* Making AVI encoding predictable (all JUNK chunks are filled with 0)
[libav.git] / postproc / swscale.c
CommitLineData
fe8054c0 1/*
5427e242 2 Copyright (C) 2001-2003 Michael Niedermayer <michaelni@gmx.at>
fe8054c0
MN
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/*
7322a67c 20 supported Input formats: YV12, I420/IYUV, YUY2, UYVY, BGR32, BGR24, BGR16, BGR15, RGB32, RGB24, Y8/Y800, YVU9/IF09
46de8b73 21 supported output formats: YV12, I420/IYUV, YUY2, {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"
5427e242 65#include "swscale_internal.h"
7630f2e0 66#include "../cpudetect.h"
a861d4d7 67#include "../bswap.h"
28bf81c9 68#include "../libvo/img_format.h"
37079906 69#include "rgb2rgb.h"
b0db4198 70#include "../libvo/fastmemcpy.h"
4a53a912 71#include "../mp_msg.h"
0d9f3d85
A
72
73#define MSG_WARN(args...) mp_msg(MSGT_SWS,MSGL_WARN, ##args )
74#define MSG_FATAL(args...) mp_msg(MSGT_SWS,MSGL_FATAL, ##args )
75#define MSG_ERR(args...) mp_msg(MSGT_SWS,MSGL_ERR, ##args )
76#define MSG_V(args...) mp_msg(MSGT_SWS,MSGL_V, ##args )
77#define MSG_DBG2(args...) mp_msg(MSGT_SWS,MSGL_DBG2, ##args )
78#define MSG_INFO(args...) mp_msg(MSGT_SWS,MSGL_INFO, ##args )
79
541c4eb9 80#undef MOVNTQ
7d7f78b5 81#undef PAVGB
d3f41512 82
783e9cc9 83//#undef HAVE_MMX2
7f56a527 84//#define HAVE_3DNOW
d3f41512 85//#undef HAVE_MMX
783e9cc9 86//#undef ARCH_X86
2ba1bff0 87//#define WORDS_BIGENDIAN
d604bab9 88#define DITHER1XBPP
d3f41512 89
ac6a2e45
MN
90#define FAST_BGR2YV12 // use 7 bit coeffs instead of 15bit
91
1e621b18 92#define RET 0xC3 //near return opcode for X86
c1b0bfb4 93
28bf81c9 94#ifdef MP_DEBUG
81b7c056 95#define ASSERT(x) assert(x);
28bf81c9 96#else
c1b0bfb4 97#define ASSERT(x) ;
28bf81c9
MN
98#endif
99
100#ifdef M_PI
101#define PI M_PI
102#else
103#define PI 3.14159265358979323846
104#endif
c1b0bfb4 105
6c7506de 106//FIXME replace this with something faster
fccb9b2b 107#define isPlanarYUV(x) ((x)==IMGFMT_YV12 || (x)==IMGFMT_YVU9 \
d80e2fa2 108 || (x)==IMGFMT_444P || (x)==IMGFMT_422P || (x)==IMGFMT_411P)
7322a67c 109#define isYUV(x) ((x)==IMGFMT_UYVY || (x)==IMGFMT_YUY2 || isPlanarYUV(x))
44c1035c 110#define isGray(x) ((x)==IMGFMT_Y800)
cf7d1c1a
MN
111#define isRGB(x) (((x)&IMGFMT_RGB_MASK)==IMGFMT_RGB)
112#define isBGR(x) (((x)&IMGFMT_BGR_MASK)==IMGFMT_BGR)
fccb9b2b 113#define isSupportedIn(x) ((x)==IMGFMT_YV12 || (x)==IMGFMT_YUY2 || (x)==IMGFMT_UYVY\
b72034dd 114 || (x)==IMGFMT_BGR32|| (x)==IMGFMT_BGR24|| (x)==IMGFMT_BGR16|| (x)==IMGFMT_BGR15\
a861d4d7 115 || (x)==IMGFMT_RGB32|| (x)==IMGFMT_RGB24\
d80e2fa2
MN
116 || (x)==IMGFMT_Y800 || (x)==IMGFMT_YVU9\
117 || (x)==IMGFMT_444P || (x)==IMGFMT_422P || (x)==IMGFMT_411P)
fccb9b2b 118#define isSupportedOut(x) ((x)==IMGFMT_YV12 || (x)==IMGFMT_YUY2\
d80e2fa2 119 || (x)==IMGFMT_444P || (x)==IMGFMT_422P || (x)==IMGFMT_411P\
cf7d1c1a 120 || isRGB(x) || isBGR(x)\
e616aa93 121 || (x)==IMGFMT_Y800 || (x)==IMGFMT_YVU9)
7322a67c 122#define isPacked(x) ((x)==IMGFMT_YUY2 || (x)==IMGFMT_UYVY ||isRGB(x) || isBGR(x))
6ff0ad6b
MN
123
124#define RGB2YUV_SHIFT 16
1e621b18
MN
125#define BY ((int)( 0.098*(1<<RGB2YUV_SHIFT)+0.5))
126#define BV ((int)(-0.071*(1<<RGB2YUV_SHIFT)+0.5))
127#define BU ((int)( 0.439*(1<<RGB2YUV_SHIFT)+0.5))
128#define GY ((int)( 0.504*(1<<RGB2YUV_SHIFT)+0.5))
129#define GV ((int)(-0.368*(1<<RGB2YUV_SHIFT)+0.5))
130#define GU ((int)(-0.291*(1<<RGB2YUV_SHIFT)+0.5))
131#define RY ((int)( 0.257*(1<<RGB2YUV_SHIFT)+0.5))
132#define RV ((int)( 0.439*(1<<RGB2YUV_SHIFT)+0.5))
133#define RU ((int)(-0.148*(1<<RGB2YUV_SHIFT)+0.5))
6c7506de 134
e3d2500f 135extern int verbose; // defined in mplayer.c
0481412a
MN
136extern const int32_t Inverse_Table_6_9[8][4];
137
783e9cc9
MN
138/*
139NOTES
d604bab9 140Special versions: fast Y 1:1 scaling (no interpolation in y direction)
31190492 141
783e9cc9 142TODO
d604bab9 143more intelligent missalignment avoidance for the horizontal scaler
c1b0bfb4
MN
144write special vertical cubic upscale version
145Optimize C code (yv12 / minmax)
ff7ba856 146add support for packed pixel yuv input & output
6ff0ad6b
MN
147add support for Y8 output
148optimize bgr24 & bgr32
ff7ba856 149add BGR4 output support
1e621b18 150write special BGR->BGR scaler
783e9cc9 151*/
31190492 152
d604bab9 153#define ABS(a) ((a) > 0 ? (a) : (-(a)))
2ff198c1
MN
154#define MIN(a,b) ((a) > (b) ? (b) : (a))
155#define MAX(a,b) ((a) < (b) ? (b) : (a))
d604bab9 156
7630f2e0
MN
157#ifdef ARCH_X86
158#define CAN_COMPILE_X86_ASM
d604bab9
MN
159#endif
160
7630f2e0 161#ifdef CAN_COMPILE_X86_ASM
f62255fb
MN
162static uint64_t __attribute__((aligned(8))) bF8= 0xF8F8F8F8F8F8F8F8LL;
163static uint64_t __attribute__((aligned(8))) bFC= 0xFCFCFCFCFCFCFCFCLL;
d604bab9 164static uint64_t __attribute__((aligned(8))) w10= 0x0010001000100010LL;
077ea8a7 165static uint64_t __attribute__((aligned(8))) w02= 0x0002000200020002LL;
d604bab9
MN
166static uint64_t __attribute__((aligned(8))) bm00001111=0x00000000FFFFFFFFLL;
167static uint64_t __attribute__((aligned(8))) bm00000111=0x0000000000FFFFFFLL;
168static uint64_t __attribute__((aligned(8))) bm11111000=0xFFFFFFFFFF000000LL;
6ff0ad6b 169static uint64_t __attribute__((aligned(8))) bm01010101=0x00FF00FF00FF00FFLL;
d604bab9 170
3fe8e8f0
MN
171static volatile uint64_t __attribute__((aligned(8))) b5Dither;
172static volatile uint64_t __attribute__((aligned(8))) g5Dither;
173static volatile uint64_t __attribute__((aligned(8))) g6Dither;
174static volatile uint64_t __attribute__((aligned(8))) r5Dither;
d8fa3c54
MN
175
176static uint64_t __attribute__((aligned(8))) dither4[2]={
177 0x0103010301030103LL,
178 0x0200020002000200LL,};
179
180static uint64_t __attribute__((aligned(8))) dither8[2]={
181 0x0602060206020602LL,
182 0x0004000400040004LL,};
d604bab9
MN
183
184static uint64_t __attribute__((aligned(8))) b16Mask= 0x001F001F001F001FLL;
185static uint64_t __attribute__((aligned(8))) g16Mask= 0x07E007E007E007E0LL;
186static uint64_t __attribute__((aligned(8))) r16Mask= 0xF800F800F800F800LL;
187static uint64_t __attribute__((aligned(8))) b15Mask= 0x001F001F001F001FLL;
188static uint64_t __attribute__((aligned(8))) g15Mask= 0x03E003E003E003E0LL;
189static uint64_t __attribute__((aligned(8))) r15Mask= 0x7C007C007C007C00LL;
190
99d2cb72
MN
191static uint64_t __attribute__((aligned(8))) M24A= 0x00FF0000FF0000FFLL;
192static uint64_t __attribute__((aligned(8))) M24B= 0xFF0000FF0000FF00LL;
193static uint64_t __attribute__((aligned(8))) M24C= 0x0000FF0000FF0000LL;
194
ac6a2e45
MN
195#ifdef FAST_BGR2YV12
196static const uint64_t bgr2YCoeff __attribute__((aligned(8))) = 0x000000210041000DULL;
4342fc14
MN
197static const uint64_t bgr2UCoeff __attribute__((aligned(8))) = 0x0000FFEEFFDC0038ULL;
198static const uint64_t bgr2VCoeff __attribute__((aligned(8))) = 0x00000038FFD2FFF8ULL;
ac6a2e45
MN
199#else
200static const uint64_t bgr2YCoeff __attribute__((aligned(8))) = 0x000020E540830C8BULL;
4342fc14
MN
201static const uint64_t bgr2UCoeff __attribute__((aligned(8))) = 0x0000ED0FDAC23831ULL;
202static const uint64_t bgr2VCoeff __attribute__((aligned(8))) = 0x00003831D0E6F6EAULL;
ac6a2e45
MN
203#endif
204static const uint64_t bgr2YOffset __attribute__((aligned(8))) = 0x1010101010101010ULL;
4342fc14 205static const uint64_t bgr2UVOffset __attribute__((aligned(8)))= 0x8080808080808080ULL;
ac6a2e45 206static const uint64_t w1111 __attribute__((aligned(8))) = 0x0001000100010001ULL;
d604bab9 207#endif
783e9cc9
MN
208
209// clipping helper table for C implementations:
210static unsigned char clip_table[768];
211
28bf81c9 212//global sws_flags from the command line
1f347f22 213int sws_flags=2;
077ea8a7 214
5cebb24b
MN
215//global srcFilter
216SwsFilter src_filter= {NULL, NULL, NULL, NULL};
217
218float sws_lum_gblur= 0.0;
219float sws_chr_gblur= 0.0;
220int sws_chr_vshift= 0;
221int sws_chr_hshift= 0;
5521b193
MN
222float sws_chr_sharpen= 0.0;
223float sws_lum_sharpen= 0.0;
5cebb24b 224
28bf81c9
MN
225/* cpuCaps combined from cpudetect and whats actually compiled in
226 (if there is no support for something compiled in it wont appear here) */
227static CpuCaps cpuCaps;
d3f41512 228
d4e24275 229int (*swScale)(SwsContext *context, uint8_t* src[], int srcStride[], int srcSliceY,
28bf81c9 230 int srcSliceH, uint8_t* dst[], int dstStride[])=NULL;
2ff198c1 231
d4e24275
MN
232static SwsVector *sws_getConvVec(SwsVector *a, SwsVector *b);
233
cf7d1c1a
MN
234extern const uint8_t dither_2x2_4[2][8];
235extern const uint8_t dither_2x2_8[2][8];
236extern const uint8_t dither_8x8_32[8][8];
237extern const uint8_t dither_8x8_73[8][8];
238extern const uint8_t dither_8x8_220[8][8];
5cebb24b 239
7630f2e0 240#ifdef CAN_COMPILE_X86_ASM
96034638
MN
241void in_asm_used_var_warning_killer()
242{
20380eb8 243 volatile int i= bF8+bFC+w10+
5ac80202 244 bm00001111+bm00000111+bm11111000+b16Mask+g16Mask+r16Mask+b15Mask+g15Mask+r15Mask+
6ff0ad6b 245 M24A+M24B+M24C+w02 + b5Dither+g5Dither+r5Dither+g6Dither+dither4[0]+dither8[0]+bm01010101;
96034638
MN
246 if(i) i=0;
247}
248#endif
d604bab9 249
5859233b 250static inline void yuv2yuvXinC(int16_t *lumFilter, int16_t **lumSrc, int lumFilterSize,
e3d2500f 251 int16_t *chrFilter, int16_t **chrSrc, int chrFilterSize,
5859233b 252 uint8_t *dest, uint8_t *uDest, uint8_t *vDest, int dstW, int chrDstW)
e3d2500f
MN
253{
254 //FIXME Optimize (just quickly writen not opti..)
255 int i;
5859233b 256 for(i=0; i<dstW; i++)
e3d2500f
MN
257 {
258 int val=0;
259 int j;
260 for(j=0; j<lumFilterSize; j++)
261 val += lumSrc[j][i] * lumFilter[j];
262
263 dest[i]= MIN(MAX(val>>19, 0), 255);
264 }
265
266 if(uDest != NULL)
5859233b 267 for(i=0; i<chrDstW; i++)
e3d2500f
MN
268 {
269 int u=0;
270 int v=0;
271 int j;
627690b5 272 for(j=0; j<chrFilterSize; j++)
e3d2500f
MN
273 {
274 u += chrSrc[j][i] * chrFilter[j];
275 v += chrSrc[j][i + 2048] * chrFilter[j];
276 }
277
278 uDest[i]= MIN(MAX(u>>19, 0), 255);
279 vDest[i]= MIN(MAX(v>>19, 0), 255);
280 }
281}
282
46de8b73 283
25593e29 284#define YSCALE_YUV_2_PACKEDX_C(type) \
cf7d1c1a
MN
285 for(i=0; i<(dstW>>1); i++){\
286 int j;\
287 int Y1=0;\
288 int Y2=0;\
289 int U=0;\
290 int V=0;\
291 type *r, *b, *g;\
292 const int i2= 2*i;\
293 \
294 for(j=0; j<lumFilterSize; j++)\
295 {\
296 Y1 += lumSrc[j][i2] * lumFilter[j];\
297 Y2 += lumSrc[j][i2+1] * lumFilter[j];\
298 }\
299 for(j=0; j<chrFilterSize; j++)\
300 {\
301 U += chrSrc[j][i] * chrFilter[j];\
302 V += chrSrc[j][i+2048] * chrFilter[j];\
303 }\
304 Y1>>=19;\
305 Y2>>=19;\
306 U >>=19;\
307 V >>=19;\
308 if((Y1|Y2|U|V)&256)\
309 {\
310 if(Y1>255) Y1=255;\
311 else if(Y1<0)Y1=0;\
312 if(Y2>255) Y2=255;\
313 else if(Y2<0)Y2=0;\
314 if(U>255) U=255;\
315 else if(U<0) U=0;\
316 if(V>255) V=255;\
317 else if(V<0) V=0;\
46de8b73
MN
318 }
319
320#define YSCALE_YUV_2_RGBX_C(type) \
25593e29 321 YSCALE_YUV_2_PACKEDX_C(type)\
cf7d1c1a
MN
322 r = c->table_rV[V];\
323 g = c->table_gU[U] + c->table_gV[V];\
324 b = c->table_bU[U];\
325
25593e29 326#define YSCALE_YUV_2_PACKED2_C \
cf7d1c1a
MN
327 for(i=0; i<(dstW>>1); i++){\
328 const int i2= 2*i;\
329 int Y1= (buf0[i2 ]*yalpha1+buf1[i2 ]*yalpha)>>19;\
330 int Y2= (buf0[i2+1]*yalpha1+buf1[i2+1]*yalpha)>>19;\
331 int U= (uvbuf0[i ]*uvalpha1+uvbuf1[i ]*uvalpha)>>19;\
332 int V= (uvbuf0[i+2048]*uvalpha1+uvbuf1[i+2048]*uvalpha)>>19;\
46de8b73
MN
333
334#define YSCALE_YUV_2_RGB2_C(type) \
25593e29 335 YSCALE_YUV_2_PACKED2_C\
cf7d1c1a
MN
336 type *r, *b, *g;\
337 r = c->table_rV[V];\
338 g = c->table_gU[U] + c->table_gV[V];\
339 b = c->table_bU[U];\
340
25593e29 341#define YSCALE_YUV_2_PACKED1_C \
cf7d1c1a
MN
342 for(i=0; i<(dstW>>1); i++){\
343 const int i2= 2*i;\
344 int Y1= buf0[i2 ]>>7;\
345 int Y2= buf0[i2+1]>>7;\
346 int U= (uvbuf1[i ])>>7;\
347 int V= (uvbuf1[i+2048])>>7;\
46de8b73
MN
348
349#define YSCALE_YUV_2_RGB1_C(type) \
25593e29 350 YSCALE_YUV_2_PACKED1_C\
cf7d1c1a
MN
351 type *r, *b, *g;\
352 r = c->table_rV[V];\
353 g = c->table_gU[U] + c->table_gV[V];\
354 b = c->table_bU[U];\
355
25593e29 356#define YSCALE_YUV_2_PACKED1B_C \
cf7d1c1a
MN
357 for(i=0; i<(dstW>>1); i++){\
358 const int i2= 2*i;\
359 int Y1= buf0[i2 ]>>7;\
360 int Y2= buf0[i2+1]>>7;\
361 int U= (uvbuf0[i ] + uvbuf1[i ])>>8;\
362 int V= (uvbuf0[i+2048] + uvbuf1[i+2048])>>8;\
46de8b73
MN
363
364#define YSCALE_YUV_2_RGB1B_C(type) \
25593e29 365 YSCALE_YUV_2_PACKED1B_C\
cf7d1c1a
MN
366 type *r, *b, *g;\
367 r = c->table_rV[V];\
368 g = c->table_gU[U] + c->table_gV[V];\
369 b = c->table_bU[U];\
370
46de8b73 371#define YSCALE_YUV_2_ANYRGB_C(func, func2)\
cf7d1c1a
MN
372 switch(c->dstFormat)\
373 {\
374 case IMGFMT_BGR32:\
375 case IMGFMT_RGB32:\
376 func(uint32_t)\
377 ((uint32_t*)dest)[i2+0]= r[Y1] + g[Y1] + b[Y1];\
378 ((uint32_t*)dest)[i2+1]= r[Y2] + g[Y2] + b[Y2];\
379 } \
380 break;\
381 case IMGFMT_RGB24:\
382 func(uint8_t)\
383 ((uint8_t*)dest)[0]= r[Y1];\
384 ((uint8_t*)dest)[1]= g[Y1];\
385 ((uint8_t*)dest)[2]= b[Y1];\
386 ((uint8_t*)dest)[3]= r[Y2];\
387 ((uint8_t*)dest)[4]= g[Y2];\
388 ((uint8_t*)dest)[5]= b[Y2];\
389 ((uint8_t*)dest)+=6;\
390 }\
391 break;\
392 case IMGFMT_BGR24:\
393 func(uint8_t)\
394 ((uint8_t*)dest)[0]= b[Y1];\
395 ((uint8_t*)dest)[1]= g[Y1];\
396 ((uint8_t*)dest)[2]= r[Y1];\
397 ((uint8_t*)dest)[3]= b[Y2];\
398 ((uint8_t*)dest)[4]= g[Y2];\
399 ((uint8_t*)dest)[5]= r[Y2];\
400 ((uint8_t*)dest)+=6;\
401 }\
402 break;\
403 case IMGFMT_RGB16:\
404 case IMGFMT_BGR16:\
405 {\
406 const int dr1= dither_2x2_8[y&1 ][0];\
407 const int dg1= dither_2x2_4[y&1 ][0];\
408 const int db1= dither_2x2_8[(y&1)^1][0];\
409 const int dr2= dither_2x2_8[y&1 ][1];\
410 const int dg2= dither_2x2_4[y&1 ][1];\
411 const int db2= dither_2x2_8[(y&1)^1][1];\
412 func(uint16_t)\
413 ((uint16_t*)dest)[i2+0]= r[Y1+dr1] + g[Y1+dg1] + b[Y1+db1];\
414 ((uint16_t*)dest)[i2+1]= r[Y2+dr2] + g[Y2+dg2] + b[Y2+db2];\
415 }\
416 }\
417 break;\
418 case IMGFMT_RGB15:\
419 case IMGFMT_BGR15:\
420 {\
421 const int dr1= dither_2x2_8[y&1 ][0];\
422 const int dg1= dither_2x2_8[y&1 ][1];\
423 const int db1= dither_2x2_8[(y&1)^1][0];\
424 const int dr2= dither_2x2_8[y&1 ][1];\
425 const int dg2= dither_2x2_8[y&1 ][0];\
426 const int db2= dither_2x2_8[(y&1)^1][1];\
427 func(uint16_t)\
428 ((uint16_t*)dest)[i2+0]= r[Y1+dr1] + g[Y1+dg1] + b[Y1+db1];\
429 ((uint16_t*)dest)[i2+1]= r[Y2+dr2] + g[Y2+dg2] + b[Y2+db2];\
430 }\
431 }\
432 break;\
433 case IMGFMT_RGB8:\
434 case IMGFMT_BGR8:\
435 {\
436 const uint8_t * const d64= dither_8x8_73[y&7];\
437 const uint8_t * const d32= dither_8x8_32[y&7];\
438 func(uint8_t)\
439 ((uint8_t*)dest)[i2+0]= r[Y1+d32[(i2+0)&7]] + g[Y1+d32[(i2+0)&7]] + b[Y1+d64[(i2+0)&7]];\
440 ((uint8_t*)dest)[i2+1]= r[Y2+d32[(i2+1)&7]] + g[Y2+d32[(i2+1)&7]] + b[Y2+d64[(i2+1)&7]];\
441 }\
442 }\
443 break;\
444 case IMGFMT_RGB4:\
445 case IMGFMT_BGR4:\
446 {\
447 const uint8_t * const d64= dither_8x8_73 [y&7];\
448 const uint8_t * const d128=dither_8x8_220[y&7];\
449 func(uint8_t)\
799fd467 450 ((uint8_t*)dest)[i]= r[Y1+d128[(i2+0)&7]] + g[Y1+d64[(i2+0)&7]] + b[Y1+d128[(i2+0)&7]]\
f17457ac
MN
451 + ((r[Y2+d128[(i2+1)&7]] + g[Y2+d64[(i2+1)&7]] + b[Y2+d128[(i2+1)&7]])<<4);\
452 }\
453 }\
454 break;\
455 case IMGFMT_RG4B:\
456 case IMGFMT_BG4B:\
457 {\
458 const uint8_t * const d64= dither_8x8_73 [y&7];\
459 const uint8_t * const d128=dither_8x8_220[y&7];\
460 func(uint8_t)\
cf7d1c1a
MN
461 ((uint8_t*)dest)[i2+0]= r[Y1+d128[(i2+0)&7]] + g[Y1+d64[(i2+0)&7]] + b[Y1+d128[(i2+0)&7]];\
462 ((uint8_t*)dest)[i2+1]= r[Y2+d128[(i2+1)&7]] + g[Y2+d64[(i2+1)&7]] + b[Y2+d128[(i2+1)&7]];\
463 }\
464 }\
465 break;\
466 case IMGFMT_RGB1:\
467 case IMGFMT_BGR1:\
468 {\
469 const uint8_t * const d128=dither_8x8_220[y&7];\
470 uint8_t *g= c->table_gU[128] + c->table_gV[128];\
471 for(i=0; i<dstW-7; i+=8){\
472 int acc;\
473 acc = g[((buf0[i ]*yalpha1+buf1[i ]*yalpha)>>19) + d128[0]];\
474 acc+= acc + g[((buf0[i+1]*yalpha1+buf1[i+1]*yalpha)>>19) + d128[1]];\
475 acc+= acc + g[((buf0[i+2]*yalpha1+buf1[i+2]*yalpha)>>19) + d128[2]];\
476 acc+= acc + g[((buf0[i+3]*yalpha1+buf1[i+3]*yalpha)>>19) + d128[3]];\
477 acc+= acc + g[((buf0[i+4]*yalpha1+buf1[i+4]*yalpha)>>19) + d128[4]];\
478 acc+= acc + g[((buf0[i+5]*yalpha1+buf1[i+5]*yalpha)>>19) + d128[5]];\
479 acc+= acc + g[((buf0[i+6]*yalpha1+buf1[i+6]*yalpha)>>19) + d128[6]];\
480 acc+= acc + g[((buf0[i+7]*yalpha1+buf1[i+7]*yalpha)>>19) + d128[7]];\
481 ((uint8_t*)dest)[0]= acc;\
482 ((uint8_t*)dest)++;\
483 }\
484\
485/*\
486((uint8_t*)dest)-= dstW>>4;\
487{\
488 int acc=0;\
489 int left=0;\
490 static int top[1024];\
491 static int last_new[1024][1024];\
492 static int last_in3[1024][1024];\
493 static int drift[1024][1024];\
494 int topLeft=0;\
495 int shift=0;\
496 int count=0;\
497 const uint8_t * const d128=dither_8x8_220[y&7];\
498 int error_new=0;\
499 int error_in3=0;\
500 int f=0;\
501 \
502 for(i=dstW>>1; i<dstW; i++){\
503 int in= ((buf0[i ]*yalpha1+buf1[i ]*yalpha)>>19);\
504 int in2 = (76309 * (in - 16) + 32768) >> 16;\
505 int in3 = (in2 < 0) ? 0 : ((in2 > 255) ? 255 : in2);\
506 int old= (left*7 + topLeft + top[i]*5 + top[i+1]*3)/20 + in3\
507 + (last_new[y][i] - in3)*f/256;\
508 int new= old> 128 ? 255 : 0;\
509\
510 error_new+= ABS(last_new[y][i] - new);\
511 error_in3+= ABS(last_in3[y][i] - in3);\
512 f= error_new - error_in3*4;\
513 if(f<0) f=0;\
514 if(f>256) f=256;\
515\
516 topLeft= top[i];\
517 left= top[i]= old - new;\
518 last_new[y][i]= new;\
519 last_in3[y][i]= in3;\
520\
521 acc+= acc + (new&1);\
522 if((i&7)==6){\
523 ((uint8_t*)dest)[0]= acc;\
524 ((uint8_t*)dest)++;\
525 }\
526 }\
527}\
528*/\
529 }\
530 break;\
46de8b73
MN
531 case IMGFMT_YUY2:\
532 func2\
533 ((uint8_t*)dest)[2*i2+0]= Y1;\
534 ((uint8_t*)dest)[2*i2+1]= U;\
535 ((uint8_t*)dest)[2*i2+2]= Y2;\
536 ((uint8_t*)dest)[2*i2+3]= V;\
537 } \
538 break;\
cf7d1c1a
MN
539 }\
540
541
25593e29 542static inline void yuv2packedXinC(SwsContext *c, int16_t *lumFilter, int16_t **lumSrc, int lumFilterSize,
e3d2500f 543 int16_t *chrFilter, int16_t **chrSrc, int chrFilterSize,
cf7d1c1a 544 uint8_t *dest, int dstW, int y)
e3d2500f 545{
cf7d1c1a
MN
546 int i;
547 switch(c->dstFormat)
e3d2500f 548 {
cf7d1c1a
MN
549 case IMGFMT_RGB32:
550 case IMGFMT_BGR32:
551 YSCALE_YUV_2_RGBX_C(uint32_t)
552 ((uint32_t*)dest)[i2+0]= r[Y1] + g[Y1] + b[Y1];
553 ((uint32_t*)dest)[i2+1]= r[Y2] + g[Y2] + b[Y2];
e3d2500f 554 }
cf7d1c1a
MN
555 break;
556 case IMGFMT_RGB24:
557 YSCALE_YUV_2_RGBX_C(uint8_t)
558 ((uint8_t*)dest)[0]= r[Y1];
559 ((uint8_t*)dest)[1]= g[Y1];
560 ((uint8_t*)dest)[2]= b[Y1];
561 ((uint8_t*)dest)[3]= r[Y2];
562 ((uint8_t*)dest)[4]= g[Y2];
563 ((uint8_t*)dest)[5]= b[Y2];
564 ((uint8_t*)dest)+=6;
565 }
566 break;
567 case IMGFMT_BGR24:
568 YSCALE_YUV_2_RGBX_C(uint8_t)
569 ((uint8_t*)dest)[0]= b[Y1];
570 ((uint8_t*)dest)[1]= g[Y1];
571 ((uint8_t*)dest)[2]= r[Y1];
572 ((uint8_t*)dest)[3]= b[Y2];
573 ((uint8_t*)dest)[4]= g[Y2];
574 ((uint8_t*)dest)[5]= r[Y2];
575 ((uint8_t*)dest)+=6;
576 }
577 break;
578 case IMGFMT_RGB16:
579 case IMGFMT_BGR16:
580 {
581 const int dr1= dither_2x2_8[y&1 ][0];
582 const int dg1= dither_2x2_4[y&1 ][0];
583 const int db1= dither_2x2_8[(y&1)^1][0];
584 const int dr2= dither_2x2_8[y&1 ][1];
585 const int dg2= dither_2x2_4[y&1 ][1];
586 const int db2= dither_2x2_8[(y&1)^1][1];
587 YSCALE_YUV_2_RGBX_C(uint16_t)
588 ((uint16_t*)dest)[i2+0]= r[Y1+dr1] + g[Y1+dg1] + b[Y1+db1];
589 ((uint16_t*)dest)[i2+1]= r[Y2+dr2] + g[Y2+dg2] + b[Y2+db2];
e3d2500f 590 }
e3d2500f 591 }
cf7d1c1a
MN
592 break;
593 case IMGFMT_RGB15:
594 case IMGFMT_BGR15:
595 {
596 const int dr1= dither_2x2_8[y&1 ][0];
597 const int dg1= dither_2x2_8[y&1 ][1];
598 const int db1= dither_2x2_8[(y&1)^1][0];
599 const int dr2= dither_2x2_8[y&1 ][1];
600 const int dg2= dither_2x2_8[y&1 ][0];
601 const int db2= dither_2x2_8[(y&1)^1][1];
602 YSCALE_YUV_2_RGBX_C(uint16_t)
603 ((uint16_t*)dest)[i2+0]= r[Y1+dr1] + g[Y1+dg1] + b[Y1+db1];
604 ((uint16_t*)dest)[i2+1]= r[Y2+dr2] + g[Y2+dg2] + b[Y2+db2];
e3d2500f 605 }
cf7d1c1a
MN
606 }
607 break;
608 case IMGFMT_RGB8:
609 case IMGFMT_BGR8:
610 {
611 const uint8_t * const d64= dither_8x8_73[y&7];
612 const uint8_t * const d32= dither_8x8_32[y&7];
613 YSCALE_YUV_2_RGBX_C(uint8_t)
614 ((uint8_t*)dest)[i2+0]= r[Y1+d32[(i2+0)&7]] + g[Y1+d32[(i2+0)&7]] + b[Y1+d64[(i2+0)&7]];
615 ((uint8_t*)dest)[i2+1]= r[Y2+d32[(i2+1)&7]] + g[Y2+d32[(i2+1)&7]] + b[Y2+d64[(i2+1)&7]];
e3d2500f 616 }
e3d2500f 617 }
cf7d1c1a
MN
618 break;
619 case IMGFMT_RGB4:
620 case IMGFMT_BGR4:
621 {
622 const uint8_t * const d64= dither_8x8_73 [y&7];
623 const uint8_t * const d128=dither_8x8_220[y&7];
624 YSCALE_YUV_2_RGBX_C(uint8_t)
799fd467 625 ((uint8_t*)dest)[i]= r[Y1+d128[(i2+0)&7]] + g[Y1+d64[(i2+0)&7]] + b[Y1+d128[(i2+0)&7]]
f17457ac
MN
626 +((r[Y2+d128[(i2+1)&7]] + g[Y2+d64[(i2+1)&7]] + b[Y2+d128[(i2+1)&7]])<<4);
627 }
628 }
629 break;
630 case IMGFMT_RG4B:
631 case IMGFMT_BG4B:
632 {
633 const uint8_t * const d64= dither_8x8_73 [y&7];
634 const uint8_t * const d128=dither_8x8_220[y&7];
635 YSCALE_YUV_2_RGBX_C(uint8_t)
cf7d1c1a
MN
636 ((uint8_t*)dest)[i2+0]= r[Y1+d128[(i2+0)&7]] + g[Y1+d64[(i2+0)&7]] + b[Y1+d128[(i2+0)&7]];
637 ((uint8_t*)dest)[i2+1]= r[Y2+d128[(i2+1)&7]] + g[Y2+d64[(i2+1)&7]] + b[Y2+d128[(i2+1)&7]];
e3d2500f 638 }
cf7d1c1a
MN
639 }
640 break;
641 case IMGFMT_RGB1:
642 case IMGFMT_BGR1:
643 {
644 const uint8_t * const d128=dither_8x8_220[y&7];
645 uint8_t *g= c->table_gU[128] + c->table_gV[128];
646 int acc=0;
647 for(i=0; i<dstW-1; i+=2){
648 int j;
649 int Y1=0;
650 int Y2=0;
651
652 for(j=0; j<lumFilterSize; j++)
653 {
654 Y1 += lumSrc[j][i] * lumFilter[j];
655 Y2 += lumSrc[j][i+1] * lumFilter[j];
656 }
657 Y1>>=19;
658 Y2>>=19;
659 if((Y1|Y2)&256)
660 {
661 if(Y1>255) Y1=255;
662 else if(Y1<0)Y1=0;
663 if(Y2>255) Y2=255;
664 else if(Y2<0)Y2=0;
665 }
666 acc+= acc + g[Y1+d128[(i+0)&7]];
667 acc+= acc + g[Y2+d128[(i+1)&7]];
668 if((i&7)==6){
669 ((uint8_t*)dest)[0]= acc;
670 ((uint8_t*)dest)++;
671 }
e3d2500f 672 }
e3d2500f 673 }
cf7d1c1a 674 break;
46de8b73 675 case IMGFMT_YUY2:
25593e29 676 YSCALE_YUV_2_PACKEDX_C(void)
46de8b73
MN
677 ((uint8_t*)dest)[2*i2+0]= Y1;
678 ((uint8_t*)dest)[2*i2+1]= U;
679 ((uint8_t*)dest)[2*i2+2]= Y2;
680 ((uint8_t*)dest)[2*i2+3]= V;
681 }
682 break;
e3d2500f
MN
683 }
684}
685
686
7630f2e0
MN
687//Note: we have C, X86, MMX, MMX2, 3DNOW version therse no 3DNOW+MMX2 one
688//Plain C versions
726a959a
MN
689#if !defined (HAVE_MMX) || defined (RUNTIME_CPUDETECT)
690#define COMPILE_C
691#endif
692
693#ifdef CAN_COMPILE_X86_ASM
694
695#if (defined (HAVE_MMX) && !defined (HAVE_3DNOW) && !defined (HAVE_MMX2)) || defined (RUNTIME_CPUDETECT)
696#define COMPILE_MMX
697#endif
698
699#if defined (HAVE_MMX2) || defined (RUNTIME_CPUDETECT)
700#define COMPILE_MMX2
701#endif
702
703#if (defined (HAVE_3DNOW) && !defined (HAVE_MMX2)) || defined (RUNTIME_CPUDETECT)
704#define COMPILE_3DNOW
705#endif
706#endif //CAN_COMPILE_X86_ASM
707
708#undef HAVE_MMX
709#undef HAVE_MMX2
710#undef HAVE_3DNOW
726a959a
MN
711
712#ifdef COMPILE_C
7630f2e0
MN
713#undef HAVE_MMX
714#undef HAVE_MMX2
715#undef HAVE_3DNOW
7630f2e0
MN
716#define RENAME(a) a ## _C
717#include "swscale_template.c"
726a959a 718#endif
397c035e 719
7630f2e0 720#ifdef CAN_COMPILE_X86_ASM
397c035e 721
7630f2e0
MN
722//X86 versions
723/*
724#undef RENAME
725#undef HAVE_MMX
726#undef HAVE_MMX2
727#undef HAVE_3DNOW
728#define ARCH_X86
729#define RENAME(a) a ## _X86
730#include "swscale_template.c"
1faf0867 731*/
7630f2e0 732//MMX versions
726a959a 733#ifdef COMPILE_MMX
7630f2e0
MN
734#undef RENAME
735#define HAVE_MMX
736#undef HAVE_MMX2
737#undef HAVE_3DNOW
7630f2e0
MN
738#define RENAME(a) a ## _MMX
739#include "swscale_template.c"
726a959a 740#endif
7630f2e0
MN
741
742//MMX2 versions
726a959a 743#ifdef COMPILE_MMX2
7630f2e0
MN
744#undef RENAME
745#define HAVE_MMX
746#define HAVE_MMX2
747#undef HAVE_3DNOW
7630f2e0
MN
748#define RENAME(a) a ## _MMX2
749#include "swscale_template.c"
726a959a 750#endif
7630f2e0
MN
751
752//3DNOW versions
726a959a 753#ifdef COMPILE_3DNOW
7630f2e0
MN
754#undef RENAME
755#define HAVE_MMX
756#undef HAVE_MMX2
757#define HAVE_3DNOW
7630f2e0
MN
758#define RENAME(a) a ## _3DNow
759#include "swscale_template.c"
726a959a 760#endif
7630f2e0
MN
761
762#endif //CAN_COMPILE_X86_ASM
763
764// minor note: the HAVE_xyz is messed up after that line so dont use it
d604bab9 765
d3f41512 766
6c7506de 767// old global scaler, dont use for new code
28bf81c9
MN
768// will use sws_flags from the command line
769void SwScale_YV12slice(unsigned char* src[], int srcStride[], int srcSliceY ,
770 int srcSliceH, uint8_t* dst[], int dstStride, int dstbpp,
d1fac6cf 771 int srcW, int srcH, int dstW, int dstH){
31190492 772
28bf81c9
MN
773 static SwsContext *context=NULL;
774 int dstFormat;
28bf81c9
MN
775 int dstStride3[3]= {dstStride, dstStride>>1, dstStride>>1};
776
6c7506de 777 switch(dstbpp)
28bf81c9 778 {
6c7506de
MN
779 case 8 : dstFormat= IMGFMT_Y8; break;
780 case 12: dstFormat= IMGFMT_YV12; break;
781 case 15: dstFormat= IMGFMT_BGR15; break;
782 case 16: dstFormat= IMGFMT_BGR16; break;
783 case 24: dstFormat= IMGFMT_BGR24; break;
784 case 32: dstFormat= IMGFMT_BGR32; break;
785 default: return;
786 }
787
d4e24275 788 if(!context) context=sws_getContextFromCmdLine(srcW, srcH, IMGFMT_YV12, dstW, dstH, dstFormat);
6c7506de 789
b6654a54 790 context->swScale(context, src, srcStride, srcSliceY, srcSliceH, dst, dstStride3);
6c7506de
MN
791}
792
d4e24275 793void sws_getFlagsAndFilterFromCmdLine(int *flags, SwsFilter **srcFilterParam, SwsFilter **dstFilterParam)
6c7506de 794{
6c7506de 795 static int firstTime=1;
5859233b 796 *flags=0;
6c7506de 797
5521b193 798#ifdef ARCH_X86
6c7506de
MN
799 if(gCpuCaps.hasMMX)
800 asm volatile("emms\n\t"::: "memory"); //FIXME this shouldnt be required but it IS (even for non mmx versions)
5521b193 801#endif
6c7506de
MN
802 if(firstTime)
803 {
28bf81c9 804 firstTime=0;
5859233b 805 *flags= SWS_PRINT_INFO;
6c7506de 806 }
5859233b 807 else if(verbose>1) *flags= SWS_PRINT_INFO;
6c7506de 808
d4e24275
MN
809 if(src_filter.lumH) sws_freeVec(src_filter.lumH);
810 if(src_filter.lumV) sws_freeVec(src_filter.lumV);
811 if(src_filter.chrH) sws_freeVec(src_filter.chrH);
812 if(src_filter.chrV) sws_freeVec(src_filter.chrV);
6c7506de
MN
813
814 if(sws_lum_gblur!=0.0){
d4e24275
MN
815 src_filter.lumH= sws_getGaussianVec(sws_lum_gblur, 3.0);
816 src_filter.lumV= sws_getGaussianVec(sws_lum_gblur, 3.0);
6c7506de 817 }else{
d4e24275
MN
818 src_filter.lumH= sws_getIdentityVec();
819 src_filter.lumV= sws_getIdentityVec();
6c7506de 820 }
c7f822d9 821
6c7506de 822 if(sws_chr_gblur!=0.0){
d4e24275
MN
823 src_filter.chrH= sws_getGaussianVec(sws_chr_gblur, 3.0);
824 src_filter.chrV= sws_getGaussianVec(sws_chr_gblur, 3.0);
6c7506de 825 }else{
d4e24275
MN
826 src_filter.chrH= sws_getIdentityVec();
827 src_filter.chrV= sws_getIdentityVec();
6c7506de 828 }
5521b193 829
6c7506de 830 if(sws_chr_sharpen!=0.0){
d4e24275
MN
831 SwsVector *g= sws_getConstVec(-1.0, 3);
832 SwsVector *id= sws_getConstVec(10.0/sws_chr_sharpen, 1);
6c7506de 833 g->coeff[1]=2.0;
d4e24275
MN
834 sws_addVec(id, g);
835 sws_convVec(src_filter.chrH, id);
836 sws_convVec(src_filter.chrV, id);
837 sws_freeVec(g);
838 sws_freeVec(id);
6c7506de 839 }
5521b193 840
6c7506de 841 if(sws_lum_sharpen!=0.0){
d4e24275
MN
842 SwsVector *g= sws_getConstVec(-1.0, 3);
843 SwsVector *id= sws_getConstVec(10.0/sws_lum_sharpen, 1);
6c7506de 844 g->coeff[1]=2.0;
d4e24275
MN
845 sws_addVec(id, g);
846 sws_convVec(src_filter.lumH, id);
847 sws_convVec(src_filter.lumV, id);
848 sws_freeVec(g);
849 sws_freeVec(id);
6c7506de 850 }
c7f822d9 851
6c7506de 852 if(sws_chr_hshift)
d4e24275 853 sws_shiftVec(src_filter.chrH, sws_chr_hshift);
c7f822d9 854
6c7506de 855 if(sws_chr_vshift)
d4e24275 856 sws_shiftVec(src_filter.chrV, sws_chr_vshift);
5521b193 857
d4e24275
MN
858 sws_normalizeVec(src_filter.chrH, 1.0);
859 sws_normalizeVec(src_filter.chrV, 1.0);
860 sws_normalizeVec(src_filter.lumH, 1.0);
861 sws_normalizeVec(src_filter.lumV, 1.0);
28bf81c9 862
d4e24275
MN
863 if(verbose > 1) sws_printVec(src_filter.chrH);
864 if(verbose > 1) sws_printVec(src_filter.lumH);
28bf81c9
MN
865
866 switch(sws_flags)
867 {
5859233b
MN
868 case 0: *flags|= SWS_FAST_BILINEAR; break;
869 case 1: *flags|= SWS_BILINEAR; break;
870 case 2: *flags|= SWS_BICUBIC; break;
871 case 3: *flags|= SWS_X; break;
872 case 4: *flags|= SWS_POINT; break;
873 case 5: *flags|= SWS_AREA; break;
81a571a8 874 case 6: *flags|= SWS_BICUBLIN; break;
a86c461c
MN
875 case 7: *flags|= SWS_GAUSS; break;
876 case 8: *flags|= SWS_SINC; break;
877 case 9: *flags|= SWS_LANCZOS; break;
878 case 10:*flags|= SWS_SPLINE; break;
5859233b 879 default:*flags|= SWS_BILINEAR; break;
28bf81c9 880 }
5859233b
MN
881
882 *srcFilterParam= &src_filter;
883 *dstFilterParam= NULL;
884}
885
886// will use sws_flags & src_filter (from cmd line)
d4e24275 887SwsContext *sws_getContextFromCmdLine(int srcW, int srcH, int srcFormat, int dstW, int dstH, int dstFormat)
5859233b
MN
888{
889 int flags;
890 SwsFilter *dstFilterParam, *srcFilterParam;
d4e24275 891 sws_getFlagsAndFilterFromCmdLine(&flags, &srcFilterParam, &dstFilterParam);
28bf81c9 892
d4e24275 893 return sws_getContext(srcW, srcH, srcFormat, dstW, dstH, dstFormat, flags, srcFilterParam, dstFilterParam);
28bf81c9
MN
894}
895
a86c461c
MN
896static double getSplineCoeff(double a, double b, double c, double d, double dist)
897{
898// printf("%f %f %f %f %f\n", a,b,c,d,dist);
899 if(dist<=1.0) return ((d*dist + c)*dist + b)*dist +a;
900 else return getSplineCoeff( 0.0,
901 b+ 2.0*c + 3.0*d,
902 c + 3.0*d,
903 -b- 3.0*c - 6.0*d,
904 dist-1.0);
905}
6c7506de 906
c7f822d9
MN
907static inline void initFilter(int16_t **outFilter, int16_t **filterPos, int *outFilterSize, int xInc,
908 int srcW, int dstW, int filterAlign, int one, int flags,
909 SwsVector *srcFilter, SwsVector *dstFilter)
28bf81c9
MN
910{
911 int i;
c7f822d9
MN
912 int filterSize;
913 int filter2Size;
914 int minFilterSize;
915 double *filter=NULL;
916 double *filter2=NULL;
28bf81c9
MN
917#ifdef ARCH_X86
918 if(gCpuCaps.hasMMX)
919 asm volatile("emms\n\t"::: "memory"); //FIXME this shouldnt be required but it IS (even for non mmx versions)
726a959a 920#endif
31190492 921
adeaecb9 922 // Note the +1 is for the MMXscaler which reads over the end
6c7506de 923 *filterPos = (int16_t*)memalign(8, (dstW+1)*sizeof(int16_t));
6c7506de 924
28bf81c9
MN
925 if(ABS(xInc - 0x10000) <10) // unscaled
926 {
927 int i;
c7f822d9
MN
928 filterSize= 1;
929 filter= (double*)memalign(8, dstW*sizeof(double)*filterSize);
930 for(i=0; i<dstW*filterSize; i++) filter[i]=0;
28bf81c9
MN
931
932 for(i=0; i<dstW; i++)
933 {
c7f822d9
MN
934 filter[i*filterSize]=1;
935 (*filterPos)[i]=i;
28bf81c9
MN
936 }
937
938 }
ff7ba856
MN
939 else if(flags&SWS_POINT) // lame looking point sampling mode
940 {
941 int i;
942 int xDstInSrc;
943 filterSize= 1;
944 filter= (double*)memalign(8, dstW*sizeof(double)*filterSize);
945
946 xDstInSrc= xInc/2 - 0x8000;
947 for(i=0; i<dstW; i++)
948 {
8a01d20c 949 int xx= (xDstInSrc - ((filterSize-1)<<15) + (1<<15))>>16;
ff7ba856
MN
950
951 (*filterPos)[i]= xx;
952 filter[i]= 1.0;
953 xDstInSrc+= xInc;
954 }
955 }
a86c461c 956 else if((xInc <= (1<<16) && (flags&SWS_AREA)) || (flags&SWS_FAST_BILINEAR)) // bilinear upscale
28bf81c9
MN
957 {
958 int i;
959 int xDstInSrc;
c7f822d9
MN
960 if (flags&SWS_BICUBIC) filterSize= 4;
961 else if(flags&SWS_X ) filterSize= 4;
d8863d37 962 else filterSize= 2; // SWS_BILINEAR / SWS_AREA
c7f822d9 963 filter= (double*)memalign(8, dstW*sizeof(double)*filterSize);
28bf81c9
MN
964
965 xDstInSrc= xInc/2 - 0x8000;
966 for(i=0; i<dstW; i++)
967 {
8a01d20c 968 int xx= (xDstInSrc - ((filterSize-1)<<15) + (1<<15))>>16;
28bf81c9
MN
969 int j;
970
c7f822d9 971 (*filterPos)[i]= xx;
d8863d37 972 //Bilinear upscale / linear interpolate / Area averaging
c7f822d9 973 for(j=0; j<filterSize; j++)
28bf81c9
MN
974 {
975 double d= ABS((xx<<16) - xDstInSrc)/(double)(1<<16);
976 double coeff= 1.0 - d;
977 if(coeff<0) coeff=0;
c7f822d9 978 filter[i*filterSize + j]= coeff;
28bf81c9
MN
979 xx++;
980 }
28bf81c9
MN
981 xDstInSrc+= xInc;
982 }
983 }
a86c461c 984 else
28bf81c9 985 {
a86c461c
MN
986 double xDstInSrc;
987 double sizeFactor, filterSizeInSrc;
988 const double xInc1= (double)xInc / (double)(1<<16);
989 int param= (flags&SWS_PARAM_MASK)>>SWS_PARAM_SHIFT;
990
991 if (flags&SWS_BICUBIC) sizeFactor= 4.0;
992 else if(flags&SWS_X) sizeFactor= 8.0;
993 else if(flags&SWS_AREA) sizeFactor= 1.0; //downscale only, for upscale it is bilinear
994 else if(flags&SWS_GAUSS) sizeFactor= 8.0; // infinite ;)
995 else if(flags&SWS_LANCZOS) sizeFactor= param ? 2.0*param : 6.0;
93768378 996 else if(flags&SWS_SINC) sizeFactor= 20.0; // infinite ;)
a86c461c
MN
997 else if(flags&SWS_SPLINE) sizeFactor= 20.0; // infinite ;)
998 else if(flags&SWS_BILINEAR) sizeFactor= 2.0;
93768378
MN
999 else {
1000 sizeFactor= 0.0; //GCC warning killer
1001 ASSERT(0)
1002 }
a86c461c
MN
1003
1004 if(xInc1 <= 1.0) filterSizeInSrc= sizeFactor; // upscale
1005 else filterSizeInSrc= sizeFactor*srcW / (double)dstW;
81b7c056 1006
a86c461c
MN
1007 filterSize= (int)ceil(1 + filterSizeInSrc); // will be reduced later if possible
1008 if(filterSize > srcW-2) filterSize=srcW-2;
28bf81c9 1009
a86c461c
MN
1010 filter= (double*)memalign(16, dstW*sizeof(double)*filterSize);
1011
1012 xDstInSrc= xInc1 / 2.0 - 0.5;
28bf81c9
MN
1013 for(i=0; i<dstW; i++)
1014 {
a86c461c 1015 int xx= (int)(xDstInSrc - (filterSize-1)*0.5 + 0.5);
28bf81c9 1016 int j;
c7f822d9
MN
1017 (*filterPos)[i]= xx;
1018 for(j=0; j<filterSize; j++)
28bf81c9 1019 {
a86c461c 1020 double d= ABS(xx - xDstInSrc)/filterSizeInSrc*sizeFactor;
28bf81c9 1021 double coeff;
a86c461c 1022 if(flags & SWS_BICUBIC)
28bf81c9 1023 {
a86c461c
MN
1024 double A= param ? -param*0.01 : -0.60;
1025
28bf81c9
MN
1026 // Equation is from VirtualDub
1027 if(d<1.0)
1028 coeff = (1.0 - (A+3.0)*d*d + (A+2.0)*d*d*d);
1029 else if(d<2.0)
1030 coeff = (-4.0*A + 8.0*A*d - 5.0*A*d*d + A*d*d*d);
1031 else
1032 coeff=0.0;
1033 }
a86c461c
MN
1034/* else if(flags & SWS_X)
1035 {
1036 double p= param ? param*0.01 : 0.3;
1037 coeff = d ? sin(d*PI)/(d*PI) : 1.0;
1038 coeff*= pow(2.0, - p*d*d);
1039 }*/
1040 else if(flags & SWS_X)
1041 {
1042 double A= param ? param*0.1 : 1.0;
1043
1044 if(d<1.0)
1045 coeff = cos(d*PI);
1046 else
1047 coeff=-1.0;
1048 if(coeff<0.0) coeff= -pow(-coeff, A);
1049 else coeff= pow( coeff, A);
1050 coeff= coeff*0.5 + 0.5;
1051 }
d8863d37 1052 else if(flags & SWS_AREA)
28bf81c9 1053 {
a86c461c 1054 double srcPixelSize= 1.0/xInc1;
d8863d37
MN
1055 if(d + srcPixelSize/2 < 0.5) coeff= 1.0;
1056 else if(d - srcPixelSize/2 < 0.5) coeff= (0.5-d)/srcPixelSize + 0.5;
1057 else coeff=0.0;
1058 }
a86c461c
MN
1059 else if(flags & SWS_GAUSS)
1060 {
1061 double p= param ? param*0.1 : 3.0;
1062 coeff = pow(2.0, - p*d*d);
1063 }
1064 else if(flags & SWS_SINC)
1065 {
1066 coeff = d ? sin(d*PI)/(d*PI) : 1.0;
1067 }
1068 else if(flags & SWS_LANCZOS)
1069 {
1070 double p= param ? param : 3.0;
1071 coeff = d ? sin(d*PI)*sin(d*PI/p)/(d*d*PI*PI/p) : 1.0;
1072 if(d>p) coeff=0;
1073 }
1074 else if(flags & SWS_BILINEAR)
28bf81c9
MN
1075 {
1076 coeff= 1.0 - d;
1077 if(coeff<0) coeff=0;
1078 }
a86c461c
MN
1079 else if(flags & SWS_SPLINE)
1080 {
1081 double p=-2.196152422706632;
1082 coeff = getSplineCoeff(1.0, 0.0, p, -p-1.0, d);
1083 }
93768378
MN
1084 else {
1085 coeff= 0.0; //GCC warning killer
1086 ASSERT(0)
1087 }
a86c461c 1088
c7f822d9 1089 filter[i*filterSize + j]= coeff;
28bf81c9
MN
1090 xx++;
1091 }
a86c461c 1092 xDstInSrc+= xInc1;
28bf81c9
MN
1093 }
1094 }
1095
c7f822d9
MN
1096 /* apply src & dst Filter to filter -> filter2
1097 free(filter);
1098 */
81b7c056 1099 ASSERT(filterSize>0)
c7f822d9
MN
1100 filter2Size= filterSize;
1101 if(srcFilter) filter2Size+= srcFilter->length - 1;
1102 if(dstFilter) filter2Size+= dstFilter->length - 1;
81b7c056 1103 ASSERT(filter2Size>0)
c7f822d9
MN
1104 filter2= (double*)memalign(8, filter2Size*dstW*sizeof(double));
1105
1106 for(i=0; i<dstW; i++)
1107 {
1108 int j;
1109 SwsVector scaleFilter;
1110 SwsVector *outVec;
1111
1112 scaleFilter.coeff= filter + i*filterSize;
1113 scaleFilter.length= filterSize;
1114
d4e24275 1115 if(srcFilter) outVec= sws_getConvVec(srcFilter, &scaleFilter);
c7f822d9
MN
1116 else outVec= &scaleFilter;
1117
1118 ASSERT(outVec->length == filter2Size)
1119 //FIXME dstFilter
1120
1121 for(j=0; j<outVec->length; j++)
1122 {
1123 filter2[i*filter2Size + j]= outVec->coeff[j];
1124 }
1125
1126 (*filterPos)[i]+= (filterSize-1)/2 - (filter2Size-1)/2;
1127
d4e24275 1128 if(outVec != &scaleFilter) sws_freeVec(outVec);
c7f822d9
MN
1129 }
1130 free(filter); filter=NULL;
1131
1132 /* try to reduce the filter-size (step1 find size and shift left) */
1133 // Assume its near normalized (*0.5 or *2.0 is ok but * 0.001 is not)
1134 minFilterSize= 0;
1135 for(i=dstW-1; i>=0; i--)
1136 {
1137 int min= filter2Size;
1138 int j;
1139 double cutOff=0.0;
1140
1141 /* get rid off near zero elements on the left by shifting left */
1142 for(j=0; j<filter2Size; j++)
1143 {
1144 int k;
1145 cutOff += ABS(filter2[i*filter2Size]);
1146
1147 if(cutOff > SWS_MAX_REDUCE_CUTOFF) break;
1148
1149 /* preserve Monotonicity because the core cant handle the filter otherwise */
1150 if(i<dstW-1 && (*filterPos)[i] >= (*filterPos)[i+1]) break;
1151
1152 // Move filter coeffs left
1153 for(k=1; k<filter2Size; k++)
1154 filter2[i*filter2Size + k - 1]= filter2[i*filter2Size + k];
1155 filter2[i*filter2Size + k - 1]= 0.0;
1156 (*filterPos)[i]++;
1157 }
1158
1159 cutOff=0.0;
1160 /* count near zeros on the right */
1161 for(j=filter2Size-1; j>0; j--)
1162 {
1163 cutOff += ABS(filter2[i*filter2Size + j]);
1164
1165 if(cutOff > SWS_MAX_REDUCE_CUTOFF) break;
1166 min--;
1167 }
1168
1169 if(min>minFilterSize) minFilterSize= min;
1170 }
1171
81b7c056 1172 ASSERT(minFilterSize > 0)
6c7506de 1173 filterSize= (minFilterSize +(filterAlign-1)) & (~(filterAlign-1));
81b7c056 1174 ASSERT(filterSize > 0)
6c7506de
MN
1175 filter= (double*)memalign(8, filterSize*dstW*sizeof(double));
1176 *outFilterSize= filterSize;
1177
4a53a912 1178 if(flags&SWS_PRINT_INFO)
0d9f3d85 1179 MSG_INFO("SwScaler: reducing / aligning filtersize %d -> %d\n", filter2Size, filterSize);
c7f822d9
MN
1180 /* try to reduce the filter-size (step2 reduce it) */
1181 for(i=0; i<dstW; i++)
1182 {
1183 int j;
1184
6c7506de
MN
1185 for(j=0; j<filterSize; j++)
1186 {
1187 if(j>=filter2Size) filter[i*filterSize + j]= 0.0;
1188 else filter[i*filterSize + j]= filter2[i*filter2Size + j];
1189 }
c7f822d9 1190 }
6c7506de
MN
1191 free(filter2); filter2=NULL;
1192
c7f822d9
MN
1193
1194 //FIXME try to align filterpos if possible
1195
28bf81c9
MN
1196 //fix borders
1197 for(i=0; i<dstW; i++)
1198 {
1199 int j;
c7f822d9 1200 if((*filterPos)[i] < 0)
28bf81c9
MN
1201 {
1202 // Move filter coeffs left to compensate for filterPos
6c7506de 1203 for(j=1; j<filterSize; j++)
28bf81c9 1204 {
c7f822d9 1205 int left= MAX(j + (*filterPos)[i], 0);
6c7506de
MN
1206 filter[i*filterSize + left] += filter[i*filterSize + j];
1207 filter[i*filterSize + j]=0;
28bf81c9 1208 }
c7f822d9 1209 (*filterPos)[i]= 0;
28bf81c9
MN
1210 }
1211
6c7506de 1212 if((*filterPos)[i] + filterSize > srcW)
28bf81c9 1213 {
6c7506de 1214 int shift= (*filterPos)[i] + filterSize - srcW;
28bf81c9 1215 // Move filter coeffs right to compensate for filterPos
6c7506de 1216 for(j=filterSize-2; j>=0; j--)
28bf81c9 1217 {
6c7506de
MN
1218 int right= MIN(j + shift, filterSize-1);
1219 filter[i*filterSize +right] += filter[i*filterSize +j];
1220 filter[i*filterSize +j]=0;
28bf81c9 1221 }
6c7506de 1222 (*filterPos)[i]= srcW - filterSize;
28bf81c9
MN
1223 }
1224 }
1225
6c7506de
MN
1226 // Note the +1 is for the MMXscaler which reads over the end
1227 *outFilter= (int16_t*)memalign(8, *outFilterSize*(dstW+1)*sizeof(int16_t));
1228 memset(*outFilter, 0, *outFilterSize*(dstW+1)*sizeof(int16_t));
c7f822d9
MN
1229
1230 /* Normalize & Store in outFilter */
28bf81c9
MN
1231 for(i=0; i<dstW; i++)
1232 {
1233 int j;
1234 double sum=0;
1235 double scale= one;
6c7506de 1236 for(j=0; j<filterSize; j++)
28bf81c9 1237 {
6c7506de 1238 sum+= filter[i*filterSize + j];
28bf81c9
MN
1239 }
1240 scale/= sum;
93768378 1241 for(j=0; j<*outFilterSize; j++)
28bf81c9 1242 {
6c7506de 1243 (*outFilter)[i*(*outFilterSize) + j]= (int)(filter[i*filterSize + j]*scale);
28bf81c9
MN
1244 }
1245 }
adeaecb9
MN
1246
1247 (*filterPos)[dstW]= (*filterPos)[dstW-1]; // the MMX scaler will read over the end
1248 for(i=0; i<*outFilterSize; i++)
1249 {
1250 int j= dstW*(*outFilterSize);
1251 (*outFilter)[j + i]= (*outFilter)[j + i - (*outFilterSize)];
1252 }
c7f822d9 1253
6c7506de 1254 free(filter);
7630f2e0 1255}
31190492 1256
28bf81c9 1257#ifdef ARCH_X86
b7dc6f66 1258static void initMMX2HScaler(int dstW, int xInc, uint8_t *funnyCode, int16_t *filter, int32_t *filterPos, int numSplits)
28bf81c9 1259{
b7dc6f66
MN
1260 uint8_t *fragmentA;
1261 int imm8OfPShufW1A;
1262 int imm8OfPShufW2A;
1263 int fragmentLengthA;
1264 uint8_t *fragmentB;
1265 int imm8OfPShufW1B;
1266 int imm8OfPShufW2B;
1267 int fragmentLengthB;
1268 int fragmentPos;
28bf81c9
MN
1269
1270 int xpos, i;
1271
1272 // create an optimized horizontal scaling routine
1273
1274 //code fragment
1275
1276 asm volatile(
1277 "jmp 9f \n\t"
1278 // Begin
1279 "0: \n\t"
b7dc6f66
MN
1280 "movq (%%edx, %%eax), %%mm3 \n\t"
1281 "movd (%%ecx, %%esi), %%mm0 \n\t"
1282 "movd 1(%%ecx, %%esi), %%mm1 \n\t"
1283 "punpcklbw %%mm7, %%mm1 \n\t"
1284 "punpcklbw %%mm7, %%mm0 \n\t"
28bf81c9
MN
1285 "pshufw $0xFF, %%mm1, %%mm1 \n\t"
1286 "1: \n\t"
28bf81c9
MN
1287 "pshufw $0xFF, %%mm0, %%mm0 \n\t"
1288 "2: \n\t"
28bf81c9 1289 "psubw %%mm1, %%mm0 \n\t"
b7dc6f66 1290 "movl 8(%%ebx, %%eax), %%esi \n\t"
28bf81c9 1291 "pmullw %%mm3, %%mm0 \n\t"
28bf81c9
MN
1292 "psllw $7, %%mm1 \n\t"
1293 "paddw %%mm1, %%mm0 \n\t"
1294
1295 "movq %%mm0, (%%edi, %%eax) \n\t"
1296
1297 "addl $8, %%eax \n\t"
1298 // End
1299 "9: \n\t"
1300// "int $3\n\t"
1301 "leal 0b, %0 \n\t"
1302 "leal 1b, %1 \n\t"
1303 "leal 2b, %2 \n\t"
1304 "decl %1 \n\t"
1305 "decl %2 \n\t"
1306 "subl %0, %1 \n\t"
1307 "subl %0, %2 \n\t"
1308 "leal 9b, %3 \n\t"
1309 "subl %0, %3 \n\t"
b7dc6f66
MN
1310
1311
1312 :"=r" (fragmentA), "=r" (imm8OfPShufW1A), "=r" (imm8OfPShufW2A),
1313 "=r" (fragmentLengthA)
28bf81c9
MN
1314 );
1315
b7dc6f66
MN
1316 asm volatile(
1317 "jmp 9f \n\t"
1318 // Begin
1319 "0: \n\t"
1320 "movq (%%edx, %%eax), %%mm3 \n\t"
1321 "movd (%%ecx, %%esi), %%mm0 \n\t"
1322 "punpcklbw %%mm7, %%mm0 \n\t"
1323 "pshufw $0xFF, %%mm0, %%mm1 \n\t"
1324 "1: \n\t"
1325 "pshufw $0xFF, %%mm0, %%mm0 \n\t"
1326 "2: \n\t"
1327 "psubw %%mm1, %%mm0 \n\t"
1328 "movl 8(%%ebx, %%eax), %%esi \n\t"
1329 "pmullw %%mm3, %%mm0 \n\t"
1330 "psllw $7, %%mm1 \n\t"
1331 "paddw %%mm1, %%mm0 \n\t"
1332
1333 "movq %%mm0, (%%edi, %%eax) \n\t"
28bf81c9 1334
b7dc6f66
MN
1335 "addl $8, %%eax \n\t"
1336 // End
1337 "9: \n\t"
1338// "int $3\n\t"
1339 "leal 0b, %0 \n\t"
1340 "leal 1b, %1 \n\t"
1341 "leal 2b, %2 \n\t"
1342 "decl %1 \n\t"
1343 "decl %2 \n\t"
1344 "subl %0, %1 \n\t"
1345 "subl %0, %2 \n\t"
1346 "leal 9b, %3 \n\t"
1347 "subl %0, %3 \n\t"
1348
1349
1350 :"=r" (fragmentB), "=r" (imm8OfPShufW1B), "=r" (imm8OfPShufW2B),
1351 "=r" (fragmentLengthB)
1352 );
1353
1354 xpos= 0; //lumXInc/2 - 0x8000; // difference between pixel centers
1355 fragmentPos=0;
1356
1357 for(i=0; i<dstW/numSplits; i++)
28bf81c9
MN
1358 {
1359 int xx=xpos>>16;
1360
1361 if((i&3) == 0)
1362 {
1363 int a=0;
1364 int b=((xpos+xInc)>>16) - xx;
1365 int c=((xpos+xInc*2)>>16) - xx;
1366 int d=((xpos+xInc*3)>>16) - xx;
1367
b7dc6f66
MN
1368 filter[i ] = (( xpos & 0xFFFF) ^ 0xFFFF)>>9;
1369 filter[i+1] = (((xpos+xInc ) & 0xFFFF) ^ 0xFFFF)>>9;
1370 filter[i+2] = (((xpos+xInc*2) & 0xFFFF) ^ 0xFFFF)>>9;
1371 filter[i+3] = (((xpos+xInc*3) & 0xFFFF) ^ 0xFFFF)>>9;
1372 filterPos[i/2]= xx;
1373
1374 if(d+1<4)
1375 {
1376 int maxShift= 3-(d+1);
1377 int shift=0;
1378
1379 memcpy(funnyCode + fragmentPos, fragmentB, fragmentLengthB);
1380
1381 funnyCode[fragmentPos + imm8OfPShufW1B]=
1382 (a+1) | ((b+1)<<2) | ((c+1)<<4) | ((d+1)<<6);
1383 funnyCode[fragmentPos + imm8OfPShufW2B]=
1384 a | (b<<2) | (c<<4) | (d<<6);
1385
1386 if(i+3>=dstW) shift=maxShift; //avoid overread
1387 else if((filterPos[i/2]&3) <= maxShift) shift=filterPos[i/2]&3; //Align
1388
1389 if(shift && i>=shift)
1390 {
1391 funnyCode[fragmentPos + imm8OfPShufW1B]+= 0x55*shift;
1392 funnyCode[fragmentPos + imm8OfPShufW2B]+= 0x55*shift;
1393 filterPos[i/2]-=shift;
1394 }
1395
1396 fragmentPos+= fragmentLengthB;
1397 }
1398 else
1399 {
1400 int maxShift= 3-d;
1401 int shift=0;
1402
1403 memcpy(funnyCode + fragmentPos, fragmentA, fragmentLengthA);
28bf81c9 1404
b7dc6f66
MN
1405 funnyCode[fragmentPos + imm8OfPShufW1A]=
1406 funnyCode[fragmentPos + imm8OfPShufW2A]=
1407 a | (b<<2) | (c<<4) | (d<<6);
28bf81c9 1408
b7dc6f66
MN
1409 if(i+4>=dstW) shift=maxShift; //avoid overread
1410 else if((filterPos[i/2]&3) <= maxShift) shift=filterPos[i/2]&3; //partial align
28bf81c9 1411
b7dc6f66
MN
1412 if(shift && i>=shift)
1413 {
1414 funnyCode[fragmentPos + imm8OfPShufW1A]+= 0x55*shift;
1415 funnyCode[fragmentPos + imm8OfPShufW2A]+= 0x55*shift;
1416 filterPos[i/2]-=shift;
1417 }
1418
1419 fragmentPos+= fragmentLengthA;
1420 }
1421
1422 funnyCode[fragmentPos]= RET;
28bf81c9
MN
1423 }
1424 xpos+=xInc;
1425 }
b7dc6f66 1426 filterPos[i/2]= xpos>>16; // needed to jump to the next part
28bf81c9
MN
1427}
1428#endif // ARCH_X86
1429
1430//FIXME remove
31190492 1431void SwScale_Init(){
28bf81c9
MN
1432}
1433
1434static void globalInit(){
31190492
A
1435 // generating tables:
1436 int i;
c1b0bfb4
MN
1437 for(i=0; i<768; i++){
1438 int c= MIN(MAX(i-256, 0), 255);
1439 clip_table[i]=c;
b18ea156 1440 }
c1b0bfb4 1441
28bf81c9
MN
1442cpuCaps= gCpuCaps;
1443
1444#ifdef RUNTIME_CPUDETECT
1445#ifdef CAN_COMPILE_X86_ASM
1446 // ordered per speed fasterst first
1447 if(gCpuCaps.hasMMX2)
1448 swScale= swScale_MMX2;
1449 else if(gCpuCaps.has3DNow)
7f56a527 1450 swScale= swScale_3DNow;
28bf81c9
MN
1451 else if(gCpuCaps.hasMMX)
1452 swScale= swScale_MMX;
1453 else
1454 swScale= swScale_C;
1455
1456#else
1457 swScale= swScale_C;
1458 cpuCaps.hasMMX2 = cpuCaps.hasMMX = cpuCaps.has3DNow = 0;
1459#endif
1460#else //RUNTIME_CPUDETECT
1461#ifdef HAVE_MMX2
1462 swScale= swScale_MMX2;
1463 cpuCaps.has3DNow = 0;
1464#elif defined (HAVE_3DNOW)
7f56a527 1465 swScale= swScale_3DNow;
28bf81c9
MN
1466 cpuCaps.hasMMX2 = 0;
1467#elif defined (HAVE_MMX)
1468 swScale= swScale_MMX;
1469 cpuCaps.hasMMX2 = cpuCaps.has3DNow = 0;
1470#else
1471 swScale= swScale_C;
1472 cpuCaps.hasMMX2 = cpuCaps.hasMMX = cpuCaps.has3DNow = 0;
1473#endif
1474#endif //!RUNTIME_CPUDETECT
31190492 1475}
7630f2e0 1476
d4e24275 1477static int PlanarToNV12Wrapper(SwsContext *c, uint8_t* src[], int srcStride[], int srcSliceY,
0d9f3d85
A
1478 int srcSliceH, uint8_t* dstParam[], int dstStride[]){
1479 uint8_t *dst=dstParam[0] + dstStride[0]*srcSliceY;
1480 /* Copy Y plane */
1481 if(dstStride[0]==srcStride[0])
1482 memcpy(dst, src[0], srcSliceH*dstStride[0]);
1483 else
1484 {
1485 int i;
1486 uint8_t *srcPtr= src[0];
1487 uint8_t *dstPtr= dst;
1488 for(i=0; i<srcSliceH; i++)
1489 {
1490 memcpy(dstPtr, srcPtr, srcStride[0]);
1491 srcPtr+= srcStride[0];
1492 dstPtr+= dstStride[0];
1493 }
1494 }
1495 dst = dstParam[1] + dstStride[1]*srcSliceY;
fccb9b2b
MN
1496 interleaveBytes( src[1],src[2],dst,c->srcW,srcSliceH,srcStride[1],srcStride[2],dstStride[0] );
1497
d4e24275 1498 return srcSliceH;
0d9f3d85
A
1499}
1500
d4e24275 1501static int PlanarToYuy2Wrapper(SwsContext *c, uint8_t* src[], int srcStride[], int srcSliceY,
0d9f3d85
A
1502 int srcSliceH, uint8_t* dstParam[], int dstStride[]){
1503 uint8_t *dst=dstParam[0] + dstStride[0]*srcSliceY;
1504
fccb9b2b
MN
1505 yv12toyuy2( src[0],src[1],src[2],dst,c->srcW,srcSliceH,srcStride[0],srcStride[1],dstStride[0] );
1506
d4e24275 1507 return srcSliceH;
0d9f3d85
A
1508}
1509
e09d12f4 1510/* {RGB,BGR}{15,16,24,32} -> {RGB,BGR}{15,16,24,32} */
d4e24275 1511static int rgb2rgbWrapper(SwsContext *c, uint8_t* src[], int srcStride[], int srcSliceY,
e09d12f4
MN
1512 int srcSliceH, uint8_t* dst[], int dstStride[]){
1513 const int srcFormat= c->srcFormat;
1514 const int dstFormat= c->dstFormat;
1515 const int srcBpp= ((srcFormat&0xFF) + 7)>>3;
1516 const int dstBpp= ((dstFormat&0xFF) + 7)>>3;
1517 const int srcId= (srcFormat&0xFF)>>2; // 1:0, 4:1, 8:2, 15:3, 16:4, 24:6, 32:8
1518 const int dstId= (dstFormat&0xFF)>>2;
1519 void (*conv)(const uint8_t *src, uint8_t *dst, unsigned src_size)=NULL;
1520
1521 /* BGR -> BGR */
20380eb8
MN
1522 if( (isBGR(srcFormat) && isBGR(dstFormat))
1523 || (isRGB(srcFormat) && isRGB(dstFormat))){
e09d12f4
MN
1524 switch(srcId | (dstId<<4)){
1525 case 0x34: conv= rgb16to15; break;
1526 case 0x36: conv= rgb24to15; break;
1527 case 0x38: conv= rgb32to15; break;
1528 case 0x43: conv= rgb15to16; break;
1529 case 0x46: conv= rgb24to16; break;
1530 case 0x48: conv= rgb32to16; break;
1531 case 0x63: conv= rgb15to24; break;
1532 case 0x64: conv= rgb16to24; break;
1533 case 0x68: conv= rgb32to24; break;
1534 case 0x83: conv= rgb15to32; break;
1535 case 0x84: conv= rgb16to32; break;
1536 case 0x86: conv= rgb24to32; break;
1537 default: MSG_ERR("swScaler: internal error %s -> %s converter\n",
1538 vo_format_name(srcFormat), vo_format_name(dstFormat)); break;
b935781b 1539 }
20380eb8
MN
1540 }else if( (isBGR(srcFormat) && isRGB(dstFormat))
1541 || (isRGB(srcFormat) && isBGR(dstFormat))){
e09d12f4
MN
1542 switch(srcId | (dstId<<4)){
1543 case 0x33: conv= rgb15tobgr15; break;
1544 case 0x34: conv= rgb16tobgr15; break;
1545 case 0x36: conv= rgb24tobgr15; break;
1546 case 0x38: conv= rgb32tobgr15; break;
1547 case 0x43: conv= rgb15tobgr16; break;
1548 case 0x44: conv= rgb16tobgr16; break;
1549 case 0x46: conv= rgb24tobgr16; break;
1550 case 0x48: conv= rgb32tobgr16; break;
1551 case 0x63: conv= rgb15tobgr24; break;
1552 case 0x64: conv= rgb16tobgr24; break;
1553 case 0x66: conv= rgb24tobgr24; break;
1554 case 0x68: conv= rgb32tobgr24; break;
1555 case 0x83: conv= rgb15tobgr32; break;
1556 case 0x84: conv= rgb16tobgr32; break;
1557 case 0x86: conv= rgb24tobgr32; break;
1558 case 0x88: conv= rgb32tobgr32; break;
1559 default: MSG_ERR("swScaler: internal error %s -> %s converter\n",
1560 vo_format_name(srcFormat), vo_format_name(dstFormat)); break;
0d9f3d85 1561 }
20380eb8
MN
1562 }else{
1563 MSG_ERR("swScaler: internal error %s -> %s converter\n",
1564 vo_format_name(srcFormat), vo_format_name(dstFormat));
e09d12f4 1565 }
20380eb8 1566
e09d12f4
MN
1567 if(dstStride[0]*srcBpp == srcStride[0]*dstBpp)
1568 conv(src[0], dst[0] + dstStride[0]*srcSliceY, srcSliceH*srcStride[0]);
0d9f3d85
A
1569 else
1570 {
1571 int i;
1572 uint8_t *srcPtr= src[0];
1573 uint8_t *dstPtr= dst[0] + dstStride[0]*srcSliceY;
1574
1575 for(i=0; i<srcSliceH; i++)
1576 {
e09d12f4 1577 conv(srcPtr, dstPtr, c->srcW*srcBpp);
0d9f3d85
A
1578 srcPtr+= srcStride[0];
1579 dstPtr+= dstStride[0];
1580 }
1581 }
d4e24275 1582 return srcSliceH;
0d9f3d85
A
1583}
1584
d4e24275 1585static int bgr24toyv12Wrapper(SwsContext *c, uint8_t* src[], int srcStride[], int srcSliceY,
ec22603f
MN
1586 int srcSliceH, uint8_t* dst[], int dstStride[]){
1587
1588 rgb24toyv12(
1589 src[0],
1590 dst[0]+ srcSliceY *dstStride[0],
1591 dst[1]+(srcSliceY>>1)*dstStride[1],
1592 dst[2]+(srcSliceY>>1)*dstStride[2],
1593 c->srcW, srcSliceH,
1594 dstStride[0], dstStride[1], srcStride[0]);
d4e24275 1595 return srcSliceH;
ec22603f
MN
1596}
1597
d4e24275 1598static int yvu9toyv12Wrapper(SwsContext *c, uint8_t* src[], int srcStride[], int srcSliceY,
b241cbf2
MN
1599 int srcSliceH, uint8_t* dst[], int dstStride[]){
1600 int i;
1601
1602 /* copy Y */
1603 if(srcStride[0]==dstStride[0])
1604 memcpy(dst[0]+ srcSliceY*dstStride[0], src[0], srcStride[0]*srcSliceH);
1605 else{
1606 uint8_t *srcPtr= src[0];
1607 uint8_t *dstPtr= dst[0] + dstStride[0]*srcSliceY;
1608
1609 for(i=0; i<srcSliceH; i++)
1610 {
1611 memcpy(dstPtr, srcPtr, c->srcW);
1612 srcPtr+= srcStride[0];
1613 dstPtr+= dstStride[0];
1614 }
1615 }
1616
1617 if(c->dstFormat==IMGFMT_YV12){
1618 planar2x(src[1], dst[1], c->chrSrcW, c->chrSrcH, srcStride[1], dstStride[1]);
1619 planar2x(src[2], dst[2], c->chrSrcW, c->chrSrcH, srcStride[2], dstStride[2]);
1620 }else{
1621 planar2x(src[1], dst[2], c->chrSrcW, c->chrSrcH, srcStride[1], dstStride[2]);
1622 planar2x(src[2], dst[1], c->chrSrcW, c->chrSrcH, srcStride[2], dstStride[1]);
1623 }
d4e24275 1624 return srcSliceH;
b241cbf2
MN
1625}
1626
44c1035c
MN
1627/**
1628 * bring pointers in YUV order instead of YVU
1629 */
fccb9b2b
MN
1630static inline void sws_orderYUV(int format, uint8_t * sortedP[], int sortedStride[], uint8_t * p[], int stride[]){
1631 if(format == IMGFMT_YV12 || format == IMGFMT_YVU9
d80e2fa2 1632 || format == IMGFMT_444P || format == IMGFMT_422P || format == IMGFMT_411P){
44c1035c 1633 sortedP[0]= p[0];
fccb9b2b
MN
1634 sortedP[1]= p[2];
1635 sortedP[2]= p[1];
c7a810cc 1636 sortedStride[0]= stride[0];
fccb9b2b
MN
1637 sortedStride[1]= stride[2];
1638 sortedStride[2]= stride[1];
44c1035c 1639 }
c7a810cc 1640 else if(isPacked(format) || isGray(format))
44c1035c
MN
1641 {
1642 sortedP[0]= p[0];
1643 sortedP[1]=
1644 sortedP[2]= NULL;
c7a810cc 1645 sortedStride[0]= stride[0];
44c1035c
MN
1646 sortedStride[1]=
1647 sortedStride[2]= 0;
1648 }
fccb9b2b 1649 else if(format == IMGFMT_I420 || format == IMGFMT_IYUV)
44c1035c
MN
1650 {
1651 sortedP[0]= p[0];
fccb9b2b
MN
1652 sortedP[1]= p[1];
1653 sortedP[2]= p[2];
c7a810cc 1654 sortedStride[0]= stride[0];
fccb9b2b
MN
1655 sortedStride[1]= stride[1];
1656 sortedStride[2]= stride[2];
e09d12f4
MN
1657 }else{
1658 MSG_ERR("internal error in orderYUV\n");
44c1035c
MN
1659 }
1660}
b935781b 1661
b6654a54 1662/* unscaled copy like stuff (assumes nearly identical formats) */
3e499f53
MN
1663static int simpleCopy(SwsContext *c, uint8_t* src[], int srcStride[], int srcSliceY,
1664 int srcSliceH, uint8_t* dst[], int dstStride[]){
b6654a54
MN
1665
1666 if(isPacked(c->srcFormat))
1667 {
1668 if(dstStride[0]==srcStride[0])
1669 memcpy(dst[0] + dstStride[0]*srcSliceY, src[0], srcSliceH*dstStride[0]);
1670 else
1671 {
1672 int i;
1673 uint8_t *srcPtr= src[0];
1674 uint8_t *dstPtr= dst[0] + dstStride[0]*srcSliceY;
a861d4d7
MN
1675 int length=0;
1676
1677 /* universal length finder */
9bd8bd1a
MN
1678 while(length+c->srcW <= ABS(dstStride[0])
1679 && length+c->srcW <= ABS(srcStride[0])) length+= c->srcW;
a861d4d7 1680 ASSERT(length!=0);
b6654a54
MN
1681
1682 for(i=0; i<srcSliceH; i++)
1683 {
1684 memcpy(dstPtr, srcPtr, length);
1685 srcPtr+= srcStride[0];
1686 dstPtr+= dstStride[0];
1687 }
1688 }
1689 }
1690 else
44c1035c 1691 { /* Planar YUV or gray */
b6654a54
MN
1692 int plane;
1693 for(plane=0; plane<3; plane++)
1694 {
e616aa93
MN
1695 int length= plane==0 ? c->srcW : -((-c->srcW )>>c->chrDstHSubSample);
1696 int y= plane==0 ? srcSliceY: -((-srcSliceY)>>c->chrDstVSubSample);
1697 int height= plane==0 ? srcSliceH: -((-srcSliceH)>>c->chrDstVSubSample);
44c1035c
MN
1698
1699 if((isGray(c->srcFormat) || isGray(c->dstFormat)) && plane>0)
0d9f3d85 1700 {
44c1035c 1701 if(!isGray(c->dstFormat))
e616aa93 1702 memset(dst[plane], 128, dstStride[plane]*height);
0d9f3d85 1703 }
b6654a54
MN
1704 else
1705 {
44c1035c
MN
1706 if(dstStride[plane]==srcStride[plane])
1707 memcpy(dst[plane] + dstStride[plane]*y, src[plane], height*dstStride[plane]);
1708 else
b6654a54 1709 {
44c1035c
MN
1710 int i;
1711 uint8_t *srcPtr= src[plane];
1712 uint8_t *dstPtr= dst[plane] + dstStride[plane]*y;
1713 for(i=0; i<height; i++)
1714 {
1715 memcpy(dstPtr, srcPtr, length);
1716 srcPtr+= srcStride[plane];
1717 dstPtr+= dstStride[plane];
1718 }
b6654a54
MN
1719 }
1720 }
1721 }
1722 }
d4e24275 1723 return srcSliceH;
37079906 1724}
28bf81c9 1725
44c1035c 1726static int remove_dup_fourcc(int fourcc)
0d9f3d85
A
1727{
1728 switch(fourcc)
1729 {
fccb9b2b
MN
1730 case IMGFMT_I420:
1731 case IMGFMT_IYUV: return IMGFMT_YV12;
0d9f3d85 1732 case IMGFMT_Y8 : return IMGFMT_Y800;
0c51ef97 1733 case IMGFMT_IF09: return IMGFMT_YVU9;
0d9f3d85
A
1734 default: return fourcc;
1735 }
1736}
1737
c7a810cc
MN
1738static void getSubSampleFactors(int *h, int *v, int format){
1739 switch(format){
7322a67c 1740 case IMGFMT_UYVY:
c7a810cc
MN
1741 case IMGFMT_YUY2:
1742 *h=1;
1743 *v=0;
1744 break;
1745 case IMGFMT_YV12:
e616aa93 1746 case IMGFMT_Y800: //FIXME remove after different subsamplings are fully implemented
c7a810cc
MN
1747 *h=1;
1748 *v=1;
1749 break;
1750 case IMGFMT_YVU9:
1751 *h=2;
1752 *v=2;
1753 break;
d80e2fa2
MN
1754 case IMGFMT_444P:
1755 *h=0;
1756 *v=0;
1757 break;
1758 case IMGFMT_422P:
1759 *h=1;
1760 *v=0;
1761 break;
1762 case IMGFMT_411P:
1763 *h=2;
1764 *v=0;
1765 break;
c7a810cc
MN
1766 default:
1767 *h=0;
1768 *v=0;
1769 break;
1770 }
1771}
1772
5427e242
MN
1773static uint16_t roundToInt16(int64_t f){
1774 int r= (f + (1<<15))>>16;
1775 if(r<-0x7FFF) return 0x8000;
1776 else if(r> 0x7FFF) return 0x7FFF;
1777 else return r;
0481412a
MN
1778}
1779
1780/**
5427e242 1781 * @param inv_table the yuv2rgb coeffs, normally Inverse_Table_6_9[x]
0481412a 1782 * @param fullRange if 1 then the luma range is 0..255 if 0 its 16..235
5427e242 1783 * @return -1 if not supported
0481412a 1784 */
5427e242
MN
1785int sws_setColorspaceDetails(SwsContext *c, const int inv_table[4], int srcRange, const int table[4], int dstRange, int brightness, int contrast, int saturation){
1786 int64_t crv = inv_table[0];
1787 int64_t cbu = inv_table[1];
1788 int64_t cgu = -inv_table[2];
1789 int64_t cgv = -inv_table[3];
1790 int64_t cy = 1<<16;
1791 int64_t oy = 0;
1792
1793 if(isYUV(c->dstFormat) || isGray(c->dstFormat)) return -1;
1794 memcpy(c->srcColorspaceTable, inv_table, sizeof(int)*4);
1795 memcpy(c->dstColorspaceTable, table, sizeof(int)*4);
1796
1797 c->brightness= brightness;
1798 c->contrast = contrast;
1799 c->saturation= saturation;
1800 c->srcRange = srcRange;
1801 c->dstRange = dstRange;
0481412a
MN
1802
1803 c->uOffset= 0x0400040004000400LL;
1804 c->vOffset= 0x0400040004000400LL;
1805
5427e242
MN
1806 if(!srcRange){
1807 cy= (cy*255) / 219;
1808 oy= 16<<16;
0481412a
MN
1809 }
1810
5427e242
MN
1811 cy = (cy *contrast )>>16;
1812 crv= (crv*contrast * saturation)>>32;
1813 cbu= (cbu*contrast * saturation)>>32;
1814 cgu= (cgu*contrast * saturation)>>32;
1815 cgv= (cgv*contrast * saturation)>>32;
0481412a 1816
5427e242 1817 oy -= 256*brightness;
0481412a
MN
1818
1819 c->yCoeff= roundToInt16(cy *8192) * 0x0001000100010001ULL;
1820 c->vrCoeff= roundToInt16(crv*8192) * 0x0001000100010001ULL;
1821 c->ubCoeff= roundToInt16(cbu*8192) * 0x0001000100010001ULL;
1822 c->vgCoeff= roundToInt16(cgv*8192) * 0x0001000100010001ULL;
1823 c->ugCoeff= roundToInt16(cgu*8192) * 0x0001000100010001ULL;
1824 c->yOffset= roundToInt16(oy * 8) * 0x0001000100010001ULL;
5427e242
MN
1825
1826 yuv2rgb_c_init_tables(c, inv_table, srcRange, brightness, contrast, saturation);
1827 //FIXME factorize
1828
1829 return 0;
1830}
1831
1832/**
1833 * @return -1 if not supported
1834 */
1835int sws_getColorspaceDetails(SwsContext *c, int **inv_table, int *srcRange, int **table, int *dstRange, int *brightness, int *contrast, int *saturation){
1836 if(isYUV(c->dstFormat) || isGray(c->dstFormat)) return -1;
1837
1838 *inv_table = c->srcColorspaceTable;
1839 *table = c->dstColorspaceTable;
1840 *srcRange = c->srcRange;
1841 *dstRange = c->dstRange;
1842 *brightness= c->brightness;
1843 *contrast = c->contrast;
1844 *saturation= c->saturation;
1845
1846 return 0;
0481412a
MN
1847}
1848
fccb9b2b 1849SwsContext *sws_getContext(int srcW, int srcH, int origSrcFormat, int dstW, int dstH, int origDstFormat, int flags,
28bf81c9
MN
1850 SwsFilter *srcFilter, SwsFilter *dstFilter){
1851
28bf81c9
MN
1852 SwsContext *c;
1853 int i;
37079906 1854 int usesFilter;
e09d12f4 1855 int unscaled, needsDither;
fccb9b2b 1856 int srcFormat, dstFormat;
c7f822d9 1857 SwsFilter dummyFilter= {NULL, NULL, NULL, NULL};
5cebb24b
MN
1858#ifdef ARCH_X86
1859 if(gCpuCaps.hasMMX)
1860 asm volatile("emms\n\t"::: "memory");
1861#endif
28bf81c9 1862 if(swScale==NULL) globalInit();
fccb9b2b 1863
6ff0ad6b 1864 /* avoid dupplicate Formats, so we dont need to check to much */
fccb9b2b
MN
1865 srcFormat = remove_dup_fourcc(origSrcFormat);
1866 dstFormat = remove_dup_fourcc(origDstFormat);
44c1035c
MN
1867
1868 unscaled = (srcW == dstW && srcH == dstH);
e09d12f4
MN
1869 needsDither= (isBGR(dstFormat) || isRGB(dstFormat))
1870 && (dstFormat&0xFF)<24
1871 && ((dstFormat&0xFF)<(srcFormat&0xFF) || (!(isRGB(srcFormat) || isBGR(srcFormat))));
44c1035c
MN
1872
1873 if(!isSupportedIn(srcFormat))
b81cf274 1874 {
44c1035c
MN
1875 MSG_ERR("swScaler: %s is not supported as input format\n", vo_format_name(srcFormat));
1876 return NULL;
1877 }
1878 if(!isSupportedOut(dstFormat))
1879 {
1880 MSG_ERR("swScaler: %s is not supported as output format\n", vo_format_name(dstFormat));
1881 return NULL;
b81cf274 1882 }
44c1035c 1883
28bf81c9 1884 /* sanity check */
b81cf274
MN
1885 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
1886 {
0d9f3d85 1887 MSG_ERR("swScaler: %dx%d -> %dx%d is invalid scaling dimension\n",
b81cf274
MN
1888 srcW, srcH, dstW, dstH);
1889 return NULL;
1890 }
28bf81c9 1891
c7f822d9
MN
1892 if(!dstFilter) dstFilter= &dummyFilter;
1893 if(!srcFilter) srcFilter= &dummyFilter;
1894
28bf81c9 1895 c= memalign(64, sizeof(SwsContext));
c7f822d9 1896 memset(c, 0, sizeof(SwsContext));
28bf81c9
MN
1897
1898 c->srcW= srcW;
1899 c->srcH= srcH;
1900 c->dstW= dstW;
1901 c->dstH= dstH;
5521b193
MN
1902 c->lumXInc= ((srcW<<16) + (dstW>>1))/dstW;
1903 c->lumYInc= ((srcH<<16) + (dstH>>1))/dstH;
28bf81c9
MN
1904 c->flags= flags;
1905 c->dstFormat= dstFormat;
1906 c->srcFormat= srcFormat;
fccb9b2b
MN
1907 c->origDstFormat= origDstFormat;
1908 c->origSrcFormat= origSrcFormat;
28bf81c9 1909
37079906
MN
1910 usesFilter=0;
1911 if(dstFilter->lumV!=NULL && dstFilter->lumV->length>1) usesFilter=1;
1912 if(dstFilter->lumH!=NULL && dstFilter->lumH->length>1) usesFilter=1;
1913 if(dstFilter->chrV!=NULL && dstFilter->chrV->length>1) usesFilter=1;
1914 if(dstFilter->chrH!=NULL && dstFilter->chrH->length>1) usesFilter=1;
1915 if(srcFilter->lumV!=NULL && srcFilter->lumV->length>1) usesFilter=1;
1916 if(srcFilter->lumH!=NULL && srcFilter->lumH->length>1) usesFilter=1;
1917 if(srcFilter->chrV!=NULL && srcFilter->chrV->length>1) usesFilter=1;
1918 if(srcFilter->chrH!=NULL && srcFilter->chrH->length>1) usesFilter=1;
e616aa93
MN
1919
1920 getSubSampleFactors(&c->chrSrcHSubSample, &c->chrSrcVSubSample, srcFormat);
1921 getSubSampleFactors(&c->chrDstHSubSample, &c->chrDstVSubSample, dstFormat);
1922
1923 // reuse chroma for 2 pixles rgb/bgr unless user wants full chroma interpolation
1924 if((isBGR(dstFormat) || isRGB(dstFormat)) && !(flags&SWS_FULL_CHR_H_INT)) c->chrDstHSubSample=1;
1925
5859233b
MN
1926 // drop some chroma lines if the user wants it
1927 c->vChrDrop= (flags&SWS_SRC_V_CHR_DROP_MASK)>>SWS_SRC_V_CHR_DROP_SHIFT;
1928 c->chrSrcVSubSample+= c->vChrDrop;
e616aa93 1929
5859233b 1930 // drop every 2. pixel for chroma calculation unless user wants full chroma
e616aa93
MN
1931 if((isBGR(srcFormat) || isRGB(srcFormat)) && !(flags&SWS_FULL_CHR_H_INP))
1932 c->chrSrcHSubSample=1;
1933
1934 c->chrIntHSubSample= c->chrDstHSubSample;
1935 c->chrIntVSubSample= c->chrSrcVSubSample;
5427e242 1936
e616aa93
MN
1937 // note the -((-x)>>y) is so that we allways round toward +inf
1938 c->chrSrcW= -((-srcW) >> c->chrSrcHSubSample);
1939 c->chrSrcH= -((-srcH) >> c->chrSrcVSubSample);
1940 c->chrDstW= -((-dstW) >> c->chrDstHSubSample);
1941 c->chrDstH= -((-dstH) >> c->chrDstVSubSample);
5427e242
MN
1942
1943 sws_setColorspaceDetails(c, Inverse_Table_6_9[SWS_CS_DEFAULT], 0, Inverse_Table_6_9[SWS_CS_DEFAULT] /* FIXME*/, 0, 0, 1<<16, 1<<16);
cf7d1c1a 1944
b935781b 1945 /* unscaled special Cases */
44c1035c 1946 if(unscaled && !usesFilter)
37079906 1947 {
0d9f3d85 1948 /* yv12_to_nv12 */
fccb9b2b 1949 if(srcFormat == IMGFMT_YV12 && dstFormat == IMGFMT_NV12)
0d9f3d85
A
1950 {
1951 c->swScale= PlanarToNV12Wrapper;
0d9f3d85 1952 }
37079906 1953 /* yuv2bgr */
fccb9b2b 1954 if((srcFormat==IMGFMT_YV12 || srcFormat==IMGFMT_422P) && (isBGR(dstFormat) || isRGB(dstFormat)))
37079906 1955 {
5427e242 1956 c->swScale= yuv2rgb_get_func_ptr(c);
37079906 1957 }
b241cbf2 1958
fccb9b2b 1959 if( srcFormat==IMGFMT_YVU9 && dstFormat==IMGFMT_YV12 )
b241cbf2
MN
1960 {
1961 c->swScale= yvu9toyv12Wrapper;
b241cbf2
MN
1962 }
1963
ec22603f
MN
1964 /* bgr24toYV12 */
1965 if(srcFormat==IMGFMT_BGR24 && dstFormat==IMGFMT_YV12)
ec22603f 1966 c->swScale= bgr24toyv12Wrapper;
e09d12f4
MN
1967
1968 /* rgb/bgr -> rgb/bgr (no dither needed forms) */
1969 if( (isBGR(srcFormat) || isRGB(srcFormat))
1970 && (isBGR(dstFormat) || isRGB(dstFormat))
1971 && !needsDither)
1972 c->swScale= rgb2rgbWrapper;
1973
1974 /* LQ converters if -sws 0 or -sws 4*/
1975 if(c->flags&(SWS_FAST_BILINEAR|SWS_POINT)){
1976 /* rgb/bgr -> rgb/bgr (dither needed forms) */
1977 if( (isBGR(srcFormat) || isRGB(srcFormat))
1978 && (isBGR(dstFormat) || isRGB(dstFormat))
1979 && needsDither)
1980 c->swScale= rgb2rgbWrapper;
2ce486d8
MN
1981
1982 /* yv12_to_yuy2 */
fccb9b2b 1983 if(srcFormat == IMGFMT_YV12 && dstFormat == IMGFMT_YUY2)
2ce486d8
MN
1984 {
1985 c->swScale= PlanarToYuy2Wrapper;
2ce486d8 1986 }
e09d12f4 1987 }
ec22603f 1988
20380eb8
MN
1989 /* simple copy */
1990 if( srcFormat == dstFormat
20380eb8
MN
1991 || (isPlanarYUV(srcFormat) && isGray(dstFormat))
1992 || (isPlanarYUV(dstFormat) && isGray(srcFormat))
1993 )
1994 {
1995 c->swScale= simpleCopy;
1996 }
1997
e09d12f4 1998 if(c->swScale){
ec22603f 1999 if(flags&SWS_PRINT_INFO)
0d9f3d85 2000 MSG_INFO("SwScaler: using unscaled %s -> %s special converter\n",
ec22603f
MN
2001 vo_format_name(srcFormat), vo_format_name(dstFormat));
2002 return c;
2003 }
37079906
MN
2004 }
2005
28bf81c9
MN
2006 if(cpuCaps.hasMMX2)
2007 {
2008 c->canMMX2BeUsed= (dstW >=srcW && (dstW&31)==0 && (srcW&15)==0) ? 1 : 0;
2009 if(!c->canMMX2BeUsed && dstW >=srcW && (srcW&15)==0 && (flags&SWS_FAST_BILINEAR))
2010 {
2011 if(flags&SWS_PRINT_INFO)
0d9f3d85 2012 MSG_INFO("SwScaler: output Width is not a multiple of 32 -> no MMX2 scaler\n");
28bf81c9
MN
2013 }
2014 }
2015 else
2016 c->canMMX2BeUsed=0;
2017
1e621b18
MN
2018 c->chrXInc= ((c->chrSrcW<<16) + (c->chrDstW>>1))/c->chrDstW;
2019 c->chrYInc= ((c->chrSrcH<<16) + (c->chrDstH>>1))/c->chrDstH;
2020
28bf81c9
MN
2021 // match pixel 0 of the src to pixel 0 of dst and match pixel n-2 of src to pixel n-2 of dst
2022 // but only for the FAST_BILINEAR mode otherwise do correct scaling
2023 // n-2 is the last chrominance sample available
2024 // this is not perfect, but noone shuld notice the difference, the more correct variant
2025 // would be like the vertical one, but that would require some special code for the
2026 // first and last pixel
2027 if(flags&SWS_FAST_BILINEAR)
2028 {
1e621b18
MN
2029 if(c->canMMX2BeUsed)
2030 {
2031 c->lumXInc+= 20;
2032 c->chrXInc+= 20;
2033 }
28bf81c9 2034 //we dont use the x86asm scaler if mmx is available
1e621b18
MN
2035 else if(cpuCaps.hasMMX)
2036 {
2037 c->lumXInc = ((srcW-2)<<16)/(dstW-2) - 20;
2038 c->chrXInc = ((c->chrSrcW-2)<<16)/(c->chrDstW-2) - 20;
2039 }
28bf81c9
MN
2040 }
2041
28bf81c9
MN
2042 /* precalculate horizontal scaler filter coefficients */
2043 {
2044 const int filterAlign= cpuCaps.hasMMX ? 4 : 1;
2045
c7f822d9 2046 initFilter(&c->hLumFilter, &c->hLumFilterPos, &c->hLumFilterSize, c->lumXInc,
81a571a8
MN
2047 srcW , dstW, filterAlign, 1<<14,
2048 (flags&SWS_BICUBLIN) ? (flags|SWS_BICUBIC) : flags,
c7f822d9
MN
2049 srcFilter->lumH, dstFilter->lumH);
2050 initFilter(&c->hChrFilter, &c->hChrFilterPos, &c->hChrFilterSize, c->chrXInc,
81a571a8
MN
2051 c->chrSrcW, c->chrDstW, filterAlign, 1<<14,
2052 (flags&SWS_BICUBLIN) ? (flags|SWS_BILINEAR) : flags,
c7f822d9 2053 srcFilter->chrH, dstFilter->chrH);
28bf81c9
MN
2054
2055#ifdef ARCH_X86
2056// cant downscale !!!
2057 if(c->canMMX2BeUsed && (flags & SWS_FAST_BILINEAR))
2058 {
b7dc6f66
MN
2059 c->lumMmx2Filter = (int16_t*)memalign(8, (dstW /8+8)*sizeof(int16_t));
2060 c->chrMmx2Filter = (int16_t*)memalign(8, (c->chrDstW /4+8)*sizeof(int16_t));
2061 c->lumMmx2FilterPos= (int32_t*)memalign(8, (dstW /2/8+8)*sizeof(int32_t));
2062 c->chrMmx2FilterPos= (int32_t*)memalign(8, (c->chrDstW/2/4+8)*sizeof(int32_t));
2063
2064 initMMX2HScaler( dstW, c->lumXInc, c->funnyYCode , c->lumMmx2Filter, c->lumMmx2FilterPos, 8);
2065 initMMX2HScaler(c->chrDstW, c->chrXInc, c->funnyUVCode, c->chrMmx2Filter, c->chrMmx2FilterPos, 4);
28bf81c9
MN
2066 }
2067#endif
2068 } // Init Horizontal stuff
2069
2070
2071
2072 /* precalculate vertical scaler filter coefficients */
c7f822d9 2073 initFilter(&c->vLumFilter, &c->vLumFilterPos, &c->vLumFilterSize, c->lumYInc,
81a571a8
MN
2074 srcH , dstH, 1, (1<<12)-4,
2075 (flags&SWS_BICUBLIN) ? (flags|SWS_BICUBIC) : flags,
c7f822d9
MN
2076 srcFilter->lumV, dstFilter->lumV);
2077 initFilter(&c->vChrFilter, &c->vChrFilterPos, &c->vChrFilterSize, c->chrYInc,
81a571a8
MN
2078 c->chrSrcH, c->chrDstH, 1, (1<<12)-4,
2079 (flags&SWS_BICUBLIN) ? (flags|SWS_BILINEAR) : flags,
2080 srcFilter->chrV, dstFilter->chrV);
28bf81c9
MN
2081
2082 // Calculate Buffer Sizes so that they wont run out while handling these damn slices
2083 c->vLumBufSize= c->vLumFilterSize;
2084 c->vChrBufSize= c->vChrFilterSize;
2085 for(i=0; i<dstH; i++)
2086 {
2087 int chrI= i*c->chrDstH / dstH;
2088 int nextSlice= MAX(c->vLumFilterPos[i ] + c->vLumFilterSize - 1,
e616aa93
MN
2089 ((c->vChrFilterPos[chrI] + c->vChrFilterSize - 1)<<c->chrSrcVSubSample));
2090 nextSlice&= ~3; // Slices start at boundaries which are divisable through 4
28bf81c9
MN
2091 if(c->vLumFilterPos[i ] + c->vLumBufSize < nextSlice)
2092 c->vLumBufSize= nextSlice - c->vLumFilterPos[i ];
e616aa93
MN
2093 if(c->vChrFilterPos[chrI] + c->vChrBufSize < (nextSlice>>c->chrSrcVSubSample))
2094 c->vChrBufSize= (nextSlice>>c->chrSrcVSubSample) - c->vChrFilterPos[chrI];
28bf81c9
MN
2095 }
2096
2097 // allocate pixbufs (we use dynamic allocation because otherwise we would need to
c7f822d9
MN
2098 c->lumPixBuf= (int16_t**)memalign(4, c->vLumBufSize*2*sizeof(int16_t*));
2099 c->chrPixBuf= (int16_t**)memalign(4, c->vChrBufSize*2*sizeof(int16_t*));
6c7506de 2100 //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
2101 for(i=0; i<c->vLumBufSize; i++)
2102 c->lumPixBuf[i]= c->lumPixBuf[i+c->vLumBufSize]= (uint16_t*)memalign(8, 4000);
2103 for(i=0; i<c->vChrBufSize; i++)
2104 c->chrPixBuf[i]= c->chrPixBuf[i+c->vChrBufSize]= (uint16_t*)memalign(8, 8000);
2105
2106 //try to avoid drawing green stuff between the right end and the stride end
2107 for(i=0; i<c->vLumBufSize; i++) memset(c->lumPixBuf[i], 0, 4000);
2108 for(i=0; i<c->vChrBufSize; i++) memset(c->chrPixBuf[i], 64, 8000);
2109
2110 ASSERT(c->chrDstH <= dstH)
28bf81c9 2111
28bf81c9
MN
2112 if(flags&SWS_PRINT_INFO)
2113 {
2114#ifdef DITHER1XBPP
5521b193
MN
2115 char *dither= " dithered";
2116#else
2117 char *dither= "";
28bf81c9
MN
2118#endif
2119 if(flags&SWS_FAST_BILINEAR)
0d9f3d85 2120 MSG_INFO("\nSwScaler: FAST_BILINEAR scaler, ");
28bf81c9 2121 else if(flags&SWS_BILINEAR)
0d9f3d85 2122 MSG_INFO("\nSwScaler: BILINEAR scaler, ");
28bf81c9 2123 else if(flags&SWS_BICUBIC)
0d9f3d85 2124 MSG_INFO("\nSwScaler: BICUBIC scaler, ");
1e621b18 2125 else if(flags&SWS_X)
0d9f3d85 2126 MSG_INFO("\nSwScaler: Experimental scaler, ");
ff7ba856 2127 else if(flags&SWS_POINT)
0d9f3d85 2128 MSG_INFO("\nSwScaler: Nearest Neighbor / POINT scaler, ");
d8863d37 2129 else if(flags&SWS_AREA)
0d9f3d85 2130 MSG_INFO("\nSwScaler: Area Averageing scaler, ");
81a571a8 2131 else if(flags&SWS_BICUBLIN)
a86c461c
MN
2132 MSG_INFO("\nSwScaler: luma BICUBIC / chroma BILINEAR scaler, ");
2133 else if(flags&SWS_GAUSS)
2134 MSG_INFO("\nSwScaler: Gaussian scaler, ");
2135 else if(flags&SWS_SINC)
2136 MSG_INFO("\nSwScaler: Sinc scaler, ");
2137 else if(flags&SWS_LANCZOS)
2138 MSG_INFO("\nSwScaler: Lanczos scaler, ");
2139 else if(flags&SWS_SPLINE)
2140 MSG_INFO("\nSwScaler: Bicubic spline scaler, ");
28bf81c9 2141 else
0d9f3d85 2142 MSG_INFO("\nSwScaler: ehh flags invalid?! ");
28bf81c9 2143
0d9f3d85
A
2144 if(dstFormat==IMGFMT_BGR15 || dstFormat==IMGFMT_BGR16)
2145 MSG_INFO("from %s to%s %s ",
2146 vo_format_name(srcFormat), dither, vo_format_name(dstFormat));
2147 else
2148 MSG_INFO("from %s to %s ",
2149 vo_format_name(srcFormat), vo_format_name(dstFormat));
28bf81c9
MN
2150
2151 if(cpuCaps.hasMMX2)
0d9f3d85 2152 MSG_INFO("using MMX2\n");
28bf81c9 2153 else if(cpuCaps.has3DNow)
0d9f3d85 2154 MSG_INFO("using 3DNOW\n");
28bf81c9 2155 else if(cpuCaps.hasMMX)
0d9f3d85 2156 MSG_INFO("using MMX\n");
28bf81c9 2157 else
0d9f3d85 2158 MSG_INFO("using C\n");
28bf81c9
MN
2159 }
2160
a749913f 2161 if((flags & SWS_PRINT_INFO) && verbose>0)
28bf81c9
MN
2162 {
2163 if(cpuCaps.hasMMX)
2164 {
2165 if(c->canMMX2BeUsed && (flags&SWS_FAST_BILINEAR))
0d9f3d85 2166 MSG_V("SwScaler: using FAST_BILINEAR MMX2 scaler for horizontal scaling\n");
28bf81c9
MN
2167 else
2168 {
2169 if(c->hLumFilterSize==4)
0d9f3d85 2170 MSG_V("SwScaler: using 4-tap MMX scaler for horizontal luminance scaling\n");
28bf81c9 2171 else if(c->hLumFilterSize==8)
0d9f3d85 2172 MSG_V("SwScaler: using 8-tap MMX scaler for horizontal luminance scaling\n");
28bf81c9 2173 else
0d9f3d85 2174 MSG_V("SwScaler: using n-tap MMX scaler for horizontal luminance scaling\n");
28bf81c9
MN
2175
2176 if(c->hChrFilterSize==4)
0d9f3d85 2177 MSG_V("SwScaler: using 4-tap MMX scaler for horizontal chrominance scaling\n");
28bf81c9 2178 else if(c->hChrFilterSize==8)
0d9f3d85 2179 MSG_V("SwScaler: using 8-tap MMX scaler for horizontal chrominance scaling\n");
28bf81c9 2180 else
0d9f3d85 2181 MSG_V("SwScaler: using n-tap MMX scaler for horizontal chrominance scaling\n");
28bf81c9
MN
2182 }
2183 }
2184 else
2185 {
2186#ifdef ARCH_X86
0d9f3d85 2187 MSG_V("SwScaler: using X86-Asm scaler for horizontal scaling\n");
28bf81c9
MN
2188#else
2189 if(flags & SWS_FAST_BILINEAR)
0d9f3d85 2190 MSG_V("SwScaler: using FAST_BILINEAR C scaler for horizontal scaling\n");
28bf81c9 2191 else
0d9f3d85 2192 MSG_V("SwScaler: using C scaler for horizontal scaling\n");
28bf81c9
MN
2193#endif
2194 }
6c7506de 2195 if(isPlanarYUV(dstFormat))
28bf81c9
MN
2196 {
2197 if(c->vLumFilterSize==1)
0d9f3d85 2198 MSG_V("SwScaler: using 1-tap %s \"scaler\" for vertical scaling (YV12 like)\n", cpuCaps.hasMMX ? "MMX" : "C");
28bf81c9 2199 else
0d9f3d85 2200 MSG_V("SwScaler: using n-tap %s scaler for vertical scaling (YV12 like)\n", cpuCaps.hasMMX ? "MMX" : "C");
28bf81c9
MN
2201 }
2202 else
2203 {
2204 if(c->vLumFilterSize==1 && c->vChrFilterSize==2)
0d9f3d85 2205 MSG_V("SwScaler: using 1-tap %s \"scaler\" for vertical luminance scaling (BGR)\n"
28bf81c9
MN
2206 "SwScaler: 2-tap scaler for vertical chrominance scaling (BGR)\n",cpuCaps.hasMMX ? "MMX" : "C");
2207 else if(c->vLumFilterSize==2 && c->vChrFilterSize==2)
0d9f3d85 2208 MSG_V("SwScaler: using 2-tap linear %s scaler for vertical scaling (BGR)\n", cpuCaps.hasMMX ? "MMX" : "C");
28bf81c9 2209 else
0d9f3d85 2210 MSG_V("SwScaler: using n-tap %s scaler for vertical scaling (BGR)\n", cpuCaps.hasMMX ? "MMX" : "C");
28bf81c9
MN
2211 }
2212
2213 if(dstFormat==IMGFMT_BGR24)
0d9f3d85 2214 MSG_V("SwScaler: using %s YV12->BGR24 Converter\n",
28bf81c9 2215 cpuCaps.hasMMX2 ? "MMX2" : (cpuCaps.hasMMX ? "MMX" : "C"));
fd284805 2216 else if(dstFormat==IMGFMT_BGR32)
0d9f3d85 2217 MSG_V("SwScaler: using %s YV12->BGR32 Converter\n", cpuCaps.hasMMX ? "MMX" : "C");
fd284805 2218 else if(dstFormat==IMGFMT_BGR16)
0d9f3d85 2219 MSG_V("SwScaler: using %s YV12->BGR16 Converter\n", cpuCaps.hasMMX ? "MMX" : "C");
fd284805 2220 else if(dstFormat==IMGFMT_BGR15)
0d9f3d85 2221 MSG_V("SwScaler: using %s YV12->BGR15 Converter\n", cpuCaps.hasMMX ? "MMX" : "C");
28bf81c9 2222
0d9f3d85 2223 MSG_V("SwScaler: %dx%d -> %dx%d\n", srcW, srcH, dstW, dstH);
28bf81c9 2224 }
1e621b18
MN
2225 if((flags & SWS_PRINT_INFO) && verbose>1)
2226 {
0d9f3d85 2227 MSG_DBG2("SwScaler:Lum srcW=%d srcH=%d dstW=%d dstH=%d xInc=%d yInc=%d\n",
1e621b18 2228 c->srcW, c->srcH, c->dstW, c->dstH, c->lumXInc, c->lumYInc);
0d9f3d85 2229 MSG_DBG2("SwScaler:Chr srcW=%d srcH=%d dstW=%d dstH=%d xInc=%d yInc=%d\n",
1e621b18
MN
2230 c->chrSrcW, c->chrSrcH, c->chrDstW, c->chrDstH, c->chrXInc, c->chrYInc);
2231 }
37079906
MN
2232
2233 c->swScale= swScale;
28bf81c9
MN
2234 return c;
2235}
2236
2237/**
fccb9b2b
MN
2238 * swscale warper, so we dont need to export the SwsContext.
2239 * assumes planar YUV to be in YUV order instead of YVU
2240 */
2241int sws_scale_ordered(SwsContext *c, uint8_t* src[], int srcStride[], int srcSliceY,
2242 int srcSliceH, uint8_t* dst[], int dstStride[]){
5bf01354 2243 return c->swScale(c, src, srcStride, srcSliceY, srcSliceH, dst, dstStride);
fccb9b2b
MN
2244}
2245
2246/**
d4e24275
MN
2247 * swscale warper, so we dont need to export the SwsContext
2248 */
3e499f53
MN
2249int sws_scale(SwsContext *c, uint8_t* srcParam[], int srcStrideParam[], int srcSliceY,
2250 int srcSliceH, uint8_t* dstParam[], int dstStrideParam[]){
2251 int srcStride[3];
2252 int dstStride[3];
2253 uint8_t *src[3];
2254 uint8_t *dst[3];
fccb9b2b
MN
2255 sws_orderYUV(c->origSrcFormat, src, srcStride, srcParam, srcStrideParam);
2256 sws_orderYUV(c->origDstFormat, dst, dstStride, dstParam, dstStrideParam);
3e499f53 2257//printf("sws: slice %d %d\n", srcSliceY, srcSliceH);
5bf01354 2258 return c->swScale(c, src, srcStride, srcSliceY, srcSliceH, dst, dstStride);
d4e24275 2259}
d4e24275
MN
2260
2261/**
28bf81c9
MN
2262 * returns a normalized gaussian curve used to filter stuff
2263 * quality=3 is high quality, lowwer is lowwer quality
2264 */
c7f822d9 2265
d4e24275 2266SwsVector *sws_getGaussianVec(double variance, double quality){
28bf81c9
MN
2267 const int length= (int)(variance*quality + 0.5) | 1;
2268 int i;
2269 double *coeff= memalign(sizeof(double), length*sizeof(double));
2270 double middle= (length-1)*0.5;
c7f822d9
MN
2271 SwsVector *vec= malloc(sizeof(SwsVector));
2272
2273 vec->coeff= coeff;
2274 vec->length= length;
28bf81c9
MN
2275
2276 for(i=0; i<length; i++)
2277 {
2278 double dist= i-middle;
2279 coeff[i]= exp( -dist*dist/(2*variance*variance) ) / sqrt(2*variance*PI);
2280 }
2281
d4e24275 2282 sws_normalizeVec(vec, 1.0);
c7f822d9
MN
2283
2284 return vec;
28bf81c9
MN
2285}
2286
d4e24275 2287SwsVector *sws_getConstVec(double c, int length){
5521b193
MN
2288 int i;
2289 double *coeff= memalign(sizeof(double), length*sizeof(double));
2290 SwsVector *vec= malloc(sizeof(SwsVector));
2291
2292 vec->coeff= coeff;
2293 vec->length= length;
2294
2295 for(i=0; i<length; i++)
2296 coeff[i]= c;
2297
2298 return vec;
2299}
2300
2301
d4e24275 2302SwsVector *sws_getIdentityVec(void){
c7f822d9
MN
2303 double *coeff= memalign(sizeof(double), sizeof(double));
2304 SwsVector *vec= malloc(sizeof(SwsVector));
2305 coeff[0]= 1.0;
2306
2307 vec->coeff= coeff;
2308 vec->length= 1;
2309
2310 return vec;
2311}
2312
d4e24275 2313void sws_normalizeVec(SwsVector *a, double height){
28bf81c9
MN
2314 int i;
2315 double sum=0;
2316 double inv;
2317
c7f822d9
MN
2318 for(i=0; i<a->length; i++)
2319 sum+= a->coeff[i];
28bf81c9
MN
2320
2321 inv= height/sum;
2322
c7f822d9 2323 for(i=0; i<a->length; i++)
8664c807 2324 a->coeff[i]*= inv;
28bf81c9
MN
2325}
2326
d4e24275 2327void sws_scaleVec(SwsVector *a, double scalar){
c7f822d9
MN
2328 int i;
2329
2330 for(i=0; i<a->length; i++)
2331 a->coeff[i]*= scalar;
2332}
2333
d4e24275 2334static SwsVector *sws_getConvVec(SwsVector *a, SwsVector *b){
c7f822d9 2335 int length= a->length + b->length - 1;
28bf81c9
MN
2336 double *coeff= memalign(sizeof(double), length*sizeof(double));
2337 int i, j;
c7f822d9
MN
2338 SwsVector *vec= malloc(sizeof(SwsVector));
2339
2340 vec->coeff= coeff;
2341 vec->length= length;
28bf81c9
MN
2342
2343 for(i=0; i<length; i++) coeff[i]= 0.0;
2344
c7f822d9 2345 for(i=0; i<a->length; i++)
28bf81c9 2346 {
c7f822d9 2347 for(j=0; j<b->length; j++)
28bf81c9 2348 {
c7f822d9 2349 coeff[i+j]+= a->coeff[i]*b->coeff[j];
28bf81c9
MN
2350 }
2351 }
2352
c7f822d9 2353 return vec;
28bf81c9
MN
2354}
2355
d4e24275 2356static SwsVector *sws_sumVec(SwsVector *a, SwsVector *b){
c7f822d9 2357 int length= MAX(a->length, b->length);
28bf81c9
MN
2358 double *coeff= memalign(sizeof(double), length*sizeof(double));
2359 int i;
c7f822d9
MN
2360 SwsVector *vec= malloc(sizeof(SwsVector));
2361
2362 vec->coeff= coeff;
2363 vec->length= length;
28bf81c9
MN
2364
2365 for(i=0; i<length; i++) coeff[i]= 0.0;
2366
c7f822d9
MN
2367 for(i=0; i<a->length; i++) coeff[i + (length-1)/2 - (a->length-1)/2]+= a->coeff[i];
2368 for(i=0; i<b->length; i++) coeff[i + (length-1)/2 - (b->length-1)/2]+= b->coeff[i];
2369
2370 return vec;
28bf81c9 2371}
c7f822d9 2372
d4e24275 2373static SwsVector *sws_diffVec(SwsVector *a, SwsVector *b){
c7f822d9
MN
2374 int length= MAX(a->length, b->length);
2375 double *coeff= memalign(sizeof(double), length*sizeof(double));
2376 int i;
2377 SwsVector *vec= malloc(sizeof(SwsVector));
2378
2379 vec->coeff= coeff;
2380 vec->length= length;
2381
2382 for(i=0; i<length; i++) coeff[i]= 0.0;
2383
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;
2388}
2389
2390/* shift left / or right if "shift" is negative */
d4e24275 2391static SwsVector *sws_getShiftedVec(SwsVector *a, int shift){
c7f822d9
MN
2392 int length= a->length + ABS(shift)*2;
2393 double *coeff= memalign(sizeof(double), length*sizeof(double));
ff7ba856 2394 int i;
c7f822d9
MN
2395 SwsVector *vec= malloc(sizeof(SwsVector));
2396
2397 vec->coeff= coeff;
2398 vec->length= length;
2399
2400 for(i=0; i<length; i++) coeff[i]= 0.0;
2401
2402 for(i=0; i<a->length; i++)
2403 {
2404 coeff[i + (length-1)/2 - (a->length-1)/2 - shift]= a->coeff[i];
2405 }
2406
2407 return vec;
2408}
2409
d4e24275
MN
2410void sws_shiftVec(SwsVector *a, int shift){
2411 SwsVector *shifted= sws_getShiftedVec(a, shift);
5cebb24b
MN
2412 free(a->coeff);
2413 a->coeff= shifted->coeff;
2414 a->length= shifted->length;
2415 free(shifted);
2416}
2417
d4e24275
MN
2418void sws_addVec(SwsVector *a, SwsVector *b){
2419 SwsVector *sum= sws_sumVec(a, b);
5cebb24b
MN
2420 free(a->coeff);
2421 a->coeff= sum->coeff;
2422 a->length= sum->length;
2423 free(sum);
2424}
2425
d4e24275
MN
2426void sws_subVec(SwsVector *a, SwsVector *b){
2427 SwsVector *diff= sws_diffVec(a, b);
5cebb24b
MN
2428 free(a->coeff);
2429 a->coeff= diff->coeff;
2430 a->length= diff->length;
2431 free(diff);
2432}
2433
d4e24275
MN
2434void sws_convVec(SwsVector *a, SwsVector *b){
2435 SwsVector *conv= sws_getConvVec(a, b);
2436 free(a->coeff);
5cebb24b
MN
2437 a->coeff= conv->coeff;
2438 a->length= conv->length;
2439 free(conv);
2440}
2441
d4e24275 2442SwsVector *sws_cloneVec(SwsVector *a){
5cebb24b
MN
2443 double *coeff= memalign(sizeof(double), a->length*sizeof(double));
2444 int i;
2445 SwsVector *vec= malloc(sizeof(SwsVector));
2446
2447 vec->coeff= coeff;
2448 vec->length= a->length;
2449
2450 for(i=0; i<a->length; i++) coeff[i]= a->coeff[i];
2451
2452 return vec;
2453}
2454
d4e24275 2455void sws_printVec(SwsVector *a){
c7f822d9
MN
2456 int i;
2457 double max=0;
2458 double min=0;
2459 double range;
2460
2461 for(i=0; i<a->length; i++)
2462 if(a->coeff[i]>max) max= a->coeff[i];
2463
2464 for(i=0; i<a->length; i++)
2465 if(a->coeff[i]<min) min= a->coeff[i];
2466
2467 range= max - min;
2468
2469 for(i=0; i<a->length; i++)
2470 {
2471 int x= (int)((a->coeff[i]-min)*60.0/range +0.5);
0d9f3d85
A
2472 MSG_DBG2("%1.3f ", a->coeff[i]);
2473 for(;x>0; x--) MSG_DBG2(" ");
2474 MSG_DBG2("|\n");
c7f822d9
MN
2475 }
2476}
2477
d4e24275 2478void sws_freeVec(SwsVector *a){
c7f822d9
MN
2479 if(!a) return;
2480 if(a->coeff) free(a->coeff);
2481 a->coeff=NULL;
2482 a->length=0;
2483 free(a);
2484}
2485
d4e24275 2486void sws_freeContext(SwsContext *c){
c7f822d9 2487 int i;
c7f822d9
MN
2488 if(!c) return;
2489
2490 if(c->lumPixBuf)
2491 {
6c7506de 2492 for(i=0; i<c->vLumBufSize; i++)
c7f822d9
MN
2493 {
2494 if(c->lumPixBuf[i]) free(c->lumPixBuf[i]);
2495 c->lumPixBuf[i]=NULL;
2496 }
2497 free(c->lumPixBuf);
2498 c->lumPixBuf=NULL;
2499 }
2500
2501 if(c->chrPixBuf)
2502 {
6c7506de 2503 for(i=0; i<c->vChrBufSize; i++)
c7f822d9
MN
2504 {
2505 if(c->chrPixBuf[i]) free(c->chrPixBuf[i]);
2506 c->chrPixBuf[i]=NULL;
2507 }
2508 free(c->chrPixBuf);
2509 c->chrPixBuf=NULL;
2510 }
2511
2512 if(c->vLumFilter) free(c->vLumFilter);
2513 c->vLumFilter = NULL;
2514 if(c->vChrFilter) free(c->vChrFilter);
2515 c->vChrFilter = NULL;
2516 if(c->hLumFilter) free(c->hLumFilter);
2517 c->hLumFilter = NULL;
2518 if(c->hChrFilter) free(c->hChrFilter);
2519 c->hChrFilter = NULL;
2520
2521 if(c->vLumFilterPos) free(c->vLumFilterPos);
2522 c->vLumFilterPos = NULL;
2523 if(c->vChrFilterPos) free(c->vChrFilterPos);
2524 c->vChrFilterPos = NULL;
2525 if(c->hLumFilterPos) free(c->hLumFilterPos);
2526 c->hLumFilterPos = NULL;
2527 if(c->hChrFilterPos) free(c->hChrFilterPos);
2528 c->hChrFilterPos = NULL;
2529
b7dc6f66
MN
2530 if(c->lumMmx2Filter) free(c->lumMmx2Filter);
2531 c->lumMmx2Filter=NULL;
2532 if(c->chrMmx2Filter) free(c->chrMmx2Filter);
2533 c->chrMmx2Filter=NULL;
2534 if(c->lumMmx2FilterPos) free(c->lumMmx2FilterPos);
2535 c->lumMmx2FilterPos=NULL;
2536 if(c->chrMmx2FilterPos) free(c->chrMmx2FilterPos);
2537 c->chrMmx2FilterPos=NULL;
cf7d1c1a
MN
2538 if(c->yuvTable) free(c->yuvTable);
2539 c->yuvTable=NULL;
b7dc6f66 2540
c7f822d9
MN
2541 free(c);
2542}
2543