Bug 572034 - Use Linux x86_64 Y'CbCr routines on x86_64 OS X. Convert Linux routines to local asm labels. r=chris.double

--HG--
extra : rebase_source : faa6af7f7529d05ec470796a369e7ea37daf3c69

gfx/ycbcr/Makefile.in

@@ -30,8 +30,13 @@ CPPSRCS += yuv_row_linux.cpp \
            $(NULL)
 else
 ifeq ($(OS_ARCH),Darwin)
+ifeq ($(OS_TEST),x86_64)
+CPPSRCS += yuv_row_linux.cpp \
+           $(NULL)
+else
 CPPSRCS += yuv_row_mac.cpp \
            $(NULL)
+endif
 else
 CPPSRCS += yuv_row_other.cpp \
            $(NULL)

gfx/ycbcr/README

@@ -19,3 +19,4 @@ export.patch: Fix export for building on comm-central
 win64_mac64.patch: Fallback to C implementation on Windows and Mac OS X 64 bit
 yv24.patch: Adds YCbCr 4:4:4 support
 row_c_fix.patch: Fix broken C fallback code (See bug 561385).
+bug572034_mac_64bit.patch: Fix x86_64 linux code so it works on OS X.

gfx/ycbcr/bug572034_mac_64bit.patch

@@ -0,0 +1,66 @@
+diff --git a/gfx/ycbcr/yuv_row_linux.cpp b/gfx/ycbcr/yuv_row_linux.cpp
+--- a/gfx/ycbcr/yuv_row_linux.cpp
++++ b/gfx/ycbcr/yuv_row_linux.cpp
+@@ -250,18 +250,18 @@ MMX_ALIGNED(int16 kCoefficientsRgbY[768]
+ 
+ // 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"
+-"convertloop:"
++  "jmp    Lconvertend\n"
++"Lconvertloop:"
+   "movzb  (%1),%%r10\n"
+   "add    $0x1,%1\n"
+   "movzb  (%2),%%r11\n"
+   "add    $0x1,%2\n"
+   "movq   2048(%5,%%r10,8),%%xmm0\n"
+   "movzb  (%0),%%r10\n"
+   "movq   4096(%5,%%r11,8),%%xmm1\n"
+   "movzb  0x1(%0),%%r11\n"
+@@ -271,36 +271,36 @@ void FastConvertYUVToRGB32Row(const uint
+   "movq   (%5,%%r11,8),%%xmm3\n"
+   "paddsw %%xmm0,%%xmm2\n"
+   "paddsw %%xmm0,%%xmm3\n"
+   "shufps $0x44,%%xmm3,%%xmm2\n"
+   "psraw  $0x6,%%xmm2\n"
+   "packuswb %%xmm2,%%xmm2\n"
+   "movq   %%xmm2,0x0(%3)\n"
+   "add    $0x8,%3\n"
+-"convertend:"
++"Lconvertend:"
+   "sub    $0x2,%4\n"
+-  "jns    convertloop\n"
++  "jns    Lconvertloop\n"
+ 
+-"convertnext:"
++"Lconvertnext:"
+   "add    $0x1,%4\n"
+-  "js     convertdone\n"
++  "js     Lconvertdone\n"
+ 
+   "movzb  (%1),%%r10\n"
+   "movq   2048(%5,%%r10,8),%%xmm0\n"
+   "movzb  (%2),%%r10\n"
+   "movq   4096(%5,%%r10,8),%%xmm1\n"
+   "paddsw %%xmm1,%%xmm0\n"
+   "movzb  (%0),%%r10\n"
+   "movq   (%5,%%r10,8),%%xmm1\n"
+   "paddsw %%xmm0,%%xmm1\n"
+   "psraw  $0x6,%%xmm1\n"
+   "packuswb %%xmm1,%%xmm1\n"
+   "movd   %%xmm1,0x0(%3)\n"
+-"convertdone:"
++"Lconvertdone:"
+   :
+   : "r"(y_buf),  // %0
+     "r"(u_buf),  // %1
+     "r"(v_buf),  // %2
+     "r"(rgb_buf),  // %3
+     "r"(width),  // %4
+     "r" (kCoefficientsRgbY)  // %5
+   : "memory", "r10", "r11", "xmm0", "xmm1", "xmm2", "xmm3"

gfx/ycbcr/update.sh

@@ -13,3 +13,4 @@ patch -p3 <export.patch
 patch -p3 <win64_mac64.patch
 patch -p3 <yv24.patch
 patch -p3 <row_c_fix.patch
+patch -p3 <bug572034_mac_64bit.patch

gfx/ycbcr/yuv_row_linux.cpp

@@ -255,8 +255,8 @@ void FastConvertYUVToRGB32Row(const uint8* y_buf,  // rdi
                               uint8* rgb_buf,      // rcx
                               int width) {         // r8
   asm(
-  "jmp    convertend\n"
-"convertloop:"
+  "jmp    Lconvertend\n"
+"Lconvertloop:"
   "movzb  (%1),%%r10\n"
   "add    $0x1,%1\n"
   "movzb  (%2),%%r11\n"
@@ -276,13 +276,13 @@ void FastConvertYUVToRGB32Row(const uint8* y_buf,  // rdi
   "packuswb %%xmm2,%%xmm2\n"
   "movq   %%xmm2,0x0(%3)\n"
   "add    $0x8,%3\n"
-"convertend:"
+"Lconvertend:"
   "sub    $0x2,%4\n"
-  "jns    convertloop\n"
+  "jns    Lconvertloop\n"
 
-"convertnext:"
+"Lconvertnext:"
   "add    $0x1,%4\n"
-  "js     convertdone\n"
+  "js     Lconvertdone\n"
 
   "movzb  (%1),%%r10\n"
   "movq   2048(%5,%%r10,8),%%xmm0\n"
@@ -295,7 +295,7 @@ void FastConvertYUVToRGB32Row(const uint8* y_buf,  // rdi
   "psraw  $0x6,%%xmm1\n"
   "packuswb %%xmm1,%%xmm1\n"
   "movd   %%xmm1,0x0(%3)\n"
-"convertdone:"
+"Lconvertdone:"
   :
   : "r"(y_buf),  // %0
     "r"(u_buf),  // %1