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/ppc/cpu.h"
50 #include "libavutil/x86/asm.h"
51 #include "libavutil/x86/cpu.h"
54 #include "swscale_internal.h"
56 static void handle_formats(SwsContext *c);
58 unsigned swscale_version(void)
60 av_assert0(LIBSWSCALE_VERSION_MICRO >= 100);
61 return LIBSWSCALE_VERSION_INT;
64 const char *swscale_configuration(void)
66 return FFMPEG_CONFIGURATION;
69 const char *swscale_license(void)
71 #define LICENSE_PREFIX "libswscale license: "
72 return LICENSE_PREFIX FFMPEG_LICENSE + sizeof(LICENSE_PREFIX) - 1;
75 #define RET 0xC3 // near return opcode for x86
77 typedef struct FormatEntry {
78 uint8_t is_supported_in :1;
79 uint8_t is_supported_out :1;
80 uint8_t is_supported_endianness :1;
83 static const FormatEntry format_entries[AV_PIX_FMT_NB] = {
84 [AV_PIX_FMT_YUV420P] = { 1, 1 },
85 [AV_PIX_FMT_YUYV422] = { 1, 1 },
86 [AV_PIX_FMT_RGB24] = { 1, 1 },
87 [AV_PIX_FMT_BGR24] = { 1, 1 },
88 [AV_PIX_FMT_YUV422P] = { 1, 1 },
89 [AV_PIX_FMT_YUV444P] = { 1, 1 },
90 [AV_PIX_FMT_YUV410P] = { 1, 1 },
91 [AV_PIX_FMT_YUV411P] = { 1, 1 },
92 [AV_PIX_FMT_GRAY8] = { 1, 1 },
93 [AV_PIX_FMT_MONOWHITE] = { 1, 1 },
94 [AV_PIX_FMT_MONOBLACK] = { 1, 1 },
95 [AV_PIX_FMT_PAL8] = { 1, 0 },
96 [AV_PIX_FMT_YUVJ420P] = { 1, 1 },
97 [AV_PIX_FMT_YUVJ411P] = { 1, 1 },
98 [AV_PIX_FMT_YUVJ422P] = { 1, 1 },
99 [AV_PIX_FMT_YUVJ444P] = { 1, 1 },
100 [AV_PIX_FMT_UYVY422] = { 1, 1 },
101 [AV_PIX_FMT_UYYVYY411] = { 0, 0 },
102 [AV_PIX_FMT_BGR8] = { 1, 1 },
103 [AV_PIX_FMT_BGR4] = { 0, 1 },
104 [AV_PIX_FMT_BGR4_BYTE] = { 1, 1 },
105 [AV_PIX_FMT_RGB8] = { 1, 1 },
106 [AV_PIX_FMT_RGB4] = { 0, 1 },
107 [AV_PIX_FMT_RGB4_BYTE] = { 1, 1 },
108 [AV_PIX_FMT_NV12] = { 1, 1 },
109 [AV_PIX_FMT_NV21] = { 1, 1 },
110 [AV_PIX_FMT_ARGB] = { 1, 1 },
111 [AV_PIX_FMT_RGBA] = { 1, 1 },
112 [AV_PIX_FMT_ABGR] = { 1, 1 },
113 [AV_PIX_FMT_BGRA] = { 1, 1 },
114 [AV_PIX_FMT_0RGB] = { 1, 1 },
115 [AV_PIX_FMT_RGB0] = { 1, 1 },
116 [AV_PIX_FMT_0BGR] = { 1, 1 },
117 [AV_PIX_FMT_BGR0] = { 1, 1 },
118 [AV_PIX_FMT_GRAY16BE] = { 1, 1 },
119 [AV_PIX_FMT_GRAY16LE] = { 1, 1 },
120 [AV_PIX_FMT_YUV440P] = { 1, 1 },
121 [AV_PIX_FMT_YUVJ440P] = { 1, 1 },
122 [AV_PIX_FMT_YUVA420P] = { 1, 1 },
123 [AV_PIX_FMT_YUVA422P] = { 1, 1 },
124 [AV_PIX_FMT_YUVA444P] = { 1, 1 },
125 [AV_PIX_FMT_YUVA420P9BE] = { 1, 1 },
126 [AV_PIX_FMT_YUVA420P9LE] = { 1, 1 },
127 [AV_PIX_FMT_YUVA422P9BE] = { 1, 1 },
128 [AV_PIX_FMT_YUVA422P9LE] = { 1, 1 },
129 [AV_PIX_FMT_YUVA444P9BE] = { 1, 1 },
130 [AV_PIX_FMT_YUVA444P9LE] = { 1, 1 },
131 [AV_PIX_FMT_YUVA420P10BE]= { 1, 1 },
132 [AV_PIX_FMT_YUVA420P10LE]= { 1, 1 },
133 [AV_PIX_FMT_YUVA422P10BE]= { 1, 1 },
134 [AV_PIX_FMT_YUVA422P10LE]= { 1, 1 },
135 [AV_PIX_FMT_YUVA444P10BE]= { 1, 1 },
136 [AV_PIX_FMT_YUVA444P10LE]= { 1, 1 },
137 [AV_PIX_FMT_YUVA420P16BE]= { 1, 1 },
138 [AV_PIX_FMT_YUVA420P16LE]= { 1, 1 },
139 [AV_PIX_FMT_YUVA422P16BE]= { 1, 1 },
140 [AV_PIX_FMT_YUVA422P16LE]= { 1, 1 },
141 [AV_PIX_FMT_YUVA444P16BE]= { 1, 1 },
142 [AV_PIX_FMT_YUVA444P16LE]= { 1, 1 },
143 [AV_PIX_FMT_RGB48BE] = { 1, 1 },
144 [AV_PIX_FMT_RGB48LE] = { 1, 1 },
145 [AV_PIX_FMT_RGBA64BE] = { 1, 1 },
146 [AV_PIX_FMT_RGBA64LE] = { 1, 1 },
147 [AV_PIX_FMT_RGB565BE] = { 1, 1 },
148 [AV_PIX_FMT_RGB565LE] = { 1, 1 },
149 [AV_PIX_FMT_RGB555BE] = { 1, 1 },
150 [AV_PIX_FMT_RGB555LE] = { 1, 1 },
151 [AV_PIX_FMT_BGR565BE] = { 1, 1 },
152 [AV_PIX_FMT_BGR565LE] = { 1, 1 },
153 [AV_PIX_FMT_BGR555BE] = { 1, 1 },
154 [AV_PIX_FMT_BGR555LE] = { 1, 1 },
155 [AV_PIX_FMT_YUV420P16LE] = { 1, 1 },
156 [AV_PIX_FMT_YUV420P16BE] = { 1, 1 },
157 [AV_PIX_FMT_YUV422P16LE] = { 1, 1 },
158 [AV_PIX_FMT_YUV422P16BE] = { 1, 1 },
159 [AV_PIX_FMT_YUV444P16LE] = { 1, 1 },
160 [AV_PIX_FMT_YUV444P16BE] = { 1, 1 },
161 [AV_PIX_FMT_RGB444LE] = { 1, 1 },
162 [AV_PIX_FMT_RGB444BE] = { 1, 1 },
163 [AV_PIX_FMT_BGR444LE] = { 1, 1 },
164 [AV_PIX_FMT_BGR444BE] = { 1, 1 },
165 [AV_PIX_FMT_Y400A] = { 1, 0 },
166 [AV_PIX_FMT_BGR48BE] = { 1, 1 },
167 [AV_PIX_FMT_BGR48LE] = { 1, 1 },
168 [AV_PIX_FMT_BGRA64BE] = { 0, 0 },
169 [AV_PIX_FMT_BGRA64LE] = { 0, 0 },
170 [AV_PIX_FMT_YUV420P9BE] = { 1, 1 },
171 [AV_PIX_FMT_YUV420P9LE] = { 1, 1 },
172 [AV_PIX_FMT_YUV420P10BE] = { 1, 1 },
173 [AV_PIX_FMT_YUV420P10LE] = { 1, 1 },
174 [AV_PIX_FMT_YUV420P12BE] = { 1, 1 },
175 [AV_PIX_FMT_YUV420P12LE] = { 1, 1 },
176 [AV_PIX_FMT_YUV420P14BE] = { 1, 1 },
177 [AV_PIX_FMT_YUV420P14LE] = { 1, 1 },
178 [AV_PIX_FMT_YUV422P9BE] = { 1, 1 },
179 [AV_PIX_FMT_YUV422P9LE] = { 1, 1 },
180 [AV_PIX_FMT_YUV422P10BE] = { 1, 1 },
181 [AV_PIX_FMT_YUV422P10LE] = { 1, 1 },
182 [AV_PIX_FMT_YUV422P12BE] = { 1, 1 },
183 [AV_PIX_FMT_YUV422P12LE] = { 1, 1 },
184 [AV_PIX_FMT_YUV422P14BE] = { 1, 1 },
185 [AV_PIX_FMT_YUV422P14LE] = { 1, 1 },
186 [AV_PIX_FMT_YUV444P9BE] = { 1, 1 },
187 [AV_PIX_FMT_YUV444P9LE] = { 1, 1 },
188 [AV_PIX_FMT_YUV444P10BE] = { 1, 1 },
189 [AV_PIX_FMT_YUV444P10LE] = { 1, 1 },
190 [AV_PIX_FMT_YUV444P12BE] = { 1, 1 },
191 [AV_PIX_FMT_YUV444P12LE] = { 1, 1 },
192 [AV_PIX_FMT_YUV444P14BE] = { 1, 1 },
193 [AV_PIX_FMT_YUV444P14LE] = { 1, 1 },
194 [AV_PIX_FMT_GBRP] = { 1, 1 },
195 [AV_PIX_FMT_GBRP9LE] = { 1, 1 },
196 [AV_PIX_FMT_GBRP9BE] = { 1, 1 },
197 [AV_PIX_FMT_GBRP10LE] = { 1, 1 },
198 [AV_PIX_FMT_GBRP10BE] = { 1, 1 },
199 [AV_PIX_FMT_GBRP12LE] = { 1, 1 },
200 [AV_PIX_FMT_GBRP12BE] = { 1, 1 },
201 [AV_PIX_FMT_GBRP14LE] = { 1, 1 },
202 [AV_PIX_FMT_GBRP14BE] = { 1, 1 },
203 [AV_PIX_FMT_GBRP16LE] = { 1, 0 },
204 [AV_PIX_FMT_GBRP16BE] = { 1, 0 },
205 [AV_PIX_FMT_XYZ12BE] = { 1, 1, 1 },
206 [AV_PIX_FMT_XYZ12LE] = { 1, 1, 1 },
207 [AV_PIX_FMT_GBRAP] = { 1, 1 },
208 [AV_PIX_FMT_GBRAP16LE] = { 1, 0 },
209 [AV_PIX_FMT_GBRAP16BE] = { 1, 0 },
210 [AV_PIX_FMT_BAYER_BGGR8] = { 1, 0 },
211 [AV_PIX_FMT_BAYER_RGGB8] = { 1, 0 },
212 [AV_PIX_FMT_BAYER_GBRG8] = { 1, 0 },
213 [AV_PIX_FMT_BAYER_GRBG8] = { 1, 0 },
214 [AV_PIX_FMT_BAYER_BGGR16LE] = { 1, 0 },
215 [AV_PIX_FMT_BAYER_BGGR16BE] = { 1, 0 },
216 [AV_PIX_FMT_BAYER_RGGB16LE] = { 1, 0 },
217 [AV_PIX_FMT_BAYER_RGGB16BE] = { 1, 0 },
218 [AV_PIX_FMT_BAYER_GBRG16LE] = { 1, 0 },
219 [AV_PIX_FMT_BAYER_GBRG16BE] = { 1, 0 },
220 [AV_PIX_FMT_BAYER_GRBG16LE] = { 1, 0 },
221 [AV_PIX_FMT_BAYER_GRBG16BE] = { 1, 0 },
224 int sws_isSupportedInput(enum AVPixelFormat pix_fmt)
226 return (unsigned)pix_fmt < AV_PIX_FMT_NB ?
227 format_entries[pix_fmt].is_supported_in : 0;
230 int sws_isSupportedOutput(enum AVPixelFormat pix_fmt)
232 return (unsigned)pix_fmt < AV_PIX_FMT_NB ?
233 format_entries[pix_fmt].is_supported_out : 0;
236 int sws_isSupportedEndiannessConversion(enum AVPixelFormat pix_fmt)
238 return (unsigned)pix_fmt < AV_PIX_FMT_NB ?
239 format_entries[pix_fmt].is_supported_endianness : 0;
242 #if FF_API_SWS_FORMAT_NAME
243 const char *sws_format_name(enum AVPixelFormat format)
245 const AVPixFmtDescriptor *desc = av_pix_fmt_desc_get(format);
249 return "Unknown format";
253 static double getSplineCoeff(double a, double b, double c, double d,
257 return ((d * dist + c) * dist + b) * dist + a;
259 return getSplineCoeff(0.0,
260 b + 2.0 * c + 3.0 * d,
262 -b - 3.0 * c - 6.0 * d,
266 static av_cold int get_local_pos(SwsContext *s, int chr_subsample, int pos, int dir)
269 pos = (128 << chr_subsample) - 128;
271 pos += 128; // relative to ideal left edge
272 return pos >> chr_subsample;
276 int flag; ///< flag associated to the algorithm
277 const char *description; ///< human-readable description
278 int size_factor; ///< size factor used when initing the filters
281 static const ScaleAlgorithm scale_algorithms[] = {
282 { SWS_AREA, "area averaging", 1 /* downscale only, for upscale it is bilinear */ },
283 { SWS_BICUBIC, "bicubic", 4 },
284 { SWS_BICUBLIN, "luma bicubic / chroma bilinear", -1 },
285 { SWS_BILINEAR, "bilinear", 2 },
286 { SWS_FAST_BILINEAR, "fast bilinear", -1 },
287 { SWS_GAUSS, "Gaussian", 8 /* infinite ;) */ },
288 { SWS_LANCZOS, "Lanczos", -1 /* custom */ },
289 { SWS_POINT, "nearest neighbor / point", -1 },
290 { SWS_SINC, "sinc", 20 /* infinite ;) */ },
291 { SWS_SPLINE, "bicubic spline", 20 /* infinite :)*/ },
292 { SWS_X, "experimental", 8 },
295 static av_cold int initFilter(int16_t **outFilter, int32_t **filterPos,
296 int *outFilterSize, int xInc, int srcW,
297 int dstW, int filterAlign, int one,
298 int flags, int cpu_flags,
299 SwsVector *srcFilter, SwsVector *dstFilter,
300 double param[2], int srcPos, int dstPos)
306 int64_t *filter = NULL;
307 int64_t *filter2 = NULL;
308 const int64_t fone = 1LL << (54 - FFMIN(av_log2(srcW/dstW), 8));
311 emms_c(); // FIXME should not be required but IS (even for non-MMX versions)
313 // NOTE: the +3 is for the MMX(+1) / SSE(+3) scaler which reads over the end
314 FF_ALLOC_OR_GOTO(NULL, *filterPos, (dstW + 3) * sizeof(**filterPos), fail);
316 if (FFABS(xInc - 0x10000) < 10 && srcPos == dstPos) { // unscaled
319 FF_ALLOCZ_OR_GOTO(NULL, filter,
320 dstW * sizeof(*filter) * filterSize, fail);
322 for (i = 0; i < dstW; i++) {
323 filter[i * filterSize] = fone;
326 } else if (flags & SWS_POINT) { // lame looking point sampling mode
330 FF_ALLOC_OR_GOTO(NULL, filter,
331 dstW * sizeof(*filter) * filterSize, fail);
333 xDstInSrc = ((dstPos*(int64_t)xInc)>>8) - ((srcPos*0x8000LL)>>7);
334 for (i = 0; i < dstW; i++) {
335 int xx = (xDstInSrc - ((filterSize - 1) << 15) + (1 << 15)) >> 16;
337 (*filterPos)[i] = xx;
341 } else if ((xInc <= (1 << 16) && (flags & SWS_AREA)) ||
342 (flags & SWS_FAST_BILINEAR)) { // bilinear upscale
346 FF_ALLOC_OR_GOTO(NULL, filter,
347 dstW * sizeof(*filter) * filterSize, fail);
349 xDstInSrc = ((dstPos*(int64_t)xInc)>>8) - ((srcPos*0x8000LL)>>7);
350 for (i = 0; i < dstW; i++) {
351 int xx = (xDstInSrc - ((filterSize - 1) << 15) + (1 << 15)) >> 16;
354 (*filterPos)[i] = xx;
355 // bilinear upscale / linear interpolate / area averaging
356 for (j = 0; j < filterSize; j++) {
357 int64_t coeff= fone - FFABS(((int64_t)xx<<16) - xDstInSrc)*(fone>>16);
360 filter[i * filterSize + j] = coeff;
369 for (i = 0; i < FF_ARRAY_ELEMS(scale_algorithms); i++) {
370 if (flags & scale_algorithms[i].flag) {
371 sizeFactor = scale_algorithms[i].size_factor;
375 if (flags & SWS_LANCZOS)
376 sizeFactor = param[0] != SWS_PARAM_DEFAULT ? ceil(2 * param[0]) : 6;
377 av_assert0(sizeFactor > 0);
380 filterSize = 1 + sizeFactor; // upscale
382 filterSize = 1 + (sizeFactor * srcW + dstW - 1) / dstW;
384 filterSize = FFMIN(filterSize, srcW - 2);
385 filterSize = FFMAX(filterSize, 1);
387 FF_ALLOC_OR_GOTO(NULL, filter,
388 dstW * sizeof(*filter) * filterSize, fail);
390 xDstInSrc = ((dstPos*(int64_t)xInc)>>7) - ((srcPos*0x10000LL)>>7);
391 for (i = 0; i < dstW; i++) {
392 int xx = (xDstInSrc - ((filterSize - 2) << 16)) / (1 << 17);
394 (*filterPos)[i] = xx;
395 for (j = 0; j < filterSize; j++) {
396 int64_t d = (FFABS(((int64_t)xx << 17) - xDstInSrc)) << 13;
402 floatd = d * (1.0 / (1 << 30));
404 if (flags & SWS_BICUBIC) {
405 int64_t B = (param[0] != SWS_PARAM_DEFAULT ? param[0] : 0) * (1 << 24);
406 int64_t C = (param[1] != SWS_PARAM_DEFAULT ? param[1] : 0.6) * (1 << 24);
408 if (d >= 1LL << 31) {
411 int64_t dd = (d * d) >> 30;
412 int64_t ddd = (dd * d) >> 30;
415 coeff = (12 * (1 << 24) - 9 * B - 6 * C) * ddd +
416 (-18 * (1 << 24) + 12 * B + 6 * C) * dd +
417 (6 * (1 << 24) - 2 * B) * (1 << 30);
419 coeff = (-B - 6 * C) * ddd +
420 (6 * B + 30 * C) * dd +
421 (-12 * B - 48 * C) * d +
422 (8 * B + 24 * C) * (1 << 30);
424 coeff /= (1LL<<54)/fone;
427 else if (flags & SWS_X) {
428 double p = param ? param * 0.01 : 0.3;
429 coeff = d ? sin(d * M_PI) / (d * M_PI) : 1.0;
430 coeff *= pow(2.0, -p * d * d);
433 else if (flags & SWS_X) {
434 double A = param[0] != SWS_PARAM_DEFAULT ? param[0] : 1.0;
438 c = cos(floatd * M_PI);
445 coeff = (c * 0.5 + 0.5) * fone;
446 } else if (flags & SWS_AREA) {
447 int64_t d2 = d - (1 << 29);
448 if (d2 * xInc < -(1LL << (29 + 16)))
449 coeff = 1.0 * (1LL << (30 + 16));
450 else if (d2 * xInc < (1LL << (29 + 16)))
451 coeff = -d2 * xInc + (1LL << (29 + 16));
454 coeff *= fone >> (30 + 16);
455 } else if (flags & SWS_GAUSS) {
456 double p = param[0] != SWS_PARAM_DEFAULT ? param[0] : 3.0;
457 coeff = (pow(2.0, -p * floatd * floatd)) * fone;
458 } else if (flags & SWS_SINC) {
459 coeff = (d ? sin(floatd * M_PI) / (floatd * M_PI) : 1.0) * fone;
460 } else if (flags & SWS_LANCZOS) {
461 double p = param[0] != SWS_PARAM_DEFAULT ? param[0] : 3.0;
462 coeff = (d ? sin(floatd * M_PI) * sin(floatd * M_PI / p) /
463 (floatd * floatd * M_PI * M_PI / p) : 1.0) * fone;
466 } else if (flags & SWS_BILINEAR) {
467 coeff = (1 << 30) - d;
471 } else if (flags & SWS_SPLINE) {
472 double p = -2.196152422706632;
473 coeff = getSplineCoeff(1.0, 0.0, p, -p - 1.0, floatd) * fone;
478 filter[i * filterSize + j] = coeff;
481 xDstInSrc += 2 * xInc;
485 /* apply src & dst Filter to filter -> filter2
488 av_assert0(filterSize > 0);
489 filter2Size = filterSize;
491 filter2Size += srcFilter->length - 1;
493 filter2Size += dstFilter->length - 1;
494 av_assert0(filter2Size > 0);
495 FF_ALLOCZ_OR_GOTO(NULL, filter2, filter2Size * dstW * sizeof(*filter2), fail);
497 for (i = 0; i < dstW; i++) {
501 for (k = 0; k < srcFilter->length; k++) {
502 for (j = 0; j < filterSize; j++)
503 filter2[i * filter2Size + k + j] +=
504 srcFilter->coeff[k] * filter[i * filterSize + j];
507 for (j = 0; j < filterSize; j++)
508 filter2[i * filter2Size + j] = filter[i * filterSize + j];
512 (*filterPos)[i] += (filterSize - 1) / 2 - (filter2Size - 1) / 2;
516 /* try to reduce the filter-size (step1 find size and shift left) */
517 // Assume it is near normalized (*0.5 or *2.0 is OK but * 0.001 is not).
519 for (i = dstW - 1; i >= 0; i--) {
520 int min = filter2Size;
522 int64_t cutOff = 0.0;
524 /* get rid of near zero elements on the left by shifting left */
525 for (j = 0; j < filter2Size; j++) {
527 cutOff += FFABS(filter2[i * filter2Size]);
529 if (cutOff > SWS_MAX_REDUCE_CUTOFF * fone)
532 /* preserve monotonicity because the core can't handle the
533 * filter otherwise */
534 if (i < dstW - 1 && (*filterPos)[i] >= (*filterPos)[i + 1])
537 // move filter coefficients left
538 for (k = 1; k < filter2Size; k++)
539 filter2[i * filter2Size + k - 1] = filter2[i * filter2Size + k];
540 filter2[i * filter2Size + k - 1] = 0;
545 /* count near zeros on the right */
546 for (j = filter2Size - 1; j > 0; j--) {
547 cutOff += FFABS(filter2[i * filter2Size + j]);
549 if (cutOff > SWS_MAX_REDUCE_CUTOFF * fone)
554 if (min > minFilterSize)
558 if (PPC_ALTIVEC(cpu_flags)) {
559 // we can handle the special case 4, so we don't want to go the full 8
560 if (minFilterSize < 5)
563 /* We really don't want to waste our time doing useless computation, so
564 * fall back on the scalar C code for very small filters.
565 * Vectorizing is worth it only if you have a decent-sized vector. */
566 if (minFilterSize < 3)
570 if (HAVE_MMX && cpu_flags & AV_CPU_FLAG_MMX) {
571 // special case for unscaled vertical filtering
572 if (minFilterSize == 1 && filterAlign == 2)
576 av_assert0(minFilterSize > 0);
577 filterSize = (minFilterSize + (filterAlign - 1)) & (~(filterAlign - 1));
578 av_assert0(filterSize > 0);
579 filter = av_malloc(filterSize * dstW * sizeof(*filter));
580 if (filterSize >= MAX_FILTER_SIZE * 16 /
581 ((flags & SWS_ACCURATE_RND) ? APCK_SIZE : 16) || !filter) {
582 av_log(NULL, AV_LOG_ERROR, "sws: filterSize %d is too large, try less extreme scaling or increase MAX_FILTER_SIZE and recompile\n", filterSize);
585 *outFilterSize = filterSize;
587 if (flags & SWS_PRINT_INFO)
588 av_log(NULL, AV_LOG_VERBOSE,
589 "SwScaler: reducing / aligning filtersize %d -> %d\n",
590 filter2Size, filterSize);
591 /* try to reduce the filter-size (step2 reduce it) */
592 for (i = 0; i < dstW; i++) {
595 for (j = 0; j < filterSize; j++) {
596 if (j >= filter2Size)
597 filter[i * filterSize + j] = 0;
599 filter[i * filterSize + j] = filter2[i * filter2Size + j];
600 if ((flags & SWS_BITEXACT) && j >= minFilterSize)
601 filter[i * filterSize + j] = 0;
605 // FIXME try to align filterPos if possible
608 for (i = 0; i < dstW; i++) {
610 if ((*filterPos)[i] < 0) {
611 // move filter coefficients left to compensate for filterPos
612 for (j = 1; j < filterSize; j++) {
613 int left = FFMAX(j + (*filterPos)[i], 0);
614 filter[i * filterSize + left] += filter[i * filterSize + j];
615 filter[i * filterSize + j] = 0;
620 if ((*filterPos)[i] + filterSize > srcW) {
621 int shift = (*filterPos)[i] + filterSize - srcW;
622 // move filter coefficients right to compensate for filterPos
623 for (j = filterSize - 2; j >= 0; j--) {
624 int right = FFMIN(j + shift, filterSize - 1);
625 filter[i * filterSize + right] += filter[i * filterSize + j];
626 filter[i * filterSize + j] = 0;
628 (*filterPos)[i]= srcW - filterSize;
632 // Note the +1 is for the MMX scaler which reads over the end
633 /* align at 16 for AltiVec (needed by hScale_altivec_real) */
634 FF_ALLOCZ_OR_GOTO(NULL, *outFilter,
635 *outFilterSize * (dstW + 3) * sizeof(int16_t), fail);
637 /* normalize & store in outFilter */
638 for (i = 0; i < dstW; i++) {
643 for (j = 0; j < filterSize; j++) {
644 sum += filter[i * filterSize + j];
646 sum = (sum + one / 2) / one;
648 av_log(NULL, AV_LOG_WARNING, "SwScaler: zero vector in scaling\n");
651 for (j = 0; j < *outFilterSize; j++) {
652 int64_t v = filter[i * filterSize + j] + error;
653 int intV = ROUNDED_DIV(v, sum);
654 (*outFilter)[i * (*outFilterSize) + j] = intV;
655 error = v - intV * sum;
659 (*filterPos)[dstW + 0] =
660 (*filterPos)[dstW + 1] =
661 (*filterPos)[dstW + 2] = (*filterPos)[dstW - 1]; /* the MMX/SSE scaler will
662 * read over the end */
663 for (i = 0; i < *outFilterSize; i++) {
664 int k = (dstW - 1) * (*outFilterSize) + i;
665 (*outFilter)[k + 1 * (*outFilterSize)] =
666 (*outFilter)[k + 2 * (*outFilterSize)] =
667 (*outFilter)[k + 3 * (*outFilterSize)] = (*outFilter)[k];
674 av_log(NULL, AV_LOG_ERROR, "sws: initFilter failed\n");
680 #if HAVE_MMXEXT_INLINE
681 static av_cold int init_hscaler_mmxext(int dstW, int xInc, uint8_t *filterCode,
682 int16_t *filter, int32_t *filterPos,
686 x86_reg imm8OfPShufW1A;
687 x86_reg imm8OfPShufW2A;
688 x86_reg fragmentLengthA;
690 x86_reg imm8OfPShufW1B;
691 x86_reg imm8OfPShufW2B;
692 x86_reg fragmentLengthB;
697 // create an optimized horizontal scaling routine
698 /* This scaler is made of runtime-generated MMXEXT code using specially tuned
699 * pshufw instructions. For every four output pixels, if four input pixels
700 * are enough for the fast bilinear scaling, then a chunk of fragmentB is
701 * used. If five input pixels are needed, then a chunk of fragmentA is used.
710 "movq (%%"REG_d", %%"REG_a"), %%mm3 \n\t"
711 "movd (%%"REG_c", %%"REG_S"), %%mm0 \n\t"
712 "movd 1(%%"REG_c", %%"REG_S"), %%mm1 \n\t"
713 "punpcklbw %%mm7, %%mm1 \n\t"
714 "punpcklbw %%mm7, %%mm0 \n\t"
715 "pshufw $0xFF, %%mm1, %%mm1 \n\t"
717 "pshufw $0xFF, %%mm0, %%mm0 \n\t"
719 "psubw %%mm1, %%mm0 \n\t"
720 "movl 8(%%"REG_b", %%"REG_a"), %%esi \n\t"
721 "pmullw %%mm3, %%mm0 \n\t"
722 "psllw $7, %%mm1 \n\t"
723 "paddw %%mm1, %%mm0 \n\t"
725 "movq %%mm0, (%%"REG_D", %%"REG_a") \n\t"
727 "add $8, %%"REG_a" \n\t"
731 "lea " LOCAL_MANGLE(0b) ", %0 \n\t"
732 "lea " LOCAL_MANGLE(1b) ", %1 \n\t"
733 "lea " LOCAL_MANGLE(2b) ", %2 \n\t"
738 "lea " LOCAL_MANGLE(9b) ", %3 \n\t"
742 : "=r" (fragmentA), "=r" (imm8OfPShufW1A), "=r" (imm8OfPShufW2A),
743 "=r" (fragmentLengthA)
750 "movq (%%"REG_d", %%"REG_a"), %%mm3 \n\t"
751 "movd (%%"REG_c", %%"REG_S"), %%mm0 \n\t"
752 "punpcklbw %%mm7, %%mm0 \n\t"
753 "pshufw $0xFF, %%mm0, %%mm1 \n\t"
755 "pshufw $0xFF, %%mm0, %%mm0 \n\t"
757 "psubw %%mm1, %%mm0 \n\t"
758 "movl 8(%%"REG_b", %%"REG_a"), %%esi \n\t"
759 "pmullw %%mm3, %%mm0 \n\t"
760 "psllw $7, %%mm1 \n\t"
761 "paddw %%mm1, %%mm0 \n\t"
763 "movq %%mm0, (%%"REG_D", %%"REG_a") \n\t"
765 "add $8, %%"REG_a" \n\t"
769 "lea " LOCAL_MANGLE(0b) ", %0 \n\t"
770 "lea " LOCAL_MANGLE(1b) ", %1 \n\t"
771 "lea " LOCAL_MANGLE(2b) ", %2 \n\t"
776 "lea " LOCAL_MANGLE(9b) ", %3 \n\t"
780 : "=r" (fragmentB), "=r" (imm8OfPShufW1B), "=r" (imm8OfPShufW2B),
781 "=r" (fragmentLengthB)
784 xpos = 0; // lumXInc/2 - 0x8000; // difference between pixel centers
787 for (i = 0; i < dstW / numSplits; i++) {
792 int b = ((xpos + xInc) >> 16) - xx;
793 int c = ((xpos + xInc * 2) >> 16) - xx;
794 int d = ((xpos + xInc * 3) >> 16) - xx;
795 int inc = (d + 1 < 4);
796 uint8_t *fragment = inc ? fragmentB : fragmentA;
797 x86_reg imm8OfPShufW1 = inc ? imm8OfPShufW1B : imm8OfPShufW1A;
798 x86_reg imm8OfPShufW2 = inc ? imm8OfPShufW2B : imm8OfPShufW2A;
799 x86_reg fragmentLength = inc ? fragmentLengthB : fragmentLengthA;
800 int maxShift = 3 - (d + inc);
804 filter[i] = ((xpos & 0xFFFF) ^ 0xFFFF) >> 9;
805 filter[i + 1] = (((xpos + xInc) & 0xFFFF) ^ 0xFFFF) >> 9;
806 filter[i + 2] = (((xpos + xInc * 2) & 0xFFFF) ^ 0xFFFF) >> 9;
807 filter[i + 3] = (((xpos + xInc * 3) & 0xFFFF) ^ 0xFFFF) >> 9;
808 filterPos[i / 2] = xx;
810 memcpy(filterCode + fragmentPos, fragment, fragmentLength);
812 filterCode[fragmentPos + imm8OfPShufW1] = (a + inc) |
816 filterCode[fragmentPos + imm8OfPShufW2] = a | (b << 2) |
820 if (i + 4 - inc >= dstW)
821 shift = maxShift; // avoid overread
822 else if ((filterPos[i / 2] & 3) <= maxShift)
823 shift = filterPos[i / 2] & 3; // align
825 if (shift && i >= shift) {
826 filterCode[fragmentPos + imm8OfPShufW1] += 0x55 * shift;
827 filterCode[fragmentPos + imm8OfPShufW2] += 0x55 * shift;
828 filterPos[i / 2] -= shift;
832 fragmentPos += fragmentLength;
835 filterCode[fragmentPos] = RET;
840 filterPos[((i / 2) + 1) & (~1)] = xpos >> 16; // needed to jump to the next part
842 return fragmentPos + 1;
844 #endif /* HAVE_MMXEXT_INLINE */
846 static void fill_rgb2yuv_table(SwsContext *c, const int table[4], int dstRange)
848 int64_t W, V, Z, Cy, Cu, Cv;
849 int64_t vr = table[0];
850 int64_t ub = table[1];
851 int64_t ug = -table[2];
852 int64_t vg = -table[3];
855 uint8_t *p = (uint8_t*)c->input_rgb2yuv_table;
857 static const int8_t map[] = {
858 BY_IDX, GY_IDX, -1 , BY_IDX, BY_IDX, GY_IDX, -1 , BY_IDX,
859 RY_IDX, -1 , GY_IDX, RY_IDX, RY_IDX, -1 , GY_IDX, RY_IDX,
860 RY_IDX, GY_IDX, -1 , RY_IDX, RY_IDX, GY_IDX, -1 , RY_IDX,
861 BY_IDX, -1 , GY_IDX, BY_IDX, BY_IDX, -1 , GY_IDX, BY_IDX,
862 BU_IDX, GU_IDX, -1 , BU_IDX, BU_IDX, GU_IDX, -1 , BU_IDX,
863 RU_IDX, -1 , GU_IDX, RU_IDX, RU_IDX, -1 , GU_IDX, RU_IDX,
864 RU_IDX, GU_IDX, -1 , RU_IDX, RU_IDX, GU_IDX, -1 , RU_IDX,
865 BU_IDX, -1 , GU_IDX, BU_IDX, BU_IDX, -1 , GU_IDX, BU_IDX,
866 BV_IDX, GV_IDX, -1 , BV_IDX, BV_IDX, GV_IDX, -1 , BV_IDX,
867 RV_IDX, -1 , GV_IDX, RV_IDX, RV_IDX, -1 , GV_IDX, RV_IDX,
868 RV_IDX, GV_IDX, -1 , RV_IDX, RV_IDX, GV_IDX, -1 , RV_IDX,
869 BV_IDX, -1 , GV_IDX, BV_IDX, BV_IDX, -1 , GV_IDX, BV_IDX,
870 RY_IDX, BY_IDX, RY_IDX, BY_IDX, RY_IDX, BY_IDX, RY_IDX, BY_IDX,
871 BY_IDX, RY_IDX, BY_IDX, RY_IDX, BY_IDX, RY_IDX, BY_IDX, RY_IDX,
872 GY_IDX, -1 , GY_IDX, -1 , GY_IDX, -1 , GY_IDX, -1 ,
873 -1 , GY_IDX, -1 , GY_IDX, -1 , GY_IDX, -1 , GY_IDX,
874 RU_IDX, BU_IDX, RU_IDX, BU_IDX, RU_IDX, BU_IDX, RU_IDX, BU_IDX,
875 BU_IDX, RU_IDX, BU_IDX, RU_IDX, BU_IDX, RU_IDX, BU_IDX, RU_IDX,
876 GU_IDX, -1 , GU_IDX, -1 , GU_IDX, -1 , GU_IDX, -1 ,
877 -1 , GU_IDX, -1 , GU_IDX, -1 , GU_IDX, -1 , GU_IDX,
878 RV_IDX, BV_IDX, RV_IDX, BV_IDX, RV_IDX, BV_IDX, RV_IDX, BV_IDX,
879 BV_IDX, RV_IDX, BV_IDX, RV_IDX, BV_IDX, RV_IDX, BV_IDX, RV_IDX,
880 GV_IDX, -1 , GV_IDX, -1 , GV_IDX, -1 , GV_IDX, -1 ,
881 -1 , GV_IDX, -1 , GV_IDX, -1 , GV_IDX, -1 , GV_IDX, //23
882 -1 , -1 , -1 , -1 , -1 , -1 , -1 , -1 , //24
883 -1 , -1 , -1 , -1 , -1 , -1 , -1 , -1 , //25
884 -1 , -1 , -1 , -1 , -1 , -1 , -1 , -1 , //26
885 -1 , -1 , -1 , -1 , -1 , -1 , -1 , -1 , //27
886 -1 , -1 , -1 , -1 , -1 , -1 , -1 , -1 , //28
887 -1 , -1 , -1 , -1 , -1 , -1 , -1 , -1 , //29
888 -1 , -1 , -1 , -1 , -1 , -1 , -1 , -1 , //30
889 -1 , -1 , -1 , -1 , -1 , -1 , -1 , -1 , //31
890 BY_IDX, GY_IDX, RY_IDX, -1 , -1 , -1 , -1 , -1 , //32
891 BU_IDX, GU_IDX, RU_IDX, -1 , -1 , -1 , -1 , -1 , //33
892 BV_IDX, GV_IDX, RV_IDX, -1 , -1 , -1 , -1 , -1 , //34
895 dstRange = 0; //FIXME range = 1 is handled elsewhere
905 W = ROUNDED_DIV(ONE*ONE*ug, ub);
906 V = ROUNDED_DIV(ONE*ONE*vg, vr);
909 Cy = ROUNDED_DIV(cy*Z, ONE);
910 Cu = ROUNDED_DIV(ub*Z, ONE);
911 Cv = ROUNDED_DIV(vr*Z, ONE);
913 c->input_rgb2yuv_table[RY_IDX] = -ROUNDED_DIV((1 << RGB2YUV_SHIFT)*V , Cy);
914 c->input_rgb2yuv_table[GY_IDX] = ROUNDED_DIV((1 << RGB2YUV_SHIFT)*ONE*ONE , Cy);
915 c->input_rgb2yuv_table[BY_IDX] = -ROUNDED_DIV((1 << RGB2YUV_SHIFT)*W , Cy);
917 c->input_rgb2yuv_table[RU_IDX] = ROUNDED_DIV((1 << RGB2YUV_SHIFT)*V , Cu);
918 c->input_rgb2yuv_table[GU_IDX] = -ROUNDED_DIV((1 << RGB2YUV_SHIFT)*ONE*ONE , Cu);
919 c->input_rgb2yuv_table[BU_IDX] = ROUNDED_DIV((1 << RGB2YUV_SHIFT)*(Z+W) , Cu);
921 c->input_rgb2yuv_table[RV_IDX] = ROUNDED_DIV((1 << RGB2YUV_SHIFT)*(V+Z) , Cv);
922 c->input_rgb2yuv_table[GV_IDX] = -ROUNDED_DIV((1 << RGB2YUV_SHIFT)*ONE*ONE , Cv);
923 c->input_rgb2yuv_table[BV_IDX] = ROUNDED_DIV((1 << RGB2YUV_SHIFT)*W , Cv);
925 if(/*!dstRange && */!memcmp(table, ff_yuv2rgb_coeffs[SWS_CS_DEFAULT], sizeof(ff_yuv2rgb_coeffs[SWS_CS_DEFAULT]))) {
926 c->input_rgb2yuv_table[BY_IDX] = ((int)(0.114 * 219 / 255 * (1 << RGB2YUV_SHIFT) + 0.5));
927 c->input_rgb2yuv_table[BV_IDX] = (-(int)(0.081 * 224 / 255 * (1 << RGB2YUV_SHIFT) + 0.5));
928 c->input_rgb2yuv_table[BU_IDX] = ((int)(0.500 * 224 / 255 * (1 << RGB2YUV_SHIFT) + 0.5));
929 c->input_rgb2yuv_table[GY_IDX] = ((int)(0.587 * 219 / 255 * (1 << RGB2YUV_SHIFT) + 0.5));
930 c->input_rgb2yuv_table[GV_IDX] = (-(int)(0.419 * 224 / 255 * (1 << RGB2YUV_SHIFT) + 0.5));
931 c->input_rgb2yuv_table[GU_IDX] = (-(int)(0.331 * 224 / 255 * (1 << RGB2YUV_SHIFT) + 0.5));
932 c->input_rgb2yuv_table[RY_IDX] = ((int)(0.299 * 219 / 255 * (1 << RGB2YUV_SHIFT) + 0.5));
933 c->input_rgb2yuv_table[RV_IDX] = ((int)(0.500 * 224 / 255 * (1 << RGB2YUV_SHIFT) + 0.5));
934 c->input_rgb2yuv_table[RU_IDX] = (-(int)(0.169 * 224 / 255 * (1 << RGB2YUV_SHIFT) + 0.5));
936 for(i=0; i<FF_ARRAY_ELEMS(map); i++)
937 AV_WL16(p + 16*4 + 2*i, map[i] >= 0 ? c->input_rgb2yuv_table[map[i]] : 0);
940 static void fill_xyztables(struct SwsContext *c)
943 double xyzgamma = XYZ_GAMMA;
944 double rgbgamma = 1.0 / RGB_GAMMA;
945 double xyzgammainv = 1.0 / XYZ_GAMMA;
946 double rgbgammainv = RGB_GAMMA;
947 static const int16_t xyz2rgb_matrix[3][4] = {
948 {13270, -6295, -2041},
950 { 228, -835, 4329} };
951 static const int16_t rgb2xyz_matrix[3][4] = {
955 static int16_t xyzgamma_tab[4096], rgbgamma_tab[4096], xyzgammainv_tab[4096], rgbgammainv_tab[4096];
957 memcpy(c->xyz2rgb_matrix, xyz2rgb_matrix, sizeof(c->xyz2rgb_matrix));
958 memcpy(c->rgb2xyz_matrix, rgb2xyz_matrix, sizeof(c->rgb2xyz_matrix));
959 c->xyzgamma = xyzgamma_tab;
960 c->rgbgamma = rgbgamma_tab;
961 c->xyzgammainv = xyzgammainv_tab;
962 c->rgbgammainv = rgbgammainv_tab;
964 if (rgbgamma_tab[4095])
967 /* set gamma vectors */
968 for (i = 0; i < 4096; i++) {
969 xyzgamma_tab[i] = lrint(pow(i / 4095.0, xyzgamma) * 4095.0);
970 rgbgamma_tab[i] = lrint(pow(i / 4095.0, rgbgamma) * 4095.0);
971 xyzgammainv_tab[i] = lrint(pow(i / 4095.0, xyzgammainv) * 4095.0);
972 rgbgammainv_tab[i] = lrint(pow(i / 4095.0, rgbgammainv) * 4095.0);
976 int sws_setColorspaceDetails(struct SwsContext *c, const int inv_table[4],
977 int srcRange, const int table[4], int dstRange,
978 int brightness, int contrast, int saturation)
980 const AVPixFmtDescriptor *desc_dst;
981 const AVPixFmtDescriptor *desc_src;
982 memmove(c->srcColorspaceTable, inv_table, sizeof(int) * 4);
983 memmove(c->dstColorspaceTable, table, sizeof(int) * 4);
986 desc_dst = av_pix_fmt_desc_get(c->dstFormat);
987 desc_src = av_pix_fmt_desc_get(c->srcFormat);
989 if(!isYUV(c->dstFormat) && !isGray(c->dstFormat))
991 if(!isYUV(c->srcFormat) && !isGray(c->srcFormat))
994 c->brightness = brightness;
995 c->contrast = contrast;
996 c->saturation = saturation;
997 c->srcRange = srcRange;
998 c->dstRange = dstRange;
1000 if ((isYUV(c->dstFormat) || isGray(c->dstFormat)) && (isYUV(c->srcFormat) || isGray(c->srcFormat)))
1003 c->dstFormatBpp = av_get_bits_per_pixel(desc_dst);
1004 c->srcFormatBpp = av_get_bits_per_pixel(desc_src);
1006 if (!isYUV(c->dstFormat) && !isGray(c->dstFormat)) {
1007 ff_yuv2rgb_c_init_tables(c, inv_table, srcRange, brightness,
1008 contrast, saturation);
1012 ff_yuv2rgb_init_tables_ppc(c, inv_table, brightness,
1013 contrast, saturation);
1016 fill_rgb2yuv_table(c, table, dstRange);
1021 int sws_getColorspaceDetails(struct SwsContext *c, int **inv_table,
1022 int *srcRange, int **table, int *dstRange,
1023 int *brightness, int *contrast, int *saturation)
1028 *inv_table = c->srcColorspaceTable;
1029 *table = c->dstColorspaceTable;
1030 *srcRange = c->srcRange;
1031 *dstRange = c->dstRange;
1032 *brightness = c->brightness;
1033 *contrast = c->contrast;
1034 *saturation = c->saturation;
1039 static int handle_jpeg(enum AVPixelFormat *format)
1042 case AV_PIX_FMT_YUVJ420P:
1043 *format = AV_PIX_FMT_YUV420P;
1045 case AV_PIX_FMT_YUVJ411P:
1046 *format = AV_PIX_FMT_YUV411P;
1048 case AV_PIX_FMT_YUVJ422P:
1049 *format = AV_PIX_FMT_YUV422P;
1051 case AV_PIX_FMT_YUVJ444P:
1052 *format = AV_PIX_FMT_YUV444P;
1054 case AV_PIX_FMT_YUVJ440P:
1055 *format = AV_PIX_FMT_YUV440P;
1057 case AV_PIX_FMT_GRAY8:
1064 static int handle_0alpha(enum AVPixelFormat *format)
1067 case AV_PIX_FMT_0BGR : *format = AV_PIX_FMT_ABGR ; return 1;
1068 case AV_PIX_FMT_BGR0 : *format = AV_PIX_FMT_BGRA ; return 4;
1069 case AV_PIX_FMT_0RGB : *format = AV_PIX_FMT_ARGB ; return 1;
1070 case AV_PIX_FMT_RGB0 : *format = AV_PIX_FMT_RGBA ; return 4;
1075 static int handle_xyz(enum AVPixelFormat *format)
1078 case AV_PIX_FMT_XYZ12BE : *format = AV_PIX_FMT_RGB48BE; return 1;
1079 case AV_PIX_FMT_XYZ12LE : *format = AV_PIX_FMT_RGB48LE; return 1;
1084 static void handle_formats(SwsContext *c)
1086 c->src0Alpha |= handle_0alpha(&c->srcFormat);
1087 c->dst0Alpha |= handle_0alpha(&c->dstFormat);
1088 c->srcXYZ |= handle_xyz(&c->srcFormat);
1089 c->dstXYZ |= handle_xyz(&c->dstFormat);
1090 if (c->srcXYZ || c->dstXYZ)
1094 SwsContext *sws_alloc_context(void)
1096 SwsContext *c = av_mallocz(sizeof(SwsContext));
1099 c->av_class = &sws_context_class;
1100 av_opt_set_defaults(c);
1106 av_cold int sws_init_context(SwsContext *c, SwsFilter *srcFilter,
1107 SwsFilter *dstFilter)
1110 int usesVFilter, usesHFilter;
1112 SwsFilter dummyFilter = { NULL, NULL, NULL, NULL };
1117 int dst_stride = FFALIGN(dstW * sizeof(int16_t) + 66, 16);
1118 int flags, cpu_flags;
1119 enum AVPixelFormat srcFormat = c->srcFormat;
1120 enum AVPixelFormat dstFormat = c->dstFormat;
1121 const AVPixFmtDescriptor *desc_src;
1122 const AVPixFmtDescriptor *desc_dst;
1124 cpu_flags = av_get_cpu_flags();
1130 unscaled = (srcW == dstW && srcH == dstH);
1132 c->srcRange |= handle_jpeg(&c->srcFormat);
1133 c->dstRange |= handle_jpeg(&c->dstFormat);
1135 if(srcFormat!=c->srcFormat || dstFormat!=c->dstFormat)
1136 av_log(c, AV_LOG_WARNING, "deprecated pixel format used, make sure you did set range correctly\n");
1138 if (!c->contrast && !c->saturation && !c->dstFormatBpp)
1139 sws_setColorspaceDetails(c, ff_yuv2rgb_coeffs[SWS_CS_DEFAULT], c->srcRange,
1140 ff_yuv2rgb_coeffs[SWS_CS_DEFAULT],
1141 c->dstRange, 0, 1 << 16, 1 << 16);
1144 srcFormat = c->srcFormat;
1145 dstFormat = c->dstFormat;
1146 desc_src = av_pix_fmt_desc_get(srcFormat);
1147 desc_dst = av_pix_fmt_desc_get(dstFormat);
1149 if (!(unscaled && sws_isSupportedEndiannessConversion(srcFormat) &&
1150 av_pix_fmt_swap_endianness(srcFormat) == dstFormat)) {
1151 if (!sws_isSupportedInput(srcFormat)) {
1152 av_log(c, AV_LOG_ERROR, "%s is not supported as input pixel format\n",
1153 av_get_pix_fmt_name(srcFormat));
1154 return AVERROR(EINVAL);
1156 if (!sws_isSupportedOutput(dstFormat)) {
1157 av_log(c, AV_LOG_ERROR, "%s is not supported as output pixel format\n",
1158 av_get_pix_fmt_name(dstFormat));
1159 return AVERROR(EINVAL);
1163 i = flags & (SWS_POINT |
1175 /* provide a default scaler if not set by caller */
1177 if (dstW < srcW && dstH < srcH)
1178 flags |= SWS_BICUBIC;
1179 else if (dstW > srcW && dstH > srcH)
1180 flags |= SWS_BICUBIC;
1182 flags |= SWS_BICUBIC;
1184 } else if (i & (i - 1)) {
1185 av_log(c, AV_LOG_ERROR,
1186 "Exactly one scaler algorithm must be chosen, got %X\n", i);
1187 return AVERROR(EINVAL);
1190 if (srcW < 1 || srcH < 1 || dstW < 1 || dstH < 1) {
1191 /* FIXME check if these are enough and try to lower them after
1192 * fixing the relevant parts of the code */
1193 av_log(c, AV_LOG_ERROR, "%dx%d -> %dx%d is invalid scaling dimension\n",
1194 srcW, srcH, dstW, dstH);
1195 return AVERROR(EINVAL);
1199 dstFilter = &dummyFilter;
1201 srcFilter = &dummyFilter;
1203 c->lumXInc = (((int64_t)srcW << 16) + (dstW >> 1)) / dstW;
1204 c->lumYInc = (((int64_t)srcH << 16) + (dstH >> 1)) / dstH;
1205 c->dstFormatBpp = av_get_bits_per_pixel(desc_dst);
1206 c->srcFormatBpp = av_get_bits_per_pixel(desc_src);
1207 c->vRounder = 4 * 0x0001000100010001ULL;
1209 usesVFilter = (srcFilter->lumV && srcFilter->lumV->length > 1) ||
1210 (srcFilter->chrV && srcFilter->chrV->length > 1) ||
1211 (dstFilter->lumV && dstFilter->lumV->length > 1) ||
1212 (dstFilter->chrV && dstFilter->chrV->length > 1);
1213 usesHFilter = (srcFilter->lumH && srcFilter->lumH->length > 1) ||
1214 (srcFilter->chrH && srcFilter->chrH->length > 1) ||
1215 (dstFilter->lumH && dstFilter->lumH->length > 1) ||
1216 (dstFilter->chrH && dstFilter->chrH->length > 1);
1218 av_pix_fmt_get_chroma_sub_sample(srcFormat, &c->chrSrcHSubSample, &c->chrSrcVSubSample);
1219 av_pix_fmt_get_chroma_sub_sample(dstFormat, &c->chrDstHSubSample, &c->chrDstVSubSample);
1221 if (isAnyRGB(dstFormat) && !(flags&SWS_FULL_CHR_H_INT)) {
1223 av_log(c, AV_LOG_DEBUG, "Forcing full internal H chroma due to odd output size\n");
1224 flags |= SWS_FULL_CHR_H_INT;
1228 if ( c->chrSrcHSubSample == 0
1229 && c->chrSrcVSubSample == 0
1230 && c->dither != SWS_DITHER_BAYER //SWS_FULL_CHR_H_INT is currently not supported with SWS_DITHER_BAYER
1231 && !(c->flags & SWS_FAST_BILINEAR)
1233 av_log(c, AV_LOG_DEBUG, "Forcing full internal H chroma due to input having non subsampled chroma\n");
1234 flags |= SWS_FULL_CHR_H_INT;
1239 if (c->dither == SWS_DITHER_AUTO) {
1240 if (flags & SWS_ERROR_DIFFUSION)
1241 c->dither = SWS_DITHER_ED;
1244 if(dstFormat == AV_PIX_FMT_BGR4_BYTE ||
1245 dstFormat == AV_PIX_FMT_RGB4_BYTE ||
1246 dstFormat == AV_PIX_FMT_BGR8 ||
1247 dstFormat == AV_PIX_FMT_RGB8) {
1248 if (c->dither == SWS_DITHER_AUTO)
1249 c->dither = (flags & SWS_FULL_CHR_H_INT) ? SWS_DITHER_ED : SWS_DITHER_BAYER;
1250 if (!(flags & SWS_FULL_CHR_H_INT)) {
1251 if (c->dither == SWS_DITHER_ED) {
1252 av_log(c, AV_LOG_DEBUG,
1253 "Desired dithering only supported in full chroma interpolation for destination format '%s'\n",
1254 av_get_pix_fmt_name(dstFormat));
1255 flags |= SWS_FULL_CHR_H_INT;
1259 if (flags & SWS_FULL_CHR_H_INT) {
1260 if (c->dither == SWS_DITHER_BAYER) {
1261 av_log(c, AV_LOG_DEBUG,
1262 "Ordered dither is not supported in full chroma interpolation for destination format '%s'\n",
1263 av_get_pix_fmt_name(dstFormat));
1264 c->dither = SWS_DITHER_ED;
1268 if (isPlanarRGB(dstFormat)) {
1269 if (!(flags & SWS_FULL_CHR_H_INT)) {
1270 av_log(c, AV_LOG_DEBUG,
1271 "%s output is not supported with half chroma resolution, switching to full\n",
1272 av_get_pix_fmt_name(dstFormat));
1273 flags |= SWS_FULL_CHR_H_INT;
1278 /* reuse chroma for 2 pixels RGB/BGR unless user wants full
1279 * chroma interpolation */
1280 if (flags & SWS_FULL_CHR_H_INT &&
1281 isAnyRGB(dstFormat) &&
1282 !isPlanarRGB(dstFormat) &&
1283 dstFormat != AV_PIX_FMT_RGBA &&
1284 dstFormat != AV_PIX_FMT_ARGB &&
1285 dstFormat != AV_PIX_FMT_BGRA &&
1286 dstFormat != AV_PIX_FMT_ABGR &&
1287 dstFormat != AV_PIX_FMT_RGB24 &&
1288 dstFormat != AV_PIX_FMT_BGR24 &&
1289 dstFormat != AV_PIX_FMT_BGR4_BYTE &&
1290 dstFormat != AV_PIX_FMT_RGB4_BYTE &&
1291 dstFormat != AV_PIX_FMT_BGR8 &&
1292 dstFormat != AV_PIX_FMT_RGB8
1294 av_log(c, AV_LOG_WARNING,
1295 "full chroma interpolation for destination format '%s' not yet implemented\n",
1296 av_get_pix_fmt_name(dstFormat));
1297 flags &= ~SWS_FULL_CHR_H_INT;
1300 if (isAnyRGB(dstFormat) && !(flags & SWS_FULL_CHR_H_INT))
1301 c->chrDstHSubSample = 1;
1303 // drop some chroma lines if the user wants it
1304 c->vChrDrop = (flags & SWS_SRC_V_CHR_DROP_MASK) >>
1305 SWS_SRC_V_CHR_DROP_SHIFT;
1306 c->chrSrcVSubSample += c->vChrDrop;
1308 /* drop every other pixel for chroma calculation unless user
1309 * wants full chroma */
1310 if (isAnyRGB(srcFormat) && !(flags & SWS_FULL_CHR_H_INP) &&
1311 srcFormat != AV_PIX_FMT_RGB8 && srcFormat != AV_PIX_FMT_BGR8 &&
1312 srcFormat != AV_PIX_FMT_RGB4 && srcFormat != AV_PIX_FMT_BGR4 &&
1313 srcFormat != AV_PIX_FMT_RGB4_BYTE && srcFormat != AV_PIX_FMT_BGR4_BYTE &&
1314 srcFormat != AV_PIX_FMT_GBRP9BE && srcFormat != AV_PIX_FMT_GBRP9LE &&
1315 srcFormat != AV_PIX_FMT_GBRP10BE && srcFormat != AV_PIX_FMT_GBRP10LE &&
1316 srcFormat != AV_PIX_FMT_GBRP12BE && srcFormat != AV_PIX_FMT_GBRP12LE &&
1317 srcFormat != AV_PIX_FMT_GBRP14BE && srcFormat != AV_PIX_FMT_GBRP14LE &&
1318 srcFormat != AV_PIX_FMT_GBRP16BE && srcFormat != AV_PIX_FMT_GBRP16LE &&
1319 ((dstW >> c->chrDstHSubSample) <= (srcW >> 1) ||
1320 (flags & SWS_FAST_BILINEAR)))
1321 c->chrSrcHSubSample = 1;
1323 // Note the FF_CEIL_RSHIFT is so that we always round toward +inf.
1324 c->chrSrcW = FF_CEIL_RSHIFT(srcW, c->chrSrcHSubSample);
1325 c->chrSrcH = FF_CEIL_RSHIFT(srcH, c->chrSrcVSubSample);
1326 c->chrDstW = FF_CEIL_RSHIFT(dstW, c->chrDstHSubSample);
1327 c->chrDstH = FF_CEIL_RSHIFT(dstH, c->chrDstVSubSample);
1329 FF_ALLOC_OR_GOTO(c, c->formatConvBuffer, FFALIGN(srcW*2+78, 16) * 2, fail);
1331 /* unscaled special cases */
1332 if (unscaled && !usesHFilter && !usesVFilter &&
1333 (c->srcRange == c->dstRange || isAnyRGB(dstFormat))) {
1334 ff_get_unscaled_swscale(c);
1337 if (flags & SWS_PRINT_INFO)
1338 av_log(c, AV_LOG_INFO,
1339 "using unscaled %s -> %s special converter\n",
1340 av_get_pix_fmt_name(srcFormat), av_get_pix_fmt_name(dstFormat));
1345 c->srcBpc = 1 + desc_src->comp[0].depth_minus1;
1348 c->dstBpc = 1 + desc_dst->comp[0].depth_minus1;
1351 if (isAnyRGB(srcFormat) || srcFormat == AV_PIX_FMT_PAL8)
1353 if (c->dstBpc == 16)
1356 if (INLINE_MMXEXT(cpu_flags) && c->srcBpc == 8 && c->dstBpc <= 14) {
1357 c->canMMXEXTBeUsed = dstW >= srcW && (dstW & 31) == 0 &&
1358 c->chrDstW >= c->chrSrcW &&
1360 if (!c->canMMXEXTBeUsed && dstW >= srcW && c->chrDstW >= c->chrSrcW && (srcW & 15) == 0
1362 && (flags & SWS_FAST_BILINEAR)) {
1363 if (flags & SWS_PRINT_INFO)
1364 av_log(c, AV_LOG_INFO,
1365 "output width is not a multiple of 32 -> no MMXEXT scaler\n");
1367 if (usesHFilter || isNBPS(c->srcFormat) || is16BPS(c->srcFormat) || isAnyRGB(c->srcFormat))
1368 c->canMMXEXTBeUsed = 0;
1370 c->canMMXEXTBeUsed = 0;
1372 c->chrXInc = (((int64_t)c->chrSrcW << 16) + (c->chrDstW >> 1)) / c->chrDstW;
1373 c->chrYInc = (((int64_t)c->chrSrcH << 16) + (c->chrDstH >> 1)) / c->chrDstH;
1375 /* Match pixel 0 of the src to pixel 0 of dst and match pixel n-2 of src
1376 * to pixel n-2 of dst, but only for the FAST_BILINEAR mode otherwise do
1378 * n-2 is the last chrominance sample available.
1379 * This is not perfect, but no one should notice the difference, the more
1380 * correct variant would be like the vertical one, but that would require
1381 * some special code for the first and last pixel */
1382 if (flags & SWS_FAST_BILINEAR) {
1383 if (c->canMMXEXTBeUsed) {
1387 // we don't use the x86 asm scaler if MMX is available
1388 else if (INLINE_MMX(cpu_flags) && c->dstBpc <= 14) {
1389 c->lumXInc = ((int64_t)(srcW - 2) << 16) / (dstW - 2) - 20;
1390 c->chrXInc = ((int64_t)(c->chrSrcW - 2) << 16) / (c->chrDstW - 2) - 20;
1394 #define USE_MMAP (HAVE_MMAP && HAVE_MPROTECT && defined MAP_ANONYMOUS)
1396 /* precalculate horizontal scaler filter coefficients */
1398 #if HAVE_MMXEXT_INLINE
1399 // can't downscale !!!
1400 if (c->canMMXEXTBeUsed && (flags & SWS_FAST_BILINEAR)) {
1401 c->lumMmxextFilterCodeSize = init_hscaler_mmxext(dstW, c->lumXInc, NULL,
1403 c->chrMmxextFilterCodeSize = init_hscaler_mmxext(c->chrDstW, c->chrXInc,
1404 NULL, NULL, NULL, 4);
1407 c->lumMmxextFilterCode = mmap(NULL, c->lumMmxextFilterCodeSize,
1408 PROT_READ | PROT_WRITE,
1409 MAP_PRIVATE | MAP_ANONYMOUS,
1411 c->chrMmxextFilterCode = mmap(NULL, c->chrMmxextFilterCodeSize,
1412 PROT_READ | PROT_WRITE,
1413 MAP_PRIVATE | MAP_ANONYMOUS,
1415 #elif HAVE_VIRTUALALLOC
1416 c->lumMmxextFilterCode = VirtualAlloc(NULL,
1417 c->lumMmxextFilterCodeSize,
1419 PAGE_EXECUTE_READWRITE);
1420 c->chrMmxextFilterCode = VirtualAlloc(NULL,
1421 c->chrMmxextFilterCodeSize,
1423 PAGE_EXECUTE_READWRITE);
1425 c->lumMmxextFilterCode = av_malloc(c->lumMmxextFilterCodeSize);
1426 c->chrMmxextFilterCode = av_malloc(c->chrMmxextFilterCodeSize);
1429 #ifdef MAP_ANONYMOUS
1430 if (c->lumMmxextFilterCode == MAP_FAILED || c->chrMmxextFilterCode == MAP_FAILED)
1432 if (!c->lumMmxextFilterCode || !c->chrMmxextFilterCode)
1435 av_log(c, AV_LOG_ERROR, "Failed to allocate MMX2FilterCode\n");
1436 return AVERROR(ENOMEM);
1439 FF_ALLOCZ_OR_GOTO(c, c->hLumFilter, (dstW / 8 + 8) * sizeof(int16_t), fail);
1440 FF_ALLOCZ_OR_GOTO(c, c->hChrFilter, (c->chrDstW / 4 + 8) * sizeof(int16_t), fail);
1441 FF_ALLOCZ_OR_GOTO(c, c->hLumFilterPos, (dstW / 2 / 8 + 8) * sizeof(int32_t), fail);
1442 FF_ALLOCZ_OR_GOTO(c, c->hChrFilterPos, (c->chrDstW / 2 / 4 + 8) * sizeof(int32_t), fail);
1444 init_hscaler_mmxext( dstW, c->lumXInc, c->lumMmxextFilterCode,
1445 c->hLumFilter, (uint32_t*)c->hLumFilterPos, 8);
1446 init_hscaler_mmxext(c->chrDstW, c->chrXInc, c->chrMmxextFilterCode,
1447 c->hChrFilter, (uint32_t*)c->hChrFilterPos, 4);
1450 if ( mprotect(c->lumMmxextFilterCode, c->lumMmxextFilterCodeSize, PROT_EXEC | PROT_READ) == -1
1451 || mprotect(c->chrMmxextFilterCode, c->chrMmxextFilterCodeSize, PROT_EXEC | PROT_READ) == -1) {
1452 av_log(c, AV_LOG_ERROR, "mprotect failed, cannot use fast bilinear scaler\n");
1457 #endif /* HAVE_MMXEXT_INLINE */
1459 const int filterAlign = X86_MMX(cpu_flags) ? 4 :
1460 PPC_ALTIVEC(cpu_flags) ? 8 : 1;
1462 if (initFilter(&c->hLumFilter, &c->hLumFilterPos,
1463 &c->hLumFilterSize, c->lumXInc,
1464 srcW, dstW, filterAlign, 1 << 14,
1465 (flags & SWS_BICUBLIN) ? (flags | SWS_BICUBIC) : flags,
1466 cpu_flags, srcFilter->lumH, dstFilter->lumH,
1468 get_local_pos(c, 0, 0, 0),
1469 get_local_pos(c, 0, 0, 0)) < 0)
1471 if (initFilter(&c->hChrFilter, &c->hChrFilterPos,
1472 &c->hChrFilterSize, c->chrXInc,
1473 c->chrSrcW, c->chrDstW, filterAlign, 1 << 14,
1474 (flags & SWS_BICUBLIN) ? (flags | SWS_BILINEAR) : flags,
1475 cpu_flags, srcFilter->chrH, dstFilter->chrH,
1477 get_local_pos(c, c->chrSrcHSubSample, c->src_h_chr_pos, 0),
1478 get_local_pos(c, c->chrDstHSubSample, c->dst_h_chr_pos, 0)) < 0)
1481 } // initialize horizontal stuff
1483 /* precalculate vertical scaler filter coefficients */
1485 const int filterAlign = X86_MMX(cpu_flags) ? 2 :
1486 PPC_ALTIVEC(cpu_flags) ? 8 : 1;
1488 if (initFilter(&c->vLumFilter, &c->vLumFilterPos, &c->vLumFilterSize,
1489 c->lumYInc, srcH, dstH, filterAlign, (1 << 12),
1490 (flags & SWS_BICUBLIN) ? (flags | SWS_BICUBIC) : flags,
1491 cpu_flags, srcFilter->lumV, dstFilter->lumV,
1493 get_local_pos(c, 0, 0, 1),
1494 get_local_pos(c, 0, 0, 1)) < 0)
1496 if (initFilter(&c->vChrFilter, &c->vChrFilterPos, &c->vChrFilterSize,
1497 c->chrYInc, c->chrSrcH, c->chrDstH,
1498 filterAlign, (1 << 12),
1499 (flags & SWS_BICUBLIN) ? (flags | SWS_BILINEAR) : flags,
1500 cpu_flags, srcFilter->chrV, dstFilter->chrV,
1502 get_local_pos(c, c->chrSrcVSubSample, c->src_v_chr_pos, 1),
1503 get_local_pos(c, c->chrDstVSubSample, c->dst_v_chr_pos, 1)) < 0)
1508 FF_ALLOC_OR_GOTO(c, c->vYCoeffsBank, sizeof(vector signed short) * c->vLumFilterSize * c->dstH, fail);
1509 FF_ALLOC_OR_GOTO(c, c->vCCoeffsBank, sizeof(vector signed short) * c->vChrFilterSize * c->chrDstH, fail);
1511 for (i = 0; i < c->vLumFilterSize * c->dstH; i++) {
1513 short *p = (short *)&c->vYCoeffsBank[i];
1514 for (j = 0; j < 8; j++)
1515 p[j] = c->vLumFilter[i];
1518 for (i = 0; i < c->vChrFilterSize * c->chrDstH; i++) {
1520 short *p = (short *)&c->vCCoeffsBank[i];
1521 for (j = 0; j < 8; j++)
1522 p[j] = c->vChrFilter[i];
1527 // calculate buffer sizes so that they won't run out while handling these damn slices
1528 c->vLumBufSize = c->vLumFilterSize;
1529 c->vChrBufSize = c->vChrFilterSize;
1530 for (i = 0; i < dstH; i++) {
1531 int chrI = (int64_t)i * c->chrDstH / dstH;
1532 int nextSlice = FFMAX(c->vLumFilterPos[i] + c->vLumFilterSize - 1,
1533 ((c->vChrFilterPos[chrI] + c->vChrFilterSize - 1)
1534 << c->chrSrcVSubSample));
1536 nextSlice >>= c->chrSrcVSubSample;
1537 nextSlice <<= c->chrSrcVSubSample;
1538 if (c->vLumFilterPos[i] + c->vLumBufSize < nextSlice)
1539 c->vLumBufSize = nextSlice - c->vLumFilterPos[i];
1540 if (c->vChrFilterPos[chrI] + c->vChrBufSize <
1541 (nextSlice >> c->chrSrcVSubSample))
1542 c->vChrBufSize = (nextSlice >> c->chrSrcVSubSample) -
1543 c->vChrFilterPos[chrI];
1546 for (i = 0; i < 4; i++)
1547 FF_ALLOCZ_OR_GOTO(c, c->dither_error[i], (c->dstW+2) * sizeof(int), fail);
1549 /* Allocate pixbufs (we use dynamic allocation because otherwise we would
1550 * need to allocate several megabytes to handle all possible cases) */
1551 FF_ALLOC_OR_GOTO(c, c->lumPixBuf, c->vLumBufSize * 3 * sizeof(int16_t *), fail);
1552 FF_ALLOC_OR_GOTO(c, c->chrUPixBuf, c->vChrBufSize * 3 * sizeof(int16_t *), fail);
1553 FF_ALLOC_OR_GOTO(c, c->chrVPixBuf, c->vChrBufSize * 3 * sizeof(int16_t *), fail);
1554 if (CONFIG_SWSCALE_ALPHA && isALPHA(c->srcFormat) && isALPHA(c->dstFormat))
1555 FF_ALLOCZ_OR_GOTO(c, c->alpPixBuf, c->vLumBufSize * 3 * sizeof(int16_t *), fail);
1556 /* Note we need at least one pixel more at the end because of the MMX code
1557 * (just in case someone wants to replace the 4000/8000). */
1558 /* align at 16 bytes for AltiVec */
1559 for (i = 0; i < c->vLumBufSize; i++) {
1560 FF_ALLOCZ_OR_GOTO(c, c->lumPixBuf[i + c->vLumBufSize],
1561 dst_stride + 16, fail);
1562 c->lumPixBuf[i] = c->lumPixBuf[i + c->vLumBufSize];
1564 // 64 / c->scalingBpp is the same as 16 / sizeof(scaling_intermediate)
1565 c->uv_off = (dst_stride>>1) + 64 / (c->dstBpc &~ 7);
1566 c->uv_offx2 = dst_stride + 16;
1567 for (i = 0; i < c->vChrBufSize; i++) {
1568 FF_ALLOC_OR_GOTO(c, c->chrUPixBuf[i + c->vChrBufSize],
1569 dst_stride * 2 + 32, fail);
1570 c->chrUPixBuf[i] = c->chrUPixBuf[i + c->vChrBufSize];
1571 c->chrVPixBuf[i] = c->chrVPixBuf[i + c->vChrBufSize]
1572 = c->chrUPixBuf[i] + (dst_stride >> 1) + 8;
1574 if (CONFIG_SWSCALE_ALPHA && c->alpPixBuf)
1575 for (i = 0; i < c->vLumBufSize; i++) {
1576 FF_ALLOCZ_OR_GOTO(c, c->alpPixBuf[i + c->vLumBufSize],
1577 dst_stride + 16, fail);
1578 c->alpPixBuf[i] = c->alpPixBuf[i + c->vLumBufSize];
1581 // try to avoid drawing green stuff between the right end and the stride end
1582 for (i = 0; i < c->vChrBufSize; i++)
1583 if(desc_dst->comp[0].depth_minus1 == 15){
1584 av_assert0(c->dstBpc > 14);
1585 for(j=0; j<dst_stride/2+1; j++)
1586 ((int32_t*)(c->chrUPixBuf[i]))[j] = 1<<18;
1588 for(j=0; j<dst_stride+1; j++)
1589 ((int16_t*)(c->chrUPixBuf[i]))[j] = 1<<14;
1591 av_assert0(c->chrDstH <= dstH);
1593 if (flags & SWS_PRINT_INFO) {
1594 const char *scaler = NULL, *cpucaps;
1596 for (i = 0; i < FF_ARRAY_ELEMS(scale_algorithms); i++) {
1597 if (flags & scale_algorithms[i].flag) {
1598 scaler = scale_algorithms[i].description;
1603 scaler = "ehh flags invalid?!";
1604 av_log(c, AV_LOG_INFO, "%s scaler, from %s to %s%s ",
1606 av_get_pix_fmt_name(srcFormat),
1608 dstFormat == AV_PIX_FMT_BGR555 || dstFormat == AV_PIX_FMT_BGR565 ||
1609 dstFormat == AV_PIX_FMT_RGB444BE || dstFormat == AV_PIX_FMT_RGB444LE ||
1610 dstFormat == AV_PIX_FMT_BGR444BE || dstFormat == AV_PIX_FMT_BGR444LE ?
1615 av_get_pix_fmt_name(dstFormat));
1617 if (INLINE_MMXEXT(cpu_flags))
1619 else if (INLINE_AMD3DNOW(cpu_flags))
1621 else if (INLINE_MMX(cpu_flags))
1623 else if (PPC_ALTIVEC(cpu_flags))
1624 cpucaps = "AltiVec";
1628 av_log(c, AV_LOG_INFO, "using %s\n", cpucaps);
1630 av_log(c, AV_LOG_VERBOSE, "%dx%d -> %dx%d\n", srcW, srcH, dstW, dstH);
1631 av_log(c, AV_LOG_DEBUG,
1632 "lum srcW=%d srcH=%d dstW=%d dstH=%d xInc=%d yInc=%d\n",
1633 c->srcW, c->srcH, c->dstW, c->dstH, c->lumXInc, c->lumYInc);
1634 av_log(c, AV_LOG_DEBUG,
1635 "chr srcW=%d srcH=%d dstW=%d dstH=%d xInc=%d yInc=%d\n",
1636 c->chrSrcW, c->chrSrcH, c->chrDstW, c->chrDstH,
1637 c->chrXInc, c->chrYInc);
1640 c->swscale = ff_getSwsFunc(c);
1642 fail: // FIXME replace things by appropriate error codes
1646 #if FF_API_SWS_GETCONTEXT
1647 SwsContext *sws_getContext(int srcW, int srcH, enum AVPixelFormat srcFormat,
1648 int dstW, int dstH, enum AVPixelFormat dstFormat,
1649 int flags, SwsFilter *srcFilter,
1650 SwsFilter *dstFilter, const double *param)
1654 if (!(c = sws_alloc_context()))
1662 c->srcFormat = srcFormat;
1663 c->dstFormat = dstFormat;
1666 c->param[0] = param[0];
1667 c->param[1] = param[1];
1670 if (sws_init_context(c, srcFilter, dstFilter) < 0) {
1679 SwsFilter *sws_getDefaultFilter(float lumaGBlur, float chromaGBlur,
1680 float lumaSharpen, float chromaSharpen,
1681 float chromaHShift, float chromaVShift,
1684 SwsFilter *filter = av_malloc(sizeof(SwsFilter));
1688 if (lumaGBlur != 0.0) {
1689 filter->lumH = sws_getGaussianVec(lumaGBlur, 3.0);
1690 filter->lumV = sws_getGaussianVec(lumaGBlur, 3.0);
1692 filter->lumH = sws_getIdentityVec();
1693 filter->lumV = sws_getIdentityVec();
1696 if (chromaGBlur != 0.0) {
1697 filter->chrH = sws_getGaussianVec(chromaGBlur, 3.0);
1698 filter->chrV = sws_getGaussianVec(chromaGBlur, 3.0);
1700 filter->chrH = sws_getIdentityVec();
1701 filter->chrV = sws_getIdentityVec();
1704 if (chromaSharpen != 0.0) {
1705 SwsVector *id = sws_getIdentityVec();
1706 sws_scaleVec(filter->chrH, -chromaSharpen);
1707 sws_scaleVec(filter->chrV, -chromaSharpen);
1708 sws_addVec(filter->chrH, id);
1709 sws_addVec(filter->chrV, id);
1713 if (lumaSharpen != 0.0) {
1714 SwsVector *id = sws_getIdentityVec();
1715 sws_scaleVec(filter->lumH, -lumaSharpen);
1716 sws_scaleVec(filter->lumV, -lumaSharpen);
1717 sws_addVec(filter->lumH, id);
1718 sws_addVec(filter->lumV, id);
1722 if (chromaHShift != 0.0)
1723 sws_shiftVec(filter->chrH, (int)(chromaHShift + 0.5));
1725 if (chromaVShift != 0.0)
1726 sws_shiftVec(filter->chrV, (int)(chromaVShift + 0.5));
1728 sws_normalizeVec(filter->chrH, 1.0);
1729 sws_normalizeVec(filter->chrV, 1.0);
1730 sws_normalizeVec(filter->lumH, 1.0);
1731 sws_normalizeVec(filter->lumV, 1.0);
1734 sws_printVec2(filter->chrH, NULL, AV_LOG_DEBUG);
1736 sws_printVec2(filter->lumH, NULL, AV_LOG_DEBUG);
1741 SwsVector *sws_allocVec(int length)
1745 if(length <= 0 || length > INT_MAX/ sizeof(double))
1748 vec = av_malloc(sizeof(SwsVector));
1751 vec->length = length;
1752 vec->coeff = av_malloc(sizeof(double) * length);
1758 SwsVector *sws_getGaussianVec(double variance, double quality)
1760 const int length = (int)(variance * quality + 0.5) | 1;
1762 double middle = (length - 1) * 0.5;
1765 if(variance < 0 || quality < 0)
1768 vec = sws_allocVec(length);
1773 for (i = 0; i < length; i++) {
1774 double dist = i - middle;
1775 vec->coeff[i] = exp(-dist * dist / (2 * variance * variance)) /
1776 sqrt(2 * variance * M_PI);
1779 sws_normalizeVec(vec, 1.0);
1784 SwsVector *sws_getConstVec(double c, int length)
1787 SwsVector *vec = sws_allocVec(length);
1792 for (i = 0; i < length; i++)
1798 SwsVector *sws_getIdentityVec(void)
1800 return sws_getConstVec(1.0, 1);
1803 static double sws_dcVec(SwsVector *a)
1808 for (i = 0; i < a->length; i++)
1814 void sws_scaleVec(SwsVector *a, double scalar)
1818 for (i = 0; i < a->length; i++)
1819 a->coeff[i] *= scalar;
1822 void sws_normalizeVec(SwsVector *a, double height)
1824 sws_scaleVec(a, height / sws_dcVec(a));
1827 static SwsVector *sws_getConvVec(SwsVector *a, SwsVector *b)
1829 int length = a->length + b->length - 1;
1831 SwsVector *vec = sws_getConstVec(0.0, length);
1836 for (i = 0; i < a->length; i++) {
1837 for (j = 0; j < b->length; j++) {
1838 vec->coeff[i + j] += a->coeff[i] * b->coeff[j];
1845 static SwsVector *sws_sumVec(SwsVector *a, SwsVector *b)
1847 int length = FFMAX(a->length, b->length);
1849 SwsVector *vec = sws_getConstVec(0.0, length);
1854 for (i = 0; i < a->length; i++)
1855 vec->coeff[i + (length - 1) / 2 - (a->length - 1) / 2] += a->coeff[i];
1856 for (i = 0; i < b->length; i++)
1857 vec->coeff[i + (length - 1) / 2 - (b->length - 1) / 2] += b->coeff[i];
1862 static SwsVector *sws_diffVec(SwsVector *a, SwsVector *b)
1864 int length = FFMAX(a->length, b->length);
1866 SwsVector *vec = sws_getConstVec(0.0, length);
1871 for (i = 0; i < a->length; i++)
1872 vec->coeff[i + (length - 1) / 2 - (a->length - 1) / 2] += a->coeff[i];
1873 for (i = 0; i < b->length; i++)
1874 vec->coeff[i + (length - 1) / 2 - (b->length - 1) / 2] -= b->coeff[i];
1879 /* shift left / or right if "shift" is negative */
1880 static SwsVector *sws_getShiftedVec(SwsVector *a, int shift)
1882 int length = a->length + FFABS(shift) * 2;
1884 SwsVector *vec = sws_getConstVec(0.0, length);
1889 for (i = 0; i < a->length; i++) {
1890 vec->coeff[i + (length - 1) / 2 -
1891 (a->length - 1) / 2 - shift] = a->coeff[i];
1897 void sws_shiftVec(SwsVector *a, int shift)
1899 SwsVector *shifted = sws_getShiftedVec(a, shift);
1901 a->coeff = shifted->coeff;
1902 a->length = shifted->length;
1906 void sws_addVec(SwsVector *a, SwsVector *b)
1908 SwsVector *sum = sws_sumVec(a, b);
1910 a->coeff = sum->coeff;
1911 a->length = sum->length;
1915 void sws_subVec(SwsVector *a, SwsVector *b)
1917 SwsVector *diff = sws_diffVec(a, b);
1919 a->coeff = diff->coeff;
1920 a->length = diff->length;
1924 void sws_convVec(SwsVector *a, SwsVector *b)
1926 SwsVector *conv = sws_getConvVec(a, b);
1928 a->coeff = conv->coeff;
1929 a->length = conv->length;
1933 SwsVector *sws_cloneVec(SwsVector *a)
1935 SwsVector *vec = sws_allocVec(a->length);
1940 memcpy(vec->coeff, a->coeff, a->length * sizeof(*a->coeff));
1945 void sws_printVec2(SwsVector *a, AVClass *log_ctx, int log_level)
1952 for (i = 0; i < a->length; i++)
1953 if (a->coeff[i] > max)
1956 for (i = 0; i < a->length; i++)
1957 if (a->coeff[i] < min)
1962 for (i = 0; i < a->length; i++) {
1963 int x = (int)((a->coeff[i] - min) * 60.0 / range + 0.5);
1964 av_log(log_ctx, log_level, "%1.3f ", a->coeff[i]);
1966 av_log(log_ctx, log_level, " ");
1967 av_log(log_ctx, log_level, "|\n");
1971 void sws_freeVec(SwsVector *a)
1975 av_freep(&a->coeff);
1980 void sws_freeFilter(SwsFilter *filter)
1985 sws_freeVec(filter->lumH);
1986 sws_freeVec(filter->lumV);
1987 sws_freeVec(filter->chrH);
1988 sws_freeVec(filter->chrV);
1992 void sws_freeContext(SwsContext *c)
1999 for (i = 0; i < c->vLumBufSize; i++)
2000 av_freep(&c->lumPixBuf[i]);
2001 av_freep(&c->lumPixBuf);
2004 if (c->chrUPixBuf) {
2005 for (i = 0; i < c->vChrBufSize; i++)
2006 av_freep(&c->chrUPixBuf[i]);
2007 av_freep(&c->chrUPixBuf);
2008 av_freep(&c->chrVPixBuf);
2011 if (CONFIG_SWSCALE_ALPHA && c->alpPixBuf) {
2012 for (i = 0; i < c->vLumBufSize; i++)
2013 av_freep(&c->alpPixBuf[i]);
2014 av_freep(&c->alpPixBuf);
2017 for (i = 0; i < 4; i++)
2018 av_freep(&c->dither_error[i]);
2020 av_freep(&c->vLumFilter);
2021 av_freep(&c->vChrFilter);
2022 av_freep(&c->hLumFilter);
2023 av_freep(&c->hChrFilter);
2025 av_freep(&c->vYCoeffsBank);
2026 av_freep(&c->vCCoeffsBank);
2029 av_freep(&c->vLumFilterPos);
2030 av_freep(&c->vChrFilterPos);
2031 av_freep(&c->hLumFilterPos);
2032 av_freep(&c->hChrFilterPos);
2036 if (c->lumMmxextFilterCode)
2037 munmap(c->lumMmxextFilterCode, c->lumMmxextFilterCodeSize);
2038 if (c->chrMmxextFilterCode)
2039 munmap(c->chrMmxextFilterCode, c->chrMmxextFilterCodeSize);
2040 #elif HAVE_VIRTUALALLOC
2041 if (c->lumMmxextFilterCode)
2042 VirtualFree(c->lumMmxextFilterCode, 0, MEM_RELEASE);
2043 if (c->chrMmxextFilterCode)
2044 VirtualFree(c->chrMmxextFilterCode, 0, MEM_RELEASE);
2046 av_free(c->lumMmxextFilterCode);
2047 av_free(c->chrMmxextFilterCode);
2049 c->lumMmxextFilterCode = NULL;
2050 c->chrMmxextFilterCode = NULL;
2051 #endif /* HAVE_MMX_INLINE */
2053 av_freep(&c->yuvTable);
2054 av_freep(&c->formatConvBuffer);
2059 struct SwsContext *sws_getCachedContext(struct SwsContext *context, int srcW,
2060 int srcH, enum AVPixelFormat srcFormat,
2062 enum AVPixelFormat dstFormat, int flags,
2063 SwsFilter *srcFilter,
2064 SwsFilter *dstFilter,
2065 const double *param)
2067 static const double default_param[2] = { SWS_PARAM_DEFAULT,
2068 SWS_PARAM_DEFAULT };
2071 param = default_param;
2074 (context->srcW != srcW ||
2075 context->srcH != srcH ||
2076 context->srcFormat != srcFormat ||
2077 context->dstW != dstW ||
2078 context->dstH != dstH ||
2079 context->dstFormat != dstFormat ||
2080 context->flags != flags ||
2081 context->param[0] != param[0] ||
2082 context->param[1] != param[1])) {
2083 sws_freeContext(context);
2088 if (!(context = sws_alloc_context()))
2090 context->srcW = srcW;
2091 context->srcH = srcH;
2092 context->srcFormat = srcFormat;
2093 context->dstW = dstW;
2094 context->dstH = dstH;
2095 context->dstFormat = dstFormat;
2096 context->flags = flags;
2097 context->param[0] = param[0];
2098 context->param[1] = param[1];
2099 if (sws_init_context(context, srcFilter, dstFilter) < 0) {
2100 sws_freeContext(context);