| author | Chris Double <chris.double@double.co.nz> |
| Thu, 11 Nov 2010 12:54:27 +1300 | |
| changeset 57390 | 64901a1fcf9339c7497c0f203d9df1f95df3c738 |
| parent 57389 | 8ecd9dc6684e04dd6a3d37b0bc8f40c5847e4e7a |
| child 57391 | d2d6455065349ad0114f134e47746f18011df7cb |
| push id | 16900 |
| push user | cdouble@mozilla.com |
| push date | Fri, 12 Nov 2010 03:14:11 +0000 |
| treeherder | mozilla-central@d2d645506534 [default view] [failures only] |
| perfherder | [talos] [build metrics] [platform microbench] (compared to previous push) |
| reviewers | roc, blocking2.0 |
| bugs | 583138 |
| milestone | 2.0b8pre |
| first release with | nightly linux32
nightly linux64
nightly mac
nightly win32
nightly win64
|
| last release without | nightly linux32
nightly linux64
nightly mac
nightly win32
nightly win64
|
--- a/gfx/layers/basic/BasicImages.cpp +++ b/gfx/layers/basic/BasicImages.cpp @@ -169,17 +169,18 @@ BasicPlanarYCbCrImage::SetData(const Dat aData.mPicSize.width, aData.mPicSize.height, size.width, size.height, aData.mYStride, aData.mCbCrStride, size.width*4, type, - gfx::ROTATE_0); + gfx::ROTATE_0, + gfx::FILTER_BILINEAR); } else { gfx::ConvertYCbCrToRGB32(aData.mYChannel, aData.mCbChannel, aData.mCrChannel, mBuffer, aData.mPicX, aData.mPicY,
--- a/gfx/ycbcr/Makefile.in +++ b/gfx/ycbcr/Makefile.in @@ -14,42 +14,38 @@ DEFINES += -D_IMPL_NS_GFX EXPORTS = chromium_types.h \ yuv_convert.h \ yuv_row.h \ $(NULL) CPPSRCS = yuv_convert.cpp \ yuv_row_c.cpp \ + yuv_row_table.cpp \ $(NULL) ifdef _MSC_VER CPPSRCS += yuv_row_win.cpp \ $(NULL) else ifeq ($(OS_ARCH),Linux) -CPPSRCS += yuv_row_linux.cpp \ +CPPSRCS += yuv_row_posix.cpp \ $(NULL) else ifeq ($(OS_ARCH),SunOS) -CPPSRCS += yuv_row_linux.cpp \ +CPPSRCS += yuv_row_posix.cpp \ $(NULL) else ifeq ($(OS_ARCH),Darwin) -ifeq ($(OS_TEST),x86_64) -CPPSRCS += yuv_row_linux.cpp \ +CPPSRCS += yuv_row_posix.cpp \ $(NULL) else -CPPSRCS += yuv_row_mac.cpp \ - $(NULL) -endif -else CPPSRCS += yuv_row_other.cpp \ $(NULL) -endif # mac +endif # Darwin endif # SunOS endif # linux endif # windows EXTRA_DSO_LDOPTS += \ $(LIBS_DIR) \ $(EXTRA_DSO_LIBS) \ $(XPCOM_LIBS) \
--- a/gfx/ycbcr/README +++ b/gfx/ycbcr/README @@ -1,24 +1,15 @@ This color conversion code is from the Chromium open source project available here: http://code.google.com/chromium/ -The code comes from svn revision 40876. +The code comes from svn revision 638400 on 2010-10-26. The code was copied from a Chromium svn checkout using the 'update.sh' script which then applies patches for our build and to add dynamic CPU detection. convert.patch: Change Chromium code to build using Mozilla build system. Add runtime CPU detection for MMX Move default C implementation to work on all platforms. - -picture_region.patch: Change Chromium code to allow a picture region. - The YUV conversion will convert within this - picture region only. - -remove_scale.patch: Removes Chromium scaling code. -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. -solaris.patch: Adds Solaris support, fallback to C implementation on SPARC -add_scale.patch: re-adds Chromium scaling code + Change Chromium code to allow a picture region. + The YUV conversion will convert within this + picture region only. + Add YCbCr 4:4:4 support
deleted file mode 100644 --- a/gfx/ycbcr/add_scale.patch +++ /dev/null @@ -1,953 +0,0 @@ -diff --git a/gfx/ycbcr/yuv_convert.cpp b/gfx/ycbcr/yuv_convert.cpp -index 40ce10f..7d46629 100644 ---- a/gfx/ycbcr/yuv_convert.cpp -+++ b/gfx/ycbcr/yuv_convert.cpp -@@ -82,10 +82,139 @@ NS_GFX_(void) ConvertYCbCrToRGB32(const uint8* y_buf, - - #ifdef ARCH_CPU_X86_FAMILY - // MMX used for FastConvertYUVToRGB32Row requires emms instruction. - if (has_mmx) - EMMS(); - #endif - } - -+// Scale a frame of YUV to 32 bit ARGB. -+NS_GFX_(void) ScaleYCbCrToRGB32(const uint8* y_buf, -+ const uint8* u_buf, -+ const uint8* v_buf, -+ uint8* rgb_buf, -+ int width, -+ int height, -+ int scaled_width, -+ int scaled_height, -+ int y_pitch, -+ int uv_pitch, -+ int rgb_pitch, -+ YUVType yuv_type, -+ Rotate view_rotate) { -+ unsigned int y_shift = yuv_type == YV12 ? 1 : 0; -+ unsigned int x_shift = yuv_type == YV24 ? 0 : 1; -+ bool has_mmx = supports_mmx(); -+ // Diagram showing origin and direction of source sampling. -+ // ->0 4<- -+ // 7 3 -+ // -+ // 6 5 -+ // ->1 2<- -+ // Rotations that start at right side of image. -+ if ((view_rotate == ROTATE_180) || -+ (view_rotate == ROTATE_270) || -+ (view_rotate == MIRROR_ROTATE_0) || -+ (view_rotate == MIRROR_ROTATE_90)) { -+ y_buf += width - 1; -+ u_buf += width / 2 - 1; -+ v_buf += width / 2 - 1; -+ width = -width; -+ } -+ // Rotations that start at bottom of image. -+ if ((view_rotate == ROTATE_90) || -+ (view_rotate == ROTATE_180) || -+ (view_rotate == MIRROR_ROTATE_90) || -+ (view_rotate == MIRROR_ROTATE_180)) { -+ y_buf += (height - 1) * y_pitch; -+ u_buf += ((height >> y_shift) - 1) * uv_pitch; -+ v_buf += ((height >> y_shift) - 1) * uv_pitch; -+ height = -height; -+ } -+ -+ // Handle zero sized destination. -+ if (scaled_width == 0 || scaled_height == 0) -+ return; -+ int scaled_dx = width * 16 / scaled_width; -+ int scaled_dy = height * 16 / scaled_height; -+ -+ int scaled_dx_uv = scaled_dx; -+ -+ if ((view_rotate == ROTATE_90) || -+ (view_rotate == ROTATE_270)) { -+ int tmp = scaled_height; -+ scaled_height = scaled_width; -+ scaled_width = tmp; -+ tmp = height; -+ height = width; -+ width = tmp; -+ int original_dx = scaled_dx; -+ int original_dy = scaled_dy; -+ scaled_dx = ((original_dy >> 4) * y_pitch) << 4; -+ scaled_dx_uv = ((original_dy >> 4) * uv_pitch) << 4; -+ scaled_dy = original_dx; -+ if (view_rotate == ROTATE_90) { -+ y_pitch = -1; -+ uv_pitch = -1; -+ height = -height; -+ } else { -+ y_pitch = 1; -+ uv_pitch = 1; -+ } -+ } -+ -+ for (int y = 0; y < scaled_height; ++y) { -+ uint8* dest_pixel = rgb_buf + y * rgb_pitch; -+ int scaled_y = (y * height / scaled_height); -+ const uint8* y_ptr = y_buf + scaled_y * y_pitch; -+ const uint8* u_ptr = u_buf + (scaled_y >> y_shift) * uv_pitch; -+ const uint8* v_ptr = v_buf + (scaled_y >> y_shift) * uv_pitch; -+ -+#if defined(_MSC_VER) && defined(_M_IX86) -+ if (scaled_width == (width * 2)) { -+ DoubleYUVToRGB32Row(y_ptr, u_ptr, v_ptr, -+ dest_pixel, scaled_width); -+ } else if ((scaled_dx & 15) == 0) { // Scaling by integer scale factor. -+ if (scaled_dx_uv == scaled_dx) { // Not rotated. -+ if (scaled_dx == 16) { // Not scaled -+ if (has_mmx) -+ FastConvertYUVToRGB32Row(y_ptr, u_ptr, v_ptr, -+ dest_pixel, scaled_width); -+ else -+ FastConvertYUVToRGB32Row_C(y_ptr, u_ptr, v_ptr, -+ dest_pixel, scaled_width, x_shift); -+ } else { // Simple scale down. ie half -+ ConvertYUVToRGB32Row(y_ptr, u_ptr, v_ptr, -+ dest_pixel, scaled_width, scaled_dx >> 4); -+ } -+ } else { -+ RotateConvertYUVToRGB32Row(y_ptr, u_ptr, v_ptr, -+ dest_pixel, scaled_width, -+ scaled_dx >> 4, scaled_dx_uv >> 4); -+ } -+#else -+ if (scaled_dx == 16) { // Not scaled -+ if (has_mmx) -+ FastConvertYUVToRGB32Row(y_ptr, u_ptr, v_ptr, -+ dest_pixel, scaled_width); -+ else -+ FastConvertYUVToRGB32Row_C(y_ptr, u_ptr, v_ptr, -+ dest_pixel, scaled_width, x_shift); -+#endif -+ } else { -+ if (has_mmx) -+ ScaleYUVToRGB32Row(y_ptr, u_ptr, v_ptr, -+ dest_pixel, scaled_width, scaled_dx); -+ else -+ ScaleYUVToRGB32Row_C(y_ptr, u_ptr, v_ptr, -+ dest_pixel, scaled_width, scaled_dx, x_shift); -+ -+ } -+ } -+ -+ // MMX used for FastConvertYUVToRGB32Row requires emms instruction. -+ if (has_mmx) -+ EMMS(); -+} -+ - } // namespace gfx - } // namespace mozilla -diff --git a/gfx/ycbcr/yuv_convert.h b/gfx/ycbcr/yuv_convert.h -index c0b678d..a7e5b68 100644 ---- a/gfx/ycbcr/yuv_convert.h -+++ b/gfx/ycbcr/yuv_convert.h -@@ -15,27 +15,56 @@ namespace gfx { - // Type of YUV surface. - // The value of these enums matter as they are used to shift vertical indices. - enum YUVType { - YV12 = 0, // YV12 is half width and half height chroma channels. - YV16 = 1, // YV16 is half width and full height chroma channels. - YV24 = 2 // YV24 is full width and full height chroma channels. - }; - -+// Mirror means flip the image horizontally, as in looking in a mirror. -+// Rotate happens after mirroring. -+enum Rotate { -+ ROTATE_0, // Rotation off. -+ ROTATE_90, // Rotate clockwise. -+ ROTATE_180, // Rotate upside down. -+ ROTATE_270, // Rotate counter clockwise. -+ MIRROR_ROTATE_0, // Mirror horizontally. -+ MIRROR_ROTATE_90, // Mirror then Rotate clockwise. -+ MIRROR_ROTATE_180, // Mirror vertically. -+ MIRROR_ROTATE_270 // Transpose. -+}; -+ - // Convert a frame of YUV to 32 bit ARGB. - // Pass in YV16/YV12 depending on source format - NS_GFX_(void) ConvertYCbCrToRGB32(const uint8* yplane, - const uint8* uplane, - const uint8* vplane, - uint8* rgbframe, - int pic_x, - int pic_y, - int pic_width, - int pic_height, - int ystride, - int uvstride, - int rgbstride, - YUVType yuv_type); - -+// Scale a frame of YUV to 32 bit ARGB. -+// Supports rotation and mirroring. -+NS_GFX_(void) ScaleYCbCrToRGB32(const uint8* yplane, -+ const uint8* uplane, -+ const uint8* vplane, -+ uint8* rgbframe, -+ int frame_width, -+ int frame_height, -+ int scaled_width, -+ int scaled_height, -+ int ystride, -+ int uvstride, -+ int rgbstride, -+ YUVType yuv_type, -+ Rotate view_rotate); -+ - } // namespace gfx - } // namespace mozilla - - #endif // MEDIA_BASE_YUV_CONVERT_H_ -diff --git a/gfx/ycbcr/yuv_row.h b/gfx/ycbcr/yuv_row.h -index 8519008..96969ec 100644 ---- a/gfx/ycbcr/yuv_row.h -+++ b/gfx/ycbcr/yuv_row.h -@@ -24,16 +24,64 @@ void FastConvertYUVToRGB32Row(const uint8* y_buf, - void FastConvertYUVToRGB32Row_C(const uint8* y_buf, - const uint8* u_buf, - const uint8* v_buf, - uint8* rgb_buf, - int width, - unsigned int x_shift); - - -+// Can do 1x, half size or any scale down by an integer amount. -+// Step can be negative (mirroring, rotate 180). -+// This is the third fastest of the scalers. -+void ConvertYUVToRGB32Row(const uint8* y_buf, -+ const uint8* u_buf, -+ const uint8* v_buf, -+ uint8* rgb_buf, -+ int width, -+ int step); -+ -+// Rotate is like Convert, but applies different step to Y versus U and V. -+// This allows rotation by 90 or 270, by stepping by stride. -+// This is the forth fastest of the scalers. -+void RotateConvertYUVToRGB32Row(const uint8* y_buf, -+ const uint8* u_buf, -+ const uint8* v_buf, -+ uint8* rgb_buf, -+ int width, -+ int ystep, -+ int uvstep); -+ -+// Doubler does 4 pixels at a time. Each pixel is replicated. -+// This is the fastest of the scalers. -+void DoubleYUVToRGB32Row(const uint8* y_buf, -+ const uint8* u_buf, -+ const uint8* v_buf, -+ uint8* rgb_buf, -+ int width); -+ -+// Handles arbitrary scaling up or down. -+// Mirroring is supported, but not 90 or 270 degree rotation. -+// Chroma is under sampled every 2 pixels for performance. -+// This is the slowest of the scalers. -+void ScaleYUVToRGB32Row(const uint8* y_buf, -+ const uint8* u_buf, -+ const uint8* v_buf, -+ uint8* rgb_buf, -+ int width, -+ int scaled_dx); -+ -+void ScaleYUVToRGB32Row_C(const uint8* y_buf, -+ const uint8* u_buf, -+ const uint8* v_buf, -+ uint8* rgb_buf, -+ int width, -+ int scaled_dx, -+ unsigned int x_shift); -+ - } // extern "C" - - // x64 uses MMX2 (SSE) so emms is not required. - #if defined(ARCH_CPU_X86) - #if defined(_MSC_VER) - #define EMMS() __asm emms - #else - #define EMMS() asm("emms") -diff --git a/gfx/ycbcr/yuv_row_c.cpp b/gfx/ycbcr/yuv_row_c.cpp -index b5c0018..49eced2 100644 ---- a/gfx/ycbcr/yuv_row_c.cpp -+++ b/gfx/ycbcr/yuv_row_c.cpp -@@ -172,10 +172,31 @@ void FastConvertYUVToRGB32Row_C(const uint8* y_buf, - v = v_buf[x + 1]; - } - YuvPixel(y1, u, v, rgb_buf + 4); - } - rgb_buf += 8; // Advance 2 pixels. - } - } - -+// 28.4 fixed point is used. A shift by 4 isolates the integer. -+// A shift by 5 is used to further subsample the chrominence channels. -+// & 15 isolates the fixed point fraction. >> 2 to get the upper 2 bits, -+// for 1/4 pixel accurate interpolation. -+void ScaleYUVToRGB32Row_C(const uint8* y_buf, -+ const uint8* u_buf, -+ const uint8* v_buf, -+ uint8* rgb_buf, -+ int width, -+ int scaled_dx, -+ unsigned int x_shift) { -+ int scaled_x = 0; -+ for (int x = 0; x < width; ++x) { -+ uint8 u = u_buf[scaled_x >> (4 + x_shift)]; -+ uint8 v = v_buf[scaled_x >> (4 + x_shift)]; -+ uint8 y0 = y_buf[scaled_x >> 4]; -+ YuvPixel(y0, u, v, rgb_buf); -+ rgb_buf += 4; -+ scaled_x += scaled_dx; -+ } -+} - } // extern "C" - -diff --git a/gfx/ycbcr/yuv_row_linux.cpp b/gfx/ycbcr/yuv_row_linux.cpp -index 9f7625c..bff02b3 100644 ---- a/gfx/ycbcr/yuv_row_linux.cpp -+++ b/gfx/ycbcr/yuv_row_linux.cpp -@@ -16,16 +16,24 @@ extern "C" { - void FastConvertYUVToRGB32Row(const uint8* y_buf, - const uint8* u_buf, - const uint8* v_buf, - uint8* rgb_buf, - int width) { - FastConvertYUVToRGB32Row_C(y_buf, u_buf, v_buf, rgb_buf, width, 1); - } - -+void ScaleYUVToRGB32Row(const uint8* y_buf, -+ const uint8* u_buf, -+ const uint8* v_buf, -+ uint8* rgb_buf, -+ int width, -+ int scaled_dx) { -+ ScaleYUVToRGB32Row_C(y_buf, u_buf, v_buf, rgb_buf, width, scaled_dx, 1); -+} - #else - - #define RGBY(i) { \ - static_cast<int16>(1.164 * 64 * (i - 16) + 0.5), \ - static_cast<int16>(1.164 * 64 * (i - 16) + 0.5), \ - static_cast<int16>(1.164 * 64 * (i - 16) + 0.5), \ - 0 \ - } -@@ -365,16 +373,86 @@ void FastConvertYUVToRGB32Row(const uint8* y_buf, // rdi - "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" - ); - } -+ -+void ScaleYUVToRGB32Row(const uint8* y_buf, // rdi -+ const uint8* u_buf, // rsi -+ const uint8* v_buf, // rdx -+ uint8* rgb_buf, // rcx -+ int width, // r8 -+ int scaled_dx) { // r9 -+ asm( -+ "xor %%r11,%%r11\n" -+ "sub $0x2,%4\n" -+ "js scalenext\n" -+ -+"scaleloop:" -+ "mov %%r11,%%r10\n" -+ "sar $0x5,%%r10\n" -+ "movzb (%1,%%r10,1),%%rax\n" -+ "movq 2048(%5,%%rax,8),%%xmm0\n" -+ "movzb (%2,%%r10,1),%%rax\n" -+ "movq 4096(%5,%%rax,8),%%xmm1\n" -+ "lea (%%r11,%6),%%r10\n" -+ "sar $0x4,%%r11\n" -+ "movzb (%0,%%r11,1),%%rax\n" -+ "paddsw %%xmm1,%%xmm0\n" -+ "movq (%5,%%rax,8),%%xmm1\n" -+ "lea (%%r10,%6),%%r11\n" -+ "sar $0x4,%%r10\n" -+ "movzb (%0,%%r10,1),%%rax\n" -+ "movq (%5,%%rax,8),%%xmm2\n" -+ "paddsw %%xmm0,%%xmm1\n" -+ "paddsw %%xmm0,%%xmm2\n" -+ "shufps $0x44,%%xmm2,%%xmm1\n" -+ "psraw $0x6,%%xmm1\n" -+ "packuswb %%xmm1,%%xmm1\n" -+ "movq %%xmm1,0x0(%3)\n" -+ "add $0x8,%3\n" -+ "sub $0x2,%4\n" -+ "jns scaleloop\n" -+ -+"scalenext:" -+ "add $0x1,%4\n" -+ "js scaledone\n" -+ -+ "mov %%r11,%%r10\n" -+ "sar $0x5,%%r10\n" -+ "movzb (%1,%%r10,1),%%rax\n" -+ "movq 2048(%5,%%rax,8),%%xmm0\n" -+ "movzb (%2,%%r10,1),%%rax\n" -+ "movq 4096(%5,%%rax,8),%%xmm1\n" -+ "paddsw %%xmm1,%%xmm0\n" -+ "sar $0x4,%%r11\n" -+ "movzb (%0,%%r11,1),%%rax\n" -+ "movq (%5,%%rax,8),%%xmm1\n" -+ "paddsw %%xmm0,%%xmm1\n" -+ "psraw $0x6,%%xmm1\n" -+ "packuswb %%xmm1,%%xmm1\n" -+ "movd %%xmm1,0x0(%3)\n" -+ -+"scaledone:" -+ : -+ : "r"(y_buf), // %0 -+ "r"(u_buf), // %1 -+ "r"(v_buf), // %2 -+ "r"(rgb_buf), // %3 -+ "r"(width), // %4 -+ "r" (kCoefficientsRgbY), // %5 -+ "r"(static_cast<long>(scaled_dx)) // %6 -+ : "memory", "r10", "r11", "rax", "xmm0", "xmm1", "xmm2" -+); -+} -+ - #endif // __SUNPRO_CC - - #else // ARCH_CPU_X86_64 - - #ifdef __SUNPRO_CC - void FastConvertYUVToRGB32Row(const uint8* y_buf, - const uint8* u_buf, - const uint8* v_buf, -@@ -493,13 +571,87 @@ void FastConvertYUVToRGB32Row(const uint8* y_buf, - "packuswb %mm1,%mm1\n" - "movd %mm1,0x0(%ebp)\n" - "2:" - "popa\n" - "ret\n" - ".previous\n" - ); - -+void ScaleYUVToRGB32Row(const uint8* y_buf, -+ const uint8* u_buf, -+ const uint8* v_buf, -+ uint8* rgb_buf, -+ int width, -+ int scaled_dx); -+ -+ asm( -+ ".global ScaleYUVToRGB32Row\n" -+"ScaleYUVToRGB32Row:\n" -+ "pusha\n" -+ "mov 0x24(%esp),%edx\n" -+ "mov 0x28(%esp),%edi\n" -+ "mov 0x2c(%esp),%esi\n" -+ "mov 0x30(%esp),%ebp\n" -+ "mov 0x34(%esp),%ecx\n" -+ "xor %ebx,%ebx\n" -+ "jmp scaleend\n" -+ -+"scaleloop:" -+ "mov %ebx,%eax\n" -+ "sar $0x5,%eax\n" -+ "movzbl (%edi,%eax,1),%eax\n" -+ "movq kCoefficientsRgbY+2048(,%eax,8),%mm0\n" -+ "mov %ebx,%eax\n" -+ "sar $0x5,%eax\n" -+ "movzbl (%esi,%eax,1),%eax\n" -+ "paddsw kCoefficientsRgbY+4096(,%eax,8),%mm0\n" -+ "mov %ebx,%eax\n" -+ "add 0x38(%esp),%ebx\n" -+ "sar $0x4,%eax\n" -+ "movzbl (%edx,%eax,1),%eax\n" -+ "movq kCoefficientsRgbY(,%eax,8),%mm1\n" -+ "mov %ebx,%eax\n" -+ "add 0x38(%esp),%ebx\n" -+ "sar $0x4,%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" -+ "movntq %mm1,0x0(%ebp)\n" -+ "add $0x8,%ebp\n" -+"scaleend:" -+ "sub $0x2,%ecx\n" -+ "jns scaleloop\n" -+ -+ "and $0x1,%ecx\n" -+ "je scaledone\n" -+ -+ "mov %ebx,%eax\n" -+ "sar $0x5,%eax\n" -+ "movzbl (%edi,%eax,1),%eax\n" -+ "movq kCoefficientsRgbY+2048(,%eax,8),%mm0\n" -+ "mov %ebx,%eax\n" -+ "sar $0x5,%eax\n" -+ "movzbl (%esi,%eax,1),%eax\n" -+ "paddsw kCoefficientsRgbY+4096(,%eax,8),%mm0\n" -+ "mov %ebx,%eax\n" -+ "sar $0x4,%eax\n" -+ "movzbl (%edx,%eax,1),%eax\n" -+ "movq kCoefficientsRgbY(,%eax,8),%mm1\n" -+ "paddsw %mm0,%mm1\n" -+ "psraw $0x6,%mm1\n" -+ "packuswb %mm1,%mm1\n" -+ "movd %mm1,0x0(%ebp)\n" -+ -+"scaledone:" -+ "popa\n" -+ "ret\n" -+); -+ - #endif // __SUNPRO_CC - #endif // ARCH_CPU_X86_64 - #endif // !ARCH_CPU_X86_FAMILY - } // extern "C" - -diff --git a/gfx/ycbcr/yuv_row_mac.cpp b/gfx/ycbcr/yuv_row_mac.cpp -index a1d0058..5acf825 100644 ---- a/gfx/ycbcr/yuv_row_mac.cpp -+++ b/gfx/ycbcr/yuv_row_mac.cpp -@@ -16,16 +16,24 @@ extern "C" { - void FastConvertYUVToRGB32Row(const uint8* y_buf, - const uint8* u_buf, - const uint8* v_buf, - uint8* rgb_buf, - int width) { - FastConvertYUVToRGB32Row_C(y_buf, u_buf, v_buf, rgb_buf, width, 1); - } - -+void ScaleYUVToRGB32Row(const uint8* y_buf, -+ const uint8* u_buf, -+ const uint8* v_buf, -+ uint8* rgb_buf, -+ int width, -+ int scaled_dx) { -+ ScaleYUVToRGB32Row_C(y_buf, u_buf, v_buf, rgb_buf, width, scaled_dx, 1); -+} - #else - - #define RGBY(i) { \ - static_cast<int16>(1.164 * 64 * (i - 16) + 0.5), \ - static_cast<int16>(1.164 * 64 * (i - 16) + 0.5), \ - static_cast<int16>(1.164 * 64 * (i - 16) + 0.5), \ - 0 \ - } -@@ -313,11 +321,96 @@ void FastConvertYUVToRGB32Row(const uint8* y_buf, - const uint8* u_buf, - const uint8* v_buf, - uint8* rgb_buf, - int width) { - MacConvertYUVToRGB32Row(y_buf, u_buf, v_buf, rgb_buf, width, - &kCoefficientsRgbY[0][0]); - } - -+extern void MacScaleYUVToRGB32Row(const uint8* y_buf, -+ const uint8* u_buf, -+ const uint8* v_buf, -+ uint8* rgb_buf, -+ int width, -+ int scaled_dx, -+ int16 *kCoefficientsRgbY); -+ -+ __asm__( -+"_MacScaleYUVToRGB32Row:\n" -+ "pusha\n" -+ "mov 0x24(%esp),%edx\n" -+ "mov 0x28(%esp),%edi\n" -+ "mov 0x2c(%esp),%esi\n" -+ "mov 0x30(%esp),%ebp\n" -+ "mov 0x3c(%esp),%ecx\n" -+ "xor %ebx,%ebx\n" -+ "jmp Lscaleend\n" -+ -+"Lscaleloop:" -+ "mov %ebx,%eax\n" -+ "sar $0x5,%eax\n" -+ "movzbl (%edi,%eax,1),%eax\n" -+ "movq 2048(%ecx,%eax,8),%mm0\n" -+ "mov %ebx,%eax\n" -+ "sar $0x5,%eax\n" -+ "movzbl (%esi,%eax,1),%eax\n" -+ "paddsw 4096(%ecx,%eax,8),%mm0\n" -+ "mov %ebx,%eax\n" -+ "add 0x38(%esp),%ebx\n" -+ "sar $0x4,%eax\n" -+ "movzbl (%edx,%eax,1),%eax\n" -+ "movq 0(%ecx,%eax,8),%mm1\n" -+ "mov %ebx,%eax\n" -+ "add 0x38(%esp),%ebx\n" -+ "sar $0x4,%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" -+ "movntq %mm1,0x0(%ebp)\n" -+ "add $0x8,%ebp\n" -+"Lscaleend:" -+ "sub $0x2,0x34(%esp)\n" -+ "jns Lscaleloop\n" -+ -+ "and $0x1,0x34(%esp)\n" -+ "je Lscaledone\n" -+ -+ "mov %ebx,%eax\n" -+ "sar $0x5,%eax\n" -+ "movzbl (%edi,%eax,1),%eax\n" -+ "movq 2048(%ecx,%eax,8),%mm0\n" -+ "mov %ebx,%eax\n" -+ "sar $0x5,%eax\n" -+ "movzbl (%esi,%eax,1),%eax\n" -+ "paddsw 4096(%ecx,%eax,8),%mm0\n" -+ "mov %ebx,%eax\n" -+ "sar $0x4,%eax\n" -+ "movzbl (%edx,%eax,1),%eax\n" -+ "movq 0(%ecx,%eax,8),%mm1\n" -+ "paddsw %mm0,%mm1\n" -+ "psraw $0x6,%mm1\n" -+ "packuswb %mm1,%mm1\n" -+ "movd %mm1,0x0(%ebp)\n" -+ -+"Lscaledone:" -+ "popa\n" -+ "ret\n" -+); -+ -+void ScaleYUVToRGB32Row(const uint8* y_buf, -+ const uint8* u_buf, -+ const uint8* v_buf, -+ uint8* rgb_buf, -+ int width, -+ int scaled_dx) { -+ -+ MacScaleYUVToRGB32Row(y_buf, u_buf, v_buf, rgb_buf, width, scaled_dx, -+ &kCoefficientsRgbY[0][0]); -+} -+ - #endif // ARCH_CPU_PPC || ARCH_CPU_64_BITS - } // extern "C" - -diff --git a/gfx/ycbcr/yuv_row_win.cpp b/gfx/ycbcr/yuv_row_win.cpp -index 699ac77..a1700fc 100644 ---- a/gfx/ycbcr/yuv_row_win.cpp -+++ b/gfx/ycbcr/yuv_row_win.cpp -@@ -11,17 +11,26 @@ extern "C" { - // PPC implementation uses C fallback - void FastConvertYUVToRGB32Row(const uint8* y_buf, - const uint8* u_buf, - const uint8* v_buf, - uint8* rgb_buf, - int width) { - FastConvertYUVToRGB32Row_C(y_buf, u_buf, v_buf, rgb_buf, width, 1); - } -- -+ -+void ScaleYUVToRGB32Row(const uint8* y_buf, -+ const uint8* u_buf, -+ const uint8* v_buf, -+ uint8* rgb_buf, -+ int width, -+ int scaled_dx) { -+ ScaleYUVToRGB32Row_C(y_buf, u_buf, v_buf, rgb_buf, width, scaled_dx, 1); -+} -+ - #else - - - #define RGBY(i) { \ - static_cast<int16>(1.164 * 64 * (i - 16) + 0.5), \ - static_cast<int16>(1.164 * 64 * (i - 16) + 0.5), \ - static_cast<int16>(1.164 * 64 * (i - 16) + 0.5), \ - 0 \ -@@ -307,11 +316,280 @@ void FastConvertYUVToRGB32Row(const uint8* y_buf, - movd [ebp], mm1 - convertdone : - - popad - ret - } - } - -+__declspec(naked) -+void ConvertYUVToRGB32Row(const uint8* y_buf, -+ const uint8* u_buf, -+ const uint8* v_buf, -+ uint8* rgb_buf, -+ int width, -+ int step) { -+ __asm { -+ pushad -+ mov edx, [esp + 32 + 4] // Y -+ mov edi, [esp + 32 + 8] // U -+ mov esi, [esp + 32 + 12] // V -+ mov ebp, [esp + 32 + 16] // rgb -+ mov ecx, [esp + 32 + 20] // width -+ mov ebx, [esp + 32 + 24] // step -+ jmp wend -+ -+ wloop : -+ movzx eax, byte ptr [edi] -+ add edi, ebx -+ movq mm0, [kCoefficientsRgbU + 8 * eax] -+ movzx eax, byte ptr [esi] -+ add esi, ebx -+ paddsw mm0, [kCoefficientsRgbV + 8 * eax] -+ movzx eax, byte ptr [edx] -+ 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 -+ movntq [ebp], mm1 -+ 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] -+ movzx eax, byte ptr [esi] -+ paddsw mm0, [kCoefficientsRgbV + 8 * eax] -+ movzx eax, byte ptr [edx] -+ movq mm1, [kCoefficientsRgbY + 8 * eax] -+ paddsw mm1, mm0 -+ psraw mm1, 6 -+ packuswb mm1, mm1 -+ movd [ebp], mm1 -+ wdone : -+ -+ popad -+ ret -+ } -+} -+ -+__declspec(naked) -+void RotateConvertYUVToRGB32Row(const uint8* y_buf, -+ const uint8* u_buf, -+ const uint8* v_buf, -+ uint8* rgb_buf, -+ int width, -+ int ystep, -+ int uvstep) { -+ __asm { -+ pushad -+ mov edx, [esp + 32 + 4] // Y -+ mov edi, [esp + 32 + 8] // U -+ mov esi, [esp + 32 + 12] // V -+ mov ebp, [esp + 32 + 16] // rgb -+ mov ecx, [esp + 32 + 20] // width -+ jmp wend -+ -+ wloop : -+ movzx eax, byte ptr [edi] -+ mov ebx, [esp + 32 + 28] // uvstep -+ add edi, ebx -+ movq mm0, [kCoefficientsRgbU + 8 * eax] -+ movzx eax, byte ptr [esi] -+ add esi, ebx -+ paddsw mm0, [kCoefficientsRgbV + 8 * eax] -+ movzx eax, byte ptr [edx] -+ mov ebx, [esp + 32 + 24] // ystep -+ 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 -+ movntq [ebp], mm1 -+ 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] -+ movzx eax, byte ptr [esi] -+ paddsw mm0, [kCoefficientsRgbV + 8 * eax] -+ movzx eax, byte ptr [edx] -+ movq mm1, [kCoefficientsRgbY + 8 * eax] -+ paddsw mm1, mm0 -+ psraw mm1, 6 -+ packuswb mm1, mm1 -+ movd [ebp], mm1 -+ wdone : -+ -+ popad -+ ret -+ } -+} -+ -+__declspec(naked) -+void DoubleYUVToRGB32Row(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 -+ mov edi, [esp + 32 + 8] // U -+ mov esi, [esp + 32 + 12] // V -+ mov ebp, [esp + 32 + 16] // rgb -+ mov ecx, [esp + 32 + 20] // width -+ jmp wend -+ -+ wloop : -+ movzx eax, byte ptr [edi] -+ add edi, 1 -+ 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 -+ movntq [ebp], mm1 -+ -+ movzx ebx, byte ptr [edx + 1] -+ add edx, 2 -+ paddsw mm0, [kCoefficientsRgbY + 8 * ebx] -+ psraw mm0, 6 -+ packuswb mm0, mm0 -+ punpckldq mm0, mm0 -+ movntq [ebp+8], mm0 -+ add ebp, 16 -+ wend : -+ sub ecx, 4 -+ jns wloop -+ -+ add ecx, 4 -+ jz wdone -+ -+ movzx eax, byte ptr [edi] -+ movq mm0, [kCoefficientsRgbU + 8 * eax] -+ movzx eax, byte ptr [esi] -+ paddsw mm0, [kCoefficientsRgbV + 8 * eax] -+ movzx eax, byte ptr [edx] -+ movq mm1, [kCoefficientsRgbY + 8 * eax] -+ paddsw mm1, mm0 -+ psraw mm1, 6 -+ packuswb mm1, mm1 -+ jmp wend1 -+ -+ wloop1 : -+ movd [ebp], mm1 -+ add ebp, 4 -+ wend1 : -+ sub ecx, 1 -+ jns wloop1 -+ wdone : -+ popad -+ ret -+ } -+} -+ -+// This version does general purpose scaling by any amount, up or down. -+// The only thing it can not do it rotation by 90 or 270. -+// For performance the chroma is under sampled, reducing cost of a 3x -+// 1080p scale from 8.4 ms to 5.4 ms. -+__declspec(naked) -+void ScaleYUVToRGB32Row(const uint8* y_buf, -+ const uint8* u_buf, -+ const uint8* v_buf, -+ uint8* rgb_buf, -+ int width, -+ int dx) { -+ __asm { -+ pushad -+ mov edx, [esp + 32 + 4] // Y -+ mov edi, [esp + 32 + 8] // U -+ mov esi, [esp + 32 + 12] // V -+ mov ebp, [esp + 32 + 16] // rgb -+ mov ecx, [esp + 32 + 20] // width -+ xor ebx, ebx // x -+ jmp scaleend -+ -+ scaleloop : -+ mov eax, ebx -+ sar eax, 5 -+ movzx eax, byte ptr [edi + eax] -+ movq mm0, [kCoefficientsRgbU + 8 * eax] -+ mov eax, ebx -+ sar eax, 5 -+ movzx eax, byte ptr [esi + eax] -+ paddsw mm0, [kCoefficientsRgbV + 8 * eax] -+ mov eax, ebx -+ add ebx, [esp + 32 + 24] // x += dx -+ sar eax, 4 -+ movzx eax, byte ptr [edx + eax] -+ movq mm1, [kCoefficientsRgbY + 8 * eax] -+ mov eax, ebx -+ add ebx, [esp + 32 + 24] // x += dx -+ sar eax, 4 -+ 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 -+ movntq [ebp], mm1 -+ add ebp, 8 -+ scaleend : -+ sub ecx, 2 -+ jns scaleloop -+ -+ and ecx, 1 // odd number of pixels? -+ jz scaledone -+ -+ mov eax, ebx -+ sar eax, 5 -+ movzx eax, byte ptr [edi + eax] -+ movq mm0, [kCoefficientsRgbU + 8 * eax] -+ mov eax, ebx -+ sar eax, 5 -+ movzx eax, byte ptr [esi + eax] -+ paddsw mm0, [kCoefficientsRgbV + 8 * eax] -+ mov eax, ebx -+ sar eax, 4 -+ movzx eax, byte ptr [edx + eax] -+ movq mm1, [kCoefficientsRgbY + 8 * eax] -+ paddsw mm1, mm0 -+ psraw mm1, 6 -+ packuswb mm1, mm1 -+ movd [ebp], mm1 -+ -+ scaledone : -+ popad -+ ret -+ } -+} -+ - #endif // ARCH_CPU_64_BITS - } // extern "C" -
deleted file mode 100644 --- a/gfx/ycbcr/bug572034_mac_64bit.patch +++ /dev/null @@ -1,144 +0,0 @@ -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 1f\n" -+"0:" - "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:" -+"1:" - "sub $0x2,%4\n" -- "jns convertloop\n" -+ "jns 0b\n" - --"convertnext:" -+"2:" - "add $0x1,%4\n" -- "js convertdone\n" -+ "js 3f\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:" -+"3:" - : - : "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" -@@ -309,28 +309,35 @@ void FastConvertYUVToRGB32Row(const uint - - #else - - void FastConvertYUVToRGB32Row(const uint8* y_buf, - const uint8* u_buf, - const uint8* v_buf, - uint8* rgb_buf, - int width); -+ -+// It's necessary to specify the correct section for the following code, -+// otherwise it will be placed in whatever the current section is as this unit -+// is compiled. Because GCC remembers the last section it emitted, we must -+// also revert to the previous section state at the end of the asm block. - asm( -+ ".section .text\n" - ".global FastConvertYUVToRGB32Row\n" -+ ".type FastConvertYUVToRGB32Row, @function\n" - "FastConvertYUVToRGB32Row:\n" - "pusha\n" - "mov 0x24(%esp),%edx\n" - "mov 0x28(%esp),%edi\n" - "mov 0x2c(%esp),%esi\n" - "mov 0x30(%esp),%ebp\n" - "mov 0x34(%esp),%ecx\n" -- "jmp convertend\n" -+ "jmp 1f\n" - --"convertloop:" -+"0:" - "movzbl (%edi),%eax\n" - "add $0x1,%edi\n" - "movzbl (%esi),%ebx\n" - "add $0x1,%esi\n" - "movq kCoefficientsRgbY+2048(,%eax,8),%mm0\n" - "movzbl (%edx),%eax\n" - "paddsw kCoefficientsRgbY+4096(,%ebx,8),%mm0\n" - "movzbl 0x1(%edx),%ebx\n" -@@ -339,34 +346,35 @@ void FastConvertYUVToRGB32Row(const uint - "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" - "movntq %mm1,0x0(%ebp)\n" - "add $0x8,%ebp\n" --"convertend:" -+"1:" - "sub $0x2,%ecx\n" -- "jns convertloop\n" -+ "jns 0b\n" - - "and $0x1,%ecx\n" -- "je convertdone\n" -+ "je 2f\n" - - "movzbl (%edi),%eax\n" - "movq kCoefficientsRgbY+2048(,%eax,8),%mm0\n" - "movzbl (%esi),%eax\n" - "paddsw kCoefficientsRgbY+4096(,%eax,8),%mm0\n" - "movzbl (%edx),%eax\n" - "movq kCoefficientsRgbY(,%eax,8),%mm1\n" - "paddsw %mm0,%mm1\n" - "psraw $0x6,%mm1\n" - "packuswb %mm1,%mm1\n" - "movd %mm1,0x0(%ebp)\n" --"convertdone:" -+"2:" - "popa\n" - "ret\n" -+ ".previous\n" - ); - - #endif - #endif // ARCH_CPU_ARM_FAMILY - } // extern "C" -
deleted file mode 100644 --- a/gfx/ycbcr/bug577645_movntq.patch +++ /dev/null @@ -1,63 +0,0 @@ -diff --git a/gfx/ycbcr/yuv_convert.cpp b/gfx/ycbcr/yuv_convert.cpp ---- a/gfx/ycbcr/yuv_convert.cpp -+++ b/gfx/ycbcr/yuv_convert.cpp -@@ -36,19 +36,21 @@ NS_GFX_(void) ConvertYCbCrToRGB32(const - int pic_width, - int pic_height, - int y_pitch, - int uv_pitch, - int rgb_pitch, - YUVType yuv_type) { - unsigned int y_shift = yuv_type == YV12 ? 1 : 0; - unsigned int x_shift = yuv_type == YV24 ? 0 : 1; -- // There is no optimized YV24 MMX routine so we check for this and -+ // Test for SSE because the optimized code uses movntq, which is not part of MMX. -+ bool has_sse = supports_mmx() && supports_sse(); -+ // There is no optimized YV24 SSE routine so we check for this and - // fall back to the C code. -- bool has_mmx = supports_mmx() && yuv_type != YV24; -+ has_sse &= yuv_type != YV24; - bool odd_pic_x = yuv_type != YV24 && pic_x % 2 != 0; - int x_width = odd_pic_x ? pic_width - 1 : pic_width; - - for (int y = pic_y; y < pic_height + pic_y; ++y) { - uint8* rgb_row = rgb_buf + (y - pic_y) * rgb_pitch; - const uint8* y_ptr = y_buf + y * y_pitch + pic_x; - const uint8* u_ptr = u_buf + (y >> y_shift) * uv_pitch + (pic_x >> x_shift); - const uint8* v_ptr = v_buf + (y >> y_shift) * uv_pitch + (pic_x >> x_shift); -@@ -60,32 +62,32 @@ NS_GFX_(void) ConvertYCbCrToRGB32(const - u_ptr++, - v_ptr++, - rgb_row, - 1, - x_shift); - rgb_row += 4; - } - -- if (has_mmx) -+ if (has_sse) - FastConvertYUVToRGB32Row(y_ptr, - u_ptr, - v_ptr, - rgb_row, - x_width); - else - FastConvertYUVToRGB32Row_C(y_ptr, - u_ptr, - v_ptr, - rgb_row, - x_width, - x_shift); - } - - #ifdef ARCH_CPU_X86_FAMILY -- // MMX used for FastConvertYUVToRGB32Row requires emms instruction. -- if (has_mmx) -+ // SSE used for FastConvertYUVToRGB32Row requires emms instruction. -+ if (has_sse) - EMMS(); - #endif - } - - } // namespace gfx - } // namespace mozilla
deleted file mode 100644 --- a/gfx/ycbcr/bustage.patch +++ /dev/null @@ -1,26 +0,0 @@ -diff --git a/gfx/ycbcr/yuv_row_linux.cpp b/gfx/ycbcr/yuv_row_linux.cpp -index ce5ee89..455dd7b 100644 ---- a/gfx/ycbcr/yuv_row_linux.cpp -+++ b/gfx/ycbcr/yuv_row_linux.cpp -@@ -18,7 +18,7 @@ void FastConvertYUVToRGB32Row(const uint8* y_buf, - const uint8* v_buf, - uint8* rgb_buf, - int width) { -- FastConvertYUVToRGB32Row_C(y_buf, u_buf, v_buf, rgb_buf, width); -+ FastConvertYUVToRGB32Row_C(y_buf, u_buf, v_buf, rgb_buf, width, 1); - } - - #else -diff --git a/gfx/ycbcr/yuv_row_mac.cpp b/gfx/ycbcr/yuv_row_mac.cpp -index 3515ada..351466c 100644 ---- a/gfx/ycbcr/yuv_row_mac.cpp -+++ b/gfx/ycbcr/yuv_row_mac.cpp -@@ -15,7 +15,7 @@ void FastConvertYUVToRGB32Row(const uint8* y_buf, - const uint8* v_buf, - uint8* rgb_buf, - int width) { -- FastConvertYUVToRGB32Row_C(y_buf, u_buf, v_buf, rgb_buf, width); -+ FastConvertYUVToRGB32Row_C(y_buf, u_buf, v_buf, rgb_buf, width, 1); - } - - #else
--- a/gfx/ycbcr/chromium_types.h +++ b/gfx/ycbcr/chromium_types.h @@ -49,23 +49,24 @@ typedef PRInt16 int16; // http://www.agner.org/optimize/calling_conventions.pdf // or with gcc, run: "echo | gcc -E -dM -" #if defined(_M_X64) || defined(__x86_64__) #define ARCH_CPU_X86_FAMILY 1 #define ARCH_CPU_X86_64 1 #define ARCH_CPU_64_BITS 1 #elif defined(_M_IX86) || defined(__i386__) || defined(__i386) #define ARCH_CPU_X86_FAMILY 1 +#define ARCH_CPU_X86_32 1 #define ARCH_CPU_X86 1 #define ARCH_CPU_32_BITS 1 #elif defined(__ARMEL__) #define ARCH_CPU_ARM_FAMILY 1 #define ARCH_CPU_ARMEL 1 #define ARCH_CPU_32_BITS 1 -#elif defined(__ppc__) +#elif defined(__ppc__) || defined(__powerpc) || defined(__PPC__) #define ARCH_CPU_PPC_FAMILY 1 #define ARCH_CPU_PPC 1 #define ARCH_CPU_32_BITS 1 #elif defined(__sparc) #define ARCH_CPU_SPARC_FAMILY 1 #define ARCH_CPU_SPARC 1 #define ARCH_CPU_32_BITS 1 #elif defined(__sparcv9)
--- a/gfx/ycbcr/convert.patch +++ b/gfx/ycbcr/convert.patch @@ -1,292 +1,438 @@ diff --git b/gfx/ycbcr/yuv_convert.cpp a/gfx/ycbcr/yuv_convert.cpp -index c73dfe4..c291d5c 100644 +index bea0e50..ab4f10a 100644 --- b/gfx/ycbcr/yuv_convert.cpp +++ a/gfx/ycbcr/yuv_convert.cpp -@@ -10,67 +10,80 @@ +@@ -6,77 +6,104 @@ + // http://www.fourcc.org/yuv.php + // The actual conversion is best described here + // http://en.wikipedia.org/wiki/YUV + // An article on optimizing YUV conversion using tables instead of multiplies // http://lestourtereaux.free.fr/papers/data/yuvrgb.pdf // // YV12 is a full plane of Y and a half height, half width chroma planes // YV16 is a full plane of Y and a full height, half width chroma planes ++// YV24 is a full plane of Y and a full height, full width chroma planes // // ARGB pixel format is output, which on little endian is stored as BGRA. // The alpha is set to 255, allowing the application to use RGBA or RGB32. -#include "media/base/yuv_convert.h" +#include "yuv_convert.h" // Header for low level row functions. -#include "media/base/yuv_row.h" +#include "yuv_row.h" ++#define MOZILLA_SSE_INCLUDE_HEADER_FOR_SSE2 ++#define MOZILLA_SSE_INCLUDE_HEADER_FOR_MMX +#include "mozilla/SSE.h" +-#if USE_MMX +-#if defined(_MSC_VER) +-#include <intrin.h> +-#else +-#include <mmintrin.h> +-#endif +-#endif +- +-#if USE_SSE2 +-#include <emmintrin.h> +-#endif +- -namespace media { +namespace mozilla { -+ + +namespace gfx { ++ + // 16.16 fixed point arithmetic + const int kFractionBits = 16; + const int kFractionMax = 1 << kFractionBits; + const int kFractionMask = ((1 << kFractionBits) - 1); // Convert a frame of YUV to 32 bit ARGB. -void ConvertYUVToRGB32(const uint8* y_buf, - const uint8* u_buf, - const uint8* v_buf, - uint8* rgb_buf, - int width, - int height, - int y_pitch, - int uv_pitch, - int rgb_pitch, - YUVType yuv_type) { -+void ConvertYCbCrToRGB32(const uint8* y_buf, -+ const uint8* u_buf, -+ const uint8* v_buf, -+ uint8* rgb_buf, -+ int width, -+ int height, -+ int y_pitch, -+ int uv_pitch, -+ int rgb_pitch, -+ YUVType yuv_type) { - unsigned int y_shift = yuv_type; -+ bool has_mmx = supports_mmx(); - for (int y = 0; y < height; ++y) { - uint8* rgb_row = rgb_buf + y * rgb_pitch; - const uint8* y_ptr = y_buf + y * y_pitch; - const uint8* u_ptr = u_buf + (y >> y_shift) * uv_pitch; - const uint8* v_ptr = v_buf + (y >> y_shift) * uv_pitch; +- unsigned int y_shift = yuv_type; +- for (int y = 0; y < height; ++y) { +- uint8* rgb_row = rgb_buf + y * rgb_pitch; +- const uint8* y_ptr = y_buf + y * y_pitch; +- const uint8* u_ptr = u_buf + (y >> y_shift) * uv_pitch; +- const uint8* v_ptr = v_buf + (y >> y_shift) * uv_pitch; ++NS_GFX_(void) ConvertYCbCrToRGB32(const uint8* y_buf, ++ const uint8* u_buf, ++ const uint8* v_buf, ++ uint8* rgb_buf, ++ int pic_x, ++ int pic_y, ++ int pic_width, ++ int pic_height, ++ int y_pitch, ++ int uv_pitch, ++ int rgb_pitch, ++ YUVType yuv_type) { ++ unsigned int y_shift = yuv_type == YV12 ? 1 : 0; ++ unsigned int x_shift = yuv_type == YV24 ? 0 : 1; ++ // Test for SSE because the optimized code uses movntq, which is not part of MMX. ++ bool has_sse = supports_mmx() && supports_sse(); ++ // There is no optimized YV24 SSE routine so we check for this and ++ // fall back to the C code. ++ has_sse &= yuv_type != YV24; ++ bool odd_pic_x = yuv_type != YV24 && pic_x % 2 != 0; ++ int x_width = odd_pic_x ? pic_width - 1 : pic_width; ++ ++ for (int y = pic_y; y < pic_height + pic_y; ++y) { ++ uint8* rgb_row = rgb_buf + (y - pic_y) * rgb_pitch; ++ const uint8* y_ptr = y_buf + y * y_pitch + pic_x; ++ const uint8* u_ptr = u_buf + (y >> y_shift) * uv_pitch + (pic_x >> x_shift); ++ const uint8* v_ptr = v_buf + (y >> y_shift) * uv_pitch + (pic_x >> x_shift); - FastConvertYUVToRGB32Row(y_ptr, - u_ptr, - v_ptr, - rgb_row, - width); -+ if (has_mmx) ++ if (odd_pic_x) { ++ // Handle the single odd pixel manually and use the ++ // fast routines for the remaining. ++ FastConvertYUVToRGB32Row_C(y_ptr++, ++ u_ptr++, ++ v_ptr++, ++ rgb_row, ++ 1, ++ x_shift); ++ rgb_row += 4; ++ } ++ ++ if (has_sse) { + FastConvertYUVToRGB32Row(y_ptr, + u_ptr, + v_ptr, + rgb_row, -+ width); -+ else ++ x_width); ++ } ++ else { + FastConvertYUVToRGB32Row_C(y_ptr, + u_ptr, + v_ptr, + rgb_row, -+ width); ++ x_width, ++ x_shift); ++ } } // MMX used for FastConvertYUVToRGB32Row requires emms instruction. - EMMS(); -+ if (has_mmx) ++ if (has_sse) + EMMS(); } +-#if USE_SSE2 ++#if defined(MOZILLA_COMPILE_WITH_SSE2) + // FilterRows combines two rows of the image using linear interpolation. + // SSE2 version does 16 pixels at a time +- + static void FilterRows(uint8* ybuf, const uint8* y0_ptr, const uint8* y1_ptr, + int source_width, int source_y_fraction) { + __m128i zero = _mm_setzero_si128(); + __m128i y1_fraction = _mm_set1_epi16(source_y_fraction); + __m128i y0_fraction = _mm_set1_epi16(256 - source_y_fraction); + + const __m128i* y0_ptr128 = reinterpret_cast<const __m128i*>(y0_ptr); + const __m128i* y1_ptr128 = reinterpret_cast<const __m128i*>(y1_ptr); +@@ -99,17 +126,17 @@ static void FilterRows(uint8* ybuf, const uint8* y0_ptr, const uint8* y1_ptr, + y0 = _mm_srli_epi16(y0, 8); + y2 = _mm_srli_epi16(y2, 8); + y0 = _mm_packus_epi16(y0, y2); + *dest128++ = y0; + ++y0_ptr128; + ++y1_ptr128; + } while (dest128 < end128); + } +-#elif USE_MMX ++#elif defined(MOZILLA_COMPILE_WITH_MMX) + // MMX version does 8 pixels at a time + static void FilterRows(uint8* ybuf, const uint8* y0_ptr, const uint8* y1_ptr, + int source_width, int source_y_fraction) { + __m64 zero = _mm_setzero_si64(); + __m64 y1_fraction = _mm_set1_pi16(source_y_fraction); + __m64 y0_fraction = _mm_set1_pi16(256 - source_y_fraction); + + const __m64* y0_ptr64 = reinterpret_cast<const __m64*>(y0_ptr); +@@ -154,44 +181,45 @@ static void FilterRows(uint8* ybuf, const uint8* y0_ptr, const uint8* y1_ptr, + ybuf[7] = (y0_ptr[7] * y0_fraction + y1_ptr[7] * y1_fraction) >> 8; + y0_ptr += 8; + y1_ptr += 8; + ybuf += 8; + } while (ybuf < end); + } + #endif + +- // Scale a frame of YUV to 32 bit ARGB. -void ScaleYUVToRGB32(const uint8* y_buf, - const uint8* u_buf, - const uint8* v_buf, - uint8* rgb_buf, +- int source_width, +- int source_height, - int width, - int height, -- int scaled_width, -- int scaled_height, - int y_pitch, - int uv_pitch, - int rgb_pitch, - YUVType yuv_type, -- Rotate view_rotate) { -+void ScaleYCbCrToRGB32(const uint8* y_buf, -+ const uint8* u_buf, -+ const uint8* v_buf, -+ uint8* rgb_buf, -+ int width, -+ int height, -+ int scaled_width, -+ int scaled_height, -+ int y_pitch, -+ int uv_pitch, -+ int rgb_pitch, -+ YUVType yuv_type, -+ Rotate view_rotate) { - unsigned int y_shift = yuv_type; +- Rotate view_rotate, +- ScaleFilter filter) { ++NS_GFX_(void) ScaleYCbCrToRGB32(const uint8* y_buf, ++ const uint8* u_buf, ++ const uint8* v_buf, ++ uint8* rgb_buf, ++ int source_width, ++ int source_height, ++ int width, ++ int height, ++ int y_pitch, ++ int uv_pitch, ++ int rgb_pitch, ++ YUVType yuv_type, ++ Rotate view_rotate, ++ ScaleFilter filter) { + bool has_mmx = supports_mmx(); ++ + // 4096 allows 3 buffers to fit in 12k. + // Helps performance on CPU with 16K L1 cache. + // Large enough for 3830x2160 and 30" displays which are 2560x1600. + const int kFilterBufferSize = 4096; + // Disable filtering if the screen is too big (to avoid buffer overflows). + // This should never happen to regular users: they don't have monitors + // wider than 4096 pixels. + // TODO(fbarchard): Allow rotated videos to filter. + if (source_width > kFilterBufferSize || view_rotate) + filter = FILTER_NONE; + +- unsigned int y_shift = yuv_type; ++ unsigned int y_shift = yuv_type == YV12 ? 1 : 0; // Diagram showing origin and direction of source sampling. // ->0 4<- // 7 3 // // 6 5 // ->1 2<- // Rotations that start at right side of image. if ((view_rotate == ROTATE_180) || -@@ -126,42 +139,57 @@ void ScaleYUVToRGB32(const uint8* y_buf, - - for (int y = 0; y < scaled_height; ++y) { - uint8* dest_pixel = rgb_buf + y * rgb_pitch; - int scaled_y = (y * height / scaled_height); - const uint8* y_ptr = y_buf + scaled_y * y_pitch; - const uint8* u_ptr = u_buf + (scaled_y >> y_shift) * uv_pitch; - const uint8* v_ptr = v_buf + (scaled_y >> y_shift) * uv_pitch; - --#if USE_MMX && defined(_MSC_VER) -+#if defined(_MSC_VER) - if (scaled_width == (width * 2)) { - DoubleYUVToRGB32Row(y_ptr, u_ptr, v_ptr, - dest_pixel, scaled_width); - } else if ((scaled_dx & 15) == 0) { // Scaling by integer scale factor. - if (scaled_dx_uv == scaled_dx) { // Not rotated. - if (scaled_dx == 16) { // Not scaled -- FastConvertYUVToRGB32Row(y_ptr, u_ptr, v_ptr, -- dest_pixel, scaled_width); -+ if (has_mmx) -+ FastConvertYUVToRGB32Row(y_ptr, u_ptr, v_ptr, -+ dest_pixel, scaled_width); -+ else -+ FastConvertYUVToRGB32Row_C(y_ptr, u_ptr, v_ptr, -+ dest_pixel, scaled_width); - } else { // Simple scale down. ie half - ConvertYUVToRGB32Row(y_ptr, u_ptr, v_ptr, - dest_pixel, scaled_width, scaled_dx >> 4); - } - } else { - RotateConvertYUVToRGB32Row(y_ptr, u_ptr, v_ptr, - dest_pixel, scaled_width, - scaled_dx >> 4, scaled_dx_uv >> 4); - } - #else - if (scaled_dx == 16) { // Not scaled -- FastConvertYUVToRGB32Row(y_ptr, u_ptr, v_ptr, -- dest_pixel, scaled_width); -+ if (has_mmx) -+ FastConvertYUVToRGB32Row(y_ptr, u_ptr, v_ptr, -+ dest_pixel, scaled_width); -+ else -+ FastConvertYUVToRGB32Row_C(y_ptr, u_ptr, v_ptr, -+ dest_pixel, scaled_width); - #endif - } else { -- ScaleYUVToRGB32Row(y_ptr, u_ptr, v_ptr, -- dest_pixel, scaled_width, scaled_dx); -- } -+ if (has_mmx) -+ ScaleYUVToRGB32Row(y_ptr, u_ptr, v_ptr, -+ dest_pixel, scaled_width, scaled_dx); -+ else -+ ScaleYUVToRGB32Row_C(y_ptr, u_ptr, v_ptr, -+ dest_pixel, scaled_width, scaled_dx); -+ -+ } +@@ -243,17 +271,17 @@ void ScaleYUVToRGB32(const uint8* y_buf, + uv_pitch = 1; + } } - // MMX used for FastConvertYUVToRGB32Row requires emms instruction. + // Need padding because FilterRows() will write 1 to 16 extra pixels + // after the end for SSE2 version. + uint8 yuvbuf[16 + kFilterBufferSize * 3 + 16]; + uint8* ybuf = +- reinterpret_cast<uint8*>(reinterpret_cast<uintptr_t>(yuvbuf + 15) & ~15); ++ reinterpret_cast<uint8*>(reinterpret_cast<PRUptrdiff>(yuvbuf + 15) & ~15); + uint8* ubuf = ybuf + kFilterBufferSize; + uint8* vbuf = ubuf + kFilterBufferSize; + // TODO(fbarchard): Fixed point math is off by 1 on negatives. + int yscale_fixed = (source_height << kFractionBits) / height; + + // TODO(fbarchard): Split this into separate function for better efficiency. + for (int y = 0; y < height; ++y) { + uint8* dest_pixel = rgb_buf + y * rgb_pitch; +@@ -276,17 +304,17 @@ void ScaleYUVToRGB32(const uint8* y_buf, + int source_uv_fraction = + ((source_y_subpixel >> y_shift) & kFractionMask) >> 8; + + const uint8* y_ptr = y0_ptr; + const uint8* u_ptr = u0_ptr; + const uint8* v_ptr = v0_ptr; + // Apply vertical filtering if necessary. + // TODO(fbarchard): Remove memcpy when not necessary. +- if (filter & media::FILTER_BILINEAR_V) { ++ if (filter & mozilla::gfx::FILTER_BILINEAR_V) { + if (yscale_fixed != kFractionMax && + source_y_fraction && ((source_y + 1) < source_height)) { + FilterRows(ybuf, y0_ptr, y1_ptr, source_width, source_y_fraction); + } else { + memcpy(ybuf, y0_ptr, source_width); + } + y_ptr = ybuf; + ybuf[source_width] = ybuf[source_width-1]; +@@ -309,17 +337,17 @@ void ScaleYUVToRGB32(const uint8* y_buf, + FastConvertYUVToRGB32Row(y_ptr, u_ptr, v_ptr, + dest_pixel, width); + } else { + if (filter & FILTER_BILINEAR_H) { + LinearScaleYUVToRGB32Row(y_ptr, u_ptr, v_ptr, + dest_pixel, width, source_dx); + } else { + // Specialized scalers and rotation. +-#if USE_MMX && defined(_MSC_VER) ++#if defined(_MSC_VER) && defined(_M_IX86) + if (width == (source_width * 2)) { + DoubleYUVToRGB32Row(y_ptr, u_ptr, v_ptr, + dest_pixel, width); + } else if ((source_dx & kFractionMask) == 0) { + // Scaling by integer scale factor. ie half. + ConvertYUVToRGB32Row(y_ptr, u_ptr, v_ptr, + dest_pixel, width, + source_dx >> kFractionBits); +@@ -331,16 +359,18 @@ void ScaleYUVToRGB32(const uint8* y_buf, + dest_pixel, width, + source_dx >> kFractionBits, + source_dx_uv >> kFractionBits); + } + #else + ScaleYUVToRGB32Row(y_ptr, u_ptr, v_ptr, + dest_pixel, width, source_dx); + #endif +- } ++ } + } + } + // MMX used for FastConvertYUVToRGB32Row and FilterRows requires emms. - EMMS(); + if (has_mmx) + EMMS(); } -} // namespace media +} // namespace gfx +} // namespace mozilla diff --git b/gfx/ycbcr/yuv_convert.h a/gfx/ycbcr/yuv_convert.h -index a757070..9d148a6 100644 +index 24a2c4e..eb99903 100644 --- b/gfx/ycbcr/yuv_convert.h +++ a/gfx/ycbcr/yuv_convert.h -@@ -1,63 +1,66 @@ - // Copyright (c) 2009 The Chromium Authors. All rights reserved. +@@ -1,72 +1,79 @@ + // 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. #ifndef MEDIA_BASE_YUV_CONVERT_H_ #define MEDIA_BASE_YUV_CONVERT_H_ -#include "base/basictypes.h" +- +-namespace media { +#include "chromium_types.h" - --namespace media { ++#include "gfxCore.h" ++ +namespace mozilla { -+ + +namespace gfx { - ++ // Type of YUV surface. // The value of these enums matter as they are used to shift vertical indices. enum YUVType { - YV16 = 0, // YV16 is half width and full height chroma channels. +- YV16 = 0, // YV16 is half width and full height chroma channels. - YV12 = 1, // YV12 is half width and half height chroma channels. -+ YV12 = 1 // YV12 is half width and half height chroma channels. ++ YV12 = 0, // YV12 is half width and half height chroma channels. ++ YV16 = 1, // YV16 is half width and full height chroma channels. ++ YV24 = 2 // YV24 is full width and full height chroma channels. }; // Mirror means flip the image horizontally, as in looking in a mirror. // Rotate happens after mirroring. enum Rotate { ROTATE_0, // Rotation off. ROTATE_90, // Rotate clockwise. ROTATE_180, // Rotate upside down. ROTATE_270, // Rotate counter clockwise. MIRROR_ROTATE_0, // Mirror horizontally. MIRROR_ROTATE_90, // Mirror then Rotate clockwise. MIRROR_ROTATE_180, // Mirror vertically. - MIRROR_ROTATE_270, // Transpose. + MIRROR_ROTATE_270 // Transpose. }; + // Filter affects how scaling looks. + enum ScaleFilter { + FILTER_NONE = 0, // No filter (point sampled). + FILTER_BILINEAR_H = 1, // Bilinear horizontal filter. + FILTER_BILINEAR_V = 2, // Bilinear vertical filter. +- FILTER_BILINEAR = 3, // Bilinear filter. ++ FILTER_BILINEAR = 3 // Bilinear filter. + }; + // Convert a frame of YUV to 32 bit ARGB. // Pass in YV16/YV12 depending on source format -void ConvertYUVToRGB32(const uint8* yplane, -+void ConvertYCbCrToRGB32(const uint8* yplane, -+ const uint8* uplane, -+ const uint8* vplane, -+ uint8* rgbframe, -+ int frame_width, -+ int frame_height, -+ int ystride, -+ int uvstride, -+ int rgbstride, -+ YUVType yuv_type); -+ -+// Scale a frame of YUV to 32 bit ARGB. -+// Supports rotation and mirroring. -+void ScaleYCbCrToRGB32(const uint8* yplane, - const uint8* uplane, - const uint8* vplane, - uint8* rgbframe, - int frame_width, - int frame_height, -+ int scaled_width, -+ int scaled_height, - int ystride, - int uvstride, - int rgbstride, +- const uint8* uplane, +- const uint8* vplane, +- uint8* rgbframe, +- int width, +- int height, +- int ystride, +- int uvstride, +- int rgbstride, - YUVType yuv_type); -+ YUVType yuv_type, -+ Rotate view_rotate); ++NS_GFX_(void) ConvertYCbCrToRGB32(const uint8* yplane, ++ const uint8* uplane, ++ const uint8* vplane, ++ uint8* rgbframe, ++ int pic_x, ++ int pic_y, ++ int pic_width, ++ int pic_height, ++ int ystride, ++ int uvstride, ++ int rgbstride, ++ YUVType yuv_type); --// Scale a frame of YUV to 32 bit ARGB. --// Supports rotation and mirroring. + // Scale a frame of YUV to 32 bit ARGB. + // Supports rotation and mirroring. -void ScaleYUVToRGB32(const uint8* yplane, - const uint8* uplane, - const uint8* vplane, - uint8* rgbframe, -- int frame_width, -- int frame_height, -- int scaled_width, -- int scaled_height, +- int source_width, +- int source_height, +- int width, +- int height, - int ystride, - int uvstride, - int rgbstride, - YUVType yuv_type, -- Rotate view_rotate); +- Rotate view_rotate, +- ScaleFilter filter); - -} // namespace media ++NS_GFX_(void) ScaleYCbCrToRGB32(const uint8* yplane, ++ const uint8* uplane, ++ const uint8* vplane, ++ uint8* rgbframe, ++ int source_width, ++ int source_height, ++ int width, ++ int height, ++ int ystride, ++ int uvstride, ++ int rgbstride, ++ YUVType yuv_type, ++ Rotate view_rotate, ++ ScaleFilter filter); + +} // namespace gfx +} // namespace mozilla - ++ #endif // MEDIA_BASE_YUV_CONVERT_H_ diff --git b/gfx/ycbcr/yuv_row.h a/gfx/ycbcr/yuv_row.h -index ac5c6fd..98efca6 100644 +index 0a2990b..4ce9eb8 100644 --- b/gfx/ycbcr/yuv_row.h +++ a/gfx/ycbcr/yuv_row.h -@@ -5,27 +5,34 @@ +@@ -5,27 +5,40 @@ // yuv_row internal functions to handle YUV conversion and scaling to RGB. // These functions are used from both yuv_convert.cc and yuv_scale.cc. // TODO(fbarchard): Write function that can handle rotation and scaling. #ifndef MEDIA_BASE_YUV_ROW_H_ #define MEDIA_BASE_YUV_ROW_H_ @@ -301,319 +447,144 @@ index ac5c6fd..98efca6 100644 const uint8* v_buf, uint8* rgb_buf, int width); +void FastConvertYUVToRGB32Row_C(const uint8* y_buf, + const uint8* u_buf, + const uint8* v_buf, + uint8* rgb_buf, -+ int width); ++ int width, ++ unsigned int x_shift); + ++void FastConvertYUVToRGB32Row(const uint8* y_buf, ++ const uint8* u_buf, ++ const uint8* v_buf, ++ uint8* rgb_buf, ++ int width); + // Can do 1x, half size or any scale down by an integer amount. // Step can be negative (mirroring, rotate 180). // This is the third fastest of the scalers. void ConvertYUVToRGB32Row(const uint8* y_buf, const uint8* u_buf, const uint8* v_buf, uint8* rgb_buf, int width, -@@ -55,29 +62,28 @@ void DoubleYUVToRGB32Row(const uint8* y_buf, +@@ -55,59 +68,67 @@ void DoubleYUVToRGB32Row(const uint8* y_buf, // Chroma is under sampled every 2 pixels for performance. - // This is the slowest of the scalers. void ScaleYUVToRGB32Row(const uint8* y_buf, const uint8* u_buf, const uint8* v_buf, uint8* rgb_buf, int width, - int scaled_dx); --} // extern "C" + int source_dx); + ++void ScaleYUVToRGB32Row(const uint8* y_buf, ++ const uint8* u_buf, ++ const uint8* v_buf, ++ uint8* rgb_buf, ++ int width, ++ int source_dx); ++ ++void ScaleYUVToRGB32Row_C(const uint8* y_buf, ++ const uint8* u_buf, ++ const uint8* v_buf, ++ uint8* rgb_buf, ++ int width, ++ int source_dx); ++ + // Handles arbitrary scaling up or down with bilinear filtering. + // Mirroring is supported, but not 90 or 270 degree rotation. + // Chroma is under sampled every 2 pixels for performance. + // This is the slowest of the scalers. + void LinearScaleYUVToRGB32Row(const uint8* y_buf, + const uint8* u_buf, + const uint8* v_buf, + uint8* rgb_buf, + int width, + int source_dx); ++void LinearScaleYUVToRGB32Row(const uint8* y_buf, ++ const uint8* u_buf, ++ const uint8* v_buf, ++ uint8* rgb_buf, ++ int width, ++ int source_dx); ++ ++void LinearScaleYUVToRGB32Row_C(const uint8* y_buf, ++ const uint8* u_buf, ++ const uint8* v_buf, ++ uint8* rgb_buf, ++ int width, ++ int source_dx); ++ ++ + #if defined(_MSC_VER) + #define SIMD_ALIGNED(var) __declspec(align(16)) var + #else + #define SIMD_ALIGNED(var) var __attribute__((aligned(16))) + #endif + extern SIMD_ALIGNED(int16 kCoefficientsRgbY[768][4]); + +-// Method to force C version. +-//#define USE_MMX 0 +-//#define USE_SSE2 0 +- -#if !defined(USE_MMX) --// Windows, Mac and Linux use MMX --#if defined(ARCH_CPU_X86) || (defined(ARCH_CPU_X86_64) && defined(OS_LINUX)) +-// Windows, Mac and Linux/BSD use MMX +-#if defined(__MMX__) || defined(_MSC_VER) -#define USE_MMX 1 -#else -#define USE_MMX 0 -#endif -#endif -+void ScaleYUVToRGB32Row_C(const uint8* y_buf, -+ const uint8* u_buf, -+ const uint8* v_buf, -+ uint8* rgb_buf, -+ int width, -+ int scaled_dx); -+ -+} // extern "C" - +- +-#if !defined(USE_SSE2) +-#if defined(__SSE2__) || defined(ARCH_CPU_X86_64) || _M_IX86_FP==2 +-#define USE_SSE2 1 +-#else +-#define USE_SSE2 0 +-#endif +-#endif +- // x64 uses MMX2 (SSE) so emms is not required. + // Warning C4799: function has no EMMS instruction. + // EMMS() is slow and should be called by the calling function once per image. -#if USE_MMX && !defined(ARCH_CPU_X86_64) +#if !defined(ARCH_CPU_X86_64) #if defined(_MSC_VER) #define EMMS() __asm emms + #pragma warning(disable: 4799) #else #define EMMS() asm("emms") #endif #else #define EMMS() - #endif diff --git b/gfx/ycbcr/yuv_row_c.cpp a/gfx/ycbcr/yuv_row_c.cpp -index 4a20777..a81416c 100644 +index a66fa7b..d327f85 100644 --- b/gfx/ycbcr/yuv_row_c.cpp +++ a/gfx/ycbcr/yuv_row_c.cpp -@@ -1,522 +1,20 @@ - // Copyright (c) 2009 The Chromium Authors. All rights reserved. +@@ -1,812 +1,18 @@ + // 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" +#include "yuv_row.h" -#ifdef _DEBUG -#include "base/logging.h" -#else #define DCHECK(a) -#endif - // TODO(fbarchard): Move this to yuv_row_posix.cc to share with Mac. - // TODO(fbarchard): Do 64 bit version. + extern "C" { - extern "C" { --#if USE_MMX -- --#define RGBY(i) { \ -- static_cast<int16>(1.164 * 64 * (i - 16) + 0.5), \ -- static_cast<int16>(1.164 * 64 * (i - 16) + 0.5), \ -- static_cast<int16>(1.164 * 64 * (i - 16) + 0.5), \ -- 0 \ --} -- --#define RGBU(i) { \ -- static_cast<int16>(2.018 * 64 * (i - 128) + 0.5), \ -- static_cast<int16>(-0.391 * 64 * (i - 128) + 0.5), \ -- 0, \ -- static_cast<int16>(256 * 64 - 1) \ --} -- --#define RGBV(i) { \ -- 0, \ -- static_cast<int16>(-0.813 * 64 * (i - 128) + 0.5), \ -- static_cast<int16>(1.596 * 64 * (i - 128) + 0.5), \ -- 0 \ --} -- --#define MMX_ALIGNED(var) var __attribute__((aligned(16))) -- -- --MMX_ALIGNED(int16 kCoefficientsRgbY[768][4]) = { -- RGBY(0x00), RGBY(0x01), RGBY(0x02), RGBY(0x03), -- RGBY(0x04), RGBY(0x05), RGBY(0x06), RGBY(0x07), -- RGBY(0x08), RGBY(0x09), RGBY(0x0A), RGBY(0x0B), -- RGBY(0x0C), RGBY(0x0D), RGBY(0x0E), RGBY(0x0F), -- RGBY(0x10), RGBY(0x11), RGBY(0x12), RGBY(0x13), -- RGBY(0x14), RGBY(0x15), RGBY(0x16), RGBY(0x17), -- RGBY(0x18), RGBY(0x19), RGBY(0x1A), RGBY(0x1B), -- RGBY(0x1C), RGBY(0x1D), RGBY(0x1E), RGBY(0x1F), -- RGBY(0x20), RGBY(0x21), RGBY(0x22), RGBY(0x23), -- RGBY(0x24), RGBY(0x25), RGBY(0x26), RGBY(0x27), -- RGBY(0x28), RGBY(0x29), RGBY(0x2A), RGBY(0x2B), -- RGBY(0x2C), RGBY(0x2D), RGBY(0x2E), RGBY(0x2F), -- RGBY(0x30), RGBY(0x31), RGBY(0x32), RGBY(0x33), -- RGBY(0x34), RGBY(0x35), RGBY(0x36), RGBY(0x37), -- RGBY(0x38), RGBY(0x39), RGBY(0x3A), RGBY(0x3B), -- RGBY(0x3C), RGBY(0x3D), RGBY(0x3E), RGBY(0x3F), -- RGBY(0x40), RGBY(0x41), RGBY(0x42), RGBY(0x43), -- RGBY(0x44), RGBY(0x45), RGBY(0x46), RGBY(0x47), -- RGBY(0x48), RGBY(0x49), RGBY(0x4A), RGBY(0x4B), -- RGBY(0x4C), RGBY(0x4D), RGBY(0x4E), RGBY(0x4F), -- RGBY(0x50), RGBY(0x51), RGBY(0x52), RGBY(0x53), -- RGBY(0x54), RGBY(0x55), RGBY(0x56), RGBY(0x57), -- RGBY(0x58), RGBY(0x59), RGBY(0x5A), RGBY(0x5B), -- RGBY(0x5C), RGBY(0x5D), RGBY(0x5E), RGBY(0x5F), -- RGBY(0x60), RGBY(0x61), RGBY(0x62), RGBY(0x63), -- RGBY(0x64), RGBY(0x65), RGBY(0x66), RGBY(0x67), -- RGBY(0x68), RGBY(0x69), RGBY(0x6A), RGBY(0x6B), -- RGBY(0x6C), RGBY(0x6D), RGBY(0x6E), RGBY(0x6F), -- RGBY(0x70), RGBY(0x71), RGBY(0x72), RGBY(0x73), -- RGBY(0x74), RGBY(0x75), RGBY(0x76), RGBY(0x77), -- RGBY(0x78), RGBY(0x79), RGBY(0x7A), RGBY(0x7B), -- RGBY(0x7C), RGBY(0x7D), RGBY(0x7E), RGBY(0x7F), -- RGBY(0x80), RGBY(0x81), RGBY(0x82), RGBY(0x83), -- RGBY(0x84), RGBY(0x85), RGBY(0x86), RGBY(0x87), -- RGBY(0x88), RGBY(0x89), RGBY(0x8A), RGBY(0x8B), -- RGBY(0x8C), RGBY(0x8D), RGBY(0x8E), RGBY(0x8F), -- RGBY(0x90), RGBY(0x91), RGBY(0x92), RGBY(0x93), -- RGBY(0x94), RGBY(0x95), RGBY(0x96), RGBY(0x97), -- RGBY(0x98), RGBY(0x99), RGBY(0x9A), RGBY(0x9B), -- RGBY(0x9C), RGBY(0x9D), RGBY(0x9E), RGBY(0x9F), -- RGBY(0xA0), RGBY(0xA1), RGBY(0xA2), RGBY(0xA3), -- RGBY(0xA4), RGBY(0xA5), RGBY(0xA6), RGBY(0xA7), -- RGBY(0xA8), RGBY(0xA9), RGBY(0xAA), RGBY(0xAB), -- RGBY(0xAC), RGBY(0xAD), RGBY(0xAE), RGBY(0xAF), -- RGBY(0xB0), RGBY(0xB1), RGBY(0xB2), RGBY(0xB3), -- RGBY(0xB4), RGBY(0xB5), RGBY(0xB6), RGBY(0xB7), -- RGBY(0xB8), RGBY(0xB9), RGBY(0xBA), RGBY(0xBB), -- RGBY(0xBC), RGBY(0xBD), RGBY(0xBE), RGBY(0xBF), -- RGBY(0xC0), RGBY(0xC1), RGBY(0xC2), RGBY(0xC3), -- RGBY(0xC4), RGBY(0xC5), RGBY(0xC6), RGBY(0xC7), -- RGBY(0xC8), RGBY(0xC9), RGBY(0xCA), RGBY(0xCB), -- RGBY(0xCC), RGBY(0xCD), RGBY(0xCE), RGBY(0xCF), -- RGBY(0xD0), RGBY(0xD1), RGBY(0xD2), RGBY(0xD3), -- RGBY(0xD4), RGBY(0xD5), RGBY(0xD6), RGBY(0xD7), -- RGBY(0xD8), RGBY(0xD9), RGBY(0xDA), RGBY(0xDB), -- RGBY(0xDC), RGBY(0xDD), RGBY(0xDE), RGBY(0xDF), -- RGBY(0xE0), RGBY(0xE1), RGBY(0xE2), RGBY(0xE3), -- RGBY(0xE4), RGBY(0xE5), RGBY(0xE6), RGBY(0xE7), -- RGBY(0xE8), RGBY(0xE9), RGBY(0xEA), RGBY(0xEB), -- RGBY(0xEC), RGBY(0xED), RGBY(0xEE), RGBY(0xEF), -- RGBY(0xF0), RGBY(0xF1), RGBY(0xF2), RGBY(0xF3), -- RGBY(0xF4), RGBY(0xF5), RGBY(0xF6), RGBY(0xF7), -- RGBY(0xF8), RGBY(0xF9), RGBY(0xFA), RGBY(0xFB), -- RGBY(0xFC), RGBY(0xFD), RGBY(0xFE), RGBY(0xFF), -- -- // Chroma U table. -- RGBU(0x00), RGBU(0x01), RGBU(0x02), RGBU(0x03), -- RGBU(0x04), RGBU(0x05), RGBU(0x06), RGBU(0x07), -- RGBU(0x08), RGBU(0x09), RGBU(0x0A), RGBU(0x0B), -- RGBU(0x0C), RGBU(0x0D), RGBU(0x0E), RGBU(0x0F), -- RGBU(0x10), RGBU(0x11), RGBU(0x12), RGBU(0x13), -- RGBU(0x14), RGBU(0x15), RGBU(0x16), RGBU(0x17), -- RGBU(0x18), RGBU(0x19), RGBU(0x1A), RGBU(0x1B), -- RGBU(0x1C), RGBU(0x1D), RGBU(0x1E), RGBU(0x1F), -- RGBU(0x20), RGBU(0x21), RGBU(0x22), RGBU(0x23), -- RGBU(0x24), RGBU(0x25), RGBU(0x26), RGBU(0x27), -- RGBU(0x28), RGBU(0x29), RGBU(0x2A), RGBU(0x2B), -- RGBU(0x2C), RGBU(0x2D), RGBU(0x2E), RGBU(0x2F), -- RGBU(0x30), RGBU(0x31), RGBU(0x32), RGBU(0x33), -- RGBU(0x34), RGBU(0x35), RGBU(0x36), RGBU(0x37), -- RGBU(0x38), RGBU(0x39), RGBU(0x3A), RGBU(0x3B), -- RGBU(0x3C), RGBU(0x3D), RGBU(0x3E), RGBU(0x3F), -- RGBU(0x40), RGBU(0x41), RGBU(0x42), RGBU(0x43), -- RGBU(0x44), RGBU(0x45), RGBU(0x46), RGBU(0x47), -- RGBU(0x48), RGBU(0x49), RGBU(0x4A), RGBU(0x4B), -- RGBU(0x4C), RGBU(0x4D), RGBU(0x4E), RGBU(0x4F), -- RGBU(0x50), RGBU(0x51), RGBU(0x52), RGBU(0x53), -- RGBU(0x54), RGBU(0x55), RGBU(0x56), RGBU(0x57), -- RGBU(0x58), RGBU(0x59), RGBU(0x5A), RGBU(0x5B), -- RGBU(0x5C), RGBU(0x5D), RGBU(0x5E), RGBU(0x5F), -- RGBU(0x60), RGBU(0x61), RGBU(0x62), RGBU(0x63), -- RGBU(0x64), RGBU(0x65), RGBU(0x66), RGBU(0x67), -- RGBU(0x68), RGBU(0x69), RGBU(0x6A), RGBU(0x6B), -- RGBU(0x6C), RGBU(0x6D), RGBU(0x6E), RGBU(0x6F), -- RGBU(0x70), RGBU(0x71), RGBU(0x72), RGBU(0x73), -- RGBU(0x74), RGBU(0x75), RGBU(0x76), RGBU(0x77), -- RGBU(0x78), RGBU(0x79), RGBU(0x7A), RGBU(0x7B), -- RGBU(0x7C), RGBU(0x7D), RGBU(0x7E), RGBU(0x7F), -- RGBU(0x80), RGBU(0x81), RGBU(0x82), RGBU(0x83), -- RGBU(0x84), RGBU(0x85), RGBU(0x86), RGBU(0x87), -- RGBU(0x88), RGBU(0x89), RGBU(0x8A), RGBU(0x8B), -- RGBU(0x8C), RGBU(0x8D), RGBU(0x8E), RGBU(0x8F), -- RGBU(0x90), RGBU(0x91), RGBU(0x92), RGBU(0x93), -- RGBU(0x94), RGBU(0x95), RGBU(0x96), RGBU(0x97), -- RGBU(0x98), RGBU(0x99), RGBU(0x9A), RGBU(0x9B), -- RGBU(0x9C), RGBU(0x9D), RGBU(0x9E), RGBU(0x9F), -- RGBU(0xA0), RGBU(0xA1), RGBU(0xA2), RGBU(0xA3), -- RGBU(0xA4), RGBU(0xA5), RGBU(0xA6), RGBU(0xA7), -- RGBU(0xA8), RGBU(0xA9), RGBU(0xAA), RGBU(0xAB), -- RGBU(0xAC), RGBU(0xAD), RGBU(0xAE), RGBU(0xAF), -- RGBU(0xB0), RGBU(0xB1), RGBU(0xB2), RGBU(0xB3), -- RGBU(0xB4), RGBU(0xB5), RGBU(0xB6), RGBU(0xB7), -- RGBU(0xB8), RGBU(0xB9), RGBU(0xBA), RGBU(0xBB), -- RGBU(0xBC), RGBU(0xBD), RGBU(0xBE), RGBU(0xBF), -- RGBU(0xC0), RGBU(0xC1), RGBU(0xC2), RGBU(0xC3), -- RGBU(0xC4), RGBU(0xC5), RGBU(0xC6), RGBU(0xC7), -- RGBU(0xC8), RGBU(0xC9), RGBU(0xCA), RGBU(0xCB), -- RGBU(0xCC), RGBU(0xCD), RGBU(0xCE), RGBU(0xCF), -- RGBU(0xD0), RGBU(0xD1), RGBU(0xD2), RGBU(0xD3), -- RGBU(0xD4), RGBU(0xD5), RGBU(0xD6), RGBU(0xD7), -- RGBU(0xD8), RGBU(0xD9), RGBU(0xDA), RGBU(0xDB), -- RGBU(0xDC), RGBU(0xDD), RGBU(0xDE), RGBU(0xDF), -- RGBU(0xE0), RGBU(0xE1), RGBU(0xE2), RGBU(0xE3), -- RGBU(0xE4), RGBU(0xE5), RGBU(0xE6), RGBU(0xE7), -- RGBU(0xE8), RGBU(0xE9), RGBU(0xEA), RGBU(0xEB), -- RGBU(0xEC), RGBU(0xED), RGBU(0xEE), RGBU(0xEF), -- RGBU(0xF0), RGBU(0xF1), RGBU(0xF2), RGBU(0xF3), -- RGBU(0xF4), RGBU(0xF5), RGBU(0xF6), RGBU(0xF7), -- RGBU(0xF8), RGBU(0xF9), RGBU(0xFA), RGBU(0xFB), -- RGBU(0xFC), RGBU(0xFD), RGBU(0xFE), RGBU(0xFF), -- -- // Chroma V table. -- RGBV(0x00), RGBV(0x01), RGBV(0x02), RGBV(0x03), -- RGBV(0x04), RGBV(0x05), RGBV(0x06), RGBV(0x07), -- RGBV(0x08), RGBV(0x09), RGBV(0x0A), RGBV(0x0B), -- RGBV(0x0C), RGBV(0x0D), RGBV(0x0E), RGBV(0x0F), -- RGBV(0x10), RGBV(0x11), RGBV(0x12), RGBV(0x13), -- RGBV(0x14), RGBV(0x15), RGBV(0x16), RGBV(0x17), -- RGBV(0x18), RGBV(0x19), RGBV(0x1A), RGBV(0x1B), -- RGBV(0x1C), RGBV(0x1D), RGBV(0x1E), RGBV(0x1F), -- RGBV(0x20), RGBV(0x21), RGBV(0x22), RGBV(0x23), -- RGBV(0x24), RGBV(0x25), RGBV(0x26), RGBV(0x27), -- RGBV(0x28), RGBV(0x29), RGBV(0x2A), RGBV(0x2B), -- RGBV(0x2C), RGBV(0x2D), RGBV(0x2E), RGBV(0x2F), -- RGBV(0x30), RGBV(0x31), RGBV(0x32), RGBV(0x33), -- RGBV(0x34), RGBV(0x35), RGBV(0x36), RGBV(0x37), -- RGBV(0x38), RGBV(0x39), RGBV(0x3A), RGBV(0x3B), -- RGBV(0x3C), RGBV(0x3D), RGBV(0x3E), RGBV(0x3F), -- RGBV(0x40), RGBV(0x41), RGBV(0x42), RGBV(0x43), -- RGBV(0x44), RGBV(0x45), RGBV(0x46), RGBV(0x47), -- RGBV(0x48), RGBV(0x49), RGBV(0x4A), RGBV(0x4B), -- RGBV(0x4C), RGBV(0x4D), RGBV(0x4E), RGBV(0x4F), -- RGBV(0x50), RGBV(0x51), RGBV(0x52), RGBV(0x53), -- RGBV(0x54), RGBV(0x55), RGBV(0x56), RGBV(0x57), -- RGBV(0x58), RGBV(0x59), RGBV(0x5A), RGBV(0x5B), -- RGBV(0x5C), RGBV(0x5D), RGBV(0x5E), RGBV(0x5F), -- RGBV(0x60), RGBV(0x61), RGBV(0x62), RGBV(0x63), -- RGBV(0x64), RGBV(0x65), RGBV(0x66), RGBV(0x67), -- RGBV(0x68), RGBV(0x69), RGBV(0x6A), RGBV(0x6B), -- RGBV(0x6C), RGBV(0x6D), RGBV(0x6E), RGBV(0x6F), -- RGBV(0x70), RGBV(0x71), RGBV(0x72), RGBV(0x73), -- RGBV(0x74), RGBV(0x75), RGBV(0x76), RGBV(0x77), -- RGBV(0x78), RGBV(0x79), RGBV(0x7A), RGBV(0x7B), -- RGBV(0x7C), RGBV(0x7D), RGBV(0x7E), RGBV(0x7F), -- RGBV(0x80), RGBV(0x81), RGBV(0x82), RGBV(0x83), -- RGBV(0x84), RGBV(0x85), RGBV(0x86), RGBV(0x87), -- RGBV(0x88), RGBV(0x89), RGBV(0x8A), RGBV(0x8B), -- RGBV(0x8C), RGBV(0x8D), RGBV(0x8E), RGBV(0x8F), -- RGBV(0x90), RGBV(0x91), RGBV(0x92), RGBV(0x93), -- RGBV(0x94), RGBV(0x95), RGBV(0x96), RGBV(0x97), -- RGBV(0x98), RGBV(0x99), RGBV(0x9A), RGBV(0x9B), -- RGBV(0x9C), RGBV(0x9D), RGBV(0x9E), RGBV(0x9F), -- RGBV(0xA0), RGBV(0xA1), RGBV(0xA2), RGBV(0xA3), -- RGBV(0xA4), RGBV(0xA5), RGBV(0xA6), RGBV(0xA7), -- RGBV(0xA8), RGBV(0xA9), RGBV(0xAA), RGBV(0xAB), -- RGBV(0xAC), RGBV(0xAD), RGBV(0xAE), RGBV(0xAF), -- RGBV(0xB0), RGBV(0xB1), RGBV(0xB2), RGBV(0xB3), -- RGBV(0xB4), RGBV(0xB5), RGBV(0xB6), RGBV(0xB7), -- RGBV(0xB8), RGBV(0xB9), RGBV(0xBA), RGBV(0xBB), -- RGBV(0xBC), RGBV(0xBD), RGBV(0xBE), RGBV(0xBF), -- RGBV(0xC0), RGBV(0xC1), RGBV(0xC2), RGBV(0xC3), -- RGBV(0xC4), RGBV(0xC5), RGBV(0xC6), RGBV(0xC7), -- RGBV(0xC8), RGBV(0xC9), RGBV(0xCA), RGBV(0xCB), -- RGBV(0xCC), RGBV(0xCD), RGBV(0xCE), RGBV(0xCF), -- RGBV(0xD0), RGBV(0xD1), RGBV(0xD2), RGBV(0xD3), -- RGBV(0xD4), RGBV(0xD5), RGBV(0xD6), RGBV(0xD7), -- RGBV(0xD8), RGBV(0xD9), RGBV(0xDA), RGBV(0xDB), -- RGBV(0xDC), RGBV(0xDD), RGBV(0xDE), RGBV(0xDF), -- RGBV(0xE0), RGBV(0xE1), RGBV(0xE2), RGBV(0xE3), -- RGBV(0xE4), RGBV(0xE5), RGBV(0xE6), RGBV(0xE7), -- RGBV(0xE8), RGBV(0xE9), RGBV(0xEA), RGBV(0xEB), -- RGBV(0xEC), RGBV(0xED), RGBV(0xEE), RGBV(0xEF), -- RGBV(0xF0), RGBV(0xF1), RGBV(0xF2), RGBV(0xF3), -- RGBV(0xF4), RGBV(0xF5), RGBV(0xF6), RGBV(0xF7), -- RGBV(0xF8), RGBV(0xF9), RGBV(0xFA), RGBV(0xFB), -- RGBV(0xFC), RGBV(0xFD), RGBV(0xFE), RGBV(0xFF), --}; -- --#undef RGBY --#undef RGBU --#undef RGBV --#undef MMX_ALIGNED -- --#if defined(ARCH_CPU_X86_64) +-#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( @@ -669,36 +640,36 @@ index 4a20777..a81416c 100644 -); -} - -void ScaleYUVToRGB32Row(const uint8* y_buf, // rdi - const uint8* u_buf, // rsi - const uint8* v_buf, // rdx - uint8* rgb_buf, // rcx - int width, // r8 -- int scaled_dx) { // r9 +- int source_dx) { // r9 - asm( - "xor %%r11,%%r11\n" - "sub $0x2,%4\n" - "js scalenext\n" - -"scaleloop:" - "mov %%r11,%%r10\n" -- "sar $0x5,%%r10\n" +- "sar $0x11,%%r10\n" - "movzb (%1,%%r10,1),%%rax\n" - "movq 2048(%5,%%rax,8),%%xmm0\n" - "movzb (%2,%%r10,1),%%rax\n" - "movq 4096(%5,%%rax,8),%%xmm1\n" - "lea (%%r11,%6),%%r10\n" -- "sar $0x4,%%r11\n" +- "sar $0x10,%%r11\n" - "movzb (%0,%%r11,1),%%rax\n" - "paddsw %%xmm1,%%xmm0\n" - "movq (%5,%%rax,8),%%xmm1\n" - "lea (%%r10,%6),%%r11\n" -- "sar $0x4,%%r10\n" +- "sar $0x10,%%r10\n" - "movzb (%0,%%r10,1),%%rax\n" - "movq (%5,%%rax,8),%%xmm2\n" - "paddsw %%xmm0,%%xmm1\n" - "paddsw %%xmm0,%%xmm2\n" - "shufps $0x44,%%xmm2,%%xmm1\n" - "psraw $0x6,%%xmm1\n" - "packuswb %%xmm1,%%xmm1\n" - "movq %%xmm1,0x0(%3)\n" @@ -706,52 +677,172 @@ index 4a20777..a81416c 100644 - "sub $0x2,%4\n" - "jns scaleloop\n" - -"scalenext:" - "add $0x1,%4\n" - "js scaledone\n" - - "mov %%r11,%%r10\n" -- "sar $0x5,%%r10\n" +- "sar $0x11,%%r10\n" - "movzb (%1,%%r10,1),%%rax\n" - "movq 2048(%5,%%rax,8),%%xmm0\n" - "movzb (%2,%%r10,1),%%rax\n" - "movq 4096(%5,%%rax,8),%%xmm1\n" - "paddsw %%xmm1,%%xmm0\n" -- "sar $0x4,%%r11\n" +- "sar $0x10,%%r11\n" - "movzb (%0,%%r11,1),%%rax\n" - "movq (%5,%%rax,8),%%xmm1\n" - "paddsw %%xmm0,%%xmm1\n" - "psraw $0x6,%%xmm1\n" - "packuswb %%xmm1,%%xmm1\n" - "movd %%xmm1,0x0(%3)\n" - -"scaledone:" - : - : "r"(y_buf), // %0 - "r"(u_buf), // %1 - "r"(v_buf), // %2 - "r"(rgb_buf), // %3 - "r"(width), // %4 - "r" (kCoefficientsRgbY), // %5 -- "r"(static_cast<long>(scaled_dx)) // %6 +- "r"(static_cast<long>(source_dx)) // %6 - : "memory", "r10", "r11", "rax", "xmm0", "xmm1", "xmm2" -); -} - --#else +-void LinearScaleYUVToRGB32Row(const uint8* y_buf, +- const uint8* u_buf, +- const uint8* v_buf, +- uint8* rgb_buf, +- int width, +- int source_dx) { +- asm( +- "xor %%r11,%%r11\n" // x = 0 +- "sub $0x2,%4\n" +- "js .lscalenext\n" +- "cmp $0x20000,%6\n" // if source_dx >= 2.0 +- "jl .lscalehalf\n" +- "mov $0x8000,%%r11\n" // x = 0.5 for 1/2 or less +-".lscalehalf:" +- +-".lscaleloop:" +- "mov %%r11,%%r10\n" +- "sar $0x11,%%r10\n" +- +- "movzb (%1, %%r10, 1), %%r13 \n" +- "movzb 1(%1, %%r10, 1), %%r14 \n" +- "mov %%r11, %%rax \n" +- "and $0x1fffe, %%rax \n" +- "imul %%rax, %%r14 \n" +- "xor $0x1fffe, %%rax \n" +- "imul %%rax, %%r13 \n" +- "add %%r14, %%r13 \n" +- "shr $17, %%r13 \n" +- "movq 2048(%5,%%r13,8), %%xmm0\n" +- +- "movzb (%2, %%r10, 1), %%r13 \n" +- "movzb 1(%2, %%r10, 1), %%r14 \n" +- "mov %%r11, %%rax \n" +- "and $0x1fffe, %%rax \n" +- "imul %%rax, %%r14 \n" +- "xor $0x1fffe, %%rax \n" +- "imul %%rax, %%r13 \n" +- "add %%r14, %%r13 \n" +- "shr $17, %%r13 \n" +- "movq 4096(%5,%%r13,8), %%xmm1\n" +- +- "mov %%r11, %%rax \n" +- "lea (%%r11,%6),%%r10\n" +- "sar $0x10,%%r11\n" +- "paddsw %%xmm1,%%xmm0\n" +- +- "movzb (%0, %%r11, 1), %%r13 \n" +- "movzb 1(%0, %%r11, 1), %%r14 \n" +- "and $0xffff, %%rax \n" +- "imul %%rax, %%r14 \n" +- "xor $0xffff, %%rax \n" +- "imul %%rax, %%r13 \n" +- "add %%r14, %%r13 \n" +- "shr $16, %%r13 \n" +- "movq (%5,%%r13,8),%%xmm1\n" +- +- "mov %%r10, %%rax \n" +- "lea (%%r10,%6),%%r11\n" +- "sar $0x10,%%r10\n" +- +- "movzb (%0,%%r10,1), %%r13 \n" +- "movzb 1(%0,%%r10,1), %%r14 \n" +- "and $0xffff, %%rax \n" +- "imul %%rax, %%r14 \n" +- "xor $0xffff, %%rax \n" +- "imul %%rax, %%r13 \n" +- "add %%r14, %%r13 \n" +- "shr $16, %%r13 \n" +- "movq (%5,%%r13,8),%%xmm2\n" +- +- "paddsw %%xmm0,%%xmm1\n" +- "paddsw %%xmm0,%%xmm2\n" +- "shufps $0x44,%%xmm2,%%xmm1\n" +- "psraw $0x6,%%xmm1\n" +- "packuswb %%xmm1,%%xmm1\n" +- "movq %%xmm1,0x0(%3)\n" +- "add $0x8,%3\n" +- "sub $0x2,%4\n" +- "jns .lscaleloop\n" +- +-".lscalenext:" +- "add $0x1,%4\n" +- "js .lscaledone\n" +- +- "mov %%r11,%%r10\n" +- "sar $0x11,%%r10\n" +- +- "movzb (%1,%%r10,1), %%r13 \n" +- "movq 2048(%5,%%r13,8),%%xmm0\n" +- +- "movzb (%2,%%r10,1), %%r13 \n" +- "movq 4096(%5,%%r13,8),%%xmm1\n" +- +- "paddsw %%xmm1,%%xmm0\n" +- "sar $0x10,%%r11\n" +- +- "movzb (%0,%%r11,1), %%r13 \n" +- "movq (%5,%%r13,8),%%xmm1\n" +- +- "paddsw %%xmm0,%%xmm1\n" +- "psraw $0x6,%%xmm1\n" +- "packuswb %%xmm1,%%xmm1\n" +- "movd %%xmm1,0x0(%3)\n" +- +-".lscaledone:" +- : +- : "r"(y_buf), // %0 +- "r"(u_buf), // %1 +- "r"(v_buf), // %2 +- "r"(rgb_buf), // %3 +- "r"(width), // %4 +- "r" (kCoefficientsRgbY), // %5 +- "r"(static_cast<long>(source_dx)) // %6 +- : "memory", "r10", "r11", "r13", "r14", "rax", "xmm0", "xmm1", "xmm2" +-); +-} +- +-#elif USE_MMX && !defined(ARCH_CPU_X86_64) && !defined(__PIC__) +- +-// PIC version is slower because less registers are available, so +-// non-PIC is used on platforms where it is possible. - -void FastConvertYUVToRGB32Row(const uint8* y_buf, - const uint8* u_buf, - const uint8* v_buf, - uint8* rgb_buf, - int width); -- - asm( +- ".text\n" - ".global FastConvertYUVToRGB32Row\n" -"FastConvertYUVToRGB32Row:\n" - "pusha\n" - "mov 0x24(%esp),%edx\n" - "mov 0x28(%esp),%edi\n" - "mov 0x2c(%esp),%esi\n" - "mov 0x30(%esp),%ebp\n" - "mov 0x34(%esp),%ecx\n" @@ -799,47 +890,47 @@ index 4a20777..a81416c 100644 -); - - -void ScaleYUVToRGB32Row(const uint8* y_buf, - const uint8* u_buf, - const uint8* v_buf, - uint8* rgb_buf, - int width, -- int scaled_dx); -- +- int source_dx); - asm( +- ".text\n" - ".global ScaleYUVToRGB32Row\n" -"ScaleYUVToRGB32Row:\n" - "pusha\n" - "mov 0x24(%esp),%edx\n" - "mov 0x28(%esp),%edi\n" - "mov 0x2c(%esp),%esi\n" - "mov 0x30(%esp),%ebp\n" - "mov 0x34(%esp),%ecx\n" - "xor %ebx,%ebx\n" - "jmp scaleend\n" - -"scaleloop:" - "mov %ebx,%eax\n" -- "sar $0x5,%eax\n" +- "sar $0x11,%eax\n" - "movzbl (%edi,%eax,1),%eax\n" - "movq kCoefficientsRgbY+2048(,%eax,8),%mm0\n" - "mov %ebx,%eax\n" -- "sar $0x5,%eax\n" +- "sar $0x11,%eax\n" - "movzbl (%esi,%eax,1),%eax\n" - "paddsw kCoefficientsRgbY+4096(,%eax,8),%mm0\n" - "mov %ebx,%eax\n" - "add 0x38(%esp),%ebx\n" -- "sar $0x4,%eax\n" +- "sar $0x10,%eax\n" - "movzbl (%edx,%eax,1),%eax\n" - "movq kCoefficientsRgbY(,%eax,8),%mm1\n" - "mov %ebx,%eax\n" - "add 0x38(%esp),%ebx\n" -- "sar $0x4,%eax\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" - "movntq %mm1,0x0(%ebp)\n" @@ -847,393 +938,1641 @@ index 4a20777..a81416c 100644 -"scaleend:" - "sub $0x2,%ecx\n" - "jns scaleloop\n" - - "and $0x1,%ecx\n" - "je scaledone\n" - - "mov %ebx,%eax\n" -- "sar $0x5,%eax\n" +- "sar $0x11,%eax\n" - "movzbl (%edi,%eax,1),%eax\n" - "movq kCoefficientsRgbY+2048(,%eax,8),%mm0\n" - "mov %ebx,%eax\n" -- "sar $0x5,%eax\n" +- "sar $0x11,%eax\n" - "movzbl (%esi,%eax,1),%eax\n" - "paddsw kCoefficientsRgbY+4096(,%eax,8),%mm0\n" - "mov %ebx,%eax\n" -- "sar $0x4,%eax\n" +- "sar $0x10,%eax\n" - "movzbl (%edx,%eax,1),%eax\n" - "movq kCoefficientsRgbY(,%eax,8),%mm1\n" - "paddsw %mm0,%mm1\n" - "psraw $0x6,%mm1\n" - "packuswb %mm1,%mm1\n" - "movd %mm1,0x0(%ebp)\n" - -"scaledone:" - "popa\n" - "ret\n" -); - +-void LinearScaleYUVToRGB32Row(const uint8* y_buf, +- const uint8* u_buf, +- const uint8* v_buf, +- uint8* rgb_buf, +- int width, +- int source_dx); +- asm( +- ".text\n" +- ".global LinearScaleYUVToRGB32Row\n" +-"LinearScaleYUVToRGB32Row:\n" +- "pusha\n" +- "mov 0x24(%esp),%edx\n" +- "mov 0x28(%esp),%edi\n" +- "mov 0x30(%esp),%ebp\n" +- +- // source_width = width * source_dx + ebx +- "mov 0x34(%esp), %ecx\n" +- "imull 0x38(%esp), %ecx\n" +- "mov %ecx, 0x34(%esp)\n" +- +- "mov 0x38(%esp), %ecx\n" +- "xor %ebx,%ebx\n" // x = 0 +- "cmp $0x20000,%ecx\n" // if source_dx >= 2.0 +- "jl .lscaleend\n" +- "mov $0x8000,%ebx\n" // x = 0.5 for 1/2 or less +- "jmp .lscaleend\n" +- +-".lscaleloop:" +- "mov %ebx,%eax\n" +- "sar $0x11,%eax\n" +- +- "movzbl (%edi,%eax,1),%ecx\n" +- "movzbl 1(%edi,%eax,1),%esi\n" +- "mov %ebx,%eax\n" +- "andl $0x1fffe, %eax \n" +- "imul %eax, %esi \n" +- "xorl $0x1fffe, %eax \n" +- "imul %eax, %ecx \n" +- "addl %esi, %ecx \n" +- "shrl $17, %ecx \n" +- "movq kCoefficientsRgbY+2048(,%ecx,8),%mm0\n" +- +- "mov 0x2c(%esp),%esi\n" +- "mov %ebx,%eax\n" +- "sar $0x11,%eax\n" +- +- "movzbl (%esi,%eax,1),%ecx\n" +- "movzbl 1(%esi,%eax,1),%esi\n" +- "mov %ebx,%eax\n" +- "andl $0x1fffe, %eax \n" +- "imul %eax, %esi \n" +- "xorl $0x1fffe, %eax \n" +- "imul %eax, %ecx \n" +- "addl %esi, %ecx \n" +- "shrl $17, %ecx \n" +- "paddsw kCoefficientsRgbY+4096(,%ecx,8),%mm0\n" +- +- "mov %ebx,%eax\n" +- "sar $0x10,%eax\n" +- "movzbl (%edx,%eax,1),%ecx\n" +- "movzbl 1(%edx,%eax,1),%esi\n" +- "mov %ebx,%eax\n" +- "add 0x38(%esp),%ebx\n" +- "andl $0xffff, %eax \n" +- "imul %eax, %esi \n" +- "xorl $0xffff, %eax \n" +- "imul %eax, %ecx \n" +- "addl %esi, %ecx \n" +- "shrl $16, %ecx \n" +- "movq kCoefficientsRgbY(,%ecx,8),%mm1\n" +- +- "cmp 0x34(%esp), %ebx\n" +- "jge .lscalelastpixel\n" +- +- "mov %ebx,%eax\n" +- "sar $0x10,%eax\n" +- "movzbl (%edx,%eax,1),%ecx\n" +- "movzbl 1(%edx,%eax,1),%esi\n" +- "mov %ebx,%eax\n" +- "add 0x38(%esp),%ebx\n" +- "andl $0xffff, %eax \n" +- "imul %eax, %esi \n" +- "xorl $0xffff, %eax \n" +- "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" +- "movntq %mm1,0x0(%ebp)\n" +- "add $0x8,%ebp\n" +- +-".lscaleend:" +- "cmp 0x34(%esp), %ebx\n" +- "jl .lscaleloop\n" +- "popa\n" +- "ret\n" +- +-".lscalelastpixel:" +- "paddsw %mm0, %mm1\n" +- "psraw $6, %mm1\n" +- "packuswb %mm1, %mm1\n" +- "movd %mm1, (%ebp)\n" +- "popa\n" +- "ret\n" +-); +- +-#elif USE_MMX && !defined(ARCH_CPU_X86_64) && defined(__PIC__) +- +-extern void PICConvertYUVToRGB32Row(const uint8* y_buf, +- const uint8* u_buf, +- const uint8* v_buf, +- uint8* rgb_buf, +- int width, +- int16 *kCoefficientsRgbY); +- asm( +- ".text\n" +-#if defined(OS_MACOSX) +-"_PICConvertYUVToRGB32Row:\n" +-#else +-"PICConvertYUVToRGB32Row:\n" -#endif +- "pusha\n" +- "mov 0x24(%esp),%edx\n" +- "mov 0x28(%esp),%edi\n" +- "mov 0x2c(%esp),%esi\n" +- "mov 0x30(%esp),%ebp\n" +- "mov 0x38(%esp),%ecx\n" +- +- "jmp .Lconvertend\n" +- +-".Lconvertloop:" +- "movzbl (%edi),%eax\n" +- "add $0x1,%edi\n" +- "movzbl (%esi),%ebx\n" +- "add $0x1,%esi\n" +- "movq 2048(%ecx,%eax,8),%mm0\n" +- "movzbl (%edx),%eax\n" +- "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" +- "movntq %mm1,0x0(%ebp)\n" +- "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" +- "movzbl (%esi),%eax\n" +- "paddsw 4096(%ecx,%eax,8),%mm0\n" +- "movzbl (%edx),%eax\n" +- "movq 0(%ecx,%eax,8),%mm1\n" +- "paddsw %mm0,%mm1\n" +- "psraw $0x6,%mm1\n" +- "packuswb %mm1,%mm1\n" +- "movd %mm1,0x0(%ebp)\n" +-".Lconvertdone:\n" +- "popa\n" +- "ret\n" +-); +- +-void FastConvertYUVToRGB32Row(const uint8* y_buf, +- const uint8* u_buf, +- const uint8* v_buf, +- uint8* rgb_buf, +- int width) { +- PICConvertYUVToRGB32Row(y_buf, u_buf, v_buf, rgb_buf, width, +- &kCoefficientsRgbY[0][0]); +-} +- +-extern void PICScaleYUVToRGB32Row(const uint8* y_buf, +- const uint8* u_buf, +- const uint8* v_buf, +- uint8* rgb_buf, +- int width, +- int source_dx, +- int16 *kCoefficientsRgbY); +- +- asm( +- ".text\n" +-#if defined(OS_MACOSX) +-"_PICScaleYUVToRGB32Row:\n" +-#else +-"PICScaleYUVToRGB32Row:\n" +-#endif +- "pusha\n" +- "mov 0x24(%esp),%edx\n" +- "mov 0x28(%esp),%edi\n" +- "mov 0x2c(%esp),%esi\n" +- "mov 0x30(%esp),%ebp\n" +- "mov 0x3c(%esp),%ecx\n" +- "xor %ebx,%ebx\n" +- "jmp Lscaleend\n" +- +-"Lscaleloop:" +- "mov %ebx,%eax\n" +- "sar $0x11,%eax\n" +- "movzbl (%edi,%eax,1),%eax\n" +- "movq 2048(%ecx,%eax,8),%mm0\n" +- "mov %ebx,%eax\n" +- "sar $0x11,%eax\n" +- "movzbl (%esi,%eax,1),%eax\n" +- "paddsw 4096(%ecx,%eax,8),%mm0\n" +- "mov %ebx,%eax\n" +- "add 0x38(%esp),%ebx\n" +- "sar $0x10,%eax\n" +- "movzbl (%edx,%eax,1),%eax\n" +- "movq 0(%ecx,%eax,8),%mm1\n" +- "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" +- "movntq %mm1,0x0(%ebp)\n" +- "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" +- "movzbl (%edi,%eax,1),%eax\n" +- "movq 2048(%ecx,%eax,8),%mm0\n" +- "mov %ebx,%eax\n" +- "sar $0x11,%eax\n" +- "movzbl (%esi,%eax,1),%eax\n" +- "paddsw 4096(%ecx,%eax,8),%mm0\n" +- "mov %ebx,%eax\n" +- "sar $0x10,%eax\n" +- "movzbl (%edx,%eax,1),%eax\n" +- "movq 0(%ecx,%eax,8),%mm1\n" +- "paddsw %mm0,%mm1\n" +- "psraw $0x6,%mm1\n" +- "packuswb %mm1,%mm1\n" +- "movd %mm1,0x0(%ebp)\n" +- +-"Lscaledone:" +- "popa\n" +- "ret\n" +-); +- +- +-void ScaleYUVToRGB32Row(const uint8* y_buf, +- const uint8* u_buf, +- const uint8* v_buf, +- uint8* rgb_buf, +- int width, +- int source_dx) { +- PICScaleYUVToRGB32Row(y_buf, u_buf, v_buf, rgb_buf, width, source_dx, +- &kCoefficientsRgbY[0][0]); +-} +- +-void PICLinearScaleYUVToRGB32Row(const uint8* y_buf, +- const uint8* u_buf, +- const uint8* v_buf, +- uint8* rgb_buf, +- int width, +- int source_dx, +- int16 *kCoefficientsRgbY); +- asm( +- ".text\n" +-#if defined(OS_MACOSX) +-"_PICLinearScaleYUVToRGB32Row:\n" +-#else +-"PICLinearScaleYUVToRGB32Row:\n" +-#endif +- "pusha\n" +- "mov 0x24(%esp),%edx\n" +- "mov 0x30(%esp),%ebp\n" +- "mov 0x34(%esp),%ecx\n" +- "mov 0x3c(%esp),%edi\n" +- "xor %ebx,%ebx\n" +- +- // source_width = width * source_dx + ebx +- "mov 0x34(%esp), %ecx\n" +- "imull 0x38(%esp), %ecx\n" +- "mov %ecx, 0x34(%esp)\n" +- +- "mov 0x38(%esp), %ecx\n" +- "xor %ebx,%ebx\n" // x = 0 +- "cmp $0x20000,%ecx\n" // if source_dx >= 2.0 +- "jl .lscaleend\n" +- "mov $0x8000,%ebx\n" // x = 0.5 for 1/2 or less +- "jmp .lscaleend\n" +- +-".lscaleloop:" +- "mov 0x28(%esp),%esi\n" +- "mov %ebx,%eax\n" +- "sar $0x11,%eax\n" +- +- "movzbl (%esi,%eax,1),%ecx\n" +- "movzbl 1(%esi,%eax,1),%esi\n" +- "mov %ebx,%eax\n" +- "andl $0x1fffe, %eax \n" +- "imul %eax, %esi \n" +- "xorl $0x1fffe, %eax \n" +- "imul %eax, %ecx \n" +- "addl %esi, %ecx \n" +- "shrl $17, %ecx \n" +- "movq 2048(%edi,%ecx,8),%mm0\n" +- +- "mov 0x2c(%esp),%esi\n" +- "mov %ebx,%eax\n" +- "sar $0x11,%eax\n" +- +- "movzbl (%esi,%eax,1),%ecx\n" +- "movzbl 1(%esi,%eax,1),%esi\n" +- "mov %ebx,%eax\n" +- "andl $0x1fffe, %eax \n" +- "imul %eax, %esi \n" +- "xorl $0x1fffe, %eax \n" +- "imul %eax, %ecx \n" +- "addl %esi, %ecx \n" +- "shrl $17, %ecx \n" +- "paddsw 4096(%edi,%ecx,8),%mm0\n" +- +- "mov %ebx,%eax\n" +- "sar $0x10,%eax\n" +- "movzbl (%edx,%eax,1),%ecx\n" +- "movzbl 1(%edx,%eax,1),%esi\n" +- "mov %ebx,%eax\n" +- "add 0x38(%esp),%ebx\n" +- "andl $0xffff, %eax \n" +- "imul %eax, %esi \n" +- "xorl $0xffff, %eax \n" +- "imul %eax, %ecx \n" +- "addl %esi, %ecx \n" +- "shrl $16, %ecx \n" +- "movq (%edi,%ecx,8),%mm1\n" +- +- "cmp 0x34(%esp), %ebx\n" +- "jge .lscalelastpixel\n" +- +- "mov %ebx,%eax\n" +- "sar $0x10,%eax\n" +- "movzbl (%edx,%eax,1),%ecx\n" +- "movzbl 1(%edx,%eax,1),%esi\n" +- "mov %ebx,%eax\n" +- "add 0x38(%esp),%ebx\n" +- "andl $0xffff, %eax \n" +- "imul %eax, %esi \n" +- "xorl $0xffff, %eax \n" +- "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" +- "movntq %mm1,0x0(%ebp)\n" +- "add $0x8,%ebp\n" +- +-".lscaleend:" +- "cmp %ebx, 0x34(%esp)\n" +- "jg .lscaleloop\n" +- "popa\n" +- "ret\n" +- +-".lscalelastpixel:" +- "paddsw %mm0, %mm1\n" +- "psraw $6, %mm1\n" +- "packuswb %mm1, %mm1\n" +- "movd %mm1, (%ebp)\n" +- "popa\n" +- "ret\n" +-); +- +-void LinearScaleYUVToRGB32Row(const uint8* y_buf, +- const uint8* u_buf, +- const uint8* v_buf, +- uint8* rgb_buf, +- int width, +- int source_dx) { +- PICLinearScaleYUVToRGB32Row(y_buf, u_buf, v_buf, rgb_buf, width, source_dx, +- &kCoefficientsRgbY[0][0]); +-} - -#else // USE_MMX - - // Reference version of YUV converter. - static const int kClipTableSize = 256; - static const int kClipOverflow = 288; // Cb max is 535. +- + // C reference code that mimic the YUV assembly. + #define packuswb(x) ((x) < 0 ? 0 : ((x) > 255 ? 255 : (x))) + #define paddsw(x, y) (((x) + (y)) < -32768 ? -32768 : \ + (((x) + (y)) > 32767 ? 32767 : ((x) + (y)))) - static uint8 kRgbClipTable[kClipOverflow + - kClipTableSize + - kClipOverflow] = { -@@ -651,17 +149,17 @@ static inline void YuvPixel(uint8 y, + static inline void YuvPixel(uint8 y, + uint8 u, + uint8 v, +@@ -833,66 +39,71 @@ static inline void YuvPixel(uint8 y, + a >>= 6; - int32 C298a = ((static_cast<int32>(y) - 16) * 298 + 128); - *reinterpret_cast<uint32*>(rgb_buf) = (clip(C298a + cb)) | - (clip(C298a + cg) << 8) | - (clip(C298a + cr) << 16) | - (0xff000000); + *reinterpret_cast<uint32*>(rgb_buf) = (packuswb(b)) | + (packuswb(g) << 8) | + (packuswb(r) << 16) | + (packuswb(a) << 24); } -void FastConvertYUVToRGB32Row(const uint8* y_buf, +- const uint8* u_buf, +- const uint8* v_buf, +- uint8* rgb_buf, +- int width) { +void FastConvertYUVToRGB32Row_C(const uint8* y_buf, - const uint8* u_buf, - const uint8* v_buf, - uint8* rgb_buf, - int width) { ++ const uint8* u_buf, ++ const uint8* v_buf, ++ uint8* rgb_buf, ++ int width, ++ unsigned int x_shift) { for (int x = 0; x < width; x += 2) { - uint8 u = u_buf[x >> 1]; - uint8 v = v_buf[x >> 1]; +- uint8 u = u_buf[x >> 1]; +- uint8 v = v_buf[x >> 1]; ++ uint8 u = u_buf[x >> x_shift]; ++ uint8 v = v_buf[x >> x_shift]; uint8 y0 = y_buf[x]; -@@ -673,27 +171,26 @@ void FastConvertYUVToRGB32Row(const uint8* y_buf, + YuvPixel(y0, u, v, rgb_buf); + if ((x + 1) < width) { + uint8 y1 = y_buf[x + 1]; ++ if (x_shift == 0) { ++ u = u_buf[x + 1]; ++ v = v_buf[x + 1]; ++ } + YuvPixel(y1, u, v, rgb_buf + 4); + } rgb_buf += 8; // Advance 2 pixels. } } - // 28.4 fixed point is used. A shift by 4 isolates the integer. - // A shift by 5 is used to further subsample the chrominence channels. - // & 15 isolates the fixed point fraction. >> 2 to get the upper 2 bits, - // for 1/4 pixel accurate interpolation. + // 16.16 fixed point is used. A shift by 16 isolates the integer. + // A shift by 17 is used to further subsample the chrominence channels. + // & 0xffff isolates the fixed point fraction. >> 2 to get the upper 2 bits, + // for 1/65536 pixel accurate interpolation. -void ScaleYUVToRGB32Row(const uint8* y_buf, +- const uint8* u_buf, +- const uint8* v_buf, +- uint8* rgb_buf, +- int width, +- int source_dx) { +void ScaleYUVToRGB32Row_C(const uint8* y_buf, - const uint8* u_buf, - const uint8* v_buf, - uint8* rgb_buf, - int width, - int scaled_dx) { - int scaled_x = 0; - for (int x = 0; x < width; ++x) { - uint8 u = u_buf[scaled_x >> 5]; - uint8 v = v_buf[scaled_x >> 5]; - uint8 y0 = y_buf[scaled_x >> 4]; - YuvPixel(y0, u, v, rgb_buf); - rgb_buf += 4; - scaled_x += scaled_dx; ++ const uint8* u_buf, ++ const uint8* v_buf, ++ uint8* rgb_buf, ++ int width, ++ int source_dx) { + int x = 0; + for (int i = 0; i < width; i += 2) { + int y = y_buf[x >> 16]; + int u = u_buf[(x >> 17)]; + int v = v_buf[(x >> 17)]; + YuvPixel(y, u, v, rgb_buf); + x += source_dx; + if ((i + 1) < width) { + y = y_buf[x >> 16]; + YuvPixel(y, u, v, rgb_buf+4); + x += source_dx; + } + rgb_buf += 8; } } + +-void LinearScaleYUVToRGB32Row(const uint8* y_buf, +- const uint8* u_buf, +- const uint8* v_buf, +- uint8* rgb_buf, +- int width, +- int source_dx) { ++void LinearScaleYUVToRGB32Row_C(const uint8* y_buf, ++ const uint8* u_buf, ++ const uint8* v_buf, ++ uint8* rgb_buf, ++ int width, ++ int source_dx) { + int x = 0; + if (source_dx >= 0x20000) { + x = 32768; + } + for (int i = 0; i < width; i += 2) { + int y0 = y_buf[x >> 16]; + int y1 = y_buf[(x >> 16) + 1]; + int u0 = u_buf[(x >> 17)]; +@@ -913,11 +124,10 @@ void LinearScaleYUVToRGB32Row(const uint8* y_buf, + 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" -diff --git b/gfx/ycbcr/yuv_row_linux.cpp a/gfx/ycbcr/yuv_row_linux.cpp -index 4a20777..eeb78f6 100644 ---- b/gfx/ycbcr/yuv_row_linux.cpp -+++ a/gfx/ycbcr/yuv_row_linux.cpp -@@ -1,25 +1,20 @@ - // Copyright (c) 2009 The Chromium Authors. All rights reserved. +diff --git b/gfx/ycbcr/yuv_row_posix.cpp a/gfx/ycbcr/yuv_row_posix.cpp +index a66fa7b..382c2bd 100644 +--- b/gfx/ycbcr/yuv_row_posix.cpp ++++ a/gfx/ycbcr/yuv_row_posix.cpp +@@ -1,33 +1,29 @@ + // 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" +#include "yuv_row.h" -#ifdef _DEBUG -#include "base/logging.h" -#else #define DCHECK(a) -#endif - // TODO(fbarchard): Move this to yuv_row_posix.cc to share with Mac. - // TODO(fbarchard): Do 64 bit version. - extern "C" { --#if USE_MMX - - #define RGBY(i) { \ - static_cast<int16>(1.164 * 64 * (i - 16) + 0.5), \ - static_cast<int16>(1.164 * 64 * (i - 16) + 0.5), \ - static_cast<int16>(1.164 * 64 * (i - 16) + 0.5), \ - 0 \ - } -@@ -234,21 +229,16 @@ MMX_ALIGNED(int16 kCoefficientsRgbY[768][4]) = { - RGBV(0xE8), RGBV(0xE9), RGBV(0xEA), RGBV(0xEB), - RGBV(0xEC), RGBV(0xED), RGBV(0xEE), RGBV(0xEF), - RGBV(0xF0), RGBV(0xF1), RGBV(0xF2), RGBV(0xF3), - RGBV(0xF4), RGBV(0xF5), RGBV(0xF6), RGBV(0xF7), - RGBV(0xF8), RGBV(0xF9), RGBV(0xFA), RGBV(0xFB), - RGBV(0xFC), RGBV(0xFD), RGBV(0xFE), RGBV(0xFF), - }; - --#undef RGBY --#undef RGBU --#undef RGBV --#undef MMX_ALIGNED -- - #if defined(ARCH_CPU_X86_64) +-#if USE_SSE2 && defined(ARCH_CPU_X86_64) ++#if 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 -@@ -376,17 +366,16 @@ void ScaleYUVToRGB32Row(const uint8* y_buf, // rdi + asm( +- "jmp convertend\n" +-"convertloop:" ++ "jmp 1f\n" ++"0:" + "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" +@@ -37,36 +33,36 @@ void FastConvertYUVToRGB32Row(const uint8* y_buf, // rdi + "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:" ++"1:" + "sub $0x2,%4\n" +- "jns convertloop\n" ++ "jns 0b\n" + +-"convertnext:" ++"2:" + "add $0x1,%4\n" +- "js convertdone\n" ++ "js 3f\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:" ++"3:" + : + : "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" +@@ -77,19 +73,19 @@ void ScaleYUVToRGB32Row(const uint8* y_buf, // rdi + const uint8* u_buf, // rsi + const uint8* v_buf, // rdx + uint8* rgb_buf, // rcx + int width, // r8 + int source_dx) { // r9 + asm( + "xor %%r11,%%r11\n" + "sub $0x2,%4\n" +- "js scalenext\n" ++ "js 1f\n" + +-"scaleloop:" ++"0:" + "mov %%r11,%%r10\n" + "sar $0x11,%%r10\n" + "movzb (%1,%%r10,1),%%rax\n" + "movq 2048(%5,%%rax,8),%%xmm0\n" + "movzb (%2,%%r10,1),%%rax\n" + "movq 4096(%5,%%rax,8),%%xmm1\n" + "lea (%%r11,%6),%%r10\n" + "sar $0x10,%%r11\n" +@@ -103,38 +99,38 @@ void ScaleYUVToRGB32Row(const uint8* y_buf, // rdi + "paddsw %%xmm0,%%xmm1\n" + "paddsw %%xmm0,%%xmm2\n" + "shufps $0x44,%%xmm2,%%xmm1\n" + "psraw $0x6,%%xmm1\n" + "packuswb %%xmm1,%%xmm1\n" + "movq %%xmm1,0x0(%3)\n" + "add $0x8,%3\n" + "sub $0x2,%4\n" +- "jns scaleloop\n" ++ "jns 0b\n" + +-"scalenext:" ++"1:" + "add $0x1,%4\n" +- "js scaledone\n" ++ "js 2f\n" - #else + "mov %%r11,%%r10\n" + "sar $0x11,%%r10\n" + "movzb (%1,%%r10,1),%%rax\n" + "movq 2048(%5,%%rax,8),%%xmm0\n" + "movzb (%2,%%r10,1),%%rax\n" + "movq 4096(%5,%%rax,8),%%xmm1\n" + "paddsw %%xmm1,%%xmm0\n" + "sar $0x10,%%r11\n" + "movzb (%0,%%r11,1),%%rax\n" + "movq (%5,%%rax,8),%%xmm1\n" + "paddsw %%xmm0,%%xmm1\n" + "psraw $0x6,%%xmm1\n" + "packuswb %%xmm1,%%xmm1\n" + "movd %%xmm1,0x0(%3)\n" + +-"scaledone:" ++"2:" + : + : "r"(y_buf), // %0 + "r"(u_buf), // %1 + "r"(v_buf), // %2 + "r"(rgb_buf), // %3 + "r"(width), // %4 + "r" (kCoefficientsRgbY), // %5 + "r"(static_cast<long>(source_dx)) // %6 +@@ -146,23 +142,23 @@ void LinearScaleYUVToRGB32Row(const uint8* y_buf, + const uint8* u_buf, + const uint8* v_buf, + uint8* rgb_buf, + int width, + int source_dx) { + asm( + "xor %%r11,%%r11\n" // x = 0 + "sub $0x2,%4\n" +- "js .lscalenext\n" ++ "js 2f\n" + "cmp $0x20000,%6\n" // if source_dx >= 2.0 +- "jl .lscalehalf\n" ++ "jl 0f\n" + "mov $0x8000,%%r11\n" // x = 0.5 for 1/2 or less +-".lscalehalf:" ++"0:" + +-".lscaleloop:" ++"1:" + "mov %%r11,%%r10\n" + "sar $0x11,%%r10\n" + "movzb (%1, %%r10, 1), %%r13 \n" + "movzb 1(%1, %%r10, 1), %%r14 \n" + "mov %%r11, %%rax \n" + "and $0x1fffe, %%rax \n" + "imul %%rax, %%r14 \n" +@@ -215,21 +211,21 @@ void LinearScaleYUVToRGB32Row(const uint8* y_buf, + "paddsw %%xmm0,%%xmm1\n" + "paddsw %%xmm0,%%xmm2\n" + "shufps $0x44,%%xmm2,%%xmm1\n" + "psraw $0x6,%%xmm1\n" + "packuswb %%xmm1,%%xmm1\n" + "movq %%xmm1,0x0(%3)\n" + "add $0x8,%3\n" + "sub $0x2,%4\n" +- "jns .lscaleloop\n" ++ "jns 1b\n" + +-".lscalenext:" ++"2:" + "add $0x1,%4\n" +- "js .lscaledone\n" ++ "js 3f\n" + + "mov %%r11,%%r10\n" + "sar $0x11,%%r10\n" + + "movzb (%1,%%r10,1), %%r13 \n" + "movq 2048(%5,%%r13,8),%%xmm0\n" + + "movzb (%2,%%r10,1), %%r13 \n" +@@ -241,52 +237,52 @@ void LinearScaleYUVToRGB32Row(const uint8* y_buf, + "movzb (%0,%%r11,1), %%r13 \n" + "movq (%5,%%r13,8),%%xmm1\n" + + "paddsw %%xmm0,%%xmm1\n" + "psraw $0x6,%%xmm1\n" + "packuswb %%xmm1,%%xmm1\n" + "movd %%xmm1,0x0(%3)\n" + +-".lscaledone:" ++"3:" + : + : "r"(y_buf), // %0 + "r"(u_buf), // %1 + "r"(v_buf), // %2 + "r"(rgb_buf), // %3 + "r"(width), // %4 + "r" (kCoefficientsRgbY), // %5 + "r"(static_cast<long>(source_dx)) // %6 + : "memory", "r10", "r11", "r13", "r14", "rax", "xmm0", "xmm1", "xmm2" + ); + } + +-#elif USE_MMX && !defined(ARCH_CPU_X86_64) && !defined(__PIC__) ++#elif defined(ARCH_CPU_X86_32) && !defined(__PIC__) + + // PIC version is slower because less registers are available, so + // non-PIC is used on platforms where it is possible. +- void FastConvertYUVToRGB32Row(const uint8* y_buf, const uint8* u_buf, const uint8* v_buf, uint8* rgb_buf, int width); -- asm( + ".text\n" ".global FastConvertYUVToRGB32Row\n" ++ ".type FastConvertYUVToRGB32Row, @function\n" "FastConvertYUVToRGB32Row:\n" "pusha\n" "mov 0x24(%esp),%edx\n" "mov 0x28(%esp),%edi\n" "mov 0x2c(%esp),%esi\n" "mov 0x30(%esp),%ebp\n" -@@ -505,195 +494,10 @@ void ScaleYUVToRGB32Row(const uint8* y_buf, + "mov 0x34(%esp),%ecx\n" +- "jmp convertend\n" ++ "jmp 1f\n" + +-"convertloop:" ++"0:" + "movzbl (%edi),%eax\n" + "add $0x1,%edi\n" + "movzbl (%esi),%ebx\n" + "add $0x1,%esi\n" + "movq kCoefficientsRgbY+2048(,%eax,8),%mm0\n" + "movzbl (%edx),%eax\n" + "paddsw kCoefficientsRgbY+4096(,%ebx,8),%mm0\n" + "movzbl 0x1(%edx),%ebx\n" +@@ -295,59 +291,63 @@ void FastConvertYUVToRGB32Row(const uint8* y_buf, + "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" + "movntq %mm1,0x0(%ebp)\n" + "add $0x8,%ebp\n" +-"convertend:" ++"1:" + "sub $0x2,%ecx\n" +- "jns convertloop\n" ++ "jns 0b\n" + + "and $0x1,%ecx\n" +- "je convertdone\n" ++ "je 2f\n" + + "movzbl (%edi),%eax\n" + "movq kCoefficientsRgbY+2048(,%eax,8),%mm0\n" + "movzbl (%esi),%eax\n" + "paddsw kCoefficientsRgbY+4096(,%eax,8),%mm0\n" + "movzbl (%edx),%eax\n" + "movq kCoefficientsRgbY(,%eax,8),%mm1\n" + "paddsw %mm0,%mm1\n" + "psraw $0x6,%mm1\n" + "packuswb %mm1,%mm1\n" + "movd %mm1,0x0(%ebp)\n" +-"convertdone:" ++"2:" + "popa\n" + "ret\n" ++#if !defined(XP_MACOSX) ++ ".previous\n" ++#endif + ); + + + void ScaleYUVToRGB32Row(const uint8* y_buf, + const uint8* u_buf, + const uint8* v_buf, + uint8* rgb_buf, + int width, + int source_dx); + asm( + ".text\n" + ".global ScaleYUVToRGB32Row\n" ++ ".type ScaleYUVToRGB32Row, @function\n" + "ScaleYUVToRGB32Row:\n" + "pusha\n" + "mov 0x24(%esp),%edx\n" + "mov 0x28(%esp),%edi\n" + "mov 0x2c(%esp),%esi\n" + "mov 0x30(%esp),%ebp\n" + "mov 0x34(%esp),%ecx\n" + "xor %ebx,%ebx\n" +- "jmp scaleend\n" ++ "jmp 1f\n" + +-"scaleloop:" ++"0:" + "mov %ebx,%eax\n" + "sar $0x11,%eax\n" + "movzbl (%edi,%eax,1),%eax\n" + "movq kCoefficientsRgbY+2048(,%eax,8),%mm0\n" + "mov %ebx,%eax\n" + "sar $0x11,%eax\n" + "movzbl (%esi,%eax,1),%eax\n" + "paddsw kCoefficientsRgbY+4096(,%eax,8),%mm0\n" +@@ -363,22 +363,22 @@ void ScaleYUVToRGB32Row(const uint8* y_buf, + "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" + "movntq %mm1,0x0(%ebp)\n" + "add $0x8,%ebp\n" +-"scaleend:" ++"1:" + "sub $0x2,%ecx\n" +- "jns scaleloop\n" ++ "jns 0b\n" + + "and $0x1,%ecx\n" +- "je scaledone\n" ++ "je 2f\n" + + "mov %ebx,%eax\n" + "sar $0x11,%eax\n" + "movzbl (%edi,%eax,1),%eax\n" + "movq kCoefficientsRgbY+2048(,%eax,8),%mm0\n" + "mov %ebx,%eax\n" + "sar $0x11,%eax\n" + "movzbl (%esi,%eax,1),%eax\n" +@@ -387,49 +387,53 @@ void ScaleYUVToRGB32Row(const uint8* y_buf, + "sar $0x10,%eax\n" + "movzbl (%edx,%eax,1),%eax\n" + "movq kCoefficientsRgbY(,%eax,8),%mm1\n" + "paddsw %mm0,%mm1\n" + "psraw $0x6,%mm1\n" + "packuswb %mm1,%mm1\n" "movd %mm1,0x0(%ebp)\n" - "scaledone:" +-"scaledone:" ++"2:" + "popa\n" + "ret\n" ++#if !defined(XP_MACOSX) ++ ".previous\n" ++#endif + ); + + void LinearScaleYUVToRGB32Row(const uint8* y_buf, + const uint8* u_buf, + const uint8* v_buf, + uint8* rgb_buf, + int width, + int source_dx); + asm( + ".text\n" + ".global LinearScaleYUVToRGB32Row\n" ++ ".type LinearScaleYUVToRGB32Row, @function\n" + "LinearScaleYUVToRGB32Row:\n" + "pusha\n" + "mov 0x24(%esp),%edx\n" + "mov 0x28(%esp),%edi\n" + "mov 0x30(%esp),%ebp\n" + + // source_width = width * source_dx + ebx + "mov 0x34(%esp), %ecx\n" + "imull 0x38(%esp), %ecx\n" + "mov %ecx, 0x34(%esp)\n" + + "mov 0x38(%esp), %ecx\n" + "xor %ebx,%ebx\n" // x = 0 + "cmp $0x20000,%ecx\n" // if source_dx >= 2.0 +- "jl .lscaleend\n" ++ "jl 1f\n" + "mov $0x8000,%ebx\n" // x = 0.5 for 1/2 or less +- "jmp .lscaleend\n" ++ "jmp 1f\n" + +-".lscaleloop:" ++"0:" + "mov %ebx,%eax\n" + "sar $0x11,%eax\n" + + "movzbl (%edi,%eax,1),%ecx\n" + "movzbl 1(%edi,%eax,1),%esi\n" + "mov %ebx,%eax\n" + "andl $0x1fffe, %eax \n" + "imul %eax, %esi \n" +@@ -464,17 +468,17 @@ void LinearScaleYUVToRGB32Row(const uint8* y_buf, + "imul %eax, %esi \n" + "xorl $0xffff, %eax \n" + "imul %eax, %ecx \n" + "addl %esi, %ecx \n" + "shrl $16, %ecx \n" + "movq kCoefficientsRgbY(,%ecx,8),%mm1\n" + + "cmp 0x34(%esp), %ebx\n" +- "jge .lscalelastpixel\n" ++ "jge 2f\n" + + "mov %ebx,%eax\n" + "sar $0x10,%eax\n" + "movzbl (%edx,%eax,1),%ecx\n" + "movzbl 1(%edx,%eax,1),%esi\n" + "mov %ebx,%eax\n" + "add 0x38(%esp),%ebx\n" + "andl $0xffff, %eax \n" +@@ -488,56 +492,60 @@ void LinearScaleYUVToRGB32Row(const uint8* y_buf, + "paddsw %mm0,%mm1\n" + "paddsw %mm0,%mm2\n" + "psraw $0x6,%mm1\n" + "psraw $0x6,%mm2\n" + "packuswb %mm2,%mm1\n" + "movntq %mm1,0x0(%ebp)\n" + "add $0x8,%ebp\n" + +-".lscaleend:" ++"1:" + "cmp 0x34(%esp), %ebx\n" +- "jl .lscaleloop\n" ++ "jl 0b\n" + "popa\n" + "ret\n" + +-".lscalelastpixel:" ++"2:" + "paddsw %mm0, %mm1\n" + "psraw $6, %mm1\n" + "packuswb %mm1, %mm1\n" + "movd %mm1, (%ebp)\n" + "popa\n" + "ret\n" ++#if !defined(XP_MACOSX) ++ ".previous\n" ++#endif + ); + +-#elif USE_MMX && !defined(ARCH_CPU_X86_64) && defined(__PIC__) ++#elif defined(ARCH_CPU_X86_32) && defined(__PIC__) ++ ++void PICConvertYUVToRGB32Row(const uint8* y_buf, ++ const uint8* u_buf, ++ const uint8* v_buf, ++ uint8* rgb_buf, ++ int width, ++ int16 *kCoefficientsRgbY); + +-extern void PICConvertYUVToRGB32Row(const uint8* y_buf, +- const uint8* u_buf, +- const uint8* v_buf, +- uint8* rgb_buf, +- int width, +- int16 *kCoefficientsRgbY); + asm( + ".text\n" +-#if defined(OS_MACOSX) ++#if defined(XP_MACOSX) + "_PICConvertYUVToRGB32Row:\n" + #else + "PICConvertYUVToRGB32Row:\n" + #endif + "pusha\n" + "mov 0x24(%esp),%edx\n" + "mov 0x28(%esp),%edi\n" + "mov 0x2c(%esp),%esi\n" + "mov 0x30(%esp),%ebp\n" + "mov 0x38(%esp),%ecx\n" + +- "jmp .Lconvertend\n" ++ "jmp 1f\n" + +-".Lconvertloop:" ++"0:" + "movzbl (%edi),%eax\n" + "add $0x1,%edi\n" + "movzbl (%esi),%ebx\n" + "add $0x1,%esi\n" + "movq 2048(%ecx,%eax,8),%mm0\n" + "movzbl (%edx),%eax\n" + "paddsw 4096(%ecx,%ebx,8),%mm0\n" + "movzbl 0x1(%edx),%ebx\n" +@@ -546,72 +554,75 @@ extern void PICConvertYUVToRGB32Row(const uint8* y_buf, + "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" + "movntq %mm1,0x0(%ebp)\n" + "add $0x8,%ebp\n" +-".Lconvertend:" ++"1:" + "subl $0x2,0x34(%esp)\n" +- "jns .Lconvertloop\n" ++ "jns 0b\n" + + "andl $0x1,0x34(%esp)\n" +- "je .Lconvertdone\n" ++ "je 2f\n" + + "movzbl (%edi),%eax\n" + "movq 2048(%ecx,%eax,8),%mm0\n" + "movzbl (%esi),%eax\n" + "paddsw 4096(%ecx,%eax,8),%mm0\n" + "movzbl (%edx),%eax\n" + "movq 0(%ecx,%eax,8),%mm1\n" + "paddsw %mm0,%mm1\n" + "psraw $0x6,%mm1\n" + "packuswb %mm1,%mm1\n" + "movd %mm1,0x0(%ebp)\n" +-".Lconvertdone:\n" ++"2:" "popa\n" "ret\n" ++#if !defined(XP_MACOSX) ++ ".previous\n" ++#endif + ); + + void FastConvertYUVToRGB32Row(const uint8* y_buf, + const uint8* u_buf, + const uint8* v_buf, + uint8* rgb_buf, + int width) { + PICConvertYUVToRGB32Row(y_buf, u_buf, v_buf, rgb_buf, width, + &kCoefficientsRgbY[0][0]); + } + +-extern void PICScaleYUVToRGB32Row(const uint8* y_buf, +- const uint8* u_buf, +- const uint8* v_buf, +- uint8* rgb_buf, +- int width, +- int source_dx, +- int16 *kCoefficientsRgbY); ++void PICScaleYUVToRGB32Row(const uint8* y_buf, ++ const uint8* u_buf, ++ const uint8* v_buf, ++ uint8* rgb_buf, ++ int width, ++ int source_dx, ++ int16 *kCoefficientsRgbY); + + asm( + ".text\n" +-#if defined(OS_MACOSX) ++#if defined(XP_MACOSX) + "_PICScaleYUVToRGB32Row:\n" + #else + "PICScaleYUVToRGB32Row:\n" + #endif + "pusha\n" + "mov 0x24(%esp),%edx\n" + "mov 0x28(%esp),%edi\n" + "mov 0x2c(%esp),%esi\n" + "mov 0x30(%esp),%ebp\n" + "mov 0x3c(%esp),%ecx\n" + "xor %ebx,%ebx\n" +- "jmp Lscaleend\n" ++ "jmp 1f\n" + +-"Lscaleloop:" ++"0:" + "mov %ebx,%eax\n" + "sar $0x11,%eax\n" + "movzbl (%edi,%eax,1),%eax\n" + "movq 2048(%ecx,%eax,8),%mm0\n" + "mov %ebx,%eax\n" + "sar $0x11,%eax\n" + "movzbl (%esi,%eax,1),%eax\n" + "paddsw 4096(%ecx,%eax,8),%mm0\n" +@@ -627,22 +638,22 @@ extern void PICScaleYUVToRGB32Row(const uint8* y_buf, + "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" + "movntq %mm1,0x0(%ebp)\n" + "add $0x8,%ebp\n" +-"Lscaleend:" ++"1:" + "subl $0x2,0x34(%esp)\n" +- "jns Lscaleloop\n" ++ "jns 0b\n" + + "andl $0x1,0x34(%esp)\n" +- "je Lscaledone\n" ++ "je 2f\n" + + "mov %ebx,%eax\n" + "sar $0x11,%eax\n" + "movzbl (%edi,%eax,1),%eax\n" + "movq 2048(%ecx,%eax,8),%mm0\n" + "mov %ebx,%eax\n" + "sar $0x11,%eax\n" + "movzbl (%esi,%eax,1),%eax\n" +@@ -651,22 +662,24 @@ extern void PICScaleYUVToRGB32Row(const uint8* y_buf, + "sar $0x10,%eax\n" + "movzbl (%edx,%eax,1),%eax\n" + "movq 0(%ecx,%eax,8),%mm1\n" + "paddsw %mm0,%mm1\n" + "psraw $0x6,%mm1\n" + "packuswb %mm1,%mm1\n" + "movd %mm1,0x0(%ebp)\n" + +-"Lscaledone:" ++"2:" + "popa\n" + "ret\n" ++#if !defined(XP_MACOSX) ++ ".previous\n" ++#endif ); +- + void ScaleYUVToRGB32Row(const uint8* y_buf, + const uint8* u_buf, + const uint8* v_buf, + uint8* rgb_buf, + int width, + int source_dx) { + PICScaleYUVToRGB32Row(y_buf, u_buf, v_buf, rgb_buf, width, source_dx, + &kCoefficientsRgbY[0][0]); +@@ -674,19 +687,20 @@ void ScaleYUVToRGB32Row(const uint8* y_buf, + + void PICLinearScaleYUVToRGB32Row(const uint8* y_buf, + const uint8* u_buf, + const uint8* v_buf, + uint8* rgb_buf, + int width, + int source_dx, + int16 *kCoefficientsRgbY); ++ + asm( + ".text\n" +-#if defined(OS_MACOSX) ++#if defined(XP_MACOSX) + "_PICLinearScaleYUVToRGB32Row:\n" + #else + "PICLinearScaleYUVToRGB32Row:\n" #endif + "pusha\n" + "mov 0x24(%esp),%edx\n" + "mov 0x30(%esp),%ebp\n" + "mov 0x34(%esp),%ecx\n" +@@ -696,21 +710,21 @@ void PICLinearScaleYUVToRGB32Row(const uint8* y_buf, + // source_width = width * source_dx + ebx + "mov 0x34(%esp), %ecx\n" + "imull 0x38(%esp), %ecx\n" + "mov %ecx, 0x34(%esp)\n" + + "mov 0x38(%esp), %ecx\n" + "xor %ebx,%ebx\n" // x = 0 + "cmp $0x20000,%ecx\n" // if source_dx >= 2.0 +- "jl .lscaleend\n" ++ "jl 1f\n" + "mov $0x8000,%ebx\n" // x = 0.5 for 1/2 or less +- "jmp .lscaleend\n" ++ "jmp 1f\n" + +-".lscaleloop:" ++"0:" + "mov 0x28(%esp),%esi\n" + "mov %ebx,%eax\n" + "sar $0x11,%eax\n" + + "movzbl (%esi,%eax,1),%ecx\n" + "movzbl 1(%esi,%eax,1),%esi\n" + "mov %ebx,%eax\n" + "andl $0x1fffe, %eax \n" +@@ -746,17 +760,17 @@ void PICLinearScaleYUVToRGB32Row(const uint8* y_buf, + "imul %eax, %esi \n" + "xorl $0xffff, %eax \n" + "imul %eax, %ecx \n" + "addl %esi, %ecx \n" + "shrl $16, %ecx \n" + "movq (%edi,%ecx,8),%mm1\n" + + "cmp 0x34(%esp), %ebx\n" +- "jge .lscalelastpixel\n" ++ "jge 2f\n" + + "mov %ebx,%eax\n" + "sar $0x10,%eax\n" + "movzbl (%edx,%eax,1),%ecx\n" + "movzbl 1(%edx,%eax,1),%esi\n" + "mov %ebx,%eax\n" + "add 0x38(%esp),%ebx\n" + "andl $0xffff, %eax \n" +@@ -770,154 +784,66 @@ void PICLinearScaleYUVToRGB32Row(const uint8* y_buf, + "paddsw %mm0,%mm1\n" + "paddsw %mm0,%mm2\n" + "psraw $0x6,%mm1\n" + "psraw $0x6,%mm2\n" + "packuswb %mm2,%mm1\n" + "movntq %mm1,0x0(%ebp)\n" + "add $0x8,%ebp\n" + +-".lscaleend:" ++"1:" + "cmp %ebx, 0x34(%esp)\n" +- "jg .lscaleloop\n" ++ "jg 0b\n" + "popa\n" + "ret\n" + +-".lscalelastpixel:" ++"2:" + "paddsw %mm0, %mm1\n" + "psraw $6, %mm1\n" + "packuswb %mm1, %mm1\n" + "movd %mm1, (%ebp)\n" + "popa\n" + "ret\n" ++#if !defined(XP_MACOSX) ++ ".previous\n" ++#endif + ); + ++ + void LinearScaleYUVToRGB32Row(const uint8* y_buf, +- const uint8* u_buf, +- const uint8* v_buf, +- uint8* rgb_buf, +- int width, +- int source_dx) { ++ const uint8* u_buf, ++ const uint8* v_buf, ++ uint8* rgb_buf, ++ int width, ++ int source_dx) { + PICLinearScaleYUVToRGB32Row(y_buf, u_buf, v_buf, rgb_buf, width, source_dx, + &kCoefficientsRgbY[0][0]); + } - -#else // USE_MMX - --// Reference version of YUV converter. --static const int kClipTableSize = 256; --static const int kClipOverflow = 288; // Cb max is 535. -- --static uint8 kRgbClipTable[kClipOverflow + -- kClipTableSize + -- kClipOverflow] = { -- 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, // 288 underflow values -- 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, // clipped to 0. -- 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, -- 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, -- 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, -- 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, -- 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, -- 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, -- 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, -- 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, -- 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, -- 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, -- 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, -- 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, -- 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, -- 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, -- 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, -- 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, -- 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, -- 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, -- 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, -- 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, -- 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, -- 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, -- 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, -- 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, -- 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, -- 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, -- 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, -- 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, -- 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, -- 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, -- 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, -- 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, -- 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, -- 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, -- 0x00, 0x01, 0x02, 0x03, 0x04, 0x05, 0x06, 0x07, // Unclipped values. -- 0x08, 0x09, 0x0A, 0x0B, 0x0C, 0x0D, 0x0E, 0x0F, -- 0x10, 0x11, 0x12, 0x13, 0x14, 0x15, 0x16, 0x17, -- 0x18, 0x19, 0x1A, 0x1B, 0x1C, 0x1D, 0x1E, 0x1F, -- 0x20, 0x21, 0x22, 0x23, 0x24, 0x25, 0x26, 0x27, -- 0x28, 0x29, 0x2A, 0x2B, 0x2C, 0x2D, 0x2E, 0x2F, -- 0x30, 0x31, 0x32, 0x33, 0x34, 0x35, 0x36, 0x37, -- 0x38, 0x39, 0x3A, 0x3B, 0x3C, 0x3D, 0x3E, 0x3F, -- 0x40, 0x41, 0x42, 0x43, 0x44, 0x45, 0x46, 0x47, -- 0x48, 0x49, 0x4A, 0x4B, 0x4C, 0x4D, 0x4E, 0x4F, -- 0x50, 0x51, 0x52, 0x53, 0x54, 0x55, 0x56, 0x57, -- 0x58, 0x59, 0x5A, 0x5B, 0x5C, 0x5D, 0x5E, 0x5F, -- 0x60, 0x61, 0x62, 0x63, 0x64, 0x65, 0x66, 0x67, -- 0x68, 0x69, 0x6A, 0x6B, 0x6C, 0x6D, 0x6E, 0x6F, -- 0x70, 0x71, 0x72, 0x73, 0x74, 0x75, 0x76, 0x77, -- 0x78, 0x79, 0x7A, 0x7B, 0x7C, 0x7D, 0x7E, 0x7F, -- 0x80, 0x81, 0x82, 0x83, 0x84, 0x85, 0x86, 0x87, -- 0x88, 0x89, 0x8A, 0x8B, 0x8C, 0x8D, 0x8E, 0x8F, -- 0x90, 0x91, 0x92, 0x93, 0x94, 0x95, 0x96, 0x97, -- 0x98, 0x99, 0x9A, 0x9B, 0x9C, 0x9D, 0x9E, 0x9F, -- 0xA0, 0xA1, 0xA2, 0xA3, 0xA4, 0xA5, 0xA6, 0xA7, -- 0xA8, 0xA9, 0xAA, 0xAB, 0xAC, 0xAD, 0xAE, 0xAF, -- 0xB0, 0xB1, 0xB2, 0xB3, 0xB4, 0xB5, 0xB6, 0xB7, -- 0xB8, 0xB9, 0xBA, 0xBB, 0xBC, 0xBD, 0xBE, 0xBF, -- 0xC0, 0xC1, 0xC2, 0xC3, 0xC4, 0xC5, 0xC6, 0xC7, -- 0xC8, 0xC9, 0xCA, 0xCB, 0xCC, 0xCD, 0xCE, 0xCF, -- 0xD0, 0xD1, 0xD2, 0xD3, 0xD4, 0xD5, 0xD6, 0xD7, -- 0xD8, 0xD9, 0xDA, 0xDB, 0xDC, 0xDD, 0xDE, 0xDF, -- 0xE0, 0xE1, 0xE2, 0xE3, 0xE4, 0xE5, 0xE6, 0xE7, -- 0xE8, 0xE9, 0xEA, 0xEB, 0xEC, 0xED, 0xEE, 0xEF, -- 0xF0, 0xF1, 0xF2, 0xF3, 0xF4, 0xF5, 0xF6, 0xF7, -- 0xF8, 0xF9, 0xFA, 0xFB, 0xFC, 0xFD, 0xFE, 0xFF, -- 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, // 288 overflow values -- 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, // clipped to 255. -- 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, -- 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, -- 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, -- 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, -- 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, -- 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, -- 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, -- 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, -- 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, -- 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, -- 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, -- 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, -- 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, -- 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, -- 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, -- 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, -- 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, -- 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, -- 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, -- 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, -- 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, -- 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, -- 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, -- 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, -- 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, -- 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, -- 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, -- 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, -- 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, -- 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, -- 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, -- 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, -- 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, -- 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, --}; -- --// Clip an rgb channel value to 0..255 range. --// Source is signed fixed point 8.8. --// Table allows for values to underflow or overflow by 128. --// Therefore source range is -128 to 384. --// Output clips to unsigned 0 to 255. --static inline uint32 clip(int32 value) { -- DCHECK(((value >> 8) + kClipOverflow) >= 0); -- DCHECK(((value >> 8) + kClipOverflow) < -- (kClipOverflow + kClipTableSize + kClipOverflow)); -- return static_cast<uint32>(kRgbClipTable[((value) >> 8) + kClipOverflow]); --} +-// C reference code that mimic the YUV assembly. +-#define packuswb(x) ((x) < 0 ? 0 : ((x) > 255 ? 255 : (x))) +-#define paddsw(x, y) (((x) + (y)) < -32768 ? -32768 : \ +- (((x) + (y)) > 32767 ? 32767 : ((x) + (y)))) - -static inline void YuvPixel(uint8 y, - uint8 u, - uint8 v, - uint8* rgb_buf) { -- int32 d = static_cast<int32>(u) - 128; -- int32 e = static_cast<int32>(v) - 128; +- +- int b = kCoefficientsRgbY[256+u][0]; +- int g = kCoefficientsRgbY[256+u][1]; +- int r = kCoefficientsRgbY[256+u][2]; +- int a = kCoefficientsRgbY[256+u][3]; +- +- b = paddsw(b, kCoefficientsRgbY[512+v][0]); +- g = paddsw(g, kCoefficientsRgbY[512+v][1]); +- r = paddsw(r, kCoefficientsRgbY[512+v][2]); +- a = paddsw(a, kCoefficientsRgbY[512+v][3]); - -- int32 cb = (516 * d + 128); -- int32 cg = (- 100 * d - 208 * e + 128); -- int32 cr = (409 * e + 128); +- b = paddsw(b, kCoefficientsRgbY[y][0]); +- g = paddsw(g, kCoefficientsRgbY[y][1]); +- r = paddsw(r, kCoefficientsRgbY[y][2]); +- a = paddsw(a, kCoefficientsRgbY[y][3]); - -- int32 C298a = ((static_cast<int32>(y) - 16) * 298 + 128); -- *reinterpret_cast<uint32*>(rgb_buf) = (clip(C298a + cb)) | -- (clip(C298a + cg) << 8) | -- (clip(C298a + cr) << 16) | -- (0xff000000); +- b >>= 6; +- g >>= 6; +- r >>= 6; +- a >>= 6; +- +- *reinterpret_cast<uint32*>(rgb_buf) = (packuswb(b)) | +- (packuswb(g) << 8) | +- (packuswb(r) << 16) | +- (packuswb(a) << 24); -} - --void FastConvertYUVToRGB32Row(const uint8* y_buf, -- const uint8* u_buf, -- const uint8* v_buf, -- uint8* rgb_buf, -- int width) { ++#else + void FastConvertYUVToRGB32Row(const uint8* y_buf, + const uint8* u_buf, + const uint8* v_buf, + uint8* rgb_buf, + int width) { - for (int x = 0; x < width; x += 2) { - uint8 u = u_buf[x >> 1]; - uint8 v = v_buf[x >> 1]; - uint8 y0 = y_buf[x]; - YuvPixel(y0, u, v, rgb_buf); - if ((x + 1) < width) { - uint8 y1 = y_buf[x + 1]; - YuvPixel(y1, u, v, rgb_buf + 4); - } - rgb_buf += 8; // Advance 2 pixels. - } --} -- --// 28.4 fixed point is used. A shift by 4 isolates the integer. --// A shift by 5 is used to further subsample the chrominence channels. --// & 15 isolates the fixed point fraction. >> 2 to get the upper 2 bits, --// for 1/4 pixel accurate interpolation. --void ScaleYUVToRGB32Row(const uint8* y_buf, -- const uint8* u_buf, -- const uint8* v_buf, -- uint8* rgb_buf, -- int width, -- int scaled_dx) { -- int scaled_x = 0; -- for (int x = 0; x < width; ++x) { -- uint8 u = u_buf[scaled_x >> 5]; -- uint8 v = v_buf[scaled_x >> 5]; -- uint8 y0 = y_buf[scaled_x >> 4]; -- YuvPixel(y0, u, v, rgb_buf); -- rgb_buf += 4; -- scaled_x += scaled_dx; ++ FastConvertYUVToRGB32Row_C(y_buf, u_buf, v_buf, rgb_buf, width, 1); + } + +-// 16.16 fixed point is used. A shift by 16 isolates the integer. +-// A shift by 17 is used to further subsample the chrominence channels. +-// & 0xffff isolates the fixed point fraction. >> 2 to get the upper 2 bits, +-// for 1/65536 pixel accurate interpolation. + void ScaleYUVToRGB32Row(const uint8* y_buf, + const uint8* u_buf, + const uint8* v_buf, + uint8* rgb_buf, + int width, + int source_dx) { +- int x = 0; +- for (int i = 0; i < width; i += 2) { +- int y = y_buf[x >> 16]; +- int u = u_buf[(x >> 17)]; +- int v = v_buf[(x >> 17)]; +- YuvPixel(y, u, v, rgb_buf); +- x += source_dx; +- if ((i + 1) < width) { +- y = y_buf[x >> 16]; +- YuvPixel(y, u, v, rgb_buf+4); +- x += source_dx; +- } +- rgb_buf += 8; - } --} ++ ScaleYUVToRGB32Row_C(y_buf, u_buf, v_buf, rgb_buf, width, source_dx); + } + + void LinearScaleYUVToRGB32Row(const uint8* y_buf, + const uint8* u_buf, + const uint8* v_buf, + uint8* rgb_buf, + int width, + int source_dx) { +- int x = 0; +- if (source_dx >= 0x20000) { +- x = 32768; +- } +- for (int i = 0; i < width; i += 2) { +- int y0 = y_buf[x >> 16]; +- int y1 = y_buf[(x >> 16) + 1]; +- int u0 = u_buf[(x >> 17)]; +- int u1 = u_buf[(x >> 17) + 1]; +- int v0 = v_buf[(x >> 17)]; +- int v1 = v_buf[(x >> 17) + 1]; +- int y_frac = (x & 65535); +- int uv_frac = ((x >> 1) & 65535); +- int y = (y_frac * y1 + (y_frac ^ 65535) * y0) >> 16; +- int u = (uv_frac * u1 + (uv_frac ^ 65535) * u0) >> 16; +- int v = (uv_frac * v1 + (uv_frac ^ 65535) * v0) >> 16; +- YuvPixel(y, u, v, rgb_buf); +- x += source_dx; +- if ((i + 1) < width) { +- y0 = y_buf[x >> 16]; +- y1 = y_buf[(x >> 16) + 1]; +- y_frac = (x & 65535); +- y = (y_frac * y1 + (y_frac ^ 65535) * y0) >> 16; +- YuvPixel(y, u, v, rgb_buf+4); +- x += source_dx; +- } +- rgb_buf += 8; +- } ++ LinearScaleYUVToRGB32Row_C(y_buf, u_buf, v_buf, rgb_buf, width, source_dx); + } ++#endif + -#endif // USE_MMX } // extern "C" -diff --git b/gfx/ycbcr/yuv_row_mac.cpp a/gfx/ycbcr/yuv_row_mac.cpp -index 39c252d..aa40aac 100644 ---- b/gfx/ycbcr/yuv_row_mac.cpp -+++ a/gfx/ycbcr/yuv_row_mac.cpp +diff --git b/gfx/ycbcr/yuv_row_table.cpp a/gfx/ycbcr/yuv_row_table.cpp +index 296380b..ad71341 100644 +--- b/gfx/ycbcr/yuv_row_table.cpp ++++ a/gfx/ycbcr/yuv_row_table.cpp @@ -1,13 +1,13 @@ - // Copyright (c) 2009 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" -+#include "yuv_row.h" - - // TODO(fbarchard): Do 64 bit version. - - extern "C" { - - #define RGBY(i) { \ - static_cast<int16>(1.164 * 64 * (i - 16) + 0.5), \ - static_cast<int16>(1.164 * 64 * (i - 16) + 0.5), \ -diff --git b/gfx/ycbcr/yuv_row_win.cpp a/gfx/ycbcr/yuv_row_win.cpp -index 1e11d94..a77a16f 100644 ---- b/gfx/ycbcr/yuv_row_win.cpp -+++ a/gfx/ycbcr/yuv_row_win.cpp -@@ -1,13 +1,13 @@ - // Copyright (c) 2009 The Chromium Authors. All rights reserved. + // 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" +#include "yuv_row.h" extern "C" { + #define RGBY(i) { \ static_cast<int16>(1.164 * 64 * (i - 16) + 0.5), \ static_cast<int16>(1.164 * 64 * (i - 16) + 0.5), \ static_cast<int16>(1.164 * 64 * (i - 16) + 0.5), \ 0 \ +diff --git b/gfx/ycbcr/yuv_row_win.cpp a/gfx/ycbcr/yuv_row_win.cpp +index b5049a5..627b8cb 100644 +--- b/gfx/ycbcr/yuv_row_win.cpp ++++ a/gfx/ycbcr/yuv_row_win.cpp +@@ -1,20 +1,23 @@ + // 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" ++#include "yuv_row.h" ++#define MOZILLA_SSE_INCLUDE_HEADER_FOR_SSE2 ++#define MOZILLA_SSE_INCLUDE_HEADER_FOR_MMX ++#include "mozilla/SSE.h" ++ + + #define kCoefficientsRgbU kCoefficientsRgbY + 2048 + #define kCoefficientsRgbV kCoefficientsRgbY + 4096 + + extern "C" { +- +-#if USE_MMX ++#if defined(MOZILLA_COMPILE_WITH_SSE2) + __declspec(naked) + void FastConvertYUVToRGB32Row(const uint8* y_buf, + const uint8* u_buf, + const uint8* v_buf, + uint8* rgb_buf, + int width) { + __asm { + pushad +@@ -438,152 +441,37 @@ lscalelastpixel: + paddsw mm1, mm0 + psraw mm1, 6 + packuswb mm1, mm1 + movd [ebp], mm1 + popad + ret + }; } +-#else // USE_MMX +- +-// C reference code that mimic the YUV assembly. +-#define packuswb(x) ((x) < 0 ? 0 : ((x) > 255 ? 255 : (x))) +-#define paddsw(x, y) (((x) + (y)) < -32768 ? -32768 : \ +- (((x) + (y)) > 32767 ? 32767 : ((x) + (y)))) +- +-static inline void YuvPixel(uint8 y, +- uint8 u, +- uint8 v, +- uint8* rgb_buf) { +- +- int b = kCoefficientsRgbY[256+u][0]; +- int g = kCoefficientsRgbY[256+u][1]; +- int r = kCoefficientsRgbY[256+u][2]; +- int a = kCoefficientsRgbY[256+u][3]; +- +- b = paddsw(b, kCoefficientsRgbY[512+v][0]); +- g = paddsw(g, kCoefficientsRgbY[512+v][1]); +- r = paddsw(r, kCoefficientsRgbY[512+v][2]); +- a = paddsw(a, kCoefficientsRgbY[512+v][3]); +- +- b = paddsw(b, kCoefficientsRgbY[y][0]); +- g = paddsw(g, kCoefficientsRgbY[y][1]); +- r = paddsw(r, kCoefficientsRgbY[y][2]); +- a = paddsw(a, kCoefficientsRgbY[y][3]); +- +- b >>= 6; +- g >>= 6; +- r >>= 6; +- a >>= 6; +- +- *reinterpret_cast<uint32*>(rgb_buf) = (packuswb(b)) | +- (packuswb(g) << 8) | +- (packuswb(r) << 16) | +- (packuswb(a) << 24); +-} +- +-#if TEST_MMX_YUV +-static inline void YuvPixel(uint8 y, +- uint8 u, +- uint8 v, +- uint8* rgb_buf) { +- +- __asm { +- movzx eax, u +- movq mm0, [kCoefficientsRgbY+2048 + 8 * eax] +- movzx eax, v +- paddsw mm0, [kCoefficientsRgbY+4096 + 8 * eax] +- movzx eax, y +- movq mm1, [kCoefficientsRgbY + 8 * eax] +- paddsw mm1, mm0 +- psraw mm1, 6 +- packuswb mm1, mm1 +- mov eax, rgb_buf +- movd [eax], mm1 +- emms +- } +-} +-#endif +- ++#else // MOZILLA_COMPILE_WITH_SSE2 + void FastConvertYUVToRGB32Row(const uint8* y_buf, + const uint8* u_buf, + const uint8* v_buf, + uint8* rgb_buf, + int width) { +- for (int x = 0; x < width; x += 2) { +- uint8 u = u_buf[x >> 1]; +- uint8 v = v_buf[x >> 1]; +- uint8 y0 = y_buf[x]; +- YuvPixel(y0, u, v, rgb_buf); +- if ((x + 1) < width) { +- uint8 y1 = y_buf[x + 1]; +- YuvPixel(y1, u, v, rgb_buf + 4); +- } +- rgb_buf += 8; // Advance 2 pixels. +- } ++ FastConvertYUVToRGB32Row_C(y_buf, u_buf, v_buf, rgb_buf, width, 1); + } + +-// 16.16 fixed point is used. A shift by 16 isolates the integer. +-// A shift by 17 is used to further subsample the chrominence channels. +-// & 0xffff isolates the fixed point fraction. >> 2 to get the upper 2 bits, +-// for 1/65536 pixel accurate interpolation. + void ScaleYUVToRGB32Row(const uint8* y_buf, + const uint8* u_buf, + const uint8* v_buf, + uint8* rgb_buf, + int width, + int source_dx) { +- int x = 0; +- for (int i = 0; i < width; i += 2) { +- int y = y_buf[x >> 16]; +- int u = u_buf[(x >> 17)]; +- int v = v_buf[(x >> 17)]; +- YuvPixel(y, u, v, rgb_buf); +- x += source_dx; +- if ((i + 1) < width) { +- y = y_buf[x >> 16]; +- YuvPixel(y, u, v, rgb_buf+4); +- x += source_dx; +- } +- rgb_buf += 8; +- } ++ ScaleYUVToRGB32Row_C(y_buf, u_buf, v_buf, rgb_buf, width, source_dx); + } + + void LinearScaleYUVToRGB32Row(const uint8* y_buf, + const uint8* u_buf, + const uint8* v_buf, + uint8* rgb_buf, + int width, + int source_dx) { +- int x = 0; +- if (source_dx >= 0x20000) { +- x = 32768; +- } +- for (int i = 0; i < width; i += 2) { +- int y0 = y_buf[x >> 16]; +- int y1 = y_buf[(x >> 16) + 1]; +- int u0 = u_buf[(x >> 17)]; +- int u1 = u_buf[(x >> 17) + 1]; +- int v0 = v_buf[(x >> 17)]; +- int v1 = v_buf[(x >> 17) + 1]; +- int y_frac = (x & 65535); +- int uv_frac = ((x >> 1) & 65535); +- int y = (y_frac * y1 + (y_frac ^ 65535) * y0) >> 16; +- int u = (uv_frac * u1 + (uv_frac ^ 65535) * u0) >> 16; +- int v = (uv_frac * v1 + (uv_frac ^ 65535) * v0) >> 16; +- YuvPixel(y, u, v, rgb_buf); +- x += source_dx; +- if ((i + 1) < width) { +- y0 = y_buf[x >> 16]; +- y1 = y_buf[(x >> 16) + 1]; +- y_frac = (x & 65535); +- y = (y_frac * y1 + (y_frac ^ 65535) * y0) >> 16; +- YuvPixel(y, u, v, rgb_buf+4); +- x += source_dx; +- } +- rgb_buf += 8; +- } ++ LinearScaleYUVToRGB32Row_C(y_buf, u_buf, v_buf, rgb_buf, width, source_dx); + } +- +-#endif // USE_MMX ++#endif + } // extern "C" +
deleted file mode 100644 --- a/gfx/ycbcr/export.patch +++ /dev/null @@ -1,43 +0,0 @@ -diff --git a/gfx/ycbcr/yuv_convert.h b/gfx/ycbcr/yuv_convert.h -index 6735b77..e624168 100644 ---- a/gfx/ycbcr/yuv_convert.h -+++ b/gfx/ycbcr/yuv_convert.h -@@ -6,6 +6,7 @@ - #define MEDIA_BASE_YUV_CONVERT_H_ - - #include "chromium_types.h" -+#include "gfxCore.h" - - namespace mozilla { - -@@ -20,18 +21,18 @@ enum YUVType { - - // Convert a frame of YUV to 32 bit ARGB. - // Pass in YV16/YV12 depending on source format --void ConvertYCbCrToRGB32(const uint8* yplane, -- const uint8* uplane, -- const uint8* vplane, -- uint8* rgbframe, -- int pic_x, -- int pic_y, -- int pic_width, -- int pic_height, -- int ystride, -- int uvstride, -- int rgbstride, -- YUVType yuv_type); -+NS_GFX_(void) ConvertYCbCrToRGB32(const uint8* yplane, -+ const uint8* uplane, -+ const uint8* vplane, -+ uint8* rgbframe, -+ int pic_x, -+ int pic_y, -+ int pic_width, -+ int pic_height, -+ int ystride, -+ int uvstride, -+ int rgbstride, -+ YUVType yuv_type); - - } // namespace gfx - } // namespace mozilla
deleted file mode 100644 --- a/gfx/ycbcr/picture_region.patch +++ /dev/null @@ -1,99 +0,0 @@ -diff --git a/gfx/ycbcr/yuv_convert.cpp b/gfx/ycbcr/yuv_convert.cpp -index c291d5c..ff7267e 100644 ---- a/gfx/ycbcr/yuv_convert.cpp -+++ b/gfx/ycbcr/yuv_convert.cpp -@@ -25,42 +25,58 @@ namespace mozilla { - - namespace gfx { - - // Convert a frame of YUV to 32 bit ARGB. - void ConvertYCbCrToRGB32(const uint8* y_buf, - const uint8* u_buf, - const uint8* v_buf, - uint8* rgb_buf, -- int width, -- int height, -+ int pic_x, -+ int pic_y, -+ int pic_width, -+ int pic_height, - int y_pitch, - int uv_pitch, - int rgb_pitch, - YUVType yuv_type) { - unsigned int y_shift = yuv_type; - bool has_mmx = supports_mmx(); -- for (int y = 0; y < height; ++y) { -- uint8* rgb_row = rgb_buf + y * rgb_pitch; -- const uint8* y_ptr = y_buf + y * y_pitch; -- const uint8* u_ptr = u_buf + (y >> y_shift) * uv_pitch; -- const uint8* v_ptr = v_buf + (y >> y_shift) * uv_pitch; -+ bool odd_pic_x = pic_x % 2 != 0; -+ int x_width = odd_pic_x ? pic_width - 1 : pic_width; -+ -+ for (int y = pic_y; y < pic_height + pic_y; ++y) { -+ uint8* rgb_row = rgb_buf + (y - pic_y) * rgb_pitch; -+ const uint8* y_ptr = y_buf + y * y_pitch + pic_x; -+ const uint8* u_ptr = u_buf + (y >> y_shift) * uv_pitch + (pic_x >> 1); -+ const uint8* v_ptr = v_buf + (y >> y_shift) * uv_pitch + (pic_x >> 1); -+ -+ if (odd_pic_x) { -+ // Handle the single odd pixel manually and use the -+ // fast routines for the remaining. -+ FastConvertYUVToRGB32Row_C(y_ptr++, -+ u_ptr++, -+ v_ptr++, -+ rgb_row, -+ 1); -+ rgb_row += 4; -+ } - - if (has_mmx) - FastConvertYUVToRGB32Row(y_ptr, - u_ptr, - v_ptr, - rgb_row, -- width); -+ x_width); - else - FastConvertYUVToRGB32Row_C(y_ptr, - u_ptr, - v_ptr, - rgb_row, -- width); -+ x_width); - } - - // MMX used for FastConvertYUVToRGB32Row requires emms instruction. - if (has_mmx) - EMMS(); - } - - // Scale a frame of YUV to 32 bit ARGB. -diff --git a/gfx/ycbcr/yuv_convert.h b/gfx/ycbcr/yuv_convert.h -index 9d148a6..77ca8e6 100644 ---- a/gfx/ycbcr/yuv_convert.h -+++ b/gfx/ycbcr/yuv_convert.h -@@ -32,18 +32,20 @@ enum Rotate { - }; - - // Convert a frame of YUV to 32 bit ARGB. - // Pass in YV16/YV12 depending on source format - void ConvertYCbCrToRGB32(const uint8* yplane, - const uint8* uplane, - const uint8* vplane, - uint8* rgbframe, -- int frame_width, -- int frame_height, -+ int pic_x, -+ int pic_y, -+ int pic_width, -+ int pic_height, - int ystride, - int uvstride, - int rgbstride, - YUVType yuv_type); - - // Scale a frame of YUV to 32 bit ARGB. - // Supports rotation and mirroring. - void ScaleYCbCrToRGB32(const uint8* yplane,
deleted file mode 100644 --- a/gfx/ycbcr/remove_scale.patch +++ /dev/null @@ -1,839 +0,0 @@ -diff --git a/gfx/ycbcr/yuv_convert.cpp b/gfx/ycbcr/yuv_convert.cpp -index eec578d..de91f79 100644 ---- a/gfx/ycbcr/yuv_convert.cpp -+++ b/gfx/ycbcr/yuv_convert.cpp -@@ -81,133 +81,5 @@ void ConvertYCbCrToRGB32(const uint8* y_buf, - EMMS(); - } - --// Scale a frame of YUV to 32 bit ARGB. --void ScaleYCbCrToRGB32(const uint8* y_buf, -- const uint8* u_buf, -- const uint8* v_buf, -- uint8* rgb_buf, -- int width, -- int height, -- int scaled_width, -- int scaled_height, -- int y_pitch, -- int uv_pitch, -- int rgb_pitch, -- YUVType yuv_type, -- Rotate view_rotate) { -- unsigned int y_shift = yuv_type; -- bool has_mmx = supports_mmx(); -- // Diagram showing origin and direction of source sampling. -- // ->0 4<- -- // 7 3 -- // -- // 6 5 -- // ->1 2<- -- // Rotations that start at right side of image. -- if ((view_rotate == ROTATE_180) || -- (view_rotate == ROTATE_270) || -- (view_rotate == MIRROR_ROTATE_0) || -- (view_rotate == MIRROR_ROTATE_90)) { -- y_buf += width - 1; -- u_buf += width / 2 - 1; -- v_buf += width / 2 - 1; -- width = -width; -- } -- // Rotations that start at bottom of image. -- if ((view_rotate == ROTATE_90) || -- (view_rotate == ROTATE_180) || -- (view_rotate == MIRROR_ROTATE_90) || -- (view_rotate == MIRROR_ROTATE_180)) { -- y_buf += (height - 1) * y_pitch; -- u_buf += ((height >> y_shift) - 1) * uv_pitch; -- v_buf += ((height >> y_shift) - 1) * uv_pitch; -- height = -height; -- } -- -- // Handle zero sized destination. -- if (scaled_width == 0 || scaled_height == 0) -- return; -- int scaled_dx = width * 16 / scaled_width; -- int scaled_dy = height * 16 / scaled_height; -- -- int scaled_dx_uv = scaled_dx; -- -- if ((view_rotate == ROTATE_90) || -- (view_rotate == ROTATE_270)) { -- int tmp = scaled_height; -- scaled_height = scaled_width; -- scaled_width = tmp; -- tmp = height; -- height = width; -- width = tmp; -- int original_dx = scaled_dx; -- int original_dy = scaled_dy; -- scaled_dx = ((original_dy >> 4) * y_pitch) << 4; -- scaled_dx_uv = ((original_dy >> 4) * uv_pitch) << 4; -- scaled_dy = original_dx; -- if (view_rotate == ROTATE_90) { -- y_pitch = -1; -- uv_pitch = -1; -- height = -height; -- } else { -- y_pitch = 1; -- uv_pitch = 1; -- } -- } -- -- for (int y = 0; y < scaled_height; ++y) { -- uint8* dest_pixel = rgb_buf + y * rgb_pitch; -- int scaled_y = (y * height / scaled_height); -- const uint8* y_ptr = y_buf + scaled_y * y_pitch; -- const uint8* u_ptr = u_buf + (scaled_y >> y_shift) * uv_pitch; -- const uint8* v_ptr = v_buf + (scaled_y >> y_shift) * uv_pitch; -- --#if defined(_MSC_VER) -- if (scaled_width == (width * 2)) { -- DoubleYUVToRGB32Row(y_ptr, u_ptr, v_ptr, -- dest_pixel, scaled_width); -- } else if ((scaled_dx & 15) == 0) { // Scaling by integer scale factor. -- if (scaled_dx_uv == scaled_dx) { // Not rotated. -- if (scaled_dx == 16) { // Not scaled -- if (has_mmx) -- FastConvertYUVToRGB32Row(y_ptr, u_ptr, v_ptr, -- dest_pixel, scaled_width); -- else -- FastConvertYUVToRGB32Row_C(y_ptr, u_ptr, v_ptr, -- dest_pixel, scaled_width); -- } else { // Simple scale down. ie half -- ConvertYUVToRGB32Row(y_ptr, u_ptr, v_ptr, -- dest_pixel, scaled_width, scaled_dx >> 4); -- } -- } else { -- RotateConvertYUVToRGB32Row(y_ptr, u_ptr, v_ptr, -- dest_pixel, scaled_width, -- scaled_dx >> 4, scaled_dx_uv >> 4); -- } --#else -- if (scaled_dx == 16) { // Not scaled -- if (has_mmx) -- FastConvertYUVToRGB32Row(y_ptr, u_ptr, v_ptr, -- dest_pixel, scaled_width); -- else -- FastConvertYUVToRGB32Row_C(y_ptr, u_ptr, v_ptr, -- dest_pixel, scaled_width); --#endif -- } else { -- if (has_mmx) -- ScaleYUVToRGB32Row(y_ptr, u_ptr, v_ptr, -- dest_pixel, scaled_width, scaled_dx); -- else -- ScaleYUVToRGB32Row_C(y_ptr, u_ptr, v_ptr, -- dest_pixel, scaled_width, scaled_dx); -- -- } -- } -- -- // MMX used for FastConvertYUVToRGB32Row requires emms instruction. -- if (has_mmx) -- EMMS(); --} -- - } // namespace gfx - } // namespace mozilla -diff --git a/gfx/ycbcr/yuv_convert.h b/gfx/ycbcr/yuv_convert.h -index 7962af7..c9bf7e0 100644 ---- a/gfx/ycbcr/yuv_convert.h -+++ b/gfx/ycbcr/yuv_convert.h -@@ -18,19 +18,6 @@ enum YUVType { - YV12 = 1 // YV12 is half width and half height chroma channels. - }; - --// Mirror means flip the image horizontally, as in looking in a mirror. --// Rotate happens after mirroring. --enum Rotate { -- ROTATE_0, // Rotation off. -- ROTATE_90, // Rotate clockwise. -- ROTATE_180, // Rotate upside down. -- ROTATE_270, // Rotate counter clockwise. -- MIRROR_ROTATE_0, // Mirror horizontally. -- MIRROR_ROTATE_90, // Mirror then Rotate clockwise. -- MIRROR_ROTATE_180, // Mirror vertically. -- MIRROR_ROTATE_270 // Transpose. --}; -- - // Convert a frame of YUV to 32 bit ARGB. - // Pass in YV16/YV12 depending on source format - void ConvertYCbCrToRGB32(const uint8* yplane, -@@ -48,22 +35,6 @@ void ConvertYCbCrToRGB32(const uint8* yplane, - int rgbstride, - YUVType yuv_type); - --// Scale a frame of YUV to 32 bit ARGB. --// Supports rotation and mirroring. --void ScaleYCbCrToRGB32(const uint8* yplane, -- const uint8* uplane, -- const uint8* vplane, -- uint8* rgbframe, -- int frame_width, -- int frame_height, -- int scaled_width, -- int scaled_height, -- int ystride, -- int uvstride, -- int rgbstride, -- YUVType yuv_type, -- Rotate view_rotate); -- - } // namespace gfx - } // namespace mozilla - -diff --git a/gfx/ycbcr/yuv_row.h b/gfx/ycbcr/yuv_row.h -index c43f713..2a82972 100644 ---- a/gfx/ycbcr/yuv_row.h -+++ b/gfx/ycbcr/yuv_row.h -@@ -28,53 +28,6 @@ void FastConvertYUVToRGB32Row_C(const uint8* y_buf, - int width); - - --// Can do 1x, half size or any scale down by an integer amount. --// Step can be negative (mirroring, rotate 180). --// This is the third fastest of the scalers. --void ConvertYUVToRGB32Row(const uint8* y_buf, -- const uint8* u_buf, -- const uint8* v_buf, -- uint8* rgb_buf, -- int width, -- int step); -- --// Rotate is like Convert, but applies different step to Y versus U and V. --// This allows rotation by 90 or 270, by stepping by stride. --// This is the forth fastest of the scalers. --void RotateConvertYUVToRGB32Row(const uint8* y_buf, -- const uint8* u_buf, -- const uint8* v_buf, -- uint8* rgb_buf, -- int width, -- int ystep, -- int uvstep); -- --// Doubler does 4 pixels at a time. Each pixel is replicated. --// This is the fastest of the scalers. --void DoubleYUVToRGB32Row(const uint8* y_buf, -- const uint8* u_buf, -- const uint8* v_buf, -- uint8* rgb_buf, -- int width); -- --// Handles arbitrary scaling up or down. --// Mirroring is supported, but not 90 or 270 degree rotation. --// Chroma is under sampled every 2 pixels for performance. --// This is the slowest of the scalers. --void ScaleYUVToRGB32Row(const uint8* y_buf, -- const uint8* u_buf, -- const uint8* v_buf, -- uint8* rgb_buf, -- int width, -- int scaled_dx); -- --void ScaleYUVToRGB32Row_C(const uint8* y_buf, -- const uint8* u_buf, -- const uint8* v_buf, -- uint8* rgb_buf, -- int width, -- int scaled_dx); -- - } // extern "C" - - // x64 uses MMX2 (SSE) so emms is not required. -diff --git a/gfx/ycbcr/yuv_row_c.cpp b/gfx/ycbcr/yuv_row_c.cpp -index a81416c..d3bdab4 100644 ---- a/gfx/ycbcr/yuv_row_c.cpp -+++ b/gfx/ycbcr/yuv_row_c.cpp -@@ -172,25 +172,5 @@ void FastConvertYUVToRGB32Row_C(const uint8* y_buf, - } - } - --// 28.4 fixed point is used. A shift by 4 isolates the integer. --// A shift by 5 is used to further subsample the chrominence channels. --// & 15 isolates the fixed point fraction. >> 2 to get the upper 2 bits, --// for 1/4 pixel accurate interpolation. --void ScaleYUVToRGB32Row_C(const uint8* y_buf, -- const uint8* u_buf, -- const uint8* v_buf, -- uint8* rgb_buf, -- int width, -- int scaled_dx) { -- int scaled_x = 0; -- for (int x = 0; x < width; ++x) { -- uint8 u = u_buf[scaled_x >> 5]; -- uint8 v = v_buf[scaled_x >> 5]; -- uint8 y0 = y_buf[scaled_x >> 4]; -- YuvPixel(y0, u, v, rgb_buf); -- rgb_buf += 4; -- scaled_x += scaled_dx; -- } --} - } // extern "C" - -diff --git a/gfx/ycbcr/yuv_row_linux.cpp b/gfx/ycbcr/yuv_row_linux.cpp -index 5fb2bc4..ce5ee89 100644 ---- a/gfx/ycbcr/yuv_row_linux.cpp -+++ b/gfx/ycbcr/yuv_row_linux.cpp -@@ -21,14 +21,6 @@ void FastConvertYUVToRGB32Row(const uint8* y_buf, - FastConvertYUVToRGB32Row_C(y_buf, u_buf, v_buf, rgb_buf, width); - } - --void ScaleYUVToRGB32Row(const uint8* y_buf, -- const uint8* u_buf, -- const uint8* v_buf, -- uint8* rgb_buf, -- int width, -- int scaled_dx) { -- ScaleYUVToRGB32Row_C(y_buf, u_buf, v_buf, rgb_buf, width, scaled_dx); --} - #else - - #define RGBY(i) { \ -@@ -315,75 +307,6 @@ void FastConvertYUVToRGB32Row(const uint8* y_buf, // rdi - ); - } - --void ScaleYUVToRGB32Row(const uint8* y_buf, // rdi -- const uint8* u_buf, // rsi -- const uint8* v_buf, // rdx -- uint8* rgb_buf, // rcx -- int width, // r8 -- int scaled_dx) { // r9 -- asm( -- "xor %%r11,%%r11\n" -- "sub $0x2,%4\n" -- "js scalenext\n" -- --"scaleloop:" -- "mov %%r11,%%r10\n" -- "sar $0x5,%%r10\n" -- "movzb (%1,%%r10,1),%%rax\n" -- "movq 2048(%5,%%rax,8),%%xmm0\n" -- "movzb (%2,%%r10,1),%%rax\n" -- "movq 4096(%5,%%rax,8),%%xmm1\n" -- "lea (%%r11,%6),%%r10\n" -- "sar $0x4,%%r11\n" -- "movzb (%0,%%r11,1),%%rax\n" -- "paddsw %%xmm1,%%xmm0\n" -- "movq (%5,%%rax,8),%%xmm1\n" -- "lea (%%r10,%6),%%r11\n" -- "sar $0x4,%%r10\n" -- "movzb (%0,%%r10,1),%%rax\n" -- "movq (%5,%%rax,8),%%xmm2\n" -- "paddsw %%xmm0,%%xmm1\n" -- "paddsw %%xmm0,%%xmm2\n" -- "shufps $0x44,%%xmm2,%%xmm1\n" -- "psraw $0x6,%%xmm1\n" -- "packuswb %%xmm1,%%xmm1\n" -- "movq %%xmm1,0x0(%3)\n" -- "add $0x8,%3\n" -- "sub $0x2,%4\n" -- "jns scaleloop\n" -- --"scalenext:" -- "add $0x1,%4\n" -- "js scaledone\n" -- -- "mov %%r11,%%r10\n" -- "sar $0x5,%%r10\n" -- "movzb (%1,%%r10,1),%%rax\n" -- "movq 2048(%5,%%rax,8),%%xmm0\n" -- "movzb (%2,%%r10,1),%%rax\n" -- "movq 4096(%5,%%rax,8),%%xmm1\n" -- "paddsw %%xmm1,%%xmm0\n" -- "sar $0x4,%%r11\n" -- "movzb (%0,%%r11,1),%%rax\n" -- "movq (%5,%%rax,8),%%xmm1\n" -- "paddsw %%xmm0,%%xmm1\n" -- "psraw $0x6,%%xmm1\n" -- "packuswb %%xmm1,%%xmm1\n" -- "movd %%xmm1,0x0(%3)\n" -- --"scaledone:" -- : -- : "r"(y_buf), // %0 -- "r"(u_buf), // %1 -- "r"(v_buf), // %2 -- "r"(rgb_buf), // %3 -- "r"(width), // %4 -- "r" (kCoefficientsRgbY), // %5 -- "r"(static_cast<long>(scaled_dx)) // %6 -- : "memory", "r10", "r11", "rax", "xmm0", "xmm1", "xmm2" --); --} -- - #else - - void FastConvertYUVToRGB32Row(const uint8* y_buf, -@@ -443,81 +366,6 @@ void FastConvertYUVToRGB32Row(const uint8* y_buf, - "ret\n" - ); - -- --void ScaleYUVToRGB32Row(const uint8* y_buf, -- const uint8* u_buf, -- const uint8* v_buf, -- uint8* rgb_buf, -- int width, -- int scaled_dx); -- -- asm( -- ".global ScaleYUVToRGB32Row\n" --"ScaleYUVToRGB32Row:\n" -- "pusha\n" -- "mov 0x24(%esp),%edx\n" -- "mov 0x28(%esp),%edi\n" -- "mov 0x2c(%esp),%esi\n" -- "mov 0x30(%esp),%ebp\n" -- "mov 0x34(%esp),%ecx\n" -- "xor %ebx,%ebx\n" -- "jmp scaleend\n" -- --"scaleloop:" -- "mov %ebx,%eax\n" -- "sar $0x5,%eax\n" -- "movzbl (%edi,%eax,1),%eax\n" -- "movq kCoefficientsRgbY+2048(,%eax,8),%mm0\n" -- "mov %ebx,%eax\n" -- "sar $0x5,%eax\n" -- "movzbl (%esi,%eax,1),%eax\n" -- "paddsw kCoefficientsRgbY+4096(,%eax,8),%mm0\n" -- "mov %ebx,%eax\n" -- "add 0x38(%esp),%ebx\n" -- "sar $0x4,%eax\n" -- "movzbl (%edx,%eax,1),%eax\n" -- "movq kCoefficientsRgbY(,%eax,8),%mm1\n" -- "mov %ebx,%eax\n" -- "add 0x38(%esp),%ebx\n" -- "sar $0x4,%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" -- "movntq %mm1,0x0(%ebp)\n" -- "add $0x8,%ebp\n" --"scaleend:" -- "sub $0x2,%ecx\n" -- "jns scaleloop\n" -- -- "and $0x1,%ecx\n" -- "je scaledone\n" -- -- "mov %ebx,%eax\n" -- "sar $0x5,%eax\n" -- "movzbl (%edi,%eax,1),%eax\n" -- "movq kCoefficientsRgbY+2048(,%eax,8),%mm0\n" -- "mov %ebx,%eax\n" -- "sar $0x5,%eax\n" -- "movzbl (%esi,%eax,1),%eax\n" -- "paddsw kCoefficientsRgbY+4096(,%eax,8),%mm0\n" -- "mov %ebx,%eax\n" -- "sar $0x4,%eax\n" -- "movzbl (%edx,%eax,1),%eax\n" -- "movq kCoefficientsRgbY(,%eax,8),%mm1\n" -- "paddsw %mm0,%mm1\n" -- "psraw $0x6,%mm1\n" -- "packuswb %mm1,%mm1\n" -- "movd %mm1,0x0(%ebp)\n" -- --"scaledone:" -- "popa\n" -- "ret\n" --); -- - #endif - #endif // ARCH_CPU_ARM_FAMILY - } // extern "C" -diff --git a/gfx/ycbcr/yuv_row_mac.cpp b/gfx/ycbcr/yuv_row_mac.cpp -index a7e8243..3515ada 100644 ---- a/gfx/ycbcr/yuv_row_mac.cpp -+++ b/gfx/ycbcr/yuv_row_mac.cpp -@@ -18,14 +18,6 @@ void FastConvertYUVToRGB32Row(const uint8* y_buf, - FastConvertYUVToRGB32Row_C(y_buf, u_buf, v_buf, rgb_buf, width); - } - --void ScaleYUVToRGB32Row(const uint8* y_buf, -- const uint8* u_buf, -- const uint8* v_buf, -- uint8* rgb_buf, -- int width, -- int scaled_dx) { -- ScaleYUVToRGB32Row_C(y_buf, u_buf, v_buf, rgb_buf, width, scaled_dx); --} - #else - - #define RGBY(i) { \ -@@ -323,91 +315,6 @@ void FastConvertYUVToRGB32Row(const uint8* y_buf, - &kCoefficientsRgbY[0][0]); - } - --extern void MacScaleYUVToRGB32Row(const uint8* y_buf, -- const uint8* u_buf, -- const uint8* v_buf, -- uint8* rgb_buf, -- int width, -- int scaled_dx, -- int16 *kCoefficientsRgbY); -- -- __asm__( --"_MacScaleYUVToRGB32Row:\n" -- "pusha\n" -- "mov 0x24(%esp),%edx\n" -- "mov 0x28(%esp),%edi\n" -- "mov 0x2c(%esp),%esi\n" -- "mov 0x30(%esp),%ebp\n" -- "mov 0x3c(%esp),%ecx\n" -- "xor %ebx,%ebx\n" -- "jmp Lscaleend\n" -- --"Lscaleloop:" -- "mov %ebx,%eax\n" -- "sar $0x5,%eax\n" -- "movzbl (%edi,%eax,1),%eax\n" -- "movq 2048(%ecx,%eax,8),%mm0\n" -- "mov %ebx,%eax\n" -- "sar $0x5,%eax\n" -- "movzbl (%esi,%eax,1),%eax\n" -- "paddsw 4096(%ecx,%eax,8),%mm0\n" -- "mov %ebx,%eax\n" -- "add 0x38(%esp),%ebx\n" -- "sar $0x4,%eax\n" -- "movzbl (%edx,%eax,1),%eax\n" -- "movq 0(%ecx,%eax,8),%mm1\n" -- "mov %ebx,%eax\n" -- "add 0x38(%esp),%ebx\n" -- "sar $0x4,%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" -- "movntq %mm1,0x0(%ebp)\n" -- "add $0x8,%ebp\n" --"Lscaleend:" -- "sub $0x2,0x34(%esp)\n" -- "jns Lscaleloop\n" -- -- "and $0x1,0x34(%esp)\n" -- "je Lscaledone\n" -- -- "mov %ebx,%eax\n" -- "sar $0x5,%eax\n" -- "movzbl (%edi,%eax,1),%eax\n" -- "movq 2048(%ecx,%eax,8),%mm0\n" -- "mov %ebx,%eax\n" -- "sar $0x5,%eax\n" -- "movzbl (%esi,%eax,1),%eax\n" -- "paddsw 4096(%ecx,%eax,8),%mm0\n" -- "mov %ebx,%eax\n" -- "sar $0x4,%eax\n" -- "movzbl (%edx,%eax,1),%eax\n" -- "movq 0(%ecx,%eax,8),%mm1\n" -- "paddsw %mm0,%mm1\n" -- "psraw $0x6,%mm1\n" -- "packuswb %mm1,%mm1\n" -- "movd %mm1,0x0(%ebp)\n" -- --"Lscaledone:" -- "popa\n" -- "ret\n" --); -- -- --void ScaleYUVToRGB32Row(const uint8* y_buf, -- const uint8* u_buf, -- const uint8* v_buf, -- uint8* rgb_buf, -- int width, -- int scaled_dx) { -- -- MacScaleYUVToRGB32Row(y_buf, u_buf, v_buf, rgb_buf, width, scaled_dx, -- &kCoefficientsRgbY[0][0]); --} - #endif // ARCH_CPU_PPC - } // extern "C" - -diff --git a/gfx/ycbcr/yuv_row_win.cpp b/gfx/ycbcr/yuv_row_win.cpp -index a77a16f..f994931 100644 ---- a/gfx/ycbcr/yuv_row_win.cpp -+++ b/gfx/ycbcr/yuv_row_win.cpp -@@ -297,273 +297,5 @@ void FastConvertYUVToRGB32Row(const uint8* y_buf, - } - } - --__declspec(naked) --void ConvertYUVToRGB32Row(const uint8* y_buf, -- const uint8* u_buf, -- const uint8* v_buf, -- uint8* rgb_buf, -- int width, -- int step) { -- __asm { -- pushad -- mov edx, [esp + 32 + 4] // Y -- mov edi, [esp + 32 + 8] // U -- mov esi, [esp + 32 + 12] // V -- mov ebp, [esp + 32 + 16] // rgb -- mov ecx, [esp + 32 + 20] // width -- mov ebx, [esp + 32 + 24] // step -- jmp wend -- -- wloop : -- movzx eax, byte ptr [edi] -- add edi, ebx -- movq mm0, [kCoefficientsRgbU + 8 * eax] -- movzx eax, byte ptr [esi] -- add esi, ebx -- paddsw mm0, [kCoefficientsRgbV + 8 * eax] -- movzx eax, byte ptr [edx] -- 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 -- movntq [ebp], mm1 -- 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] -- movzx eax, byte ptr [esi] -- paddsw mm0, [kCoefficientsRgbV + 8 * eax] -- movzx eax, byte ptr [edx] -- movq mm1, [kCoefficientsRgbY + 8 * eax] -- paddsw mm1, mm0 -- psraw mm1, 6 -- packuswb mm1, mm1 -- movd [ebp], mm1 -- wdone : -- -- popad -- ret -- } --} -- --__declspec(naked) --void RotateConvertYUVToRGB32Row(const uint8* y_buf, -- const uint8* u_buf, -- const uint8* v_buf, -- uint8* rgb_buf, -- int width, -- int ystep, -- int uvstep) { -- __asm { -- pushad -- mov edx, [esp + 32 + 4] // Y -- mov edi, [esp + 32 + 8] // U -- mov esi, [esp + 32 + 12] // V -- mov ebp, [esp + 32 + 16] // rgb -- mov ecx, [esp + 32 + 20] // width -- jmp wend -- -- wloop : -- movzx eax, byte ptr [edi] -- mov ebx, [esp + 32 + 28] // uvstep -- add edi, ebx -- movq mm0, [kCoefficientsRgbU + 8 * eax] -- movzx eax, byte ptr [esi] -- add esi, ebx -- paddsw mm0, [kCoefficientsRgbV + 8 * eax] -- movzx eax, byte ptr [edx] -- mov ebx, [esp + 32 + 24] // ystep -- 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 -- movntq [ebp], mm1 -- 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] -- movzx eax, byte ptr [esi] -- paddsw mm0, [kCoefficientsRgbV + 8 * eax] -- movzx eax, byte ptr [edx] -- movq mm1, [kCoefficientsRgbY + 8 * eax] -- paddsw mm1, mm0 -- psraw mm1, 6 -- packuswb mm1, mm1 -- movd [ebp], mm1 -- wdone : -- -- popad -- ret -- } --} -- --__declspec(naked) --void DoubleYUVToRGB32Row(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 -- mov edi, [esp + 32 + 8] // U -- mov esi, [esp + 32 + 12] // V -- mov ebp, [esp + 32 + 16] // rgb -- mov ecx, [esp + 32 + 20] // width -- jmp wend -- -- wloop : -- movzx eax, byte ptr [edi] -- add edi, 1 -- 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 -- movntq [ebp], mm1 -- -- movzx ebx, byte ptr [edx + 1] -- add edx, 2 -- paddsw mm0, [kCoefficientsRgbY + 8 * ebx] -- psraw mm0, 6 -- packuswb mm0, mm0 -- punpckldq mm0, mm0 -- movntq [ebp+8], mm0 -- add ebp, 16 -- wend : -- sub ecx, 4 -- jns wloop -- -- add ecx, 4 -- jz wdone -- -- movzx eax, byte ptr [edi] -- movq mm0, [kCoefficientsRgbU + 8 * eax] -- movzx eax, byte ptr [esi] -- paddsw mm0, [kCoefficientsRgbV + 8 * eax] -- movzx eax, byte ptr [edx] -- movq mm1, [kCoefficientsRgbY + 8 * eax] -- paddsw mm1, mm0 -- psraw mm1, 6 -- packuswb mm1, mm1 -- jmp wend1 -- -- wloop1 : -- movd [ebp], mm1 -- add ebp, 4 -- wend1 : -- sub ecx, 1 -- jns wloop1 -- wdone : -- popad -- ret -- } --} -- --// This version does general purpose scaling by any amount, up or down. --// The only thing it can not do it rotation by 90 or 270. --// For performance the chroma is under sampled, reducing cost of a 3x --// 1080p scale from 8.4 ms to 5.4 ms. --__declspec(naked) --void ScaleYUVToRGB32Row(const uint8* y_buf, -- const uint8* u_buf, -- const uint8* v_buf, -- uint8* rgb_buf, -- int width, -- int dx) { -- __asm { -- pushad -- mov edx, [esp + 32 + 4] // Y -- mov edi, [esp + 32 + 8] // U -- mov esi, [esp + 32 + 12] // V -- mov ebp, [esp + 32 + 16] // rgb -- mov ecx, [esp + 32 + 20] // width -- xor ebx, ebx // x -- jmp scaleend -- -- scaleloop : -- mov eax, ebx -- sar eax, 5 -- movzx eax, byte ptr [edi + eax] -- movq mm0, [kCoefficientsRgbU + 8 * eax] -- mov eax, ebx -- sar eax, 5 -- movzx eax, byte ptr [esi + eax] -- paddsw mm0, [kCoefficientsRgbV + 8 * eax] -- mov eax, ebx -- add ebx, [esp + 32 + 24] // x += dx -- sar eax, 4 -- movzx eax, byte ptr [edx + eax] -- movq mm1, [kCoefficientsRgbY + 8 * eax] -- mov eax, ebx -- add ebx, [esp + 32 + 24] // x += dx -- sar eax, 4 -- 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 -- movntq [ebp], mm1 -- add ebp, 8 -- scaleend : -- sub ecx, 2 -- jns scaleloop -- -- and ecx, 1 // odd number of pixels? -- jz scaledone -- -- mov eax, ebx -- sar eax, 5 -- movzx eax, byte ptr [edi + eax] -- movq mm0, [kCoefficientsRgbU + 8 * eax] -- mov eax, ebx -- sar eax, 5 -- movzx eax, byte ptr [esi + eax] -- paddsw mm0, [kCoefficientsRgbV + 8 * eax] -- mov eax, ebx -- sar eax, 4 -- movzx eax, byte ptr [edx + eax] -- movq mm1, [kCoefficientsRgbY + 8 * eax] -- paddsw mm1, mm0 -- psraw mm1, 6 -- packuswb mm1, mm1 -- movd [ebp], mm1 -- -- scaledone : -- popad -- ret -- } --} - } // extern "C" -
deleted file mode 100644 --- a/gfx/ycbcr/row_c_fix.patch +++ /dev/null @@ -1,23 +0,0 @@ -diff --git a/gfx/ycbcr/yuv_row_c.cpp b/gfx/ycbcr/yuv_row_c.cpp -index 36d9bda..b5c0018 100644 ---- a/gfx/ycbcr/yuv_row_c.cpp -+++ b/gfx/ycbcr/yuv_row_c.cpp -@@ -142,17 +142,17 @@ static inline void YuvPixel(uint8 y, - uint8* rgb_buf) { - int32 d = static_cast<int32>(u) - 128; - int32 e = static_cast<int32>(v) - 128; - - int32 cb = (516 * d + 128); - int32 cg = (- 100 * d - 208 * e + 128); - int32 cr = (409 * e + 128); - -- int32 C298a = ((static_cast<int32>(y) - 16) * 298 + 128); -+ int32 C298a = ((static_cast<int32>(y) - 16) * 298); - *reinterpret_cast<uint32*>(rgb_buf) = (clip(C298a + cb)) | - (clip(C298a + cg) << 8) | - (clip(C298a + cr) << 16) | - (0xff000000); - } - - void FastConvertYUVToRGB32Row_C(const uint8* y_buf, - const uint8* u_buf,
--- a/gfx/ycbcr/update.sh +++ b/gfx/ycbcr/update.sh @@ -1,18 +1,9 @@ # update.sh <chromium-src-directory> cp $1/media/base/yuv_convert.h . cp $1/media/base/yuv_convert.cc yuv_convert.cpp cp $1/media/base/yuv_row.h . -cp $1/media/base/yuv_row_linux.cc yuv_row_linux.cpp -cp $1/media/base/yuv_row_mac.cc yuv_row_mac.cpp +cp $1/media/base/yuv_row_table.cc yuv_row_table.cpp +cp $1/media/base/yuv_row_posix.cc yuv_row_posix.cpp cp $1/media/base/yuv_row_win.cc yuv_row_win.cpp -cp $1/media/base/yuv_row_linux.cc yuv_row_c.cpp +cp $1/media/base/yuv_row_posix.cc yuv_row_c.cpp patch -p3 <convert.patch -patch -p3 <picture_region.patch -patch -p3 <remove_scale.patch -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 -patch -p3 <bug577645_movntq.patch -patch -p3 <add_scale.patch
deleted file mode 100644 --- a/gfx/ycbcr/win64_mac64.patch +++ /dev/null @@ -1,57 +0,0 @@ -diff --git a/gfx/ycbcr/yuv_row_mac.cpp b/gfx/ycbcr/yuv_row_mac.cpp -index 351466c..2a679cc 100644 ---- a/gfx/ycbcr/yuv_row_mac.cpp -+++ b/gfx/ycbcr/yuv_row_mac.cpp -@@ -8,7 +8,10 @@ - - extern "C" { - --#if defined(ARCH_CPU_PPC) -+// PPC and 64 Bit builds use the C fallback. Optimized code -+// needs to be fixed for 64 bit builds. PPC has no optimized code -+// option at all. -+#if defined(ARCH_CPU_PPC) || defined(ARCH_CPU_64_BITS) - // PPC implementation uses C fallback - void FastConvertYUVToRGB32Row(const uint8* y_buf, - const uint8* u_buf, -@@ -315,6 +318,6 @@ void FastConvertYUVToRGB32Row(const uint8* y_buf, - &kCoefficientsRgbY[0][0]); - } - --#endif // ARCH_CPU_PPC -+#endif // ARCH_CPU_PPC || ARCH_CPU_64_BITS - } // extern "C" - -diff --git a/gfx/ycbcr/yuv_row_win.cpp b/gfx/ycbcr/yuv_row_win.cpp -index f994931..708ef14 100644 ---- a/gfx/ycbcr/yuv_row_win.cpp -+++ b/gfx/ycbcr/yuv_row_win.cpp -@@ -5,6 +5,21 @@ - #include "yuv_row.h" - - extern "C" { -+// 64 Bit builds use the C fallback. Optimized code -+// needs to be fixed for 64 bit builds. -+#if defined(ARCH_CPU_64_BITS) -+// PPC implementation uses C fallback -+void FastConvertYUVToRGB32Row(const uint8* y_buf, -+ const uint8* u_buf, -+ const uint8* v_buf, -+ uint8* rgb_buf, -+ int width) { -+ FastConvertYUVToRGB32Row_C(y_buf, u_buf, v_buf, rgb_buf, width); -+} -+ -+#else -+ -+ - #define RGBY(i) { \ - static_cast<int16>(1.164 * 64 * (i - 16) + 0.5), \ - static_cast<int16>(1.164 * 64 * (i - 16) + 0.5), \ -@@ -297,5 +312,6 @@ void FastConvertYUVToRGB32Row(const uint8* y_buf, - } - } - -+#endif // ARCH_CPU_64_BITS - } // extern "C" -
--- a/gfx/ycbcr/yuv_convert.cpp +++ b/gfx/ycbcr/yuv_convert.cpp @@ -1,9 +1,9 @@ -// Copyright (c) 2009 The Chromium Authors. All rights reserved. +// 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. // This webpage shows layout of YV12 and other YUV formats // http://www.fourcc.org/yuv.php // The actual conversion is best described here // http://en.wikipedia.org/wiki/YUV // An article on optimizing YUV conversion using tables instead of multiplies @@ -15,21 +15,28 @@ // // ARGB pixel format is output, which on little endian is stored as BGRA. // The alpha is set to 255, allowing the application to use RGBA or RGB32. #include "yuv_convert.h" // Header for low level row functions. #include "yuv_row.h" +#define MOZILLA_SSE_INCLUDE_HEADER_FOR_SSE2 +#define MOZILLA_SSE_INCLUDE_HEADER_FOR_MMX #include "mozilla/SSE.h" namespace mozilla { namespace gfx { + +// 16.16 fixed point arithmetic +const int kFractionBits = 16; +const int kFractionMax = 1 << kFractionBits; +const int kFractionMask = ((1 << kFractionBits) - 1); // Convert a frame of YUV to 32 bit ARGB. NS_GFX_(void) ConvertYCbCrToRGB32(const uint8* y_buf, const uint8* u_buf, const uint8* v_buf, uint8* rgb_buf, int pic_x, int pic_y, @@ -62,161 +69,308 @@ NS_GFX_(void) ConvertYCbCrToRGB32(const u_ptr++, v_ptr++, rgb_row, 1, x_shift); rgb_row += 4; } - if (has_sse) + if (has_sse) { FastConvertYUVToRGB32Row(y_ptr, u_ptr, v_ptr, rgb_row, x_width); - else + } + else { FastConvertYUVToRGB32Row_C(y_ptr, u_ptr, v_ptr, rgb_row, x_width, x_shift); + } } -#ifdef ARCH_CPU_X86_FAMILY - // SSE used for FastConvertYUVToRGB32Row requires emms instruction. + // MMX used for FastConvertYUVToRGB32Row requires emms instruction. if (has_sse) EMMS(); +} + +#if defined(MOZILLA_COMPILE_WITH_SSE2) +// FilterRows combines two rows of the image using linear interpolation. +// SSE2 version does 16 pixels at a time +static void FilterRows(uint8* ybuf, const uint8* y0_ptr, const uint8* y1_ptr, + int source_width, int source_y_fraction) { + __m128i zero = _mm_setzero_si128(); + __m128i y1_fraction = _mm_set1_epi16(source_y_fraction); + __m128i y0_fraction = _mm_set1_epi16(256 - source_y_fraction); + + const __m128i* y0_ptr128 = reinterpret_cast<const __m128i*>(y0_ptr); + const __m128i* y1_ptr128 = reinterpret_cast<const __m128i*>(y1_ptr); + __m128i* dest128 = reinterpret_cast<__m128i*>(ybuf); + __m128i* end128 = reinterpret_cast<__m128i*>(ybuf + source_width); + + do { + __m128i y0 = _mm_loadu_si128(y0_ptr128); + __m128i y1 = _mm_loadu_si128(y1_ptr128); + __m128i y2 = _mm_unpackhi_epi8(y0, zero); + __m128i y3 = _mm_unpackhi_epi8(y1, zero); + y0 = _mm_unpacklo_epi8(y0, zero); + y1 = _mm_unpacklo_epi8(y1, zero); + y0 = _mm_mullo_epi16(y0, y0_fraction); + y1 = _mm_mullo_epi16(y1, y1_fraction); + y2 = _mm_mullo_epi16(y2, y0_fraction); + y3 = _mm_mullo_epi16(y3, y1_fraction); + y0 = _mm_add_epi16(y0, y1); + y2 = _mm_add_epi16(y2, y3); + y0 = _mm_srli_epi16(y0, 8); + y2 = _mm_srli_epi16(y2, 8); + y0 = _mm_packus_epi16(y0, y2); + *dest128++ = y0; + ++y0_ptr128; + ++y1_ptr128; + } while (dest128 < end128); +} +#elif defined(MOZILLA_COMPILE_WITH_MMX) +// MMX version does 8 pixels at a time +static void FilterRows(uint8* ybuf, const uint8* y0_ptr, const uint8* y1_ptr, + int source_width, int source_y_fraction) { + __m64 zero = _mm_setzero_si64(); + __m64 y1_fraction = _mm_set1_pi16(source_y_fraction); + __m64 y0_fraction = _mm_set1_pi16(256 - source_y_fraction); + + const __m64* y0_ptr64 = reinterpret_cast<const __m64*>(y0_ptr); + const __m64* y1_ptr64 = reinterpret_cast<const __m64*>(y1_ptr); + __m64* dest64 = reinterpret_cast<__m64*>(ybuf); + __m64* end64 = reinterpret_cast<__m64*>(ybuf + source_width); + + do { + __m64 y0 = *y0_ptr64++; + __m64 y1 = *y1_ptr64++; + __m64 y2 = _mm_unpackhi_pi8(y0, zero); + __m64 y3 = _mm_unpackhi_pi8(y1, zero); + y0 = _mm_unpacklo_pi8(y0, zero); + y1 = _mm_unpacklo_pi8(y1, zero); + y0 = _mm_mullo_pi16(y0, y0_fraction); + y1 = _mm_mullo_pi16(y1, y1_fraction); + y2 = _mm_mullo_pi16(y2, y0_fraction); + y3 = _mm_mullo_pi16(y3, y1_fraction); + y0 = _mm_add_pi16(y0, y1); + y2 = _mm_add_pi16(y2, y3); + y0 = _mm_srli_pi16(y0, 8); + y2 = _mm_srli_pi16(y2, 8); + y0 = _mm_packs_pu16(y0, y2); + *dest64++ = y0; + } while (dest64 < end64); +} +#else // no MMX or SSE2 +// C version does 8 at a time to mimic MMX code +static void FilterRows(uint8* ybuf, const uint8* y0_ptr, const uint8* y1_ptr, + int source_width, int source_y_fraction) { + int y1_fraction = source_y_fraction; + int y0_fraction = 256 - y1_fraction; + uint8* end = ybuf + source_width; + do { + ybuf[0] = (y0_ptr[0] * y0_fraction + y1_ptr[0] * y1_fraction) >> 8; + ybuf[1] = (y0_ptr[1] * y0_fraction + y1_ptr[1] * y1_fraction) >> 8; + ybuf[2] = (y0_ptr[2] * y0_fraction + y1_ptr[2] * y1_fraction) >> 8; + ybuf[3] = (y0_ptr[3] * y0_fraction + y1_ptr[3] * y1_fraction) >> 8; + ybuf[4] = (y0_ptr[4] * y0_fraction + y1_ptr[4] * y1_fraction) >> 8; + ybuf[5] = (y0_ptr[5] * y0_fraction + y1_ptr[5] * y1_fraction) >> 8; + ybuf[6] = (y0_ptr[6] * y0_fraction + y1_ptr[6] * y1_fraction) >> 8; + ybuf[7] = (y0_ptr[7] * y0_fraction + y1_ptr[7] * y1_fraction) >> 8; + y0_ptr += 8; + y1_ptr += 8; + ybuf += 8; + } while (ybuf < end); +} #endif -} // Scale a frame of YUV to 32 bit ARGB. NS_GFX_(void) ScaleYCbCrToRGB32(const uint8* y_buf, const uint8* u_buf, const uint8* v_buf, uint8* rgb_buf, + int source_width, + int source_height, int width, int height, - int scaled_width, - int scaled_height, int y_pitch, int uv_pitch, int rgb_pitch, YUVType yuv_type, - Rotate view_rotate) { + Rotate view_rotate, + ScaleFilter filter) { + bool has_mmx = supports_mmx(); + + // 4096 allows 3 buffers to fit in 12k. + // Helps performance on CPU with 16K L1 cache. + // Large enough for 3830x2160 and 30" displays which are 2560x1600. + const int kFilterBufferSize = 4096; + // Disable filtering if the screen is too big (to avoid buffer overflows). + // This should never happen to regular users: they don't have monitors + // wider than 4096 pixels. + // TODO(fbarchard): Allow rotated videos to filter. + if (source_width > kFilterBufferSize || view_rotate) + filter = FILTER_NONE; + unsigned int y_shift = yuv_type == YV12 ? 1 : 0; - unsigned int x_shift = yuv_type == YV24 ? 0 : 1; - bool has_mmx = supports_mmx(); // Diagram showing origin and direction of source sampling. // ->0 4<- // 7 3 // // 6 5 // ->1 2<- // Rotations that start at right side of image. if ((view_rotate == ROTATE_180) || (view_rotate == ROTATE_270) || (view_rotate == MIRROR_ROTATE_0) || (view_rotate == MIRROR_ROTATE_90)) { - y_buf += width - 1; - u_buf += width / 2 - 1; - v_buf += width / 2 - 1; - width = -width; + y_buf += source_width - 1; + u_buf += source_width / 2 - 1; + v_buf += source_width / 2 - 1; + source_width = -source_width; } // Rotations that start at bottom of image. if ((view_rotate == ROTATE_90) || (view_rotate == ROTATE_180) || (view_rotate == MIRROR_ROTATE_90) || (view_rotate == MIRROR_ROTATE_180)) { - y_buf += (height - 1) * y_pitch; - u_buf += ((height >> y_shift) - 1) * uv_pitch; - v_buf += ((height >> y_shift) - 1) * uv_pitch; - height = -height; + y_buf += (source_height - 1) * y_pitch; + u_buf += ((source_height >> y_shift) - 1) * uv_pitch; + v_buf += ((source_height >> y_shift) - 1) * uv_pitch; + source_height = -source_height; } // Handle zero sized destination. - if (scaled_width == 0 || scaled_height == 0) + if (width == 0 || height == 0) return; - int scaled_dx = width * 16 / scaled_width; - int scaled_dy = height * 16 / scaled_height; - - int scaled_dx_uv = scaled_dx; + int source_dx = source_width * kFractionMax / width; + int source_dy = source_height * kFractionMax / height; + int source_dx_uv = source_dx; if ((view_rotate == ROTATE_90) || (view_rotate == ROTATE_270)) { - int tmp = scaled_height; - scaled_height = scaled_width; - scaled_width = tmp; - tmp = height; + int tmp = height; height = width; width = tmp; - int original_dx = scaled_dx; - int original_dy = scaled_dy; - scaled_dx = ((original_dy >> 4) * y_pitch) << 4; - scaled_dx_uv = ((original_dy >> 4) * uv_pitch) << 4; - scaled_dy = original_dx; + tmp = source_height; + source_height = source_width; + source_width = tmp; + int original_dx = source_dx; + int original_dy = source_dy; + source_dx = ((original_dy >> kFractionBits) * y_pitch) << kFractionBits; + source_dx_uv = ((original_dy >> kFractionBits) * uv_pitch) << kFractionBits; + source_dy = original_dx; if (view_rotate == ROTATE_90) { y_pitch = -1; uv_pitch = -1; - height = -height; + source_height = -source_height; } else { y_pitch = 1; uv_pitch = 1; } } - for (int y = 0; y < scaled_height; ++y) { + // Need padding because FilterRows() will write 1 to 16 extra pixels + // after the end for SSE2 version. + uint8 yuvbuf[16 + kFilterBufferSize * 3 + 16]; + uint8* ybuf = + reinterpret_cast<uint8*>(reinterpret_cast<PRUptrdiff>(yuvbuf + 15) & ~15); + uint8* ubuf = ybuf + kFilterBufferSize; + uint8* vbuf = ubuf + kFilterBufferSize; + // TODO(fbarchard): Fixed point math is off by 1 on negatives. + int yscale_fixed = (source_height << kFractionBits) / height; + + // TODO(fbarchard): Split this into separate function for better efficiency. + for (int y = 0; y < height; ++y) { uint8* dest_pixel = rgb_buf + y * rgb_pitch; - int scaled_y = (y * height / scaled_height); - const uint8* y_ptr = y_buf + scaled_y * y_pitch; - const uint8* u_ptr = u_buf + (scaled_y >> y_shift) * uv_pitch; - const uint8* v_ptr = v_buf + (scaled_y >> y_shift) * uv_pitch; + int source_y_subpixel = (y * yscale_fixed); + if (yscale_fixed >= (kFractionMax * 2)) { + source_y_subpixel += kFractionMax / 2; // For 1/2 or less, center filter. + } + int source_y = source_y_subpixel >> kFractionBits; -#if defined(_MSC_VER) && defined(_M_IX86) - if (scaled_width == (width * 2)) { - DoubleYUVToRGB32Row(y_ptr, u_ptr, v_ptr, - dest_pixel, scaled_width); - } else if ((scaled_dx & 15) == 0) { // Scaling by integer scale factor. - if (scaled_dx_uv == scaled_dx) { // Not rotated. - if (scaled_dx == 16) { // Not scaled - if (has_mmx) - FastConvertYUVToRGB32Row(y_ptr, u_ptr, v_ptr, - dest_pixel, scaled_width); - else - FastConvertYUVToRGB32Row_C(y_ptr, u_ptr, v_ptr, - dest_pixel, scaled_width, x_shift); - } else { // Simple scale down. ie half - ConvertYUVToRGB32Row(y_ptr, u_ptr, v_ptr, - dest_pixel, scaled_width, scaled_dx >> 4); - } + const uint8* y0_ptr = y_buf + source_y * y_pitch; + const uint8* y1_ptr = y0_ptr + y_pitch; + + const uint8* u0_ptr = u_buf + (source_y >> y_shift) * uv_pitch; + const uint8* u1_ptr = u0_ptr + uv_pitch; + const uint8* v0_ptr = v_buf + (source_y >> y_shift) * uv_pitch; + const uint8* v1_ptr = v0_ptr + uv_pitch; + + // vertical scaler uses 16.8 fixed point + int source_y_fraction = (source_y_subpixel & kFractionMask) >> 8; + int source_uv_fraction = + ((source_y_subpixel >> y_shift) & kFractionMask) >> 8; + + const uint8* y_ptr = y0_ptr; + const uint8* u_ptr = u0_ptr; + const uint8* v_ptr = v0_ptr; + // Apply vertical filtering if necessary. + // TODO(fbarchard): Remove memcpy when not necessary. + if (filter & mozilla::gfx::FILTER_BILINEAR_V) { + if (yscale_fixed != kFractionMax && + source_y_fraction && ((source_y + 1) < source_height)) { + FilterRows(ybuf, y0_ptr, y1_ptr, source_width, source_y_fraction); } else { - RotateConvertYUVToRGB32Row(y_ptr, u_ptr, v_ptr, - dest_pixel, scaled_width, - scaled_dx >> 4, scaled_dx_uv >> 4); + memcpy(ybuf, y0_ptr, source_width); + } + y_ptr = ybuf; + ybuf[source_width] = ybuf[source_width-1]; + int uv_source_width = (source_width + 1) / 2; + if (yscale_fixed != kFractionMax && + source_uv_fraction && + (((source_y >> y_shift) + 1) < (source_height >> y_shift))) { + FilterRows(ubuf, u0_ptr, u1_ptr, uv_source_width, source_uv_fraction); + FilterRows(vbuf, v0_ptr, v1_ptr, uv_source_width, source_uv_fraction); + } else { + memcpy(ubuf, u0_ptr, uv_source_width); + memcpy(vbuf, v0_ptr, uv_source_width); } + u_ptr = ubuf; + v_ptr = vbuf; + ubuf[uv_source_width] = ubuf[uv_source_width - 1]; + vbuf[uv_source_width] = vbuf[uv_source_width - 1]; + } + if (source_dx == kFractionMax) { // Not scaled + FastConvertYUVToRGB32Row(y_ptr, u_ptr, v_ptr, + dest_pixel, width); + } else { + if (filter & FILTER_BILINEAR_H) { + LinearScaleYUVToRGB32Row(y_ptr, u_ptr, v_ptr, + dest_pixel, width, source_dx); + } else { +// Specialized scalers and rotation. +#if defined(_MSC_VER) && defined(_M_IX86) + if (width == (source_width * 2)) { + DoubleYUVToRGB32Row(y_ptr, u_ptr, v_ptr, + dest_pixel, width); + } else if ((source_dx & kFractionMask) == 0) { + // Scaling by integer scale factor. ie half. + ConvertYUVToRGB32Row(y_ptr, u_ptr, v_ptr, + dest_pixel, width, + source_dx >> kFractionBits); + } else if (source_dx_uv == source_dx) { // Not rotated. + ScaleYUVToRGB32Row(y_ptr, u_ptr, v_ptr, + dest_pixel, width, source_dx); + } else { + RotateConvertYUVToRGB32Row(y_ptr, u_ptr, v_ptr, + dest_pixel, width, + source_dx >> kFractionBits, + source_dx_uv >> kFractionBits); + } #else - if (scaled_dx == 16) { // Not scaled - if (has_mmx) - FastConvertYUVToRGB32Row(y_ptr, u_ptr, v_ptr, - dest_pixel, scaled_width); - else - FastConvertYUVToRGB32Row_C(y_ptr, u_ptr, v_ptr, - dest_pixel, scaled_width, x_shift); + ScaleYUVToRGB32Row(y_ptr, u_ptr, v_ptr, + dest_pixel, width, source_dx); #endif - } else { - if (has_mmx) - ScaleYUVToRGB32Row(y_ptr, u_ptr, v_ptr, - dest_pixel, scaled_width, scaled_dx); - else - ScaleYUVToRGB32Row_C(y_ptr, u_ptr, v_ptr, - dest_pixel, scaled_width, scaled_dx, x_shift); - - } + } + } } - - // MMX used for FastConvertYUVToRGB32Row requires emms instruction. + // MMX used for FastConvertYUVToRGB32Row and FilterRows requires emms. if (has_mmx) EMMS(); } } // namespace gfx } // namespace mozilla
--- a/gfx/ycbcr/yuv_convert.h +++ b/gfx/ycbcr/yuv_convert.h @@ -1,22 +1,22 @@ -// Copyright (c) 2009 The Chromium Authors. All rights reserved. +// 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. #ifndef MEDIA_BASE_YUV_CONVERT_H_ #define MEDIA_BASE_YUV_CONVERT_H_ #include "chromium_types.h" #include "gfxCore.h" - + namespace mozilla { namespace gfx { - + // Type of YUV surface. // The value of these enums matter as they are used to shift vertical indices. enum YUVType { YV12 = 0, // YV12 is half width and half height chroma channels. YV16 = 1, // YV16 is half width and full height chroma channels. YV24 = 2 // YV24 is full width and full height chroma channels. }; @@ -28,16 +28,24 @@ enum Rotate { ROTATE_180, // Rotate upside down. ROTATE_270, // Rotate counter clockwise. MIRROR_ROTATE_0, // Mirror horizontally. MIRROR_ROTATE_90, // Mirror then Rotate clockwise. MIRROR_ROTATE_180, // Mirror vertically. MIRROR_ROTATE_270 // Transpose. }; +// Filter affects how scaling looks. +enum ScaleFilter { + FILTER_NONE = 0, // No filter (point sampled). + FILTER_BILINEAR_H = 1, // Bilinear horizontal filter. + FILTER_BILINEAR_V = 2, // Bilinear vertical filter. + FILTER_BILINEAR = 3 // Bilinear filter. +}; + // Convert a frame of YUV to 32 bit ARGB. // Pass in YV16/YV12 depending on source format NS_GFX_(void) ConvertYCbCrToRGB32(const uint8* yplane, const uint8* uplane, const uint8* vplane, uint8* rgbframe, int pic_x, int pic_y, @@ -49,22 +57,23 @@ NS_GFX_(void) ConvertYCbCrToRGB32(const YUVType yuv_type); // Scale a frame of YUV to 32 bit ARGB. // Supports rotation and mirroring. NS_GFX_(void) ScaleYCbCrToRGB32(const uint8* yplane, const uint8* uplane, const uint8* vplane, uint8* rgbframe, - int frame_width, - int frame_height, - int scaled_width, - int scaled_height, + int source_width, + int source_height, + int width, + int height, int ystride, int uvstride, int rgbstride, YUVType yuv_type, - Rotate view_rotate); + Rotate view_rotate, + ScaleFilter filter); } // namespace gfx } // namespace mozilla - + #endif // MEDIA_BASE_YUV_CONVERT_H_
--- a/gfx/ycbcr/yuv_row.h +++ b/gfx/ycbcr/yuv_row.h @@ -1,9 +1,9 @@ -// Copyright (c) 2009 The Chromium Authors. All rights reserved. +// 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. // yuv_row internal functions to handle YUV conversion and scaling to RGB. // These functions are used from both yuv_convert.cc and yuv_scale.cc. // TODO(fbarchard): Write function that can handle rotation and scaling. @@ -23,16 +23,21 @@ void FastConvertYUVToRGB32Row(const uint void FastConvertYUVToRGB32Row_C(const uint8* y_buf, const uint8* u_buf, const uint8* v_buf, uint8* rgb_buf, int width, unsigned int x_shift); +void FastConvertYUVToRGB32Row(const uint8* y_buf, + const uint8* u_buf, + const uint8* v_buf, + uint8* rgb_buf, + int width); // Can do 1x, half size or any scale down by an integer amount. // Step can be negative (mirroring, rotate 180). // This is the third fastest of the scalers. void ConvertYUVToRGB32Row(const uint8* y_buf, const uint8* u_buf, const uint8* v_buf, uint8* rgb_buf, @@ -56,38 +61,79 @@ void DoubleYUVToRGB32Row(const uint8* y_ const uint8* u_buf, const uint8* v_buf, uint8* rgb_buf, int width); // Handles arbitrary scaling up or down. // Mirroring is supported, but not 90 or 270 degree rotation. // Chroma is under sampled every 2 pixels for performance. -// This is the slowest of the scalers. void ScaleYUVToRGB32Row(const uint8* y_buf, const uint8* u_buf, const uint8* v_buf, uint8* rgb_buf, int width, - int scaled_dx); + int source_dx); + +void ScaleYUVToRGB32Row(const uint8* y_buf, + const uint8* u_buf, + const uint8* v_buf, + uint8* rgb_buf, + int width, + int source_dx); void ScaleYUVToRGB32Row_C(const uint8* y_buf, const uint8* u_buf, const uint8* v_buf, uint8* rgb_buf, int width, - int scaled_dx, - unsigned int x_shift); + int source_dx); + +// Handles arbitrary scaling up or down with bilinear filtering. +// Mirroring is supported, but not 90 or 270 degree rotation. +// Chroma is under sampled every 2 pixels for performance. +// This is the slowest of the scalers. +void LinearScaleYUVToRGB32Row(const uint8* y_buf, + const uint8* u_buf, + const uint8* v_buf, + uint8* rgb_buf, + int width, + int source_dx); -} // extern "C" +void LinearScaleYUVToRGB32Row(const uint8* y_buf, + const uint8* u_buf, + const uint8* v_buf, + uint8* rgb_buf, + int width, + int source_dx); + +void LinearScaleYUVToRGB32Row_C(const uint8* y_buf, + const uint8* u_buf, + const uint8* v_buf, + uint8* rgb_buf, + int width, + int source_dx); + + +#if defined(_MSC_VER) +#define SIMD_ALIGNED(var) __declspec(align(16)) var +#else +#define SIMD_ALIGNED(var) var __attribute__((aligned(16))) +#endif +extern SIMD_ALIGNED(int16 kCoefficientsRgbY[768][4]); // x64 uses MMX2 (SSE) so emms is not required. -#if defined(ARCH_CPU_X86) +// Warning C4799: function has no EMMS instruction. +// EMMS() is slow and should be called by the calling function once per image. +#if !defined(ARCH_CPU_X86_64) #if defined(_MSC_VER) #define EMMS() __asm emms +#pragma warning(disable: 4799) #else #define EMMS() asm("emms") #endif #else #define EMMS() #endif +} // extern "C" + #endif // MEDIA_BASE_YUV_ROW_H_
--- a/gfx/ycbcr/yuv_row_c.cpp +++ b/gfx/ycbcr/yuv_row_c.cpp @@ -1,170 +1,60 @@ -// Copyright (c) 2009 The Chromium Authors. All rights reserved. +// 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 "yuv_row.h" #define DCHECK(a) -// TODO(fbarchard): Move this to yuv_row_posix.cc to share with Mac. -// TODO(fbarchard): Do 64 bit version. - extern "C" { -// Reference version of YUV converter. -static const int kClipTableSize = 256; -static const int kClipOverflow = 288; // Cb max is 535. - -static uint8 kRgbClipTable[kClipOverflow + - kClipTableSize + - kClipOverflow] = { - 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, // 288 underflow values - 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, // clipped to 0. - 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, - 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, - 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, - 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, - 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, - 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, - 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, - 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, - 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, - 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, - 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, - 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, - 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, - 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, - 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, - 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, - 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, - 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, - 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, - 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, - 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, - 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, - 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, - 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, - 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, - 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, - 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, - 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, - 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, - 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, - 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, - 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, - 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, - 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, - 0x00, 0x01, 0x02, 0x03, 0x04, 0x05, 0x06, 0x07, // Unclipped values. - 0x08, 0x09, 0x0A, 0x0B, 0x0C, 0x0D, 0x0E, 0x0F, - 0x10, 0x11, 0x12, 0x13, 0x14, 0x15, 0x16, 0x17, - 0x18, 0x19, 0x1A, 0x1B, 0x1C, 0x1D, 0x1E, 0x1F, - 0x20, 0x21, 0x22, 0x23, 0x24, 0x25, 0x26, 0x27, - 0x28, 0x29, 0x2A, 0x2B, 0x2C, 0x2D, 0x2E, 0x2F, - 0x30, 0x31, 0x32, 0x33, 0x34, 0x35, 0x36, 0x37, - 0x38, 0x39, 0x3A, 0x3B, 0x3C, 0x3D, 0x3E, 0x3F, - 0x40, 0x41, 0x42, 0x43, 0x44, 0x45, 0x46, 0x47, - 0x48, 0x49, 0x4A, 0x4B, 0x4C, 0x4D, 0x4E, 0x4F, - 0x50, 0x51, 0x52, 0x53, 0x54, 0x55, 0x56, 0x57, - 0x58, 0x59, 0x5A, 0x5B, 0x5C, 0x5D, 0x5E, 0x5F, - 0x60, 0x61, 0x62, 0x63, 0x64, 0x65, 0x66, 0x67, - 0x68, 0x69, 0x6A, 0x6B, 0x6C, 0x6D, 0x6E, 0x6F, - 0x70, 0x71, 0x72, 0x73, 0x74, 0x75, 0x76, 0x77, - 0x78, 0x79, 0x7A, 0x7B, 0x7C, 0x7D, 0x7E, 0x7F, - 0x80, 0x81, 0x82, 0x83, 0x84, 0x85, 0x86, 0x87, - 0x88, 0x89, 0x8A, 0x8B, 0x8C, 0x8D, 0x8E, 0x8F, - 0x90, 0x91, 0x92, 0x93, 0x94, 0x95, 0x96, 0x97, - 0x98, 0x99, 0x9A, 0x9B, 0x9C, 0x9D, 0x9E, 0x9F, - 0xA0, 0xA1, 0xA2, 0xA3, 0xA4, 0xA5, 0xA6, 0xA7, - 0xA8, 0xA9, 0xAA, 0xAB, 0xAC, 0xAD, 0xAE, 0xAF, - 0xB0, 0xB1, 0xB2, 0xB3, 0xB4, 0xB5, 0xB6, 0xB7, - 0xB8, 0xB9, 0xBA, 0xBB, 0xBC, 0xBD, 0xBE, 0xBF, - 0xC0, 0xC1, 0xC2, 0xC3, 0xC4, 0xC5, 0xC6, 0xC7, - 0xC8, 0xC9, 0xCA, 0xCB, 0xCC, 0xCD, 0xCE, 0xCF, - 0xD0, 0xD1, 0xD2, 0xD3, 0xD4, 0xD5, 0xD6, 0xD7, - 0xD8, 0xD9, 0xDA, 0xDB, 0xDC, 0xDD, 0xDE, 0xDF, - 0xE0, 0xE1, 0xE2, 0xE3, 0xE4, 0xE5, 0xE6, 0xE7, - 0xE8, 0xE9, 0xEA, 0xEB, 0xEC, 0xED, 0xEE, 0xEF, - 0xF0, 0xF1, 0xF2, 0xF3, 0xF4, 0xF5, 0xF6, 0xF7, - 0xF8, 0xF9, 0xFA, 0xFB, 0xFC, 0xFD, 0xFE, 0xFF, - 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, // 288 overflow values - 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, // clipped to 255. - 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, - 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, - 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, - 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, - 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, - 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, - 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, - 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, - 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, - 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, - 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, - 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, - 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, - 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, - 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, - 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, - 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, - 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, - 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, - 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, - 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, - 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, - 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, - 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, - 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, - 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, - 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, - 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, - 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, - 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, - 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, - 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, - 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, - 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, -}; - -// Clip an rgb channel value to 0..255 range. -// Source is signed fixed point 8.8. -// Table allows for values to underflow or overflow by 128. -// Therefore source range is -128 to 384. -// Output clips to unsigned 0 to 255. -static inline uint32 clip(int32 value) { - DCHECK(((value >> 8) + kClipOverflow) >= 0); - DCHECK(((value >> 8) + kClipOverflow) < - (kClipOverflow + kClipTableSize + kClipOverflow)); - return static_cast<uint32>(kRgbClipTable[((value) >> 8) + kClipOverflow]); -} +// C reference code that mimic the YUV assembly. +#define packuswb(x) ((x) < 0 ? 0 : ((x) > 255 ? 255 : (x))) +#define paddsw(x, y) (((x) + (y)) < -32768 ? -32768 : \ + (((x) + (y)) > 32767 ? 32767 : ((x) + (y)))) static inline void YuvPixel(uint8 y, uint8 u, uint8 v, uint8* rgb_buf) { - int32 d = static_cast<int32>(u) - 128; - int32 e = static_cast<int32>(v) - 128; + + int b = kCoefficientsRgbY[256+u][0]; + int g = kCoefficientsRgbY[256+u][1]; + int r = kCoefficientsRgbY[256+u][2]; + int a = kCoefficientsRgbY[256+u][3]; + + b = paddsw(b, kCoefficientsRgbY[512+v][0]); + g = paddsw(g, kCoefficientsRgbY[512+v][1]); + r = paddsw(r, kCoefficientsRgbY[512+v][2]); + a = paddsw(a, kCoefficientsRgbY[512+v][3]); - int32 cb = (516 * d + 128); - int32 cg = (- 100 * d - 208 * e + 128); - int32 cr = (409 * e + 128); + b = paddsw(b, kCoefficientsRgbY[y][0]); + g = paddsw(g, kCoefficientsRgbY[y][1]); + r = paddsw(r, kCoefficientsRgbY[y][2]); + a = paddsw(a, kCoefficientsRgbY[y][3]); - int32 C298a = ((static_cast<int32>(y) - 16) * 298); - *reinterpret_cast<uint32*>(rgb_buf) = (clip(C298a + cb)) | - (clip(C298a + cg) << 8) | - (clip(C298a + cr) << 16) | - (0xff000000); + b >>= 6; + g >>= 6; + r >>= 6; + a >>= 6; + + *reinterpret_cast<uint32*>(rgb_buf) = (packuswb(b)) | + (packuswb(g) << 8) | + (packuswb(r) << 16) | + (packuswb(a) << 24); } void FastConvertYUVToRGB32Row_C(const uint8* y_buf, - const uint8* u_buf, - const uint8* v_buf, - uint8* rgb_buf, - int width, - unsigned int x_shift) { + const uint8* u_buf, + const uint8* v_buf, + uint8* rgb_buf, + int width, + unsigned int x_shift) { for (int x = 0; x < width; x += 2) { uint8 u = u_buf[x >> x_shift]; uint8 v = v_buf[x >> x_shift]; uint8 y0 = y_buf[x]; YuvPixel(y0, u, v, rgb_buf); if ((x + 1) < width) { uint8 y1 = y_buf[x + 1]; if (x_shift == 0) { @@ -172,31 +62,72 @@ void FastConvertYUVToRGB32Row_C(const ui v = v_buf[x + 1]; } YuvPixel(y1, u, v, rgb_buf + 4); } rgb_buf += 8; // Advance 2 pixels. } } -// 28.4 fixed point is used. A shift by 4 isolates the integer. -// A shift by 5 is used to further subsample the chrominence channels. -// & 15 isolates the fixed point fraction. >> 2 to get the upper 2 bits, -// for 1/4 pixel accurate interpolation. +// 16.16 fixed point is used. A shift by 16 isolates the integer. +// A shift by 17 is used to further subsample the chrominence channels. +// & 0xffff isolates the fixed point fraction. >> 2 to get the upper 2 bits, +// for 1/65536 pixel accurate interpolation. void ScaleYUVToRGB32Row_C(const uint8* y_buf, - const uint8* u_buf, - const uint8* v_buf, - uint8* rgb_buf, - int width, - int scaled_dx, - unsigned int x_shift) { - int scaled_x = 0; - for (int x = 0; x < width; ++x) { - uint8 u = u_buf[scaled_x >> (4 + x_shift)]; - uint8 v = v_buf[scaled_x >> (4 + x_shift)]; - uint8 y0 = y_buf[scaled_x >> 4]; - YuvPixel(y0, u, v, rgb_buf); - rgb_buf += 4; - scaled_x += scaled_dx; + const uint8* u_buf, + const uint8* v_buf, + uint8* rgb_buf, + int width, + int source_dx) { + int x = 0; + for (int i = 0; i < width; i += 2) { + int y = y_buf[x >> 16]; + int u = u_buf[(x >> 17)]; + int v = v_buf[(x >> 17)]; + YuvPixel(y, u, v, rgb_buf); + x += source_dx; + if ((i + 1) < width) { + y = y_buf[x >> 16]; + YuvPixel(y, u, v, rgb_buf+4); + x += source_dx; + } + rgb_buf += 8; } } + +void LinearScaleYUVToRGB32Row_C(const uint8* y_buf, + const uint8* u_buf, + const uint8* v_buf, + uint8* rgb_buf, + int width, + int source_dx) { + int x = 0; + if (source_dx >= 0x20000) { + x = 32768; + } + for (int i = 0; i < width; i += 2) { + int y0 = y_buf[x >> 16]; + int y1 = y_buf[(x >> 16) + 1]; + int u0 = u_buf[(x >> 17)]; + int u1 = u_buf[(x >> 17) + 1]; + int v0 = v_buf[(x >> 17)]; + int v1 = v_buf[(x >> 17) + 1]; + int y_frac = (x & 65535); + int uv_frac = ((x >> 1) & 65535); + int y = (y_frac * y1 + (y_frac ^ 65535) * y0) >> 16; + int u = (uv_frac * u1 + (uv_frac ^ 65535) * u0) >> 16; + int v = (uv_frac * v1 + (uv_frac ^ 65535) * v0) >> 16; + YuvPixel(y, u, v, rgb_buf); + x += source_dx; + if ((i + 1) < width) { + y0 = y_buf[x >> 16]; + y1 = y_buf[(x >> 16) + 1]; + y_frac = (x & 65535); + y = (y_frac * y1 + (y_frac ^ 65535) * y0) >> 16; + YuvPixel(y, u, v, rgb_buf+4); + x += source_dx; + } + rgb_buf += 8; + } +} + } // extern "C"
deleted file mode 100644 --- a/gfx/ycbcr/yuv_row_linux.cpp +++ /dev/null @@ -1,657 +0,0 @@ -// Copyright (c) 2009 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 "yuv_row.h" - -#define DCHECK(a) - -// TODO(fbarchard): Move this to yuv_row_posix.cc to share with Mac. -// TODO(fbarchard): Do 64 bit version. - -extern "C" { - -#ifndef ARCH_CPU_X86_FAMILY -// non-x86 implementation uses C fallback -void FastConvertYUVToRGB32Row(const uint8* y_buf, - const uint8* u_buf, - const uint8* v_buf, - uint8* rgb_buf, - int width) { - FastConvertYUVToRGB32Row_C(y_buf, u_buf, v_buf, rgb_buf, width, 1); -} - -void ScaleYUVToRGB32Row(const uint8* y_buf, - const uint8* u_buf, - const uint8* v_buf, - uint8* rgb_buf, - int width, - int scaled_dx) { - ScaleYUVToRGB32Row_C(y_buf, u_buf, v_buf, rgb_buf, width, scaled_dx, 1); -} -#else - -#define RGBY(i) { \ - static_cast<int16>(1.164 * 64 * (i - 16) + 0.5), \ - static_cast<int16>(1.164 * 64 * (i - 16) + 0.5), \ - static_cast<int16>(1.164 * 64 * (i - 16) + 0.5), \ - 0 \ -} - -#define RGBU(i) { \ - static_cast<int16>(2.018 * 64 * (i - 128) + 0.5), \ - static_cast<int16>(-0.391 * 64 * (i - 128) + 0.5), \ - 0, \ - static_cast<int16>(256 * 64 - 1) \ -} - -#define RGBV(i) { \ - 0, \ - static_cast<int16>(-0.813 * 64 * (i - 128) + 0.5), \ - static_cast<int16>(1.596 * 64 * (i - 128) + 0.5), \ - 0 \ -} - -#define MMX_ALIGNED(var) var __attribute__((aligned(16))) - - -MMX_ALIGNED(int16 kCoefficientsRgbY[768][4]) = { - RGBY(0x00), RGBY(0x01), RGBY(0x02), RGBY(0x03), - RGBY(0x04), RGBY(0x05), RGBY(0x06), RGBY(0x07), - RGBY(0x08), RGBY(0x09), RGBY(0x0A), RGBY(0x0B), - RGBY(0x0C), RGBY(0x0D), RGBY(0x0E), RGBY(0x0F), - RGBY(0x10), RGBY(0x11), RGBY(0x12), RGBY(0x13), - RGBY(0x14), RGBY(0x15), RGBY(0x16), RGBY(0x17), - RGBY(0x18), RGBY(0x19), RGBY(0x1A), RGBY(0x1B), - RGBY(0x1C), RGBY(0x1D), RGBY(0x1E), RGBY(0x1F), - RGBY(0x20), RGBY(0x21), RGBY(0x22), RGBY(0x23), - RGBY(0x24), RGBY(0x25), RGBY(0x26), RGBY(0x27), - RGBY(0x28), RGBY(0x29), RGBY(0x2A), RGBY(0x2B), - RGBY(0x2C), RGBY(0x2D), RGBY(0x2E), RGBY(0x2F), - RGBY(0x30), RGBY(0x31), RGBY(0x32), RGBY(0x33), - RGBY(0x34), RGBY(0x35), RGBY(0x36), RGBY(0x37), - RGBY(0x38), RGBY(0x39), RGBY(0x3A), RGBY(0x3B), - RGBY(0x3C), RGBY(0x3D), RGBY(0x3E), RGBY(0x3F), - RGBY(0x40), RGBY(0x41), RGBY(0x42), RGBY(0x43), - RGBY(0x44), RGBY(0x45), RGBY(0x46), RGBY(0x47), - RGBY(0x48), RGBY(0x49), RGBY(0x4A), RGBY(0x4B), - RGBY(0x4C), RGBY(0x4D), RGBY(0x4E), RGBY(0x4F), - RGBY(0x50), RGBY(0x51), RGBY(0x52), RGBY(0x53), - RGBY(0x54), RGBY(0x55), RGBY(0x56), RGBY(0x57), - RGBY(0x58), RGBY(0x59), RGBY(0x5A), RGBY(0x5B), - RGBY(0x5C), RGBY(0x5D), RGBY(0x5E), RGBY(0x5F), - RGBY(0x60), RGBY(0x61), RGBY(0x62), RGBY(0x63), - RGBY(0x64), RGBY(0x65), RGBY(0x66), RGBY(0x67), - RGBY(0x68), RGBY(0x69), RGBY(0x6A), RGBY(0x6B), - RGBY(0x6C), RGBY(0x6D), RGBY(0x6E), RGBY(0x6F), - RGBY(0x70), RGBY(0x71), RGBY(0x72), RGBY(0x73), - RGBY(0x74), RGBY(0x75), RGBY(0x76), RGBY(0x77), - RGBY(0x78), RGBY(0x79), RGBY(0x7A), RGBY(0x7B), - RGBY(0x7C), RGBY(0x7D), RGBY(0x7E), RGBY(0x7F), - RGBY(0x80), RGBY(0x81), RGBY(0x82), RGBY(0x83), - RGBY(0x84), RGBY(0x85), RGBY(0x86), RGBY(0x87), - RGBY(0x88), RGBY(0x89), RGBY(0x8A), RGBY(0x8B), - RGBY(0x8C), RGBY(0x8D), RGBY(0x8E), RGBY(0x8F), - RGBY(0x90), RGBY(0x91), RGBY(0x92), RGBY(0x93), - RGBY(0x94), RGBY(0x95), RGBY(0x96), RGBY(0x97), - RGBY(0x98), RGBY(0x99), RGBY(0x9A), RGBY(0x9B), - RGBY(0x9C), RGBY(0x9D), RGBY(0x9E), RGBY(0x9F), - RGBY(0xA0), RGBY(0xA1), RGBY(0xA2), RGBY(0xA3), - RGBY(0xA4), RGBY(0xA5), RGBY(0xA6), RGBY(0xA7), - RGBY(0xA8), RGBY(0xA9), RGBY(0xAA), RGBY(0xAB), - RGBY(0xAC), RGBY(0xAD), RGBY(0xAE), RGBY(0xAF), - RGBY(0xB0), RGBY(0xB1), RGBY(0xB2), RGBY(0xB3), - RGBY(0xB4), RGBY(0xB5), RGBY(0xB6), RGBY(0xB7), - RGBY(0xB8), RGBY(0xB9), RGBY(0xBA), RGBY(0xBB), - RGBY(0xBC), RGBY(0xBD), RGBY(0xBE), RGBY(0xBF), - RGBY(0xC0), RGBY(0xC1), RGBY(0xC2), RGBY(0xC3), - RGBY(0xC4), RGBY(0xC5), RGBY(0xC6), RGBY(0xC7), - RGBY(0xC8), RGBY(0xC9), RGBY(0xCA), RGBY(0xCB), - RGBY(0xCC), RGBY(0xCD), RGBY(0xCE), RGBY(0xCF), - RGBY(0xD0), RGBY(0xD1), RGBY(0xD2), RGBY(0xD3), - RGBY(0xD4), RGBY(0xD5), RGBY(0xD6), RGBY(0xD7), - RGBY(0xD8), RGBY(0xD9), RGBY(0xDA), RGBY(0xDB), - RGBY(0xDC), RGBY(0xDD), RGBY(0xDE), RGBY(0xDF), - RGBY(0xE0), RGBY(0xE1), RGBY(0xE2), RGBY(0xE3), - RGBY(0xE4), RGBY(0xE5), RGBY(0xE6), RGBY(0xE7), - RGBY(0xE8), RGBY(0xE9), RGBY(0xEA), RGBY(0xEB), - RGBY(0xEC), RGBY(0xED), RGBY(0xEE), RGBY(0xEF), - RGBY(0xF0), RGBY(0xF1), RGBY(0xF2), RGBY(0xF3), - RGBY(0xF4), RGBY(0xF5), RGBY(0xF6), RGBY(0xF7), - RGBY(0xF8), RGBY(0xF9), RGBY(0xFA), RGBY(0xFB), - RGBY(0xFC), RGBY(0xFD), RGBY(0xFE), RGBY(0xFF), - - // Chroma U table. - RGBU(0x00), RGBU(0x01), RGBU(0x02), RGBU(0x03), - RGBU(0x04), RGBU(0x05), RGBU(0x06), RGBU(0x07), - RGBU(0x08), RGBU(0x09), RGBU(0x0A), RGBU(0x0B), - RGBU(0x0C), RGBU(0x0D), RGBU(0x0E), RGBU(0x0F), - RGBU(0x10), RGBU(0x11), RGBU(0x12), RGBU(0x13), - RGBU(0x14), RGBU(0x15), RGBU(0x16), RGBU(0x17), - RGBU(0x18), RGBU(0x19), RGBU(0x1A), RGBU(0x1B), - RGBU(0x1C), RGBU(0x1D), RGBU(0x1E), RGBU(0x1F), - RGBU(0x20), RGBU(0x21), RGBU(0x22), RGBU(0x23), - RGBU(0x24), RGBU(0x25), RGBU(0x26), RGBU(0x27), - RGBU(0x28), RGBU(0x29), RGBU(0x2A), RGBU(0x2B), - RGBU(0x2C), RGBU(0x2D), RGBU(0x2E), RGBU(0x2F), - RGBU(0x30), RGBU(0x31), RGBU(0x32), RGBU(0x33), - RGBU(0x34), RGBU(0x35), RGBU(0x36), RGBU(0x37), - RGBU(0x38), RGBU(0x39), RGBU(0x3A), RGBU(0x3B), - RGBU(0x3C), RGBU(0x3D), RGBU(0x3E), RGBU(0x3F), - RGBU(0x40), RGBU(0x41), RGBU(0x42), RGBU(0x43), - RGBU(0x44), RGBU(0x45), RGBU(0x46), RGBU(0x47), - RGBU(0x48), RGBU(0x49), RGBU(0x4A), RGBU(0x4B), - RGBU(0x4C), RGBU(0x4D), RGBU(0x4E), RGBU(0x4F), - RGBU(0x50), RGBU(0x51), RGBU(0x52), RGBU(0x53), - RGBU(0x54), RGBU(0x55), RGBU(0x56), RGBU(0x57), - RGBU(0x58), RGBU(0x59), RGBU(0x5A), RGBU(0x5B), - RGBU(0x5C), RGBU(0x5D), RGBU(0x5E), RGBU(0x5F), - RGBU(0x60), RGBU(0x61), RGBU(0x62), RGBU(0x63), - RGBU(0x64), RGBU(0x65), RGBU(0x66), RGBU(0x67), - RGBU(0x68), RGBU(0x69), RGBU(0x6A), RGBU(0x6B), - RGBU(0x6C), RGBU(0x6D), RGBU(0x6E), RGBU(0x6F), - RGBU(0x70), RGBU(0x71), RGBU(0x72), RGBU(0x73), - RGBU(0x74), RGBU(0x75), RGBU(0x76), RGBU(0x77), - RGBU(0x78), RGBU(0x79), RGBU(0x7A), RGBU(0x7B), - RGBU(0x7C), RGBU(0x7D), RGBU(0x7E), RGBU(0x7F), - RGBU(0x80), RGBU(0x81), RGBU(0x82), RGBU(0x83), - RGBU(0x84), RGBU(0x85), RGBU(0x86), RGBU(0x87), - RGBU(0x88), RGBU(0x89), RGBU(0x8A), RGBU(0x8B), - RGBU(0x8C), RGBU(0x8D), RGBU(0x8E), RGBU(0x8F), - RGBU(0x90), RGBU(0x91), RGBU(0x92), RGBU(0x93), - RGBU(0x94), RGBU(0x95), RGBU(0x96), RGBU(0x97), - RGBU(0x98), RGBU(0x99), RGBU(0x9A), RGBU(0x9B), - RGBU(0x9C), RGBU(0x9D), RGBU(0x9E), RGBU(0x9F), - RGBU(0xA0), RGBU(0xA1), RGBU(0xA2), RGBU(0xA3), - RGBU(0xA4), RGBU(0xA5), RGBU(0xA6), RGBU(0xA7), - RGBU(0xA8), RGBU(0xA9), RGBU(0xAA), RGBU(0xAB), - RGBU(0xAC), RGBU(0xAD), RGBU(0xAE), RGBU(0xAF), - RGBU(0xB0), RGBU(0xB1), RGBU(0xB2), RGBU(0xB3), - RGBU(0xB4), RGBU(0xB5), RGBU(0xB6), RGBU(0xB7), - RGBU(0xB8), RGBU(0xB9), RGBU(0xBA), RGBU(0xBB), - RGBU(0xBC), RGBU(0xBD), RGBU(0xBE), RGBU(0xBF), - RGBU(0xC0), RGBU(0xC1), RGBU(0xC2), RGBU(0xC3), - RGBU(0xC4), RGBU(0xC5), RGBU(0xC6), RGBU(0xC7), - RGBU(0xC8), RGBU(0xC9), RGBU(0xCA), RGBU(0xCB), - RGBU(0xCC), RGBU(0xCD), RGBU(0xCE), RGBU(0xCF), - RGBU(0xD0), RGBU(0xD1), RGBU(0xD2), RGBU(0xD3), - RGBU(0xD4), RGBU(0xD5), RGBU(0xD6), RGBU(0xD7), - RGBU(0xD8), RGBU(0xD9), RGBU(0xDA), RGBU(0xDB), - RGBU(0xDC), RGBU(0xDD), RGBU(0xDE), RGBU(0xDF), - RGBU(0xE0), RGBU(0xE1), RGBU(0xE2), RGBU(0xE3), - RGBU(0xE4), RGBU(0xE5), RGBU(0xE6), RGBU(0xE7), - RGBU(0xE8), RGBU(0xE9), RGBU(0xEA), RGBU(0xEB), - RGBU(0xEC), RGBU(0xED), RGBU(0xEE), RGBU(0xEF), - RGBU(0xF0), RGBU(0xF1), RGBU(0xF2), RGBU(0xF3), - RGBU(0xF4), RGBU(0xF5), RGBU(0xF6), RGBU(0xF7), - RGBU(0xF8), RGBU(0xF9), RGBU(0xFA), RGBU(0xFB), - RGBU(0xFC), RGBU(0xFD), RGBU(0xFE), RGBU(0xFF), - - // Chroma V table. - RGBV(0x00), RGBV(0x01), RGBV(0x02), RGBV(0x03), - RGBV(0x04), RGBV(0x05), RGBV(0x06), RGBV(0x07), - RGBV(0x08), RGBV(0x09), RGBV(0x0A), RGBV(0x0B), - RGBV(0x0C), RGBV(0x0D), RGBV(0x0E), RGBV(0x0F), - RGBV(0x10), RGBV(0x11), RGBV(0x12), RGBV(0x13), - RGBV(0x14), RGBV(0x15), RGBV(0x16), RGBV(0x17), - RGBV(0x18), RGBV(0x19), RGBV(0x1A), RGBV(0x1B), - RGBV(0x1C), RGBV(0x1D), RGBV(0x1E), RGBV(0x1F), - RGBV(0x20), RGBV(0x21), RGBV(0x22), RGBV(0x23), - RGBV(0x24), RGBV(0x25), RGBV(0x26), RGBV(0x27), - RGBV(0x28), RGBV(0x29), RGBV(0x2A), RGBV(0x2B), - RGBV(0x2C), RGBV(0x2D), RGBV(0x2E), RGBV(0x2F), - RGBV(0x30), RGBV(0x31), RGBV(0x32), RGBV(0x33), - RGBV(0x34), RGBV(0x35), RGBV(0x36), RGBV(0x37), - RGBV(0x38), RGBV(0x39), RGBV(0x3A), RGBV(0x3B), - RGBV(0x3C), RGBV(0x3D), RGBV(0x3E), RGBV(0x3F), - RGBV(0x40), RGBV(0x41), RGBV(0x42), RGBV(0x43), - RGBV(0x44), RGBV(0x45), RGBV(0x46), RGBV(0x47), - RGBV(0x48), RGBV(0x49), RGBV(0x4A), RGBV(0x4B), - RGBV(0x4C), RGBV(0x4D), RGBV(0x4E), RGBV(0x4F), - RGBV(0x50), RGBV(0x51), RGBV(0x52), RGBV(0x53), - RGBV(0x54), RGBV(0x55), RGBV(0x56), RGBV(0x57), - RGBV(0x58), RGBV(0x59), RGBV(0x5A), RGBV(0x5B), - RGBV(0x5C), RGBV(0x5D), RGBV(0x5E), RGBV(0x5F), - RGBV(0x60), RGBV(0x61), RGBV(0x62), RGBV(0x63), - RGBV(0x64), RGBV(0x65), RGBV(0x66), RGBV(0x67), - RGBV(0x68), RGBV(0x69), RGBV(0x6A), RGBV(0x6B), - RGBV(0x6C), RGBV(0x6D), RGBV(0x6E), RGBV(0x6F), - RGBV(0x70), RGBV(0x71), RGBV(0x72), RGBV(0x73), - RGBV(0x74), RGBV(0x75), RGBV(0x76), RGBV(0x77), - RGBV(0x78), RGBV(0x79), RGBV(0x7A), RGBV(0x7B), - RGBV(0x7C), RGBV(0x7D), RGBV(0x7E), RGBV(0x7F), - RGBV(0x80), RGBV(0x81), RGBV(0x82), RGBV(0x83), - RGBV(0x84), RGBV(0x85), RGBV(0x86), RGBV(0x87), - RGBV(0x88), RGBV(0x89), RGBV(0x8A), RGBV(0x8B), - RGBV(0x8C), RGBV(0x8D), RGBV(0x8E), RGBV(0x8F), - RGBV(0x90), RGBV(0x91), RGBV(0x92), RGBV(0x93), - RGBV(0x94), RGBV(0x95), RGBV(0x96), RGBV(0x97), - RGBV(0x98), RGBV(0x99), RGBV(0x9A), RGBV(0x9B), - RGBV(0x9C), RGBV(0x9D), RGBV(0x9E), RGBV(0x9F), - RGBV(0xA0), RGBV(0xA1), RGBV(0xA2), RGBV(0xA3), - RGBV(0xA4), RGBV(0xA5), RGBV(0xA6), RGBV(0xA7), - RGBV(0xA8), RGBV(0xA9), RGBV(0xAA), RGBV(0xAB), - RGBV(0xAC), RGBV(0xAD), RGBV(0xAE), RGBV(0xAF), - RGBV(0xB0), RGBV(0xB1), RGBV(0xB2), RGBV(0xB3), - RGBV(0xB4), RGBV(0xB5), RGBV(0xB6), RGBV(0xB7), - RGBV(0xB8), RGBV(0xB9), RGBV(0xBA), RGBV(0xBB), - RGBV(0xBC), RGBV(0xBD), RGBV(0xBE), RGBV(0xBF), - RGBV(0xC0), RGBV(0xC1), RGBV(0xC2), RGBV(0xC3), - RGBV(0xC4), RGBV(0xC5), RGBV(0xC6), RGBV(0xC7), - RGBV(0xC8), RGBV(0xC9), RGBV(0xCA), RGBV(0xCB), - RGBV(0xCC), RGBV(0xCD), RGBV(0xCE), RGBV(0xCF), - RGBV(0xD0), RGBV(0xD1), RGBV(0xD2), RGBV(0xD3), - RGBV(0xD4), RGBV(0xD5), RGBV(0xD6), RGBV(0xD7), - RGBV(0xD8), RGBV(0xD9), RGBV(0xDA), RGBV(0xDB), - RGBV(0xDC), RGBV(0xDD), RGBV(0xDE), RGBV(0xDF), - RGBV(0xE0), RGBV(0xE1), RGBV(0xE2), RGBV(0xE3), - RGBV(0xE4), RGBV(0xE5), RGBV(0xE6), RGBV(0xE7), - RGBV(0xE8), RGBV(0xE9), RGBV(0xEA), RGBV(0xEB), - RGBV(0xEC), RGBV(0xED), RGBV(0xEE), RGBV(0xEF), - RGBV(0xF0), RGBV(0xF1), RGBV(0xF2), RGBV(0xF3), - RGBV(0xF4), RGBV(0xF5), RGBV(0xF6), RGBV(0xF7), - RGBV(0xF8), RGBV(0xF9), RGBV(0xFA), RGBV(0xFB), - RGBV(0xFC), RGBV(0xFD), RGBV(0xFE), RGBV(0xFF), -}; - -#ifdef __SUNPRO_CC -#pragma align 16 (kCoefficientsRgbY) -#endif - -#if defined(ARCH_CPU_X86_64) - -#ifdef __SUNPRO_CC -// AMD64 ABI uses register parameters. -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:" - "movzbq (%1),%%r10\n" - "add $0x1,%1\n" - "movzbq (%2),%%r11\n" - "add $0x1,%2\n" - "movq 2048(%5,%%r10,8),%%xmm0\n" - "movzbq (%0),%%r10\n" - "movq 4096(%5,%%r11,8),%%xmm1\n" - "movzbq 0x1(%0),%%r11\n" - "paddsw %%xmm1,%%xmm0\n" - "movq (%5,%%r10,8),%%xmm2\n" - "add $0x2,%0\n" - "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:" - "sub $0x2,%4\n" - "jns convertloop\n" - -"convertnext:" - "add $0x1,%4\n" - "js convertdone\n" - - "movzbq (%1),%%r10\n" - "movq 2048(%5,%%r10,8),%%xmm0\n" - "movzbq (%2),%%r10\n" - "movq 4096(%5,%%r10,8),%%xmm1\n" - "paddsw %%xmm1,%%xmm0\n" - "movzbq (%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:" - : - : "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" -); -} -#else // __SUNPRO_CC -// 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 1f\n" -"0:" - "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" - "paddsw %%xmm1,%%xmm0\n" - "movq (%5,%%r10,8),%%xmm2\n" - "add $0x2,%0\n" - "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" -"1:" - "sub $0x2,%4\n" - "jns 0b\n" - -"2:" - "add $0x1,%4\n" - "js 3f\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" -"3:" - : - : "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" -); -} - -void ScaleYUVToRGB32Row(const uint8* y_buf, // rdi - const uint8* u_buf, // rsi - const uint8* v_buf, // rdx - uint8* rgb_buf, // rcx - int width, // r8 - int scaled_dx) { // r9 - asm( - "xor %%r11,%%r11\n" - "sub $0x2,%4\n" - "js scalenext\n" - -"scaleloop:" - "mov %%r11,%%r10\n" - "sar $0x5,%%r10\n" - "movzb (%1,%%r10,1),%%rax\n" - "movq 2048(%5,%%rax,8),%%xmm0\n" - "movzb (%2,%%r10,1),%%rax\n" - "movq 4096(%5,%%rax,8),%%xmm1\n" - "lea (%%r11,%6),%%r10\n" - "sar $0x4,%%r11\n" - "movzb (%0,%%r11,1),%%rax\n" - "paddsw %%xmm1,%%xmm0\n" - "movq (%5,%%rax,8),%%xmm1\n" - "lea (%%r10,%6),%%r11\n" - "sar $0x4,%%r10\n" - "movzb (%0,%%r10,1),%%rax\n" - "movq (%5,%%rax,8),%%xmm2\n" - "paddsw %%xmm0,%%xmm1\n" - "paddsw %%xmm0,%%xmm2\n" - "shufps $0x44,%%xmm2,%%xmm1\n" - "psraw $0x6,%%xmm1\n" - "packuswb %%xmm1,%%xmm1\n" - "movq %%xmm1,0x0(%3)\n" - "add $0x8,%3\n" - "sub $0x2,%4\n" - "jns scaleloop\n" - -"scalenext:" - "add $0x1,%4\n" - "js scaledone\n" - - "mov %%r11,%%r10\n" - "sar $0x5,%%r10\n" - "movzb (%1,%%r10,1),%%rax\n" - "movq 2048(%5,%%rax,8),%%xmm0\n" - "movzb (%2,%%r10,1),%%rax\n" - "movq 4096(%5,%%rax,8),%%xmm1\n" - "paddsw %%xmm1,%%xmm0\n" - "sar $0x4,%%r11\n" - "movzb (%0,%%r11,1),%%rax\n" - "movq (%5,%%rax,8),%%xmm1\n" - "paddsw %%xmm0,%%xmm1\n" - "psraw $0x6,%%xmm1\n" - "packuswb %%xmm1,%%xmm1\n" - "movd %%xmm1,0x0(%3)\n" - -"scaledone:" - : - : "r"(y_buf), // %0 - "r"(u_buf), // %1 - "r"(v_buf), // %2 - "r"(rgb_buf), // %3 - "r"(width), // %4 - "r" (kCoefficientsRgbY), // %5 - "r"(static_cast<long>(scaled_dx)) // %6 - : "memory", "r10", "r11", "rax", "xmm0", "xmm1", "xmm2" -); -} - -#endif // __SUNPRO_CC - -#else // ARCH_CPU_X86_64 - -#ifdef __SUNPRO_CC -void FastConvertYUVToRGB32Row(const uint8* y_buf, - const uint8* u_buf, - const uint8* v_buf, - uint8* rgb_buf, - int width) { - asm( - "pusha\n" - "mov %eax,%ebp\n" - "jmp convertend\n" - -"convertloop:" - "movzbl (%edi),%eax\n" - "add $0x1,%edi\n" - "movzbl (%esi),%ebx\n" - "add $0x1,%esi\n" - "movq kCoefficientsRgbY+2048(,%eax,8),%mm0\n" - "movzbl (%edx),%eax\n" - "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" - "movntq %mm1,0x0(%ebp)\n" - "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" - "movzbl (%esi),%eax\n" - "paddsw kCoefficientsRgbY+4096(,%eax,8),%mm0\n" - "movzbl (%edx),%eax\n" - "movq kCoefficientsRgbY(,%eax,8),%mm1\n" - "paddsw %mm0,%mm1\n" - "psraw $0x6,%mm1\n" - "packuswb %mm1,%mm1\n" - "movd %mm1,0x0(%ebp)\n" -"convertdone:" - "popa\n" - : - : "d"(y_buf), // %edx - "D"(u_buf), // %edi - "S"(v_buf), // %esi - "a"(rgb_buf), // %eax - "c"(width) // %ecx - : "memory" -); -} -#else // __SUNPRO_CC -void FastConvertYUVToRGB32Row(const uint8* y_buf, - const uint8* u_buf, - const uint8* v_buf, - uint8* rgb_buf, - int width); - -// It's necessary to specify the correct section for the following code, -// otherwise it will be placed in whatever the current section is as this unit -// is compiled. Because GCC remembers the last section it emitted, we must -// also revert to the previous section state at the end of the asm block. - asm( - ".section .text\n" - ".global FastConvertYUVToRGB32Row\n" - ".type FastConvertYUVToRGB32Row, @function\n" -"FastConvertYUVToRGB32Row:\n" - "pusha\n" - "mov 0x24(%esp),%edx\n" - "mov 0x28(%esp),%edi\n" - "mov 0x2c(%esp),%esi\n" - "mov 0x30(%esp),%ebp\n" - "mov 0x34(%esp),%ecx\n" - "jmp 1f\n" - -"0:" - "movzbl (%edi),%eax\n" - "add $0x1,%edi\n" - "movzbl (%esi),%ebx\n" - "add $0x1,%esi\n" - "movq kCoefficientsRgbY+2048(,%eax,8),%mm0\n" - "movzbl (%edx),%eax\n" - "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" - "movntq %mm1,0x0(%ebp)\n" - "add $0x8,%ebp\n" -"1:" - "sub $0x2,%ecx\n" - "jns 0b\n" - - "and $0x1,%ecx\n" - "je 2f\n" - - "movzbl (%edi),%eax\n" - "movq kCoefficientsRgbY+2048(,%eax,8),%mm0\n" - "movzbl (%esi),%eax\n" - "paddsw kCoefficientsRgbY+4096(,%eax,8),%mm0\n" - "movzbl (%edx),%eax\n" - "movq kCoefficientsRgbY(,%eax,8),%mm1\n" - "paddsw %mm0,%mm1\n" - "psraw $0x6,%mm1\n" - "packuswb %mm1,%mm1\n" - "movd %mm1,0x0(%ebp)\n" -"2:" - "popa\n" - "ret\n" - ".previous\n" -); - -void ScaleYUVToRGB32Row(const uint8* y_buf, - const uint8* u_buf, - const uint8* v_buf, - uint8* rgb_buf, - int width, - int scaled_dx); - - asm( - ".global ScaleYUVToRGB32Row\n" -"ScaleYUVToRGB32Row:\n" - "pusha\n" - "mov 0x24(%esp),%edx\n" - "mov 0x28(%esp),%edi\n" - "mov 0x2c(%esp),%esi\n" - "mov 0x30(%esp),%ebp\n" - "mov 0x34(%esp),%ecx\n" - "xor %ebx,%ebx\n" - "jmp scaleend\n" - -"scaleloop:" - "mov %ebx,%eax\n" - "sar $0x5,%eax\n" - "movzbl (%edi,%eax,1),%eax\n" - "movq kCoefficientsRgbY+2048(,%eax,8),%mm0\n" - "mov %ebx,%eax\n" - "sar $0x5,%eax\n" - "movzbl (%esi,%eax,1),%eax\n" - "paddsw kCoefficientsRgbY+4096(,%eax,8),%mm0\n" - "mov %ebx,%eax\n" - "add 0x38(%esp),%ebx\n" - "sar $0x4,%eax\n" - "movzbl (%edx,%eax,1),%eax\n" - "movq kCoefficientsRgbY(,%eax,8),%mm1\n" - "mov %ebx,%eax\n" - "add 0x38(%esp),%ebx\n" - "sar $0x4,%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" - "movntq %mm1,0x0(%ebp)\n" - "add $0x8,%ebp\n" -"scaleend:" - "sub $0x2,%ecx\n" - "jns scaleloop\n" - - "and $0x1,%ecx\n" - "je scaledone\n" - - "mov %ebx,%eax\n" - "sar $0x5,%eax\n" - "movzbl (%edi,%eax,1),%eax\n" - "movq kCoefficientsRgbY+2048(,%eax,8),%mm0\n" - "mov %ebx,%eax\n" - "sar $0x5,%eax\n" - "movzbl (%esi,%eax,1),%eax\n" - "paddsw kCoefficientsRgbY+4096(,%eax,8),%mm0\n" - "mov %ebx,%eax\n" - "sar $0x4,%eax\n" - "movzbl (%edx,%eax,1),%eax\n" - "movq kCoefficientsRgbY(,%eax,8),%mm1\n" - "paddsw %mm0,%mm1\n" - "psraw $0x6,%mm1\n" - "packuswb %mm1,%mm1\n" - "movd %mm1,0x0(%ebp)\n" - -"scaledone:" - "popa\n" - "ret\n" -); - -#endif // __SUNPRO_CC -#endif // ARCH_CPU_X86_64 -#endif // !ARCH_CPU_X86_FAMILY -} // extern "C" -
deleted file mode 100644 --- a/gfx/ycbcr/yuv_row_mac.cpp +++ /dev/null @@ -1,416 +0,0 @@ -// Copyright (c) 2009 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 "yuv_row.h" - -// TODO(fbarchard): Do 64 bit version. - -extern "C" { - -// PPC and 64 Bit builds use the C fallback. Optimized code -// needs to be fixed for 64 bit builds. PPC has no optimized code -// option at all. -#if defined(ARCH_CPU_PPC) || defined(ARCH_CPU_64_BITS) -// PPC implementation uses C fallback -void FastConvertYUVToRGB32Row(const uint8* y_buf, - const uint8* u_buf, - const uint8* v_buf, - uint8* rgb_buf, - int width) { - FastConvertYUVToRGB32Row_C(y_buf, u_buf, v_buf, rgb_buf, width, 1); -} - -void ScaleYUVToRGB32Row(const uint8* y_buf, - const uint8* u_buf, - const uint8* v_buf, - uint8* rgb_buf, - int width, - int scaled_dx) { - ScaleYUVToRGB32Row_C(y_buf, u_buf, v_buf, rgb_buf, width, scaled_dx, 1); -} -#else - -#define RGBY(i) { \ - static_cast<int16>(1.164 * 64 * (i - 16) + 0.5), \ - static_cast<int16>(1.164 * 64 * (i - 16) + 0.5), \ - static_cast<int16>(1.164 * 64 * (i - 16) + 0.5), \ - 0 \ -} - -#define RGBU(i) { \ - static_cast<int16>(2.018 * 64 * (i - 128) + 0.5), \ - static_cast<int16>(-0.391 * 64 * (i - 128) + 0.5), \ - 0, \ - static_cast<int16>(256 * 64 - 1) \ -} - -#define RGBV(i) { \ - 0, \ - static_cast<int16>(-0.813 * 64 * (i - 128) + 0.5), \ - static_cast<int16>(1.596 * 64 * (i - 128) + 0.5), \ - 0 \ -} - -#define MMX_ALIGNED(var) \ - var __attribute__ ((section ("__TEXT,__text"))) __attribute__ ((aligned(16))) - -MMX_ALIGNED(int16 kCoefficientsRgbY[768][4]) = { - RGBY(0x00), RGBY(0x01), RGBY(0x02), RGBY(0x03), - RGBY(0x04), RGBY(0x05), RGBY(0x06), RGBY(0x07), - RGBY(0x08), RGBY(0x09), RGBY(0x0A), RGBY(0x0B), - RGBY(0x0C), RGBY(0x0D), RGBY(0x0E), RGBY(0x0F), - RGBY(0x10), RGBY(0x11), RGBY(0x12), RGBY(0x13), - RGBY(0x14), RGBY(0x15), RGBY(0x16), RGBY(0x17), - RGBY(0x18), RGBY(0x19), RGBY(0x1A), RGBY(0x1B), - RGBY(0x1C), RGBY(0x1D), RGBY(0x1E), RGBY(0x1F), - RGBY(0x20), RGBY(0x21), RGBY(0x22), RGBY(0x23), - RGBY(0x24), RGBY(0x25), RGBY(0x26), RGBY(0x27), - RGBY(0x28), RGBY(0x29), RGBY(0x2A), RGBY(0x2B), - RGBY(0x2C), RGBY(0x2D), RGBY(0x2E), RGBY(0x2F), - RGBY(0x30), RGBY(0x31), RGBY(0x32), RGBY(0x33), - RGBY(0x34), RGBY(0x35), RGBY(0x36), RGBY(0x37), - RGBY(0x38), RGBY(0x39), RGBY(0x3A), RGBY(0x3B), - RGBY(0x3C), RGBY(0x3D), RGBY(0x3E), RGBY(0x3F), - RGBY(0x40), RGBY(0x41), RGBY(0x42), RGBY(0x43), - RGBY(0x44), RGBY(0x45), RGBY(0x46), RGBY(0x47), - RGBY(0x48), RGBY(0x49), RGBY(0x4A), RGBY(0x4B), - RGBY(0x4C), RGBY(0x4D), RGBY(0x4E), RGBY(0x4F), - RGBY(0x50), RGBY(0x51), RGBY(0x52), RGBY(0x53), - RGBY(0x54), RGBY(0x55), RGBY(0x56), RGBY(0x57), - RGBY(0x58), RGBY(0x59), RGBY(0x5A), RGBY(0x5B), - RGBY(0x5C), RGBY(0x5D), RGBY(0x5E), RGBY(0x5F), - RGBY(0x60), RGBY(0x61), RGBY(0x62), RGBY(0x63), - RGBY(0x64), RGBY(0x65), RGBY(0x66), RGBY(0x67), - RGBY(0x68), RGBY(0x69), RGBY(0x6A), RGBY(0x6B), - RGBY(0x6C), RGBY(0x6D), RGBY(0x6E), RGBY(0x6F), - RGBY(0x70), RGBY(0x71), RGBY(0x72), RGBY(0x73), - RGBY(0x74), RGBY(0x75), RGBY(0x76), RGBY(0x77), - RGBY(0x78), RGBY(0x79), RGBY(0x7A), RGBY(0x7B), - RGBY(0x7C), RGBY(0x7D), RGBY(0x7E), RGBY(0x7F), - RGBY(0x80), RGBY(0x81), RGBY(0x82), RGBY(0x83), - RGBY(0x84), RGBY(0x85), RGBY(0x86), RGBY(0x87), - RGBY(0x88), RGBY(0x89), RGBY(0x8A), RGBY(0x8B), - RGBY(0x8C), RGBY(0x8D), RGBY(0x8E), RGBY(0x8F), - RGBY(0x90), RGBY(0x91), RGBY(0x92), RGBY(0x93), - RGBY(0x94), RGBY(0x95), RGBY(0x96), RGBY(0x97), - RGBY(0x98), RGBY(0x99), RGBY(0x9A), RGBY(0x9B), - RGBY(0x9C), RGBY(0x9D), RGBY(0x9E), RGBY(0x9F), - RGBY(0xA0), RGBY(0xA1), RGBY(0xA2), RGBY(0xA3), - RGBY(0xA4), RGBY(0xA5), RGBY(0xA6), RGBY(0xA7), - RGBY(0xA8), RGBY(0xA9), RGBY(0xAA), RGBY(0xAB), - RGBY(0xAC), RGBY(0xAD), RGBY(0xAE), RGBY(0xAF), - RGBY(0xB0), RGBY(0xB1), RGBY(0xB2), RGBY(0xB3), - RGBY(0xB4), RGBY(0xB5), RGBY(0xB6), RGBY(0xB7), - RGBY(0xB8), RGBY(0xB9), RGBY(0xBA), RGBY(0xBB), - RGBY(0xBC), RGBY(0xBD), RGBY(0xBE), RGBY(0xBF), - RGBY(0xC0), RGBY(0xC1), RGBY(0xC2), RGBY(0xC3), - RGBY(0xC4), RGBY(0xC5), RGBY(0xC6), RGBY(0xC7), - RGBY(0xC8), RGBY(0xC9), RGBY(0xCA), RGBY(0xCB), - RGBY(0xCC), RGBY(0xCD), RGBY(0xCE), RGBY(0xCF), - RGBY(0xD0), RGBY(0xD1), RGBY(0xD2), RGBY(0xD3), - RGBY(0xD4), RGBY(0xD5), RGBY(0xD6), RGBY(0xD7), - RGBY(0xD8), RGBY(0xD9), RGBY(0xDA), RGBY(0xDB), - RGBY(0xDC), RGBY(0xDD), RGBY(0xDE), RGBY(0xDF), - RGBY(0xE0), RGBY(0xE1), RGBY(0xE2), RGBY(0xE3), - RGBY(0xE4), RGBY(0xE5), RGBY(0xE6), RGBY(0xE7), - RGBY(0xE8), RGBY(0xE9), RGBY(0xEA), RGBY(0xEB), - RGBY(0xEC), RGBY(0xED), RGBY(0xEE), RGBY(0xEF), - RGBY(0xF0), RGBY(0xF1), RGBY(0xF2), RGBY(0xF3), - RGBY(0xF4), RGBY(0xF5), RGBY(0xF6), RGBY(0xF7), - RGBY(0xF8), RGBY(0xF9), RGBY(0xFA), RGBY(0xFB), - RGBY(0xFC), RGBY(0xFD), RGBY(0xFE), RGBY(0xFF), - - // Chroma U table. - RGBU(0x00), RGBU(0x01), RGBU(0x02), RGBU(0x03), - RGBU(0x04), RGBU(0x05), RGBU(0x06), RGBU(0x07), - RGBU(0x08), RGBU(0x09), RGBU(0x0A), RGBU(0x0B), - RGBU(0x0C), RGBU(0x0D), RGBU(0x0E), RGBU(0x0F), - RGBU(0x10), RGBU(0x11), RGBU(0x12), RGBU(0x13), - RGBU(0x14), RGBU(0x15), RGBU(0x16), RGBU(0x17), - RGBU(0x18), RGBU(0x19), RGBU(0x1A), RGBU(0x1B), - RGBU(0x1C), RGBU(0x1D), RGBU(0x1E), RGBU(0x1F), - RGBU(0x20), RGBU(0x21), RGBU(0x22), RGBU(0x23), - RGBU(0x24), RGBU(0x25), RGBU(0x26), RGBU(0x27), - RGBU(0x28), RGBU(0x29), RGBU(0x2A), RGBU(0x2B), - RGBU(0x2C), RGBU(0x2D), RGBU(0x2E), RGBU(0x2F), - RGBU(0x30), RGBU(0x31), RGBU(0x32), RGBU(0x33), - RGBU(0x34), RGBU(0x35), RGBU(0x36), RGBU(0x37), - RGBU(0x38), RGBU(0x39), RGBU(0x3A), RGBU(0x3B), - RGBU(0x3C), RGBU(0x3D), RGBU(0x3E), RGBU(0x3F), - RGBU(0x40), RGBU(0x41), RGBU(0x42), RGBU(0x43), - RGBU(0x44), RGBU(0x45), RGBU(0x46), RGBU(0x47), - RGBU(0x48), RGBU(0x49), RGBU(0x4A), RGBU(0x4B), - RGBU(0x4C), RGBU(0x4D), RGBU(0x4E), RGBU(0x4F), - RGBU(0x50), RGBU(0x51), RGBU(0x52), RGBU(0x53), - RGBU(0x54), RGBU(0x55), RGBU(0x56), RGBU(0x57), - RGBU(0x58), RGBU(0x59), RGBU(0x5A), RGBU(0x5B), - RGBU(0x5C), RGBU(0x5D), RGBU(0x5E), RGBU(0x5F), - RGBU(0x60), RGBU(0x61), RGBU(0x62), RGBU(0x63), - RGBU(0x64), RGBU(0x65), RGBU(0x66), RGBU(0x67), - RGBU(0x68), RGBU(0x69), RGBU(0x6A), RGBU(0x6B), - RGBU(0x6C), RGBU(0x6D), RGBU(0x6E), RGBU(0x6F), - RGBU(0x70), RGBU(0x71), RGBU(0x72), RGBU(0x73), - RGBU(0x74), RGBU(0x75), RGBU(0x76), RGBU(0x77), - RGBU(0x78), RGBU(0x79), RGBU(0x7A), RGBU(0x7B), - RGBU(0x7C), RGBU(0x7D), RGBU(0x7E), RGBU(0x7F), - RGBU(0x80), RGBU(0x81), RGBU(0x82), RGBU(0x83), - RGBU(0x84), RGBU(0x85), RGBU(0x86), RGBU(0x87), - RGBU(0x88), RGBU(0x89), RGBU(0x8A), RGBU(0x8B), - RGBU(0x8C), RGBU(0x8D), RGBU(0x8E), RGBU(0x8F), - RGBU(0x90), RGBU(0x91), RGBU(0x92), RGBU(0x93), - RGBU(0x94), RGBU(0x95), RGBU(0x96), RGBU(0x97), - RGBU(0x98), RGBU(0x99), RGBU(0x9A), RGBU(0x9B), - RGBU(0x9C), RGBU(0x9D), RGBU(0x9E), RGBU(0x9F), - RGBU(0xA0), RGBU(0xA1), RGBU(0xA2), RGBU(0xA3), - RGBU(0xA4), RGBU(0xA5), RGBU(0xA6), RGBU(0xA7), - RGBU(0xA8), RGBU(0xA9), RGBU(0xAA), RGBU(0xAB), - RGBU(0xAC), RGBU(0xAD), RGBU(0xAE), RGBU(0xAF), - RGBU(0xB0), RGBU(0xB1), RGBU(0xB2), RGBU(0xB3), - RGBU(0xB4), RGBU(0xB5), RGBU(0xB6), RGBU(0xB7), - RGBU(0xB8), RGBU(0xB9), RGBU(0xBA), RGBU(0xBB), - RGBU(0xBC), RGBU(0xBD), RGBU(0xBE), RGBU(0xBF), - RGBU(0xC0), RGBU(0xC1), RGBU(0xC2), RGBU(0xC3), - RGBU(0xC4), RGBU(0xC5), RGBU(0xC6), RGBU(0xC7), - RGBU(0xC8), RGBU(0xC9), RGBU(0xCA), RGBU(0xCB), - RGBU(0xCC), RGBU(0xCD), RGBU(0xCE), RGBU(0xCF), - RGBU(0xD0), RGBU(0xD1), RGBU(0xD2), RGBU(0xD3), - RGBU(0xD4), RGBU(0xD5), RGBU(0xD6), RGBU(0xD7), - RGBU(0xD8), RGBU(0xD9), RGBU(0xDA), RGBU(0xDB), - RGBU(0xDC), RGBU(0xDD), RGBU(0xDE), RGBU(0xDF), - RGBU(0xE0), RGBU(0xE1), RGBU(0xE2), RGBU(0xE3), - RGBU(0xE4), RGBU(0xE5), RGBU(0xE6), RGBU(0xE7), - RGBU(0xE8), RGBU(0xE9), RGBU(0xEA), RGBU(0xEB), - RGBU(0xEC), RGBU(0xED), RGBU(0xEE), RGBU(0xEF), - RGBU(0xF0), RGBU(0xF1), RGBU(0xF2), RGBU(0xF3), - RGBU(0xF4), RGBU(0xF5), RGBU(0xF6), RGBU(0xF7), - RGBU(0xF8), RGBU(0xF9), RGBU(0xFA), RGBU(0xFB), - RGBU(0xFC), RGBU(0xFD), RGBU(0xFE), RGBU(0xFF), - - // Chroma V table. - RGBV(0x00), RGBV(0x01), RGBV(0x02), RGBV(0x03), - RGBV(0x04), RGBV(0x05), RGBV(0x06), RGBV(0x07), - RGBV(0x08), RGBV(0x09), RGBV(0x0A), RGBV(0x0B), - RGBV(0x0C), RGBV(0x0D), RGBV(0x0E), RGBV(0x0F), - RGBV(0x10), RGBV(0x11), RGBV(0x12), RGBV(0x13), - RGBV(0x14), RGBV(0x15), RGBV(0x16), RGBV(0x17), - RGBV(0x18), RGBV(0x19), RGBV(0x1A), RGBV(0x1B), - RGBV(0x1C), RGBV(0x1D), RGBV(0x1E), RGBV(0x1F), - RGBV(0x20), RGBV(0x21), RGBV(0x22), RGBV(0x23), - RGBV(0x24), RGBV(0x25), RGBV(0x26), RGBV(0x27), - RGBV(0x28), RGBV(0x29), RGBV(0x2A), RGBV(0x2B), - RGBV(0x2C), RGBV(0x2D), RGBV(0x2E), RGBV(0x2F), - RGBV(0x30), RGBV(0x31), RGBV(0x32), RGBV(0x33), - RGBV(0x34), RGBV(0x35), RGBV(0x36), RGBV(0x37), - RGBV(0x38), RGBV(0x39), RGBV(0x3A), RGBV(0x3B), - RGBV(0x3C), RGBV(0x3D), RGBV(0x3E), RGBV(0x3F), - RGBV(0x40), RGBV(0x41), RGBV(0x42), RGBV(0x43), - RGBV(0x44), RGBV(0x45), RGBV(0x46), RGBV(0x47), - RGBV(0x48), RGBV(0x49), RGBV(0x4A), RGBV(0x4B), - RGBV(0x4C), RGBV(0x4D), RGBV(0x4E), RGBV(0x4F), - RGBV(0x50), RGBV(0x51), RGBV(0x52), RGBV(0x53), - RGBV(0x54), RGBV(0x55), RGBV(0x56), RGBV(0x57), - RGBV(0x58), RGBV(0x59), RGBV(0x5A), RGBV(0x5B), - RGBV(0x5C), RGBV(0x5D), RGBV(0x5E), RGBV(0x5F), - RGBV(0x60), RGBV(0x61), RGBV(0x62), RGBV(0x63), - RGBV(0x64), RGBV(0x65), RGBV(0x66), RGBV(0x67), - RGBV(0x68), RGBV(0x69), RGBV(0x6A), RGBV(0x6B), - RGBV(0x6C), RGBV(0x6D), RGBV(0x6E), RGBV(0x6F), - RGBV(0x70), RGBV(0x71), RGBV(0x72), RGBV(0x73), - RGBV(0x74), RGBV(0x75), RGBV(0x76), RGBV(0x77), - RGBV(0x78), RGBV(0x79), RGBV(0x7A), RGBV(0x7B), - RGBV(0x7C), RGBV(0x7D), RGBV(0x7E), RGBV(0x7F), - RGBV(0x80), RGBV(0x81), RGBV(0x82), RGBV(0x83), - RGBV(0x84), RGBV(0x85), RGBV(0x86), RGBV(0x87), - RGBV(0x88), RGBV(0x89), RGBV(0x8A), RGBV(0x8B), - RGBV(0x8C), RGBV(0x8D), RGBV(0x8E), RGBV(0x8F), - RGBV(0x90), RGBV(0x91), RGBV(0x92), RGBV(0x93), - RGBV(0x94), RGBV(0x95), RGBV(0x96), RGBV(0x97), - RGBV(0x98), RGBV(0x99), RGBV(0x9A), RGBV(0x9B), - RGBV(0x9C), RGBV(0x9D), RGBV(0x9E), RGBV(0x9F), - RGBV(0xA0), RGBV(0xA1), RGBV(0xA2), RGBV(0xA3), - RGBV(0xA4), RGBV(0xA5), RGBV(0xA6), RGBV(0xA7), - RGBV(0xA8), RGBV(0xA9), RGBV(0xAA), RGBV(0xAB), - RGBV(0xAC), RGBV(0xAD), RGBV(0xAE), RGBV(0xAF), - RGBV(0xB0), RGBV(0xB1), RGBV(0xB2), RGBV(0xB3), - RGBV(0xB4), RGBV(0xB5), RGBV(0xB6), RGBV(0xB7), - RGBV(0xB8), RGBV(0xB9), RGBV(0xBA), RGBV(0xBB), - RGBV(0xBC), RGBV(0xBD), RGBV(0xBE), RGBV(0xBF), - RGBV(0xC0), RGBV(0xC1), RGBV(0xC2), RGBV(0xC3), - RGBV(0xC4), RGBV(0xC5), RGBV(0xC6), RGBV(0xC7), - RGBV(0xC8), RGBV(0xC9), RGBV(0xCA), RGBV(0xCB), - RGBV(0xCC), RGBV(0xCD), RGBV(0xCE), RGBV(0xCF), - RGBV(0xD0), RGBV(0xD1), RGBV(0xD2), RGBV(0xD3), - RGBV(0xD4), RGBV(0xD5), RGBV(0xD6), RGBV(0xD7), - RGBV(0xD8), RGBV(0xD9), RGBV(0xDA), RGBV(0xDB), - RGBV(0xDC), RGBV(0xDD), RGBV(0xDE), RGBV(0xDF), - RGBV(0xE0), RGBV(0xE1), RGBV(0xE2), RGBV(0xE3), - RGBV(0xE4), RGBV(0xE5), RGBV(0xE6), RGBV(0xE7), - RGBV(0xE8), RGBV(0xE9), RGBV(0xEA), RGBV(0xEB), - RGBV(0xEC), RGBV(0xED), RGBV(0xEE), RGBV(0xEF), - RGBV(0xF0), RGBV(0xF1), RGBV(0xF2), RGBV(0xF3), - RGBV(0xF4), RGBV(0xF5), RGBV(0xF6), RGBV(0xF7), - RGBV(0xF8), RGBV(0xF9), RGBV(0xFA), RGBV(0xFB), - RGBV(0xFC), RGBV(0xFD), RGBV(0xFE), RGBV(0xFF), -}; - -#undef RGBY -#undef RGBU -#undef RGBV -#undef MMX_ALIGNED - -extern void MacConvertYUVToRGB32Row(const uint8* y_buf, - const uint8* u_buf, - const uint8* v_buf, - uint8* rgb_buf, - int width, - int16 *kCoefficientsRgbY); - __asm__( -"_MacConvertYUVToRGB32Row:\n" - "pusha\n" - "mov 0x24(%esp),%edx\n" - "mov 0x28(%esp),%edi\n" - "mov 0x2c(%esp),%esi\n" - "mov 0x30(%esp),%ebp\n" - "mov 0x38(%esp),%ecx\n" - - "jmp Lconvertend\n" - -"Lconvertloop:" - "movzbl (%edi),%eax\n" - "add $0x1,%edi\n" - "movzbl (%esi),%ebx\n" - "add $0x1,%esi\n" - "movq 2048(%ecx,%eax,8),%mm0\n" - "movzbl (%edx),%eax\n" - "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" - "movntq %mm1,0x0(%ebp)\n" - "add $0x8,%ebp\n" -"Lconvertend:" - "sub $0x2,0x34(%esp)\n" - "jns Lconvertloop\n" - - "and $0x1,0x34(%esp)\n" - "je Lconvertdone\n" - - "movzbl (%edi),%eax\n" - "movq 2048(%ecx,%eax,8),%mm0\n" - "movzbl (%esi),%eax\n" - "paddsw 4096(%ecx,%eax,8),%mm0\n" - "movzbl (%edx),%eax\n" - "movq 0(%ecx,%eax,8),%mm1\n" - "paddsw %mm0,%mm1\n" - "psraw $0x6,%mm1\n" - "packuswb %mm1,%mm1\n" - "movd %mm1,0x0(%ebp)\n" -"Lconvertdone:\n" - "popa\n" - "ret\n" -); - -void FastConvertYUVToRGB32Row(const uint8* y_buf, - const uint8* u_buf, - const uint8* v_buf, - uint8* rgb_buf, - int width) { - MacConvertYUVToRGB32Row(y_buf, u_buf, v_buf, rgb_buf, width, - &kCoefficientsRgbY[0][0]); -} - -extern void MacScaleYUVToRGB32Row(const uint8* y_buf, - const uint8* u_buf, - const uint8* v_buf, - uint8* rgb_buf, - int width, - int scaled_dx, - int16 *kCoefficientsRgbY); - - __asm__( -"_MacScaleYUVToRGB32Row:\n" - "pusha\n" - "mov 0x24(%esp),%edx\n" - "mov 0x28(%esp),%edi\n" - "mov 0x2c(%esp),%esi\n" - "mov 0x30(%esp),%ebp\n" - "mov 0x3c(%esp),%ecx\n" - "xor %ebx,%ebx\n" - "jmp Lscaleend\n" - -"Lscaleloop:" - "mov %ebx,%eax\n" - "sar $0x5,%eax\n" - "movzbl (%edi,%eax,1),%eax\n" - "movq 2048(%ecx,%eax,8),%mm0\n" - "mov %ebx,%eax\n" - "sar $0x5,%eax\n" - "movzbl (%esi,%eax,1),%eax\n" - "paddsw 4096(%ecx,%eax,8),%mm0\n" - "mov %ebx,%eax\n" - "add 0x38(%esp),%ebx\n" - "sar $0x4,%eax\n" - "movzbl (%edx,%eax,1),%eax\n" - "movq 0(%ecx,%eax,8),%mm1\n" - "mov %ebx,%eax\n" - "add 0x38(%esp),%ebx\n" - "sar $0x4,%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" - "movntq %mm1,0x0(%ebp)\n" - "add $0x8,%ebp\n" -"Lscaleend:" - "sub $0x2,0x34(%esp)\n" - "jns Lscaleloop\n" - - "and $0x1,0x34(%esp)\n" - "je Lscaledone\n" - - "mov %ebx,%eax\n" - "sar $0x5,%eax\n" - "movzbl (%edi,%eax,1),%eax\n" - "movq 2048(%ecx,%eax,8),%mm0\n" - "mov %ebx,%eax\n" - "sar $0x5,%eax\n" - "movzbl (%esi,%eax,1),%eax\n" - "paddsw 4096(%ecx,%eax,8),%mm0\n" - "mov %ebx,%eax\n" - "sar $0x4,%eax\n" - "movzbl (%edx,%eax,1),%eax\n" - "movq 0(%ecx,%eax,8),%mm1\n" - "paddsw %mm0,%mm1\n" - "psraw $0x6,%mm1\n" - "packuswb %mm1,%mm1\n" - "movd %mm1,0x0(%ebp)\n" - -"Lscaledone:" - "popa\n" - "ret\n" -); - -void ScaleYUVToRGB32Row(const uint8* y_buf, - const uint8* u_buf, - const uint8* v_buf, - uint8* rgb_buf, - int width, - int scaled_dx) { - - MacScaleYUVToRGB32Row(y_buf, u_buf, v_buf, rgb_buf, width, scaled_dx, - &kCoefficientsRgbY[0][0]); -} - -#endif // ARCH_CPU_PPC || ARCH_CPU_64_BITS -} // extern "C" -
--- a/gfx/ycbcr/yuv_row_other.cpp +++ b/gfx/ycbcr/yuv_row_other.cpp @@ -7,9 +7,28 @@ extern "C" { void FastConvertYUVToRGB32Row(const uint8* y_buf, const uint8* u_buf, const uint8* v_buf, uint8* rgb_buf, int width) { FastConvertYUVToRGB32Row_C(y_buf, u_buf, v_buf, rgb_buf, width, 1); } + +void ScaleYUVToRGB32Row(const uint8* y_buf, + const uint8* u_buf, + const uint8* v_buf, + uint8* rgb_buf, + int width, + int source_dx) { + ScaleYUVToRGB32Row_C(y_buf, u_buf, v_buf, rgb_buf, width, source_dx); +} + +void LinearScaleYUVToRGB32Row(const uint8* y_buf, + const uint8* u_buf, + const uint8* v_buf, + uint8* rgb_buf, + int width, + int source_dx) { + LinearScaleYUVToRGB32Row_C(y_buf, u_buf, v_buf, rgb_buf, width, source_dx); +} + }
new file mode 100644 --- /dev/null +++ b/gfx/ycbcr/yuv_row_posix.cpp @@ -0,0 +1,849 @@ +// 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 "yuv_row.h" + +#define DCHECK(a) + +extern "C" { + +#if 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 1f\n" +"0:" + "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" + "paddsw %%xmm1,%%xmm0\n" + "movq (%5,%%r10,8),%%xmm2\n" + "add $0x2,%0\n" + "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" +"1:" + "sub $0x2,%4\n" + "jns 0b\n" + +"2:" + "add $0x1,%4\n" + "js 3f\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" +"3:" + : + : "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" +); +} + +void ScaleYUVToRGB32Row(const uint8* y_buf, // rdi + const uint8* u_buf, // rsi + const uint8* v_buf, // rdx + uint8* rgb_buf, // rcx + int width, // r8 + int source_dx) { // r9 + asm( + "xor %%r11,%%r11\n" + "sub $0x2,%4\n" + "js 1f\n" + +"0:" + "mov %%r11,%%r10\n" + "sar $0x11,%%r10\n" + "movzb (%1,%%r10,1),%%rax\n" + "movq 2048(%5,%%rax,8),%%xmm0\n" + "movzb (%2,%%r10,1),%%rax\n" + "movq 4096(%5,%%rax,8),%%xmm1\n" + "lea (%%r11,%6),%%r10\n" + "sar $0x10,%%r11\n" + "movzb (%0,%%r11,1),%%rax\n" + "paddsw %%xmm1,%%xmm0\n" + "movq (%5,%%rax,8),%%xmm1\n" + "lea (%%r10,%6),%%r11\n" + "sar $0x10,%%r10\n" + "movzb (%0,%%r10,1),%%rax\n" + "movq (%5,%%rax,8),%%xmm2\n" + "paddsw %%xmm0,%%xmm1\n" + "paddsw %%xmm0,%%xmm2\n" + "shufps $0x44,%%xmm2,%%xmm1\n" + "psraw $0x6,%%xmm1\n" + "packuswb %%xmm1,%%xmm1\n" + "movq %%xmm1,0x0(%3)\n" + "add $0x8,%3\n" + "sub $0x2,%4\n" + "jns 0b\n" + +"1:" + "add $0x1,%4\n" + "js 2f\n" + + "mov %%r11,%%r10\n" + "sar $0x11,%%r10\n" + "movzb (%1,%%r10,1),%%rax\n" + "movq 2048(%5,%%rax,8),%%xmm0\n" + "movzb (%2,%%r10,1),%%rax\n" + "movq 4096(%5,%%rax,8),%%xmm1\n" + "paddsw %%xmm1,%%xmm0\n" + "sar $0x10,%%r11\n" + "movzb (%0,%%r11,1),%%rax\n" + "movq (%5,%%rax,8),%%xmm1\n" + "paddsw %%xmm0,%%xmm1\n" + "psraw $0x6,%%xmm1\n" + "packuswb %%xmm1,%%xmm1\n" + "movd %%xmm1,0x0(%3)\n" + +"2:" + : + : "r"(y_buf), // %0 + "r"(u_buf), // %1 + "r"(v_buf), // %2 + "r"(rgb_buf), // %3 + "r"(width), // %4 + "r" (kCoefficientsRgbY), // %5 + "r"(static_cast<long>(source_dx)) // %6 + : "memory", "r10", "r11", "rax", "xmm0", "xmm1", "xmm2" +); +} + +void LinearScaleYUVToRGB32Row(const uint8* y_buf, + const uint8* u_buf, + const uint8* v_buf, + uint8* rgb_buf, + int width, + int source_dx) { + asm( + "xor %%r11,%%r11\n" // x = 0 + "sub $0x2,%4\n" + "js 2f\n" + "cmp $0x20000,%6\n" // if source_dx >= 2.0 + "jl 0f\n" + "mov $0x8000,%%r11\n" // x = 0.5 for 1/2 or less +"0:" + +"1:" + "mov %%r11,%%r10\n" + "sar $0x11,%%r10\n" + + "movzb (%1, %%r10, 1), %%r13 \n" + "movzb 1(%1, %%r10, 1), %%r14 \n" + "mov %%r11, %%rax \n" + "and $0x1fffe, %%rax \n" + "imul %%rax, %%r14 \n" + "xor $0x1fffe, %%rax \n" + "imul %%rax, %%r13 \n" + "add %%r14, %%r13 \n" + "shr $17, %%r13 \n" + "movq 2048(%5,%%r13,8), %%xmm0\n" + + "movzb (%2, %%r10, 1), %%r13 \n" + "movzb 1(%2, %%r10, 1), %%r14 \n" + "mov %%r11, %%rax \n" + "and $0x1fffe, %%rax \n" + "imul %%rax, %%r14 \n" + "xor $0x1fffe, %%rax \n" + "imul %%rax, %%r13 \n" + "add %%r14, %%r13 \n" + "shr $17, %%r13 \n" + "movq 4096(%5,%%r13,8), %%xmm1\n" + + "mov %%r11, %%rax \n" + "lea (%%r11,%6),%%r10\n" + "sar $0x10,%%r11\n" + "paddsw %%xmm1,%%xmm0\n" + + "movzb (%0, %%r11, 1), %%r13 \n" + "movzb 1(%0, %%r11, 1), %%r14 \n" + "and $0xffff, %%rax \n" + "imul %%rax, %%r14 \n" + "xor $0xffff, %%rax \n" + "imul %%rax, %%r13 \n" + "add %%r14, %%r13 \n" + "shr $16, %%r13 \n" + "movq (%5,%%r13,8),%%xmm1\n" + + "mov %%r10, %%rax \n" + "lea (%%r10,%6),%%r11\n" + "sar $0x10,%%r10\n" + + "movzb (%0,%%r10,1), %%r13 \n" + "movzb 1(%0,%%r10,1), %%r14 \n" + "and $0xffff, %%rax \n" + "imul %%rax, %%r14 \n" + "xor $0xffff, %%rax \n" + "imul %%rax, %%r13 \n" + "add %%r14, %%r13 \n" + "shr $16, %%r13 \n" + "movq (%5,%%r13,8),%%xmm2\n" + + "paddsw %%xmm0,%%xmm1\n" + "paddsw %%xmm0,%%xmm2\n" + "shufps $0x44,%%xmm2,%%xmm1\n" + "psraw $0x6,%%xmm1\n" + "packuswb %%xmm1,%%xmm1\n" + "movq %%xmm1,0x0(%3)\n" + "add $0x8,%3\n" + "sub $0x2,%4\n" + "jns 1b\n" + +"2:" + "add $0x1,%4\n" + "js 3f\n" + + "mov %%r11,%%r10\n" + "sar $0x11,%%r10\n" + + "movzb (%1,%%r10,1), %%r13 \n" + "movq 2048(%5,%%r13,8),%%xmm0\n" + + "movzb (%2,%%r10,1), %%r13 \n" + "movq 4096(%5,%%r13,8),%%xmm1\n" + + "paddsw %%xmm1,%%xmm0\n" + "sar $0x10,%%r11\n" + + "movzb (%0,%%r11,1), %%r13 \n" + "movq (%5,%%r13,8),%%xmm1\n" + + "paddsw %%xmm0,%%xmm1\n" + "psraw $0x6,%%xmm1\n" + "packuswb %%xmm1,%%xmm1\n" + "movd %%xmm1,0x0(%3)\n" + +"3:" + : + : "r"(y_buf), // %0 + "r"(u_buf), // %1 + "r"(v_buf), // %2 + "r"(rgb_buf), // %3 + "r"(width), // %4 + "r" (kCoefficientsRgbY), // %5 + "r"(static_cast<long>(source_dx)) // %6 + : "memory", "r10", "r11", "r13", "r14", "rax", "xmm0", "xmm1", "xmm2" +); +} + +#elif defined(ARCH_CPU_X86_32) && !defined(__PIC__) + +// PIC version is slower because less registers are available, so +// non-PIC is used on platforms where it is possible. +void FastConvertYUVToRGB32Row(const uint8* y_buf, + const uint8* u_buf, + const uint8* v_buf, + uint8* rgb_buf, + int width); + asm( + ".text\n" + ".global FastConvertYUVToRGB32Row\n" + ".type FastConvertYUVToRGB32Row, @function\n" +"FastConvertYUVToRGB32Row:\n" + "pusha\n" + "mov 0x24(%esp),%edx\n" + "mov 0x28(%esp),%edi\n" + "mov 0x2c(%esp),%esi\n" + "mov 0x30(%esp),%ebp\n" + "mov 0x34(%esp),%ecx\n" + "jmp 1f\n" + +"0:" + "movzbl (%edi),%eax\n" + "add $0x1,%edi\n" + "movzbl (%esi),%ebx\n" + "add $0x1,%esi\n" + "movq kCoefficientsRgbY+2048(,%eax,8),%mm0\n" + "movzbl (%edx),%eax\n" + "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" + "movntq %mm1,0x0(%ebp)\n" + "add $0x8,%ebp\n" +"1:" + "sub $0x2,%ecx\n" + "jns 0b\n" + + "and $0x1,%ecx\n" + "je 2f\n" + + "movzbl (%edi),%eax\n" + "movq kCoefficientsRgbY+2048(,%eax,8),%mm0\n" + "movzbl (%esi),%eax\n" + "paddsw kCoefficientsRgbY+4096(,%eax,8),%mm0\n" + "movzbl (%edx),%eax\n" + "movq kCoefficientsRgbY(,%eax,8),%mm1\n" + "paddsw %mm0,%mm1\n" + "psraw $0x6,%mm1\n" + "packuswb %mm1,%mm1\n" + "movd %mm1,0x0(%ebp)\n" +"2:" + "popa\n" + "ret\n" +#if !defined(XP_MACOSX) + ".previous\n" +#endif +); + + +void ScaleYUVToRGB32Row(const uint8* y_buf, + const uint8* u_buf, + const uint8* v_buf, + uint8* rgb_buf, + int width, + int source_dx); + asm( + ".text\n" + ".global ScaleYUVToRGB32Row\n" + ".type ScaleYUVToRGB32Row, @function\n" +"ScaleYUVToRGB32Row:\n" + "pusha\n" + "mov 0x24(%esp),%edx\n" + "mov 0x28(%esp),%edi\n" + "mov 0x2c(%esp),%esi\n" + "mov 0x30(%esp),%ebp\n" + "mov 0x34(%esp),%ecx\n" + "xor %ebx,%ebx\n" + "jmp 1f\n" + +"0:" + "mov %ebx,%eax\n" + "sar $0x11,%eax\n" + "movzbl (%edi,%eax,1),%eax\n" + "movq kCoefficientsRgbY+2048(,%eax,8),%mm0\n" + "mov %ebx,%eax\n" + "sar $0x11,%eax\n" + "movzbl (%esi,%eax,1),%eax\n" + "paddsw kCoefficientsRgbY+4096(,%eax,8),%mm0\n" + "mov %ebx,%eax\n" + "add 0x38(%esp),%ebx\n" + "sar $0x10,%eax\n" + "movzbl (%edx,%eax,1),%eax\n" + "movq kCoefficientsRgbY(,%eax,8),%mm1\n" + "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" + "movntq %mm1,0x0(%ebp)\n" + "add $0x8,%ebp\n" +"1:" + "sub $0x2,%ecx\n" + "jns 0b\n" + + "and $0x1,%ecx\n" + "je 2f\n" + + "mov %ebx,%eax\n" + "sar $0x11,%eax\n" + "movzbl (%edi,%eax,1),%eax\n" + "movq kCoefficientsRgbY+2048(,%eax,8),%mm0\n" + "mov %ebx,%eax\n" + "sar $0x11,%eax\n" + "movzbl (%esi,%eax,1),%eax\n" + "paddsw kCoefficientsRgbY+4096(,%eax,8),%mm0\n" + "mov %ebx,%eax\n" + "sar $0x10,%eax\n" + "movzbl (%edx,%eax,1),%eax\n" + "movq kCoefficientsRgbY(,%eax,8),%mm1\n" + "paddsw %mm0,%mm1\n" + "psraw $0x6,%mm1\n" + "packuswb %mm1,%mm1\n" + "movd %mm1,0x0(%ebp)\n" + +"2:" + "popa\n" + "ret\n" +#if !defined(XP_MACOSX) + ".previous\n" +#endif +); + +void LinearScaleYUVToRGB32Row(const uint8* y_buf, + const uint8* u_buf, + const uint8* v_buf, + uint8* rgb_buf, + int width, + int source_dx); + asm( + ".text\n" + ".global LinearScaleYUVToRGB32Row\n" + ".type LinearScaleYUVToRGB32Row, @function\n" +"LinearScaleYUVToRGB32Row:\n" + "pusha\n" + "mov 0x24(%esp),%edx\n" + "mov 0x28(%esp),%edi\n" + "mov 0x30(%esp),%ebp\n" + + // source_width = width * source_dx + ebx + "mov 0x34(%esp), %ecx\n" + "imull 0x38(%esp), %ecx\n" + "mov %ecx, 0x34(%esp)\n" + + "mov 0x38(%esp), %ecx\n" + "xor %ebx,%ebx\n" // x = 0 + "cmp $0x20000,%ecx\n" // if source_dx >= 2.0 + "jl 1f\n" + "mov $0x8000,%ebx\n" // x = 0.5 for 1/2 or less + "jmp 1f\n" + +"0:" + "mov %ebx,%eax\n" + "sar $0x11,%eax\n" + + "movzbl (%edi,%eax,1),%ecx\n" + "movzbl 1(%edi,%eax,1),%esi\n" + "mov %ebx,%eax\n" + "andl $0x1fffe, %eax \n" + "imul %eax, %esi \n" + "xorl $0x1fffe, %eax \n" + "imul %eax, %ecx \n" + "addl %esi, %ecx \n" + "shrl $17, %ecx \n" + "movq kCoefficientsRgbY+2048(,%ecx,8),%mm0\n" + + "mov 0x2c(%esp),%esi\n" + "mov %ebx,%eax\n" + "sar $0x11,%eax\n" + + "movzbl (%esi,%eax,1),%ecx\n" + "movzbl 1(%esi,%eax,1),%esi\n" + "mov %ebx,%eax\n" + "andl $0x1fffe, %eax \n" + "imul %eax, %esi \n" + "xorl $0x1fffe, %eax \n" + "imul %eax, %ecx \n" + "addl %esi, %ecx \n" + "shrl $17, %ecx \n" + "paddsw kCoefficientsRgbY+4096(,%ecx,8),%mm0\n" + + "mov %ebx,%eax\n" + "sar $0x10,%eax\n" + "movzbl (%edx,%eax,1),%ecx\n" + "movzbl 1(%edx,%eax,1),%esi\n" + "mov %ebx,%eax\n" + "add 0x38(%esp),%ebx\n" + "andl $0xffff, %eax \n" + "imul %eax, %esi \n" + "xorl $0xffff, %eax \n" + "imul %eax, %ecx \n" + "addl %esi, %ecx \n" + "shrl $16, %ecx \n" + "movq kCoefficientsRgbY(,%ecx,8),%mm1\n" + + "cmp 0x34(%esp), %ebx\n" + "jge 2f\n" + + "mov %ebx,%eax\n" + "sar $0x10,%eax\n" + "movzbl (%edx,%eax,1),%ecx\n" + "movzbl 1(%edx,%eax,1),%esi\n" + "mov %ebx,%eax\n" + "add 0x38(%esp),%ebx\n" + "andl $0xffff, %eax \n" + "imul %eax, %esi \n" + "xorl $0xffff, %eax \n" + "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" + "movntq %mm1,0x0(%ebp)\n" + "add $0x8,%ebp\n" + +"1:" + "cmp 0x34(%esp), %ebx\n" + "jl 0b\n" + "popa\n" + "ret\n" + +"2:" + "paddsw %mm0, %mm1\n" + "psraw $6, %mm1\n" + "packuswb %mm1, %mm1\n" + "movd %mm1, (%ebp)\n" + "popa\n" + "ret\n" +#if !defined(XP_MACOSX) + ".previous\n" +#endif +); + +#elif defined(ARCH_CPU_X86_32) && defined(__PIC__) + +void PICConvertYUVToRGB32Row(const uint8* y_buf, + const uint8* u_buf, + const uint8* v_buf, + uint8* rgb_buf, + int width, + int16 *kCoefficientsRgbY); + + asm( + ".text\n" +#if defined(XP_MACOSX) +"_PICConvertYUVToRGB32Row:\n" +#else +"PICConvertYUVToRGB32Row:\n" +#endif + "pusha\n" + "mov 0x24(%esp),%edx\n" + "mov 0x28(%esp),%edi\n" + "mov 0x2c(%esp),%esi\n" + "mov 0x30(%esp),%ebp\n" + "mov 0x38(%esp),%ecx\n" + + "jmp 1f\n" + +"0:" + "movzbl (%edi),%eax\n" + "add $0x1,%edi\n" + "movzbl (%esi),%ebx\n" + "add $0x1,%esi\n" + "movq 2048(%ecx,%eax,8),%mm0\n" + "movzbl (%edx),%eax\n" + "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" + "movntq %mm1,0x0(%ebp)\n" + "add $0x8,%ebp\n" +"1:" + "subl $0x2,0x34(%esp)\n" + "jns 0b\n" + + "andl $0x1,0x34(%esp)\n" + "je 2f\n" + + "movzbl (%edi),%eax\n" + "movq 2048(%ecx,%eax,8),%mm0\n" + "movzbl (%esi),%eax\n" + "paddsw 4096(%ecx,%eax,8),%mm0\n" + "movzbl (%edx),%eax\n" + "movq 0(%ecx,%eax,8),%mm1\n" + "paddsw %mm0,%mm1\n" + "psraw $0x6,%mm1\n" + "packuswb %mm1,%mm1\n" + "movd %mm1,0x0(%ebp)\n" +"2:" + "popa\n" + "ret\n" +#if !defined(XP_MACOSX) + ".previous\n" +#endif +); + +void FastConvertYUVToRGB32Row(const uint8* y_buf, + const uint8* u_buf, + const uint8* v_buf, + uint8* rgb_buf, + int width) { + PICConvertYUVToRGB32Row(y_buf, u_buf, v_buf, rgb_buf, width, + &kCoefficientsRgbY[0][0]); +} + +void PICScaleYUVToRGB32Row(const uint8* y_buf, + const uint8* u_buf, + const uint8* v_buf, + uint8* rgb_buf, + int width, + int source_dx, + int16 *kCoefficientsRgbY); + + asm( + ".text\n" +#if defined(XP_MACOSX) +"_PICScaleYUVToRGB32Row:\n" +#else +"PICScaleYUVToRGB32Row:\n" +#endif + "pusha\n" + "mov 0x24(%esp),%edx\n" + "mov 0x28(%esp),%edi\n" + "mov 0x2c(%esp),%esi\n" + "mov 0x30(%esp),%ebp\n" + "mov 0x3c(%esp),%ecx\n" + "xor %ebx,%ebx\n" + "jmp 1f\n" + +"0:" + "mov %ebx,%eax\n" + "sar $0x11,%eax\n" + "movzbl (%edi,%eax,1),%eax\n" + "movq 2048(%ecx,%eax,8),%mm0\n" + "mov %ebx,%eax\n" + "sar $0x11,%eax\n" + "movzbl (%esi,%eax,1),%eax\n" + "paddsw 4096(%ecx,%eax,8),%mm0\n" + "mov %ebx,%eax\n" + "add 0x38(%esp),%ebx\n" + "sar $0x10,%eax\n" + "movzbl (%edx,%eax,1),%eax\n" + "movq 0(%ecx,%eax,8),%mm1\n" + "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" + "movntq %mm1,0x0(%ebp)\n" + "add $0x8,%ebp\n" +"1:" + "subl $0x2,0x34(%esp)\n" + "jns 0b\n" + + "andl $0x1,0x34(%esp)\n" + "je 2f\n" + + "mov %ebx,%eax\n" + "sar $0x11,%eax\n" + "movzbl (%edi,%eax,1),%eax\n" + "movq 2048(%ecx,%eax,8),%mm0\n" + "mov %ebx,%eax\n" + "sar $0x11,%eax\n" + "movzbl (%esi,%eax,1),%eax\n" + "paddsw 4096(%ecx,%eax,8),%mm0\n" + "mov %ebx,%eax\n" + "sar $0x10,%eax\n" + "movzbl (%edx,%eax,1),%eax\n" + "movq 0(%ecx,%eax,8),%mm1\n" + "paddsw %mm0,%mm1\n" + "psraw $0x6,%mm1\n" + "packuswb %mm1,%mm1\n" + "movd %mm1,0x0(%ebp)\n" + +"2:" + "popa\n" + "ret\n" +#if !defined(XP_MACOSX) + ".previous\n" +#endif +); + +void ScaleYUVToRGB32Row(const uint8* y_buf, + const uint8* u_buf, + const uint8* v_buf, + uint8* rgb_buf, + int width, + int source_dx) { + PICScaleYUVToRGB32Row(y_buf, u_buf, v_buf, rgb_buf, width, source_dx, + &kCoefficientsRgbY[0][0]); +} + +void PICLinearScaleYUVToRGB32Row(const uint8* y_buf, + const uint8* u_buf, + const uint8* v_buf, + uint8* rgb_buf, + int width, + int source_dx, + int16 *kCoefficientsRgbY); + + asm( + ".text\n" +#if defined(XP_MACOSX) +"_PICLinearScaleYUVToRGB32Row:\n" +#else +"PICLinearScaleYUVToRGB32Row:\n" +#endif + "pusha\n" + "mov 0x24(%esp),%edx\n" + "mov 0x30(%esp),%ebp\n" + "mov 0x34(%esp),%ecx\n" + "mov 0x3c(%esp),%edi\n" + "xor %ebx,%ebx\n" + + // source_width = width * source_dx + ebx + "mov 0x34(%esp), %ecx\n" + "imull 0x38(%esp), %ecx\n" + "mov %ecx, 0x34(%esp)\n" + + "mov 0x38(%esp), %ecx\n" + "xor %ebx,%ebx\n" // x = 0 + "cmp $0x20000,%ecx\n" // if source_dx >= 2.0 + "jl 1f\n" + "mov $0x8000,%ebx\n" // x = 0.5 for 1/2 or less + "jmp 1f\n" + +"0:" + "mov 0x28(%esp),%esi\n" + "mov %ebx,%eax\n" + "sar $0x11,%eax\n" + + "movzbl (%esi,%eax,1),%ecx\n" + "movzbl 1(%esi,%eax,1),%esi\n" + "mov %ebx,%eax\n" + "andl $0x1fffe, %eax \n" + "imul %eax, %esi \n" + "xorl $0x1fffe, %eax \n" + "imul %eax, %ecx \n" + "addl %esi, %ecx \n" + "shrl $17, %ecx \n" + "movq 2048(%edi,%ecx,8),%mm0\n" + + "mov 0x2c(%esp),%esi\n" + "mov %ebx,%eax\n" + "sar $0x11,%eax\n" + + "movzbl (%esi,%eax,1),%ecx\n" + "movzbl 1(%esi,%eax,1),%esi\n" + "mov %ebx,%eax\n" + "andl $0x1fffe, %eax \n" + "imul %eax, %esi \n" + "xorl $0x1fffe, %eax \n" + "imul %eax, %ecx \n" + "addl %esi, %ecx \n" + "shrl $17, %ecx \n" + "paddsw 4096(%edi,%ecx,8),%mm0\n" + + "mov %ebx,%eax\n" + "sar $0x10,%eax\n" + "movzbl (%edx,%eax,1),%ecx\n" + "movzbl 1(%edx,%eax,1),%esi\n" + "mov %ebx,%eax\n" + "add 0x38(%esp),%ebx\n" + "andl $0xffff, %eax \n" + "imul %eax, %esi \n" + "xorl $0xffff, %eax \n" + "imul %eax, %ecx \n" + "addl %esi, %ecx \n" + "shrl $16, %ecx \n" + "movq (%edi,%ecx,8),%mm1\n" + + "cmp 0x34(%esp), %ebx\n" + "jge 2f\n" + + "mov %ebx,%eax\n" + "sar $0x10,%eax\n" + "movzbl (%edx,%eax,1),%ecx\n" + "movzbl 1(%edx,%eax,1),%esi\n" + "mov %ebx,%eax\n" + "add 0x38(%esp),%ebx\n" + "andl $0xffff, %eax \n" + "imul %eax, %esi \n" + "xorl $0xffff, %eax \n" + "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" + "movntq %mm1,0x0(%ebp)\n" + "add $0x8,%ebp\n" + +"1:" + "cmp %ebx, 0x34(%esp)\n" + "jg 0b\n" + "popa\n" + "ret\n" + +"2:" + "paddsw %mm0, %mm1\n" + "psraw $6, %mm1\n" + "packuswb %mm1, %mm1\n" + "movd %mm1, (%ebp)\n" + "popa\n" + "ret\n" +#if !defined(XP_MACOSX) + ".previous\n" +#endif +); + + +void LinearScaleYUVToRGB32Row(const uint8* y_buf, + const uint8* u_buf, + const uint8* v_buf, + uint8* rgb_buf, + int width, + int source_dx) { + PICLinearScaleYUVToRGB32Row(y_buf, u_buf, v_buf, rgb_buf, width, source_dx, + &kCoefficientsRgbY[0][0]); +} +#else +void FastConvertYUVToRGB32Row(const uint8* y_buf, + const uint8* u_buf, + const uint8* v_buf, + uint8* rgb_buf, + int width) { + FastConvertYUVToRGB32Row_C(y_buf, u_buf, v_buf, rgb_buf, width, 1); +} + +void ScaleYUVToRGB32Row(const uint8* y_buf, + const uint8* u_buf, + const uint8* v_buf, + uint8* rgb_buf, + int width, + int source_dx) { + ScaleYUVToRGB32Row_C(y_buf, u_buf, v_buf, rgb_buf, width, source_dx); +} + +void LinearScaleYUVToRGB32Row(const uint8* y_buf, + const uint8* u_buf, + const uint8* v_buf, + uint8* rgb_buf, + int width, + int source_dx) { + LinearScaleYUVToRGB32Row_C(y_buf, u_buf, v_buf, rgb_buf, width, source_dx); +} +#endif + +} // extern "C" +
new file mode 100644 --- /dev/null +++ b/gfx/ycbcr/yuv_row_table.cpp @@ -0,0 +1,233 @@ +// 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 "yuv_row.h" + +extern "C" { + +#define RGBY(i) { \ + static_cast<int16>(1.164 * 64 * (i - 16) + 0.5), \ + static_cast<int16>(1.164 * 64 * (i - 16) + 0.5), \ + static_cast<int16>(1.164 * 64 * (i - 16) + 0.5), \ + 0 \ +} + +#define RGBU(i) { \ + static_cast<int16>(2.018 * 64 * (i - 128) + 0.5), \ + static_cast<int16>(-0.391 * 64 * (i - 128) + 0.5), \ + 0, \ + static_cast<int16>(256 * 64 - 1) \ +} + +#define RGBV(i) { \ + 0, \ + static_cast<int16>(-0.813 * 64 * (i - 128) + 0.5), \ + static_cast<int16>(1.596 * 64 * (i - 128) + 0.5), \ + 0 \ +} + +SIMD_ALIGNED(int16 kCoefficientsRgbY[256 * 3][4]) = { + RGBY(0x00), RGBY(0x01), RGBY(0x02), RGBY(0x03), + RGBY(0x04), RGBY(0x05), RGBY(0x06), RGBY(0x07), + RGBY(0x08), RGBY(0x09), RGBY(0x0A), RGBY(0x0B), + RGBY(0x0C), RGBY(0x0D), RGBY(0x0E), RGBY(0x0F), + RGBY(0x10), RGBY(0x11), RGBY(0x12), RGBY(0x13), + RGBY(0x14), RGBY(0x15), RGBY(0x16), RGBY(0x17), + RGBY(0x18), RGBY(0x19), RGBY(0x1A), RGBY(0x1B), + RGBY(0x1C), RGBY(0x1D), RGBY(0x1E), RGBY(0x1F), + RGBY(0x20), RGBY(0x21), RGBY(0x22), RGBY(0x23), + RGBY(0x24), RGBY(0x25), RGBY(0x26), RGBY(0x27), + RGBY(0x28), RGBY(0x29), RGBY(0x2A), RGBY(0x2B), + RGBY(0x2C), RGBY(0x2D), RGBY(0x2E), RGBY(0x2F), + RGBY(0x30), RGBY(0x31), RGBY(0x32), RGBY(0x33), + RGBY(0x34), RGBY(0x35), RGBY(0x36), RGBY(0x37), + RGBY(0x38), RGBY(0x39), RGBY(0x3A), RGBY(0x3B), + RGBY(0x3C), RGBY(0x3D), RGBY(0x3E), RGBY(0x3F), + RGBY(0x40), RGBY(0x41), RGBY(0x42), RGBY(0x43), + RGBY(0x44), RGBY(0x45), RGBY(0x46), RGBY(0x47), + RGBY(0x48), RGBY(0x49), RGBY(0x4A), RGBY(0x4B), + RGBY(0x4C), RGBY(0x4D), RGBY(0x4E), RGBY(0x4F), + RGBY(0x50), RGBY(0x51), RGBY(0x52), RGBY(0x53), + RGBY(0x54), RGBY(0x55), RGBY(0x56), RGBY(0x57), + RGBY(0x58), RGBY(0x59), RGBY(0x5A), RGBY(0x5B), + RGBY(0x5C), RGBY(0x5D), RGBY(0x5E), RGBY(0x5F), + RGBY(0x60), RGBY(0x61), RGBY(0x62), RGBY(0x63), + RGBY(0x64), RGBY(0x65), RGBY(0x66), RGBY(0x67), + RGBY(0x68), RGBY(0x69), RGBY(0x6A), RGBY(0x6B), + RGBY(0x6C), RGBY(0x6D), RGBY(0x6E), RGBY(0x6F), + RGBY(0x70), RGBY(0x71), RGBY(0x72), RGBY(0x73), + RGBY(0x74), RGBY(0x75), RGBY(0x76), RGBY(0x77), + RGBY(0x78), RGBY(0x79), RGBY(0x7A), RGBY(0x7B), + RGBY(0x7C), RGBY(0x7D), RGBY(0x7E), RGBY(0x7F), + RGBY(0x80), RGBY(0x81), RGBY(0x82), RGBY(0x83), + RGBY(0x84), RGBY(0x85), RGBY(0x86), RGBY(0x87), + RGBY(0x88), RGBY(0x89), RGBY(0x8A), RGBY(0x8B), + RGBY(0x8C), RGBY(0x8D), RGBY(0x8E), RGBY(0x8F), + RGBY(0x90), RGBY(0x91), RGBY(0x92), RGBY(0x93), + RGBY(0x94), RGBY(0x95), RGBY(0x96), RGBY(0x97), + RGBY(0x98), RGBY(0x99), RGBY(0x9A), RGBY(0x9B), + RGBY(0x9C), RGBY(0x9D), RGBY(0x9E), RGBY(0x9F), + RGBY(0xA0), RGBY(0xA1), RGBY(0xA2), RGBY(0xA3), + RGBY(0xA4), RGBY(0xA5), RGBY(0xA6), RGBY(0xA7), + RGBY(0xA8), RGBY(0xA9), RGBY(0xAA), RGBY(0xAB), + RGBY(0xAC), RGBY(0xAD), RGBY(0xAE), RGBY(0xAF), + RGBY(0xB0), RGBY(0xB1), RGBY(0xB2), RGBY(0xB3), + RGBY(0xB4), RGBY(0xB5), RGBY(0xB6), RGBY(0xB7), + RGBY(0xB8), RGBY(0xB9), RGBY(0xBA), RGBY(0xBB), + RGBY(0xBC), RGBY(0xBD), RGBY(0xBE), RGBY(0xBF), + RGBY(0xC0), RGBY(0xC1), RGBY(0xC2), RGBY(0xC3), + RGBY(0xC4), RGBY(0xC5), RGBY(0xC6), RGBY(0xC7), + RGBY(0xC8), RGBY(0xC9), RGBY(0xCA), RGBY(0xCB), + RGBY(0xCC), RGBY(0xCD), RGBY(0xCE), RGBY(0xCF), + RGBY(0xD0), RGBY(0xD1), RGBY(0xD2), RGBY(0xD3), + RGBY(0xD4), RGBY(0xD5), RGBY(0xD6), RGBY(0xD7), + RGBY(0xD8), RGBY(0xD9), RGBY(0xDA), RGBY(0xDB), + RGBY(0xDC), RGBY(0xDD), RGBY(0xDE), RGBY(0xDF), + RGBY(0xE0), RGBY(0xE1), RGBY(0xE2), RGBY(0xE3), + RGBY(0xE4), RGBY(0xE5), RGBY(0xE6), RGBY(0xE7), + RGBY(0xE8), RGBY(0xE9), RGBY(0xEA), RGBY(0xEB), + RGBY(0xEC), RGBY(0xED), RGBY(0xEE), RGBY(0xEF), + RGBY(0xF0), RGBY(0xF1), RGBY(0xF2), RGBY(0xF3), + RGBY(0xF4), RGBY(0xF5), RGBY(0xF6), RGBY(0xF7), + RGBY(0xF8), RGBY(0xF9), RGBY(0xFA), RGBY(0xFB), + RGBY(0xFC), RGBY(0xFD), RGBY(0xFE), RGBY(0xFF), + + // Chroma U table. + RGBU(0x00), RGBU(0x01), RGBU(0x02), RGBU(0x03), + RGBU(0x04), RGBU(0x05), RGBU(0x06), RGBU(0x07), + RGBU(0x08), RGBU(0x09), RGBU(0x0A), RGBU(0x0B), + RGBU(0x0C), RGBU(0x0D), RGBU(0x0E), RGBU(0x0F), + RGBU(0x10), RGBU(0x11), RGBU(0x12), RGBU(0x13), + RGBU(0x14), RGBU(0x15), RGBU(0x16), RGBU(0x17), + RGBU(0x18), RGBU(0x19), RGBU(0x1A), RGBU(0x1B), + RGBU(0x1C), RGBU(0x1D), RGBU(0x1E), RGBU(0x1F), + RGBU(0x20), RGBU(0x21), RGBU(0x22), RGBU(0x23), + RGBU(0x24), RGBU(0x25), RGBU(0x26), RGBU(0x27), + RGBU(0x28), RGBU(0x29), RGBU(0x2A), RGBU(0x2B), + RGBU(0x2C), RGBU(0x2D), RGBU(0x2E), RGBU(0x2F), + RGBU(0x30), RGBU(0x31), RGBU(0x32), RGBU(0x33), + RGBU(0x34), RGBU(0x35), RGBU(0x36), RGBU(0x37), + RGBU(0x38), RGBU(0x39), RGBU(0x3A), RGBU(0x3B), + RGBU(0x3C), RGBU(0x3D), RGBU(0x3E), RGBU(0x3F), + RGBU(0x40), RGBU(0x41), RGBU(0x42), RGBU(0x43), + RGBU(0x44), RGBU(0x45), RGBU(0x46), RGBU(0x47), + RGBU(0x48), RGBU(0x49), RGBU(0x4A), RGBU(0x4B), + RGBU(0x4C), RGBU(0x4D), RGBU(0x4E), RGBU(0x4F), + RGBU(0x50), RGBU(0x51), RGBU(0x52), RGBU(0x53), + RGBU(0x54), RGBU(0x55), RGBU(0x56), RGBU(0x57), + RGBU(0x58), RGBU(0x59), RGBU(0x5A), RGBU(0x5B), + RGBU(0x5C), RGBU(0x5D), RGBU(0x5E), RGBU(0x5F), + RGBU(0x60), RGBU(0x61), RGBU(0x62), RGBU(0x63), + RGBU(0x64), RGBU(0x65), RGBU(0x66), RGBU(0x67), + RGBU(0x68), RGBU(0x69), RGBU(0x6A), RGBU(0x6B), + RGBU(0x6C), RGBU(0x6D), RGBU(0x6E), RGBU(0x6F), + RGBU(0x70), RGBU(0x71), RGBU(0x72), RGBU(0x73), + RGBU(0x74), RGBU(0x75), RGBU(0x76), RGBU(0x77), + RGBU(0x78), RGBU(0x79), RGBU(0x7A), RGBU(0x7B), + RGBU(0x7C), RGBU(0x7D), RGBU(0x7E), RGBU(0x7F), + RGBU(0x80), RGBU(0x81), RGBU(0x82), RGBU(0x83), + RGBU(0x84), RGBU(0x85), RGBU(0x86), RGBU(0x87), + RGBU(0x88), RGBU(0x89), RGBU(0x8A), RGBU(0x8B), + RGBU(0x8C), RGBU(0x8D), RGBU(0x8E), RGBU(0x8F), + RGBU(0x90), RGBU(0x91), RGBU(0x92), RGBU(0x93), + RGBU(0x94), RGBU(0x95), RGBU(0x96), RGBU(0x97), + RGBU(0x98), RGBU(0x99), RGBU(0x9A), RGBU(0x9B), + RGBU(0x9C), RGBU(0x9D), RGBU(0x9E), RGBU(0x9F), + RGBU(0xA0), RGBU(0xA1), RGBU(0xA2), RGBU(0xA3), + RGBU(0xA4), RGBU(0xA5), RGBU(0xA6), RGBU(0xA7), + RGBU(0xA8), RGBU(0xA9), RGBU(0xAA), RGBU(0xAB), + RGBU(0xAC), RGBU(0xAD), RGBU(0xAE), RGBU(0xAF), + RGBU(0xB0), RGBU(0xB1), RGBU(0xB2), RGBU(0xB3), + RGBU(0xB4), RGBU(0xB5), RGBU(0xB6), RGBU(0xB7), + RGBU(0xB8), RGBU(0xB9), RGBU(0xBA), RGBU(0xBB), + RGBU(0xBC), RGBU(0xBD), RGBU(0xBE), RGBU(0xBF), + RGBU(0xC0), RGBU(0xC1), RGBU(0xC2), RGBU(0xC3), + RGBU(0xC4), RGBU(0xC5), RGBU(0xC6), RGBU(0xC7), + RGBU(0xC8), RGBU(0xC9), RGBU(0xCA), RGBU(0xCB), + RGBU(0xCC), RGBU(0xCD), RGBU(0xCE), RGBU(0xCF), + RGBU(0xD0), RGBU(0xD1), RGBU(0xD2), RGBU(0xD3), + RGBU(0xD4), RGBU(0xD5), RGBU(0xD6), RGBU(0xD7), + RGBU(0xD8), RGBU(0xD9), RGBU(0xDA), RGBU(0xDB), + RGBU(0xDC), RGBU(0xDD), RGBU(0xDE), RGBU(0xDF), + RGBU(0xE0), RGBU(0xE1), RGBU(0xE2), RGBU(0xE3), + RGBU(0xE4), RGBU(0xE5), RGBU(0xE6), RGBU(0xE7), + RGBU(0xE8), RGBU(0xE9), RGBU(0xEA), RGBU(0xEB), + RGBU(0xEC), RGBU(0xED), RGBU(0xEE), RGBU(0xEF), + RGBU(0xF0), RGBU(0xF1), RGBU(0xF2), RGBU(0xF3), + RGBU(0xF4), RGBU(0xF5), RGBU(0xF6), RGBU(0xF7), + RGBU(0xF8), RGBU(0xF9), RGBU(0xFA), RGBU(0xFB), + RGBU(0xFC), RGBU(0xFD), RGBU(0xFE), RGBU(0xFF), + + // Chroma V table. + RGBV(0x00), RGBV(0x01), RGBV(0x02), RGBV(0x03), + RGBV(0x04), RGBV(0x05), RGBV(0x06), RGBV(0x07), + RGBV(0x08), RGBV(0x09), RGBV(0x0A), RGBV(0x0B), + RGBV(0x0C), RGBV(0x0D), RGBV(0x0E), RGBV(0x0F), + RGBV(0x10), RGBV(0x11), RGBV(0x12), RGBV(0x13), + RGBV(0x14), RGBV(0x15), RGBV(0x16), RGBV(0x17), + RGBV(0x18), RGBV(0x19), RGBV(0x1A), RGBV(0x1B), + RGBV(0x1C), RGBV(0x1D), RGBV(0x1E), RGBV(0x1F), + RGBV(0x20), RGBV(0x21), RGBV(0x22), RGBV(0x23), + RGBV(0x24), RGBV(0x25), RGBV(0x26), RGBV(0x27), + RGBV(0x28), RGBV(0x29), RGBV(0x2A), RGBV(0x2B), + RGBV(0x2C), RGBV(0x2D), RGBV(0x2E), RGBV(0x2F), + RGBV(0x30), RGBV(0x31), RGBV(0x32), RGBV(0x33), + RGBV(0x34), RGBV(0x35), RGBV(0x36), RGBV(0x37), + RGBV(0x38), RGBV(0x39), RGBV(0x3A), RGBV(0x3B), + RGBV(0x3C), RGBV(0x3D), RGBV(0x3E), RGBV(0x3F), + RGBV(0x40), RGBV(0x41), RGBV(0x42), RGBV(0x43), + RGBV(0x44), RGBV(0x45), RGBV(0x46), RGBV(0x47), + RGBV(0x48), RGBV(0x49), RGBV(0x4A), RGBV(0x4B), + RGBV(0x4C), RGBV(0x4D), RGBV(0x4E), RGBV(0x4F), + RGBV(0x50), RGBV(0x51), RGBV(0x52), RGBV(0x53), + RGBV(0x54), RGBV(0x55), RGBV(0x56), RGBV(0x57), + RGBV(0x58), RGBV(0x59), RGBV(0x5A), RGBV(0x5B), + RGBV(0x5C), RGBV(0x5D), RGBV(0x5E), RGBV(0x5F), + RGBV(0x60), RGBV(0x61), RGBV(0x62), RGBV(0x63), + RGBV(0x64), RGBV(0x65), RGBV(0x66), RGBV(0x67), + RGBV(0x68), RGBV(0x69), RGBV(0x6A), RGBV(0x6B), + RGBV(0x6C), RGBV(0x6D), RGBV(0x6E), RGBV(0x6F), + RGBV(0x70), RGBV(0x71), RGBV(0x72), RGBV(0x73), + RGBV(0x74), RGBV(0x75), RGBV(0x76), RGBV(0x77), + RGBV(0x78), RGBV(0x79), RGBV(0x7A), RGBV(0x7B), + RGBV(0x7C), RGBV(0x7D), RGBV(0x7E), RGBV(0x7F), + RGBV(0x80), RGBV(0x81), RGBV(0x82), RGBV(0x83), + RGBV(0x84), RGBV(0x85), RGBV(0x86), RGBV(0x87), + RGBV(0x88), RGBV(0x89), RGBV(0x8A), RGBV(0x8B), + RGBV(0x8C), RGBV(0x8D), RGBV(0x8E), RGBV(0x8F), + RGBV(0x90), RGBV(0x91), RGBV(0x92), RGBV(0x93), + RGBV(0x94), RGBV(0x95), RGBV(0x96), RGBV(0x97), + RGBV(0x98), RGBV(0x99), RGBV(0x9A), RGBV(0x9B), + RGBV(0x9C), RGBV(0x9D), RGBV(0x9E), RGBV(0x9F), + RGBV(0xA0), RGBV(0xA1), RGBV(0xA2), RGBV(0xA3), + RGBV(0xA4), RGBV(0xA5), RGBV(0xA6), RGBV(0xA7), + RGBV(0xA8), RGBV(0xA9), RGBV(0xAA), RGBV(0xAB), + RGBV(0xAC), RGBV(0xAD), RGBV(0xAE), RGBV(0xAF), + RGBV(0xB0), RGBV(0xB1), RGBV(0xB2), RGBV(0xB3), + RGBV(0xB4), RGBV(0xB5), RGBV(0xB6), RGBV(0xB7), + RGBV(0xB8), RGBV(0xB9), RGBV(0xBA), RGBV(0xBB), + RGBV(0xBC), RGBV(0xBD), RGBV(0xBE), RGBV(0xBF), + RGBV(0xC0), RGBV(0xC1), RGBV(0xC2), RGBV(0xC3), + RGBV(0xC4), RGBV(0xC5), RGBV(0xC6), RGBV(0xC7), + RGBV(0xC8), RGBV(0xC9), RGBV(0xCA), RGBV(0xCB), + RGBV(0xCC), RGBV(0xCD), RGBV(0xCE), RGBV(0xCF), + RGBV(0xD0), RGBV(0xD1), RGBV(0xD2), RGBV(0xD3), + RGBV(0xD4), RGBV(0xD5), RGBV(0xD6), RGBV(0xD7), + RGBV(0xD8), RGBV(0xD9), RGBV(0xDA), RGBV(0xDB), + RGBV(0xDC), RGBV(0xDD), RGBV(0xDE), RGBV(0xDF), + RGBV(0xE0), RGBV(0xE1), RGBV(0xE2), RGBV(0xE3), + RGBV(0xE4), RGBV(0xE5), RGBV(0xE6), RGBV(0xE7), + RGBV(0xE8), RGBV(0xE9), RGBV(0xEA), RGBV(0xEB), + RGBV(0xEC), RGBV(0xED), RGBV(0xEE), RGBV(0xEF), + RGBV(0xF0), RGBV(0xF1), RGBV(0xF2), RGBV(0xF3), + RGBV(0xF4), RGBV(0xF5), RGBV(0xF6), RGBV(0xF7), + RGBV(0xF8), RGBV(0xF9), RGBV(0xFA), RGBV(0xFB), + RGBV(0xFC), RGBV(0xFD), RGBV(0xFE), RGBV(0xFF), +}; + +#undef RGBY +#undef RGBU +#undef RGBV + +} // extern "C"
--- a/gfx/ycbcr/yuv_row_win.cpp +++ b/gfx/ycbcr/yuv_row_win.cpp @@ -1,273 +1,23 @@ -// Copyright (c) 2009 The Chromium Authors. All rights reserved. +// 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 "yuv_row.h" - -extern "C" { -// 64 Bit builds use the C fallback. Optimized code -// needs to be fixed for 64 bit builds. -#if defined(ARCH_CPU_64_BITS) -// PPC implementation uses C fallback -void FastConvertYUVToRGB32Row(const uint8* y_buf, - const uint8* u_buf, - const uint8* v_buf, - uint8* rgb_buf, - int width) { - FastConvertYUVToRGB32Row_C(y_buf, u_buf, v_buf, rgb_buf, width, 1); -} - -void ScaleYUVToRGB32Row(const uint8* y_buf, - const uint8* u_buf, - const uint8* v_buf, - uint8* rgb_buf, - int width, - int scaled_dx) { - ScaleYUVToRGB32Row_C(y_buf, u_buf, v_buf, rgb_buf, width, scaled_dx, 1); -} - -#else +#define MOZILLA_SSE_INCLUDE_HEADER_FOR_SSE2 +#define MOZILLA_SSE_INCLUDE_HEADER_FOR_MMX +#include "mozilla/SSE.h" -#define RGBY(i) { \ - static_cast<int16>(1.164 * 64 * (i - 16) + 0.5), \ - static_cast<int16>(1.164 * 64 * (i - 16) + 0.5), \ - static_cast<int16>(1.164 * 64 * (i - 16) + 0.5), \ - 0 \ -} - -#define RGBU(i) { \ - static_cast<int16>(2.018 * 64 * (i - 128) + 0.5), \ - static_cast<int16>(-0.391 * 64 * (i - 128) + 0.5), \ - 0, \ - static_cast<int16>(256 * 64 - 1) \ -} - -#define RGBV(i) { \ - 0, \ - static_cast<int16>(-0.813 * 64 * (i - 128) + 0.5), \ - static_cast<int16>(1.596 * 64 * (i - 128) + 0.5), \ - 0 \ -} - -#define MMX_ALIGNED(var) __declspec(align(16)) var - -MMX_ALIGNED(int16 kCoefficientsRgbY[256][4]) = { - RGBY(0x00), RGBY(0x01), RGBY(0x02), RGBY(0x03), - RGBY(0x04), RGBY(0x05), RGBY(0x06), RGBY(0x07), - RGBY(0x08), RGBY(0x09), RGBY(0x0A), RGBY(0x0B), - RGBY(0x0C), RGBY(0x0D), RGBY(0x0E), RGBY(0x0F), - RGBY(0x10), RGBY(0x11), RGBY(0x12), RGBY(0x13), - RGBY(0x14), RGBY(0x15), RGBY(0x16), RGBY(0x17), - RGBY(0x18), RGBY(0x19), RGBY(0x1A), RGBY(0x1B), - RGBY(0x1C), RGBY(0x1D), RGBY(0x1E), RGBY(0x1F), - RGBY(0x20), RGBY(0x21), RGBY(0x22), RGBY(0x23), - RGBY(0x24), RGBY(0x25), RGBY(0x26), RGBY(0x27), - RGBY(0x28), RGBY(0x29), RGBY(0x2A), RGBY(0x2B), - RGBY(0x2C), RGBY(0x2D), RGBY(0x2E), RGBY(0x2F), - RGBY(0x30), RGBY(0x31), RGBY(0x32), RGBY(0x33), - RGBY(0x34), RGBY(0x35), RGBY(0x36), RGBY(0x37), - RGBY(0x38), RGBY(0x39), RGBY(0x3A), RGBY(0x3B), - RGBY(0x3C), RGBY(0x3D), RGBY(0x3E), RGBY(0x3F), - RGBY(0x40), RGBY(0x41), RGBY(0x42), RGBY(0x43), - RGBY(0x44), RGBY(0x45), RGBY(0x46), RGBY(0x47), - RGBY(0x48), RGBY(0x49), RGBY(0x4A), RGBY(0x4B), - RGBY(0x4C), RGBY(0x4D), RGBY(0x4E), RGBY(0x4F), - RGBY(0x50), RGBY(0x51), RGBY(0x52), RGBY(0x53), - RGBY(0x54), RGBY(0x55), RGBY(0x56), RGBY(0x57), - RGBY(0x58), RGBY(0x59), RGBY(0x5A), RGBY(0x5B), - RGBY(0x5C), RGBY(0x5D), RGBY(0x5E), RGBY(0x5F), - RGBY(0x60), RGBY(0x61), RGBY(0x62), RGBY(0x63), - RGBY(0x64), RGBY(0x65), RGBY(0x66), RGBY(0x67), - RGBY(0x68), RGBY(0x69), RGBY(0x6A), RGBY(0x6B), - RGBY(0x6C), RGBY(0x6D), RGBY(0x6E), RGBY(0x6F), - RGBY(0x70), RGBY(0x71), RGBY(0x72), RGBY(0x73), - RGBY(0x74), RGBY(0x75), RGBY(0x76), RGBY(0x77), - RGBY(0x78), RGBY(0x79), RGBY(0x7A), RGBY(0x7B), - RGBY(0x7C), RGBY(0x7D), RGBY(0x7E), RGBY(0x7F), - RGBY(0x80), RGBY(0x81), RGBY(0x82), RGBY(0x83), - RGBY(0x84), RGBY(0x85), RGBY(0x86), RGBY(0x87), - RGBY(0x88), RGBY(0x89), RGBY(0x8A), RGBY(0x8B), - RGBY(0x8C), RGBY(0x8D), RGBY(0x8E), RGBY(0x8F), - RGBY(0x90), RGBY(0x91), RGBY(0x92), RGBY(0x93), - RGBY(0x94), RGBY(0x95), RGBY(0x96), RGBY(0x97), - RGBY(0x98), RGBY(0x99), RGBY(0x9A), RGBY(0x9B), - RGBY(0x9C), RGBY(0x9D), RGBY(0x9E), RGBY(0x9F), - RGBY(0xA0), RGBY(0xA1), RGBY(0xA2), RGBY(0xA3), - RGBY(0xA4), RGBY(0xA5), RGBY(0xA6), RGBY(0xA7), - RGBY(0xA8), RGBY(0xA9), RGBY(0xAA), RGBY(0xAB), - RGBY(0xAC), RGBY(0xAD), RGBY(0xAE), RGBY(0xAF), - RGBY(0xB0), RGBY(0xB1), RGBY(0xB2), RGBY(0xB3), - RGBY(0xB4), RGBY(0xB5), RGBY(0xB6), RGBY(0xB7), - RGBY(0xB8), RGBY(0xB9), RGBY(0xBA), RGBY(0xBB), - RGBY(0xBC), RGBY(0xBD), RGBY(0xBE), RGBY(0xBF), - RGBY(0xC0), RGBY(0xC1), RGBY(0xC2), RGBY(0xC3), - RGBY(0xC4), RGBY(0xC5), RGBY(0xC6), RGBY(0xC7), - RGBY(0xC8), RGBY(0xC9), RGBY(0xCA), RGBY(0xCB), - RGBY(0xCC), RGBY(0xCD), RGBY(0xCE), RGBY(0xCF), - RGBY(0xD0), RGBY(0xD1), RGBY(0xD2), RGBY(0xD3), - RGBY(0xD4), RGBY(0xD5), RGBY(0xD6), RGBY(0xD7), - RGBY(0xD8), RGBY(0xD9), RGBY(0xDA), RGBY(0xDB), - RGBY(0xDC), RGBY(0xDD), RGBY(0xDE), RGBY(0xDF), - RGBY(0xE0), RGBY(0xE1), RGBY(0xE2), RGBY(0xE3), - RGBY(0xE4), RGBY(0xE5), RGBY(0xE6), RGBY(0xE7), - RGBY(0xE8), RGBY(0xE9), RGBY(0xEA), RGBY(0xEB), - RGBY(0xEC), RGBY(0xED), RGBY(0xEE), RGBY(0xEF), - RGBY(0xF0), RGBY(0xF1), RGBY(0xF2), RGBY(0xF3), - RGBY(0xF4), RGBY(0xF5), RGBY(0xF6), RGBY(0xF7), - RGBY(0xF8), RGBY(0xF9), RGBY(0xFA), RGBY(0xFB), - RGBY(0xFC), RGBY(0xFD), RGBY(0xFE), RGBY(0xFF), -}; +#define kCoefficientsRgbU kCoefficientsRgbY + 2048 +#define kCoefficientsRgbV kCoefficientsRgbY + 4096 -MMX_ALIGNED(int16 kCoefficientsRgbU[256][4]) = { - RGBU(0x00), RGBU(0x01), RGBU(0x02), RGBU(0x03), - RGBU(0x04), RGBU(0x05), RGBU(0x06), RGBU(0x07), - RGBU(0x08), RGBU(0x09), RGBU(0x0A), RGBU(0x0B), - RGBU(0x0C), RGBU(0x0D), RGBU(0x0E), RGBU(0x0F), - RGBU(0x10), RGBU(0x11), RGBU(0x12), RGBU(0x13), - RGBU(0x14), RGBU(0x15), RGBU(0x16), RGBU(0x17), - RGBU(0x18), RGBU(0x19), RGBU(0x1A), RGBU(0x1B), - RGBU(0x1C), RGBU(0x1D), RGBU(0x1E), RGBU(0x1F), - RGBU(0x20), RGBU(0x21), RGBU(0x22), RGBU(0x23), - RGBU(0x24), RGBU(0x25), RGBU(0x26), RGBU(0x27), - RGBU(0x28), RGBU(0x29), RGBU(0x2A), RGBU(0x2B), - RGBU(0x2C), RGBU(0x2D), RGBU(0x2E), RGBU(0x2F), - RGBU(0x30), RGBU(0x31), RGBU(0x32), RGBU(0x33), - RGBU(0x34), RGBU(0x35), RGBU(0x36), RGBU(0x37), - RGBU(0x38), RGBU(0x39), RGBU(0x3A), RGBU(0x3B), - RGBU(0x3C), RGBU(0x3D), RGBU(0x3E), RGBU(0x3F), - RGBU(0x40), RGBU(0x41), RGBU(0x42), RGBU(0x43), - RGBU(0x44), RGBU(0x45), RGBU(0x46), RGBU(0x47), - RGBU(0x48), RGBU(0x49), RGBU(0x4A), RGBU(0x4B), - RGBU(0x4C), RGBU(0x4D), RGBU(0x4E), RGBU(0x4F), - RGBU(0x50), RGBU(0x51), RGBU(0x52), RGBU(0x53), - RGBU(0x54), RGBU(0x55), RGBU(0x56), RGBU(0x57), - RGBU(0x58), RGBU(0x59), RGBU(0x5A), RGBU(0x5B), - RGBU(0x5C), RGBU(0x5D), RGBU(0x5E), RGBU(0x5F), - RGBU(0x60), RGBU(0x61), RGBU(0x62), RGBU(0x63), - RGBU(0x64), RGBU(0x65), RGBU(0x66), RGBU(0x67), - RGBU(0x68), RGBU(0x69), RGBU(0x6A), RGBU(0x6B), - RGBU(0x6C), RGBU(0x6D), RGBU(0x6E), RGBU(0x6F), - RGBU(0x70), RGBU(0x71), RGBU(0x72), RGBU(0x73), - RGBU(0x74), RGBU(0x75), RGBU(0x76), RGBU(0x77), - RGBU(0x78), RGBU(0x79), RGBU(0x7A), RGBU(0x7B), - RGBU(0x7C), RGBU(0x7D), RGBU(0x7E), RGBU(0x7F), - RGBU(0x80), RGBU(0x81), RGBU(0x82), RGBU(0x83), - RGBU(0x84), RGBU(0x85), RGBU(0x86), RGBU(0x87), - RGBU(0x88), RGBU(0x89), RGBU(0x8A), RGBU(0x8B), - RGBU(0x8C), RGBU(0x8D), RGBU(0x8E), RGBU(0x8F), - RGBU(0x90), RGBU(0x91), RGBU(0x92), RGBU(0x93), - RGBU(0x94), RGBU(0x95), RGBU(0x96), RGBU(0x97), - RGBU(0x98), RGBU(0x99), RGBU(0x9A), RGBU(0x9B), - RGBU(0x9C), RGBU(0x9D), RGBU(0x9E), RGBU(0x9F), - RGBU(0xA0), RGBU(0xA1), RGBU(0xA2), RGBU(0xA3), - RGBU(0xA4), RGBU(0xA5), RGBU(0xA6), RGBU(0xA7), - RGBU(0xA8), RGBU(0xA9), RGBU(0xAA), RGBU(0xAB), - RGBU(0xAC), RGBU(0xAD), RGBU(0xAE), RGBU(0xAF), - RGBU(0xB0), RGBU(0xB1), RGBU(0xB2), RGBU(0xB3), - RGBU(0xB4), RGBU(0xB5), RGBU(0xB6), RGBU(0xB7), - RGBU(0xB8), RGBU(0xB9), RGBU(0xBA), RGBU(0xBB), - RGBU(0xBC), RGBU(0xBD), RGBU(0xBE), RGBU(0xBF), - RGBU(0xC0), RGBU(0xC1), RGBU(0xC2), RGBU(0xC3), - RGBU(0xC4), RGBU(0xC5), RGBU(0xC6), RGBU(0xC7), - RGBU(0xC8), RGBU(0xC9), RGBU(0xCA), RGBU(0xCB), - RGBU(0xCC), RGBU(0xCD), RGBU(0xCE), RGBU(0xCF), - RGBU(0xD0), RGBU(0xD1), RGBU(0xD2), RGBU(0xD3), - RGBU(0xD4), RGBU(0xD5), RGBU(0xD6), RGBU(0xD7), - RGBU(0xD8), RGBU(0xD9), RGBU(0xDA), RGBU(0xDB), - RGBU(0xDC), RGBU(0xDD), RGBU(0xDE), RGBU(0xDF), - RGBU(0xE0), RGBU(0xE1), RGBU(0xE2), RGBU(0xE3), - RGBU(0xE4), RGBU(0xE5), RGBU(0xE6), RGBU(0xE7), - RGBU(0xE8), RGBU(0xE9), RGBU(0xEA), RGBU(0xEB), - RGBU(0xEC), RGBU(0xED), RGBU(0xEE), RGBU(0xEF), - RGBU(0xF0), RGBU(0xF1), RGBU(0xF2), RGBU(0xF3), - RGBU(0xF4), RGBU(0xF5), RGBU(0xF6), RGBU(0xF7), - RGBU(0xF8), RGBU(0xF9), RGBU(0xFA), RGBU(0xFB), - RGBU(0xFC), RGBU(0xFD), RGBU(0xFE), RGBU(0xFF), -}; - -MMX_ALIGNED(int16 kCoefficientsRgbV[256][4]) = { - RGBV(0x00), RGBV(0x01), RGBV(0x02), RGBV(0x03), - RGBV(0x04), RGBV(0x05), RGBV(0x06), RGBV(0x07), - RGBV(0x08), RGBV(0x09), RGBV(0x0A), RGBV(0x0B), - RGBV(0x0C), RGBV(0x0D), RGBV(0x0E), RGBV(0x0F), - RGBV(0x10), RGBV(0x11), RGBV(0x12), RGBV(0x13), - RGBV(0x14), RGBV(0x15), RGBV(0x16), RGBV(0x17), - RGBV(0x18), RGBV(0x19), RGBV(0x1A), RGBV(0x1B), - RGBV(0x1C), RGBV(0x1D), RGBV(0x1E), RGBV(0x1F), - RGBV(0x20), RGBV(0x21), RGBV(0x22), RGBV(0x23), - RGBV(0x24), RGBV(0x25), RGBV(0x26), RGBV(0x27), - RGBV(0x28), RGBV(0x29), RGBV(0x2A), RGBV(0x2B), - RGBV(0x2C), RGBV(0x2D), RGBV(0x2E), RGBV(0x2F), - RGBV(0x30), RGBV(0x31), RGBV(0x32), RGBV(0x33), - RGBV(0x34), RGBV(0x35), RGBV(0x36), RGBV(0x37), - RGBV(0x38), RGBV(0x39), RGBV(0x3A), RGBV(0x3B), - RGBV(0x3C), RGBV(0x3D), RGBV(0x3E), RGBV(0x3F), - RGBV(0x40), RGBV(0x41), RGBV(0x42), RGBV(0x43), - RGBV(0x44), RGBV(0x45), RGBV(0x46), RGBV(0x47), - RGBV(0x48), RGBV(0x49), RGBV(0x4A), RGBV(0x4B), - RGBV(0x4C), RGBV(0x4D), RGBV(0x4E), RGBV(0x4F), - RGBV(0x50), RGBV(0x51), RGBV(0x52), RGBV(0x53), - RGBV(0x54), RGBV(0x55), RGBV(0x56), RGBV(0x57), - RGBV(0x58), RGBV(0x59), RGBV(0x5A), RGBV(0x5B), - RGBV(0x5C), RGBV(0x5D), RGBV(0x5E), RGBV(0x5F), - RGBV(0x60), RGBV(0x61), RGBV(0x62), RGBV(0x63), - RGBV(0x64), RGBV(0x65), RGBV(0x66), RGBV(0x67), - RGBV(0x68), RGBV(0x69), RGBV(0x6A), RGBV(0x6B), - RGBV(0x6C), RGBV(0x6D), RGBV(0x6E), RGBV(0x6F), - RGBV(0x70), RGBV(0x71), RGBV(0x72), RGBV(0x73), - RGBV(0x74), RGBV(0x75), RGBV(0x76), RGBV(0x77), - RGBV(0x78), RGBV(0x79), RGBV(0x7A), RGBV(0x7B), - RGBV(0x7C), RGBV(0x7D), RGBV(0x7E), RGBV(0x7F), - RGBV(0x80), RGBV(0x81), RGBV(0x82), RGBV(0x83), - RGBV(0x84), RGBV(0x85), RGBV(0x86), RGBV(0x87), - RGBV(0x88), RGBV(0x89), RGBV(0x8A), RGBV(0x8B), - RGBV(0x8C), RGBV(0x8D), RGBV(0x8E), RGBV(0x8F), - RGBV(0x90), RGBV(0x91), RGBV(0x92), RGBV(0x93), - RGBV(0x94), RGBV(0x95), RGBV(0x96), RGBV(0x97), - RGBV(0x98), RGBV(0x99), RGBV(0x9A), RGBV(0x9B), - RGBV(0x9C), RGBV(0x9D), RGBV(0x9E), RGBV(0x9F), - RGBV(0xA0), RGBV(0xA1), RGBV(0xA2), RGBV(0xA3), - RGBV(0xA4), RGBV(0xA5), RGBV(0xA6), RGBV(0xA7), - RGBV(0xA8), RGBV(0xA9), RGBV(0xAA), RGBV(0xAB), - RGBV(0xAC), RGBV(0xAD), RGBV(0xAE), RGBV(0xAF), - RGBV(0xB0), RGBV(0xB1), RGBV(0xB2), RGBV(0xB3), - RGBV(0xB4), RGBV(0xB5), RGBV(0xB6), RGBV(0xB7), - RGBV(0xB8), RGBV(0xB9), RGBV(0xBA), RGBV(0xBB), - RGBV(0xBC), RGBV(0xBD), RGBV(0xBE), RGBV(0xBF), - RGBV(0xC0), RGBV(0xC1), RGBV(0xC2), RGBV(0xC3), - RGBV(0xC4), RGBV(0xC5), RGBV(0xC6), RGBV(0xC7), - RGBV(0xC8), RGBV(0xC9), RGBV(0xCA), RGBV(0xCB), - RGBV(0xCC), RGBV(0xCD), RGBV(0xCE), RGBV(0xCF), - RGBV(0xD0), RGBV(0xD1), RGBV(0xD2), RGBV(0xD3), - RGBV(0xD4), RGBV(0xD5), RGBV(0xD6), RGBV(0xD7), - RGBV(0xD8), RGBV(0xD9), RGBV(0xDA), RGBV(0xDB), - RGBV(0xDC), RGBV(0xDD), RGBV(0xDE), RGBV(0xDF), - RGBV(0xE0), RGBV(0xE1), RGBV(0xE2), RGBV(0xE3), - RGBV(0xE4), RGBV(0xE5), RGBV(0xE6), RGBV(0xE7), - RGBV(0xE8), RGBV(0xE9), RGBV(0xEA), RGBV(0xEB), - RGBV(0xEC), RGBV(0xED), RGBV(0xEE), RGBV(0xEF), - RGBV(0xF0), RGBV(0xF1), RGBV(0xF2), RGBV(0xF3), - RGBV(0xF4), RGBV(0xF5), RGBV(0xF6), RGBV(0xF7), - RGBV(0xF8), RGBV(0xF9), RGBV(0xFA), RGBV(0xFB), - RGBV(0xFC), RGBV(0xFD), RGBV(0xFE), RGBV(0xFF), -}; - -#undef RGBHY -#undef RGBY -#undef RGBU -#undef RGBV -#undef MMX_ALIGNED - -// Warning C4799: function has no EMMS instruction. -// EMMS() is slow and should be called by the calling function once per image. -#pragma warning(disable: 4799) - +extern "C" { +#if defined(MOZILLA_COMPILE_WITH_SSE2) __declspec(naked) void FastConvertYUVToRGB32Row(const uint8* y_buf, const uint8* u_buf, const uint8* v_buf, uint8* rgb_buf, int width) { __asm { pushad @@ -518,44 +268,44 @@ void DoubleYUVToRGB32Row(const uint8* y_ // For performance the chroma is under sampled, reducing cost of a 3x // 1080p scale from 8.4 ms to 5.4 ms. __declspec(naked) void ScaleYUVToRGB32Row(const uint8* y_buf, const uint8* u_buf, const uint8* v_buf, uint8* rgb_buf, int width, - int dx) { + int source_dx) { __asm { pushad mov edx, [esp + 32 + 4] // Y mov edi, [esp + 32 + 8] // U mov esi, [esp + 32 + 12] // V mov ebp, [esp + 32 + 16] // rgb mov ecx, [esp + 32 + 20] // width xor ebx, ebx // x jmp scaleend scaleloop : mov eax, ebx - sar eax, 5 + sar eax, 17 movzx eax, byte ptr [edi + eax] movq mm0, [kCoefficientsRgbU + 8 * eax] mov eax, ebx - sar eax, 5 + sar eax, 17 movzx eax, byte ptr [esi + eax] paddsw mm0, [kCoefficientsRgbV + 8 * eax] mov eax, ebx - add ebx, [esp + 32 + 24] // x += dx - sar eax, 4 + add ebx, [esp + 32 + 24] // x += source_dx + sar eax, 16 movzx eax, byte ptr [edx + eax] movq mm1, [kCoefficientsRgbY + 8 * eax] mov eax, ebx - add ebx, [esp + 32 + 24] // x += dx - sar eax, 4 + 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 movntq [ebp], mm1 @@ -563,33 +313,165 @@ void ScaleYUVToRGB32Row(const uint8* y_b scaleend : sub ecx, 2 jns scaleloop and ecx, 1 // odd number of pixels? jz scaledone mov eax, ebx - sar eax, 5 + sar eax, 17 movzx eax, byte ptr [edi + eax] movq mm0, [kCoefficientsRgbU + 8 * eax] mov eax, ebx - sar eax, 5 + sar eax, 17 movzx eax, byte ptr [esi + eax] paddsw mm0, [kCoefficientsRgbV + 8 * eax] mov eax, ebx - sar eax, 4 + sar eax, 16 movzx eax, byte ptr [edx + eax] movq mm1, [kCoefficientsRgbY + 8 * eax] paddsw mm1, mm0 psraw mm1, 6 packuswb mm1, mm1 movd [ebp], mm1 scaledone : popad ret } } -#endif // ARCH_CPU_64_BITS +__declspec(naked) +void LinearScaleYUVToRGB32Row(const uint8* y_buf, + const uint8* u_buf, + const uint8* v_buf, + uint8* rgb_buf, + int width, + int source_dx) { + __asm { + pushad + mov edx, [esp + 32 + 4] // Y + mov edi, [esp + 32 + 8] // U + // [esp + 32 + 12] // V + mov ebp, [esp + 32 + 16] // rgb + mov ecx, [esp + 32 + 20] // width + imul ecx, [esp + 32 + 24] // source_dx + mov [esp + 32 + 20], ecx // source_width = width * source_dx + mov ecx, [esp + 32 + 24] // source_dx + xor ebx, ebx // x = 0 + cmp ecx, 0x20000 + jl lscaleend + mov ebx, 0x8000 // x = 0.5 for 1/2 or less + jmp lscaleend +lscaleloop: + mov eax, ebx + sar eax, 0x11 + + movzx ecx, byte ptr [edi + eax] + movzx esi, byte ptr [edi + eax + 1] + mov eax, ebx + and eax, 0x1fffe + imul esi, eax + xor eax, 0x1fffe + imul ecx, eax + add ecx, esi + shr ecx, 17 + movq mm0, [kCoefficientsRgbU + 8 * ecx] + + mov esi, [esp + 32 + 12] + mov eax, ebx + sar eax, 0x11 + + movzx ecx, byte ptr [esi + eax] + movzx esi, byte ptr [esi + eax + 1] + mov eax, ebx + and eax, 0x1fffe + imul esi, eax + xor eax, 0x1fffe + imul ecx, eax + add ecx, esi + shr ecx, 17 + paddsw mm0, [kCoefficientsRgbV + 8 * ecx] + + mov eax, ebx + sar eax, 0x10 + movzx ecx, byte ptr [edx + eax] + movzx esi, byte ptr [1 + edx + eax] + mov eax, ebx + add ebx, [esp + 32 + 24] + and eax, 0xffff + imul esi, eax + xor eax, 0xffff + imul ecx, eax + add ecx, esi + shr ecx, 16 + movq mm1, [kCoefficientsRgbY + 8 * ecx] + + cmp ebx, [esp + 32 + 20] + jge lscalelastpixel + + mov eax, ebx + sar eax, 0x10 + movzx ecx, byte ptr [edx + eax] + movzx esi, byte ptr [edx + eax + 1] + mov eax, ebx + add ebx, [esp + 32 + 24] + and eax, 0xffff + imul esi, eax + xor eax, 0xffff + 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 + movntq [ebp], mm1 + add ebp, 0x8 + +lscaleend: + cmp ebx, [esp + 32 + 20] + jl lscaleloop + popad + ret + +lscalelastpixel: + paddsw mm1, mm0 + psraw mm1, 6 + packuswb mm1, mm1 + movd [ebp], mm1 + popad + ret + }; +} +#else // MOZILLA_COMPILE_WITH_SSE2 +void FastConvertYUVToRGB32Row(const uint8* y_buf, + const uint8* u_buf, + const uint8* v_buf, + uint8* rgb_buf, + int width) { + FastConvertYUVToRGB32Row_C(y_buf, u_buf, v_buf, rgb_buf, width, 1); +} + +void ScaleYUVToRGB32Row(const uint8* y_buf, + const uint8* u_buf, + const uint8* v_buf, + uint8* rgb_buf, + int width, + int source_dx) { + ScaleYUVToRGB32Row_C(y_buf, u_buf, v_buf, rgb_buf, width, source_dx); +} + +void LinearScaleYUVToRGB32Row(const uint8* y_buf, + const uint8* u_buf, + const uint8* v_buf, + uint8* rgb_buf, + int width, + int source_dx) { + LinearScaleYUVToRGB32Row_C(y_buf, u_buf, v_buf, rgb_buf, width, source_dx); +} +#endif } // extern "C"
deleted file mode 100644 --- a/gfx/ycbcr/yv24.patch +++ /dev/null @@ -1,241 +0,0 @@ -diff --git a/gfx/ycbcr/yuv_convert.cpp b/gfx/ycbcr/yuv_convert.cpp -index b22e778..cdbb040 100644 ---- a/gfx/ycbcr/yuv_convert.cpp -+++ b/gfx/ycbcr/yuv_convert.cpp -@@ -6,16 +6,17 @@ - // http://www.fourcc.org/yuv.php - // The actual conversion is best described here - // http://en.wikipedia.org/wiki/YUV - // An article on optimizing YUV conversion using tables instead of multiplies - // http://lestourtereaux.free.fr/papers/data/yuvrgb.pdf - // - // YV12 is a full plane of Y and a half height, half width chroma planes - // YV16 is a full plane of Y and a full height, half width chroma planes -+// YV24 is a full plane of Y and a full height, full width chroma planes - // - // ARGB pixel format is output, which on little endian is stored as BGRA. - // The alpha is set to 255, allowing the application to use RGBA or RGB32. - - #include "yuv_convert.h" - - // Header for low level row functions. - #include "yuv_row.h" -@@ -33,50 +34,55 @@ void ConvertYCbCrToRGB32(const uint8* y_buf, - int pic_x, - int pic_y, - int pic_width, - int pic_height, - int y_pitch, - int uv_pitch, - int rgb_pitch, - YUVType yuv_type) { -- unsigned int y_shift = yuv_type; -- bool has_mmx = supports_mmx(); -- bool odd_pic_x = pic_x % 2 != 0; -+ unsigned int y_shift = yuv_type == YV12 ? 1 : 0; -+ unsigned int x_shift = yuv_type == YV24 ? 0 : 1; -+ // There is no optimized YV24 MMX routine so we check for this and -+ // fall back to the C code. -+ bool has_mmx = supports_mmx() && yuv_type != YV24; -+ bool odd_pic_x = yuv_type != YV24 && pic_x % 2 != 0; - int x_width = odd_pic_x ? pic_width - 1 : pic_width; - - for (int y = pic_y; y < pic_height + pic_y; ++y) { - uint8* rgb_row = rgb_buf + (y - pic_y) * rgb_pitch; - const uint8* y_ptr = y_buf + y * y_pitch + pic_x; -- const uint8* u_ptr = u_buf + (y >> y_shift) * uv_pitch + (pic_x >> 1); -- const uint8* v_ptr = v_buf + (y >> y_shift) * uv_pitch + (pic_x >> 1); -+ const uint8* u_ptr = u_buf + (y >> y_shift) * uv_pitch + (pic_x >> x_shift); -+ const uint8* v_ptr = v_buf + (y >> y_shift) * uv_pitch + (pic_x >> x_shift); - - if (odd_pic_x) { - // Handle the single odd pixel manually and use the - // fast routines for the remaining. - FastConvertYUVToRGB32Row_C(y_ptr++, - u_ptr++, - v_ptr++, - rgb_row, -- 1); -+ 1, -+ x_shift); - rgb_row += 4; - } - - if (has_mmx) - FastConvertYUVToRGB32Row(y_ptr, - u_ptr, - v_ptr, - rgb_row, - x_width); - else - FastConvertYUVToRGB32Row_C(y_ptr, - u_ptr, - v_ptr, - rgb_row, -- x_width); -+ x_width, -+ x_shift); - } - - // MMX used for FastConvertYUVToRGB32Row requires emms instruction. - if (has_mmx) - EMMS(); - } - - } // namespace gfx -diff --git a/gfx/ycbcr/yuv_convert.h b/gfx/ycbcr/yuv_convert.h -index e624168..c0b678d 100644 ---- a/gfx/ycbcr/yuv_convert.h -+++ b/gfx/ycbcr/yuv_convert.h -@@ -10,18 +10,19 @@ - - namespace mozilla { - - namespace gfx { - - // Type of YUV surface. - // The value of these enums matter as they are used to shift vertical indices. - enum YUVType { -- YV16 = 0, // YV16 is half width and full height chroma channels. -- YV12 = 1 // YV12 is half width and half height chroma channels. -+ YV12 = 0, // YV12 is half width and half height chroma channels. -+ YV16 = 1, // YV16 is half width and full height chroma channels. -+ YV24 = 2 // YV24 is full width and full height chroma channels. - }; - - // Convert a frame of YUV to 32 bit ARGB. - // Pass in YV16/YV12 depending on source format - NS_GFX_(void) ConvertYCbCrToRGB32(const uint8* yplane, - const uint8* uplane, - const uint8* vplane, - uint8* rgbframe, -diff --git a/gfx/ycbcr/yuv_row.h b/gfx/ycbcr/yuv_row.h -index 2a82972..d776dac 100644 ---- a/gfx/ycbcr/yuv_row.h -+++ b/gfx/ycbcr/yuv_row.h -@@ -20,17 +20,18 @@ void FastConvertYUVToRGB32Row(const uint8* y_buf, - const uint8* v_buf, - uint8* rgb_buf, - int width); - - void FastConvertYUVToRGB32Row_C(const uint8* y_buf, - const uint8* u_buf, - const uint8* v_buf, - uint8* rgb_buf, -- int width); -+ int width, -+ unsigned int x_shift); - - - } // extern "C" - - // x64 uses MMX2 (SSE) so emms is not required. - #if !defined(ARCH_CPU_X86_64) && !defined(ARCH_CPU_PPC) - #if defined(_MSC_VER) - #define EMMS() __asm emms -diff --git a/gfx/ycbcr/yuv_row_c.cpp b/gfx/ycbcr/yuv_row_c.cpp -index d3bdab4..36d9bda 100644 ---- a/gfx/ycbcr/yuv_row_c.cpp -+++ b/gfx/ycbcr/yuv_row_c.cpp -@@ -153,24 +153,29 @@ static inline void YuvPixel(uint8 y, - (clip(C298a + cr) << 16) | - (0xff000000); - } - - void FastConvertYUVToRGB32Row_C(const uint8* y_buf, - const uint8* u_buf, - const uint8* v_buf, - uint8* rgb_buf, -- int width) { -+ int width, -+ unsigned int x_shift) { - for (int x = 0; x < width; x += 2) { -- uint8 u = u_buf[x >> 1]; -- uint8 v = v_buf[x >> 1]; -+ uint8 u = u_buf[x >> x_shift]; -+ uint8 v = v_buf[x >> x_shift]; - uint8 y0 = y_buf[x]; - YuvPixel(y0, u, v, rgb_buf); - if ((x + 1) < width) { - uint8 y1 = y_buf[x + 1]; -+ if (x_shift == 0) { -+ u = u_buf[x + 1]; -+ v = v_buf[x + 1]; -+ } - YuvPixel(y1, u, v, rgb_buf + 4); - } - rgb_buf += 8; // Advance 2 pixels. - } - } - - } // extern "C" - -diff --git a/gfx/ycbcr/yuv_row_linux.cpp b/gfx/ycbcr/yuv_row_linux.cpp -index ce5ee89..455dd7b 100644 ---- a/gfx/ycbcr/yuv_row_linux.cpp -+++ b/gfx/ycbcr/yuv_row_linux.cpp -@@ -13,17 +13,17 @@ extern "C" { - - #if defined(ARCH_CPU_ARM_FAMILY) - // ARM implementation uses C fallback - void FastConvertYUVToRGB32Row(const uint8* y_buf, - const uint8* u_buf, - const uint8* v_buf, - uint8* rgb_buf, - int width) { -- FastConvertYUVToRGB32Row_C(y_buf, u_buf, v_buf, rgb_buf, width); -+ FastConvertYUVToRGB32Row_C(y_buf, u_buf, v_buf, rgb_buf, width, 1); - } - - #else - - #define RGBY(i) { \ - static_cast<int16>(1.164 * 64 * (i - 16) + 0.5), \ - static_cast<int16>(1.164 * 64 * (i - 16) + 0.5), \ - static_cast<int16>(1.164 * 64 * (i - 16) + 0.5), \ -diff --git a/gfx/ycbcr/yuv_row_mac.cpp b/gfx/ycbcr/yuv_row_mac.cpp -index 34ecdc1..2a679cc 100644 ---- a/gfx/ycbcr/yuv_row_mac.cpp -+++ b/gfx/ycbcr/yuv_row_mac.cpp -@@ -13,17 +13,17 @@ extern "C" { - // option at all. - #if defined(ARCH_CPU_PPC) || defined(ARCH_CPU_64_BITS) - // PPC implementation uses C fallback - void FastConvertYUVToRGB32Row(const uint8* y_buf, - const uint8* u_buf, - const uint8* v_buf, - uint8* rgb_buf, - int width) { -- FastConvertYUVToRGB32Row_C(y_buf, u_buf, v_buf, rgb_buf, width); -+ FastConvertYUVToRGB32Row_C(y_buf, u_buf, v_buf, rgb_buf, width, 1); - } - - #else - - #define RGBY(i) { \ - static_cast<int16>(1.164 * 64 * (i - 16) + 0.5), \ - static_cast<int16>(1.164 * 64 * (i - 16) + 0.5), \ - static_cast<int16>(1.164 * 64 * (i - 16) + 0.5), \ -diff --git a/gfx/ycbcr/yuv_row_win.cpp b/gfx/ycbcr/yuv_row_win.cpp -index d2b82c4..708ef14 100644 ---- a/gfx/ycbcr/yuv_row_win.cpp -+++ b/gfx/ycbcr/yuv_row_win.cpp -@@ -9,17 +9,17 @@ extern "C" { - // needs to be fixed for 64 bit builds. - #if defined(ARCH_CPU_64_BITS) - // PPC implementation uses C fallback - void FastConvertYUVToRGB32Row(const uint8* y_buf, - const uint8* u_buf, - const uint8* v_buf, - uint8* rgb_buf, - int width) { -- FastConvertYUVToRGB32Row_C(y_buf, u_buf, v_buf, rgb_buf, width); -+ FastConvertYUVToRGB32Row_C(y_buf, u_buf, v_buf, rgb_buf, width, 1); - } - - #else - - - #define RGBY(i) { \ - static_cast<int16>(1.164 * 64 * (i - 16) + 0.5), \ - static_cast<int16>(1.164 * 64 * (i - 16) + 0.5), \