port I411 movzx 2 byte reader to gcc

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 1500 matching lines @@
     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"            \
     "movq       " MEMACCESS([y_buf]) ",%%xmm4                   \n"            \
     "punpcklbw  %%xmm4,%%xmm4                                   \n"            \
     "lea        " MEMLEA(0x8, [y_buf]) ",%[y_buf]               \n"            \
     "movq       " MEMACCESS([a_buf]) ",%%xmm5                   \n"            \
     "lea        " MEMLEA(0x8, [a_buf]) ",%[a_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)                            \
+// 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_EBX                                                         \
+    "movzw      " MEMACCESS([u_buf]) ",%%ebx                    \n"            \
+    "movd       %%ebx,%%xmm0                                    \n"            \
+    MEMOPREG(movzw,0x00,[u_buf],[v_buf],1,ebx) "                \n"            \
+    "movd       %%ebx,%%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
 #define READNV12                                                               \
@@ skipping 258 matching lines @@
                                 const uint8* v_buf,
                                 uint8* dst_argb,
                                 const struct YuvConstants* yuvconstants,
                                 int width) {
   asm volatile (
     YUVTORGB_SETUP(yuvconstants)
     "sub       %[u_buf],%[v_buf]               \n"
     "pcmpeqb   %%xmm5,%%xmm5                   \n"
     LABELALIGN
   "1:                                          \n"
-    READYUV411
+    READYUV411_EBX
     YUVTORGB(yuvconstants)
     STOREARGB
     "sub       $0x8,%[width]                   \n"
     "jg        1b                              \n"
   : [y_buf]"+r"(y_buf),    // %[y_buf]
     [u_buf]"+r"(u_buf),    // %[u_buf]
     [v_buf]"+r"(v_buf),    // %[v_buf]
     [dst_argb]"+r"(dst_argb),  // %[dst_argb]
     [width]"+rm"(width)    // %[width]
   : [yuvconstants]"r"(yuvconstants)  // %[yuvconstants]
-  : "memory", "cc", NACL_R14 YUVTORGB_REGS
+  : "memory", "cc", "ebx", NACL_R14 YUVTORGB_REGS
     "xmm0", "xmm1", "xmm2", "xmm3", "xmm4", "xmm5"
   );
 }
 
 void OMITFP NV12ToARGBRow_SSSE3(const uint8* y_buf,
                                 const uint8* uv_buf,
                                 uint8* dst_argb,
                                 const struct YuvConstants* yuvconstants,
                                 int width) {
   asm volatile (
@@ skipping 3653 matching lines @@
   );
 }
 #endif  // HAS_ARGBLUMACOLORTABLEROW_SSSE3
 
 #endif  // defined(__x86_64__) || defined(__i386__)
 
 #ifdef __cplusplus
 }  // extern "C"
 }  // namespace libyuv
 #endif