max_b_frame=4 bugfix
[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;
a86c461c
MN
961 case 7: *flags|= SWS_GAUSS; break;
962 case 8: *flags|= SWS_SINC; break;
963 case 9: *flags|= SWS_LANCZOS; break;
964 case 10:*flags|= SWS_SPLINE; break;
5859233b 965 default:*flags|= SWS_BILINEAR; break;
28bf81c9 966 }
5859233b
MN
967
968 *srcFilterParam= &src_filter;
969 *dstFilterParam= NULL;
970}
971
972// will use sws_flags & src_filter (from cmd line)
973SwsContext *getSwsContextFromCmdLine(int srcW, int srcH, int srcFormat, int dstW, int dstH, int dstFormat)
974{
975 int flags;
976 SwsFilter *dstFilterParam, *srcFilterParam;
977 swsGetFlagsAndFilterFromCmdLine(&flags, &srcFilterParam, &dstFilterParam);
28bf81c9 978
5859233b 979 return getSwsContext(srcW, srcH, srcFormat, dstW, dstH, dstFormat, flags, srcFilterParam, dstFilterParam);
28bf81c9
MN
980}
981
a86c461c
MN
982static double getSplineCoeff(double a, double b, double c, double d, double dist)
983{
984// printf("%f %f %f %f %f\n", a,b,c,d,dist);
985 if(dist<=1.0) return ((d*dist + c)*dist + b)*dist +a;
986 else return getSplineCoeff( 0.0,
987 b+ 2.0*c + 3.0*d,
988 c + 3.0*d,
989 -b- 3.0*c - 6.0*d,
990 dist-1.0);
991}
6c7506de 992
c7f822d9
MN
993static inline void initFilter(int16_t **outFilter, int16_t **filterPos, int *outFilterSize, int xInc,
994 int srcW, int dstW, int filterAlign, int one, int flags,
995 SwsVector *srcFilter, SwsVector *dstFilter)
28bf81c9
MN
996{
997 int i;
c7f822d9
MN
998 int filterSize;
999 int filter2Size;
1000 int minFilterSize;
1001 double *filter=NULL;
1002 double *filter2=NULL;
28bf81c9
MN
1003#ifdef ARCH_X86
1004 if(gCpuCaps.hasMMX)
1005 asm volatile("emms\n\t"::: "memory"); //FIXME this shouldnt be required but it IS (even for non mmx versions)
726a959a 1006#endif
31190492 1007
adeaecb9 1008 // Note the +1 is for the MMXscaler which reads over the end
6c7506de 1009 *filterPos = (int16_t*)memalign(8, (dstW+1)*sizeof(int16_t));
6c7506de 1010
28bf81c9
MN
1011 if(ABS(xInc - 0x10000) <10) // unscaled
1012 {
1013 int i;
c7f822d9
MN
1014 filterSize= 1;
1015 filter= (double*)memalign(8, dstW*sizeof(double)*filterSize);
1016 for(i=0; i<dstW*filterSize; i++) filter[i]=0;
28bf81c9
MN
1017
1018 for(i=0; i<dstW; i++)
1019 {
c7f822d9
MN
1020 filter[i*filterSize]=1;
1021 (*filterPos)[i]=i;
28bf81c9
MN
1022 }
1023
1024 }
ff7ba856
MN
1025 else if(flags&SWS_POINT) // lame looking point sampling mode
1026 {
1027 int i;
1028 int xDstInSrc;
1029 filterSize= 1;
1030 filter= (double*)memalign(8, dstW*sizeof(double)*filterSize);
1031
1032 xDstInSrc= xInc/2 - 0x8000;
1033 for(i=0; i<dstW; i++)
1034 {
8a01d20c 1035 int xx= (xDstInSrc - ((filterSize-1)<<15) + (1<<15))>>16;
ff7ba856
MN
1036
1037 (*filterPos)[i]= xx;
1038 filter[i]= 1.0;
1039 xDstInSrc+= xInc;
1040 }
1041 }
a86c461c 1042 else if((xInc <= (1<<16) && (flags&SWS_AREA)) || (flags&SWS_FAST_BILINEAR)) // bilinear upscale
28bf81c9
MN
1043 {
1044 int i;
1045 int xDstInSrc;
c7f822d9
MN
1046 if (flags&SWS_BICUBIC) filterSize= 4;
1047 else if(flags&SWS_X ) filterSize= 4;
d8863d37 1048 else filterSize= 2; // SWS_BILINEAR / SWS_AREA
c7f822d9 1049 filter= (double*)memalign(8, dstW*sizeof(double)*filterSize);
28bf81c9
MN
1050
1051 xDstInSrc= xInc/2 - 0x8000;
1052 for(i=0; i<dstW; i++)
1053 {
8a01d20c 1054 int xx= (xDstInSrc - ((filterSize-1)<<15) + (1<<15))>>16;
28bf81c9
MN
1055 int j;
1056
c7f822d9 1057 (*filterPos)[i]= xx;
d8863d37 1058 //Bilinear upscale / linear interpolate / Area averaging
c7f822d9 1059 for(j=0; j<filterSize; j++)
28bf81c9
MN
1060 {
1061 double d= ABS((xx<<16) - xDstInSrc)/(double)(1<<16);
1062 double coeff= 1.0 - d;
1063 if(coeff<0) coeff=0;
c7f822d9 1064 filter[i*filterSize + j]= coeff;
28bf81c9
MN
1065 xx++;
1066 }
28bf81c9
MN
1067 xDstInSrc+= xInc;
1068 }
1069 }
a86c461c 1070 else
28bf81c9 1071 {
a86c461c
MN
1072 double xDstInSrc;
1073 double sizeFactor, filterSizeInSrc;
1074 const double xInc1= (double)xInc / (double)(1<<16);
1075 int param= (flags&SWS_PARAM_MASK)>>SWS_PARAM_SHIFT;
1076
1077 if (flags&SWS_BICUBIC) sizeFactor= 4.0;
1078 else if(flags&SWS_X) sizeFactor= 8.0;
1079 else if(flags&SWS_AREA) sizeFactor= 1.0; //downscale only, for upscale it is bilinear
1080 else if(flags&SWS_GAUSS) sizeFactor= 8.0; // infinite ;)
1081 else if(flags&SWS_LANCZOS) sizeFactor= param ? 2.0*param : 6.0;
93768378 1082 else if(flags&SWS_SINC) sizeFactor= 20.0; // infinite ;)
a86c461c
MN
1083 else if(flags&SWS_SPLINE) sizeFactor= 20.0; // infinite ;)
1084 else if(flags&SWS_BILINEAR) sizeFactor= 2.0;
93768378
MN
1085 else {
1086 sizeFactor= 0.0; //GCC warning killer
1087 ASSERT(0)
1088 }
a86c461c
MN
1089
1090 if(xInc1 <= 1.0) filterSizeInSrc= sizeFactor; // upscale
1091 else filterSizeInSrc= sizeFactor*srcW / (double)dstW;
81b7c056 1092
a86c461c
MN
1093 filterSize= (int)ceil(1 + filterSizeInSrc); // will be reduced later if possible
1094 if(filterSize > srcW-2) filterSize=srcW-2;
28bf81c9 1095
a86c461c
MN
1096 filter= (double*)memalign(16, dstW*sizeof(double)*filterSize);
1097
1098 xDstInSrc= xInc1 / 2.0 - 0.5;
28bf81c9
MN
1099 for(i=0; i<dstW; i++)
1100 {
a86c461c 1101 int xx= (int)(xDstInSrc - (filterSize-1)*0.5 + 0.5);
28bf81c9 1102 int j;
c7f822d9
MN
1103 (*filterPos)[i]= xx;
1104 for(j=0; j<filterSize; j++)
28bf81c9 1105 {
a86c461c 1106 double d= ABS(xx - xDstInSrc)/filterSizeInSrc*sizeFactor;
28bf81c9 1107 double coeff;
a86c461c 1108 if(flags & SWS_BICUBIC)
28bf81c9 1109 {
a86c461c
MN
1110 double A= param ? -param*0.01 : -0.60;
1111
28bf81c9
MN
1112 // Equation is from VirtualDub
1113 if(d<1.0)
1114 coeff = (1.0 - (A+3.0)*d*d + (A+2.0)*d*d*d);
1115 else if(d<2.0)
1116 coeff = (-4.0*A + 8.0*A*d - 5.0*A*d*d + A*d*d*d);
1117 else
1118 coeff=0.0;
1119 }
a86c461c
MN
1120/* else if(flags & SWS_X)
1121 {
1122 double p= param ? param*0.01 : 0.3;
1123 coeff = d ? sin(d*PI)/(d*PI) : 1.0;
1124 coeff*= pow(2.0, - p*d*d);
1125 }*/
1126 else if(flags & SWS_X)
1127 {
1128 double A= param ? param*0.1 : 1.0;
1129
1130 if(d<1.0)
1131 coeff = cos(d*PI);
1132 else
1133 coeff=-1.0;
1134 if(coeff<0.0) coeff= -pow(-coeff, A);
1135 else coeff= pow( coeff, A);
1136 coeff= coeff*0.5 + 0.5;
1137 }
d8863d37 1138 else if(flags & SWS_AREA)
28bf81c9 1139 {
a86c461c 1140 double srcPixelSize= 1.0/xInc1;
d8863d37
MN
1141 if(d + srcPixelSize/2 < 0.5) coeff= 1.0;
1142 else if(d - srcPixelSize/2 < 0.5) coeff= (0.5-d)/srcPixelSize + 0.5;
1143 else coeff=0.0;
1144 }
a86c461c
MN
1145 else if(flags & SWS_GAUSS)
1146 {
1147 double p= param ? param*0.1 : 3.0;
1148 coeff = pow(2.0, - p*d*d);
1149 }
1150 else if(flags & SWS_SINC)
1151 {
1152 coeff = d ? sin(d*PI)/(d*PI) : 1.0;
1153 }
1154 else if(flags & SWS_LANCZOS)
1155 {
1156 double p= param ? param : 3.0;
1157 coeff = d ? sin(d*PI)*sin(d*PI/p)/(d*d*PI*PI/p) : 1.0;
1158 if(d>p) coeff=0;
1159 }
1160 else if(flags & SWS_BILINEAR)
28bf81c9
MN
1161 {
1162 coeff= 1.0 - d;
1163 if(coeff<0) coeff=0;
1164 }
a86c461c
MN
1165 else if(flags & SWS_SPLINE)
1166 {
1167 double p=-2.196152422706632;
1168 coeff = getSplineCoeff(1.0, 0.0, p, -p-1.0, d);
1169 }
93768378
MN
1170 else {
1171 coeff= 0.0; //GCC warning killer
1172 ASSERT(0)
1173 }
a86c461c 1174
c7f822d9 1175 filter[i*filterSize + j]= coeff;
28bf81c9
MN
1176 xx++;
1177 }
a86c461c 1178 xDstInSrc+= xInc1;
28bf81c9
MN
1179 }
1180 }
1181
c7f822d9
MN
1182 /* apply src & dst Filter to filter -> filter2
1183 free(filter);
1184 */
81b7c056 1185 ASSERT(filterSize>0)
c7f822d9
MN
1186 filter2Size= filterSize;
1187 if(srcFilter) filter2Size+= srcFilter->length - 1;
1188 if(dstFilter) filter2Size+= dstFilter->length - 1;
81b7c056 1189 ASSERT(filter2Size>0)
c7f822d9
MN
1190 filter2= (double*)memalign(8, filter2Size*dstW*sizeof(double));
1191
1192 for(i=0; i<dstW; i++)
1193 {
1194 int j;
1195 SwsVector scaleFilter;
1196 SwsVector *outVec;
1197
1198 scaleFilter.coeff= filter + i*filterSize;
1199 scaleFilter.length= filterSize;
1200
5cebb24b 1201 if(srcFilter) outVec= getConvVec(srcFilter, &scaleFilter);
c7f822d9
MN
1202 else outVec= &scaleFilter;
1203
1204 ASSERT(outVec->length == filter2Size)
1205 //FIXME dstFilter
1206
1207 for(j=0; j<outVec->length; j++)
1208 {
1209 filter2[i*filter2Size + j]= outVec->coeff[j];
1210 }
1211
1212 (*filterPos)[i]+= (filterSize-1)/2 - (filter2Size-1)/2;
1213
1214 if(outVec != &scaleFilter) freeVec(outVec);
1215 }
1216 free(filter); filter=NULL;
1217
1218 /* try to reduce the filter-size (step1 find size and shift left) */
1219 // Assume its near normalized (*0.5 or *2.0 is ok but * 0.001 is not)
1220 minFilterSize= 0;
1221 for(i=dstW-1; i>=0; i--)
1222 {
1223 int min= filter2Size;
1224 int j;
1225 double cutOff=0.0;
1226
1227 /* get rid off near zero elements on the left by shifting left */
1228 for(j=0; j<filter2Size; j++)
1229 {
1230 int k;
1231 cutOff += ABS(filter2[i*filter2Size]);
1232
1233 if(cutOff > SWS_MAX_REDUCE_CUTOFF) break;
1234
1235 /* preserve Monotonicity because the core cant handle the filter otherwise */
1236 if(i<dstW-1 && (*filterPos)[i] >= (*filterPos)[i+1]) break;
1237
1238 // Move filter coeffs left
1239 for(k=1; k<filter2Size; k++)
1240 filter2[i*filter2Size + k - 1]= filter2[i*filter2Size + k];
1241 filter2[i*filter2Size + k - 1]= 0.0;
1242 (*filterPos)[i]++;
1243 }
1244
1245 cutOff=0.0;
1246 /* count near zeros on the right */
1247 for(j=filter2Size-1; j>0; j--)
1248 {
1249 cutOff += ABS(filter2[i*filter2Size + j]);
1250
1251 if(cutOff > SWS_MAX_REDUCE_CUTOFF) break;
1252 min--;
1253 }
1254
1255 if(min>minFilterSize) minFilterSize= min;
1256 }
1257
81b7c056 1258 ASSERT(minFilterSize > 0)
6c7506de 1259 filterSize= (minFilterSize +(filterAlign-1)) & (~(filterAlign-1));
81b7c056 1260 ASSERT(filterSize > 0)
6c7506de
MN
1261 filter= (double*)memalign(8, filterSize*dstW*sizeof(double));
1262 *outFilterSize= filterSize;
1263
4a53a912 1264 if(flags&SWS_PRINT_INFO)
0d9f3d85 1265 MSG_INFO("SwScaler: reducing / aligning filtersize %d -> %d\n", filter2Size, filterSize);
c7f822d9
MN
1266 /* try to reduce the filter-size (step2 reduce it) */
1267 for(i=0; i<dstW; i++)
1268 {
1269 int j;
1270
6c7506de
MN
1271 for(j=0; j<filterSize; j++)
1272 {
1273 if(j>=filter2Size) filter[i*filterSize + j]= 0.0;
1274 else filter[i*filterSize + j]= filter2[i*filter2Size + j];
1275 }
c7f822d9 1276 }
6c7506de
MN
1277 free(filter2); filter2=NULL;
1278
c7f822d9
MN
1279
1280 //FIXME try to align filterpos if possible
1281
28bf81c9
MN
1282 //fix borders
1283 for(i=0; i<dstW; i++)
1284 {
1285 int j;
c7f822d9 1286 if((*filterPos)[i] < 0)
28bf81c9
MN
1287 {
1288 // Move filter coeffs left to compensate for filterPos
6c7506de 1289 for(j=1; j<filterSize; j++)
28bf81c9 1290 {
c7f822d9 1291 int left= MAX(j + (*filterPos)[i], 0);
6c7506de
MN
1292 filter[i*filterSize + left] += filter[i*filterSize + j];
1293 filter[i*filterSize + j]=0;
28bf81c9 1294 }
c7f822d9 1295 (*filterPos)[i]= 0;
28bf81c9
MN
1296 }
1297
6c7506de 1298 if((*filterPos)[i] + filterSize > srcW)
28bf81c9 1299 {
6c7506de 1300 int shift= (*filterPos)[i] + filterSize - srcW;
28bf81c9 1301 // Move filter coeffs right to compensate for filterPos
6c7506de 1302 for(j=filterSize-2; j>=0; j--)
28bf81c9 1303 {
6c7506de
MN
1304 int right= MIN(j + shift, filterSize-1);
1305 filter[i*filterSize +right] += filter[i*filterSize +j];
1306 filter[i*filterSize +j]=0;
28bf81c9 1307 }
6c7506de 1308 (*filterPos)[i]= srcW - filterSize;
28bf81c9
MN
1309 }
1310 }
1311
6c7506de
MN
1312 // Note the +1 is for the MMXscaler which reads over the end
1313 *outFilter= (int16_t*)memalign(8, *outFilterSize*(dstW+1)*sizeof(int16_t));
1314 memset(*outFilter, 0, *outFilterSize*(dstW+1)*sizeof(int16_t));
c7f822d9
MN
1315
1316 /* Normalize & Store in outFilter */
28bf81c9
MN
1317 for(i=0; i<dstW; i++)
1318 {
1319 int j;
1320 double sum=0;
1321 double scale= one;
6c7506de 1322 for(j=0; j<filterSize; j++)
28bf81c9 1323 {
6c7506de 1324 sum+= filter[i*filterSize + j];
28bf81c9
MN
1325 }
1326 scale/= sum;
93768378 1327 for(j=0; j<*outFilterSize; j++)
28bf81c9 1328 {
6c7506de 1329 (*outFilter)[i*(*outFilterSize) + j]= (int)(filter[i*filterSize + j]*scale);
28bf81c9
MN
1330 }
1331 }
adeaecb9
MN
1332
1333 (*filterPos)[dstW]= (*filterPos)[dstW-1]; // the MMX scaler will read over the end
1334 for(i=0; i<*outFilterSize; i++)
1335 {
1336 int j= dstW*(*outFilterSize);
1337 (*outFilter)[j + i]= (*outFilter)[j + i - (*outFilterSize)];
1338 }
c7f822d9 1339
6c7506de 1340 free(filter);
7630f2e0 1341}
31190492 1342
28bf81c9 1343#ifdef ARCH_X86
b7dc6f66 1344static void initMMX2HScaler(int dstW, int xInc, uint8_t *funnyCode, int16_t *filter, int32_t *filterPos, int numSplits)
28bf81c9 1345{
b7dc6f66
MN
1346 uint8_t *fragmentA;
1347 int imm8OfPShufW1A;
1348 int imm8OfPShufW2A;
1349 int fragmentLengthA;
1350 uint8_t *fragmentB;
1351 int imm8OfPShufW1B;
1352 int imm8OfPShufW2B;
1353 int fragmentLengthB;
1354 int fragmentPos;
28bf81c9
MN
1355
1356 int xpos, i;
1357
1358 // create an optimized horizontal scaling routine
1359
1360 //code fragment
1361
1362 asm volatile(
1363 "jmp 9f \n\t"
1364 // Begin
1365 "0: \n\t"
b7dc6f66
MN
1366 "movq (%%edx, %%eax), %%mm3 \n\t"
1367 "movd (%%ecx, %%esi), %%mm0 \n\t"
1368 "movd 1(%%ecx, %%esi), %%mm1 \n\t"
1369 "punpcklbw %%mm7, %%mm1 \n\t"
1370 "punpcklbw %%mm7, %%mm0 \n\t"
28bf81c9
MN
1371 "pshufw $0xFF, %%mm1, %%mm1 \n\t"
1372 "1: \n\t"
28bf81c9
MN
1373 "pshufw $0xFF, %%mm0, %%mm0 \n\t"
1374 "2: \n\t"
28bf81c9 1375 "psubw %%mm1, %%mm0 \n\t"
b7dc6f66 1376 "movl 8(%%ebx, %%eax), %%esi \n\t"
28bf81c9 1377 "pmullw %%mm3, %%mm0 \n\t"
28bf81c9
MN
1378 "psllw $7, %%mm1 \n\t"
1379 "paddw %%mm1, %%mm0 \n\t"
1380
1381 "movq %%mm0, (%%edi, %%eax) \n\t"
1382
1383 "addl $8, %%eax \n\t"
1384 // End
1385 "9: \n\t"
1386// "int $3\n\t"
1387 "leal 0b, %0 \n\t"
1388 "leal 1b, %1 \n\t"
1389 "leal 2b, %2 \n\t"
1390 "decl %1 \n\t"
1391 "decl %2 \n\t"
1392 "subl %0, %1 \n\t"
1393 "subl %0, %2 \n\t"
1394 "leal 9b, %3 \n\t"
1395 "subl %0, %3 \n\t"
b7dc6f66
MN
1396
1397
1398 :"=r" (fragmentA), "=r" (imm8OfPShufW1A), "=r" (imm8OfPShufW2A),
1399 "=r" (fragmentLengthA)
28bf81c9
MN
1400 );
1401
b7dc6f66
MN
1402 asm volatile(
1403 "jmp 9f \n\t"
1404 // Begin
1405 "0: \n\t"
1406 "movq (%%edx, %%eax), %%mm3 \n\t"
1407 "movd (%%ecx, %%esi), %%mm0 \n\t"
1408 "punpcklbw %%mm7, %%mm0 \n\t"
1409 "pshufw $0xFF, %%mm0, %%mm1 \n\t"
1410 "1: \n\t"
1411 "pshufw $0xFF, %%mm0, %%mm0 \n\t"
1412 "2: \n\t"
1413 "psubw %%mm1, %%mm0 \n\t"
1414 "movl 8(%%ebx, %%eax), %%esi \n\t"
1415 "pmullw %%mm3, %%mm0 \n\t"
1416 "psllw $7, %%mm1 \n\t"
1417 "paddw %%mm1, %%mm0 \n\t"
1418
1419 "movq %%mm0, (%%edi, %%eax) \n\t"
28bf81c9 1420
b7dc6f66
MN
1421 "addl $8, %%eax \n\t"
1422 // End
1423 "9: \n\t"
1424// "int $3\n\t"
1425 "leal 0b, %0 \n\t"
1426 "leal 1b, %1 \n\t"
1427 "leal 2b, %2 \n\t"
1428 "decl %1 \n\t"
1429 "decl %2 \n\t"
1430 "subl %0, %1 \n\t"
1431 "subl %0, %2 \n\t"
1432 "leal 9b, %3 \n\t"
1433 "subl %0, %3 \n\t"
1434
1435
1436 :"=r" (fragmentB), "=r" (imm8OfPShufW1B), "=r" (imm8OfPShufW2B),
1437 "=r" (fragmentLengthB)
1438 );
1439
1440 xpos= 0; //lumXInc/2 - 0x8000; // difference between pixel centers
1441 fragmentPos=0;
1442
1443 for(i=0; i<dstW/numSplits; i++)
28bf81c9
MN
1444 {
1445 int xx=xpos>>16;
1446
1447 if((i&3) == 0)
1448 {
1449 int a=0;
1450 int b=((xpos+xInc)>>16) - xx;
1451 int c=((xpos+xInc*2)>>16) - xx;
1452 int d=((xpos+xInc*3)>>16) - xx;
1453
b7dc6f66
MN
1454 filter[i ] = (( xpos & 0xFFFF) ^ 0xFFFF)>>9;
1455 filter[i+1] = (((xpos+xInc ) & 0xFFFF) ^ 0xFFFF)>>9;
1456 filter[i+2] = (((xpos+xInc*2) & 0xFFFF) ^ 0xFFFF)>>9;
1457 filter[i+3] = (((xpos+xInc*3) & 0xFFFF) ^ 0xFFFF)>>9;
1458 filterPos[i/2]= xx;
1459
1460 if(d+1<4)
1461 {
1462 int maxShift= 3-(d+1);
1463 int shift=0;
1464
1465 memcpy(funnyCode + fragmentPos, fragmentB, fragmentLengthB);
1466
1467 funnyCode[fragmentPos + imm8OfPShufW1B]=
1468 (a+1) | ((b+1)<<2) | ((c+1)<<4) | ((d+1)<<6);
1469 funnyCode[fragmentPos + imm8OfPShufW2B]=
1470 a | (b<<2) | (c<<4) | (d<<6);
1471
1472 if(i+3>=dstW) shift=maxShift; //avoid overread
1473 else if((filterPos[i/2]&3) <= maxShift) shift=filterPos[i/2]&3; //Align
1474
1475 if(shift && i>=shift)
1476 {
1477 funnyCode[fragmentPos + imm8OfPShufW1B]+= 0x55*shift;
1478 funnyCode[fragmentPos + imm8OfPShufW2B]+= 0x55*shift;
1479 filterPos[i/2]-=shift;
1480 }
1481
1482 fragmentPos+= fragmentLengthB;
1483 }
1484 else
1485 {
1486 int maxShift= 3-d;
1487 int shift=0;
1488
1489 memcpy(funnyCode + fragmentPos, fragmentA, fragmentLengthA);
28bf81c9 1490
b7dc6f66
MN
1491 funnyCode[fragmentPos + imm8OfPShufW1A]=
1492 funnyCode[fragmentPos + imm8OfPShufW2A]=
1493 a | (b<<2) | (c<<4) | (d<<6);
28bf81c9 1494
b7dc6f66
MN
1495 if(i+4>=dstW) shift=maxShift; //avoid overread
1496 else if((filterPos[i/2]&3) <= maxShift) shift=filterPos[i/2]&3; //partial align
28bf81c9 1497
b7dc6f66
MN
1498 if(shift && i>=shift)
1499 {
1500 funnyCode[fragmentPos + imm8OfPShufW1A]+= 0x55*shift;
1501 funnyCode[fragmentPos + imm8OfPShufW2A]+= 0x55*shift;
1502 filterPos[i/2]-=shift;
1503 }
1504
1505 fragmentPos+= fragmentLengthA;
1506 }
1507
1508 funnyCode[fragmentPos]= RET;
28bf81c9
MN
1509 }
1510 xpos+=xInc;
1511 }
b7dc6f66 1512 filterPos[i/2]= xpos>>16; // needed to jump to the next part
28bf81c9
MN
1513}
1514#endif // ARCH_X86
1515
1516//FIXME remove
31190492 1517void SwScale_Init(){
28bf81c9
MN
1518}
1519
1520static void globalInit(){
31190492
A
1521 // generating tables:
1522 int i;
c1b0bfb4
MN
1523 for(i=0; i<768; i++){
1524 int c= MIN(MAX(i-256, 0), 255);
1525 clip_table[i]=c;
b18ea156 1526 }
c1b0bfb4 1527
28bf81c9
MN
1528cpuCaps= gCpuCaps;
1529
1530#ifdef RUNTIME_CPUDETECT
1531#ifdef CAN_COMPILE_X86_ASM
1532 // ordered per speed fasterst first
1533 if(gCpuCaps.hasMMX2)
1534 swScale= swScale_MMX2;
1535 else if(gCpuCaps.has3DNow)
7f56a527 1536 swScale= swScale_3DNow;
28bf81c9
MN
1537 else if(gCpuCaps.hasMMX)
1538 swScale= swScale_MMX;
1539 else
1540 swScale= swScale_C;
1541
1542#else
1543 swScale= swScale_C;
1544 cpuCaps.hasMMX2 = cpuCaps.hasMMX = cpuCaps.has3DNow = 0;
1545#endif
1546#else //RUNTIME_CPUDETECT
1547#ifdef HAVE_MMX2
1548 swScale= swScale_MMX2;
1549 cpuCaps.has3DNow = 0;
1550#elif defined (HAVE_3DNOW)
7f56a527 1551 swScale= swScale_3DNow;
28bf81c9
MN
1552 cpuCaps.hasMMX2 = 0;
1553#elif defined (HAVE_MMX)
1554 swScale= swScale_MMX;
1555 cpuCaps.hasMMX2 = cpuCaps.has3DNow = 0;
1556#else
1557 swScale= swScale_C;
1558 cpuCaps.hasMMX2 = cpuCaps.hasMMX = cpuCaps.has3DNow = 0;
1559#endif
1560#endif //!RUNTIME_CPUDETECT
31190492 1561}
7630f2e0 1562
0d9f3d85
A
1563static void PlanarToNV12Wrapper(SwsContext *c, uint8_t* src[], int srcStride[], int srcSliceY,
1564 int srcSliceH, uint8_t* dstParam[], int dstStride[]){
1565 uint8_t *dst=dstParam[0] + dstStride[0]*srcSliceY;
1566 /* Copy Y plane */
1567 if(dstStride[0]==srcStride[0])
1568 memcpy(dst, src[0], srcSliceH*dstStride[0]);
1569 else
1570 {
1571 int i;
1572 uint8_t *srcPtr= src[0];
1573 uint8_t *dstPtr= dst;
1574 for(i=0; i<srcSliceH; i++)
1575 {
1576 memcpy(dstPtr, srcPtr, srcStride[0]);
1577 srcPtr+= srcStride[0];
1578 dstPtr+= dstStride[0];
1579 }
1580 }
1581 dst = dstParam[1] + dstStride[1]*srcSliceY;
1582 if(c->srcFormat==IMGFMT_YV12)
1583 interleaveBytes( src[1],src[2],dst,c->srcW,srcSliceH,srcStride[1],srcStride[2],dstStride[0] );
1584 else /* I420 & IYUV */
1585 interleaveBytes( src[2],src[1],dst,c->srcW,srcSliceH,srcStride[2],srcStride[1],dstStride[0] );
1586}
1587
1588
37079906
MN
1589/* Warper functions for yuv2bgr */
1590static void planarYuvToBgr(SwsContext *c, uint8_t* src[], int srcStride[], int srcSliceY,
b6654a54
MN
1591 int srcSliceH, uint8_t* dstParam[], int dstStride[]){
1592 uint8_t *dst=dstParam[0] + dstStride[0]*srcSliceY;
37079906
MN
1593
1594 if(c->srcFormat==IMGFMT_YV12)
b6654a54 1595 yuv2rgb( dst,src[0],src[1],src[2],c->srcW,srcSliceH,dstStride[0],srcStride[0],srcStride[1] );
37079906 1596 else /* I420 & IYUV */
b6654a54
MN
1597 yuv2rgb( dst,src[0],src[2],src[1],c->srcW,srcSliceH,dstStride[0],srcStride[0],srcStride[1] );
1598}
1599
44c1035c 1600static void PlanarToYuy2Wrapper(SwsContext *c, uint8_t* src[], int srcStride[], int srcSliceY,
0d9f3d85
A
1601 int srcSliceH, uint8_t* dstParam[], int dstStride[]){
1602 uint8_t *dst=dstParam[0] + dstStride[0]*srcSliceY;
1603
1604 if(c->srcFormat==IMGFMT_YV12)
1605 yv12toyuy2( src[0],src[1],src[2],dst,c->srcW,srcSliceH,srcStride[0],srcStride[1],dstStride[0] );
1606 else /* I420 & IYUV */
1607 yv12toyuy2( src[0],src[2],src[1],dst,c->srcW,srcSliceH,srcStride[0],srcStride[1],dstStride[0] );
1608}
1609
e09d12f4
MN
1610/* {RGB,BGR}{15,16,24,32} -> {RGB,BGR}{15,16,24,32} */
1611static void rgb2rgbWrapper(SwsContext *c, uint8_t* src[], int srcStride[], int srcSliceY,
1612 int srcSliceH, uint8_t* dst[], int dstStride[]){
1613 const int srcFormat= c->srcFormat;
1614 const int dstFormat= c->dstFormat;
1615 const int srcBpp= ((srcFormat&0xFF) + 7)>>3;
1616 const int dstBpp= ((dstFormat&0xFF) + 7)>>3;
1617 const int srcId= (srcFormat&0xFF)>>2; // 1:0, 4:1, 8:2, 15:3, 16:4, 24:6, 32:8
1618 const int dstId= (dstFormat&0xFF)>>2;
1619 void (*conv)(const uint8_t *src, uint8_t *dst, unsigned src_size)=NULL;
1620
1621 /* BGR -> BGR */
1622 if(isBGR(srcFormat) && isBGR(dstFormat)){
1623 switch(srcId | (dstId<<4)){
1624 case 0x34: conv= rgb16to15; break;
1625 case 0x36: conv= rgb24to15; break;
1626 case 0x38: conv= rgb32to15; break;
1627 case 0x43: conv= rgb15to16; break;
1628 case 0x46: conv= rgb24to16; break;
1629 case 0x48: conv= rgb32to16; break;
1630 case 0x63: conv= rgb15to24; break;
1631 case 0x64: conv= rgb16to24; break;
1632 case 0x68: conv= rgb32to24; break;
1633 case 0x83: conv= rgb15to32; break;
1634 case 0x84: conv= rgb16to32; break;
1635 case 0x86: conv= rgb24to32; break;
1636 default: MSG_ERR("swScaler: internal error %s -> %s converter\n",
1637 vo_format_name(srcFormat), vo_format_name(dstFormat)); break;
b935781b 1638 }
e09d12f4
MN
1639 }else if(isBGR(srcFormat) && isRGB(dstFormat)){
1640 switch(srcId | (dstId<<4)){
1641 case 0x33: conv= rgb15tobgr15; break;
1642 case 0x34: conv= rgb16tobgr15; break;
1643 case 0x36: conv= rgb24tobgr15; break;
1644 case 0x38: conv= rgb32tobgr15; break;
1645 case 0x43: conv= rgb15tobgr16; break;
1646 case 0x44: conv= rgb16tobgr16; break;
1647 case 0x46: conv= rgb24tobgr16; break;
1648 case 0x48: conv= rgb32tobgr16; break;
1649 case 0x63: conv= rgb15tobgr24; break;
1650 case 0x64: conv= rgb16tobgr24; break;
1651 case 0x66: conv= rgb24tobgr24; break;
1652 case 0x68: conv= rgb32tobgr24; break;
1653 case 0x83: conv= rgb15tobgr32; break;
1654 case 0x84: conv= rgb16tobgr32; break;
1655 case 0x86: conv= rgb24tobgr32; break;
1656 case 0x88: conv= rgb32tobgr32; break;
1657 default: MSG_ERR("swScaler: internal error %s -> %s converter\n",
1658 vo_format_name(srcFormat), vo_format_name(dstFormat)); break;
0d9f3d85 1659 }
e09d12f4
MN
1660 }else if(isRGB(srcFormat) && isRGB(dstFormat)){
1661 switch(srcId | (dstId<<4)){
1662 case 0x34: conv= rgb16to15; break;
1663 case 0x36: conv= rgb24to15; break;
1664 case 0x38: conv= rgb32to15; break;
1665 case 0x43: conv= rgb15to16; break;
1666 case 0x46: conv= rgb24to16; break;
1667 case 0x48: conv= rgb32to16; break;
1668 case 0x63: conv= rgb15to24; break;
1669 case 0x64: conv= rgb16to24; break;
1670 case 0x68: conv= rgb32to24; break;
1671 case 0x83: conv= rgb15to32; break;
1672 case 0x84: conv= rgb16to32; break;
1673 case 0x86: conv= rgb24to32; break;
1674 default: MSG_ERR("swScaler: internal error %s -> %s converter\n",
1675 vo_format_name(srcFormat), vo_format_name(dstFormat)); break;
0d9f3d85 1676 }
e09d12f4
MN
1677 }else if(isRGB(srcFormat) && isBGR(dstFormat)){
1678 switch(srcId | (dstId<<4)){
1679 case 0x33: conv= rgb15tobgr15; break;
1680 case 0x34: conv= rgb16tobgr15; break;
1681 case 0x36: conv= rgb24tobgr15; break;
1682 case 0x38: conv= rgb32tobgr15; break;
1683 case 0x43: conv= rgb15tobgr16; break;
1684 case 0x44: conv= rgb16tobgr16; break;
1685 case 0x46: conv= rgb24tobgr16; break;
1686 case 0x48: conv= rgb32tobgr16; break;
1687 case 0x63: conv= rgb15tobgr24; break;
1688 case 0x64: conv= rgb16tobgr24; break;
1689 case 0x66: conv= rgb24tobgr24; break;
1690 case 0x68: conv= rgb32tobgr24; break;
1691 case 0x83: conv= rgb15tobgr32; break;
1692 case 0x84: conv= rgb16tobgr32; break;
1693 case 0x86: conv= rgb24tobgr32; break;
1694 case 0x88: conv= rgb32tobgr32; break;
1695 default: MSG_ERR("swScaler: internal error %s -> %s converter\n",
1696 vo_format_name(srcFormat), vo_format_name(dstFormat)); break;
0d9f3d85 1697 }
e09d12f4
MN
1698 }
1699 if(dstStride[0]*srcBpp == srcStride[0]*dstBpp)
1700 conv(src[0], dst[0] + dstStride[0]*srcSliceY, srcSliceH*srcStride[0]);
0d9f3d85
A
1701 else
1702 {
1703 int i;
1704 uint8_t *srcPtr= src[0];
1705 uint8_t *dstPtr= dst[0] + dstStride[0]*srcSliceY;
1706
1707 for(i=0; i<srcSliceH; i++)
1708 {
e09d12f4 1709 conv(srcPtr, dstPtr, c->srcW*srcBpp);
0d9f3d85
A
1710 srcPtr+= srcStride[0];
1711 dstPtr+= dstStride[0];
1712 }
1713 }
1714}
1715
ec22603f
MN
1716static void bgr24toyv12Wrapper(SwsContext *c, uint8_t* src[], int srcStride[], int srcSliceY,
1717 int srcSliceH, uint8_t* dst[], int dstStride[]){
1718
1719 rgb24toyv12(
1720 src[0],
1721 dst[0]+ srcSliceY *dstStride[0],
1722 dst[1]+(srcSliceY>>1)*dstStride[1],
1723 dst[2]+(srcSliceY>>1)*dstStride[2],
1724 c->srcW, srcSliceH,
1725 dstStride[0], dstStride[1], srcStride[0]);
1726}
1727
b241cbf2
MN
1728static void yvu9toyv12Wrapper(SwsContext *c, uint8_t* src[], int srcStride[], int srcSliceY,
1729 int srcSliceH, uint8_t* dst[], int dstStride[]){
1730 int i;
1731
1732 /* copy Y */
1733 if(srcStride[0]==dstStride[0])
1734 memcpy(dst[0]+ srcSliceY*dstStride[0], src[0], srcStride[0]*srcSliceH);
1735 else{
1736 uint8_t *srcPtr= src[0];
1737 uint8_t *dstPtr= dst[0] + dstStride[0]*srcSliceY;
1738
1739 for(i=0; i<srcSliceH; i++)
1740 {
1741 memcpy(dstPtr, srcPtr, c->srcW);
1742 srcPtr+= srcStride[0];
1743 dstPtr+= dstStride[0];
1744 }
1745 }
1746
1747 if(c->dstFormat==IMGFMT_YV12){
1748 planar2x(src[1], dst[1], c->chrSrcW, c->chrSrcH, srcStride[1], dstStride[1]);
1749 planar2x(src[2], dst[2], c->chrSrcW, c->chrSrcH, srcStride[2], dstStride[2]);
1750 }else{
1751 planar2x(src[1], dst[2], c->chrSrcW, c->chrSrcH, srcStride[1], dstStride[2]);
1752 planar2x(src[2], dst[1], c->chrSrcW, c->chrSrcH, srcStride[2], dstStride[1]);
1753 }
1754}
1755
44c1035c
MN
1756/**
1757 * bring pointers in YUV order instead of YVU
1758 */
c7a810cc 1759static inline void orderYUV(int format, uint8_t * sortedP[], int sortedStride[], uint8_t * p[], int stride[]){
e616aa93 1760 if(format == IMGFMT_YV12 || format == IMGFMT_YVU9){
44c1035c
MN
1761 sortedP[0]= p[0];
1762 sortedP[1]= p[1];
1763 sortedP[2]= p[2];
c7a810cc
MN
1764 sortedStride[0]= stride[0];
1765 sortedStride[1]= stride[1];
1766 sortedStride[2]= stride[2];
44c1035c 1767 }
c7a810cc 1768 else if(isPacked(format) || isGray(format))
44c1035c
MN
1769 {
1770 sortedP[0]= p[0];
1771 sortedP[1]=
1772 sortedP[2]= NULL;
c7a810cc 1773 sortedStride[0]= stride[0];
44c1035c
MN
1774 sortedStride[1]=
1775 sortedStride[2]= 0;
1776 }
e09d12f4 1777 else if(format == IMGFMT_I420)
44c1035c
MN
1778 {
1779 sortedP[0]= p[0];
1780 sortedP[1]= p[2];
1781 sortedP[2]= p[1];
c7a810cc
MN
1782 sortedStride[0]= stride[0];
1783 sortedStride[1]= stride[2];
1784 sortedStride[2]= stride[1];
e09d12f4
MN
1785 }else{
1786 MSG_ERR("internal error in orderYUV\n");
44c1035c
MN
1787 }
1788}
b935781b 1789
b6654a54
MN
1790/* unscaled copy like stuff (assumes nearly identical formats) */
1791static void simpleCopy(SwsContext *c, uint8_t* srcParam[], int srcStrideParam[], int srcSliceY,
44c1035c 1792 int srcSliceH, uint8_t* dstParam[], int dstStrideParam[]){
b6654a54
MN
1793
1794 int srcStride[3];
44c1035c 1795 int dstStride[3];
b6654a54
MN
1796 uint8_t *src[3];
1797 uint8_t *dst[3];
1798
c7a810cc
MN
1799 orderYUV(c->srcFormat, src, srcStride, srcParam, srcStrideParam);
1800 orderYUV(c->dstFormat, dst, dstStride, dstParam, dstStrideParam);
b6654a54
MN
1801
1802 if(isPacked(c->srcFormat))
1803 {
1804 if(dstStride[0]==srcStride[0])
1805 memcpy(dst[0] + dstStride[0]*srcSliceY, src[0], srcSliceH*dstStride[0]);
1806 else
1807 {
1808 int i;
1809 uint8_t *srcPtr= src[0];
1810 uint8_t *dstPtr= dst[0] + dstStride[0]*srcSliceY;
a861d4d7
MN
1811 int length=0;
1812
1813 /* universal length finder */
9bd8bd1a
MN
1814 while(length+c->srcW <= ABS(dstStride[0])
1815 && length+c->srcW <= ABS(srcStride[0])) length+= c->srcW;
a861d4d7 1816 ASSERT(length!=0);
b6654a54
MN
1817
1818 for(i=0; i<srcSliceH; i++)
1819 {
1820 memcpy(dstPtr, srcPtr, length);
1821 srcPtr+= srcStride[0];
1822 dstPtr+= dstStride[0];
1823 }
1824 }
1825 }
1826 else
44c1035c 1827 { /* Planar YUV or gray */
b6654a54
MN
1828 int plane;
1829 for(plane=0; plane<3; plane++)
1830 {
e616aa93
MN
1831 int length= plane==0 ? c->srcW : -((-c->srcW )>>c->chrDstHSubSample);
1832 int y= plane==0 ? srcSliceY: -((-srcSliceY)>>c->chrDstVSubSample);
1833 int height= plane==0 ? srcSliceH: -((-srcSliceH)>>c->chrDstVSubSample);
44c1035c
MN
1834
1835 if((isGray(c->srcFormat) || isGray(c->dstFormat)) && plane>0)
0d9f3d85 1836 {
44c1035c 1837 if(!isGray(c->dstFormat))
e616aa93 1838 memset(dst[plane], 128, dstStride[plane]*height);
0d9f3d85 1839 }
b6654a54
MN
1840 else
1841 {
44c1035c
MN
1842 if(dstStride[plane]==srcStride[plane])
1843 memcpy(dst[plane] + dstStride[plane]*y, src[plane], height*dstStride[plane]);
1844 else
b6654a54 1845 {
44c1035c
MN
1846 int i;
1847 uint8_t *srcPtr= src[plane];
1848 uint8_t *dstPtr= dst[plane] + dstStride[plane]*y;
1849 for(i=0; i<height; i++)
1850 {
1851 memcpy(dstPtr, srcPtr, length);
1852 srcPtr+= srcStride[plane];
1853 dstPtr+= dstStride[plane];
1854 }
b6654a54
MN
1855 }
1856 }
1857 }
1858 }
37079906 1859}
28bf81c9 1860
44c1035c 1861static int remove_dup_fourcc(int fourcc)
0d9f3d85
A
1862{
1863 switch(fourcc)
1864 {
1865 case IMGFMT_IYUV: return IMGFMT_I420;
1866 case IMGFMT_Y8 : return IMGFMT_Y800;
0c51ef97 1867 case IMGFMT_IF09: return IMGFMT_YVU9;
0d9f3d85
A
1868 default: return fourcc;
1869 }
1870}
1871
c7a810cc
MN
1872static void getSubSampleFactors(int *h, int *v, int format){
1873 switch(format){
1874 case IMGFMT_YUY2:
1875 *h=1;
1876 *v=0;
1877 break;
1878 case IMGFMT_YV12:
1879 case IMGFMT_I420:
e616aa93 1880 case IMGFMT_Y800: //FIXME remove after different subsamplings are fully implemented
c7a810cc
MN
1881 *h=1;
1882 *v=1;
1883 break;
1884 case IMGFMT_YVU9:
1885 *h=2;
1886 *v=2;
1887 break;
1888 default:
1889 *h=0;
1890 *v=0;
1891 break;
1892 }
1893}
1894
28bf81c9
MN
1895SwsContext *getSwsContext(int srcW, int srcH, int srcFormat, int dstW, int dstH, int dstFormat, int flags,
1896 SwsFilter *srcFilter, SwsFilter *dstFilter){
1897
28bf81c9
MN
1898 SwsContext *c;
1899 int i;
37079906 1900 int usesFilter;
e09d12f4 1901 int unscaled, needsDither;
c7f822d9 1902 SwsFilter dummyFilter= {NULL, NULL, NULL, NULL};
5cebb24b
MN
1903#ifdef ARCH_X86
1904 if(gCpuCaps.hasMMX)
1905 asm volatile("emms\n\t"::: "memory");
1906#endif
28bf81c9 1907 if(swScale==NULL) globalInit();
e616aa93 1908//srcFormat= IMGFMT_Y800;
e09d12f4 1909//dstFormat= IMGFMT_Y800;
6ff0ad6b 1910 /* avoid dupplicate Formats, so we dont need to check to much */
0d9f3d85
A
1911 srcFormat = remove_dup_fourcc(srcFormat);
1912 dstFormat = remove_dup_fourcc(dstFormat);
44c1035c
MN
1913
1914 unscaled = (srcW == dstW && srcH == dstH);
e09d12f4
MN
1915 needsDither= (isBGR(dstFormat) || isRGB(dstFormat))
1916 && (dstFormat&0xFF)<24
1917 && ((dstFormat&0xFF)<(srcFormat&0xFF) || (!(isRGB(srcFormat) || isBGR(srcFormat))));
44c1035c
MN
1918
1919 if(!isSupportedIn(srcFormat))
b81cf274 1920 {
44c1035c
MN
1921 MSG_ERR("swScaler: %s is not supported as input format\n", vo_format_name(srcFormat));
1922 return NULL;
1923 }
1924 if(!isSupportedOut(dstFormat))
1925 {
1926 MSG_ERR("swScaler: %s is not supported as output format\n", vo_format_name(dstFormat));
1927 return NULL;
b81cf274 1928 }
44c1035c 1929
28bf81c9 1930 /* sanity check */
b81cf274
MN
1931 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
1932 {
0d9f3d85 1933 MSG_ERR("swScaler: %dx%d -> %dx%d is invalid scaling dimension\n",
b81cf274
MN
1934 srcW, srcH, dstW, dstH);
1935 return NULL;
1936 }
28bf81c9 1937
c7f822d9
MN
1938 if(!dstFilter) dstFilter= &dummyFilter;
1939 if(!srcFilter) srcFilter= &dummyFilter;
1940
28bf81c9 1941 c= memalign(64, sizeof(SwsContext));
c7f822d9 1942 memset(c, 0, sizeof(SwsContext));
28bf81c9
MN
1943
1944 c->srcW= srcW;
1945 c->srcH= srcH;
1946 c->dstW= dstW;
1947 c->dstH= dstH;
5521b193
MN
1948 c->lumXInc= ((srcW<<16) + (dstW>>1))/dstW;
1949 c->lumYInc= ((srcH<<16) + (dstH>>1))/dstH;
28bf81c9
MN
1950 c->flags= flags;
1951 c->dstFormat= dstFormat;
1952 c->srcFormat= srcFormat;
1953
37079906
MN
1954 usesFilter=0;
1955 if(dstFilter->lumV!=NULL && dstFilter->lumV->length>1) usesFilter=1;
1956 if(dstFilter->lumH!=NULL && dstFilter->lumH->length>1) usesFilter=1;
1957 if(dstFilter->chrV!=NULL && dstFilter->chrV->length>1) usesFilter=1;
1958 if(dstFilter->chrH!=NULL && dstFilter->chrH->length>1) usesFilter=1;
1959 if(srcFilter->lumV!=NULL && srcFilter->lumV->length>1) usesFilter=1;
1960 if(srcFilter->lumH!=NULL && srcFilter->lumH->length>1) usesFilter=1;
1961 if(srcFilter->chrV!=NULL && srcFilter->chrV->length>1) usesFilter=1;
1962 if(srcFilter->chrH!=NULL && srcFilter->chrH->length>1) usesFilter=1;
e616aa93
MN
1963
1964 getSubSampleFactors(&c->chrSrcHSubSample, &c->chrSrcVSubSample, srcFormat);
1965 getSubSampleFactors(&c->chrDstHSubSample, &c->chrDstVSubSample, dstFormat);
1966
1967 // reuse chroma for 2 pixles rgb/bgr unless user wants full chroma interpolation
1968 if((isBGR(dstFormat) || isRGB(dstFormat)) && !(flags&SWS_FULL_CHR_H_INT)) c->chrDstHSubSample=1;
1969
5859233b
MN
1970 // drop some chroma lines if the user wants it
1971 c->vChrDrop= (flags&SWS_SRC_V_CHR_DROP_MASK)>>SWS_SRC_V_CHR_DROP_SHIFT;
1972 c->chrSrcVSubSample+= c->vChrDrop;
e616aa93 1973
5859233b 1974 // drop every 2. pixel for chroma calculation unless user wants full chroma
e616aa93
MN
1975 if((isBGR(srcFormat) || isRGB(srcFormat)) && !(flags&SWS_FULL_CHR_H_INP))
1976 c->chrSrcHSubSample=1;
1977
1978 c->chrIntHSubSample= c->chrDstHSubSample;
1979 c->chrIntVSubSample= c->chrSrcVSubSample;
1980
1981 // note the -((-x)>>y) is so that we allways round toward +inf
1982 c->chrSrcW= -((-srcW) >> c->chrSrcHSubSample);
1983 c->chrSrcH= -((-srcH) >> c->chrSrcVSubSample);
1984 c->chrDstW= -((-dstW) >> c->chrDstHSubSample);
1985 c->chrDstH= -((-dstH) >> c->chrDstVSubSample);
37079906 1986
cf7d1c1a
MN
1987 if(isBGR(dstFormat))
1988 c->yuvTable= yuv2rgb_c_init(dstFormat & 0xFF, MODE_RGB, c->table_rV, c->table_gU, c->table_gV, c->table_bU);
1989 if(isRGB(dstFormat))
1990 c->yuvTable= yuv2rgb_c_init(dstFormat & 0xFF, MODE_BGR, c->table_rV, c->table_gU, c->table_gV, c->table_bU);
1991
b935781b 1992 /* unscaled special Cases */
44c1035c 1993 if(unscaled && !usesFilter)
37079906 1994 {
0d9f3d85
A
1995 /* yv12_to_nv12 */
1996 if((srcFormat == IMGFMT_YV12||srcFormat==IMGFMT_I420)&&dstFormat == IMGFMT_NV12)
1997 {
1998 c->swScale= PlanarToNV12Wrapper;
1999
2000 if(flags&SWS_PRINT_INFO)
2001 MSG_INFO("SwScaler: using unscaled %s -> %s special converter\n",
2002 vo_format_name(srcFormat), vo_format_name(dstFormat));
2003 return c;
2004 }
2005 /* yv12_to_yuy2 */
2006 if((srcFormat == IMGFMT_YV12||srcFormat==IMGFMT_I420)&&dstFormat == IMGFMT_YUY2)
2007 {
44c1035c 2008 c->swScale= PlanarToYuy2Wrapper;
0d9f3d85
A
2009
2010 if(flags&SWS_PRINT_INFO)
2011 MSG_INFO("SwScaler: using unscaled %s -> %s special converter\n",
2012 vo_format_name(srcFormat), vo_format_name(dstFormat));
2013 return c;
2014 }
37079906 2015 /* yuv2bgr */
e616aa93 2016 if((srcFormat==IMGFMT_YV12 || srcFormat==IMGFMT_I420) && isBGR(dstFormat))
37079906
MN
2017 {
2018 // FIXME multiple yuv2rgb converters wont work that way cuz that thing is full of globals&statics
cf7d1c1a 2019 //FIXME rgb vs. bgr ?
1e1c4fe9 2020#ifdef WORDS_BIGENDIAN
daa57641
MN
2021 if(dstFormat==IMGFMT_BGR32)
2022 yuv2rgb_init( dstFormat&0xFF /* =bpp */, MODE_BGR);
2023 else
2024 yuv2rgb_init( dstFormat&0xFF /* =bpp */, MODE_RGB);
1e1c4fe9 2025#else
b6654a54 2026 yuv2rgb_init( dstFormat&0xFF /* =bpp */, MODE_RGB);
1e1c4fe9 2027#endif
37079906 2028 c->swScale= planarYuvToBgr;
b6654a54
MN
2029
2030 if(flags&SWS_PRINT_INFO)
0d9f3d85 2031 MSG_INFO("SwScaler: using unscaled %s -> %s special converter\n",
b6654a54
MN
2032 vo_format_name(srcFormat), vo_format_name(dstFormat));
2033 return c;
2034 }
b241cbf2 2035
b6654a54 2036 /* simple copy */
e616aa93
MN
2037 if( srcFormat == dstFormat
2038 || (srcFormat==IMGFMT_YV12 && dstFormat==IMGFMT_I420)
2039 || (srcFormat==IMGFMT_I420 && dstFormat==IMGFMT_YV12)
2040 || (isPlanarYUV(srcFormat) && isGray(dstFormat))
2041 || (isPlanarYUV(dstFormat) && isGray(srcFormat))
2042 )
b6654a54
MN
2043 {
2044 c->swScale= simpleCopy;
2045
37079906 2046 if(flags&SWS_PRINT_INFO)
0d9f3d85 2047 MSG_INFO("SwScaler: using unscaled %s -> %s special converter\n",
37079906
MN
2048 vo_format_name(srcFormat), vo_format_name(dstFormat));
2049 return c;
2050 }
b241cbf2
MN
2051
2052 if( srcFormat==IMGFMT_YVU9 && (dstFormat==IMGFMT_YV12 || dstFormat==IMGFMT_I420) )
2053 {
2054 c->swScale= yvu9toyv12Wrapper;
2055
2056 if(flags&SWS_PRINT_INFO)
2057 MSG_INFO("SwScaler: using unscaled %s -> %s special converter\n",
2058 vo_format_name(srcFormat), vo_format_name(dstFormat));
2059 return c;
2060 }
2061
ec22603f
MN
2062 /* bgr24toYV12 */
2063 if(srcFormat==IMGFMT_BGR24 && dstFormat==IMGFMT_YV12)
ec22603f 2064 c->swScale= bgr24toyv12Wrapper;
e09d12f4
MN
2065
2066 /* rgb/bgr -> rgb/bgr (no dither needed forms) */
2067 if( (isBGR(srcFormat) || isRGB(srcFormat))
2068 && (isBGR(dstFormat) || isRGB(dstFormat))
2069 && !needsDither)
2070 c->swScale= rgb2rgbWrapper;
2071
2072 /* LQ converters if -sws 0 or -sws 4*/
2073 if(c->flags&(SWS_FAST_BILINEAR|SWS_POINT)){
2074 /* rgb/bgr -> rgb/bgr (dither needed forms) */
2075 if( (isBGR(srcFormat) || isRGB(srcFormat))
2076 && (isBGR(dstFormat) || isRGB(dstFormat))
2077 && needsDither)
2078 c->swScale= rgb2rgbWrapper;
2079 }
ec22603f 2080
e09d12f4 2081 if(c->swScale){
ec22603f 2082 if(flags&SWS_PRINT_INFO)
0d9f3d85 2083 MSG_INFO("SwScaler: using unscaled %s -> %s special converter\n",
ec22603f
MN
2084 vo_format_name(srcFormat), vo_format_name(dstFormat));
2085 return c;
2086 }
37079906
MN
2087 }
2088
28bf81c9
MN
2089 if(cpuCaps.hasMMX2)
2090 {
2091 c->canMMX2BeUsed= (dstW >=srcW && (dstW&31)==0 && (srcW&15)==0) ? 1 : 0;
2092 if(!c->canMMX2BeUsed && dstW >=srcW && (srcW&15)==0 && (flags&SWS_FAST_BILINEAR))
2093 {
2094 if(flags&SWS_PRINT_INFO)
0d9f3d85 2095 MSG_INFO("SwScaler: output Width is not a multiple of 32 -> no MMX2 scaler\n");
28bf81c9
MN
2096 }
2097 }
2098 else
2099 c->canMMX2BeUsed=0;
2100
1e621b18
MN
2101 c->chrXInc= ((c->chrSrcW<<16) + (c->chrDstW>>1))/c->chrDstW;
2102 c->chrYInc= ((c->chrSrcH<<16) + (c->chrDstH>>1))/c->chrDstH;
2103
28bf81c9
MN
2104 // match pixel 0 of the src to pixel 0 of dst and match pixel n-2 of src to pixel n-2 of dst
2105 // but only for the FAST_BILINEAR mode otherwise do correct scaling
2106 // n-2 is the last chrominance sample available
2107 // this is not perfect, but noone shuld notice the difference, the more correct variant
2108 // would be like the vertical one, but that would require some special code for the
2109 // first and last pixel
2110 if(flags&SWS_FAST_BILINEAR)
2111 {
1e621b18
MN
2112 if(c->canMMX2BeUsed)
2113 {
2114 c->lumXInc+= 20;
2115 c->chrXInc+= 20;
2116 }
28bf81c9 2117 //we dont use the x86asm scaler if mmx is available
1e621b18
MN
2118 else if(cpuCaps.hasMMX)
2119 {
2120 c->lumXInc = ((srcW-2)<<16)/(dstW-2) - 20;
2121 c->chrXInc = ((c->chrSrcW-2)<<16)/(c->chrDstW-2) - 20;
2122 }
28bf81c9
MN
2123 }
2124
28bf81c9
MN
2125 /* precalculate horizontal scaler filter coefficients */
2126 {
2127 const int filterAlign= cpuCaps.hasMMX ? 4 : 1;
2128
c7f822d9 2129 initFilter(&c->hLumFilter, &c->hLumFilterPos, &c->hLumFilterSize, c->lumXInc,
81a571a8
MN
2130 srcW , dstW, filterAlign, 1<<14,
2131 (flags&SWS_BICUBLIN) ? (flags|SWS_BICUBIC) : flags,
c7f822d9
MN
2132 srcFilter->lumH, dstFilter->lumH);
2133 initFilter(&c->hChrFilter, &c->hChrFilterPos, &c->hChrFilterSize, c->chrXInc,
81a571a8
MN
2134 c->chrSrcW, c->chrDstW, filterAlign, 1<<14,
2135 (flags&SWS_BICUBLIN) ? (flags|SWS_BILINEAR) : flags,
c7f822d9 2136 srcFilter->chrH, dstFilter->chrH);
28bf81c9
MN
2137
2138#ifdef ARCH_X86
2139// cant downscale !!!
2140 if(c->canMMX2BeUsed && (flags & SWS_FAST_BILINEAR))
2141 {
b7dc6f66
MN
2142 c->lumMmx2Filter = (int16_t*)memalign(8, (dstW /8+8)*sizeof(int16_t));
2143 c->chrMmx2Filter = (int16_t*)memalign(8, (c->chrDstW /4+8)*sizeof(int16_t));
2144 c->lumMmx2FilterPos= (int32_t*)memalign(8, (dstW /2/8+8)*sizeof(int32_t));
2145 c->chrMmx2FilterPos= (int32_t*)memalign(8, (c->chrDstW/2/4+8)*sizeof(int32_t));
2146
2147 initMMX2HScaler( dstW, c->lumXInc, c->funnyYCode , c->lumMmx2Filter, c->lumMmx2FilterPos, 8);
2148 initMMX2HScaler(c->chrDstW, c->chrXInc, c->funnyUVCode, c->chrMmx2Filter, c->chrMmx2FilterPos, 4);
28bf81c9
MN
2149 }
2150#endif
2151 } // Init Horizontal stuff
2152
2153
2154
2155 /* precalculate vertical scaler filter coefficients */
c7f822d9 2156 initFilter(&c->vLumFilter, &c->vLumFilterPos, &c->vLumFilterSize, c->lumYInc,
81a571a8
MN
2157 srcH , dstH, 1, (1<<12)-4,
2158 (flags&SWS_BICUBLIN) ? (flags|SWS_BICUBIC) : flags,
c7f822d9
MN
2159 srcFilter->lumV, dstFilter->lumV);
2160 initFilter(&c->vChrFilter, &c->vChrFilterPos, &c->vChrFilterSize, c->chrYInc,
81a571a8
MN
2161 c->chrSrcH, c->chrDstH, 1, (1<<12)-4,
2162 (flags&SWS_BICUBLIN) ? (flags|SWS_BILINEAR) : flags,
2163 srcFilter->chrV, dstFilter->chrV);
28bf81c9
MN
2164
2165 // Calculate Buffer Sizes so that they wont run out while handling these damn slices
2166 c->vLumBufSize= c->vLumFilterSize;
2167 c->vChrBufSize= c->vChrFilterSize;
2168 for(i=0; i<dstH; i++)
2169 {
2170 int chrI= i*c->chrDstH / dstH;
2171 int nextSlice= MAX(c->vLumFilterPos[i ] + c->vLumFilterSize - 1,
e616aa93
MN
2172 ((c->vChrFilterPos[chrI] + c->vChrFilterSize - 1)<<c->chrSrcVSubSample));
2173 nextSlice&= ~3; // Slices start at boundaries which are divisable through 4
28bf81c9
MN
2174 if(c->vLumFilterPos[i ] + c->vLumBufSize < nextSlice)
2175 c->vLumBufSize= nextSlice - c->vLumFilterPos[i ];
e616aa93
MN
2176 if(c->vChrFilterPos[chrI] + c->vChrBufSize < (nextSlice>>c->chrSrcVSubSample))
2177 c->vChrBufSize= (nextSlice>>c->chrSrcVSubSample) - c->vChrFilterPos[chrI];
28bf81c9
MN
2178 }
2179
2180 // allocate pixbufs (we use dynamic allocation because otherwise we would need to
c7f822d9
MN
2181 c->lumPixBuf= (int16_t**)memalign(4, c->vLumBufSize*2*sizeof(int16_t*));
2182 c->chrPixBuf= (int16_t**)memalign(4, c->vChrBufSize*2*sizeof(int16_t*));
6c7506de 2183 //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
2184 for(i=0; i<c->vLumBufSize; i++)
2185 c->lumPixBuf[i]= c->lumPixBuf[i+c->vLumBufSize]= (uint16_t*)memalign(8, 4000);
2186 for(i=0; i<c->vChrBufSize; i++)
2187 c->chrPixBuf[i]= c->chrPixBuf[i+c->vChrBufSize]= (uint16_t*)memalign(8, 8000);
2188
2189 //try to avoid drawing green stuff between the right end and the stride end
2190 for(i=0; i<c->vLumBufSize; i++) memset(c->lumPixBuf[i], 0, 4000);
2191 for(i=0; i<c->vChrBufSize; i++) memset(c->chrPixBuf[i], 64, 8000);
2192
2193 ASSERT(c->chrDstH <= dstH)
28bf81c9
MN
2194
2195 // pack filter data for mmx code
2196 if(cpuCaps.hasMMX)
2197 {
c7f822d9
MN
2198 c->lumMmxFilter= (int16_t*)memalign(8, c->vLumFilterSize* dstH*4*sizeof(int16_t));
2199 c->chrMmxFilter= (int16_t*)memalign(8, c->vChrFilterSize*c->chrDstH*4*sizeof(int16_t));
28bf81c9
MN
2200 for(i=0; i<c->vLumFilterSize*dstH; i++)
2201 c->lumMmxFilter[4*i]=c->lumMmxFilter[4*i+1]=c->lumMmxFilter[4*i+2]=c->lumMmxFilter[4*i+3]=
2202 c->vLumFilter[i];
2203 for(i=0; i<c->vChrFilterSize*c->chrDstH; i++)
2204 c->chrMmxFilter[4*i]=c->chrMmxFilter[4*i+1]=c->chrMmxFilter[4*i+2]=c->chrMmxFilter[4*i+3]=
2205 c->vChrFilter[i];
2206 }
2207
2208 if(flags&SWS_PRINT_INFO)
2209 {
2210#ifdef DITHER1XBPP
5521b193
MN
2211 char *dither= " dithered";
2212#else
2213 char *dither= "";
28bf81c9
MN
2214#endif
2215 if(flags&SWS_FAST_BILINEAR)
0d9f3d85 2216 MSG_INFO("\nSwScaler: FAST_BILINEAR scaler, ");
28bf81c9 2217 else if(flags&SWS_BILINEAR)
0d9f3d85 2218 MSG_INFO("\nSwScaler: BILINEAR scaler, ");
28bf81c9 2219 else if(flags&SWS_BICUBIC)
0d9f3d85 2220 MSG_INFO("\nSwScaler: BICUBIC scaler, ");
1e621b18 2221 else if(flags&SWS_X)
0d9f3d85 2222 MSG_INFO("\nSwScaler: Experimental scaler, ");
ff7ba856 2223 else if(flags&SWS_POINT)
0d9f3d85 2224 MSG_INFO("\nSwScaler: Nearest Neighbor / POINT scaler, ");
d8863d37 2225 else if(flags&SWS_AREA)
0d9f3d85 2226 MSG_INFO("\nSwScaler: Area Averageing scaler, ");
81a571a8 2227 else if(flags&SWS_BICUBLIN)
a86c461c
MN
2228 MSG_INFO("\nSwScaler: luma BICUBIC / chroma BILINEAR scaler, ");
2229 else if(flags&SWS_GAUSS)
2230 MSG_INFO("\nSwScaler: Gaussian scaler, ");
2231 else if(flags&SWS_SINC)
2232 MSG_INFO("\nSwScaler: Sinc scaler, ");
2233 else if(flags&SWS_LANCZOS)
2234 MSG_INFO("\nSwScaler: Lanczos scaler, ");
2235 else if(flags&SWS_SPLINE)
2236 MSG_INFO("\nSwScaler: Bicubic spline scaler, ");
28bf81c9 2237 else
0d9f3d85 2238 MSG_INFO("\nSwScaler: ehh flags invalid?! ");
28bf81c9 2239
0d9f3d85
A
2240 if(dstFormat==IMGFMT_BGR15 || dstFormat==IMGFMT_BGR16)
2241 MSG_INFO("from %s to%s %s ",
2242 vo_format_name(srcFormat), dither, vo_format_name(dstFormat));
2243 else
2244 MSG_INFO("from %s to %s ",
2245 vo_format_name(srcFormat), vo_format_name(dstFormat));
28bf81c9
MN
2246
2247 if(cpuCaps.hasMMX2)
0d9f3d85 2248 MSG_INFO("using MMX2\n");
28bf81c9 2249 else if(cpuCaps.has3DNow)
0d9f3d85 2250 MSG_INFO("using 3DNOW\n");
28bf81c9 2251 else if(cpuCaps.hasMMX)
0d9f3d85 2252 MSG_INFO("using MMX\n");
28bf81c9 2253 else
0d9f3d85 2254 MSG_INFO("using C\n");
28bf81c9
MN
2255 }
2256
2257 if((flags & SWS_PRINT_INFO) && verbose)
2258 {
2259 if(cpuCaps.hasMMX)
2260 {
2261 if(c->canMMX2BeUsed && (flags&SWS_FAST_BILINEAR))
0d9f3d85 2262 MSG_V("SwScaler: using FAST_BILINEAR MMX2 scaler for horizontal scaling\n");
28bf81c9
MN
2263 else
2264 {
2265 if(c->hLumFilterSize==4)
0d9f3d85 2266 MSG_V("SwScaler: using 4-tap MMX scaler for horizontal luminance scaling\n");
28bf81c9 2267 else if(c->hLumFilterSize==8)
0d9f3d85 2268 MSG_V("SwScaler: using 8-tap MMX scaler for horizontal luminance scaling\n");
28bf81c9 2269 else
0d9f3d85 2270 MSG_V("SwScaler: using n-tap MMX scaler for horizontal luminance scaling\n");
28bf81c9
MN
2271
2272 if(c->hChrFilterSize==4)
0d9f3d85 2273 MSG_V("SwScaler: using 4-tap MMX scaler for horizontal chrominance scaling\n");
28bf81c9 2274 else if(c->hChrFilterSize==8)
0d9f3d85 2275 MSG_V("SwScaler: using 8-tap MMX scaler for horizontal chrominance scaling\n");
28bf81c9 2276 else
0d9f3d85 2277 MSG_V("SwScaler: using n-tap MMX scaler for horizontal chrominance scaling\n");
28bf81c9
MN
2278 }
2279 }
2280 else
2281 {
2282#ifdef ARCH_X86
0d9f3d85 2283 MSG_V("SwScaler: using X86-Asm scaler for horizontal scaling\n");
28bf81c9
MN
2284#else
2285 if(flags & SWS_FAST_BILINEAR)
0d9f3d85 2286 MSG_V("SwScaler: using FAST_BILINEAR C scaler for horizontal scaling\n");
28bf81c9 2287 else
0d9f3d85 2288 MSG_V("SwScaler: using C scaler for horizontal scaling\n");
28bf81c9
MN
2289#endif
2290 }
6c7506de 2291 if(isPlanarYUV(dstFormat))
28bf81c9
MN
2292 {
2293 if(c->vLumFilterSize==1)
0d9f3d85 2294 MSG_V("SwScaler: using 1-tap %s \"scaler\" for vertical scaling (YV12 like)\n", cpuCaps.hasMMX ? "MMX" : "C");
28bf81c9 2295 else
0d9f3d85 2296 MSG_V("SwScaler: using n-tap %s scaler for vertical scaling (YV12 like)\n", cpuCaps.hasMMX ? "MMX" : "C");
28bf81c9
MN
2297 }
2298 else
2299 {
2300 if(c->vLumFilterSize==1 && c->vChrFilterSize==2)
0d9f3d85 2301 MSG_V("SwScaler: using 1-tap %s \"scaler\" for vertical luminance scaling (BGR)\n"
28bf81c9
MN
2302 "SwScaler: 2-tap scaler for vertical chrominance scaling (BGR)\n",cpuCaps.hasMMX ? "MMX" : "C");
2303 else if(c->vLumFilterSize==2 && c->vChrFilterSize==2)
0d9f3d85 2304 MSG_V("SwScaler: using 2-tap linear %s scaler for vertical scaling (BGR)\n", cpuCaps.hasMMX ? "MMX" : "C");
28bf81c9 2305 else
0d9f3d85 2306 MSG_V("SwScaler: using n-tap %s scaler for vertical scaling (BGR)\n", cpuCaps.hasMMX ? "MMX" : "C");
28bf81c9
MN
2307 }
2308
2309 if(dstFormat==IMGFMT_BGR24)
0d9f3d85 2310 MSG_V("SwScaler: using %s YV12->BGR24 Converter\n",
28bf81c9 2311 cpuCaps.hasMMX2 ? "MMX2" : (cpuCaps.hasMMX ? "MMX" : "C"));
fd284805 2312 else if(dstFormat==IMGFMT_BGR32)
0d9f3d85 2313 MSG_V("SwScaler: using %s YV12->BGR32 Converter\n", cpuCaps.hasMMX ? "MMX" : "C");
fd284805 2314 else if(dstFormat==IMGFMT_BGR16)
0d9f3d85 2315 MSG_V("SwScaler: using %s YV12->BGR16 Converter\n", cpuCaps.hasMMX ? "MMX" : "C");
fd284805 2316 else if(dstFormat==IMGFMT_BGR15)
0d9f3d85 2317 MSG_V("SwScaler: using %s YV12->BGR15 Converter\n", cpuCaps.hasMMX ? "MMX" : "C");
28bf81c9 2318
0d9f3d85 2319 MSG_V("SwScaler: %dx%d -> %dx%d\n", srcW, srcH, dstW, dstH);
28bf81c9 2320 }
1e621b18
MN
2321 if((flags & SWS_PRINT_INFO) && verbose>1)
2322 {
0d9f3d85 2323 MSG_DBG2("SwScaler:Lum srcW=%d srcH=%d dstW=%d dstH=%d xInc=%d yInc=%d\n",
1e621b18 2324 c->srcW, c->srcH, c->dstW, c->dstH, c->lumXInc, c->lumYInc);
0d9f3d85 2325 MSG_DBG2("SwScaler:Chr srcW=%d srcH=%d dstW=%d dstH=%d xInc=%d yInc=%d\n",
1e621b18
MN
2326 c->chrSrcW, c->chrSrcH, c->chrDstW, c->chrDstH, c->chrXInc, c->chrYInc);
2327 }
37079906
MN
2328
2329 c->swScale= swScale;
28bf81c9
MN
2330 return c;
2331}
2332
2333/**
2334 * returns a normalized gaussian curve used to filter stuff
2335 * quality=3 is high quality, lowwer is lowwer quality
2336 */
c7f822d9
MN
2337
2338SwsVector *getGaussianVec(double variance, double quality){
28bf81c9
MN
2339 const int length= (int)(variance*quality + 0.5) | 1;
2340 int i;
2341 double *coeff= memalign(sizeof(double), length*sizeof(double));
2342 double middle= (length-1)*0.5;
c7f822d9
MN
2343 SwsVector *vec= malloc(sizeof(SwsVector));
2344
2345 vec->coeff= coeff;
2346 vec->length= length;
28bf81c9
MN
2347
2348 for(i=0; i<length; i++)
2349 {
2350 double dist= i-middle;
2351 coeff[i]= exp( -dist*dist/(2*variance*variance) ) / sqrt(2*variance*PI);
2352 }
2353
c7f822d9
MN
2354 normalizeVec(vec, 1.0);
2355
2356 return vec;
28bf81c9
MN
2357}
2358
5521b193
MN
2359SwsVector *getConstVec(double c, int length){
2360 int i;
2361 double *coeff= memalign(sizeof(double), length*sizeof(double));
2362 SwsVector *vec= malloc(sizeof(SwsVector));
2363
2364 vec->coeff= coeff;
2365 vec->length= length;
2366
2367 for(i=0; i<length; i++)
2368 coeff[i]= c;
2369
2370 return vec;
2371}
2372
2373
c7f822d9
MN
2374SwsVector *getIdentityVec(void){
2375 double *coeff= memalign(sizeof(double), sizeof(double));
2376 SwsVector *vec= malloc(sizeof(SwsVector));
2377 coeff[0]= 1.0;
2378
2379 vec->coeff= coeff;
2380 vec->length= 1;
2381
2382 return vec;
2383}
2384
2385void normalizeVec(SwsVector *a, double height){
28bf81c9
MN
2386 int i;
2387 double sum=0;
2388 double inv;
2389
c7f822d9
MN
2390 for(i=0; i<a->length; i++)
2391 sum+= a->coeff[i];
28bf81c9
MN
2392
2393 inv= height/sum;
2394
c7f822d9
MN
2395 for(i=0; i<a->length; i++)
2396 a->coeff[i]*= height;
28bf81c9
MN
2397}
2398
c7f822d9
MN
2399void scaleVec(SwsVector *a, double scalar){
2400 int i;
2401
2402 for(i=0; i<a->length; i++)
2403 a->coeff[i]*= scalar;
2404}
2405
5cebb24b 2406static SwsVector *getConvVec(SwsVector *a, SwsVector *b){
c7f822d9 2407 int length= a->length + b->length - 1;
28bf81c9
MN
2408 double *coeff= memalign(sizeof(double), length*sizeof(double));
2409 int i, j;
c7f822d9
MN
2410 SwsVector *vec= malloc(sizeof(SwsVector));
2411
2412 vec->coeff= coeff;
2413 vec->length= length;
28bf81c9
MN
2414
2415 for(i=0; i<length; i++) coeff[i]= 0.0;
2416
c7f822d9 2417 for(i=0; i<a->length; i++)
28bf81c9 2418 {
c7f822d9 2419 for(j=0; j<b->length; j++)
28bf81c9 2420 {
c7f822d9 2421 coeff[i+j]+= a->coeff[i]*b->coeff[j];
28bf81c9
MN
2422 }
2423 }
2424
c7f822d9 2425 return vec;
28bf81c9
MN
2426}
2427
5cebb24b 2428static SwsVector *sumVec(SwsVector *a, SwsVector *b){
c7f822d9 2429 int length= MAX(a->length, b->length);
28bf81c9
MN
2430 double *coeff= memalign(sizeof(double), length*sizeof(double));
2431 int i;
c7f822d9
MN
2432 SwsVector *vec= malloc(sizeof(SwsVector));
2433
2434 vec->coeff= coeff;
2435 vec->length= length;
28bf81c9
MN
2436
2437 for(i=0; i<length; i++) coeff[i]= 0.0;
2438
c7f822d9
MN
2439 for(i=0; i<a->length; i++) coeff[i + (length-1)/2 - (a->length-1)/2]+= a->coeff[i];
2440 for(i=0; i<b->length; i++) coeff[i + (length-1)/2 - (b->length-1)/2]+= b->coeff[i];
2441
2442 return vec;
28bf81c9 2443}
c7f822d9 2444
5cebb24b 2445static SwsVector *diffVec(SwsVector *a, SwsVector *b){
c7f822d9
MN
2446 int length= MAX(a->length, b->length);
2447 double *coeff= memalign(sizeof(double), length*sizeof(double));
2448 int i;
2449 SwsVector *vec= malloc(sizeof(SwsVector));
2450
2451 vec->coeff= coeff;
2452 vec->length= length;
2453
2454 for(i=0; i<length; i++) coeff[i]= 0.0;
2455
2456 for(i=0; i<a->length; i++) coeff[i + (length-1)/2 - (a->length-1)/2]+= a->coeff[i];
2457 for(i=0; i<b->length; i++) coeff[i + (length-1)/2 - (b->length-1)/2]-= b->coeff[i];
2458
2459 return vec;
2460}
2461
2462/* shift left / or right if "shift" is negative */
5cebb24b 2463static SwsVector *getShiftedVec(SwsVector *a, int shift){
c7f822d9
MN
2464 int length= a->length + ABS(shift)*2;
2465 double *coeff= memalign(sizeof(double), length*sizeof(double));
ff7ba856 2466 int i;
c7f822d9
MN
2467 SwsVector *vec= malloc(sizeof(SwsVector));
2468
2469 vec->coeff= coeff;
2470 vec->length= length;
2471
2472 for(i=0; i<length; i++) coeff[i]= 0.0;
2473
2474 for(i=0; i<a->length; i++)
2475 {
2476 coeff[i + (length-1)/2 - (a->length-1)/2 - shift]= a->coeff[i];
2477 }
2478
2479 return vec;
2480}
2481
5cebb24b
MN
2482void shiftVec(SwsVector *a, int shift){
2483 SwsVector *shifted= getShiftedVec(a, shift);
2484 free(a->coeff);
2485 a->coeff= shifted->coeff;
2486 a->length= shifted->length;
2487 free(shifted);
2488}
2489
2490void addVec(SwsVector *a, SwsVector *b){
2491 SwsVector *sum= sumVec(a, b);
2492 free(a->coeff);
2493 a->coeff= sum->coeff;
2494 a->length= sum->length;
2495 free(sum);
2496}
2497
2498void subVec(SwsVector *a, SwsVector *b){
2499 SwsVector *diff= diffVec(a, b);
2500 free(a->coeff);
2501 a->coeff= diff->coeff;
2502 a->length= diff->length;
2503 free(diff);
2504}
2505
2506void convVec(SwsVector *a, SwsVector *b){
2507 SwsVector *conv= getConvVec(a, b);
2508 free(a->coeff);
2509 a->coeff= conv->coeff;
2510 a->length= conv->length;
2511 free(conv);
2512}
2513
2514SwsVector *cloneVec(SwsVector *a){
2515 double *coeff= memalign(sizeof(double), a->length*sizeof(double));
2516 int i;
2517 SwsVector *vec= malloc(sizeof(SwsVector));
2518
2519 vec->coeff= coeff;
2520 vec->length= a->length;
2521
2522 for(i=0; i<a->length; i++) coeff[i]= a->coeff[i];
2523
2524 return vec;
2525}
2526
c7f822d9
MN
2527void printVec(SwsVector *a){
2528 int i;
2529 double max=0;
2530 double min=0;
2531 double range;
2532
2533 for(i=0; i<a->length; i++)
2534 if(a->coeff[i]>max) max= a->coeff[i];
2535
2536 for(i=0; i<a->length; i++)
2537 if(a->coeff[i]<min) min= a->coeff[i];
2538
2539 range= max - min;
2540
2541 for(i=0; i<a->length; i++)
2542 {
2543 int x= (int)((a->coeff[i]-min)*60.0/range +0.5);
0d9f3d85
A
2544 MSG_DBG2("%1.3f ", a->coeff[i]);
2545 for(;x>0; x--) MSG_DBG2(" ");
2546 MSG_DBG2("|\n");
c7f822d9
MN
2547 }
2548}
2549
2550void freeVec(SwsVector *a){
2551 if(!a) return;
2552 if(a->coeff) free(a->coeff);
2553 a->coeff=NULL;
2554 a->length=0;
2555 free(a);
2556}
2557
2558void freeSwsContext(SwsContext *c){
2559 int i;
c7f822d9
MN
2560 if(!c) return;
2561
2562 if(c->lumPixBuf)
2563 {
6c7506de 2564 for(i=0; i<c->vLumBufSize; i++)
c7f822d9
MN
2565 {
2566 if(c->lumPixBuf[i]) free(c->lumPixBuf[i]);
2567 c->lumPixBuf[i]=NULL;
2568 }
2569 free(c->lumPixBuf);
2570 c->lumPixBuf=NULL;
2571 }
2572
2573 if(c->chrPixBuf)
2574 {
6c7506de 2575 for(i=0; i<c->vChrBufSize; i++)
c7f822d9
MN
2576 {
2577 if(c->chrPixBuf[i]) free(c->chrPixBuf[i]);
2578 c->chrPixBuf[i]=NULL;
2579 }
2580 free(c->chrPixBuf);
2581 c->chrPixBuf=NULL;
2582 }
2583
2584 if(c->vLumFilter) free(c->vLumFilter);
2585 c->vLumFilter = NULL;
2586 if(c->vChrFilter) free(c->vChrFilter);
2587 c->vChrFilter = NULL;
2588 if(c->hLumFilter) free(c->hLumFilter);
2589 c->hLumFilter = NULL;
2590 if(c->hChrFilter) free(c->hChrFilter);
2591 c->hChrFilter = NULL;
2592
2593 if(c->vLumFilterPos) free(c->vLumFilterPos);
2594 c->vLumFilterPos = NULL;
2595 if(c->vChrFilterPos) free(c->vChrFilterPos);
2596 c->vChrFilterPos = NULL;
2597 if(c->hLumFilterPos) free(c->hLumFilterPos);
2598 c->hLumFilterPos = NULL;
2599 if(c->hChrFilterPos) free(c->hChrFilterPos);
2600 c->hChrFilterPos = NULL;
2601
2602 if(c->lumMmxFilter) free(c->lumMmxFilter);
2603 c->lumMmxFilter = NULL;
2604 if(c->chrMmxFilter) free(c->chrMmxFilter);
2605 c->chrMmxFilter = NULL;
2606
b7dc6f66
MN
2607 if(c->lumMmx2Filter) free(c->lumMmx2Filter);
2608 c->lumMmx2Filter=NULL;
2609 if(c->chrMmx2Filter) free(c->chrMmx2Filter);
2610 c->chrMmx2Filter=NULL;
2611 if(c->lumMmx2FilterPos) free(c->lumMmx2FilterPos);
2612 c->lumMmx2FilterPos=NULL;
2613 if(c->chrMmx2FilterPos) free(c->chrMmx2FilterPos);
2614 c->chrMmx2FilterPos=NULL;
cf7d1c1a
MN
2615 if(c->yuvTable) free(c->yuvTable);
2616 c->yuvTable=NULL;
b7dc6f66 2617
c7f822d9
MN
2618 free(c);
2619}
2620
7f56a527 2621