Framework for supporting SIMD acceleration

Designed to impose minimal changes on the "normal" code.


git-svn-id: https://libjpeg-turbo.svn.sourceforge.net/svnroot/libjpeg-turbo@14 3789f03b-4d11-0410-bbf8-ca57d06f2519

13 files changed, 674 insertions, 29 deletions

diff --git a/trunk/Makefile.am b/trunk/Makefile.am
index 7404d10..114dd7d 100644
--- a/trunk/Makefile.am
+++ b/trunk/Makefile.am
@@ -1,7 +1,7 @@
 noinst_LTLIBRARIES = libjpeg.la
 
 HDRS = jchuff.h jdct.h jdhuff.h jerror.h jinclude.h jmemsys.h jmorecfg.h \
-	jpegint.h jpeglib.h jversion.h
+	jpegint.h jpeglib.h jversion.h jsimd.h jsimddct.h
 
 libjpeg_la_SOURCES = $(HDRS) jcapimin.c jcapistd.c jccoefct.c jccolor.c \
 	jcdctmgr.c jchuff.c jcinit.c jcmainct.c jcmarker.c jcmaster.c \
@@ -10,5 +10,6 @@ libjpeg_la_SOURCES = $(HDRS) jcapimin.c jcapistd.c jccoefct.c jccolor.c \
 	jddctmgr.c jdhuff.c jdinput.c jdmainct.c jdmarker.c jdmaster.c \
 	jdmerge.c jdphuff.c jdpostct.c jdsample.c jdtrans.c jerror.c \
 	jfdctflt.c jfdctfst.c jfdctint.c jidctflt.c jidctfst.c jidctint.c \
-	jidctred.c jquant1.c jquant2.c jutils.c jmemmgr.c jmemnobs.c
+	jidctred.c jquant1.c jquant2.c jutils.c jmemmgr.c jmemnobs.c \
+	jsimd.c
 
diff --git a/trunk/configure.ac b/trunk/configure.ac
index 79bdb45..c2263f2 100644
--- a/trunk/configure.ac
+++ b/trunk/configure.ac
@@ -72,6 +72,14 @@ AC_CHECK_FUNCS([memset memcpy], [],
 	[AC_DEFINE([NEED_BSD_STRINGS], 1,
 		   [Define if you have BSD-like bzero and bcopy])])
 
+# SIMD is optional
+# todo: detect automatically when it can be enabled
+AC_ARG_WITH([simd],
+    AC_HELP_STRING([--with-simd],[Include accelerated SIMD routines.]))
+if test "x${with_simd}" != "xno"; then
+	AC_DEFINE([WITH_SIMD], [1], [Use accelerated SIMD routines.])
+fi
+
 # 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
 # API headers, which can screw things up for users of the lib.
diff --git a/trunk/jccolor.c b/trunk/jccolor.c
index 0a8a4b5..2407456 100644
--- a/trunk/jccolor.c
+++ b/trunk/jccolor.c
@@ -2,6 +2,7 @@
  * jccolor.c
  *
  * Copyright (C) 1991-1996, Thomas G. Lane.
+ * Copyright 2009 Pierre Ossman <ossman@cendio.se> for Cendio AB
  * This file is part of the Independent JPEG Group's software.
  * For conditions of distribution and use, see the accompanying README file.
  *
@@ -11,6 +12,7 @@
 #define JPEG_INTERNALS
 #include "jinclude.h"
 #include "jpeglib.h"
+#include "jsimd.h"
 
 
 /* Private subobject */
@@ -420,8 +422,12 @@ jinit_color_converter (j_compress_ptr cinfo)
     if (cinfo->num_components != 3)
       ERREXIT(cinfo, JERR_BAD_J_COLORSPACE);
     if (cinfo->in_color_space == JCS_RGB) {
-      cconvert->pub.start_pass = rgb_ycc_start;
-      cconvert->pub.color_convert = rgb_ycc_convert;
+      if (jsimd_can_rgb_ycc())
+        cconvert->pub.color_convert = jsimd_rgb_ycc_convert;
+      else {
+        cconvert->pub.start_pass = rgb_ycc_start;
+        cconvert->pub.color_convert = rgb_ycc_convert;
+      }
     } else if (cinfo->in_color_space == JCS_YCbCr)
       cconvert->pub.color_convert = null_convert;
     else
diff --git a/trunk/jcdctmgr.c b/trunk/jcdctmgr.c
index 45b9c0d..75d48e0 100644
--- a/trunk/jcdctmgr.c
+++ b/trunk/jcdctmgr.c
@@ -2,6 +2,7 @@
  * jcdctmgr.c
  *
  * Copyright (C) 1994-1996, Thomas G. Lane.
+ * Copyright 2009 Pierre Ossman <ossman@cendio.se> for Cendio AB
  * This file is part of the Independent JPEG Group's software.
  * For conditions of distribution and use, see the accompanying README file.
  *
@@ -15,6 +16,7 @@
 #include "jinclude.h"
 #include "jpeglib.h"
 #include "jdct.h"		/* Private declarations for DCT subsystem */
+#include "jsimddct.h"
 
 
 /* Private subobject for this module */
@@ -424,19 +426,28 @@ jinit_forward_dct (j_compress_ptr cinfo)
 #ifdef DCT_ISLOW_SUPPORTED
   case JDCT_ISLOW:
     fdct->pub.forward_DCT = forward_DCT;
-    fdct->dct = jpeg_fdct_islow;
+    if (jsimd_can_fdct_islow())
+      fdct->dct = jsimd_fdct_islow;
+    else
+      fdct->dct = jpeg_fdct_islow;
     break;
 #endif
 #ifdef DCT_IFAST_SUPPORTED
   case JDCT_IFAST:
     fdct->pub.forward_DCT = forward_DCT;
-    fdct->dct = jpeg_fdct_ifast;
+    if (jsimd_can_fdct_ifast())
+      fdct->dct = jsimd_fdct_ifast;
+    else
+      fdct->dct = jpeg_fdct_ifast;
     break;
 #endif
 #ifdef DCT_FLOAT_SUPPORTED
   case JDCT_FLOAT:
     fdct->pub.forward_DCT = forward_DCT_float;
-    fdct->float_dct = jpeg_fdct_float;
+    if (jsimd_can_fdct_float())
+      fdct->float_dct = jsimd_fdct_float;
+    else
+      fdct->float_dct = jpeg_fdct_float;
     break;
 #endif
   default:
@@ -453,14 +464,26 @@ jinit_forward_dct (j_compress_ptr cinfo)
   case JDCT_IFAST:
 #endif
 #if defined(DCT_ISLOW_SUPPORTED) || defined(DCT_IFAST_SUPPORTED)
-    fdct->convsamp = convsamp;
-    fdct->quantize = quantize;
+    if (jsimd_can_convsamp())
+      fdct->convsamp = jsimd_convsamp;
+    else
+      fdct->convsamp = convsamp;
+    if (jsimd_can_quantize())
+      fdct->quantize = jsimd_quantize;
+    else
+      fdct->quantize = quantize;
     break;
 #endif
 #ifdef DCT_FLOAT_SUPPORTED
   case JDCT_FLOAT:
-    fdct->float_convsamp = convsamp_float;
-    fdct->float_quantize = quantize_float;
+    if (jsimd_can_convsamp_float())
+      fdct->float_convsamp = jsimd_convsamp_float;
+    else
+      fdct->float_convsamp = convsamp_float;
+    if (jsimd_can_quantize_float())
+      fdct->float_quantize = jsimd_quantize_float;
+    else
+      fdct->float_quantize = quantize_float;
     break;
 #endif
   default:
diff --git a/trunk/jconfig.h.in b/trunk/jconfig.h.in
index 4a10697..4e5e80e 100644
--- a/trunk/jconfig.h.in
+++ b/trunk/jconfig.h.in
@@ -28,6 +28,9 @@
 /* Define if shift is unsigned */
 #undef RIGHT_SHIFT_IS_UNSIGNED
 
+/* Use accelerated SIMD routines. */
+#undef WITH_SIMD
+
 /* Define to 1 if type `char' is unsigned and you are not using gcc.  */
 #ifndef __CHAR_UNSIGNED__
 # undef __CHAR_UNSIGNED__
diff --git a/trunk/jcsample.c b/trunk/jcsample.c
index 212ec87..eea376f 100644
--- a/trunk/jcsample.c
+++ b/trunk/jcsample.c
@@ -2,6 +2,7 @@
  * jcsample.c
  *
  * Copyright (C) 1991-1996, Thomas G. Lane.
+ * Copyright 2009 Pierre Ossman <ossman@cendio.se> for Cendio AB
  * This file is part of the Independent JPEG Group's software.
  * For conditions of distribution and use, see the accompanying README file.
  *
@@ -48,6 +49,7 @@
 #define JPEG_INTERNALS
 #include "jinclude.h"
 #include "jpeglib.h"
+#include "jsimd.h"
 
 
 /* Pointer to routine to downsample a single component */
@@ -494,7 +496,10 @@ jinit_downsampler (j_compress_ptr cinfo)
     } else if (compptr->h_samp_factor * 2 == cinfo->max_h_samp_factor &&
 	       compptr->v_samp_factor == cinfo->max_v_samp_factor) {
       smoothok = FALSE;
-      downsample->methods[ci] = h2v1_downsample;
+      if (jsimd_can_h2v1_downsample())
+        downsample->methods[ci] = jsimd_h2v1_downsample;
+      else
+        downsample->methods[ci] = h2v1_downsample;
     } else if (compptr->h_samp_factor * 2 == cinfo->max_h_samp_factor &&
 	       compptr->v_samp_factor * 2 == cinfo->max_v_samp_factor) {
 #ifdef INPUT_SMOOTHING_SUPPORTED
@@ -503,7 +508,10 @@ jinit_downsampler (j_compress_ptr cinfo)
 	downsample->pub.need_context_rows = TRUE;
       } else
 #endif
-	downsample->methods[ci] = h2v2_downsample;
+	if (jsimd_can_h2v2_downsample())
+	  downsample->methods[ci] = jsimd_h2v2_downsample;
+	else
+	  downsample->methods[ci] = h2v2_downsample;
     } else if ((cinfo->max_h_samp_factor % compptr->h_samp_factor) == 0 &&
 	       (cinfo->max_v_samp_factor % compptr->v_samp_factor) == 0) {
       smoothok = FALSE;
diff --git a/trunk/jdcolor.c b/trunk/jdcolor.c
index 6c04dfe..d2d1483 100644
--- a/trunk/jdcolor.c
+++ b/trunk/jdcolor.c
@@ -2,6 +2,7 @@
  * jdcolor.c
  *
  * Copyright (C) 1991-1997, Thomas G. Lane.
+ * Copyright 2009 Pierre Ossman <ossman@cendio.se> for Cendio AB
  * This file is part of the Independent JPEG Group's software.
  * For conditions of distribution and use, see the accompanying README file.
  *
@@ -11,6 +12,7 @@
 #define JPEG_INTERNALS
 #include "jinclude.h"
 #include "jpeglib.h"
+#include "jsimd.h"
 
 
 /* Private subobject */
@@ -358,8 +360,12 @@ jinit_color_deconverter (j_decompress_ptr cinfo)
   case JCS_RGB:
     cinfo->out_color_components = RGB_PIXELSIZE;
     if (cinfo->jpeg_color_space == JCS_YCbCr) {
-      cconvert->pub.color_convert = ycc_rgb_convert;
-      build_ycc_rgb_table(cinfo);
+      if (jsimd_can_ycc_rgb())
+        cconvert->pub.color_convert = jsimd_ycc_rgb_convert;
+      else {
+        cconvert->pub.color_convert = ycc_rgb_convert;
+        build_ycc_rgb_table(cinfo);
+      }
     } else if (cinfo->jpeg_color_space == JCS_GRAYSCALE) {
       cconvert->pub.color_convert = gray_rgb_convert;
     } else if (cinfo->jpeg_color_space == JCS_RGB && RGB_PIXELSIZE == 3) {
diff --git a/trunk/jddctmgr.c b/trunk/jddctmgr.c
index bbf8d0e..52f5090 100644
--- a/trunk/jddctmgr.c
+++ b/trunk/jddctmgr.c
@@ -2,6 +2,7 @@
  * jddctmgr.c
  *
  * Copyright (C) 1994-1996, Thomas G. Lane.
+ * Copyright 2009 Pierre Ossman <ossman@cendio.se> for Cendio AB
  * This file is part of the Independent JPEG Group's software.
  * For conditions of distribution and use, see the accompanying README file.
  *
@@ -19,6 +20,7 @@
 #include "jinclude.h"
 #include "jpeglib.h"
 #include "jdct.h"		/* Private declarations for DCT subsystem */
+#include "jsimddct.h"
 
 
 /*
@@ -105,11 +107,17 @@ start_pass (j_decompress_ptr cinfo)
       method = JDCT_ISLOW;	/* jidctred uses islow-style table */
       break;
     case 2:
-      method_ptr = jpeg_idct_2x2;
+      if (jsimd_can_idct_2x2())
+        method_ptr = jsimd_idct_2x2;
+      else
+        method_ptr = jpeg_idct_2x2;
       method = JDCT_ISLOW;	/* jidctred uses islow-style table */
       break;
     case 4:
-      method_ptr = jpeg_idct_4x4;
+      if (jsimd_can_idct_4x4())
+        method_ptr = jsimd_idct_4x4;
+      else
+        method_ptr = jpeg_idct_4x4;
       method = JDCT_ISLOW;	/* jidctred uses islow-style table */
       break;
 #endif
@@ -117,19 +125,28 @@ start_pass (j_decompress_ptr cinfo)
       switch (cinfo->dct_method) {
 #ifdef DCT_ISLOW_SUPPORTED
       case JDCT_ISLOW:
-	method_ptr = jpeg_idct_islow;
+	if (jsimd_can_idct_islow())
+	  method_ptr = jsimd_idct_islow;
+	else
+	  method_ptr = jpeg_idct_islow;
 	method = JDCT_ISLOW;
 	break;
 #endif
 #ifdef DCT_IFAST_SUPPORTED
       case JDCT_IFAST:
-	method_ptr = jpeg_idct_ifast;
+	if (jsimd_can_idct_ifast())
+	  method_ptr = jsimd_idct_ifast;
+	else
+	  method_ptr = jpeg_idct_ifast;
 	method = JDCT_IFAST;
 	break;
 #endif
 #ifdef DCT_FLOAT_SUPPORTED
       case JDCT_FLOAT:
-	method_ptr = jpeg_idct_float;
+	if (jsimd_can_idct_float())
+	  method_ptr = jsimd_idct_float;
+	else
+	  method_ptr = jpeg_idct_float;
 	method = JDCT_FLOAT;
 	break;
 #endif
diff --git a/trunk/jdmerge.c b/trunk/jdmerge.c
index 3744446..7742941 100644
--- a/trunk/jdmerge.c
+++ b/trunk/jdmerge.c
@@ -2,6 +2,7 @@
  * jdmerge.c
  *
  * Copyright (C) 1994-1996, Thomas G. Lane.
+ * Copyright 2009 Pierre Ossman <ossman@cendio.se> for Cendio AB
  * This file is part of the Independent JPEG Group's software.
  * For conditions of distribution and use, see the accompanying README file.
  *
@@ -35,6 +36,7 @@
 #define JPEG_INTERNALS
 #include "jinclude.h"
 #include "jpeglib.h"
+#include "jsimd.h"
 
 #ifdef UPSAMPLE_MERGING_SUPPORTED
 
@@ -382,14 +384,20 @@ jinit_merged_upsampler (j_decompress_ptr cinfo)
 
   if (cinfo->max_v_samp_factor == 2) {
     upsample->pub.upsample = merged_2v_upsample;
-    upsample->upmethod = h2v2_merged_upsample;
+    if (jsimd_can_h2v2_merged_upsample())
+      upsample->upmethod = jsimd_h2v2_merged_upsample;
+    else
+      upsample->upmethod = h2v2_merged_upsample;
     /* Allocate a spare row buffer */
     upsample->spare_row = (JSAMPROW)
       (*cinfo->mem->alloc_large) ((j_common_ptr) cinfo, JPOOL_IMAGE,
 		(size_t) (upsample->out_row_width * SIZEOF(JSAMPLE)));
   } else {
     upsample->pub.upsample = merged_1v_upsample;
-    upsample->upmethod = h2v1_merged_upsample;
+    if (jsimd_can_h2v1_merged_upsample())
+      upsample->upmethod = jsimd_h2v1_merged_upsample;
+    else
+      upsample->upmethod = h2v1_merged_upsample;
     /* No spare row needed */
     upsample->spare_row = NULL;
   }
diff --git a/trunk/jdsample.c b/trunk/jdsample.c
index 80ffefb..4e0b8b4 100644
--- a/trunk/jdsample.c
+++ b/trunk/jdsample.c
@@ -2,6 +2,7 @@
  * jdsample.c
  *
  * Copyright (C) 1991-1996, Thomas G. Lane.
+ * Copyright 2009 Pierre Ossman <ossman@cendio.se> for Cendio AB
  * This file is part of the Independent JPEG Group's software.
  * For conditions of distribution and use, see the accompanying README file.
  *
@@ -21,6 +22,7 @@
 #define JPEG_INTERNALS
 #include "jinclude.h"
 #include "jpeglib.h"
+#include "jsimd.h"
 
 
 /* Pointer to routine to upsample a single component */
@@ -447,18 +449,32 @@ jinit_upsampler (j_decompress_ptr cinfo)
     } else if (h_in_group * 2 == h_out_group &&
 	       v_in_group == v_out_group) {
       /* Special cases for 2h1v upsampling */
-      if (do_fancy && compptr->downsampled_width > 2)
-	upsample->methods[ci] = h2v1_fancy_upsample;
-      else
-	upsample->methods[ci] = h2v1_upsample;
+      if (do_fancy && compptr->downsampled_width > 2) {
+	if (jsimd_can_h2v1_fancy_upsample())
+	  upsample->methods[ci] = jsimd_h2v1_fancy_upsample;
+	else
+	  upsample->methods[ci] = h2v1_fancy_upsample;
+      } else {
+	if (jsimd_can_h2v1_upsample())
+	  upsample->methods[ci] = jsimd_h2v1_upsample;
+	else
+	  upsample->methods[ci] = h2v1_upsample;
+      }
     } else if (h_in_group * 2 == h_out_group &&
 	       v_in_group * 2 == v_out_group) {
       /* Special cases for 2h2v upsampling */
       if (do_fancy && compptr->downsampled_width > 2) {
-	upsample->methods[ci] = h2v2_fancy_upsample;
+	if (jsimd_can_h2v2_fancy_upsample())
+	  upsample->methods[ci] = jsimd_h2v2_fancy_upsample;
+	else
+	  upsample->methods[ci] = h2v2_fancy_upsample;
 	upsample->pub.need_context_rows = TRUE;
-      } else
-	upsample->methods[ci] = h2v2_upsample;
+      } else {
+	if (jsimd_can_h2v2_upsample())
+	  upsample->methods[ci] = jsimd_h2v2_upsample;
+	else
+	  upsample->methods[ci] = h2v2_upsample;
+      }
     } else if ((h_out_group % h_in_group) == 0 &&
 	       (v_out_group % v_in_group) == 0) {
       /* Generic integral-factors upsampling method */
diff --git a/trunk/jsimd.c b/trunk/jsimd.c
new file mode 100644
index 0000000..c6378e8
--- /dev/null
+++ b/trunk/jsimd.c
@@ -0,0 +1,359 @@
+/*
+ * jsimd.c
+ *
+ * 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.
+ *
+ * This file contains the interface between the "normal" portions
+ * of the library and the SIMD implementations.
+ */
+
+#define JPEG_INTERNALS
+#include "jinclude.h"
+#include "jpeglib.h"
+#include "jdct.h"
+
+#define JSIMD_NONE    0x00
+
+static unsigned int simd_support = ~0;
+
+/*
+ * Check what SIMD accelerations are supported.
+ *
+ * FIXME: This code is racy under a multi-threaded environment.
+ */
+LOCAL(void)
+init_simd (void)
+{
+  if (simd_support != ~0)
+    return;
+
+  simd_support = JSIMD_NONE;
+}
+
+GLOBAL(int)
+jsimd_can_rgb_ycc (void)
+{
+  init_simd();
+
+  return 0;
+}
+
+GLOBAL(int)
+jsimd_can_ycc_rgb (void)
+{
+  init_simd();
+
+  return 0;
+}
+
+GLOBAL(void)
+jsimd_rgb_ycc_convert (j_compress_ptr cinfo,
+                       JSAMPARRAY input_buf, JSAMPIMAGE output_buf,
+                       JDIMENSION output_row, int num_rows)
+{
+}
+
+GLOBAL(void)
+jsimd_ycc_rgb_convert (j_decompress_ptr cinfo,
+                       JSAMPIMAGE input_buf, JDIMENSION input_row,
+                       JSAMPARRAY output_buf, int num_rows)
+{
+}
+
+GLOBAL(int)
+jsimd_can_h2v2_downsample (void)
+{
+  init_simd();
+
+  return 0;
+}
+
+GLOBAL(int)
+jsimd_can_h2v1_downsample (void)
+{
+  init_simd();
+
+  return 0;
+}
+
+GLOBAL(void)
+jsimd_h2v2_downsample (j_compress_ptr cinfo, jpeg_component_info * compptr,
+                       JSAMPARRAY input_data, JSAMPARRAY output_data)
+{
+}
+
+GLOBAL(void)
+jsimd_h2v1_downsample (j_compress_ptr cinfo, jpeg_component_info * compptr,
+                       JSAMPARRAY input_data, JSAMPARRAY output_data)
+{
+}
+
+GLOBAL(int)
+jsimd_can_h2v2_upsample (void)
+{
+  init_simd();
+
+  return 0;
+}
+
+GLOBAL(int)
+jsimd_can_h2v1_upsample (void)
+{
+  init_simd();
+
+  return 0;
+}
+
+GLOBAL(void)
+jsimd_h2v2_upsample (j_decompress_ptr cinfo,
+                     jpeg_component_info * compptr, 
+                     JSAMPARRAY input_data,
+                     JSAMPARRAY * output_data_ptr)
+{
+}
+
+GLOBAL(void)
+jsimd_h2v1_upsample (j_decompress_ptr cinfo,
+                     jpeg_component_info * compptr, 
+                     JSAMPARRAY input_data,
+                     JSAMPARRAY * output_data_ptr)
+{
+}
+
+GLOBAL(int)
+jsimd_can_h2v2_fancy_upsample (void)
+{
+  init_simd();
+
+  return 0;
+}
+
+GLOBAL(int)
+jsimd_can_h2v1_fancy_upsample (void)
+{
+  init_simd();
+
+  return 0;
+}
+
+GLOBAL(void)
+jsimd_h2v2_fancy_upsample (j_decompress_ptr cinfo,
+                           jpeg_component_info * compptr, 
+                           JSAMPARRAY input_data,
+                           JSAMPARRAY * output_data_ptr)
+{
+}
+
+GLOBAL(void)
+jsimd_h2v1_fancy_upsample (j_decompress_ptr cinfo,
+                           jpeg_component_info * compptr, 
+                           JSAMPARRAY input_data,
+                           JSAMPARRAY * output_data_ptr)
+{
+}
+
+GLOBAL(int)
+jsimd_can_h2v2_merged_upsample (void)
+{
+  init_simd();
+
+  return 0;
+}
+
+GLOBAL(int)
+jsimd_can_h2v1_merged_upsample (void)
+{
+  init_simd();
+
+  return 0;
+}
+
+GLOBAL(void)
+jsimd_h2v2_merged_upsample (j_decompress_ptr cinfo,
+                            JSAMPIMAGE input_buf,
+                            JDIMENSION in_row_group_ctr,
+                            JSAMPARRAY output_buf)
+{
+}
+
+GLOBAL(void)
+jsimd_h2v1_merged_upsample (j_decompress_ptr cinfo,
+                            JSAMPIMAGE input_buf,
+                            JDIMENSION in_row_group_ctr,
+                            JSAMPARRAY output_buf)
+{
+}
+
+GLOBAL(int)
+jsimd_can_convsamp (void)
+{
+  init_simd();
+
+  return 0;
+}
+
+GLOBAL(int)
+jsimd_can_convsamp_float (void)
+{
+  init_simd();
+
+  return 0;
+}
+
+GLOBAL(void)
+jsimd_convsamp (JSAMPARRAY sample_data, JDIMENSION start_col,
+                DCTELEM * workspace)
+{
+}
+
+GLOBAL(void)
+jsimd_convsamp_float (JSAMPARRAY sample_data, JDIMENSION start_col,
+                      FAST_FLOAT * workspace)
+{
+}
+
+GLOBAL(int)
+jsimd_can_fdct_islow (void)
+{
+  init_simd();
+
+  return 0;
+}
+
+GLOBAL(int)
+jsimd_can_fdct_ifast (void)
+{
+  init_simd();
+
+  return 0;
+}
+
+GLOBAL(int)
+jsimd_can_fdct_float (void)
+{
+  init_simd();
+
+  return 0;
+}
+
+GLOBAL(void)
+jsimd_fdct_islow (DCTELEM * data)
+{
+}
+
+GLOBAL(void)
+jsimd_fdct_ifast (DCTELEM * data)
+{
+}
+
+GLOBAL(void)
+jsimd_fdct_float (FAST_FLOAT * data)
+{
+}
+
+GLOBAL(int)
+jsimd_can_quantize (void)
+{
+  init_simd();
+
+  return 0;
+}
+
+GLOBAL(int)
+jsimd_can_quantize_float (void)
+{
+  init_simd();
+
+  return 0;
+}
+
+GLOBAL(void)
+jsimd_quantize (JCOEFPTR coef_block, DCTELEM * divisors,
+                DCTELEM * workspace)
+{
+}
+
+GLOBAL(void)
+jsimd_quantize_float (JCOEFPTR coef_block, FAST_FLOAT * divisors,
+                      FAST_FLOAT * workspace)
+{
+}
+
+GLOBAL(int)
+jsimd_can_idct_2x2 (void)
+{
+  init_simd();
+
+  return 0;
+}
+
+GLOBAL(int)
+jsimd_can_idct_4x4 (void)
+{
+  init_simd();
+
+  return 0;
+}
+
+GLOBAL(void)
+jsimd_idct_2x2 (j_decompress_ptr cinfo, jpeg_component_info * compptr,
+                JCOEFPTR coef_block, JSAMPARRAY output_buf,
+                JDIMENSION output_col)
+{
+}
+
+GLOBAL(void)
+jsimd_idct_4x4 (j_decompress_ptr cinfo, jpeg_component_info * compptr,
+                JCOEFPTR coef_block, JSAMPARRAY output_buf,
+                JDIMENSION output_col)
+{
+}
+
+GLOBAL(int)
+jsimd_can_idct_islow (void)
+{
+  init_simd();
+
+  return 0;
+}
+
+GLOBAL(int)
+jsimd_can_idct_ifast (void)
+{
+  init_simd();
+
+  return 0;
+}
+
+GLOBAL(int)
+jsimd_can_idct_float (void)
+{
+  init_simd();
+
+  return 0;
+}
+
+GLOBAL(void)
+jsimd_idct_islow (j_decompress_ptr cinfo, jpeg_component_info * compptr,
+                JCOEFPTR coef_block, JSAMPARRAY output_buf,
+                JDIMENSION output_col)
+{
+}
+
+GLOBAL(void)
+jsimd_idct_ifast (j_decompress_ptr cinfo, jpeg_component_info * compptr,
+                JCOEFPTR coef_block, JSAMPARRAY output_buf,
+                JDIMENSION output_col)
+{
+}
+
+GLOBAL(void)
+jsimd_idct_float (j_decompress_ptr cinfo, jpeg_component_info * compptr,
+                JCOEFPTR coef_block, JSAMPARRAY output_buf,
+                JDIMENSION output_col)
+{
+}
+
diff --git a/trunk/jsimd.h b/trunk/jsimd.h
new file mode 100644
index 0000000..90be19e
--- /dev/null
+++ b/trunk/jsimd.h
@@ -0,0 +1,89 @@
+/*
+ * 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.
+ *
+ */
+
+/* Short forms of external names for systems with brain-damaged linkers. */
+
+#ifdef NEED_SHORT_EXTERNAL_NAMES
+#define jsimd_can_rgb_ycc                 jSCanRgbYcc
+#define jsimd_can_ycc_rgb                 jSCanYccRgb
+#define jsimd_rgb_ycc_convert             jSRgbYccConv
+#define jsimd_ycc_rgb_convert             jSYccRgbConv
+#define jsimd_can_h2v2_downsample         jSCanH2V2Down
+#define jsimd_can_h2v1_downsample         jSCanH2V1Down
+#define jsimd_h2v2_downsample             jSH2V2Down
+#define jsimd_h2v1_downsample             jSH2V1Down
+#define jsimd_can_h2v2_upsample           jSCanH2V2Up
+#define jsimd_can_h2v1_upsample           jSCanH2V1Up
+#define jsimd_h2v2_upsample               jSH2V2Up
+#define jsimd_h2v1_upsample               jSH2V1Up
+#define jsimd_can_h2v2_fancy_upsample     jSCanH2V2FUp
+#define jsimd_can_h2v1_fancy_upsample     jSCanH2V1FUp
+#define jsimd_h2v2_fancy_upsample         jSH2V2FUp
+#define jsimd_h2v1_fancy_upsample         jSH2V1FUp
+#define jsimd_can_h2v2_merged_upsample    jSCanH2V2MUp
+#define jsimd_can_h2v1_merged_upsample    jSCanH2V1MUp
+#define jsimd_h2v2_merged_upsample        jSH2V2MUp
+#define jsimd_h2v1_merged_upsample        jSH2V1MUp
+#endif /* NEED_SHORT_EXTERNAL_NAMES */
+
+EXTERN(int) jsimd_can_rgb_ycc JPP((void));
+EXTERN(int) jsimd_can_ycc_rgb JPP((void));
+
+EXTERN(void) jsimd_rgb_ycc_convert
+        JPP((j_compress_ptr cinfo,
+             JSAMPARRAY input_buf, JSAMPIMAGE output_buf,
+             JDIMENSION output_row, int num_rows));
+EXTERN(void) jsimd_ycc_rgb_convert
+        JPP((j_decompress_ptr cinfo,
+             JSAMPIMAGE input_buf, JDIMENSION input_row,
+             JSAMPARRAY output_buf, int num_rows));
+
+EXTERN(int) jsimd_can_h2v2_downsample JPP((void));
+EXTERN(int) jsimd_can_h2v1_downsample JPP((void));
+
+EXTERN(void) jsimd_h2v2_downsample
+        JPP((j_compress_ptr cinfo, jpeg_component_info * compptr,
+             JSAMPARRAY input_data, JSAMPARRAY output_data));
+EXTERN(void) jsimd_h2v1_downsample
+        JPP((j_compress_ptr cinfo, jpeg_component_info * compptr,
+             JSAMPARRAY input_data, JSAMPARRAY output_data));
+
+EXTERN(int) jsimd_can_h2v2_upsample JPP((void));
+EXTERN(int) jsimd_can_h2v1_upsample JPP((void));
+
+EXTERN(void) jsimd_h2v2_upsample
+        JPP((j_decompress_ptr cinfo, jpeg_component_info * compptr,
+             JSAMPARRAY input_data, JSAMPARRAY * output_data_ptr));
+EXTERN(void) jsimd_h2v1_upsample
+        JPP((j_decompress_ptr cinfo, jpeg_component_info * compptr,
+             JSAMPARRAY input_data, JSAMPARRAY * output_data_ptr));
+
+EXTERN(int) jsimd_can_h2v2_fancy_upsample JPP((void));
+EXTERN(int) jsimd_can_h2v1_fancy_upsample JPP((void));
+
+EXTERN(void) jsimd_h2v2_fancy_upsample
+        JPP((j_decompress_ptr cinfo, jpeg_component_info * compptr,
+             JSAMPARRAY input_data, JSAMPARRAY * output_data_ptr));
+EXTERN(void) jsimd_h2v1_fancy_upsample
+        JPP((j_decompress_ptr cinfo, jpeg_component_info * compptr,
+             JSAMPARRAY input_data, JSAMPARRAY * output_data_ptr));
+
+EXTERN(int) jsimd_can_h2v2_merged_upsample JPP((void));
+EXTERN(int) jsimd_can_h2v1_merged_upsample JPP((void));
+
+EXTERN(void) jsimd_h2v2_merged_upsample
+        JPP((j_decompress_ptr cinfo,
+             JSAMPIMAGE input_buf, JDIMENSION in_row_group_ctr,
+             JSAMPARRAY output_buf));
+EXTERN(void) jsimd_h2v1_merged_upsample
+        JPP((j_decompress_ptr cinfo,
+             JSAMPIMAGE input_buf, JDIMENSION in_row_group_ctr,
+             JSAMPARRAY output_buf));
+
diff --git a/trunk/jsimddct.h b/trunk/jsimddct.h
new file mode 100644
index 0000000..d73d0c4
--- /dev/null
+++ b/trunk/jsimddct.h
@@ -0,0 +1,101 @@
+/*
+ * jsimddct.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.
+ *
+ */
+
+/* Short forms of external names for systems with brain-damaged linkers. */
+
+#ifdef NEED_SHORT_EXTERNAL_NAMES
+#define jsimd_can_convsamp                jSCanConv
+#define jsimd_can_convsamp_float          jSCanConvF
+#define jsimd_convsamp                    jSConv
+#define jsimd_convsamp_float              jSConvF
+#define jsimd_can_fdct_islow              jSCanFDCTIS
+#define jsimd_can_fdct_ifast              jSCanFDCTIF
+#define jsimd_can_fdct_float              jSCanFDCTFl
+#define jsimd_fdct_islow                  jSFDCTIS
+#define jsimd_fdct_ifast                  jSFDCTIF
+#define jsimd_fdct_float                  jSFDCTFl
+#define jsimd_can_quantize                jSCanQuant
+#define jsimd_can_quantize_float          jSCanQuantF
+#define jsimd_quantize                    jSQuant
+#define jsimd_quantize_float              jSQuantF
+#define jsimd_can_idct_2x2                jSCanIDCT22
+#define jsimd_can_idct_4x4                jSCanIDCT44
+#define jsimd_idct_2x2                    jSIDCT22
+#define jsimd_idct_4x4                    jSIDCT44
+#define jsimd_can_idct_islow              jSCanIDCTIS
+#define jsimd_can_idct_ifast              jSCanIDCTIF
+#define jsimd_can_idct_float              jSCanIDCTFl
+#define jsimd_idct_islow                  jSIDCTIS
+#define jsimd_idct_ifast                  jSIDCTIF
+#define jsimd_idct_float                  jSIDCTFl
+#endif /* NEED_SHORT_EXTERNAL_NAMES */
+
+EXTERN(int) jsimd_can_convsamp JPP((void));
+EXTERN(int) jsimd_can_convsamp_float JPP((void));
+
+EXTERN(void) jsimd_convsamp JPP((JSAMPARRAY sample_data,
+                                 JDIMENSION start_col,
+                                 DCTELEM * workspace));
+EXTERN(void) jsimd_convsamp_float JPP((JSAMPARRAY sample_data,
+                                       JDIMENSION start_col,
+                                       FAST_FLOAT * workspace));
+
+EXTERN(int) jsimd_can_fdct_islow JPP((void));
+EXTERN(int) jsimd_can_fdct_ifast JPP((void));
+EXTERN(int) jsimd_can_fdct_float JPP((void));
+
+EXTERN(void) jsimd_fdct_islow JPP((DCTELEM * data));
+EXTERN(void) jsimd_fdct_ifast JPP((DCTELEM * data));
+EXTERN(void) jsimd_fdct_float JPP((FAST_FLOAT * data));
+
+EXTERN(int) jsimd_can_quantize JPP((void));
+EXTERN(int) jsimd_can_quantize_float JPP((void));
+
+EXTERN(void) jsimd_quantize JPP((JCOEFPTR coef_block,
+                                 DCTELEM * divisors,
+                                 DCTELEM * workspace));
+EXTERN(void) jsimd_quantize_float JPP((JCOEFPTR coef_block,
+                                       FAST_FLOAT * divisors,
+                                       FAST_FLOAT * workspace));
+
+EXTERN(int) jsimd_can_idct_2x2 JPP((void));
+EXTERN(int) jsimd_can_idct_4x4 JPP((void));
+
+EXTERN(void) jsimd_idct_2x2 JPP((j_decompress_ptr cinfo,
+                                 jpeg_component_info * compptr,
+                                 JCOEFPTR coef_block,
+                                 JSAMPARRAY output_buf,
+                                 JDIMENSION output_col));
+EXTERN(void) jsimd_idct_4x4 JPP((j_decompress_ptr cinfo,
+                                 jpeg_component_info * compptr,
+                                 JCOEFPTR coef_block,
+                                 JSAMPARRAY output_buf,
+                                 JDIMENSION output_col));
+
+EXTERN(int) jsimd_can_idct_islow JPP((void));
+EXTERN(int) jsimd_can_idct_ifast JPP((void));
+EXTERN(int) jsimd_can_idct_float JPP((void));
+
+EXTERN(void) jsimd_idct_islow JPP((j_decompress_ptr cinfo,
+                                   jpeg_component_info * compptr,
+                                   JCOEFPTR coef_block,
+                                   JSAMPARRAY output_buf,
+                                   JDIMENSION output_col));
+EXTERN(void) jsimd_idct_ifast JPP((j_decompress_ptr cinfo,
+                                   jpeg_component_info * compptr,
+                                   JCOEFPTR coef_block,
+                                   JSAMPARRAY output_buf,
+                                   JDIMENSION output_col));
+EXTERN(void) jsimd_idct_float JPP((j_decompress_ptr cinfo,
+                                   jpeg_component_info * compptr,
+                                   JCOEFPTR coef_block,
+                                   JSAMPARRAY output_buf,
+                                   JDIMENSION output_col));
+