2 * Copyright (C) 2001-2003 Michael Niedermayer <michaelni@gmx.at>
4 * This file is part of FFmpeg.
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.
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.
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
23 #define _SVID_SOURCE // needed for MAP_ANONYMOUS
24 #define _DARWIN_C_SOURCE // needed for MAP_ANON
31 #if defined(MAP_ANON) && !defined(MAP_ANONYMOUS)
32 #define MAP_ANONYMOUS MAP_ANON
36 #define WIN32_LEAN_AND_MEAN
40 #include "libavutil/attributes.h"
41 #include "libavutil/avassert.h"
42 #include "libavutil/avutil.h"
43 #include "libavutil/bswap.h"
44 #include "libavutil/cpu.h"
45 #include "libavutil/intreadwrite.h"
46 #include "libavutil/mathematics.h"
47 #include "libavutil/opt.h"
48 #include "libavutil/pixdesc.h"
49 #include "libavutil/x86/asm.h"
50 #include "libavutil/x86/cpu.h"
53 #include "swscale_internal.h"
55 unsigned swscale_version(void)
57 av_assert0(LIBSWSCALE_VERSION_MICRO >= 100);
58 return LIBSWSCALE_VERSION_INT;
61 const char *swscale_configuration(void)
63 return FFMPEG_CONFIGURATION;
66 const char *swscale_license(void)
68 #define LICENSE_PREFIX "libswscale license: "
69 return LICENSE_PREFIX FFMPEG_LICENSE + sizeof(LICENSE_PREFIX) - 1;
72 #define RET 0xC3 // near return opcode for x86
74 typedef struct FormatEntry {
75 int is_supported_in, is_supported_out;
78 static const FormatEntry format_entries[AV_PIX_FMT_NB] = {
79 [AV_PIX_FMT_YUV420P] = { 1, 1 },
80 [AV_PIX_FMT_YUYV422] = { 1, 1 },
81 [AV_PIX_FMT_RGB24] = { 1, 1 },
82 [AV_PIX_FMT_BGR24] = { 1, 1 },
83 [AV_PIX_FMT_YUV422P] = { 1, 1 },
84 [AV_PIX_FMT_YUV444P] = { 1, 1 },
85 [AV_PIX_FMT_YUV410P] = { 1, 1 },
86 [AV_PIX_FMT_YUV411P] = { 1, 1 },
87 [AV_PIX_FMT_GRAY8] = { 1, 1 },
88 [AV_PIX_FMT_MONOWHITE] = { 1, 1 },
89 [AV_PIX_FMT_MONOBLACK] = { 1, 1 },
90 [AV_PIX_FMT_PAL8] = { 1, 0 },
91 [AV_PIX_FMT_YUVJ420P] = { 1, 1 },
92 [AV_PIX_FMT_YUVJ422P] = { 1, 1 },
93 [AV_PIX_FMT_YUVJ444P] = { 1, 1 },
94 [AV_PIX_FMT_UYVY422] = { 1, 1 },
95 [AV_PIX_FMT_UYYVYY411] = { 0, 0 },
96 [AV_PIX_FMT_BGR8] = { 1, 1 },
97 [AV_PIX_FMT_BGR4] = { 0, 1 },
98 [AV_PIX_FMT_BGR4_BYTE] = { 1, 1 },
99 [AV_PIX_FMT_RGB8] = { 1, 1 },
100 [AV_PIX_FMT_RGB4] = { 0, 1 },
101 [AV_PIX_FMT_RGB4_BYTE] = { 1, 1 },
102 [AV_PIX_FMT_NV12] = { 1, 1 },
103 [AV_PIX_FMT_NV21] = { 1, 1 },
104 [AV_PIX_FMT_ARGB] = { 1, 1 },
105 [AV_PIX_FMT_RGBA] = { 1, 1 },
106 [AV_PIX_FMT_ABGR] = { 1, 1 },
107 [AV_PIX_FMT_BGRA] = { 1, 1 },
108 [AV_PIX_FMT_0RGB] = { 1, 1 },
109 [AV_PIX_FMT_RGB0] = { 1, 1 },
110 [AV_PIX_FMT_0BGR] = { 1, 1 },
111 [AV_PIX_FMT_BGR0] = { 1, 1 },
112 [AV_PIX_FMT_GRAY16BE] = { 1, 1 },
113 [AV_PIX_FMT_GRAY16LE] = { 1, 1 },
114 [AV_PIX_FMT_YUV440P] = { 1, 1 },
115 [AV_PIX_FMT_YUVJ440P] = { 1, 1 },
116 [AV_PIX_FMT_YUVA420P] = { 1, 1 },
117 [AV_PIX_FMT_YUVA422P] = { 1, 1 },
118 [AV_PIX_FMT_YUVA444P] = { 1, 1 },
119 [AV_PIX_FMT_RGB48BE] = { 1, 1 },
120 [AV_PIX_FMT_RGB48LE] = { 1, 1 },
121 [AV_PIX_FMT_RGBA64BE] = { 1, 0 },
122 [AV_PIX_FMT_RGBA64LE] = { 1, 0 },
123 [AV_PIX_FMT_RGB565BE] = { 1, 1 },
124 [AV_PIX_FMT_RGB565LE] = { 1, 1 },
125 [AV_PIX_FMT_RGB555BE] = { 1, 1 },
126 [AV_PIX_FMT_RGB555LE] = { 1, 1 },
127 [AV_PIX_FMT_BGR565BE] = { 1, 1 },
128 [AV_PIX_FMT_BGR565LE] = { 1, 1 },
129 [AV_PIX_FMT_BGR555BE] = { 1, 1 },
130 [AV_PIX_FMT_BGR555LE] = { 1, 1 },
131 [AV_PIX_FMT_YUV420P16LE] = { 1, 1 },
132 [AV_PIX_FMT_YUV420P16BE] = { 1, 1 },
133 [AV_PIX_FMT_YUV422P16LE] = { 1, 1 },
134 [AV_PIX_FMT_YUV422P16BE] = { 1, 1 },
135 [AV_PIX_FMT_YUV444P16LE] = { 1, 1 },
136 [AV_PIX_FMT_YUV444P16BE] = { 1, 1 },
137 [AV_PIX_FMT_RGB444LE] = { 1, 1 },
138 [AV_PIX_FMT_RGB444BE] = { 1, 1 },
139 [AV_PIX_FMT_BGR444LE] = { 1, 1 },
140 [AV_PIX_FMT_BGR444BE] = { 1, 1 },
141 [AV_PIX_FMT_Y400A] = { 1, 0 },
142 [AV_PIX_FMT_BGR48BE] = { 1, 1 },
143 [AV_PIX_FMT_BGR48LE] = { 1, 1 },
144 [AV_PIX_FMT_BGRA64BE] = { 0, 0 },
145 [AV_PIX_FMT_BGRA64LE] = { 0, 0 },
146 [AV_PIX_FMT_YUV420P9BE] = { 1, 1 },
147 [AV_PIX_FMT_YUV420P9LE] = { 1, 1 },
148 [AV_PIX_FMT_YUV420P10BE] = { 1, 1 },
149 [AV_PIX_FMT_YUV420P10LE] = { 1, 1 },
150 [AV_PIX_FMT_YUV420P12BE] = { 1, 1 },
151 [AV_PIX_FMT_YUV420P12LE] = { 1, 1 },
152 [AV_PIX_FMT_YUV420P14BE] = { 1, 1 },
153 [AV_PIX_FMT_YUV420P14LE] = { 1, 1 },
154 [AV_PIX_FMT_YUV422P9BE] = { 1, 1 },
155 [AV_PIX_FMT_YUV422P9LE] = { 1, 1 },
156 [AV_PIX_FMT_YUV422P10BE] = { 1, 1 },
157 [AV_PIX_FMT_YUV422P10LE] = { 1, 1 },
158 [AV_PIX_FMT_YUV422P12BE] = { 1, 1 },
159 [AV_PIX_FMT_YUV422P12LE] = { 1, 1 },
160 [AV_PIX_FMT_YUV422P14BE] = { 1, 1 },
161 [AV_PIX_FMT_YUV422P14LE] = { 1, 1 },
162 [AV_PIX_FMT_YUV444P9BE] = { 1, 1 },
163 [AV_PIX_FMT_YUV444P9LE] = { 1, 1 },
164 [AV_PIX_FMT_YUV444P10BE] = { 1, 1 },
165 [AV_PIX_FMT_YUV444P10LE] = { 1, 1 },
166 [AV_PIX_FMT_YUV444P12BE] = { 1, 1 },
167 [AV_PIX_FMT_YUV444P12LE] = { 1, 1 },
168 [AV_PIX_FMT_YUV444P14BE] = { 1, 1 },
169 [AV_PIX_FMT_YUV444P14LE] = { 1, 1 },
170 [AV_PIX_FMT_GBRP] = { 1, 0 },
171 [AV_PIX_FMT_GBRP9LE] = { 1, 0 },
172 [AV_PIX_FMT_GBRP9BE] = { 1, 0 },
173 [AV_PIX_FMT_GBRP10LE] = { 1, 0 },
174 [AV_PIX_FMT_GBRP10BE] = { 1, 0 },
175 [AV_PIX_FMT_GBRP12LE] = { 1, 0 },
176 [AV_PIX_FMT_GBRP12BE] = { 1, 0 },
177 [AV_PIX_FMT_GBRP14LE] = { 1, 0 },
178 [AV_PIX_FMT_GBRP14BE] = { 1, 0 },
179 [AV_PIX_FMT_GBRP16LE] = { 1, 0 },
180 [AV_PIX_FMT_GBRP16BE] = { 1, 0 },
183 int sws_isSupportedInput(enum AVPixelFormat pix_fmt)
185 return (unsigned)pix_fmt < AV_PIX_FMT_NB ?
186 format_entries[pix_fmt].is_supported_in : 0;
189 int sws_isSupportedOutput(enum AVPixelFormat pix_fmt)
191 return (unsigned)pix_fmt < AV_PIX_FMT_NB ?
192 format_entries[pix_fmt].is_supported_out : 0;
195 extern const int32_t ff_yuv2rgb_coeffs[8][4];
197 #if FF_API_SWS_FORMAT_NAME
198 const char *sws_format_name(enum AVPixelFormat format)
200 const AVPixFmtDescriptor *desc = av_pix_fmt_desc_get(format);
204 return "Unknown format";
208 static double getSplineCoeff(double a, double b, double c, double d,
212 return ((d * dist + c) * dist + b) * dist + a;
214 return getSplineCoeff(0.0,
215 b + 2.0 * c + 3.0 * d,
217 -b - 3.0 * c - 6.0 * d,
221 static int initFilter(int16_t **outFilter, int32_t **filterPos,
222 int *outFilterSize, int xInc, int srcW, int dstW,
223 int filterAlign, int one, int flags, int cpu_flags,
224 SwsVector *srcFilter, SwsVector *dstFilter,
231 int64_t *filter = NULL;
232 int64_t *filter2 = NULL;
233 const int64_t fone = 1LL << 54;
236 emms_c(); // FIXME should not be required but IS (even for non-MMX versions)
238 // NOTE: the +3 is for the MMX(+1) / SSE(+3) scaler which reads over the end
239 FF_ALLOC_OR_GOTO(NULL, *filterPos, (dstW + 3) * sizeof(**filterPos), fail);
241 if (FFABS(xInc - 0x10000) < 10) { // unscaled
244 FF_ALLOCZ_OR_GOTO(NULL, filter,
245 dstW * sizeof(*filter) * filterSize, fail);
247 for (i = 0; i < dstW; i++) {
248 filter[i * filterSize] = fone;
251 } else if (flags & SWS_POINT) { // lame looking point sampling mode
255 FF_ALLOC_OR_GOTO(NULL, filter,
256 dstW * sizeof(*filter) * filterSize, fail);
258 xDstInSrc = xInc / 2 - 0x8000;
259 for (i = 0; i < dstW; i++) {
260 int xx = (xDstInSrc - ((filterSize - 1) << 15) + (1 << 15)) >> 16;
262 (*filterPos)[i] = xx;
266 } else if ((xInc <= (1 << 16) && (flags & SWS_AREA)) ||
267 (flags & SWS_FAST_BILINEAR)) { // bilinear upscale
271 FF_ALLOC_OR_GOTO(NULL, filter,
272 dstW * sizeof(*filter) * filterSize, fail);
274 xDstInSrc = xInc / 2 - 0x8000;
275 for (i = 0; i < dstW; i++) {
276 int xx = (xDstInSrc - ((filterSize - 1) << 15) + (1 << 15)) >> 16;
279 (*filterPos)[i] = xx;
280 // bilinear upscale / linear interpolate / area averaging
281 for (j = 0; j < filterSize; j++) {
282 int64_t coeff= fone - FFABS(((int64_t)xx<<16) - xDstInSrc)*(fone>>16);
285 filter[i * filterSize + j] = coeff;
294 if (flags & SWS_BICUBIC)
296 else if (flags & SWS_X)
298 else if (flags & SWS_AREA)
299 sizeFactor = 1; // downscale only, for upscale it is bilinear
300 else if (flags & SWS_GAUSS)
301 sizeFactor = 8; // infinite ;)
302 else if (flags & SWS_LANCZOS)
303 sizeFactor = param[0] != SWS_PARAM_DEFAULT ? ceil(2 * param[0]) : 6;
304 else if (flags & SWS_SINC)
305 sizeFactor = 20; // infinite ;)
306 else if (flags & SWS_SPLINE)
307 sizeFactor = 20; // infinite ;)
308 else if (flags & SWS_BILINEAR)
315 filterSize = 1 + sizeFactor; // upscale
317 filterSize = 1 + (sizeFactor * srcW + dstW - 1) / dstW;
319 filterSize = FFMIN(filterSize, srcW - 2);
320 filterSize = FFMAX(filterSize, 1);
322 FF_ALLOC_OR_GOTO(NULL, filter,
323 dstW * sizeof(*filter) * filterSize, fail);
325 xDstInSrc = xInc - 0x10000;
326 for (i = 0; i < dstW; i++) {
327 int xx = (xDstInSrc - ((filterSize - 2) << 16)) / (1 << 17);
329 (*filterPos)[i] = xx;
330 for (j = 0; j < filterSize; j++) {
331 int64_t d = (FFABS(((int64_t)xx << 17) - xDstInSrc)) << 13;
337 floatd = d * (1.0 / (1 << 30));
339 if (flags & SWS_BICUBIC) {
340 int64_t B = (param[0] != SWS_PARAM_DEFAULT ? param[0] : 0) * (1 << 24);
341 int64_t C = (param[1] != SWS_PARAM_DEFAULT ? param[1] : 0.6) * (1 << 24);
343 if (d >= 1LL << 31) {
346 int64_t dd = (d * d) >> 30;
347 int64_t ddd = (dd * d) >> 30;
350 coeff = (12 * (1 << 24) - 9 * B - 6 * C) * ddd +
351 (-18 * (1 << 24) + 12 * B + 6 * C) * dd +
352 (6 * (1 << 24) - 2 * B) * (1 << 30);
354 coeff = (-B - 6 * C) * ddd +
355 (6 * B + 30 * C) * dd +
356 (-12 * B - 48 * C) * d +
357 (8 * B + 24 * C) * (1 << 30);
359 coeff *= fone >> (30 + 24);
362 else if (flags & SWS_X) {
363 double p = param ? param * 0.01 : 0.3;
364 coeff = d ? sin(d * M_PI) / (d * M_PI) : 1.0;
365 coeff *= pow(2.0, -p * d * d);
368 else if (flags & SWS_X) {
369 double A = param[0] != SWS_PARAM_DEFAULT ? param[0] : 1.0;
373 c = cos(floatd * M_PI);
380 coeff = (c * 0.5 + 0.5) * fone;
381 } else if (flags & SWS_AREA) {
382 int64_t d2 = d - (1 << 29);
383 if (d2 * xInc < -(1LL << (29 + 16)))
384 coeff = 1.0 * (1LL << (30 + 16));
385 else if (d2 * xInc < (1LL << (29 + 16)))
386 coeff = -d2 * xInc + (1LL << (29 + 16));
389 coeff *= fone >> (30 + 16);
390 } else if (flags & SWS_GAUSS) {
391 double p = param[0] != SWS_PARAM_DEFAULT ? param[0] : 3.0;
392 coeff = (pow(2.0, -p * floatd * floatd)) * fone;
393 } else if (flags & SWS_SINC) {
394 coeff = (d ? sin(floatd * M_PI) / (floatd * M_PI) : 1.0) * fone;
395 } else if (flags & SWS_LANCZOS) {
396 double p = param[0] != SWS_PARAM_DEFAULT ? param[0] : 3.0;
397 coeff = (d ? sin(floatd * M_PI) * sin(floatd * M_PI / p) /
398 (floatd * floatd * M_PI * M_PI / p) : 1.0) * fone;
401 } else if (flags & SWS_BILINEAR) {
402 coeff = (1 << 30) - d;
406 } else if (flags & SWS_SPLINE) {
407 double p = -2.196152422706632;
408 coeff = getSplineCoeff(1.0, 0.0, p, -p - 1.0, floatd) * fone;
413 filter[i * filterSize + j] = coeff;
416 xDstInSrc += 2 * xInc;
420 /* apply src & dst Filter to filter -> filter2
423 av_assert0(filterSize > 0);
424 filter2Size = filterSize;
426 filter2Size += srcFilter->length - 1;
428 filter2Size += dstFilter->length - 1;
429 av_assert0(filter2Size > 0);
430 FF_ALLOCZ_OR_GOTO(NULL, filter2, filter2Size * dstW * sizeof(*filter2), fail);
432 for (i = 0; i < dstW; i++) {
436 for (k = 0; k < srcFilter->length; k++) {
437 for (j = 0; j < filterSize; j++)
438 filter2[i * filter2Size + k + j] +=
439 srcFilter->coeff[k] * filter[i * filterSize + j];
442 for (j = 0; j < filterSize; j++)
443 filter2[i * filter2Size + j] = filter[i * filterSize + j];
447 (*filterPos)[i] += (filterSize - 1) / 2 - (filter2Size - 1) / 2;
451 /* try to reduce the filter-size (step1 find size and shift left) */
452 // Assume it is near normalized (*0.5 or *2.0 is OK but * 0.001 is not).
454 for (i = dstW - 1; i >= 0; i--) {
455 int min = filter2Size;
457 int64_t cutOff = 0.0;
459 /* get rid of near zero elements on the left by shifting left */
460 for (j = 0; j < filter2Size; j++) {
462 cutOff += FFABS(filter2[i * filter2Size]);
464 if (cutOff > SWS_MAX_REDUCE_CUTOFF * fone)
467 /* preserve monotonicity because the core can't handle the
468 * filter otherwise */
469 if (i < dstW - 1 && (*filterPos)[i] >= (*filterPos)[i + 1])
472 // move filter coefficients left
473 for (k = 1; k < filter2Size; k++)
474 filter2[i * filter2Size + k - 1] = filter2[i * filter2Size + k];
475 filter2[i * filter2Size + k - 1] = 0;
480 /* count near zeros on the right */
481 for (j = filter2Size - 1; j > 0; j--) {
482 cutOff += FFABS(filter2[i * filter2Size + j]);
484 if (cutOff > SWS_MAX_REDUCE_CUTOFF * fone)
489 if (min > minFilterSize)
493 if (HAVE_ALTIVEC && cpu_flags & AV_CPU_FLAG_ALTIVEC) {
494 // we can handle the special case 4, so we don't want to go the full 8
495 if (minFilterSize < 5)
498 /* We really don't want to waste our time doing useless computation, so
499 * fall back on the scalar C code for very small filters.
500 * Vectorizing is worth it only if you have a decent-sized vector. */
501 if (minFilterSize < 3)
505 if (INLINE_MMX(cpu_flags)) {
506 // special case for unscaled vertical filtering
507 if (minFilterSize == 1 && filterAlign == 2)
511 av_assert0(minFilterSize > 0);
512 filterSize = (minFilterSize + (filterAlign - 1)) & (~(filterAlign - 1));
513 av_assert0(filterSize > 0);
514 filter = av_malloc(filterSize * dstW * sizeof(*filter));
515 if (filterSize >= MAX_FILTER_SIZE * 16 /
516 ((flags & SWS_ACCURATE_RND) ? APCK_SIZE : 16) || !filter)
518 *outFilterSize = filterSize;
520 if (flags & SWS_PRINT_INFO)
521 av_log(NULL, AV_LOG_VERBOSE,
522 "SwScaler: reducing / aligning filtersize %d -> %d\n",
523 filter2Size, filterSize);
524 /* try to reduce the filter-size (step2 reduce it) */
525 for (i = 0; i < dstW; i++) {
528 for (j = 0; j < filterSize; j++) {
529 if (j >= filter2Size)
530 filter[i * filterSize + j] = 0;
532 filter[i * filterSize + j] = filter2[i * filter2Size + j];
533 if ((flags & SWS_BITEXACT) && j >= minFilterSize)
534 filter[i * filterSize + j] = 0;
538 // FIXME try to align filterPos if possible
541 for (i = 0; i < dstW; i++) {
543 if ((*filterPos)[i] < 0) {
544 // move filter coefficients left to compensate for filterPos
545 for (j = 1; j < filterSize; j++) {
546 int left = FFMAX(j + (*filterPos)[i], 0);
547 filter[i * filterSize + left] += filter[i * filterSize + j];
548 filter[i * filterSize + j] = 0;
553 if ((*filterPos)[i] + filterSize > srcW) {
554 int shift = (*filterPos)[i] + filterSize - srcW;
555 // move filter coefficients right to compensate for filterPos
556 for (j = filterSize - 2; j >= 0; j--) {
557 int right = FFMIN(j + shift, filterSize - 1);
558 filter[i * filterSize + right] += filter[i * filterSize + j];
559 filter[i * filterSize + j] = 0;
561 (*filterPos)[i]= srcW - filterSize;
565 // Note the +1 is for the MMX scaler which reads over the end
566 /* align at 16 for AltiVec (needed by hScale_altivec_real) */
567 FF_ALLOCZ_OR_GOTO(NULL, *outFilter,
568 *outFilterSize * (dstW + 3) * sizeof(int16_t), fail);
570 /* normalize & store in outFilter */
571 for (i = 0; i < dstW; i++) {
576 for (j = 0; j < filterSize; j++) {
577 sum += filter[i * filterSize + j];
579 sum = (sum + one / 2) / one;
580 for (j = 0; j < *outFilterSize; j++) {
581 int64_t v = filter[i * filterSize + j] + error;
582 int intV = ROUNDED_DIV(v, sum);
583 (*outFilter)[i * (*outFilterSize) + j] = intV;
584 error = v - intV * sum;
588 (*filterPos)[dstW + 0] =
589 (*filterPos)[dstW + 1] =
590 (*filterPos)[dstW + 2] = (*filterPos)[dstW - 1]; /* the MMX/SSE scaler will
591 * read over the end */
592 for (i = 0; i < *outFilterSize; i++) {
593 int k = (dstW - 1) * (*outFilterSize) + i;
594 (*outFilter)[k + 1 * (*outFilterSize)] =
595 (*outFilter)[k + 2 * (*outFilterSize)] =
596 (*outFilter)[k + 3 * (*outFilterSize)] = (*outFilter)[k];
607 #if HAVE_MMXEXT_INLINE
608 static int initMMX2HScaler(int dstW, int xInc, uint8_t *filterCode,
609 int16_t *filter, int32_t *filterPos, int numSplits)
612 x86_reg imm8OfPShufW1A;
613 x86_reg imm8OfPShufW2A;
614 x86_reg fragmentLengthA;
616 x86_reg imm8OfPShufW1B;
617 x86_reg imm8OfPShufW2B;
618 x86_reg fragmentLengthB;
623 // create an optimized horizontal scaling routine
624 /* This scaler is made of runtime-generated MMX2 code using specially tuned
625 * pshufw instructions. For every four output pixels, if four input pixels
626 * are enough for the fast bilinear scaling, then a chunk of fragmentB is
627 * used. If five input pixels are needed, then a chunk of fragmentA is used.
636 "movq (%%"REG_d", %%"REG_a"), %%mm3 \n\t"
637 "movd (%%"REG_c", %%"REG_S"), %%mm0 \n\t"
638 "movd 1(%%"REG_c", %%"REG_S"), %%mm1 \n\t"
639 "punpcklbw %%mm7, %%mm1 \n\t"
640 "punpcklbw %%mm7, %%mm0 \n\t"
641 "pshufw $0xFF, %%mm1, %%mm1 \n\t"
643 "pshufw $0xFF, %%mm0, %%mm0 \n\t"
645 "psubw %%mm1, %%mm0 \n\t"
646 "movl 8(%%"REG_b", %%"REG_a"), %%esi \n\t"
647 "pmullw %%mm3, %%mm0 \n\t"
648 "psllw $7, %%mm1 \n\t"
649 "paddw %%mm1, %%mm0 \n\t"
651 "movq %%mm0, (%%"REG_D", %%"REG_a") \n\t"
653 "add $8, %%"REG_a" \n\t"
657 "lea " LOCAL_MANGLE(0b) ", %0 \n\t"
658 "lea " LOCAL_MANGLE(1b) ", %1 \n\t"
659 "lea " LOCAL_MANGLE(2b) ", %2 \n\t"
664 "lea " LOCAL_MANGLE(9b) ", %3 \n\t"
668 : "=r" (fragmentA), "=r" (imm8OfPShufW1A), "=r" (imm8OfPShufW2A),
669 "=r" (fragmentLengthA)
676 "movq (%%"REG_d", %%"REG_a"), %%mm3 \n\t"
677 "movd (%%"REG_c", %%"REG_S"), %%mm0 \n\t"
678 "punpcklbw %%mm7, %%mm0 \n\t"
679 "pshufw $0xFF, %%mm0, %%mm1 \n\t"
681 "pshufw $0xFF, %%mm0, %%mm0 \n\t"
683 "psubw %%mm1, %%mm0 \n\t"
684 "movl 8(%%"REG_b", %%"REG_a"), %%esi \n\t"
685 "pmullw %%mm3, %%mm0 \n\t"
686 "psllw $7, %%mm1 \n\t"
687 "paddw %%mm1, %%mm0 \n\t"
689 "movq %%mm0, (%%"REG_D", %%"REG_a") \n\t"
691 "add $8, %%"REG_a" \n\t"
695 "lea " LOCAL_MANGLE(0b) ", %0 \n\t"
696 "lea " LOCAL_MANGLE(1b) ", %1 \n\t"
697 "lea " LOCAL_MANGLE(2b) ", %2 \n\t"
702 "lea " LOCAL_MANGLE(9b) ", %3 \n\t"
706 : "=r" (fragmentB), "=r" (imm8OfPShufW1B), "=r" (imm8OfPShufW2B),
707 "=r" (fragmentLengthB)
710 xpos = 0; // lumXInc/2 - 0x8000; // difference between pixel centers
713 for (i = 0; i < dstW / numSplits; i++) {
718 int b = ((xpos + xInc) >> 16) - xx;
719 int c = ((xpos + xInc * 2) >> 16) - xx;
720 int d = ((xpos + xInc * 3) >> 16) - xx;
721 int inc = (d + 1 < 4);
722 uint8_t *fragment = (d + 1 < 4) ? fragmentB : fragmentA;
723 x86_reg imm8OfPShufW1 = (d + 1 < 4) ? imm8OfPShufW1B : imm8OfPShufW1A;
724 x86_reg imm8OfPShufW2 = (d + 1 < 4) ? imm8OfPShufW2B : imm8OfPShufW2A;
725 x86_reg fragmentLength = (d + 1 < 4) ? fragmentLengthB : fragmentLengthA;
726 int maxShift = 3 - (d + inc);
730 filter[i] = ((xpos & 0xFFFF) ^ 0xFFFF) >> 9;
731 filter[i + 1] = (((xpos + xInc) & 0xFFFF) ^ 0xFFFF) >> 9;
732 filter[i + 2] = (((xpos + xInc * 2) & 0xFFFF) ^ 0xFFFF) >> 9;
733 filter[i + 3] = (((xpos + xInc * 3) & 0xFFFF) ^ 0xFFFF) >> 9;
734 filterPos[i / 2] = xx;
736 memcpy(filterCode + fragmentPos, fragment, fragmentLength);
738 filterCode[fragmentPos + imm8OfPShufW1] = (a + inc) |
742 filterCode[fragmentPos + imm8OfPShufW2] = a | (b << 2) |
746 if (i + 4 - inc >= dstW)
747 shift = maxShift; // avoid overread
748 else if ((filterPos[i / 2] & 3) <= maxShift)
749 shift = filterPos[i / 2] & 3; // align
751 if (shift && i >= shift) {
752 filterCode[fragmentPos + imm8OfPShufW1] += 0x55 * shift;
753 filterCode[fragmentPos + imm8OfPShufW2] += 0x55 * shift;
754 filterPos[i / 2] -= shift;
758 fragmentPos += fragmentLength;
761 filterCode[fragmentPos] = RET;
766 filterPos[((i / 2) + 1) & (~1)] = xpos >> 16; // needed to jump to the next part
768 return fragmentPos + 1;
770 #endif /* HAVE_MMXEXT_INLINE */
772 static void getSubSampleFactors(int *h, int *v, enum AVPixelFormat format)
774 const AVPixFmtDescriptor *desc = av_pix_fmt_desc_get(format);
775 *h = desc->log2_chroma_w;
776 *v = desc->log2_chroma_h;
779 int sws_setColorspaceDetails(struct SwsContext *c, const int inv_table[4],
780 int srcRange, const int table[4], int dstRange,
781 int brightness, int contrast, int saturation)
783 const AVPixFmtDescriptor *desc_dst = av_pix_fmt_desc_get(c->dstFormat);
784 const AVPixFmtDescriptor *desc_src = av_pix_fmt_desc_get(c->srcFormat);
785 memcpy(c->srcColorspaceTable, inv_table, sizeof(int) * 4);
786 memcpy(c->dstColorspaceTable, table, sizeof(int) * 4);
788 c->brightness = brightness;
789 c->contrast = contrast;
790 c->saturation = saturation;
791 c->srcRange = srcRange;
792 c->dstRange = dstRange;
793 if (isYUV(c->dstFormat) || isGray(c->dstFormat))
796 c->dstFormatBpp = av_get_bits_per_pixel(desc_dst);
797 c->srcFormatBpp = av_get_bits_per_pixel(desc_src);
799 ff_yuv2rgb_c_init_tables(c, inv_table, srcRange, brightness,
800 contrast, saturation);
803 if (HAVE_ALTIVEC && av_get_cpu_flags() & AV_CPU_FLAG_ALTIVEC)
804 ff_yuv2rgb_init_tables_altivec(c, inv_table, brightness,
805 contrast, saturation);
809 int sws_getColorspaceDetails(struct SwsContext *c, int **inv_table,
810 int *srcRange, int **table, int *dstRange,
811 int *brightness, int *contrast, int *saturation)
813 if (!c || isYUV(c->dstFormat) || isGray(c->dstFormat))
816 *inv_table = c->srcColorspaceTable;
817 *table = c->dstColorspaceTable;
818 *srcRange = c->srcRange;
819 *dstRange = c->dstRange;
820 *brightness = c->brightness;
821 *contrast = c->contrast;
822 *saturation = c->saturation;
827 static int handle_jpeg(enum AVPixelFormat *format)
830 case AV_PIX_FMT_YUVJ420P:
831 *format = AV_PIX_FMT_YUV420P;
833 case AV_PIX_FMT_YUVJ422P:
834 *format = AV_PIX_FMT_YUV422P;
836 case AV_PIX_FMT_YUVJ444P:
837 *format = AV_PIX_FMT_YUV444P;
839 case AV_PIX_FMT_YUVJ440P:
840 *format = AV_PIX_FMT_YUV440P;
847 static int handle_0alpha(enum AVPixelFormat *format)
850 case AV_PIX_FMT_0BGR : *format = AV_PIX_FMT_ABGR ; return 1;
851 case AV_PIX_FMT_BGR0 : *format = AV_PIX_FMT_BGRA ; return 4;
852 case AV_PIX_FMT_0RGB : *format = AV_PIX_FMT_ARGB ; return 1;
853 case AV_PIX_FMT_RGB0 : *format = AV_PIX_FMT_RGBA ; return 4;
858 SwsContext *sws_alloc_context(void)
860 SwsContext *c = av_mallocz(sizeof(SwsContext));
862 c->av_class = &sws_context_class;
863 av_opt_set_defaults(c);
868 av_cold int sws_init_context(SwsContext *c, SwsFilter *srcFilter,
869 SwsFilter *dstFilter)
872 int usesVFilter, usesHFilter;
874 SwsFilter dummyFilter = { NULL, NULL, NULL, NULL };
879 int dst_stride = FFALIGN(dstW * sizeof(int16_t) + 66, 16);
880 int flags, cpu_flags;
881 enum AVPixelFormat srcFormat = c->srcFormat;
882 enum AVPixelFormat dstFormat = c->dstFormat;
883 const AVPixFmtDescriptor *desc_src = av_pix_fmt_desc_get(srcFormat);
884 const AVPixFmtDescriptor *desc_dst = av_pix_fmt_desc_get(dstFormat);
886 cpu_flags = av_get_cpu_flags();
892 unscaled = (srcW == dstW && srcH == dstH);
894 handle_jpeg(&srcFormat);
895 handle_jpeg(&dstFormat);
896 handle_0alpha(&srcFormat);
897 handle_0alpha(&dstFormat);
899 if(srcFormat!=c->srcFormat || dstFormat!=c->dstFormat){
900 av_log(c, AV_LOG_WARNING, "deprecated pixel format used, make sure you did set range correctly\n");
901 c->srcFormat= srcFormat;
902 c->dstFormat= dstFormat;
905 if (!sws_isSupportedInput(srcFormat)) {
906 av_log(c, AV_LOG_ERROR, "%s is not supported as input pixel format\n",
907 av_get_pix_fmt_name(srcFormat));
908 return AVERROR(EINVAL);
910 if (!sws_isSupportedOutput(dstFormat)) {
911 av_log(c, AV_LOG_ERROR, "%s is not supported as output pixel format\n",
912 av_get_pix_fmt_name(dstFormat));
913 return AVERROR(EINVAL);
916 i = flags & (SWS_POINT |
927 if (!i || (i & (i - 1))) {
928 av_log(c, AV_LOG_ERROR, "Exactly one scaler algorithm must be chosen, got %X\n", i);
929 return AVERROR(EINVAL);
932 if (srcW < 4 || srcH < 1 || dstW < 8 || dstH < 1) {
933 /* FIXME check if these are enough and try to lower them after
934 * fixing the relevant parts of the code */
935 av_log(c, AV_LOG_ERROR, "%dx%d -> %dx%d is invalid scaling dimension\n",
936 srcW, srcH, dstW, dstH);
937 return AVERROR(EINVAL);
941 dstFilter = &dummyFilter;
943 srcFilter = &dummyFilter;
945 c->lumXInc = (((int64_t)srcW << 16) + (dstW >> 1)) / dstW;
946 c->lumYInc = (((int64_t)srcH << 16) + (dstH >> 1)) / dstH;
947 c->dstFormatBpp = av_get_bits_per_pixel(desc_dst);
948 c->srcFormatBpp = av_get_bits_per_pixel(desc_src);
949 c->vRounder = 4 * 0x0001000100010001ULL;
951 usesVFilter = (srcFilter->lumV && srcFilter->lumV->length > 1) ||
952 (srcFilter->chrV && srcFilter->chrV->length > 1) ||
953 (dstFilter->lumV && dstFilter->lumV->length > 1) ||
954 (dstFilter->chrV && dstFilter->chrV->length > 1);
955 usesHFilter = (srcFilter->lumH && srcFilter->lumH->length > 1) ||
956 (srcFilter->chrH && srcFilter->chrH->length > 1) ||
957 (dstFilter->lumH && dstFilter->lumH->length > 1) ||
958 (dstFilter->chrH && dstFilter->chrH->length > 1);
960 getSubSampleFactors(&c->chrSrcHSubSample, &c->chrSrcVSubSample, srcFormat);
961 getSubSampleFactors(&c->chrDstHSubSample, &c->chrDstVSubSample, dstFormat);
964 if (isAnyRGB(dstFormat) && !(flags&SWS_FULL_CHR_H_INT)) {
966 av_log(c, AV_LOG_DEBUG, "Forcing full internal H chroma due to odd output size\n");
967 flags |= SWS_FULL_CHR_H_INT;
971 /* reuse chroma for 2 pixels RGB/BGR unless user wants full
972 * chroma interpolation */
973 if (flags & SWS_FULL_CHR_H_INT &&
974 isAnyRGB(dstFormat) &&
975 dstFormat != AV_PIX_FMT_RGBA &&
976 dstFormat != AV_PIX_FMT_ARGB &&
977 dstFormat != AV_PIX_FMT_BGRA &&
978 dstFormat != AV_PIX_FMT_ABGR &&
979 dstFormat != AV_PIX_FMT_RGB24 &&
980 dstFormat != AV_PIX_FMT_BGR24) {
981 av_log(c, AV_LOG_WARNING,
982 "full chroma interpolation for destination format '%s' not yet implemented\n",
983 av_get_pix_fmt_name(dstFormat));
984 flags &= ~SWS_FULL_CHR_H_INT;
987 if (isAnyRGB(dstFormat) && !(flags & SWS_FULL_CHR_H_INT))
988 c->chrDstHSubSample = 1;
990 // drop some chroma lines if the user wants it
991 c->vChrDrop = (flags & SWS_SRC_V_CHR_DROP_MASK) >>
992 SWS_SRC_V_CHR_DROP_SHIFT;
993 c->chrSrcVSubSample += c->vChrDrop;
995 /* drop every other pixel for chroma calculation unless user
996 * wants full chroma */
997 if (isAnyRGB(srcFormat) && !(flags & SWS_FULL_CHR_H_INP) &&
998 srcFormat != AV_PIX_FMT_RGB8 && srcFormat != AV_PIX_FMT_BGR8 &&
999 srcFormat != AV_PIX_FMT_RGB4 && srcFormat != AV_PIX_FMT_BGR4 &&
1000 srcFormat != AV_PIX_FMT_RGB4_BYTE && srcFormat != AV_PIX_FMT_BGR4_BYTE &&
1001 ((dstW >> c->chrDstHSubSample) <= (srcW >> 1) ||
1002 (flags & SWS_FAST_BILINEAR)))
1003 c->chrSrcHSubSample = 1;
1005 // Note the -((-x)>>y) is so that we always round toward +inf.
1006 c->chrSrcW = -((-srcW) >> c->chrSrcHSubSample);
1007 c->chrSrcH = -((-srcH) >> c->chrSrcVSubSample);
1008 c->chrDstW = -((-dstW) >> c->chrDstHSubSample);
1009 c->chrDstH = -((-dstH) >> c->chrDstVSubSample);
1011 FF_ALLOC_OR_GOTO(c, c->formatConvBuffer, FFALIGN(srcW*2+78, 16) * 2, fail);
1013 /* unscaled special cases */
1014 if (unscaled && !usesHFilter && !usesVFilter &&
1015 (c->srcRange == c->dstRange || isAnyRGB(dstFormat))) {
1016 ff_get_unscaled_swscale(c);
1019 if (flags & SWS_PRINT_INFO)
1020 av_log(c, AV_LOG_INFO,
1021 "using unscaled %s -> %s special converter\n",
1022 av_get_pix_fmt_name(srcFormat), av_get_pix_fmt_name(dstFormat));
1027 c->srcBpc = 1 + desc_src->comp[0].depth_minus1;
1030 c->dstBpc = 1 + desc_dst->comp[0].depth_minus1;
1033 if (isAnyRGB(srcFormat) || srcFormat == AV_PIX_FMT_PAL8)
1035 if (c->dstBpc == 16)
1037 if (INLINE_MMXEXT(cpu_flags) && c->srcBpc == 8 && c->dstBpc <= 14) {
1038 c->canMMX2BeUsed = (dstW >= srcW && (dstW & 31) == 0 &&
1039 (srcW & 15) == 0) ? 1 : 0;
1040 if (!c->canMMX2BeUsed && dstW >= srcW && (srcW & 15) == 0
1041 && (flags & SWS_FAST_BILINEAR)) {
1042 if (flags & SWS_PRINT_INFO)
1043 av_log(c, AV_LOG_INFO,
1044 "output width is not a multiple of 32 -> no MMX2 scaler\n");
1046 if (usesHFilter || isNBPS(c->srcFormat) || is16BPS(c->srcFormat) || isAnyRGB(c->srcFormat))
1049 c->canMMX2BeUsed = 0;
1051 c->chrXInc = (((int64_t)c->chrSrcW << 16) + (c->chrDstW >> 1)) / c->chrDstW;
1052 c->chrYInc = (((int64_t)c->chrSrcH << 16) + (c->chrDstH >> 1)) / c->chrDstH;
1054 /* Match pixel 0 of the src to pixel 0 of dst and match pixel n-2 of src
1055 * to pixel n-2 of dst, but only for the FAST_BILINEAR mode otherwise do
1057 * n-2 is the last chrominance sample available.
1058 * This is not perfect, but no one should notice the difference, the more
1059 * correct variant would be like the vertical one, but that would require
1060 * some special code for the first and last pixel */
1061 if (flags & SWS_FAST_BILINEAR) {
1062 if (c->canMMX2BeUsed) {
1066 // we don't use the x86 asm scaler if MMX is available
1067 else if (INLINE_MMX(cpu_flags) && c->dstBpc <= 14) {
1068 c->lumXInc = ((int64_t)(srcW - 2) << 16) / (dstW - 2) - 20;
1069 c->chrXInc = ((int64_t)(c->chrSrcW - 2) << 16) / (c->chrDstW - 2) - 20;
1073 #define USE_MMAP (HAVE_MMAP && HAVE_MPROTECT && defined MAP_ANONYMOUS)
1075 /* precalculate horizontal scaler filter coefficients */
1077 #if HAVE_MMXEXT_INLINE
1078 // can't downscale !!!
1079 if (c->canMMX2BeUsed && (flags & SWS_FAST_BILINEAR)) {
1080 c->lumMmx2FilterCodeSize = initMMX2HScaler(dstW, c->lumXInc, NULL,
1082 c->chrMmx2FilterCodeSize = initMMX2HScaler(c->chrDstW, c->chrXInc,
1083 NULL, NULL, NULL, 4);
1086 c->lumMmx2FilterCode = mmap(NULL, c->lumMmx2FilterCodeSize, PROT_READ | PROT_WRITE, MAP_PRIVATE | MAP_ANONYMOUS, -1, 0);
1087 c->chrMmx2FilterCode = mmap(NULL, c->chrMmx2FilterCodeSize, PROT_READ | PROT_WRITE, MAP_PRIVATE | MAP_ANONYMOUS, -1, 0);
1088 #elif HAVE_VIRTUALALLOC
1089 c->lumMmx2FilterCode = VirtualAlloc(NULL, c->lumMmx2FilterCodeSize, MEM_COMMIT, PAGE_EXECUTE_READWRITE);
1090 c->chrMmx2FilterCode = VirtualAlloc(NULL, c->chrMmx2FilterCodeSize, MEM_COMMIT, PAGE_EXECUTE_READWRITE);
1092 c->lumMmx2FilterCode = av_malloc(c->lumMmx2FilterCodeSize);
1093 c->chrMmx2FilterCode = av_malloc(c->chrMmx2FilterCodeSize);
1096 #ifdef MAP_ANONYMOUS
1097 if (c->lumMmx2FilterCode == MAP_FAILED || c->chrMmx2FilterCode == MAP_FAILED)
1099 if (!c->lumMmx2FilterCode || !c->chrMmx2FilterCode)
1102 av_log(c, AV_LOG_ERROR, "Failed to allocate MMX2FilterCode\n");
1103 return AVERROR(ENOMEM);
1106 FF_ALLOCZ_OR_GOTO(c, c->hLumFilter, (dstW / 8 + 8) * sizeof(int16_t), fail);
1107 FF_ALLOCZ_OR_GOTO(c, c->hChrFilter, (c->chrDstW / 4 + 8) * sizeof(int16_t), fail);
1108 FF_ALLOCZ_OR_GOTO(c, c->hLumFilterPos, (dstW / 2 / 8 + 8) * sizeof(int32_t), fail);
1109 FF_ALLOCZ_OR_GOTO(c, c->hChrFilterPos, (c->chrDstW / 2 / 4 + 8) * sizeof(int32_t), fail);
1111 initMMX2HScaler( dstW, c->lumXInc, c->lumMmx2FilterCode,
1112 c->hLumFilter, (uint32_t*)c->hLumFilterPos, 8);
1113 initMMX2HScaler(c->chrDstW, c->chrXInc, c->chrMmx2FilterCode,
1114 c->hChrFilter, (uint32_t*)c->hChrFilterPos, 4);
1117 mprotect(c->lumMmx2FilterCode, c->lumMmx2FilterCodeSize, PROT_EXEC | PROT_READ);
1118 mprotect(c->chrMmx2FilterCode, c->chrMmx2FilterCodeSize, PROT_EXEC | PROT_READ);
1121 #endif /* HAVE_MMXEXT_INLINE */
1123 const int filterAlign =
1124 (HAVE_MMX && cpu_flags & AV_CPU_FLAG_MMX) ? 4 :
1125 (HAVE_ALTIVEC && cpu_flags & AV_CPU_FLAG_ALTIVEC) ? 8 :
1128 if (initFilter(&c->hLumFilter, &c->hLumFilterPos,
1129 &c->hLumFilterSize, c->lumXInc,
1130 srcW, dstW, filterAlign, 1 << 14,
1131 (flags & SWS_BICUBLIN) ? (flags | SWS_BICUBIC) : flags,
1132 cpu_flags, srcFilter->lumH, dstFilter->lumH,
1135 if (initFilter(&c->hChrFilter, &c->hChrFilterPos,
1136 &c->hChrFilterSize, c->chrXInc,
1137 c->chrSrcW, c->chrDstW, filterAlign, 1 << 14,
1138 (flags & SWS_BICUBLIN) ? (flags | SWS_BILINEAR) : flags,
1139 cpu_flags, srcFilter->chrH, dstFilter->chrH,
1143 } // initialize horizontal stuff
1145 /* precalculate vertical scaler filter coefficients */
1147 const int filterAlign =
1148 (HAVE_MMX && cpu_flags & AV_CPU_FLAG_MMX) ? 2 :
1149 (HAVE_ALTIVEC && cpu_flags & AV_CPU_FLAG_ALTIVEC) ? 8 :
1152 if (initFilter(&c->vLumFilter, &c->vLumFilterPos, &c->vLumFilterSize,
1153 c->lumYInc, srcH, dstH, filterAlign, (1 << 12),
1154 (flags & SWS_BICUBLIN) ? (flags | SWS_BICUBIC) : flags,
1155 cpu_flags, srcFilter->lumV, dstFilter->lumV,
1158 if (initFilter(&c->vChrFilter, &c->vChrFilterPos, &c->vChrFilterSize,
1159 c->chrYInc, c->chrSrcH, c->chrDstH,
1160 filterAlign, (1 << 12),
1161 (flags & SWS_BICUBLIN) ? (flags | SWS_BILINEAR) : flags,
1162 cpu_flags, srcFilter->chrV, dstFilter->chrV,
1167 FF_ALLOC_OR_GOTO(c, c->vYCoeffsBank, sizeof(vector signed short) * c->vLumFilterSize * c->dstH, fail);
1168 FF_ALLOC_OR_GOTO(c, c->vCCoeffsBank, sizeof(vector signed short) * c->vChrFilterSize * c->chrDstH, fail);
1170 for (i = 0; i < c->vLumFilterSize * c->dstH; i++) {
1172 short *p = (short *)&c->vYCoeffsBank[i];
1173 for (j = 0; j < 8; j++)
1174 p[j] = c->vLumFilter[i];
1177 for (i = 0; i < c->vChrFilterSize * c->chrDstH; i++) {
1179 short *p = (short *)&c->vCCoeffsBank[i];
1180 for (j = 0; j < 8; j++)
1181 p[j] = c->vChrFilter[i];
1186 // calculate buffer sizes so that they won't run out while handling these damn slices
1187 c->vLumBufSize = c->vLumFilterSize;
1188 c->vChrBufSize = c->vChrFilterSize;
1189 for (i = 0; i < dstH; i++) {
1190 int chrI = (int64_t)i * c->chrDstH / dstH;
1191 int nextSlice = FFMAX(c->vLumFilterPos[i] + c->vLumFilterSize - 1,
1192 ((c->vChrFilterPos[chrI] + c->vChrFilterSize - 1)
1193 << c->chrSrcVSubSample));
1195 nextSlice >>= c->chrSrcVSubSample;
1196 nextSlice <<= c->chrSrcVSubSample;
1197 if (c->vLumFilterPos[i] + c->vLumBufSize < nextSlice)
1198 c->vLumBufSize = nextSlice - c->vLumFilterPos[i];
1199 if (c->vChrFilterPos[chrI] + c->vChrBufSize <
1200 (nextSlice >> c->chrSrcVSubSample))
1201 c->vChrBufSize = (nextSlice >> c->chrSrcVSubSample) -
1202 c->vChrFilterPos[chrI];
1205 /* Allocate pixbufs (we use dynamic allocation because otherwise we would
1206 * need to allocate several megabytes to handle all possible cases) */
1207 FF_ALLOC_OR_GOTO(c, c->lumPixBuf, c->vLumBufSize * 3 * sizeof(int16_t *), fail);
1208 FF_ALLOC_OR_GOTO(c, c->chrUPixBuf, c->vChrBufSize * 3 * sizeof(int16_t *), fail);
1209 FF_ALLOC_OR_GOTO(c, c->chrVPixBuf, c->vChrBufSize * 3 * sizeof(int16_t *), fail);
1210 if (CONFIG_SWSCALE_ALPHA && isALPHA(c->srcFormat) && isALPHA(c->dstFormat))
1211 FF_ALLOCZ_OR_GOTO(c, c->alpPixBuf, c->vLumBufSize * 3 * sizeof(int16_t *), fail);
1212 /* Note we need at least one pixel more at the end because of the MMX code
1213 * (just in case someone wants to replace the 4000/8000). */
1214 /* align at 16 bytes for AltiVec */
1215 for (i = 0; i < c->vLumBufSize; i++) {
1216 FF_ALLOCZ_OR_GOTO(c, c->lumPixBuf[i + c->vLumBufSize],
1217 dst_stride + 16, fail);
1218 c->lumPixBuf[i] = c->lumPixBuf[i + c->vLumBufSize];
1220 // 64 / c->scalingBpp is the same as 16 / sizeof(scaling_intermediate)
1221 c->uv_off = (dst_stride>>1) + 64 / (c->dstBpc &~ 7);
1222 c->uv_offx2 = dst_stride + 16;
1223 for (i = 0; i < c->vChrBufSize; i++) {
1224 FF_ALLOC_OR_GOTO(c, c->chrUPixBuf[i + c->vChrBufSize],
1225 dst_stride * 2 + 32, fail);
1226 c->chrUPixBuf[i] = c->chrUPixBuf[i + c->vChrBufSize];
1227 c->chrVPixBuf[i] = c->chrVPixBuf[i + c->vChrBufSize]
1228 = c->chrUPixBuf[i] + (dst_stride >> 1) + 8;
1230 if (CONFIG_SWSCALE_ALPHA && c->alpPixBuf)
1231 for (i = 0; i < c->vLumBufSize; i++) {
1232 FF_ALLOCZ_OR_GOTO(c, c->alpPixBuf[i + c->vLumBufSize],
1233 dst_stride + 16, fail);
1234 c->alpPixBuf[i] = c->alpPixBuf[i + c->vLumBufSize];
1237 // try to avoid drawing green stuff between the right end and the stride end
1238 for (i = 0; i < c->vChrBufSize; i++)
1239 if(desc_dst->comp[0].depth_minus1 == 15){
1240 av_assert0(c->dstBpc > 14);
1241 for(j=0; j<dst_stride/2+1; j++)
1242 ((int32_t*)(c->chrUPixBuf[i]))[j] = 1<<18;
1244 for(j=0; j<dst_stride+1; j++)
1245 ((int16_t*)(c->chrUPixBuf[i]))[j] = 1<<14;
1247 av_assert0(c->chrDstH <= dstH);
1249 if (flags & SWS_PRINT_INFO) {
1250 if (flags & SWS_FAST_BILINEAR)
1251 av_log(c, AV_LOG_INFO, "FAST_BILINEAR scaler, ");
1252 else if (flags & SWS_BILINEAR)
1253 av_log(c, AV_LOG_INFO, "BILINEAR scaler, ");
1254 else if (flags & SWS_BICUBIC)
1255 av_log(c, AV_LOG_INFO, "BICUBIC scaler, ");
1256 else if (flags & SWS_X)
1257 av_log(c, AV_LOG_INFO, "Experimental scaler, ");
1258 else if (flags & SWS_POINT)
1259 av_log(c, AV_LOG_INFO, "Nearest Neighbor / POINT scaler, ");
1260 else if (flags & SWS_AREA)
1261 av_log(c, AV_LOG_INFO, "Area Averaging scaler, ");
1262 else if (flags & SWS_BICUBLIN)
1263 av_log(c, AV_LOG_INFO, "luma BICUBIC / chroma BILINEAR scaler, ");
1264 else if (flags & SWS_GAUSS)
1265 av_log(c, AV_LOG_INFO, "Gaussian scaler, ");
1266 else if (flags & SWS_SINC)
1267 av_log(c, AV_LOG_INFO, "Sinc scaler, ");
1268 else if (flags & SWS_LANCZOS)
1269 av_log(c, AV_LOG_INFO, "Lanczos scaler, ");
1270 else if (flags & SWS_SPLINE)
1271 av_log(c, AV_LOG_INFO, "Bicubic spline scaler, ");
1273 av_log(c, AV_LOG_INFO, "ehh flags invalid?! ");
1275 av_log(c, AV_LOG_INFO, "from %s to %s%s ",
1276 av_get_pix_fmt_name(srcFormat),
1278 dstFormat == AV_PIX_FMT_BGR555 || dstFormat == AV_PIX_FMT_BGR565 ||
1279 dstFormat == AV_PIX_FMT_RGB444BE || dstFormat == AV_PIX_FMT_RGB444LE ||
1280 dstFormat == AV_PIX_FMT_BGR444BE || dstFormat == AV_PIX_FMT_BGR444LE ?
1285 av_get_pix_fmt_name(dstFormat));
1287 if (INLINE_MMXEXT(cpu_flags))
1288 av_log(c, AV_LOG_INFO, "using MMX2\n");
1289 else if (INLINE_AMD3DNOW(cpu_flags))
1290 av_log(c, AV_LOG_INFO, "using 3DNOW\n");
1291 else if (INLINE_MMX(cpu_flags))
1292 av_log(c, AV_LOG_INFO, "using MMX\n");
1293 else if (HAVE_ALTIVEC && cpu_flags & AV_CPU_FLAG_ALTIVEC)
1294 av_log(c, AV_LOG_INFO, "using AltiVec\n");
1296 av_log(c, AV_LOG_INFO, "using C\n");
1298 av_log(c, AV_LOG_VERBOSE, "%dx%d -> %dx%d\n", srcW, srcH, dstW, dstH);
1299 av_log(c, AV_LOG_DEBUG,
1300 "lum srcW=%d srcH=%d dstW=%d dstH=%d xInc=%d yInc=%d\n",
1301 c->srcW, c->srcH, c->dstW, c->dstH, c->lumXInc, c->lumYInc);
1302 av_log(c, AV_LOG_DEBUG,
1303 "chr srcW=%d srcH=%d dstW=%d dstH=%d xInc=%d yInc=%d\n",
1304 c->chrSrcW, c->chrSrcH, c->chrDstW, c->chrDstH,
1305 c->chrXInc, c->chrYInc);
1308 c->swScale = ff_getSwsFunc(c);
1310 fail: // FIXME replace things by appropriate error codes
1314 #if FF_API_SWS_GETCONTEXT
1315 SwsContext *sws_getContext(int srcW, int srcH, enum AVPixelFormat srcFormat,
1316 int dstW, int dstH, enum AVPixelFormat dstFormat,
1317 int flags, SwsFilter *srcFilter,
1318 SwsFilter *dstFilter, const double *param)
1322 if (!(c = sws_alloc_context()))
1330 c->srcRange = handle_jpeg(&srcFormat);
1331 c->dstRange = handle_jpeg(&dstFormat);
1332 c->src0Alpha = handle_0alpha(&srcFormat);
1333 c->dst0Alpha = handle_0alpha(&dstFormat);
1334 c->srcFormat = srcFormat;
1335 c->dstFormat = dstFormat;
1338 c->param[0] = param[0];
1339 c->param[1] = param[1];
1341 sws_setColorspaceDetails(c, ff_yuv2rgb_coeffs[SWS_CS_DEFAULT], c->srcRange,
1342 ff_yuv2rgb_coeffs[SWS_CS_DEFAULT] /* FIXME*/,
1343 c->dstRange, 0, 1 << 16, 1 << 16);
1345 if (sws_init_context(c, srcFilter, dstFilter) < 0) {
1354 SwsFilter *sws_getDefaultFilter(float lumaGBlur, float chromaGBlur,
1355 float lumaSharpen, float chromaSharpen,
1356 float chromaHShift, float chromaVShift,
1359 SwsFilter *filter = av_malloc(sizeof(SwsFilter));
1363 if (lumaGBlur != 0.0) {
1364 filter->lumH = sws_getGaussianVec(lumaGBlur, 3.0);
1365 filter->lumV = sws_getGaussianVec(lumaGBlur, 3.0);
1367 filter->lumH = sws_getIdentityVec();
1368 filter->lumV = sws_getIdentityVec();
1371 if (chromaGBlur != 0.0) {
1372 filter->chrH = sws_getGaussianVec(chromaGBlur, 3.0);
1373 filter->chrV = sws_getGaussianVec(chromaGBlur, 3.0);
1375 filter->chrH = sws_getIdentityVec();
1376 filter->chrV = sws_getIdentityVec();
1379 if (chromaSharpen != 0.0) {
1380 SwsVector *id = sws_getIdentityVec();
1381 sws_scaleVec(filter->chrH, -chromaSharpen);
1382 sws_scaleVec(filter->chrV, -chromaSharpen);
1383 sws_addVec(filter->chrH, id);
1384 sws_addVec(filter->chrV, id);
1388 if (lumaSharpen != 0.0) {
1389 SwsVector *id = sws_getIdentityVec();
1390 sws_scaleVec(filter->lumH, -lumaSharpen);
1391 sws_scaleVec(filter->lumV, -lumaSharpen);
1392 sws_addVec(filter->lumH, id);
1393 sws_addVec(filter->lumV, id);
1397 if (chromaHShift != 0.0)
1398 sws_shiftVec(filter->chrH, (int)(chromaHShift + 0.5));
1400 if (chromaVShift != 0.0)
1401 sws_shiftVec(filter->chrV, (int)(chromaVShift + 0.5));
1403 sws_normalizeVec(filter->chrH, 1.0);
1404 sws_normalizeVec(filter->chrV, 1.0);
1405 sws_normalizeVec(filter->lumH, 1.0);
1406 sws_normalizeVec(filter->lumV, 1.0);
1409 sws_printVec2(filter->chrH, NULL, AV_LOG_DEBUG);
1411 sws_printVec2(filter->lumH, NULL, AV_LOG_DEBUG);
1416 SwsVector *sws_allocVec(int length)
1418 SwsVector *vec = av_malloc(sizeof(SwsVector));
1421 vec->length = length;
1422 vec->coeff = av_malloc(sizeof(double) * length);
1428 SwsVector *sws_getGaussianVec(double variance, double quality)
1430 const int length = (int)(variance * quality + 0.5) | 1;
1432 double middle = (length - 1) * 0.5;
1433 SwsVector *vec = sws_allocVec(length);
1438 for (i = 0; i < length; i++) {
1439 double dist = i - middle;
1440 vec->coeff[i] = exp(-dist * dist / (2 * variance * variance)) /
1441 sqrt(2 * variance * M_PI);
1444 sws_normalizeVec(vec, 1.0);
1449 SwsVector *sws_getConstVec(double c, int length)
1452 SwsVector *vec = sws_allocVec(length);
1457 for (i = 0; i < length; i++)
1463 SwsVector *sws_getIdentityVec(void)
1465 return sws_getConstVec(1.0, 1);
1468 static double sws_dcVec(SwsVector *a)
1473 for (i = 0; i < a->length; i++)
1479 void sws_scaleVec(SwsVector *a, double scalar)
1483 for (i = 0; i < a->length; i++)
1484 a->coeff[i] *= scalar;
1487 void sws_normalizeVec(SwsVector *a, double height)
1489 sws_scaleVec(a, height / sws_dcVec(a));
1492 static SwsVector *sws_getConvVec(SwsVector *a, SwsVector *b)
1494 int length = a->length + b->length - 1;
1496 SwsVector *vec = sws_getConstVec(0.0, length);
1501 for (i = 0; i < a->length; i++) {
1502 for (j = 0; j < b->length; j++) {
1503 vec->coeff[i + j] += a->coeff[i] * b->coeff[j];
1510 static SwsVector *sws_sumVec(SwsVector *a, SwsVector *b)
1512 int length = FFMAX(a->length, b->length);
1514 SwsVector *vec = sws_getConstVec(0.0, length);
1519 for (i = 0; i < a->length; i++)
1520 vec->coeff[i + (length - 1) / 2 - (a->length - 1) / 2] += a->coeff[i];
1521 for (i = 0; i < b->length; i++)
1522 vec->coeff[i + (length - 1) / 2 - (b->length - 1) / 2] += b->coeff[i];
1527 static SwsVector *sws_diffVec(SwsVector *a, SwsVector *b)
1529 int length = FFMAX(a->length, b->length);
1531 SwsVector *vec = sws_getConstVec(0.0, length);
1536 for (i = 0; i < a->length; i++)
1537 vec->coeff[i + (length - 1) / 2 - (a->length - 1) / 2] += a->coeff[i];
1538 for (i = 0; i < b->length; i++)
1539 vec->coeff[i + (length - 1) / 2 - (b->length - 1) / 2] -= b->coeff[i];
1544 /* shift left / or right if "shift" is negative */
1545 static SwsVector *sws_getShiftedVec(SwsVector *a, int shift)
1547 int length = a->length + FFABS(shift) * 2;
1549 SwsVector *vec = sws_getConstVec(0.0, length);
1554 for (i = 0; i < a->length; i++) {
1555 vec->coeff[i + (length - 1) / 2 -
1556 (a->length - 1) / 2 - shift] = a->coeff[i];
1562 void sws_shiftVec(SwsVector *a, int shift)
1564 SwsVector *shifted = sws_getShiftedVec(a, shift);
1566 a->coeff = shifted->coeff;
1567 a->length = shifted->length;
1571 void sws_addVec(SwsVector *a, SwsVector *b)
1573 SwsVector *sum = sws_sumVec(a, b);
1575 a->coeff = sum->coeff;
1576 a->length = sum->length;
1580 void sws_subVec(SwsVector *a, SwsVector *b)
1582 SwsVector *diff = sws_diffVec(a, b);
1584 a->coeff = diff->coeff;
1585 a->length = diff->length;
1589 void sws_convVec(SwsVector *a, SwsVector *b)
1591 SwsVector *conv = sws_getConvVec(a, b);
1593 a->coeff = conv->coeff;
1594 a->length = conv->length;
1598 SwsVector *sws_cloneVec(SwsVector *a)
1601 SwsVector *vec = sws_allocVec(a->length);
1606 for (i = 0; i < a->length; i++)
1607 vec->coeff[i] = a->coeff[i];
1612 void sws_printVec2(SwsVector *a, AVClass *log_ctx, int log_level)
1619 for (i = 0; i < a->length; i++)
1620 if (a->coeff[i] > max)
1623 for (i = 0; i < a->length; i++)
1624 if (a->coeff[i] < min)
1629 for (i = 0; i < a->length; i++) {
1630 int x = (int)((a->coeff[i] - min) * 60.0 / range + 0.5);
1631 av_log(log_ctx, log_level, "%1.3f ", a->coeff[i]);
1633 av_log(log_ctx, log_level, " ");
1634 av_log(log_ctx, log_level, "|\n");
1638 void sws_freeVec(SwsVector *a)
1642 av_freep(&a->coeff);
1647 void sws_freeFilter(SwsFilter *filter)
1653 sws_freeVec(filter->lumH);
1655 sws_freeVec(filter->lumV);
1657 sws_freeVec(filter->chrH);
1659 sws_freeVec(filter->chrV);
1663 void sws_freeContext(SwsContext *c)
1670 for (i = 0; i < c->vLumBufSize; i++)
1671 av_freep(&c->lumPixBuf[i]);
1672 av_freep(&c->lumPixBuf);
1675 if (c->chrUPixBuf) {
1676 for (i = 0; i < c->vChrBufSize; i++)
1677 av_freep(&c->chrUPixBuf[i]);
1678 av_freep(&c->chrUPixBuf);
1679 av_freep(&c->chrVPixBuf);
1682 if (CONFIG_SWSCALE_ALPHA && c->alpPixBuf) {
1683 for (i = 0; i < c->vLumBufSize; i++)
1684 av_freep(&c->alpPixBuf[i]);
1685 av_freep(&c->alpPixBuf);
1688 av_freep(&c->vLumFilter);
1689 av_freep(&c->vChrFilter);
1690 av_freep(&c->hLumFilter);
1691 av_freep(&c->hChrFilter);
1693 av_freep(&c->vYCoeffsBank);
1694 av_freep(&c->vCCoeffsBank);
1697 av_freep(&c->vLumFilterPos);
1698 av_freep(&c->vChrFilterPos);
1699 av_freep(&c->hLumFilterPos);
1700 av_freep(&c->hChrFilterPos);
1704 if (c->lumMmx2FilterCode)
1705 munmap(c->lumMmx2FilterCode, c->lumMmx2FilterCodeSize);
1706 if (c->chrMmx2FilterCode)
1707 munmap(c->chrMmx2FilterCode, c->chrMmx2FilterCodeSize);
1708 #elif HAVE_VIRTUALALLOC
1709 if (c->lumMmx2FilterCode)
1710 VirtualFree(c->lumMmx2FilterCode, 0, MEM_RELEASE);
1711 if (c->chrMmx2FilterCode)
1712 VirtualFree(c->chrMmx2FilterCode, 0, MEM_RELEASE);
1714 av_free(c->lumMmx2FilterCode);
1715 av_free(c->chrMmx2FilterCode);
1717 c->lumMmx2FilterCode = NULL;
1718 c->chrMmx2FilterCode = NULL;
1719 #endif /* HAVE_MMX_INLINE */
1721 av_freep(&c->yuvTable);
1722 av_freep(&c->formatConvBuffer);
1727 struct SwsContext *sws_getCachedContext(struct SwsContext *context, int srcW,
1728 int srcH, enum AVPixelFormat srcFormat,
1730 enum AVPixelFormat dstFormat, int flags,
1731 SwsFilter *srcFilter,
1732 SwsFilter *dstFilter,
1733 const double *param)
1735 static const double default_param[2] = { SWS_PARAM_DEFAULT,
1736 SWS_PARAM_DEFAULT };
1739 param = default_param;
1742 (context->srcW != srcW ||
1743 context->srcH != srcH ||
1744 context->srcFormat != srcFormat ||
1745 context->dstW != dstW ||
1746 context->dstH != dstH ||
1747 context->dstFormat != dstFormat ||
1748 context->flags != flags ||
1749 context->param[0] != param[0] ||
1750 context->param[1] != param[1])) {
1751 sws_freeContext(context);
1756 if (!(context = sws_alloc_context()))
1758 context->srcW = srcW;
1759 context->srcH = srcH;
1760 context->srcRange = handle_jpeg(&srcFormat);
1761 context->src0Alpha = handle_0alpha(&srcFormat);
1762 context->srcFormat = srcFormat;
1763 context->dstW = dstW;
1764 context->dstH = dstH;
1765 context->dstRange = handle_jpeg(&dstFormat);
1766 context->dst0Alpha = handle_0alpha(&dstFormat);
1767 context->dstFormat = dstFormat;
1768 context->flags = flags;
1769 context->param[0] = param[0];
1770 context->param[1] = param[1];
1771 sws_setColorspaceDetails(context, ff_yuv2rgb_coeffs[SWS_CS_DEFAULT],
1773 ff_yuv2rgb_coeffs[SWS_CS_DEFAULT] /* FIXME*/,
1774 context->dstRange, 0, 1 << 16, 1 << 16);
1775 if (sws_init_context(context, srcFilter, dstFilter) < 0) {
1776 sws_freeContext(context);