read in read function for yuv conversion

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 1306 matching lines @@
   );
 }
 
 #if defined(HAS_I422TOARGBROW_SSSE3) || defined(HAS_I422TOARGBROW_AVX2)
 
 // Read 8 UV from 411
 #define READYUV444                                                             \
     "movq       " MEMACCESS([u_buf]) ",%%xmm0                   \n"            \
     MEMOPREG(movq, 0x00, [u_buf], [v_buf], 1, xmm1)                            \
     "lea        " MEMLEA(0x8, [u_buf]) ",%[u_buf]               \n"            \
-    "punpcklbw  %%xmm1,%%xmm0                                   \n"
+    "punpcklbw  %%xmm1,%%xmm0                                   \n"            \
+    "movq       " MEMACCESS([y_buf]) ",%%xmm4                   \n"            \
+    "lea        " MEMLEA(0x8, [y_buf]) ",%[y_buf]               \n"
 
 // Read 4 UV from 422, upsample to 8 UV
 #define READYUV422                                                             \
     "movd       " MEMACCESS([u_buf]) ",%%xmm0                   \n"            \
     MEMOPREG(movd, 0x00, [u_buf], [v_buf], 1, xmm1)                            \
     "lea        " MEMLEA(0x4, [u_buf]) ",%[u_buf]               \n"            \
     "punpcklbw  %%xmm1,%%xmm0                                   \n"            \
-    "punpcklwd  %%xmm0,%%xmm0                                   \n"
+    "punpcklwd  %%xmm0,%%xmm0                                   \n"            \
+    "movq       " MEMACCESS([y_buf]) ",%%xmm4                   \n"            \
+    "lea        " MEMLEA(0x8, [y_buf]) ",%[y_buf]               \n"
 
 // Read 2 UV from 411, upsample to 8 UV
 #define READYUV411                                                             \
     "movd       " MEMACCESS([u_buf]) ",%%xmm0                   \n"            \
     MEMOPREG(movd, 0x00, [u_buf], [v_buf], 1, xmm1)                            \
     "lea        " MEMLEA(0x2, [u_buf]) ",%[u_buf]               \n"            \
     "punpcklbw  %%xmm1,%%xmm0                                   \n"            \
     "punpcklwd  %%xmm0,%%xmm0                                   \n"            \
-    "punpckldq  %%xmm0,%%xmm0                                   \n"
+    "punpckldq  %%xmm0,%%xmm0                                   \n"            \
+    "movq       " MEMACCESS([y_buf]) ",%%xmm4                   \n"            \
+    "lea        " MEMLEA(0x8, [y_buf]) ",%[y_buf]               \n"
 
 // Read 4 UV from NV12, upsample to 8 UV
 #define READNV12                                                               \
     "movq       " MEMACCESS([uv_buf]) ",%%xmm0                  \n"            \
     "lea        " MEMLEA(0x8, [uv_buf]) ",%[uv_buf]             \n"            \
-    "punpcklwd  %%xmm0,%%xmm0                                   \n"
+    "punpcklwd  %%xmm0,%%xmm0                                   \n"            \
+    "movq       " MEMACCESS([y_buf]) ",%%xmm4                   \n"            \
+    "lea        " MEMLEA(0x8, [y_buf]) ",%[y_buf]               \n"
 
 // Convert 8 pixels: 8 UV and 8 Y
 #define YUVTORGB(yuvconstants)                                                 \
     "movdqa     %%xmm0,%%xmm1                                   \n"            \
     "movdqa     %%xmm0,%%xmm2                                   \n"            \
     "movdqa     %%xmm0,%%xmm3                                   \n"            \
     "movdqa     " MEMACCESS2(96, [yuvconstants]) ",%%xmm0       \n"            \
     "pmaddubsw  " MEMACCESS([yuvconstants]) ",%%xmm1            \n"            \
     "psubw      %%xmm1,%%xmm0                                   \n"            \
     "movdqa     " MEMACCESS2(128, [yuvconstants]) ",%%xmm1      \n"            \
     "pmaddubsw  " MEMACCESS2(32, [yuvconstants]) ",%%xmm2       \n"            \
     "psubw      %%xmm2,%%xmm1                                   \n"            \
     "movdqa     " MEMACCESS2(160, [yuvconstants]) ",%%xmm2      \n"            \
     "pmaddubsw  " MEMACCESS2(64, [yuvconstants]) ",%%xmm3       \n"            \
     "psubw      %%xmm3,%%xmm2                                   \n"            \
-    "movq       " MEMACCESS([y_buf]) ",%%xmm3                   \n"            \
-    "lea        " MEMLEA(0x8, [y_buf]) ",%[y_buf]               \n"            \
-    "punpcklbw  %%xmm3,%%xmm3                                   \n"            \
-    "pmulhuw    " MEMACCESS2(192, [yuvconstants]) ",%%xmm3      \n"            \
-    "paddsw     %%xmm3,%%xmm0                                   \n"            \
-    "paddsw     %%xmm3,%%xmm1                                   \n"            \
-    "paddsw     %%xmm3,%%xmm2                                   \n"            \
+    "punpcklbw  %%xmm4,%%xmm4                                   \n"            \
+    "pmulhuw    " MEMACCESS2(192, [yuvconstants]) ",%%xmm4      \n"            \
+    "paddsw     %%xmm4,%%xmm0                                   \n"            \
+    "paddsw     %%xmm4,%%xmm1                                   \n"            \
+    "paddsw     %%xmm4,%%xmm2                                   \n"            \
     "psraw      $0x6,%%xmm0                                     \n"            \
     "psraw      $0x6,%%xmm1                                     \n"            \
     "psraw      $0x6,%%xmm2                                     \n"            \
     "packuswb   %%xmm0,%%xmm0                                   \n"            \
     "packuswb   %%xmm1,%%xmm1                                   \n"            \
     "packuswb   %%xmm2,%%xmm2                                   \n"
 
 // Store 8 ARGB values. Assumes XMM5 is zero.
 #define STOREARGB                                                              \
     "punpcklbw  %%xmm1,%%xmm0                                    \n"           \
@@ skipping 3818 matching lines @@
   );
 }
 #endif  // HAS_ARGBLUMACOLORTABLEROW_SSSE3
 
 #endif  // defined(__x86_64__) || defined(__i386__)
 
 #ifdef __cplusplus
 }  // extern "C"
 }  // namespace libyuv
 #endif