passing an array or double precission parameters for the scaling function, instead...
[libav.git] / postproc / swscale.c
1 /*
2 Copyright (C) 2001-2003 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.
8
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.
13
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 */
18
19 /*
20 supported Input formats: YV12, I420/IYUV, YUY2, UYVY, BGR32, BGR24, BGR16, BGR15, RGB32, RGB24, Y8/Y800, YVU9/IF09
21 supported output formats: YV12, I420/IYUV, YUY2, UYVY, {BGR,RGB}{1,4,8,15,16,24,32}, Y8/Y800, YVU9/IF09
22 {BGR,RGB}{1,4,8,15,16} support dithering
23
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
30 BGR24 -> BGR32 & RGB24 -> RGB32
31 BGR32 -> BGR24 & RGB32 -> RGB24
32 BGR15 -> BGR16
33 */
34
35 /*
36 tested special converters (most are tested actually but i didnt write it down ...)
37 YV12 -> BGR16
38 YV12 -> YV12
39 BGR15 -> BGR16
40 BGR16 -> BGR16
41 YVU9 -> YV12
42
43 untested special converters
44 YV12/I420 -> BGR15/BGR24/BGR32 (its the yuv2rgb stuff, so it should be ok)
45 YV12/I420 -> YV12/I420
46 YUY2/BGR15/BGR24/BGR32/RGB24/RGB32 -> same format
47 BGR24 -> BGR32 & RGB24 -> RGB32
48 BGR32 -> BGR24 & RGB32 -> RGB24
49 BGR24 -> YV12
50 */
51
52 #include <inttypes.h>
53 #include <string.h>
54 #include <math.h>
55 #include <stdio.h>
56 #include "../config.h"
57 #include "../mangle.h"
58 #include <assert.h>
59 #ifdef HAVE_MALLOC_H
60 #include <malloc.h>
61 #else
62 #include <stdlib.h>
63 #endif
64 #include "swscale.h"
65 #include "swscale_internal.h"
66 #include "../cpudetect.h"
67 #include "../bswap.h"
68 #include "../libvo/img_format.h"
69 #include "rgb2rgb.h"
70 #include "../libvo/fastmemcpy.h"
71
72 #undef MOVNTQ
73 #undef PAVGB
74
75 //#undef HAVE_MMX2
76 //#define HAVE_3DNOW
77 //#undef HAVE_MMX
78 //#undef ARCH_X86
79 //#define WORDS_BIGENDIAN
80 #define DITHER1XBPP
81
82 #define FAST_BGR2YV12 // use 7 bit coeffs instead of 15bit
83
84 #define RET 0xC3 //near return opcode for X86
85
86 #ifdef MP_DEBUG
87 #define ASSERT(x) assert(x);
88 #else
89 #define ASSERT(x) ;
90 #endif
91
92 #ifdef M_PI
93 #define PI M_PI
94 #else
95 #define PI 3.14159265358979323846
96 #endif
97
98 //FIXME replace this with something faster
99 #define isPlanarYUV(x) ((x)==IMGFMT_YV12 || (x)==IMGFMT_YVU9 \
100 || (x)==IMGFMT_444P || (x)==IMGFMT_422P || (x)==IMGFMT_411P)
101 #define isYUV(x) ((x)==IMGFMT_UYVY || (x)==IMGFMT_YUY2 || isPlanarYUV(x))
102 #define isGray(x) ((x)==IMGFMT_Y800)
103 #define isRGB(x) (((x)&IMGFMT_RGB_MASK)==IMGFMT_RGB)
104 #define isBGR(x) (((x)&IMGFMT_BGR_MASK)==IMGFMT_BGR)
105 #define isSupportedIn(x) ((x)==IMGFMT_YV12 || (x)==IMGFMT_YUY2 || (x)==IMGFMT_UYVY\
106 || (x)==IMGFMT_BGR32|| (x)==IMGFMT_BGR24|| (x)==IMGFMT_BGR16|| (x)==IMGFMT_BGR15\
107 || (x)==IMGFMT_RGB32|| (x)==IMGFMT_RGB24\
108 || (x)==IMGFMT_Y800 || (x)==IMGFMT_YVU9\
109 || (x)==IMGFMT_444P || (x)==IMGFMT_422P || (x)==IMGFMT_411P)
110 #define isSupportedOut(x) ((x)==IMGFMT_YV12 || (x)==IMGFMT_YUY2 || (x)==IMGFMT_UYVY\
111 || (x)==IMGFMT_444P || (x)==IMGFMT_422P || (x)==IMGFMT_411P\
112 || isRGB(x) || isBGR(x)\
113 || (x)==IMGFMT_Y800 || (x)==IMGFMT_YVU9)
114 #define isPacked(x) ((x)==IMGFMT_YUY2 || (x)==IMGFMT_UYVY ||isRGB(x) || isBGR(x))
115
116 #define RGB2YUV_SHIFT 16
117 #define BY ((int)( 0.098*(1<<RGB2YUV_SHIFT)+0.5))
118 #define BV ((int)(-0.071*(1<<RGB2YUV_SHIFT)+0.5))
119 #define BU ((int)( 0.439*(1<<RGB2YUV_SHIFT)+0.5))
120 #define GY ((int)( 0.504*(1<<RGB2YUV_SHIFT)+0.5))
121 #define GV ((int)(-0.368*(1<<RGB2YUV_SHIFT)+0.5))
122 #define GU ((int)(-0.291*(1<<RGB2YUV_SHIFT)+0.5))
123 #define RY ((int)( 0.257*(1<<RGB2YUV_SHIFT)+0.5))
124 #define RV ((int)( 0.439*(1<<RGB2YUV_SHIFT)+0.5))
125 #define RU ((int)(-0.148*(1<<RGB2YUV_SHIFT)+0.5))
126
127 extern const int32_t Inverse_Table_6_9[8][4];
128
129 /*
130 NOTES
131 Special versions: fast Y 1:1 scaling (no interpolation in y direction)
132
133 TODO
134 more intelligent missalignment avoidance for the horizontal scaler
135 write special vertical cubic upscale version
136 Optimize C code (yv12 / minmax)
137 add support for packed pixel yuv input & output
138 add support for Y8 output
139 optimize bgr24 & bgr32
140 add BGR4 output support
141 write special BGR->BGR scaler
142 */
143
144 #define ABS(a) ((a) > 0 ? (a) : (-(a)))
145 #define MIN(a,b) ((a) > (b) ? (b) : (a))
146 #define MAX(a,b) ((a) < (b) ? (b) : (a))
147
148 #ifdef ARCH_X86
149 static uint64_t attribute_used __attribute__((aligned(8))) bF8= 0xF8F8F8F8F8F8F8F8LL;
150 static uint64_t attribute_used __attribute__((aligned(8))) bFC= 0xFCFCFCFCFCFCFCFCLL;
151 static uint64_t __attribute__((aligned(8))) w10= 0x0010001000100010LL;
152 static uint64_t attribute_used __attribute__((aligned(8))) w02= 0x0002000200020002LL;
153 static uint64_t attribute_used __attribute__((aligned(8))) bm00001111=0x00000000FFFFFFFFLL;
154 static uint64_t attribute_used __attribute__((aligned(8))) bm00000111=0x0000000000FFFFFFLL;
155 static uint64_t attribute_used __attribute__((aligned(8))) bm11111000=0xFFFFFFFFFF000000LL;
156 static uint64_t attribute_used __attribute__((aligned(8))) bm01010101=0x00FF00FF00FF00FFLL;
157
158 static volatile uint64_t attribute_used __attribute__((aligned(8))) b5Dither;
159 static volatile uint64_t attribute_used __attribute__((aligned(8))) g5Dither;
160 static volatile uint64_t attribute_used __attribute__((aligned(8))) g6Dither;
161 static volatile uint64_t attribute_used __attribute__((aligned(8))) r5Dither;
162
163 static uint64_t __attribute__((aligned(8))) dither4[2]={
164 0x0103010301030103LL,
165 0x0200020002000200LL,};
166
167 static uint64_t __attribute__((aligned(8))) dither8[2]={
168 0x0602060206020602LL,
169 0x0004000400040004LL,};
170
171 static uint64_t __attribute__((aligned(8))) b16Mask= 0x001F001F001F001FLL;
172 static uint64_t attribute_used __attribute__((aligned(8))) g16Mask= 0x07E007E007E007E0LL;
173 static uint64_t attribute_used __attribute__((aligned(8))) r16Mask= 0xF800F800F800F800LL;
174 static uint64_t __attribute__((aligned(8))) b15Mask= 0x001F001F001F001FLL;
175 static uint64_t attribute_used __attribute__((aligned(8))) g15Mask= 0x03E003E003E003E0LL;
176 static uint64_t attribute_used __attribute__((aligned(8))) r15Mask= 0x7C007C007C007C00LL;
177
178 static uint64_t attribute_used __attribute__((aligned(8))) M24A= 0x00FF0000FF0000FFLL;
179 static uint64_t attribute_used __attribute__((aligned(8))) M24B= 0xFF0000FF0000FF00LL;
180 static uint64_t attribute_used __attribute__((aligned(8))) M24C= 0x0000FF0000FF0000LL;
181
182 #ifdef FAST_BGR2YV12
183 static const uint64_t bgr2YCoeff attribute_used __attribute__((aligned(8))) = 0x000000210041000DULL;
184 static const uint64_t bgr2UCoeff attribute_used __attribute__((aligned(8))) = 0x0000FFEEFFDC0038ULL;
185 static const uint64_t bgr2VCoeff attribute_used __attribute__((aligned(8))) = 0x00000038FFD2FFF8ULL;
186 #else
187 static const uint64_t bgr2YCoeff attribute_used __attribute__((aligned(8))) = 0x000020E540830C8BULL;
188 static const uint64_t bgr2UCoeff attribute_used __attribute__((aligned(8))) = 0x0000ED0FDAC23831ULL;
189 static const uint64_t bgr2VCoeff attribute_used __attribute__((aligned(8))) = 0x00003831D0E6F6EAULL;
190 #endif
191 static const uint64_t bgr2YOffset attribute_used __attribute__((aligned(8))) = 0x1010101010101010ULL;
192 static const uint64_t bgr2UVOffset attribute_used __attribute__((aligned(8)))= 0x8080808080808080ULL;
193 static const uint64_t w1111 attribute_used __attribute__((aligned(8))) = 0x0001000100010001ULL;
194 #endif
195
196 // clipping helper table for C implementations:
197 static unsigned char clip_table[768];
198
199 static SwsVector *sws_getConvVec(SwsVector *a, SwsVector *b);
200
201 extern const uint8_t dither_2x2_4[2][8];
202 extern const uint8_t dither_2x2_8[2][8];
203 extern const uint8_t dither_8x8_32[8][8];
204 extern const uint8_t dither_8x8_73[8][8];
205 extern const uint8_t dither_8x8_220[8][8];
206
207 #ifdef ARCH_X86
208 void in_asm_used_var_warning_killer()
209 {
210 volatile int i= bF8+bFC+w10+
211 bm00001111+bm00000111+bm11111000+b16Mask+g16Mask+r16Mask+b15Mask+g15Mask+r15Mask+
212 M24A+M24B+M24C+w02 + b5Dither+g5Dither+r5Dither+g6Dither+dither4[0]+dither8[0]+bm01010101;
213 if(i) i=0;
214 }
215 #endif
216
217 static inline void yuv2yuvXinC(int16_t *lumFilter, int16_t **lumSrc, int lumFilterSize,
218 int16_t *chrFilter, int16_t **chrSrc, int chrFilterSize,
219 uint8_t *dest, uint8_t *uDest, uint8_t *vDest, int dstW, int chrDstW)
220 {
221 //FIXME Optimize (just quickly writen not opti..)
222 int i;
223 for(i=0; i<dstW; i++)
224 {
225 int val=1<<18;
226 int j;
227 for(j=0; j<lumFilterSize; j++)
228 val += lumSrc[j][i] * lumFilter[j];
229
230 dest[i]= MIN(MAX(val>>19, 0), 255);
231 }
232
233 if(uDest != NULL)
234 for(i=0; i<chrDstW; i++)
235 {
236 int u=1<<18;
237 int v=1<<18;
238 int j;
239 for(j=0; j<chrFilterSize; j++)
240 {
241 u += chrSrc[j][i] * chrFilter[j];
242 v += chrSrc[j][i + 2048] * chrFilter[j];
243 }
244
245 uDest[i]= MIN(MAX(u>>19, 0), 255);
246 vDest[i]= MIN(MAX(v>>19, 0), 255);
247 }
248 }
249
250
251 #define YSCALE_YUV_2_PACKEDX_C(type) \
252 for(i=0; i<(dstW>>1); i++){\
253 int j;\
254 int Y1=1<<18;\
255 int Y2=1<<18;\
256 int U=1<<18;\
257 int V=1<<18;\
258 type *r, *b, *g;\
259 const int i2= 2*i;\
260 \
261 for(j=0; j<lumFilterSize; j++)\
262 {\
263 Y1 += lumSrc[j][i2] * lumFilter[j];\
264 Y2 += lumSrc[j][i2+1] * lumFilter[j];\
265 }\
266 for(j=0; j<chrFilterSize; j++)\
267 {\
268 U += chrSrc[j][i] * chrFilter[j];\
269 V += chrSrc[j][i+2048] * chrFilter[j];\
270 }\
271 Y1>>=19;\
272 Y2>>=19;\
273 U >>=19;\
274 V >>=19;\
275 if((Y1|Y2|U|V)&256)\
276 {\
277 if(Y1>255) Y1=255;\
278 else if(Y1<0)Y1=0;\
279 if(Y2>255) Y2=255;\
280 else if(Y2<0)Y2=0;\
281 if(U>255) U=255;\
282 else if(U<0) U=0;\
283 if(V>255) V=255;\
284 else if(V<0) V=0;\
285 }
286
287 #define YSCALE_YUV_2_RGBX_C(type) \
288 YSCALE_YUV_2_PACKEDX_C(type)\
289 r = c->table_rV[V];\
290 g = c->table_gU[U] + c->table_gV[V];\
291 b = c->table_bU[U];\
292
293 #define YSCALE_YUV_2_PACKED2_C \
294 for(i=0; i<(dstW>>1); i++){\
295 const int i2= 2*i;\
296 int Y1= (buf0[i2 ]*yalpha1+buf1[i2 ]*yalpha)>>19;\
297 int Y2= (buf0[i2+1]*yalpha1+buf1[i2+1]*yalpha)>>19;\
298 int U= (uvbuf0[i ]*uvalpha1+uvbuf1[i ]*uvalpha)>>19;\
299 int V= (uvbuf0[i+2048]*uvalpha1+uvbuf1[i+2048]*uvalpha)>>19;\
300
301 #define YSCALE_YUV_2_RGB2_C(type) \
302 YSCALE_YUV_2_PACKED2_C\
303 type *r, *b, *g;\
304 r = c->table_rV[V];\
305 g = c->table_gU[U] + c->table_gV[V];\
306 b = c->table_bU[U];\
307
308 #define YSCALE_YUV_2_PACKED1_C \
309 for(i=0; i<(dstW>>1); i++){\
310 const int i2= 2*i;\
311 int Y1= buf0[i2 ]>>7;\
312 int Y2= buf0[i2+1]>>7;\
313 int U= (uvbuf1[i ])>>7;\
314 int V= (uvbuf1[i+2048])>>7;\
315
316 #define YSCALE_YUV_2_RGB1_C(type) \
317 YSCALE_YUV_2_PACKED1_C\
318 type *r, *b, *g;\
319 r = c->table_rV[V];\
320 g = c->table_gU[U] + c->table_gV[V];\
321 b = c->table_bU[U];\
322
323 #define YSCALE_YUV_2_PACKED1B_C \
324 for(i=0; i<(dstW>>1); i++){\
325 const int i2= 2*i;\
326 int Y1= buf0[i2 ]>>7;\
327 int Y2= buf0[i2+1]>>7;\
328 int U= (uvbuf0[i ] + uvbuf1[i ])>>8;\
329 int V= (uvbuf0[i+2048] + uvbuf1[i+2048])>>8;\
330
331 #define YSCALE_YUV_2_RGB1B_C(type) \
332 YSCALE_YUV_2_PACKED1B_C\
333 type *r, *b, *g;\
334 r = c->table_rV[V];\
335 g = c->table_gU[U] + c->table_gV[V];\
336 b = c->table_bU[U];\
337
338 #define YSCALE_YUV_2_ANYRGB_C(func, func2)\
339 switch(c->dstFormat)\
340 {\
341 case IMGFMT_BGR32:\
342 case IMGFMT_RGB32:\
343 func(uint32_t)\
344 ((uint32_t*)dest)[i2+0]= r[Y1] + g[Y1] + b[Y1];\
345 ((uint32_t*)dest)[i2+1]= r[Y2] + g[Y2] + b[Y2];\
346 } \
347 break;\
348 case IMGFMT_RGB24:\
349 func(uint8_t)\
350 ((uint8_t*)dest)[0]= r[Y1];\
351 ((uint8_t*)dest)[1]= g[Y1];\
352 ((uint8_t*)dest)[2]= b[Y1];\
353 ((uint8_t*)dest)[3]= r[Y2];\
354 ((uint8_t*)dest)[4]= g[Y2];\
355 ((uint8_t*)dest)[5]= b[Y2];\
356 dest+=6;\
357 }\
358 break;\
359 case IMGFMT_BGR24:\
360 func(uint8_t)\
361 ((uint8_t*)dest)[0]= b[Y1];\
362 ((uint8_t*)dest)[1]= g[Y1];\
363 ((uint8_t*)dest)[2]= r[Y1];\
364 ((uint8_t*)dest)[3]= b[Y2];\
365 ((uint8_t*)dest)[4]= g[Y2];\
366 ((uint8_t*)dest)[5]= r[Y2];\
367 dest+=6;\
368 }\
369 break;\
370 case IMGFMT_RGB16:\
371 case IMGFMT_BGR16:\
372 {\
373 const int dr1= dither_2x2_8[y&1 ][0];\
374 const int dg1= dither_2x2_4[y&1 ][0];\
375 const int db1= dither_2x2_8[(y&1)^1][0];\
376 const int dr2= dither_2x2_8[y&1 ][1];\
377 const int dg2= dither_2x2_4[y&1 ][1];\
378 const int db2= dither_2x2_8[(y&1)^1][1];\
379 func(uint16_t)\
380 ((uint16_t*)dest)[i2+0]= r[Y1+dr1] + g[Y1+dg1] + b[Y1+db1];\
381 ((uint16_t*)dest)[i2+1]= r[Y2+dr2] + g[Y2+dg2] + b[Y2+db2];\
382 }\
383 }\
384 break;\
385 case IMGFMT_RGB15:\
386 case IMGFMT_BGR15:\
387 {\
388 const int dr1= dither_2x2_8[y&1 ][0];\
389 const int dg1= dither_2x2_8[y&1 ][1];\
390 const int db1= dither_2x2_8[(y&1)^1][0];\
391 const int dr2= dither_2x2_8[y&1 ][1];\
392 const int dg2= dither_2x2_8[y&1 ][0];\
393 const int db2= dither_2x2_8[(y&1)^1][1];\
394 func(uint16_t)\
395 ((uint16_t*)dest)[i2+0]= r[Y1+dr1] + g[Y1+dg1] + b[Y1+db1];\
396 ((uint16_t*)dest)[i2+1]= r[Y2+dr2] + g[Y2+dg2] + b[Y2+db2];\
397 }\
398 }\
399 break;\
400 case IMGFMT_RGB8:\
401 case IMGFMT_BGR8:\
402 {\
403 const uint8_t * const d64= dither_8x8_73[y&7];\
404 const uint8_t * const d32= dither_8x8_32[y&7];\
405 func(uint8_t)\
406 ((uint8_t*)dest)[i2+0]= r[Y1+d32[(i2+0)&7]] + g[Y1+d32[(i2+0)&7]] + b[Y1+d64[(i2+0)&7]];\
407 ((uint8_t*)dest)[i2+1]= r[Y2+d32[(i2+1)&7]] + g[Y2+d32[(i2+1)&7]] + b[Y2+d64[(i2+1)&7]];\
408 }\
409 }\
410 break;\
411 case IMGFMT_RGB4:\
412 case IMGFMT_BGR4:\
413 {\
414 const uint8_t * const d64= dither_8x8_73 [y&7];\
415 const uint8_t * const d128=dither_8x8_220[y&7];\
416 func(uint8_t)\
417 ((uint8_t*)dest)[i]= r[Y1+d128[(i2+0)&7]] + g[Y1+d64[(i2+0)&7]] + b[Y1+d128[(i2+0)&7]]\
418 + ((r[Y2+d128[(i2+1)&7]] + g[Y2+d64[(i2+1)&7]] + b[Y2+d128[(i2+1)&7]])<<4);\
419 }\
420 }\
421 break;\
422 case IMGFMT_RG4B:\
423 case IMGFMT_BG4B:\
424 {\
425 const uint8_t * const d64= dither_8x8_73 [y&7];\
426 const uint8_t * const d128=dither_8x8_220[y&7];\
427 func(uint8_t)\
428 ((uint8_t*)dest)[i2+0]= r[Y1+d128[(i2+0)&7]] + g[Y1+d64[(i2+0)&7]] + b[Y1+d128[(i2+0)&7]];\
429 ((uint8_t*)dest)[i2+1]= r[Y2+d128[(i2+1)&7]] + g[Y2+d64[(i2+1)&7]] + b[Y2+d128[(i2+1)&7]];\
430 }\
431 }\
432 break;\
433 case IMGFMT_RGB1:\
434 case IMGFMT_BGR1:\
435 {\
436 const uint8_t * const d128=dither_8x8_220[y&7];\
437 uint8_t *g= c->table_gU[128] + c->table_gV[128];\
438 for(i=0; i<dstW-7; i+=8){\
439 int acc;\
440 acc = g[((buf0[i ]*yalpha1+buf1[i ]*yalpha)>>19) + d128[0]];\
441 acc+= acc + g[((buf0[i+1]*yalpha1+buf1[i+1]*yalpha)>>19) + d128[1]];\
442 acc+= acc + g[((buf0[i+2]*yalpha1+buf1[i+2]*yalpha)>>19) + d128[2]];\
443 acc+= acc + g[((buf0[i+3]*yalpha1+buf1[i+3]*yalpha)>>19) + d128[3]];\
444 acc+= acc + g[((buf0[i+4]*yalpha1+buf1[i+4]*yalpha)>>19) + d128[4]];\
445 acc+= acc + g[((buf0[i+5]*yalpha1+buf1[i+5]*yalpha)>>19) + d128[5]];\
446 acc+= acc + g[((buf0[i+6]*yalpha1+buf1[i+6]*yalpha)>>19) + d128[6]];\
447 acc+= acc + g[((buf0[i+7]*yalpha1+buf1[i+7]*yalpha)>>19) + d128[7]];\
448 ((uint8_t*)dest)[0]= acc;\
449 dest++;\
450 }\
451 \
452 /*\
453 ((uint8_t*)dest)-= dstW>>4;\
454 {\
455 int acc=0;\
456 int left=0;\
457 static int top[1024];\
458 static int last_new[1024][1024];\
459 static int last_in3[1024][1024];\
460 static int drift[1024][1024];\
461 int topLeft=0;\
462 int shift=0;\
463 int count=0;\
464 const uint8_t * const d128=dither_8x8_220[y&7];\
465 int error_new=0;\
466 int error_in3=0;\
467 int f=0;\
468 \
469 for(i=dstW>>1; i<dstW; i++){\
470 int in= ((buf0[i ]*yalpha1+buf1[i ]*yalpha)>>19);\
471 int in2 = (76309 * (in - 16) + 32768) >> 16;\
472 int in3 = (in2 < 0) ? 0 : ((in2 > 255) ? 255 : in2);\
473 int old= (left*7 + topLeft + top[i]*5 + top[i+1]*3)/20 + in3\
474 + (last_new[y][i] - in3)*f/256;\
475 int new= old> 128 ? 255 : 0;\
476 \
477 error_new+= ABS(last_new[y][i] - new);\
478 error_in3+= ABS(last_in3[y][i] - in3);\
479 f= error_new - error_in3*4;\
480 if(f<0) f=0;\
481 if(f>256) f=256;\
482 \
483 topLeft= top[i];\
484 left= top[i]= old - new;\
485 last_new[y][i]= new;\
486 last_in3[y][i]= in3;\
487 \
488 acc+= acc + (new&1);\
489 if((i&7)==6){\
490 ((uint8_t*)dest)[0]= acc;\
491 ((uint8_t*)dest)++;\
492 }\
493 }\
494 }\
495 */\
496 }\
497 break;\
498 case IMGFMT_YUY2:\
499 func2\
500 ((uint8_t*)dest)[2*i2+0]= Y1;\
501 ((uint8_t*)dest)[2*i2+1]= U;\
502 ((uint8_t*)dest)[2*i2+2]= Y2;\
503 ((uint8_t*)dest)[2*i2+3]= V;\
504 } \
505 break;\
506 case IMGFMT_UYVY:\
507 func2\
508 ((uint8_t*)dest)[2*i2+0]= U;\
509 ((uint8_t*)dest)[2*i2+1]= Y1;\
510 ((uint8_t*)dest)[2*i2+2]= V;\
511 ((uint8_t*)dest)[2*i2+3]= Y2;\
512 } \
513 break;\
514 }\
515
516
517 static inline void yuv2packedXinC(SwsContext *c, int16_t *lumFilter, int16_t **lumSrc, int lumFilterSize,
518 int16_t *chrFilter, int16_t **chrSrc, int chrFilterSize,
519 uint8_t *dest, int dstW, int y)
520 {
521 int i;
522 switch(c->dstFormat)
523 {
524 case IMGFMT_RGB32:
525 case IMGFMT_BGR32:
526 YSCALE_YUV_2_RGBX_C(uint32_t)
527 ((uint32_t*)dest)[i2+0]= r[Y1] + g[Y1] + b[Y1];
528 ((uint32_t*)dest)[i2+1]= r[Y2] + g[Y2] + b[Y2];
529 }
530 break;
531 case IMGFMT_RGB24:
532 YSCALE_YUV_2_RGBX_C(uint8_t)
533 ((uint8_t*)dest)[0]= r[Y1];
534 ((uint8_t*)dest)[1]= g[Y1];
535 ((uint8_t*)dest)[2]= b[Y1];
536 ((uint8_t*)dest)[3]= r[Y2];
537 ((uint8_t*)dest)[4]= g[Y2];
538 ((uint8_t*)dest)[5]= b[Y2];
539 dest+=6;
540 }
541 break;
542 case IMGFMT_BGR24:
543 YSCALE_YUV_2_RGBX_C(uint8_t)
544 ((uint8_t*)dest)[0]= b[Y1];
545 ((uint8_t*)dest)[1]= g[Y1];
546 ((uint8_t*)dest)[2]= r[Y1];
547 ((uint8_t*)dest)[3]= b[Y2];
548 ((uint8_t*)dest)[4]= g[Y2];
549 ((uint8_t*)dest)[5]= r[Y2];
550 dest+=6;
551 }
552 break;
553 case IMGFMT_RGB16:
554 case IMGFMT_BGR16:
555 {
556 const int dr1= dither_2x2_8[y&1 ][0];
557 const int dg1= dither_2x2_4[y&1 ][0];
558 const int db1= dither_2x2_8[(y&1)^1][0];
559 const int dr2= dither_2x2_8[y&1 ][1];
560 const int dg2= dither_2x2_4[y&1 ][1];
561 const int db2= dither_2x2_8[(y&1)^1][1];
562 YSCALE_YUV_2_RGBX_C(uint16_t)
563 ((uint16_t*)dest)[i2+0]= r[Y1+dr1] + g[Y1+dg1] + b[Y1+db1];
564 ((uint16_t*)dest)[i2+1]= r[Y2+dr2] + g[Y2+dg2] + b[Y2+db2];
565 }
566 }
567 break;
568 case IMGFMT_RGB15:
569 case IMGFMT_BGR15:
570 {
571 const int dr1= dither_2x2_8[y&1 ][0];
572 const int dg1= dither_2x2_8[y&1 ][1];
573 const int db1= dither_2x2_8[(y&1)^1][0];
574 const int dr2= dither_2x2_8[y&1 ][1];
575 const int dg2= dither_2x2_8[y&1 ][0];
576 const int db2= dither_2x2_8[(y&1)^1][1];
577 YSCALE_YUV_2_RGBX_C(uint16_t)
578 ((uint16_t*)dest)[i2+0]= r[Y1+dr1] + g[Y1+dg1] + b[Y1+db1];
579 ((uint16_t*)dest)[i2+1]= r[Y2+dr2] + g[Y2+dg2] + b[Y2+db2];
580 }
581 }
582 break;
583 case IMGFMT_RGB8:
584 case IMGFMT_BGR8:
585 {
586 const uint8_t * const d64= dither_8x8_73[y&7];
587 const uint8_t * const d32= dither_8x8_32[y&7];
588 YSCALE_YUV_2_RGBX_C(uint8_t)
589 ((uint8_t*)dest)[i2+0]= r[Y1+d32[(i2+0)&7]] + g[Y1+d32[(i2+0)&7]] + b[Y1+d64[(i2+0)&7]];
590 ((uint8_t*)dest)[i2+1]= r[Y2+d32[(i2+1)&7]] + g[Y2+d32[(i2+1)&7]] + b[Y2+d64[(i2+1)&7]];
591 }
592 }
593 break;
594 case IMGFMT_RGB4:
595 case IMGFMT_BGR4:
596 {
597 const uint8_t * const d64= dither_8x8_73 [y&7];
598 const uint8_t * const d128=dither_8x8_220[y&7];
599 YSCALE_YUV_2_RGBX_C(uint8_t)
600 ((uint8_t*)dest)[i]= r[Y1+d128[(i2+0)&7]] + g[Y1+d64[(i2+0)&7]] + b[Y1+d128[(i2+0)&7]]
601 +((r[Y2+d128[(i2+1)&7]] + g[Y2+d64[(i2+1)&7]] + b[Y2+d128[(i2+1)&7]])<<4);
602 }
603 }
604 break;
605 case IMGFMT_RG4B:
606 case IMGFMT_BG4B:
607 {
608 const uint8_t * const d64= dither_8x8_73 [y&7];
609 const uint8_t * const d128=dither_8x8_220[y&7];
610 YSCALE_YUV_2_RGBX_C(uint8_t)
611 ((uint8_t*)dest)[i2+0]= r[Y1+d128[(i2+0)&7]] + g[Y1+d64[(i2+0)&7]] + b[Y1+d128[(i2+0)&7]];
612 ((uint8_t*)dest)[i2+1]= r[Y2+d128[(i2+1)&7]] + g[Y2+d64[(i2+1)&7]] + b[Y2+d128[(i2+1)&7]];
613 }
614 }
615 break;
616 case IMGFMT_RGB1:
617 case IMGFMT_BGR1:
618 {
619 const uint8_t * const d128=dither_8x8_220[y&7];
620 uint8_t *g= c->table_gU[128] + c->table_gV[128];
621 int acc=0;
622 for(i=0; i<dstW-1; i+=2){
623 int j;
624 int Y1=1<<18;
625 int Y2=1<<18;
626
627 for(j=0; j<lumFilterSize; j++)
628 {
629 Y1 += lumSrc[j][i] * lumFilter[j];
630 Y2 += lumSrc[j][i+1] * lumFilter[j];
631 }
632 Y1>>=19;
633 Y2>>=19;
634 if((Y1|Y2)&256)
635 {
636 if(Y1>255) Y1=255;
637 else if(Y1<0)Y1=0;
638 if(Y2>255) Y2=255;
639 else if(Y2<0)Y2=0;
640 }
641 acc+= acc + g[Y1+d128[(i+0)&7]];
642 acc+= acc + g[Y2+d128[(i+1)&7]];
643 if((i&7)==6){
644 ((uint8_t*)dest)[0]= acc;
645 dest++;
646 }
647 }
648 }
649 break;
650 case IMGFMT_YUY2:
651 YSCALE_YUV_2_PACKEDX_C(void)
652 ((uint8_t*)dest)[2*i2+0]= Y1;
653 ((uint8_t*)dest)[2*i2+1]= U;
654 ((uint8_t*)dest)[2*i2+2]= Y2;
655 ((uint8_t*)dest)[2*i2+3]= V;
656 }
657 break;
658 case IMGFMT_UYVY:
659 YSCALE_YUV_2_PACKEDX_C(void)
660 ((uint8_t*)dest)[2*i2+0]= U;
661 ((uint8_t*)dest)[2*i2+1]= Y1;
662 ((uint8_t*)dest)[2*i2+2]= V;
663 ((uint8_t*)dest)[2*i2+3]= Y2;
664 }
665 break;
666 }
667 }
668
669
670 //Note: we have C, X86, MMX, MMX2, 3DNOW version therse no 3DNOW+MMX2 one
671 //Plain C versions
672 #if !defined (HAVE_MMX) || defined (RUNTIME_CPUDETECT)
673 #define COMPILE_C
674 #endif
675
676 #ifdef ARCH_POWERPC
677 #ifdef HAVE_ALTIVEC
678 #define COMPILE_ALTIVEC
679 #endif //HAVE_ALTIVEC
680 #endif //ARCH_POWERPC
681
682 #ifdef ARCH_X86
683
684 #if (defined (HAVE_MMX) && !defined (HAVE_3DNOW) && !defined (HAVE_MMX2)) || defined (RUNTIME_CPUDETECT)
685 #define COMPILE_MMX
686 #endif
687
688 #if defined (HAVE_MMX2) || defined (RUNTIME_CPUDETECT)
689 #define COMPILE_MMX2
690 #endif
691
692 #if (defined (HAVE_3DNOW) && !defined (HAVE_MMX2)) || defined (RUNTIME_CPUDETECT)
693 #define COMPILE_3DNOW
694 #endif
695 #endif //ARCH_X86
696
697 #undef HAVE_MMX
698 #undef HAVE_MMX2
699 #undef HAVE_3DNOW
700
701 #ifdef COMPILE_C
702 #undef HAVE_MMX
703 #undef HAVE_MMX2
704 #undef HAVE_3DNOW
705 #undef HAVE_ALTIVEC
706 #define RENAME(a) a ## _C
707 #include "swscale_template.c"
708 #endif
709
710 #ifdef ARCH_POWERPC
711 #ifdef COMPILE_ALTIVEC
712 #undef RENAME
713 #define HAVE_ALTIVEC
714 #define RENAME(a) a ## _altivec
715 #include "swscale_template.c"
716 #endif
717 #endif //ARCH_POWERPC
718
719 #ifdef ARCH_X86
720
721 //X86 versions
722 /*
723 #undef RENAME
724 #undef HAVE_MMX
725 #undef HAVE_MMX2
726 #undef HAVE_3DNOW
727 #define ARCH_X86
728 #define RENAME(a) a ## _X86
729 #include "swscale_template.c"
730 */
731 //MMX versions
732 #ifdef COMPILE_MMX
733 #undef RENAME
734 #define HAVE_MMX
735 #undef HAVE_MMX2
736 #undef HAVE_3DNOW
737 #define RENAME(a) a ## _MMX
738 #include "swscale_template.c"
739 #endif
740
741 //MMX2 versions
742 #ifdef COMPILE_MMX2
743 #undef RENAME
744 #define HAVE_MMX
745 #define HAVE_MMX2
746 #undef HAVE_3DNOW
747 #define RENAME(a) a ## _MMX2
748 #include "swscale_template.c"
749 #endif
750
751 //3DNOW versions
752 #ifdef COMPILE_3DNOW
753 #undef RENAME
754 #define HAVE_MMX
755 #undef HAVE_MMX2
756 #define HAVE_3DNOW
757 #define RENAME(a) a ## _3DNow
758 #include "swscale_template.c"
759 #endif
760
761 #endif //ARCH_X86
762
763 // minor note: the HAVE_xyz is messed up after that line so don't use it
764
765 static double getSplineCoeff(double a, double b, double c, double d, double dist)
766 {
767 // printf("%f %f %f %f %f\n", a,b,c,d,dist);
768 if(dist<=1.0) return ((d*dist + c)*dist + b)*dist +a;
769 else return getSplineCoeff( 0.0,
770 b+ 2.0*c + 3.0*d,
771 c + 3.0*d,
772 -b- 3.0*c - 6.0*d,
773 dist-1.0);
774 }
775
776 static inline void initFilter(int16_t **outFilter, int16_t **filterPos, int *outFilterSize, int xInc,
777 int srcW, int dstW, int filterAlign, int one, int flags,
778 SwsVector *srcFilter, SwsVector *dstFilter, double param[2])
779 {
780 int i;
781 int filterSize;
782 int filter2Size;
783 int minFilterSize;
784 double *filter=NULL;
785 double *filter2=NULL;
786 #ifdef ARCH_X86
787 if(flags & SWS_CPU_CAPS_MMX)
788 asm volatile("emms\n\t"::: "memory"); //FIXME this shouldnt be required but it IS (even for non mmx versions)
789 #endif
790
791 // Note the +1 is for the MMXscaler which reads over the end
792 *filterPos = (int16_t*)memalign(8, (dstW+1)*sizeof(int16_t));
793
794 if(ABS(xInc - 0x10000) <10) // unscaled
795 {
796 int i;
797 filterSize= 1;
798 filter= (double*)memalign(8, dstW*sizeof(double)*filterSize);
799 for(i=0; i<dstW*filterSize; i++) filter[i]=0;
800
801 for(i=0; i<dstW; i++)
802 {
803 filter[i*filterSize]=1;
804 (*filterPos)[i]=i;
805 }
806
807 }
808 else if(flags&SWS_POINT) // lame looking point sampling mode
809 {
810 int i;
811 int xDstInSrc;
812 filterSize= 1;
813 filter= (double*)memalign(8, dstW*sizeof(double)*filterSize);
814
815 xDstInSrc= xInc/2 - 0x8000;
816 for(i=0; i<dstW; i++)
817 {
818 int xx= (xDstInSrc - ((filterSize-1)<<15) + (1<<15))>>16;
819
820 (*filterPos)[i]= xx;
821 filter[i]= 1.0;
822 xDstInSrc+= xInc;
823 }
824 }
825 else if((xInc <= (1<<16) && (flags&SWS_AREA)) || (flags&SWS_FAST_BILINEAR)) // bilinear upscale
826 {
827 int i;
828 int xDstInSrc;
829 if (flags&SWS_BICUBIC) filterSize= 4;
830 else if(flags&SWS_X ) filterSize= 4;
831 else filterSize= 2; // SWS_BILINEAR / SWS_AREA
832 filter= (double*)memalign(8, dstW*sizeof(double)*filterSize);
833
834 xDstInSrc= xInc/2 - 0x8000;
835 for(i=0; i<dstW; i++)
836 {
837 int xx= (xDstInSrc - ((filterSize-1)<<15) + (1<<15))>>16;
838 int j;
839
840 (*filterPos)[i]= xx;
841 //Bilinear upscale / linear interpolate / Area averaging
842 for(j=0; j<filterSize; j++)
843 {
844 double d= ABS((xx<<16) - xDstInSrc)/(double)(1<<16);
845 double coeff= 1.0 - d;
846 if(coeff<0) coeff=0;
847 filter[i*filterSize + j]= coeff;
848 xx++;
849 }
850 xDstInSrc+= xInc;
851 }
852 }
853 else
854 {
855 double xDstInSrc;
856 double sizeFactor, filterSizeInSrc;
857 const double xInc1= (double)xInc / (double)(1<<16);
858
859 if (flags&SWS_BICUBIC) sizeFactor= 4.0;
860 else if(flags&SWS_X) sizeFactor= 8.0;
861 else if(flags&SWS_AREA) sizeFactor= 1.0; //downscale only, for upscale it is bilinear
862 else if(flags&SWS_GAUSS) sizeFactor= 8.0; // infinite ;)
863 else if(flags&SWS_LANCZOS) sizeFactor= param[0] != SWS_PARAM_DEFAULT ? 2.0*param[0] : 6.0;
864 else if(flags&SWS_SINC) sizeFactor= 20.0; // infinite ;)
865 else if(flags&SWS_SPLINE) sizeFactor= 20.0; // infinite ;)
866 else if(flags&SWS_BILINEAR) sizeFactor= 2.0;
867 else {
868 sizeFactor= 0.0; //GCC warning killer
869 ASSERT(0)
870 }
871
872 if(xInc1 <= 1.0) filterSizeInSrc= sizeFactor; // upscale
873 else filterSizeInSrc= sizeFactor*srcW / (double)dstW;
874
875 filterSize= (int)ceil(1 + filterSizeInSrc); // will be reduced later if possible
876 if(filterSize > srcW-2) filterSize=srcW-2;
877
878 filter= (double*)memalign(16, dstW*sizeof(double)*filterSize);
879
880 xDstInSrc= xInc1 / 2.0 - 0.5;
881 for(i=0; i<dstW; i++)
882 {
883 int xx= (int)(xDstInSrc - (filterSize-1)*0.5 + 0.5);
884 int j;
885 (*filterPos)[i]= xx;
886 for(j=0; j<filterSize; j++)
887 {
888 double d= ABS(xx - xDstInSrc)/filterSizeInSrc*sizeFactor;
889 double coeff;
890 if(flags & SWS_BICUBIC)
891 {
892 double B= param[0] != SWS_PARAM_DEFAULT ? param[0] : 0.0;
893 double C= param[1] != SWS_PARAM_DEFAULT ? param[1] : 0.6;
894
895 if(d<1.0)
896 coeff = (12-9*B-6*C)*d*d*d + (-18+12*B+6*C)*d*d + 6-2*B;
897 else if(d<2.0)
898 coeff = (-B-6*C)*d*d*d + (6*B+30*C)*d*d + (-12*B-48*C)*d +8*B+24*C;
899 else
900 coeff=0.0;
901 }
902 /* else if(flags & SWS_X)
903 {
904 double p= param ? param*0.01 : 0.3;
905 coeff = d ? sin(d*PI)/(d*PI) : 1.0;
906 coeff*= pow(2.0, - p*d*d);
907 }*/
908 else if(flags & SWS_X)
909 {
910 double A= param[0] != SWS_PARAM_DEFAULT ? param[0] : 1.0;
911
912 if(d<1.0)
913 coeff = cos(d*PI);
914 else
915 coeff=-1.0;
916 if(coeff<0.0) coeff= -pow(-coeff, A);
917 else coeff= pow( coeff, A);
918 coeff= coeff*0.5 + 0.5;
919 }
920 else if(flags & SWS_AREA)
921 {
922 double srcPixelSize= 1.0/xInc1;
923 if(d + srcPixelSize/2 < 0.5) coeff= 1.0;
924 else if(d - srcPixelSize/2 < 0.5) coeff= (0.5-d)/srcPixelSize + 0.5;
925 else coeff=0.0;
926 }
927 else if(flags & SWS_GAUSS)
928 {
929 double p= param[0] != SWS_PARAM_DEFAULT ? param[0] : 3.0;
930 coeff = pow(2.0, - p*d*d);
931 }
932 else if(flags & SWS_SINC)
933 {
934 coeff = d ? sin(d*PI)/(d*PI) : 1.0;
935 }
936 else if(flags & SWS_LANCZOS)
937 {
938 double p= param[0] != SWS_PARAM_DEFAULT ? param[0] : 3.0;
939 coeff = d ? sin(d*PI)*sin(d*PI/p)/(d*d*PI*PI/p) : 1.0;
940 if(d>p) coeff=0;
941 }
942 else if(flags & SWS_BILINEAR)
943 {
944 coeff= 1.0 - d;
945 if(coeff<0) coeff=0;
946 }
947 else if(flags & SWS_SPLINE)
948 {
949 double p=-2.196152422706632;
950 coeff = getSplineCoeff(1.0, 0.0, p, -p-1.0, d);
951 }
952 else {
953 coeff= 0.0; //GCC warning killer
954 ASSERT(0)
955 }
956
957 filter[i*filterSize + j]= coeff;
958 xx++;
959 }
960 xDstInSrc+= xInc1;
961 }
962 }
963
964 /* apply src & dst Filter to filter -> filter2
965 free(filter);
966 */
967 ASSERT(filterSize>0)
968 filter2Size= filterSize;
969 if(srcFilter) filter2Size+= srcFilter->length - 1;
970 if(dstFilter) filter2Size+= dstFilter->length - 1;
971 ASSERT(filter2Size>0)
972 filter2= (double*)memalign(8, filter2Size*dstW*sizeof(double));
973
974 for(i=0; i<dstW; i++)
975 {
976 int j;
977 SwsVector scaleFilter;
978 SwsVector *outVec;
979
980 scaleFilter.coeff= filter + i*filterSize;
981 scaleFilter.length= filterSize;
982
983 if(srcFilter) outVec= sws_getConvVec(srcFilter, &scaleFilter);
984 else outVec= &scaleFilter;
985
986 ASSERT(outVec->length == filter2Size)
987 //FIXME dstFilter
988
989 for(j=0; j<outVec->length; j++)
990 {
991 filter2[i*filter2Size + j]= outVec->coeff[j];
992 }
993
994 (*filterPos)[i]+= (filterSize-1)/2 - (filter2Size-1)/2;
995
996 if(outVec != &scaleFilter) sws_freeVec(outVec);
997 }
998 free(filter); filter=NULL;
999
1000 /* try to reduce the filter-size (step1 find size and shift left) */
1001 // Assume its near normalized (*0.5 or *2.0 is ok but * 0.001 is not)
1002 minFilterSize= 0;
1003 for(i=dstW-1; i>=0; i--)
1004 {
1005 int min= filter2Size;
1006 int j;
1007 double cutOff=0.0;
1008
1009 /* get rid off near zero elements on the left by shifting left */
1010 for(j=0; j<filter2Size; j++)
1011 {
1012 int k;
1013 cutOff += ABS(filter2[i*filter2Size]);
1014
1015 if(cutOff > SWS_MAX_REDUCE_CUTOFF) break;
1016
1017 /* preserve Monotonicity because the core can't handle the filter otherwise */
1018 if(i<dstW-1 && (*filterPos)[i] >= (*filterPos)[i+1]) break;
1019
1020 // Move filter coeffs left
1021 for(k=1; k<filter2Size; k++)
1022 filter2[i*filter2Size + k - 1]= filter2[i*filter2Size + k];
1023 filter2[i*filter2Size + k - 1]= 0.0;
1024 (*filterPos)[i]++;
1025 }
1026
1027 cutOff=0.0;
1028 /* count near zeros on the right */
1029 for(j=filter2Size-1; j>0; j--)
1030 {
1031 cutOff += ABS(filter2[i*filter2Size + j]);
1032
1033 if(cutOff > SWS_MAX_REDUCE_CUTOFF) break;
1034 min--;
1035 }
1036
1037 if(min>minFilterSize) minFilterSize= min;
1038 }
1039
1040 if (flags & SWS_CPU_CAPS_ALTIVEC) {
1041 // we can handle the special case 4,
1042 // so we don't want to go to the full 8
1043 if (minFilterSize < 5)
1044 filterAlign = 4;
1045
1046 // we really don't want to waste our time
1047 // doing useless computation, so fall-back on
1048 // the scalar C code for very small filter.
1049 // vectorizing is worth it only if you have
1050 // decent-sized vector.
1051 if (minFilterSize < 3)
1052 filterAlign = 1;
1053 }
1054
1055 ASSERT(minFilterSize > 0)
1056 filterSize= (minFilterSize +(filterAlign-1)) & (~(filterAlign-1));
1057 ASSERT(filterSize > 0)
1058 filter= (double*)memalign(8, filterSize*dstW*sizeof(double));
1059 *outFilterSize= filterSize;
1060
1061 if(flags&SWS_PRINT_INFO)
1062 MSG_INFO("SwScaler: reducing / aligning filtersize %d -> %d\n", filter2Size, filterSize);
1063 /* try to reduce the filter-size (step2 reduce it) */
1064 for(i=0; i<dstW; i++)
1065 {
1066 int j;
1067
1068 for(j=0; j<filterSize; j++)
1069 {
1070 if(j>=filter2Size) filter[i*filterSize + j]= 0.0;
1071 else filter[i*filterSize + j]= filter2[i*filter2Size + j];
1072 }
1073 }
1074 free(filter2); filter2=NULL;
1075
1076
1077 //FIXME try to align filterpos if possible
1078
1079 //fix borders
1080 for(i=0; i<dstW; i++)
1081 {
1082 int j;
1083 if((*filterPos)[i] < 0)
1084 {
1085 // Move filter coeffs left to compensate for filterPos
1086 for(j=1; j<filterSize; j++)
1087 {
1088 int left= MAX(j + (*filterPos)[i], 0);
1089 filter[i*filterSize + left] += filter[i*filterSize + j];
1090 filter[i*filterSize + j]=0;
1091 }
1092 (*filterPos)[i]= 0;
1093 }
1094
1095 if((*filterPos)[i] + filterSize > srcW)
1096 {
1097 int shift= (*filterPos)[i] + filterSize - srcW;
1098 // Move filter coeffs right to compensate for filterPos
1099 for(j=filterSize-2; j>=0; j--)
1100 {
1101 int right= MIN(j + shift, filterSize-1);
1102 filter[i*filterSize +right] += filter[i*filterSize +j];
1103 filter[i*filterSize +j]=0;
1104 }
1105 (*filterPos)[i]= srcW - filterSize;
1106 }
1107 }
1108
1109 // Note the +1 is for the MMXscaler which reads over the end
1110 *outFilter= (int16_t*)memalign(8, *outFilterSize*(dstW+1)*sizeof(int16_t));
1111 memset(*outFilter, 0, *outFilterSize*(dstW+1)*sizeof(int16_t));
1112
1113 /* Normalize & Store in outFilter */
1114 for(i=0; i<dstW; i++)
1115 {
1116 int j;
1117 double error=0;
1118 double sum=0;
1119 double scale= one;
1120
1121 for(j=0; j<filterSize; j++)
1122 {
1123 sum+= filter[i*filterSize + j];
1124 }
1125 scale/= sum;
1126 for(j=0; j<*outFilterSize; j++)
1127 {
1128 double v= filter[i*filterSize + j]*scale + error;
1129 int intV= floor(v + 0.5);
1130 (*outFilter)[i*(*outFilterSize) + j]= intV;
1131 error = v - intV;
1132 }
1133 }
1134
1135 (*filterPos)[dstW]= (*filterPos)[dstW-1]; // the MMX scaler will read over the end
1136 for(i=0; i<*outFilterSize; i++)
1137 {
1138 int j= dstW*(*outFilterSize);
1139 (*outFilter)[j + i]= (*outFilter)[j + i - (*outFilterSize)];
1140 }
1141
1142 free(filter);
1143 }
1144
1145 #ifdef ARCH_X86
1146 static void initMMX2HScaler(int dstW, int xInc, uint8_t *funnyCode, int16_t *filter, int32_t *filterPos, int numSplits)
1147 {
1148 uint8_t *fragmentA;
1149 int imm8OfPShufW1A;
1150 int imm8OfPShufW2A;
1151 int fragmentLengthA;
1152 uint8_t *fragmentB;
1153 int imm8OfPShufW1B;
1154 int imm8OfPShufW2B;
1155 int fragmentLengthB;
1156 int fragmentPos;
1157
1158 int xpos, i;
1159
1160 // create an optimized horizontal scaling routine
1161
1162 //code fragment
1163
1164 asm volatile(
1165 "jmp 9f \n\t"
1166 // Begin
1167 "0: \n\t"
1168 "movq (%%edx, %%eax), %%mm3 \n\t"
1169 "movd (%%ecx, %%esi), %%mm0 \n\t"
1170 "movd 1(%%ecx, %%esi), %%mm1 \n\t"
1171 "punpcklbw %%mm7, %%mm1 \n\t"
1172 "punpcklbw %%mm7, %%mm0 \n\t"
1173 "pshufw $0xFF, %%mm1, %%mm1 \n\t"
1174 "1: \n\t"
1175 "pshufw $0xFF, %%mm0, %%mm0 \n\t"
1176 "2: \n\t"
1177 "psubw %%mm1, %%mm0 \n\t"
1178 "movl 8(%%ebx, %%eax), %%esi \n\t"
1179 "pmullw %%mm3, %%mm0 \n\t"
1180 "psllw $7, %%mm1 \n\t"
1181 "paddw %%mm1, %%mm0 \n\t"
1182
1183 "movq %%mm0, (%%edi, %%eax) \n\t"
1184
1185 "addl $8, %%eax \n\t"
1186 // End
1187 "9: \n\t"
1188 // "int $3\n\t"
1189 "leal 0b, %0 \n\t"
1190 "leal 1b, %1 \n\t"
1191 "leal 2b, %2 \n\t"
1192 "decl %1 \n\t"
1193 "decl %2 \n\t"
1194 "subl %0, %1 \n\t"
1195 "subl %0, %2 \n\t"
1196 "leal 9b, %3 \n\t"
1197 "subl %0, %3 \n\t"
1198
1199
1200 :"=r" (fragmentA), "=r" (imm8OfPShufW1A), "=r" (imm8OfPShufW2A),
1201 "=r" (fragmentLengthA)
1202 );
1203
1204 asm volatile(
1205 "jmp 9f \n\t"
1206 // Begin
1207 "0: \n\t"
1208 "movq (%%edx, %%eax), %%mm3 \n\t"
1209 "movd (%%ecx, %%esi), %%mm0 \n\t"
1210 "punpcklbw %%mm7, %%mm0 \n\t"
1211 "pshufw $0xFF, %%mm0, %%mm1 \n\t"
1212 "1: \n\t"
1213 "pshufw $0xFF, %%mm0, %%mm0 \n\t"
1214 "2: \n\t"
1215 "psubw %%mm1, %%mm0 \n\t"
1216 "movl 8(%%ebx, %%eax), %%esi \n\t"
1217 "pmullw %%mm3, %%mm0 \n\t"
1218 "psllw $7, %%mm1 \n\t"
1219 "paddw %%mm1, %%mm0 \n\t"
1220
1221 "movq %%mm0, (%%edi, %%eax) \n\t"
1222
1223 "addl $8, %%eax \n\t"
1224 // End
1225 "9: \n\t"
1226 // "int $3\n\t"
1227 "leal 0b, %0 \n\t"
1228 "leal 1b, %1 \n\t"
1229 "leal 2b, %2 \n\t"
1230 "decl %1 \n\t"
1231 "decl %2 \n\t"
1232 "subl %0, %1 \n\t"
1233 "subl %0, %2 \n\t"
1234 "leal 9b, %3 \n\t"
1235 "subl %0, %3 \n\t"
1236
1237
1238 :"=r" (fragmentB), "=r" (imm8OfPShufW1B), "=r" (imm8OfPShufW2B),
1239 "=r" (fragmentLengthB)
1240 );
1241
1242 xpos= 0; //lumXInc/2 - 0x8000; // difference between pixel centers
1243 fragmentPos=0;
1244
1245 for(i=0; i<dstW/numSplits; i++)
1246 {
1247 int xx=xpos>>16;
1248
1249 if((i&3) == 0)
1250 {
1251 int a=0;
1252 int b=((xpos+xInc)>>16) - xx;
1253 int c=((xpos+xInc*2)>>16) - xx;
1254 int d=((xpos+xInc*3)>>16) - xx;
1255
1256 filter[i ] = (( xpos & 0xFFFF) ^ 0xFFFF)>>9;
1257 filter[i+1] = (((xpos+xInc ) & 0xFFFF) ^ 0xFFFF)>>9;
1258 filter[i+2] = (((xpos+xInc*2) & 0xFFFF) ^ 0xFFFF)>>9;
1259 filter[i+3] = (((xpos+xInc*3) & 0xFFFF) ^ 0xFFFF)>>9;
1260 filterPos[i/2]= xx;
1261
1262 if(d+1<4)
1263 {
1264 int maxShift= 3-(d+1);
1265 int shift=0;
1266
1267 memcpy(funnyCode + fragmentPos, fragmentB, fragmentLengthB);
1268
1269 funnyCode[fragmentPos + imm8OfPShufW1B]=
1270 (a+1) | ((b+1)<<2) | ((c+1)<<4) | ((d+1)<<6);
1271 funnyCode[fragmentPos + imm8OfPShufW2B]=
1272 a | (b<<2) | (c<<4) | (d<<6);
1273
1274 if(i+3>=dstW) shift=maxShift; //avoid overread
1275 else if((filterPos[i/2]&3) <= maxShift) shift=filterPos[i/2]&3; //Align
1276
1277 if(shift && i>=shift)
1278 {
1279 funnyCode[fragmentPos + imm8OfPShufW1B]+= 0x55*shift;
1280 funnyCode[fragmentPos + imm8OfPShufW2B]+= 0x55*shift;
1281 filterPos[i/2]-=shift;
1282 }
1283
1284 fragmentPos+= fragmentLengthB;
1285 }
1286 else
1287 {
1288 int maxShift= 3-d;
1289 int shift=0;
1290
1291 memcpy(funnyCode + fragmentPos, fragmentA, fragmentLengthA);
1292
1293 funnyCode[fragmentPos + imm8OfPShufW1A]=
1294 funnyCode[fragmentPos + imm8OfPShufW2A]=
1295 a | (b<<2) | (c<<4) | (d<<6);
1296
1297 if(i+4>=dstW) shift=maxShift; //avoid overread
1298 else if((filterPos[i/2]&3) <= maxShift) shift=filterPos[i/2]&3; //partial align
1299
1300 if(shift && i>=shift)
1301 {
1302 funnyCode[fragmentPos + imm8OfPShufW1A]+= 0x55*shift;
1303 funnyCode[fragmentPos + imm8OfPShufW2A]+= 0x55*shift;
1304 filterPos[i/2]-=shift;
1305 }
1306
1307 fragmentPos+= fragmentLengthA;
1308 }
1309
1310 funnyCode[fragmentPos]= RET;
1311 }
1312 xpos+=xInc;
1313 }
1314 filterPos[i/2]= xpos>>16; // needed to jump to the next part
1315 }
1316 #endif // ARCH_X86
1317
1318 static void globalInit(){
1319 // generating tables:
1320 int i;
1321 for(i=0; i<768; i++){
1322 int c= MIN(MAX(i-256, 0), 255);
1323 clip_table[i]=c;
1324 }
1325 }
1326
1327 static SwsFunc getSwsFunc(int flags){
1328
1329 #ifdef RUNTIME_CPUDETECT
1330 #ifdef ARCH_X86
1331 // ordered per speed fasterst first
1332 if(flags & SWS_CPU_CAPS_MMX2)
1333 return swScale_MMX2;
1334 else if(flags & SWS_CPU_CAPS_3DNOW)
1335 return swScale_3DNow;
1336 else if(flags & SWS_CPU_CAPS_MMX)
1337 return swScale_MMX;
1338 else
1339 return swScale_C;
1340
1341 #else
1342 #ifdef ARCH_POWERPC
1343 if(flags & SWS_CPU_CAPS_ALTIVEC)
1344 return swScale_altivec;
1345 else
1346 return swScale_C;
1347 #endif
1348 return swScale_C;
1349 #endif
1350 #else //RUNTIME_CPUDETECT
1351 #ifdef HAVE_MMX2
1352 return swScale_MMX2;
1353 #elif defined (HAVE_3DNOW)
1354 return swScale_3DNow;
1355 #elif defined (HAVE_MMX)
1356 return swScale_MMX;
1357 #elif defined (HAVE_ALTIVEC)
1358 return swScale_altivec;
1359 #else
1360 return swScale_C;
1361 #endif
1362 #endif //!RUNTIME_CPUDETECT
1363 }
1364
1365 static int PlanarToNV12Wrapper(SwsContext *c, uint8_t* src[], int srcStride[], int srcSliceY,
1366 int srcSliceH, uint8_t* dstParam[], int dstStride[]){
1367 uint8_t *dst=dstParam[0] + dstStride[0]*srcSliceY;
1368 /* Copy Y plane */
1369 if(dstStride[0]==srcStride[0])
1370 memcpy(dst, src[0], srcSliceH*dstStride[0]);
1371 else
1372 {
1373 int i;
1374 uint8_t *srcPtr= src[0];
1375 uint8_t *dstPtr= dst;
1376 for(i=0; i<srcSliceH; i++)
1377 {
1378 memcpy(dstPtr, srcPtr, srcStride[0]);
1379 srcPtr+= srcStride[0];
1380 dstPtr+= dstStride[0];
1381 }
1382 }
1383 dst = dstParam[1] + dstStride[1]*srcSliceY;
1384 interleaveBytes( src[1],src[2],dst,c->srcW,srcSliceH,srcStride[1],srcStride[2],dstStride[0] );
1385
1386 return srcSliceH;
1387 }
1388
1389 static int PlanarToYuy2Wrapper(SwsContext *c, uint8_t* src[], int srcStride[], int srcSliceY,
1390 int srcSliceH, uint8_t* dstParam[], int dstStride[]){
1391 uint8_t *dst=dstParam[0] + dstStride[0]*srcSliceY;
1392
1393 yv12toyuy2( src[0],src[1],src[2],dst,c->srcW,srcSliceH,srcStride[0],srcStride[1],dstStride[0] );
1394
1395 return srcSliceH;
1396 }
1397
1398 static int PlanarToUyvyWrapper(SwsContext *c, uint8_t* src[], int srcStride[], int srcSliceY,
1399 int srcSliceH, uint8_t* dstParam[], int dstStride[]){
1400 uint8_t *dst=dstParam[0] + dstStride[0]*srcSliceY;
1401
1402 yv12touyvy( src[0],src[1],src[2],dst,c->srcW,srcSliceH,srcStride[0],srcStride[1],dstStride[0] );
1403
1404 return srcSliceH;
1405 }
1406
1407 /* {RGB,BGR}{15,16,24,32} -> {RGB,BGR}{15,16,24,32} */
1408 static int rgb2rgbWrapper(SwsContext *c, uint8_t* src[], int srcStride[], int srcSliceY,
1409 int srcSliceH, uint8_t* dst[], int dstStride[]){
1410 const int srcFormat= c->srcFormat;
1411 const int dstFormat= c->dstFormat;
1412 const int srcBpp= ((srcFormat&0xFF) + 7)>>3;
1413 const int dstBpp= ((dstFormat&0xFF) + 7)>>3;
1414 const int srcId= (srcFormat&0xFF)>>2; // 1:0, 4:1, 8:2, 15:3, 16:4, 24:6, 32:8
1415 const int dstId= (dstFormat&0xFF)>>2;
1416 void (*conv)(const uint8_t *src, uint8_t *dst, unsigned src_size)=NULL;
1417
1418 /* BGR -> BGR */
1419 if( (isBGR(srcFormat) && isBGR(dstFormat))
1420 || (isRGB(srcFormat) && isRGB(dstFormat))){
1421 switch(srcId | (dstId<<4)){
1422 case 0x34: conv= rgb16to15; break;
1423 case 0x36: conv= rgb24to15; break;
1424 case 0x38: conv= rgb32to15; break;
1425 case 0x43: conv= rgb15to16; break;
1426 case 0x46: conv= rgb24to16; break;
1427 case 0x48: conv= rgb32to16; break;
1428 case 0x63: conv= rgb15to24; break;
1429 case 0x64: conv= rgb16to24; break;
1430 case 0x68: conv= rgb32to24; break;
1431 case 0x83: conv= rgb15to32; break;
1432 case 0x84: conv= rgb16to32; break;
1433 case 0x86: conv= rgb24to32; break;
1434 default: MSG_ERR("swScaler: internal error %s -> %s converter\n",
1435 vo_format_name(srcFormat), vo_format_name(dstFormat)); break;
1436 }
1437 }else if( (isBGR(srcFormat) && isRGB(dstFormat))
1438 || (isRGB(srcFormat) && isBGR(dstFormat))){
1439 switch(srcId | (dstId<<4)){
1440 case 0x33: conv= rgb15tobgr15; break;
1441 case 0x34: conv= rgb16tobgr15; break;
1442 case 0x36: conv= rgb24tobgr15; break;
1443 case 0x38: conv= rgb32tobgr15; break;
1444 case 0x43: conv= rgb15tobgr16; break;
1445 case 0x44: conv= rgb16tobgr16; break;
1446 case 0x46: conv= rgb24tobgr16; break;
1447 case 0x48: conv= rgb32tobgr16; break;
1448 case 0x63: conv= rgb15tobgr24; break;
1449 case 0x64: conv= rgb16tobgr24; break;
1450 case 0x66: conv= rgb24tobgr24; break;
1451 case 0x68: conv= rgb32tobgr24; break;
1452 case 0x83: conv= rgb15tobgr32; break;
1453 case 0x84: conv= rgb16tobgr32; break;
1454 case 0x86: conv= rgb24tobgr32; break;
1455 case 0x88: conv= rgb32tobgr32; break;
1456 default: MSG_ERR("swScaler: internal error %s -> %s converter\n",
1457 vo_format_name(srcFormat), vo_format_name(dstFormat)); break;
1458 }
1459 }else{
1460 MSG_ERR("swScaler: internal error %s -> %s converter\n",
1461 vo_format_name(srcFormat), vo_format_name(dstFormat));
1462 }
1463
1464 if(dstStride[0]*srcBpp == srcStride[0]*dstBpp)
1465 conv(src[0], dst[0] + dstStride[0]*srcSliceY, srcSliceH*srcStride[0]);
1466 else
1467 {
1468 int i;
1469 uint8_t *srcPtr= src[0];
1470 uint8_t *dstPtr= dst[0] + dstStride[0]*srcSliceY;
1471
1472 for(i=0; i<srcSliceH; i++)
1473 {
1474 conv(srcPtr, dstPtr, c->srcW*srcBpp);
1475 srcPtr+= srcStride[0];
1476 dstPtr+= dstStride[0];
1477 }
1478 }
1479 return srcSliceH;
1480 }
1481
1482 static int bgr24toyv12Wrapper(SwsContext *c, uint8_t* src[], int srcStride[], int srcSliceY,
1483 int srcSliceH, uint8_t* dst[], int dstStride[]){
1484
1485 rgb24toyv12(
1486 src[0],
1487 dst[0]+ srcSliceY *dstStride[0],
1488 dst[1]+(srcSliceY>>1)*dstStride[1],
1489 dst[2]+(srcSliceY>>1)*dstStride[2],
1490 c->srcW, srcSliceH,
1491 dstStride[0], dstStride[1], srcStride[0]);
1492 return srcSliceH;
1493 }
1494
1495 static int yvu9toyv12Wrapper(SwsContext *c, uint8_t* src[], int srcStride[], int srcSliceY,
1496 int srcSliceH, uint8_t* dst[], int dstStride[]){
1497 int i;
1498
1499 /* copy Y */
1500 if(srcStride[0]==dstStride[0])
1501 memcpy(dst[0]+ srcSliceY*dstStride[0], src[0], srcStride[0]*srcSliceH);
1502 else{
1503 uint8_t *srcPtr= src[0];
1504 uint8_t *dstPtr= dst[0] + dstStride[0]*srcSliceY;
1505
1506 for(i=0; i<srcSliceH; i++)
1507 {
1508 memcpy(dstPtr, srcPtr, c->srcW);
1509 srcPtr+= srcStride[0];
1510 dstPtr+= dstStride[0];
1511 }
1512 }
1513
1514 if(c->dstFormat==IMGFMT_YV12){
1515 planar2x(src[1], dst[1], c->chrSrcW, c->chrSrcH, srcStride[1], dstStride[1]);
1516 planar2x(src[2], dst[2], c->chrSrcW, c->chrSrcH, srcStride[2], dstStride[2]);
1517 }else{
1518 planar2x(src[1], dst[2], c->chrSrcW, c->chrSrcH, srcStride[1], dstStride[2]);
1519 planar2x(src[2], dst[1], c->chrSrcW, c->chrSrcH, srcStride[2], dstStride[1]);
1520 }
1521 return srcSliceH;
1522 }
1523
1524 /**
1525 * bring pointers in YUV order instead of YVU
1526 */
1527 static inline void sws_orderYUV(int format, uint8_t * sortedP[], int sortedStride[], uint8_t * p[], int stride[]){
1528 if(format == IMGFMT_YV12 || format == IMGFMT_YVU9
1529 || format == IMGFMT_444P || format == IMGFMT_422P || format == IMGFMT_411P){
1530 sortedP[0]= p[0];
1531 sortedP[1]= p[2];
1532 sortedP[2]= p[1];
1533 sortedStride[0]= stride[0];
1534 sortedStride[1]= stride[2];
1535 sortedStride[2]= stride[1];
1536 }
1537 else if(isPacked(format) || isGray(format) || format == IMGFMT_Y8)
1538 {
1539 sortedP[0]= p[0];
1540 sortedP[1]=
1541 sortedP[2]= NULL;
1542 sortedStride[0]= stride[0];
1543 sortedStride[1]=
1544 sortedStride[2]= 0;
1545 }
1546 else if(format == IMGFMT_I420 || format == IMGFMT_IYUV)
1547 {
1548 sortedP[0]= p[0];
1549 sortedP[1]= p[1];
1550 sortedP[2]= p[2];
1551 sortedStride[0]= stride[0];
1552 sortedStride[1]= stride[1];
1553 sortedStride[2]= stride[2];
1554 }else{
1555 MSG_ERR("internal error in orderYUV\n");
1556 }
1557 }
1558
1559 /* unscaled copy like stuff (assumes nearly identical formats) */
1560 static int simpleCopy(SwsContext *c, uint8_t* src[], int srcStride[], int srcSliceY,
1561 int srcSliceH, uint8_t* dst[], int dstStride[]){
1562
1563 if(isPacked(c->srcFormat))
1564 {
1565 if(dstStride[0]==srcStride[0])
1566 memcpy(dst[0] + dstStride[0]*srcSliceY, src[0], srcSliceH*dstStride[0]);
1567 else
1568 {
1569 int i;
1570 uint8_t *srcPtr= src[0];
1571 uint8_t *dstPtr= dst[0] + dstStride[0]*srcSliceY;
1572 int length=0;
1573
1574 /* universal length finder */
1575 while(length+c->srcW <= ABS(dstStride[0])
1576 && length+c->srcW <= ABS(srcStride[0])) length+= c->srcW;
1577 ASSERT(length!=0);
1578
1579 for(i=0; i<srcSliceH; i++)
1580 {
1581 memcpy(dstPtr, srcPtr, length);
1582 srcPtr+= srcStride[0];
1583 dstPtr+= dstStride[0];
1584 }
1585 }
1586 }
1587 else
1588 { /* Planar YUV or gray */
1589 int plane;
1590 for(plane=0; plane<3; plane++)
1591 {
1592 int length= plane==0 ? c->srcW : -((-c->srcW )>>c->chrDstHSubSample);
1593 int y= plane==0 ? srcSliceY: -((-srcSliceY)>>c->chrDstVSubSample);
1594 int height= plane==0 ? srcSliceH: -((-srcSliceH)>>c->chrDstVSubSample);
1595
1596 if((isGray(c->srcFormat) || isGray(c->dstFormat)) && plane>0)
1597 {
1598 if(!isGray(c->dstFormat))
1599 memset(dst[plane], 128, dstStride[plane]*height);
1600 }
1601 else
1602 {
1603 if(dstStride[plane]==srcStride[plane])
1604 memcpy(dst[plane] + dstStride[plane]*y, src[plane], height*dstStride[plane]);
1605 else
1606 {
1607 int i;
1608 uint8_t *srcPtr= src[plane];
1609 uint8_t *dstPtr= dst[plane] + dstStride[plane]*y;
1610 for(i=0; i<height; i++)
1611 {
1612 memcpy(dstPtr, srcPtr, length);
1613 srcPtr+= srcStride[plane];
1614 dstPtr+= dstStride[plane];
1615 }
1616 }
1617 }
1618 }
1619 }
1620 return srcSliceH;
1621 }
1622
1623 static int remove_dup_fourcc(int fourcc)
1624 {
1625 switch(fourcc)
1626 {
1627 case IMGFMT_I420:
1628 case IMGFMT_IYUV: return IMGFMT_YV12;
1629 case IMGFMT_Y8 : return IMGFMT_Y800;
1630 case IMGFMT_IF09: return IMGFMT_YVU9;
1631 default: return fourcc;
1632 }
1633 }
1634
1635 static void getSubSampleFactors(int *h, int *v, int format){
1636 switch(format){
1637 case IMGFMT_UYVY:
1638 case IMGFMT_YUY2:
1639 *h=1;
1640 *v=0;
1641 break;
1642 case IMGFMT_YV12:
1643 case IMGFMT_Y800: //FIXME remove after different subsamplings are fully implemented
1644 *h=1;
1645 *v=1;
1646 break;
1647 case IMGFMT_YVU9:
1648 *h=2;
1649 *v=2;
1650 break;
1651 case IMGFMT_444P:
1652 *h=0;
1653 *v=0;
1654 break;
1655 case IMGFMT_422P:
1656 *h=1;
1657 *v=0;
1658 break;
1659 case IMGFMT_411P:
1660 *h=2;
1661 *v=0;
1662 break;
1663 default:
1664 *h=0;
1665 *v=0;
1666 break;
1667 }
1668 }
1669
1670 static uint16_t roundToInt16(int64_t f){
1671 int r= (f + (1<<15))>>16;
1672 if(r<-0x7FFF) return 0x8000;
1673 else if(r> 0x7FFF) return 0x7FFF;
1674 else return r;
1675 }
1676
1677 /**
1678 * @param inv_table the yuv2rgb coeffs, normally Inverse_Table_6_9[x]
1679 * @param fullRange if 1 then the luma range is 0..255 if 0 its 16..235
1680 * @return -1 if not supported
1681 */
1682 int sws_setColorspaceDetails(SwsContext *c, const int inv_table[4], int srcRange, const int table[4], int dstRange, int brightness, int contrast, int saturation){
1683 int64_t crv = inv_table[0];
1684 int64_t cbu = inv_table[1];
1685 int64_t cgu = -inv_table[2];
1686 int64_t cgv = -inv_table[3];
1687 int64_t cy = 1<<16;
1688 int64_t oy = 0;
1689
1690 if(isYUV(c->dstFormat) || isGray(c->dstFormat)) return -1;
1691 memcpy(c->srcColorspaceTable, inv_table, sizeof(int)*4);
1692 memcpy(c->dstColorspaceTable, table, sizeof(int)*4);
1693
1694 c->brightness= brightness;
1695 c->contrast = contrast;
1696 c->saturation= saturation;
1697 c->srcRange = srcRange;
1698 c->dstRange = dstRange;
1699
1700 c->uOffset= 0x0400040004000400LL;
1701 c->vOffset= 0x0400040004000400LL;
1702
1703 if(!srcRange){
1704 cy= (cy*255) / 219;
1705 oy= 16<<16;
1706 }
1707
1708 cy = (cy *contrast )>>16;
1709 crv= (crv*contrast * saturation)>>32;
1710 cbu= (cbu*contrast * saturation)>>32;
1711 cgu= (cgu*contrast * saturation)>>32;
1712 cgv= (cgv*contrast * saturation)>>32;
1713
1714 oy -= 256*brightness;
1715
1716 c->yCoeff= roundToInt16(cy *8192) * 0x0001000100010001ULL;
1717 c->vrCoeff= roundToInt16(crv*8192) * 0x0001000100010001ULL;
1718 c->ubCoeff= roundToInt16(cbu*8192) * 0x0001000100010001ULL;
1719 c->vgCoeff= roundToInt16(cgv*8192) * 0x0001000100010001ULL;
1720 c->ugCoeff= roundToInt16(cgu*8192) * 0x0001000100010001ULL;
1721 c->yOffset= roundToInt16(oy * 8) * 0x0001000100010001ULL;
1722
1723 yuv2rgb_c_init_tables(c, inv_table, srcRange, brightness, contrast, saturation);
1724 //FIXME factorize
1725
1726 #ifdef HAVE_ALTIVEC
1727 yuv2rgb_altivec_init_tables (c, inv_table);
1728 #endif
1729 return 0;
1730 }
1731
1732 /**
1733 * @return -1 if not supported
1734 */
1735 int sws_getColorspaceDetails(SwsContext *c, int **inv_table, int *srcRange, int **table, int *dstRange, int *brightness, int *contrast, int *saturation){
1736 if(isYUV(c->dstFormat) || isGray(c->dstFormat)) return -1;
1737
1738 *inv_table = c->srcColorspaceTable;
1739 *table = c->dstColorspaceTable;
1740 *srcRange = c->srcRange;
1741 *dstRange = c->dstRange;
1742 *brightness= c->brightness;
1743 *contrast = c->contrast;
1744 *saturation= c->saturation;
1745
1746 return 0;
1747 }
1748
1749 SwsContext *sws_getContext(int srcW, int srcH, int origSrcFormat, int dstW, int dstH, int origDstFormat, int flags,
1750 SwsFilter *srcFilter, SwsFilter *dstFilter, double *param){
1751
1752 SwsContext *c;
1753 int i;
1754 int usesVFilter, usesHFilter;
1755 int unscaled, needsDither;
1756 int srcFormat, dstFormat;
1757 SwsFilter dummyFilter= {NULL, NULL, NULL, NULL};
1758 #ifdef ARCH_X86
1759 if(flags & SWS_CPU_CAPS_MMX)
1760 asm volatile("emms\n\t"::: "memory");
1761 #endif
1762
1763 #ifndef RUNTIME_CPUDETECT //ensure that the flags match the compiled variant if cpudetect is off
1764 flags &= ~(SWS_CPU_CAPS_MMX|SWS_CPU_CAPS_MMX2|SWS_CPU_CAPS_3DNOW|SWS_CPU_CAPS_ALTIVEC);
1765 #ifdef HAVE_MMX2
1766 flags |= SWS_CPU_CAPS_MMX|SWS_CPU_CAPS_MMX2;
1767 #elif defined (HAVE_3DNOW)
1768 flags |= SWS_CPU_CAPS_MMX|SWS_CPU_CAPS_3DNOW;
1769 #elif defined (HAVE_MMX)
1770 flags |= SWS_CPU_CAPS_MMX;
1771 #elif defined (HAVE_ALTIVEC)
1772 flags |= SWS_CPU_CAPS_ALTIVEC;
1773 #endif
1774 #endif
1775 if(clip_table[512] != 255) globalInit();
1776 if(rgb15to16 == NULL) sws_rgb2rgb_init(flags);
1777
1778 /* avoid duplicate Formats, so we don't need to check to much */
1779 srcFormat = remove_dup_fourcc(origSrcFormat);
1780 dstFormat = remove_dup_fourcc(origDstFormat);
1781
1782 unscaled = (srcW == dstW && srcH == dstH);
1783 needsDither= (isBGR(dstFormat) || isRGB(dstFormat))
1784 && (dstFormat&0xFF)<24
1785 && ((dstFormat&0xFF)<(srcFormat&0xFF) || (!(isRGB(srcFormat) || isBGR(srcFormat))));
1786
1787 if(!isSupportedIn(srcFormat))
1788 {
1789 MSG_ERR("swScaler: %s is not supported as input format\n", vo_format_name(srcFormat));
1790 return NULL;
1791 }
1792 if(!isSupportedOut(dstFormat))
1793 {
1794 MSG_ERR("swScaler: %s is not supported as output format\n", vo_format_name(dstFormat));
1795 return NULL;
1796 }
1797
1798 /* sanity check */
1799 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
1800 {
1801 MSG_ERR("swScaler: %dx%d -> %dx%d is invalid scaling dimension\n",
1802 srcW, srcH, dstW, dstH);
1803 return NULL;
1804 }
1805
1806 if(!dstFilter) dstFilter= &dummyFilter;
1807 if(!srcFilter) srcFilter= &dummyFilter;
1808
1809 c= memalign(64, sizeof(SwsContext));
1810 memset(c, 0, sizeof(SwsContext));
1811
1812 c->srcW= srcW;
1813 c->srcH= srcH;
1814 c->dstW= dstW;
1815 c->dstH= dstH;
1816 c->lumXInc= ((srcW<<16) + (dstW>>1))/dstW;
1817 c->lumYInc= ((srcH<<16) + (dstH>>1))/dstH;
1818 c->flags= flags;
1819 c->dstFormat= dstFormat;
1820 c->srcFormat= srcFormat;
1821 c->origDstFormat= origDstFormat;
1822 c->origSrcFormat= origSrcFormat;
1823 c->vRounder= 4* 0x0001000100010001ULL;
1824
1825 usesHFilter= usesVFilter= 0;
1826 if(dstFilter->lumV!=NULL && dstFilter->lumV->length>1) usesVFilter=1;
1827 if(dstFilter->lumH!=NULL && dstFilter->lumH->length>1) usesHFilter=1;
1828 if(dstFilter->chrV!=NULL && dstFilter->chrV->length>1) usesVFilter=1;
1829 if(dstFilter->chrH!=NULL && dstFilter->chrH->length>1) usesHFilter=1;
1830 if(srcFilter->lumV!=NULL && srcFilter->lumV->length>1) usesVFilter=1;
1831 if(srcFilter->lumH!=NULL && srcFilter->lumH->length>1) usesHFilter=1;
1832 if(srcFilter->chrV!=NULL && srcFilter->chrV->length>1) usesVFilter=1;
1833 if(srcFilter->chrH!=NULL && srcFilter->chrH->length>1) usesHFilter=1;
1834
1835 getSubSampleFactors(&c->chrSrcHSubSample, &c->chrSrcVSubSample, srcFormat);
1836 getSubSampleFactors(&c->chrDstHSubSample, &c->chrDstVSubSample, dstFormat);
1837
1838 // reuse chroma for 2 pixles rgb/bgr unless user wants full chroma interpolation
1839 if((isBGR(dstFormat) || isRGB(dstFormat)) && !(flags&SWS_FULL_CHR_H_INT)) c->chrDstHSubSample=1;
1840
1841 // drop some chroma lines if the user wants it
1842 c->vChrDrop= (flags&SWS_SRC_V_CHR_DROP_MASK)>>SWS_SRC_V_CHR_DROP_SHIFT;
1843 c->chrSrcVSubSample+= c->vChrDrop;
1844
1845 // drop every 2. pixel for chroma calculation unless user wants full chroma
1846 if((isBGR(srcFormat) || isRGB(srcFormat)) && !(flags&SWS_FULL_CHR_H_INP))
1847 c->chrSrcHSubSample=1;
1848
1849 if(param){
1850 c->param[0] = param[0];
1851 c->param[1] = param[1];
1852 }else{
1853 c->param[0] =
1854 c->param[1] = SWS_PARAM_DEFAULT;
1855 }
1856
1857 c->chrIntHSubSample= c->chrDstHSubSample;
1858 c->chrIntVSubSample= c->chrSrcVSubSample;
1859
1860 // note the -((-x)>>y) is so that we allways round toward +inf
1861 c->chrSrcW= -((-srcW) >> c->chrSrcHSubSample);
1862 c->chrSrcH= -((-srcH) >> c->chrSrcVSubSample);
1863 c->chrDstW= -((-dstW) >> c->chrDstHSubSample);
1864 c->chrDstH= -((-dstH) >> c->chrDstVSubSample);
1865
1866 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);
1867
1868 /* unscaled special Cases */
1869 if(unscaled && !usesHFilter && !usesVFilter)
1870 {
1871 /* yv12_to_nv12 */
1872 if(srcFormat == IMGFMT_YV12 && dstFormat == IMGFMT_NV12)
1873 {
1874 c->swScale= PlanarToNV12Wrapper;
1875 }
1876 /* yuv2bgr */
1877 if((srcFormat==IMGFMT_YV12 || srcFormat==IMGFMT_422P) && (isBGR(dstFormat) || isRGB(dstFormat)))
1878 {
1879 c->swScale= yuv2rgb_get_func_ptr(c);
1880 }
1881
1882 if( srcFormat==IMGFMT_YVU9 && dstFormat==IMGFMT_YV12 )
1883 {
1884 c->swScale= yvu9toyv12Wrapper;
1885 }
1886
1887 /* bgr24toYV12 */
1888 if(srcFormat==IMGFMT_BGR24 && dstFormat==IMGFMT_YV12)
1889 c->swScale= bgr24toyv12Wrapper;
1890
1891 /* rgb/bgr -> rgb/bgr (no dither needed forms) */
1892 if( (isBGR(srcFormat) || isRGB(srcFormat))
1893 && (isBGR(dstFormat) || isRGB(dstFormat))
1894 && !needsDither)
1895 c->swScale= rgb2rgbWrapper;
1896
1897 /* LQ converters if -sws 0 or -sws 4*/
1898 if(c->flags&(SWS_FAST_BILINEAR|SWS_POINT)){
1899 /* rgb/bgr -> rgb/bgr (dither needed forms) */
1900 if( (isBGR(srcFormat) || isRGB(srcFormat))
1901 && (isBGR(dstFormat) || isRGB(dstFormat))
1902 && needsDither)
1903 c->swScale= rgb2rgbWrapper;
1904
1905 /* yv12_to_yuy2 */
1906 if(srcFormat == IMGFMT_YV12 &&
1907 (dstFormat == IMGFMT_YUY2 || dstFormat == IMGFMT_UYVY))
1908 {
1909 if (dstFormat == IMGFMT_YUY2)
1910 c->swScale= PlanarToYuy2Wrapper;
1911 else
1912 c->swScale= PlanarToUyvyWrapper;
1913 }
1914 }
1915
1916 #ifdef HAVE_ALTIVEC
1917 if ((c->flags & SWS_CPU_CAPS_ALTIVEC) &&
1918 ((srcFormat == IMGFMT_YV12 &&
1919 (dstFormat == IMGFMT_YUY2 || dstFormat == IMGFMT_UYVY)))) {
1920 // unscaled YV12 -> packed YUV, we want speed
1921 if (dstFormat == IMGFMT_YUY2)
1922 c->swScale= yv12toyuy2_unscaled_altivec;
1923 else
1924 c->swScale= yv12touyvy_unscaled_altivec;
1925 }
1926 #endif
1927
1928 /* simple copy */
1929 if( srcFormat == dstFormat
1930 || (isPlanarYUV(srcFormat) && isGray(dstFormat))
1931 || (isPlanarYUV(dstFormat) && isGray(srcFormat))
1932 )
1933 {
1934 c->swScale= simpleCopy;
1935 }
1936
1937 if(c->swScale){
1938 if(flags&SWS_PRINT_INFO)
1939 MSG_INFO("SwScaler: using unscaled %s -> %s special converter\n",
1940 vo_format_name(srcFormat), vo_format_name(dstFormat));
1941 return c;
1942 }
1943 }
1944
1945 if(flags & SWS_CPU_CAPS_MMX2)
1946 {
1947 c->canMMX2BeUsed= (dstW >=srcW && (dstW&31)==0 && (srcW&15)==0) ? 1 : 0;
1948 if(!c->canMMX2BeUsed && dstW >=srcW && (srcW&15)==0 && (flags&SWS_FAST_BILINEAR))
1949 {
1950 if(flags&SWS_PRINT_INFO)
1951 MSG_INFO("SwScaler: output Width is not a multiple of 32 -> no MMX2 scaler\n");
1952 }
1953 if(usesHFilter) c->canMMX2BeUsed=0;
1954 }
1955 else
1956 c->canMMX2BeUsed=0;
1957
1958 c->chrXInc= ((c->chrSrcW<<16) + (c->chrDstW>>1))/c->chrDstW;
1959 c->chrYInc= ((c->chrSrcH<<16) + (c->chrDstH>>1))/c->chrDstH;
1960
1961 // match pixel 0 of the src to pixel 0 of dst and match pixel n-2 of src to pixel n-2 of dst
1962 // but only for the FAST_BILINEAR mode otherwise do correct scaling
1963 // n-2 is the last chrominance sample available
1964 // this is not perfect, but noone shuld notice the difference, the more correct variant
1965 // would be like the vertical one, but that would require some special code for the
1966 // first and last pixel
1967 if(flags&SWS_FAST_BILINEAR)
1968 {
1969 if(c->canMMX2BeUsed)
1970 {
1971 c->lumXInc+= 20;
1972 c->chrXInc+= 20;
1973 }
1974 //we don't use the x86asm scaler if mmx is available
1975 else if(flags & SWS_CPU_CAPS_MMX)
1976 {
1977 c->lumXInc = ((srcW-2)<<16)/(dstW-2) - 20;
1978 c->chrXInc = ((c->chrSrcW-2)<<16)/(c->chrDstW-2) - 20;
1979 }
1980 }
1981
1982 /* precalculate horizontal scaler filter coefficients */
1983 {
1984 const int filterAlign=
1985 (flags & SWS_CPU_CAPS_MMX) ? 4 :
1986 (flags & SWS_CPU_CAPS_ALTIVEC) ? 8 :
1987 1;
1988
1989 initFilter(&c->hLumFilter, &c->hLumFilterPos, &c->hLumFilterSize, c->lumXInc,
1990 srcW , dstW, filterAlign, 1<<14,
1991 (flags&SWS_BICUBLIN) ? (flags|SWS_BICUBIC) : flags,
1992 srcFilter->lumH, dstFilter->lumH, c->param);
1993 initFilter(&c->hChrFilter, &c->hChrFilterPos, &c->hChrFilterSize, c->chrXInc,
1994 c->chrSrcW, c->chrDstW, filterAlign, 1<<14,
1995 (flags&SWS_BICUBLIN) ? (flags|SWS_BILINEAR) : flags,
1996 srcFilter->chrH, dstFilter->chrH, c->param);
1997
1998 #ifdef ARCH_X86
1999 // can't downscale !!!
2000 if(c->canMMX2BeUsed && (flags & SWS_FAST_BILINEAR))
2001 {
2002 c->lumMmx2Filter = (int16_t*)memalign(8, (dstW /8+8)*sizeof(int16_t));
2003 c->chrMmx2Filter = (int16_t*)memalign(8, (c->chrDstW /4+8)*sizeof(int16_t));
2004 c->lumMmx2FilterPos= (int32_t*)memalign(8, (dstW /2/8+8)*sizeof(int32_t));
2005 c->chrMmx2FilterPos= (int32_t*)memalign(8, (c->chrDstW/2/4+8)*sizeof(int32_t));
2006
2007 initMMX2HScaler( dstW, c->lumXInc, c->funnyYCode , c->lumMmx2Filter, c->lumMmx2FilterPos, 8);
2008 initMMX2HScaler(c->chrDstW, c->chrXInc, c->funnyUVCode, c->chrMmx2Filter, c->chrMmx2FilterPos, 4);
2009 }
2010 #endif
2011 } // Init Horizontal stuff
2012
2013
2014
2015 /* precalculate vertical scaler filter coefficients */
2016 {
2017 const int filterAlign=
2018 (flags & SWS_CPU_CAPS_ALTIVEC) ? 8 :
2019 1;
2020
2021 initFilter(&c->vLumFilter, &c->vLumFilterPos, &c->vLumFilterSize, c->lumYInc,
2022 srcH , dstH, filterAlign, (1<<12)-4,
2023 (flags&SWS_BICUBLIN) ? (flags|SWS_BICUBIC) : flags,
2024 srcFilter->lumV, dstFilter->lumV, c->param);
2025 initFilter(&c->vChrFilter, &c->vChrFilterPos, &c->vChrFilterSize, c->chrYInc,
2026 c->chrSrcH, c->chrDstH, filterAlign, (1<<12)-4,
2027 (flags&SWS_BICUBLIN) ? (flags|SWS_BILINEAR) : flags,
2028 srcFilter->chrV, dstFilter->chrV, c->param);
2029 }
2030
2031 // Calculate Buffer Sizes so that they won't run out while handling these damn slices
2032 c->vLumBufSize= c->vLumFilterSize;
2033 c->vChrBufSize= c->vChrFilterSize;
2034 for(i=0; i<dstH; i++)
2035 {
2036 int chrI= i*c->chrDstH / dstH;
2037 int nextSlice= MAX(c->vLumFilterPos[i ] + c->vLumFilterSize - 1,
2038 ((c->vChrFilterPos[chrI] + c->vChrFilterSize - 1)<<c->chrSrcVSubSample));
2039
2040 nextSlice>>= c->chrSrcVSubSample;
2041 nextSlice<<= c->chrSrcVSubSample;
2042 if(c->vLumFilterPos[i ] + c->vLumBufSize < nextSlice)
2043 c->vLumBufSize= nextSlice - c->vLumFilterPos[i ];
2044 if(c->vChrFilterPos[chrI] + c->vChrBufSize < (nextSlice>>c->chrSrcVSubSample))
2045 c->vChrBufSize= (nextSlice>>c->chrSrcVSubSample) - c->vChrFilterPos[chrI];
2046 }
2047
2048 // allocate pixbufs (we use dynamic allocation because otherwise we would need to
2049 c->lumPixBuf= (int16_t**)memalign(4, c->vLumBufSize*2*sizeof(int16_t*));
2050 c->chrPixBuf= (int16_t**)memalign(4, c->vChrBufSize*2*sizeof(int16_t*));
2051 //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)
2052 for(i=0; i<c->vLumBufSize; i++)
2053 c->lumPixBuf[i]= c->lumPixBuf[i+c->vLumBufSize]= (uint16_t*)memalign(8, 4000);
2054 for(i=0; i<c->vChrBufSize; i++)
2055 c->chrPixBuf[i]= c->chrPixBuf[i+c->vChrBufSize]= (uint16_t*)memalign(8, 8000);
2056
2057 //try to avoid drawing green stuff between the right end and the stride end
2058 for(i=0; i<c->vLumBufSize; i++) memset(c->lumPixBuf[i], 0, 4000);
2059 for(i=0; i<c->vChrBufSize; i++) memset(c->chrPixBuf[i], 64, 8000);
2060
2061 ASSERT(c->chrDstH <= dstH)
2062
2063 if(flags&SWS_PRINT_INFO)
2064 {
2065 #ifdef DITHER1XBPP
2066 char *dither= " dithered";
2067 #else
2068 char *dither= "";
2069 #endif
2070 if(flags&SWS_FAST_BILINEAR)
2071 MSG_INFO("\nSwScaler: FAST_BILINEAR scaler, ");
2072 else if(flags&SWS_BILINEAR)
2073 MSG_INFO("\nSwScaler: BILINEAR scaler, ");
2074 else if(flags&SWS_BICUBIC)
2075 MSG_INFO("\nSwScaler: BICUBIC scaler, ");
2076 else if(flags&SWS_X)
2077 MSG_INFO("\nSwScaler: Experimental scaler, ");
2078 else if(flags&SWS_POINT)
2079 MSG_INFO("\nSwScaler: Nearest Neighbor / POINT scaler, ");
2080 else if(flags&SWS_AREA)
2081 MSG_INFO("\nSwScaler: Area Averageing scaler, ");
2082 else if(flags&SWS_BICUBLIN)
2083 MSG_INFO("\nSwScaler: luma BICUBIC / chroma BILINEAR scaler, ");
2084 else if(flags&SWS_GAUSS)
2085 MSG_INFO("\nSwScaler: Gaussian scaler, ");
2086 else if(flags&SWS_SINC)
2087 MSG_INFO("\nSwScaler: Sinc scaler, ");
2088 else if(flags&SWS_LANCZOS)
2089 MSG_INFO("\nSwScaler: Lanczos scaler, ");
2090 else if(flags&SWS_SPLINE)
2091 MSG_INFO("\nSwScaler: Bicubic spline scaler, ");
2092 else
2093 MSG_INFO("\nSwScaler: ehh flags invalid?! ");
2094
2095 if(dstFormat==IMGFMT_BGR15 || dstFormat==IMGFMT_BGR16)
2096 MSG_INFO("from %s to%s %s ",
2097 vo_format_name(srcFormat), dither, vo_format_name(dstFormat));
2098 else
2099 MSG_INFO("from %s to %s ",
2100 vo_format_name(srcFormat), vo_format_name(dstFormat));
2101
2102 if(flags & SWS_CPU_CAPS_MMX2)
2103 MSG_INFO("using MMX2\n");
2104 else if(flags & SWS_CPU_CAPS_3DNOW)
2105 MSG_INFO("using 3DNOW\n");
2106 else if(flags & SWS_CPU_CAPS_MMX)
2107 MSG_INFO("using MMX\n");
2108 else if(flags & SWS_CPU_CAPS_ALTIVEC)
2109 MSG_INFO("using AltiVec\n");
2110 else
2111 MSG_INFO("using C\n");
2112 }
2113
2114 if(flags & SWS_PRINT_INFO)
2115 {
2116 if(flags & SWS_CPU_CAPS_MMX)
2117 {
2118 if(c->canMMX2BeUsed && (flags&SWS_FAST_BILINEAR))
2119 MSG_V("SwScaler: using FAST_BILINEAR MMX2 scaler for horizontal scaling\n");
2120 else
2121 {
2122 if(c->hLumFilterSize==4)
2123 MSG_V("SwScaler: using 4-tap MMX scaler for horizontal luminance scaling\n");
2124 else if(c->hLumFilterSize==8)
2125 MSG_V("SwScaler: using 8-tap MMX scaler for horizontal luminance scaling\n");
2126 else
2127 MSG_V("SwScaler: using n-tap MMX scaler for horizontal luminance scaling\n");
2128
2129 if(c->hChrFilterSize==4)
2130 MSG_V("SwScaler: using 4-tap MMX scaler for horizontal chrominance scaling\n");
2131 else if(c->hChrFilterSize==8)
2132 MSG_V("SwScaler: using 8-tap MMX scaler for horizontal chrominance scaling\n");
2133 else
2134 MSG_V("SwScaler: using n-tap MMX scaler for horizontal chrominance scaling\n");
2135 }
2136 }
2137 else
2138 {
2139 #ifdef ARCH_X86
2140 MSG_V("SwScaler: using X86-Asm scaler for horizontal scaling\n");
2141 #else
2142 if(flags & SWS_FAST_BILINEAR)
2143 MSG_V("SwScaler: using FAST_BILINEAR C scaler for horizontal scaling\n");
2144 else
2145 MSG_V("SwScaler: using C scaler for horizontal scaling\n");
2146 #endif
2147 }
2148 if(isPlanarYUV(dstFormat))
2149 {
2150 if(c->vLumFilterSize==1)
2151 MSG_V("SwScaler: using 1-tap %s \"scaler\" for vertical scaling (YV12 like)\n", (flags & SWS_CPU_CAPS_MMX) ? "MMX" : "C");
2152 else
2153 MSG_V("SwScaler: using n-tap %s scaler for vertical scaling (YV12 like)\n", (flags & SWS_CPU_CAPS_MMX) ? "MMX" : "C");
2154 }
2155 else
2156 {
2157 if(c->vLumFilterSize==1 && c->vChrFilterSize==2)
2158 MSG_V("SwScaler: using 1-tap %s \"scaler\" for vertical luminance scaling (BGR)\n"
2159 "SwScaler: 2-tap scaler for vertical chrominance scaling (BGR)\n",(flags & SWS_CPU_CAPS_MMX) ? "MMX" : "C");
2160 else if(c->vLumFilterSize==2 && c->vChrFilterSize==2)
2161 MSG_V("SwScaler: using 2-tap linear %s scaler for vertical scaling (BGR)\n", (flags & SWS_CPU_CAPS_MMX) ? "MMX" : "C");
2162 else
2163 MSG_V("SwScaler: using n-tap %s scaler for vertical scaling (BGR)\n", (flags & SWS_CPU_CAPS_MMX) ? "MMX" : "C");
2164 }
2165
2166 if(dstFormat==IMGFMT_BGR24)
2167 MSG_V("SwScaler: using %s YV12->BGR24 Converter\n",
2168 (flags & SWS_CPU_CAPS_MMX2) ? "MMX2" : ((flags & SWS_CPU_CAPS_MMX) ? "MMX" : "C"));
2169 else if(dstFormat==IMGFMT_BGR32)
2170 MSG_V("SwScaler: using %s YV12->BGR32 Converter\n", (flags & SWS_CPU_CAPS_MMX) ? "MMX" : "C");
2171 else if(dstFormat==IMGFMT_BGR16)
2172 MSG_V("SwScaler: using %s YV12->BGR16 Converter\n", (flags & SWS_CPU_CAPS_MMX) ? "MMX" : "C");
2173 else if(dstFormat==IMGFMT_BGR15)
2174 MSG_V("SwScaler: using %s YV12->BGR15 Converter\n", (flags & SWS_CPU_CAPS_MMX) ? "MMX" : "C");
2175
2176 MSG_V("SwScaler: %dx%d -> %dx%d\n", srcW, srcH, dstW, dstH);
2177 }
2178 if(flags & SWS_PRINT_INFO)
2179 {
2180 MSG_DBG2("SwScaler:Lum srcW=%d srcH=%d dstW=%d dstH=%d xInc=%d yInc=%d\n",
2181 c->srcW, c->srcH, c->dstW, c->dstH, c->lumXInc, c->lumYInc);
2182 MSG_DBG2("SwScaler:Chr srcW=%d srcH=%d dstW=%d dstH=%d xInc=%d yInc=%d\n",
2183 c->chrSrcW, c->chrSrcH, c->chrDstW, c->chrDstH, c->chrXInc, c->chrYInc);
2184 }
2185
2186 c->swScale= getSwsFunc(flags);
2187 return c;
2188 }
2189
2190 /**
2191 * swscale warper, so we don't need to export the SwsContext.
2192 * assumes planar YUV to be in YUV order instead of YVU
2193 */
2194 int sws_scale_ordered(SwsContext *c, uint8_t* src[], int srcStride[], int srcSliceY,
2195 int srcSliceH, uint8_t* dst[], int dstStride[]){
2196 //copy strides, so they can safely be modified
2197 int srcStride2[3]= {srcStride[0], srcStride[1], srcStride[2]};
2198 int dstStride2[3]= {dstStride[0], dstStride[1], dstStride[2]};
2199 return c->swScale(c, src, srcStride2, srcSliceY, srcSliceH, dst, dstStride2);
2200 }
2201
2202 /**
2203 * swscale warper, so we don't need to export the SwsContext
2204 */
2205 int sws_scale(SwsContext *c, uint8_t* srcParam[], int srcStrideParam[], int srcSliceY,
2206 int srcSliceH, uint8_t* dstParam[], int dstStrideParam[]){
2207 int srcStride[3];
2208 int dstStride[3];
2209 uint8_t *src[3];
2210 uint8_t *dst[3];
2211 sws_orderYUV(c->origSrcFormat, src, srcStride, srcParam, srcStrideParam);
2212 sws_orderYUV(c->origDstFormat, dst, dstStride, dstParam, dstStrideParam);
2213 //printf("sws: slice %d %d\n", srcSliceY, srcSliceH);
2214
2215 return c->swScale(c, src, srcStride, srcSliceY, srcSliceH, dst, dstStride);
2216 }
2217
2218 SwsFilter *sws_getDefaultFilter(float lumaGBlur, float chromaGBlur,
2219 float lumaSharpen, float chromaSharpen,
2220 float chromaHShift, float chromaVShift,
2221 int verbose)
2222 {
2223 SwsFilter *filter= malloc(sizeof(SwsFilter));
2224
2225 if(lumaGBlur!=0.0){
2226 filter->lumH= sws_getGaussianVec(lumaGBlur, 3.0);
2227 filter->lumV= sws_getGaussianVec(lumaGBlur, 3.0);
2228 }else{
2229 filter->lumH= sws_getIdentityVec();
2230 filter->lumV= sws_getIdentityVec();
2231 }
2232
2233 if(chromaGBlur!=0.0){
2234 filter->chrH= sws_getGaussianVec(chromaGBlur, 3.0);
2235 filter->chrV= sws_getGaussianVec(chromaGBlur, 3.0);
2236 }else{
2237 filter->chrH= sws_getIdentityVec();
2238 filter->chrV= sws_getIdentityVec();
2239 }
2240
2241 if(chromaSharpen!=0.0){
2242 SwsVector *g= sws_getConstVec(-1.0, 3);
2243 SwsVector *id= sws_getConstVec(10.0/chromaSharpen, 1);
2244 g->coeff[1]=2.0;
2245 sws_addVec(id, g);
2246 sws_convVec(filter->chrH, id);
2247 sws_convVec(filter->chrV, id);
2248 sws_freeVec(g);
2249 sws_freeVec(id);
2250 }
2251
2252 if(lumaSharpen!=0.0){
2253 SwsVector *g= sws_getConstVec(-1.0, 3);
2254 SwsVector *id= sws_getConstVec(10.0/lumaSharpen, 1);
2255 g->coeff[1]=2.0;
2256 sws_addVec(id, g);
2257 sws_convVec(filter->lumH, id);
2258 sws_convVec(filter->lumV, id);
2259 sws_freeVec(g);
2260 sws_freeVec(id);
2261 }
2262
2263 if(chromaHShift != 0.0)
2264 sws_shiftVec(filter->chrH, (int)(chromaHShift+0.5));
2265
2266 if(chromaVShift != 0.0)
2267 sws_shiftVec(filter->chrV, (int)(chromaVShift+0.5));
2268
2269 sws_normalizeVec(filter->chrH, 1.0);
2270 sws_normalizeVec(filter->chrV, 1.0);
2271 sws_normalizeVec(filter->lumH, 1.0);
2272 sws_normalizeVec(filter->lumV, 1.0);
2273
2274 if(verbose) sws_printVec(filter->chrH);
2275 if(verbose) sws_printVec(filter->lumH);
2276
2277 return filter;
2278 }
2279
2280 /**
2281 * returns a normalized gaussian curve used to filter stuff
2282 * quality=3 is high quality, lowwer is lowwer quality
2283 */
2284 SwsVector *sws_getGaussianVec(double variance, double quality){
2285 const int length= (int)(variance*quality + 0.5) | 1;
2286 int i;
2287 double *coeff= memalign(sizeof(double), length*sizeof(double));
2288 double middle= (length-1)*0.5;
2289 SwsVector *vec= malloc(sizeof(SwsVector));
2290
2291 vec->coeff= coeff;
2292 vec->length= length;
2293
2294 for(i=0; i<length; i++)
2295 {
2296 double dist= i-middle;
2297 coeff[i]= exp( -dist*dist/(2*variance*variance) ) / sqrt(2*variance*PI);
2298 }
2299
2300 sws_normalizeVec(vec, 1.0);
2301
2302 return vec;
2303 }
2304
2305 SwsVector *sws_getConstVec(double c, int length){
2306 int i;
2307 double *coeff= memalign(sizeof(double), length*sizeof(double));
2308 SwsVector *vec= malloc(sizeof(SwsVector));
2309
2310 vec->coeff= coeff;
2311 vec->length= length;
2312
2313 for(i=0; i<length; i++)
2314 coeff[i]= c;
2315
2316 return vec;
2317 }
2318
2319
2320 SwsVector *sws_getIdentityVec(void){
2321 double *coeff= memalign(sizeof(double), sizeof(double));
2322 SwsVector *vec= malloc(sizeof(SwsVector));
2323 coeff[0]= 1.0;
2324
2325 vec->coeff= coeff;
2326 vec->length= 1;
2327
2328 return vec;
2329 }
2330
2331 void sws_normalizeVec(SwsVector *a, double height){
2332 int i;
2333 double sum=0;
2334 double inv;
2335
2336 for(i=0; i<a->length; i++)
2337 sum+= a->coeff[i];
2338
2339 inv= height/sum;
2340
2341 for(i=0; i<a->length; i++)
2342 a->coeff[i]*= inv;
2343 }
2344
2345 void sws_scaleVec(SwsVector *a, double scalar){
2346 int i;
2347
2348 for(i=0; i<a->length; i++)
2349 a->coeff[i]*= scalar;
2350 }
2351
2352 static SwsVector *sws_getConvVec(SwsVector *a, SwsVector *b){
2353 int length= a->length + b->length - 1;
2354 double *coeff= memalign(sizeof(double), length*sizeof(double));
2355 int i, j;
2356 SwsVector *vec= malloc(sizeof(SwsVector));
2357
2358 vec->coeff= coeff;
2359 vec->length= length;
2360
2361 for(i=0; i<length; i++) coeff[i]= 0.0;
2362
2363 for(i=0; i<a->length; i++)
2364 {
2365 for(j=0; j<b->length; j++)
2366 {
2367 coeff[i+j]+= a->coeff[i]*b->coeff[j];
2368 }
2369 }
2370
2371 return vec;
2372 }
2373
2374 static SwsVector *sws_sumVec(SwsVector *a, SwsVector *b){
2375 int length= MAX(a->length, b->length);
2376 double *coeff= memalign(sizeof(double), length*sizeof(double));
2377 int i;
2378 SwsVector *vec= malloc(sizeof(SwsVector));
2379
2380 vec->coeff= coeff;
2381 vec->length= length;
2382
2383 for(i=0; i<length; i++) coeff[i]= 0.0;
2384
2385 for(i=0; i<a->length; i++) coeff[i + (length-1)/2 - (a->length-1)/2]+= a->coeff[i];
2386 for(i=0; i<b->length; i++) coeff[i + (length-1)/2 - (b->length-1)/2]+= b->coeff[i];
2387
2388 return vec;
2389 }
2390
2391 static SwsVector *sws_diffVec(SwsVector *a, SwsVector *b){
2392 int length= MAX(a->length, b->length);
2393 double *coeff= memalign(sizeof(double), length*sizeof(double));
2394 int i;
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++) coeff[i + (length-1)/2 - (a->length-1)/2]+= a->coeff[i];
2403 for(i=0; i<b->length; i++) coeff[i + (length-1)/2 - (b->length-1)/2]-= b->coeff[i];
2404
2405 return vec;
2406 }
2407
2408 /* shift left / or right if "shift" is negative */
2409 static SwsVector *sws_getShiftedVec(SwsVector *a, int shift){
2410 int length= a->length + ABS(shift)*2;
2411 double *coeff= memalign(sizeof(double), length*sizeof(double));
2412 int i;
2413 SwsVector *vec= malloc(sizeof(SwsVector));
2414
2415 vec->coeff= coeff;
2416 vec->length= length;
2417
2418 for(i=0; i<length; i++) coeff[i]= 0.0;
2419
2420 for(i=0; i<a->length; i++)
2421 {
2422 coeff[i + (length-1)/2 - (a->length-1)/2 - shift]= a->coeff[i];
2423 }
2424
2425 return vec;
2426 }
2427
2428 void sws_shiftVec(SwsVector *a, int shift){
2429 SwsVector *shifted= sws_getShiftedVec(a, shift);
2430 free(a->coeff);
2431 a->coeff= shifted->coeff;
2432 a->length= shifted->length;
2433 free(shifted);
2434 }
2435
2436 void sws_addVec(SwsVector *a, SwsVector *b){
2437 SwsVector *sum= sws_sumVec(a, b);
2438 free(a->coeff);
2439 a->coeff= sum->coeff;
2440 a->length= sum->length;
2441 free(sum);
2442 }
2443
2444 void sws_subVec(SwsVector *a, SwsVector *b){
2445 SwsVector *diff= sws_diffVec(a, b);
2446 free(a->coeff);
2447 a->coeff= diff->coeff;
2448 a->length= diff->length;
2449 free(diff);
2450 }
2451
2452 void sws_convVec(SwsVector *a, SwsVector *b){
2453 SwsVector *conv= sws_getConvVec(a, b);
2454 free(a->coeff);
2455 a->coeff= conv->coeff;
2456 a->length= conv->length;
2457 free(conv);
2458 }
2459
2460 SwsVector *sws_cloneVec(SwsVector *a){
2461 double *coeff= memalign(sizeof(double), a->length*sizeof(double));
2462 int i;
2463 SwsVector *vec= malloc(sizeof(SwsVector));
2464
2465 vec->coeff= coeff;
2466 vec->length= a->length;
2467
2468 for(i=0; i<a->length; i++) coeff[i]= a->coeff[i];
2469
2470 return vec;
2471 }
2472
2473 void sws_printVec(SwsVector *a){
2474 int i;
2475 double max=0;
2476 double min=0;
2477 double range;
2478
2479 for(i=0; i<a->length; i++)
2480 if(a->coeff[i]>max) max= a->coeff[i];
2481
2482 for(i=0; i<a->length; i++)
2483 if(a->coeff[i]<min) min= a->coeff[i];
2484
2485 range= max - min;
2486
2487 for(i=0; i<a->length; i++)
2488 {
2489 int x= (int)((a->coeff[i]-min)*60.0/range +0.5);
2490 MSG_DBG2("%1.3f ", a->coeff[i]);
2491 for(;x>0; x--) MSG_DBG2(" ");
2492 MSG_DBG2("|\n");
2493 }
2494 }
2495
2496 void sws_freeVec(SwsVector *a){
2497 if(!a) return;
2498 if(a->coeff) free(a->coeff);
2499 a->coeff=NULL;
2500 a->length=0;
2501 free(a);
2502 }
2503
2504 void sws_freeFilter(SwsFilter *filter){
2505 if(!filter) return;
2506
2507 if(filter->lumH) sws_freeVec(filter->lumH);
2508 if(filter->lumV) sws_freeVec(filter->lumV);
2509 if(filter->chrH) sws_freeVec(filter->chrH);
2510 if(filter->chrV) sws_freeVec(filter->chrV);
2511 free(filter);
2512 }
2513
2514
2515 void sws_freeContext(SwsContext *c){
2516 int i;
2517 if(!c) return;
2518
2519 if(c->lumPixBuf)
2520 {
2521 for(i=0; i<c->vLumBufSize; i++)
2522 {
2523 if(c->lumPixBuf[i]) free(c->lumPixBuf[i]);
2524 c->lumPixBuf[i]=NULL;
2525 }
2526 free(c->lumPixBuf);
2527 c->lumPixBuf=NULL;
2528 }
2529
2530 if(c->chrPixBuf)
2531 {
2532 for(i=0; i<c->vChrBufSize; i++)
2533 {
2534 if(c->chrPixBuf[i]) free(c->chrPixBuf[i]);
2535 c->chrPixBuf[i]=NULL;
2536 }
2537 free(c->chrPixBuf);
2538 c->chrPixBuf=NULL;
2539 }
2540
2541 if(c->vLumFilter) free(c->vLumFilter);
2542 c->vLumFilter = NULL;
2543 if(c->vChrFilter) free(c->vChrFilter);
2544 c->vChrFilter = NULL;
2545 if(c->hLumFilter) free(c->hLumFilter);
2546 c->hLumFilter = NULL;
2547 if(c->hChrFilter) free(c->hChrFilter);
2548 c->hChrFilter = NULL;
2549
2550 if(c->vLumFilterPos) free(c->vLumFilterPos);
2551 c->vLumFilterPos = NULL;
2552 if(c->vChrFilterPos) free(c->vChrFilterPos);
2553 c->vChrFilterPos = NULL;
2554 if(c->hLumFilterPos) free(c->hLumFilterPos);
2555 c->hLumFilterPos = NULL;
2556 if(c->hChrFilterPos) free(c->hChrFilterPos);
2557 c->hChrFilterPos = NULL;
2558
2559 if(c->lumMmx2Filter) free(c->lumMmx2Filter);
2560 c->lumMmx2Filter=NULL;
2561 if(c->chrMmx2Filter) free(c->chrMmx2Filter);
2562 c->chrMmx2Filter=NULL;
2563 if(c->lumMmx2FilterPos) free(c->lumMmx2FilterPos);
2564 c->lumMmx2FilterPos=NULL;
2565 if(c->chrMmx2FilterPos) free(c->chrMmx2FilterPos);
2566 c->chrMmx2FilterPos=NULL;
2567 if(c->yuvTable) free(c->yuvTable);
2568 c->yuvTable=NULL;
2569
2570 free(c);
2571 }
2572