Revert "Mac YUV assembly"

This reverts commit r26614.

Review URL: http://codereview.chromium.org/213034

git-svn-id: svn://svn.chromium.org/chrome/trunk/src@26621 0039d316-1c4b-4281-b951-d872f2087c98

2 files changed, 273 insertions, 63 deletions

diff --git a/media/base/yuv_row.h b/media/base/yuv_row.h
index 37bd935..104b9c7 100644
--- a/media/base/yuv_row.h
+++ b/media/base/yuv_row.h
@@ -63,8 +63,9 @@ void ScaleYUVToRGB32Row(const uint8* y_buf,
 }  // extern "C"
 
 #if !defined(USE_MMX)
-// Windows, Mac and Linux x86 use MMX; x64 and other CPUs do not.
-#if defined(OS_WIN) || defined(ARCH_CPU_X86)
+#if defined(_MSC_VER)
+#define USE_MMX 1
+#elif defined(OS_LINUX) && defined(ARCH_CPU_X86)
 #define USE_MMX 1
 #else
 #define USE_MMX 0
diff --git a/media/base/yuv_row_mac.cc b/media/base/yuv_row_mac.cc
index 1f33317..c80b213 100644
--- a/media/base/yuv_row_mac.cc
+++ b/media/base/yuv_row_mac.cc
@@ -4,9 +4,17 @@
 
 #include "media/base/yuv_row.h"
 
+#ifdef _DEBUG
+#include "base/logging.h"
+#else
+#define DCHECK(a)
+#endif
+
+// TODO(fbarchard): Make MMX work in DLLs.  Currently only works in unittests.
 // TODO(fbarchard): Do 64 bit version.
 
 extern "C" {
+#if USE_MMX
 
 #define RGBY(i) { \
   static_cast<int16>(1.164 * 64 * (i - 16) + 0.5), \
@@ -29,8 +37,7 @@ extern "C" {
   0 \
 }
 
-#define MMX_ALIGNED(var) \
-  var __attribute__ ((section ("__TEXT,__text"))) __attribute__ ((aligned(16)))
+#define MMX_ALIGNED(var) var __attribute__((aligned(16)))
 
 MMX_ALIGNED(int16 kCoefficientsRgbY[256][4]) = {
   RGBY(0x00), RGBY(0x01), RGBY(0x02), RGBY(0x03),
@@ -238,36 +245,75 @@ MMX_ALIGNED(int16 kCoefficientsRgbV[256][4]) = {
 #undef RGBV
 #undef MMX_ALIGNED
 
-extern void MacConvertYUVToRGB32Row(const uint8* y_buf,
-                                    const uint8* u_buf,
-                                    const uint8* v_buf,
-                                    uint8* rgb_buf,
-                                    int width,
-                                    int16 *kCoefficientsRgbY);
+// TODO(fbarchard): Use the following function instead of
+// pure assembly to help make code more portable to 64 bit
+// and Mac, which has different labels.
+// no-gcse eliminates the frame pointer, freeing up ebp.
+
+#if defined(FUTURE_64BIT_VERSION)
+void __attribute__((optimize("O2", "no-gcse")))
+    NewFastConvertYUVToRGB32Row(const uint8* y_buf,
+                                const uint8* u_buf,
+                                const uint8* v_buf,
+                                uint8* rgb_buf,
+                                int width)  {
+  asm(
+  "shr    %4\n"
+"1:\n"
+  "movzb  (%1),%%eax\n"
+  "add    $0x1,%1\n"
+  "movzb  (%2),%%ebx\n"
+  "add    $0x1,%2\n"
+  "movq   kCoefficientsRgbU(,%%eax,8),%%mm0\n"
+  "movzb  (%0),%%eax\n"
+  "paddsw kCoefficientsRgbV(,%%ebx,8),%%mm0\n"
+  "movzb  0x1(%0),%%ebx\n"
+  "movq   kCoefficientsRgbY(,%%eax,8),%%mm1\n"
+  "add    $0x2,%0\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(%3)\n"
+  "add    $0x8,%3\n"
+  "sub    $0x1,%4\n"
+  "jne    1b\n"
+      : : "r"(y_buf),"r"(u_buf),"r"(v_buf),"r"(rgb_buf),"r"(width)
+      : "eax","ebx");
+}
+#endif
+
+extern void FastConvertYUVToRGB32Row(const uint8* y_buf,
+                                     const uint8* u_buf,
+                                     const uint8* v_buf,
+                                     uint8* rgb_buf,
+                                     int width);
+
   __asm__(
-" .globl _MacConvertYUVToRGB32Row\n"
-"_MacConvertYUVToRGB32Row:\n"
+" .globl _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    0x38(%esp),%ecx\n"
-
-  "jmp    Lconvertend\n"
+  "mov    0x34(%esp),%ecx\n"
+  "jmp    convertend\n"
 
-"Lconvertloop:"
+"convertloop:"
   "movzbl (%edi),%eax\n"
   "add    $0x1,%edi\n"
   "movzbl (%esi),%ebx\n"
   "add    $0x1,%esi\n"
-  "movq   2048(%ecx,%eax,8),%mm0\n"
+  "movq   _kCoefficientsRgbU(,%eax,8),%mm0\n"
   "movzbl (%edx),%eax\n"
-  "paddsw 4096(%ecx,%ebx,8),%mm0\n"
+  "paddsw _kCoefficientsRgbV(,%ebx,8),%mm0\n"
   "movzbl 0x1(%edx),%ebx\n"
-  "movq   0(%ecx,%eax,8),%mm1\n"
+  "movq   _kCoefficientsRgbY(,%eax,8),%mm1\n"
   "add    $0x2,%edx\n"
-  "movq   0(%ecx,%ebx,8),%mm2\n"
+  "movq   _kCoefficientsRgbY(,%ebx,8),%mm2\n"
   "paddsw %mm0,%mm1\n"
   "paddsw %mm0,%mm2\n"
   "psraw  $0x6,%mm1\n"
@@ -275,76 +321,67 @@ extern void MacConvertYUVToRGB32Row(const uint8* y_buf,
   "packuswb %mm2,%mm1\n"
   "movntq %mm1,0x0(%ebp)\n"
   "add    $0x8,%ebp\n"
-"Lconvertend:"
-  "sub    $0x2,0x34(%esp)\n"
-  "jns    Lconvertloop\n"
+"convertend:"
+  "sub    $0x2,%ecx\n"
+  "jns    convertloop\n"
 
-  "and    $0x1,0x34(%esp)\n"
-  "je     Lconvertdone\n"
+  "and    $0x1,%ecx\n"
+  "je     convertdone\n"
 
   "movzbl (%edi),%eax\n"
-  "movq   2048(%ecx,%eax,8),%mm0\n"
+  "movq   _kCoefficientsRgbU(,%eax,8),%mm0\n"
   "movzbl (%esi),%eax\n"
-  "paddsw 4096(%ecx,%eax,8),%mm0\n"
+  "paddsw _kCoefficientsRgbV(,%eax,8),%mm0\n"
   "movzbl (%edx),%eax\n"
-  "movq   0(%ecx,%eax,8),%mm1\n"
+  "movq   _kCoefficientsRgbY(,%eax,8),%mm1\n"
   "paddsw %mm0,%mm1\n"
   "psraw  $0x6,%mm1\n"
   "packuswb %mm1,%mm1\n"
   "movd   %mm1,0x0(%ebp)\n"
-"Lconvertdone:\n"
+"convertdone:"
   "popa\n"
   "ret\n"
 );
 
-void FastConvertYUVToRGB32Row(const uint8* y_buf,
-                              const uint8* u_buf,
-                              const uint8* v_buf,
-                              uint8* rgb_buf,
-                              int width) {
-  MacConvertYUVToRGB32Row(y_buf, u_buf, v_buf, rgb_buf, width,
-                          &kCoefficientsRgbY[0][0]);
-}
 
-extern void MacScaleYUVToRGB32Row(const uint8* y_buf,
+extern void ScaleYUVToRGB32Row(const uint8* y_buf,
                                const uint8* u_buf,
                                const uint8* v_buf,
                                uint8* rgb_buf,
                                int width,
-                               int scaled_dx,
-                               int16 *kCoefficientsRgbY);
+                               int scaled_dx);
 
   __asm__(
-" .globl _MacScaleYUVToRGB32Row\n"
-"_MacScaleYUVToRGB32Row:\n"
+" .globl _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    0x3c(%esp),%ecx\n"
+  "mov    0x34(%esp),%ecx\n"
   "xor    %ebx,%ebx\n"
-  "jmp    Lscaleend\n"
+  "jmp    scaleend\n"
 
-"Lscaleloop:"
+"scaleloop:"
   "mov    %ebx,%eax\n"
   "sar    $0x5,%eax\n"
   "movzbl (%edi,%eax,1),%eax\n"
-  "movq   2048(%ecx,%eax,8),%mm0\n"
+  "movq   kCoefficientsRgbU(,%eax,8),%mm0\n"
   "mov    %ebx,%eax\n"
   "sar    $0x5,%eax\n"
   "movzbl (%esi,%eax,1),%eax\n"
-  "paddsw 4096(%ecx,%eax,8),%mm0\n"
+  "paddsw kCoefficientsRgbV(,%eax,8),%mm0\n"
   "mov    %ebx,%eax\n"
   "add    0x38(%esp),%ebx\n"
   "sar    $0x4,%eax\n"
   "movzbl (%edx,%eax,1),%eax\n"
-  "movq   0(%ecx,%eax,8),%mm1\n"
+  "movq   kCoefficientsRgbY(,%eax,8),%mm1\n"
   "mov    %ebx,%eax\n"
   "add    0x38(%esp),%ebx\n"
   "sar    $0x4,%eax\n"
   "movzbl (%edx,%eax,1),%eax\n"
-  "movq   0(%ecx,%eax,8),%mm2\n"
+  "movq   kCoefficientsRgbY(,%eax,8),%mm2\n"
   "paddsw %mm0,%mm1\n"
   "paddsw %mm0,%mm2\n"
   "psraw  $0x6,%mm1\n"
@@ -352,29 +389,29 @@ extern void MacScaleYUVToRGB32Row(const uint8* y_buf,
   "packuswb %mm2,%mm1\n"
   "movntq %mm1,0x0(%ebp)\n"
   "add    $0x8,%ebp\n"
-"Lscaleend:"
-  "sub    $0x2,0x34(%esp)\n"
-  "jns    Lscaleloop\n"
+"scaleend:"
+  "sub    $0x2,%ecx\n"
+  "jns    scaleloop\n"
 
-  "and    $0x1,0x34(%esp)\n"
-  "je     Lscaledone\n"
+  "and    $0x1,%ecx\n"
+  "je     scaledone\n"
 
   "mov    %ebx,%eax\n"
   "sar    $0x5,%eax\n"
   "movzbl (%edi,%eax,1),%eax\n"
-  "movq   2048(%ecx,%eax,8),%mm0\n"
+  "movq   kCoefficientsRgbU(,%eax,8),%mm0\n"
   "mov    %ebx,%eax\n"
   "sar    $0x5,%eax\n"
   "movzbl (%esi,%eax,1),%eax\n"
-  "paddsw 4096(%ecx,%eax,8),%mm0\n"
+  "paddsw kCoefficientsRgbV(,%eax,8),%mm0\n"
   "mov    %ebx,%eax\n"
   "sar    $0x4,%eax\n"
   "movzbl (%edx,%eax,1),%eax\n"
-  "movq   0(%ecx,%eax,8),%mm1\n"
+  "movq   kCoefficientsRgbY(,%eax,8),%mm1\n"
   "mov    %ebx,%eax\n"
   "sar    $0x4,%eax\n"
   "movzbl (%edx,%eax,1),%eax\n"
-  "movq   0(%ecx,%eax,8),%mm2\n"
+  "movq   kCoefficientsRgbY(,%eax,8),%mm2\n"
   "paddsw %mm0,%mm1\n"
   "paddsw %mm0,%mm2\n"
   "psraw  $0x6,%mm1\n"
@@ -382,22 +419,194 @@ extern void MacScaleYUVToRGB32Row(const uint8* y_buf,
   "packuswb %mm2,%mm1\n"
   "movd   %mm1,0x0(%ebp)\n"
 
-"Lscaledone:"
+"scaledone:"
   "popa\n"
   "ret\n"
 );
 
+#else  // USE_MMX
+
+// Reference version of YUV converter.
+static const int kClipTableSize = 256;
+static const int kClipOverflow = 288;  // Cb max is 535.
+
+static uint8 kRgbClipTable[kClipOverflow +
+                           kClipTableSize +
+                           kClipOverflow] = {
+  0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,  // 288 underflow values
+  0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,  // clipped to 0.
+  0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
+  0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
+  0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
+  0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
+  0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
+  0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
+  0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
+  0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
+  0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
+  0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
+  0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
+  0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
+  0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
+  0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
+  0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
+  0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
+  0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
+  0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
+  0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
+  0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
+  0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
+  0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
+  0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
+  0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
+  0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
+  0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
+  0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
+  0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
+  0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
+  0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
+  0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
+  0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
+  0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
+  0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
+  0x00, 0x01, 0x02, 0x03, 0x04, 0x05, 0x06, 0x07,  // Unclipped values.
+  0x08, 0x09, 0x0A, 0x0B, 0x0C, 0x0D, 0x0E, 0x0F,
+  0x10, 0x11, 0x12, 0x13, 0x14, 0x15, 0x16, 0x17,
+  0x18, 0x19, 0x1A, 0x1B, 0x1C, 0x1D, 0x1E, 0x1F,
+  0x20, 0x21, 0x22, 0x23, 0x24, 0x25, 0x26, 0x27,
+  0x28, 0x29, 0x2A, 0x2B, 0x2C, 0x2D, 0x2E, 0x2F,
+  0x30, 0x31, 0x32, 0x33, 0x34, 0x35, 0x36, 0x37,
+  0x38, 0x39, 0x3A, 0x3B, 0x3C, 0x3D, 0x3E, 0x3F,
+  0x40, 0x41, 0x42, 0x43, 0x44, 0x45, 0x46, 0x47,
+  0x48, 0x49, 0x4A, 0x4B, 0x4C, 0x4D, 0x4E, 0x4F,
+  0x50, 0x51, 0x52, 0x53, 0x54, 0x55, 0x56, 0x57,
+  0x58, 0x59, 0x5A, 0x5B, 0x5C, 0x5D, 0x5E, 0x5F,
+  0x60, 0x61, 0x62, 0x63, 0x64, 0x65, 0x66, 0x67,
+  0x68, 0x69, 0x6A, 0x6B, 0x6C, 0x6D, 0x6E, 0x6F,
+  0x70, 0x71, 0x72, 0x73, 0x74, 0x75, 0x76, 0x77,
+  0x78, 0x79, 0x7A, 0x7B, 0x7C, 0x7D, 0x7E, 0x7F,
+  0x80, 0x81, 0x82, 0x83, 0x84, 0x85, 0x86, 0x87,
+  0x88, 0x89, 0x8A, 0x8B, 0x8C, 0x8D, 0x8E, 0x8F,
+  0x90, 0x91, 0x92, 0x93, 0x94, 0x95, 0x96, 0x97,
+  0x98, 0x99, 0x9A, 0x9B, 0x9C, 0x9D, 0x9E, 0x9F,
+  0xA0, 0xA1, 0xA2, 0xA3, 0xA4, 0xA5, 0xA6, 0xA7,
+  0xA8, 0xA9, 0xAA, 0xAB, 0xAC, 0xAD, 0xAE, 0xAF,
+  0xB0, 0xB1, 0xB2, 0xB3, 0xB4, 0xB5, 0xB6, 0xB7,
+  0xB8, 0xB9, 0xBA, 0xBB, 0xBC, 0xBD, 0xBE, 0xBF,
+  0xC0, 0xC1, 0xC2, 0xC3, 0xC4, 0xC5, 0xC6, 0xC7,
+  0xC8, 0xC9, 0xCA, 0xCB, 0xCC, 0xCD, 0xCE, 0xCF,
+  0xD0, 0xD1, 0xD2, 0xD3, 0xD4, 0xD5, 0xD6, 0xD7,
+  0xD8, 0xD9, 0xDA, 0xDB, 0xDC, 0xDD, 0xDE, 0xDF,
+  0xE0, 0xE1, 0xE2, 0xE3, 0xE4, 0xE5, 0xE6, 0xE7,
+  0xE8, 0xE9, 0xEA, 0xEB, 0xEC, 0xED, 0xEE, 0xEF,
+  0xF0, 0xF1, 0xF2, 0xF3, 0xF4, 0xF5, 0xF6, 0xF7,
+  0xF8, 0xF9, 0xFA, 0xFB, 0xFC, 0xFD, 0xFE, 0xFF,
+  0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF,  // 288 overflow values
+  0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF,  // clipped to 255.
+  0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF,
+  0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF,
+  0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF,
+  0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF,
+  0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF,
+  0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF,
+  0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF,
+  0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF,
+  0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF,
+  0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF,
+  0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF,
+  0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF,
+  0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF,
+  0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF,
+  0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF,
+  0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF,
+  0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF,
+  0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF,
+  0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF,
+  0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF,
+  0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF,
+  0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF,
+  0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF,
+  0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF,
+  0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF,
+  0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF,
+  0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF,
+  0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF,
+  0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF,
+  0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF,
+  0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF,
+  0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF,
+  0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF,
+  0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF,
+};
+
+// Clip an rgb channel value to 0..255 range.
+// Source is signed fixed point 8.8.
+// Table allows for values to underflow or overflow by 128.
+// Therefore source range is -128 to 384.
+// Output clips to unsigned 0 to 255.
+static inline uint32 clip(int32 value) {
+  DCHECK(((value >> 8) + kClipOverflow) >= 0);
+  DCHECK(((value >> 8) + kClipOverflow) <
+         (kClipOverflow + kClipTableSize + kClipOverflow));
+  return static_cast<uint32>(kRgbClipTable[((value) >> 8) + kClipOverflow]);
+}
+
+static inline void YuvPixel(uint8 y,
+                            uint8 u,
+                            uint8 v,
+                            uint8* rgb_buf) {
+  int32 d = static_cast<int32>(u) - 128;
+  int32 e = static_cast<int32>(v) - 128;
+
+  int32 cb =   (516 * d + 128);
+  int32 cg = (- 100 * d - 208 * e + 128);
+  int32 cr =             (409 * e + 128);
 
+  int32 C298a = ((static_cast<int32>(y) - 16) * 298 + 128);
+  *reinterpret_cast<uint32*>(rgb_buf) = (clip(C298a + cb)) |
+                                        (clip(C298a + cg) << 8) |
+                                        (clip(C298a + cr) << 16) |
+                                        (0xff000000);
+}
+
+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.
+  }
+}
+
+// 28.4 fixed point is used.  A shift by 4 isolates the integer.
+// A shift by 5 is used to further subsample the chrominence channels.
+// & 15 isolates the fixed point fraction.  >> 2 to get the upper 2 bits,
+// for 1/4 pixel accurate interpolation.
 void ScaleYUVToRGB32Row(const uint8* y_buf,
                         const uint8* u_buf,
                         const uint8* v_buf,
                         uint8* rgb_buf,
                         int width,
                         int scaled_dx) {
-
-  MacScaleYUVToRGB32Row(y_buf, u_buf, v_buf, rgb_buf, width, scaled_dx,
-                        &kCoefficientsRgbY[0][0]);
+  int scaled_x = 0;
+  for (int x = 0; x < width; ++x) {
+    uint8 u = u_buf[scaled_x >> 5];
+    uint8 v = v_buf[scaled_x >> 5];
+    uint8 y0 = y_buf[scaled_x >> 4];
+    YuvPixel(y0, u, v, rgb_buf);
+    rgb_buf += 4;
+    scaled_x += scaled_dx;
+  }
 }
-
+#endif // USE_MMX
 }  // extern "C"