1.1 --- /dev/null Thu Jan 01 00:00:00 1970 +0000
1.2 +++ b/gfx/ycbcr/win64.patch Wed Dec 31 06:09:35 2014 +0100
1.3 @@ -0,0 +1,210 @@
1.4 +diff --git a/gfx/ycbcr/yuv_row_win64.cpp b/gfx/ycbcr/yuv_row_win64.cpp
1.5 +new file mode 100644
1.6 +--- /dev/null
1.7 ++++ b/gfx/ycbcr/yuv_row_win64.cpp
1.8 +@@ -0,0 +1,205 @@
1.9 ++// Copyright (c) 2010 The Chromium Authors. All rights reserved.
1.10 ++// Use of this source code is governed by a BSD-style license that can be
1.11 ++// found in the LICENSE file.
1.12 ++
1.13 ++#include "yuv_row.h"
1.14 ++
1.15 ++extern "C" {
1.16 ++
1.17 ++// x64 compiler doesn't support MMX and inline assembler. Use SSE2 intrinsics.
1.18 ++
1.19 ++#define kCoefficientsRgbU (reinterpret_cast<uint8*>(kCoefficientsRgbY) + 2048)
1.20 ++#define kCoefficientsRgbV (reinterpret_cast<uint8*>(kCoefficientsRgbY) + 4096)
1.21 ++
1.22 ++#include <emmintrin.h>
1.23 ++
1.24 ++static void FastConvertYUVToRGB32Row_SSE2(const uint8* y_buf,
1.25 ++ const uint8* u_buf,
1.26 ++ const uint8* v_buf,
1.27 ++ uint8* rgb_buf,
1.28 ++ int width) {
1.29 ++ __m128i xmm0, xmmY1, xmmY2;
1.30 ++ __m128 xmmY;
1.31 ++
1.32 ++ while (width >= 2) {
1.33 ++ xmm0 = _mm_adds_epi16(_mm_loadl_epi64(reinterpret_cast<__m128i*>(kCoefficientsRgbU + 8 * *u_buf++)),
1.34 ++ _mm_loadl_epi64(reinterpret_cast<__m128i*>(kCoefficientsRgbV + 8 * *v_buf++)));
1.35 ++
1.36 ++ xmmY1 = _mm_loadl_epi64(reinterpret_cast<__m128i*>(reinterpret_cast<uint8*>(kCoefficientsRgbY) + 8 * *y_buf++));
1.37 ++ xmmY1 = _mm_adds_epi16(xmmY1, xmm0);
1.38 ++
1.39 ++ xmmY2 = _mm_loadl_epi64(reinterpret_cast<__m128i*>(reinterpret_cast<uint8*>(kCoefficientsRgbY) + 8 * *y_buf++));
1.40 ++ xmmY2 = _mm_adds_epi16(xmmY2, xmm0);
1.41 ++
1.42 ++ xmmY = _mm_shuffle_ps(_mm_castsi128_ps(xmmY1), _mm_castsi128_ps(xmmY2),
1.43 ++ 0x44);
1.44 ++ xmmY1 = _mm_srai_epi16(_mm_castps_si128(xmmY), 6);
1.45 ++ xmmY1 = _mm_packus_epi16(xmmY1, xmmY1);
1.46 ++
1.47 ++ _mm_storel_epi64(reinterpret_cast<__m128i*>(rgb_buf), xmmY1);
1.48 ++ rgb_buf += 8;
1.49 ++ width -= 2;
1.50 ++ }
1.51 ++
1.52 ++ if (width) {
1.53 ++ xmm0 = _mm_adds_epi16(_mm_loadl_epi64(reinterpret_cast<__m128i*>(kCoefficientsRgbU + 8 * *u_buf)),
1.54 ++ _mm_loadl_epi64(reinterpret_cast<__m128i*>(kCoefficientsRgbV + 8 * *v_buf)));
1.55 ++ xmmY1 = _mm_loadl_epi64(reinterpret_cast<__m128i*>(reinterpret_cast<uint8*>(kCoefficientsRgbY) + 8 * *y_buf));
1.56 ++ xmmY1 = _mm_adds_epi16(xmmY1, xmm0);
1.57 ++ xmmY1 = _mm_srai_epi16(xmmY1, 6);
1.58 ++ xmmY1 = _mm_packus_epi16(xmmY1, xmmY1);
1.59 ++ *reinterpret_cast<uint32*>(rgb_buf) = _mm_cvtsi128_si32(xmmY1);
1.60 ++ }
1.61 ++}
1.62 ++
1.63 ++static void ScaleYUVToRGB32Row_SSE2(const uint8* y_buf,
1.64 ++ const uint8* u_buf,
1.65 ++ const uint8* v_buf,
1.66 ++ uint8* rgb_buf,
1.67 ++ int width,
1.68 ++ int source_dx) {
1.69 ++ __m128i xmm0, xmmY1, xmmY2;
1.70 ++ __m128 xmmY;
1.71 ++ uint8 u, v, y;
1.72 ++ int x = 0;
1.73 ++
1.74 ++ while (width >= 2) {
1.75 ++ u = u_buf[x >> 17];
1.76 ++ v = v_buf[x >> 17];
1.77 ++ y = y_buf[x >> 16];
1.78 ++ x += source_dx;
1.79 ++
1.80 ++ xmm0 = _mm_adds_epi16(_mm_loadl_epi64(reinterpret_cast<__m128i*>(kCoefficientsRgbU + 8 * u)),
1.81 ++ _mm_loadl_epi64(reinterpret_cast<__m128i*>(kCoefficientsRgbV + 8 * v)));
1.82 ++ xmmY1 = _mm_loadl_epi64(reinterpret_cast<__m128i*>(reinterpret_cast<uint8*>(kCoefficientsRgbY) + 8 * y));
1.83 ++ xmmY1 = _mm_adds_epi16(xmmY1, xmm0);
1.84 ++
1.85 ++ y = y_buf[x >> 16];
1.86 ++ x += source_dx;
1.87 ++
1.88 ++ xmmY2 = _mm_loadl_epi64(reinterpret_cast<__m128i*>(reinterpret_cast<uint8*>(kCoefficientsRgbY) + 8 * y));
1.89 ++ xmmY2 = _mm_adds_epi16(xmmY2, xmm0);
1.90 ++
1.91 ++ xmmY = _mm_shuffle_ps(_mm_castsi128_ps(xmmY1), _mm_castsi128_ps(xmmY2),
1.92 ++ 0x44);
1.93 ++ xmmY1 = _mm_srai_epi16(_mm_castps_si128(xmmY), 6);
1.94 ++ xmmY1 = _mm_packus_epi16(xmmY1, xmmY1);
1.95 ++
1.96 ++ _mm_storel_epi64(reinterpret_cast<__m128i*>(rgb_buf), xmmY1);
1.97 ++ rgb_buf += 8;
1.98 ++ width -= 2;
1.99 ++ }
1.100 ++
1.101 ++ if (width) {
1.102 ++ u = u_buf[x >> 17];
1.103 ++ v = v_buf[x >> 17];
1.104 ++ y = y_buf[x >> 16];
1.105 ++
1.106 ++ xmm0 = _mm_adds_epi16(_mm_loadl_epi64(reinterpret_cast<__m128i*>(kCoefficientsRgbU + 8 * u)),
1.107 ++ _mm_loadl_epi64(reinterpret_cast<__m128i*>(kCoefficientsRgbV + 8 * v)));
1.108 ++ xmmY1 = _mm_loadl_epi64(reinterpret_cast<__m128i*>(reinterpret_cast<uint8*>(kCoefficientsRgbY) + 8 * y));
1.109 ++ xmmY1 = _mm_adds_epi16(xmmY1, xmm0);
1.110 ++ xmmY1 = _mm_srai_epi16(xmmY1, 6);
1.111 ++ xmmY1 = _mm_packus_epi16(xmmY1, xmmY1);
1.112 ++ *reinterpret_cast<uint32*>(rgb_buf) = _mm_cvtsi128_si32(xmmY1);
1.113 ++ }
1.114 ++}
1.115 ++
1.116 ++static void LinearScaleYUVToRGB32Row_SSE2(const uint8* y_buf,
1.117 ++ const uint8* u_buf,
1.118 ++ const uint8* v_buf,
1.119 ++ uint8* rgb_buf,
1.120 ++ int width,
1.121 ++ int source_dx) {
1.122 ++ __m128i xmm0, xmmY1, xmmY2;
1.123 ++ __m128 xmmY;
1.124 ++ uint8 u0, u1, v0, v1, y0, y1;
1.125 ++ uint32 uv_frac, y_frac, u, v, y;
1.126 ++ int x = 0;
1.127 ++
1.128 ++ if (source_dx >= 0x20000) {
1.129 ++ x = 32768;
1.130 ++ }
1.131 ++
1.132 ++ while(width >= 2) {
1.133 ++ u0 = u_buf[x >> 17];
1.134 ++ u1 = u_buf[(x >> 17) + 1];
1.135 ++ v0 = v_buf[x >> 17];
1.136 ++ v1 = v_buf[(x >> 17) + 1];
1.137 ++ y0 = y_buf[x >> 16];
1.138 ++ y1 = y_buf[(x >> 16) + 1];
1.139 ++ uv_frac = (x & 0x1fffe);
1.140 ++ y_frac = (x & 0xffff);
1.141 ++ u = (uv_frac * u1 + (uv_frac ^ 0x1fffe) * u0) >> 17;
1.142 ++ v = (uv_frac * v1 + (uv_frac ^ 0x1fffe) * v0) >> 17;
1.143 ++ y = (y_frac * y1 + (y_frac ^ 0xffff) * y0) >> 16;
1.144 ++ x += source_dx;
1.145 ++
1.146 ++ xmm0 = _mm_adds_epi16(_mm_loadl_epi64(reinterpret_cast<__m128i*>(kCoefficientsRgbU + 8 * u)),
1.147 ++ _mm_loadl_epi64(reinterpret_cast<__m128i*>(kCoefficientsRgbV + 8 * v)));
1.148 ++ xmmY1 = _mm_loadl_epi64(reinterpret_cast<__m128i*>(reinterpret_cast<uint8*>(kCoefficientsRgbY) + 8 * y));
1.149 ++ xmmY1 = _mm_adds_epi16(xmmY1, xmm0);
1.150 ++
1.151 ++ y0 = y_buf[x >> 16];
1.152 ++ y1 = y_buf[(x >> 16) + 1];
1.153 ++ y_frac = (x & 0xffff);
1.154 ++ y = (y_frac * y1 + (y_frac ^ 0xffff) * y0) >> 16;
1.155 ++ x += source_dx;
1.156 ++
1.157 ++ xmmY2 = _mm_loadl_epi64(reinterpret_cast<__m128i*>(reinterpret_cast<uint8*>(kCoefficientsRgbY) + 8 * y));
1.158 ++ xmmY2 = _mm_adds_epi16(xmmY2, xmm0);
1.159 ++
1.160 ++ xmmY = _mm_shuffle_ps(_mm_castsi128_ps(xmmY1), _mm_castsi128_ps(xmmY2),
1.161 ++ 0x44);
1.162 ++ xmmY1 = _mm_srai_epi16(_mm_castps_si128(xmmY), 6);
1.163 ++ xmmY1 = _mm_packus_epi16(xmmY1, xmmY1);
1.164 ++
1.165 ++ _mm_storel_epi64(reinterpret_cast<__m128i*>(rgb_buf), xmmY1);
1.166 ++ rgb_buf += 8;
1.167 ++ width -= 2;
1.168 ++ }
1.169 ++
1.170 ++ if (width) {
1.171 ++ u = u_buf[x >> 17];
1.172 ++ v = v_buf[x >> 17];
1.173 ++ y = y_buf[x >> 16];
1.174 ++
1.175 ++ xmm0 = _mm_adds_epi16(_mm_loadl_epi64(reinterpret_cast<__m128i*>(kCoefficientsRgbU + 8 * u)),
1.176 ++ _mm_loadl_epi64(reinterpret_cast<__m128i*>(kCoefficientsRgbV + 8 * v)));
1.177 ++ xmmY1 = _mm_loadl_epi64(reinterpret_cast<__m128i*>(reinterpret_cast<uint8*>(kCoefficientsRgbY) + 8 * y));
1.178 ++
1.179 ++ xmmY1 = _mm_adds_epi16(xmmY1, xmm0);
1.180 ++ xmmY1 = _mm_srai_epi16(xmmY1, 6);
1.181 ++ xmmY1 = _mm_packus_epi16(xmmY1, xmmY1);
1.182 ++ *reinterpret_cast<uint32*>(rgb_buf) = _mm_cvtsi128_si32(xmmY1);
1.183 ++ }
1.184 ++}
1.185 ++
1.186 ++void FastConvertYUVToRGB32Row(const uint8* y_buf,
1.187 ++ const uint8* u_buf,
1.188 ++ const uint8* v_buf,
1.189 ++ uint8* rgb_buf,
1.190 ++ int width) {
1.191 ++ FastConvertYUVToRGB32Row_SSE2(y_buf, u_buf, v_buf, rgb_buf, width);
1.192 ++}
1.193 ++
1.194 ++void ScaleYUVToRGB32Row(const uint8* y_buf,
1.195 ++ const uint8* u_buf,
1.196 ++ const uint8* v_buf,
1.197 ++ uint8* rgb_buf,
1.198 ++ int width,
1.199 ++ int source_dx) {
1.200 ++ ScaleYUVToRGB32Row_SSE2(y_buf, u_buf, v_buf, rgb_buf, width, source_dx);
1.201 ++}
1.202 ++
1.203 ++void LinearScaleYUVToRGB32Row(const uint8* y_buf,
1.204 ++ const uint8* u_buf,
1.205 ++ const uint8* v_buf,
1.206 ++ uint8* rgb_buf,
1.207 ++ int width,
1.208 ++ int source_dx) {
1.209 ++ LinearScaleYUVToRGB32Row_SSE2(y_buf, u_buf, v_buf, rgb_buf, width,
1.210 ++ source_dx);
1.211 ++}
1.212 ++
1.213 ++} // extern "C"
