2 * Copyright (C) 2001-2011 Michael Niedermayer <michaelni@gmx.at>
4 * This file is part of FFmpeg.
6 * FFmpeg is free software; you can redistribute it and/or
7 * modify it under the terms of the GNU Lesser General Public
8 * License as published by the Free Software Foundation; either
9 * version 2.1 of the License, or (at your option) any later version.
11 * FFmpeg is distributed in the hope that it will be useful,
12 * but WITHOUT ANY WARRANTY; without even the implied warranty of
13 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
14 * Lesser General Public License for more details.
16 * You should have received a copy of the GNU Lesser General Public
17 * License along with FFmpeg; if not, write to the Free Software
18 * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
26 #include "libavutil/avassert.h"
27 #include "libavutil/avutil.h"
28 #include "libavutil/bswap.h"
29 #include "libavutil/cpu.h"
30 #include "libavutil/imgutils.h"
31 #include "libavutil/intreadwrite.h"
32 #include "libavutil/mathematics.h"
33 #include "libavutil/pixdesc.h"
36 #include "swscale_internal.h"
39 DECLARE_ALIGNED(8, const uint8_t, ff_dither_8x8_128)[9][8] = {
40 { 36, 68, 60, 92, 34, 66, 58, 90, },
41 { 100, 4, 124, 28, 98, 2, 122, 26, },
42 { 52, 84, 44, 76, 50, 82, 42, 74, },
43 { 116, 20, 108, 12, 114, 18, 106, 10, },
44 { 32, 64, 56, 88, 38, 70, 62, 94, },
45 { 96, 0, 120, 24, 102, 6, 126, 30, },
46 { 48, 80, 40, 72, 54, 86, 46, 78, },
47 { 112, 16, 104, 8, 118, 22, 110, 14, },
48 { 36, 68, 60, 92, 34, 66, 58, 90, },
51 DECLARE_ALIGNED(8, static const uint8_t, sws_pb_64)[8] = {
52 64, 64, 64, 64, 64, 64, 64, 64
55 static av_always_inline void fillPlane(uint8_t *plane, int stride, int width,
56 int height, int y, uint8_t val)
59 uint8_t *ptr = plane + stride * y;
60 for (i = 0; i < height; i++) {
61 memset(ptr, val, width);
66 static void hScale16To19_c(SwsContext *c, int16_t *_dst, int dstW,
67 const uint8_t *_src, const int16_t *filter,
68 const int32_t *filterPos, int filterSize)
70 const AVPixFmtDescriptor *desc = av_pix_fmt_desc_get(c->srcFormat);
72 int32_t *dst = (int32_t *) _dst;
73 const uint16_t *src = (const uint16_t *) _src;
74 int bits = desc->comp[0].depth - 1;
77 if ((isAnyRGB(c->srcFormat) || c->srcFormat==AV_PIX_FMT_PAL8) && desc->comp[0].depth<16) {
79 } else if (desc->flags & AV_PIX_FMT_FLAG_FLOAT) { /* float input are process like uint 16bpc */
83 for (i = 0; i < dstW; i++) {
85 int srcPos = filterPos[i];
88 for (j = 0; j < filterSize; j++) {
89 val += src[srcPos + j] * filter[filterSize * i + j];
91 // filter=14 bit, input=16 bit, output=30 bit, >> 11 makes 19 bit
92 dst[i] = FFMIN(val >> sh, (1 << 19) - 1);
96 static void hScale16To15_c(SwsContext *c, int16_t *dst, int dstW,
97 const uint8_t *_src, const int16_t *filter,
98 const int32_t *filterPos, int filterSize)
100 const AVPixFmtDescriptor *desc = av_pix_fmt_desc_get(c->srcFormat);
102 const uint16_t *src = (const uint16_t *) _src;
103 int sh = desc->comp[0].depth - 1;
106 sh = isAnyRGB(c->srcFormat) || c->srcFormat==AV_PIX_FMT_PAL8 ? 13 : (desc->comp[0].depth - 1);
107 } else if (desc->flags & AV_PIX_FMT_FLAG_FLOAT) { /* float input are process like uint 16bpc */
111 for (i = 0; i < dstW; i++) {
113 int srcPos = filterPos[i];
116 for (j = 0; j < filterSize; j++) {
117 val += src[srcPos + j] * filter[filterSize * i + j];
119 // filter=14 bit, input=16 bit, output=30 bit, >> 15 makes 15 bit
120 dst[i] = FFMIN(val >> sh, (1 << 15) - 1);
124 // bilinear / bicubic scaling
125 static void hScale8To15_c(SwsContext *c, int16_t *dst, int dstW,
126 const uint8_t *src, const int16_t *filter,
127 const int32_t *filterPos, int filterSize)
130 for (i = 0; i < dstW; i++) {
132 int srcPos = filterPos[i];
134 for (j = 0; j < filterSize; j++) {
135 val += ((int)src[srcPos + j]) * filter[filterSize * i + j];
137 dst[i] = FFMIN(val >> 7, (1 << 15) - 1); // the cubic equation does overflow ...
141 static void hScale8To19_c(SwsContext *c, int16_t *_dst, int dstW,
142 const uint8_t *src, const int16_t *filter,
143 const int32_t *filterPos, int filterSize)
146 int32_t *dst = (int32_t *) _dst;
147 for (i = 0; i < dstW; i++) {
149 int srcPos = filterPos[i];
151 for (j = 0; j < filterSize; j++) {
152 val += ((int)src[srcPos + j]) * filter[filterSize * i + j];
154 dst[i] = FFMIN(val >> 3, (1 << 19) - 1); // the cubic equation does overflow ...
158 // FIXME all pal and rgb srcFormats could do this conversion as well
159 // FIXME all scalers more complex than bilinear could do half of this transform
160 static void chrRangeToJpeg_c(int16_t *dstU, int16_t *dstV, int width)
163 for (i = 0; i < width; i++) {
164 dstU[i] = (FFMIN(dstU[i], 30775) * 4663 - 9289992) >> 12; // -264
165 dstV[i] = (FFMIN(dstV[i], 30775) * 4663 - 9289992) >> 12; // -264
169 static void chrRangeFromJpeg_c(int16_t *dstU, int16_t *dstV, int width)
172 for (i = 0; i < width; i++) {
173 dstU[i] = (dstU[i] * 1799 + 4081085) >> 11; // 1469
174 dstV[i] = (dstV[i] * 1799 + 4081085) >> 11; // 1469
178 static void lumRangeToJpeg_c(int16_t *dst, int width)
181 for (i = 0; i < width; i++)
182 dst[i] = (FFMIN(dst[i], 30189) * 19077 - 39057361) >> 14;
185 static void lumRangeFromJpeg_c(int16_t *dst, int width)
188 for (i = 0; i < width; i++)
189 dst[i] = (dst[i] * 14071 + 33561947) >> 14;
192 static void chrRangeToJpeg16_c(int16_t *_dstU, int16_t *_dstV, int width)
195 int32_t *dstU = (int32_t *) _dstU;
196 int32_t *dstV = (int32_t *) _dstV;
197 for (i = 0; i < width; i++) {
198 dstU[i] = (FFMIN(dstU[i], 30775 << 4) * 4663 - (9289992 << 4)) >> 12; // -264
199 dstV[i] = (FFMIN(dstV[i], 30775 << 4) * 4663 - (9289992 << 4)) >> 12; // -264
203 static void chrRangeFromJpeg16_c(int16_t *_dstU, int16_t *_dstV, int width)
206 int32_t *dstU = (int32_t *) _dstU;
207 int32_t *dstV = (int32_t *) _dstV;
208 for (i = 0; i < width; i++) {
209 dstU[i] = (dstU[i] * 1799 + (4081085 << 4)) >> 11; // 1469
210 dstV[i] = (dstV[i] * 1799 + (4081085 << 4)) >> 11; // 1469
214 static void lumRangeToJpeg16_c(int16_t *_dst, int width)
217 int32_t *dst = (int32_t *) _dst;
218 for (i = 0; i < width; i++) {
219 dst[i] = ((int)(FFMIN(dst[i], 30189 << 4) * 4769U - (39057361 << 2))) >> 12;
223 static void lumRangeFromJpeg16_c(int16_t *_dst, int width)
226 int32_t *dst = (int32_t *) _dst;
227 for (i = 0; i < width; i++)
228 dst[i] = (dst[i]*(14071/4) + (33561947<<4)/4)>>12;
232 #define DEBUG_SWSCALE_BUFFERS 0
233 #define DEBUG_BUFFERS(...) \
234 if (DEBUG_SWSCALE_BUFFERS) \
235 av_log(c, AV_LOG_DEBUG, __VA_ARGS__)
237 static int swscale(SwsContext *c, const uint8_t *src[],
238 int srcStride[], int srcSliceY,
239 int srcSliceH, uint8_t *dst[], int dstStride[])
241 /* load a few things into local vars to make the code more readable?
243 const int dstW = c->dstW;
244 const int dstH = c->dstH;
246 const enum AVPixelFormat dstFormat = c->dstFormat;
247 const int flags = c->flags;
248 int32_t *vLumFilterPos = c->vLumFilterPos;
249 int32_t *vChrFilterPos = c->vChrFilterPos;
251 const int vLumFilterSize = c->vLumFilterSize;
252 const int vChrFilterSize = c->vChrFilterSize;
254 yuv2planar1_fn yuv2plane1 = c->yuv2plane1;
255 yuv2planarX_fn yuv2planeX = c->yuv2planeX;
256 yuv2interleavedX_fn yuv2nv12cX = c->yuv2nv12cX;
257 yuv2packed1_fn yuv2packed1 = c->yuv2packed1;
258 yuv2packed2_fn yuv2packed2 = c->yuv2packed2;
259 yuv2packedX_fn yuv2packedX = c->yuv2packedX;
260 yuv2anyX_fn yuv2anyX = c->yuv2anyX;
261 const int chrSrcSliceY = srcSliceY >> c->chrSrcVSubSample;
262 const int chrSrcSliceH = AV_CEIL_RSHIFT(srcSliceH, c->chrSrcVSubSample);
263 int should_dither = isNBPS(c->srcFormat) ||
264 is16BPS(c->srcFormat);
267 /* vars which will change and which we need to store back in the context */
269 int lumBufIndex = c->lumBufIndex;
270 int chrBufIndex = c->chrBufIndex;
271 int lastInLumBuf = c->lastInLumBuf;
272 int lastInChrBuf = c->lastInChrBuf;
275 int lumEnd = c->descIndex[0];
276 int chrStart = lumEnd;
277 int chrEnd = c->descIndex[1];
279 int vEnd = c->numDesc;
280 SwsSlice *src_slice = &c->slice[lumStart];
281 SwsSlice *hout_slice = &c->slice[c->numSlice-2];
282 SwsSlice *vout_slice = &c->slice[c->numSlice-1];
283 SwsFilterDescriptor *desc = c->desc;
285 int needAlpha = c->needAlpha;
290 if (isPacked(c->srcFormat)) {
296 srcStride[3] = srcStride[0];
298 srcStride[1] <<= c->vChrDrop;
299 srcStride[2] <<= c->vChrDrop;
301 DEBUG_BUFFERS("swscale() %p[%d] %p[%d] %p[%d] %p[%d] -> %p[%d] %p[%d] %p[%d] %p[%d]\n",
302 src[0], srcStride[0], src[1], srcStride[1],
303 src[2], srcStride[2], src[3], srcStride[3],
304 dst[0], dstStride[0], dst[1], dstStride[1],
305 dst[2], dstStride[2], dst[3], dstStride[3]);
306 DEBUG_BUFFERS("srcSliceY: %d srcSliceH: %d dstY: %d dstH: %d\n",
307 srcSliceY, srcSliceH, dstY, dstH);
308 DEBUG_BUFFERS("vLumFilterSize: %d vChrFilterSize: %d\n",
309 vLumFilterSize, vChrFilterSize);
311 if (dstStride[0]&15 || dstStride[1]&15 ||
312 dstStride[2]&15 || dstStride[3]&15) {
313 static int warnedAlready = 0; // FIXME maybe move this into the context
314 if (flags & SWS_PRINT_INFO && !warnedAlready) {
315 av_log(c, AV_LOG_WARNING,
316 "Warning: dstStride is not aligned!\n"
317 " ->cannot do aligned memory accesses anymore\n");
322 if ( (uintptr_t)dst[0]&15 || (uintptr_t)dst[1]&15 || (uintptr_t)dst[2]&15
323 || (uintptr_t)src[0]&15 || (uintptr_t)src[1]&15 || (uintptr_t)src[2]&15
324 || dstStride[0]&15 || dstStride[1]&15 || dstStride[2]&15 || dstStride[3]&15
325 || srcStride[0]&15 || srcStride[1]&15 || srcStride[2]&15 || srcStride[3]&15
327 static int warnedAlready=0;
328 int cpu_flags = av_get_cpu_flags();
329 if (HAVE_MMXEXT && (cpu_flags & AV_CPU_FLAG_SSE2) && !warnedAlready){
330 av_log(c, AV_LOG_WARNING, "Warning: data is not aligned! This can lead to a speed loss\n");
335 /* Note the user might start scaling the picture in the middle so this
336 * will not get executed. This is not really intended but works
337 * currently, so people might do it. */
338 if (srcSliceY == 0) {
346 if (!should_dither) {
347 c->chrDither8 = c->lumDither8 = sws_pb_64;
351 ff_init_vscale_pfn(c, yuv2plane1, yuv2planeX, yuv2nv12cX,
352 yuv2packed1, yuv2packed2, yuv2packedX, yuv2anyX, c->use_mmx_vfilter);
354 ff_init_slice_from_src(src_slice, (uint8_t**)src, srcStride, c->srcW,
355 srcSliceY, srcSliceH, chrSrcSliceY, chrSrcSliceH, 1);
357 ff_init_slice_from_src(vout_slice, (uint8_t**)dst, dstStride, c->dstW,
358 dstY, dstH, dstY >> c->chrDstVSubSample,
359 AV_CEIL_RSHIFT(dstH, c->chrDstVSubSample), 0);
360 if (srcSliceY == 0) {
361 hout_slice->plane[0].sliceY = lastInLumBuf + 1;
362 hout_slice->plane[1].sliceY = lastInChrBuf + 1;
363 hout_slice->plane[2].sliceY = lastInChrBuf + 1;
364 hout_slice->plane[3].sliceY = lastInLumBuf + 1;
366 hout_slice->plane[0].sliceH =
367 hout_slice->plane[1].sliceH =
368 hout_slice->plane[2].sliceH =
369 hout_slice->plane[3].sliceH = 0;
370 hout_slice->width = dstW;
373 for (; dstY < dstH; dstY++) {
374 const int chrDstY = dstY >> c->chrDstVSubSample;
375 int use_mmx_vfilter= c->use_mmx_vfilter;
377 // First line needed as input
378 const int firstLumSrcY = FFMAX(1 - vLumFilterSize, vLumFilterPos[dstY]);
379 const int firstLumSrcY2 = FFMAX(1 - vLumFilterSize, vLumFilterPos[FFMIN(dstY | ((1 << c->chrDstVSubSample) - 1), dstH - 1)]);
380 // First line needed as input
381 const int firstChrSrcY = FFMAX(1 - vChrFilterSize, vChrFilterPos[chrDstY]);
383 // Last line needed as input
384 int lastLumSrcY = FFMIN(c->srcH, firstLumSrcY + vLumFilterSize) - 1;
385 int lastLumSrcY2 = FFMIN(c->srcH, firstLumSrcY2 + vLumFilterSize) - 1;
386 int lastChrSrcY = FFMIN(c->chrSrcH, firstChrSrcY + vChrFilterSize) - 1;
390 int posY, cPosY, firstPosY, lastPosY, firstCPosY, lastCPosY;
392 // handle holes (FAST_BILINEAR & weird filters)
393 if (firstLumSrcY > lastInLumBuf) {
395 hasLumHoles = lastInLumBuf != firstLumSrcY - 1;
397 hout_slice->plane[0].sliceY = firstLumSrcY;
398 hout_slice->plane[3].sliceY = firstLumSrcY;
399 hout_slice->plane[0].sliceH =
400 hout_slice->plane[3].sliceH = 0;
403 lastInLumBuf = firstLumSrcY - 1;
405 if (firstChrSrcY > lastInChrBuf) {
407 hasChrHoles = lastInChrBuf != firstChrSrcY - 1;
409 hout_slice->plane[1].sliceY = firstChrSrcY;
410 hout_slice->plane[2].sliceY = firstChrSrcY;
411 hout_slice->plane[1].sliceH =
412 hout_slice->plane[2].sliceH = 0;
415 lastInChrBuf = firstChrSrcY - 1;
418 DEBUG_BUFFERS("dstY: %d\n", dstY);
419 DEBUG_BUFFERS("\tfirstLumSrcY: %d lastLumSrcY: %d lastInLumBuf: %d\n",
420 firstLumSrcY, lastLumSrcY, lastInLumBuf);
421 DEBUG_BUFFERS("\tfirstChrSrcY: %d lastChrSrcY: %d lastInChrBuf: %d\n",
422 firstChrSrcY, lastChrSrcY, lastInChrBuf);
424 // Do we have enough lines in this slice to output the dstY line
425 enough_lines = lastLumSrcY2 < srcSliceY + srcSliceH &&
426 lastChrSrcY < AV_CEIL_RSHIFT(srcSliceY + srcSliceH, c->chrSrcVSubSample);
429 lastLumSrcY = srcSliceY + srcSliceH - 1;
430 lastChrSrcY = chrSrcSliceY + chrSrcSliceH - 1;
431 DEBUG_BUFFERS("buffering slice: lastLumSrcY %d lastChrSrcY %d\n",
432 lastLumSrcY, lastChrSrcY);
435 av_assert0((lastLumSrcY - firstLumSrcY + 1) <= hout_slice->plane[0].available_lines);
436 av_assert0((lastChrSrcY - firstChrSrcY + 1) <= hout_slice->plane[1].available_lines);
439 posY = hout_slice->plane[0].sliceY + hout_slice->plane[0].sliceH;
440 if (posY <= lastLumSrcY && !hasLumHoles) {
441 firstPosY = FFMAX(firstLumSrcY, posY);
442 lastPosY = FFMIN(firstLumSrcY + hout_slice->plane[0].available_lines - 1, srcSliceY + srcSliceH - 1);
445 lastPosY = lastLumSrcY;
448 cPosY = hout_slice->plane[1].sliceY + hout_slice->plane[1].sliceH;
449 if (cPosY <= lastChrSrcY && !hasChrHoles) {
450 firstCPosY = FFMAX(firstChrSrcY, cPosY);
451 lastCPosY = FFMIN(firstChrSrcY + hout_slice->plane[1].available_lines - 1, AV_CEIL_RSHIFT(srcSliceY + srcSliceH, c->chrSrcVSubSample) - 1);
454 lastCPosY = lastChrSrcY;
457 ff_rotate_slice(hout_slice, lastPosY, lastCPosY);
459 if (posY < lastLumSrcY + 1) {
460 for (i = lumStart; i < lumEnd; ++i)
461 desc[i].process(c, &desc[i], firstPosY, lastPosY - firstPosY + 1);
464 lumBufIndex += lastLumSrcY - lastInLumBuf;
465 lastInLumBuf = lastLumSrcY;
467 if (cPosY < lastChrSrcY + 1) {
468 for (i = chrStart; i < chrEnd; ++i)
469 desc[i].process(c, &desc[i], firstCPosY, lastCPosY - firstCPosY + 1);
472 chrBufIndex += lastChrSrcY - lastInChrBuf;
473 lastInChrBuf = lastChrSrcY;
475 // wrap buf index around to stay inside the ring buffer
476 if (lumBufIndex >= vLumFilterSize)
477 lumBufIndex -= vLumFilterSize;
478 if (chrBufIndex >= vChrFilterSize)
479 chrBufIndex -= vChrFilterSize;
481 break; // we can't output a dstY line so let's try with the next slice
484 ff_updateMMXDitherTables(c, dstY, lumBufIndex, chrBufIndex,
485 lastInLumBuf, lastInChrBuf);
488 c->chrDither8 = ff_dither_8x8_128[chrDstY & 7];
489 c->lumDither8 = ff_dither_8x8_128[dstY & 7];
491 if (dstY >= dstH - 2) {
492 /* hmm looks like we can't use MMX here without overwriting
493 * this array's tail */
494 ff_sws_init_output_funcs(c, &yuv2plane1, &yuv2planeX, &yuv2nv12cX,
495 &yuv2packed1, &yuv2packed2, &yuv2packedX, &yuv2anyX);
497 ff_init_vscale_pfn(c, yuv2plane1, yuv2planeX, yuv2nv12cX,
498 yuv2packed1, yuv2packed2, yuv2packedX, yuv2anyX, use_mmx_vfilter);
502 for (i = vStart; i < vEnd; ++i)
503 desc[i].process(c, &desc[i], dstY, 1);
506 if (isPlanar(dstFormat) && isALPHA(dstFormat) && !needAlpha) {
508 int height = dstY - lastDstY;
510 if (is16BPS(dstFormat) || isNBPS(dstFormat)) {
511 const AVPixFmtDescriptor *desc = av_pix_fmt_desc_get(dstFormat);
512 fillPlane16(dst[3], dstStride[3], length, height, lastDstY,
513 1, desc->comp[3].depth,
516 fillPlane(dst[3], dstStride[3], length, height, lastDstY, 255);
519 #if HAVE_MMXEXT_INLINE
520 if (av_get_cpu_flags() & AV_CPU_FLAG_MMXEXT)
521 __asm__ volatile ("sfence" ::: "memory");
525 /* store changed local vars back in the context */
527 c->lumBufIndex = lumBufIndex;
528 c->chrBufIndex = chrBufIndex;
529 c->lastInLumBuf = lastInLumBuf;
530 c->lastInChrBuf = lastInChrBuf;
532 return dstY - lastDstY;
535 av_cold void ff_sws_init_range_convert(SwsContext *c)
537 c->lumConvertRange = NULL;
538 c->chrConvertRange = NULL;
539 if (c->srcRange != c->dstRange && !isAnyRGB(c->dstFormat)) {
540 if (c->dstBpc <= 14) {
542 c->lumConvertRange = lumRangeFromJpeg_c;
543 c->chrConvertRange = chrRangeFromJpeg_c;
545 c->lumConvertRange = lumRangeToJpeg_c;
546 c->chrConvertRange = chrRangeToJpeg_c;
550 c->lumConvertRange = lumRangeFromJpeg16_c;
551 c->chrConvertRange = chrRangeFromJpeg16_c;
553 c->lumConvertRange = lumRangeToJpeg16_c;
554 c->chrConvertRange = chrRangeToJpeg16_c;
560 static av_cold void sws_init_swscale(SwsContext *c)
562 enum AVPixelFormat srcFormat = c->srcFormat;
564 ff_sws_init_output_funcs(c, &c->yuv2plane1, &c->yuv2planeX,
565 &c->yuv2nv12cX, &c->yuv2packed1,
566 &c->yuv2packed2, &c->yuv2packedX, &c->yuv2anyX);
568 ff_sws_init_input_funcs(c);
570 if (c->srcBpc == 8) {
571 if (c->dstBpc <= 14) {
572 c->hyScale = c->hcScale = hScale8To15_c;
573 if (c->flags & SWS_FAST_BILINEAR) {
574 c->hyscale_fast = ff_hyscale_fast_c;
575 c->hcscale_fast = ff_hcscale_fast_c;
578 c->hyScale = c->hcScale = hScale8To19_c;
581 c->hyScale = c->hcScale = c->dstBpc > 14 ? hScale16To19_c
585 ff_sws_init_range_convert(c);
587 if (!(isGray(srcFormat) || isGray(c->dstFormat) ||
588 srcFormat == AV_PIX_FMT_MONOBLACK || srcFormat == AV_PIX_FMT_MONOWHITE))
589 c->needs_hcscale = 1;
592 SwsFunc ff_getSwsFunc(SwsContext *c)
597 ff_sws_init_swscale_ppc(c);
599 ff_sws_init_swscale_x86(c);
601 ff_sws_init_swscale_aarch64(c);
603 ff_sws_init_swscale_arm(c);
608 static void reset_ptr(const uint8_t *src[], enum AVPixelFormat format)
610 if (!isALPHA(format))
612 if (!isPlanar(format)) {
613 src[3] = src[2] = NULL;
620 static int check_image_pointers(const uint8_t * const data[4], enum AVPixelFormat pix_fmt,
621 const int linesizes[4])
623 const AVPixFmtDescriptor *desc = av_pix_fmt_desc_get(pix_fmt);
628 for (i = 0; i < 4; i++) {
629 int plane = desc->comp[i].plane;
630 if (!data[plane] || !linesizes[plane])
637 static void xyz12Torgb48(struct SwsContext *c, uint16_t *dst,
638 const uint16_t *src, int stride, int h)
641 const AVPixFmtDescriptor *desc = av_pix_fmt_desc_get(c->srcFormat);
643 for (yp=0; yp<h; yp++) {
644 for (xp=0; xp+2<stride; xp+=3) {
645 int x, y, z, r, g, b;
647 if (desc->flags & AV_PIX_FMT_FLAG_BE) {
648 x = AV_RB16(src + xp + 0);
649 y = AV_RB16(src + xp + 1);
650 z = AV_RB16(src + xp + 2);
652 x = AV_RL16(src + xp + 0);
653 y = AV_RL16(src + xp + 1);
654 z = AV_RL16(src + xp + 2);
657 x = c->xyzgamma[x>>4];
658 y = c->xyzgamma[y>>4];
659 z = c->xyzgamma[z>>4];
661 // convert from XYZlinear to sRGBlinear
662 r = c->xyz2rgb_matrix[0][0] * x +
663 c->xyz2rgb_matrix[0][1] * y +
664 c->xyz2rgb_matrix[0][2] * z >> 12;
665 g = c->xyz2rgb_matrix[1][0] * x +
666 c->xyz2rgb_matrix[1][1] * y +
667 c->xyz2rgb_matrix[1][2] * z >> 12;
668 b = c->xyz2rgb_matrix[2][0] * x +
669 c->xyz2rgb_matrix[2][1] * y +
670 c->xyz2rgb_matrix[2][2] * z >> 12;
672 // limit values to 12-bit depth
673 r = av_clip_uintp2(r, 12);
674 g = av_clip_uintp2(g, 12);
675 b = av_clip_uintp2(b, 12);
677 // convert from sRGBlinear to RGB and scale from 12bit to 16bit
678 if (desc->flags & AV_PIX_FMT_FLAG_BE) {
679 AV_WB16(dst + xp + 0, c->rgbgamma[r] << 4);
680 AV_WB16(dst + xp + 1, c->rgbgamma[g] << 4);
681 AV_WB16(dst + xp + 2, c->rgbgamma[b] << 4);
683 AV_WL16(dst + xp + 0, c->rgbgamma[r] << 4);
684 AV_WL16(dst + xp + 1, c->rgbgamma[g] << 4);
685 AV_WL16(dst + xp + 2, c->rgbgamma[b] << 4);
693 static void rgb48Toxyz12(struct SwsContext *c, uint16_t *dst,
694 const uint16_t *src, int stride, int h)
697 const AVPixFmtDescriptor *desc = av_pix_fmt_desc_get(c->dstFormat);
699 for (yp=0; yp<h; yp++) {
700 for (xp=0; xp+2<stride; xp+=3) {
701 int x, y, z, r, g, b;
703 if (desc->flags & AV_PIX_FMT_FLAG_BE) {
704 r = AV_RB16(src + xp + 0);
705 g = AV_RB16(src + xp + 1);
706 b = AV_RB16(src + xp + 2);
708 r = AV_RL16(src + xp + 0);
709 g = AV_RL16(src + xp + 1);
710 b = AV_RL16(src + xp + 2);
713 r = c->rgbgammainv[r>>4];
714 g = c->rgbgammainv[g>>4];
715 b = c->rgbgammainv[b>>4];
717 // convert from sRGBlinear to XYZlinear
718 x = c->rgb2xyz_matrix[0][0] * r +
719 c->rgb2xyz_matrix[0][1] * g +
720 c->rgb2xyz_matrix[0][2] * b >> 12;
721 y = c->rgb2xyz_matrix[1][0] * r +
722 c->rgb2xyz_matrix[1][1] * g +
723 c->rgb2xyz_matrix[1][2] * b >> 12;
724 z = c->rgb2xyz_matrix[2][0] * r +
725 c->rgb2xyz_matrix[2][1] * g +
726 c->rgb2xyz_matrix[2][2] * b >> 12;
728 // limit values to 12-bit depth
729 x = av_clip_uintp2(x, 12);
730 y = av_clip_uintp2(y, 12);
731 z = av_clip_uintp2(z, 12);
733 // convert from XYZlinear to X'Y'Z' and scale from 12bit to 16bit
734 if (desc->flags & AV_PIX_FMT_FLAG_BE) {
735 AV_WB16(dst + xp + 0, c->xyzgammainv[x] << 4);
736 AV_WB16(dst + xp + 1, c->xyzgammainv[y] << 4);
737 AV_WB16(dst + xp + 2, c->xyzgammainv[z] << 4);
739 AV_WL16(dst + xp + 0, c->xyzgammainv[x] << 4);
740 AV_WL16(dst + xp + 1, c->xyzgammainv[y] << 4);
741 AV_WL16(dst + xp + 2, c->xyzgammainv[z] << 4);
750 * swscale wrapper, so we don't need to export the SwsContext.
751 * Assumes planar YUV to be in YUV order instead of YVU.
753 int attribute_align_arg sws_scale(struct SwsContext *c,
754 const uint8_t * const srcSlice[],
755 const int srcStride[], int srcSliceY,
756 int srcSliceH, uint8_t *const dst[],
757 const int dstStride[])
760 const uint8_t *src2[4];
762 uint8_t *rgb0_tmp = NULL;
763 int macro_height = isBayer(c->srcFormat) ? 2 : (1 << c->chrSrcVSubSample);
764 // copy strides, so they can safely be modified
767 int srcSliceY_internal = srcSliceY;
769 if (!srcStride || !dstStride || !dst || !srcSlice) {
770 av_log(c, AV_LOG_ERROR, "One of the input parameters to sws_scale() is NULL, please check the calling code\n");
774 for (i=0; i<4; i++) {
775 srcStride2[i] = srcStride[i];
776 dstStride2[i] = dstStride[i];
779 if ((srcSliceY & (macro_height-1)) ||
780 ((srcSliceH& (macro_height-1)) && srcSliceY + srcSliceH != c->srcH) ||
781 srcSliceY + srcSliceH > c->srcH) {
782 av_log(c, AV_LOG_ERROR, "Slice parameters %d, %d are invalid\n", srcSliceY, srcSliceH);
783 return AVERROR(EINVAL);
786 if (c->gamma_flag && c->cascaded_context[0]) {
787 ret = sws_scale(c->cascaded_context[0],
788 srcSlice, srcStride, srcSliceY, srcSliceH,
789 c->cascaded_tmp, c->cascaded_tmpStride);
794 if (c->cascaded_context[2])
795 ret = sws_scale(c->cascaded_context[1], (const uint8_t * const *)c->cascaded_tmp, c->cascaded_tmpStride, srcSliceY, srcSliceH, c->cascaded1_tmp, c->cascaded1_tmpStride);
797 ret = sws_scale(c->cascaded_context[1], (const uint8_t * const *)c->cascaded_tmp, c->cascaded_tmpStride, srcSliceY, srcSliceH, dst, dstStride);
802 if (c->cascaded_context[2]) {
803 ret = sws_scale(c->cascaded_context[2],
804 (const uint8_t * const *)c->cascaded1_tmp, c->cascaded1_tmpStride, c->cascaded_context[1]->dstY - ret, c->cascaded_context[1]->dstY,
810 if (c->cascaded_context[0] && srcSliceY == 0 && srcSliceH == c->cascaded_context[0]->srcH) {
811 ret = sws_scale(c->cascaded_context[0],
812 srcSlice, srcStride, srcSliceY, srcSliceH,
813 c->cascaded_tmp, c->cascaded_tmpStride);
816 ret = sws_scale(c->cascaded_context[1],
817 (const uint8_t * const * )c->cascaded_tmp, c->cascaded_tmpStride, 0, c->cascaded_context[0]->dstH,
822 memcpy(src2, srcSlice, sizeof(src2));
823 memcpy(dst2, dst, sizeof(dst2));
825 // do not mess up sliceDir if we have a "trailing" 0-size slice
829 if (!check_image_pointers(srcSlice, c->srcFormat, srcStride)) {
830 av_log(c, AV_LOG_ERROR, "bad src image pointers\n");
833 if (!check_image_pointers((const uint8_t* const*)dst, c->dstFormat, dstStride)) {
834 av_log(c, AV_LOG_ERROR, "bad dst image pointers\n");
838 if (c->sliceDir == 0 && srcSliceY != 0 && srcSliceY + srcSliceH != c->srcH) {
839 av_log(c, AV_LOG_ERROR, "Slices start in the middle!\n");
842 if (c->sliceDir == 0) {
843 if (srcSliceY == 0) c->sliceDir = 1; else c->sliceDir = -1;
846 if (usePal(c->srcFormat)) {
847 for (i = 0; i < 256; i++) {
848 int r, g, b, y, u, v, a = 0xff;
849 if (c->srcFormat == AV_PIX_FMT_PAL8) {
850 uint32_t p = ((const uint32_t *)(srcSlice[1]))[i];
851 a = (p >> 24) & 0xFF;
852 r = (p >> 16) & 0xFF;
855 } else if (c->srcFormat == AV_PIX_FMT_RGB8) {
857 g = ((i >> 2) & 7) * 36;
859 } else if (c->srcFormat == AV_PIX_FMT_BGR8) {
861 g = ((i >> 3) & 7) * 36;
863 } else if (c->srcFormat == AV_PIX_FMT_RGB4_BYTE) {
864 r = ( i >> 3 ) * 255;
865 g = ((i >> 1) & 3) * 85;
867 } else if (c->srcFormat == AV_PIX_FMT_GRAY8 || c->srcFormat == AV_PIX_FMT_GRAY8A) {
870 av_assert1(c->srcFormat == AV_PIX_FMT_BGR4_BYTE);
871 b = ( i >> 3 ) * 255;
872 g = ((i >> 1) & 3) * 85;
875 #define RGB2YUV_SHIFT 15
876 #define BY ( (int) (0.114 * 219 / 255 * (1 << RGB2YUV_SHIFT) + 0.5))
877 #define BV (-(int) (0.081 * 224 / 255 * (1 << RGB2YUV_SHIFT) + 0.5))
878 #define BU ( (int) (0.500 * 224 / 255 * (1 << RGB2YUV_SHIFT) + 0.5))
879 #define GY ( (int) (0.587 * 219 / 255 * (1 << RGB2YUV_SHIFT) + 0.5))
880 #define GV (-(int) (0.419 * 224 / 255 * (1 << RGB2YUV_SHIFT) + 0.5))
881 #define GU (-(int) (0.331 * 224 / 255 * (1 << RGB2YUV_SHIFT) + 0.5))
882 #define RY ( (int) (0.299 * 219 / 255 * (1 << RGB2YUV_SHIFT) + 0.5))
883 #define RV ( (int) (0.500 * 224 / 255 * (1 << RGB2YUV_SHIFT) + 0.5))
884 #define RU (-(int) (0.169 * 224 / 255 * (1 << RGB2YUV_SHIFT) + 0.5))
886 y = av_clip_uint8((RY * r + GY * g + BY * b + ( 33 << (RGB2YUV_SHIFT - 1))) >> RGB2YUV_SHIFT);
887 u = av_clip_uint8((RU * r + GU * g + BU * b + (257 << (RGB2YUV_SHIFT - 1))) >> RGB2YUV_SHIFT);
888 v = av_clip_uint8((RV * r + GV * g + BV * b + (257 << (RGB2YUV_SHIFT - 1))) >> RGB2YUV_SHIFT);
889 c->pal_yuv[i]= y + (u<<8) + (v<<16) + ((unsigned)a<<24);
891 switch (c->dstFormat) {
892 case AV_PIX_FMT_BGR32:
894 case AV_PIX_FMT_RGB24:
896 c->pal_rgb[i]= r + (g<<8) + (b<<16) + ((unsigned)a<<24);
898 case AV_PIX_FMT_BGR32_1:
900 case AV_PIX_FMT_BGR24:
902 c->pal_rgb[i]= a + (r<<8) + (g<<16) + ((unsigned)b<<24);
904 case AV_PIX_FMT_RGB32_1:
906 case AV_PIX_FMT_RGB24:
908 c->pal_rgb[i]= a + (b<<8) + (g<<16) + ((unsigned)r<<24);
910 case AV_PIX_FMT_RGB32:
912 case AV_PIX_FMT_BGR24:
915 c->pal_rgb[i]= b + (g<<8) + (r<<16) + ((unsigned)a<<24);
920 if (c->src0Alpha && !c->dst0Alpha && isALPHA(c->dstFormat)) {
923 rgb0_tmp = av_malloc(FFABS(srcStride[0]) * srcSliceH + 32);
925 return AVERROR(ENOMEM);
927 base = srcStride[0] < 0 ? rgb0_tmp - srcStride[0] * (srcSliceH-1) : rgb0_tmp;
928 for (y=0; y<srcSliceH; y++){
929 memcpy(base + srcStride[0]*y, src2[0] + srcStride[0]*y, 4*c->srcW);
930 for (x=c->src0Alpha-1; x<4*c->srcW; x+=4) {
931 base[ srcStride[0]*y + x] = 0xFF;
937 if (c->srcXYZ && !(c->dstXYZ && c->srcW==c->dstW && c->srcH==c->dstH)) {
939 rgb0_tmp = av_malloc(FFABS(srcStride[0]) * srcSliceH + 32);
941 return AVERROR(ENOMEM);
943 base = srcStride[0] < 0 ? rgb0_tmp - srcStride[0] * (srcSliceH-1) : rgb0_tmp;
945 xyz12Torgb48(c, (uint16_t*)base, (const uint16_t*)src2[0], srcStride[0]/2, srcSliceH);
949 if (!srcSliceY && (c->flags & SWS_BITEXACT) && c->dither == SWS_DITHER_ED && c->dither_error[0])
950 for (i = 0; i < 4; i++)
951 memset(c->dither_error[i], 0, sizeof(c->dither_error[0][0]) * (c->dstW+2));
953 if (c->sliceDir != 1) {
954 // slices go from bottom to top => we flip the image internally
955 for (i=0; i<4; i++) {
960 src2[0] += (srcSliceH - 1) * srcStride[0];
961 if (!usePal(c->srcFormat))
962 src2[1] += ((srcSliceH >> c->chrSrcVSubSample) - 1) * srcStride[1];
963 src2[2] += ((srcSliceH >> c->chrSrcVSubSample) - 1) * srcStride[2];
964 src2[3] += (srcSliceH - 1) * srcStride[3];
965 dst2[0] += ( c->dstH - 1) * dstStride[0];
966 dst2[1] += ((c->dstH >> c->chrDstVSubSample) - 1) * dstStride[1];
967 dst2[2] += ((c->dstH >> c->chrDstVSubSample) - 1) * dstStride[2];
968 dst2[3] += ( c->dstH - 1) * dstStride[3];
970 srcSliceY_internal = c->srcH-srcSliceY-srcSliceH;
972 reset_ptr(src2, c->srcFormat);
973 reset_ptr((void*)dst2, c->dstFormat);
975 /* reset slice direction at end of frame */
976 if (srcSliceY_internal + srcSliceH == c->srcH)
978 ret = c->swscale(c, src2, srcStride2, srcSliceY_internal, srcSliceH, dst2, dstStride2);
980 if (c->dstXYZ && !(c->srcXYZ && c->srcW==c->dstW && c->srcH==c->dstH)) {
981 int dstY = c->dstY ? c->dstY : srcSliceY + srcSliceH;
982 uint16_t *dst16 = (uint16_t*)(dst2[0] + (dstY - ret) * dstStride2[0]);
983 av_assert0(dstY >= ret);
984 av_assert0(ret >= 0);
985 av_assert0(c->dstH >= dstY);
987 /* replace on the same data */
988 rgb48Toxyz12(c, dst16, dst16, dstStride2[0]/2, ret);