Clean up SIMD glue code

The SIMD glue code has gotten a bit #ifdef heavy so clean it up by having
one file for each possible SIMD arch. This also allows a simplification of
the x86_64 code as SSE/SSE2 is always known to exist on that arch.


git-svn-id: svn://svn.code.sf.net/p/libjpeg-turbo/code/trunk@49 632fc199-4ca6-4c93-a231-07263d6284db

4 files changed, 1645 insertions, 100 deletions

diff --git a/simd/Makefile.am b/simd/Makefile.am
index da7d684..8e7469f 100644
--- a/simd/Makefile.am
+++ b/simd/Makefile.am
@@ -4,11 +4,12 @@ BUILT_SOURCES = jsimdcfg.inc
 
 EXTRA_DIST = nasm_lt.sh
 
-if WITH_SIMD64
+if SIMD_X86_64
 
-libsimd_la_SOURCES = jsimd.h jsimdcfg.inc.h \
+libsimd_la_SOURCES = jsimd_x86_64.c \
+	jsimd.h jsimdcfg.inc.h \
 	jsimdext.inc jcolsamp.inc jdct.inc \
-	jsimdcpu-64.asm jfsseflt-64.asm \
+	jfsseflt-64.asm \
 	jccolss2-64.asm jdcolss2-64.asm \
 	jcsamss2-64.asm jdsamss2-64.asm jdmerss2-64.asm \
 	jcqnts2i-64.asm jfss2fst-64.asm jfss2int-64.asm \
@@ -19,9 +20,12 @@ jccolss2-64.lo: jcclrss2-64.asm
 jdcolss2-64.lo: jdclrss2-64.asm
 jdmerss2-64.lo: jdmrgss2-64.asm
 
-else
+endif
+
+if SIMD_I386
 
-libsimd_la_SOURCES = jsimd.h jsimdcfg.inc.h \
+libsimd_la_SOURCES = jsimd_i386.c \
+	jsimd.h jsimdcfg.inc.h \
 	jsimdext.inc jcolsamp.inc jdct.inc \
 	jsimdcpu.asm \
 	jccolmmx.asm jdcolmmx.asm \
diff --git a/simd/jsimd_i386.c b/simd/jsimd_i386.c
new file mode 100644
index 0000000..751fe2d
--- /dev/null
+++ b/simd/jsimd_i386.c
@@ -0,0 +1,956 @@
+/*
+ * jsimd_i386.c
+ *
+ * Copyright 2009 Pierre Ossman <ossman@cendio.se> for Cendio AB
+ * Copyright 2009 D. R. Commander
+ * 
+ * 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 when running on a
+ * 32-bit x86 architecture.
+ */
+
+#define JPEG_INTERNALS
+#include "../jinclude.h"
+#include "../jpeglib.h"
+#include "../jsimd.h"
+#include "../jdct.h"
+#include "../jsimddct.h"
+#include "simd/jsimd.h"
+
+/*
+ * In the PIC cases, we have no guarantee that constants will keep
+ * their alignment. This macro allows us to verify it at runtime.
+ */
+#define IS_ALIGNED(ptr, order) (((unsigned)ptr & ((1 << order) - 1)) == 0)
+
+#define IS_ALIGNED_SSE(ptr) (IS_ALIGNED(ptr, 4)) /* 16 byte alignment */
+
+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)
+{
+  char *env = NULL;
+
+  if (simd_support != ~0)
+    return;
+
+  simd_support = jpeg_simd_cpu_support();
+
+  /* Force different settings through environment variables */
+  env = getenv("JSIMD_FORCEMMX");
+  if ((env != NULL) && (strcmp(env, "1") == 0))
+    simd_support &= JSIMD_MMX;
+  env = getenv("JSIMD_FORCE3DNOW");
+  if ((env != NULL) && (strcmp(env, "1") == 0))
+    simd_support &= JSIMD_3DNOW;
+  env = getenv("JSIMD_FORCESSE");
+  if ((env != NULL) && (strcmp(env, "1") == 0))
+    simd_support &= JSIMD_SSE;
+  env = getenv("JSIMD_FORCESSE2");
+  if ((env != NULL) && (strcmp(env, "1") == 0))
+    simd_support &= JSIMD_SSE2;
+}
+
+GLOBAL(int)
+jsimd_can_rgb_ycc (void)
+{
+  init_simd();
+
+  /* The code is optimised for these values only */
+  if (BITS_IN_JSAMPLE != 8)
+    return 0;
+  if (sizeof(JDIMENSION) != 4)
+    return 0;
+  if ((RGB_PIXELSIZE != 3) && (RGB_PIXELSIZE != 4))
+    return 0;
+
+  if ((simd_support & JSIMD_SSE2) &&
+      IS_ALIGNED_SSE(jconst_rgb_ycc_convert_sse2))
+    return 1;
+  if (simd_support & JSIMD_MMX)
+    return 1;
+
+  return 0;
+}
+
+GLOBAL(int)
+jsimd_can_ycc_rgb (void)
+{
+  init_simd();
+
+  /* The code is optimised for these values only */
+  if (BITS_IN_JSAMPLE != 8)
+    return 0;
+  if (sizeof(JDIMENSION) != 4)
+    return 0;
+  if ((RGB_PIXELSIZE != 3) && (RGB_PIXELSIZE != 4))
+    return 0;
+
+  if ((simd_support & JSIMD_SSE2) &&
+      IS_ALIGNED_SSE(jconst_ycc_rgb_convert_sse2))
+    return 1;
+  if (simd_support & JSIMD_MMX)
+    return 1;
+
+  return 0;
+}
+
+GLOBAL(void)
+jsimd_rgb_ycc_convert (j_compress_ptr cinfo,
+                       JSAMPARRAY input_buf, JSAMPIMAGE output_buf,
+                       JDIMENSION output_row, int num_rows)
+{
+  void (*sse2fct)(JDIMENSION, JSAMPARRAY, JSAMPIMAGE, JDIMENSION, int);
+  void (*mmxfct)(JDIMENSION, JSAMPARRAY, JSAMPIMAGE, JDIMENSION, int);
+
+  switch(cinfo->in_color_space)
+  {
+    case JCS_EXT_RGB:
+      sse2fct=jsimd_extrgb_ycc_convert_sse2;
+      mmxfct=jsimd_extrgb_ycc_convert_mmx;
+      break;
+    case JCS_EXT_RGBX:
+      sse2fct=jsimd_extrgbx_ycc_convert_sse2;
+      mmxfct=jsimd_extrgbx_ycc_convert_mmx;
+      break;
+    case JCS_EXT_BGR:
+      sse2fct=jsimd_extbgr_ycc_convert_sse2;
+      mmxfct=jsimd_extbgr_ycc_convert_mmx;
+      break;
+    case JCS_EXT_BGRX:
+      sse2fct=jsimd_extbgrx_ycc_convert_sse2;
+      mmxfct=jsimd_extbgrx_ycc_convert_mmx;
+      break;
+    case JCS_EXT_XBGR:
+      sse2fct=jsimd_extxbgr_ycc_convert_sse2;
+      mmxfct=jsimd_extxbgr_ycc_convert_mmx;
+      break;
+    case JCS_EXT_XRGB:
+      sse2fct=jsimd_extxrgb_ycc_convert_sse2;
+      mmxfct=jsimd_extxrgb_ycc_convert_mmx;
+      break;
+    default:
+      sse2fct=jsimd_rgb_ycc_convert_sse2;
+      mmxfct=jsimd_rgb_ycc_convert_mmx;
+      break;
+  }
+
+  if ((simd_support & JSIMD_SSE2) &&
+      IS_ALIGNED_SSE(jconst_rgb_ycc_convert_sse2))
+    sse2fct(cinfo->image_width, input_buf,
+        output_buf, output_row, num_rows);
+  else if (simd_support & JSIMD_MMX)
+    mmxfct(cinfo->image_width, input_buf,
+        output_buf, output_row, num_rows);
+}
+
+GLOBAL(void)
+jsimd_ycc_rgb_convert (j_decompress_ptr cinfo,
+                       JSAMPIMAGE input_buf, JDIMENSION input_row,
+                       JSAMPARRAY output_buf, int num_rows)
+{
+  void (*sse2fct)(JDIMENSION, JSAMPIMAGE, JDIMENSION, JSAMPARRAY, int);
+  void (*mmxfct)(JDIMENSION, JSAMPIMAGE, JDIMENSION, JSAMPARRAY, int);
+
+  switch(cinfo->out_color_space)
+  {
+    case JCS_EXT_RGB:
+      sse2fct=jsimd_ycc_extrgb_convert_sse2;
+      mmxfct=jsimd_ycc_extrgb_convert_mmx;
+      break;
+    case JCS_EXT_RGBX:
+      sse2fct=jsimd_ycc_extrgbx_convert_sse2;
+      mmxfct=jsimd_ycc_extrgbx_convert_mmx;
+      break;
+    case JCS_EXT_BGR:
+      sse2fct=jsimd_ycc_extbgr_convert_sse2;
+      mmxfct=jsimd_ycc_extbgr_convert_mmx;
+      break;
+    case JCS_EXT_BGRX:
+      sse2fct=jsimd_ycc_extbgrx_convert_sse2;
+      mmxfct=jsimd_ycc_extbgrx_convert_mmx;
+      break;
+    case JCS_EXT_XBGR:
+      sse2fct=jsimd_ycc_extxbgr_convert_sse2;
+      mmxfct=jsimd_ycc_extxbgr_convert_mmx;
+      break;
+    case JCS_EXT_XRGB:
+      sse2fct=jsimd_ycc_extxrgb_convert_sse2;
+      mmxfct=jsimd_ycc_extxrgb_convert_mmx;
+      break;
+    default:
+      sse2fct=jsimd_ycc_rgb_convert_sse2;
+      mmxfct=jsimd_ycc_rgb_convert_mmx;
+      break;
+  }
+
+  if ((simd_support & JSIMD_SSE2) &&
+      IS_ALIGNED_SSE(jconst_ycc_rgb_convert_sse2))
+    sse2fct(cinfo->output_width, input_buf,
+        input_row, output_buf, num_rows);
+  else if (simd_support & JSIMD_MMX)
+    mmxfct(cinfo->output_width, input_buf,
+        input_row, output_buf, num_rows);
+}
+
+GLOBAL(int)
+jsimd_can_h2v2_downsample (void)
+{
+  init_simd();
+
+  /* The code is optimised for these values only */
+  if (BITS_IN_JSAMPLE != 8)
+    return 0;
+  if (sizeof(JDIMENSION) != 4)
+    return 0;
+
+  if (simd_support & JSIMD_SSE2)
+    return 1;
+  if (simd_support & JSIMD_MMX)
+    return 1;
+
+  return 0;
+}
+
+GLOBAL(int)
+jsimd_can_h2v1_downsample (void)
+{
+  init_simd();
+
+  /* The code is optimised for these values only */
+  if (BITS_IN_JSAMPLE != 8)
+    return 0;
+  if (sizeof(JDIMENSION) != 4)
+    return 0;
+
+  if (simd_support & JSIMD_SSE2)
+    return 1;
+  if (simd_support & JSIMD_MMX)
+    return 1;
+
+  return 0;
+}
+
+GLOBAL(void)
+jsimd_h2v2_downsample (j_compress_ptr cinfo, jpeg_component_info * compptr,
+                       JSAMPARRAY input_data, JSAMPARRAY output_data)
+{
+  if (simd_support & JSIMD_SSE2)
+    jsimd_h2v2_downsample_sse2(cinfo->image_width, cinfo->max_v_samp_factor,
+        compptr->v_samp_factor, compptr->width_in_blocks,
+        input_data, output_data);
+  else if (simd_support & JSIMD_MMX)
+    jsimd_h2v2_downsample_mmx(cinfo->image_width, cinfo->max_v_samp_factor,
+        compptr->v_samp_factor, compptr->width_in_blocks,
+        input_data, output_data);
+}
+
+GLOBAL(void)
+jsimd_h2v1_downsample (j_compress_ptr cinfo, jpeg_component_info * compptr,
+                       JSAMPARRAY input_data, JSAMPARRAY output_data)
+{
+  if (simd_support & JSIMD_SSE2)
+    jsimd_h2v1_downsample_sse2(cinfo->image_width, cinfo->max_v_samp_factor,
+        compptr->v_samp_factor, compptr->width_in_blocks,
+        input_data, output_data);
+  else if (simd_support & JSIMD_MMX)
+    jsimd_h2v1_downsample_mmx(cinfo->image_width, cinfo->max_v_samp_factor,
+        compptr->v_samp_factor, compptr->width_in_blocks,
+        input_data, output_data);
+}
+
+GLOBAL(int)
+jsimd_can_h2v2_upsample (void)
+{
+  init_simd();
+
+  /* The code is optimised for these values only */
+  if (BITS_IN_JSAMPLE != 8)
+    return 0;
+  if (sizeof(JDIMENSION) != 4)
+    return 0;
+
+  if (simd_support & JSIMD_SSE2)
+    return 1;
+  if (simd_support & JSIMD_MMX)
+    return 1;
+
+  return 0;
+}
+
+GLOBAL(int)
+jsimd_can_h2v1_upsample (void)
+{
+  init_simd();
+
+  /* The code is optimised for these values only */
+  if (BITS_IN_JSAMPLE != 8)
+    return 0;
+  if (sizeof(JDIMENSION) != 4)
+    return 0;
+
+  if (simd_support & JSIMD_SSE2)
+    return 1;
+  if (simd_support & JSIMD_MMX)
+    return 1;
+
+  return 0;
+}
+
+GLOBAL(void)
+jsimd_h2v2_upsample (j_decompress_ptr cinfo,
+                     jpeg_component_info * compptr, 
+                     JSAMPARRAY input_data,
+                     JSAMPARRAY * output_data_ptr)
+{
+  if (simd_support & JSIMD_SSE2)
+    jsimd_h2v2_upsample_sse2(cinfo->max_v_samp_factor,
+        cinfo->output_width, input_data, output_data_ptr);
+  else if (simd_support & JSIMD_MMX)
+    jsimd_h2v2_upsample_mmx(cinfo->max_v_samp_factor,
+        cinfo->output_width, input_data, output_data_ptr);
+}
+
+GLOBAL(void)
+jsimd_h2v1_upsample (j_decompress_ptr cinfo,
+                     jpeg_component_info * compptr, 
+                     JSAMPARRAY input_data,
+                     JSAMPARRAY * output_data_ptr)
+{
+  if (simd_support & JSIMD_SSE2)
+    jsimd_h2v1_upsample_sse2(cinfo->max_v_samp_factor,
+        cinfo->output_width, input_data, output_data_ptr);
+  else if (simd_support & JSIMD_MMX)
+    jsimd_h2v1_upsample_mmx(cinfo->max_v_samp_factor,
+        cinfo->output_width, input_data, output_data_ptr);
+}
+
+GLOBAL(int)
+jsimd_can_h2v2_fancy_upsample (void)
+{
+  init_simd();
+
+  /* The code is optimised for these values only */
+  if (BITS_IN_JSAMPLE != 8)
+    return 0;
+  if (sizeof(JDIMENSION) != 4)
+    return 0;
+
+  if ((simd_support & JSIMD_SSE2) &&
+      IS_ALIGNED_SSE(jconst_fancy_upsample_sse2))
+    return 1;
+  if (simd_support & JSIMD_MMX)
+    return 1;
+
+  return 0;
+}
+
+GLOBAL(int)
+jsimd_can_h2v1_fancy_upsample (void)
+{
+  init_simd();
+
+  /* The code is optimised for these values only */
+  if (BITS_IN_JSAMPLE != 8)
+    return 0;
+  if (sizeof(JDIMENSION) != 4)
+    return 0;
+
+  if ((simd_support & JSIMD_SSE2) &&
+      IS_ALIGNED_SSE(jconst_fancy_upsample_sse2))
+    return 1;
+  if (simd_support & JSIMD_MMX)
+    return 1;
+
+  return 0;
+}
+
+GLOBAL(void)
+jsimd_h2v2_fancy_upsample (j_decompress_ptr cinfo,
+                           jpeg_component_info * compptr, 
+                           JSAMPARRAY input_data,
+                           JSAMPARRAY * output_data_ptr)
+{
+  if ((simd_support & JSIMD_SSE2) &&
+      IS_ALIGNED_SSE(jconst_fancy_upsample_sse2))
+    jsimd_h2v1_fancy_upsample_sse2(cinfo->max_v_samp_factor,
+        compptr->downsampled_width, input_data, output_data_ptr);
+  else if (simd_support & JSIMD_MMX)
+    jsimd_h2v2_fancy_upsample_mmx(cinfo->max_v_samp_factor,
+        compptr->downsampled_width, input_data, output_data_ptr);
+}
+
+GLOBAL(void)
+jsimd_h2v1_fancy_upsample (j_decompress_ptr cinfo,
+                           jpeg_component_info * compptr, 
+                           JSAMPARRAY input_data,
+                           JSAMPARRAY * output_data_ptr)
+{
+  if ((simd_support & JSIMD_SSE2) &&
+      IS_ALIGNED_SSE(jconst_fancy_upsample_sse2))
+    jsimd_h2v1_fancy_upsample_sse2(cinfo->max_v_samp_factor,
+        compptr->downsampled_width, input_data, output_data_ptr);
+  else if (simd_support & JSIMD_MMX)
+    jsimd_h2v1_fancy_upsample_mmx(cinfo->max_v_samp_factor,
+        compptr->downsampled_width, input_data, output_data_ptr);
+}
+
+GLOBAL(int)
+jsimd_can_h2v2_merged_upsample (void)
+{
+  init_simd();
+
+  /* The code is optimised for these values only */
+  if (BITS_IN_JSAMPLE != 8)
+    return 0;
+  if (sizeof(JDIMENSION) != 4)
+    return 0;
+
+  if ((simd_support & JSIMD_SSE2) &&
+      IS_ALIGNED_SSE(jconst_merged_upsample_sse2))
+    return 1;
+  if (simd_support & JSIMD_MMX)
+    return 1;
+
+  return 0;
+}
+
+GLOBAL(int)
+jsimd_can_h2v1_merged_upsample (void)
+{
+  init_simd();
+
+  /* The code is optimised for these values only */
+  if (BITS_IN_JSAMPLE != 8)
+    return 0;
+  if (sizeof(JDIMENSION) != 4)
+    return 0;
+
+  if ((simd_support & JSIMD_SSE2) &&
+      IS_ALIGNED_SSE(jconst_merged_upsample_sse2))
+    return 1;
+  if (simd_support & JSIMD_MMX)
+    return 1;
+
+  return 0;
+}
+
+GLOBAL(void)
+jsimd_h2v2_merged_upsample (j_decompress_ptr cinfo,
+                            JSAMPIMAGE input_buf,
+                            JDIMENSION in_row_group_ctr,
+                            JSAMPARRAY output_buf)
+{
+  void (*sse2fct)(JDIMENSION, JSAMPIMAGE, JDIMENSION, JSAMPARRAY);
+  void (*mmxfct)(JDIMENSION, JSAMPIMAGE, JDIMENSION, JSAMPARRAY);
+
+  switch(cinfo->out_color_space)
+  {
+    case JCS_EXT_RGB:
+      sse2fct=jsimd_h2v2_extrgb_merged_upsample_sse2;
+      mmxfct=jsimd_h2v2_extrgb_merged_upsample_mmx;
+      break;
+    case JCS_EXT_RGBX:
+      sse2fct=jsimd_h2v2_extrgbx_merged_upsample_sse2;
+      mmxfct=jsimd_h2v2_extrgbx_merged_upsample_mmx;
+      break;
+    case JCS_EXT_BGR:
+      sse2fct=jsimd_h2v2_extbgr_merged_upsample_sse2;
+      mmxfct=jsimd_h2v2_extbgr_merged_upsample_mmx;
+      break;
+    case JCS_EXT_BGRX:
+      sse2fct=jsimd_h2v2_extbgrx_merged_upsample_sse2;
+      mmxfct=jsimd_h2v2_extbgrx_merged_upsample_mmx;
+      break;
+    case JCS_EXT_XBGR:
+      sse2fct=jsimd_h2v2_extxbgr_merged_upsample_sse2;
+      mmxfct=jsimd_h2v2_extxbgr_merged_upsample_mmx;
+      break;
+    case JCS_EXT_XRGB:
+      sse2fct=jsimd_h2v2_extxrgb_merged_upsample_sse2;
+      mmxfct=jsimd_h2v2_extxrgb_merged_upsample_mmx;
+      break;
+    default:
+      sse2fct=jsimd_h2v2_merged_upsample_sse2;
+      mmxfct=jsimd_h2v2_merged_upsample_mmx;
+      break;
+  }
+
+  if ((simd_support & JSIMD_SSE2) &&
+      IS_ALIGNED_SSE(jconst_merged_upsample_sse2))
+    sse2fct(cinfo->output_width, input_buf,
+        in_row_group_ctr, output_buf);
+  else if (simd_support & JSIMD_MMX)
+    mmxfct(cinfo->output_width, input_buf,
+        in_row_group_ctr, output_buf);
+}
+
+GLOBAL(void)
+jsimd_h2v1_merged_upsample (j_decompress_ptr cinfo,
+                            JSAMPIMAGE input_buf,
+                            JDIMENSION in_row_group_ctr,
+                            JSAMPARRAY output_buf)
+{
+  void (*sse2fct)(JDIMENSION, JSAMPIMAGE, JDIMENSION, JSAMPARRAY);
+  void (*mmxfct)(JDIMENSION, JSAMPIMAGE, JDIMENSION, JSAMPARRAY);
+
+  switch(cinfo->out_color_space)
+  {
+    case JCS_EXT_RGB:
+      sse2fct=jsimd_h2v1_extrgb_merged_upsample_sse2;
+      mmxfct=jsimd_h2v1_extrgb_merged_upsample_mmx;
+      break;
+    case JCS_EXT_RGBX:
+      sse2fct=jsimd_h2v1_extrgbx_merged_upsample_sse2;
+      mmxfct=jsimd_h2v1_extrgbx_merged_upsample_mmx;
+      break;
+    case JCS_EXT_BGR:
+      sse2fct=jsimd_h2v1_extbgr_merged_upsample_sse2;
+      mmxfct=jsimd_h2v1_extbgr_merged_upsample_mmx;
+      break;
+    case JCS_EXT_BGRX:
+      sse2fct=jsimd_h2v1_extbgrx_merged_upsample_sse2;
+      mmxfct=jsimd_h2v1_extbgrx_merged_upsample_mmx;
+      break;
+    case JCS_EXT_XBGR:
+      sse2fct=jsimd_h2v1_extxbgr_merged_upsample_sse2;
+      mmxfct=jsimd_h2v1_extxbgr_merged_upsample_mmx;
+      break;
+    case JCS_EXT_XRGB:
+      sse2fct=jsimd_h2v1_extxrgb_merged_upsample_sse2;
+      mmxfct=jsimd_h2v1_extxrgb_merged_upsample_mmx;
+      break;
+    default:
+      sse2fct=jsimd_h2v1_merged_upsample_sse2;
+      mmxfct=jsimd_h2v1_merged_upsample_mmx;
+      break;
+  }
+
+  if ((simd_support & JSIMD_SSE2) &&
+      IS_ALIGNED_SSE(jconst_merged_upsample_sse2))
+    sse2fct(cinfo->output_width, input_buf,
+        in_row_group_ctr, output_buf);
+  else if (simd_support & JSIMD_MMX)
+    mmxfct(cinfo->output_width, input_buf,
+        in_row_group_ctr, output_buf);
+}
+
+GLOBAL(int)
+jsimd_can_convsamp (void)
+{
+  init_simd();
+
+  /* The code is optimised for these values only */
+  if (DCTSIZE != 8)
+    return 0;
+  if (BITS_IN_JSAMPLE != 8)
+    return 0;
+  if (sizeof(JDIMENSION) != 4)
+    return 0;
+  if (sizeof(DCTELEM) != 2)
+    return 0;
+
+  if (simd_support & JSIMD_SSE2)
+    return 1;
+  if (simd_support & JSIMD_MMX)
+    return 1;
+
+  return 0;
+}
+
+GLOBAL(int)
+jsimd_can_convsamp_float (void)
+{
+  init_simd();
+
+  /* The code is optimised for these values only */
+  if (DCTSIZE != 8)
+    return 0;
+  if (BITS_IN_JSAMPLE != 8)
+    return 0;
+  if (sizeof(JDIMENSION) != 4)
+    return 0;
+  if (sizeof(FAST_FLOAT) != 4)
+    return 0;
+
+  if (simd_support & JSIMD_SSE2)
+    return 1;
+  if (simd_support & JSIMD_SSE)
+    return 1;
+  if (simd_support & JSIMD_3DNOW)
+    return 1;
+
+  return 0;
+}
+
+GLOBAL(void)
+jsimd_convsamp (JSAMPARRAY sample_data, JDIMENSION start_col,
+                DCTELEM * workspace)
+{
+  if (simd_support & JSIMD_SSE2)
+    jsimd_convsamp_sse2(sample_data, start_col, workspace);
+  else if (simd_support & JSIMD_MMX)
+    jsimd_convsamp_mmx(sample_data, start_col, workspace);
+}
+
+GLOBAL(void)
+jsimd_convsamp_float (JSAMPARRAY sample_data, JDIMENSION start_col,
+                      FAST_FLOAT * workspace)
+{
+  if (simd_support & JSIMD_SSE2)
+    jsimd_convsamp_float_sse2(sample_data, start_col, workspace);
+  else if (simd_support & JSIMD_SSE)
+    jsimd_convsamp_float_sse(sample_data, start_col, workspace);
+  else if (simd_support & JSIMD_3DNOW)
+    jsimd_convsamp_float_3dnow(sample_data, start_col, workspace);
+}
+
+GLOBAL(int)
+jsimd_can_fdct_islow (void)
+{
+  init_simd();
+
+  /* The code is optimised for these values only */
+  if (DCTSIZE != 8)
+    return 0;
+  if (sizeof(DCTELEM) != 2)
+    return 0;
+
+  if ((simd_support & JSIMD_SSE2) && IS_ALIGNED_SSE(jconst_fdct_islow_sse2))
+    return 1;
+  if (simd_support & JSIMD_MMX)
+    return 1;
+
+  return 0;
+}
+
+GLOBAL(int)
+jsimd_can_fdct_ifast (void)
+{
+  init_simd();
+
+  /* The code is optimised for these values only */
+  if (DCTSIZE != 8)
+    return 0;
+  if (sizeof(DCTELEM) != 2)
+    return 0;
+
+  if ((simd_support & JSIMD_SSE2) && IS_ALIGNED_SSE(jconst_fdct_ifast_sse2))
+    return 1;
+  if (simd_support & JSIMD_MMX)
+    return 1;
+
+  return 0;
+}
+
+GLOBAL(int)
+jsimd_can_fdct_float (void)
+{
+  init_simd();
+
+  /* The code is optimised for these values only */
+  if (DCTSIZE != 8)
+    return 0;
+  if (sizeof(FAST_FLOAT) != 4)
+    return 0;
+
+  if ((simd_support & JSIMD_SSE) && IS_ALIGNED_SSE(jconst_fdct_float_sse))
+    return 1;
+  if (simd_support & JSIMD_3DNOW)
+    return 1;
+
+  return 0;
+}
+
+GLOBAL(void)
+jsimd_fdct_islow (DCTELEM * data)
+{
+  if ((simd_support & JSIMD_SSE2) && IS_ALIGNED_SSE(jconst_fdct_islow_sse2))
+    jsimd_fdct_islow_sse2(data);
+  else if (simd_support & JSIMD_MMX)
+    jsimd_fdct_islow_mmx(data);
+}
+
+GLOBAL(void)
+jsimd_fdct_ifast (DCTELEM * data)
+{
+  if ((simd_support & JSIMD_SSE2) && IS_ALIGNED_SSE(jconst_fdct_islow_sse2))
+    jsimd_fdct_ifast_sse2(data);
+  else if (simd_support & JSIMD_MMX)
+    jsimd_fdct_ifast_mmx(data);
+}
+
+GLOBAL(void)
+jsimd_fdct_float (FAST_FLOAT * data)
+{
+  if ((simd_support & JSIMD_SSE) && IS_ALIGNED_SSE(jconst_fdct_float_sse))
+    jsimd_fdct_float_sse(data);
+  else if (simd_support & JSIMD_3DNOW)
+    jsimd_fdct_float_3dnow(data);
+}
+
+GLOBAL(int)
+jsimd_can_quantize (void)
+{
+  init_simd();
+
+  /* The code is optimised for these values only */
+  if (DCTSIZE != 8)
+    return 0;
+  if (sizeof(JCOEF) != 2)
+    return 0;
+  if (sizeof(DCTELEM) != 2)
+    return 0;
+
+  if (simd_support & JSIMD_SSE2)
+    return 1;
+  if (simd_support & JSIMD_MMX)
+    return 1;
+
+  return 0;
+}
+
+GLOBAL(int)
+jsimd_can_quantize_float (void)
+{
+  init_simd();
+
+  /* The code is optimised for these values only */
+  if (DCTSIZE != 8)
+    return 0;
+  if (sizeof(JCOEF) != 2)
+    return 0;
+  if (sizeof(FAST_FLOAT) != 4)
+    return 0;
+
+  if (simd_support & JSIMD_SSE2)
+    return 1;
+  if (simd_support & JSIMD_SSE)
+    return 1;
+  if (simd_support & JSIMD_3DNOW)
+    return 1;
+
+  return 0;
+}
+
+GLOBAL(void)
+jsimd_quantize (JCOEFPTR coef_block, DCTELEM * divisors,
+                DCTELEM * workspace)
+{
+  if (simd_support & JSIMD_SSE2)
+    jsimd_quantize_sse2(coef_block, divisors, workspace);
+  else if (simd_support & JSIMD_MMX)
+    jsimd_quantize_mmx(coef_block, divisors, workspace);
+}
+
+GLOBAL(void)
+jsimd_quantize_float (JCOEFPTR coef_block, FAST_FLOAT * divisors,
+                      FAST_FLOAT * workspace)
+{
+  if (simd_support & JSIMD_SSE2)
+    jsimd_quantize_float_sse2(coef_block, divisors, workspace);
+  else if (simd_support & JSIMD_SSE)
+    jsimd_quantize_float_sse(coef_block, divisors, workspace);
+  else if (simd_support & JSIMD_3DNOW)
+    jsimd_quantize_float_3dnow(coef_block, divisors, workspace);
+}
+
+GLOBAL(int)
+jsimd_can_idct_2x2 (void)
+{
+  init_simd();
+
+  /* The code is optimised for these values only */
+  if (DCTSIZE != 8)
+    return 0;
+  if (sizeof(JCOEF) != 2)
+    return 0;
+  if (BITS_IN_JSAMPLE != 8)
+    return 0;
+  if (sizeof(JDIMENSION) != 4)
+    return 0;
+  if (sizeof(ISLOW_MULT_TYPE) != 2)
+    return 0;
+
+  if ((simd_support & JSIMD_SSE2) && IS_ALIGNED_SSE(jconst_idct_red_sse2))
+    return 1;
+  if (simd_support & JSIMD_MMX)
+    return 1;
+
+  return 0;
+}
+
+GLOBAL(int)
+jsimd_can_idct_4x4 (void)
+{
+  init_simd();
+
+  /* The code is optimised for these values only */
+  if (DCTSIZE != 8)
+    return 0;
+  if (sizeof(JCOEF) != 2)
+    return 0;
+  if (BITS_IN_JSAMPLE != 8)
+    return 0;
+  if (sizeof(JDIMENSION) != 4)
+    return 0;
+  if (sizeof(ISLOW_MULT_TYPE) != 2)
+    return 0;
+
+  if ((simd_support & JSIMD_SSE2) && IS_ALIGNED_SSE(jconst_idct_red_sse2))
+    return 1;
+  if (simd_support & JSIMD_MMX)
+    return 1;
+
+  return 0;
+}
+
+GLOBAL(void)
+jsimd_idct_2x2 (j_decompress_ptr cinfo, jpeg_component_info * compptr,
+                JCOEFPTR coef_block, JSAMPARRAY output_buf,
+                JDIMENSION output_col)
+{
+  if ((simd_support & JSIMD_SSE2) && IS_ALIGNED_SSE(jconst_idct_red_sse2))
+    jsimd_idct_2x2_sse2(compptr->dct_table, coef_block, output_buf, output_col);
+  else if (simd_support & JSIMD_MMX)
+    jsimd_idct_2x2_mmx(compptr->dct_table, coef_block, output_buf, output_col);
+}
+
+GLOBAL(void)
+jsimd_idct_4x4 (j_decompress_ptr cinfo, jpeg_component_info * compptr,
+                JCOEFPTR coef_block, JSAMPARRAY output_buf,
+                JDIMENSION output_col)
+{
+  if ((simd_support & JSIMD_SSE2) && IS_ALIGNED_SSE(jconst_idct_red_sse2))
+    jsimd_idct_4x4_sse2(compptr->dct_table, coef_block, output_buf, output_col);
+  else if (simd_support & JSIMD_MMX)
+    jsimd_idct_4x4_mmx(compptr->dct_table, coef_block, output_buf, output_col);
+}
+
+GLOBAL(int)
+jsimd_can_idct_islow (void)
+{
+  init_simd();
+
+  /* The code is optimised for these values only */
+  if (DCTSIZE != 8)
+    return 0;
+  if (sizeof(JCOEF) != 2)
+    return 0;
+  if (BITS_IN_JSAMPLE != 8)
+    return 0;
+  if (sizeof(JDIMENSION) != 4)
+    return 0;
+  if (sizeof(ISLOW_MULT_TYPE) != 2)
+    return 0;
+
+  if ((simd_support & JSIMD_SSE2) && IS_ALIGNED_SSE(jconst_idct_islow_sse2))
+    return 1;
+  if (simd_support & JSIMD_MMX)
+    return 1;
+
+  return 0;
+}
+
+GLOBAL(int)
+jsimd_can_idct_ifast (void)
+{
+  init_simd();
+
+  /* The code is optimised for these values only */
+  if (DCTSIZE != 8)
+    return 0;
+  if (sizeof(JCOEF) != 2)
+    return 0;
+  if (BITS_IN_JSAMPLE != 8)
+    return 0;
+  if (sizeof(JDIMENSION) != 4)
+    return 0;
+  if (sizeof(IFAST_MULT_TYPE) != 2)
+    return 0;
+  if (IFAST_SCALE_BITS != 2)
+    return 0;
+
+  if ((simd_support & JSIMD_SSE2) && IS_ALIGNED_SSE(jconst_idct_ifast_sse2))
+    return 1;
+  if (simd_support & JSIMD_MMX)
+    return 1;
+
+  return 0;
+}
+
+GLOBAL(int)
+jsimd_can_idct_float (void)
+{
+  init_simd();
+
+  if (DCTSIZE != 8)
+    return 0;
+  if (sizeof(JCOEF) != 2)
+    return 0;
+  if (BITS_IN_JSAMPLE != 8)
+    return 0;
+  if (sizeof(JDIMENSION) != 4)
+    return 0;
+  if (sizeof(FAST_FLOAT) != 4)
+    return 0;
+  if (sizeof(FLOAT_MULT_TYPE) != 4)
+    return 0;
+
+  if ((simd_support & JSIMD_SSE2) && IS_ALIGNED_SSE(jconst_idct_float_sse2))
+    return 1;
+  if ((simd_support & JSIMD_SSE) && IS_ALIGNED_SSE(jconst_idct_float_sse))
+    return 1;
+  if (simd_support & JSIMD_3DNOW)
+    return 1;
+
+  return 0;
+}
+
+GLOBAL(void)
+jsimd_idct_islow (j_decompress_ptr cinfo, jpeg_component_info * compptr,
+                JCOEFPTR coef_block, JSAMPARRAY output_buf,
+                JDIMENSION output_col)
+{
+  if ((simd_support & JSIMD_SSE2) && IS_ALIGNED_SSE(jconst_idct_islow_sse2))
+    jsimd_idct_islow_sse2(compptr->dct_table, coef_block, output_buf, output_col);
+  else if (simd_support & JSIMD_MMX)
+    jsimd_idct_islow_mmx(compptr->dct_table, coef_block, output_buf, output_col);
+}
+
+GLOBAL(void)
+jsimd_idct_ifast (j_decompress_ptr cinfo, jpeg_component_info * compptr,
+                JCOEFPTR coef_block, JSAMPARRAY output_buf,
+                JDIMENSION output_col)
+{
+  if ((simd_support & JSIMD_SSE2) && IS_ALIGNED_SSE(jconst_idct_ifast_sse2))
+    jsimd_idct_ifast_sse2(compptr->dct_table, coef_block, output_buf, output_col);
+  else if (simd_support & JSIMD_MMX)
+    jsimd_idct_ifast_mmx(compptr->dct_table, coef_block, output_buf, output_col);
+}
+
+GLOBAL(void)
+jsimd_idct_float (j_decompress_ptr cinfo, jpeg_component_info * compptr,
+                JCOEFPTR coef_block, JSAMPARRAY output_buf,
+                JDIMENSION output_col)
+{
+  if ((simd_support & JSIMD_SSE2) && IS_ALIGNED_SSE(jconst_idct_float_sse2))
+    jsimd_idct_float_sse2(compptr->dct_table, coef_block,
+        output_buf, output_col);
+  else if ((simd_support & JSIMD_SSE) && IS_ALIGNED_SSE(jconst_idct_float_sse))
+    jsimd_idct_float_sse(compptr->dct_table, coef_block,
+        output_buf, output_col);
+  else if (simd_support & JSIMD_3DNOW)
+    jsimd_idct_float_3dnow(compptr->dct_table, coef_block,
+        output_buf, output_col);
+}
+
diff --git a/simd/jsimd_x86_64.c b/simd/jsimd_x86_64.c
new file mode 100644
index 0000000..4c1ef7d
--- /dev/null
+++ b/simd/jsimd_x86_64.c
@@ -0,0 +1,680 @@
+/*
+ * jsimd_x86_64.c
+ *
+ * Copyright 2009 Pierre Ossman <ossman@cendio.se> for Cendio AB
+ * Copyright 2009 D. R. Commander
+ * 
+ * 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 when running on a
+ * x86_64 architecture.
+ */
+
+#define JPEG_INTERNALS
+#include "../jinclude.h"
+#include "../jpeglib.h"
+#include "../jsimd.h"
+#include "../jdct.h"
+#include "../jsimddct.h"
+#include "simd/jsimd.h"
+
+/*
+ * In the PIC cases, we have no guarantee that constants will keep
+ * their alignment. This macro allows us to verify it at runtime.
+ */
+#define IS_ALIGNED(ptr, order) (((unsigned)ptr & ((1 << order) - 1)) == 0)
+
+#define IS_ALIGNED_SSE(ptr) (IS_ALIGNED(ptr, 4)) /* 16 byte alignment */
+
+GLOBAL(int)
+jsimd_can_rgb_ycc (void)
+{
+  /* The code is optimised for these values only */
+  if (BITS_IN_JSAMPLE != 8)
+    return 0;
+  if (sizeof(JDIMENSION) != 4)
+    return 0;
+  if ((RGB_PIXELSIZE != 3) && (RGB_PIXELSIZE != 4))
+    return 0;
+
+  if (!IS_ALIGNED_SSE(jconst_rgb_ycc_convert_sse2))
+    return 0;
+
+  return 1;
+}
+
+GLOBAL(int)
+jsimd_can_ycc_rgb (void)
+{
+  /* The code is optimised for these values only */
+  if (BITS_IN_JSAMPLE != 8)
+    return 0;
+  if (sizeof(JDIMENSION) != 4)
+    return 0;
+  if ((RGB_PIXELSIZE != 3) && (RGB_PIXELSIZE != 4))
+    return 0;
+
+  if (!IS_ALIGNED_SSE(jconst_ycc_rgb_convert_sse2))
+    return 0;
+
+  return 1;
+}
+
+GLOBAL(void)
+jsimd_rgb_ycc_convert (j_compress_ptr cinfo,
+                       JSAMPARRAY input_buf, JSAMPIMAGE output_buf,
+                       JDIMENSION output_row, int num_rows)
+{
+  void (*sse2fct)(JDIMENSION, JSAMPARRAY, JSAMPIMAGE, JDIMENSION, int);
+
+  switch(cinfo->in_color_space)
+  {
+    case JCS_EXT_RGB:
+      sse2fct=jsimd_extrgb_ycc_convert_sse2;
+      break;
+    case JCS_EXT_RGBX:
+      sse2fct=jsimd_extrgbx_ycc_convert_sse2;
+      break;
+    case JCS_EXT_BGR:
+      sse2fct=jsimd_extbgr_ycc_convert_sse2;
+      break;
+    case JCS_EXT_BGRX:
+      sse2fct=jsimd_extbgrx_ycc_convert_sse2;
+      break;
+    case JCS_EXT_XBGR:
+      sse2fct=jsimd_extxbgr_ycc_convert_sse2;
+      break;
+    case JCS_EXT_XRGB:
+      sse2fct=jsimd_extxrgb_ycc_convert_sse2;
+      break;
+    default:
+      sse2fct=jsimd_rgb_ycc_convert_sse2;
+      break;
+  }
+
+  sse2fct(cinfo->image_width, input_buf, output_buf, output_row, num_rows);
+}
+
+GLOBAL(void)
+jsimd_ycc_rgb_convert (j_decompress_ptr cinfo,
+                       JSAMPIMAGE input_buf, JDIMENSION input_row,
+                       JSAMPARRAY output_buf, int num_rows)
+{
+  void (*sse2fct)(JDIMENSION, JSAMPIMAGE, JDIMENSION, JSAMPARRAY, int);
+
+  switch(cinfo->out_color_space)
+  {
+    case JCS_EXT_RGB:
+      sse2fct=jsimd_ycc_extrgb_convert_sse2;
+      break;
+    case JCS_EXT_RGBX:
+      sse2fct=jsimd_ycc_extrgbx_convert_sse2;
+      break;
+    case JCS_EXT_BGR:
+      sse2fct=jsimd_ycc_extbgr_convert_sse2;
+      break;
+    case JCS_EXT_BGRX:
+      sse2fct=jsimd_ycc_extbgrx_convert_sse2;
+      break;
+    case JCS_EXT_XBGR:
+      sse2fct=jsimd_ycc_extxbgr_convert_sse2;
+      break;
+    case JCS_EXT_XRGB:
+      sse2fct=jsimd_ycc_extxrgb_convert_sse2;
+      break;
+    default:
+      sse2fct=jsimd_ycc_rgb_convert_sse2;
+      break;
+  }
+
+  sse2fct(cinfo->output_width, input_buf, input_row, output_buf, num_rows);
+}
+
+GLOBAL(int)
+jsimd_can_h2v2_downsample (void)
+{
+  /* The code is optimised for these values only */
+  if (BITS_IN_JSAMPLE != 8)
+    return 0;
+  if (sizeof(JDIMENSION) != 4)
+    return 0;
+
+  return 1;
+}
+
+GLOBAL(int)
+jsimd_can_h2v1_downsample (void)
+{
+  /* The code is optimised for these values only */
+  if (BITS_IN_JSAMPLE != 8)
+    return 0;
+  if (sizeof(JDIMENSION) != 4)
+    return 0;
+
+  return 1;
+}
+
+GLOBAL(void)
+jsimd_h2v2_downsample (j_compress_ptr cinfo, jpeg_component_info * compptr,
+                       JSAMPARRAY input_data, JSAMPARRAY output_data)
+{
+  jsimd_h2v2_downsample_sse2(cinfo->image_width,
+                             cinfo->max_v_samp_factor,
+                             compptr->v_samp_factor,
+                             compptr->width_in_blocks,
+                             input_data, output_data);
+}
+
+GLOBAL(void)
+jsimd_h2v1_downsample (j_compress_ptr cinfo, jpeg_component_info * compptr,
+                       JSAMPARRAY input_data, JSAMPARRAY output_data)
+{
+  jsimd_h2v1_downsample_sse2(cinfo->image_width,
+                             cinfo->max_v_samp_factor,
+                             compptr->v_samp_factor,
+                             compptr->width_in_blocks,
+                             input_data, output_data);
+}
+
+GLOBAL(int)
+jsimd_can_h2v2_upsample (void)
+{
+  /* The code is optimised for these values only */
+  if (BITS_IN_JSAMPLE != 8)
+    return 0;
+  if (sizeof(JDIMENSION) != 4)
+    return 0;
+
+  return 1;
+}
+
+GLOBAL(int)
+jsimd_can_h2v1_upsample (void)
+{
+  /* The code is optimised for these values only */
+  if (BITS_IN_JSAMPLE != 8)
+    return 0;
+  if (sizeof(JDIMENSION) != 4)
+    return 0;
+
+  return 1;
+}
+
+GLOBAL(void)
+jsimd_h2v2_upsample (j_decompress_ptr cinfo,
+                     jpeg_component_info * compptr, 
+                     JSAMPARRAY input_data,
+                     JSAMPARRAY * output_data_ptr)
+{
+  jsimd_h2v2_upsample_sse2(cinfo->max_v_samp_factor,
+                           cinfo->output_width,
+                           input_data, output_data_ptr);
+}
+
+GLOBAL(void)
+jsimd_h2v1_upsample (j_decompress_ptr cinfo,
+                     jpeg_component_info * compptr, 
+                     JSAMPARRAY input_data,
+                     JSAMPARRAY * output_data_ptr)
+{
+  jsimd_h2v1_upsample_sse2(cinfo->max_v_samp_factor,
+                           cinfo->output_width,
+                           input_data, output_data_ptr);
+}
+
+GLOBAL(int)
+jsimd_can_h2v2_fancy_upsample (void)
+{
+  /* The code is optimised for these values only */
+  if (BITS_IN_JSAMPLE != 8)
+    return 0;
+  if (sizeof(JDIMENSION) != 4)
+    return 0;
+
+  if (!IS_ALIGNED_SSE(jconst_fancy_upsample_sse2))
+    return 0;
+
+  return 1;
+}
+
+GLOBAL(int)
+jsimd_can_h2v1_fancy_upsample (void)
+{
+  /* The code is optimised for these values only */
+  if (BITS_IN_JSAMPLE != 8)
+    return 0;
+  if (sizeof(JDIMENSION) != 4)
+    return 0;
+
+  if (!IS_ALIGNED_SSE(jconst_fancy_upsample_sse2))
+    return 0;
+
+  return 1;
+}
+
+GLOBAL(void)
+jsimd_h2v2_fancy_upsample (j_decompress_ptr cinfo,
+                           jpeg_component_info * compptr, 
+                           JSAMPARRAY input_data,
+                           JSAMPARRAY * output_data_ptr)
+{
+  jsimd_h2v1_fancy_upsample_sse2(cinfo->max_v_samp_factor,
+                                 compptr->downsampled_width,
+                                 input_data, output_data_ptr);
+}
+
+GLOBAL(void)
+jsimd_h2v1_fancy_upsample (j_decompress_ptr cinfo,
+                           jpeg_component_info * compptr, 
+                           JSAMPARRAY input_data,
+                           JSAMPARRAY * output_data_ptr)
+{
+  jsimd_h2v1_fancy_upsample_sse2(cinfo->max_v_samp_factor,
+                                 compptr->downsampled_width,
+                                 input_data, output_data_ptr);
+}
+
+GLOBAL(int)
+jsimd_can_h2v2_merged_upsample (void)
+{
+  /* The code is optimised for these values only */
+  if (BITS_IN_JSAMPLE != 8)
+    return 0;
+  if (sizeof(JDIMENSION) != 4)
+    return 0;
+
+  if (!IS_ALIGNED_SSE(jconst_merged_upsample_sse2))
+    return 0;
+
+  return 1;
+}
+
+GLOBAL(int)
+jsimd_can_h2v1_merged_upsample (void)
+{
+  /* The code is optimised for these values only */
+  if (BITS_IN_JSAMPLE != 8)
+    return 0;
+  if (sizeof(JDIMENSION) != 4)
+    return 0;
+
+  if (!IS_ALIGNED_SSE(jconst_merged_upsample_sse2))
+    return 0;
+
+  return 1;
+}
+
+GLOBAL(void)
+jsimd_h2v2_merged_upsample (j_decompress_ptr cinfo,
+                            JSAMPIMAGE input_buf,
+                            JDIMENSION in_row_group_ctr,
+                            JSAMPARRAY output_buf)
+{
+  void (*sse2fct)(JDIMENSION, JSAMPIMAGE, JDIMENSION, JSAMPARRAY);
+
+  switch(cinfo->out_color_space)
+  {
+    case JCS_EXT_RGB:
+      sse2fct=jsimd_h2v2_extrgb_merged_upsample_sse2;
+      break;
+    case JCS_EXT_RGBX:
+      sse2fct=jsimd_h2v2_extrgbx_merged_upsample_sse2;
+      break;
+    case JCS_EXT_BGR:
+      sse2fct=jsimd_h2v2_extbgr_merged_upsample_sse2;
+      break;
+    case JCS_EXT_BGRX:
+      sse2fct=jsimd_h2v2_extbgrx_merged_upsample_sse2;
+      break;
+    case JCS_EXT_XBGR:
+      sse2fct=jsimd_h2v2_extxbgr_merged_upsample_sse2;
+      break;
+    case JCS_EXT_XRGB:
+      sse2fct=jsimd_h2v2_extxrgb_merged_upsample_sse2;
+      break;
+    default:
+      sse2fct=jsimd_h2v2_merged_upsample_sse2;
+      break;
+  }
+
+  sse2fct(cinfo->output_width, input_buf, in_row_group_ctr, output_buf);
+}
+
+GLOBAL(void)
+jsimd_h2v1_merged_upsample (j_decompress_ptr cinfo,
+                            JSAMPIMAGE input_buf,
+                            JDIMENSION in_row_group_ctr,
+                            JSAMPARRAY output_buf)
+{
+  void (*sse2fct)(JDIMENSION, JSAMPIMAGE, JDIMENSION, JSAMPARRAY);
+
+  switch(cinfo->out_color_space)
+  {
+    case JCS_EXT_RGB:
+      sse2fct=jsimd_h2v1_extrgb_merged_upsample_sse2;
+      break;
+    case JCS_EXT_RGBX:
+      sse2fct=jsimd_h2v1_extrgbx_merged_upsample_sse2;
+      break;
+    case JCS_EXT_BGR:
+      sse2fct=jsimd_h2v1_extbgr_merged_upsample_sse2;
+      break;
+    case JCS_EXT_BGRX:
+      sse2fct=jsimd_h2v1_extbgrx_merged_upsample_sse2;
+      break;
+    case JCS_EXT_XBGR:
+      sse2fct=jsimd_h2v1_extxbgr_merged_upsample_sse2;
+      break;
+    case JCS_EXT_XRGB:
+      sse2fct=jsimd_h2v1_extxrgb_merged_upsample_sse2;
+      break;
+    default:
+      sse2fct=jsimd_h2v1_merged_upsample_sse2;
+      break;
+  }
+
+  sse2fct(cinfo->output_width, input_buf, in_row_group_ctr, output_buf);
+}
+
+GLOBAL(int)
+jsimd_can_convsamp (void)
+{
+  /* The code is optimised for these values only */
+  if (DCTSIZE != 8)
+    return 0;
+  if (BITS_IN_JSAMPLE != 8)
+    return 0;
+  if (sizeof(JDIMENSION) != 4)
+    return 0;
+  if (sizeof(DCTELEM) != 2)
+    return 0;
+
+  return 1;
+}
+
+GLOBAL(int)
+jsimd_can_convsamp_float (void)
+{
+  /* The code is optimised for these values only */
+  if (DCTSIZE != 8)
+    return 0;
+  if (BITS_IN_JSAMPLE != 8)
+    return 0;
+  if (sizeof(JDIMENSION) != 4)
+    return 0;
+  if (sizeof(FAST_FLOAT) != 4)
+    return 0;
+
+  return 1;
+}
+
+GLOBAL(void)
+jsimd_convsamp (JSAMPARRAY sample_data, JDIMENSION start_col,
+                DCTELEM * workspace)
+{
+  jsimd_convsamp_sse2(sample_data, start_col, workspace);
+}
+
+GLOBAL(void)
+jsimd_convsamp_float (JSAMPARRAY sample_data, JDIMENSION start_col,
+                      FAST_FLOAT * workspace)
+{
+  jsimd_convsamp_float_sse2(sample_data, start_col, workspace);
+}
+
+GLOBAL(int)
+jsimd_can_fdct_islow (void)
+{
+  /* The code is optimised for these values only */
+  if (DCTSIZE != 8)
+    return 0;
+  if (sizeof(DCTELEM) != 2)
+    return 0;
+
+  if (!IS_ALIGNED_SSE(jconst_fdct_islow_sse2))
+    return 0;
+
+  return 1;
+}
+
+GLOBAL(int)
+jsimd_can_fdct_ifast (void)
+{
+  /* The code is optimised for these values only */
+  if (DCTSIZE != 8)
+    return 0;
+  if (sizeof(DCTELEM) != 2)
+    return 0;
+
+  if (!IS_ALIGNED_SSE(jconst_fdct_ifast_sse2))
+    return 0;
+
+  return 1;
+}
+
+GLOBAL(int)
+jsimd_can_fdct_float (void)
+{
+  /* The code is optimised for these values only */
+  if (DCTSIZE != 8)
+    return 0;
+  if (sizeof(FAST_FLOAT) != 4)
+    return 0;
+
+  if (!IS_ALIGNED_SSE(jconst_fdct_float_sse))
+    return 0;
+
+  return 1;
+}
+
+GLOBAL(void)
+jsimd_fdct_islow (DCTELEM * data)
+{
+  jsimd_fdct_islow_sse2(data);
+}
+
+GLOBAL(void)
+jsimd_fdct_ifast (DCTELEM * data)
+{
+  jsimd_fdct_ifast_sse2(data);
+}
+
+GLOBAL(void)
+jsimd_fdct_float (FAST_FLOAT * data)
+{
+  jsimd_fdct_float_sse(data);
+}
+
+GLOBAL(int)
+jsimd_can_quantize (void)
+{
+  /* The code is optimised for these values only */
+  if (DCTSIZE != 8)
+    return 0;
+  if (sizeof(JCOEF) != 2)
+    return 0;
+  if (sizeof(DCTELEM) != 2)
+    return 0;
+
+  return 1;
+}
+
+GLOBAL(int)
+jsimd_can_quantize_float (void)
+{
+  /* The code is optimised for these values only */
+  if (DCTSIZE != 8)
+    return 0;
+  if (sizeof(JCOEF) != 2)
+    return 0;
+  if (sizeof(FAST_FLOAT) != 4)
+    return 0;
+
+  return 1;
+}
+
+GLOBAL(void)
+jsimd_quantize (JCOEFPTR coef_block, DCTELEM * divisors,
+                DCTELEM * workspace)
+{
+  jsimd_quantize_sse2(coef_block, divisors, workspace);
+}
+
+GLOBAL(void)
+jsimd_quantize_float (JCOEFPTR coef_block, FAST_FLOAT * divisors,
+                      FAST_FLOAT * workspace)
+{
+  jsimd_quantize_float_sse2(coef_block, divisors, workspace);
+}
+
+GLOBAL(int)
+jsimd_can_idct_2x2 (void)
+{
+  /* The code is optimised for these values only */
+  if (DCTSIZE != 8)
+    return 0;
+  if (sizeof(JCOEF) != 2)
+    return 0;
+  if (BITS_IN_JSAMPLE != 8)
+    return 0;
+  if (sizeof(JDIMENSION) != 4)
+    return 0;
+  if (sizeof(ISLOW_MULT_TYPE) != 2)
+    return 0;
+
+  if (!IS_ALIGNED_SSE(jconst_idct_red_sse2))
+    return 0;
+
+  return 1;
+}
+
+GLOBAL(int)
+jsimd_can_idct_4x4 (void)
+{
+  /* The code is optimised for these values only */
+  if (DCTSIZE != 8)
+    return 0;
+  if (sizeof(JCOEF) != 2)
+    return 0;
+  if (BITS_IN_JSAMPLE != 8)
+    return 0;
+  if (sizeof(JDIMENSION) != 4)
+    return 0;
+  if (sizeof(ISLOW_MULT_TYPE) != 2)
+    return 0;
+
+  if (!IS_ALIGNED_SSE(jconst_idct_red_sse2))
+    return 0;
+
+  return 1;
+}
+
+GLOBAL(void)
+jsimd_idct_2x2 (j_decompress_ptr cinfo, jpeg_component_info * compptr,
+                JCOEFPTR coef_block, JSAMPARRAY output_buf,
+                JDIMENSION output_col)
+{
+  jsimd_idct_2x2_sse2(compptr->dct_table, coef_block, output_buf, output_col);
+}
+
+GLOBAL(void)
+jsimd_idct_4x4 (j_decompress_ptr cinfo, jpeg_component_info * compptr,
+                JCOEFPTR coef_block, JSAMPARRAY output_buf,
+                JDIMENSION output_col)
+{
+  jsimd_idct_4x4_sse2(compptr->dct_table, coef_block, output_buf, output_col);
+}
+
+GLOBAL(int)
+jsimd_can_idct_islow (void)
+{
+  /* The code is optimised for these values only */
+  if (DCTSIZE != 8)
+    return 0;
+  if (sizeof(JCOEF) != 2)
+    return 0;
+  if (BITS_IN_JSAMPLE != 8)
+    return 0;
+  if (sizeof(JDIMENSION) != 4)
+    return 0;
+  if (sizeof(ISLOW_MULT_TYPE) != 2)
+    return 0;
+
+  if (!IS_ALIGNED_SSE(jconst_idct_islow_sse2))
+    return 0;
+
+  return 1;
+}
+
+GLOBAL(int)
+jsimd_can_idct_ifast (void)
+{
+  /* The code is optimised for these values only */
+  if (DCTSIZE != 8)
+    return 0;
+  if (sizeof(JCOEF) != 2)
+    return 0;
+  if (BITS_IN_JSAMPLE != 8)
+    return 0;
+  if (sizeof(JDIMENSION) != 4)
+    return 0;
+  if (sizeof(IFAST_MULT_TYPE) != 2)
+    return 0;
+  if (IFAST_SCALE_BITS != 2)
+    return 0;
+
+  if (!IS_ALIGNED_SSE(jconst_idct_ifast_sse2))
+    return 0;
+
+  return 1;
+}
+
+GLOBAL(int)
+jsimd_can_idct_float (void)
+{
+  if (DCTSIZE != 8)
+    return 0;
+  if (sizeof(JCOEF) != 2)
+    return 0;
+  if (BITS_IN_JSAMPLE != 8)
+    return 0;
+  if (sizeof(JDIMENSION) != 4)
+    return 0;
+  if (sizeof(FAST_FLOAT) != 4)
+    return 0;
+  if (sizeof(FLOAT_MULT_TYPE) != 4)
+    return 0;
+
+  if (!IS_ALIGNED_SSE(jconst_idct_float_sse2))
+    return 0;
+
+  return 1;
+}
+
+GLOBAL(void)
+jsimd_idct_islow (j_decompress_ptr cinfo, jpeg_component_info * compptr,
+                JCOEFPTR coef_block, JSAMPARRAY output_buf,
+                JDIMENSION output_col)
+{
+  jsimd_idct_islow_sse2(compptr->dct_table, coef_block, output_buf, output_col);
+}
+
+GLOBAL(void)
+jsimd_idct_ifast (j_decompress_ptr cinfo, jpeg_component_info * compptr,
+                JCOEFPTR coef_block, JSAMPARRAY output_buf,
+                JDIMENSION output_col)
+{
+  jsimd_idct_ifast_sse2(compptr->dct_table, coef_block, output_buf, output_col);
+}
+
+GLOBAL(void)
+jsimd_idct_float (j_decompress_ptr cinfo, jpeg_component_info * compptr,
+                JCOEFPTR coef_block, JSAMPARRAY output_buf,
+                JDIMENSION output_col)
+{
+  jsimd_idct_float_sse2(compptr->dct_table, coef_block,
+                        output_buf, output_col);
+}
+
diff --git a/simd/jsimdcpu-64.asm b/simd/jsimdcpu-64.asm
deleted file mode 100644
index 6eaf629..0000000
--- a/simd/jsimdcpu-64.asm
+++ /dev/null
@@ -1,95 +0,0 @@
-;
-; jsimdcpu.asm - SIMD instruction support check
-;
-; Copyright 2009 Pierre Ossman <ossman@cendio.se> for Cendio AB
-; Copyright 2009 D. R. Commander
-;
-; Based on
-; x86 SIMD extension for IJG JPEG library
-; Copyright (C) 1999-2006, MIYASAKA Masaru.
-; For conditions of distribution and use, see copyright notice in jsimdext.inc
-;
-; This file should be assembled with NASM (Netwide Assembler),
-; can *not* be assembled with Microsoft's MASM or any compatible
-; assembler (including Borland's Turbo Assembler).
-; NASM is available from http://nasm.sourceforge.net/ or
-; http://sourceforge.net/project/showfiles.php?group_id=6208
-;
-; [TAB8]
-
-%include "jsimdext.inc"
-
-; --------------------------------------------------------------------------
-	SECTION	SEG_TEXT
-	BITS	64
-;
-; Check if the CPU supports SIMD instructions
-;
-; GLOBAL(unsigned int)
-; jpeg_simd_cpu_support (void)
-;
-
-	align	16
-	global	EXTN(jpeg_simd_cpu_support)
-
-EXTN(jpeg_simd_cpu_support):
-	push	rbx
-
-	xor	rdi,rdi			; simd support flag
-
-	pushfq
-	pop	rax
-	mov	rdx,rax
-	xor	rax, 1<<21		; flip ID bit in EFLAGS
-	push	rax
-	popfq
-	pushfq
-	pop	rax
-	xor	rax,rdx
-	jz	short .return		; CPUID is not supported
-
-	; Check for MMX instruction support
-	xor	rax,rax
-	cpuid
-	test	rax,rax
-	jz	short .return
-
-	xor	rax,rax
-	inc	rax
-	cpuid
-	mov	rax,rdx			; rax = Standard feature flags
-
-	test	rax, 1<<23		; bit23:MMX
-	jz	short .no_mmx
-	or	rdi, byte JSIMD_MMX
-.no_mmx:
-	test	rax, 1<<25		; bit25:SSE
-	jz	short .no_sse
-	or	rdi, byte JSIMD_SSE
-.no_sse:
-	test	rax, 1<<26		; bit26:SSE2
-	jz	short .no_sse2
-	or	rdi, byte JSIMD_SSE2
-.no_sse2:
-
-	; Check for 3DNow! instruction support
-	mov	eax, 0x80000000
-	cpuid
-	cmp	eax, 0x80000000
-	jbe	short .return
-
-	mov	rax, 0x80000001
-	cpuid
-	mov	rax,rdx			; eax = Extended feature flags
-
-	test	eax, 1<<31		; bit31:3DNow!(vendor independent)
-	jz	short .no_3dnow
-	or	edi, byte JSIMD_3DNOW
-.no_3dnow:
-
-.return:
-	mov	rax,rdi
-
-	pop	rbx
-	ret
-