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
92 #define ASSERT(x) assert(x);
100 #define PI 3.14159265358979323846
103 #define isSupportedIn(x) ( \
104 (x)==PIX_FMT_YUV420P \
105 || (x)==PIX_FMT_YUVA420P \
106 || (x)==PIX_FMT_YUYV422 \
107 || (x)==PIX_FMT_UYVY422 \
108 || (x)==PIX_FMT_RGB32 \
109 || (x)==PIX_FMT_BGR24 \
110 || (x)==PIX_FMT_BGR565 \
111 || (x)==PIX_FMT_BGR555 \
112 || (x)==PIX_FMT_BGR32 \
113 || (x)==PIX_FMT_RGB24 \
114 || (x)==PIX_FMT_RGB565 \
115 || (x)==PIX_FMT_RGB555 \
116 || (x)==PIX_FMT_GRAY8 \
117 || (x)==PIX_FMT_YUV410P \
118 || (x)==PIX_FMT_GRAY16BE \
119 || (x)==PIX_FMT_GRAY16LE \
120 || (x)==PIX_FMT_YUV444P \
121 || (x)==PIX_FMT_YUV422P \
122 || (x)==PIX_FMT_YUV411P \
123 || (x)==PIX_FMT_PAL8 \
124 || (x)==PIX_FMT_BGR8 \
125 || (x)==PIX_FMT_RGB8 \
126 || (x)==PIX_FMT_BGR4_BYTE \
127 || (x)==PIX_FMT_RGB4_BYTE \
128 || (x)==PIX_FMT_YUV440P \
130 #define isSupportedOut(x) ( \
131 (x)==PIX_FMT_YUV420P \
132 || (x)==PIX_FMT_YUYV422 \
133 || (x)==PIX_FMT_UYVY422 \
134 || (x)==PIX_FMT_YUV444P \
135 || (x)==PIX_FMT_YUV422P \
136 || (x)==PIX_FMT_YUV411P \
139 || (x)==PIX_FMT_NV12 \
140 || (x)==PIX_FMT_NV21 \
141 || (x)==PIX_FMT_GRAY16BE \
142 || (x)==PIX_FMT_GRAY16LE \
143 || (x)==PIX_FMT_GRAY8 \
144 || (x)==PIX_FMT_YUV410P \
146 #define isPacked(x) ( \
148 || (x)==PIX_FMT_YUYV422 \
149 || (x)==PIX_FMT_UYVY422 \
154 #define RGB2YUV_SHIFT 16
155 #define BY ((int)( 0.098*(1<<RGB2YUV_SHIFT)+0.5))
156 #define BV ((int)(-0.071*(1<<RGB2YUV_SHIFT)+0.5))
157 #define BU ((int)( 0.439*(1<<RGB2YUV_SHIFT)+0.5))
158 #define GY ((int)( 0.504*(1<<RGB2YUV_SHIFT)+0.5))
159 #define GV ((int)(-0.368*(1<<RGB2YUV_SHIFT)+0.5))
160 #define GU ((int)(-0.291*(1<<RGB2YUV_SHIFT)+0.5))
161 #define RY ((int)( 0.257*(1<<RGB2YUV_SHIFT)+0.5))
162 #define RV ((int)( 0.439*(1<<RGB2YUV_SHIFT)+0.5))
163 #define RU ((int)(-0.148*(1<<RGB2YUV_SHIFT)+0.5))
165 extern const int32_t Inverse_Table_6_9[8][4];
169 Special versions: fast Y 1:1 scaling (no interpolation in y direction)
172 more intelligent misalignment avoidance for the horizontal scaler
173 write special vertical cubic upscale version
174 Optimize C code (yv12 / minmax)
175 add support for packed pixel yuv input & output
176 add support for Y8 output
177 optimize bgr24 & bgr32
178 add BGR4 output support
179 write special BGR->BGR scaler
182 #if defined(ARCH_X86) && defined (CONFIG_GPL)
183 DECLARE_ASM_CONST(8, uint64_t, bF8)= 0xF8F8F8F8F8F8F8F8LL;
184 DECLARE_ASM_CONST(8, uint64_t, bFC)= 0xFCFCFCFCFCFCFCFCLL;
185 DECLARE_ASM_CONST(8, uint64_t, w10)= 0x0010001000100010LL;
186 DECLARE_ASM_CONST(8, uint64_t, w02)= 0x0002000200020002LL;
187 DECLARE_ASM_CONST(8, uint64_t, bm00001111)=0x00000000FFFFFFFFLL;
188 DECLARE_ASM_CONST(8, uint64_t, bm00000111)=0x0000000000FFFFFFLL;
189 DECLARE_ASM_CONST(8, uint64_t, bm11111000)=0xFFFFFFFFFF000000LL;
190 DECLARE_ASM_CONST(8, uint64_t, bm01010101)=0x00FF00FF00FF00FFLL;
192 static volatile uint64_t attribute_used __attribute__((aligned(8))) b5Dither;
193 static volatile uint64_t attribute_used __attribute__((aligned(8))) g5Dither;
194 static volatile uint64_t attribute_used __attribute__((aligned(8))) g6Dither;
195 static volatile uint64_t attribute_used __attribute__((aligned(8))) r5Dither;
197 const DECLARE_ALIGNED(8, uint64_t, ff_dither4[2]) = {
198 0x0103010301030103LL,
199 0x0200020002000200LL,};
201 const DECLARE_ALIGNED(8, uint64_t, ff_dither8[2]) = {
202 0x0602060206020602LL,
203 0x0004000400040004LL,};
205 DECLARE_ASM_CONST(8, uint64_t, b16Mask)= 0x001F001F001F001FLL;
206 DECLARE_ASM_CONST(8, uint64_t, g16Mask)= 0x07E007E007E007E0LL;
207 DECLARE_ASM_CONST(8, uint64_t, r16Mask)= 0xF800F800F800F800LL;
208 DECLARE_ASM_CONST(8, uint64_t, b15Mask)= 0x001F001F001F001FLL;
209 DECLARE_ASM_CONST(8, uint64_t, g15Mask)= 0x03E003E003E003E0LL;
210 DECLARE_ASM_CONST(8, uint64_t, r15Mask)= 0x7C007C007C007C00LL;
212 DECLARE_ALIGNED(8, const uint64_t, ff_M24A) = 0x00FF0000FF0000FFLL;
213 DECLARE_ALIGNED(8, const uint64_t, ff_M24B) = 0xFF0000FF0000FF00LL;
214 DECLARE_ALIGNED(8, const uint64_t, ff_M24C) = 0x0000FF0000FF0000LL;
217 DECLARE_ALIGNED(8, const uint64_t, ff_bgr2YCoeff) = 0x000000210041000DULL;
218 DECLARE_ALIGNED(8, const uint64_t, ff_bgr2UCoeff) = 0x0000FFEEFFDC0038ULL;
219 DECLARE_ALIGNED(8, const uint64_t, ff_bgr2VCoeff) = 0x00000038FFD2FFF8ULL;
221 DECLARE_ALIGNED(8, const uint64_t, ff_bgr2YCoeff) = 0x000020E540830C8BULL;
222 DECLARE_ALIGNED(8, const uint64_t, ff_bgr2UCoeff) = 0x0000ED0FDAC23831ULL;
223 DECLARE_ALIGNED(8, const uint64_t, ff_bgr2VCoeff) = 0x00003831D0E6F6EAULL;
224 #endif /* FAST_BGR2YV12 */
225 DECLARE_ALIGNED(8, const uint64_t, ff_bgr2YOffset) = 0x1010101010101010ULL;
226 DECLARE_ALIGNED(8, const uint64_t, ff_bgr2UVOffset) = 0x8080808080808080ULL;
227 DECLARE_ALIGNED(8, const uint64_t, ff_w1111) = 0x0001000100010001ULL;
228 #endif /* defined(ARCH_X86) */
230 // clipping helper table for C implementations:
231 static unsigned char clip_table[768];
233 static SwsVector *sws_getConvVec(SwsVector *a, SwsVector *b);
235 extern const uint8_t dither_2x2_4[2][8];
236 extern const uint8_t dither_2x2_8[2][8];
237 extern const uint8_t dither_8x8_32[8][8];
238 extern const uint8_t dither_8x8_73[8][8];
239 extern const uint8_t dither_8x8_220[8][8];
241 static const char * sws_context_to_name(void * ptr) {
245 #define OFFSET(x) offsetof(SwsContext, x)
247 #define VE AV_OPT_FLAG_VIDEO_PARAM | AV_OPT_FLAG_ENCODING_PARAM
249 static const AVOption options[] = {
250 { "sws_flags", "scaler/cpu flags", OFFSET(flags), FF_OPT_TYPE_FLAGS, DEFAULT, 0, UINT_MAX, VE, "sws_flags" },
251 { "fast_bilinear", "fast bilinear", 0, FF_OPT_TYPE_CONST, SWS_FAST_BILINEAR, INT_MIN, INT_MAX, VE, "sws_flags" },
252 { "bilinear", "bilinear", 0, FF_OPT_TYPE_CONST, SWS_BILINEAR, INT_MIN, INT_MAX, VE, "sws_flags" },
253 { "bicubic", "bicubic", 0, FF_OPT_TYPE_CONST, SWS_BICUBIC, INT_MIN, INT_MAX, VE, "sws_flags" },
254 { "experimental", "experimental", 0, FF_OPT_TYPE_CONST, SWS_X, INT_MIN, INT_MAX, VE, "sws_flags" },
255 { "neighbor", "nearest neighbor", 0, FF_OPT_TYPE_CONST, SWS_POINT, INT_MIN, INT_MAX, VE, "sws_flags" },
256 { "area", "averaging area", 0, FF_OPT_TYPE_CONST, SWS_AREA, INT_MIN, INT_MAX, VE, "sws_flags" },
257 { "bicublin", "luma bicubic, chroma bilinear", 0, FF_OPT_TYPE_CONST, SWS_BICUBLIN, INT_MIN, INT_MAX, VE, "sws_flags" },
258 { "gauss", "gaussian", 0, FF_OPT_TYPE_CONST, SWS_GAUSS, INT_MIN, INT_MAX, VE, "sws_flags" },
259 { "sinc", "sinc", 0, FF_OPT_TYPE_CONST, SWS_SINC, INT_MIN, INT_MAX, VE, "sws_flags" },
260 { "lanczos", "lanczos", 0, FF_OPT_TYPE_CONST, SWS_LANCZOS, INT_MIN, INT_MAX, VE, "sws_flags" },
261 { "spline", "natural bicubic spline", 0, FF_OPT_TYPE_CONST, SWS_SPLINE, INT_MIN, INT_MAX, VE, "sws_flags" },
262 { "print_info", "print info", 0, FF_OPT_TYPE_CONST, SWS_PRINT_INFO, INT_MIN, INT_MAX, VE, "sws_flags" },
263 { "accurate_rnd", "accurate rounding", 0, FF_OPT_TYPE_CONST, SWS_ACCURATE_RND, INT_MIN, INT_MAX, VE, "sws_flags" },
264 { "mmx", "MMX SIMD acceleration", 0, FF_OPT_TYPE_CONST, SWS_CPU_CAPS_MMX, INT_MIN, INT_MAX, VE, "sws_flags" },
265 { "mmx2", "MMX2 SIMD acceleration", 0, FF_OPT_TYPE_CONST, SWS_CPU_CAPS_MMX2, INT_MIN, INT_MAX, VE, "sws_flags" },
266 { "3dnow", "3DNOW SIMD acceleration", 0, FF_OPT_TYPE_CONST, SWS_CPU_CAPS_3DNOW, INT_MIN, INT_MAX, VE, "sws_flags" },
267 { "altivec", "AltiVec SIMD acceleration", 0, FF_OPT_TYPE_CONST, SWS_CPU_CAPS_ALTIVEC, INT_MIN, INT_MAX, VE, "sws_flags" },
268 { "bfin", "Blackfin SIMD acceleration", 0, FF_OPT_TYPE_CONST, SWS_CPU_CAPS_BFIN, INT_MIN, INT_MAX, VE, "sws_flags" },
269 { "full_chroma_int", "full chroma interpolation", 0 , FF_OPT_TYPE_CONST, SWS_FULL_CHR_H_INT, INT_MIN, INT_MAX, VE, "sws_flags" },
270 { "full_chroma_inp", "full chroma input", 0 , FF_OPT_TYPE_CONST, SWS_FULL_CHR_H_INP, INT_MIN, INT_MAX, VE, "sws_flags" },
277 static const AVClass sws_context_class = { "SWScaler", sws_context_to_name, options };
279 const char *sws_format_name(enum PixelFormat format)
282 case PIX_FMT_YUV420P:
284 case PIX_FMT_YUVA420P:
286 case PIX_FMT_YUYV422:
292 case PIX_FMT_YUV422P:
294 case PIX_FMT_YUV444P:
298 case PIX_FMT_YUV410P:
300 case PIX_FMT_YUV411P:
306 case PIX_FMT_GRAY16BE:
308 case PIX_FMT_GRAY16LE:
312 case PIX_FMT_MONOWHITE:
314 case PIX_FMT_MONOBLACK:
318 case PIX_FMT_YUVJ420P:
320 case PIX_FMT_YUVJ422P:
322 case PIX_FMT_YUVJ444P:
324 case PIX_FMT_XVMC_MPEG2_MC:
325 return "xvmc_mpeg2_mc";
326 case PIX_FMT_XVMC_MPEG2_IDCT:
327 return "xvmc_mpeg2_idct";
328 case PIX_FMT_UYVY422:
330 case PIX_FMT_UYYVYY411:
332 case PIX_FMT_RGB32_1:
334 case PIX_FMT_BGR32_1:
346 case PIX_FMT_BGR4_BYTE:
352 case PIX_FMT_RGB4_BYTE:
358 case PIX_FMT_YUV440P:
361 return "Unknown format";
365 static inline void yuv2yuvXinC(int16_t *lumFilter, int16_t **lumSrc, int lumFilterSize,
366 int16_t *chrFilter, int16_t **chrSrc, int chrFilterSize,
367 uint8_t *dest, uint8_t *uDest, uint8_t *vDest, int dstW, int chrDstW)
369 //FIXME Optimize (just quickly writen not opti..)
371 for (i=0; i<dstW; i++)
375 for (j=0; j<lumFilterSize; j++)
376 val += lumSrc[j][i] * lumFilter[j];
378 dest[i]= av_clip_uint8(val>>19);
382 for (i=0; i<chrDstW; i++)
387 for (j=0; j<chrFilterSize; j++)
389 u += chrSrc[j][i] * chrFilter[j];
390 v += chrSrc[j][i + VOFW] * chrFilter[j];
393 uDest[i]= av_clip_uint8(u>>19);
394 vDest[i]= av_clip_uint8(v>>19);
398 static inline void yuv2nv12XinC(int16_t *lumFilter, int16_t **lumSrc, int lumFilterSize,
399 int16_t *chrFilter, int16_t **chrSrc, int chrFilterSize,
400 uint8_t *dest, uint8_t *uDest, int dstW, int chrDstW, int dstFormat)
402 //FIXME Optimize (just quickly writen not opti..)
404 for (i=0; i<dstW; i++)
408 for (j=0; j<lumFilterSize; j++)
409 val += lumSrc[j][i] * lumFilter[j];
411 dest[i]= av_clip_uint8(val>>19);
417 if (dstFormat == PIX_FMT_NV12)
418 for (i=0; i<chrDstW; i++)
423 for (j=0; j<chrFilterSize; j++)
425 u += chrSrc[j][i] * chrFilter[j];
426 v += chrSrc[j][i + VOFW] * chrFilter[j];
429 uDest[2*i]= av_clip_uint8(u>>19);
430 uDest[2*i+1]= av_clip_uint8(v>>19);
433 for (i=0; i<chrDstW; i++)
438 for (j=0; j<chrFilterSize; j++)
440 u += chrSrc[j][i] * chrFilter[j];
441 v += chrSrc[j][i + VOFW] * chrFilter[j];
444 uDest[2*i]= av_clip_uint8(v>>19);
445 uDest[2*i+1]= av_clip_uint8(u>>19);
449 #define YSCALE_YUV_2_PACKEDX_C(type) \
450 for (i=0; i<(dstW>>1); i++){\
456 type av_unused *r, *b, *g;\
459 for (j=0; j<lumFilterSize; j++)\
461 Y1 += lumSrc[j][i2] * lumFilter[j];\
462 Y2 += lumSrc[j][i2+1] * lumFilter[j];\
464 for (j=0; j<chrFilterSize; j++)\
466 U += chrSrc[j][i] * chrFilter[j];\
467 V += chrSrc[j][i+VOFW] * chrFilter[j];\
473 if ((Y1|Y2|U|V)&256)\
475 if (Y1>255) Y1=255; \
476 else if (Y1<0)Y1=0; \
477 if (Y2>255) Y2=255; \
478 else if (Y2<0)Y2=0; \
485 #define YSCALE_YUV_2_RGBX_C(type) \
486 YSCALE_YUV_2_PACKEDX_C(type) \
487 r = (type *)c->table_rV[V]; \
488 g = (type *)(c->table_gU[U] + c->table_gV[V]); \
489 b = (type *)c->table_bU[U]; \
491 #define YSCALE_YUV_2_PACKED2_C \
492 for (i=0; i<(dstW>>1); i++){ \
494 int Y1= (buf0[i2 ]*yalpha1+buf1[i2 ]*yalpha)>>19; \
495 int Y2= (buf0[i2+1]*yalpha1+buf1[i2+1]*yalpha)>>19; \
496 int U= (uvbuf0[i ]*uvalpha1+uvbuf1[i ]*uvalpha)>>19; \
497 int V= (uvbuf0[i+VOFW]*uvalpha1+uvbuf1[i+VOFW]*uvalpha)>>19; \
499 #define YSCALE_YUV_2_RGB2_C(type) \
500 YSCALE_YUV_2_PACKED2_C\
502 r = (type *)c->table_rV[V];\
503 g = (type *)(c->table_gU[U] + c->table_gV[V]);\
504 b = (type *)c->table_bU[U];\
506 #define YSCALE_YUV_2_PACKED1_C \
507 for (i=0; i<(dstW>>1); i++){\
509 int Y1= buf0[i2 ]>>7;\
510 int Y2= buf0[i2+1]>>7;\
511 int U= (uvbuf1[i ])>>7;\
512 int V= (uvbuf1[i+VOFW])>>7;\
514 #define YSCALE_YUV_2_RGB1_C(type) \
515 YSCALE_YUV_2_PACKED1_C\
517 r = (type *)c->table_rV[V];\
518 g = (type *)(c->table_gU[U] + c->table_gV[V]);\
519 b = (type *)c->table_bU[U];\
521 #define YSCALE_YUV_2_PACKED1B_C \
522 for (i=0; i<(dstW>>1); i++){\
524 int Y1= buf0[i2 ]>>7;\
525 int Y2= buf0[i2+1]>>7;\
526 int U= (uvbuf0[i ] + uvbuf1[i ])>>8;\
527 int V= (uvbuf0[i+VOFW] + uvbuf1[i+VOFW])>>8;\
529 #define YSCALE_YUV_2_RGB1B_C(type) \
530 YSCALE_YUV_2_PACKED1B_C\
532 r = (type *)c->table_rV[V];\
533 g = (type *)(c->table_gU[U] + c->table_gV[V]);\
534 b = (type *)c->table_bU[U];\
536 #define YSCALE_YUV_2_ANYRGB_C(func, func2)\
537 switch(c->dstFormat)\
542 ((uint32_t*)dest)[i2+0]= r[Y1] + g[Y1] + b[Y1];\
543 ((uint32_t*)dest)[i2+1]= r[Y2] + g[Y2] + b[Y2];\
548 ((uint8_t*)dest)[0]= r[Y1];\
549 ((uint8_t*)dest)[1]= g[Y1];\
550 ((uint8_t*)dest)[2]= b[Y1];\
551 ((uint8_t*)dest)[3]= r[Y2];\
552 ((uint8_t*)dest)[4]= g[Y2];\
553 ((uint8_t*)dest)[5]= b[Y2];\
559 ((uint8_t*)dest)[0]= b[Y1];\
560 ((uint8_t*)dest)[1]= g[Y1];\
561 ((uint8_t*)dest)[2]= r[Y1];\
562 ((uint8_t*)dest)[3]= b[Y2];\
563 ((uint8_t*)dest)[4]= g[Y2];\
564 ((uint8_t*)dest)[5]= r[Y2];\
568 case PIX_FMT_RGB565:\
569 case PIX_FMT_BGR565:\
571 const int dr1= dither_2x2_8[y&1 ][0];\
572 const int dg1= dither_2x2_4[y&1 ][0];\
573 const int db1= dither_2x2_8[(y&1)^1][0];\
574 const int dr2= dither_2x2_8[y&1 ][1];\
575 const int dg2= dither_2x2_4[y&1 ][1];\
576 const int db2= dither_2x2_8[(y&1)^1][1];\
578 ((uint16_t*)dest)[i2+0]= r[Y1+dr1] + g[Y1+dg1] + b[Y1+db1];\
579 ((uint16_t*)dest)[i2+1]= r[Y2+dr2] + g[Y2+dg2] + b[Y2+db2];\
583 case PIX_FMT_RGB555:\
584 case PIX_FMT_BGR555:\
586 const int dr1= dither_2x2_8[y&1 ][0];\
587 const int dg1= dither_2x2_8[y&1 ][1];\
588 const int db1= dither_2x2_8[(y&1)^1][0];\
589 const int dr2= dither_2x2_8[y&1 ][1];\
590 const int dg2= dither_2x2_8[y&1 ][0];\
591 const int db2= dither_2x2_8[(y&1)^1][1];\
593 ((uint16_t*)dest)[i2+0]= r[Y1+dr1] + g[Y1+dg1] + b[Y1+db1];\
594 ((uint16_t*)dest)[i2+1]= r[Y2+dr2] + g[Y2+dg2] + b[Y2+db2];\
601 const uint8_t * const d64= dither_8x8_73[y&7];\
602 const uint8_t * const d32= dither_8x8_32[y&7];\
604 ((uint8_t*)dest)[i2+0]= r[Y1+d32[(i2+0)&7]] + g[Y1+d32[(i2+0)&7]] + b[Y1+d64[(i2+0)&7]];\
605 ((uint8_t*)dest)[i2+1]= r[Y2+d32[(i2+1)&7]] + g[Y2+d32[(i2+1)&7]] + b[Y2+d64[(i2+1)&7]];\
612 const uint8_t * const d64= dither_8x8_73 [y&7];\
613 const uint8_t * const d128=dither_8x8_220[y&7];\
615 ((uint8_t*)dest)[i]= r[Y1+d128[(i2+0)&7]] + g[Y1+d64[(i2+0)&7]] + b[Y1+d128[(i2+0)&7]]\
616 + ((r[Y2+d128[(i2+1)&7]] + g[Y2+d64[(i2+1)&7]] + b[Y2+d128[(i2+1)&7]])<<4);\
620 case PIX_FMT_RGB4_BYTE:\
621 case PIX_FMT_BGR4_BYTE:\
623 const uint8_t * const d64= dither_8x8_73 [y&7];\
624 const uint8_t * const d128=dither_8x8_220[y&7];\
626 ((uint8_t*)dest)[i2+0]= r[Y1+d128[(i2+0)&7]] + g[Y1+d64[(i2+0)&7]] + b[Y1+d128[(i2+0)&7]];\
627 ((uint8_t*)dest)[i2+1]= r[Y2+d128[(i2+1)&7]] + g[Y2+d64[(i2+1)&7]] + b[Y2+d128[(i2+1)&7]];\
631 case PIX_FMT_MONOBLACK:\
633 const uint8_t * const d128=dither_8x8_220[y&7];\
634 uint8_t *g= c->table_gU[128] + c->table_gV[128];\
635 for (i=0; i<dstW-7; i+=8){\
637 acc = g[((buf0[i ]*yalpha1+buf1[i ]*yalpha)>>19) + d128[0]];\
638 acc+= acc + g[((buf0[i+1]*yalpha1+buf1[i+1]*yalpha)>>19) + d128[1]];\
639 acc+= acc + g[((buf0[i+2]*yalpha1+buf1[i+2]*yalpha)>>19) + d128[2]];\
640 acc+= acc + g[((buf0[i+3]*yalpha1+buf1[i+3]*yalpha)>>19) + d128[3]];\
641 acc+= acc + g[((buf0[i+4]*yalpha1+buf1[i+4]*yalpha)>>19) + d128[4]];\
642 acc+= acc + g[((buf0[i+5]*yalpha1+buf1[i+5]*yalpha)>>19) + d128[5]];\
643 acc+= acc + g[((buf0[i+6]*yalpha1+buf1[i+6]*yalpha)>>19) + d128[6]];\
644 acc+= acc + g[((buf0[i+7]*yalpha1+buf1[i+7]*yalpha)>>19) + d128[7]];\
645 ((uint8_t*)dest)[0]= acc;\
650 ((uint8_t*)dest)-= dstW>>4;\
654 static int top[1024];\
655 static int last_new[1024][1024];\
656 static int last_in3[1024][1024];\
657 static int drift[1024][1024];\
661 const uint8_t * const d128=dither_8x8_220[y&7];\
666 for (i=dstW>>1; i<dstW; i++){\
667 int in= ((buf0[i ]*yalpha1+buf1[i ]*yalpha)>>19);\
668 int in2 = (76309 * (in - 16) + 32768) >> 16;\
669 int in3 = (in2 < 0) ? 0 : ((in2 > 255) ? 255 : in2);\
670 int old= (left*7 + topLeft + top[i]*5 + top[i+1]*3)/20 + in3\
671 + (last_new[y][i] - in3)*f/256;\
672 int new= old> 128 ? 255 : 0;\
674 error_new+= FFABS(last_new[y][i] - new);\
675 error_in3+= FFABS(last_in3[y][i] - in3);\
676 f= error_new - error_in3*4;\
681 left= top[i]= old - new;\
682 last_new[y][i]= new;\
683 last_in3[y][i]= in3;\
685 acc+= acc + (new&1);\
687 ((uint8_t*)dest)[0]= acc;\
695 case PIX_FMT_YUYV422:\
697 ((uint8_t*)dest)[2*i2+0]= Y1;\
698 ((uint8_t*)dest)[2*i2+1]= U;\
699 ((uint8_t*)dest)[2*i2+2]= Y2;\
700 ((uint8_t*)dest)[2*i2+3]= V;\
703 case PIX_FMT_UYVY422:\
705 ((uint8_t*)dest)[2*i2+0]= U;\
706 ((uint8_t*)dest)[2*i2+1]= Y1;\
707 ((uint8_t*)dest)[2*i2+2]= V;\
708 ((uint8_t*)dest)[2*i2+3]= Y2;\
714 static inline void yuv2packedXinC(SwsContext *c, int16_t *lumFilter, int16_t **lumSrc, int lumFilterSize,
715 int16_t *chrFilter, int16_t **chrSrc, int chrFilterSize,
716 uint8_t *dest, int dstW, int y)
723 YSCALE_YUV_2_RGBX_C(uint32_t)
724 ((uint32_t*)dest)[i2+0]= r[Y1] + g[Y1] + b[Y1];
725 ((uint32_t*)dest)[i2+1]= r[Y2] + g[Y2] + b[Y2];
729 YSCALE_YUV_2_RGBX_C(uint8_t)
730 ((uint8_t*)dest)[0]= r[Y1];
731 ((uint8_t*)dest)[1]= g[Y1];
732 ((uint8_t*)dest)[2]= b[Y1];
733 ((uint8_t*)dest)[3]= r[Y2];
734 ((uint8_t*)dest)[4]= g[Y2];
735 ((uint8_t*)dest)[5]= b[Y2];
740 YSCALE_YUV_2_RGBX_C(uint8_t)
741 ((uint8_t*)dest)[0]= b[Y1];
742 ((uint8_t*)dest)[1]= g[Y1];
743 ((uint8_t*)dest)[2]= r[Y1];
744 ((uint8_t*)dest)[3]= b[Y2];
745 ((uint8_t*)dest)[4]= g[Y2];
746 ((uint8_t*)dest)[5]= r[Y2];
753 const int dr1= dither_2x2_8[y&1 ][0];
754 const int dg1= dither_2x2_4[y&1 ][0];
755 const int db1= dither_2x2_8[(y&1)^1][0];
756 const int dr2= dither_2x2_8[y&1 ][1];
757 const int dg2= dither_2x2_4[y&1 ][1];
758 const int db2= dither_2x2_8[(y&1)^1][1];
759 YSCALE_YUV_2_RGBX_C(uint16_t)
760 ((uint16_t*)dest)[i2+0]= r[Y1+dr1] + g[Y1+dg1] + b[Y1+db1];
761 ((uint16_t*)dest)[i2+1]= r[Y2+dr2] + g[Y2+dg2] + b[Y2+db2];
768 const int dr1= dither_2x2_8[y&1 ][0];
769 const int dg1= dither_2x2_8[y&1 ][1];
770 const int db1= dither_2x2_8[(y&1)^1][0];
771 const int dr2= dither_2x2_8[y&1 ][1];
772 const int dg2= dither_2x2_8[y&1 ][0];
773 const int db2= dither_2x2_8[(y&1)^1][1];
774 YSCALE_YUV_2_RGBX_C(uint16_t)
775 ((uint16_t*)dest)[i2+0]= r[Y1+dr1] + g[Y1+dg1] + b[Y1+db1];
776 ((uint16_t*)dest)[i2+1]= r[Y2+dr2] + g[Y2+dg2] + b[Y2+db2];
783 const uint8_t * const d64= dither_8x8_73[y&7];
784 const uint8_t * const d32= dither_8x8_32[y&7];
785 YSCALE_YUV_2_RGBX_C(uint8_t)
786 ((uint8_t*)dest)[i2+0]= r[Y1+d32[(i2+0)&7]] + g[Y1+d32[(i2+0)&7]] + b[Y1+d64[(i2+0)&7]];
787 ((uint8_t*)dest)[i2+1]= r[Y2+d32[(i2+1)&7]] + g[Y2+d32[(i2+1)&7]] + b[Y2+d64[(i2+1)&7]];
794 const uint8_t * const d64= dither_8x8_73 [y&7];
795 const uint8_t * const d128=dither_8x8_220[y&7];
796 YSCALE_YUV_2_RGBX_C(uint8_t)
797 ((uint8_t*)dest)[i]= r[Y1+d128[(i2+0)&7]] + g[Y1+d64[(i2+0)&7]] + b[Y1+d128[(i2+0)&7]]
798 +((r[Y2+d128[(i2+1)&7]] + g[Y2+d64[(i2+1)&7]] + b[Y2+d128[(i2+1)&7]])<<4);
802 case PIX_FMT_RGB4_BYTE:
803 case PIX_FMT_BGR4_BYTE:
805 const uint8_t * const d64= dither_8x8_73 [y&7];
806 const uint8_t * const d128=dither_8x8_220[y&7];
807 YSCALE_YUV_2_RGBX_C(uint8_t)
808 ((uint8_t*)dest)[i2+0]= r[Y1+d128[(i2+0)&7]] + g[Y1+d64[(i2+0)&7]] + b[Y1+d128[(i2+0)&7]];
809 ((uint8_t*)dest)[i2+1]= r[Y2+d128[(i2+1)&7]] + g[Y2+d64[(i2+1)&7]] + b[Y2+d128[(i2+1)&7]];
813 case PIX_FMT_MONOBLACK:
815 const uint8_t * const d128=dither_8x8_220[y&7];
816 uint8_t *g= c->table_gU[128] + c->table_gV[128];
818 for (i=0; i<dstW-1; i+=2){
823 for (j=0; j<lumFilterSize; j++)
825 Y1 += lumSrc[j][i] * lumFilter[j];
826 Y2 += lumSrc[j][i+1] * lumFilter[j];
837 acc+= acc + g[Y1+d128[(i+0)&7]];
838 acc+= acc + g[Y2+d128[(i+1)&7]];
840 ((uint8_t*)dest)[0]= acc;
846 case PIX_FMT_YUYV422:
847 YSCALE_YUV_2_PACKEDX_C(void)
848 ((uint8_t*)dest)[2*i2+0]= Y1;
849 ((uint8_t*)dest)[2*i2+1]= U;
850 ((uint8_t*)dest)[2*i2+2]= Y2;
851 ((uint8_t*)dest)[2*i2+3]= V;
854 case PIX_FMT_UYVY422:
855 YSCALE_YUV_2_PACKEDX_C(void)
856 ((uint8_t*)dest)[2*i2+0]= U;
857 ((uint8_t*)dest)[2*i2+1]= Y1;
858 ((uint8_t*)dest)[2*i2+2]= V;
859 ((uint8_t*)dest)[2*i2+3]= Y2;
866 //Note: we have C, X86, MMX, MMX2, 3DNOW version therse no 3DNOW+MMX2 one
868 #if !defined (HAVE_MMX) || defined (RUNTIME_CPUDETECT) || !defined(CONFIG_GPL)
873 #if (defined (HAVE_ALTIVEC) || defined (RUNTIME_CPUDETECT)) && defined (CONFIG_GPL)
874 #define COMPILE_ALTIVEC
875 #endif //HAVE_ALTIVEC
876 #endif //ARCH_POWERPC
878 #if defined(ARCH_X86)
880 #if ((defined (HAVE_MMX) && !defined (HAVE_3DNOW) && !defined (HAVE_MMX2)) || defined (RUNTIME_CPUDETECT)) && defined (CONFIG_GPL)
884 #if (defined (HAVE_MMX2) || defined (RUNTIME_CPUDETECT)) && defined (CONFIG_GPL)
888 #if ((defined (HAVE_3DNOW) && !defined (HAVE_MMX2)) || defined (RUNTIME_CPUDETECT)) && defined (CONFIG_GPL)
889 #define COMPILE_3DNOW
891 #endif //ARCH_X86 || ARCH_X86_64
902 #define RENAME(a) a ## _C
903 #include "swscale_template.c"
906 #ifdef COMPILE_ALTIVEC
909 #define RENAME(a) a ## _altivec
910 #include "swscale_template.c"
913 #if defined(ARCH_X86)
922 #define RENAME(a) a ## _X86
923 #include "swscale_template.c"
931 #define RENAME(a) a ## _MMX
932 #include "swscale_template.c"
941 #define RENAME(a) a ## _MMX2
942 #include "swscale_template.c"
951 #define RENAME(a) a ## _3DNow
952 #include "swscale_template.c"
955 #endif //ARCH_X86 || ARCH_X86_64
957 // minor note: the HAVE_xyz is messed up after that line so don't use it
959 static double getSplineCoeff(double a, double b, double c, double d, double dist)
961 // printf("%f %f %f %f %f\n", a,b,c,d,dist);
962 if (dist<=1.0) return ((d*dist + c)*dist + b)*dist +a;
963 else return getSplineCoeff( 0.0,
970 static inline int initFilter(int16_t **outFilter, int16_t **filterPos, int *outFilterSize, int xInc,
971 int srcW, int dstW, int filterAlign, int one, int flags,
972 SwsVector *srcFilter, SwsVector *dstFilter, double param[2])
979 double *filter2=NULL;
980 #if defined(ARCH_X86)
981 if (flags & SWS_CPU_CAPS_MMX)
982 asm volatile("emms\n\t"::: "memory"); //FIXME this should not be required but it IS (even for non-MMX versions)
985 // Note the +1 is for the MMXscaler which reads over the end
986 *filterPos = av_malloc((dstW+1)*sizeof(int16_t));
988 if (FFABS(xInc - 0x10000) <10) // unscaled
992 filter= av_malloc(dstW*sizeof(double)*filterSize);
993 for (i=0; i<dstW*filterSize; i++) filter[i]=0;
995 for (i=0; i<dstW; i++)
997 filter[i*filterSize]=1;
1002 else if (flags&SWS_POINT) // lame looking point sampling mode
1007 filter= av_malloc(dstW*sizeof(double)*filterSize);
1009 xDstInSrc= xInc/2 - 0x8000;
1010 for (i=0; i<dstW; i++)
1012 int xx= (xDstInSrc - ((filterSize-1)<<15) + (1<<15))>>16;
1014 (*filterPos)[i]= xx;
1019 else if ((xInc <= (1<<16) && (flags&SWS_AREA)) || (flags&SWS_FAST_BILINEAR)) // bilinear upscale
1023 if (flags&SWS_BICUBIC) filterSize= 4;
1024 else if (flags&SWS_X ) filterSize= 4;
1025 else filterSize= 2; // SWS_BILINEAR / SWS_AREA
1026 filter= av_malloc(dstW*sizeof(double)*filterSize);
1028 xDstInSrc= xInc/2 - 0x8000;
1029 for (i=0; i<dstW; i++)
1031 int xx= (xDstInSrc - ((filterSize-1)<<15) + (1<<15))>>16;
1034 (*filterPos)[i]= xx;
1035 //Bilinear upscale / linear interpolate / Area averaging
1036 for (j=0; j<filterSize; j++)
1038 double d= FFABS((xx<<16) - xDstInSrc)/(double)(1<<16);
1039 double coeff= 1.0 - d;
1040 if (coeff<0) coeff=0;
1041 filter[i*filterSize + j]= coeff;
1050 double sizeFactor, filterSizeInSrc;
1051 const double xInc1= (double)xInc / (double)(1<<16);
1053 if (flags&SWS_BICUBIC) sizeFactor= 4.0;
1054 else if (flags&SWS_X) sizeFactor= 8.0;
1055 else if (flags&SWS_AREA) sizeFactor= 1.0; //downscale only, for upscale it is bilinear
1056 else if (flags&SWS_GAUSS) sizeFactor= 8.0; // infinite ;)
1057 else if (flags&SWS_LANCZOS) sizeFactor= param[0] != SWS_PARAM_DEFAULT ? 2.0*param[0] : 6.0;
1058 else if (flags&SWS_SINC) sizeFactor= 20.0; // infinite ;)
1059 else if (flags&SWS_SPLINE) sizeFactor= 20.0; // infinite ;)
1060 else if (flags&SWS_BILINEAR) sizeFactor= 2.0;
1062 sizeFactor= 0.0; //GCC warning killer
1066 if (xInc1 <= 1.0) filterSizeInSrc= sizeFactor; // upscale
1067 else filterSizeInSrc= sizeFactor*srcW / (double)dstW;
1069 filterSize= (int)ceil(1 + filterSizeInSrc); // will be reduced later if possible
1070 if (filterSize > srcW-2) filterSize=srcW-2;
1072 filter= av_malloc(dstW*sizeof(double)*filterSize);
1074 xDstInSrc= xInc1 / 2.0 - 0.5;
1075 for (i=0; i<dstW; i++)
1077 int xx= (int)(xDstInSrc - (filterSize-1)*0.5 + 0.5);
1079 (*filterPos)[i]= xx;
1080 for (j=0; j<filterSize; j++)
1082 double d= FFABS(xx - xDstInSrc)/filterSizeInSrc*sizeFactor;
1084 if (flags & SWS_BICUBIC)
1086 double B= param[0] != SWS_PARAM_DEFAULT ? param[0] : 0.0;
1087 double C= param[1] != SWS_PARAM_DEFAULT ? param[1] : 0.6;
1090 coeff = (12-9*B-6*C)*d*d*d + (-18+12*B+6*C)*d*d + 6-2*B;
1092 coeff = (-B-6*C)*d*d*d + (6*B+30*C)*d*d + (-12*B-48*C)*d +8*B+24*C;
1096 /* else if (flags & SWS_X)
1098 double p= param ? param*0.01 : 0.3;
1099 coeff = d ? sin(d*PI)/(d*PI) : 1.0;
1100 coeff*= pow(2.0, - p*d*d);
1102 else if (flags & SWS_X)
1104 double A= param[0] != SWS_PARAM_DEFAULT ? param[0] : 1.0;
1110 if (coeff<0.0) coeff= -pow(-coeff, A);
1111 else coeff= pow( coeff, A);
1112 coeff= coeff*0.5 + 0.5;
1114 else if (flags & SWS_AREA)
1116 double srcPixelSize= 1.0/xInc1;
1117 if (d + srcPixelSize/2 < 0.5) coeff= 1.0;
1118 else if (d - srcPixelSize/2 < 0.5) coeff= (0.5-d)/srcPixelSize + 0.5;
1121 else if (flags & SWS_GAUSS)
1123 double p= param[0] != SWS_PARAM_DEFAULT ? param[0] : 3.0;
1124 coeff = pow(2.0, - p*d*d);
1126 else if (flags & SWS_SINC)
1128 coeff = d ? sin(d*PI)/(d*PI) : 1.0;
1130 else if (flags & SWS_LANCZOS)
1132 double p= param[0] != SWS_PARAM_DEFAULT ? param[0] : 3.0;
1133 coeff = d ? sin(d*PI)*sin(d*PI/p)/(d*d*PI*PI/p) : 1.0;
1136 else if (flags & SWS_BILINEAR)
1139 if (coeff<0) coeff=0;
1141 else if (flags & SWS_SPLINE)
1143 double p=-2.196152422706632;
1144 coeff = getSplineCoeff(1.0, 0.0, p, -p-1.0, d);
1147 coeff= 0.0; //GCC warning killer
1151 filter[i*filterSize + j]= coeff;
1158 /* apply src & dst Filter to filter -> filter2
1161 ASSERT(filterSize>0)
1162 filter2Size= filterSize;
1163 if (srcFilter) filter2Size+= srcFilter->length - 1;
1164 if (dstFilter) filter2Size+= dstFilter->length - 1;
1165 ASSERT(filter2Size>0)
1166 filter2= av_malloc(filter2Size*dstW*sizeof(double));
1168 for (i=0; i<dstW; i++)
1171 SwsVector scaleFilter;
1174 scaleFilter.coeff= filter + i*filterSize;
1175 scaleFilter.length= filterSize;
1177 if (srcFilter) outVec= sws_getConvVec(srcFilter, &scaleFilter);
1178 else outVec= &scaleFilter;
1180 ASSERT(outVec->length == filter2Size)
1183 for (j=0; j<outVec->length; j++)
1185 filter2[i*filter2Size + j]= outVec->coeff[j];
1188 (*filterPos)[i]+= (filterSize-1)/2 - (filter2Size-1)/2;
1190 if (outVec != &scaleFilter) sws_freeVec(outVec);
1192 av_free(filter); filter=NULL;
1194 /* try to reduce the filter-size (step1 find size and shift left) */
1195 // Assume it is near normalized (*0.5 or *2.0 is OK but * 0.001 is not).
1197 for (i=dstW-1; i>=0; i--)
1199 int min= filter2Size;
1203 /* get rid off near zero elements on the left by shifting left */
1204 for (j=0; j<filter2Size; j++)
1207 cutOff += FFABS(filter2[i*filter2Size]);
1209 if (cutOff > SWS_MAX_REDUCE_CUTOFF) break;
1211 /* preserve monotonicity because the core can't handle the filter otherwise */
1212 if (i<dstW-1 && (*filterPos)[i] >= (*filterPos)[i+1]) break;
1214 // Move filter coeffs left
1215 for (k=1; k<filter2Size; k++)
1216 filter2[i*filter2Size + k - 1]= filter2[i*filter2Size + k];
1217 filter2[i*filter2Size + k - 1]= 0.0;
1222 /* count near zeros on the right */
1223 for (j=filter2Size-1; j>0; j--)
1225 cutOff += FFABS(filter2[i*filter2Size + j]);
1227 if (cutOff > SWS_MAX_REDUCE_CUTOFF) break;
1231 if (min>minFilterSize) minFilterSize= min;
1234 if (flags & SWS_CPU_CAPS_ALTIVEC) {
1235 // we can handle the special case 4,
1236 // so we don't want to go to the full 8
1237 if (minFilterSize < 5)
1240 // we really don't want to waste our time
1241 // doing useless computation, so fall-back on
1242 // the scalar C code for very small filter.
1243 // vectorizing is worth it only if you have
1244 // decent-sized vector.
1245 if (minFilterSize < 3)
1249 if (flags & SWS_CPU_CAPS_MMX) {
1250 // special case for unscaled vertical filtering
1251 if (minFilterSize == 1 && filterAlign == 2)
1255 ASSERT(minFilterSize > 0)
1256 filterSize= (minFilterSize +(filterAlign-1)) & (~(filterAlign-1));
1257 ASSERT(filterSize > 0)
1258 filter= av_malloc(filterSize*dstW*sizeof(double));
1259 if (filterSize >= MAX_FILTER_SIZE)
1261 *outFilterSize= filterSize;
1263 if (flags&SWS_PRINT_INFO)
1264 av_log(NULL, AV_LOG_VERBOSE, "SwScaler: reducing / aligning filtersize %d -> %d\n", filter2Size, filterSize);
1265 /* try to reduce the filter-size (step2 reduce it) */
1266 for (i=0; i<dstW; i++)
1270 for (j=0; j<filterSize; j++)
1272 if (j>=filter2Size) filter[i*filterSize + j]= 0.0;
1273 else filter[i*filterSize + j]= filter2[i*filter2Size + j];
1276 av_free(filter2); filter2=NULL;
1279 //FIXME try to align filterpos if possible
1282 for (i=0; i<dstW; i++)
1285 if ((*filterPos)[i] < 0)
1287 // Move filter coeffs left to compensate for filterPos
1288 for (j=1; j<filterSize; j++)
1290 int left= FFMAX(j + (*filterPos)[i], 0);
1291 filter[i*filterSize + left] += filter[i*filterSize + j];
1292 filter[i*filterSize + j]=0;
1297 if ((*filterPos)[i] + filterSize > srcW)
1299 int shift= (*filterPos)[i] + filterSize - srcW;
1300 // Move filter coeffs right to compensate for filterPos
1301 for (j=filterSize-2; j>=0; j--)
1303 int right= FFMIN(j + shift, filterSize-1);
1304 filter[i*filterSize +right] += filter[i*filterSize +j];
1305 filter[i*filterSize +j]=0;
1307 (*filterPos)[i]= srcW - filterSize;
1311 // Note the +1 is for the MMXscaler which reads over the end
1312 /* align at 16 for AltiVec (needed by hScale_altivec_real) */
1313 *outFilter= av_mallocz(*outFilterSize*(dstW+1)*sizeof(int16_t));
1315 /* Normalize & Store in outFilter */
1316 for (i=0; i<dstW; i++)
1323 for (j=0; j<filterSize; j++)
1325 sum+= filter[i*filterSize + j];
1328 for (j=0; j<*outFilterSize; j++)
1330 double v= filter[i*filterSize + j]*scale + error;
1331 int intV= floor(v + 0.5);
1332 (*outFilter)[i*(*outFilterSize) + j]= intV;
1337 (*filterPos)[dstW]= (*filterPos)[dstW-1]; // the MMX scaler will read over the end
1338 for (i=0; i<*outFilterSize; i++)
1340 int j= dstW*(*outFilterSize);
1341 (*outFilter)[j + i]= (*outFilter)[j + i - (*outFilterSize)];
1349 static void initMMX2HScaler(int dstW, int xInc, uint8_t *funnyCode, int16_t *filter, int32_t *filterPos, int numSplits)
1352 long imm8OfPShufW1A;
1353 long imm8OfPShufW2A;
1354 long fragmentLengthA;
1356 long imm8OfPShufW1B;
1357 long imm8OfPShufW2B;
1358 long fragmentLengthB;
1363 // create an optimized horizontal scaling routine
1371 "movq (%%"REG_d", %%"REG_a"), %%mm3 \n\t"
1372 "movd (%%"REG_c", %%"REG_S"), %%mm0 \n\t"
1373 "movd 1(%%"REG_c", %%"REG_S"), %%mm1 \n\t"
1374 "punpcklbw %%mm7, %%mm1 \n\t"
1375 "punpcklbw %%mm7, %%mm0 \n\t"
1376 "pshufw $0xFF, %%mm1, %%mm1 \n\t"
1378 "pshufw $0xFF, %%mm0, %%mm0 \n\t"
1380 "psubw %%mm1, %%mm0 \n\t"
1381 "movl 8(%%"REG_b", %%"REG_a"), %%esi \n\t"
1382 "pmullw %%mm3, %%mm0 \n\t"
1383 "psllw $7, %%mm1 \n\t"
1384 "paddw %%mm1, %%mm0 \n\t"
1386 "movq %%mm0, (%%"REG_D", %%"REG_a") \n\t"
1388 "add $8, %%"REG_a" \n\t"
1392 "lea " LOCAL_MANGLE(0b) ", %0 \n\t"
1393 "lea " LOCAL_MANGLE(1b) ", %1 \n\t"
1394 "lea " LOCAL_MANGLE(2b) ", %2 \n\t"
1399 "lea " LOCAL_MANGLE(9b) ", %3 \n\t"
1403 :"=r" (fragmentA), "=r" (imm8OfPShufW1A), "=r" (imm8OfPShufW2A),
1404 "=r" (fragmentLengthA)
1411 "movq (%%"REG_d", %%"REG_a"), %%mm3 \n\t"
1412 "movd (%%"REG_c", %%"REG_S"), %%mm0 \n\t"
1413 "punpcklbw %%mm7, %%mm0 \n\t"
1414 "pshufw $0xFF, %%mm0, %%mm1 \n\t"
1416 "pshufw $0xFF, %%mm0, %%mm0 \n\t"
1418 "psubw %%mm1, %%mm0 \n\t"
1419 "movl 8(%%"REG_b", %%"REG_a"), %%esi \n\t"
1420 "pmullw %%mm3, %%mm0 \n\t"
1421 "psllw $7, %%mm1 \n\t"
1422 "paddw %%mm1, %%mm0 \n\t"
1424 "movq %%mm0, (%%"REG_D", %%"REG_a") \n\t"
1426 "add $8, %%"REG_a" \n\t"
1430 "lea " LOCAL_MANGLE(0b) ", %0 \n\t"
1431 "lea " LOCAL_MANGLE(1b) ", %1 \n\t"
1432 "lea " LOCAL_MANGLE(2b) ", %2 \n\t"
1437 "lea " LOCAL_MANGLE(9b) ", %3 \n\t"
1441 :"=r" (fragmentB), "=r" (imm8OfPShufW1B), "=r" (imm8OfPShufW2B),
1442 "=r" (fragmentLengthB)
1445 xpos= 0; //lumXInc/2 - 0x8000; // difference between pixel centers
1448 for (i=0; i<dstW/numSplits; i++)
1455 int b=((xpos+xInc)>>16) - xx;
1456 int c=((xpos+xInc*2)>>16) - xx;
1457 int d=((xpos+xInc*3)>>16) - xx;
1459 filter[i ] = (( xpos & 0xFFFF) ^ 0xFFFF)>>9;
1460 filter[i+1] = (((xpos+xInc ) & 0xFFFF) ^ 0xFFFF)>>9;
1461 filter[i+2] = (((xpos+xInc*2) & 0xFFFF) ^ 0xFFFF)>>9;
1462 filter[i+3] = (((xpos+xInc*3) & 0xFFFF) ^ 0xFFFF)>>9;
1467 int maxShift= 3-(d+1);
1470 memcpy(funnyCode + fragmentPos, fragmentB, fragmentLengthB);
1472 funnyCode[fragmentPos + imm8OfPShufW1B]=
1473 (a+1) | ((b+1)<<2) | ((c+1)<<4) | ((d+1)<<6);
1474 funnyCode[fragmentPos + imm8OfPShufW2B]=
1475 a | (b<<2) | (c<<4) | (d<<6);
1477 if (i+3>=dstW) shift=maxShift; //avoid overread
1478 else if ((filterPos[i/2]&3) <= maxShift) shift=filterPos[i/2]&3; //Align
1480 if (shift && i>=shift)
1482 funnyCode[fragmentPos + imm8OfPShufW1B]+= 0x55*shift;
1483 funnyCode[fragmentPos + imm8OfPShufW2B]+= 0x55*shift;
1484 filterPos[i/2]-=shift;
1487 fragmentPos+= fragmentLengthB;
1494 memcpy(funnyCode + fragmentPos, fragmentA, fragmentLengthA);
1496 funnyCode[fragmentPos + imm8OfPShufW1A]=
1497 funnyCode[fragmentPos + imm8OfPShufW2A]=
1498 a | (b<<2) | (c<<4) | (d<<6);
1500 if (i+4>=dstW) shift=maxShift; //avoid overread
1501 else if ((filterPos[i/2]&3) <= maxShift) shift=filterPos[i/2]&3; //partial align
1503 if (shift && i>=shift)
1505 funnyCode[fragmentPos + imm8OfPShufW1A]+= 0x55*shift;
1506 funnyCode[fragmentPos + imm8OfPShufW2A]+= 0x55*shift;
1507 filterPos[i/2]-=shift;
1510 fragmentPos+= fragmentLengthA;
1513 funnyCode[fragmentPos]= RET;
1517 filterPos[i/2]= xpos>>16; // needed to jump to the next part
1519 #endif /* COMPILE_MMX2 */
1521 static void globalInit(void){
1522 // generating tables:
1524 for (i=0; i<768; i++){
1525 int c= av_clip_uint8(i-256);
1530 static SwsFunc getSwsFunc(int flags){
1532 #if defined(RUNTIME_CPUDETECT) && defined (CONFIG_GPL)
1533 #if defined(ARCH_X86)
1534 // ordered per speed fastest first
1535 if (flags & SWS_CPU_CAPS_MMX2)
1536 return swScale_MMX2;
1537 else if (flags & SWS_CPU_CAPS_3DNOW)
1538 return swScale_3DNow;
1539 else if (flags & SWS_CPU_CAPS_MMX)
1546 if (flags & SWS_CPU_CAPS_ALTIVEC)
1547 return swScale_altivec;
1552 #endif /* defined(ARCH_X86) */
1553 #else //RUNTIME_CPUDETECT
1555 return swScale_MMX2;
1556 #elif defined (HAVE_3DNOW)
1557 return swScale_3DNow;
1558 #elif defined (HAVE_MMX)
1560 #elif defined (HAVE_ALTIVEC)
1561 return swScale_altivec;
1565 #endif //!RUNTIME_CPUDETECT
1568 static int PlanarToNV12Wrapper(SwsContext *c, uint8_t* src[], int srcStride[], int srcSliceY,
1569 int srcSliceH, uint8_t* dstParam[], int dstStride[]){
1570 uint8_t *dst=dstParam[0] + dstStride[0]*srcSliceY;
1572 if (dstStride[0]==srcStride[0] && srcStride[0] > 0)
1573 memcpy(dst, src[0], srcSliceH*dstStride[0]);
1577 uint8_t *srcPtr= src[0];
1578 uint8_t *dstPtr= dst;
1579 for (i=0; i<srcSliceH; i++)
1581 memcpy(dstPtr, srcPtr, c->srcW);
1582 srcPtr+= srcStride[0];
1583 dstPtr+= dstStride[0];
1586 dst = dstParam[1] + dstStride[1]*srcSliceY/2;
1587 if (c->dstFormat == PIX_FMT_NV12)
1588 interleaveBytes(src[1], src[2], dst, c->srcW/2, srcSliceH/2, srcStride[1], srcStride[2], dstStride[0]);
1590 interleaveBytes(src[2], src[1], dst, c->srcW/2, srcSliceH/2, srcStride[2], srcStride[1], dstStride[0]);
1595 static int PlanarToYuy2Wrapper(SwsContext *c, uint8_t* src[], int srcStride[], int srcSliceY,
1596 int srcSliceH, uint8_t* dstParam[], int dstStride[]){
1597 uint8_t *dst=dstParam[0] + dstStride[0]*srcSliceY;
1599 yv12toyuy2(src[0], src[1], src[2], dst, c->srcW, srcSliceH, srcStride[0], srcStride[1], dstStride[0]);
1604 static int PlanarToUyvyWrapper(SwsContext *c, uint8_t* src[], int srcStride[], int srcSliceY,
1605 int srcSliceH, uint8_t* dstParam[], int dstStride[]){
1606 uint8_t *dst=dstParam[0] + dstStride[0]*srcSliceY;
1608 yv12touyvy(src[0], src[1], src[2], dst, c->srcW, srcSliceH, srcStride[0], srcStride[1], dstStride[0]);
1613 /* {RGB,BGR}{15,16,24,32} -> {RGB,BGR}{15,16,24,32} */
1614 static int rgb2rgbWrapper(SwsContext *c, uint8_t* src[], int srcStride[], int srcSliceY,
1615 int srcSliceH, uint8_t* dst[], int dstStride[]){
1616 const int srcFormat= c->srcFormat;
1617 const int dstFormat= c->dstFormat;
1618 const int srcBpp= (fmt_depth(srcFormat) + 7) >> 3;
1619 const int dstBpp= (fmt_depth(dstFormat) + 7) >> 3;
1620 const int srcId= fmt_depth(srcFormat) >> 2; /* 1:0, 4:1, 8:2, 15:3, 16:4, 24:6, 32:8 */
1621 const int dstId= fmt_depth(dstFormat) >> 2;
1622 void (*conv)(const uint8_t *src, uint8_t *dst, long src_size)=NULL;
1625 if ( (isBGR(srcFormat) && isBGR(dstFormat))
1626 || (isRGB(srcFormat) && isRGB(dstFormat))){
1627 switch(srcId | (dstId<<4)){
1628 case 0x34: conv= rgb16to15; break;
1629 case 0x36: conv= rgb24to15; break;
1630 case 0x38: conv= rgb32to15; break;
1631 case 0x43: conv= rgb15to16; break;
1632 case 0x46: conv= rgb24to16; break;
1633 case 0x48: conv= rgb32to16; break;
1634 case 0x63: conv= rgb15to24; break;
1635 case 0x64: conv= rgb16to24; break;
1636 case 0x68: conv= rgb32to24; break;
1637 case 0x83: conv= rgb15to32; break;
1638 case 0x84: conv= rgb16to32; break;
1639 case 0x86: conv= rgb24to32; break;
1640 default: av_log(c, AV_LOG_ERROR, "internal error %s -> %s converter\n",
1641 sws_format_name(srcFormat), sws_format_name(dstFormat)); break;
1643 }else if ( (isBGR(srcFormat) && isRGB(dstFormat))
1644 || (isRGB(srcFormat) && isBGR(dstFormat))){
1645 switch(srcId | (dstId<<4)){
1646 case 0x33: conv= rgb15tobgr15; break;
1647 case 0x34: conv= rgb16tobgr15; break;
1648 case 0x36: conv= rgb24tobgr15; break;
1649 case 0x38: conv= rgb32tobgr15; break;
1650 case 0x43: conv= rgb15tobgr16; break;
1651 case 0x44: conv= rgb16tobgr16; break;
1652 case 0x46: conv= rgb24tobgr16; break;
1653 case 0x48: conv= rgb32tobgr16; break;
1654 case 0x63: conv= rgb15tobgr24; break;
1655 case 0x64: conv= rgb16tobgr24; break;
1656 case 0x66: conv= rgb24tobgr24; break;
1657 case 0x68: conv= rgb32tobgr24; break;
1658 case 0x83: conv= rgb15tobgr32; break;
1659 case 0x84: conv= rgb16tobgr32; break;
1660 case 0x86: conv= rgb24tobgr32; break;
1661 case 0x88: conv= rgb32tobgr32; break;
1662 default: av_log(c, AV_LOG_ERROR, "internal error %s -> %s converter\n",
1663 sws_format_name(srcFormat), sws_format_name(dstFormat)); break;
1666 av_log(c, AV_LOG_ERROR, "internal error %s -> %s converter\n",
1667 sws_format_name(srcFormat), sws_format_name(dstFormat));
1672 if (dstStride[0]*srcBpp == srcStride[0]*dstBpp && srcStride[0] > 0)
1673 conv(src[0], dst[0] + dstStride[0]*srcSliceY, srcSliceH*srcStride[0]);
1677 uint8_t *srcPtr= src[0];
1678 uint8_t *dstPtr= dst[0] + dstStride[0]*srcSliceY;
1680 for (i=0; i<srcSliceH; i++)
1682 conv(srcPtr, dstPtr, c->srcW*srcBpp);
1683 srcPtr+= srcStride[0];
1684 dstPtr+= dstStride[0];
1691 static int bgr24toyv12Wrapper(SwsContext *c, uint8_t* src[], int srcStride[], int srcSliceY,
1692 int srcSliceH, uint8_t* dst[], int dstStride[]){
1696 dst[0]+ srcSliceY *dstStride[0],
1697 dst[1]+(srcSliceY>>1)*dstStride[1],
1698 dst[2]+(srcSliceY>>1)*dstStride[2],
1700 dstStride[0], dstStride[1], srcStride[0]);
1704 static int yvu9toyv12Wrapper(SwsContext *c, uint8_t* src[], int srcStride[], int srcSliceY,
1705 int srcSliceH, uint8_t* dst[], int dstStride[]){
1709 if (srcStride[0]==dstStride[0] && srcStride[0] > 0)
1710 memcpy(dst[0]+ srcSliceY*dstStride[0], src[0], srcStride[0]*srcSliceH);
1712 uint8_t *srcPtr= src[0];
1713 uint8_t *dstPtr= dst[0] + dstStride[0]*srcSliceY;
1715 for (i=0; i<srcSliceH; i++)
1717 memcpy(dstPtr, srcPtr, c->srcW);
1718 srcPtr+= srcStride[0];
1719 dstPtr+= dstStride[0];
1723 if (c->dstFormat==PIX_FMT_YUV420P){
1724 planar2x(src[1], dst[1], c->chrSrcW, c->chrSrcH, srcStride[1], dstStride[1]);
1725 planar2x(src[2], dst[2], c->chrSrcW, c->chrSrcH, srcStride[2], dstStride[2]);
1727 planar2x(src[1], dst[2], c->chrSrcW, c->chrSrcH, srcStride[1], dstStride[2]);
1728 planar2x(src[2], dst[1], c->chrSrcW, c->chrSrcH, srcStride[2], dstStride[1]);
1733 /* unscaled copy like stuff (assumes nearly identical formats) */
1734 static int simpleCopy(SwsContext *c, uint8_t* src[], int srcStride[], int srcSliceY,
1735 int srcSliceH, uint8_t* dst[], int dstStride[]){
1737 if (isPacked(c->srcFormat))
1739 if (dstStride[0]==srcStride[0] && srcStride[0] > 0)
1740 memcpy(dst[0] + dstStride[0]*srcSliceY, src[0], srcSliceH*dstStride[0]);
1744 uint8_t *srcPtr= src[0];
1745 uint8_t *dstPtr= dst[0] + dstStride[0]*srcSliceY;
1748 /* universal length finder */
1749 while(length+c->srcW <= FFABS(dstStride[0])
1750 && length+c->srcW <= FFABS(srcStride[0])) length+= c->srcW;
1753 for (i=0; i<srcSliceH; i++)
1755 memcpy(dstPtr, srcPtr, length);
1756 srcPtr+= srcStride[0];
1757 dstPtr+= dstStride[0];
1762 { /* Planar YUV or gray */
1764 for (plane=0; plane<3; plane++)
1766 int length= plane==0 ? c->srcW : -((-c->srcW )>>c->chrDstHSubSample);
1767 int y= plane==0 ? srcSliceY: -((-srcSliceY)>>c->chrDstVSubSample);
1768 int height= plane==0 ? srcSliceH: -((-srcSliceH)>>c->chrDstVSubSample);
1770 if ((isGray(c->srcFormat) || isGray(c->dstFormat)) && plane>0)
1772 if (!isGray(c->dstFormat))
1773 memset(dst[plane], 128, dstStride[plane]*height);
1777 if (dstStride[plane]==srcStride[plane] && srcStride[plane] > 0)
1778 memcpy(dst[plane] + dstStride[plane]*y, src[plane], height*dstStride[plane]);
1782 uint8_t *srcPtr= src[plane];
1783 uint8_t *dstPtr= dst[plane] + dstStride[plane]*y;
1784 for (i=0; i<height; i++)
1786 memcpy(dstPtr, srcPtr, length);
1787 srcPtr+= srcStride[plane];
1788 dstPtr+= dstStride[plane];
1797 static int gray16togray(SwsContext *c, uint8_t* src[], int srcStride[], int srcSliceY,
1798 int srcSliceH, uint8_t* dst[], int dstStride[]){
1800 int length= c->srcW;
1802 int height= srcSliceH;
1804 uint8_t *srcPtr= src[0];
1805 uint8_t *dstPtr= dst[0] + dstStride[0]*y;
1807 if (!isGray(c->dstFormat)){
1808 int height= -((-srcSliceH)>>c->chrDstVSubSample);
1809 memset(dst[1], 128, dstStride[1]*height);
1810 memset(dst[2], 128, dstStride[2]*height);
1812 if (c->srcFormat == PIX_FMT_GRAY16LE) srcPtr++;
1813 for (i=0; i<height; i++)
1815 for (j=0; j<length; j++) dstPtr[j] = srcPtr[j<<1];
1816 srcPtr+= srcStride[0];
1817 dstPtr+= dstStride[0];
1822 static int graytogray16(SwsContext *c, uint8_t* src[], int srcStride[], int srcSliceY,
1823 int srcSliceH, uint8_t* dst[], int dstStride[]){
1825 int length= c->srcW;
1827 int height= srcSliceH;
1829 uint8_t *srcPtr= src[0];
1830 uint8_t *dstPtr= dst[0] + dstStride[0]*y;
1831 for (i=0; i<height; i++)
1833 for (j=0; j<length; j++)
1835 dstPtr[j<<1] = srcPtr[j];
1836 dstPtr[(j<<1)+1] = srcPtr[j];
1838 srcPtr+= srcStride[0];
1839 dstPtr+= dstStride[0];
1844 static int gray16swap(SwsContext *c, uint8_t* src[], int srcStride[], int srcSliceY,
1845 int srcSliceH, uint8_t* dst[], int dstStride[]){
1847 int length= c->srcW;
1849 int height= srcSliceH;
1851 uint16_t *srcPtr= (uint16_t*)src[0];
1852 uint16_t *dstPtr= (uint16_t*)(dst[0] + dstStride[0]*y/2);
1853 for (i=0; i<height; i++)
1855 for (j=0; j<length; j++) dstPtr[j] = bswap_16(srcPtr[j]);
1856 srcPtr+= srcStride[0]/2;
1857 dstPtr+= dstStride[0]/2;
1863 static void getSubSampleFactors(int *h, int *v, int format){
1865 case PIX_FMT_UYVY422:
1866 case PIX_FMT_YUYV422:
1870 case PIX_FMT_YUV420P:
1871 case PIX_FMT_YUVA420P:
1872 case PIX_FMT_GRAY16BE:
1873 case PIX_FMT_GRAY16LE:
1874 case PIX_FMT_GRAY8: //FIXME remove after different subsamplings are fully implemented
1880 case PIX_FMT_YUV440P:
1884 case PIX_FMT_YUV410P:
1888 case PIX_FMT_YUV444P:
1892 case PIX_FMT_YUV422P:
1896 case PIX_FMT_YUV411P:
1907 static uint16_t roundToInt16(int64_t f){
1908 int r= (f + (1<<15))>>16;
1909 if (r<-0x7FFF) return 0x8000;
1910 else if (r> 0x7FFF) return 0x7FFF;
1915 * @param inv_table the yuv2rgb coeffs, normally Inverse_Table_6_9[x]
1916 * @param fullRange if 1 then the luma range is 0..255 if 0 it is 16..235
1917 * @return -1 if not supported
1919 int sws_setColorspaceDetails(SwsContext *c, const int inv_table[4], int srcRange, const int table[4], int dstRange, int brightness, int contrast, int saturation){
1920 int64_t crv = inv_table[0];
1921 int64_t cbu = inv_table[1];
1922 int64_t cgu = -inv_table[2];
1923 int64_t cgv = -inv_table[3];
1927 if (isYUV(c->dstFormat) || isGray(c->dstFormat)) return -1;
1928 memcpy(c->srcColorspaceTable, inv_table, sizeof(int)*4);
1929 memcpy(c->dstColorspaceTable, table, sizeof(int)*4);
1931 c->brightness= brightness;
1932 c->contrast = contrast;
1933 c->saturation= saturation;
1934 c->srcRange = srcRange;
1935 c->dstRange = dstRange;
1937 c->uOffset= 0x0400040004000400LL;
1938 c->vOffset= 0x0400040004000400LL;
1944 crv= (crv*224) / 255;
1945 cbu= (cbu*224) / 255;
1946 cgu= (cgu*224) / 255;
1947 cgv= (cgv*224) / 255;
1950 cy = (cy *contrast )>>16;
1951 crv= (crv*contrast * saturation)>>32;
1952 cbu= (cbu*contrast * saturation)>>32;
1953 cgu= (cgu*contrast * saturation)>>32;
1954 cgv= (cgv*contrast * saturation)>>32;
1956 oy -= 256*brightness;
1958 c->yCoeff= roundToInt16(cy *8192) * 0x0001000100010001ULL;
1959 c->vrCoeff= roundToInt16(crv*8192) * 0x0001000100010001ULL;
1960 c->ubCoeff= roundToInt16(cbu*8192) * 0x0001000100010001ULL;
1961 c->vgCoeff= roundToInt16(cgv*8192) * 0x0001000100010001ULL;
1962 c->ugCoeff= roundToInt16(cgu*8192) * 0x0001000100010001ULL;
1963 c->yOffset= roundToInt16(oy * 8) * 0x0001000100010001ULL;
1965 yuv2rgb_c_init_tables(c, inv_table, srcRange, brightness, contrast, saturation);
1968 #ifdef COMPILE_ALTIVEC
1969 if (c->flags & SWS_CPU_CAPS_ALTIVEC)
1970 yuv2rgb_altivec_init_tables (c, inv_table, brightness, contrast, saturation);
1976 * @return -1 if not supported
1978 int sws_getColorspaceDetails(SwsContext *c, int **inv_table, int *srcRange, int **table, int *dstRange, int *brightness, int *contrast, int *saturation){
1979 if (isYUV(c->dstFormat) || isGray(c->dstFormat)) return -1;
1981 *inv_table = c->srcColorspaceTable;
1982 *table = c->dstColorspaceTable;
1983 *srcRange = c->srcRange;
1984 *dstRange = c->dstRange;
1985 *brightness= c->brightness;
1986 *contrast = c->contrast;
1987 *saturation= c->saturation;
1992 static int handle_jpeg(int *format)
1995 case PIX_FMT_YUVJ420P:
1996 *format = PIX_FMT_YUV420P;
1998 case PIX_FMT_YUVJ422P:
1999 *format = PIX_FMT_YUV422P;
2001 case PIX_FMT_YUVJ444P:
2002 *format = PIX_FMT_YUV444P;
2004 case PIX_FMT_YUVJ440P:
2005 *format = PIX_FMT_YUV440P;
2012 SwsContext *sws_getContext(int srcW, int srcH, int srcFormat, int dstW, int dstH, int dstFormat, int flags,
2013 SwsFilter *srcFilter, SwsFilter *dstFilter, double *param){
2017 int usesVFilter, usesHFilter;
2018 int unscaled, needsDither;
2019 int srcRange, dstRange;
2020 SwsFilter dummyFilter= {NULL, NULL, NULL, NULL};
2021 #if defined(ARCH_X86)
2022 if (flags & SWS_CPU_CAPS_MMX)
2023 asm volatile("emms\n\t"::: "memory");
2026 #if !defined(RUNTIME_CPUDETECT) || !defined (CONFIG_GPL) //ensure that the flags match the compiled variant if cpudetect is off
2027 flags &= ~(SWS_CPU_CAPS_MMX|SWS_CPU_CAPS_MMX2|SWS_CPU_CAPS_3DNOW|SWS_CPU_CAPS_ALTIVEC|SWS_CPU_CAPS_BFIN);
2029 flags |= SWS_CPU_CAPS_MMX|SWS_CPU_CAPS_MMX2;
2030 #elif defined (HAVE_3DNOW)
2031 flags |= SWS_CPU_CAPS_MMX|SWS_CPU_CAPS_3DNOW;
2032 #elif defined (HAVE_MMX)
2033 flags |= SWS_CPU_CAPS_MMX;
2034 #elif defined (HAVE_ALTIVEC)
2035 flags |= SWS_CPU_CAPS_ALTIVEC;
2036 #elif defined (ARCH_BFIN)
2037 flags |= SWS_CPU_CAPS_BFIN;
2039 #endif /* RUNTIME_CPUDETECT */
2040 if (clip_table[512] != 255) globalInit();
2041 if (!rgb15to16) sws_rgb2rgb_init(flags);
2043 unscaled = (srcW == dstW && srcH == dstH);
2044 needsDither= (isBGR(dstFormat) || isRGB(dstFormat))
2045 && (fmt_depth(dstFormat))<24
2046 && ((fmt_depth(dstFormat))<(fmt_depth(srcFormat)) || (!(isRGB(srcFormat) || isBGR(srcFormat))));
2048 srcRange = handle_jpeg(&srcFormat);
2049 dstRange = handle_jpeg(&dstFormat);
2051 if (!isSupportedIn(srcFormat))
2053 av_log(NULL, AV_LOG_ERROR, "swScaler: %s is not supported as input pixel format\n", sws_format_name(srcFormat));
2056 if (!isSupportedOut(dstFormat))
2058 av_log(NULL, AV_LOG_ERROR, "swScaler: %s is not supported as output pixel format\n", sws_format_name(dstFormat));
2063 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
2065 av_log(NULL, AV_LOG_ERROR, "swScaler: %dx%d -> %dx%d is invalid scaling dimension\n",
2066 srcW, srcH, dstW, dstH);
2069 if(srcW > VOFW || dstW > VOFW){
2070 av_log(NULL, AV_LOG_ERROR, "swScaler: Compile time max width is "AV_STRINGIFY(VOFW)" change VOF/VOFW and recompile\n");
2074 if (!dstFilter) dstFilter= &dummyFilter;
2075 if (!srcFilter) srcFilter= &dummyFilter;
2077 c= av_mallocz(sizeof(SwsContext));
2079 c->av_class = &sws_context_class;
2084 c->lumXInc= ((srcW<<16) + (dstW>>1))/dstW;
2085 c->lumYInc= ((srcH<<16) + (dstH>>1))/dstH;
2087 c->dstFormat= dstFormat;
2088 c->srcFormat= srcFormat;
2089 c->vRounder= 4* 0x0001000100010001ULL;
2091 usesHFilter= usesVFilter= 0;
2092 if (dstFilter->lumV && dstFilter->lumV->length>1) usesVFilter=1;
2093 if (dstFilter->lumH && dstFilter->lumH->length>1) usesHFilter=1;
2094 if (dstFilter->chrV && dstFilter->chrV->length>1) usesVFilter=1;
2095 if (dstFilter->chrH && dstFilter->chrH->length>1) usesHFilter=1;
2096 if (srcFilter->lumV && srcFilter->lumV->length>1) usesVFilter=1;
2097 if (srcFilter->lumH && srcFilter->lumH->length>1) usesHFilter=1;
2098 if (srcFilter->chrV && srcFilter->chrV->length>1) usesVFilter=1;
2099 if (srcFilter->chrH && srcFilter->chrH->length>1) usesHFilter=1;
2101 getSubSampleFactors(&c->chrSrcHSubSample, &c->chrSrcVSubSample, srcFormat);
2102 getSubSampleFactors(&c->chrDstHSubSample, &c->chrDstVSubSample, dstFormat);
2104 // reuse chroma for 2 pixles rgb/bgr unless user wants full chroma interpolation
2105 if ((isBGR(dstFormat) || isRGB(dstFormat)) && !(flags&SWS_FULL_CHR_H_INT)) c->chrDstHSubSample=1;
2107 // drop some chroma lines if the user wants it
2108 c->vChrDrop= (flags&SWS_SRC_V_CHR_DROP_MASK)>>SWS_SRC_V_CHR_DROP_SHIFT;
2109 c->chrSrcVSubSample+= c->vChrDrop;
2111 // drop every 2. pixel for chroma calculation unless user wants full chroma
2112 if ((isBGR(srcFormat) || isRGB(srcFormat)) && !(flags&SWS_FULL_CHR_H_INP)
2113 && srcFormat!=PIX_FMT_RGB8 && srcFormat!=PIX_FMT_BGR8
2114 && srcFormat!=PIX_FMT_RGB4 && srcFormat!=PIX_FMT_BGR4
2115 && srcFormat!=PIX_FMT_RGB4_BYTE && srcFormat!=PIX_FMT_BGR4_BYTE)
2116 c->chrSrcHSubSample=1;
2119 c->param[0] = param[0];
2120 c->param[1] = param[1];
2123 c->param[1] = SWS_PARAM_DEFAULT;
2126 c->chrIntHSubSample= c->chrDstHSubSample;
2127 c->chrIntVSubSample= c->chrSrcVSubSample;
2129 // Note the -((-x)>>y) is so that we always round toward +inf.
2130 c->chrSrcW= -((-srcW) >> c->chrSrcHSubSample);
2131 c->chrSrcH= -((-srcH) >> c->chrSrcVSubSample);
2132 c->chrDstW= -((-dstW) >> c->chrDstHSubSample);
2133 c->chrDstH= -((-dstH) >> c->chrDstVSubSample);
2135 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);
2137 /* unscaled special Cases */
2138 if (unscaled && !usesHFilter && !usesVFilter)
2141 if (srcFormat == PIX_FMT_YUV420P && (dstFormat == PIX_FMT_NV12 || dstFormat == PIX_FMT_NV21))
2143 c->swScale= PlanarToNV12Wrapper;
2147 if ((srcFormat==PIX_FMT_YUV420P || srcFormat==PIX_FMT_YUV422P) && (isBGR(dstFormat) || isRGB(dstFormat)))
2149 c->swScale= yuv2rgb_get_func_ptr(c);
2153 if (srcFormat==PIX_FMT_YUV410P && dstFormat==PIX_FMT_YUV420P)
2155 c->swScale= yvu9toyv12Wrapper;
2159 if (srcFormat==PIX_FMT_BGR24 && dstFormat==PIX_FMT_YUV420P)
2160 c->swScale= bgr24toyv12Wrapper;
2162 /* rgb/bgr -> rgb/bgr (no dither needed forms) */
2163 if ( (isBGR(srcFormat) || isRGB(srcFormat))
2164 && (isBGR(dstFormat) || isRGB(dstFormat))
2165 && srcFormat != PIX_FMT_BGR8 && dstFormat != PIX_FMT_BGR8
2166 && srcFormat != PIX_FMT_RGB8 && dstFormat != PIX_FMT_RGB8
2167 && srcFormat != PIX_FMT_BGR4 && dstFormat != PIX_FMT_BGR4
2168 && srcFormat != PIX_FMT_RGB4 && dstFormat != PIX_FMT_RGB4
2169 && srcFormat != PIX_FMT_BGR4_BYTE && dstFormat != PIX_FMT_BGR4_BYTE
2170 && srcFormat != PIX_FMT_RGB4_BYTE && dstFormat != PIX_FMT_RGB4_BYTE
2171 && srcFormat != PIX_FMT_MONOBLACK && dstFormat != PIX_FMT_MONOBLACK
2173 c->swScale= rgb2rgbWrapper;
2175 /* LQ converters if -sws 0 or -sws 4*/
2176 if (c->flags&(SWS_FAST_BILINEAR|SWS_POINT)){
2177 /* rgb/bgr -> rgb/bgr (dither needed forms) */
2178 if ( (isBGR(srcFormat) || isRGB(srcFormat))
2179 && (isBGR(dstFormat) || isRGB(dstFormat))
2181 c->swScale= rgb2rgbWrapper;
2184 if (srcFormat == PIX_FMT_YUV420P &&
2185 (dstFormat == PIX_FMT_YUYV422 || dstFormat == PIX_FMT_UYVY422))
2187 if (dstFormat == PIX_FMT_YUYV422)
2188 c->swScale= PlanarToYuy2Wrapper;
2190 c->swScale= PlanarToUyvyWrapper;
2194 #ifdef COMPILE_ALTIVEC
2195 if ((c->flags & SWS_CPU_CAPS_ALTIVEC) &&
2196 ((srcFormat == PIX_FMT_YUV420P &&
2197 (dstFormat == PIX_FMT_YUYV422 || dstFormat == PIX_FMT_UYVY422)))) {
2198 // unscaled YV12 -> packed YUV, we want speed
2199 if (dstFormat == PIX_FMT_YUYV422)
2200 c->swScale= yv12toyuy2_unscaled_altivec;
2202 c->swScale= yv12touyvy_unscaled_altivec;
2207 if ( srcFormat == dstFormat
2208 || (isPlanarYUV(srcFormat) && isGray(dstFormat))
2209 || (isPlanarYUV(dstFormat) && isGray(srcFormat)))
2211 c->swScale= simpleCopy;
2214 /* gray16{le,be} conversions */
2215 if (isGray16(srcFormat) && (isPlanarYUV(dstFormat) || (dstFormat == PIX_FMT_GRAY8)))
2217 c->swScale= gray16togray;
2219 if ((isPlanarYUV(srcFormat) || (srcFormat == PIX_FMT_GRAY8)) && isGray16(dstFormat))
2221 c->swScale= graytogray16;
2223 if (srcFormat != dstFormat && isGray16(srcFormat) && isGray16(dstFormat))
2225 c->swScale= gray16swap;
2229 if (flags & SWS_CPU_CAPS_BFIN)
2230 ff_bfin_get_unscaled_swscale (c);
2234 if (flags&SWS_PRINT_INFO)
2235 av_log(c, AV_LOG_INFO, "using unscaled %s -> %s special converter\n",
2236 sws_format_name(srcFormat), sws_format_name(dstFormat));
2241 if (flags & SWS_CPU_CAPS_MMX2)
2243 c->canMMX2BeUsed= (dstW >=srcW && (dstW&31)==0 && (srcW&15)==0) ? 1 : 0;
2244 if (!c->canMMX2BeUsed && dstW >=srcW && (srcW&15)==0 && (flags&SWS_FAST_BILINEAR))
2246 if (flags&SWS_PRINT_INFO)
2247 av_log(c, AV_LOG_INFO, "output Width is not a multiple of 32 -> no MMX2 scaler\n");
2249 if (usesHFilter) c->canMMX2BeUsed=0;
2254 c->chrXInc= ((c->chrSrcW<<16) + (c->chrDstW>>1))/c->chrDstW;
2255 c->chrYInc= ((c->chrSrcH<<16) + (c->chrDstH>>1))/c->chrDstH;
2257 // match pixel 0 of the src to pixel 0 of dst and match pixel n-2 of src to pixel n-2 of dst
2258 // but only for the FAST_BILINEAR mode otherwise do correct scaling
2259 // n-2 is the last chrominance sample available
2260 // this is not perfect, but no one should notice the difference, the more correct variant
2261 // would be like the vertical one, but that would require some special code for the
2262 // first and last pixel
2263 if (flags&SWS_FAST_BILINEAR)
2265 if (c->canMMX2BeUsed)
2270 //we don't use the x86asm scaler if mmx is available
2271 else if (flags & SWS_CPU_CAPS_MMX)
2273 c->lumXInc = ((srcW-2)<<16)/(dstW-2) - 20;
2274 c->chrXInc = ((c->chrSrcW-2)<<16)/(c->chrDstW-2) - 20;
2278 /* precalculate horizontal scaler filter coefficients */
2280 const int filterAlign=
2281 (flags & SWS_CPU_CAPS_MMX) ? 4 :
2282 (flags & SWS_CPU_CAPS_ALTIVEC) ? 8 :
2285 initFilter(&c->hLumFilter, &c->hLumFilterPos, &c->hLumFilterSize, c->lumXInc,
2286 srcW , dstW, filterAlign, 1<<14,
2287 (flags&SWS_BICUBLIN) ? (flags|SWS_BICUBIC) : flags,
2288 srcFilter->lumH, dstFilter->lumH, c->param);
2289 initFilter(&c->hChrFilter, &c->hChrFilterPos, &c->hChrFilterSize, c->chrXInc,
2290 c->chrSrcW, c->chrDstW, filterAlign, 1<<14,
2291 (flags&SWS_BICUBLIN) ? (flags|SWS_BILINEAR) : flags,
2292 srcFilter->chrH, dstFilter->chrH, c->param);
2294 #define MAX_FUNNY_CODE_SIZE 10000
2295 #if defined(COMPILE_MMX2)
2296 // can't downscale !!!
2297 if (c->canMMX2BeUsed && (flags & SWS_FAST_BILINEAR))
2299 #ifdef MAP_ANONYMOUS
2300 c->funnyYCode = (uint8_t*)mmap(NULL, MAX_FUNNY_CODE_SIZE, PROT_EXEC | PROT_READ | PROT_WRITE, MAP_PRIVATE | MAP_ANONYMOUS, 0, 0);
2301 c->funnyUVCode = (uint8_t*)mmap(NULL, MAX_FUNNY_CODE_SIZE, PROT_EXEC | PROT_READ | PROT_WRITE, MAP_PRIVATE | MAP_ANONYMOUS, 0, 0);
2303 c->funnyYCode = av_malloc(MAX_FUNNY_CODE_SIZE);
2304 c->funnyUVCode = av_malloc(MAX_FUNNY_CODE_SIZE);
2307 c->lumMmx2Filter = av_malloc((dstW /8+8)*sizeof(int16_t));
2308 c->chrMmx2Filter = av_malloc((c->chrDstW /4+8)*sizeof(int16_t));
2309 c->lumMmx2FilterPos= av_malloc((dstW /2/8+8)*sizeof(int32_t));
2310 c->chrMmx2FilterPos= av_malloc((c->chrDstW/2/4+8)*sizeof(int32_t));
2312 initMMX2HScaler( dstW, c->lumXInc, c->funnyYCode , c->lumMmx2Filter, c->lumMmx2FilterPos, 8);
2313 initMMX2HScaler(c->chrDstW, c->chrXInc, c->funnyUVCode, c->chrMmx2Filter, c->chrMmx2FilterPos, 4);
2315 #endif /* defined(COMPILE_MMX2) */
2316 } // Init Horizontal stuff
2320 /* precalculate vertical scaler filter coefficients */
2322 const int filterAlign=
2323 (flags & SWS_CPU_CAPS_MMX) && (flags & SWS_ACCURATE_RND) ? 2 :
2324 (flags & SWS_CPU_CAPS_ALTIVEC) ? 8 :
2327 initFilter(&c->vLumFilter, &c->vLumFilterPos, &c->vLumFilterSize, c->lumYInc,
2328 srcH , dstH, filterAlign, (1<<12)-4,
2329 (flags&SWS_BICUBLIN) ? (flags|SWS_BICUBIC) : flags,
2330 srcFilter->lumV, dstFilter->lumV, c->param);
2331 initFilter(&c->vChrFilter, &c->vChrFilterPos, &c->vChrFilterSize, c->chrYInc,
2332 c->chrSrcH, c->chrDstH, filterAlign, (1<<12)-4,
2333 (flags&SWS_BICUBLIN) ? (flags|SWS_BILINEAR) : flags,
2334 srcFilter->chrV, dstFilter->chrV, c->param);
2337 c->vYCoeffsBank = av_malloc(sizeof (vector signed short)*c->vLumFilterSize*c->dstH);
2338 c->vCCoeffsBank = av_malloc(sizeof (vector signed short)*c->vChrFilterSize*c->chrDstH);
2340 for (i=0;i<c->vLumFilterSize*c->dstH;i++) {
2342 short *p = (short *)&c->vYCoeffsBank[i];
2344 p[j] = c->vLumFilter[i];
2347 for (i=0;i<c->vChrFilterSize*c->chrDstH;i++) {
2349 short *p = (short *)&c->vCCoeffsBank[i];
2351 p[j] = c->vChrFilter[i];
2356 // Calculate Buffer Sizes so that they won't run out while handling these damn slices
2357 c->vLumBufSize= c->vLumFilterSize;
2358 c->vChrBufSize= c->vChrFilterSize;
2359 for (i=0; i<dstH; i++)
2361 int chrI= i*c->chrDstH / dstH;
2362 int nextSlice= FFMAX(c->vLumFilterPos[i ] + c->vLumFilterSize - 1,
2363 ((c->vChrFilterPos[chrI] + c->vChrFilterSize - 1)<<c->chrSrcVSubSample));
2365 nextSlice>>= c->chrSrcVSubSample;
2366 nextSlice<<= c->chrSrcVSubSample;
2367 if (c->vLumFilterPos[i ] + c->vLumBufSize < nextSlice)
2368 c->vLumBufSize= nextSlice - c->vLumFilterPos[i];
2369 if (c->vChrFilterPos[chrI] + c->vChrBufSize < (nextSlice>>c->chrSrcVSubSample))
2370 c->vChrBufSize= (nextSlice>>c->chrSrcVSubSample) - c->vChrFilterPos[chrI];
2373 // allocate pixbufs (we use dynamic allocation because otherwise we would need to
2374 c->lumPixBuf= av_malloc(c->vLumBufSize*2*sizeof(int16_t*));
2375 c->chrPixBuf= av_malloc(c->vChrBufSize*2*sizeof(int16_t*));
2376 //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)
2377 /* align at 16 bytes for AltiVec */
2378 for (i=0; i<c->vLumBufSize; i++)
2379 c->lumPixBuf[i]= c->lumPixBuf[i+c->vLumBufSize]= av_mallocz(VOF+1);
2380 for (i=0; i<c->vChrBufSize; i++)
2381 c->chrPixBuf[i]= c->chrPixBuf[i+c->vChrBufSize]= av_malloc((VOF+1)*2);
2383 //try to avoid drawing green stuff between the right end and the stride end
2384 for (i=0; i<c->vChrBufSize; i++) memset(c->chrPixBuf[i], 64, (VOF+1)*2);
2386 assert(2*VOFW == VOF);
2388 ASSERT(c->chrDstH <= dstH)
2390 if (flags&SWS_PRINT_INFO)
2393 const char *dither= " dithered";
2395 const char *dither= "";
2397 if (flags&SWS_FAST_BILINEAR)
2398 av_log(c, AV_LOG_INFO, "FAST_BILINEAR scaler, ");
2399 else if (flags&SWS_BILINEAR)
2400 av_log(c, AV_LOG_INFO, "BILINEAR scaler, ");
2401 else if (flags&SWS_BICUBIC)
2402 av_log(c, AV_LOG_INFO, "BICUBIC scaler, ");
2403 else if (flags&SWS_X)
2404 av_log(c, AV_LOG_INFO, "Experimental scaler, ");
2405 else if (flags&SWS_POINT)
2406 av_log(c, AV_LOG_INFO, "Nearest Neighbor / POINT scaler, ");
2407 else if (flags&SWS_AREA)
2408 av_log(c, AV_LOG_INFO, "Area Averageing scaler, ");
2409 else if (flags&SWS_BICUBLIN)
2410 av_log(c, AV_LOG_INFO, "luma BICUBIC / chroma BILINEAR scaler, ");
2411 else if (flags&SWS_GAUSS)
2412 av_log(c, AV_LOG_INFO, "Gaussian scaler, ");
2413 else if (flags&SWS_SINC)
2414 av_log(c, AV_LOG_INFO, "Sinc scaler, ");
2415 else if (flags&SWS_LANCZOS)
2416 av_log(c, AV_LOG_INFO, "Lanczos scaler, ");
2417 else if (flags&SWS_SPLINE)
2418 av_log(c, AV_LOG_INFO, "Bicubic spline scaler, ");
2420 av_log(c, AV_LOG_INFO, "ehh flags invalid?! ");
2422 if (dstFormat==PIX_FMT_BGR555 || dstFormat==PIX_FMT_BGR565)
2423 av_log(c, AV_LOG_INFO, "from %s to%s %s ",
2424 sws_format_name(srcFormat), dither, sws_format_name(dstFormat));
2426 av_log(c, AV_LOG_INFO, "from %s to %s ",
2427 sws_format_name(srcFormat), sws_format_name(dstFormat));
2429 if (flags & SWS_CPU_CAPS_MMX2)
2430 av_log(c, AV_LOG_INFO, "using MMX2\n");
2431 else if (flags & SWS_CPU_CAPS_3DNOW)
2432 av_log(c, AV_LOG_INFO, "using 3DNOW\n");
2433 else if (flags & SWS_CPU_CAPS_MMX)
2434 av_log(c, AV_LOG_INFO, "using MMX\n");
2435 else if (flags & SWS_CPU_CAPS_ALTIVEC)
2436 av_log(c, AV_LOG_INFO, "using AltiVec\n");
2438 av_log(c, AV_LOG_INFO, "using C\n");
2441 if (flags & SWS_PRINT_INFO)
2443 if (flags & SWS_CPU_CAPS_MMX)
2445 if (c->canMMX2BeUsed && (flags&SWS_FAST_BILINEAR))
2446 av_log(c, AV_LOG_VERBOSE, "using FAST_BILINEAR MMX2 scaler for horizontal scaling\n");
2449 if (c->hLumFilterSize==4)
2450 av_log(c, AV_LOG_VERBOSE, "using 4-tap MMX scaler for horizontal luminance scaling\n");
2451 else if (c->hLumFilterSize==8)
2452 av_log(c, AV_LOG_VERBOSE, "using 8-tap MMX scaler for horizontal luminance scaling\n");
2454 av_log(c, AV_LOG_VERBOSE, "using n-tap MMX scaler for horizontal luminance scaling\n");
2456 if (c->hChrFilterSize==4)
2457 av_log(c, AV_LOG_VERBOSE, "using 4-tap MMX scaler for horizontal chrominance scaling\n");
2458 else if (c->hChrFilterSize==8)
2459 av_log(c, AV_LOG_VERBOSE, "using 8-tap MMX scaler for horizontal chrominance scaling\n");
2461 av_log(c, AV_LOG_VERBOSE, "using n-tap MMX scaler for horizontal chrominance scaling\n");
2466 #if defined(ARCH_X86)
2467 av_log(c, AV_LOG_VERBOSE, "using X86-Asm scaler for horizontal scaling\n");
2469 if (flags & SWS_FAST_BILINEAR)
2470 av_log(c, AV_LOG_VERBOSE, "using FAST_BILINEAR C scaler for horizontal scaling\n");
2472 av_log(c, AV_LOG_VERBOSE, "using C scaler for horizontal scaling\n");
2475 if (isPlanarYUV(dstFormat))
2477 if (c->vLumFilterSize==1)
2478 av_log(c, AV_LOG_VERBOSE, "using 1-tap %s \"scaler\" for vertical scaling (YV12 like)\n", (flags & SWS_CPU_CAPS_MMX) ? "MMX" : "C");
2480 av_log(c, AV_LOG_VERBOSE, "using n-tap %s scaler for vertical scaling (YV12 like)\n", (flags & SWS_CPU_CAPS_MMX) ? "MMX" : "C");
2484 if (c->vLumFilterSize==1 && c->vChrFilterSize==2)
2485 av_log(c, AV_LOG_VERBOSE, "using 1-tap %s \"scaler\" for vertical luminance scaling (BGR)\n"
2486 " 2-tap scaler for vertical chrominance scaling (BGR)\n", (flags & SWS_CPU_CAPS_MMX) ? "MMX" : "C");
2487 else if (c->vLumFilterSize==2 && c->vChrFilterSize==2)
2488 av_log(c, AV_LOG_VERBOSE, "using 2-tap linear %s scaler for vertical scaling (BGR)\n", (flags & SWS_CPU_CAPS_MMX) ? "MMX" : "C");
2490 av_log(c, AV_LOG_VERBOSE, "using n-tap %s scaler for vertical scaling (BGR)\n", (flags & SWS_CPU_CAPS_MMX) ? "MMX" : "C");
2493 if (dstFormat==PIX_FMT_BGR24)
2494 av_log(c, AV_LOG_VERBOSE, "using %s YV12->BGR24 Converter\n",
2495 (flags & SWS_CPU_CAPS_MMX2) ? "MMX2" : ((flags & SWS_CPU_CAPS_MMX) ? "MMX" : "C"));
2496 else if (dstFormat==PIX_FMT_RGB32)
2497 av_log(c, AV_LOG_VERBOSE, "using %s YV12->BGR32 Converter\n", (flags & SWS_CPU_CAPS_MMX) ? "MMX" : "C");
2498 else if (dstFormat==PIX_FMT_BGR565)
2499 av_log(c, AV_LOG_VERBOSE, "using %s YV12->BGR16 Converter\n", (flags & SWS_CPU_CAPS_MMX) ? "MMX" : "C");
2500 else if (dstFormat==PIX_FMT_BGR555)
2501 av_log(c, AV_LOG_VERBOSE, "using %s YV12->BGR15 Converter\n", (flags & SWS_CPU_CAPS_MMX) ? "MMX" : "C");
2503 av_log(c, AV_LOG_VERBOSE, "%dx%d -> %dx%d\n", srcW, srcH, dstW, dstH);
2505 if (flags & SWS_PRINT_INFO)
2507 av_log(c, AV_LOG_DEBUG, "Lum srcW=%d srcH=%d dstW=%d dstH=%d xInc=%d yInc=%d\n",
2508 c->srcW, c->srcH, c->dstW, c->dstH, c->lumXInc, c->lumYInc);
2509 av_log(c, AV_LOG_DEBUG, "Chr srcW=%d srcH=%d dstW=%d dstH=%d xInc=%d yInc=%d\n",
2510 c->chrSrcW, c->chrSrcH, c->chrDstW, c->chrDstH, c->chrXInc, c->chrYInc);
2513 c->swScale= getSwsFunc(flags);
2518 * swscale wrapper, so we don't need to export the SwsContext.
2519 * assumes planar YUV to be in YUV order instead of YVU
2521 int sws_scale(SwsContext *c, uint8_t* src[], int srcStride[], int srcSliceY,
2522 int srcSliceH, uint8_t* dst[], int dstStride[]){
2524 uint8_t* src2[4]= {src[0], src[1], src[2]};
2526 if (c->sliceDir == 0 && srcSliceY != 0 && srcSliceY + srcSliceH != c->srcH) {
2527 av_log(c, AV_LOG_ERROR, "Slices start in the middle!\n");
2530 if (c->sliceDir == 0) {
2531 if (srcSliceY == 0) c->sliceDir = 1; else c->sliceDir = -1;
2534 if (c->srcFormat == PIX_FMT_PAL8){
2535 for (i=0; i<256; i++){
2536 int p= ((uint32_t*)(src[1]))[i];
2537 int r= (p>>16)&0xFF;
2538 int g= (p>> 8)&0xFF;
2540 int y= av_clip_uint8(((RY*r + GY*g + BY*b)>>RGB2YUV_SHIFT) + 16 );
2541 int u= av_clip_uint8(((RU*r + GU*g + BU*b)>>RGB2YUV_SHIFT) + 128);
2542 int v= av_clip_uint8(((RV*r + GV*g + BV*b)>>RGB2YUV_SHIFT) + 128);
2543 pal[i]= y + (u<<8) + (v<<16);
2545 src2[1]= (uint8_t*)pal;
2548 // copy strides, so they can safely be modified
2549 if (c->sliceDir == 1) {
2550 // slices go from top to bottom
2551 int srcStride2[4]= {srcStride[0], srcStride[1], srcStride[2]};
2552 int dstStride2[4]= {dstStride[0], dstStride[1], dstStride[2]};
2553 return c->swScale(c, src2, srcStride2, srcSliceY, srcSliceH, dst, dstStride2);
2555 // slices go from bottom to top => we flip the image internally
2556 uint8_t* dst2[4]= {dst[0] + (c->dstH-1)*dstStride[0],
2557 dst[1] + ((c->dstH>>c->chrDstVSubSample)-1)*dstStride[1],
2558 dst[2] + ((c->dstH>>c->chrDstVSubSample)-1)*dstStride[2]};
2559 int srcStride2[4]= {-srcStride[0], -srcStride[1], -srcStride[2]};
2560 int dstStride2[4]= {-dstStride[0], -dstStride[1], -dstStride[2]};
2562 src2[0] += (srcSliceH-1)*srcStride[0];
2563 if (c->srcFormat != PIX_FMT_PAL8)
2564 src2[1] += ((srcSliceH>>c->chrSrcVSubSample)-1)*srcStride[1];
2565 src2[2] += ((srcSliceH>>c->chrSrcVSubSample)-1)*srcStride[2];
2567 return c->swScale(c, src2, srcStride2, c->srcH-srcSliceY-srcSliceH, srcSliceH, dst2, dstStride2);
2572 * swscale wrapper, so we don't need to export the SwsContext
2574 int sws_scale_ordered(SwsContext *c, uint8_t* src[], int srcStride[], int srcSliceY,
2575 int srcSliceH, uint8_t* dst[], int dstStride[]){
2576 return sws_scale(c, src, srcStride, srcSliceY, srcSliceH, dst, dstStride);
2579 SwsFilter *sws_getDefaultFilter(float lumaGBlur, float chromaGBlur,
2580 float lumaSharpen, float chromaSharpen,
2581 float chromaHShift, float chromaVShift,
2584 SwsFilter *filter= av_malloc(sizeof(SwsFilter));
2586 if (lumaGBlur!=0.0){
2587 filter->lumH= sws_getGaussianVec(lumaGBlur, 3.0);
2588 filter->lumV= sws_getGaussianVec(lumaGBlur, 3.0);
2590 filter->lumH= sws_getIdentityVec();
2591 filter->lumV= sws_getIdentityVec();
2594 if (chromaGBlur!=0.0){
2595 filter->chrH= sws_getGaussianVec(chromaGBlur, 3.0);
2596 filter->chrV= sws_getGaussianVec(chromaGBlur, 3.0);
2598 filter->chrH= sws_getIdentityVec();
2599 filter->chrV= sws_getIdentityVec();
2602 if (chromaSharpen!=0.0){
2603 SwsVector *id= sws_getIdentityVec();
2604 sws_scaleVec(filter->chrH, -chromaSharpen);
2605 sws_scaleVec(filter->chrV, -chromaSharpen);
2606 sws_addVec(filter->chrH, id);
2607 sws_addVec(filter->chrV, id);
2611 if (lumaSharpen!=0.0){
2612 SwsVector *id= sws_getIdentityVec();
2613 sws_scaleVec(filter->lumH, -lumaSharpen);
2614 sws_scaleVec(filter->lumV, -lumaSharpen);
2615 sws_addVec(filter->lumH, id);
2616 sws_addVec(filter->lumV, id);
2620 if (chromaHShift != 0.0)
2621 sws_shiftVec(filter->chrH, (int)(chromaHShift+0.5));
2623 if (chromaVShift != 0.0)
2624 sws_shiftVec(filter->chrV, (int)(chromaVShift+0.5));
2626 sws_normalizeVec(filter->chrH, 1.0);
2627 sws_normalizeVec(filter->chrV, 1.0);
2628 sws_normalizeVec(filter->lumH, 1.0);
2629 sws_normalizeVec(filter->lumV, 1.0);
2631 if (verbose) sws_printVec(filter->chrH);
2632 if (verbose) sws_printVec(filter->lumH);
2638 * returns a normalized gaussian curve used to filter stuff
2639 * quality=3 is high quality, lowwer is lowwer quality
2641 SwsVector *sws_getGaussianVec(double variance, double quality){
2642 const int length= (int)(variance*quality + 0.5) | 1;
2644 double *coeff= av_malloc(length*sizeof(double));
2645 double middle= (length-1)*0.5;
2646 SwsVector *vec= av_malloc(sizeof(SwsVector));
2649 vec->length= length;
2651 for (i=0; i<length; i++)
2653 double dist= i-middle;
2654 coeff[i]= exp(-dist*dist/(2*variance*variance)) / sqrt(2*variance*PI);
2657 sws_normalizeVec(vec, 1.0);
2662 SwsVector *sws_getConstVec(double c, int length){
2664 double *coeff= av_malloc(length*sizeof(double));
2665 SwsVector *vec= av_malloc(sizeof(SwsVector));
2668 vec->length= length;
2670 for (i=0; i<length; i++)
2677 SwsVector *sws_getIdentityVec(void){
2678 return sws_getConstVec(1.0, 1);
2681 double sws_dcVec(SwsVector *a){
2685 for (i=0; i<a->length; i++)
2691 void sws_scaleVec(SwsVector *a, double scalar){
2694 for (i=0; i<a->length; i++)
2695 a->coeff[i]*= scalar;
2698 void sws_normalizeVec(SwsVector *a, double height){
2699 sws_scaleVec(a, height/sws_dcVec(a));
2702 static SwsVector *sws_getConvVec(SwsVector *a, SwsVector *b){
2703 int length= a->length + b->length - 1;
2704 double *coeff= av_malloc(length*sizeof(double));
2706 SwsVector *vec= av_malloc(sizeof(SwsVector));
2709 vec->length= length;
2711 for (i=0; i<length; i++) coeff[i]= 0.0;
2713 for (i=0; i<a->length; i++)
2715 for (j=0; j<b->length; j++)
2717 coeff[i+j]+= a->coeff[i]*b->coeff[j];
2724 static SwsVector *sws_sumVec(SwsVector *a, SwsVector *b){
2725 int length= FFMAX(a->length, b->length);
2726 double *coeff= av_malloc(length*sizeof(double));
2728 SwsVector *vec= av_malloc(sizeof(SwsVector));
2731 vec->length= length;
2733 for (i=0; i<length; i++) coeff[i]= 0.0;
2735 for (i=0; i<a->length; i++) coeff[i + (length-1)/2 - (a->length-1)/2]+= a->coeff[i];
2736 for (i=0; i<b->length; i++) coeff[i + (length-1)/2 - (b->length-1)/2]+= b->coeff[i];
2741 static SwsVector *sws_diffVec(SwsVector *a, SwsVector *b){
2742 int length= FFMAX(a->length, b->length);
2743 double *coeff= av_malloc(length*sizeof(double));
2745 SwsVector *vec= av_malloc(sizeof(SwsVector));
2748 vec->length= length;
2750 for (i=0; i<length; i++) coeff[i]= 0.0;
2752 for (i=0; i<a->length; i++) coeff[i + (length-1)/2 - (a->length-1)/2]+= a->coeff[i];
2753 for (i=0; i<b->length; i++) coeff[i + (length-1)/2 - (b->length-1)/2]-= b->coeff[i];
2758 /* shift left / or right if "shift" is negative */
2759 static SwsVector *sws_getShiftedVec(SwsVector *a, int shift){
2760 int length= a->length + FFABS(shift)*2;
2761 double *coeff= av_malloc(length*sizeof(double));
2763 SwsVector *vec= av_malloc(sizeof(SwsVector));
2766 vec->length= length;
2768 for (i=0; i<length; i++) coeff[i]= 0.0;
2770 for (i=0; i<a->length; i++)
2772 coeff[i + (length-1)/2 - (a->length-1)/2 - shift]= a->coeff[i];
2778 void sws_shiftVec(SwsVector *a, int shift){
2779 SwsVector *shifted= sws_getShiftedVec(a, shift);
2781 a->coeff= shifted->coeff;
2782 a->length= shifted->length;
2786 void sws_addVec(SwsVector *a, SwsVector *b){
2787 SwsVector *sum= sws_sumVec(a, b);
2789 a->coeff= sum->coeff;
2790 a->length= sum->length;
2794 void sws_subVec(SwsVector *a, SwsVector *b){
2795 SwsVector *diff= sws_diffVec(a, b);
2797 a->coeff= diff->coeff;
2798 a->length= diff->length;
2802 void sws_convVec(SwsVector *a, SwsVector *b){
2803 SwsVector *conv= sws_getConvVec(a, b);
2805 a->coeff= conv->coeff;
2806 a->length= conv->length;
2810 SwsVector *sws_cloneVec(SwsVector *a){
2811 double *coeff= av_malloc(a->length*sizeof(double));
2813 SwsVector *vec= av_malloc(sizeof(SwsVector));
2816 vec->length= a->length;
2818 for (i=0; i<a->length; i++) coeff[i]= a->coeff[i];
2823 void sws_printVec(SwsVector *a){
2829 for (i=0; i<a->length; i++)
2830 if (a->coeff[i]>max) max= a->coeff[i];
2832 for (i=0; i<a->length; i++)
2833 if (a->coeff[i]<min) min= a->coeff[i];
2837 for (i=0; i<a->length; i++)
2839 int x= (int)((a->coeff[i]-min)*60.0/range +0.5);
2840 av_log(NULL, AV_LOG_DEBUG, "%1.3f ", a->coeff[i]);
2841 for (;x>0; x--) av_log(NULL, AV_LOG_DEBUG, " ");
2842 av_log(NULL, AV_LOG_DEBUG, "|\n");
2846 void sws_freeVec(SwsVector *a){
2854 void sws_freeFilter(SwsFilter *filter){
2855 if (!filter) return;
2857 if (filter->lumH) sws_freeVec(filter->lumH);
2858 if (filter->lumV) sws_freeVec(filter->lumV);
2859 if (filter->chrH) sws_freeVec(filter->chrH);
2860 if (filter->chrV) sws_freeVec(filter->chrV);
2865 void sws_freeContext(SwsContext *c){
2871 for (i=0; i<c->vLumBufSize; i++)
2873 av_free(c->lumPixBuf[i]);
2874 c->lumPixBuf[i]=NULL;
2876 av_free(c->lumPixBuf);
2882 for (i=0; i<c->vChrBufSize; i++)
2884 av_free(c->chrPixBuf[i]);
2885 c->chrPixBuf[i]=NULL;
2887 av_free(c->chrPixBuf);
2891 av_free(c->vLumFilter);
2892 c->vLumFilter = NULL;
2893 av_free(c->vChrFilter);
2894 c->vChrFilter = NULL;
2895 av_free(c->hLumFilter);
2896 c->hLumFilter = NULL;
2897 av_free(c->hChrFilter);
2898 c->hChrFilter = NULL;
2900 av_free(c->vYCoeffsBank);
2901 c->vYCoeffsBank = NULL;
2902 av_free(c->vCCoeffsBank);
2903 c->vCCoeffsBank = NULL;
2906 av_free(c->vLumFilterPos);
2907 c->vLumFilterPos = NULL;
2908 av_free(c->vChrFilterPos);
2909 c->vChrFilterPos = NULL;
2910 av_free(c->hLumFilterPos);
2911 c->hLumFilterPos = NULL;
2912 av_free(c->hChrFilterPos);
2913 c->hChrFilterPos = NULL;
2915 #if defined(ARCH_X86) && defined(CONFIG_GPL)
2916 #ifdef MAP_ANONYMOUS
2917 if (c->funnyYCode) munmap(c->funnyYCode, MAX_FUNNY_CODE_SIZE);
2918 if (c->funnyUVCode) munmap(c->funnyUVCode, MAX_FUNNY_CODE_SIZE);
2920 av_free(c->funnyYCode);
2921 av_free(c->funnyUVCode);
2924 c->funnyUVCode=NULL;
2925 #endif /* defined(ARCH_X86) */
2927 av_free(c->lumMmx2Filter);
2928 c->lumMmx2Filter=NULL;
2929 av_free(c->chrMmx2Filter);
2930 c->chrMmx2Filter=NULL;
2931 av_free(c->lumMmx2FilterPos);
2932 c->lumMmx2FilterPos=NULL;
2933 av_free(c->chrMmx2FilterPos);
2934 c->chrMmx2FilterPos=NULL;
2935 av_free(c->yuvTable);
2942 * Checks if context is valid or reallocs a new one instead.
2943 * If context is NULL, just calls sws_getContext() to get a new one.
2944 * Otherwise, checks if the parameters are the same already saved in context.
2945 * If that is the case, returns the current context.
2946 * Otherwise, frees context and gets a new one.
2948 * Be warned that srcFilter, dstFilter are not checked, they are
2949 * asumed to remain valid.
2951 struct SwsContext *sws_getCachedContext(struct SwsContext *context,
2952 int srcW, int srcH, int srcFormat,
2953 int dstW, int dstH, int dstFormat, int flags,
2954 SwsFilter *srcFilter, SwsFilter *dstFilter, double *param)
2956 static const double default_param[2] = {SWS_PARAM_DEFAULT, SWS_PARAM_DEFAULT};
2959 param = default_param;
2962 if (context->srcW != srcW || context->srcH != srcH ||
2963 context->srcFormat != srcFormat ||
2964 context->dstW != dstW || context->dstH != dstH ||
2965 context->dstFormat != dstFormat || context->flags != flags ||
2966 context->param[0] != param[0] || context->param[1] != param[1])
2968 sws_freeContext(context);
2973 return sws_getContext(srcW, srcH, srcFormat,
2974 dstW, dstH, dstFormat, flags,
2975 srcFilter, dstFilter, param);