added rtp protocol
[libav.git] / postproc / swscale.c
CommitLineData
fe8054c0
MN
1/*
2 Copyright (C) 2001-2002 Michael Niedermayer <michaelni@gmx.at>
3
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.
31190492 8
fe8054c0
MN
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.
31190492 13
fe8054c0
MN
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
17*/
783e9cc9 18
28bf81c9 19/*
0c51ef97 20 supported Input formats: YV12, I420/IYUV, YUY2, BGR32, BGR24, BGR16, BGR15, RGB32, RGB24, Y8/Y800, YVU9/IF09
cf7d1c1a 21 supported output formats: YV12, I420/IYUV, {BGR,RGB}{1,4,8,15,16,24,32}, Y8/Y800, YVU9/IF09
e09d12f4 22 {BGR,RGB}{1,4,8,15,16} support dithering
a861d4d7 23
e09d12f4
MN
24 unscaled special converters (YV12=I420=IYUV, Y800=Y8)
25 YV12 -> {BGR,RGB}{1,4,8,15,16,24,32}
26 x -> x
27 YUV9 -> YV12
28 YUV9/YV12 -> Y800
29 Y800 -> YUV9/YV12
b935781b
MN
30 BGR24 -> BGR32 & RGB24 -> RGB32
31 BGR32 -> BGR24 & RGB32 -> RGB24
4bb3fa5e 32 BGR15 -> BGR16
b935781b
MN
33*/
34
35/*
e09d12f4
MN
36tested special converters (most are tested actually but i didnt write it down ...)
37 YV12 -> BGR16
b935781b 38 YV12 -> YV12
4bb3fa5e 39 BGR15 -> BGR16
1e1c4fe9 40 BGR16 -> BGR16
e09d12f4 41 YVU9 -> YV12
b935781b
MN
42
43untested special converters
1e1c4fe9
MN
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
b935781b
MN
47 BGR24 -> BGR32 & RGB24 -> RGB32
48 BGR32 -> BGR24 & RGB32 -> RGB24
ec22603f 49 BGR24 -> YV12
28bf81c9
MN
50*/
51
d3f41512 52#include <inttypes.h>
dda87e9f 53#include <string.h>
077ea8a7 54#include <math.h>
c1b0bfb4 55#include <stdio.h>
d3f41512 56#include "../config.h"
9b464428 57#include "../mangle.h"
81b7c056 58#include <assert.h>
c1b0bfb4
MN
59#ifdef HAVE_MALLOC_H
60#include <malloc.h>
b6acbc3c
BS
61#else
62#include <stdlib.h>
c1b0bfb4 63#endif
d604bab9 64#include "swscale.h"
7630f2e0 65#include "../cpudetect.h"
a861d4d7 66#include "../bswap.h"
28bf81c9 67#include "../libvo/img_format.h"
37079906 68#include "rgb2rgb.h"
b0db4198 69#include "../libvo/fastmemcpy.h"
4a53a912 70#include "../mp_msg.h"
0d9f3d85
A
71
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 )
78
541c4eb9 79#undef MOVNTQ
7d7f78b5 80#undef PAVGB
d3f41512 81
783e9cc9 82//#undef HAVE_MMX2
7f56a527 83//#define HAVE_3DNOW
d3f41512 84//#undef HAVE_MMX
783e9cc9 85//#undef ARCH_X86
2ba1bff0 86//#define WORDS_BIGENDIAN
d604bab9 87#define DITHER1XBPP
d3f41512 88
ac6a2e45
MN
89#define FAST_BGR2YV12 // use 7 bit coeffs instead of 15bit
90
1e621b18 91#define RET 0xC3 //near return opcode for X86
c1b0bfb4 92
28bf81c9 93#ifdef MP_DEBUG
81b7c056 94#define ASSERT(x) assert(x);
28bf81c9 95#else
c1b0bfb4 96#define ASSERT(x) ;
28bf81c9
MN
97#endif
98
99#ifdef M_PI
100#define PI M_PI
101#else
102#define PI 3.14159265358979323846
103#endif
c1b0bfb4 104
6c7506de 105//FIXME replace this with something faster
44c1035c
MN
106#define isPlanarYUV(x) ((x)==IMGFMT_YV12 || (x)==IMGFMT_I420 || (x)==IMGFMT_YVU9)
107#define isYUV(x) ((x)==IMGFMT_YUY2 || isPlanarYUV(x))
44c1035c 108#define isGray(x) ((x)==IMGFMT_Y800)
cf7d1c1a
MN
109#define isRGB(x) (((x)&IMGFMT_RGB_MASK)==IMGFMT_RGB)
110#define isBGR(x) (((x)&IMGFMT_BGR_MASK)==IMGFMT_BGR)
6ff0ad6b 111#define isSupportedIn(x) ((x)==IMGFMT_YV12 || (x)==IMGFMT_I420 || (x)==IMGFMT_YUY2 \
b72034dd 112 || (x)==IMGFMT_BGR32|| (x)==IMGFMT_BGR24|| (x)==IMGFMT_BGR16|| (x)==IMGFMT_BGR15\
a861d4d7 113 || (x)==IMGFMT_RGB32|| (x)==IMGFMT_RGB24\
e616aa93 114 || (x)==IMGFMT_Y800 || (x)==IMGFMT_YVU9)
6ff0ad6b 115#define isSupportedOut(x) ((x)==IMGFMT_YV12 || (x)==IMGFMT_I420 \
cf7d1c1a 116 || isRGB(x) || isBGR(x)\
e616aa93 117 || (x)==IMGFMT_Y800 || (x)==IMGFMT_YVU9)
44c1035c 118#define isPacked(x) ((x)==IMGFMT_YUY2 || isRGB(x) || isBGR(x))
6ff0ad6b
MN
119
120#define RGB2YUV_SHIFT 16
1e621b18
MN
121#define BY ((int)( 0.098*(1<<RGB2YUV_SHIFT)+0.5))
122#define BV ((int)(-0.071*(1<<RGB2YUV_SHIFT)+0.5))
123#define BU ((int)( 0.439*(1<<RGB2YUV_SHIFT)+0.5))
124#define GY ((int)( 0.504*(1<<RGB2YUV_SHIFT)+0.5))
125#define GV ((int)(-0.368*(1<<RGB2YUV_SHIFT)+0.5))
126#define GU ((int)(-0.291*(1<<RGB2YUV_SHIFT)+0.5))
127#define RY ((int)( 0.257*(1<<RGB2YUV_SHIFT)+0.5))
128#define RV ((int)( 0.439*(1<<RGB2YUV_SHIFT)+0.5))
129#define RU ((int)(-0.148*(1<<RGB2YUV_SHIFT)+0.5))
6c7506de 130
e3d2500f 131extern int verbose; // defined in mplayer.c
783e9cc9
MN
132/*
133NOTES
d604bab9 134Special versions: fast Y 1:1 scaling (no interpolation in y direction)
31190492 135
783e9cc9 136TODO
d604bab9 137more intelligent missalignment avoidance for the horizontal scaler
c1b0bfb4
MN
138write special vertical cubic upscale version
139Optimize C code (yv12 / minmax)
ff7ba856 140add support for packed pixel yuv input & output
6ff0ad6b
MN
141add support for Y8 output
142optimize bgr24 & bgr32
ff7ba856 143add BGR4 output support
1e621b18 144write special BGR->BGR scaler
37079906 145deglobalize yuv2rgb*.c
783e9cc9 146*/
31190492 147
d604bab9 148#define ABS(a) ((a) > 0 ? (a) : (-(a)))
2ff198c1
MN
149#define MIN(a,b) ((a) > (b) ? (b) : (a))
150#define MAX(a,b) ((a) < (b) ? (b) : (a))
d604bab9 151
7630f2e0
MN
152#ifdef ARCH_X86
153#define CAN_COMPILE_X86_ASM
d604bab9
MN
154#endif
155
7630f2e0 156#ifdef CAN_COMPILE_X86_ASM
d604bab9 157static uint64_t __attribute__((aligned(8))) yCoeff= 0x2568256825682568LL;
390b20a6
MN
158static uint64_t __attribute__((aligned(8))) vrCoeff= 0x3343334333433343LL;
159static uint64_t __attribute__((aligned(8))) ubCoeff= 0x40cf40cf40cf40cfLL;
160static uint64_t __attribute__((aligned(8))) vgCoeff= 0xE5E2E5E2E5E2E5E2LL;
161static uint64_t __attribute__((aligned(8))) ugCoeff= 0xF36EF36EF36EF36ELL;
f62255fb
MN
162static uint64_t __attribute__((aligned(8))) bF8= 0xF8F8F8F8F8F8F8F8LL;
163static uint64_t __attribute__((aligned(8))) bFC= 0xFCFCFCFCFCFCFCFCLL;
d604bab9
MN
164static uint64_t __attribute__((aligned(8))) w400= 0x0400040004000400LL;
165static uint64_t __attribute__((aligned(8))) w80= 0x0080008000800080LL;
166static uint64_t __attribute__((aligned(8))) w10= 0x0010001000100010LL;
077ea8a7 167static uint64_t __attribute__((aligned(8))) w02= 0x0002000200020002LL;
d604bab9
MN
168static uint64_t __attribute__((aligned(8))) bm00001111=0x00000000FFFFFFFFLL;
169static uint64_t __attribute__((aligned(8))) bm00000111=0x0000000000FFFFFFLL;
170static uint64_t __attribute__((aligned(8))) bm11111000=0xFFFFFFFFFF000000LL;
6ff0ad6b 171static uint64_t __attribute__((aligned(8))) bm01010101=0x00FF00FF00FF00FFLL;
d604bab9 172
3fe8e8f0
MN
173static volatile uint64_t __attribute__((aligned(8))) b5Dither;
174static volatile uint64_t __attribute__((aligned(8))) g5Dither;
175static volatile uint64_t __attribute__((aligned(8))) g6Dither;
176static volatile uint64_t __attribute__((aligned(8))) r5Dither;
d8fa3c54
MN
177
178static uint64_t __attribute__((aligned(8))) dither4[2]={
179 0x0103010301030103LL,
180 0x0200020002000200LL,};
181
182static uint64_t __attribute__((aligned(8))) dither8[2]={
183 0x0602060206020602LL,
184 0x0004000400040004LL,};
d604bab9
MN
185
186static uint64_t __attribute__((aligned(8))) b16Mask= 0x001F001F001F001FLL;
187static uint64_t __attribute__((aligned(8))) g16Mask= 0x07E007E007E007E0LL;
188static uint64_t __attribute__((aligned(8))) r16Mask= 0xF800F800F800F800LL;
189static uint64_t __attribute__((aligned(8))) b15Mask= 0x001F001F001F001FLL;
190static uint64_t __attribute__((aligned(8))) g15Mask= 0x03E003E003E003E0LL;
191static uint64_t __attribute__((aligned(8))) r15Mask= 0x7C007C007C007C00LL;
192
99d2cb72
MN
193static uint64_t __attribute__((aligned(8))) M24A= 0x00FF0000FF0000FFLL;
194static uint64_t __attribute__((aligned(8))) M24B= 0xFF0000FF0000FF00LL;
195static uint64_t __attribute__((aligned(8))) M24C= 0x0000FF0000FF0000LL;
196
ac6a2e45
MN
197#ifdef FAST_BGR2YV12
198static const uint64_t bgr2YCoeff __attribute__((aligned(8))) = 0x000000210041000DULL;
4342fc14
MN
199static const uint64_t bgr2UCoeff __attribute__((aligned(8))) = 0x0000FFEEFFDC0038ULL;
200static const uint64_t bgr2VCoeff __attribute__((aligned(8))) = 0x00000038FFD2FFF8ULL;
ac6a2e45
MN
201#else
202static const uint64_t bgr2YCoeff __attribute__((aligned(8))) = 0x000020E540830C8BULL;
4342fc14
MN
203static const uint64_t bgr2UCoeff __attribute__((aligned(8))) = 0x0000ED0FDAC23831ULL;
204static const uint64_t bgr2VCoeff __attribute__((aligned(8))) = 0x00003831D0E6F6EAULL;
ac6a2e45
MN
205#endif
206static const uint64_t bgr2YOffset __attribute__((aligned(8))) = 0x1010101010101010ULL;
4342fc14 207static const uint64_t bgr2UVOffset __attribute__((aligned(8)))= 0x8080808080808080ULL;
ac6a2e45 208static const uint64_t w1111 __attribute__((aligned(8))) = 0x0001000100010001ULL;
d604bab9 209#endif
783e9cc9
MN
210
211// clipping helper table for C implementations:
212static unsigned char clip_table[768];
213
28bf81c9 214//global sws_flags from the command line
1f347f22 215int sws_flags=2;
077ea8a7 216
5cebb24b
MN
217//global srcFilter
218SwsFilter src_filter= {NULL, NULL, NULL, NULL};
219
220float sws_lum_gblur= 0.0;
221float sws_chr_gblur= 0.0;
222int sws_chr_vshift= 0;
223int sws_chr_hshift= 0;
5521b193
MN
224float sws_chr_sharpen= 0.0;
225float sws_lum_sharpen= 0.0;
5cebb24b 226
28bf81c9
MN
227/* cpuCaps combined from cpudetect and whats actually compiled in
228 (if there is no support for something compiled in it wont appear here) */
229static CpuCaps cpuCaps;
d3f41512 230
28bf81c9
MN
231void (*swScale)(SwsContext *context, uint8_t* src[], int srcStride[], int srcSliceY,
232 int srcSliceH, uint8_t* dst[], int dstStride[])=NULL;
2ff198c1 233
5cebb24b 234static SwsVector *getConvVec(SwsVector *a, SwsVector *b);
5859233b 235static inline void orderYUV(int format, uint8_t * sortedP[], int sortedStride[], uint8_t * p[], int stride[]);
cf7d1c1a 236void *yuv2rgb_c_init (unsigned bpp, int mode, void *table_rV[256], void *table_gU[256], int table_gV[256], void *table_bU[256]);
5859233b 237
cf7d1c1a
MN
238extern const uint8_t dither_2x2_4[2][8];
239extern const uint8_t dither_2x2_8[2][8];
240extern const uint8_t dither_8x8_32[8][8];
241extern const uint8_t dither_8x8_73[8][8];
242extern const uint8_t dither_8x8_220[8][8];
5cebb24b 243
7630f2e0 244#ifdef CAN_COMPILE_X86_ASM
96034638
MN
245void in_asm_used_var_warning_killer()
246{
077ea8a7 247 volatile int i= yCoeff+vrCoeff+ubCoeff+vgCoeff+ugCoeff+bF8+bFC+w400+w80+w10+
5ac80202 248 bm00001111+bm00000111+bm11111000+b16Mask+g16Mask+r16Mask+b15Mask+g15Mask+r15Mask+
6ff0ad6b 249 M24A+M24B+M24C+w02 + b5Dither+g5Dither+r5Dither+g6Dither+dither4[0]+dither8[0]+bm01010101;
96034638
MN
250 if(i) i=0;
251}
252#endif
d604bab9 253
c7a810cc 254static int testFormat[]={
e616aa93 255IMGFMT_YVU9,
c7a810cc
MN
256IMGFMT_YV12,
257//IMGFMT_IYUV,
258IMGFMT_I420,
259IMGFMT_BGR15,
260IMGFMT_BGR16,
261IMGFMT_BGR24,
262IMGFMT_BGR32,
81a571a8
MN
263IMGFMT_RGB24,
264IMGFMT_RGB32,
c7a810cc
MN
265//IMGFMT_Y8,
266IMGFMT_Y800,
267//IMGFMT_YUY2,
2680
269};
270
271static uint64_t getSSD(uint8_t *src1, uint8_t *src2, int stride1, int stride2, int w, int h){
272 int x,y;
273 uint64_t ssd=0;
274
275 for(y=0; y<h; y++){
276 for(x=0; x<w; x++){
277 int d= src1[x + y*stride1] - src2[x + y*stride2];
278 ssd+= d*d;
279 }
280 }
281 return ssd;
282}
283
284// test by ref -> src -> dst -> out & compare out against ref
285// ref & out are YV12
286static void doTest(uint8_t *ref[3], int refStride[3], int w, int h, int srcFormat, int dstFormat,
287 int srcW, int srcH, int dstW, int dstH, int flags){
288 uint8_t *src[3];
289 uint8_t *dst[3];
290 uint8_t *out[3];
291 int srcStride[3], dstStride[3];
292 int i;
293 uint64_t ssdY, ssdU, ssdV;
294 SwsContext *srcContext, *dstContext, *outContext;
295
296 for(i=0; i<3; i++){
e09d12f4
MN
297 // avoid stride % bpp != 0
298 if(srcFormat==IMGFMT_RGB24 || srcFormat==IMGFMT_BGR24)
299 srcStride[i]= srcW*3;
300 else
301 srcStride[i]= srcW*4;
302
303 if(dstFormat==IMGFMT_RGB24 || dstFormat==IMGFMT_BGR24)
304 dstStride[i]= dstW*3;
305 else
306 dstStride[i]= dstW*4;
307
c7a810cc
MN
308 src[i]= malloc(srcStride[i]*srcH);
309 dst[i]= malloc(dstStride[i]*dstH);
310 out[i]= malloc(refStride[i]*h);
311 }
312
313 srcContext= getSwsContext(w, h, IMGFMT_YV12, srcW, srcH, srcFormat, flags, NULL, NULL);
314 dstContext= getSwsContext(srcW, srcH, srcFormat, dstW, dstH, dstFormat, flags, NULL, NULL);
315 outContext= getSwsContext(dstW, dstH, dstFormat, w, h, IMGFMT_YV12, flags, NULL, NULL);
316 if(srcContext==NULL ||dstContext==NULL ||outContext==NULL){
317 printf("Failed allocating swsContext\n");
318 goto end;
319 }
320// printf("test %X %X %X -> %X %X %X\n", (int)ref[0], (int)ref[1], (int)ref[2],
321// (int)src[0], (int)src[1], (int)src[2]);
322
323 srcContext->swScale(srcContext, ref, refStride, 0, h , src, srcStride);
324 dstContext->swScale(dstContext, src, srcStride, 0, srcH, dst, dstStride);
325 outContext->swScale(outContext, dst, dstStride, 0, dstH, out, refStride);
326
327 ssdY= getSSD(ref[0], out[0], refStride[0], refStride[0], w, h);
328 ssdU= getSSD(ref[1], out[1], refStride[1], refStride[1], (w+1)>>1, (h+1)>>1);
329 ssdV= getSSD(ref[2], out[2], refStride[2], refStride[2], (w+1)>>1, (h+1)>>1);
330
331 if(isGray(srcFormat) || isGray(dstFormat)) ssdU=ssdV=0; //FIXME check that output is really gray
332
333 ssdY/= w*h;
334 ssdU/= w*h/4;
335 ssdV/= w*h/4;
336
e616aa93 337 if(ssdY>100 || ssdU>50 || ssdV>50){
c7a810cc
MN
338 printf(" %s %dx%d -> %s %4dx%4d flags=%2d SSD=%5lld,%5lld,%5lld\n",
339 vo_format_name(srcFormat), srcW, srcH,
340 vo_format_name(dstFormat), dstW, dstH,
341 flags,
342 ssdY, ssdU, ssdV);
343 }
344
345 end:
346
347 freeSwsContext(srcContext);
348 freeSwsContext(dstContext);
349 freeSwsContext(outContext);
350
351 for(i=0; i<3; i++){
352 free(src[i]);
353 free(dst[i]);
354 free(out[i]);
355 }
356}
357
358static void selfTest(uint8_t *src[3], int stride[3], int w, int h){
359 int srcFormat, dstFormat, srcFormatIndex, dstFormatIndex;
360 int srcW, srcH, dstW, dstH;
361 int flags;
362
363 for(srcFormatIndex=0; ;srcFormatIndex++){
364 srcFormat= testFormat[srcFormatIndex];
365 if(!srcFormat) break;
366 for(dstFormatIndex=0; ;dstFormatIndex++){
367 dstFormat= testFormat[dstFormatIndex];
368 if(!dstFormat) break;
369 if(!isSupportedOut(dstFormat)) continue;
e09d12f4
MN
370printf("%s -> %s\n",
371 vo_format_name(srcFormat),
372 vo_format_name(dstFormat));
c7a810cc
MN
373
374 srcW= w+w/3;
375 srcH= h+h/3;
376 for(dstW=w; dstW<w*2; dstW+= dstW/3){
377 for(dstH=h; dstH<h*2; dstH+= dstH/3){
378 for(flags=1; flags<33; flags*=2)
379 doTest(src, stride, w, h, srcFormat, dstFormat,
380 srcW, srcH, dstW, dstH, flags);
381 }
382 }
383 }
384 }
385}
386
5859233b 387static inline void yuv2yuvXinC(int16_t *lumFilter, int16_t **lumSrc, int lumFilterSize,
e3d2500f 388 int16_t *chrFilter, int16_t **chrSrc, int chrFilterSize,
5859233b 389 uint8_t *dest, uint8_t *uDest, uint8_t *vDest, int dstW, int chrDstW)
e3d2500f
MN
390{
391 //FIXME Optimize (just quickly writen not opti..)
392 int i;
5859233b 393 for(i=0; i<dstW; i++)
e3d2500f
MN
394 {
395 int val=0;
396 int j;
397 for(j=0; j<lumFilterSize; j++)
398 val += lumSrc[j][i] * lumFilter[j];
399
400 dest[i]= MIN(MAX(val>>19, 0), 255);
401 }
402
403 if(uDest != NULL)
5859233b 404 for(i=0; i<chrDstW; i++)
e3d2500f
MN
405 {
406 int u=0;
407 int v=0;
408 int j;
627690b5 409 for(j=0; j<chrFilterSize; j++)
e3d2500f
MN
410 {
411 u += chrSrc[j][i] * chrFilter[j];
412 v += chrSrc[j][i + 2048] * chrFilter[j];
413 }
414
415 uDest[i]= MIN(MAX(u>>19, 0), 255);
416 vDest[i]= MIN(MAX(v>>19, 0), 255);
417 }
418}
419
cf7d1c1a
MN
420#define YSCALE_YUV_2_RGBX_C(type) \
421 for(i=0; i<(dstW>>1); i++){\
422 int j;\
423 int Y1=0;\
424 int Y2=0;\
425 int U=0;\
426 int V=0;\
427 type *r, *b, *g;\
428 const int i2= 2*i;\
429 \
430 for(j=0; j<lumFilterSize; j++)\
431 {\
432 Y1 += lumSrc[j][i2] * lumFilter[j];\
433 Y2 += lumSrc[j][i2+1] * lumFilter[j];\
434 }\
435 for(j=0; j<chrFilterSize; j++)\
436 {\
437 U += chrSrc[j][i] * chrFilter[j];\
438 V += chrSrc[j][i+2048] * chrFilter[j];\
439 }\
440 Y1>>=19;\
441 Y2>>=19;\
442 U >>=19;\
443 V >>=19;\
444 if((Y1|Y2|U|V)&256)\
445 {\
446 if(Y1>255) Y1=255;\
447 else if(Y1<0)Y1=0;\
448 if(Y2>255) Y2=255;\
449 else if(Y2<0)Y2=0;\
450 if(U>255) U=255;\
451 else if(U<0) U=0;\
452 if(V>255) V=255;\
453 else if(V<0) V=0;\
454 }\
455 r = c->table_rV[V];\
456 g = c->table_gU[U] + c->table_gV[V];\
457 b = c->table_bU[U];\
458
459#define YSCALE_YUV_2_RGB2_C(type) \
460 for(i=0; i<(dstW>>1); i++){\
461 const int i2= 2*i;\
462 int Y1= (buf0[i2 ]*yalpha1+buf1[i2 ]*yalpha)>>19;\
463 int Y2= (buf0[i2+1]*yalpha1+buf1[i2+1]*yalpha)>>19;\
464 int U= (uvbuf0[i ]*uvalpha1+uvbuf1[i ]*uvalpha)>>19;\
465 int V= (uvbuf0[i+2048]*uvalpha1+uvbuf1[i+2048]*uvalpha)>>19;\
466 type *r, *b, *g;\
467 r = c->table_rV[V];\
468 g = c->table_gU[U] + c->table_gV[V];\
469 b = c->table_bU[U];\
470
471#define YSCALE_YUV_2_RGB1_C(type) \
472 for(i=0; i<(dstW>>1); i++){\
473 const int i2= 2*i;\
474 int Y1= buf0[i2 ]>>7;\
475 int Y2= buf0[i2+1]>>7;\
476 int U= (uvbuf1[i ])>>7;\
477 int V= (uvbuf1[i+2048])>>7;\
478 type *r, *b, *g;\
479 r = c->table_rV[V];\
480 g = c->table_gU[U] + c->table_gV[V];\
481 b = c->table_bU[U];\
482
483#define YSCALE_YUV_2_RGB1B_C(type) \
484 for(i=0; i<(dstW>>1); i++){\
485 const int i2= 2*i;\
486 int Y1= buf0[i2 ]>>7;\
487 int Y2= buf0[i2+1]>>7;\
488 int U= (uvbuf0[i ] + uvbuf1[i ])>>8;\
489 int V= (uvbuf0[i+2048] + uvbuf1[i+2048])>>8;\
490 type *r, *b, *g;\
491 r = c->table_rV[V];\
492 g = c->table_gU[U] + c->table_gV[V];\
493 b = c->table_bU[U];\
494
495#define YSCALE_YUV_2_ANYRGB_C(func)\
496 switch(c->dstFormat)\
497 {\
498 case IMGFMT_BGR32:\
499 case IMGFMT_RGB32:\
500 func(uint32_t)\
501 ((uint32_t*)dest)[i2+0]= r[Y1] + g[Y1] + b[Y1];\
502 ((uint32_t*)dest)[i2+1]= r[Y2] + g[Y2] + b[Y2];\
503 } \
504 break;\
505 case IMGFMT_RGB24:\
506 func(uint8_t)\
507 ((uint8_t*)dest)[0]= r[Y1];\
508 ((uint8_t*)dest)[1]= g[Y1];\
509 ((uint8_t*)dest)[2]= b[Y1];\
510 ((uint8_t*)dest)[3]= r[Y2];\
511 ((uint8_t*)dest)[4]= g[Y2];\
512 ((uint8_t*)dest)[5]= b[Y2];\
513 ((uint8_t*)dest)+=6;\
514 }\
515 break;\
516 case IMGFMT_BGR24:\
517 func(uint8_t)\
518 ((uint8_t*)dest)[0]= b[Y1];\
519 ((uint8_t*)dest)[1]= g[Y1];\
520 ((uint8_t*)dest)[2]= r[Y1];\
521 ((uint8_t*)dest)[3]= b[Y2];\
522 ((uint8_t*)dest)[4]= g[Y2];\
523 ((uint8_t*)dest)[5]= r[Y2];\
524 ((uint8_t*)dest)+=6;\
525 }\
526 break;\
527 case IMGFMT_RGB16:\
528 case IMGFMT_BGR16:\
529 {\
530 const int dr1= dither_2x2_8[y&1 ][0];\
531 const int dg1= dither_2x2_4[y&1 ][0];\
532 const int db1= dither_2x2_8[(y&1)^1][0];\
533 const int dr2= dither_2x2_8[y&1 ][1];\
534 const int dg2= dither_2x2_4[y&1 ][1];\
535 const int db2= dither_2x2_8[(y&1)^1][1];\
536 func(uint16_t)\
537 ((uint16_t*)dest)[i2+0]= r[Y1+dr1] + g[Y1+dg1] + b[Y1+db1];\
538 ((uint16_t*)dest)[i2+1]= r[Y2+dr2] + g[Y2+dg2] + b[Y2+db2];\
539 }\
540 }\
541 break;\
542 case IMGFMT_RGB15:\
543 case IMGFMT_BGR15:\
544 {\
545 const int dr1= dither_2x2_8[y&1 ][0];\
546 const int dg1= dither_2x2_8[y&1 ][1];\
547 const int db1= dither_2x2_8[(y&1)^1][0];\
548 const int dr2= dither_2x2_8[y&1 ][1];\
549 const int dg2= dither_2x2_8[y&1 ][0];\
550 const int db2= dither_2x2_8[(y&1)^1][1];\
551 func(uint16_t)\
552 ((uint16_t*)dest)[i2+0]= r[Y1+dr1] + g[Y1+dg1] + b[Y1+db1];\
553 ((uint16_t*)dest)[i2+1]= r[Y2+dr2] + g[Y2+dg2] + b[Y2+db2];\
554 }\
555 }\
556 break;\
557 case IMGFMT_RGB8:\
558 case IMGFMT_BGR8:\
559 {\
560 const uint8_t * const d64= dither_8x8_73[y&7];\
561 const uint8_t * const d32= dither_8x8_32[y&7];\
562 func(uint8_t)\
563 ((uint8_t*)dest)[i2+0]= r[Y1+d32[(i2+0)&7]] + g[Y1+d32[(i2+0)&7]] + b[Y1+d64[(i2+0)&7]];\
564 ((uint8_t*)dest)[i2+1]= r[Y2+d32[(i2+1)&7]] + g[Y2+d32[(i2+1)&7]] + b[Y2+d64[(i2+1)&7]];\
565 }\
566 }\
567 break;\
568 case IMGFMT_RGB4:\
569 case IMGFMT_BGR4:\
570 {\
571 const uint8_t * const d64= dither_8x8_73 [y&7];\
572 const uint8_t * const d128=dither_8x8_220[y&7];\
573 func(uint8_t)\
574 ((uint8_t*)dest)[i2+0]= r[Y1+d128[(i2+0)&7]] + g[Y1+d64[(i2+0)&7]] + b[Y1+d128[(i2+0)&7]];\
575 ((uint8_t*)dest)[i2+1]= r[Y2+d128[(i2+1)&7]] + g[Y2+d64[(i2+1)&7]] + b[Y2+d128[(i2+1)&7]];\
576 }\
577 }\
578 break;\
579 case IMGFMT_RGB1:\
580 case IMGFMT_BGR1:\
581 {\
582 const uint8_t * const d128=dither_8x8_220[y&7];\
583 uint8_t *g= c->table_gU[128] + c->table_gV[128];\
584 for(i=0; i<dstW-7; i+=8){\
585 int acc;\
586 acc = g[((buf0[i ]*yalpha1+buf1[i ]*yalpha)>>19) + d128[0]];\
587 acc+= acc + g[((buf0[i+1]*yalpha1+buf1[i+1]*yalpha)>>19) + d128[1]];\
588 acc+= acc + g[((buf0[i+2]*yalpha1+buf1[i+2]*yalpha)>>19) + d128[2]];\
589 acc+= acc + g[((buf0[i+3]*yalpha1+buf1[i+3]*yalpha)>>19) + d128[3]];\
590 acc+= acc + g[((buf0[i+4]*yalpha1+buf1[i+4]*yalpha)>>19) + d128[4]];\
591 acc+= acc + g[((buf0[i+5]*yalpha1+buf1[i+5]*yalpha)>>19) + d128[5]];\
592 acc+= acc + g[((buf0[i+6]*yalpha1+buf1[i+6]*yalpha)>>19) + d128[6]];\
593 acc+= acc + g[((buf0[i+7]*yalpha1+buf1[i+7]*yalpha)>>19) + d128[7]];\
594 ((uint8_t*)dest)[0]= acc;\
595 ((uint8_t*)dest)++;\
596 }\
597\
598/*\
599((uint8_t*)dest)-= dstW>>4;\
600{\
601 int acc=0;\
602 int left=0;\
603 static int top[1024];\
604 static int last_new[1024][1024];\
605 static int last_in3[1024][1024];\
606 static int drift[1024][1024];\
607 int topLeft=0;\
608 int shift=0;\
609 int count=0;\
610 const uint8_t * const d128=dither_8x8_220[y&7];\
611 int error_new=0;\
612 int error_in3=0;\
613 int f=0;\
614 \
615 for(i=dstW>>1; i<dstW; i++){\
616 int in= ((buf0[i ]*yalpha1+buf1[i ]*yalpha)>>19);\
617 int in2 = (76309 * (in - 16) + 32768) >> 16;\
618 int in3 = (in2 < 0) ? 0 : ((in2 > 255) ? 255 : in2);\
619 int old= (left*7 + topLeft + top[i]*5 + top[i+1]*3)/20 + in3\
620 + (last_new[y][i] - in3)*f/256;\
621 int new= old> 128 ? 255 : 0;\
622\
623 error_new+= ABS(last_new[y][i] - new);\
624 error_in3+= ABS(last_in3[y][i] - in3);\
625 f= error_new - error_in3*4;\
626 if(f<0) f=0;\
627 if(f>256) f=256;\
628\
629 topLeft= top[i];\
630 left= top[i]= old - new;\
631 last_new[y][i]= new;\
632 last_in3[y][i]= in3;\
633\
634 acc+= acc + (new&1);\
635 if((i&7)==6){\
636 ((uint8_t*)dest)[0]= acc;\
637 ((uint8_t*)dest)++;\
638 }\
639 }\
640}\
641*/\
642 }\
643 break;\
644 }\
645
646
647static inline void yuv2rgbXinC(SwsContext *c, int16_t *lumFilter, int16_t **lumSrc, int lumFilterSize,
e3d2500f 648 int16_t *chrFilter, int16_t **chrSrc, int chrFilterSize,
cf7d1c1a 649 uint8_t *dest, int dstW, int y)
e3d2500f 650{
cf7d1c1a
MN
651 int i;
652 switch(c->dstFormat)
e3d2500f 653 {
cf7d1c1a
MN
654 case IMGFMT_RGB32:
655 case IMGFMT_BGR32:
656 YSCALE_YUV_2_RGBX_C(uint32_t)
657 ((uint32_t*)dest)[i2+0]= r[Y1] + g[Y1] + b[Y1];
658 ((uint32_t*)dest)[i2+1]= r[Y2] + g[Y2] + b[Y2];
e3d2500f 659 }
cf7d1c1a
MN
660 break;
661 case IMGFMT_RGB24:
662 YSCALE_YUV_2_RGBX_C(uint8_t)
663 ((uint8_t*)dest)[0]= r[Y1];
664 ((uint8_t*)dest)[1]= g[Y1];
665 ((uint8_t*)dest)[2]= b[Y1];
666 ((uint8_t*)dest)[3]= r[Y2];
667 ((uint8_t*)dest)[4]= g[Y2];
668 ((uint8_t*)dest)[5]= b[Y2];
669 ((uint8_t*)dest)+=6;
670 }
671 break;
672 case IMGFMT_BGR24:
673 YSCALE_YUV_2_RGBX_C(uint8_t)
674 ((uint8_t*)dest)[0]= b[Y1];
675 ((uint8_t*)dest)[1]= g[Y1];
676 ((uint8_t*)dest)[2]= r[Y1];
677 ((uint8_t*)dest)[3]= b[Y2];
678 ((uint8_t*)dest)[4]= g[Y2];
679 ((uint8_t*)dest)[5]= r[Y2];
680 ((uint8_t*)dest)+=6;
681 }
682 break;
683 case IMGFMT_RGB16:
684 case IMGFMT_BGR16:
685 {
686 const int dr1= dither_2x2_8[y&1 ][0];
687 const int dg1= dither_2x2_4[y&1 ][0];
688 const int db1= dither_2x2_8[(y&1)^1][0];
689 const int dr2= dither_2x2_8[y&1 ][1];
690 const int dg2= dither_2x2_4[y&1 ][1];
691 const int db2= dither_2x2_8[(y&1)^1][1];
692 YSCALE_YUV_2_RGBX_C(uint16_t)
693 ((uint16_t*)dest)[i2+0]= r[Y1+dr1] + g[Y1+dg1] + b[Y1+db1];
694 ((uint16_t*)dest)[i2+1]= r[Y2+dr2] + g[Y2+dg2] + b[Y2+db2];
e3d2500f 695 }
e3d2500f 696 }
cf7d1c1a
MN
697 break;
698 case IMGFMT_RGB15:
699 case IMGFMT_BGR15:
700 {
701 const int dr1= dither_2x2_8[y&1 ][0];
702 const int dg1= dither_2x2_8[y&1 ][1];
703 const int db1= dither_2x2_8[(y&1)^1][0];
704 const int dr2= dither_2x2_8[y&1 ][1];
705 const int dg2= dither_2x2_8[y&1 ][0];
706 const int db2= dither_2x2_8[(y&1)^1][1];
707 YSCALE_YUV_2_RGBX_C(uint16_t)
708 ((uint16_t*)dest)[i2+0]= r[Y1+dr1] + g[Y1+dg1] + b[Y1+db1];
709 ((uint16_t*)dest)[i2+1]= r[Y2+dr2] + g[Y2+dg2] + b[Y2+db2];
e3d2500f 710 }
cf7d1c1a
MN
711 }
712 break;
713 case IMGFMT_RGB8:
714 case IMGFMT_BGR8:
715 {
716 const uint8_t * const d64= dither_8x8_73[y&7];
717 const uint8_t * const d32= dither_8x8_32[y&7];
718 YSCALE_YUV_2_RGBX_C(uint8_t)
719 ((uint8_t*)dest)[i2+0]= r[Y1+d32[(i2+0)&7]] + g[Y1+d32[(i2+0)&7]] + b[Y1+d64[(i2+0)&7]];
720 ((uint8_t*)dest)[i2+1]= r[Y2+d32[(i2+1)&7]] + g[Y2+d32[(i2+1)&7]] + b[Y2+d64[(i2+1)&7]];
e3d2500f 721 }
e3d2500f 722 }
cf7d1c1a
MN
723 break;
724 case IMGFMT_RGB4:
725 case IMGFMT_BGR4:
726 {
727 const uint8_t * const d64= dither_8x8_73 [y&7];
728 const uint8_t * const d128=dither_8x8_220[y&7];
729 YSCALE_YUV_2_RGBX_C(uint8_t)
730 ((uint8_t*)dest)[i2+0]= r[Y1+d128[(i2+0)&7]] + g[Y1+d64[(i2+0)&7]] + b[Y1+d128[(i2+0)&7]];
731 ((uint8_t*)dest)[i2+1]= r[Y2+d128[(i2+1)&7]] + g[Y2+d64[(i2+1)&7]] + b[Y2+d128[(i2+1)&7]];
e3d2500f 732 }
cf7d1c1a
MN
733 }
734 break;
735 case IMGFMT_RGB1:
736 case IMGFMT_BGR1:
737 {
738 const uint8_t * const d128=dither_8x8_220[y&7];
739 uint8_t *g= c->table_gU[128] + c->table_gV[128];
740 int acc=0;
741 for(i=0; i<dstW-1; i+=2){
742 int j;
743 int Y1=0;
744 int Y2=0;
745
746 for(j=0; j<lumFilterSize; j++)
747 {
748 Y1 += lumSrc[j][i] * lumFilter[j];
749 Y2 += lumSrc[j][i+1] * lumFilter[j];
750 }
751 Y1>>=19;
752 Y2>>=19;
753 if((Y1|Y2)&256)
754 {
755 if(Y1>255) Y1=255;
756 else if(Y1<0)Y1=0;
757 if(Y2>255) Y2=255;
758 else if(Y2<0)Y2=0;
759 }
760 acc+= acc + g[Y1+d128[(i+0)&7]];
761 acc+= acc + g[Y2+d128[(i+1)&7]];
762 if((i&7)==6){
763 ((uint8_t*)dest)[0]= acc;
764 ((uint8_t*)dest)++;
765 }
e3d2500f 766 }
e3d2500f 767 }
cf7d1c1a 768 break;
e3d2500f
MN
769 }
770}
771
772
7630f2e0
MN
773//Note: we have C, X86, MMX, MMX2, 3DNOW version therse no 3DNOW+MMX2 one
774//Plain C versions
726a959a
MN
775#if !defined (HAVE_MMX) || defined (RUNTIME_CPUDETECT)
776#define COMPILE_C
777#endif
778
779#ifdef CAN_COMPILE_X86_ASM
780
781#if (defined (HAVE_MMX) && !defined (HAVE_3DNOW) && !defined (HAVE_MMX2)) || defined (RUNTIME_CPUDETECT)
782#define COMPILE_MMX
783#endif
784
785#if defined (HAVE_MMX2) || defined (RUNTIME_CPUDETECT)
786#define COMPILE_MMX2
787#endif
788
789#if (defined (HAVE_3DNOW) && !defined (HAVE_MMX2)) || defined (RUNTIME_CPUDETECT)
790#define COMPILE_3DNOW
791#endif
792#endif //CAN_COMPILE_X86_ASM
793
794#undef HAVE_MMX
795#undef HAVE_MMX2
796#undef HAVE_3DNOW
726a959a
MN
797
798#ifdef COMPILE_C
7630f2e0
MN
799#undef HAVE_MMX
800#undef HAVE_MMX2
801#undef HAVE_3DNOW
7630f2e0
MN
802#define RENAME(a) a ## _C
803#include "swscale_template.c"
726a959a 804#endif
397c035e 805
7630f2e0 806#ifdef CAN_COMPILE_X86_ASM
397c035e 807
7630f2e0
MN
808//X86 versions
809/*
810#undef RENAME
811#undef HAVE_MMX
812#undef HAVE_MMX2
813#undef HAVE_3DNOW
814#define ARCH_X86
815#define RENAME(a) a ## _X86
816#include "swscale_template.c"
1faf0867 817*/
7630f2e0 818//MMX versions
726a959a 819#ifdef COMPILE_MMX
7630f2e0
MN
820#undef RENAME
821#define HAVE_MMX
822#undef HAVE_MMX2
823#undef HAVE_3DNOW
7630f2e0
MN
824#define RENAME(a) a ## _MMX
825#include "swscale_template.c"
726a959a 826#endif
7630f2e0
MN
827
828//MMX2 versions
726a959a 829#ifdef COMPILE_MMX2
7630f2e0
MN
830#undef RENAME
831#define HAVE_MMX
832#define HAVE_MMX2
833#undef HAVE_3DNOW
7630f2e0
MN
834#define RENAME(a) a ## _MMX2
835#include "swscale_template.c"
726a959a 836#endif
7630f2e0
MN
837
838//3DNOW versions
726a959a 839#ifdef COMPILE_3DNOW
7630f2e0
MN
840#undef RENAME
841#define HAVE_MMX
842#undef HAVE_MMX2
843#define HAVE_3DNOW
7630f2e0
MN
844#define RENAME(a) a ## _3DNow
845#include "swscale_template.c"
726a959a 846#endif
7630f2e0
MN
847
848#endif //CAN_COMPILE_X86_ASM
849
850// minor note: the HAVE_xyz is messed up after that line so dont use it
d604bab9 851
d3f41512 852
6c7506de 853// old global scaler, dont use for new code
28bf81c9
MN
854// will use sws_flags from the command line
855void SwScale_YV12slice(unsigned char* src[], int srcStride[], int srcSliceY ,
856 int srcSliceH, uint8_t* dst[], int dstStride, int dstbpp,
d1fac6cf 857 int srcW, int srcH, int dstW, int dstH){
31190492 858
28bf81c9
MN
859 static SwsContext *context=NULL;
860 int dstFormat;
28bf81c9
MN
861 int dstStride3[3]= {dstStride, dstStride>>1, dstStride>>1};
862
6c7506de 863 switch(dstbpp)
28bf81c9 864 {
6c7506de
MN
865 case 8 : dstFormat= IMGFMT_Y8; break;
866 case 12: dstFormat= IMGFMT_YV12; break;
867 case 15: dstFormat= IMGFMT_BGR15; break;
868 case 16: dstFormat= IMGFMT_BGR16; break;
869 case 24: dstFormat= IMGFMT_BGR24; break;
870 case 32: dstFormat= IMGFMT_BGR32; break;
871 default: return;
872 }
873
874 if(!context) context=getSwsContextFromCmdLine(srcW, srcH, IMGFMT_YV12, dstW, dstH, dstFormat);
875
b6654a54 876 context->swScale(context, src, srcStride, srcSliceY, srcSliceH, dst, dstStride3);
6c7506de
MN
877}
878
5859233b 879void swsGetFlagsAndFilterFromCmdLine(int *flags, SwsFilter **srcFilterParam, SwsFilter **dstFilterParam)
6c7506de 880{
6c7506de 881 static int firstTime=1;
5859233b 882 *flags=0;
6c7506de 883
5521b193 884#ifdef ARCH_X86
6c7506de
MN
885 if(gCpuCaps.hasMMX)
886 asm volatile("emms\n\t"::: "memory"); //FIXME this shouldnt be required but it IS (even for non mmx versions)
5521b193 887#endif
6c7506de
MN
888 if(firstTime)
889 {
28bf81c9 890 firstTime=0;
5859233b 891 *flags= SWS_PRINT_INFO;
6c7506de 892 }
5859233b 893 else if(verbose>1) *flags= SWS_PRINT_INFO;
6c7506de
MN
894
895 if(src_filter.lumH) freeVec(src_filter.lumH);
896 if(src_filter.lumV) freeVec(src_filter.lumV);
897 if(src_filter.chrH) freeVec(src_filter.chrH);
898 if(src_filter.chrV) freeVec(src_filter.chrV);
899
900 if(sws_lum_gblur!=0.0){
901 src_filter.lumH= getGaussianVec(sws_lum_gblur, 3.0);
902 src_filter.lumV= getGaussianVec(sws_lum_gblur, 3.0);
903 }else{
904 src_filter.lumH= getIdentityVec();
905 src_filter.lumV= getIdentityVec();
906 }
c7f822d9 907
6c7506de
MN
908 if(sws_chr_gblur!=0.0){
909 src_filter.chrH= getGaussianVec(sws_chr_gblur, 3.0);
910 src_filter.chrV= getGaussianVec(sws_chr_gblur, 3.0);
911 }else{
912 src_filter.chrH= getIdentityVec();
913 src_filter.chrV= getIdentityVec();
914 }
5521b193 915
6c7506de
MN
916 if(sws_chr_sharpen!=0.0){
917 SwsVector *g= getConstVec(-1.0, 3);
918 SwsVector *id= getConstVec(10.0/sws_chr_sharpen, 1);
919 g->coeff[1]=2.0;
920 addVec(id, g);
921 convVec(src_filter.chrH, id);
922 convVec(src_filter.chrV, id);
923 freeVec(g);
924 freeVec(id);
925 }
5521b193 926
6c7506de
MN
927 if(sws_lum_sharpen!=0.0){
928 SwsVector *g= getConstVec(-1.0, 3);
929 SwsVector *id= getConstVec(10.0/sws_lum_sharpen, 1);
930 g->coeff[1]=2.0;
931 addVec(id, g);
932 convVec(src_filter.lumH, id);
933 convVec(src_filter.lumV, id);
934 freeVec(g);
935 freeVec(id);
936 }
c7f822d9 937
6c7506de
MN
938 if(sws_chr_hshift)
939 shiftVec(src_filter.chrH, sws_chr_hshift);
c7f822d9 940
6c7506de
MN
941 if(sws_chr_vshift)
942 shiftVec(src_filter.chrV, sws_chr_vshift);
5521b193 943
6c7506de
MN
944 normalizeVec(src_filter.chrH, 1.0);
945 normalizeVec(src_filter.chrV, 1.0);
946 normalizeVec(src_filter.lumH, 1.0);
947 normalizeVec(src_filter.lumV, 1.0);
28bf81c9 948
6c7506de
MN
949 if(verbose > 1) printVec(src_filter.chrH);
950 if(verbose > 1) printVec(src_filter.lumH);
28bf81c9
MN
951
952 switch(sws_flags)
953 {
5859233b
MN
954 case 0: *flags|= SWS_FAST_BILINEAR; break;
955 case 1: *flags|= SWS_BILINEAR; break;
956 case 2: *flags|= SWS_BICUBIC; break;
957 case 3: *flags|= SWS_X; break;
958 case 4: *flags|= SWS_POINT; break;
959 case 5: *flags|= SWS_AREA; break;
81a571a8 960 case 6: *flags|= SWS_BICUBLIN; break;
5859233b 961 default:*flags|= SWS_BILINEAR; break;
28bf81c9 962 }
5859233b
MN
963
964 *srcFilterParam= &src_filter;
965 *dstFilterParam= NULL;
966}
967
968// will use sws_flags & src_filter (from cmd line)
969SwsContext *getSwsContextFromCmdLine(int srcW, int srcH, int srcFormat, int dstW, int dstH, int dstFormat)
970{
971 int flags;
972 SwsFilter *dstFilterParam, *srcFilterParam;
973 swsGetFlagsAndFilterFromCmdLine(&flags, &srcFilterParam, &dstFilterParam);
28bf81c9 974
5859233b 975 return getSwsContext(srcW, srcH, srcFormat, dstW, dstH, dstFormat, flags, srcFilterParam, dstFilterParam);
28bf81c9
MN
976}
977
6c7506de 978
c7f822d9
MN
979static inline void initFilter(int16_t **outFilter, int16_t **filterPos, int *outFilterSize, int xInc,
980 int srcW, int dstW, int filterAlign, int one, int flags,
981 SwsVector *srcFilter, SwsVector *dstFilter)
28bf81c9
MN
982{
983 int i;
c7f822d9
MN
984 int filterSize;
985 int filter2Size;
986 int minFilterSize;
987 double *filter=NULL;
988 double *filter2=NULL;
28bf81c9
MN
989#ifdef ARCH_X86
990 if(gCpuCaps.hasMMX)
991 asm volatile("emms\n\t"::: "memory"); //FIXME this shouldnt be required but it IS (even for non mmx versions)
726a959a 992#endif
31190492 993
adeaecb9 994 // Note the +1 is for the MMXscaler which reads over the end
6c7506de 995 *filterPos = (int16_t*)memalign(8, (dstW+1)*sizeof(int16_t));
6c7506de 996
28bf81c9
MN
997 if(ABS(xInc - 0x10000) <10) // unscaled
998 {
999 int i;
c7f822d9
MN
1000 filterSize= 1;
1001 filter= (double*)memalign(8, dstW*sizeof(double)*filterSize);
1002 for(i=0; i<dstW*filterSize; i++) filter[i]=0;
28bf81c9
MN
1003
1004 for(i=0; i<dstW; i++)
1005 {
c7f822d9
MN
1006 filter[i*filterSize]=1;
1007 (*filterPos)[i]=i;
28bf81c9
MN
1008 }
1009
1010 }
ff7ba856
MN
1011 else if(flags&SWS_POINT) // lame looking point sampling mode
1012 {
1013 int i;
1014 int xDstInSrc;
1015 filterSize= 1;
1016 filter= (double*)memalign(8, dstW*sizeof(double)*filterSize);
1017
1018 xDstInSrc= xInc/2 - 0x8000;
1019 for(i=0; i<dstW; i++)
1020 {
8a01d20c 1021 int xx= (xDstInSrc - ((filterSize-1)<<15) + (1<<15))>>16;
ff7ba856
MN
1022
1023 (*filterPos)[i]= xx;
1024 filter[i]= 1.0;
1025 xDstInSrc+= xInc;
1026 }
1027 }
28bf81c9
MN
1028 else if(xInc <= (1<<16) || (flags&SWS_FAST_BILINEAR)) // upscale
1029 {
1030 int i;
1031 int xDstInSrc;
c7f822d9
MN
1032 if (flags&SWS_BICUBIC) filterSize= 4;
1033 else if(flags&SWS_X ) filterSize= 4;
d8863d37 1034 else filterSize= 2; // SWS_BILINEAR / SWS_AREA
c7f822d9 1035 filter= (double*)memalign(8, dstW*sizeof(double)*filterSize);
28bf81c9
MN
1036
1037 xDstInSrc= xInc/2 - 0x8000;
1038 for(i=0; i<dstW; i++)
1039 {
8a01d20c 1040 int xx= (xDstInSrc - ((filterSize-1)<<15) + (1<<15))>>16;
28bf81c9
MN
1041 int j;
1042
c7f822d9 1043 (*filterPos)[i]= xx;
28bf81c9
MN
1044 if((flags & SWS_BICUBIC) || (flags & SWS_X))
1045 {
1046 double d= ABS(((xx+1)<<16) - xDstInSrc)/(double)(1<<16);
1047 double y1,y2,y3,y4;
1048 double A= -0.6;
1049 if(flags & SWS_BICUBIC){
1050 // Equation is from VirtualDub
1051 y1 = ( + A*d - 2.0*A*d*d + A*d*d*d);
1052 y2 = (+ 1.0 - (A+3.0)*d*d + (A+2.0)*d*d*d);
1053 y3 = ( - A*d + (2.0*A+3.0)*d*d - (A+2.0)*d*d*d);
1054 y4 = ( + A*d*d - A*d*d*d);
1055 }else{
1056 // cubic interpolation (derived it myself)
1057 y1 = ( -2.0*d + 3.0*d*d - 1.0*d*d*d)/6.0;
1058 y2 = (6.0 -3.0*d - 6.0*d*d + 3.0*d*d*d)/6.0;
1059 y3 = ( +6.0*d + 3.0*d*d - 3.0*d*d*d)/6.0;
1060 y4 = ( -1.0*d + 1.0*d*d*d)/6.0;
1061 }
1062
c7f822d9
MN
1063 filter[i*filterSize + 0]= y1;
1064 filter[i*filterSize + 1]= y2;
1065 filter[i*filterSize + 2]= y3;
1066 filter[i*filterSize + 3]= y4;
28bf81c9
MN
1067 }
1068 else
1069 {
d8863d37 1070 //Bilinear upscale / linear interpolate / Area averaging
c7f822d9 1071 for(j=0; j<filterSize; j++)
28bf81c9
MN
1072 {
1073 double d= ABS((xx<<16) - xDstInSrc)/(double)(1<<16);
1074 double coeff= 1.0 - d;
1075 if(coeff<0) coeff=0;
c7f822d9 1076 filter[i*filterSize + j]= coeff;
28bf81c9
MN
1077 xx++;
1078 }
1079 }
1080 xDstInSrc+= xInc;
1081 }
1082 }
1083 else // downscale
1084 {
1085 int xDstInSrc;
81b7c056
MN
1086 ASSERT(dstW <= srcW)
1087
d8863d37
MN
1088 if(flags&SWS_BICUBIC) filterSize= (int)ceil(1 + 4.0*srcW / (double)dstW);
1089 else if(flags&SWS_X) filterSize= (int)ceil(1 + 4.0*srcW / (double)dstW);
1090 else if(flags&SWS_AREA) filterSize= (int)ceil(1 + 1.0*srcW / (double)dstW);
1091 else /* BILINEAR */ filterSize= (int)ceil(1 + 2.0*srcW / (double)dstW);
c7f822d9 1092 filter= (double*)memalign(8, dstW*sizeof(double)*filterSize);
28bf81c9
MN
1093
1094 xDstInSrc= xInc/2 - 0x8000;
1095 for(i=0; i<dstW; i++)
1096 {
c7f822d9 1097 int xx= (int)((double)xDstInSrc/(double)(1<<16) - (filterSize-1)*0.5 + 0.5);
28bf81c9 1098 int j;
c7f822d9
MN
1099 (*filterPos)[i]= xx;
1100 for(j=0; j<filterSize; j++)
28bf81c9
MN
1101 {
1102 double d= ABS((xx<<16) - xDstInSrc)/(double)xInc;
1103 double coeff;
1104 if((flags & SWS_BICUBIC) || (flags & SWS_X))
1105 {
1106 double A= -0.75;
1107// d*=2;
1108 // Equation is from VirtualDub
1109 if(d<1.0)
1110 coeff = (1.0 - (A+3.0)*d*d + (A+2.0)*d*d*d);
1111 else if(d<2.0)
1112 coeff = (-4.0*A + 8.0*A*d - 5.0*A*d*d + A*d*d*d);
1113 else
1114 coeff=0.0;
1115 }
d8863d37 1116 else if(flags & SWS_AREA)
28bf81c9 1117 {
d8863d37
MN
1118 double srcPixelSize= (1<<16)/(double)xInc;
1119 if(d + srcPixelSize/2 < 0.5) coeff= 1.0;
1120 else if(d - srcPixelSize/2 < 0.5) coeff= (0.5-d)/srcPixelSize + 0.5;
1121 else coeff=0.0;
1122 }
28bf81c9
MN
1123 else
1124 {
1125 coeff= 1.0 - d;
1126 if(coeff<0) coeff=0;
1127 }
c7f822d9 1128 filter[i*filterSize + j]= coeff;
28bf81c9
MN
1129 xx++;
1130 }
1131 xDstInSrc+= xInc;
1132 }
1133 }
1134
c7f822d9
MN
1135 /* apply src & dst Filter to filter -> filter2
1136 free(filter);
1137 */
81b7c056 1138 ASSERT(filterSize>0)
c7f822d9
MN
1139 filter2Size= filterSize;
1140 if(srcFilter) filter2Size+= srcFilter->length - 1;
1141 if(dstFilter) filter2Size+= dstFilter->length - 1;
81b7c056 1142 ASSERT(filter2Size>0)
c7f822d9
MN
1143 filter2= (double*)memalign(8, filter2Size*dstW*sizeof(double));
1144
1145 for(i=0; i<dstW; i++)
1146 {
1147 int j;
1148 SwsVector scaleFilter;
1149 SwsVector *outVec;
1150
1151 scaleFilter.coeff= filter + i*filterSize;
1152 scaleFilter.length= filterSize;
1153
5cebb24b 1154 if(srcFilter) outVec= getConvVec(srcFilter, &scaleFilter);
c7f822d9
MN
1155 else outVec= &scaleFilter;
1156
1157 ASSERT(outVec->length == filter2Size)
1158 //FIXME dstFilter
1159
1160 for(j=0; j<outVec->length; j++)
1161 {
1162 filter2[i*filter2Size + j]= outVec->coeff[j];
1163 }
1164
1165 (*filterPos)[i]+= (filterSize-1)/2 - (filter2Size-1)/2;
1166
1167 if(outVec != &scaleFilter) freeVec(outVec);
1168 }
1169 free(filter); filter=NULL;
1170
1171 /* try to reduce the filter-size (step1 find size and shift left) */
1172 // Assume its near normalized (*0.5 or *2.0 is ok but * 0.001 is not)
1173 minFilterSize= 0;
1174 for(i=dstW-1; i>=0; i--)
1175 {
1176 int min= filter2Size;
1177 int j;
1178 double cutOff=0.0;
1179
1180 /* get rid off near zero elements on the left by shifting left */
1181 for(j=0; j<filter2Size; j++)
1182 {
1183 int k;
1184 cutOff += ABS(filter2[i*filter2Size]);
1185
1186 if(cutOff > SWS_MAX_REDUCE_CUTOFF) break;
1187
1188 /* preserve Monotonicity because the core cant handle the filter otherwise */
1189 if(i<dstW-1 && (*filterPos)[i] >= (*filterPos)[i+1]) break;
1190
1191 // Move filter coeffs left
1192 for(k=1; k<filter2Size; k++)
1193 filter2[i*filter2Size + k - 1]= filter2[i*filter2Size + k];
1194 filter2[i*filter2Size + k - 1]= 0.0;
1195 (*filterPos)[i]++;
1196 }
1197
1198 cutOff=0.0;
1199 /* count near zeros on the right */
1200 for(j=filter2Size-1; j>0; j--)
1201 {
1202 cutOff += ABS(filter2[i*filter2Size + j]);
1203
1204 if(cutOff > SWS_MAX_REDUCE_CUTOFF) break;
1205 min--;
1206 }
1207
1208 if(min>minFilterSize) minFilterSize= min;
1209 }
1210
81b7c056 1211 ASSERT(minFilterSize > 0)
6c7506de 1212 filterSize= (minFilterSize +(filterAlign-1)) & (~(filterAlign-1));
81b7c056 1213 ASSERT(filterSize > 0)
6c7506de
MN
1214 filter= (double*)memalign(8, filterSize*dstW*sizeof(double));
1215 *outFilterSize= filterSize;
1216
4a53a912 1217 if(flags&SWS_PRINT_INFO)
0d9f3d85 1218 MSG_INFO("SwScaler: reducing / aligning filtersize %d -> %d\n", filter2Size, filterSize);
c7f822d9
MN
1219 /* try to reduce the filter-size (step2 reduce it) */
1220 for(i=0; i<dstW; i++)
1221 {
1222 int j;
1223
6c7506de
MN
1224 for(j=0; j<filterSize; j++)
1225 {
1226 if(j>=filter2Size) filter[i*filterSize + j]= 0.0;
1227 else filter[i*filterSize + j]= filter2[i*filter2Size + j];
1228 }
c7f822d9 1229 }
6c7506de
MN
1230 free(filter2); filter2=NULL;
1231
c7f822d9
MN
1232
1233 //FIXME try to align filterpos if possible
1234
28bf81c9
MN
1235 //fix borders
1236 for(i=0; i<dstW; i++)
1237 {
1238 int j;
c7f822d9 1239 if((*filterPos)[i] < 0)
28bf81c9
MN
1240 {
1241 // Move filter coeffs left to compensate for filterPos
6c7506de 1242 for(j=1; j<filterSize; j++)
28bf81c9 1243 {
c7f822d9 1244 int left= MAX(j + (*filterPos)[i], 0);
6c7506de
MN
1245 filter[i*filterSize + left] += filter[i*filterSize + j];
1246 filter[i*filterSize + j]=0;
28bf81c9 1247 }
c7f822d9 1248 (*filterPos)[i]= 0;
28bf81c9
MN
1249 }
1250
6c7506de 1251 if((*filterPos)[i] + filterSize > srcW)
28bf81c9 1252 {
6c7506de 1253 int shift= (*filterPos)[i] + filterSize - srcW;
28bf81c9 1254 // Move filter coeffs right to compensate for filterPos
6c7506de 1255 for(j=filterSize-2; j>=0; j--)
28bf81c9 1256 {
6c7506de
MN
1257 int right= MIN(j + shift, filterSize-1);
1258 filter[i*filterSize +right] += filter[i*filterSize +j];
1259 filter[i*filterSize +j]=0;
28bf81c9 1260 }
6c7506de 1261 (*filterPos)[i]= srcW - filterSize;
28bf81c9
MN
1262 }
1263 }
1264
6c7506de
MN
1265 // Note the +1 is for the MMXscaler which reads over the end
1266 *outFilter= (int16_t*)memalign(8, *outFilterSize*(dstW+1)*sizeof(int16_t));
1267 memset(*outFilter, 0, *outFilterSize*(dstW+1)*sizeof(int16_t));
c7f822d9
MN
1268
1269 /* Normalize & Store in outFilter */
28bf81c9
MN
1270 for(i=0; i<dstW; i++)
1271 {
1272 int j;
1273 double sum=0;
1274 double scale= one;
6c7506de 1275 for(j=0; j<filterSize; j++)
28bf81c9 1276 {
6c7506de 1277 sum+= filter[i*filterSize + j];
28bf81c9
MN
1278 }
1279 scale/= sum;
6c7506de 1280 for(j=0; j<filterSize; j++)
28bf81c9 1281 {
6c7506de 1282 (*outFilter)[i*(*outFilterSize) + j]= (int)(filter[i*filterSize + j]*scale);
28bf81c9
MN
1283 }
1284 }
adeaecb9
MN
1285
1286 (*filterPos)[dstW]= (*filterPos)[dstW-1]; // the MMX scaler will read over the end
1287 for(i=0; i<*outFilterSize; i++)
1288 {
1289 int j= dstW*(*outFilterSize);
1290 (*outFilter)[j + i]= (*outFilter)[j + i - (*outFilterSize)];
1291 }
c7f822d9 1292
6c7506de 1293 free(filter);
7630f2e0 1294}
31190492 1295
28bf81c9 1296#ifdef ARCH_X86
b7dc6f66 1297static void initMMX2HScaler(int dstW, int xInc, uint8_t *funnyCode, int16_t *filter, int32_t *filterPos, int numSplits)
28bf81c9 1298{
b7dc6f66
MN
1299 uint8_t *fragmentA;
1300 int imm8OfPShufW1A;
1301 int imm8OfPShufW2A;
1302 int fragmentLengthA;
1303 uint8_t *fragmentB;
1304 int imm8OfPShufW1B;
1305 int imm8OfPShufW2B;
1306 int fragmentLengthB;
1307 int fragmentPos;
28bf81c9
MN
1308
1309 int xpos, i;
1310
1311 // create an optimized horizontal scaling routine
1312
1313 //code fragment
1314
1315 asm volatile(
1316 "jmp 9f \n\t"
1317 // Begin
1318 "0: \n\t"
b7dc6f66
MN
1319 "movq (%%edx, %%eax), %%mm3 \n\t"
1320 "movd (%%ecx, %%esi), %%mm0 \n\t"
1321 "movd 1(%%ecx, %%esi), %%mm1 \n\t"
1322 "punpcklbw %%mm7, %%mm1 \n\t"
1323 "punpcklbw %%mm7, %%mm0 \n\t"
28bf81c9
MN
1324 "pshufw $0xFF, %%mm1, %%mm1 \n\t"
1325 "1: \n\t"
28bf81c9
MN
1326 "pshufw $0xFF, %%mm0, %%mm0 \n\t"
1327 "2: \n\t"
28bf81c9 1328 "psubw %%mm1, %%mm0 \n\t"
b7dc6f66 1329 "movl 8(%%ebx, %%eax), %%esi \n\t"
28bf81c9 1330 "pmullw %%mm3, %%mm0 \n\t"
28bf81c9
MN
1331 "psllw $7, %%mm1 \n\t"
1332 "paddw %%mm1, %%mm0 \n\t"
1333
1334 "movq %%mm0, (%%edi, %%eax) \n\t"
1335
1336 "addl $8, %%eax \n\t"
1337 // End
1338 "9: \n\t"
1339// "int $3\n\t"
1340 "leal 0b, %0 \n\t"
1341 "leal 1b, %1 \n\t"
1342 "leal 2b, %2 \n\t"
1343 "decl %1 \n\t"
1344 "decl %2 \n\t"
1345 "subl %0, %1 \n\t"
1346 "subl %0, %2 \n\t"
1347 "leal 9b, %3 \n\t"
1348 "subl %0, %3 \n\t"
b7dc6f66
MN
1349
1350
1351 :"=r" (fragmentA), "=r" (imm8OfPShufW1A), "=r" (imm8OfPShufW2A),
1352 "=r" (fragmentLengthA)
28bf81c9
MN
1353 );
1354
b7dc6f66
MN
1355 asm volatile(
1356 "jmp 9f \n\t"
1357 // Begin
1358 "0: \n\t"
1359 "movq (%%edx, %%eax), %%mm3 \n\t"
1360 "movd (%%ecx, %%esi), %%mm0 \n\t"
1361 "punpcklbw %%mm7, %%mm0 \n\t"
1362 "pshufw $0xFF, %%mm0, %%mm1 \n\t"
1363 "1: \n\t"
1364 "pshufw $0xFF, %%mm0, %%mm0 \n\t"
1365 "2: \n\t"
1366 "psubw %%mm1, %%mm0 \n\t"
1367 "movl 8(%%ebx, %%eax), %%esi \n\t"
1368 "pmullw %%mm3, %%mm0 \n\t"
1369 "psllw $7, %%mm1 \n\t"
1370 "paddw %%mm1, %%mm0 \n\t"
1371
1372 "movq %%mm0, (%%edi, %%eax) \n\t"
28bf81c9 1373
b7dc6f66
MN
1374 "addl $8, %%eax \n\t"
1375 // End
1376 "9: \n\t"
1377// "int $3\n\t"
1378 "leal 0b, %0 \n\t"
1379 "leal 1b, %1 \n\t"
1380 "leal 2b, %2 \n\t"
1381 "decl %1 \n\t"
1382 "decl %2 \n\t"
1383 "subl %0, %1 \n\t"
1384 "subl %0, %2 \n\t"
1385 "leal 9b, %3 \n\t"
1386 "subl %0, %3 \n\t"
1387
1388
1389 :"=r" (fragmentB), "=r" (imm8OfPShufW1B), "=r" (imm8OfPShufW2B),
1390 "=r" (fragmentLengthB)
1391 );
1392
1393 xpos= 0; //lumXInc/2 - 0x8000; // difference between pixel centers
1394 fragmentPos=0;
1395
1396 for(i=0; i<dstW/numSplits; i++)
28bf81c9
MN
1397 {
1398 int xx=xpos>>16;
1399
1400 if((i&3) == 0)
1401 {
1402 int a=0;
1403 int b=((xpos+xInc)>>16) - xx;
1404 int c=((xpos+xInc*2)>>16) - xx;
1405 int d=((xpos+xInc*3)>>16) - xx;
1406
b7dc6f66
MN
1407 filter[i ] = (( xpos & 0xFFFF) ^ 0xFFFF)>>9;
1408 filter[i+1] = (((xpos+xInc ) & 0xFFFF) ^ 0xFFFF)>>9;
1409 filter[i+2] = (((xpos+xInc*2) & 0xFFFF) ^ 0xFFFF)>>9;
1410 filter[i+3] = (((xpos+xInc*3) & 0xFFFF) ^ 0xFFFF)>>9;
1411 filterPos[i/2]= xx;
1412
1413 if(d+1<4)
1414 {
1415 int maxShift= 3-(d+1);
1416 int shift=0;
1417
1418 memcpy(funnyCode + fragmentPos, fragmentB, fragmentLengthB);
1419
1420 funnyCode[fragmentPos + imm8OfPShufW1B]=
1421 (a+1) | ((b+1)<<2) | ((c+1)<<4) | ((d+1)<<6);
1422 funnyCode[fragmentPos + imm8OfPShufW2B]=
1423 a | (b<<2) | (c<<4) | (d<<6);
1424
1425 if(i+3>=dstW) shift=maxShift; //avoid overread
1426 else if((filterPos[i/2]&3) <= maxShift) shift=filterPos[i/2]&3; //Align
1427
1428 if(shift && i>=shift)
1429 {
1430 funnyCode[fragmentPos + imm8OfPShufW1B]+= 0x55*shift;
1431 funnyCode[fragmentPos + imm8OfPShufW2B]+= 0x55*shift;
1432 filterPos[i/2]-=shift;
1433 }
1434
1435 fragmentPos+= fragmentLengthB;
1436 }
1437 else
1438 {
1439 int maxShift= 3-d;
1440 int shift=0;
1441
1442 memcpy(funnyCode + fragmentPos, fragmentA, fragmentLengthA);
28bf81c9 1443
b7dc6f66
MN
1444 funnyCode[fragmentPos + imm8OfPShufW1A]=
1445 funnyCode[fragmentPos + imm8OfPShufW2A]=
1446 a | (b<<2) | (c<<4) | (d<<6);
28bf81c9 1447
b7dc6f66
MN
1448 if(i+4>=dstW) shift=maxShift; //avoid overread
1449 else if((filterPos[i/2]&3) <= maxShift) shift=filterPos[i/2]&3; //partial align
28bf81c9 1450
b7dc6f66
MN
1451 if(shift && i>=shift)
1452 {
1453 funnyCode[fragmentPos + imm8OfPShufW1A]+= 0x55*shift;
1454 funnyCode[fragmentPos + imm8OfPShufW2A]+= 0x55*shift;
1455 filterPos[i/2]-=shift;
1456 }
1457
1458 fragmentPos+= fragmentLengthA;
1459 }
1460
1461 funnyCode[fragmentPos]= RET;
28bf81c9
MN
1462 }
1463 xpos+=xInc;
1464 }
b7dc6f66 1465 filterPos[i/2]= xpos>>16; // needed to jump to the next part
28bf81c9
MN
1466}
1467#endif // ARCH_X86
1468
1469//FIXME remove
31190492 1470void SwScale_Init(){
28bf81c9
MN
1471}
1472
1473static void globalInit(){
31190492
A
1474 // generating tables:
1475 int i;
c1b0bfb4
MN
1476 for(i=0; i<768; i++){
1477 int c= MIN(MAX(i-256, 0), 255);
1478 clip_table[i]=c;
b18ea156 1479 }
c1b0bfb4 1480
28bf81c9
MN
1481cpuCaps= gCpuCaps;
1482
1483#ifdef RUNTIME_CPUDETECT
1484#ifdef CAN_COMPILE_X86_ASM
1485 // ordered per speed fasterst first
1486 if(gCpuCaps.hasMMX2)
1487 swScale= swScale_MMX2;
1488 else if(gCpuCaps.has3DNow)
7f56a527 1489 swScale= swScale_3DNow;
28bf81c9
MN
1490 else if(gCpuCaps.hasMMX)
1491 swScale= swScale_MMX;
1492 else
1493 swScale= swScale_C;
1494
1495#else
1496 swScale= swScale_C;
1497 cpuCaps.hasMMX2 = cpuCaps.hasMMX = cpuCaps.has3DNow = 0;
1498#endif
1499#else //RUNTIME_CPUDETECT
1500#ifdef HAVE_MMX2
1501 swScale= swScale_MMX2;
1502 cpuCaps.has3DNow = 0;
1503#elif defined (HAVE_3DNOW)
7f56a527 1504 swScale= swScale_3DNow;
28bf81c9
MN
1505 cpuCaps.hasMMX2 = 0;
1506#elif defined (HAVE_MMX)
1507 swScale= swScale_MMX;
1508 cpuCaps.hasMMX2 = cpuCaps.has3DNow = 0;
1509#else
1510 swScale= swScale_C;
1511 cpuCaps.hasMMX2 = cpuCaps.hasMMX = cpuCaps.has3DNow = 0;
1512#endif
1513#endif //!RUNTIME_CPUDETECT
31190492 1514}
7630f2e0 1515
0d9f3d85
A
1516static void PlanarToNV12Wrapper(SwsContext *c, uint8_t* src[], int srcStride[], int srcSliceY,
1517 int srcSliceH, uint8_t* dstParam[], int dstStride[]){
1518 uint8_t *dst=dstParam[0] + dstStride[0]*srcSliceY;
1519 /* Copy Y plane */
1520 if(dstStride[0]==srcStride[0])
1521 memcpy(dst, src[0], srcSliceH*dstStride[0]);
1522 else
1523 {
1524 int i;
1525 uint8_t *srcPtr= src[0];
1526 uint8_t *dstPtr= dst;
1527 for(i=0; i<srcSliceH; i++)
1528 {
1529 memcpy(dstPtr, srcPtr, srcStride[0]);
1530 srcPtr+= srcStride[0];
1531 dstPtr+= dstStride[0];
1532 }
1533 }
1534 dst = dstParam[1] + dstStride[1]*srcSliceY;
1535 if(c->srcFormat==IMGFMT_YV12)
1536 interleaveBytes( src[1],src[2],dst,c->srcW,srcSliceH,srcStride[1],srcStride[2],dstStride[0] );
1537 else /* I420 & IYUV */
1538 interleaveBytes( src[2],src[1],dst,c->srcW,srcSliceH,srcStride[2],srcStride[1],dstStride[0] );
1539}
1540
1541
37079906
MN
1542/* Warper functions for yuv2bgr */
1543static void planarYuvToBgr(SwsContext *c, uint8_t* src[], int srcStride[], int srcSliceY,
b6654a54
MN
1544 int srcSliceH, uint8_t* dstParam[], int dstStride[]){
1545 uint8_t *dst=dstParam[0] + dstStride[0]*srcSliceY;
37079906
MN
1546
1547 if(c->srcFormat==IMGFMT_YV12)
b6654a54 1548 yuv2rgb( dst,src[0],src[1],src[2],c->srcW,srcSliceH,dstStride[0],srcStride[0],srcStride[1] );
37079906 1549 else /* I420 & IYUV */
b6654a54
MN
1550 yuv2rgb( dst,src[0],src[2],src[1],c->srcW,srcSliceH,dstStride[0],srcStride[0],srcStride[1] );
1551}
1552
44c1035c 1553static void PlanarToYuy2Wrapper(SwsContext *c, uint8_t* src[], int srcStride[], int srcSliceY,
0d9f3d85
A
1554 int srcSliceH, uint8_t* dstParam[], int dstStride[]){
1555 uint8_t *dst=dstParam[0] + dstStride[0]*srcSliceY;
1556
1557 if(c->srcFormat==IMGFMT_YV12)
1558 yv12toyuy2( src[0],src[1],src[2],dst,c->srcW,srcSliceH,srcStride[0],srcStride[1],dstStride[0] );
1559 else /* I420 & IYUV */
1560 yv12toyuy2( src[0],src[2],src[1],dst,c->srcW,srcSliceH,srcStride[0],srcStride[1],dstStride[0] );
1561}
1562
e09d12f4
MN
1563/* {RGB,BGR}{15,16,24,32} -> {RGB,BGR}{15,16,24,32} */
1564static void rgb2rgbWrapper(SwsContext *c, uint8_t* src[], int srcStride[], int srcSliceY,
1565 int srcSliceH, uint8_t* dst[], int dstStride[]){
1566 const int srcFormat= c->srcFormat;
1567 const int dstFormat= c->dstFormat;
1568 const int srcBpp= ((srcFormat&0xFF) + 7)>>3;
1569 const int dstBpp= ((dstFormat&0xFF) + 7)>>3;
1570 const int srcId= (srcFormat&0xFF)>>2; // 1:0, 4:1, 8:2, 15:3, 16:4, 24:6, 32:8
1571 const int dstId= (dstFormat&0xFF)>>2;
1572 void (*conv)(const uint8_t *src, uint8_t *dst, unsigned src_size)=NULL;
1573
1574 /* BGR -> BGR */
1575 if(isBGR(srcFormat) && isBGR(dstFormat)){
1576 switch(srcId | (dstId<<4)){
1577 case 0x34: conv= rgb16to15; break;
1578 case 0x36: conv= rgb24to15; break;
1579 case 0x38: conv= rgb32to15; break;
1580 case 0x43: conv= rgb15to16; break;
1581 case 0x46: conv= rgb24to16; break;
1582 case 0x48: conv= rgb32to16; break;
1583 case 0x63: conv= rgb15to24; break;
1584 case 0x64: conv= rgb16to24; break;
1585 case 0x68: conv= rgb32to24; break;
1586 case 0x83: conv= rgb15to32; break;
1587 case 0x84: conv= rgb16to32; break;
1588 case 0x86: conv= rgb24to32; break;
1589 default: MSG_ERR("swScaler: internal error %s -> %s converter\n",
1590 vo_format_name(srcFormat), vo_format_name(dstFormat)); break;
b935781b 1591 }
e09d12f4
MN
1592 }else if(isBGR(srcFormat) && isRGB(dstFormat)){
1593 switch(srcId | (dstId<<4)){
1594 case 0x33: conv= rgb15tobgr15; break;
1595 case 0x34: conv= rgb16tobgr15; break;
1596 case 0x36: conv= rgb24tobgr15; break;
1597 case 0x38: conv= rgb32tobgr15; break;
1598 case 0x43: conv= rgb15tobgr16; break;
1599 case 0x44: conv= rgb16tobgr16; break;
1600 case 0x46: conv= rgb24tobgr16; break;
1601 case 0x48: conv= rgb32tobgr16; break;
1602 case 0x63: conv= rgb15tobgr24; break;
1603 case 0x64: conv= rgb16tobgr24; break;
1604 case 0x66: conv= rgb24tobgr24; break;
1605 case 0x68: conv= rgb32tobgr24; break;
1606 case 0x83: conv= rgb15tobgr32; break;
1607 case 0x84: conv= rgb16tobgr32; break;
1608 case 0x86: conv= rgb24tobgr32; break;
1609 case 0x88: conv= rgb32tobgr32; break;
1610 default: MSG_ERR("swScaler: internal error %s -> %s converter\n",
1611 vo_format_name(srcFormat), vo_format_name(dstFormat)); break;
0d9f3d85 1612 }
e09d12f4
MN
1613 }else if(isRGB(srcFormat) && isRGB(dstFormat)){
1614 switch(srcId | (dstId<<4)){
1615 case 0x34: conv= rgb16to15; break;
1616 case 0x36: conv= rgb24to15; break;
1617 case 0x38: conv= rgb32to15; break;
1618 case 0x43: conv= rgb15to16; break;
1619 case 0x46: conv= rgb24to16; break;
1620 case 0x48: conv= rgb32to16; break;
1621 case 0x63: conv= rgb15to24; break;
1622 case 0x64: conv= rgb16to24; break;
1623 case 0x68: conv= rgb32to24; break;
1624 case 0x83: conv= rgb15to32; break;
1625 case 0x84: conv= rgb16to32; break;
1626 case 0x86: conv= rgb24to32; break;
1627 default: MSG_ERR("swScaler: internal error %s -> %s converter\n",
1628 vo_format_name(srcFormat), vo_format_name(dstFormat)); break;
0d9f3d85 1629 }
e09d12f4
MN
1630 }else if(isRGB(srcFormat) && isBGR(dstFormat)){
1631 switch(srcId | (dstId<<4)){
1632 case 0x33: conv= rgb15tobgr15; break;
1633 case 0x34: conv= rgb16tobgr15; break;
1634 case 0x36: conv= rgb24tobgr15; break;
1635 case 0x38: conv= rgb32tobgr15; break;
1636 case 0x43: conv= rgb15tobgr16; break;
1637 case 0x44: conv= rgb16tobgr16; break;
1638 case 0x46: conv= rgb24tobgr16; break;
1639 case 0x48: conv= rgb32tobgr16; break;
1640 case 0x63: conv= rgb15tobgr24; break;
1641 case 0x64: conv= rgb16tobgr24; break;
1642 case 0x66: conv= rgb24tobgr24; break;
1643 case 0x68: conv= rgb32tobgr24; break;
1644 case 0x83: conv= rgb15tobgr32; break;
1645 case 0x84: conv= rgb16tobgr32; break;
1646 case 0x86: conv= rgb24tobgr32; break;
1647 case 0x88: conv= rgb32tobgr32; break;
1648 default: MSG_ERR("swScaler: internal error %s -> %s converter\n",
1649 vo_format_name(srcFormat), vo_format_name(dstFormat)); break;
0d9f3d85 1650 }
e09d12f4
MN
1651 }
1652 if(dstStride[0]*srcBpp == srcStride[0]*dstBpp)
1653 conv(src[0], dst[0] + dstStride[0]*srcSliceY, srcSliceH*srcStride[0]);
0d9f3d85
A
1654 else
1655 {
1656 int i;
1657 uint8_t *srcPtr= src[0];
1658 uint8_t *dstPtr= dst[0] + dstStride[0]*srcSliceY;
1659
1660 for(i=0; i<srcSliceH; i++)
1661 {
e09d12f4 1662 conv(srcPtr, dstPtr, c->srcW*srcBpp);
0d9f3d85
A
1663 srcPtr+= srcStride[0];
1664 dstPtr+= dstStride[0];
1665 }
1666 }
1667}
1668
ec22603f
MN
1669static void bgr24toyv12Wrapper(SwsContext *c, uint8_t* src[], int srcStride[], int srcSliceY,
1670 int srcSliceH, uint8_t* dst[], int dstStride[]){
1671
1672 rgb24toyv12(
1673 src[0],
1674 dst[0]+ srcSliceY *dstStride[0],
1675 dst[1]+(srcSliceY>>1)*dstStride[1],
1676 dst[2]+(srcSliceY>>1)*dstStride[2],
1677 c->srcW, srcSliceH,
1678 dstStride[0], dstStride[1], srcStride[0]);
1679}
1680
b241cbf2
MN
1681static void yvu9toyv12Wrapper(SwsContext *c, uint8_t* src[], int srcStride[], int srcSliceY,
1682 int srcSliceH, uint8_t* dst[], int dstStride[]){
1683 int i;
1684
1685 /* copy Y */
1686 if(srcStride[0]==dstStride[0])
1687 memcpy(dst[0]+ srcSliceY*dstStride[0], src[0], srcStride[0]*srcSliceH);
1688 else{
1689 uint8_t *srcPtr= src[0];
1690 uint8_t *dstPtr= dst[0] + dstStride[0]*srcSliceY;
1691
1692 for(i=0; i<srcSliceH; i++)
1693 {
1694 memcpy(dstPtr, srcPtr, c->srcW);
1695 srcPtr+= srcStride[0];
1696 dstPtr+= dstStride[0];
1697 }
1698 }
1699
1700 if(c->dstFormat==IMGFMT_YV12){
1701 planar2x(src[1], dst[1], c->chrSrcW, c->chrSrcH, srcStride[1], dstStride[1]);
1702 planar2x(src[2], dst[2], c->chrSrcW, c->chrSrcH, srcStride[2], dstStride[2]);
1703 }else{
1704 planar2x(src[1], dst[2], c->chrSrcW, c->chrSrcH, srcStride[1], dstStride[2]);
1705 planar2x(src[2], dst[1], c->chrSrcW, c->chrSrcH, srcStride[2], dstStride[1]);
1706 }
1707}
1708
44c1035c
MN
1709/**
1710 * bring pointers in YUV order instead of YVU
1711 */
c7a810cc 1712static inline void orderYUV(int format, uint8_t * sortedP[], int sortedStride[], uint8_t * p[], int stride[]){
e616aa93 1713 if(format == IMGFMT_YV12 || format == IMGFMT_YVU9){
44c1035c
MN
1714 sortedP[0]= p[0];
1715 sortedP[1]= p[1];
1716 sortedP[2]= p[2];
c7a810cc
MN
1717 sortedStride[0]= stride[0];
1718 sortedStride[1]= stride[1];
1719 sortedStride[2]= stride[2];
44c1035c 1720 }
c7a810cc 1721 else if(isPacked(format) || isGray(format))
44c1035c
MN
1722 {
1723 sortedP[0]= p[0];
1724 sortedP[1]=
1725 sortedP[2]= NULL;
c7a810cc 1726 sortedStride[0]= stride[0];
44c1035c
MN
1727 sortedStride[1]=
1728 sortedStride[2]= 0;
1729 }
e09d12f4 1730 else if(format == IMGFMT_I420)
44c1035c
MN
1731 {
1732 sortedP[0]= p[0];
1733 sortedP[1]= p[2];
1734 sortedP[2]= p[1];
c7a810cc
MN
1735 sortedStride[0]= stride[0];
1736 sortedStride[1]= stride[2];
1737 sortedStride[2]= stride[1];
e09d12f4
MN
1738 }else{
1739 MSG_ERR("internal error in orderYUV\n");
44c1035c
MN
1740 }
1741}
b935781b 1742
b6654a54
MN
1743/* unscaled copy like stuff (assumes nearly identical formats) */
1744static void simpleCopy(SwsContext *c, uint8_t* srcParam[], int srcStrideParam[], int srcSliceY,
44c1035c 1745 int srcSliceH, uint8_t* dstParam[], int dstStrideParam[]){
b6654a54
MN
1746
1747 int srcStride[3];
44c1035c 1748 int dstStride[3];
b6654a54
MN
1749 uint8_t *src[3];
1750 uint8_t *dst[3];
1751
c7a810cc
MN
1752 orderYUV(c->srcFormat, src, srcStride, srcParam, srcStrideParam);
1753 orderYUV(c->dstFormat, dst, dstStride, dstParam, dstStrideParam);
b6654a54
MN
1754
1755 if(isPacked(c->srcFormat))
1756 {
1757 if(dstStride[0]==srcStride[0])
1758 memcpy(dst[0] + dstStride[0]*srcSliceY, src[0], srcSliceH*dstStride[0]);
1759 else
1760 {
1761 int i;
1762 uint8_t *srcPtr= src[0];
1763 uint8_t *dstPtr= dst[0] + dstStride[0]*srcSliceY;
a861d4d7
MN
1764 int length=0;
1765
1766 /* universal length finder */
9bd8bd1a
MN
1767 while(length+c->srcW <= ABS(dstStride[0])
1768 && length+c->srcW <= ABS(srcStride[0])) length+= c->srcW;
a861d4d7 1769 ASSERT(length!=0);
b6654a54
MN
1770
1771 for(i=0; i<srcSliceH; i++)
1772 {
1773 memcpy(dstPtr, srcPtr, length);
1774 srcPtr+= srcStride[0];
1775 dstPtr+= dstStride[0];
1776 }
1777 }
1778 }
1779 else
44c1035c 1780 { /* Planar YUV or gray */
b6654a54
MN
1781 int plane;
1782 for(plane=0; plane<3; plane++)
1783 {
e616aa93
MN
1784 int length= plane==0 ? c->srcW : -((-c->srcW )>>c->chrDstHSubSample);
1785 int y= plane==0 ? srcSliceY: -((-srcSliceY)>>c->chrDstVSubSample);
1786 int height= plane==0 ? srcSliceH: -((-srcSliceH)>>c->chrDstVSubSample);
44c1035c
MN
1787
1788 if((isGray(c->srcFormat) || isGray(c->dstFormat)) && plane>0)
0d9f3d85 1789 {
44c1035c 1790 if(!isGray(c->dstFormat))
e616aa93 1791 memset(dst[plane], 128, dstStride[plane]*height);
0d9f3d85 1792 }
b6654a54
MN
1793 else
1794 {
44c1035c
MN
1795 if(dstStride[plane]==srcStride[plane])
1796 memcpy(dst[plane] + dstStride[plane]*y, src[plane], height*dstStride[plane]);
1797 else
b6654a54 1798 {
44c1035c
MN
1799 int i;
1800 uint8_t *srcPtr= src[plane];
1801 uint8_t *dstPtr= dst[plane] + dstStride[plane]*y;
1802 for(i=0; i<height; i++)
1803 {
1804 memcpy(dstPtr, srcPtr, length);
1805 srcPtr+= srcStride[plane];
1806 dstPtr+= dstStride[plane];
1807 }
b6654a54
MN
1808 }
1809 }
1810 }
1811 }
37079906 1812}
28bf81c9 1813
44c1035c 1814static int remove_dup_fourcc(int fourcc)
0d9f3d85
A
1815{
1816 switch(fourcc)
1817 {
1818 case IMGFMT_IYUV: return IMGFMT_I420;
1819 case IMGFMT_Y8 : return IMGFMT_Y800;
0c51ef97 1820 case IMGFMT_IF09: return IMGFMT_YVU9;
0d9f3d85
A
1821 default: return fourcc;
1822 }
1823}
1824
c7a810cc
MN
1825static void getSubSampleFactors(int *h, int *v, int format){
1826 switch(format){
1827 case IMGFMT_YUY2:
1828 *h=1;
1829 *v=0;
1830 break;
1831 case IMGFMT_YV12:
1832 case IMGFMT_I420:
e616aa93 1833 case IMGFMT_Y800: //FIXME remove after different subsamplings are fully implemented
c7a810cc
MN
1834 *h=1;
1835 *v=1;
1836 break;
1837 case IMGFMT_YVU9:
1838 *h=2;
1839 *v=2;
1840 break;
1841 default:
1842 *h=0;
1843 *v=0;
1844 break;
1845 }
1846}
1847
28bf81c9
MN
1848SwsContext *getSwsContext(int srcW, int srcH, int srcFormat, int dstW, int dstH, int dstFormat, int flags,
1849 SwsFilter *srcFilter, SwsFilter *dstFilter){
1850
28bf81c9
MN
1851 SwsContext *c;
1852 int i;
37079906 1853 int usesFilter;
e09d12f4 1854 int unscaled, needsDither;
c7f822d9 1855 SwsFilter dummyFilter= {NULL, NULL, NULL, NULL};
5cebb24b
MN
1856#ifdef ARCH_X86
1857 if(gCpuCaps.hasMMX)
1858 asm volatile("emms\n\t"::: "memory");
1859#endif
28bf81c9 1860 if(swScale==NULL) globalInit();
e616aa93 1861//srcFormat= IMGFMT_Y800;
e09d12f4 1862//dstFormat= IMGFMT_Y800;
6ff0ad6b 1863 /* avoid dupplicate Formats, so we dont need to check to much */
0d9f3d85
A
1864 srcFormat = remove_dup_fourcc(srcFormat);
1865 dstFormat = remove_dup_fourcc(dstFormat);
44c1035c
MN
1866
1867 unscaled = (srcW == dstW && srcH == dstH);
e09d12f4
MN
1868 needsDither= (isBGR(dstFormat) || isRGB(dstFormat))
1869 && (dstFormat&0xFF)<24
1870 && ((dstFormat&0xFF)<(srcFormat&0xFF) || (!(isRGB(srcFormat) || isBGR(srcFormat))));
44c1035c
MN
1871
1872 if(!isSupportedIn(srcFormat))
b81cf274 1873 {
44c1035c
MN
1874 MSG_ERR("swScaler: %s is not supported as input format\n", vo_format_name(srcFormat));
1875 return NULL;
1876 }
1877 if(!isSupportedOut(dstFormat))
1878 {
1879 MSG_ERR("swScaler: %s is not supported as output format\n", vo_format_name(dstFormat));
1880 return NULL;
b81cf274 1881 }
44c1035c 1882
28bf81c9 1883 /* sanity check */
b81cf274
MN
1884 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
1885 {
0d9f3d85 1886 MSG_ERR("swScaler: %dx%d -> %dx%d is invalid scaling dimension\n",
b81cf274
MN
1887 srcW, srcH, dstW, dstH);
1888 return NULL;
1889 }
28bf81c9 1890
c7f822d9
MN
1891 if(!dstFilter) dstFilter= &dummyFilter;
1892 if(!srcFilter) srcFilter= &dummyFilter;
1893
28bf81c9 1894 c= memalign(64, sizeof(SwsContext));
c7f822d9 1895 memset(c, 0, sizeof(SwsContext));
28bf81c9
MN
1896
1897 c->srcW= srcW;
1898 c->srcH= srcH;
1899 c->dstW= dstW;
1900 c->dstH= dstH;
5521b193
MN
1901 c->lumXInc= ((srcW<<16) + (dstW>>1))/dstW;
1902 c->lumYInc= ((srcH<<16) + (dstH>>1))/dstH;
28bf81c9
MN
1903 c->flags= flags;
1904 c->dstFormat= dstFormat;
1905 c->srcFormat= srcFormat;
1906
37079906
MN
1907 usesFilter=0;
1908 if(dstFilter->lumV!=NULL && dstFilter->lumV->length>1) usesFilter=1;
1909 if(dstFilter->lumH!=NULL && dstFilter->lumH->length>1) usesFilter=1;
1910 if(dstFilter->chrV!=NULL && dstFilter->chrV->length>1) usesFilter=1;
1911 if(dstFilter->chrH!=NULL && dstFilter->chrH->length>1) usesFilter=1;
1912 if(srcFilter->lumV!=NULL && srcFilter->lumV->length>1) usesFilter=1;
1913 if(srcFilter->lumH!=NULL && srcFilter->lumH->length>1) usesFilter=1;
1914 if(srcFilter->chrV!=NULL && srcFilter->chrV->length>1) usesFilter=1;
1915 if(srcFilter->chrH!=NULL && srcFilter->chrH->length>1) usesFilter=1;
e616aa93
MN
1916
1917 getSubSampleFactors(&c->chrSrcHSubSample, &c->chrSrcVSubSample, srcFormat);
1918 getSubSampleFactors(&c->chrDstHSubSample, &c->chrDstVSubSample, dstFormat);
1919
1920 // reuse chroma for 2 pixles rgb/bgr unless user wants full chroma interpolation
1921 if((isBGR(dstFormat) || isRGB(dstFormat)) && !(flags&SWS_FULL_CHR_H_INT)) c->chrDstHSubSample=1;
1922
5859233b
MN
1923 // drop some chroma lines if the user wants it
1924 c->vChrDrop= (flags&SWS_SRC_V_CHR_DROP_MASK)>>SWS_SRC_V_CHR_DROP_SHIFT;
1925 c->chrSrcVSubSample+= c->vChrDrop;
e616aa93 1926
5859233b 1927 // drop every 2. pixel for chroma calculation unless user wants full chroma
e616aa93
MN
1928 if((isBGR(srcFormat) || isRGB(srcFormat)) && !(flags&SWS_FULL_CHR_H_INP))
1929 c->chrSrcHSubSample=1;
1930
1931 c->chrIntHSubSample= c->chrDstHSubSample;
1932 c->chrIntVSubSample= c->chrSrcVSubSample;
1933
1934 // note the -((-x)>>y) is so that we allways round toward +inf
1935 c->chrSrcW= -((-srcW) >> c->chrSrcHSubSample);
1936 c->chrSrcH= -((-srcH) >> c->chrSrcVSubSample);
1937 c->chrDstW= -((-dstW) >> c->chrDstHSubSample);
1938 c->chrDstH= -((-dstH) >> c->chrDstVSubSample);
37079906 1939
cf7d1c1a
MN
1940 if(isBGR(dstFormat))
1941 c->yuvTable= yuv2rgb_c_init(dstFormat & 0xFF, MODE_RGB, c->table_rV, c->table_gU, c->table_gV, c->table_bU);
1942 if(isRGB(dstFormat))
1943 c->yuvTable= yuv2rgb_c_init(dstFormat & 0xFF, MODE_BGR, c->table_rV, c->table_gU, c->table_gV, c->table_bU);
1944
b935781b 1945 /* unscaled special Cases */
44c1035c 1946 if(unscaled && !usesFilter)
37079906 1947 {
0d9f3d85
A
1948 /* yv12_to_nv12 */
1949 if((srcFormat == IMGFMT_YV12||srcFormat==IMGFMT_I420)&&dstFormat == IMGFMT_NV12)
1950 {
1951 c->swScale= PlanarToNV12Wrapper;
1952
1953 if(flags&SWS_PRINT_INFO)
1954 MSG_INFO("SwScaler: using unscaled %s -> %s special converter\n",
1955 vo_format_name(srcFormat), vo_format_name(dstFormat));
1956 return c;
1957 }
1958 /* yv12_to_yuy2 */
1959 if((srcFormat == IMGFMT_YV12||srcFormat==IMGFMT_I420)&&dstFormat == IMGFMT_YUY2)
1960 {
44c1035c 1961 c->swScale= PlanarToYuy2Wrapper;
0d9f3d85
A
1962
1963 if(flags&SWS_PRINT_INFO)
1964 MSG_INFO("SwScaler: using unscaled %s -> %s special converter\n",
1965 vo_format_name(srcFormat), vo_format_name(dstFormat));
1966 return c;
1967 }
37079906 1968 /* yuv2bgr */
e616aa93 1969 if((srcFormat==IMGFMT_YV12 || srcFormat==IMGFMT_I420) && isBGR(dstFormat))
37079906
MN
1970 {
1971 // FIXME multiple yuv2rgb converters wont work that way cuz that thing is full of globals&statics
cf7d1c1a 1972 //FIXME rgb vs. bgr ?
1e1c4fe9 1973#ifdef WORDS_BIGENDIAN
daa57641
MN
1974 if(dstFormat==IMGFMT_BGR32)
1975 yuv2rgb_init( dstFormat&0xFF /* =bpp */, MODE_BGR);
1976 else
1977 yuv2rgb_init( dstFormat&0xFF /* =bpp */, MODE_RGB);
1e1c4fe9 1978#else
b6654a54 1979 yuv2rgb_init( dstFormat&0xFF /* =bpp */, MODE_RGB);
1e1c4fe9 1980#endif
37079906 1981 c->swScale= planarYuvToBgr;
b6654a54
MN
1982
1983 if(flags&SWS_PRINT_INFO)
0d9f3d85 1984 MSG_INFO("SwScaler: using unscaled %s -> %s special converter\n",
b6654a54
MN
1985 vo_format_name(srcFormat), vo_format_name(dstFormat));
1986 return c;
1987 }
b241cbf2 1988
b6654a54 1989 /* simple copy */
e616aa93
MN
1990 if( srcFormat == dstFormat
1991 || (srcFormat==IMGFMT_YV12 && dstFormat==IMGFMT_I420)
1992 || (srcFormat==IMGFMT_I420 && dstFormat==IMGFMT_YV12)
1993 || (isPlanarYUV(srcFormat) && isGray(dstFormat))
1994 || (isPlanarYUV(dstFormat) && isGray(srcFormat))
1995 )
b6654a54
MN
1996 {
1997 c->swScale= simpleCopy;
1998
37079906 1999 if(flags&SWS_PRINT_INFO)
0d9f3d85 2000 MSG_INFO("SwScaler: using unscaled %s -> %s special converter\n",
37079906
MN
2001 vo_format_name(srcFormat), vo_format_name(dstFormat));
2002 return c;
2003 }
b241cbf2
MN
2004
2005 if( srcFormat==IMGFMT_YVU9 && (dstFormat==IMGFMT_YV12 || dstFormat==IMGFMT_I420) )
2006 {
2007 c->swScale= yvu9toyv12Wrapper;
2008
2009 if(flags&SWS_PRINT_INFO)
2010 MSG_INFO("SwScaler: using unscaled %s -> %s special converter\n",
2011 vo_format_name(srcFormat), vo_format_name(dstFormat));
2012 return c;
2013 }
2014
ec22603f
MN
2015 /* bgr24toYV12 */
2016 if(srcFormat==IMGFMT_BGR24 && dstFormat==IMGFMT_YV12)
ec22603f 2017 c->swScale= bgr24toyv12Wrapper;
e09d12f4
MN
2018
2019 /* rgb/bgr -> rgb/bgr (no dither needed forms) */
2020 if( (isBGR(srcFormat) || isRGB(srcFormat))
2021 && (isBGR(dstFormat) || isRGB(dstFormat))
2022 && !needsDither)
2023 c->swScale= rgb2rgbWrapper;
2024
2025 /* LQ converters if -sws 0 or -sws 4*/
2026 if(c->flags&(SWS_FAST_BILINEAR|SWS_POINT)){
2027 /* rgb/bgr -> rgb/bgr (dither needed forms) */
2028 if( (isBGR(srcFormat) || isRGB(srcFormat))
2029 && (isBGR(dstFormat) || isRGB(dstFormat))
2030 && needsDither)
2031 c->swScale= rgb2rgbWrapper;
2032 }
ec22603f 2033
e09d12f4 2034 if(c->swScale){
ec22603f 2035 if(flags&SWS_PRINT_INFO)
0d9f3d85 2036 MSG_INFO("SwScaler: using unscaled %s -> %s special converter\n",
ec22603f
MN
2037 vo_format_name(srcFormat), vo_format_name(dstFormat));
2038 return c;
2039 }
37079906
MN
2040 }
2041
28bf81c9
MN
2042 if(cpuCaps.hasMMX2)
2043 {
2044 c->canMMX2BeUsed= (dstW >=srcW && (dstW&31)==0 && (srcW&15)==0) ? 1 : 0;
2045 if(!c->canMMX2BeUsed && dstW >=srcW && (srcW&15)==0 && (flags&SWS_FAST_BILINEAR))
2046 {
2047 if(flags&SWS_PRINT_INFO)
0d9f3d85 2048 MSG_INFO("SwScaler: output Width is not a multiple of 32 -> no MMX2 scaler\n");
28bf81c9
MN
2049 }
2050 }
2051 else
2052 c->canMMX2BeUsed=0;
2053
1e621b18
MN
2054 c->chrXInc= ((c->chrSrcW<<16) + (c->chrDstW>>1))/c->chrDstW;
2055 c->chrYInc= ((c->chrSrcH<<16) + (c->chrDstH>>1))/c->chrDstH;
2056
28bf81c9
MN
2057 // match pixel 0 of the src to pixel 0 of dst and match pixel n-2 of src to pixel n-2 of dst
2058 // but only for the FAST_BILINEAR mode otherwise do correct scaling
2059 // n-2 is the last chrominance sample available
2060 // this is not perfect, but noone shuld notice the difference, the more correct variant
2061 // would be like the vertical one, but that would require some special code for the
2062 // first and last pixel
2063 if(flags&SWS_FAST_BILINEAR)
2064 {
1e621b18
MN
2065 if(c->canMMX2BeUsed)
2066 {
2067 c->lumXInc+= 20;
2068 c->chrXInc+= 20;
2069 }
28bf81c9 2070 //we dont use the x86asm scaler if mmx is available
1e621b18
MN
2071 else if(cpuCaps.hasMMX)
2072 {
2073 c->lumXInc = ((srcW-2)<<16)/(dstW-2) - 20;
2074 c->chrXInc = ((c->chrSrcW-2)<<16)/(c->chrDstW-2) - 20;
2075 }
28bf81c9
MN
2076 }
2077
28bf81c9
MN
2078 /* precalculate horizontal scaler filter coefficients */
2079 {
2080 const int filterAlign= cpuCaps.hasMMX ? 4 : 1;
2081
c7f822d9 2082 initFilter(&c->hLumFilter, &c->hLumFilterPos, &c->hLumFilterSize, c->lumXInc,
81a571a8
MN
2083 srcW , dstW, filterAlign, 1<<14,
2084 (flags&SWS_BICUBLIN) ? (flags|SWS_BICUBIC) : flags,
c7f822d9
MN
2085 srcFilter->lumH, dstFilter->lumH);
2086 initFilter(&c->hChrFilter, &c->hChrFilterPos, &c->hChrFilterSize, c->chrXInc,
81a571a8
MN
2087 c->chrSrcW, c->chrDstW, filterAlign, 1<<14,
2088 (flags&SWS_BICUBLIN) ? (flags|SWS_BILINEAR) : flags,
c7f822d9 2089 srcFilter->chrH, dstFilter->chrH);
28bf81c9
MN
2090
2091#ifdef ARCH_X86
2092// cant downscale !!!
2093 if(c->canMMX2BeUsed && (flags & SWS_FAST_BILINEAR))
2094 {
b7dc6f66
MN
2095 c->lumMmx2Filter = (int16_t*)memalign(8, (dstW /8+8)*sizeof(int16_t));
2096 c->chrMmx2Filter = (int16_t*)memalign(8, (c->chrDstW /4+8)*sizeof(int16_t));
2097 c->lumMmx2FilterPos= (int32_t*)memalign(8, (dstW /2/8+8)*sizeof(int32_t));
2098 c->chrMmx2FilterPos= (int32_t*)memalign(8, (c->chrDstW/2/4+8)*sizeof(int32_t));
2099
2100 initMMX2HScaler( dstW, c->lumXInc, c->funnyYCode , c->lumMmx2Filter, c->lumMmx2FilterPos, 8);
2101 initMMX2HScaler(c->chrDstW, c->chrXInc, c->funnyUVCode, c->chrMmx2Filter, c->chrMmx2FilterPos, 4);
28bf81c9
MN
2102 }
2103#endif
2104 } // Init Horizontal stuff
2105
2106
2107
2108 /* precalculate vertical scaler filter coefficients */
c7f822d9 2109 initFilter(&c->vLumFilter, &c->vLumFilterPos, &c->vLumFilterSize, c->lumYInc,
81a571a8
MN
2110 srcH , dstH, 1, (1<<12)-4,
2111 (flags&SWS_BICUBLIN) ? (flags|SWS_BICUBIC) : flags,
c7f822d9
MN
2112 srcFilter->lumV, dstFilter->lumV);
2113 initFilter(&c->vChrFilter, &c->vChrFilterPos, &c->vChrFilterSize, c->chrYInc,
81a571a8
MN
2114 c->chrSrcH, c->chrDstH, 1, (1<<12)-4,
2115 (flags&SWS_BICUBLIN) ? (flags|SWS_BILINEAR) : flags,
2116 srcFilter->chrV, dstFilter->chrV);
28bf81c9
MN
2117
2118 // Calculate Buffer Sizes so that they wont run out while handling these damn slices
2119 c->vLumBufSize= c->vLumFilterSize;
2120 c->vChrBufSize= c->vChrFilterSize;
2121 for(i=0; i<dstH; i++)
2122 {
2123 int chrI= i*c->chrDstH / dstH;
2124 int nextSlice= MAX(c->vLumFilterPos[i ] + c->vLumFilterSize - 1,
e616aa93
MN
2125 ((c->vChrFilterPos[chrI] + c->vChrFilterSize - 1)<<c->chrSrcVSubSample));
2126 nextSlice&= ~3; // Slices start at boundaries which are divisable through 4
28bf81c9
MN
2127 if(c->vLumFilterPos[i ] + c->vLumBufSize < nextSlice)
2128 c->vLumBufSize= nextSlice - c->vLumFilterPos[i ];
e616aa93
MN
2129 if(c->vChrFilterPos[chrI] + c->vChrBufSize < (nextSlice>>c->chrSrcVSubSample))
2130 c->vChrBufSize= (nextSlice>>c->chrSrcVSubSample) - c->vChrFilterPos[chrI];
28bf81c9
MN
2131 }
2132
2133 // allocate pixbufs (we use dynamic allocation because otherwise we would need to
c7f822d9
MN
2134 c->lumPixBuf= (int16_t**)memalign(4, c->vLumBufSize*2*sizeof(int16_t*));
2135 c->chrPixBuf= (int16_t**)memalign(4, c->vChrBufSize*2*sizeof(int16_t*));
6c7506de 2136 //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)
28bf81c9
MN
2137 for(i=0; i<c->vLumBufSize; i++)
2138 c->lumPixBuf[i]= c->lumPixBuf[i+c->vLumBufSize]= (uint16_t*)memalign(8, 4000);
2139 for(i=0; i<c->vChrBufSize; i++)
2140 c->chrPixBuf[i]= c->chrPixBuf[i+c->vChrBufSize]= (uint16_t*)memalign(8, 8000);
2141
2142 //try to avoid drawing green stuff between the right end and the stride end
2143 for(i=0; i<c->vLumBufSize; i++) memset(c->lumPixBuf[i], 0, 4000);
2144 for(i=0; i<c->vChrBufSize; i++) memset(c->chrPixBuf[i], 64, 8000);
2145
2146 ASSERT(c->chrDstH <= dstH)
28bf81c9
MN
2147
2148 // pack filter data for mmx code
2149 if(cpuCaps.hasMMX)
2150 {
c7f822d9
MN
2151 c->lumMmxFilter= (int16_t*)memalign(8, c->vLumFilterSize* dstH*4*sizeof(int16_t));
2152 c->chrMmxFilter= (int16_t*)memalign(8, c->vChrFilterSize*c->chrDstH*4*sizeof(int16_t));
28bf81c9
MN
2153 for(i=0; i<c->vLumFilterSize*dstH; i++)
2154 c->lumMmxFilter[4*i]=c->lumMmxFilter[4*i+1]=c->lumMmxFilter[4*i+2]=c->lumMmxFilter[4*i+3]=
2155 c->vLumFilter[i];
2156 for(i=0; i<c->vChrFilterSize*c->chrDstH; i++)
2157 c->chrMmxFilter[4*i]=c->chrMmxFilter[4*i+1]=c->chrMmxFilter[4*i+2]=c->chrMmxFilter[4*i+3]=
2158 c->vChrFilter[i];
2159 }
2160
2161 if(flags&SWS_PRINT_INFO)
2162 {
2163#ifdef DITHER1XBPP
5521b193
MN
2164 char *dither= " dithered";
2165#else
2166 char *dither= "";
28bf81c9
MN
2167#endif
2168 if(flags&SWS_FAST_BILINEAR)
0d9f3d85 2169 MSG_INFO("\nSwScaler: FAST_BILINEAR scaler, ");
28bf81c9 2170 else if(flags&SWS_BILINEAR)
0d9f3d85 2171 MSG_INFO("\nSwScaler: BILINEAR scaler, ");
28bf81c9 2172 else if(flags&SWS_BICUBIC)
0d9f3d85 2173 MSG_INFO("\nSwScaler: BICUBIC scaler, ");
1e621b18 2174 else if(flags&SWS_X)
0d9f3d85 2175 MSG_INFO("\nSwScaler: Experimental scaler, ");
ff7ba856 2176 else if(flags&SWS_POINT)
0d9f3d85 2177 MSG_INFO("\nSwScaler: Nearest Neighbor / POINT scaler, ");
d8863d37 2178 else if(flags&SWS_AREA)
0d9f3d85 2179 MSG_INFO("\nSwScaler: Area Averageing scaler, ");
81a571a8
MN
2180 else if(flags&SWS_BICUBLIN)
2181 MSG_INFO("\nSwScaler: luma BICUBIC / chroma BILINEAR, ");
28bf81c9 2182 else
0d9f3d85 2183 MSG_INFO("\nSwScaler: ehh flags invalid?! ");
28bf81c9 2184
0d9f3d85
A
2185 if(dstFormat==IMGFMT_BGR15 || dstFormat==IMGFMT_BGR16)
2186 MSG_INFO("from %s to%s %s ",
2187 vo_format_name(srcFormat), dither, vo_format_name(dstFormat));
2188 else
2189 MSG_INFO("from %s to %s ",
2190 vo_format_name(srcFormat), vo_format_name(dstFormat));
28bf81c9
MN
2191
2192 if(cpuCaps.hasMMX2)
0d9f3d85 2193 MSG_INFO("using MMX2\n");
28bf81c9 2194 else if(cpuCaps.has3DNow)
0d9f3d85 2195 MSG_INFO("using 3DNOW\n");
28bf81c9 2196 else if(cpuCaps.hasMMX)
0d9f3d85 2197 MSG_INFO("using MMX\n");
28bf81c9 2198 else
0d9f3d85 2199 MSG_INFO("using C\n");
28bf81c9
MN
2200 }
2201
2202 if((flags & SWS_PRINT_INFO) && verbose)
2203 {
2204 if(cpuCaps.hasMMX)
2205 {
2206 if(c->canMMX2BeUsed && (flags&SWS_FAST_BILINEAR))
0d9f3d85 2207 MSG_V("SwScaler: using FAST_BILINEAR MMX2 scaler for horizontal scaling\n");
28bf81c9
MN
2208 else
2209 {
2210 if(c->hLumFilterSize==4)
0d9f3d85 2211 MSG_V("SwScaler: using 4-tap MMX scaler for horizontal luminance scaling\n");
28bf81c9 2212 else if(c->hLumFilterSize==8)
0d9f3d85 2213 MSG_V("SwScaler: using 8-tap MMX scaler for horizontal luminance scaling\n");
28bf81c9 2214 else
0d9f3d85 2215 MSG_V("SwScaler: using n-tap MMX scaler for horizontal luminance scaling\n");
28bf81c9
MN
2216
2217 if(c->hChrFilterSize==4)
0d9f3d85 2218 MSG_V("SwScaler: using 4-tap MMX scaler for horizontal chrominance scaling\n");
28bf81c9 2219 else if(c->hChrFilterSize==8)
0d9f3d85 2220 MSG_V("SwScaler: using 8-tap MMX scaler for horizontal chrominance scaling\n");
28bf81c9 2221 else
0d9f3d85 2222 MSG_V("SwScaler: using n-tap MMX scaler for horizontal chrominance scaling\n");
28bf81c9
MN
2223 }
2224 }
2225 else
2226 {
2227#ifdef ARCH_X86
0d9f3d85 2228 MSG_V("SwScaler: using X86-Asm scaler for horizontal scaling\n");
28bf81c9
MN
2229#else
2230 if(flags & SWS_FAST_BILINEAR)
0d9f3d85 2231 MSG_V("SwScaler: using FAST_BILINEAR C scaler for horizontal scaling\n");
28bf81c9 2232 else
0d9f3d85 2233 MSG_V("SwScaler: using C scaler for horizontal scaling\n");
28bf81c9
MN
2234#endif
2235 }
6c7506de 2236 if(isPlanarYUV(dstFormat))
28bf81c9
MN
2237 {
2238 if(c->vLumFilterSize==1)
0d9f3d85 2239 MSG_V("SwScaler: using 1-tap %s \"scaler\" for vertical scaling (YV12 like)\n", cpuCaps.hasMMX ? "MMX" : "C");
28bf81c9 2240 else
0d9f3d85 2241 MSG_V("SwScaler: using n-tap %s scaler for vertical scaling (YV12 like)\n", cpuCaps.hasMMX ? "MMX" : "C");
28bf81c9
MN
2242 }
2243 else
2244 {
2245 if(c->vLumFilterSize==1 && c->vChrFilterSize==2)
0d9f3d85 2246 MSG_V("SwScaler: using 1-tap %s \"scaler\" for vertical luminance scaling (BGR)\n"
28bf81c9
MN
2247 "SwScaler: 2-tap scaler for vertical chrominance scaling (BGR)\n",cpuCaps.hasMMX ? "MMX" : "C");
2248 else if(c->vLumFilterSize==2 && c->vChrFilterSize==2)
0d9f3d85 2249 MSG_V("SwScaler: using 2-tap linear %s scaler for vertical scaling (BGR)\n", cpuCaps.hasMMX ? "MMX" : "C");
28bf81c9 2250 else
0d9f3d85 2251 MSG_V("SwScaler: using n-tap %s scaler for vertical scaling (BGR)\n", cpuCaps.hasMMX ? "MMX" : "C");
28bf81c9
MN
2252 }
2253
2254 if(dstFormat==IMGFMT_BGR24)
0d9f3d85 2255 MSG_V("SwScaler: using %s YV12->BGR24 Converter\n",
28bf81c9 2256 cpuCaps.hasMMX2 ? "MMX2" : (cpuCaps.hasMMX ? "MMX" : "C"));
fd284805 2257 else if(dstFormat==IMGFMT_BGR32)
0d9f3d85 2258 MSG_V("SwScaler: using %s YV12->BGR32 Converter\n", cpuCaps.hasMMX ? "MMX" : "C");
fd284805 2259 else if(dstFormat==IMGFMT_BGR16)
0d9f3d85 2260 MSG_V("SwScaler: using %s YV12->BGR16 Converter\n", cpuCaps.hasMMX ? "MMX" : "C");
fd284805 2261 else if(dstFormat==IMGFMT_BGR15)
0d9f3d85 2262 MSG_V("SwScaler: using %s YV12->BGR15 Converter\n", cpuCaps.hasMMX ? "MMX" : "C");
28bf81c9 2263
0d9f3d85 2264 MSG_V("SwScaler: %dx%d -> %dx%d\n", srcW, srcH, dstW, dstH);
28bf81c9 2265 }
1e621b18
MN
2266 if((flags & SWS_PRINT_INFO) && verbose>1)
2267 {
0d9f3d85 2268 MSG_DBG2("SwScaler:Lum srcW=%d srcH=%d dstW=%d dstH=%d xInc=%d yInc=%d\n",
1e621b18 2269 c->srcW, c->srcH, c->dstW, c->dstH, c->lumXInc, c->lumYInc);
0d9f3d85 2270 MSG_DBG2("SwScaler:Chr srcW=%d srcH=%d dstW=%d dstH=%d xInc=%d yInc=%d\n",
1e621b18
MN
2271 c->chrSrcW, c->chrSrcH, c->chrDstW, c->chrDstH, c->chrXInc, c->chrYInc);
2272 }
37079906
MN
2273
2274 c->swScale= swScale;
28bf81c9
MN
2275 return c;
2276}
2277
2278/**
2279 * returns a normalized gaussian curve used to filter stuff
2280 * quality=3 is high quality, lowwer is lowwer quality
2281 */
c7f822d9
MN
2282
2283SwsVector *getGaussianVec(double variance, double quality){
28bf81c9
MN
2284 const int length= (int)(variance*quality + 0.5) | 1;
2285 int i;
2286 double *coeff= memalign(sizeof(double), length*sizeof(double));
2287 double middle= (length-1)*0.5;
c7f822d9
MN
2288 SwsVector *vec= malloc(sizeof(SwsVector));
2289
2290 vec->coeff= coeff;
2291 vec->length= length;
28bf81c9
MN
2292
2293 for(i=0; i<length; i++)
2294 {
2295 double dist= i-middle;
2296 coeff[i]= exp( -dist*dist/(2*variance*variance) ) / sqrt(2*variance*PI);
2297 }
2298
c7f822d9
MN
2299 normalizeVec(vec, 1.0);
2300
2301 return vec;
28bf81c9
MN
2302}
2303
5521b193
MN
2304SwsVector *getConstVec(double c, int length){
2305 int i;
2306 double *coeff= memalign(sizeof(double), length*sizeof(double));
2307 SwsVector *vec= malloc(sizeof(SwsVector));
2308
2309 vec->coeff= coeff;
2310 vec->length= length;
2311
2312 for(i=0; i<length; i++)
2313 coeff[i]= c;
2314
2315 return vec;
2316}
2317
2318
c7f822d9
MN
2319SwsVector *getIdentityVec(void){
2320 double *coeff= memalign(sizeof(double), sizeof(double));
2321 SwsVector *vec= malloc(sizeof(SwsVector));
2322 coeff[0]= 1.0;
2323
2324 vec->coeff= coeff;
2325 vec->length= 1;
2326
2327 return vec;
2328}
2329
2330void normalizeVec(SwsVector *a, double height){
28bf81c9
MN
2331 int i;
2332 double sum=0;
2333 double inv;
2334
c7f822d9
MN
2335 for(i=0; i<a->length; i++)
2336 sum+= a->coeff[i];
28bf81c9
MN
2337
2338 inv= height/sum;
2339
c7f822d9
MN
2340 for(i=0; i<a->length; i++)
2341 a->coeff[i]*= height;
28bf81c9
MN
2342}
2343
c7f822d9
MN
2344void scaleVec(SwsVector *a, double scalar){
2345 int i;
2346
2347 for(i=0; i<a->length; i++)
2348 a->coeff[i]*= scalar;
2349}
2350
5cebb24b 2351static SwsVector *getConvVec(SwsVector *a, SwsVector *b){
c7f822d9 2352 int length= a->length + b->length - 1;
28bf81c9
MN
2353 double *coeff= memalign(sizeof(double), length*sizeof(double));
2354 int i, j;
c7f822d9
MN
2355 SwsVector *vec= malloc(sizeof(SwsVector));
2356
2357 vec->coeff= coeff;
2358 vec->length= length;
28bf81c9
MN
2359
2360 for(i=0; i<length; i++) coeff[i]= 0.0;
2361
c7f822d9 2362 for(i=0; i<a->length; i++)
28bf81c9 2363 {
c7f822d9 2364 for(j=0; j<b->length; j++)
28bf81c9 2365 {
c7f822d9 2366 coeff[i+j]+= a->coeff[i]*b->coeff[j];
28bf81c9
MN
2367 }
2368 }
2369
c7f822d9 2370 return vec;
28bf81c9
MN
2371}
2372
5cebb24b 2373static SwsVector *sumVec(SwsVector *a, SwsVector *b){
c7f822d9 2374 int length= MAX(a->length, b->length);
28bf81c9
MN
2375 double *coeff= memalign(sizeof(double), length*sizeof(double));
2376 int i;
c7f822d9
MN
2377 SwsVector *vec= malloc(sizeof(SwsVector));
2378
2379 vec->coeff= coeff;
2380 vec->length= length;
28bf81c9
MN
2381
2382 for(i=0; i<length; i++) coeff[i]= 0.0;
2383
c7f822d9
MN
2384 for(i=0; i<a->length; i++) coeff[i + (length-1)/2 - (a->length-1)/2]+= a->coeff[i];
2385 for(i=0; i<b->length; i++) coeff[i + (length-1)/2 - (b->length-1)/2]+= b->coeff[i];
2386
2387 return vec;
28bf81c9 2388}
c7f822d9 2389
5cebb24b 2390static SwsVector *diffVec(SwsVector *a, SwsVector *b){
c7f822d9
MN
2391 int length= MAX(a->length, b->length);
2392 double *coeff= memalign(sizeof(double), length*sizeof(double));
2393 int i;
2394 SwsVector *vec= malloc(sizeof(SwsVector));
2395
2396 vec->coeff= coeff;
2397 vec->length= length;
2398
2399 for(i=0; i<length; i++) coeff[i]= 0.0;
2400
2401 for(i=0; i<a->length; i++) coeff[i + (length-1)/2 - (a->length-1)/2]+= a->coeff[i];
2402 for(i=0; i<b->length; i++) coeff[i + (length-1)/2 - (b->length-1)/2]-= b->coeff[i];
2403
2404 return vec;
2405}
2406
2407/* shift left / or right if "shift" is negative */
5cebb24b 2408static SwsVector *getShiftedVec(SwsVector *a, int shift){
c7f822d9
MN
2409 int length= a->length + ABS(shift)*2;
2410 double *coeff= memalign(sizeof(double), length*sizeof(double));
ff7ba856 2411 int i;
c7f822d9
MN
2412 SwsVector *vec= malloc(sizeof(SwsVector));
2413
2414 vec->coeff= coeff;
2415 vec->length= length;
2416
2417 for(i=0; i<length; i++) coeff[i]= 0.0;
2418
2419 for(i=0; i<a->length; i++)
2420 {
2421 coeff[i + (length-1)/2 - (a->length-1)/2 - shift]= a->coeff[i];
2422 }
2423
2424 return vec;
2425}
2426
5cebb24b
MN
2427void shiftVec(SwsVector *a, int shift){
2428 SwsVector *shifted= getShiftedVec(a, shift);
2429 free(a->coeff);
2430 a->coeff= shifted->coeff;
2431 a->length= shifted->length;
2432 free(shifted);
2433}
2434
2435void addVec(SwsVector *a, SwsVector *b){
2436 SwsVector *sum= sumVec(a, b);
2437 free(a->coeff);
2438 a->coeff= sum->coeff;
2439 a->length= sum->length;
2440 free(sum);
2441}
2442
2443void subVec(SwsVector *a, SwsVector *b){
2444 SwsVector *diff= diffVec(a, b);
2445 free(a->coeff);
2446 a->coeff= diff->coeff;
2447 a->length= diff->length;
2448 free(diff);
2449}
2450
2451void convVec(SwsVector *a, SwsVector *b){
2452 SwsVector *conv= getConvVec(a, b);
2453 free(a->coeff);
2454 a->coeff= conv->coeff;
2455 a->length= conv->length;
2456 free(conv);
2457}
2458
2459SwsVector *cloneVec(SwsVector *a){
2460 double *coeff= memalign(sizeof(double), a->length*sizeof(double));
2461 int i;
2462 SwsVector *vec= malloc(sizeof(SwsVector));
2463
2464 vec->coeff= coeff;
2465 vec->length= a->length;
2466
2467 for(i=0; i<a->length; i++) coeff[i]= a->coeff[i];
2468
2469 return vec;
2470}
2471
c7f822d9
MN
2472void printVec(SwsVector *a){
2473 int i;
2474 double max=0;
2475 double min=0;
2476 double range;
2477
2478 for(i=0; i<a->length; i++)
2479 if(a->coeff[i]>max) max= a->coeff[i];
2480
2481 for(i=0; i<a->length; i++)
2482 if(a->coeff[i]<min) min= a->coeff[i];
2483
2484 range= max - min;
2485
2486 for(i=0; i<a->length; i++)
2487 {
2488 int x= (int)((a->coeff[i]-min)*60.0/range +0.5);
0d9f3d85
A
2489 MSG_DBG2("%1.3f ", a->coeff[i]);
2490 for(;x>0; x--) MSG_DBG2(" ");
2491 MSG_DBG2("|\n");
c7f822d9
MN
2492 }
2493}
2494
2495void freeVec(SwsVector *a){
2496 if(!a) return;
2497 if(a->coeff) free(a->coeff);
2498 a->coeff=NULL;
2499 a->length=0;
2500 free(a);
2501}
2502
2503void freeSwsContext(SwsContext *c){
2504 int i;
c7f822d9
MN
2505 if(!c) return;
2506
2507 if(c->lumPixBuf)
2508 {
6c7506de 2509 for(i=0; i<c->vLumBufSize; i++)
c7f822d9
MN
2510 {
2511 if(c->lumPixBuf[i]) free(c->lumPixBuf[i]);
2512 c->lumPixBuf[i]=NULL;
2513 }
2514 free(c->lumPixBuf);
2515 c->lumPixBuf=NULL;
2516 }
2517
2518 if(c->chrPixBuf)
2519 {
6c7506de 2520 for(i=0; i<c->vChrBufSize; i++)
c7f822d9
MN
2521 {
2522 if(c->chrPixBuf[i]) free(c->chrPixBuf[i]);
2523 c->chrPixBuf[i]=NULL;
2524 }
2525 free(c->chrPixBuf);
2526 c->chrPixBuf=NULL;
2527 }
2528
2529 if(c->vLumFilter) free(c->vLumFilter);
2530 c->vLumFilter = NULL;
2531 if(c->vChrFilter) free(c->vChrFilter);
2532 c->vChrFilter = NULL;
2533 if(c->hLumFilter) free(c->hLumFilter);
2534 c->hLumFilter = NULL;
2535 if(c->hChrFilter) free(c->hChrFilter);
2536 c->hChrFilter = NULL;
2537
2538 if(c->vLumFilterPos) free(c->vLumFilterPos);
2539 c->vLumFilterPos = NULL;
2540 if(c->vChrFilterPos) free(c->vChrFilterPos);
2541 c->vChrFilterPos = NULL;
2542 if(c->hLumFilterPos) free(c->hLumFilterPos);
2543 c->hLumFilterPos = NULL;
2544 if(c->hChrFilterPos) free(c->hChrFilterPos);
2545 c->hChrFilterPos = NULL;
2546
2547 if(c->lumMmxFilter) free(c->lumMmxFilter);
2548 c->lumMmxFilter = NULL;
2549 if(c->chrMmxFilter) free(c->chrMmxFilter);
2550 c->chrMmxFilter = NULL;
2551
b7dc6f66
MN
2552 if(c->lumMmx2Filter) free(c->lumMmx2Filter);
2553 c->lumMmx2Filter=NULL;
2554 if(c->chrMmx2Filter) free(c->chrMmx2Filter);
2555 c->chrMmx2Filter=NULL;
2556 if(c->lumMmx2FilterPos) free(c->lumMmx2FilterPos);
2557 c->lumMmx2FilterPos=NULL;
2558 if(c->chrMmx2FilterPos) free(c->chrMmx2FilterPos);
2559 c->chrMmx2FilterPos=NULL;
cf7d1c1a
MN
2560 if(c->yuvTable) free(c->yuvTable);
2561 c->yuvTable=NULL;
b7dc6f66 2562
c7f822d9
MN
2563 free(c);
2564}
2565
7f56a527 2566