Fix #endif comments.
[libav.git] / libswscale / swscale.c
1 /*
2 * Copyright (C) 2001-2003 Michael Niedermayer <michaelni@gmx.at>
3 *
4 * This file is part of FFmpeg.
5 *
6 * FFmpeg is free software; you can redistribute it and/or modify
7 * it under the terms of the GNU General Public License as published by
8 * the Free Software Foundation; either version 2 of the License, or
9 * (at your option) any later version.
10 *
11 * FFmpeg is distributed in the hope that it will be useful,
12 * but WITHOUT ANY WARRANTY; without even the implied warranty of
13 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
14 * GNU General Public License for more details.
15 *
16 * You should have received a copy of the GNU General Public License
17 * along with FFmpeg; if not, write to the Free Software
18 * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
19 *
20 * the C code (not assembly, mmx, ...) of this file can be used
21 * under the LGPL license too
22 */
23
24 /*
25 supported Input formats: YV12, I420/IYUV, YUY2, UYVY, BGR32, BGR32_1, BGR24, BGR16, BGR15, RGB32, RGB32_1, RGB24, Y8/Y800, YVU9/IF09, PAL8
26 supported output formats: YV12, I420/IYUV, YUY2, UYVY, {BGR,RGB}{1,4,8,15,16,24,32}, Y8/Y800, YVU9/IF09
27 {BGR,RGB}{1,4,8,15,16} support dithering
28
29 unscaled special converters (YV12=I420=IYUV, Y800=Y8)
30 YV12 -> {BGR,RGB}{1,4,8,15,16,24,32}
31 x -> x
32 YUV9 -> YV12
33 YUV9/YV12 -> Y800
34 Y800 -> YUV9/YV12
35 BGR24 -> BGR32 & RGB24 -> RGB32
36 BGR32 -> BGR24 & RGB32 -> RGB24
37 BGR15 -> BGR16
38 */
39
40 /*
41 tested special converters (most are tested actually, but I did not write it down ...)
42 YV12 -> BGR16
43 YV12 -> YV12
44 BGR15 -> BGR16
45 BGR16 -> BGR16
46 YVU9 -> YV12
47
48 untested special converters
49 YV12/I420 -> BGR15/BGR24/BGR32 (it is the yuv2rgb stuff, so it should be ok)
50 YV12/I420 -> YV12/I420
51 YUY2/BGR15/BGR24/BGR32/RGB24/RGB32 -> same format
52 BGR24 -> BGR32 & RGB24 -> RGB32
53 BGR32 -> BGR24 & RGB32 -> RGB24
54 BGR24 -> YV12
55 */
56
57 #define _SVID_SOURCE //needed for MAP_ANONYMOUS
58 #include <inttypes.h>
59 #include <string.h>
60 #include <math.h>
61 #include <stdio.h>
62 #include <unistd.h>
63 #include "config.h"
64 #include <assert.h>
65 #if HAVE_SYS_MMAN_H
66 #include <sys/mman.h>
67 #if defined(MAP_ANON) && !defined(MAP_ANONYMOUS)
68 #define MAP_ANONYMOUS MAP_ANON
69 #endif
70 #endif
71 #include "swscale.h"
72 #include "swscale_internal.h"
73 #include "rgb2rgb.h"
74 #include "libavutil/x86_cpu.h"
75 #include "libavutil/bswap.h"
76
77 unsigned swscale_version(void)
78 {
79 return LIBSWSCALE_VERSION_INT;
80 }
81
82 #undef MOVNTQ
83 #undef PAVGB
84
85 //#undef HAVE_MMX2
86 //#define HAVE_3DNOW
87 //#undef HAVE_MMX
88 //#undef ARCH_X86
89 //#define WORDS_BIGENDIAN
90 #define DITHER1XBPP
91
92 #define FAST_BGR2YV12 // use 7 bit coeffs instead of 15bit
93
94 #define RET 0xC3 //near return opcode for X86
95
96 #ifdef M_PI
97 #define PI M_PI
98 #else
99 #define PI 3.14159265358979323846
100 #endif
101
102 #define isSupportedIn(x) ( \
103 (x)==PIX_FMT_YUV420P \
104 || (x)==PIX_FMT_YUVA420P \
105 || (x)==PIX_FMT_YUYV422 \
106 || (x)==PIX_FMT_UYVY422 \
107 || (x)==PIX_FMT_RGB32 \
108 || (x)==PIX_FMT_RGB32_1 \
109 || (x)==PIX_FMT_BGR24 \
110 || (x)==PIX_FMT_BGR565 \
111 || (x)==PIX_FMT_BGR555 \
112 || (x)==PIX_FMT_BGR32 \
113 || (x)==PIX_FMT_BGR32_1 \
114 || (x)==PIX_FMT_RGB24 \
115 || (x)==PIX_FMT_RGB565 \
116 || (x)==PIX_FMT_RGB555 \
117 || (x)==PIX_FMT_GRAY8 \
118 || (x)==PIX_FMT_YUV410P \
119 || (x)==PIX_FMT_YUV440P \
120 || (x)==PIX_FMT_GRAY16BE \
121 || (x)==PIX_FMT_GRAY16LE \
122 || (x)==PIX_FMT_YUV444P \
123 || (x)==PIX_FMT_YUV422P \
124 || (x)==PIX_FMT_YUV411P \
125 || (x)==PIX_FMT_PAL8 \
126 || (x)==PIX_FMT_BGR8 \
127 || (x)==PIX_FMT_RGB8 \
128 || (x)==PIX_FMT_BGR4_BYTE \
129 || (x)==PIX_FMT_RGB4_BYTE \
130 || (x)==PIX_FMT_YUV440P \
131 || (x)==PIX_FMT_MONOWHITE \
132 || (x)==PIX_FMT_MONOBLACK \
133 )
134 #define isSupportedOut(x) ( \
135 (x)==PIX_FMT_YUV420P \
136 || (x)==PIX_FMT_YUYV422 \
137 || (x)==PIX_FMT_UYVY422 \
138 || (x)==PIX_FMT_YUV444P \
139 || (x)==PIX_FMT_YUV422P \
140 || (x)==PIX_FMT_YUV411P \
141 || isRGB(x) \
142 || isBGR(x) \
143 || (x)==PIX_FMT_NV12 \
144 || (x)==PIX_FMT_NV21 \
145 || (x)==PIX_FMT_GRAY16BE \
146 || (x)==PIX_FMT_GRAY16LE \
147 || (x)==PIX_FMT_GRAY8 \
148 || (x)==PIX_FMT_YUV410P \
149 || (x)==PIX_FMT_YUV440P \
150 )
151 #define isPacked(x) ( \
152 (x)==PIX_FMT_PAL8 \
153 || (x)==PIX_FMT_YUYV422 \
154 || (x)==PIX_FMT_UYVY422 \
155 || isRGB(x) \
156 || isBGR(x) \
157 )
158 #define usePal(x) ( \
159 (x)==PIX_FMT_PAL8 \
160 || (x)==PIX_FMT_BGR4_BYTE \
161 || (x)==PIX_FMT_RGB4_BYTE \
162 || (x)==PIX_FMT_BGR8 \
163 || (x)==PIX_FMT_RGB8 \
164 )
165
166 #define RGB2YUV_SHIFT 15
167 #define BY ( (int)(0.114*219/255*(1<<RGB2YUV_SHIFT)+0.5))
168 #define BV (-(int)(0.081*224/255*(1<<RGB2YUV_SHIFT)+0.5))
169 #define BU ( (int)(0.500*224/255*(1<<RGB2YUV_SHIFT)+0.5))
170 #define GY ( (int)(0.587*219/255*(1<<RGB2YUV_SHIFT)+0.5))
171 #define GV (-(int)(0.419*224/255*(1<<RGB2YUV_SHIFT)+0.5))
172 #define GU (-(int)(0.331*224/255*(1<<RGB2YUV_SHIFT)+0.5))
173 #define RY ( (int)(0.299*219/255*(1<<RGB2YUV_SHIFT)+0.5))
174 #define RV ( (int)(0.500*224/255*(1<<RGB2YUV_SHIFT)+0.5))
175 #define RU (-(int)(0.169*224/255*(1<<RGB2YUV_SHIFT)+0.5))
176
177 extern const int32_t Inverse_Table_6_9[8][4];
178
179 static const double rgb2yuv_table[8][9]={
180 {0.7152, 0.0722, 0.2126, -0.386, 0.5, -0.115, -0.454, -0.046, 0.5},
181 {0.7152, 0.0722, 0.2126, -0.386, 0.5, -0.115, -0.454, -0.046, 0.5},
182 {0.587 , 0.114 , 0.299 , -0.331, 0.5, -0.169, -0.419, -0.081, 0.5},
183 {0.587 , 0.114 , 0.299 , -0.331, 0.5, -0.169, -0.419, -0.081, 0.5},
184 {0.59 , 0.11 , 0.30 , -0.331, 0.5, -0.169, -0.421, -0.079, 0.5}, //FCC
185 {0.587 , 0.114 , 0.299 , -0.331, 0.5, -0.169, -0.419, -0.081, 0.5},
186 {0.587 , 0.114 , 0.299 , -0.331, 0.5, -0.169, -0.419, -0.081, 0.5}, //SMPTE 170M
187 {0.701 , 0.087 , 0.212 , -0.384, 0.5 -0.116, -0.445, -0.055, 0.5}, //SMPTE 240M
188 };
189
190 /*
191 NOTES
192 Special versions: fast Y 1:1 scaling (no interpolation in y direction)
193
194 TODO
195 more intelligent misalignment avoidance for the horizontal scaler
196 write special vertical cubic upscale version
197 Optimize C code (yv12 / minmax)
198 add support for packed pixel yuv input & output
199 add support for Y8 output
200 optimize bgr24 & bgr32
201 add BGR4 output support
202 write special BGR->BGR scaler
203 */
204
205 #if ARCH_X86 && CONFIG_GPL
206 DECLARE_ASM_CONST(8, uint64_t, bF8)= 0xF8F8F8F8F8F8F8F8LL;
207 DECLARE_ASM_CONST(8, uint64_t, bFC)= 0xFCFCFCFCFCFCFCFCLL;
208 DECLARE_ASM_CONST(8, uint64_t, w10)= 0x0010001000100010LL;
209 DECLARE_ASM_CONST(8, uint64_t, w02)= 0x0002000200020002LL;
210 DECLARE_ASM_CONST(8, uint64_t, bm00001111)=0x00000000FFFFFFFFLL;
211 DECLARE_ASM_CONST(8, uint64_t, bm00000111)=0x0000000000FFFFFFLL;
212 DECLARE_ASM_CONST(8, uint64_t, bm11111000)=0xFFFFFFFFFF000000LL;
213 DECLARE_ASM_CONST(8, uint64_t, bm01010101)=0x00FF00FF00FF00FFLL;
214
215 const DECLARE_ALIGNED(8, uint64_t, ff_dither4[2]) = {
216 0x0103010301030103LL,
217 0x0200020002000200LL,};
218
219 const DECLARE_ALIGNED(8, uint64_t, ff_dither8[2]) = {
220 0x0602060206020602LL,
221 0x0004000400040004LL,};
222
223 DECLARE_ASM_CONST(8, uint64_t, b16Mask)= 0x001F001F001F001FLL;
224 DECLARE_ASM_CONST(8, uint64_t, g16Mask)= 0x07E007E007E007E0LL;
225 DECLARE_ASM_CONST(8, uint64_t, r16Mask)= 0xF800F800F800F800LL;
226 DECLARE_ASM_CONST(8, uint64_t, b15Mask)= 0x001F001F001F001FLL;
227 DECLARE_ASM_CONST(8, uint64_t, g15Mask)= 0x03E003E003E003E0LL;
228 DECLARE_ASM_CONST(8, uint64_t, r15Mask)= 0x7C007C007C007C00LL;
229
230 DECLARE_ALIGNED(8, const uint64_t, ff_M24A) = 0x00FF0000FF0000FFLL;
231 DECLARE_ALIGNED(8, const uint64_t, ff_M24B) = 0xFF0000FF0000FF00LL;
232 DECLARE_ALIGNED(8, const uint64_t, ff_M24C) = 0x0000FF0000FF0000LL;
233
234 #ifdef FAST_BGR2YV12
235 DECLARE_ALIGNED(8, const uint64_t, ff_bgr2YCoeff) = 0x000000210041000DULL;
236 DECLARE_ALIGNED(8, const uint64_t, ff_bgr2UCoeff) = 0x0000FFEEFFDC0038ULL;
237 DECLARE_ALIGNED(8, const uint64_t, ff_bgr2VCoeff) = 0x00000038FFD2FFF8ULL;
238 #else
239 DECLARE_ALIGNED(8, const uint64_t, ff_bgr2YCoeff) = 0x000020E540830C8BULL;
240 DECLARE_ALIGNED(8, const uint64_t, ff_bgr2UCoeff) = 0x0000ED0FDAC23831ULL;
241 DECLARE_ALIGNED(8, const uint64_t, ff_bgr2VCoeff) = 0x00003831D0E6F6EAULL;
242 #endif /* FAST_BGR2YV12 */
243 DECLARE_ALIGNED(8, const uint64_t, ff_bgr2YOffset) = 0x1010101010101010ULL;
244 DECLARE_ALIGNED(8, const uint64_t, ff_bgr2UVOffset) = 0x8080808080808080ULL;
245 DECLARE_ALIGNED(8, const uint64_t, ff_w1111) = 0x0001000100010001ULL;
246
247 DECLARE_ASM_CONST(8, uint64_t, ff_bgr24toY1Coeff) = 0x0C88000040870C88ULL;
248 DECLARE_ASM_CONST(8, uint64_t, ff_bgr24toY2Coeff) = 0x20DE4087000020DEULL;
249 DECLARE_ASM_CONST(8, uint64_t, ff_rgb24toY1Coeff) = 0x20DE0000408720DEULL;
250 DECLARE_ASM_CONST(8, uint64_t, ff_rgb24toY2Coeff) = 0x0C88408700000C88ULL;
251 DECLARE_ASM_CONST(8, uint64_t, ff_bgr24toYOffset) = 0x0008400000084000ULL;
252
253 DECLARE_ASM_CONST(8, uint64_t, ff_bgr24toUV[2][4]) = {
254 {0x38380000DAC83838ULL, 0xECFFDAC80000ECFFULL, 0xF6E40000D0E3F6E4ULL, 0x3838D0E300003838ULL},
255 {0xECFF0000DAC8ECFFULL, 0x3838DAC800003838ULL, 0x38380000D0E33838ULL, 0xF6E4D0E30000F6E4ULL},
256 };
257
258 DECLARE_ASM_CONST(8, uint64_t, ff_bgr24toUVOffset)= 0x0040400000404000ULL;
259
260 #endif /* ARCH_X86 && CONFIG_GPL */
261
262 // clipping helper table for C implementations:
263 static unsigned char clip_table[768];
264
265 static SwsVector *sws_getConvVec(SwsVector *a, SwsVector *b);
266
267 static const uint8_t __attribute__((aligned(8))) dither_2x2_4[2][8]={
268 { 1, 3, 1, 3, 1, 3, 1, 3, },
269 { 2, 0, 2, 0, 2, 0, 2, 0, },
270 };
271
272 static const uint8_t __attribute__((aligned(8))) dither_2x2_8[2][8]={
273 { 6, 2, 6, 2, 6, 2, 6, 2, },
274 { 0, 4, 0, 4, 0, 4, 0, 4, },
275 };
276
277 const uint8_t __attribute__((aligned(8))) dither_8x8_32[8][8]={
278 { 17, 9, 23, 15, 16, 8, 22, 14, },
279 { 5, 29, 3, 27, 4, 28, 2, 26, },
280 { 21, 13, 19, 11, 20, 12, 18, 10, },
281 { 0, 24, 6, 30, 1, 25, 7, 31, },
282 { 16, 8, 22, 14, 17, 9, 23, 15, },
283 { 4, 28, 2, 26, 5, 29, 3, 27, },
284 { 20, 12, 18, 10, 21, 13, 19, 11, },
285 { 1, 25, 7, 31, 0, 24, 6, 30, },
286 };
287
288 #if 0
289 const uint8_t __attribute__((aligned(8))) dither_8x8_64[8][8]={
290 { 0, 48, 12, 60, 3, 51, 15, 63, },
291 { 32, 16, 44, 28, 35, 19, 47, 31, },
292 { 8, 56, 4, 52, 11, 59, 7, 55, },
293 { 40, 24, 36, 20, 43, 27, 39, 23, },
294 { 2, 50, 14, 62, 1, 49, 13, 61, },
295 { 34, 18, 46, 30, 33, 17, 45, 29, },
296 { 10, 58, 6, 54, 9, 57, 5, 53, },
297 { 42, 26, 38, 22, 41, 25, 37, 21, },
298 };
299 #endif
300
301 const uint8_t __attribute__((aligned(8))) dither_8x8_73[8][8]={
302 { 0, 55, 14, 68, 3, 58, 17, 72, },
303 { 37, 18, 50, 32, 40, 22, 54, 35, },
304 { 9, 64, 5, 59, 13, 67, 8, 63, },
305 { 46, 27, 41, 23, 49, 31, 44, 26, },
306 { 2, 57, 16, 71, 1, 56, 15, 70, },
307 { 39, 21, 52, 34, 38, 19, 51, 33, },
308 { 11, 66, 7, 62, 10, 65, 6, 60, },
309 { 48, 30, 43, 25, 47, 29, 42, 24, },
310 };
311
312 #if 0
313 const uint8_t __attribute__((aligned(8))) dither_8x8_128[8][8]={
314 { 68, 36, 92, 60, 66, 34, 90, 58, },
315 { 20, 116, 12, 108, 18, 114, 10, 106, },
316 { 84, 52, 76, 44, 82, 50, 74, 42, },
317 { 0, 96, 24, 120, 6, 102, 30, 126, },
318 { 64, 32, 88, 56, 70, 38, 94, 62, },
319 { 16, 112, 8, 104, 22, 118, 14, 110, },
320 { 80, 48, 72, 40, 86, 54, 78, 46, },
321 { 4, 100, 28, 124, 2, 98, 26, 122, },
322 };
323 #endif
324
325 #if 1
326 const uint8_t __attribute__((aligned(8))) dither_8x8_220[8][8]={
327 {117, 62, 158, 103, 113, 58, 155, 100, },
328 { 34, 199, 21, 186, 31, 196, 17, 182, },
329 {144, 89, 131, 76, 141, 86, 127, 72, },
330 { 0, 165, 41, 206, 10, 175, 52, 217, },
331 {110, 55, 151, 96, 120, 65, 162, 107, },
332 { 28, 193, 14, 179, 38, 203, 24, 189, },
333 {138, 83, 124, 69, 148, 93, 134, 79, },
334 { 7, 172, 48, 213, 3, 168, 45, 210, },
335 };
336 #elif 1
337 // tries to correct a gamma of 1.5
338 const uint8_t __attribute__((aligned(8))) dither_8x8_220[8][8]={
339 { 0, 143, 18, 200, 2, 156, 25, 215, },
340 { 78, 28, 125, 64, 89, 36, 138, 74, },
341 { 10, 180, 3, 161, 16, 195, 8, 175, },
342 {109, 51, 93, 38, 121, 60, 105, 47, },
343 { 1, 152, 23, 210, 0, 147, 20, 205, },
344 { 85, 33, 134, 71, 81, 30, 130, 67, },
345 { 14, 190, 6, 171, 12, 185, 5, 166, },
346 {117, 57, 101, 44, 113, 54, 97, 41, },
347 };
348 #elif 1
349 // tries to correct a gamma of 2.0
350 const uint8_t __attribute__((aligned(8))) dither_8x8_220[8][8]={
351 { 0, 124, 8, 193, 0, 140, 12, 213, },
352 { 55, 14, 104, 42, 66, 19, 119, 52, },
353 { 3, 168, 1, 145, 6, 187, 3, 162, },
354 { 86, 31, 70, 21, 99, 39, 82, 28, },
355 { 0, 134, 11, 206, 0, 129, 9, 200, },
356 { 62, 17, 114, 48, 58, 16, 109, 45, },
357 { 5, 181, 2, 157, 4, 175, 1, 151, },
358 { 95, 36, 78, 26, 90, 34, 74, 24, },
359 };
360 #else
361 // tries to correct a gamma of 2.5
362 const uint8_t __attribute__((aligned(8))) dither_8x8_220[8][8]={
363 { 0, 107, 3, 187, 0, 125, 6, 212, },
364 { 39, 7, 86, 28, 49, 11, 102, 36, },
365 { 1, 158, 0, 131, 3, 180, 1, 151, },
366 { 68, 19, 52, 12, 81, 25, 64, 17, },
367 { 0, 119, 5, 203, 0, 113, 4, 195, },
368 { 45, 9, 96, 33, 42, 8, 91, 30, },
369 { 2, 172, 1, 144, 2, 165, 0, 137, },
370 { 77, 23, 60, 15, 72, 21, 56, 14, },
371 };
372 #endif
373
374 const char *sws_format_name(enum PixelFormat format)
375 {
376 switch (format) {
377 case PIX_FMT_YUV420P:
378 return "yuv420p";
379 case PIX_FMT_YUVA420P:
380 return "yuva420p";
381 case PIX_FMT_YUYV422:
382 return "yuyv422";
383 case PIX_FMT_RGB24:
384 return "rgb24";
385 case PIX_FMT_BGR24:
386 return "bgr24";
387 case PIX_FMT_YUV422P:
388 return "yuv422p";
389 case PIX_FMT_YUV444P:
390 return "yuv444p";
391 case PIX_FMT_RGB32:
392 return "rgb32";
393 case PIX_FMT_YUV410P:
394 return "yuv410p";
395 case PIX_FMT_YUV411P:
396 return "yuv411p";
397 case PIX_FMT_RGB565:
398 return "rgb565";
399 case PIX_FMT_RGB555:
400 return "rgb555";
401 case PIX_FMT_GRAY16BE:
402 return "gray16be";
403 case PIX_FMT_GRAY16LE:
404 return "gray16le";
405 case PIX_FMT_GRAY8:
406 return "gray8";
407 case PIX_FMT_MONOWHITE:
408 return "mono white";
409 case PIX_FMT_MONOBLACK:
410 return "mono black";
411 case PIX_FMT_PAL8:
412 return "Palette";
413 case PIX_FMT_YUVJ420P:
414 return "yuvj420p";
415 case PIX_FMT_YUVJ422P:
416 return "yuvj422p";
417 case PIX_FMT_YUVJ444P:
418 return "yuvj444p";
419 case PIX_FMT_XVMC_MPEG2_MC:
420 return "xvmc_mpeg2_mc";
421 case PIX_FMT_XVMC_MPEG2_IDCT:
422 return "xvmc_mpeg2_idct";
423 case PIX_FMT_UYVY422:
424 return "uyvy422";
425 case PIX_FMT_UYYVYY411:
426 return "uyyvyy411";
427 case PIX_FMT_RGB32_1:
428 return "rgb32x";
429 case PIX_FMT_BGR32_1:
430 return "bgr32x";
431 case PIX_FMT_BGR32:
432 return "bgr32";
433 case PIX_FMT_BGR565:
434 return "bgr565";
435 case PIX_FMT_BGR555:
436 return "bgr555";
437 case PIX_FMT_BGR8:
438 return "bgr8";
439 case PIX_FMT_BGR4:
440 return "bgr4";
441 case PIX_FMT_BGR4_BYTE:
442 return "bgr4 byte";
443 case PIX_FMT_RGB8:
444 return "rgb8";
445 case PIX_FMT_RGB4:
446 return "rgb4";
447 case PIX_FMT_RGB4_BYTE:
448 return "rgb4 byte";
449 case PIX_FMT_NV12:
450 return "nv12";
451 case PIX_FMT_NV21:
452 return "nv21";
453 case PIX_FMT_YUV440P:
454 return "yuv440p";
455 case PIX_FMT_VDPAU_H264:
456 return "vdpau_h264";
457 case PIX_FMT_VDPAU_MPEG1:
458 return "vdpau_mpeg1";
459 case PIX_FMT_VDPAU_MPEG2:
460 return "vdpau_mpeg2";
461 case PIX_FMT_VDPAU_WMV3:
462 return "vdpau_wmv3";
463 case PIX_FMT_VDPAU_VC1:
464 return "vdpau_vc1";
465 default:
466 return "Unknown format";
467 }
468 }
469
470 static inline void yuv2yuvXinC(int16_t *lumFilter, int16_t **lumSrc, int lumFilterSize,
471 int16_t *chrFilter, int16_t **chrSrc, int chrFilterSize,
472 uint8_t *dest, uint8_t *uDest, uint8_t *vDest, int dstW, int chrDstW)
473 {
474 //FIXME Optimize (just quickly writen not opti..)
475 int i;
476 for (i=0; i<dstW; i++)
477 {
478 int val=1<<18;
479 int j;
480 for (j=0; j<lumFilterSize; j++)
481 val += lumSrc[j][i] * lumFilter[j];
482
483 dest[i]= av_clip_uint8(val>>19);
484 }
485
486 if (uDest)
487 for (i=0; i<chrDstW; i++)
488 {
489 int u=1<<18;
490 int v=1<<18;
491 int j;
492 for (j=0; j<chrFilterSize; j++)
493 {
494 u += chrSrc[j][i] * chrFilter[j];
495 v += chrSrc[j][i + VOFW] * chrFilter[j];
496 }
497
498 uDest[i]= av_clip_uint8(u>>19);
499 vDest[i]= av_clip_uint8(v>>19);
500 }
501 }
502
503 static inline void yuv2nv12XinC(int16_t *lumFilter, int16_t **lumSrc, int lumFilterSize,
504 int16_t *chrFilter, int16_t **chrSrc, int chrFilterSize,
505 uint8_t *dest, uint8_t *uDest, int dstW, int chrDstW, int dstFormat)
506 {
507 //FIXME Optimize (just quickly writen not opti..)
508 int i;
509 for (i=0; i<dstW; i++)
510 {
511 int val=1<<18;
512 int j;
513 for (j=0; j<lumFilterSize; j++)
514 val += lumSrc[j][i] * lumFilter[j];
515
516 dest[i]= av_clip_uint8(val>>19);
517 }
518
519 if (!uDest)
520 return;
521
522 if (dstFormat == PIX_FMT_NV12)
523 for (i=0; i<chrDstW; i++)
524 {
525 int u=1<<18;
526 int v=1<<18;
527 int j;
528 for (j=0; j<chrFilterSize; j++)
529 {
530 u += chrSrc[j][i] * chrFilter[j];
531 v += chrSrc[j][i + VOFW] * chrFilter[j];
532 }
533
534 uDest[2*i]= av_clip_uint8(u>>19);
535 uDest[2*i+1]= av_clip_uint8(v>>19);
536 }
537 else
538 for (i=0; i<chrDstW; i++)
539 {
540 int u=1<<18;
541 int v=1<<18;
542 int j;
543 for (j=0; j<chrFilterSize; j++)
544 {
545 u += chrSrc[j][i] * chrFilter[j];
546 v += chrSrc[j][i + VOFW] * chrFilter[j];
547 }
548
549 uDest[2*i]= av_clip_uint8(v>>19);
550 uDest[2*i+1]= av_clip_uint8(u>>19);
551 }
552 }
553
554 #define YSCALE_YUV_2_PACKEDX_NOCLIP_C(type) \
555 for (i=0; i<(dstW>>1); i++){\
556 int j;\
557 int Y1 = 1<<18;\
558 int Y2 = 1<<18;\
559 int U = 1<<18;\
560 int V = 1<<18;\
561 type av_unused *r, *b, *g;\
562 const int i2= 2*i;\
563 \
564 for (j=0; j<lumFilterSize; j++)\
565 {\
566 Y1 += lumSrc[j][i2] * lumFilter[j];\
567 Y2 += lumSrc[j][i2+1] * lumFilter[j];\
568 }\
569 for (j=0; j<chrFilterSize; j++)\
570 {\
571 U += chrSrc[j][i] * chrFilter[j];\
572 V += chrSrc[j][i+VOFW] * chrFilter[j];\
573 }\
574 Y1>>=19;\
575 Y2>>=19;\
576 U >>=19;\
577 V >>=19;\
578
579 #define YSCALE_YUV_2_PACKEDX_C(type) \
580 YSCALE_YUV_2_PACKEDX_NOCLIP_C(type)\
581 if ((Y1|Y2|U|V)&256)\
582 {\
583 if (Y1>255) Y1=255; \
584 else if (Y1<0)Y1=0; \
585 if (Y2>255) Y2=255; \
586 else if (Y2<0)Y2=0; \
587 if (U>255) U=255; \
588 else if (U<0) U=0; \
589 if (V>255) V=255; \
590 else if (V<0) V=0; \
591 }
592
593 #define YSCALE_YUV_2_PACKEDX_FULL_C \
594 for (i=0; i<dstW; i++){\
595 int j;\
596 int Y = 0;\
597 int U = -128<<19;\
598 int V = -128<<19;\
599 int R,G,B;\
600 \
601 for (j=0; j<lumFilterSize; j++){\
602 Y += lumSrc[j][i ] * lumFilter[j];\
603 }\
604 for (j=0; j<chrFilterSize; j++){\
605 U += chrSrc[j][i ] * chrFilter[j];\
606 V += chrSrc[j][i+VOFW] * chrFilter[j];\
607 }\
608 Y >>=10;\
609 U >>=10;\
610 V >>=10;\
611
612 #define YSCALE_YUV_2_RGBX_FULL_C(rnd) \
613 YSCALE_YUV_2_PACKEDX_FULL_C\
614 Y-= c->yuv2rgb_y_offset;\
615 Y*= c->yuv2rgb_y_coeff;\
616 Y+= rnd;\
617 R= Y + V*c->yuv2rgb_v2r_coeff;\
618 G= Y + V*c->yuv2rgb_v2g_coeff + U*c->yuv2rgb_u2g_coeff;\
619 B= Y + U*c->yuv2rgb_u2b_coeff;\
620 if ((R|G|B)&(0xC0000000)){\
621 if (R>=(256<<22)) R=(256<<22)-1; \
622 else if (R<0)R=0; \
623 if (G>=(256<<22)) G=(256<<22)-1; \
624 else if (G<0)G=0; \
625 if (B>=(256<<22)) B=(256<<22)-1; \
626 else if (B<0)B=0; \
627 }\
628
629
630 #define YSCALE_YUV_2_GRAY16_C \
631 for (i=0; i<(dstW>>1); i++){\
632 int j;\
633 int Y1 = 1<<18;\
634 int Y2 = 1<<18;\
635 int U = 1<<18;\
636 int V = 1<<18;\
637 \
638 const int i2= 2*i;\
639 \
640 for (j=0; j<lumFilterSize; j++)\
641 {\
642 Y1 += lumSrc[j][i2] * lumFilter[j];\
643 Y2 += lumSrc[j][i2+1] * lumFilter[j];\
644 }\
645 Y1>>=11;\
646 Y2>>=11;\
647 if ((Y1|Y2|U|V)&65536)\
648 {\
649 if (Y1>65535) Y1=65535; \
650 else if (Y1<0)Y1=0; \
651 if (Y2>65535) Y2=65535; \
652 else if (Y2<0)Y2=0; \
653 }
654
655 #define YSCALE_YUV_2_RGBX_C(type) \
656 YSCALE_YUV_2_PACKEDX_C(type) /* FIXME fix tables so that cliping is not needed and then use _NOCLIP*/\
657 r = (type *)c->table_rV[V]; \
658 g = (type *)(c->table_gU[U] + c->table_gV[V]); \
659 b = (type *)c->table_bU[U]; \
660
661 #define YSCALE_YUV_2_PACKED2_C \
662 for (i=0; i<(dstW>>1); i++){ \
663 const int i2= 2*i; \
664 int Y1= (buf0[i2 ]*yalpha1+buf1[i2 ]*yalpha)>>19; \
665 int Y2= (buf0[i2+1]*yalpha1+buf1[i2+1]*yalpha)>>19; \
666 int U= (uvbuf0[i ]*uvalpha1+uvbuf1[i ]*uvalpha)>>19; \
667 int V= (uvbuf0[i+VOFW]*uvalpha1+uvbuf1[i+VOFW]*uvalpha)>>19; \
668
669 #define YSCALE_YUV_2_GRAY16_2_C \
670 for (i=0; i<(dstW>>1); i++){ \
671 const int i2= 2*i; \
672 int Y1= (buf0[i2 ]*yalpha1+buf1[i2 ]*yalpha)>>11; \
673 int Y2= (buf0[i2+1]*yalpha1+buf1[i2+1]*yalpha)>>11; \
674
675 #define YSCALE_YUV_2_RGB2_C(type) \
676 YSCALE_YUV_2_PACKED2_C\
677 type *r, *b, *g;\
678 r = (type *)c->table_rV[V];\
679 g = (type *)(c->table_gU[U] + c->table_gV[V]);\
680 b = (type *)c->table_bU[U];\
681
682 #define YSCALE_YUV_2_PACKED1_C \
683 for (i=0; i<(dstW>>1); i++){\
684 const int i2= 2*i;\
685 int Y1= buf0[i2 ]>>7;\
686 int Y2= buf0[i2+1]>>7;\
687 int U= (uvbuf1[i ])>>7;\
688 int V= (uvbuf1[i+VOFW])>>7;\
689
690 #define YSCALE_YUV_2_GRAY16_1_C \
691 for (i=0; i<(dstW>>1); i++){\
692 const int i2= 2*i;\
693 int Y1= buf0[i2 ]<<1;\
694 int Y2= buf0[i2+1]<<1;\
695
696 #define YSCALE_YUV_2_RGB1_C(type) \
697 YSCALE_YUV_2_PACKED1_C\
698 type *r, *b, *g;\
699 r = (type *)c->table_rV[V];\
700 g = (type *)(c->table_gU[U] + c->table_gV[V]);\
701 b = (type *)c->table_bU[U];\
702
703 #define YSCALE_YUV_2_PACKED1B_C \
704 for (i=0; i<(dstW>>1); i++){\
705 const int i2= 2*i;\
706 int Y1= buf0[i2 ]>>7;\
707 int Y2= buf0[i2+1]>>7;\
708 int U= (uvbuf0[i ] + uvbuf1[i ])>>8;\
709 int V= (uvbuf0[i+VOFW] + uvbuf1[i+VOFW])>>8;\
710
711 #define YSCALE_YUV_2_RGB1B_C(type) \
712 YSCALE_YUV_2_PACKED1B_C\
713 type *r, *b, *g;\
714 r = (type *)c->table_rV[V];\
715 g = (type *)(c->table_gU[U] + c->table_gV[V]);\
716 b = (type *)c->table_bU[U];\
717
718 #define YSCALE_YUV_2_MONO2_C \
719 const uint8_t * const d128=dither_8x8_220[y&7];\
720 uint8_t *g= c->table_gU[128] + c->table_gV[128];\
721 for (i=0; i<dstW-7; i+=8){\
722 int acc;\
723 acc = g[((buf0[i ]*yalpha1+buf1[i ]*yalpha)>>19) + d128[0]];\
724 acc+= acc + g[((buf0[i+1]*yalpha1+buf1[i+1]*yalpha)>>19) + d128[1]];\
725 acc+= acc + g[((buf0[i+2]*yalpha1+buf1[i+2]*yalpha)>>19) + d128[2]];\
726 acc+= acc + g[((buf0[i+3]*yalpha1+buf1[i+3]*yalpha)>>19) + d128[3]];\
727 acc+= acc + g[((buf0[i+4]*yalpha1+buf1[i+4]*yalpha)>>19) + d128[4]];\
728 acc+= acc + g[((buf0[i+5]*yalpha1+buf1[i+5]*yalpha)>>19) + d128[5]];\
729 acc+= acc + g[((buf0[i+6]*yalpha1+buf1[i+6]*yalpha)>>19) + d128[6]];\
730 acc+= acc + g[((buf0[i+7]*yalpha1+buf1[i+7]*yalpha)>>19) + d128[7]];\
731 ((uint8_t*)dest)[0]= c->dstFormat == PIX_FMT_MONOBLACK ? acc : ~acc;\
732 dest++;\
733 }\
734
735
736 #define YSCALE_YUV_2_MONOX_C \
737 const uint8_t * const d128=dither_8x8_220[y&7];\
738 uint8_t *g= c->table_gU[128] + c->table_gV[128];\
739 int acc=0;\
740 for (i=0; i<dstW-1; i+=2){\
741 int j;\
742 int Y1=1<<18;\
743 int Y2=1<<18;\
744 \
745 for (j=0; j<lumFilterSize; j++)\
746 {\
747 Y1 += lumSrc[j][i] * lumFilter[j];\
748 Y2 += lumSrc[j][i+1] * lumFilter[j];\
749 }\
750 Y1>>=19;\
751 Y2>>=19;\
752 if ((Y1|Y2)&256)\
753 {\
754 if (Y1>255) Y1=255;\
755 else if (Y1<0)Y1=0;\
756 if (Y2>255) Y2=255;\
757 else if (Y2<0)Y2=0;\
758 }\
759 acc+= acc + g[Y1+d128[(i+0)&7]];\
760 acc+= acc + g[Y2+d128[(i+1)&7]];\
761 if ((i&7)==6){\
762 ((uint8_t*)dest)[0]= c->dstFormat == PIX_FMT_MONOBLACK ? acc : ~acc;\
763 dest++;\
764 }\
765 }
766
767
768 #define YSCALE_YUV_2_ANYRGB_C(func, func2, func_g16, func_monoblack)\
769 switch(c->dstFormat)\
770 {\
771 case PIX_FMT_RGB32:\
772 case PIX_FMT_BGR32:\
773 case PIX_FMT_RGB32_1:\
774 case PIX_FMT_BGR32_1:\
775 func(uint32_t)\
776 ((uint32_t*)dest)[i2+0]= r[Y1] + g[Y1] + b[Y1];\
777 ((uint32_t*)dest)[i2+1]= r[Y2] + g[Y2] + b[Y2];\
778 } \
779 break;\
780 case PIX_FMT_RGB24:\
781 func(uint8_t)\
782 ((uint8_t*)dest)[0]= r[Y1];\
783 ((uint8_t*)dest)[1]= g[Y1];\
784 ((uint8_t*)dest)[2]= b[Y1];\
785 ((uint8_t*)dest)[3]= r[Y2];\
786 ((uint8_t*)dest)[4]= g[Y2];\
787 ((uint8_t*)dest)[5]= b[Y2];\
788 dest+=6;\
789 }\
790 break;\
791 case PIX_FMT_BGR24:\
792 func(uint8_t)\
793 ((uint8_t*)dest)[0]= b[Y1];\
794 ((uint8_t*)dest)[1]= g[Y1];\
795 ((uint8_t*)dest)[2]= r[Y1];\
796 ((uint8_t*)dest)[3]= b[Y2];\
797 ((uint8_t*)dest)[4]= g[Y2];\
798 ((uint8_t*)dest)[5]= r[Y2];\
799 dest+=6;\
800 }\
801 break;\
802 case PIX_FMT_RGB565:\
803 case PIX_FMT_BGR565:\
804 {\
805 const int dr1= dither_2x2_8[y&1 ][0];\
806 const int dg1= dither_2x2_4[y&1 ][0];\
807 const int db1= dither_2x2_8[(y&1)^1][0];\
808 const int dr2= dither_2x2_8[y&1 ][1];\
809 const int dg2= dither_2x2_4[y&1 ][1];\
810 const int db2= dither_2x2_8[(y&1)^1][1];\
811 func(uint16_t)\
812 ((uint16_t*)dest)[i2+0]= r[Y1+dr1] + g[Y1+dg1] + b[Y1+db1];\
813 ((uint16_t*)dest)[i2+1]= r[Y2+dr2] + g[Y2+dg2] + b[Y2+db2];\
814 }\
815 }\
816 break;\
817 case PIX_FMT_RGB555:\
818 case PIX_FMT_BGR555:\
819 {\
820 const int dr1= dither_2x2_8[y&1 ][0];\
821 const int dg1= dither_2x2_8[y&1 ][1];\
822 const int db1= dither_2x2_8[(y&1)^1][0];\
823 const int dr2= dither_2x2_8[y&1 ][1];\
824 const int dg2= dither_2x2_8[y&1 ][0];\
825 const int db2= dither_2x2_8[(y&1)^1][1];\
826 func(uint16_t)\
827 ((uint16_t*)dest)[i2+0]= r[Y1+dr1] + g[Y1+dg1] + b[Y1+db1];\
828 ((uint16_t*)dest)[i2+1]= r[Y2+dr2] + g[Y2+dg2] + b[Y2+db2];\
829 }\
830 }\
831 break;\
832 case PIX_FMT_RGB8:\
833 case PIX_FMT_BGR8:\
834 {\
835 const uint8_t * const d64= dither_8x8_73[y&7];\
836 const uint8_t * const d32= dither_8x8_32[y&7];\
837 func(uint8_t)\
838 ((uint8_t*)dest)[i2+0]= r[Y1+d32[(i2+0)&7]] + g[Y1+d32[(i2+0)&7]] + b[Y1+d64[(i2+0)&7]];\
839 ((uint8_t*)dest)[i2+1]= r[Y2+d32[(i2+1)&7]] + g[Y2+d32[(i2+1)&7]] + b[Y2+d64[(i2+1)&7]];\
840 }\
841 }\
842 break;\
843 case PIX_FMT_RGB4:\
844 case PIX_FMT_BGR4:\
845 {\
846 const uint8_t * const d64= dither_8x8_73 [y&7];\
847 const uint8_t * const d128=dither_8x8_220[y&7];\
848 func(uint8_t)\
849 ((uint8_t*)dest)[i]= r[Y1+d128[(i2+0)&7]] + g[Y1+d64[(i2+0)&7]] + b[Y1+d128[(i2+0)&7]]\
850 + ((r[Y2+d128[(i2+1)&7]] + g[Y2+d64[(i2+1)&7]] + b[Y2+d128[(i2+1)&7]])<<4);\
851 }\
852 }\
853 break;\
854 case PIX_FMT_RGB4_BYTE:\
855 case PIX_FMT_BGR4_BYTE:\
856 {\
857 const uint8_t * const d64= dither_8x8_73 [y&7];\
858 const uint8_t * const d128=dither_8x8_220[y&7];\
859 func(uint8_t)\
860 ((uint8_t*)dest)[i2+0]= r[Y1+d128[(i2+0)&7]] + g[Y1+d64[(i2+0)&7]] + b[Y1+d128[(i2+0)&7]];\
861 ((uint8_t*)dest)[i2+1]= r[Y2+d128[(i2+1)&7]] + g[Y2+d64[(i2+1)&7]] + b[Y2+d128[(i2+1)&7]];\
862 }\
863 }\
864 break;\
865 case PIX_FMT_MONOBLACK:\
866 case PIX_FMT_MONOWHITE:\
867 {\
868 func_monoblack\
869 }\
870 break;\
871 case PIX_FMT_YUYV422:\
872 func2\
873 ((uint8_t*)dest)[2*i2+0]= Y1;\
874 ((uint8_t*)dest)[2*i2+1]= U;\
875 ((uint8_t*)dest)[2*i2+2]= Y2;\
876 ((uint8_t*)dest)[2*i2+3]= V;\
877 } \
878 break;\
879 case PIX_FMT_UYVY422:\
880 func2\
881 ((uint8_t*)dest)[2*i2+0]= U;\
882 ((uint8_t*)dest)[2*i2+1]= Y1;\
883 ((uint8_t*)dest)[2*i2+2]= V;\
884 ((uint8_t*)dest)[2*i2+3]= Y2;\
885 } \
886 break;\
887 case PIX_FMT_GRAY16BE:\
888 func_g16\
889 ((uint8_t*)dest)[2*i2+0]= Y1>>8;\
890 ((uint8_t*)dest)[2*i2+1]= Y1;\
891 ((uint8_t*)dest)[2*i2+2]= Y2>>8;\
892 ((uint8_t*)dest)[2*i2+3]= Y2;\
893 } \
894 break;\
895 case PIX_FMT_GRAY16LE:\
896 func_g16\
897 ((uint8_t*)dest)[2*i2+0]= Y1;\
898 ((uint8_t*)dest)[2*i2+1]= Y1>>8;\
899 ((uint8_t*)dest)[2*i2+2]= Y2;\
900 ((uint8_t*)dest)[2*i2+3]= Y2>>8;\
901 } \
902 break;\
903 }\
904
905
906 static inline void yuv2packedXinC(SwsContext *c, int16_t *lumFilter, int16_t **lumSrc, int lumFilterSize,
907 int16_t *chrFilter, int16_t **chrSrc, int chrFilterSize,
908 uint8_t *dest, int dstW, int y)
909 {
910 int i;
911 YSCALE_YUV_2_ANYRGB_C(YSCALE_YUV_2_RGBX_C, YSCALE_YUV_2_PACKEDX_C(void), YSCALE_YUV_2_GRAY16_C, YSCALE_YUV_2_MONOX_C)
912 }
913
914 static inline void yuv2rgbXinC_full(SwsContext *c, int16_t *lumFilter, int16_t **lumSrc, int lumFilterSize,
915 int16_t *chrFilter, int16_t **chrSrc, int chrFilterSize,
916 uint8_t *dest, int dstW, int y)
917 {
918 int i;
919 int step= fmt_depth(c->dstFormat)/8;
920 int aidx= 3;
921
922 switch(c->dstFormat){
923 case PIX_FMT_ARGB:
924 dest++;
925 aidx= 0;
926 case PIX_FMT_RGB24:
927 aidx--;
928 case PIX_FMT_RGBA:
929 YSCALE_YUV_2_RGBX_FULL_C(1<<21)
930 dest[aidx]= 0;
931 dest[0]= R>>22;
932 dest[1]= G>>22;
933 dest[2]= B>>22;
934 dest+= step;
935 }
936 break;
937 case PIX_FMT_ABGR:
938 dest++;
939 aidx= 0;
940 case PIX_FMT_BGR24:
941 aidx--;
942 case PIX_FMT_BGRA:
943 YSCALE_YUV_2_RGBX_FULL_C(1<<21)
944 dest[aidx]= 0;
945 dest[0]= B>>22;
946 dest[1]= G>>22;
947 dest[2]= R>>22;
948 dest+= step;
949 }
950 break;
951 default:
952 assert(0);
953 }
954 }
955
956 //Note: we have C, X86, MMX, MMX2, 3DNOW version therse no 3DNOW+MMX2 one
957 //Plain C versions
958 #if !HAVE_MMX || defined (RUNTIME_CPUDETECT) || !CONFIG_GPL
959 #define COMPILE_C
960 #endif
961
962 #if ARCH_PPC
963 #if (HAVE_ALTIVEC || defined (RUNTIME_CPUDETECT)) && CONFIG_GPL
964 #define COMPILE_ALTIVEC
965 #endif
966 #endif //ARCH_PPC
967
968 #if ARCH_X86
969
970 #if ((HAVE_MMX && !HAVE_3DNOW && !HAVE_MMX2) || defined (RUNTIME_CPUDETECT)) && CONFIG_GPL
971 #define COMPILE_MMX
972 #endif
973
974 #if (HAVE_MMX2 || defined (RUNTIME_CPUDETECT)) && CONFIG_GPL
975 #define COMPILE_MMX2
976 #endif
977
978 #if ((HAVE_3DNOW && !HAVE_MMX2) || defined (RUNTIME_CPUDETECT)) && CONFIG_GPL
979 #define COMPILE_3DNOW
980 #endif
981 #endif //ARCH_X86
982
983 #undef HAVE_MMX
984 #undef HAVE_MMX2
985 #undef HAVE_3DNOW
986 #undef HAVE_ALTIVEC
987 #define HAVE_MMX 0
988 #define HAVE_MMX2 0
989 #define HAVE_3DNOW 0
990 #define HAVE_ALTIVEC 0
991
992 #ifdef COMPILE_C
993 #define RENAME(a) a ## _C
994 #include "swscale_template.c"
995 #endif
996
997 #ifdef COMPILE_ALTIVEC
998 #undef RENAME
999 #undef HAVE_ALTIVEC
1000 #define HAVE_ALTIVEC 1
1001 #define RENAME(a) a ## _altivec
1002 #include "swscale_template.c"
1003 #endif
1004
1005 #if ARCH_X86
1006
1007 //X86 versions
1008 /*
1009 #undef RENAME
1010 #undef HAVE_MMX
1011 #undef HAVE_MMX2
1012 #undef HAVE_3DNOW
1013 #define ARCH_X86
1014 #define RENAME(a) a ## _X86
1015 #include "swscale_template.c"
1016 */
1017 //MMX versions
1018 #ifdef COMPILE_MMX
1019 #undef RENAME
1020 #undef HAVE_MMX
1021 #undef HAVE_MMX2
1022 #undef HAVE_3DNOW
1023 #define HAVE_MMX 1
1024 #define HAVE_MMX2 0
1025 #define HAVE_3DNOW 0
1026 #define RENAME(a) a ## _MMX
1027 #include "swscale_template.c"
1028 #endif
1029
1030 //MMX2 versions
1031 #ifdef COMPILE_MMX2
1032 #undef RENAME
1033 #undef HAVE_MMX
1034 #undef HAVE_MMX2
1035 #undef HAVE_3DNOW
1036 #define HAVE_MMX 1
1037 #define HAVE_MMX2 1
1038 #define HAVE_3DNOW 0
1039 #define RENAME(a) a ## _MMX2
1040 #include "swscale_template.c"
1041 #endif
1042
1043 //3DNOW versions
1044 #ifdef COMPILE_3DNOW
1045 #undef RENAME
1046 #undef HAVE_MMX
1047 #undef HAVE_MMX2
1048 #undef HAVE_3DNOW
1049 #define HAVE_MMX 1
1050 #define HAVE_MMX2 0
1051 #define HAVE_3DNOW 1
1052 #define RENAME(a) a ## _3DNow
1053 #include "swscale_template.c"
1054 #endif
1055
1056 #endif //ARCH_X86
1057
1058 // minor note: the HAVE_xyz is messed up after that line so don't use it
1059
1060 static double getSplineCoeff(double a, double b, double c, double d, double dist)
1061 {
1062 // printf("%f %f %f %f %f\n", a,b,c,d,dist);
1063 if (dist<=1.0) return ((d*dist + c)*dist + b)*dist +a;
1064 else return getSplineCoeff( 0.0,
1065 b+ 2.0*c + 3.0*d,
1066 c + 3.0*d,
1067 -b- 3.0*c - 6.0*d,
1068 dist-1.0);
1069 }
1070
1071 static inline int initFilter(int16_t **outFilter, int16_t **filterPos, int *outFilterSize, int xInc,
1072 int srcW, int dstW, int filterAlign, int one, int flags,
1073 SwsVector *srcFilter, SwsVector *dstFilter, double param[2])
1074 {
1075 int i;
1076 int filterSize;
1077 int filter2Size;
1078 int minFilterSize;
1079 int64_t *filter=NULL;
1080 int64_t *filter2=NULL;
1081 const int64_t fone= 1LL<<54;
1082 int ret= -1;
1083 #if ARCH_X86
1084 if (flags & SWS_CPU_CAPS_MMX)
1085 __asm__ volatile("emms\n\t"::: "memory"); //FIXME this should not be required but it IS (even for non-MMX versions)
1086 #endif
1087
1088 // Note the +1 is for the MMXscaler which reads over the end
1089 *filterPos = av_malloc((dstW+1)*sizeof(int16_t));
1090
1091 if (FFABS(xInc - 0x10000) <10) // unscaled
1092 {
1093 int i;
1094 filterSize= 1;
1095 filter= av_mallocz(dstW*sizeof(*filter)*filterSize);
1096
1097 for (i=0; i<dstW; i++)
1098 {
1099 filter[i*filterSize]= fone;
1100 (*filterPos)[i]=i;
1101 }
1102
1103 }
1104 else if (flags&SWS_POINT) // lame looking point sampling mode
1105 {
1106 int i;
1107 int xDstInSrc;
1108 filterSize= 1;
1109 filter= av_malloc(dstW*sizeof(*filter)*filterSize);
1110
1111 xDstInSrc= xInc/2 - 0x8000;
1112 for (i=0; i<dstW; i++)
1113 {
1114 int xx= (xDstInSrc - ((filterSize-1)<<15) + (1<<15))>>16;
1115
1116 (*filterPos)[i]= xx;
1117 filter[i]= fone;
1118 xDstInSrc+= xInc;
1119 }
1120 }
1121 else if ((xInc <= (1<<16) && (flags&SWS_AREA)) || (flags&SWS_FAST_BILINEAR)) // bilinear upscale
1122 {
1123 int i;
1124 int xDstInSrc;
1125 if (flags&SWS_BICUBIC) filterSize= 4;
1126 else if (flags&SWS_X ) filterSize= 4;
1127 else filterSize= 2; // SWS_BILINEAR / SWS_AREA
1128 filter= av_malloc(dstW*sizeof(*filter)*filterSize);
1129
1130 xDstInSrc= xInc/2 - 0x8000;
1131 for (i=0; i<dstW; i++)
1132 {
1133 int xx= (xDstInSrc - ((filterSize-1)<<15) + (1<<15))>>16;
1134 int j;
1135
1136 (*filterPos)[i]= xx;
1137 //Bilinear upscale / linear interpolate / Area averaging
1138 for (j=0; j<filterSize; j++)
1139 {
1140 int64_t coeff= fone - FFABS((xx<<16) - xDstInSrc)*(fone>>16);
1141 if (coeff<0) coeff=0;
1142 filter[i*filterSize + j]= coeff;
1143 xx++;
1144 }
1145 xDstInSrc+= xInc;
1146 }
1147 }
1148 else
1149 {
1150 int xDstInSrc;
1151 int sizeFactor;
1152
1153 if (flags&SWS_BICUBIC) sizeFactor= 4;
1154 else if (flags&SWS_X) sizeFactor= 8;
1155 else if (flags&SWS_AREA) sizeFactor= 1; //downscale only, for upscale it is bilinear
1156 else if (flags&SWS_GAUSS) sizeFactor= 8; // infinite ;)
1157 else if (flags&SWS_LANCZOS) sizeFactor= param[0] != SWS_PARAM_DEFAULT ? ceil(2*param[0]) : 6;
1158 else if (flags&SWS_SINC) sizeFactor= 20; // infinite ;)
1159 else if (flags&SWS_SPLINE) sizeFactor= 20; // infinite ;)
1160 else if (flags&SWS_BILINEAR) sizeFactor= 2;
1161 else {
1162 sizeFactor= 0; //GCC warning killer
1163 assert(0);
1164 }
1165
1166 if (xInc <= 1<<16) filterSize= 1 + sizeFactor; // upscale
1167 else filterSize= 1 + (sizeFactor*srcW + dstW - 1)/ dstW;
1168
1169 if (filterSize > srcW-2) filterSize=srcW-2;
1170
1171 filter= av_malloc(dstW*sizeof(*filter)*filterSize);
1172
1173 xDstInSrc= xInc - 0x10000;
1174 for (i=0; i<dstW; i++)
1175 {
1176 int xx= (xDstInSrc - ((filterSize-2)<<16)) / (1<<17);
1177 int j;
1178 (*filterPos)[i]= xx;
1179 for (j=0; j<filterSize; j++)
1180 {
1181 int64_t d= ((int64_t)FFABS((xx<<17) - xDstInSrc))<<13;
1182 double floatd;
1183 int64_t coeff;
1184
1185 if (xInc > 1<<16)
1186 d= d*dstW/srcW;
1187 floatd= d * (1.0/(1<<30));
1188
1189 if (flags & SWS_BICUBIC)
1190 {
1191 int64_t B= (param[0] != SWS_PARAM_DEFAULT ? param[0] : 0) * (1<<24);
1192 int64_t C= (param[1] != SWS_PARAM_DEFAULT ? param[1] : 0.6) * (1<<24);
1193 int64_t dd = ( d*d)>>30;
1194 int64_t ddd= (dd*d)>>30;
1195
1196 if (d < 1LL<<30)
1197 coeff = (12*(1<<24)-9*B-6*C)*ddd + (-18*(1<<24)+12*B+6*C)*dd + (6*(1<<24)-2*B)*(1<<30);
1198 else if (d < 1LL<<31)
1199 coeff = (-B-6*C)*ddd + (6*B+30*C)*dd + (-12*B-48*C)*d + (8*B+24*C)*(1<<30);
1200 else
1201 coeff=0.0;
1202 coeff *= fone>>(30+24);
1203 }
1204 /* else if (flags & SWS_X)
1205 {
1206 double p= param ? param*0.01 : 0.3;
1207 coeff = d ? sin(d*PI)/(d*PI) : 1.0;
1208 coeff*= pow(2.0, - p*d*d);
1209 }*/
1210 else if (flags & SWS_X)
1211 {
1212 double A= param[0] != SWS_PARAM_DEFAULT ? param[0] : 1.0;
1213 double c;
1214
1215 if (floatd<1.0)
1216 c = cos(floatd*PI);
1217 else
1218 c=-1.0;
1219 if (c<0.0) c= -pow(-c, A);
1220 else c= pow( c, A);
1221 coeff= (c*0.5 + 0.5)*fone;
1222 }
1223 else if (flags & SWS_AREA)
1224 {
1225 int64_t d2= d - (1<<29);
1226 if (d2*xInc < -(1LL<<(29+16))) coeff= 1.0 * (1LL<<(30+16));
1227 else if (d2*xInc < (1LL<<(29+16))) coeff= -d2*xInc + (1LL<<(29+16));
1228 else coeff=0.0;
1229 coeff *= fone>>(30+16);
1230 }
1231 else if (flags & SWS_GAUSS)
1232 {
1233 double p= param[0] != SWS_PARAM_DEFAULT ? param[0] : 3.0;
1234 coeff = (pow(2.0, - p*floatd*floatd))*fone;
1235 }
1236 else if (flags & SWS_SINC)
1237 {
1238 coeff = (d ? sin(floatd*PI)/(floatd*PI) : 1.0)*fone;
1239 }
1240 else if (flags & SWS_LANCZOS)
1241 {
1242 double p= param[0] != SWS_PARAM_DEFAULT ? param[0] : 3.0;
1243 coeff = (d ? sin(floatd*PI)*sin(floatd*PI/p)/(floatd*floatd*PI*PI/p) : 1.0)*fone;
1244 if (floatd>p) coeff=0;
1245 }
1246 else if (flags & SWS_BILINEAR)
1247 {
1248 coeff= (1<<30) - d;
1249 if (coeff<0) coeff=0;
1250 coeff *= fone >> 30;
1251 }
1252 else if (flags & SWS_SPLINE)
1253 {
1254 double p=-2.196152422706632;
1255 coeff = getSplineCoeff(1.0, 0.0, p, -p-1.0, floatd) * fone;
1256 }
1257 else {
1258 coeff= 0.0; //GCC warning killer
1259 assert(0);
1260 }
1261
1262 filter[i*filterSize + j]= coeff;
1263 xx++;
1264 }
1265 xDstInSrc+= 2*xInc;
1266 }
1267 }
1268
1269 /* apply src & dst Filter to filter -> filter2
1270 av_free(filter);
1271 */
1272 assert(filterSize>0);
1273 filter2Size= filterSize;
1274 if (srcFilter) filter2Size+= srcFilter->length - 1;
1275 if (dstFilter) filter2Size+= dstFilter->length - 1;
1276 assert(filter2Size>0);
1277 filter2= av_mallocz(filter2Size*dstW*sizeof(*filter2));
1278
1279 for (i=0; i<dstW; i++)
1280 {
1281 int j, k;
1282
1283 if(srcFilter){
1284 for (k=0; k<srcFilter->length; k++){
1285 for (j=0; j<filterSize; j++)
1286 filter2[i*filter2Size + k + j] += srcFilter->coeff[k]*filter[i*filterSize + j];
1287 }
1288 }else{
1289 for (j=0; j<filterSize; j++)
1290 filter2[i*filter2Size + j]= filter[i*filterSize + j];
1291 }
1292 //FIXME dstFilter
1293
1294 (*filterPos)[i]+= (filterSize-1)/2 - (filter2Size-1)/2;
1295 }
1296 av_freep(&filter);
1297
1298 /* try to reduce the filter-size (step1 find size and shift left) */
1299 // Assume it is near normalized (*0.5 or *2.0 is OK but * 0.001 is not).
1300 minFilterSize= 0;
1301 for (i=dstW-1; i>=0; i--)
1302 {
1303 int min= filter2Size;
1304 int j;
1305 int64_t cutOff=0.0;
1306
1307 /* get rid off near zero elements on the left by shifting left */
1308 for (j=0; j<filter2Size; j++)
1309 {
1310 int k;
1311 cutOff += FFABS(filter2[i*filter2Size]);
1312
1313 if (cutOff > SWS_MAX_REDUCE_CUTOFF*fone) break;
1314
1315 /* preserve monotonicity because the core can't handle the filter otherwise */
1316 if (i<dstW-1 && (*filterPos)[i] >= (*filterPos)[i+1]) break;
1317
1318 // Move filter coeffs left
1319 for (k=1; k<filter2Size; k++)
1320 filter2[i*filter2Size + k - 1]= filter2[i*filter2Size + k];
1321 filter2[i*filter2Size + k - 1]= 0;
1322 (*filterPos)[i]++;
1323 }
1324
1325 cutOff=0;
1326 /* count near zeros on the right */
1327 for (j=filter2Size-1; j>0; j--)
1328 {
1329 cutOff += FFABS(filter2[i*filter2Size + j]);
1330
1331 if (cutOff > SWS_MAX_REDUCE_CUTOFF*fone) break;
1332 min--;
1333 }
1334
1335 if (min>minFilterSize) minFilterSize= min;
1336 }
1337
1338 if (flags & SWS_CPU_CAPS_ALTIVEC) {
1339 // we can handle the special case 4,
1340 // so we don't want to go to the full 8
1341 if (minFilterSize < 5)
1342 filterAlign = 4;
1343
1344 // we really don't want to waste our time
1345 // doing useless computation, so fall-back on
1346 // the scalar C code for very small filter.
1347 // vectorizing is worth it only if you have
1348 // decent-sized vector.
1349 if (minFilterSize < 3)
1350 filterAlign = 1;
1351 }
1352
1353 if (flags & SWS_CPU_CAPS_MMX) {
1354 // special case for unscaled vertical filtering
1355 if (minFilterSize == 1 && filterAlign == 2)
1356 filterAlign= 1;
1357 }
1358
1359 assert(minFilterSize > 0);
1360 filterSize= (minFilterSize +(filterAlign-1)) & (~(filterAlign-1));
1361 assert(filterSize > 0);
1362 filter= av_malloc(filterSize*dstW*sizeof(*filter));
1363 if (filterSize >= MAX_FILTER_SIZE*16/((flags&SWS_ACCURATE_RND) ? APCK_SIZE : 16) || !filter)
1364 goto error;
1365 *outFilterSize= filterSize;
1366
1367 if (flags&SWS_PRINT_INFO)
1368 av_log(NULL, AV_LOG_VERBOSE, "SwScaler: reducing / aligning filtersize %d -> %d\n", filter2Size, filterSize);
1369 /* try to reduce the filter-size (step2 reduce it) */
1370 for (i=0; i<dstW; i++)
1371 {
1372 int j;
1373
1374 for (j=0; j<filterSize; j++)
1375 {
1376 if (j>=filter2Size) filter[i*filterSize + j]= 0;
1377 else filter[i*filterSize + j]= filter2[i*filter2Size + j];
1378 if((flags & SWS_BITEXACT) && j>=minFilterSize)
1379 filter[i*filterSize + j]= 0;
1380 }
1381 }
1382
1383
1384 //FIXME try to align filterpos if possible
1385
1386 //fix borders
1387 for (i=0; i<dstW; i++)
1388 {
1389 int j;
1390 if ((*filterPos)[i] < 0)
1391 {
1392 // Move filter coeffs left to compensate for filterPos
1393 for (j=1; j<filterSize; j++)
1394 {
1395 int left= FFMAX(j + (*filterPos)[i], 0);
1396 filter[i*filterSize + left] += filter[i*filterSize + j];
1397 filter[i*filterSize + j]=0;
1398 }
1399 (*filterPos)[i]= 0;
1400 }
1401
1402 if ((*filterPos)[i] + filterSize > srcW)
1403 {
1404 int shift= (*filterPos)[i] + filterSize - srcW;
1405 // Move filter coeffs right to compensate for filterPos
1406 for (j=filterSize-2; j>=0; j--)
1407 {
1408 int right= FFMIN(j + shift, filterSize-1);
1409 filter[i*filterSize +right] += filter[i*filterSize +j];
1410 filter[i*filterSize +j]=0;
1411 }
1412 (*filterPos)[i]= srcW - filterSize;
1413 }
1414 }
1415
1416 // Note the +1 is for the MMXscaler which reads over the end
1417 /* align at 16 for AltiVec (needed by hScale_altivec_real) */
1418 *outFilter= av_mallocz(*outFilterSize*(dstW+1)*sizeof(int16_t));
1419
1420 /* Normalize & Store in outFilter */
1421 for (i=0; i<dstW; i++)
1422 {
1423 int j;
1424 int64_t error=0;
1425 int64_t sum=0;
1426
1427 for (j=0; j<filterSize; j++)
1428 {
1429 sum+= filter[i*filterSize + j];
1430 }
1431 sum= (sum + one/2)/ one;
1432 for (j=0; j<*outFilterSize; j++)
1433 {
1434 int64_t v= filter[i*filterSize + j] + error;
1435 int intV= ROUNDED_DIV(v, sum);
1436 (*outFilter)[i*(*outFilterSize) + j]= intV;
1437 error= v - intV*sum;
1438 }
1439 }
1440
1441 (*filterPos)[dstW]= (*filterPos)[dstW-1]; // the MMX scaler will read over the end
1442 for (i=0; i<*outFilterSize; i++)
1443 {
1444 int j= dstW*(*outFilterSize);
1445 (*outFilter)[j + i]= (*outFilter)[j + i - (*outFilterSize)];
1446 }
1447
1448 ret=0;
1449 error:
1450 av_free(filter);
1451 av_free(filter2);
1452 return ret;
1453 }
1454
1455 #ifdef COMPILE_MMX2
1456 static void initMMX2HScaler(int dstW, int xInc, uint8_t *funnyCode, int16_t *filter, int32_t *filterPos, int numSplits)
1457 {
1458 uint8_t *fragmentA;
1459 long imm8OfPShufW1A;
1460 long imm8OfPShufW2A;
1461 long fragmentLengthA;
1462 uint8_t *fragmentB;
1463 long imm8OfPShufW1B;
1464 long imm8OfPShufW2B;
1465 long fragmentLengthB;
1466 int fragmentPos;
1467
1468 int xpos, i;
1469
1470 // create an optimized horizontal scaling routine
1471
1472 //code fragment
1473
1474 __asm__ volatile(
1475 "jmp 9f \n\t"
1476 // Begin
1477 "0: \n\t"
1478 "movq (%%"REG_d", %%"REG_a"), %%mm3 \n\t"
1479 "movd (%%"REG_c", %%"REG_S"), %%mm0 \n\t"
1480 "movd 1(%%"REG_c", %%"REG_S"), %%mm1 \n\t"
1481 "punpcklbw %%mm7, %%mm1 \n\t"
1482 "punpcklbw %%mm7, %%mm0 \n\t"
1483 "pshufw $0xFF, %%mm1, %%mm1 \n\t"
1484 "1: \n\t"
1485 "pshufw $0xFF, %%mm0, %%mm0 \n\t"
1486 "2: \n\t"
1487 "psubw %%mm1, %%mm0 \n\t"
1488 "movl 8(%%"REG_b", %%"REG_a"), %%esi \n\t"
1489 "pmullw %%mm3, %%mm0 \n\t"
1490 "psllw $7, %%mm1 \n\t"
1491 "paddw %%mm1, %%mm0 \n\t"
1492
1493 "movq %%mm0, (%%"REG_D", %%"REG_a") \n\t"
1494
1495 "add $8, %%"REG_a" \n\t"
1496 // End
1497 "9: \n\t"
1498 // "int $3 \n\t"
1499 "lea " LOCAL_MANGLE(0b) ", %0 \n\t"
1500 "lea " LOCAL_MANGLE(1b) ", %1 \n\t"
1501 "lea " LOCAL_MANGLE(2b) ", %2 \n\t"
1502 "dec %1 \n\t"
1503 "dec %2 \n\t"
1504 "sub %0, %1 \n\t"
1505 "sub %0, %2 \n\t"
1506 "lea " LOCAL_MANGLE(9b) ", %3 \n\t"
1507 "sub %0, %3 \n\t"
1508
1509
1510 :"=r" (fragmentA), "=r" (imm8OfPShufW1A), "=r" (imm8OfPShufW2A),
1511 "=r" (fragmentLengthA)
1512 );
1513
1514 __asm__ volatile(
1515 "jmp 9f \n\t"
1516 // Begin
1517 "0: \n\t"
1518 "movq (%%"REG_d", %%"REG_a"), %%mm3 \n\t"
1519 "movd (%%"REG_c", %%"REG_S"), %%mm0 \n\t"
1520 "punpcklbw %%mm7, %%mm0 \n\t"
1521 "pshufw $0xFF, %%mm0, %%mm1 \n\t"
1522 "1: \n\t"
1523 "pshufw $0xFF, %%mm0, %%mm0 \n\t"
1524 "2: \n\t"
1525 "psubw %%mm1, %%mm0 \n\t"
1526 "movl 8(%%"REG_b", %%"REG_a"), %%esi \n\t"
1527 "pmullw %%mm3, %%mm0 \n\t"
1528 "psllw $7, %%mm1 \n\t"
1529 "paddw %%mm1, %%mm0 \n\t"
1530
1531 "movq %%mm0, (%%"REG_D", %%"REG_a") \n\t"
1532
1533 "add $8, %%"REG_a" \n\t"
1534 // End
1535 "9: \n\t"
1536 // "int $3 \n\t"
1537 "lea " LOCAL_MANGLE(0b) ", %0 \n\t"
1538 "lea " LOCAL_MANGLE(1b) ", %1 \n\t"
1539 "lea " LOCAL_MANGLE(2b) ", %2 \n\t"
1540 "dec %1 \n\t"
1541 "dec %2 \n\t"
1542 "sub %0, %1 \n\t"
1543 "sub %0, %2 \n\t"
1544 "lea " LOCAL_MANGLE(9b) ", %3 \n\t"
1545 "sub %0, %3 \n\t"
1546
1547
1548 :"=r" (fragmentB), "=r" (imm8OfPShufW1B), "=r" (imm8OfPShufW2B),
1549 "=r" (fragmentLengthB)
1550 );
1551
1552 xpos= 0; //lumXInc/2 - 0x8000; // difference between pixel centers
1553 fragmentPos=0;
1554
1555 for (i=0; i<dstW/numSplits; i++)
1556 {
1557 int xx=xpos>>16;
1558
1559 if ((i&3) == 0)
1560 {
1561 int a=0;
1562 int b=((xpos+xInc)>>16) - xx;
1563 int c=((xpos+xInc*2)>>16) - xx;
1564 int d=((xpos+xInc*3)>>16) - xx;
1565
1566 filter[i ] = (( xpos & 0xFFFF) ^ 0xFFFF)>>9;
1567 filter[i+1] = (((xpos+xInc ) & 0xFFFF) ^ 0xFFFF)>>9;
1568 filter[i+2] = (((xpos+xInc*2) & 0xFFFF) ^ 0xFFFF)>>9;
1569 filter[i+3] = (((xpos+xInc*3) & 0xFFFF) ^ 0xFFFF)>>9;
1570 filterPos[i/2]= xx;
1571
1572 if (d+1<4)
1573 {
1574 int maxShift= 3-(d+1);
1575 int shift=0;
1576
1577 memcpy(funnyCode + fragmentPos, fragmentB, fragmentLengthB);
1578
1579 funnyCode[fragmentPos + imm8OfPShufW1B]=
1580 (a+1) | ((b+1)<<2) | ((c+1)<<4) | ((d+1)<<6);
1581 funnyCode[fragmentPos + imm8OfPShufW2B]=
1582 a | (b<<2) | (c<<4) | (d<<6);
1583
1584 if (i+3>=dstW) shift=maxShift; //avoid overread
1585 else if ((filterPos[i/2]&3) <= maxShift) shift=filterPos[i/2]&3; //Align
1586
1587 if (shift && i>=shift)
1588 {
1589 funnyCode[fragmentPos + imm8OfPShufW1B]+= 0x55*shift;
1590 funnyCode[fragmentPos + imm8OfPShufW2B]+= 0x55*shift;
1591 filterPos[i/2]-=shift;
1592 }
1593
1594 fragmentPos+= fragmentLengthB;
1595 }
1596 else
1597 {
1598 int maxShift= 3-d;
1599 int shift=0;
1600
1601 memcpy(funnyCode + fragmentPos, fragmentA, fragmentLengthA);
1602
1603 funnyCode[fragmentPos + imm8OfPShufW1A]=
1604 funnyCode[fragmentPos + imm8OfPShufW2A]=
1605 a | (b<<2) | (c<<4) | (d<<6);
1606
1607 if (i+4>=dstW) shift=maxShift; //avoid overread
1608 else if ((filterPos[i/2]&3) <= maxShift) shift=filterPos[i/2]&3; //partial align
1609
1610 if (shift && i>=shift)
1611 {
1612 funnyCode[fragmentPos + imm8OfPShufW1A]+= 0x55*shift;
1613 funnyCode[fragmentPos + imm8OfPShufW2A]+= 0x55*shift;
1614 filterPos[i/2]-=shift;
1615 }
1616
1617 fragmentPos+= fragmentLengthA;
1618 }
1619
1620 funnyCode[fragmentPos]= RET;
1621 }
1622 xpos+=xInc;
1623 }
1624 filterPos[i/2]= xpos>>16; // needed to jump to the next part
1625 }
1626 #endif /* COMPILE_MMX2 */
1627
1628 static void globalInit(void){
1629 // generating tables:
1630 int i;
1631 for (i=0; i<768; i++){
1632 int c= av_clip_uint8(i-256);
1633 clip_table[i]=c;
1634 }
1635 }
1636
1637 static SwsFunc getSwsFunc(int flags){
1638
1639 #if defined(RUNTIME_CPUDETECT) && CONFIG_GPL
1640 #if ARCH_X86
1641 // ordered per speed fastest first
1642 if (flags & SWS_CPU_CAPS_MMX2)
1643 return swScale_MMX2;
1644 else if (flags & SWS_CPU_CAPS_3DNOW)
1645 return swScale_3DNow;
1646 else if (flags & SWS_CPU_CAPS_MMX)
1647 return swScale_MMX;
1648 else
1649 return swScale_C;
1650
1651 #else
1652 #if ARCH_PPC
1653 if (flags & SWS_CPU_CAPS_ALTIVEC)
1654 return swScale_altivec;
1655 else
1656 return swScale_C;
1657 #endif
1658 return swScale_C;
1659 #endif /* ARCH_X86 */
1660 #else //RUNTIME_CPUDETECT
1661 #if HAVE_MMX2
1662 return swScale_MMX2;
1663 #elif HAVE_3DNOW
1664 return swScale_3DNow;
1665 #elif HAVE_MMX
1666 return swScale_MMX;
1667 #elif HAVE_ALTIVEC
1668 return swScale_altivec;
1669 #else
1670 return swScale_C;
1671 #endif
1672 #endif //!RUNTIME_CPUDETECT
1673 }
1674
1675 static int PlanarToNV12Wrapper(SwsContext *c, uint8_t* src[], int srcStride[], int srcSliceY,
1676 int srcSliceH, uint8_t* dstParam[], int dstStride[]){
1677 uint8_t *dst=dstParam[0] + dstStride[0]*srcSliceY;
1678 /* Copy Y plane */
1679 if (dstStride[0]==srcStride[0] && srcStride[0] > 0)
1680 memcpy(dst, src[0], srcSliceH*dstStride[0]);
1681 else
1682 {
1683 int i;
1684 uint8_t *srcPtr= src[0];
1685 uint8_t *dstPtr= dst;
1686 for (i=0; i<srcSliceH; i++)
1687 {
1688 memcpy(dstPtr, srcPtr, c->srcW);
1689 srcPtr+= srcStride[0];
1690 dstPtr+= dstStride[0];
1691 }
1692 }
1693 dst = dstParam[1] + dstStride[1]*srcSliceY/2;
1694 if (c->dstFormat == PIX_FMT_NV12)
1695 interleaveBytes(src[1], src[2], dst, c->srcW/2, srcSliceH/2, srcStride[1], srcStride[2], dstStride[0]);
1696 else
1697 interleaveBytes(src[2], src[1], dst, c->srcW/2, srcSliceH/2, srcStride[2], srcStride[1], dstStride[0]);
1698
1699 return srcSliceH;
1700 }
1701
1702 static int PlanarToYuy2Wrapper(SwsContext *c, uint8_t* src[], int srcStride[], int srcSliceY,
1703 int srcSliceH, uint8_t* dstParam[], int dstStride[]){
1704 uint8_t *dst=dstParam[0] + dstStride[0]*srcSliceY;
1705
1706 yv12toyuy2(src[0], src[1], src[2], dst, c->srcW, srcSliceH, srcStride[0], srcStride[1], dstStride[0]);
1707
1708 return srcSliceH;
1709 }
1710
1711 static int PlanarToUyvyWrapper(SwsContext *c, uint8_t* src[], int srcStride[], int srcSliceY,
1712 int srcSliceH, uint8_t* dstParam[], int dstStride[]){
1713 uint8_t *dst=dstParam[0] + dstStride[0]*srcSliceY;
1714
1715 yv12touyvy(src[0], src[1], src[2], dst, c->srcW, srcSliceH, srcStride[0], srcStride[1], dstStride[0]);
1716
1717 return srcSliceH;
1718 }
1719
1720 static int YUV422PToYuy2Wrapper(SwsContext *c, uint8_t* src[], int srcStride[], int srcSliceY,
1721 int srcSliceH, uint8_t* dstParam[], int dstStride[]){
1722 uint8_t *dst=dstParam[0] + dstStride[0]*srcSliceY;
1723
1724 yuv422ptoyuy2(src[0],src[1],src[2],dst,c->srcW,srcSliceH,srcStride[0],srcStride[1],dstStride[0]);
1725
1726 return srcSliceH;
1727 }
1728
1729 static int YUV422PToUyvyWrapper(SwsContext *c, uint8_t* src[], int srcStride[], int srcSliceY,
1730 int srcSliceH, uint8_t* dstParam[], int dstStride[]){
1731 uint8_t *dst=dstParam[0] + dstStride[0]*srcSliceY;
1732
1733 yuv422ptouyvy(src[0],src[1],src[2],dst,c->srcW,srcSliceH,srcStride[0],srcStride[1],dstStride[0]);
1734
1735 return srcSliceH;
1736 }
1737
1738 static int pal2rgbWrapper(SwsContext *c, uint8_t* src[], int srcStride[], int srcSliceY,
1739 int srcSliceH, uint8_t* dst[], int dstStride[]){
1740 const enum PixelFormat srcFormat= c->srcFormat;
1741 const enum PixelFormat dstFormat= c->dstFormat;
1742 void (*conv)(const uint8_t *src, uint8_t *dst, long num_pixels,
1743 const uint8_t *palette)=NULL;
1744 int i;
1745 uint8_t *dstPtr= dst[0] + dstStride[0]*srcSliceY;
1746 uint8_t *srcPtr= src[0];
1747
1748 if (!usePal(srcFormat))
1749 av_log(c, AV_LOG_ERROR, "internal error %s -> %s converter\n",
1750 sws_format_name(srcFormat), sws_format_name(dstFormat));
1751
1752 switch(dstFormat){
1753 case PIX_FMT_RGB32 : conv = palette8topacked32; break;
1754 case PIX_FMT_BGR32 : conv = palette8topacked32; break;
1755 case PIX_FMT_BGR32_1: conv = palette8topacked32; break;
1756 case PIX_FMT_RGB32_1: conv = palette8topacked32; break;
1757 case PIX_FMT_RGB24 : conv = palette8topacked24; break;
1758 case PIX_FMT_BGR24 : conv = palette8topacked24; break;
1759 default: av_log(c, AV_LOG_ERROR, "internal error %s -> %s converter\n",
1760 sws_format_name(srcFormat), sws_format_name(dstFormat)); break;
1761 }
1762
1763
1764 for (i=0; i<srcSliceH; i++) {
1765 conv(srcPtr, dstPtr, c->srcW, (uint8_t *) c->pal_rgb);
1766 srcPtr+= srcStride[0];
1767 dstPtr+= dstStride[0];
1768 }
1769
1770 return srcSliceH;
1771 }
1772
1773 /* {RGB,BGR}{15,16,24,32,32_1} -> {RGB,BGR}{15,16,24,32} */
1774 static int rgb2rgbWrapper(SwsContext *c, uint8_t* src[], int srcStride[], int srcSliceY,
1775 int srcSliceH, uint8_t* dst[], int dstStride[]){
1776 const enum PixelFormat srcFormat= c->srcFormat;
1777 const enum PixelFormat dstFormat= c->dstFormat;
1778 const int srcBpp= (fmt_depth(srcFormat) + 7) >> 3;
1779 const int dstBpp= (fmt_depth(dstFormat) + 7) >> 3;
1780 const int srcId= fmt_depth(srcFormat) >> 2; /* 1:0, 4:1, 8:2, 15:3, 16:4, 24:6, 32:8 */
1781 const int dstId= fmt_depth(dstFormat) >> 2;
1782 void (*conv)(const uint8_t *src, uint8_t *dst, long src_size)=NULL;
1783
1784 /* BGR -> BGR */
1785 if ( (isBGR(srcFormat) && isBGR(dstFormat))
1786 || (isRGB(srcFormat) && isRGB(dstFormat))){
1787 switch(srcId | (dstId<<4)){
1788 case 0x34: conv= rgb16to15; break;
1789 case 0x36: conv= rgb24to15; break;
1790 case 0x38: conv= rgb32to15; break;
1791 case 0x43: conv= rgb15to16; break;
1792 case 0x46: conv= rgb24to16; break;
1793 case 0x48: conv= rgb32to16; break;
1794 case 0x63: conv= rgb15to24; break;
1795 case 0x64: conv= rgb16to24; break;
1796 case 0x68: conv= rgb32to24; break;
1797 case 0x83: conv= rgb15to32; break;
1798 case 0x84: conv= rgb16to32; break;
1799 case 0x86: conv= rgb24to32; break;
1800 default: av_log(c, AV_LOG_ERROR, "internal error %s -> %s converter\n",
1801 sws_format_name(srcFormat), sws_format_name(dstFormat)); break;
1802 }
1803 }else if ( (isBGR(srcFormat) && isRGB(dstFormat))
1804 || (isRGB(srcFormat) && isBGR(dstFormat))){
1805 switch(srcId | (dstId<<4)){
1806 case 0x33: conv= rgb15tobgr15; break;
1807 case 0x34: conv= rgb16tobgr15; break;
1808 case 0x36: conv= rgb24tobgr15; break;
1809 case 0x38: conv= rgb32tobgr15; break;
1810 case 0x43: conv= rgb15tobgr16; break;
1811 case 0x44: conv= rgb16tobgr16; break;
1812 case 0x46: conv= rgb24tobgr16; break;
1813 case 0x48: conv= rgb32tobgr16; break;
1814 case 0x63: conv= rgb15tobgr24; break;
1815 case 0x64: conv= rgb16tobgr24; break;
1816 case 0x66: conv= rgb24tobgr24; break;
1817 case 0x68: conv= rgb32tobgr24; break;
1818 case 0x83: conv= rgb15tobgr32; break;
1819 case 0x84: conv= rgb16tobgr32; break;
1820 case 0x86: conv= rgb24tobgr32; break;
1821 case 0x88: conv= rgb32tobgr32; break;
1822 default: av_log(c, AV_LOG_ERROR, "internal error %s -> %s converter\n",
1823 sws_format_name(srcFormat), sws_format_name(dstFormat)); break;
1824 }
1825 }else{
1826 av_log(c, AV_LOG_ERROR, "internal error %s -> %s converter\n",
1827 sws_format_name(srcFormat), sws_format_name(dstFormat));
1828 }
1829
1830 if(conv)
1831 {
1832 uint8_t *srcPtr= src[0];
1833 if(srcFormat == PIX_FMT_RGB32_1 || srcFormat == PIX_FMT_BGR32_1)
1834 srcPtr += ALT32_CORR;
1835
1836 if (dstStride[0]*srcBpp == srcStride[0]*dstBpp && srcStride[0] > 0)
1837 conv(srcPtr, dst[0] + dstStride[0]*srcSliceY, srcSliceH*srcStride[0]);
1838 else
1839 {
1840 int i;
1841 uint8_t *dstPtr= dst[0] + dstStride[0]*srcSliceY;
1842
1843 for (i=0; i<srcSliceH; i++)
1844 {
1845 conv(srcPtr, dstPtr, c->srcW*srcBpp);
1846 srcPtr+= srcStride[0];
1847 dstPtr+= dstStride[0];
1848 }
1849 }
1850 }
1851 return srcSliceH;
1852 }
1853
1854 static int bgr24toyv12Wrapper(SwsContext *c, uint8_t* src[], int srcStride[], int srcSliceY,
1855 int srcSliceH, uint8_t* dst[], int dstStride[]){
1856
1857 rgb24toyv12(
1858 src[0],
1859 dst[0]+ srcSliceY *dstStride[0],
1860 dst[1]+(srcSliceY>>1)*dstStride[1],
1861 dst[2]+(srcSliceY>>1)*dstStride[2],
1862 c->srcW, srcSliceH,
1863 dstStride[0], dstStride[1], srcStride[0]);
1864 return srcSliceH;
1865 }
1866
1867 static int yvu9toyv12Wrapper(SwsContext *c, uint8_t* src[], int srcStride[], int srcSliceY,
1868 int srcSliceH, uint8_t* dst[], int dstStride[]){
1869 int i;
1870
1871 /* copy Y */
1872 if (srcStride[0]==dstStride[0] && srcStride[0] > 0)
1873 memcpy(dst[0]+ srcSliceY*dstStride[0], src[0], srcStride[0]*srcSliceH);
1874 else{
1875 uint8_t *srcPtr= src[0];
1876 uint8_t *dstPtr= dst[0] + dstStride[0]*srcSliceY;
1877
1878 for (i=0; i<srcSliceH; i++)
1879 {
1880 memcpy(dstPtr, srcPtr, c->srcW);
1881 srcPtr+= srcStride[0];
1882 dstPtr+= dstStride[0];
1883 }
1884 }
1885
1886 if (c->dstFormat==PIX_FMT_YUV420P){
1887 planar2x(src[1], dst[1], c->chrSrcW, c->chrSrcH, srcStride[1], dstStride[1]);
1888 planar2x(src[2], dst[2], c->chrSrcW, c->chrSrcH, srcStride[2], dstStride[2]);
1889 }else{
1890 planar2x(src[1], dst[2], c->chrSrcW, c->chrSrcH, srcStride[1], dstStride[2]);
1891 planar2x(src[2], dst[1], c->chrSrcW, c->chrSrcH, srcStride[2], dstStride[1]);
1892 }
1893 return srcSliceH;
1894 }
1895
1896 /* unscaled copy like stuff (assumes nearly identical formats) */
1897 static int packedCopy(SwsContext *c, uint8_t* src[], int srcStride[], int srcSliceY,
1898 int srcSliceH, uint8_t* dst[], int dstStride[])
1899 {
1900 if (dstStride[0]==srcStride[0] && srcStride[0] > 0)
1901 memcpy(dst[0] + dstStride[0]*srcSliceY, src[0], srcSliceH*dstStride[0]);
1902 else
1903 {
1904 int i;
1905 uint8_t *srcPtr= src[0];
1906 uint8_t *dstPtr= dst[0] + dstStride[0]*srcSliceY;
1907 int length=0;
1908
1909 /* universal length finder */
1910 while(length+c->srcW <= FFABS(dstStride[0])
1911 && length+c->srcW <= FFABS(srcStride[0])) length+= c->srcW;
1912 assert(length!=0);
1913
1914 for (i=0; i<srcSliceH; i++)
1915 {
1916 memcpy(dstPtr, srcPtr, length);
1917 srcPtr+= srcStride[0];
1918 dstPtr+= dstStride[0];
1919 }
1920 }
1921 return srcSliceH;
1922 }
1923
1924 static int planarCopy(SwsContext *c, uint8_t* src[], int srcStride[], int srcSliceY,
1925 int srcSliceH, uint8_t* dst[], int dstStride[])
1926 {
1927 int plane;
1928 for (plane=0; plane<3; plane++)
1929 {
1930 int length= plane==0 ? c->srcW : -((-c->srcW )>>c->chrDstHSubSample);
1931 int y= plane==0 ? srcSliceY: -((-srcSliceY)>>c->chrDstVSubSample);
1932 int height= plane==0 ? srcSliceH: -((-srcSliceH)>>c->chrDstVSubSample);
1933
1934 if ((isGray(c->srcFormat) || isGray(c->dstFormat)) && plane>0)
1935 {
1936 if (!isGray(c->dstFormat))
1937 memset(dst[plane], 128, dstStride[plane]*height);
1938 }
1939 else
1940 {
1941 if (dstStride[plane]==srcStride[plane] && srcStride[plane] > 0)
1942 memcpy(dst[plane] + dstStride[plane]*y, src[plane], height*dstStride[plane]);
1943 else
1944 {
1945 int i;
1946 uint8_t *srcPtr= src[plane];
1947 uint8_t *dstPtr= dst[plane] + dstStride[plane]*y;
1948 for (i=0; i<height; i++)
1949 {
1950 memcpy(dstPtr, srcPtr, length);
1951 srcPtr+= srcStride[plane];
1952 dstPtr+= dstStride[plane];
1953 }
1954 }
1955 }
1956 }
1957 return srcSliceH;
1958 }
1959
1960 static int gray16togray(SwsContext *c, uint8_t* src[], int srcStride[], int srcSliceY,
1961 int srcSliceH, uint8_t* dst[], int dstStride[]){
1962
1963 int length= c->srcW;
1964 int y= srcSliceY;
1965 int height= srcSliceH;
1966 int i, j;
1967 uint8_t *srcPtr= src[0];
1968 uint8_t *dstPtr= dst[0] + dstStride[0]*y;
1969
1970 if (!isGray(c->dstFormat)){
1971 int height= -((-srcSliceH)>>c->chrDstVSubSample);
1972 memset(dst[1], 128, dstStride[1]*height);
1973 memset(dst[2], 128, dstStride[2]*height);
1974 }
1975 if (c->srcFormat == PIX_FMT_GRAY16LE) srcPtr++;
1976 for (i=0; i<height; i++)
1977 {
1978 for (j=0; j<length; j++) dstPtr[j] = srcPtr[j<<1];
1979 srcPtr+= srcStride[0];
1980 dstPtr+= dstStride[0];
1981 }
1982 return srcSliceH;
1983 }
1984
1985 static int graytogray16(SwsContext *c, uint8_t* src[], int srcStride[], int srcSliceY,
1986 int srcSliceH, uint8_t* dst[], int dstStride[]){
1987
1988 int length= c->srcW;
1989 int y= srcSliceY;
1990 int height= srcSliceH;
1991 int i, j;
1992 uint8_t *srcPtr= src[0];
1993 uint8_t *dstPtr= dst[0] + dstStride[0]*y;
1994 for (i=0; i<height; i++)
1995 {
1996 for (j=0; j<length; j++)
1997 {
1998 dstPtr[j<<1] = srcPtr[j];
1999 dstPtr[(j<<1)+1] = srcPtr[j];
2000 }
2001 srcPtr+= srcStride[0];
2002 dstPtr+= dstStride[0];
2003 }
2004 return srcSliceH;
2005 }
2006
2007 static int gray16swap(SwsContext *c, uint8_t* src[], int srcStride[], int srcSliceY,
2008 int srcSliceH, uint8_t* dst[], int dstStride[]){
2009
2010 int length= c->srcW;
2011 int y= srcSliceY;
2012 int height= srcSliceH;
2013 int i, j;
2014 uint16_t *srcPtr= (uint16_t*)src[0];
2015 uint16_t *dstPtr= (uint16_t*)(dst[0] + dstStride[0]*y/2);
2016 for (i=0; i<height; i++)
2017 {
2018 for (j=0; j<length; j++) dstPtr[j] = bswap_16(srcPtr[j]);
2019 srcPtr+= srcStride[0]/2;
2020 dstPtr+= dstStride[0]/2;
2021 }
2022 return srcSliceH;
2023 }
2024
2025
2026 static void getSubSampleFactors(int *h, int *v, int format){
2027 switch(format){
2028 case PIX_FMT_UYVY422:
2029 case PIX_FMT_YUYV422:
2030 *h=1;
2031 *v=0;
2032 break;
2033 case PIX_FMT_YUV420P:
2034 case PIX_FMT_YUVA420P:
2035 case PIX_FMT_GRAY16BE:
2036 case PIX_FMT_GRAY16LE:
2037 case PIX_FMT_GRAY8: //FIXME remove after different subsamplings are fully implemented
2038 case PIX_FMT_NV12:
2039 case PIX_FMT_NV21:
2040 *h=1;
2041 *v=1;
2042 break;
2043 case PIX_FMT_YUV440P:
2044 *h=0;
2045 *v=1;
2046 break;
2047 case PIX_FMT_YUV410P:
2048 *h=2;
2049 *v=2;
2050 break;
2051 case PIX_FMT_YUV444P:
2052 *h=0;
2053 *v=0;
2054 break;
2055 case PIX_FMT_YUV422P:
2056 *h=1;
2057 *v=0;
2058 break;
2059 case PIX_FMT_YUV411P:
2060 *h=2;
2061 *v=0;
2062 break;
2063 default:
2064 *h=0;
2065 *v=0;
2066 break;
2067 }
2068 }
2069
2070 static uint16_t roundToInt16(int64_t f){
2071 int r= (f + (1<<15))>>16;
2072 if (r<-0x7FFF) return 0x8000;
2073 else if (r> 0x7FFF) return 0x7FFF;
2074 else return r;
2075 }
2076
2077 /**
2078 * @param inv_table the yuv2rgb coeffs, normally Inverse_Table_6_9[x]
2079 * @param fullRange if 1 then the luma range is 0..255 if 0 it is 16..235
2080 * @return -1 if not supported
2081 */
2082 int sws_setColorspaceDetails(SwsContext *c, const int inv_table[4], int srcRange, const int table[4], int dstRange, int brightness, int contrast, int saturation){
2083 int64_t crv = inv_table[0];
2084 int64_t cbu = inv_table[1];
2085 int64_t cgu = -inv_table[2];
2086 int64_t cgv = -inv_table[3];
2087 int64_t cy = 1<<16;
2088 int64_t oy = 0;
2089
2090 memcpy(c->srcColorspaceTable, inv_table, sizeof(int)*4);
2091 memcpy(c->dstColorspaceTable, table, sizeof(int)*4);
2092
2093 c->brightness= brightness;
2094 c->contrast = contrast;
2095 c->saturation= saturation;
2096 c->srcRange = srcRange;
2097 c->dstRange = dstRange;
2098 if (isYUV(c->dstFormat) || isGray(c->dstFormat)) return 0;
2099
2100 c->uOffset= 0x0400040004000400LL;
2101 c->vOffset= 0x0400040004000400LL;
2102
2103 if (!srcRange){
2104 cy= (cy*255) / 219;
2105 oy= 16<<16;
2106 }else{
2107 crv= (crv*224) / 255;
2108 cbu= (cbu*224) / 255;
2109 cgu= (cgu*224) / 255;
2110 cgv= (cgv*224) / 255;
2111 }
2112
2113 cy = (cy *contrast )>>16;
2114 crv= (crv*contrast * saturation)>>32;
2115 cbu= (cbu*contrast * saturation)>>32;
2116 cgu= (cgu*contrast * saturation)>>32;
2117 cgv= (cgv*contrast * saturation)>>32;
2118
2119 oy -= 256*brightness;
2120
2121 c->yCoeff= roundToInt16(cy *8192) * 0x0001000100010001ULL;
2122 c->vrCoeff= roundToInt16(crv*8192) * 0x0001000100010001ULL;
2123 c->ubCoeff= roundToInt16(cbu*8192) * 0x0001000100010001ULL;
2124 c->vgCoeff= roundToInt16(cgv*8192) * 0x0001000100010001ULL;
2125 c->ugCoeff= roundToInt16(cgu*8192) * 0x0001000100010001ULL;
2126 c->yOffset= roundToInt16(oy * 8) * 0x0001000100010001ULL;
2127
2128 c->yuv2rgb_y_coeff = (int16_t)roundToInt16(cy <<13);
2129 c->yuv2rgb_y_offset = (int16_t)roundToInt16(oy << 9);
2130 c->yuv2rgb_v2r_coeff= (int16_t)roundToInt16(crv<<13);
2131 c->yuv2rgb_v2g_coeff= (int16_t)roundToInt16(cgv<<13);
2132 c->yuv2rgb_u2g_coeff= (int16_t)roundToInt16(cgu<<13);
2133 c->yuv2rgb_u2b_coeff= (int16_t)roundToInt16(cbu<<13);
2134
2135 yuv2rgb_c_init_tables(c, inv_table, srcRange, brightness, contrast, saturation);
2136 //FIXME factorize
2137
2138 #ifdef COMPILE_ALTIVEC
2139 if (c->flags & SWS_CPU_CAPS_ALTIVEC)
2140 yuv2rgb_altivec_init_tables (c, inv_table, brightness, contrast, saturation);
2141 #endif
2142 return 0;
2143 }
2144
2145 /**
2146 * @return -1 if not supported
2147 */
2148 int sws_getColorspaceDetails(SwsContext *c, int **inv_table, int *srcRange, int **table, int *dstRange, int *brightness, int *contrast, int *saturation){
2149 if (isYUV(c->dstFormat) || isGray(c->dstFormat)) return -1;
2150
2151 *inv_table = c->srcColorspaceTable;
2152 *table = c->dstColorspaceTable;
2153 *srcRange = c->srcRange;
2154 *dstRange = c->dstRange;
2155 *brightness= c->brightness;
2156 *contrast = c->contrast;
2157 *saturation= c->saturation;
2158
2159 return 0;
2160 }
2161
2162 static int handle_jpeg(enum PixelFormat *format)
2163 {
2164 switch (*format) {
2165 case PIX_FMT_YUVJ420P:
2166 *format = PIX_FMT_YUV420P;
2167 return 1;
2168 case PIX_FMT_YUVJ422P:
2169 *format = PIX_FMT_YUV422P;
2170 return 1;
2171 case PIX_FMT_YUVJ444P:
2172 *format = PIX_FMT_YUV444P;
2173 return 1;
2174 case PIX_FMT_YUVJ440P:
2175 *format = PIX_FMT_YUV440P;
2176 return 1;
2177 default:
2178 return 0;
2179 }
2180 }
2181
2182 SwsContext *sws_getContext(int srcW, int srcH, enum PixelFormat srcFormat, int dstW, int dstH, enum PixelFormat dstFormat, int flags,
2183 SwsFilter *srcFilter, SwsFilter *dstFilter, double *param){
2184
2185 SwsContext *c;
2186 int i;
2187 int usesVFilter, usesHFilter;
2188 int unscaled, needsDither;
2189 int srcRange, dstRange;
2190 SwsFilter dummyFilter= {NULL, NULL, NULL, NULL};
2191 #if ARCH_X86
2192 if (flags & SWS_CPU_CAPS_MMX)
2193 __asm__ volatile("emms\n\t"::: "memory");
2194 #endif
2195
2196 #if !defined(RUNTIME_CPUDETECT) || !CONFIG_GPL //ensure that the flags match the compiled variant if cpudetect is off
2197 flags &= ~(SWS_CPU_CAPS_MMX|SWS_CPU_CAPS_MMX2|SWS_CPU_CAPS_3DNOW|SWS_CPU_CAPS_ALTIVEC|SWS_CPU_CAPS_BFIN);
2198 #if HAVE_MMX2
2199 flags |= SWS_CPU_CAPS_MMX|SWS_CPU_CAPS_MMX2;
2200 #elif HAVE_3DNOW
2201 flags |= SWS_CPU_CAPS_MMX|SWS_CPU_CAPS_3DNOW;
2202 #elif HAVE_MMX
2203 flags |= SWS_CPU_CAPS_MMX;
2204 #elif HAVE_ALTIVEC
2205 flags |= SWS_CPU_CAPS_ALTIVEC;
2206 #elif ARCH_BFIN
2207 flags |= SWS_CPU_CAPS_BFIN;
2208 #endif
2209 #endif /* RUNTIME_CPUDETECT */
2210 if (clip_table[512] != 255) globalInit();
2211 if (!rgb15to16) sws_rgb2rgb_init(flags);
2212
2213 unscaled = (srcW == dstW && srcH == dstH);
2214 needsDither= (isBGR(dstFormat) || isRGB(dstFormat))
2215 && (fmt_depth(dstFormat))<24
2216 && ((fmt_depth(dstFormat))<(fmt_depth(srcFormat)) || (!(isRGB(srcFormat) || isBGR(srcFormat))));
2217
2218 srcRange = handle_jpeg(&srcFormat);
2219 dstRange = handle_jpeg(&dstFormat);
2220
2221 if (!isSupportedIn(srcFormat))
2222 {
2223 av_log(NULL, AV_LOG_ERROR, "swScaler: %s is not supported as input pixel format\n", sws_format_name(srcFormat));
2224 return NULL;
2225 }
2226 if (!isSupportedOut(dstFormat))
2227 {
2228 av_log(NULL, AV_LOG_ERROR, "swScaler: %s is not supported as output pixel format\n", sws_format_name(dstFormat));
2229 return NULL;
2230 }
2231
2232 i= flags & ( SWS_POINT
2233 |SWS_AREA
2234 |SWS_BILINEAR
2235 |SWS_FAST_BILINEAR
2236 |SWS_BICUBIC
2237 |SWS_X
2238 |SWS_GAUSS
2239 |SWS_LANCZOS
2240 |SWS_SINC
2241 |SWS_SPLINE
2242 |SWS_BICUBLIN);
2243 if(!i || (i & (i-1)))
2244 {
2245 av_log(NULL, AV_LOG_ERROR, "swScaler: Exactly one scaler algorithm must be choosen\n");
2246 return NULL;
2247 }
2248
2249 /* sanity check */
2250 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
2251 {
2252 av_log(NULL, AV_LOG_ERROR, "swScaler: %dx%d -> %dx%d is invalid scaling dimension\n",
2253 srcW, srcH, dstW, dstH);
2254 return NULL;
2255 }
2256 if(srcW > VOFW || dstW > VOFW){
2257 av_log(NULL, AV_LOG_ERROR, "swScaler: Compile time max width is "AV_STRINGIFY(VOFW)" change VOF/VOFW and recompile\n");
2258 return NULL;
2259 }
2260
2261 if (!dstFilter) dstFilter= &dummyFilter;
2262 if (!srcFilter) srcFilter= &dummyFilter;
2263
2264 c= av_mallocz(sizeof(SwsContext));
2265
2266 c->av_class = &sws_context_class;
2267 c->srcW= srcW;
2268 c->srcH= srcH;
2269 c->dstW= dstW;
2270 c->dstH= dstH;
2271 c->lumXInc= ((srcW<<16) + (dstW>>1))/dstW;
2272 c->lumYInc= ((srcH<<16) + (dstH>>1))/dstH;
2273 c->flags= flags;
2274 c->dstFormat= dstFormat;
2275 c->srcFormat= srcFormat;
2276 c->vRounder= 4* 0x0001000100010001ULL;
2277
2278 usesHFilter= usesVFilter= 0;
2279 if (dstFilter->lumV && dstFilter->lumV->length>1) usesVFilter=1;
2280 if (dstFilter->lumH && dstFilter->lumH->length>1) usesHFilter=1;
2281 if (dstFilter->chrV && dstFilter->chrV->length>1) usesVFilter=1;
2282 if (dstFilter->chrH && dstFilter->chrH->length>1) usesHFilter=1;
2283 if (srcFilter->lumV && srcFilter->lumV->length>1) usesVFilter=1;
2284 if (srcFilter->lumH && srcFilter->lumH->length>1) usesHFilter=1;
2285 if (srcFilter->chrV && srcFilter->chrV->length>1) usesVFilter=1;
2286 if (srcFilter->chrH && srcFilter->chrH->length>1) usesHFilter=1;
2287
2288 getSubSampleFactors(&c->chrSrcHSubSample, &c->chrSrcVSubSample, srcFormat);
2289 getSubSampleFactors(&c->chrDstHSubSample, &c->chrDstVSubSample, dstFormat);
2290
2291 // reuse chroma for 2 pixles rgb/bgr unless user wants full chroma interpolation
2292 if ((isBGR(dstFormat) || isRGB(dstFormat)) && !(flags&SWS_FULL_CHR_H_INT)) c->chrDstHSubSample=1;
2293
2294 // drop some chroma lines if the user wants it
2295 c->vChrDrop= (flags&SWS_SRC_V_CHR_DROP_MASK)>>SWS_SRC_V_CHR_DROP_SHIFT;
2296 c->chrSrcVSubSample+= c->vChrDrop;
2297
2298 // drop every 2. pixel for chroma calculation unless user wants full chroma
2299 if ((isBGR(srcFormat) || isRGB(srcFormat)) && !(flags&SWS_FULL_CHR_H_INP)
2300 && srcFormat!=PIX_FMT_RGB8 && srcFormat!=PIX_FMT_BGR8
2301 && srcFormat!=PIX_FMT_RGB4 && srcFormat!=PIX_FMT_BGR4
2302 && srcFormat!=PIX_FMT_RGB4_BYTE && srcFormat!=PIX_FMT_BGR4_BYTE
2303 && ((dstW>>c->chrDstHSubSample) <= (srcW>>1) || (flags&(SWS_FAST_BILINEAR|SWS_POINT))))
2304 c->chrSrcHSubSample=1;
2305
2306 if (param){
2307 c->param[0] = param[0];
2308 c->param[1] = param[1];
2309 }else{
2310 c->param[0] =
2311 c->param[1] = SWS_PARAM_DEFAULT;
2312 }
2313
2314 c->chrIntHSubSample= c->chrDstHSubSample;
2315 c->chrIntVSubSample= c->chrSrcVSubSample;
2316
2317 // Note the -((-x)>>y) is so that we always round toward +inf.
2318 c->chrSrcW= -((-srcW) >> c->chrSrcHSubSample);
2319 c->chrSrcH= -((-srcH) >> c->chrSrcVSubSample);
2320 c->chrDstW= -((-dstW) >> c->chrDstHSubSample);
2321 c->chrDstH= -((-dstH) >> c->chrDstVSubSample);
2322
2323 sws_setColorspaceDetails(c, Inverse_Table_6_9[SWS_CS_DEFAULT], srcRange, Inverse_Table_6_9[SWS_CS_DEFAULT] /* FIXME*/, dstRange, 0, 1<<16, 1<<16);
2324
2325 /* unscaled special Cases */
2326 if (unscaled && !usesHFilter && !usesVFilter && (srcRange == dstRange || isBGR(dstFormat) || isRGB(dstFormat)))
2327 {
2328 /* yv12_to_nv12 */
2329 if (srcFormat == PIX_FMT_YUV420P && (dstFormat == PIX_FMT_NV12 || dstFormat == PIX_FMT_NV21))
2330 {
2331 c->swScale= PlanarToNV12Wrapper;
2332 }
2333 #if CONFIG_GPL
2334 /* yuv2bgr */
2335 if ((srcFormat==PIX_FMT_YUV420P || srcFormat==PIX_FMT_YUV422P) && (isBGR(dstFormat) || isRGB(dstFormat))
2336 && !(flags & SWS_ACCURATE_RND) && !(dstH&1))
2337 {
2338 c->swScale= yuv2rgb_get_func_ptr(c);
2339 }
2340 #endif
2341
2342 if (srcFormat==PIX_FMT_YUV410P && dstFormat==PIX_FMT_YUV420P && !(flags & SWS_BITEXACT))
2343 {
2344 c->swScale= yvu9toyv12Wrapper;
2345 }
2346
2347 /* bgr24toYV12 */
2348 if (srcFormat==PIX_FMT_BGR24 && dstFormat==PIX_FMT_YUV420P && !(flags & SWS_ACCURATE_RND))
2349 c->swScale= bgr24toyv12Wrapper;
2350
2351 /* rgb/bgr -> rgb/bgr (no dither needed forms) */
2352 if ( (isBGR(srcFormat) || isRGB(srcFormat))
2353 && (isBGR(dstFormat) || isRGB(dstFormat))
2354 && srcFormat != PIX_FMT_BGR8 && dstFormat != PIX_FMT_BGR8
2355 && srcFormat != PIX_FMT_RGB8 && dstFormat != PIX_FMT_RGB8
2356 && srcFormat != PIX_FMT_BGR4 && dstFormat != PIX_FMT_BGR4
2357 && srcFormat != PIX_FMT_RGB4 && dstFormat != PIX_FMT_RGB4
2358 && srcFormat != PIX_FMT_BGR4_BYTE && dstFormat != PIX_FMT_BGR4_BYTE
2359 && srcFormat != PIX_FMT_RGB4_BYTE && dstFormat != PIX_FMT_RGB4_BYTE
2360 && srcFormat != PIX_FMT_MONOBLACK && dstFormat != PIX_FMT_MONOBLACK
2361 && srcFormat != PIX_FMT_MONOWHITE && dstFormat != PIX_FMT_MONOWHITE
2362 && dstFormat != PIX_FMT_RGB32_1
2363 && dstFormat != PIX_FMT_BGR32_1
2364 && (!needsDither || (c->flags&(SWS_FAST_BILINEAR|SWS_POINT))))
2365 c->swScale= rgb2rgbWrapper;
2366
2367 if ((usePal(srcFormat) && (
2368 dstFormat == PIX_FMT_RGB32 ||
2369 dstFormat == PIX_FMT_RGB32_1 ||
2370 dstFormat == PIX_FMT_RGB24 ||
2371 dstFormat == PIX_FMT_BGR32 ||
2372 dstFormat == PIX_FMT_BGR32_1 ||
2373 dstFormat == PIX_FMT_BGR24)))
2374 c->swScale= pal2rgbWrapper;
2375
2376 if (srcFormat == PIX_FMT_YUV422P)
2377 {
2378 if (dstFormat == PIX_FMT_YUYV422)
2379 c->swScale= YUV422PToYuy2Wrapper;
2380 else if (dstFormat == PIX_FMT_UYVY422)
2381 c->swScale= YUV422PToUyvyWrapper;
2382 }
2383
2384 /* LQ converters if -sws 0 or -sws 4*/
2385 if (c->flags&(SWS_FAST_BILINEAR|SWS_POINT)){
2386 /* yv12_to_yuy2 */
2387 if (srcFormat == PIX_FMT_YUV420P)
2388 {
2389 if (dstFormat == PIX_FMT_YUYV422)
2390 c->swScale= PlanarToYuy2Wrapper;
2391 else if (dstFormat == PIX_FMT_UYVY422)
2392 c->swScale= PlanarToUyvyWrapper;
2393 }
2394 }
2395
2396 #ifdef COMPILE_ALTIVEC
2397 if ((c->flags & SWS_CPU_CAPS_ALTIVEC) &&
2398 srcFormat == PIX_FMT_YUV420P) {
2399 // unscaled YV12 -> packed YUV, we want speed
2400 if (dstFormat == PIX_FMT_YUYV422)
2401 c->swScale= yv12toyuy2_unscaled_altivec;
2402 else if (dstFormat == PIX_FMT_UYVY422)
2403 c->swScale= yv12touyvy_unscaled_altivec;
2404 }
2405 #endif
2406
2407 /* simple copy */
2408 if ( srcFormat == dstFormat
2409 || (isPlanarYUV(srcFormat) && isGray(dstFormat))
2410 || (isPlanarYUV(dstFormat) && isGray(srcFormat)))
2411 {
2412 if (isPacked(c->srcFormat))
2413 c->swScale= packedCopy;
2414 else /* Planar YUV or gray */
2415 c->swScale= planarCopy;
2416 }
2417
2418 /* gray16{le,be} conversions */
2419 if (isGray16(srcFormat) && (isPlanarYUV(dstFormat) || (dstFormat == PIX_FMT_GRAY8)))
2420 {
2421 c->swScale= gray16togray;
2422 }
2423 if ((isPlanarYUV(srcFormat) || (srcFormat == PIX_FMT_GRAY8)) && isGray16(dstFormat))
2424 {
2425 c->swScale= graytogray16;
2426 }
2427 if (srcFormat != dstFormat && isGray16(srcFormat) && isGray16(dstFormat))
2428 {
2429 c->swScale= gray16swap;
2430 }
2431
2432 #if ARCH_BFIN
2433 if (flags & SWS_CPU_CAPS_BFIN)
2434 ff_bfin_get_unscaled_swscale (c);
2435 #endif
2436
2437 if (c->swScale){
2438 if (flags&SWS_PRINT_INFO)
2439 av_log(c, AV_LOG_INFO, "using unscaled %s -> %s special converter\n",
2440 sws_format_name(srcFormat), sws_format_name(dstFormat));
2441 return c;
2442 }
2443 }
2444
2445 if (flags & SWS_CPU_CAPS_MMX2)
2446 {
2447 c->canMMX2BeUsed= (dstW >=srcW && (dstW&31)==0 && (srcW&15)==0) ? 1 : 0;
2448 if (!c->canMMX2BeUsed && dstW >=srcW && (srcW&15)==0 && (flags&SWS_FAST_BILINEAR))
2449 {
2450 if (flags&SWS_PRINT_INFO)
2451 av_log(c, AV_LOG_INFO, "output Width is not a multiple of 32 -> no MMX2 scaler\n");
2452 }
2453 if (usesHFilter) c->canMMX2BeUsed=0;
2454 }
2455 else
2456 c->canMMX2BeUsed=0;
2457
2458 c->chrXInc= ((c->chrSrcW<<16) + (c->chrDstW>>1))/c->chrDstW;
2459 c->chrYInc= ((c->chrSrcH<<16) + (c->chrDstH>>1))/c->chrDstH;
2460
2461 // match pixel 0 of the src to pixel 0 of dst and match pixel n-2 of src to pixel n-2 of dst
2462 // but only for the FAST_BILINEAR mode otherwise do correct scaling
2463 // n-2 is the last chrominance sample available
2464 // this is not perfect, but no one should notice the difference, the more correct variant
2465 // would be like the vertical one, but that would require some special code for the
2466 // first and last pixel
2467 if (flags&SWS_FAST_BILINEAR)
2468 {
2469 if (c->canMMX2BeUsed)
2470 {
2471 c->lumXInc+= 20;
2472 c->chrXInc+= 20;
2473 }
2474 //we don't use the x86asm scaler if mmx is available
2475 else if (flags & SWS_CPU_CAPS_MMX)
2476 {
2477 c->lumXInc = ((srcW-2)<<16)/(dstW-2) - 20;
2478 c->chrXInc = ((c->chrSrcW-2)<<16)/(c->chrDstW-2) - 20;
2479 }
2480 }
2481
2482 /* precalculate horizontal scaler filter coefficients */
2483 {
2484 const int filterAlign=
2485 (flags & SWS_CPU_CAPS_MMX) ? 4 :
2486 (flags & SWS_CPU_CAPS_ALTIVEC) ? 8 :
2487 1;
2488
2489 initFilter(&c->hLumFilter, &c->hLumFilterPos, &c->hLumFilterSize, c->lumXInc,
2490 srcW , dstW, filterAlign, 1<<14,
2491 (flags&SWS_BICUBLIN) ? (flags|SWS_BICUBIC) : flags,
2492 srcFilter->lumH, dstFilter->lumH, c->param);
2493 initFilter(&c->hChrFilter, &c->hChrFilterPos, &c->hChrFilterSize, c->chrXInc,
2494 c->chrSrcW, c->chrDstW, filterAlign, 1<<14,
2495 (flags&SWS_BICUBLIN) ? (flags|SWS_BILINEAR) : flags,
2496 srcFilter->chrH, dstFilter->chrH, c->param);
2497
2498 #define MAX_FUNNY_CODE_SIZE 10000
2499 #if defined(COMPILE_MMX2)
2500 // can't downscale !!!
2501 if (c->canMMX2BeUsed && (flags & SWS_FAST_BILINEAR))
2502 {
2503 #ifdef MAP_ANONYMOUS
2504 c->funnyYCode = (uint8_t*)mmap(NULL, MAX_FUNNY_CODE_SIZE, PROT_EXEC | PROT_READ | PROT_WRITE, MAP_PRIVATE | MAP_ANONYMOUS, 0, 0);
2505 c->funnyUVCode = (uint8_t*)mmap(NULL, MAX_FUNNY_CODE_SIZE, PROT_EXEC | PROT_READ | PROT_WRITE, MAP_PRIVATE | MAP_ANONYMOUS, 0, 0);
2506 #else
2507 c->funnyYCode = av_malloc(MAX_FUNNY_CODE_SIZE);
2508 c->funnyUVCode = av_malloc(MAX_FUNNY_CODE_SIZE);
2509 #endif
2510
2511 c->lumMmx2Filter = av_malloc((dstW /8+8)*sizeof(int16_t));
2512 c->chrMmx2Filter = av_malloc((c->chrDstW /4+8)*sizeof(int16_t));
2513 c->lumMmx2FilterPos= av_malloc((dstW /2/8+8)*sizeof(int32_t));
2514 c->chrMmx2FilterPos= av_malloc((c->chrDstW/2/4+8)*sizeof(int32_t));
2515
2516 initMMX2HScaler( dstW, c->lumXInc, c->funnyYCode , c->lumMmx2Filter, c->lumMmx2FilterPos, 8);
2517 initMMX2HScaler(c->chrDstW, c->chrXInc, c->funnyUVCode, c->chrMmx2Filter, c->chrMmx2FilterPos, 4);
2518 }
2519 #endif /* defined(COMPILE_MMX2) */
2520 } // Init Horizontal stuff
2521
2522
2523
2524 /* precalculate vertical scaler filter coefficients */
2525 {
2526 const int filterAlign=
2527 (flags & SWS_CPU_CAPS_MMX) && (flags & SWS_ACCURATE_RND) ? 2 :
2528 (flags & SWS_CPU_CAPS_ALTIVEC) ? 8 :
2529 1;
2530
2531 initFilter(&c->vLumFilter, &c->vLumFilterPos, &c->vLumFilterSize, c->lumYInc,
2532 srcH , dstH, filterAlign, (1<<12),
2533 (flags&SWS_BICUBLIN) ? (flags|SWS_BICUBIC) : flags,
2534 srcFilter->lumV, dstFilter->lumV, c->param);
2535 initFilter(&c->vChrFilter, &c->vChrFilterPos, &c->vChrFilterSize, c->chrYInc,
2536 c->chrSrcH, c->chrDstH, filterAlign, (1<<12),
2537 (flags&SWS_BICUBLIN) ? (flags|SWS_BILINEAR) : flags,
2538 srcFilter->chrV, dstFilter->chrV, c->param);
2539
2540 #if HAVE_ALTIVEC
2541 c->vYCoeffsBank = av_malloc(sizeof (vector signed short)*c->vLumFilterSize*c->dstH);
2542 c->vCCoeffsBank = av_malloc(sizeof (vector signed short)*c->vChrFilterSize*c->chrDstH);
2543
2544 for (i=0;i<c->vLumFilterSize*c->dstH;i++) {
2545 int j;
2546 short *p = (short *)&c->vYCoeffsBank[i];
2547 for (j=0;j<8;j++)
2548 p[j] = c->vLumFilter[i];
2549 }
2550
2551 for (i=0;i<c->vChrFilterSize*c->chrDstH;i++) {
2552 int j;
2553 short *p = (short *)&c->vCCoeffsBank[i];
2554 for (j=0;j<8;j++)
2555 p[j] = c->vChrFilter[i];
2556 }
2557 #endif
2558 }
2559
2560 // Calculate Buffer Sizes so that they won't run out while handling these damn slices
2561 c->vLumBufSize= c->vLumFilterSize;
2562 c->vChrBufSize= c->vChrFilterSize;
2563 for (i=0; i<dstH; i++)
2564 {
2565 int chrI= i*c->chrDstH / dstH;
2566 int nextSlice= FFMAX(c->vLumFilterPos[i ] + c->vLumFilterSize - 1,
2567 ((c->vChrFilterPos[chrI] + c->vChrFilterSize - 1)<<c->chrSrcVSubSample));
2568
2569 nextSlice>>= c->chrSrcVSubSample;
2570 nextSlice<<= c->chrSrcVSubSample;
2571 if (c->vLumFilterPos[i ] + c->vLumBufSize < nextSlice)
2572 c->vLumBufSize= nextSlice - c->vLumFilterPos[i];
2573 if (c->vChrFilterPos[chrI] + c->vChrBufSize < (nextSlice>>c->chrSrcVSubSample))
2574 c->vChrBufSize= (nextSlice>>c->chrSrcVSubSample) - c->vChrFilterPos[chrI];
2575 }
2576
2577 // allocate pixbufs (we use dynamic allocation because otherwise we would need to
2578 c->lumPixBuf= av_malloc(c->vLumBufSize*2*sizeof(int16_t*));
2579 c->chrPixBuf= av_malloc(c->vChrBufSize*2*sizeof(int16_t*));
2580 //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)
2581 /* align at 16 bytes for AltiVec */
2582 for (i=0; i<c->vLumBufSize; i++)
2583 c->lumPixBuf[i]= c->lumPixBuf[i+c->vLumBufSize]= av_mallocz(VOF+1);
2584 for (i=0; i<c->vChrBufSize; i++)
2585 c->chrPixBuf[i]= c->chrPixBuf[i+c->vChrBufSize]= av_malloc((VOF+1)*2);
2586
2587 //try to avoid drawing green stuff between the right end and the stride end
2588 for (i=0; i<c->vChrBufSize; i++) memset(c->chrPixBuf[i], 64, (VOF+1)*2);
2589
2590 assert(2*VOFW == VOF);
2591
2592 assert(c->chrDstH <= dstH);
2593
2594 if (flags&SWS_PRINT_INFO)
2595 {
2596 #ifdef DITHER1XBPP
2597 const char *dither= " dithered";
2598 #else
2599 const char *dither= "";
2600 #endif
2601 if (flags&SWS_FAST_BILINEAR)
2602 av_log(c, AV_LOG_INFO, "FAST_BILINEAR scaler, ");
2603 else if (flags&SWS_BILINEAR)
2604 av_log(c, AV_LOG_INFO, "BILINEAR scaler, ");
2605 else if (flags&SWS_BICUBIC)
2606 av_log(c, AV_LOG_INFO, "BICUBIC scaler, ");
2607 else if (flags&SWS_X)
2608 av_log(c, AV_LOG_INFO, "Experimental scaler, ");
2609 else if (flags&SWS_POINT)
2610 av_log(c, AV_LOG_INFO, "Nearest Neighbor / POINT scaler, ");
2611 else if (flags&SWS_AREA)
2612 av_log(c, AV_LOG_INFO, "Area Averageing scaler, ");
2613 else if (flags&SWS_BICUBLIN)
2614 av_log(c, AV_LOG_INFO, "luma BICUBIC / chroma BILINEAR scaler, ");
2615 else if (flags&SWS_GAUSS)
2616 av_log(c, AV_LOG_INFO, "Gaussian scaler, ");
2617 else if (flags&SWS_SINC)
2618 av_log(c, AV_LOG_INFO, "Sinc scaler, ");
2619 else if (flags&SWS_LANCZOS)
2620 av_log(c, AV_LOG_INFO, "Lanczos scaler, ");
2621 else if (flags&SWS_SPLINE)
2622 av_log(c, AV_LOG_INFO, "Bicubic spline scaler, ");
2623 else
2624 av_log(c, AV_LOG_INFO, "ehh flags invalid?! ");
2625
2626 if (dstFormat==PIX_FMT_BGR555 || dstFormat==PIX_FMT_BGR565)
2627 av_log(c, AV_LOG_INFO, "from %s to%s %s ",
2628 sws_format_name(srcFormat), dither, sws_format_name(dstFormat));
2629 else
2630 av_log(c, AV_LOG_INFO, "from %s to %s ",
2631 sws_format_name(srcFormat), sws_format_name(dstFormat));
2632
2633 if (flags & SWS_CPU_CAPS_MMX2)
2634 av_log(c, AV_LOG_INFO, "using MMX2\n");
2635 else if (flags & SWS_CPU_CAPS_3DNOW)
2636 av_log(c, AV_LOG_INFO, "using 3DNOW\n");
2637 else if (flags & SWS_CPU_CAPS_MMX)
2638 av_log(c, AV_LOG_INFO, "using MMX\n");
2639 else if (flags & SWS_CPU_CAPS_ALTIVEC)
2640 av_log(c, AV_LOG_INFO, "using AltiVec\n");
2641 else
2642 av_log(c, AV_LOG_INFO, "using C\n");
2643 }
2644
2645 if (flags & SWS_PRINT_INFO)
2646 {
2647 if (flags & SWS_CPU_CAPS_MMX)
2648 {
2649 if (c->canMMX2BeUsed && (flags&SWS_FAST_BILINEAR))
2650 av_log(c, AV_LOG_VERBOSE, "using FAST_BILINEAR MMX2 scaler for horizontal scaling\n");
2651 else
2652 {
2653 if (c->hLumFilterSize==4)
2654 av_log(c, AV_LOG_VERBOSE, "using 4-tap MMX scaler for horizontal luminance scaling\n");
2655 else if (c->hLumFilterSize==8)
2656 av_log(c, AV_LOG_VERBOSE, "using 8-tap MMX scaler for horizontal luminance scaling\n");
2657 else
2658 av_log(c, AV_LOG_VERBOSE, "using n-tap MMX scaler for horizontal luminance scaling\n");
2659
2660 if (c->hChrFilterSize==4)
2661 av_log(c, AV_LOG_VERBOSE, "using 4-tap MMX scaler for horizontal chrominance scaling\n");
2662 else if (c->hChrFilterSize==8)
2663 av_log(c, AV_LOG_VERBOSE, "using 8-tap MMX scaler for horizontal chrominance scaling\n");
2664 else
2665 av_log(c, AV_LOG_VERBOSE, "using n-tap MMX scaler for horizontal chrominance scaling\n");
2666 }
2667 }
2668 else
2669 {
2670 #if ARCH_X86
2671 av_log(c, AV_LOG_VERBOSE, "using X86-Asm scaler for horizontal scaling\n");
2672 #else
2673 if (flags & SWS_FAST_BILINEAR)
2674 av_log(c, AV_LOG_VERBOSE, "using FAST_BILINEAR C scaler for horizontal scaling\n");
2675 else
2676 av_log(c, AV_LOG_VERBOSE, "using C scaler for horizontal scaling\n");
2677 #endif
2678 }
2679 if (isPlanarYUV(dstFormat))
2680 {
2681 if (c->vLumFilterSize==1)
2682 av_log(c, AV_LOG_VERBOSE, "using 1-tap %s \"scaler\" for vertical scaling (YV12 like)\n", (flags & SWS_CPU_CAPS_MMX) ? "MMX" : "C");
2683 else
2684 av_log(c, AV_LOG_VERBOSE, "using n-tap %s scaler for vertical scaling (YV12 like)\n", (flags & SWS_CPU_CAPS_MMX) ? "MMX" : "C");
2685 }
2686 else
2687 {
2688 if (c->vLumFilterSize==1 && c->vChrFilterSize==2)
2689 av_log(c, AV_LOG_VERBOSE, "using 1-tap %s \"scaler\" for vertical luminance scaling (BGR)\n"
2690 " 2-tap scaler for vertical chrominance scaling (BGR)\n", (flags & SWS_CPU_CAPS_MMX) ? "MMX" : "C");
2691 else if (c->vLumFilterSize==2 && c->vChrFilterSize==2)
2692 av_log(c, AV_LOG_VERBOSE, "using 2-tap linear %s scaler for vertical scaling (BGR)\n", (flags & SWS_CPU_CAPS_MMX) ? "MMX" : "C");
2693 else
2694 av_log(c, AV_LOG_VERBOSE, "using n-tap %s scaler for vertical scaling (BGR)\n", (flags & SWS_CPU_CAPS_MMX) ? "MMX" : "C");
2695 }
2696
2697 if (dstFormat==PIX_FMT_BGR24)
2698 av_log(c, AV_LOG_VERBOSE, "using %s YV12->BGR24 Converter\n",
2699 (flags & SWS_CPU_CAPS_MMX2) ? "MMX2" : ((flags & SWS_CPU_CAPS_MMX) ? "MMX" : "C"));
2700 else if (dstFormat==PIX_FMT_RGB32)
2701 av_log(c, AV_LOG_VERBOSE, "using %s YV12->BGR32 Converter\n", (flags & SWS_CPU_CAPS_MMX) ? "MMX" : "C");
2702 else if (dstFormat==PIX_FMT_BGR565)
2703 av_log(c, AV_LOG_VERBOSE, "using %s YV12->BGR16 Converter\n", (flags & SWS_CPU_CAPS_MMX) ? "MMX" : "C");
2704 else if (dstFormat==PIX_FMT_BGR555)
2705 av_log(c, AV_LOG_VERBOSE, "using %s YV12->BGR15 Converter\n", (flags & SWS_CPU_CAPS_MMX) ? "MMX" : "C");
2706
2707 av_log(c, AV_LOG_VERBOSE, "%dx%d -> %dx%d\n", srcW, srcH, dstW, dstH);
2708 }
2709 if (flags & SWS_PRINT_INFO)
2710 {
2711 av_log(c, AV_LOG_DEBUG, "Lum srcW=%d srcH=%d dstW=%d dstH=%d xInc=%d yInc=%d\n",
2712 c->srcW, c->srcH, c->dstW, c->dstH, c->lumXInc, c->lumYInc);
2713 av_log(c, AV_LOG_DEBUG, "Chr srcW=%d srcH=%d dstW=%d dstH=%d xInc=%d yInc=%d\n",
2714 c->chrSrcW, c->chrSrcH, c->chrDstW, c->chrDstH, c->chrXInc, c->chrYInc);
2715 }
2716
2717 c->swScale= getSwsFunc(flags);
2718 return c;
2719 }
2720
2721 /**
2722 * swscale wrapper, so we don't need to export the SwsContext.
2723 * assumes planar YUV to be in YUV order instead of YVU
2724 */
2725 int sws_scale(SwsContext *c, uint8_t* src[], int srcStride[], int srcSliceY,
2726 int srcSliceH, uint8_t* dst[], int dstStride[]){
2727 int i;
2728 uint8_t* src2[4]= {src[0], src[1], src[2]};
2729
2730 if (c->sliceDir == 0 && srcSliceY != 0 && srcSliceY + srcSliceH != c->srcH) {
2731 av_log(c, AV_LOG_ERROR, "Slices start in the middle!\n");
2732 return 0;
2733 }
2734 if (c->sliceDir == 0) {
2735 if (srcSliceY == 0) c->sliceDir = 1; else c->sliceDir = -1;
2736 }
2737
2738 if (usePal(c->srcFormat)){
2739 for (i=0; i<256; i++){
2740 int p, r, g, b,y,u,v;
2741 if(c->srcFormat == PIX_FMT_PAL8){
2742 p=((uint32_t*)(src[1]))[i];
2743 r= (p>>16)&0xFF;
2744 g= (p>> 8)&0xFF;
2745 b= p &0xFF;
2746 }else if(c->srcFormat == PIX_FMT_RGB8){
2747 r= (i>>5 )*36;
2748 g= ((i>>2)&7)*36;
2749 b= (i&3 )*85;
2750 }else if(c->srcFormat == PIX_FMT_BGR8){
2751 b= (i>>6 )*85;
2752 g= ((i>>3)&7)*36;
2753 r= (i&7 )*36;
2754 }else if(c->srcFormat == PIX_FMT_RGB4_BYTE){
2755 r= (i>>3 )*255;
2756 g= ((i>>1)&3)*85;
2757 b= (i&1 )*255;
2758 }else {
2759 assert(c->srcFormat == PIX_FMT_BGR4_BYTE);
2760 b= (i>>3 )*255;
2761 g= ((i>>1)&3)*85;
2762 r= (i&1 )*255;
2763 }
2764 y= av_clip_uint8((RY*r + GY*g + BY*b + ( 33<<(RGB2YUV_SHIFT-1)))>>RGB2YUV_SHIFT);
2765 u= av_clip_uint8((RU*r + GU*g + BU*b + (257<<(RGB2YUV_SHIFT-1)))>>RGB2YUV_SHIFT);
2766 v= av_clip_uint8((RV*r + GV*g + BV*b + (257<<(RGB2YUV_SHIFT-1)))>>RGB2YUV_SHIFT);
2767 c->pal_yuv[i]= y + (u<<8) + (v<<16);
2768
2769
2770 switch(c->dstFormat) {
2771 case PIX_FMT_BGR32:
2772 #ifndef WORDS_BIGENDIAN
2773 case PIX_FMT_RGB24:
2774 #endif
2775 c->pal_rgb[i]= r + (g<<8) + (b<<16);
2776 break;
2777 case PIX_FMT_BGR32_1:
2778 #ifdef WORDS_BIGENDIAN
2779 case PIX_FMT_BGR24:
2780 #endif
2781 c->pal_rgb[i]= (r + (g<<8) + (b<<16)) << 8;
2782 break;
2783 case PIX_FMT_RGB32_1:
2784 #ifdef WORDS_BIGENDIAN
2785 case PIX_FMT_RGB24:
2786 #endif
2787 c->pal_rgb[i]= (b + (g<<8) + (r<<16)) << 8;
2788 break;
2789 case PIX_FMT_RGB32:
2790 #ifndef WORDS_BIGENDIAN
2791 case PIX_FMT_BGR24:
2792 #endif
2793 default:
2794 c->pal_rgb[i]= b + (g<<8) + (r<<16);
2795 }
2796 }
2797 }
2798
2799 // copy strides, so they can safely be modified
2800 if (c->sliceDir == 1) {
2801 // slices go from top to bottom
2802 int srcStride2[4]= {srcStride[0], srcStride[1], srcStride[2]};
2803 int dstStride2[4]= {dstStride[0], dstStride[1], dstStride[2]};
2804 return c->swScale(c, src2, srcStride2, srcSliceY, srcSliceH, dst, dstStride2);
2805 } else {
2806 // slices go from bottom to top => we flip the image internally
2807 uint8_t* dst2[4]= {dst[0] + (c->dstH-1)*dstStride[0],
2808 dst[1] + ((c->dstH>>c->chrDstVSubSample)-1)*dstStride[1],
2809 dst[2] + ((c->dstH>>c->chrDstVSubSample)-1)*dstStride[2]};
2810 int srcStride2[4]= {-srcStride[0], -srcStride[1], -srcStride[2]};
2811 int dstStride2[4]= {-dstStride[0], -dstStride[1], -dstStride[2]};
2812
2813 src2[0] += (srcSliceH-1)*srcStride[0];
2814 if (!usePal(c->srcFormat))
2815 src2[1] += ((srcSliceH>>c->chrSrcVSubSample)-1)*srcStride[1];
2816 src2[2] += ((srcSliceH>>c->chrSrcVSubSample)-1)*srcStride[2];
2817
2818 return c->swScale(c, src2, srcStride2, c->srcH-srcSliceY-srcSliceH, srcSliceH, dst2, dstStride2);
2819 }
2820 }
2821
2822 /**
2823 * swscale wrapper, so we don't need to export the SwsContext
2824 */
2825 int sws_scale_ordered(SwsContext *c, uint8_t* src[], int srcStride[], int srcSliceY,
2826 int srcSliceH, uint8_t* dst[], int dstStride[]){
2827 return sws_scale(c, src, srcStride, srcSliceY, srcSliceH, dst, dstStride);
2828 }
2829
2830 SwsFilter *sws_getDefaultFilter(float lumaGBlur, float chromaGBlur,
2831 float lumaSharpen, float chromaSharpen,
2832 float chromaHShift, float chromaVShift,
2833 int verbose)
2834 {
2835 SwsFilter *filter= av_malloc(sizeof(SwsFilter));
2836
2837 if (lumaGBlur!=0.0){
2838 filter->lumH= sws_getGaussianVec(lumaGBlur, 3.0);
2839 filter->lumV= sws_getGaussianVec(lumaGBlur, 3.0);
2840 }else{
2841 filter->lumH= sws_getIdentityVec();
2842 filter->lumV= sws_getIdentityVec();
2843 }
2844
2845 if (chromaGBlur!=0.0){
2846 filter->chrH= sws_getGaussianVec(chromaGBlur, 3.0);
2847 filter->chrV= sws_getGaussianVec(chromaGBlur, 3.0);
2848 }else{
2849 filter->chrH= sws_getIdentityVec();
2850 filter->chrV= sws_getIdentityVec();
2851 }
2852
2853 if (chromaSharpen!=0.0){
2854 SwsVector *id= sws_getIdentityVec();
2855 sws_scaleVec(filter->chrH, -chromaSharpen);
2856 sws_scaleVec(filter->chrV, -chromaSharpen);
2857 sws_addVec(filter->chrH, id);
2858 sws_addVec(filter->chrV, id);
2859 sws_freeVec(id);
2860 }
2861
2862 if (lumaSharpen!=0.0){
2863 SwsVector *id= sws_getIdentityVec();
2864 sws_scaleVec(filter->lumH, -lumaSharpen);
2865 sws_scaleVec(filter->lumV, -lumaSharpen);
2866 sws_addVec(filter->lumH, id);
2867 sws_addVec(filter->lumV, id);
2868 sws_freeVec(id);
2869 }
2870
2871 if (chromaHShift != 0.0)
2872 sws_shiftVec(filter->chrH, (int)(chromaHShift+0.5));
2873
2874 if (chromaVShift != 0.0)
2875 sws_shiftVec(filter->chrV, (int)(chromaVShift+0.5));
2876
2877 sws_normalizeVec(filter->chrH, 1.0);
2878 sws_normalizeVec(filter->chrV, 1.0);
2879 sws_normalizeVec(filter->lumH, 1.0);
2880 sws_normalizeVec(filter->lumV, 1.0);
2881
2882 if (verbose) sws_printVec(filter->chrH);
2883 if (verbose) sws_printVec(filter->lumH);
2884
2885 return filter;
2886 }
2887
2888 /**
2889 * returns a normalized gaussian curve used to filter stuff
2890 * quality=3 is high quality, lowwer is lowwer quality
2891 */
2892 SwsVector *sws_getGaussianVec(double variance, double quality){
2893 const int length= (int)(variance*quality + 0.5) | 1;
2894 int i;
2895 double *coeff= av_malloc(length*sizeof(double));
2896 double middle= (length-1)*0.5;
2897 SwsVector *vec= av_malloc(sizeof(SwsVector));
2898
2899 vec->coeff= coeff;
2900 vec->length= length;
2901
2902 for (i=0; i<length; i++)
2903 {
2904 double dist= i-middle;
2905 coeff[i]= exp(-dist*dist/(2*variance*variance)) / sqrt(2*variance*PI);
2906 }
2907
2908 sws_normalizeVec(vec, 1.0);
2909
2910 return vec;
2911 }
2912
2913 SwsVector *sws_getConstVec(double c, int length){
2914 int i;
2915 double *coeff= av_malloc(length*sizeof(double));
2916 SwsVector *vec= av_malloc(sizeof(SwsVector));
2917
2918 vec->coeff= coeff;
2919 vec->length= length;
2920
2921 for (i=0; i<length; i++)
2922 coeff[i]= c;
2923
2924 return vec;
2925 }
2926
2927
2928 SwsVector *sws_getIdentityVec(void){
2929 return sws_getConstVec(1.0, 1);
2930 }
2931
2932 double sws_dcVec(SwsVector *a){
2933 int i;
2934 double sum=0;
2935
2936 for (i=0; i<a->length; i++)
2937 sum+= a->coeff[i];
2938
2939 return sum;
2940 }
2941
2942 void sws_scaleVec(SwsVector *a, double scalar){
2943 int i;
2944
2945 for (i=0; i<a->length; i++)
2946 a->coeff[i]*= scalar;
2947 }
2948
2949 void sws_normalizeVec(SwsVector *a, double height){
2950 sws_scaleVec(a, height/sws_dcVec(a));
2951 }
2952
2953 static SwsVector *sws_getConvVec(SwsVector *a, SwsVector *b){
2954 int length= a->length + b->length - 1;
2955 double *coeff= av_malloc(length*sizeof(double));
2956 int i, j;
2957 SwsVector *vec= av_malloc(sizeof(SwsVector));
2958
2959 vec->coeff= coeff;
2960 vec->length= length;
2961
2962 for (i=0; i<length; i++) coeff[i]= 0.0;
2963
2964 for (i=0; i<a->length; i++)
2965 {
2966 for (j=0; j<b->length; j++)
2967 {
2968 coeff[i+j]+= a->coeff[i]*b->coeff[j];
2969 }
2970 }
2971
2972 return vec;
2973 }
2974
2975 static SwsVector *sws_sumVec(SwsVector *a, SwsVector *b){
2976 int length= FFMAX(a->length, b->length);
2977 double *coeff= av_malloc(length*sizeof(double));
2978 int i;
2979 SwsVector *vec= av_malloc(sizeof(SwsVector));
2980
2981 vec->coeff= coeff;
2982 vec->length= length;
2983
2984 for (i=0; i<length; i++) coeff[i]= 0.0;
2985
2986 for (<