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/intreadwrite.h"
31 #include "libavutil/mathematics.h"
32 #include "libavutil/pixdesc.h"
35 #include "swscale_internal.h"
38 DECLARE_ALIGNED(8, const uint8_t, dither_8x8_128)[8][8] = {
39 { 36, 68, 60, 92, 34, 66, 58, 90, },
40 { 100, 4, 124, 28, 98, 2, 122, 26, },
41 { 52, 84, 44, 76, 50, 82, 42, 74, },
42 { 116, 20, 108, 12, 114, 18, 106, 10, },
43 { 32, 64, 56, 88, 38, 70, 62, 94, },
44 { 96, 0, 120, 24, 102, 6, 126, 30, },
45 { 48, 80, 40, 72, 54, 86, 46, 78, },
46 { 112, 16, 104, 8, 118, 22, 110, 14, },
49 DECLARE_ALIGNED(8, const uint8_t, ff_sws_pb_64)[8] = {
50 64, 64, 64, 64, 64, 64, 64, 64
53 static av_always_inline void fillPlane(uint8_t *plane, int stride, int width,
54 int height, int y, uint8_t val)
57 uint8_t *ptr = plane + stride * y;
58 for (i = 0; i < height; i++) {
59 memset(ptr, val, width);
64 static void hScale16To19_c(SwsContext *c, int16_t *_dst, int dstW,
65 const uint8_t *_src, const int16_t *filter,
66 const int32_t *filterPos, int filterSize)
68 const AVPixFmtDescriptor *desc = av_pix_fmt_desc_get(c->srcFormat);
70 int32_t *dst = (int32_t *) _dst;
71 const uint16_t *src = (const uint16_t *) _src;
72 int bits = desc->comp[0].depth_minus1;
75 if((isAnyRGB(c->srcFormat) || c->srcFormat==AV_PIX_FMT_PAL8) && desc->comp[0].depth_minus1<15)
78 for (i = 0; i < dstW; i++) {
80 int srcPos = filterPos[i];
83 for (j = 0; j < filterSize; j++) {
84 val += src[srcPos + j] * filter[filterSize * i + j];
86 // filter=14 bit, input=16 bit, output=30 bit, >> 11 makes 19 bit
87 dst[i] = FFMIN(val >> sh, (1 << 19) - 1);
91 static void hScale16To15_c(SwsContext *c, int16_t *dst, int dstW,
92 const uint8_t *_src, const int16_t *filter,
93 const int32_t *filterPos, int filterSize)
95 const AVPixFmtDescriptor *desc = av_pix_fmt_desc_get(c->srcFormat);
97 const uint16_t *src = (const uint16_t *) _src;
98 int sh = desc->comp[0].depth_minus1;
101 sh= isAnyRGB(c->srcFormat) || c->srcFormat==AV_PIX_FMT_PAL8 ? 13 : desc->comp[0].depth_minus1;
103 for (i = 0; i < dstW; i++) {
105 int srcPos = filterPos[i];
108 for (j = 0; j < filterSize; j++) {
109 val += src[srcPos + j] * filter[filterSize * i + j];
111 // filter=14 bit, input=16 bit, output=30 bit, >> 15 makes 15 bit
112 dst[i] = FFMIN(val >> sh, (1 << 15) - 1);
116 // bilinear / bicubic scaling
117 static void hScale8To15_c(SwsContext *c, int16_t *dst, int dstW,
118 const uint8_t *src, const int16_t *filter,
119 const int32_t *filterPos, int filterSize)
122 for (i = 0; i < dstW; i++) {
124 int srcPos = filterPos[i];
126 for (j = 0; j < filterSize; j++) {
127 val += ((int)src[srcPos + j]) * filter[filterSize * i + j];
129 dst[i] = FFMIN(val >> 7, (1 << 15) - 1); // the cubic equation does overflow ...
133 static void hScale8To19_c(SwsContext *c, int16_t *_dst, int dstW,
134 const uint8_t *src, const int16_t *filter,
135 const int32_t *filterPos, int filterSize)
138 int32_t *dst = (int32_t *) _dst;
139 for (i = 0; i < dstW; i++) {
141 int srcPos = filterPos[i];
143 for (j = 0; j < filterSize; j++) {
144 val += ((int)src[srcPos + j]) * filter[filterSize * i + j];
146 dst[i] = FFMIN(val >> 3, (1 << 19) - 1); // the cubic equation does overflow ...
150 // FIXME all pal and rgb srcFormats could do this conversion as well
151 // FIXME all scalers more complex than bilinear could do half of this transform
152 static void chrRangeToJpeg_c(int16_t *dstU, int16_t *dstV, int width)
155 for (i = 0; i < width; i++) {
156 dstU[i] = (FFMIN(dstU[i], 30775) * 4663 - 9289992) >> 12; // -264
157 dstV[i] = (FFMIN(dstV[i], 30775) * 4663 - 9289992) >> 12; // -264
161 static void chrRangeFromJpeg_c(int16_t *dstU, int16_t *dstV, int width)
164 for (i = 0; i < width; i++) {
165 dstU[i] = (dstU[i] * 1799 + 4081085) >> 11; // 1469
166 dstV[i] = (dstV[i] * 1799 + 4081085) >> 11; // 1469
170 static void lumRangeToJpeg_c(int16_t *dst, int width)
173 for (i = 0; i < width; i++)
174 dst[i] = (FFMIN(dst[i], 30189) * 19077 - 39057361) >> 14;
177 static void lumRangeFromJpeg_c(int16_t *dst, int width)
180 for (i = 0; i < width; i++)
181 dst[i] = (dst[i] * 14071 + 33561947) >> 14;
184 static void chrRangeToJpeg16_c(int16_t *_dstU, int16_t *_dstV, int width)
187 int32_t *dstU = (int32_t *) _dstU;
188 int32_t *dstV = (int32_t *) _dstV;
189 for (i = 0; i < width; i++) {
190 dstU[i] = (FFMIN(dstU[i], 30775 << 4) * 4663 - (9289992 << 4)) >> 12; // -264
191 dstV[i] = (FFMIN(dstV[i], 30775 << 4) * 4663 - (9289992 << 4)) >> 12; // -264
195 static void chrRangeFromJpeg16_c(int16_t *_dstU, int16_t *_dstV, int width)
198 int32_t *dstU = (int32_t *) _dstU;
199 int32_t *dstV = (int32_t *) _dstV;
200 for (i = 0; i < width; i++) {
201 dstU[i] = (dstU[i] * 1799 + (4081085 << 4)) >> 11; // 1469
202 dstV[i] = (dstV[i] * 1799 + (4081085 << 4)) >> 11; // 1469
206 static void lumRangeToJpeg16_c(int16_t *_dst, int width)
209 int32_t *dst = (int32_t *) _dst;
210 for (i = 0; i < width; i++)
211 dst[i] = (FFMIN(dst[i], 30189 << 4) * 4769 - (39057361 << 2)) >> 12;
214 static void lumRangeFromJpeg16_c(int16_t *_dst, int width)
217 int32_t *dst = (int32_t *) _dst;
218 for (i = 0; i < width; i++)
219 dst[i] = (dst[i]*(14071/4) + (33561947<<4)/4)>>12;
222 static void hyscale_fast_c(SwsContext *c, int16_t *dst, int dstWidth,
223 const uint8_t *src, int srcW, int xInc)
226 unsigned int xpos = 0;
227 for (i = 0; i < dstWidth; i++) {
228 register unsigned int xx = xpos >> 16;
229 register unsigned int xalpha = (xpos & 0xFFFF) >> 9;
230 dst[i] = (src[xx] << 7) + (src[xx + 1] - src[xx]) * xalpha;
233 for (i=dstWidth-1; (i*xInc)>>16 >=srcW-1; i--)
234 dst[i] = src[srcW-1]*128;
237 // *** horizontal scale Y line to temp buffer
238 static av_always_inline void hyscale(SwsContext *c, int16_t *dst, int dstWidth,
239 const uint8_t *src_in[4],
241 const int16_t *hLumFilter,
242 const int32_t *hLumFilterPos,
244 uint8_t *formatConvBuffer,
245 uint32_t *pal, int isAlpha)
247 void (*toYV12)(uint8_t *, const uint8_t *, const uint8_t *, const uint8_t *, int, uint32_t *) =
248 isAlpha ? c->alpToYV12 : c->lumToYV12;
249 void (*convertRange)(int16_t *, int) = isAlpha ? NULL : c->lumConvertRange;
250 const uint8_t *src = src_in[isAlpha ? 3 : 0];
253 toYV12(formatConvBuffer, src, src_in[1], src_in[2], srcW, pal);
254 src = formatConvBuffer;
255 } else if (c->readLumPlanar && !isAlpha) {
256 c->readLumPlanar(formatConvBuffer, src_in, srcW, c->input_rgb2yuv_table);
257 src = formatConvBuffer;
258 } else if (c->readAlpPlanar && isAlpha) {
259 c->readAlpPlanar(formatConvBuffer, src_in, srcW, NULL);
260 src = formatConvBuffer;
263 if (!c->hyscale_fast) {
264 c->hyScale(c, dst, dstWidth, src, hLumFilter,
265 hLumFilterPos, hLumFilterSize);
266 } else { // fast bilinear upscale / crap downscale
267 c->hyscale_fast(c, dst, dstWidth, src, srcW, xInc);
271 convertRange(dst, dstWidth);
274 static void hcscale_fast_c(SwsContext *c, int16_t *dst1, int16_t *dst2,
275 int dstWidth, const uint8_t *src1,
276 const uint8_t *src2, int srcW, int xInc)
279 unsigned int xpos = 0;
280 for (i = 0; i < dstWidth; i++) {
281 register unsigned int xx = xpos >> 16;
282 register unsigned int xalpha = (xpos & 0xFFFF) >> 9;
283 dst1[i] = (src1[xx] * (xalpha ^ 127) + src1[xx + 1] * xalpha);
284 dst2[i] = (src2[xx] * (xalpha ^ 127) + src2[xx + 1] * xalpha);
287 for (i=dstWidth-1; (i*xInc)>>16 >=srcW-1; i--) {
288 dst1[i] = src1[srcW-1]*128;
289 dst2[i] = src2[srcW-1]*128;
293 static av_always_inline void hcscale(SwsContext *c, int16_t *dst1,
294 int16_t *dst2, int dstWidth,
295 const uint8_t *src_in[4],
297 const int16_t *hChrFilter,
298 const int32_t *hChrFilterPos,
300 uint8_t *formatConvBuffer, uint32_t *pal)
302 const uint8_t *src1 = src_in[1], *src2 = src_in[2];
304 uint8_t *buf2 = formatConvBuffer +
305 FFALIGN(srcW*2+78, 16);
306 c->chrToYV12(formatConvBuffer, buf2, src_in[0], src1, src2, srcW, pal);
307 src1= formatConvBuffer;
309 } else if (c->readChrPlanar) {
310 uint8_t *buf2 = formatConvBuffer +
311 FFALIGN(srcW*2+78, 16);
312 c->readChrPlanar(formatConvBuffer, buf2, src_in, srcW, c->input_rgb2yuv_table);
313 src1 = formatConvBuffer;
317 if (!c->hcscale_fast) {
318 c->hcScale(c, dst1, dstWidth, src1, hChrFilter, hChrFilterPos, hChrFilterSize);
319 c->hcScale(c, dst2, dstWidth, src2, hChrFilter, hChrFilterPos, hChrFilterSize);
320 } else { // fast bilinear upscale / crap downscale
321 c->hcscale_fast(c, dst1, dst2, dstWidth, src1, src2, srcW, xInc);
324 if (c->chrConvertRange)
325 c->chrConvertRange(dst1, dst2, dstWidth);
328 #define DEBUG_SWSCALE_BUFFERS 0
329 #define DEBUG_BUFFERS(...) \
330 if (DEBUG_SWSCALE_BUFFERS) \
331 av_log(c, AV_LOG_DEBUG, __VA_ARGS__)
333 static int swScale(SwsContext *c, const uint8_t *src[],
334 int srcStride[], int srcSliceY,
335 int srcSliceH, uint8_t *dst[], int dstStride[])
337 /* load a few things into local vars to make the code more readable?
339 const int srcW = c->srcW;
340 const int dstW = c->dstW;
341 const int dstH = c->dstH;
342 const int chrDstW = c->chrDstW;
343 const int chrSrcW = c->chrSrcW;
344 const int lumXInc = c->lumXInc;
345 const int chrXInc = c->chrXInc;
346 const enum AVPixelFormat dstFormat = c->dstFormat;
347 const int flags = c->flags;
348 int32_t *vLumFilterPos = c->vLumFilterPos;
349 int32_t *vChrFilterPos = c->vChrFilterPos;
350 int32_t *hLumFilterPos = c->hLumFilterPos;
351 int32_t *hChrFilterPos = c->hChrFilterPos;
352 int16_t *hLumFilter = c->hLumFilter;
353 int16_t *hChrFilter = c->hChrFilter;
354 int32_t *lumMmxFilter = c->lumMmxFilter;
355 int32_t *chrMmxFilter = c->chrMmxFilter;
356 const int vLumFilterSize = c->vLumFilterSize;
357 const int vChrFilterSize = c->vChrFilterSize;
358 const int hLumFilterSize = c->hLumFilterSize;
359 const int hChrFilterSize = c->hChrFilterSize;
360 int16_t **lumPixBuf = c->lumPixBuf;
361 int16_t **chrUPixBuf = c->chrUPixBuf;
362 int16_t **chrVPixBuf = c->chrVPixBuf;
363 int16_t **alpPixBuf = c->alpPixBuf;
364 const int vLumBufSize = c->vLumBufSize;
365 const int vChrBufSize = c->vChrBufSize;
366 uint8_t *formatConvBuffer = c->formatConvBuffer;
367 uint32_t *pal = c->pal_yuv;
368 yuv2planar1_fn yuv2plane1 = c->yuv2plane1;
369 yuv2planarX_fn yuv2planeX = c->yuv2planeX;
370 yuv2interleavedX_fn yuv2nv12cX = c->yuv2nv12cX;
371 yuv2packed1_fn yuv2packed1 = c->yuv2packed1;
372 yuv2packed2_fn yuv2packed2 = c->yuv2packed2;
373 yuv2packedX_fn yuv2packedX = c->yuv2packedX;
374 yuv2anyX_fn yuv2anyX = c->yuv2anyX;
375 const int chrSrcSliceY = srcSliceY >> c->chrSrcVSubSample;
376 const int chrSrcSliceH = FF_CEIL_RSHIFT(srcSliceH, c->chrSrcVSubSample);
377 int should_dither = is9_OR_10BPS(c->srcFormat) ||
378 is16BPS(c->srcFormat);
381 /* vars which will change and which we need to store back in the context */
383 int lumBufIndex = c->lumBufIndex;
384 int chrBufIndex = c->chrBufIndex;
385 int lastInLumBuf = c->lastInLumBuf;
386 int lastInChrBuf = c->lastInChrBuf;
388 if (!usePal(c->srcFormat)) {
389 pal = c->input_rgb2yuv_table;
392 if (isPacked(c->srcFormat)) {
400 srcStride[3] = srcStride[0];
402 srcStride[1] <<= c->vChrDrop;
403 srcStride[2] <<= c->vChrDrop;
405 DEBUG_BUFFERS("swScale() %p[%d] %p[%d] %p[%d] %p[%d] -> %p[%d] %p[%d] %p[%d] %p[%d]\n",
406 src[0], srcStride[0], src[1], srcStride[1],
407 src[2], srcStride[2], src[3], srcStride[3],
408 dst[0], dstStride[0], dst[1], dstStride[1],
409 dst[2], dstStride[2], dst[3], dstStride[3]);
410 DEBUG_BUFFERS("srcSliceY: %d srcSliceH: %d dstY: %d dstH: %d\n",
411 srcSliceY, srcSliceH, dstY, dstH);
412 DEBUG_BUFFERS("vLumFilterSize: %d vLumBufSize: %d vChrFilterSize: %d vChrBufSize: %d\n",
413 vLumFilterSize, vLumBufSize, vChrFilterSize, vChrBufSize);
415 if (dstStride[0]%16 !=0 || dstStride[1]%16 !=0 ||
416 dstStride[2]%16 !=0 || dstStride[3]%16 != 0) {
417 static int warnedAlready = 0; // FIXME maybe move this into the context
418 if (flags & SWS_PRINT_INFO && !warnedAlready) {
419 av_log(c, AV_LOG_WARNING,
420 "Warning: dstStride is not aligned!\n"
421 " ->cannot do aligned memory accesses anymore\n");
426 if ( (uintptr_t)dst[0]%16 || (uintptr_t)dst[1]%16 || (uintptr_t)dst[2]%16
427 || (uintptr_t)src[0]%16 || (uintptr_t)src[1]%16 || (uintptr_t)src[2]%16
428 || dstStride[0]%16 || dstStride[1]%16 || dstStride[2]%16 || dstStride[3]%16
429 || srcStride[0]%16 || srcStride[1]%16 || srcStride[2]%16 || srcStride[3]%16
431 static int warnedAlready=0;
432 int cpu_flags = av_get_cpu_flags();
433 if (HAVE_MMXEXT && (cpu_flags & AV_CPU_FLAG_SSE2) && !warnedAlready){
434 av_log(c, AV_LOG_WARNING, "Warning: data is not aligned! This can lead to a speedloss\n");
439 /* Note the user might start scaling the picture in the middle so this
440 * will not get executed. This is not really intended but works
441 * currently, so people might do it. */
442 if (srcSliceY == 0) {
450 if (!should_dither) {
451 c->chrDither8 = c->lumDither8 = ff_sws_pb_64;
455 for (; dstY < dstH; dstY++) {
456 const int chrDstY = dstY >> c->chrDstVSubSample;
458 dst[0] + dstStride[0] * dstY,
459 dst[1] + dstStride[1] * chrDstY,
460 dst[2] + dstStride[2] * chrDstY,
461 (CONFIG_SWSCALE_ALPHA && alpPixBuf) ? dst[3] + dstStride[3] * dstY : NULL,
463 int use_mmx_vfilter= c->use_mmx_vfilter;
465 // First line needed as input
466 const int firstLumSrcY = FFMAX(1 - vLumFilterSize, vLumFilterPos[dstY]);
467 const int firstLumSrcY2 = FFMAX(1 - vLumFilterSize, vLumFilterPos[FFMIN(dstY | ((1 << c->chrDstVSubSample) - 1), dstH - 1)]);
468 // First line needed as input
469 const int firstChrSrcY = FFMAX(1 - vChrFilterSize, vChrFilterPos[chrDstY]);
471 // Last line needed as input
472 int lastLumSrcY = FFMIN(c->srcH, firstLumSrcY + vLumFilterSize) - 1;
473 int lastLumSrcY2 = FFMIN(c->srcH, firstLumSrcY2 + vLumFilterSize) - 1;
474 int lastChrSrcY = FFMIN(c->chrSrcH, firstChrSrcY + vChrFilterSize) - 1;
477 // handle holes (FAST_BILINEAR & weird filters)
478 if (firstLumSrcY > lastInLumBuf)
479 lastInLumBuf = firstLumSrcY - 1;
480 if (firstChrSrcY > lastInChrBuf)
481 lastInChrBuf = firstChrSrcY - 1;
482 av_assert0(firstLumSrcY >= lastInLumBuf - vLumBufSize + 1);
483 av_assert0(firstChrSrcY >= lastInChrBuf - vChrBufSize + 1);
485 DEBUG_BUFFERS("dstY: %d\n", dstY);
486 DEBUG_BUFFERS("\tfirstLumSrcY: %d lastLumSrcY: %d lastInLumBuf: %d\n",
487 firstLumSrcY, lastLumSrcY, lastInLumBuf);
488 DEBUG_BUFFERS("\tfirstChrSrcY: %d lastChrSrcY: %d lastInChrBuf: %d\n",
489 firstChrSrcY, lastChrSrcY, lastInChrBuf);
491 // Do we have enough lines in this slice to output the dstY line
492 enough_lines = lastLumSrcY2 < srcSliceY + srcSliceH &&
493 lastChrSrcY < FF_CEIL_RSHIFT(srcSliceY + srcSliceH, c->chrSrcVSubSample);
496 lastLumSrcY = srcSliceY + srcSliceH - 1;
497 lastChrSrcY = chrSrcSliceY + chrSrcSliceH - 1;
498 DEBUG_BUFFERS("buffering slice: lastLumSrcY %d lastChrSrcY %d\n",
499 lastLumSrcY, lastChrSrcY);
502 // Do horizontal scaling
503 while (lastInLumBuf < lastLumSrcY) {
504 const uint8_t *src1[4] = {
505 src[0] + (lastInLumBuf + 1 - srcSliceY) * srcStride[0],
506 src[1] + (lastInLumBuf + 1 - srcSliceY) * srcStride[1],
507 src[2] + (lastInLumBuf + 1 - srcSliceY) * srcStride[2],
508 src[3] + (lastInLumBuf + 1 - srcSliceY) * srcStride[3],
511 av_assert0(lumBufIndex < 2 * vLumBufSize);
512 av_assert0(lastInLumBuf + 1 - srcSliceY < srcSliceH);
513 av_assert0(lastInLumBuf + 1 - srcSliceY >= 0);
514 hyscale(c, lumPixBuf[lumBufIndex], dstW, src1, srcW, lumXInc,
515 hLumFilter, hLumFilterPos, hLumFilterSize,
516 formatConvBuffer, pal, 0);
517 if (CONFIG_SWSCALE_ALPHA && alpPixBuf)
518 hyscale(c, alpPixBuf[lumBufIndex], dstW, src1, srcW,
519 lumXInc, hLumFilter, hLumFilterPos, hLumFilterSize,
520 formatConvBuffer, pal, 1);
522 DEBUG_BUFFERS("\t\tlumBufIndex %d: lastInLumBuf: %d\n",
523 lumBufIndex, lastInLumBuf);
525 while (lastInChrBuf < lastChrSrcY) {
526 const uint8_t *src1[4] = {
527 src[0] + (lastInChrBuf + 1 - chrSrcSliceY) * srcStride[0],
528 src[1] + (lastInChrBuf + 1 - chrSrcSliceY) * srcStride[1],
529 src[2] + (lastInChrBuf + 1 - chrSrcSliceY) * srcStride[2],
530 src[3] + (lastInChrBuf + 1 - chrSrcSliceY) * srcStride[3],
533 av_assert0(chrBufIndex < 2 * vChrBufSize);
534 av_assert0(lastInChrBuf + 1 - chrSrcSliceY < (chrSrcSliceH));
535 av_assert0(lastInChrBuf + 1 - chrSrcSliceY >= 0);
536 // FIXME replace parameters through context struct (some at least)
538 if (c->needs_hcscale)
539 hcscale(c, chrUPixBuf[chrBufIndex], chrVPixBuf[chrBufIndex],
540 chrDstW, src1, chrSrcW, chrXInc,
541 hChrFilter, hChrFilterPos, hChrFilterSize,
542 formatConvBuffer, pal);
544 DEBUG_BUFFERS("\t\tchrBufIndex %d: lastInChrBuf: %d\n",
545 chrBufIndex, lastInChrBuf);
547 // wrap buf index around to stay inside the ring buffer
548 if (lumBufIndex >= vLumBufSize)
549 lumBufIndex -= vLumBufSize;
550 if (chrBufIndex >= vChrBufSize)
551 chrBufIndex -= vChrBufSize;
553 break; // we can't output a dstY line so let's try with the next slice
556 updateMMXDitherTables(c, dstY, lumBufIndex, chrBufIndex,
557 lastInLumBuf, lastInChrBuf);
560 c->chrDither8 = dither_8x8_128[chrDstY & 7];
561 c->lumDither8 = dither_8x8_128[dstY & 7];
563 if (dstY >= dstH - 2) {
564 /* hmm looks like we can't use MMX here without overwriting
565 * this array's tail */
566 ff_sws_init_output_funcs(c, &yuv2plane1, &yuv2planeX, &yuv2nv12cX,
567 &yuv2packed1, &yuv2packed2, &yuv2packedX, &yuv2anyX);
572 const int16_t **lumSrcPtr = (const int16_t **)(void*) lumPixBuf + lumBufIndex + firstLumSrcY - lastInLumBuf + vLumBufSize;
573 const int16_t **chrUSrcPtr = (const int16_t **)(void*) chrUPixBuf + chrBufIndex + firstChrSrcY - lastInChrBuf + vChrBufSize;
574 const int16_t **chrVSrcPtr = (const int16_t **)(void*) chrVPixBuf + chrBufIndex + firstChrSrcY - lastInChrBuf + vChrBufSize;
575 const int16_t **alpSrcPtr = (CONFIG_SWSCALE_ALPHA && alpPixBuf) ?
576 (const int16_t **)(void*) alpPixBuf + lumBufIndex + firstLumSrcY - lastInLumBuf + vLumBufSize : NULL;
577 int16_t *vLumFilter = c->vLumFilter;
578 int16_t *vChrFilter = c->vChrFilter;
580 if (isPlanarYUV(dstFormat) ||
581 (isGray(dstFormat) && !isALPHA(dstFormat))) { // YV12 like
582 const int chrSkipMask = (1 << c->chrDstVSubSample) - 1;
584 vLumFilter += dstY * vLumFilterSize;
585 vChrFilter += chrDstY * vChrFilterSize;
587 // av_assert0(use_mmx_vfilter != (
588 // yuv2planeX == yuv2planeX_10BE_c
589 // || yuv2planeX == yuv2planeX_10LE_c
590 // || yuv2planeX == yuv2planeX_9BE_c
591 // || yuv2planeX == yuv2planeX_9LE_c
592 // || yuv2planeX == yuv2planeX_16BE_c
593 // || yuv2planeX == yuv2planeX_16LE_c
594 // || yuv2planeX == yuv2planeX_8_c) || !ARCH_X86);
597 vLumFilter= (int16_t *)c->lumMmxFilter;
598 vChrFilter= (int16_t *)c->chrMmxFilter;
601 if (vLumFilterSize == 1) {
602 yuv2plane1(lumSrcPtr[0], dest[0], dstW, c->lumDither8, 0);
604 yuv2planeX(vLumFilter, vLumFilterSize,
606 dstW, c->lumDither8, 0);
609 if (!((dstY & chrSkipMask) || isGray(dstFormat))) {
611 yuv2nv12cX(c, vChrFilter,
612 vChrFilterSize, chrUSrcPtr, chrVSrcPtr,
614 } else if (vChrFilterSize == 1) {
615 yuv2plane1(chrUSrcPtr[0], dest[1], chrDstW, c->chrDither8, 0);
616 yuv2plane1(chrVSrcPtr[0], dest[2], chrDstW, c->chrDither8, 3);
618 yuv2planeX(vChrFilter,
619 vChrFilterSize, chrUSrcPtr, dest[1],
620 chrDstW, c->chrDither8, 0);
621 yuv2planeX(vChrFilter,
622 vChrFilterSize, chrVSrcPtr, dest[2],
623 chrDstW, c->chrDither8, use_mmx_vfilter ? (c->uv_offx2 >> 1) : 3);
627 if (CONFIG_SWSCALE_ALPHA && alpPixBuf) {
629 vLumFilter= (int16_t *)c->alpMmxFilter;
631 if (vLumFilterSize == 1) {
632 yuv2plane1(alpSrcPtr[0], dest[3], dstW,
635 yuv2planeX(vLumFilter,
636 vLumFilterSize, alpSrcPtr, dest[3],
637 dstW, c->lumDither8, 0);
640 } else if (yuv2packedX) {
641 av_assert1(lumSrcPtr + vLumFilterSize - 1 < (const int16_t **)lumPixBuf + vLumBufSize * 2);
642 av_assert1(chrUSrcPtr + vChrFilterSize - 1 < (const int16_t **)chrUPixBuf + vChrBufSize * 2);
643 if (c->yuv2packed1 && vLumFilterSize == 1 &&
644 vChrFilterSize <= 2) { // unscaled RGB
645 int chrAlpha = vChrFilterSize == 1 ? 0 : vChrFilter[2 * dstY + 1];
646 yuv2packed1(c, *lumSrcPtr, chrUSrcPtr, chrVSrcPtr,
647 alpPixBuf ? *alpSrcPtr : NULL,
648 dest[0], dstW, chrAlpha, dstY);
649 } else if (c->yuv2packed2 && vLumFilterSize == 2 &&
650 vChrFilterSize == 2) { // bilinear upscale RGB
651 int lumAlpha = vLumFilter[2 * dstY + 1];
652 int chrAlpha = vChrFilter[2 * dstY + 1];
654 lumMmxFilter[3] = vLumFilter[2 * dstY] * 0x10001;
656 chrMmxFilter[3] = vChrFilter[2 * chrDstY] * 0x10001;
657 yuv2packed2(c, lumSrcPtr, chrUSrcPtr, chrVSrcPtr,
658 alpPixBuf ? alpSrcPtr : NULL,
659 dest[0], dstW, lumAlpha, chrAlpha, dstY);
660 } else { // general RGB
661 yuv2packedX(c, vLumFilter + dstY * vLumFilterSize,
662 lumSrcPtr, vLumFilterSize,
663 vChrFilter + dstY * vChrFilterSize,
664 chrUSrcPtr, chrVSrcPtr, vChrFilterSize,
665 alpSrcPtr, dest[0], dstW, dstY);
668 av_assert1(!yuv2packed1 && !yuv2packed2);
669 yuv2anyX(c, vLumFilter + dstY * vLumFilterSize,
670 lumSrcPtr, vLumFilterSize,
671 vChrFilter + dstY * vChrFilterSize,
672 chrUSrcPtr, chrVSrcPtr, vChrFilterSize,
673 alpSrcPtr, dest, dstW, dstY);
677 if (isPlanar(dstFormat) && isALPHA(dstFormat) && !alpPixBuf) {
679 int height = dstY - lastDstY;
681 if (is16BPS(dstFormat) || isNBPS(dstFormat)) {
682 const AVPixFmtDescriptor *desc = av_pix_fmt_desc_get(dstFormat);
683 fillPlane16(dst[3], dstStride[3], length, height, lastDstY,
684 1, desc->comp[3].depth_minus1,
687 fillPlane(dst[3], dstStride[3], length, height, lastDstY, 255);
690 #if HAVE_MMXEXT_INLINE
691 if (av_get_cpu_flags() & AV_CPU_FLAG_MMXEXT)
692 __asm__ volatile ("sfence" ::: "memory");
696 /* store changed local vars back in the context */
698 c->lumBufIndex = lumBufIndex;
699 c->chrBufIndex = chrBufIndex;
700 c->lastInLumBuf = lastInLumBuf;
701 c->lastInChrBuf = lastInChrBuf;
703 return dstY - lastDstY;
706 static av_cold void sws_init_swScale_c(SwsContext *c)
708 enum AVPixelFormat srcFormat = c->srcFormat;
710 ff_sws_init_output_funcs(c, &c->yuv2plane1, &c->yuv2planeX,
711 &c->yuv2nv12cX, &c->yuv2packed1,
712 &c->yuv2packed2, &c->yuv2packedX, &c->yuv2anyX);
714 ff_sws_init_input_funcs(c);
717 if (c->srcBpc == 8) {
718 if (c->dstBpc <= 14) {
719 c->hyScale = c->hcScale = hScale8To15_c;
720 if (c->flags & SWS_FAST_BILINEAR) {
721 c->hyscale_fast = hyscale_fast_c;
722 c->hcscale_fast = hcscale_fast_c;
725 c->hyScale = c->hcScale = hScale8To19_c;
728 c->hyScale = c->hcScale = c->dstBpc > 14 ? hScale16To19_c
732 if (c->srcRange != c->dstRange && !isAnyRGB(c->dstFormat)) {
733 if (c->dstBpc <= 14) {
735 c->lumConvertRange = lumRangeFromJpeg_c;
736 c->chrConvertRange = chrRangeFromJpeg_c;
738 c->lumConvertRange = lumRangeToJpeg_c;
739 c->chrConvertRange = chrRangeToJpeg_c;
743 c->lumConvertRange = lumRangeFromJpeg16_c;
744 c->chrConvertRange = chrRangeFromJpeg16_c;
746 c->lumConvertRange = lumRangeToJpeg16_c;
747 c->chrConvertRange = chrRangeToJpeg16_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)
759 sws_init_swScale_c(c);
762 ff_sws_init_swScale_mmx(c);
764 ff_sws_init_swScale_altivec(c);
769 static void reset_ptr(const uint8_t *src[], int 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);
787 for (i = 0; i < 4; i++) {
788 int plane = desc->comp[i].plane;
789 if (!data[plane] || !linesizes[plane])
796 static void xyz12Torgb48(struct SwsContext *c, uint16_t *dst,
797 const uint16_t *src, int stride, int h)
800 const AVPixFmtDescriptor *desc = av_pix_fmt_desc_get(c->srcFormat);
802 for (yp=0; yp<h; yp++) {
803 for (xp=0; xp+2<stride; xp+=3) {
804 int x, y, z, r, g, b;
806 if (desc->flags & AV_PIX_FMT_FLAG_BE) {
807 x = AV_RB16(src + xp + 0);
808 y = AV_RB16(src + xp + 1);
809 z = AV_RB16(src + xp + 2);
811 x = AV_RL16(src + xp + 0);
812 y = AV_RL16(src + xp + 1);
813 z = AV_RL16(src + xp + 2);
816 x = c->xyzgamma[x>>4];
817 y = c->xyzgamma[y>>4];
818 z = c->xyzgamma[z>>4];
820 // convert from XYZlinear to sRGBlinear
821 r = c->xyz2rgb_matrix[0][0] * x +
822 c->xyz2rgb_matrix[0][1] * y +
823 c->xyz2rgb_matrix[0][2] * z >> 12;
824 g = c->xyz2rgb_matrix[1][0] * x +
825 c->xyz2rgb_matrix[1][1] * y +
826 c->xyz2rgb_matrix[1][2] * z >> 12;
827 b = c->xyz2rgb_matrix[2][0] * x +
828 c->xyz2rgb_matrix[2][1] * y +
829 c->xyz2rgb_matrix[2][2] * z >> 12;
831 // limit values to 12-bit depth
832 r = av_clip_c(r,0,4095);
833 g = av_clip_c(g,0,4095);
834 b = av_clip_c(b,0,4095);
836 // convert from sRGBlinear to RGB and scale from 12bit to 16bit
837 if (desc->flags & AV_PIX_FMT_FLAG_BE) {
838 AV_WB16(dst + xp + 0, c->rgbgamma[r] << 4);
839 AV_WB16(dst + xp + 1, c->rgbgamma[g] << 4);
840 AV_WB16(dst + xp + 2, c->rgbgamma[b] << 4);
842 AV_WL16(dst + xp + 0, c->rgbgamma[r] << 4);
843 AV_WL16(dst + xp + 1, c->rgbgamma[g] << 4);
844 AV_WL16(dst + xp + 2, c->rgbgamma[b] << 4);
852 static void rgb48Toxyz12(struct SwsContext *c, uint16_t *dst,
853 const uint16_t *src, int stride, int h)
856 const AVPixFmtDescriptor *desc = av_pix_fmt_desc_get(c->srcFormat);
858 for (yp=0; yp<h; yp++) {
859 for (xp=0; xp+2<stride; xp+=3) {
860 int x, y, z, r, g, b;
862 if (desc->flags & AV_PIX_FMT_FLAG_BE) {
863 r = AV_RB16(src + xp + 0);
864 g = AV_RB16(src + xp + 1);
865 b = AV_RB16(src + xp + 2);
867 r = AV_RL16(src + xp + 0);
868 g = AV_RL16(src + xp + 1);
869 b = AV_RL16(src + xp + 2);
872 r = c->rgbgammainv[r>>4];
873 g = c->rgbgammainv[g>>4];
874 b = c->rgbgammainv[b>>4];
876 // convert from sRGBlinear to XYZlinear
877 x = c->rgb2xyz_matrix[0][0] * r +
878 c->rgb2xyz_matrix[0][1] * g +
879 c->rgb2xyz_matrix[0][2] * b >> 12;
880 y = c->rgb2xyz_matrix[1][0] * r +
881 c->rgb2xyz_matrix[1][1] * g +
882 c->rgb2xyz_matrix[1][2] * b >> 12;
883 z = c->rgb2xyz_matrix[2][0] * r +
884 c->rgb2xyz_matrix[2][1] * g +
885 c->rgb2xyz_matrix[2][2] * b >> 12;
887 // limit values to 12-bit depth
888 x = av_clip_c(x,0,4095);
889 y = av_clip_c(y,0,4095);
890 z = av_clip_c(z,0,4095);
892 // convert from XYZlinear to X'Y'Z' and scale from 12bit to 16bit
893 if (desc->flags & AV_PIX_FMT_FLAG_BE) {
894 AV_WB16(dst + xp + 0, c->xyzgammainv[x] << 4);
895 AV_WB16(dst + xp + 1, c->xyzgammainv[y] << 4);
896 AV_WB16(dst + xp + 2, c->xyzgammainv[z] << 4);
898 AV_WL16(dst + xp + 0, c->xyzgammainv[x] << 4);
899 AV_WL16(dst + xp + 1, c->xyzgammainv[y] << 4);
900 AV_WL16(dst + xp + 2, c->xyzgammainv[z] << 4);
909 * swscale wrapper, so we don't need to export the SwsContext.
910 * Assumes planar YUV to be in YUV order instead of YVU.
912 int attribute_align_arg sws_scale(struct SwsContext *c,
913 const uint8_t * const srcSlice[],
914 const int srcStride[], int srcSliceY,
915 int srcSliceH, uint8_t *const dst[],
916 const int dstStride[])
919 const uint8_t *src2[4];
921 uint8_t *rgb0_tmp = NULL;
923 if (!srcSlice || !dstStride || !dst || !srcSlice) {
924 av_log(c, AV_LOG_ERROR, "One of the input parameters to sws_scale() is NULL, please check the calling code\n");
927 memcpy(src2, srcSlice, sizeof(src2));
928 memcpy(dst2, dst, sizeof(dst2));
930 // do not mess up sliceDir if we have a "trailing" 0-size slice
934 if (!check_image_pointers(srcSlice, c->srcFormat, srcStride)) {
935 av_log(c, AV_LOG_ERROR, "bad src image pointers\n");
938 if (!check_image_pointers((const uint8_t* const*)dst, c->dstFormat, dstStride)) {
939 av_log(c, AV_LOG_ERROR, "bad dst image pointers\n");
943 if (c->sliceDir == 0 && srcSliceY != 0 && srcSliceY + srcSliceH != c->srcH) {
944 av_log(c, AV_LOG_ERROR, "Slices start in the middle!\n");
947 if (c->sliceDir == 0) {
948 if (srcSliceY == 0) c->sliceDir = 1; else c->sliceDir = -1;
951 if (usePal(c->srcFormat)) {
952 for (i = 0; i < 256; i++) {
953 int p, r, g, b, y, u, v, a = 0xff;
954 if (c->srcFormat == AV_PIX_FMT_PAL8) {
955 p = ((const uint32_t *)(srcSlice[1]))[i];
956 a = (p >> 24) & 0xFF;
957 r = (p >> 16) & 0xFF;
960 } else if (c->srcFormat == AV_PIX_FMT_RGB8) {
962 g = ((i >> 2) & 7) * 36;
964 } else if (c->srcFormat == AV_PIX_FMT_BGR8) {
966 g = ((i >> 3) & 7) * 36;
968 } else if (c->srcFormat == AV_PIX_FMT_RGB4_BYTE) {
969 r = ( i >> 3 ) * 255;
970 g = ((i >> 1) & 3) * 85;
972 } else if (c->srcFormat == AV_PIX_FMT_GRAY8 || c->srcFormat == AV_PIX_FMT_GRAY8A) {
975 av_assert1(c->srcFormat == AV_PIX_FMT_BGR4_BYTE);
976 b = ( i >> 3 ) * 255;
977 g = ((i >> 1) & 3) * 85;
980 #define RGB2YUV_SHIFT 15
981 #define BY ( (int) (0.114 * 219 / 255 * (1 << RGB2YUV_SHIFT) + 0.5))
982 #define BV (-(int) (0.081 * 224 / 255 * (1 << RGB2YUV_SHIFT) + 0.5))
983 #define BU ( (int) (0.500 * 224 / 255 * (1 << RGB2YUV_SHIFT) + 0.5))
984 #define GY ( (int) (0.587 * 219 / 255 * (1 << RGB2YUV_SHIFT) + 0.5))
985 #define GV (-(int) (0.419 * 224 / 255 * (1 << RGB2YUV_SHIFT) + 0.5))
986 #define GU (-(int) (0.331 * 224 / 255 * (1 << RGB2YUV_SHIFT) + 0.5))
987 #define RY ( (int) (0.299 * 219 / 255 * (1 << RGB2YUV_SHIFT) + 0.5))
988 #define RV ( (int) (0.500 * 224 / 255 * (1 << RGB2YUV_SHIFT) + 0.5))
989 #define RU (-(int) (0.169 * 224 / 255 * (1 << RGB2YUV_SHIFT) + 0.5))
991 y = av_clip_uint8((RY * r + GY * g + BY * b + ( 33 << (RGB2YUV_SHIFT - 1))) >> RGB2YUV_SHIFT);
992 u = av_clip_uint8((RU * r + GU * g + BU * b + (257 << (RGB2YUV_SHIFT - 1))) >> RGB2YUV_SHIFT);
993 v = av_clip_uint8((RV * r + GV * g + BV * b + (257 << (RGB2YUV_SHIFT - 1))) >> RGB2YUV_SHIFT);
994 c->pal_yuv[i]= y + (u<<8) + (v<<16) + ((unsigned)a<<24);
996 switch (c->dstFormat) {
997 case AV_PIX_FMT_BGR32:
999 case AV_PIX_FMT_RGB24:
1001 c->pal_rgb[i]= r + (g<<8) + (b<<16) + ((unsigned)a<<24);
1003 case AV_PIX_FMT_BGR32_1:
1005 case AV_PIX_FMT_BGR24:
1007 c->pal_rgb[i]= a + (r<<8) + (g<<16) + ((unsigned)b<<24);
1009 case AV_PIX_FMT_RGB32_1:
1011 case AV_PIX_FMT_RGB24:
1013 c->pal_rgb[i]= a + (b<<8) + (g<<16) + ((unsigned)r<<24);
1015 case AV_PIX_FMT_RGB32:
1017 case AV_PIX_FMT_BGR24:
1020 c->pal_rgb[i]= b + (g<<8) + (r<<16) + ((unsigned)a<<24);
1025 if (c->src0Alpha && !c->dst0Alpha && isALPHA(c->dstFormat)) {
1028 rgb0_tmp = av_malloc(FFABS(srcStride[0]) * srcSliceH + 32);
1030 return AVERROR(ENOMEM);
1032 base = srcStride[0] < 0 ? rgb0_tmp - srcStride[0] * (srcSliceH-1) : rgb0_tmp;
1033 for (y=0; y<srcSliceH; y++){
1034 memcpy(base + srcStride[0]*y, src2[0] + srcStride[0]*y, 4*c->srcW);
1035 for (x=c->src0Alpha-1; x<4*c->srcW; x+=4) {
1036 base[ srcStride[0]*y + x] = 0xFF;
1042 if (c->srcXYZ && !(c->dstXYZ && c->srcW==c->dstW && c->srcH==c->dstH)) {
1044 rgb0_tmp = av_malloc(FFABS(srcStride[0]) * srcSliceH + 32);
1046 return AVERROR(ENOMEM);
1048 base = srcStride[0] < 0 ? rgb0_tmp - srcStride[0] * (srcSliceH-1) : rgb0_tmp;
1050 xyz12Torgb48(c, (uint16_t*)base, (const uint16_t*)src2[0], srcStride[0]/2, srcSliceH);
1054 if (!srcSliceY && (c->flags & SWS_BITEXACT) && c->dither == SWS_DITHER_ED && c->dither_error[0])
1055 for (i = 0; i < 4; i++)
1056 memset(c->dither_error[i], 0, sizeof(c->dither_error[0][0]) * (c->dstW+2));
1059 // copy strides, so they can safely be modified
1060 if (c->sliceDir == 1) {
1061 // slices go from top to bottom
1062 int srcStride2[4] = { srcStride[0], srcStride[1], srcStride[2],
1064 int dstStride2[4] = { dstStride[0], dstStride[1], dstStride[2],
1067 reset_ptr(src2, c->srcFormat);
1068 reset_ptr((void*)dst2, c->dstFormat);
1070 /* reset slice direction at end of frame */
1071 if (srcSliceY + srcSliceH == c->srcH)
1074 ret = c->swScale(c, src2, srcStride2, srcSliceY, srcSliceH, dst2,
1077 // slices go from bottom to top => we flip the image internally
1078 int srcStride2[4] = { -srcStride[0], -srcStride[1], -srcStride[2],
1080 int dstStride2[4] = { -dstStride[0], -dstStride[1], -dstStride[2],
1083 src2[0] += (srcSliceH - 1) * srcStride[0];
1084 if (!usePal(c->srcFormat))
1085 src2[1] += ((srcSliceH >> c->chrSrcVSubSample) - 1) * srcStride[1];
1086 src2[2] += ((srcSliceH >> c->chrSrcVSubSample) - 1) * srcStride[2];
1087 src2[3] += (srcSliceH - 1) * srcStride[3];
1088 dst2[0] += ( c->dstH - 1) * dstStride[0];
1089 dst2[1] += ((c->dstH >> c->chrDstVSubSample) - 1) * dstStride[1];
1090 dst2[2] += ((c->dstH >> c->chrDstVSubSample) - 1) * dstStride[2];
1091 dst2[3] += ( c->dstH - 1) * dstStride[3];
1093 reset_ptr(src2, c->srcFormat);
1094 reset_ptr((void*)dst2, c->dstFormat);
1096 /* reset slice direction at end of frame */
1100 ret = c->swScale(c, src2, srcStride2, c->srcH-srcSliceY-srcSliceH,
1101 srcSliceH, dst2, dstStride2);
1105 if (c->dstXYZ && !(c->srcXYZ && c->srcW==c->dstW && c->srcH==c->dstH)) {
1106 /* replace on the same data */
1107 rgb48Toxyz12(c, (uint16_t*)dst2[0], (const uint16_t*)dst2[0], dstStride[0]/2, ret);