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 <stdint.h>
22 #include <stdio.h>
23 #include <string.h>
24
25 #include "libavutil/avassert.h"
26 #include "libavutil/bswap.h"
27 #include "libavutil/common.h"
28 #include "libavutil/cpu.h"
29 #include "libavutil/intreadwrite.h"
30 #include "libavutil/mem_internal.h"
31 #include "libavutil/pixdesc.h"
32 #include "config.h"
33 #include "swscale_internal.h"
34 #include "swscale.h"
35
36 DECLARE_ALIGNED(8, const uint8_t, ff_dither_8x8_128)[9][8] = {
37 { 36, 68, 60, 92, 34, 66, 58, 90, },
38 { 100, 4, 124, 28, 98, 2, 122, 26, },
39 { 52, 84, 44, 76, 50, 82, 42, 74, },
40 { 116, 20, 108, 12, 114, 18, 106, 10, },
41 { 32, 64, 56, 88, 38, 70, 62, 94, },
42 { 96, 0, 120, 24, 102, 6, 126, 30, },
43 { 48, 80, 40, 72, 54, 86, 46, 78, },
44 { 112, 16, 104, 8, 118, 22, 110, 14, },
45 { 36, 68, 60, 92, 34, 66, 58, 90, },
46 };
47
48 DECLARE_ALIGNED(8, static const uint8_t, sws_pb_64)[8] = {
49 64, 64, 64, 64, 64, 64, 64, 64
50 };
51
fillPlane(uint8_t * plane,int stride,int width,int height,int y,uint8_t val)52 static av_always_inline void fillPlane(uint8_t *plane, int stride, int width,
53 int height, int y, uint8_t val)
54 {
55 int i;
56 uint8_t *ptr = plane + stride * y;
57 for (i = 0; i < height; i++) {
58 memset(ptr, val, width);
59 ptr += stride;
60 }
61 }
62
hScale16To19_c(SwsContext * c,int16_t * _dst,int dstW,const uint8_t * _src,const int16_t * filter,const int32_t * filterPos,int filterSize)63 static void hScale16To19_c(SwsContext *c, int16_t *_dst, int dstW,
64 const uint8_t *_src, const int16_t *filter,
65 const int32_t *filterPos, int filterSize)
66 {
67 const AVPixFmtDescriptor *desc = av_pix_fmt_desc_get(c->srcFormat);
68 int i;
69 int32_t *dst = (int32_t *) _dst;
70 const uint16_t *src = (const uint16_t *) _src;
71 int bits = desc->comp[0].depth - 1;
72 int sh = bits - 4;
73
74 if ((isAnyRGB(c->srcFormat) || c->srcFormat==AV_PIX_FMT_PAL8) && desc->comp[0].depth<16) {
75 sh = 9;
76 } else if (desc->flags & AV_PIX_FMT_FLAG_FLOAT) { /* float input are process like uint 16bpc */
77 sh = 16 - 1 - 4;
78 }
79
80 for (i = 0; i < dstW; i++) {
81 int j;
82 int srcPos = filterPos[i];
83 int val = 0;
84
85 for (j = 0; j < filterSize; j++) {
86 val += src[srcPos + j] * filter[filterSize * i + j];
87 }
88 // filter=14 bit, input=16 bit, output=30 bit, >> 11 makes 19 bit
89 dst[i] = FFMIN(val >> sh, (1 << 19) - 1);
90 }
91 }
92
hScale16To15_c(SwsContext * c,int16_t * dst,int dstW,const uint8_t * _src,const int16_t * filter,const int32_t * filterPos,int filterSize)93 static void hScale16To15_c(SwsContext *c, int16_t *dst, int dstW,
94 const uint8_t *_src, const int16_t *filter,
95 const int32_t *filterPos, int filterSize)
96 {
97 const AVPixFmtDescriptor *desc = av_pix_fmt_desc_get(c->srcFormat);
98 int i;
99 const uint16_t *src = (const uint16_t *) _src;
100 int sh = desc->comp[0].depth - 1;
101
102 if (sh<15) {
103 sh = isAnyRGB(c->srcFormat) || c->srcFormat==AV_PIX_FMT_PAL8 ? 13 : (desc->comp[0].depth - 1);
104 } else if (desc->flags & AV_PIX_FMT_FLAG_FLOAT) { /* float input are process like uint 16bpc */
105 sh = 16 - 1;
106 }
107
108 for (i = 0; i < dstW; i++) {
109 int j;
110 int srcPos = filterPos[i];
111 int val = 0;
112
113 for (j = 0; j < filterSize; j++) {
114 val += src[srcPos + j] * filter[filterSize * i + j];
115 }
116 // filter=14 bit, input=16 bit, output=30 bit, >> 15 makes 15 bit
117 dst[i] = FFMIN(val >> sh, (1 << 15) - 1);
118 }
119 }
120
121 // 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)122 static void hScale8To15_c(SwsContext *c, int16_t *dst, int dstW,
123 const uint8_t *src, const int16_t *filter,
124 const int32_t *filterPos, int filterSize)
125 {
126 int i;
127 for (i = 0; i < dstW; i++) {
128 int j;
129 int srcPos = filterPos[i];
130 int val = 0;
131 for (j = 0; j < filterSize; j++) {
132 val += ((int)src[srcPos + j]) * filter[filterSize * i + j];
133 }
134 dst[i] = FFMIN(val >> 7, (1 << 15) - 1); // the cubic equation does overflow ...
135 }
136 }
137
hScale8To19_c(SwsContext * c,int16_t * _dst,int dstW,const uint8_t * src,const int16_t * filter,const int32_t * filterPos,int filterSize)138 static void hScale8To19_c(SwsContext *c, int16_t *_dst, int dstW,
139 const uint8_t *src, const int16_t *filter,
140 const int32_t *filterPos, int filterSize)
141 {
142 int i;
143 int32_t *dst = (int32_t *) _dst;
144 for (i = 0; i < dstW; i++) {
145 int j;
146 int srcPos = filterPos[i];
147 int val = 0;
148 for (j = 0; j < filterSize; j++) {
149 val += ((int)src[srcPos + j]) * filter[filterSize * i + j];
150 }
151 dst[i] = FFMIN(val >> 3, (1 << 19) - 1); // the cubic equation does overflow ...
152 }
153 }
154
155 // FIXME all pal and rgb srcFormats could do this conversion as well
156 // FIXME all scalers more complex than bilinear could do half of this transform
chrRangeToJpeg_c(int16_t * dstU,int16_t * dstV,int width)157 static void chrRangeToJpeg_c(int16_t *dstU, int16_t *dstV, int width)
158 {
159 int i;
160 for (i = 0; i < width; i++) {
161 dstU[i] = (FFMIN(dstU[i], 30775) * 4663 - 9289992) >> 12; // -264
162 dstV[i] = (FFMIN(dstV[i], 30775) * 4663 - 9289992) >> 12; // -264
163 }
164 }
165
chrRangeFromJpeg_c(int16_t * dstU,int16_t * dstV,int width)166 static void chrRangeFromJpeg_c(int16_t *dstU, int16_t *dstV, int width)
167 {
168 int i;
169 for (i = 0; i < width; i++) {
170 dstU[i] = (dstU[i] * 1799 + 4081085) >> 11; // 1469
171 dstV[i] = (dstV[i] * 1799 + 4081085) >> 11; // 1469
172 }
173 }
174
lumRangeToJpeg_c(int16_t * dst,int width)175 static void lumRangeToJpeg_c(int16_t *dst, int width)
176 {
177 int i;
178 for (i = 0; i < width; i++)
179 dst[i] = (FFMIN(dst[i], 30189) * 19077 - 39057361) >> 14;
180 }
181
lumRangeFromJpeg_c(int16_t * dst,int width)182 static void lumRangeFromJpeg_c(int16_t *dst, int width)
183 {
184 int i;
185 for (i = 0; i < width; i++)
186 dst[i] = (dst[i] * 14071 + 33561947) >> 14;
187 }
188
chrRangeToJpeg16_c(int16_t * _dstU,int16_t * _dstV,int width)189 static void chrRangeToJpeg16_c(int16_t *_dstU, int16_t *_dstV, int width)
190 {
191 int i;
192 int32_t *dstU = (int32_t *) _dstU;
193 int32_t *dstV = (int32_t *) _dstV;
194 for (i = 0; i < width; i++) {
195 dstU[i] = (FFMIN(dstU[i], 30775 << 4) * 4663 - (9289992 << 4)) >> 12; // -264
196 dstV[i] = (FFMIN(dstV[i], 30775 << 4) * 4663 - (9289992 << 4)) >> 12; // -264
197 }
198 }
199
chrRangeFromJpeg16_c(int16_t * _dstU,int16_t * _dstV,int width)200 static void chrRangeFromJpeg16_c(int16_t *_dstU, int16_t *_dstV, int width)
201 {
202 int i;
203 int32_t *dstU = (int32_t *) _dstU;
204 int32_t *dstV = (int32_t *) _dstV;
205 for (i = 0; i < width; i++) {
206 dstU[i] = (dstU[i] * 1799 + (4081085 << 4)) >> 11; // 1469
207 dstV[i] = (dstV[i] * 1799 + (4081085 << 4)) >> 11; // 1469
208 }
209 }
210
lumRangeToJpeg16_c(int16_t * _dst,int width)211 static void lumRangeToJpeg16_c(int16_t *_dst, int width)
212 {
213 int i;
214 int32_t *dst = (int32_t *) _dst;
215 for (i = 0; i < width; i++) {
216 dst[i] = ((int)(FFMIN(dst[i], 30189 << 4) * 4769U - (39057361 << 2))) >> 12;
217 }
218 }
219
lumRangeFromJpeg16_c(int16_t * _dst,int width)220 static void lumRangeFromJpeg16_c(int16_t *_dst, int width)
221 {
222 int i;
223 int32_t *dst = (int32_t *) _dst;
224 for (i = 0; i < width; i++)
225 dst[i] = (dst[i]*(14071/4) + (33561947<<4)/4)>>12;
226 }
227
228
229 #define DEBUG_SWSCALE_BUFFERS 0
230 #define DEBUG_BUFFERS(...) \
231 if (DEBUG_SWSCALE_BUFFERS) \
232 av_log(c, AV_LOG_DEBUG, __VA_ARGS__)
233
swscale(SwsContext * c,const uint8_t * src[],int srcStride[],int srcSliceY,int srcSliceH,uint8_t * dst[],int dstStride[],int dstSliceY,int dstSliceH)234 static int swscale(SwsContext *c, const uint8_t *src[],
235 int srcStride[], int srcSliceY, int srcSliceH,
236 uint8_t *dst[], int dstStride[],
237 int dstSliceY, int dstSliceH)
238 {
239 const int scale_dst = dstSliceY > 0 || dstSliceH < c->dstH;
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 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 SwsContext *const ctx = c->parent ? c->parent : c;
312 if (flags & SWS_PRINT_INFO &&
313 !atomic_exchange_explicit(&ctx->stride_unaligned_warned, 1, memory_order_relaxed)) {
314 av_log(c, AV_LOG_WARNING,
315 "Warning: dstStride is not aligned!\n"
316 " ->cannot do aligned memory accesses anymore\n");
317 }
318 }
319
320 #if ARCH_X86
321 if ( (uintptr_t)dst[0]&15 || (uintptr_t)dst[1]&15 || (uintptr_t)dst[2]&15
322 || (uintptr_t)src[0]&15 || (uintptr_t)src[1]&15 || (uintptr_t)src[2]&15
323 || dstStride[0]&15 || dstStride[1]&15 || dstStride[2]&15 || dstStride[3]&15
324 || srcStride[0]&15 || srcStride[1]&15 || srcStride[2]&15 || srcStride[3]&15
325 ) {
326 SwsContext *const ctx = c->parent ? c->parent : c;
327 int cpu_flags = av_get_cpu_flags();
328 if (flags & SWS_PRINT_INFO && HAVE_MMXEXT && (cpu_flags & AV_CPU_FLAG_SSE2) &&
329 !atomic_exchange_explicit(&ctx->stride_unaligned_warned,1, memory_order_relaxed)) {
330 av_log(c, AV_LOG_WARNING, "Warning: data is not aligned! This can lead to a speed loss\n");
331 }
332 }
333 #endif
334
335 if (scale_dst) {
336 dstY = dstSliceY;
337 dstH = dstY + dstSliceH;
338 lastInLumBuf = -1;
339 lastInChrBuf = -1;
340 } else if (srcSliceY == 0) {
341 /* Note the user might start scaling the picture in the middle so this
342 * will not get executed. This is not really intended but works
343 * currently, so people might do it. */
344 dstY = 0;
345 lastInLumBuf = -1;
346 lastInChrBuf = -1;
347 }
348
349 if (!should_dither) {
350 c->chrDither8 = c->lumDither8 = sws_pb_64;
351 }
352 lastDstY = dstY;
353
354 ff_init_vscale_pfn(c, yuv2plane1, yuv2planeX, yuv2nv12cX,
355 yuv2packed1, yuv2packed2, yuv2packedX, yuv2anyX, c->use_mmx_vfilter);
356
357 ff_init_slice_from_src(src_slice, (uint8_t**)src, srcStride, c->srcW,
358 srcSliceY, srcSliceH, chrSrcSliceY, chrSrcSliceH, 1);
359
360 ff_init_slice_from_src(vout_slice, (uint8_t**)dst, dstStride, c->dstW,
361 dstY, dstSliceH, dstY >> c->chrDstVSubSample,
362 AV_CEIL_RSHIFT(dstSliceH, c->chrDstVSubSample), scale_dst);
363 if (srcSliceY == 0) {
364 hout_slice->plane[0].sliceY = lastInLumBuf + 1;
365 hout_slice->plane[1].sliceY = lastInChrBuf + 1;
366 hout_slice->plane[2].sliceY = lastInChrBuf + 1;
367 hout_slice->plane[3].sliceY = lastInLumBuf + 1;
368
369 hout_slice->plane[0].sliceH =
370 hout_slice->plane[1].sliceH =
371 hout_slice->plane[2].sliceH =
372 hout_slice->plane[3].sliceH = 0;
373 hout_slice->width = dstW;
374 }
375
376 for (; dstY < dstH; dstY++) {
377 const int chrDstY = dstY >> c->chrDstVSubSample;
378 int use_mmx_vfilter= c->use_mmx_vfilter;
379
380 // First line needed as input
381 const int firstLumSrcY = FFMAX(1 - vLumFilterSize, vLumFilterPos[dstY]);
382 const int firstLumSrcY2 = FFMAX(1 - vLumFilterSize, vLumFilterPos[FFMIN(dstY | ((1 << c->chrDstVSubSample) - 1), c->dstH - 1)]);
383 // First line needed as input
384 const int firstChrSrcY = FFMAX(1 - vChrFilterSize, vChrFilterPos[chrDstY]);
385
386 // Last line needed as input
387 int lastLumSrcY = FFMIN(c->srcH, firstLumSrcY + vLumFilterSize) - 1;
388 int lastLumSrcY2 = FFMIN(c->srcH, firstLumSrcY2 + vLumFilterSize) - 1;
389 int lastChrSrcY = FFMIN(c->chrSrcH, firstChrSrcY + vChrFilterSize) - 1;
390 int enough_lines;
391
392 int i;
393 int posY, cPosY, firstPosY, lastPosY, firstCPosY, lastCPosY;
394
395 // handle holes (FAST_BILINEAR & weird filters)
396 if (firstLumSrcY > lastInLumBuf) {
397
398 hasLumHoles = lastInLumBuf != firstLumSrcY - 1;
399 if (hasLumHoles) {
400 hout_slice->plane[0].sliceY = firstLumSrcY;
401 hout_slice->plane[3].sliceY = firstLumSrcY;
402 hout_slice->plane[0].sliceH =
403 hout_slice->plane[3].sliceH = 0;
404 }
405
406 lastInLumBuf = firstLumSrcY - 1;
407 }
408 if (firstChrSrcY > lastInChrBuf) {
409
410 hasChrHoles = lastInChrBuf != firstChrSrcY - 1;
411 if (hasChrHoles) {
412 hout_slice->plane[1].sliceY = firstChrSrcY;
413 hout_slice->plane[2].sliceY = firstChrSrcY;
414 hout_slice->plane[1].sliceH =
415 hout_slice->plane[2].sliceH = 0;
416 }
417
418 lastInChrBuf = firstChrSrcY - 1;
419 }
420
421 DEBUG_BUFFERS("dstY: %d\n", dstY);
422 DEBUG_BUFFERS("\tfirstLumSrcY: %d lastLumSrcY: %d lastInLumBuf: %d\n",
423 firstLumSrcY, lastLumSrcY, lastInLumBuf);
424 DEBUG_BUFFERS("\tfirstChrSrcY: %d lastChrSrcY: %d lastInChrBuf: %d\n",
425 firstChrSrcY, lastChrSrcY, lastInChrBuf);
426
427 // Do we have enough lines in this slice to output the dstY line
428 enough_lines = lastLumSrcY2 < srcSliceY + srcSliceH &&
429 lastChrSrcY < AV_CEIL_RSHIFT(srcSliceY + srcSliceH, c->chrSrcVSubSample);
430
431 if (!enough_lines) {
432 lastLumSrcY = srcSliceY + srcSliceH - 1;
433 lastChrSrcY = chrSrcSliceY + chrSrcSliceH - 1;
434 DEBUG_BUFFERS("buffering slice: lastLumSrcY %d lastChrSrcY %d\n",
435 lastLumSrcY, lastChrSrcY);
436 }
437
438 av_assert0((lastLumSrcY - firstLumSrcY + 1) <= hout_slice->plane[0].available_lines);
439 av_assert0((lastChrSrcY - firstChrSrcY + 1) <= hout_slice->plane[1].available_lines);
440
441
442 posY = hout_slice->plane[0].sliceY + hout_slice->plane[0].sliceH;
443 if (posY <= lastLumSrcY && !hasLumHoles) {
444 firstPosY = FFMAX(firstLumSrcY, posY);
445 lastPosY = FFMIN(firstLumSrcY + hout_slice->plane[0].available_lines - 1, srcSliceY + srcSliceH - 1);
446 } else {
447 firstPosY = posY;
448 lastPosY = lastLumSrcY;
449 }
450
451 cPosY = hout_slice->plane[1].sliceY + hout_slice->plane[1].sliceH;
452 if (cPosY <= lastChrSrcY && !hasChrHoles) {
453 firstCPosY = FFMAX(firstChrSrcY, cPosY);
454 lastCPosY = FFMIN(firstChrSrcY + hout_slice->plane[1].available_lines - 1, AV_CEIL_RSHIFT(srcSliceY + srcSliceH, c->chrSrcVSubSample) - 1);
455 } else {
456 firstCPosY = cPosY;
457 lastCPosY = lastChrSrcY;
458 }
459
460 ff_rotate_slice(hout_slice, lastPosY, lastCPosY);
461
462 if (posY < lastLumSrcY + 1) {
463 for (i = lumStart; i < lumEnd; ++i)
464 desc[i].process(c, &desc[i], firstPosY, lastPosY - firstPosY + 1);
465 }
466
467 lastInLumBuf = lastLumSrcY;
468
469 if (cPosY < lastChrSrcY + 1) {
470 for (i = chrStart; i < chrEnd; ++i)
471 desc[i].process(c, &desc[i], firstCPosY, lastCPosY - firstCPosY + 1);
472 }
473
474 lastInChrBuf = lastChrSrcY;
475
476 if (!enough_lines)
477 break; // we can't output a dstY line so let's try with the next slice
478
479 #if HAVE_MMX_INLINE
480 ff_updateMMXDitherTables(c, dstY);
481 #endif
482 if (should_dither) {
483 c->chrDither8 = ff_dither_8x8_128[chrDstY & 7];
484 c->lumDither8 = ff_dither_8x8_128[dstY & 7];
485 }
486 if (dstY >= c->dstH - 2) {
487 /* hmm looks like we can't use MMX here without overwriting
488 * this array's tail */
489 ff_sws_init_output_funcs(c, &yuv2plane1, &yuv2planeX, &yuv2nv12cX,
490 &yuv2packed1, &yuv2packed2, &yuv2packedX, &yuv2anyX);
491 use_mmx_vfilter= 0;
492 ff_init_vscale_pfn(c, yuv2plane1, yuv2planeX, yuv2nv12cX,
493 yuv2packed1, yuv2packed2, yuv2packedX, yuv2anyX, use_mmx_vfilter);
494 }
495
496 for (i = vStart; i < vEnd; ++i)
497 desc[i].process(c, &desc[i], dstY, 1);
498 }
499 if (isPlanar(dstFormat) && isALPHA(dstFormat) && !needAlpha) {
500 int offset = lastDstY - dstSliceY;
501 int length = dstW;
502 int height = dstY - lastDstY;
503
504 if (is16BPS(dstFormat) || isNBPS(dstFormat)) {
505 const AVPixFmtDescriptor *desc = av_pix_fmt_desc_get(dstFormat);
506 fillPlane16(dst[3], dstStride[3], length, height, offset,
507 1, desc->comp[3].depth,
508 isBE(dstFormat));
509 } else if (is32BPS(dstFormat)) {
510 const AVPixFmtDescriptor *desc = av_pix_fmt_desc_get(dstFormat);
511 fillPlane32(dst[3], dstStride[3], length, height, offset,
512 1, desc->comp[3].depth,
513 isBE(dstFormat), desc->flags & AV_PIX_FMT_FLAG_FLOAT);
514 } else
515 fillPlane(dst[3], dstStride[3], length, height, offset, 255);
516 }
517
518 #if HAVE_MMXEXT_INLINE
519 if (av_get_cpu_flags() & AV_CPU_FLAG_MMXEXT)
520 __asm__ volatile ("sfence" ::: "memory");
521 #endif
522 emms_c();
523
524 /* store changed local vars back in the context */
525 c->dstY = dstY;
526 c->lastInLumBuf = lastInLumBuf;
527 c->lastInChrBuf = lastInChrBuf;
528
529 return dstY - lastDstY;
530 }
531
ff_sws_init_range_convert(SwsContext * c)532 av_cold void ff_sws_init_range_convert(SwsContext *c)
533 {
534 c->lumConvertRange = NULL;
535 c->chrConvertRange = NULL;
536 if (c->srcRange != c->dstRange && !isAnyRGB(c->dstFormat)) {
537 if (c->dstBpc <= 14) {
538 if (c->srcRange) {
539 c->lumConvertRange = lumRangeFromJpeg_c;
540 c->chrConvertRange = chrRangeFromJpeg_c;
541 } else {
542 c->lumConvertRange = lumRangeToJpeg_c;
543 c->chrConvertRange = chrRangeToJpeg_c;
544 }
545 } else {
546 if (c->srcRange) {
547 c->lumConvertRange = lumRangeFromJpeg16_c;
548 c->chrConvertRange = chrRangeFromJpeg16_c;
549 } else {
550 c->lumConvertRange = lumRangeToJpeg16_c;
551 c->chrConvertRange = chrRangeToJpeg16_c;
552 }
553 }
554 }
555 }
556
sws_init_swscale(SwsContext * c)557 static av_cold void sws_init_swscale(SwsContext *c)
558 {
559 enum AVPixelFormat srcFormat = c->srcFormat;
560
561 ff_sws_init_output_funcs(c, &c->yuv2plane1, &c->yuv2planeX,
562 &c->yuv2nv12cX, &c->yuv2packed1,
563 &c->yuv2packed2, &c->yuv2packedX, &c->yuv2anyX);
564
565 ff_sws_init_input_funcs(c);
566
567 if (c->srcBpc == 8) {
568 if (c->dstBpc <= 14) {
569 c->hyScale = c->hcScale = hScale8To15_c;
570 if (c->flags & SWS_FAST_BILINEAR) {
571 c->hyscale_fast = ff_hyscale_fast_c;
572 c->hcscale_fast = ff_hcscale_fast_c;
573 }
574 } else {
575 c->hyScale = c->hcScale = hScale8To19_c;
576 }
577 } else {
578 c->hyScale = c->hcScale = c->dstBpc > 14 ? hScale16To19_c
579 : hScale16To15_c;
580 }
581
582 ff_sws_init_range_convert(c);
583
584 if (!(isGray(srcFormat) || isGray(c->dstFormat) ||
585 srcFormat == AV_PIX_FMT_MONOBLACK || srcFormat == AV_PIX_FMT_MONOWHITE))
586 c->needs_hcscale = 1;
587 }
588
ff_sws_init_scale(SwsContext * c)589 void ff_sws_init_scale(SwsContext *c)
590 {
591 sws_init_swscale(c);
592
593 #if ARCH_PPC
594 ff_sws_init_swscale_ppc(c);
595 #elif ARCH_X86
596 ff_sws_init_swscale_x86(c);
597 #elif ARCH_AARCH64
598 ff_sws_init_swscale_aarch64(c);
599 #elif ARCH_ARM
600 ff_sws_init_swscale_arm(c);
601 #endif
602 }
603
reset_ptr(const uint8_t * src[],enum AVPixelFormat format)604 static void reset_ptr(const uint8_t *src[], enum AVPixelFormat format)
605 {
606 if (!isALPHA(format))
607 src[3] = NULL;
608 if (!isPlanar(format)) {
609 src[3] = src[2] = NULL;
610
611 if (!usePal(format))
612 src[1] = NULL;
613 }
614 }
615
check_image_pointers(const uint8_t * const data[4],enum AVPixelFormat pix_fmt,const int linesizes[4])616 static int check_image_pointers(const uint8_t * const data[4], enum AVPixelFormat pix_fmt,
617 const int linesizes[4])
618 {
619 const AVPixFmtDescriptor *desc = av_pix_fmt_desc_get(pix_fmt);
620 int i;
621
622 av_assert2(desc);
623
624 for (i = 0; i < 4; i++) {
625 int plane = desc->comp[i].plane;
626 if (!data[plane] || !linesizes[plane])
627 return 0;
628 }
629
630 return 1;
631 }
632
xyz12Torgb48(struct SwsContext * c,uint16_t * dst,const uint16_t * src,int stride,int h)633 static void xyz12Torgb48(struct SwsContext *c, uint16_t *dst,
634 const uint16_t *src, int stride, int h)
635 {
636 int xp,yp;
637 const AVPixFmtDescriptor *desc = av_pix_fmt_desc_get(c->srcFormat);
638
639 for (yp=0; yp<h; yp++) {
640 for (xp=0; xp+2<stride; xp+=3) {
641 int x, y, z, r, g, b;
642
643 if (desc->flags & AV_PIX_FMT_FLAG_BE) {
644 x = AV_RB16(src + xp + 0);
645 y = AV_RB16(src + xp + 1);
646 z = AV_RB16(src + xp + 2);
647 } else {
648 x = AV_RL16(src + xp + 0);
649 y = AV_RL16(src + xp + 1);
650 z = AV_RL16(src + xp + 2);
651 }
652
653 x = c->xyzgamma[x>>4];
654 y = c->xyzgamma[y>>4];
655 z = c->xyzgamma[z>>4];
656
657 // convert from XYZlinear to sRGBlinear
658 r = c->xyz2rgb_matrix[0][0] * x +
659 c->xyz2rgb_matrix[0][1] * y +
660 c->xyz2rgb_matrix[0][2] * z >> 12;
661 g = c->xyz2rgb_matrix[1][0] * x +
662 c->xyz2rgb_matrix[1][1] * y +
663 c->xyz2rgb_matrix[1][2] * z >> 12;
664 b = c->xyz2rgb_matrix[2][0] * x +
665 c->xyz2rgb_matrix[2][1] * y +
666 c->xyz2rgb_matrix[2][2] * z >> 12;
667
668 // limit values to 12-bit depth
669 r = av_clip_uintp2(r, 12);
670 g = av_clip_uintp2(g, 12);
671 b = av_clip_uintp2(b, 12);
672
673 // convert from sRGBlinear to RGB and scale from 12bit to 16bit
674 if (desc->flags & AV_PIX_FMT_FLAG_BE) {
675 AV_WB16(dst + xp + 0, c->rgbgamma[r] << 4);
676 AV_WB16(dst + xp + 1, c->rgbgamma[g] << 4);
677 AV_WB16(dst + xp + 2, c->rgbgamma[b] << 4);
678 } else {
679 AV_WL16(dst + xp + 0, c->rgbgamma[r] << 4);
680 AV_WL16(dst + xp + 1, c->rgbgamma[g] << 4);
681 AV_WL16(dst + xp + 2, c->rgbgamma[b] << 4);
682 }
683 }
684 src += stride;
685 dst += stride;
686 }
687 }
688
rgb48Toxyz12(struct SwsContext * c,uint16_t * dst,const uint16_t * src,int stride,int h)689 static void rgb48Toxyz12(struct SwsContext *c, uint16_t *dst,
690 const uint16_t *src, int stride, int h)
691 {
692 int xp,yp;
693 const AVPixFmtDescriptor *desc = av_pix_fmt_desc_get(c->dstFormat);
694
695 for (yp=0; yp<h; yp++) {
696 for (xp=0; xp+2<stride; xp+=3) {
697 int x, y, z, r, g, b;
698
699 if (desc->flags & AV_PIX_FMT_FLAG_BE) {
700 r = AV_RB16(src + xp + 0);
701 g = AV_RB16(src + xp + 1);
702 b = AV_RB16(src + xp + 2);
703 } else {
704 r = AV_RL16(src + xp + 0);
705 g = AV_RL16(src + xp + 1);
706 b = AV_RL16(src + xp + 2);
707 }
708
709 r = c->rgbgammainv[r>>4];
710 g = c->rgbgammainv[g>>4];
711 b = c->rgbgammainv[b>>4];
712
713 // convert from sRGBlinear to XYZlinear
714 x = c->rgb2xyz_matrix[0][0] * r +
715 c->rgb2xyz_matrix[0][1] * g +
716 c->rgb2xyz_matrix[0][2] * b >> 12;
717 y = c->rgb2xyz_matrix[1][0] * r +
718 c->rgb2xyz_matrix[1][1] * g +
719 c->rgb2xyz_matrix[1][2] * b >> 12;
720 z = c->rgb2xyz_matrix[2][0] * r +
721 c->rgb2xyz_matrix[2][1] * g +
722 c->rgb2xyz_matrix[2][2] * b >> 12;
723
724 // limit values to 12-bit depth
725 x = av_clip_uintp2(x, 12);
726 y = av_clip_uintp2(y, 12);
727 z = av_clip_uintp2(z, 12);
728
729 // convert from XYZlinear to X'Y'Z' and scale from 12bit to 16bit
730 if (desc->flags & AV_PIX_FMT_FLAG_BE) {
731 AV_WB16(dst + xp + 0, c->xyzgammainv[x] << 4);
732 AV_WB16(dst + xp + 1, c->xyzgammainv[y] << 4);
733 AV_WB16(dst + xp + 2, c->xyzgammainv[z] << 4);
734 } else {
735 AV_WL16(dst + xp + 0, c->xyzgammainv[x] << 4);
736 AV_WL16(dst + xp + 1, c->xyzgammainv[y] << 4);
737 AV_WL16(dst + xp + 2, c->xyzgammainv[z] << 4);
738 }
739 }
740 src += stride;
741 dst += stride;
742 }
743 }
744
update_palette(SwsContext * c,const uint32_t * pal)745 static void update_palette(SwsContext *c, const uint32_t *pal)
746 {
747 for (int i = 0; i < 256; i++) {
748 int r, g, b, y, u, v, a = 0xff;
749 if (c->srcFormat == AV_PIX_FMT_PAL8) {
750 uint32_t p = pal[i];
751 a = (p >> 24) & 0xFF;
752 r = (p >> 16) & 0xFF;
753 g = (p >> 8) & 0xFF;
754 b = p & 0xFF;
755 } else if (c->srcFormat == AV_PIX_FMT_RGB8) {
756 r = ( i >> 5 ) * 36;
757 g = ((i >> 2) & 7) * 36;
758 b = ( i & 3) * 85;
759 } else if (c->srcFormat == AV_PIX_FMT_BGR8) {
760 b = ( i >> 6 ) * 85;
761 g = ((i >> 3) & 7) * 36;
762 r = ( i & 7) * 36;
763 } else if (c->srcFormat == AV_PIX_FMT_RGB4_BYTE) {
764 r = ( i >> 3 ) * 255;
765 g = ((i >> 1) & 3) * 85;
766 b = ( i & 1) * 255;
767 } else if (c->srcFormat == AV_PIX_FMT_GRAY8 || c->srcFormat == AV_PIX_FMT_GRAY8A) {
768 r = g = b = i;
769 } else {
770 av_assert1(c->srcFormat == AV_PIX_FMT_BGR4_BYTE);
771 b = ( i >> 3 ) * 255;
772 g = ((i >> 1) & 3) * 85;
773 r = ( i & 1) * 255;
774 }
775 #define RGB2YUV_SHIFT 15
776 #define BY ( (int) (0.114 * 219 / 255 * (1 << RGB2YUV_SHIFT) + 0.5))
777 #define BV (-(int) (0.081 * 224 / 255 * (1 << RGB2YUV_SHIFT) + 0.5))
778 #define BU ( (int) (0.500 * 224 / 255 * (1 << RGB2YUV_SHIFT) + 0.5))
779 #define GY ( (int) (0.587 * 219 / 255 * (1 << RGB2YUV_SHIFT) + 0.5))
780 #define GV (-(int) (0.419 * 224 / 255 * (1 << RGB2YUV_SHIFT) + 0.5))
781 #define GU (-(int) (0.331 * 224 / 255 * (1 << RGB2YUV_SHIFT) + 0.5))
782 #define RY ( (int) (0.299 * 219 / 255 * (1 << RGB2YUV_SHIFT) + 0.5))
783 #define RV ( (int) (0.500 * 224 / 255 * (1 << RGB2YUV_SHIFT) + 0.5))
784 #define RU (-(int) (0.169 * 224 / 255 * (1 << RGB2YUV_SHIFT) + 0.5))
785
786 y = av_clip_uint8((RY * r + GY * g + BY * b + ( 33 << (RGB2YUV_SHIFT - 1))) >> RGB2YUV_SHIFT);
787 u = av_clip_uint8((RU * r + GU * g + BU * b + (257 << (RGB2YUV_SHIFT - 1))) >> RGB2YUV_SHIFT);
788 v = av_clip_uint8((RV * r + GV * g + BV * b + (257 << (RGB2YUV_SHIFT - 1))) >> RGB2YUV_SHIFT);
789 c->pal_yuv[i]= y + (u<<8) + (v<<16) + ((unsigned)a<<24);
790
791 switch (c->dstFormat) {
792 case AV_PIX_FMT_BGR32:
793 #if !HAVE_BIGENDIAN
794 case AV_PIX_FMT_RGB24:
795 #endif
796 c->pal_rgb[i]= r + (g<<8) + (b<<16) + ((unsigned)a<<24);
797 break;
798 case AV_PIX_FMT_BGR32_1:
799 #if HAVE_BIGENDIAN
800 case AV_PIX_FMT_BGR24:
801 #endif
802 c->pal_rgb[i]= a + (r<<8) + (g<<16) + ((unsigned)b<<24);
803 break;
804 case AV_PIX_FMT_RGB32_1:
805 #if HAVE_BIGENDIAN
806 case AV_PIX_FMT_RGB24:
807 #endif
808 c->pal_rgb[i]= a + (b<<8) + (g<<16) + ((unsigned)r<<24);
809 break;
810 case AV_PIX_FMT_RGB32:
811 #if !HAVE_BIGENDIAN
812 case AV_PIX_FMT_BGR24:
813 #endif
814 default:
815 c->pal_rgb[i]= b + (g<<8) + (r<<16) + ((unsigned)a<<24);
816 }
817 }
818 }
819
820 static int scale_internal(SwsContext *c,
821 const uint8_t * const srcSlice[], const int srcStride[],
822 int srcSliceY, int srcSliceH,
823 uint8_t *const dstSlice[], const int dstStride[],
824 int dstSliceY, int dstSliceH);
825
scale_gamma(SwsContext * c,const uint8_t * const srcSlice[],const int srcStride[],int srcSliceY,int srcSliceH,uint8_t * const dstSlice[],const int dstStride[],int dstSliceY,int dstSliceH)826 static int scale_gamma(SwsContext *c,
827 const uint8_t * const srcSlice[], const int srcStride[],
828 int srcSliceY, int srcSliceH,
829 uint8_t * const dstSlice[], const int dstStride[],
830 int dstSliceY, int dstSliceH)
831 {
832 int ret = scale_internal(c->cascaded_context[0],
833 srcSlice, srcStride, srcSliceY, srcSliceH,
834 c->cascaded_tmp, c->cascaded_tmpStride, 0, c->srcH);
835
836 if (ret < 0)
837 return ret;
838
839 if (c->cascaded_context[2])
840 ret = scale_internal(c->cascaded_context[1], (const uint8_t * const *)c->cascaded_tmp,
841 c->cascaded_tmpStride, srcSliceY, srcSliceH,
842 c->cascaded1_tmp, c->cascaded1_tmpStride, 0, c->dstH);
843 else
844 ret = scale_internal(c->cascaded_context[1], (const uint8_t * const *)c->cascaded_tmp,
845 c->cascaded_tmpStride, srcSliceY, srcSliceH,
846 dstSlice, dstStride, dstSliceY, dstSliceH);
847
848 if (ret < 0)
849 return ret;
850
851 if (c->cascaded_context[2]) {
852 ret = scale_internal(c->cascaded_context[2], (const uint8_t * const *)c->cascaded1_tmp,
853 c->cascaded1_tmpStride, c->cascaded_context[1]->dstY - ret,
854 c->cascaded_context[1]->dstY,
855 dstSlice, dstStride, dstSliceY, dstSliceH);
856 }
857 return ret;
858 }
859
scale_cascaded(SwsContext * c,const uint8_t * const srcSlice[],const int srcStride[],int srcSliceY,int srcSliceH,uint8_t * const dstSlice[],const int dstStride[],int dstSliceY,int dstSliceH)860 static int scale_cascaded(SwsContext *c,
861 const uint8_t * const srcSlice[], const int srcStride[],
862 int srcSliceY, int srcSliceH,
863 uint8_t * const dstSlice[], const int dstStride[],
864 int dstSliceY, int dstSliceH)
865 {
866 int ret = scale_internal(c->cascaded_context[0],
867 srcSlice, srcStride, srcSliceY, srcSliceH,
868 c->cascaded_tmp, c->cascaded_tmpStride,
869 0, c->cascaded_context[0]->dstH);
870 if (ret < 0)
871 return ret;
872 ret = scale_internal(c->cascaded_context[1],
873 (const uint8_t * const * )c->cascaded_tmp, c->cascaded_tmpStride,
874 0, c->cascaded_context[0]->dstH,
875 dstSlice, dstStride, dstSliceY, dstSliceH);
876 return ret;
877 }
878
scale_internal(SwsContext * c,const uint8_t * const srcSlice[],const int srcStride[],int srcSliceY,int srcSliceH,uint8_t * const dstSlice[],const int dstStride[],int dstSliceY,int dstSliceH)879 static int scale_internal(SwsContext *c,
880 const uint8_t * const srcSlice[], const int srcStride[],
881 int srcSliceY, int srcSliceH,
882 uint8_t *const dstSlice[], const int dstStride[],
883 int dstSliceY, int dstSliceH)
884 {
885 const int scale_dst = dstSliceY > 0 || dstSliceH < c->dstH;
886 const int frame_start = scale_dst || !c->sliceDir;
887 int i, ret;
888 const uint8_t *src2[4];
889 uint8_t *dst2[4];
890 int macro_height_src = isBayer(c->srcFormat) ? 2 : (1 << c->chrSrcVSubSample);
891 int macro_height_dst = isBayer(c->dstFormat) ? 2 : (1 << c->chrDstVSubSample);
892 // copy strides, so they can safely be modified
893 int srcStride2[4];
894 int dstStride2[4];
895 int srcSliceY_internal = srcSliceY;
896
897 if (!srcStride || !dstStride || !dstSlice || !srcSlice) {
898 av_log(c, AV_LOG_ERROR, "One of the input parameters to sws_scale() is NULL, please check the calling code\n");
899 return AVERROR(EINVAL);
900 }
901
902 if ((srcSliceY & (macro_height_src - 1)) ||
903 ((srcSliceH & (macro_height_src - 1)) && srcSliceY + srcSliceH != c->srcH) ||
904 srcSliceY + srcSliceH > c->srcH) {
905 av_log(c, AV_LOG_ERROR, "Slice parameters %d, %d are invalid\n", srcSliceY, srcSliceH);
906 return AVERROR(EINVAL);
907 }
908
909 if ((dstSliceY & (macro_height_dst - 1)) ||
910 ((dstSliceH & (macro_height_dst - 1)) && dstSliceY + dstSliceH != c->dstH) ||
911 dstSliceY + dstSliceH > c->dstH) {
912 av_log(c, AV_LOG_ERROR, "Slice parameters %d, %d are invalid\n", dstSliceY, dstSliceH);
913 return AVERROR(EINVAL);
914 }
915
916 if (!check_image_pointers(srcSlice, c->srcFormat, srcStride)) {
917 av_log(c, AV_LOG_ERROR, "bad src image pointers\n");
918 return AVERROR(EINVAL);
919 }
920 if (!check_image_pointers((const uint8_t* const*)dstSlice, c->dstFormat, dstStride)) {
921 av_log(c, AV_LOG_ERROR, "bad dst image pointers\n");
922 return AVERROR(EINVAL);
923 }
924
925 // do not mess up sliceDir if we have a "trailing" 0-size slice
926 if (srcSliceH == 0)
927 return 0;
928
929 if (c->gamma_flag && c->cascaded_context[0])
930 return scale_gamma(c, srcSlice, srcStride, srcSliceY, srcSliceH,
931 dstSlice, dstStride, dstSliceY, dstSliceH);
932
933 if (c->cascaded_context[0] && srcSliceY == 0 && srcSliceH == c->cascaded_context[0]->srcH)
934 return scale_cascaded(c, srcSlice, srcStride, srcSliceY, srcSliceH,
935 dstSlice, dstStride, dstSliceY, dstSliceH);
936
937 if (!srcSliceY && (c->flags & SWS_BITEXACT) && c->dither == SWS_DITHER_ED && c->dither_error[0])
938 for (i = 0; i < 4; i++)
939 memset(c->dither_error[i], 0, sizeof(c->dither_error[0][0]) * (c->dstW+2));
940
941 if (usePal(c->srcFormat))
942 update_palette(c, (const uint32_t *)srcSlice[1]);
943
944 memcpy(src2, srcSlice, sizeof(src2));
945 memcpy(dst2, dstSlice, sizeof(dst2));
946 memcpy(srcStride2, srcStride, sizeof(srcStride2));
947 memcpy(dstStride2, dstStride, sizeof(dstStride2));
948
949 if (frame_start && !scale_dst) {
950 if (srcSliceY != 0 && srcSliceY + srcSliceH != c->srcH) {
951 av_log(c, AV_LOG_ERROR, "Slices start in the middle!\n");
952 return AVERROR(EINVAL);
953 }
954
955 c->sliceDir = (srcSliceY == 0) ? 1 : -1;
956 } else if (scale_dst)
957 c->sliceDir = 1;
958
959 if (c->src0Alpha && !c->dst0Alpha && isALPHA(c->dstFormat)) {
960 uint8_t *base;
961 int x,y;
962
963 av_fast_malloc(&c->rgb0_scratch, &c->rgb0_scratch_allocated,
964 FFABS(srcStride[0]) * srcSliceH + 32);
965 if (!c->rgb0_scratch)
966 return AVERROR(ENOMEM);
967
968 base = srcStride[0] < 0 ? c->rgb0_scratch - srcStride[0] * (srcSliceH-1) :
969 c->rgb0_scratch;
970 for (y=0; y<srcSliceH; y++){
971 memcpy(base + srcStride[0]*y, src2[0] + srcStride[0]*y, 4*c->srcW);
972 for (x=c->src0Alpha-1; x<4*c->srcW; x+=4) {
973 base[ srcStride[0]*y + x] = 0xFF;
974 }
975 }
976 src2[0] = base;
977 }
978
979 if (c->srcXYZ && !(c->dstXYZ && c->srcW==c->dstW && c->srcH==c->dstH)) {
980 uint8_t *base;
981
982 av_fast_malloc(&c->xyz_scratch, &c->xyz_scratch_allocated,
983 FFABS(srcStride[0]) * srcSliceH + 32);
984 if (!c->xyz_scratch)
985 return AVERROR(ENOMEM);
986
987 base = srcStride[0] < 0 ? c->xyz_scratch - srcStride[0] * (srcSliceH-1) :
988 c->xyz_scratch;
989
990 xyz12Torgb48(c, (uint16_t*)base, (const uint16_t*)src2[0], srcStride[0]/2, srcSliceH);
991 src2[0] = base;
992 }
993
994 if (c->sliceDir != 1) {
995 // slices go from bottom to top => we flip the image internally
996 for (i=0; i<4; i++) {
997 srcStride2[i] *= -1;
998 dstStride2[i] *= -1;
999 }
1000
1001 src2[0] += (srcSliceH - 1) * srcStride[0];
1002 if (!usePal(c->srcFormat))
1003 src2[1] += ((srcSliceH >> c->chrSrcVSubSample) - 1) * srcStride[1];
1004 src2[2] += ((srcSliceH >> c->chrSrcVSubSample) - 1) * srcStride[2];
1005 src2[3] += (srcSliceH - 1) * srcStride[3];
1006 dst2[0] += ( c->dstH - 1) * dstStride[0];
1007 dst2[1] += ((c->dstH >> c->chrDstVSubSample) - 1) * dstStride[1];
1008 dst2[2] += ((c->dstH >> c->chrDstVSubSample) - 1) * dstStride[2];
1009 dst2[3] += ( c->dstH - 1) * dstStride[3];
1010
1011 srcSliceY_internal = c->srcH-srcSliceY-srcSliceH;
1012 }
1013 reset_ptr(src2, c->srcFormat);
1014 reset_ptr((void*)dst2, c->dstFormat);
1015
1016 if (c->convert_unscaled) {
1017 int offset = srcSliceY_internal;
1018 int slice_h = srcSliceH;
1019
1020 // for dst slice scaling, offset the pointers to match the unscaled API
1021 if (scale_dst) {
1022 av_assert0(offset == 0);
1023 for (i = 0; i < 4 && src2[i]; i++) {
1024 if (!src2[i] || (i > 0 && usePal(c->srcFormat)))
1025 break;
1026 src2[i] += (dstSliceY >> ((i == 1 || i == 2) ? c->chrSrcVSubSample : 0)) * srcStride2[i];
1027 }
1028
1029 for (i = 0; i < 4 && dst2[i]; i++) {
1030 if (!dst2[i] || (i > 0 && usePal(c->dstFormat)))
1031 break;
1032 dst2[i] -= (dstSliceY >> ((i == 1 || i == 2) ? c->chrDstVSubSample : 0)) * dstStride2[i];
1033 }
1034 offset = dstSliceY;
1035 slice_h = dstSliceH;
1036 }
1037
1038 ret = c->convert_unscaled(c, src2, srcStride2, offset, slice_h,
1039 dst2, dstStride2);
1040 if (scale_dst)
1041 dst2[0] += dstSliceY * dstStride2[0];
1042 } else {
1043 ret = swscale(c, src2, srcStride2, srcSliceY_internal, srcSliceH,
1044 dst2, dstStride2, dstSliceY, dstSliceH);
1045 }
1046
1047 if (c->dstXYZ && !(c->srcXYZ && c->srcW==c->dstW && c->srcH==c->dstH)) {
1048 uint16_t *dst16;
1049
1050 if (scale_dst) {
1051 dst16 = (uint16_t *)dst2[0];
1052 } else {
1053 int dstY = c->dstY ? c->dstY : srcSliceY + srcSliceH;
1054
1055 av_assert0(dstY >= ret);
1056 av_assert0(ret >= 0);
1057 av_assert0(c->dstH >= dstY);
1058 dst16 = (uint16_t*)(dst2[0] + (dstY - ret) * dstStride2[0]);
1059 }
1060
1061 /* replace on the same data */
1062 rgb48Toxyz12(c, dst16, dst16, dstStride2[0]/2, ret);
1063 }
1064
1065 /* reset slice direction at end of frame */
1066 if ((srcSliceY_internal + srcSliceH == c->srcH) || scale_dst)
1067 c->sliceDir = 0;
1068
1069 return ret;
1070 }
1071
sws_frame_end(struct SwsContext * c)1072 void sws_frame_end(struct SwsContext *c)
1073 {
1074 av_frame_unref(c->frame_src);
1075 av_frame_unref(c->frame_dst);
1076 c->src_ranges.nb_ranges = 0;
1077 }
1078
sws_frame_start(struct SwsContext * c,AVFrame * dst,const AVFrame * src)1079 int sws_frame_start(struct SwsContext *c, AVFrame *dst, const AVFrame *src)
1080 {
1081 int ret, allocated = 0;
1082
1083 ret = av_frame_ref(c->frame_src, src);
1084 if (ret < 0)
1085 return ret;
1086
1087 if (!dst->buf[0]) {
1088 dst->width = c->dstW;
1089 dst->height = c->dstH;
1090 dst->format = c->dstFormat;
1091
1092 ret = av_frame_get_buffer(dst, 0);
1093 if (ret < 0)
1094 return ret;
1095 allocated = 1;
1096 }
1097
1098 ret = av_frame_ref(c->frame_dst, dst);
1099 if (ret < 0) {
1100 if (allocated)
1101 av_frame_unref(dst);
1102
1103 return ret;
1104 }
1105
1106 return 0;
1107 }
1108
sws_send_slice(struct SwsContext * c,unsigned int slice_start,unsigned int slice_height)1109 int sws_send_slice(struct SwsContext *c, unsigned int slice_start,
1110 unsigned int slice_height)
1111 {
1112 int ret;
1113
1114 ret = ff_range_add(&c->src_ranges, slice_start, slice_height);
1115 if (ret < 0)
1116 return ret;
1117
1118 return 0;
1119 }
1120
sws_receive_slice_alignment(const struct SwsContext * c)1121 unsigned int sws_receive_slice_alignment(const struct SwsContext *c)
1122 {
1123 if (c->slice_ctx)
1124 return c->slice_ctx[0]->dst_slice_align;
1125
1126 return c->dst_slice_align;
1127 }
1128
sws_receive_slice(struct SwsContext * c,unsigned int slice_start,unsigned int slice_height)1129 int sws_receive_slice(struct SwsContext *c, unsigned int slice_start,
1130 unsigned int slice_height)
1131 {
1132 unsigned int align = sws_receive_slice_alignment(c);
1133 uint8_t *dst[4];
1134
1135 /* wait until complete input has been received */
1136 if (!(c->src_ranges.nb_ranges == 1 &&
1137 c->src_ranges.ranges[0].start == 0 &&
1138 c->src_ranges.ranges[0].len == c->srcH))
1139 return AVERROR(EAGAIN);
1140
1141 if ((slice_start > 0 || slice_height < c->dstH) &&
1142 (slice_start % align || slice_height % align)) {
1143 av_log(c, AV_LOG_ERROR,
1144 "Incorrectly aligned output: %u/%u not multiples of %u\n",
1145 slice_start, slice_height, align);
1146 return AVERROR(EINVAL);
1147 }
1148
1149 if (c->slicethread) {
1150 int nb_jobs = c->slice_ctx[0]->dither == SWS_DITHER_ED ? 1 : c->nb_slice_ctx;
1151 int ret = 0;
1152
1153 c->dst_slice_start = slice_start;
1154 c->dst_slice_height = slice_height;
1155
1156 avpriv_slicethread_execute(c->slicethread, nb_jobs, 0);
1157
1158 for (int i = 0; i < c->nb_slice_ctx; i++) {
1159 if (c->slice_err[i] < 0) {
1160 ret = c->slice_err[i];
1161 break;
1162 }
1163 }
1164
1165 memset(c->slice_err, 0, c->nb_slice_ctx * sizeof(*c->slice_err));
1166
1167 return ret;
1168 }
1169
1170 for (int i = 0; i < FF_ARRAY_ELEMS(dst); i++) {
1171 ptrdiff_t offset = c->frame_dst->linesize[i] * (slice_start >> c->chrDstVSubSample);
1172 dst[i] = FF_PTR_ADD(c->frame_dst->data[i], offset);
1173 }
1174
1175 return scale_internal(c, (const uint8_t * const *)c->frame_src->data,
1176 c->frame_src->linesize, 0, c->srcH,
1177 dst, c->frame_dst->linesize, slice_start, slice_height);
1178 }
1179
sws_scale_frame(struct SwsContext * c,AVFrame * dst,const AVFrame * src)1180 int sws_scale_frame(struct SwsContext *c, AVFrame *dst, const AVFrame *src)
1181 {
1182 int ret;
1183
1184 ret = sws_frame_start(c, dst, src);
1185 if (ret < 0)
1186 return ret;
1187
1188 ret = sws_send_slice(c, 0, src->height);
1189 if (ret >= 0)
1190 ret = sws_receive_slice(c, 0, dst->height);
1191
1192 sws_frame_end(c);
1193
1194 return ret;
1195 }
1196
1197 /**
1198 * swscale wrapper, so we don't need to export the SwsContext.
1199 * Assumes planar YUV to be in YUV order instead of YVU.
1200 */
sws_scale(struct SwsContext * c,const uint8_t * const srcSlice[],const int srcStride[],int srcSliceY,int srcSliceH,uint8_t * const dst[],const int dstStride[])1201 int attribute_align_arg sws_scale(struct SwsContext *c,
1202 const uint8_t * const srcSlice[],
1203 const int srcStride[], int srcSliceY,
1204 int srcSliceH, uint8_t *const dst[],
1205 const int dstStride[])
1206 {
1207 if (c->nb_slice_ctx)
1208 c = c->slice_ctx[0];
1209
1210 return scale_internal(c, srcSlice, srcStride, srcSliceY, srcSliceH,
1211 dst, dstStride, 0, c->dstH);
1212 }
1213
ff_sws_slice_worker(void * priv,int jobnr,int threadnr,int nb_jobs,int nb_threads)1214 void ff_sws_slice_worker(void *priv, int jobnr, int threadnr,
1215 int nb_jobs, int nb_threads)
1216 {
1217 SwsContext *parent = priv;
1218 SwsContext *c = parent->slice_ctx[threadnr];
1219
1220 const int slice_height = FFALIGN(FFMAX((parent->dst_slice_height + nb_jobs - 1) / nb_jobs, 1),
1221 c->dst_slice_align);
1222 const int slice_start = jobnr * slice_height;
1223 const int slice_end = FFMIN((jobnr + 1) * slice_height, parent->dst_slice_height);
1224 int err = 0;
1225
1226 if (slice_end > slice_start) {
1227 uint8_t *dst[4] = { NULL };
1228
1229 for (int i = 0; i < FF_ARRAY_ELEMS(dst) && parent->frame_dst->data[i]; i++) {
1230 const int vshift = (i == 1 || i == 2) ? c->chrDstVSubSample : 0;
1231 const ptrdiff_t offset = parent->frame_dst->linesize[i] *
1232 ((slice_start + parent->dst_slice_start) >> vshift);
1233
1234 dst[i] = parent->frame_dst->data[i] + offset;
1235 }
1236
1237 err = scale_internal(c, (const uint8_t * const *)parent->frame_src->data,
1238 parent->frame_src->linesize, 0, c->srcH,
1239 dst, parent->frame_dst->linesize,
1240 parent->dst_slice_start + slice_start, slice_end - slice_start);
1241 }
1242
1243 parent->slice_err[threadnr] = err;
1244 }
1245