Switch from xor/mov bx, to movzx ebx, which still passes drmemory and valgrind.

BUG=none
TESTED=drmemory

Review URL: https://webrtc-codereview.appspot.com/5339004

git-svn-id: http://libyuv.googlecode.com/svn/trunk@904 16f28f9a-4ce2-e073-06de-1de4eb20be90

README.chromium

 Name: libyuv
 URL: http://code.google.com/p/libyuv/
-Version: 902
+Version: 904
 License: BSD
 License File: LICENSE
 

include/libyuv/version.h

@@ -11,6 +11,6 @@
 #ifndef INCLUDE_LIBYUV_VERSION_H_  // NOLINT
 #define INCLUDE_LIBYUV_VERSION_H_
 
-#define LIBYUV_VERSION 902
+#define LIBYUV_VERSION 904
 
 #endif  // INCLUDE_LIBYUV_VERSION_H_  NOLINT

source/scale_posix.cc

@@ -857,13 +857,11 @@ void ScaleFilterCols_SSSE3(uint8* dst_ptr, const uint8* src_ptr,
   "2:                                          \n"
     "movdqa    %%xmm2,%%xmm1                   \n"
     "paddd     %%xmm3,%%xmm2                   \n"
-    "xor       %2,%2                           \n"
-    MEMOP(mov,0x00,1,3,1) ",%w2                \n"  //  mov  (%1,%3,1),%w2
+    MEMOP(movzwl,0x00,1,3,1) ",%k2             \n"  //  movzwl  (%1,%3,1),%k2
     "movd      %k2,%%xmm0                      \n"
     "psrlw     $0x9,%%xmm1                     \n"
     BUNDLEALIGN
-    "xor       %2,%2                           \n"
-    MEMOP(mov,0x00,1,4,1) ",%w2                \n"  //  mov  (%1,%4,1),%w2
+    MEMOP(movzwl,0x00,1,4,1) ",%k2             \n"  //  movzwl  (%1,%4,1),%k2
     "movd      %k2,%%xmm4                      \n"
     "pshufb    %%xmm5,%%xmm1                   \n"
     "punpcklwd %%xmm4,%%xmm0                   \n"
@@ -883,8 +881,7 @@ void ScaleFilterCols_SSSE3(uint8* dst_ptr, const uint8* src_ptr,
   "29:                                         \n"
     "addl      $0x1,%5                         \n"
     "jl        99f                             \n"
-    "xor       %2,%2                           \n"
-    MEMOP(mov,0x00,1,3,1) ",%w2                \n"  //  mov  (%1,%3,1),%w2
+    MEMOP(movzwl,0x00,1,3,1) ",%k2             \n"  //  movzwl  (%1,%3,1),%k2
     "movd      %k2,%%xmm0                      \n"
     "psrlw     $0x9,%%xmm2                     \n"
     "pshufb    %%xmm5,%%xmm2                   \n"

source/scale_win.cc

@@ -831,12 +831,10 @@ void ScaleFilterCols_SSSE3(uint8* dst_ptr, const uint8* src_ptr,
   xloop2:
     movdqa     xmm1, xmm2           // x0, x1 fractions.
     paddd      xmm2, xmm3           // x += dx
-    xor        ebx, ebx
-    mov        bx, word ptr [esi + eax]  // 2 source x0 pixels
+    movzx      ebx, word ptr [esi + eax]  // 2 source x0 pixels
     movd       xmm0, ebx
     psrlw      xmm1, 9              // 7 bit fractions.
-    xor        ebx, ebx
-    mov        bx, word ptr [esi + edx]  // 2 source x1 pixels
+    movzx      ebx, word ptr [esi + edx]  // 2 source x1 pixels
     movd       xmm4, ebx
     pshufb     xmm1, xmm5           // 0011
     punpcklwd  xmm0, xmm4
@@ -859,8 +857,7 @@ void ScaleFilterCols_SSSE3(uint8* dst_ptr, const uint8* src_ptr,
     jl         xloop99
 
     // 1 pixel remainder
-    xor        ebx, ebx
-    mov        bx, word ptr [esi + eax]  // 2 source x0 pixels
+    movzx      ebx, word ptr [esi + eax]  // 2 source x0 pixels
     movd       xmm0, ebx
     psrlw      xmm2, 9              // 7 bit fractions.
     pshufb     xmm2, xmm5           // 0011

unit_test/unit_test.cc

@@ -19,8 +19,8 @@
 #define BENCHMARK_ITERATIONS 1
 
 libyuvTest::libyuvTest() : rotate_max_w_(128), rotate_max_h_(128),
-    benchmark_iterations_(BENCHMARK_ITERATIONS), benchmark_width_(353),
-    benchmark_height_(264) {
+    benchmark_iterations_(BENCHMARK_ITERATIONS), benchmark_width_(33),
+    benchmark_height_(17) {
     const char* repeat = getenv("LIBYUV_REPEAT");
     if (repeat) {
       benchmark_iterations_ = atoi(repeat);  // NOLINT