diff --git a/Makefile.am b/Makefile.am
index 114dd7d..85dcd84 100644
--- a/Makefile.am
+++ b/Makefile.am
@@ -13,3 +13,20 @@
 	jidctred.c jquant1.c jquant2.c jutils.c jmemmgr.c jmemnobs.c \
 	jsimd.c
 
+if WITH_SIMD
+
+BUILT_SOURCES = simd/jsimdcfg.inc
+
+EXTRA_DIST = nasm_lt.sh
+
+libjpeg_la_SOURCES += simd/jsimd.h simd/jsimdcfg.inc.h \
+	simd/jsimdext.inc simd/jsimdcpu.asm
+
+endif
+
+.asm.lo:
+	$(LIBTOOL) --mode=compile --tag NASM ./nasm_lt.sh $(NASM) $(NAFLAGS) $< -o $@
+
+simd/jsimdcfg.inc: simd/jsimdcfg.inc.h jpeglib.h jconfig.h jmorecfg.h
+	$(CPP) $< | grep ^[\;%] | sed 's%_cpp_protection_%%' > $@
+
diff --git a/acinclude.m4 b/acinclude.m4
new file mode 100644
index 0000000..9150799
--- /dev/null
+++ b/acinclude.m4
@@ -0,0 +1,99 @@
+# AC_PROG_NASM
+# --------------------------
+# Check that NASM exists and determine flags
+AC_DEFUN([AC_PROG_NASM],[
+
+AC_CHECK_PROGS(NASM, [nasm nasmw])
+test -z "$NASM" && AC_MSG_ERROR([no nasm (Netwide Assembler) found])
+
+AC_MSG_CHECKING([for object file format of host system])
+case "$host_os" in
+  cygwin* | mingw* | pw32* | interix*)
+    objfmt='Win32-COFF'
+  ;;
+  msdosdjgpp* | go32*)
+    objfmt='COFF'
+  ;;
+  os2-emx*)			# not tested
+    objfmt='MSOMF'		# obj
+  ;;
+  linux*coff* | linux*oldld*)
+    objfmt='COFF'		# ???
+  ;;
+  linux*aout*)
+    objfmt='a.out'
+  ;;
+  linux*)
+    objfmt='ELF'
+  ;;
+  freebsd* | netbsd* | openbsd*)
+    if echo __ELF__ | $CC -E - | grep __ELF__ > /dev/null; then
+      objfmt='BSD-a.out'
+    else
+      objfmt='ELF'
+    fi
+  ;;
+  solaris* | sunos* | sysv* | sco*)
+    objfmt='ELF'
+  ;;
+  darwin* | rhapsody* | nextstep* | openstep* | macos*)
+    objfmt='Mach-O'
+  ;;
+  *)
+    objfmt='ELF ?'
+  ;;
+esac
+
+AC_MSG_RESULT([$objfmt])
+if test "$objfmt" = 'ELF ?'; then
+  objfmt='ELF'
+  AC_MSG_WARN([unexpected host system. assumed that the format is $objfmt.])
+fi
+
+AC_MSG_CHECKING([for object file format specifier (NAFLAGS) ])
+case "$objfmt" in
+  MSOMF)      NAFLAGS='-fobj -DOBJ32';;
+  Win32-COFF) NAFLAGS='-fwin32 -DWIN32';;
+  COFF)       NAFLAGS='-fcoff -DCOFF';;
+  a.out)      NAFLAGS='-faout -DAOUT';;
+  BSD-a.out)  NAFLAGS='-faoutb -DAOUT';;
+  ELF)        NAFLAGS='-felf -DELF';;
+  RDF)        NAFLAGS='-frdf -DRDF';;
+  Mach-O)     NAFLAGS='-fmacho -DMACHO';;
+esac
+AC_MSG_RESULT([$NAFLAGS])
+AC_SUBST([NAFLAGS])
+
+AC_MSG_CHECKING([whether the assembler ($NASM $NAFLAGS) works])
+cat > conftest.asm <<EOF
+[%line __oline__ "configure"
+        section .text
+        bits    32
+        global  _main,main
+_main:
+main:   xor     eax,eax
+        ret
+]EOF
+try_nasm='$NASM $NAFLAGS -o conftest.o conftest.asm'
+if AC_TRY_EVAL(try_nasm) && test -s conftest.o; then
+  AC_MSG_RESULT(yes)
+else
+  echo "configure: failed program was:" >&AC_FD_CC
+  cat conftest.asm >&AC_FD_CC
+  rm -rf conftest*
+  AC_MSG_RESULT(no)
+  AC_MSG_ERROR([installation or configuration problem: assembler cannot create object files.])
+fi
+
+AC_MSG_CHECKING([whether the linker accepts assembler output])
+try_nasm='${CC-cc} -o conftest${ac_exeext} $LDFLAGS conftest.o $LIBS 1>&AC_FD_CC'
+if AC_TRY_EVAL(try_nasm) && test -s conftest${ac_exeext}; then
+  rm -rf conftest*
+  AC_MSG_RESULT(yes)
+else
+  rm -rf conftest*
+  AC_MSG_RESULT(no)
+  AC_MSG_ERROR([configuration problem: maybe object file format mismatch.])
+fi
+
+])
diff --git a/configure.ac b/configure.ac
index c2263f2..683ca7e 100644
--- a/configure.ac
+++ b/configure.ac
@@ -12,6 +12,7 @@
 AC_DEFINE([INCOMPLETE_TYPES_BROKEN], 1, [Define if you want use complete types])
 
 # Checks for programs.
+AC_PROG_CPP
 AC_PROG_CC
 AC_PROG_INSTALL
 AC_PROG_LIBTOOL
@@ -72,13 +73,33 @@
 	[AC_DEFINE([NEED_BSD_STRINGS], 1,
 		   [Define if you have BSD-like bzero and bcopy])])
 
+# Set flags to indicate platform
+case "$host_os" in
+  cygwin* | mingw* | pw32* | interix*)
+    is_win32=1
+  ;;
+esac
+AM_CONDITIONAL([IS_WIN32], [test "x$is_win32" = "x1"])
+
 # SIMD is optional
-# todo: detect automatically when it can be enabled
 AC_ARG_WITH([simd],
-    AC_HELP_STRING([--with-simd],[Include accelerated SIMD routines.]))
+    AC_HELP_STRING([--without-simd],[Omit accelerated SIMD routines.]))
 if test "x${with_simd}" != "xno"; then
-	AC_DEFINE([WITH_SIMD], [1], [Use accelerated SIMD routines.])
+  # Check if we're on a supported CPU
+  AC_MSG_CHECKING([if host cpu type is i386 or compatible])
+  case "$host_cpu" in
+    i*86 | x86 | ia32)
+      AC_MSG_RESULT(yes)
+      AC_PROG_NASM
+      AC_DEFINE([WITH_SIMD], [1], [Use accelerated SIMD routines.])
+    ;;
+    *)
+      AC_MSG_RESULT([no ("$host_cpu")])
+      with_simd=no
+    ;;
+  esac
 fi
+AM_CONDITIONAL([WITH_SIMD], [test "x$with_simd" != "xno"])
 
 # jconfig.h is the file we use, but we have another before that to
 # fool autoheader. the reason is that we include this header in our
diff --git a/jsimd.c b/jsimd.c
index c6378e8..c92ae60 100644
--- a/jsimd.c
+++ b/jsimd.c
@@ -13,9 +13,10 @@
 #define JPEG_INTERNALS
 #include "jinclude.h"
 #include "jpeglib.h"
+#include "jsimd.h"
 #include "jdct.h"
-
-#define JSIMD_NONE    0x00
+#include "jsimddct.h"
+#include "simd/jsimd.h"
 
 static unsigned int simd_support = ~0;
 
@@ -30,7 +31,11 @@
   if (simd_support != ~0)
     return;
 
+#ifdef WITH_SIMD
+  simd_support = jpeg_simd_cpu_support();
+#else
   simd_support = JSIMD_NONE;
+#endif
 }
 
 GLOBAL(int)
diff --git a/nasm_lt.sh b/nasm_lt.sh
new file mode 100755
index 0000000..ef5a591
--- /dev/null
+++ b/nasm_lt.sh
@@ -0,0 +1,57 @@
+#! /bin/sh
+command=""
+infile=""
+o_opt=no
+pic=no
+while [ $# -gt 0 ]; do
+    case "$1" in
+        -DPIC|-fPIC|-fpic)
+            if [ "$pic" != "yes" ] ; then
+                command="$command -DPIC"
+                pic=yes
+            fi
+            ;;
+        -f|-fbin|-faout|-faoutb|-fcoff|-felf|-fas86| \
+        -fobj|-fwin32|-frdf|-fieee|-fmacho)
+            # it's a file format specifier for nasm.
+            command="$command $1"
+            ;;
+        -f*)
+            # maybe a code-generation flag for gcc.
+            ;;
+        -[Ii]*)
+            incdir=`echo "$1" | sed 's/^-[Ii]//'`
+            if [ "x$incdir" = x -a "x$2" != x ] ; then
+                case "$2" in
+                    -*) ;;
+                    *) incdir="$2"; shift;;
+                esac
+            fi
+            if [ "x$incdir" != x ] ; then
+                # In the case of NASM, the trailing slash is necessary.
+                incdir=`echo "$incdir" | sed 's%/*$%/%'`
+                command="$command -I$incdir"
+            fi
+            ;;
+        -o*)
+            o_opt=yes
+            command="$command $1"
+            ;;
+        *.asm)
+            infile=$1
+            command="$command $1"
+            ;;
+        *)
+            command="$command $1"
+            ;;
+    esac
+    shift
+done
+if [ "$o_opt" != yes ] ; then
+    # By default, NASM creates an output file
+    # in the same directory as the input file.
+    outfile="-o `echo $infile | sed -e 's%^.*/%%' -e 's%\.[^.]*$%%'`.o"
+    command="$command $outfile"
+fi
+echo $command
+exec $command
diff --git a/simd/jsimd.h b/simd/jsimd.h
new file mode 100644
index 0000000..09ec46d
--- /dev/null
+++ b/simd/jsimd.h
@@ -0,0 +1,23 @@
+/*
+ * simd/jsimd.h
+ *
+ * Copyright 2009 Pierre Ossman <ossman@cendio.se> for Cendio AB
+ * 
+ * Based on the x86 SIMD extension for IJG JPEG library,
+ * Copyright (C) 1999-2006, MIYASAKA Masaru.
+ *
+ */
+
+/* Bitmask for supported acceleration methods */
+
+#define JSIMD_NONE    0x00
+
+/* Short forms of external names for systems with brain-damaged linkers. */
+
+#ifdef NEED_SHORT_EXTERNAL_NAMES
+#define jpeg_simd_cpu_support	jSiCpuSupport
+#endif /* NEED_SHORT_EXTERNAL_NAMES */
+
+/* SIMD Ext: retrieve SIMD/CPU information */
+EXTERN(unsigned int) jpeg_simd_cpu_support JPP((void));
+
diff --git a/simd/jsimdcfg.inc.h b/simd/jsimdcfg.inc.h
new file mode 100644
index 0000000..bc2ff9f
--- /dev/null
+++ b/simd/jsimdcfg.inc.h
@@ -0,0 +1,29 @@
+// This file generates the include file for the assembly
+// implementations by abusing the C preprocessor.
+
+;
+; Automatically generated include file from jsimdcfg.inc.h
+;
+
+#define JPEG_INTERNALS
+
+#include "../jpeglib.h"
+#include "../jconfig.h"
+#include "../jmorecfg.h"
+#include "jsimd.h"
+
+#define define(var) %define _cpp_protection_##var
+#define definev(var) %define _cpp_protection_##var var
+
+;
+; -- jsimd.h
+;
+
+definev(JSIMD_NONE)
+
+; Short forms of external names for systems with brain-damaged linkers.
+;
+#ifdef NEED_SHORT_EXTERNAL_NAMES
+definev(jpeg_simd_cpu_support)
+#endif /* NEED_SHORT_EXTERNAL_NAMES */
+
diff --git a/simd/jsimdcpu.asm b/simd/jsimdcpu.asm
new file mode 100644
index 0000000..6bb4e32
--- /dev/null
+++ b/simd/jsimdcpu.asm
@@ -0,0 +1,63 @@
+;
+; jsimdcpu.asm - SIMD instruction support check
+;
+; Copyright 2009 Pierre Ossman <ossman@cendio.se> for Cendio AB
+;
+; Based on
+; x86 SIMD extension for IJG JPEG library
+; Copyright (C) 1999-2006, MIYASAKA Masaru.
+; For conditions of distribution and use, see copyright notice in jsimdext.inc
+;
+; This file should be assembled with NASM (Netwide Assembler),
+; can *not* be assembled with Microsoft's MASM or any compatible
+; assembler (including Borland's Turbo Assembler).
+; NASM is available from http://nasm.sourceforge.net/ or
+; http://sourceforge.net/project/showfiles.php?group_id=6208
+;
+; [TAB8]
+
+%include "simd/jsimdext.inc"
+
+; --------------------------------------------------------------------------
+	SECTION	SEG_TEXT
+	BITS	32
+;
+; Check if the CPU supports SIMD instructions
+;
+; GLOBAL(unsigned int)
+; jpeg_simd_cpu_support (void)
+;
+
+	align	16
+	global	EXTN(jpeg_simd_cpu_support)
+
+EXTN(jpeg_simd_cpu_support):
+	push	ebx
+;	push	ecx		; need not be preserved
+;	push	edx		; need not be preserved
+;	push	esi		; unused
+	push	edi
+
+	xor	edi,edi			; simd support flag
+
+	pushfd
+	pop	eax
+	mov	edx,eax
+	xor	eax, 1<<21		; flip ID bit in EFLAGS
+	push	eax
+	popfd
+	pushfd
+	pop	eax
+	xor	eax,edx
+	jz	short .return		; CPUID is not supported
+
+.return:
+	mov	eax,edi
+
+	pop	edi
+;	pop	esi		; unused
+;	pop	edx		; need not be preserved
+;	pop	ecx		; need not be preserved
+	pop	ebx
+	ret
+
diff --git a/simd/jsimdext.inc b/simd/jsimdext.inc
new file mode 100644
index 0000000..c0bd54c
--- /dev/null
+++ b/simd/jsimdext.inc
@@ -0,0 +1,243 @@
+;
+; jsimdext.inc - common declarations
+;
+; Copyright 2009 Pierre Ossman <ossman@cendio.se> for Cendio AB
+;
+; Based on
+; x86 SIMD extension for IJG JPEG library - version 1.02
+;
+; Copyright (C) 1999-2006, MIYASAKA Masaru.
+;
+; This software is provided 'as-is', without any express or implied
+; warranty.  In no event will the authors be held liable for any damages
+; arising from the use of this software.
+;
+; Permission is granted to anyone to use this software for any purpose,
+; including commercial applications, and to alter it and redistribute it
+; freely, subject to the following restrictions:
+;
+; 1. The origin of this software must not be misrepresented; you must not
+;    claim that you wrote the original software. If you use this software
+;    in a product, an acknowledgment in the product documentation would be
+;    appreciated but is not required.
+; 2. Altered source versions must be plainly marked as such, and must not be
+;    misrepresented as being the original software.
+; 3. This notice may not be removed or altered from any source distribution.
+;
+; [TAB8]
+
+; ==========================================================================
+;  System-dependent configurations
+
+%ifdef WIN32	; ----(nasm -fwin32 -DWIN32 ...)--------
+; * Microsoft Visual C++
+; * MinGW (Minimalist GNU for Windows)
+; * CygWin
+; * LCC-Win32
+
+; -- segment definition --
+;
+%define SEG_TEXT    .text  align=16 public use32 class=CODE
+%define SEG_CONST   .rdata align=16 public use32 class=CONST
+
+%elifdef OBJ32	; ----(nasm -fobj -DOBJ32 ...)----------
+; * Borland C++ (Win32)
+
+; -- segment definition --
+;
+%define SEG_TEXT    .text  align=16 public use32 class=CODE
+%define SEG_CONST   .data  align=16 public use32 class=DATA
+
+%elifdef ELF	; ----(nasm -felf -DELF ...)------------
+; * Linux
+; * *BSD family Unix using elf format
+; * Unix System V, including Solaris x86, UnixWare and SCO Unix
+
+; -- segment definition --
+;
+%define SEG_TEXT    .text   progbits alloc exec   nowrite align=16
+%define SEG_CONST   .rodata progbits alloc noexec nowrite align=16
+
+; To make the code position-independent, append -DPIC to the commandline
+;
+%define GOT_SYMBOL  _GLOBAL_OFFSET_TABLE_	; ELF supports PIC
+%define EXTN(name)  name			; foo() -> foo
+
+%elifdef AOUT	; ----(nasm -faoutb/aout -DAOUT ...)----
+; * Older Linux using a.out format  (nasm -f aout -DAOUT ...)
+; * *BSD family Unix using a.out format  (nasm -f aoutb -DAOUT ...)
+
+; -- segment definition --
+;
+%define SEG_TEXT    .text
+%define SEG_CONST   .data
+
+; To make the code position-independent, append -DPIC to the commandline
+;
+%define GOT_SYMBOL  __GLOBAL_OFFSET_TABLE_	; BSD-style a.out supports PIC
+
+%elifdef MACHO	; ----(nasm -fmacho -DMACHO ...)--------
+; * NeXTstep/OpenStep/Rhapsody/Darwin/MacOS X (Mach-O format)
+
+; -- segment definition --
+;
+%define SEG_TEXT    .text  ;align=16	; nasm doesn't accept align=16. why?
+%define SEG_CONST   .rodata align=16
+
+; The generation of position-independent code (PIC) is the default on Darwin.
+;
+%define PIC
+%define GOT_SYMBOL  _MACHO_PIC_		; Mach-O style code-relative addressing
+
+%else		; ----(Other case)----------------------
+
+; -- segment definition --
+;
+%define SEG_TEXT    .text
+%define SEG_CONST   .data
+
+%endif	; ----------------------------------------------
+
+; ==========================================================================
+
+; --------------------------------------------------------------------------
+;  Common types
+;
+%define POINTER                 dword           ; general pointer type
+%define SIZEOF_POINTER          SIZEOF_DWORD    ; sizeof(POINTER)
+%define POINTER_BIT             DWORD_BIT       ; sizeof(POINTER)*BYTE_BIT
+
+; --------------------------------------------------------------------------
+;  External Symbol Name
+;
+%ifndef EXTN
+%define EXTN(name)   _ %+ name		; foo() -> _foo
+%endif
+
+; --------------------------------------------------------------------------
+;  Macros for position-independent code (PIC) support
+;
+%ifndef GOT_SYMBOL
+%undef PIC
+%endif
+
+%ifdef PIC ; -------------------------------------------
+
+%ifidn GOT_SYMBOL,_MACHO_PIC_ ; --------------------
+
+; At present, nasm doesn't seem to support PIC generation for Mach-O.
+; The PIC support code below is a little tricky.
+
+	SECTION	SEG_CONST
+const_base:
+
+%define GOTOFF(got,sym) (got) + (sym) - const_base
+
+%imacro get_GOT	1
+	; NOTE: this macro destroys ecx resister.
+	call	%%geteip
+	add	ecx, byte (%%ref - $)
+	jmp	short %%adjust
+%%geteip:
+	mov	ecx, POINTER [esp]
+	ret
+%%adjust:
+	push	ebp
+	xor	ebp,ebp		; ebp = 0
+%ifidni %1,ebx	; (%1 == ebx)
+	; db 0x8D,0x9C + jmp near const_base =
+	;   lea ebx, [ecx+ebp*8+(const_base-%%ref)] ; 8D,9C,E9,(offset32)
+	db	0x8D,0x9C		; 8D,9C
+	jmp	near const_base		; E9,(const_base-%%ref)
+%%ref:
+%else  ; (%1 != ebx)
+	; db 0x8D,0x8C + jmp near const_base =
+	;   lea ecx, [ecx+ebp*8+(const_base-%%ref)] ; 8D,8C,E9,(offset32)
+	db	0x8D,0x8C		; 8D,8C
+	jmp	near const_base		; E9,(const_base-%%ref)
+%%ref:	mov	%1, ecx
+%endif ; (%1 == ebx)
+	pop	ebp
+%endmacro
+
+%else	; GOT_SYMBOL != _MACHO_PIC_ ----------------
+
+%define GOTOFF(got,sym) (got) + (sym) wrt ..gotoff
+
+%imacro get_GOT	1
+	extern	GOT_SYMBOL
+	call	%%geteip
+	add	%1, GOT_SYMBOL + $$ - $ wrt ..gotpc
+	jmp	short %%done
+%%geteip:
+	mov	%1, POINTER [esp]
+	ret
+%%done:
+%endmacro
+
+%endif	; GOT_SYMBOL == _MACHO_PIC_ ----------------
+
+%imacro pushpic	1.nolist
+	push	%1
+%endmacro
+%imacro poppic	1.nolist
+	pop	%1
+%endmacro
+%imacro movpic	2.nolist
+	mov	%1,%2
+%endmacro
+
+%else	; !PIC -----------------------------------------
+
+%define GOTOFF(got,sym) (sym)
+
+%imacro get_GOT	1.nolist
+%endmacro
+%imacro pushpic	1.nolist
+%endmacro
+%imacro poppic	1.nolist
+%endmacro
+%imacro movpic	2.nolist
+%endmacro
+
+%endif	;  PIC -----------------------------------------
+
+; --------------------------------------------------------------------------
+;  Align the next instruction on {2,4,8,16,..}-byte boundary.
+;  ".balign n,,m" in GNU as
+;
+%define MSKLE(x,y)  (~(((y) & 0xFFFF) - ((x) & 0xFFFF)) >> 16)
+%define FILLB(b,n)  (($$-(b)) & ((n)-1))
+
+%imacro alignx 1-2.nolist 0xFFFF
+%%bs:	times MSKLE(FILLB(%%bs,%1),%2) & MSKLE(16,FILLB($,%1)) & FILLB($,%1) \
+	       db 0x90                               ; nop
+	times MSKLE(FILLB(%%bs,%1),%2) & FILLB($,%1)/9 \
+	       db 0x8D,0x9C,0x23,0x00,0x00,0x00,0x00 ; lea ebx,[ebx+0x00000000]
+	times MSKLE(FILLB(%%bs,%1),%2) & FILLB($,%1)/7 \
+	       db 0x8D,0xAC,0x25,0x00,0x00,0x00,0x00 ; lea ebp,[ebp+0x00000000]
+	times MSKLE(FILLB(%%bs,%1),%2) & FILLB($,%1)/6 \
+	       db 0x8D,0xAD,0x00,0x00,0x00,0x00      ; lea ebp,[ebp+0x00000000]
+	times MSKLE(FILLB(%%bs,%1),%2) & FILLB($,%1)/4 \
+	       db 0x8D,0x6C,0x25,0x00                ; lea ebp,[ebp+0x00]
+	times MSKLE(FILLB(%%bs,%1),%2) & FILLB($,%1)/3 \
+	       db 0x8D,0x6D,0x00                     ; lea ebp,[ebp+0x00]
+	times MSKLE(FILLB(%%bs,%1),%2) & FILLB($,%1)/2 \
+	       db 0x8B,0xED                          ; mov ebp,ebp
+	times MSKLE(FILLB(%%bs,%1),%2) & FILLB($,%1)/1 \
+	       db 0x90                               ; nop
+%endmacro
+
+; Align the next data on {2,4,8,16,..}-byte boundary.
+;
+%imacro alignz 1.nolist
+	align %1, db 0		; filling zeros
+%endmacro
+
+
+; --------------------------------------------------------------------------
+;  Defines picked up from the C headers
+;
+%include "simd/jsimdcfg.inc"
+
+; --------------------------------------------------------------------------