2 Copyright (C) 2001-2003 Michael Niedermayer <michaelni@gmx.at>
4 This program is free software; you can redistribute it and/or modify
5 it under the terms of the GNU General Public License as published by
6 the Free Software Foundation; either version 2 of the License, or
7 (at your option) any later version.
9 This program is distributed in the hope that it will be useful,
10 but WITHOUT ANY WARRANTY; without even the implied warranty of
11 MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
12 GNU General Public License for more details.
14 You should have received a copy of the GNU General Public License
15 along with this program; if not, write to the Free Software
16 Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA 02111-1307 USA
20 supported Input formats: YV12, I420/IYUV, YUY2, UYVY, BGR32, BGR24, BGR16, BGR15, RGB32, RGB24, Y8/Y800, YVU9/IF09
21 supported output formats: YV12, I420/IYUV, YUY2, UYVY, {BGR,RGB}{1,4,8,15,16,24,32}, Y8/Y800, YVU9/IF09
22 {BGR,RGB}{1,4,8,15,16} support dithering
24 unscaled special converters (YV12=I420=IYUV, Y800=Y8)
25 YV12 -> {BGR,RGB}{1,4,8,15,16,24,32}
30 BGR24 -> BGR32 & RGB24 -> RGB32
31 BGR32 -> BGR24 & RGB32 -> RGB24
36 tested special converters (most are tested actually but i didnt write it down ...)
43 untested special converters
44 YV12/I420 -> BGR15/BGR24/BGR32 (its the yuv2rgb stuff, so it should be ok)
45 YV12/I420 -> YV12/I420
46 YUY2/BGR15/BGR24/BGR32/RGB24/RGB32 -> same format
47 BGR24 -> BGR32 & RGB24 -> RGB32
48 BGR32 -> BGR24 & RGB32 -> RGB24
64 #include "swscale_internal.h"
67 #define RUNTIME_CPUDETECT 1
76 //#define WORDS_BIGENDIAN
79 #define FAST_BGR2YV12 // use 7 bit coeffs instead of 15bit
81 #define RET 0xC3 //near return opcode for X86
84 #define ASSERT(x) assert(x);
92 #define PI 3.14159265358979323846
95 //FIXME replace this with something faster
96 #define isPlanarYUV(x) ((x)==IMGFMT_YV12 || (x)==IMGFMT_YVU9 \
97 || (x)==IMGFMT_444P || (x)==IMGFMT_422P || (x)==IMGFMT_411P)
98 #define isYUV(x) ((x)==IMGFMT_UYVY || (x)==IMGFMT_YUY2 || isPlanarYUV(x))
99 #define isGray(x) ((x)==IMGFMT_Y800)
100 #define isRGB(x) (((x)&IMGFMT_RGB_MASK)==IMGFMT_RGB)
101 #define isBGR(x) (((x)&IMGFMT_BGR_MASK)==IMGFMT_BGR)
102 #define isSupportedIn(x) ((x)==IMGFMT_YV12 || (x)==IMGFMT_YUY2 || (x)==IMGFMT_UYVY\
103 || (x)==IMGFMT_BGR32|| (x)==IMGFMT_BGR24|| (x)==IMGFMT_BGR16|| (x)==IMGFMT_BGR15\
104 || (x)==IMGFMT_RGB32|| (x)==IMGFMT_RGB24\
105 || (x)==IMGFMT_Y800 || (x)==IMGFMT_YVU9\
106 || (x)==IMGFMT_444P || (x)==IMGFMT_422P || (x)==IMGFMT_411P)
107 #define isSupportedOut(x) ((x)==IMGFMT_YV12 || (x)==IMGFMT_YUY2 || (x)==IMGFMT_UYVY\
108 || (x)==IMGFMT_444P || (x)==IMGFMT_422P || (x)==IMGFMT_411P\
109 || isRGB(x) || isBGR(x)\
110 || (x)==IMGFMT_Y800 || (x)==IMGFMT_YVU9)
111 #define isPacked(x) ((x)==IMGFMT_YUY2 || (x)==IMGFMT_UYVY ||isRGB(x) || isBGR(x))
113 #define RGB2YUV_SHIFT 16
114 #define BY ((int)( 0.098*(1<<RGB2YUV_SHIFT)+0.5))
115 #define BV ((int)(-0.071*(1<<RGB2YUV_SHIFT)+0.5))
116 #define BU ((int)( 0.439*(1<<RGB2YUV_SHIFT)+0.5))
117 #define GY ((int)( 0.504*(1<<RGB2YUV_SHIFT)+0.5))
118 #define GV ((int)(-0.368*(1<<RGB2YUV_SHIFT)+0.5))
119 #define GU ((int)(-0.291*(1<<RGB2YUV_SHIFT)+0.5))
120 #define RY ((int)( 0.257*(1<<RGB2YUV_SHIFT)+0.5))
121 #define RV ((int)( 0.439*(1<<RGB2YUV_SHIFT)+0.5))
122 #define RU ((int)(-0.148*(1<<RGB2YUV_SHIFT)+0.5))
124 extern const int32_t Inverse_Table_6_9[8][4];
128 Special versions: fast Y 1:1 scaling (no interpolation in y direction)
131 more intelligent missalignment avoidance for the horizontal scaler
132 write special vertical cubic upscale version
133 Optimize C code (yv12 / minmax)
134 add support for packed pixel yuv input & output
135 add support for Y8 output
136 optimize bgr24 & bgr32
137 add BGR4 output support
138 write special BGR->BGR scaler
141 #define ABS(a) ((a) > 0 ? (a) : (-(a)))
142 #define MIN(a,b) ((a) > (b) ? (b) : (a))
143 #define MAX(a,b) ((a) < (b) ? (b) : (a))
146 static uint64_t attribute_used __attribute__((aligned(8))) bF8= 0xF8F8F8F8F8F8F8F8LL;
147 static uint64_t attribute_used __attribute__((aligned(8))) bFC= 0xFCFCFCFCFCFCFCFCLL;
148 static uint64_t __attribute__((aligned(8))) w10= 0x0010001000100010LL;
149 static uint64_t attribute_used __attribute__((aligned(8))) w02= 0x0002000200020002LL;
150 static uint64_t attribute_used __attribute__((aligned(8))) bm00001111=0x00000000FFFFFFFFLL;
151 static uint64_t attribute_used __attribute__((aligned(8))) bm00000111=0x0000000000FFFFFFLL;
152 static uint64_t attribute_used __attribute__((aligned(8))) bm11111000=0xFFFFFFFFFF000000LL;
153 static uint64_t attribute_used __attribute__((aligned(8))) bm01010101=0x00FF00FF00FF00FFLL;
155 static volatile uint64_t attribute_used __attribute__((aligned(8))) b5Dither;
156 static volatile uint64_t attribute_used __attribute__((aligned(8))) g5Dither;
157 static volatile uint64_t attribute_used __attribute__((aligned(8))) g6Dither;
158 static volatile uint64_t attribute_used __attribute__((aligned(8))) r5Dither;
160 static uint64_t __attribute__((aligned(8))) dither4[2]={
161 0x0103010301030103LL,
162 0x0200020002000200LL,};
164 static uint64_t __attribute__((aligned(8))) dither8[2]={
165 0x0602060206020602LL,
166 0x0004000400040004LL,};
168 static uint64_t __attribute__((aligned(8))) b16Mask= 0x001F001F001F001FLL;
169 static uint64_t attribute_used __attribute__((aligned(8))) g16Mask= 0x07E007E007E007E0LL;
170 static uint64_t attribute_used __attribute__((aligned(8))) r16Mask= 0xF800F800F800F800LL;
171 static uint64_t __attribute__((aligned(8))) b15Mask= 0x001F001F001F001FLL;
172 static uint64_t attribute_used __attribute__((aligned(8))) g15Mask= 0x03E003E003E003E0LL;
173 static uint64_t attribute_used __attribute__((aligned(8))) r15Mask= 0x7C007C007C007C00LL;
175 static uint64_t attribute_used __attribute__((aligned(8))) M24A= 0x00FF0000FF0000FFLL;
176 static uint64_t attribute_used __attribute__((aligned(8))) M24B= 0xFF0000FF0000FF00LL;
177 static uint64_t attribute_used __attribute__((aligned(8))) M24C= 0x0000FF0000FF0000LL;
180 static const uint64_t bgr2YCoeff attribute_used __attribute__((aligned(8))) = 0x000000210041000DULL;
181 static const uint64_t bgr2UCoeff attribute_used __attribute__((aligned(8))) = 0x0000FFEEFFDC0038ULL;
182 static const uint64_t bgr2VCoeff attribute_used __attribute__((aligned(8))) = 0x00000038FFD2FFF8ULL;
184 static const uint64_t bgr2YCoeff attribute_used __attribute__((aligned(8))) = 0x000020E540830C8BULL;
185 static const uint64_t bgr2UCoeff attribute_used __attribute__((aligned(8))) = 0x0000ED0FDAC23831ULL;
186 static const uint64_t bgr2VCoeff attribute_used __attribute__((aligned(8))) = 0x00003831D0E6F6EAULL;
188 static const uint64_t bgr2YOffset attribute_used __attribute__((aligned(8))) = 0x1010101010101010ULL;
189 static const uint64_t bgr2UVOffset attribute_used __attribute__((aligned(8)))= 0x8080808080808080ULL;
190 static const uint64_t w1111 attribute_used __attribute__((aligned(8))) = 0x0001000100010001ULL;
193 // clipping helper table for C implementations:
194 static unsigned char clip_table[768];
196 static SwsVector *sws_getConvVec(SwsVector *a, SwsVector *b);
198 extern const uint8_t dither_2x2_4[2][8];
199 extern const uint8_t dither_2x2_8[2][8];
200 extern const uint8_t dither_8x8_32[8][8];
201 extern const uint8_t dither_8x8_73[8][8];
202 extern const uint8_t dither_8x8_220[8][8];
205 void in_asm_used_var_warning_killer()
207 volatile int i= bF8+bFC+w10+
208 bm00001111+bm00000111+bm11111000+b16Mask+g16Mask+r16Mask+b15Mask+g15Mask+r15Mask+
209 M24A+M24B+M24C+w02 + b5Dither+g5Dither+r5Dither+g6Dither+dither4[0]+dither8[0]+bm01010101;
214 static inline void yuv2yuvXinC(int16_t *lumFilter, int16_t **lumSrc, int lumFilterSize,
215 int16_t *chrFilter, int16_t **chrSrc, int chrFilterSize,
216 uint8_t *dest, uint8_t *uDest, uint8_t *vDest, int dstW, int chrDstW)
218 //FIXME Optimize (just quickly writen not opti..)
220 for(i=0; i<dstW; i++)
224 for(j=0; j<lumFilterSize; j++)
225 val += lumSrc[j][i] * lumFilter[j];
227 dest[i]= MIN(MAX(val>>19, 0), 255);
231 for(i=0; i<chrDstW; i++)
236 for(j=0; j<chrFilterSize; j++)
238 u += chrSrc[j][i] * chrFilter[j];
239 v += chrSrc[j][i + 2048] * chrFilter[j];
242 uDest[i]= MIN(MAX(u>>19, 0), 255);
243 vDest[i]= MIN(MAX(v>>19, 0), 255);
248 #define YSCALE_YUV_2_PACKEDX_C(type) \
249 for(i=0; i<(dstW>>1); i++){\
258 for(j=0; j<lumFilterSize; j++)\
260 Y1 += lumSrc[j][i2] * lumFilter[j];\
261 Y2 += lumSrc[j][i2+1] * lumFilter[j];\
263 for(j=0; j<chrFilterSize; j++)\
265 U += chrSrc[j][i] * chrFilter[j];\
266 V += chrSrc[j][i+2048] * chrFilter[j];\
284 #define YSCALE_YUV_2_RGBX_C(type) \
285 YSCALE_YUV_2_PACKEDX_C(type)\
287 g = c->table_gU[U] + c->table_gV[V];\
290 #define YSCALE_YUV_2_PACKED2_C \
291 for(i=0; i<(dstW>>1); i++){\
293 int Y1= (buf0[i2 ]*yalpha1+buf1[i2 ]*yalpha)>>19;\
294 int Y2= (buf0[i2+1]*yalpha1+buf1[i2+1]*yalpha)>>19;\
295 int U= (uvbuf0[i ]*uvalpha1+uvbuf1[i ]*uvalpha)>>19;\
296 int V= (uvbuf0[i+2048]*uvalpha1+uvbuf1[i+2048]*uvalpha)>>19;\
298 #define YSCALE_YUV_2_RGB2_C(type) \
299 YSCALE_YUV_2_PACKED2_C\
302 g = c->table_gU[U] + c->table_gV[V];\
305 #define YSCALE_YUV_2_PACKED1_C \
306 for(i=0; i<(dstW>>1); i++){\
308 int Y1= buf0[i2 ]>>7;\
309 int Y2= buf0[i2+1]>>7;\
310 int U= (uvbuf1[i ])>>7;\
311 int V= (uvbuf1[i+2048])>>7;\
313 #define YSCALE_YUV_2_RGB1_C(type) \
314 YSCALE_YUV_2_PACKED1_C\
317 g = c->table_gU[U] + c->table_gV[V];\
320 #define YSCALE_YUV_2_PACKED1B_C \
321 for(i=0; i<(dstW>>1); i++){\
323 int Y1= buf0[i2 ]>>7;\
324 int Y2= buf0[i2+1]>>7;\
325 int U= (uvbuf0[i ] + uvbuf1[i ])>>8;\
326 int V= (uvbuf0[i+2048] + uvbuf1[i+2048])>>8;\
328 #define YSCALE_YUV_2_RGB1B_C(type) \
329 YSCALE_YUV_2_PACKED1B_C\
332 g = c->table_gU[U] + c->table_gV[V];\
335 #define YSCALE_YUV_2_ANYRGB_C(func, func2)\
336 switch(c->dstFormat)\
341 ((uint32_t*)dest)[i2+0]= r[Y1] + g[Y1] + b[Y1];\
342 ((uint32_t*)dest)[i2+1]= r[Y2] + g[Y2] + b[Y2];\
347 ((uint8_t*)dest)[0]= r[Y1];\
348 ((uint8_t*)dest)[1]= g[Y1];\
349 ((uint8_t*)dest)[2]= b[Y1];\
350 ((uint8_t*)dest)[3]= r[Y2];\
351 ((uint8_t*)dest)[4]= g[Y2];\
352 ((uint8_t*)dest)[5]= b[Y2];\
358 ((uint8_t*)dest)[0]= b[Y1];\
359 ((uint8_t*)dest)[1]= g[Y1];\
360 ((uint8_t*)dest)[2]= r[Y1];\
361 ((uint8_t*)dest)[3]= b[Y2];\
362 ((uint8_t*)dest)[4]= g[Y2];\
363 ((uint8_t*)dest)[5]= r[Y2];\
370 const int dr1= dither_2x2_8[y&1 ][0];\
371 const int dg1= dither_2x2_4[y&1 ][0];\
372 const int db1= dither_2x2_8[(y&1)^1][0];\
373 const int dr2= dither_2x2_8[y&1 ][1];\
374 const int dg2= dither_2x2_4[y&1 ][1];\
375 const int db2= dither_2x2_8[(y&1)^1][1];\
377 ((uint16_t*)dest)[i2+0]= r[Y1+dr1] + g[Y1+dg1] + b[Y1+db1];\
378 ((uint16_t*)dest)[i2+1]= r[Y2+dr2] + g[Y2+dg2] + b[Y2+db2];\
385 const int dr1= dither_2x2_8[y&1 ][0];\
386 const int dg1= dither_2x2_8[y&1 ][1];\
387 const int db1= dither_2x2_8[(y&1)^1][0];\
388 const int dr2= dither_2x2_8[y&1 ][1];\
389 const int dg2= dither_2x2_8[y&1 ][0];\
390 const int db2= dither_2x2_8[(y&1)^1][1];\
392 ((uint16_t*)dest)[i2+0]= r[Y1+dr1] + g[Y1+dg1] + b[Y1+db1];\
393 ((uint16_t*)dest)[i2+1]= r[Y2+dr2] + g[Y2+dg2] + b[Y2+db2];\
400 const uint8_t * const d64= dither_8x8_73[y&7];\
401 const uint8_t * const d32= dither_8x8_32[y&7];\
403 ((uint8_t*)dest)[i2+0]= r[Y1+d32[(i2+0)&7]] + g[Y1+d32[(i2+0)&7]] + b[Y1+d64[(i2+0)&7]];\
404 ((uint8_t*)dest)[i2+1]= r[Y2+d32[(i2+1)&7]] + g[Y2+d32[(i2+1)&7]] + b[Y2+d64[(i2+1)&7]];\
411 const uint8_t * const d64= dither_8x8_73 [y&7];\
412 const uint8_t * const d128=dither_8x8_220[y&7];\
414 ((uint8_t*)dest)[i]= r[Y1+d128[(i2+0)&7]] + g[Y1+d64[(i2+0)&7]] + b[Y1+d128[(i2+0)&7]]\
415 + ((r[Y2+d128[(i2+1)&7]] + g[Y2+d64[(i2+1)&7]] + b[Y2+d128[(i2+1)&7]])<<4);\
422 const uint8_t * const d64= dither_8x8_73 [y&7];\
423 const uint8_t * const d128=dither_8x8_220[y&7];\
425 ((uint8_t*)dest)[i2+0]= r[Y1+d128[(i2+0)&7]] + g[Y1+d64[(i2+0)&7]] + b[Y1+d128[(i2+0)&7]];\
426 ((uint8_t*)dest)[i2+1]= r[Y2+d128[(i2+1)&7]] + g[Y2+d64[(i2+1)&7]] + b[Y2+d128[(i2+1)&7]];\
433 const uint8_t * const d128=dither_8x8_220[y&7];\
434 uint8_t *g= c->table_gU[128] + c->table_gV[128];\
435 for(i=0; i<dstW-7; i+=8){\
437 acc = g[((buf0[i ]*yalpha1+buf1[i ]*yalpha)>>19) + d128[0]];\
438 acc+= acc + g[((buf0[i+1]*yalpha1+buf1[i+1]*yalpha)>>19) + d128[1]];\
439 acc+= acc + g[((buf0[i+2]*yalpha1+buf1[i+2]*yalpha)>>19) + d128[2]];\
440 acc+= acc + g[((buf0[i+3]*yalpha1+buf1[i+3]*yalpha)>>19) + d128[3]];\
441 acc+= acc + g[((buf0[i+4]*yalpha1+buf1[i+4]*yalpha)>>19) + d128[4]];\
442 acc+= acc + g[((buf0[i+5]*yalpha1+buf1[i+5]*yalpha)>>19) + d128[5]];\
443 acc+= acc + g[((buf0[i+6]*yalpha1+buf1[i+6]*yalpha)>>19) + d128[6]];\
444 acc+= acc + g[((buf0[i+7]*yalpha1+buf1[i+7]*yalpha)>>19) + d128[7]];\
445 ((uint8_t*)dest)[0]= acc;\
450 ((uint8_t*)dest)-= dstW>>4;\
454 static int top[1024];\
455 static int last_new[1024][1024];\
456 static int last_in3[1024][1024];\
457 static int drift[1024][1024];\
461 const uint8_t * const d128=dither_8x8_220[y&7];\
466 for(i=dstW>>1; i<dstW; i++){\
467 int in= ((buf0[i ]*yalpha1+buf1[i ]*yalpha)>>19);\
468 int in2 = (76309 * (in - 16) + 32768) >> 16;\
469 int in3 = (in2 < 0) ? 0 : ((in2 > 255) ? 255 : in2);\
470 int old= (left*7 + topLeft + top[i]*5 + top[i+1]*3)/20 + in3\
471 + (last_new[y][i] - in3)*f/256;\
472 int new= old> 128 ? 255 : 0;\
474 error_new+= ABS(last_new[y][i] - new);\
475 error_in3+= ABS(last_in3[y][i] - in3);\
476 f= error_new - error_in3*4;\
481 left= top[i]= old - new;\
482 last_new[y][i]= new;\
483 last_in3[y][i]= in3;\
485 acc+= acc + (new&1);\
487 ((uint8_t*)dest)[0]= acc;\
497 ((uint8_t*)dest)[2*i2+0]= Y1;\
498 ((uint8_t*)dest)[2*i2+1]= U;\
499 ((uint8_t*)dest)[2*i2+2]= Y2;\
500 ((uint8_t*)dest)[2*i2+3]= V;\
505 ((uint8_t*)dest)[2*i2+0]= U;\
506 ((uint8_t*)dest)[2*i2+1]= Y1;\
507 ((uint8_t*)dest)[2*i2+2]= V;\
508 ((uint8_t*)dest)[2*i2+3]= Y2;\
514 static inline void yuv2packedXinC(SwsContext *c, int16_t *lumFilter, int16_t **lumSrc, int lumFilterSize,
515 int16_t *chrFilter, int16_t **chrSrc, int chrFilterSize,
516 uint8_t *dest, int dstW, int y)
523 YSCALE_YUV_2_RGBX_C(uint32_t)
524 ((uint32_t*)dest)[i2+0]= r[Y1] + g[Y1] + b[Y1];
525 ((uint32_t*)dest)[i2+1]= r[Y2] + g[Y2] + b[Y2];
529 YSCALE_YUV_2_RGBX_C(uint8_t)
530 ((uint8_t*)dest)[0]= r[Y1];
531 ((uint8_t*)dest)[1]= g[Y1];
532 ((uint8_t*)dest)[2]= b[Y1];
533 ((uint8_t*)dest)[3]= r[Y2];
534 ((uint8_t*)dest)[4]= g[Y2];
535 ((uint8_t*)dest)[5]= b[Y2];
540 YSCALE_YUV_2_RGBX_C(uint8_t)
541 ((uint8_t*)dest)[0]= b[Y1];
542 ((uint8_t*)dest)[1]= g[Y1];
543 ((uint8_t*)dest)[2]= r[Y1];
544 ((uint8_t*)dest)[3]= b[Y2];
545 ((uint8_t*)dest)[4]= g[Y2];
546 ((uint8_t*)dest)[5]= r[Y2];
553 const int dr1= dither_2x2_8[y&1 ][0];
554 const int dg1= dither_2x2_4[y&1 ][0];
555 const int db1= dither_2x2_8[(y&1)^1][0];
556 const int dr2= dither_2x2_8[y&1 ][1];
557 const int dg2= dither_2x2_4[y&1 ][1];
558 const int db2= dither_2x2_8[(y&1)^1][1];
559 YSCALE_YUV_2_RGBX_C(uint16_t)
560 ((uint16_t*)dest)[i2+0]= r[Y1+dr1] + g[Y1+dg1] + b[Y1+db1];
561 ((uint16_t*)dest)[i2+1]= r[Y2+dr2] + g[Y2+dg2] + b[Y2+db2];
568 const int dr1= dither_2x2_8[y&1 ][0];
569 const int dg1= dither_2x2_8[y&1 ][1];
570 const int db1= dither_2x2_8[(y&1)^1][0];
571 const int dr2= dither_2x2_8[y&1 ][1];
572 const int dg2= dither_2x2_8[y&1 ][0];
573 const int db2= dither_2x2_8[(y&1)^1][1];
574 YSCALE_YUV_2_RGBX_C(uint16_t)
575 ((uint16_t*)dest)[i2+0]= r[Y1+dr1] + g[Y1+dg1] + b[Y1+db1];
576 ((uint16_t*)dest)[i2+1]= r[Y2+dr2] + g[Y2+dg2] + b[Y2+db2];
583 const uint8_t * const d64= dither_8x8_73[y&7];
584 const uint8_t * const d32= dither_8x8_32[y&7];
585 YSCALE_YUV_2_RGBX_C(uint8_t)
586 ((uint8_t*)dest)[i2+0]= r[Y1+d32[(i2+0)&7]] + g[Y1+d32[(i2+0)&7]] + b[Y1+d64[(i2+0)&7]];
587 ((uint8_t*)dest)[i2+1]= r[Y2+d32[(i2+1)&7]] + g[Y2+d32[(i2+1)&7]] + b[Y2+d64[(i2+1)&7]];
594 const uint8_t * const d64= dither_8x8_73 [y&7];
595 const uint8_t * const d128=dither_8x8_220[y&7];
596 YSCALE_YUV_2_RGBX_C(uint8_t)
597 ((uint8_t*)dest)[i]= r[Y1+d128[(i2+0)&7]] + g[Y1+d64[(i2+0)&7]] + b[Y1+d128[(i2+0)&7]]
598 +((r[Y2+d128[(i2+1)&7]] + g[Y2+d64[(i2+1)&7]] + b[Y2+d128[(i2+1)&7]])<<4);
605 const uint8_t * const d64= dither_8x8_73 [y&7];
606 const uint8_t * const d128=dither_8x8_220[y&7];
607 YSCALE_YUV_2_RGBX_C(uint8_t)
608 ((uint8_t*)dest)[i2+0]= r[Y1+d128[(i2+0)&7]] + g[Y1+d64[(i2+0)&7]] + b[Y1+d128[(i2+0)&7]];
609 ((uint8_t*)dest)[i2+1]= r[Y2+d128[(i2+1)&7]] + g[Y2+d64[(i2+1)&7]] + b[Y2+d128[(i2+1)&7]];
616 const uint8_t * const d128=dither_8x8_220[y&7];
617 uint8_t *g= c->table_gU[128] + c->table_gV[128];
619 for(i=0; i<dstW-1; i+=2){
624 for(j=0; j<lumFilterSize; j++)
626 Y1 += lumSrc[j][i] * lumFilter[j];
627 Y2 += lumSrc[j][i+1] * lumFilter[j];
638 acc+= acc + g[Y1+d128[(i+0)&7]];
639 acc+= acc + g[Y2+d128[(i+1)&7]];
641 ((uint8_t*)dest)[0]= acc;
648 YSCALE_YUV_2_PACKEDX_C(void)
649 ((uint8_t*)dest)[2*i2+0]= Y1;
650 ((uint8_t*)dest)[2*i2+1]= U;
651 ((uint8_t*)dest)[2*i2+2]= Y2;
652 ((uint8_t*)dest)[2*i2+3]= V;
656 YSCALE_YUV_2_PACKEDX_C(void)
657 ((uint8_t*)dest)[2*i2+0]= U;
658 ((uint8_t*)dest)[2*i2+1]= Y1;
659 ((uint8_t*)dest)[2*i2+2]= V;
660 ((uint8_t*)dest)[2*i2+3]= Y2;
667 //Note: we have C, X86, MMX, MMX2, 3DNOW version therse no 3DNOW+MMX2 one
669 #if !defined (HAVE_MMX) || defined (RUNTIME_CPUDETECT)
675 #define COMPILE_ALTIVEC
676 #endif //HAVE_ALTIVEC
677 #endif //ARCH_POWERPC
681 #if (defined (HAVE_MMX) && !defined (HAVE_3DNOW) && !defined (HAVE_MMX2)) || defined (RUNTIME_CPUDETECT)
685 #if defined (HAVE_MMX2) || defined (RUNTIME_CPUDETECT)
689 #if (defined (HAVE_3DNOW) && !defined (HAVE_MMX2)) || defined (RUNTIME_CPUDETECT)
690 #define COMPILE_3DNOW
703 #define RENAME(a) a ## _C
704 #include "swscale_template.c"
708 #ifdef COMPILE_ALTIVEC
711 #define RENAME(a) a ## _altivec
712 #include "swscale_template.c"
714 #endif //ARCH_POWERPC
725 #define RENAME(a) a ## _X86
726 #include "swscale_template.c"
734 #define RENAME(a) a ## _MMX
735 #include "swscale_template.c"
744 #define RENAME(a) a ## _MMX2
745 #include "swscale_template.c"
754 #define RENAME(a) a ## _3DNow
755 #include "swscale_template.c"
760 // minor note: the HAVE_xyz is messed up after that line so don't use it
762 static double getSplineCoeff(double a, double b, double c, double d, double dist)
764 // printf("%f %f %f %f %f\n", a,b,c,d,dist);
765 if(dist<=1.0) return ((d*dist + c)*dist + b)*dist +a;
766 else return getSplineCoeff( 0.0,
773 static inline void initFilter(int16_t **outFilter, int16_t **filterPos, int *outFilterSize, int xInc,
774 int srcW, int dstW, int filterAlign, int one, int flags,
775 SwsVector *srcFilter, SwsVector *dstFilter)
782 double *filter2=NULL;
784 if(flags & SWS_CPU_CAPS_MMX)
785 asm volatile("emms\n\t"::: "memory"); //FIXME this shouldnt be required but it IS (even for non mmx versions)
788 // Note the +1 is for the MMXscaler which reads over the end
789 *filterPos = (int16_t*)memalign(8, (dstW+1)*sizeof(int16_t));
791 if(ABS(xInc - 0x10000) <10) // unscaled
795 filter= (double*)memalign(8, dstW*sizeof(double)*filterSize);
796 for(i=0; i<dstW*filterSize; i++) filter[i]=0;
798 for(i=0; i<dstW; i++)
800 filter[i*filterSize]=1;
805 else if(flags&SWS_POINT) // lame looking point sampling mode
810 filter= (double*)memalign(8, dstW*sizeof(double)*filterSize);
812 xDstInSrc= xInc/2 - 0x8000;
813 for(i=0; i<dstW; i++)
815 int xx= (xDstInSrc - ((filterSize-1)<<15) + (1<<15))>>16;
822 else if((xInc <= (1<<16) && (flags&SWS_AREA)) || (flags&SWS_FAST_BILINEAR)) // bilinear upscale
826 if (flags&SWS_BICUBIC) filterSize= 4;
827 else if(flags&SWS_X ) filterSize= 4;
828 else filterSize= 2; // SWS_BILINEAR / SWS_AREA
829 filter= (double*)memalign(8, dstW*sizeof(double)*filterSize);
831 xDstInSrc= xInc/2 - 0x8000;
832 for(i=0; i<dstW; i++)
834 int xx= (xDstInSrc - ((filterSize-1)<<15) + (1<<15))>>16;
838 //Bilinear upscale / linear interpolate / Area averaging
839 for(j=0; j<filterSize; j++)
841 double d= ABS((xx<<16) - xDstInSrc)/(double)(1<<16);
842 double coeff= 1.0 - d;
844 filter[i*filterSize + j]= coeff;
853 double sizeFactor, filterSizeInSrc;
854 const double xInc1= (double)xInc / (double)(1<<16);
855 int param= (flags&SWS_PARAM_MASK)>>SWS_PARAM_SHIFT;
857 if (flags&SWS_BICUBIC) sizeFactor= 4.0;
858 else if(flags&SWS_X) sizeFactor= 8.0;
859 else if(flags&SWS_AREA) sizeFactor= 1.0; //downscale only, for upscale it is bilinear
860 else if(flags&SWS_GAUSS) sizeFactor= 8.0; // infinite ;)
861 else if(flags&SWS_LANCZOS) sizeFactor= param ? 2.0*param : 6.0;
862 else if(flags&SWS_SINC) sizeFactor= 20.0; // infinite ;)
863 else if(flags&SWS_SPLINE) sizeFactor= 20.0; // infinite ;)
864 else if(flags&SWS_BILINEAR) sizeFactor= 2.0;
866 sizeFactor= 0.0; //GCC warning killer
870 if(xInc1 <= 1.0) filterSizeInSrc= sizeFactor; // upscale
871 else filterSizeInSrc= sizeFactor*srcW / (double)dstW;
873 filterSize= (int)ceil(1 + filterSizeInSrc); // will be reduced later if possible
874 if(filterSize > srcW-2) filterSize=srcW-2;
876 filter= (double*)memalign(16, dstW*sizeof(double)*filterSize);
878 xDstInSrc= xInc1 / 2.0 - 0.5;
879 for(i=0; i<dstW; i++)
881 int xx= (int)(xDstInSrc - (filterSize-1)*0.5 + 0.5);
884 for(j=0; j<filterSize; j++)
886 double d= ABS(xx - xDstInSrc)/filterSizeInSrc*sizeFactor;
888 if(flags & SWS_BICUBIC)
890 double A= param ? -param*0.01 : -0.60;
892 // Equation is from VirtualDub
894 coeff = (1.0 - (A+3.0)*d*d + (A+2.0)*d*d*d);
896 coeff = (-4.0*A + 8.0*A*d - 5.0*A*d*d + A*d*d*d);
900 /* else if(flags & SWS_X)
902 double p= param ? param*0.01 : 0.3;
903 coeff = d ? sin(d*PI)/(d*PI) : 1.0;
904 coeff*= pow(2.0, - p*d*d);
906 else if(flags & SWS_X)
908 double A= param ? param*0.1 : 1.0;
914 if(coeff<0.0) coeff= -pow(-coeff, A);
915 else coeff= pow( coeff, A);
916 coeff= coeff*0.5 + 0.5;
918 else if(flags & SWS_AREA)
920 double srcPixelSize= 1.0/xInc1;
921 if(d + srcPixelSize/2 < 0.5) coeff= 1.0;
922 else if(d - srcPixelSize/2 < 0.5) coeff= (0.5-d)/srcPixelSize + 0.5;
925 else if(flags & SWS_GAUSS)
927 double p= param ? param*0.1 : 3.0;
928 coeff = pow(2.0, - p*d*d);
930 else if(flags & SWS_SINC)
932 coeff = d ? sin(d*PI)/(d*PI) : 1.0;
934 else if(flags & SWS_LANCZOS)
936 double p= param ? param : 3.0;
937 coeff = d ? sin(d*PI)*sin(d*PI/p)/(d*d*PI*PI/p) : 1.0;
940 else if(flags & SWS_BILINEAR)
945 else if(flags & SWS_SPLINE)
947 double p=-2.196152422706632;
948 coeff = getSplineCoeff(1.0, 0.0, p, -p-1.0, d);
951 coeff= 0.0; //GCC warning killer
955 filter[i*filterSize + j]= coeff;
962 /* apply src & dst Filter to filter -> filter2
966 filter2Size= filterSize;
967 if(srcFilter) filter2Size+= srcFilter->length - 1;
968 if(dstFilter) filter2Size+= dstFilter->length - 1;
969 ASSERT(filter2Size>0)
970 filter2= (double*)memalign(8, filter2Size*dstW*sizeof(double));
972 for(i=0; i<dstW; i++)
975 SwsVector scaleFilter;
978 scaleFilter.coeff= filter + i*filterSize;
979 scaleFilter.length= filterSize;
981 if(srcFilter) outVec= sws_getConvVec(srcFilter, &scaleFilter);
982 else outVec= &scaleFilter;
984 ASSERT(outVec->length == filter2Size)
987 for(j=0; j<outVec->length; j++)
989 filter2[i*filter2Size + j]= outVec->coeff[j];
992 (*filterPos)[i]+= (filterSize-1)/2 - (filter2Size-1)/2;
994 if(outVec != &scaleFilter) sws_freeVec(outVec);
996 free(filter); filter=NULL;
998 /* try to reduce the filter-size (step1 find size and shift left) */
999 // Assume its near normalized (*0.5 or *2.0 is ok but * 0.001 is not)
1001 for(i=dstW-1; i>=0; i--)
1003 int min= filter2Size;
1007 /* get rid off near zero elements on the left by shifting left */
1008 for(j=0; j<filter2Size; j++)
1011 cutOff += ABS(filter2[i*filter2Size]);
1013 if(cutOff > SWS_MAX_REDUCE_CUTOFF) break;
1015 /* preserve Monotonicity because the core can't handle the filter otherwise */
1016 if(i<dstW-1 && (*filterPos)[i] >= (*filterPos)[i+1]) break;
1018 // Move filter coeffs left
1019 for(k=1; k<filter2Size; k++)
1020 filter2[i*filter2Size + k - 1]= filter2[i*filter2Size + k];
1021 filter2[i*filter2Size + k - 1]= 0.0;
1026 /* count near zeros on the right */
1027 for(j=filter2Size-1; j>0; j--)
1029 cutOff += ABS(filter2[i*filter2Size + j]);
1031 if(cutOff > SWS_MAX_REDUCE_CUTOFF) break;
1035 if(min>minFilterSize) minFilterSize= min;
1038 if (flags & SWS_CPU_CAPS_ALTIVEC) {
1039 // we can handle the special case 4,
1040 // so we don't want to go to the full 8
1041 if (minFilterSize < 5)
1044 // we really don't want to waste our time
1045 // doing useless computation, so fall-back on
1046 // the scalar C code for very small filter.
1047 // vectorizing is worth it only if you have
1048 // decent-sized vector.
1049 if (minFilterSize < 3)
1053 ASSERT(minFilterSize > 0)
1054 filterSize= (minFilterSize +(filterAlign-1)) & (~(filterAlign-1));
1055 ASSERT(filterSize > 0)
1056 filter= (double*)memalign(8, filterSize*dstW*sizeof(double));
1057 *outFilterSize= filterSize;
1059 if(flags&SWS_PRINT_INFO)
1060 MSG_INFO("SwScaler: reducing / aligning filtersize %d -> %d\n", filter2Size, filterSize);
1061 /* try to reduce the filter-size (step2 reduce it) */
1062 for(i=0; i<dstW; i++)
1066 for(j=0; j<filterSize; j++)
1068 if(j>=filter2Size) filter[i*filterSize + j]= 0.0;
1069 else filter[i*filterSize + j]= filter2[i*filter2Size + j];
1072 free(filter2); filter2=NULL;
1075 //FIXME try to align filterpos if possible
1078 for(i=0; i<dstW; i++)
1081 if((*filterPos)[i] < 0)
1083 // Move filter coeffs left to compensate for filterPos
1084 for(j=1; j<filterSize; j++)
1086 int left= MAX(j + (*filterPos)[i], 0);
1087 filter[i*filterSize + left] += filter[i*filterSize + j];
1088 filter[i*filterSize + j]=0;
1093 if((*filterPos)[i] + filterSize > srcW)
1095 int shift= (*filterPos)[i] + filterSize - srcW;
1096 // Move filter coeffs right to compensate for filterPos
1097 for(j=filterSize-2; j>=0; j--)
1099 int right= MIN(j + shift, filterSize-1);
1100 filter[i*filterSize +right] += filter[i*filterSize +j];
1101 filter[i*filterSize +j]=0;
1103 (*filterPos)[i]= srcW - filterSize;
1107 // Note the +1 is for the MMXscaler which reads over the end
1108 *outFilter= (int16_t*)memalign(8, *outFilterSize*(dstW+1)*sizeof(int16_t));
1109 memset(*outFilter, 0, *outFilterSize*(dstW+1)*sizeof(int16_t));
1111 /* Normalize & Store in outFilter */
1112 for(i=0; i<dstW; i++)
1119 for(j=0; j<filterSize; j++)
1121 sum+= filter[i*filterSize + j];
1124 for(j=0; j<*outFilterSize; j++)
1126 double v= filter[i*filterSize + j]*scale + error;
1127 int intV= floor(v + 0.5);
1128 (*outFilter)[i*(*outFilterSize) + j]= intV;
1133 (*filterPos)[dstW]= (*filterPos)[dstW-1]; // the MMX scaler will read over the end
1134 for(i=0; i<*outFilterSize; i++)
1136 int j= dstW*(*outFilterSize);
1137 (*outFilter)[j + i]= (*outFilter)[j + i - (*outFilterSize)];
1144 static void initMMX2HScaler(int dstW, int xInc, uint8_t *funnyCode, int16_t *filter, int32_t *filterPos, int numSplits)
1149 int fragmentLengthA;
1153 int fragmentLengthB;
1158 // create an optimized horizontal scaling routine
1166 "movq (%%edx, %%eax), %%mm3 \n\t"
1167 "movd (%%ecx, %%esi), %%mm0 \n\t"
1168 "movd 1(%%ecx, %%esi), %%mm1 \n\t"
1169 "punpcklbw %%mm7, %%mm1 \n\t"
1170 "punpcklbw %%mm7, %%mm0 \n\t"
1171 "pshufw $0xFF, %%mm1, %%mm1 \n\t"
1173 "pshufw $0xFF, %%mm0, %%mm0 \n\t"
1175 "psubw %%mm1, %%mm0 \n\t"
1176 "movl 8(%%ebx, %%eax), %%esi \n\t"
1177 "pmullw %%mm3, %%mm0 \n\t"
1178 "psllw $7, %%mm1 \n\t"
1179 "paddw %%mm1, %%mm0 \n\t"
1181 "movq %%mm0, (%%edi, %%eax) \n\t"
1183 "addl $8, %%eax \n\t"
1198 :"=r" (fragmentA), "=r" (imm8OfPShufW1A), "=r" (imm8OfPShufW2A),
1199 "=r" (fragmentLengthA)
1206 "movq (%%edx, %%eax), %%mm3 \n\t"
1207 "movd (%%ecx, %%esi), %%mm0 \n\t"
1208 "punpcklbw %%mm7, %%mm0 \n\t"
1209 "pshufw $0xFF, %%mm0, %%mm1 \n\t"
1211 "pshufw $0xFF, %%mm0, %%mm0 \n\t"
1213 "psubw %%mm1, %%mm0 \n\t"
1214 "movl 8(%%ebx, %%eax), %%esi \n\t"
1215 "pmullw %%mm3, %%mm0 \n\t"
1216 "psllw $7, %%mm1 \n\t"
1217 "paddw %%mm1, %%mm0 \n\t"
1219 "movq %%mm0, (%%edi, %%eax) \n\t"
1221 "addl $8, %%eax \n\t"
1236 :"=r" (fragmentB), "=r" (imm8OfPShufW1B), "=r" (imm8OfPShufW2B),
1237 "=r" (fragmentLengthB)
1240 xpos= 0; //lumXInc/2 - 0x8000; // difference between pixel centers
1243 for(i=0; i<dstW/numSplits; i++)
1250 int b=((xpos+xInc)>>16) - xx;
1251 int c=((xpos+xInc*2)>>16) - xx;
1252 int d=((xpos+xInc*3)>>16) - xx;
1254 filter[i ] = (( xpos & 0xFFFF) ^ 0xFFFF)>>9;
1255 filter[i+1] = (((xpos+xInc ) & 0xFFFF) ^ 0xFFFF)>>9;
1256 filter[i+2] = (((xpos+xInc*2) & 0xFFFF) ^ 0xFFFF)>>9;
1257 filter[i+3] = (((xpos+xInc*3) & 0xFFFF) ^ 0xFFFF)>>9;
1262 int maxShift= 3-(d+1);
1265 memcpy(funnyCode + fragmentPos, fragmentB, fragmentLengthB);
1267 funnyCode[fragmentPos + imm8OfPShufW1B]=
1268 (a+1) | ((b+1)<<2) | ((c+1)<<4) | ((d+1)<<6);
1269 funnyCode[fragmentPos + imm8OfPShufW2B]=
1270 a | (b<<2) | (c<<4) | (d<<6);
1272 if(i+3>=dstW) shift=maxShift; //avoid overread
1273 else if((filterPos[i/2]&3) <= maxShift) shift=filterPos[i/2]&3; //Align
1275 if(shift && i>=shift)
1277 funnyCode[fragmentPos + imm8OfPShufW1B]+= 0x55*shift;
1278 funnyCode[fragmentPos + imm8OfPShufW2B]+= 0x55*shift;
1279 filterPos[i/2]-=shift;
1282 fragmentPos+= fragmentLengthB;
1289 memcpy(funnyCode + fragmentPos, fragmentA, fragmentLengthA);
1291 funnyCode[fragmentPos + imm8OfPShufW1A]=
1292 funnyCode[fragmentPos + imm8OfPShufW2A]=
1293 a | (b<<2) | (c<<4) | (d<<6);
1295 if(i+4>=dstW) shift=maxShift; //avoid overread
1296 else if((filterPos[i/2]&3) <= maxShift) shift=filterPos[i/2]&3; //partial align
1298 if(shift && i>=shift)
1300 funnyCode[fragmentPos + imm8OfPShufW1A]+= 0x55*shift;
1301 funnyCode[fragmentPos + imm8OfPShufW2A]+= 0x55*shift;
1302 filterPos[i/2]-=shift;
1305 fragmentPos+= fragmentLengthA;
1308 funnyCode[fragmentPos]= RET;
1312 filterPos[i/2]= xpos>>16; // needed to jump to the next part
1316 static void globalInit(){
1317 // generating tables:
1319 for(i=0; i<768; i++){
1320 int c= MIN(MAX(i-256, 0), 255);
1325 static SwsFunc getSwsFunc(int flags){
1327 #ifdef RUNTIME_CPUDETECT
1329 // ordered per speed fasterst first
1330 if(flags & SWS_CPU_CAPS_MMX2)
1331 return swScale_MMX2;
1332 else if(flags & SWS_CPU_CAPS_3DNOW)
1333 return swScale_3DNow;
1334 else if(flags & SWS_CPU_CAPS_MMX)
1341 if(flags & SWS_CPU_CAPS_ALTIVEC)
1342 return swScale_altivec;
1348 #else //RUNTIME_CPUDETECT
1350 return swScale_MMX2;
1351 #elif defined (HAVE_3DNOW)
1352 return swScale_3DNow;
1353 #elif defined (HAVE_MMX)
1355 #elif defined (HAVE_ALTIVEC)
1356 return swScale_altivec;
1360 #endif //!RUNTIME_CPUDETECT
1363 static int PlanarToNV12Wrapper(SwsContext *c, uint8_t* src[], int srcStride[], int srcSliceY,
1364 int srcSliceH, uint8_t* dstParam[], int dstStride[]){
1365 uint8_t *dst=dstParam[0] + dstStride[0]*srcSliceY;
1367 if(dstStride[0]==srcStride[0])
1368 memcpy(dst, src[0], srcSliceH*dstStride[0]);
1372 uint8_t *srcPtr= src[0];
1373 uint8_t *dstPtr= dst;
1374 for(i=0; i<srcSliceH; i++)
1376 memcpy(dstPtr, srcPtr, srcStride[0]);
1377 srcPtr+= srcStride[0];
1378 dstPtr+= dstStride[0];
1381 dst = dstParam[1] + dstStride[1]*srcSliceY;
1382 interleaveBytes( src[1],src[2],dst,c->srcW,srcSliceH,srcStride[1],srcStride[2],dstStride[0] );
1387 static int PlanarToYuy2Wrapper(SwsContext *c, uint8_t* src[], int srcStride[], int srcSliceY,
1388 int srcSliceH, uint8_t* dstParam[], int dstStride[]){
1389 uint8_t *dst=dstParam[0] + dstStride[0]*srcSliceY;
1391 yv12toyuy2( src[0],src[1],src[2],dst,c->srcW,srcSliceH,srcStride[0],srcStride[1],dstStride[0] );
1396 static int PlanarToUyvyWrapper(SwsContext *c, uint8_t* src[], int srcStride[], int srcSliceY,
1397 int srcSliceH, uint8_t* dstParam[], int dstStride[]){
1398 uint8_t *dst=dstParam[0] + dstStride[0]*srcSliceY;
1400 yv12touyvy( src[0],src[1],src[2],dst,c->srcW,srcSliceH,srcStride[0],srcStride[1],dstStride[0] );
1405 /* {RGB,BGR}{15,16,24,32} -> {RGB,BGR}{15,16,24,32} */
1406 static int rgb2rgbWrapper(SwsContext *c, uint8_t* src[], int srcStride[], int srcSliceY,
1407 int srcSliceH, uint8_t* dst[], int dstStride[]){
1408 const int srcFormat= c->srcFormat;
1409 const int dstFormat= c->dstFormat;
1410 const int srcBpp= ((srcFormat&0xFF) + 7)>>3;
1411 const int dstBpp= ((dstFormat&0xFF) + 7)>>3;
1412 const int srcId= (srcFormat&0xFF)>>2; // 1:0, 4:1, 8:2, 15:3, 16:4, 24:6, 32:8
1413 const int dstId= (dstFormat&0xFF)>>2;
1414 void (*conv)(const uint8_t *src, uint8_t *dst, unsigned src_size)=NULL;
1417 if( (isBGR(srcFormat) && isBGR(dstFormat))
1418 || (isRGB(srcFormat) && isRGB(dstFormat))){
1419 switch(srcId | (dstId<<4)){
1420 case 0x34: conv= rgb16to15; break;
1421 case 0x36: conv= rgb24to15; break;
1422 case 0x38: conv= rgb32to15; break;
1423 case 0x43: conv= rgb15to16; break;
1424 case 0x46: conv= rgb24to16; break;
1425 case 0x48: conv= rgb32to16; break;
1426 case 0x63: conv= rgb15to24; break;
1427 case 0x64: conv= rgb16to24; break;
1428 case 0x68: conv= rgb32to24; break;
1429 case 0x83: conv= rgb15to32; break;
1430 case 0x84: conv= rgb16to32; break;
1431 case 0x86: conv= rgb24to32; break;
1432 default: MSG_ERR("swScaler: internal error %s -> %s converter\n",
1433 vo_format_name(srcFormat), vo_format_name(dstFormat)); break;
1435 }else if( (isBGR(srcFormat) && isRGB(dstFormat))
1436 || (isRGB(srcFormat) && isBGR(dstFormat))){
1437 switch(srcId | (dstId<<4)){
1438 case 0x33: conv= rgb15tobgr15; break;
1439 case 0x34: conv= rgb16tobgr15; break;
1440 case 0x36: conv= rgb24tobgr15; break;
1441 case 0x38: conv= rgb32tobgr15; break;
1442 case 0x43: conv= rgb15tobgr16; break;
1443 case 0x44: conv= rgb16tobgr16; break;
1444 case 0x46: conv= rgb24tobgr16; break;
1445 case 0x48: conv= rgb32tobgr16; break;
1446 case 0x63: conv= rgb15tobgr24; break;
1447 case 0x64: conv= rgb16tobgr24; break;
1448 case 0x66: conv= rgb24tobgr24; break;
1449 case 0x68: conv= rgb32tobgr24; break;
1450 case 0x83: conv= rgb15tobgr32; break;
1451 case 0x84: conv= rgb16tobgr32; break;
1452 case 0x86: conv= rgb24tobgr32; break;
1453 case 0x88: conv= rgb32tobgr32; break;
1454 default: MSG_ERR("swScaler: internal error %s -> %s converter\n",
1455 vo_format_name(srcFormat), vo_format_name(dstFormat)); break;
1458 MSG_ERR("swScaler: internal error %s -> %s converter\n",
1459 vo_format_name(srcFormat), vo_format_name(dstFormat));
1462 if(dstStride[0]*srcBpp == srcStride[0]*dstBpp)
1463 conv(src[0], dst[0] + dstStride[0]*srcSliceY, srcSliceH*srcStride[0]);
1467 uint8_t *srcPtr= src[0];
1468 uint8_t *dstPtr= dst[0] + dstStride[0]*srcSliceY;
1470 for(i=0; i<srcSliceH; i++)
1472 conv(srcPtr, dstPtr, c->srcW*srcBpp);
1473 srcPtr+= srcStride[0];
1474 dstPtr+= dstStride[0];
1480 static int bgr24toyv12Wrapper(SwsContext *c, uint8_t* src[], int srcStride[], int srcSliceY,
1481 int srcSliceH, uint8_t* dst[], int dstStride[]){
1485 dst[0]+ srcSliceY *dstStride[0],
1486 dst[1]+(srcSliceY>>1)*dstStride[1],
1487 dst[2]+(srcSliceY>>1)*dstStride[2],
1489 dstStride[0], dstStride[1], srcStride[0]);
1493 static int yvu9toyv12Wrapper(SwsContext *c, uint8_t* src[], int srcStride[], int srcSliceY,
1494 int srcSliceH, uint8_t* dst[], int dstStride[]){
1498 if(srcStride[0]==dstStride[0])
1499 memcpy(dst[0]+ srcSliceY*dstStride[0], src[0], srcStride[0]*srcSliceH);
1501 uint8_t *srcPtr= src[0];
1502 uint8_t *dstPtr= dst[0] + dstStride[0]*srcSliceY;
1504 for(i=0; i<srcSliceH; i++)
1506 memcpy(dstPtr, srcPtr, c->srcW);
1507 srcPtr+= srcStride[0];
1508 dstPtr+= dstStride[0];
1512 if(c->dstFormat==IMGFMT_YV12){
1513 planar2x(src[1], dst[1], c->chrSrcW, c->chrSrcH, srcStride[1], dstStride[1]);
1514 planar2x(src[2], dst[2], c->chrSrcW, c->chrSrcH, srcStride[2], dstStride[2]);
1516 planar2x(src[1], dst[2], c->chrSrcW, c->chrSrcH, srcStride[1], dstStride[2]);
1517 planar2x(src[2], dst[1], c->chrSrcW, c->chrSrcH, srcStride[2], dstStride[1]);
1523 * bring pointers in YUV order instead of YVU
1525 static inline void sws_orderYUV(int format, uint8_t * sortedP[], int sortedStride[], uint8_t * p[], int stride[]){
1526 if(format == IMGFMT_YV12 || format == IMGFMT_YVU9
1527 || format == IMGFMT_444P || format == IMGFMT_422P || format == IMGFMT_411P){
1531 sortedStride[0]= stride[0];
1532 sortedStride[1]= stride[2];
1533 sortedStride[2]= stride[1];
1535 else if(isPacked(format) || isGray(format) || format == IMGFMT_Y8)
1540 sortedStride[0]= stride[0];
1544 else if(format == IMGFMT_I420 || format == IMGFMT_IYUV)
1549 sortedStride[0]= stride[0];
1550 sortedStride[1]= stride[1];
1551 sortedStride[2]= stride[2];
1553 MSG_ERR("internal error in orderYUV\n");
1557 /* unscaled copy like stuff (assumes nearly identical formats) */
1558 static int simpleCopy(SwsContext *c, uint8_t* src[], int srcStride[], int srcSliceY,
1559 int srcSliceH, uint8_t* dst[], int dstStride[]){
1561 if(isPacked(c->srcFormat))
1563 if(dstStride[0]==srcStride[0])
1564 memcpy(dst[0] + dstStride[0]*srcSliceY, src[0], srcSliceH*dstStride[0]);
1568 uint8_t *srcPtr= src[0];
1569 uint8_t *dstPtr= dst[0] + dstStride[0]*srcSliceY;
1572 /* universal length finder */
1573 while(length+c->srcW <= ABS(dstStride[0])
1574 && length+c->srcW <= ABS(srcStride[0])) length+= c->srcW;
1577 for(i=0; i<srcSliceH; i++)
1579 memcpy(dstPtr, srcPtr, length);
1580 srcPtr+= srcStride[0];
1581 dstPtr+= dstStride[0];
1586 { /* Planar YUV or gray */
1588 for(plane=0; plane<3; plane++)
1590 int length= plane==0 ? c->srcW : -((-c->srcW )>>c->chrDstHSubSample);
1591 int y= plane==0 ? srcSliceY: -((-srcSliceY)>>c->chrDstVSubSample);
1592 int height= plane==0 ? srcSliceH: -((-srcSliceH)>>c->chrDstVSubSample);
1594 if((isGray(c->srcFormat) || isGray(c->dstFormat)) && plane>0)
1596 if(!isGray(c->dstFormat))
1597 memset(dst[plane], 128, dstStride[plane]*height);
1601 if(dstStride[plane]==srcStride[plane])
1602 memcpy(dst[plane] + dstStride[plane]*y, src[plane], height*dstStride[plane]);
1606 uint8_t *srcPtr= src[plane];
1607 uint8_t *dstPtr= dst[plane] + dstStride[plane]*y;
1608 for(i=0; i<height; i++)
1610 memcpy(dstPtr, srcPtr, length);
1611 srcPtr+= srcStride[plane];
1612 dstPtr+= dstStride[plane];
1621 static int remove_dup_fourcc(int fourcc)
1626 case IMGFMT_IYUV: return IMGFMT_YV12;
1627 case IMGFMT_Y8 : return IMGFMT_Y800;
1628 case IMGFMT_IF09: return IMGFMT_YVU9;
1629 default: return fourcc;
1633 static void getSubSampleFactors(int *h, int *v, int format){
1641 case IMGFMT_Y800: //FIXME remove after different subsamplings are fully implemented
1668 static uint16_t roundToInt16(int64_t f){
1669 int r= (f + (1<<15))>>16;
1670 if(r<-0x7FFF) return 0x8000;
1671 else if(r> 0x7FFF) return 0x7FFF;
1676 * @param inv_table the yuv2rgb coeffs, normally Inverse_Table_6_9[x]
1677 * @param fullRange if 1 then the luma range is 0..255 if 0 its 16..235
1678 * @return -1 if not supported
1680 int sws_setColorspaceDetails(SwsContext *c, const int inv_table[4], int srcRange, const int table[4], int dstRange, int brightness, int contrast, int saturation){
1681 int64_t crv = inv_table[0];
1682 int64_t cbu = inv_table[1];
1683 int64_t cgu = -inv_table[2];
1684 int64_t cgv = -inv_table[3];
1688 if(isYUV(c->dstFormat) || isGray(c->dstFormat)) return -1;
1689 memcpy(c->srcColorspaceTable, inv_table, sizeof(int)*4);
1690 memcpy(c->dstColorspaceTable, table, sizeof(int)*4);
1692 c->brightness= brightness;
1693 c->contrast = contrast;
1694 c->saturation= saturation;
1695 c->srcRange = srcRange;
1696 c->dstRange = dstRange;
1698 c->uOffset= 0x0400040004000400LL;
1699 c->vOffset= 0x0400040004000400LL;
1706 cy = (cy *contrast )>>16;
1707 crv= (crv*contrast * saturation)>>32;
1708 cbu= (cbu*contrast * saturation)>>32;
1709 cgu= (cgu*contrast * saturation)>>32;
1710 cgv= (cgv*contrast * saturation)>>32;
1712 oy -= 256*brightness;
1714 c->yCoeff= roundToInt16(cy *8192) * 0x0001000100010001ULL;
1715 c->vrCoeff= roundToInt16(crv*8192) * 0x0001000100010001ULL;
1716 c->ubCoeff= roundToInt16(cbu*8192) * 0x0001000100010001ULL;
1717 c->vgCoeff= roundToInt16(cgv*8192) * 0x0001000100010001ULL;
1718 c->ugCoeff= roundToInt16(cgu*8192) * 0x0001000100010001ULL;
1719 c->yOffset= roundToInt16(oy * 8) * 0x0001000100010001ULL;
1721 yuv2rgb_c_init_tables(c, inv_table, srcRange, brightness, contrast, saturation);
1725 yuv2rgb_altivec_init_tables (c, inv_table);
1731 * @return -1 if not supported
1733 int sws_getColorspaceDetails(SwsContext *c, int **inv_table, int *srcRange, int **table, int *dstRange, int *brightness, int *contrast, int *saturation){
1734 if(isYUV(c->dstFormat) || isGray(c->dstFormat)) return -1;
1736 *inv_table = c->srcColorspaceTable;
1737 *table = c->dstColorspaceTable;
1738 *srcRange = c->srcRange;
1739 *dstRange = c->dstRange;
1740 *brightness= c->brightness;
1741 *contrast = c->contrast;
1742 *saturation= c->saturation;
1747 SwsContext *sws_getContext(int srcW, int srcH, int origSrcFormat, int dstW, int dstH, int origDstFormat, int flags,
1748 SwsFilter *srcFilter, SwsFilter *dstFilter){
1752 int usesVFilter, usesHFilter;
1753 int unscaled, needsDither;
1754 int srcFormat, dstFormat;
1755 SwsFilter dummyFilter= {NULL, NULL, NULL, NULL};
1757 if(flags & SWS_CPU_CAPS_MMX)
1758 asm volatile("emms\n\t"::: "memory");
1761 #ifndef RUNTIME_CPUDETECT //ensure that the flags match the compiled variant if cpudetect is off
1762 flags &= ~(SWS_CPU_CAPS_MMX|SWS_CPU_CAPS_MMX2|SWS_CPU_CAPS_3DNOW|SWS_CPU_CAPS_ALTIVEC);
1764 flags |= SWS_CPU_CAPS_MMX|SWS_CPU_CAPS_MMX2;
1765 #elif defined (HAVE_3DNOW)
1766 flags |= SWS_CPU_CAPS_MMX|SWS_CPU_CAPS_3DNOW;
1767 #elif defined (HAVE_MMX)
1768 flags |= SWS_CPU_CAPS_MMX;
1769 #elif defined (HAVE_ALTIVEC)
1770 flags |= SWS_CPU_CAPS_ALTIVEC;
1773 if(clip_table[512] != 255) globalInit();
1774 if(rgb15to16 == NULL) sws_rgb2rgb_init(flags);
1776 /* avoid duplicate Formats, so we don't need to check to much */
1777 srcFormat = remove_dup_fourcc(origSrcFormat);
1778 dstFormat = remove_dup_fourcc(origDstFormat);
1780 unscaled = (srcW == dstW && srcH == dstH);
1781 needsDither= (isBGR(dstFormat) || isRGB(dstFormat))
1782 && (dstFormat&0xFF)<24
1783 && ((dstFormat&0xFF)<(srcFormat&0xFF) || (!(isRGB(srcFormat) || isBGR(srcFormat))));
1785 if(!isSupportedIn(srcFormat))
1787 MSG_ERR("swScaler: %s is not supported as input format\n", vo_format_name(srcFormat));
1790 if(!isSupportedOut(dstFormat))
1792 MSG_ERR("swScaler: %s is not supported as output format\n", vo_format_name(dstFormat));
1797 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
1799 MSG_ERR("swScaler: %dx%d -> %dx%d is invalid scaling dimension\n",
1800 srcW, srcH, dstW, dstH);
1804 if(!dstFilter) dstFilter= &dummyFilter;
1805 if(!srcFilter) srcFilter= &dummyFilter;
1807 c= memalign(64, sizeof(SwsContext));
1808 memset(c, 0, sizeof(SwsContext));
1814 c->lumXInc= ((srcW<<16) + (dstW>>1))/dstW;
1815 c->lumYInc= ((srcH<<16) + (dstH>>1))/dstH;
1817 c->dstFormat= dstFormat;
1818 c->srcFormat= srcFormat;
1819 c->origDstFormat= origDstFormat;
1820 c->origSrcFormat= origSrcFormat;
1821 c->vRounder= 4* 0x0001000100010001ULL;
1823 usesHFilter= usesVFilter= 0;
1824 if(dstFilter->lumV!=NULL && dstFilter->lumV->length>1) usesVFilter=1;
1825 if(dstFilter->lumH!=NULL && dstFilter->lumH->length>1) usesHFilter=1;
1826 if(dstFilter->chrV!=NULL && dstFilter->chrV->length>1) usesVFilter=1;
1827 if(dstFilter->chrH!=NULL && dstFilter->chrH->length>1) usesHFilter=1;
1828 if(srcFilter->lumV!=NULL && srcFilter->lumV->length>1) usesVFilter=1;
1829 if(srcFilter->lumH!=NULL && srcFilter->lumH->length>1) usesHFilter=1;
1830 if(srcFilter->chrV!=NULL && srcFilter->chrV->length>1) usesVFilter=1;
1831 if(srcFilter->chrH!=NULL && srcFilter->chrH->length>1) usesHFilter=1;
1833 getSubSampleFactors(&c->chrSrcHSubSample, &c->chrSrcVSubSample, srcFormat);
1834 getSubSampleFactors(&c->chrDstHSubSample, &c->chrDstVSubSample, dstFormat);
1836 // reuse chroma for 2 pixles rgb/bgr unless user wants full chroma interpolation
1837 if((isBGR(dstFormat) || isRGB(dstFormat)) && !(flags&SWS_FULL_CHR_H_INT)) c->chrDstHSubSample=1;
1839 // drop some chroma lines if the user wants it
1840 c->vChrDrop= (flags&SWS_SRC_V_CHR_DROP_MASK)>>SWS_SRC_V_CHR_DROP_SHIFT;
1841 c->chrSrcVSubSample+= c->vChrDrop;
1843 // drop every 2. pixel for chroma calculation unless user wants full chroma
1844 if((isBGR(srcFormat) || isRGB(srcFormat)) && !(flags&SWS_FULL_CHR_H_INP))
1845 c->chrSrcHSubSample=1;
1847 c->chrIntHSubSample= c->chrDstHSubSample;
1848 c->chrIntVSubSample= c->chrSrcVSubSample;
1850 // note the -((-x)>>y) is so that we allways round toward +inf
1851 c->chrSrcW= -((-srcW) >> c->chrSrcHSubSample);
1852 c->chrSrcH= -((-srcH) >> c->chrSrcVSubSample);
1853 c->chrDstW= -((-dstW) >> c->chrDstHSubSample);
1854 c->chrDstH= -((-dstH) >> c->chrDstVSubSample);
1856 sws_setColorspaceDetails(c, Inverse_Table_6_9[SWS_CS_DEFAULT], 0, Inverse_Table_6_9[SWS_CS_DEFAULT] /* FIXME*/, 0, 0, 1<<16, 1<<16);
1858 /* unscaled special Cases */
1859 if(unscaled && !usesHFilter && !usesVFilter)
1862 if(srcFormat == IMGFMT_YV12 && dstFormat == IMGFMT_NV12)
1864 c->swScale= PlanarToNV12Wrapper;
1867 if((srcFormat==IMGFMT_YV12 || srcFormat==IMGFMT_422P) && (isBGR(dstFormat) || isRGB(dstFormat)))
1869 c->swScale= yuv2rgb_get_func_ptr(c);
1872 if( srcFormat==IMGFMT_YVU9 && dstFormat==IMGFMT_YV12 )
1874 c->swScale= yvu9toyv12Wrapper;
1878 if(srcFormat==IMGFMT_BGR24 && dstFormat==IMGFMT_YV12)
1879 c->swScale= bgr24toyv12Wrapper;
1881 /* rgb/bgr -> rgb/bgr (no dither needed forms) */
1882 if( (isBGR(srcFormat) || isRGB(srcFormat))
1883 && (isBGR(dstFormat) || isRGB(dstFormat))
1885 c->swScale= rgb2rgbWrapper;
1887 /* LQ converters if -sws 0 or -sws 4*/
1888 if(c->flags&(SWS_FAST_BILINEAR|SWS_POINT)){
1889 /* rgb/bgr -> rgb/bgr (dither needed forms) */
1890 if( (isBGR(srcFormat) || isRGB(srcFormat))
1891 && (isBGR(dstFormat) || isRGB(dstFormat))
1893 c->swScale= rgb2rgbWrapper;
1896 if(srcFormat == IMGFMT_YV12 &&
1897 (dstFormat == IMGFMT_YUY2 || dstFormat == IMGFMT_UYVY))
1899 if (dstFormat == IMGFMT_YUY2)
1900 c->swScale= PlanarToYuy2Wrapper;
1902 c->swScale= PlanarToUyvyWrapper;
1907 if ((c->flags & SWS_CPU_CAPS_ALTIVEC) &&
1908 ((srcFormat == IMGFMT_YV12 &&
1909 (dstFormat == IMGFMT_YUY2 || dstFormat == IMGFMT_UYVY)))) {
1910 // unscaled YV12 -> packed YUV, we want speed
1911 if (dstFormat == IMGFMT_YUY2)
1912 c->swScale= yv12toyuy2_unscaled_altivec;
1914 c->swScale= yv12touyvy_unscaled_altivec;
1919 if( srcFormat == dstFormat
1920 || (isPlanarYUV(srcFormat) && isGray(dstFormat))
1921 || (isPlanarYUV(dstFormat) && isGray(srcFormat))
1924 c->swScale= simpleCopy;
1928 if(flags&SWS_PRINT_INFO)
1929 MSG_INFO("SwScaler: using unscaled %s -> %s special converter\n",
1930 vo_format_name(srcFormat), vo_format_name(dstFormat));
1935 if(flags & SWS_CPU_CAPS_MMX2)
1937 c->canMMX2BeUsed= (dstW >=srcW && (dstW&31)==0 && (srcW&15)==0) ? 1 : 0;
1938 if(!c->canMMX2BeUsed && dstW >=srcW && (srcW&15)==0 && (flags&SWS_FAST_BILINEAR))
1940 if(flags&SWS_PRINT_INFO)
1941 MSG_INFO("SwScaler: output Width is not a multiple of 32 -> no MMX2 scaler\n");
1943 if(usesHFilter) c->canMMX2BeUsed=0;
1948 c->chrXInc= ((c->chrSrcW<<16) + (c->chrDstW>>1))/c->chrDstW;
1949 c->chrYInc= ((c->chrSrcH<<16) + (c->chrDstH>>1))/c->chrDstH;
1951 // match pixel 0 of the src to pixel 0 of dst and match pixel n-2 of src to pixel n-2 of dst
1952 // but only for the FAST_BILINEAR mode otherwise do correct scaling
1953 // n-2 is the last chrominance sample available
1954 // this is not perfect, but noone shuld notice the difference, the more correct variant
1955 // would be like the vertical one, but that would require some special code for the
1956 // first and last pixel
1957 if(flags&SWS_FAST_BILINEAR)
1959 if(c->canMMX2BeUsed)
1964 //we don't use the x86asm scaler if mmx is available
1965 else if(flags & SWS_CPU_CAPS_MMX)
1967 c->lumXInc = ((srcW-2)<<16)/(dstW-2) - 20;
1968 c->chrXInc = ((c->chrSrcW-2)<<16)/(c->chrDstW-2) - 20;
1972 /* precalculate horizontal scaler filter coefficients */
1974 const int filterAlign=
1975 (flags & SWS_CPU_CAPS_MMX) ? 4 :
1976 (flags & SWS_CPU_CAPS_ALTIVEC) ? 8 :
1979 initFilter(&c->hLumFilter, &c->hLumFilterPos, &c->hLumFilterSize, c->lumXInc,
1980 srcW , dstW, filterAlign, 1<<14,
1981 (flags&SWS_BICUBLIN) ? (flags|SWS_BICUBIC) : flags,
1982 srcFilter->lumH, dstFilter->lumH);
1983 initFilter(&c->hChrFilter, &c->hChrFilterPos, &c->hChrFilterSize, c->chrXInc,
1984 c->chrSrcW, c->chrDstW, filterAlign, 1<<14,
1985 (flags&SWS_BICUBLIN) ? (flags|SWS_BILINEAR) : flags,
1986 srcFilter->chrH, dstFilter->chrH);
1989 // can't downscale !!!
1990 if(c->canMMX2BeUsed && (flags & SWS_FAST_BILINEAR))
1992 c->lumMmx2Filter = (int16_t*)memalign(8, (dstW /8+8)*sizeof(int16_t));
1993 c->chrMmx2Filter = (int16_t*)memalign(8, (c->chrDstW /4+8)*sizeof(int16_t));
1994 c->lumMmx2FilterPos= (int32_t*)memalign(8, (dstW /2/8+8)*sizeof(int32_t));
1995 c->chrMmx2FilterPos= (int32_t*)memalign(8, (c->chrDstW/2/4+8)*sizeof(int32_t));
1997 initMMX2HScaler( dstW, c->lumXInc, c->funnyYCode , c->lumMmx2Filter, c->lumMmx2FilterPos, 8);
1998 initMMX2HScaler(c->chrDstW, c->chrXInc, c->funnyUVCode, c->chrMmx2Filter, c->chrMmx2FilterPos, 4);
2001 } // Init Horizontal stuff
2005 /* precalculate vertical scaler filter coefficients */
2007 const int filterAlign=
2008 (flags & SWS_CPU_CAPS_ALTIVEC) ? 8 :
2011 initFilter(&c->vLumFilter, &c->vLumFilterPos, &c->vLumFilterSize, c->lumYInc,
2012 srcH , dstH, filterAlign, (1<<12)-4,
2013 (flags&SWS_BICUBLIN) ? (flags|SWS_BICUBIC) : flags,
2014 srcFilter->lumV, dstFilter->lumV);
2015 initFilter(&c->vChrFilter, &c->vChrFilterPos, &c->vChrFilterSize, c->chrYInc,
2016 c->chrSrcH, c->chrDstH, filterAlign, (1<<12)-4,
2017 (flags&SWS_BICUBLIN) ? (flags|SWS_BILINEAR) : flags,
2018 srcFilter->chrV, dstFilter->chrV);
2021 // Calculate Buffer Sizes so that they won't run out while handling these damn slices
2022 c->vLumBufSize= c->vLumFilterSize;
2023 c->vChrBufSize= c->vChrFilterSize;
2024 for(i=0; i<dstH; i++)
2026 int chrI= i*c->chrDstH / dstH;
2027 int nextSlice= MAX(c->vLumFilterPos[i ] + c->vLumFilterSize - 1,
2028 ((c->vChrFilterPos[chrI] + c->vChrFilterSize - 1)<<c->chrSrcVSubSample));
2030 nextSlice>>= c->chrSrcVSubSample;
2031 nextSlice<<= c->chrSrcVSubSample;
2032 if(c->vLumFilterPos[i ] + c->vLumBufSize < nextSlice)
2033 c->vLumBufSize= nextSlice - c->vLumFilterPos[i ];
2034 if(c->vChrFilterPos[chrI] + c->vChrBufSize < (nextSlice>>c->chrSrcVSubSample))
2035 c->vChrBufSize= (nextSlice>>c->chrSrcVSubSample) - c->vChrFilterPos[chrI];
2038 // allocate pixbufs (we use dynamic allocation because otherwise we would need to
2039 c->lumPixBuf= (int16_t**)memalign(4, c->vLumBufSize*2*sizeof(int16_t*));
2040 c->chrPixBuf= (int16_t**)memalign(4, c->vChrBufSize*2*sizeof(int16_t*));
2041 //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)
2042 for(i=0; i<c->vLumBufSize; i++)
2043 c->lumPixBuf[i]= c->lumPixBuf[i+c->vLumBufSize]= (uint16_t*)memalign(8, 4000);
2044 for(i=0; i<c->vChrBufSize; i++)
2045 c->chrPixBuf[i]= c->chrPixBuf[i+c->vChrBufSize]= (uint16_t*)memalign(8, 8000);
2047 //try to avoid drawing green stuff between the right end and the stride end
2048 for(i=0; i<c->vLumBufSize; i++) memset(c->lumPixBuf[i], 0, 4000);
2049 for(i=0; i<c->vChrBufSize; i++) memset(c->chrPixBuf[i], 64, 8000);
2051 ASSERT(c->chrDstH <= dstH)
2053 if(flags&SWS_PRINT_INFO)
2056 char *dither= " dithered";
2060 if(flags&SWS_FAST_BILINEAR)
2061 MSG_INFO("\nSwScaler: FAST_BILINEAR scaler, ");
2062 else if(flags&SWS_BILINEAR)
2063 MSG_INFO("\nSwScaler: BILINEAR scaler, ");
2064 else if(flags&SWS_BICUBIC)
2065 MSG_INFO("\nSwScaler: BICUBIC scaler, ");
2066 else if(flags&SWS_X)
2067 MSG_INFO("\nSwScaler: Experimental scaler, ");
2068 else if(flags&SWS_POINT)
2069 MSG_INFO("\nSwScaler: Nearest Neighbor / POINT scaler, ");
2070 else if(flags&SWS_AREA)
2071 MSG_INFO("\nSwScaler: Area Averageing scaler, ");
2072 else if(flags&SWS_BICUBLIN)
2073 MSG_INFO("\nSwScaler: luma BICUBIC / chroma BILINEAR scaler, ");
2074 else if(flags&SWS_GAUSS)
2075 MSG_INFO("\nSwScaler: Gaussian scaler, ");
2076 else if(flags&SWS_SINC)
2077 MSG_INFO("\nSwScaler: Sinc scaler, ");
2078 else if(flags&SWS_LANCZOS)
2079 MSG_INFO("\nSwScaler: Lanczos scaler, ");
2080 else if(flags&SWS_SPLINE)
2081 MSG_INFO("\nSwScaler: Bicubic spline scaler, ");
2083 MSG_INFO("\nSwScaler: ehh flags invalid?! ");
2085 if(dstFormat==IMGFMT_BGR15 || dstFormat==IMGFMT_BGR16)
2086 MSG_INFO("from %s to%s %s ",
2087 vo_format_name(srcFormat), dither, vo_format_name(dstFormat));
2089 MSG_INFO("from %s to %s ",
2090 vo_format_name(srcFormat), vo_format_name(dstFormat));
2092 if(flags & SWS_CPU_CAPS_MMX2)
2093 MSG_INFO("using MMX2\n");
2094 else if(flags & SWS_CPU_CAPS_3DNOW)
2095 MSG_INFO("using 3DNOW\n");
2096 else if(flags & SWS_CPU_CAPS_MMX)
2097 MSG_INFO("using MMX\n");
2098 else if(flags & SWS_CPU_CAPS_ALTIVEC)
2099 MSG_INFO("using AltiVec\n");
2101 MSG_INFO("using C\n");
2104 if(flags & SWS_PRINT_INFO)
2106 if(flags & SWS_CPU_CAPS_MMX)
2108 if(c->canMMX2BeUsed && (flags&SWS_FAST_BILINEAR))
2109 MSG_V("SwScaler: using FAST_BILINEAR MMX2 scaler for horizontal scaling\n");
2112 if(c->hLumFilterSize==4)
2113 MSG_V("SwScaler: using 4-tap MMX scaler for horizontal luminance scaling\n");
2114 else if(c->hLumFilterSize==8)
2115 MSG_V("SwScaler: using 8-tap MMX scaler for horizontal luminance scaling\n");
2117 MSG_V("SwScaler: using n-tap MMX scaler for horizontal luminance scaling\n");
2119 if(c->hChrFilterSize==4)
2120 MSG_V("SwScaler: using 4-tap MMX scaler for horizontal chrominance scaling\n");
2121 else if(c->hChrFilterSize==8)
2122 MSG_V("SwScaler: using 8-tap MMX scaler for horizontal chrominance scaling\n");
2124 MSG_V("SwScaler: using n-tap MMX scaler for horizontal chrominance scaling\n");
2130 MSG_V("SwScaler: using X86-Asm scaler for horizontal scaling\n");
2132 if(flags & SWS_FAST_BILINEAR)
2133 MSG_V("SwScaler: using FAST_BILINEAR C scaler for horizontal scaling\n");
2135 MSG_V("SwScaler: using C scaler for horizontal scaling\n");
2138 if(isPlanarYUV(dstFormat))
2140 if(c->vLumFilterSize==1)
2141 MSG_V("SwScaler: using 1-tap %s \"scaler\" for vertical scaling (YV12 like)\n", (flags & SWS_CPU_CAPS_MMX) ? "MMX" : "C");
2143 MSG_V("SwScaler: using n-tap %s scaler for vertical scaling (YV12 like)\n", (flags & SWS_CPU_CAPS_MMX) ? "MMX" : "C");
2147 if(c->vLumFilterSize==1 && c->vChrFilterSize==2)
2148 MSG_V("SwScaler: using 1-tap %s \"scaler\" for vertical luminance scaling (BGR)\n"
2149 "SwScaler: 2-tap scaler for vertical chrominance scaling (BGR)\n",(flags & SWS_CPU_CAPS_MMX) ? "MMX" : "C");
2150 else if(c->vLumFilterSize==2 && c->vChrFilterSize==2)
2151 MSG_V("SwScaler: using 2-tap linear %s scaler for vertical scaling (BGR)\n", (flags & SWS_CPU_CAPS_MMX) ? "MMX" : "C");
2153 MSG_V("SwScaler: using n-tap %s scaler for vertical scaling (BGR)\n", (flags & SWS_CPU_CAPS_MMX) ? "MMX" : "C");
2156 if(dstFormat==IMGFMT_BGR24)
2157 MSG_V("SwScaler: using %s YV12->BGR24 Converter\n",
2158 (flags & SWS_CPU_CAPS_MMX2) ? "MMX2" : ((flags & SWS_CPU_CAPS_MMX) ? "MMX" : "C"));
2159 else if(dstFormat==IMGFMT_BGR32)
2160 MSG_V("SwScaler: using %s YV12->BGR32 Converter\n", (flags & SWS_CPU_CAPS_MMX) ? "MMX" : "C");
2161 else if(dstFormat==IMGFMT_BGR16)
2162 MSG_V("SwScaler: using %s YV12->BGR16 Converter\n", (flags & SWS_CPU_CAPS_MMX) ? "MMX" : "C");
2163 else if(dstFormat==IMGFMT_BGR15)
2164 MSG_V("SwScaler: using %s YV12->BGR15 Converter\n", (flags & SWS_CPU_CAPS_MMX) ? "MMX" : "C");
2166 MSG_V("SwScaler: %dx%d -> %dx%d\n", srcW, srcH, dstW, dstH);
2168 if(flags & SWS_PRINT_INFO)
2170 MSG_DBG2("SwScaler:Lum srcW=%d srcH=%d dstW=%d dstH=%d xInc=%d yInc=%d\n",
2171 c->srcW, c->srcH, c->dstW, c->dstH, c->lumXInc, c->lumYInc);
2172 MSG_DBG2("SwScaler:Chr srcW=%d srcH=%d dstW=%d dstH=%d xInc=%d yInc=%d\n",
2173 c->chrSrcW, c->chrSrcH, c->chrDstW, c->chrDstH, c->chrXInc, c->chrYInc);
2176 c->swScale= getSwsFunc(flags);
2181 * swscale warper, so we don't need to export the SwsContext.
2182 * assumes planar YUV to be in YUV order instead of YVU
2184 int sws_scale_ordered(SwsContext *c, uint8_t* src[], int srcStride[], int srcSliceY,
2185 int srcSliceH, uint8_t* dst[], int dstStride[]){
2186 //copy strides, so they can safely be modified
2187 int srcStride2[3]= {srcStride[0], srcStride[1], srcStride[2]};
2188 int dstStride2[3]= {dstStride[0], dstStride[1], dstStride[2]};
2189 return c->swScale(c, src, srcStride2, srcSliceY, srcSliceH, dst, dstStride2);
2193 * swscale warper, so we don't need to export the SwsContext
2195 int sws_scale(SwsContext *c, uint8_t* srcParam[], int srcStrideParam[], int srcSliceY,
2196 int srcSliceH, uint8_t* dstParam[], int dstStrideParam[]){
2201 sws_orderYUV(c->origSrcFormat, src, srcStride, srcParam, srcStrideParam);
2202 sws_orderYUV(c->origDstFormat, dst, dstStride, dstParam, dstStrideParam);
2203 //printf("sws: slice %d %d\n", srcSliceY, srcSliceH);
2205 return c->swScale(c, src, srcStride, srcSliceY, srcSliceH, dst, dstStride);
2208 SwsFilter *sws_getDefaultFilter(float lumaGBlur, float chromaGBlur,
2209 float lumaSharpen, float chromaSharpen,
2210 float chromaHShift, float chromaVShift,
2213 SwsFilter *filter= malloc(sizeof(SwsFilter));
2216 filter->lumH= sws_getGaussianVec(lumaGBlur, 3.0);
2217 filter->lumV= sws_getGaussianVec(lumaGBlur, 3.0);
2219 filter->lumH= sws_getIdentityVec();
2220 filter->lumV= sws_getIdentityVec();
2223 if(chromaGBlur!=0.0){
2224 filter->chrH= sws_getGaussianVec(chromaGBlur, 3.0);
2225 filter->chrV= sws_getGaussianVec(chromaGBlur, 3.0);
2227 filter->chrH= sws_getIdentityVec();
2228 filter->chrV= sws_getIdentityVec();
2231 if(chromaSharpen!=0.0){
2232 SwsVector *g= sws_getConstVec(-1.0, 3);
2233 SwsVector *id= sws_getConstVec(10.0/chromaSharpen, 1);
2236 sws_convVec(filter->chrH, id);
2237 sws_convVec(filter->chrV, id);
2242 if(lumaSharpen!=0.0){
2243 SwsVector *g= sws_getConstVec(-1.0, 3);
2244 SwsVector *id= sws_getConstVec(10.0/lumaSharpen, 1);
2247 sws_convVec(filter->lumH, id);
2248 sws_convVec(filter->lumV, id);
2253 if(chromaHShift != 0.0)
2254 sws_shiftVec(filter->chrH, (int)(chromaHShift+0.5));
2256 if(chromaVShift != 0.0)
2257 sws_shiftVec(filter->chrV, (int)(chromaVShift+0.5));
2259 sws_normalizeVec(filter->chrH, 1.0);
2260 sws_normalizeVec(filter->chrV, 1.0);
2261 sws_normalizeVec(filter->lumH, 1.0);
2262 sws_normalizeVec(filter->lumV, 1.0);
2264 if(verbose) sws_printVec(filter->chrH);
2265 if(verbose) sws_printVec(filter->lumH);
2271 * returns a normalized gaussian curve used to filter stuff
2272 * quality=3 is high quality, lowwer is lowwer quality
2274 SwsVector *sws_getGaussianVec(double variance, double quality){
2275 const int length= (int)(variance*quality + 0.5) | 1;
2277 double *coeff= memalign(sizeof(double), length*sizeof(double));
2278 double middle= (length-1)*0.5;
2279 SwsVector *vec= malloc(sizeof(SwsVector));
2282 vec->length= length;
2284 for(i=0; i<length; i++)
2286 double dist= i-middle;
2287 coeff[i]= exp( -dist*dist/(2*variance*variance) ) / sqrt(2*variance*PI);
2290 sws_normalizeVec(vec, 1.0);
2295 SwsVector *sws_getConstVec(double c, int length){
2297 double *coeff= memalign(sizeof(double), length*sizeof(double));
2298 SwsVector *vec= malloc(sizeof(SwsVector));
2301 vec->length= length;
2303 for(i=0; i<length; i++)
2310 SwsVector *sws_getIdentityVec(void){
2311 double *coeff= memalign(sizeof(double), sizeof(double));
2312 SwsVector *vec= malloc(sizeof(SwsVector));
2321 void sws_normalizeVec(SwsVector *a, double height){
2326 for(i=0; i<a->length; i++)
2331 for(i=0; i<a->length; i++)
2335 void sws_scaleVec(SwsVector *a, double scalar){
2338 for(i=0; i<a->length; i++)
2339 a->coeff[i]*= scalar;
2342 static SwsVector *sws_getConvVec(SwsVector *a, SwsVector *b){
2343 int length= a->length + b->length - 1;
2344 double *coeff= memalign(sizeof(double), length*sizeof(double));
2346 SwsVector *vec= malloc(sizeof(SwsVector));
2349 vec->length= length;
2351 for(i=0; i<length; i++) coeff[i]= 0.0;
2353 for(i=0; i<a->length; i++)
2355 for(j=0; j<b->length; j++)
2357 coeff[i+j]+= a->coeff[i]*b->coeff[j];
2364 static SwsVector *sws_sumVec(SwsVector *a, SwsVector *b){
2365 int length= MAX(a->length, b->length);
2366 double *coeff= memalign(sizeof(double), length*sizeof(double));
2368 SwsVector *vec= malloc(sizeof(SwsVector));
2371 vec->length= length;
2373 for(i=0; i<length; i++) coeff[i]= 0.0;
2375 for(i=0; i<a->length; i++) coeff[i + (length-1)/2 - (a->length-1)/2]+= a->coeff[i];
2376 for(i=0; i<b->length; i++) coeff[i + (length-1)/2 - (b->length-1)/2]+= b->coeff[i];
2381 static SwsVector *sws_diffVec(SwsVector *a, SwsVector *b){
2382 int length= MAX(a->length, b->length);
2383 double *coeff= memalign(sizeof(double), length*sizeof(double));
2385 SwsVector *vec= malloc(sizeof(SwsVector));
2388 vec->length= length;
2390 for(i=0; i<length; i++) coeff[i]= 0.0;
2392 for(i=0; i<a->length; i++) coeff[i + (length-1)/2 - (a->length-1)/2]+= a->coeff[i];
2393 for(i=0; i<b->length; i++) coeff[i + (length-1)/2 - (b->length-1)/2]-= b->coeff[i];
2398 /* shift left / or right if "shift" is negative */
2399 static SwsVector *sws_getShiftedVec(SwsVector *a, int shift){
2400 int length= a->length + ABS(shift)*2;
2401 double *coeff= memalign(sizeof(double), length*sizeof(double));
2403 SwsVector *vec= malloc(sizeof(SwsVector));
2406 vec->length= length;
2408 for(i=0; i<length; i++) coeff[i]= 0.0;
2410 for(i=0; i<a->length; i++)
2412 coeff[i + (length-1)/2 - (a->length-1)/2 - shift]= a->coeff[i];
2418 void sws_shiftVec(SwsVector *a, int shift){
2419 SwsVector *shifted= sws_getShiftedVec(a, shift);
2421 a->coeff= shifted->coeff;
2422 a->length= shifted->length;
2426 void sws_addVec(SwsVector *a, SwsVector *b){
2427 SwsVector *sum= sws_sumVec(a, b);
2429 a->coeff= sum->coeff;
2430 a->length= sum->length;
2434 void sws_subVec(SwsVector *a, SwsVector *b){
2435 SwsVector *diff= sws_diffVec(a, b);
2437 a->coeff= diff->coeff;
2438 a->length= diff->length;
2442 void sws_convVec(SwsVector *a, SwsVector *b){
2443 SwsVector *conv= sws_getConvVec(a, b);
2445 a->coeff= conv->coeff;
2446 a->length= conv->length;
2450 SwsVector *sws_cloneVec(SwsVector *a){
2451 double *coeff= memalign(sizeof(double), a->length*sizeof(double));
2453 SwsVector *vec= malloc(sizeof(SwsVector));
2456 vec->length= a->length;
2458 for(i=0; i<a->length; i++) coeff[i]= a->coeff[i];
2463 void sws_printVec(SwsVector *a){
2469 for(i=0; i<a->length; i++)
2470 if(a->coeff[i]>max) max= a->coeff[i];
2472 for(i=0; i<a->length; i++)
2473 if(a->coeff[i]<min) min= a->coeff[i];
2477 for(i=0; i<a->length; i++)
2479 int x= (int)((a->coeff[i]-min)*60.0/range +0.5);
2480 MSG_DBG2("%1.3f ", a->coeff[i]);
2481 for(;x>0; x--) MSG_DBG2(" ");
2486 void sws_freeVec(SwsVector *a){
2488 if(a->coeff) free(a->coeff);
2494 void sws_freeFilter(SwsFilter *filter){
2497 if(filter->lumH) sws_freeVec(filter->lumH);
2498 if(filter->lumV) sws_freeVec(filter->lumV);
2499 if(filter->chrH) sws_freeVec(filter->chrH);
2500 if(filter->chrV) sws_freeVec(filter->chrV);
2505 void sws_freeContext(SwsContext *c){
2511 for(i=0; i<c->vLumBufSize; i++)
2513 if(c->lumPixBuf[i]) free(c->lumPixBuf[i]);
2514 c->lumPixBuf[i]=NULL;
2522 for(i=0; i<c->vChrBufSize; i++)
2524 if(c->chrPixBuf[i]) free(c->chrPixBuf[i]);
2525 c->chrPixBuf[i]=NULL;
2531 if(c->vLumFilter) free(c->vLumFilter);
2532 c->vLumFilter = NULL;
2533 if(c->vChrFilter) free(c->vChrFilter);
2534 c->vChrFilter = NULL;
2535 if(c->hLumFilter) free(c->hLumFilter);
2536 c->hLumFilter = NULL;
2537 if(c->hChrFilter) free(c->hChrFilter);
2538 c->hChrFilter = NULL;
2540 if(c->vLumFilterPos) free(c->vLumFilterPos);
2541 c->vLumFilterPos = NULL;
2542 if(c->vChrFilterPos) free(c->vChrFilterPos);
2543 c->vChrFilterPos = NULL;
2544 if(c->hLumFilterPos) free(c->hLumFilterPos);
2545 c->hLumFilterPos = NULL;
2546 if(c->hChrFilterPos) free(c->hChrFilterPos);
2547 c->hChrFilterPos = NULL;
2549 if(c->lumMmx2Filter) free(c->lumMmx2Filter);
2550 c->lumMmx2Filter=NULL;
2551 if(c->chrMmx2Filter) free(c->chrMmx2Filter);
2552 c->chrMmx2Filter=NULL;
2553 if(c->lumMmx2FilterPos) free(c->lumMmx2FilterPos);
2554 c->lumMmx2FilterPos=NULL;
2555 if(c->chrMmx2FilterPos) free(c->chrMmx2FilterPos);
2556 c->chrMmx2FilterPos=NULL;
2557 if(c->yuvTable) free(c->yuvTable);