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/mem_internal.h"
34 #include "libavutil/pixdesc.h"
35 #include "config.h"
36 #include "rgb2rgb.h"
37 #include "swscale_internal.h"
38 #include "swscale.h"
39
40 DECLARE_ALIGNED(8, const uint8_t, ff_dither_8x8_128)[9][8] = {
41 { 36, 68, 60, 92, 34, 66, 58, 90, },
42 { 100, 4, 124, 28, 98, 2, 122, 26, },
43 { 52, 84, 44, 76, 50, 82, 42, 74, },
44 { 116, 20, 108, 12, 114, 18, 106, 10, },
45 { 32, 64, 56, 88, 38, 70, 62, 94, },
46 { 96, 0, 120, 24, 102, 6, 126, 30, },
47 { 48, 80, 40, 72, 54, 86, 46, 78, },
48 { 112, 16, 104, 8, 118, 22, 110, 14, },
49 { 36, 68, 60, 92, 34, 66, 58, 90, },
50 };
51
52 DECLARE_ALIGNED(8, static const uint8_t, sws_pb_64)[8] = {
53 64, 64, 64, 64, 64, 64, 64, 64
54 };
55
fillPlane(uint8_t * plane,int stride,int width,int height,int y,uint8_t val)56 static av_always_inline void fillPlane(uint8_t *plane, int stride, int width,
57 int height, int y, uint8_t val)
58 {
59 int i;
60 uint8_t *ptr = plane + stride * y;
61 for (i = 0; i < height; i++) {
62 memset(ptr, val, width);
63 ptr += stride;
64 }
65 }
66
hScale16To19_c(SwsContext * c,int16_t * _dst,int dstW,const uint8_t * _src,const int16_t * filter,const int32_t * filterPos,int filterSize)67 static void hScale16To19_c(SwsContext *c, int16_t *_dst, int dstW,
68 const uint8_t *_src, const int16_t *filter,
69 const int32_t *filterPos, int filterSize)
70 {
71 const AVPixFmtDescriptor *desc = av_pix_fmt_desc_get(c->srcFormat);
72 int i;
73 int32_t *dst = (int32_t *) _dst;
74 const uint16_t *src = (const uint16_t *) _src;
75 int bits = desc->comp[0].depth - 1;
76 int sh = bits - 4;
77
78 if ((isAnyRGB(c->srcFormat) || c->srcFormat==AV_PIX_FMT_PAL8) && desc->comp[0].depth<16) {
79 sh = 9;
80 } else if (desc->flags & AV_PIX_FMT_FLAG_FLOAT) { /* float input are process like uint 16bpc */
81 sh = 16 - 1 - 4;
82 }
83
84 for (i = 0; i < dstW; i++) {
85 int j;
86 int srcPos = filterPos[i];
87 int val = 0;
88
89 for (j = 0; j < filterSize; j++) {
90 val += src[srcPos + j] * filter[filterSize * i + j];
91 }
92 // filter=14 bit, input=16 bit, output=30 bit, >> 11 makes 19 bit
93 dst[i] = FFMIN(val >> sh, (1 << 19) - 1);
94 }
95 }
96
hScale16To15_c(SwsContext * c,int16_t * dst,int dstW,const uint8_t * _src,const int16_t * filter,const int32_t * filterPos,int filterSize)97 static void hScale16To15_c(SwsContext *c, int16_t *dst, int dstW,
98 const uint8_t *_src, const int16_t *filter,
99 const int32_t *filterPos, int filterSize)
100 {
101 const AVPixFmtDescriptor *desc = av_pix_fmt_desc_get(c->srcFormat);
102 int i;
103 const uint16_t *src = (const uint16_t *) _src;
104 int sh = desc->comp[0].depth - 1;
105
106 if (sh<15) {
107 sh = isAnyRGB(c->srcFormat) || c->srcFormat==AV_PIX_FMT_PAL8 ? 13 : (desc->comp[0].depth - 1);
108 } else if (desc->flags & AV_PIX_FMT_FLAG_FLOAT) { /* float input are process like uint 16bpc */
109 sh = 16 - 1;
110 }
111
112 for (i = 0; i < dstW; i++) {
113 int j;
114 int srcPos = filterPos[i];
115 int val = 0;
116
117 for (j = 0; j < filterSize; j++) {
118 val += src[srcPos + j] * filter[filterSize * i + j];
119 }
120 // filter=14 bit, input=16 bit, output=30 bit, >> 15 makes 15 bit
121 dst[i] = FFMIN(val >> sh, (1 << 15) - 1);
122 }
123 }
124
125 // 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)126 static void hScale8To15_c(SwsContext *c, int16_t *dst, int dstW,
127 const uint8_t *src, const int16_t *filter,
128 const int32_t *filterPos, int filterSize)
129 {
130 int i;
131 for (i = 0; i < dstW; i++) {
132 int j;
133 int srcPos = filterPos[i];
134 int val = 0;
135 for (j = 0; j < filterSize; j++) {
136 val += ((int)src[srcPos + j]) * filter[filterSize * i + j];
137 }
138 dst[i] = FFMIN(val >> 7, (1 << 15) - 1); // the cubic equation does overflow ...
139 }
140 }
141
hScale8To19_c(SwsContext * c,int16_t * _dst,int dstW,const uint8_t * src,const int16_t * filter,const int32_t * filterPos,int filterSize)142 static void hScale8To19_c(SwsContext *c, int16_t *_dst, int dstW,
143 const uint8_t *src, const int16_t *filter,
144 const int32_t *filterPos, int filterSize)
145 {
146 int i;
147 int32_t *dst = (int32_t *) _dst;
148 for (i = 0; i < dstW; i++) {
149 int j;
150 int srcPos = filterPos[i];
151 int val = 0;
152 for (j = 0; j < filterSize; j++) {
153 val += ((int)src[srcPos + j]) * filter[filterSize * i + j];
154 }
155 dst[i] = FFMIN(val >> 3, (1 << 19) - 1); // the cubic equation does overflow ...
156 }
157 }
158
159 // FIXME all pal and rgb srcFormats could do this conversion as well
160 // FIXME all scalers more complex than bilinear could do half of this transform
chrRangeToJpeg_c(int16_t * dstU,int16_t * dstV,int width)161 static void chrRangeToJpeg_c(int16_t *dstU, int16_t *dstV, int width)
162 {
163 int i;
164 for (i = 0; i < width; i++) {
165 dstU[i] = (FFMIN(dstU[i], 30775) * 4663 - 9289992) >> 12; // -264
166 dstV[i] = (FFMIN(dstV[i], 30775) * 4663 - 9289992) >> 12; // -264
167 }
168 }
169
chrRangeFromJpeg_c(int16_t * dstU,int16_t * dstV,int width)170 static void chrRangeFromJpeg_c(int16_t *dstU, int16_t *dstV, int width)
171 {
172 int i;
173 for (i = 0; i < width; i++) {
174 dstU[i] = (dstU[i] * 1799 + 4081085) >> 11; // 1469
175 dstV[i] = (dstV[i] * 1799 + 4081085) >> 11; // 1469
176 }
177 }
178
lumRangeToJpeg_c(int16_t * dst,int width)179 static void lumRangeToJpeg_c(int16_t *dst, int width)
180 {
181 int i;
182 for (i = 0; i < width; i++)
183 dst[i] = (FFMIN(dst[i], 30189) * 19077 - 39057361) >> 14;
184 }
185
lumRangeFromJpeg_c(int16_t * dst,int width)186 static void lumRangeFromJpeg_c(int16_t *dst, int width)
187 {
188 int i;
189 for (i = 0; i < width; i++)
190 dst[i] = (dst[i] * 14071 + 33561947) >> 14;
191 }
192
chrRangeToJpeg16_c(int16_t * _dstU,int16_t * _dstV,int width)193 static void chrRangeToJpeg16_c(int16_t *_dstU, int16_t *_dstV, int width)
194 {
195 int i;
196 int32_t *dstU = (int32_t *) _dstU;
197 int32_t *dstV = (int32_t *) _dstV;
198 for (i = 0; i < width; i++) {
199 dstU[i] = (FFMIN(dstU[i], 30775 << 4) * 4663 - (9289992 << 4)) >> 12; // -264
200 dstV[i] = (FFMIN(dstV[i], 30775 << 4) * 4663 - (9289992 << 4)) >> 12; // -264
201 }
202 }
203
chrRangeFromJpeg16_c(int16_t * _dstU,int16_t * _dstV,int width)204 static void chrRangeFromJpeg16_c(int16_t *_dstU, int16_t *_dstV, int width)
205 {
206 int i;
207 int32_t *dstU = (int32_t *) _dstU;
208 int32_t *dstV = (int32_t *) _dstV;
209 for (i = 0; i < width; i++) {
210 dstU[i] = (dstU[i] * 1799 + (4081085 << 4)) >> 11; // 1469
211 dstV[i] = (dstV[i] * 1799 + (4081085 << 4)) >> 11; // 1469
212 }
213 }
214
lumRangeToJpeg16_c(int16_t * _dst,int width)215 static void lumRangeToJpeg16_c(int16_t *_dst, int width)
216 {
217 int i;
218 int32_t *dst = (int32_t *) _dst;
219 for (i = 0; i < width; i++) {
220 dst[i] = ((int)(FFMIN(dst[i], 30189 << 4) * 4769U - (39057361 << 2))) >> 12;
221 }
222 }
223
lumRangeFromJpeg16_c(int16_t * _dst,int width)224 static void lumRangeFromJpeg16_c(int16_t *_dst, int width)
225 {
226 int i;
227 int32_t *dst = (int32_t *) _dst;
228 for (i = 0; i < width; i++)
229 dst[i] = (dst[i]*(14071/4) + (33561947<<4)/4)>>12;
230 }
231
232
233 #define DEBUG_SWSCALE_BUFFERS 0
234 #define DEBUG_BUFFERS(...) \
235 if (DEBUG_SWSCALE_BUFFERS) \
236 av_log(c, AV_LOG_DEBUG, __VA_ARGS__)
237
swscale(SwsContext * c,const uint8_t * src[],int srcStride[],int srcSliceY,int srcSliceH,uint8_t * dst[],int dstStride[])238 static int swscale(SwsContext *c, const uint8_t *src[],
239 int srcStride[], int srcSliceY,
240 int srcSliceH, uint8_t *dst[], int dstStride[])
241 {
242 /* load a few things into local vars to make the code more readable?
243 * and faster */
244 const int dstW = c->dstW;
245 const int dstH = c->dstH;
246
247 const enum AVPixelFormat dstFormat = c->dstFormat;
248 const int flags = c->flags;
249 int32_t *vLumFilterPos = c->vLumFilterPos;
250 int32_t *vChrFilterPos = c->vChrFilterPos;
251
252 const int vLumFilterSize = c->vLumFilterSize;
253 const int vChrFilterSize = c->vChrFilterSize;
254
255 yuv2planar1_fn yuv2plane1 = c->yuv2plane1;
256 yuv2planarX_fn yuv2planeX = c->yuv2planeX;
257 yuv2interleavedX_fn yuv2nv12cX = c->yuv2nv12cX;
258 yuv2packed1_fn yuv2packed1 = c->yuv2packed1;
259 yuv2packed2_fn yuv2packed2 = c->yuv2packed2;
260 yuv2packedX_fn yuv2packedX = c->yuv2packedX;
261 yuv2anyX_fn yuv2anyX = c->yuv2anyX;
262 const int chrSrcSliceY = srcSliceY >> c->chrSrcVSubSample;
263 const int chrSrcSliceH = AV_CEIL_RSHIFT(srcSliceH, c->chrSrcVSubSample);
264 int should_dither = isNBPS(c->srcFormat) ||
265 is16BPS(c->srcFormat);
266 int lastDstY;
267
268 /* vars which will change and which we need to store back in the context */
269 int dstY = c->dstY;
270 int lastInLumBuf = c->lastInLumBuf;
271 int lastInChrBuf = c->lastInChrBuf;
272
273 int lumStart = 0;
274 int lumEnd = c->descIndex[0];
275 int chrStart = lumEnd;
276 int chrEnd = c->descIndex[1];
277 int vStart = chrEnd;
278 int vEnd = c->numDesc;
279 SwsSlice *src_slice = &c->slice[lumStart];
280 SwsSlice *hout_slice = &c->slice[c->numSlice-2];
281 SwsSlice *vout_slice = &c->slice[c->numSlice-1];
282 SwsFilterDescriptor *desc = c->desc;
283
284 int needAlpha = c->needAlpha;
285
286 int hasLumHoles = 1;
287 int hasChrHoles = 1;
288
289 if (isPacked(c->srcFormat)) {
290 src[1] =
291 src[2] =
292 src[3] = src[0];
293 srcStride[1] =
294 srcStride[2] =
295 srcStride[3] = srcStride[0];
296 }
297 srcStride[1] *= 1 << c->vChrDrop;
298 srcStride[2] *= 1 << c->vChrDrop;
299
300 DEBUG_BUFFERS("swscale() %p[%d] %p[%d] %p[%d] %p[%d] -> %p[%d] %p[%d] %p[%d] %p[%d]\n",
301 src[0], srcStride[0], src[1], srcStride[1],
302 src[2], srcStride[2], src[3], srcStride[3],
303 dst[0], dstStride[0], dst[1], dstStride[1],
304 dst[2], dstStride[2], dst[3], dstStride[3]);
305 DEBUG_BUFFERS("srcSliceY: %d srcSliceH: %d dstY: %d dstH: %d\n",
306 srcSliceY, srcSliceH, dstY, dstH);
307 DEBUG_BUFFERS("vLumFilterSize: %d vChrFilterSize: %d\n",
308 vLumFilterSize, vChrFilterSize);
309
310 if (dstStride[0]&15 || dstStride[1]&15 ||
311 dstStride[2]&15 || dstStride[3]&15) {
312 static int warnedAlready = 0; // FIXME maybe move this into the context
313 if (flags & SWS_PRINT_INFO && !warnedAlready) {
314 av_log(c, AV_LOG_WARNING,
315 "Warning: dstStride is not aligned!\n"
316 " ->cannot do aligned memory accesses anymore\n");
317 warnedAlready = 1;
318 }
319 }
320
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 static int warnedAlready=0;
327 int cpu_flags = av_get_cpu_flags();
328 if (HAVE_MMXEXT && (cpu_flags & AV_CPU_FLAG_SSE2) && !warnedAlready){
329 av_log(c, AV_LOG_WARNING, "Warning: data is not aligned! This can lead to a speed loss\n");
330 warnedAlready=1;
331 }
332 }
333
334 /* Note the user might start scaling the picture in the middle so this
335 * will not get executed. This is not really intended but works
336 * currently, so people might do it. */
337 if (srcSliceY == 0) {
338 dstY = 0;
339 lastInLumBuf = -1;
340 lastInChrBuf = -1;
341 }
342
343 if (!should_dither) {
344 c->chrDither8 = c->lumDither8 = sws_pb_64;
345 }
346 lastDstY = dstY;
347
348 ff_init_vscale_pfn(c, yuv2plane1, yuv2planeX, yuv2nv12cX,
349 yuv2packed1, yuv2packed2, yuv2packedX, yuv2anyX, c->use_mmx_vfilter);
350
351 ff_init_slice_from_src(src_slice, (uint8_t**)src, srcStride, c->srcW,
352 srcSliceY, srcSliceH, chrSrcSliceY, chrSrcSliceH, 1);
353
354 ff_init_slice_from_src(vout_slice, (uint8_t**)dst, dstStride, c->dstW,
355 dstY, dstH, dstY >> c->chrDstVSubSample,
356 AV_CEIL_RSHIFT(dstH, c->chrDstVSubSample), 0);
357 if (srcSliceY == 0) {
358 hout_slice->plane[0].sliceY = lastInLumBuf + 1;
359 hout_slice->plane[1].sliceY = lastInChrBuf + 1;
360 hout_slice->plane[2].sliceY = lastInChrBuf + 1;
361 hout_slice->plane[3].sliceY = lastInLumBuf + 1;
362
363 hout_slice->plane[0].sliceH =
364 hout_slice->plane[1].sliceH =
365 hout_slice->plane[2].sliceH =
366 hout_slice->plane[3].sliceH = 0;
367 hout_slice->width = dstW;
368 }
369
370 for (; dstY < dstH; dstY++) {
371 const int chrDstY = dstY >> c->chrDstVSubSample;
372 int use_mmx_vfilter= c->use_mmx_vfilter;
373
374 // First line needed as input
375 const int firstLumSrcY = FFMAX(1 - vLumFilterSize, vLumFilterPos[dstY]);
376 const int firstLumSrcY2 = FFMAX(1 - vLumFilterSize, vLumFilterPos[FFMIN(dstY | ((1 << c->chrDstVSubSample) - 1), dstH - 1)]);
377 // First line needed as input
378 const int firstChrSrcY = FFMAX(1 - vChrFilterSize, vChrFilterPos[chrDstY]);
379
380 // Last line needed as input
381 int lastLumSrcY = FFMIN(c->srcH, firstLumSrcY + vLumFilterSize) - 1;
382 int lastLumSrcY2 = FFMIN(c->srcH, firstLumSrcY2 + vLumFilterSize) - 1;
383 int lastChrSrcY = FFMIN(c->chrSrcH, firstChrSrcY + vChrFilterSize) - 1;
384 int enough_lines;
385
386 int i;
387 int posY, cPosY, firstPosY, lastPosY, firstCPosY, lastCPosY;
388
389 // handle holes (FAST_BILINEAR & weird filters)
390 if (firstLumSrcY > lastInLumBuf) {
391
392 hasLumHoles = lastInLumBuf != firstLumSrcY - 1;
393 if (hasLumHoles) {
394 hout_slice->plane[0].sliceY = firstLumSrcY;
395 hout_slice->plane[3].sliceY = firstLumSrcY;
396 hout_slice->plane[0].sliceH =
397 hout_slice->plane[3].sliceH = 0;
398 }
399
400 lastInLumBuf = firstLumSrcY - 1;
401 }
402 if (firstChrSrcY > lastInChrBuf) {
403
404 hasChrHoles = lastInChrBuf != firstChrSrcY - 1;
405 if (hasChrHoles) {
406 hout_slice->plane[1].sliceY = firstChrSrcY;
407 hout_slice->plane[2].sliceY = firstChrSrcY;
408 hout_slice->plane[1].sliceH =
409 hout_slice->plane[2].sliceH = 0;
410 }
411
412 lastInChrBuf = firstChrSrcY - 1;
413 }
414
415 DEBUG_BUFFERS("dstY: %d\n", dstY);
416 DEBUG_BUFFERS("\tfirstLumSrcY: %d lastLumSrcY: %d lastInLumBuf: %d\n",
417 firstLumSrcY, lastLumSrcY, lastInLumBuf);
418 DEBUG_BUFFERS("\tfirstChrSrcY: %d lastChrSrcY: %d lastInChrBuf: %d\n",
419 firstChrSrcY, lastChrSrcY, lastInChrBuf);
420
421 // Do we have enough lines in this slice to output the dstY line
422 enough_lines = lastLumSrcY2 < srcSliceY + srcSliceH &&
423 lastChrSrcY < AV_CEIL_RSHIFT(srcSliceY + srcSliceH, c->chrSrcVSubSample);
424
425 if (!enough_lines) {
426 lastLumSrcY = srcSliceY + srcSliceH - 1;
427 lastChrSrcY = chrSrcSliceY + chrSrcSliceH - 1;
428 DEBUG_BUFFERS("buffering slice: lastLumSrcY %d lastChrSrcY %d\n",
429 lastLumSrcY, lastChrSrcY);
430 }
431
432 av_assert0((lastLumSrcY - firstLumSrcY + 1) <= hout_slice->plane[0].available_lines);
433 av_assert0((lastChrSrcY - firstChrSrcY + 1) <= hout_slice->plane[1].available_lines);
434
435
436 posY = hout_slice->plane[0].sliceY + hout_slice->plane[0].sliceH;
437 if (posY <= lastLumSrcY && !hasLumHoles) {
438 firstPosY = FFMAX(firstLumSrcY, posY);
439 lastPosY = FFMIN(firstLumSrcY + hout_slice->plane[0].available_lines - 1, srcSliceY + srcSliceH - 1);
440 } else {
441 firstPosY = posY;
442 lastPosY = lastLumSrcY;
443 }
444
445 cPosY = hout_slice->plane[1].sliceY + hout_slice->plane[1].sliceH;
446 if (cPosY <= lastChrSrcY && !hasChrHoles) {
447 firstCPosY = FFMAX(firstChrSrcY, cPosY);
448 lastCPosY = FFMIN(firstChrSrcY + hout_slice->plane[1].available_lines - 1, AV_CEIL_RSHIFT(srcSliceY + srcSliceH, c->chrSrcVSubSample) - 1);
449 } else {
450 firstCPosY = cPosY;
451 lastCPosY = lastChrSrcY;
452 }
453
454 ff_rotate_slice(hout_slice, lastPosY, lastCPosY);
455
456 if (posY < lastLumSrcY + 1) {
457 for (i = lumStart; i < lumEnd; ++i)
458 desc[i].process(c, &desc[i], firstPosY, lastPosY - firstPosY + 1);
459 }
460
461 lastInLumBuf = lastLumSrcY;
462
463 if (cPosY < lastChrSrcY + 1) {
464 for (i = chrStart; i < chrEnd; ++i)
465 desc[i].process(c, &desc[i], firstCPosY, lastCPosY - firstCPosY + 1);
466 }
467
468 lastInChrBuf = lastChrSrcY;
469
470 if (!enough_lines)
471 break; // we can't output a dstY line so let's try with the next slice
472
473 #if HAVE_MMX_INLINE
474 ff_updateMMXDitherTables(c, dstY);
475 #endif
476 if (should_dither) {
477 c->chrDither8 = ff_dither_8x8_128[chrDstY & 7];
478 c->lumDither8 = ff_dither_8x8_128[dstY & 7];
479 }
480 if (dstY >= dstH - 2) {
481 /* hmm looks like we can't use MMX here without overwriting
482 * this array's tail */
483 ff_sws_init_output_funcs(c, &yuv2plane1, &yuv2planeX, &yuv2nv12cX,
484 &yuv2packed1, &yuv2packed2, &yuv2packedX, &yuv2anyX);
485 use_mmx_vfilter= 0;
486 ff_init_vscale_pfn(c, yuv2plane1, yuv2planeX, yuv2nv12cX,
487 yuv2packed1, yuv2packed2, yuv2packedX, yuv2anyX, use_mmx_vfilter);
488 }
489
490 {
491 for (i = vStart; i < vEnd; ++i)
492 desc[i].process(c, &desc[i], dstY, 1);
493 }
494 }
495 if (isPlanar(dstFormat) && isALPHA(dstFormat) && !needAlpha) {
496 int length = dstW;
497 int height = dstY - lastDstY;
498
499 if (is16BPS(dstFormat) || isNBPS(dstFormat)) {
500 const AVPixFmtDescriptor *desc = av_pix_fmt_desc_get(dstFormat);
501 fillPlane16(dst[3], dstStride[3], length, height, lastDstY,
502 1, desc->comp[3].depth,
503 isBE(dstFormat));
504 } else if (is32BPS(dstFormat)) {
505 const AVPixFmtDescriptor *desc = av_pix_fmt_desc_get(dstFormat);
506 fillPlane32(dst[3], dstStride[3], length, height, lastDstY,
507 1, desc->comp[3].depth,
508 isBE(dstFormat), desc->flags & AV_PIX_FMT_FLAG_FLOAT);
509 } else
510 fillPlane(dst[3], dstStride[3], length, height, lastDstY, 255);
511 }
512
513 #if HAVE_MMXEXT_INLINE
514 if (av_get_cpu_flags() & AV_CPU_FLAG_MMXEXT)
515 __asm__ volatile ("sfence" ::: "memory");
516 #endif
517 emms_c();
518
519 /* store changed local vars back in the context */
520 c->dstY = dstY;
521 c->lastInLumBuf = lastInLumBuf;
522 c->lastInChrBuf = lastInChrBuf;
523
524 return dstY - lastDstY;
525 }
526
ff_sws_init_range_convert(SwsContext * c)527 av_cold void ff_sws_init_range_convert(SwsContext *c)
528 {
529 c->lumConvertRange = NULL;
530 c->chrConvertRange = NULL;
531 if (c->srcRange != c->dstRange && !isAnyRGB(c->dstFormat)) {
532 if (c->dstBpc <= 14) {
533 if (c->srcRange) {
534 c->lumConvertRange = lumRangeFromJpeg_c;
535 c->chrConvertRange = chrRangeFromJpeg_c;
536 } else {
537 c->lumConvertRange = lumRangeToJpeg_c;
538 c->chrConvertRange = chrRangeToJpeg_c;
539 }
540 } else {
541 if (c->srcRange) {
542 c->lumConvertRange = lumRangeFromJpeg16_c;
543 c->chrConvertRange = chrRangeFromJpeg16_c;
544 } else {
545 c->lumConvertRange = lumRangeToJpeg16_c;
546 c->chrConvertRange = chrRangeToJpeg16_c;
547 }
548 }
549 }
550 }
551
sws_init_swscale(SwsContext * c)552 static av_cold void sws_init_swscale(SwsContext *c)
553 {
554 enum AVPixelFormat srcFormat = c->srcFormat;
555
556 ff_sws_init_output_funcs(c, &c->yuv2plane1, &c->yuv2planeX,
557 &c->yuv2nv12cX, &c->yuv2packed1,
558 &c->yuv2packed2, &c->yuv2packedX, &c->yuv2anyX);
559
560 ff_sws_init_input_funcs(c);
561
562 if (c->srcBpc == 8) {
563 if (c->dstBpc <= 14) {
564 c->hyScale = c->hcScale = hScale8To15_c;
565 if (c->flags & SWS_FAST_BILINEAR) {
566 c->hyscale_fast = ff_hyscale_fast_c;
567 c->hcscale_fast = ff_hcscale_fast_c;
568 }
569 } else {
570 c->hyScale = c->hcScale = hScale8To19_c;
571 }
572 } else {
573 c->hyScale = c->hcScale = c->dstBpc > 14 ? hScale16To19_c
574 : hScale16To15_c;
575 }
576
577 ff_sws_init_range_convert(c);
578
579 if (!(isGray(srcFormat) || isGray(c->dstFormat) ||
580 srcFormat == AV_PIX_FMT_MONOBLACK || srcFormat == AV_PIX_FMT_MONOWHITE))
581 c->needs_hcscale = 1;
582 }
583
ff_getSwsFunc(SwsContext * c)584 SwsFunc ff_getSwsFunc(SwsContext *c)
585 {
586 sws_init_swscale(c);
587
588 if (ARCH_PPC)
589 ff_sws_init_swscale_ppc(c);
590 if (ARCH_X86)
591 ff_sws_init_swscale_x86(c);
592 if (ARCH_AARCH64)
593 ff_sws_init_swscale_aarch64(c);
594 if (ARCH_ARM)
595 ff_sws_init_swscale_arm(c);
596
597 return swscale;
598 }
599
reset_ptr(const uint8_t * src[],enum AVPixelFormat format)600 static void reset_ptr(const uint8_t *src[], enum AVPixelFormat format)
601 {
602 if (!isALPHA(format))
603 src[3] = NULL;
604 if (!isPlanar(format)) {
605 src[3] = src[2] = NULL;
606
607 if (!usePal(format))
608 src[1] = NULL;
609 }
610 }
611
check_image_pointers(const uint8_t * const data[4],enum AVPixelFormat pix_fmt,const int linesizes[4])612 static int check_image_pointers(const uint8_t * const data[4], enum AVPixelFormat pix_fmt,
613 const int linesizes[4])
614 {
615 const AVPixFmtDescriptor *desc = av_pix_fmt_desc_get(pix_fmt);
616 int i;
617
618 av_assert2(desc);
619
620 for (i = 0; i < 4; i++) {
621 int plane = desc->comp[i].plane;
622 if (!data[plane] || !linesizes[plane])
623 return 0;
624 }
625
626 return 1;
627 }
628
xyz12Torgb48(struct SwsContext * c,uint16_t * dst,const uint16_t * src,int stride,int h)629 static void xyz12Torgb48(struct SwsContext *c, uint16_t *dst,
630 const uint16_t *src, int stride, int h)
631 {
632 int xp,yp;
633 const AVPixFmtDescriptor *desc = av_pix_fmt_desc_get(c->srcFormat);
634
635 for (yp=0; yp<h; yp++) {
636 for (xp=0; xp+2<stride; xp+=3) {
637 int x, y, z, r, g, b;
638
639 if (desc->flags & AV_PIX_FMT_FLAG_BE) {
640 x = AV_RB16(src + xp + 0);
641 y = AV_RB16(src + xp + 1);
642 z = AV_RB16(src + xp + 2);
643 } else {
644 x = AV_RL16(src + xp + 0);
645 y = AV_RL16(src + xp + 1);
646 z = AV_RL16(src + xp + 2);
647 }
648
649 x = c->xyzgamma[x>>4];
650 y = c->xyzgamma[y>>4];
651 z = c->xyzgamma[z>>4];
652
653 // convert from XYZlinear to sRGBlinear
654 r = c->xyz2rgb_matrix[0][0] * x +
655 c->xyz2rgb_matrix[0][1] * y +
656 c->xyz2rgb_matrix[0][2] * z >> 12;
657 g = c->xyz2rgb_matrix[1][0] * x +
658 c->xyz2rgb_matrix[1][1] * y +
659 c->xyz2rgb_matrix[1][2] * z >> 12;
660 b = c->xyz2rgb_matrix[2][0] * x +
661 c->xyz2rgb_matrix[2][1] * y +
662 c->xyz2rgb_matrix[2][2] * z >> 12;
663
664 // limit values to 12-bit depth
665 r = av_clip_uintp2(r, 12);
666 g = av_clip_uintp2(g, 12);
667 b = av_clip_uintp2(b, 12);
668
669 // convert from sRGBlinear to RGB and scale from 12bit to 16bit
670 if (desc->flags & AV_PIX_FMT_FLAG_BE) {
671 AV_WB16(dst + xp + 0, c->rgbgamma[r] << 4);
672 AV_WB16(dst + xp + 1, c->rgbgamma[g] << 4);
673 AV_WB16(dst + xp + 2, c->rgbgamma[b] << 4);
674 } else {
675 AV_WL16(dst + xp + 0, c->rgbgamma[r] << 4);
676 AV_WL16(dst + xp + 1, c->rgbgamma[g] << 4);
677 AV_WL16(dst + xp + 2, c->rgbgamma[b] << 4);
678 }
679 }
680 src += stride;
681 dst += stride;
682 }
683 }
684
rgb48Toxyz12(struct SwsContext * c,uint16_t * dst,const uint16_t * src,int stride,int h)685 static void rgb48Toxyz12(struct SwsContext *c, uint16_t *dst,
686 const uint16_t *src, int stride, int h)
687 {
688 int xp,yp;
689 const AVPixFmtDescriptor *desc = av_pix_fmt_desc_get(c->dstFormat);
690
691 for (yp=0; yp<h; yp++) {
692 for (xp=0; xp+2<stride; xp+=3) {
693 int x, y, z, r, g, b;
694
695 if (desc->flags & AV_PIX_FMT_FLAG_BE) {
696 r = AV_RB16(src + xp + 0);
697 g = AV_RB16(src + xp + 1);
698 b = AV_RB16(src + xp + 2);
699 } else {
700 r = AV_RL16(src + xp + 0);
701 g = AV_RL16(src + xp + 1);
702 b = AV_RL16(src + xp + 2);
703 }
704
705 r = c->rgbgammainv[r>>4];
706 g = c->rgbgammainv[g>>4];
707 b = c->rgbgammainv[b>>4];
708
709 // convert from sRGBlinear to XYZlinear
710 x = c->rgb2xyz_matrix[0][0] * r +
711 c->rgb2xyz_matrix[0][1] * g +
712 c->rgb2xyz_matrix[0][2] * b >> 12;
713 y = c->rgb2xyz_matrix[1][0] * r +
714 c->rgb2xyz_matrix[1][1] * g +
715 c->rgb2xyz_matrix[1][2] * b >> 12;
716 z = c->rgb2xyz_matrix[2][0] * r +
717 c->rgb2xyz_matrix[2][1] * g +
718 c->rgb2xyz_matrix[2][2] * b >> 12;
719
720 // limit values to 12-bit depth
721 x = av_clip_uintp2(x, 12);
722 y = av_clip_uintp2(y, 12);
723 z = av_clip_uintp2(z, 12);
724
725 // convert from XYZlinear to X'Y'Z' and scale from 12bit to 16bit
726 if (desc->flags & AV_PIX_FMT_FLAG_BE) {
727 AV_WB16(dst + xp + 0, c->xyzgammainv[x] << 4);
728 AV_WB16(dst + xp + 1, c->xyzgammainv[y] << 4);
729 AV_WB16(dst + xp + 2, c->xyzgammainv[z] << 4);
730 } else {
731 AV_WL16(dst + xp + 0, c->xyzgammainv[x] << 4);
732 AV_WL16(dst + xp + 1, c->xyzgammainv[y] << 4);
733 AV_WL16(dst + xp + 2, c->xyzgammainv[z] << 4);
734 }
735 }
736 src += stride;
737 dst += stride;
738 }
739 }
740
741 /**
742 * swscale wrapper, so we don't need to export the SwsContext.
743 * Assumes planar YUV to be in YUV order instead of YVU.
744 */
sws_scale(struct SwsContext * c,const uint8_t * const srcSlice[],const int srcStride[],int srcSliceY,int srcSliceH,uint8_t * const dst[],const int dstStride[])745 int attribute_align_arg sws_scale(struct SwsContext *c,
746 const uint8_t * const srcSlice[],
747 const int srcStride[], int srcSliceY,
748 int srcSliceH, uint8_t *const dst[],
749 const int dstStride[])
750 {
751 int i, ret;
752 const uint8_t *src2[4];
753 uint8_t *dst2[4];
754 uint8_t *rgb0_tmp = NULL;
755 int macro_height = isBayer(c->srcFormat) ? 2 : (1 << c->chrSrcVSubSample);
756 // copy strides, so they can safely be modified
757 int srcStride2[4];
758 int dstStride2[4];
759 int srcSliceY_internal = srcSliceY;
760
761 if (!srcStride || !dstStride || !dst || !srcSlice) {
762 av_log(c, AV_LOG_ERROR, "One of the input parameters to sws_scale() is NULL, please check the calling code\n");
763 return 0;
764 }
765
766 for (i=0; i<4; i++) {
767 srcStride2[i] = srcStride[i];
768 dstStride2[i] = dstStride[i];
769 }
770
771 if ((srcSliceY & (macro_height-1)) ||
772 ((srcSliceH& (macro_height-1)) && srcSliceY + srcSliceH != c->srcH) ||
773 srcSliceY + srcSliceH > c->srcH) {
774 av_log(c, AV_LOG_ERROR, "Slice parameters %d, %d are invalid\n", srcSliceY, srcSliceH);
775 return AVERROR(EINVAL);
776 }
777
778 if (c->gamma_flag && c->cascaded_context[0]) {
779 ret = sws_scale(c->cascaded_context[0],
780 srcSlice, srcStride, srcSliceY, srcSliceH,
781 c->cascaded_tmp, c->cascaded_tmpStride);
782
783 if (ret < 0)
784 return ret;
785
786 if (c->cascaded_context[2])
787 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);
788 else
789 ret = sws_scale(c->cascaded_context[1], (const uint8_t * const *)c->cascaded_tmp, c->cascaded_tmpStride, srcSliceY, srcSliceH, dst, dstStride);
790
791 if (ret < 0)
792 return ret;
793
794 if (c->cascaded_context[2]) {
795 ret = sws_scale(c->cascaded_context[2],
796 (const uint8_t * const *)c->cascaded1_tmp, c->cascaded1_tmpStride, c->cascaded_context[1]->dstY - ret, c->cascaded_context[1]->dstY,
797 dst, dstStride);
798 }
799 return ret;
800 }
801
802 if (c->cascaded_context[0] && srcSliceY == 0 && srcSliceH == c->cascaded_context[0]->srcH) {
803 ret = sws_scale(c->cascaded_context[0],
804 srcSlice, srcStride, srcSliceY, srcSliceH,
805 c->cascaded_tmp, c->cascaded_tmpStride);
806 if (ret < 0)
807 return ret;
808 ret = sws_scale(c->cascaded_context[1],
809 (const uint8_t * const * )c->cascaded_tmp, c->cascaded_tmpStride, 0, c->cascaded_context[0]->dstH,
810 dst, dstStride);
811 return ret;
812 }
813
814 memcpy(src2, srcSlice, sizeof(src2));
815 memcpy(dst2, dst, sizeof(dst2));
816
817 // do not mess up sliceDir if we have a "trailing" 0-size slice
818 if (srcSliceH == 0)
819 return 0;
820
821 if (!check_image_pointers(srcSlice, c->srcFormat, srcStride)) {
822 av_log(c, AV_LOG_ERROR, "bad src image pointers\n");
823 return 0;
824 }
825 if (!check_image_pointers((const uint8_t* const*)dst, c->dstFormat, dstStride)) {
826 av_log(c, AV_LOG_ERROR, "bad dst image pointers\n");
827 return 0;
828 }
829
830 if (c->sliceDir == 0 && srcSliceY != 0 && srcSliceY + srcSliceH != c->srcH) {
831 av_log(c, AV_LOG_ERROR, "Slices start in the middle!\n");
832 return 0;
833 }
834 if (c->sliceDir == 0) {
835 if (srcSliceY == 0) c->sliceDir = 1; else c->sliceDir = -1;
836 }
837
838 if (usePal(c->srcFormat)) {
839 for (i = 0; i < 256; i++) {
840 int r, g, b, y, u, v, a = 0xff;
841 if (c->srcFormat == AV_PIX_FMT_PAL8) {
842 uint32_t p = ((const uint32_t *)(srcSlice[1]))[i];
843 a = (p >> 24) & 0xFF;
844 r = (p >> 16) & 0xFF;
845 g = (p >> 8) & 0xFF;
846 b = p & 0xFF;
847 } else if (c->srcFormat == AV_PIX_FMT_RGB8) {
848 r = ( i >> 5 ) * 36;
849 g = ((i >> 2) & 7) * 36;
850 b = ( i & 3) * 85;
851 } else if (c->srcFormat == AV_PIX_FMT_BGR8) {
852 b = ( i >> 6 ) * 85;
853 g = ((i >> 3) & 7) * 36;
854 r = ( i & 7) * 36;
855 } else if (c->srcFormat == AV_PIX_FMT_RGB4_BYTE) {
856 r = ( i >> 3 ) * 255;
857 g = ((i >> 1) & 3) * 85;
858 b = ( i & 1) * 255;
859 } else if (c->srcFormat == AV_PIX_FMT_GRAY8 || c->srcFormat == AV_PIX_FMT_GRAY8A) {
860 r = g = b = i;
861 } else {
862 av_assert1(c->srcFormat == AV_PIX_FMT_BGR4_BYTE);
863 b = ( i >> 3 ) * 255;
864 g = ((i >> 1) & 3) * 85;
865 r = ( i & 1) * 255;
866 }
867 #define RGB2YUV_SHIFT 15
868 #define BY ( (int) (0.114 * 219 / 255 * (1 << RGB2YUV_SHIFT) + 0.5))
869 #define BV (-(int) (0.081 * 224 / 255 * (1 << RGB2YUV_SHIFT) + 0.5))
870 #define BU ( (int) (0.500 * 224 / 255 * (1 << RGB2YUV_SHIFT) + 0.5))
871 #define GY ( (int) (0.587 * 219 / 255 * (1 << RGB2YUV_SHIFT) + 0.5))
872 #define GV (-(int) (0.419 * 224 / 255 * (1 << RGB2YUV_SHIFT) + 0.5))
873 #define GU (-(int) (0.331 * 224 / 255 * (1 << RGB2YUV_SHIFT) + 0.5))
874 #define RY ( (int) (0.299 * 219 / 255 * (1 << RGB2YUV_SHIFT) + 0.5))
875 #define RV ( (int) (0.500 * 224 / 255 * (1 << RGB2YUV_SHIFT) + 0.5))
876 #define RU (-(int) (0.169 * 224 / 255 * (1 << RGB2YUV_SHIFT) + 0.5))
877
878 y = av_clip_uint8((RY * r + GY * g + BY * b + ( 33 << (RGB2YUV_SHIFT - 1))) >> RGB2YUV_SHIFT);
879 u = av_clip_uint8((RU * r + GU * g + BU * b + (257 << (RGB2YUV_SHIFT - 1))) >> RGB2YUV_SHIFT);
880 v = av_clip_uint8((RV * r + GV * g + BV * b + (257 << (RGB2YUV_SHIFT - 1))) >> RGB2YUV_SHIFT);
881 c->pal_yuv[i]= y + (u<<8) + (v<<16) + ((unsigned)a<<24);
882
883 switch (c->dstFormat) {
884 case AV_PIX_FMT_BGR32:
885 #if !HAVE_BIGENDIAN
886 case AV_PIX_FMT_RGB24:
887 #endif
888 c->pal_rgb[i]= r + (g<<8) + (b<<16) + ((unsigned)a<<24);
889 break;
890 case AV_PIX_FMT_BGR32_1:
891 #if HAVE_BIGENDIAN
892 case AV_PIX_FMT_BGR24:
893 #endif
894 c->pal_rgb[i]= a + (r<<8) + (g<<16) + ((unsigned)b<<24);
895 break;
896 case AV_PIX_FMT_RGB32_1:
897 #if HAVE_BIGENDIAN
898 case AV_PIX_FMT_RGB24:
899 #endif
900 c->pal_rgb[i]= a + (b<<8) + (g<<16) + ((unsigned)r<<24);
901 break;
902 case AV_PIX_FMT_RGB32:
903 #if !HAVE_BIGENDIAN
904 case AV_PIX_FMT_BGR24:
905 #endif
906 default:
907 c->pal_rgb[i]= b + (g<<8) + (r<<16) + ((unsigned)a<<24);
908 }
909 }
910 }
911
912 if (c->src0Alpha && !c->dst0Alpha && isALPHA(c->dstFormat)) {
913 uint8_t *base;
914 int x,y;
915 rgb0_tmp = av_malloc(FFABS(srcStride[0]) * srcSliceH + 32);
916 if (!rgb0_tmp)
917 return AVERROR(ENOMEM);
918
919 base = srcStride[0] < 0 ? rgb0_tmp - srcStride[0] * (srcSliceH-1) : rgb0_tmp;
920 for (y=0; y<srcSliceH; y++){
921 memcpy(base + srcStride[0]*y, src2[0] + srcStride[0]*y, 4*c->srcW);
922 for (x=c->src0Alpha-1; x<4*c->srcW; x+=4) {
923 base[ srcStride[0]*y + x] = 0xFF;
924 }
925 }
926 src2[0] = base;
927 }
928
929 if (c->srcXYZ && !(c->dstXYZ && c->srcW==c->dstW && c->srcH==c->dstH)) {
930 uint8_t *base;
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
937 xyz12Torgb48(c, (uint16_t*)base, (const uint16_t*)src2[0], srcStride[0]/2, srcSliceH);
938 src2[0] = base;
939 }
940
941 if (!srcSliceY && (c->flags & SWS_BITEXACT) && c->dither == SWS_DITHER_ED && c->dither_error[0])
942 for (i = 0; i < 4; i++)
943 memset(c->dither_error[i], 0, sizeof(c->dither_error[0][0]) * (c->dstW+2));
944
945 if (c->sliceDir != 1) {
946 // slices go from bottom to top => we flip the image internally
947 for (i=0; i<4; i++) {
948 srcStride2[i] *= -1;
949 dstStride2[i] *= -1;
950 }
951
952 src2[0] += (srcSliceH - 1) * srcStride[0];
953 if (!usePal(c->srcFormat))
954 src2[1] += ((srcSliceH >> c->chrSrcVSubSample) - 1) * srcStride[1];
955 src2[2] += ((srcSliceH >> c->chrSrcVSubSample) - 1) * srcStride[2];
956 src2[3] += (srcSliceH - 1) * srcStride[3];
957 dst2[0] += ( c->dstH - 1) * dstStride[0];
958 dst2[1] += ((c->dstH >> c->chrDstVSubSample) - 1) * dstStride[1];
959 dst2[2] += ((c->dstH >> c->chrDstVSubSample) - 1) * dstStride[2];
960 dst2[3] += ( c->dstH - 1) * dstStride[3];
961
962 srcSliceY_internal = c->srcH-srcSliceY-srcSliceH;
963 }
964 reset_ptr(src2, c->srcFormat);
965 reset_ptr((void*)dst2, c->dstFormat);
966
967 /* reset slice direction at end of frame */
968 if (srcSliceY_internal + srcSliceH == c->srcH)
969 c->sliceDir = 0;
970 ret = c->swscale(c, src2, srcStride2, srcSliceY_internal, srcSliceH, dst2, dstStride2);
971
972 if (c->dstXYZ && !(c->srcXYZ && c->srcW==c->dstW && c->srcH==c->dstH)) {
973 int dstY = c->dstY ? c->dstY : srcSliceY + srcSliceH;
974 uint16_t *dst16 = (uint16_t*)(dst2[0] + (dstY - ret) * dstStride2[0]);
975 av_assert0(dstY >= ret);
976 av_assert0(ret >= 0);
977 av_assert0(c->dstH >= dstY);
978
979 /* replace on the same data */
980 rgb48Toxyz12(c, dst16, dst16, dstStride2[0]/2, ret);
981 }
982
983 av_free(rgb0_tmp);
984 return ret;
985 }
986