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 modify
7 * it under the terms of the GNU General Public License as published by
8 * the Free Software Foundation; either version 2 of the License, or
9 * (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
14 * GNU General Public License for more details.
16 * You should have received a copy of the GNU General Public License
17 * along with FFmpeg; if not, write to the Free Software
18 * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
20 * the C code (not assembly, mmx, ...) of this file can be used
21 * under the LGPL license too
25 supported Input formats: YV12, I420/IYUV, YUY2, UYVY, BGR32, BGR24, BGR16, BGR15, RGB32, RGB24, Y8/Y800, YVU9/IF09, PAL8
26 supported output formats: YV12, I420/IYUV, YUY2, UYVY, {BGR,RGB}{1,4,8,15,16,24,32}, Y8/Y800, YVU9/IF09
27 {BGR,RGB}{1,4,8,15,16} support dithering
29 unscaled special converters (YV12=I420=IYUV, Y800=Y8)
30 YV12 -> {BGR,RGB}{1,4,8,15,16,24,32}
35 BGR24 -> BGR32 & RGB24 -> RGB32
36 BGR32 -> BGR24 & RGB32 -> RGB24
41 tested special converters (most are tested actually, but I did not write it down ...)
48 untested special converters
49 YV12/I420 -> BGR15/BGR24/BGR32 (it is the yuv2rgb stuff, so it should be ok)
50 YV12/I420 -> YV12/I420
51 YUY2/BGR15/BGR24/BGR32/RGB24/RGB32 -> same format
52 BGR24 -> BGR32 & RGB24 -> RGB32
53 BGR32 -> BGR24 & RGB32 -> RGB24
64 #ifdef HAVE_SYS_MMAN_H
66 #if defined(MAP_ANON) && !defined(MAP_ANONYMOUS)
67 #define MAP_ANONYMOUS MAP_ANON
71 #include "swscale_internal.h"
73 #include "libavutil/x86_cpu.h"
74 #include "libavutil/bswap.h"
75 #include "libavcodec/opt.h"
84 //#define WORDS_BIGENDIAN
87 #define FAST_BGR2YV12 // use 7 bit coeffs instead of 15bit
89 #define RET 0xC3 //near return opcode for X86
94 #define PI 3.14159265358979323846
97 #define isSupportedIn(x) ( \
98 (x)==PIX_FMT_YUV420P \
99 || (x)==PIX_FMT_YUVA420P \
100 || (x)==PIX_FMT_YUYV422 \
101 || (x)==PIX_FMT_UYVY422 \
102 || (x)==PIX_FMT_RGB32 \
103 || (x)==PIX_FMT_BGR24 \
104 || (x)==PIX_FMT_BGR565 \
105 || (x)==PIX_FMT_BGR555 \
106 || (x)==PIX_FMT_BGR32 \
107 || (x)==PIX_FMT_RGB24 \
108 || (x)==PIX_FMT_RGB565 \
109 || (x)==PIX_FMT_RGB555 \
110 || (x)==PIX_FMT_GRAY8 \
111 || (x)==PIX_FMT_YUV410P \
112 || (x)==PIX_FMT_GRAY16BE \
113 || (x)==PIX_FMT_GRAY16LE \
114 || (x)==PIX_FMT_YUV444P \
115 || (x)==PIX_FMT_YUV422P \
116 || (x)==PIX_FMT_YUV411P \
117 || (x)==PIX_FMT_PAL8 \
118 || (x)==PIX_FMT_BGR8 \
119 || (x)==PIX_FMT_RGB8 \
120 || (x)==PIX_FMT_BGR4_BYTE \
121 || (x)==PIX_FMT_RGB4_BYTE \
122 || (x)==PIX_FMT_YUV440P \
124 #define isSupportedOut(x) ( \
125 (x)==PIX_FMT_YUV420P \
126 || (x)==PIX_FMT_YUYV422 \
127 || (x)==PIX_FMT_UYVY422 \
128 || (x)==PIX_FMT_YUV444P \
129 || (x)==PIX_FMT_YUV422P \
130 || (x)==PIX_FMT_YUV411P \
133 || (x)==PIX_FMT_NV12 \
134 || (x)==PIX_FMT_NV21 \
135 || (x)==PIX_FMT_GRAY16BE \
136 || (x)==PIX_FMT_GRAY16LE \
137 || (x)==PIX_FMT_GRAY8 \
138 || (x)==PIX_FMT_YUV410P \
140 #define isPacked(x) ( \
142 || (x)==PIX_FMT_YUYV422 \
143 || (x)==PIX_FMT_UYVY422 \
148 #define RGB2YUV_SHIFT 16
149 #define BY ((int)( 0.098*(1<<RGB2YUV_SHIFT)+0.5))
150 #define BV ((int)(-0.071*(1<<RGB2YUV_SHIFT)+0.5))
151 #define BU ((int)( 0.439*(1<<RGB2YUV_SHIFT)+0.5))
152 #define GY ((int)( 0.504*(1<<RGB2YUV_SHIFT)+0.5))
153 #define GV ((int)(-0.368*(1<<RGB2YUV_SHIFT)+0.5))
154 #define GU ((int)(-0.291*(1<<RGB2YUV_SHIFT)+0.5))
155 #define RY ((int)( 0.257*(1<<RGB2YUV_SHIFT)+0.5))
156 #define RV ((int)( 0.439*(1<<RGB2YUV_SHIFT)+0.5))
157 #define RU ((int)(-0.148*(1<<RGB2YUV_SHIFT)+0.5))
159 extern const int32_t Inverse_Table_6_9[8][4];
163 Special versions: fast Y 1:1 scaling (no interpolation in y direction)
166 more intelligent misalignment avoidance for the horizontal scaler
167 write special vertical cubic upscale version
168 Optimize C code (yv12 / minmax)
169 add support for packed pixel yuv input & output
170 add support for Y8 output
171 optimize bgr24 & bgr32
172 add BGR4 output support
173 write special BGR->BGR scaler
176 #if defined(ARCH_X86) && defined (CONFIG_GPL)
177 DECLARE_ASM_CONST(8, uint64_t, bF8)= 0xF8F8F8F8F8F8F8F8LL;
178 DECLARE_ASM_CONST(8, uint64_t, bFC)= 0xFCFCFCFCFCFCFCFCLL;
179 DECLARE_ASM_CONST(8, uint64_t, w10)= 0x0010001000100010LL;
180 DECLARE_ASM_CONST(8, uint64_t, w02)= 0x0002000200020002LL;
181 DECLARE_ASM_CONST(8, uint64_t, bm00001111)=0x00000000FFFFFFFFLL;
182 DECLARE_ASM_CONST(8, uint64_t, bm00000111)=0x0000000000FFFFFFLL;
183 DECLARE_ASM_CONST(8, uint64_t, bm11111000)=0xFFFFFFFFFF000000LL;
184 DECLARE_ASM_CONST(8, uint64_t, bm01010101)=0x00FF00FF00FF00FFLL;
186 static volatile uint64_t attribute_used __attribute__((aligned(8))) b5Dither;
187 static volatile uint64_t attribute_used __attribute__((aligned(8))) g5Dither;
188 static volatile uint64_t attribute_used __attribute__((aligned(8))) g6Dither;
189 static volatile uint64_t attribute_used __attribute__((aligned(8))) r5Dither;
191 const DECLARE_ALIGNED(8, uint64_t, ff_dither4[2]) = {
192 0x0103010301030103LL,
193 0x0200020002000200LL,};
195 const DECLARE_ALIGNED(8, uint64_t, ff_dither8[2]) = {
196 0x0602060206020602LL,
197 0x0004000400040004LL,};
199 DECLARE_ASM_CONST(8, uint64_t, b16Mask)= 0x001F001F001F001FLL;
200 DECLARE_ASM_CONST(8, uint64_t, g16Mask)= 0x07E007E007E007E0LL;
201 DECLARE_ASM_CONST(8, uint64_t, r16Mask)= 0xF800F800F800F800LL;
202 DECLARE_ASM_CONST(8, uint64_t, b15Mask)= 0x001F001F001F001FLL;
203 DECLARE_ASM_CONST(8, uint64_t, g15Mask)= 0x03E003E003E003E0LL;
204 DECLARE_ASM_CONST(8, uint64_t, r15Mask)= 0x7C007C007C007C00LL;
206 DECLARE_ALIGNED(8, const uint64_t, ff_M24A) = 0x00FF0000FF0000FFLL;
207 DECLARE_ALIGNED(8, const uint64_t, ff_M24B) = 0xFF0000FF0000FF00LL;
208 DECLARE_ALIGNED(8, const uint64_t, ff_M24C) = 0x0000FF0000FF0000LL;
211 DECLARE_ALIGNED(8, const uint64_t, ff_bgr2YCoeff) = 0x000000210041000DULL;
212 DECLARE_ALIGNED(8, const uint64_t, ff_bgr2UCoeff) = 0x0000FFEEFFDC0038ULL;
213 DECLARE_ALIGNED(8, const uint64_t, ff_bgr2VCoeff) = 0x00000038FFD2FFF8ULL;
215 DECLARE_ALIGNED(8, const uint64_t, ff_bgr2YCoeff) = 0x000020E540830C8BULL;
216 DECLARE_ALIGNED(8, const uint64_t, ff_bgr2UCoeff) = 0x0000ED0FDAC23831ULL;
217 DECLARE_ALIGNED(8, const uint64_t, ff_bgr2VCoeff) = 0x00003831D0E6F6EAULL;
218 #endif /* FAST_BGR2YV12 */
219 DECLARE_ALIGNED(8, const uint64_t, ff_bgr2YOffset) = 0x1010101010101010ULL;
220 DECLARE_ALIGNED(8, const uint64_t, ff_bgr2UVOffset) = 0x8080808080808080ULL;
221 DECLARE_ALIGNED(8, const uint64_t, ff_w1111) = 0x0001000100010001ULL;
222 #endif /* defined(ARCH_X86) */
224 // clipping helper table for C implementations:
225 static unsigned char clip_table[768];
227 static SwsVector *sws_getConvVec(SwsVector *a, SwsVector *b);
229 extern const uint8_t dither_2x2_4[2][8];
230 extern const uint8_t dither_2x2_8[2][8];
231 extern const uint8_t dither_8x8_32[8][8];
232 extern const uint8_t dither_8x8_73[8][8];
233 extern const uint8_t dither_8x8_220[8][8];
235 static const char * sws_context_to_name(void * ptr) {
239 #define OFFSET(x) offsetof(SwsContext, x)
241 #define VE AV_OPT_FLAG_VIDEO_PARAM | AV_OPT_FLAG_ENCODING_PARAM
243 static const AVOption options[] = {
244 { "sws_flags", "scaler/cpu flags", OFFSET(flags), FF_OPT_TYPE_FLAGS, DEFAULT, 0, UINT_MAX, VE, "sws_flags" },
245 { "fast_bilinear", "fast bilinear", 0, FF_OPT_TYPE_CONST, SWS_FAST_BILINEAR, INT_MIN, INT_MAX, VE, "sws_flags" },
246 { "bilinear", "bilinear", 0, FF_OPT_TYPE_CONST, SWS_BILINEAR, INT_MIN, INT_MAX, VE, "sws_flags" },
247 { "bicubic", "bicubic", 0, FF_OPT_TYPE_CONST, SWS_BICUBIC, INT_MIN, INT_MAX, VE, "sws_flags" },
248 { "experimental", "experimental", 0, FF_OPT_TYPE_CONST, SWS_X, INT_MIN, INT_MAX, VE, "sws_flags" },
249 { "neighbor", "nearest neighbor", 0, FF_OPT_TYPE_CONST, SWS_POINT, INT_MIN, INT_MAX, VE, "sws_flags" },
250 { "area", "averaging area", 0, FF_OPT_TYPE_CONST, SWS_AREA, INT_MIN, INT_MAX, VE, "sws_flags" },
251 { "bicublin", "luma bicubic, chroma bilinear", 0, FF_OPT_TYPE_CONST, SWS_BICUBLIN, INT_MIN, INT_MAX, VE, "sws_flags" },
252 { "gauss", "gaussian", 0, FF_OPT_TYPE_CONST, SWS_GAUSS, INT_MIN, INT_MAX, VE, "sws_flags" },
253 { "sinc", "sinc", 0, FF_OPT_TYPE_CONST, SWS_SINC, INT_MIN, INT_MAX, VE, "sws_flags" },
254 { "lanczos", "lanczos", 0, FF_OPT_TYPE_CONST, SWS_LANCZOS, INT_MIN, INT_MAX, VE, "sws_flags" },
255 { "spline", "natural bicubic spline", 0, FF_OPT_TYPE_CONST, SWS_SPLINE, INT_MIN, INT_MAX, VE, "sws_flags" },
256 { "print_info", "print info", 0, FF_OPT_TYPE_CONST, SWS_PRINT_INFO, INT_MIN, INT_MAX, VE, "sws_flags" },
257 { "accurate_rnd", "accurate rounding", 0, FF_OPT_TYPE_CONST, SWS_ACCURATE_RND, INT_MIN, INT_MAX, VE, "sws_flags" },
258 { "mmx", "MMX SIMD acceleration", 0, FF_OPT_TYPE_CONST, SWS_CPU_CAPS_MMX, INT_MIN, INT_MAX, VE, "sws_flags" },
259 { "mmx2", "MMX2 SIMD acceleration", 0, FF_OPT_TYPE_CONST, SWS_CPU_CAPS_MMX2, INT_MIN, INT_MAX, VE, "sws_flags" },
260 { "3dnow", "3DNOW SIMD acceleration", 0, FF_OPT_TYPE_CONST, SWS_CPU_CAPS_3DNOW, INT_MIN, INT_MAX, VE, "sws_flags" },
261 { "altivec", "AltiVec SIMD acceleration", 0, FF_OPT_TYPE_CONST, SWS_CPU_CAPS_ALTIVEC, INT_MIN, INT_MAX, VE, "sws_flags" },
262 { "bfin", "Blackfin SIMD acceleration", 0, FF_OPT_TYPE_CONST, SWS_CPU_CAPS_BFIN, INT_MIN, INT_MAX, VE, "sws_flags" },
263 { "full_chroma_int", "full chroma interpolation", 0 , FF_OPT_TYPE_CONST, SWS_FULL_CHR_H_INT, INT_MIN, INT_MAX, VE, "sws_flags" },
264 { "full_chroma_inp", "full chroma input", 0 , FF_OPT_TYPE_CONST, SWS_FULL_CHR_H_INP, INT_MIN, INT_MAX, VE, "sws_flags" },
271 static const AVClass sws_context_class = { "SWScaler", sws_context_to_name, options };
273 const char *sws_format_name(enum PixelFormat format)
276 case PIX_FMT_YUV420P:
278 case PIX_FMT_YUVA420P:
280 case PIX_FMT_YUYV422:
286 case PIX_FMT_YUV422P:
288 case PIX_FMT_YUV444P:
292 case PIX_FMT_YUV410P:
294 case PIX_FMT_YUV411P:
300 case PIX_FMT_GRAY16BE:
302 case PIX_FMT_GRAY16LE:
306 case PIX_FMT_MONOWHITE:
308 case PIX_FMT_MONOBLACK:
312 case PIX_FMT_YUVJ420P:
314 case PIX_FMT_YUVJ422P:
316 case PIX_FMT_YUVJ444P:
318 case PIX_FMT_XVMC_MPEG2_MC:
319 return "xvmc_mpeg2_mc";
320 case PIX_FMT_XVMC_MPEG2_IDCT:
321 return "xvmc_mpeg2_idct";
322 case PIX_FMT_UYVY422:
324 case PIX_FMT_UYYVYY411:
326 case PIX_FMT_RGB32_1:
328 case PIX_FMT_BGR32_1:
340 case PIX_FMT_BGR4_BYTE:
346 case PIX_FMT_RGB4_BYTE:
352 case PIX_FMT_YUV440P:
355 return "Unknown format";
359 static inline void yuv2yuvXinC(int16_t *lumFilter, int16_t **lumSrc, int lumFilterSize,
360 int16_t *chrFilter, int16_t **chrSrc, int chrFilterSize,
361 uint8_t *dest, uint8_t *uDest, uint8_t *vDest, int dstW, int chrDstW)
363 //FIXME Optimize (just quickly writen not opti..)
365 for (i=0; i<dstW; i++)
369 for (j=0; j<lumFilterSize; j++)
370 val += lumSrc[j][i] * lumFilter[j];
372 dest[i]= av_clip_uint8(val>>19);
376 for (i=0; i<chrDstW; i++)
381 for (j=0; j<chrFilterSize; j++)
383 u += chrSrc[j][i] * chrFilter[j];
384 v += chrSrc[j][i + VOFW] * chrFilter[j];
387 uDest[i]= av_clip_uint8(u>>19);
388 vDest[i]= av_clip_uint8(v>>19);
392 static inline void yuv2nv12XinC(int16_t *lumFilter, int16_t **lumSrc, int lumFilterSize,
393 int16_t *chrFilter, int16_t **chrSrc, int chrFilterSize,
394 uint8_t *dest, uint8_t *uDest, int dstW, int chrDstW, int dstFormat)
396 //FIXME Optimize (just quickly writen not opti..)
398 for (i=0; i<dstW; i++)
402 for (j=0; j<lumFilterSize; j++)
403 val += lumSrc[j][i] * lumFilter[j];
405 dest[i]= av_clip_uint8(val>>19);
411 if (dstFormat == PIX_FMT_NV12)
412 for (i=0; i<chrDstW; i++)
417 for (j=0; j<chrFilterSize; j++)
419 u += chrSrc[j][i] * chrFilter[j];
420 v += chrSrc[j][i + VOFW] * chrFilter[j];
423 uDest[2*i]= av_clip_uint8(u>>19);
424 uDest[2*i+1]= av_clip_uint8(v>>19);
427 for (i=0; i<chrDstW; i++)
432 for (j=0; j<chrFilterSize; j++)
434 u += chrSrc[j][i] * chrFilter[j];
435 v += chrSrc[j][i + VOFW] * chrFilter[j];
438 uDest[2*i]= av_clip_uint8(v>>19);
439 uDest[2*i+1]= av_clip_uint8(u>>19);
443 #define YSCALE_YUV_2_PACKEDX_C(type) \
444 for (i=0; i<(dstW>>1); i++){\
450 type av_unused *r, *b, *g;\
453 for (j=0; j<lumFilterSize; j++)\
455 Y1 += lumSrc[j][i2] * lumFilter[j];\
456 Y2 += lumSrc[j][i2+1] * lumFilter[j];\
458 for (j=0; j<chrFilterSize; j++)\
460 U += chrSrc[j][i] * chrFilter[j];\
461 V += chrSrc[j][i+VOFW] * chrFilter[j];\
467 if ((Y1|Y2|U|V)&256)\
469 if (Y1>255) Y1=255; \
470 else if (Y1<0)Y1=0; \
471 if (Y2>255) Y2=255; \
472 else if (Y2<0)Y2=0; \
479 #define YSCALE_YUV_2_RGBX_C(type) \
480 YSCALE_YUV_2_PACKEDX_C(type) \
481 r = (type *)c->table_rV[V]; \
482 g = (type *)(c->table_gU[U] + c->table_gV[V]); \
483 b = (type *)c->table_bU[U]; \
485 #define YSCALE_YUV_2_PACKED2_C \
486 for (i=0; i<(dstW>>1); i++){ \
488 int Y1= (buf0[i2 ]*yalpha1+buf1[i2 ]*yalpha)>>19; \
489 int Y2= (buf0[i2+1]*yalpha1+buf1[i2+1]*yalpha)>>19; \
490 int U= (uvbuf0[i ]*uvalpha1+uvbuf1[i ]*uvalpha)>>19; \
491 int V= (uvbuf0[i+VOFW]*uvalpha1+uvbuf1[i+VOFW]*uvalpha)>>19; \
493 #define YSCALE_YUV_2_RGB2_C(type) \
494 YSCALE_YUV_2_PACKED2_C\
496 r = (type *)c->table_rV[V];\
497 g = (type *)(c->table_gU[U] + c->table_gV[V]);\
498 b = (type *)c->table_bU[U];\
500 #define YSCALE_YUV_2_PACKED1_C \
501 for (i=0; i<(dstW>>1); i++){\
503 int Y1= buf0[i2 ]>>7;\
504 int Y2= buf0[i2+1]>>7;\
505 int U= (uvbuf1[i ])>>7;\
506 int V= (uvbuf1[i+VOFW])>>7;\
508 #define YSCALE_YUV_2_RGB1_C(type) \
509 YSCALE_YUV_2_PACKED1_C\
511 r = (type *)c->table_rV[V];\
512 g = (type *)(c->table_gU[U] + c->table_gV[V]);\
513 b = (type *)c->table_bU[U];\
515 #define YSCALE_YUV_2_PACKED1B_C \
516 for (i=0; i<(dstW>>1); i++){\
518 int Y1= buf0[i2 ]>>7;\
519 int Y2= buf0[i2+1]>>7;\
520 int U= (uvbuf0[i ] + uvbuf1[i ])>>8;\
521 int V= (uvbuf0[i+VOFW] + uvbuf1[i+VOFW])>>8;\
523 #define YSCALE_YUV_2_RGB1B_C(type) \
524 YSCALE_YUV_2_PACKED1B_C\
526 r = (type *)c->table_rV[V];\
527 g = (type *)(c->table_gU[U] + c->table_gV[V]);\
528 b = (type *)c->table_bU[U];\
530 #define YSCALE_YUV_2_ANYRGB_C(func, func2)\
531 switch(c->dstFormat)\
536 ((uint32_t*)dest)[i2+0]= r[Y1] + g[Y1] + b[Y1];\
537 ((uint32_t*)dest)[i2+1]= r[Y2] + g[Y2] + b[Y2];\
542 ((uint8_t*)dest)[0]= r[Y1];\
543 ((uint8_t*)dest)[1]= g[Y1];\
544 ((uint8_t*)dest)[2]= b[Y1];\
545 ((uint8_t*)dest)[3]= r[Y2];\
546 ((uint8_t*)dest)[4]= g[Y2];\
547 ((uint8_t*)dest)[5]= b[Y2];\
553 ((uint8_t*)dest)[0]= b[Y1];\
554 ((uint8_t*)dest)[1]= g[Y1];\
555 ((uint8_t*)dest)[2]= r[Y1];\
556 ((uint8_t*)dest)[3]= b[Y2];\
557 ((uint8_t*)dest)[4]= g[Y2];\
558 ((uint8_t*)dest)[5]= r[Y2];\
562 case PIX_FMT_RGB565:\
563 case PIX_FMT_BGR565:\
565 const int dr1= dither_2x2_8[y&1 ][0];\
566 const int dg1= dither_2x2_4[y&1 ][0];\
567 const int db1= dither_2x2_8[(y&1)^1][0];\
568 const int dr2= dither_2x2_8[y&1 ][1];\
569 const int dg2= dither_2x2_4[y&1 ][1];\
570 const int db2= dither_2x2_8[(y&1)^1][1];\
572 ((uint16_t*)dest)[i2+0]= r[Y1+dr1] + g[Y1+dg1] + b[Y1+db1];\
573 ((uint16_t*)dest)[i2+1]= r[Y2+dr2] + g[Y2+dg2] + b[Y2+db2];\
577 case PIX_FMT_RGB555:\
578 case PIX_FMT_BGR555:\
580 const int dr1= dither_2x2_8[y&1 ][0];\
581 const int dg1= dither_2x2_8[y&1 ][1];\
582 const int db1= dither_2x2_8[(y&1)^1][0];\
583 const int dr2= dither_2x2_8[y&1 ][1];\
584 const int dg2= dither_2x2_8[y&1 ][0];\
585 const int db2= dither_2x2_8[(y&1)^1][1];\
587 ((uint16_t*)dest)[i2+0]= r[Y1+dr1] + g[Y1+dg1] + b[Y1+db1];\
588 ((uint16_t*)dest)[i2+1]= r[Y2+dr2] + g[Y2+dg2] + b[Y2+db2];\
595 const uint8_t * const d64= dither_8x8_73[y&7];\
596 const uint8_t * const d32= dither_8x8_32[y&7];\
598 ((uint8_t*)dest)[i2+0]= r[Y1+d32[(i2+0)&7]] + g[Y1+d32[(i2+0)&7]] + b[Y1+d64[(i2+0)&7]];\
599 ((uint8_t*)dest)[i2+1]= r[Y2+d32[(i2+1)&7]] + g[Y2+d32[(i2+1)&7]] + b[Y2+d64[(i2+1)&7]];\
606 const uint8_t * const d64= dither_8x8_73 [y&7];\
607 const uint8_t * const d128=dither_8x8_220[y&7];\
609 ((uint8_t*)dest)[i]= r[Y1+d128[(i2+0)&7]] + g[Y1+d64[(i2+0)&7]] + b[Y1+d128[(i2+0)&7]]\
610 + ((r[Y2+d128[(i2+1)&7]] + g[Y2+d64[(i2+1)&7]] + b[Y2+d128[(i2+1)&7]])<<4);\
614 case PIX_FMT_RGB4_BYTE:\
615 case PIX_FMT_BGR4_BYTE:\
617 const uint8_t * const d64= dither_8x8_73 [y&7];\
618 const uint8_t * const d128=dither_8x8_220[y&7];\
620 ((uint8_t*)dest)[i2+0]= r[Y1+d128[(i2+0)&7]] + g[Y1+d64[(i2+0)&7]] + b[Y1+d128[(i2+0)&7]];\
621 ((uint8_t*)dest)[i2+1]= r[Y2+d128[(i2+1)&7]] + g[Y2+d64[(i2+1)&7]] + b[Y2+d128[(i2+1)&7]];\
625 case PIX_FMT_MONOBLACK:\
627 const uint8_t * const d128=dither_8x8_220[y&7];\
628 uint8_t *g= c->table_gU[128] + c->table_gV[128];\
629 for (i=0; i<dstW-7; i+=8){\
631 acc = g[((buf0[i ]*yalpha1+buf1[i ]*yalpha)>>19) + d128[0]];\
632 acc+= acc + g[((buf0[i+1]*yalpha1+buf1[i+1]*yalpha)>>19) + d128[1]];\
633 acc+= acc + g[((buf0[i+2]*yalpha1+buf1[i+2]*yalpha)>>19) + d128[2]];\
634 acc+= acc + g[((buf0[i+3]*yalpha1+buf1[i+3]*yalpha)>>19) + d128[3]];\
635 acc+= acc + g[((buf0[i+4]*yalpha1+buf1[i+4]*yalpha)>>19) + d128[4]];\
636 acc+= acc + g[((buf0[i+5]*yalpha1+buf1[i+5]*yalpha)>>19) + d128[5]];\
637 acc+= acc + g[((buf0[i+6]*yalpha1+buf1[i+6]*yalpha)>>19) + d128[6]];\
638 acc+= acc + g[((buf0[i+7]*yalpha1+buf1[i+7]*yalpha)>>19) + d128[7]];\
639 ((uint8_t*)dest)[0]= acc;\
644 ((uint8_t*)dest)-= dstW>>4;\
648 static int top[1024];\
649 static int last_new[1024][1024];\
650 static int last_in3[1024][1024];\
651 static int drift[1024][1024];\
655 const uint8_t * const d128=dither_8x8_220[y&7];\
660 for (i=dstW>>1; i<dstW; i++){\
661 int in= ((buf0[i ]*yalpha1+buf1[i ]*yalpha)>>19);\
662 int in2 = (76309 * (in - 16) + 32768) >> 16;\
663 int in3 = (in2 < 0) ? 0 : ((in2 > 255) ? 255 : in2);\
664 int old= (left*7 + topLeft + top[i]*5 + top[i+1]*3)/20 + in3\
665 + (last_new[y][i] - in3)*f/256;\
666 int new= old> 128 ? 255 : 0;\
668 error_new+= FFABS(last_new[y][i] - new);\
669 error_in3+= FFABS(last_in3[y][i] - in3);\
670 f= error_new - error_in3*4;\
675 left= top[i]= old - new;\
676 last_new[y][i]= new;\
677 last_in3[y][i]= in3;\
679 acc+= acc + (new&1);\
681 ((uint8_t*)dest)[0]= acc;\
689 case PIX_FMT_YUYV422:\
691 ((uint8_t*)dest)[2*i2+0]= Y1;\
692 ((uint8_t*)dest)[2*i2+1]= U;\
693 ((uint8_t*)dest)[2*i2+2]= Y2;\
694 ((uint8_t*)dest)[2*i2+3]= V;\
697 case PIX_FMT_UYVY422:\
699 ((uint8_t*)dest)[2*i2+0]= U;\
700 ((uint8_t*)dest)[2*i2+1]= Y1;\
701 ((uint8_t*)dest)[2*i2+2]= V;\
702 ((uint8_t*)dest)[2*i2+3]= Y2;\
708 static inline void yuv2packedXinC(SwsContext *c, int16_t *lumFilter, int16_t **lumSrc, int lumFilterSize,
709 int16_t *chrFilter, int16_t **chrSrc, int chrFilterSize,
710 uint8_t *dest, int dstW, int y)
717 YSCALE_YUV_2_RGBX_C(uint32_t)
718 ((uint32_t*)dest)[i2+0]= r[Y1] + g[Y1] + b[Y1];
719 ((uint32_t*)dest)[i2+1]= r[Y2] + g[Y2] + b[Y2];
723 YSCALE_YUV_2_RGBX_C(uint8_t)
724 ((uint8_t*)dest)[0]= r[Y1];
725 ((uint8_t*)dest)[1]= g[Y1];
726 ((uint8_t*)dest)[2]= b[Y1];
727 ((uint8_t*)dest)[3]= r[Y2];
728 ((uint8_t*)dest)[4]= g[Y2];
729 ((uint8_t*)dest)[5]= b[Y2];
734 YSCALE_YUV_2_RGBX_C(uint8_t)
735 ((uint8_t*)dest)[0]= b[Y1];
736 ((uint8_t*)dest)[1]= g[Y1];
737 ((uint8_t*)dest)[2]= r[Y1];
738 ((uint8_t*)dest)[3]= b[Y2];
739 ((uint8_t*)dest)[4]= g[Y2];
740 ((uint8_t*)dest)[5]= r[Y2];
747 const int dr1= dither_2x2_8[y&1 ][0];
748 const int dg1= dither_2x2_4[y&1 ][0];
749 const int db1= dither_2x2_8[(y&1)^1][0];
750 const int dr2= dither_2x2_8[y&1 ][1];
751 const int dg2= dither_2x2_4[y&1 ][1];
752 const int db2= dither_2x2_8[(y&1)^1][1];
753 YSCALE_YUV_2_RGBX_C(uint16_t)
754 ((uint16_t*)dest)[i2+0]= r[Y1+dr1] + g[Y1+dg1] + b[Y1+db1];
755 ((uint16_t*)dest)[i2+1]= r[Y2+dr2] + g[Y2+dg2] + b[Y2+db2];
762 const int dr1= dither_2x2_8[y&1 ][0];
763 const int dg1= dither_2x2_8[y&1 ][1];
764 const int db1= dither_2x2_8[(y&1)^1][0];
765 const int dr2= dither_2x2_8[y&1 ][1];
766 const int dg2= dither_2x2_8[y&1 ][0];
767 const int db2= dither_2x2_8[(y&1)^1][1];
768 YSCALE_YUV_2_RGBX_C(uint16_t)
769 ((uint16_t*)dest)[i2+0]= r[Y1+dr1] + g[Y1+dg1] + b[Y1+db1];
770 ((uint16_t*)dest)[i2+1]= r[Y2+dr2] + g[Y2+dg2] + b[Y2+db2];
777 const uint8_t * const d64= dither_8x8_73[y&7];
778 const uint8_t * const d32= dither_8x8_32[y&7];
779 YSCALE_YUV_2_RGBX_C(uint8_t)
780 ((uint8_t*)dest)[i2+0]= r[Y1+d32[(i2+0)&7]] + g[Y1+d32[(i2+0)&7]] + b[Y1+d64[(i2+0)&7]];
781 ((uint8_t*)dest)[i2+1]= r[Y2+d32[(i2+1)&7]] + g[Y2+d32[(i2+1)&7]] + b[Y2+d64[(i2+1)&7]];
788 const uint8_t * const d64= dither_8x8_73 [y&7];
789 const uint8_t * const d128=dither_8x8_220[y&7];
790 YSCALE_YUV_2_RGBX_C(uint8_t)
791 ((uint8_t*)dest)[i]= r[Y1+d128[(i2+0)&7]] + g[Y1+d64[(i2+0)&7]] + b[Y1+d128[(i2+0)&7]]
792 +((r[Y2+d128[(i2+1)&7]] + g[Y2+d64[(i2+1)&7]] + b[Y2+d128[(i2+1)&7]])<<4);
796 case PIX_FMT_RGB4_BYTE:
797 case PIX_FMT_BGR4_BYTE:
799 const uint8_t * const d64= dither_8x8_73 [y&7];
800 const uint8_t * const d128=dither_8x8_220[y&7];
801 YSCALE_YUV_2_RGBX_C(uint8_t)
802 ((uint8_t*)dest)[i2+0]= r[Y1+d128[(i2+0)&7]] + g[Y1+d64[(i2+0)&7]] + b[Y1+d128[(i2+0)&7]];
803 ((uint8_t*)dest)[i2+1]= r[Y2+d128[(i2+1)&7]] + g[Y2+d64[(i2+1)&7]] + b[Y2+d128[(i2+1)&7]];
807 case PIX_FMT_MONOBLACK:
809 const uint8_t * const d128=dither_8x8_220[y&7];
810 uint8_t *g= c->table_gU[128] + c->table_gV[128];
812 for (i=0; i<dstW-1; i+=2){
817 for (j=0; j<lumFilterSize; j++)
819 Y1 += lumSrc[j][i] * lumFilter[j];
820 Y2 += lumSrc[j][i+1] * lumFilter[j];
831 acc+= acc + g[Y1+d128[(i+0)&7]];
832 acc+= acc + g[Y2+d128[(i+1)&7]];
834 ((uint8_t*)dest)[0]= acc;
840 case PIX_FMT_YUYV422:
841 YSCALE_YUV_2_PACKEDX_C(void)
842 ((uint8_t*)dest)[2*i2+0]= Y1;
843 ((uint8_t*)dest)[2*i2+1]= U;
844 ((uint8_t*)dest)[2*i2+2]= Y2;
845 ((uint8_t*)dest)[2*i2+3]= V;
848 case PIX_FMT_UYVY422:
849 YSCALE_YUV_2_PACKEDX_C(void)
850 ((uint8_t*)dest)[2*i2+0]= U;
851 ((uint8_t*)dest)[2*i2+1]= Y1;
852 ((uint8_t*)dest)[2*i2+2]= V;
853 ((uint8_t*)dest)[2*i2+3]= Y2;
860 //Note: we have C, X86, MMX, MMX2, 3DNOW version therse no 3DNOW+MMX2 one
862 #if !defined (HAVE_MMX) || defined (RUNTIME_CPUDETECT) || !defined(CONFIG_GPL)
867 #if (defined (HAVE_ALTIVEC) || defined (RUNTIME_CPUDETECT)) && defined (CONFIG_GPL)
868 #define COMPILE_ALTIVEC
869 #endif //HAVE_ALTIVEC
870 #endif //ARCH_POWERPC
872 #if defined(ARCH_X86)
874 #if ((defined (HAVE_MMX) && !defined (HAVE_3DNOW) && !defined (HAVE_MMX2)) || defined (RUNTIME_CPUDETECT)) && defined (CONFIG_GPL)
878 #if (defined (HAVE_MMX2) || defined (RUNTIME_CPUDETECT)) && defined (CONFIG_GPL)
882 #if ((defined (HAVE_3DNOW) && !defined (HAVE_MMX2)) || defined (RUNTIME_CPUDETECT)) && defined (CONFIG_GPL)
883 #define COMPILE_3DNOW
885 #endif //ARCH_X86 || ARCH_X86_64
896 #define RENAME(a) a ## _C
897 #include "swscale_template.c"
900 #ifdef COMPILE_ALTIVEC
903 #define RENAME(a) a ## _altivec
904 #include "swscale_template.c"
907 #if defined(ARCH_X86)
916 #define RENAME(a) a ## _X86
917 #include "swscale_template.c"
925 #define RENAME(a) a ## _MMX
926 #include "swscale_template.c"
935 #define RENAME(a) a ## _MMX2
936 #include "swscale_template.c"
945 #define RENAME(a) a ## _3DNow
946 #include "swscale_template.c"
949 #endif //ARCH_X86 || ARCH_X86_64
951 // minor note: the HAVE_xyz is messed up after that line so don't use it
953 static double getSplineCoeff(double a, double b, double c, double d, double dist)
955 // printf("%f %f %f %f %f\n", a,b,c,d,dist);
956 if (dist<=1.0) return ((d*dist + c)*dist + b)*dist +a;
957 else return getSplineCoeff( 0.0,
964 static inline int initFilter(int16_t **outFilter, int16_t **filterPos, int *outFilterSize, int xInc,
965 int srcW, int dstW, int filterAlign, int one, int flags,
966 SwsVector *srcFilter, SwsVector *dstFilter, double param[2])
973 double *filter2=NULL;
974 #if defined(ARCH_X86)
975 if (flags & SWS_CPU_CAPS_MMX)
976 asm volatile("emms\n\t"::: "memory"); //FIXME this should not be required but it IS (even for non-MMX versions)
979 // Note the +1 is for the MMXscaler which reads over the end
980 *filterPos = av_malloc((dstW+1)*sizeof(int16_t));
982 if (FFABS(xInc - 0x10000) <10) // unscaled
986 filter= av_malloc(dstW*sizeof(double)*filterSize);
987 for (i=0; i<dstW*filterSize; i++) filter[i]=0;
989 for (i=0; i<dstW; i++)
991 filter[i*filterSize]=1;
996 else if (flags&SWS_POINT) // lame looking point sampling mode
1001 filter= av_malloc(dstW*sizeof(double)*filterSize);
1003 xDstInSrc= xInc/2 - 0x8000;
1004 for (i=0; i<dstW; i++)
1006 int xx= (xDstInSrc - ((filterSize-1)<<15) + (1<<15))>>16;
1008 (*filterPos)[i]= xx;
1013 else if ((xInc <= (1<<16) && (flags&SWS_AREA)) || (flags&SWS_FAST_BILINEAR)) // bilinear upscale
1017 if (flags&SWS_BICUBIC) filterSize= 4;
1018 else if (flags&SWS_X ) filterSize= 4;
1019 else filterSize= 2; // SWS_BILINEAR / SWS_AREA
1020 filter= av_malloc(dstW*sizeof(double)*filterSize);
1022 xDstInSrc= xInc/2 - 0x8000;
1023 for (i=0; i<dstW; i++)
1025 int xx= (xDstInSrc - ((filterSize-1)<<15) + (1<<15))>>16;
1028 (*filterPos)[i]= xx;
1029 //Bilinear upscale / linear interpolate / Area averaging
1030 for (j=0; j<filterSize; j++)
1032 double d= FFABS((xx<<16) - xDstInSrc)/(double)(1<<16);
1033 double coeff= 1.0 - d;
1034 if (coeff<0) coeff=0;
1035 filter[i*filterSize + j]= coeff;
1044 double sizeFactor, filterSizeInSrc;
1045 const double xInc1= (double)xInc / (double)(1<<16);
1047 if (flags&SWS_BICUBIC) sizeFactor= 4.0;
1048 else if (flags&SWS_X) sizeFactor= 8.0;
1049 else if (flags&SWS_AREA) sizeFactor= 1.0; //downscale only, for upscale it is bilinear
1050 else if (flags&SWS_GAUSS) sizeFactor= 8.0; // infinite ;)
1051 else if (flags&SWS_LANCZOS) sizeFactor= param[0] != SWS_PARAM_DEFAULT ? 2.0*param[0] : 6.0;
1052 else if (flags&SWS_SINC) sizeFactor= 20.0; // infinite ;)
1053 else if (flags&SWS_SPLINE) sizeFactor= 20.0; // infinite ;)
1054 else if (flags&SWS_BILINEAR) sizeFactor= 2.0;
1056 sizeFactor= 0.0; //GCC warning killer
1060 if (xInc1 <= 1.0) filterSizeInSrc= sizeFactor; // upscale
1061 else filterSizeInSrc= sizeFactor*srcW / (double)dstW;
1063 filterSize= (int)ceil(1 + filterSizeInSrc); // will be reduced later if possible
1064 if (filterSize > srcW-2) filterSize=srcW-2;
1066 filter= av_malloc(dstW*sizeof(double)*filterSize);
1068 xDstInSrc= xInc1 / 2.0 - 0.5;
1069 for (i=0; i<dstW; i++)
1071 int xx= (int)(xDstInSrc - (filterSize-1)*0.5 + 0.5);
1073 (*filterPos)[i]= xx;
1074 for (j=0; j<filterSize; j++)
1076 double d= FFABS(xx - xDstInSrc)/filterSizeInSrc*sizeFactor;
1078 if (flags & SWS_BICUBIC)
1080 double B= param[0] != SWS_PARAM_DEFAULT ? param[0] : 0.0;
1081 double C= param[1] != SWS_PARAM_DEFAULT ? param[1] : 0.6;
1084 coeff = (12-9*B-6*C)*d*d*d + (-18+12*B+6*C)*d*d + 6-2*B;
1086 coeff = (-B-6*C)*d*d*d + (6*B+30*C)*d*d + (-12*B-48*C)*d +8*B+24*C;
1090 /* else if (flags & SWS_X)
1092 double p= param ? param*0.01 : 0.3;
1093 coeff = d ? sin(d*PI)/(d*PI) : 1.0;
1094 coeff*= pow(2.0, - p*d*d);
1096 else if (flags & SWS_X)
1098 double A= param[0] != SWS_PARAM_DEFAULT ? param[0] : 1.0;
1104 if (coeff<0.0) coeff= -pow(-coeff, A);
1105 else coeff= pow( coeff, A);
1106 coeff= coeff*0.5 + 0.5;
1108 else if (flags & SWS_AREA)
1110 double srcPixelSize= 1.0/xInc1;
1111 if (d + srcPixelSize/2 < 0.5) coeff= 1.0;
1112 else if (d - srcPixelSize/2 < 0.5) coeff= (0.5-d)/srcPixelSize + 0.5;
1115 else if (flags & SWS_GAUSS)
1117 double p= param[0] != SWS_PARAM_DEFAULT ? param[0] : 3.0;
1118 coeff = pow(2.0, - p*d*d);
1120 else if (flags & SWS_SINC)
1122 coeff = d ? sin(d*PI)/(d*PI) : 1.0;
1124 else if (flags & SWS_LANCZOS)
1126 double p= param[0] != SWS_PARAM_DEFAULT ? param[0] : 3.0;
1127 coeff = d ? sin(d*PI)*sin(d*PI/p)/(d*d*PI*PI/p) : 1.0;
1130 else if (flags & SWS_BILINEAR)
1133 if (coeff<0) coeff=0;
1135 else if (flags & SWS_SPLINE)
1137 double p=-2.196152422706632;
1138 coeff = getSplineCoeff(1.0, 0.0, p, -p-1.0, d);
1141 coeff= 0.0; //GCC warning killer
1145 filter[i*filterSize + j]= coeff;
1152 /* apply src & dst Filter to filter -> filter2
1155 assert(filterSize>0);
1156 filter2Size= filterSize;
1157 if (srcFilter) filter2Size+= srcFilter->length - 1;
1158 if (dstFilter) filter2Size+= dstFilter->length - 1;
1159 assert(filter2Size>0);
1160 filter2= av_malloc(filter2Size*dstW*sizeof(double));
1162 for (i=0; i<dstW; i++)
1165 SwsVector scaleFilter;
1168 scaleFilter.coeff= filter + i*filterSize;
1169 scaleFilter.length= filterSize;
1171 if (srcFilter) outVec= sws_getConvVec(srcFilter, &scaleFilter);
1172 else outVec= &scaleFilter;
1174 assert(outVec->length == filter2Size);
1177 for (j=0; j<outVec->length; j++)
1179 filter2[i*filter2Size + j]= outVec->coeff[j];
1182 (*filterPos)[i]+= (filterSize-1)/2 - (filter2Size-1)/2;
1184 if (outVec != &scaleFilter) sws_freeVec(outVec);
1186 av_free(filter); filter=NULL;
1188 /* try to reduce the filter-size (step1 find size and shift left) */
1189 // Assume it is near normalized (*0.5 or *2.0 is OK but * 0.001 is not).
1191 for (i=dstW-1; i>=0; i--)
1193 int min= filter2Size;
1197 /* get rid off near zero elements on the left by shifting left */
1198 for (j=0; j<filter2Size; j++)
1201 cutOff += FFABS(filter2[i*filter2Size]);
1203 if (cutOff > SWS_MAX_REDUCE_CUTOFF) break;
1205 /* preserve monotonicity because the core can't handle the filter otherwise */
1206 if (i<dstW-1 && (*filterPos)[i] >= (*filterPos)[i+1]) break;
1208 // Move filter coeffs left
1209 for (k=1; k<filter2Size; k++)
1210 filter2[i*filter2Size + k - 1]= filter2[i*filter2Size + k];
1211 filter2[i*filter2Size + k - 1]= 0.0;
1216 /* count near zeros on the right */
1217 for (j=filter2Size-1; j>0; j--)
1219 cutOff += FFABS(filter2[i*filter2Size + j]);
1221 if (cutOff > SWS_MAX_REDUCE_CUTOFF) break;
1225 if (min>minFilterSize) minFilterSize= min;
1228 if (flags & SWS_CPU_CAPS_ALTIVEC) {
1229 // we can handle the special case 4,
1230 // so we don't want to go to the full 8
1231 if (minFilterSize < 5)
1234 // we really don't want to waste our time
1235 // doing useless computation, so fall-back on
1236 // the scalar C code for very small filter.
1237 // vectorizing is worth it only if you have
1238 // decent-sized vector.
1239 if (minFilterSize < 3)
1243 if (flags & SWS_CPU_CAPS_MMX) {
1244 // special case for unscaled vertical filtering
1245 if (minFilterSize == 1 && filterAlign == 2)
1249 assert(minFilterSize > 0);
1250 filterSize= (minFilterSize +(filterAlign-1)) & (~(filterAlign-1));
1251 assert(filterSize > 0);
1252 filter= av_malloc(filterSize*dstW*sizeof(double));
1253 if (filterSize >= MAX_FILTER_SIZE)
1255 *outFilterSize= filterSize;
1257 if (flags&SWS_PRINT_INFO)
1258 av_log(NULL, AV_LOG_VERBOSE, "SwScaler: reducing / aligning filtersize %d -> %d\n", filter2Size, filterSize);
1259 /* try to reduce the filter-size (step2 reduce it) */
1260 for (i=0; i<dstW; i++)
1264 for (j=0; j<filterSize; j++)
1266 if (j>=filter2Size) filter[i*filterSize + j]= 0.0;
1267 else filter[i*filterSize + j]= filter2[i*filter2Size + j];
1270 av_free(filter2); filter2=NULL;
1273 //FIXME try to align filterpos if possible
1276 for (i=0; i<dstW; i++)
1279 if ((*filterPos)[i] < 0)
1281 // Move filter coeffs left to compensate for filterPos
1282 for (j=1; j<filterSize; j++)
1284 int left= FFMAX(j + (*filterPos)[i], 0);
1285 filter[i*filterSize + left] += filter[i*filterSize + j];
1286 filter[i*filterSize + j]=0;
1291 if ((*filterPos)[i] + filterSize > srcW)
1293 int shift= (*filterPos)[i] + filterSize - srcW;
1294 // Move filter coeffs right to compensate for filterPos
1295 for (j=filterSize-2; j>=0; j--)
1297 int right= FFMIN(j + shift, filterSize-1);
1298 filter[i*filterSize +right] += filter[i*filterSize +j];
1299 filter[i*filterSize +j]=0;
1301 (*filterPos)[i]= srcW - filterSize;
1305 // Note the +1 is for the MMXscaler which reads over the end
1306 /* align at 16 for AltiVec (needed by hScale_altivec_real) */
1307 *outFilter= av_mallocz(*outFilterSize*(dstW+1)*sizeof(int16_t));
1309 /* Normalize & Store in outFilter */
1310 for (i=0; i<dstW; i++)
1317 for (j=0; j<filterSize; j++)
1319 sum+= filter[i*filterSize + j];
1322 for (j=0; j<*outFilterSize; j++)
1324 double v= filter[i*filterSize + j]*scale + error;
1325 int intV= floor(v + 0.5);
1326 (*outFilter)[i*(*outFilterSize) + j]= intV;
1331 (*filterPos)[dstW]= (*filterPos)[dstW-1]; // the MMX scaler will read over the end
1332 for (i=0; i<*outFilterSize; i++)
1334 int j= dstW*(*outFilterSize);
1335 (*outFilter)[j + i]= (*outFilter)[j + i - (*outFilterSize)];
1343 static void initMMX2HScaler(int dstW, int xInc, uint8_t *funnyCode, int16_t *filter, int32_t *filterPos, int numSplits)
1346 long imm8OfPShufW1A;
1347 long imm8OfPShufW2A;
1348 long fragmentLengthA;
1350 long imm8OfPShufW1B;
1351 long imm8OfPShufW2B;
1352 long fragmentLengthB;
1357 // create an optimized horizontal scaling routine
1365 "movq (%%"REG_d", %%"REG_a"), %%mm3 \n\t"
1366 "movd (%%"REG_c", %%"REG_S"), %%mm0 \n\t"
1367 "movd 1(%%"REG_c", %%"REG_S"), %%mm1 \n\t"
1368 "punpcklbw %%mm7, %%mm1 \n\t"
1369 "punpcklbw %%mm7, %%mm0 \n\t"
1370 "pshufw $0xFF, %%mm1, %%mm1 \n\t"
1372 "pshufw $0xFF, %%mm0, %%mm0 \n\t"
1374 "psubw %%mm1, %%mm0 \n\t"
1375 "movl 8(%%"REG_b", %%"REG_a"), %%esi \n\t"
1376 "pmullw %%mm3, %%mm0 \n\t"
1377 "psllw $7, %%mm1 \n\t"
1378 "paddw %%mm1, %%mm0 \n\t"
1380 "movq %%mm0, (%%"REG_D", %%"REG_a") \n\t"
1382 "add $8, %%"REG_a" \n\t"
1386 "lea " LOCAL_MANGLE(0b) ", %0 \n\t"
1387 "lea " LOCAL_MANGLE(1b) ", %1 \n\t"
1388 "lea " LOCAL_MANGLE(2b) ", %2 \n\t"
1393 "lea " LOCAL_MANGLE(9b) ", %3 \n\t"
1397 :"=r" (fragmentA), "=r" (imm8OfPShufW1A), "=r" (imm8OfPShufW2A),
1398 "=r" (fragmentLengthA)
1405 "movq (%%"REG_d", %%"REG_a"), %%mm3 \n\t"
1406 "movd (%%"REG_c", %%"REG_S"), %%mm0 \n\t"
1407 "punpcklbw %%mm7, %%mm0 \n\t"
1408 "pshufw $0xFF, %%mm0, %%mm1 \n\t"
1410 "pshufw $0xFF, %%mm0, %%mm0 \n\t"
1412 "psubw %%mm1, %%mm0 \n\t"
1413 "movl 8(%%"REG_b", %%"REG_a"), %%esi \n\t"
1414 "pmullw %%mm3, %%mm0 \n\t"
1415 "psllw $7, %%mm1 \n\t"
1416 "paddw %%mm1, %%mm0 \n\t"
1418 "movq %%mm0, (%%"REG_D", %%"REG_a") \n\t"
1420 "add $8, %%"REG_a" \n\t"
1424 "lea " LOCAL_MANGLE(0b) ", %0 \n\t"
1425 "lea " LOCAL_MANGLE(1b) ", %1 \n\t"
1426 "lea " LOCAL_MANGLE(2b) ", %2 \n\t"
1431 "lea " LOCAL_MANGLE(9b) ", %3 \n\t"
1435 :"=r" (fragmentB), "=r" (imm8OfPShufW1B), "=r" (imm8OfPShufW2B),
1436 "=r" (fragmentLengthB)
1439 xpos= 0; //lumXInc/2 - 0x8000; // difference between pixel centers
1442 for (i=0; i<dstW/numSplits; i++)
1449 int b=((xpos+xInc)>>16) - xx;
1450 int c=((xpos+xInc*2)>>16) - xx;
1451 int d=((xpos+xInc*3)>>16) - xx;
1453 filter[i ] = (( xpos & 0xFFFF) ^ 0xFFFF)>>9;
1454 filter[i+1] = (((xpos+xInc ) & 0xFFFF) ^ 0xFFFF)>>9;
1455 filter[i+2] = (((xpos+xInc*2) & 0xFFFF) ^ 0xFFFF)>>9;
1456 filter[i+3] = (((xpos+xInc*3) & 0xFFFF) ^ 0xFFFF)>>9;
1461 int maxShift= 3-(d+1);
1464 memcpy(funnyCode + fragmentPos, fragmentB, fragmentLengthB);
1466 funnyCode[fragmentPos + imm8OfPShufW1B]=
1467 (a+1) | ((b+1)<<2) | ((c+1)<<4) | ((d+1)<<6);
1468 funnyCode[fragmentPos + imm8OfPShufW2B]=
1469 a | (b<<2) | (c<<4) | (d<<6);
1471 if (i+3>=dstW) shift=maxShift; //avoid overread
1472 else if ((filterPos[i/2]&3) <= maxShift) shift=filterPos[i/2]&3; //Align
1474 if (shift && i>=shift)
1476 funnyCode[fragmentPos + imm8OfPShufW1B]+= 0x55*shift;
1477 funnyCode[fragmentPos + imm8OfPShufW2B]+= 0x55*shift;
1478 filterPos[i/2]-=shift;
1481 fragmentPos+= fragmentLengthB;
1488 memcpy(funnyCode + fragmentPos, fragmentA, fragmentLengthA);
1490 funnyCode[fragmentPos + imm8OfPShufW1A]=
1491 funnyCode[fragmentPos + imm8OfPShufW2A]=
1492 a | (b<<2) | (c<<4) | (d<<6);
1494 if (i+4>=dstW) shift=maxShift; //avoid overread
1495 else if ((filterPos[i/2]&3) <= maxShift) shift=filterPos[i/2]&3; //partial align
1497 if (shift && i>=shift)
1499 funnyCode[fragmentPos + imm8OfPShufW1A]+= 0x55*shift;
1500 funnyCode[fragmentPos + imm8OfPShufW2A]+= 0x55*shift;
1501 filterPos[i/2]-=shift;
1504 fragmentPos+= fragmentLengthA;
1507 funnyCode[fragmentPos]= RET;
1511 filterPos[i/2]= xpos>>16; // needed to jump to the next part
1513 #endif /* COMPILE_MMX2 */
1515 static void globalInit(void){
1516 // generating tables:
1518 for (i=0; i<768; i++){
1519 int c= av_clip_uint8(i-256);
1524 static SwsFunc getSwsFunc(int flags){
1526 #if defined(RUNTIME_CPUDETECT) && defined (CONFIG_GPL)
1527 #if defined(ARCH_X86)
1528 // ordered per speed fastest first
1529 if (flags & SWS_CPU_CAPS_MMX2)
1530 return swScale_MMX2;
1531 else if (flags & SWS_CPU_CAPS_3DNOW)
1532 return swScale_3DNow;
1533 else if (flags & SWS_CPU_CAPS_MMX)
1540 if (flags & SWS_CPU_CAPS_ALTIVEC)
1541 return swScale_altivec;
1546 #endif /* defined(ARCH_X86) */
1547 #else //RUNTIME_CPUDETECT
1549 return swScale_MMX2;
1550 #elif defined (HAVE_3DNOW)
1551 return swScale_3DNow;
1552 #elif defined (HAVE_MMX)
1554 #elif defined (HAVE_ALTIVEC)
1555 return swScale_altivec;
1559 #endif //!RUNTIME_CPUDETECT
1562 static int PlanarToNV12Wrapper(SwsContext *c, uint8_t* src[], int srcStride[], int srcSliceY,
1563 int srcSliceH, uint8_t* dstParam[], int dstStride[]){
1564 uint8_t *dst=dstParam[0] + dstStride[0]*srcSliceY;
1566 if (dstStride[0]==srcStride[0] && srcStride[0] > 0)
1567 memcpy(dst, src[0], srcSliceH*dstStride[0]);
1571 uint8_t *srcPtr= src[0];
1572 uint8_t *dstPtr= dst;
1573 for (i=0; i<srcSliceH; i++)
1575 memcpy(dstPtr, srcPtr, c->srcW);
1576 srcPtr+= srcStride[0];
1577 dstPtr+= dstStride[0];
1580 dst = dstParam[1] + dstStride[1]*srcSliceY/2;
1581 if (c->dstFormat == PIX_FMT_NV12)
1582 interleaveBytes(src[1], src[2], dst, c->srcW/2, srcSliceH/2, srcStride[1], srcStride[2], dstStride[0]);
1584 interleaveBytes(src[2], src[1], dst, c->srcW/2, srcSliceH/2, srcStride[2], srcStride[1], dstStride[0]);
1589 static int PlanarToYuy2Wrapper(SwsContext *c, uint8_t* src[], int srcStride[], int srcSliceY,
1590 int srcSliceH, uint8_t* dstParam[], int dstStride[]){
1591 uint8_t *dst=dstParam[0] + dstStride[0]*srcSliceY;
1593 yv12toyuy2(src[0], src[1], src[2], dst, c->srcW, srcSliceH, srcStride[0], srcStride[1], dstStride[0]);
1598 static int PlanarToUyvyWrapper(SwsContext *c, uint8_t* src[], int srcStride[], int srcSliceY,
1599 int srcSliceH, uint8_t* dstParam[], int dstStride[]){
1600 uint8_t *dst=dstParam[0] + dstStride[0]*srcSliceY;
1602 yv12touyvy(src[0], src[1], src[2], dst, c->srcW, srcSliceH, srcStride[0], srcStride[1], dstStride[0]);
1607 /* {RGB,BGR}{15,16,24,32} -> {RGB,BGR}{15,16,24,32} */
1608 static int rgb2rgbWrapper(SwsContext *c, uint8_t* src[], int srcStride[], int srcSliceY,
1609 int srcSliceH, uint8_t* dst[], int dstStride[]){
1610 const int srcFormat= c->srcFormat;
1611 const int dstFormat= c->dstFormat;
1612 const int srcBpp= (fmt_depth(srcFormat) + 7) >> 3;
1613 const int dstBpp= (fmt_depth(dstFormat) + 7) >> 3;
1614 const int srcId= fmt_depth(srcFormat) >> 2; /* 1:0, 4:1, 8:2, 15:3, 16:4, 24:6, 32:8 */
1615 const int dstId= fmt_depth(dstFormat) >> 2;
1616 void (*conv)(const uint8_t *src, uint8_t *dst, long src_size)=NULL;
1619 if ( (isBGR(srcFormat) && isBGR(dstFormat))
1620 || (isRGB(srcFormat) && isRGB(dstFormat))){
1621 switch(srcId | (dstId<<4)){
1622 case 0x34: conv= rgb16to15; break;
1623 case 0x36: conv= rgb24to15; break;
1624 case 0x38: conv= rgb32to15; break;
1625 case 0x43: conv= rgb15to16; break;
1626 case 0x46: conv= rgb24to16; break;
1627 case 0x48: conv= rgb32to16; break;
1628 case 0x63: conv= rgb15to24; break;
1629 case 0x64: conv= rgb16to24; break;
1630 case 0x68: conv= rgb32to24; break;
1631 case 0x83: conv= rgb15to32; break;
1632 case 0x84: conv= rgb16to32; break;
1633 case 0x86: conv= rgb24to32; break;
1634 default: av_log(c, AV_LOG_ERROR, "internal error %s -> %s converter\n",
1635 sws_format_name(srcFormat), sws_format_name(dstFormat)); break;
1637 }else if ( (isBGR(srcFormat) && isRGB(dstFormat))
1638 || (isRGB(srcFormat) && isBGR(dstFormat))){
1639 switch(srcId | (dstId<<4)){
1640 case 0x33: conv= rgb15tobgr15; break;
1641 case 0x34: conv= rgb16tobgr15; break;
1642 case 0x36: conv= rgb24tobgr15; break;
1643 case 0x38: conv= rgb32tobgr15; break;
1644 case 0x43: conv= rgb15tobgr16; break;
1645 case 0x44: conv= rgb16tobgr16; break;
1646 case 0x46: conv= rgb24tobgr16; break;
1647 case 0x48: conv= rgb32tobgr16; break;
1648 case 0x63: conv= rgb15tobgr24; break;
1649 case 0x64: conv= rgb16tobgr24; break;
1650 case 0x66: conv= rgb24tobgr24; break;
1651 case 0x68: conv= rgb32tobgr24; break;
1652 case 0x83: conv= rgb15tobgr32; break;
1653 case 0x84: conv= rgb16tobgr32; break;
1654 case 0x86: conv= rgb24tobgr32; break;
1655 case 0x88: conv= rgb32tobgr32; break;
1656 default: av_log(c, AV_LOG_ERROR, "internal error %s -> %s converter\n",
1657 sws_format_name(srcFormat), sws_format_name(dstFormat)); break;
1660 av_log(c, AV_LOG_ERROR, "internal error %s -> %s converter\n",
1661 sws_format_name(srcFormat), sws_format_name(dstFormat));
1666 if (dstStride[0]*srcBpp == srcStride[0]*dstBpp && srcStride[0] > 0)
1667 conv(src[0], dst[0] + dstStride[0]*srcSliceY, srcSliceH*srcStride[0]);
1671 uint8_t *srcPtr= src[0];
1672 uint8_t *dstPtr= dst[0] + dstStride[0]*srcSliceY;
1674 for (i=0; i<srcSliceH; i++)
1676 conv(srcPtr, dstPtr, c->srcW*srcBpp);
1677 srcPtr+= srcStride[0];
1678 dstPtr+= dstStride[0];
1685 static int bgr24toyv12Wrapper(SwsContext *c, uint8_t* src[], int srcStride[], int srcSliceY,
1686 int srcSliceH, uint8_t* dst[], int dstStride[]){
1690 dst[0]+ srcSliceY *dstStride[0],
1691 dst[1]+(srcSliceY>>1)*dstStride[1],
1692 dst[2]+(srcSliceY>>1)*dstStride[2],
1694 dstStride[0], dstStride[1], srcStride[0]);
1698 static int yvu9toyv12Wrapper(SwsContext *c, uint8_t* src[], int srcStride[], int srcSliceY,
1699 int srcSliceH, uint8_t* dst[], int dstStride[]){
1703 if (srcStride[0]==dstStride[0] && srcStride[0] > 0)
1704 memcpy(dst[0]+ srcSliceY*dstStride[0], src[0], srcStride[0]*srcSliceH);
1706 uint8_t *srcPtr= src[0];
1707 uint8_t *dstPtr= dst[0] + dstStride[0]*srcSliceY;
1709 for (i=0; i<srcSliceH; i++)
1711 memcpy(dstPtr, srcPtr, c->srcW);
1712 srcPtr+= srcStride[0];
1713 dstPtr+= dstStride[0];
1717 if (c->dstFormat==PIX_FMT_YUV420P){
1718 planar2x(src[1], dst[1], c->chrSrcW, c->chrSrcH, srcStride[1], dstStride[1]);
1719 planar2x(src[2], dst[2], c->chrSrcW, c->chrSrcH, srcStride[2], dstStride[2]);
1721 planar2x(src[1], dst[2], c->chrSrcW, c->chrSrcH, srcStride[1], dstStride[2]);
1722 planar2x(src[2], dst[1], c->chrSrcW, c->chrSrcH, srcStride[2], dstStride[1]);
1727 /* unscaled copy like stuff (assumes nearly identical formats) */
1728 static int packedCopy(SwsContext *c, uint8_t* src[], int srcStride[], int srcSliceY,
1729 int srcSliceH, uint8_t* dst[], int dstStride[])
1731 if (dstStride[0]==srcStride[0] && srcStride[0] > 0)
1732 memcpy(dst[0] + dstStride[0]*srcSliceY, src[0], srcSliceH*dstStride[0]);
1736 uint8_t *srcPtr= src[0];
1737 uint8_t *dstPtr= dst[0] + dstStride[0]*srcSliceY;
1740 /* universal length finder */
1741 while(length+c->srcW <= FFABS(dstStride[0])
1742 && length+c->srcW <= FFABS(srcStride[0])) length+= c->srcW;
1745 for (i=0; i<srcSliceH; i++)
1747 memcpy(dstPtr, srcPtr, length);
1748 srcPtr+= srcStride[0];
1749 dstPtr+= dstStride[0];
1755 static int planarCopy(SwsContext *c, uint8_t* src[], int srcStride[], int srcSliceY,
1756 int srcSliceH, uint8_t* dst[], int dstStride[])
1759 for (plane=0; plane<3; plane++)
1761 int length= plane==0 ? c->srcW : -((-c->srcW )>>c->chrDstHSubSample);
1762 int y= plane==0 ? srcSliceY: -((-srcSliceY)>>c->chrDstVSubSample);
1763 int height= plane==0 ? srcSliceH: -((-srcSliceH)>>c->chrDstVSubSample);
1765 if ((isGray(c->srcFormat) || isGray(c->dstFormat)) && plane>0)
1767 if (!isGray(c->dstFormat))
1768 memset(dst[plane], 128, dstStride[plane]*height);
1772 if (dstStride[plane]==srcStride[plane] && srcStride[plane] > 0)
1773 memcpy(dst[plane] + dstStride[plane]*y, src[plane], height*dstStride[plane]);
1777 uint8_t *srcPtr= src[plane];
1778 uint8_t *dstPtr= dst[plane] + dstStride[plane]*y;
1779 for (i=0; i<height; i++)
1781 memcpy(dstPtr, srcPtr, length);
1782 srcPtr+= srcStride[plane];
1783 dstPtr+= dstStride[plane];
1791 static int gray16togray(SwsContext *c, uint8_t* src[], int srcStride[], int srcSliceY,
1792 int srcSliceH, uint8_t* dst[], int dstStride[]){
1794 int length= c->srcW;
1796 int height= srcSliceH;
1798 uint8_t *srcPtr= src[0];
1799 uint8_t *dstPtr= dst[0] + dstStride[0]*y;
1801 if (!isGray(c->dstFormat)){
1802 int height= -((-srcSliceH)>>c->chrDstVSubSample);
1803 memset(dst[1], 128, dstStride[1]*height);
1804 memset(dst[2], 128, dstStride[2]*height);
1806 if (c->srcFormat == PIX_FMT_GRAY16LE) srcPtr++;
1807 for (i=0; i<height; i++)
1809 for (j=0; j<length; j++) dstPtr[j] = srcPtr[j<<1];
1810 srcPtr+= srcStride[0];
1811 dstPtr+= dstStride[0];
1816 static int graytogray16(SwsContext *c, uint8_t* src[], int srcStride[], int srcSliceY,
1817 int srcSliceH, uint8_t* dst[], int dstStride[]){
1819 int length= c->srcW;
1821 int height= srcSliceH;
1823 uint8_t *srcPtr= src[0];
1824 uint8_t *dstPtr= dst[0] + dstStride[0]*y;
1825 for (i=0; i<height; i++)
1827 for (j=0; j<length; j++)
1829 dstPtr[j<<1] = srcPtr[j];
1830 dstPtr[(j<<1)+1] = srcPtr[j];
1832 srcPtr+= srcStride[0];
1833 dstPtr+= dstStride[0];
1838 static int gray16swap(SwsContext *c, uint8_t* src[], int srcStride[], int srcSliceY,
1839 int srcSliceH, uint8_t* dst[], int dstStride[]){
1841 int length= c->srcW;
1843 int height= srcSliceH;
1845 uint16_t *srcPtr= (uint16_t*)src[0];
1846 uint16_t *dstPtr= (uint16_t*)(dst[0] + dstStride[0]*y/2);
1847 for (i=0; i<height; i++)
1849 for (j=0; j<length; j++) dstPtr[j] = bswap_16(srcPtr[j]);
1850 srcPtr+= srcStride[0]/2;
1851 dstPtr+= dstStride[0]/2;
1857 static void getSubSampleFactors(int *h, int *v, int format){
1859 case PIX_FMT_UYVY422:
1860 case PIX_FMT_YUYV422:
1864 case PIX_FMT_YUV420P:
1865 case PIX_FMT_YUVA420P:
1866 case PIX_FMT_GRAY16BE:
1867 case PIX_FMT_GRAY16LE:
1868 case PIX_FMT_GRAY8: //FIXME remove after different subsamplings are fully implemented
1874 case PIX_FMT_YUV440P:
1878 case PIX_FMT_YUV410P:
1882 case PIX_FMT_YUV444P:
1886 case PIX_FMT_YUV422P:
1890 case PIX_FMT_YUV411P:
1901 static uint16_t roundToInt16(int64_t f){
1902 int r= (f + (1<<15))>>16;
1903 if (r<-0x7FFF) return 0x8000;
1904 else if (r> 0x7FFF) return 0x7FFF;
1909 * @param inv_table the yuv2rgb coeffs, normally Inverse_Table_6_9[x]
1910 * @param fullRange if 1 then the luma range is 0..255 if 0 it is 16..235
1911 * @return -1 if not supported
1913 int sws_setColorspaceDetails(SwsContext *c, const int inv_table[4], int srcRange, const int table[4], int dstRange, int brightness, int contrast, int saturation){
1914 int64_t crv = inv_table[0];
1915 int64_t cbu = inv_table[1];
1916 int64_t cgu = -inv_table[2];
1917 int64_t cgv = -inv_table[3];
1921 if (isYUV(c->dstFormat) || isGray(c->dstFormat)) return -1;
1922 memcpy(c->srcColorspaceTable, inv_table, sizeof(int)*4);
1923 memcpy(c->dstColorspaceTable, table, sizeof(int)*4);
1925 c->brightness= brightness;
1926 c->contrast = contrast;
1927 c->saturation= saturation;
1928 c->srcRange = srcRange;
1929 c->dstRange = dstRange;
1931 c->uOffset= 0x0400040004000400LL;
1932 c->vOffset= 0x0400040004000400LL;
1938 crv= (crv*224) / 255;
1939 cbu= (cbu*224) / 255;
1940 cgu= (cgu*224) / 255;
1941 cgv= (cgv*224) / 255;
1944 cy = (cy *contrast )>>16;
1945 crv= (crv*contrast * saturation)>>32;
1946 cbu= (cbu*contrast * saturation)>>32;
1947 cgu= (cgu*contrast * saturation)>>32;
1948 cgv= (cgv*contrast * saturation)>>32;
1950 oy -= 256*brightness;
1952 c->yCoeff= roundToInt16(cy *8192) * 0x0001000100010001ULL;
1953 c->vrCoeff= roundToInt16(crv*8192) * 0x0001000100010001ULL;
1954 c->ubCoeff= roundToInt16(cbu*8192) * 0x0001000100010001ULL;
1955 c->vgCoeff= roundToInt16(cgv*8192) * 0x0001000100010001ULL;
1956 c->ugCoeff= roundToInt16(cgu*8192) * 0x0001000100010001ULL;
1957 c->yOffset= roundToInt16(oy * 8) * 0x0001000100010001ULL;
1959 yuv2rgb_c_init_tables(c, inv_table, srcRange, brightness, contrast, saturation);
1962 #ifdef COMPILE_ALTIVEC
1963 if (c->flags & SWS_CPU_CAPS_ALTIVEC)
1964 yuv2rgb_altivec_init_tables (c, inv_table, brightness, contrast, saturation);
1970 * @return -1 if not supported
1972 int sws_getColorspaceDetails(SwsContext *c, int **inv_table, int *srcRange, int **table, int *dstRange, int *brightness, int *contrast, int *saturation){
1973 if (isYUV(c->dstFormat) || isGray(c->dstFormat)) return -1;
1975 *inv_table = c->srcColorspaceTable;
1976 *table = c->dstColorspaceTable;
1977 *srcRange = c->srcRange;
1978 *dstRange = c->dstRange;
1979 *brightness= c->brightness;
1980 *contrast = c->contrast;
1981 *saturation= c->saturation;
1986 static int handle_jpeg(int *format)
1989 case PIX_FMT_YUVJ420P:
1990 *format = PIX_FMT_YUV420P;
1992 case PIX_FMT_YUVJ422P:
1993 *format = PIX_FMT_YUV422P;
1995 case PIX_FMT_YUVJ444P:
1996 *format = PIX_FMT_YUV444P;
1998 case PIX_FMT_YUVJ440P:
1999 *format = PIX_FMT_YUV440P;
2006 SwsContext *sws_getContext(int srcW, int srcH, int srcFormat, int dstW, int dstH, int dstFormat, int flags,
2007 SwsFilter *srcFilter, SwsFilter *dstFilter, double *param){
2011 int usesVFilter, usesHFilter;
2012 int unscaled, needsDither;
2013 int srcRange, dstRange;
2014 SwsFilter dummyFilter= {NULL, NULL, NULL, NULL};
2015 #if defined(ARCH_X86)
2016 if (flags & SWS_CPU_CAPS_MMX)
2017 asm volatile("emms\n\t"::: "memory");
2020 #if !defined(RUNTIME_CPUDETECT) || !defined (CONFIG_GPL) //ensure that the flags match the compiled variant if cpudetect is off
2021 flags &= ~(SWS_CPU_CAPS_MMX|SWS_CPU_CAPS_MMX2|SWS_CPU_CAPS_3DNOW|SWS_CPU_CAPS_ALTIVEC|SWS_CPU_CAPS_BFIN);
2023 flags |= SWS_CPU_CAPS_MMX|SWS_CPU_CAPS_MMX2;
2024 #elif defined (HAVE_3DNOW)
2025 flags |= SWS_CPU_CAPS_MMX|SWS_CPU_CAPS_3DNOW;
2026 #elif defined (HAVE_MMX)
2027 flags |= SWS_CPU_CAPS_MMX;
2028 #elif defined (HAVE_ALTIVEC)
2029 flags |= SWS_CPU_CAPS_ALTIVEC;
2030 #elif defined (ARCH_BFIN)
2031 flags |= SWS_CPU_CAPS_BFIN;
2033 #endif /* RUNTIME_CPUDETECT */
2034 if (clip_table[512] != 255) globalInit();
2035 if (!rgb15to16) sws_rgb2rgb_init(flags);
2037 unscaled = (srcW == dstW && srcH == dstH);
2038 needsDither= (isBGR(dstFormat) || isRGB(dstFormat))
2039 && (fmt_depth(dstFormat))<24
2040 && ((fmt_depth(dstFormat))<(fmt_depth(srcFormat)) || (!(isRGB(srcFormat) || isBGR(srcFormat))));
2042 srcRange = handle_jpeg(&srcFormat);
2043 dstRange = handle_jpeg(&dstFormat);
2045 if (!isSupportedIn(srcFormat))
2047 av_log(NULL, AV_LOG_ERROR, "swScaler: %s is not supported as input pixel format\n", sws_format_name(srcFormat));
2050 if (!isSupportedOut(dstFormat))
2052 av_log(NULL, AV_LOG_ERROR, "swScaler: %s is not supported as output pixel format\n", sws_format_name(dstFormat));
2057 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
2059 av_log(NULL, AV_LOG_ERROR, "swScaler: %dx%d -> %dx%d is invalid scaling dimension\n",
2060 srcW, srcH, dstW, dstH);
2063 if(srcW > VOFW || dstW > VOFW){
2064 av_log(NULL, AV_LOG_ERROR, "swScaler: Compile time max width is "AV_STRINGIFY(VOFW)" change VOF/VOFW and recompile\n");
2068 if (!dstFilter) dstFilter= &dummyFilter;
2069 if (!srcFilter) srcFilter= &dummyFilter;
2071 c= av_mallocz(sizeof(SwsContext));
2073 c->av_class = &sws_context_class;
2078 c->lumXInc= ((srcW<<16) + (dstW>>1))/dstW;
2079 c->lumYInc= ((srcH<<16) + (dstH>>1))/dstH;
2081 c->dstFormat= dstFormat;
2082 c->srcFormat= srcFormat;
2083 c->vRounder= 4* 0x0001000100010001ULL;
2085 usesHFilter= usesVFilter= 0;
2086 if (dstFilter->lumV && dstFilter->lumV->length>1) usesVFilter=1;
2087 if (dstFilter->lumH && dstFilter->lumH->length>1) usesHFilter=1;
2088 if (dstFilter->chrV && dstFilter->chrV->length>1) usesVFilter=1;
2089 if (dstFilter->chrH && dstFilter->chrH->length>1) usesHFilter=1;
2090 if (srcFilter->lumV && srcFilter->lumV->length>1) usesVFilter=1;
2091 if (srcFilter->lumH && srcFilter->lumH->length>1) usesHFilter=1;
2092 if (srcFilter->chrV && srcFilter->chrV->length>1) usesVFilter=1;
2093 if (srcFilter->chrH && srcFilter->chrH->length>1) usesHFilter=1;
2095 getSubSampleFactors(&c->chrSrcHSubSample, &c->chrSrcVSubSample, srcFormat);
2096 getSubSampleFactors(&c->chrDstHSubSample, &c->chrDstVSubSample, dstFormat);
2098 // reuse chroma for 2 pixles rgb/bgr unless user wants full chroma interpolation
2099 if ((isBGR(dstFormat) || isRGB(dstFormat)) && !(flags&SWS_FULL_CHR_H_INT)) c->chrDstHSubSample=1;
2101 // drop some chroma lines if the user wants it
2102 c->vChrDrop= (flags&SWS_SRC_V_CHR_DROP_MASK)>>SWS_SRC_V_CHR_DROP_SHIFT;
2103 c->chrSrcVSubSample+= c->vChrDrop;
2105 // drop every 2. pixel for chroma calculation unless user wants full chroma
2106 if ((isBGR(srcFormat) || isRGB(srcFormat)) && !(flags&SWS_FULL_CHR_H_INP)
2107 && srcFormat!=PIX_FMT_RGB8 && srcFormat!=PIX_FMT_BGR8
2108 && srcFormat!=PIX_FMT_RGB4 && srcFormat!=PIX_FMT_BGR4
2109 && srcFormat!=PIX_FMT_RGB4_BYTE && srcFormat!=PIX_FMT_BGR4_BYTE)
2110 c->chrSrcHSubSample=1;
2113 c->param[0] = param[0];
2114 c->param[1] = param[1];
2117 c->param[1] = SWS_PARAM_DEFAULT;
2120 c->chrIntHSubSample= c->chrDstHSubSample;
2121 c->chrIntVSubSample= c->chrSrcVSubSample;
2123 // Note the -((-x)>>y) is so that we always round toward +inf.
2124 c->chrSrcW= -((-srcW) >> c->chrSrcHSubSample);
2125 c->chrSrcH= -((-srcH) >> c->chrSrcVSubSample);
2126 c->chrDstW= -((-dstW) >> c->chrDstHSubSample);
2127 c->chrDstH= -((-dstH) >> c->chrDstVSubSample);
2129 sws_setColorspaceDetails(c, Inverse_Table_6_9[SWS_CS_DEFAULT], srcRange, Inverse_Table_6_9[SWS_CS_DEFAULT] /* FIXME*/, dstRange, 0, 1<<16, 1<<16);
2131 /* unscaled special Cases */
2132 if (unscaled && !usesHFilter && !usesVFilter)
2135 if (srcFormat == PIX_FMT_YUV420P && (dstFormat == PIX_FMT_NV12 || dstFormat == PIX_FMT_NV21))
2137 c->swScale= PlanarToNV12Wrapper;
2141 if ((srcFormat==PIX_FMT_YUV420P || srcFormat==PIX_FMT_YUV422P) && (isBGR(dstFormat) || isRGB(dstFormat)))
2143 c->swScale= yuv2rgb_get_func_ptr(c);
2147 if (srcFormat==PIX_FMT_YUV410P && dstFormat==PIX_FMT_YUV420P)
2149 c->swScale= yvu9toyv12Wrapper;
2153 if (srcFormat==PIX_FMT_BGR24 && dstFormat==PIX_FMT_YUV420P)
2154 c->swScale= bgr24toyv12Wrapper;
2156 /* rgb/bgr -> rgb/bgr (no dither needed forms) */
2157 if ( (isBGR(srcFormat) || isRGB(srcFormat))
2158 && (isBGR(dstFormat) || isRGB(dstFormat))
2159 && srcFormat != PIX_FMT_BGR8 && dstFormat != PIX_FMT_BGR8
2160 && srcFormat != PIX_FMT_RGB8 && dstFormat != PIX_FMT_RGB8
2161 && srcFormat != PIX_FMT_BGR4 && dstFormat != PIX_FMT_BGR4
2162 && srcFormat != PIX_FMT_RGB4 && dstFormat != PIX_FMT_RGB4
2163 && srcFormat != PIX_FMT_BGR4_BYTE && dstFormat != PIX_FMT_BGR4_BYTE
2164 && srcFormat != PIX_FMT_RGB4_BYTE && dstFormat != PIX_FMT_RGB4_BYTE
2165 && srcFormat != PIX_FMT_MONOBLACK && dstFormat != PIX_FMT_MONOBLACK
2167 c->swScale= rgb2rgbWrapper;
2169 /* LQ converters if -sws 0 or -sws 4*/
2170 if (c->flags&(SWS_FAST_BILINEAR|SWS_POINT)){
2171 /* rgb/bgr -> rgb/bgr (dither needed forms) */
2172 if ( (isBGR(srcFormat) || isRGB(srcFormat))
2173 && (isBGR(dstFormat) || isRGB(dstFormat))
2175 c->swScale= rgb2rgbWrapper;
2178 if (srcFormat == PIX_FMT_YUV420P &&
2179 (dstFormat == PIX_FMT_YUYV422 || dstFormat == PIX_FMT_UYVY422))
2181 if (dstFormat == PIX_FMT_YUYV422)
2182 c->swScale= PlanarToYuy2Wrapper;
2184 c->swScale= PlanarToUyvyWrapper;
2188 #ifdef COMPILE_ALTIVEC
2189 if ((c->flags & SWS_CPU_CAPS_ALTIVEC) &&
2190 ((srcFormat == PIX_FMT_YUV420P &&
2191 (dstFormat == PIX_FMT_YUYV422 || dstFormat == PIX_FMT_UYVY422)))) {
2192 // unscaled YV12 -> packed YUV, we want speed
2193 if (dstFormat == PIX_FMT_YUYV422)
2194 c->swScale= yv12toyuy2_unscaled_altivec;
2196 c->swScale= yv12touyvy_unscaled_altivec;
2201 if ( srcFormat == dstFormat
2202 || (isPlanarYUV(srcFormat) && isGray(dstFormat))
2203 || (isPlanarYUV(dstFormat) && isGray(srcFormat)))
2205 if (isPacked(c->srcFormat))
2206 c->swScale= packedCopy;
2207 else /* Planar YUV or gray */
2208 c->swScale= planarCopy;
2211 /* gray16{le,be} conversions */
2212 if (isGray16(srcFormat) && (isPlanarYUV(dstFormat) || (dstFormat == PIX_FMT_GRAY8)))
2214 c->swScale= gray16togray;
2216 if ((isPlanarYUV(srcFormat) || (srcFormat == PIX_FMT_GRAY8)) && isGray16(dstFormat))
2218 c->swScale= graytogray16;
2220 if (srcFormat != dstFormat && isGray16(srcFormat) && isGray16(dstFormat))
2222 c->swScale= gray16swap;
2226 if (flags & SWS_CPU_CAPS_BFIN)
2227 ff_bfin_get_unscaled_swscale (c);
2231 if (flags&SWS_PRINT_INFO)
2232 av_log(c, AV_LOG_INFO, "using unscaled %s -> %s special converter\n",
2233 sws_format_name(srcFormat), sws_format_name(dstFormat));
2238 if (flags & SWS_CPU_CAPS_MMX2)
2240 c->canMMX2BeUsed= (dstW >=srcW && (dstW&31)==0 && (srcW&15)==0) ? 1 : 0;
2241 if (!c->canMMX2BeUsed && dstW >=srcW && (srcW&15)==0 && (flags&SWS_FAST_BILINEAR))
2243 if (flags&SWS_PRINT_INFO)
2244 av_log(c, AV_LOG_INFO, "output Width is not a multiple of 32 -> no MMX2 scaler\n");
2246 if (usesHFilter) c->canMMX2BeUsed=0;
2251 c->chrXInc= ((c->chrSrcW<<16) + (c->chrDstW>>1))/c->chrDstW;
2252 c->chrYInc= ((c->chrSrcH<<16) + (c->chrDstH>>1))/c->chrDstH;
2254 // match pixel 0 of the src to pixel 0 of dst and match pixel n-2 of src to pixel n-2 of dst
2255 // but only for the FAST_BILINEAR mode otherwise do correct scaling
2256 // n-2 is the last chrominance sample available
2257 // this is not perfect, but no one should notice the difference, the more correct variant
2258 // would be like the vertical one, but that would require some special code for the
2259 // first and last pixel
2260 if (flags&SWS_FAST_BILINEAR)
2262 if (c->canMMX2BeUsed)
2267 //we don't use the x86asm scaler if mmx is available
2268 else if (flags & SWS_CPU_CAPS_MMX)
2270 c->lumXInc = ((srcW-2)<<16)/(dstW-2) - 20;
2271 c->chrXInc = ((c->chrSrcW-2)<<16)/(c->chrDstW-2) - 20;
2275 /* precalculate horizontal scaler filter coefficients */
2277 const int filterAlign=
2278 (flags & SWS_CPU_CAPS_MMX) ? 4 :
2279 (flags & SWS_CPU_CAPS_ALTIVEC) ? 8 :
2282 initFilter(&c->hLumFilter, &c->hLumFilterPos, &c->hLumFilterSize, c->lumXInc,
2283 srcW , dstW, filterAlign, 1<<14,
2284 (flags&SWS_BICUBLIN) ? (flags|SWS_BICUBIC) : flags,
2285 srcFilter->lumH, dstFilter->lumH, c->param);
2286 initFilter(&c->hChrFilter, &c->hChrFilterPos, &c->hChrFilterSize, c->chrXInc,
2287 c->chrSrcW, c->chrDstW, filterAlign, 1<<14,
2288 (flags&SWS_BICUBLIN) ? (flags|SWS_BILINEAR) : flags,
2289 srcFilter->chrH, dstFilter->chrH, c->param);
2291 #define MAX_FUNNY_CODE_SIZE 10000
2292 #if defined(COMPILE_MMX2)
2293 // can't downscale !!!
2294 if (c->canMMX2BeUsed && (flags & SWS_FAST_BILINEAR))
2296 #ifdef MAP_ANONYMOUS
2297 c->funnyYCode = (uint8_t*)mmap(NULL, MAX_FUNNY_CODE_SIZE, PROT_EXEC | PROT_READ | PROT_WRITE, MAP_PRIVATE | MAP_ANONYMOUS, 0, 0);
2298 c->funnyUVCode = (uint8_t*)mmap(NULL, MAX_FUNNY_CODE_SIZE, PROT_EXEC | PROT_READ | PROT_WRITE, MAP_PRIVATE | MAP_ANONYMOUS, 0, 0);
2300 c->funnyYCode = av_malloc(MAX_FUNNY_CODE_SIZE);
2301 c->funnyUVCode = av_malloc(MAX_FUNNY_CODE_SIZE);
2304 c->lumMmx2Filter = av_malloc((dstW /8+8)*sizeof(int16_t));
2305 c->chrMmx2Filter = av_malloc((c->chrDstW /4+8)*sizeof(int16_t));
2306 c->lumMmx2FilterPos= av_malloc((dstW /2/8+8)*sizeof(int32_t));
2307 c->chrMmx2FilterPos= av_malloc((c->chrDstW/2/4+8)*sizeof(int32_t));
2309 initMMX2HScaler( dstW, c->lumXInc, c->funnyYCode , c->lumMmx2Filter, c->lumMmx2FilterPos, 8);
2310 initMMX2HScaler(c->chrDstW, c->chrXInc, c->funnyUVCode, c->chrMmx2Filter, c->chrMmx2FilterPos, 4);
2312 #endif /* defined(COMPILE_MMX2) */
2313 } // Init Horizontal stuff
2317 /* precalculate vertical scaler filter coefficients */
2319 const int filterAlign=
2320 (flags & SWS_CPU_CAPS_MMX) && (flags & SWS_ACCURATE_RND) ? 2 :
2321 (flags & SWS_CPU_CAPS_ALTIVEC) ? 8 :
2324 initFilter(&c->vLumFilter, &c->vLumFilterPos, &c->vLumFilterSize, c->lumYInc,
2325 srcH , dstH, filterAlign, (1<<12)-4,
2326 (flags&SWS_BICUBLIN) ? (flags|SWS_BICUBIC) : flags,
2327 srcFilter->lumV, dstFilter->lumV, c->param);
2328 initFilter(&c->vChrFilter, &c->vChrFilterPos, &c->vChrFilterSize, c->chrYInc,
2329 c->chrSrcH, c->chrDstH, filterAlign, (1<<12)-4,
2330 (flags&SWS_BICUBLIN) ? (flags|SWS_BILINEAR) : flags,
2331 srcFilter->chrV, dstFilter->chrV, c->param);
2334 c->vYCoeffsBank = av_malloc(sizeof (vector signed short)*c->vLumFilterSize*c->dstH);
2335 c->vCCoeffsBank = av_malloc(sizeof (vector signed short)*c->vChrFilterSize*c->chrDstH);
2337 for (i=0;i<c->vLumFilterSize*c->dstH;i++) {
2339 short *p = (short *)&c->vYCoeffsBank[i];
2341 p[j] = c->vLumFilter[i];
2344 for (i=0;i<c->vChrFilterSize*c->chrDstH;i++) {
2346 short *p = (short *)&c->vCCoeffsBank[i];
2348 p[j] = c->vChrFilter[i];
2353 // Calculate Buffer Sizes so that they won't run out while handling these damn slices
2354 c->vLumBufSize= c->vLumFilterSize;
2355 c->vChrBufSize= c->vChrFilterSize;
2356 for (i=0; i<dstH; i++)
2358 int chrI= i*c->chrDstH / dstH;
2359 int nextSlice= FFMAX(c->vLumFilterPos[i ] + c->vLumFilterSize - 1,
2360 ((c->vChrFilterPos[chrI] + c->vChrFilterSize - 1)<<c->chrSrcVSubSample));
2362 nextSlice>>= c->chrSrcVSubSample;
2363 nextSlice<<= c->chrSrcVSubSample;
2364 if (c->vLumFilterPos[i ] + c->vLumBufSize < nextSlice)
2365 c->vLumBufSize= nextSlice - c->vLumFilterPos[i];
2366 if (c->vChrFilterPos[chrI] + c->vChrBufSize < (nextSlice>>c->chrSrcVSubSample))
2367 c->vChrBufSize= (nextSlice>>c->chrSrcVSubSample) - c->vChrFilterPos[chrI];
2370 // allocate pixbufs (we use dynamic allocation because otherwise we would need to
2371 c->lumPixBuf= av_malloc(c->vLumBufSize*2*sizeof(int16_t*));
2372 c->chrPixBuf= av_malloc(c->vChrBufSize*2*sizeof(int16_t*));
2373 //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)
2374 /* align at 16 bytes for AltiVec */
2375 for (i=0; i<c->vLumBufSize; i++)
2376 c->lumPixBuf[i]= c->lumPixBuf[i+c->vLumBufSize]= av_mallocz(VOF+1);
2377 for (i=0; i<c->vChrBufSize; i++)
2378 c->chrPixBuf[i]= c->chrPixBuf[i+c->vChrBufSize]= av_malloc((VOF+1)*2);
2380 //try to avoid drawing green stuff between the right end and the stride end
2381 for (i=0; i<c->vChrBufSize; i++) memset(c->chrPixBuf[i], 64, (VOF+1)*2);
2383 assert(2*VOFW == VOF);
2385 assert(c->chrDstH <= dstH);
2387 if (flags&SWS_PRINT_INFO)
2390 const char *dither= " dithered";
2392 const char *dither= "";
2394 if (flags&SWS_FAST_BILINEAR)
2395 av_log(c, AV_LOG_INFO, "FAST_BILINEAR scaler, ");
2396 else if (flags&SWS_BILINEAR)
2397 av_log(c, AV_LOG_INFO, "BILINEAR scaler, ");
2398 else if (flags&SWS_BICUBIC)
2399 av_log(c, AV_LOG_INFO, "BICUBIC scaler, ");
2400 else if (flags&SWS_X)
2401 av_log(c, AV_LOG_INFO, "Experimental scaler, ");
2402 else if (flags&SWS_POINT)
2403 av_log(c, AV_LOG_INFO, "Nearest Neighbor / POINT scaler, ");
2404 else if (flags&SWS_AREA)
2405 av_log(c, AV_LOG_INFO, "Area Averageing scaler, ");
2406 else if (flags&SWS_BICUBLIN)
2407 av_log(c, AV_LOG_INFO, "luma BICUBIC / chroma BILINEAR scaler, ");
2408 else if (flags&SWS_GAUSS)
2409 av_log(c, AV_LOG_INFO, "Gaussian scaler, ");
2410 else if (flags&SWS_SINC)
2411 av_log(c, AV_LOG_INFO, "Sinc scaler, ");
2412 else if (flags&SWS_LANCZOS)
2413 av_log(c, AV_LOG_INFO, "Lanczos scaler, ");
2414 else if (flags&SWS_SPLINE)
2415 av_log(c, AV_LOG_INFO, "Bicubic spline scaler, ");
2417 av_log(c, AV_LOG_INFO, "ehh flags invalid?! ");
2419 if (dstFormat==PIX_FMT_BGR555 || dstFormat==PIX_FMT_BGR565)
2420 av_log(c, AV_LOG_INFO, "from %s to%s %s ",
2421 sws_format_name(srcFormat), dither, sws_format_name(dstFormat));
2423 av_log(c, AV_LOG_INFO, "from %s to %s ",
2424 sws_format_name(srcFormat), sws_format_name(dstFormat));
2426 if (flags & SWS_CPU_CAPS_MMX2)
2427 av_log(c, AV_LOG_INFO, "using MMX2\n");
2428 else if (flags & SWS_CPU_CAPS_3DNOW)
2429 av_log(c, AV_LOG_INFO, "using 3DNOW\n");
2430 else if (flags & SWS_CPU_CAPS_MMX)
2431 av_log(c, AV_LOG_INFO, "using MMX\n");
2432 else if (flags & SWS_CPU_CAPS_ALTIVEC)
2433 av_log(c, AV_LOG_INFO, "using AltiVec\n");
2435 av_log(c, AV_LOG_INFO, "using C\n");
2438 if (flags & SWS_PRINT_INFO)
2440 if (flags & SWS_CPU_CAPS_MMX)
2442 if (c->canMMX2BeUsed && (flags&SWS_FAST_BILINEAR))
2443 av_log(c, AV_LOG_VERBOSE, "using FAST_BILINEAR MMX2 scaler for horizontal scaling\n");
2446 if (c->hLumFilterSize==4)
2447 av_log(c, AV_LOG_VERBOSE, "using 4-tap MMX scaler for horizontal luminance scaling\n");
2448 else if (c->hLumFilterSize==8)
2449 av_log(c, AV_LOG_VERBOSE, "using 8-tap MMX scaler for horizontal luminance scaling\n");
2451 av_log(c, AV_LOG_VERBOSE, "using n-tap MMX scaler for horizontal luminance scaling\n");
2453 if (c->hChrFilterSize==4)
2454 av_log(c, AV_LOG_VERBOSE, "using 4-tap MMX scaler for horizontal chrominance scaling\n");
2455 else if (c->hChrFilterSize==8)
2456 av_log(c, AV_LOG_VERBOSE, "using 8-tap MMX scaler for horizontal chrominance scaling\n");
2458 av_log(c, AV_LOG_VERBOSE, "using n-tap MMX scaler for horizontal chrominance scaling\n");
2463 #if defined(ARCH_X86)
2464 av_log(c, AV_LOG_VERBOSE, "using X86-Asm scaler for horizontal scaling\n");
2466 if (flags & SWS_FAST_BILINEAR)
2467 av_log(c, AV_LOG_VERBOSE, "using FAST_BILINEAR C scaler for horizontal scaling\n");
2469 av_log(c, AV_LOG_VERBOSE, "using C scaler for horizontal scaling\n");
2472 if (isPlanarYUV(dstFormat))
2474 if (c->vLumFilterSize==1)
2475 av_log(c, AV_LOG_VERBOSE, "using 1-tap %s \"scaler\" for vertical scaling (YV12 like)\n", (flags & SWS_CPU_CAPS_MMX) ? "MMX" : "C");
2477 av_log(c, AV_LOG_VERBOSE, "using n-tap %s scaler for vertical scaling (YV12 like)\n", (flags & SWS_CPU_CAPS_MMX) ? "MMX" : "C");
2481 if (c->vLumFilterSize==1 && c->vChrFilterSize==2)
2482 av_log(c, AV_LOG_VERBOSE, "using 1-tap %s \"scaler\" for vertical luminance scaling (BGR)\n"
2483 " 2-tap scaler for vertical chrominance scaling (BGR)\n", (flags & SWS_CPU_CAPS_MMX) ? "MMX" : "C");
2484 else if (c->vLumFilterSize==2 && c->vChrFilterSize==2)
2485 av_log(c, AV_LOG_VERBOSE, "using 2-tap linear %s scaler for vertical scaling (BGR)\n", (flags & SWS_CPU_CAPS_MMX) ? "MMX" : "C");
2487 av_log(c, AV_LOG_VERBOSE, "using n-tap %s scaler for vertical scaling (BGR)\n", (flags & SWS_CPU_CAPS_MMX) ? "MMX" : "C");
2490 if (dstFormat==PIX_FMT_BGR24)
2491 av_log(c, AV_LOG_VERBOSE, "using %s YV12->BGR24 Converter\n",
2492 (flags & SWS_CPU_CAPS_MMX2) ? "MMX2" : ((flags & SWS_CPU_CAPS_MMX) ? "MMX" : "C"));
2493 else if (dstFormat==PIX_FMT_RGB32)
2494 av_log(c, AV_LOG_VERBOSE, "using %s YV12->BGR32 Converter\n", (flags & SWS_CPU_CAPS_MMX) ? "MMX" : "C");
2495 else if (dstFormat==PIX_FMT_BGR565)
2496 av_log(c, AV_LOG_VERBOSE, "using %s YV12->BGR16 Converter\n", (flags & SWS_CPU_CAPS_MMX) ? "MMX" : "C");
2497 else if (dstFormat==PIX_FMT_BGR555)
2498 av_log(c, AV_LOG_VERBOSE, "using %s YV12->BGR15 Converter\n", (flags & SWS_CPU_CAPS_MMX) ? "MMX" : "C");
2500 av_log(c, AV_LOG_VERBOSE, "%dx%d -> %dx%d\n", srcW, srcH, dstW, dstH);
2502 if (flags & SWS_PRINT_INFO)
2504 av_log(c, AV_LOG_DEBUG, "Lum srcW=%d srcH=%d dstW=%d dstH=%d xInc=%d yInc=%d\n",
2505 c->srcW, c->srcH, c->dstW, c->dstH, c->lumXInc, c->lumYInc);
2506 av_log(c, AV_LOG_DEBUG, "Chr srcW=%d srcH=%d dstW=%d dstH=%d xInc=%d yInc=%d\n",
2507 c->chrSrcW, c->chrSrcH, c->chrDstW, c->chrDstH, c->chrXInc, c->chrYInc);
2510 c->swScale= getSwsFunc(flags);
2515 * swscale wrapper, so we don't need to export the SwsContext.
2516 * assumes planar YUV to be in YUV order instead of YVU
2518 int sws_scale(SwsContext *c, uint8_t* src[], int srcStride[], int srcSliceY,
2519 int srcSliceH, uint8_t* dst[], int dstStride[]){
2521 uint8_t* src2[4]= {src[0], src[1], src[2]};
2523 if (c->sliceDir == 0 && srcSliceY != 0 && srcSliceY + srcSliceH != c->srcH) {
2524 av_log(c, AV_LOG_ERROR, "Slices start in the middle!\n");
2527 if (c->sliceDir == 0) {
2528 if (srcSliceY == 0) c->sliceDir = 1; else c->sliceDir = -1;
2531 if (c->srcFormat == PIX_FMT_PAL8){
2532 for (i=0; i<256; i++){
2533 int p= ((uint32_t*)(src[1]))[i];
2534 int r= (p>>16)&0xFF;
2535 int g= (p>> 8)&0xFF;
2537 int y= av_clip_uint8(((RY*r + GY*g + BY*b)>>RGB2YUV_SHIFT) + 16 );
2538 int u= av_clip_uint8(((RU*r + GU*g + BU*b)>>RGB2YUV_SHIFT) + 128);
2539 int v= av_clip_uint8(((RV*r + GV*g + BV*b)>>RGB2YUV_SHIFT) + 128);
2540 pal[i]= y + (u<<8) + (v<<16);
2542 src2[1]= (uint8_t*)pal;
2545 // copy strides, so they can safely be modified
2546 if (c->sliceDir == 1) {
2547 // slices go from top to bottom
2548 int srcStride2[4]= {srcStride[0], srcStride[1], srcStride[2]};
2549 int dstStride2[4]= {dstStride[0], dstStride[1], dstStride[2]};
2550 return c->swScale(c, src2, srcStride2, srcSliceY, srcSliceH, dst, dstStride2);
2552 // slices go from bottom to top => we flip the image internally
2553 uint8_t* dst2[4]= {dst[0] + (c->dstH-1)*dstStride[0],
2554 dst[1] + ((c->dstH>>c->chrDstVSubSample)-1)*dstStride[1],
2555 dst[2] + ((c->dstH>>c->chrDstVSubSample)-1)*dstStride[2]};
2556 int srcStride2[4]= {-srcStride[0], -srcStride[1], -srcStride[2]};
2557 int dstStride2[4]= {-dstStride[0], -dstStride[1], -dstStride[2]};
2559 src2[0] += (srcSliceH-1)*srcStride[0];
2560 if (c->srcFormat != PIX_FMT_PAL8)
2561 src2[1] += ((srcSliceH>>c->chrSrcVSubSample)-1)*srcStride[1];
2562 src2[2] += ((srcSliceH>>c->chrSrcVSubSample)-1)*srcStride[2];
2564 return c->swScale(c, src2, srcStride2, c->srcH-srcSliceY-srcSliceH, srcSliceH, dst2, dstStride2);
2569 * swscale wrapper, so we don't need to export the SwsContext
2571 int sws_scale_ordered(SwsContext *c, uint8_t* src[], int srcStride[], int srcSliceY,
2572 int srcSliceH, uint8_t* dst[], int dstStride[]){
2573 return sws_scale(c, src, srcStride, srcSliceY, srcSliceH, dst, dstStride);
2576 SwsFilter *sws_getDefaultFilter(float lumaGBlur, float chromaGBlur,
2577 float lumaSharpen, float chromaSharpen,
2578 float chromaHShift, float chromaVShift,
2581 SwsFilter *filter= av_malloc(sizeof(SwsFilter));
2583 if (lumaGBlur!=0.0){
2584 filter->lumH= sws_getGaussianVec(lumaGBlur, 3.0);
2585 filter->lumV= sws_getGaussianVec(lumaGBlur, 3.0);
2587 filter->lumH= sws_getIdentityVec();
2588 filter->lumV= sws_getIdentityVec();
2591 if (chromaGBlur!=0.0){
2592 filter->chrH= sws_getGaussianVec(chromaGBlur, 3.0);
2593 filter->chrV= sws_getGaussianVec(chromaGBlur, 3.0);
2595 filter->chrH= sws_getIdentityVec();
2596 filter->chrV= sws_getIdentityVec();
2599 if (chromaSharpen!=0.0){
2600 SwsVector *id= sws_getIdentityVec();
2601 sws_scaleVec(filter->chrH, -chromaSharpen);
2602 sws_scaleVec(filter->chrV, -chromaSharpen);
2603 sws_addVec(filter->chrH, id);
2604 sws_addVec(filter->chrV, id);
2608 if (lumaSharpen!=0.0){
2609 SwsVector *id= sws_getIdentityVec();
2610 sws_scaleVec(filter->lumH, -lumaSharpen);
2611 sws_scaleVec(filter->lumV, -lumaSharpen);
2612 sws_addVec(filter->lumH, id);
2613 sws_addVec(filter->lumV, id);
2617 if (chromaHShift != 0.0)
2618 sws_shiftVec(filter->chrH, (int)(chromaHShift+0.5));
2620 if (chromaVShift != 0.0)
2621 sws_shiftVec(filter->chrV, (int)(chromaVShift+0.5));
2623 sws_normalizeVec(filter->chrH, 1.0);
2624 sws_normalizeVec(filter->chrV, 1.0);
2625 sws_normalizeVec(filter->lumH, 1.0);
2626 sws_normalizeVec(filter->lumV, 1.0);
2628 if (verbose) sws_printVec(filter->chrH);
2629 if (verbose) sws_printVec(filter->lumH);
2635 * returns a normalized gaussian curve used to filter stuff
2636 * quality=3 is high quality, lowwer is lowwer quality
2638 SwsVector *sws_getGaussianVec(double variance, double quality){
2639 const int length= (int)(variance*quality + 0.5) | 1;
2641 double *coeff= av_malloc(length*sizeof(double));
2642 double middle= (length-1)*0.5;
2643 SwsVector *vec= av_malloc(sizeof(SwsVector));
2646 vec->length= length;
2648 for (i=0; i<length; i++)
2650 double dist= i-middle;
2651 coeff[i]= exp(-dist*dist/(2*variance*variance)) / sqrt(2*variance*PI);
2654 sws_normalizeVec(vec, 1.0);
2659 SwsVector *sws_getConstVec(double c, int length){
2661 double *coeff= av_malloc(length*sizeof(double));
2662 SwsVector *vec= av_malloc(sizeof(SwsVector));
2665 vec->length= length;
2667 for (i=0; i<length; i++)
2674 SwsVector *sws_getIdentityVec(void){
2675 return sws_getConstVec(1.0, 1);
2678 double sws_dcVec(SwsVector *a){
2682 for (i=0; i<a->length; i++)
2688 void sws_scaleVec(SwsVector *a, double scalar){
2691 for (i=0; i<a->length; i++)
2692 a->coeff[i]*= scalar;
2695 void sws_normalizeVec(SwsVector *a, double height){
2696 sws_scaleVec(a, height/sws_dcVec(a));
2699 static SwsVector *sws_getConvVec(SwsVector *a, SwsVector *b){
2700 int length= a->length + b->length - 1;
2701 double *coeff= av_malloc(length*sizeof(double));
2703 SwsVector *vec= av_malloc(sizeof(SwsVector));
2706 vec->length= length;
2708 for (i=0; i<length; i++) coeff[i]= 0.0;
2710 for (i=0; i<a->length; i++)
2712 for (j=0; j<b->length; j++)
2714 coeff[i+j]+= a->coeff[i]*b->coeff[j];
2721 static SwsVector *sws_sumVec(SwsVector *a, SwsVector *b){
2722 int length= FFMAX(a->length, b->length);
2723 double *coeff= av_malloc(length*sizeof(double));
2725 SwsVector *vec= av_malloc(sizeof(SwsVector));
2728 vec->length= length;
2730 for (i=0; i<length; i++) coeff[i]= 0.0;
2732 for (i=0; i<a->length; i++) coeff[i + (length-1)/2 - (a->length-1)/2]+= a->coeff[i];
2733 for (i=0; i<b->length; i++) coeff[i + (length-1)/2 - (b->length-1)/2]+= b->coeff[i];
2738 static SwsVector *sws_diffVec(SwsVector *a, SwsVector *b){
2739 int length= FFMAX(a->length, b->length);
2740 double *coeff= av_malloc(length*sizeof(double));
2742 SwsVector *vec= av_malloc(sizeof(SwsVector));
2745 vec->length= length;
2747 for (i=0; i<length; i++) coeff[i]= 0.0;
2749 for (i=0; i<a->length; i++) coeff[i + (length-1)/2 - (a->length-1)/2]+= a->coeff[i];
2750 for (i=0; i<b->length; i++) coeff[i + (length-1)/2 - (b->length-1)/2]-= b->coeff[i];
2755 /* shift left / or right if "shift" is negative */
2756 static SwsVector *sws_getShiftedVec(SwsVector *a, int shift){
2757 int length= a->length + FFABS(shift)*2;
2758 double *coeff= av_malloc(length*sizeof(double));
2760 SwsVector *vec= av_malloc(sizeof(SwsVector));
2763 vec->length= length;
2765 for (i=0; i<length; i++) coeff[i]= 0.0;
2767 for (i=0; i<a->length; i++)
2769 coeff[i + (length-1)/2 - (a->length-1)/2 - shift]= a->coeff[i];
2775 void sws_shiftVec(SwsVector *a, int shift){
2776 SwsVector *shifted= sws_getShiftedVec(a, shift);
2778 a->coeff= shifted->coeff;
2779 a->length= shifted->length;
2783 void sws_addVec(SwsVector *a, SwsVector *b){
2784 SwsVector *sum= sws_sumVec(a, b);
2786 a->coeff= sum->coeff;
2787 a->length= sum->length;
2791 void sws_subVec(SwsVector *a, SwsVector *b){
2792 SwsVector *diff= sws_diffVec(a, b);
2794 a->coeff= diff->coeff;
2795 a->length= diff->length;
2799 void sws_convVec(SwsVector *a, SwsVector *b){
2800 SwsVector *conv= sws_getConvVec(a, b);
2802 a->coeff= conv->coeff;
2803 a->length= conv->length;
2807 SwsVector *sws_cloneVec(SwsVector *a){
2808 double *coeff= av_malloc(a->length*sizeof(double));
2810 SwsVector *vec= av_malloc(sizeof(SwsVector));
2813 vec->length= a->length;
2815 for (i=0; i<a->length; i++) coeff[i]= a->coeff[i];
2820 void sws_printVec(SwsVector *a){
2826 for (i=0; i<a->length; i++)
2827 if (a->coeff[i]>max) max= a->coeff[i];
2829 for (i=0; i<a->length; i++)
2830 if (a->coeff[i]<min) min= a->coeff[i];
2834 for (i=0; i<a->length; i++)
2836 int x= (int)((a->coeff[i]-min)*60.0/range +0.5);
2837 av_log(NULL, AV_LOG_DEBUG, "%1.3f ", a->coeff[i]);
2838 for (;x>0; x--) av_log(NULL, AV_LOG_DEBUG, " ");
2839 av_log(NULL, AV_LOG_DEBUG, "|\n");
2843 void sws_freeVec(SwsVector *a){
2851 void sws_freeFilter(SwsFilter *filter){
2852 if (!filter) return;
2854 if (filter->lumH) sws_freeVec(filter->lumH);
2855 if (filter->lumV) sws_freeVec(filter->lumV);
2856 if (filter->chrH) sws_freeVec(filter->chrH);
2857 if (filter->chrV) sws_freeVec(filter->chrV);
2862 void sws_freeContext(SwsContext *c){
2868 for (i=0; i<c->vLumBufSize; i++)
2870 av_free(c->lumPixBuf[i]);
2871 c->lumPixBuf[i]=NULL;
2873 av_free(c->lumPixBuf);
2879 for (i=0; i<c->vChrBufSize; i++)
2881 av_free(c->chrPixBuf[i]);
2882 c->chrPixBuf[i]=NULL;
2884 av_free(c->chrPixBuf);
2888 av_free(c->vLumFilter);
2889 c->vLumFilter = NULL;
2890 av_free(c->vChrFilter);
2891 c->vChrFilter = NULL;
2892 av_free(c->hLumFilter);
2893 c->hLumFilter = NULL;
2894 av_free(c->hChrFilter);
2895 c->hChrFilter = NULL;
2897 av_free(c->vYCoeffsBank);
2898 c->vYCoeffsBank = NULL;
2899 av_free(c->vCCoeffsBank);
2900 c->vCCoeffsBank = NULL;
2903 av_free(c->vLumFilterPos);
2904 c->vLumFilterPos = NULL;
2905 av_free(c->vChrFilterPos);
2906 c->vChrFilterPos = NULL;
2907 av_free(c->hLumFilterPos);
2908 c->hLumFilterPos = NULL;
2909 av_free(c->hChrFilterPos);
2910 c->hChrFilterPos = NULL;
2912 #if defined(ARCH_X86) && defined(CONFIG_GPL)
2913 #ifdef MAP_ANONYMOUS
2914 if (c->funnyYCode) munmap(c->funnyYCode, MAX_FUNNY_CODE_SIZE);
2915 if (c->funnyUVCode) munmap(c->funnyUVCode, MAX_FUNNY_CODE_SIZE);
2917 av_free(c->funnyYCode);
2918 av_free(c->funnyUVCode);
2921 c->funnyUVCode=NULL;
2922 #endif /* defined(ARCH_X86) */
2924 av_free(c->lumMmx2Filter);
2925 c->lumMmx2Filter=NULL;
2926 av_free(c->chrMmx2Filter);
2927 c->chrMmx2Filter=NULL;
2928 av_free(c->lumMmx2FilterPos);
2929 c->lumMmx2FilterPos=NULL;
2930 av_free(c->chrMmx2FilterPos);
2931 c->chrMmx2FilterPos=NULL;
2932 av_free(c->yuvTable);
2939 * Checks if context is valid or reallocs a new one instead.
2940 * If context is NULL, just calls sws_getContext() to get a new one.
2941 * Otherwise, checks if the parameters are the same already saved in context.
2942 * If that is the case, returns the current context.
2943 * Otherwise, frees context and gets a new one.
2945 * Be warned that srcFilter, dstFilter are not checked, they are
2946 * asumed to remain valid.
2948 struct SwsContext *sws_getCachedContext(struct SwsContext *context,
2949 int srcW, int srcH, int srcFormat,
2950 int dstW, int dstH, int dstFormat, int flags,
2951 SwsFilter *srcFilter, SwsFilter *dstFilter, double *param)
2953 static const double default_param[2] = {SWS_PARAM_DEFAULT, SWS_PARAM_DEFAULT};
2956 param = default_param;
2959 if (context->srcW != srcW || context->srcH != srcH ||
2960 context->srcFormat != srcFormat ||
2961 context->dstW != dstW || context->dstH != dstH ||
2962 context->dstFormat != dstFormat || context->flags != flags ||
2963 context->param[0] != param[0] || context->param[1] != param[1])
2965 sws_freeContext(context);
2970 return sws_getContext(srcW, srcH, srcFormat,
2971 dstW, dstH, dstFormat, flags,
2972 srcFilter, dstFilter, param);