2 Copyright (C) 2001-2002 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, BGR32, BGR24, BGR16, BGR15, RGB32, RGB24, Y8/Y800, YVU9/IF09
21 supported output formats: YV12, I420/IYUV, YUY2, {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
56 #include "../config.h"
57 #include "../mangle.h"
65 #include "../cpudetect.h"
67 #include "../libvo/img_format.h"
69 #include "../libvo/fastmemcpy.h"
70 #include "../mp_msg.h"
72 #define MSG_WARN(args...) mp_msg(MSGT_SWS,MSGL_WARN, ##args )
73 #define MSG_FATAL(args...) mp_msg(MSGT_SWS,MSGL_FATAL, ##args )
74 #define MSG_ERR(args...) mp_msg(MSGT_SWS,MSGL_ERR, ##args )
75 #define MSG_V(args...) mp_msg(MSGT_SWS,MSGL_V, ##args )
76 #define MSG_DBG2(args...) mp_msg(MSGT_SWS,MSGL_DBG2, ##args )
77 #define MSG_INFO(args...) mp_msg(MSGT_SWS,MSGL_INFO, ##args )
86 //#define WORDS_BIGENDIAN
89 #define FAST_BGR2YV12 // use 7 bit coeffs instead of 15bit
91 #define RET 0xC3 //near return opcode for X86
94 #define ASSERT(x) assert(x);
102 #define PI 3.14159265358979323846
105 //FIXME replace this with something faster
106 #define isPlanarYUV(x) ((x)==IMGFMT_YV12 || (x)==IMGFMT_I420 || (x)==IMGFMT_YVU9 \
107 || (x)==IMGFMT_444P || (x)==IMGFMT_422P || (x)==IMGFMT_411P)
108 #define isYUV(x) ((x)==IMGFMT_YUY2 || isPlanarYUV(x))
109 #define isGray(x) ((x)==IMGFMT_Y800)
110 #define isRGB(x) (((x)&IMGFMT_RGB_MASK)==IMGFMT_RGB)
111 #define isBGR(x) (((x)&IMGFMT_BGR_MASK)==IMGFMT_BGR)
112 #define isSupportedIn(x) ((x)==IMGFMT_YV12 || (x)==IMGFMT_I420 || (x)==IMGFMT_YUY2 \
113 || (x)==IMGFMT_BGR32|| (x)==IMGFMT_BGR24|| (x)==IMGFMT_BGR16|| (x)==IMGFMT_BGR15\
114 || (x)==IMGFMT_RGB32|| (x)==IMGFMT_RGB24\
115 || (x)==IMGFMT_Y800 || (x)==IMGFMT_YVU9\
116 || (x)==IMGFMT_444P || (x)==IMGFMT_422P || (x)==IMGFMT_411P)
117 #define isSupportedOut(x) ((x)==IMGFMT_YV12 || (x)==IMGFMT_I420 || (x)==IMGFMT_YUY2\
118 || (x)==IMGFMT_444P || (x)==IMGFMT_422P || (x)==IMGFMT_411P\
119 || isRGB(x) || isBGR(x)\
120 || (x)==IMGFMT_Y800 || (x)==IMGFMT_YVU9)
121 #define isPacked(x) ((x)==IMGFMT_YUY2 || isRGB(x) || isBGR(x))
123 #define RGB2YUV_SHIFT 16
124 #define BY ((int)( 0.098*(1<<RGB2YUV_SHIFT)+0.5))
125 #define BV ((int)(-0.071*(1<<RGB2YUV_SHIFT)+0.5))
126 #define BU ((int)( 0.439*(1<<RGB2YUV_SHIFT)+0.5))
127 #define GY ((int)( 0.504*(1<<RGB2YUV_SHIFT)+0.5))
128 #define GV ((int)(-0.368*(1<<RGB2YUV_SHIFT)+0.5))
129 #define GU ((int)(-0.291*(1<<RGB2YUV_SHIFT)+0.5))
130 #define RY ((int)( 0.257*(1<<RGB2YUV_SHIFT)+0.5))
131 #define RV ((int)( 0.439*(1<<RGB2YUV_SHIFT)+0.5))
132 #define RU ((int)(-0.148*(1<<RGB2YUV_SHIFT)+0.5))
134 extern int verbose; // defined in mplayer.c
137 Special versions: fast Y 1:1 scaling (no interpolation in y direction)
140 more intelligent missalignment avoidance for the horizontal scaler
141 write special vertical cubic upscale version
142 Optimize C code (yv12 / minmax)
143 add support for packed pixel yuv input & output
144 add support for Y8 output
145 optimize bgr24 & bgr32
146 add BGR4 output support
147 write special BGR->BGR scaler
148 deglobalize yuv2rgb*.c
151 #define ABS(a) ((a) > 0 ? (a) : (-(a)))
152 #define MIN(a,b) ((a) > (b) ? (b) : (a))
153 #define MAX(a,b) ((a) < (b) ? (b) : (a))
156 #define CAN_COMPILE_X86_ASM
159 #ifdef CAN_COMPILE_X86_ASM
160 static uint64_t __attribute__((aligned(8))) yCoeff= 0x2568256825682568LL;
161 static uint64_t __attribute__((aligned(8))) vrCoeff= 0x3343334333433343LL;
162 static uint64_t __attribute__((aligned(8))) ubCoeff= 0x40cf40cf40cf40cfLL;
163 static uint64_t __attribute__((aligned(8))) vgCoeff= 0xE5E2E5E2E5E2E5E2LL;
164 static uint64_t __attribute__((aligned(8))) ugCoeff= 0xF36EF36EF36EF36ELL;
165 static uint64_t __attribute__((aligned(8))) bF8= 0xF8F8F8F8F8F8F8F8LL;
166 static uint64_t __attribute__((aligned(8))) bFC= 0xFCFCFCFCFCFCFCFCLL;
167 static uint64_t __attribute__((aligned(8))) w400= 0x0400040004000400LL;
168 static uint64_t __attribute__((aligned(8))) w80= 0x0080008000800080LL;
169 static uint64_t __attribute__((aligned(8))) w10= 0x0010001000100010LL;
170 static uint64_t __attribute__((aligned(8))) w02= 0x0002000200020002LL;
171 static uint64_t __attribute__((aligned(8))) bm00001111=0x00000000FFFFFFFFLL;
172 static uint64_t __attribute__((aligned(8))) bm00000111=0x0000000000FFFFFFLL;
173 static uint64_t __attribute__((aligned(8))) bm11111000=0xFFFFFFFFFF000000LL;
174 static uint64_t __attribute__((aligned(8))) bm01010101=0x00FF00FF00FF00FFLL;
176 static volatile uint64_t __attribute__((aligned(8))) b5Dither;
177 static volatile uint64_t __attribute__((aligned(8))) g5Dither;
178 static volatile uint64_t __attribute__((aligned(8))) g6Dither;
179 static volatile uint64_t __attribute__((aligned(8))) r5Dither;
181 static uint64_t __attribute__((aligned(8))) dither4[2]={
182 0x0103010301030103LL,
183 0x0200020002000200LL,};
185 static uint64_t __attribute__((aligned(8))) dither8[2]={
186 0x0602060206020602LL,
187 0x0004000400040004LL,};
189 static uint64_t __attribute__((aligned(8))) b16Mask= 0x001F001F001F001FLL;
190 static uint64_t __attribute__((aligned(8))) g16Mask= 0x07E007E007E007E0LL;
191 static uint64_t __attribute__((aligned(8))) r16Mask= 0xF800F800F800F800LL;
192 static uint64_t __attribute__((aligned(8))) b15Mask= 0x001F001F001F001FLL;
193 static uint64_t __attribute__((aligned(8))) g15Mask= 0x03E003E003E003E0LL;
194 static uint64_t __attribute__((aligned(8))) r15Mask= 0x7C007C007C007C00LL;
196 static uint64_t __attribute__((aligned(8))) M24A= 0x00FF0000FF0000FFLL;
197 static uint64_t __attribute__((aligned(8))) M24B= 0xFF0000FF0000FF00LL;
198 static uint64_t __attribute__((aligned(8))) M24C= 0x0000FF0000FF0000LL;
201 static const uint64_t bgr2YCoeff __attribute__((aligned(8))) = 0x000000210041000DULL;
202 static const uint64_t bgr2UCoeff __attribute__((aligned(8))) = 0x0000FFEEFFDC0038ULL;
203 static const uint64_t bgr2VCoeff __attribute__((aligned(8))) = 0x00000038FFD2FFF8ULL;
205 static const uint64_t bgr2YCoeff __attribute__((aligned(8))) = 0x000020E540830C8BULL;
206 static const uint64_t bgr2UCoeff __attribute__((aligned(8))) = 0x0000ED0FDAC23831ULL;
207 static const uint64_t bgr2VCoeff __attribute__((aligned(8))) = 0x00003831D0E6F6EAULL;
209 static const uint64_t bgr2YOffset __attribute__((aligned(8))) = 0x1010101010101010ULL;
210 static const uint64_t bgr2UVOffset __attribute__((aligned(8)))= 0x8080808080808080ULL;
211 static const uint64_t w1111 __attribute__((aligned(8))) = 0x0001000100010001ULL;
214 // clipping helper table for C implementations:
215 static unsigned char clip_table[768];
217 //global sws_flags from the command line
221 SwsFilter src_filter= {NULL, NULL, NULL, NULL};
223 float sws_lum_gblur= 0.0;
224 float sws_chr_gblur= 0.0;
225 int sws_chr_vshift= 0;
226 int sws_chr_hshift= 0;
227 float sws_chr_sharpen= 0.0;
228 float sws_lum_sharpen= 0.0;
230 /* cpuCaps combined from cpudetect and whats actually compiled in
231 (if there is no support for something compiled in it wont appear here) */
232 static CpuCaps cpuCaps;
234 void (*swScale)(SwsContext *context, uint8_t* src[], int srcStride[], int srcSliceY,
235 int srcSliceH, uint8_t* dst[], int dstStride[])=NULL;
237 static SwsVector *getConvVec(SwsVector *a, SwsVector *b);
238 static inline void orderYUV(int format, uint8_t * sortedP[], int sortedStride[], uint8_t * p[], int stride[]);
239 void *yuv2rgb_c_init (unsigned bpp, int mode, void *table_rV[256], void *table_gU[256], int table_gV[256], void *table_bU[256]);
241 extern const uint8_t dither_2x2_4[2][8];
242 extern const uint8_t dither_2x2_8[2][8];
243 extern const uint8_t dither_8x8_32[8][8];
244 extern const uint8_t dither_8x8_73[8][8];
245 extern const uint8_t dither_8x8_220[8][8];
247 #ifdef CAN_COMPILE_X86_ASM
248 void in_asm_used_var_warning_killer()
250 volatile int i= yCoeff+vrCoeff+ubCoeff+vgCoeff+ugCoeff+bF8+bFC+w400+w80+w10+
251 bm00001111+bm00000111+bm11111000+b16Mask+g16Mask+r16Mask+b15Mask+g15Mask+r15Mask+
252 M24A+M24B+M24C+w02 + b5Dither+g5Dither+r5Dither+g6Dither+dither4[0]+dither8[0]+bm01010101;
257 static int testFormat[]={
274 static uint64_t getSSD(uint8_t *src1, uint8_t *src2, int stride1, int stride2, int w, int h){
280 int d= src1[x + y*stride1] - src2[x + y*stride2];
287 // test by ref -> src -> dst -> out & compare out against ref
288 // ref & out are YV12
289 static void doTest(uint8_t *ref[3], int refStride[3], int w, int h, int srcFormat, int dstFormat,
290 int srcW, int srcH, int dstW, int dstH, int flags){
294 int srcStride[3], dstStride[3];
296 uint64_t ssdY, ssdU, ssdV;
297 SwsContext *srcContext, *dstContext, *outContext;
300 // avoid stride % bpp != 0
301 if(srcFormat==IMGFMT_RGB24 || srcFormat==IMGFMT_BGR24)
302 srcStride[i]= srcW*3;
304 srcStride[i]= srcW*4;
306 if(dstFormat==IMGFMT_RGB24 || dstFormat==IMGFMT_BGR24)
307 dstStride[i]= dstW*3;
309 dstStride[i]= dstW*4;
311 src[i]= malloc(srcStride[i]*srcH);
312 dst[i]= malloc(dstStride[i]*dstH);
313 out[i]= malloc(refStride[i]*h);
316 srcContext= getSwsContext(w, h, IMGFMT_YV12, srcW, srcH, srcFormat, flags, NULL, NULL);
317 dstContext= getSwsContext(srcW, srcH, srcFormat, dstW, dstH, dstFormat, flags, NULL, NULL);
318 outContext= getSwsContext(dstW, dstH, dstFormat, w, h, IMGFMT_YV12, flags, NULL, NULL);
319 if(srcContext==NULL ||dstContext==NULL ||outContext==NULL){
320 printf("Failed allocating swsContext\n");
323 // printf("test %X %X %X -> %X %X %X\n", (int)ref[0], (int)ref[1], (int)ref[2],
324 // (int)src[0], (int)src[1], (int)src[2]);
326 srcContext->swScale(srcContext, ref, refStride, 0, h , src, srcStride);
327 dstContext->swScale(dstContext, src, srcStride, 0, srcH, dst, dstStride);
328 outContext->swScale(outContext, dst, dstStride, 0, dstH, out, refStride);
330 ssdY= getSSD(ref[0], out[0], refStride[0], refStride[0], w, h);
331 ssdU= getSSD(ref[1], out[1], refStride[1], refStride[1], (w+1)>>1, (h+1)>>1);
332 ssdV= getSSD(ref[2], out[2], refStride[2], refStride[2], (w+1)>>1, (h+1)>>1);
334 if(isGray(srcFormat) || isGray(dstFormat)) ssdU=ssdV=0; //FIXME check that output is really gray
340 if(ssdY>100 || ssdU>50 || ssdV>50){
341 printf(" %s %dx%d -> %s %4dx%4d flags=%2d SSD=%5lld,%5lld,%5lld\n",
342 vo_format_name(srcFormat), srcW, srcH,
343 vo_format_name(dstFormat), dstW, dstH,
350 freeSwsContext(srcContext);
351 freeSwsContext(dstContext);
352 freeSwsContext(outContext);
361 static void selfTest(uint8_t *src[3], int stride[3], int w, int h){
362 int srcFormat, dstFormat, srcFormatIndex, dstFormatIndex;
363 int srcW, srcH, dstW, dstH;
366 for(srcFormatIndex=0; ;srcFormatIndex++){
367 srcFormat= testFormat[srcFormatIndex];
368 if(!srcFormat) break;
369 for(dstFormatIndex=0; ;dstFormatIndex++){
370 dstFormat= testFormat[dstFormatIndex];
371 if(!dstFormat) break;
372 if(!isSupportedOut(dstFormat)) continue;
374 vo_format_name(srcFormat),
375 vo_format_name(dstFormat));
379 for(dstW=w; dstW<w*2; dstW+= dstW/3){
380 for(dstH=h; dstH<h*2; dstH+= dstH/3){
381 for(flags=1; flags<33; flags*=2)
382 doTest(src, stride, w, h, srcFormat, dstFormat,
383 srcW, srcH, dstW, dstH, flags);
390 static inline void yuv2yuvXinC(int16_t *lumFilter, int16_t **lumSrc, int lumFilterSize,
391 int16_t *chrFilter, int16_t **chrSrc, int chrFilterSize,
392 uint8_t *dest, uint8_t *uDest, uint8_t *vDest, int dstW, int chrDstW)
394 //FIXME Optimize (just quickly writen not opti..)
396 for(i=0; i<dstW; i++)
400 for(j=0; j<lumFilterSize; j++)
401 val += lumSrc[j][i] * lumFilter[j];
403 dest[i]= MIN(MAX(val>>19, 0), 255);
407 for(i=0; i<chrDstW; i++)
412 for(j=0; j<chrFilterSize; j++)
414 u += chrSrc[j][i] * chrFilter[j];
415 v += chrSrc[j][i + 2048] * chrFilter[j];
418 uDest[i]= MIN(MAX(u>>19, 0), 255);
419 vDest[i]= MIN(MAX(v>>19, 0), 255);
424 #define YSCALE_YUV_2_PACKEDX_C(type) \
425 for(i=0; i<(dstW>>1); i++){\
434 for(j=0; j<lumFilterSize; j++)\
436 Y1 += lumSrc[j][i2] * lumFilter[j];\
437 Y2 += lumSrc[j][i2+1] * lumFilter[j];\
439 for(j=0; j<chrFilterSize; j++)\
441 U += chrSrc[j][i] * chrFilter[j];\
442 V += chrSrc[j][i+2048] * chrFilter[j];\
460 #define YSCALE_YUV_2_RGBX_C(type) \
461 YSCALE_YUV_2_PACKEDX_C(type)\
463 g = c->table_gU[U] + c->table_gV[V];\
466 #define YSCALE_YUV_2_PACKED2_C \
467 for(i=0; i<(dstW>>1); i++){\
469 int Y1= (buf0[i2 ]*yalpha1+buf1[i2 ]*yalpha)>>19;\
470 int Y2= (buf0[i2+1]*yalpha1+buf1[i2+1]*yalpha)>>19;\
471 int U= (uvbuf0[i ]*uvalpha1+uvbuf1[i ]*uvalpha)>>19;\
472 int V= (uvbuf0[i+2048]*uvalpha1+uvbuf1[i+2048]*uvalpha)>>19;\
474 #define YSCALE_YUV_2_RGB2_C(type) \
475 YSCALE_YUV_2_PACKED2_C\
478 g = c->table_gU[U] + c->table_gV[V];\
481 #define YSCALE_YUV_2_PACKED1_C \
482 for(i=0; i<(dstW>>1); i++){\
484 int Y1= buf0[i2 ]>>7;\
485 int Y2= buf0[i2+1]>>7;\
486 int U= (uvbuf1[i ])>>7;\
487 int V= (uvbuf1[i+2048])>>7;\
489 #define YSCALE_YUV_2_RGB1_C(type) \
490 YSCALE_YUV_2_PACKED1_C\
493 g = c->table_gU[U] + c->table_gV[V];\
496 #define YSCALE_YUV_2_PACKED1B_C \
497 for(i=0; i<(dstW>>1); i++){\
499 int Y1= buf0[i2 ]>>7;\
500 int Y2= buf0[i2+1]>>7;\
501 int U= (uvbuf0[i ] + uvbuf1[i ])>>8;\
502 int V= (uvbuf0[i+2048] + uvbuf1[i+2048])>>8;\
504 #define YSCALE_YUV_2_RGB1B_C(type) \
505 YSCALE_YUV_2_PACKED1B_C\
508 g = c->table_gU[U] + c->table_gV[V];\
511 #define YSCALE_YUV_2_ANYRGB_C(func, func2)\
512 switch(c->dstFormat)\
517 ((uint32_t*)dest)[i2+0]= r[Y1] + g[Y1] + b[Y1];\
518 ((uint32_t*)dest)[i2+1]= r[Y2] + g[Y2] + b[Y2];\
523 ((uint8_t*)dest)[0]= r[Y1];\
524 ((uint8_t*)dest)[1]= g[Y1];\
525 ((uint8_t*)dest)[2]= b[Y1];\
526 ((uint8_t*)dest)[3]= r[Y2];\
527 ((uint8_t*)dest)[4]= g[Y2];\
528 ((uint8_t*)dest)[5]= b[Y2];\
529 ((uint8_t*)dest)+=6;\
534 ((uint8_t*)dest)[0]= b[Y1];\
535 ((uint8_t*)dest)[1]= g[Y1];\
536 ((uint8_t*)dest)[2]= r[Y1];\
537 ((uint8_t*)dest)[3]= b[Y2];\
538 ((uint8_t*)dest)[4]= g[Y2];\
539 ((uint8_t*)dest)[5]= r[Y2];\
540 ((uint8_t*)dest)+=6;\
546 const int dr1= dither_2x2_8[y&1 ][0];\
547 const int dg1= dither_2x2_4[y&1 ][0];\
548 const int db1= dither_2x2_8[(y&1)^1][0];\
549 const int dr2= dither_2x2_8[y&1 ][1];\
550 const int dg2= dither_2x2_4[y&1 ][1];\
551 const int db2= dither_2x2_8[(y&1)^1][1];\
553 ((uint16_t*)dest)[i2+0]= r[Y1+dr1] + g[Y1+dg1] + b[Y1+db1];\
554 ((uint16_t*)dest)[i2+1]= r[Y2+dr2] + g[Y2+dg2] + b[Y2+db2];\
561 const int dr1= dither_2x2_8[y&1 ][0];\
562 const int dg1= dither_2x2_8[y&1 ][1];\
563 const int db1= dither_2x2_8[(y&1)^1][0];\
564 const int dr2= dither_2x2_8[y&1 ][1];\
565 const int dg2= dither_2x2_8[y&1 ][0];\
566 const int db2= dither_2x2_8[(y&1)^1][1];\
568 ((uint16_t*)dest)[i2+0]= r[Y1+dr1] + g[Y1+dg1] + b[Y1+db1];\
569 ((uint16_t*)dest)[i2+1]= r[Y2+dr2] + g[Y2+dg2] + b[Y2+db2];\
576 const uint8_t * const d64= dither_8x8_73[y&7];\
577 const uint8_t * const d32= dither_8x8_32[y&7];\
579 ((uint8_t*)dest)[i2+0]= r[Y1+d32[(i2+0)&7]] + g[Y1+d32[(i2+0)&7]] + b[Y1+d64[(i2+0)&7]];\
580 ((uint8_t*)dest)[i2+1]= r[Y2+d32[(i2+1)&7]] + g[Y2+d32[(i2+1)&7]] + b[Y2+d64[(i2+1)&7]];\
587 const uint8_t * const d64= dither_8x8_73 [y&7];\
588 const uint8_t * const d128=dither_8x8_220[y&7];\
590 ((uint8_t*)dest)[i2+0]= r[Y1+d128[(i2+0)&7]] + g[Y1+d64[(i2+0)&7]] + b[Y1+d128[(i2+0)&7]];\
591 ((uint8_t*)dest)[i2+1]= r[Y2+d128[(i2+1)&7]] + g[Y2+d64[(i2+1)&7]] + b[Y2+d128[(i2+1)&7]];\
598 const uint8_t * const d128=dither_8x8_220[y&7];\
599 uint8_t *g= c->table_gU[128] + c->table_gV[128];\
600 for(i=0; i<dstW-7; i+=8){\
602 acc = g[((buf0[i ]*yalpha1+buf1[i ]*yalpha)>>19) + d128[0]];\
603 acc+= acc + g[((buf0[i+1]*yalpha1+buf1[i+1]*yalpha)>>19) + d128[1]];\
604 acc+= acc + g[((buf0[i+2]*yalpha1+buf1[i+2]*yalpha)>>19) + d128[2]];\
605 acc+= acc + g[((buf0[i+3]*yalpha1+buf1[i+3]*yalpha)>>19) + d128[3]];\
606 acc+= acc + g[((buf0[i+4]*yalpha1+buf1[i+4]*yalpha)>>19) + d128[4]];\
607 acc+= acc + g[((buf0[i+5]*yalpha1+buf1[i+5]*yalpha)>>19) + d128[5]];\
608 acc+= acc + g[((buf0[i+6]*yalpha1+buf1[i+6]*yalpha)>>19) + d128[6]];\
609 acc+= acc + g[((buf0[i+7]*yalpha1+buf1[i+7]*yalpha)>>19) + d128[7]];\
610 ((uint8_t*)dest)[0]= acc;\
615 ((uint8_t*)dest)-= dstW>>4;\
619 static int top[1024];\
620 static int last_new[1024][1024];\
621 static int last_in3[1024][1024];\
622 static int drift[1024][1024];\
626 const uint8_t * const d128=dither_8x8_220[y&7];\
631 for(i=dstW>>1; i<dstW; i++){\
632 int in= ((buf0[i ]*yalpha1+buf1[i ]*yalpha)>>19);\
633 int in2 = (76309 * (in - 16) + 32768) >> 16;\
634 int in3 = (in2 < 0) ? 0 : ((in2 > 255) ? 255 : in2);\
635 int old= (left*7 + topLeft + top[i]*5 + top[i+1]*3)/20 + in3\
636 + (last_new[y][i] - in3)*f/256;\
637 int new= old> 128 ? 255 : 0;\
639 error_new+= ABS(last_new[y][i] - new);\
640 error_in3+= ABS(last_in3[y][i] - in3);\
641 f= error_new - error_in3*4;\
646 left= top[i]= old - new;\
647 last_new[y][i]= new;\
648 last_in3[y][i]= in3;\
650 acc+= acc + (new&1);\
652 ((uint8_t*)dest)[0]= acc;\
662 ((uint8_t*)dest)[2*i2+0]= Y1;\
663 ((uint8_t*)dest)[2*i2+1]= U;\
664 ((uint8_t*)dest)[2*i2+2]= Y2;\
665 ((uint8_t*)dest)[2*i2+3]= V;\
671 static inline void yuv2packedXinC(SwsContext *c, int16_t *lumFilter, int16_t **lumSrc, int lumFilterSize,
672 int16_t *chrFilter, int16_t **chrSrc, int chrFilterSize,
673 uint8_t *dest, int dstW, int y)
680 YSCALE_YUV_2_RGBX_C(uint32_t)
681 ((uint32_t*)dest)[i2+0]= r[Y1] + g[Y1] + b[Y1];
682 ((uint32_t*)dest)[i2+1]= r[Y2] + g[Y2] + b[Y2];
686 YSCALE_YUV_2_RGBX_C(uint8_t)
687 ((uint8_t*)dest)[0]= r[Y1];
688 ((uint8_t*)dest)[1]= g[Y1];
689 ((uint8_t*)dest)[2]= b[Y1];
690 ((uint8_t*)dest)[3]= r[Y2];
691 ((uint8_t*)dest)[4]= g[Y2];
692 ((uint8_t*)dest)[5]= b[Y2];
697 YSCALE_YUV_2_RGBX_C(uint8_t)
698 ((uint8_t*)dest)[0]= b[Y1];
699 ((uint8_t*)dest)[1]= g[Y1];
700 ((uint8_t*)dest)[2]= r[Y1];
701 ((uint8_t*)dest)[3]= b[Y2];
702 ((uint8_t*)dest)[4]= g[Y2];
703 ((uint8_t*)dest)[5]= r[Y2];
710 const int dr1= dither_2x2_8[y&1 ][0];
711 const int dg1= dither_2x2_4[y&1 ][0];
712 const int db1= dither_2x2_8[(y&1)^1][0];
713 const int dr2= dither_2x2_8[y&1 ][1];
714 const int dg2= dither_2x2_4[y&1 ][1];
715 const int db2= dither_2x2_8[(y&1)^1][1];
716 YSCALE_YUV_2_RGBX_C(uint16_t)
717 ((uint16_t*)dest)[i2+0]= r[Y1+dr1] + g[Y1+dg1] + b[Y1+db1];
718 ((uint16_t*)dest)[i2+1]= r[Y2+dr2] + g[Y2+dg2] + b[Y2+db2];
725 const int dr1= dither_2x2_8[y&1 ][0];
726 const int dg1= dither_2x2_8[y&1 ][1];
727 const int db1= dither_2x2_8[(y&1)^1][0];
728 const int dr2= dither_2x2_8[y&1 ][1];
729 const int dg2= dither_2x2_8[y&1 ][0];
730 const int db2= dither_2x2_8[(y&1)^1][1];
731 YSCALE_YUV_2_RGBX_C(uint16_t)
732 ((uint16_t*)dest)[i2+0]= r[Y1+dr1] + g[Y1+dg1] + b[Y1+db1];
733 ((uint16_t*)dest)[i2+1]= r[Y2+dr2] + g[Y2+dg2] + b[Y2+db2];
740 const uint8_t * const d64= dither_8x8_73[y&7];
741 const uint8_t * const d32= dither_8x8_32[y&7];
742 YSCALE_YUV_2_RGBX_C(uint8_t)
743 ((uint8_t*)dest)[i2+0]= r[Y1+d32[(i2+0)&7]] + g[Y1+d32[(i2+0)&7]] + b[Y1+d64[(i2+0)&7]];
744 ((uint8_t*)dest)[i2+1]= r[Y2+d32[(i2+1)&7]] + g[Y2+d32[(i2+1)&7]] + b[Y2+d64[(i2+1)&7]];
751 const uint8_t * const d64= dither_8x8_73 [y&7];
752 const uint8_t * const d128=dither_8x8_220[y&7];
753 YSCALE_YUV_2_RGBX_C(uint8_t)
754 ((uint8_t*)dest)[i2+0]= r[Y1+d128[(i2+0)&7]] + g[Y1+d64[(i2+0)&7]] + b[Y1+d128[(i2+0)&7]];
755 ((uint8_t*)dest)[i2+1]= r[Y2+d128[(i2+1)&7]] + g[Y2+d64[(i2+1)&7]] + b[Y2+d128[(i2+1)&7]];
762 const uint8_t * const d128=dither_8x8_220[y&7];
763 uint8_t *g= c->table_gU[128] + c->table_gV[128];
765 for(i=0; i<dstW-1; i+=2){
770 for(j=0; j<lumFilterSize; j++)
772 Y1 += lumSrc[j][i] * lumFilter[j];
773 Y2 += lumSrc[j][i+1] * lumFilter[j];
784 acc+= acc + g[Y1+d128[(i+0)&7]];
785 acc+= acc + g[Y2+d128[(i+1)&7]];
787 ((uint8_t*)dest)[0]= acc;
794 YSCALE_YUV_2_PACKEDX_C(void)
795 ((uint8_t*)dest)[2*i2+0]= Y1;
796 ((uint8_t*)dest)[2*i2+1]= U;
797 ((uint8_t*)dest)[2*i2+2]= Y2;
798 ((uint8_t*)dest)[2*i2+3]= V;
805 //Note: we have C, X86, MMX, MMX2, 3DNOW version therse no 3DNOW+MMX2 one
807 #if !defined (HAVE_MMX) || defined (RUNTIME_CPUDETECT)
811 #ifdef CAN_COMPILE_X86_ASM
813 #if (defined (HAVE_MMX) && !defined (HAVE_3DNOW) && !defined (HAVE_MMX2)) || defined (RUNTIME_CPUDETECT)
817 #if defined (HAVE_MMX2) || defined (RUNTIME_CPUDETECT)
821 #if (defined (HAVE_3DNOW) && !defined (HAVE_MMX2)) || defined (RUNTIME_CPUDETECT)
822 #define COMPILE_3DNOW
824 #endif //CAN_COMPILE_X86_ASM
834 #define RENAME(a) a ## _C
835 #include "swscale_template.c"
838 #ifdef CAN_COMPILE_X86_ASM
847 #define RENAME(a) a ## _X86
848 #include "swscale_template.c"
856 #define RENAME(a) a ## _MMX
857 #include "swscale_template.c"
866 #define RENAME(a) a ## _MMX2
867 #include "swscale_template.c"
876 #define RENAME(a) a ## _3DNow
877 #include "swscale_template.c"
880 #endif //CAN_COMPILE_X86_ASM
882 // minor note: the HAVE_xyz is messed up after that line so dont use it
885 // old global scaler, dont use for new code
886 // will use sws_flags from the command line
887 void SwScale_YV12slice(unsigned char* src[], int srcStride[], int srcSliceY ,
888 int srcSliceH, uint8_t* dst[], int dstStride, int dstbpp,
889 int srcW, int srcH, int dstW, int dstH){
891 static SwsContext *context=NULL;
893 int dstStride3[3]= {dstStride, dstStride>>1, dstStride>>1};
897 case 8 : dstFormat= IMGFMT_Y8; break;
898 case 12: dstFormat= IMGFMT_YV12; break;
899 case 15: dstFormat= IMGFMT_BGR15; break;
900 case 16: dstFormat= IMGFMT_BGR16; break;
901 case 24: dstFormat= IMGFMT_BGR24; break;
902 case 32: dstFormat= IMGFMT_BGR32; break;
906 if(!context) context=getSwsContextFromCmdLine(srcW, srcH, IMGFMT_YV12, dstW, dstH, dstFormat);
908 context->swScale(context, src, srcStride, srcSliceY, srcSliceH, dst, dstStride3);
911 void swsGetFlagsAndFilterFromCmdLine(int *flags, SwsFilter **srcFilterParam, SwsFilter **dstFilterParam)
913 static int firstTime=1;
918 asm volatile("emms\n\t"::: "memory"); //FIXME this shouldnt be required but it IS (even for non mmx versions)
923 *flags= SWS_PRINT_INFO;
925 else if(verbose>1) *flags= SWS_PRINT_INFO;
927 if(src_filter.lumH) freeVec(src_filter.lumH);
928 if(src_filter.lumV) freeVec(src_filter.lumV);
929 if(src_filter.chrH) freeVec(src_filter.chrH);
930 if(src_filter.chrV) freeVec(src_filter.chrV);
932 if(sws_lum_gblur!=0.0){
933 src_filter.lumH= getGaussianVec(sws_lum_gblur, 3.0);
934 src_filter.lumV= getGaussianVec(sws_lum_gblur, 3.0);
936 src_filter.lumH= getIdentityVec();
937 src_filter.lumV= getIdentityVec();
940 if(sws_chr_gblur!=0.0){
941 src_filter.chrH= getGaussianVec(sws_chr_gblur, 3.0);
942 src_filter.chrV= getGaussianVec(sws_chr_gblur, 3.0);
944 src_filter.chrH= getIdentityVec();
945 src_filter.chrV= getIdentityVec();
948 if(sws_chr_sharpen!=0.0){
949 SwsVector *g= getConstVec(-1.0, 3);
950 SwsVector *id= getConstVec(10.0/sws_chr_sharpen, 1);
953 convVec(src_filter.chrH, id);
954 convVec(src_filter.chrV, id);
959 if(sws_lum_sharpen!=0.0){
960 SwsVector *g= getConstVec(-1.0, 3);
961 SwsVector *id= getConstVec(10.0/sws_lum_sharpen, 1);
964 convVec(src_filter.lumH, id);
965 convVec(src_filter.lumV, id);
971 shiftVec(src_filter.chrH, sws_chr_hshift);
974 shiftVec(src_filter.chrV, sws_chr_vshift);
976 normalizeVec(src_filter.chrH, 1.0);
977 normalizeVec(src_filter.chrV, 1.0);
978 normalizeVec(src_filter.lumH, 1.0);
979 normalizeVec(src_filter.lumV, 1.0);
981 if(verbose > 1) printVec(src_filter.chrH);
982 if(verbose > 1) printVec(src_filter.lumH);
986 case 0: *flags|= SWS_FAST_BILINEAR; break;
987 case 1: *flags|= SWS_BILINEAR; break;
988 case 2: *flags|= SWS_BICUBIC; break;
989 case 3: *flags|= SWS_X; break;
990 case 4: *flags|= SWS_POINT; break;
991 case 5: *flags|= SWS_AREA; break;
992 case 6: *flags|= SWS_BICUBLIN; break;
993 case 7: *flags|= SWS_GAUSS; break;
994 case 8: *flags|= SWS_SINC; break;
995 case 9: *flags|= SWS_LANCZOS; break;
996 case 10:*flags|= SWS_SPLINE; break;
997 default:*flags|= SWS_BILINEAR; break;
1000 *srcFilterParam= &src_filter;
1001 *dstFilterParam= NULL;
1004 // will use sws_flags & src_filter (from cmd line)
1005 SwsContext *getSwsContextFromCmdLine(int srcW, int srcH, int srcFormat, int dstW, int dstH, int dstFormat)
1008 SwsFilter *dstFilterParam, *srcFilterParam;
1009 swsGetFlagsAndFilterFromCmdLine(&flags, &srcFilterParam, &dstFilterParam);
1011 return getSwsContext(srcW, srcH, srcFormat, dstW, dstH, dstFormat, flags, srcFilterParam, dstFilterParam);
1014 static double getSplineCoeff(double a, double b, double c, double d, double dist)
1016 // printf("%f %f %f %f %f\n", a,b,c,d,dist);
1017 if(dist<=1.0) return ((d*dist + c)*dist + b)*dist +a;
1018 else return getSplineCoeff( 0.0,
1025 static inline void initFilter(int16_t **outFilter, int16_t **filterPos, int *outFilterSize, int xInc,
1026 int srcW, int dstW, int filterAlign, int one, int flags,
1027 SwsVector *srcFilter, SwsVector *dstFilter)
1033 double *filter=NULL;
1034 double *filter2=NULL;
1037 asm volatile("emms\n\t"::: "memory"); //FIXME this shouldnt be required but it IS (even for non mmx versions)
1040 // Note the +1 is for the MMXscaler which reads over the end
1041 *filterPos = (int16_t*)memalign(8, (dstW+1)*sizeof(int16_t));
1043 if(ABS(xInc - 0x10000) <10) // unscaled
1047 filter= (double*)memalign(8, dstW*sizeof(double)*filterSize);
1048 for(i=0; i<dstW*filterSize; i++) filter[i]=0;
1050 for(i=0; i<dstW; i++)
1052 filter[i*filterSize]=1;
1057 else if(flags&SWS_POINT) // lame looking point sampling mode
1062 filter= (double*)memalign(8, dstW*sizeof(double)*filterSize);
1064 xDstInSrc= xInc/2 - 0x8000;
1065 for(i=0; i<dstW; i++)
1067 int xx= (xDstInSrc - ((filterSize-1)<<15) + (1<<15))>>16;
1069 (*filterPos)[i]= xx;
1074 else if((xInc <= (1<<16) && (flags&SWS_AREA)) || (flags&SWS_FAST_BILINEAR)) // bilinear upscale
1078 if (flags&SWS_BICUBIC) filterSize= 4;
1079 else if(flags&SWS_X ) filterSize= 4;
1080 else filterSize= 2; // SWS_BILINEAR / SWS_AREA
1081 filter= (double*)memalign(8, dstW*sizeof(double)*filterSize);
1083 xDstInSrc= xInc/2 - 0x8000;
1084 for(i=0; i<dstW; i++)
1086 int xx= (xDstInSrc - ((filterSize-1)<<15) + (1<<15))>>16;
1089 (*filterPos)[i]= xx;
1090 //Bilinear upscale / linear interpolate / Area averaging
1091 for(j=0; j<filterSize; j++)
1093 double d= ABS((xx<<16) - xDstInSrc)/(double)(1<<16);
1094 double coeff= 1.0 - d;
1095 if(coeff<0) coeff=0;
1096 filter[i*filterSize + j]= coeff;
1105 double sizeFactor, filterSizeInSrc;
1106 const double xInc1= (double)xInc / (double)(1<<16);
1107 int param= (flags&SWS_PARAM_MASK)>>SWS_PARAM_SHIFT;
1109 if (flags&SWS_BICUBIC) sizeFactor= 4.0;
1110 else if(flags&SWS_X) sizeFactor= 8.0;
1111 else if(flags&SWS_AREA) sizeFactor= 1.0; //downscale only, for upscale it is bilinear
1112 else if(flags&SWS_GAUSS) sizeFactor= 8.0; // infinite ;)
1113 else if(flags&SWS_LANCZOS) sizeFactor= param ? 2.0*param : 6.0;
1114 else if(flags&SWS_SINC) sizeFactor= 20.0; // infinite ;)
1115 else if(flags&SWS_SPLINE) sizeFactor= 20.0; // infinite ;)
1116 else if(flags&SWS_BILINEAR) sizeFactor= 2.0;
1118 sizeFactor= 0.0; //GCC warning killer
1122 if(xInc1 <= 1.0) filterSizeInSrc= sizeFactor; // upscale
1123 else filterSizeInSrc= sizeFactor*srcW / (double)dstW;
1125 filterSize= (int)ceil(1 + filterSizeInSrc); // will be reduced later if possible
1126 if(filterSize > srcW-2) filterSize=srcW-2;
1128 filter= (double*)memalign(16, dstW*sizeof(double)*filterSize);
1130 xDstInSrc= xInc1 / 2.0 - 0.5;
1131 for(i=0; i<dstW; i++)
1133 int xx= (int)(xDstInSrc - (filterSize-1)*0.5 + 0.5);
1135 (*filterPos)[i]= xx;
1136 for(j=0; j<filterSize; j++)
1138 double d= ABS(xx - xDstInSrc)/filterSizeInSrc*sizeFactor;
1140 if(flags & SWS_BICUBIC)
1142 double A= param ? -param*0.01 : -0.60;
1144 // Equation is from VirtualDub
1146 coeff = (1.0 - (A+3.0)*d*d + (A+2.0)*d*d*d);
1148 coeff = (-4.0*A + 8.0*A*d - 5.0*A*d*d + A*d*d*d);
1152 /* else if(flags & SWS_X)
1154 double p= param ? param*0.01 : 0.3;
1155 coeff = d ? sin(d*PI)/(d*PI) : 1.0;
1156 coeff*= pow(2.0, - p*d*d);
1158 else if(flags & SWS_X)
1160 double A= param ? param*0.1 : 1.0;
1166 if(coeff<0.0) coeff= -pow(-coeff, A);
1167 else coeff= pow( coeff, A);
1168 coeff= coeff*0.5 + 0.5;
1170 else if(flags & SWS_AREA)
1172 double srcPixelSize= 1.0/xInc1;
1173 if(d + srcPixelSize/2 < 0.5) coeff= 1.0;
1174 else if(d - srcPixelSize/2 < 0.5) coeff= (0.5-d)/srcPixelSize + 0.5;
1177 else if(flags & SWS_GAUSS)
1179 double p= param ? param*0.1 : 3.0;
1180 coeff = pow(2.0, - p*d*d);
1182 else if(flags & SWS_SINC)
1184 coeff = d ? sin(d*PI)/(d*PI) : 1.0;
1186 else if(flags & SWS_LANCZOS)
1188 double p= param ? param : 3.0;
1189 coeff = d ? sin(d*PI)*sin(d*PI/p)/(d*d*PI*PI/p) : 1.0;
1192 else if(flags & SWS_BILINEAR)
1195 if(coeff<0) coeff=0;
1197 else if(flags & SWS_SPLINE)
1199 double p=-2.196152422706632;
1200 coeff = getSplineCoeff(1.0, 0.0, p, -p-1.0, d);
1203 coeff= 0.0; //GCC warning killer
1207 filter[i*filterSize + j]= coeff;
1214 /* apply src & dst Filter to filter -> filter2
1217 ASSERT(filterSize>0)
1218 filter2Size= filterSize;
1219 if(srcFilter) filter2Size+= srcFilter->length - 1;
1220 if(dstFilter) filter2Size+= dstFilter->length - 1;
1221 ASSERT(filter2Size>0)
1222 filter2= (double*)memalign(8, filter2Size*dstW*sizeof(double));
1224 for(i=0; i<dstW; i++)
1227 SwsVector scaleFilter;
1230 scaleFilter.coeff= filter + i*filterSize;
1231 scaleFilter.length= filterSize;
1233 if(srcFilter) outVec= getConvVec(srcFilter, &scaleFilter);
1234 else outVec= &scaleFilter;
1236 ASSERT(outVec->length == filter2Size)
1239 for(j=0; j<outVec->length; j++)
1241 filter2[i*filter2Size + j]= outVec->coeff[j];
1244 (*filterPos)[i]+= (filterSize-1)/2 - (filter2Size-1)/2;
1246 if(outVec != &scaleFilter) freeVec(outVec);
1248 free(filter); filter=NULL;
1250 /* try to reduce the filter-size (step1 find size and shift left) */
1251 // Assume its near normalized (*0.5 or *2.0 is ok but * 0.001 is not)
1253 for(i=dstW-1; i>=0; i--)
1255 int min= filter2Size;
1259 /* get rid off near zero elements on the left by shifting left */
1260 for(j=0; j<filter2Size; j++)
1263 cutOff += ABS(filter2[i*filter2Size]);
1265 if(cutOff > SWS_MAX_REDUCE_CUTOFF) break;
1267 /* preserve Monotonicity because the core cant handle the filter otherwise */
1268 if(i<dstW-1 && (*filterPos)[i] >= (*filterPos)[i+1]) break;
1270 // Move filter coeffs left
1271 for(k=1; k<filter2Size; k++)
1272 filter2[i*filter2Size + k - 1]= filter2[i*filter2Size + k];
1273 filter2[i*filter2Size + k - 1]= 0.0;
1278 /* count near zeros on the right */
1279 for(j=filter2Size-1; j>0; j--)
1281 cutOff += ABS(filter2[i*filter2Size + j]);
1283 if(cutOff > SWS_MAX_REDUCE_CUTOFF) break;
1287 if(min>minFilterSize) minFilterSize= min;
1290 ASSERT(minFilterSize > 0)
1291 filterSize= (minFilterSize +(filterAlign-1)) & (~(filterAlign-1));
1292 ASSERT(filterSize > 0)
1293 filter= (double*)memalign(8, filterSize*dstW*sizeof(double));
1294 *outFilterSize= filterSize;
1296 if(flags&SWS_PRINT_INFO)
1297 MSG_INFO("SwScaler: reducing / aligning filtersize %d -> %d\n", filter2Size, filterSize);
1298 /* try to reduce the filter-size (step2 reduce it) */
1299 for(i=0; i<dstW; i++)
1303 for(j=0; j<filterSize; j++)
1305 if(j>=filter2Size) filter[i*filterSize + j]= 0.0;
1306 else filter[i*filterSize + j]= filter2[i*filter2Size + j];
1309 free(filter2); filter2=NULL;
1312 //FIXME try to align filterpos if possible
1315 for(i=0; i<dstW; i++)
1318 if((*filterPos)[i] < 0)
1320 // Move filter coeffs left to compensate for filterPos
1321 for(j=1; j<filterSize; j++)
1323 int left= MAX(j + (*filterPos)[i], 0);
1324 filter[i*filterSize + left] += filter[i*filterSize + j];
1325 filter[i*filterSize + j]=0;
1330 if((*filterPos)[i] + filterSize > srcW)
1332 int shift= (*filterPos)[i] + filterSize - srcW;
1333 // Move filter coeffs right to compensate for filterPos
1334 for(j=filterSize-2; j>=0; j--)
1336 int right= MIN(j + shift, filterSize-1);
1337 filter[i*filterSize +right] += filter[i*filterSize +j];
1338 filter[i*filterSize +j]=0;
1340 (*filterPos)[i]= srcW - filterSize;
1344 // Note the +1 is for the MMXscaler which reads over the end
1345 *outFilter= (int16_t*)memalign(8, *outFilterSize*(dstW+1)*sizeof(int16_t));
1346 memset(*outFilter, 0, *outFilterSize*(dstW+1)*sizeof(int16_t));
1348 /* Normalize & Store in outFilter */
1349 for(i=0; i<dstW; i++)
1354 for(j=0; j<filterSize; j++)
1356 sum+= filter[i*filterSize + j];
1359 for(j=0; j<*outFilterSize; j++)
1361 (*outFilter)[i*(*outFilterSize) + j]= (int)(filter[i*filterSize + j]*scale);
1365 (*filterPos)[dstW]= (*filterPos)[dstW-1]; // the MMX scaler will read over the end
1366 for(i=0; i<*outFilterSize; i++)
1368 int j= dstW*(*outFilterSize);
1369 (*outFilter)[j + i]= (*outFilter)[j + i - (*outFilterSize)];
1376 static void initMMX2HScaler(int dstW, int xInc, uint8_t *funnyCode, int16_t *filter, int32_t *filterPos, int numSplits)
1381 int fragmentLengthA;
1385 int fragmentLengthB;
1390 // create an optimized horizontal scaling routine
1398 "movq (%%edx, %%eax), %%mm3 \n\t"
1399 "movd (%%ecx, %%esi), %%mm0 \n\t"
1400 "movd 1(%%ecx, %%esi), %%mm1 \n\t"
1401 "punpcklbw %%mm7, %%mm1 \n\t"
1402 "punpcklbw %%mm7, %%mm0 \n\t"
1403 "pshufw $0xFF, %%mm1, %%mm1 \n\t"
1405 "pshufw $0xFF, %%mm0, %%mm0 \n\t"
1407 "psubw %%mm1, %%mm0 \n\t"
1408 "movl 8(%%ebx, %%eax), %%esi \n\t"
1409 "pmullw %%mm3, %%mm0 \n\t"
1410 "psllw $7, %%mm1 \n\t"
1411 "paddw %%mm1, %%mm0 \n\t"
1413 "movq %%mm0, (%%edi, %%eax) \n\t"
1415 "addl $8, %%eax \n\t"
1430 :"=r" (fragmentA), "=r" (imm8OfPShufW1A), "=r" (imm8OfPShufW2A),
1431 "=r" (fragmentLengthA)
1438 "movq (%%edx, %%eax), %%mm3 \n\t"
1439 "movd (%%ecx, %%esi), %%mm0 \n\t"
1440 "punpcklbw %%mm7, %%mm0 \n\t"
1441 "pshufw $0xFF, %%mm0, %%mm1 \n\t"
1443 "pshufw $0xFF, %%mm0, %%mm0 \n\t"
1445 "psubw %%mm1, %%mm0 \n\t"
1446 "movl 8(%%ebx, %%eax), %%esi \n\t"
1447 "pmullw %%mm3, %%mm0 \n\t"
1448 "psllw $7, %%mm1 \n\t"
1449 "paddw %%mm1, %%mm0 \n\t"
1451 "movq %%mm0, (%%edi, %%eax) \n\t"
1453 "addl $8, %%eax \n\t"
1468 :"=r" (fragmentB), "=r" (imm8OfPShufW1B), "=r" (imm8OfPShufW2B),
1469 "=r" (fragmentLengthB)
1472 xpos= 0; //lumXInc/2 - 0x8000; // difference between pixel centers
1475 for(i=0; i<dstW/numSplits; i++)
1482 int b=((xpos+xInc)>>16) - xx;
1483 int c=((xpos+xInc*2)>>16) - xx;
1484 int d=((xpos+xInc*3)>>16) - xx;
1486 filter[i ] = (( xpos & 0xFFFF) ^ 0xFFFF)>>9;
1487 filter[i+1] = (((xpos+xInc ) & 0xFFFF) ^ 0xFFFF)>>9;
1488 filter[i+2] = (((xpos+xInc*2) & 0xFFFF) ^ 0xFFFF)>>9;
1489 filter[i+3] = (((xpos+xInc*3) & 0xFFFF) ^ 0xFFFF)>>9;
1494 int maxShift= 3-(d+1);
1497 memcpy(funnyCode + fragmentPos, fragmentB, fragmentLengthB);
1499 funnyCode[fragmentPos + imm8OfPShufW1B]=
1500 (a+1) | ((b+1)<<2) | ((c+1)<<4) | ((d+1)<<6);
1501 funnyCode[fragmentPos + imm8OfPShufW2B]=
1502 a | (b<<2) | (c<<4) | (d<<6);
1504 if(i+3>=dstW) shift=maxShift; //avoid overread
1505 else if((filterPos[i/2]&3) <= maxShift) shift=filterPos[i/2]&3; //Align
1507 if(shift && i>=shift)
1509 funnyCode[fragmentPos + imm8OfPShufW1B]+= 0x55*shift;
1510 funnyCode[fragmentPos + imm8OfPShufW2B]+= 0x55*shift;
1511 filterPos[i/2]-=shift;
1514 fragmentPos+= fragmentLengthB;
1521 memcpy(funnyCode + fragmentPos, fragmentA, fragmentLengthA);
1523 funnyCode[fragmentPos + imm8OfPShufW1A]=
1524 funnyCode[fragmentPos + imm8OfPShufW2A]=
1525 a | (b<<2) | (c<<4) | (d<<6);
1527 if(i+4>=dstW) shift=maxShift; //avoid overread
1528 else if((filterPos[i/2]&3) <= maxShift) shift=filterPos[i/2]&3; //partial align
1530 if(shift && i>=shift)
1532 funnyCode[fragmentPos + imm8OfPShufW1A]+= 0x55*shift;
1533 funnyCode[fragmentPos + imm8OfPShufW2A]+= 0x55*shift;
1534 filterPos[i/2]-=shift;
1537 fragmentPos+= fragmentLengthA;
1540 funnyCode[fragmentPos]= RET;
1544 filterPos[i/2]= xpos>>16; // needed to jump to the next part
1549 void SwScale_Init(){
1552 static void globalInit(){
1553 // generating tables:
1555 for(i=0; i<768; i++){
1556 int c= MIN(MAX(i-256, 0), 255);
1562 #ifdef RUNTIME_CPUDETECT
1563 #ifdef CAN_COMPILE_X86_ASM
1564 // ordered per speed fasterst first
1565 if(gCpuCaps.hasMMX2)
1566 swScale= swScale_MMX2;
1567 else if(gCpuCaps.has3DNow)
1568 swScale= swScale_3DNow;
1569 else if(gCpuCaps.hasMMX)
1570 swScale= swScale_MMX;
1576 cpuCaps.hasMMX2 = cpuCaps.hasMMX = cpuCaps.has3DNow = 0;
1578 #else //RUNTIME_CPUDETECT
1580 swScale= swScale_MMX2;
1581 cpuCaps.has3DNow = 0;
1582 #elif defined (HAVE_3DNOW)
1583 swScale= swScale_3DNow;
1584 cpuCaps.hasMMX2 = 0;
1585 #elif defined (HAVE_MMX)
1586 swScale= swScale_MMX;
1587 cpuCaps.hasMMX2 = cpuCaps.has3DNow = 0;
1590 cpuCaps.hasMMX2 = cpuCaps.hasMMX = cpuCaps.has3DNow = 0;
1592 #endif //!RUNTIME_CPUDETECT
1595 static void PlanarToNV12Wrapper(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 if(dstStride[0]==srcStride[0])
1600 memcpy(dst, src[0], srcSliceH*dstStride[0]);
1604 uint8_t *srcPtr= src[0];
1605 uint8_t *dstPtr= dst;
1606 for(i=0; i<srcSliceH; i++)
1608 memcpy(dstPtr, srcPtr, srcStride[0]);
1609 srcPtr+= srcStride[0];
1610 dstPtr+= dstStride[0];
1613 dst = dstParam[1] + dstStride[1]*srcSliceY;
1614 if(c->srcFormat==IMGFMT_YV12)
1615 interleaveBytes( src[1],src[2],dst,c->srcW,srcSliceH,srcStride[1],srcStride[2],dstStride[0] );
1616 else /* I420 & IYUV */
1617 interleaveBytes( src[2],src[1],dst,c->srcW,srcSliceH,srcStride[2],srcStride[1],dstStride[0] );
1621 /* Warper functions for yuv2bgr */
1622 static void planarYuvToBgr(SwsContext *c, uint8_t* src[], int srcStride[], int srcSliceY,
1623 int srcSliceH, uint8_t* dstParam[], int dstStride[]){
1624 uint8_t *dst=dstParam[0] + dstStride[0]*srcSliceY;
1626 if(c->srcFormat==IMGFMT_YV12)
1627 yuv2rgb( dst,src[0],src[1],src[2],c->srcW,srcSliceH,dstStride[0],srcStride[0],srcStride[1] );
1628 else /* I420 & IYUV */
1629 yuv2rgb( dst,src[0],src[2],src[1],c->srcW,srcSliceH,dstStride[0],srcStride[0],srcStride[1] );
1632 static void PlanarToYuy2Wrapper(SwsContext *c, uint8_t* src[], int srcStride[], int srcSliceY,
1633 int srcSliceH, uint8_t* dstParam[], int dstStride[]){
1634 uint8_t *dst=dstParam[0] + dstStride[0]*srcSliceY;
1636 if(c->srcFormat==IMGFMT_YV12)
1637 yv12toyuy2( src[0],src[1],src[2],dst,c->srcW,srcSliceH,srcStride[0],srcStride[1],dstStride[0] );
1638 else /* I420 & IYUV */
1639 yv12toyuy2( src[0],src[2],src[1],dst,c->srcW,srcSliceH,srcStride[0],srcStride[1],dstStride[0] );
1642 /* {RGB,BGR}{15,16,24,32} -> {RGB,BGR}{15,16,24,32} */
1643 static void rgb2rgbWrapper(SwsContext *c, uint8_t* src[], int srcStride[], int srcSliceY,
1644 int srcSliceH, uint8_t* dst[], int dstStride[]){
1645 const int srcFormat= c->srcFormat;
1646 const int dstFormat= c->dstFormat;
1647 const int srcBpp= ((srcFormat&0xFF) + 7)>>3;
1648 const int dstBpp= ((dstFormat&0xFF) + 7)>>3;
1649 const int srcId= (srcFormat&0xFF)>>2; // 1:0, 4:1, 8:2, 15:3, 16:4, 24:6, 32:8
1650 const int dstId= (dstFormat&0xFF)>>2;
1651 void (*conv)(const uint8_t *src, uint8_t *dst, unsigned src_size)=NULL;
1654 if(isBGR(srcFormat) && isBGR(dstFormat)){
1655 switch(srcId | (dstId<<4)){
1656 case 0x34: conv= rgb16to15; break;
1657 case 0x36: conv= rgb24to15; break;
1658 case 0x38: conv= rgb32to15; break;
1659 case 0x43: conv= rgb15to16; break;
1660 case 0x46: conv= rgb24to16; break;
1661 case 0x48: conv= rgb32to16; break;
1662 case 0x63: conv= rgb15to24; break;
1663 case 0x64: conv= rgb16to24; break;
1664 case 0x68: conv= rgb32to24; break;
1665 case 0x83: conv= rgb15to32; break;
1666 case 0x84: conv= rgb16to32; break;
1667 case 0x86: conv= rgb24to32; break;
1668 default: MSG_ERR("swScaler: internal error %s -> %s converter\n",
1669 vo_format_name(srcFormat), vo_format_name(dstFormat)); break;
1671 }else if(isBGR(srcFormat) && isRGB(dstFormat)){
1672 switch(srcId | (dstId<<4)){
1673 case 0x33: conv= rgb15tobgr15; break;
1674 case 0x34: conv= rgb16tobgr15; break;
1675 case 0x36: conv= rgb24tobgr15; break;
1676 case 0x38: conv= rgb32tobgr15; break;
1677 case 0x43: conv= rgb15tobgr16; break;
1678 case 0x44: conv= rgb16tobgr16; break;
1679 case 0x46: conv= rgb24tobgr16; break;
1680 case 0x48: conv= rgb32tobgr16; break;
1681 case 0x63: conv= rgb15tobgr24; break;
1682 case 0x64: conv= rgb16tobgr24; break;
1683 case 0x66: conv= rgb24tobgr24; break;
1684 case 0x68: conv= rgb32tobgr24; break;
1685 case 0x83: conv= rgb15tobgr32; break;
1686 case 0x84: conv= rgb16tobgr32; break;
1687 case 0x86: conv= rgb24tobgr32; break;
1688 case 0x88: conv= rgb32tobgr32; break;
1689 default: MSG_ERR("swScaler: internal error %s -> %s converter\n",
1690 vo_format_name(srcFormat), vo_format_name(dstFormat)); break;
1692 }else if(isRGB(srcFormat) && isRGB(dstFormat)){
1693 switch(srcId | (dstId<<4)){
1694 case 0x34: conv= rgb16to15; break;
1695 case 0x36: conv= rgb24to15; break;
1696 case 0x38: conv= rgb32to15; break;
1697 case 0x43: conv= rgb15to16; break;
1698 case 0x46: conv= rgb24to16; break;
1699 case 0x48: conv= rgb32to16; break;
1700 case 0x63: conv= rgb15to24; break;
1701 case 0x64: conv= rgb16to24; break;
1702 case 0x68: conv= rgb32to24; break;
1703 case 0x83: conv= rgb15to32; break;
1704 case 0x84: conv= rgb16to32; break;
1705 case 0x86: conv= rgb24to32; break;
1706 default: MSG_ERR("swScaler: internal error %s -> %s converter\n",
1707 vo_format_name(srcFormat), vo_format_name(dstFormat)); break;
1709 }else if(isRGB(srcFormat) && isBGR(dstFormat)){
1710 switch(srcId | (dstId<<4)){
1711 case 0x33: conv= rgb15tobgr15; break;
1712 case 0x34: conv= rgb16tobgr15; break;
1713 case 0x36: conv= rgb24tobgr15; break;
1714 case 0x38: conv= rgb32tobgr15; break;
1715 case 0x43: conv= rgb15tobgr16; break;
1716 case 0x44: conv= rgb16tobgr16; break;
1717 case 0x46: conv= rgb24tobgr16; break;
1718 case 0x48: conv= rgb32tobgr16; break;
1719 case 0x63: conv= rgb15tobgr24; break;
1720 case 0x64: conv= rgb16tobgr24; break;
1721 case 0x66: conv= rgb24tobgr24; break;
1722 case 0x68: conv= rgb32tobgr24; break;
1723 case 0x83: conv= rgb15tobgr32; break;
1724 case 0x84: conv= rgb16tobgr32; break;
1725 case 0x86: conv= rgb24tobgr32; break;
1726 case 0x88: conv= rgb32tobgr32; break;
1727 default: MSG_ERR("swScaler: internal error %s -> %s converter\n",
1728 vo_format_name(srcFormat), vo_format_name(dstFormat)); break;
1731 if(dstStride[0]*srcBpp == srcStride[0]*dstBpp)
1732 conv(src[0], dst[0] + dstStride[0]*srcSliceY, srcSliceH*srcStride[0]);
1736 uint8_t *srcPtr= src[0];
1737 uint8_t *dstPtr= dst[0] + dstStride[0]*srcSliceY;
1739 for(i=0; i<srcSliceH; i++)
1741 conv(srcPtr, dstPtr, c->srcW*srcBpp);
1742 srcPtr+= srcStride[0];
1743 dstPtr+= dstStride[0];
1748 static void bgr24toyv12Wrapper(SwsContext *c, uint8_t* src[], int srcStride[], int srcSliceY,
1749 int srcSliceH, uint8_t* dst[], int dstStride[]){
1753 dst[0]+ srcSliceY *dstStride[0],
1754 dst[1]+(srcSliceY>>1)*dstStride[1],
1755 dst[2]+(srcSliceY>>1)*dstStride[2],
1757 dstStride[0], dstStride[1], srcStride[0]);
1760 static void yvu9toyv12Wrapper(SwsContext *c, uint8_t* src[], int srcStride[], int srcSliceY,
1761 int srcSliceH, uint8_t* dst[], int dstStride[]){
1765 if(srcStride[0]==dstStride[0])
1766 memcpy(dst[0]+ srcSliceY*dstStride[0], src[0], srcStride[0]*srcSliceH);
1768 uint8_t *srcPtr= src[0];
1769 uint8_t *dstPtr= dst[0] + dstStride[0]*srcSliceY;
1771 for(i=0; i<srcSliceH; i++)
1773 memcpy(dstPtr, srcPtr, c->srcW);
1774 srcPtr+= srcStride[0];
1775 dstPtr+= dstStride[0];
1779 if(c->dstFormat==IMGFMT_YV12){
1780 planar2x(src[1], dst[1], c->chrSrcW, c->chrSrcH, srcStride[1], dstStride[1]);
1781 planar2x(src[2], dst[2], c->chrSrcW, c->chrSrcH, srcStride[2], dstStride[2]);
1783 planar2x(src[1], dst[2], c->chrSrcW, c->chrSrcH, srcStride[1], dstStride[2]);
1784 planar2x(src[2], dst[1], c->chrSrcW, c->chrSrcH, srcStride[2], dstStride[1]);
1789 * bring pointers in YUV order instead of YVU
1791 static inline void orderYUV(int format, uint8_t * sortedP[], int sortedStride[], uint8_t * p[], int stride[]){
1792 if(format == IMGFMT_YV12 || format == IMGFMT_YVU9
1793 || format == IMGFMT_444P || format == IMGFMT_422P || format == IMGFMT_411P){
1797 sortedStride[0]= stride[0];
1798 sortedStride[1]= stride[1];
1799 sortedStride[2]= stride[2];
1801 else if(isPacked(format) || isGray(format))
1806 sortedStride[0]= stride[0];
1810 else if(format == IMGFMT_I420)
1815 sortedStride[0]= stride[0];
1816 sortedStride[1]= stride[2];
1817 sortedStride[2]= stride[1];
1819 MSG_ERR("internal error in orderYUV\n");
1823 /* unscaled copy like stuff (assumes nearly identical formats) */
1824 static void simpleCopy(SwsContext *c, uint8_t* srcParam[], int srcStrideParam[], int srcSliceY,
1825 int srcSliceH, uint8_t* dstParam[], int dstStrideParam[]){
1832 orderYUV(c->srcFormat, src, srcStride, srcParam, srcStrideParam);
1833 orderYUV(c->dstFormat, dst, dstStride, dstParam, dstStrideParam);
1835 if(isPacked(c->srcFormat))
1837 if(dstStride[0]==srcStride[0])
1838 memcpy(dst[0] + dstStride[0]*srcSliceY, src[0], srcSliceH*dstStride[0]);
1842 uint8_t *srcPtr= src[0];
1843 uint8_t *dstPtr= dst[0] + dstStride[0]*srcSliceY;
1846 /* universal length finder */
1847 while(length+c->srcW <= ABS(dstStride[0])
1848 && length+c->srcW <= ABS(srcStride[0])) length+= c->srcW;
1851 for(i=0; i<srcSliceH; i++)
1853 memcpy(dstPtr, srcPtr, length);
1854 srcPtr+= srcStride[0];
1855 dstPtr+= dstStride[0];
1860 { /* Planar YUV or gray */
1862 for(plane=0; plane<3; plane++)
1864 int length= plane==0 ? c->srcW : -((-c->srcW )>>c->chrDstHSubSample);
1865 int y= plane==0 ? srcSliceY: -((-srcSliceY)>>c->chrDstVSubSample);
1866 int height= plane==0 ? srcSliceH: -((-srcSliceH)>>c->chrDstVSubSample);
1868 if((isGray(c->srcFormat) || isGray(c->dstFormat)) && plane>0)
1870 if(!isGray(c->dstFormat))
1871 memset(dst[plane], 128, dstStride[plane]*height);
1875 if(dstStride[plane]==srcStride[plane])
1876 memcpy(dst[plane] + dstStride[plane]*y, src[plane], height*dstStride[plane]);
1880 uint8_t *srcPtr= src[plane];
1881 uint8_t *dstPtr= dst[plane] + dstStride[plane]*y;
1882 for(i=0; i<height; i++)
1884 memcpy(dstPtr, srcPtr, length);
1885 srcPtr+= srcStride[plane];
1886 dstPtr+= dstStride[plane];
1894 static int remove_dup_fourcc(int fourcc)
1898 case IMGFMT_IYUV: return IMGFMT_I420;
1899 case IMGFMT_Y8 : return IMGFMT_Y800;
1900 case IMGFMT_IF09: return IMGFMT_YVU9;
1901 default: return fourcc;
1905 static void getSubSampleFactors(int *h, int *v, int format){
1913 case IMGFMT_Y800: //FIXME remove after different subsamplings are fully implemented
1940 SwsContext *getSwsContext(int srcW, int srcH, int srcFormat, int dstW, int dstH, int dstFormat, int flags,
1941 SwsFilter *srcFilter, SwsFilter *dstFilter){
1946 int unscaled, needsDither;
1947 SwsFilter dummyFilter= {NULL, NULL, NULL, NULL};
1950 asm volatile("emms\n\t"::: "memory");
1952 if(swScale==NULL) globalInit();
1953 //srcFormat= IMGFMT_Y800;
1954 //dstFormat= IMGFMT_Y800;
1955 /* avoid dupplicate Formats, so we dont need to check to much */
1956 srcFormat = remove_dup_fourcc(srcFormat);
1957 dstFormat = remove_dup_fourcc(dstFormat);
1959 unscaled = (srcW == dstW && srcH == dstH);
1960 needsDither= (isBGR(dstFormat) || isRGB(dstFormat))
1961 && (dstFormat&0xFF)<24
1962 && ((dstFormat&0xFF)<(srcFormat&0xFF) || (!(isRGB(srcFormat) || isBGR(srcFormat))));
1964 if(!isSupportedIn(srcFormat))
1966 MSG_ERR("swScaler: %s is not supported as input format\n", vo_format_name(srcFormat));
1969 if(!isSupportedOut(dstFormat))
1971 MSG_ERR("swScaler: %s is not supported as output format\n", vo_format_name(dstFormat));
1976 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
1978 MSG_ERR("swScaler: %dx%d -> %dx%d is invalid scaling dimension\n",
1979 srcW, srcH, dstW, dstH);
1983 if(!dstFilter) dstFilter= &dummyFilter;
1984 if(!srcFilter) srcFilter= &dummyFilter;
1986 c= memalign(64, sizeof(SwsContext));
1987 memset(c, 0, sizeof(SwsContext));
1993 c->lumXInc= ((srcW<<16) + (dstW>>1))/dstW;
1994 c->lumYInc= ((srcH<<16) + (dstH>>1))/dstH;
1996 c->dstFormat= dstFormat;
1997 c->srcFormat= srcFormat;
2000 if(dstFilter->lumV!=NULL && dstFilter->lumV->length>1) usesFilter=1;
2001 if(dstFilter->lumH!=NULL && dstFilter->lumH->length>1) usesFilter=1;
2002 if(dstFilter->chrV!=NULL && dstFilter->chrV->length>1) usesFilter=1;
2003 if(dstFilter->chrH!=NULL && dstFilter->chrH->length>1) usesFilter=1;
2004 if(srcFilter->lumV!=NULL && srcFilter->lumV->length>1) usesFilter=1;
2005 if(srcFilter->lumH!=NULL && srcFilter->lumH->length>1) usesFilter=1;
2006 if(srcFilter->chrV!=NULL && srcFilter->chrV->length>1) usesFilter=1;
2007 if(srcFilter->chrH!=NULL && srcFilter->chrH->length>1) usesFilter=1;
2009 getSubSampleFactors(&c->chrSrcHSubSample, &c->chrSrcVSubSample, srcFormat);
2010 getSubSampleFactors(&c->chrDstHSubSample, &c->chrDstVSubSample, dstFormat);
2012 // reuse chroma for 2 pixles rgb/bgr unless user wants full chroma interpolation
2013 if((isBGR(dstFormat) || isRGB(dstFormat)) && !(flags&SWS_FULL_CHR_H_INT)) c->chrDstHSubSample=1;
2015 // drop some chroma lines if the user wants it
2016 c->vChrDrop= (flags&SWS_SRC_V_CHR_DROP_MASK)>>SWS_SRC_V_CHR_DROP_SHIFT;
2017 c->chrSrcVSubSample+= c->vChrDrop;
2019 // drop every 2. pixel for chroma calculation unless user wants full chroma
2020 if((isBGR(srcFormat) || isRGB(srcFormat)) && !(flags&SWS_FULL_CHR_H_INP))
2021 c->chrSrcHSubSample=1;
2023 c->chrIntHSubSample= c->chrDstHSubSample;
2024 c->chrIntVSubSample= c->chrSrcVSubSample;
2026 // note the -((-x)>>y) is so that we allways round toward +inf
2027 c->chrSrcW= -((-srcW) >> c->chrSrcHSubSample);
2028 c->chrSrcH= -((-srcH) >> c->chrSrcVSubSample);
2029 c->chrDstW= -((-dstW) >> c->chrDstHSubSample);
2030 c->chrDstH= -((-dstH) >> c->chrDstVSubSample);
2032 if(isBGR(dstFormat))
2033 c->yuvTable= yuv2rgb_c_init(dstFormat & 0xFF, MODE_RGB, c->table_rV, c->table_gU, c->table_gV, c->table_bU);
2034 if(isRGB(dstFormat))
2035 c->yuvTable= yuv2rgb_c_init(dstFormat & 0xFF, MODE_BGR, c->table_rV, c->table_gU, c->table_gV, c->table_bU);
2037 /* unscaled special Cases */
2038 if(unscaled && !usesFilter)
2041 if((srcFormat == IMGFMT_YV12||srcFormat==IMGFMT_I420)&&dstFormat == IMGFMT_NV12)
2043 c->swScale= PlanarToNV12Wrapper;
2045 if(flags&SWS_PRINT_INFO)
2046 MSG_INFO("SwScaler: using unscaled %s -> %s special converter\n",
2047 vo_format_name(srcFormat), vo_format_name(dstFormat));
2051 if((srcFormat==IMGFMT_YV12 || srcFormat==IMGFMT_I420) && isBGR(dstFormat))
2053 // FIXME multiple yuv2rgb converters wont work that way cuz that thing is full of globals&statics
2054 //FIXME rgb vs. bgr ?
2055 #ifdef WORDS_BIGENDIAN
2056 if(dstFormat==IMGFMT_BGR32)
2057 yuv2rgb_init( dstFormat&0xFF /* =bpp */, MODE_BGR);
2059 yuv2rgb_init( dstFormat&0xFF /* =bpp */, MODE_RGB);
2061 yuv2rgb_init( dstFormat&0xFF /* =bpp */, MODE_RGB);
2063 c->swScale= planarYuvToBgr;
2065 if(flags&SWS_PRINT_INFO)
2066 MSG_INFO("SwScaler: using unscaled %s -> %s special converter\n",
2067 vo_format_name(srcFormat), vo_format_name(dstFormat));
2072 if( srcFormat == dstFormat
2073 || (srcFormat==IMGFMT_YV12 && dstFormat==IMGFMT_I420)
2074 || (srcFormat==IMGFMT_I420 && dstFormat==IMGFMT_YV12)
2075 || (isPlanarYUV(srcFormat) && isGray(dstFormat))
2076 || (isPlanarYUV(dstFormat) && isGray(srcFormat))
2079 c->swScale= simpleCopy;
2081 if(flags&SWS_PRINT_INFO)
2082 MSG_INFO("SwScaler: using unscaled %s -> %s special converter\n",
2083 vo_format_name(srcFormat), vo_format_name(dstFormat));
2087 if( srcFormat==IMGFMT_YVU9 && (dstFormat==IMGFMT_YV12 || dstFormat==IMGFMT_I420) )
2089 c->swScale= yvu9toyv12Wrapper;
2091 if(flags&SWS_PRINT_INFO)
2092 MSG_INFO("SwScaler: using unscaled %s -> %s special converter\n",
2093 vo_format_name(srcFormat), vo_format_name(dstFormat));
2098 if(srcFormat==IMGFMT_BGR24 && dstFormat==IMGFMT_YV12)
2099 c->swScale= bgr24toyv12Wrapper;
2101 /* rgb/bgr -> rgb/bgr (no dither needed forms) */
2102 if( (isBGR(srcFormat) || isRGB(srcFormat))
2103 && (isBGR(dstFormat) || isRGB(dstFormat))
2105 c->swScale= rgb2rgbWrapper;
2107 /* LQ converters if -sws 0 or -sws 4*/
2108 if(c->flags&(SWS_FAST_BILINEAR|SWS_POINT)){
2109 /* rgb/bgr -> rgb/bgr (dither needed forms) */
2110 if( (isBGR(srcFormat) || isRGB(srcFormat))
2111 && (isBGR(dstFormat) || isRGB(dstFormat))
2113 c->swScale= rgb2rgbWrapper;
2116 if((srcFormat == IMGFMT_YV12||srcFormat==IMGFMT_I420)&&dstFormat == IMGFMT_YUY2)
2118 c->swScale= PlanarToYuy2Wrapper;
2120 if(flags&SWS_PRINT_INFO)
2121 MSG_INFO("SwScaler: using unscaled %s -> %s special converter\n",
2122 vo_format_name(srcFormat), vo_format_name(dstFormat));
2128 if(flags&SWS_PRINT_INFO)
2129 MSG_INFO("SwScaler: using unscaled %s -> %s special converter\n",
2130 vo_format_name(srcFormat), vo_format_name(dstFormat));
2137 c->canMMX2BeUsed= (dstW >=srcW && (dstW&31)==0 && (srcW&15)==0) ? 1 : 0;
2138 if(!c->canMMX2BeUsed && dstW >=srcW && (srcW&15)==0 && (flags&SWS_FAST_BILINEAR))
2140 if(flags&SWS_PRINT_INFO)
2141 MSG_INFO("SwScaler: output Width is not a multiple of 32 -> no MMX2 scaler\n");
2147 c->chrXInc= ((c->chrSrcW<<16) + (c->chrDstW>>1))/c->chrDstW;
2148 c->chrYInc= ((c->chrSrcH<<16) + (c->chrDstH>>1))/c->chrDstH;
2150 // match pixel 0 of the src to pixel 0 of dst and match pixel n-2 of src to pixel n-2 of dst
2151 // but only for the FAST_BILINEAR mode otherwise do correct scaling
2152 // n-2 is the last chrominance sample available
2153 // this is not perfect, but noone shuld notice the difference, the more correct variant
2154 // would be like the vertical one, but that would require some special code for the
2155 // first and last pixel
2156 if(flags&SWS_FAST_BILINEAR)
2158 if(c->canMMX2BeUsed)
2163 //we dont use the x86asm scaler if mmx is available
2164 else if(cpuCaps.hasMMX)
2166 c->lumXInc = ((srcW-2)<<16)/(dstW-2) - 20;
2167 c->chrXInc = ((c->chrSrcW-2)<<16)/(c->chrDstW-2) - 20;
2171 /* precalculate horizontal scaler filter coefficients */
2173 const int filterAlign= cpuCaps.hasMMX ? 4 : 1;
2175 initFilter(&c->hLumFilter, &c->hLumFilterPos, &c->hLumFilterSize, c->lumXInc,
2176 srcW , dstW, filterAlign, 1<<14,
2177 (flags&SWS_BICUBLIN) ? (flags|SWS_BICUBIC) : flags,
2178 srcFilter->lumH, dstFilter->lumH);
2179 initFilter(&c->hChrFilter, &c->hChrFilterPos, &c->hChrFilterSize, c->chrXInc,
2180 c->chrSrcW, c->chrDstW, filterAlign, 1<<14,
2181 (flags&SWS_BICUBLIN) ? (flags|SWS_BILINEAR) : flags,
2182 srcFilter->chrH, dstFilter->chrH);
2185 // cant downscale !!!
2186 if(c->canMMX2BeUsed && (flags & SWS_FAST_BILINEAR))
2188 c->lumMmx2Filter = (int16_t*)memalign(8, (dstW /8+8)*sizeof(int16_t));
2189 c->chrMmx2Filter = (int16_t*)memalign(8, (c->chrDstW /4+8)*sizeof(int16_t));
2190 c->lumMmx2FilterPos= (int32_t*)memalign(8, (dstW /2/8+8)*sizeof(int32_t));
2191 c->chrMmx2FilterPos= (int32_t*)memalign(8, (c->chrDstW/2/4+8)*sizeof(int32_t));
2193 initMMX2HScaler( dstW, c->lumXInc, c->funnyYCode , c->lumMmx2Filter, c->lumMmx2FilterPos, 8);
2194 initMMX2HScaler(c->chrDstW, c->chrXInc, c->funnyUVCode, c->chrMmx2Filter, c->chrMmx2FilterPos, 4);
2197 } // Init Horizontal stuff
2201 /* precalculate vertical scaler filter coefficients */
2202 initFilter(&c->vLumFilter, &c->vLumFilterPos, &c->vLumFilterSize, c->lumYInc,
2203 srcH , dstH, 1, (1<<12)-4,
2204 (flags&SWS_BICUBLIN) ? (flags|SWS_BICUBIC) : flags,
2205 srcFilter->lumV, dstFilter->lumV);
2206 initFilter(&c->vChrFilter, &c->vChrFilterPos, &c->vChrFilterSize, c->chrYInc,
2207 c->chrSrcH, c->chrDstH, 1, (1<<12)-4,
2208 (flags&SWS_BICUBLIN) ? (flags|SWS_BILINEAR) : flags,
2209 srcFilter->chrV, dstFilter->chrV);
2211 // Calculate Buffer Sizes so that they wont run out while handling these damn slices
2212 c->vLumBufSize= c->vLumFilterSize;
2213 c->vChrBufSize= c->vChrFilterSize;
2214 for(i=0; i<dstH; i++)
2216 int chrI= i*c->chrDstH / dstH;
2217 int nextSlice= MAX(c->vLumFilterPos[i ] + c->vLumFilterSize - 1,
2218 ((c->vChrFilterPos[chrI] + c->vChrFilterSize - 1)<<c->chrSrcVSubSample));
2219 nextSlice&= ~3; // Slices start at boundaries which are divisable through 4
2220 if(c->vLumFilterPos[i ] + c->vLumBufSize < nextSlice)
2221 c->vLumBufSize= nextSlice - c->vLumFilterPos[i ];
2222 if(c->vChrFilterPos[chrI] + c->vChrBufSize < (nextSlice>>c->chrSrcVSubSample))
2223 c->vChrBufSize= (nextSlice>>c->chrSrcVSubSample) - c->vChrFilterPos[chrI];
2226 // allocate pixbufs (we use dynamic allocation because otherwise we would need to
2227 c->lumPixBuf= (int16_t**)memalign(4, c->vLumBufSize*2*sizeof(int16_t*));
2228 c->chrPixBuf= (int16_t**)memalign(4, c->vChrBufSize*2*sizeof(int16_t*));
2229 //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)
2230 for(i=0; i<c->vLumBufSize; i++)
2231 c->lumPixBuf[i]= c->lumPixBuf[i+c->vLumBufSize]= (uint16_t*)memalign(8, 4000);
2232 for(i=0; i<c->vChrBufSize; i++)
2233 c->chrPixBuf[i]= c->chrPixBuf[i+c->vChrBufSize]= (uint16_t*)memalign(8, 8000);
2235 //try to avoid drawing green stuff between the right end and the stride end
2236 for(i=0; i<c->vLumBufSize; i++) memset(c->lumPixBuf[i], 0, 4000);
2237 for(i=0; i<c->vChrBufSize; i++) memset(c->chrPixBuf[i], 64, 8000);
2239 ASSERT(c->chrDstH <= dstH)
2241 // pack filter data for mmx code
2244 c->lumMmxFilter= (int16_t*)memalign(8, c->vLumFilterSize* dstH*4*sizeof(int16_t));
2245 c->chrMmxFilter= (int16_t*)memalign(8, c->vChrFilterSize*c->chrDstH*4*sizeof(int16_t));
2246 for(i=0; i<c->vLumFilterSize*dstH; i++)
2247 c->lumMmxFilter[4*i]=c->lumMmxFilter[4*i+1]=c->lumMmxFilter[4*i+2]=c->lumMmxFilter[4*i+3]=
2249 for(i=0; i<c->vChrFilterSize*c->chrDstH; i++)
2250 c->chrMmxFilter[4*i]=c->chrMmxFilter[4*i+1]=c->chrMmxFilter[4*i+2]=c->chrMmxFilter[4*i+3]=
2254 if(flags&SWS_PRINT_INFO)
2257 char *dither= " dithered";
2261 if(flags&SWS_FAST_BILINEAR)
2262 MSG_INFO("\nSwScaler: FAST_BILINEAR scaler, ");
2263 else if(flags&SWS_BILINEAR)
2264 MSG_INFO("\nSwScaler: BILINEAR scaler, ");
2265 else if(flags&SWS_BICUBIC)
2266 MSG_INFO("\nSwScaler: BICUBIC scaler, ");
2267 else if(flags&SWS_X)
2268 MSG_INFO("\nSwScaler: Experimental scaler, ");
2269 else if(flags&SWS_POINT)
2270 MSG_INFO("\nSwScaler: Nearest Neighbor / POINT scaler, ");
2271 else if(flags&SWS_AREA)
2272 MSG_INFO("\nSwScaler: Area Averageing scaler, ");
2273 else if(flags&SWS_BICUBLIN)
2274 MSG_INFO("\nSwScaler: luma BICUBIC / chroma BILINEAR scaler, ");
2275 else if(flags&SWS_GAUSS)
2276 MSG_INFO("\nSwScaler: Gaussian scaler, ");
2277 else if(flags&SWS_SINC)
2278 MSG_INFO("\nSwScaler: Sinc scaler, ");
2279 else if(flags&SWS_LANCZOS)
2280 MSG_INFO("\nSwScaler: Lanczos scaler, ");
2281 else if(flags&SWS_SPLINE)
2282 MSG_INFO("\nSwScaler: Bicubic spline scaler, ");
2284 MSG_INFO("\nSwScaler: ehh flags invalid?! ");
2286 if(dstFormat==IMGFMT_BGR15 || dstFormat==IMGFMT_BGR16)
2287 MSG_INFO("from %s to%s %s ",
2288 vo_format_name(srcFormat), dither, vo_format_name(dstFormat));
2290 MSG_INFO("from %s to %s ",
2291 vo_format_name(srcFormat), vo_format_name(dstFormat));
2294 MSG_INFO("using MMX2\n");
2295 else if(cpuCaps.has3DNow)
2296 MSG_INFO("using 3DNOW\n");
2297 else if(cpuCaps.hasMMX)
2298 MSG_INFO("using MMX\n");
2300 MSG_INFO("using C\n");
2303 if((flags & SWS_PRINT_INFO) && verbose>0)
2307 if(c->canMMX2BeUsed && (flags&SWS_FAST_BILINEAR))
2308 MSG_V("SwScaler: using FAST_BILINEAR MMX2 scaler for horizontal scaling\n");
2311 if(c->hLumFilterSize==4)
2312 MSG_V("SwScaler: using 4-tap MMX scaler for horizontal luminance scaling\n");
2313 else if(c->hLumFilterSize==8)
2314 MSG_V("SwScaler: using 8-tap MMX scaler for horizontal luminance scaling\n");
2316 MSG_V("SwScaler: using n-tap MMX scaler for horizontal luminance scaling\n");
2318 if(c->hChrFilterSize==4)
2319 MSG_V("SwScaler: using 4-tap MMX scaler for horizontal chrominance scaling\n");
2320 else if(c->hChrFilterSize==8)
2321 MSG_V("SwScaler: using 8-tap MMX scaler for horizontal chrominance scaling\n");
2323 MSG_V("SwScaler: using n-tap MMX scaler for horizontal chrominance scaling\n");
2329 MSG_V("SwScaler: using X86-Asm scaler for horizontal scaling\n");
2331 if(flags & SWS_FAST_BILINEAR)
2332 MSG_V("SwScaler: using FAST_BILINEAR C scaler for horizontal scaling\n");
2334 MSG_V("SwScaler: using C scaler for horizontal scaling\n");
2337 if(isPlanarYUV(dstFormat))
2339 if(c->vLumFilterSize==1)
2340 MSG_V("SwScaler: using 1-tap %s \"scaler\" for vertical scaling (YV12 like)\n", cpuCaps.hasMMX ? "MMX" : "C");
2342 MSG_V("SwScaler: using n-tap %s scaler for vertical scaling (YV12 like)\n", cpuCaps.hasMMX ? "MMX" : "C");
2346 if(c->vLumFilterSize==1 && c->vChrFilterSize==2)
2347 MSG_V("SwScaler: using 1-tap %s \"scaler\" for vertical luminance scaling (BGR)\n"
2348 "SwScaler: 2-tap scaler for vertical chrominance scaling (BGR)\n",cpuCaps.hasMMX ? "MMX" : "C");
2349 else if(c->vLumFilterSize==2 && c->vChrFilterSize==2)
2350 MSG_V("SwScaler: using 2-tap linear %s scaler for vertical scaling (BGR)\n", cpuCaps.hasMMX ? "MMX" : "C");
2352 MSG_V("SwScaler: using n-tap %s scaler for vertical scaling (BGR)\n", cpuCaps.hasMMX ? "MMX" : "C");
2355 if(dstFormat==IMGFMT_BGR24)
2356 MSG_V("SwScaler: using %s YV12->BGR24 Converter\n",
2357 cpuCaps.hasMMX2 ? "MMX2" : (cpuCaps.hasMMX ? "MMX" : "C"));
2358 else if(dstFormat==IMGFMT_BGR32)
2359 MSG_V("SwScaler: using %s YV12->BGR32 Converter\n", cpuCaps.hasMMX ? "MMX" : "C");
2360 else if(dstFormat==IMGFMT_BGR16)
2361 MSG_V("SwScaler: using %s YV12->BGR16 Converter\n", cpuCaps.hasMMX ? "MMX" : "C");
2362 else if(dstFormat==IMGFMT_BGR15)
2363 MSG_V("SwScaler: using %s YV12->BGR15 Converter\n", cpuCaps.hasMMX ? "MMX" : "C");
2365 MSG_V("SwScaler: %dx%d -> %dx%d\n", srcW, srcH, dstW, dstH);
2367 if((flags & SWS_PRINT_INFO) && verbose>1)
2369 MSG_DBG2("SwScaler:Lum srcW=%d srcH=%d dstW=%d dstH=%d xInc=%d yInc=%d\n",
2370 c->srcW, c->srcH, c->dstW, c->dstH, c->lumXInc, c->lumYInc);
2371 MSG_DBG2("SwScaler:Chr srcW=%d srcH=%d dstW=%d dstH=%d xInc=%d yInc=%d\n",
2372 c->chrSrcW, c->chrSrcH, c->chrDstW, c->chrDstH, c->chrXInc, c->chrYInc);
2375 c->swScale= swScale;
2380 * returns a normalized gaussian curve used to filter stuff
2381 * quality=3 is high quality, lowwer is lowwer quality
2384 SwsVector *getGaussianVec(double variance, double quality){
2385 const int length= (int)(variance*quality + 0.5) | 1;
2387 double *coeff= memalign(sizeof(double), length*sizeof(double));
2388 double middle= (length-1)*0.5;
2389 SwsVector *vec= malloc(sizeof(SwsVector));
2392 vec->length= length;
2394 for(i=0; i<length; i++)
2396 double dist= i-middle;
2397 coeff[i]= exp( -dist*dist/(2*variance*variance) ) / sqrt(2*variance*PI);
2400 normalizeVec(vec, 1.0);
2405 SwsVector *getConstVec(double c, int length){
2407 double *coeff= memalign(sizeof(double), length*sizeof(double));
2408 SwsVector *vec= malloc(sizeof(SwsVector));
2411 vec->length= length;
2413 for(i=0; i<length; i++)
2420 SwsVector *getIdentityVec(void){
2421 double *coeff= memalign(sizeof(double), sizeof(double));
2422 SwsVector *vec= malloc(sizeof(SwsVector));
2431 void normalizeVec(SwsVector *a, double height){
2436 for(i=0; i<a->length; i++)
2441 for(i=0; i<a->length; i++)
2445 void scaleVec(SwsVector *a, double scalar){
2448 for(i=0; i<a->length; i++)
2449 a->coeff[i]*= scalar;
2452 static SwsVector *getConvVec(SwsVector *a, SwsVector *b){
2453 int length= a->length + b->length - 1;
2454 double *coeff= memalign(sizeof(double), length*sizeof(double));
2456 SwsVector *vec= malloc(sizeof(SwsVector));
2459 vec->length= length;
2461 for(i=0; i<length; i++) coeff[i]= 0.0;
2463 for(i=0; i<a->length; i++)
2465 for(j=0; j<b->length; j++)
2467 coeff[i+j]+= a->coeff[i]*b->coeff[j];
2474 static SwsVector *sumVec(SwsVector *a, SwsVector *b){
2475 int length= MAX(a->length, b->length);
2476 double *coeff= memalign(sizeof(double), length*sizeof(double));
2478 SwsVector *vec= malloc(sizeof(SwsVector));
2481 vec->length= length;
2483 for(i=0; i<length; i++) coeff[i]= 0.0;
2485 for(i=0; i<a->length; i++) coeff[i + (length-1)/2 - (a->length-1)/2]+= a->coeff[i];
2486 for(i=0; i<b->length; i++) coeff[i + (length-1)/2 - (b->length-1)/2]+= b->coeff[i];
2491 static SwsVector *diffVec(SwsVector *a, SwsVector *b){
2492 int length= MAX(a->length, b->length);
2493 double *coeff= memalign(sizeof(double), length*sizeof(double));
2495 SwsVector *vec= malloc(sizeof(SwsVector));
2498 vec->length= length;
2500 for(i=0; i<length; i++) coeff[i]= 0.0;
2502 for(i=0; i<a->length; i++) coeff[i + (length-1)/2 - (a->length-1)/2]+= a->coeff[i];
2503 for(i=0; i<b->length; i++) coeff[i + (length-1)/2 - (b->length-1)/2]-= b->coeff[i];
2508 /* shift left / or right if "shift" is negative */
2509 static SwsVector *getShiftedVec(SwsVector *a, int shift){
2510 int length= a->length + ABS(shift)*2;
2511 double *coeff= memalign(sizeof(double), length*sizeof(double));
2513 SwsVector *vec= malloc(sizeof(SwsVector));
2516 vec->length= length;
2518 for(i=0; i<length; i++) coeff[i]= 0.0;
2520 for(i=0; i<a->length; i++)
2522 coeff[i + (length-1)/2 - (a->length-1)/2 - shift]= a->coeff[i];
2528 void shiftVec(SwsVector *a, int shift){
2529 SwsVector *shifted= getShiftedVec(a, shift);
2531 a->coeff= shifted->coeff;
2532 a->length= shifted->length;
2536 void addVec(SwsVector *a, SwsVector *b){
2537 SwsVector *sum= sumVec(a, b);
2539 a->coeff= sum->coeff;
2540 a->length= sum->length;
2544 void subVec(SwsVector *a, SwsVector *b){
2545 SwsVector *diff= diffVec(a, b);
2547 a->coeff= diff->coeff;
2548 a->length= diff->length;
2552 void convVec(SwsVector *a, SwsVector *b){
2553 SwsVector *conv= getConvVec(a, b);
2555 a->coeff= conv->coeff;
2556 a->length= conv->length;
2560 SwsVector *cloneVec(SwsVector *a){
2561 double *coeff= memalign(sizeof(double), a->length*sizeof(double));
2563 SwsVector *vec= malloc(sizeof(SwsVector));
2566 vec->length= a->length;
2568 for(i=0; i<a->length; i++) coeff[i]= a->coeff[i];
2573 void printVec(SwsVector *a){
2579 for(i=0; i<a->length; i++)
2580 if(a->coeff[i]>max) max= a->coeff[i];
2582 for(i=0; i<a->length; i++)
2583 if(a->coeff[i]<min) min= a->coeff[i];
2587 for(i=0; i<a->length; i++)
2589 int x= (int)((a->coeff[i]-min)*60.0/range +0.5);
2590 MSG_DBG2("%1.3f ", a->coeff[i]);
2591 for(;x>0; x--) MSG_DBG2(" ");
2596 void freeVec(SwsVector *a){
2598 if(a->coeff) free(a->coeff);
2604 void freeSwsContext(SwsContext *c){
2610 for(i=0; i<c->vLumBufSize; i++)
2612 if(c->lumPixBuf[i]) free(c->lumPixBuf[i]);
2613 c->lumPixBuf[i]=NULL;
2621 for(i=0; i<c->vChrBufSize; i++)
2623 if(c->chrPixBuf[i]) free(c->chrPixBuf[i]);
2624 c->chrPixBuf[i]=NULL;
2630 if(c->vLumFilter) free(c->vLumFilter);
2631 c->vLumFilter = NULL;
2632 if(c->vChrFilter) free(c->vChrFilter);
2633 c->vChrFilter = NULL;
2634 if(c->hLumFilter) free(c->hLumFilter);
2635 c->hLumFilter = NULL;
2636 if(c->hChrFilter) free(c->hChrFilter);
2637 c->hChrFilter = NULL;
2639 if(c->vLumFilterPos) free(c->vLumFilterPos);
2640 c->vLumFilterPos = NULL;
2641 if(c->vChrFilterPos) free(c->vChrFilterPos);
2642 c->vChrFilterPos = NULL;
2643 if(c->hLumFilterPos) free(c->hLumFilterPos);
2644 c->hLumFilterPos = NULL;
2645 if(c->hChrFilterPos) free(c->hChrFilterPos);
2646 c->hChrFilterPos = NULL;
2648 if(c->lumMmxFilter) free(c->lumMmxFilter);
2649 c->lumMmxFilter = NULL;
2650 if(c->chrMmxFilter) free(c->chrMmxFilter);
2651 c->chrMmxFilter = NULL;
2653 if(c->lumMmx2Filter) free(c->lumMmx2Filter);
2654 c->lumMmx2Filter=NULL;
2655 if(c->chrMmx2Filter) free(c->chrMmx2Filter);
2656 c->chrMmx2Filter=NULL;
2657 if(c->lumMmx2FilterPos) free(c->lumMmx2FilterPos);
2658 c->lumMmx2FilterPos=NULL;
2659 if(c->chrMmx2FilterPos) free(c->chrMmx2FilterPos);
2660 c->chrMmx2FilterPos=NULL;
2661 if(c->yuvTable) free(c->yuvTable);