2 * Copyright (C) 2001-2003 Michael Niedermayer <michaelni@gmx.at>
4 * This file is part of Libav.
6 * Libav 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 * Libav 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 Libav; if not, write to the Free Software
18 * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
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, ff_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, static const uint8_t, 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 fill_plane9or10(uint8_t *plane, int stride, int width,
65 int height, int y, uint8_t val,
66 const int dst_depth, const int big_endian)
69 uint16_t *dst = (uint16_t *) (plane + stride * y);
70 #define FILL8TO9_OR_10(wfunc) \
71 for (i = 0; i < height; i++) { \
72 for (j = 0; j < width; j++) { \
73 wfunc(&dst[j], (val << (dst_depth - 8)) | \
74 (val >> (16 - dst_depth))); \
79 FILL8TO9_OR_10(AV_WB16);
81 FILL8TO9_OR_10(AV_WL16);
86 static void hScale16To19_c(SwsContext *c, int16_t *_dst, int dstW,
87 const uint8_t *_src, const int16_t *filter,
88 const int32_t *filterPos, int filterSize)
90 const AVPixFmtDescriptor *desc = av_pix_fmt_desc_get(c->srcFormat);
92 int32_t *dst = (int32_t *) _dst;
93 const uint16_t *src = (const uint16_t *) _src;
94 int bits = desc->comp[0].depth_minus1;
97 for (i = 0; i < dstW; i++) {
99 int srcPos = filterPos[i];
102 for (j = 0; j < filterSize; j++) {
103 val += src[srcPos + j] * filter[filterSize * i + j];
105 // filter=14 bit, input=16 bit, output=30 bit, >> 11 makes 19 bit
106 dst[i] = FFMIN(val >> sh, (1 << 19) - 1);
110 static void hScale16To15_c(SwsContext *c, int16_t *dst, int dstW,
111 const uint8_t *_src, const int16_t *filter,
112 const int32_t *filterPos, int filterSize)
114 const AVPixFmtDescriptor *desc = av_pix_fmt_desc_get(c->srcFormat);
116 const uint16_t *src = (const uint16_t *) _src;
117 int sh = desc->comp[0].depth_minus1;
119 for (i = 0; i < dstW; i++) {
121 int srcPos = filterPos[i];
124 for (j = 0; j < filterSize; j++) {
125 val += src[srcPos + j] * filter[filterSize * i + j];
127 // filter=14 bit, input=16 bit, output=30 bit, >> 15 makes 15 bit
128 dst[i] = FFMIN(val >> sh, (1 << 15) - 1);
132 // bilinear / bicubic scaling
133 static void hScale8To15_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 for (i = 0; i < dstW; i++) {
140 int srcPos = filterPos[i];
142 for (j = 0; j < filterSize; j++) {
143 val += ((int)src[srcPos + j]) * filter[filterSize * i + j];
145 dst[i] = FFMIN(val >> 7, (1 << 15) - 1); // the cubic equation does overflow ...
149 static void hScale8To19_c(SwsContext *c, int16_t *_dst, int dstW,
150 const uint8_t *src, const int16_t *filter,
151 const int32_t *filterPos, int filterSize)
154 int32_t *dst = (int32_t *) _dst;
155 for (i = 0; i < dstW; i++) {
157 int srcPos = filterPos[i];
159 for (j = 0; j < filterSize; j++) {
160 val += ((int)src[srcPos + j]) * filter[filterSize * i + j];
162 dst[i] = FFMIN(val >> 3, (1 << 19) - 1); // the cubic equation does overflow ...
166 // FIXME all pal and rgb srcFormats could do this conversion as well
167 // FIXME all scalers more complex than bilinear could do half of this transform
168 static void chrRangeToJpeg_c(int16_t *dstU, int16_t *dstV, int width)
171 for (i = 0; i < width; i++) {
172 dstU[i] = (FFMIN(dstU[i], 30775) * 4663 - 9289992) >> 12; // -264
173 dstV[i] = (FFMIN(dstV[i], 30775) * 4663 - 9289992) >> 12; // -264
177 static void chrRangeFromJpeg_c(int16_t *dstU, int16_t *dstV, int width)
180 for (i = 0; i < width; i++) {
181 dstU[i] = (dstU[i] * 1799 + 4081085) >> 11; // 1469
182 dstV[i] = (dstV[i] * 1799 + 4081085) >> 11; // 1469
186 static void lumRangeToJpeg_c(int16_t *dst, int width)
189 for (i = 0; i < width; i++)
190 dst[i] = (FFMIN(dst[i], 30189) * 19077 - 39057361) >> 14;
193 static void lumRangeFromJpeg_c(int16_t *dst, int width)
196 for (i = 0; i < width; i++)
197 dst[i] = (dst[i] * 14071 + 33561947) >> 14;
200 static void chrRangeToJpeg16_c(int16_t *_dstU, int16_t *_dstV, int width)
203 int32_t *dstU = (int32_t *) _dstU;
204 int32_t *dstV = (int32_t *) _dstV;
205 for (i = 0; i < width; i++) {
206 dstU[i] = (FFMIN(dstU[i], 30775 << 4) * 4663 - (9289992 << 4)) >> 12; // -264
207 dstV[i] = (FFMIN(dstV[i], 30775 << 4) * 4663 - (9289992 << 4)) >> 12; // -264
211 static void chrRangeFromJpeg16_c(int16_t *_dstU, int16_t *_dstV, int width)
214 int32_t *dstU = (int32_t *) _dstU;
215 int32_t *dstV = (int32_t *) _dstV;
216 for (i = 0; i < width; i++) {
217 dstU[i] = (dstU[i] * 1799 + (4081085 << 4)) >> 11; // 1469
218 dstV[i] = (dstV[i] * 1799 + (4081085 << 4)) >> 11; // 1469
222 static void lumRangeToJpeg16_c(int16_t *_dst, int width)
225 int32_t *dst = (int32_t *) _dst;
226 for (i = 0; i < width; i++)
227 dst[i] = (FFMIN(dst[i], 30189 << 4) * 4769 - (39057361 << 2)) >> 12;
230 static void lumRangeFromJpeg16_c(int16_t *_dst, int width)
233 int32_t *dst = (int32_t *) _dst;
234 for (i = 0; i < width; i++)
235 dst[i] = (dst[i] * 14071 + (33561947 << 4)) >> 14;
238 static void hyscale_fast_c(SwsContext *c, int16_t *dst, int dstWidth,
239 const uint8_t *src, int srcW, int xInc)
242 unsigned int xpos = 0;
243 for (i = 0; i < dstWidth; i++) {
244 register unsigned int xx = xpos >> 16;
245 register unsigned int xalpha = (xpos & 0xFFFF) >> 9;
246 dst[i] = (src[xx] << 7) + (src[xx + 1] - src[xx]) * xalpha;
251 // *** horizontal scale Y line to temp buffer
252 static av_always_inline void hyscale(SwsContext *c, int16_t *dst, int dstWidth,
253 const uint8_t *src_in[4],
255 const int16_t *hLumFilter,
256 const int32_t *hLumFilterPos,
258 uint8_t *formatConvBuffer,
259 uint32_t *pal, int isAlpha)
261 void (*toYV12)(uint8_t *, const uint8_t *, int, uint32_t *) =
262 isAlpha ? c->alpToYV12 : c->lumToYV12;
263 void (*convertRange)(int16_t *, int) = isAlpha ? NULL : c->lumConvertRange;
264 const uint8_t *src = src_in[isAlpha ? 3 : 0];
267 toYV12(formatConvBuffer, src, srcW, pal);
268 src = formatConvBuffer;
269 } else if (c->readLumPlanar && !isAlpha) {
270 c->readLumPlanar(formatConvBuffer, src_in, srcW);
271 src = formatConvBuffer;
274 if (!c->hyscale_fast) {
275 c->hyScale(c, dst, dstWidth, src, hLumFilter,
276 hLumFilterPos, hLumFilterSize);
277 } else { // fast bilinear upscale / crap downscale
278 c->hyscale_fast(c, dst, dstWidth, src, srcW, xInc);
282 convertRange(dst, dstWidth);
285 static void hcscale_fast_c(SwsContext *c, int16_t *dst1, int16_t *dst2,
286 int dstWidth, const uint8_t *src1,
287 const uint8_t *src2, int srcW, int xInc)
290 unsigned int xpos = 0;
291 for (i = 0; i < dstWidth; i++) {
292 register unsigned int xx = xpos >> 16;
293 register unsigned int xalpha = (xpos & 0xFFFF) >> 9;
294 dst1[i] = (src1[xx] * (xalpha ^ 127) + src1[xx + 1] * xalpha);
295 dst2[i] = (src2[xx] * (xalpha ^ 127) + src2[xx + 1] * xalpha);
300 static av_always_inline void hcscale(SwsContext *c, int16_t *dst1,
301 int16_t *dst2, int dstWidth,
302 const uint8_t *src_in[4],
304 const int16_t *hChrFilter,
305 const int32_t *hChrFilterPos,
307 uint8_t *formatConvBuffer, uint32_t *pal)
309 const uint8_t *src1 = src_in[1], *src2 = src_in[2];
311 uint8_t *buf2 = formatConvBuffer +
312 FFALIGN(srcW * FFALIGN(c->srcBpc, 8) >> 3, 16);
313 c->chrToYV12(formatConvBuffer, buf2, src1, src2, srcW, pal);
314 src1 = formatConvBuffer;
316 } else if (c->readChrPlanar) {
317 uint8_t *buf2 = formatConvBuffer +
318 FFALIGN(srcW * FFALIGN(c->srcBpc, 8) >> 3, 16);
319 c->readChrPlanar(formatConvBuffer, buf2, src_in, srcW);
320 src1 = formatConvBuffer;
324 if (!c->hcscale_fast) {
325 c->hcScale(c, dst1, dstWidth, src1, hChrFilter, hChrFilterPos, hChrFilterSize);
326 c->hcScale(c, dst2, dstWidth, src2, hChrFilter, hChrFilterPos, hChrFilterSize);
327 } else { // fast bilinear upscale / crap downscale
328 c->hcscale_fast(c, dst1, dst2, dstWidth, src1, src2, srcW, xInc);
331 if (c->chrConvertRange)
332 c->chrConvertRange(dst1, dst2, dstWidth);
335 #define DEBUG_SWSCALE_BUFFERS 0
336 #define DEBUG_BUFFERS(...) \
337 if (DEBUG_SWSCALE_BUFFERS) \
338 av_log(c, AV_LOG_DEBUG, __VA_ARGS__)
340 static int swscale(SwsContext *c, const uint8_t *src[],
341 int srcStride[], int srcSliceY,
342 int srcSliceH, uint8_t *dst[], int dstStride[])
344 /* load a few things into local vars to make the code more readable?
346 const int srcW = c->srcW;
347 const int dstW = c->dstW;
348 const int dstH = c->dstH;
349 const int chrDstW = c->chrDstW;
350 const int chrSrcW = c->chrSrcW;
351 const int lumXInc = c->lumXInc;
352 const int chrXInc = c->chrXInc;
353 const enum AVPixelFormat dstFormat = c->dstFormat;
354 const int flags = c->flags;
355 int32_t *vLumFilterPos = c->vLumFilterPos;
356 int32_t *vChrFilterPos = c->vChrFilterPos;
357 int32_t *hLumFilterPos = c->hLumFilterPos;
358 int32_t *hChrFilterPos = c->hChrFilterPos;
359 int16_t *vLumFilter = c->vLumFilter;
360 int16_t *vChrFilter = c->vChrFilter;
361 int16_t *hLumFilter = c->hLumFilter;
362 int16_t *hChrFilter = c->hChrFilter;
363 int32_t *lumMmxFilter = c->lumMmxFilter;
364 int32_t *chrMmxFilter = c->chrMmxFilter;
365 const int vLumFilterSize = c->vLumFilterSize;
366 const int vChrFilterSize = c->vChrFilterSize;
367 const int hLumFilterSize = c->hLumFilterSize;
368 const int hChrFilterSize = c->hChrFilterSize;
369 int16_t **lumPixBuf = c->lumPixBuf;
370 int16_t **chrUPixBuf = c->chrUPixBuf;
371 int16_t **chrVPixBuf = c->chrVPixBuf;
372 int16_t **alpPixBuf = c->alpPixBuf;
373 const int vLumBufSize = c->vLumBufSize;
374 const int vChrBufSize = c->vChrBufSize;
375 uint8_t *formatConvBuffer = c->formatConvBuffer;
376 uint32_t *pal = c->pal_yuv;
377 yuv2planar1_fn yuv2plane1 = c->yuv2plane1;
378 yuv2planarX_fn yuv2planeX = c->yuv2planeX;
379 yuv2interleavedX_fn yuv2nv12cX = c->yuv2nv12cX;
380 yuv2packed1_fn yuv2packed1 = c->yuv2packed1;
381 yuv2packed2_fn yuv2packed2 = c->yuv2packed2;
382 yuv2packedX_fn yuv2packedX = c->yuv2packedX;
383 yuv2anyX_fn yuv2anyX = c->yuv2anyX;
384 const int chrSrcSliceY = srcSliceY >> c->chrSrcVSubSample;
385 const int chrSrcSliceH = -((-srcSliceH) >> c->chrSrcVSubSample);
386 int should_dither = is9_OR_10BPS(c->srcFormat) ||
387 is16BPS(c->srcFormat);
390 /* vars which will change and which we need to store back in the context */
392 int lumBufIndex = c->lumBufIndex;
393 int chrBufIndex = c->chrBufIndex;
394 int lastInLumBuf = c->lastInLumBuf;
395 int lastInChrBuf = c->lastInChrBuf;
397 if (isPacked(c->srcFormat)) {
405 srcStride[3] = srcStride[0];
407 srcStride[1] <<= c->vChrDrop;
408 srcStride[2] <<= c->vChrDrop;
410 DEBUG_BUFFERS("swscale() %p[%d] %p[%d] %p[%d] %p[%d] -> %p[%d] %p[%d] %p[%d] %p[%d]\n",
411 src[0], srcStride[0], src[1], srcStride[1],
412 src[2], srcStride[2], src[3], srcStride[3],
413 dst[0], dstStride[0], dst[1], dstStride[1],
414 dst[2], dstStride[2], dst[3], dstStride[3]);
415 DEBUG_BUFFERS("srcSliceY: %d srcSliceH: %d dstY: %d dstH: %d\n",
416 srcSliceY, srcSliceH, dstY, dstH);
417 DEBUG_BUFFERS("vLumFilterSize: %d vLumBufSize: %d vChrFilterSize: %d vChrBufSize: %d\n",
418 vLumFilterSize, vLumBufSize, vChrFilterSize, vChrBufSize);
420 if (dstStride[0] % 8 != 0 || dstStride[1] % 8 != 0 ||
421 dstStride[2] % 8 != 0 || dstStride[3] % 8 != 0) {
422 static int warnedAlready = 0; // FIXME maybe move this into the context
423 if (flags & SWS_PRINT_INFO && !warnedAlready) {
424 av_log(c, AV_LOG_WARNING,
425 "Warning: dstStride is not aligned!\n"
426 " ->cannot do aligned memory accesses anymore\n");
431 /* Note the user might start scaling the picture in the middle so this
432 * will not get executed. This is not really intended but works
433 * currently, so people might do it. */
434 if (srcSliceY == 0) {
442 if (!should_dither) {
443 c->chrDither8 = c->lumDither8 = sws_pb_64;
447 for (; dstY < dstH; dstY++) {
448 const int chrDstY = dstY >> c->chrDstVSubSample;
450 dst[0] + dstStride[0] * dstY,
451 dst[1] + dstStride[1] * chrDstY,
452 dst[2] + dstStride[2] * chrDstY,
453 (CONFIG_SWSCALE_ALPHA && alpPixBuf) ? dst[3] + dstStride[3] * dstY : NULL,
456 // First line needed as input
457 const int firstLumSrcY = FFMAX(1 - vLumFilterSize, vLumFilterPos[dstY]);
458 const int firstLumSrcY2 = FFMAX(1 - vLumFilterSize, vLumFilterPos[FFMIN(dstY | ((1 << c->chrDstVSubSample) - 1), dstH - 1)]);
459 // First line needed as input
460 const int firstChrSrcY = FFMAX(1 - vChrFilterSize, vChrFilterPos[chrDstY]);
462 // Last line needed as input
463 int lastLumSrcY = FFMIN(c->srcH, firstLumSrcY + vLumFilterSize) - 1;
464 int lastLumSrcY2 = FFMIN(c->srcH, firstLumSrcY2 + vLumFilterSize) - 1;
465 int lastChrSrcY = FFMIN(c->chrSrcH, firstChrSrcY + vChrFilterSize) - 1;
468 // handle holes (FAST_BILINEAR & weird filters)
469 if (firstLumSrcY > lastInLumBuf)
470 lastInLumBuf = firstLumSrcY - 1;
471 if (firstChrSrcY > lastInChrBuf)
472 lastInChrBuf = firstChrSrcY - 1;
473 assert(firstLumSrcY >= lastInLumBuf - vLumBufSize + 1);
474 assert(firstChrSrcY >= lastInChrBuf - vChrBufSize + 1);
476 DEBUG_BUFFERS("dstY: %d\n", dstY);
477 DEBUG_BUFFERS("\tfirstLumSrcY: %d lastLumSrcY: %d lastInLumBuf: %d\n",
478 firstLumSrcY, lastLumSrcY, lastInLumBuf);
479 DEBUG_BUFFERS("\tfirstChrSrcY: %d lastChrSrcY: %d lastInChrBuf: %d\n",
480 firstChrSrcY, lastChrSrcY, lastInChrBuf);
482 // Do we have enough lines in this slice to output the dstY line
483 enough_lines = lastLumSrcY2 < srcSliceY + srcSliceH &&
484 lastChrSrcY < -((-srcSliceY - srcSliceH) >> c->chrSrcVSubSample);
487 lastLumSrcY = srcSliceY + srcSliceH - 1;
488 lastChrSrcY = chrSrcSliceY + chrSrcSliceH - 1;
489 DEBUG_BUFFERS("buffering slice: lastLumSrcY %d lastChrSrcY %d\n",
490 lastLumSrcY, lastChrSrcY);
493 // Do horizontal scaling
494 while (lastInLumBuf < lastLumSrcY) {
495 const uint8_t *src1[4] = {
496 src[0] + (lastInLumBuf + 1 - srcSliceY) * srcStride[0],
497 src[1] + (lastInLumBuf + 1 - srcSliceY) * srcStride[1],
498 src[2] + (lastInLumBuf + 1 - srcSliceY) * srcStride[2],
499 src[3] + (lastInLumBuf + 1 - srcSliceY) * srcStride[3],
502 assert(lumBufIndex < 2 * vLumBufSize);
503 assert(lastInLumBuf + 1 - srcSliceY < srcSliceH);
504 assert(lastInLumBuf + 1 - srcSliceY >= 0);
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 assert(chrBufIndex < 2 * vChrBufSize);
525 assert(lastInChrBuf + 1 - chrSrcSliceY < (chrSrcSliceH));
526 assert(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 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);
562 const int16_t **lumSrcPtr = (const int16_t **)lumPixBuf + lumBufIndex + firstLumSrcY - lastInLumBuf + vLumBufSize;
563 const int16_t **chrUSrcPtr = (const int16_t **)chrUPixBuf + chrBufIndex + firstChrSrcY - lastInChrBuf + vChrBufSize;
564 const int16_t **chrVSrcPtr = (const int16_t **)chrVPixBuf + chrBufIndex + firstChrSrcY - lastInChrBuf + vChrBufSize;
565 const int16_t **alpSrcPtr = (CONFIG_SWSCALE_ALPHA && alpPixBuf) ?
566 (const int16_t **)alpPixBuf + lumBufIndex + firstLumSrcY - lastInLumBuf + vLumBufSize : NULL;
568 if (firstLumSrcY < 0 || firstLumSrcY + vLumFilterSize > c->srcH) {
569 const int16_t **tmpY = (const int16_t **)lumPixBuf +
571 int neg = -firstLumSrcY, i;
572 int end = FFMIN(c->srcH - firstLumSrcY, vLumFilterSize);
573 for (i = 0; i < neg; i++)
574 tmpY[i] = lumSrcPtr[neg];
576 tmpY[i] = lumSrcPtr[i];
577 for (; i < vLumFilterSize; i++)
578 tmpY[i] = tmpY[i - 1];
582 const int16_t **tmpA = (const int16_t **)alpPixBuf +
584 for (i = 0; i < neg; i++)
585 tmpA[i] = alpSrcPtr[neg];
587 tmpA[i] = alpSrcPtr[i];
588 for (; i < vLumFilterSize; i++)
589 tmpA[i] = tmpA[i - 1];
593 if (firstChrSrcY < 0 ||
594 firstChrSrcY + vChrFilterSize > c->chrSrcH) {
595 const int16_t **tmpU = (const int16_t **)chrUPixBuf + 2 * vChrBufSize,
596 **tmpV = (const int16_t **)chrVPixBuf + 2 * vChrBufSize;
597 int neg = -firstChrSrcY, i;
598 int end = FFMIN(c->chrSrcH - firstChrSrcY, vChrFilterSize);
599 for (i = 0; i < neg; i++) {
600 tmpU[i] = chrUSrcPtr[neg];
601 tmpV[i] = chrVSrcPtr[neg];
603 for (; i < end; i++) {
604 tmpU[i] = chrUSrcPtr[i];
605 tmpV[i] = chrVSrcPtr[i];
607 for (; i < vChrFilterSize; i++) {
608 tmpU[i] = tmpU[i - 1];
609 tmpV[i] = tmpV[i - 1];
615 if (isPlanarYUV(dstFormat) ||
616 (isGray(dstFormat) && !isALPHA(dstFormat))) { // YV12 like
617 const int chrSkipMask = (1 << c->chrDstVSubSample) - 1;
619 if (vLumFilterSize == 1) {
620 yuv2plane1(lumSrcPtr[0], dest[0], dstW, c->lumDither8, 0);
622 yuv2planeX(vLumFilter + dstY * vLumFilterSize,
623 vLumFilterSize, lumSrcPtr, dest[0],
624 dstW, c->lumDither8, 0);
627 if (!((dstY & chrSkipMask) || isGray(dstFormat))) {
629 yuv2nv12cX(c, vChrFilter + chrDstY * vChrFilterSize,
630 vChrFilterSize, chrUSrcPtr, chrVSrcPtr,
632 } else if (vChrFilterSize == 1) {
633 yuv2plane1(chrUSrcPtr[0], dest[1], chrDstW, c->chrDither8, 0);
634 yuv2plane1(chrVSrcPtr[0], dest[2], chrDstW, c->chrDither8, 3);
636 yuv2planeX(vChrFilter + chrDstY * vChrFilterSize,
637 vChrFilterSize, chrUSrcPtr, dest[1],
638 chrDstW, c->chrDither8, 0);
639 yuv2planeX(vChrFilter + chrDstY * vChrFilterSize,
640 vChrFilterSize, chrVSrcPtr, dest[2],
641 chrDstW, c->chrDither8, 3);
645 if (CONFIG_SWSCALE_ALPHA && alpPixBuf) {
646 if (vLumFilterSize == 1) {
647 yuv2plane1(alpSrcPtr[0], dest[3], dstW,
650 yuv2planeX(vLumFilter + dstY * vLumFilterSize,
651 vLumFilterSize, alpSrcPtr, dest[3],
652 dstW, c->lumDither8, 0);
655 } else if (yuv2packedX) {
656 if (c->yuv2packed1 && vLumFilterSize == 1 &&
657 vChrFilterSize <= 2) { // unscaled RGB
658 int chrAlpha = vChrFilterSize == 1 ? 0 : vChrFilter[2 * dstY + 1];
659 yuv2packed1(c, *lumSrcPtr, chrUSrcPtr, chrVSrcPtr,
660 alpPixBuf ? *alpSrcPtr : NULL,
661 dest[0], dstW, chrAlpha, dstY);
662 } else if (c->yuv2packed2 && vLumFilterSize == 2 &&
663 vChrFilterSize == 2) { // bilinear upscale RGB
664 int lumAlpha = vLumFilter[2 * dstY + 1];
665 int chrAlpha = vChrFilter[2 * dstY + 1];
667 lumMmxFilter[3] = vLumFilter[2 * dstY] * 0x10001;
669 chrMmxFilter[3] = vChrFilter[2 * chrDstY] * 0x10001;
670 yuv2packed2(c, lumSrcPtr, chrUSrcPtr, chrVSrcPtr,
671 alpPixBuf ? alpSrcPtr : NULL,
672 dest[0], dstW, lumAlpha, chrAlpha, dstY);
673 } else { // general RGB
674 yuv2packedX(c, vLumFilter + dstY * vLumFilterSize,
675 lumSrcPtr, vLumFilterSize,
676 vChrFilter + dstY * vChrFilterSize,
677 chrUSrcPtr, chrVSrcPtr, vChrFilterSize,
678 alpSrcPtr, dest[0], dstW, dstY);
681 yuv2anyX(c, vLumFilter + dstY * vLumFilterSize,
682 lumSrcPtr, vLumFilterSize,
683 vChrFilter + dstY * vChrFilterSize,
684 chrUSrcPtr, chrVSrcPtr, vChrFilterSize,
685 alpSrcPtr, dest, dstW, dstY);
690 if (isPlanar(dstFormat) && isALPHA(dstFormat) && !alpPixBuf) {
692 int height = dstY - lastDstY;
693 if (is16BPS(c->dstFormat))
696 if (is9_OR_10BPS(dstFormat)) {
697 const AVPixFmtDescriptor *desc = av_pix_fmt_desc_get(dstFormat);
698 fill_plane9or10(dst[3], dstStride[3], length, height, lastDstY,
699 255, desc->comp[3].depth_minus1 + 1,
702 fillPlane(dst[3], dstStride[3], length, height, lastDstY, 255);
705 #if HAVE_MMXEXT_INLINE
706 if (av_get_cpu_flags() & AV_CPU_FLAG_MMXEXT)
707 __asm__ volatile ("sfence" ::: "memory");
711 /* store changed local vars back in the context */
713 c->lumBufIndex = lumBufIndex;
714 c->chrBufIndex = chrBufIndex;
715 c->lastInLumBuf = lastInLumBuf;
716 c->lastInChrBuf = lastInChrBuf;
718 return dstY - lastDstY;
721 static av_cold void sws_init_swscale(SwsContext *c)
723 enum AVPixelFormat srcFormat = c->srcFormat;
725 ff_sws_init_output_funcs(c, &c->yuv2plane1, &c->yuv2planeX,
726 &c->yuv2nv12cX, &c->yuv2packed1,
727 &c->yuv2packed2, &c->yuv2packedX, &c->yuv2anyX);
729 ff_sws_init_input_funcs(c);
731 if (c->srcBpc == 8) {
732 if (c->dstBpc <= 10) {
733 c->hyScale = c->hcScale = hScale8To15_c;
734 if (c->flags & SWS_FAST_BILINEAR) {
735 c->hyscale_fast = hyscale_fast_c;
736 c->hcscale_fast = hcscale_fast_c;
739 c->hyScale = c->hcScale = hScale8To19_c;
742 c->hyScale = c->hcScale = c->dstBpc > 10 ? hScale16To19_c
746 if (c->srcRange != c->dstRange && !isAnyRGB(c->dstFormat)) {
747 if (c->dstBpc <= 10) {
749 c->lumConvertRange = lumRangeFromJpeg_c;
750 c->chrConvertRange = chrRangeFromJpeg_c;
752 c->lumConvertRange = lumRangeToJpeg_c;
753 c->chrConvertRange = chrRangeToJpeg_c;
757 c->lumConvertRange = lumRangeFromJpeg16_c;
758 c->chrConvertRange = chrRangeFromJpeg16_c;
760 c->lumConvertRange = lumRangeToJpeg16_c;
761 c->chrConvertRange = chrRangeToJpeg16_c;
766 if (!(isGray(srcFormat) || isGray(c->dstFormat) ||
767 srcFormat == AV_PIX_FMT_MONOBLACK || srcFormat == AV_PIX_FMT_MONOWHITE))
768 c->needs_hcscale = 1;
771 SwsFunc ff_getSwsFunc(SwsContext *c)
776 ff_sws_init_swscale_ppc(c);
778 ff_sws_init_swscale_x86(c);