Replacing movntq instruction with movq for 874 branch.

media/base/yuv_row_win.cc

 // Copyright (c) 2010 The Chromium Authors. All rights reserved.
 // Use of this source code is governed by a BSD-style license that can be
 // found in the LICENSE file.
 
 #include "media/base/yuv_row.h"
 
 #define kCoefficientsRgbU kCoefficientsRgbY + 2048
 #define kCoefficientsRgbV kCoefficientsRgbY + 4096
 
 extern "C" {
 
+// Branch 874 specific fix to disable movntq to prevent crashes on Pentium IIs.
+#define USE_MOVNTQ 0
+
 #if USE_MMX
 __declspec(naked)
 void FastConvertYUVToRGB32Row(const uint8* y_buf,
                               const uint8* u_buf,
                               const uint8* v_buf,
                               uint8* rgb_buf,
                               int width) {
   __asm {
     pushad
     mov       edx, [esp + 32 + 4]   // Y
@@ skipping 13 matching lines @@
     paddsw    mm0, [kCoefficientsRgbV + 8 * ebx]
     movzx     ebx, byte ptr [edx + 1]
     movq      mm1, [kCoefficientsRgbY + 8 * eax]
     add       edx, 2
     movq      mm2, [kCoefficientsRgbY + 8 * ebx]
     paddsw    mm1, mm0
     paddsw    mm2, mm0
     psraw     mm1, 6
     psraw     mm2, 6
     packuswb  mm1, mm2
+#if USE_MOVNTQ
     movntq    [ebp], mm1
+#else
+    movq      [ebp], mm1
+#endif
     add       ebp, 8
  convertend :
     sub       ecx, 2
     jns       convertloop
 
     and       ecx, 1  // odd number of pixels?
     jz        convertdone
 
     movzx     eax, byte ptr [edi]
     movq      mm0, [kCoefficientsRgbU + 8 * eax]
@@ skipping 40 matching lines @@
     add       edx, ebx
     movq      mm1, [kCoefficientsRgbY + 8 * eax]
     movzx     eax, byte ptr [edx]
     add       edx, ebx
     movq      mm2, [kCoefficientsRgbY + 8 * eax]
     paddsw    mm1, mm0
     paddsw    mm2, mm0
     psraw     mm1, 6
     psraw     mm2, 6
     packuswb  mm1, mm2
+#if USE_MOVNTQ
     movntq    [ebp], mm1
+#else
+    movq      [ebp], mm1
+#endif
     add       ebp, 8
  wend :
     sub       ecx, 2
     jns       wloop
 
     and       ecx, 1  // odd number of pixels?
     jz        wdone
 
     movzx     eax, byte ptr [edi]
     movq      mm0, [kCoefficientsRgbU + 8 * eax]
@@ skipping 42 matching lines @@
     add       edx, ebx
     movq      mm1, [kCoefficientsRgbY + 8 * eax]
     movzx     eax, byte ptr [edx]
     add       edx, ebx
     movq      mm2, [kCoefficientsRgbY + 8 * eax]
     paddsw    mm1, mm0
     paddsw    mm2, mm0
     psraw     mm1, 6
     psraw     mm2, 6
     packuswb  mm1, mm2
+#if USE_MOVNTQ
     movntq    [ebp], mm1
+#else
+    movq      [ebp], mm1
+#endif
     add       ebp, 8
  wend :
     sub       ecx, 2
     jns       wloop
 
     and       ecx, 1  // odd number of pixels?
     jz        wdone
 
     movzx     eax, byte ptr [edi]
     movq      mm0, [kCoefficientsRgbU + 8 * eax]
@@ skipping 33 matching lines @@
     movzx     ebx, byte ptr [esi]
     add       esi, 1
     movq      mm0, [kCoefficientsRgbU + 8 * eax]
     movzx     eax, byte ptr [edx]
     paddsw    mm0, [kCoefficientsRgbV + 8 * ebx]
     movq      mm1, [kCoefficientsRgbY + 8 * eax]
     paddsw    mm1, mm0
     psraw     mm1, 6
     packuswb  mm1, mm1
     punpckldq mm1, mm1
+#if USE_MOVNTQ
     movntq    [ebp], mm1
+#else
+    movq      [ebp], mm1
+#endif
 
     movzx     ebx, byte ptr [edx + 1]
     add       edx, 2
     paddsw    mm0, [kCoefficientsRgbY + 8 * ebx]
     psraw     mm0, 6
     packuswb  mm0, mm0
     punpckldq mm0, mm0
+#if USE_MOVNTQ
     movntq    [ebp+8], mm0
+#else
+    movq      [ebp+8], mm0
+#endif
     add       ebp, 16
  wend :
     sub       ecx, 4
     jns       wloop
 
     add       ecx, 4
     jz        wdone
 
     movzx     eax, byte ptr [edi]
     movq      mm0, [kCoefficientsRgbU + 8 * eax]
@@ skipping 56 matching lines @@
     mov       eax, ebx
     add       ebx, [esp + 32 + 24]  // x += source_dx
     sar       eax, 16
     movzx     eax, byte ptr [edx + eax]
     movq      mm2, [kCoefficientsRgbY + 8 * eax]
     paddsw    mm1, mm0
     paddsw    mm2, mm0
     psraw     mm1, 6
     psraw     mm2, 6
     packuswb  mm1, mm2
+#if USE_MOVNTQ
     movntq    [ebp], mm1
+#else
+    movq      [ebp], mm1
+#endif
     add       ebp, 8
  scaleend :
     sub       ecx, 2
     jns       scaleloop
 
     and       ecx, 1  // odd number of pixels?
     jz        scaledone
 
     mov       eax, ebx
     sar       eax, 17
@@ skipping 99 matching lines @@
     imul      ecx, eax
     add       ecx, esi
     shr       ecx, 16
     movq      mm2, [kCoefficientsRgbY + 8 * ecx]
 
     paddsw    mm1, mm0
     paddsw    mm2, mm0
     psraw     mm1, 0x6
     psraw     mm2, 0x6
     packuswb  mm1, mm2
+#if USE_MOVNTQ
     movntq    [ebp], mm1
+#else
+    movq      [ebp], mm1
+#endif
     add       ebp, 0x8
 
 lscaleend:
     cmp       ebx, [esp + 32 + 20]
     jl        lscaleloop
     popad
     ret
 
 lscalelastpixel:
     paddsw    mm1, mm0
@@ skipping 140 matching lines @@
       y = (y_frac * y1 + (y_frac ^ 65535) * y0) >> 16;
       YuvPixel(y, u, v, rgb_buf+4);
       x += source_dx;
     }
     rgb_buf += 8;
   }
 }
 
 #endif  // USE_MMX
 }  // extern "C"
-