7e25afb6e52bf65ec7f1bf737e8454ccc3519181
[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 #if HAVE_VIRTUALALLOC
72 #define WIN32_LEAN_AND_MEAN
73 #include <windows.h>
74 #endif
75 #include "swscale.h"
76 #include "swscale_internal.h"
77 #include "rgb2rgb.h"
78 #include "libavutil/x86_cpu.h"
79 #include "libavutil/bswap.h"
80
81 unsigned swscale_version(void)
82 {
83 return LIBSWSCALE_VERSION_INT;
84 }
85
86 #undef MOVNTQ
87 #undef PAVGB
88
89 //#undef HAVE_MMX2
90 //#define HAVE_AMD3DNOW
91 //#undef HAVE_MMX
92 //#undef ARCH_X86
93 //#define WORDS_BIGENDIAN
94 #define DITHER1XBPP
95
96 #define FAST_BGR2YV12 // use 7 bit coefficients instead of 15 bit
97
98 #define RET 0xC3 //near return opcode for x86
99
100 #ifdef M_PI
101 #define PI M_PI
102 #else
103 #define PI 3.14159265358979323846
104 #endif
105
106 #define isSupportedIn(x) ( \
107 (x)==PIX_FMT_YUV420P \
108 || (x)==PIX_FMT_YUVA420P \
109 || (x)==PIX_FMT_YUYV422 \
110 || (x)==PIX_FMT_UYVY422 \
111 || (x)==PIX_FMT_RGB32 \
112 || (x)==PIX_FMT_RGB32_1 \
113 || (x)==PIX_FMT_BGR24 \
114 || (x)==PIX_FMT_BGR565 \
115 || (x)==PIX_FMT_BGR555 \
116 || (x)==PIX_FMT_BGR32 \
117 || (x)==PIX_FMT_BGR32_1 \
118 || (x)==PIX_FMT_RGB24 \
119 || (x)==PIX_FMT_RGB565 \
120 || (x)==PIX_FMT_RGB555 \
121 || (x)==PIX_FMT_GRAY8 \
122 || (x)==PIX_FMT_YUV410P \
123 || (x)==PIX_FMT_YUV440P \
124 || (x)==PIX_FMT_GRAY16BE \
125 || (x)==PIX_FMT_GRAY16LE \
126 || (x)==PIX_FMT_YUV444P \
127 || (x)==PIX_FMT_YUV422P \
128 || (x)==PIX_FMT_YUV411P \
129 || (x)==PIX_FMT_PAL8 \
130 || (x)==PIX_FMT_BGR8 \
131 || (x)==PIX_FMT_RGB8 \
132 || (x)==PIX_FMT_BGR4_BYTE \
133 || (x)==PIX_FMT_RGB4_BYTE \
134 || (x)==PIX_FMT_YUV440P \
135 || (x)==PIX_FMT_MONOWHITE \
136 || (x)==PIX_FMT_MONOBLACK \
137 )
138 #define isSupportedOut(x) ( \
139 (x)==PIX_FMT_YUV420P \
140 || (x)==PIX_FMT_YUVA420P \
141 || (x)==PIX_FMT_YUYV422 \
142 || (x)==PIX_FMT_UYVY422 \
143 || (x)==PIX_FMT_YUV444P \
144 || (x)==PIX_FMT_YUV422P \
145 || (x)==PIX_FMT_YUV411P \
146 || isRGB(x) \
147 || isBGR(x) \
148 || (x)==PIX_FMT_NV12 \
149 || (x)==PIX_FMT_NV21 \
150 || (x)==PIX_FMT_GRAY16BE \
151 || (x)==PIX_FMT_GRAY16LE \
152 || (x)==PIX_FMT_GRAY8 \
153 || (x)==PIX_FMT_YUV410P \
154 || (x)==PIX_FMT_YUV440P \
155 )
156 #define isPacked(x) ( \
157 (x)==PIX_FMT_PAL8 \
158 || (x)==PIX_FMT_YUYV422 \
159 || (x)==PIX_FMT_UYVY422 \
160 || isRGB(x) \
161 || isBGR(x) \
162 )
163 #define usePal(x) ( \
164 (x)==PIX_FMT_PAL8 \
165 || (x)==PIX_FMT_BGR4_BYTE \
166 || (x)==PIX_FMT_RGB4_BYTE \
167 || (x)==PIX_FMT_BGR8 \
168 || (x)==PIX_FMT_RGB8 \
169 )
170
171 #define RGB2YUV_SHIFT 15
172 #define BY ( (int)(0.114*219/255*(1<<RGB2YUV_SHIFT)+0.5))
173 #define BV (-(int)(0.081*224/255*(1<<RGB2YUV_SHIFT)+0.5))
174 #define BU ( (int)(0.500*224/255*(1<<RGB2YUV_SHIFT)+0.5))
175 #define GY ( (int)(0.587*219/255*(1<<RGB2YUV_SHIFT)+0.5))
176 #define GV (-(int)(0.419*224/255*(1<<RGB2YUV_SHIFT)+0.5))
177 #define GU (-(int)(0.331*224/255*(1<<RGB2YUV_SHIFT)+0.5))
178 #define RY ( (int)(0.299*219/255*(1<<RGB2YUV_SHIFT)+0.5))
179 #define RV ( (int)(0.500*224/255*(1<<RGB2YUV_SHIFT)+0.5))
180 #define RU (-(int)(0.169*224/255*(1<<RGB2YUV_SHIFT)+0.5))
181
182 extern const int32_t ff_yuv2rgb_coeffs[8][4];
183
184 static const double rgb2yuv_table[8][9]={
185 {0.7152, 0.0722, 0.2126, -0.386, 0.5, -0.115, -0.454, -0.046, 0.5},
186 {0.7152, 0.0722, 0.2126, -0.386, 0.5, -0.115, -0.454, -0.046, 0.5},
187 {0.587 , 0.114 , 0.299 , -0.331, 0.5, -0.169, -0.419, -0.081, 0.5},
188 {0.587 , 0.114 , 0.299 , -0.331, 0.5, -0.169, -0.419, -0.081, 0.5},
189 {0.59 , 0.11 , 0.30 , -0.331, 0.5, -0.169, -0.421, -0.079, 0.5}, //FCC
190 {0.587 , 0.114 , 0.299 , -0.331, 0.5, -0.169, -0.419, -0.081, 0.5},
191 {0.587 , 0.114 , 0.299 , -0.331, 0.5, -0.169, -0.419, -0.081, 0.5}, //SMPTE 170M
192 {0.701 , 0.087 , 0.212 , -0.384, 0.5 -0.116, -0.445, -0.055, 0.5}, //SMPTE 240M
193 };
194
195 /*
196 NOTES
197 Special versions: fast Y 1:1 scaling (no interpolation in y direction)
198
199 TODO
200 more intelligent misalignment avoidance for the horizontal scaler
201 write special vertical cubic upscale version
202 optimize C code (YV12 / minmax)
203 add support for packed pixel YUV input & output
204 add support for Y8 output
205 optimize BGR24 & BGR32
206 add BGR4 output support
207 write special BGR->BGR scaler
208 */
209
210 #if ARCH_X86 && CONFIG_GPL
211 DECLARE_ASM_CONST(8, uint64_t, bF8)= 0xF8F8F8F8F8F8F8F8LL;
212 DECLARE_ASM_CONST(8, uint64_t, bFC)= 0xFCFCFCFCFCFCFCFCLL;
213 DECLARE_ASM_CONST(8, uint64_t, w10)= 0x0010001000100010LL;
214 DECLARE_ASM_CONST(8, uint64_t, w02)= 0x0002000200020002LL;
215 DECLARE_ASM_CONST(8, uint64_t, bm00001111)=0x00000000FFFFFFFFLL;
216 DECLARE_ASM_CONST(8, uint64_t, bm00000111)=0x0000000000FFFFFFLL;
217 DECLARE_ASM_CONST(8, uint64_t, bm11111000)=0xFFFFFFFFFF000000LL;
218 DECLARE_ASM_CONST(8, uint64_t, bm01010101)=0x00FF00FF00FF00FFLL;
219
220 const DECLARE_ALIGNED(8, uint64_t, ff_dither4[2]) = {
221 0x0103010301030103LL,
222 0x0200020002000200LL,};
223
224 const DECLARE_ALIGNED(8, uint64_t, ff_dither8[2]) = {
225 0x0602060206020602LL,
226 0x0004000400040004LL,};
227
228 DECLARE_ASM_CONST(8, uint64_t, b16Mask)= 0x001F001F001F001FLL;
229 DECLARE_ASM_CONST(8, uint64_t, g16Mask)= 0x07E007E007E007E0LL;
230 DECLARE_ASM_CONST(8, uint64_t, r16Mask)= 0xF800F800F800F800LL;
231 DECLARE_ASM_CONST(8, uint64_t, b15Mask)= 0x001F001F001F001FLL;
232 DECLARE_ASM_CONST(8, uint64_t, g15Mask)= 0x03E003E003E003E0LL;
233 DECLARE_ASM_CONST(8, uint64_t, r15Mask)= 0x7C007C007C007C00LL;
234
235 DECLARE_ALIGNED(8, const uint64_t, ff_M24A) = 0x00FF0000FF0000FFLL;
236 DECLARE_ALIGNED(8, const uint64_t, ff_M24B) = 0xFF0000FF0000FF00LL;
237 DECLARE_ALIGNED(8, const uint64_t, ff_M24C) = 0x0000FF0000FF0000LL;
238
239 #ifdef FAST_BGR2YV12
240 DECLARE_ALIGNED(8, const uint64_t, ff_bgr2YCoeff) = 0x000000210041000DULL;
241 DECLARE_ALIGNED(8, const uint64_t, ff_bgr2UCoeff) = 0x0000FFEEFFDC0038ULL;
242 DECLARE_ALIGNED(8, const uint64_t, ff_bgr2VCoeff) = 0x00000038FFD2FFF8ULL;
243 #else
244 DECLARE_ALIGNED(8, const uint64_t, ff_bgr2YCoeff) = 0x000020E540830C8BULL;
245 DECLARE_ALIGNED(8, const uint64_t, ff_bgr2UCoeff) = 0x0000ED0FDAC23831ULL;
246 DECLARE_ALIGNED(8, const uint64_t, ff_bgr2VCoeff) = 0x00003831D0E6F6EAULL;
247 #endif /* FAST_BGR2YV12 */
248 DECLARE_ALIGNED(8, const uint64_t, ff_bgr2YOffset) = 0x1010101010101010ULL;
249 DECLARE_ALIGNED(8, const uint64_t, ff_bgr2UVOffset) = 0x8080808080808080ULL;
250 DECLARE_ALIGNED(8, const uint64_t, ff_w1111) = 0x0001000100010001ULL;
251
252 DECLARE_ASM_CONST(8, uint64_t, ff_bgr24toY1Coeff) = 0x0C88000040870C88ULL;
253 DECLARE_ASM_CONST(8, uint64_t, ff_bgr24toY2Coeff) = 0x20DE4087000020DEULL;
254 DECLARE_ASM_CONST(8, uint64_t, ff_rgb24toY1Coeff) = 0x20DE0000408720DEULL;
255 DECLARE_ASM_CONST(8, uint64_t, ff_rgb24toY2Coeff) = 0x0C88408700000C88ULL;
256 DECLARE_ASM_CONST(8, uint64_t, ff_bgr24toYOffset) = 0x0008400000084000ULL;
257
258 DECLARE_ASM_CONST(8, uint64_t, ff_bgr24toUV[2][4]) = {
259 {0x38380000DAC83838ULL, 0xECFFDAC80000ECFFULL, 0xF6E40000D0E3F6E4ULL, 0x3838D0E300003838ULL},
260 {0xECFF0000DAC8ECFFULL, 0x3838DAC800003838ULL, 0x38380000D0E33838ULL, 0xF6E4D0E30000F6E4ULL},
261 };
262
263 DECLARE_ASM_CONST(8, uint64_t, ff_bgr24toUVOffset)= 0x0040400000404000ULL;
264
265 #endif /* ARCH_X86 && CONFIG_GPL */
266
267 // clipping helper table for C implementations:
268 static unsigned char clip_table[768];
269
270 static SwsVector *sws_getConvVec(SwsVector *a, SwsVector *b);
271
272 static const uint8_t __attribute__((aligned(8))) dither_2x2_4[2][8]={
273 { 1, 3, 1, 3, 1, 3, 1, 3, },
274 { 2, 0, 2, 0, 2, 0, 2, 0, },
275 };
276
277 static const uint8_t __attribute__((aligned(8))) dither_2x2_8[2][8]={
278 { 6, 2, 6, 2, 6, 2, 6, 2, },
279 { 0, 4, 0, 4, 0, 4, 0, 4, },
280 };
281
282 const uint8_t __attribute__((aligned(8))) dither_8x8_32[8][8]={
283 { 17, 9, 23, 15, 16, 8, 22, 14, },
284 { 5, 29, 3, 27, 4, 28, 2, 26, },
285 { 21, 13, 19, 11, 20, 12, 18, 10, },
286 { 0, 24, 6, 30, 1, 25, 7, 31, },
287 { 16, 8, 22, 14, 17, 9, 23, 15, },
288 { 4, 28, 2, 26, 5, 29, 3, 27, },
289 { 20, 12, 18, 10, 21, 13, 19, 11, },
290 { 1, 25, 7, 31, 0, 24, 6, 30, },
291 };
292
293 #if 0
294 const uint8_t __attribute__((aligned(8))) dither_8x8_64[8][8]={
295 { 0, 48, 12, 60, 3, 51, 15, 63, },
296 { 32, 16, 44, 28, 35, 19, 47, 31, },
297 { 8, 56, 4, 52, 11, 59, 7, 55, },
298 { 40, 24, 36, 20, 43, 27, 39, 23, },
299 { 2, 50, 14, 62, 1, 49, 13, 61, },
300 { 34, 18, 46, 30, 33, 17, 45, 29, },
301 { 10, 58, 6, 54, 9, 57, 5, 53, },
302 { 42, 26, 38, 22, 41, 25, 37, 21, },
303 };
304 #endif
305
306 const uint8_t __attribute__((aligned(8))) dither_8x8_73[8][8]={
307 { 0, 55, 14, 68, 3, 58, 17, 72, },
308 { 37, 18, 50, 32, 40, 22, 54, 35, },
309 { 9, 64, 5, 59, 13, 67, 8, 63, },
310 { 46, 27, 41, 23, 49, 31, 44, 26, },
311 { 2, 57, 16, 71, 1, 56, 15, 70, },
312 { 39, 21, 52, 34, 38, 19, 51, 33, },
313 { 11, 66, 7, 62, 10, 65, 6, 60, },
314 { 48, 30, 43, 25, 47, 29, 42, 24, },
315 };
316
317 #if 0
318 const uint8_t __attribute__((aligned(8))) dither_8x8_128[8][8]={
319 { 68, 36, 92, 60, 66, 34, 90, 58, },
320 { 20, 116, 12, 108, 18, 114, 10, 106, },
321 { 84, 52, 76, 44, 82, 50, 74, 42, },
322 { 0, 96, 24, 120, 6, 102, 30, 126, },
323 { 64, 32, 88, 56, 70, 38, 94, 62, },
324 { 16, 112, 8, 104, 22, 118, 14, 110, },
325 { 80, 48, 72, 40, 86, 54, 78, 46, },
326 { 4, 100, 28, 124, 2, 98, 26, 122, },
327 };
328 #endif
329
330 #if 1
331 const uint8_t __attribute__((aligned(8))) dither_8x8_220[8][8]={
332 {117, 62, 158, 103, 113, 58, 155, 100, },
333 { 34, 199, 21, 186, 31, 196, 17, 182, },
334 {144, 89, 131, 76, 141, 86, 127, 72, },
335 { 0, 165, 41, 206, 10, 175, 52, 217, },
336 {110, 55, 151, 96, 120, 65, 162, 107, },
337 { 28, 193, 14, 179, 38, 203, 24, 189, },
338 {138, 83, 124, 69, 148, 93, 134, 79, },
339 { 7, 172, 48, 213, 3, 168, 45, 210, },
340 };
341 #elif 1
342 // tries to correct a gamma of 1.5
343 const uint8_t __attribute__((aligned(8))) dither_8x8_220[8][8]={
344 { 0, 143, 18, 200, 2, 156, 25, 215, },
345 { 78, 28, 125, 64, 89, 36, 138, 74, },
346 { 10, 180, 3, 161, 16, 195, 8, 175, },
347 {109, 51, 93, 38, 121, 60, 105, 47, },
348 { 1, 152, 23, 210, 0, 147, 20, 205, },
349 { 85, 33, 134, 71, 81, 30, 130, 67, },
350 { 14, 190, 6, 171, 12, 185, 5, 166, },
351 {117, 57, 101, 44, 113, 54, 97, 41, },
352 };
353 #elif 1
354 // tries to correct a gamma of 2.0
355 const uint8_t __attribute__((aligned(8))) dither_8x8_220[8][8]={
356 { 0, 124, 8, 193, 0, 140, 12, 213, },
357 { 55, 14, 104, 42, 66, 19, 119, 52, },
358 { 3, 168, 1, 145, 6, 187, 3, 162, },
359 { 86, 31, 70, 21, 99, 39, 82, 28, },
360 { 0, 134, 11, 206, 0, 129, 9, 200, },
361 { 62, 17, 114, 48, 58, 16, 109, 45, },
362 { 5, 181, 2, 157, 4, 175, 1, 151, },
363 { 95, 36, 78, 26, 90, 34, 74, 24, },
364 };
365 #else
366 // tries to correct a gamma of 2.5
367 const uint8_t __attribute__((aligned(8))) dither_8x8_220[8][8]={
368 { 0, 107, 3, 187, 0, 125, 6, 212, },
369 { 39, 7, 86, 28, 49, 11, 102, 36, },
370 { 1, 158, 0, 131, 3, 180, 1, 151, },
371 { 68, 19, 52, 12, 81, 25, 64, 17, },
372 { 0, 119, 5, 203, 0, 113, 4, 195, },
373 { 45, 9, 96, 33, 42, 8, 91, 30, },
374 { 2, 172, 1, 144, 2, 165, 0, 137, },
375 { 77, 23, 60, 15, 72, 21, 56, 14, },
376 };
377 #endif
378
379 const char *sws_format_name(enum PixelFormat format)
380 {
381 switch (format) {
382 case PIX_FMT_YUV420P:
383 return "yuv420p";
384 case PIX_FMT_YUVA420P:
385 return "yuva420p";
386 case PIX_FMT_YUYV422:
387 return "yuyv422";
388 case PIX_FMT_RGB24:
389 return "rgb24";
390 case PIX_FMT_BGR24:
391 return "bgr24";
392 case PIX_FMT_YUV422P:
393 return "yuv422p";
394 case PIX_FMT_YUV444P:
395 return "yuv444p";
396 case PIX_FMT_RGB32:
397 return "rgb32";
398 case PIX_FMT_YUV410P:
399 return "yuv410p";
400 case PIX_FMT_YUV411P:
401 return "yuv411p";
402 case PIX_FMT_RGB565:
403 return "rgb565";
404 case PIX_FMT_RGB555:
405 return "rgb555";
406 case PIX_FMT_GRAY16BE:
407 return "gray16be";
408 case PIX_FMT_GRAY16LE:
409 return "gray16le";
410 case PIX_FMT_GRAY8:
411 return "gray8";
412 case PIX_FMT_MONOWHITE:
413 return "mono white";
414 case PIX_FMT_MONOBLACK:
415 return "mono black";
416 case PIX_FMT_PAL8:
417 return "Palette";
418 case PIX_FMT_YUVJ420P:
419 return "yuvj420p";
420 case PIX_FMT_YUVJ422P:
421 return "yuvj422p";
422 case PIX_FMT_YUVJ444P:
423 return "yuvj444p";
424 case PIX_FMT_XVMC_MPEG2_MC:
425 return "xvmc_mpeg2_mc";
426 case PIX_FMT_XVMC_MPEG2_IDCT:
427 return "xvmc_mpeg2_idct";
428 case PIX_FMT_UYVY422:
429 return "uyvy422";
430 case PIX_FMT_UYYVYY411:
431 return "uyyvyy411";
432 case PIX_FMT_RGB32_1:
433 return "rgb32x";
434 case PIX_FMT_BGR32_1:
435 return "bgr32x";
436 case PIX_FMT_BGR32:
437 return "bgr32";
438 case PIX_FMT_BGR565:
439 return "bgr565";
440 case PIX_FMT_BGR555:
441 return "bgr555";
442 case PIX_FMT_BGR8:
443 return "bgr8";
444 case PIX_FMT_BGR4:
445 return "bgr4";
446 case PIX_FMT_BGR4_BYTE:
447 return "bgr4 byte";
448 case PIX_FMT_RGB8:
449 return "rgb8";
450 case PIX_FMT_RGB4:
451 return "rgb4";
452 case PIX_FMT_RGB4_BYTE:
453 return "rgb4 byte";
454 case PIX_FMT_NV12:
455 return "nv12";
456 case PIX_FMT_NV21:
457 return "nv21";
458 case PIX_FMT_YUV440P:
459 return "yuv440p";
460 case PIX_FMT_VDPAU_H264:
461 return "vdpau_h264";
462 case PIX_FMT_VDPAU_MPEG1:
463 return "vdpau_mpeg1";
464 case PIX_FMT_VDPAU_MPEG2:
465 return "vdpau_mpeg2";
466 case PIX_FMT_VDPAU_WMV3:
467 return "vdpau_wmv3";
468 case PIX_FMT_VDPAU_VC1:
469 return "vdpau_vc1";
470 default:
471 return "Unknown format";
472 }
473 }
474
475 static inline void yuv2yuvXinC(const int16_t *lumFilter, const int16_t **lumSrc, int lumFilterSize,
476 const int16_t *chrFilter, const int16_t **chrSrc, int chrFilterSize,
477 const int16_t **alpSrc, uint8_t *dest, uint8_t *uDest, uint8_t *vDest, uint8_t *aDest, int dstW, int chrDstW)
478 {
479 //FIXME Optimize (just quickly written not optimized..)
480 int i;
481 for (i=0; i<dstW; i++)
482 {
483 int val=1<<18;
484 int j;
485 for (j=0; j<lumFilterSize; j++)
486 val += lumSrc[j][i] * lumFilter[j];
487
488 dest[i]= av_clip_uint8(val>>19);
489 }
490
491 if (uDest)
492 for (i=0; i<chrDstW; i++)
493 {
494 int u=1<<18;
495 int v=1<<18;
496 int j;
497 for (j=0; j<chrFilterSize; j++)
498 {
499 u += chrSrc[j][i] * chrFilter[j];
500 v += chrSrc[j][i + VOFW] * chrFilter[j];
501 }
502
503 uDest[i]= av_clip_uint8(u>>19);
504 vDest[i]= av_clip_uint8(v>>19);
505 }
506
507 if (CONFIG_SWSCALE_ALPHA && aDest)
508 for (i=0; i<dstW; i++){
509 int val=1<<18;
510 int j;
511 for (j=0; j<lumFilterSize; j++)
512 val += alpSrc[j][i] * lumFilter[j];
513
514 aDest[i]= av_clip_uint8(val>>19);
515 }
516
517 }
518
519 static inline void yuv2nv12XinC(const int16_t *lumFilter, const int16_t **lumSrc, int lumFilterSize,
520 const int16_t *chrFilter, const int16_t **chrSrc, int chrFilterSize,
521 uint8_t *dest, uint8_t *uDest, int dstW, int chrDstW, int dstFormat)
522 {
523 //FIXME Optimize (just quickly written not optimized..)
524 int i;
525 for (i=0; i<dstW; i++)
526 {
527 int val=1<<18;
528 int j;
529 for (j=0; j<lumFilterSize; j++)
530 val += lumSrc[j][i] * lumFilter[j];
531
532 dest[i]= av_clip_uint8(val>>19);
533 }
534
535 if (!uDest)
536 return;
537
538 if (dstFormat == PIX_FMT_NV12)
539 for (i=0; i<chrDstW; i++)
540 {
541 int u=1<<18;
542 int v=1<<18;
543 int j;
544 for (j=0; j<chrFilterSize; j++)
545 {
546 u += chrSrc[j][i] * chrFilter[j];
547 v += chrSrc[j][i + VOFW] * chrFilter[j];
548 }
549
550 uDest[2*i]= av_clip_uint8(u>>19);
551 uDest[2*i+1]= av_clip_uint8(v>>19);
552 }
553 else
554 for (i=0; i<chrDstW; i++)
555 {
556 int u=1<<18;
557 int v=1<<18;
558 int j;
559 for (j=0; j<chrFilterSize; j++)
560 {
561 u += chrSrc[j][i] * chrFilter[j];
562 v += chrSrc[j][i + VOFW] * chrFilter[j];
563 }
564
565 uDest[2*i]= av_clip_uint8(v>>19);
566 uDest[2*i+1]= av_clip_uint8(u>>19);
567 }
568 }
569
570 #define YSCALE_YUV_2_PACKEDX_NOCLIP_C(type,alpha) \
571 for (i=0; i<(dstW>>1); i++){\
572 int j;\
573 int Y1 = 1<<18;\
574 int Y2 = 1<<18;\
575 int U = 1<<18;\
576 int V = 1<<18;\
577 int av_unused A1, A2;\
578 type av_unused *r, *b, *g;\
579 const int i2= 2*i;\
580 \
581 for (j=0; j<lumFilterSize; j++)\
582 {\
583 Y1 += lumSrc[j][i2] * lumFilter[j];\
584 Y2 += lumSrc[j][i2+1] * lumFilter[j];\
585 }\
586 for (j=0; j<chrFilterSize; j++)\
587 {\
588 U += chrSrc[j][i] * chrFilter[j];\
589 V += chrSrc[j][i+VOFW] * chrFilter[j];\
590 }\
591 Y1>>=19;\
592 Y2>>=19;\
593 U >>=19;\
594 V >>=19;\
595 if (alpha){\
596 A1 = 1<<18;\
597 A2 = 1<<18;\
598 for (j=0; j<lumFilterSize; j++){\
599 A1 += alpSrc[j][i2 ] * lumFilter[j];\
600 A2 += alpSrc[j][i2+1] * lumFilter[j];\
601 }\
602 A1>>=19;\
603 A2>>=19;\
604 }\
605
606 #define YSCALE_YUV_2_PACKEDX_C(type,alpha) \
607 YSCALE_YUV_2_PACKEDX_NOCLIP_C(type,alpha)\
608 if ((Y1|Y2|U|V)&256)\
609 {\
610 if (Y1>255) Y1=255; \
611 else if (Y1<0)Y1=0; \
612 if (Y2>255) Y2=255; \
613 else if (Y2<0)Y2=0; \
614 if (U>255) U=255; \
615 else if (U<0) U=0; \
616 if (V>255) V=255; \
617 else if (V<0) V=0; \
618 }\
619 if (alpha && ((A1|A2)&256)){\
620 A1=av_clip_uint8(A1);\
621 A2=av_clip_uint8(A2);\
622 }
623
624 #define YSCALE_YUV_2_PACKEDX_FULL_C(rnd,alpha) \
625 for (i=0; i<dstW; i++){\
626 int j;\
627 int Y = 0;\
628 int U = -128<<19;\
629 int V = -128<<19;\
630 int av_unused A;\
631 int R,G,B;\
632 \
633 for (j=0; j<lumFilterSize; j++){\
634 Y += lumSrc[j][i ] * lumFilter[j];\
635 }\
636 for (j=0; j<chrFilterSize; j++){\
637 U += chrSrc[j][i ] * chrFilter[j];\
638 V += chrSrc[j][i+VOFW] * chrFilter[j];\
639 }\
640 Y >>=10;\
641 U >>=10;\
642 V >>=10;\
643 if (alpha){\
644 A = rnd;\
645 for (j=0; j<lumFilterSize; j++)\
646 A += alpSrc[j][i ] * lumFilter[j];\
647 A >>=19;\
648 if (A&256)\
649 A = av_clip_uint8(A);\
650 }\
651
652 #define YSCALE_YUV_2_RGBX_FULL_C(rnd,alpha) \
653 YSCALE_YUV_2_PACKEDX_FULL_C(rnd>>3,alpha)\
654 Y-= c->yuv2rgb_y_offset;\
655 Y*= c->yuv2rgb_y_coeff;\
656 Y+= rnd;\
657 R= Y + V*c->yuv2rgb_v2r_coeff;\
658 G= Y + V*c->yuv2rgb_v2g_coeff + U*c->yuv2rgb_u2g_coeff;\
659 B= Y + U*c->yuv2rgb_u2b_coeff;\
660 if ((R|G|B)&(0xC0000000)){\
661 if (R>=(256<<22)) R=(256<<22)-1; \
662 else if (R<0)R=0; \
663 if (G>=(256<<22)) G=(256<<22)-1; \
664 else if (G<0)G=0; \
665 if (B>=(256<<22)) B=(256<<22)-1; \
666 else if (B<0)B=0; \
667 }\
668
669
670 #define YSCALE_YUV_2_GRAY16_C \
671 for (i=0; i<(dstW>>1); i++){\
672 int j;\
673 int Y1 = 1<<18;\
674 int Y2 = 1<<18;\
675 int U = 1<<18;\
676 int V = 1<<18;\
677 \
678 const int i2= 2*i;\
679 \
680 for (j=0; j<lumFilterSize; j++)\
681 {\
682 Y1 += lumSrc[j][i2] * lumFilter[j];\
683 Y2 += lumSrc[j][i2+1] * lumFilter[j];\
684 }\
685 Y1>>=11;\
686 Y2>>=11;\
687 if ((Y1|Y2|U|V)&65536)\
688 {\
689 if (Y1>65535) Y1=65535; \
690 else if (Y1<0)Y1=0; \
691 if (Y2>65535) Y2=65535; \
692 else if (Y2<0)Y2=0; \
693 }
694
695 #define YSCALE_YUV_2_RGBX_C(type,alpha) \
696 YSCALE_YUV_2_PACKEDX_C(type,alpha) /* FIXME fix tables so that clipping is not needed and then use _NOCLIP*/\
697 r = (type *)c->table_rV[V]; \
698 g = (type *)(c->table_gU[U] + c->table_gV[V]); \
699 b = (type *)c->table_bU[U]; \
700
701 #define YSCALE_YUV_2_PACKED2_C(type,alpha) \
702 for (i=0; i<(dstW>>1); i++){ \
703 const int i2= 2*i; \
704 int Y1= (buf0[i2 ]*yalpha1+buf1[i2 ]*yalpha)>>19; \
705 int Y2= (buf0[i2+1]*yalpha1+buf1[i2+1]*yalpha)>>19; \
706 int U= (uvbuf0[i ]*uvalpha1+uvbuf1[i ]*uvalpha)>>19; \
707 int V= (uvbuf0[i+VOFW]*uvalpha1+uvbuf1[i+VOFW]*uvalpha)>>19; \
708 type av_unused *r, *b, *g; \
709 int av_unused A1, A2; \
710 if (alpha){\
711 A1= (abuf0[i2 ]*yalpha1+abuf1[i2 ]*yalpha)>>19; \
712 A2= (abuf0[i2+1]*yalpha1+abuf1[i2+1]*yalpha)>>19; \
713 }\
714
715 #define YSCALE_YUV_2_GRAY16_2_C \
716 for (i=0; i<(dstW>>1); i++){ \
717 const int i2= 2*i; \
718 int Y1= (buf0[i2 ]*yalpha1+buf1[i2 ]*yalpha)>>11; \
719 int Y2= (buf0[i2+1]*yalpha1+buf1[i2+1]*yalpha)>>11; \
720
721 #define YSCALE_YUV_2_RGB2_C(type,alpha) \
722 YSCALE_YUV_2_PACKED2_C(type,alpha)\
723 r = (type *)c->table_rV[V];\
724 g = (type *)(c->table_gU[U] + c->table_gV[V]);\
725 b = (type *)c->table_bU[U];\
726
727 #define YSCALE_YUV_2_PACKED1_C(type,alpha) \
728 for (i=0; i<(dstW>>1); i++){\
729 const int i2= 2*i;\
730 int Y1= buf0[i2 ]>>7;\
731 int Y2= buf0[i2+1]>>7;\
732 int U= (uvbuf1[i ])>>7;\
733 int V= (uvbuf1[i+VOFW])>>7;\
734 type av_unused *r, *b, *g;\
735 int av_unused A1, A2;\
736 if (alpha){\
737 A1= abuf0[i2 ]>>7;\
738 A2= abuf0[i2+1]>>7;\
739 }\
740
741 #define YSCALE_YUV_2_GRAY16_1_C \
742 for (i=0; i<(dstW>>1); i++){\
743 const int i2= 2*i;\
744 int Y1= buf0[i2 ]<<1;\
745 int Y2= buf0[i2+1]<<1;\
746
747 #define YSCALE_YUV_2_RGB1_C(type,alpha) \
748 YSCALE_YUV_2_PACKED1_C(type,alpha)\
749 r = (type *)c->table_rV[V];\
750 g = (type *)(c->table_gU[U] + c->table_gV[V]);\
751 b = (type *)c->table_bU[U];\
752
753 #define YSCALE_YUV_2_PACKED1B_C(type,alpha) \
754 for (i=0; i<(dstW>>1); i++){\
755 const int i2= 2*i;\
756 int Y1= buf0[i2 ]>>7;\
757 int Y2= buf0[i2+1]>>7;\
758 int U= (uvbuf0[i ] + uvbuf1[i ])>>8;\
759 int V= (uvbuf0[i+VOFW] + uvbuf1[i+VOFW])>>8;\
760 type av_unused *r, *b, *g;\
761 int av_unused A1, A2;\
762 if (alpha){\
763 A1= abuf0[i2 ]>>7;\
764 A2= abuf0[i2+1]>>7;\
765 }\
766
767 #define YSCALE_YUV_2_RGB1B_C(type,alpha) \
768 YSCALE_YUV_2_PACKED1B_C(type,alpha)\
769 r = (type *)c->table_rV[V];\
770 g = (type *)(c->table_gU[U] + c->table_gV[V]);\
771 b = (type *)c->table_bU[U];\
772
773 #define YSCALE_YUV_2_MONO2_C \
774 const uint8_t * const d128=dither_8x8_220[y&7];\
775 uint8_t *g= c->table_gU[128] + c->table_gV[128];\
776 for (i=0; i<dstW-7; i+=8){\
777 int acc;\
778 acc = g[((buf0[i ]*yalpha1+buf1[i ]*yalpha)>>19) + d128[0]];\
779 acc+= acc + g[((buf0[i+1]*yalpha1+buf1[i+1]*yalpha)>>19) + d128[1]];\
780 acc+= acc + g[((buf0[i+2]*yalpha1+buf1[i+2]*yalpha)>>19) + d128[2]];\
781 acc+= acc + g[((buf0[i+3]*yalpha1+buf1[i+3]*yalpha)>>19) + d128[3]];\
782 acc+= acc + g[((buf0[i+4]*yalpha1+buf1[i+4]*yalpha)>>19) + d128[4]];\
783 acc+= acc + g[((buf0[i+5]*yalpha1+buf1[i+5]*yalpha)>>19) + d128[5]];\
784 acc+= acc + g[((buf0[i+6]*yalpha1+buf1[i+6]*yalpha)>>19) + d128[6]];\
785 acc+= acc + g[((buf0[i+7]*yalpha1+buf1[i+7]*yalpha)>>19) + d128[7]];\
786 ((uint8_t*)dest)[0]= c->dstFormat == PIX_FMT_MONOBLACK ? acc : ~acc;\
787 dest++;\
788 }\
789
790
791 #define YSCALE_YUV_2_MONOX_C \
792 const uint8_t * const d128=dither_8x8_220[y&7];\
793 uint8_t *g= c->table_gU[128] + c->table_gV[128];\
794 int acc=0;\
795 for (i=0; i<dstW-1; i+=2){\
796 int j;\
797 int Y1=1<<18;\
798 int Y2=1<<18;\
799 \
800 for (j=0; j<lumFilterSize; j++)\
801 {\
802 Y1 += lumSrc[j][i] * lumFilter[j];\
803 Y2 += lumSrc[j][i+1] * lumFilter[j];\
804 }\
805 Y1>>=19;\
806 Y2>>=19;\
807 if ((Y1|Y2)&256)\
808 {\
809 if (Y1>255) Y1=255;\
810 else if (Y1<0)Y1=0;\
811 if (Y2>255) Y2=255;\
812 else if (Y2<0)Y2=0;\
813 }\
814 acc+= acc + g[Y1+d128[(i+0)&7]];\
815 acc+= acc + g[Y2+d128[(i+1)&7]];\
816 if ((i&7)==6){\
817 ((uint8_t*)dest)[0]= c->dstFormat == PIX_FMT_MONOBLACK ? acc : ~acc;\
818 dest++;\
819 }\
820 }
821
822
823 #define YSCALE_YUV_2_ANYRGB_C(func, func2, func_g16, func_monoblack)\
824 switch(c->dstFormat)\
825 {\
826 case PIX_FMT_RGBA:\
827 case PIX_FMT_BGRA:\
828 if (CONFIG_SMALL){\
829 int needAlpha = CONFIG_SWSCALE_ALPHA && c->alpPixBuf;\
830 func(uint32_t,needAlpha)\
831 ((uint32_t*)dest)[i2+0]= r[Y1] + g[Y1] + b[Y1] + (needAlpha ? (A1<<24) : 0);\
832 ((uint32_t*)dest)[i2+1]= r[Y2] + g[Y2] + b[Y2] + (needAlpha ? (A2<<24) : 0);\
833 }\
834 }else{\
835 if (CONFIG_SWSCALE_ALPHA && c->alpPixBuf){\
836 func(uint32_t,1)\
837 ((uint32_t*)dest)[i2+0]= r[Y1] + g[Y1] + b[Y1] + (A1<<24);\
838 ((uint32_t*)dest)[i2+1]= r[Y2] + g[Y2] + b[Y2] + (A2<<24);\
839 }\
840 }else{\
841 func(uint32_t,0)\
842 ((uint32_t*)dest)[i2+0]= r[Y1] + g[Y1] + b[Y1];\
843 ((uint32_t*)dest)[i2+1]= r[Y2] + g[Y2] + b[Y2];\
844 }\
845 }\
846 }\
847 break;\
848 case PIX_FMT_ARGB:\
849 case PIX_FMT_ABGR:\
850 if (CONFIG_SMALL){\
851 int needAlpha = CONFIG_SWSCALE_ALPHA && c->alpPixBuf;\
852 func(uint32_t,needAlpha)\
853 ((uint32_t*)dest)[i2+0]= r[Y1] + g[Y1] + b[Y1] + (needAlpha ? A1 : 0);\
854 ((uint32_t*)dest)[i2+1]= r[Y2] + g[Y2] + b[Y2] + (needAlpha ? A2 : 0);\
855 }\
856 }else{\
857 if (CONFIG_SWSCALE_ALPHA && c->alpPixBuf){\
858 func(uint32_t,1)\
859 ((uint32_t*)dest)[i2+0]= r[Y1] + g[Y1] + b[Y1] + A1;\
860 ((uint32_t*)dest)[i2+1]= r[Y2] + g[Y2] + b[Y2] + A2;\
861 }\
862 }else{\
863 func(uint32_t,0)\
864 ((uint32_t*)dest)[i2+0]= r[Y1] + g[Y1] + b[Y1];\
865 ((uint32_t*)dest)[i2+1]= r[Y2] + g[Y2] + b[Y2];\
866 }\
867 }\
868 } \
869 break;\
870 case PIX_FMT_RGB24:\
871 func(uint8_t,0)\
872 ((uint8_t*)dest)[0]= r[Y1];\
873 ((uint8_t*)dest)[1]= g[Y1];\
874 ((uint8_t*)dest)[2]= b[Y1];\
875 ((uint8_t*)dest)[3]= r[Y2];\
876 ((uint8_t*)dest)[4]= g[Y2];\
877 ((uint8_t*)dest)[5]= b[Y2];\
878 dest+=6;\
879 }\
880 break;\
881 case PIX_FMT_BGR24:\
882 func(uint8_t,0)\
883 ((uint8_t*)dest)[0]= b[Y1];\
884 ((uint8_t*)dest)[1]= g[Y1];\
885 ((uint8_t*)dest)[2]= r[Y1];\
886 ((uint8_t*)dest)[3]= b[Y2];\
887 ((uint8_t*)dest)[4]= g[Y2];\
888 ((uint8_t*)dest)[5]= r[Y2];\
889 dest+=6;\
890 }\
891 break;\
892 case PIX_FMT_RGB565:\
893 case PIX_FMT_BGR565:\
894 {\
895 const int dr1= dither_2x2_8[y&1 ][0];\
896 const int dg1= dither_2x2_4[y&1 ][0];\
897 const int db1= dither_2x2_8[(y&1)^1][0];\
898 const int dr2= dither_2x2_8[y&1 ][1];\
899 const int dg2= dither_2x2_4[y&1 ][1];\
900 const int db2= dither_2x2_8[(y&1)^1][1];\
901 func(uint16_t,0)\
902 ((uint16_t*)dest)[i2+0]= r[Y1+dr1] + g[Y1+dg1] + b[Y1+db1];\
903 ((uint16_t*)dest)[i2+1]= r[Y2+dr2] + g[Y2+dg2] + b[Y2+db2];\
904 }\
905 }\
906 break;\
907 case PIX_FMT_RGB555:\
908 case PIX_FMT_BGR555:\
909 {\
910 const int dr1= dither_2x2_8[y&1 ][0];\
911 const int dg1= dither_2x2_8[y&1 ][1];\
912 const int db1= dither_2x2_8[(y&1)^1][0];\
913 const int dr2= dither_2x2_8[y&1 ][1];\
914 const int dg2= dither_2x2_8[y&1 ][0];\
915 const int db2= dither_2x2_8[(y&1)^1][1];\
916 func(uint16_t,0)\
917 ((uint16_t*)dest)[i2+0]= r[Y1+dr1] + g[Y1+dg1] + b[Y1+db1];\
918 ((uint16_t*)dest)[i2+1]= r[Y2+dr2] + g[Y2+dg2] + b[Y2+db2];\
919 }\
920 }\
921 break;\
922 case PIX_FMT_RGB8:\
923 case PIX_FMT_BGR8:\
924 {\
925 const uint8_t * const d64= dither_8x8_73[y&7];\
926 const uint8_t * const d32= dither_8x8_32[y&7];\
927 func(uint8_t,0)\
928 ((uint8_t*)dest)[i2+0]= r[Y1+d32[(i2+0)&7]] + g[Y1+d32[(i2+0)&7]] + b[Y1+d64[(i2+0)&7]];\
929 ((uint8_t*)dest)[i2+1]= r[Y2+d32[(i2+1)&7]] + g[Y2+d32[(i2+1)&7]] + b[Y2+d64[(i2+1)&7]];\
930 }\
931 }\
932 break;\
933 case PIX_FMT_RGB4:\
934 case PIX_FMT_BGR4:\
935 {\
936 const uint8_t * const d64= dither_8x8_73 [y&7];\
937 const uint8_t * const d128=dither_8x8_220[y&7];\
938 func(uint8_t,0)\
939 ((uint8_t*)dest)[i]= r[Y1+d128[(i2+0)&7]] + g[Y1+d64[(i2+0)&7]] + b[Y1+d128[(i2+0)&7]]\
940 + ((r[Y2+d128[(i2+1)&7]] + g[Y2+d64[(i2+1)&7]] + b[Y2+d128[(i2+1)&7]])<<4);\
941 }\
942 }\
943 break;\
944 case PIX_FMT_RGB4_BYTE:\
945 case PIX_FMT_BGR4_BYTE:\
946 {\
947 const uint8_t * const d64= dither_8x8_73 [y&7];\
948 const uint8_t * const d128=dither_8x8_220[y&7];\
949 func(uint8_t,0)\
950 ((uint8_t*)dest)[i2+0]= r[Y1+d128[(i2+0)&7]] + g[Y1+d64[(i2+0)&7]] + b[Y1+d128[(i2+0)&7]];\
951 ((uint8_t*)dest)[i2+1]= r[Y2+d128[(i2+1)&7]] + g[Y2+d64[(i2+1)&7]] + b[Y2+d128[(i2+1)&7]];\
952 }\
953 }\
954 break;\
955 case PIX_FMT_MONOBLACK:\
956 case PIX_FMT_MONOWHITE:\
957 {\
958 func_monoblack\
959 }\
960 break;\
961 case PIX_FMT_YUYV422:\
962 func2\
963 ((uint8_t*)dest)[2*i2+0]= Y1;\
964 ((uint8_t*)dest)[2*i2+1]= U;\
965 ((uint8_t*)dest)[2*i2+2]= Y2;\
966 ((uint8_t*)dest)[2*i2+3]= V;\
967 } \
968 break;\
969 case PIX_FMT_UYVY422:\
970 func2\
971 ((uint8_t*)dest)[2*i2+0]= U;\
972 ((uint8_t*)dest)[2*i2+1]= Y1;\
973 ((uint8_t*)dest)[2*i2+2]= V;\
974 ((uint8_t*)dest)[2*i2+3]= Y2;\
975 } \
976 break;\
977 case PIX_FMT_GRAY16BE:\
978 func_g16\
979 ((uint8_t*)dest)[2*i2+0]= Y1>>8;\
980 ((uint8_t*)dest)[2*i2+1]= Y1;\
981 ((uint8_t*)dest)[2*i2+2]= Y2>>8;\
982 ((uint8_t*)dest)[2*i2+3]= Y2;\
983 } \
984 break;\
985 case PIX_FMT_GRAY16LE:\
986 func_g16\
987 ((uint8_t*)dest)[2*i2+0]= Y1;\
988 ((uint8_t*)dest)[2*i2+1]= Y1>>8;\
989 ((uint8_t*)dest)[2*i2+2]= Y2;\
990 ((uint8_t*)dest)[2*i2+3]= Y2>>8;\
991 } \
992 break;\
993 }\
994
995
996 static inline void yuv2packedXinC(SwsContext *c, const int16_t *lumFilter, const int16_t **lumSrc, int lumFilterSize,
997 const int16_t *chrFilter, const int16_t **chrSrc, int chrFilterSize,
998 const int16_t **alpSrc, uint8_t *dest, int dstW, int y)
999 {
1000 int i;
1001 YSCALE_YUV_2_ANYRGB_C(YSCALE_YUV_2_RGBX_C, YSCALE_YUV_2_PACKEDX_C(void,0), YSCALE_YUV_2_GRAY16_C, YSCALE_YUV_2_MONOX_C)
1002 }
1003
1004 static inline void yuv2rgbXinC_full(SwsContext *c, const int16_t *lumFilter, const int16_t **lumSrc, int lumFilterSize,
1005 const int16_t *chrFilter, const int16_t **chrSrc, int chrFilterSize,
1006 const int16_t **alpSrc, uint8_t *dest, int dstW, int y)
1007 {
1008 int i;
1009 int step= fmt_depth(c->dstFormat)/8;
1010 int aidx= 3;
1011
1012 switch(c->dstFormat){
1013 case PIX_FMT_ARGB:
1014 dest++;
1015 aidx= 0;
1016 case PIX_FMT_RGB24:
1017 aidx--;
1018 case PIX_FMT_RGBA:
1019 if (CONFIG_SMALL){
1020 int needAlpha = CONFIG_SWSCALE_ALPHA && c->alpPixBuf;
1021 YSCALE_YUV_2_RGBX_FULL_C(1<<21, needAlpha)
1022 dest[aidx]= needAlpha ? A : 255;
1023 dest[0]= R>>22;
1024 dest[1]= G>>22;
1025 dest[2]= B>>22;
1026 dest+= step;
1027 }
1028 }else{
1029 if (CONFIG_SWSCALE_ALPHA && c->alpPixBuf){
1030 YSCALE_YUV_2_RGBX_FULL_C(1<<21, 1)
1031 dest[aidx]= A;
1032 dest[0]= R>>22;
1033 dest[1]= G>>22;
1034 dest[2]= B>>22;
1035 dest+= step;
1036 }
1037 }else{
1038 YSCALE_YUV_2_RGBX_FULL_C(1<<21, 0)
1039 dest[aidx]= 255;
1040 dest[0]= R>>22;
1041 dest[1]= G>>22;
1042 dest[2]= B>>22;
1043 dest+= step;
1044 }
1045 }
1046 }
1047 break;
1048 case PIX_FMT_ABGR:
1049 dest++;
1050 aidx= 0;
1051 case PIX_FMT_BGR24:
1052 aidx--;
1053 case PIX_FMT_BGRA:
1054 if (CONFIG_SMALL){
1055 int needAlpha = CONFIG_SWSCALE_ALPHA && c->alpPixBuf;
1056 YSCALE_YUV_2_RGBX_FULL_C(1<<21, needAlpha)
1057 dest[aidx]= needAlpha ? A : 255;
1058 dest[0]= B>>22;
1059 dest[1]= G>>22;
1060 dest[2]= R>>22;
1061 dest+= step;
1062 }
1063 }else{
1064 if (CONFIG_SWSCALE_ALPHA && c->alpPixBuf){
1065 YSCALE_YUV_2_RGBX_FULL_C(1<<21, 1)
1066 dest[aidx]= A;
1067 dest[0]= B>>22;
1068 dest[1]= G>>22;
1069 dest[2]= R>>22;
1070 dest+= step;
1071 }
1072 }else{
1073 YSCALE_YUV_2_RGBX_FULL_C(1<<21, 0)
1074 dest[aidx]= 255;
1075 dest[0]= B>>22;
1076 dest[1]= G>>22;
1077 dest[2]= R>>22;
1078 dest+= step;
1079 }
1080 }
1081 }
1082 break;
1083 default:
1084 assert(0);
1085 }
1086 }
1087
1088 static void fillPlane(uint8_t* plane, int stride, int width, int height, int y, uint8_t val){
1089 int i;
1090 uint8_t *ptr = plane + stride*y;
1091 for (i=0; i<height; i++){
1092 memset(ptr, val, width);
1093 ptr += stride;
1094 }
1095 }
1096
1097 //Note: we have C, X86, MMX, MMX2, 3DNOW versions, there is no 3DNOW+MMX2 one
1098 //Plain C versions
1099 #if !HAVE_MMX || defined (RUNTIME_CPUDETECT) || !CONFIG_GPL
1100 #define COMPILE_C
1101 #endif
1102
1103 #if ARCH_PPC
1104 #if (HAVE_ALTIVEC || defined (RUNTIME_CPUDETECT)) && CONFIG_GPL
1105 #undef COMPILE_C
1106 #define COMPILE_ALTIVEC
1107 #endif
1108 #endif //ARCH_PPC
1109
1110 #if ARCH_X86
1111
1112 #if ((HAVE_MMX && !HAVE_AMD3DNOW && !HAVE_MMX2) || defined (RUNTIME_CPUDETECT)) && CONFIG_GPL
1113 #define COMPILE_MMX
1114 #endif
1115
1116 #if (HAVE_MMX2 || defined (RUNTIME_CPUDETECT)) && CONFIG_GPL
1117 #define COMPILE_MMX2
1118 #endif
1119
1120 #if ((HAVE_AMD3DNOW && !HAVE_MMX2) || defined (RUNTIME_CPUDETECT)) && CONFIG_GPL
1121 #define COMPILE_3DNOW
1122 #endif
1123 #endif //ARCH_X86
1124
1125 #undef HAVE_MMX
1126 #undef HAVE_MMX2
1127 #undef HAVE_AMD3DNOW
1128 #undef HAVE_ALTIVEC
1129 #define HAVE_MMX 0
1130 #define HAVE_MMX2 0
1131 #define HAVE_AMD3DNOW 0
1132 #define HAVE_ALTIVEC 0
1133
1134 #ifdef COMPILE_C
1135 #define RENAME(a) a ## _C
1136 #include "swscale_template.c"
1137 #endif
1138
1139 #ifdef COMPILE_ALTIVEC
1140 #undef RENAME
1141 #undef HAVE_ALTIVEC
1142 #define HAVE_ALTIVEC 1
1143 #define RENAME(a) a ## _altivec
1144 #include "swscale_template.c"
1145 #endif
1146
1147 #if ARCH_X86
1148
1149 //x86 versions
1150 /*
1151 #undef RENAME
1152 #undef HAVE_MMX
1153 #undef HAVE_MMX2
1154 #undef HAVE_AMD3DNOW
1155 #define ARCH_X86
1156 #define RENAME(a) a ## _X86
1157 #include "swscale_template.c"
1158 */
1159 //MMX versions
1160 #ifdef COMPILE_MMX
1161 #undef RENAME
1162 #undef HAVE_MMX
1163 #undef HAVE_MMX2
1164 #undef HAVE_AMD3DNOW
1165 #define HAVE_MMX 1
1166 #define HAVE_MMX2 0
1167 #define HAVE_AMD3DNOW 0
1168 #define RENAME(a) a ## _MMX
1169 #include "swscale_template.c"
1170 #endif
1171
1172 //MMX2 versions
1173 #ifdef COMPILE_MMX2
1174 #undef RENAME
1175 #undef HAVE_MMX
1176 #undef HAVE_MMX2
1177 #undef HAVE_AMD3DNOW
1178 #define HAVE_MMX 1
1179 #define HAVE_MMX2 1
1180 #define HAVE_AMD3DNOW 0
1181 #define RENAME(a) a ## _MMX2
1182 #include "swscale_template.c"
1183 #endif
1184
1185 //3DNOW versions
1186 #ifdef COMPILE_3DNOW
1187 #undef RENAME
1188 #undef HAVE_MMX
1189 #undef HAVE_MMX2
1190 #undef HAVE_AMD3DNOW
1191 #define HAVE_MMX 1
1192 #define HAVE_MMX2 0
1193 #define HAVE_AMD3DNOW 1
1194 #define RENAME(a) a ## _3DNow
1195 #include "swscale_template.c"
1196 #endif
1197
1198 #endif //ARCH_X86
1199
1200 // minor note: the HAVE_xyz are messed up after this line so don't use them
1201
1202 static double getSplineCoeff(double a, double b, double c, double d, double dist)
1203 {
1204 // printf("%f %f %f %f %f\n", a,b,c,d,dist);
1205 if (dist<=1.0) return ((d*dist + c)*dist + b)*dist +a;
1206 else return getSplineCoeff( 0.0,
1207 b+ 2.0*c + 3.0*d,
1208 c + 3.0*d,
1209 -b- 3.0*c - 6.0*d,
1210 dist-1.0);
1211 }
1212
1213 static inline int initFilter(int16_t **outFilter, int16_t **filterPos, int *outFilterSize, int xInc,
1214 int srcW, int dstW, int filterAlign, int one, int flags,
1215 SwsVector *srcFilter, SwsVector *dstFilter, double param[2])
1216 {
1217 int i;
1218 int filterSize;
1219 int filter2Size;
1220 int minFilterSize;
1221 int64_t *filter=NULL;
1222 int64_t *filter2=NULL;
1223 const int64_t fone= 1LL<<54;
1224 int ret= -1;
1225 #if ARCH_X86
1226 if (flags & SWS_CPU_CAPS_MMX)
1227 __asm__ volatile("emms\n\t"::: "memory"); //FIXME this should not be required but it IS (even for non-MMX versions)
1228 #endif
1229
1230 // NOTE: the +1 is for the MMX scaler which reads over the end
1231 *filterPos = av_malloc((dstW+1)*sizeof(int16_t));
1232
1233 if (FFABS(xInc - 0x10000) <10) // unscaled
1234 {
1235 int i;
1236 filterSize= 1;
1237 filter= av_mallocz(dstW*sizeof(*filter)*filterSize);
1238
1239 for (i=0; i<dstW; i++)
1240 {
1241 filter[i*filterSize]= fone;
1242 (*filterPos)[i]=i;
1243 }
1244
1245 }
1246 else if (flags&SWS_POINT) // lame looking point sampling mode
1247 {
1248 int i;
1249 int xDstInSrc;
1250 filterSize= 1;
1251 filter= av_malloc(dstW*sizeof(*filter)*filterSize);
1252
1253 xDstInSrc= xInc/2 - 0x8000;
1254 for (i=0; i<dstW; i++)
1255 {
1256 int xx= (xDstInSrc - ((filterSize-1)<<15) + (1<<15))>>16;
1257
1258 (*filterPos)[i]= xx;
1259 filter[i]= fone;
1260 xDstInSrc+= xInc;
1261 }
1262 }
1263 else if ((xInc <= (1<<16) && (flags&SWS_AREA)) || (flags&SWS_FAST_BILINEAR)) // bilinear upscale
1264 {
1265 int i;
1266 int xDstInSrc;
1267 if (flags&SWS_BICUBIC) filterSize= 4;
1268 else if (flags&SWS_X ) filterSize= 4;
1269 else filterSize= 2; // SWS_BILINEAR / SWS_AREA
1270 filter= av_malloc(dstW*sizeof(*filter)*filterSize);
1271
1272 xDstInSrc= xInc/2 - 0x8000;
1273 for (i=0; i<dstW; i++)
1274 {
1275 int xx= (xDstInSrc - ((filterSize-1)<<15) + (1<<15))>>16;
1276 int j;
1277
1278 (*filterPos)[i]= xx;
1279 //bilinear upscale / linear interpolate / area averaging
1280 for (j=0; j<filterSize; j++)
1281 {
1282 int64_t coeff= fone - FFABS((xx<<16) - xDstInSrc)*(fone>>16);
1283 if (coeff<0) coeff=0;
1284 filter[i*filterSize + j]= coeff;
1285 xx++;
1286 }
1287 xDstInSrc+= xInc;
1288 }
1289 }
1290 else
1291 {
1292 int xDstInSrc;
1293 int sizeFactor;
1294
1295 if (flags&SWS_BICUBIC) sizeFactor= 4;
1296 else if (flags&SWS_X) sizeFactor= 8;
1297 else if (flags&SWS_AREA) sizeFactor= 1; //downscale only, for upscale it is bilinear
1298 else if (flags&SWS_GAUSS) sizeFactor= 8; // infinite ;)
1299 else if (flags&SWS_LANCZOS) sizeFactor= param[0] != SWS_PARAM_DEFAULT ? ceil(2*param[0]) : 6;
1300 else if (flags&SWS_SINC) sizeFactor= 20; // infinite ;)
1301 else if (flags&SWS_SPLINE) sizeFactor= 20; // infinite ;)
1302 else if (flags&SWS_BILINEAR) sizeFactor= 2;
1303 else {
1304 sizeFactor= 0; //GCC warning killer
1305 assert(0);
1306 }
1307
1308 if (xInc <= 1<<16) filterSize= 1 + sizeFactor; // upscale
1309 else filterSize= 1 + (sizeFactor*srcW + dstW - 1)/ dstW;
1310
1311 if (filterSize > srcW-2) filterSize=srcW-2;
1312
1313 filter= av_malloc(dstW*sizeof(*filter)*filterSize);
1314
1315 xDstInSrc= xInc - 0x10000;
1316 for (i=0; i<dstW; i++)
1317 {
1318 int xx= (xDstInSrc - ((filterSize-2)<<16)) / (1<<17);
1319 int j;
1320 (*filterPos)[i]= xx;
1321 for (j=0; j<filterSize; j++)
1322 {
1323 int64_t d= ((int64_t)FFABS((xx<<17) - xDstInSrc))<<13;
1324 double floatd;
1325 int64_t coeff;
1326
1327 if (xInc > 1<<16)
1328 d= d*dstW/srcW;
1329 floatd= d * (1.0/(1<<30));
1330
1331 if (flags & SWS_BICUBIC)
1332 {
1333 int64_t B= (param[0] != SWS_PARAM_DEFAULT ? param[0] : 0) * (1<<24);
1334 int64_t C= (param[1] != SWS_PARAM_DEFAULT ? param[1] : 0.6) * (1<<24);
1335 int64_t dd = ( d*d)>>30;
1336 int64_t ddd= (dd*d)>>30;
1337
1338 if (d < 1LL<<30)
1339 coeff = (12*(1<<24)-9*B-6*C)*ddd + (-18*(1<<24)+12*B+6*C)*dd + (6*(1<<24)-2*B)*(1<<30);
1340 else if (d < 1LL<<31)
1341 coeff = (-B-6*C)*ddd + (6*B+30*C)*dd + (-12*B-48*C)*d + (8*B+24*C)*(1<<30);
1342 else
1343 coeff=0.0;
1344 coeff *= fone>>(30+24);
1345 }
1346 /* else if (flags & SWS_X)
1347 {
1348 double p= param ? param*0.01 : 0.3;
1349 coeff = d ? sin(d*PI)/(d*PI) : 1.0;
1350 coeff*= pow(2.0, - p*d*d);
1351 }*/
1352 else if (flags & SWS_X)
1353 {
1354 double A= param[0] != SWS_PARAM_DEFAULT ? param[0] : 1.0;
1355 double c;
1356
1357 if (floatd<1.0)
1358 c = cos(floatd*PI);
1359 else
1360 c=-1.0;
1361 if (c<0.0) c= -pow(-c, A);
1362 else c= pow( c, A);
1363 coeff= (c*0.5 + 0.5)*fone;
1364 }
1365 else if (flags & SWS_AREA)
1366 {
1367 int64_t d2= d - (1<<29);
1368 if (d2*xInc < -(1LL<<(29+16))) coeff= 1.0 * (1LL<<(30+16));
1369 else if (d2*xInc < (1LL<<(29+16))) coeff= -d2*xInc + (1LL<<(29+16));
1370 else coeff=0.0;
1371 coeff *= fone>>(30+16);
1372 }
1373 else if (flags & SWS_GAUSS)
1374 {
1375 double p= param[0] != SWS_PARAM_DEFAULT ? param[0] : 3.0;
1376 coeff = (pow(2.0, - p*floatd*floatd))*fone;
1377 }
1378 else if (flags & SWS_SINC)
1379 {
1380 coeff = (d ? sin(floatd*PI)/(floatd*PI) : 1.0)*fone;
1381 }
1382 else if (flags & SWS_LANCZOS)
1383 {
1384 double p= param[0] != SWS_PARAM_DEFAULT ? param[0] : 3.0;
1385 coeff = (d ? sin(floatd*PI)*sin(floatd*PI/p)/(floatd*floatd*PI*PI/p) : 1.0)*fone;
1386 if (floatd>p) coeff=0;
1387 }
1388 else if (flags & SWS_BILINEAR)
1389 {
1390 coeff= (1<<30) - d;
1391 if (coeff<0) coeff=0;
1392 coeff *= fone >> 30;
1393 }
1394 else if (flags & SWS_SPLINE)
1395 {
1396 double p=-2.196152422706632;
1397 coeff = getSplineCoeff(1.0, 0.0, p, -p-1.0, floatd) * fone;
1398 }
1399 else {
1400 coeff= 0.0; //GCC warning killer
1401 assert(0);
1402 }
1403
1404 filter[i*filterSize + j]= coeff;
1405 xx++;
1406 }
1407 xDstInSrc+= 2*xInc;
1408 }
1409 }
1410
1411 /* apply src & dst Filter to filter -> filter2
1412 av_free(filter);
1413 */
1414 assert(filterSize>0);
1415 filter2Size= filterSize;
1416 if (srcFilter) filter2Size+= srcFilter->length - 1;
1417 if (dstFilter) filter2Size+= dstFilter->length - 1;
1418 assert(filter2Size>0);
1419 filter2= av_mallocz(filter2Size*dstW*sizeof(*filter2));
1420
1421 for (i=0; i<dstW; i++)
1422 {
1423 int j, k;
1424
1425 if(srcFilter){
1426 for (k=0; k<srcFilter->length; k++){
1427 for (j=0; j<filterSize; j++)
1428 filter2[i*filter2Size + k + j] += srcFilter->coeff[k]*filter[i*filterSize + j];
1429 }
1430 }else{
1431 for (j=0; j<filterSize; j++)
1432 filter2[i*filter2Size + j]= filter[i*filterSize + j];
1433 }
1434 //FIXME dstFilter
1435
1436 (*filterPos)[i]+= (filterSize-1)/2 - (filter2Size-1)/2;
1437 }
1438 av_freep(&filter);
1439
1440 /* try to reduce the filter-size (step1 find size and shift left) */
1441 // Assume it is near normalized (*0.5 or *2.0 is OK but * 0.001 is not).
1442 minFilterSize= 0;
1443 for (i=dstW-1; i>=0; i--)
1444 {
1445 int min= filter2Size;
1446 int j;
1447 int64_t cutOff=0.0;
1448
1449 /* get rid off near zero elements on the left by shifting left */
1450 for (j=0; j<filter2Size; j++)
1451 {
1452 int k;
1453 cutOff += FFABS(filter2[i*filter2Size]);
1454
1455 if (cutOff > SWS_MAX_REDUCE_CUTOFF*fone) break;
1456
1457 /* preserve monotonicity because the core can't handle the filter otherwise */
1458 if (i<dstW-1 && (*filterPos)[i] >= (*filterPos)[i+1]) break;
1459
1460 // move filter coefficients left
1461 for (k=1; k<filter2Size; k++)
1462 filter2[i*filter2Size + k - 1]= filter2[i*filter2Size + k];
1463 filter2[i*filter2Size + k - 1]= 0;
1464 (*filterPos)[i]++;
1465 }
1466
1467 cutOff=0;
1468 /* count near zeros on the right */
1469 for (j=filter2Size-1; j>0; j--)
1470 {
1471 cutOff += FFABS(filter2[i*filter2Size + j]);
1472
1473 if (cutOff > SWS_MAX_REDUCE_CUTOFF*fone) break;
1474 min--;
1475 }
1476
1477 if (min>minFilterSize) minFilterSize= min;
1478 }
1479
1480 if (flags & SWS_CPU_CAPS_ALTIVEC) {
1481 // we can handle the special case 4,
1482 // so we don't want to go to the full 8
1483 if (minFilterSize < 5)
1484 filterAlign = 4;
1485
1486 // We really don't want to waste our time
1487 // doing useless computation, so fall back on
1488 // the scalar C code for very small filters.
1489 // Vectorizing is worth it only if you have a
1490 // decent-sized vector.
1491 if (minFilterSize < 3)
1492 filterAlign = 1;
1493 }
1494
1495 if (flags & SWS_CPU_CAPS_MMX) {
1496 // special case for unscaled vertical filtering
1497 if (minFilterSize == 1 && filterAlign == 2)
1498 filterAlign= 1;
1499 }
1500
1501 assert(minFilterSize > 0);
1502 filterSize= (minFilterSize +(filterAlign-1)) & (~(filterAlign-1));
1503 assert(filterSize > 0);
1504 filter= av_malloc(filterSize*dstW*sizeof(*filter));
1505 if (filterSize >= MAX_FILTER_SIZE*16/((flags&SWS_ACCURATE_RND) ? APCK_SIZE : 16) || !filter)
1506 goto error;
1507 *outFilterSize= filterSize;
1508
1509 if (flags&SWS_PRINT_INFO)
1510 av_log(NULL, AV_LOG_VERBOSE, "SwScaler: reducing / aligning filtersize %d -> %d\n", filter2Size, filterSize);
1511 /* try to reduce the filter-size (step2 reduce it) */
1512 for (i=0; i<dstW; i++)
1513 {
1514 int j;
1515
1516 for (j=0; j<filterSize; j++)
1517 {
1518 if (j>=filter2Size) filter[i*filterSize + j]= 0;
1519 else filter[i*filterSize + j]= filter2[i*filter2Size + j];
1520 if((flags & SWS_BITEXACT) && j>=minFilterSize)
1521 filter[i*filterSize + j]= 0;
1522 }
1523 }
1524
1525
1526 //FIXME try to align filterPos if possible
1527
1528 //fix borders
1529 for (i=0; i<dstW; i++)
1530 {
1531 int j;
1532 if ((*filterPos)[i] < 0)
1533 {
1534 // move filter coefficients left to compensate for filterPos
1535 for (j=1; j<filterSize; j++)
1536 {
1537 int left= FFMAX(j + (*filterPos)[i], 0);
1538 filter[i*filterSize + left] += filter[i*filterSize + j];
1539 filter[i*filterSize + j]=0;
1540 }
1541 (*filterPos)[i]= 0;
1542 }
1543
1544 if ((*filterPos)[i] + filterSize > srcW)
1545 {
1546 int shift= (*filterPos)[i] + filterSize - srcW;
1547 // move filter coefficients right to compensate for filterPos
1548 for (j=filterSize-2; j>=0; j--)
1549 {
1550 int right= FFMIN(j + shift, filterSize-1);
1551 filter[i*filterSize +right] += filter[i*filterSize +j];
1552 filter[i*filterSize +j]=0;
1553 }
1554 (*filterPos)[i]= srcW - filterSize;
1555 }
1556 }
1557
1558 // Note the +1 is for the MMX scaler which reads over the end
1559 /* align at 16 for AltiVec (needed by hScale_altivec_real) */
1560 *outFilter= av_mallocz(*outFilterSize*(dstW+1)*sizeof(int16_t));
1561
1562 /* normalize & store in outFilter */
1563 for (i=0; i<dstW; i++)
1564 {
1565 int j;
1566 int64_t error=0;
1567 int64_t sum=0;
1568
1569 for (j=0; j<filterSize; j++)
1570 {
1571 sum+= filter[i*filterSize + j];
1572 }
1573 sum= (sum + one/2)/ one;
1574 for (j=0; j<*outFilterSize; j++)
1575 {
1576 int64_t v= filter[i*filterSize + j] + error;
1577 int intV= ROUNDED_DIV(v, sum);
1578 (*outFilter)[i*(*outFilterSize) + j]= intV;
1579 error= v - intV*sum;
1580 }
1581 }
1582
1583 (*filterPos)[dstW]= (*filterPos)[dstW-1]; // the MMX scaler will read over the end
1584 for (i=0; i<*outFilterSize; i++)
1585 {
1586 int j= dstW*(*outFilterSize);
1587 (*outFilter)[j + i]= (*outFilter)[j + i - (*outFilterSize)];
1588 }
1589
1590 ret=0;
1591 error:
1592 av_free(filter);
1593 av_free(filter2);
1594 return ret;
1595 }
1596
1597 #ifdef COMPILE_MMX2
1598 static void initMMX2HScaler(int dstW, int xInc, uint8_t *funnyCode, int16_t *filter, int32_t *filterPos, int numSplits)
1599 {
1600 uint8_t *fragmentA;
1601 x86_reg imm8OfPShufW1A;
1602 x86_reg imm8OfPShufW2A;
1603 x86_reg fragmentLengthA;
1604 uint8_t *fragmentB;
1605 x86_reg imm8OfPShufW1B;
1606 x86_reg imm8OfPShufW2B;
1607 x86_reg fragmentLengthB;
1608 int fragmentPos;
1609
1610 int xpos, i;
1611
1612 // create an optimized horizontal scaling routine
1613
1614 //code fragment
1615
1616 __asm__ volatile(
1617 "jmp 9f \n\t"
1618 // Begin
1619 "0: \n\t"
1620 "movq (%%"REG_d", %%"REG_a"), %%mm3 \n\t"
1621 "movd (%%"REG_c", %%"REG_S"), %%mm0 \n\t"
1622 "movd 1(%%"REG_c", %%"REG_S"), %%mm1 \n\t"
1623 "punpcklbw %%mm7, %%mm1 \n\t"
1624 "punpcklbw %%mm7, %%mm0 \n\t"
1625 "pshufw $0xFF, %%mm1, %%mm1 \n\t"
1626 "1: \n\t"
1627 "pshufw $0xFF, %%mm0, %%mm0 \n\t"
1628 "2: \n\t"
1629 "psubw %%mm1, %%mm0 \n\t"
1630 "movl 8(%%"REG_b", %%"REG_a"), %%esi \n\t"
1631 "pmullw %%mm3, %%mm0 \n\t"
1632 "psllw $7, %%mm1 \n\t"
1633 "paddw %%mm1, %%mm0 \n\t"
1634
1635 "movq %%mm0, (%%"REG_D", %%"REG_a") \n\t"
1636
1637 "add $8, %%"REG_a" \n\t"
1638 // End
1639 "9: \n\t"
1640 // "int $3 \n\t"
1641 "lea " LOCAL_MANGLE(0b) ", %0 \n\t"
1642 "lea " LOCAL_MANGLE(1b) ", %1 \n\t"
1643 "lea " LOCAL_MANGLE(2b) ", %2 \n\t"
1644 "dec %1 \n\t"
1645 "dec %2 \n\t"
1646 "sub %0, %1 \n\t"
1647 "sub %0, %2 \n\t"
1648 "lea " LOCAL_MANGLE(9b) ", %3 \n\t"
1649 "sub %0, %3 \n\t"
1650
1651
1652 :"=r" (fragmentA), "=r" (imm8OfPShufW1A), "=r" (imm8OfPShufW2A),
1653 "=r" (fragmentLengthA)
1654 );
1655
1656 __asm__ volatile(
1657 "jmp 9f \n\t"
1658 // Begin
1659 "0: \n\t"
1660 "movq (%%"REG_d", %%"REG_a"), %%mm3 \n\t"
1661 "movd (%%"REG_c", %%"REG_S"), %%mm0 \n\t"
1662 "punpcklbw %%mm7, %%mm0 \n\t"
1663 "pshufw $0xFF, %%mm0, %%mm1 \n\t"
1664 "1: \n\t"
1665 "pshufw $0xFF, %%mm0, %%mm0 \n\t"
1666 "2: \n\t"
1667 "psubw %%mm1, %%mm0 \n\t"
1668 "movl 8(%%"REG_b", %%"REG_a"), %%esi \n\t"
1669 "pmullw %%mm3, %%mm0 \n\t"
1670 "psllw $7, %%mm1 \n\t"
1671 "paddw %%mm1, %%mm0 \n\t"
1672
1673 "movq %%mm0, (%%"REG_D", %%"REG_a") \n\t"
1674
1675 "add $8, %%"REG_a" \n\t"
1676 // End
1677 "9: \n\t"
1678 // "int $3 \n\t"
1679 "lea " LOCAL_MANGLE(0b) ", %0 \n\t"
1680 "lea " LOCAL_MANGLE(1b) ", %1 \n\t"
1681 "lea " LOCAL_MANGLE(2b) ", %2 \n\t"
1682 "dec %1 \n\t"
1683 "dec %2 \n\t"
1684 "sub %0, %1 \n\t"
1685 "sub %0, %2 \n\t"
1686 "lea " LOCAL_MANGLE(9b) ", %3 \n\t"
1687 "sub %0, %3 \n\t"
1688
1689
1690 :"=r" (fragmentB), "=r" (imm8OfPShufW1B), "=r" (imm8OfPShufW2B),
1691 "=r" (fragmentLengthB)
1692 );
1693
1694 xpos= 0; //lumXInc/2 - 0x8000; // difference between pixel centers
1695 fragmentPos=0;
1696
1697 for (i=0; i<dstW/numSplits; i++)
1698 {
1699 int xx=xpos>>16;
1700
1701 if ((i&3) == 0)
1702 {
1703 int a=0;
1704 int b=((xpos+xInc)>>16) - xx;
1705 int c=((xpos+xInc*2)>>16) - xx;
1706 int d=((xpos+xInc*3)>>16) - xx;
1707
1708 filter[i ] = (( xpos & 0xFFFF) ^ 0xFFFF)>>9;
1709 filter[i+1] = (((xpos+xInc ) & 0xFFFF) ^ 0xFFFF)>>9;
1710 filter[i+2] = (((xpos+xInc*2) & 0xFFFF) ^ 0xFFFF)>>9;
1711 filter[i+3] = (((xpos+xInc*3) & 0xFFFF) ^ 0xFFFF)>>9;
1712 filterPos[i/2]= xx;
1713
1714 if (d+1<4)
1715 {
1716 int maxShift= 3-(d+1);
1717 int shift=0;
1718
1719 memcpy(funnyCode + fragmentPos, fragmentB, fragmentLengthB);
1720
1721 funnyCode[fragmentPos + imm8OfPShufW1B]=
1722 (a+1) | ((b+1)<<2) | ((c+1)<<4) | ((d+1)<<6);
1723 funnyCode[fragmentPos + imm8OfPShufW2B]=
1724 a | (b<<2) | (c<<4) | (d<<6);
1725
1726 if (i+3>=dstW) shift=maxShift; //avoid overread
1727 else if ((filterPos[i/2]&3) <= maxShift) shift=filterPos[i/2]&3; //Align
1728
1729 if (shift && i>=shift)
1730 {
1731 funnyCode[fragmentPos + imm8OfPShufW1B]+= 0x55*shift;
1732 funnyCode[fragmentPos + imm8OfPShufW2B]+= 0x55*shift;
1733 filterPos[i/2]-=shift;
1734 }
1735
1736 fragmentPos+= fragmentLengthB;
1737 }
1738 else
1739 {
1740 int maxShift= 3-d;
1741 int shift=0;
1742
1743 memcpy(funnyCode + fragmentPos, fragmentA, fragmentLengthA);
1744
1745 funnyCode[fragmentPos + imm8OfPShufW1A]=
1746 funnyCode[fragmentPos + imm8OfPShufW2A]=
1747 a | (b<<2) | (c<<4) | (d<<6);
1748
1749 if (i+4>=dstW) shift=maxShift; //avoid overread
1750 else if ((filterPos[i/2]&3) <= maxShift) shift=filterPos[i/2]&3; //partial align
1751
1752 if (shift && i>=shift)
1753 {
1754 funnyCode[fragmentPos + imm8OfPShufW1A]+= 0x55*shift;
1755 funnyCode[fragmentPos + imm8OfPShufW2A]+= 0x55*shift;
1756 filterPos[i/2]-=shift;
1757 }
1758
1759 fragmentPos+= fragmentLengthA;
1760 }
1761
1762 funnyCode[fragmentPos]= RET;
1763 }
1764 xpos+=xInc;
1765 }
1766 filterPos[((i/2)+1)&(~1)]= xpos>>16; // needed to jump to the next part
1767 }
1768 #endif /* COMPILE_MMX2 */
1769
1770 static void globalInit(void){
1771 // generating tables:
1772 int i;
1773 for (i=0; i<768; i++){
1774 int c= av_clip_uint8(i-256);
1775 clip_table[i]=c;
1776 }
1777 }
1778
1779 static SwsFunc getSwsFunc(int flags)
1780 {
1781
1782 #if defined(RUNTIME_CPUDETECT) && CONFIG_GPL
1783 #if ARCH_X86
1784 // ordered per speed fastest first
1785 if (flags & SWS_CPU_CAPS_MMX2) {
1786 return swScale_MMX2;
1787 } else if (flags & SWS_CPU_CAPS_3DNOW) {
1788 return swScale_3DNow;
1789 } else if (flags & SWS_CPU_CAPS_MMX) {
1790 return swScale_MMX;
1791 } else {
1792 return swScale_C;
1793 }
1794
1795 #else
1796 #if ARCH_PPC
1797 if (flags & SWS_CPU_CAPS_ALTIVEC) {
1798 return swScale_altivec;
1799 } else {
1800 return swScale_C;
1801 }
1802 #endif
1803 return swScale_C;
1804 #endif /* ARCH_X86 */
1805 #else //RUNTIME_CPUDETECT
1806 #if HAVE_MMX2
1807 return swScale_MMX2;
1808 #elif HAVE_AMD3DNOW
1809 return swScale_3DNow;
1810 #elif HAVE_MMX
1811 return swScale_MMX;
1812 #elif HAVE_ALTIVEC
1813 return swScale_altivec;
1814 #else
1815 return swScale_C;
1816 #endif
1817 #endif //!RUNTIME_CPUDETECT
1818 }
1819
1820 static int PlanarToNV12Wrapper(SwsContext *c, uint8_t* src[], int srcStride[], int srcSliceY,
1821 int srcSliceH, uint8_t* dstParam[], int dstStride[]){
1822 uint8_t *dst=dstParam[0] + dstStride[0]*srcSliceY;
1823 /* Copy Y plane */
1824 if (dstStride[0]==srcStride[0] && srcStride[0] > 0)
1825 memcpy(dst, src[0], srcSliceH*dstStride[0]);
1826 else
1827 {
1828 int i;
1829 const uint8_t *srcPtr= src[0];
1830 uint8_t *dstPtr= dst;
1831 for (i=0; i<srcSliceH; i++)
1832 {
1833 memcpy(dstPtr, srcPtr, c->srcW);
1834 srcPtr+= srcStride[0];
1835 dstPtr+= dstStride[0];
1836 }
1837 }
1838 dst = dstParam[1] + dstStride[1]*srcSliceY/2;
1839 if (c->dstFormat == PIX_FMT_NV12)
1840 interleaveBytes(src[1], src[2], dst, c->srcW/2, srcSliceH/2, srcStride[1], srcStride[2], dstStride[0]);
1841 else
1842 interleaveBytes(src[2], src[1], dst, c->srcW/2, srcSliceH/2, srcStride[2], srcStride[1], dstStride[0]);
1843
1844 return srcSliceH;
1845 }
1846
1847 static int PlanarToYuy2Wrapper(SwsContext *c, uint8_t* src[], int srcStride[], int srcSliceY,
1848 int srcSliceH, uint8_t* dstParam[], int dstStride[]){
1849 uint8_t *dst=dstParam[0] + dstStride[0]*srcSliceY;
1850
1851 yv12toyuy2(src[0], src[1], src[2], dst, c->srcW, srcSliceH, srcStride[0], srcStride[1], dstStride[0]);
1852
1853 return srcSliceH;
1854 }
1855
1856 static int PlanarToUyvyWrapper(SwsContext *c, uint8_t* src[], int srcStride[], int srcSliceY,
1857 int srcSliceH, uint8_t* dstParam[], int dstStride[]){
1858 uint8_t *dst=dstParam[0] + dstStride[0]*srcSliceY;
1859
1860 yv12touyvy(src[0], src[1], src[2], dst, c->srcW, srcSliceH, srcStride[0], srcStride[1], dstStride[0]);
1861
1862 return srcSliceH;
1863 }
1864
1865 static int YUV422PToYuy2Wrapper(SwsContext *c, uint8_t* src[], int srcStride[], int srcSliceY,
1866 int srcSliceH, uint8_t* dstParam[], int dstStride[]){
1867 uint8_t *dst=dstParam[0] + dstStride[0]*srcSliceY;
1868
1869 yuv422ptoyuy2(src[0],src[1],src[2],dst,c->srcW,srcSliceH,srcStride[0],srcStride[1],dstStride[0]);
1870
1871 return srcSliceH;
1872 }
1873
1874 static int YUV422PToUyvyWrapper(SwsContext *c, uint8_t* src[], int srcStride[], int srcSliceY,
1875 int srcSliceH, uint8_t* dstParam[], int dstStride[]){
1876 uint8_t *dst=dstParam[0] + dstStride[0]*srcSliceY;
1877
1878 yuv422ptouyvy(src[0],src[1],src[2],dst,c->srcW,srcSliceH,srcStride[0],srcStride[1],dstStride[0]);
1879
1880 return srcSliceH;
1881 }
1882
1883 static int YUYV2YUV420Wrapper(SwsContext *c, uint8_t* src[], int srcStride[], int srcSliceY,
1884 int srcSliceH, uint8_t* dstParam[], int dstStride[]){
1885 uint8_t *ydst=dstParam[0] + dstStride[0]*srcSliceY;
1886 uint8_t *udst=dstParam[1] + dstStride[1]*srcSliceY/2;
1887 uint8_t *vdst=dstParam[2] + dstStride[2]*srcSliceY/2;
1888
1889 yuyvtoyuv420(ydst, udst, vdst, src[0], c->srcW, srcSliceH, dstStride[0], dstStride[1], srcStride[0]);
1890
1891 if (dstParam[3])
1892 fillPlane(dstParam[3], dstStride[3], c->srcW, srcSliceH, srcSliceY, 255);
1893
1894 return srcSliceH;
1895 }
1896
1897 static int YUYV2YUV422Wrapper(SwsContext *c, uint8_t* src[], int srcStride[], int srcSliceY,
1898 int srcSliceH, uint8_t* dstParam[], int dstStride[]){
1899 uint8_t *ydst=dstParam[0] + dstStride[0]*srcSliceY;
1900 uint8_t *udst=dstParam[1] + dstStride[1]*srcSliceY;
1901 uint8_t *vdst=dstParam[2] + dstStride[2]*srcSliceY;
1902
1903 yuyvtoyuv422(ydst, udst, vdst, src[0], c->srcW, srcSliceH, dstStride[0], dstStride[1], srcStride[0]);
1904
1905 return srcSliceH;
1906 }
1907
1908 static int UYVY2YUV420Wrapper(SwsContext *c, uint8_t* src[], int srcStride[], int srcSliceY,
1909 int srcSliceH, uint8_t* dstParam[], int dstStride[]){
1910 uint8_t *ydst=dstParam[0] + dstStride[0]*srcSliceY;
1911 uint8_t *udst=dstParam[1] + dstStride[1]*srcSliceY/2;
1912 uint8_t *vdst=dstParam[2] + dstStride[2]*srcSliceY/2;
1913
1914 uyvytoyuv420(ydst, udst, vdst, src[0], c->srcW, srcSliceH, dstStride[0], dstStride[1], srcStride[0]);
1915
1916 if (dstParam[3])
1917 fillPlane(dstParam[3], dstStride[3], c->srcW, srcSliceH, srcSliceY, 255);
1918
1919 return srcSliceH;
1920 }
1921
1922 static int UYVY2YUV422Wrapper(SwsContext *c, uint8_t* src[], int srcStride[], int srcSliceY,
1923 int srcSliceH, uint8_t* dstParam[], int dstStride[]){
1924 uint8_t *ydst=dstParam[0] + dstStride[0]*srcSliceY;
1925 uint8_t *udst=dstParam[1] + dstStride[1]*srcSliceY;
1926 uint8_t *vdst=dstParam[2] + dstStride[2]*srcSliceY;
1927
1928 uyvytoyuv422(ydst, udst, vdst, src[0], c->srcW, srcSliceH, dstStride[0], dstStride[1], srcStride[0]);
1929
1930 return srcSliceH;
1931 }
1932
1933 static int pal2rgbWrapper(SwsContext *c, uint8_t* src[], int srcStride[], int srcSliceY,
1934 int srcSliceH, uint8_t* dst[], int dstStride[]){
1935 const enum PixelFormat srcFormat= c->srcFormat;
1936 const enum PixelFormat dstFormat= c->dstFormat;
1937 void (*conv)(const uint8_t *src, uint8_t *dst, long num_pixels,
1938 const uint8_t *palette)=NULL;
1939 int i;
1940 uint8_t *dstPtr= dst[0] + dstStride[0]*srcSliceY;
1941 uint8_t *srcPtr= src[0];
1942
1943 if (!usePal(srcFormat))
1944 av_log(c, AV_LOG_ERROR, "internal error %s -> %s converter\n",
1945 sws_format_name(srcFormat), sws_format_name(dstFormat));
1946
1947 switch(dstFormat){
1948 case PIX_FMT_RGB32 : conv = palette8topacked32; break;
1949 case PIX_FMT_BGR32 : conv = palette8topacked32; break;
1950 case PIX_FMT_BGR32_1: conv = palette8topacked32; break;
1951 case PIX_FMT_RGB32_1: conv = palette8topacked32; break;
1952 case PIX_FMT_RGB24 : conv = palette8topacked24; break;
1953 case PIX_FMT_BGR24 : conv = palette8topacked24; break;
1954 default: av_log(c, AV_LOG_ERROR, "internal error %s -> %s converter\n",
1955 sws_format_name(srcFormat), sws_format_name(dstFormat)); break;
1956 }
1957
1958
1959 for (i=0; i<srcSliceH; i++) {
1960 conv(srcPtr, dstPtr, c->srcW, (uint8_t *) c->pal_rgb);
1961 srcPtr+= srcStride[0];
1962 dstPtr+= dstStride[0];
1963 }
1964
1965 return srcSliceH;
1966 }
1967
1968 /* {RGB,BGR}{15,16,24,32,32_1} -> {RGB,BGR}{15,16,24,32} */
1969 static int rgb2rgbWrapper(SwsContext *c, uint8_t* src[], int srcStride[], int srcSliceY,
1970 int srcSliceH, uint8_t* dst[], int dstStride[]){
1971 const enum PixelFormat srcFormat= c->srcFormat;
1972 const enum PixelFormat dstFormat= c->dstFormat;
1973 const int srcBpp= (fmt_depth(srcFormat) + 7) >> 3;
1974 const int dstBpp= (fmt_depth(dstFormat) + 7) >> 3;
1975 const int srcId= fmt_depth(srcFormat) >> 2; /* 1:0, 4:1, 8:2, 15:3, 16:4, 24:6, 32:8 */
1976 const int dstId= fmt_depth(dstFormat) >> 2;
1977 void (*conv)(const uint8_t *src, uint8_t *dst, long src_size)=NULL;
1978
1979 /* BGR -> BGR */
1980 if ( (isBGR(srcFormat) && isBGR(dstFormat))
1981 || (isRGB(srcFormat) && isRGB(dstFormat))){
1982 switch(srcId | (dstId<<4)){
1983 case 0x34: conv= rgb16to15; break;
1984 case 0x36: conv= rgb24to15; break;
1985 case 0x38: conv= rgb32to15; break;
1986 case 0x43: conv= rgb15to16; break;
1987 case 0x46: conv= rgb24to16; break;
1988 case 0x48: conv= rgb32to16; break;
1989 case 0x63: conv= rgb15to24; break;
1990 case 0x64: conv= rgb16to24; break;
1991 case 0x68: conv= rgb32to24; break;
1992 case 0x83: conv= rgb15to32; break;
1993 case 0x84: conv= rgb16to32; break;
1994 case 0x86: conv= rgb24to32; break;
1995 default: av_log(c, AV_LOG_ERROR, "internal error %s -> %s converter\n",
1996 sws_format_name(srcFormat), sws_format_name(dstFormat)); break;
1997 }
1998 }else if ( (isBGR(srcFormat) && isRGB(dstFormat))
1999 || (isRGB(srcFormat) && isBGR(dstFormat))){
2000 switch(srcId | (dstId<<4)){
2001 case 0x33: conv= rgb15tobgr15; break;
2002 case 0x34: conv= rgb16tobgr15; break;
2003 case 0x36: conv= rgb24tobgr15; break;
2004 case 0x38: conv= rgb32tobgr15; break;
2005 case 0x43: conv= rgb15tobgr16; break;
2006 case 0x44: conv= rgb16tobgr16; break;
2007 case 0x46: conv= rgb24tobgr16; break;
2008 case 0x48: conv= rgb32tobgr16; break;
2009 case 0x63: conv= rgb15tobgr24; break;
2010 case 0x64: conv= rgb16tobgr24; break;
2011 case 0x66: conv= rgb24tobgr24; break;
2012 case 0x68: conv= rgb32tobgr24; break;
2013 case 0x83: conv= rgb15tobgr32; break;
2014 case 0x84: conv= rgb16tobgr32; break;
2015 case 0x86: conv= rgb24tobgr32; break;
2016 case 0x88: conv= rgb32tobgr32; break;
2017 default: av_log(c, AV_LOG_ERROR, "internal error %s -> %s converter\n",
2018 sws_format_name(srcFormat), sws_format_name(dstFormat)); break;
2019 }
2020 }else{
2021 av_log(c, AV_LOG_ERROR, "internal error %s -> %s converter\n",
2022 sws_format_name(srcFormat), sws_format_name(dstFormat));
2023 }
2024
2025 if(conv)
2026 {
2027 uint8_t *srcPtr= src[0];
2028 if(srcFormat == PIX_FMT_RGB32_1 || srcFormat == PIX_FMT_BGR32_1)
2029 srcPtr += ALT32_CORR;
2030
2031 if (dstStride[0]*srcBpp == srcStride[0]*dstBpp && srcStride[0] > 0)
2032 conv(srcPtr, dst[0] + dstStride[0]*srcSliceY, srcSliceH*srcStride[0]);
2033 else
2034 {
2035 int i;
2036 uint8_t *dstPtr= dst[0] + dstStride[0]*srcSliceY;
2037
2038 for (i=0; i<srcSliceH; i++)
2039 {
2040 conv(srcPtr, dstPtr, c->srcW*srcBpp);
2041 srcPtr+= srcStride[0];
2042 dstPtr+= dstStride[0];
2043 }
2044 }
2045 }
2046 return srcSliceH;
2047 }
2048
2049 static int bgr24toyv12Wrapper(SwsContext *c, uint8_t* src[], int srcStride[], int srcSliceY,
2050 int srcSliceH, uint8_t* dst[], int dstStride[]){
2051
2052 rgb24toyv12(
2053 src[0],
2054 dst[0]+ srcSliceY *dstStride[0],
2055 dst[1]+(srcSliceY>>1)*dstStride[1],
2056 dst[2]+(srcSliceY>>1)*dstStride[2],
2057 c->srcW, srcSliceH,
2058 dstStride[0], dstStride[1], srcStride[0]);
2059 if (dst[3])
2060 fillPlane(dst[3], dstStride[3], c->srcW, srcSliceH, srcSliceY, 255);
2061 return srcSliceH;
2062 }
2063
2064 static int yvu9toyv12Wrapper(SwsContext *c, uint8_t* src[], int srcStride[], int srcSliceY,
2065 int srcSliceH, uint8_t* dst[], int dstStride[]){
2066 int i;
2067
2068 /* copy Y */
2069 if (srcStride[0]==dstStride[0] && srcStride[0] > 0)
2070 memcpy(dst[0]+ srcSliceY*dstStride[0], src[0], srcStride[0]*srcSliceH);
2071 else{
2072 uint8_t *srcPtr= src[0];
2073 uint8_t *dstPtr= dst[0] + dstStride[0]*srcSliceY;
2074
2075 for (i=0; i<srcSliceH; i++)
2076 {
2077 memcpy(dstPtr, srcPtr, c->srcW);
2078 srcPtr+= srcStride[0];
2079 dstPtr+= dstStride[0];
2080 }
2081 }
2082
2083 if (c->dstFormat==PIX_FMT_YUV420P || c->dstFormat==PIX_FMT_YUVA420P){
2084 planar2x(src[1], dst[1], c->chrSrcW, c->chrSrcH, srcStride[1], dstStride[1]);
2085 planar2x(src[2], dst[2], c->chrSrcW, c->chrSrcH, srcStride[2], dstStride[2]);
2086 }else{
2087 planar2x(src[1], dst[2], c->chrSrcW, c->chrSrcH, srcStride[1], dstStride[2]);
2088 planar2x(src[2], dst[1], c->chrSrcW, c->chrSrcH, srcStride[2], dstStride[1]);
2089 }
2090 if (dst[3])
2091 fillPlane(dst[3], dstStride[3], c->srcW, srcSliceH, srcSliceY, 255);
2092 return srcSliceH;
2093 }
2094
2095 /* unscaled copy like stuff (assumes nearly identical formats) */
2096 static int packedCopy(SwsContext *c, uint8_t* src[], int srcStride[], int srcSliceY,
2097 int srcSliceH, uint8_t* dst[], int dstStride[])
2098 {
2099 if (dstStride[0]==srcStride[0] && srcStride[0] > 0)
2100 memcpy(dst[0] + dstStride[0]*srcSliceY, src[0], srcSliceH*dstStride[0]);
2101 else
2102 {
2103 int i;
2104 uint8_t *srcPtr= src[0];
2105 uint8_t *dstPtr= dst[0] + dstStride[0]*srcSliceY;
2106 int length=0;
2107
2108 /* universal length finder */
2109 while(length+c->srcW <= FFABS(dstStride[0])
2110 && length+c->srcW <= FFABS(srcStride[0])) length+= c->srcW;
2111 assert(length!=0);
2112
2113 for (i=0; i<srcSliceH; i++)
2114 {
2115 memcpy(dstPtr, srcPtr, length);
2116 srcPtr+= srcStride[0];
2117 dstPtr+= dstStride[0];
2118 }
2119 }
2120 return srcSliceH;
2121 }
2122
2123 static int planarCopy(SwsContext *c, uint8_t* src[], int srcStride[], int srcSliceY,
2124 int srcSliceH, uint8_t* dst[], int dstStride[])
2125 {
2126 int plane;
2127 for (plane=0; plane<4; plane++)
2128 {
2129 int length= (plane==0 || plane==3) ? c->srcW : -((-c->srcW )>>c->chrDstHSubSample);
2130 int y= (plane==0 || plane==3) ? srcSliceY: -((-srcSliceY)>>c->chrDstVSubSample);
2131 int height= (plane==0 || plane==3) ? srcSliceH: -((-srcSliceH)>>c->chrDstVSubSample);
2132
2133 if (!dst[plane]) continue;
2134 // ignore palette for GRAY8
2135 if (plane == 1 && !dst[2]) continue;
2136 if (!src[plane] || (plane == 1 && !src[2]))
2137 fillPlane(dst[plane], dstStride[plane], length, height, y, (plane==3) ? 255 : 128);
2138 else
2139 {
2140 if (dstStride[plane]==srcStride[plane] && srcStride[plane] > 0)
2141 memcpy(dst[plane] + dstStride[plane]*y, src[plane], height*dstStride[plane]);
2142 else
2143 {
2144 int i;
2145 uint8_t *srcPtr= src[plane];
2146 uint8_t *dstPtr= dst[plane] + dstStride[plane]*y;
2147 for (i=0; i<height; i++)
2148 {
2149 memcpy(dstPtr, srcPtr, length);
2150 srcPtr+= srcStride[plane];
2151 dstPtr+= dstStride[plane];
2152 }
2153 }
2154 }
2155 }
2156 return srcSliceH;
2157 }
2158
2159 static int gray16togray(SwsContext *c, uint8_t* src[], int srcStride[], int srcSliceY,
2160 int srcSliceH, uint8_t* dst[], int dstStride[]){
2161
2162 int length= c->srcW;
2163 int y= srcSliceY;
2164 int height= srcSliceH;
2165 int i, j;
2166 uint8_t *srcPtr= src[0];
2167 uint8_t *dstPtr= dst[0] + dstStride[0]*y;
2168
2169 if (!isGray(c->dstFormat)){
2170 int height= -((-srcSliceH)>>c->chrDstVSubSample);
2171 memset(dst[1], 128, dstStride[1]*height);
2172 memset(dst[2], 128, dstStride[2]*height);
2173 }
2174 if (c->srcFormat == PIX_FMT_GRAY16LE) srcPtr++;
2175 for (i=0; i<height; i++)
2176 {
2177 for (j=0; j<length; j++) dstPtr[j] = srcPtr[j<<1];
2178 srcPtr+= srcStride[0];
2179 dstPtr+= dstStride[0];
2180 }
2181 if (dst[3])
2182 fillPlane(dst[3], dstStride[3], length, height, y, 255);
2183 return srcSliceH;
2184 }
2185
2186 static int graytogray16(SwsContext *c, uint8_t* src[], int srcStride[], int srcSliceY,
2187 int srcSliceH, uint8_t* dst[], int dstStride[]){
2188
2189 int length= c->srcW;
2190 int y= srcSliceY;
2191 int height= srcSliceH;
2192 int i, j;
2193 uint8_t *srcPtr= src[0];
2194 uint8_t *dstPtr= dst[0] + dstStride[0]*y;
2195 for (i=0; i<height; i++)
2196 {
2197 for (j=0; j<length; j++)
2198 {
2199 dstPtr[j<<1] = srcPtr[j];
2200 dstPtr[(j<<1)+1] = srcPtr[j];
2201 }
2202 srcPtr+= srcStride[0];
2203 dstPtr+= dstStride[0];
2204 }
2205 return srcSliceH;
2206 }
2207
2208 static int gray16swap(SwsContext *c, uint8_t* src[], int srcStride[], int srcSliceY,
2209 int srcSliceH, uint8_t* dst[], int dstStride[]){
2210
2211 int length= c->srcW;
2212 int y= srcSliceY;
2213 int height= srcSliceH;
2214 int i, j;
2215 uint16_t *srcPtr= (uint16_t*)src[0];
2216 uint16_t *dstPtr= (uint16_t*)(dst[0] + dstStride[0]*y/2);
2217 for (i=0; i<height; i++)
2218 {
2219 for (j=0; j<length; j++) dstPtr[j] = bswap_16(srcPtr[j]);
2220 srcPtr+= srcStride[0]/2;
2221 dstPtr+= dstStride[0]/2;
2222 }
2223 return srcSliceH;
2224 }
2225
2226
2227 static void getSubSampleFactors(int *h, int *v, int format){
2228 switch(format){
2229 case PIX_FMT_UYVY422:
2230 case PIX_FMT_YUYV422:
2231 *h=1;
2232 *v=0;
2233 break;
2234 case PIX_FMT_YUV420P:
2235 case PIX_FMT_YUVA420P:
2236 case PIX_FMT_GRAY16BE:
2237 case PIX_FMT_GRAY16LE:
2238 case PIX_FMT_GRAY8: //FIXME remove after different subsamplings are fully implemented
2239 case PIX_FMT_NV12:
2240 case PIX_FMT_NV21:
2241 *h=1;
2242 *v=1;
2243 break;
2244 case PIX_FMT_YUV440P:
2245 *h=0;
2246 *v=1;
2247 break;
2248 case PIX_FMT_YUV410P:
2249 *h=2;
2250 *v=2;
2251 break;
2252 case PIX_FMT_YUV444P:
2253 *h=0;
2254 *v=0;
2255 break;
2256 case PIX_FMT_YUV422P:
2257 *h=1;
2258 *v=0;
2259 break;
2260 case PIX_FMT_YUV411P:
2261 *h=2;
2262 *v=0;
2263 break;
2264 default:
2265 *h=0;
2266 *v=0;
2267 break;
2268 }
2269 }
2270
2271 static uint16_t roundToInt16(int64_t f){
2272 int r= (f + (1<<15))>>16;
2273 if (r<-0x7FFF) return 0x8000;
2274 else if (r> 0x7FFF) return 0x7FFF;
2275 else return r;
2276 }
2277
2278 /**
2279 * @param inv_table the yuv2rgb coefficients, normally ff_yuv2rgb_coeffs[x]
2280 * @param fullRange if 1 then the luma range is 0..255 if 0 it is 16..235
2281 * @return -1 if not supported
2282 */
2283 int sws_setColorspaceDetails(SwsContext *c, const int inv_table[4], int srcRange, const int table[4], int dstRange, int brightness, int contrast, int saturation){
2284 int64_t crv = inv_table[0];
2285 int64_t cbu = inv_table[1];
2286 int64_t cgu = -inv_table[2];
2287 int64_t cgv = -inv_table[3];
2288 int64_t cy = 1<<16;
2289 int64_t oy = 0;
2290
2291 memcpy(c->srcColorspaceTable, inv_table, sizeof(int)*4);
2292 memcpy(c->dstColorspaceTable, table, sizeof(int)*4);
2293
2294 c->brightness= brightness;
2295 c->contrast = contrast;
2296 c->saturation= saturation;
2297 c->srcRange = srcRange;
2298 c->dstRange = dstRange;
2299 if (isYUV(c->dstFormat) || isGray(c->dstFormat)) return 0;
2300
2301 c->uOffset= 0x0400040004000400LL;
2302 c->vOffset= 0x0400040004000400LL;
2303
2304 if (!srcRange){
2305 cy= (cy*255) / 219;
2306 oy= 16<<16;
2307 }else{
2308 crv= (crv*224) / 255;
2309 cbu= (cbu*224) / 255;
2310 cgu= (cgu*224) / 255;
2311 cgv= (cgv*224) / 255;
2312 }
2313
2314 cy = (cy *contrast )>>16;
2315 crv= (crv*contrast * saturation)>>32;
2316 cbu= (cbu*contrast * saturation)>>32;
2317 cgu= (cgu*contrast * saturation)>>32;
2318 cgv= (cgv*contrast * saturation)>>32;
2319
2320 oy -= 256*brightness;
2321
2322 c->yCoeff= roundToInt16(cy *8192) * 0x0001000100010001ULL;
2323 c->vrCoeff= roundToInt16(crv*8192) * 0x0001000100010001ULL;
2324 c->ubCoeff= roundToInt16(cbu*8192) * 0x0001000100010001ULL;
2325 c->vgCoeff= roundToInt16(cgv*8192) * 0x0001000100010001ULL;
2326 c->ugCoeff= roundToInt16(cgu*8192) * 0x0001000100010001ULL;
2327 c->yOffset= roundToInt16(oy * 8) * 0x0001000100010001ULL;
2328
2329 c->yuv2rgb_y_coeff = (int16_t)roundToInt16(cy <<13);
2330 c->yuv2rgb_y_offset = (int16_t)roundToInt16(oy << 9);
2331 c->yuv2rgb_v2r_coeff= (int16_t)roundToInt16(crv<<13);
2332 c->yuv2rgb_v2g_coeff= (int16_t)roundToInt16(cgv<<13);
2333 c->yuv2rgb_u2g_coeff= (int16_t)roundToInt16(cgu<<13);
2334 c->yuv2rgb_u2b_coeff= (int16_t)roundToInt16(cbu<<13);
2335
2336 ff_yuv2rgb_c_init_tables(c, inv_table, srcRange, brightness, contrast, saturation);
2337 //FIXME factorize
2338
2339 #ifdef COMPILE_ALTIVEC
2340 if (c->flags & SWS_CPU_CAPS_ALTIVEC)
2341 ff_yuv2rgb_init_tables_altivec(c, inv_table, brightness, contrast, saturation);
2342 #endif
2343 return 0;
2344 }
2345
2346 /**
2347 * @return -1 if not supported
2348 */
2349 int sws_getColorspaceDetails(SwsContext *c, int **inv_table, int *srcRange, int **table, int *dstRange, int *brightness, int *contrast, int *saturation){
2350 if (isYUV(c->dstFormat) || isGray(c->dstFormat)) return -1;
2351
2352 *inv_table = c->srcColorspaceTable;
2353 *table = c->dstColorspaceTable;
2354 *srcRange = c->srcRange;
2355 *dstRange = c->dstRange;
2356 *brightness= c->brightness;
2357 *contrast = c->contrast;
2358 *saturation= c->saturation;
2359
2360 return 0;
2361 }
2362
2363 static int handle_jpeg(enum PixelFormat *format)
2364 {
2365 switch (*format) {
2366 case PIX_FMT_YUVJ420P:
2367 *format = PIX_FMT_YUV420P;
2368 return 1;
2369 case PIX_FMT_YUVJ422P:
2370 *format = PIX_FMT_YUV422P;
2371 return 1;
2372 case PIX_FMT_YUVJ444P:
2373 *format = PIX_FMT_YUV444P;
2374 return 1;
2375 case PIX_FMT_YUVJ440P:
2376 *format = PIX_FMT_YUV440P;
2377 return 1;
2378 default:
2379 return 0;
2380 }
2381 }
2382
2383 SwsContext *sws_getContext(int srcW, int srcH, enum PixelFormat srcFormat, int dstW, int dstH, enum PixelFormat dstFormat, int flags,
2384 SwsFilter *srcFilter, SwsFilter *dstFilter, double *param){
2385
2386 SwsContext *c;
2387 int i;
2388 int usesVFilter, usesHFilter;
2389 int unscaled, needsDither;
2390 int srcRange, dstRange;
2391 SwsFilter dummyFilter= {NULL, NULL, NULL, NULL};
2392 #if ARCH_X86
2393 if (flags & SWS_CPU_CAPS_MMX)
2394 __asm__ volatile("emms\n\t"::: "memory");
2395 #endif
2396
2397 #if !defined(RUNTIME_CPUDETECT) || !CONFIG_GPL //ensure that the flags match the compiled variant if cpudetect is off
2398 flags &= ~(SWS_CPU_CAPS_MMX|SWS_CPU_CAPS_MMX2|SWS_CPU_CAPS_3DNOW|SWS_CPU_CAPS_ALTIVEC|SWS_CPU_CAPS_BFIN);
2399 #if HAVE_MMX2
2400 flags |= SWS_CPU_CAPS_MMX|SWS_CPU_CAPS_MMX2;
2401 #elif HAVE_AMD3DNOW
2402 flags |= SWS_CPU_CAPS_MMX|SWS_CPU_CAPS_3DNOW;
2403 #elif HAVE_MMX
2404 flags |= SWS_CPU_CAPS_MMX;
2405 #elif HAVE_ALTIVEC
2406 flags |= SWS_CPU_CAPS_ALTIVEC;
2407 #elif ARCH_BFIN
2408 flags |= SWS_CPU_CAPS_BFIN;
2409 #endif
2410 #endif /* RUNTIME_CPUDETECT */
2411 if (clip_table[512] != 255) globalInit();
2412 if (!rgb15to16) sws_rgb2rgb_init(flags);
2413
2414 unscaled = (srcW == dstW && srcH == dstH);
2415 needsDither= (isBGR(dstFormat) || isRGB(dstFormat))
2416 && (fmt_depth(dstFormat))<24
2417 && ((fmt_depth(dstFormat))<(fmt_depth(srcFormat)) || (!(isRGB(srcFormat) || isBGR(srcFormat))));
2418
2419 srcRange = handle_jpeg(&srcFormat);
2420 dstRange = handle_jpeg(&dstFormat);
2421
2422 if (!isSupportedIn(srcFormat))
2423 {
2424 av_log(NULL, AV_LOG_ERROR, "swScaler: %s is not supported as input pixel format\n", sws_format_name(srcFormat));
2425 return NULL;
2426 }
2427 if (!isSupportedOut(dstFormat))
2428 {
2429 av_log(NULL, AV_LOG_ERROR, "swScaler: %s is not supported as output pixel format\n", sws_format_name(dstFormat));
2430 return NULL;
2431 }
2432
2433 i= flags & ( SWS_POINT
2434 |SWS_AREA
2435 |SWS_BILINEAR
2436 |SWS_FAST_BILINEAR
2437 |SWS_BICUBIC
2438 |SWS_X
2439 |SWS_GAUSS
2440 |SWS_LANCZOS
2441 |SWS_SINC
2442 |SWS_SPLINE
2443 |SWS_BICUBLIN);
2444 if(!i || (i & (i-1)))
2445 {
2446 av_log(NULL, AV_LOG_ERROR, "swScaler: Exactly one scaler algorithm must be chosen\n");
2447 return NULL;
2448 }
2449
2450 /* sanity check */
2451 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
2452 {
2453 av_log(NULL, AV_LOG_ERROR, "swScaler: %dx%d -> %dx%d is invalid scaling dimension\n",
2454 srcW, srcH, dstW, dstH);
2455 return NULL;
2456 }
2457 if(srcW > VOFW || dstW > VOFW){
2458 av_log(NULL, AV_LOG_ERROR, "swScaler: Compile-time maximum width is "AV_STRINGIFY(VOFW)" change VOF/VOFW and recompile\n");
2459 return NULL;
2460 }
2461
2462 if (!dstFilter) dstFilter= &dummyFilter;
2463 if (!srcFilter) srcFilter= &dummyFilter;
2464
2465 c= av_mallocz(sizeof(SwsContext));
2466
2467 c->av_class = &sws_context_class;
2468 c->srcW= srcW;
2469 c->srcH= srcH;
2470 c->dstW= dstW;
2471 c->dstH= dstH;
2472 c->lumXInc= ((srcW<<16) + (dstW>>1))/dstW;
2473 c->lumYInc= ((srcH<<16) + (dstH>>1))/dstH;
2474 c->flags= flags;
2475 c->dstFormat= dstFormat;
2476 c->srcFormat= srcFormat;
2477 c->vRounder= 4* 0x0001000100010001ULL;
2478
2479 usesHFilter= usesVFilter= 0;
2480 if (dstFilter->lumV && dstFilter->lumV->length>1) usesVFilter=1;
2481 if (dstFilter->lumH && dstFilter->lumH->length>1) usesHFilter=1;
2482 if (dstFilter->chrV && dstFilter->chrV->length>1) usesVFilter=1;
2483 if (dstFilter->chrH && dstFilter->chrH->length>1) usesHFilter=1;
2484 if (srcFilter->lumV && srcFilter->lumV->length>1) usesVFilter=1;
2485 if (srcFilter->lumH && srcFilter->lumH->length>1) usesHFilter=1;
2486 if (srcFilter->chrV && srcFilter->chrV->length>1) usesVFilter=1;
2487 if (srcFilter->chrH && srcFilter->chrH->length>1) usesHFilter=1;
2488
2489 getSubSampleFactors(&c->chrSrcHSubSample, &c->chrSrcVSubSample, srcFormat);
2490 getSubSampleFactors(&c->chrDstHSubSample, &c->chrDstVSubSample, dstFormat);
2491
2492 // reuse chroma for 2 pixels RGB/BGR unless user wants full chroma interpolation
2493 if ((isBGR(dstFormat) || isRGB(dstFormat)) && !(flags&SWS_FULL_CHR_H_INT)) c->chrDstHSubSample=1;
2494
2495 // drop some chroma lines if the user wants it
2496 c->vChrDrop= (flags&SWS_SRC_V_CHR_DROP_MASK)>>SWS_SRC_V_CHR_DROP_SHIFT;
2497 c->chrSrcVSubSample+= c->vChrDrop;
2498
2499 // drop every other pixel for chroma calculation unless user wants full chroma
2500 if ((isBGR(srcFormat) || isRGB(srcFormat)) && !(flags&SWS_FULL_CHR_H_INP)
2501 && srcFormat!=PIX_FMT_RGB8 && srcFormat!=PIX_FMT_BGR8
2502 && srcFormat!=PIX_FMT_RGB4 && srcFormat!=PIX_FMT_BGR4
2503 && srcFormat!=PIX_FMT_RGB4_BYTE && srcFormat!=PIX_FMT_BGR4_BYTE
2504 && ((dstW>>c->chrDstHSubSample) <= (srcW>>1) || (flags&(SWS_FAST_BILINEAR|SWS_POINT))))
2505 c->chrSrcHSubSample=1;
2506
2507 if (param){
2508 c->param[0] = param[0];
2509 c->param[1] = param[1];
2510 }else{
2511 c->param[0] =
2512 c->param[1] = SWS_PARAM_DEFAULT;
2513 }
2514
2515 c->chrIntHSubSample= c->chrDstHSubSample;
2516 c->chrIntVSubSample= c->chrSrcVSubSample;
2517
2518 // Note the -((-x)>>y) is so that we always round toward +inf.
2519 c->chrSrcW= -((-srcW) >> c->chrSrcHSubSample);
2520 c->chrSrcH= -((-srcH) >> c->chrSrcVSubSample);
2521 c->chrDstW= -((-dstW) >> c->chrDstHSubSample);
2522 c->chrDstH= -((-dstH) >> c->chrDstVSubSample);
2523
2524 sws_setColorspaceDetails(c, ff_yuv2rgb_coeffs[SWS_CS_DEFAULT], srcRange, ff_yuv2rgb_coeffs[SWS_CS_DEFAULT] /* FIXME*/, dstRange, 0, 1<<16, 1<<16);
2525
2526 /* unscaled special cases */
2527 if (unscaled && !usesHFilter && !usesVFilter && (srcRange == dstRange || isBGR(dstFormat) || isRGB(dstFormat)))
2528 {
2529 /* yv12_to_nv12 */
2530 if ((srcFormat == PIX_FMT_YUV420P || srcFormat == PIX_FMT_YUVA420P) && (dstFormat == PIX_FMT_NV12 || dstFormat == PIX_FMT_NV21))
2531 {
2532 c->swScale= PlanarToNV12Wrapper;
2533 }
2534 /* yuv2bgr */
2535 if ((srcFormat==PIX_FMT_YUV420P || srcFormat==PIX_FMT_YUV422P || srcFormat==PIX_FMT_YUVA420P) && (isBGR(dstFormat) || isRGB(dstFormat))
2536 && !(flags & SWS_ACCURATE_RND) && !(dstH&1))
2537 {
2538 c->swScale= ff_yuv2rgb_get_func_ptr(c);
2539 }
2540
2541 if (srcFormat==PIX_FMT_YUV410P && (dstFormat==PIX_FMT_YUV420P || dstFormat==PIX_FMT_YUVA420P) && !(flags & SWS_BITEXACT))
2542 {
2543 c->swScale= yvu9toyv12Wrapper;
2544 }
2545
2546 /* bgr24toYV12 */
2547 if (srcFormat==PIX_FMT_BGR24 && (dstFormat==PIX_FMT_YUV420P || dstFormat==PIX_FMT_YUVA420P) && !(flags & SWS_ACCURATE_RND))
2548 c->swScale= bgr24toyv12Wrapper;
2549
2550 /* RGB/BGR -> RGB/BGR (no dither needed forms) */
2551 if ( (isBGR(srcFormat) || isRGB(srcFormat))
2552 && (isBGR(dstFormat) || isRGB(dstFormat))
2553 && srcFormat != PIX_FMT_BGR8 && dstFormat != PIX_FMT_BGR8
2554 && srcFormat != PIX_FMT_RGB8 && dstFormat != PIX_FMT_RGB8
2555 && srcFormat != PIX_FMT_BGR4 && dstFormat != PIX_FMT_BGR4
2556 && srcFormat != PIX_FMT_RGB4 && dstFormat != PIX_FMT_RGB4
2557 && srcFormat != PIX_FMT_BGR4_BYTE && dstFormat != PIX_FMT_BGR4_BYTE
2558 && srcFormat != PIX_FMT_RGB4_BYTE && dstFormat != PIX_FMT_RGB4_BYTE
2559 && srcFormat != PIX_FMT_MONOBLACK && dstFormat != PIX_FMT_MONOBLACK
2560 && srcFormat != PIX_FMT_MONOWHITE && dstFormat != PIX_FMT_MONOWHITE
2561 && dstFormat != PIX_FMT_RGB32_1
2562 && dstFormat != PIX_FMT_BGR32_1
2563 && (!needsDither || (c->flags&(SWS_FAST_BILINEAR|SWS_POINT))))
2564 c->swScale= rgb2rgbWrapper;
2565
2566 if ((usePal(srcFormat) && (
2567 dstFormat == PIX_FMT_RGB32 ||
2568 dstFormat == PIX_FMT_RGB32_1 ||
2569 dstFormat == PIX_FMT_RGB24 ||
2570 dstFormat == PIX_FMT_BGR32 ||
2571 dstFormat == PIX_FMT_BGR32_1 ||
2572 dstFormat == PIX_FMT_BGR24)))
2573 c->swScale= pal2rgbWrapper;
2574
2575 if (srcFormat == PIX_FMT_YUV422P)
2576 {
2577 if (dstFormat == PIX_FMT_YUYV422)
2578 c->swScale= YUV422PToYuy2Wrapper;
2579 else if (dstFormat == PIX_FMT_UYVY422)
2580 c->swScale= YUV422PToUyvyWrapper;
2581 }
2582
2583 /* LQ converters if -sws 0 or -sws 4*/
2584 if (c->flags&(SWS_FAST_BILINEAR|SWS_POINT)){
2585 /* yv12_to_yuy2 */
2586 if (srcFormat == PIX_FMT_YUV420P || srcFormat == PIX_FMT_YUVA420P)
2587 {
2588 if (dstFormat == PIX_FMT_YUYV422)
2589 c->swScale= PlanarToYuy2Wrapper;
2590 else if (dstFormat == PIX_FMT_UYVY422)
2591 c->swScale= PlanarToUyvyWrapper;
2592 }
2593 }
2594 if(srcFormat == PIX_FMT_YUYV422 && (dstFormat == PIX_FMT_YUV420P || dstFormat == PIX_FMT_YUVA420P))
2595 c->swScale= YUYV2YUV420Wrapper;
2596 if(srcFormat == PIX_FMT_UYVY422 && (dstFormat == PIX_FMT_YUV420P || dstFormat == PIX_FMT_YUVA420P))
2597 c->swScale= UYVY2YUV420Wrapper;
2598 if(srcFormat == PIX_FMT_YUYV422 && dstFormat == PIX_FMT_YUV422P)
2599 c->swScale= YUYV2YUV422Wrapper;
2600 if(srcFormat == PIX_FMT_UYVY422 && dstFormat == PIX_FMT_YUV422P)
2601 c->swScale= UYVY2YUV422Wrapper;
2602
2603 #ifdef COMPILE_ALTIVEC
2604 if ((c->flags & SWS_CPU_CAPS_ALTIVEC) &&
2605 !(c->flags & SWS_BITEXACT) &&
2606 srcFormat == PIX_FMT_YUV420P) {
2607 // unscaled YV12 -> packed YUV, we want speed
2608 if (dstFormat == PIX_FMT_YUYV422)
2609 c->swScale= yv12toyuy2_unscaled_altivec;
2610 else if (dstFormat == PIX_FMT_UYVY422)
2611 c->swScale= yv12touyvy_unscaled_altivec;
2612 }
2613 #endif
2614
2615 /* simple copy */
2616 if ( srcFormat == dstFormat
2617 || (srcFormat == PIX_FMT_YUVA420P && dstFormat == PIX_FMT_YUV420P)
2618 || (srcFormat == PIX_FMT_YUV420P && dstFormat == PIX_FMT_YUVA420P)
2619 || (isPlanarYUV(srcFormat) && isGray(dstFormat))
2620 || (isPlanarYUV(dstFormat) && isGray(srcFormat)))
2621 {
2622 if (isPacked(c->srcFormat))
2623 c->swScale= packedCopy;
2624 else /* Planar YUV or gray */
2625 c->swScale= planarCopy;
2626 }
2627
2628 /* gray16{le,be} conversions */
2629 if (isGray16(srcFormat) && (isPlanarYUV(dstFormat) || (dstFormat == PIX_FMT_GRAY8)))
2630 {
2631 c->swScale= gray16togray;
2632 }
2633 if ((isPlanarYUV(srcFormat) || (srcFormat == PIX_FMT_GRAY8)) && isGray16(dstFormat))
2634 {
2635 c->swScale= graytogray16;
2636 }
2637 if (srcFormat != dstFormat && isGray16(srcFormat) && isGray16(dstFormat))
2638 {
2639 c->swScale= gray16swap;
2640 }
2641
2642 #if ARCH_BFIN
2643 if (flags & SWS_CPU_CAPS_BFIN)
2644 ff_bfin_get_unscaled_swscale (c);
2645 #endif
2646
2647 if (c->swScale){
2648 if (flags&SWS_PRINT_INFO)
2649 av_log(c, AV_LOG_INFO, "using unscaled %s -> %s special converter\n",
2650 sws_format_name(srcFormat), sws_format_name(dstFormat));
2651 return c;
2652 }
2653 }
2654
2655 if (flags & SWS_CPU_CAPS_MMX2)
2656 {
2657 c->canMMX2BeUsed= (dstW >=srcW && (dstW&31)==0 && (srcW&15)==0) ? 1 : 0;
2658 if (!c->canMMX2BeUsed && dstW >=srcW && (srcW&15)==0 && (flags&SWS_FAST_BILINEAR))
2659 {
2660 if (flags&SWS_PRINT_INFO)
2661 av_log(c, AV_LOG_INFO, "output width is not a multiple of 32 -> no MMX2 scaler\n");
2662 }
2663 if (usesHFilter) c->canMMX2BeUsed=0;
2664 }
2665 else
2666 c->canMMX2BeUsed=0;
2667
2668 c->chrXInc= ((c->chrSrcW<<16) + (c->chrDstW>>1))/c->chrDstW;
2669 c->chrYInc= ((c->chrSrcH<<16) + (c->chrDstH>>1))/c->chrDstH;
2670
2671 // match pixel 0 of the src to pixel 0 of dst and match pixel n-2 of src to pixel n-2 of dst
2672 // but only for the FAST_BILINEAR mode otherwise do correct scaling
2673 // n-2 is the last chrominance sample available
2674 // this is not perfect, but no one should notice the difference, the more correct variant
2675 // would be like the vertical one, but that would require some special code for the
2676 // first and last pixel
2677 if (flags&SWS_FAST_BILINEAR)
2678 {
2679 if (c->canMMX2BeUsed)
2680 {
2681 c->lumXInc+= 20;
2682 c->chrXInc+= 20;
2683 }
2684 //we don't use the x86 asm scaler if MMX is available
2685 else if (flags & SWS_CPU_CAPS_MMX)
2686 {
2687 c->lumXInc = ((srcW-2)<<16)/(dstW-2) - 20;
2688 c->chrXInc = ((c->chrSrcW-2)<<16)/(c->chrDstW-2) - 20;
2689 }
2690 }
2691
2692 /* precalculate horizontal scaler filter coefficients */
2693 {
2694 const int filterAlign=
2695 (flags & SWS_CPU_CAPS_MMX) ? 4 :
2696 (flags & SWS_CPU_CAPS_ALTIVEC) ? 8 :
2697 1;
2698
2699 initFilter(&c->hLumFilter, &c->hLumFilterPos, &c->hLumFilterSize, c->lumXInc,
2700 srcW , dstW, filterAlign, 1<<14,
2701 (flags&SWS_BICUBLIN) ? (flags|SWS_BICUBIC) : flags,
2702 srcFilter->lumH, dstFilter->lumH, c->param);
2703 initFilter(&c->hChrFilter, &c->hChrFilterPos, &c->hChrFilterSize, c->chrXInc,
2704 c->chrSrcW, c->chrDstW, filterAlign, 1<<14,
2705 (flags&SWS_BICUBLIN) ? (flags|SWS_BILINEAR) : flags,
2706 srcFilter->chrH, dstFilter->chrH, c->param);
2707
2708 #define MAX_FUNNY_CODE_SIZE 10000
2709 #if defined(COMPILE_MMX2)
2710 // can't downscale !!!
2711 if (c->canMMX2BeUsed && (flags & SWS_FAST_BILINEAR))
2712 {
2713 #ifdef MAP_ANONYMOUS
2714 c->funnyYCode = mmap(NULL, MAX_FUNNY_CODE_SIZE, PROT_EXEC | PROT_READ | PROT_WRITE, MAP_PRIVATE | MAP_ANONYMOUS, 0, 0);
2715 c->funnyUVCode = mmap(NULL, MAX_FUNNY_CODE_SIZE, PROT_EXEC | PROT_READ | PROT_WRITE, MAP_PRIVATE | MAP_ANONYMOUS, 0, 0);
2716 #elif HAVE_VIRTUALALLOC
2717 c->funnyYCode = VirtualAlloc(NULL, MAX_FUNNY_CODE_SIZE, MEM_COMMIT, PAGE_EXECUTE_READWRITE);
2718 c->funnyUVCode = VirtualAlloc(NULL, MAX_FUNNY_CODE_SIZE, MEM_COMMIT, PAGE_EXECUTE_READWRITE);
2719 #else
2720 c->funnyYCode = av_malloc(MAX_FUNNY_CODE_SIZE);
2721 c->funnyUVCode = av_malloc(MAX_FUNNY_CODE_SIZE);
2722 #endif
2723
2724 c->lumMmx2Filter = av_malloc((dstW /8+8)*sizeof(int16_t));
2725 c->chrMmx2Filter = av_malloc((c->chrDstW /4+8)*sizeof(int16_t));
2726 c->lumMmx2FilterPos= av_malloc((dstW /2/8+8)*sizeof(int32_t));
2727 c->chrMmx2FilterPos= av_malloc((c->chrDstW/2/4+8)*sizeof(int32_t));
2728
2729 initMMX2HScaler( dstW, c->lumXInc, c->funnyYCode , c->lumMmx2Filter, c->lumMmx2FilterPos, 8);
2730 initMMX2HScaler(c->chrDstW, c->chrXInc, c->funnyUVCode, c->chrMmx2Filter, c->chrMmx2FilterPos, 4);
2731 }
2732 #endif /* defined(COMPILE_MMX2) */
2733 } // initialize horizontal stuff
2734
2735
2736
2737 /* precalculate vertical scaler filter coefficients */
2738 {
2739 const int filterAlign=
2740 (flags & SWS_CPU_CAPS_MMX) && (flags & SWS_ACCURATE_RND) ? 2 :
2741 (flags & SWS_CPU_CAPS_ALTIVEC) ? 8 :
2742 1;
2743
2744 initFilter(&c->vLumFilter, &c->vLumFilterPos, &c->vLumFilterSize, c->lumYInc,
2745 srcH , dstH, filterAlign, (1<<12),
2746 (flags&SWS_BICUBLIN) ? (flags|SWS_BICUBIC) : flags,
2747 srcFilter->lumV, dstFilter->lumV, c->param);
2748 initFilter(&c->vChrFilter, &c->vChrFilterPos, &c->vChrFilterSize, c->chrYInc,
2749 c->chrSrcH, c->chrDstH, filterAlign, (1<<12),
2750 (flags&SWS_BICUBLIN) ? (flags|SWS_BILINEAR) : flags,
2751 srcFilter->chrV, dstFilter->chrV, c->param);
2752
2753 #if HAVE_ALTIVEC
2754 c->vYCoeffsBank = av_malloc(sizeof (vector signed short)*c->vLumFilterSize*c->dstH);
2755 c->vCCoeffsBank = av_malloc(sizeof (vector signed short)*c->vChrFilterSize*c->chrDstH);
2756
2757 for (i=0;i<c->vLumFilterSize*c->dstH;i++) {
2758 int j;
2759 short *p = (short *)&c->vYCoeffsBank[i];
2760 for (j=0;j<8;j++)
2761 p[j] = c->vLumFilter[i];
2762 }
2763
2764 for (i=0;i<c->vChrFilterSize*c->chrDstH;i++) {
2765 int j;
2766 short *p = (short *)&c->vCCoeffsBank[i];
2767 for (j=0;j<8;j++)
2768 p[j] = c->vChrFilter[i];
2769 }
2770 #endif
2771 }
2772
2773 // calculate buffer sizes so that they won't run out while handling these damn slices
2774 c->vLumBufSize= c->vLumFilterSize;
2775 c->vChrBufSize= c->vChrFilterSize;
2776 for (i=0; i<dstH; i++)
2777 {
2778 int chrI= i*c->chrDstH / dstH;
2779 int nextSlice= FFMAX(c->vLumFilterPos[i ] + c->vLumFilterSize - 1,
2780 ((c->vChrFilterPos[chrI] + c->vChrFilterSize - 1)<<c->chrSrcVSubSample));
2781
2782 nextSlice>>= c->chrSrcVSubSample;
2783 nextSlice<<= c->chrSrcVSubSample;
2784 if (c->vLumFilterPos[i ] + c->vLumBufSize < nextSlice)
2785 c->vLumBufSize= nextSlice - c->vLumFilterPos[i];
2786 if (c->vChrFilterPos[chrI] + c->vChrBufSize < (nextSlice>>c->chrSrcVSubSample))
2787 c->vChrBufSize= (nextSlice>>c->chrSrcVSubSample) - c->vChrFilterPos[chrI];
2788 }
2789
2790 // allocate pixbufs (we use dynamic allocation because otherwise we would need to
2791 c->lumPixBuf= av_malloc(c->vLumBufSize*2*sizeof(int16_t*));
2792 c->chrPixBuf= av_malloc(c->vChrBufSize*2*sizeof(int16_t*));
2793 if (CONFIG_SWSCALE_ALPHA && isALPHA(c->srcFormat) && isALPHA(c->dstFormat))
2794 c->alpPixBuf= av_malloc(c->vLumBufSize*2*sizeof(int16_t*));
2795 //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)
2796 /* align at 16 bytes for AltiVec */
2797 for (i=0; i<c->vLumBufSize; i++)
2798 c->lumPixBuf[i]= c->lumPixBuf[i+c->vLumBufSize]= av_mallocz(VOF+1);
2799 for (i=0; i<c->vChrBufSize; i++)
2800 c->chrPixBuf[i]= c->chrPixBuf[i+c->vChrBufSize]= av_malloc((VOF+1)*2);
2801 if (CONFIG_SWSCALE_ALPHA && c->alpPixBuf)
2802 for (i=0; i<c->vLumBufSize; i++)
2803 c->alpPixBuf[i]= c->alpPixBuf[i+c->vLumBufSize]= av_mallocz(VOF+1);
2804
2805 //try to avoid drawing green stuff between the right end and the stride end
2806 for (i=0; i<c->vChrBufSize; i++) memset(c->chrPixBuf[i], 64, (VOF+1)*2);
2807
2808 assert(2*VOFW == VOF);
2809
2810 assert(c->chrDstH <= dstH);
2811
2812 if (flags&SWS_PRINT_INFO)
2813 {
2814 #ifdef DITHER1XBPP
2815 const char *dither= " dithered";
2816 #else
2817 const char *dither= "";
2818 #endif
2819 if (flags&SWS_FAST_BILINEAR)
2820 av_log(c, AV_LOG_INFO, "FAST_BILINEAR scaler, ");
2821 else if (flags&SWS_BILINEAR)
2822 av_log(c, AV_LOG_INFO, "BILINEAR scaler, ");
2823 else if (flags&SWS_BICUBIC)
2824 av_log(c, AV_LOG_INFO, "BICUBIC scaler, ");
2825 else if (flags&SWS_X)
2826 av_log(c, AV_LOG_INFO, "Experimental scaler, ");
2827 else if (flags&SWS_POINT)
2828 av_log(c, AV_LOG_INFO, "Nearest Neighbor / POINT scaler, ");
2829 else if (flags&SWS_AREA)
2830 av_log(c, AV_LOG_INFO, "Area Averageing scaler, ");
2831 else if (flags&SWS_BICUBLIN)
2832 av_log(c, AV_LOG_INFO, "luma BICUBIC / chroma BILINEAR scaler, ");
2833 else if (flags&SWS_GAUSS)
2834 av_log(c, AV_LOG_INFO, "Gaussian scaler, ");
2835 else if (flags&SWS_SINC)
2836 av_log(c, AV_LOG_INFO, "Sinc scaler, ");
2837 else if (flags&SWS_LANCZOS)
2838 av_log(c, AV_LOG_INFO, "Lanczos scaler, ");
2839 else if (flags&SWS_SPLINE)
2840 av_log(c, AV_LOG_INFO, "Bicubic spline scaler, ");
2841 else
2842 av_log(c, AV_LOG_INFO, "ehh flags invalid?! ");
2843
2844 if (dstFormat==PIX_FMT_BGR555 || dstFormat==PIX_FMT_BGR565)
2845 av_log(c, AV_LOG_INFO, "from %s to%s %s ",
2846 sws_format_name(srcFormat), dither, sws_format_name(dstFormat));
2847 else
2848 av_log(c, AV_LOG_INFO, "from %s to %s ",
2849 sws_format_name(srcFormat), sws_format_name(dstFormat));
2850
2851 if (flags & SWS_CPU_CAPS_MMX2)
2852 av_log(c, AV_LOG_INFO, "using MMX2\n");
2853 else if (flags & SWS_CPU_CAPS_3DNOW)
2854 av_log(c, AV_LOG_INFO, "using 3DNOW\n");
2855 else if (flags & SWS_CPU_CAPS_MMX)
2856 av_log(c, AV_LOG_INFO, "using MMX\n");
2857 else if (flags & SWS_CPU_CAPS_ALTIVEC)
2858 av_log(c, AV_LOG_INFO, "using AltiVec\n");
2859 else
2860 av_log(c, AV_LOG_INFO, "using C\n");
2861 }
2862
2863 if (flags & SWS_PRINT_INFO)
2864 {
2865 if (flags & SWS_CPU_CAPS_MMX)
2866 {
2867 if (c->canMMX2BeUsed && (flags&SWS_FAST_BILINEAR))
2868 av_log(c, AV_LOG_VERBOSE, "using FAST_BILINEAR MMX2 scaler for horizontal scaling\n");
2869 else
2870 {
2871 if (c->hLumFilterSize==4)
2872 av_log(c, AV_LOG_VERBOSE, "using 4-tap MMX scaler for horizontal luminance scaling\n");
2873 else if (c->hLumFilterSize==8)
2874 av_log(c, AV_LOG_VERBOSE, "using 8-tap MMX scaler for horizontal luminance scaling\n");
2875 else
2876 av_log(c, AV_LOG_VERBOSE, "using n-tap MMX scaler for horizontal luminance scaling\n");
2877
2878 if (c->hChrFilterSize==4)
2879 av_log(c, AV_LOG_VERBOSE, "using 4-tap MMX scaler for horizontal chrominance scaling\n");
2880 else if (c->hChrFilterSize==8)
2881 av_log(c, AV_LOG_VERBOSE, "using 8-tap MMX scaler for horizontal chrominance scaling\n");
2882 else
2883 av_log(c, AV_LOG_VERBOSE, "using n-tap MMX scaler for horizontal chrominance scaling\n");
2884 }
2885 }
2886 else
2887 {
2888 #if ARCH_X86
2889 av_log(c, AV_LOG_VERBOSE, "using x86 asm scaler for horizontal scaling\n");
2890 #else
2891 if (flags & SWS_FAST_BILINEAR)
2892 av_log(c, AV_LOG_VERBOSE, "using FAST_BILINEAR C scaler for horizontal scaling\n");
2893 else
2894 av_log(c, AV_LOG_VERBOSE, "using C scaler for horizontal scaling\n");
2895 #endif
2896 }
2897 if (isPlanarYUV(dstFormat))
2898 {
2899 if (c->vLumFilterSize==1)
2900 av_log(c, AV_LOG_VERBOSE, "using 1-tap %s \"scaler\" for vertical scaling (YV12 like)\n", (flags & SWS_CPU_CAPS_MMX) ? "MMX" : "C");
2901 else
2902 av_log(c, AV_LOG_VERBOSE, "using n-tap %s scaler for vertical scaling (YV12 like)\n", (flags & SWS_CPU_CAPS_MMX) ? "MMX" : "C");
2903 }
2904 else
2905 {
2906 if (c->vLumFilterSize==1 && c->vChrFilterSize==2)
2907 av_log(c, AV_LOG_VERBOSE, "using 1-tap %s \"scaler\" for vertical luminance scaling (BGR)\n"
2908 " 2-tap scaler for vertical chrominance scaling (BGR)\n", (flags & SWS_CPU_CAPS_MMX) ? "MMX" : "C");
2909 else if (c->vLumFilterSize==2 && c->vChrFilterSize==2)
2910 av_log(c, AV_LOG_VERBOSE, "using 2-tap linear %s scaler for vertical scaling (BGR)\n", (flags & SWS_CPU_CAPS_MMX) ? "MMX" : "C");
2911 else
2912 av_log(c, AV_LOG_VERBOSE, "using n-tap %s scaler for vertical scaling (BGR)\n", (flags & SWS_CPU_CAPS_MMX) ? "MMX" : "C");
2913 }
2914
2915 if (dstFormat==PIX_FMT_BGR24)
2916 av_log(c, AV_LOG_VERBOSE, "using %s YV12->BGR24 converter\n",
2917 (flags & SWS_CPU_CAPS_MMX2) ? "MMX2" : ((flags & SWS_CPU_CAPS_MMX) ? "MMX" : "C"));
2918 else if (dstFormat==PIX_FMT_RGB32)
2919 av_log(c, AV_LOG_VERBOSE, "using %s YV12->BGR32 converter\n", (flags & SWS_CPU_CAPS_MMX) ? "MMX" : "C");
2920 else if (dstFormat==PIX_FMT_BGR565)
2921 av_log(c, AV_LOG_VERBOSE, "using %s YV12->BGR16 converter\n", (flags & SWS_CPU_CAPS_MMX) ? "MMX" : "C");
2922 else if (dstFormat==PIX_FMT_BGR555)
2923 av_log(c, AV_LOG_VERBOSE, "using %s YV12->BGR15 converter\n", (flags & SWS_CPU_CAPS_MMX) ? "MMX" : "C");
2924
2925 av_log(c, AV_LOG_VERBOSE, "%dx%d -> %dx%d\n", srcW, srcH, dstW, dstH);
2926 }
2927 if (flags & SWS_PRINT_INFO)
2928 {
2929 av_log(c, AV_LOG_DEBUG, "lum srcW=%d srcH=%d dstW=%d dstH=%d xInc=%d yInc=%d\n",
2930 c->srcW, c->srcH, c->dstW, c->dstH, c->lumXInc, c->lumYInc);
2931 av_log(c, AV_LOG_DEBUG, "chr srcW=%d srcH=%d dstW=%d dstH=%d xInc=%d yInc=%d\n",
2932 c->chrSrcW, c->chrSrcH, c->chrDstW, c->chrDstH, c->chrXInc, c->chrYInc);
2933 }
2934
2935 c->swScale= getSwsFunc(flags);
2936 return c;
2937 }
2938
2939 /**
2940 * swscale wrapper, so we don't need to export the SwsContext.
2941 * Assumes planar YUV to be in YUV order instead of YVU.
2942 */
2943 int sws_scale(SwsContext *c, uint8_t* src[], int srcStride[], int srcSliceY,
2944 int srcSliceH, uint8_t* dst[], int dstStride[]){
2945 int i;
2946 uint8_t* src2[4]= {src[0], src[1], src[2], src[3]};
2947
2948 if (c->sliceDir == 0 && srcSliceY != 0 && srcSliceY + srcSliceH != c->srcH) {
2949 av_log(c, AV_LOG_ERROR, "Slices start in the middle!\n");
2950 return 0;
2951 }
2952 if (c->sliceDir == 0) {
2953 if (srcSliceY == 0) c->sliceDir = 1; else c->sliceDir = -1;
2954 }
2955
2956 if (usePal(c->srcFormat)){
2957 for (i=0; i<256; i++){
2958 int p, r, g, b,y,u,v;
2959 if(c->srcFormat == PIX_FMT_PAL8){
2960 p=((uint32_t*)(src[1]))[i];
2961 r= (p>>16)&0xFF;
2962 g= (p>> 8)&0xFF;
2963 b= p &0xFF;
2964 }else if(c->srcFormat == PIX_FMT_RGB8){
2965 r= (i>>5 )*36;
2966 g= ((i>>2)&7)*36;
2967 b= (i&3 )*85;
2968 }else if(c->srcFormat == PIX_FMT_BGR8){
2969 b= (i>>6 )*85;
2970 g= ((i>>3)&7)*36;
2971 r= (i&7 )*36;
2972 }else if(c->srcFormat == PIX_FMT_RGB4_BYTE){
2973 r= (i>>3 )*255;
2974 g= ((i>>1)&3)*85;
2975 b= (i&1 )*255;
2976 }else {
2977 assert(c->srcFormat == PIX_FMT_BGR4_BYTE);
2978 b= (i>>3 )*255;
2979 g= ((i>>1)&3)*85;
2980 r= (i&1 )*255;
2981 }
2982 y= av_clip_uint8((RY*r + GY*g + BY*b + ( 33<<(RGB2YUV_SHIFT-1)))>>RGB2YUV_SHIFT);
2983 u= av_clip_uint8((RU*r + GU*g + BU*b + (257<<(RGB2YUV_SHIFT-1)))>>RGB2YUV_SHIFT);
2984 v= av_clip_uint8((RV*r + GV*g + BV*b + (257<<(RGB2YUV_SHIFT-1)))>>RGB2YUV_SHIFT);
2985 c->pal_yuv[i]= y + (u<<8) + (v<<16);
2986
2987
2988 switch(c->dstFormat) {
2989 case PIX_FMT_BGR32:
2990 #ifndef WORDS_BIGENDIAN
2991 case PIX_FMT_RGB24:
2992 #endif
2993 c->pal_rgb[i]= r + (g<<8) + (b<<16);
2994 break;
2995 case PIX_FMT_BGR32_1:
2996 #ifdef WORDS_BIGENDIAN
2997 case PIX_FMT_BGR24:
2998 #endif
2999 c->pal_rgb[i]= (r + (g<<8) + (b<<16)) << 8;
3000 break;
3001 case PIX_FMT_RGB32_1:
3002 #ifdef WORDS_BIGENDIAN
3003 case PIX_FMT_RGB24:
3004 #endif
3005 c->pal_rgb[i]= (b + (g<<8) + (r<<16)) << 8;
3006 break;
3007 case PIX_FMT_RGB32:
3008 #ifndef WORDS_BIGENDIAN
3009 case PIX_FMT_BGR24:
3010 #endif
3011 default:
3012 c->pal_rgb[i]= b + (g<<8) + (r<<16);
3013 }
3014 }
3015 }
3016
3017 // copy strides, so they can safely be modified
3018 if (c->sliceDir == 1) {
3019 // slices go from top to bottom
3020 int srcStride2[4]= {srcStride[0], srcStride[1], srcStride[2], srcStride[3]};
3021 int dstStride2[4]= {dstStride[0], dstStride[1], dstStride[2], dstStride[3]};
3022 return c->swScale(c, src2, srcStride2, srcSliceY, srcSliceH, dst, dstStride2);
3023 } else {
3024 // slices go from bottom to top => we flip the image internally
3025 uint8_t* dst2[4]= {dst[0] + (c->dstH-1)*dstStride[0],
3026 dst[1] + ((c->dstH>>c->chrDstVSubSample)-1)*dstStride[1],
3027 dst[2] + ((c->dstH>>c->chrDstVSubSample)-1)*dstStride[2],
3028 dst[3] + (c->dstH-1)*dstStride[3]};
3029 int srcStride2[4]= {-srcStride[0], -srcStride[1], -srcStride[2], -srcStride[3]};
3030 int dstStride2[4]= {-dstStride[0], -dstStride[1], -dstStride[2], -dstStride[3]};
3031
3032 src2[0] += (srcSliceH-1)*srcStride[0];
3033 if (!usePal(c->srcFormat))
3034 src2[1] += ((srcSliceH>>c->chrSrcVSubSample)-1)*srcStride[1];
3035 src2[2] += ((srcSliceH>>c->chrSrcVSubSample)-1)*srcStride[2];
3036 src2[3] += (srcSliceH-1)*srcStride[3];
3037
3038 return c->swScale(c, src2, srcStride2, c->srcH-srcSliceY-srcSliceH, srcSliceH, dst2, dstStride2);
3039 }
3040 }
3041
3042 #if LIBSWSCALE_VERSION_MAJOR < 1
3043 int sws_scale_ordered(SwsContext *c, uint8_t* src[], int srcStride[], int srcSliceY,
3044 int srcSliceH, uint8_t* dst[], int dstStride[]){
3045 return sws_scale(c, src, srcStride, srcSliceY, srcSliceH, dst, dstStride);
3046 }
3047 #endif
3048
3049 SwsFilter *sws_getDefaultFilter(float lumaGBlur, float chromaGBlur,
3050 float lumaSharpen, float chromaSharpen,
3051 float chromaHShift, float chromaVShift,
3052 int verbose)
3053 {
3054 SwsFilter *filter= av_malloc(sizeof(SwsFilter));
3055
3056 if (lumaGBlur!=0.0){
3057 filter->lumH= sws_getGaussianVec(lumaGBlur, 3.0);
3058 filter->lumV= sws_getGaussianVec(lumaGBlur, 3.0);
3059 }else{
3060 filter->lumH= sws_getIdentityVec();
3061 filter->lumV= sws_getIdentityVec();
3062 }
3063
3064 if (chromaGBlur!=0.0){
3065 filter->chrH= sws_getGaussianVec(chromaGBlur, 3.0);
3066 filter->chrV= sws_getGaussianVec(chromaGBlur, 3.0);
3067 }else{
3068 filter->chrH= sws_getIdentityVec();
3069 filter->chrV= sws_getIdentityVec();
3070 }
3071
3072 if (chromaSharpen!=0.0){
3073 SwsVector *id= sws_getIdentityVec();
3074 sws_scaleVec(filter->chrH, -chromaSharpen);
3075 sws_scaleVec(filter->chrV, -chromaSharpen);
3076 sws_addVec(filter->chrH, id);
3077 sws_addVec(filter->chrV, id);
3078 sws_freeVec(id);
3079 }
3080
3081 if (lumaSharpen!=0.0){
3082 SwsVector *id= sws_getIdentityVec();
3083 sws_scaleVec(filter->lumH, -lumaSharpen);
3084 sws_scaleVec(filter->lumV, -lumaSharpen);
3085 sws_addVec(filter->lumH, id);
3086 sws_addVec(filter->lumV, id);
3087 sws_freeVec(id);
3088 }
3089
3090 if (chromaHShift != 0.0)
3091 sws_shiftVec(filter->chrH, (int)(chromaHShift+0.5));
3092
3093 if (chromaVShift != 0.0)
3094 sws_shiftVec(filter->chrV, (int)(chromaVShift+0.5));
3095
3096 sws_normalizeVec(filter->chrH, 1.0);
3097 sws_normalizeVec(filter->chrV, 1.0);
3098 sws_normalizeVec(filter->lumH, 1.0);
3099 sws_normalizeVec(filter->lumV, 1.0);
3100
3101 if (verbose) sws_printVec2(filter->chrH, NULL, AV_LOG_DEBUG);
3102 if (verbose) sws_printVec2(filter->lumH, NULL, AV_LOG_DEBUG);
3103
3104 return filter;
3105 }
3106
3107 SwsVector *sws_getGaussianVec(double variance, double quality){
3108 const int length= (int)(variance*quality + 0.5) | 1;
3109 int i;
3110 double *coeff= av_malloc(length*sizeof(double));
3111 double middle= (length-1)*0.5;
3112 SwsVector *vec= av_malloc(sizeof(SwsVector));
3113
3114 vec->coeff= coeff;
3115 vec->length= length;
3116
3117 for (i=0; i<length; i++)
3118 {
3119 double dist= i-middle;
3120 coeff[i]= exp(-dist*dist/(2*variance*variance)) / sqrt(2*variance*PI);
3121 }
3122
3123 sws_normalizeVec(vec, 1.0);
3124
3125 return vec;
3126 }
3127
3128 SwsVector *sws_getConstVec(double c, int length){
3129 int i;
3130 double *coeff= av_malloc(length*sizeof(double));
3131 SwsVector *vec= av_malloc(sizeof(SwsVector));
3132
3133 vec->coeff= coeff;
3134 vec->length= length;
3135
3136 for (i=0; i<length; i++)
3137 coeff[i]= c;
3138
3139 return vec;
3140 }
3141
3142
3143 SwsVector *sws_getIdentityVec(void){
3144 return sws_getConstVec(1.0, 1);
3145 }
3146
3147 double sws_dcVec(SwsVector *a){
3148 int i;
3149 double sum=0;
3150
3151 for (i=0; i<a->length; i++)
3152 sum+= a->coeff[i];
3153
3154 return sum;
3155 }
3156
3157 void sws_scaleVec(SwsVector *a, double scalar){
3158 int i;
3159
3160 for (i=0; i<a->length; i++)
3161 a->coeff[i]*= scalar;
3162 }
3163
3164 void sws_normalizeVec(SwsVector *a, double height){
3165 sws_scaleVec(a, height/sws_dcVec(a));
3166 }
3167
3168 static SwsVector *sws_getConvVec(SwsVector *a, SwsVector *b){
3169 int length= a->length + b->length - 1;
3170 double *coeff= av_malloc(length*sizeof(double));
3171 int i, j;
3172 SwsVector *vec= av_malloc(sizeof(SwsVector));
3173
3174 vec->coeff= coeff;
3175 vec->length= length;
3176
3177 for (i=0; i<length; i++) coeff[i]= 0.0;
3178
3179 for (i=0; i<a->length; i++)
3180 {
3181 for (j=0; j<b->length; j++)
3182 {
3183 coeff[i+j]+= a->coeff[i]*b->coeff[j];
3184 }
3185 }
3186
3187 return vec;
3188 }
3189
3190 static SwsVector *sws_sumVec(SwsVector *a, SwsVector *b){
3191 int length= FFMAX(a->length, b->length);
3192 double *coeff= av_malloc(length*sizeof(double));
3193 int i;
3194 SwsVector *vec= av_malloc(sizeof(SwsVector));
3195
3196 vec->coeff= coeff;
3197 vec->length= length;
3198
3199 for (i=0; i<length; i++) coeff[i]= 0.0;
3200
3201 for (i=0; i<a->length; i++) coeff[i + (length-1)/2 - (a->length-1)/2]+= a->coeff[i];
3202 for (i=0; i<b->length; i++) coeff[i + (length-1)/2 - (b->length-1)/2]+= b->coeff[i];
3203
3204 return vec;
3205 }
3206
3207 static SwsVector *sws_diffVec(SwsVector *a, SwsVector *b){
3208 int length= FFMAX(a->length, b->length);
3209 double *coeff= av_malloc(length*sizeof(double));
3210 int i;
3211 SwsVector *vec= av_malloc(sizeof(SwsVector));
3212
3213 vec->coeff= coeff;
3214 vec->length= length;
3215
3216 for (i=0; i<length; i++) coeff[i]= 0.0;
3217
3218 for (i=0; i<a->length; i++) coeff[i + (length-1)/2 - (a->length-1)/2]+= a->coeff[i];
3219 for (i=0; i<b->length; i++) coeff[i + (length-1)/2 - (b->length-1)/2]-= b->coeff[i];
3220
3221 return vec;
3222 }
3223
3224 /* shift left / or right if "shift" is negative */
3225 static SwsVector *sws_getShiftedVec(SwsVector *a, int shift){
3226 int length= a->length + FFABS(shift)*2;
3227 double *coeff= av_malloc(length*sizeof(double));
3228 int i;
3229 SwsVector *vec= av_malloc(sizeof(SwsVector));
3230
3231 vec->coeff= coeff;
3232 vec->length= length;
3233
3234 for (i=0; i<length; i++) coeff[i]= 0.0;
3235
3236 for (i=0; i<a->length; i++)
3237 {
3238 coeff[i + (length-1)/2 - (a->length-1)/2 - shift]= a->coeff[i];
3239 }
3240
3241 return vec;
3242 }
3243
3244 void sws_shiftVec(SwsVector *a, int shift){
3245 SwsVector *shifted= sws_getShiftedVec(a, shift);
3246 av_free(a->coeff);
3247 a->coeff= shifted->coeff;
3248 a->length= shifted->length;
3249 av_free(shifted);
3250 }
3251
3252 void sws_addVec(SwsVector *a, SwsVector *b){
3253 SwsVector *sum= sws_sumVec(a, b);
3254 av_free(a->coeff);
3255 a->coeff= sum->coeff;
3256 a->length= sum->length;
3257 av_free(sum);
3258 }
3259
3260 void sws_subVec(SwsVector *a, SwsVector *b){
3261 SwsVector *diff= sws_diffVec(a, b);
3262 av_free(a->coeff);
3263 a->coeff= diff->coeff;
3264 a->length= diff->length;
3265 av_free(diff);
3266 }
3267
3268 void sws_convVec(SwsVector *a, SwsVector *b){
3269 SwsVector *conv= sws_getConvVec(a, b);
3270 av_free(a->coeff);
3271 a->coeff= conv->coeff;
3272 a->length= conv->length;
3273 av_free(conv);
3274 }
3275
3276 SwsVector *sws_cloneVec(SwsVector *a){
3277 double *coeff= av_malloc(a->length*sizeof(double));
3278 int i;
3279 SwsVector *vec= av_malloc(sizeof(SwsVector));
3280
3281 vec->coeff= coeff;
3282 vec->length= a->length;
3283
3284 for (i=0; i<a->length; i++) coeff[i]= a->coeff[i];
3285
3286 return vec;
3287 }
3288
3289 void sws_printVec2(SwsVector *a, AVClass *log_ctx, int log_level){
3290 int i;
3291 double max=0;
3292 double min=0;
3293 double range;
3294
3295 for (i=0; i<a->length; i++)
3296 if (a->coeff[i]>max) max= a->coeff[i];
3297
3298 for (i=0; i<a->length; i++)
3299 if (a->coeff[i]<min) min= a->coeff[i];
3300
3301 range= max - min;
3302
3303 for (i=0; i<a->length; i++)
3304 {
3305 int x= (int)((a->coeff[i]-min)*60.0/range +0.5);
3306 av_log(log_ctx, log_level, "%1.3f ", a->coeff[i]);
3307 for (;x>0; x--) av_log(log_ctx, log_level, " ");
3308 av_log(log_ctx, log_level, "|\n");
3309 }
3310 }
3311
3312 #if LIBSWSCALE_VERSION_MAJOR < 1
3313 void sws_printVec(SwsVector *a){
3314 sws_printVec2(a, NULL, AV_LOG_DEBUG);
3315 }
3316 #endif
3317
3318 void sws_freeVec(SwsVector *a){
3319 if (!a) return;
3320 av_freep(&a->coeff);
3321 a->length=0;
3322 av_free(a);
3323 }
3324
3325 void sws_freeFilter(SwsFilter *filter){
3326 if (!filter) return;
3327
3328 if (filter->lumH) sws_freeVec(filter->lumH);
3329 if (filter->lumV) sws_freeVec(filter->lumV);
3330 if (filter->chrH) sws_freeVec(filter->chrH);
3331 if (filter->chrV) sws_freeVec(filter->chrV);
3332 av_free(filter);
3333 }
3334
3335
3336 void sws_freeContext(SwsContext *c){
3337 int i;
3338 if (!c) return;
3339
3340 if (c->lumPixBuf)
3341 {
3342 for (i=0; i<c->vLumBufSize; i++)
3343 av_freep(&c->lumPixBuf[i]);
3344 av_freep(&c->lumPixBuf);
3345 }
3346
3347 if (c->chrPixBuf)
3348 {
3349 for (i=0; i<c->vChrBufSize; i++)
3350 av_freep(&c->chrPixBuf[i]);
3351 av_freep(&c->chrPixBuf);
3352 }
3353
3354 if (CONFIG_SWSCALE_ALPHA && c->alpPixBuf){
3355 for (i=0; i<c->vLumBufSize; i++)
3356 av_freep(&c->alpPixBuf[i]);
3357 av_freep(&c->alpPixBuf);
3358 }
3359
3360 av_freep(&c->vLumFilter);
3361 av_freep(&c->vChrFilter);
3362 av_freep(&c->hLumFilter);
3363 av_freep(&c->hChrFilter);
3364 #if HAVE_ALTIVEC
3365 av_freep(&c->vYCoeffsBank);
3366 av_freep(&c->vCCoeffsBank);
3367 #endif
3368
3369 av_freep(&c->vLumFilterPos);
3370 av_freep(&c->vChrFilterPos);
3371 av_freep(&c->hLumFilterPos);
3372 av_freep(&c->hChrFilterPos);
3373
3374 #if ARCH_X86 && CONFIG_GPL
3375 #ifdef MAP_ANONYMOUS
3376 if (c->funnyYCode ) munmap(c->funnyYCode , MAX_FUNNY_CODE_SIZE);
3377 if (c->funnyUVCode) munmap(c->funnyUVCode, MAX_FUNNY_CODE_SIZE);
3378 #elif HAVE_VIRTUALALLOC
3379 if (c->funnyYCode ) VirtualFree(c->funnyYCode , MAX_FUNNY_CODE_SIZE, MEM_RELEASE);
3380 if (c->funnyUVCode) VirtualFree(c->funnyUVCode, MAX_FUNNY_CODE_SIZE, MEM_RELEASE);
3381 #else
3382 av_free(c->funnyYCode );
3383 av_free(c->funnyUVCode);
3384 #endif
3385 c->funnyYCode=NULL;
3386 c->funnyUVCode=NULL;
3387 #endif /* ARCH_X86 && CONFIG_GPL */
3388
3389 av_freep(&c->lumMmx2Filter);
3390 av_freep(&c->chrMmx2Filter);
3391 av_freep(&c->lumMmx2FilterPos);
3392 av_freep(&c->chrMmx2FilterPos);
3393 av_freep(&c->yuvTable);
3394
3395 av_free(c);
3396 }
3397
3398 struct SwsContext *sws_getCachedContext(struct SwsContext *context,
3399 int srcW, int srcH, enum PixelFormat srcFormat,
3400 int dstW, int dstH, enum PixelFormat dstFormat, int flags,
3401 SwsFilter *srcFilter, SwsFilter *dstFilter, double *param)