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 void gamma_convert(uint8_t * src[], int width, uint16_t *gamma)
58 uint16_t *src1 = (uint16_t*)src[0];
60 for (i = 0; i < width; ++i) {
61 uint16_t r = AV_RL16(src1 + i*4 + 0);
62 uint16_t g = AV_RL16(src1 + i*4 + 1);
63 uint16_t b = AV_RL16(src1 + i*4 + 2);
65 AV_WL16(src1 + i*4 + 0, gamma[r]);
66 AV_WL16(src1 + i*4 + 1, gamma[g]);
67 AV_WL16(src1 + i*4 + 2, gamma[b]);
71 static av_always_inline void fillPlane(uint8_t *plane, int stride, int width,
72 int height, int y, uint8_t val)
75 uint8_t *ptr = plane + stride * y;
76 for (i = 0; i < height; i++) {
77 memset(ptr, val, width);
82 static void hScale16To19_c(SwsContext *c, int16_t *_dst, int dstW,
83 const uint8_t *_src, const int16_t *filter,
84 const int32_t *filterPos, int filterSize)
86 const AVPixFmtDescriptor *desc = av_pix_fmt_desc_get(c->srcFormat);
88 int32_t *dst = (int32_t *) _dst;
89 const uint16_t *src = (const uint16_t *) _src;
90 int bits = desc->comp[0].depth_minus1;
93 if((isAnyRGB(c->srcFormat) || c->srcFormat==AV_PIX_FMT_PAL8) && desc->comp[0].depth_minus1<15)
96 for (i = 0; i < dstW; i++) {
98 int srcPos = filterPos[i];
101 for (j = 0; j < filterSize; j++) {
102 val += src[srcPos + j] * filter[filterSize * i + j];
104 // filter=14 bit, input=16 bit, output=30 bit, >> 11 makes 19 bit
105 dst[i] = FFMIN(val >> sh, (1 << 19) - 1);
109 static void hScale16To15_c(SwsContext *c, int16_t *dst, int dstW,
110 const uint8_t *_src, const int16_t *filter,
111 const int32_t *filterPos, int filterSize)
113 const AVPixFmtDescriptor *desc = av_pix_fmt_desc_get(c->srcFormat);
115 const uint16_t *src = (const uint16_t *) _src;
116 int sh = desc->comp[0].depth_minus1;
119 sh= isAnyRGB(c->srcFormat) || c->srcFormat==AV_PIX_FMT_PAL8 ? 13 : desc->comp[0].depth_minus1;
121 for (i = 0; i < dstW; i++) {
123 int srcPos = filterPos[i];
126 for (j = 0; j < filterSize; j++) {
127 val += src[srcPos + j] * filter[filterSize * i + j];
129 // filter=14 bit, input=16 bit, output=30 bit, >> 15 makes 15 bit
130 dst[i] = FFMIN(val >> sh, (1 << 15) - 1);
134 // bilinear / bicubic scaling
135 static void hScale8To15_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 for (i = 0; i < dstW; i++) {
142 int srcPos = filterPos[i];
144 for (j = 0; j < filterSize; j++) {
145 val += ((int)src[srcPos + j]) * filter[filterSize * i + j];
147 dst[i] = FFMIN(val >> 7, (1 << 15) - 1); // the cubic equation does overflow ...
151 static void hScale8To19_c(SwsContext *c, int16_t *_dst, int dstW,
152 const uint8_t *src, const int16_t *filter,
153 const int32_t *filterPos, int filterSize)
156 int32_t *dst = (int32_t *) _dst;
157 for (i = 0; i < dstW; i++) {
159 int srcPos = filterPos[i];
161 for (j = 0; j < filterSize; j++) {
162 val += ((int)src[srcPos + j]) * filter[filterSize * i + j];
164 dst[i] = FFMIN(val >> 3, (1 << 19) - 1); // the cubic equation does overflow ...
168 // FIXME all pal and rgb srcFormats could do this conversion as well
169 // FIXME all scalers more complex than bilinear could do half of this transform
170 static void chrRangeToJpeg_c(int16_t *dstU, int16_t *dstV, int width)
173 for (i = 0; i < width; i++) {
174 dstU[i] = (FFMIN(dstU[i], 30775) * 4663 - 9289992) >> 12; // -264
175 dstV[i] = (FFMIN(dstV[i], 30775) * 4663 - 9289992) >> 12; // -264
179 static void chrRangeFromJpeg_c(int16_t *dstU, int16_t *dstV, int width)
182 for (i = 0; i < width; i++) {
183 dstU[i] = (dstU[i] * 1799 + 4081085) >> 11; // 1469
184 dstV[i] = (dstV[i] * 1799 + 4081085) >> 11; // 1469
188 static void lumRangeToJpeg_c(int16_t *dst, int width)
191 for (i = 0; i < width; i++)
192 dst[i] = (FFMIN(dst[i], 30189) * 19077 - 39057361) >> 14;
195 static void lumRangeFromJpeg_c(int16_t *dst, int width)
198 for (i = 0; i < width; i++)
199 dst[i] = (dst[i] * 14071 + 33561947) >> 14;
202 static void chrRangeToJpeg16_c(int16_t *_dstU, int16_t *_dstV, int width)
205 int32_t *dstU = (int32_t *) _dstU;
206 int32_t *dstV = (int32_t *) _dstV;
207 for (i = 0; i < width; i++) {
208 dstU[i] = (FFMIN(dstU[i], 30775 << 4) * 4663 - (9289992 << 4)) >> 12; // -264
209 dstV[i] = (FFMIN(dstV[i], 30775 << 4) * 4663 - (9289992 << 4)) >> 12; // -264
213 static void chrRangeFromJpeg16_c(int16_t *_dstU, int16_t *_dstV, int width)
216 int32_t *dstU = (int32_t *) _dstU;
217 int32_t *dstV = (int32_t *) _dstV;
218 for (i = 0; i < width; i++) {
219 dstU[i] = (dstU[i] * 1799 + (4081085 << 4)) >> 11; // 1469
220 dstV[i] = (dstV[i] * 1799 + (4081085 << 4)) >> 11; // 1469
224 static void lumRangeToJpeg16_c(int16_t *_dst, int width)
227 int32_t *dst = (int32_t *) _dst;
228 for (i = 0; i < width; i++) {
229 dst[i] = ((int)(FFMIN(dst[i], 30189 << 4) * 4769U - (39057361 << 2))) >> 12;
233 static void lumRangeFromJpeg16_c(int16_t *_dst, int width)
236 int32_t *dst = (int32_t *) _dst;
237 for (i = 0; i < width; i++)
238 dst[i] = (dst[i]*(14071/4) + (33561947<<4)/4)>>12;
241 // *** horizontal scale Y line to temp buffer
242 static av_always_inline void hyscale(SwsContext *c, int16_t *dst, int dstWidth,
243 const uint8_t *src_in[4],
245 const int16_t *hLumFilter,
246 const int32_t *hLumFilterPos,
248 uint8_t *formatConvBuffer,
249 uint32_t *pal, int isAlpha)
251 void (*toYV12)(uint8_t *, const uint8_t *, const uint8_t *, const uint8_t *, int, uint32_t *) =
252 isAlpha ? c->alpToYV12 : c->lumToYV12;
253 void (*convertRange)(int16_t *, int) = isAlpha ? NULL : c->lumConvertRange;
254 const uint8_t *src = src_in[isAlpha ? 3 : 0];
257 toYV12(formatConvBuffer, src, src_in[1], src_in[2], srcW, pal);
258 src = formatConvBuffer;
259 } else if (c->readLumPlanar && !isAlpha) {
260 c->readLumPlanar(formatConvBuffer, src_in, srcW, c->input_rgb2yuv_table);
261 src = formatConvBuffer;
262 } else if (c->readAlpPlanar && isAlpha) {
263 c->readAlpPlanar(formatConvBuffer, src_in, srcW, NULL);
264 src = formatConvBuffer;
267 if (!c->hyscale_fast) {
268 c->hyScale(c, dst, dstWidth, src, hLumFilter,
269 hLumFilterPos, hLumFilterSize);
270 } else { // fast bilinear upscale / crap downscale
271 c->hyscale_fast(c, dst, dstWidth, src, srcW, xInc);
275 convertRange(dst, dstWidth);
278 static av_always_inline void hcscale(SwsContext *c, int16_t *dst1,
279 int16_t *dst2, int dstWidth,
280 const uint8_t *src_in[4],
282 const int16_t *hChrFilter,
283 const int32_t *hChrFilterPos,
285 uint8_t *formatConvBuffer, uint32_t *pal)
287 const uint8_t *src1 = src_in[1], *src2 = src_in[2];
289 uint8_t *buf2 = formatConvBuffer +
290 FFALIGN(srcW*2+78, 16);
291 c->chrToYV12(formatConvBuffer, buf2, src_in[0], src1, src2, srcW, pal);
292 src1= formatConvBuffer;
294 } else if (c->readChrPlanar) {
295 uint8_t *buf2 = formatConvBuffer +
296 FFALIGN(srcW*2+78, 16);
297 c->readChrPlanar(formatConvBuffer, buf2, src_in, srcW, c->input_rgb2yuv_table);
298 src1 = formatConvBuffer;
302 if (!c->hcscale_fast) {
303 c->hcScale(c, dst1, dstWidth, src1, hChrFilter, hChrFilterPos, hChrFilterSize);
304 c->hcScale(c, dst2, dstWidth, src2, hChrFilter, hChrFilterPos, hChrFilterSize);
305 } else { // fast bilinear upscale / crap downscale
306 c->hcscale_fast(c, dst1, dst2, dstWidth, src1, src2, srcW, xInc);
309 if (c->chrConvertRange)
310 c->chrConvertRange(dst1, dst2, dstWidth);
313 #define DEBUG_SWSCALE_BUFFERS 0
314 #define DEBUG_BUFFERS(...) \
315 if (DEBUG_SWSCALE_BUFFERS) \
316 av_log(c, AV_LOG_DEBUG, __VA_ARGS__)
318 static int swscale(SwsContext *c, const uint8_t *src[],
319 int srcStride[], int srcSliceY,
320 int srcSliceH, uint8_t *dst[], int dstStride[])
322 /* load a few things into local vars to make the code more readable?
324 const int srcW = c->srcW;
325 const int dstW = c->dstW;
326 const int dstH = c->dstH;
327 const int chrDstW = c->chrDstW;
328 const int chrSrcW = c->chrSrcW;
329 const int lumXInc = c->lumXInc;
330 const int chrXInc = c->chrXInc;
331 const enum AVPixelFormat dstFormat = c->dstFormat;
332 const int flags = c->flags;
333 int32_t *vLumFilterPos = c->vLumFilterPos;
334 int32_t *vChrFilterPos = c->vChrFilterPos;
335 int32_t *hLumFilterPos = c->hLumFilterPos;
336 int32_t *hChrFilterPos = c->hChrFilterPos;
337 int16_t *hLumFilter = c->hLumFilter;
338 int16_t *hChrFilter = c->hChrFilter;
339 int32_t *lumMmxFilter = c->lumMmxFilter;
340 int32_t *chrMmxFilter = c->chrMmxFilter;
341 const int vLumFilterSize = c->vLumFilterSize;
342 const int vChrFilterSize = c->vChrFilterSize;
343 const int hLumFilterSize = c->hLumFilterSize;
344 const int hChrFilterSize = c->hChrFilterSize;
345 int16_t **lumPixBuf = c->lumPixBuf;
346 int16_t **chrUPixBuf = c->chrUPixBuf;
347 int16_t **chrVPixBuf = c->chrVPixBuf;
348 int16_t **alpPixBuf = c->alpPixBuf;
349 const int vLumBufSize = c->vLumBufSize;
350 const int vChrBufSize = c->vChrBufSize;
351 uint8_t *formatConvBuffer = c->formatConvBuffer;
352 uint32_t *pal = c->pal_yuv;
353 yuv2planar1_fn yuv2plane1 = c->yuv2plane1;
354 yuv2planarX_fn yuv2planeX = c->yuv2planeX;
355 yuv2interleavedX_fn yuv2nv12cX = c->yuv2nv12cX;
356 yuv2packed1_fn yuv2packed1 = c->yuv2packed1;
357 yuv2packed2_fn yuv2packed2 = c->yuv2packed2;
358 yuv2packedX_fn yuv2packedX = c->yuv2packedX;
359 yuv2anyX_fn yuv2anyX = c->yuv2anyX;
360 const int chrSrcSliceY = srcSliceY >> c->chrSrcVSubSample;
361 const int chrSrcSliceH = FF_CEIL_RSHIFT(srcSliceH, c->chrSrcVSubSample);
362 int should_dither = is9_OR_10BPS(c->srcFormat) ||
363 is16BPS(c->srcFormat);
366 /* vars which will change and which we need to store back in the context */
368 int lumBufIndex = c->lumBufIndex;
369 int chrBufIndex = c->chrBufIndex;
370 int lastInLumBuf = c->lastInLumBuf;
371 int lastInChrBuf = c->lastInChrBuf;
372 int perform_gamma = c->is_internal_gamma;
375 if (!usePal(c->srcFormat)) {
376 pal = c->input_rgb2yuv_table;
379 if (isPacked(c->srcFormat)) {
387 srcStride[3] = srcStride[0];
389 srcStride[1] <<= c->vChrDrop;
390 srcStride[2] <<= c->vChrDrop;
392 DEBUG_BUFFERS("swscale() %p[%d] %p[%d] %p[%d] %p[%d] -> %p[%d] %p[%d] %p[%d] %p[%d]\n",
393 src[0], srcStride[0], src[1], srcStride[1],
394 src[2], srcStride[2], src[3], srcStride[3],
395 dst[0], dstStride[0], dst[1], dstStride[1],
396 dst[2], dstStride[2], dst[3], dstStride[3]);
397 DEBUG_BUFFERS("srcSliceY: %d srcSliceH: %d dstY: %d dstH: %d\n",
398 srcSliceY, srcSliceH, dstY, dstH);
399 DEBUG_BUFFERS("vLumFilterSize: %d vLumBufSize: %d vChrFilterSize: %d vChrBufSize: %d\n",
400 vLumFilterSize, vLumBufSize, vChrFilterSize, vChrBufSize);
402 if (dstStride[0]&15 || dstStride[1]&15 ||
403 dstStride[2]&15 || dstStride[3]&15) {
404 static int warnedAlready = 0; // FIXME maybe move this into the context
405 if (flags & SWS_PRINT_INFO && !warnedAlready) {
406 av_log(c, AV_LOG_WARNING,
407 "Warning: dstStride is not aligned!\n"
408 " ->cannot do aligned memory accesses anymore\n");
413 if ( (uintptr_t)dst[0]&15 || (uintptr_t)dst[1]&15 || (uintptr_t)dst[2]&15
414 || (uintptr_t)src[0]&15 || (uintptr_t)src[1]&15 || (uintptr_t)src[2]&15
415 || dstStride[0]&15 || dstStride[1]&15 || dstStride[2]&15 || dstStride[3]&15
416 || srcStride[0]&15 || srcStride[1]&15 || srcStride[2]&15 || srcStride[3]&15
418 static int warnedAlready=0;
419 int cpu_flags = av_get_cpu_flags();
420 if (HAVE_MMXEXT && (cpu_flags & AV_CPU_FLAG_SSE2) && !warnedAlready){
421 av_log(c, AV_LOG_WARNING, "Warning: data is not aligned! This can lead to a speedloss\n");
426 /* Note the user might start scaling the picture in the middle so this
427 * will not get executed. This is not really intended but works
428 * currently, so people might do it. */
429 if (srcSliceY == 0) {
437 if (!should_dither) {
438 c->chrDither8 = c->lumDither8 = sws_pb_64;
442 for (; dstY < dstH; dstY++) {
443 const int chrDstY = dstY >> c->chrDstVSubSample;
445 dst[0] + dstStride[0] * dstY,
446 dst[1] + dstStride[1] * chrDstY,
447 dst[2] + dstStride[2] * chrDstY,
448 (CONFIG_SWSCALE_ALPHA && alpPixBuf) ? dst[3] + dstStride[3] * dstY : NULL,
450 int use_mmx_vfilter= c->use_mmx_vfilter;
452 // First line needed as input
453 const int firstLumSrcY = FFMAX(1 - vLumFilterSize, vLumFilterPos[dstY]);
454 const int firstLumSrcY2 = FFMAX(1 - vLumFilterSize, vLumFilterPos[FFMIN(dstY | ((1 << c->chrDstVSubSample) - 1), dstH - 1)]);
455 // First line needed as input
456 const int firstChrSrcY = FFMAX(1 - vChrFilterSize, vChrFilterPos[chrDstY]);
458 // Last line needed as input
459 int lastLumSrcY = FFMIN(c->srcH, firstLumSrcY + vLumFilterSize) - 1;
460 int lastLumSrcY2 = FFMIN(c->srcH, firstLumSrcY2 + vLumFilterSize) - 1;
461 int lastChrSrcY = FFMIN(c->chrSrcH, firstChrSrcY + vChrFilterSize) - 1;
464 // handle holes (FAST_BILINEAR & weird filters)
465 if (firstLumSrcY > lastInLumBuf)
466 lastInLumBuf = firstLumSrcY - 1;
467 if (firstChrSrcY > lastInChrBuf)
468 lastInChrBuf = firstChrSrcY - 1;
469 av_assert0(firstLumSrcY >= lastInLumBuf - vLumBufSize + 1);
470 av_assert0(firstChrSrcY >= lastInChrBuf - vChrBufSize + 1);
472 DEBUG_BUFFERS("dstY: %d\n", dstY);
473 DEBUG_BUFFERS("\tfirstLumSrcY: %d lastLumSrcY: %d lastInLumBuf: %d\n",
474 firstLumSrcY, lastLumSrcY, lastInLumBuf);
475 DEBUG_BUFFERS("\tfirstChrSrcY: %d lastChrSrcY: %d lastInChrBuf: %d\n",
476 firstChrSrcY, lastChrSrcY, lastInChrBuf);
478 // Do we have enough lines in this slice to output the dstY line
479 enough_lines = lastLumSrcY2 < srcSliceY + srcSliceH &&
480 lastChrSrcY < FF_CEIL_RSHIFT(srcSliceY + srcSliceH, c->chrSrcVSubSample);
483 lastLumSrcY = srcSliceY + srcSliceH - 1;
484 lastChrSrcY = chrSrcSliceY + chrSrcSliceH - 1;
485 DEBUG_BUFFERS("buffering slice: lastLumSrcY %d lastChrSrcY %d\n",
486 lastLumSrcY, lastChrSrcY);
489 // Do horizontal scaling
490 while (lastInLumBuf < lastLumSrcY) {
491 const uint8_t *src1[4] = {
492 src[0] + (lastInLumBuf + 1 - srcSliceY) * srcStride[0],
493 src[1] + (lastInLumBuf + 1 - srcSliceY) * srcStride[1],
494 src[2] + (lastInLumBuf + 1 - srcSliceY) * srcStride[2],
495 src[3] + (lastInLumBuf + 1 - srcSliceY) * srcStride[3],
498 av_assert0(lumBufIndex < 2 * vLumBufSize);
499 av_assert0(lastInLumBuf + 1 - srcSliceY < srcSliceH);
500 av_assert0(lastInLumBuf + 1 - srcSliceY >= 0);
503 gamma_convert((uint8_t **)src1, srcW, c->inv_gamma);
505 hyscale(c, lumPixBuf[lumBufIndex], dstW, src1, srcW, lumXInc,
506 hLumFilter, hLumFilterPos, hLumFilterSize,
507 formatConvBuffer, pal, 0);
508 if (CONFIG_SWSCALE_ALPHA && alpPixBuf)
509 hyscale(c, alpPixBuf[lumBufIndex], dstW, src1, srcW,
510 lumXInc, hLumFilter, hLumFilterPos, hLumFilterSize,
511 formatConvBuffer, pal, 1);
513 DEBUG_BUFFERS("\t\tlumBufIndex %d: lastInLumBuf: %d\n",
514 lumBufIndex, lastInLumBuf);
516 while (lastInChrBuf < lastChrSrcY) {
517 const uint8_t *src1[4] = {
518 src[0] + (lastInChrBuf + 1 - chrSrcSliceY) * srcStride[0],
519 src[1] + (lastInChrBuf + 1 - chrSrcSliceY) * srcStride[1],
520 src[2] + (lastInChrBuf + 1 - chrSrcSliceY) * srcStride[2],
521 src[3] + (lastInChrBuf + 1 - chrSrcSliceY) * srcStride[3],
524 av_assert0(chrBufIndex < 2 * vChrBufSize);
525 av_assert0(lastInChrBuf + 1 - chrSrcSliceY < (chrSrcSliceH));
526 av_assert0(lastInChrBuf + 1 - chrSrcSliceY >= 0);
527 // FIXME replace parameters through context struct (some at least)
529 if (c->needs_hcscale)
530 hcscale(c, chrUPixBuf[chrBufIndex], chrVPixBuf[chrBufIndex],
531 chrDstW, src1, chrSrcW, chrXInc,
532 hChrFilter, hChrFilterPos, hChrFilterSize,
533 formatConvBuffer, pal);
535 DEBUG_BUFFERS("\t\tchrBufIndex %d: lastInChrBuf: %d\n",
536 chrBufIndex, lastInChrBuf);
538 // wrap buf index around to stay inside the ring buffer
539 if (lumBufIndex >= vLumBufSize)
540 lumBufIndex -= vLumBufSize;
541 if (chrBufIndex >= vChrBufSize)
542 chrBufIndex -= vChrBufSize;
544 break; // we can't output a dstY line so let's try with the next slice
547 ff_updateMMXDitherTables(c, dstY, lumBufIndex, chrBufIndex,
548 lastInLumBuf, lastInChrBuf);
551 c->chrDither8 = ff_dither_8x8_128[chrDstY & 7];
552 c->lumDither8 = ff_dither_8x8_128[dstY & 7];
554 if (dstY >= dstH - 2) {
555 /* hmm looks like we can't use MMX here without overwriting
556 * this array's tail */
557 ff_sws_init_output_funcs(c, &yuv2plane1, &yuv2planeX, &yuv2nv12cX,
558 &yuv2packed1, &yuv2packed2, &yuv2packedX, &yuv2anyX);
563 const int16_t **lumSrcPtr = (const int16_t **)(void*) lumPixBuf + lumBufIndex + firstLumSrcY - lastInLumBuf + vLumBufSize;
564 const int16_t **chrUSrcPtr = (const int16_t **)(void*) chrUPixBuf + chrBufIndex + firstChrSrcY - lastInChrBuf + vChrBufSize;
565 const int16_t **chrVSrcPtr = (const int16_t **)(void*) chrVPixBuf + chrBufIndex + firstChrSrcY - lastInChrBuf + vChrBufSize;
566 const int16_t **alpSrcPtr = (CONFIG_SWSCALE_ALPHA && alpPixBuf) ?
567 (const int16_t **)(void*) alpPixBuf + lumBufIndex + firstLumSrcY - lastInLumBuf + vLumBufSize : NULL;
568 int16_t *vLumFilter = c->vLumFilter;
569 int16_t *vChrFilter = c->vChrFilter;
571 if (isPlanarYUV(dstFormat) ||
572 (isGray(dstFormat) && !isALPHA(dstFormat))) { // YV12 like
573 const int chrSkipMask = (1 << c->chrDstVSubSample) - 1;
575 vLumFilter += dstY * vLumFilterSize;
576 vChrFilter += chrDstY * vChrFilterSize;
578 // av_assert0(use_mmx_vfilter != (
579 // yuv2planeX == yuv2planeX_10BE_c
580 // || yuv2planeX == yuv2planeX_10LE_c
581 // || yuv2planeX == yuv2planeX_9BE_c
582 // || yuv2planeX == yuv2planeX_9LE_c
583 // || yuv2planeX == yuv2planeX_16BE_c
584 // || yuv2planeX == yuv2planeX_16LE_c
585 // || yuv2planeX == yuv2planeX_8_c) || !ARCH_X86);
588 vLumFilter= (int16_t *)c->lumMmxFilter;
589 vChrFilter= (int16_t *)c->chrMmxFilter;
592 if (vLumFilterSize == 1) {
593 yuv2plane1(lumSrcPtr[0], dest[0], dstW, c->lumDither8, 0);
595 yuv2planeX(vLumFilter, vLumFilterSize,
597 dstW, c->lumDither8, 0);
600 if (!((dstY & chrSkipMask) || isGray(dstFormat))) {
602 yuv2nv12cX(c, vChrFilter,
603 vChrFilterSize, chrUSrcPtr, chrVSrcPtr,
605 } else if (vChrFilterSize == 1) {
606 yuv2plane1(chrUSrcPtr[0], dest[1], chrDstW, c->chrDither8, 0);
607 yuv2plane1(chrVSrcPtr[0], dest[2], chrDstW, c->chrDither8, 3);
609 yuv2planeX(vChrFilter,
610 vChrFilterSize, chrUSrcPtr, dest[1],
611 chrDstW, c->chrDither8, 0);
612 yuv2planeX(vChrFilter,
613 vChrFilterSize, chrVSrcPtr, dest[2],
614 chrDstW, c->chrDither8, use_mmx_vfilter ? (c->uv_offx2 >> 1) : 3);
618 if (CONFIG_SWSCALE_ALPHA && alpPixBuf) {
620 vLumFilter= (int16_t *)c->alpMmxFilter;
622 if (vLumFilterSize == 1) {
623 yuv2plane1(alpSrcPtr[0], dest[3], dstW,
626 yuv2planeX(vLumFilter,
627 vLumFilterSize, alpSrcPtr, dest[3],
628 dstW, c->lumDither8, 0);
631 } else if (yuv2packedX) {
632 av_assert1(lumSrcPtr + vLumFilterSize - 1 < (const int16_t **)lumPixBuf + vLumBufSize * 2);
633 av_assert1(chrUSrcPtr + vChrFilterSize - 1 < (const int16_t **)chrUPixBuf + vChrBufSize * 2);
634 if (c->yuv2packed1 && vLumFilterSize == 1 &&
635 vChrFilterSize <= 2) { // unscaled RGB
636 int chrAlpha = vChrFilterSize == 1 ? 0 : vChrFilter[2 * dstY + 1];
637 yuv2packed1(c, *lumSrcPtr, chrUSrcPtr, chrVSrcPtr,
638 alpPixBuf ? *alpSrcPtr : NULL,
639 dest[0], dstW, chrAlpha, dstY);
640 } else if (c->yuv2packed2 && vLumFilterSize == 2 &&
641 vChrFilterSize == 2) { // bilinear upscale RGB
642 int lumAlpha = vLumFilter[2 * dstY + 1];
643 int chrAlpha = vChrFilter[2 * dstY + 1];
645 lumMmxFilter[3] = vLumFilter[2 * dstY] * 0x10001;
647 chrMmxFilter[3] = vChrFilter[2 * chrDstY] * 0x10001;
648 yuv2packed2(c, lumSrcPtr, chrUSrcPtr, chrVSrcPtr,
649 alpPixBuf ? alpSrcPtr : NULL,
650 dest[0], dstW, lumAlpha, chrAlpha, dstY);
651 } else { // general RGB
652 yuv2packedX(c, vLumFilter + dstY * vLumFilterSize,
653 lumSrcPtr, vLumFilterSize,
654 vChrFilter + dstY * vChrFilterSize,
655 chrUSrcPtr, chrVSrcPtr, vChrFilterSize,
656 alpSrcPtr, dest[0], dstW, dstY);
659 av_assert1(!yuv2packed1 && !yuv2packed2);
660 yuv2anyX(c, vLumFilter + dstY * vLumFilterSize,
661 lumSrcPtr, vLumFilterSize,
662 vChrFilter + dstY * vChrFilterSize,
663 chrUSrcPtr, chrVSrcPtr, vChrFilterSize,
664 alpSrcPtr, dest, dstW, dstY);
667 gamma_convert(dest, dstW, c->gamma);
670 if (isPlanar(dstFormat) && isALPHA(dstFormat) && !alpPixBuf) {
672 int height = dstY - lastDstY;
674 if (is16BPS(dstFormat) || isNBPS(dstFormat)) {
675 const AVPixFmtDescriptor *desc = av_pix_fmt_desc_get(dstFormat);
676 fillPlane16(dst[3], dstStride[3], length, height, lastDstY,
677 1, desc->comp[3].depth_minus1,
680 fillPlane(dst[3], dstStride[3], length, height, lastDstY, 255);
683 #if HAVE_MMXEXT_INLINE
684 if (av_get_cpu_flags() & AV_CPU_FLAG_MMXEXT)
685 __asm__ volatile ("sfence" ::: "memory");
689 /* store changed local vars back in the context */
691 c->lumBufIndex = lumBufIndex;
692 c->chrBufIndex = chrBufIndex;
693 c->lastInLumBuf = lastInLumBuf;
694 c->lastInChrBuf = lastInChrBuf;
696 return dstY - lastDstY;
699 av_cold void ff_sws_init_range_convert(SwsContext *c)
701 c->lumConvertRange = NULL;
702 c->chrConvertRange = NULL;
703 if (c->srcRange != c->dstRange && !isAnyRGB(c->dstFormat)) {
704 if (c->dstBpc <= 14) {
706 c->lumConvertRange = lumRangeFromJpeg_c;
707 c->chrConvertRange = chrRangeFromJpeg_c;
709 c->lumConvertRange = lumRangeToJpeg_c;
710 c->chrConvertRange = chrRangeToJpeg_c;
714 c->lumConvertRange = lumRangeFromJpeg16_c;
715 c->chrConvertRange = chrRangeFromJpeg16_c;
717 c->lumConvertRange = lumRangeToJpeg16_c;
718 c->chrConvertRange = chrRangeToJpeg16_c;
724 static av_cold void sws_init_swscale(SwsContext *c)
726 enum AVPixelFormat srcFormat = c->srcFormat;
728 ff_sws_init_output_funcs(c, &c->yuv2plane1, &c->yuv2planeX,
729 &c->yuv2nv12cX, &c->yuv2packed1,
730 &c->yuv2packed2, &c->yuv2packedX, &c->yuv2anyX);
732 ff_sws_init_input_funcs(c);
735 if (c->srcBpc == 8) {
736 if (c->dstBpc <= 14) {
737 c->hyScale = c->hcScale = hScale8To15_c;
738 if (c->flags & SWS_FAST_BILINEAR) {
739 c->hyscale_fast = ff_hyscale_fast_c;
740 c->hcscale_fast = ff_hcscale_fast_c;
743 c->hyScale = c->hcScale = hScale8To19_c;
746 c->hyScale = c->hcScale = c->dstBpc > 14 ? hScale16To19_c
750 ff_sws_init_range_convert(c);
752 if (!(isGray(srcFormat) || isGray(c->dstFormat) ||
753 srcFormat == AV_PIX_FMT_MONOBLACK || srcFormat == AV_PIX_FMT_MONOWHITE))
754 c->needs_hcscale = 1;
757 SwsFunc ff_getSwsFunc(SwsContext *c)
762 ff_sws_init_swscale_ppc(c);
764 ff_sws_init_swscale_x86(c);
769 static void reset_ptr(const uint8_t *src[], enum AVPixelFormat format)
771 if (!isALPHA(format))
773 if (!isPlanar(format)) {
774 src[3] = src[2] = NULL;
781 static int check_image_pointers(const uint8_t * const data[4], enum AVPixelFormat pix_fmt,
782 const int linesizes[4])
784 const AVPixFmtDescriptor *desc = av_pix_fmt_desc_get(pix_fmt);
789 for (i = 0; i < 4; i++) {
790 int plane = desc->comp[i].plane;
791 if (!data[plane] || !linesizes[plane])
798 static void xyz12Torgb48(struct SwsContext *c, uint16_t *dst,
799 const uint16_t *src, int stride, int h)
802 const AVPixFmtDescriptor *desc = av_pix_fmt_desc_get(c->srcFormat);
804 for (yp=0; yp<h; yp++) {
805 for (xp=0; xp+2<stride; xp+=3) {
806 int x, y, z, r, g, b;
808 if (desc->flags & AV_PIX_FMT_FLAG_BE) {
809 x = AV_RB16(src + xp + 0);
810 y = AV_RB16(src + xp + 1);
811 z = AV_RB16(src + xp + 2);
813 x = AV_RL16(src + xp + 0);
814 y = AV_RL16(src + xp + 1);
815 z = AV_RL16(src + xp + 2);
818 x = c->xyzgamma[x>>4];
819 y = c->xyzgamma[y>>4];
820 z = c->xyzgamma[z>>4];
822 // convert from XYZlinear to sRGBlinear
823 r = c->xyz2rgb_matrix[0][0] * x +
824 c->xyz2rgb_matrix[0][1] * y +
825 c->xyz2rgb_matrix[0][2] * z >> 12;
826 g = c->xyz2rgb_matrix[1][0] * x +
827 c->xyz2rgb_matrix[1][1] * y +
828 c->xyz2rgb_matrix[1][2] * z >> 12;
829 b = c->xyz2rgb_matrix[2][0] * x +
830 c->xyz2rgb_matrix[2][1] * y +
831 c->xyz2rgb_matrix[2][2] * z >> 12;
833 // limit values to 12-bit depth
834 r = av_clip_uintp2(r, 12);
835 g = av_clip_uintp2(g, 12);
836 b = av_clip_uintp2(b, 12);
838 // convert from sRGBlinear to RGB and scale from 12bit to 16bit
839 if (desc->flags & AV_PIX_FMT_FLAG_BE) {
840 AV_WB16(dst + xp + 0, c->rgbgamma[r] << 4);
841 AV_WB16(dst + xp + 1, c->rgbgamma[g] << 4);
842 AV_WB16(dst + xp + 2, c->rgbgamma[b] << 4);
844 AV_WL16(dst + xp + 0, c->rgbgamma[r] << 4);
845 AV_WL16(dst + xp + 1, c->rgbgamma[g] << 4);
846 AV_WL16(dst + xp + 2, c->rgbgamma[b] << 4);
854 static void rgb48Toxyz12(struct SwsContext *c, uint16_t *dst,
855 const uint16_t *src, int stride, int h)
858 const AVPixFmtDescriptor *desc = av_pix_fmt_desc_get(c->dstFormat);
860 for (yp=0; yp<h; yp++) {
861 for (xp=0; xp+2<stride; xp+=3) {
862 int x, y, z, r, g, b;
864 if (desc->flags & AV_PIX_FMT_FLAG_BE) {
865 r = AV_RB16(src + xp + 0);
866 g = AV_RB16(src + xp + 1);
867 b = AV_RB16(src + xp + 2);
869 r = AV_RL16(src + xp + 0);
870 g = AV_RL16(src + xp + 1);
871 b = AV_RL16(src + xp + 2);
874 r = c->rgbgammainv[r>>4];
875 g = c->rgbgammainv[g>>4];
876 b = c->rgbgammainv[b>>4];
878 // convert from sRGBlinear to XYZlinear
879 x = c->rgb2xyz_matrix[0][0] * r +
880 c->rgb2xyz_matrix[0][1] * g +
881 c->rgb2xyz_matrix[0][2] * b >> 12;
882 y = c->rgb2xyz_matrix[1][0] * r +
883 c->rgb2xyz_matrix[1][1] * g +
884 c->rgb2xyz_matrix[1][2] * b >> 12;
885 z = c->rgb2xyz_matrix[2][0] * r +
886 c->rgb2xyz_matrix[2][1] * g +
887 c->rgb2xyz_matrix[2][2] * b >> 12;
889 // limit values to 12-bit depth
890 x = av_clip_uintp2(x, 12);
891 y = av_clip_uintp2(y, 12);
892 z = av_clip_uintp2(z, 12);
894 // convert from XYZlinear to X'Y'Z' and scale from 12bit to 16bit
895 if (desc->flags & AV_PIX_FMT_FLAG_BE) {
896 AV_WB16(dst + xp + 0, c->xyzgammainv[x] << 4);
897 AV_WB16(dst + xp + 1, c->xyzgammainv[y] << 4);
898 AV_WB16(dst + xp + 2, c->xyzgammainv[z] << 4);
900 AV_WL16(dst + xp + 0, c->xyzgammainv[x] << 4);
901 AV_WL16(dst + xp + 1, c->xyzgammainv[y] << 4);
902 AV_WL16(dst + xp + 2, c->xyzgammainv[z] << 4);
911 * swscale wrapper, so we don't need to export the SwsContext.
912 * Assumes planar YUV to be in YUV order instead of YVU.
914 int attribute_align_arg sws_scale(struct SwsContext *c,
915 const uint8_t * const srcSlice[],
916 const int srcStride[], int srcSliceY,
917 int srcSliceH, uint8_t *const dst[],
918 const int dstStride[])
921 const uint8_t *src2[4];
923 uint8_t *rgb0_tmp = NULL;
925 if (!srcStride || !dstStride || !dst || !srcSlice) {
926 av_log(c, AV_LOG_ERROR, "One of the input parameters to sws_scale() is NULL, please check the calling code\n");
930 if (c->gamma_flag && c->cascaded_context[0]) {
933 ret = sws_scale(c->cascaded_context[0],
934 srcSlice, srcStride, srcSliceY, srcSliceH,
935 c->cascaded_tmp, c->cascaded_tmpStride);
940 if (c->cascaded_context[2])
941 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);
943 ret = sws_scale(c->cascaded_context[1], (const uint8_t * const *)c->cascaded_tmp, c->cascaded_tmpStride, srcSliceY, srcSliceH, dst, dstStride);
948 if (c->cascaded_context[2]) {
949 ret = sws_scale(c->cascaded_context[2],
950 (const uint8_t * const *)c->cascaded1_tmp, c->cascaded1_tmpStride, c->cascaded_context[1]->dstY - ret, c->cascaded_context[1]->dstY,
956 if (c->cascaded_context[0] && srcSliceY == 0 && srcSliceH == c->cascaded_context[0]->srcH) {
957 ret = sws_scale(c->cascaded_context[0],
958 srcSlice, srcStride, srcSliceY, srcSliceH,
959 c->cascaded_tmp, c->cascaded_tmpStride);
962 ret = sws_scale(c->cascaded_context[1],
963 (const uint8_t * const * )c->cascaded_tmp, c->cascaded_tmpStride, 0, c->cascaded_context[0]->dstH,
968 memcpy(src2, srcSlice, sizeof(src2));
969 memcpy(dst2, dst, sizeof(dst2));
971 // do not mess up sliceDir if we have a "trailing" 0-size slice
975 if (!check_image_pointers(srcSlice, c->srcFormat, srcStride)) {
976 av_log(c, AV_LOG_ERROR, "bad src image pointers\n");
979 if (!check_image_pointers((const uint8_t* const*)dst, c->dstFormat, dstStride)) {
980 av_log(c, AV_LOG_ERROR, "bad dst image pointers\n");
984 if (c->sliceDir == 0 && srcSliceY != 0 && srcSliceY + srcSliceH != c->srcH) {
985 av_log(c, AV_LOG_ERROR, "Slices start in the middle!\n");
988 if (c->sliceDir == 0) {
989 if (srcSliceY == 0) c->sliceDir = 1; else c->sliceDir = -1;
992 if (usePal(c->srcFormat)) {
993 for (i = 0; i < 256; i++) {
994 int r, g, b, y, u, v, a = 0xff;
995 if (c->srcFormat == AV_PIX_FMT_PAL8) {
996 uint32_t p = ((const uint32_t *)(srcSlice[1]))[i];
997 a = (p >> 24) & 0xFF;
998 r = (p >> 16) & 0xFF;
1001 } else if (c->srcFormat == AV_PIX_FMT_RGB8) {
1002 r = ( i >> 5 ) * 36;
1003 g = ((i >> 2) & 7) * 36;
1005 } else if (c->srcFormat == AV_PIX_FMT_BGR8) {
1006 b = ( i >> 6 ) * 85;
1007 g = ((i >> 3) & 7) * 36;
1009 } else if (c->srcFormat == AV_PIX_FMT_RGB4_BYTE) {
1010 r = ( i >> 3 ) * 255;
1011 g = ((i >> 1) & 3) * 85;
1013 } else if (c->srcFormat == AV_PIX_FMT_GRAY8 || c->srcFormat == AV_PIX_FMT_GRAY8A) {
1016 av_assert1(c->srcFormat == AV_PIX_FMT_BGR4_BYTE);
1017 b = ( i >> 3 ) * 255;
1018 g = ((i >> 1) & 3) * 85;
1021 #define RGB2YUV_SHIFT 15
1022 #define BY ( (int) (0.114 * 219 / 255 * (1 << RGB2YUV_SHIFT) + 0.5))
1023 #define BV (-(int) (0.081 * 224 / 255 * (1 << RGB2YUV_SHIFT) + 0.5))
1024 #define BU ( (int) (0.500 * 224 / 255 * (1 << RGB2YUV_SHIFT) + 0.5))
1025 #define GY ( (int) (0.587 * 219 / 255 * (1 << RGB2YUV_SHIFT) + 0.5))
1026 #define GV (-(int) (0.419 * 224 / 255 * (1 << RGB2YUV_SHIFT) + 0.5))
1027 #define GU (-(int) (0.331 * 224 / 255 * (1 << RGB2YUV_SHIFT) + 0.5))
1028 #define RY ( (int) (0.299 * 219 / 255 * (1 << RGB2YUV_SHIFT) + 0.5))
1029 #define RV ( (int) (0.500 * 224 / 255 * (1 << RGB2YUV_SHIFT) + 0.5))
1030 #define RU (-(int) (0.169 * 224 / 255 * (1 << RGB2YUV_SHIFT) + 0.5))
1032 y = av_clip_uint8((RY * r + GY * g + BY * b + ( 33 << (RGB2YUV_SHIFT - 1))) >> RGB2YUV_SHIFT);
1033 u = av_clip_uint8((RU * r + GU * g + BU * b + (257 << (RGB2YUV_SHIFT - 1))) >> RGB2YUV_SHIFT);
1034 v = av_clip_uint8((RV * r + GV * g + BV * b + (257 << (RGB2YUV_SHIFT - 1))) >> RGB2YUV_SHIFT);
1035 c->pal_yuv[i]= y + (u<<8) + (v<<16) + ((unsigned)a<<24);
1037 switch (c->dstFormat) {
1038 case AV_PIX_FMT_BGR32:
1040 case AV_PIX_FMT_RGB24:
1042 c->pal_rgb[i]= r + (g<<8) + (b<<16) + ((unsigned)a<<24);
1044 case AV_PIX_FMT_BGR32_1:
1046 case AV_PIX_FMT_BGR24:
1048 c->pal_rgb[i]= a + (r<<8) + (g<<16) + ((unsigned)b<<24);
1050 case AV_PIX_FMT_RGB32_1:
1052 case AV_PIX_FMT_RGB24:
1054 c->pal_rgb[i]= a + (b<<8) + (g<<16) + ((unsigned)r<<24);
1056 case AV_PIX_FMT_RGB32:
1058 case AV_PIX_FMT_BGR24:
1061 c->pal_rgb[i]= b + (g<<8) + (r<<16) + ((unsigned)a<<24);
1066 if (c->src0Alpha && !c->dst0Alpha && isALPHA(c->dstFormat)) {
1069 rgb0_tmp = av_malloc(FFABS(srcStride[0]) * srcSliceH + 32);
1071 return AVERROR(ENOMEM);
1073 base = srcStride[0] < 0 ? rgb0_tmp - srcStride[0] * (srcSliceH-1) : rgb0_tmp;
1074 for (y=0; y<srcSliceH; y++){
1075 memcpy(base + srcStride[0]*y, src2[0] + srcStride[0]*y, 4*c->srcW);
1076 for (x=c->src0Alpha-1; x<4*c->srcW; x+=4) {
1077 base[ srcStride[0]*y + x] = 0xFF;
1083 if (c->srcXYZ && !(c->dstXYZ && c->srcW==c->dstW && c->srcH==c->dstH)) {
1085 rgb0_tmp = av_malloc(FFABS(srcStride[0]) * srcSliceH + 32);
1087 return AVERROR(ENOMEM);
1089 base = srcStride[0] < 0 ? rgb0_tmp - srcStride[0] * (srcSliceH-1) : rgb0_tmp;
1091 xyz12Torgb48(c, (uint16_t*)base, (const uint16_t*)src2[0], srcStride[0]/2, srcSliceH);
1095 if (!srcSliceY && (c->flags & SWS_BITEXACT) && c->dither == SWS_DITHER_ED && c->dither_error[0])
1096 for (i = 0; i < 4; i++)
1097 memset(c->dither_error[i], 0, sizeof(c->dither_error[0][0]) * (c->dstW+2));
1100 // copy strides, so they can safely be modified
1101 if (c->sliceDir == 1) {
1102 // slices go from top to bottom
1103 int srcStride2[4] = { srcStride[0], srcStride[1], srcStride[2],
1105 int dstStride2[4] = { dstStride[0], dstStride[1], dstStride[2],
1108 reset_ptr(src2, c->srcFormat);
1109 reset_ptr((void*)dst2, c->dstFormat);
1111 /* reset slice direction at end of frame */
1112 if (srcSliceY + srcSliceH == c->srcH)
1115 ret = c->swscale(c, src2, srcStride2, srcSliceY, srcSliceH, dst2,
1118 // slices go from bottom to top => we flip the image internally
1119 int srcStride2[4] = { -srcStride[0], -srcStride[1], -srcStride[2],
1121 int dstStride2[4] = { -dstStride[0], -dstStride[1], -dstStride[2],
1124 src2[0] += (srcSliceH - 1) * srcStride[0];
1125 if (!usePal(c->srcFormat))
1126 src2[1] += ((srcSliceH >> c->chrSrcVSubSample) - 1) * srcStride[1];
1127 src2[2] += ((srcSliceH >> c->chrSrcVSubSample) - 1) * srcStride[2];
1128 src2[3] += (srcSliceH - 1) * srcStride[3];
1129 dst2[0] += ( c->dstH - 1) * dstStride[0];
1130 dst2[1] += ((c->dstH >> c->chrDstVSubSample) - 1) * dstStride[1];
1131 dst2[2] += ((c->dstH >> c->chrDstVSubSample) - 1) * dstStride[2];
1132 dst2[3] += ( c->dstH - 1) * dstStride[3];
1134 reset_ptr(src2, c->srcFormat);
1135 reset_ptr((void*)dst2, c->dstFormat);
1137 /* reset slice direction at end of frame */
1141 ret = c->swscale(c, src2, srcStride2, c->srcH-srcSliceY-srcSliceH,
1142 srcSliceH, dst2, dstStride2);
1146 if (c->dstXYZ && !(c->srcXYZ && c->srcW==c->dstW && c->srcH==c->dstH)) {
1147 /* replace on the same data */
1148 rgb48Toxyz12(c, (uint16_t*)dst2[0], (const uint16_t*)dst2[0], dstStride[0]/2, ret);