Implements RGB to YV12 conversion in YASM.

media/base/simd/convert_row.asm

+; Copyright (c) 2011 The Chromium Authors. All rights reserved.
+; Use of this source code is governed by a BSD-style license that can be
+; found in the LICENSE file.
+
+%include "convert_row.inc"
+
+;
+; The numer of fractional bits used for fixed-point calculation.
+;
+%define SHIFT_BITS  12
+
+;
+; Tables used for RGB-to-YV12 conversion and YV12-to-RGB conversion. These
+; tables represent the conversion matrices used for the conversions as listed in
+; following C++ snippet. Unfortunately, we need to use precalculated constants
+; here since the yasm preprocessor cannot handle complicated floating-point
+; calculation.
+;
+; #define FIX(n) ((n) * (1 << SHIFT_BITS) + 0.5)
+;
+; int16 SIMD_ConvertRGBAToYUV_kTable[8 * 3] = {

[fbarchard, 2011/06/10 17:29:45]

I suggest calling this SIMD_ConvertARGBToYUV which matches the fourcc's in
WebRTC/Talk.  RGBA is also defined and has a deferent definition.

+;     FIX(0.098), FIX(0.504), FIX(0.257), 0,
+;     FIX(0.098), FIX(0.504), FIX(0.257), 0,
+;     FIX(0.5), 0, FIX(0.5), 0,
+;     FIX(0.5), 0, FIX(0.5), 0,
+;     FIX(0.439), -FIX(0.291), -FIX(0.148), 0,
+;     -FIX(0.071), -FIX(0.368), FIX(0.439), 0,
+; };
+;
+; int16 SIMD_ConvertYUVtoRGBA_kTable[8 * 2] = {
+;     -16, -128, -128, FIX(0.5),
+;     FIX(1.164), 0, FIX(1.596), 1,
+;     FIX(1.164), -FIX(0.391), -FIX(0.813), 1,
+;     FIX(1.164), FIX(2.018), 0, 1,
+; };
+;
+  SECTION   SEG_CONST
+
+  global    EXTN(SIMD_ConvertRGBAToYUVRow_kTable) PRIVATE
+
+EXTN(SIMD_ConvertRGBAToYUVRow_kTable):
+%if SHIFT_BITS == 8
+  dw        25, 129, 66, 0, 25, 129, 66, 0
+  dw        128, 0, 128, 0, 128, 0, 128, 0
+  dw        112, -74, -38, 0, -18, -94, 112, 0
+%elif SHIFT_BITS == 12
+  dw        401, 2064, 1053, 0
+  dw        401, 2064, 1053, 0
+  dw        2048, 0, 2048, 0

[rbultje, 2011/06/10 18:26:33]

You can "simplify" these tables using preprocessor magic:
401 >> (12 - SHIFT_BITS)
You can make that a macro to make it less characters to type.

Also, the times element is nice:
times 2 dw 401, 2064, 1053, 0
times 4 dw 2048, 0

+  dw        2048, 0, 2048, 0
+  dw        1798, -1191, -605, 0
+  dw        -290, -1506, 1798, 0
+%endif
+
+  global EXTN(SIMD_ConvertYUVtoRGBA_kTable) PRIVATE
+
+EXTN(SIMD_ConvertYUVtoRGBA_kTable):
+%if SHIFT_BITS == 8
+  dw        -16, -128, -128, 128
+  dw        298, 0, 409, 1
+  dw        298, -100, -208, 1
+  dw        298, 516, 0, 1,
+%elif SHIFT_BITS == 12
+  dw        -16, -128, -128, 2048
+  dw        4768, 0, 6537, 1
+  dw        4768, -1601, -3329, 1
+  dw        4768, 8266, 0, 1
+%endif
+
+  SECTION   SEG_TEXT
+  CPU       PENTIUM4
+;
+; void media::simd::ConvertRGBAToYUVRow(const uint8* rgba,
+;                                       uint8* y,
+;                                       uint8* u,
+;                                       uint8* v,
+;                                       int width);
+;
+%ifdef MACHO
+%define SYMBOL_NAME     __ZN5media4simd19ConvertRGBAtoYUVRowEPKhPhS3_S3_i
+%elifdef ELF
+%define SYMBOL_NAME     _ZN5media4simd19ConvertRGBAtoYUVRowEPKhPhS3_S3_i
+%elifdef MSVC
+%define SYMBOL_NAME     ?ConvertRGBAtoYUVRow@simd@media@@YAXPBEPAE11H@Z
+%endif
+
+%define PIXELSIZE       4
+%include "convert_row_rgb.inc"
+
+;
+; void media::simd::ConvertRGBAToYUVRow(const uint8* rgba,
+;                                       uint8* y,
+;                                       uint8* u,
+;                                       uint8* v,
+;                                       int width);
+;
+%ifdef MACHO
+%define SYMBOL_NAME     __ZN5media4simd18ConvertRGBtoYUVRowEPKhPhS3_S3_i
+%elifdef ELF
+%define SYMBOL_NAME     _ZN5media4simd18ConvertRGBtoYUVRowEPKhPhS3_S3_i
+%elifdef MSVC
+%define SYMBOL_NAME     ?ConvertRGBtoYUVRow@simd@media@@YAXPBEPAE11H@Z
+%endif
+
+%define PIXELSIZE       3
+%include "convert_row_rgb.inc"
+
+;
+; void media::simd::ConvertYUVtoRGBARow(const uint8* y,
+;                                       const uint8* u,
+;                                       const uint8* v,
+;                                       const uint8* rgba,
+;                                       int width);
+;
+%ifdef MACHO
+%define SYMBOL_NAME     __ZN5media4simd19ConvertYUVtoRGBARowEPKhS2_S2_Phi
+%elifdef ELF
+%define SYMBOL_NAME     _ZN5media4simd19ConvertYUVtoRGBARowEPKhS2_S2_Phi
+%elifdef MSVC
+%define SYMBOL_NAME     ?ConvertYUVtoRGBARow@simd@media@@YAXPBE00PAEH@Z
+%endif
+
+%define PIXELSIZE       4
+%define ALPHA           255
+%include "convert_row_yuv.inc"
+
+;
+; void media::simd::ConvertYUVtoRGBRow(const uint8* y,
+;                                      const uint8* u,
+;                                      const uint8* v,
+;                                      const uint8* rgb,
+;                                      int width);
+;
+%ifdef MACHO
+%define SYMBOL_NAME     __ZN5media4simd18ConvertYUVtoRGBRowEPKhS2_S2_Phi
+%elifdef ELF
+%define SYMBOL_NAME     _ZN5media4simd18ConvertYUVtoRGBRowEPKhS2_S2_Phi
+%elifdef MSVC
+%define SYMBOL_NAME     ?ConvertYUVtoRGBRow@simd@media@@YAXPBE00PAEH@Z
+%endif
+
+%define PIXELSIZE       3
+%define ALPHA           0
+%include "convert_row_yuv.inc"