Change semantic of CONFIG_*, HAVE_* and ARCH_*.
[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 */
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 default:
458 return "Unknown format";
459 }
460 }
461
462 static inline void yuv2yuvXinC(int16_t *lumFilter, int16_t **lumSrc, int lumFilterSize,
463 int16_t *chrFilter, int16_t **chrSrc, int chrFilterSize,
464 uint8_t *dest, uint8_t *uDest, uint8_t *vDest, int dstW, int chrDstW)
465 {
466 //FIXME Optimize (just quickly writen not opti..)
467 int i;
468 for (i=0; i<dstW; i++)
469 {
470 int val=1<<18;
471 int j;
472 for (j=0; j<lumFilterSize; j++)
473 val += lumSrc[j][i] * lumFilter[j];
474
475 dest[i]= av_clip_uint8(val>>19);
476 }
477
478 if (uDest)
479 for (i=0; i<chrDstW; i++)
480 {
481 int u=1<<18;
482 int v=1<<18;
483 int j;
484 for (j=0; j<chrFilterSize; j++)
485 {
486 u += chrSrc[j][i] * chrFilter[j];
487 v += chrSrc[j][i + VOFW] * chrFilter[j];
488 }
489
490 uDest[i]= av_clip_uint8(u>>19);
491 vDest[i]= av_clip_uint8(v>>19);
492 }
493 }
494
495 static inline void yuv2nv12XinC(int16_t *lumFilter, int16_t **lumSrc, int lumFilterSize,
496 int16_t *chrFilter, int16_t **chrSrc, int chrFilterSize,
497 uint8_t *dest, uint8_t *uDest, int dstW, int chrDstW, int dstFormat)
498 {
499 //FIXME Optimize (just quickly writen not opti..)
500 int i;
501 for (i=0; i<dstW; i++)
502 {
503 int val=1<<18;
504 int j;
505 for (j=0; j<lumFilterSize; j++)
506 val += lumSrc[j][i] * lumFilter[j];
507
508 dest[i]= av_clip_uint8(val>>19);
509 }
510
511 if (!uDest)
512 return;
513
514 if (dstFormat == PIX_FMT_NV12)
515 for (i=0; i<chrDstW; i++)
516 {
517 int u=1<<18;
518 int v=1<<18;
519 int j;
520 for (j=0; j<chrFilterSize; j++)
521 {
522 u += chrSrc[j][i] * chrFilter[j];
523 v += chrSrc[j][i + VOFW] * chrFilter[j];
524 }
525
526 uDest[2*i]= av_clip_uint8(u>>19);
527 uDest[2*i+1]= av_clip_uint8(v>>19);
528 }
529 else
530 for (i=0; i<chrDstW; i++)
531 {
532 int u=1<<18;
533 int v=1<<18;
534 int j;
535 for (j=0; j<chrFilterSize; j++)
536 {
537 u += chrSrc[j][i] * chrFilter[j];
538 v += chrSrc[j][i + VOFW] * chrFilter[j];
539 }
540
541 uDest[2*i]= av_clip_uint8(v>>19);
542 uDest[2*i+1]= av_clip_uint8(u>>19);
543 }
544 }
545
546 #define YSCALE_YUV_2_PACKEDX_NOCLIP_C(type) \
547 for (i=0; i<(dstW>>1); i++){\
548 int j;\
549 int Y1 = 1<<18;\
550 int Y2 = 1<<18;\
551 int U = 1<<18;\
552 int V = 1<<18;\
553 type av_unused *r, *b, *g;\
554 const int i2= 2*i;\
555 \
556 for (j=0; j<lumFilterSize; j++)\
557 {\
558 Y1 += lumSrc[j][i2] * lumFilter[j];\
559 Y2 += lumSrc[j][i2+1] * lumFilter[j];\
560 }\
561 for (j=0; j<chrFilterSize; j++)\
562 {\
563 U += chrSrc[j][i] * chrFilter[j];\
564 V += chrSrc[j][i+VOFW] * chrFilter[j];\
565 }\
566 Y1>>=19;\
567 Y2>>=19;\
568 U >>=19;\
569 V >>=19;\
570
571 #define YSCALE_YUV_2_PACKEDX_C(type) \
572 YSCALE_YUV_2_PACKEDX_NOCLIP_C(type)\
573 if ((Y1|Y2|U|V)&256)\
574 {\
575 if (Y1>255) Y1=255; \
576 else if (Y1<0)Y1=0; \
577 if (Y2>255) Y2=255; \
578 else if (Y2<0)Y2=0; \
579 if (U>255) U=255; \
580 else if (U<0) U=0; \
581 if (V>255) V=255; \
582 else if (V<0) V=0; \
583 }
584
585 #define YSCALE_YUV_2_PACKEDX_FULL_C \
586 for (i=0; i<dstW; i++){\
587 int j;\
588 int Y = 0;\
589 int U = -128<<19;\
590 int V = -128<<19;\
591 int R,G,B;\
592 \
593 for (j=0; j<lumFilterSize; j++){\
594 Y += lumSrc[j][i ] * lumFilter[j];\
595 }\
596 for (j=0; j<chrFilterSize; j++){\
597 U += chrSrc[j][i ] * chrFilter[j];\
598 V += chrSrc[j][i+VOFW] * chrFilter[j];\
599 }\
600 Y >>=10;\
601 U >>=10;\
602 V >>=10;\
603
604 #define YSCALE_YUV_2_RGBX_FULL_C(rnd) \
605 YSCALE_YUV_2_PACKEDX_FULL_C\
606 Y-= c->yuv2rgb_y_offset;\
607 Y*= c->yuv2rgb_y_coeff;\
608 Y+= rnd;\
609 R= Y + V*c->yuv2rgb_v2r_coeff;\
610 G= Y + V*c->yuv2rgb_v2g_coeff + U*c->yuv2rgb_u2g_coeff;\
611 B= Y + U*c->yuv2rgb_u2b_coeff;\
612 if ((R|G|B)&(0xC0000000)){\
613 if (R>=(256<<22)) R=(256<<22)-1; \
614 else if (R<0)R=0; \
615 if (G>=(256<<22)) G=(256<<22)-1; \
616 else if (G<0)G=0; \
617 if (B>=(256<<22)) B=(256<<22)-1; \
618 else if (B<0)B=0; \
619 }\
620
621
622 #define YSCALE_YUV_2_GRAY16_C \
623 for (i=0; i<(dstW>>1); i++){\
624 int j;\
625 int Y1 = 1<<18;\
626 int Y2 = 1<<18;\
627 int U = 1<<18;\
628 int V = 1<<18;\
629 \
630 const int i2= 2*i;\
631 \
632 for (j=0; j<lumFilterSize; j++)\
633 {\
634 Y1 += lumSrc[j][i2] * lumFilter[j];\
635 Y2 += lumSrc[j][i2+1] * lumFilter[j];\
636 }\
637 Y1>>=11;\
638 Y2>>=11;\
639 if ((Y1|Y2|U|V)&65536)\
640 {\
641 if (Y1>65535) Y1=65535; \
642 else if (Y1<0)Y1=0; \
643 if (Y2>65535) Y2=65535; \
644 else if (Y2<0)Y2=0; \
645 }
646
647 #define YSCALE_YUV_2_RGBX_C(type) \
648 YSCALE_YUV_2_PACKEDX_C(type) /* FIXME fix tables so that cliping is not needed and then use _NOCLIP*/\
649 r = (type *)c->table_rV[V]; \
650 g = (type *)(c->table_gU[U] + c->table_gV[V]); \
651 b = (type *)c->table_bU[U]; \
652
653 #define YSCALE_YUV_2_PACKED2_C \
654 for (i=0; i<(dstW>>1); i++){ \
655 const int i2= 2*i; \
656 int Y1= (buf0[i2 ]*yalpha1+buf1[i2 ]*yalpha)>>19; \
657 int Y2= (buf0[i2+1]*yalpha1+buf1[i2+1]*yalpha)>>19; \
658 int U= (uvbuf0[i ]*uvalpha1+uvbuf1[i ]*uvalpha)>>19; \
659 int V= (uvbuf0[i+VOFW]*uvalpha1+uvbuf1[i+VOFW]*uvalpha)>>19; \
660
661 #define YSCALE_YUV_2_GRAY16_2_C \
662 for (i=0; i<(dstW>>1); i++){ \
663 const int i2= 2*i; \
664 int Y1= (buf0[i2 ]*yalpha1+buf1[i2 ]*yalpha)>>11; \
665 int Y2= (buf0[i2+1]*yalpha1+buf1[i2+1]*yalpha)>>11; \
666
667 #define YSCALE_YUV_2_RGB2_C(type) \
668 YSCALE_YUV_2_PACKED2_C\
669 type *r, *b, *g;\
670 r = (type *)c->table_rV[V];\
671 g = (type *)(c->table_gU[U] + c->table_gV[V]);\
672 b = (type *)c->table_bU[U];\
673
674 #define YSCALE_YUV_2_PACKED1_C \
675 for (i=0; i<(dstW>>1); i++){\
676 const int i2= 2*i;\
677 int Y1= buf0[i2 ]>>7;\
678 int Y2= buf0[i2+1]>>7;\
679 int U= (uvbuf1[i ])>>7;\
680 int V= (uvbuf1[i+VOFW])>>7;\
681
682 #define YSCALE_YUV_2_GRAY16_1_C \
683 for (i=0; i<(dstW>>1); i++){\
684 const int i2= 2*i;\
685 int Y1= buf0[i2 ]<<1;\
686 int Y2= buf0[i2+1]<<1;\
687
688 #define YSCALE_YUV_2_RGB1_C(type) \
689 YSCALE_YUV_2_PACKED1_C\
690 type *r, *b, *g;\
691 r = (type *)c->table_rV[V];\
692 g = (type *)(c->table_gU[U] + c->table_gV[V]);\
693 b = (type *)c->table_bU[U];\
694
695 #define YSCALE_YUV_2_PACKED1B_C \
696 for (i=0; i<(dstW>>1); i++){\
697 const int i2= 2*i;\
698 int Y1= buf0[i2 ]>>7;\
699 int Y2= buf0[i2+1]>>7;\
700 int U= (uvbuf0[i ] + uvbuf1[i ])>>8;\
701 int V= (uvbuf0[i+VOFW] + uvbuf1[i+VOFW])>>8;\
702
703 #define YSCALE_YUV_2_RGB1B_C(type) \
704 YSCALE_YUV_2_PACKED1B_C\
705 type *r, *b, *g;\
706 r = (type *)c->table_rV[V];\
707 g = (type *)(c->table_gU[U] + c->table_gV[V]);\
708 b = (type *)c->table_bU[U];\
709
710 #define YSCALE_YUV_2_MONO2_C \
711 const uint8_t * const d128=dither_8x8_220[y&7];\
712 uint8_t *g= c->table_gU[128] + c->table_gV[128];\
713 for (i=0; i<dstW-7; i+=8){\
714 int acc;\
715 acc = g[((buf0[i ]*yalpha1+buf1[i ]*yalpha)>>19) + d128[0]];\
716 acc+= acc + g[((buf0[i+1]*yalpha1+buf1[i+1]*yalpha)>>19) + d128[1]];\
717 acc+= acc + g[((buf0[i+2]*yalpha1+buf1[i+2]*yalpha)>>19) + d128[2]];\
718 acc+= acc + g[((buf0[i+3]*yalpha1+buf1[i+3]*yalpha)>>19) + d128[3]];\
719 acc+= acc + g[((buf0[i+4]*yalpha1+buf1[i+4]*yalpha)>>19) + d128[4]];\
720 acc+= acc + g[((buf0[i+5]*yalpha1+buf1[i+5]*yalpha)>>19) + d128[5]];\
721 acc+= acc + g[((buf0[i+6]*yalpha1+buf1[i+6]*yalpha)>>19) + d128[6]];\
722 acc+= acc + g[((buf0[i+7]*yalpha1+buf1[i+7]*yalpha)>>19) + d128[7]];\
723 ((uint8_t*)dest)[0]= c->dstFormat == PIX_FMT_MONOBLACK ? acc : ~acc;\
724 dest++;\
725 }\
726
727
728 #define YSCALE_YUV_2_MONOX_C \
729 const uint8_t * const d128=dither_8x8_220[y&7];\
730 uint8_t *g= c->table_gU[128] + c->table_gV[128];\
731 int acc=0;\
732 for (i=0; i<dstW-1; i+=2){\
733 int j;\
734 int Y1=1<<18;\
735 int Y2=1<<18;\
736 \
737 for (j=0; j<lumFilterSize; j++)\
738 {\
739 Y1 += lumSrc[j][i] * lumFilter[j];\
740 Y2 += lumSrc[j][i+1] * lumFilter[j];\
741 }\
742 Y1>>=19;\
743 Y2>>=19;\
744 if ((Y1|Y2)&256)\
745 {\
746 if (Y1>255) Y1=255;\
747 else if (Y1<0)Y1=0;\
748 if (Y2>255) Y2=255;\
749 else if (Y2<0)Y2=0;\
750 }\
751 acc+= acc + g[Y1+d128[(i+0)&7]];\
752 acc+= acc + g[Y2+d128[(i+1)&7]];\
753 if ((i&7)==6){\
754 ((uint8_t*)dest)[0]= c->dstFormat == PIX_FMT_MONOBLACK ? acc : ~acc;\
755 dest++;\
756 }\
757 }
758
759
760 #define YSCALE_YUV_2_ANYRGB_C(func, func2, func_g16, func_monoblack)\
761 switch(c->dstFormat)\
762 {\
763 case PIX_FMT_RGB32:\
764 case PIX_FMT_BGR32:\
765 case PIX_FMT_RGB32_1:\
766 case PIX_FMT_BGR32_1:\
767 func(uint32_t)\
768 ((uint32_t*)dest)[i2+0]= r[Y1] + g[Y1] + b[Y1];\
769 ((uint32_t*)dest)[i2+1]= r[Y2] + g[Y2] + b[Y2];\
770 } \
771 break;\
772 case PIX_FMT_RGB24:\
773 func(uint8_t)\
774 ((uint8_t*)dest)[0]= r[Y1];\
775 ((uint8_t*)dest)[1]= g[Y1];\
776 ((uint8_t*)dest)[2]= b[Y1];\
777 ((uint8_t*)dest)[3]= r[Y2];\
778 ((uint8_t*)dest)[4]= g[Y2];\
779 ((uint8_t*)dest)[5]= b[Y2];\
780 dest+=6;\
781 }\
782 break;\
783 case PIX_FMT_BGR24:\
784 func(uint8_t)\
785 ((uint8_t*)dest)[0]= b[Y1];\
786 ((uint8_t*)dest)[1]= g[Y1];\
787 ((uint8_t*)dest)[2]= r[Y1];\
788 ((uint8_t*)dest)[3]= b[Y2];\
789 ((uint8_t*)dest)[4]= g[Y2];\
790 ((uint8_t*)dest)[5]= r[Y2];\
791 dest+=6;\
792 }\
793 break;\
794 case PIX_FMT_RGB565:\
795 case PIX_FMT_BGR565:\
796 {\
797 const int dr1= dither_2x2_8[y&1 ][0];\
798 const int dg1= dither_2x2_4[y&1 ][0];\
799 const int db1= dither_2x2_8[(y&1)^1][0];\
800 const int dr2= dither_2x2_8[y&1 ][1];\
801 const int dg2= dither_2x2_4[y&1 ][1];\
802 const int db2= dither_2x2_8[(y&1)^1][1];\
803 func(uint16_t)\
804 ((uint16_t*)dest)[i2+0]= r[Y1+dr1] + g[Y1+dg1] + b[Y1+db1];\
805 ((uint16_t*)dest)[i2+1]= r[Y2+dr2] + g[Y2+dg2] + b[Y2+db2];\
806 }\
807 }\
808 break;\
809 case PIX_FMT_RGB555:\
810 case PIX_FMT_BGR555:\
811 {\
812 const int dr1= dither_2x2_8[y&1 ][0];\
813 const int dg1= dither_2x2_8[y&1 ][1];\
814 const int db1= dither_2x2_8[(y&1)^1][0];\
815 const int dr2= dither_2x2_8[y&1 ][1];\
816 const int dg2= dither_2x2_8[y&1 ][0];\
817 const int db2= dither_2x2_8[(y&1)^1][1];\
818 func(uint16_t)\
819 ((uint16_t*)dest)[i2+0]= r[Y1+dr1] + g[Y1+dg1] + b[Y1+db1];\
820 ((uint16_t*)dest)[i2+1]= r[Y2+dr2] + g[Y2+dg2] + b[Y2+db2];\
821 }\
822 }\
823 break;\
824 case PIX_FMT_RGB8:\
825 case PIX_FMT_BGR8:\
826 {\
827 const uint8_t * const d64= dither_8x8_73[y&7];\
828 const uint8_t * const d32= dither_8x8_32[y&7];\
829 func(uint8_t)\
830 ((uint8_t*)dest)[i2+0]= r[Y1+d32[(i2+0)&7]] + g[Y1+d32[(i2+0)&7]] + b[Y1+d64[(i2+0)&7]];\
831 ((uint8_t*)dest)[i2+1]= r[Y2+d32[(i2+1)&7]] + g[Y2+d32[(i2+1)&7]] + b[Y2+d64[(i2+1)&7]];\
832 }\
833 }\
834 break;\
835 case PIX_FMT_RGB4:\
836 case PIX_FMT_BGR4:\
837 {\
838 const uint8_t * const d64= dither_8x8_73 [y&7];\
839 const uint8_t * const d128=dither_8x8_220[y&7];\
840 func(uint8_t)\
841 ((uint8_t*)dest)[i]= r[Y1+d128[(i2+0)&7]] + g[Y1+d64[(i2+0)&7]] + b[Y1+d128[(i2+0)&7]]\
842 + ((r[Y2+d128[(i2+1)&7]] + g[Y2+d64[(i2+1)&7]] + b[Y2+d128[(i2+1)&7]])<<4);\
843 }\
844 }\
845 break;\
846 case PIX_FMT_RGB4_BYTE:\
847 case PIX_FMT_BGR4_BYTE:\
848 {\
849 const uint8_t * const d64= dither_8x8_73 [y&7];\
850 const uint8_t * const d128=dither_8x8_220[y&7];\
851 func(uint8_t)\
852 ((uint8_t*)dest)[i2+0]= r[Y1+d128[(i2+0)&7]] + g[Y1+d64[(i2+0)&7]] + b[Y1+d128[(i2+0)&7]];\
853 ((uint8_t*)dest)[i2+1]= r[Y2+d128[(i2+1)&7]] + g[Y2+d64[(i2+1)&7]] + b[Y2+d128[(i2+1)&7]];\
854 }\
855 }\
856 break;\
857 case PIX_FMT_MONOBLACK:\
858 case PIX_FMT_MONOWHITE:\
859 {\
860 func_monoblack\
861 }\
862 break;\
863 case PIX_FMT_YUYV422:\
864 func2\
865 ((uint8_t*)dest)[2*i2+0]= Y1;\
866 ((uint8_t*)dest)[2*i2+1]= U;\
867 ((uint8_t*)dest)[2*i2+2]= Y2;\
868 ((uint8_t*)dest)[2*i2+3]= V;\
869 } \
870 break;\
871 case PIX_FMT_UYVY422:\
872 func2\
873 ((uint8_t*)dest)[2*i2+0]= U;\
874 ((uint8_t*)dest)[2*i2+1]= Y1;\
875 ((uint8_t*)dest)[2*i2+2]= V;\
876 ((uint8_t*)dest)[2*i2+3]= Y2;\
877 } \
878 break;\
879 case PIX_FMT_GRAY16BE:\
880 func_g16\
881 ((uint8_t*)dest)[2*i2+0]= Y1>>8;\
882 ((uint8_t*)dest)[2*i2+1]= Y1;\
883 ((uint8_t*)dest)[2*i2+2]= Y2>>8;\
884 ((uint8_t*)dest)[2*i2+3]= Y2;\
885 } \
886 break;\
887 case PIX_FMT_GRAY16LE:\
888 func_g16\
889 ((uint8_t*)dest)[2*i2+0]= Y1;\
890 ((uint8_t*)dest)[2*i2+1]= Y1>>8;\
891 ((uint8_t*)dest)[2*i2+2]= Y2;\
892 ((uint8_t*)dest)[2*i2+3]= Y2>>8;\
893 } \
894 break;\
895 }\
896
897
898 static inline void yuv2packedXinC(SwsContext *c, int16_t *lumFilter, int16_t **lumSrc, int lumFilterSize,
899 int16_t *chrFilter, int16_t **chrSrc, int chrFilterSize,
900 uint8_t *dest, int dstW, int y)
901 {
902 int i;
903 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)
904 }
905
906 static inline void yuv2rgbXinC_full(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 int step= fmt_depth(c->dstFormat)/8;
912 int aidx= 3;
913
914 switch(c->dstFormat){
915 case PIX_FMT_ARGB:
916 dest++;
917 aidx= 0;
918 case PIX_FMT_RGB24:
919 aidx--;
920 case PIX_FMT_RGBA:
921 YSCALE_YUV_2_RGBX_FULL_C(1<<21)
922 dest[aidx]= 0;
923 dest[0]= R>>22;
924 dest[1]= G>>22;
925 dest[2]= B>>22;
926 dest+= step;
927 }
928 break;
929 case PIX_FMT_ABGR:
930 dest++;
931 aidx= 0;
932 case PIX_FMT_BGR24:
933 aidx--;
934 case PIX_FMT_BGRA:
935 YSCALE_YUV_2_RGBX_FULL_C(1<<21)
936 dest[aidx]= 0;
937 dest[0]= B>>22;
938 dest[1]= G>>22;
939 dest[2]= R>>22;
940 dest+= step;
941 }
942 break;
943 default:
944 assert(0);
945 }
946 }
947
948 //Note: we have C, X86, MMX, MMX2, 3DNOW version therse no 3DNOW+MMX2 one
949 //Plain C versions
950 #if !HAVE_MMX || defined (RUNTIME_CPUDETECT) || !CONFIG_GPL
951 #define COMPILE_C
952 #endif
953
954 #if ARCH_PPC
955 #if (HAVE_ALTIVEC || defined (RUNTIME_CPUDETECT)) && CONFIG_GPL
956 #define COMPILE_ALTIVEC
957 #endif //HAVE_ALTIVEC
958 #endif //ARCH_PPC
959
960 #if ARCH_X86
961
962 #if ((HAVE_MMX && !HAVE_3DNOW && !HAVE_MMX2) || defined (RUNTIME_CPUDETECT)) && CONFIG_GPL
963 #define COMPILE_MMX
964 #endif
965
966 #if (HAVE_MMX2 || defined (RUNTIME_CPUDETECT)) && CONFIG_GPL
967 #define COMPILE_MMX2
968 #endif
969
970 #if ((HAVE_3DNOW && !HAVE_MMX2) || defined (RUNTIME_CPUDETECT)) && CONFIG_GPL
971 #define COMPILE_3DNOW
972 #endif
973 #endif //ARCH_X86 || ARCH_X86_64
974
975 #undef HAVE_MMX
976 #undef HAVE_MMX2
977 #undef HAVE_3DNOW
978 #undef HAVE_ALTIVEC
979 #define HAVE_MMX 0
980 #define HAVE_MMX2 0
981 #define HAVE_3DNOW 0
982 #define HAVE_ALTIVEC 0
983
984 #ifdef COMPILE_C
985 #define RENAME(a) a ## _C
986 #include "swscale_template.c"
987 #endif
988
989 #ifdef COMPILE_ALTIVEC
990 #undef RENAME
991 #undef HAVE_ALTIVEC
992 #define HAVE_ALTIVEC 1
993 #define RENAME(a) a ## _altivec
994 #include "swscale_template.c"
995 #endif
996
997 #if ARCH_X86
998
999 //X86 versions
1000 /*
1001 #undef RENAME
1002 #undef HAVE_MMX
1003 #undef HAVE_MMX2
1004 #undef HAVE_3DNOW
1005 #define ARCH_X86
1006 #define RENAME(a) a ## _X86
1007 #include "swscale_template.c"
1008 */
1009 //MMX versions
1010 #ifdef COMPILE_MMX
1011 #undef RENAME
1012 #undef HAVE_MMX
1013 #undef HAVE_MMX2
1014 #undef HAVE_3DNOW
1015 #define HAVE_MMX 1
1016 #define HAVE_MMX2 0
1017 #define HAVE_3DNOW 0
1018 #define RENAME(a) a ## _MMX
1019 #include "swscale_template.c"
1020 #endif
1021
1022 //MMX2 versions
1023 #ifdef COMPILE_MMX2
1024 #undef RENAME
1025 #undef HAVE_MMX
1026 #undef HAVE_MMX2
1027 #undef HAVE_3DNOW
1028 #define HAVE_MMX 1
1029 #define HAVE_MMX2 1
1030 #define HAVE_3DNOW 0
1031 #define RENAME(a) a ## _MMX2
1032 #include "swscale_template.c"
1033 #endif
1034
1035 //3DNOW versions
1036 #ifdef COMPILE_3DNOW
1037 #undef RENAME
1038 #undef HAVE_MMX
1039 #undef HAVE_MMX2
1040 #undef HAVE_3DNOW
1041 #define HAVE_MMX 1
1042 #define HAVE_MMX2 0
1043 #define HAVE_3DNOW 1
1044 #define RENAME(a) a ## _3DNow
1045 #include "swscale_template.c"
1046 #endif
1047
1048 #endif //ARCH_X86 || ARCH_X86_64
1049
1050 // minor note: the HAVE_xyz is messed up after that line so don't use it
1051
1052 static double getSplineCoeff(double a, double b, double c, double d, double dist)
1053 {
1054 // printf("%f %f %f %f %f\n", a,b,c,d,dist);
1055 if (dist<=1.0) return ((d*dist + c)*dist + b)*dist +a;
1056 else return getSplineCoeff( 0.0,
1057 b+ 2.0*c + 3.0*d,
1058 c + 3.0*d,
1059 -b- 3.0*c - 6.0*d,
1060 dist-1.0);
1061 }
1062
1063 static inline int initFilter(int16_t **outFilter, int16_t **filterPos, int *outFilterSize, int xInc,
1064 int srcW, int dstW, int filterAlign, int one, int flags,
1065 SwsVector *srcFilter, SwsVector *dstFilter, double param[2])
1066 {
1067 int i;
1068 int filterSize;
1069 int filter2Size;
1070 int minFilterSize;
1071 int64_t *filter=NULL;
1072 int64_t *filter2=NULL;
1073 const int64_t fone= 1LL<<54;
1074 int ret= -1;
1075 #if ARCH_X86
1076 if (flags & SWS_CPU_CAPS_MMX)
1077 __asm__ volatile("emms\n\t"::: "memory"); //FIXME this should not be required but it IS (even for non-MMX versions)
1078 #endif
1079
1080 // Note the +1 is for the MMXscaler which reads over the end
1081 *filterPos = av_malloc((dstW+1)*sizeof(int16_t));
1082
1083 if (FFABS(xInc - 0x10000) <10) // unscaled
1084 {
1085 int i;
1086 filterSize= 1;
1087 filter= av_mallocz(dstW*sizeof(*filter)*filterSize);
1088
1089 for (i=0; i<dstW; i++)
1090 {
1091 filter[i*filterSize]= fone;
1092 (*filterPos)[i]=i;
1093 }
1094
1095 }
1096 else if (flags&SWS_POINT) // lame looking point sampling mode
1097 {
1098 int i;
1099 int xDstInSrc;
1100 filterSize= 1;
1101 filter= av_malloc(dstW*sizeof(*filter)*filterSize);
1102
1103 xDstInSrc= xInc/2 - 0x8000;
1104 for (i=0; i<dstW; i++)
1105 {
1106 int xx= (xDstInSrc - ((filterSize-1)<<15) + (1<<15))>>16;
1107
1108 (*filterPos)[i]= xx;
1109 filter[i]= fone;
1110 xDstInSrc+= xInc;
1111 }
1112 }
1113 else if ((xInc <= (1<<16) && (flags&SWS_AREA)) || (flags&SWS_FAST_BILINEAR)) // bilinear upscale
1114 {
1115 int i;
1116 int xDstInSrc;
1117 if (flags&SWS_BICUBIC) filterSize= 4;
1118 else if (flags&SWS_X ) filterSize= 4;
1119 else filterSize= 2; // SWS_BILINEAR / SWS_AREA
1120 filter= av_malloc(dstW*sizeof(*filter)*filterSize);
1121
1122 xDstInSrc= xInc/2 - 0x8000;
1123 for (i=0; i<dstW; i++)
1124 {
1125 int xx= (xDstInSrc - ((filterSize-1)<<15) + (1<<15))>>16;
1126 int j;
1127
1128 (*filterPos)[i]= xx;
1129 //Bilinear upscale / linear interpolate / Area averaging
1130 for (j=0; j<filterSize; j++)
1131 {
1132 int64_t coeff= fone - FFABS((xx<<16) - xDstInSrc)*(fone>>16);
1133 if (coeff<0) coeff=0;
1134 filter[i*filterSize + j]= coeff;
1135 xx++;
1136 }
1137 xDstInSrc+= xInc;
1138 }
1139 }
1140 else
1141 {
1142 int xDstInSrc;
1143 int sizeFactor;
1144
1145 if (flags&SWS_BICUBIC) sizeFactor= 4;
1146 else if (flags&SWS_X) sizeFactor= 8;
1147 else if (flags&SWS_AREA) sizeFactor= 1; //downscale only, for upscale it is bilinear
1148 else if (flags&SWS_GAUSS) sizeFactor= 8; // infinite ;)
1149 else if (flags&SWS_LANCZOS) sizeFactor= param[0] != SWS_PARAM_DEFAULT ? ceil(2*param[0]) : 6;
1150 else if (flags&SWS_SINC) sizeFactor= 20; // infinite ;)
1151 else if (flags&SWS_SPLINE) sizeFactor= 20; // infinite ;)
1152 else if (flags&SWS_BILINEAR) sizeFactor= 2;
1153 else {
1154 sizeFactor= 0; //GCC warning killer
1155 assert(0);
1156 }
1157
1158 if (xInc <= 1<<16) filterSize= 1 + sizeFactor; // upscale
1159 else filterSize= 1 + (sizeFactor*srcW + dstW - 1)/ dstW;
1160
1161 if (filterSize > srcW-2) filterSize=srcW-2;
1162
1163 filter= av_malloc(dstW*sizeof(*filter)*filterSize);
1164
1165 xDstInSrc= xInc - 0x10000;
1166 for (i=0; i<dstW; i++)
1167 {
1168 int xx= (xDstInSrc - ((filterSize-2)<<16)) / (1<<17);
1169 int j;
1170 (*filterPos)[i]= xx;
1171 for (j=0; j<filterSize; j++)
1172 {
1173 int64_t d= ((int64_t)FFABS((xx<<17) - xDstInSrc))<<13;
1174 double floatd;
1175 int64_t coeff;
1176
1177 if (xInc > 1<<16)
1178 d= d*dstW/srcW;
1179 floatd= d * (1.0/(1<<30));
1180
1181 if (flags & SWS_BICUBIC)
1182 {
1183 int64_t B= (param[0] != SWS_PARAM_DEFAULT ? param[0] : 0) * (1<<24);
1184 int64_t C= (param[1] != SWS_PARAM_DEFAULT ? param[1] : 0.6) * (1<<24);
1185 int64_t dd = ( d*d)>>30;
1186 int64_t ddd= (dd*d)>>30;
1187
1188 if (d < 1LL<<30)
1189 coeff = (12*(1<<24)-9*B-6*C)*ddd + (-18*(1<<24)+12*B+6*C)*dd + (6*(1<<24)-2*B)*(1<<30);
1190 else if (d < 1LL<<31)
1191 coeff = (-B-6*C)*ddd + (6*B+30*C)*dd + (-12*B-48*C)*d + (8*B+24*C)*(1<<30);
1192 else
1193 coeff=0.0;
1194 coeff *= fone>>(30+24);
1195 }
1196 /* else if (flags & SWS_X)
1197 {
1198 double p= param ? param*0.01 : 0.3;
1199 coeff = d ? sin(d*PI)/(d*PI) : 1.0;
1200 coeff*= pow(2.0, - p*d*d);
1201 }*/
1202 else if (flags & SWS_X)
1203 {
1204 double A= param[0] != SWS_PARAM_DEFAULT ? param[0] : 1.0;
1205 double c;
1206
1207 if (floatd<1.0)
1208 c = cos(floatd*PI);
1209 else
1210 c=-1.0;
1211 if (c<0.0) c= -pow(-c, A);
1212 else c= pow( c, A);
1213 coeff= (c*0.5 + 0.5)*fone;
1214 }
1215 else if (flags & SWS_AREA)
1216 {
1217 int64_t d2= d - (1<<29);
1218 if (d2*xInc < -(1LL<<(29+16))) coeff= 1.0 * (1LL<<(30+16));
1219 else if (d2*xInc < (1LL<<(29+16))) coeff= -d2*xInc + (1LL<<(29+16));
1220 else coeff=0.0;
1221 coeff *= fone>>(30+16);
1222 }
1223 else if (flags & SWS_GAUSS)
1224 {
1225 double p= param[0] != SWS_PARAM_DEFAULT ? param[0] : 3.0;
1226 coeff = (pow(2.0, - p*floatd*floatd))*fone;
1227 }
1228 else if (flags & SWS_SINC)
1229 {
1230 coeff = (d ? sin(floatd*PI)/(floatd*PI) : 1.0)*fone;
1231 }
1232 else if (flags & SWS_LANCZOS)
1233 {
1234 double p= param[0] != SWS_PARAM_DEFAULT ? param[0] : 3.0;
1235 coeff = (d ? sin(floatd*PI)*sin(floatd*PI/p)/(floatd*floatd*PI*PI/p) : 1.0)*fone;
1236 if (floatd>p) coeff=0;
1237 }
1238 else if (flags & SWS_BILINEAR)
1239 {
1240 coeff= (1<<30) - d;
1241 if (coeff<0) coeff=0;
1242 coeff *= fone >> 30;
1243 }
1244 else if (flags & SWS_SPLINE)
1245 {
1246 double p=-2.196152422706632;
1247 coeff = getSplineCoeff(1.0, 0.0, p, -p-1.0, floatd) * fone;
1248 }
1249 else {
1250 coeff= 0.0; //GCC warning killer
1251 assert(0);
1252 }
1253
1254 filter[i*filterSize + j]= coeff;
1255 xx++;
1256 }
1257 xDstInSrc+= 2*xInc;
1258 }
1259 }
1260
1261 /* apply src & dst Filter to filter -> filter2
1262 av_free(filter);
1263 */
1264 assert(filterSize>0);
1265 filter2Size= filterSize;
1266 if (srcFilter) filter2Size+= srcFilter->length - 1;
1267 if (dstFilter) filter2Size+= dstFilter->length - 1;
1268 assert(filter2Size>0);
1269 filter2= av_mallocz(filter2Size*dstW*sizeof(*filter2));
1270
1271 for (i=0; i<dstW; i++)
1272 {
1273 int j, k;
1274
1275 if(srcFilter){
1276 for (k=0; k<srcFilter->length; k++){
1277 for (j=0; j<filterSize; j++)
1278 filter2[i*filter2Size + k + j] += srcFilter->coeff[k]*filter[i*filterSize + j];
1279 }
1280 }else{
1281 for (j=0; j<filterSize; j++)
1282 filter2[i*filter2Size + j]= filter[i*filterSize + j];
1283 }
1284 //FIXME dstFilter
1285
1286 (*filterPos)[i]+= (filterSize-1)/2 - (filter2Size-1)/2;
1287 }
1288 av_freep(&filter);
1289
1290 /* try to reduce the filter-size (step1 find size and shift left) */
1291 // Assume it is near normalized (*0.5 or *2.0 is OK but * 0.001 is not).
1292 minFilterSize= 0;
1293 for (i=dstW-1; i>=0; i--)
1294 {
1295 int min= filter2Size;
1296 int j;
1297 int64_t cutOff=0.0;
1298
1299 /* get rid off near zero elements on the left by shifting left */
1300 for (j=0; j<filter2Size; j++)
1301 {
1302 int k;
1303 cutOff += FFABS(filter2[i*filter2Size]);
1304
1305 if (cutOff > SWS_MAX_REDUCE_CUTOFF*fone) break;
1306
1307 /* preserve monotonicity because the core can't handle the filter otherwise */
1308 if (i<dstW-1 && (*filterPos)[i] >= (*filterPos)[i+1]) break;
1309
1310 // Move filter coeffs left
1311 for (k=1; k<filter2Size; k++)
1312 filter2[i*filter2Size + k - 1]= filter2[i*filter2Size + k];
1313 filter2[i*filter2Size + k - 1]= 0;
1314 (*filterPos)[i]++;
1315 }
1316
1317 cutOff=0;
1318 /* count near zeros on the right */
1319 for (j=filter2Size-1; j>0; j--)
1320 {
1321 cutOff += FFABS(filter2[i*filter2Size + j]);
1322
1323 if (cutOff > SWS_MAX_REDUCE_CUTOFF*fone) break;
1324 min--;
1325 }
1326
1327 if (min>minFilterSize) minFilterSize= min;
1328 }
1329
1330 if (flags & SWS_CPU_CAPS_ALTIVEC) {
1331 // we can handle the special case 4,
1332 // so we don't want to go to the full 8
1333 if (minFilterSize < 5)
1334 filterAlign = 4;
1335
1336 // we really don't want to waste our time
1337 // doing useless computation, so fall-back on
1338 // the scalar C code for very small filter.
1339 // vectorizing is worth it only if you have
1340 // decent-sized vector.
1341 if (minFilterSize < 3)
1342 filterAlign = 1;
1343 }
1344
1345 if (flags & SWS_CPU_CAPS_MMX) {
1346 // special case for unscaled vertical filtering
1347 if (minFilterSize == 1 && filterAlign == 2)
1348 filterAlign= 1;
1349 }
1350
1351 assert(minFilterSize > 0);
1352 filterSize= (minFilterSize +(filterAlign-1)) & (~(filterAlign-1));
1353 assert(filterSize > 0);
1354 filter= av_malloc(filterSize*dstW*sizeof(*filter));
1355 if (filterSize >= MAX_FILTER_SIZE*16/((flags&SWS_ACCURATE_RND) ? APCK_SIZE : 16) || !filter)
1356 goto error;
1357 *outFilterSize= filterSize;
1358
1359 if (flags&SWS_PRINT_INFO)
1360 av_log(NULL, AV_LOG_VERBOSE, "SwScaler: reducing / aligning filtersize %d -> %d\n", filter2Size, filterSize);
1361 /* try to reduce the filter-size (step2 reduce it) */
1362 for (i=0; i<dstW; i++)
1363 {
1364 int j;
1365
1366 for (j=0; j<filterSize; j++)
1367 {
1368 if (j>=filter2Size) filter[i*filterSize + j]= 0;
1369 else filter[i*filterSize + j]= filter2[i*filter2Size + j];
1370 if((flags & SWS_BITEXACT) && j>=minFilterSize)
1371 filter[i*filterSize + j]= 0;
1372 }
1373 }
1374
1375
1376 //FIXME try to align filterpos if possible
1377
1378 //fix borders
1379 for (i=0; i<dstW; i++)
1380 {
1381 int j;
1382 if ((*filterPos)[i] < 0)
1383 {
1384 // Move filter coeffs left to compensate for filterPos
1385 for (j=1; j<filterSize; j++)
1386 {
1387 int left= FFMAX(j + (*filterPos)[i], 0);
1388 filter[i*filterSize + left] += filter[i*filterSize + j];
1389 filter[i*filterSize + j]=0;
1390 }
1391 (*filterPos)[i]= 0;
1392 }
1393
1394 if ((*filterPos)[i] + filterSize > srcW)
1395 {
1396 int shift= (*filterPos)[i] + filterSize - srcW;
1397 // Move filter coeffs right to compensate for filterPos
1398 for (j=filterSize-2; j>=0; j--)
1399 {
1400 int right= FFMIN(j + shift, filterSize-1);
1401 filter[i*filterSize +right] += filter[i*filterSize +j];
1402 filter[i*filterSize +j]=0;
1403 }
1404 (*filterPos)[i]= srcW - filterSize;
1405 }
1406 }
1407
1408 // Note the +1 is for the MMXscaler which reads over the end
1409 /* align at 16 for AltiVec (needed by hScale_altivec_real) */
1410 *outFilter= av_mallocz(*outFilterSize*(dstW+1)*sizeof(int16_t));
1411
1412 /* Normalize & Store in outFilter */
1413 for (i=0; i<dstW; i++)
1414 {
1415 int j;
1416 int64_t error=0;
1417 int64_t sum=0;
1418
1419 for (j=0; j<filterSize; j++)
1420 {
1421 sum+= filter[i*filterSize + j];
1422 }
1423 sum= (sum + one/2)/ one;
1424 for (j=0; j<*outFilterSize; j++)
1425 {
1426 int64_t v= filter[i*filterSize + j] + error;
1427 int intV= ROUNDED_DIV(v, sum);
1428 (*outFilter)[i*(*outFilterSize) + j]= intV;
1429 error= v - intV*sum;
1430 }
1431 }
1432
1433 (*filterPos)[dstW]= (*filterPos)[dstW-1]; // the MMX scaler will read over the end
1434 for (i=0; i<*outFilterSize; i++)
1435 {
1436 int j= dstW*(*outFilterSize);
1437 (*outFilter)[j + i]= (*outFilter)[j + i - (*outFilterSize)];
1438 }
1439
1440 ret=0;
1441 error:
1442 av_free(filter);
1443 av_free(filter2);
1444 return ret;
1445 }
1446
1447 #ifdef COMPILE_MMX2
1448 static void initMMX2HScaler(int dstW, int xInc, uint8_t *funnyCode, int16_t *filter, int32_t *filterPos, int numSplits)
1449 {
1450 uint8_t *fragmentA;
1451 long imm8OfPShufW1A;
1452 long imm8OfPShufW2A;
1453 long fragmentLengthA;
1454 uint8_t *fragmentB;
1455 long imm8OfPShufW1B;
1456 long imm8OfPShufW2B;
1457 long fragmentLengthB;
1458 int fragmentPos;
1459
1460 int xpos, i;
1461
1462 // create an optimized horizontal scaling routine
1463
1464 //code fragment
1465
1466 __asm__ volatile(
1467 "jmp 9f \n\t"
1468 // Begin
1469 "0: \n\t"
1470 "movq (%%"REG_d", %%"REG_a"), %%mm3 \n\t"
1471 "movd (%%"REG_c", %%"REG_S"), %%mm0 \n\t"
1472 "movd 1(%%"REG_c", %%"REG_S"), %%mm1 \n\t"
1473 "punpcklbw %%mm7, %%mm1 \n\t"
1474 "punpcklbw %%mm7, %%mm0 \n\t"
1475 "pshufw $0xFF, %%mm1, %%mm1 \n\t"
1476 "1: \n\t"
1477 "pshufw $0xFF, %%mm0, %%mm0 \n\t"
1478 "2: \n\t"
1479 "psubw %%mm1, %%mm0 \n\t"
1480 "movl 8(%%"REG_b", %%"REG_a"), %%esi \n\t"
1481 "pmullw %%mm3, %%mm0 \n\t"
1482 "psllw $7, %%mm1 \n\t"
1483 "paddw %%mm1, %%mm0 \n\t"
1484
1485 "movq %%mm0, (%%"REG_D", %%"REG_a") \n\t"
1486
1487 "add $8, %%"REG_a" \n\t"
1488 // End
1489 "9: \n\t"
1490 // "int $3 \n\t"
1491 "lea " LOCAL_MANGLE(0b) ", %0 \n\t"
1492 "lea " LOCAL_MANGLE(1b) ", %1 \n\t"
1493 "lea " LOCAL_MANGLE(2b) ", %2 \n\t"
1494 "dec %1 \n\t"
1495 "dec %2 \n\t"
1496 "sub %0, %1 \n\t"
1497 "sub %0, %2 \n\t"
1498 "lea " LOCAL_MANGLE(9b) ", %3 \n\t"
1499 "sub %0, %3 \n\t"
1500
1501
1502 :"=r" (fragmentA), "=r" (imm8OfPShufW1A), "=r" (imm8OfPShufW2A),
1503 "=r" (fragmentLengthA)
1504 );
1505
1506 __asm__ volatile(
1507 "jmp 9f \n\t"
1508 // Begin
1509 "0: \n\t"
1510 "movq (%%"REG_d", %%"REG_a"), %%mm3 \n\t"
1511 "movd (%%"REG_c", %%"REG_S"), %%mm0 \n\t"
1512 "punpcklbw %%mm7, %%mm0 \n\t"
1513 "pshufw $0xFF, %%mm0, %%mm1 \n\t"
1514 "1: \n\t"
1515 "pshufw $0xFF, %%mm0, %%mm0 \n\t"
1516 "2: \n\t"
1517 "psubw %%mm1, %%mm0 \n\t"
1518 "movl 8(%%"REG_b", %%"REG_a"), %%esi \n\t"
1519 "pmullw %%mm3, %%mm0 \n\t"
1520 "psllw $7, %%mm1 \n\t"
1521 "paddw %%mm1, %%mm0 \n\t"
1522
1523 "movq %%mm0, (%%"REG_D", %%"REG_a") \n\t"
1524
1525 "add $8, %%"REG_a" \n\t"
1526 // End
1527 "9: \n\t"
1528 // "int $3 \n\t"
1529 "lea " LOCAL_MANGLE(0b) ", %0 \n\t"
1530 "lea " LOCAL_MANGLE(1b) ", %1 \n\t"
1531 "lea " LOCAL_MANGLE(2b) ", %2 \n\t"
1532 "dec %1 \n\t"
1533 "dec %2 \n\t"
1534 "sub %0, %1 \n\t"
1535 "sub %0, %2 \n\t"
1536 "lea " LOCAL_MANGLE(9b) ", %3 \n\t"
1537 "sub %0, %3 \n\t"
1538
1539
1540 :"=r" (fragmentB), "=r" (imm8OfPShufW1B), "=r" (imm8OfPShufW2B),
1541 "=r" (fragmentLengthB)
1542 );
1543
1544 xpos= 0; //lumXInc/2 - 0x8000; // difference between pixel centers
1545 fragmentPos=0;
1546
1547 for (i=0; i<dstW/numSplits; i++)
1548 {
1549 int xx=xpos>>16;
1550
1551 if ((i&3) == 0)
1552 {
1553 int a=0;
1554 int b=((xpos+xInc)>>16) - xx;
1555 int c=((xpos+xInc*2)>>16) - xx;
1556 int d=((xpos+xInc*3)>>16) - xx;
1557
1558 filter[i ] = (( xpos & 0xFFFF) ^ 0xFFFF)>>9;
1559 filter[i+1] = (((xpos+xInc ) & 0xFFFF) ^ 0xFFFF)>>9;
1560 filter[i+2] = (((xpos+xInc*2) & 0xFFFF) ^ 0xFFFF)>>9;
1561 filter[i+3] = (((xpos+xInc*3) & 0xFFFF) ^ 0xFFFF)>>9;
1562 filterPos[i/2]= xx;
1563
1564 if (d+1<4)
1565 {
1566 int maxShift= 3-(d+1);
1567 int shift=0;
1568
1569 memcpy(funnyCode + fragmentPos, fragmentB, fragmentLengthB);
1570
1571 funnyCode[fragmentPos + imm8OfPShufW1B]=
1572 (a+1) | ((b+1)<<2) | ((c+1)<<4) | ((d+1)<<6);
1573 funnyCode[fragmentPos + imm8OfPShufW2B]=
1574 a | (b<<2) | (c<<4) | (d<<6);
1575
1576 if (i+3>=dstW) shift=maxShift; //avoid overread
1577 else if ((filterPos[i/2]&3) <= maxShift) shift=filterPos[i/2]&3; //Align
1578
1579 if (shift && i>=shift)
1580 {
1581 funnyCode[fragmentPos + imm8OfPShufW1B]+= 0x55*shift;
1582 funnyCode[fragmentPos + imm8OfPShufW2B]+= 0x55*shift;
1583 filterPos[i/2]-=shift;
1584 }
1585
1586 fragmentPos+= fragmentLengthB;
1587 }
1588 else
1589 {
1590 int maxShift= 3-d;
1591 int shift=0;
1592
1593 memcpy(funnyCode + fragmentPos, fragmentA, fragmentLengthA);
1594
1595 funnyCode[fragmentPos + imm8OfPShufW1A]=
1596 funnyCode[fragmentPos + imm8OfPShufW2A]=
1597 a | (b<<2) | (c<<4) | (d<<6);
1598
1599 if (i+4>=dstW) shift=maxShift; //avoid overread
1600 else if ((filterPos[i/2]&3) <= maxShift) shift=filterPos[i/2]&3; //partial align
1601
1602 if (shift && i>=shift)
1603 {
1604 funnyCode[fragmentPos + imm8OfPShufW1A]+= 0x55*shift;
1605 funnyCode[fragmentPos + imm8OfPShufW2A]+= 0x55*shift;
1606 filterPos[i/2]-=shift;
1607 }
1608
1609 fragmentPos+= fragmentLengthA;
1610 }
1611
1612 funnyCode[fragmentPos]= RET;
1613 }
1614 xpos+=xInc;
1615 }
1616 filterPos[i/2]= xpos>>16; // needed to jump to the next part
1617 }
1618 #endif /* COMPILE_MMX2 */
1619
1620 static void globalInit(void){
1621 // generating tables:
1622 int i;
1623 for (i=0; i<768; i++){
1624 int c= av_clip_uint8(i-256);
1625 clip_table[i]=c;
1626 }
1627 }
1628
1629 static SwsFunc getSwsFunc(int flags){
1630
1631 #if defined(RUNTIME_CPUDETECT) && CONFIG_GPL
1632 #if ARCH_X86
1633 // ordered per speed fastest first
1634 if (flags & SWS_CPU_CAPS_MMX2)
1635 return swScale_MMX2;
1636 else if (flags & SWS_CPU_CAPS_3DNOW)
1637 return swScale_3DNow;
1638 else if (flags & SWS_CPU_CAPS_MMX)
1639 return swScale_MMX;
1640 else
1641 return swScale_C;
1642
1643 #else
1644 #if ARCH_PPC
1645 if (flags & SWS_CPU_CAPS_ALTIVEC)
1646 return swScale_altivec;
1647 else
1648 return swScale_C;
1649 #endif
1650 return swScale_C;
1651 #endif /* ARCH_X86 */
1652 #else //RUNTIME_CPUDETECT
1653 #if HAVE_MMX2
1654 return swScale_MMX2;
1655 #elif HAVE_3DNOW
1656 return swScale_3DNow;
1657 #elif HAVE_MMX
1658 return swScale_MMX;
1659 #elif HAVE_ALTIVEC
1660 return swScale_altivec;
1661 #else
1662 return swScale_C;
1663 #endif
1664 #endif //!RUNTIME_CPUDETECT
1665 }
1666
1667 static int PlanarToNV12Wrapper(SwsContext *c, uint8_t* src[], int srcStride[], int srcSliceY,
1668 int srcSliceH, uint8_t* dstParam[], int dstStride[]){
1669 uint8_t *dst=dstParam[0] + dstStride[0]*srcSliceY;
1670 /* Copy Y plane */
1671 if (dstStride[0]==srcStride[0] && srcStride[0] > 0)
1672 memcpy(dst, src[0], srcSliceH*dstStride[0]);
1673 else
1674 {
1675 int i;
1676 uint8_t *srcPtr= src[0];
1677 uint8_t *dstPtr= dst;
1678 for (i=0; i<srcSliceH; i++)
1679 {
1680 memcpy(dstPtr, srcPtr, c->srcW);
1681 srcPtr+= srcStride[0];
1682 dstPtr+= dstStride[0];
1683 }
1684 }
1685 dst = dstParam[1] + dstStride[1]*srcSliceY/2;
1686 if (c->dstFormat == PIX_FMT_NV12)
1687 interleaveBytes(src[1], src[2], dst, c->srcW/2, srcSliceH/2, srcStride[1], srcStride[2], dstStride[0]);
1688 else
1689 interleaveBytes(src[2], src[1], dst, c->srcW/2, srcSliceH/2, srcStride[2], srcStride[1], dstStride[0]);
1690
1691 return srcSliceH;
1692 }
1693
1694 static int PlanarToYuy2Wrapper(SwsContext *c, uint8_t* src[], int srcStride[], int srcSliceY,
1695 int srcSliceH, uint8_t* dstParam[], int dstStride[]){
1696 uint8_t *dst=dstParam[0] + dstStride[0]*srcSliceY;
1697
1698 yv12toyuy2(src[0], src[1], src[2], dst, c->srcW, srcSliceH, srcStride[0], srcStride[1], dstStride[0]);
1699
1700 return srcSliceH;
1701 }
1702
1703 static int PlanarToUyvyWrapper(SwsContext *c, uint8_t* src[], int srcStride[], int srcSliceY,
1704 int srcSliceH, uint8_t* dstParam[], int dstStride[]){
1705 uint8_t *dst=dstParam[0] + dstStride[0]*srcSliceY;
1706
1707 yv12touyvy(src[0], src[1], src[2], dst, c->srcW, srcSliceH, srcStride[0], srcStride[1], dstStride[0]);
1708
1709 return srcSliceH;
1710 }
1711
1712 static int YUV422PToYuy2Wrapper(SwsContext *c, uint8_t* src[], int srcStride[], int srcSliceY,
1713 int srcSliceH, uint8_t* dstParam[], int dstStride[]){
1714 uint8_t *dst=dstParam[0] + dstStride[0]*srcSliceY;
1715
1716 yuv422ptoyuy2(src[0],src[1],src[2],dst,c->srcW,srcSliceH,srcStride[0],srcStride[1],dstStride[0]);
1717
1718 return srcSliceH;
1719 }
1720
1721 static int YUV422PToUyvyWrapper(SwsContext *c, uint8_t* src[], int srcStride[], int srcSliceY,
1722 int srcSliceH, uint8_t* dstParam[], int dstStride[]){
1723 uint8_t *dst=dstParam[0] + dstStride[0]*srcSliceY;
1724
1725 yuv422ptouyvy(src[0],src[1],src[2],dst,c->srcW,srcSliceH,srcStride[0],srcStride[1],dstStride[0]);
1726
1727 return srcSliceH;
1728 }
1729
1730 static int pal2rgbWrapper(SwsContext *c, uint8_t* src[], int srcStride[], int srcSliceY,
1731 int srcSliceH, uint8_t* dst[], int dstStride[]){
1732 const enum PixelFormat srcFormat= c->srcFormat;
1733 const enum PixelFormat dstFormat= c->dstFormat;
1734 void (*conv)(const uint8_t *src, uint8_t *dst, long num_pixels,
1735 const uint8_t *palette)=NULL;
1736 int i;
1737 uint8_t *dstPtr= dst[0] + dstStride[0]*srcSliceY;
1738 uint8_t *srcPtr= src[0];
1739
1740 if (!usePal(srcFormat))
1741 av_log(c, AV_LOG_ERROR, "internal error %s -> %s converter\n",
1742 sws_format_name(srcFormat), sws_format_name(dstFormat));
1743
1744 switch(dstFormat){
1745 case PIX_FMT_RGB32 : conv = palette8topacked32; break;
1746 case PIX_FMT_BGR32 : conv = palette8topacked32; break;
1747 case PIX_FMT_BGR32_1: conv = palette8topacked32; break;
1748 case PIX_FMT_RGB32_1: conv = palette8topacked32; break;
1749 case PIX_FMT_RGB24 : conv = palette8topacked24; break;
1750 case PIX_FMT_BGR24 : conv = palette8topacked24; break;
1751 default: av_log(c, AV_LOG_ERROR, "internal error %s -> %s converter\n",
1752 sws_format_name(srcFormat), sws_format_name(dstFormat)); break;
1753 }
1754
1755
1756 for (i=0; i<srcSliceH; i++) {
1757 conv(srcPtr, dstPtr, c->srcW, (uint8_t *) c->pal_rgb);
1758 srcPtr+= srcStride[0];
1759 dstPtr+= dstStride[0];
1760 }
1761
1762 return srcSliceH;
1763 }
1764
1765 /* {RGB,BGR}{15,16,24,32,32_1} -> {RGB,BGR}{15,16,24,32} */
1766 static int rgb2rgbWrapper(SwsContext *c, uint8_t* src[], int srcStride[], int srcSliceY,
1767 int srcSliceH, uint8_t* dst[], int dstStride[]){
1768 const enum PixelFormat srcFormat= c->srcFormat;
1769 const enum PixelFormat dstFormat= c->dstFormat;
1770 const int srcBpp= (fmt_depth(srcFormat) + 7) >> 3;
1771 const int dstBpp= (fmt_depth(dstFormat) + 7) >> 3;
1772 const int srcId= fmt_depth(srcFormat) >> 2; /* 1:0, 4:1, 8:2, 15:3, 16:4, 24:6, 32:8 */
1773 const int dstId= fmt_depth(dstFormat) >> 2;
1774 void (*conv)(const uint8_t *src, uint8_t *dst, long src_size)=NULL;
1775
1776 /* BGR -> BGR */
1777 if ( (isBGR(srcFormat) && isBGR(dstFormat))
1778 || (isRGB(srcFormat) && isRGB(dstFormat))){
1779 switch(srcId | (dstId<<4)){
1780 case 0x34: conv= rgb16to15; break;
1781 case 0x36: conv= rgb24to15; break;
1782 case 0x38: conv= rgb32to15; break;
1783 case 0x43: conv= rgb15to16; break;
1784 case 0x46: conv= rgb24to16; break;
1785 case 0x48: conv= rgb32to16; break;
1786 case 0x63: conv= rgb15to24; break;
1787 case 0x64: conv= rgb16to24; break;
1788 case 0x68: conv= rgb32to24; break;
1789 case 0x83: conv= rgb15to32; break;
1790 case 0x84: conv= rgb16to32; break;
1791 case 0x86: conv= rgb24to32; break;
1792 default: av_log(c, AV_LOG_ERROR, "internal error %s -> %s converter\n",
1793 sws_format_name(srcFormat), sws_format_name(dstFormat)); break;
1794 }
1795 }else if ( (isBGR(srcFormat) && isRGB(dstFormat))
1796 || (isRGB(srcFormat) && isBGR(dstFormat))){
1797 switch(srcId | (dstId<<4)){
1798 case 0x33: conv= rgb15tobgr15; break;
1799 case 0x34: conv= rgb16tobgr15; break;
1800 case 0x36: conv= rgb24tobgr15; break;
1801 case 0x38: conv= rgb32tobgr15; break;
1802 case 0x43: conv= rgb15tobgr16; break;
1803 case 0x44: conv= rgb16tobgr16; break;
1804 case 0x46: conv= rgb24tobgr16; break;
1805 case 0x48: conv= rgb32tobgr16; break;
1806 case 0x63: conv= rgb15tobgr24; break;
1807 case 0x64: conv= rgb16tobgr24; break;
1808 case 0x66: conv= rgb24tobgr24; break;
1809 case 0x68: conv= rgb32tobgr24; break;
1810 case 0x83: conv= rgb15tobgr32; break;
1811 case 0x84: conv= rgb16tobgr32; break;
1812 case 0x86: conv= rgb24tobgr32; break;
1813 case 0x88: conv= rgb32tobgr32; break;
1814 default: av_log(c, AV_LOG_ERROR, "internal error %s -> %s converter\n",
1815 sws_format_name(srcFormat), sws_format_name(dstFormat)); break;
1816 }
1817 }else{
1818 av_log(c, AV_LOG_ERROR, "internal error %s -> %s converter\n",
1819 sws_format_name(srcFormat), sws_format_name(dstFormat));
1820 }
1821
1822 if(conv)
1823 {
1824 uint8_t *srcPtr= src[0];
1825 if(srcFormat == PIX_FMT_RGB32_1 || srcFormat == PIX_FMT_BGR32_1)
1826 srcPtr += ALT32_CORR;
1827
1828 if (dstStride[0]*srcBpp == srcStride[0]*dstBpp && srcStride[0] > 0)
1829 conv(srcPtr, dst[0] + dstStride[0]*srcSliceY, srcSliceH*srcStride[0]);
1830 else
1831 {
1832 int i;
1833 uint8_t *dstPtr= dst[0] + dstStride[0]*srcSliceY;
1834
1835 for (i=0; i<srcSliceH; i++)
1836 {
1837 conv(srcPtr, dstPtr, c->srcW*srcBpp);
1838 srcPtr+= srcStride[0];
1839 dstPtr+= dstStride[0];
1840 }
1841 }
1842 }
1843 return srcSliceH;
1844 }
1845
1846 static int bgr24toyv12Wrapper(SwsContext *c, uint8_t* src[], int srcStride[], int srcSliceY,
1847 int srcSliceH, uint8_t* dst[], int dstStride[]){
1848
1849 rgb24toyv12(
1850 src[0],
1851 dst[0]+ srcSliceY *dstStride[0],
1852 dst[1]+(srcSliceY>>1)*dstStride[1],
1853 dst[2]+(srcSliceY>>1)*dstStride[2],
1854 c->srcW, srcSliceH,
1855 dstStride[0], dstStride[1], srcStride[0]);
1856 return srcSliceH;
1857 }
1858
1859 static int yvu9toyv12Wrapper(SwsContext *c, uint8_t* src[], int srcStride[], int srcSliceY,
1860 int srcSliceH, uint8_t* dst[], int dstStride[]){
1861 int i;
1862
1863 /* copy Y */
1864 if (srcStride[0]==dstStride[0] && srcStride[0] > 0)
1865 memcpy(dst[0]+ srcSliceY*dstStride[0], src[0], srcStride[0]*srcSliceH);
1866 else{
1867 uint8_t *srcPtr= src[0];
1868 uint8_t *dstPtr= dst[0] + dstStride[0]*srcSliceY;
1869
1870 for (i=0; i<srcSliceH; i++)
1871 {
1872 memcpy(dstPtr, srcPtr, c->srcW);
1873 srcPtr+= srcStride[0];
1874 dstPtr+= dstStride[0];
1875 }
1876 }
1877
1878 if (c->dstFormat==PIX_FMT_YUV420P){
1879 planar2x(src[1], dst[1], c->chrSrcW, c->chrSrcH, srcStride[1], dstStride[1]);
1880 planar2x(src[2], dst[2], c->chrSrcW, c->chrSrcH, srcStride[2], dstStride[2]);
1881 }else{
1882 planar2x(src[1], dst[2], c->chrSrcW, c->chrSrcH, srcStride[1], dstStride[2]);
1883 planar2x(src[2], dst[1], c->chrSrcW, c->chrSrcH, srcStride[2], dstStride[1]);
1884 }
1885 return srcSliceH;
1886 }
1887
1888 /* unscaled copy like stuff (assumes nearly identical formats) */
1889 static int packedCopy(SwsContext *c, uint8_t* src[], int srcStride[], int srcSliceY,
1890 int srcSliceH, uint8_t* dst[], int dstStride[])
1891 {
1892 if (dstStride[0]==srcStride[0] && srcStride[0] > 0)
1893 memcpy(dst[0] + dstStride[0]*srcSliceY, src[0], srcSliceH*dstStride[0]);
1894 else
1895 {
1896 int i;
1897 uint8_t *srcPtr= src[0];
1898 uint8_t *dstPtr= dst[0] + dstStride[0]*srcSliceY;
1899 int length=0;
1900
1901 /* universal length finder */
1902 while(length+c->srcW <= FFABS(dstStride[0])
1903 && length+c->srcW <= FFABS(srcStride[0])) length+= c->srcW;
1904 assert(length!=0);
1905
1906 for (i=0; i<srcSliceH; i++)
1907 {
1908 memcpy(dstPtr, srcPtr, length);
1909 srcPtr+= srcStride[0];
1910 dstPtr+= dstStride[0];
1911 }
1912 }
1913 return srcSliceH;
1914 }
1915
1916 static int planarCopy(SwsContext *c, uint8_t* src[], int srcStride[], int srcSliceY,
1917 int srcSliceH, uint8_t* dst[], int dstStride[])
1918 {
1919 int plane;
1920 for (plane=0; plane<3; plane++)
1921 {
1922 int length= plane==0 ? c->srcW : -((-c->srcW )>>c->chrDstHSubSample);
1923 int y= plane==0 ? srcSliceY: -((-srcSliceY)>>c->chrDstVSubSample);
1924 int height= plane==0 ? srcSliceH: -((-srcSliceH)>>c->chrDstVSubSample);
1925
1926 if ((isGray(c->srcFormat) || isGray(c->dstFormat)) && plane>0)
1927 {
1928 if (!isGray(c->dstFormat))
1929 memset(dst[plane], 128, dstStride[plane]*height);
1930 }
1931 else
1932 {
1933 if (dstStride[plane]==srcStride[plane] && srcStride[plane] > 0)
1934 memcpy(dst[plane] + dstStride[plane]*y, src[plane], height*dstStride[plane]);
1935 else
1936 {
1937 int i;
1938 uint8_t *srcPtr= src[plane];
1939 uint8_t *dstPtr= dst[plane] + dstStride[plane]*y;
1940 for (i=0; i<height; i++)
1941 {
1942 memcpy(dstPtr, srcPtr, length);
1943 srcPtr+= srcStride[plane];
1944 dstPtr+= dstStride[plane];
1945 }
1946 }
1947 }
1948 }
1949 return srcSliceH;
1950 }
1951
1952 static int gray16togray(SwsContext *c, uint8_t* src[], int srcStride[], int srcSliceY,
1953 int srcSliceH, uint8_t* dst[], int dstStride[]){
1954
1955 int length= c->srcW;
1956 int y= srcSliceY;
1957 int height= srcSliceH;
1958 int i, j;
1959 uint8_t *srcPtr= src[0];
1960 uint8_t *dstPtr= dst[0] + dstStride[0]*y;
1961
1962 if (!isGray(c->dstFormat)){
1963 int height= -((-srcSliceH)>>c->chrDstVSubSample);
1964 memset(dst[1], 128, dstStride[1]*height);
1965 memset(dst[2], 128, dstStride[2]*height);
1966 }
1967 if (c->srcFormat == PIX_FMT_GRAY16LE) srcPtr++;
1968 for (i=0; i<height; i++)
1969 {
1970 for (j=0; j<length; j++) dstPtr[j] = srcPtr[j<<1];
1971 srcPtr+= srcStride[0];
1972 dstPtr+= dstStride[0];
1973 }
1974 return srcSliceH;
1975 }
1976
1977 static int graytogray16(SwsContext *c, uint8_t* src[], int srcStride[], int srcSliceY,
1978 int srcSliceH, uint8_t* dst[], int dstStride[]){
1979
1980 int length= c->srcW;
1981 int y= srcSliceY;
1982 int height= srcSliceH;
1983 int i, j;
1984 uint8_t *srcPtr= src[0];
1985 uint8_t *dstPtr= dst[0] + dstStride[0]*y;
1986 for (i=0; i<height; i++)
1987 {
1988 for (j=0; j<length; j++)
1989 {
1990 dstPtr[j<<1] = srcPtr[j];
1991 dstPtr[(j<<1)+1] = srcPtr[j];
1992 }
1993 srcPtr+= srcStride[0];
1994 dstPtr+= dstStride[0];
1995 }
1996 return srcSliceH;
1997 }
1998
1999 static int gray16swap(SwsContext *c, uint8_t* src[], int srcStride[], int srcSliceY,
2000 int srcSliceH, uint8_t* dst[], int dstStride[]){
2001
2002 int length= c->srcW;
2003 int y= srcSliceY;
2004 int height= srcSliceH;
2005 int i, j;
2006 uint16_t *srcPtr= (uint16_t*)src[0];
2007 uint16_t *dstPtr= (uint16_t*)(dst[0] + dstStride[0]*y/2);
2008 for (i=0; i<height; i++)
2009 {
2010 for (j=0; j<length; j++) dstPtr[j] = bswap_16(srcPtr[j]);
2011 srcPtr+= srcStride[0]/2;
2012 dstPtr+= dstStride[0]/2;
2013 }
2014 return srcSliceH;
2015 }
2016
2017
2018 static void getSubSampleFactors(int *h, int *v, int format){
2019 switch(format){
2020 case PIX_FMT_UYVY422:
2021 case PIX_FMT_YUYV422:
2022 *h=1;
2023 *v=0;
2024 break;
2025 case PIX_FMT_YUV420P:
2026 case PIX_FMT_YUVA420P:
2027 case PIX_FMT_GRAY16BE:
2028 case PIX_FMT_GRAY16LE:
2029 case PIX_FMT_GRAY8: //FIXME remove after different subsamplings are fully implemented
2030 case PIX_FMT_NV12:
2031 case PIX_FMT_NV21:
2032 *h=1;
2033 *v=1;
2034 break;
2035 case PIX_FMT_YUV440P:
2036 *h=0;
2037 *v=1;
2038 break;
2039 case PIX_FMT_YUV410P:
2040 *h=2;
2041 *v=2;
2042 break;
2043 case PIX_FMT_YUV444P:
2044 *h=0;
2045 *v=0;
2046 break;
2047 case PIX_FMT_YUV422P:
2048 *h=1;
2049 *v=0;
2050 break;
2051 case PIX_FMT_YUV411P:
2052 *h=2;
2053 *v=0;
2054 break;
2055 default:
2056 *h=0;
2057 *v=0;
2058 break;
2059 }
2060 }
2061
2062 static uint16_t roundToInt16(int64_t f){
2063 int r= (f + (1<<15))>>16;
2064 if (r<-0x7FFF) return 0x8000;
2065 else if (r> 0x7FFF) return 0x7FFF;
2066 else return r;
2067 }
2068
2069 /**
2070 * @param inv_table the yuv2rgb coeffs, normally Inverse_Table_6_9[x]
2071 * @param fullRange if 1 then the luma range is 0..255 if 0 it is 16..235
2072 * @return -1 if not supported
2073 */
2074 int sws_setColorspaceDetails(SwsContext *c, const int inv_table[4], int srcRange, const int table[4], int dstRange, int brightness, int contrast, int saturation){
2075 int64_t crv = inv_table[0];
2076 int64_t cbu = inv_table[1];
2077 int64_t cgu = -inv_table[2];
2078 int64_t cgv = -inv_table[3];
2079 int64_t cy = 1<<16;
2080 int64_t oy = 0;
2081
2082 memcpy(c->srcColorspaceTable, inv_table, sizeof(int)*4);
2083 memcpy(c->dstColorspaceTable, table, sizeof(int)*4);
2084
2085 c->brightness= brightness;
2086 c->contrast = contrast;
2087 c->saturation= saturation;
2088 c->srcRange = srcRange;
2089 c->dstRange = dstRange;
2090 if (isYUV(c->dstFormat) || isGray(c->dstFormat)) return 0;
2091
2092 c->uOffset= 0x0400040004000400LL;
2093 c->vOffset= 0x0400040004000400LL;
2094
2095 if (!srcRange){
2096 cy= (cy*255) / 219;
2097 oy= 16<<16;
2098 }else{
2099 crv= (crv*224) / 255;
2100 cbu= (cbu*224) / 255;
2101 cgu= (cgu*224) / 255;
2102 cgv= (cgv*224) / 255;
2103 }
2104
2105 cy = (cy *contrast )>>16;
2106 crv= (crv*contrast * saturation)>>32;
2107 cbu= (cbu*contrast * saturation)>>32;
2108 cgu= (cgu*contrast * saturation)>>32;
2109 cgv= (cgv*contrast * saturation)>>32;
2110
2111 oy -= 256*brightness;
2112
2113 c->yCoeff= roundToInt16(cy *8192) * 0x0001000100010001ULL;
2114 c->vrCoeff= roundToInt16(crv*8192) * 0x0001000100010001ULL;
2115 c->ubCoeff= roundToInt16(cbu*8192) * 0x0001000100010001ULL;
2116 c->vgCoeff= roundToInt16(cgv*8192) * 0x0001000100010001ULL;
2117 c->ugCoeff= roundToInt16(cgu*8192) * 0x0001000100010001ULL;
2118 c->yOffset= roundToInt16(oy * 8) * 0x0001000100010001ULL;
2119
2120 c->yuv2rgb_y_coeff = (int16_t)roundToInt16(cy <<13);
2121 c->yuv2rgb_y_offset = (int16_t)roundToInt16(oy << 9);
2122 c->yuv2rgb_v2r_coeff= (int16_t)roundToInt16(crv<<13);
2123 c->yuv2rgb_v2g_coeff= (int16_t)roundToInt16(cgv<<13);
2124 c->yuv2rgb_u2g_coeff= (int16_t)roundToInt16(cgu<<13);
2125 c->yuv2rgb_u2b_coeff= (int16_t)roundToInt16(cbu<<13);
2126
2127 yuv2rgb_c_init_tables(c, inv_table, srcRange, brightness, contrast, saturation);
2128 //FIXME factorize
2129
2130 #ifdef COMPILE_ALTIVEC
2131 if (c->flags & SWS_CPU_CAPS_ALTIVEC)
2132 yuv2rgb_altivec_init_tables (c, inv_table, brightness, contrast, saturation);
2133 #endif
2134 return 0;
2135 }
2136
2137 /**
2138 * @return -1 if not supported
2139 */
2140 int sws_getColorspaceDetails(SwsContext *c, int **inv_table, int *srcRange, int **table, int *dstRange, int *brightness, int *contrast, int *saturation){
2141 if (isYUV(c->dstFormat) || isGray(c->dstFormat)) return -1;
2142
2143 *inv_table = c->srcColorspaceTable;
2144 *table = c->dstColorspaceTable;
2145 *srcRange = c->srcRange;
2146 *dstRange = c->dstRange;
2147 *brightness= c->brightness;
2148 *contrast = c->contrast;
2149 *saturation= c->saturation;
2150
2151 return 0;
2152 }
2153
2154 static int handle_jpeg(enum PixelFormat *format)
2155 {
2156 switch (*format) {
2157 case PIX_FMT_YUVJ420P:
2158 *format = PIX_FMT_YUV420P;
2159 return 1;
2160 case PIX_FMT_YUVJ422P:
2161 *format = PIX_FMT_YUV422P;
2162 return 1;
2163 case PIX_FMT_YUVJ444P:
2164 *format = PIX_FMT_YUV444P;
2165 return 1;
2166 case PIX_FMT_YUVJ440P:
2167 *format = PIX_FMT_YUV440P;
2168 return 1;
2169 default:
2170 return 0;
2171 }
2172 }
2173
2174 SwsContext *sws_getContext(int srcW, int srcH, enum PixelFormat srcFormat, int dstW, int dstH, enum PixelFormat dstFormat, int flags,
2175 SwsFilter *srcFilter, SwsFilter *dstFilter, double *param){
2176
2177 SwsContext *c;
2178 int i;
2179 int usesVFilter, usesHFilter;
2180 int unscaled, needsDither;
2181 int srcRange, dstRange;
2182 SwsFilter dummyFilter= {NULL, NULL, NULL, NULL};
2183 #if ARCH_X86
2184 if (flags & SWS_CPU_CAPS_MMX)
2185 __asm__ volatile("emms\n\t"::: "memory");
2186 #endif
2187
2188 #if !defined(RUNTIME_CPUDETECT) || !CONFIG_GPL //ensure that the flags match the compiled variant if cpudetect is off
2189 flags &= ~(SWS_CPU_CAPS_MMX|SWS_CPU_CAPS_MMX2|SWS_CPU_CAPS_3DNOW|SWS_CPU_CAPS_ALTIVEC|SWS_CPU_CAPS_BFIN);
2190 #if HAVE_MMX2
2191 flags |= SWS_CPU_CAPS_MMX|SWS_CPU_CAPS_MMX2;
2192 #elif HAVE_3DNOW
2193 flags |= SWS_CPU_CAPS_MMX|SWS_CPU_CAPS_3DNOW;
2194 #elif HAVE_MMX
2195 flags |= SWS_CPU_CAPS_MMX;
2196 #elif HAVE_ALTIVEC
2197 flags |= SWS_CPU_CAPS_ALTIVEC;
2198 #elif ARCH_BFIN
2199 flags |= SWS_CPU_CAPS_BFIN;
2200 #endif
2201 #endif /* RUNTIME_CPUDETECT */
2202 if (clip_table[512] != 255) globalInit();
2203 if (!rgb15to16) sws_rgb2rgb_init(flags);
2204
2205 unscaled = (srcW == dstW && srcH == dstH);
2206 needsDither= (isBGR(dstFormat) || isRGB(dstFormat))
2207 && (fmt_depth(dstFormat))<24
2208 && ((fmt_depth(dstFormat))<(fmt_depth(srcFormat)) || (!(isRGB(srcFormat) || isBGR(srcFormat))));
2209
2210 srcRange = handle_jpeg(&srcFormat);
2211 dstRange = handle_jpeg(&dstFormat);
2212
2213 if (!isSupportedIn(srcFormat))
2214 {
2215 av_log(NULL, AV_LOG_ERROR, "swScaler: %s is not supported as input pixel format\n", sws_format_name(srcFormat));
2216 return NULL;
2217 }
2218 if (!isSupportedOut(dstFormat))
2219 {
2220 av_log(NULL, AV_LOG_ERROR, "swScaler: %s is not supported as output pixel format\n", sws_format_name(dstFormat));
2221 return NULL;
2222 }
2223
2224 i= flags & ( SWS_POINT
2225 |SWS_AREA
2226 |SWS_BILINEAR
2227 |SWS_FAST_BILINEAR
2228 |SWS_BICUBIC
2229 |SWS_X
2230 |SWS_GAUSS
2231 |SWS_LANCZOS
2232 |SWS_SINC
2233 |SWS_SPLINE
2234 |SWS_BICUBLIN);
2235 if(!i || (i & (i-1)))
2236 {
2237 av_log(NULL, AV_LOG_ERROR, "swScaler: Exactly one scaler algorithm must be choosen\n");
2238 return NULL;
2239 }
2240
2241 /* sanity check */
2242 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
2243 {
2244 av_log(NULL, AV_LOG_ERROR, "swScaler: %dx%d -> %dx%d is invalid scaling dimension\n",
2245 srcW, srcH, dstW, dstH);
2246 return NULL;
2247 }
2248 if(srcW > VOFW || dstW > VOFW){
2249 av_log(NULL, AV_LOG_ERROR, "swScaler: Compile time max width is "AV_STRINGIFY(VOFW)" change VOF/VOFW and recompile\n");
2250 return NULL;
2251 }
2252
2253 if (!dstFilter) dstFilter= &dummyFilter;
2254 if (!srcFilter) srcFilter= &dummyFilter;
2255
2256 c= av_mallocz(sizeof(SwsContext));
2257
2258 c->av_class = &sws_context_class;
2259 c->srcW= srcW;
2260 c->srcH= srcH;
2261 c->dstW= dstW;
2262 c->dstH= dstH;
2263 c->lumXInc= ((srcW<<16) + (dstW>>1))/dstW;
2264 c->lumYInc= ((srcH<<16) + (dstH>>1))/dstH;
2265 c->flags= flags;
2266 c->dstFormat= dstFormat;
2267 c->srcFormat= srcFormat;
2268 c->vRounder= 4* 0x0001000100010001ULL;
2269
2270 usesHFilter= usesVFilter= 0;
2271 if (dstFilter->lumV && dstFilter->lumV->length>1) usesVFilter=1;
2272 if (dstFilter->lumH && dstFilter->lumH->length>1) usesHFilter=1;
2273 if (dstFilter->chrV && dstFilter->chrV->length>1) usesVFilter=1;
2274 if (dstFilter->chrH && dstFilter->chrH->length>1) usesHFilter=1;
2275 if (srcFilter->lumV && srcFilter->lumV->length>1) usesVFilter=1;
2276 if (srcFilter->lumH && srcFilter->lumH->length>1) usesHFilter=1;
2277 if (srcFilter->chrV && srcFilter->chrV->length>1) usesVFilter=1;
2278 if (srcFilter->chrH && srcFilter->chrH->length>1) usesHFilter=1;
2279
2280 getSubSampleFactors(&c->chrSrcHSubSample, &c->chrSrcVSubSample, srcFormat);
2281 getSubSampleFactors(&c->chrDstHSubSample, &c->chrDstVSubSample, dstFormat);
2282
2283 // reuse chroma for 2 pixles rgb/bgr unless user wants full chroma interpolation
2284 if ((isBGR(dstFormat) || isRGB(dstFormat)) && !(flags&SWS_FULL_CHR_H_INT)) c->chrDstHSubSample=1;
2285
2286 // drop some chroma lines if the user wants it
2287 c->vChrDrop= (flags&SWS_SRC_V_CHR_DROP_MASK)>>SWS_SRC_V_CHR_DROP_SHIFT;
2288 c->chrSrcVSubSample+= c->vChrDrop;
2289
2290 // drop every 2. pixel for chroma calculation unless user wants full chroma
2291 if ((isBGR(srcFormat) || isRGB(srcFormat)) && !(flags&SWS_FULL_CHR_H_INP)
2292 && srcFormat!=PIX_FMT_RGB8 && srcFormat!=PIX_FMT_BGR8
2293 && srcFormat!=PIX_FMT_RGB4 && srcFormat!=PIX_FMT_BGR4
2294 && srcFormat!=PIX_FMT_RGB4_BYTE && srcFormat!=PIX_FMT_BGR4_BYTE
2295 && ((dstW>>c->chrDstHSubSample) <= (srcW>>1) || (flags&(SWS_FAST_BILINEAR|SWS_POINT))))
2296 c->chrSrcHSubSample=1;
2297
2298 if (param){
2299 c->param[0] = param[0];
2300 c->param[1] = param[1];
2301 }else{
2302 c->param[0] =
2303 c->param[1] = SWS_PARAM_DEFAULT;
2304 }
2305
2306 c->chrIntHSubSample= c->chrDstHSubSample;
2307 c->chrIntVSubSample= c->chrSrcVSubSample;
2308
2309 // Note the -((-x)>>y) is so that we always round toward +inf.
2310 c->chrSrcW= -((-srcW) >> c->chrSrcHSubSample);
2311 c->chrSrcH= -((-srcH) >> c->chrSrcVSubSample);
2312 c->chrDstW= -((-dstW) >> c->chrDstHSubSample);
2313 c->chrDstH= -((-dstH) >> c->chrDstVSubSample);
2314
2315 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);
2316
2317 /* unscaled special Cases */
2318 if (unscaled && !usesHFilter && !usesVFilter && (srcRange == dstRange || isBGR(dstFormat) || isRGB(dstFormat)))
2319 {
2320 /* yv12_to_nv12 */
2321 if (srcFormat == PIX_FMT_YUV420P && (dstFormat == PIX_FMT_NV12 || dstFormat == PIX_FMT_NV21))
2322 {
2323 c->swScale= PlanarToNV12Wrapper;
2324 }
2325 #if CONFIG_GPL
2326 /* yuv2bgr */
2327 if ((srcFormat==PIX_FMT_YUV420P || srcFormat==PIX_FMT_YUV422P) && (isBGR(dstFormat) || isRGB(dstFormat))
2328 && !(flags & SWS_ACCURATE_RND) && !(dstH&1))
2329 {
2330 c->swScale= yuv2rgb_get_func_ptr(c);
2331 }
2332 #endif
2333
2334 if (srcFormat==PIX_FMT_YUV410P && dstFormat==PIX_FMT_YUV420P && !(flags & SWS_BITEXACT))
2335 {
2336 c->swScale= yvu9toyv12Wrapper;
2337 }
2338
2339 /* bgr24toYV12 */
2340 if (srcFormat==PIX_FMT_BGR24 && dstFormat==PIX_FMT_YUV420P && !(flags & SWS_ACCURATE_RND))
2341 c->swScale= bgr24toyv12Wrapper;
2342
2343 /* rgb/bgr -> rgb/bgr (no dither needed forms) */
2344 if ( (isBGR(srcFormat) || isRGB(srcFormat))
2345 && (isBGR(dstFormat) || isRGB(dstFormat))
2346 && srcFormat != PIX_FMT_BGR8 && dstFormat != PIX_FMT_BGR8
2347 && srcFormat != PIX_FMT_RGB8 && dstFormat != PIX_FMT_RGB8
2348 && srcFormat != PIX_FMT_BGR4 && dstFormat != PIX_FMT_BGR4
2349 && srcFormat != PIX_FMT_RGB4 && dstFormat != PIX_FMT_RGB4
2350 && srcFormat != PIX_FMT_BGR4_BYTE && dstFormat != PIX_FMT_BGR4_BYTE
2351 && srcFormat != PIX_FMT_RGB4_BYTE && dstFormat != PIX_FMT_RGB4_BYTE
2352 && srcFormat != PIX_FMT_MONOBLACK && dstFormat != PIX_FMT_MONOBLACK
2353 && srcFormat != PIX_FMT_MONOWHITE && dstFormat != PIX_FMT_MONOWHITE
2354 && dstFormat != PIX_FMT_RGB32_1
2355 && dstFormat != PIX_FMT_BGR32_1
2356 && (!needsDither || (c->flags&(SWS_FAST_BILINEAR|SWS_POINT))))
2357 c->swScale= rgb2rgbWrapper;
2358
2359 if ((usePal(srcFormat) && (
2360 dstFormat == PIX_FMT_RGB32 ||
2361 dstFormat == PIX_FMT_RGB32_1 ||
2362 dstFormat == PIX_FMT_RGB24 ||
2363 dstFormat == PIX_FMT_BGR32 ||
2364 dstFormat == PIX_FMT_BGR32_1 ||
2365 dstFormat == PIX_FMT_BGR24)))
2366 c->swScale= pal2rgbWrapper;
2367
2368 if (srcFormat == PIX_FMT_YUV422P)
2369 {
2370 if (dstFormat == PIX_FMT_YUYV422)
2371 c->swScale= YUV422PToYuy2Wrapper;
2372 else if (dstFormat == PIX_FMT_UYVY422)
2373 c->swScale= YUV422PToUyvyWrapper;
2374 }
2375
2376 /* LQ converters if -sws 0 or -sws 4*/
2377 if (c->flags&(SWS_FAST_BILINEAR|SWS_POINT)){
2378 /* yv12_to_yuy2 */
2379 if (srcFormat == PIX_FMT_YUV420P)
2380 {
2381 if (dstFormat == PIX_FMT_YUYV422)
2382 c->swScale= PlanarToYuy2Wrapper;
2383 else if (dstFormat == PIX_FMT_UYVY422)
2384 c->swScale= PlanarToUyvyWrapper;
2385 }
2386 }
2387
2388 #ifdef COMPILE_ALTIVEC
2389 if ((c->flags & SWS_CPU_CAPS_ALTIVEC) &&
2390 srcFormat == PIX_FMT_YUV420P) {
2391 // unscaled YV12 -> packed YUV, we want speed
2392 if (dstFormat == PIX_FMT_YUYV422)
2393 c->swScale= yv12toyuy2_unscaled_altivec;
2394 else if (dstFormat == PIX_FMT_UYVY422)
2395 c->swScale= yv12touyvy_unscaled_altivec;
2396 }
2397 #endif
2398
2399 /* simple copy */
2400 if ( srcFormat == dstFormat
2401 || (isPlanarYUV(srcFormat) && isGray(dstFormat))
2402 || (isPlanarYUV(dstFormat) && isGray(srcFormat)))
2403 {
2404 if (isPacked(c->srcFormat))
2405 c->swScale= packedCopy;
2406 else /* Planar YUV or gray */
2407 c->swScale= planarCopy;
2408 }
2409
2410 /* gray16{le,be} conversions */
2411 if (isGray16(srcFormat) && (isPlanarYUV(dstFormat) || (dstFormat == PIX_FMT_GRAY8)))
2412 {
2413 c->swScale= gray16togray;
2414 }
2415 if ((isPlanarYUV(srcFormat) || (srcFormat == PIX_FMT_GRAY8)) && isGray16(dstFormat))
2416 {
2417 c->swScale= graytogray16;
2418 }
2419 if (srcFormat != dstFormat && isGray16(srcFormat) && isGray16(dstFormat))
2420 {
2421 c->swScale= gray16swap;
2422 }
2423
2424 #if ARCH_BFIN
2425 if (flags & SWS_CPU_CAPS_BFIN)
2426 ff_bfin_get_unscaled_swscale (c);
2427 #endif
2428
2429 if (c->swScale){
2430 if (flags&SWS_PRINT_INFO)
2431 av_log(c, AV_LOG_INFO, "using unscaled %s -> %s special converter\n",
2432 sws_format_name(srcFormat), sws_format_name(dstFormat));
2433 return c;
2434 }
2435 }
2436
2437 if (flags & SWS_CPU_CAPS_MMX2)
2438 {
2439 c->canMMX2BeUsed= (dstW >=srcW && (dstW&31)==0 && (srcW&15)==0) ? 1 : 0;
2440 if (!c->canMMX2BeUsed && dstW >=srcW && (srcW&15)==0 && (flags&SWS_FAST_BILINEAR))
2441 {
2442 if (flags&SWS_PRINT_INFO)
2443 av_log(c, AV_LOG_INFO, "output Width is not a multiple of 32 -> no MMX2 scaler\n");
2444 }
2445 if (usesHFilter) c->canMMX2BeUsed=0;
2446 }
2447 else
2448 c->canMMX2BeUsed=0;
2449
2450 c->chrXInc= ((c->chrSrcW<<16) + (c->chrDstW>>1))/c->chrDstW;
2451 c->chrYInc= ((c->chrSrcH<<16) + (c->chrDstH>>1))/c->chrDstH;
2452
2453 // match pixel 0 of the src to pixel 0 of dst and match pixel n-2 of src to pixel n-2 of dst
2454 // but only for the FAST_BILINEAR mode otherwise do correct scaling
2455 // n-2 is the last chrominance sample available
2456 // this is not perfect, but no one should notice the difference, the more correct variant
2457 // would be like the vertical one, but that would require some special code for the
2458 // first and last pixel
2459 if (flags&SWS_FAST_BILINEAR)
2460 {
2461 if (c->canMMX2BeUsed)
2462 {
2463 c->lumXInc+= 20;
2464 c->chrXInc+= 20;
2465 }
2466 //we don't use the x86asm scaler if mmx is available
2467 else if (flags & SWS_CPU_CAPS_MMX)
2468 {
2469 c->lumXInc = ((srcW-2)<<16)/(dstW-2) - 20;
2470 c->chrXInc = ((c->chrSrcW-2)<<16)/(c->chrDstW-2) - 20;
2471 }
2472 }
2473
2474 /* precalculate horizontal scaler filter coefficients */
2475 {
2476 const int filterAlign=
2477 (flags & SWS_CPU_CAPS_MMX) ? 4 :
2478 (flags & SWS_CPU_CAPS_ALTIVEC) ? 8 :
2479 1;
2480
2481 initFilter(&c->hLumFilter, &c->hLumFilterPos, &c->hLumFilterSize, c->lumXInc,
2482 srcW , dstW, filterAlign, 1<<14,
2483 (flags&SWS_BICUBLIN) ? (flags|SWS_BICUBIC) : flags,
2484 srcFilter->lumH, dstFilter->lumH, c->param);
2485 initFilter(&c->hChrFilter, &c->hChrFilterPos, &c->hChrFilterSize, c->chrXInc,
2486 c->chrSrcW, c->chrDstW, filterAlign, 1<<14,
2487 (flags&SWS_BICUBLIN) ? (flags|SWS_BILINEAR) : flags,
2488 srcFilter->chrH, dstFilter->chrH, c->param);
2489
2490 #define MAX_FUNNY_CODE_SIZE 10000
2491 #if defined(COMPILE_MMX2)
2492 // can't downscale !!!
2493 if (c->canMMX2BeUsed && (flags & SWS_FAST_BILINEAR))
2494 {
2495 #ifdef MAP_ANONYMOUS
2496 c->funnyYCode = (uint8_t*)mmap(NULL, MAX_FUNNY_CODE_SIZE, PROT_EXEC | PROT_READ | PROT_WRITE, MAP_PRIVATE | MAP_ANONYMOUS, 0, 0);
2497 c->funnyUVCode = (uint8_t*)mmap(NULL, MAX_FUNNY_CODE_SIZE, PROT_EXEC | PROT_READ | PROT_WRITE, MAP_PRIVATE | MAP_ANONYMOUS, 0, 0);
2498 #else
2499 c->funnyYCode = av_malloc(MAX_FUNNY_CODE_SIZE);
2500 c->funnyUVCode = av_malloc(MAX_FUNNY_CODE_SIZE);
2501 #endif
2502
2503 c->lumMmx2Filter = av_malloc((dstW /8+8)*sizeof(int16_t));
2504 c->chrMmx2Filter = av_malloc((c->chrDstW /4+8)*sizeof(int16_t));
2505 c->lumMmx2FilterPos= av_malloc((dstW /2/8+8)*sizeof(int32_t));
2506 c->chrMmx2FilterPos= av_malloc((c->chrDstW/2/4+8)*sizeof(int32_t));
2507
2508 initMMX2HScaler( dstW, c->lumXInc, c->funnyYCode , c->lumMmx2Filter, c->lumMmx2FilterPos, 8);
2509 initMMX2HScaler(c->chrDstW, c->chrXInc, c->funnyUVCode, c->chrMmx2Filter, c->chrMmx2FilterPos, 4);
2510 }
2511 #endif /* defined(COMPILE_MMX2) */
2512 } // Init Horizontal stuff
2513
2514
2515
2516 /* precalculate vertical scaler filter coefficients */
2517 {
2518 const int filterAlign=
2519 (flags & SWS_CPU_CAPS_MMX) && (flags & SWS_ACCURATE_RND) ? 2 :
2520 (flags & SWS_CPU_CAPS_ALTIVEC) ? 8 :
2521 1;
2522
2523 initFilter(&c->vLumFilter, &c->vLumFilterPos, &c->vLumFilterSize, c->lumYInc,
2524 srcH , dstH, filterAlign, (1<<12),
2525 (flags&SWS_BICUBLIN) ? (flags|SWS_BICUBIC) : flags,
2526 srcFilter->lumV, dstFilter->lumV, c->param);
2527 initFilter(&c->vChrFilter, &c->vChrFilterPos, &c->vChrFilterSize, c->chrYInc,
2528 c->chrSrcH, c->chrDstH, filterAlign, (1<<12),
2529 (flags&SWS_BICUBLIN) ? (flags|SWS_BILINEAR) : flags,
2530 srcFilter->chrV, dstFilter->chrV, c->param);
2531
2532 #if HAVE_ALTIVEC
2533 c->vYCoeffsBank = av_malloc(sizeof (vector signed short)*c->vLumFilterSize*c->dstH);
2534 c->vCCoeffsBank = av_malloc(sizeof (vector signed short)*c->vChrFilterSize*c->chrDstH);
2535
2536 for (i=0;i<c->vLumFilterSize*c->dstH;i++) {
2537 int j;
2538 short *p = (short *)&c->vYCoeffsBank[i];
2539 for (j=0;j<8;j++)
2540 p[j] = c->vLumFilter[i];
2541 }
2542
2543 for (i=0;i<c->vChrFilterSize*c->chrDstH;i++) {
2544 int j;
2545 short *p = (short *)&c->vCCoeffsBank[i];
2546 for (j=0;j<8;j++)
2547 p[j] = c->vChrFilter[i];
2548 }
2549 #endif
2550 }
2551
2552 // Calculate Buffer Sizes so that they won't run out while handling these damn slices
2553 c->vLumBufSize= c->vLumFilterSize;
2554 c->vChrBufSize= c->vChrFilterSize;
2555 for (i=0; i<dstH; i++)
2556 {
2557 int chrI= i*c->chrDstH / dstH;
2558 int nextSlice= FFMAX(c->vLumFilterPos[i ] + c->vLumFilterSize - 1,
2559 ((c->vChrFilterPos[chrI] + c->vChrFilterSize - 1)<<c->chrSrcVSubSample));
2560
2561 nextSlice>>= c->chrSrcVSubSample;
2562 nextSlice<<= c->chrSrcVSubSample;
2563 if (c->vLumFilterPos[i ] + c->vLumBufSize < nextSlice)
2564 c->vLumBufSize= nextSlice - c->vLumFilterPos[i];
2565 if (c->vChrFilterPos[chrI] + c->vChrBufSize < (nextSlice>>c->chrSrcVSubSample))
2566 c->vChrBufSize= (nextSlice>>c->chrSrcVSubSample) - c->vChrFilterPos[chrI];
2567 }
2568
2569 // allocate pixbufs (we use dynamic allocation because otherwise we would need to
2570 c->lumPixBuf= av_malloc(c->vLumBufSize*2*sizeof(int16_t*));
2571 c->chrPixBuf= av_malloc(c->vChrBufSize*2*sizeof(int16_t*));
2572 //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)
2573 /* align at 16 bytes for AltiVec */
2574 for (i=0; i<c->vLumBufSize; i++)
2575 c->lumPixBuf[i]= c->lumPixBuf[i+c->vLumBufSize]= av_mallocz(VOF+1);
2576 for (i=0; i<c->vChrBufSize; i++)
2577 c->chrPixBuf[i]= c->chrPixBuf[i+c->vChrBufSize]= av_malloc((VOF+1)*2);
2578
2579 //try to avoid drawing green stuff between the right end and the stride end
2580 for (i=0; i<c->vChrBufSize; i++) memset(c->chrPixBuf[i], 64, (VOF+1)*2);
2581
2582 assert(2*VOFW == VOF);
2583
2584 assert(c->chrDstH <= dstH);
2585
2586 if (flags&SWS_PRINT_INFO)
2587 {
2588 #ifdef DITHER1XBPP
2589 const char *dither= " dithered";
2590 #else
2591 const char *dither= "";
2592 #endif
2593 if (flags&SWS_FAST_BILINEAR)
2594 av_log(c, AV_LOG_INFO, "FAST_BILINEAR scaler, ");
2595 else if (flags&SWS_BILINEAR)
2596 av_log(c, AV_LOG_INFO, "BILINEAR scaler, ");
2597 else if (flags&SWS_BICUBIC)
2598 av_log(c, AV_LOG_INFO, "BICUBIC scaler, ");
2599 else if (flags&SWS_X)
2600 av_log(c, AV_LOG_INFO, "Experimental scaler, ");
2601 else if (flags&SWS_POINT)
2602 av_log(c, AV_LOG_INFO, "Nearest Neighbor / POINT scaler, ");
2603 else if (flags&SWS_AREA)
2604 av_log(c, AV_LOG_INFO, "Area Averageing scaler, ");
2605 else if (flags&SWS_BICUBLIN)
2606 av_log(c, AV_LOG_INFO, "luma BICUBIC / chroma BILINEAR scaler, ");
2607 else if (flags&SWS_GAUSS)
2608 av_log(c, AV_LOG_INFO, "Gaussian scaler, ");
2609 else if (flags&SWS_SINC)
2610 av_log(c, AV_LOG_INFO, "Sinc scaler, ");
2611 else if (flags&SWS_LANCZOS)
2612 av_log(c, AV_LOG_INFO, "Lanczos scaler, ");
2613 else if (flags&SWS_SPLINE)
2614 av_log(c, AV_LOG_INFO, "Bicubic spline scaler, ");
2615 else
2616 av_log(c, AV_LOG_INFO, "ehh flags invalid?! ");
2617
2618 if (dstFormat==PIX_FMT_BGR555 || dstFormat==PIX_FMT_BGR565)
2619 av_log(c, AV_LOG_INFO, "from %s to%s %s ",
2620 sws_format_name(srcFormat), dither, sws_format_name(dstFormat));
2621 else
2622 av_log(c, AV_LOG_INFO, "from %s to %s ",
2623 sws_format_name(srcFormat), sws_format_name(dstFormat));
2624
2625 if (flags & SWS_CPU_CAPS_MMX2)
2626 av_log(c, AV_LOG_INFO, "using MMX2\n");
2627 else if (flags & SWS_CPU_CAPS_3DNOW)
2628 av_log(c, AV_LOG_INFO, "using 3DNOW\n");
2629 else if (flags & SWS_CPU_CAPS_MMX)
2630 av_log(c, AV_LOG_INFO, "using MMX\n");
2631 else if (flags & SWS_CPU_CAPS_ALTIVEC)
2632 av_log(c, AV_LOG_INFO, "using AltiVec\n");
2633 else
2634 av_log(c, AV_LOG_INFO, "using C\n");
2635 }
2636
2637 if (flags & SWS_PRINT_INFO)
2638 {
2639 if (flags & SWS_CPU_CAPS_MMX)
2640 {
2641 if (c->canMMX2BeUsed && (flags&SWS_FAST_BILINEAR))
2642 av_log(c, AV_LOG_VERBOSE, "using FAST_BILINEAR MMX2 scaler for horizontal scaling\n");
2643 else
2644 {
2645 if (c->hLumFilterSize==4)
2646 av_log(c, AV_LOG_VERBOSE, "using 4-tap MMX scaler for horizontal luminance scaling\n");
2647 else if (c->hLumFilterSize==8)
2648 av_log(c, AV_LOG_VERBOSE, "using 8-tap MMX scaler for horizontal luminance scaling\n");
2649 else
2650 av_log(c, AV_LOG_VERBOSE, "using n-tap MMX scaler for horizontal luminance scaling\n");
2651
2652 if (c->hChrFilterSize==4)
2653 av_log(c, AV_LOG_VERBOSE, "using 4-tap MMX scaler for horizontal chrominance scaling\n");
2654 else if (c->hChrFilterSize==8)
2655 av_log(c, AV_LOG_VERBOSE, "using 8-tap MMX scaler for horizontal chrominance scaling\n");
2656 else
2657 av_log(c, AV_LOG_VERBOSE, "using n-tap MMX scaler for horizontal chrominance scaling\n");
2658 }
2659 }
2660 else
2661 {
2662 #if ARCH_X86
2663 av_log(c, AV_LOG_VERBOSE, "using X86-Asm scaler for horizontal scaling\n");
2664 #else
2665 if (flags & SWS_FAST_BILINEAR)
2666 av_log(c, AV_LOG_VERBOSE, "using FAST_BILINEAR C scaler for horizontal scaling\n");
2667 else
2668 av_log(c, AV_LOG_VERBOSE, "using C scaler for horizontal scaling\n");
2669 #endif
2670 }
2671 if (isPlanarYUV(dstFormat))
2672 {
2673 if (c->vLumFilterSize==1)
2674 av_log(c, AV_LOG_VERBOSE, "using 1-tap %s \"scaler\" for vertical scaling (YV12 like)\n", (flags & SWS_CPU_CAPS_MMX) ? "MMX" : "C");
2675 else
2676 av_log(c, AV_LOG_VERBOSE, "using n-tap %s scaler for vertical scaling (YV12 like)\n", (flags & SWS_CPU_CAPS_MMX) ? "MMX" : "C");
2677 }
2678 else
2679 {
2680 if (c->vLumFilterSize==1 && c->vChrFilterSize==2)
2681 av_log(c, AV_LOG_VERBOSE, "using 1-tap %s \"scaler\" for vertical luminance scaling (BGR)\n"
2682 " 2-tap scaler for vertical chrominance scaling (BGR)\n", (flags & SWS_CPU_CAPS_MMX) ? "MMX" : "C");
2683 else if (c->vLumFilterSize==2 && c->vChrFilterSize==2)
2684 av_log(c, AV_LOG_VERBOSE, "using 2-tap linear %s scaler for vertical scaling (BGR)\n", (flags & SWS_CPU_CAPS_MMX) ? "MMX" : "C");
2685 else
2686 av_log(c, AV_LOG_VERBOSE, "using n-tap %s scaler for vertical scaling (BGR)\n", (flags & SWS_CPU_CAPS_MMX) ? "MMX" : "C");
2687 }
2688
2689 if (dstFormat==PIX_FMT_BGR24)
2690 av_log(c, AV_LOG_VERBOSE, "using %s YV12->BGR24 Converter\n",
2691 (flags & SWS_CPU_CAPS_MMX2) ? "MMX2" : ((flags & SWS_CPU_CAPS_MMX) ? "MMX" : "C"));
2692 else if (dstFormat==PIX_FMT_RGB32)
2693 av_log(c, AV_LOG_VERBOSE, "using %s YV12->BGR32 Converter\n", (flags & SWS_CPU_CAPS_MMX) ? "MMX" : "C");
2694 else if (dstFormat==PIX_FMT_BGR565)
2695 av_log(c, AV_LOG_VERBOSE, "using %s YV12->BGR16 Converter\n", (flags & SWS_CPU_CAPS_MMX) ? "MMX" : "C");
2696 else if (dstFormat==PIX_FMT_BGR555)
2697 av_log(c, AV_LOG_VERBOSE, "using %s YV12->BGR15 Converter\n", (flags & SWS_CPU_CAPS_MMX) ? "MMX" : "C");
2698
2699 av_log(c, AV_LOG_VERBOSE, "%dx%d -> %dx%d\n", srcW, srcH, dstW, dstH);
2700 }
2701 if (flags & SWS_PRINT_INFO)
2702 {
2703 av_log(c, AV_LOG_DEBUG, "Lum srcW=%d srcH=%d dstW=%d dstH=%d xInc=%d yInc=%d\n",
2704 c->srcW, c->srcH, c->dstW, c->dstH, c->lumXInc, c->lumYInc);
2705 av_log(c, AV_LOG_DEBUG, "Chr srcW=%d srcH=%d dstW=%d dstH=%d xInc=%d yInc=%d\n",
2706 c->chrSrcW, c->chrSrcH, c->chrDstW, c->chrDstH, c->chrXInc, c->chrYInc);
2707 }
2708
2709 c->swScale= getSwsFunc(flags);
2710 return c;
2711 }
2712
2713 /**
2714 * swscale wrapper, so we don't need to export the SwsContext.
2715 * assumes planar YUV to be in YUV order instead of YVU
2716 */
2717 int sws_scale(SwsContext *c, uint8_t* src[], int srcStride[], int srcSliceY,
2718 int srcSliceH, uint8_t* dst[], int dstStride[]){
2719 int i;
2720 uint8_t* src2[4]= {src[0], src[1], src[2]};
2721
2722 if (c->sliceDir == 0 && srcSliceY != 0 && srcSliceY + srcSliceH != c->srcH) {
2723 av_log(c, AV_LOG_ERROR, "Slices start in the middle!\n");
2724 return 0;
2725 }
2726 if (c->sliceDir == 0) {
2727 if (srcSliceY == 0) c->sliceDir = 1; else c->sliceDir = -1;
2728 }
2729
2730 if (usePal(c->srcFormat)){
2731 for (i=0; i<256; i++){
2732 int p, r, g, b,y,u,v;
2733 if(c->srcFormat == PIX_FMT_PAL8){
2734 p=((uint32_t*)(src[1]))[i];
2735 r= (p>>16)&0xFF;
2736 g= (p>> 8)&0xFF;
2737 b= p &0xFF;
2738 }else if(c->srcFormat == PIX_FMT_RGB8){
2739 r= (i>>5 )*36;
2740 g= ((i>>2)&7)*36;
2741 b= (i&3 )*85;
2742 }else if(c->srcFormat == PIX_FMT_BGR8){
2743 b= (i>>6 )*85;
2744 g= ((i>>3)&7)*36;
2745 r= (i&7 )*36;
2746 }else if(c->srcFormat == PIX_FMT_RGB4_BYTE){
2747 r= (i>>3 )*255;
2748 g= ((i>>1)&3)*85;
2749 b= (i&1 )*255;
2750 }else {
2751 assert(c->srcFormat == PIX_FMT_BGR4_BYTE);
2752 b= (i>>3 )*255;
2753 g= ((i>>1)&3)*85;
2754 r= (i&1 )*255;
2755 }
2756 y= av_clip_uint8((RY*r + GY*g + BY*b + ( 33<<(RGB2YUV_SHIFT-1)))>>RGB2YUV_SHIFT);
2757 u= av_clip_uint8((RU*r + GU*g + BU*b + (257<<(RGB2YUV_SHIFT-1)))>>RGB2YUV_SHIFT);
2758 v= av_clip_uint8((RV*r + GV*g + BV*b + (257<<(RGB2YUV_SHIFT-1)))>>RGB2YUV_SHIFT);
2759 c->pal_yuv[i]= y + (u<<8) + (v<<16);
2760
2761
2762 switch(c->dstFormat) {
2763 case PIX_FMT_BGR32:
2764 #ifndef WORDS_BIGENDIAN
2765 case PIX_FMT_RGB24:
2766 #endif
2767 c->pal_rgb[i]= r + (g<<8) + (b<<16);
2768 break;
2769 case PIX_FMT_BGR32_1:
2770 #ifdef WORDS_BIGENDIAN
2771 case PIX_FMT_BGR24:
2772 #endif
2773 c->pal_rgb[i]= (r + (g<<8) + (b<<16)) << 8;
2774 break;
2775 case PIX_FMT_RGB32_1:
2776 #ifdef WORDS_BIGENDIAN
2777 case PIX_FMT_RGB24:
2778 #endif
2779 c->pal_rgb[i]= (b + (g<<8) + (r<<16)) << 8;
2780 break;
2781 case PIX_FMT_RGB32:
2782 #ifndef WORDS_BIGENDIAN
2783 case PIX_FMT_BGR24:
2784 #endif
2785 default:
2786 c->pal_rgb[i]= b + (g<<8) + (r<<16);
2787 }
2788 }
2789 }
2790
2791 // copy strides, so they can safely be modified
2792 if (c->sliceDir == 1) {
2793 // slices go from top to bottom
2794 int srcStride2[4]= {srcStride[0], srcStride[1], srcStride[2]};
2795 int dstStride2[4]= {dstStride[0], dstStride[1], dstStride[2]};
2796 return c->swScale(c, src2, srcStride2, srcSliceY, srcSliceH, dst, dstStride2);
2797 } else {
2798 // slices go from bottom to top => we flip the image internally
2799 uint8_t* dst2[4]= {dst[0] + (c->dstH-1)*dstStride[0],
2800 dst[1] + ((c->dstH>>c->chrDstVSubSample)-1)*dstStride[1],
2801 dst[2] + ((c->dstH>>c->chrDstVSubSample)-1)*dstStride[2]};
2802 int srcStride2[4]= {-srcStride[0], -srcStride[1], -srcStride[2]};
2803 int dstStride2[4]= {-dstStride[0], -dstStride[1], -dstStride[2]};
2804
2805 src2[0] += (srcSliceH-1)*srcStride[0];
2806 if (!usePal(c->srcFormat))
2807 src2[1] += ((srcSliceH>>c->chrSrcVSubSample)-1)*srcStride[1];
2808 src2[2] += ((srcSliceH>>c->chrSrcVSubSample)-1)*srcStride[2];
2809
2810 return c->swScale(c, src2, srcStride2, c->srcH-srcSliceY-srcSliceH, srcSliceH, dst2, dstStride2);
2811 }
2812 }
2813
2814 /**
2815 * swscale wrapper, so we don't need to export the SwsContext
2816 */
2817 int sws_scale_ordered(SwsContext *c, uint8_t* src[], int srcStride[], int srcSliceY,
2818 int srcSliceH, uint8_t* dst[], int dstStride[]){
2819 return sws_scale(c, src, srcStride, srcSliceY, srcSliceH, dst, dstStride);
2820 }
2821
2822 SwsFilter *sws_getDefaultFilter(float lumaGBlur, float chromaGBlur,
2823 float lumaSharpen, float chromaSharpen,
2824 float chromaHShift, float chromaVShift,
2825 int verbose)
2826 {
2827 SwsFilter *filter= av_malloc(sizeof(SwsFilter));
2828
2829 if (lumaGBlur!=0.0){
2830 filter->lumH= sws_getGaussianVec(lumaGBlur, 3.0);
2831 filter->lumV= sws_getGaussianVec(lumaGBlur, 3.0);
2832 }else{
2833 filter->lumH= sws_getIdentityVec();
2834 filter->lumV= sws_getIdentityVec();
2835 }
2836
2837 if (chromaGBlur!=0.0){
2838 filter->chrH= sws_getGaussianVec(chromaGBlur, 3.0);
2839 filter->chrV= sws_getGaussianVec(chromaGBlur, 3.0);
2840 }else{
2841 filter->chrH= sws_getIdentityVec();
2842 filter->chrV= sws_getIdentityVec();
2843 }
2844
2845 if (chromaSharpen!=0.0){
2846 SwsVector *id= sws_getIdentityVec();
2847 sws_scaleVec(filter->chrH, -chromaSharpen);
2848 sws_scaleVec(filter->chrV, -chromaSharpen);
2849 sws_addVec(filter->chrH, id);
2850 sws_addVec(filter->chrV, id);
2851 sws_freeVec(id);
2852 }
2853
2854 if (lumaSharpen!=0.0){
2855 SwsVector *id= sws_getIdentityVec();
2856 sws_scaleVec(filter->lumH, -lumaSharpen);
2857 sws_scaleVec(filter->lumV, -lumaSharpen);
2858 sws_addVec(filter->lumH, id);
2859 sws_addVec(filter->lumV, id);
2860 sws_freeVec(id);
2861 }
2862
2863 if (chromaHShift != 0.0)
2864 sws_shiftVec(filter->chrH, (int)(chromaHShift+0.5));
2865
2866 if (chromaVShift != 0.0)
2867 sws_shiftVec(filter->chrV, (int)(chromaVShift+0.5));
2868
2869 sws_normalizeVec(filter->chrH, 1.0);
2870 sws_normalizeVec(filter->chrV, 1.0);
2871 sws_normalizeVec(filter->lumH, 1.0);
2872 sws_normalizeVec(filter->lumV, 1.0);
2873
2874 if (verbose) sws_printVec(filter->chrH);
2875 if (verbose) sws_printVec(filter->lumH);
2876
2877 return filter;
2878 }
2879
2880 /**
2881 * returns a normalized gaussian curve used to filter stuff
2882 * quality=3 is high quality, lowwer is lowwer quality
2883 */
2884 SwsVector *sws_getGaussianVec(double variance, double quality){
2885 const int length= (int)(variance*quality + 0.5) | 1;
2886 int i;
2887 double *coeff= av_malloc(length*sizeof(double));
2888 double middle= (length-1)*0.5;
2889 SwsVector *vec= av_malloc(sizeof(SwsVector));
2890
2891 vec->coeff= coeff;
2892 vec->length= length;
2893
2894 for (i=0; i<length; i++)
2895 {
2896 double dist= i-middle;
2897 coeff[i]= exp(-dist*dist/(2*variance*variance)) / sqrt(2*variance*PI);
2898 }
2899
2900 sws_normalizeVec(vec, 1.0);
2901
2902 return vec;
2903 }
2904
2905 SwsVector *sws_getConstVec(double c, int length){
2906 int i;
2907 double *coeff= av_malloc(length*sizeof(double));
2908 SwsVector *vec= av_malloc(sizeof(SwsVector));
2909
2910 vec->coeff= coeff;
2911 vec->length= length;
2912
2913 for (i=0; i<length; i++)
2914 coeff[i]= c;
2915
2916 return vec;
2917 }
2918
2919
2920 SwsVector *sws_getIdentityVec(void){
2921 return sws_getConstVec(1.0, 1);
2922 }
2923
2924 double sws_dcVec(SwsVector *a){
2925 int i;
2926 double sum=0;
2927
2928 for (i=0; i<a->length; i++)
2929 sum+= a->coeff[i];
2930
2931 return sum;
2932 }
2933
2934 void sws_scaleVec(SwsVector *a, double scalar){
2935 int i;
2936
2937 for (i=0; i<a->length; i++)
2938 a->coeff[i]*= scalar;
2939 }
2940
2941 void sws_normalizeVec(SwsVector *a, double height){
2942 sws_scaleVec(a, height/sws_dcVec(a));
2943 }
2944
2945 static SwsVector *sws_getConvVec(SwsVector *a, SwsVector *b){
2946 int length= a->length + b->length - 1;
2947 double *coeff= av_malloc(length*sizeof(double));
2948 int i, j;
2949 SwsVector *vec= av_malloc(sizeof(SwsVector));
2950
2951 vec->coeff= coeff;
2952 vec->length= length;
2953
2954 for (i=0; i<length; i++) coeff[i]= 0.0;
2955
2956 for (i=0; i<a->length; i++)
2957 {
2958 for (j=0; j<b->length; j++)
2959 {
2960 coeff[i+j]+= a->coeff[i]*b->coeff[j];
2961 }
2962 }
2963
2964 return vec;
2965 }
2966
2967 static SwsVector *sws_sumVec(SwsVector *a, SwsVector *b){
2968 int length= FFMAX(a->length, b->length);
2969 double *coeff= av_malloc(length*sizeof(double));
2970 int i;
2971 SwsVector *vec= av_malloc(sizeof(SwsVector));
2972
2973 vec->coeff= coeff;
2974 vec->length= length;
2975
2976 for (i=0; i<length; i++) coeff[i]= 0.0;
2977
2978 for (i=0; i<a->length; i++) coeff[i + (length-1)/2 - (a->length-1)/2]+= a->coeff[i];
2979 for (i=0; i<b->length; i++) coeff[i + (length-1)/2 - (b->length-1)/2]+= b->coeff[i];
2980
2981 return vec;