remove mmx functions

BUG=none
TEST=builds
Review URL: http://webrtc-codereview.appspot.com/269010

git-svn-id: http://libyuv.googlecode.com/svn/trunk@77 16f28f9a-4ce2-e073-06de-1de4eb20be90

README.chromium

@@ -1,6 +1,6 @@
 Name: libyuv
 URL: http://code.google.com/p/libyuv/
-Version: 69
+Version: 77
 License: BSD
 License File: LICENSE
 

source/planar_functions.cc

@@ -1149,11 +1149,6 @@ int I420ToARGB(const uint8* src_y, int src_stride_y,
       (width % 2 == 0)) {
     FastConvertYUVToARGBRow = FastConvertYUVToARGBRow_SSE2;
   } else
-#endif
-#if defined(HAS_FASTCONVERTYUVTOARGBROW_MMX)
-  if (width % 2 == 0) {
-    FastConvertYUVToARGBRow = FastConvertYUVToARGBRow_MMX;
-  } else
 #endif
   {
     FastConvertYUVToARGBRow = FastConvertYUVToARGBRow_C;
@@ -1167,8 +1162,6 @@ int I420ToARGB(const uint8* src_y, int src_stride_y,
       src_v += src_stride_v;
     }
   }
-  // MMX used for FastConvertYUVToARGBRow requires an emms instruction.
-  EMMS();
   return 0;
 }
 
@@ -1201,11 +1194,6 @@ int I420ToBGRA(const uint8* src_y, int src_stride_y,
       (width % 2 == 0)) {
     FastConvertYUVToBGRARow = FastConvertYUVToBGRARow_SSE2;
   } else
-#endif
-#if defined(HAS_FASTCONVERTYUVTOBGRAROW_MMX)
-  if (width % 2 == 0) {
-    FastConvertYUVToBGRARow = FastConvertYUVToBGRARow_MMX;
-  } else
 #endif
   {
     FastConvertYUVToBGRARow = FastConvertYUVToBGRARow_C;
@@ -1219,7 +1207,6 @@ int I420ToBGRA(const uint8* src_y, int src_stride_y,
       src_v += src_stride_v;
     }
   }
-  EMMS();
   return 0;
 }
 
@@ -1252,11 +1239,6 @@ int I420ToABGR(const uint8* src_y, int src_stride_y,
       (width % 2 == 0)) {
     FastConvertYUVToABGRRow = FastConvertYUVToABGRRow_SSE2;
   } else
-#endif
-#if defined(HAS_FASTCONVERTYUVTOABGRROW_MMX)
-  if (width % 2 == 0) {
-    FastConvertYUVToABGRRow = FastConvertYUVToABGRRow_MMX;
-  } else
 #endif
   {
     FastConvertYUVToABGRRow = FastConvertYUVToABGRRow_C;
@@ -1270,7 +1252,6 @@ int I420ToABGR(const uint8* src_y, int src_stride_y,
       src_v += src_stride_v;
     }
   }
-  EMMS();
   return 0;
 }
 
@@ -1303,11 +1284,6 @@ int I422ToARGB(const uint8* src_y, int src_stride_y,
       (width % 2 == 0)) {
     FastConvertYUVToARGBRow = FastConvertYUVToARGBRow_SSE2;
   } else
-#endif
-#if defined(HAS_FASTCONVERTYUVTOARGBROW_MMX)
-  if (width % 2 == 0) {
-    FastConvertYUVToARGBRow = FastConvertYUVToARGBRow_MMX;
-  } else
 #endif
   {
     FastConvertYUVToARGBRow = FastConvertYUVToARGBRow_C;
@@ -1319,8 +1295,6 @@ int I422ToARGB(const uint8* src_y, int src_stride_y,
     src_u += src_stride_u;
     src_v += src_stride_v;
   }
-  // MMX used for FastConvertYUVToARGBRow requires an emms instruction.
-  EMMS();
   return 0;
 }
 
@@ -1353,13 +1327,9 @@ int I444ToARGB(const uint8* src_y, int src_stride_y,
     FastConvertYUV444ToARGBRow = FastConvertYUV444ToARGBRow_SSE2;
   } else
 #endif
-#if defined(HAS_FASTCONVERTYUVTOARGBROW_MMX)
-    FastConvertYUV444ToARGBRow = FastConvertYUV444ToARGBRow_MMX;
-#else
   {
     FastConvertYUV444ToARGBRow = FastConvertYUV444ToARGBRow_C;
   }
-#endif
   for (int y = 0; y < height; ++y) {
     FastConvertYUV444ToARGBRow(src_y, src_u, src_v, dst_argb, width);
     dst_argb += dst_stride_argb;
@@ -1367,8 +1337,6 @@ int I444ToARGB(const uint8* src_y, int src_stride_y,
     src_u += src_stride_u;
     src_v += src_stride_v;
   }
-  // MMX used for FastConvertYUVToARGBRow requires an emms instruction.
-  EMMS();
   return 0;
 }
 
@@ -1391,11 +1359,6 @@ int I400ToARGB_Reference(const uint8* src_y, int src_stride_y,
       IS_ALIGNED(dst_argb, 16) && (dst_stride_argb % 16 == 0)) {
     FastConvertYToARGBRow = FastConvertYToARGBRow_SSE2;
   } else
-#endif
-#if defined(HAS_FASTCONVERTYUVTOARGBROW_MMX)
-  if (width % 2 == 0) {
-    FastConvertYToARGBRow = FastConvertYToARGBRow_MMX;
-  } else
 #endif
   {
     FastConvertYToARGBRow = FastConvertYToARGBRow_C;
@@ -1405,8 +1368,6 @@ int I400ToARGB_Reference(const uint8* src_y, int src_stride_y,
     dst_argb += dst_stride_argb;
     src_y += src_stride_y;
   }
-  // MMX used for FastConvertYUVToARGBRow requires an emms instruction.
-  EMMS();
   return 0;
 }
 

source/rotate.cc

@@ -13,21 +13,19 @@
 #include "libyuv/cpu_id.h"
 #include "libyuv/planar_functions.h"
 #include "rotate_priv.h"
+#include "row.h"
 
 namespace libyuv {
 
-#if (defined(WIN32) || defined(__x86_64__) || defined(__i386__)) \
+#if (defined(WIN32) || defined(__x86_64__) || defined(__i386__)) && \
-    && !defined(COVERAGE_ENABLED) && !defined(TARGET_IPHONE_SIMULATOR)
+    !defined(__APPLE__) && \
-#if defined(_MSC_VER)
+    !defined(COVERAGE_ENABLED) && !defined(TARGET_IPHONE_SIMULATOR)
-#define TALIGN16(t, var) static __declspec(align(16)) t _ ## var
+
-#else
-#define TALIGN16(t, var) t var __attribute__((aligned(16)))
-#endif
 // Shuffle table for reversing the bytes.
-extern "C" TALIGN16(const uint8, kShuffleReverse[16]) =
+static const uvec8 kShuffleReverse =
   { 15u, 14u, 13u, 12u, 11u, 10u, 9u, 8u, 7u, 6u, 5u, 4u, 3u, 2u, 1u, 0u };
 // Shuffle table for reversing the bytes of UV channels.
-extern "C" TALIGN16(const uint8, kShuffleReverseUV[16]) =
+static const uvec8 kShuffleReverseUV =
   { 14u, 12u, 10u, 8u, 6u, 4u, 2u, 0u, 15u, 13u, 11u, 9u, 7u, 5u, 3u, 1u };
 #endif
 
@@ -73,7 +71,7 @@ __asm {
     mov       edx, [esp + 12 + 12]  // dst
     mov       esi, [esp + 12 + 16]  // dst_stride
     mov       ecx, [esp + 12 + 20]  // width
- convertloop :
+ convertloop:
     // Read in the data from the source pointer.
     // First round of bit swap.
     movq      xmm0, qword ptr [eax]
@@ -172,7 +170,7 @@ __asm {
     and       esp, ~15
     mov       [esp + 16], ecx
     mov       ecx, [ecx + 16 + 28]  // w
- convertloop :
+ convertloop:
     // Read in the data from the source pointer.
     // First round of bit swap.
     movdqa    xmm0, [eax]
@@ -863,9 +861,9 @@ __asm {
     mov       eax, [esp + 4]   // src
     mov       edx, [esp + 8]   // dst
     mov       ecx, [esp + 12]  // width
-    movdqa    xmm5, _kShuffleReverse
+    movdqa    xmm5, kShuffleReverse
     lea       eax, [eax + ecx - 16]
- convertloop :
+ convertloop:
     movdqa    xmm0, [eax]
     lea       eax, [eax - 16]
     pshufb    xmm0, xmm5
@@ -878,12 +876,16 @@ __asm {
 }
 
 #elif (defined(__i386__) || defined(__x86_64__)) && \
+    !defined(__APPLE__) && \
     !defined(COVERAGE_ENABLED) && !defined(TARGET_IPHONE_SIMULATOR)
 #define HAS_REVERSE_LINE_SSSE3
 static void ReverseLine_SSSE3(const uint8* src, uint8* dst, int width) {
   intptr_t temp_width = static_cast<intptr_t>(width);
   asm volatile (
-  "movdqa     (%3),%%xmm5                      \n"
+  "movdqa     %0,%%xmm5                        \n"
+  :: "m"(kShuffleReverse)
+  );
+  asm volatile (
   "lea        -0x10(%0,%2,1),%0                \n"
 "1:                                            \n"
   "movdqa     (%0),%%xmm0                      \n"
@@ -896,12 +898,12 @@ static void ReverseLine_SSSE3(const uint8* src, uint8* dst, int width) {
   : "+r"(src),  // %0
     "+r"(dst),  // %1
     "+r"(temp_width)  // %2
-  : "r"(kShuffleReverse)  // %3
+  :
   : "memory", "cc"
 #if defined(__SSE2__)
     , "xmm0", "xmm5"
 #endif
-);
+  );
 }
 #endif
 
@@ -1066,10 +1068,10 @@ __asm {
     mov       edx, [esp + 4 + 8]   // dst_a
     mov       edi, [esp + 4 + 12]  // dst_b
     mov       ecx, [esp + 4 + 16]  // width
-    movdqa    xmm5, _kShuffleReverseUV
+    movdqa    xmm5, kShuffleReverseUV
     lea       eax, [eax + ecx * 2 - 16]
 
- convertloop :
+ convertloop:
     movdqa    xmm0, [eax]
     lea       eax, [eax - 16]
     pshufb    xmm0, xmm5
@@ -1085,6 +1087,7 @@ __asm {
 }
 
 #elif (defined(__i386__) || defined(__x86_64__)) && \
+    !defined(__APPLE__) && \
     !defined(COVERAGE_ENABLED) && !defined(TARGET_IPHONE_SIMULATOR)
 #define HAS_REVERSE_LINE_UV_SSSE3
 void ReverseLineUV_SSSE3(const uint8* src,
@@ -1092,28 +1095,31 @@ void ReverseLineUV_SSSE3(const uint8* src,
                          int width) {
   intptr_t temp_width = static_cast<intptr_t>(width);
   asm volatile (
-  "movdqa     (%4),%%xmm5                      \n"
+  "movdqa     %0,%%xmm5                        \n"
-  "lea        -0x10(%0,%3,2),%0                \n"
+  :: "m"(kShuffleReverseUV)
+  );
+  asm volatile (
+  "lea        -16(%0,%3,2),%0                  \n"
 "1:                                            \n"
   "movdqa     (%0),%%xmm0                      \n"
-  "lea        -0x10(%0),%0                     \n"
+  "lea        -16(%0),%0                       \n"
   "pshufb     %%xmm5,%%xmm0                    \n"
   "movlpd     %%xmm0,(%1)                      \n"
-  "lea        0x8(%1),%1                       \n"
+  "lea        8(%1),%1                         \n"
   "movhpd     %%xmm0,(%2)                      \n"
-  "lea        0x8(%2),%2                       \n"
+  "lea        8(%2),%2                         \n"
-  "sub        $0x8,%3                          \n"
+  "sub        $8,%3                            \n"
   "ja         1b                               \n"
   : "+r"(src),      // %0
     "+r"(dst_a),    // %1
     "+r"(dst_b),    // %2
     "+r"(temp_width)  // %3
-  : "r"(kShuffleReverseUV)  // %4
+  :
   : "memory", "cc"
 #if defined(__SSE2__)
     , "xmm0", "xmm5"
 #endif
-);
+  );
 }
 #endif
 

source/row.h

@@ -51,15 +51,6 @@
 #define HAS_FASTCONVERTYTOARGBROW_SSE2
 #endif
 
-// The following are available on Windows and GCC 32 bit
-#if (defined(WIN32) || \
-    defined(__i386__)) && \
-    !defined(LIBYUV_DISABLE_ASM)
-#define HAS_FASTCONVERTYUVTOARGBROW_MMX
-#define HAS_FASTCONVERTYUVTOBGRAROW_MMX
-#define HAS_FASTCONVERTYUVTOABGRROW_MMX
-#endif
-
 // The following are available on Windows
 #if defined(WIN32) && \
     !defined(LIBYUV_DISABLE_ASM)
@@ -128,12 +119,14 @@ void I400ToARGBRow_C(const uint8* src_y, uint8* dst_argb, int pix);
 
 #if defined(_MSC_VER)
 #define SIMD_ALIGNED(var) __declspec(align(16)) var
-#define TALIGN16(t, var) static __declspec(align(16)) t _ ## var
+typedef __declspec(align(16)) signed char vec8[16];
+typedef __declspec(align(16)) unsigned char uvec8[16];
+typedef __declspec(align(16)) signed short vec16[8];
 #else // __GNUC__
 #define SIMD_ALIGNED(var) var __attribute__((aligned(16)))
-#define TALIGN16(t, var) t var __attribute__((aligned(16)))
 typedef signed char __attribute__((vector_size(16))) vec8;
 typedef unsigned char __attribute__((vector_size(16))) uvec8;
+typedef signed short __attribute__((vector_size(16))) vec16;
 #endif
 
 extern "C" SIMD_ALIGNED(const int16 kCoefficientsRgbY[768][4]);
@@ -204,36 +197,6 @@ void FastConvertYToARGBRow_SSE2(const uint8* y_buf,
                                 int width);
 #endif
 
-#ifdef HAS_FASTCONVERTYUVTOARGBROW_MMX
-void FastConvertYUVToARGBRow_MMX(const uint8* y_buf,
-                                 const uint8* u_buf,
-                                 const uint8* v_buf,
-                                 uint8* rgb_buf,
-                                 int width);
-
-void FastConvertYUVToBGRARow_MMX(const uint8* y_buf,
-                                 const uint8* u_buf,
-                                 const uint8* v_buf,
-                                 uint8* rgb_buf,
-                                 int width);
-
-void FastConvertYUVToABGRRow_MMX(const uint8* y_buf,
-                                 const uint8* u_buf,
-                                 const uint8* v_buf,
-                                 uint8* rgb_buf,
-                                 int width);
-
-void FastConvertYUV444ToARGBRow_MMX(const uint8* y_buf,
-                                    const uint8* u_buf,
-                                    const uint8* v_buf,
-                                    uint8* rgb_buf,
-                                    int width);
-
-void FastConvertYToARGBRow_MMX(const uint8* y_buf,
-                               uint8* rgb_buf,
-                               int width);
-#endif
-
 #ifdef HAS_FASTCONVERTYUVTOARGBROW_SSSE3
 void FastConvertYUVToARGBRow_SSSE3(const uint8* y_buf,
                                    const uint8* u_buf,
@@ -268,42 +231,6 @@ void FastConvertYToARGBRow_SSE2(const uint8* y_buf,
 
 #endif
 
-// Method to force C version.
-//#define USE_MMX 0
-//#define USE_SSE2 0
-
-#if !defined(USE_MMX)
-// Windows, Mac and Linux use MMX
-#if defined(__i386__) || 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(_MSC_VER)
-#define EMMS() __asm emms
-#pragma warning(disable: 4799)
-#else
-#define EMMS() asm("emms")
-#endif
-#else
-#define EMMS()
-#endif
-
-
 }  // extern "C"
 
 #endif  // LIBYUV_SOURCE_ROW_H_

source/row_posix.cc

@@ -542,231 +542,6 @@ void FastConvertYToARGBRow_SSE2(const uint8* y_buf,  // rdi
 #endif
 );
 }
-
-#endif
-
-#ifdef HAS_FASTCONVERTYUVTOARGBROW_MMX
-// 32 bit mmx gcc version
-
-#ifdef OSX
-#define UNDERSCORE "_"
-#else
-#define UNDERSCORE ""
-#endif
-
-void FastConvertYUVToARGBRow_MMX(const uint8* y_buf,
-                                 const uint8* u_buf,
-                                 const uint8* v_buf,
-                                 uint8* rgb_buf,
-                                 int width);
-  asm(
-  ".text                                       \n"
-#if defined(OSX) || defined(IOS)
-  ".globl _FastConvertYUVToARGBRow_MMX         \n"
-"_FastConvertYUVToARGBRow_MMX:                 \n"
-#else
-  ".global FastConvertYUVToARGBRow_MMX         \n"
-"FastConvertYUVToARGBRow_MMX:                  \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    0x34(%esp),%ecx                      \n"
-
-"1:                                            \n"
-  "movzbl (%edi),%eax                          \n"
-  "lea    1(%edi),%edi                         \n"
-  "movzbl (%esi),%ebx                          \n"
-  "lea    1(%esi),%esi                         \n"
-  "movq   " UNDERSCORE "kCoefficientsRgbY+2048(,%eax,8),%mm0\n"
-  "movzbl (%edx),%eax                          \n"
-  "paddsw " UNDERSCORE "kCoefficientsRgbY+4096(,%ebx,8),%mm0\n"
-  "movzbl 0x1(%edx),%ebx                       \n"
-  "movq   " UNDERSCORE "kCoefficientsRgbY(,%eax,8),%mm1\n"
-  "lea    2(%edx),%edx                         \n"
-  "movq   " UNDERSCORE "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"
-  "movq   %mm1,0x0(%ebp)                       \n"
-  "lea    8(%ebp),%ebp                         \n"
-  "sub    $0x2,%ecx                            \n"
-  "ja     1b                                   \n"
-  "popa                                        \n"
-  "ret                                         \n"
-);
-
-void FastConvertYUVToBGRARow_MMX(const uint8* y_buf,
-                                 const uint8* u_buf,
-                                 const uint8* v_buf,
-                                 uint8* rgb_buf,
-                                 int width);
-  asm(
-  ".text                                       \n"
-#if defined(OSX) || defined(IOS)
-  ".globl _FastConvertYUVToBGRARow_MMX         \n"
-"_FastConvertYUVToBGRARow_MMX:                 \n"
-#else
-  ".global FastConvertYUVToBGRARow_MMX         \n"
-"FastConvertYUVToBGRARow_MMX:                  \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    0x34(%esp),%ecx                      \n"
-
-"1:                                            \n"
-  "movzbl (%edi),%eax                          \n"
-  "lea    1(%edi),%edi                         \n"
-  "movzbl (%esi),%ebx                          \n"
-  "lea    1(%esi),%esi                         \n"
-  "movq   " UNDERSCORE "kCoefficientsBgraY+2048(,%eax,8),%mm0\n"
-  "movzbl (%edx),%eax                          \n"
-  "paddsw " UNDERSCORE "kCoefficientsBgraY+4096(,%ebx,8),%mm0\n"
-  "movzbl 0x1(%edx),%ebx                       \n"
-  "movq   " UNDERSCORE "kCoefficientsBgraY(,%eax,8),%mm1\n"
-  "lea    2(%edx),%edx                         \n"
-  "movq   " UNDERSCORE "kCoefficientsBgraY(,%ebx,8),%mm2\n"
-  "paddsw %mm0,%mm1                            \n"
-  "paddsw %mm0,%mm2                            \n"
-  "psraw  $0x6,%mm1                            \n"
-  "psraw  $0x6,%mm2                            \n"
-  "packuswb %mm2,%mm1                          \n"
-  "movq   %mm1,0x0(%ebp)                       \n"
-  "lea    8(%ebp),%ebp                         \n"
-  "sub    $0x2,%ecx                            \n"
-  "ja     1b                                   \n"
-  "popa                                        \n"
-  "ret                                         \n"
-);
-
-void FastConvertYUVToABGRRow_MMX(const uint8* y_buf,
-                                 const uint8* u_buf,
-                                 const uint8* v_buf,
-                                 uint8* rgb_buf,
-                                 int width);
-  asm(
-  ".text                                       \n"
-#if defined(OSX) || defined(IOS)
-  ".globl _FastConvertYUVToABGRRow_MMX         \n"
-"_FastConvertYUVToABGRRow_MMX:                 \n"
-#else
-  ".global FastConvertYUVToABGRRow_MMX         \n"
-"FastConvertYUVToABGRRow_MMX:                  \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    0x34(%esp),%ecx                      \n"
-
-"1:                                            \n"
-  "movzbl (%edi),%eax                          \n"
-  "lea    1(%edi),%edi                         \n"
-  "movzbl (%esi),%ebx                          \n"
-  "lea    1(%esi),%esi                         \n"
-  "movq   " UNDERSCORE "kCoefficientsAbgrY+2048(,%eax,8),%mm0\n"
-  "movzbl (%edx),%eax                          \n"
-  "paddsw " UNDERSCORE "kCoefficientsAbgrY+4096(,%ebx,8),%mm0\n"
-  "movzbl 0x1(%edx),%ebx                       \n"
-  "movq   " UNDERSCORE "kCoefficientsAbgrY(,%eax,8),%mm1\n"
-  "lea    2(%edx),%edx                         \n"
-  "movq   " UNDERSCORE "kCoefficientsAbgrY(,%ebx,8),%mm2\n"
-  "paddsw %mm0,%mm1                            \n"
-  "paddsw %mm0,%mm2                            \n"
-  "psraw  $0x6,%mm1                            \n"
-  "psraw  $0x6,%mm2                            \n"
-  "packuswb %mm2,%mm1                          \n"
-  "movq   %mm1,0x0(%ebp)                       \n"
-  "lea    8(%ebp),%ebp                         \n"
-  "sub    $0x2,%ecx                            \n"
-  "ja     1b                                   \n"
-  "popa                                        \n"
-  "ret                                         \n"
-);
-
-void FastConvertYUV444ToARGBRow_MMX(const uint8* y_buf,
-                                    const uint8* u_buf,
-                                    const uint8* v_buf,
-                                    uint8* rgb_buf,
-                                    int width);
-  asm(
-  ".text                                       \n"
-#if defined(OSX) || defined(IOS)
-  ".globl _FastConvertYUV444ToARGBRow_MMX      \n"
-"_FastConvertYUV444ToARGBRow_MMX:              \n"
-#else
-  ".global FastConvertYUV444ToARGBRow_MMX      \n"
-"FastConvertYUV444ToARGBRow_MMX:               \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    0x34(%esp),%ecx                      \n"
-
-"1:                                            \n"
-  "movzbl (%edi),%eax                          \n"
-  "lea    1(%edi),%edi                         \n"
-  "movzbl (%esi),%ebx                          \n"
-  "lea    1(%esi),%esi                         \n"
-  "movq   " UNDERSCORE "kCoefficientsRgbY+2048(,%eax,8),%mm0\n"
-  "movzbl (%edx),%eax                          \n"
-  "paddsw " UNDERSCORE "kCoefficientsRgbY+4096(,%ebx,8),%mm0\n"
-  "lea    1(%edx),%edx                         \n"
-  "paddsw " UNDERSCORE "kCoefficientsRgbY(,%eax,8),%mm0\n"
-  "psraw  $0x6,%mm0                            \n"
-  "packuswb %mm0,%mm0                          \n"
-  "movd   %mm0,0x0(%ebp)                       \n"
-  "lea    4(%ebp),%ebp                         \n"
-  "sub    $0x1,%ecx                            \n"
-  "ja     1b                                   \n"
-  "popa                                        \n"
-  "ret                                         \n"
-);
-
-void FastConvertYToARGBRow_MMX(const uint8* y_buf,
-                               uint8* rgb_buf,
-                               int width);
-  asm(
-  ".text                                       \n"
-#if defined(OSX) || defined(IOS)
-  ".globl _FastConvertYToARGBRow_MMX           \n"
-"_FastConvertYToARGBRow_MMX:                   \n"
-#else
-  ".global FastConvertYToARGBRow_MMX           \n"
-"FastConvertYToARGBRow_MMX:                    \n"
-#endif
-  "push   %ebx                                 \n"
-  "mov    0x8(%esp),%eax                       \n"
-  "mov    0xc(%esp),%edx                       \n"
-  "mov    0x10(%esp),%ecx                      \n"
-
-"1:                                            \n"
-  "movzbl (%eax),%ebx                          \n"
-  "movq   " UNDERSCORE "kCoefficientsRgbY(,%ebx,8),%mm0\n"
-  "psraw  $0x6,%mm0                            \n"
-  "movzbl 0x1(%eax),%ebx                       \n"
-  "movq   " UNDERSCORE "kCoefficientsRgbY(,%ebx,8),%mm1\n"
-  "psraw  $0x6,%mm1                            \n"
-  "packuswb %mm1,%mm0                          \n"
-  "lea    0x2(%eax),%eax                       \n"
-  "movq   %mm0,(%edx)                          \n"
-  "lea    0x8(%edx),%edx                       \n"
-  "sub    $0x2,%ecx                            \n"
-  "ja     1b                                   \n"
-  "pop    %ebx                                 \n"
-  "ret                                         \n"
-);
-
 #endif
 
 #ifdef HAS_ARGBTOYROW_SSSE3

source/row_win.cc

@@ -15,71 +15,71 @@ extern "C" {
 #ifdef HAS_ARGBTOYROW_SSSE3
 
 // Constant multiplication table for converting ARGB to I400.
-SIMD_ALIGNED(const int8 kARGBToY[16]) = {
+static const vec8 kARGBToY = {
   13, 65, 33, 0, 13, 65, 33, 0, 13, 65, 33, 0, 13, 65, 33, 0
 };
 
-SIMD_ALIGNED(const int8 kARGBToU[16]) = {
+static const vec8 kARGBToU = {
   112, -74, -38, 0, 112, -74, -38, 0, 112, -74, -38, 0, 112, -74, -38, 0
 };
 
-SIMD_ALIGNED(const int8 kARGBToV[16]) = {
+static const vec8 kARGBToV = {
   -18, -94, 112, 0, -18, -94, 112, 0, -18, -94, 112, 0, -18, -94, 112, 0,
 };
 
 // Constants for BGRA
-SIMD_ALIGNED(const int8 kBGRAToY[16]) = {
+static const vec8 kBGRAToY = {
   0, 33, 65, 13, 0, 33, 65, 13, 0, 33, 65, 13, 0, 33, 65, 13
 };
 
-SIMD_ALIGNED(const int8 kBGRAToU[16]) = {
+static const vec8 kBGRAToU = {
   0, -38, -74, 112, 0, -38, -74, 112, 0, -38, -74, 112, 0, -38, -74, 112
 };
 
-SIMD_ALIGNED(const int8 kBGRAToV[16]) = {
+static const vec8 kBGRAToV = {
   0, 112, -94, -18, 0, 112, -94, -18, 0, 112, -94, -18, 0, 112, -94, -18
 };
 
 // Constants for ABGR
-SIMD_ALIGNED(const int8 kABGRToY[16]) = {
+static const vec8 kABGRToY = {
   33, 65, 13, 0, 33, 65, 13, 0, 33, 65, 13, 0, 33, 65, 13, 0
 };
 
-SIMD_ALIGNED(const int8 kABGRToU[16]) = {
+static const vec8 kABGRToU = {
   -38, -74, 112, 0, -38, -74, 112, 0, -38, -74, 112, 0, -38, -74, 112, 0
 };
 
-SIMD_ALIGNED(const int8 kABGRToV[16]) = {
+static const vec8 kABGRToV = {
   112, -94, -18, 0, 112, -94, -18, 0, 112, -94, -18, 0, 112, -94, -18, 0
 };
 
-SIMD_ALIGNED(const uint8 kAddY16[16]) = {
+static const uvec8 kAddY16 = {
   16u, 16u, 16u, 16u, 16u, 16u, 16u, 16u,
   16u, 16u, 16u, 16u, 16u, 16u, 16u, 16u,
 };
 
-SIMD_ALIGNED(const uint8 kAddUV128[16]) = {
+static const uvec8 kAddUV128 = {
   128u, 128u, 128u, 128u, 128u, 128u, 128u, 128u,
   128u, 128u, 128u, 128u, 128u, 128u, 128u, 128u
 };
 
 // Shuffle table for converting BG24 to ARGB.
-SIMD_ALIGNED(const uint8 kShuffleMaskBG24ToARGB[16]) = {
+static const uvec8 kShuffleMaskBG24ToARGB = {
   0u, 1u, 2u, 12u, 3u, 4u, 5u, 13u, 6u, 7u, 8u, 14u, 9u, 10u, 11u, 15u
 };
 
 // Shuffle table for converting RAW to ARGB.
-SIMD_ALIGNED(const uint8 kShuffleMaskRAWToARGB[16]) = {
+static const uvec8 kShuffleMaskRAWToARGB = {
   2u, 1u, 0u, 12u, 5u, 4u, 3u, 13u, 8u, 7u, 6u, 14u, 11u, 10u, 9u, 15u
 };
 
 // Shuffle table for converting ABGR to ARGB.
-SIMD_ALIGNED(const uint8 kShuffleMaskABGRToARGB[16]) = {
+static const uvec8 kShuffleMaskABGRToARGB = {
   2u, 1u, 0u, 3u, 6u, 5u, 4u, 7u, 10u, 9u, 8u, 11u, 14u, 13u, 12u, 15u
 };
 
 // Shuffle table for converting BGRA to ARGB.
-SIMD_ALIGNED(const uint8 kShuffleMaskBGRAToARGB[16]) = {
+static const uvec8 kShuffleMaskBGRAToARGB = {
   3u, 2u, 1u, 0u, 7u, 6u, 5u, 4u, 11u, 10u, 9u, 8u, 15u, 14u, 13u, 12u
 };
 
@@ -118,7 +118,7 @@ __asm {
     mov       ecx, [esp + 12]  // pix
     movdqa    xmm5, kShuffleMaskABGRToARGB
 
- convertloop :
+ convertloop:
     movdqa    xmm0, [eax]
     lea       eax, [eax + 16]
     pshufb    xmm0, xmm5
@@ -138,7 +138,7 @@ __asm {
     mov       ecx, [esp + 12]  // pix
     movdqa    xmm5, kShuffleMaskBGRAToARGB
 
- convertloop :
+ convertloop:
     movdqa    xmm0, [eax]
     lea       eax, [eax + 16]
     pshufb    xmm0, xmm5
@@ -160,7 +160,7 @@ __asm {
     pslld     xmm5, 24
     movdqa    xmm4, kShuffleMaskBG24ToARGB
 
- convertloop :
+ convertloop:
     movdqa    xmm0, [eax]
     movdqa    xmm1, [eax + 16]
     movdqa    xmm3, [eax + 32]
@@ -199,7 +199,7 @@ __asm {
     pslld     xmm5, 24
     movdqa    xmm4, kShuffleMaskRAWToARGB
 
- convertloop :
+ convertloop:
     movdqa    xmm0, [eax]
     movdqa    xmm1, [eax + 16]
     movdqa    xmm3, [eax + 32]
@@ -237,7 +237,7 @@ __asm {
     movdqa     xmm5, kAddY16
     movdqa     xmm4, kARGBToY
 
- convertloop :
+ convertloop:
     movdqa     xmm0, [eax]
     movdqa     xmm1, [eax + 16]
     movdqa     xmm2, [eax + 32]
@@ -270,7 +270,7 @@ __asm {
     movdqa     xmm5, kAddY16
     movdqa     xmm4, kBGRAToY
 
- convertloop :
+ convertloop:
     movdqa     xmm0, [eax]
     movdqa     xmm1, [eax + 16]
     movdqa     xmm2, [eax + 32]
@@ -303,7 +303,7 @@ __asm {
     movdqa     xmm5, kAddY16
     movdqa     xmm4, kABGRToY
 
- convertloop :
+ convertloop:
     movdqa     xmm0, [eax]
     movdqa     xmm1, [eax + 16]
     movdqa     xmm2, [eax + 32]
@@ -343,7 +343,7 @@ __asm {
     movdqa     xmm5, kAddUV128
     sub        edi, edx             // stride from u to v
 
- convertloop :
+ convertloop:
     /* step 1 - subsample 16x2 argb pixels to 8x1 */
     movdqa     xmm0, [eax]
     movdqa     xmm1, [eax + 16]
@@ -407,7 +407,7 @@ __asm {
     movdqa     xmm5, kAddUV128
     sub        edi, edx             // stride from u to v
 
- convertloop :
+ convertloop:
     /* step 1 - subsample 16x2 argb pixels to 8x1 */
     movdqa     xmm0, [eax]
     movdqa     xmm1, [eax + 16]
@@ -471,7 +471,7 @@ __asm {
     movdqa     xmm5, kAddUV128
     sub        edi, edx             // stride from u to v
 
- convertloop :
+ convertloop:
     /* step 1 - subsample 16x2 argb pixels to 8x1 */
     movdqa     xmm0, [eax]
     movdqa     xmm1, [eax + 16]
@@ -519,182 +519,6 @@ __asm {
   }
 }
 
-#define YUVTORGB_MMX(TABLE) __asm {                                            \
-    __asm convertloop :                                                        \
-    __asm movzx     eax, byte ptr [edi]                                        \
-    __asm lea       edi, [edi + 1]                                             \
-    __asm movzx     ebx, byte ptr [esi]                                        \
-    __asm lea       esi, [esi + 1]                                             \
-    __asm movq      mm0, [TABLE + 2048 + 8 * eax]                              \
-    __asm movzx     eax, byte ptr [edx]                                        \
-    __asm paddsw    mm0, [TABLE + 4096 + 8 * ebx]                              \
-    __asm movzx     ebx, byte ptr [edx + 1]                                    \
-    __asm movq      mm1, [TABLE + 8 * eax]                                     \
-    __asm lea       edx, [edx + 2]                                             \
-    __asm movq      mm2, [TABLE + 8 * ebx]                                     \
-    __asm paddsw    mm1, mm0                                                   \
-    __asm paddsw    mm2, mm0                                                   \
-    __asm psraw     mm1, 6                                                     \
-    __asm psraw     mm2, 6                                                     \
-    __asm packuswb  mm1, mm2                                                   \
-    __asm movq      [ebp], mm1                                                 \
-    __asm lea       ebp, [ebp + 8]                                             \
-    __asm sub       ecx, 2                                                     \
-    __asm ja        convertloop                                                \
-  }
-
-__declspec(naked)
-void FastConvertYUVToARGBRow_MMX(const uint8* y_buf,
-                                 const uint8* u_buf,
-                                 const uint8* v_buf,
-                                 uint8* rgb_buf,
-                                 int width) {
-  __asm {
-    push      ebx
-    push      esi
-    push      edi
-    push      ebp
-    mov       edx, [esp + 16 + 4]
-    mov       edi, [esp + 16 + 8]
-    mov       esi, [esp + 16 + 12]
-    mov       ebp, [esp + 16 + 16]
-    mov       ecx, [esp + 16 + 20]
-
-    YUVTORGB_MMX(kCoefficientsRgbY)
-
-    pop       ebp
-    pop       edi
-    pop       esi
-    pop       ebx
-    ret
-  }
-}
-
-__declspec(naked)
-void FastConvertYUVToBGRARow_MMX(const uint8* y_buf,
-                                 const uint8* u_buf,
-                                 const uint8* v_buf,
-                                 uint8* rgb_buf,
-                                 int width) {
-  __asm {
-    push      ebx
-    push      esi
-    push      edi
-    push      ebp
-    mov       edx, [esp + 16 + 4]
-    mov       edi, [esp + 16 + 8]
-    mov       esi, [esp + 16 + 12]
-    mov       ebp, [esp + 16 + 16]
-    mov       ecx, [esp + 16 + 20]
-
-    YUVTORGB_MMX(kCoefficientsBgraY)
-
-    pop       ebp
-    pop       edi
-    pop       esi
-    pop       ebx
-    ret
-  }
-}
-
-__declspec(naked)
-void FastConvertYUVToABGRRow_MMX(const uint8* y_buf,
-                                 const uint8* u_buf,
-                                 const uint8* v_buf,
-                                 uint8* rgb_buf,
-                                 int width) {
-  __asm {
-    push      ebx
-    push      esi
-    push      edi
-    push      ebp
-    mov       edx, [esp + 16 + 4]
-    mov       edi, [esp + 16 + 8]
-    mov       esi, [esp + 16 + 12]
-    mov       ebp, [esp + 16 + 16]
-    mov       ecx, [esp + 16 + 20]
-
-    YUVTORGB_MMX(kCoefficientsAbgrY)
-
-    pop       ebp
-    pop       edi
-    pop       esi
-    pop       ebx
-    ret
-  }
-}
-
-__declspec(naked)
-void FastConvertYUV444ToARGBRow_MMX(const uint8* y_buf,
-                                    const uint8* u_buf,
-                                    const uint8* v_buf,
-                                    uint8* rgb_buf,
-                                    int width) {
-  __asm {
-    push      ebx
-    push      esi
-    push      edi
-    push      ebp
-    mov       edx, [esp + 16 + 4]
-    mov       edi, [esp + 16 + 8]
-    mov       esi, [esp + 16 + 12]
-    mov       ebp, [esp + 16 + 16]
-    mov       ecx, [esp + 16 + 20]
-
- convertloop :
-    movzx     eax, byte ptr [edi]
-    lea       edi, [edi + 1]
-    movzx     ebx, byte ptr [esi]
-    lea       esi, [esi + 1]
-    movq      mm0, [kCoefficientsRgbY + 2048 + 8 * eax]
-    movzx     eax, byte ptr [edx]
-    paddsw    mm0, [kCoefficientsRgbY + 4096 + 8 * ebx]
-    lea       edx, [edx + 1]
-    paddsw    mm0, [kCoefficientsRgbY + 8 * eax]
-    psraw     mm0, 6
-    packuswb  mm0, mm0
-    movd      [ebp], mm0
-    lea       ebp, [ebp + 4]
-    sub       ecx, 1
-    ja        convertloop
-
-    pop       ebp
-    pop       edi
-    pop       esi
-    pop       ebx
-    ret
-  }
-}
-
-__declspec(naked)
-void FastConvertYToARGBRow_MMX(const uint8* y_buf,
-                               uint8* rgb_buf,
-                               int width) {
-  __asm {
-    push      ebx
-    mov       eax, [esp + 4 + 4]   // Y
-    mov       edx, [esp + 4 + 8]   // rgb
-    mov       ecx, [esp + 4 + 12]  // width
-
- convertloop :
-    movzx     ebx, byte ptr [eax]
-    movq      mm0, [kCoefficientsRgbY + 8 * ebx]
-    psraw     mm0, 6
-    movzx     ebx, byte ptr [eax + 1]
-    movq      mm1, [kCoefficientsRgbY + 8 * ebx]
-    psraw     mm1, 6
-    packuswb  mm0, mm1
-    lea       eax, [eax + 2]
-    movq      [edx], mm0
-    lea       edx, [edx + 8]
-    sub       ecx, 2
-    ja        convertloop
-
-    pop       ebx
-    ret
-  }
-}
-
 #ifdef HAS_FASTCONVERTYUVTOARGBROW_SSSE3
 
 #define YG 74 /* static_cast<int8>(1.164 * 64 + 0.5) */
@@ -712,35 +536,35 @@ void FastConvertYToARGBRow_MMX(const uint8* y_buf,
 #define BG UG * 128 + VG * 128
 #define BR UR * 128 + VR * 128
 
-SIMD_ALIGNED(const int8 kUVToB[16]) = {
+static const vec8 kUVToB = {
   UB, VB, UB, VB, UB, VB, UB, VB, UB, VB, UB, VB, UB, VB, UB, VB
 };
 
-SIMD_ALIGNED(const int8 kUVToR[16]) = {
+static const vec8 kUVToR = {
   UR, VR, UR, VR, UR, VR, UR, VR, UR, VR, UR, VR, UR, VR, UR, VR
 };
 
-SIMD_ALIGNED(const int8 kUVToG[16]) = {
+static const vec8 kUVToG = {
   UG, VG, UG, VG, UG, VG, UG, VG, UG, VG, UG, VG, UG, VG, UG, VG
 };
 
-SIMD_ALIGNED(const int16 kYToRgb[8]) = {
+static const vec16 kYToRgb = {
   YG, YG, YG, YG, YG, YG, YG, YG
 };
 
-SIMD_ALIGNED(const int16 kYSub16[8]) = {
+static const vec16 kYSub16 = {
   16, 16, 16, 16, 16, 16, 16, 16
 };
 
-SIMD_ALIGNED(const int16 kUVBiasB[8]) = {
+static const vec16 kUVBiasB = {
   BB, BB, BB, BB, BB, BB, BB, BB
 };
 
-SIMD_ALIGNED(const int16 kUVBiasG[8]) = {
+static const vec16 kUVBiasG = {
   BG, BG, BG, BG, BG, BG, BG, BG
 };
 
-SIMD_ALIGNED(const int16 kUVBiasR[8]) = {
+static const vec16 kUVBiasR = {
   BR, BR, BR, BR, BR, BR, BR, BR
 };
 
@@ -794,7 +618,7 @@ void FastConvertYUVToARGBRow_SSSE3(const uint8* y_buf,
     pcmpeqb    xmm5, xmm5           // generate 0xffffffff for alpha
     pxor       xmm4, xmm4
 
- convertloop :
+ convertloop:
     YUVTORGB_SSSE3
 
     // Step 3: Weave into ARGB
@@ -833,7 +657,7 @@ void FastConvertYUVToBGRARow_SSSE3(const uint8* y_buf,
     sub        edi, esi
     pxor       xmm4, xmm4
 
- convertloop :
+ convertloop:
     YUVTORGB_SSSE3
 
     // Step 3: Weave into BGRA
@@ -874,7 +698,7 @@ void FastConvertYUVToABGRRow_SSSE3(const uint8* y_buf,
     pcmpeqb    xmm5, xmm5           // generate 0xffffffff for alpha
     pxor       xmm4, xmm4
 
- convertloop :
+ convertloop:
     YUVTORGB_SSSE3
 
     // Step 3: Weave into ARGB
@@ -914,7 +738,7 @@ void FastConvertYUV444ToARGBRow_SSSE3(const uint8* y_buf,
     pcmpeqb    xmm5, xmm5            // generate 0xffffffff for alpha
     pxor       xmm4, xmm4
 
- convertloop :
+ convertloop:
     // Step 1: Find 4 UV contributions to 4 R,G,B values
     movd       xmm0, [esi]          // U
     movd       xmm1, [esi + edi]    // V
@@ -978,7 +802,7 @@ void FastConvertYToARGBRow_SSE2(const uint8* y_buf,
     movdqa     xmm3, kYSub16
     movdqa     xmm2, kYToRgb
 
- convertloop :
+ convertloop:
     // Step 1: Scale Y contribution to 8 G values. G = (y - 16) * 1.164
     movq       xmm0, qword ptr [eax]
     lea        eax, [eax + 8]

source/scale.cc

@@ -14,6 +14,7 @@
 #include <string.h>
 
 #include "libyuv/cpu_id.h"
+#include "row.h"
 
 #if defined(_MSC_VER)
 #define ALIGN16(var) __declspec(align(16)) var
@@ -21,6 +22,7 @@
 #define ALIGN16(var) var __attribute__((aligned(16)))
 #endif
 
+
 // Note: A Neon reference manual
 // http://infocenter.arm.com/help/index.jsp?topic=/com.arm.doc.dui0204j/CJAJIIGG.html
 // Note: Some SSE2 reference manuals