Update libjpeg_turbo to 1.4.90 from https://github.com/libjpeg-turbo/

simd/jsimd_i386.c

 /*
  * jsimd_i386.c
  *
  * Copyright 2009 Pierre Ossman <ossman@cendio.se> for Cendio AB
- * Copyright 2009-2011, 2013-2014 D. R. Commander
- * 
+ * Copyright 2009-2011, 2013-2014, 2016 D. R. Commander
+ * Copyright 2015 Matthieu Darbois
+ *
  * Based on the 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 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 "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;
+static unsigned int simd_huffman = 1;
 
 /*
  * 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 != ~0U)
     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|JSIMD_MMX;
   env = getenv("JSIMD_FORCESSE");
   if ((env != NULL) && (strcmp(env, "1") == 0))
     simd_support &= JSIMD_SSE|JSIMD_MMX;
   env = getenv("JSIMD_FORCESSE2");
   if ((env != NULL) && (strcmp(env, "1") == 0))
     simd_support &= JSIMD_SSE2;
+  env = getenv("JSIMD_FORCENONE");
+  if ((env != NULL) && (strcmp(env, "1") == 0))
+    simd_support = 0;
+  env = getenv("JSIMD_NOHUFFENC");
+  if ((env != NULL) && (strcmp(env, "1") == 0))
+    simd_huffman = 0;
 }
 
-#ifndef JPEG_DECODE_ONLY
 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;
 }
-#endif
 
 GLOBAL(int)
 jsimd_can_rgb_gray (void)
 {
   init_simd();
 
   /* The code is optimised for these values only */
   if (BITS_IN_JSAMPLE != 8)
     return 0;
   if (sizeof(JDIMENSION) != 4)
@@ skipping 31 matching lines @@
 
   return 0;
 }
 
 GLOBAL(int)
 jsimd_can_ycc_rgb565 (void)
 {
   return 0;
 }
 
-#ifndef JPEG_DECODE_ONLY
 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)
-  {
+  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:
     case JCS_EXT_RGBA:
       sse2fct=jsimd_extrgbx_ycc_convert_sse2;
       mmxfct=jsimd_extrgbx_ycc_convert_mmx;
       break;
     case JCS_EXT_BGR:
@@ skipping 16 matching lines @@
       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);
+    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);
-}
-#endif
-
-GLOBAL(void)
-jsimd_ycc_rgb565_convert (j_decompress_ptr cinfo,
-                          JSAMPIMAGE input_buf, JDIMENSION input_row,
-                          JSAMPARRAY output_buf, int num_rows)
-{
+    mmxfct(cinfo->image_width, input_buf, output_buf, output_row, num_rows);
 }
 
 GLOBAL(void)
 jsimd_rgb_gray_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)
-  {
+  switch(cinfo->in_color_space) {
     case JCS_EXT_RGB:
       sse2fct=jsimd_extrgb_gray_convert_sse2;
       mmxfct=jsimd_extrgb_gray_convert_mmx;
       break;
     case JCS_EXT_RGBX:
     case JCS_EXT_RGBA:
       sse2fct=jsimd_extrgbx_gray_convert_sse2;
       mmxfct=jsimd_extrgbx_gray_convert_mmx;
       break;
     case JCS_EXT_BGR:
@@ skipping 16 matching lines @@
       mmxfct=jsimd_extxrgb_gray_convert_mmx;
       break;
     default:
       sse2fct=jsimd_rgb_gray_convert_sse2;
       mmxfct=jsimd_rgb_gray_convert_mmx;
       break;
   }
 
   if ((simd_support & JSIMD_SSE2) &&
       IS_ALIGNED_SSE(jconst_rgb_gray_convert_sse2))
-    sse2fct(cinfo->image_width, input_buf,
-        output_buf, output_row, num_rows);
+    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);
+    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)
-  {
+  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:
     case JCS_EXT_RGBA:
       sse2fct=jsimd_ycc_extrgbx_convert_sse2;
       mmxfct=jsimd_ycc_extrgbx_convert_mmx;
       break;
     case JCS_EXT_BGR:
@@ skipping 16 matching lines @@
       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);
+    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);
+    mmxfct(cinfo->output_width, input_buf, input_row, output_buf, num_rows);
 }
 
-#ifndef JPEG_DECODE_ONLY
+GLOBAL(void)
+jsimd_ycc_rgb565_convert (j_decompress_ptr cinfo,
+                          JSAMPIMAGE input_buf, JDIMENSION input_row,
+                          JSAMPARRAY output_buf, int num_rows)
+{
+}
+
 GLOBAL(int)
 jsimd_can_h2v2_downsample (void)
 {
   init_simd();
 
   /* The code is optimised for these values only */
   if (BITS_IN_JSAMPLE != 8)
     return 0;
   if (sizeof(JDIMENSION) != 4)
     return 0;
@@ skipping 19 matching lines @@
 
   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,
+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);
+                               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);
+                              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,
+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);
+                               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);
+                              compptr->v_samp_factor, compptr->width_in_blocks,
+                              input_data, output_data);
 }
-#endif
 
 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)
@@ skipping 21 matching lines @@
   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, 
+                     jpeg_component_info *compptr,
                      JSAMPARRAY input_data,
-                     JSAMPARRAY * output_data_ptr)
+                     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);
+    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);
+    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, 
+                     jpeg_component_info *compptr,
                      JSAMPARRAY input_data,
-                     JSAMPARRAY * output_data_ptr)
+                     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);
+    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);
+    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;
@@ skipping 24 matching lines @@
       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, 
+                           jpeg_component_info *compptr,
                            JSAMPARRAY input_data,
-                           JSAMPARRAY * output_data_ptr)
+                           JSAMPARRAY *output_data_ptr)
 {
   if ((simd_support & JSIMD_SSE2) &&
       IS_ALIGNED_SSE(jconst_fancy_upsample_sse2))
     jsimd_h2v2_fancy_upsample_sse2(cinfo->max_v_samp_factor,
-        compptr->downsampled_width, input_data, output_data_ptr);
+                                   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);
+                                  compptr->downsampled_width, input_data,
+                                  output_data_ptr);
 }
 
 GLOBAL(void)
 jsimd_h2v1_fancy_upsample (j_decompress_ptr cinfo,
-                           jpeg_component_info * compptr, 
+                           jpeg_component_info *compptr,
                            JSAMPARRAY input_data,
-                           JSAMPARRAY * output_data_ptr)
+                           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);
+                                   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);
+                                  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;
@@ skipping 31 matching lines @@
 
 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)
-  {
+  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:
     case JCS_EXT_RGBA:
       sse2fct=jsimd_h2v2_extrgbx_merged_upsample_sse2;
       mmxfct=jsimd_h2v2_extrgbx_merged_upsample_mmx;
       break;
     case JCS_EXT_BGR:
@@ skipping 16 matching lines @@
       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);
+    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);
+    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)
-  {
+  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:
     case JCS_EXT_RGBA:
       sse2fct=jsimd_h2v1_extrgbx_merged_upsample_sse2;
       mmxfct=jsimd_h2v1_extrgbx_merged_upsample_mmx;
       break;
     case JCS_EXT_BGR:
@@ skipping 16 matching lines @@
       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);
+    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);
+    mmxfct(cinfo->output_width, input_buf, in_row_group_ctr, output_buf);
 }
 
-#ifndef JPEG_DECODE_ONLY
 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;
@@ skipping 30 matching lines @@
   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)
+                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)
+                      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)
@@ skipping 47 matching lines @@
 
   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)
+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)
+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)
+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)
 {
@@ skipping 32 matching lines @@
     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)
+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)
+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);
 }
-#endif
 
 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)
@@ skipping 32 matching lines @@
 
   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,
+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);
+    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,
+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);
+    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 */
@@ skipping 65 matching lines @@
     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)
+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);
+    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);
+    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)
+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);
+    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);
+    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)
+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);
+    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);
+    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);
+    jsimd_idct_float_3dnow(compptr->dct_table, coef_block, output_buf,
+                           output_col);
 }
+
+GLOBAL(int)
+jsimd_can_huff_encode_one_block (void)
+{
+  init_simd();
+
+  if (DCTSIZE != 8)
+    return 0;
+  if (sizeof(JCOEF) != 2)
+    return 0;
+
+  if ((simd_support & JSIMD_SSE2) && simd_huffman &&
+      IS_ALIGNED_SSE(jconst_huff_encode_one_block))
+    return 1;
+
+  return 0;
+}
+
+GLOBAL(JOCTET*)
+jsimd_huff_encode_one_block (void *state, JOCTET *buffer, JCOEFPTR block,
+                             int last_dc_val, c_derived_tbl *dctbl,
+                             c_derived_tbl *actbl)
+{
+  return jsimd_huff_encode_one_block_sse2(state, buffer, block, last_dc_val,
+                                          dctbl, actbl);
+}