1 /*
2 * Copyright (C) 2001-2011 Michael Niedermayer <michaelni@gmx.at>
3 *
4 * This file is part of FFmpeg.
5 *
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.
10 *
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.
15 *
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
19 */
20
21 #include <inttypes.h>
22 #include <math.h>
23 #include <stdio.h>
24 #include <string.h>
25
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"
34 #include "config.h"
35 #include "rgb2rgb.h"
36 #include "swscale_internal.h"
37 #include "swscale.h"
38
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, },
49 };
50
51 DECLARE_ALIGNED(8, static const uint8_t, sws_pb_64)[8] = {
52 64, 64, 64, 64, 64, 64, 64, 64
53 };
54
fillPlane(uint8_t * plane,int stride,int width,int height,int y,uint8_t val)55 static av_always_inline void fillPlane(uint8_t *plane, int stride, int width,
56 int height, int y, uint8_t val)
57 {
58 int i;
59 uint8_t *ptr = plane + stride * y;
60 for (i = 0; i < height; i++) {
61 memset(ptr, val, width);
62 ptr += stride;
63 }
64 }
65
hScale16To19_c(SwsContext * c,int16_t * _dst,int dstW,const uint8_t * _src,const int16_t * filter,const int32_t * filterPos,int filterSize)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)
69 {
70 const AVPixFmtDescriptor *desc = av_pix_fmt_desc_get(c->srcFormat);
71 int i;
72 int32_t *dst = (int32_t *) _dst;
73 const uint16_t *src = (const uint16_t *) _src;
74 int bits = desc->comp[0].depth - 1;
75 int sh = bits - 4;
76
77 if ((isAnyRGB(c->srcFormat) || c->srcFormat==AV_PIX_FMT_PAL8) && desc->comp[0].depth<16) {
78 sh = 9;
79 } else if (desc->flags & AV_PIX_FMT_FLAG_FLOAT) { /* float input are process like uint 16bpc */
80 sh = 16 - 1 - 4;
81 }
82
83 for (i = 0; i < dstW; i++) {
84 int j;
85 int srcPos = filterPos[i];
86 int val = 0;
87
88 for (j = 0; j < filterSize; j++) {
89 val += src[srcPos + j] * filter[filterSize * i + j];
90 }
91 // filter=14 bit, input=16 bit, output=30 bit, >> 11 makes 19 bit
92 dst[i] = FFMIN(val >> sh, (1 << 19) - 1);
93 }
94 }
95
hScale16To15_c(SwsContext * c,int16_t * dst,int dstW,const uint8_t * _src,const int16_t * filter,const int32_t * filterPos,int filterSize)96 static void hScale16To15_c(SwsContext *c, int16_t *dst, int dstW,
97 const uint8_t *_src, const int16_t *filter,
98 const int32_t *filterPos, int filterSize)
99 {
100 const AVPixFmtDescriptor *desc = av_pix_fmt_desc_get(c->srcFormat);
101 int i;
102 const uint16_t *src = (const uint16_t *) _src;
103 int sh = desc->comp[0].depth - 1;
104
105 if (sh<15) {
106 sh = isAnyRGB(c->srcFormat) || c->srcFormat==AV_PIX_FMT_PAL8 ? 13 : (desc->comp[0].depth - 1);
107 } else if (desc->flags & AV_PIX_FMT_FLAG_FLOAT) { /* float input are process like uint 16bpc */
108 sh = 16 - 1;
109 }
110
111 for (i = 0; i < dstW; i++) {
112 int j;
113 int srcPos = filterPos[i];
114 int val = 0;
115
116 for (j = 0; j < filterSize; j++) {
117 val += src[srcPos + j] * filter[filterSize * i + j];
118 }
119 // filter=14 bit, input=16 bit, output=30 bit, >> 15 makes 15 bit
120 dst[i] = FFMIN(val >> sh, (1 << 15) - 1);
121 }
122 }
123
124 // bilinear / bicubic scaling
hScale8To15_c(SwsContext * c,int16_t * dst,int dstW,const uint8_t * src,const int16_t * filter,const int32_t * filterPos,int filterSize)125 static void hScale8To15_c(SwsContext *c, int16_t *dst, int dstW,
126 const uint8_t *src, const int16_t *filter,
127 const int32_t *filterPos, int filterSize)
128 {
129 int i;
130 for (i = 0; i < dstW; i++) {
131 int j;
132 int srcPos = filterPos[i];
133 int val = 0;
134 for (j = 0; j < filterSize; j++) {
135 val += ((int)src[srcPos + j]) * filter[filterSize * i + j];
136 }
137 dst[i] = FFMIN(val >> 7, (1 << 15) - 1); // the cubic equation does overflow ...
138 }
139 }
140
hScale8To19_c(SwsContext * c,int16_t * _dst,int dstW,const uint8_t * src,const int16_t * filter,const int32_t * filterPos,int filterSize)141 static void hScale8To19_c(SwsContext *c, int16_t *_dst, int dstW,
142 const uint8_t *src, const int16_t *filter,
143 const int32_t *filterPos, int filterSize)
144 {
145 int i;
146 int32_t *dst = (int32_t *) _dst;
147 for (i = 0; i < dstW; i++) {
148 int j;
149 int srcPos = filterPos[i];
150 int val = 0;
151 for (j = 0; j < filterSize; j++) {
152 val += ((int)src[srcPos + j]) * filter[filterSize * i + j];
153 }
154 dst[i] = FFMIN(val >> 3, (1 << 19) - 1); // the cubic equation does overflow ...
155 }
156 }
157
158 // FIXME all pal and rgb srcFormats could do this conversion as well
159 // FIXME all scalers more complex than bilinear could do half of this transform
chrRangeToJpeg_c(int16_t * dstU,int16_t * dstV,int width)160 static void chrRangeToJpeg_c(int16_t *dstU, int16_t *dstV, int width)
161 {
162 int i;
163 for (i = 0; i < width; i++) {
164 dstU[i] = (FFMIN(dstU[i], 30775) * 4663 - 9289992) >> 12; // -264
165 dstV[i] = (FFMIN(dstV[i], 30775) * 4663 - 9289992) >> 12; // -264
166 }
167 }
168
chrRangeFromJpeg_c(int16_t * dstU,int16_t * dstV,int width)169 static void chrRangeFromJpeg_c(int16_t *dstU, int16_t *dstV, int width)
170 {
171 int i;
172 for (i = 0; i < width; i++) {
173 dstU[i] = (dstU[i] * 1799 + 4081085) >> 11; // 1469
174 dstV[i] = (dstV[i] * 1799 + 4081085) >> 11; // 1469
175 }
176 }
177
lumRangeToJpeg_c(int16_t * dst,int width)178 static void lumRangeToJpeg_c(int16_t *dst, int width)
179 {
180 int i;
181 for (i = 0; i < width; i++)
182 dst[i] = (FFMIN(dst[i], 30189) * 19077 - 39057361) >> 14;
183 }
184
lumRangeFromJpeg_c(int16_t * dst,int width)185 static void lumRangeFromJpeg_c(int16_t *dst, int width)
186 {
187 int i;
188 for (i = 0; i < width; i++)
189 dst[i] = (dst[i] * 14071 + 33561947) >> 14;
190 }
191
chrRangeToJpeg16_c(int16_t * _dstU,int16_t * _dstV,int width)192 static void chrRangeToJpeg16_c(int16_t *_dstU, int16_t *_dstV, int width)
193 {
194 int i;
195 int32_t *dstU = (int32_t *) _dstU;
196 int32_t *dstV = (int32_t *) _dstV;
197 for (i = 0; i < width; i++) {
198 dstU[i] = (FFMIN(dstU[i], 30775 << 4) * 4663 - (9289992 << 4)) >> 12; // -264
199 dstV[i] = (FFMIN(dstV[i], 30775 << 4) * 4663 - (9289992 << 4)) >> 12; // -264
200 }
201 }
202
chrRangeFromJpeg16_c(int16_t * _dstU,int16_t * _dstV,int width)203 static void chrRangeFromJpeg16_c(int16_t *_dstU, int16_t *_dstV, int width)
204 {
205 int i;
206 int32_t *dstU = (int32_t *) _dstU;
207 int32_t *dstV = (int32_t *) _dstV;
208 for (i = 0; i < width; i++) {
209 dstU[i] = (dstU[i] * 1799 + (4081085 << 4)) >> 11; // 1469
210 dstV[i] = (dstV[i] * 1799 + (4081085 << 4)) >> 11; // 1469
211 }
212 }
213
lumRangeToJpeg16_c(int16_t * _dst,int width)214 static void lumRangeToJpeg16_c(int16_t *_dst, int width)
215 {
216 int i;
217 int32_t *dst = (int32_t *) _dst;
218 for (i = 0; i < width; i++) {
219 dst[i] = ((int)(FFMIN(dst[i], 30189 << 4) * 4769U - (39057361 << 2))) >> 12;
220 }
221 }
222
lumRangeFromJpeg16_c(int16_t * _dst,int width)223 static void lumRangeFromJpeg16_c(int16_t *_dst, int width)
224 {
225 int i;
226 int32_t *dst = (int32_t *) _dst;
227 for (i = 0; i < width; i++)
228 dst[i] = (dst[i]*(14071/4) + (33561947<<4)/4)>>12;
229 }
230
231
232 #define DEBUG_SWSCALE_BUFFERS 0
233 #define DEBUG_BUFFERS(...) \
234 if (DEBUG_SWSCALE_BUFFERS) \
235 av_log(c, AV_LOG_DEBUG, __VA_ARGS__)
236
swscale(SwsContext * c,const uint8_t * src[],int srcStride[],int srcSliceY,int srcSliceH,uint8_t * dst[],int dstStride[])237 static int swscale(SwsContext *c, const uint8_t *src[],
238 int srcStride[], int srcSliceY,
239 int srcSliceH, uint8_t *dst[], int dstStride[])
240 {
241 /* load a few things into local vars to make the code more readable?
242 * and faster */
243 const int dstW = c->dstW;
244 const int dstH = c->dstH;
245
246 const enum AVPixelFormat dstFormat = c->dstFormat;
247 const int flags = c->flags;
248 int32_t *vLumFilterPos = c->vLumFilterPos;
249 int32_t *vChrFilterPos = c->vChrFilterPos;
250
251 const int vLumFilterSize = c->vLumFilterSize;
252 const int vChrFilterSize = c->vChrFilterSize;
253
254 yuv2planar1_fn yuv2plane1 = c->yuv2plane1;
255 yuv2planarX_fn yuv2planeX = c->yuv2planeX;
256 yuv2interleavedX_fn yuv2nv12cX = c->yuv2nv12cX;
257 yuv2packed1_fn yuv2packed1 = c->yuv2packed1;
258 yuv2packed2_fn yuv2packed2 = c->yuv2packed2;
259 yuv2packedX_fn yuv2packedX = c->yuv2packedX;
260 yuv2anyX_fn yuv2anyX = c->yuv2anyX;
261 const int chrSrcSliceY = srcSliceY >> c->chrSrcVSubSample;
262 const int chrSrcSliceH = AV_CEIL_RSHIFT(srcSliceH, c->chrSrcVSubSample);
263 int should_dither = isNBPS(c->srcFormat) ||
264 is16BPS(c->srcFormat);
265 int lastDstY;
266
267 /* vars which will change and which we need to store back in the context */
268 int dstY = c->dstY;
269 int lumBufIndex = c->lumBufIndex;
270 int chrBufIndex = c->chrBufIndex;
271 int lastInLumBuf = c->lastInLumBuf;
272 int lastInChrBuf = c->lastInChrBuf;
273
274
275 int lumStart = 0;
276 int lumEnd = c->descIndex[0];
277 int chrStart = lumEnd;
278 int chrEnd = c->descIndex[1];
279 int vStart = chrEnd;
280 int vEnd = c->numDesc;
281 SwsSlice *src_slice = &c->slice[lumStart];
282 SwsSlice *hout_slice = &c->slice[c->numSlice-2];
283 SwsSlice *vout_slice = &c->slice[c->numSlice-1];
284 SwsFilterDescriptor *desc = c->desc;
285
286
287 int needAlpha = c->needAlpha;
288
289 int hasLumHoles = 1;
290 int hasChrHoles = 1;
291
292
293 if (isPacked(c->srcFormat)) {
294 src[0] =
295 src[1] =
296 src[2] =
297 src[3] = src[0];
298 srcStride[0] =
299 srcStride[1] =
300 srcStride[2] =
301 srcStride[3] = srcStride[0];
302 }
303 srcStride[1] <<= c->vChrDrop;
304 srcStride[2] <<= c->vChrDrop;
305
306 DEBUG_BUFFERS("swscale() %p[%d] %p[%d] %p[%d] %p[%d] -> %p[%d] %p[%d] %p[%d] %p[%d]\n",
307 src[0], srcStride[0], src[1], srcStride[1],
308 src[2], srcStride[2], src[3], srcStride[3],
309 dst[0], dstStride[0], dst[1], dstStride[1],
310 dst[2], dstStride[2], dst[3], dstStride[3]);
311 DEBUG_BUFFERS("srcSliceY: %d srcSliceH: %d dstY: %d dstH: %d\n",
312 srcSliceY, srcSliceH, dstY, dstH);
313 DEBUG_BUFFERS("vLumFilterSize: %d vChrFilterSize: %d\n",
314 vLumFilterSize, vChrFilterSize);
315
316 if (dstStride[0]&15 || dstStride[1]&15 ||
317 dstStride[2]&15 || dstStride[3]&15) {
318 static int warnedAlready = 0; // FIXME maybe move this into the context
319 if (flags & SWS_PRINT_INFO && !warnedAlready) {
320 av_log(c, AV_LOG_WARNING,
321 "Warning: dstStride is not aligned!\n"
322 " ->cannot do aligned memory accesses anymore\n");
323 warnedAlready = 1;
324 }
325 }
326
327 if ( (uintptr_t)dst[0]&15 || (uintptr_t)dst[1]&15 || (uintptr_t)dst[2]&15
328 || (uintptr_t)src[0]&15 || (uintptr_t)src[1]&15 || (uintptr_t)src[2]&15
329 || dstStride[0]&15 || dstStride[1]&15 || dstStride[2]&15 || dstStride[3]&15
330 || srcStride[0]&15 || srcStride[1]&15 || srcStride[2]&15 || srcStride[3]&15
331 ) {
332 static int warnedAlready=0;
333 int cpu_flags = av_get_cpu_flags();
334 if (HAVE_MMXEXT && (cpu_flags & AV_CPU_FLAG_SSE2) && !warnedAlready){
335 av_log(c, AV_LOG_WARNING, "Warning: data is not aligned! This can lead to a speed loss\n");
336 warnedAlready=1;
337 }
338 }
339
340 /* Note the user might start scaling the picture in the middle so this
341 * will not get executed. This is not really intended but works
342 * currently, so people might do it. */
343 if (srcSliceY == 0) {
344 lumBufIndex = -1;
345 chrBufIndex = -1;
346 dstY = 0;
347 lastInLumBuf = -1;
348 lastInChrBuf = -1;
349 }
350
351 if (!should_dither) {
352 c->chrDither8 = c->lumDither8 = sws_pb_64;
353 }
354 lastDstY = dstY;
355
356 ff_init_vscale_pfn(c, yuv2plane1, yuv2planeX, yuv2nv12cX,
357 yuv2packed1, yuv2packed2, yuv2packedX, yuv2anyX, c->use_mmx_vfilter);
358
359 ff_init_slice_from_src(src_slice, (uint8_t**)src, srcStride, c->srcW,
360 srcSliceY, srcSliceH, chrSrcSliceY, chrSrcSliceH, 1);
361
362 ff_init_slice_from_src(vout_slice, (uint8_t**)dst, dstStride, c->dstW,
363 dstY, dstH, dstY >> c->chrDstVSubSample,
364 AV_CEIL_RSHIFT(dstH, c->chrDstVSubSample), 0);
365 if (srcSliceY == 0) {
366 hout_slice->plane[0].sliceY = lastInLumBuf + 1;
367 hout_slice->plane[1].sliceY = lastInChrBuf + 1;
368 hout_slice->plane[2].sliceY = lastInChrBuf + 1;
369 hout_slice->plane[3].sliceY = lastInLumBuf + 1;
370
371 hout_slice->plane[0].sliceH =
372 hout_slice->plane[1].sliceH =
373 hout_slice->plane[2].sliceH =
374 hout_slice->plane[3].sliceH = 0;
375 hout_slice->width = dstW;
376 }
377
378 for (; dstY < dstH; dstY++) {
379 const int chrDstY = dstY >> c->chrDstVSubSample;
380 int use_mmx_vfilter= c->use_mmx_vfilter;
381
382 // First line needed as input
383 const int firstLumSrcY = FFMAX(1 - vLumFilterSize, vLumFilterPos[dstY]);
384 const int firstLumSrcY2 = FFMAX(1 - vLumFilterSize, vLumFilterPos[FFMIN(dstY | ((1 << c->chrDstVSubSample) - 1), dstH - 1)]);
385 // First line needed as input
386 const int firstChrSrcY = FFMAX(1 - vChrFilterSize, vChrFilterPos[chrDstY]);
387
388 // Last line needed as input
389 int lastLumSrcY = FFMIN(c->srcH, firstLumSrcY + vLumFilterSize) - 1;
390 int lastLumSrcY2 = FFMIN(c->srcH, firstLumSrcY2 + vLumFilterSize) - 1;
391 int lastChrSrcY = FFMIN(c->chrSrcH, firstChrSrcY + vChrFilterSize) - 1;
392 int enough_lines;
393
394 int i;
395 int posY, cPosY, firstPosY, lastPosY, firstCPosY, lastCPosY;
396
397 // handle holes (FAST_BILINEAR & weird filters)
398 if (firstLumSrcY > lastInLumBuf) {
399
400 hasLumHoles = lastInLumBuf != firstLumSrcY - 1;
401 if (hasLumHoles) {
402 hout_slice->plane[0].sliceY = firstLumSrcY;
403 hout_slice->plane[3].sliceY = firstLumSrcY;
404 hout_slice->plane[0].sliceH =
405 hout_slice->plane[3].sliceH = 0;
406 }
407
408 lastInLumBuf = firstLumSrcY - 1;
409 }
410 if (firstChrSrcY > lastInChrBuf) {
411
412 hasChrHoles = lastInChrBuf != firstChrSrcY - 1;
413 if (hasChrHoles) {
414 hout_slice->plane[1].sliceY = firstChrSrcY;
415 hout_slice->plane[2].sliceY = firstChrSrcY;
416 hout_slice->plane[1].sliceH =
417 hout_slice->plane[2].sliceH = 0;
418 }
419
420 lastInChrBuf = firstChrSrcY - 1;
421 }
422
423 DEBUG_BUFFERS("dstY: %d\n", dstY);
424 DEBUG_BUFFERS("\tfirstLumSrcY: %d lastLumSrcY: %d lastInLumBuf: %d\n",
425 firstLumSrcY, lastLumSrcY, lastInLumBuf);
426 DEBUG_BUFFERS("\tfirstChrSrcY: %d lastChrSrcY: %d lastInChrBuf: %d\n",
427 firstChrSrcY, lastChrSrcY, lastInChrBuf);
428
429 // Do we have enough lines in this slice to output the dstY line
430 enough_lines = lastLumSrcY2 < srcSliceY + srcSliceH &&
431 lastChrSrcY < AV_CEIL_RSHIFT(srcSliceY + srcSliceH, c->chrSrcVSubSample);
432
433 if (!enough_lines) {
434 lastLumSrcY = srcSliceY + srcSliceH - 1;
435 lastChrSrcY = chrSrcSliceY + chrSrcSliceH - 1;
436 DEBUG_BUFFERS("buffering slice: lastLumSrcY %d lastChrSrcY %d\n",
437 lastLumSrcY, lastChrSrcY);
438 }
439
440 av_assert0((lastLumSrcY - firstLumSrcY + 1) <= hout_slice->plane[0].available_lines);
441 av_assert0((lastChrSrcY - firstChrSrcY + 1) <= hout_slice->plane[1].available_lines);
442
443
444 posY = hout_slice->plane[0].sliceY + hout_slice->plane[0].sliceH;
445 if (posY <= lastLumSrcY && !hasLumHoles) {
446 firstPosY = FFMAX(firstLumSrcY, posY);
447 lastPosY = FFMIN(firstLumSrcY + hout_slice->plane[0].available_lines - 1, srcSliceY + srcSliceH - 1);
448 } else {
449 firstPosY = posY;
450 lastPosY = lastLumSrcY;
451 }
452
453 cPosY = hout_slice->plane[1].sliceY + hout_slice->plane[1].sliceH;
454 if (cPosY <= lastChrSrcY && !hasChrHoles) {
455 firstCPosY = FFMAX(firstChrSrcY, cPosY);
456 lastCPosY = FFMIN(firstChrSrcY + hout_slice->plane[1].available_lines - 1, AV_CEIL_RSHIFT(srcSliceY + srcSliceH, c->chrSrcVSubSample) - 1);
457 } else {
458 firstCPosY = cPosY;
459 lastCPosY = lastChrSrcY;
460 }
461
462 ff_rotate_slice(hout_slice, lastPosY, lastCPosY);
463
464 if (posY < lastLumSrcY + 1) {
465 for (i = lumStart; i < lumEnd; ++i)
466 desc[i].process(c, &desc[i], firstPosY, lastPosY - firstPosY + 1);
467 }
468
469 lumBufIndex += lastLumSrcY - lastInLumBuf;
470 lastInLumBuf = lastLumSrcY;
471
472 if (cPosY < lastChrSrcY + 1) {
473 for (i = chrStart; i < chrEnd; ++i)
474 desc[i].process(c, &desc[i], firstCPosY, lastCPosY - firstCPosY + 1);
475 }
476
477 chrBufIndex += lastChrSrcY - lastInChrBuf;
478 lastInChrBuf = lastChrSrcY;
479
480 // wrap buf index around to stay inside the ring buffer
481 if (lumBufIndex >= vLumFilterSize)
482 lumBufIndex -= vLumFilterSize;
483 if (chrBufIndex >= vChrFilterSize)
484 chrBufIndex -= vChrFilterSize;
485 if (!enough_lines)
486 break; // we can't output a dstY line so let's try with the next slice
487
488 #if HAVE_MMX_INLINE
489 ff_updateMMXDitherTables(c, dstY, lumBufIndex, chrBufIndex,
490 lastInLumBuf, lastInChrBuf);
491 #endif
492 if (should_dither) {
493 c->chrDither8 = ff_dither_8x8_128[chrDstY & 7];
494 c->lumDither8 = ff_dither_8x8_128[dstY & 7];
495 }
496 if (dstY >= dstH - 2) {
497 /* hmm looks like we can't use MMX here without overwriting
498 * this array's tail */
499 ff_sws_init_output_funcs(c, &yuv2plane1, &yuv2planeX, &yuv2nv12cX,
500 &yuv2packed1, &yuv2packed2, &yuv2packedX, &yuv2anyX);
501 use_mmx_vfilter= 0;
502 ff_init_vscale_pfn(c, yuv2plane1, yuv2planeX, yuv2nv12cX,
503 yuv2packed1, yuv2packed2, yuv2packedX, yuv2anyX, use_mmx_vfilter);
504 }
505
506 {
507 for (i = vStart; i < vEnd; ++i)
508 desc[i].process(c, &desc[i], dstY, 1);
509 }
510 }
511 if (isPlanar(dstFormat) && isALPHA(dstFormat) && !needAlpha) {
512 int length = dstW;
513 int height = dstY - lastDstY;
514
515 if (is16BPS(dstFormat) || isNBPS(dstFormat)) {
516 const AVPixFmtDescriptor *desc = av_pix_fmt_desc_get(dstFormat);
517 fillPlane16(dst[3], dstStride[3], length, height, lastDstY,
518 1, desc->comp[3].depth,
519 isBE(dstFormat));
520 } else
521 fillPlane(dst[3], dstStride[3], length, height, lastDstY, 255);
522 }
523
524 #if HAVE_MMXEXT_INLINE
525 if (av_get_cpu_flags() & AV_CPU_FLAG_MMXEXT)
526 __asm__ volatile ("sfence" ::: "memory");
527 #endif
528 emms_c();
529
530 /* store changed local vars back in the context */
531 c->dstY = dstY;
532 c->lumBufIndex = lumBufIndex;
533 c->chrBufIndex = chrBufIndex;
534 c->lastInLumBuf = lastInLumBuf;
535 c->lastInChrBuf = lastInChrBuf;
536
537 return dstY - lastDstY;
538 }
539
ff_sws_init_range_convert(SwsContext * c)540 av_cold void ff_sws_init_range_convert(SwsContext *c)
541 {
542 c->lumConvertRange = NULL;
543 c->chrConvertRange = NULL;
544 if (c->srcRange != c->dstRange && !isAnyRGB(c->dstFormat)) {
545 if (c->dstBpc <= 14) {
546 if (c->srcRange) {
547 c->lumConvertRange = lumRangeFromJpeg_c;
548 c->chrConvertRange = chrRangeFromJpeg_c;
549 } else {
550 c->lumConvertRange = lumRangeToJpeg_c;
551 c->chrConvertRange = chrRangeToJpeg_c;
552 }
553 } else {
554 if (c->srcRange) {
555 c->lumConvertRange = lumRangeFromJpeg16_c;
556 c->chrConvertRange = chrRangeFromJpeg16_c;
557 } else {
558 c->lumConvertRange = lumRangeToJpeg16_c;
559 c->chrConvertRange = chrRangeToJpeg16_c;
560 }
561 }
562 }
563 }
564
sws_init_swscale(SwsContext * c)565 static av_cold void sws_init_swscale(SwsContext *c)
566 {
567 enum AVPixelFormat srcFormat = c->srcFormat;
568
569 ff_sws_init_output_funcs(c, &c->yuv2plane1, &c->yuv2planeX,
570 &c->yuv2nv12cX, &c->yuv2packed1,
571 &c->yuv2packed2, &c->yuv2packedX, &c->yuv2anyX);
572
573 ff_sws_init_input_funcs(c);
574
575
576 if (c->srcBpc == 8) {
577 if (c->dstBpc <= 14) {
578 c->hyScale = c->hcScale = hScale8To15_c;
579 if (c->flags & SWS_FAST_BILINEAR) {
580 c->hyscale_fast = ff_hyscale_fast_c;
581 c->hcscale_fast = ff_hcscale_fast_c;
582 }
583 } else {
584 c->hyScale = c->hcScale = hScale8To19_c;
585 }
586 } else {
587 c->hyScale = c->hcScale = c->dstBpc > 14 ? hScale16To19_c
588 : hScale16To15_c;
589 }
590
591 ff_sws_init_range_convert(c);
592
593 if (!(isGray(srcFormat) || isGray(c->dstFormat) ||
594 srcFormat == AV_PIX_FMT_MONOBLACK || srcFormat == AV_PIX_FMT_MONOWHITE))
595 c->needs_hcscale = 1;
596 }
597
ff_getSwsFunc(SwsContext * c)598 SwsFunc ff_getSwsFunc(SwsContext *c)
599 {
600 sws_init_swscale(c);
601
602 if (ARCH_PPC)
603 ff_sws_init_swscale_ppc(c);
604 if (ARCH_X86)
605 ff_sws_init_swscale_x86(c);
606 if (ARCH_AARCH64)
607 ff_sws_init_swscale_aarch64(c);
608 if (ARCH_ARM)
609 ff_sws_init_swscale_arm(c);
610
611 return swscale;
612 }
613
reset_ptr(const uint8_t * src[],enum AVPixelFormat format)614 static void reset_ptr(const uint8_t *src[], enum AVPixelFormat format)
615 {
616 if (!isALPHA(format))
617 src[3] = NULL;
618 if (!isPlanar(format)) {
619 src[3] = src[2] = NULL;
620
621 if (!usePal(format))
622 src[1] = NULL;
623 }
624 }
625
check_image_pointers(const uint8_t * const data[4],enum AVPixelFormat pix_fmt,const int linesizes[4])626 static int check_image_pointers(const uint8_t * const data[4], enum AVPixelFormat pix_fmt,
627 const int linesizes[4])
628 {
629 const AVPixFmtDescriptor *desc = av_pix_fmt_desc_get(pix_fmt);
630 int i;
631
632 av_assert2(desc);
633
634 for (i = 0; i < 4; i++) {
635 int plane = desc->comp[i].plane;
636 if (!data[plane] || !linesizes[plane])
637 return 0;
638 }
639
640 return 1;
641 }
642
xyz12Torgb48(struct SwsContext * c,uint16_t * dst,const uint16_t * src,int stride,int h)643 static void xyz12Torgb48(struct SwsContext *c, uint16_t *dst,
644 const uint16_t *src, int stride, int h)
645 {
646 int xp,yp;
647 const AVPixFmtDescriptor *desc = av_pix_fmt_desc_get(c->srcFormat);
648
649 for (yp=0; yp<h; yp++) {
650 for (xp=0; xp+2<stride; xp+=3) {
651 int x, y, z, r, g, b;
652
653 if (desc->flags & AV_PIX_FMT_FLAG_BE) {
654 x = AV_RB16(src + xp + 0);
655 y = AV_RB16(src + xp + 1);
656 z = AV_RB16(src + xp + 2);
657 } else {
658 x = AV_RL16(src + xp + 0);
659 y = AV_RL16(src + xp + 1);
660 z = AV_RL16(src + xp + 2);
661 }
662
663 x = c->xyzgamma[x>>4];
664 y = c->xyzgamma[y>>4];
665 z = c->xyzgamma[z>>4];
666
667 // convert from XYZlinear to sRGBlinear
668 r = c->xyz2rgb_matrix[0][0] * x +
669 c->xyz2rgb_matrix[0][1] * y +
670 c->xyz2rgb_matrix[0][2] * z >> 12;
671 g = c->xyz2rgb_matrix[1][0] * x +
672 c->xyz2rgb_matrix[1][1] * y +
673 c->xyz2rgb_matrix[1][2] * z >> 12;
674 b = c->xyz2rgb_matrix[2][0] * x +
675 c->xyz2rgb_matrix[2][1] * y +
676 c->xyz2rgb_matrix[2][2] * z >> 12;
677
678 // limit values to 12-bit depth
679 r = av_clip_uintp2(r, 12);
680 g = av_clip_uintp2(g, 12);
681 b = av_clip_uintp2(b, 12);
682
683 // convert from sRGBlinear to RGB and scale from 12bit to 16bit
684 if (desc->flags & AV_PIX_FMT_FLAG_BE) {
685 AV_WB16(dst + xp + 0, c->rgbgamma[r] << 4);
686 AV_WB16(dst + xp + 1, c->rgbgamma[g] << 4);
687 AV_WB16(dst + xp + 2, c->rgbgamma[b] << 4);
688 } else {
689 AV_WL16(dst + xp + 0, c->rgbgamma[r] << 4);
690 AV_WL16(dst + xp + 1, c->rgbgamma[g] << 4);
691 AV_WL16(dst + xp + 2, c->rgbgamma[b] << 4);
692 }
693 }
694 src += stride;
695 dst += stride;
696 }
697 }
698
rgb48Toxyz12(struct SwsContext * c,uint16_t * dst,const uint16_t * src,int stride,int h)699 static void rgb48Toxyz12(struct SwsContext *c, uint16_t *dst,
700 const uint16_t *src, int stride, int h)
701 {
702 int xp,yp;
703 const AVPixFmtDescriptor *desc = av_pix_fmt_desc_get(c->dstFormat);
704
705 for (yp=0; yp<h; yp++) {
706 for (xp=0; xp+2<stride; xp+=3) {
707 int x, y, z, r, g, b;
708
709 if (desc->flags & AV_PIX_FMT_FLAG_BE) {
710 r = AV_RB16(src + xp + 0);
711 g = AV_RB16(src + xp + 1);
712 b = AV_RB16(src + xp + 2);
713 } else {
714 r = AV_RL16(src + xp + 0);
715 g = AV_RL16(src + xp + 1);
716 b = AV_RL16(src + xp + 2);
717 }
718
719 r = c->rgbgammainv[r>>4];
720 g = c->rgbgammainv[g>>4];
721 b = c->rgbgammainv[b>>4];
722
723 // convert from sRGBlinear to XYZlinear
724 x = c->rgb2xyz_matrix[0][0] * r +
725 c->rgb2xyz_matrix[0][1] * g +
726 c->rgb2xyz_matrix[0][2] * b >> 12;
727 y = c->rgb2xyz_matrix[1][0] * r +
728 c->rgb2xyz_matrix[1][1] * g +
729 c->rgb2xyz_matrix[1][2] * b >> 12;
730 z = c->rgb2xyz_matrix[2][0] * r +
731 c->rgb2xyz_matrix[2][1] * g +
732 c->rgb2xyz_matrix[2][2] * b >> 12;
733
734 // limit values to 12-bit depth
735 x = av_clip_uintp2(x, 12);
736 y = av_clip_uintp2(y, 12);
737 z = av_clip_uintp2(z, 12);
738
739 // convert from XYZlinear to X'Y'Z' and scale from 12bit to 16bit
740 if (desc->flags & AV_PIX_FMT_FLAG_BE) {
741 AV_WB16(dst + xp + 0, c->xyzgammainv[x] << 4);
742 AV_WB16(dst + xp + 1, c->xyzgammainv[y] << 4);
743 AV_WB16(dst + xp + 2, c->xyzgammainv[z] << 4);
744 } else {
745 AV_WL16(dst + xp + 0, c->xyzgammainv[x] << 4);
746 AV_WL16(dst + xp + 1, c->xyzgammainv[y] << 4);
747 AV_WL16(dst + xp + 2, c->xyzgammainv[z] << 4);
748 }
749 }
750 src += stride;
751 dst += stride;
752 }
753 }
754
755 /**
756 * swscale wrapper, so we don't need to export the SwsContext.
757 * Assumes planar YUV to be in YUV order instead of YVU.
758 */
sws_scale(struct SwsContext * c,const uint8_t * const srcSlice[],const int srcStride[],int srcSliceY,int srcSliceH,uint8_t * const dst[],const int dstStride[])759 int attribute_align_arg sws_scale(struct SwsContext *c,
760 const uint8_t * const srcSlice[],
761 const int srcStride[], int srcSliceY,
762 int srcSliceH, uint8_t *const dst[],
763 const int dstStride[])
764 {
765 int i, ret;
766 const uint8_t *src2[4];
767 uint8_t *dst2[4];
768 uint8_t *rgb0_tmp = NULL;
769 int macro_height = isBayer(c->srcFormat) ? 2 : (1 << c->chrSrcVSubSample);
770 // copy strides, so they can safely be modified
771 int srcStride2[4];
772 int dstStride2[4];
773 int srcSliceY_internal = srcSliceY;
774
775 if (!srcStride || !dstStride || !dst || !srcSlice) {
776 av_log(c, AV_LOG_ERROR, "One of the input parameters to sws_scale() is NULL, please check the calling code\n");
777 return 0;
778 }
779
780 for (i=0; i<4; i++) {
781 srcStride2[i] = srcStride[i];
782 dstStride2[i] = dstStride[i];
783 }
784
785 if ((srcSliceY & (macro_height-1)) ||
786 ((srcSliceH& (macro_height-1)) && srcSliceY + srcSliceH != c->srcH) ||
787 srcSliceY + srcSliceH > c->srcH) {
788 av_log(c, AV_LOG_ERROR, "Slice parameters %d, %d are invalid\n", srcSliceY, srcSliceH);
789 return AVERROR(EINVAL);
790 }
791
792 if (c->gamma_flag && c->cascaded_context[0]) {
793
794
795 ret = sws_scale(c->cascaded_context[0],
796 srcSlice, srcStride, srcSliceY, srcSliceH,
797 c->cascaded_tmp, c->cascaded_tmpStride);
798
799 if (ret < 0)
800 return ret;
801
802 if (c->cascaded_context[2])
803 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);
804 else
805 ret = sws_scale(c->cascaded_context[1], (const uint8_t * const *)c->cascaded_tmp, c->cascaded_tmpStride, srcSliceY, srcSliceH, dst, dstStride);
806
807 if (ret < 0)
808 return ret;
809
810 if (c->cascaded_context[2]) {
811 ret = sws_scale(c->cascaded_context[2],
812 (const uint8_t * const *)c->cascaded1_tmp, c->cascaded1_tmpStride, c->cascaded_context[1]->dstY - ret, c->cascaded_context[1]->dstY,
813 dst, dstStride);
814 }
815 return ret;
816 }
817
818 if (c->cascaded_context[0] && srcSliceY == 0 && srcSliceH == c->cascaded_context[0]->srcH) {
819 ret = sws_scale(c->cascaded_context[0],
820 srcSlice, srcStride, srcSliceY, srcSliceH,
821 c->cascaded_tmp, c->cascaded_tmpStride);
822 if (ret < 0)
823 return ret;
824 ret = sws_scale(c->cascaded_context[1],
825 (const uint8_t * const * )c->cascaded_tmp, c->cascaded_tmpStride, 0, c->cascaded_context[0]->dstH,
826 dst, dstStride);
827 return ret;
828 }
829
830 memcpy(src2, srcSlice, sizeof(src2));
831 memcpy(dst2, dst, sizeof(dst2));
832
833 // do not mess up sliceDir if we have a "trailing" 0-size slice
834 if (srcSliceH == 0)
835 return 0;
836
837 if (!check_image_pointers(srcSlice, c->srcFormat, srcStride)) {
838 av_log(c, AV_LOG_ERROR, "bad src image pointers\n");
839 return 0;
840 }
841 if (!check_image_pointers((const uint8_t* const*)dst, c->dstFormat, dstStride)) {
842 av_log(c, AV_LOG_ERROR, "bad dst image pointers\n");
843 return 0;
844 }
845
846 if (c->sliceDir == 0 && srcSliceY != 0 && srcSliceY + srcSliceH != c->srcH) {
847 av_log(c, AV_LOG_ERROR, "Slices start in the middle!\n");
848 return 0;
849 }
850 if (c->sliceDir == 0) {
851 if (srcSliceY == 0) c->sliceDir = 1; else c->sliceDir = -1;
852 }
853
854 if (usePal(c->srcFormat)) {
855 for (i = 0; i < 256; i++) {
856 int r, g, b, y, u, v, a = 0xff;
857 if (c->srcFormat == AV_PIX_FMT_PAL8) {
858 uint32_t p = ((const uint32_t *)(srcSlice[1]))[i];
859 a = (p >> 24) & 0xFF;
860 r = (p >> 16) & 0xFF;
861 g = (p >> 8) & 0xFF;
862 b = p & 0xFF;
863 } else if (c->srcFormat == AV_PIX_FMT_RGB8) {
864 r = ( i >> 5 ) * 36;
865 g = ((i >> 2) & 7) * 36;
866 b = ( i & 3) * 85;
867 } else if (c->srcFormat == AV_PIX_FMT_BGR8) {
868 b = ( i >> 6 ) * 85;
869 g = ((i >> 3) & 7) * 36;
870 r = ( i & 7) * 36;
871 } else if (c->srcFormat == AV_PIX_FMT_RGB4_BYTE) {
872 r = ( i >> 3 ) * 255;
873 g = ((i >> 1) & 3) * 85;
874 b = ( i & 1) * 255;
875 } else if (c->srcFormat == AV_PIX_FMT_GRAY8 || c->srcFormat == AV_PIX_FMT_GRAY8A) {
876 r = g = b = i;
877 } else {
878 av_assert1(c->srcFormat == AV_PIX_FMT_BGR4_BYTE);
879 b = ( i >> 3 ) * 255;
880 g = ((i >> 1) & 3) * 85;
881 r = ( i & 1) * 255;
882 }
883 #define RGB2YUV_SHIFT 15
884 #define BY ( (int) (0.114 * 219 / 255 * (1 << RGB2YUV_SHIFT) + 0.5))
885 #define BV (-(int) (0.081 * 224 / 255 * (1 << RGB2YUV_SHIFT) + 0.5))
886 #define BU ( (int) (0.500 * 224 / 255 * (1 << RGB2YUV_SHIFT) + 0.5))
887 #define GY ( (int) (0.587 * 219 / 255 * (1 << RGB2YUV_SHIFT) + 0.5))
888 #define GV (-(int) (0.419 * 224 / 255 * (1 << RGB2YUV_SHIFT) + 0.5))
889 #define GU (-(int) (0.331 * 224 / 255 * (1 << RGB2YUV_SHIFT) + 0.5))
890 #define RY ( (int) (0.299 * 219 / 255 * (1 << RGB2YUV_SHIFT) + 0.5))
891 #define RV ( (int) (0.500 * 224 / 255 * (1 << RGB2YUV_SHIFT) + 0.5))
892 #define RU (-(int) (0.169 * 224 / 255 * (1 << RGB2YUV_SHIFT) + 0.5))
893
894 y = av_clip_uint8((RY * r + GY * g + BY * b + ( 33 << (RGB2YUV_SHIFT - 1))) >> RGB2YUV_SHIFT);
895 u = av_clip_uint8((RU * r + GU * g + BU * b + (257 << (RGB2YUV_SHIFT - 1))) >> RGB2YUV_SHIFT);
896 v = av_clip_uint8((RV * r + GV * g + BV * b + (257 << (RGB2YUV_SHIFT - 1))) >> RGB2YUV_SHIFT);
897 c->pal_yuv[i]= y + (u<<8) + (v<<16) + ((unsigned)a<<24);
898
899 switch (c->dstFormat) {
900 case AV_PIX_FMT_BGR32:
901 #if !HAVE_BIGENDIAN
902 case AV_PIX_FMT_RGB24:
903 #endif
904 c->pal_rgb[i]= r + (g<<8) + (b<<16) + ((unsigned)a<<24);
905 break;
906 case AV_PIX_FMT_BGR32_1:
907 #if HAVE_BIGENDIAN
908 case AV_PIX_FMT_BGR24:
909 #endif
910 c->pal_rgb[i]= a + (r<<8) + (g<<16) + ((unsigned)b<<24);
911 break;
912 case AV_PIX_FMT_RGB32_1:
913 #if HAVE_BIGENDIAN
914 case AV_PIX_FMT_RGB24:
915 #endif
916 c->pal_rgb[i]= a + (b<<8) + (g<<16) + ((unsigned)r<<24);
917 break;
918 case AV_PIX_FMT_RGB32:
919 #if !HAVE_BIGENDIAN
920 case AV_PIX_FMT_BGR24:
921 #endif
922 default:
923 c->pal_rgb[i]= b + (g<<8) + (r<<16) + ((unsigned)a<<24);
924 }
925 }
926 }
927
928 if (c->src0Alpha && !c->dst0Alpha && isALPHA(c->dstFormat)) {
929 uint8_t *base;
930 int x,y;
931 rgb0_tmp = av_malloc(FFABS(srcStride[0]) * srcSliceH + 32);
932 if (!rgb0_tmp)
933 return AVERROR(ENOMEM);
934
935 base = srcStride[0] < 0 ? rgb0_tmp - srcStride[0] * (srcSliceH-1) : rgb0_tmp;
936 for (y=0; y<srcSliceH; y++){
937 memcpy(base + srcStride[0]*y, src2[0] + srcStride[0]*y, 4*c->srcW);
938 for (x=c->src0Alpha-1; x<4*c->srcW; x+=4) {
939 base[ srcStride[0]*y + x] = 0xFF;
940 }
941 }
942 src2[0] = base;
943 }
944
945 if (c->srcXYZ && !(c->dstXYZ && c->srcW==c->dstW && c->srcH==c->dstH)) {
946 uint8_t *base;
947 rgb0_tmp = av_malloc(FFABS(srcStride[0]) * srcSliceH + 32);
948 if (!rgb0_tmp)
949 return AVERROR(ENOMEM);
950
951 base = srcStride[0] < 0 ? rgb0_tmp - srcStride[0] * (srcSliceH-1) : rgb0_tmp;
952
953 xyz12Torgb48(c, (uint16_t*)base, (const uint16_t*)src2[0], srcStride[0]/2, srcSliceH);
954 src2[0] = base;
955 }
956
957 if (!srcSliceY && (c->flags & SWS_BITEXACT) && c->dither == SWS_DITHER_ED && c->dither_error[0])
958 for (i = 0; i < 4; i++)
959 memset(c->dither_error[i], 0, sizeof(c->dither_error[0][0]) * (c->dstW+2));
960
961 if (c->sliceDir != 1) {
962 // slices go from bottom to top => we flip the image internally
963 for (i=0; i<4; i++) {
964 srcStride2[i] *= -1;
965 dstStride2[i] *= -1;
966 }
967
968 src2[0] += (srcSliceH - 1) * srcStride[0];
969 if (!usePal(c->srcFormat))
970 src2[1] += ((srcSliceH >> c->chrSrcVSubSample) - 1) * srcStride[1];
971 src2[2] += ((srcSliceH >> c->chrSrcVSubSample) - 1) * srcStride[2];
972 src2[3] += (srcSliceH - 1) * srcStride[3];
973 dst2[0] += ( c->dstH - 1) * dstStride[0];
974 dst2[1] += ((c->dstH >> c->chrDstVSubSample) - 1) * dstStride[1];
975 dst2[2] += ((c->dstH >> c->chrDstVSubSample) - 1) * dstStride[2];
976 dst2[3] += ( c->dstH - 1) * dstStride[3];
977
978 srcSliceY_internal = c->srcH-srcSliceY-srcSliceH;
979 }
980 reset_ptr(src2, c->srcFormat);
981 reset_ptr((void*)dst2, c->dstFormat);
982
983 /* reset slice direction at end of frame */
984 if (srcSliceY_internal + srcSliceH == c->srcH)
985 c->sliceDir = 0;
986 ret = c->swscale(c, src2, srcStride2, srcSliceY_internal, srcSliceH, dst2, dstStride2);
987
988
989 if (c->dstXYZ && !(c->srcXYZ && c->srcW==c->dstW && c->srcH==c->dstH)) {
990 int dstY = c->dstY ? c->dstY : srcSliceY + srcSliceH;
991 uint16_t *dst16 = (uint16_t*)(dst2[0] + (dstY - ret) * dstStride2[0]);
992 av_assert0(dstY >= ret);
993 av_assert0(ret >= 0);
994 av_assert0(c->dstH >= dstY);
995
996 /* replace on the same data */
997 rgb48Toxyz12(c, dst16, dst16, dstStride2[0]/2, ret);
998 }
999
1000 av_free(rgb0_tmp);
1001 return ret;
1002 }
1003