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)
80 for (i = 0; i < dstW; i++) {
82 int srcPos = filterPos[i];
85 for (j = 0; j < filterSize; j++) {
86 val += src[srcPos + j] * filter[filterSize * i + j];
88 // filter=14 bit, input=16 bit, output=30 bit, >> 11 makes 19 bit
89 dst[i] = FFMIN(val >> sh, (1 << 19) - 1);
93 static void hScale16To15_c(SwsContext *c, int16_t *dst, int dstW,
94 const uint8_t *_src, const int16_t *filter,
95 const int32_t *filterPos, int filterSize)
97 const AVPixFmtDescriptor *desc = av_pix_fmt_desc_get(c->srcFormat);
99 const uint16_t *src = (const uint16_t *) _src;
100 int sh = desc->comp[0].depth - 1;
103 sh= isAnyRGB(c->srcFormat) || c->srcFormat==AV_PIX_FMT_PAL8 ? 13 : (desc->comp[0].depth - 1);
105 for (i = 0; i < dstW; i++) {
107 int srcPos = filterPos[i];
110 for (j = 0; j < filterSize; j++) {
111 val += src[srcPos + j] * filter[filterSize * i + j];
113 // filter=14 bit, input=16 bit, output=30 bit, >> 15 makes 15 bit
114 dst[i] = FFMIN(val >> sh, (1 << 15) - 1);
118 // bilinear / bicubic scaling
119 static void hScale8To15_c(SwsContext *c, int16_t *dst, int dstW,
120 const uint8_t *src, const int16_t *filter,
121 const int32_t *filterPos, int filterSize)
124 for (i = 0; i < dstW; i++) {
126 int srcPos = filterPos[i];
128 for (j = 0; j < filterSize; j++) {
129 val += ((int)src[srcPos + j]) * filter[filterSize * i + j];
131 dst[i] = FFMIN(val >> 7, (1 << 15) - 1); // the cubic equation does overflow ...
135 static void hScale8To19_c(SwsContext *c, int16_t *_dst, int dstW,
136 const uint8_t *src, const int16_t *filter,
137 const int32_t *filterPos, int filterSize)
140 int32_t *dst = (int32_t *) _dst;
141 for (i = 0; i < dstW; i++) {
143 int srcPos = filterPos[i];
145 for (j = 0; j < filterSize; j++) {
146 val += ((int)src[srcPos + j]) * filter[filterSize * i + j];
148 dst[i] = FFMIN(val >> 3, (1 << 19) - 1); // the cubic equation does overflow ...
152 // FIXME all pal and rgb srcFormats could do this conversion as well
153 // FIXME all scalers more complex than bilinear could do half of this transform
154 static void chrRangeToJpeg_c(int16_t *dstU, int16_t *dstV, int width)
157 for (i = 0; i < width; i++) {
158 dstU[i] = (FFMIN(dstU[i], 30775) * 4663 - 9289992) >> 12; // -264
159 dstV[i] = (FFMIN(dstV[i], 30775) * 4663 - 9289992) >> 12; // -264
163 static void chrRangeFromJpeg_c(int16_t *dstU, int16_t *dstV, int width)
166 for (i = 0; i < width; i++) {
167 dstU[i] = (dstU[i] * 1799 + 4081085) >> 11; // 1469
168 dstV[i] = (dstV[i] * 1799 + 4081085) >> 11; // 1469
172 static void lumRangeToJpeg_c(int16_t *dst, int width)
175 for (i = 0; i < width; i++)
176 dst[i] = (FFMIN(dst[i], 30189) * 19077 - 39057361) >> 14;
179 static void lumRangeFromJpeg_c(int16_t *dst, int width)
182 for (i = 0; i < width; i++)
183 dst[i] = (dst[i] * 14071 + 33561947) >> 14;
186 static void chrRangeToJpeg16_c(int16_t *_dstU, int16_t *_dstV, int width)
189 int32_t *dstU = (int32_t *) _dstU;
190 int32_t *dstV = (int32_t *) _dstV;
191 for (i = 0; i < width; i++) {
192 dstU[i] = (FFMIN(dstU[i], 30775 << 4) * 4663 - (9289992 << 4)) >> 12; // -264
193 dstV[i] = (FFMIN(dstV[i], 30775 << 4) * 4663 - (9289992 << 4)) >> 12; // -264
197 static void chrRangeFromJpeg16_c(int16_t *_dstU, int16_t *_dstV, int width)
200 int32_t *dstU = (int32_t *) _dstU;
201 int32_t *dstV = (int32_t *) _dstV;
202 for (i = 0; i < width; i++) {
203 dstU[i] = (dstU[i] * 1799 + (4081085 << 4)) >> 11; // 1469
204 dstV[i] = (dstV[i] * 1799 + (4081085 << 4)) >> 11; // 1469
208 static void lumRangeToJpeg16_c(int16_t *_dst, int width)
211 int32_t *dst = (int32_t *) _dst;
212 for (i = 0; i < width; i++) {
213 dst[i] = ((int)(FFMIN(dst[i], 30189 << 4) * 4769U - (39057361 << 2))) >> 12;
217 static void lumRangeFromJpeg16_c(int16_t *_dst, int width)
220 int32_t *dst = (int32_t *) _dst;
221 for (i = 0; i < width; i++)
222 dst[i] = (dst[i]*(14071/4) + (33561947<<4)/4)>>12;
226 #define DEBUG_SWSCALE_BUFFERS 0
227 #define DEBUG_BUFFERS(...) \
228 if (DEBUG_SWSCALE_BUFFERS) \
229 av_log(c, AV_LOG_DEBUG, __VA_ARGS__)
231 static int swscale(SwsContext *c, const uint8_t *src[],
232 int srcStride[], int srcSliceY,
233 int srcSliceH, uint8_t *dst[], int dstStride[])
235 /* load a few things into local vars to make the code more readable?
237 const int dstW = c->dstW;
238 const int dstH = c->dstH;
240 const enum AVPixelFormat dstFormat = c->dstFormat;
241 const int flags = c->flags;
242 int32_t *vLumFilterPos = c->vLumFilterPos;
243 int32_t *vChrFilterPos = c->vChrFilterPos;
245 const int vLumFilterSize = c->vLumFilterSize;
246 const int vChrFilterSize = c->vChrFilterSize;
248 yuv2planar1_fn yuv2plane1 = c->yuv2plane1;
249 yuv2planarX_fn yuv2planeX = c->yuv2planeX;
250 yuv2interleavedX_fn yuv2nv12cX = c->yuv2nv12cX;
251 yuv2packed1_fn yuv2packed1 = c->yuv2packed1;
252 yuv2packed2_fn yuv2packed2 = c->yuv2packed2;
253 yuv2packedX_fn yuv2packedX = c->yuv2packedX;
254 yuv2anyX_fn yuv2anyX = c->yuv2anyX;
255 const int chrSrcSliceY = srcSliceY >> c->chrSrcVSubSample;
256 const int chrSrcSliceH = AV_CEIL_RSHIFT(srcSliceH, c->chrSrcVSubSample);
257 int should_dither = is9_OR_10BPS(c->srcFormat) ||
258 is16BPS(c->srcFormat);
261 /* vars which will change and which we need to store back in the context */
263 int lumBufIndex = c->lumBufIndex;
264 int chrBufIndex = c->chrBufIndex;
265 int lastInLumBuf = c->lastInLumBuf;
266 int lastInChrBuf = c->lastInChrBuf;
270 int lumEnd = c->descIndex[0];
271 int chrStart = lumEnd;
272 int chrEnd = c->descIndex[1];
274 int vEnd = c->numDesc;
275 SwsSlice *src_slice = &c->slice[lumStart];
276 SwsSlice *hout_slice = &c->slice[c->numSlice-2];
277 SwsSlice *vout_slice = &c->slice[c->numSlice-1];
278 SwsFilterDescriptor *desc = c->desc;
281 int needAlpha = c->needAlpha;
287 if (isPacked(c->srcFormat)) {
295 srcStride[3] = srcStride[0];
297 srcStride[1] <<= c->vChrDrop;
298 srcStride[2] <<= c->vChrDrop;
300 DEBUG_BUFFERS("swscale() %p[%d] %p[%d] %p[%d] %p[%d] -> %p[%d] %p[%d] %p[%d] %p[%d]\n",
301 src[0], srcStride[0], src[1], srcStride[1],
302 src[2], srcStride[2], src[3], srcStride[3],
303 dst[0], dstStride[0], dst[1], dstStride[1],
304 dst[2], dstStride[2], dst[3], dstStride[3]);
305 DEBUG_BUFFERS("srcSliceY: %d srcSliceH: %d dstY: %d dstH: %d\n",
306 srcSliceY, srcSliceH, dstY, dstH);
307 DEBUG_BUFFERS("vLumFilterSize: %d vChrFilterSize: %d\n",
308 vLumFilterSize, vChrFilterSize);
310 if (dstStride[0]&15 || dstStride[1]&15 ||
311 dstStride[2]&15 || dstStride[3]&15) {
312 static int warnedAlready = 0; // FIXME maybe move this into the context
313 if (flags & SWS_PRINT_INFO && !warnedAlready) {
314 av_log(c, AV_LOG_WARNING,
315 "Warning: dstStride is not aligned!\n"
316 " ->cannot do aligned memory accesses anymore\n");
321 if ( (uintptr_t)dst[0]&15 || (uintptr_t)dst[1]&15 || (uintptr_t)dst[2]&15
322 || (uintptr_t)src[0]&15 || (uintptr_t)src[1]&15 || (uintptr_t)src[2]&15
323 || dstStride[0]&15 || dstStride[1]&15 || dstStride[2]&15 || dstStride[3]&15
324 || srcStride[0]&15 || srcStride[1]&15 || srcStride[2]&15 || srcStride[3]&15
326 static int warnedAlready=0;
327 int cpu_flags = av_get_cpu_flags();
328 if (HAVE_MMXEXT && (cpu_flags & AV_CPU_FLAG_SSE2) && !warnedAlready){
329 av_log(c, AV_LOG_WARNING, "Warning: data is not aligned! This can lead to a speedloss\n");
334 /* Note the user might start scaling the picture in the middle so this
335 * will not get executed. This is not really intended but works
336 * currently, so people might do it. */
337 if (srcSliceY == 0) {
345 if (!should_dither) {
346 c->chrDither8 = c->lumDither8 = sws_pb_64;
350 ff_init_vscale_pfn(c, yuv2plane1, yuv2planeX, yuv2nv12cX,
351 yuv2packed1, yuv2packed2, yuv2packedX, yuv2anyX, c->use_mmx_vfilter);
353 ff_init_slice_from_src(src_slice, (uint8_t**)src, srcStride, c->srcW,
354 srcSliceY, srcSliceH, chrSrcSliceY, chrSrcSliceH, 1);
356 ff_init_slice_from_src(vout_slice, (uint8_t**)dst, dstStride, c->dstW,
357 dstY, dstH, dstY >> c->chrDstVSubSample,
358 AV_CEIL_RSHIFT(dstH, c->chrDstVSubSample), 0);
359 if (srcSliceY == 0) {
360 hout_slice->plane[0].sliceY = lastInLumBuf + 1;
361 hout_slice->plane[1].sliceY = lastInChrBuf + 1;
362 hout_slice->plane[2].sliceY = lastInChrBuf + 1;
363 hout_slice->plane[3].sliceY = lastInLumBuf + 1;
365 hout_slice->plane[0].sliceH =
366 hout_slice->plane[1].sliceH =
367 hout_slice->plane[2].sliceH =
368 hout_slice->plane[3].sliceH = 0;
369 hout_slice->width = dstW;
372 for (; dstY < dstH; dstY++) {
373 const int chrDstY = dstY >> c->chrDstVSubSample;
374 int use_mmx_vfilter= c->use_mmx_vfilter;
376 // First line needed as input
377 const int firstLumSrcY = FFMAX(1 - vLumFilterSize, vLumFilterPos[dstY]);
378 const int firstLumSrcY2 = FFMAX(1 - vLumFilterSize, vLumFilterPos[FFMIN(dstY | ((1 << c->chrDstVSubSample) - 1), dstH - 1)]);
379 // First line needed as input
380 const int firstChrSrcY = FFMAX(1 - vChrFilterSize, vChrFilterPos[chrDstY]);
382 // Last line needed as input
383 int lastLumSrcY = FFMIN(c->srcH, firstLumSrcY + vLumFilterSize) - 1;
384 int lastLumSrcY2 = FFMIN(c->srcH, firstLumSrcY2 + vLumFilterSize) - 1;
385 int lastChrSrcY = FFMIN(c->chrSrcH, firstChrSrcY + vChrFilterSize) - 1;
389 int posY, cPosY, firstPosY, lastPosY, firstCPosY, lastCPosY;
391 // handle holes (FAST_BILINEAR & weird filters)
392 if (firstLumSrcY > lastInLumBuf) {
394 hasLumHoles = lastInLumBuf != firstLumSrcY - 1;
396 hout_slice->plane[0].sliceY = firstLumSrcY;
397 hout_slice->plane[3].sliceY = firstLumSrcY;
398 hout_slice->plane[0].sliceH =
399 hout_slice->plane[3].sliceH = 0;
402 lastInLumBuf = firstLumSrcY - 1;
404 if (firstChrSrcY > lastInChrBuf) {
406 hasChrHoles = lastInChrBuf != firstChrSrcY - 1;
408 hout_slice->plane[1].sliceY = firstChrSrcY;
409 hout_slice->plane[2].sliceY = firstChrSrcY;
410 hout_slice->plane[1].sliceH =
411 hout_slice->plane[2].sliceH = 0;
414 lastInChrBuf = firstChrSrcY - 1;
417 DEBUG_BUFFERS("dstY: %d\n", dstY);
418 DEBUG_BUFFERS("\tfirstLumSrcY: %d lastLumSrcY: %d lastInLumBuf: %d\n",
419 firstLumSrcY, lastLumSrcY, lastInLumBuf);
420 DEBUG_BUFFERS("\tfirstChrSrcY: %d lastChrSrcY: %d lastInChrBuf: %d\n",
421 firstChrSrcY, lastChrSrcY, lastInChrBuf);
423 // Do we have enough lines in this slice to output the dstY line
424 enough_lines = lastLumSrcY2 < srcSliceY + srcSliceH &&
425 lastChrSrcY < AV_CEIL_RSHIFT(srcSliceY + srcSliceH, c->chrSrcVSubSample);
428 lastLumSrcY = srcSliceY + srcSliceH - 1;
429 lastChrSrcY = chrSrcSliceY + chrSrcSliceH - 1;
430 DEBUG_BUFFERS("buffering slice: lastLumSrcY %d lastChrSrcY %d\n",
431 lastLumSrcY, lastChrSrcY);
434 av_assert0((lastLumSrcY - firstLumSrcY + 1) <= hout_slice->plane[0].available_lines);
435 av_assert0((lastChrSrcY - firstChrSrcY + 1) <= hout_slice->plane[1].available_lines);
438 posY = hout_slice->plane[0].sliceY + hout_slice->plane[0].sliceH;
439 if (posY <= lastLumSrcY && !hasLumHoles) {
440 firstPosY = FFMAX(firstLumSrcY, posY);
441 lastPosY = FFMIN(firstLumSrcY + hout_slice->plane[0].available_lines - 1, srcSliceY + srcSliceH - 1);
444 lastPosY = lastLumSrcY;
447 cPosY = hout_slice->plane[1].sliceY + hout_slice->plane[1].sliceH;
448 if (cPosY <= lastChrSrcY && !hasChrHoles) {
449 firstCPosY = FFMAX(firstChrSrcY, cPosY);
450 lastCPosY = FFMIN(firstChrSrcY + hout_slice->plane[1].available_lines - 1, AV_CEIL_RSHIFT(srcSliceY + srcSliceH, c->chrSrcVSubSample) - 1);
453 lastCPosY = lastChrSrcY;
456 ff_rotate_slice(hout_slice, lastPosY, lastCPosY);
458 if (posY < lastLumSrcY + 1) {
459 for (i = lumStart; i < lumEnd; ++i)
460 desc[i].process(c, &desc[i], firstPosY, lastPosY - firstPosY + 1);
463 lumBufIndex += lastLumSrcY - lastInLumBuf;
464 lastInLumBuf = lastLumSrcY;
466 if (cPosY < lastChrSrcY + 1) {
467 for (i = chrStart; i < chrEnd; ++i)
468 desc[i].process(c, &desc[i], firstCPosY, lastCPosY - firstCPosY + 1);
471 chrBufIndex += lastChrSrcY - lastInChrBuf;
472 lastInChrBuf = lastChrSrcY;
474 // wrap buf index around to stay inside the ring buffer
475 if (lumBufIndex >= vLumFilterSize)
476 lumBufIndex -= vLumFilterSize;
477 if (chrBufIndex >= vChrFilterSize)
478 chrBufIndex -= vChrFilterSize;
480 break; // we can't output a dstY line so let's try with the next slice
483 ff_updateMMXDitherTables(c, dstY, lumBufIndex, chrBufIndex,
484 lastInLumBuf, lastInChrBuf);
487 c->chrDither8 = ff_dither_8x8_128[chrDstY & 7];
488 c->lumDither8 = ff_dither_8x8_128[dstY & 7];
490 if (dstY >= dstH - 2) {
491 /* hmm looks like we can't use MMX here without overwriting
492 * this array's tail */
493 ff_sws_init_output_funcs(c, &yuv2plane1, &yuv2planeX, &yuv2nv12cX,
494 &yuv2packed1, &yuv2packed2, &yuv2packedX, &yuv2anyX);
496 ff_init_vscale_pfn(c, yuv2plane1, yuv2planeX, yuv2nv12cX,
497 yuv2packed1, yuv2packed2, yuv2packedX, yuv2anyX, use_mmx_vfilter);
501 for (i = vStart; i < vEnd; ++i)
502 desc[i].process(c, &desc[i], dstY, 1);
505 if (isPlanar(dstFormat) && isALPHA(dstFormat) && !needAlpha) {
507 int height = dstY - lastDstY;
509 if (is16BPS(dstFormat) || isNBPS(dstFormat)) {
510 const AVPixFmtDescriptor *desc = av_pix_fmt_desc_get(dstFormat);
511 fillPlane16(dst[3], dstStride[3], length, height, lastDstY,
512 1, desc->comp[3].depth,
515 fillPlane(dst[3], dstStride[3], length, height, lastDstY, 255);
518 #if HAVE_MMXEXT_INLINE
519 if (av_get_cpu_flags() & AV_CPU_FLAG_MMXEXT)
520 __asm__ volatile ("sfence" ::: "memory");
524 /* store changed local vars back in the context */
526 c->lumBufIndex = lumBufIndex;
527 c->chrBufIndex = chrBufIndex;
528 c->lastInLumBuf = lastInLumBuf;
529 c->lastInChrBuf = lastInChrBuf;
531 return dstY - lastDstY;
534 av_cold void ff_sws_init_range_convert(SwsContext *c)
536 c->lumConvertRange = NULL;
537 c->chrConvertRange = NULL;
538 if (c->srcRange != c->dstRange && !isAnyRGB(c->dstFormat)) {
539 if (c->dstBpc <= 14) {
541 c->lumConvertRange = lumRangeFromJpeg_c;
542 c->chrConvertRange = chrRangeFromJpeg_c;
544 c->lumConvertRange = lumRangeToJpeg_c;
545 c->chrConvertRange = chrRangeToJpeg_c;
549 c->lumConvertRange = lumRangeFromJpeg16_c;
550 c->chrConvertRange = chrRangeFromJpeg16_c;
552 c->lumConvertRange = lumRangeToJpeg16_c;
553 c->chrConvertRange = chrRangeToJpeg16_c;
559 static av_cold void sws_init_swscale(SwsContext *c)
561 enum AVPixelFormat srcFormat = c->srcFormat;
563 ff_sws_init_output_funcs(c, &c->yuv2plane1, &c->yuv2planeX,
564 &c->yuv2nv12cX, &c->yuv2packed1,
565 &c->yuv2packed2, &c->yuv2packedX, &c->yuv2anyX);
567 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
765 int srcStride2[4] = { srcStride[0], srcStride[1], srcStride[2],
767 int dstStride2[4] = { dstStride[0], dstStride[1], dstStride[2],
769 int srcSliceY_internal = srcSliceY;
771 if (!srcStride || !dstStride || !dst || !srcSlice) {
772 av_log(c, AV_LOG_ERROR, "One of the input parameters to sws_scale() is NULL, please check the calling code\n");
776 if ((srcSliceY & (macro_height-1)) ||
777 ((srcSliceH& (macro_height-1)) && srcSliceY + srcSliceH != c->srcH) ||
778 srcSliceY + srcSliceH > c->srcH) {
779 av_log(c, AV_LOG_ERROR, "Slice parameters %d, %d are invalid\n", srcSliceY, srcSliceH);
780 return AVERROR(EINVAL);
783 if (c->gamma_flag && c->cascaded_context[0]) {
786 ret = sws_scale(c->cascaded_context[0],
787 srcSlice, srcStride, srcSliceY, srcSliceH,
788 c->cascaded_tmp, c->cascaded_tmpStride);
793 if (c->cascaded_context[2])
794 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);
796 ret = sws_scale(c->cascaded_context[1], (const uint8_t * const *)c->cascaded_tmp, c->cascaded_tmpStride, srcSliceY, srcSliceH, dst, dstStride);
801 if (c->cascaded_context[2]) {
802 ret = sws_scale(c->cascaded_context[2],
803 (const uint8_t * const *)c->cascaded1_tmp, c->cascaded1_tmpStride, c->cascaded_context[1]->dstY - ret, c->cascaded_context[1]->dstY,
809 if (c->cascaded_context[0] && srcSliceY == 0 && srcSliceH == c->cascaded_context[0]->srcH) {
810 ret = sws_scale(c->cascaded_context[0],
811 srcSlice, srcStride, srcSliceY, srcSliceH,
812 c->cascaded_tmp, c->cascaded_tmpStride);
815 ret = sws_scale(c->cascaded_context[1],
816 (const uint8_t * const * )c->cascaded_tmp, c->cascaded_tmpStride, 0, c->cascaded_context[0]->dstH,
821 memcpy(src2, srcSlice, sizeof(src2));
822 memcpy(dst2, dst, sizeof(dst2));
824 // do not mess up sliceDir if we have a "trailing" 0-size slice
828 if (!check_image_pointers(srcSlice, c->srcFormat, srcStride)) {
829 av_log(c, AV_LOG_ERROR, "bad src image pointers\n");
832 if (!check_image_pointers((const uint8_t* const*)dst, c->dstFormat, dstStride)) {
833 av_log(c, AV_LOG_ERROR, "bad dst image pointers\n");
837 if (c->sliceDir == 0 && srcSliceY != 0 && srcSliceY + srcSliceH != c->srcH) {
838 av_log(c, AV_LOG_ERROR, "Slices start in the middle!\n");
841 if (c->sliceDir == 0) {
842 if (srcSliceY == 0) c->sliceDir = 1; else c->sliceDir = -1;
845 if (usePal(c->srcFormat)) {
846 for (i = 0; i < 256; i++) {
847 int r, g, b, y, u, v, a = 0xff;
848 if (c->srcFormat == AV_PIX_FMT_PAL8) {
849 uint32_t p = ((const uint32_t *)(srcSlice[1]))[i];
850 a = (p >> 24) & 0xFF;
851 r = (p >> 16) & 0xFF;
854 } else if (c->srcFormat == AV_PIX_FMT_RGB8) {
856 g = ((i >> 2) & 7) * 36;
858 } else if (c->srcFormat == AV_PIX_FMT_BGR8) {
860 g = ((i >> 3) & 7) * 36;
862 } else if (c->srcFormat == AV_PIX_FMT_RGB4_BYTE) {
863 r = ( i >> 3 ) * 255;
864 g = ((i >> 1) & 3) * 85;
866 } else if (c->srcFormat == AV_PIX_FMT_GRAY8 || c->srcFormat == AV_PIX_FMT_GRAY8A) {
869 av_assert1(c->srcFormat == AV_PIX_FMT_BGR4_BYTE);
870 b = ( i >> 3 ) * 255;
871 g = ((i >> 1) & 3) * 85;
874 #define RGB2YUV_SHIFT 15
875 #define BY ( (int) (0.114 * 219 / 255 * (1 << RGB2YUV_SHIFT) + 0.5))
876 #define BV (-(int) (0.081 * 224 / 255 * (1 << RGB2YUV_SHIFT) + 0.5))
877 #define BU ( (int) (0.500 * 224 / 255 * (1 << RGB2YUV_SHIFT) + 0.5))
878 #define GY ( (int) (0.587 * 219 / 255 * (1 << RGB2YUV_SHIFT) + 0.5))
879 #define GV (-(int) (0.419 * 224 / 255 * (1 << RGB2YUV_SHIFT) + 0.5))
880 #define GU (-(int) (0.331 * 224 / 255 * (1 << RGB2YUV_SHIFT) + 0.5))
881 #define RY ( (int) (0.299 * 219 / 255 * (1 << RGB2YUV_SHIFT) + 0.5))
882 #define RV ( (int) (0.500 * 224 / 255 * (1 << RGB2YUV_SHIFT) + 0.5))
883 #define RU (-(int) (0.169 * 224 / 255 * (1 << RGB2YUV_SHIFT) + 0.5))
885 y = av_clip_uint8((RY * r + GY * g + BY * b + ( 33 << (RGB2YUV_SHIFT - 1))) >> RGB2YUV_SHIFT);
886 u = av_clip_uint8((RU * r + GU * g + BU * b + (257 << (RGB2YUV_SHIFT - 1))) >> RGB2YUV_SHIFT);
887 v = av_clip_uint8((RV * r + GV * g + BV * b + (257 << (RGB2YUV_SHIFT - 1))) >> RGB2YUV_SHIFT);
888 c->pal_yuv[i]= y + (u<<8) + (v<<16) + ((unsigned)a<<24);
890 switch (c->dstFormat) {
891 case AV_PIX_FMT_BGR32:
893 case AV_PIX_FMT_RGB24:
895 c->pal_rgb[i]= r + (g<<8) + (b<<16) + ((unsigned)a<<24);
897 case AV_PIX_FMT_BGR32_1:
899 case AV_PIX_FMT_BGR24:
901 c->pal_rgb[i]= a + (r<<8) + (g<<16) + ((unsigned)b<<24);
903 case AV_PIX_FMT_RGB32_1:
905 case AV_PIX_FMT_RGB24:
907 c->pal_rgb[i]= a + (b<<8) + (g<<16) + ((unsigned)r<<24);
909 case AV_PIX_FMT_RGB32:
911 case AV_PIX_FMT_BGR24:
914 c->pal_rgb[i]= b + (g<<8) + (r<<16) + ((unsigned)a<<24);
919 if (c->src0Alpha && !c->dst0Alpha && isALPHA(c->dstFormat)) {
922 rgb0_tmp = av_malloc(FFABS(srcStride[0]) * srcSliceH + 32);
924 return AVERROR(ENOMEM);
926 base = srcStride[0] < 0 ? rgb0_tmp - srcStride[0] * (srcSliceH-1) : rgb0_tmp;
927 for (y=0; y<srcSliceH; y++){
928 memcpy(base + srcStride[0]*y, src2[0] + srcStride[0]*y, 4*c->srcW);
929 for (x=c->src0Alpha-1; x<4*c->srcW; x+=4) {
930 base[ srcStride[0]*y + x] = 0xFF;
936 if (c->srcXYZ && !(c->dstXYZ && c->srcW==c->dstW && c->srcH==c->dstH)) {
938 rgb0_tmp = av_malloc(FFABS(srcStride[0]) * srcSliceH + 32);
940 return AVERROR(ENOMEM);
942 base = srcStride[0] < 0 ? rgb0_tmp - srcStride[0] * (srcSliceH-1) : rgb0_tmp;
944 xyz12Torgb48(c, (uint16_t*)base, (const uint16_t*)src2[0], srcStride[0]/2, srcSliceH);
948 if (!srcSliceY && (c->flags & SWS_BITEXACT) && c->dither == SWS_DITHER_ED && c->dither_error[0])
949 for (i = 0; i < 4; i++)
950 memset(c->dither_error[i], 0, sizeof(c->dither_error[0][0]) * (c->dstW+2));
952 if (c->sliceDir != 1) {
953 // slices go from bottom to top => we flip the image internally
954 for (i=0; i<4; i++) {
959 src2[0] += (srcSliceH - 1) * srcStride[0];
960 if (!usePal(c->srcFormat))
961 src2[1] += ((srcSliceH >> c->chrSrcVSubSample) - 1) * srcStride[1];
962 src2[2] += ((srcSliceH >> c->chrSrcVSubSample) - 1) * srcStride[2];
963 src2[3] += (srcSliceH - 1) * srcStride[3];
964 dst2[0] += ( c->dstH - 1) * dstStride[0];
965 dst2[1] += ((c->dstH >> c->chrDstVSubSample) - 1) * dstStride[1];
966 dst2[2] += ((c->dstH >> c->chrDstVSubSample) - 1) * dstStride[2];
967 dst2[3] += ( c->dstH - 1) * dstStride[3];
969 srcSliceY_internal = c->srcH-srcSliceY-srcSliceH;
971 reset_ptr(src2, c->srcFormat);
972 reset_ptr((void*)dst2, c->dstFormat);
974 /* reset slice direction at end of frame */
975 if (srcSliceY_internal + srcSliceH == c->srcH)
977 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);