]> granicus.if.org Git - libjpeg-turbo/commitdiff
Clean up SIMD glue code
authorPierre Ossman <ossman@cendio.se>
Mon, 29 Jun 2009 11:20:42 +0000 (11:20 +0000)
committerPierre Ossman <ossman@cendio.se>
Mon, 29 Jun 2009 11:20:42 +0000 (11:20 +0000)
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+ssh://svn.code.sf.net/p/libjpeg-turbo/code/trunk@49 632fc199-4ca6-4c93-a231-07263d6284db

Makefile.am
configure.ac
jsimd_none.c [new file with mode: 0644]
simd/Makefile.am
simd/jsimd_i386.c [moved from jsimd.c with 87% similarity]
simd/jsimd_x86_64.c [new file with mode: 0644]
simd/jsimdcpu-64.asm [deleted file]

index 57081f7e1397a5e56abd79645f73e33809a39a4e..0e7ae4d65311a8d68a973d5ed50981888829ef0d 100644 (file)
@@ -10,14 +10,17 @@ 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 \
-       jsimd.c
+       jidctred.c jquant1.c jquant2.c jutils.c jmemmgr.c jmemnobs.c
 
 if WITH_SIMD
 
 SUBDIRS = simd
 libjpeg_la_LIBADD = simd/libsimd.la
 
+else
+
+libjpeg_la_SOURCES += jsimd_none.c
+
 endif
 
 TSTHDRS = turbojpeg.h rrutil.h rrtimer.h
index a97a4e995e66c775cbb7549fb2e5a034d1c9b50d..0b371329a727d3d0efd9bc1ba6a519a44552803c 100644 (file)
@@ -92,22 +92,27 @@ if test "x${with_simd}" != "xno"; then
     x86_64)
       AC_MSG_RESULT([yes (x86_64)])
       AC_PROG_NASM
-      AC_DEFINE([WITH_SIMD], [1], [Use accelerated SIMD routines.])
-      AC_DEFINE([WITH_SIMD64], [1], [Use x86-64 accelerated SIMD routines.])
+      simd_arch=x86_64
     ;;
     i*86 | x86 | ia32)
       AC_MSG_RESULT([yes (i386)])
       AC_PROG_NASM
-      AC_DEFINE([WITH_SIMD], [1], [Use accelerated SIMD routines.])
+      simd_arch=i386
     ;;
     *)
       AC_MSG_RESULT([no ("$host_cpu")])
       with_simd=no
     ;;
   esac
+
+  if test "x${with_simd}" != "xno"; then
+    AC_DEFINE([WITH_SIMD], [1], [Use accelerated SIMD routines.])
+  fi
 fi
-AM_CONDITIONAL([WITH_SIMD64], [test "x$with_simd64" != "xno"])
+
 AM_CONDITIONAL([WITH_SIMD], [test "x$with_simd" != "xno"])
+AM_CONDITIONAL([SIMD_I386], [test "x$simd_arch" = "xi386"])
+AM_CONDITIONAL([SIMD_X86_64], [test "x$simd_arch" = "xx86_64"])
 
 # 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_none.c b/jsimd_none.c
new file mode 100644 (file)
index 0000000..8960802
--- /dev/null
@@ -0,0 +1,299 @@
+/*
+ * jsimd_none.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 stubs for when there is no SIMD support available.
+ */
+
+#define JPEG_INTERNALS
+#include "jinclude.h"
+#include "jpeglib.h"
+#include "jsimd.h"
+#include "jdct.h"
+#include "jsimddct.h"
+
+GLOBAL(int)
+jsimd_can_rgb_ycc (void)
+{
+  return 0;
+}
+
+GLOBAL(int)
+jsimd_can_ycc_rgb (void)
+{
+  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)
+{
+  return 0;
+}
+
+GLOBAL(int)
+jsimd_can_h2v1_downsample (void)
+{
+  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)
+{
+  return 0;
+}
+
+GLOBAL(int)
+jsimd_can_h2v1_upsample (void)
+{
+  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)
+{
+  return 0;
+}
+
+GLOBAL(int)
+jsimd_can_h2v1_fancy_upsample (void)
+{
+  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)
+{
+  return 0;
+}
+
+GLOBAL(int)
+jsimd_can_h2v1_merged_upsample (void)
+{
+  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)
+{
+  return 0;
+}
+
+GLOBAL(int)
+jsimd_can_convsamp_float (void)
+{
+  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)
+{
+  return 0;
+}
+
+GLOBAL(int)
+jsimd_can_fdct_ifast (void)
+{
+  return 0;
+}
+
+GLOBAL(int)
+jsimd_can_fdct_float (void)
+{
+  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)
+{
+  return 0;
+}
+
+GLOBAL(int)
+jsimd_can_quantize_float (void)
+{
+  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)
+{
+  return 0;
+}
+
+GLOBAL(int)
+jsimd_can_idct_4x4 (void)
+{
+  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)
+{
+  return 0;
+}
+
+GLOBAL(int)
+jsimd_can_idct_ifast (void)
+{
+  return 0;
+}
+
+GLOBAL(int)
+jsimd_can_idct_float (void)
+{
+  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)
+{
+}
+
index da7d6842dfe12c2d6770c88a86f7442c60c8e0f5..8e7469fa847e2b6df9fa1622334bb457749d97df 100644 (file)
@@ -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 \
similarity index 87%
rename from jsimd.c
rename to simd/jsimd_i386.c
index 20e83fc59da788577b0769d413001a09d4e7f350..751fe2dee8eb640463f16a5d4a0d75790033b05d 100644 (file)
--- a/jsimd.c
@@ -1,5 +1,5 @@
 /*
- * jsimd.c
+ * jsimd_i386.c
  *
  * Copyright 2009 Pierre Ossman <ossman@cendio.se> for Cendio AB
  * Copyright 2009 D. R. Commander
@@ -8,26 +8,23 @@
  * Copyright (C) 1999-2006, MIYASAKA Masaru.
  *
  * This file contains the interface between the "normal" portions
- * of the library and the SIMD implementations.
+ * 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 "../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.
  */
-#ifdef WITH_SIMD
 #define IS_ALIGNED(ptr, order) (((unsigned)ptr & ((1 << order) - 1)) == 0)
-#else
-#define IS_ALIGNED(ptr, order) (0)
-#endif
 
 #define IS_ALIGNED_SSE(ptr) (IS_ALIGNED(ptr, 4)) /* 16 byte alignment */
 
@@ -41,23 +38,26 @@ static unsigned int simd_support = ~0;
 LOCAL(void)
 init_simd (void)
 {
-#ifdef WITH_SIMD
   char *env = NULL;
-#endif
+
   if (simd_support != ~0)
     return;
 
-#ifdef WITH_SIMD
   simd_support = jpeg_simd_cpu_support();
-  #ifndef __x86_64__
-  if((env=getenv("JSIMD_FORCEMMX"))!=NULL && !strcmp(env, "1"))
-    simd_support = JSIMD_MMX;
-  else if((env=getenv("JSIMD_FORCESSE2"))!=NULL && !strcmp(env, "1"))
-    simd_support = JSIMD_SSE2;
-  #endif
-#else
-  simd_support = JSIMD_NONE;
-#endif
+
+  /* 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)
@@ -109,66 +109,48 @@ jsimd_rgb_ycc_convert (j_compress_ptr cinfo,
                        JSAMPARRAY input_buf, JSAMPIMAGE output_buf,
                        JDIMENSION output_row, int num_rows)
 {
-#ifdef WITH_SIMD
   void (*sse2fct)(JDIMENSION, JSAMPARRAY, JSAMPIMAGE, JDIMENSION, int);
-  #ifndef __x86_64__
   void (*mmxfct)(JDIMENSION, JSAMPARRAY, JSAMPIMAGE, JDIMENSION, int);
-  #endif
+
   switch(cinfo->in_color_space)
   {
     case JCS_EXT_RGB:
       sse2fct=jsimd_extrgb_ycc_convert_sse2;
-      #ifndef __x86_64__
       mmxfct=jsimd_extrgb_ycc_convert_mmx;
-      #endif
       break;
     case JCS_EXT_RGBX:
       sse2fct=jsimd_extrgbx_ycc_convert_sse2;
-      #ifndef __x86_64__
       mmxfct=jsimd_extrgbx_ycc_convert_mmx;
-      #endif
       break;
     case JCS_EXT_BGR:
       sse2fct=jsimd_extbgr_ycc_convert_sse2;
-      #ifndef __x86_64__
       mmxfct=jsimd_extbgr_ycc_convert_mmx;
-      #endif
       break;
     case JCS_EXT_BGRX:
       sse2fct=jsimd_extbgrx_ycc_convert_sse2;
-      #ifndef __x86_64__
       mmxfct=jsimd_extbgrx_ycc_convert_mmx;
-      #endif
       break;
     case JCS_EXT_XBGR:
       sse2fct=jsimd_extxbgr_ycc_convert_sse2;
-      #ifndef __x86_64__
       mmxfct=jsimd_extxbgr_ycc_convert_mmx;
-      #endif
       break;
     case JCS_EXT_XRGB:
       sse2fct=jsimd_extxrgb_ycc_convert_sse2;
-      #ifndef __x86_64__
       mmxfct=jsimd_extxrgb_ycc_convert_mmx;
-      #endif
       break;
     default:
       sse2fct=jsimd_rgb_ycc_convert_sse2;
-      #ifndef __x86_64__
       mmxfct=jsimd_rgb_ycc_convert_mmx;
-      #endif
       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);
-  #ifndef __x86_64__
   else if (simd_support & JSIMD_MMX)
     mmxfct(cinfo->image_width, input_buf,
         output_buf, output_row, num_rows);
-  #endif
-#endif
 }
 
 GLOBAL(void)
@@ -176,66 +158,48 @@ jsimd_ycc_rgb_convert (j_decompress_ptr cinfo,
                        JSAMPIMAGE input_buf, JDIMENSION input_row,
                        JSAMPARRAY output_buf, int num_rows)
 {
-#ifdef WITH_SIMD
   void (*sse2fct)(JDIMENSION, JSAMPIMAGE, JDIMENSION, JSAMPARRAY, int);
-  #ifndef __x86_64__
   void (*mmxfct)(JDIMENSION, JSAMPIMAGE, JDIMENSION, JSAMPARRAY, int);
-  #endif
+
   switch(cinfo->out_color_space)
   {
     case JCS_EXT_RGB:
       sse2fct=jsimd_ycc_extrgb_convert_sse2;
-      #ifndef __x86_64__
       mmxfct=jsimd_ycc_extrgb_convert_mmx;
-      #endif
       break;
     case JCS_EXT_RGBX:
       sse2fct=jsimd_ycc_extrgbx_convert_sse2;
-      #ifndef __x86_64__
       mmxfct=jsimd_ycc_extrgbx_convert_mmx;
-      #endif
       break;
     case JCS_EXT_BGR:
       sse2fct=jsimd_ycc_extbgr_convert_sse2;
-      #ifndef __x86_64__
       mmxfct=jsimd_ycc_extbgr_convert_mmx;
-      #endif
       break;
     case JCS_EXT_BGRX:
       sse2fct=jsimd_ycc_extbgrx_convert_sse2;
-      #ifndef __x86_64__
       mmxfct=jsimd_ycc_extbgrx_convert_mmx;
-      #endif
       break;
     case JCS_EXT_XBGR:
       sse2fct=jsimd_ycc_extxbgr_convert_sse2;
-      #ifndef __x86_64__
       mmxfct=jsimd_ycc_extxbgr_convert_mmx;
-      #endif
       break;
     case JCS_EXT_XRGB:
       sse2fct=jsimd_ycc_extxrgb_convert_sse2;
-      #ifndef __x86_64__
       mmxfct=jsimd_ycc_extxrgb_convert_mmx;
-      #endif
       break;
     default:
       sse2fct=jsimd_ycc_rgb_convert_sse2;
-      #ifndef __x86_64__
       mmxfct=jsimd_ycc_rgb_convert_mmx;
-      #endif
       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);
-  #ifndef __x86_64__
   else if (simd_support & JSIMD_MMX)
     mmxfct(cinfo->output_width, input_buf,
         input_row, output_buf, num_rows);
-  #endif
-#endif
 }
 
 GLOBAL(int)
@@ -251,10 +215,8 @@ jsimd_can_h2v2_downsample (void)
 
   if (simd_support & JSIMD_SSE2)
     return 1;
-  #ifndef __x86_64__
   if (simd_support & JSIMD_MMX)
     return 1;
-  #endif
 
   return 0;
 }
@@ -272,10 +234,8 @@ jsimd_can_h2v1_downsample (void)
 
   if (simd_support & JSIMD_SSE2)
     return 1;
-  #ifndef __x86_64__
   if (simd_support & JSIMD_MMX)
     return 1;
-  #endif
 
   return 0;
 }
@@ -284,36 +244,28 @@ GLOBAL(void)
 jsimd_h2v2_downsample (j_compress_ptr cinfo, jpeg_component_info * compptr,
                        JSAMPARRAY input_data, JSAMPARRAY output_data)
 {
-#ifdef WITH_SIMD
   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);
-  #ifndef __x86_64__
   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);
-  #endif
-#endif
 }
 
 GLOBAL(void)
 jsimd_h2v1_downsample (j_compress_ptr cinfo, jpeg_component_info * compptr,
                        JSAMPARRAY input_data, JSAMPARRAY output_data)
 {
-#ifdef WITH_SIMD
   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);
-  #ifndef __x86_64__
   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);
-  #endif
-#endif
 }
 
 GLOBAL(int)
@@ -329,10 +281,8 @@ jsimd_can_h2v2_upsample (void)
 
   if (simd_support & JSIMD_SSE2)
     return 1;
-  #ifndef __x86_64__
   if (simd_support & JSIMD_MMX)
     return 1;
-  #endif
 
   return 0;
 }
@@ -350,10 +300,8 @@ jsimd_can_h2v1_upsample (void)
 
   if (simd_support & JSIMD_SSE2)
     return 1;
-  #ifndef __x86_64__
   if (simd_support & JSIMD_MMX)
     return 1;
-  #endif
 
   return 0;
 }
@@ -364,16 +312,12 @@ jsimd_h2v2_upsample (j_decompress_ptr cinfo,
                      JSAMPARRAY input_data,
                      JSAMPARRAY * output_data_ptr)
 {
-#ifdef WITH_SIMD
   if (simd_support & JSIMD_SSE2)
     jsimd_h2v2_upsample_sse2(cinfo->max_v_samp_factor,
         cinfo->output_width, input_data, output_data_ptr);
-  #ifndef __x86_64__
   else if (simd_support & JSIMD_MMX)
     jsimd_h2v2_upsample_mmx(cinfo->max_v_samp_factor,
         cinfo->output_width, input_data, output_data_ptr);
-  #endif
-#endif
 }
 
 GLOBAL(void)
@@ -382,16 +326,12 @@ jsimd_h2v1_upsample (j_decompress_ptr cinfo,
                      JSAMPARRAY input_data,
                      JSAMPARRAY * output_data_ptr)
 {
-#ifdef WITH_SIMD
   if (simd_support & JSIMD_SSE2)
     jsimd_h2v1_upsample_sse2(cinfo->max_v_samp_factor,
         cinfo->output_width, input_data, output_data_ptr);
-  #ifndef __x86_64__
   else if (simd_support & JSIMD_MMX)
     jsimd_h2v1_upsample_mmx(cinfo->max_v_samp_factor,
         cinfo->output_width, input_data, output_data_ptr);
-  #endif
-#endif
 }
 
 GLOBAL(int)
@@ -408,10 +348,8 @@ jsimd_can_h2v2_fancy_upsample (void)
   if ((simd_support & JSIMD_SSE2) &&
       IS_ALIGNED_SSE(jconst_fancy_upsample_sse2))
     return 1;
-  #ifndef __x86_64__
   if (simd_support & JSIMD_MMX)
     return 1;
-  #endif
 
   return 0;
 }
@@ -430,10 +368,8 @@ jsimd_can_h2v1_fancy_upsample (void)
   if ((simd_support & JSIMD_SSE2) &&
       IS_ALIGNED_SSE(jconst_fancy_upsample_sse2))
     return 1;
-  #ifndef __x86_64__
   if (simd_support & JSIMD_MMX)
     return 1;
-  #endif
 
   return 0;
 }
@@ -444,17 +380,13 @@ jsimd_h2v2_fancy_upsample (j_decompress_ptr cinfo,
                            JSAMPARRAY input_data,
                            JSAMPARRAY * output_data_ptr)
 {
-#ifdef WITH_SIMD
   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);
-  #ifndef __x86_64__
   else if (simd_support & JSIMD_MMX)
     jsimd_h2v2_fancy_upsample_mmx(cinfo->max_v_samp_factor,
         compptr->downsampled_width, input_data, output_data_ptr);
-  #endif
-#endif
 }
 
 GLOBAL(void)
@@ -463,17 +395,13 @@ jsimd_h2v1_fancy_upsample (j_decompress_ptr cinfo,
                            JSAMPARRAY input_data,
                            JSAMPARRAY * output_data_ptr)
 {
-#ifdef WITH_SIMD
   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);
-  #ifndef __x86_64__
   else if (simd_support & JSIMD_MMX)
     jsimd_h2v1_fancy_upsample_mmx(cinfo->max_v_samp_factor,
         compptr->downsampled_width, input_data, output_data_ptr);
-  #endif
-#endif
 }
 
 GLOBAL(int)
@@ -490,10 +418,8 @@ jsimd_can_h2v2_merged_upsample (void)
   if ((simd_support & JSIMD_SSE2) &&
       IS_ALIGNED_SSE(jconst_merged_upsample_sse2))
     return 1;
-  #ifndef __x86_64__
   if (simd_support & JSIMD_MMX)
     return 1;
-  #endif
 
   return 0;
 }
@@ -512,10 +438,8 @@ jsimd_can_h2v1_merged_upsample (void)
   if ((simd_support & JSIMD_SSE2) &&
       IS_ALIGNED_SSE(jconst_merged_upsample_sse2))
     return 1;
-  #ifndef __x86_64__
   if (simd_support & JSIMD_MMX)
     return 1;
-  #endif
 
   return 0;
 }
@@ -526,66 +450,48 @@ jsimd_h2v2_merged_upsample (j_decompress_ptr cinfo,
                             JDIMENSION in_row_group_ctr,
                             JSAMPARRAY output_buf)
 {
-#ifdef WITH_SIMD
   void (*sse2fct)(JDIMENSION, JSAMPIMAGE, JDIMENSION, JSAMPARRAY);
-  #ifndef __x86_64__
   void (*mmxfct)(JDIMENSION, JSAMPIMAGE, JDIMENSION, JSAMPARRAY);
-  #endif
+
   switch(cinfo->out_color_space)
   {
     case JCS_EXT_RGB:
       sse2fct=jsimd_h2v2_extrgb_merged_upsample_sse2;
-      #ifndef __x86_64__
       mmxfct=jsimd_h2v2_extrgb_merged_upsample_mmx;
-      #endif
       break;
     case JCS_EXT_RGBX:
       sse2fct=jsimd_h2v2_extrgbx_merged_upsample_sse2;
-      #ifndef __x86_64__
       mmxfct=jsimd_h2v2_extrgbx_merged_upsample_mmx;
-      #endif
       break;
     case JCS_EXT_BGR:
       sse2fct=jsimd_h2v2_extbgr_merged_upsample_sse2;
-      #ifndef __x86_64__
       mmxfct=jsimd_h2v2_extbgr_merged_upsample_mmx;
-      #endif
       break;
     case JCS_EXT_BGRX:
       sse2fct=jsimd_h2v2_extbgrx_merged_upsample_sse2;
-      #ifndef __x86_64__
       mmxfct=jsimd_h2v2_extbgrx_merged_upsample_mmx;
-      #endif
       break;
     case JCS_EXT_XBGR:
       sse2fct=jsimd_h2v2_extxbgr_merged_upsample_sse2;
-      #ifndef __x86_64__
       mmxfct=jsimd_h2v2_extxbgr_merged_upsample_mmx;
-      #endif
       break;
     case JCS_EXT_XRGB:
       sse2fct=jsimd_h2v2_extxrgb_merged_upsample_sse2;
-      #ifndef __x86_64__
       mmxfct=jsimd_h2v2_extxrgb_merged_upsample_mmx;
-      #endif
       break;
     default:
       sse2fct=jsimd_h2v2_merged_upsample_sse2;
-      #ifndef __x86_64__
       mmxfct=jsimd_h2v2_merged_upsample_mmx;
-      #endif
       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);
-  #ifndef __x86_64__
   else if (simd_support & JSIMD_MMX)
     mmxfct(cinfo->output_width, input_buf,
         in_row_group_ctr, output_buf);
-  #endif
-#endif
 }
 
 GLOBAL(void)
@@ -594,66 +500,48 @@ jsimd_h2v1_merged_upsample (j_decompress_ptr cinfo,
                             JDIMENSION in_row_group_ctr,
                             JSAMPARRAY output_buf)
 {
-#ifdef WITH_SIMD
   void (*sse2fct)(JDIMENSION, JSAMPIMAGE, JDIMENSION, JSAMPARRAY);
-  #ifndef __x86_64__
   void (*mmxfct)(JDIMENSION, JSAMPIMAGE, JDIMENSION, JSAMPARRAY);
-  #endif
+
   switch(cinfo->out_color_space)
   {
     case JCS_EXT_RGB:
       sse2fct=jsimd_h2v1_extrgb_merged_upsample_sse2;
-      #ifndef __x86_64__
       mmxfct=jsimd_h2v1_extrgb_merged_upsample_mmx;
-      #endif
       break;
     case JCS_EXT_RGBX:
       sse2fct=jsimd_h2v1_extrgbx_merged_upsample_sse2;
-      #ifndef __x86_64__
       mmxfct=jsimd_h2v1_extrgbx_merged_upsample_mmx;
-      #endif
       break;
     case JCS_EXT_BGR:
       sse2fct=jsimd_h2v1_extbgr_merged_upsample_sse2;
-      #ifndef __x86_64__
       mmxfct=jsimd_h2v1_extbgr_merged_upsample_mmx;
-      #endif
       break;
     case JCS_EXT_BGRX:
       sse2fct=jsimd_h2v1_extbgrx_merged_upsample_sse2;
-      #ifndef __x86_64__
       mmxfct=jsimd_h2v1_extbgrx_merged_upsample_mmx;
-      #endif
       break;
     case JCS_EXT_XBGR:
       sse2fct=jsimd_h2v1_extxbgr_merged_upsample_sse2;
-      #ifndef __x86_64__
       mmxfct=jsimd_h2v1_extxbgr_merged_upsample_mmx;
-      #endif
       break;
     case JCS_EXT_XRGB:
       sse2fct=jsimd_h2v1_extxrgb_merged_upsample_sse2;
-      #ifndef __x86_64__
       mmxfct=jsimd_h2v1_extxrgb_merged_upsample_mmx;
-      #endif
       break;
     default:
       sse2fct=jsimd_h2v1_merged_upsample_sse2;
-      #ifndef __x86_64__
       mmxfct=jsimd_h2v1_merged_upsample_mmx;
-      #endif
       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);
-  #ifndef __x86_64__
   else if (simd_support & JSIMD_MMX)
     mmxfct(cinfo->output_width, input_buf,
         in_row_group_ctr, output_buf);
-  #endif
-#endif
 }
 
 GLOBAL(int)
@@ -673,10 +561,8 @@ jsimd_can_convsamp (void)
 
   if (simd_support & JSIMD_SSE2)
     return 1;
-  #ifndef __x86_64__
   if (simd_support & JSIMD_MMX)
     return 1;
-  #endif
 
   return 0;
 }
@@ -700,10 +586,8 @@ jsimd_can_convsamp_float (void)
     return 1;
   if (simd_support & JSIMD_SSE)
     return 1;
-  #ifndef __x86_64__
   if (simd_support & JSIMD_3DNOW)
     return 1;
-  #endif
 
   return 0;
 }
@@ -712,30 +596,22 @@ GLOBAL(void)
 jsimd_convsamp (JSAMPARRAY sample_data, JDIMENSION start_col,
                 DCTELEM * workspace)
 {
-#ifdef WITH_SIMD
   if (simd_support & JSIMD_SSE2)
     jsimd_convsamp_sse2(sample_data, start_col, workspace);
-  #ifndef __x86_64__
   else if (simd_support & JSIMD_MMX)
     jsimd_convsamp_mmx(sample_data, start_col, workspace);
-  #endif
-#endif
 }
 
 GLOBAL(void)
 jsimd_convsamp_float (JSAMPARRAY sample_data, JDIMENSION start_col,
                       FAST_FLOAT * workspace)
 {
-#ifdef WITH_SIMD
   if (simd_support & JSIMD_SSE2)
     jsimd_convsamp_float_sse2(sample_data, start_col, workspace);
-  #ifndef __x86_64__
   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);
-  #endif
-#endif
 }
 
 GLOBAL(int)
@@ -751,10 +627,8 @@ jsimd_can_fdct_islow (void)
 
   if ((simd_support & JSIMD_SSE2) && IS_ALIGNED_SSE(jconst_fdct_islow_sse2))
     return 1;
-  #ifndef __x86_64__
   if (simd_support & JSIMD_MMX)
     return 1;
-  #endif
 
   return 0;
 }
@@ -772,10 +646,8 @@ jsimd_can_fdct_ifast (void)
 
   if ((simd_support & JSIMD_SSE2) && IS_ALIGNED_SSE(jconst_fdct_ifast_sse2))
     return 1;
-  #ifndef __x86_64__
   if (simd_support & JSIMD_MMX)
     return 1;
-  #endif
 
   return 0;
 }
@@ -793,10 +665,8 @@ jsimd_can_fdct_float (void)
 
   if ((simd_support & JSIMD_SSE) && IS_ALIGNED_SSE(jconst_fdct_float_sse))
     return 1;
-  #ifndef __x86_64__
   if (simd_support & JSIMD_3DNOW)
     return 1;
-  #endif
 
   return 0;
 }
@@ -804,40 +674,28 @@ jsimd_can_fdct_float (void)
 GLOBAL(void)
 jsimd_fdct_islow (DCTELEM * data)
 {
-#ifdef WITH_SIMD
   if ((simd_support & JSIMD_SSE2) && IS_ALIGNED_SSE(jconst_fdct_islow_sse2))
     jsimd_fdct_islow_sse2(data);
-  #ifndef __x86_64__
   else if (simd_support & JSIMD_MMX)
     jsimd_fdct_islow_mmx(data);
-  #endif
-#endif
 }
 
 GLOBAL(void)
 jsimd_fdct_ifast (DCTELEM * data)
 {
-#ifdef WITH_SIMD
   if ((simd_support & JSIMD_SSE2) && IS_ALIGNED_SSE(jconst_fdct_islow_sse2))
     jsimd_fdct_ifast_sse2(data);
-  #ifndef __x86_64__
   else if (simd_support & JSIMD_MMX)
     jsimd_fdct_ifast_mmx(data);
-  #endif
-#endif
 }
 
 GLOBAL(void)
 jsimd_fdct_float (FAST_FLOAT * data)
 {
-#ifdef WITH_SIMD
   if ((simd_support & JSIMD_SSE) && IS_ALIGNED_SSE(jconst_fdct_float_sse))
     jsimd_fdct_float_sse(data);
-  #ifndef __x86_64__
   else if (simd_support & JSIMD_3DNOW)
     jsimd_fdct_float_3dnow(data);
-  #endif
-#endif
 }
 
 GLOBAL(int)
@@ -855,10 +713,8 @@ jsimd_can_quantize (void)
 
   if (simd_support & JSIMD_SSE2)
     return 1;
-  #ifndef __x86_64__
   if (simd_support & JSIMD_MMX)
     return 1;
-  #endif
 
   return 0;
 }
@@ -880,10 +736,8 @@ jsimd_can_quantize_float (void)
     return 1;
   if (simd_support & JSIMD_SSE)
     return 1;
-  #ifndef __x86_64__
   if (simd_support & JSIMD_3DNOW)
     return 1;
-  #endif
 
   return 0;
 }
@@ -892,30 +746,22 @@ GLOBAL(void)
 jsimd_quantize (JCOEFPTR coef_block, DCTELEM * divisors,
                 DCTELEM * workspace)
 {
-#ifdef WITH_SIMD
   if (simd_support & JSIMD_SSE2)
     jsimd_quantize_sse2(coef_block, divisors, workspace);
-  #ifndef __x86_64__
   else if (simd_support & JSIMD_MMX)
     jsimd_quantize_mmx(coef_block, divisors, workspace);
-  #endif
-#endif
 }
 
 GLOBAL(void)
 jsimd_quantize_float (JCOEFPTR coef_block, FAST_FLOAT * divisors,
                       FAST_FLOAT * workspace)
 {
-#ifdef WITH_SIMD
   if (simd_support & JSIMD_SSE2)
     jsimd_quantize_float_sse2(coef_block, divisors, workspace);
-  #ifndef __x86_64__
   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);
-  #endif
-#endif
 }
 
 GLOBAL(int)
@@ -937,10 +783,8 @@ jsimd_can_idct_2x2 (void)
 
   if ((simd_support & JSIMD_SSE2) && IS_ALIGNED_SSE(jconst_idct_red_sse2))
     return 1;
-  #ifndef __x86_64__
   if (simd_support & JSIMD_MMX)
     return 1;
-  #endif
 
   return 0;
 }
@@ -964,10 +808,8 @@ jsimd_can_idct_4x4 (void)
 
   if ((simd_support & JSIMD_SSE2) && IS_ALIGNED_SSE(jconst_idct_red_sse2))
     return 1;
-  #ifndef __x86_64__
   if (simd_support & JSIMD_MMX)
     return 1;
-  #endif
 
   return 0;
 }
@@ -977,14 +819,10 @@ jsimd_idct_2x2 (j_decompress_ptr cinfo, jpeg_component_info * compptr,
                 JCOEFPTR coef_block, JSAMPARRAY output_buf,
                 JDIMENSION output_col)
 {
-#if WITH_SIMD
   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);
-  #ifndef __x86_64__
   else if (simd_support & JSIMD_MMX)
     jsimd_idct_2x2_mmx(compptr->dct_table, coef_block, output_buf, output_col);
-  #endif
-#endif
 }
 
 GLOBAL(void)
@@ -992,14 +830,10 @@ jsimd_idct_4x4 (j_decompress_ptr cinfo, jpeg_component_info * compptr,
                 JCOEFPTR coef_block, JSAMPARRAY output_buf,
                 JDIMENSION output_col)
 {
-#if WITH_SIMD
   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);
-  #ifndef __x86_64__
   else if (simd_support & JSIMD_MMX)
     jsimd_idct_4x4_mmx(compptr->dct_table, coef_block, output_buf, output_col);
-  #endif
-#endif
 }
 
 GLOBAL(int)
@@ -1021,10 +855,8 @@ jsimd_can_idct_islow (void)
 
   if ((simd_support & JSIMD_SSE2) && IS_ALIGNED_SSE(jconst_idct_islow_sse2))
     return 1;
-  #ifndef __x86_64__
   if (simd_support & JSIMD_MMX)
     return 1;
-  #endif
 
   return 0;
 }
@@ -1050,10 +882,8 @@ jsimd_can_idct_ifast (void)
 
   if ((simd_support & JSIMD_SSE2) && IS_ALIGNED_SSE(jconst_idct_ifast_sse2))
     return 1;
-  #ifndef __x86_64__
   if (simd_support & JSIMD_MMX)
     return 1;
-  #endif
 
   return 0;
 }
@@ -1078,12 +908,10 @@ jsimd_can_idct_float (void)
 
   if ((simd_support & JSIMD_SSE2) && IS_ALIGNED_SSE(jconst_idct_float_sse2))
     return 1;
-  #ifndef __x86_64__
   if ((simd_support & JSIMD_SSE) && IS_ALIGNED_SSE(jconst_idct_float_sse))
     return 1;
   if (simd_support & JSIMD_3DNOW)
     return 1;
-  #endif
 
   return 0;
 }
@@ -1093,14 +921,10 @@ jsimd_idct_islow (j_decompress_ptr cinfo, jpeg_component_info * compptr,
                 JCOEFPTR coef_block, JSAMPARRAY output_buf,
                 JDIMENSION output_col)
 {
-#if WITH_SIMD
   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);
-  #ifndef __x86_64__
   else if (simd_support & JSIMD_MMX)
     jsimd_idct_islow_mmx(compptr->dct_table, coef_block, output_buf, output_col);
-  #endif
-#endif
 }
 
 GLOBAL(void)
@@ -1108,14 +932,10 @@ jsimd_idct_ifast (j_decompress_ptr cinfo, jpeg_component_info * compptr,
                 JCOEFPTR coef_block, JSAMPARRAY output_buf,
                 JDIMENSION output_col)
 {
-#if WITH_SIMD
   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);
-  #ifndef __x86_64__
   else if (simd_support & JSIMD_MMX)
     jsimd_idct_ifast_mmx(compptr->dct_table, coef_block, output_buf, output_col);
-  #endif
-#endif
 }
 
 GLOBAL(void)
@@ -1123,18 +943,14 @@ jsimd_idct_float (j_decompress_ptr cinfo, jpeg_component_info * compptr,
                 JCOEFPTR coef_block, JSAMPARRAY output_buf,
                 JDIMENSION output_col)
 {
-#if WITH_SIMD
   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);
-  #ifndef __x86_64__
   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);
-  #endif
-#endif
 }
 
diff --git a/simd/jsimd_x86_64.c b/simd/jsimd_x86_64.c
new file mode 100644 (file)
index 0000000..4c1ef7d
--- /dev/null
@@ -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 (file)
index 6eaf629..0000000
+++ /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
-