define IN_FUNC_IN_IN_BODY(fnc, body) \
XN_FUNC_XN_XN_BODY(uchar, fnc, body) \
XN_FUNC_XN_XN_BODY(char, fnc, body) \
@@ -791,6 +823,7 @@
XN_FUNC_XN_XN_BODY(float, fnc, body)
/**
+ * \fn uchar abs(char)
* Return the absolute value of a value.
*
* Supports 1,2,3,4 components of char, short, int.
@@ -827,16 +860,41 @@
* @param low Lower bound, must be scalar or matching vector.
* @param high High bound, must match type of low
*/
-_RS_RUNTIME float __attribute__((overloadable)) clamp(float amount, float low, float high);
+
+#if !defined(RS_VERSION) || (RS_VERSION < 19)
+_RS_RUNTIME float __attribute__((const, overloadable)) clamp(float amount, float low, float high);
FN_FUNC_FN_FN_FN(clamp)
FN_FUNC_FN_F_F(clamp)
+#else
+#define _CLAMP(T) \
+extern T __attribute__((overloadable)) clamp(T amount, T low, T high); \
+extern T##2 __attribute__((overloadable)) clamp(T##2 amount, T##2 low, T##2 high); \
+extern T##3 __attribute__((overloadable)) clamp(T##3 amount, T##3 low, T##3 high); \
+extern T##4 __attribute__((overloadable)) clamp(T##4 amount, T##4 low, T##4 high); \
+extern T##2 __attribute__((overloadable)) clamp(T##2 amount, T low, T high); \
+extern T##3 __attribute__((overloadable)) clamp(T##3 amount, T low, T high); \
+extern T##4 __attribute__((overloadable)) clamp(T##4 amount, T low, T high)
+
+_CLAMP(float);
+_CLAMP(double);
+_CLAMP(char);
+_CLAMP(uchar);
+_CLAMP(short);
+_CLAMP(ushort);
+_CLAMP(int);
+_CLAMP(uint);
+_CLAMP(long);
+_CLAMP(ulong);
+
+#undef _CLAMP
+#endif
/**
* Convert from radians to degrees.
*
* Supports 1,2,3,4 components
*/
-_RS_RUNTIME float __attribute__((overloadable)) degrees(float radians);
+_RS_RUNTIME float __attribute__((const, overloadable)) degrees(float radians);
FN_FUNC_FN(degrees)
/**
@@ -844,7 +902,7 @@
*
* Supports 1,2,3,4 components
*/
-_RS_RUNTIME float __attribute__((overloadable)) mix(float start, float stop, float amount);
+_RS_RUNTIME float __attribute__((const, overloadable)) mix(float start, float stop, float amount);
FN_FUNC_FN_FN_FN(mix)
FN_FUNC_FN_FN_F(mix)
@@ -853,7 +911,7 @@
*
* Supports 1,2,3,4 components
*/
-_RS_RUNTIME float __attribute__((overloadable)) radians(float degrees);
+_RS_RUNTIME float __attribute__((const, overloadable)) radians(float degrees);
FN_FUNC_FN(radians)
/**
@@ -864,18 +922,18 @@
*
* Supports 1,2,3,4 components
*/
-_RS_RUNTIME float __attribute__((overloadable)) step(float edge, float v);
+_RS_RUNTIME float __attribute__((const, overloadable)) step(float edge, float v);
FN_FUNC_FN_FN(step)
FN_FUNC_FN_F(step)
// not implemented
-extern float __attribute__((overloadable)) smoothstep(float, float, float);
-extern float2 __attribute__((overloadable)) smoothstep(float2, float2, float2);
-extern float3 __attribute__((overloadable)) smoothstep(float3, float3, float3);
-extern float4 __attribute__((overloadable)) smoothstep(float4, float4, float4);
-extern float2 __attribute__((overloadable)) smoothstep(float, float, float2);
-extern float3 __attribute__((overloadable)) smoothstep(float, float, float3);
-extern float4 __attribute__((overloadable)) smoothstep(float, float, float4);
+extern float __attribute__((const, overloadable)) smoothstep(float, float, float);
+extern float2 __attribute__((const, overloadable)) smoothstep(float2, float2, float2);
+extern float3 __attribute__((const, overloadable)) smoothstep(float3, float3, float3);
+extern float4 __attribute__((const, overloadable)) smoothstep(float4, float4, float4);
+extern float2 __attribute__((const, overloadable)) smoothstep(float, float, float2);
+extern float3 __attribute__((const, overloadable)) smoothstep(float, float, float3);
+extern float4 __attribute__((const, overloadable)) smoothstep(float, float, float4);
/**
* Return the sign of a value.
@@ -886,7 +944,7 @@
*
* Supports 1,2,3,4 components
*/
-_RS_RUNTIME float __attribute__((overloadable)) sign(float v);
+_RS_RUNTIME float __attribute__((const, overloadable)) sign(float v);
FN_FUNC_FN(sign)
/**
@@ -894,15 +952,15 @@
*
* Supports 3,4 components
*/
-_RS_RUNTIME float3 __attribute__((overloadable)) cross(float3 lhs, float3 rhs);
-_RS_RUNTIME float4 __attribute__((overloadable)) cross(float4 lhs, float4 rhs);
+_RS_RUNTIME float3 __attribute__((const, overloadable)) cross(float3 lhs, float3 rhs);
+_RS_RUNTIME float4 __attribute__((const, overloadable)) cross(float4 lhs, float4 rhs);
/**
* Compute the dot product of two vectors.
*
* Supports 1,2,3,4 components
*/
-_RS_RUNTIME float __attribute__((overloadable)) dot(float lhs, float rhs);
+_RS_RUNTIME float __attribute__((const, overloadable)) dot(float lhs, float rhs);
F_FUNC_FN_FN(dot)
/**
@@ -910,7 +968,7 @@
*
* Supports 1,2,3,4 components
*/
-_RS_RUNTIME float __attribute__((overloadable)) length(float v);
+_RS_RUNTIME float __attribute__((const, overloadable)) length(float v);
F_FUNC_FN(length)
/**
@@ -918,7 +976,7 @@
*
* Supports 1,2,3,4 components
*/
-_RS_RUNTIME float __attribute__((overloadable)) distance(float lhs, float rhs);
+_RS_RUNTIME float __attribute__((const, overloadable)) distance(float lhs, float rhs);
F_FUNC_FN_FN(distance)
/**
@@ -926,7 +984,7 @@
*
* Supports 1,2,3,4 components
*/
-_RS_RUNTIME float __attribute__((overloadable)) normalize(float v);
+_RS_RUNTIME float __attribute__((const, overloadable)) normalize(float v);
FN_FUNC_FN(normalize)
@@ -938,7 +996,7 @@
*
* Supports 1,2,3,4 components
*/
-_RS_RUNTIME float __attribute__((overloadable)) half_recip(float);
+_RS_RUNTIME float __attribute__((const, overloadable)) half_recip(float);
FN_FUNC_FN(half_recip)
/**
@@ -946,7 +1004,7 @@
*
* Supports 1,2,3,4 components
*/
-_RS_RUNTIME float __attribute__((overloadable)) half_sqrt(float);
+_RS_RUNTIME float __attribute__((const, overloadable)) half_sqrt(float);
FN_FUNC_FN(half_sqrt)
/**
@@ -954,7 +1012,7 @@
*
* Supports 1,2,3,4 components
*/
-_RS_RUNTIME float __attribute__((overloadable)) half_rsqrt(float v);
+_RS_RUNTIME float __attribute__((const, overloadable)) half_rsqrt(float v);
FN_FUNC_FN(half_rsqrt)
/**
@@ -962,7 +1020,7 @@
*
* Supports 1,2,3,4 components
*/
-_RS_RUNTIME float __attribute__((overloadable)) fast_length(float v);
+_RS_RUNTIME float __attribute__((const, overloadable)) fast_length(float v);
F_FUNC_FN(fast_length)
/**
@@ -970,7 +1028,7 @@
*
* Supports 1,2,3,4 components
*/
-_RS_RUNTIME float __attribute__((overloadable)) fast_distance(float lhs, float rhs);
+_RS_RUNTIME float __attribute__((const, overloadable)) fast_distance(float lhs, float rhs);
F_FUNC_FN_FN(fast_distance)
/**
@@ -978,7 +1036,7 @@
*
* Supports 1,2,3,4 components
*/
-_RS_RUNTIME float __attribute__((overloadable)) fast_normalize(float v);
+_RS_RUNTIME float __attribute__((const, overloadable)) fast_normalize(float v);
F_FUNC_FN(fast_normalize)
#endif // (defined(RS_VERSION) && (RS_VERSION >= 17))
@@ -996,7 +1054,7 @@
*
* Supports 1,2,3,4 components
*/
-_RS_RUNTIME float __attribute__((overloadable)) native_exp2(float v);
+_RS_RUNTIME float __attribute__((const, overloadable)) native_exp2(float v);
FN_FUNC_FN(native_exp2)
/**
@@ -1006,7 +1064,7 @@
*
* Supports 1,2,3,4 components
*/
-_RS_RUNTIME float __attribute__((overloadable)) native_exp(float v);
+_RS_RUNTIME float __attribute__((const, overloadable)) native_exp(float v);
FN_FUNC_FN(native_exp)
/**
@@ -1016,21 +1074,21 @@
*
* Supports 1,2,3,4 components
*/
-_RS_RUNTIME float __attribute__((overloadable)) native_exp10(float v);
+_RS_RUNTIME float __attribute__((const, overloadable)) native_exp10(float v);
FN_FUNC_FN(native_exp10)
-_RS_RUNTIME float __attribute__((overloadable)) native_log2(float v);
+_RS_RUNTIME float __attribute__((const, overloadable)) native_log2(float v);
FN_FUNC_FN(native_log2)
-_RS_RUNTIME float __attribute__((overloadable)) native_log(float v);
+_RS_RUNTIME float __attribute__((const, overloadable)) native_log(float v);
FN_FUNC_FN(native_log)
-_RS_RUNTIME float __attribute__((overloadable)) native_log10(float v);
+_RS_RUNTIME float __attribute__((const, overloadable)) native_log10(float v);
FN_FUNC_FN(native_log10)
-_RS_RUNTIME float __attribute__((overloadable)) native_powr(float v, float y);
+_RS_RUNTIME float __attribute__((const, overloadable)) native_powr(float v, float y);
FN_FUNC_FN_FN(native_powr)
diff --git a/scriptc/rs_math.rsh b/scriptc/rs_math.rsh
index 73040b3..4d3124c 100644
--- a/scriptc/rs_math.rsh
+++ b/scriptc/rs_math.rsh
@@ -49,7 +49,7 @@
/**
* Returns the fractional part of a float
*/
-extern float __attribute__((overloadable))
+extern float __attribute__((const, overloadable))
rsFrac(float);
@@ -64,28 +64,28 @@
* @param low
* @param high
*/
-_RS_RUNTIME uint __attribute__((overloadable, always_inline)) rsClamp(uint amount, uint low, uint high);
+_RS_RUNTIME uint __attribute__((const, overloadable, always_inline)) rsClamp(uint amount, uint low, uint high);
/**
* \overload
*/
-_RS_RUNTIME int __attribute__((overloadable, always_inline)) rsClamp(int amount, int low, int high);
+_RS_RUNTIME int __attribute__((const, overloadable, always_inline)) rsClamp(int amount, int low, int high);
/**
* \overload
*/
-_RS_RUNTIME ushort __attribute__((overloadable, always_inline)) rsClamp(ushort amount, ushort low, ushort high);
+_RS_RUNTIME ushort __attribute__((const, overloadable, always_inline)) rsClamp(ushort amount, ushort low, ushort high);
/**
* \overload
*/
-_RS_RUNTIME short __attribute__((overloadable, always_inline)) rsClamp(short amount, short low, short high);
+_RS_RUNTIME short __attribute__((const, overloadable, always_inline)) rsClamp(short amount, short low, short high);
/**
* \overload
*/
-_RS_RUNTIME uchar __attribute__((overloadable, always_inline)) rsClamp(uchar amount, uchar low, uchar high);
+_RS_RUNTIME uchar __attribute__((const, overloadable, always_inline)) rsClamp(uchar amount, uchar low, uchar high);
/**
* \overload
*/
-_RS_RUNTIME char __attribute__((overloadable, always_inline)) rsClamp(char amount, char low, char high);
+_RS_RUNTIME char __attribute__((const, overloadable, always_inline)) rsClamp(char amount, char low, char high);
/**
@@ -202,7 +202,7 @@
*
* @return uchar4
*/
-_RS_RUNTIME uchar4 __attribute__((overloadable)) rsPackColorTo8888(float r, float g, float b);
+_RS_RUNTIME uchar4 __attribute__((const, overloadable)) rsPackColorTo8888(float r, float g, float b);
/**
* Pack floating point (0-1) RGBA values into a uchar4.
@@ -214,7 +214,7 @@
*
* @return uchar4
*/
-_RS_RUNTIME uchar4 __attribute__((overloadable)) rsPackColorTo8888(float r, float g, float b, float a);
+_RS_RUNTIME uchar4 __attribute__((const, overloadable)) rsPackColorTo8888(float r, float g, float b, float a);
/**
* Pack floating point (0-1) RGB values into a uchar4. The alpha component is
@@ -224,7 +224,7 @@
*
* @return uchar4
*/
-_RS_RUNTIME uchar4 __attribute__((overloadable)) rsPackColorTo8888(float3 color);
+_RS_RUNTIME uchar4 __attribute__((const, overloadable)) rsPackColorTo8888(float3 color);
/**
* Pack floating point (0-1) RGBA values into a uchar4.
@@ -233,7 +233,7 @@
*
* @return uchar4
*/
-_RS_RUNTIME uchar4 __attribute__((overloadable)) rsPackColorTo8888(float4 color);
+_RS_RUNTIME uchar4 __attribute__((const, overloadable)) rsPackColorTo8888(float4 color);
/**
* Unpack a uchar4 color to float4. The resulting float range will be (0-1).
@@ -242,10 +242,10 @@
*
* @return float4
*/
-_RS_RUNTIME float4 rsUnpackColor8888(uchar4 c);
+_RS_RUNTIME float4 __attribute__((const)) rsUnpackColor8888(uchar4 c);
-_RS_RUNTIME uchar4 __attribute__((overloadable)) rsYuvToRGBA_uchar4(uchar y, uchar u, uchar v);
-_RS_RUNTIME float4 __attribute__((overloadable)) rsYuvToRGBA_float4(uchar y, uchar u, uchar v);
+_RS_RUNTIME uchar4 __attribute__((const, overloadable)) rsYuvToRGBA_uchar4(uchar y, uchar u, uchar v);
+_RS_RUNTIME float4 __attribute__((const, overloadable)) rsYuvToRGBA_float4(uchar y, uchar u, uchar v);
#endif
diff --git a/tests/cppbasic/Android.mk b/tests/cppbasic/Android.mk
index 5c2c53a..bf7725a 100644
--- a/tests/cppbasic/Android.mk
+++ b/tests/cppbasic/Android.mk
@@ -6,18 +6,18 @@
compute.cpp
LOCAL_SHARED_LIBRARIES := \
- libRS \
- libRScpp \
libz \
- libcutils \
- libutils \
libEGL \
libGLESv1_CM \
libGLESv2 \
libui \
libbcc \
libbcinfo \
- libgui
+ libgui \
+ libdl \
+ libRScpp \
+ libstlport
+
LOCAL_MODULE:= rstest-compute
@@ -25,6 +25,7 @@
intermediates := $(call intermediates-dir-for,STATIC_LIBRARIES,libRS,TARGET,)
+LOCAL_C_INCLUDES += external/stlport/stlport bionic/ bionic/libstdc++/include
LOCAL_C_INCLUDES += frameworks/rs/cpp
LOCAL_C_INCLUDES += frameworks/rs
LOCAL_C_INCLUDES += $(intermediates)
diff --git a/tests/cppbasic/compute.cpp b/tests/cppbasic/compute.cpp
index d1647c4..bc80c7e 100644
--- a/tests/cppbasic/compute.cpp
+++ b/tests/cppbasic/compute.cpp
@@ -32,7 +32,7 @@
sp<Allocation> aout = Allocation::createTyped(rs, t);
printf("Allocation %p %p\n", ain.get(), aout.get());
- sp<ScriptC_mono> sc = new ScriptC_mono(rs, NULL, 0);
+ ScriptC_mono* sc = new ScriptC_mono(rs);
printf("new script\n");
// We read back the status from the script-side via a "failed" allocation.
@@ -66,8 +66,8 @@
{
sp<const Element> e = Element::I32(rs);
Type::Builder tb(rs, e);
- tb.setX(5);
- tb.setY(5);
+ tb.setX(8);
+ tb.setY(8);
sp<const Type> t = tb.create();
sp<Allocation> kern1_in = Allocation::createTyped(rs, t);
sp<Allocation> kern1_out = Allocation::createTyped(rs, t);
@@ -84,21 +84,10 @@
rs->finish();
failed_alloc->copy1DTo(&failed);
-
- e.clear();
- t.clear();
- kern1_in.clear();
- kern1_out.clear();
}
printf("Deleting stuff\n");
- sc.clear();
- t.clear();
- a1.clear();
- e.clear();
- ain.clear();
- aout.clear();
- // delete rs;
+ delete sc;
printf("Delete OK\n");
if (failed) {
diff --git a/tests/cppbasic/mono.rs b/tests/cppbasic/mono.rs
index d49d0f7..f134b36 100644
--- a/tests/cppbasic/mono.rs
+++ b/tests/cppbasic/mono.rs
@@ -15,7 +15,7 @@
*/
#pragma version(1)
-#pragma rs java_package_name(com.android.rs.image)
+#pragma rs java_package_name(com.android.rs.cppbasic)
#pragma rs_fp_relaxed
int g_i = 4;
