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
21 #define _SVID_SOURCE //needed for MAP_ANONYMOUS
22 #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 "swscale_internal.h"
42 #include "libavutil/intreadwrite.h"
43 #include "libavutil/x86_cpu.h"
44 #include "libavutil/cpu.h"
45 #include "libavutil/avutil.h"
46 #include "libavutil/bswap.h"
47 #include "libavutil/mathematics.h"
48 #include "libavutil/opt.h"
49 #include "libavutil/pixdesc.h"
50 #include "libavutil/avassert.h"
52 unsigned swscale_version(void)
54 av_assert0(LIBSWSCALE_VERSION_MICRO >= 100);
55 return LIBSWSCALE_VERSION_INT;
58 const char *swscale_configuration(void)
60 return FFMPEG_CONFIGURATION;
63 const char *swscale_license(void)
65 #define LICENSE_PREFIX "libswscale license: "
66 return LICENSE_PREFIX FFMPEG_LICENSE + sizeof(LICENSE_PREFIX) - 1;
69 #define RET 0xC3 //near return opcode for x86
71 typedef struct FormatEntry {
72 int is_supported_in, is_supported_out;
75 static const FormatEntry format_entries[PIX_FMT_NB] = {
76 [PIX_FMT_YUV420P] = { 1 , 1 },
77 [PIX_FMT_YUYV422] = { 1 , 1 },
78 [PIX_FMT_RGB24] = { 1 , 1 },
79 [PIX_FMT_BGR24] = { 1 , 1 },
80 [PIX_FMT_YUV422P] = { 1 , 1 },
81 [PIX_FMT_YUV444P] = { 1 , 1 },
82 [PIX_FMT_YUV410P] = { 1 , 1 },
83 [PIX_FMT_YUV411P] = { 1 , 1 },
84 [PIX_FMT_GRAY8] = { 1 , 1 },
85 [PIX_FMT_MONOWHITE] = { 1 , 1 },
86 [PIX_FMT_MONOBLACK] = { 1 , 1 },
87 [PIX_FMT_PAL8] = { 1 , 0 },
88 [PIX_FMT_YUVJ420P] = { 1 , 1 },
89 [PIX_FMT_YUVJ422P] = { 1 , 1 },
90 [PIX_FMT_YUVJ444P] = { 1 , 1 },
91 [PIX_FMT_UYVY422] = { 1 , 1 },
92 [PIX_FMT_UYYVYY411] = { 0 , 0 },
93 [PIX_FMT_BGR8] = { 1 , 1 },
94 [PIX_FMT_BGR4] = { 0 , 1 },
95 [PIX_FMT_BGR4_BYTE] = { 1 , 1 },
96 [PIX_FMT_RGB8] = { 1 , 1 },
97 [PIX_FMT_RGB4] = { 0 , 1 },
98 [PIX_FMT_RGB4_BYTE] = { 1 , 1 },
99 [PIX_FMT_NV12] = { 1 , 1 },
100 [PIX_FMT_NV21] = { 1 , 1 },
101 [PIX_FMT_ARGB] = { 1 , 1 },
102 [PIX_FMT_RGBA] = { 1 , 1 },
103 [PIX_FMT_ABGR] = { 1 , 1 },
104 [PIX_FMT_BGRA] = { 1 , 1 },
105 [PIX_FMT_0RGB] = { 1 , 1 },
106 [PIX_FMT_RGB0] = { 1 , 1 },
107 [PIX_FMT_0BGR] = { 1 , 1 },
108 [PIX_FMT_BGR0] = { 1 , 1 },
109 [PIX_FMT_GRAY16BE] = { 1 , 1 },
110 [PIX_FMT_GRAY16LE] = { 1 , 1 },
111 [PIX_FMT_YUV440P] = { 1 , 1 },
112 [PIX_FMT_YUVJ440P] = { 1 , 1 },
113 [PIX_FMT_YUVA420P] = { 1 , 1 },
114 [PIX_FMT_YUVA444P] = { 1 , 1 },
115 [PIX_FMT_RGB48BE] = { 1 , 1 },
116 [PIX_FMT_RGB48LE] = { 1 , 1 },
117 [PIX_FMT_RGBA64BE] = { 0 , 0 },
118 [PIX_FMT_RGBA64LE] = { 0 , 0 },
119 [PIX_FMT_RGB565BE] = { 1 , 1 },
120 [PIX_FMT_RGB565LE] = { 1 , 1 },
121 [PIX_FMT_RGB555BE] = { 1 , 1 },
122 [PIX_FMT_RGB555LE] = { 1 , 1 },
123 [PIX_FMT_BGR565BE] = { 1 , 1 },
124 [PIX_FMT_BGR565LE] = { 1 , 1 },
125 [PIX_FMT_BGR555BE] = { 1 , 1 },
126 [PIX_FMT_BGR555LE] = { 1 , 1 },
127 [PIX_FMT_YUV420P16LE] = { 1 , 1 },
128 [PIX_FMT_YUV420P16BE] = { 1 , 1 },
129 [PIX_FMT_YUV422P16LE] = { 1 , 1 },
130 [PIX_FMT_YUV422P16BE] = { 1 , 1 },
131 [PIX_FMT_YUV444P16LE] = { 1 , 1 },
132 [PIX_FMT_YUV444P16BE] = { 1 , 1 },
133 [PIX_FMT_RGB444LE] = { 1 , 1 },
134 [PIX_FMT_RGB444BE] = { 1 , 1 },
135 [PIX_FMT_BGR444LE] = { 1 , 1 },
136 [PIX_FMT_BGR444BE] = { 1 , 1 },
137 [PIX_FMT_Y400A] = { 1 , 0 },
138 [PIX_FMT_BGR48BE] = { 1 , 1 },
139 [PIX_FMT_BGR48LE] = { 1 , 1 },
140 [PIX_FMT_BGRA64BE] = { 0 , 0 },
141 [PIX_FMT_BGRA64LE] = { 0 , 0 },
142 [PIX_FMT_YUV420P9BE] = { 1 , 1 },
143 [PIX_FMT_YUV420P9LE] = { 1 , 1 },
144 [PIX_FMT_YUV420P10BE] = { 1 , 1 },
145 [PIX_FMT_YUV420P10LE] = { 1 , 1 },
146 [PIX_FMT_YUV422P9BE] = { 1 , 1 },
147 [PIX_FMT_YUV422P9LE] = { 1 , 1 },
148 [PIX_FMT_YUV422P10BE] = { 1 , 1 },
149 [PIX_FMT_YUV422P10LE] = { 1 , 1 },
150 [PIX_FMT_YUV444P9BE] = { 1 , 1 },
151 [PIX_FMT_YUV444P9LE] = { 1 , 1 },
152 [PIX_FMT_YUV444P10BE] = { 1 , 1 },
153 [PIX_FMT_YUV444P10LE] = { 1 , 1 },
154 [PIX_FMT_GBR24P] = { 1 , 0 },
155 [PIX_FMT_GBRP] = { 1 , 0 },
156 [PIX_FMT_GBRP9LE] = { 1 , 0 },
157 [PIX_FMT_GBRP9BE] = { 1 , 0 },
158 [PIX_FMT_GBRP10LE] = { 1 , 0 },
159 [PIX_FMT_GBRP10BE] = { 1 , 0 },
160 [PIX_FMT_GBRP16LE] = { 1 , 0 },
161 [PIX_FMT_GBRP16BE] = { 1 , 0 },
164 int sws_isSupportedInput(enum PixelFormat pix_fmt)
166 return (unsigned)pix_fmt < PIX_FMT_NB ?
167 format_entries[pix_fmt].is_supported_in : 0;
170 int sws_isSupportedOutput(enum PixelFormat pix_fmt)
172 return (unsigned)pix_fmt < PIX_FMT_NB ?
173 format_entries[pix_fmt].is_supported_out : 0;
176 extern const int32_t ff_yuv2rgb_coeffs[8][4];
178 #if FF_API_SWS_FORMAT_NAME
179 const char *sws_format_name(enum PixelFormat format)
181 return av_get_pix_fmt_name(format);
185 static double getSplineCoeff(double a, double b, double c, double d, double dist)
187 if (dist<=1.0) return ((d*dist + c)*dist + b)*dist +a;
188 else return getSplineCoeff( 0.0,
195 static int initFilter(int16_t **outFilter, int16_t **filterPos, int *outFilterSize, int xInc,
196 int srcW, int dstW, int filterAlign, int one, int flags, int cpu_flags,
197 SwsVector *srcFilter, SwsVector *dstFilter, double param[2])
203 int64_t *filter=NULL;
204 int64_t *filter2=NULL;
205 const int64_t fone= 1LL<<54;
208 emms_c(); //FIXME this should not be required but it IS (even for non-MMX versions)
210 // NOTE: the +3 is for the MMX(+1)/SSE(+3) scaler which reads over the end
211 FF_ALLOC_OR_GOTO(NULL, *filterPos, (dstW+3)*sizeof(int16_t), fail);
213 if (FFABS(xInc - 0x10000) <10) { // unscaled
216 FF_ALLOCZ_OR_GOTO(NULL, filter, dstW*sizeof(*filter)*filterSize, fail);
218 for (i=0; i<dstW; i++) {
219 filter[i*filterSize]= fone;
223 } else if (flags&SWS_POINT) { // lame looking point sampling mode
227 FF_ALLOC_OR_GOTO(NULL, filter, dstW*sizeof(*filter)*filterSize, fail);
229 xDstInSrc= xInc/2 - 0x8000;
230 for (i=0; i<dstW; i++) {
231 int xx= (xDstInSrc - ((filterSize-1)<<15) + (1<<15))>>16;
237 } else if ((xInc <= (1<<16) && (flags&SWS_AREA)) || (flags&SWS_FAST_BILINEAR)) { // bilinear upscale
241 FF_ALLOC_OR_GOTO(NULL, filter, dstW*sizeof(*filter)*filterSize, fail);
243 xDstInSrc= xInc/2 - 0x8000;
244 for (i=0; i<dstW; i++) {
245 int xx= (xDstInSrc - ((filterSize-1)<<15) + (1<<15))>>16;
249 //bilinear upscale / linear interpolate / area averaging
250 for (j=0; j<filterSize; j++) {
251 int64_t coeff= fone - FFABS(((int64_t)xx<<16) - xDstInSrc)*(fone>>16);
252 if (coeff<0) coeff=0;
253 filter[i*filterSize + j]= coeff;
262 if (flags&SWS_BICUBIC) sizeFactor= 4;
263 else if (flags&SWS_X) sizeFactor= 8;
264 else if (flags&SWS_AREA) sizeFactor= 1; //downscale only, for upscale it is bilinear
265 else if (flags&SWS_GAUSS) sizeFactor= 8; // infinite ;)
266 else if (flags&SWS_LANCZOS) sizeFactor= param[0] != SWS_PARAM_DEFAULT ? ceil(2*param[0]) : 6;
267 else if (flags&SWS_SINC) sizeFactor= 20; // infinite ;)
268 else if (flags&SWS_SPLINE) sizeFactor= 20; // infinite ;)
269 else if (flags&SWS_BILINEAR) sizeFactor= 2;
271 sizeFactor= 0; //GCC warning killer
275 if (xInc <= 1<<16) filterSize= 1 + sizeFactor; // upscale
276 else filterSize= 1 + (sizeFactor*srcW + dstW - 1)/ dstW;
278 filterSize = FFMIN(filterSize, srcW - 2);
279 filterSize = FFMAX(filterSize, 1);
281 FF_ALLOC_OR_GOTO(NULL, filter, dstW*sizeof(*filter)*filterSize, fail);
283 xDstInSrc= xInc - 0x10000;
284 for (i=0; i<dstW; i++) {
285 int xx= (xDstInSrc - ((filterSize-2)<<16)) / (1<<17);
288 for (j=0; j<filterSize; j++) {
289 int64_t d= (FFABS(((int64_t)xx<<17) - xDstInSrc))<<13;
295 floatd= d * (1.0/(1<<30));
297 if (flags & SWS_BICUBIC) {
298 int64_t B= (param[0] != SWS_PARAM_DEFAULT ? param[0] : 0) * (1<<24);
299 int64_t C= (param[1] != SWS_PARAM_DEFAULT ? param[1] : 0.6) * (1<<24);
304 int64_t dd = (d * d) >> 30;
305 int64_t ddd = (dd * d) >> 30;
308 coeff = (12*(1<<24)-9*B-6*C)*ddd + (-18*(1<<24)+12*B+6*C)*dd + (6*(1<<24)-2*B)*(1<<30);
310 coeff = (-B-6*C)*ddd + (6*B+30*C)*dd + (-12*B-48*C)*d + (8*B+24*C)*(1<<30);
312 coeff *= fone>>(30+24);
314 /* else if (flags & SWS_X) {
315 double p= param ? param*0.01 : 0.3;
316 coeff = d ? sin(d*M_PI)/(d*M_PI) : 1.0;
317 coeff*= pow(2.0, - p*d*d);
319 else if (flags & SWS_X) {
320 double A= param[0] != SWS_PARAM_DEFAULT ? param[0] : 1.0;
324 c = cos(floatd*M_PI);
327 if (c<0.0) c= -pow(-c, A);
329 coeff= (c*0.5 + 0.5)*fone;
330 } else if (flags & SWS_AREA) {
331 int64_t d2= d - (1<<29);
332 if (d2*xInc < -(1LL<<(29+16))) coeff= 1.0 * (1LL<<(30+16));
333 else if (d2*xInc < (1LL<<(29+16))) coeff= -d2*xInc + (1LL<<(29+16));
335 coeff *= fone>>(30+16);
336 } else if (flags & SWS_GAUSS) {
337 double p= param[0] != SWS_PARAM_DEFAULT ? param[0] : 3.0;
338 coeff = (pow(2.0, - p*floatd*floatd))*fone;
339 } else if (flags & SWS_SINC) {
340 coeff = (d ? sin(floatd*M_PI)/(floatd*M_PI) : 1.0)*fone;
341 } else if (flags & SWS_LANCZOS) {
342 double p= param[0] != SWS_PARAM_DEFAULT ? param[0] : 3.0;
343 coeff = (d ? sin(floatd*M_PI)*sin(floatd*M_PI/p)/(floatd*floatd*M_PI*M_PI/p) : 1.0)*fone;
344 if (floatd>p) coeff=0;
345 } else if (flags & SWS_BILINEAR) {
347 if (coeff<0) coeff=0;
349 } else if (flags & SWS_SPLINE) {
350 double p=-2.196152422706632;
351 coeff = getSplineCoeff(1.0, 0.0, p, -p-1.0, floatd) * fone;
353 coeff= 0.0; //GCC warning killer
357 filter[i*filterSize + j]= coeff;
364 /* apply src & dst Filter to filter -> filter2
367 assert(filterSize>0);
368 filter2Size= filterSize;
369 if (srcFilter) filter2Size+= srcFilter->length - 1;
370 if (dstFilter) filter2Size+= dstFilter->length - 1;
371 assert(filter2Size>0);
372 FF_ALLOCZ_OR_GOTO(NULL, filter2, filter2Size*dstW*sizeof(*filter2), fail);
374 for (i=0; i<dstW; i++) {
378 for (k=0; k<srcFilter->length; k++) {
379 for (j=0; j<filterSize; j++)
380 filter2[i*filter2Size + k + j] += srcFilter->coeff[k]*filter[i*filterSize + j];
383 for (j=0; j<filterSize; j++)
384 filter2[i*filter2Size + j]= filter[i*filterSize + j];
388 (*filterPos)[i]+= (filterSize-1)/2 - (filter2Size-1)/2;
392 /* try to reduce the filter-size (step1 find size and shift left) */
393 // Assume it is near normalized (*0.5 or *2.0 is OK but * 0.001 is not).
395 for (i=dstW-1; i>=0; i--) {
396 int min= filter2Size;
400 /* get rid of near zero elements on the left by shifting left */
401 for (j=0; j<filter2Size; j++) {
403 cutOff += FFABS(filter2[i*filter2Size]);
405 if (cutOff > SWS_MAX_REDUCE_CUTOFF*fone) break;
407 /* preserve monotonicity because the core can't handle the filter otherwise */
408 if (i<dstW-1 && (*filterPos)[i] >= (*filterPos)[i+1]) break;
410 // move filter coefficients left
411 for (k=1; k<filter2Size; k++)
412 filter2[i*filter2Size + k - 1]= filter2[i*filter2Size + k];
413 filter2[i*filter2Size + k - 1]= 0;
418 /* count near zeros on the right */
419 for (j=filter2Size-1; j>0; j--) {
420 cutOff += FFABS(filter2[i*filter2Size + j]);
422 if (cutOff > SWS_MAX_REDUCE_CUTOFF*fone) break;
426 if (min>minFilterSize) minFilterSize= min;
429 if (HAVE_ALTIVEC && cpu_flags & AV_CPU_FLAG_ALTIVEC) {
430 // we can handle the special case 4,
431 // so we don't want to go to the full 8
432 if (minFilterSize < 5)
435 // We really don't want to waste our time
436 // doing useless computation, so fall back on
437 // the scalar C code for very small filters.
438 // Vectorizing is worth it only if you have a
439 // decent-sized vector.
440 if (minFilterSize < 3)
444 if (HAVE_MMX && cpu_flags & AV_CPU_FLAG_MMX) {
445 // special case for unscaled vertical filtering
446 if (minFilterSize == 1 && filterAlign == 2)
450 assert(minFilterSize > 0);
451 filterSize= (minFilterSize +(filterAlign-1)) & (~(filterAlign-1));
452 assert(filterSize > 0);
453 filter= av_malloc(filterSize*dstW*sizeof(*filter));
454 if (filterSize >= MAX_FILTER_SIZE*16/((flags&SWS_ACCURATE_RND) ? APCK_SIZE : 16) || !filter)
456 *outFilterSize= filterSize;
458 if (flags&SWS_PRINT_INFO)
459 av_log(NULL, AV_LOG_VERBOSE, "SwScaler: reducing / aligning filtersize %d -> %d\n", filter2Size, filterSize);
460 /* try to reduce the filter-size (step2 reduce it) */
461 for (i=0; i<dstW; i++) {
464 for (j=0; j<filterSize; j++) {
465 if (j>=filter2Size) filter[i*filterSize + j]= 0;
466 else filter[i*filterSize + j]= filter2[i*filter2Size + j];
467 if((flags & SWS_BITEXACT) && j>=minFilterSize)
468 filter[i*filterSize + j]= 0;
472 //FIXME try to align filterPos if possible
475 for (i=0; i<dstW; i++) {
477 if ((*filterPos)[i] < 0) {
478 // move filter coefficients left to compensate for filterPos
479 for (j=1; j<filterSize; j++) {
480 int left= FFMAX(j + (*filterPos)[i], 0);
481 filter[i*filterSize + left] += filter[i*filterSize + j];
482 filter[i*filterSize + j]=0;
487 if ((*filterPos)[i] + filterSize > srcW) {
488 int shift= (*filterPos)[i] + filterSize - srcW;
489 // move filter coefficients right to compensate for filterPos
490 for (j=filterSize-2; j>=0; j--) {
491 int right= FFMIN(j + shift, filterSize-1);
492 filter[i*filterSize +right] += filter[i*filterSize +j];
493 filter[i*filterSize +j]=0;
495 (*filterPos)[i]= srcW - filterSize;
499 // Note the +1 is for the MMX scaler which reads over the end
500 /* align at 16 for AltiVec (needed by hScale_altivec_real) */
501 FF_ALLOCZ_OR_GOTO(NULL, *outFilter, *outFilterSize*(dstW+3)*sizeof(int16_t), fail);
503 /* normalize & store in outFilter */
504 for (i=0; i<dstW; i++) {
509 for (j=0; j<filterSize; j++) {
510 sum+= filter[i*filterSize + j];
512 sum= (sum + one/2)/ one;
513 for (j=0; j<*outFilterSize; j++) {
514 int64_t v= filter[i*filterSize + j] + error;
515 int intV= ROUNDED_DIV(v, sum);
516 (*outFilter)[i*(*outFilterSize) + j]= intV;
521 (*filterPos)[dstW+0] =
522 (*filterPos)[dstW+1] =
523 (*filterPos)[dstW+2] = (*filterPos)[dstW-1]; // the MMX/SSE scaler will read over the end
524 for (i=0; i<*outFilterSize; i++) {
525 int k= (dstW - 1) * (*outFilterSize) + i;
526 (*outFilter)[k + 1 * (*outFilterSize)] =
527 (*outFilter)[k + 2 * (*outFilterSize)] =
528 (*outFilter)[k + 3 * (*outFilterSize)] = (*outFilter)[k];
539 static int initMMX2HScaler(int dstW, int xInc, uint8_t *filterCode, int16_t *filter, int32_t *filterPos, int numSplits)
542 x86_reg imm8OfPShufW1A;
543 x86_reg imm8OfPShufW2A;
544 x86_reg fragmentLengthA;
546 x86_reg imm8OfPShufW1B;
547 x86_reg imm8OfPShufW2B;
548 x86_reg fragmentLengthB;
553 // create an optimized horizontal scaling routine
554 /* This scaler is made of runtime-generated MMX2 code using specially
555 * tuned pshufw instructions. For every four output pixels, if four
556 * input pixels are enough for the fast bilinear scaling, then a chunk
557 * of fragmentB is used. If five input pixels are needed, then a chunk
558 * of fragmentA is used.
567 "movq (%%"REG_d", %%"REG_a"), %%mm3 \n\t"
568 "movd (%%"REG_c", %%"REG_S"), %%mm0 \n\t"
569 "movd 1(%%"REG_c", %%"REG_S"), %%mm1 \n\t"
570 "punpcklbw %%mm7, %%mm1 \n\t"
571 "punpcklbw %%mm7, %%mm0 \n\t"
572 "pshufw $0xFF, %%mm1, %%mm1 \n\t"
574 "pshufw $0xFF, %%mm0, %%mm0 \n\t"
576 "psubw %%mm1, %%mm0 \n\t"
577 "movl 8(%%"REG_b", %%"REG_a"), %%esi \n\t"
578 "pmullw %%mm3, %%mm0 \n\t"
579 "psllw $7, %%mm1 \n\t"
580 "paddw %%mm1, %%mm0 \n\t"
582 "movq %%mm0, (%%"REG_D", %%"REG_a") \n\t"
584 "add $8, %%"REG_a" \n\t"
588 "lea " LOCAL_MANGLE(0b) ", %0 \n\t"
589 "lea " LOCAL_MANGLE(1b) ", %1 \n\t"
590 "lea " LOCAL_MANGLE(2b) ", %2 \n\t"
595 "lea " LOCAL_MANGLE(9b) ", %3 \n\t"
599 :"=r" (fragmentA), "=r" (imm8OfPShufW1A), "=r" (imm8OfPShufW2A),
600 "=r" (fragmentLengthA)
607 "movq (%%"REG_d", %%"REG_a"), %%mm3 \n\t"
608 "movd (%%"REG_c", %%"REG_S"), %%mm0 \n\t"
609 "punpcklbw %%mm7, %%mm0 \n\t"
610 "pshufw $0xFF, %%mm0, %%mm1 \n\t"
612 "pshufw $0xFF, %%mm0, %%mm0 \n\t"
614 "psubw %%mm1, %%mm0 \n\t"
615 "movl 8(%%"REG_b", %%"REG_a"), %%esi \n\t"
616 "pmullw %%mm3, %%mm0 \n\t"
617 "psllw $7, %%mm1 \n\t"
618 "paddw %%mm1, %%mm0 \n\t"
620 "movq %%mm0, (%%"REG_D", %%"REG_a") \n\t"
622 "add $8, %%"REG_a" \n\t"
626 "lea " LOCAL_MANGLE(0b) ", %0 \n\t"
627 "lea " LOCAL_MANGLE(1b) ", %1 \n\t"
628 "lea " LOCAL_MANGLE(2b) ", %2 \n\t"
633 "lea " LOCAL_MANGLE(9b) ", %3 \n\t"
637 :"=r" (fragmentB), "=r" (imm8OfPShufW1B), "=r" (imm8OfPShufW2B),
638 "=r" (fragmentLengthB)
641 xpos= 0; //lumXInc/2 - 0x8000; // difference between pixel centers
644 for (i=0; i<dstW/numSplits; i++) {
649 int b=((xpos+xInc)>>16) - xx;
650 int c=((xpos+xInc*2)>>16) - xx;
651 int d=((xpos+xInc*3)>>16) - xx;
653 uint8_t *fragment = (d+1<4) ? fragmentB : fragmentA;
654 x86_reg imm8OfPShufW1 = (d+1<4) ? imm8OfPShufW1B : imm8OfPShufW1A;
655 x86_reg imm8OfPShufW2 = (d+1<4) ? imm8OfPShufW2B : imm8OfPShufW2A;
656 x86_reg fragmentLength = (d+1<4) ? fragmentLengthB : fragmentLengthA;
657 int maxShift= 3-(d+inc);
661 filter[i ] = (( xpos & 0xFFFF) ^ 0xFFFF)>>9;
662 filter[i+1] = (((xpos+xInc ) & 0xFFFF) ^ 0xFFFF)>>9;
663 filter[i+2] = (((xpos+xInc*2) & 0xFFFF) ^ 0xFFFF)>>9;
664 filter[i+3] = (((xpos+xInc*3) & 0xFFFF) ^ 0xFFFF)>>9;
667 memcpy(filterCode + fragmentPos, fragment, fragmentLength);
669 filterCode[fragmentPos + imm8OfPShufW1]=
670 (a+inc) | ((b+inc)<<2) | ((c+inc)<<4) | ((d+inc)<<6);
671 filterCode[fragmentPos + imm8OfPShufW2]=
672 a | (b<<2) | (c<<4) | (d<<6);
674 if (i+4-inc>=dstW) shift=maxShift; //avoid overread
675 else if ((filterPos[i/2]&3) <= maxShift) shift=filterPos[i/2]&3; //Align
677 if (shift && i>=shift) {
678 filterCode[fragmentPos + imm8OfPShufW1]+= 0x55*shift;
679 filterCode[fragmentPos + imm8OfPShufW2]+= 0x55*shift;
680 filterPos[i/2]-=shift;
684 fragmentPos+= fragmentLength;
687 filterCode[fragmentPos]= RET;
692 filterPos[((i/2)+1)&(~1)]= xpos>>16; // needed to jump to the next part
694 return fragmentPos + 1;
696 #endif /* HAVE_MMX2 */
698 static void getSubSampleFactors(int *h, int *v, enum PixelFormat format)
700 *h = av_pix_fmt_descriptors[format].log2_chroma_w;
701 *v = av_pix_fmt_descriptors[format].log2_chroma_h;
704 int sws_setColorspaceDetails(struct SwsContext *c, const int inv_table[4],
705 int srcRange, const int table[4], int dstRange,
706 int brightness, int contrast, int saturation)
708 memcpy(c->srcColorspaceTable, inv_table, sizeof(int)*4);
709 memcpy(c->dstColorspaceTable, table, sizeof(int)*4);
711 c->brightness= brightness;
712 c->contrast = contrast;
713 c->saturation= saturation;
714 c->srcRange = srcRange;
715 c->dstRange = dstRange;
716 if (isYUV(c->dstFormat) || isGray(c->dstFormat)) return -1;
718 c->dstFormatBpp = av_get_bits_per_pixel(&av_pix_fmt_descriptors[c->dstFormat]);
719 c->srcFormatBpp = av_get_bits_per_pixel(&av_pix_fmt_descriptors[c->srcFormat]);
721 ff_yuv2rgb_c_init_tables(c, inv_table, srcRange, brightness, contrast, saturation);
724 if (HAVE_ALTIVEC && av_get_cpu_flags() & AV_CPU_FLAG_ALTIVEC)
725 ff_yuv2rgb_init_tables_altivec(c, inv_table, brightness, contrast, saturation);
729 int sws_getColorspaceDetails(struct SwsContext *c, int **inv_table,
730 int *srcRange, int **table, int *dstRange,
731 int *brightness, int *contrast, int *saturation)
733 if (!c || isYUV(c->dstFormat) || isGray(c->dstFormat)) return -1;
735 *inv_table = c->srcColorspaceTable;
736 *table = c->dstColorspaceTable;
737 *srcRange = c->srcRange;
738 *dstRange = c->dstRange;
739 *brightness= c->brightness;
740 *contrast = c->contrast;
741 *saturation= c->saturation;
746 static int handle_jpeg(enum PixelFormat *format)
749 case PIX_FMT_YUVJ420P: *format = PIX_FMT_YUV420P; return 1;
750 case PIX_FMT_YUVJ422P: *format = PIX_FMT_YUV422P; return 1;
751 case PIX_FMT_YUVJ444P: *format = PIX_FMT_YUV444P; return 1;
752 case PIX_FMT_YUVJ440P: *format = PIX_FMT_YUV440P; return 1;
757 static int handle_0alpha(enum PixelFormat *format)
760 case PIX_FMT_0BGR : *format = PIX_FMT_ABGR ; return 1;
761 case PIX_FMT_BGR0 : *format = PIX_FMT_BGRA ; return 4;
762 case PIX_FMT_0RGB : *format = PIX_FMT_ARGB ; return 1;
763 case PIX_FMT_RGB0 : *format = PIX_FMT_RGBA ; return 4;
768 SwsContext *sws_alloc_context(void)
770 SwsContext *c= av_mallocz(sizeof(SwsContext));
772 c->av_class = &sws_context_class;
773 av_opt_set_defaults(c);
778 int sws_init_context(SwsContext *c, SwsFilter *srcFilter, SwsFilter *dstFilter)
781 int usesVFilter, usesHFilter;
783 SwsFilter dummyFilter= {NULL, NULL, NULL, NULL};
788 int dst_stride = FFALIGN(dstW * sizeof(int16_t)+66, 16);
789 int flags, cpu_flags;
790 enum PixelFormat srcFormat= c->srcFormat;
791 enum PixelFormat dstFormat= c->dstFormat;
793 cpu_flags = av_get_cpu_flags();
796 if (!rgb15to16) sws_rgb2rgb_init();
798 unscaled = (srcW == dstW && srcH == dstH);
800 handle_jpeg(&srcFormat);
801 handle_jpeg(&dstFormat);
802 handle_0alpha(&srcFormat);
803 handle_0alpha(&dstFormat);
805 if(srcFormat!=c->srcFormat || dstFormat!=c->dstFormat){
806 av_log(c, AV_LOG_WARNING, "deprecated pixel format used, make sure you did set range correctly\n");
807 c->srcFormat= srcFormat;
808 c->dstFormat= dstFormat;
811 if (!sws_isSupportedInput(srcFormat)) {
812 av_log(c, AV_LOG_ERROR, "%s is not supported as input pixel format\n", av_get_pix_fmt_name(srcFormat));
813 return AVERROR(EINVAL);
815 if (!sws_isSupportedOutput(dstFormat)) {
816 av_log(c, AV_LOG_ERROR, "%s is not supported as output pixel format\n", av_get_pix_fmt_name(dstFormat));
817 return AVERROR(EINVAL);
820 i= flags & ( SWS_POINT
831 if(!i || (i & (i-1))) {
832 av_log(c, AV_LOG_ERROR, "Exactly one scaler algorithm must be chosen, got %X\n", i);
833 return AVERROR(EINVAL);
836 if (srcW<4 || srcH<1 || dstW<8 || dstH<1) { //FIXME check if these are enough and try to lowwer them after fixing the relevant parts of the code
837 av_log(c, AV_LOG_ERROR, "%dx%d -> %dx%d is invalid scaling dimension\n",
838 srcW, srcH, dstW, dstH);
839 return AVERROR(EINVAL);
842 if (!dstFilter) dstFilter= &dummyFilter;
843 if (!srcFilter) srcFilter= &dummyFilter;
845 c->lumXInc= (((int64_t)srcW<<16) + (dstW>>1))/dstW;
846 c->lumYInc= (((int64_t)srcH<<16) + (dstH>>1))/dstH;
847 c->dstFormatBpp = av_get_bits_per_pixel(&av_pix_fmt_descriptors[dstFormat]);
848 c->srcFormatBpp = av_get_bits_per_pixel(&av_pix_fmt_descriptors[srcFormat]);
849 c->vRounder= 4* 0x0001000100010001ULL;
851 usesVFilter = (srcFilter->lumV && srcFilter->lumV->length>1) ||
852 (srcFilter->chrV && srcFilter->chrV->length>1) ||
853 (dstFilter->lumV && dstFilter->lumV->length>1) ||
854 (dstFilter->chrV && dstFilter->chrV->length>1);
855 usesHFilter = (srcFilter->lumH && srcFilter->lumH->length>1) ||
856 (srcFilter->chrH && srcFilter->chrH->length>1) ||
857 (dstFilter->lumH && dstFilter->lumH->length>1) ||
858 (dstFilter->chrH && dstFilter->chrH->length>1);
860 getSubSampleFactors(&c->chrSrcHSubSample, &c->chrSrcVSubSample, srcFormat);
861 getSubSampleFactors(&c->chrDstHSubSample, &c->chrDstVSubSample, dstFormat);
863 // reuse chroma for 2 pixels RGB/BGR unless user wants full chroma interpolation
864 if (isAnyRGB(dstFormat) && !(flags&SWS_FULL_CHR_H_INT)) {
866 av_log(c, AV_LOG_DEBUG, "Forcing full internal H chroma due to odd output size\n");
867 flags |= SWS_FULL_CHR_H_INT;
870 c->chrDstHSubSample = 1;
873 // drop some chroma lines if the user wants it
874 c->vChrDrop= (flags&SWS_SRC_V_CHR_DROP_MASK)>>SWS_SRC_V_CHR_DROP_SHIFT;
875 c->chrSrcVSubSample+= c->vChrDrop;
877 // drop every other pixel for chroma calculation unless user wants full chroma
878 if (isAnyRGB(srcFormat) && !(flags&SWS_FULL_CHR_H_INP)
879 && srcFormat!=PIX_FMT_RGB8 && srcFormat!=PIX_FMT_BGR8
880 && srcFormat!=PIX_FMT_RGB4 && srcFormat!=PIX_FMT_BGR4
881 && srcFormat!=PIX_FMT_RGB4_BYTE && srcFormat!=PIX_FMT_BGR4_BYTE
882 && ((dstW>>c->chrDstHSubSample) <= (srcW>>1) || (flags&SWS_FAST_BILINEAR)))
883 c->chrSrcHSubSample=1;
885 // Note the -((-x)>>y) is so that we always round toward +inf.
886 c->chrSrcW= -((-srcW) >> c->chrSrcHSubSample);
887 c->chrSrcH= -((-srcH) >> c->chrSrcVSubSample);
888 c->chrDstW= -((-dstW) >> c->chrDstHSubSample);
889 c->chrDstH= -((-dstH) >> c->chrDstVSubSample);
891 /* unscaled special cases */
892 if (unscaled && !usesHFilter && !usesVFilter && (c->srcRange == c->dstRange || isAnyRGB(dstFormat))) {
893 ff_get_unscaled_swscale(c);
896 if (flags&SWS_PRINT_INFO)
897 av_log(c, AV_LOG_INFO, "using unscaled %s -> %s special converter\n",
898 av_get_pix_fmt_name(srcFormat), av_get_pix_fmt_name(dstFormat));
903 c->srcBpc = 1 + av_pix_fmt_descriptors[srcFormat].comp[0].depth_minus1;
906 c->dstBpc = 1 + av_pix_fmt_descriptors[dstFormat].comp[0].depth_minus1;
909 if (isAnyRGB(srcFormat) || srcFormat == PIX_FMT_PAL8)
913 FF_ALLOC_OR_GOTO(c, c->formatConvBuffer, FFALIGN(srcW*2+78, 16) * 2, fail);
914 if (HAVE_MMX2 && cpu_flags & AV_CPU_FLAG_MMX2 && c->srcBpc == 8 && c->dstBpc <= 10) {
915 c->canMMX2BeUsed= (dstW >=srcW && (dstW&31)==0 && (srcW&15)==0) ? 1 : 0;
916 if (!c->canMMX2BeUsed && dstW >=srcW && (srcW&15)==0 && (flags&SWS_FAST_BILINEAR)) {
917 if (flags&SWS_PRINT_INFO)
918 av_log(c, AV_LOG_INFO, "output width is not a multiple of 32 -> no MMX2 scaler\n");
920 if (usesHFilter || isNBPS(c->srcFormat) || is16BPS(c->srcFormat) || isAnyRGB(c->srcFormat)) c->canMMX2BeUsed=0;
925 c->chrXInc= (((int64_t)c->chrSrcW<<16) + (c->chrDstW>>1))/c->chrDstW;
926 c->chrYInc= (((int64_t)c->chrSrcH<<16) + (c->chrDstH>>1))/c->chrDstH;
928 // match pixel 0 of the src to pixel 0 of dst and match pixel n-2 of src to pixel n-2 of dst
929 // but only for the FAST_BILINEAR mode otherwise do correct scaling
930 // n-2 is the last chrominance sample available
931 // this is not perfect, but no one should notice the difference, the more correct variant
932 // would be like the vertical one, but that would require some special code for the
933 // first and last pixel
934 if (flags&SWS_FAST_BILINEAR) {
935 if (c->canMMX2BeUsed) {
939 //we don't use the x86 asm scaler if MMX is available
940 else if (HAVE_MMX && cpu_flags & AV_CPU_FLAG_MMX && c->dstBpc <= 10) {
941 c->lumXInc = ((int64_t)(srcW-2)<<16)/(dstW-2) - 20;
942 c->chrXInc = ((int64_t)(c->chrSrcW-2)<<16)/(c->chrDstW-2) - 20;
946 /* precalculate horizontal scaler filter coefficients */
949 // can't downscale !!!
950 if (c->canMMX2BeUsed && (flags & SWS_FAST_BILINEAR)) {
951 c->lumMmx2FilterCodeSize = initMMX2HScaler( dstW, c->lumXInc, NULL, NULL, NULL, 8);
952 c->chrMmx2FilterCodeSize = initMMX2HScaler(c->chrDstW, c->chrXInc, NULL, NULL, NULL, 4);
955 c->lumMmx2FilterCode = mmap(NULL, c->lumMmx2FilterCodeSize, PROT_READ | PROT_WRITE, MAP_PRIVATE | MAP_ANONYMOUS, -1, 0);
956 c->chrMmx2FilterCode = mmap(NULL, c->chrMmx2FilterCodeSize, PROT_READ | PROT_WRITE, MAP_PRIVATE | MAP_ANONYMOUS, -1, 0);
957 #elif HAVE_VIRTUALALLOC
958 c->lumMmx2FilterCode = VirtualAlloc(NULL, c->lumMmx2FilterCodeSize, MEM_COMMIT, PAGE_EXECUTE_READWRITE);
959 c->chrMmx2FilterCode = VirtualAlloc(NULL, c->chrMmx2FilterCodeSize, MEM_COMMIT, PAGE_EXECUTE_READWRITE);
961 c->lumMmx2FilterCode = av_malloc(c->lumMmx2FilterCodeSize);
962 c->chrMmx2FilterCode = av_malloc(c->chrMmx2FilterCodeSize);
966 if (c->lumMmx2FilterCode == MAP_FAILED || c->chrMmx2FilterCode == MAP_FAILED)
968 if (!c->lumMmx2FilterCode || !c->chrMmx2FilterCode)
971 av_log(c, AV_LOG_ERROR, "Failed to allocate MMX2FilterCode\n");
972 return AVERROR(ENOMEM);
974 FF_ALLOCZ_OR_GOTO(c, c->hLumFilter , (dstW /8+8)*sizeof(int16_t), fail);
975 FF_ALLOCZ_OR_GOTO(c, c->hChrFilter , (c->chrDstW /4+8)*sizeof(int16_t), fail);
976 FF_ALLOCZ_OR_GOTO(c, c->hLumFilterPos, (dstW /2/8+8)*sizeof(int32_t), fail);
977 FF_ALLOCZ_OR_GOTO(c, c->hChrFilterPos, (c->chrDstW/2/4+8)*sizeof(int32_t), fail);
979 initMMX2HScaler( dstW, c->lumXInc, c->lumMmx2FilterCode, c->hLumFilter, (uint32_t*)c->hLumFilterPos, 8);
980 initMMX2HScaler(c->chrDstW, c->chrXInc, c->chrMmx2FilterCode, c->hChrFilter, (uint32_t*)c->hChrFilterPos, 4);
983 mprotect(c->lumMmx2FilterCode, c->lumMmx2FilterCodeSize, PROT_EXEC | PROT_READ);
984 mprotect(c->chrMmx2FilterCode, c->chrMmx2FilterCodeSize, PROT_EXEC | PROT_READ);
987 #endif /* HAVE_MMX2 */
989 const int filterAlign=
990 (HAVE_MMX && cpu_flags & AV_CPU_FLAG_MMX) ? 4 :
991 (HAVE_ALTIVEC && cpu_flags & AV_CPU_FLAG_ALTIVEC) ? 8 :
994 if (initFilter(&c->hLumFilter, &c->hLumFilterPos, &c->hLumFilterSize, c->lumXInc,
995 srcW , dstW, filterAlign, 1<<14,
996 (flags&SWS_BICUBLIN) ? (flags|SWS_BICUBIC) : flags, cpu_flags,
997 srcFilter->lumH, dstFilter->lumH, c->param) < 0)
999 if (initFilter(&c->hChrFilter, &c->hChrFilterPos, &c->hChrFilterSize, c->chrXInc,
1000 c->chrSrcW, c->chrDstW, filterAlign, 1<<14,
1001 (flags&SWS_BICUBLIN) ? (flags|SWS_BILINEAR) : flags, cpu_flags,
1002 srcFilter->chrH, dstFilter->chrH, c->param) < 0)
1005 } // initialize horizontal stuff
1007 /* precalculate vertical scaler filter coefficients */
1009 const int filterAlign=
1010 (HAVE_MMX && cpu_flags & AV_CPU_FLAG_MMX) ? 2 :
1011 (HAVE_ALTIVEC && cpu_flags & AV_CPU_FLAG_ALTIVEC) ? 8 :
1014 if (initFilter(&c->vLumFilter, &c->vLumFilterPos, &c->vLumFilterSize, c->lumYInc,
1015 srcH , dstH, filterAlign, (1<<12),
1016 (flags&SWS_BICUBLIN) ? (flags|SWS_BICUBIC) : flags, cpu_flags,
1017 srcFilter->lumV, dstFilter->lumV, c->param) < 0)
1019 if (initFilter(&c->vChrFilter, &c->vChrFilterPos, &c->vChrFilterSize, c->chrYInc,
1020 c->chrSrcH, c->chrDstH, filterAlign, (1<<12),
1021 (flags&SWS_BICUBLIN) ? (flags|SWS_BILINEAR) : flags, cpu_flags,
1022 srcFilter->chrV, dstFilter->chrV, c->param) < 0)
1026 FF_ALLOC_OR_GOTO(c, c->vYCoeffsBank, sizeof (vector signed short)*c->vLumFilterSize*c->dstH, fail);
1027 FF_ALLOC_OR_GOTO(c, c->vCCoeffsBank, sizeof (vector signed short)*c->vChrFilterSize*c->chrDstH, fail);
1029 for (i=0;i<c->vLumFilterSize*c->dstH;i++) {
1031 short *p = (short *)&c->vYCoeffsBank[i];
1033 p[j] = c->vLumFilter[i];
1036 for (i=0;i<c->vChrFilterSize*c->chrDstH;i++) {
1038 short *p = (short *)&c->vCCoeffsBank[i];
1040 p[j] = c->vChrFilter[i];
1045 // calculate buffer sizes so that they won't run out while handling these damn slices
1046 c->vLumBufSize= c->vLumFilterSize;
1047 c->vChrBufSize= c->vChrFilterSize;
1048 for (i=0; i<dstH; i++) {
1049 int chrI = (int64_t) i * c->chrDstH / dstH;
1050 int nextSlice= FFMAX(c->vLumFilterPos[i ] + c->vLumFilterSize - 1,
1051 ((c->vChrFilterPos[chrI] + c->vChrFilterSize - 1)<<c->chrSrcVSubSample));
1053 nextSlice>>= c->chrSrcVSubSample;
1054 nextSlice<<= c->chrSrcVSubSample;
1055 if (c->vLumFilterPos[i ] + c->vLumBufSize < nextSlice)
1056 c->vLumBufSize= nextSlice - c->vLumFilterPos[i];
1057 if (c->vChrFilterPos[chrI] + c->vChrBufSize < (nextSlice>>c->chrSrcVSubSample))
1058 c->vChrBufSize= (nextSlice>>c->chrSrcVSubSample) - c->vChrFilterPos[chrI];
1061 // allocate pixbufs (we use dynamic allocation because otherwise we would need to
1062 // allocate several megabytes to handle all possible cases)
1063 FF_ALLOC_OR_GOTO(c, c->lumPixBuf, c->vLumBufSize*2*sizeof(int16_t*), fail);
1064 FF_ALLOC_OR_GOTO(c, c->chrUPixBuf, c->vChrBufSize*2*sizeof(int16_t*), fail);
1065 FF_ALLOC_OR_GOTO(c, c->chrVPixBuf, c->vChrBufSize*2*sizeof(int16_t*), fail);
1066 if (CONFIG_SWSCALE_ALPHA && isALPHA(c->srcFormat) && isALPHA(c->dstFormat))
1067 FF_ALLOCZ_OR_GOTO(c, c->alpPixBuf, c->vLumBufSize*2*sizeof(int16_t*), fail);
1068 //Note we need at least one pixel more at the end because of the MMX code (just in case someone wanna replace the 4000/8000)
1069 /* align at 16 bytes for AltiVec */
1070 for (i=0; i<c->vLumBufSize; i++) {
1071 FF_ALLOCZ_OR_GOTO(c, c->lumPixBuf[i+c->vLumBufSize], dst_stride+16, fail);
1072 c->lumPixBuf[i] = c->lumPixBuf[i+c->vLumBufSize];
1074 // 64 / c->scalingBpp is the same as 16 / sizeof(scaling_intermediate)
1075 c->uv_off = (dst_stride>>1) + 64 / (c->dstBpc &~ 7);
1076 c->uv_offx2 = dst_stride + 16;
1077 for (i=0; i<c->vChrBufSize; i++) {
1078 FF_ALLOC_OR_GOTO(c, c->chrUPixBuf[i+c->vChrBufSize], dst_stride*2+32, fail);
1079 c->chrUPixBuf[i] = c->chrUPixBuf[i+c->vChrBufSize];
1080 c->chrVPixBuf[i] = c->chrVPixBuf[i+c->vChrBufSize] = c->chrUPixBuf[i] + (dst_stride >> 1) + 8;
1082 if (CONFIG_SWSCALE_ALPHA && c->alpPixBuf)
1083 for (i=0; i<c->vLumBufSize; i++) {
1084 FF_ALLOCZ_OR_GOTO(c, c->alpPixBuf[i+c->vLumBufSize], dst_stride+16, fail);
1085 c->alpPixBuf[i] = c->alpPixBuf[i+c->vLumBufSize];
1088 //try to avoid drawing green stuff between the right end and the stride end
1089 for (i=0; i<c->vChrBufSize; i++)
1090 if(av_pix_fmt_descriptors[c->dstFormat].comp[0].depth_minus1 == 15){
1091 av_assert0(c->dstBpc > 10);
1092 for(j=0; j<dst_stride/2+1; j++)
1093 ((int32_t*)(c->chrUPixBuf[i]))[j] = 1<<18;
1095 for(j=0; j<dst_stride+1; j++)
1096 ((int16_t*)(c->chrUPixBuf[i]))[j] = 1<<14;
1098 assert(c->chrDstH <= dstH);
1100 if (flags&SWS_PRINT_INFO) {
1101 if (flags&SWS_FAST_BILINEAR) av_log(c, AV_LOG_INFO, "FAST_BILINEAR scaler, ");
1102 else if (flags&SWS_BILINEAR) av_log(c, AV_LOG_INFO, "BILINEAR scaler, ");
1103 else if (flags&SWS_BICUBIC) av_log(c, AV_LOG_INFO, "BICUBIC scaler, ");
1104 else if (flags&SWS_X) av_log(c, AV_LOG_INFO, "Experimental scaler, ");
1105 else if (flags&SWS_POINT) av_log(c, AV_LOG_INFO, "Nearest Neighbor / POINT scaler, ");
1106 else if (flags&SWS_AREA) av_log(c, AV_LOG_INFO, "Area Averaging scaler, ");
1107 else if (flags&SWS_BICUBLIN) av_log(c, AV_LOG_INFO, "luma BICUBIC / chroma BILINEAR scaler, ");
1108 else if (flags&SWS_GAUSS) av_log(c, AV_LOG_INFO, "Gaussian scaler, ");
1109 else if (flags&SWS_SINC) av_log(c, AV_LOG_INFO, "Sinc scaler, ");
1110 else if (flags&SWS_LANCZOS) av_log(c, AV_LOG_INFO, "Lanczos scaler, ");
1111 else if (flags&SWS_SPLINE) av_log(c, AV_LOG_INFO, "Bicubic spline scaler, ");
1112 else av_log(c, AV_LOG_INFO, "ehh flags invalid?! ");
1114 av_log(c, AV_LOG_INFO, "from %s to %s%s ",
1115 av_get_pix_fmt_name(srcFormat),
1117 dstFormat == PIX_FMT_BGR555 || dstFormat == PIX_FMT_BGR565 ||
1118 dstFormat == PIX_FMT_RGB444BE || dstFormat == PIX_FMT_RGB444LE ||
1119 dstFormat == PIX_FMT_BGR444BE || dstFormat == PIX_FMT_BGR444LE ? "dithered " : "",
1123 av_get_pix_fmt_name(dstFormat));
1125 if (HAVE_MMX2 && cpu_flags & AV_CPU_FLAG_MMX2) av_log(c, AV_LOG_INFO, "using MMX2\n");
1126 else if (HAVE_AMD3DNOW && cpu_flags & AV_CPU_FLAG_3DNOW) av_log(c, AV_LOG_INFO, "using 3DNOW\n");
1127 else if (HAVE_MMX && cpu_flags & AV_CPU_FLAG_MMX) av_log(c, AV_LOG_INFO, "using MMX\n");
1128 else if (HAVE_ALTIVEC && cpu_flags & AV_CPU_FLAG_ALTIVEC) av_log(c, AV_LOG_INFO, "using AltiVec\n");
1129 else av_log(c, AV_LOG_INFO, "using C\n");
1131 av_log(c, AV_LOG_VERBOSE, "%dx%d -> %dx%d\n", srcW, srcH, dstW, dstH);
1132 av_log(c, AV_LOG_DEBUG, "lum srcW=%d srcH=%d dstW=%d dstH=%d xInc=%d yInc=%d\n",
1133 c->srcW, c->srcH, c->dstW, c->dstH, c->lumXInc, c->lumYInc);
1134 av_log(c, AV_LOG_DEBUG, "chr srcW=%d srcH=%d dstW=%d dstH=%d xInc=%d yInc=%d\n",
1135 c->chrSrcW, c->chrSrcH, c->chrDstW, c->chrDstH, c->chrXInc, c->chrYInc);
1138 c->swScale= ff_getSwsFunc(c);
1140 fail: //FIXME replace things by appropriate error codes
1144 #if FF_API_SWS_GETCONTEXT
1145 SwsContext *sws_getContext(int srcW, int srcH, enum PixelFormat srcFormat,
1146 int dstW, int dstH, enum PixelFormat dstFormat, int flags,
1147 SwsFilter *srcFilter, SwsFilter *dstFilter, const double *param)
1151 if(!(c=sws_alloc_context()))
1159 c->srcRange = handle_jpeg(&srcFormat);
1160 c->dstRange = handle_jpeg(&dstFormat);
1161 c->src0Alpha = handle_0alpha(&srcFormat);
1162 c->dst0Alpha = handle_0alpha(&dstFormat);
1163 c->srcFormat= srcFormat;
1164 c->dstFormat= dstFormat;
1167 c->param[0] = param[0];
1168 c->param[1] = param[1];
1170 sws_setColorspaceDetails(c, ff_yuv2rgb_coeffs[SWS_CS_DEFAULT], c->srcRange, ff_yuv2rgb_coeffs[SWS_CS_DEFAULT] /* FIXME*/, c->dstRange, 0, 1<<16, 1<<16);
1172 if(sws_init_context(c, srcFilter, dstFilter) < 0){
1181 SwsFilter *sws_getDefaultFilter(float lumaGBlur, float chromaGBlur,
1182 float lumaSharpen, float chromaSharpen,
1183 float chromaHShift, float chromaVShift,
1186 SwsFilter *filter= av_malloc(sizeof(SwsFilter));
1190 if (lumaGBlur!=0.0) {
1191 filter->lumH= sws_getGaussianVec(lumaGBlur, 3.0);
1192 filter->lumV= sws_getGaussianVec(lumaGBlur, 3.0);
1194 filter->lumH= sws_getIdentityVec();
1195 filter->lumV= sws_getIdentityVec();
1198 if (chromaGBlur!=0.0) {
1199 filter->chrH= sws_getGaussianVec(chromaGBlur, 3.0);
1200 filter->chrV= sws_getGaussianVec(chromaGBlur, 3.0);
1202 filter->chrH= sws_getIdentityVec();
1203 filter->chrV= sws_getIdentityVec();
1206 if (chromaSharpen!=0.0) {
1207 SwsVector *id= sws_getIdentityVec();
1208 sws_scaleVec(filter->chrH, -chromaSharpen);
1209 sws_scaleVec(filter->chrV, -chromaSharpen);
1210 sws_addVec(filter->chrH, id);
1211 sws_addVec(filter->chrV, id);
1215 if (lumaSharpen!=0.0) {
1216 SwsVector *id= sws_getIdentityVec();
1217 sws_scaleVec(filter->lumH, -lumaSharpen);
1218 sws_scaleVec(filter->lumV, -lumaSharpen);
1219 sws_addVec(filter->lumH, id);
1220 sws_addVec(filter->lumV, id);
1224 if (chromaHShift != 0.0)
1225 sws_shiftVec(filter->chrH, (int)(chromaHShift+0.5));
1227 if (chromaVShift != 0.0)
1228 sws_shiftVec(filter->chrV, (int)(chromaVShift+0.5));
1230 sws_normalizeVec(filter->chrH, 1.0);
1231 sws_normalizeVec(filter->chrV, 1.0);
1232 sws_normalizeVec(filter->lumH, 1.0);
1233 sws_normalizeVec(filter->lumV, 1.0);
1235 if (verbose) sws_printVec2(filter->chrH, NULL, AV_LOG_DEBUG);
1236 if (verbose) sws_printVec2(filter->lumH, NULL, AV_LOG_DEBUG);
1241 SwsVector *sws_allocVec(int length)
1243 SwsVector *vec = av_malloc(sizeof(SwsVector));
1246 vec->length = length;
1247 vec->coeff = av_malloc(sizeof(double) * length);
1253 SwsVector *sws_getGaussianVec(double variance, double quality)
1255 const int length= (int)(variance*quality + 0.5) | 1;
1257 double middle= (length-1)*0.5;
1258 SwsVector *vec= sws_allocVec(length);
1263 for (i=0; i<length; i++) {
1264 double dist= i-middle;
1265 vec->coeff[i]= exp(-dist*dist/(2*variance*variance)) / sqrt(2*variance*M_PI);
1268 sws_normalizeVec(vec, 1.0);
1273 SwsVector *sws_getConstVec(double c, int length)
1276 SwsVector *vec= sws_allocVec(length);
1281 for (i=0; i<length; i++)
1287 SwsVector *sws_getIdentityVec(void)
1289 return sws_getConstVec(1.0, 1);
1292 static double sws_dcVec(SwsVector *a)
1297 for (i=0; i<a->length; i++)
1303 void sws_scaleVec(SwsVector *a, double scalar)
1307 for (i=0; i<a->length; i++)
1308 a->coeff[i]*= scalar;
1311 void sws_normalizeVec(SwsVector *a, double height)
1313 sws_scaleVec(a, height/sws_dcVec(a));
1316 static SwsVector *sws_getConvVec(SwsVector *a, SwsVector *b)
1318 int length= a->length + b->length - 1;
1320 SwsVector *vec= sws_getConstVec(0.0, length);
1325 for (i=0; i<a->length; i++) {
1326 for (j=0; j<b->length; j++) {
1327 vec->coeff[i+j]+= a->coeff[i]*b->coeff[j];
1334 static SwsVector *sws_sumVec(SwsVector *a, SwsVector *b)
1336 int length= FFMAX(a->length, b->length);
1338 SwsVector *vec= sws_getConstVec(0.0, length);
1343 for (i=0; i<a->length; i++) vec->coeff[i + (length-1)/2 - (a->length-1)/2]+= a->coeff[i];
1344 for (i=0; i<b->length; i++) vec->coeff[i + (length-1)/2 - (b->length-1)/2]+= b->coeff[i];
1349 static SwsVector *sws_diffVec(SwsVector *a, SwsVector *b)
1351 int length= FFMAX(a->length, b->length);
1353 SwsVector *vec= sws_getConstVec(0.0, length);
1358 for (i=0; i<a->length; i++) vec->coeff[i + (length-1)/2 - (a->length-1)/2]+= a->coeff[i];
1359 for (i=0; i<b->length; i++) vec->coeff[i + (length-1)/2 - (b->length-1)/2]-= b->coeff[i];
1364 /* shift left / or right if "shift" is negative */
1365 static SwsVector *sws_getShiftedVec(SwsVector *a, int shift)
1367 int length= a->length + FFABS(shift)*2;
1369 SwsVector *vec= sws_getConstVec(0.0, length);
1374 for (i=0; i<a->length; i++) {
1375 vec->coeff[i + (length-1)/2 - (a->length-1)/2 - shift]= a->coeff[i];
1381 void sws_shiftVec(SwsVector *a, int shift)
1383 SwsVector *shifted= sws_getShiftedVec(a, shift);
1385 a->coeff= shifted->coeff;
1386 a->length= shifted->length;
1390 void sws_addVec(SwsVector *a, SwsVector *b)
1392 SwsVector *sum= sws_sumVec(a, b);
1394 a->coeff= sum->coeff;
1395 a->length= sum->length;
1399 void sws_subVec(SwsVector *a, SwsVector *b)
1401 SwsVector *diff= sws_diffVec(a, b);
1403 a->coeff= diff->coeff;
1404 a->length= diff->length;
1408 void sws_convVec(SwsVector *a, SwsVector *b)
1410 SwsVector *conv= sws_getConvVec(a, b);
1412 a->coeff= conv->coeff;
1413 a->length= conv->length;
1417 SwsVector *sws_cloneVec(SwsVector *a)
1420 SwsVector *vec= sws_allocVec(a->length);
1425 for (i=0; i<a->length; i++) vec->coeff[i]= a->coeff[i];
1430 void sws_printVec2(SwsVector *a, AVClass *log_ctx, int log_level)
1437 for (i=0; i<a->length; i++)
1438 if (a->coeff[i]>max) max= a->coeff[i];
1440 for (i=0; i<a->length; i++)
1441 if (a->coeff[i]<min) min= a->coeff[i];
1445 for (i=0; i<a->length; i++) {
1446 int x= (int)((a->coeff[i]-min)*60.0/range +0.5);
1447 av_log(log_ctx, log_level, "%1.3f ", a->coeff[i]);
1448 for (;x>0; x--) av_log(log_ctx, log_level, " ");
1449 av_log(log_ctx, log_level, "|\n");
1453 void sws_freeVec(SwsVector *a)
1456 av_freep(&a->coeff);
1461 void sws_freeFilter(SwsFilter *filter)
1463 if (!filter) return;
1465 if (filter->lumH) sws_freeVec(filter->lumH);
1466 if (filter->lumV) sws_freeVec(filter->lumV);
1467 if (filter->chrH) sws_freeVec(filter->chrH);
1468 if (filter->chrV) sws_freeVec(filter->chrV);
1472 void sws_freeContext(SwsContext *c)
1478 for (i=0; i<c->vLumBufSize; i++)
1479 av_freep(&c->lumPixBuf[i]);
1480 av_freep(&c->lumPixBuf);
1483 if (c->chrUPixBuf) {
1484 for (i=0; i<c->vChrBufSize; i++)
1485 av_freep(&c->chrUPixBuf[i]);
1486 av_freep(&c->chrUPixBuf);
1487 av_freep(&c->chrVPixBuf);
1490 if (CONFIG_SWSCALE_ALPHA && c->alpPixBuf) {
1491 for (i=0; i<c->vLumBufSize; i++)
1492 av_freep(&c->alpPixBuf[i]);
1493 av_freep(&c->alpPixBuf);
1496 av_freep(&c->vLumFilter);
1497 av_freep(&c->vChrFilter);
1498 av_freep(&c->hLumFilter);
1499 av_freep(&c->hChrFilter);
1501 av_freep(&c->vYCoeffsBank);
1502 av_freep(&c->vCCoeffsBank);
1505 av_freep(&c->vLumFilterPos);
1506 av_freep(&c->vChrFilterPos);
1507 av_freep(&c->hLumFilterPos);
1508 av_freep(&c->hChrFilterPos);
1511 #ifdef MAP_ANONYMOUS
1512 if (c->lumMmx2FilterCode) munmap(c->lumMmx2FilterCode, c->lumMmx2FilterCodeSize);
1513 if (c->chrMmx2FilterCode) munmap(c->chrMmx2FilterCode, c->chrMmx2FilterCodeSize);
1514 #elif HAVE_VIRTUALALLOC
1515 if (c->lumMmx2FilterCode) VirtualFree(c->lumMmx2FilterCode, 0, MEM_RELEASE);
1516 if (c->chrMmx2FilterCode) VirtualFree(c->chrMmx2FilterCode, 0, MEM_RELEASE);
1518 av_free(c->lumMmx2FilterCode);
1519 av_free(c->chrMmx2FilterCode);
1521 c->lumMmx2FilterCode=NULL;
1522 c->chrMmx2FilterCode=NULL;
1523 #endif /* HAVE_MMX */
1525 av_freep(&c->yuvTable);
1526 av_freep(&c->formatConvBuffer);
1531 struct SwsContext *sws_getCachedContext(struct SwsContext *context,
1532 int srcW, int srcH, enum PixelFormat srcFormat,
1533 int dstW, int dstH, enum PixelFormat dstFormat, int flags,
1534 SwsFilter *srcFilter, SwsFilter *dstFilter, const double *param)
1536 static const double default_param[2] = {SWS_PARAM_DEFAULT, SWS_PARAM_DEFAULT};
1539 param = default_param;
1542 (context->srcW != srcW ||
1543 context->srcH != srcH ||
1544 context->srcFormat != srcFormat ||
1545 context->dstW != dstW ||
1546 context->dstH != dstH ||
1547 context->dstFormat != dstFormat ||
1548 context->flags != flags ||
1549 context->param[0] != param[0] ||
1550 context->param[1] != param[1])) {
1551 sws_freeContext(context);
1556 if (!(context = sws_alloc_context()))
1558 context->srcW = srcW;
1559 context->srcH = srcH;
1560 context->srcRange = handle_jpeg(&srcFormat);
1561 context->src0Alpha = handle_0alpha(&srcFormat);
1562 context->srcFormat = srcFormat;
1563 context->dstW = dstW;
1564 context->dstH = dstH;
1565 context->dstRange = handle_jpeg(&dstFormat);
1566 context->dst0Alpha = handle_0alpha(&dstFormat);
1567 context->dstFormat = dstFormat;
1568 context->flags = flags;
1569 context->param[0] = param[0];
1570 context->param[1] = param[1];
1571 sws_setColorspaceDetails(context, ff_yuv2rgb_coeffs[SWS_CS_DEFAULT], context->srcRange, ff_yuv2rgb_coeffs[SWS_CS_DEFAULT] /* FIXME*/, context->dstRange, 0, 1<<16, 1<<16);
1572 if (sws_init_context(context, srcFilter, dstFilter) < 0) {
1573 sws_freeContext(context);