gecko/gfx/ycbcr/convert.patch

diff --git a/gfx/ycbcr/yuv_convert.cpp b/gfx/ycbcr/yuv_convert.cpp
--- a/gfx/ycbcr/yuv_convert.cpp
+++ b/gfx/ycbcr/yuv_convert.cpp
@@ -6,145 +6,102 @@
 // http://www.fourcc.org/yuv.php
 // The actual conversion is best described here
 // http://en.wikipedia.org/wiki/YUV
 // An article on optimizing YUV conversion using tables instead of multiplies
 // http://lestourtereaux.free.fr/papers/data/yuvrgb.pdf
 //
 // YV12 is a full plane of Y and a half height, half width chroma planes
 // YV16 is a full plane of Y and a full height, half width chroma planes
+// YV24 is a full plane of Y and a full height, full width chroma planes
 //
 // ARGB pixel format is output, which on little endian is stored as BGRA.
 // The alpha is set to 255, allowing the application to use RGBA or RGB32.
 
-#include "media/base/yuv_convert.h"
+#include "yuv_convert.h"
 
 // Header for low level row functions.
-#include "media/base/yuv_row.h"
-
-#if USE_MMX
-#if defined(_MSC_VER)
-#include <intrin.h>
-#else
-#include <mmintrin.h>
-#endif
-#endif
-
-#if USE_SSE2
-#include <emmintrin.h>
-#endif
-
-namespace media {
-
+#include "yuv_row.h"
+#include "mozilla/SSE.h"
+
+namespace mozilla {
+
+namespace gfx {
+ 
 // 16.16 fixed point arithmetic
 const int kFractionBits = 16;
 const int kFractionMax = 1 << kFractionBits;
 const int kFractionMask = ((1 << kFractionBits) - 1);
 
 // Convert a frame of YUV to 32 bit ARGB.
-void ConvertYUVToRGB32(const uint8* y_buf,
-                       const uint8* u_buf,
-                       const uint8* v_buf,
-                       uint8* rgb_buf,
-                       int width,
-                       int height,
-                       int y_pitch,
-                       int uv_pitch,
-                       int rgb_pitch,
-                       YUVType yuv_type) {
-  unsigned int y_shift = yuv_type;
-  for (int y = 0; y < height; ++y) {
-    uint8* rgb_row = rgb_buf + y * rgb_pitch;
-    const uint8* y_ptr = y_buf + y * y_pitch;
-    const uint8* u_ptr = u_buf + (y >> y_shift) * uv_pitch;
-    const uint8* v_ptr = v_buf + (y >> y_shift) * uv_pitch;
-
-    FastConvertYUVToRGB32Row(y_ptr,
-                             u_ptr,
-                             v_ptr,
-                             rgb_row,
-                             width);
-  }
+NS_GFX_(void) ConvertYCbCrToRGB32(const uint8* y_buf,
+                                  const uint8* u_buf,
+                                  const uint8* v_buf,
+                                  uint8* rgb_buf,
+                                  int pic_x,
+                                  int pic_y,
+                                  int pic_width,
+                                  int pic_height,
+                                  int y_pitch,
+                                  int uv_pitch,
+                                  int rgb_pitch,
+                                  YUVType yuv_type) {
+  unsigned int y_shift = yuv_type == YV12 ? 1 : 0;
+  unsigned int x_shift = yuv_type == YV24 ? 0 : 1;
+  // Test for SSE because the optimized code uses movntq, which is not part of MMX.
+  bool has_sse = supports_mmx() && supports_sse();
+  // There is no optimized YV24 SSE routine so we check for this and
+  // fall back to the C code.
+  has_sse &= yuv_type != YV24;
+  bool odd_pic_x = yuv_type != YV24 && pic_x % 2 != 0;
+  int x_width = odd_pic_x ? pic_width - 1 : pic_width;
+
+  for (int y = pic_y; y < pic_height + pic_y; ++y) {
+    uint8* rgb_row = rgb_buf + (y - pic_y) * rgb_pitch;
+    const uint8* y_ptr = y_buf + y * y_pitch + pic_x;
+    const uint8* u_ptr = u_buf + (y >> y_shift) * uv_pitch + (pic_x >> x_shift);
+    const uint8* v_ptr = v_buf + (y >> y_shift) * uv_pitch + (pic_x >> x_shift);
+
+    if (odd_pic_x) {
+      // Handle the single odd pixel manually and use the
+      // fast routines for the remaining.
+      FastConvertYUVToRGB32Row_C(y_ptr++,
+                                 u_ptr++,
+                                 v_ptr++,
+                                 rgb_row,
+                                 1,
+                                 x_shift);
+      rgb_row += 4;
+    }
+
+    if (has_sse) {
+      FastConvertYUVToRGB32Row(y_ptr,
+                               u_ptr,
+                               v_ptr,
+                               rgb_row,
+                               x_width);
+    }
+    else {
+      FastConvertYUVToRGB32Row_C(y_ptr,
+                                 u_ptr,
+                                 v_ptr,
+                                 rgb_row,
+                                 x_width,
+                                 x_shift);
+    }
+  }
 
   // MMX used for FastConvertYUVToRGB32Row requires emms instruction.
-  EMMS();
-}
-
-#if USE_SSE2
-// FilterRows combines two rows of the image using linear interpolation.
-// SSE2 version does 16 pixels at a time
-
-static void FilterRows(uint8* ybuf, const uint8* y0_ptr, const uint8* y1_ptr,
-                       int source_width, int source_y_fraction) {
-  __m128i zero = _mm_setzero_si128();
-  __m128i y1_fraction = _mm_set1_epi16(source_y_fraction);
-  __m128i y0_fraction = _mm_set1_epi16(256 - source_y_fraction);
-
-  const __m128i* y0_ptr128 = reinterpret_cast<const __m128i*>(y0_ptr);
-  const __m128i* y1_ptr128 = reinterpret_cast<const __m128i*>(y1_ptr);
-  __m128i* dest128 = reinterpret_cast<__m128i*>(ybuf);
-  __m128i* end128 = reinterpret_cast<__m128i*>(ybuf + source_width);
-
-  do {
-    __m128i y0 = _mm_loadu_si128(y0_ptr128);
-    __m128i y1 = _mm_loadu_si128(y1_ptr128);
-    __m128i y2 = _mm_unpackhi_epi8(y0, zero);
-    __m128i y3 = _mm_unpackhi_epi8(y1, zero);
-    y0 = _mm_unpacklo_epi8(y0, zero);
-    y1 = _mm_unpacklo_epi8(y1, zero);
-    y0 = _mm_mullo_epi16(y0, y0_fraction);
-    y1 = _mm_mullo_epi16(y1, y1_fraction);
-    y2 = _mm_mullo_epi16(y2, y0_fraction);
-    y3 = _mm_mullo_epi16(y3, y1_fraction);
-    y0 = _mm_add_epi16(y0, y1);
-    y2 = _mm_add_epi16(y2, y3);
-    y0 = _mm_srli_epi16(y0, 8);
-    y2 = _mm_srli_epi16(y2, 8);
-    y0 = _mm_packus_epi16(y0, y2);
-    *dest128++ = y0;
-    ++y0_ptr128;
-    ++y1_ptr128;
-  } while (dest128 < end128);
-}
-#elif USE_MMX
-// MMX version does 8 pixels at a time
-static void FilterRows(uint8* ybuf, const uint8* y0_ptr, const uint8* y1_ptr,
-                       int source_width, int source_y_fraction) {
-  __m64 zero = _mm_setzero_si64();
-  __m64 y1_fraction = _mm_set1_pi16(source_y_fraction);
-  __m64 y0_fraction = _mm_set1_pi16(256 - source_y_fraction);
-
-  const __m64* y0_ptr64 = reinterpret_cast<const __m64*>(y0_ptr);
-  const __m64* y1_ptr64 = reinterpret_cast<const __m64*>(y1_ptr);
-  __m64* dest64 = reinterpret_cast<__m64*>(ybuf);
-  __m64* end64 = reinterpret_cast<__m64*>(ybuf + source_width);
-
-  do {
-    __m64 y0 = *y0_ptr64++;
-    __m64 y1 = *y1_ptr64++;
-    __m64 y2 = _mm_unpackhi_pi8(y0, zero);
-    __m64 y3 = _mm_unpackhi_pi8(y1, zero);
-    y0 = _mm_unpacklo_pi8(y0, zero);
-    y1 = _mm_unpacklo_pi8(y1, zero);
-    y0 = _mm_mullo_pi16(y0, y0_fraction);
-    y1 = _mm_mullo_pi16(y1, y1_fraction);
-    y2 = _mm_mullo_pi16(y2, y0_fraction);
-    y3 = _mm_mullo_pi16(y3, y1_fraction);
-    y0 = _mm_add_pi16(y0, y1);
-    y2 = _mm_add_pi16(y2, y3);
-    y0 = _mm_srli_pi16(y0, 8);
-    y2 = _mm_srli_pi16(y2, 8);
-    y0 = _mm_packs_pu16(y0, y2);
-    *dest64++ = y0;
-  } while (dest64 < end64);
-}
-#else  // no MMX or SSE2
+  if (has_sse)
+    EMMS();
+}
+
 // C version does 8 at a time to mimic MMX code
-static void FilterRows(uint8* ybuf, const uint8* y0_ptr, const uint8* y1_ptr,
-                       int source_width, int source_y_fraction) {
+static void FilterRows_C(uint8* ybuf, const uint8* y0_ptr, const uint8* y1_ptr,
+                         int source_width, int source_y_fraction) {
   int y1_fraction = source_y_fraction;
   int y0_fraction = 256 - y1_fraction;
   uint8* end = ybuf + source_width;
   do {
     ybuf[0] = (y0_ptr[0] * y0_fraction + y1_ptr[0] * y1_fraction) >> 8;
     ybuf[1] = (y0_ptr[1] * y0_fraction + y1_ptr[1] * y1_fraction) >> 8;
     ybuf[2] = (y0_ptr[2] * y0_fraction + y1_ptr[2] * y1_fraction) >> 8;
     ybuf[3] = (y0_ptr[3] * y0_fraction + y1_ptr[3] * y1_fraction) >> 8;
@@ -152,46 +140,77 @@ static void FilterRows(uint8* ybuf, cons
     ybuf[5] = (y0_ptr[5] * y0_fraction + y1_ptr[5] * y1_fraction) >> 8;
     ybuf[6] = (y0_ptr[6] * y0_fraction + y1_ptr[6] * y1_fraction) >> 8;
     ybuf[7] = (y0_ptr[7] * y0_fraction + y1_ptr[7] * y1_fraction) >> 8;
     y0_ptr += 8;
     y1_ptr += 8;
     ybuf += 8;
   } while (ybuf < end);
 }
-#endif
+
+#ifdef MOZILLA_MAY_SUPPORT_MMX
+void FilterRows_MMX(uint8* ybuf, const uint8* y0_ptr, const uint8* y1_ptr,
+                    int source_width, int source_y_fraction);
+#endif
+
+#ifdef MOZILLA_MAY_SUPPORT_SSE2
+void FilterRows_SSE2(uint8* ybuf, const uint8* y0_ptr, const uint8* y1_ptr,
+                     int source_width, int source_y_fraction);
+#endif
+
+static inline void FilterRows(uint8* ybuf, const uint8* y0_ptr,
+                              const uint8* y1_ptr, int source_width,
+                              int source_y_fraction) {
+#ifdef MOZILLA_MAY_SUPPORT_SSE2
+  if (mozilla::supports_sse2()) {
+    FilterRows_SSE2(ybuf, y0_ptr, y1_ptr, source_width, source_y_fraction);
+    return;
+  }
+#endif
+
+#ifdef MOZILLA_MAY_SUPPORT_MMX
+  if (mozilla::supports_mmx()) {
+    FilterRows_MMX(ybuf, y0_ptr, y1_ptr, source_width, source_y_fraction);
+    return;
+  }
+#endif
+
+  FilterRows_C(ybuf, y0_ptr, y1_ptr, source_width, source_y_fraction);
+}
 
 
 // Scale a frame of YUV to 32 bit ARGB.
-void ScaleYUVToRGB32(const uint8* y_buf,
-                     const uint8* u_buf,
-                     const uint8* v_buf,
-                     uint8* rgb_buf,
-                     int source_width,
-                     int source_height,
-                     int width,
-                     int height,
-                     int y_pitch,
-                     int uv_pitch,
-                     int rgb_pitch,
-                     YUVType yuv_type,
-                     Rotate view_rotate,
-                     ScaleFilter filter) {
+NS_GFX_(void) ScaleYCbCrToRGB32(const uint8* y_buf,
+                                const uint8* u_buf,
+                                const uint8* v_buf,
+                                uint8* rgb_buf,
+                                int source_width,
+                                int source_height,
+                                int width,
+                                int height,
+                                int y_pitch,
+                                int uv_pitch,
+                                int rgb_pitch,
+                                YUVType yuv_type,
+                                Rotate view_rotate,
+                                ScaleFilter filter) {
+  bool has_mmx = supports_mmx();
+
   // 4096 allows 3 buffers to fit in 12k.
   // Helps performance on CPU with 16K L1 cache.
   // Large enough for 3830x2160 and 30" displays which are 2560x1600.
   const int kFilterBufferSize = 4096;
   // Disable filtering if the screen is too big (to avoid buffer overflows).
   // This should never happen to regular users: they don't have monitors
   // wider than 4096 pixels.
   // TODO(fbarchard): Allow rotated videos to filter.
   if (source_width > kFilterBufferSize || view_rotate)
     filter = FILTER_NONE;
 
-  unsigned int y_shift = yuv_type;
+  unsigned int y_shift = yuv_type == YV12 ? 1 : 0;
   // Diagram showing origin and direction of source sampling.
   // ->0   4<-
   // 7       3
   //
   // 6       5
   // ->1   2<-
   // Rotations that start at right side of image.
   if ((view_rotate == ROTATE_180) ||
@@ -276,17 +295,17 @@ void ScaleYUVToRGB32(const uint8* y_buf,
     int source_uv_fraction =
         ((source_y_subpixel >> y_shift) & kFractionMask) >> 8;
 
     const uint8* y_ptr = y0_ptr;
     const uint8* u_ptr = u0_ptr;
     const uint8* v_ptr = v0_ptr;
     // Apply vertical filtering if necessary.
     // TODO(fbarchard): Remove memcpy when not necessary.
-    if (filter & media::FILTER_BILINEAR_V) {
+    if (filter & mozilla::gfx::FILTER_BILINEAR_V) {
       if (yscale_fixed != kFractionMax &&
           source_y_fraction && ((source_y + 1) < source_height)) {
         FilterRows(ybuf, y0_ptr, y1_ptr, source_width, source_y_fraction);
       } else {
         memcpy(ybuf, y0_ptr, source_width);
       }
       y_ptr = ybuf;
       ybuf[source_width] = ybuf[source_width-1];
@@ -303,44 +322,50 @@ void ScaleYUVToRGB32(const uint8* y_buf,
       u_ptr = ubuf;
       v_ptr = vbuf;
       ubuf[uv_source_width] = ubuf[uv_source_width - 1];
       vbuf[uv_source_width] = vbuf[uv_source_width - 1];
     }
     if (source_dx == kFractionMax) {  // Not scaled
       FastConvertYUVToRGB32Row(y_ptr, u_ptr, v_ptr,
                                dest_pixel, width);
-    } else {
-      if (filter & FILTER_BILINEAR_H) {
+    } else if (filter & FILTER_BILINEAR_H) {
         LinearScaleYUVToRGB32Row(y_ptr, u_ptr, v_ptr,
                                  dest_pixel, width, source_dx);
     } else {
 // Specialized scalers and rotation.
-#if USE_MMX && defined(_MSC_VER)
+#if defined(MOZILLA_MAY_SUPPORT_SSE) && defined(_MSC_VER) && defined(_M_IX86)
+      if(mozilla::supports_sse()) {
         if (width == (source_width * 2)) {
-          DoubleYUVToRGB32Row(y_ptr, u_ptr, v_ptr,
-                              dest_pixel, width);
+          DoubleYUVToRGB32Row_SSE(y_ptr, u_ptr, v_ptr,
+                                  dest_pixel, width);
         } else if ((source_dx & kFractionMask) == 0) {
           // Scaling by integer scale factor. ie half.
-          ConvertYUVToRGB32Row(y_ptr, u_ptr, v_ptr,
-                               dest_pixel, width,
-                               source_dx >> kFractionBits);
+          ConvertYUVToRGB32Row_SSE(y_ptr, u_ptr, v_ptr,
+                                   dest_pixel, width,
+                                   source_dx >> kFractionBits);
         } else if (source_dx_uv == source_dx) {  // Not rotated.
           ScaleYUVToRGB32Row(y_ptr, u_ptr, v_ptr,
                              dest_pixel, width, source_dx);
         } else {
-          RotateConvertYUVToRGB32Row(y_ptr, u_ptr, v_ptr,
-                                     dest_pixel, width,
-                                     source_dx >> kFractionBits,
-                                     source_dx_uv >> kFractionBits);
+          RotateConvertYUVToRGB32Row_SSE(y_ptr, u_ptr, v_ptr,
+                                         dest_pixel, width,
+                                         source_dx >> kFractionBits,
+                                         source_dx_uv >> kFractionBits);
         }
+      }
+      else {
+        ScaleYUVToRGB32Row_C(y_ptr, u_ptr, v_ptr,
+                             dest_pixel, width, source_dx);
+      }
 #else
-        ScaleYUVToRGB32Row(y_ptr, u_ptr, v_ptr,
-                           dest_pixel, width, source_dx);
-#endif
-      }
+      ScaleYUVToRGB32Row(y_ptr, u_ptr, v_ptr,
+                         dest_pixel, width, source_dx);
+#endif
     }
   }
   // MMX used for FastConvertYUVToRGB32Row and FilterRows requires emms.
-  EMMS();
-}
-
-}  // namespace media
+  if (has_mmx)
+    EMMS();
+}
+
+}  // namespace gfx
+}  // namespace mozilla
diff --git a/gfx/ycbcr/yuv_convert.h b/gfx/ycbcr/yuv_convert.h
--- a/gfx/ycbcr/yuv_convert.h
+++ b/gfx/ycbcr/yuv_convert.h
@@ -1,72 +1,79 @@
 // Copyright (c) 2010 The Chromium Authors. All rights reserved.
 // Use of this source code is governed by a BSD-style license that can be
 // found in the LICENSE file.
 
 #ifndef MEDIA_BASE_YUV_CONVERT_H_
 #define MEDIA_BASE_YUV_CONVERT_H_
 
-#include "base/basictypes.h"
-
-namespace media {
-
+#include "chromium_types.h"
+#include "gfxCore.h"
+
+namespace mozilla {
+
+namespace gfx {
+ 
 // Type of YUV surface.
 // The value of these enums matter as they are used to shift vertical indices.
 enum YUVType {
-  YV16 = 0,           // YV16 is half width and full height chroma channels.
-  YV12 = 1,           // YV12 is half width and half height chroma channels.
+  YV12 = 0,           // YV12 is half width and half height chroma channels.
+  YV16 = 1,           // YV16 is half width and full height chroma channels.
+  YV24 = 2            // YV24 is full width and full height chroma channels.
 };
 
 // Mirror means flip the image horizontally, as in looking in a mirror.
 // Rotate happens after mirroring.
 enum Rotate {
   ROTATE_0,           // Rotation off.
   ROTATE_90,          // Rotate clockwise.
   ROTATE_180,         // Rotate upside down.
   ROTATE_270,         // Rotate counter clockwise.
   MIRROR_ROTATE_0,    // Mirror horizontally.
   MIRROR_ROTATE_90,   // Mirror then Rotate clockwise.
   MIRROR_ROTATE_180,  // Mirror vertically.
-  MIRROR_ROTATE_270,  // Transpose.
+  MIRROR_ROTATE_270   // Transpose.
 };
 
 // Filter affects how scaling looks.
 enum ScaleFilter {
   FILTER_NONE = 0,        // No filter (point sampled).
   FILTER_BILINEAR_H = 1,  // Bilinear horizontal filter.
   FILTER_BILINEAR_V = 2,  // Bilinear vertical filter.
-  FILTER_BILINEAR = 3,    // Bilinear filter.
+  FILTER_BILINEAR = 3     // Bilinear filter.
 };
 
 // Convert a frame of YUV to 32 bit ARGB.
 // Pass in YV16/YV12 depending on source format
-void ConvertYUVToRGB32(const uint8* yplane,
-                       const uint8* uplane,
-                       const uint8* vplane,
-                       uint8* rgbframe,
-                       int width,
-                       int height,
-                       int ystride,
-                       int uvstride,
-                       int rgbstride,
-                       YUVType yuv_type);
+NS_GFX_(void) ConvertYCbCrToRGB32(const uint8* yplane,
+                                  const uint8* uplane,
+                                  const uint8* vplane,
+                                  uint8* rgbframe,
+                                  int pic_x,
+                                  int pic_y,
+                                  int pic_width,
+                                  int pic_height,
+                                  int ystride,
+                                  int uvstride,
+                                  int rgbstride,
+                                  YUVType yuv_type);
 
 // Scale a frame of YUV to 32 bit ARGB.
 // Supports rotation and mirroring.
-void ScaleYUVToRGB32(const uint8* yplane,
-                     const uint8* uplane,
-                     const uint8* vplane,
-                     uint8* rgbframe,
-                     int source_width,
-                     int source_height,
-                     int width,
-                     int height,
-                     int ystride,
-                     int uvstride,
-                     int rgbstride,
-                     YUVType yuv_type,
-                     Rotate view_rotate,
-                     ScaleFilter filter);
-
-}  // namespace media
-
+NS_GFX_(void) ScaleYCbCrToRGB32(const uint8* yplane,
+                                const uint8* uplane,
+                                const uint8* vplane,
+                                uint8* rgbframe,
+                                int source_width,
+                                int source_height,
+                                int width,
+                                int height,
+                                int ystride,
+                                int uvstride,
+                                int rgbstride,
+                                YUVType yuv_type,
+                                Rotate view_rotate,
+                                ScaleFilter filter);
+
+}  // namespace gfx
+}  // namespace mozilla
+ 
 #endif  // MEDIA_BASE_YUV_CONVERT_H_
diff --git a/gfx/ycbcr/yuv_convert_mmx.cpp b/gfx/ycbcr/yuv_convert_mmx.cpp
new file mode 100644
--- /dev/null
+++ b/gfx/ycbcr/yuv_convert_mmx.cpp
@@ -0,0 +1,45 @@
+// Copyright (c) 2010 The Chromium Authors. All rights reserved.
+// Use of this source code is governed by a BSD-style license that can be
+// found in the LICENSE file.
+
+#include <mmintrin.h>
+#include "yuv_row.h"
+
+namespace mozilla {
+namespace gfx {
+
+// FilterRows combines two rows of the image using linear interpolation.
+// MMX version does 8 pixels at a time.
+void FilterRows_MMX(uint8* ybuf, const uint8* y0_ptr, const uint8* y1_ptr,
+                    int source_width, int source_y_fraction) {
+  __m64 zero = _mm_setzero_si64();
+  __m64 y1_fraction = _mm_set1_pi16(source_y_fraction);
+  __m64 y0_fraction = _mm_set1_pi16(256 - source_y_fraction);
+
+  const __m64* y0_ptr64 = reinterpret_cast<const __m64*>(y0_ptr);
+  const __m64* y1_ptr64 = reinterpret_cast<const __m64*>(y1_ptr);
+  __m64* dest64 = reinterpret_cast<__m64*>(ybuf);
+  __m64* end64 = reinterpret_cast<__m64*>(ybuf + source_width);
+
+  do {
+    __m64 y0 = *y0_ptr64++;
+    __m64 y1 = *y1_ptr64++;
+    __m64 y2 = _mm_unpackhi_pi8(y0, zero);
+    __m64 y3 = _mm_unpackhi_pi8(y1, zero);
+    y0 = _mm_unpacklo_pi8(y0, zero);
+    y1 = _mm_unpacklo_pi8(y1, zero);
+    y0 = _mm_mullo_pi16(y0, y0_fraction);
+    y1 = _mm_mullo_pi16(y1, y1_fraction);
+    y2 = _mm_mullo_pi16(y2, y0_fraction);
+    y3 = _mm_mullo_pi16(y3, y1_fraction);
+    y0 = _mm_add_pi16(y0, y1);
+    y2 = _mm_add_pi16(y2, y3);
+    y0 = _mm_srli_pi16(y0, 8);
+    y2 = _mm_srli_pi16(y2, 8);
+    y0 = _mm_packs_pu16(y0, y2);
+    *dest64++ = y0;
+  } while (dest64 < end64);
+}
+
+}
+}
diff --git a/gfx/ycbcr/yuv_convert_sse2.cpp b/gfx/ycbcr/yuv_convert_sse2.cpp
new file mode 100644
--- /dev/null
+++ b/gfx/ycbcr/yuv_convert_sse2.cpp
@@ -0,0 +1,47 @@
+// Copyright (c) 2010 The Chromium Authors. All rights reserved.
+// Use of this source code is governed by a BSD-style license that can be
+// found in the LICENSE file.
+
+#include <emmintrin.h>
+#include "yuv_row.h"
+
+namespace mozilla {
+namespace gfx {
+
+// FilterRows combines two rows of the image using linear interpolation.
+// SSE2 version does 16 pixels at a time.
+void FilterRows_SSE2(uint8* ybuf, const uint8* y0_ptr, const uint8* y1_ptr,
+                     int source_width, int source_y_fraction) {
+  __m128i zero = _mm_setzero_si128();
+  __m128i y1_fraction = _mm_set1_epi16(source_y_fraction);
+  __m128i y0_fraction = _mm_set1_epi16(256 - source_y_fraction);
+
+  const __m128i* y0_ptr128 = reinterpret_cast<const __m128i*>(y0_ptr);
+  const __m128i* y1_ptr128 = reinterpret_cast<const __m128i*>(y1_ptr);
+  __m128i* dest128 = reinterpret_cast<__m128i*>(ybuf);
+  __m128i* end128 = reinterpret_cast<__m128i*>(ybuf + source_width);
+
+  do {
+    __m128i y0 = _mm_loadu_si128(y0_ptr128);
+    __m128i y1 = _mm_loadu_si128(y1_ptr128);
+    __m128i y2 = _mm_unpackhi_epi8(y0, zero);
+    __m128i y3 = _mm_unpackhi_epi8(y1, zero);
+    y0 = _mm_unpacklo_epi8(y0, zero);
+    y1 = _mm_unpacklo_epi8(y1, zero);
+    y0 = _mm_mullo_epi16(y0, y0_fraction);
+    y1 = _mm_mullo_epi16(y1, y1_fraction);
+    y2 = _mm_mullo_epi16(y2, y0_fraction);
+    y3 = _mm_mullo_epi16(y3, y1_fraction);
+    y0 = _mm_add_epi16(y0, y1);
+    y2 = _mm_add_epi16(y2, y3);
+    y0 = _mm_srli_epi16(y0, 8);
+    y2 = _mm_srli_epi16(y2, 8);
+    y0 = _mm_packus_epi16(y0, y2);
+    *dest128++ = y0;
+    ++y0_ptr128;
+    ++y1_ptr128;
+  } while (dest128 < end128);
+}
+
+}
+}
diff --git a/gfx/ycbcr/yuv_row.h b/gfx/ycbcr/yuv_row.h
--- a/gfx/ycbcr/yuv_row.h
+++ b/gfx/ycbcr/yuv_row.h
@@ -5,109 +5,133 @@
 // yuv_row internal functions to handle YUV conversion and scaling to RGB.
 // These functions are used from both yuv_convert.cc and yuv_scale.cc.
 
 // TODO(fbarchard): Write function that can handle rotation and scaling.
 
 #ifndef MEDIA_BASE_YUV_ROW_H_
 #define MEDIA_BASE_YUV_ROW_H_
 
-#include "base/basictypes.h"
+#include "chromium_types.h"
 
 extern "C" {
 // Can only do 1x.
 // This is the second fastest of the scalers.
 void FastConvertYUVToRGB32Row(const uint8* y_buf,
                               const uint8* u_buf,
                               const uint8* v_buf,
                               uint8* rgb_buf,
                               int width);
 
-// Can do 1x, half size or any scale down by an integer amount.
-// Step can be negative (mirroring, rotate 180).
-// This is the third fastest of the scalers.
-void ConvertYUVToRGB32Row(const uint8* y_buf,
-                          const uint8* u_buf,
-                          const uint8* v_buf,
-                          uint8* rgb_buf,
-                          int width,
-                          int step);
-
-// Rotate is like Convert, but applies different step to Y versus U and V.
-// This allows rotation by 90 or 270, by stepping by stride.
-// This is the forth fastest of the scalers.
-void RotateConvertYUVToRGB32Row(const uint8* y_buf,
+void FastConvertYUVToRGB32Row_C(const uint8* y_buf,
                                 const uint8* u_buf,
                                 const uint8* v_buf,
                                 uint8* rgb_buf,
                                 int width,
-                                int ystep,
-                                int uvstep);
+                                unsigned int x_shift);
+
+void FastConvertYUVToRGB32Row(const uint8* y_buf,
+                              const uint8* u_buf,
+                              const uint8* v_buf,
+                              uint8* rgb_buf,
+                              int width);
+
+// Can do 1x, half size or any scale down by an integer amount.
+// Step can be negative (mirroring, rotate 180).
+// This is the third fastest of the scalers.
+// Only defined on Windows x86-32.
+void ConvertYUVToRGB32Row_SSE(const uint8* y_buf,
+                              const uint8* u_buf,
+                              const uint8* v_buf,
+                              uint8* rgb_buf,
+                              int width,
+                              int step);
+
+// Rotate is like Convert, but applies different step to Y versus U and V.
+// This allows rotation by 90 or 270, by stepping by stride.
+// This is the forth fastest of the scalers.
+// Only defined on Windows x86-32.
+void RotateConvertYUVToRGB32Row_SSE(const uint8* y_buf,
+                                    const uint8* u_buf,
+                                    const uint8* v_buf,
+                                    uint8* rgb_buf,
+                                    int width,
+                                    int ystep,
+                                    int uvstep);
 
 // Doubler does 4 pixels at a time.  Each pixel is replicated.
 // This is the fastest of the scalers.
-void DoubleYUVToRGB32Row(const uint8* y_buf,
-                         const uint8* u_buf,
-                         const uint8* v_buf,
-                         uint8* rgb_buf,
-                         int width);
+// Only defined on Windows x86-32.
+void DoubleYUVToRGB32Row_SSE(const uint8* y_buf,
+                             const uint8* u_buf,
+                             const uint8* v_buf,
+                             uint8* rgb_buf,
+                             int width);
 
 // Handles arbitrary scaling up or down.
 // Mirroring is supported, but not 90 or 270 degree rotation.
 // Chroma is under sampled every 2 pixels for performance.
 void ScaleYUVToRGB32Row(const uint8* y_buf,
                         const uint8* u_buf,
                         const uint8* v_buf,
                         uint8* rgb_buf,
                         int width,
                         int source_dx);
 
+void ScaleYUVToRGB32Row(const uint8* y_buf,
+                        const uint8* u_buf,
+                        const uint8* v_buf,
+                        uint8* rgb_buf,
+                        int width,
+                        int source_dx);
+
+void ScaleYUVToRGB32Row_C(const uint8* y_buf,
+                          const uint8* u_buf,
+                          const uint8* v_buf,
+                          uint8* rgb_buf,
+                          int width,
+                          int source_dx);
+
 // Handles arbitrary scaling up or down with bilinear filtering.
 // Mirroring is supported, but not 90 or 270 degree rotation.
 // Chroma is under sampled every 2 pixels for performance.
 // This is the slowest of the scalers.
 void LinearScaleYUVToRGB32Row(const uint8* y_buf,
                               const uint8* u_buf,
                               const uint8* v_buf,
                               uint8* rgb_buf,
                               int width,
                               int source_dx);
 
+void LinearScaleYUVToRGB32Row(const uint8* y_buf,
+                              const uint8* u_buf,
+                              const uint8* v_buf,
+                              uint8* rgb_buf,
+                              int width,
+                              int source_dx);
+
+void LinearScaleYUVToRGB32Row_C(const uint8* y_buf,
+                                const uint8* u_buf,
+                                const uint8* v_buf,
+                                uint8* rgb_buf,
+                                int width,
+                                int source_dx);
+
+
 #if defined(_MSC_VER)
 #define SIMD_ALIGNED(var) __declspec(align(16)) var
 #else
 #define SIMD_ALIGNED(var) var __attribute__((aligned(16)))
 #endif
 extern SIMD_ALIGNED(int16 kCoefficientsRgbY[768][4]);
 
-// Method to force C version.
-//#define USE_MMX 0
-//#define USE_SSE2 0
-
-#if !defined(USE_MMX)
-// Windows, Mac and Linux/BSD use MMX
-#if defined(__MMX__) || defined(_MSC_VER)
-#define USE_MMX 1
-#else
-#define USE_MMX 0
-#endif
-#endif
-
-#if !defined(USE_SSE2)
-#if defined(__SSE2__) || defined(ARCH_CPU_X86_64) || _M_IX86_FP==2
-#define USE_SSE2 1
-#else
-#define USE_SSE2 0
-#endif
-#endif
-
 // x64 uses MMX2 (SSE) so emms is not required.
 // Warning C4799: function has no EMMS instruction.
 // EMMS() is slow and should be called by the calling function once per image.
-#if USE_MMX && !defined(ARCH_CPU_X86_64)
+#if defined(ARCH_CPU_X86) && !defined(ARCH_CPU_X86_64)
 #if defined(_MSC_VER)
 #define EMMS() __asm emms
 #pragma warning(disable: 4799)
 #else
 #define EMMS() asm("emms")
 #endif
 #else
 #define EMMS()
diff --git a/gfx/ycbcr/yuv_row_c.cpp b/gfx/ycbcr/yuv_row_c.cpp
--- a/gfx/ycbcr/yuv_row_c.cpp
+++ b/gfx/ycbcr/yuv_row_c.cpp
@@ -1,812 +1,18 @@
 // Copyright (c) 2010 The Chromium Authors. All rights reserved.
 // Use of this source code is governed by a BSD-style license that can be
 // found in the LICENSE file.
 
-#include "media/base/yuv_row.h"
-
-#ifdef _DEBUG
-#include "base/logging.h"
-#else
+#include "yuv_row.h"
+
 #define DCHECK(a)
-#endif
 
 extern "C" {
 
-#if USE_SSE2 && defined(ARCH_CPU_X86_64)
-
-// AMD64 ABI uses register paremters.
-void FastConvertYUVToRGB32Row(const uint8* y_buf,  // rdi
-                              const uint8* u_buf,  // rsi
-                              const uint8* v_buf,  // rdx
-                              uint8* rgb_buf,      // rcx
-                              int width) {         // r8
-  asm(
-  "jmp    convertend\n"
-"convertloop:"
-  "movzb  (%1),%%r10\n"
-  "add    $0x1,%1\n"
-  "movzb  (%2),%%r11\n"
-  "add    $0x1,%2\n"
-  "movq   2048(%5,%%r10,8),%%xmm0\n"
-  "movzb  (%0),%%r10\n"
-  "movq   4096(%5,%%r11,8),%%xmm1\n"
-  "movzb  0x1(%0),%%r11\n"
-  "paddsw %%xmm1,%%xmm0\n"
-  "movq   (%5,%%r10,8),%%xmm2\n"
-  "add    $0x2,%0\n"
-  "movq   (%5,%%r11,8),%%xmm3\n"
-  "paddsw %%xmm0,%%xmm2\n"
-  "paddsw %%xmm0,%%xmm3\n"
-  "shufps $0x44,%%xmm3,%%xmm2\n"
-  "psraw  $0x6,%%xmm2\n"
-  "packuswb %%xmm2,%%xmm2\n"
-  "movq   %%xmm2,0x0(%3)\n"
-  "add    $0x8,%3\n"
-"convertend:"
-  "sub    $0x2,%4\n"
-  "jns    convertloop\n"
-
-"convertnext:"
-  "add    $0x1,%4\n"
-  "js     convertdone\n"
-
-  "movzb  (%1),%%r10\n"
-  "movq   2048(%5,%%r10,8),%%xmm0\n"
-  "movzb  (%2),%%r10\n"
-  "movq   4096(%5,%%r10,8),%%xmm1\n"
-  "paddsw %%xmm1,%%xmm0\n"
-  "movzb  (%0),%%r10\n"
-  "movq   (%5,%%r10,8),%%xmm1\n"
-  "paddsw %%xmm0,%%xmm1\n"
-  "psraw  $0x6,%%xmm1\n"
-  "packuswb %%xmm1,%%xmm1\n"
-  "movd   %%xmm1,0x0(%3)\n"
-"convertdone:"
-  :
-  : "r"(y_buf),  // %0
-    "r"(u_buf),  // %1
-    "r"(v_buf),  // %2
-    "r"(rgb_buf),  // %3
-    "r"(width),  // %4
-    "r" (kCoefficientsRgbY)  // %5
-  : "memory", "r10", "r11", "xmm0", "xmm1", "xmm2", "xmm3"
-);
-}
-
-void ScaleYUVToRGB32Row(const uint8* y_buf,  // rdi
-                        const uint8* u_buf,  // rsi
-                        const uint8* v_buf,  // rdx
-                        uint8* rgb_buf,      // rcx
-                        int width,           // r8
-                        int source_dx) {     // r9
-  asm(
-  "xor    %%r11,%%r11\n"
-  "sub    $0x2,%4\n"
-  "js     scalenext\n"
-
-"scaleloop:"
-  "mov    %%r11,%%r10\n"
-  "sar    $0x11,%%r10\n"
-  "movzb  (%1,%%r10,1),%%rax\n"
-  "movq   2048(%5,%%rax,8),%%xmm0\n"
-  "movzb  (%2,%%r10,1),%%rax\n"
-  "movq   4096(%5,%%rax,8),%%xmm1\n"
-  "lea    (%%r11,%6),%%r10\n"
-  "sar    $0x10,%%r11\n"
-  "movzb  (%0,%%r11,1),%%rax\n"
-  "paddsw %%xmm1,%%xmm0\n"
-  "movq   (%5,%%rax,8),%%xmm1\n"
-  "lea    (%%r10,%6),%%r11\n"
-  "sar    $0x10,%%r10\n"
-  "movzb  (%0,%%r10,1),%%rax\n"
-  "movq   (%5,%%rax,8),%%xmm2\n"
-  "paddsw %%xmm0,%%xmm1\n"
-  "paddsw %%xmm0,%%xmm2\n"
-  "shufps $0x44,%%xmm2,%%xmm1\n"
-  "psraw  $0x6,%%xmm1\n"
-  "packuswb %%xmm1,%%xmm1\n"
-  "movq   %%xmm1,0x0(%3)\n"
-  "add    $0x8,%3\n"
-  "sub    $0x2,%4\n"
-  "jns    scaleloop\n"
-
-"scalenext:"
-  "add    $0x1,%4\n"
-  "js     scaledone\n"
-
-  "mov    %%r11,%%r10\n"
-  "sar    $0x11,%%r10\n"
-  "movzb  (%1,%%r10,1),%%rax\n"
-  "movq   2048(%5,%%rax,8),%%xmm0\n"
-  "movzb  (%2,%%r10,1),%%rax\n"
-  "movq   4096(%5,%%rax,8),%%xmm1\n"
-  "paddsw %%xmm1,%%xmm0\n"
-  "sar    $0x10,%%r11\n"
-  "movzb  (%0,%%r11,1),%%rax\n"
-  "movq   (%5,%%rax,8),%%xmm1\n"
-  "paddsw %%xmm0,%%xmm1\n"
-  "psraw  $0x6,%%xmm1\n"
-  "packuswb %%xmm1,%%xmm1\n"
-  "movd   %%xmm1,0x0(%3)\n"
-
-"scaledone:"
-  :
-  : "r"(y_buf),  // %0
-    "r"(u_buf),  // %1
-    "r"(v_buf),  // %2
-    "r"(rgb_buf),  // %3
-    "r"(width),  // %4
-    "r" (kCoefficientsRgbY),  // %5
-    "r"(static_cast<long>(source_dx))  // %6
-  : "memory", "r10", "r11", "rax", "xmm0", "xmm1", "xmm2"
-);
-}
-
-void LinearScaleYUVToRGB32Row(const uint8* y_buf,
-                              const uint8* u_buf,
-                              const uint8* v_buf,
-                              uint8* rgb_buf,
-                              int width,
-                              int source_dx) {
-  asm(
-  "xor    %%r11,%%r11\n"   // x = 0
-  "sub    $0x2,%4\n"
-  "js     .lscalenext\n"
-  "cmp    $0x20000,%6\n"   // if source_dx >= 2.0
-  "jl     .lscalehalf\n"
-  "mov    $0x8000,%%r11\n" // x = 0.5 for 1/2 or less
-".lscalehalf:"
-
-".lscaleloop:"
-  "mov    %%r11,%%r10\n"
-  "sar    $0x11,%%r10\n"
-
-  "movzb  (%1, %%r10, 1), %%r13 \n"
-  "movzb  1(%1, %%r10, 1), %%r14 \n"
-  "mov    %%r11, %%rax \n"
-  "and    $0x1fffe, %%rax \n"
-  "imul   %%rax, %%r14 \n"
-  "xor    $0x1fffe, %%rax \n"
-  "imul   %%rax, %%r13 \n"
-  "add    %%r14, %%r13 \n"
-  "shr    $17, %%r13 \n"
-  "movq   2048(%5,%%r13,8), %%xmm0\n"
-
-  "movzb  (%2, %%r10, 1), %%r13 \n"
-  "movzb  1(%2, %%r10, 1), %%r14 \n"
-  "mov    %%r11, %%rax \n"
-  "and    $0x1fffe, %%rax \n"
-  "imul   %%rax, %%r14 \n"
-  "xor    $0x1fffe, %%rax \n"
-  "imul   %%rax, %%r13 \n"
-  "add    %%r14, %%r13 \n"
-  "shr    $17, %%r13 \n"
-  "movq   4096(%5,%%r13,8), %%xmm1\n"
-
-  "mov    %%r11, %%rax \n"
-  "lea    (%%r11,%6),%%r10\n"
-  "sar    $0x10,%%r11\n"
-  "paddsw %%xmm1,%%xmm0\n"
-
-  "movzb  (%0, %%r11, 1), %%r13 \n"
-  "movzb  1(%0, %%r11, 1), %%r14 \n"
-  "and    $0xffff, %%rax \n"
-  "imul   %%rax, %%r14 \n"
-  "xor    $0xffff, %%rax \n"
-  "imul   %%rax, %%r13 \n"
-  "add    %%r14, %%r13 \n"
-  "shr    $16, %%r13 \n"
-  "movq   (%5,%%r13,8),%%xmm1\n"
-
-  "mov    %%r10, %%rax \n"
-  "lea    (%%r10,%6),%%r11\n"
-  "sar    $0x10,%%r10\n"
-
-  "movzb  (%0,%%r10,1), %%r13 \n"
-  "movzb  1(%0,%%r10,1), %%r14 \n"
-  "and    $0xffff, %%rax \n"
-  "imul   %%rax, %%r14 \n"
-  "xor    $0xffff, %%rax \n"
-  "imul   %%rax, %%r13 \n"
-  "add    %%r14, %%r13 \n"
-  "shr    $16, %%r13 \n"
-  "movq   (%5,%%r13,8),%%xmm2\n"
-
-  "paddsw %%xmm0,%%xmm1\n"
-  "paddsw %%xmm0,%%xmm2\n"
-  "shufps $0x44,%%xmm2,%%xmm1\n"
-  "psraw  $0x6,%%xmm1\n"
-  "packuswb %%xmm1,%%xmm1\n"
-  "movq   %%xmm1,0x0(%3)\n"
-  "add    $0x8,%3\n"
-  "sub    $0x2,%4\n"
-  "jns    .lscaleloop\n"
-
-".lscalenext:"
-  "add    $0x1,%4\n"
-  "js     .lscaledone\n"
-
-  "mov    %%r11,%%r10\n"
-  "sar    $0x11,%%r10\n"
-
-  "movzb  (%1,%%r10,1), %%r13 \n"
-  "movq   2048(%5,%%r13,8),%%xmm0\n"
-
-  "movzb  (%2,%%r10,1), %%r13 \n"
-  "movq   4096(%5,%%r13,8),%%xmm1\n"
-
-  "paddsw %%xmm1,%%xmm0\n"
-  "sar    $0x10,%%r11\n"
-
-  "movzb  (%0,%%r11,1), %%r13 \n"
-  "movq   (%5,%%r13,8),%%xmm1\n"
-
-  "paddsw %%xmm0,%%xmm1\n"
-  "psraw  $0x6,%%xmm1\n"
-  "packuswb %%xmm1,%%xmm1\n"
-  "movd   %%xmm1,0x0(%3)\n"
-
-".lscaledone:"
-  :
-  : "r"(y_buf),  // %0
-    "r"(u_buf),  // %1
-    "r"(v_buf),  // %2
-    "r"(rgb_buf),  // %3
-    "r"(width),  // %4
-    "r" (kCoefficientsRgbY),  // %5
-    "r"(static_cast<long>(source_dx))  // %6
-  : "memory", "r10", "r11", "r13", "r14", "rax", "xmm0", "xmm1", "xmm2"
-);
-}
-
-#elif USE_MMX && !defined(ARCH_CPU_X86_64) && !defined(__PIC__)
-
-// PIC version is slower because less registers are available, so
-// non-PIC is used on platforms where it is possible.
-
-void FastConvertYUVToRGB32Row(const uint8* y_buf,
-                              const uint8* u_buf,
-                              const uint8* v_buf,
-                              uint8* rgb_buf,
-                              int width);
-  asm(
-  ".text\n"
-  ".global FastConvertYUVToRGB32Row\n"
-"FastConvertYUVToRGB32Row:\n"
-  "pusha\n"
-  "mov    0x24(%esp),%edx\n"
-  "mov    0x28(%esp),%edi\n"
-  "mov    0x2c(%esp),%esi\n"
-  "mov    0x30(%esp),%ebp\n"
-  "mov    0x34(%esp),%ecx\n"
-  "jmp    convertend\n"
-
-"convertloop:"
-  "movzbl (%edi),%eax\n"
-  "add    $0x1,%edi\n"
-  "movzbl (%esi),%ebx\n"
-  "add    $0x1,%esi\n"
-  "movq   kCoefficientsRgbY+2048(,%eax,8),%mm0\n"
-  "movzbl (%edx),%eax\n"
-  "paddsw kCoefficientsRgbY+4096(,%ebx,8),%mm0\n"
-  "movzbl 0x1(%edx),%ebx\n"
-  "movq   kCoefficientsRgbY(,%eax,8),%mm1\n"
-  "add    $0x2,%edx\n"
-  "movq   kCoefficientsRgbY(,%ebx,8),%mm2\n"
-  "paddsw %mm0,%mm1\n"
-  "paddsw %mm0,%mm2\n"
-  "psraw  $0x6,%mm1\n"
-  "psraw  $0x6,%mm2\n"
-  "packuswb %mm2,%mm1\n"
-  "movntq %mm1,0x0(%ebp)\n"
-  "add    $0x8,%ebp\n"
-"convertend:"
-  "sub    $0x2,%ecx\n"
-  "jns    convertloop\n"
-
-  "and    $0x1,%ecx\n"
-  "je     convertdone\n"
-
-  "movzbl (%edi),%eax\n"
-  "movq   kCoefficientsRgbY+2048(,%eax,8),%mm0\n"
-  "movzbl (%esi),%eax\n"
-  "paddsw kCoefficientsRgbY+4096(,%eax,8),%mm0\n"
-  "movzbl (%edx),%eax\n"
-  "movq   kCoefficientsRgbY(,%eax,8),%mm1\n"
-  "paddsw %mm0,%mm1\n"
-  "psraw  $0x6,%mm1\n"
-  "packuswb %mm1,%mm1\n"
-  "movd   %mm1,0x0(%ebp)\n"
-"convertdone:"
-  "popa\n"
-  "ret\n"
-);
-
-
-void ScaleYUVToRGB32Row(const uint8* y_buf,
-                        const uint8* u_buf,
-                        const uint8* v_buf,
-                        uint8* rgb_buf,
-                        int width,
-                        int source_dx);
-  asm(
-  ".text\n"
-  ".global ScaleYUVToRGB32Row\n"
-"ScaleYUVToRGB32Row:\n"
-  "pusha\n"
-  "mov    0x24(%esp),%edx\n"
-  "mov    0x28(%esp),%edi\n"
-  "mov    0x2c(%esp),%esi\n"
-  "mov    0x30(%esp),%ebp\n"
-  "mov    0x34(%esp),%ecx\n"
-  "xor    %ebx,%ebx\n"
-  "jmp    scaleend\n"
-
-"scaleloop:"
-  "mov    %ebx,%eax\n"
-  "sar    $0x11,%eax\n"
-  "movzbl (%edi,%eax,1),%eax\n"
-  "movq   kCoefficientsRgbY+2048(,%eax,8),%mm0\n"
-  "mov    %ebx,%eax\n"
-  "sar    $0x11,%eax\n"
-  "movzbl (%esi,%eax,1),%eax\n"
-  "paddsw kCoefficientsRgbY+4096(,%eax,8),%mm0\n"
-  "mov    %ebx,%eax\n"
-  "add    0x38(%esp),%ebx\n"
-  "sar    $0x10,%eax\n"
-  "movzbl (%edx,%eax,1),%eax\n"
-  "movq   kCoefficientsRgbY(,%eax,8),%mm1\n"
-  "mov    %ebx,%eax\n"
-  "add    0x38(%esp),%ebx\n"
-  "sar    $0x10,%eax\n"
-  "movzbl (%edx,%eax,1),%eax\n"
-  "movq   kCoefficientsRgbY(,%eax,8),%mm2\n"
-  "paddsw %mm0,%mm1\n"
-  "paddsw %mm0,%mm2\n"
-  "psraw  $0x6,%mm1\n"
-  "psraw  $0x6,%mm2\n"
-  "packuswb %mm2,%mm1\n"
-  "movntq %mm1,0x0(%ebp)\n"
-  "add    $0x8,%ebp\n"
-"scaleend:"
-  "sub    $0x2,%ecx\n"
-  "jns    scaleloop\n"
-
-  "and    $0x1,%ecx\n"
-  "je     scaledone\n"
-
-  "mov    %ebx,%eax\n"
-  "sar    $0x11,%eax\n"
-  "movzbl (%edi,%eax,1),%eax\n"
-  "movq   kCoefficientsRgbY+2048(,%eax,8),%mm0\n"
-  "mov    %ebx,%eax\n"
-  "sar    $0x11,%eax\n"
-  "movzbl (%esi,%eax,1),%eax\n"
-  "paddsw kCoefficientsRgbY+4096(,%eax,8),%mm0\n"
-  "mov    %ebx,%eax\n"
-  "sar    $0x10,%eax\n"
-  "movzbl (%edx,%eax,1),%eax\n"
-  "movq   kCoefficientsRgbY(,%eax,8),%mm1\n"
-  "paddsw %mm0,%mm1\n"
-  "psraw  $0x6,%mm1\n"
-  "packuswb %mm1,%mm1\n"
-  "movd   %mm1,0x0(%ebp)\n"
-
-"scaledone:"
-  "popa\n"
-  "ret\n"
-);
-
-void LinearScaleYUVToRGB32Row(const uint8* y_buf,
-                              const uint8* u_buf,
-                              const uint8* v_buf,
-                              uint8* rgb_buf,
-                              int width,
-                              int source_dx);
-  asm(
-  ".text\n"
-  ".global LinearScaleYUVToRGB32Row\n"
-"LinearScaleYUVToRGB32Row:\n"
-  "pusha\n"
-  "mov    0x24(%esp),%edx\n"
-  "mov    0x28(%esp),%edi\n"
-  "mov    0x30(%esp),%ebp\n"
-
-  // source_width = width * source_dx + ebx
-  "mov    0x34(%esp), %ecx\n"
-  "imull  0x38(%esp), %ecx\n"
-  "mov    %ecx, 0x34(%esp)\n"
-
-  "mov    0x38(%esp), %ecx\n"
-  "xor    %ebx,%ebx\n"     // x = 0
-  "cmp    $0x20000,%ecx\n" // if source_dx >= 2.0
-  "jl     .lscaleend\n"
-  "mov    $0x8000,%ebx\n"  // x = 0.5 for 1/2 or less
-  "jmp    .lscaleend\n"
-
-".lscaleloop:"
-  "mov    %ebx,%eax\n"
-  "sar    $0x11,%eax\n"
-
-  "movzbl (%edi,%eax,1),%ecx\n"
-  "movzbl 1(%edi,%eax,1),%esi\n"
-  "mov    %ebx,%eax\n"
-  "andl   $0x1fffe, %eax \n"
-  "imul   %eax, %esi \n"
-  "xorl   $0x1fffe, %eax \n"
-  "imul   %eax, %ecx \n"
-  "addl   %esi, %ecx \n"
-  "shrl   $17, %ecx \n"
-  "movq   kCoefficientsRgbY+2048(,%ecx,8),%mm0\n"
-
-  "mov    0x2c(%esp),%esi\n"
-  "mov    %ebx,%eax\n"
-  "sar    $0x11,%eax\n"
-
-  "movzbl (%esi,%eax,1),%ecx\n"
-  "movzbl 1(%esi,%eax,1),%esi\n"
-  "mov    %ebx,%eax\n"
-  "andl   $0x1fffe, %eax \n"
-  "imul   %eax, %esi \n"
-  "xorl   $0x1fffe, %eax \n"
-  "imul   %eax, %ecx \n"
-  "addl   %esi, %ecx \n"
-  "shrl   $17, %ecx \n"
-  "paddsw kCoefficientsRgbY+4096(,%ecx,8),%mm0\n"
-
-  "mov    %ebx,%eax\n"
-  "sar    $0x10,%eax\n"
-  "movzbl (%edx,%eax,1),%ecx\n"
-  "movzbl 1(%edx,%eax,1),%esi\n"
-  "mov    %ebx,%eax\n"
-  "add    0x38(%esp),%ebx\n"
-  "andl   $0xffff, %eax \n"
-  "imul   %eax, %esi \n"
-  "xorl   $0xffff, %eax \n"
-  "imul   %eax, %ecx \n"
-  "addl   %esi, %ecx \n"
-  "shrl   $16, %ecx \n"
-  "movq   kCoefficientsRgbY(,%ecx,8),%mm1\n"
-
-  "cmp    0x34(%esp), %ebx\n"
-  "jge    .lscalelastpixel\n"
-
-  "mov    %ebx,%eax\n"
-  "sar    $0x10,%eax\n"
-  "movzbl (%edx,%eax,1),%ecx\n"
-  "movzbl 1(%edx,%eax,1),%esi\n"
-  "mov    %ebx,%eax\n"
-  "add    0x38(%esp),%ebx\n"
-  "andl   $0xffff, %eax \n"
-  "imul   %eax, %esi \n"
-  "xorl   $0xffff, %eax \n"
-  "imul   %eax, %ecx \n"
-  "addl   %esi, %ecx \n"
-  "shrl   $16, %ecx \n"
-  "movq   kCoefficientsRgbY(,%ecx,8),%mm2\n"
-
-  "paddsw %mm0,%mm1\n"
-  "paddsw %mm0,%mm2\n"
-  "psraw  $0x6,%mm1\n"
-  "psraw  $0x6,%mm2\n"
-  "packuswb %mm2,%mm1\n"
-  "movntq %mm1,0x0(%ebp)\n"
-  "add    $0x8,%ebp\n"
-
-".lscaleend:"
-  "cmp    0x34(%esp), %ebx\n"
-  "jl     .lscaleloop\n"
-  "popa\n"
-  "ret\n"
-
-".lscalelastpixel:"
-  "paddsw %mm0, %mm1\n"
-  "psraw $6, %mm1\n"
-  "packuswb %mm1, %mm1\n"
-  "movd %mm1, (%ebp)\n"
-  "popa\n"
-  "ret\n"
-);
-
-#elif USE_MMX && !defined(ARCH_CPU_X86_64) && defined(__PIC__)
-
-extern void PICConvertYUVToRGB32Row(const uint8* y_buf,
-                                    const uint8* u_buf,
-                                    const uint8* v_buf,
-                                    uint8* rgb_buf,
-                                    int width,
-                                    int16 *kCoefficientsRgbY);
-  asm(
-  ".text\n"
-#if defined(OS_MACOSX)
-"_PICConvertYUVToRGB32Row:\n"
-#else
-"PICConvertYUVToRGB32Row:\n"
-#endif
-  "pusha\n"
-  "mov    0x24(%esp),%edx\n"
-  "mov    0x28(%esp),%edi\n"
-  "mov    0x2c(%esp),%esi\n"
-  "mov    0x30(%esp),%ebp\n"
-  "mov    0x38(%esp),%ecx\n"
-
-  "jmp    .Lconvertend\n"
-
-".Lconvertloop:"
-  "movzbl (%edi),%eax\n"
-  "add    $0x1,%edi\n"
-  "movzbl (%esi),%ebx\n"
-  "add    $0x1,%esi\n"
-  "movq   2048(%ecx,%eax,8),%mm0\n"
-  "movzbl (%edx),%eax\n"
-  "paddsw 4096(%ecx,%ebx,8),%mm0\n"
-  "movzbl 0x1(%edx),%ebx\n"
-  "movq   0(%ecx,%eax,8),%mm1\n"
-  "add    $0x2,%edx\n"
-  "movq   0(%ecx,%ebx,8),%mm2\n"
-  "paddsw %mm0,%mm1\n"
-  "paddsw %mm0,%mm2\n"
-  "psraw  $0x6,%mm1\n"
-  "psraw  $0x6,%mm2\n"
-  "packuswb %mm2,%mm1\n"
-  "movntq %mm1,0x0(%ebp)\n"
-  "add    $0x8,%ebp\n"
-".Lconvertend:"
-  "subl   $0x2,0x34(%esp)\n"
-  "jns    .Lconvertloop\n"
-
-  "andl   $0x1,0x34(%esp)\n"
-  "je     .Lconvertdone\n"
-
-  "movzbl (%edi),%eax\n"
-  "movq   2048(%ecx,%eax,8),%mm0\n"
-  "movzbl (%esi),%eax\n"
-  "paddsw 4096(%ecx,%eax,8),%mm0\n"
-  "movzbl (%edx),%eax\n"
-  "movq   0(%ecx,%eax,8),%mm1\n"
-  "paddsw %mm0,%mm1\n"
-  "psraw  $0x6,%mm1\n"
-  "packuswb %mm1,%mm1\n"
-  "movd   %mm1,0x0(%ebp)\n"
-".Lconvertdone:\n"
-  "popa\n"
-  "ret\n"
-);
-
-void FastConvertYUVToRGB32Row(const uint8* y_buf,
-                              const uint8* u_buf,
-                              const uint8* v_buf,
-                              uint8* rgb_buf,
-                              int width) {
-  PICConvertYUVToRGB32Row(y_buf, u_buf, v_buf, rgb_buf, width,
-                          &kCoefficientsRgbY[0][0]);
-}
-
-extern void PICScaleYUVToRGB32Row(const uint8* y_buf,
-                               const uint8* u_buf,
-                               const uint8* v_buf,
-                               uint8* rgb_buf,
-                               int width,
-                               int source_dx,
-                               int16 *kCoefficientsRgbY);
-
-  asm(
-  ".text\n"
-#if defined(OS_MACOSX)
-"_PICScaleYUVToRGB32Row:\n"
-#else
-"PICScaleYUVToRGB32Row:\n"
-#endif
-  "pusha\n"
-  "mov    0x24(%esp),%edx\n"
-  "mov    0x28(%esp),%edi\n"
-  "mov    0x2c(%esp),%esi\n"
-  "mov    0x30(%esp),%ebp\n"
-  "mov    0x3c(%esp),%ecx\n"
-  "xor    %ebx,%ebx\n"
-  "jmp    Lscaleend\n"
-
-"Lscaleloop:"
-  "mov    %ebx,%eax\n"
-  "sar    $0x11,%eax\n"
-  "movzbl (%edi,%eax,1),%eax\n"
-  "movq   2048(%ecx,%eax,8),%mm0\n"
-  "mov    %ebx,%eax\n"
-  "sar    $0x11,%eax\n"
-  "movzbl (%esi,%eax,1),%eax\n"
-  "paddsw 4096(%ecx,%eax,8),%mm0\n"
-  "mov    %ebx,%eax\n"
-  "add    0x38(%esp),%ebx\n"
-  "sar    $0x10,%eax\n"
-  "movzbl (%edx,%eax,1),%eax\n"
-  "movq   0(%ecx,%eax,8),%mm1\n"
-  "mov    %ebx,%eax\n"
-  "add    0x38(%esp),%ebx\n"
-  "sar    $0x10,%eax\n"
-  "movzbl (%edx,%eax,1),%eax\n"
-  "movq   0(%ecx,%eax,8),%mm2\n"
-  "paddsw %mm0,%mm1\n"
-  "paddsw %mm0,%mm2\n"
-  "psraw  $0x6,%mm1\n"
-  "psraw  $0x6,%mm2\n"
-  "packuswb %mm2,%mm1\n"
-  "movntq %mm1,0x0(%ebp)\n"
-  "add    $0x8,%ebp\n"
-"Lscaleend:"
-  "subl   $0x2,0x34(%esp)\n"
-  "jns    Lscaleloop\n"
-
-  "andl   $0x1,0x34(%esp)\n"
-  "je     Lscaledone\n"
-
-  "mov    %ebx,%eax\n"
-  "sar    $0x11,%eax\n"
-  "movzbl (%edi,%eax,1),%eax\n"
-  "movq   2048(%ecx,%eax,8),%mm0\n"
-  "mov    %ebx,%eax\n"
-  "sar    $0x11,%eax\n"
-  "movzbl (%esi,%eax,1),%eax\n"
-  "paddsw 4096(%ecx,%eax,8),%mm0\n"
-  "mov    %ebx,%eax\n"
-  "sar    $0x10,%eax\n"
-  "movzbl (%edx,%eax,1),%eax\n"
-  "movq   0(%ecx,%eax,8),%mm1\n"
-  "paddsw %mm0,%mm1\n"
-  "psraw  $0x6,%mm1\n"
-  "packuswb %mm1,%mm1\n"
-  "movd   %mm1,0x0(%ebp)\n"
-
-"Lscaledone:"
-  "popa\n"
-  "ret\n"
-);
-
-
-void ScaleYUVToRGB32Row(const uint8* y_buf,
-                        const uint8* u_buf,
-                        const uint8* v_buf,
-                        uint8* rgb_buf,
-                        int width,
-                        int source_dx) {
-  PICScaleYUVToRGB32Row(y_buf, u_buf, v_buf, rgb_buf, width, source_dx,
-                        &kCoefficientsRgbY[0][0]);
-}
-
-void PICLinearScaleYUVToRGB32Row(const uint8* y_buf,
-                                 const uint8* u_buf,
-                                 const uint8* v_buf,
-                                 uint8* rgb_buf,
-                                 int width,
-                                 int source_dx,
-                                 int16 *kCoefficientsRgbY);
-  asm(
-  ".text\n"
-#if defined(OS_MACOSX)
-"_PICLinearScaleYUVToRGB32Row:\n"
-#else
-"PICLinearScaleYUVToRGB32Row:\n"
-#endif
-  "pusha\n"
-  "mov    0x24(%esp),%edx\n"
-  "mov    0x30(%esp),%ebp\n"
-  "mov    0x34(%esp),%ecx\n"
-  "mov    0x3c(%esp),%edi\n"
-  "xor    %ebx,%ebx\n"
-
-  // source_width = width * source_dx + ebx
-  "mov    0x34(%esp), %ecx\n"
-  "imull  0x38(%esp), %ecx\n"
-  "mov    %ecx, 0x34(%esp)\n"
-
-  "mov    0x38(%esp), %ecx\n"
-  "xor    %ebx,%ebx\n"     // x = 0
-  "cmp    $0x20000,%ecx\n" // if source_dx >= 2.0
-  "jl     .lscaleend\n"
-  "mov    $0x8000,%ebx\n"  // x = 0.5 for 1/2 or less
-  "jmp    .lscaleend\n"
-
-".lscaleloop:"
-  "mov    0x28(%esp),%esi\n"
-  "mov    %ebx,%eax\n"
-  "sar    $0x11,%eax\n"
-
-  "movzbl (%esi,%eax,1),%ecx\n"
-  "movzbl 1(%esi,%eax,1),%esi\n"
-  "mov    %ebx,%eax\n"
-  "andl   $0x1fffe, %eax \n"
-  "imul   %eax, %esi \n"
-  "xorl   $0x1fffe, %eax \n"
-  "imul   %eax, %ecx \n"
-  "addl   %esi, %ecx \n"
-  "shrl   $17, %ecx \n"
-  "movq   2048(%edi,%ecx,8),%mm0\n"
-
-  "mov    0x2c(%esp),%esi\n"
-  "mov    %ebx,%eax\n"
-  "sar    $0x11,%eax\n"
-
-  "movzbl (%esi,%eax,1),%ecx\n"
-  "movzbl 1(%esi,%eax,1),%esi\n"
-  "mov    %ebx,%eax\n"
-  "andl   $0x1fffe, %eax \n"
-  "imul   %eax, %esi \n"
-  "xorl   $0x1fffe, %eax \n"
-  "imul   %eax, %ecx \n"
-  "addl   %esi, %ecx \n"
-  "shrl   $17, %ecx \n"
-  "paddsw 4096(%edi,%ecx,8),%mm0\n"
-
-  "mov    %ebx,%eax\n"
-  "sar    $0x10,%eax\n"
-  "movzbl (%edx,%eax,1),%ecx\n"
-  "movzbl 1(%edx,%eax,1),%esi\n"
-  "mov    %ebx,%eax\n"
-  "add    0x38(%esp),%ebx\n"
-  "andl   $0xffff, %eax \n"
-  "imul   %eax, %esi \n"
-  "xorl   $0xffff, %eax \n"
-  "imul   %eax, %ecx \n"
-  "addl   %esi, %ecx \n"
-  "shrl   $16, %ecx \n"
-  "movq   (%edi,%ecx,8),%mm1\n"
-
-  "cmp    0x34(%esp), %ebx\n"
-  "jge    .lscalelastpixel\n"
-
-  "mov    %ebx,%eax\n"
-  "sar    $0x10,%eax\n"
-  "movzbl (%edx,%eax,1),%ecx\n"
-  "movzbl 1(%edx,%eax,1),%esi\n"
-  "mov    %ebx,%eax\n"
-  "add    0x38(%esp),%ebx\n"
-  "andl   $0xffff, %eax \n"
-  "imul   %eax, %esi \n"
-  "xorl   $0xffff, %eax \n"
-  "imul   %eax, %ecx \n"
-  "addl   %esi, %ecx \n"
-  "shrl   $16, %ecx \n"
-  "movq   (%edi,%ecx,8),%mm2\n"
-
-  "paddsw %mm0,%mm1\n"
-  "paddsw %mm0,%mm2\n"
-  "psraw  $0x6,%mm1\n"
-  "psraw  $0x6,%mm2\n"
-  "packuswb %mm2,%mm1\n"
-  "movntq %mm1,0x0(%ebp)\n"
-  "add    $0x8,%ebp\n"
-
-".lscaleend:"
-  "cmp    %ebx, 0x34(%esp)\n"
-  "jg     .lscaleloop\n"
-  "popa\n"
-  "ret\n"
-
-".lscalelastpixel:"
-  "paddsw %mm0, %mm1\n"
-  "psraw $6, %mm1\n"
-  "packuswb %mm1, %mm1\n"
-  "movd %mm1, (%ebp)\n"
-  "popa\n"
-  "ret\n"
-);
-
-void LinearScaleYUVToRGB32Row(const uint8* y_buf,
-                        const uint8* u_buf,
-                        const uint8* v_buf,
-                        uint8* rgb_buf,
-                        int width,
-                        int source_dx) {
-  PICLinearScaleYUVToRGB32Row(y_buf, u_buf, v_buf, rgb_buf, width, source_dx,
-                              &kCoefficientsRgbY[0][0]);
-}
-
-#else  // USE_MMX
-
 // C reference code that mimic the YUV assembly.
 #define packuswb(x) ((x) < 0 ? 0 : ((x) > 255 ? 255 : (x)))
 #define paddsw(x, y) (((x) + (y)) < -32768 ? -32768 : \
     (((x) + (y)) > 32767 ? 32767 : ((x) + (y))))
 
 static inline void YuvPixel(uint8 y,
                             uint8 u,
                             uint8 v,
@@ -833,66 +39,71 @@ static inline void YuvPixel(uint8 y,
   a >>= 6;
 
   *reinterpret_cast<uint32*>(rgb_buf) = (packuswb(b)) |
                                         (packuswb(g) << 8) |
                                         (packuswb(r) << 16) |
                                         (packuswb(a) << 24);
 }
 
-void FastConvertYUVToRGB32Row(const uint8* y_buf,
-                              const uint8* u_buf,
-                              const uint8* v_buf,
-                              uint8* rgb_buf,
-                              int width) {
+void FastConvertYUVToRGB32Row_C(const uint8* y_buf,
+                                const uint8* u_buf,
+                                const uint8* v_buf,
+                                uint8* rgb_buf,
+                                int width,
+                                unsigned int x_shift) {
   for (int x = 0; x < width; x += 2) {
-    uint8 u = u_buf[x >> 1];
-    uint8 v = v_buf[x >> 1];
+    uint8 u = u_buf[x >> x_shift];
+    uint8 v = v_buf[x >> x_shift];
     uint8 y0 = y_buf[x];
     YuvPixel(y0, u, v, rgb_buf);
     if ((x + 1) < width) {
       uint8 y1 = y_buf[x + 1];
+      if (x_shift == 0) {
+        u = u_buf[x + 1];
+        v = v_buf[x + 1];
+      }
       YuvPixel(y1, u, v, rgb_buf + 4);
     }
     rgb_buf += 8;  // Advance 2 pixels.
   }
 }
 
 // 16.16 fixed point is used.  A shift by 16 isolates the integer.
 // A shift by 17 is used to further subsample the chrominence channels.
 // & 0xffff isolates the fixed point fraction.  >> 2 to get the upper 2 bits,
 // for 1/65536 pixel accurate interpolation.
-void ScaleYUVToRGB32Row(const uint8* y_buf,
-                        const uint8* u_buf,
-                        const uint8* v_buf,
-                        uint8* rgb_buf,
-                        int width,
-                        int source_dx) {
+void ScaleYUVToRGB32Row_C(const uint8* y_buf,
+                          const uint8* u_buf,
+                          const uint8* v_buf,
+                          uint8* rgb_buf,
+                          int width,
+                          int source_dx) {
   int x = 0;
   for (int i = 0; i < width; i += 2) {
     int y = y_buf[x >> 16];
     int u = u_buf[(x >> 17)];
     int v = v_buf[(x >> 17)];
     YuvPixel(y, u, v, rgb_buf);
     x += source_dx;
     if ((i + 1) < width) {
       y = y_buf[x >> 16];
       YuvPixel(y, u, v, rgb_buf+4);
       x += source_dx;
     }
     rgb_buf += 8;
   }
 }
 
-void LinearScaleYUVToRGB32Row(const uint8* y_buf,
-                              const uint8* u_buf,
-                              const uint8* v_buf,
-                              uint8* rgb_buf,
-                              int width,
-                              int source_dx) {
+void LinearScaleYUVToRGB32Row_C(const uint8* y_buf,
+                                const uint8* u_buf,
+                                const uint8* v_buf,
+                                uint8* rgb_buf,
+                                int width,
+                                int source_dx) {
   int x = 0;
   if (source_dx >= 0x20000) {
     x = 32768;
   }
   for (int i = 0; i < width; i += 2) {
     int y0 = y_buf[x >> 16];
     int y1 = y_buf[(x >> 16) + 1];
     int u0 = u_buf[(x >> 17)];
@@ -913,11 +124,10 @@ void LinearScaleYUVToRGB32Row(const uint
       y = (y_frac * y1 + (y_frac ^ 65535) * y0) >> 16;
       YuvPixel(y, u, v, rgb_buf+4);
       x += source_dx;
     }
     rgb_buf += 8;
   }
 }
 
-#endif  // USE_MMX
 }  // extern "C"
 
diff --git a/gfx/ycbcr/yuv_row_posix.cpp b/gfx/ycbcr/yuv_row_posix.cpp
--- a/gfx/ycbcr/yuv_row_posix.cpp
+++ b/gfx/ycbcr/yuv_row_posix.cpp
@@ -1,33 +1,32 @@
 // Copyright (c) 2010 The Chromium Authors. All rights reserved.
 // Use of this source code is governed by a BSD-style license that can be
 // found in the LICENSE file.
 
-#include "media/base/yuv_row.h"
-
-#ifdef _DEBUG
-#include "base/logging.h"
-#else
+#include "yuv_row.h"
+#include "mozilla/SSE.h"
+
 #define DCHECK(a)
-#endif
 
 extern "C" {
 
-#if USE_SSE2 && defined(ARCH_CPU_X86_64)
+#if defined(ARCH_CPU_X86_64)
+
+// We don't need CPUID guards here, since x86-64 implies SSE2.
 
 // AMD64 ABI uses register paremters.
 void FastConvertYUVToRGB32Row(const uint8* y_buf,  // rdi
                               const uint8* u_buf,  // rsi
                               const uint8* v_buf,  // rdx
                               uint8* rgb_buf,      // rcx
                               int width) {         // r8
   asm(
-  "jmp    convertend\n"
-"convertloop:"
+  "jmp    1f\n"
+"0:"
   "movzb  (%1),%%r10\n"
   "add    $0x1,%1\n"
   "movzb  (%2),%%r11\n"
   "add    $0x1,%2\n"
   "movq   2048(%5,%%r10,8),%%xmm0\n"
   "movzb  (%0),%%r10\n"
   "movq   4096(%5,%%r11,8),%%xmm1\n"
   "movzb  0x1(%0),%%r11\n"
@@ -37,36 +36,36 @@ void FastConvertYUVToRGB32Row(const uint
   "movq   (%5,%%r11,8),%%xmm3\n"
   "paddsw %%xmm0,%%xmm2\n"
   "paddsw %%xmm0,%%xmm3\n"
   "shufps $0x44,%%xmm3,%%xmm2\n"
   "psraw  $0x6,%%xmm2\n"
   "packuswb %%xmm2,%%xmm2\n"
   "movq   %%xmm2,0x0(%3)\n"
   "add    $0x8,%3\n"
-"convertend:"
+"1:"
   "sub    $0x2,%4\n"
-  "jns    convertloop\n"
-
-"convertnext:"
+  "jns    0b\n"
+
+"2:"
   "add    $0x1,%4\n"
-  "js     convertdone\n"
+  "js     3f\n"
 
   "movzb  (%1),%%r10\n"
   "movq   2048(%5,%%r10,8),%%xmm0\n"
   "movzb  (%2),%%r10\n"
   "movq   4096(%5,%%r10,8),%%xmm1\n"
   "paddsw %%xmm1,%%xmm0\n"
   "movzb  (%0),%%r10\n"
   "movq   (%5,%%r10,8),%%xmm1\n"
   "paddsw %%xmm0,%%xmm1\n"
   "psraw  $0x6,%%xmm1\n"
   "packuswb %%xmm1,%%xmm1\n"
   "movd   %%xmm1,0x0(%3)\n"
-"convertdone:"
+"3:"
   :
   : "r"(y_buf),  // %0
     "r"(u_buf),  // %1
     "r"(v_buf),  // %2
     "r"(rgb_buf),  // %3
     "r"(width),  // %4
     "r" (kCoefficientsRgbY)  // %5
   : "memory", "r10", "r11", "xmm0", "xmm1", "xmm2", "xmm3"
@@ -77,19 +76,19 @@ void ScaleYUVToRGB32Row(const uint8* y_b
                         const uint8* u_buf,  // rsi
                         const uint8* v_buf,  // rdx
                         uint8* rgb_buf,      // rcx
                         int width,           // r8
                         int source_dx) {     // r9
   asm(
   "xor    %%r11,%%r11\n"
   "sub    $0x2,%4\n"
-  "js     scalenext\n"
-
-"scaleloop:"
+  "js     1f\n"
+
+"0:"
   "mov    %%r11,%%r10\n"
   "sar    $0x11,%%r10\n"
   "movzb  (%1,%%r10,1),%%rax\n"
   "movq   2048(%5,%%rax,8),%%xmm0\n"
   "movzb  (%2,%%r10,1),%%rax\n"
   "movq   4096(%5,%%rax,8),%%xmm1\n"
   "lea    (%%r11,%6),%%r10\n"
   "sar    $0x10,%%r11\n"
@@ -103,38 +102,38 @@ void ScaleYUVToRGB32Row(const uint8* y_b
   "paddsw %%xmm0,%%xmm1\n"
   "paddsw %%xmm0,%%xmm2\n"
   "shufps $0x44,%%xmm2,%%xmm1\n"
   "psraw  $0x6,%%xmm1\n"
   "packuswb %%xmm1,%%xmm1\n"
   "movq   %%xmm1,0x0(%3)\n"
   "add    $0x8,%3\n"
   "sub    $0x2,%4\n"
-  "jns    scaleloop\n"
-
-"scalenext:"
+  "jns    0b\n"
+
+"1:"
   "add    $0x1,%4\n"
-  "js     scaledone\n"
+  "js     2f\n"
 
   "mov    %%r11,%%r10\n"
   "sar    $0x11,%%r10\n"
   "movzb  (%1,%%r10,1),%%rax\n"
   "movq   2048(%5,%%rax,8),%%xmm0\n"
   "movzb  (%2,%%r10,1),%%rax\n"
   "movq   4096(%5,%%rax,8),%%xmm1\n"
   "paddsw %%xmm1,%%xmm0\n"
   "sar    $0x10,%%r11\n"
   "movzb  (%0,%%r11,1),%%rax\n"
   "movq   (%5,%%rax,8),%%xmm1\n"
   "paddsw %%xmm0,%%xmm1\n"
   "psraw  $0x6,%%xmm1\n"
   "packuswb %%xmm1,%%xmm1\n"
   "movd   %%xmm1,0x0(%3)\n"
 
-"scaledone:"
+"2:"
   :
   : "r"(y_buf),  // %0
     "r"(u_buf),  // %1
     "r"(v_buf),  // %2
     "r"(rgb_buf),  // %3
     "r"(width),  // %4
     "r" (kCoefficientsRgbY),  // %5
     "r"(static_cast<long>(source_dx))  // %6
@@ -146,23 +145,23 @@ void LinearScaleYUVToRGB32Row(const uint
                               const uint8* u_buf,
                               const uint8* v_buf,
                               uint8* rgb_buf,
                               int width,
                               int source_dx) {
   asm(
   "xor    %%r11,%%r11\n"   // x = 0
   "sub    $0x2,%4\n"
-  "js     .lscalenext\n"
+  "js     2f\n"
   "cmp    $0x20000,%6\n"   // if source_dx >= 2.0
-  "jl     .lscalehalf\n"
+  "jl     0f\n"
   "mov    $0x8000,%%r11\n" // x = 0.5 for 1/2 or less
-".lscalehalf:"
-
-".lscaleloop:"
+"0:"
+
+"1:"
   "mov    %%r11,%%r10\n"
   "sar    $0x11,%%r10\n"
 
   "movzb  (%1, %%r10, 1), %%r13 \n"
   "movzb  1(%1, %%r10, 1), %%r14 \n"
   "mov    %%r11, %%rax \n"
   "and    $0x1fffe, %%rax \n"
   "imul   %%rax, %%r14 \n"
@@ -215,21 +214,21 @@ void LinearScaleYUVToRGB32Row(const uint
   "paddsw %%xmm0,%%xmm1\n"
   "paddsw %%xmm0,%%xmm2\n"
   "shufps $0x44,%%xmm2,%%xmm1\n"
   "psraw  $0x6,%%xmm1\n"
   "packuswb %%xmm1,%%xmm1\n"
   "movq   %%xmm1,0x0(%3)\n"
   "add    $0x8,%3\n"
   "sub    $0x2,%4\n"
-  "jns    .lscaleloop\n"
-
-".lscalenext:"
+  "jns    1b\n"
+
+"2:"
   "add    $0x1,%4\n"
-  "js     .lscaledone\n"
+  "js     3f\n"
 
   "mov    %%r11,%%r10\n"
   "sar    $0x11,%%r10\n"
 
   "movzb  (%1,%%r10,1), %%r13 \n"
   "movq   2048(%5,%%r13,8),%%xmm0\n"
 
   "movzb  (%2,%%r10,1), %%r13 \n"
@@ -241,52 +240,52 @@ void LinearScaleYUVToRGB32Row(const uint
   "movzb  (%0,%%r11,1), %%r13 \n"
   "movq   (%5,%%r13,8),%%xmm1\n"
 
   "paddsw %%xmm0,%%xmm1\n"
   "psraw  $0x6,%%xmm1\n"
   "packuswb %%xmm1,%%xmm1\n"
   "movd   %%xmm1,0x0(%3)\n"
 
-".lscaledone:"
+"3:"
   :
   : "r"(y_buf),  // %0
     "r"(u_buf),  // %1
     "r"(v_buf),  // %2
     "r"(rgb_buf),  // %3
     "r"(width),  // %4
     "r" (kCoefficientsRgbY),  // %5
     "r"(static_cast<long>(source_dx))  // %6
   : "memory", "r10", "r11", "r13", "r14", "rax", "xmm0", "xmm1", "xmm2"
 );
 }
 
-#elif USE_MMX && !defined(ARCH_CPU_X86_64) && !defined(__PIC__)
+#elif defined(MOZILLA_MAY_SUPPORT_SSE) && defined(ARCH_CPU_X86_32) && !defined(__PIC__)
 
 // PIC version is slower because less registers are available, so
 // non-PIC is used on platforms where it is possible.
-
-void FastConvertYUVToRGB32Row(const uint8* y_buf,
-                              const uint8* u_buf,
-                              const uint8* v_buf,
-                              uint8* rgb_buf,
-                              int width);
+void FastConvertYUVToRGB32Row_SSE(const uint8* y_buf,
+                                  const uint8* u_buf,
+                                  const uint8* v_buf,
+                                  uint8* rgb_buf,
+                                  int width);
   asm(
   ".text\n"
-  ".global FastConvertYUVToRGB32Row\n"
-"FastConvertYUVToRGB32Row:\n"
+  ".global FastConvertYUVToRGB32Row_SSE\n"
+  ".type FastConvertYUVToRGB32Row_SSE, @function\n"
+"FastConvertYUVToRGB32Row_SSE:\n"
   "pusha\n"
   "mov    0x24(%esp),%edx\n"
   "mov    0x28(%esp),%edi\n"
   "mov    0x2c(%esp),%esi\n"
   "mov    0x30(%esp),%ebp\n"
   "mov    0x34(%esp),%ecx\n"
-  "jmp    convertend\n"
-
-"convertloop:"
+  "jmp    1f\n"
+
+"0:"
   "movzbl (%edi),%eax\n"
   "add    $0x1,%edi\n"
   "movzbl (%esi),%ebx\n"
   "add    $0x1,%esi\n"
   "movq   kCoefficientsRgbY+2048(,%eax,8),%mm0\n"
   "movzbl (%edx),%eax\n"
   "paddsw kCoefficientsRgbY+4096(,%ebx,8),%mm0\n"
   "movzbl 0x1(%edx),%ebx\n"
@@ -295,59 +294,77 @@ void FastConvertYUVToRGB32Row(const uint
   "movq   kCoefficientsRgbY(,%ebx,8),%mm2\n"
   "paddsw %mm0,%mm1\n"
   "paddsw %mm0,%mm2\n"
   "psraw  $0x6,%mm1\n"
   "psraw  $0x6,%mm2\n"
   "packuswb %mm2,%mm1\n"
   "movntq %mm1,0x0(%ebp)\n"
   "add    $0x8,%ebp\n"
-"convertend:"
+"1:"
   "sub    $0x2,%ecx\n"
-  "jns    convertloop\n"
+  "jns    0b\n"
 
   "and    $0x1,%ecx\n"
-  "je     convertdone\n"
+  "je     2f\n"
 
   "movzbl (%edi),%eax\n"
   "movq   kCoefficientsRgbY+2048(,%eax,8),%mm0\n"
   "movzbl (%esi),%eax\n"
   "paddsw kCoefficientsRgbY+4096(,%eax,8),%mm0\n"
   "movzbl (%edx),%eax\n"
   "movq   kCoefficientsRgbY(,%eax,8),%mm1\n"
   "paddsw %mm0,%mm1\n"
   "psraw  $0x6,%mm1\n"
   "packuswb %mm1,%mm1\n"
   "movd   %mm1,0x0(%ebp)\n"
-"convertdone:"
+"2:"
   "popa\n"
   "ret\n"
+#if !defined(XP_MACOSX)
+  ".previous\n"
+#endif
 );
 
-
-void ScaleYUVToRGB32Row(const uint8* y_buf,
-                        const uint8* u_buf,
-                        const uint8* v_buf,
-                        uint8* rgb_buf,
-                        int width,
-                        int source_dx);
+void FastConvertYUVToRGB32Row(const uint8* y_buf,
+                              const uint8* u_buf,
+                              const uint8* v_buf,
+                              uint8* rgb_buf,
+                              int width)
+{
+  if (mozilla::supports_sse()) {
+    FastConvertYUVToRGB32Row_SSE(y_buf, u_buf, v_buf, rgb_buf, width);
+    return;
+  }
+
+  FastConvertYUVToRGB32Row_C(y_buf, u_buf, v_buf, rgb_buf, width, 1);
+}
+
+
+void ScaleYUVToRGB32Row_SSE(const uint8* y_buf,
+                            const uint8* u_buf,
+                            const uint8* v_buf,
+                            uint8* rgb_buf,
+                            int width,
+                            int source_dx);
   asm(
   ".text\n"
-  ".global ScaleYUVToRGB32Row\n"
-"ScaleYUVToRGB32Row:\n"
+  ".global ScaleYUVToRGB32Row_SSE\n"
+  ".type ScaleYUVToRGB32Row_SSE, @function\n"
+"ScaleYUVToRGB32Row_SSE:\n"
   "pusha\n"
   "mov    0x24(%esp),%edx\n"
   "mov    0x28(%esp),%edi\n"
   "mov    0x2c(%esp),%esi\n"
   "mov    0x30(%esp),%ebp\n"
   "mov    0x34(%esp),%ecx\n"
   "xor    %ebx,%ebx\n"
-  "jmp    scaleend\n"
-
-"scaleloop:"
+  "jmp    1f\n"
+
+"0:"
   "mov    %ebx,%eax\n"
   "sar    $0x11,%eax\n"
   "movzbl (%edi,%eax,1),%eax\n"
   "movq   kCoefficientsRgbY+2048(,%eax,8),%mm0\n"
   "mov    %ebx,%eax\n"
   "sar    $0x11,%eax\n"
   "movzbl (%esi,%eax,1),%eax\n"
   "paddsw kCoefficientsRgbY+4096(,%eax,8),%mm0\n"
@@ -363,22 +380,22 @@ void ScaleYUVToRGB32Row(const uint8* y_b
   "movq   kCoefficientsRgbY(,%eax,8),%mm2\n"
   "paddsw %mm0,%mm1\n"
   "paddsw %mm0,%mm2\n"
   "psraw  $0x6,%mm1\n"
   "psraw  $0x6,%mm2\n"
   "packuswb %mm2,%mm1\n"
   "movntq %mm1,0x0(%ebp)\n"
   "add    $0x8,%ebp\n"
-"scaleend:"
+"1:"
   "sub    $0x2,%ecx\n"
-  "jns    scaleloop\n"
+  "jns    0b\n"
 
   "and    $0x1,%ecx\n"
-  "je     scaledone\n"
+  "je     2f\n"
 
   "mov    %ebx,%eax\n"
   "sar    $0x11,%eax\n"
   "movzbl (%edi,%eax,1),%eax\n"
   "movq   kCoefficientsRgbY+2048(,%eax,8),%mm0\n"
   "mov    %ebx,%eax\n"
   "sar    $0x11,%eax\n"
   "movzbl (%esi,%eax,1),%eax\n"
@@ -387,51 +404,71 @@ void ScaleYUVToRGB32Row(const uint8* y_b
   "sar    $0x10,%eax\n"
   "movzbl (%edx,%eax,1),%eax\n"
   "movq   kCoefficientsRgbY(,%eax,8),%mm1\n"
   "paddsw %mm0,%mm1\n"
   "psraw  $0x6,%mm1\n"
   "packuswb %mm1,%mm1\n"
   "movd   %mm1,0x0(%ebp)\n"
 
-"scaledone:"
+"2:"
   "popa\n"
   "ret\n"
+#if !defined(XP_MACOSX)
+  ".previous\n"
+#endif
 );
 
-void LinearScaleYUVToRGB32Row(const uint8* y_buf,
-                              const uint8* u_buf,
-                              const uint8* v_buf,
-                              uint8* rgb_buf,
-                              int width,
-                              int source_dx);
+void ScaleYUVToRGB32Row(const uint8* y_buf,
+                        const uint8* u_buf,
+                        const uint8* v_buf,
+                        uint8* rgb_buf,
+                        int width,
+                        int source_dx)
+{
+  if (mozilla::supports_sse()) {
+    ScaleYUVToRGB32Row_SSE(y_buf, u_buf, v_buf, rgb_buf,
+                           width, source_dx);
+  }
+
+  ScaleYUVToRGB32Row_C(y_buf, u_buf, v_buf, rgb_buf,
+                       width, source_dx);
+}
+
+void LinearScaleYUVToRGB32Row_SSE(const uint8* y_buf,
+                                  const uint8* u_buf,
+                                  const uint8* v_buf,
+                                  uint8* rgb_buf,
+                                  int width,
+                                  int source_dx);
   asm(
   ".text\n"
-  ".global LinearScaleYUVToRGB32Row\n"
-"LinearScaleYUVToRGB32Row:\n"
+  ".global LinearScaleYUVToRGB32Row_SSE\n"
+  ".type LinearScaleYUVToRGB32Row_SSE, @function\n"
+"LinearScaleYUVToRGB32Row_SSE:\n"
   "pusha\n"
   "mov    0x24(%esp),%edx\n"
   "mov    0x28(%esp),%edi\n"
   "mov    0x30(%esp),%ebp\n"
 
   // source_width = width * source_dx + ebx
   "mov    0x34(%esp), %ecx\n"
   "imull  0x38(%esp), %ecx\n"
   "mov    %ecx, 0x34(%esp)\n"
 
   "mov    0x38(%esp), %ecx\n"
   "xor    %ebx,%ebx\n"     // x = 0
   "cmp    $0x20000,%ecx\n" // if source_dx >= 2.0
-  "jl     .lscaleend\n"
+  "jl     1f\n"
   "mov    $0x8000,%ebx\n"  // x = 0.5 for 1/2 or less
-  "jmp    .lscaleend\n"
-
-".lscaleloop:"
-  "mov    %ebx,%eax\n"
-  "sar    $0x11,%eax\n"
+  "jmp    1f\n"
+
+"0:"
+  "mov    %ebx,%eax\n"
+  "sar    $0x11,%eax\n"
 
   "movzbl (%edi,%eax,1),%ecx\n"
   "movzbl 1(%edi,%eax,1),%esi\n"
   "mov    %ebx,%eax\n"
   "andl   $0x1fffe, %eax \n"
   "imul   %eax, %esi \n"
   "xorl   $0x1fffe, %eax \n"
   "imul   %eax, %ecx \n"
@@ -464,17 +501,17 @@ void LinearScaleYUVToRGB32Row(const uint
   "imul   %eax, %esi \n"
   "xorl   $0xffff, %eax \n"
   "imul   %eax, %ecx \n"
   "addl   %esi, %ecx \n"
   "shrl   $16, %ecx \n"
   "movq   kCoefficientsRgbY(,%ecx,8),%mm1\n"
 
   "cmp    0x34(%esp), %ebx\n"
-  "jge    .lscalelastpixel\n"
+  "jge    2f\n"
 
   "mov    %ebx,%eax\n"
   "sar    $0x10,%eax\n"
   "movzbl (%edx,%eax,1),%ecx\n"
   "movzbl 1(%edx,%eax,1),%esi\n"
   "mov    %ebx,%eax\n"
   "add    0x38(%esp),%ebx\n"
   "andl   $0xffff, %eax \n"
@@ -488,56 +525,76 @@ void LinearScaleYUVToRGB32Row(const uint
   "paddsw %mm0,%mm1\n"
   "paddsw %mm0,%mm2\n"
   "psraw  $0x6,%mm1\n"
   "psraw  $0x6,%mm2\n"
   "packuswb %mm2,%mm1\n"
   "movntq %mm1,0x0(%ebp)\n"
   "add    $0x8,%ebp\n"
 
-".lscaleend:"
+"1:"
   "cmp    0x34(%esp), %ebx\n"
-  "jl     .lscaleloop\n"
+  "jl     0b\n"
   "popa\n"
   "ret\n"
 
-".lscalelastpixel:"
+"2:"
   "paddsw %mm0, %mm1\n"
   "psraw $6, %mm1\n"
   "packuswb %mm1, %mm1\n"
   "movd %mm1, (%ebp)\n"
   "popa\n"
   "ret\n"
+#if !defined(XP_MACOSX)
+  ".previous\n"
+#endif
 );
 
-#elif USE_MMX && !defined(ARCH_CPU_X86_64) && defined(__PIC__)
-
-extern void PICConvertYUVToRGB32Row(const uint8* y_buf,
-                                    const uint8* u_buf,
-                                    const uint8* v_buf,
-                                    uint8* rgb_buf,
-                                    int width,
-                                    int16 *kCoefficientsRgbY);
+void LinearScaleYUVToRGB32Row(const uint8* y_buf,
+                              const uint8* u_buf,
+                              const uint8* v_buf,
+                              uint8* rgb_buf,
+                              int width,
+                              int source_dx)
+{
+  if (mozilla::supports_sse()) {
+    LinearScaleYUVToRGB32Row_SSE(y_buf, u_buf, v_buf, rgb_buf,
+                                 width, source_dx);
+  }
+
+  LinearScaleYUVToRGB32Row_C(y_buf, u_buf, v_buf, rgb_buf,
+                             width, source_dx);
+}
+
+#elif defined(MOZILLA_MAY_SUPPORT_SSE) && defined(ARCH_CPU_X86_32) && defined(__PIC__)
+
+void PICConvertYUVToRGB32Row_SSE(const uint8* y_buf,
+                                 const uint8* u_buf,
+                                 const uint8* v_buf,
+                                 uint8* rgb_buf,
+                                 int width,
+                                 int16 *kCoefficientsRgbY);
+
   asm(
   ".text\n"
-#if defined(OS_MACOSX)
-"_PICConvertYUVToRGB32Row:\n"
+#if defined(XP_MACOSX)
+"_PICConvertYUVToRGB32Row_SSE:\n"
 #else
-"PICConvertYUVToRGB32Row:\n"
+"PICConvertYUVToRGB32Row_SSE:\n"
 #endif
   "pusha\n"
   "mov    0x24(%esp),%edx\n"
   "mov    0x28(%esp),%edi\n"
   "mov    0x2c(%esp),%esi\n"
   "mov    0x30(%esp),%ebp\n"
   "mov    0x38(%esp),%ecx\n"
 
-  "jmp    .Lconvertend\n"
-
-".Lconvertloop:"
+  "jmp    1f\n"
+
+"0:"
   "movzbl (%edi),%eax\n"
   "add    $0x1,%edi\n"
   "movzbl (%esi),%ebx\n"
   "add    $0x1,%esi\n"
   "movq   2048(%ecx,%eax,8),%mm0\n"
   "movzbl (%edx),%eax\n"
   "paddsw 4096(%ecx,%ebx,8),%mm0\n"
   "movzbl 0x1(%edx),%ebx\n"
@@ -546,72 +603,81 @@ extern void PICConvertYUVToRGB32Row(cons
   "movq   0(%ecx,%ebx,8),%mm2\n"
   "paddsw %mm0,%mm1\n"
   "paddsw %mm0,%mm2\n"
   "psraw  $0x6,%mm1\n"
   "psraw  $0x6,%mm2\n"
   "packuswb %mm2,%mm1\n"
   "movntq %mm1,0x0(%ebp)\n"
   "add    $0x8,%ebp\n"
-".Lconvertend:"
+"1:"
   "subl   $0x2,0x34(%esp)\n"
-  "jns    .Lconvertloop\n"
+  "jns    0b\n"
 
   "andl   $0x1,0x34(%esp)\n"
-  "je     .Lconvertdone\n"
+  "je     2f\n"
 
   "movzbl (%edi),%eax\n"
   "movq   2048(%ecx,%eax,8),%mm0\n"
   "movzbl (%esi),%eax\n"
   "paddsw 4096(%ecx,%eax,8),%mm0\n"
   "movzbl (%edx),%eax\n"
   "movq   0(%ecx,%eax,8),%mm1\n"
   "paddsw %mm0,%mm1\n"
   "psraw  $0x6,%mm1\n"
   "packuswb %mm1,%mm1\n"
   "movd   %mm1,0x0(%ebp)\n"
-".Lconvertdone:\n"
+"2:"
   "popa\n"
   "ret\n"
+#if !defined(XP_MACOSX)
+  ".previous\n"
+#endif
 );
 
 void FastConvertYUVToRGB32Row(const uint8* y_buf,
                               const uint8* u_buf,
                               const uint8* v_buf,
                               uint8* rgb_buf,
-                              int width) {
-  PICConvertYUVToRGB32Row(y_buf, u_buf, v_buf, rgb_buf, width,
-                          &kCoefficientsRgbY[0][0]);
-}
-
-extern void PICScaleYUVToRGB32Row(const uint8* y_buf,
+                              int width)
+{
+  if (mozilla::supports_sse()) {
+    PICConvertYUVToRGB32Row_SSE(y_buf, u_buf, v_buf, rgb_buf, width,
+                                &kCoefficientsRgbY[0][0]);
+    return;
+  }
+
+  FastConvertYUVToRGB32Row_C(y_buf, u_buf, v_buf, rgb_buf, width, 1);
+}
+
+void PICScaleYUVToRGB32Row_SSE(const uint8* y_buf,
                                const uint8* u_buf,
                                const uint8* v_buf,
                                uint8* rgb_buf,
                                int width,
                                int source_dx,
                                int16 *kCoefficientsRgbY);
 
   asm(
   ".text\n"
-#if defined(OS_MACOSX)
-"_PICScaleYUVToRGB32Row:\n"
+#if defined(XP_MACOSX)
+"_PICScaleYUVToRGB32Row_SSE:\n"
 #else
-"PICScaleYUVToRGB32Row:\n"
+"PICScaleYUVToRGB32Row_SSE:\n"
 #endif
   "pusha\n"
   "mov    0x24(%esp),%edx\n"
   "mov    0x28(%esp),%edi\n"
   "mov    0x2c(%esp),%esi\n"
   "mov    0x30(%esp),%ebp\n"
   "mov    0x3c(%esp),%ecx\n"
   "xor    %ebx,%ebx\n"
-  "jmp    Lscaleend\n"
-
-"Lscaleloop:"
+  "jmp    1f\n"
+
+"0:"
   "mov    %ebx,%eax\n"
   "sar    $0x11,%eax\n"
   "movzbl (%edi,%eax,1),%eax\n"
   "movq   2048(%ecx,%eax,8),%mm0\n"
   "mov    %ebx,%eax\n"
   "sar    $0x11,%eax\n"
   "movzbl (%esi,%eax,1),%eax\n"
   "paddsw 4096(%ecx,%eax,8),%mm0\n"
@@ -627,22 +693,22 @@ extern void PICScaleYUVToRGB32Row(const 
   "movq   0(%ecx,%eax,8),%mm2\n"
   "paddsw %mm0,%mm1\n"
   "paddsw %mm0,%mm2\n"
   "psraw  $0x6,%mm1\n"
   "psraw  $0x6,%mm2\n"
   "packuswb %mm2,%mm1\n"
   "movntq %mm1,0x0(%ebp)\n"
   "add    $0x8,%ebp\n"
-"Lscaleend:"
+"1:"
   "subl   $0x2,0x34(%esp)\n"
-  "jns    Lscaleloop\n"
+  "jns    0b\n"
 
   "andl   $0x1,0x34(%esp)\n"
-  "je     Lscaledone\n"
+  "je     2f\n"
 
   "mov    %ebx,%eax\n"
   "sar    $0x11,%eax\n"
   "movzbl (%edi,%eax,1),%eax\n"
   "movq   2048(%ecx,%eax,8),%mm0\n"
   "mov    %ebx,%eax\n"
   "sar    $0x11,%eax\n"
   "movzbl (%esi,%eax,1),%eax\n"
@@ -651,66 +717,75 @@ extern void PICScaleYUVToRGB32Row(const 
   "sar    $0x10,%eax\n"
   "movzbl (%edx,%eax,1),%eax\n"
   "movq   0(%ecx,%eax,8),%mm1\n"
   "paddsw %mm0,%mm1\n"
   "psraw  $0x6,%mm1\n"
   "packuswb %mm1,%mm1\n"
   "movd   %mm1,0x0(%ebp)\n"
 
-"Lscaledone:"
+"2:"
   "popa\n"
   "ret\n"
+#if !defined(XP_MACOSX)
+  ".previous\n"
+#endif
 );
 
-
 void ScaleYUVToRGB32Row(const uint8* y_buf,
                         const uint8* u_buf,
                         const uint8* v_buf,
                         uint8* rgb_buf,
                         int width,
-                        int source_dx) {
-  PICScaleYUVToRGB32Row(y_buf, u_buf, v_buf, rgb_buf, width, source_dx,
-                        &kCoefficientsRgbY[0][0]);
-}
-
-void PICLinearScaleYUVToRGB32Row(const uint8* y_buf,
-                                 const uint8* u_buf,
-                                 const uint8* v_buf,
-                                 uint8* rgb_buf,
-                                 int width,
-                                 int source_dx,
-                                 int16 *kCoefficientsRgbY);
+                        int source_dx)
+{
+  if (mozilla::supports_sse()) {
+    PICScaleYUVToRGB32Row_SSE(y_buf, u_buf, v_buf, rgb_buf, width, source_dx,
+                              &kCoefficientsRgbY[0][0]);
+    return;
+  }
+
+  ScaleYUVToRGB32Row_C(y_buf, u_buf, v_buf, rgb_buf, width, source_dx);
+}
+
+void PICLinearScaleYUVToRGB32Row_SSE(const uint8* y_buf,
+                                     const uint8* u_buf,
+                                     const uint8* v_buf,
+                                     uint8* rgb_buf,
+                                     int width,
+                                     int source_dx,
+                                     int16 *kCoefficientsRgbY);
+
   asm(
   ".text\n"
-#if defined(OS_MACOSX)
-"_PICLinearScaleYUVToRGB32Row:\n"
+#if defined(XP_MACOSX)
+"_PICLinearScaleYUVToRGB32Row_SSE:\n"
 #else
-"PICLinearScaleYUVToRGB32Row:\n"
+"PICLinearScaleYUVToRGB32Row_SSE:\n"
 #endif
   "pusha\n"
   "mov    0x24(%esp),%edx\n"
   "mov    0x30(%esp),%ebp\n"
   "mov    0x34(%esp),%ecx\n"
   "mov    0x3c(%esp),%edi\n"
   "xor    %ebx,%ebx\n"
 
   // source_width = width * source_dx + ebx
   "mov    0x34(%esp), %ecx\n"
   "imull  0x38(%esp), %ecx\n"
   "mov    %ecx, 0x34(%esp)\n"
 
   "mov    0x38(%esp), %ecx\n"
   "xor    %ebx,%ebx\n"     // x = 0
   "cmp    $0x20000,%ecx\n" // if source_dx >= 2.0
-  "jl     .lscaleend\n"
+  "jl     1f\n"
   "mov    $0x8000,%ebx\n"  // x = 0.5 for 1/2 or less
-  "jmp    .lscaleend\n"
-
-".lscaleloop:"
+  "jmp    1f\n"
+
+"0:"
   "mov    0x28(%esp),%esi\n"
   "mov    %ebx,%eax\n"
   "sar    $0x11,%eax\n"
 
   "movzbl (%esi,%eax,1),%ecx\n"
   "movzbl 1(%esi,%eax,1),%esi\n"
   "mov    %ebx,%eax\n"
   "andl   $0x1fffe, %eax \n"
@@ -746,17 +821,17 @@ void PICLinearScaleYUVToRGB32Row(const u
   "imul   %eax, %esi \n"
   "xorl   $0xffff, %eax \n"
   "imul   %eax, %ecx \n"
   "addl   %esi, %ecx \n"
   "shrl   $16, %ecx \n"
   "movq   (%edi,%ecx,8),%mm1\n"
 
   "cmp    0x34(%esp), %ebx\n"
-  "jge    .lscalelastpixel\n"
+  "jge    2f\n"
 
   "mov    %ebx,%eax\n"
   "sar    $0x10,%eax\n"
   "movzbl (%edx,%eax,1),%ecx\n"
   "movzbl 1(%edx,%eax,1),%esi\n"
   "mov    %ebx,%eax\n"
   "add    0x38(%esp),%ebx\n"
   "andl   $0xffff, %eax \n"
@@ -770,154 +845,71 @@ void PICLinearScaleYUVToRGB32Row(const u
   "paddsw %mm0,%mm1\n"
   "paddsw %mm0,%mm2\n"
   "psraw  $0x6,%mm1\n"
   "psraw  $0x6,%mm2\n"
   "packuswb %mm2,%mm1\n"
   "movntq %mm1,0x0(%ebp)\n"
   "add    $0x8,%ebp\n"
 
-".lscaleend:"
+"1:"
   "cmp    %ebx, 0x34(%esp)\n"
-  "jg     .lscaleloop\n"
+  "jg     0b\n"
   "popa\n"
   "ret\n"
 
-".lscalelastpixel:"
+"2:"
   "paddsw %mm0, %mm1\n"
   "psraw $6, %mm1\n"
   "packuswb %mm1, %mm1\n"
   "movd %mm1, (%ebp)\n"
   "popa\n"
   "ret\n"
+#if !defined(XP_MACOSX)
+  ".previous\n"
+#endif
 );
 
+
 void LinearScaleYUVToRGB32Row(const uint8* y_buf,
-                        const uint8* u_buf,
-                        const uint8* v_buf,
-                        uint8* rgb_buf,
-                        int width,
-                        int source_dx) {
-  PICLinearScaleYUVToRGB32Row(y_buf, u_buf, v_buf, rgb_buf, width, source_dx,
-                              &kCoefficientsRgbY[0][0]);
-}
-
-#else  // USE_MMX
-
-// C reference code that mimic the YUV assembly.
-#define packuswb(x) ((x) < 0 ? 0 : ((x) > 255 ? 255 : (x)))
-#define paddsw(x, y) (((x) + (y)) < -32768 ? -32768 : \
-    (((x) + (y)) > 32767 ? 32767 : ((x) + (y))))
-
-static inline void YuvPixel(uint8 y,
-                            uint8 u,
-                            uint8 v,
-                            uint8* rgb_buf) {
-
-  int b = kCoefficientsRgbY[256+u][0];
-  int g = kCoefficientsRgbY[256+u][1];
-  int r = kCoefficientsRgbY[256+u][2];
-  int a = kCoefficientsRgbY[256+u][3];
-
-  b = paddsw(b, kCoefficientsRgbY[512+v][0]);
-  g = paddsw(g, kCoefficientsRgbY[512+v][1]);
-  r = paddsw(r, kCoefficientsRgbY[512+v][2]);
-  a = paddsw(a, kCoefficientsRgbY[512+v][3]);
-
-  b = paddsw(b, kCoefficientsRgbY[y][0]);
-  g = paddsw(g, kCoefficientsRgbY[y][1]);
-  r = paddsw(r, kCoefficientsRgbY[y][2]);
-  a = paddsw(a, kCoefficientsRgbY[y][3]);
-
-  b >>= 6;
-  g >>= 6;
-  r >>= 6;
-  a >>= 6;
-
-  *reinterpret_cast<uint32*>(rgb_buf) = (packuswb(b)) |
-                                        (packuswb(g) << 8) |
-                                        (packuswb(r) << 16) |
-                                        (packuswb(a) << 24);
-}
-
+                              const uint8* u_buf,
+                              const uint8* v_buf,
+                              uint8* rgb_buf,
+                              int width,
+                              int source_dx)
+{
+  if (mozilla::supports_sse()) {
+    PICLinearScaleYUVToRGB32Row_SSE(y_buf, u_buf, v_buf, rgb_buf, width,
+                                    source_dx, &kCoefficientsRgbY[0][0]);
+    return;
+  }
+
+  LinearScaleYUVToRGB32Row_C(y_buf, u_buf, v_buf, rgb_buf, width, source_dx);
+}
+#else
 void FastConvertYUVToRGB32Row(const uint8* y_buf,
                               const uint8* u_buf,
                               const uint8* v_buf,
                               uint8* rgb_buf,
                               int width) {
-  for (int x = 0; x < width; x += 2) {
-    uint8 u = u_buf[x >> 1];
-    uint8 v = v_buf[x >> 1];
-    uint8 y0 = y_buf[x];
-    YuvPixel(y0, u, v, rgb_buf);
-    if ((x + 1) < width) {
-      uint8 y1 = y_buf[x + 1];
-      YuvPixel(y1, u, v, rgb_buf + 4);
-    }
-    rgb_buf += 8;  // Advance 2 pixels.
-  }
-}
-
-// 16.16 fixed point is used.  A shift by 16 isolates the integer.
-// A shift by 17 is used to further subsample the chrominence channels.
-// & 0xffff isolates the fixed point fraction.  >> 2 to get the upper 2 bits,
-// for 1/65536 pixel accurate interpolation.
+  FastConvertYUVToRGB32Row_C(y_buf, u_buf, v_buf, rgb_buf, width, 1);
+}
+
 void ScaleYUVToRGB32Row(const uint8* y_buf,
                         const uint8* u_buf,
                         const uint8* v_buf,
                         uint8* rgb_buf,
                         int width,
                         int source_dx) {
-  int x = 0;
-  for (int i = 0; i < width; i += 2) {
-    int y = y_buf[x >> 16];
-    int u = u_buf[(x >> 17)];
-    int v = v_buf[(x >> 17)];
-    YuvPixel(y, u, v, rgb_buf);
-    x += source_dx;
-    if ((i + 1) < width) {
-      y = y_buf[x >> 16];
-      YuvPixel(y, u, v, rgb_buf+4);
-      x += source_dx;
-    }
-    rgb_buf += 8;
-  }
-}
+  ScaleYUVToRGB32Row_C(y_buf, u_buf, v_buf, rgb_buf, width, source_dx);
+}
 
 void LinearScaleYUVToRGB32Row(const uint8* y_buf,
                               const uint8* u_buf,
                               const uint8* v_buf,
                               uint8* rgb_buf,
                               int width,
                               int source_dx) {
-  int x = 0;
-  if (source_dx >= 0x20000) {
-    x = 32768;
-  }
-  for (int i = 0; i < width; i += 2) {
-    int y0 = y_buf[x >> 16];
-    int y1 = y_buf[(x >> 16) + 1];
-    int u0 = u_buf[(x >> 17)];
-    int u1 = u_buf[(x >> 17) + 1];
-    int v0 = v_buf[(x >> 17)];
-    int v1 = v_buf[(x >> 17) + 1];
-    int y_frac = (x & 65535);
-    int uv_frac = ((x >> 1) & 65535);
-    int y = (y_frac * y1 + (y_frac ^ 65535) * y0) >> 16;
-    int u = (uv_frac * u1 + (uv_frac ^ 65535) * u0) >> 16;
-    int v = (uv_frac * v1 + (uv_frac ^ 65535) * v0) >> 16;
-    YuvPixel(y, u, v, rgb_buf);
-    x += source_dx;
-    if ((i + 1) < width) {
-      y0 = y_buf[x >> 16];
-      y1 = y_buf[(x >> 16) + 1];
-      y_frac = (x & 65535);
-      y = (y_frac * y1 + (y_frac ^ 65535) * y0) >> 16;
-      YuvPixel(y, u, v, rgb_buf+4);
-      x += source_dx;
-    }
-    rgb_buf += 8;
-  }
-}
-
-#endif  // USE_MMX
-}  // extern "C"
-
+  LinearScaleYUVToRGB32Row_C(y_buf, u_buf, v_buf, rgb_buf, width, source_dx);
+}
+#endif
+
+}
diff --git a/gfx/ycbcr/yuv_row_table.cpp b/gfx/ycbcr/yuv_row_table.cpp
--- a/gfx/ycbcr/yuv_row_table.cpp
+++ b/gfx/ycbcr/yuv_row_table.cpp
@@ -1,13 +1,13 @@
 // Copyright (c) 2010 The Chromium Authors. All rights reserved.
 // Use of this source code is governed by a BSD-style license that can be
 // found in the LICENSE file.
 
-#include "media/base/yuv_row.h"
+#include "yuv_row.h"
 
 extern "C" {
 
 #define RGBY(i) { \
   static_cast<int16>(1.164 * 64 * (i - 16) + 0.5), \
   static_cast<int16>(1.164 * 64 * (i - 16) + 0.5), \
   static_cast<int16>(1.164 * 64 * (i - 16) + 0.5), \
   0 \
diff --git a/gfx/ycbcr/yuv_row_win.cpp b/gfx/ycbcr/yuv_row_win.cpp
--- a/gfx/ycbcr/yuv_row_win.cpp
+++ b/gfx/ycbcr/yuv_row_win.cpp
@@ -1,26 +1,27 @@
 // Copyright (c) 2010 The Chromium Authors. All rights reserved.
 // Use of this source code is governed by a BSD-style license that can be
 // found in the LICENSE file.
 
-#include "media/base/yuv_row.h"
+#include "yuv_row.h"
+#include "mozilla/SSE.h"
 
 #define kCoefficientsRgbU kCoefficientsRgbY + 2048
 #define kCoefficientsRgbV kCoefficientsRgbY + 4096
 
 extern "C" {
 
-#if USE_MMX
-__declspec(naked)
-void FastConvertYUVToRGB32Row(const uint8* y_buf,
-                              const uint8* u_buf,
-                              const uint8* v_buf,
-                              uint8* rgb_buf,
-                              int width) {
+#if defined(MOZILLA_MAY_SUPPORT_SSE) && defined(_M_IX86)
+__declspec(naked)
+void FastConvertYUVToRGB32Row_SSE(const uint8* y_buf,
+                                  const uint8* u_buf,
+                                  const uint8* v_buf,
+                                  uint8* rgb_buf,
+                                  int width) {
   __asm {
     pushad
     mov       edx, [esp + 32 + 4]   // Y
     mov       edi, [esp + 32 + 8]   // U
     mov       esi, [esp + 32 + 12]  // V
     mov       ebp, [esp + 32 + 16]  // rgb
     mov       ecx, [esp + 32 + 20]  // width
     jmp       convertend
@@ -64,22 +65,22 @@ void FastConvertYUVToRGB32Row(const uint
  convertdone :
 
     popad
     ret
   }
 }
 
 __declspec(naked)
-void ConvertYUVToRGB32Row(const uint8* y_buf,
-                          const uint8* u_buf,
-                          const uint8* v_buf,
-                          uint8* rgb_buf,
-                          int width,
-                          int step) {
+void ConvertYUVToRGB32Row_SSE(const uint8* y_buf,
+                              const uint8* u_buf,
+                              const uint8* v_buf,
+                              uint8* rgb_buf,
+                              int width,
+                              int step) {
   __asm {
     pushad
     mov       edx, [esp + 32 + 4]   // Y
     mov       edi, [esp + 32 + 8]   // U
     mov       esi, [esp + 32 + 12]  // V
     mov       ebp, [esp + 32 + 16]  // rgb
     mov       ecx, [esp + 32 + 20]  // width
     mov       ebx, [esp + 32 + 24]  // step
@@ -125,23 +126,23 @@ void ConvertYUVToRGB32Row(const uint8* y
  wdone :
 
     popad
     ret
   }
 }
 
 __declspec(naked)
-void RotateConvertYUVToRGB32Row(const uint8* y_buf,
-                                const uint8* u_buf,
-                                const uint8* v_buf,
-                                uint8* rgb_buf,
-                                int width,
-                                int ystep,
-                                int uvstep) {
+void RotateConvertYUVToRGB32Row_SSE(const uint8* y_buf,
+                                    const uint8* u_buf,
+                                    const uint8* v_buf,
+                                    uint8* rgb_buf,
+                                    int width,
+                                    int ystep,
+                                    int uvstep) {
   __asm {
     pushad
     mov       edx, [esp + 32 + 4]   // Y
     mov       edi, [esp + 32 + 8]   // U
     mov       esi, [esp + 32 + 12]  // V
     mov       ebp, [esp + 32 + 16]  // rgb
     mov       ecx, [esp + 32 + 20]  // width
     jmp       wend
@@ -188,21 +189,21 @@ void RotateConvertYUVToRGB32Row(const ui
  wdone :
 
     popad
     ret
   }
 }
 
 __declspec(naked)
-void DoubleYUVToRGB32Row(const uint8* y_buf,
-                         const uint8* u_buf,
-                         const uint8* v_buf,
-                         uint8* rgb_buf,
-                         int width) {
+void DoubleYUVToRGB32Row_SSE(const uint8* y_buf,
+                             const uint8* u_buf,
+                             const uint8* v_buf,
+                             uint8* rgb_buf,
+                             int width) {
   __asm {
     pushad
     mov       edx, [esp + 32 + 4]   // Y
     mov       edi, [esp + 32 + 8]   // U
     mov       esi, [esp + 32 + 12]  // V
     mov       ebp, [esp + 32 + 16]  // rgb
     mov       ecx, [esp + 32 + 20]  // width
     jmp       wend
@@ -256,26 +257,26 @@ void DoubleYUVToRGB32Row(const uint8* y_
     jns       wloop1
  wdone :
     popad
     ret
   }
 }
 
 // This version does general purpose scaling by any amount, up or down.
-// The only thing it can not do it rotation by 90 or 270.
-// For performance the chroma is under sampled, reducing cost of a 3x
+// The only thing it cannot do is rotation by 90 or 270.
+// For performance the chroma is under-sampled, reducing cost of a 3x
 // 1080p scale from 8.4 ms to 5.4 ms.
 __declspec(naked)
-void ScaleYUVToRGB32Row(const uint8* y_buf,
-                        const uint8* u_buf,
-                        const uint8* v_buf,
-                        uint8* rgb_buf,
-                        int width,
-                        int source_dx) {
+void ScaleYUVToRGB32Row_SSE(const uint8* y_buf,
+                            const uint8* u_buf,
+                            const uint8* v_buf,
+                            uint8* rgb_buf,
+                            int width,
+                            int source_dx) {
   __asm {
     pushad
     mov       edx, [esp + 32 + 4]   // Y
     mov       edi, [esp + 32 + 8]   // U
     mov       esi, [esp + 32 + 12]  // V
     mov       ebp, [esp + 32 + 16]  // rgb
     mov       ecx, [esp + 32 + 20]  // width
     xor       ebx, ebx              // x
@@ -333,22 +334,22 @@ void ScaleYUVToRGB32Row(const uint8* y_b
 
  scaledone :
     popad
     ret
   }
 }
 
 __declspec(naked)
-void LinearScaleYUVToRGB32Row(const uint8* y_buf,
-                              const uint8* u_buf,
-                              const uint8* v_buf,
-                              uint8* rgb_buf,
-                              int width,
-                              int source_dx) {
+void LinearScaleYUVToRGB32Row_SSE(const uint8* y_buf,
+                                  const uint8* u_buf,
+                                  const uint8* v_buf,
+                                  uint8* rgb_buf,
+                                  int width,
+                                  int source_dx) {
   __asm {
     pushad
     mov       edx, [esp + 32 + 4]  // Y
     mov       edi, [esp + 32 + 8]  // U
                 // [esp + 32 + 12] // V
     mov       ebp, [esp + 32 + 16] // rgb
     mov       ecx, [esp + 32 + 20] // width
     imul      ecx, [esp + 32 + 24] // source_dx
@@ -438,152 +439,60 @@ lscalelastpixel:
     paddsw    mm1, mm0
     psraw     mm1, 6
     packuswb  mm1, mm1
     movd      [ebp], mm1
     popad
     ret
   };
 }
-#else  // USE_MMX
-
-// C reference code that mimic the YUV assembly.
-#define packuswb(x) ((x) < 0 ? 0 : ((x) > 255 ? 255 : (x)))
-#define paddsw(x, y) (((x) + (y)) < -32768 ? -32768 : \
-    (((x) + (y)) > 32767 ? 32767 : ((x) + (y))))
-
-static inline void YuvPixel(uint8 y,
-                            uint8 u,
-                            uint8 v,
-                            uint8* rgb_buf) {
-
-  int b = kCoefficientsRgbY[256+u][0];
-  int g = kCoefficientsRgbY[256+u][1];
-  int r = kCoefficientsRgbY[256+u][2];
-  int a = kCoefficientsRgbY[256+u][3];
-
-  b = paddsw(b, kCoefficientsRgbY[512+v][0]);
-  g = paddsw(g, kCoefficientsRgbY[512+v][1]);
-  r = paddsw(r, kCoefficientsRgbY[512+v][2]);
-  a = paddsw(a, kCoefficientsRgbY[512+v][3]);
-
-  b = paddsw(b, kCoefficientsRgbY[y][0]);
-  g = paddsw(g, kCoefficientsRgbY[y][1]);
-  r = paddsw(r, kCoefficientsRgbY[y][2]);
-  a = paddsw(a, kCoefficientsRgbY[y][3]);
-
-  b >>= 6;
-  g >>= 6;
-  r >>= 6;
-  a >>= 6;
-
-  *reinterpret_cast<uint32*>(rgb_buf) = (packuswb(b)) |
-                                        (packuswb(g) << 8) |
-                                        (packuswb(r) << 16) |
-                                        (packuswb(a) << 24);
-}
-
-#if TEST_MMX_YUV
-static inline void YuvPixel(uint8 y,
-                            uint8 u,
-                            uint8 v,
-                            uint8* rgb_buf) {
-
-  __asm {
-    movzx     eax, u
-    movq      mm0, [kCoefficientsRgbY+2048 + 8 * eax]
-    movzx     eax, v
-    paddsw    mm0, [kCoefficientsRgbY+4096 + 8 * eax]
-    movzx     eax, y
-    movq      mm1, [kCoefficientsRgbY + 8 * eax]
-    paddsw    mm1, mm0
-    psraw     mm1, 6
-    packuswb  mm1, mm1
-    mov       eax, rgb_buf
-    movd      [eax], mm1
-    emms
-  }
-}
-#endif
+#endif // if defined(MOZILLA_MAY_SUPPORT_SSE) && defined(_M_IX86)
 
 void FastConvertYUVToRGB32Row(const uint8* y_buf,
                               const uint8* u_buf,
                               const uint8* v_buf,
                               uint8* rgb_buf,
                               int width) {
-  for (int x = 0; x < width; x += 2) {
-    uint8 u = u_buf[x >> 1];
-    uint8 v = v_buf[x >> 1];
-    uint8 y0 = y_buf[x];
-    YuvPixel(y0, u, v, rgb_buf);
-    if ((x + 1) < width) {
-      uint8 y1 = y_buf[x + 1];
-      YuvPixel(y1, u, v, rgb_buf + 4);
-    }
-    rgb_buf += 8;  // Advance 2 pixels.
-  }
-}
-
-// 16.16 fixed point is used.  A shift by 16 isolates the integer.
-// A shift by 17 is used to further subsample the chrominence channels.
-// & 0xffff isolates the fixed point fraction.  >> 2 to get the upper 2 bits,
-// for 1/65536 pixel accurate interpolation.
+#if defined(MOZILLA_MAY_SUPPORT_SSE) && defined(_M_IX86)
+  if (mozilla::supports_sse()) {
+    FastConvertYUVToRGB32Row_SSE(y_buf, u_buf, v_buf, rgb_buf, width);
+    return;
+  }
+#endif
+
+  FastConvertYUVToRGB32Row_C(y_buf, u_buf, v_buf, rgb_buf, width, 1);
+}
+
 void ScaleYUVToRGB32Row(const uint8* y_buf,
                         const uint8* u_buf,
                         const uint8* v_buf,
                         uint8* rgb_buf,
                         int width,
                         int source_dx) {
-  int x = 0;
-  for (int i = 0; i < width; i += 2) {
-    int y = y_buf[x >> 16];
-    int u = u_buf[(x >> 17)];
-    int v = v_buf[(x >> 17)];
-    YuvPixel(y, u, v, rgb_buf);
-    x += source_dx;
-    if ((i + 1) < width) {
-      y = y_buf[x >> 16];
-      YuvPixel(y, u, v, rgb_buf+4);
-      x += source_dx;
-    }
-    rgb_buf += 8;
-  }
-}
+
+#if defined(MOZILLA_MAY_SUPPORT_SSE) && defined(_M_IX86)
+  if (mozilla::supports_sse()) {
+    ScaleYUVToRGB32Row_SSE(y_buf, u_buf, v_buf, rgb_buf, width, source_dx);
+    return;
+  }
+#endif
+
+  ScaleYUVToRGB32Row_C(y_buf, u_buf, v_buf, rgb_buf, width, source_dx);
+}
 
 void LinearScaleYUVToRGB32Row(const uint8* y_buf,
                               const uint8* u_buf,
                               const uint8* v_buf,
                               uint8* rgb_buf,
                               int width,
                               int source_dx) {
-  int x = 0;
-  if (source_dx >= 0x20000) {
-    x = 32768;
-  }
-  for (int i = 0; i < width; i += 2) {
-    int y0 = y_buf[x >> 16];
-    int y1 = y_buf[(x >> 16) + 1];
-    int u0 = u_buf[(x >> 17)];
-    int u1 = u_buf[(x >> 17) + 1];
-    int v0 = v_buf[(x >> 17)];
-    int v1 = v_buf[(x >> 17) + 1];
-    int y_frac = (x & 65535);
-    int uv_frac = ((x >> 1) & 65535);
-    int y = (y_frac * y1 + (y_frac ^ 65535) * y0) >> 16;
-    int u = (uv_frac * u1 + (uv_frac ^ 65535) * u0) >> 16;
-    int v = (uv_frac * v1 + (uv_frac ^ 65535) * v0) >> 16;
-    YuvPixel(y, u, v, rgb_buf);
-    x += source_dx;
-    if ((i + 1) < width) {
-      y0 = y_buf[x >> 16];
-      y1 = y_buf[(x >> 16) + 1];
-      y_frac = (x & 65535);
-      y = (y_frac * y1 + (y_frac ^ 65535) * y0) >> 16;
-      YuvPixel(y, u, v, rgb_buf+4);
-      x += source_dx;
-    }
-    rgb_buf += 8;
-  }
-}
-
-#endif  // USE_MMX
-}  // extern "C"
-
+#if defined(MOZILLA_MAY_SUPPORT_SSE) && defined(_M_IX86)
+  if (mozilla::supports_sse()) {
+    LinearScaleYUVToRGB32Row_SSE(y_buf, u_buf, v_buf, rgb_buf, width,
+                                 source_dx);
+    return;
+  }
+#endif
+
+  LinearScaleYUVToRGB32Row_C(y_buf, u_buf, v_buf, rgb_buf, width, source_dx);
+}
+
+} // extern "C"