Replacing movntq instruction with movq for 874 branch.

media/base/yuv_row_posix.cc

 // Copyright (c) 2011 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"
 
 #ifndef NDEBUG
 #include "base/logging.h"
 #else
 #define DCHECK(a)
 #endif
 
 extern "C" {
 
+// Branch 874 specific fix to disable movntq to prevent crashes on Pentium IIs.
+#define USE_MOVNTQ 0
+
 #if USE_SSE2 && defined(ARCH_CPU_X86_64)
 
 // AMD64 ABI uses register paremters.
 void FastConvertYUVToRGB32Row(const uint8* y_buf,  // rdi
                               const uint8* u_buf,  // rsi
                               const uint8* v_buf,  // rdx
                               uint8* rgb_buf,      // rcx
                               int width) {         // r8
   asm(
   "jmp    convertend\n"
@@ skipping 266 matching lines @@
   "paddsw kCoefficientsRgbY+4096(,%ebx,8),%mm0\n"
   "movzbl 0x1(%edx),%ebx\n"
   "movq   kCoefficientsRgbY(,%eax,8),%mm1\n"
   "add    $0x2,%edx\n"
   "movq   kCoefficientsRgbY(,%ebx,8),%mm2\n"
   "paddsw %mm0,%mm1\n"
   "paddsw %mm0,%mm2\n"
   "psraw  $0x6,%mm1\n"
   "psraw  $0x6,%mm2\n"
   "packuswb %mm2,%mm1\n"
+#if USE_MOVNTQ
   "movntq %mm1,0x0(%ebp)\n"
+#else
+  "movq   %mm1,0x0(%ebp)\n"
+#endif
   "add    $0x8,%ebp\n"
 "convertend:"
   "sub    $0x2,%ecx\n"
   "jns    convertloop\n"
 
   "and    $0x1,%ecx\n"
   "je     convertdone\n"
 
   "movzbl (%edi),%eax\n"
   "movq   kCoefficientsRgbY+2048(,%eax,8),%mm0\n"
@@ skipping 47 matching lines @@
   "mov    %ebx,%eax\n"
   "add    0x38(%esp),%ebx\n"
   "sar    $0x10,%eax\n"
   "movzbl (%edx,%eax,1),%eax\n"
   "movq   kCoefficientsRgbY(,%eax,8),%mm2\n"
   "paddsw %mm0,%mm1\n"
   "paddsw %mm0,%mm2\n"
   "psraw  $0x6,%mm1\n"
   "psraw  $0x6,%mm2\n"
   "packuswb %mm2,%mm1\n"
+#if USE_MOVNTQ
   "movntq %mm1,0x0(%ebp)\n"
+#else
+  "movq   %mm1,0x0(%ebp)\n"
+#endif
   "add    $0x8,%ebp\n"
 "scaleend:"
   "sub    $0x2,%ecx\n"
   "jns    scaleloop\n"
 
   "and    $0x1,%ecx\n"
   "je     scaledone\n"
 
   "mov    %ebx,%eax\n"
   "sar    $0x11,%eax\n"
@@ skipping 103 matching lines @@
   "imul   %eax, %ecx \n"
   "addl   %esi, %ecx \n"
   "shrl   $16, %ecx \n"
   "movq   kCoefficientsRgbY(,%ecx,8),%mm2\n"
 
   "paddsw %mm0,%mm1\n"
   "paddsw %mm0,%mm2\n"
   "psraw  $0x6,%mm1\n"
   "psraw  $0x6,%mm2\n"
   "packuswb %mm2,%mm1\n"
+#if USE_MOVNTQ
   "movntq %mm1,0x0(%ebp)\n"
+#else
+  "movq   %mm1,0x0(%ebp)\n"
+#endif
   "add    $0x8,%ebp\n"
 
 ".lscaleend:"
   "cmp    0x34(%esp), %ebx\n"
   "jl     .lscaleloop\n"
   "popa\n"
   "ret\n"
 
 ".lscalelastpixel:"
   "paddsw %mm0, %mm1\n"
@@ skipping 38 matching lines @@
   "paddsw 4096(%ecx,%ebx,8),%mm0\n"
   "movzbl 0x1(%edx),%ebx\n"
   "movq   0(%ecx,%eax,8),%mm1\n"
   "add    $0x2,%edx\n"
   "movq   0(%ecx,%ebx,8),%mm2\n"
   "paddsw %mm0,%mm1\n"
   "paddsw %mm0,%mm2\n"
   "psraw  $0x6,%mm1\n"
   "psraw  $0x6,%mm2\n"
   "packuswb %mm2,%mm1\n"
+#if USE_MOVNTQ
   "movntq %mm1,0x0(%ebp)\n"
+#else
+  "movq   %mm1,0x0(%ebp)\n"
+#endif
   "add    $0x8,%ebp\n"
 ".Lconvertend:"
   "subl   $0x2,0x34(%esp)\n"
   "jns    .Lconvertloop\n"
 
   "andl   $0x1,0x34(%esp)\n"
   "je     .Lconvertdone\n"
 
   "movzbl (%edi),%eax\n"
   "movq   2048(%ecx,%eax,8),%mm0\n"
@@ skipping 60 matching lines @@
   "mov    %ebx,%eax\n"
   "add    0x38(%esp),%ebx\n"
   "sar    $0x10,%eax\n"
   "movzbl (%edx,%eax,1),%eax\n"
   "movq   0(%ecx,%eax,8),%mm2\n"
   "paddsw %mm0,%mm1\n"
   "paddsw %mm0,%mm2\n"
   "psraw  $0x6,%mm1\n"
   "psraw  $0x6,%mm2\n"
   "packuswb %mm2,%mm1\n"
+#if USE_MOVNTQ
   "movntq %mm1,0x0(%ebp)\n"
+#else
+  "movq   %mm1,0x0(%ebp)\n"
+#endif
   "add    $0x8,%ebp\n"
 "Lscaleend:"
   "subl   $0x2,0x34(%esp)\n"
   "jns    Lscaleloop\n"
 
   "andl   $0x1,0x34(%esp)\n"
   "je     Lscaledone\n"
 
   "mov    %ebx,%eax\n"
   "sar    $0x11,%eax\n"
@@ skipping 121 matching lines @@
   "imul   %eax, %ecx \n"
   "addl   %esi, %ecx \n"
   "shrl   $16, %ecx \n"
   "movq   (%edi,%ecx,8),%mm2\n"
 
   "paddsw %mm0,%mm1\n"
   "paddsw %mm0,%mm2\n"
   "psraw  $0x6,%mm1\n"
   "psraw  $0x6,%mm2\n"
   "packuswb %mm2,%mm1\n"
+#if USE_MOVNTQ
   "movntq %mm1,0x0(%ebp)\n"
+#else
+  "movq   %mm1,0x0(%ebp)\n"
+#endif
   "add    $0x8,%ebp\n"
 
 ".lscaleend:"
   "cmp    %ebx, 0x34(%esp)\n"
   "jg     .lscaleloop\n"
   "popa\n"
   "ret\n"
 
 ".lscalelastpixel:"
   "paddsw %mm0, %mm1\n"
@@ skipping 127 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"
-