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 lastInLumBuf = c->lastInLumBuf;
270 int lastInChrBuf = c->lastInChrBuf;
271
272 int lumStart = 0;
273 int lumEnd = c->descIndex[0];
274 int chrStart = lumEnd;
275 int chrEnd = c->descIndex[1];
276 int vStart = chrEnd;
277 int vEnd = c->numDesc;
278 SwsSlice *src_slice = &c->slice[lumStart];
279 SwsSlice *hout_slice = &c->slice[c->numSlice-2];
280 SwsSlice *vout_slice = &c->slice[c->numSlice-1];
281 SwsFilterDescriptor *desc = c->desc;
282
283 int needAlpha = c->needAlpha;
284
285 int hasLumHoles = 1;
286 int hasChrHoles = 1;
287
288 if (isPacked(c->srcFormat)) {
289 src[1] =
290 src[2] =
291 src[3] = src[0];
292 srcStride[1] =
293 srcStride[2] =
294 srcStride[3] = srcStride[0];
295 }
296 srcStride[1] *= 1 << c->vChrDrop;
297 srcStride[2] *= 1 << c->vChrDrop;
298
299 DEBUG_BUFFERS("swscale() %p[%d] %p[%d] %p[%d] %p[%d] -> %p[%d] %p[%d] %p[%d] %p[%d]\n",
300 src[0], srcStride[0], src[1], srcStride[1],
301 src[2], srcStride[2], src[3], srcStride[3],
302 dst[0], dstStride[0], dst[1], dstStride[1],
303 dst[2], dstStride[2], dst[3], dstStride[3]);
304 DEBUG_BUFFERS("srcSliceY: %d srcSliceH: %d dstY: %d dstH: %d\n",
305 srcSliceY, srcSliceH, dstY, dstH);
306 DEBUG_BUFFERS("vLumFilterSize: %d vChrFilterSize: %d\n",
307 vLumFilterSize, vChrFilterSize);
308
309 if (dstStride[0]&15 || dstStride[1]&15 ||
310 dstStride[2]&15 || dstStride[3]&15) {
311 static int warnedAlready = 0; // FIXME maybe move this into the context
312 if (flags & SWS_PRINT_INFO && !warnedAlready) {
313 av_log(c, AV_LOG_WARNING,
314 "Warning: dstStride is not aligned!\n"
315 " ->cannot do aligned memory accesses anymore\n");
316 warnedAlready = 1;
317 }
318 }
319
320 if ( (uintptr_t)dst[0]&15 || (uintptr_t)dst[1]&15 || (uintptr_t)dst[2]&15
321 || (uintptr_t)src[0]&15 || (uintptr_t)src[1]&15 || (uintptr_t)src[2]&15
322 || dstStride[0]&15 || dstStride[1]&15 || dstStride[2]&15 || dstStride[3]&15
323 || srcStride[0]&15 || srcStride[1]&15 || srcStride[2]&15 || srcStride[3]&15
324 ) {
325 static int warnedAlready=0;
326 int cpu_flags = av_get_cpu_flags();
327 if (HAVE_MMXEXT && (cpu_flags & AV_CPU_FLAG_SSE2) && !warnedAlready){
328 av_log(c, AV_LOG_WARNING, "Warning: data is not aligned! This can lead to a speed loss\n");
329 warnedAlready=1;
330 }
331 }
332
333 /* Note the user might start scaling the picture in the middle so this
334 * will not get executed. This is not really intended but works
335 * currently, so people might do it. */
336 if (srcSliceY == 0) {
337 dstY = 0;
338 lastInLumBuf = -1;
339 lastInChrBuf = -1;
340 }
341
342 if (!should_dither) {
343 c->chrDither8 = c->lumDither8 = sws_pb_64;
344 }
345 lastDstY = dstY;
346
347 ff_init_vscale_pfn(c, yuv2plane1, yuv2planeX, yuv2nv12cX,
348 yuv2packed1, yuv2packed2, yuv2packedX, yuv2anyX, c->use_mmx_vfilter);
349
350 ff_init_slice_from_src(src_slice, (uint8_t**)src, srcStride, c->srcW,
351 srcSliceY, srcSliceH, chrSrcSliceY, chrSrcSliceH, 1);
352
353 ff_init_slice_from_src(vout_slice, (uint8_t**)dst, dstStride, c->dstW,
354 dstY, dstH, dstY >> c->chrDstVSubSample,
355 AV_CEIL_RSHIFT(dstH, c->chrDstVSubSample), 0);
356 if (srcSliceY == 0) {
357 hout_slice->plane[0].sliceY = lastInLumBuf + 1;
358 hout_slice->plane[1].sliceY = lastInChrBuf + 1;
359 hout_slice->plane[2].sliceY = lastInChrBuf + 1;
360 hout_slice->plane[3].sliceY = lastInLumBuf + 1;
361
362 hout_slice->plane[0].sliceH =
363 hout_slice->plane[1].sliceH =
364 hout_slice->plane[2].sliceH =
365 hout_slice->plane[3].sliceH = 0;
366 hout_slice->width = dstW;
367 }
368
369 for (; dstY < dstH; dstY++) {
370 const int chrDstY = dstY >> c->chrDstVSubSample;
371 int use_mmx_vfilter= c->use_mmx_vfilter;
372
373 // First line needed as input
374 const int firstLumSrcY = FFMAX(1 - vLumFilterSize, vLumFilterPos[dstY]);
375 const int firstLumSrcY2 = FFMAX(1 - vLumFilterSize, vLumFilterPos[FFMIN(dstY | ((1 << c->chrDstVSubSample) - 1), dstH - 1)]);
376 // First line needed as input
377 const int firstChrSrcY = FFMAX(1 - vChrFilterSize, vChrFilterPos[chrDstY]);
378
379 // Last line needed as input
380 int lastLumSrcY = FFMIN(c->srcH, firstLumSrcY + vLumFilterSize) - 1;
381 int lastLumSrcY2 = FFMIN(c->srcH, firstLumSrcY2 + vLumFilterSize) - 1;
382 int lastChrSrcY = FFMIN(c->chrSrcH, firstChrSrcY + vChrFilterSize) - 1;
383 int enough_lines;
384
385 int i;
386 int posY, cPosY, firstPosY, lastPosY, firstCPosY, lastCPosY;
387
388 // handle holes (FAST_BILINEAR & weird filters)
389 if (firstLumSrcY > lastInLumBuf) {
390
391 hasLumHoles = lastInLumBuf != firstLumSrcY - 1;
392 if (hasLumHoles) {
393 hout_slice->plane[0].sliceY = firstLumSrcY;
394 hout_slice->plane[3].sliceY = firstLumSrcY;
395 hout_slice->plane[0].sliceH =
396 hout_slice->plane[3].sliceH = 0;
397 }
398
399 lastInLumBuf = firstLumSrcY - 1;
400 }
401 if (firstChrSrcY > lastInChrBuf) {
402
403 hasChrHoles = lastInChrBuf != firstChrSrcY - 1;
404 if (hasChrHoles) {
405 hout_slice->plane[1].sliceY = firstChrSrcY;
406 hout_slice->plane[2].sliceY = firstChrSrcY;
407 hout_slice->plane[1].sliceH =
408 hout_slice->plane[2].sliceH = 0;
409 }
410
411 lastInChrBuf = firstChrSrcY - 1;
412 }
413
414 DEBUG_BUFFERS("dstY: %d\n", dstY);
415 DEBUG_BUFFERS("\tfirstLumSrcY: %d lastLumSrcY: %d lastInLumBuf: %d\n",
416 firstLumSrcY, lastLumSrcY, lastInLumBuf);
417 DEBUG_BUFFERS("\tfirstChrSrcY: %d lastChrSrcY: %d lastInChrBuf: %d\n",
418 firstChrSrcY, lastChrSrcY, lastInChrBuf);
419
420 // Do we have enough lines in this slice to output the dstY line
421 enough_lines = lastLumSrcY2 < srcSliceY + srcSliceH &&
422 lastChrSrcY < AV_CEIL_RSHIFT(srcSliceY + srcSliceH, c->chrSrcVSubSample);
423
424 if (!enough_lines) {
425 lastLumSrcY = srcSliceY + srcSliceH - 1;
426 lastChrSrcY = chrSrcSliceY + chrSrcSliceH - 1;
427 DEBUG_BUFFERS("buffering slice: lastLumSrcY %d lastChrSrcY %d\n",
428 lastLumSrcY, lastChrSrcY);
429 }
430
431 av_assert0((lastLumSrcY - firstLumSrcY + 1) <= hout_slice->plane[0].available_lines);
432 av_assert0((lastChrSrcY - firstChrSrcY + 1) <= hout_slice->plane[1].available_lines);
433
434
435 posY = hout_slice->plane[0].sliceY + hout_slice->plane[0].sliceH;
436 if (posY <= lastLumSrcY && !hasLumHoles) {
437 firstPosY = FFMAX(firstLumSrcY, posY);
438 lastPosY = FFMIN(firstLumSrcY + hout_slice->plane[0].available_lines - 1, srcSliceY + srcSliceH - 1);
439 } else {
440 firstPosY = posY;
441 lastPosY = lastLumSrcY;
442 }
443
444 cPosY = hout_slice->plane[1].sliceY + hout_slice->plane[1].sliceH;
445 if (cPosY <= lastChrSrcY && !hasChrHoles) {
446 firstCPosY = FFMAX(firstChrSrcY, cPosY);
447 lastCPosY = FFMIN(firstChrSrcY + hout_slice->plane[1].available_lines - 1, AV_CEIL_RSHIFT(srcSliceY + srcSliceH, c->chrSrcVSubSample) - 1);
448 } else {
449 firstCPosY = cPosY;
450 lastCPosY = lastChrSrcY;
451 }
452
453 ff_rotate_slice(hout_slice, lastPosY, lastCPosY);
454
455 if (posY < lastLumSrcY + 1) {
456 for (i = lumStart; i < lumEnd; ++i)
457 desc[i].process(c, &desc[i], firstPosY, lastPosY - firstPosY + 1);
458 }
459
460 lastInLumBuf = lastLumSrcY;
461
462 if (cPosY < lastChrSrcY + 1) {
463 for (i = chrStart; i < chrEnd; ++i)
464 desc[i].process(c, &desc[i], firstCPosY, lastCPosY - firstCPosY + 1);
465 }
466
467 lastInChrBuf = lastChrSrcY;
468
469 if (!enough_lines)
470 break; // we can't output a dstY line so let's try with the next slice
471
472 #if HAVE_MMX_INLINE
473 ff_updateMMXDitherTables(c, dstY);
474 #endif
475 if (should_dither) {
476 c->chrDither8 = ff_dither_8x8_128[chrDstY & 7];
477 c->lumDither8 = ff_dither_8x8_128[dstY & 7];
478 }
479 if (dstY >= dstH - 2) {
480 /* hmm looks like we can't use MMX here without overwriting
481 * this array's tail */
482 ff_sws_init_output_funcs(c, &yuv2plane1, &yuv2planeX, &yuv2nv12cX,
483 &yuv2packed1, &yuv2packed2, &yuv2packedX, &yuv2anyX);
484 use_mmx_vfilter= 0;
485 ff_init_vscale_pfn(c, yuv2plane1, yuv2planeX, yuv2nv12cX,
486 yuv2packed1, yuv2packed2, yuv2packedX, yuv2anyX, use_mmx_vfilter);
487 }
488
489 {
490 for (i = vStart; i < vEnd; ++i)
491 desc[i].process(c, &desc[i], dstY, 1);
492 }
493 }
494 if (isPlanar(dstFormat) && isALPHA(dstFormat) && !needAlpha) {
495 int length = dstW;
496 int height = dstY - lastDstY;
497
498 if (is16BPS(dstFormat) || isNBPS(dstFormat)) {
499 const AVPixFmtDescriptor *desc = av_pix_fmt_desc_get(dstFormat);
500 fillPlane16(dst[3], dstStride[3], length, height, lastDstY,
501 1, desc->comp[3].depth,
502 isBE(dstFormat));
503 } else if (is32BPS(dstFormat)) {
504 const AVPixFmtDescriptor *desc = av_pix_fmt_desc_get(dstFormat);
505 fillPlane32(dst[3], dstStride[3], length, height, lastDstY,
506 1, desc->comp[3].depth,
507 isBE(dstFormat), desc->flags & AV_PIX_FMT_FLAG_FLOAT);
508 } else
509 fillPlane(dst[3], dstStride[3], length, height, lastDstY, 255);
510 }
511
512 #if HAVE_MMXEXT_INLINE
513 if (av_get_cpu_flags() & AV_CPU_FLAG_MMXEXT)
514 __asm__ volatile ("sfence" ::: "memory");
515 #endif
516 emms_c();
517
518 /* store changed local vars back in the context */
519 c->dstY = dstY;
520 c->lastInLumBuf = lastInLumBuf;
521 c->lastInChrBuf = lastInChrBuf;
522
523 return dstY - lastDstY;
524 }
525
ff_sws_init_range_convert(SwsContext * c)526 av_cold void ff_sws_init_range_convert(SwsContext *c)
527 {
528 c->lumConvertRange = NULL;
529 c->chrConvertRange = NULL;
530 if (c->srcRange != c->dstRange && !isAnyRGB(c->dstFormat)) {
531 if (c->dstBpc <= 14) {
532 if (c->srcRange) {
533 c->lumConvertRange = lumRangeFromJpeg_c;
534 c->chrConvertRange = chrRangeFromJpeg_c;
535 } else {
536 c->lumConvertRange = lumRangeToJpeg_c;
537 c->chrConvertRange = chrRangeToJpeg_c;
538 }
539 } else {
540 if (c->srcRange) {
541 c->lumConvertRange = lumRangeFromJpeg16_c;
542 c->chrConvertRange = chrRangeFromJpeg16_c;
543 } else {
544 c->lumConvertRange = lumRangeToJpeg16_c;
545 c->chrConvertRange = chrRangeToJpeg16_c;
546 }
547 }
548 }
549 }
550
sws_init_swscale(SwsContext * c)551 static av_cold void sws_init_swscale(SwsContext *c)
552 {
553 enum AVPixelFormat srcFormat = c->srcFormat;
554
555 ff_sws_init_output_funcs(c, &c->yuv2plane1, &c->yuv2planeX,
556 &c->yuv2nv12cX, &c->yuv2packed1,
557 &c->yuv2packed2, &c->yuv2packedX, &c->yuv2anyX);
558
559 ff_sws_init_input_funcs(c);
560
561 if (c->srcBpc == 8) {
562 if (c->dstBpc <= 14) {
563 c->hyScale = c->hcScale = hScale8To15_c;
564 if (c->flags & SWS_FAST_BILINEAR) {
565 c->hyscale_fast = ff_hyscale_fast_c;
566 c->hcscale_fast = ff_hcscale_fast_c;
567 }
568 } else {
569 c->hyScale = c->hcScale = hScale8To19_c;
570 }
571 } else {
572 c->hyScale = c->hcScale = c->dstBpc > 14 ? hScale16To19_c
573 : hScale16To15_c;
574 }
575
576 ff_sws_init_range_convert(c);
577
578 if (!(isGray(srcFormat) || isGray(c->dstFormat) ||
579 srcFormat == AV_PIX_FMT_MONOBLACK || srcFormat == AV_PIX_FMT_MONOWHITE))
580 c->needs_hcscale = 1;
581 }
582
ff_getSwsFunc(SwsContext * c)583 SwsFunc ff_getSwsFunc(SwsContext *c)
584 {
585 sws_init_swscale(c);
586
587 if (ARCH_PPC)
588 ff_sws_init_swscale_ppc(c);
589 if (ARCH_X86)
590 ff_sws_init_swscale_x86(c);
591 if (ARCH_AARCH64)
592 ff_sws_init_swscale_aarch64(c);
593 if (ARCH_ARM)
594 ff_sws_init_swscale_arm(c);
595
596 return swscale;
597 }
598
reset_ptr(const uint8_t * src[],enum AVPixelFormat format)599 static void reset_ptr(const uint8_t *src[], enum AVPixelFormat format)
600 {
601 if (!isALPHA(format))
602 src[3] = NULL;
603 if (!isPlanar(format)) {
604 src[3] = src[2] = NULL;
605
606 if (!usePal(format))
607 src[1] = NULL;
608 }
609 }
610
check_image_pointers(const uint8_t * const data[4],enum AVPixelFormat pix_fmt,const int linesizes[4])611 static int check_image_pointers(const uint8_t * const data[4], enum AVPixelFormat pix_fmt,
612 const int linesizes[4])
613 {
614 const AVPixFmtDescriptor *desc = av_pix_fmt_desc_get(pix_fmt);
615 int i;
616
617 av_assert2(desc);
618
619 for (i = 0; i < 4; i++) {
620 int plane = desc->comp[i].plane;
621 if (!data[plane] || !linesizes[plane])
622 return 0;
623 }
624
625 return 1;
626 }
627
xyz12Torgb48(struct SwsContext * c,uint16_t * dst,const uint16_t * src,int stride,int h)628 static void xyz12Torgb48(struct SwsContext *c, uint16_t *dst,
629 const uint16_t *src, int stride, int h)
630 {
631 int xp,yp;
632 const AVPixFmtDescriptor *desc = av_pix_fmt_desc_get(c->srcFormat);
633
634 for (yp=0; yp<h; yp++) {
635 for (xp=0; xp+2<stride; xp+=3) {
636 int x, y, z, r, g, b;
637
638 if (desc->flags & AV_PIX_FMT_FLAG_BE) {
639 x = AV_RB16(src + xp + 0);
640 y = AV_RB16(src + xp + 1);
641 z = AV_RB16(src + xp + 2);
642 } else {
643 x = AV_RL16(src + xp + 0);
644 y = AV_RL16(src + xp + 1);
645 z = AV_RL16(src + xp + 2);
646 }
647
648 x = c->xyzgamma[x>>4];
649 y = c->xyzgamma[y>>4];
650 z = c->xyzgamma[z>>4];
651
652 // convert from XYZlinear to sRGBlinear
653 r = c->xyz2rgb_matrix[0][0] * x +
654 c->xyz2rgb_matrix[0][1] * y +
655 c->xyz2rgb_matrix[0][2] * z >> 12;
656 g = c->xyz2rgb_matrix[1][0] * x +
657 c->xyz2rgb_matrix[1][1] * y +
658 c->xyz2rgb_matrix[1][2] * z >> 12;
659 b = c->xyz2rgb_matrix[2][0] * x +
660 c->xyz2rgb_matrix[2][1] * y +
661 c->xyz2rgb_matrix[2][2] * z >> 12;
662
663 // limit values to 12-bit depth
664 r = av_clip_uintp2(r, 12);
665 g = av_clip_uintp2(g, 12);
666 b = av_clip_uintp2(b, 12);
667
668 // convert from sRGBlinear to RGB and scale from 12bit to 16bit
669 if (desc->flags & AV_PIX_FMT_FLAG_BE) {
670 AV_WB16(dst + xp + 0, c->rgbgamma[r] << 4);
671 AV_WB16(dst + xp + 1, c->rgbgamma[g] << 4);
672 AV_WB16(dst + xp + 2, c->rgbgamma[b] << 4);
673 } else {
674 AV_WL16(dst + xp + 0, c->rgbgamma[r] << 4);
675 AV_WL16(dst + xp + 1, c->rgbgamma[g] << 4);
676 AV_WL16(dst + xp + 2, c->rgbgamma[b] << 4);
677 }
678 }
679 src += stride;
680 dst += stride;
681 }
682 }
683
rgb48Toxyz12(struct SwsContext * c,uint16_t * dst,const uint16_t * src,int stride,int h)684 static void rgb48Toxyz12(struct SwsContext *c, uint16_t *dst,
685 const uint16_t *src, int stride, int h)
686 {
687 int xp,yp;
688 const AVPixFmtDescriptor *desc = av_pix_fmt_desc_get(c->dstFormat);
689
690 for (yp=0; yp<h; yp++) {
691 for (xp=0; xp+2<stride; xp+=3) {
692 int x, y, z, r, g, b;
693
694 if (desc->flags & AV_PIX_FMT_FLAG_BE) {
695 r = AV_RB16(src + xp + 0);
696 g = AV_RB16(src + xp + 1);
697 b = AV_RB16(src + xp + 2);
698 } else {
699 r = AV_RL16(src + xp + 0);
700 g = AV_RL16(src + xp + 1);
701 b = AV_RL16(src + xp + 2);
702 }
703
704 r = c->rgbgammainv[r>>4];
705 g = c->rgbgammainv[g>>4];
706 b = c->rgbgammainv[b>>4];
707
708 // convert from sRGBlinear to XYZlinear
709 x = c->rgb2xyz_matrix[0][0] * r +
710 c->rgb2xyz_matrix[0][1] * g +
711 c->rgb2xyz_matrix[0][2] * b >> 12;
712 y = c->rgb2xyz_matrix[1][0] * r +
713 c->rgb2xyz_matrix[1][1] * g +
714 c->rgb2xyz_matrix[1][2] * b >> 12;
715 z = c->rgb2xyz_matrix[2][0] * r +
716 c->rgb2xyz_matrix[2][1] * g +
717 c->rgb2xyz_matrix[2][2] * b >> 12;
718
719 // limit values to 12-bit depth
720 x = av_clip_uintp2(x, 12);
721 y = av_clip_uintp2(y, 12);
722 z = av_clip_uintp2(z, 12);
723
724 // convert from XYZlinear to X'Y'Z' and scale from 12bit to 16bit
725 if (desc->flags & AV_PIX_FMT_FLAG_BE) {
726 AV_WB16(dst + xp + 0, c->xyzgammainv[x] << 4);
727 AV_WB16(dst + xp + 1, c->xyzgammainv[y] << 4);
728 AV_WB16(dst + xp + 2, c->xyzgammainv[z] << 4);
729 } else {
730 AV_WL16(dst + xp + 0, c->xyzgammainv[x] << 4);
731 AV_WL16(dst + xp + 1, c->xyzgammainv[y] << 4);
732 AV_WL16(dst + xp + 2, c->xyzgammainv[z] << 4);
733 }
734 }
735 src += stride;
736 dst += stride;
737 }
738 }
739
740 /**
741 * swscale wrapper, so we don't need to export the SwsContext.
742 * Assumes planar YUV to be in YUV order instead of YVU.
743 */
sws_scale(struct SwsContext * c,const uint8_t * const srcSlice[],const int srcStride[],int srcSliceY,int srcSliceH,uint8_t * const dst[],const int dstStride[])744 int attribute_align_arg sws_scale(struct SwsContext *c,
745 const uint8_t * const srcSlice[],
746 const int srcStride[], int srcSliceY,
747 int srcSliceH, uint8_t *const dst[],
748 const int dstStride[])
749 {
750 int i, ret;
751 const uint8_t *src2[4];
752 uint8_t *dst2[4];
753 uint8_t *rgb0_tmp = NULL;
754 int macro_height = isBayer(c->srcFormat) ? 2 : (1 << c->chrSrcVSubSample);
755 // copy strides, so they can safely be modified
756 int srcStride2[4];
757 int dstStride2[4];
758 int srcSliceY_internal = srcSliceY;
759
760 if (!srcStride || !dstStride || !dst || !srcSlice) {
761 av_log(c, AV_LOG_ERROR, "One of the input parameters to sws_scale() is NULL, please check the calling code\n");
762 return 0;
763 }
764
765 for (i=0; i<4; i++) {
766 srcStride2[i] = srcStride[i];
767 dstStride2[i] = dstStride[i];
768 }
769
770 if ((srcSliceY & (macro_height-1)) ||
771 ((srcSliceH& (macro_height-1)) && srcSliceY + srcSliceH != c->srcH) ||
772 srcSliceY + srcSliceH > c->srcH) {
773 av_log(c, AV_LOG_ERROR, "Slice parameters %d, %d are invalid\n", srcSliceY, srcSliceH);
774 return AVERROR(EINVAL);
775 }
776
777 if (c->gamma_flag && c->cascaded_context[0]) {
778 ret = sws_scale(c->cascaded_context[0],
779 srcSlice, srcStride, srcSliceY, srcSliceH,
780 c->cascaded_tmp, c->cascaded_tmpStride);
781
782 if (ret < 0)
783 return ret;
784
785 if (c->cascaded_context[2])
786 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);
787 else
788 ret = sws_scale(c->cascaded_context[1], (const uint8_t * const *)c->cascaded_tmp, c->cascaded_tmpStride, srcSliceY, srcSliceH, dst, dstStride);
789
790 if (ret < 0)
791 return ret;
792
793 if (c->cascaded_context[2]) {
794 ret = sws_scale(c->cascaded_context[2],
795 (const uint8_t * const *)c->cascaded1_tmp, c->cascaded1_tmpStride, c->cascaded_context[1]->dstY - ret, c->cascaded_context[1]->dstY,
796 dst, dstStride);
797 }
798 return ret;
799 }
800
801 if (c->cascaded_context[0] && srcSliceY == 0 && srcSliceH == c->cascaded_context[0]->srcH) {
802 ret = sws_scale(c->cascaded_context[0],
803 srcSlice, srcStride, srcSliceY, srcSliceH,
804 c->cascaded_tmp, c->cascaded_tmpStride);
805 if (ret < 0)
806 return ret;
807 ret = sws_scale(c->cascaded_context[1],
808 (const uint8_t * const * )c->cascaded_tmp, c->cascaded_tmpStride, 0, c->cascaded_context[0]->dstH,
809 dst, dstStride);
810 return ret;
811 }
812
813 memcpy(src2, srcSlice, sizeof(src2));
814 memcpy(dst2, dst, sizeof(dst2));
815
816 // do not mess up sliceDir if we have a "trailing" 0-size slice
817 if (srcSliceH == 0)
818 return 0;
819
820 if (!check_image_pointers(srcSlice, c->srcFormat, srcStride)) {
821 av_log(c, AV_LOG_ERROR, "bad src image pointers\n");
822 return 0;
823 }
824 if (!check_image_pointers((const uint8_t* const*)dst, c->dstFormat, dstStride)) {
825 av_log(c, AV_LOG_ERROR, "bad dst image pointers\n");
826 return 0;
827 }
828
829 if (c->sliceDir == 0 && srcSliceY != 0 && srcSliceY + srcSliceH != c->srcH) {
830 av_log(c, AV_LOG_ERROR, "Slices start in the middle!\n");
831 return 0;
832 }
833 if (c->sliceDir == 0) {
834 if (srcSliceY == 0) c->sliceDir = 1; else c->sliceDir = -1;
835 }
836
837 if (usePal(c->srcFormat)) {
838 for (i = 0; i < 256; i++) {
839 int r, g, b, y, u, v, a = 0xff;
840 if (c->srcFormat == AV_PIX_FMT_PAL8) {
841 uint32_t p = ((const uint32_t *)(srcSlice[1]))[i];
842 a = (p >> 24) & 0xFF;
843 r = (p >> 16) & 0xFF;
844 g = (p >> 8) & 0xFF;
845 b = p & 0xFF;
846 } else if (c->srcFormat == AV_PIX_FMT_RGB8) {
847 r = ( i >> 5 ) * 36;
848 g = ((i >> 2) & 7) * 36;
849 b = ( i & 3) * 85;
850 } else if (c->srcFormat == AV_PIX_FMT_BGR8) {
851 b = ( i >> 6 ) * 85;
852 g = ((i >> 3) & 7) * 36;
853 r = ( i & 7) * 36;
854 } else if (c->srcFormat == AV_PIX_FMT_RGB4_BYTE) {
855 r = ( i >> 3 ) * 255;
856 g = ((i >> 1) & 3) * 85;
857 b = ( i & 1) * 255;
858 } else if (c->srcFormat == AV_PIX_FMT_GRAY8 || c->srcFormat == AV_PIX_FMT_GRAY8A) {
859 r = g = b = i;
860 } else {
861 av_assert1(c->srcFormat == AV_PIX_FMT_BGR4_BYTE);
862 b = ( i >> 3 ) * 255;
863 g = ((i >> 1) & 3) * 85;
864 r = ( i & 1) * 255;
865 }
866 #define RGB2YUV_SHIFT 15
867 #define BY ( (int) (0.114 * 219 / 255 * (1 << RGB2YUV_SHIFT) + 0.5))
868 #define BV (-(int) (0.081 * 224 / 255 * (1 << RGB2YUV_SHIFT) + 0.5))
869 #define BU ( (int) (0.500 * 224 / 255 * (1 << RGB2YUV_SHIFT) + 0.5))
870 #define GY ( (int) (0.587 * 219 / 255 * (1 << RGB2YUV_SHIFT) + 0.5))
871 #define GV (-(int) (0.419 * 224 / 255 * (1 << RGB2YUV_SHIFT) + 0.5))
872 #define GU (-(int) (0.331 * 224 / 255 * (1 << RGB2YUV_SHIFT) + 0.5))
873 #define RY ( (int) (0.299 * 219 / 255 * (1 << RGB2YUV_SHIFT) + 0.5))
874 #define RV ( (int) (0.500 * 224 / 255 * (1 << RGB2YUV_SHIFT) + 0.5))
875 #define RU (-(int) (0.169 * 224 / 255 * (1 << RGB2YUV_SHIFT) + 0.5))
876
877 y = av_clip_uint8((RY * r + GY * g + BY * b + ( 33 << (RGB2YUV_SHIFT - 1))) >> RGB2YUV_SHIFT);
878 u = av_clip_uint8((RU * r + GU * g + BU * b + (257 << (RGB2YUV_SHIFT - 1))) >> RGB2YUV_SHIFT);
879 v = av_clip_uint8((RV * r + GV * g + BV * b + (257 << (RGB2YUV_SHIFT - 1))) >> RGB2YUV_SHIFT);
880 c->pal_yuv[i]= y + (u<<8) + (v<<16) + ((unsigned)a<<24);
881
882 switch (c->dstFormat) {
883 case AV_PIX_FMT_BGR32:
884 #if !HAVE_BIGENDIAN
885 case AV_PIX_FMT_RGB24:
886 #endif
887 c->pal_rgb[i]= r + (g<<8) + (b<<16) + ((unsigned)a<<24);
888 break;
889 case AV_PIX_FMT_BGR32_1:
890 #if HAVE_BIGENDIAN
891 case AV_PIX_FMT_BGR24:
892 #endif
893 c->pal_rgb[i]= a + (r<<8) + (g<<16) + ((unsigned)b<<24);
894 break;
895 case AV_PIX_FMT_RGB32_1:
896 #if HAVE_BIGENDIAN
897 case AV_PIX_FMT_RGB24:
898 #endif
899 c->pal_rgb[i]= a + (b<<8) + (g<<16) + ((unsigned)r<<24);
900 break;
901 case AV_PIX_FMT_RGB32:
902 #if !HAVE_BIGENDIAN
903 case AV_PIX_FMT_BGR24:
904 #endif
905 default:
906 c->pal_rgb[i]= b + (g<<8) + (r<<16) + ((unsigned)a<<24);
907 }
908 }
909 }
910
911 if (c->src0Alpha && !c->dst0Alpha && isALPHA(c->dstFormat)) {
912 uint8_t *base;
913 int x,y;
914 rgb0_tmp = av_malloc(FFABS(srcStride[0]) * srcSliceH + 32);
915 if (!rgb0_tmp)
916 return AVERROR(ENOMEM);
917
918 base = srcStride[0] < 0 ? rgb0_tmp - srcStride[0] * (srcSliceH-1) : rgb0_tmp;
919 for (y=0; y<srcSliceH; y++){
920 memcpy(base + srcStride[0]*y, src2[0] + srcStride[0]*y, 4*c->srcW);
921 for (x=c->src0Alpha-1; x<4*c->srcW; x+=4) {
922 base[ srcStride[0]*y + x] = 0xFF;
923 }
924 }
925 src2[0] = base;
926 }
927
928 if (c->srcXYZ && !(c->dstXYZ && c->srcW==c->dstW && c->srcH==c->dstH)) {
929 uint8_t *base;
930 rgb0_tmp = av_malloc(FFABS(srcStride[0]) * srcSliceH + 32);
931 if (!rgb0_tmp)
932 return AVERROR(ENOMEM);
933
934 base = srcStride[0] < 0 ? rgb0_tmp - srcStride[0] * (srcSliceH-1) : rgb0_tmp;
935
936 xyz12Torgb48(c, (uint16_t*)base, (const uint16_t*)src2[0], srcStride[0]/2, srcSliceH);
937 src2[0] = base;
938 }
939
940 if (!srcSliceY && (c->flags & SWS_BITEXACT) && c->dither == SWS_DITHER_ED && c->dither_error[0])
941 for (i = 0; i < 4; i++)
942 memset(c->dither_error[i], 0, sizeof(c->dither_error[0][0]) * (c->dstW+2));
943
944 if (c->sliceDir != 1) {
945 // slices go from bottom to top => we flip the image internally
946 for (i=0; i<4; i++) {
947 srcStride2[i] *= -1;
948 dstStride2[i] *= -1;
949 }
950
951 src2[0] += (srcSliceH - 1) * srcStride[0];
952 if (!usePal(c->srcFormat))
953 src2[1] += ((srcSliceH >> c->chrSrcVSubSample) - 1) * srcStride[1];
954 src2[2] += ((srcSliceH >> c->chrSrcVSubSample) - 1) * srcStride[2];
955 src2[3] += (srcSliceH - 1) * srcStride[3];
956 dst2[0] += ( c->dstH - 1) * dstStride[0];
957 dst2[1] += ((c->dstH >> c->chrDstVSubSample) - 1) * dstStride[1];
958 dst2[2] += ((c->dstH >> c->chrDstVSubSample) - 1) * dstStride[2];
959 dst2[3] += ( c->dstH - 1) * dstStride[3];
960
961 srcSliceY_internal = c->srcH-srcSliceY-srcSliceH;
962 }
963 reset_ptr(src2, c->srcFormat);
964 reset_ptr((void*)dst2, c->dstFormat);
965
966 /* reset slice direction at end of frame */
967 if (srcSliceY_internal + srcSliceH == c->srcH)
968 c->sliceDir = 0;
969 ret = c->swscale(c, src2, srcStride2, srcSliceY_internal, srcSliceH, dst2, dstStride2);
970
971 if (c->dstXYZ && !(c->srcXYZ && c->srcW==c->dstW && c->srcH==c->dstH)) {
972 int dstY = c->dstY ? c->dstY : srcSliceY + srcSliceH;
973 uint16_t *dst16 = (uint16_t*)(dst2[0] + (dstY - ret) * dstStride2[0]);
974 av_assert0(dstY >= ret);
975 av_assert0(ret >= 0);
976 av_assert0(c->dstH >= dstY);
977
978 /* replace on the same data */
979 rgb48Toxyz12(c, dst16, dst16, dstStride2[0]/2, ret);
980 }
981
982 av_free(rgb0_tmp);
983 return ret;
984 }
985