1 // Copyright (c) 2010 The Chromium Authors. All rights reserved.
2 // Use of this source code is governed by a BSD-style license that can be
3 // found in the LICENSE file.
4
5 // This webpage shows layout of YV12 and other YUV formats
6 // http://www.fourcc.org/yuv.php
7 // The actual conversion is best described here
8 // http://en.wikipedia.org/wiki/YUV
9 // An article on optimizing YUV conversion using tables instead of multiplies
10 // http://lestourtereaux.free.fr/papers/data/yuvrgb.pdf
11 //
12 // YV12 is a full plane of Y and a half height, half width chroma planes
13 // YV16 is a full plane of Y and a full height, half width chroma planes
14 // YV24 is a full plane of Y and a full height, full width chroma planes
15 // Y8 is a full plane of Y and no chroma planes (i.e., monochrome)
16 //
17 // ARGB pixel format is output, which on little endian is stored as BGRA.
18 // The alpha is set to 255, allowing the application to use RGBA or RGB32.
19
20 #include "yuv_convert.h"
21
22 #include "mozilla/StaticPrefs_gfx.h"
23 #include "libyuv.h"
24 #include "scale_yuv_argb.h"
25 // Header for low level row functions.
26 #include "yuv_row.h"
27 #include "mozilla/SSE.h"
28 #include "mozilla/IntegerRange.h"
29
30 namespace mozilla {
31
32 namespace gfx {
33
34 // 16.16 fixed point arithmetic
35 const int kFractionBits = 16;
36 const int kFractionMax = 1 << kFractionBits;
37 const int kFractionMask = ((1 << kFractionBits) - 1);
38
39 // clang-format off
40
TypeFromSize(int ywidth,int yheight,int cbcrwidth,int cbcrheight)41 YUVType TypeFromSize(int ywidth,
42 int yheight,
43 int cbcrwidth,
44 int cbcrheight)
45 {
46 if (ywidth == cbcrwidth && yheight == cbcrheight) {
47 return YV24;
48 }
49 else if ((ywidth + 1) / 2 == cbcrwidth && yheight == cbcrheight) {
50 return YV16;
51 }
52 else if ((ywidth + 1) / 2 == cbcrwidth && (yheight + 1) / 2 == cbcrheight) {
53 return YV12;
54 }
55 else if (cbcrwidth == 0 && cbcrheight == 0) {
56 return Y8;
57 }
58 else {
59 MOZ_CRASH("Can't determine YUV type from size");
60 }
61 }
62
FourCCFromYUVType(YUVType aYUVType)63 libyuv::FourCC FourCCFromYUVType(YUVType aYUVType) {
64 switch (aYUVType) {
65 case YV24: return libyuv::FOURCC_I444;
66 case YV16: return libyuv::FOURCC_I422;
67 case YV12: return libyuv::FOURCC_I420;
68 case Y8: return libyuv::FOURCC_I400;
69 default: return libyuv::FOURCC_ANY;
70 }
71 }
72
GBRPlanarToARGB(const uint8_t * src_y,int y_pitch,const uint8_t * src_u,int u_pitch,const uint8_t * src_v,int v_pitch,uint8_t * rgb_buf,int rgb_pitch,int pic_width,int pic_height)73 int GBRPlanarToARGB(const uint8_t* src_y, int y_pitch,
74 const uint8_t* src_u, int u_pitch,
75 const uint8_t* src_v, int v_pitch,
76 uint8_t* rgb_buf, int rgb_pitch,
77 int pic_width, int pic_height) {
78 // libyuv has no native conversion function for this
79 // fixme: replace with something less awful
80 for (const auto row : IntegerRange(pic_height)) {
81 for (const auto col : IntegerRange(pic_width)) {
82 rgb_buf[rgb_pitch * row + col * 4 + 0] = src_u[u_pitch * row + col];
83 rgb_buf[rgb_pitch * row + col * 4 + 1] = src_y[y_pitch * row + col];
84 rgb_buf[rgb_pitch * row + col * 4 + 2] = src_v[v_pitch * row + col];
85 rgb_buf[rgb_pitch * row + col * 4 + 3] = 255;
86 }
87 }
88 return 0;
89 }
90
91 // Convert a frame of YUV to 32 bit ARGB.
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,YUVColorSpace yuv_color_space,ColorRange color_range)92 void ConvertYCbCrToRGB32(const uint8* y_buf, const uint8* u_buf,
93 const uint8* v_buf, uint8* rgb_buf, int pic_x,
94 int pic_y, int pic_width, int pic_height, int y_pitch,
95 int uv_pitch, int rgb_pitch, YUVType yuv_type,
96 YUVColorSpace yuv_color_space,
97 ColorRange color_range) {
98 // Deprecated function's conversion is accurate.
99 // libyuv converion is a bit inaccurate to get performance. It dynamically
100 // calculates RGB from YUV to use simd. In it, signed byte is used for
101 // conversion's coefficient, but it requests 129. libyuv cut 129 to 127. And
102 // only 6 bits are used for a decimal part during the dynamic calculation.
103 //
104 // The function is still fast on some old intel chips.
105 // See Bug 1256475.
106 bool use_deprecated = StaticPrefs::gfx_ycbcr_accurate_conversion() ||
107 (supports_mmx() && supports_sse() && !supports_sse3() &&
108 yuv_color_space == YUVColorSpace::BT601 &&
109 color_range == ColorRange::LIMITED);
110 // The deprecated function only support BT601.
111 // See Bug 1210357.
112 if (yuv_color_space != YUVColorSpace::BT601) {
113 use_deprecated = false;
114 }
115 if (use_deprecated) {
116 ConvertYCbCrToRGB32_deprecated(y_buf, u_buf, v_buf, rgb_buf, pic_x, pic_y,
117 pic_width, pic_height, y_pitch, uv_pitch,
118 rgb_pitch, yuv_type);
119 return;
120 }
121
122 decltype(libyuv::I420ToARGBMatrix)* fConvertYUVToARGB = nullptr;
123 const uint8* src_y = nullptr;
124 const uint8* src_u = nullptr;
125 const uint8* src_v = nullptr;
126 const libyuv::YuvConstants* yuv_constant = nullptr;
127
128 switch (yuv_color_space) {
129 case YUVColorSpace::BT2020:
130 yuv_constant = color_range == ColorRange::LIMITED
131 ? &libyuv::kYuv2020Constants
132 : &libyuv::kYuvV2020Constants;
133 break;
134 case YUVColorSpace::BT709:
135 yuv_constant = color_range == ColorRange::LIMITED
136 ? &libyuv::kYuvH709Constants
137 : &libyuv::kYuvF709Constants;
138 break;
139 case YUVColorSpace::Identity:
140 MOZ_ASSERT(yuv_type == YV24, "Identity (aka RGB) with chroma subsampling is unsupported");
141 if (yuv_type == YV24) {
142 break;
143 }
144 [[fallthrough]]; // Assuming BT601 for unsupported input is better than crashing
145 default:
146 MOZ_FALLTHROUGH_ASSERT("Unsupported YUVColorSpace");
147 case YUVColorSpace::BT601:
148 yuv_constant = color_range == ColorRange::LIMITED
149 ? &libyuv::kYuvI601Constants
150 : &libyuv::kYuvJPEGConstants;
151 break;
152 }
153
154 switch (yuv_type) {
155 case YV24: {
156 src_y = y_buf + y_pitch * pic_y + pic_x;
157 src_u = u_buf + uv_pitch * pic_y + pic_x;
158 src_v = v_buf + uv_pitch * pic_y + pic_x;
159
160 if (yuv_color_space == YUVColorSpace::Identity) {
161 // Special case for RGB image
162 DebugOnly<int> err =
163 GBRPlanarToARGB(src_y, y_pitch, src_u, uv_pitch, src_v, uv_pitch,
164 rgb_buf, rgb_pitch, pic_width, pic_height);
165 MOZ_ASSERT(!err);
166 return;
167 }
168
169 fConvertYUVToARGB = libyuv::I444ToARGBMatrix;
170 break;
171 }
172 case YV16: {
173 src_y = y_buf + y_pitch * pic_y + pic_x;
174 src_u = u_buf + uv_pitch * pic_y + pic_x / 2;
175 src_v = v_buf + uv_pitch * pic_y + pic_x / 2;
176
177 fConvertYUVToARGB = libyuv::I422ToARGBMatrix;
178 break;
179 }
180 case YV12: {
181 src_y = y_buf + y_pitch * pic_y + pic_x;
182 src_u = u_buf + (uv_pitch * pic_y + pic_x) / 2;
183 src_v = v_buf + (uv_pitch * pic_y + pic_x) / 2;
184
185 fConvertYUVToARGB = libyuv::I420ToARGBMatrix;
186 break;
187 }
188 case Y8: {
189 src_y = y_buf + y_pitch * pic_y + pic_x;
190 MOZ_ASSERT(u_buf == nullptr);
191 MOZ_ASSERT(v_buf == nullptr);
192
193 if (color_range == ColorRange::LIMITED) {
194 DebugOnly<int> err =
195 libyuv::I400ToARGB(src_y, y_pitch, rgb_buf, rgb_pitch, pic_width,
196 pic_height);
197 MOZ_ASSERT(!err);
198 } else {
199 DebugOnly<int> err =
200 libyuv::J400ToARGB(src_y, y_pitch, rgb_buf, rgb_pitch, pic_width,
201 pic_height);
202 MOZ_ASSERT(!err);
203 }
204
205 return;
206 }
207 default:
208 MOZ_ASSERT_UNREACHABLE("Unsupported YUV type");
209 }
210
211 DebugOnly<int> err =
212 fConvertYUVToARGB(src_y, y_pitch, src_u, uv_pitch, src_v, uv_pitch,
213 rgb_buf, rgb_pitch, yuv_constant, pic_width, pic_height);
214 MOZ_ASSERT(!err);
215 }
216
217 // Convert a frame of YUV to 32 bit ARGB.
ConvertYCbCrToRGB32_deprecated(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)218 void ConvertYCbCrToRGB32_deprecated(const uint8* y_buf,
219 const uint8* u_buf,
220 const uint8* v_buf,
221 uint8* rgb_buf,
222 int pic_x,
223 int pic_y,
224 int pic_width,
225 int pic_height,
226 int y_pitch,
227 int uv_pitch,
228 int rgb_pitch,
229 YUVType yuv_type) {
230 unsigned int y_shift = yuv_type == YV12 ? 1 : 0;
231 unsigned int x_shift = yuv_type == YV24 ? 0 : 1;
232 // Test for SSE because the optimized code uses movntq, which is not part of MMX.
233 bool has_sse = supports_mmx() && supports_sse();
234 // There is no optimized YV24 SSE routine so we check for this and
235 // fall back to the C code.
236 has_sse &= yuv_type != YV24;
237 bool odd_pic_x = yuv_type != YV24 && pic_x % 2 != 0;
238 int x_width = odd_pic_x ? pic_width - 1 : pic_width;
239
240 for (int y = pic_y; y < pic_height + pic_y; ++y) {
241 uint8* rgb_row = rgb_buf + (y - pic_y) * rgb_pitch;
242 const uint8* y_ptr = y_buf + y * y_pitch + pic_x;
243 const uint8* u_ptr = u_buf + (y >> y_shift) * uv_pitch + (pic_x >> x_shift);
244 const uint8* v_ptr = v_buf + (y >> y_shift) * uv_pitch + (pic_x >> x_shift);
245
246 if (odd_pic_x) {
247 // Handle the single odd pixel manually and use the
248 // fast routines for the remaining.
249 FastConvertYUVToRGB32Row_C(y_ptr++,
250 u_ptr++,
251 v_ptr++,
252 rgb_row,
253 1,
254 x_shift);
255 rgb_row += 4;
256 }
257
258 if (has_sse) {
259 FastConvertYUVToRGB32Row(y_ptr,
260 u_ptr,
261 v_ptr,
262 rgb_row,
263 x_width);
264 }
265 else {
266 FastConvertYUVToRGB32Row_C(y_ptr,
267 u_ptr,
268 v_ptr,
269 rgb_row,
270 x_width,
271 x_shift);
272 }
273 }
274
275 // MMX used for FastConvertYUVToRGB32Row requires emms instruction.
276 if (has_sse)
277 EMMS();
278 }
279
280 // C version does 8 at a time to mimic MMX code
FilterRows_C(uint8 * ybuf,const uint8 * y0_ptr,const uint8 * y1_ptr,int source_width,int source_y_fraction)281 static void FilterRows_C(uint8* ybuf, const uint8* y0_ptr, const uint8* y1_ptr,
282 int source_width, int source_y_fraction) {
283 int y1_fraction = source_y_fraction;
284 int y0_fraction = 256 - y1_fraction;
285 uint8* end = ybuf + source_width;
286 do {
287 ybuf[0] = (y0_ptr[0] * y0_fraction + y1_ptr[0] * y1_fraction) >> 8;
288 ybuf[1] = (y0_ptr[1] * y0_fraction + y1_ptr[1] * y1_fraction) >> 8;
289 ybuf[2] = (y0_ptr[2] * y0_fraction + y1_ptr[2] * y1_fraction) >> 8;
290 ybuf[3] = (y0_ptr[3] * y0_fraction + y1_ptr[3] * y1_fraction) >> 8;
291 ybuf[4] = (y0_ptr[4] * y0_fraction + y1_ptr[4] * y1_fraction) >> 8;
292 ybuf[5] = (y0_ptr[5] * y0_fraction + y1_ptr[5] * y1_fraction) >> 8;
293 ybuf[6] = (y0_ptr[6] * y0_fraction + y1_ptr[6] * y1_fraction) >> 8;
294 ybuf[7] = (y0_ptr[7] * y0_fraction + y1_ptr[7] * y1_fraction) >> 8;
295 y0_ptr += 8;
296 y1_ptr += 8;
297 ybuf += 8;
298 } while (ybuf < end);
299 }
300
301 #ifdef MOZILLA_MAY_SUPPORT_MMX
302 void FilterRows_MMX(uint8* ybuf, const uint8* y0_ptr, const uint8* y1_ptr,
303 int source_width, int source_y_fraction);
304 #endif
305
306 #ifdef MOZILLA_MAY_SUPPORT_SSE2
307 void FilterRows_SSE2(uint8* ybuf, const uint8* y0_ptr, const uint8* y1_ptr,
308 int source_width, int source_y_fraction);
309 #endif
310
FilterRows(uint8 * ybuf,const uint8 * y0_ptr,const uint8 * y1_ptr,int source_width,int source_y_fraction)311 static inline void FilterRows(uint8* ybuf, const uint8* y0_ptr,
312 const uint8* y1_ptr, int source_width,
313 int source_y_fraction) {
314 #ifdef MOZILLA_MAY_SUPPORT_SSE2
315 if (mozilla::supports_sse2()) {
316 FilterRows_SSE2(ybuf, y0_ptr, y1_ptr, source_width, source_y_fraction);
317 return;
318 }
319 #endif
320
321 #ifdef MOZILLA_MAY_SUPPORT_MMX
322 if (mozilla::supports_mmx()) {
323 FilterRows_MMX(ybuf, y0_ptr, y1_ptr, source_width, source_y_fraction);
324 return;
325 }
326 #endif
327
328 FilterRows_C(ybuf, y0_ptr, y1_ptr, source_width, source_y_fraction);
329 }
330
331
332 // Scale a frame of YUV to 32 bit ARGB.
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,YUVColorSpace yuv_color_space,ScaleFilter filter)333 void ScaleYCbCrToRGB32(const uint8* y_buf,
334 const uint8* u_buf,
335 const uint8* v_buf,
336 uint8* rgb_buf,
337 int source_width,
338 int source_height,
339 int width,
340 int height,
341 int y_pitch,
342 int uv_pitch,
343 int rgb_pitch,
344 YUVType yuv_type,
345 YUVColorSpace yuv_color_space,
346 ScaleFilter filter) {
347 bool use_deprecated =
348 StaticPrefs::gfx_ycbcr_accurate_conversion() ||
349 #if defined(XP_WIN) && defined(_M_X64)
350 // libyuv does not support SIMD scaling on win 64bit. See Bug 1295927.
351 supports_sse3() ||
352 #endif
353 (supports_mmx() && supports_sse() && !supports_sse3());
354 // The deprecated function only support BT601.
355 // See Bug 1210357.
356 if (yuv_color_space != YUVColorSpace::BT601) {
357 use_deprecated = false;
358 }
359 if (use_deprecated) {
360 ScaleYCbCrToRGB32_deprecated(y_buf, u_buf, v_buf,
361 rgb_buf,
362 source_width, source_height,
363 width, height,
364 y_pitch, uv_pitch,
365 rgb_pitch,
366 yuv_type,
367 ROTATE_0,
368 filter);
369 return;
370 }
371
372 DebugOnly<int> err =
373 libyuv::YUVToARGBScale(y_buf, y_pitch,
374 u_buf, uv_pitch,
375 v_buf, uv_pitch,
376 FourCCFromYUVType(yuv_type),
377 yuv_color_space,
378 source_width, source_height,
379 rgb_buf, rgb_pitch,
380 width, height,
381 libyuv::kFilterBilinear);
382 MOZ_ASSERT(!err);
383 return;
384 }
385
386 // Scale a frame of YUV to 32 bit ARGB.
ScaleYCbCrToRGB32_deprecated(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)387 void ScaleYCbCrToRGB32_deprecated(const uint8* y_buf,
388 const uint8* u_buf,
389 const uint8* v_buf,
390 uint8* rgb_buf,
391 int source_width,
392 int source_height,
393 int width,
394 int height,
395 int y_pitch,
396 int uv_pitch,
397 int rgb_pitch,
398 YUVType yuv_type,
399 Rotate view_rotate,
400 ScaleFilter filter) {
401 bool has_mmx = supports_mmx();
402
403 // 4096 allows 3 buffers to fit in 12k.
404 // Helps performance on CPU with 16K L1 cache.
405 // Large enough for 3830x2160 and 30" displays which are 2560x1600.
406 const int kFilterBufferSize = 4096;
407 // Disable filtering if the screen is too big (to avoid buffer overflows).
408 // This should never happen to regular users: they don't have monitors
409 // wider than 4096 pixels.
410 // TODO(fbarchard): Allow rotated videos to filter.
411 if (source_width > kFilterBufferSize || view_rotate)
412 filter = FILTER_NONE;
413
414 unsigned int y_shift = yuv_type == YV12 ? 1 : 0;
415 // Diagram showing origin and direction of source sampling.
416 // ->0 4<-
417 // 7 3
418 //
419 // 6 5
420 // ->1 2<-
421 // Rotations that start at right side of image.
422 if ((view_rotate == ROTATE_180) ||
423 (view_rotate == ROTATE_270) ||
424 (view_rotate == MIRROR_ROTATE_0) ||
425 (view_rotate == MIRROR_ROTATE_90)) {
426 y_buf += source_width - 1;
427 u_buf += source_width / 2 - 1;
428 v_buf += source_width / 2 - 1;
429 source_width = -source_width;
430 }
431 // Rotations that start at bottom of image.
432 if ((view_rotate == ROTATE_90) ||
433 (view_rotate == ROTATE_180) ||
434 (view_rotate == MIRROR_ROTATE_90) ||
435 (view_rotate == MIRROR_ROTATE_180)) {
436 y_buf += (source_height - 1) * y_pitch;
437 u_buf += ((source_height >> y_shift) - 1) * uv_pitch;
438 v_buf += ((source_height >> y_shift) - 1) * uv_pitch;
439 source_height = -source_height;
440 }
441
442 // Handle zero sized destination.
443 if (width == 0 || height == 0)
444 return;
445 int source_dx = source_width * kFractionMax / width;
446 int source_dy = source_height * kFractionMax / height;
447 int source_dx_uv = source_dx;
448
449 if ((view_rotate == ROTATE_90) ||
450 (view_rotate == ROTATE_270)) {
451 int tmp = height;
452 height = width;
453 width = tmp;
454 tmp = source_height;
455 source_height = source_width;
456 source_width = tmp;
457 int original_dx = source_dx;
458 int original_dy = source_dy;
459 source_dx = ((original_dy >> kFractionBits) * y_pitch) << kFractionBits;
460 source_dx_uv = ((original_dy >> kFractionBits) * uv_pitch) << kFractionBits;
461 source_dy = original_dx;
462 if (view_rotate == ROTATE_90) {
463 y_pitch = -1;
464 uv_pitch = -1;
465 source_height = -source_height;
466 } else {
467 y_pitch = 1;
468 uv_pitch = 1;
469 }
470 }
471
472 // Need padding because FilterRows() will write 1 to 16 extra pixels
473 // after the end for SSE2 version.
474 uint8 yuvbuf[16 + kFilterBufferSize * 3 + 16];
475 uint8* ybuf =
476 reinterpret_cast<uint8*>(reinterpret_cast<uintptr_t>(yuvbuf + 15) & ~15);
477 uint8* ubuf = ybuf + kFilterBufferSize;
478 uint8* vbuf = ubuf + kFilterBufferSize;
479 // TODO(fbarchard): Fixed point math is off by 1 on negatives.
480 int yscale_fixed = (source_height << kFractionBits) / height;
481
482 // TODO(fbarchard): Split this into separate function for better efficiency.
483 for (int y = 0; y < height; ++y) {
484 uint8* dest_pixel = rgb_buf + y * rgb_pitch;
485 int source_y_subpixel = (y * yscale_fixed);
486 if (yscale_fixed >= (kFractionMax * 2)) {
487 source_y_subpixel += kFractionMax / 2; // For 1/2 or less, center filter.
488 }
489 int source_y = source_y_subpixel >> kFractionBits;
490
491 const uint8* y0_ptr = y_buf + source_y * y_pitch;
492 const uint8* y1_ptr = y0_ptr + y_pitch;
493
494 const uint8* u0_ptr = u_buf + (source_y >> y_shift) * uv_pitch;
495 const uint8* u1_ptr = u0_ptr + uv_pitch;
496 const uint8* v0_ptr = v_buf + (source_y >> y_shift) * uv_pitch;
497 const uint8* v1_ptr = v0_ptr + uv_pitch;
498
499 // vertical scaler uses 16.8 fixed point
500 int source_y_fraction = (source_y_subpixel & kFractionMask) >> 8;
501 int source_uv_fraction =
502 ((source_y_subpixel >> y_shift) & kFractionMask) >> 8;
503
504 const uint8* y_ptr = y0_ptr;
505 const uint8* u_ptr = u0_ptr;
506 const uint8* v_ptr = v0_ptr;
507 // Apply vertical filtering if necessary.
508 // TODO(fbarchard): Remove memcpy when not necessary.
509 if (filter & mozilla::gfx::FILTER_BILINEAR_V) {
510 if (yscale_fixed != kFractionMax &&
511 source_y_fraction && ((source_y + 1) < source_height)) {
512 FilterRows(ybuf, y0_ptr, y1_ptr, source_width, source_y_fraction);
513 } else {
514 memcpy(ybuf, y0_ptr, source_width);
515 }
516 y_ptr = ybuf;
517 ybuf[source_width] = ybuf[source_width-1];
518 int uv_source_width = (source_width + 1) / 2;
519 if (yscale_fixed != kFractionMax &&
520 source_uv_fraction &&
521 (((source_y >> y_shift) + 1) < (source_height >> y_shift))) {
522 FilterRows(ubuf, u0_ptr, u1_ptr, uv_source_width, source_uv_fraction);
523 FilterRows(vbuf, v0_ptr, v1_ptr, uv_source_width, source_uv_fraction);
524 } else {
525 memcpy(ubuf, u0_ptr, uv_source_width);
526 memcpy(vbuf, v0_ptr, uv_source_width);
527 }
528 u_ptr = ubuf;
529 v_ptr = vbuf;
530 ubuf[uv_source_width] = ubuf[uv_source_width - 1];
531 vbuf[uv_source_width] = vbuf[uv_source_width - 1];
532 }
533 if (source_dx == kFractionMax) { // Not scaled
534 FastConvertYUVToRGB32Row(y_ptr, u_ptr, v_ptr,
535 dest_pixel, width);
536 } else if (filter & FILTER_BILINEAR_H) {
537 LinearScaleYUVToRGB32Row(y_ptr, u_ptr, v_ptr,
538 dest_pixel, width, source_dx);
539 } else {
540 // Specialized scalers and rotation.
541 #if defined(MOZILLA_MAY_SUPPORT_SSE) && defined(_MSC_VER) && defined(_M_IX86) && !defined(__clang__)
542 if(mozilla::supports_sse()) {
543 if (width == (source_width * 2)) {
544 DoubleYUVToRGB32Row_SSE(y_ptr, u_ptr, v_ptr,
545 dest_pixel, width);
546 } else if ((source_dx & kFractionMask) == 0) {
547 // Scaling by integer scale factor. ie half.
548 ConvertYUVToRGB32Row_SSE(y_ptr, u_ptr, v_ptr,
549 dest_pixel, width,
550 source_dx >> kFractionBits);
551 } else if (source_dx_uv == source_dx) { // Not rotated.
552 ScaleYUVToRGB32Row(y_ptr, u_ptr, v_ptr,
553 dest_pixel, width, source_dx);
554 } else {
555 RotateConvertYUVToRGB32Row_SSE(y_ptr, u_ptr, v_ptr,
556 dest_pixel, width,
557 source_dx >> kFractionBits,
558 source_dx_uv >> kFractionBits);
559 }
560 }
561 else {
562 ScaleYUVToRGB32Row_C(y_ptr, u_ptr, v_ptr,
563 dest_pixel, width, source_dx);
564 }
565 #else
566 (void)source_dx_uv;
567 ScaleYUVToRGB32Row(y_ptr, u_ptr, v_ptr,
568 dest_pixel, width, source_dx);
569 #endif
570 }
571 }
572 // MMX used for FastConvertYUVToRGB32Row and FilterRows requires emms.
573 if (has_mmx)
574 EMMS();
575 }
ConvertI420AlphaToARGB32(const uint8 * y_buf,const uint8 * u_buf,const uint8 * v_buf,const uint8 * a_buf,uint8 * argb_buf,int pic_width,int pic_height,int ya_pitch,int uv_pitch,int argb_pitch)576 void ConvertI420AlphaToARGB32(const uint8* y_buf,
577 const uint8* u_buf,
578 const uint8* v_buf,
579 const uint8* a_buf,
580 uint8* argb_buf,
581 int pic_width,
582 int pic_height,
583 int ya_pitch,
584 int uv_pitch,
585 int argb_pitch) {
586
587 // The downstream graphics stack expects an attenuated input, hence why the
588 // attenuation parameter is set.
589 DebugOnly<int> err = libyuv::I420AlphaToARGB(y_buf, ya_pitch,
590 u_buf, uv_pitch,
591 v_buf, uv_pitch,
592 a_buf, ya_pitch,
593 argb_buf, argb_pitch,
594 pic_width, pic_height, 1);
595 MOZ_ASSERT(!err);
596 }
597
598 } // namespace gfx
599 } // namespace mozilla
600