swscale.c
Go to the documentation of this file.
1 /*
2  * Copyright (C) 2001-2003 Michael Niedermayer <michaelni@gmx.at>
3  *
4  * This file is part of Libav.
5  *
6  * Libav is free software; you can redistribute it and/or
7  * modify it under the terms of the GNU Lesser General Public
8  * License as published by the Free Software Foundation; either
9  * version 2.1 of the License, or (at your option) any later version.
10  *
11  * Libav is distributed in the hope that it will be useful,
12  * but WITHOUT ANY WARRANTY; without even the implied warranty of
13  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
14  * Lesser General Public License for more details.
15  *
16  * You should have received a copy of the GNU Lesser General Public
17  * License along with Libav; if not, write to the Free Software
18  * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
19  */
20 
21 #include <assert.h>
22 #include <inttypes.h>
23 #include <math.h>
24 #include <stdio.h>
25 #include <string.h>
26 
27 #include "libavutil/avutil.h"
28 #include "libavutil/bswap.h"
29 #include "libavutil/cpu.h"
30 #include "libavutil/intreadwrite.h"
31 #include "libavutil/mathematics.h"
32 #include "libavutil/pixdesc.h"
33 #include "config.h"
34 #include "rgb2rgb.h"
35 #include "swscale_internal.h"
36 #include "swscale.h"
37 
39  { 36, 68, 60, 92, 34, 66, 58, 90, },
40  { 100, 4, 124, 28, 98, 2, 122, 26, },
41  { 52, 84, 44, 76, 50, 82, 42, 74, },
42  { 116, 20, 108, 12, 114, 18, 106, 10, },
43  { 32, 64, 56, 88, 38, 70, 62, 94, },
44  { 96, 0, 120, 24, 102, 6, 126, 30, },
45  { 48, 80, 40, 72, 54, 86, 46, 78, },
46  { 112, 16, 104, 8, 118, 22, 110, 14, },
47 };
48 
50  64, 64, 64, 64, 64, 64, 64, 64
51 };
52 
53 static av_always_inline void fillPlane(uint8_t *plane, int stride, int width,
54  int height, int y, uint8_t val)
55 {
56  int i;
57  uint8_t *ptr = plane + stride * y;
58  for (i = 0; i < height; i++) {
59  memset(ptr, val, width);
60  ptr += stride;
61  }
62 }
63 
64 static void fill_plane9or10(uint8_t *plane, int stride, int width,
65  int height, int y, uint8_t val,
66  const int dst_depth, const int big_endian)
67 {
68  int i, j;
69  uint16_t *dst = (uint16_t *) (plane + stride * y);
70 #define FILL8TO9_OR_10(wfunc) \
71  for (i = 0; i < height; i++) { \
72  for (j = 0; j < width; j++) { \
73  wfunc(&dst[j], (val << (dst_depth - 8)) | \
74  (val >> (16 - dst_depth))); \
75  } \
76  dst += stride / 2; \
77  }
78  if (big_endian) {
80  } else {
82  }
83 }
84 
85 
86 static void hScale16To19_c(SwsContext *c, int16_t *_dst, int dstW,
87  const uint8_t *_src, const int16_t *filter,
88  const int32_t *filterPos, int filterSize)
89 {
91  int i;
92  int32_t *dst = (int32_t *) _dst;
93  const uint16_t *src = (const uint16_t *) _src;
94  int bits = desc->comp[0].depth_minus1;
95  int sh = bits - 4;
96 
97  for (i = 0; i < dstW; i++) {
98  int j;
99  int srcPos = filterPos[i];
100  int val = 0;
101 
102  for (j = 0; j < filterSize; j++) {
103  val += src[srcPos + j] * filter[filterSize * i + j];
104  }
105  // filter=14 bit, input=16 bit, output=30 bit, >> 11 makes 19 bit
106  dst[i] = FFMIN(val >> sh, (1 << 19) - 1);
107  }
108 }
109 
110 static void hScale16To15_c(SwsContext *c, int16_t *dst, int dstW,
111  const uint8_t *_src, const int16_t *filter,
112  const int32_t *filterPos, int filterSize)
113 {
115  int i;
116  const uint16_t *src = (const uint16_t *) _src;
117  int sh = desc->comp[0].depth_minus1;
118 
119  for (i = 0; i < dstW; i++) {
120  int j;
121  int srcPos = filterPos[i];
122  int val = 0;
123 
124  for (j = 0; j < filterSize; j++) {
125  val += src[srcPos + j] * filter[filterSize * i + j];
126  }
127  // filter=14 bit, input=16 bit, output=30 bit, >> 15 makes 15 bit
128  dst[i] = FFMIN(val >> sh, (1 << 15) - 1);
129  }
130 }
131 
132 // bilinear / bicubic scaling
133 static void hScale8To15_c(SwsContext *c, int16_t *dst, int dstW,
134  const uint8_t *src, const int16_t *filter,
135  const int32_t *filterPos, int filterSize)
136 {
137  int i;
138  for (i = 0; i < dstW; i++) {
139  int j;
140  int srcPos = filterPos[i];
141  int val = 0;
142  for (j = 0; j < filterSize; j++) {
143  val += ((int)src[srcPos + j]) * filter[filterSize * i + j];
144  }
145  dst[i] = FFMIN(val >> 7, (1 << 15) - 1); // the cubic equation does overflow ...
146  }
147 }
148 
149 static void hScale8To19_c(SwsContext *c, int16_t *_dst, int dstW,
150  const uint8_t *src, const int16_t *filter,
151  const int32_t *filterPos, int filterSize)
152 {
153  int i;
154  int32_t *dst = (int32_t *) _dst;
155  for (i = 0; i < dstW; i++) {
156  int j;
157  int srcPos = filterPos[i];
158  int val = 0;
159  for (j = 0; j < filterSize; j++) {
160  val += ((int)src[srcPos + j]) * filter[filterSize * i + j];
161  }
162  dst[i] = FFMIN(val >> 3, (1 << 19) - 1); // the cubic equation does overflow ...
163  }
164 }
165 
166 // FIXME all pal and rgb srcFormats could do this conversion as well
167 // FIXME all scalers more complex than bilinear could do half of this transform
168 static void chrRangeToJpeg_c(int16_t *dstU, int16_t *dstV, int width)
169 {
170  int i;
171  for (i = 0; i < width; i++) {
172  dstU[i] = (FFMIN(dstU[i], 30775) * 4663 - 9289992) >> 12; // -264
173  dstV[i] = (FFMIN(dstV[i], 30775) * 4663 - 9289992) >> 12; // -264
174  }
175 }
176 
177 static void chrRangeFromJpeg_c(int16_t *dstU, int16_t *dstV, int width)
178 {
179  int i;
180  for (i = 0; i < width; i++) {
181  dstU[i] = (dstU[i] * 1799 + 4081085) >> 11; // 1469
182  dstV[i] = (dstV[i] * 1799 + 4081085) >> 11; // 1469
183  }
184 }
185 
186 static void lumRangeToJpeg_c(int16_t *dst, int width)
187 {
188  int i;
189  for (i = 0; i < width; i++)
190  dst[i] = (FFMIN(dst[i], 30189) * 19077 - 39057361) >> 14;
191 }
192 
193 static void lumRangeFromJpeg_c(int16_t *dst, int width)
194 {
195  int i;
196  for (i = 0; i < width; i++)
197  dst[i] = (dst[i] * 14071 + 33561947) >> 14;
198 }
199 
200 static void chrRangeToJpeg16_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] = (FFMIN(dstU[i], 30775 << 4) * 4663 - (9289992 << 4)) >> 12; // -264
207  dstV[i] = (FFMIN(dstV[i], 30775 << 4) * 4663 - (9289992 << 4)) >> 12; // -264
208  }
209 }
210 
211 static void chrRangeFromJpeg16_c(int16_t *_dstU, int16_t *_dstV, int width)
212 {
213  int i;
214  int32_t *dstU = (int32_t *) _dstU;
215  int32_t *dstV = (int32_t *) _dstV;
216  for (i = 0; i < width; i++) {
217  dstU[i] = (dstU[i] * 1799 + (4081085 << 4)) >> 11; // 1469
218  dstV[i] = (dstV[i] * 1799 + (4081085 << 4)) >> 11; // 1469
219  }
220 }
221 
222 static void lumRangeToJpeg16_c(int16_t *_dst, int width)
223 {
224  int i;
225  int32_t *dst = (int32_t *) _dst;
226  for (i = 0; i < width; i++)
227  dst[i] = (FFMIN(dst[i], 30189 << 4) * 4769 - (39057361 << 2)) >> 12;
228 }
229 
230 static void lumRangeFromJpeg16_c(int16_t *_dst, int width)
231 {
232  int i;
233  int32_t *dst = (int32_t *) _dst;
234  for (i = 0; i < width; i++)
235  dst[i] = (dst[i] * 14071 + (33561947 << 4)) >> 14;
236 }
237 
238 static void hyscale_fast_c(SwsContext *c, int16_t *dst, int dstWidth,
239  const uint8_t *src, int srcW, int xInc)
240 {
241  int i;
242  unsigned int xpos = 0;
243  for (i = 0; i < dstWidth; i++) {
244  register unsigned int xx = xpos >> 16;
245  register unsigned int xalpha = (xpos & 0xFFFF) >> 9;
246  dst[i] = (src[xx] << 7) + (src[xx + 1] - src[xx]) * xalpha;
247  xpos += xInc;
248  }
249 }
250 
251 // *** horizontal scale Y line to temp buffer
252 static av_always_inline void hyscale(SwsContext *c, int16_t *dst, int dstWidth,
253  const uint8_t *src_in[4],
254  int srcW, int xInc,
255  const int16_t *hLumFilter,
256  const int32_t *hLumFilterPos,
257  int hLumFilterSize,
259  uint32_t *pal, int isAlpha)
260 {
261  void (*toYV12)(uint8_t *, const uint8_t *, int, uint32_t *) =
262  isAlpha ? c->alpToYV12 : c->lumToYV12;
263  void (*convertRange)(int16_t *, int) = isAlpha ? NULL : c->lumConvertRange;
264  const uint8_t *src = src_in[isAlpha ? 3 : 0];
265 
266  if (toYV12) {
267  toYV12(formatConvBuffer, src, srcW, pal);
268  src = formatConvBuffer;
269  } else if (c->readLumPlanar && !isAlpha) {
270  c->readLumPlanar(formatConvBuffer, src_in, srcW);
271  src = formatConvBuffer;
272  }
273 
274  if (!c->hyscale_fast) {
275  c->hyScale(c, dst, dstWidth, src, hLumFilter,
276  hLumFilterPos, hLumFilterSize);
277  } else { // fast bilinear upscale / crap downscale
278  c->hyscale_fast(c, dst, dstWidth, src, srcW, xInc);
279  }
280 
281  if (convertRange)
282  convertRange(dst, dstWidth);
283 }
284 
285 static void hcscale_fast_c(SwsContext *c, int16_t *dst1, int16_t *dst2,
286  int dstWidth, const uint8_t *src1,
287  const uint8_t *src2, int srcW, int xInc)
288 {
289  int i;
290  unsigned int xpos = 0;
291  for (i = 0; i < dstWidth; i++) {
292  register unsigned int xx = xpos >> 16;
293  register unsigned int xalpha = (xpos & 0xFFFF) >> 9;
294  dst1[i] = (src1[xx] * (xalpha ^ 127) + src1[xx + 1] * xalpha);
295  dst2[i] = (src2[xx] * (xalpha ^ 127) + src2[xx + 1] * xalpha);
296  xpos += xInc;
297  }
298 }
299 
300 static av_always_inline void hcscale(SwsContext *c, int16_t *dst1,
301  int16_t *dst2, int dstWidth,
302  const uint8_t *src_in[4],
303  int srcW, int xInc,
304  const int16_t *hChrFilter,
305  const int32_t *hChrFilterPos,
306  int hChrFilterSize,
307  uint8_t *formatConvBuffer, uint32_t *pal)
308 {
309  const uint8_t *src1 = src_in[1], *src2 = src_in[2];
310  if (c->chrToYV12) {
311  uint8_t *buf2 = formatConvBuffer +
312  FFALIGN(srcW * FFALIGN(c->srcBpc, 8) >> 3, 16);
313  c->chrToYV12(formatConvBuffer, buf2, src1, src2, srcW, pal);
314  src1 = formatConvBuffer;
315  src2 = buf2;
316  } else if (c->readChrPlanar) {
317  uint8_t *buf2 = formatConvBuffer +
318  FFALIGN(srcW * FFALIGN(c->srcBpc, 8) >> 3, 16);
319  c->readChrPlanar(formatConvBuffer, buf2, src_in, srcW);
320  src1 = formatConvBuffer;
321  src2 = buf2;
322  }
323 
324  if (!c->hcscale_fast) {
325  c->hcScale(c, dst1, dstWidth, src1, hChrFilter, hChrFilterPos, hChrFilterSize);
326  c->hcScale(c, dst2, dstWidth, src2, hChrFilter, hChrFilterPos, hChrFilterSize);
327  } else { // fast bilinear upscale / crap downscale
328  c->hcscale_fast(c, dst1, dst2, dstWidth, src1, src2, srcW, xInc);
329  }
330 
331  if (c->chrConvertRange)
332  c->chrConvertRange(dst1, dst2, dstWidth);
333 }
334 
335 #define DEBUG_SWSCALE_BUFFERS 0
336 #define DEBUG_BUFFERS(...) \
337  if (DEBUG_SWSCALE_BUFFERS) \
338  av_log(c, AV_LOG_DEBUG, __VA_ARGS__)
339 
340 static int swScale(SwsContext *c, const uint8_t *src[],
341  int srcStride[], int srcSliceY,
342  int srcSliceH, uint8_t *dst[], int dstStride[])
343 {
344  /* load a few things into local vars to make the code more readable?
345  * and faster */
346  const int srcW = c->srcW;
347  const int dstW = c->dstW;
348  const int dstH = c->dstH;
349  const int chrDstW = c->chrDstW;
350  const int chrSrcW = c->chrSrcW;
351  const int lumXInc = c->lumXInc;
352  const int chrXInc = c->chrXInc;
353  const enum AVPixelFormat dstFormat = c->dstFormat;
354  const int flags = c->flags;
359  int16_t *vLumFilter = c->vLumFilter;
360  int16_t *vChrFilter = c->vChrFilter;
361  int16_t *hLumFilter = c->hLumFilter;
362  int16_t *hChrFilter = c->hChrFilter;
365  const int vLumFilterSize = c->vLumFilterSize;
366  const int vChrFilterSize = c->vChrFilterSize;
367  const int hLumFilterSize = c->hLumFilterSize;
368  const int hChrFilterSize = c->hChrFilterSize;
369  int16_t **lumPixBuf = c->lumPixBuf;
370  int16_t **chrUPixBuf = c->chrUPixBuf;
371  int16_t **chrVPixBuf = c->chrVPixBuf;
372  int16_t **alpPixBuf = c->alpPixBuf;
373  const int vLumBufSize = c->vLumBufSize;
374  const int vChrBufSize = c->vChrBufSize;
376  uint32_t *pal = c->pal_yuv;
383  const int chrSrcSliceY = srcSliceY >> c->chrSrcVSubSample;
384  const int chrSrcSliceH = -((-srcSliceH) >> c->chrSrcVSubSample);
385  int should_dither = is9_OR_10BPS(c->srcFormat) ||
386  is16BPS(c->srcFormat);
387  int lastDstY;
388 
389  /* vars which will change and which we need to store back in the context */
390  int dstY = c->dstY;
391  int lumBufIndex = c->lumBufIndex;
392  int chrBufIndex = c->chrBufIndex;
393  int lastInLumBuf = c->lastInLumBuf;
394  int lastInChrBuf = c->lastInChrBuf;
395 
396  if (isPacked(c->srcFormat)) {
397  src[0] =
398  src[1] =
399  src[2] =
400  src[3] = src[0];
401  srcStride[0] =
402  srcStride[1] =
403  srcStride[2] =
404  srcStride[3] = srcStride[0];
405  }
406  srcStride[1] <<= c->vChrDrop;
407  srcStride[2] <<= c->vChrDrop;
408 
409  DEBUG_BUFFERS("swScale() %p[%d] %p[%d] %p[%d] %p[%d] -> %p[%d] %p[%d] %p[%d] %p[%d]\n",
410  src[0], srcStride[0], src[1], srcStride[1],
411  src[2], srcStride[2], src[3], srcStride[3],
412  dst[0], dstStride[0], dst[1], dstStride[1],
413  dst[2], dstStride[2], dst[3], dstStride[3]);
414  DEBUG_BUFFERS("srcSliceY: %d srcSliceH: %d dstY: %d dstH: %d\n",
415  srcSliceY, srcSliceH, dstY, dstH);
416  DEBUG_BUFFERS("vLumFilterSize: %d vLumBufSize: %d vChrFilterSize: %d vChrBufSize: %d\n",
417  vLumFilterSize, vLumBufSize, vChrFilterSize, vChrBufSize);
418 
419  if (dstStride[0] % 8 != 0 || dstStride[1] % 8 != 0 ||
420  dstStride[2] % 8 != 0 || dstStride[3] % 8 != 0) {
421  static int warnedAlready = 0; // FIXME maybe move this into the context
422  if (flags & SWS_PRINT_INFO && !warnedAlready) {
424  "Warning: dstStride is not aligned!\n"
425  " ->cannot do aligned memory accesses anymore\n");
426  warnedAlready = 1;
427  }
428  }
429 
430  /* Note the user might start scaling the picture in the middle so this
431  * will not get executed. This is not really intended but works
432  * currently, so people might do it. */
433  if (srcSliceY == 0) {
434  lumBufIndex = -1;
435  chrBufIndex = -1;
436  dstY = 0;
437  lastInLumBuf = -1;
438  lastInChrBuf = -1;
439  }
440 
441  if (!should_dither) {
443  }
444  lastDstY = dstY;
445 
446  for (; dstY < dstH; dstY++) {
447  const int chrDstY = dstY >> c->chrDstVSubSample;
448  uint8_t *dest[4] = {
449  dst[0] + dstStride[0] * dstY,
450  dst[1] + dstStride[1] * chrDstY,
451  dst[2] + dstStride[2] * chrDstY,
452  (CONFIG_SWSCALE_ALPHA && alpPixBuf) ? dst[3] + dstStride[3] * dstY : NULL,
453  };
454 
455  // First line needed as input
456  const int firstLumSrcY = FFMAX(1 - vLumFilterSize, vLumFilterPos[dstY]);
457  const int firstLumSrcY2 = FFMAX(1 - vLumFilterSize, vLumFilterPos[FFMIN(dstY | ((1 << c->chrDstVSubSample) - 1), dstH - 1)]);
458  // First line needed as input
459  const int firstChrSrcY = FFMAX(1 - vChrFilterSize, vChrFilterPos[chrDstY]);
460 
461  // Last line needed as input
462  int lastLumSrcY = FFMIN(c->srcH, firstLumSrcY + vLumFilterSize) - 1;
463  int lastLumSrcY2 = FFMIN(c->srcH, firstLumSrcY2 + vLumFilterSize) - 1;
464  int lastChrSrcY = FFMIN(c->chrSrcH, firstChrSrcY + vChrFilterSize) - 1;
465  int enough_lines;
466 
467  // handle holes (FAST_BILINEAR & weird filters)
468  if (firstLumSrcY > lastInLumBuf)
469  lastInLumBuf = firstLumSrcY - 1;
470  if (firstChrSrcY > lastInChrBuf)
471  lastInChrBuf = firstChrSrcY - 1;
472  assert(firstLumSrcY >= lastInLumBuf - vLumBufSize + 1);
473  assert(firstChrSrcY >= lastInChrBuf - vChrBufSize + 1);
474 
475  DEBUG_BUFFERS("dstY: %d\n", dstY);
476  DEBUG_BUFFERS("\tfirstLumSrcY: %d lastLumSrcY: %d lastInLumBuf: %d\n",
477  firstLumSrcY, lastLumSrcY, lastInLumBuf);
478  DEBUG_BUFFERS("\tfirstChrSrcY: %d lastChrSrcY: %d lastInChrBuf: %d\n",
479  firstChrSrcY, lastChrSrcY, lastInChrBuf);
480 
481  // Do we have enough lines in this slice to output the dstY line
482  enough_lines = lastLumSrcY2 < srcSliceY + srcSliceH &&
483  lastChrSrcY < -((-srcSliceY - srcSliceH) >> c->chrSrcVSubSample);
484 
485  if (!enough_lines) {
486  lastLumSrcY = srcSliceY + srcSliceH - 1;
487  lastChrSrcY = chrSrcSliceY + chrSrcSliceH - 1;
488  DEBUG_BUFFERS("buffering slice: lastLumSrcY %d lastChrSrcY %d\n",
489  lastLumSrcY, lastChrSrcY);
490  }
491 
492  // Do horizontal scaling
493  while (lastInLumBuf < lastLumSrcY) {
494  const uint8_t *src1[4] = {
495  src[0] + (lastInLumBuf + 1 - srcSliceY) * srcStride[0],
496  src[1] + (lastInLumBuf + 1 - srcSliceY) * srcStride[1],
497  src[2] + (lastInLumBuf + 1 - srcSliceY) * srcStride[2],
498  src[3] + (lastInLumBuf + 1 - srcSliceY) * srcStride[3],
499  };
500  lumBufIndex++;
501  assert(lumBufIndex < 2 * vLumBufSize);
502  assert(lastInLumBuf + 1 - srcSliceY < srcSliceH);
503  assert(lastInLumBuf + 1 - srcSliceY >= 0);
504  hyscale(c, lumPixBuf[lumBufIndex], dstW, src1, srcW, lumXInc,
505  hLumFilter, hLumFilterPos, hLumFilterSize,
506  formatConvBuffer, pal, 0);
507  if (CONFIG_SWSCALE_ALPHA && alpPixBuf)
508  hyscale(c, alpPixBuf[lumBufIndex], dstW, src1, srcW,
509  lumXInc, hLumFilter, hLumFilterPos, hLumFilterSize,
510  formatConvBuffer, pal, 1);
511  lastInLumBuf++;
512  DEBUG_BUFFERS("\t\tlumBufIndex %d: lastInLumBuf: %d\n",
513  lumBufIndex, lastInLumBuf);
514  }
515  while (lastInChrBuf < lastChrSrcY) {
516  const uint8_t *src1[4] = {
517  src[0] + (lastInChrBuf + 1 - chrSrcSliceY) * srcStride[0],
518  src[1] + (lastInChrBuf + 1 - chrSrcSliceY) * srcStride[1],
519  src[2] + (lastInChrBuf + 1 - chrSrcSliceY) * srcStride[2],
520  src[3] + (lastInChrBuf + 1 - chrSrcSliceY) * srcStride[3],
521  };
522  chrBufIndex++;
523  assert(chrBufIndex < 2 * vChrBufSize);
524  assert(lastInChrBuf + 1 - chrSrcSliceY < (chrSrcSliceH));
525  assert(lastInChrBuf + 1 - chrSrcSliceY >= 0);
526  // FIXME replace parameters through context struct (some at least)
527 
528  if (c->needs_hcscale)
529  hcscale(c, chrUPixBuf[chrBufIndex], chrVPixBuf[chrBufIndex],
530  chrDstW, src1, chrSrcW, chrXInc,
531  hChrFilter, hChrFilterPos, hChrFilterSize,
532  formatConvBuffer, pal);
533  lastInChrBuf++;
534  DEBUG_BUFFERS("\t\tchrBufIndex %d: lastInChrBuf: %d\n",
535  chrBufIndex, lastInChrBuf);
536  }
537  // wrap buf index around to stay inside the ring buffer
538  if (lumBufIndex >= vLumBufSize)
539  lumBufIndex -= vLumBufSize;
540  if (chrBufIndex >= vChrBufSize)
541  chrBufIndex -= vChrBufSize;
542  if (!enough_lines)
543  break; // we can't output a dstY line so let's try with the next slice
544 
545 #if HAVE_MMX_INLINE
546  updateMMXDitherTables(c, dstY, lumBufIndex, chrBufIndex,
547  lastInLumBuf, lastInChrBuf);
548 #endif
549  if (should_dither) {
550  c->chrDither8 = dither_8x8_128[chrDstY & 7];
551  c->lumDither8 = dither_8x8_128[dstY & 7];
552  }
553  if (dstY >= dstH - 2) {
554  /* hmm looks like we can't use MMX here without overwriting
555  * this array's tail */
556  ff_sws_init_output_funcs(c, &yuv2plane1, &yuv2planeX, &yuv2nv12cX,
557  &yuv2packed1, &yuv2packed2, &yuv2packedX);
558  }
559 
560  {
561  const int16_t **lumSrcPtr = (const int16_t **)lumPixBuf + lumBufIndex + firstLumSrcY - lastInLumBuf + vLumBufSize;
562  const int16_t **chrUSrcPtr = (const int16_t **)chrUPixBuf + chrBufIndex + firstChrSrcY - lastInChrBuf + vChrBufSize;
563  const int16_t **chrVSrcPtr = (const int16_t **)chrVPixBuf + chrBufIndex + firstChrSrcY - lastInChrBuf + vChrBufSize;
564  const int16_t **alpSrcPtr = (CONFIG_SWSCALE_ALPHA && alpPixBuf) ?
565  (const int16_t **)alpPixBuf + lumBufIndex + firstLumSrcY - lastInLumBuf + vLumBufSize : NULL;
566 
567  if (firstLumSrcY < 0 || firstLumSrcY + vLumFilterSize > c->srcH) {
568  const int16_t **tmpY = (const int16_t **)lumPixBuf +
569  2 * vLumBufSize;
570  int neg = -firstLumSrcY, i;
571  int end = FFMIN(c->srcH - firstLumSrcY, vLumFilterSize);
572  for (i = 0; i < neg; i++)
573  tmpY[i] = lumSrcPtr[neg];
574  for (; i < end; i++)
575  tmpY[i] = lumSrcPtr[i];
576  for (; i < vLumFilterSize; i++)
577  tmpY[i] = tmpY[i - 1];
578  lumSrcPtr = tmpY;
579 
580  if (alpSrcPtr) {
581  const int16_t **tmpA = (const int16_t **)alpPixBuf +
582  2 * vLumBufSize;
583  for (i = 0; i < neg; i++)
584  tmpA[i] = alpSrcPtr[neg];
585  for (; i < end; i++)
586  tmpA[i] = alpSrcPtr[i];
587  for (; i < vLumFilterSize; i++)
588  tmpA[i] = tmpA[i - 1];
589  alpSrcPtr = tmpA;
590  }
591  }
592  if (firstChrSrcY < 0 ||
593  firstChrSrcY + vChrFilterSize > c->chrSrcH) {
594  const int16_t **tmpU = (const int16_t **)chrUPixBuf + 2 * vChrBufSize,
595  **tmpV = (const int16_t **)chrVPixBuf + 2 * vChrBufSize;
596  int neg = -firstChrSrcY, i;
597  int end = FFMIN(c->chrSrcH - firstChrSrcY, vChrFilterSize);
598  for (i = 0; i < neg; i++) {
599  tmpU[i] = chrUSrcPtr[neg];
600  tmpV[i] = chrVSrcPtr[neg];
601  }
602  for (; i < end; i++) {
603  tmpU[i] = chrUSrcPtr[i];
604  tmpV[i] = chrVSrcPtr[i];
605  }
606  for (; i < vChrFilterSize; i++) {
607  tmpU[i] = tmpU[i - 1];
608  tmpV[i] = tmpV[i - 1];
609  }
610  chrUSrcPtr = tmpU;
611  chrVSrcPtr = tmpV;
612  }
613 
614  if (isPlanarYUV(dstFormat) ||
615  (isGray(dstFormat) && !isALPHA(dstFormat))) { // YV12 like
616  const int chrSkipMask = (1 << c->chrDstVSubSample) - 1;
617 
618  if (vLumFilterSize == 1) {
619  yuv2plane1(lumSrcPtr[0], dest[0], dstW, c->lumDither8, 0);
620  } else {
621  yuv2planeX(vLumFilter + dstY * vLumFilterSize,
622  vLumFilterSize, lumSrcPtr, dest[0],
623  dstW, c->lumDither8, 0);
624  }
625 
626  if (!((dstY & chrSkipMask) || isGray(dstFormat))) {
627  if (yuv2nv12cX) {
628  yuv2nv12cX(c, vChrFilter + chrDstY * vChrFilterSize,
629  vChrFilterSize, chrUSrcPtr, chrVSrcPtr,
630  dest[1], chrDstW);
631  } else if (vChrFilterSize == 1) {
632  yuv2plane1(chrUSrcPtr[0], dest[1], chrDstW, c->chrDither8, 0);
633  yuv2plane1(chrVSrcPtr[0], dest[2], chrDstW, c->chrDither8, 3);
634  } else {
635  yuv2planeX(vChrFilter + chrDstY * vChrFilterSize,
636  vChrFilterSize, chrUSrcPtr, dest[1],
637  chrDstW, c->chrDither8, 0);
638  yuv2planeX(vChrFilter + chrDstY * vChrFilterSize,
639  vChrFilterSize, chrVSrcPtr, dest[2],
640  chrDstW, c->chrDither8, 3);
641  }
642  }
643 
644  if (CONFIG_SWSCALE_ALPHA && alpPixBuf) {
645  if (vLumFilterSize == 1) {
646  yuv2plane1(alpSrcPtr[0], dest[3], dstW,
647  c->lumDither8, 0);
648  } else {
649  yuv2planeX(vLumFilter + dstY * vLumFilterSize,
650  vLumFilterSize, alpSrcPtr, dest[3],
651  dstW, c->lumDither8, 0);
652  }
653  }
654  } else {
655  if (c->yuv2packed1 && vLumFilterSize == 1 &&
656  vChrFilterSize <= 2) { // unscaled RGB
657  int chrAlpha = vChrFilterSize == 1 ? 0 : vChrFilter[2 * dstY + 1];
658  yuv2packed1(c, *lumSrcPtr, chrUSrcPtr, chrVSrcPtr,
659  alpPixBuf ? *alpSrcPtr : NULL,
660  dest[0], dstW, chrAlpha, dstY);
661  } else if (c->yuv2packed2 && vLumFilterSize == 2 &&
662  vChrFilterSize == 2) { // bilinear upscale RGB
663  int lumAlpha = vLumFilter[2 * dstY + 1];
664  int chrAlpha = vChrFilter[2 * dstY + 1];
665  lumMmxFilter[2] =
666  lumMmxFilter[3] = vLumFilter[2 * dstY] * 0x10001;
667  chrMmxFilter[2] =
668  chrMmxFilter[3] = vChrFilter[2 * chrDstY] * 0x10001;
669  yuv2packed2(c, lumSrcPtr, chrUSrcPtr, chrVSrcPtr,
670  alpPixBuf ? alpSrcPtr : NULL,
671  dest[0], dstW, lumAlpha, chrAlpha, dstY);
672  } else { // general RGB
673  yuv2packedX(c, vLumFilter + dstY * vLumFilterSize,
674  lumSrcPtr, vLumFilterSize,
675  vChrFilter + dstY * vChrFilterSize,
676  chrUSrcPtr, chrVSrcPtr, vChrFilterSize,
677  alpSrcPtr, dest[0], dstW, dstY);
678  }
679  }
680  }
681  }
682 
683  if (isPlanar(dstFormat) && isALPHA(dstFormat) && !alpPixBuf) {
684  int length = dstW;
685  int height = dstY - lastDstY;
686  if (is16BPS(c->dstFormat))
687  length *= 2;
688 
689  if (is9_OR_10BPS(dstFormat)) {
690  const AVPixFmtDescriptor *desc = av_pix_fmt_desc_get(dstFormat);
691  fill_plane9or10(dst[3], dstStride[3], length, height, lastDstY,
692  255, desc->comp[3].depth_minus1 + 1,
693  isBE(dstFormat));
694  } else
695  fillPlane(dst[3], dstStride[3], length, height, lastDstY, 255);
696  }
697 
698 #if HAVE_MMXEXT_INLINE
700  __asm__ volatile ("sfence" ::: "memory");
701 #endif
702  emms_c();
703 
704  /* store changed local vars back in the context */
705  c->dstY = dstY;
710 
711  return dstY - lastDstY;
712 }
713 
715 {
717 
719  &c->yuv2nv12cX, &c->yuv2packed1,
720  &c->yuv2packed2, &c->yuv2packedX);
721 
723 
724  if (c->srcBpc == 8) {
725  if (c->dstBpc <= 10) {
726  c->hyScale = c->hcScale = hScale8To15_c;
727  if (c->flags & SWS_FAST_BILINEAR) {
730  }
731  } else {
732  c->hyScale = c->hcScale = hScale8To19_c;
733  }
734  } else {
735  c->hyScale = c->hcScale = c->dstBpc > 10 ? hScale16To19_c
736  : hScale16To15_c;
737  }
738 
739  if (c->srcRange != c->dstRange && !isAnyRGB(c->dstFormat)) {
740  if (c->dstBpc <= 10) {
741  if (c->srcRange) {
744  } else {
747  }
748  } else {
749  if (c->srcRange) {
752  } else {
755  }
756  }
757  }
758 
759  if (!(isGray(srcFormat) || isGray(c->dstFormat) ||
760  srcFormat == AV_PIX_FMT_MONOBLACK || srcFormat == AV_PIX_FMT_MONOWHITE))
761  c->needs_hcscale = 1;
762 }
763 
765 {
767 
768  if (HAVE_MMX)
770  if (HAVE_ALTIVEC)
772 
773  return swScale;
774 }