Port ARGBToUV to AVX2.

BUG=269
TESTED=ncval
R=brucedawson@google.com, tpsiaki@google.com

Review URL: https://webrtc-codereview.appspot.com/35449004

git-svn-id: http://libyuv.googlecode.com/svn/trunk@1194 16f28f9a-4ce2-e073-06de-1de4eb20be90

README.chromium

 Name: libyuv
 URL: http://code.google.com/p/libyuv/
-Version: 1193
+Version: 1194
 License: BSD
 License File: LICENSE
 

include/libyuv/row.h

@@ -189,6 +189,7 @@ extern "C" {
 #define HAS_ARGBMIRRORROW_AVX2
 #define HAS_ARGBPOLYNOMIALROW_AVX2
 #define HAS_ARGBSHUFFLEROW_AVX2
+#define HAS_ARGBTOUVROW_AVX2
 #define HAS_ARGBTOYJROW_AVX2
 #define HAS_ARGBTOYROW_AVX2
 #define HAS_COPYROW_AVX
@@ -215,12 +216,6 @@ extern "C" {
 #define HAS_ARGBUNATTENUATEROW_AVX2
 #endif
 
-// The following are require VS2012.
-// TODO(fbarchard): Port to gcc.
-#if !defined(LIBYUV_DISABLE_X86) && defined(VISUALC_HAS_AVX2)
-#define HAS_ARGBTOUVROW_AVX2
-#endif  // defined(VISUALC_HAS_AVX2)
-
 // The following are Yasm x86 only:
 // TODO(fbarchard): Port AVX2 to inline.
 #if !defined(LIBYUV_DISABLE_X86) && defined(HAVE_YASM)

include/libyuv/version.h

@@ -11,6 +11,6 @@
 #ifndef INCLUDE_LIBYUV_VERSION_H_  // NOLINT
 #define INCLUDE_LIBYUV_VERSION_H_
 
-#define LIBYUV_VERSION 1193
+#define LIBYUV_VERSION 1194
 
 #endif  // INCLUDE_LIBYUV_VERSION_H_  NOLINT

source/row_posix.cc

@@ -953,6 +953,74 @@ void ARGBToUVRow_SSSE3(const uint8* src_argb0, int src_stride_argb,
 }
 #endif  // HAS_ARGBTOUVROW_SSSE3
 
+#ifdef HAS_ARGBTOUVROW_AVX2
+// vpshufb for vphaddw + vpackuswb packed to shorts.
+static const lvec8 kShufARGBToUV_AVX = {
+  0, 1, 8, 9, 2, 3, 10, 11, 4, 5, 12, 13, 6, 7, 14, 15,
+  0, 1, 8, 9, 2, 3, 10, 11, 4, 5, 12, 13, 6, 7, 14, 15
+};
+void ARGBToUVRow_AVX2(const uint8* src_argb0, int src_stride_argb,
+                      uint8* dst_u, uint8* dst_v, int width) {
+  asm volatile (
+    "vbroadcastf128 %5,%%ymm5                  \n"
+    "vbroadcastf128 %6,%%ymm6                  \n"
+    "vbroadcastf128 %7,%%ymm7                  \n"
+    "sub       %1,%2                           \n"
+    LABELALIGN
+  "1:                                          \n"
+    "vmovdqu    " MEMACCESS(0) ",%%ymm0        \n"
+    "vmovdqu    " MEMACCESS2(0x20,0) ",%%ymm1  \n"
+    "vmovdqu    " MEMACCESS2(0x40,0) ",%%ymm2  \n"
+    "vmovdqu    " MEMACCESS2(0x60,0) ",%%ymm3  \n"
+    VMEMOPREG(vpavgb,0x00,0,4,1,ymm0,ymm0)     // vpavgb (%0,%4,1),%%ymm0,%%ymm0
+    VMEMOPREG(vpavgb,0x20,0,4,1,ymm1,ymm1)
+    VMEMOPREG(vpavgb,0x40,0,4,1,ymm2,ymm2)
+    VMEMOPREG(vpavgb,0x60,0,4,1,ymm3,ymm3)
+    "lea       " MEMLEA(0x80,0) ",%0           \n"
+    "vshufps    $0x88,%%ymm1,%%ymm0,%%ymm4     \n"
+    "vshufps    $0xdd,%%ymm1,%%ymm0,%%ymm0     \n"
+    "vpavgb     %%ymm4,%%ymm0,%%ymm0           \n"
+    "vshufps    $0x88,%%ymm3,%%ymm2,%%ymm4     \n"
+    "vshufps    $0xdd,%%ymm3,%%ymm2,%%ymm2     \n"
+    "vpavgb     %%ymm4,%%ymm2,%%ymm2           \n"
+    "vpmaddubsw %%ymm7,%%ymm0,%%ymm1           \n"
+    "vpmaddubsw %%ymm7,%%ymm2,%%ymm3           \n"
+    "vpmaddubsw %%ymm6,%%ymm0,%%ymm0           \n"
+    "vpmaddubsw %%ymm6,%%ymm2,%%ymm2           \n"
+    "vphaddw    %%ymm3,%%ymm1,%%ymm1           \n"
+    "vphaddw    %%ymm2,%%ymm0,%%ymm0           \n"
+    "vpsraw     $0x8,%%ymm1,%%ymm1             \n"
+    "vpsraw     $0x8,%%ymm0,%%ymm0             \n"
+    "vpacksswb  %%ymm0,%%ymm1,%%ymm0           \n"
+    "vpermq     $0xd8,%%ymm0,%%ymm0            \n"
+    "vpshufb    %8,%%ymm0,%%ymm0               \n"
+    "vpaddb     %%ymm5,%%ymm0,%%ymm0           \n"
+    "vextractf128 $0x0,%%ymm0," MEMACCESS(1) " \n"
+    VEXTOPMEM(vextractf128,ymm0,0x1,1,2,1) // vextractf128 $0x1,%%ymm0,(%1,%2,1)
+    "lea       " MEMLEA(0x10,1) ",%1           \n"
+    "sub       $0x20,%3                        \n"
+    "jg        1b                              \n"
+    "vzeroupper                                \n"
+  : "+r"(src_argb0),       // %0
+    "+r"(dst_u),           // %1
+    "+r"(dst_v),           // %2
+    "+rm"(width)           // %3
+  : "r"((intptr_t)(src_stride_argb)), // %4
+    "m"(kARGBToV),   // %5
+    "m"(kARGBToU),   // %6
+    "m"(kAddUV128),  // %7
+    "m"(kShufARGBToUV_AVX)  // %8
+  : "memory", "cc"
+#if defined(__native_client__) && defined(__x86_64__)
+    , "r14"
+#endif
+#if defined(__SSE2__)
+    , "xmm0", "xmm1", "xmm2", "xmm3", "xmm4", "xmm5", "xmm6", "xmm7"
+#endif
+  );
+}
+#endif  // HAS_ARGBTOUVROW_AVX2
+
 #ifdef HAS_ARGBTOUVJROW_SSSE3
 // TODO(fbarchard): Share code with ARGBToUVRow_SSSE3.
 void ARGBToUVJRow_SSSE3(const uint8* src_argb0, int src_stride_argb,