Optimize yuv alpha blend AVX2 code to do 32 pixels at time.

source/row_gcc.cc

 // VERSION 2
 /*
  *  Copyright 2011 The LibYuv Project Authors. All rights reserved.
  *
  *  Use of this source code is governed by a BSD-style license
  *  that can be found in the LICENSE file in the root of the source
  *  tree. An additional intellectual property rights grant can be found
  *  in the file PATENTS. All contributing project authors may
  *  be found in the AUTHORS file in the root of the source tree.
  */
@@ skipping 3513 matching lines @@
     "mov        $0x80808080,%%eax              \n"
     "vmovd      %%eax,%%xmm6                   \n"
     "vbroadcastss %%xmm6,%%ymm6                \n"
     "mov        $0x807f807f,%%eax              \n"
     "vmovd      %%eax,%%xmm7                   \n"
     "vbroadcastss %%xmm7,%%ymm7                \n"
     "sub        %2,%0                          \n"
     "sub        %2,%1                          \n"
     "sub        %2,%3                          \n"
 
-    // 16 pixel loop.
+    // 32 pixel loop.
     LABELALIGN
   "1:                                          \n"
-    "vmovdqu    (%2),%%xmm0                    \n"
-    "vpermq     $0xd8,%%ymm0,%%ymm0            \n"
+    "vmovdqu    (%2),%%ymm0                    \n"
+    "vpunpckhbw %%ymm0,%%ymm0,%%ymm3           \n"
     "vpunpcklbw %%ymm0,%%ymm0,%%ymm0           \n"
+    "vpxor      %%ymm5,%%ymm3,%%ymm3           \n"
     "vpxor      %%ymm5,%%ymm0,%%ymm0           \n"
-    "vmovdqu    (%0,%2,1),%%xmm1               \n"
-    "vmovdqu    (%1,%2,1),%%xmm2               \n"
-    "vpermq     $0xd8,%%ymm1,%%ymm1            \n"
-    "vpermq     $0xd8,%%ymm2,%%ymm2            \n"
+    "vmovdqu    (%0,%2,1),%%ymm1               \n"
+    "vmovdqu    (%1,%2,1),%%ymm2               \n"
+    "vpunpckhbw %%ymm2,%%ymm1,%%ymm4           \n"
     "vpunpcklbw %%ymm2,%%ymm1,%%ymm1           \n"
+    "vpsubb     %%ymm6,%%ymm1,%%ymm4           \n"
     "vpsubb     %%ymm6,%%ymm1,%%ymm1           \n"
+    "vpmaddubsw %%ymm4,%%ymm3,%%ymm3           \n"
     "vpmaddubsw %%ymm1,%%ymm0,%%ymm0           \n"
+    "vpaddw     %%ymm7,%%ymm3,%%ymm3           \n"
     "vpaddw     %%ymm7,%%ymm0,%%ymm0           \n"
+    "vpsrlw     $0x8,%%ymm3,%%ymm3             \n"
     "vpsrlw     $0x8,%%ymm0,%%ymm0             \n"
-    "vpackuswb  %%ymm0,%%ymm0,%%ymm0           \n"
-    "vpermq     $0xd8,%%ymm0,%%ymm0            \n"
-    "vmovdqu    %%xmm0,(%3,%2,1)               \n"
-    "lea        0x10(%2),%2                    \n"
-    "sub        $0x10,%4                       \n"
+    "vpackuswb  %%ymm3,%%ymm0,%%ymm0           \n"
+    "vmovdqu    %%ymm0,(%3,%2,1)               \n"
+    "lea        0x20(%2),%2                    \n"
+    "sub        $0x20,%4                       \n"
     "jg        1b                              \n"
     "vzeroupper                                \n"
   : "+r"(src0),       // %0
     "+r"(src1),       // %1
     "+r"(alpha),      // %2
     "+r"(dst),        // %3
     "+r"(width)       // %4
-  :: "memory", "cc", "eax", "xmm0", "xmm1", "xmm2", "xmm5", "xmm6", "xmm7"
+  :: "memory", "cc", "eax",
+     "xmm0", "xmm1", "xmm2", "xmm3", "xmm4", "xmm5", "xmm6", "xmm7"
   );
 }
 #endif  // HAS_BLENDPLANEROW_AVX2
 
 #ifdef HAS_ARGBATTENUATEROW_SSSE3
 // Shuffle table duplicating alpha
 static uvec8 kShuffleAlpha0 = {
   3u, 3u, 3u, 3u, 3u, 3u, 128u, 128u, 7u, 7u, 7u, 7u, 7u, 7u, 128u, 128u
 };
 static uvec8 kShuffleAlpha1 = {
@@ skipping 2014 matching lines @@
   );
 }
 #endif  // HAS_ARGBLUMACOLORTABLEROW_SSSE3
 
 #endif  // defined(__x86_64__) || defined(__i386__)
 
 #ifdef __cplusplus
 }  // extern "C"
 }  // namespace libyuv
 #endif