Port ARGBToJ420 to AVX2

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 1005 matching lines @@
     "m"(kAddUV128),  // %5
     "m"(kARGBToV),   // %6
     "m"(kARGBToU),   // %7
     "m"(kShufARGBToUV_AVX)  // %8
   : "memory", "cc", NACL_R14
     "xmm0", "xmm1", "xmm2", "xmm3", "xmm4", "xmm5", "xmm6", "xmm7"
   );
 }
 #endif  // HAS_ARGBTOUVROW_AVX2
 
+#ifdef HAS_ARGBTOUVJROW_AVX2
+void ARGBToUVJRow_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"
+    "vpaddw     %%ymm5,%%ymm0,%%ymm0           \n"
+    "vpaddw     %%ymm5,%%ymm1,%%ymm1           \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"
+
+    "vextractf128 $0x0,%%ymm0," MEMACCESS(1) " \n"
+    VEXTOPMEM(vextractf128,1,ymm0,0x0,1,2,1) // vextractf128 $1,%%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"(kAddUVJ128),  // %5
+    "m"(kARGBToVJ),  // %6
+    "m"(kARGBToUJ),  // %7
+    "m"(kShufARGBToUV_AVX)  // %8
+  : "memory", "cc", NACL_R14
+    "xmm0", "xmm1", "xmm2", "xmm3", "xmm4", "xmm5", "xmm6", "xmm7"
+  );
+}
+#endif  // HAS_ARGBTOUVJROW_AVX2
+
 #ifdef HAS_ARGBTOUVJROW_SSSE3
 void ARGBToUVJRow_SSSE3(const uint8* src_argb0, int src_stride_argb,
                         uint8* dst_u, uint8* dst_v, int width) {
   asm volatile (
     "movdqa    %5,%%xmm3                       \n"
     "movdqa    %6,%%xmm4                       \n"
     "movdqa    %7,%%xmm5                       \n"
     "sub       %1,%2                           \n"
     LABELALIGN
   "1:                                          \n"
@@ skipping 432 matching lines @@
 // Read 2 UV from 411, upsample to 8 UV.
 // reading 4 bytes is an msan violation.
 //    "movd       " MEMACCESS([u_buf]) ",%%xmm0                   \n"
 //    MEMOPREG(movd, 0x00, [u_buf], [v_buf], 1, xmm1)
 // pinsrw fails with drmemory
 //  __asm pinsrw     xmm0, [esi], 0        /* U */
 //  __asm pinsrw     xmm1, [esi + edi], 0  /* V */
 #define READYUV411_TEMP                                                        \
     "movzwl     " MEMACCESS([u_buf]) ",%[temp]                  \n"            \
     "movd       %[temp],%%xmm0                                  \n"            \
-    MEMOPARG(movzwl,0x00,[u_buf],[v_buf],1,[temp]) "            \n"            \
+    MEMOPARG(movzwl, 0x00, [u_buf], [v_buf], 1, [temp]) "       \n"            \
     "movd       %[temp],%%xmm1                                  \n"            \
     "lea        " MEMLEA(0x2, [u_buf]) ",%[u_buf]               \n"            \
     "punpcklbw  %%xmm1,%%xmm0                                   \n"            \
     "punpcklwd  %%xmm0,%%xmm0                                   \n"            \
     "punpckldq  %%xmm0,%%xmm0                                   \n"            \
     "movq       " MEMACCESS([y_buf]) ",%%xmm4                   \n"            \
     "punpcklbw  %%xmm4,%%xmm4                                   \n"            \
     "lea        " MEMLEA(0x8, [y_buf]) ",%[y_buf]               \n"
 
 // Read 4 UV from NV12, upsample to 8 UV
@@ skipping 536 matching lines @@
     "vpaddsw     %%ymm4,%%ymm1,%%ymm1                               \n"        \
     "vpaddsw     %%ymm4,%%ymm2,%%ymm2                               \n"        \
     "vpsraw      $0x6,%%ymm0,%%ymm0                                 \n"        \
     "vpsraw      $0x6,%%ymm1,%%ymm1                                 \n"        \
     "vpsraw      $0x6,%%ymm2,%%ymm2                                 \n"        \
     "vpackuswb   %%ymm0,%%ymm0,%%ymm0                               \n"        \
     "vpackuswb   %%ymm1,%%ymm1,%%ymm1                               \n"        \
     "vpackuswb   %%ymm2,%%ymm2,%%ymm2                               \n"
 #define YUVTORGB_REGS_AVX2 \
     "xmm8", "xmm9", "xmm10", "xmm11", "xmm12", "xmm13", "xmm14",
-#else// Convert 16 pixels: 16 UV and 16 Y.
+#else  // Convert 16 pixels: 16 UV and 16 Y.
 #define YUVTORGB_SETUP_AVX2(yuvconstants)
 #define YUVTORGB_AVX2(yuvconstants)                                            \
     "vpmaddubsw  " MEMACCESS2(64, [yuvconstants]) ",%%ymm0,%%ymm2   \n"        \
     "vpmaddubsw  " MEMACCESS2(32, [yuvconstants]) ",%%ymm0,%%ymm1   \n"        \
     "vpmaddubsw  " MEMACCESS([yuvconstants]) ",%%ymm0,%%ymm0        \n"        \
     "vmovdqu     " MEMACCESS2(160, [yuvconstants]) ",%%ymm3         \n"        \
     "vpsubw      %%ymm2,%%ymm3,%%ymm2                               \n"        \
     "vmovdqu     " MEMACCESS2(128, [yuvconstants]) ",%%ymm3         \n"        \
     "vpsubw      %%ymm1,%%ymm3,%%ymm1                               \n"        \
     "vmovdqu     " MEMACCESS2(96, [yuvconstants]) ",%%ymm3          \n"        \
@@ skipping 3391 matching lines @@
   );
 }
 #endif  // HAS_ARGBLUMACOLORTABLEROW_SSSE3
 
 #endif  // defined(__x86_64__) || defined(__i386__)
 
 #ifdef __cplusplus
 }  // extern "C"
 }  // namespace libyuv
 #endif