1.1 --- /dev/null Thu Jan 01 00:00:00 1970 +0000
1.2 +++ b/gfx/ycbcr/yuv_row_posix.cpp Wed Dec 31 06:09:35 2014 +0100
1.3 @@ -0,0 +1,915 @@
1.4 +// Copyright (c) 2010 The Chromium Authors. All rights reserved.
1.5 +// Use of this source code is governed by a BSD-style license that can be
1.6 +// found in the LICENSE file.
1.7 +
1.8 +#include "yuv_row.h"
1.9 +#include "mozilla/SSE.h"
1.10 +
1.11 +#define DCHECK(a)
1.12 +
1.13 +extern "C" {
1.14 +
1.15 +#if defined(ARCH_CPU_X86_64)
1.16 +
1.17 +// We don't need CPUID guards here, since x86-64 implies SSE2.
1.18 +
1.19 +// AMD64 ABI uses register paremters.
1.20 +void FastConvertYUVToRGB32Row(const uint8* y_buf, // rdi
1.21 + const uint8* u_buf, // rsi
1.22 + const uint8* v_buf, // rdx
1.23 + uint8* rgb_buf, // rcx
1.24 + int width) { // r8
1.25 + asm(
1.26 + "jmp 1f\n"
1.27 +"0:"
1.28 + "movzb (%1),%%r10\n"
1.29 + "add $0x1,%1\n"
1.30 + "movzb (%2),%%r11\n"
1.31 + "add $0x1,%2\n"
1.32 + "movq 2048(%5,%%r10,8),%%xmm0\n"
1.33 + "movzb (%0),%%r10\n"
1.34 + "movq 4096(%5,%%r11,8),%%xmm1\n"
1.35 + "movzb 0x1(%0),%%r11\n"
1.36 + "paddsw %%xmm1,%%xmm0\n"
1.37 + "movq (%5,%%r10,8),%%xmm2\n"
1.38 + "add $0x2,%0\n"
1.39 + "movq (%5,%%r11,8),%%xmm3\n"
1.40 + "paddsw %%xmm0,%%xmm2\n"
1.41 + "paddsw %%xmm0,%%xmm3\n"
1.42 + "shufps $0x44,%%xmm3,%%xmm2\n"
1.43 + "psraw $0x6,%%xmm2\n"
1.44 + "packuswb %%xmm2,%%xmm2\n"
1.45 + "movq %%xmm2,0x0(%3)\n"
1.46 + "add $0x8,%3\n"
1.47 +"1:"
1.48 + "sub $0x2,%4\n"
1.49 + "jns 0b\n"
1.50 +
1.51 +"2:"
1.52 + "add $0x1,%4\n"
1.53 + "js 3f\n"
1.54 +
1.55 + "movzb (%1),%%r10\n"
1.56 + "movq 2048(%5,%%r10,8),%%xmm0\n"
1.57 + "movzb (%2),%%r10\n"
1.58 + "movq 4096(%5,%%r10,8),%%xmm1\n"
1.59 + "paddsw %%xmm1,%%xmm0\n"
1.60 + "movzb (%0),%%r10\n"
1.61 + "movq (%5,%%r10,8),%%xmm1\n"
1.62 + "paddsw %%xmm0,%%xmm1\n"
1.63 + "psraw $0x6,%%xmm1\n"
1.64 + "packuswb %%xmm1,%%xmm1\n"
1.65 + "movd %%xmm1,0x0(%3)\n"
1.66 +"3:"
1.67 + :
1.68 + : "r"(y_buf), // %0
1.69 + "r"(u_buf), // %1
1.70 + "r"(v_buf), // %2
1.71 + "r"(rgb_buf), // %3
1.72 + "r"(width), // %4
1.73 + "r" (kCoefficientsRgbY) // %5
1.74 + : "memory", "r10", "r11", "xmm0", "xmm1", "xmm2", "xmm3"
1.75 +);
1.76 +}
1.77 +
1.78 +void ScaleYUVToRGB32Row(const uint8* y_buf, // rdi
1.79 + const uint8* u_buf, // rsi
1.80 + const uint8* v_buf, // rdx
1.81 + uint8* rgb_buf, // rcx
1.82 + int width, // r8
1.83 + int source_dx) { // r9
1.84 + asm(
1.85 + "xor %%r11,%%r11\n"
1.86 + "sub $0x2,%4\n"
1.87 + "js 1f\n"
1.88 +
1.89 +"0:"
1.90 + "mov %%r11,%%r10\n"
1.91 + "sar $0x11,%%r10\n"
1.92 + "movzb (%1,%%r10,1),%%rax\n"
1.93 + "movq 2048(%5,%%rax,8),%%xmm0\n"
1.94 + "movzb (%2,%%r10,1),%%rax\n"
1.95 + "movq 4096(%5,%%rax,8),%%xmm1\n"
1.96 + "lea (%%r11,%6),%%r10\n"
1.97 + "sar $0x10,%%r11\n"
1.98 + "movzb (%0,%%r11,1),%%rax\n"
1.99 + "paddsw %%xmm1,%%xmm0\n"
1.100 + "movq (%5,%%rax,8),%%xmm1\n"
1.101 + "lea (%%r10,%6),%%r11\n"
1.102 + "sar $0x10,%%r10\n"
1.103 + "movzb (%0,%%r10,1),%%rax\n"
1.104 + "movq (%5,%%rax,8),%%xmm2\n"
1.105 + "paddsw %%xmm0,%%xmm1\n"
1.106 + "paddsw %%xmm0,%%xmm2\n"
1.107 + "shufps $0x44,%%xmm2,%%xmm1\n"
1.108 + "psraw $0x6,%%xmm1\n"
1.109 + "packuswb %%xmm1,%%xmm1\n"
1.110 + "movq %%xmm1,0x0(%3)\n"
1.111 + "add $0x8,%3\n"
1.112 + "sub $0x2,%4\n"
1.113 + "jns 0b\n"
1.114 +
1.115 +"1:"
1.116 + "add $0x1,%4\n"
1.117 + "js 2f\n"
1.118 +
1.119 + "mov %%r11,%%r10\n"
1.120 + "sar $0x11,%%r10\n"
1.121 + "movzb (%1,%%r10,1),%%rax\n"
1.122 + "movq 2048(%5,%%rax,8),%%xmm0\n"
1.123 + "movzb (%2,%%r10,1),%%rax\n"
1.124 + "movq 4096(%5,%%rax,8),%%xmm1\n"
1.125 + "paddsw %%xmm1,%%xmm0\n"
1.126 + "sar $0x10,%%r11\n"
1.127 + "movzb (%0,%%r11,1),%%rax\n"
1.128 + "movq (%5,%%rax,8),%%xmm1\n"
1.129 + "paddsw %%xmm0,%%xmm1\n"
1.130 + "psraw $0x6,%%xmm1\n"
1.131 + "packuswb %%xmm1,%%xmm1\n"
1.132 + "movd %%xmm1,0x0(%3)\n"
1.133 +
1.134 +"2:"
1.135 + :
1.136 + : "r"(y_buf), // %0
1.137 + "r"(u_buf), // %1
1.138 + "r"(v_buf), // %2
1.139 + "r"(rgb_buf), // %3
1.140 + "r"(width), // %4
1.141 + "r" (kCoefficientsRgbY), // %5
1.142 + "r"(static_cast<long>(source_dx)) // %6
1.143 + : "memory", "r10", "r11", "rax", "xmm0", "xmm1", "xmm2"
1.144 +);
1.145 +}
1.146 +
1.147 +void LinearScaleYUVToRGB32Row(const uint8* y_buf,
1.148 + const uint8* u_buf,
1.149 + const uint8* v_buf,
1.150 + uint8* rgb_buf,
1.151 + int width,
1.152 + int source_dx) {
1.153 + asm(
1.154 + "xor %%r11,%%r11\n" // x = 0
1.155 + "sub $0x2,%4\n"
1.156 + "js 2f\n"
1.157 + "cmp $0x20000,%6\n" // if source_dx >= 2.0
1.158 + "jl 0f\n"
1.159 + "mov $0x8000,%%r11\n" // x = 0.5 for 1/2 or less
1.160 +"0:"
1.161 +
1.162 +"1:"
1.163 + "mov %%r11,%%r10\n"
1.164 + "sar $0x11,%%r10\n"
1.165 +
1.166 + "movzb (%1, %%r10, 1), %%r13 \n"
1.167 + "movzb 1(%1, %%r10, 1), %%r14 \n"
1.168 + "mov %%r11, %%rax \n"
1.169 + "and $0x1fffe, %%rax \n"
1.170 + "imul %%rax, %%r14 \n"
1.171 + "xor $0x1fffe, %%rax \n"
1.172 + "imul %%rax, %%r13 \n"
1.173 + "add %%r14, %%r13 \n"
1.174 + "shr $17, %%r13 \n"
1.175 + "movq 2048(%5,%%r13,8), %%xmm0\n"
1.176 +
1.177 + "movzb (%2, %%r10, 1), %%r13 \n"
1.178 + "movzb 1(%2, %%r10, 1), %%r14 \n"
1.179 + "mov %%r11, %%rax \n"
1.180 + "and $0x1fffe, %%rax \n"
1.181 + "imul %%rax, %%r14 \n"
1.182 + "xor $0x1fffe, %%rax \n"
1.183 + "imul %%rax, %%r13 \n"
1.184 + "add %%r14, %%r13 \n"
1.185 + "shr $17, %%r13 \n"
1.186 + "movq 4096(%5,%%r13,8), %%xmm1\n"
1.187 +
1.188 + "mov %%r11, %%rax \n"
1.189 + "lea (%%r11,%6),%%r10\n"
1.190 + "sar $0x10,%%r11\n"
1.191 + "paddsw %%xmm1,%%xmm0\n"
1.192 +
1.193 + "movzb (%0, %%r11, 1), %%r13 \n"
1.194 + "movzb 1(%0, %%r11, 1), %%r14 \n"
1.195 + "and $0xffff, %%rax \n"
1.196 + "imul %%rax, %%r14 \n"
1.197 + "xor $0xffff, %%rax \n"
1.198 + "imul %%rax, %%r13 \n"
1.199 + "add %%r14, %%r13 \n"
1.200 + "shr $16, %%r13 \n"
1.201 + "movq (%5,%%r13,8),%%xmm1\n"
1.202 +
1.203 + "mov %%r10, %%rax \n"
1.204 + "lea (%%r10,%6),%%r11\n"
1.205 + "sar $0x10,%%r10\n"
1.206 +
1.207 + "movzb (%0,%%r10,1), %%r13 \n"
1.208 + "movzb 1(%0,%%r10,1), %%r14 \n"
1.209 + "and $0xffff, %%rax \n"
1.210 + "imul %%rax, %%r14 \n"
1.211 + "xor $0xffff, %%rax \n"
1.212 + "imul %%rax, %%r13 \n"
1.213 + "add %%r14, %%r13 \n"
1.214 + "shr $16, %%r13 \n"
1.215 + "movq (%5,%%r13,8),%%xmm2\n"
1.216 +
1.217 + "paddsw %%xmm0,%%xmm1\n"
1.218 + "paddsw %%xmm0,%%xmm2\n"
1.219 + "shufps $0x44,%%xmm2,%%xmm1\n"
1.220 + "psraw $0x6,%%xmm1\n"
1.221 + "packuswb %%xmm1,%%xmm1\n"
1.222 + "movq %%xmm1,0x0(%3)\n"
1.223 + "add $0x8,%3\n"
1.224 + "sub $0x2,%4\n"
1.225 + "jns 1b\n"
1.226 +
1.227 +"2:"
1.228 + "add $0x1,%4\n"
1.229 + "js 3f\n"
1.230 +
1.231 + "mov %%r11,%%r10\n"
1.232 + "sar $0x11,%%r10\n"
1.233 +
1.234 + "movzb (%1,%%r10,1), %%r13 \n"
1.235 + "movq 2048(%5,%%r13,8),%%xmm0\n"
1.236 +
1.237 + "movzb (%2,%%r10,1), %%r13 \n"
1.238 + "movq 4096(%5,%%r13,8),%%xmm1\n"
1.239 +
1.240 + "paddsw %%xmm1,%%xmm0\n"
1.241 + "sar $0x10,%%r11\n"
1.242 +
1.243 + "movzb (%0,%%r11,1), %%r13 \n"
1.244 + "movq (%5,%%r13,8),%%xmm1\n"
1.245 +
1.246 + "paddsw %%xmm0,%%xmm1\n"
1.247 + "psraw $0x6,%%xmm1\n"
1.248 + "packuswb %%xmm1,%%xmm1\n"
1.249 + "movd %%xmm1,0x0(%3)\n"
1.250 +
1.251 +"3:"
1.252 + :
1.253 + : "r"(y_buf), // %0
1.254 + "r"(u_buf), // %1
1.255 + "r"(v_buf), // %2
1.256 + "r"(rgb_buf), // %3
1.257 + "r"(width), // %4
1.258 + "r" (kCoefficientsRgbY), // %5
1.259 + "r"(static_cast<long>(source_dx)) // %6
1.260 + : "memory", "r10", "r11", "r13", "r14", "rax", "xmm0", "xmm1", "xmm2"
1.261 +);
1.262 +}
1.263 +
1.264 +#elif defined(MOZILLA_MAY_SUPPORT_SSE) && defined(ARCH_CPU_X86_32) && !defined(__PIC__)
1.265 +
1.266 +// PIC version is slower because less registers are available, so
1.267 +// non-PIC is used on platforms where it is possible.
1.268 +void FastConvertYUVToRGB32Row_SSE(const uint8* y_buf,
1.269 + const uint8* u_buf,
1.270 + const uint8* v_buf,
1.271 + uint8* rgb_buf,
1.272 + int width);
1.273 + asm(
1.274 + ".text\n"
1.275 + ".global FastConvertYUVToRGB32Row_SSE\n"
1.276 + ".type FastConvertYUVToRGB32Row_SSE, @function\n"
1.277 +"FastConvertYUVToRGB32Row_SSE:\n"
1.278 + "pusha\n"
1.279 + "mov 0x24(%esp),%edx\n"
1.280 + "mov 0x28(%esp),%edi\n"
1.281 + "mov 0x2c(%esp),%esi\n"
1.282 + "mov 0x30(%esp),%ebp\n"
1.283 + "mov 0x34(%esp),%ecx\n"
1.284 + "jmp 1f\n"
1.285 +
1.286 +"0:"
1.287 + "movzbl (%edi),%eax\n"
1.288 + "add $0x1,%edi\n"
1.289 + "movzbl (%esi),%ebx\n"
1.290 + "add $0x1,%esi\n"
1.291 + "movq kCoefficientsRgbY+2048(,%eax,8),%mm0\n"
1.292 + "movzbl (%edx),%eax\n"
1.293 + "paddsw kCoefficientsRgbY+4096(,%ebx,8),%mm0\n"
1.294 + "movzbl 0x1(%edx),%ebx\n"
1.295 + "movq kCoefficientsRgbY(,%eax,8),%mm1\n"
1.296 + "add $0x2,%edx\n"
1.297 + "movq kCoefficientsRgbY(,%ebx,8),%mm2\n"
1.298 + "paddsw %mm0,%mm1\n"
1.299 + "paddsw %mm0,%mm2\n"
1.300 + "psraw $0x6,%mm1\n"
1.301 + "psraw $0x6,%mm2\n"
1.302 + "packuswb %mm2,%mm1\n"
1.303 + "movntq %mm1,0x0(%ebp)\n"
1.304 + "add $0x8,%ebp\n"
1.305 +"1:"
1.306 + "sub $0x2,%ecx\n"
1.307 + "jns 0b\n"
1.308 +
1.309 + "and $0x1,%ecx\n"
1.310 + "je 2f\n"
1.311 +
1.312 + "movzbl (%edi),%eax\n"
1.313 + "movq kCoefficientsRgbY+2048(,%eax,8),%mm0\n"
1.314 + "movzbl (%esi),%eax\n"
1.315 + "paddsw kCoefficientsRgbY+4096(,%eax,8),%mm0\n"
1.316 + "movzbl (%edx),%eax\n"
1.317 + "movq kCoefficientsRgbY(,%eax,8),%mm1\n"
1.318 + "paddsw %mm0,%mm1\n"
1.319 + "psraw $0x6,%mm1\n"
1.320 + "packuswb %mm1,%mm1\n"
1.321 + "movd %mm1,0x0(%ebp)\n"
1.322 +"2:"
1.323 + "popa\n"
1.324 + "ret\n"
1.325 +#if !defined(XP_MACOSX)
1.326 + ".previous\n"
1.327 +#endif
1.328 +);
1.329 +
1.330 +void FastConvertYUVToRGB32Row(const uint8* y_buf,
1.331 + const uint8* u_buf,
1.332 + const uint8* v_buf,
1.333 + uint8* rgb_buf,
1.334 + int width)
1.335 +{
1.336 + if (mozilla::supports_sse()) {
1.337 + FastConvertYUVToRGB32Row_SSE(y_buf, u_buf, v_buf, rgb_buf, width);
1.338 + return;
1.339 + }
1.340 +
1.341 + FastConvertYUVToRGB32Row_C(y_buf, u_buf, v_buf, rgb_buf, width, 1);
1.342 +}
1.343 +
1.344 +
1.345 +void ScaleYUVToRGB32Row_SSE(const uint8* y_buf,
1.346 + const uint8* u_buf,
1.347 + const uint8* v_buf,
1.348 + uint8* rgb_buf,
1.349 + int width,
1.350 + int source_dx);
1.351 + asm(
1.352 + ".text\n"
1.353 + ".global ScaleYUVToRGB32Row_SSE\n"
1.354 + ".type ScaleYUVToRGB32Row_SSE, @function\n"
1.355 +"ScaleYUVToRGB32Row_SSE:\n"
1.356 + "pusha\n"
1.357 + "mov 0x24(%esp),%edx\n"
1.358 + "mov 0x28(%esp),%edi\n"
1.359 + "mov 0x2c(%esp),%esi\n"
1.360 + "mov 0x30(%esp),%ebp\n"
1.361 + "mov 0x34(%esp),%ecx\n"
1.362 + "xor %ebx,%ebx\n"
1.363 + "jmp 1f\n"
1.364 +
1.365 +"0:"
1.366 + "mov %ebx,%eax\n"
1.367 + "sar $0x11,%eax\n"
1.368 + "movzbl (%edi,%eax,1),%eax\n"
1.369 + "movq kCoefficientsRgbY+2048(,%eax,8),%mm0\n"
1.370 + "mov %ebx,%eax\n"
1.371 + "sar $0x11,%eax\n"
1.372 + "movzbl (%esi,%eax,1),%eax\n"
1.373 + "paddsw kCoefficientsRgbY+4096(,%eax,8),%mm0\n"
1.374 + "mov %ebx,%eax\n"
1.375 + "add 0x38(%esp),%ebx\n"
1.376 + "sar $0x10,%eax\n"
1.377 + "movzbl (%edx,%eax,1),%eax\n"
1.378 + "movq kCoefficientsRgbY(,%eax,8),%mm1\n"
1.379 + "mov %ebx,%eax\n"
1.380 + "add 0x38(%esp),%ebx\n"
1.381 + "sar $0x10,%eax\n"
1.382 + "movzbl (%edx,%eax,1),%eax\n"
1.383 + "movq kCoefficientsRgbY(,%eax,8),%mm2\n"
1.384 + "paddsw %mm0,%mm1\n"
1.385 + "paddsw %mm0,%mm2\n"
1.386 + "psraw $0x6,%mm1\n"
1.387 + "psraw $0x6,%mm2\n"
1.388 + "packuswb %mm2,%mm1\n"
1.389 + "movntq %mm1,0x0(%ebp)\n"
1.390 + "add $0x8,%ebp\n"
1.391 +"1:"
1.392 + "sub $0x2,%ecx\n"
1.393 + "jns 0b\n"
1.394 +
1.395 + "and $0x1,%ecx\n"
1.396 + "je 2f\n"
1.397 +
1.398 + "mov %ebx,%eax\n"
1.399 + "sar $0x11,%eax\n"
1.400 + "movzbl (%edi,%eax,1),%eax\n"
1.401 + "movq kCoefficientsRgbY+2048(,%eax,8),%mm0\n"
1.402 + "mov %ebx,%eax\n"
1.403 + "sar $0x11,%eax\n"
1.404 + "movzbl (%esi,%eax,1),%eax\n"
1.405 + "paddsw kCoefficientsRgbY+4096(,%eax,8),%mm0\n"
1.406 + "mov %ebx,%eax\n"
1.407 + "sar $0x10,%eax\n"
1.408 + "movzbl (%edx,%eax,1),%eax\n"
1.409 + "movq kCoefficientsRgbY(,%eax,8),%mm1\n"
1.410 + "paddsw %mm0,%mm1\n"
1.411 + "psraw $0x6,%mm1\n"
1.412 + "packuswb %mm1,%mm1\n"
1.413 + "movd %mm1,0x0(%ebp)\n"
1.414 +
1.415 +"2:"
1.416 + "popa\n"
1.417 + "ret\n"
1.418 +#if !defined(XP_MACOSX)
1.419 + ".previous\n"
1.420 +#endif
1.421 +);
1.422 +
1.423 +void ScaleYUVToRGB32Row(const uint8* y_buf,
1.424 + const uint8* u_buf,
1.425 + const uint8* v_buf,
1.426 + uint8* rgb_buf,
1.427 + int width,
1.428 + int source_dx)
1.429 +{
1.430 + if (mozilla::supports_sse()) {
1.431 + ScaleYUVToRGB32Row_SSE(y_buf, u_buf, v_buf, rgb_buf,
1.432 + width, source_dx);
1.433 + }
1.434 +
1.435 + ScaleYUVToRGB32Row_C(y_buf, u_buf, v_buf, rgb_buf,
1.436 + width, source_dx);
1.437 +}
1.438 +
1.439 +void LinearScaleYUVToRGB32Row_SSE(const uint8* y_buf,
1.440 + const uint8* u_buf,
1.441 + const uint8* v_buf,
1.442 + uint8* rgb_buf,
1.443 + int width,
1.444 + int source_dx);
1.445 + asm(
1.446 + ".text\n"
1.447 + ".global LinearScaleYUVToRGB32Row_SSE\n"
1.448 + ".type LinearScaleYUVToRGB32Row_SSE, @function\n"
1.449 +"LinearScaleYUVToRGB32Row_SSE:\n"
1.450 + "pusha\n"
1.451 + "mov 0x24(%esp),%edx\n"
1.452 + "mov 0x28(%esp),%edi\n"
1.453 + "mov 0x30(%esp),%ebp\n"
1.454 +
1.455 + // source_width = width * source_dx + ebx
1.456 + "mov 0x34(%esp), %ecx\n"
1.457 + "imull 0x38(%esp), %ecx\n"
1.458 + "mov %ecx, 0x34(%esp)\n"
1.459 +
1.460 + "mov 0x38(%esp), %ecx\n"
1.461 + "xor %ebx,%ebx\n" // x = 0
1.462 + "cmp $0x20000,%ecx\n" // if source_dx >= 2.0
1.463 + "jl 1f\n"
1.464 + "mov $0x8000,%ebx\n" // x = 0.5 for 1/2 or less
1.465 + "jmp 1f\n"
1.466 +
1.467 +"0:"
1.468 + "mov %ebx,%eax\n"
1.469 + "sar $0x11,%eax\n"
1.470 +
1.471 + "movzbl (%edi,%eax,1),%ecx\n"
1.472 + "movzbl 1(%edi,%eax,1),%esi\n"
1.473 + "mov %ebx,%eax\n"
1.474 + "andl $0x1fffe, %eax \n"
1.475 + "imul %eax, %esi \n"
1.476 + "xorl $0x1fffe, %eax \n"
1.477 + "imul %eax, %ecx \n"
1.478 + "addl %esi, %ecx \n"
1.479 + "shrl $17, %ecx \n"
1.480 + "movq kCoefficientsRgbY+2048(,%ecx,8),%mm0\n"
1.481 +
1.482 + "mov 0x2c(%esp),%esi\n"
1.483 + "mov %ebx,%eax\n"
1.484 + "sar $0x11,%eax\n"
1.485 +
1.486 + "movzbl (%esi,%eax,1),%ecx\n"
1.487 + "movzbl 1(%esi,%eax,1),%esi\n"
1.488 + "mov %ebx,%eax\n"
1.489 + "andl $0x1fffe, %eax \n"
1.490 + "imul %eax, %esi \n"
1.491 + "xorl $0x1fffe, %eax \n"
1.492 + "imul %eax, %ecx \n"
1.493 + "addl %esi, %ecx \n"
1.494 + "shrl $17, %ecx \n"
1.495 + "paddsw kCoefficientsRgbY+4096(,%ecx,8),%mm0\n"
1.496 +
1.497 + "mov %ebx,%eax\n"
1.498 + "sar $0x10,%eax\n"
1.499 + "movzbl (%edx,%eax,1),%ecx\n"
1.500 + "movzbl 1(%edx,%eax,1),%esi\n"
1.501 + "mov %ebx,%eax\n"
1.502 + "add 0x38(%esp),%ebx\n"
1.503 + "andl $0xffff, %eax \n"
1.504 + "imul %eax, %esi \n"
1.505 + "xorl $0xffff, %eax \n"
1.506 + "imul %eax, %ecx \n"
1.507 + "addl %esi, %ecx \n"
1.508 + "shrl $16, %ecx \n"
1.509 + "movq kCoefficientsRgbY(,%ecx,8),%mm1\n"
1.510 +
1.511 + "cmp 0x34(%esp), %ebx\n"
1.512 + "jge 2f\n"
1.513 +
1.514 + "mov %ebx,%eax\n"
1.515 + "sar $0x10,%eax\n"
1.516 + "movzbl (%edx,%eax,1),%ecx\n"
1.517 + "movzbl 1(%edx,%eax,1),%esi\n"
1.518 + "mov %ebx,%eax\n"
1.519 + "add 0x38(%esp),%ebx\n"
1.520 + "andl $0xffff, %eax \n"
1.521 + "imul %eax, %esi \n"
1.522 + "xorl $0xffff, %eax \n"
1.523 + "imul %eax, %ecx \n"
1.524 + "addl %esi, %ecx \n"
1.525 + "shrl $16, %ecx \n"
1.526 + "movq kCoefficientsRgbY(,%ecx,8),%mm2\n"
1.527 +
1.528 + "paddsw %mm0,%mm1\n"
1.529 + "paddsw %mm0,%mm2\n"
1.530 + "psraw $0x6,%mm1\n"
1.531 + "psraw $0x6,%mm2\n"
1.532 + "packuswb %mm2,%mm1\n"
1.533 + "movntq %mm1,0x0(%ebp)\n"
1.534 + "add $0x8,%ebp\n"
1.535 +
1.536 +"1:"
1.537 + "cmp 0x34(%esp), %ebx\n"
1.538 + "jl 0b\n"
1.539 + "popa\n"
1.540 + "ret\n"
1.541 +
1.542 +"2:"
1.543 + "paddsw %mm0, %mm1\n"
1.544 + "psraw $6, %mm1\n"
1.545 + "packuswb %mm1, %mm1\n"
1.546 + "movd %mm1, (%ebp)\n"
1.547 + "popa\n"
1.548 + "ret\n"
1.549 +#if !defined(XP_MACOSX)
1.550 + ".previous\n"
1.551 +#endif
1.552 +);
1.553 +
1.554 +void LinearScaleYUVToRGB32Row(const uint8* y_buf,
1.555 + const uint8* u_buf,
1.556 + const uint8* v_buf,
1.557 + uint8* rgb_buf,
1.558 + int width,
1.559 + int source_dx)
1.560 +{
1.561 + if (mozilla::supports_sse()) {
1.562 + LinearScaleYUVToRGB32Row_SSE(y_buf, u_buf, v_buf, rgb_buf,
1.563 + width, source_dx);
1.564 + }
1.565 +
1.566 + LinearScaleYUVToRGB32Row_C(y_buf, u_buf, v_buf, rgb_buf,
1.567 + width, source_dx);
1.568 +}
1.569 +
1.570 +#elif defined(MOZILLA_MAY_SUPPORT_SSE) && defined(ARCH_CPU_X86_32) && defined(__PIC__)
1.571 +
1.572 +void PICConvertYUVToRGB32Row_SSE(const uint8* y_buf,
1.573 + const uint8* u_buf,
1.574 + const uint8* v_buf,
1.575 + uint8* rgb_buf,
1.576 + int width,
1.577 + int16 *kCoefficientsRgbY);
1.578 +
1.579 + asm(
1.580 + ".text\n"
1.581 +#if defined(XP_MACOSX)
1.582 +"_PICConvertYUVToRGB32Row_SSE:\n"
1.583 +#else
1.584 +"PICConvertYUVToRGB32Row_SSE:\n"
1.585 +#endif
1.586 + "pusha\n"
1.587 + "mov 0x24(%esp),%edx\n"
1.588 + "mov 0x28(%esp),%edi\n"
1.589 + "mov 0x2c(%esp),%esi\n"
1.590 + "mov 0x30(%esp),%ebp\n"
1.591 + "mov 0x38(%esp),%ecx\n"
1.592 +
1.593 + "jmp 1f\n"
1.594 +
1.595 +"0:"
1.596 + "movzbl (%edi),%eax\n"
1.597 + "add $0x1,%edi\n"
1.598 + "movzbl (%esi),%ebx\n"
1.599 + "add $0x1,%esi\n"
1.600 + "movq 2048(%ecx,%eax,8),%mm0\n"
1.601 + "movzbl (%edx),%eax\n"
1.602 + "paddsw 4096(%ecx,%ebx,8),%mm0\n"
1.603 + "movzbl 0x1(%edx),%ebx\n"
1.604 + "movq 0(%ecx,%eax,8),%mm1\n"
1.605 + "add $0x2,%edx\n"
1.606 + "movq 0(%ecx,%ebx,8),%mm2\n"
1.607 + "paddsw %mm0,%mm1\n"
1.608 + "paddsw %mm0,%mm2\n"
1.609 + "psraw $0x6,%mm1\n"
1.610 + "psraw $0x6,%mm2\n"
1.611 + "packuswb %mm2,%mm1\n"
1.612 + "movntq %mm1,0x0(%ebp)\n"
1.613 + "add $0x8,%ebp\n"
1.614 +"1:"
1.615 + "subl $0x2,0x34(%esp)\n"
1.616 + "jns 0b\n"
1.617 +
1.618 + "andl $0x1,0x34(%esp)\n"
1.619 + "je 2f\n"
1.620 +
1.621 + "movzbl (%edi),%eax\n"
1.622 + "movq 2048(%ecx,%eax,8),%mm0\n"
1.623 + "movzbl (%esi),%eax\n"
1.624 + "paddsw 4096(%ecx,%eax,8),%mm0\n"
1.625 + "movzbl (%edx),%eax\n"
1.626 + "movq 0(%ecx,%eax,8),%mm1\n"
1.627 + "paddsw %mm0,%mm1\n"
1.628 + "psraw $0x6,%mm1\n"
1.629 + "packuswb %mm1,%mm1\n"
1.630 + "movd %mm1,0x0(%ebp)\n"
1.631 +"2:"
1.632 + "popa\n"
1.633 + "ret\n"
1.634 +#if !defined(XP_MACOSX)
1.635 + ".previous\n"
1.636 +#endif
1.637 +);
1.638 +
1.639 +void FastConvertYUVToRGB32Row(const uint8* y_buf,
1.640 + const uint8* u_buf,
1.641 + const uint8* v_buf,
1.642 + uint8* rgb_buf,
1.643 + int width)
1.644 +{
1.645 + if (mozilla::supports_sse()) {
1.646 + PICConvertYUVToRGB32Row_SSE(y_buf, u_buf, v_buf, rgb_buf, width,
1.647 + &kCoefficientsRgbY[0][0]);
1.648 + return;
1.649 + }
1.650 +
1.651 + FastConvertYUVToRGB32Row_C(y_buf, u_buf, v_buf, rgb_buf, width, 1);
1.652 +}
1.653 +
1.654 +void PICScaleYUVToRGB32Row_SSE(const uint8* y_buf,
1.655 + const uint8* u_buf,
1.656 + const uint8* v_buf,
1.657 + uint8* rgb_buf,
1.658 + int width,
1.659 + int source_dx,
1.660 + int16 *kCoefficientsRgbY);
1.661 +
1.662 + asm(
1.663 + ".text\n"
1.664 +#if defined(XP_MACOSX)
1.665 +"_PICScaleYUVToRGB32Row_SSE:\n"
1.666 +#else
1.667 +"PICScaleYUVToRGB32Row_SSE:\n"
1.668 +#endif
1.669 + "pusha\n"
1.670 + "mov 0x24(%esp),%edx\n"
1.671 + "mov 0x28(%esp),%edi\n"
1.672 + "mov 0x2c(%esp),%esi\n"
1.673 + "mov 0x30(%esp),%ebp\n"
1.674 + "mov 0x3c(%esp),%ecx\n"
1.675 + "xor %ebx,%ebx\n"
1.676 + "jmp 1f\n"
1.677 +
1.678 +"0:"
1.679 + "mov %ebx,%eax\n"
1.680 + "sar $0x11,%eax\n"
1.681 + "movzbl (%edi,%eax,1),%eax\n"
1.682 + "movq 2048(%ecx,%eax,8),%mm0\n"
1.683 + "mov %ebx,%eax\n"
1.684 + "sar $0x11,%eax\n"
1.685 + "movzbl (%esi,%eax,1),%eax\n"
1.686 + "paddsw 4096(%ecx,%eax,8),%mm0\n"
1.687 + "mov %ebx,%eax\n"
1.688 + "add 0x38(%esp),%ebx\n"
1.689 + "sar $0x10,%eax\n"
1.690 + "movzbl (%edx,%eax,1),%eax\n"
1.691 + "movq 0(%ecx,%eax,8),%mm1\n"
1.692 + "mov %ebx,%eax\n"
1.693 + "add 0x38(%esp),%ebx\n"
1.694 + "sar $0x10,%eax\n"
1.695 + "movzbl (%edx,%eax,1),%eax\n"
1.696 + "movq 0(%ecx,%eax,8),%mm2\n"
1.697 + "paddsw %mm0,%mm1\n"
1.698 + "paddsw %mm0,%mm2\n"
1.699 + "psraw $0x6,%mm1\n"
1.700 + "psraw $0x6,%mm2\n"
1.701 + "packuswb %mm2,%mm1\n"
1.702 + "movntq %mm1,0x0(%ebp)\n"
1.703 + "add $0x8,%ebp\n"
1.704 +"1:"
1.705 + "subl $0x2,0x34(%esp)\n"
1.706 + "jns 0b\n"
1.707 +
1.708 + "andl $0x1,0x34(%esp)\n"
1.709 + "je 2f\n"
1.710 +
1.711 + "mov %ebx,%eax\n"
1.712 + "sar $0x11,%eax\n"
1.713 + "movzbl (%edi,%eax,1),%eax\n"
1.714 + "movq 2048(%ecx,%eax,8),%mm0\n"
1.715 + "mov %ebx,%eax\n"
1.716 + "sar $0x11,%eax\n"
1.717 + "movzbl (%esi,%eax,1),%eax\n"
1.718 + "paddsw 4096(%ecx,%eax,8),%mm0\n"
1.719 + "mov %ebx,%eax\n"
1.720 + "sar $0x10,%eax\n"
1.721 + "movzbl (%edx,%eax,1),%eax\n"
1.722 + "movq 0(%ecx,%eax,8),%mm1\n"
1.723 + "paddsw %mm0,%mm1\n"
1.724 + "psraw $0x6,%mm1\n"
1.725 + "packuswb %mm1,%mm1\n"
1.726 + "movd %mm1,0x0(%ebp)\n"
1.727 +
1.728 +"2:"
1.729 + "popa\n"
1.730 + "ret\n"
1.731 +#if !defined(XP_MACOSX)
1.732 + ".previous\n"
1.733 +#endif
1.734 +);
1.735 +
1.736 +void ScaleYUVToRGB32Row(const uint8* y_buf,
1.737 + const uint8* u_buf,
1.738 + const uint8* v_buf,
1.739 + uint8* rgb_buf,
1.740 + int width,
1.741 + int source_dx)
1.742 +{
1.743 + if (mozilla::supports_sse()) {
1.744 + PICScaleYUVToRGB32Row_SSE(y_buf, u_buf, v_buf, rgb_buf, width, source_dx,
1.745 + &kCoefficientsRgbY[0][0]);
1.746 + return;
1.747 + }
1.748 +
1.749 + ScaleYUVToRGB32Row_C(y_buf, u_buf, v_buf, rgb_buf, width, source_dx);
1.750 +}
1.751 +
1.752 +void PICLinearScaleYUVToRGB32Row_SSE(const uint8* y_buf,
1.753 + const uint8* u_buf,
1.754 + const uint8* v_buf,
1.755 + uint8* rgb_buf,
1.756 + int width,
1.757 + int source_dx,
1.758 + int16 *kCoefficientsRgbY);
1.759 +
1.760 + asm(
1.761 + ".text\n"
1.762 +#if defined(XP_MACOSX)
1.763 +"_PICLinearScaleYUVToRGB32Row_SSE:\n"
1.764 +#else
1.765 +"PICLinearScaleYUVToRGB32Row_SSE:\n"
1.766 +#endif
1.767 + "pusha\n"
1.768 + "mov 0x24(%esp),%edx\n"
1.769 + "mov 0x30(%esp),%ebp\n"
1.770 + "mov 0x34(%esp),%ecx\n"
1.771 + "mov 0x3c(%esp),%edi\n"
1.772 + "xor %ebx,%ebx\n"
1.773 +
1.774 + // source_width = width * source_dx + ebx
1.775 + "mov 0x34(%esp), %ecx\n"
1.776 + "imull 0x38(%esp), %ecx\n"
1.777 + "mov %ecx, 0x34(%esp)\n"
1.778 +
1.779 + "mov 0x38(%esp), %ecx\n"
1.780 + "xor %ebx,%ebx\n" // x = 0
1.781 + "cmp $0x20000,%ecx\n" // if source_dx >= 2.0
1.782 + "jl 1f\n"
1.783 + "mov $0x8000,%ebx\n" // x = 0.5 for 1/2 or less
1.784 + "jmp 1f\n"
1.785 +
1.786 +"0:"
1.787 + "mov 0x28(%esp),%esi\n"
1.788 + "mov %ebx,%eax\n"
1.789 + "sar $0x11,%eax\n"
1.790 +
1.791 + "movzbl (%esi,%eax,1),%ecx\n"
1.792 + "movzbl 1(%esi,%eax,1),%esi\n"
1.793 + "mov %ebx,%eax\n"
1.794 + "andl $0x1fffe, %eax \n"
1.795 + "imul %eax, %esi \n"
1.796 + "xorl $0x1fffe, %eax \n"
1.797 + "imul %eax, %ecx \n"
1.798 + "addl %esi, %ecx \n"
1.799 + "shrl $17, %ecx \n"
1.800 + "movq 2048(%edi,%ecx,8),%mm0\n"
1.801 +
1.802 + "mov 0x2c(%esp),%esi\n"
1.803 + "mov %ebx,%eax\n"
1.804 + "sar $0x11,%eax\n"
1.805 +
1.806 + "movzbl (%esi,%eax,1),%ecx\n"
1.807 + "movzbl 1(%esi,%eax,1),%esi\n"
1.808 + "mov %ebx,%eax\n"
1.809 + "andl $0x1fffe, %eax \n"
1.810 + "imul %eax, %esi \n"
1.811 + "xorl $0x1fffe, %eax \n"
1.812 + "imul %eax, %ecx \n"
1.813 + "addl %esi, %ecx \n"
1.814 + "shrl $17, %ecx \n"
1.815 + "paddsw 4096(%edi,%ecx,8),%mm0\n"
1.816 +
1.817 + "mov %ebx,%eax\n"
1.818 + "sar $0x10,%eax\n"
1.819 + "movzbl (%edx,%eax,1),%ecx\n"
1.820 + "movzbl 1(%edx,%eax,1),%esi\n"
1.821 + "mov %ebx,%eax\n"
1.822 + "add 0x38(%esp),%ebx\n"
1.823 + "andl $0xffff, %eax \n"
1.824 + "imul %eax, %esi \n"
1.825 + "xorl $0xffff, %eax \n"
1.826 + "imul %eax, %ecx \n"
1.827 + "addl %esi, %ecx \n"
1.828 + "shrl $16, %ecx \n"
1.829 + "movq (%edi,%ecx,8),%mm1\n"
1.830 +
1.831 + "cmp 0x34(%esp), %ebx\n"
1.832 + "jge 2f\n"
1.833 +
1.834 + "mov %ebx,%eax\n"
1.835 + "sar $0x10,%eax\n"
1.836 + "movzbl (%edx,%eax,1),%ecx\n"
1.837 + "movzbl 1(%edx,%eax,1),%esi\n"
1.838 + "mov %ebx,%eax\n"
1.839 + "add 0x38(%esp),%ebx\n"
1.840 + "andl $0xffff, %eax \n"
1.841 + "imul %eax, %esi \n"
1.842 + "xorl $0xffff, %eax \n"
1.843 + "imul %eax, %ecx \n"
1.844 + "addl %esi, %ecx \n"
1.845 + "shrl $16, %ecx \n"
1.846 + "movq (%edi,%ecx,8),%mm2\n"
1.847 +
1.848 + "paddsw %mm0,%mm1\n"
1.849 + "paddsw %mm0,%mm2\n"
1.850 + "psraw $0x6,%mm1\n"
1.851 + "psraw $0x6,%mm2\n"
1.852 + "packuswb %mm2,%mm1\n"
1.853 + "movntq %mm1,0x0(%ebp)\n"
1.854 + "add $0x8,%ebp\n"
1.855 +
1.856 +"1:"
1.857 + "cmp %ebx, 0x34(%esp)\n"
1.858 + "jg 0b\n"
1.859 + "popa\n"
1.860 + "ret\n"
1.861 +
1.862 +"2:"
1.863 + "paddsw %mm0, %mm1\n"
1.864 + "psraw $6, %mm1\n"
1.865 + "packuswb %mm1, %mm1\n"
1.866 + "movd %mm1, (%ebp)\n"
1.867 + "popa\n"
1.868 + "ret\n"
1.869 +#if !defined(XP_MACOSX)
1.870 + ".previous\n"
1.871 +#endif
1.872 +);
1.873 +
1.874 +
1.875 +void LinearScaleYUVToRGB32Row(const uint8* y_buf,
1.876 + const uint8* u_buf,
1.877 + const uint8* v_buf,
1.878 + uint8* rgb_buf,
1.879 + int width,
1.880 + int source_dx)
1.881 +{
1.882 + if (mozilla::supports_sse()) {
1.883 + PICLinearScaleYUVToRGB32Row_SSE(y_buf, u_buf, v_buf, rgb_buf, width,
1.884 + source_dx, &kCoefficientsRgbY[0][0]);
1.885 + return;
1.886 + }
1.887 +
1.888 + LinearScaleYUVToRGB32Row_C(y_buf, u_buf, v_buf, rgb_buf, width, source_dx);
1.889 +}
1.890 +#else
1.891 +void FastConvertYUVToRGB32Row(const uint8* y_buf,
1.892 + const uint8* u_buf,
1.893 + const uint8* v_buf,
1.894 + uint8* rgb_buf,
1.895 + int width) {
1.896 + FastConvertYUVToRGB32Row_C(y_buf, u_buf, v_buf, rgb_buf, width, 1);
1.897 +}
1.898 +
1.899 +void ScaleYUVToRGB32Row(const uint8* y_buf,
1.900 + const uint8* u_buf,
1.901 + const uint8* v_buf,
1.902 + uint8* rgb_buf,
1.903 + int width,
1.904 + int source_dx) {
1.905 + ScaleYUVToRGB32Row_C(y_buf, u_buf, v_buf, rgb_buf, width, source_dx);
1.906 +}
1.907 +
1.908 +void LinearScaleYUVToRGB32Row(const uint8* y_buf,
1.909 + const uint8* u_buf,
1.910 + const uint8* v_buf,
1.911 + uint8* rgb_buf,
1.912 + int width,
1.913 + int source_dx) {
1.914 + LinearScaleYUVToRGB32Row_C(y_buf, u_buf, v_buf, rgb_buf, width, source_dx);
1.915 +}
1.916 +#endif
1.917 +
1.918 +}
