vextractf128 requuires a constant argument for which dqword to extract, so add a new macro.

BUG=none
TESTED=local build on clang for osx
R=harryjin@google.com

Review URL: https://webrtc-codereview.appspot.com/30869004

git-svn-id: http://libyuv.googlecode.com/svn/trunk@1153 16f28f9a-4ce2-e073-06de-1de4eb20be90

include/libyuv/row.h

@@ -536,6 +536,11 @@ typedef uint8 ulvec8[32];
     "lea " #offset "(%q" #base ",%q" #index "," #scale "),%%r14d\n" \
     #opcode " (%%r15,%%r14),%%" #reg1 ",%%" #reg2 "\n" \
     BUNDLEUNLOCK
+#define VEXTOPMEM(opcode, reg, offset, base, index, scale) \
+    BUNDLELOCK \
+    "lea " #offset "(%q" #base ",%q" #index "," #scale "),%%r14d\n" \
+    #opcode " $0x0,%%" #reg ",(%%r15,%%r14)\n" \
+    BUNDLEUNLOCK
 #else  // defined(__native_client__) && defined(__x86_64__)
 #define BUNDLEALIGN "\n"
 #define MEMACCESS(base) "(%" #base ")"
@@ -556,6 +561,8 @@ typedef uint8 ulvec8[32];
 #define VMEMOPREG(opcode, offset, base, index, scale, reg1, reg2) \
     #opcode " " #offset "(%" #base ",%" #index "," #scale "),%%" #reg1 ",%%" \
     #reg2 "\n"
+#define VEXTOPMEM(opcode, reg, offset, base, index, scale) \
+    #opcode " $0x0,%%" #reg ","#offset "(%" #base ",%" #index "," #scale ")\n"
 #endif  // defined(__native_client__) && defined(__x86_64__)
 
 #if defined(__arm__) || defined(__aarch64__)

source/row_any.cc

@@ -358,8 +358,6 @@ YANY(ARGBAttenuateRow_Any_NEON, ARGBAttenuateRow_NEON, ARGBAttenuateRow_C,
 
 #ifdef HAS_ARGBTOUVROW_AVX2
 UVANY(ARGBToUVRow_Any_AVX2, ARGBToUVRow_AVX2, ARGBToUVRow_C, 4, 31)
-UVANY(YUY2ToUVRow_Any_AVX2, YUY2ToUVRow_AVX2, YUY2ToUVRow_C, 2, 31)
-UVANY(UYVYToUVRow_Any_AVX2, UYVYToUVRow_AVX2, UYVYToUVRow_C, 2, 31)
 #endif
 #ifdef HAS_ARGBTOUVROW_SSSE3
 UVANY(ARGBToUVRow_Any_SSSE3, ARGBToUVRow_SSSE3, ARGBToUVRow_C, 4, 15)
@@ -368,6 +366,10 @@ UVANY(BGRAToUVRow_Any_SSSE3, BGRAToUVRow_SSSE3, BGRAToUVRow_C, 4, 15)
 UVANY(ABGRToUVRow_Any_SSSE3, ABGRToUVRow_SSSE3, ABGRToUVRow_C, 4, 15)
 UVANY(RGBAToUVRow_Any_SSSE3, RGBAToUVRow_SSSE3, RGBAToUVRow_C, 4, 15)
 #endif
+#ifdef HAS_YUY2TOUVROW_AVX2
+UVANY(YUY2ToUVRow_Any_AVX2, YUY2ToUVRow_AVX2, YUY2ToUVRow_C, 2, 31)
+UVANY(UYVYToUVRow_Any_AVX2, UYVYToUVRow_AVX2, UYVYToUVRow_C, 2, 31)
+#endif
 #ifdef HAS_YUY2TOUVROW_SSE2
 UVANY(YUY2ToUVRow_Any_SSE2, YUY2ToUVRow_SSE2, YUY2ToUVRow_C, 2, 15)
 UVANY(UYVYToUVRow_Any_SSE2, UYVYToUVRow_SSE2, UYVYToUVRow_C, 2, 15)

source/row_posix.cc

@@ -2907,7 +2907,7 @@ void YUY2ToUVRow_AVX2(const uint8* src_yuy2, int stride_yuy2,
     "vpermq    $0xd8,%%ymm1,%%ymm1             \n"
     "vpermq    $0xd8,%%ymm0,%%ymm0             \n"
     "vextractf128 $0x0,%%ymm1," MEMACCESS(1) " \n"
-    MEMOPMEM(vextractf128,ymm0,0x00,1,2,1) // vextractf128 $0x0,%%ymm0,(%1,%2,1)
+    VEXTOPMEM(vextractf128,ymm0,0x00,1,2,1) // vextractf128 $0x0,%%ymm0,(%1,%2,1)
     "lea       0x10" MEMACCESS(1) ",%1         \n"
     "sub       $0x20,%3                        \n"
     "jg        1b                              \n"
@@ -2949,7 +2949,7 @@ void YUY2ToUV422Row_AVX2(const uint8* src_yuy2,
     "vpermq    $0xd8,%%ymm1,%%ymm1             \n"
     "vpermq    $0xd8,%%ymm0,%%ymm0             \n"
     "vextractf128 $0x0,%%ymm1," MEMACCESS(1) " \n"
-    MEMOPMEM(vextractf128,ymm0,0x00,1,2,1) // vextractf128 $0x0,%%ymm0,(%1,%2,1)
+    VEXTOPMEM(vextractf128,ymm0,0x00,1,2,1) // vextractf128 $0x0,%%ymm0,(%1,%2,1)
     "lea       0x10" MEMACCESS(1) ",%1         \n"
     "sub       $0x20,%3                        \n"
     "jg        1b                              \n"
@@ -3020,7 +3020,7 @@ void UYVYToUVRow_AVX2(const uint8* src_uyvy, int stride_uyvy,
     "vpermq    $0xd8,%%ymm1,%%ymm1             \n"
     "vpermq    $0xd8,%%ymm0,%%ymm0             \n"
     "vextractf128 $0x0,%%ymm1," MEMACCESS(1) " \n"
-    MEMOPMEM(vextractf128,ymm0,0x00,1,2,1) // vextractf128 $0x0,%%ymm0,(%1,%2,1)
+    VEXTOPMEM(vextractf128,ymm0,0x00,1,2,1) // vextractf128 $0x0,%%ymm0,(%1,%2,1)
     "lea       0x10" MEMACCESS(1) ",%1         \n"
     "sub       $0x20,%3                        \n"
     "jg        1b                              \n"
@@ -3062,7 +3062,7 @@ void UYVYToUV422Row_AVX2(const uint8* src_uyvy,
     "vpermq    $0xd8,%%ymm1,%%ymm1             \n"
     "vpermq    $0xd8,%%ymm0,%%ymm0             \n"
     "vextractf128 $0x0,%%ymm1," MEMACCESS(1) " \n"
-    MEMOPMEM(vextractf128,ymm0,0x00,1,2,1) // vextractf128 $0x0,%%ymm0,(%1,%2,1)
+    VEXTOPMEM(vextractf128,ymm0,0x00,1,2,1) // vextractf128 $0x0,%%ymm0,(%1,%2,1)
     "lea       0x10" MEMACCESS(1) ",%1         \n"
     "sub       $0x20,%3                        \n"
     "jg        1b                              \n"