Mercurial > mplayer.hg
annotate libswscale/swscale.c @ 19172:bae6c99a99cc
vertical scaler with accurate rounding, some people on doom9 can see +-1 errors
the +-1 issue is limited to >2tap vertical filters, so bilinear upscale was unaffected
the new code is sometime faster sometimes slower but the difference is significant (~20%) so its optional and enabled with arnd=1
author | michael |
---|---|
date | Mon, 24 Jul 2006 10:36:06 +0000 |
parents | 70ea0a8d3b4a |
children | e40cf0305d4e |
rev | line source |
---|---|
18861 | 1 /* |
2 Copyright (C) 2001-2003 Michael Niedermayer <michaelni@gmx.at> | |
3 | |
4 This program is free software; you can redistribute it and/or modify | |
5 it under the terms of the GNU General Public License as published by | |
6 the Free Software Foundation; either version 2 of the License, or | |
7 (at your option) any later version. | |
8 | |
9 This program is distributed in the hope that it will be useful, | |
10 but WITHOUT ANY WARRANTY; without even the implied warranty of | |
11 MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the | |
12 GNU General Public License for more details. | |
13 | |
14 You should have received a copy of the GNU General Public License | |
15 along with this program; if not, write to the Free Software | |
16 Foundation, Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA | |
17 */ | |
18 | |
19 /* | |
20 supported Input formats: YV12, I420/IYUV, YUY2, UYVY, BGR32, BGR24, BGR16, BGR15, RGB32, RGB24, Y8/Y800, YVU9/IF09 | |
21 supported output formats: YV12, I420/IYUV, YUY2, UYVY, {BGR,RGB}{1,4,8,15,16,24,32}, Y8/Y800, YVU9/IF09 | |
22 {BGR,RGB}{1,4,8,15,16} support dithering | |
23 | |
24 unscaled special converters (YV12=I420=IYUV, Y800=Y8) | |
25 YV12 -> {BGR,RGB}{1,4,8,15,16,24,32} | |
26 x -> x | |
27 YUV9 -> YV12 | |
28 YUV9/YV12 -> Y800 | |
29 Y800 -> YUV9/YV12 | |
30 BGR24 -> BGR32 & RGB24 -> RGB32 | |
31 BGR32 -> BGR24 & RGB32 -> RGB24 | |
32 BGR15 -> BGR16 | |
33 */ | |
34 | |
35 /* | |
36 tested special converters (most are tested actually but i didnt write it down ...) | |
37 YV12 -> BGR16 | |
38 YV12 -> YV12 | |
39 BGR15 -> BGR16 | |
40 BGR16 -> BGR16 | |
41 YVU9 -> YV12 | |
42 | |
43 untested special converters | |
44 YV12/I420 -> BGR15/BGR24/BGR32 (its the yuv2rgb stuff, so it should be ok) | |
45 YV12/I420 -> YV12/I420 | |
46 YUY2/BGR15/BGR24/BGR32/RGB24/RGB32 -> same format | |
47 BGR24 -> BGR32 & RGB24 -> RGB32 | |
48 BGR32 -> BGR24 & RGB32 -> RGB24 | |
49 BGR24 -> YV12 | |
50 */ | |
51 | |
52 #include <inttypes.h> | |
53 #include <string.h> | |
54 #include <math.h> | |
55 #include <stdio.h> | |
56 #include <unistd.h> | |
57 #include "config.h" | |
58 #include <assert.h> | |
59 #ifdef HAVE_MALLOC_H | |
60 #include <malloc.h> | |
61 #else | |
62 #include <stdlib.h> | |
63 #endif | |
64 #ifdef HAVE_SYS_MMAN_H | |
65 #include <sys/mman.h> | |
66 #if defined(MAP_ANON) && !defined(MAP_ANONYMOUS) | |
67 #define MAP_ANONYMOUS MAP_ANON | |
68 #endif | |
69 #endif | |
70 #include "swscale.h" | |
71 #include "swscale_internal.h" | |
19143
c4dac777b44c
Use libavutil in libswscale, and allow it to be built out of the mplayer tree
lucabe
parents:
18861
diff
changeset
|
72 #include "x86_cpu.h" |
18861 | 73 #include "bswap.h" |
19143
c4dac777b44c
Use libavutil in libswscale, and allow it to be built out of the mplayer tree
lucabe
parents:
18861
diff
changeset
|
74 #include "img_format.h" |
18861 | 75 #include "rgb2rgb.h" |
19143
c4dac777b44c
Use libavutil in libswscale, and allow it to be built out of the mplayer tree
lucabe
parents:
18861
diff
changeset
|
76 #ifdef USE_FASTMEMCPY |
18861 | 77 #include "libvo/fastmemcpy.h" |
19143
c4dac777b44c
Use libavutil in libswscale, and allow it to be built out of the mplayer tree
lucabe
parents:
18861
diff
changeset
|
78 #endif |
18861 | 79 |
80 #undef MOVNTQ | |
81 #undef PAVGB | |
82 | |
83 //#undef HAVE_MMX2 | |
84 //#define HAVE_3DNOW | |
85 //#undef HAVE_MMX | |
86 //#undef ARCH_X86 | |
87 //#define WORDS_BIGENDIAN | |
88 #define DITHER1XBPP | |
89 | |
90 #define FAST_BGR2YV12 // use 7 bit coeffs instead of 15bit | |
91 | |
92 #define RET 0xC3 //near return opcode for X86 | |
93 | |
94 #ifdef MP_DEBUG | |
95 #define ASSERT(x) assert(x); | |
96 #else | |
97 #define ASSERT(x) ; | |
98 #endif | |
99 | |
100 #ifdef M_PI | |
101 #define PI M_PI | |
102 #else | |
103 #define PI 3.14159265358979323846 | |
104 #endif | |
105 | |
106 //FIXME replace this with something faster | |
107 #define isPlanarYUV(x) ((x)==IMGFMT_YV12 || (x)==IMGFMT_YVU9 \ | |
108 || (x)==IMGFMT_NV12 || (x)==IMGFMT_NV21 \ | |
109 || (x)==IMGFMT_444P || (x)==IMGFMT_422P || (x)==IMGFMT_411P) | |
110 #define isYUV(x) ((x)==IMGFMT_UYVY || (x)==IMGFMT_YUY2 || isPlanarYUV(x)) | |
111 #define isGray(x) ((x)==IMGFMT_Y800) | |
112 #define isRGB(x) (((x)&IMGFMT_RGB_MASK)==IMGFMT_RGB) | |
113 #define isBGR(x) (((x)&IMGFMT_BGR_MASK)==IMGFMT_BGR) | |
114 #define isSupportedIn(x) ((x)==IMGFMT_YV12 || (x)==IMGFMT_YUY2 || (x)==IMGFMT_UYVY\ | |
115 || (x)==IMGFMT_BGR32|| (x)==IMGFMT_BGR24|| (x)==IMGFMT_BGR16|| (x)==IMGFMT_BGR15\ | |
116 || (x)==IMGFMT_RGB32|| (x)==IMGFMT_RGB24\ | |
117 || (x)==IMGFMT_Y800 || (x)==IMGFMT_YVU9\ | |
118 || (x)==IMGFMT_444P || (x)==IMGFMT_422P || (x)==IMGFMT_411P) | |
119 #define isSupportedOut(x) ((x)==IMGFMT_YV12 || (x)==IMGFMT_YUY2 || (x)==IMGFMT_UYVY\ | |
120 || (x)==IMGFMT_444P || (x)==IMGFMT_422P || (x)==IMGFMT_411P\ | |
121 || isRGB(x) || isBGR(x)\ | |
122 || (x)==IMGFMT_NV12 || (x)==IMGFMT_NV21\ | |
123 || (x)==IMGFMT_Y800 || (x)==IMGFMT_YVU9) | |
124 #define isPacked(x) ((x)==IMGFMT_YUY2 || (x)==IMGFMT_UYVY ||isRGB(x) || isBGR(x)) | |
125 | |
126 #define RGB2YUV_SHIFT 16 | |
127 #define BY ((int)( 0.098*(1<<RGB2YUV_SHIFT)+0.5)) | |
128 #define BV ((int)(-0.071*(1<<RGB2YUV_SHIFT)+0.5)) | |
129 #define BU ((int)( 0.439*(1<<RGB2YUV_SHIFT)+0.5)) | |
130 #define GY ((int)( 0.504*(1<<RGB2YUV_SHIFT)+0.5)) | |
131 #define GV ((int)(-0.368*(1<<RGB2YUV_SHIFT)+0.5)) | |
132 #define GU ((int)(-0.291*(1<<RGB2YUV_SHIFT)+0.5)) | |
133 #define RY ((int)( 0.257*(1<<RGB2YUV_SHIFT)+0.5)) | |
134 #define RV ((int)( 0.439*(1<<RGB2YUV_SHIFT)+0.5)) | |
135 #define RU ((int)(-0.148*(1<<RGB2YUV_SHIFT)+0.5)) | |
136 | |
137 extern const int32_t Inverse_Table_6_9[8][4]; | |
138 | |
139 /* | |
140 NOTES | |
141 Special versions: fast Y 1:1 scaling (no interpolation in y direction) | |
142 | |
143 TODO | |
144 more intelligent missalignment avoidance for the horizontal scaler | |
145 write special vertical cubic upscale version | |
146 Optimize C code (yv12 / minmax) | |
147 add support for packed pixel yuv input & output | |
148 add support for Y8 output | |
149 optimize bgr24 & bgr32 | |
150 add BGR4 output support | |
151 write special BGR->BGR scaler | |
152 */ | |
153 | |
154 #define MIN(a,b) ((a) > (b) ? (b) : (a)) | |
155 #define MAX(a,b) ((a) < (b) ? (b) : (a)) | |
156 | |
157 #if defined(ARCH_X86) || defined(ARCH_X86_64) | |
158 static uint64_t attribute_used __attribute__((aligned(8))) bF8= 0xF8F8F8F8F8F8F8F8LL; | |
159 static uint64_t attribute_used __attribute__((aligned(8))) bFC= 0xFCFCFCFCFCFCFCFCLL; | |
160 static uint64_t __attribute__((aligned(8))) w10= 0x0010001000100010LL; | |
161 static uint64_t attribute_used __attribute__((aligned(8))) w02= 0x0002000200020002LL; | |
162 static uint64_t attribute_used __attribute__((aligned(8))) bm00001111=0x00000000FFFFFFFFLL; | |
163 static uint64_t attribute_used __attribute__((aligned(8))) bm00000111=0x0000000000FFFFFFLL; | |
164 static uint64_t attribute_used __attribute__((aligned(8))) bm11111000=0xFFFFFFFFFF000000LL; | |
165 static uint64_t attribute_used __attribute__((aligned(8))) bm01010101=0x00FF00FF00FF00FFLL; | |
166 | |
167 static volatile uint64_t attribute_used __attribute__((aligned(8))) b5Dither; | |
168 static volatile uint64_t attribute_used __attribute__((aligned(8))) g5Dither; | |
169 static volatile uint64_t attribute_used __attribute__((aligned(8))) g6Dither; | |
170 static volatile uint64_t attribute_used __attribute__((aligned(8))) r5Dither; | |
171 | |
172 static uint64_t __attribute__((aligned(8))) dither4[2]={ | |
173 0x0103010301030103LL, | |
174 0x0200020002000200LL,}; | |
175 | |
176 static uint64_t __attribute__((aligned(8))) dither8[2]={ | |
177 0x0602060206020602LL, | |
178 0x0004000400040004LL,}; | |
179 | |
180 static uint64_t __attribute__((aligned(8))) b16Mask= 0x001F001F001F001FLL; | |
181 static uint64_t attribute_used __attribute__((aligned(8))) g16Mask= 0x07E007E007E007E0LL; | |
182 static uint64_t attribute_used __attribute__((aligned(8))) r16Mask= 0xF800F800F800F800LL; | |
183 static uint64_t __attribute__((aligned(8))) b15Mask= 0x001F001F001F001FLL; | |
184 static uint64_t attribute_used __attribute__((aligned(8))) g15Mask= 0x03E003E003E003E0LL; | |
185 static uint64_t attribute_used __attribute__((aligned(8))) r15Mask= 0x7C007C007C007C00LL; | |
186 | |
187 static uint64_t attribute_used __attribute__((aligned(8))) M24A= 0x00FF0000FF0000FFLL; | |
188 static uint64_t attribute_used __attribute__((aligned(8))) M24B= 0xFF0000FF0000FF00LL; | |
189 static uint64_t attribute_used __attribute__((aligned(8))) M24C= 0x0000FF0000FF0000LL; | |
190 | |
191 #ifdef FAST_BGR2YV12 | |
192 static const uint64_t bgr2YCoeff attribute_used __attribute__((aligned(8))) = 0x000000210041000DULL; | |
193 static const uint64_t bgr2UCoeff attribute_used __attribute__((aligned(8))) = 0x0000FFEEFFDC0038ULL; | |
194 static const uint64_t bgr2VCoeff attribute_used __attribute__((aligned(8))) = 0x00000038FFD2FFF8ULL; | |
195 #else | |
196 static const uint64_t bgr2YCoeff attribute_used __attribute__((aligned(8))) = 0x000020E540830C8BULL; | |
197 static const uint64_t bgr2UCoeff attribute_used __attribute__((aligned(8))) = 0x0000ED0FDAC23831ULL; | |
198 static const uint64_t bgr2VCoeff attribute_used __attribute__((aligned(8))) = 0x00003831D0E6F6EAULL; | |
199 #endif | |
200 static const uint64_t bgr2YOffset attribute_used __attribute__((aligned(8))) = 0x1010101010101010ULL; | |
201 static const uint64_t bgr2UVOffset attribute_used __attribute__((aligned(8)))= 0x8080808080808080ULL; | |
202 static const uint64_t w1111 attribute_used __attribute__((aligned(8))) = 0x0001000100010001ULL; | |
203 #endif | |
204 | |
205 // clipping helper table for C implementations: | |
206 static unsigned char clip_table[768]; | |
207 | |
208 static SwsVector *sws_getConvVec(SwsVector *a, SwsVector *b); | |
209 | |
210 extern const uint8_t dither_2x2_4[2][8]; | |
211 extern const uint8_t dither_2x2_8[2][8]; | |
212 extern const uint8_t dither_8x8_32[8][8]; | |
213 extern const uint8_t dither_8x8_73[8][8]; | |
214 extern const uint8_t dither_8x8_220[8][8]; | |
215 | |
19143
c4dac777b44c
Use libavutil in libswscale, and allow it to be built out of the mplayer tree
lucabe
parents:
18861
diff
changeset
|
216 char *sws_format_name(int format) |
c4dac777b44c
Use libavutil in libswscale, and allow it to be built out of the mplayer tree
lucabe
parents:
18861
diff
changeset
|
217 { |
c4dac777b44c
Use libavutil in libswscale, and allow it to be built out of the mplayer tree
lucabe
parents:
18861
diff
changeset
|
218 static char fmt_name[64]; |
c4dac777b44c
Use libavutil in libswscale, and allow it to be built out of the mplayer tree
lucabe
parents:
18861
diff
changeset
|
219 char *res; |
c4dac777b44c
Use libavutil in libswscale, and allow it to be built out of the mplayer tree
lucabe
parents:
18861
diff
changeset
|
220 static int buffer; |
c4dac777b44c
Use libavutil in libswscale, and allow it to be built out of the mplayer tree
lucabe
parents:
18861
diff
changeset
|
221 |
c4dac777b44c
Use libavutil in libswscale, and allow it to be built out of the mplayer tree
lucabe
parents:
18861
diff
changeset
|
222 res = fmt_name + buffer * 32; |
c4dac777b44c
Use libavutil in libswscale, and allow it to be built out of the mplayer tree
lucabe
parents:
18861
diff
changeset
|
223 buffer = 1 - buffer; |
c4dac777b44c
Use libavutil in libswscale, and allow it to be built out of the mplayer tree
lucabe
parents:
18861
diff
changeset
|
224 snprintf(res, 32, "0x%x (%c%c%c%c)", format, |
c4dac777b44c
Use libavutil in libswscale, and allow it to be built out of the mplayer tree
lucabe
parents:
18861
diff
changeset
|
225 format >> 24, (format >> 16) & 0xFF, |
c4dac777b44c
Use libavutil in libswscale, and allow it to be built out of the mplayer tree
lucabe
parents:
18861
diff
changeset
|
226 (format >> 8) & 0xFF, |
c4dac777b44c
Use libavutil in libswscale, and allow it to be built out of the mplayer tree
lucabe
parents:
18861
diff
changeset
|
227 format & 0xFF); |
c4dac777b44c
Use libavutil in libswscale, and allow it to be built out of the mplayer tree
lucabe
parents:
18861
diff
changeset
|
228 |
c4dac777b44c
Use libavutil in libswscale, and allow it to be built out of the mplayer tree
lucabe
parents:
18861
diff
changeset
|
229 return res; |
c4dac777b44c
Use libavutil in libswscale, and allow it to be built out of the mplayer tree
lucabe
parents:
18861
diff
changeset
|
230 } |
c4dac777b44c
Use libavutil in libswscale, and allow it to be built out of the mplayer tree
lucabe
parents:
18861
diff
changeset
|
231 |
18861 | 232 #if defined(ARCH_X86) || defined(ARCH_X86_64) |
233 void in_asm_used_var_warning_killer() | |
234 { | |
235 volatile int i= bF8+bFC+w10+ | |
236 bm00001111+bm00000111+bm11111000+b16Mask+g16Mask+r16Mask+b15Mask+g15Mask+r15Mask+ | |
237 M24A+M24B+M24C+w02 + b5Dither+g5Dither+r5Dither+g6Dither+dither4[0]+dither8[0]+bm01010101; | |
238 if(i) i=0; | |
239 } | |
240 #endif | |
241 | |
242 static inline void yuv2yuvXinC(int16_t *lumFilter, int16_t **lumSrc, int lumFilterSize, | |
243 int16_t *chrFilter, int16_t **chrSrc, int chrFilterSize, | |
244 uint8_t *dest, uint8_t *uDest, uint8_t *vDest, int dstW, int chrDstW) | |
245 { | |
246 //FIXME Optimize (just quickly writen not opti..) | |
247 int i; | |
248 for(i=0; i<dstW; i++) | |
249 { | |
250 int val=1<<18; | |
251 int j; | |
252 for(j=0; j<lumFilterSize; j++) | |
253 val += lumSrc[j][i] * lumFilter[j]; | |
254 | |
255 dest[i]= MIN(MAX(val>>19, 0), 255); | |
256 } | |
257 | |
258 if(uDest != NULL) | |
259 for(i=0; i<chrDstW; i++) | |
260 { | |
261 int u=1<<18; | |
262 int v=1<<18; | |
263 int j; | |
264 for(j=0; j<chrFilterSize; j++) | |
265 { | |
266 u += chrSrc[j][i] * chrFilter[j]; | |
267 v += chrSrc[j][i + 2048] * chrFilter[j]; | |
268 } | |
269 | |
270 uDest[i]= MIN(MAX(u>>19, 0), 255); | |
271 vDest[i]= MIN(MAX(v>>19, 0), 255); | |
272 } | |
273 } | |
274 | |
275 static inline void yuv2nv12XinC(int16_t *lumFilter, int16_t **lumSrc, int lumFilterSize, | |
276 int16_t *chrFilter, int16_t **chrSrc, int chrFilterSize, | |
277 uint8_t *dest, uint8_t *uDest, int dstW, int chrDstW, int dstFormat) | |
278 { | |
279 //FIXME Optimize (just quickly writen not opti..) | |
280 int i; | |
281 for(i=0; i<dstW; i++) | |
282 { | |
283 int val=1<<18; | |
284 int j; | |
285 for(j=0; j<lumFilterSize; j++) | |
286 val += lumSrc[j][i] * lumFilter[j]; | |
287 | |
288 dest[i]= MIN(MAX(val>>19, 0), 255); | |
289 } | |
290 | |
291 if(uDest == NULL) | |
292 return; | |
293 | |
294 if(dstFormat == IMGFMT_NV12) | |
295 for(i=0; i<chrDstW; i++) | |
296 { | |
297 int u=1<<18; | |
298 int v=1<<18; | |
299 int j; | |
300 for(j=0; j<chrFilterSize; j++) | |
301 { | |
302 u += chrSrc[j][i] * chrFilter[j]; | |
303 v += chrSrc[j][i + 2048] * chrFilter[j]; | |
304 } | |
305 | |
306 uDest[2*i]= MIN(MAX(u>>19, 0), 255); | |
307 uDest[2*i+1]= MIN(MAX(v>>19, 0), 255); | |
308 } | |
309 else | |
310 for(i=0; i<chrDstW; i++) | |
311 { | |
312 int u=1<<18; | |
313 int v=1<<18; | |
314 int j; | |
315 for(j=0; j<chrFilterSize; j++) | |
316 { | |
317 u += chrSrc[j][i] * chrFilter[j]; | |
318 v += chrSrc[j][i + 2048] * chrFilter[j]; | |
319 } | |
320 | |
321 uDest[2*i]= MIN(MAX(v>>19, 0), 255); | |
322 uDest[2*i+1]= MIN(MAX(u>>19, 0), 255); | |
323 } | |
324 } | |
325 | |
326 #define YSCALE_YUV_2_PACKEDX_C(type) \ | |
327 for(i=0; i<(dstW>>1); i++){\ | |
328 int j;\ | |
329 int Y1=1<<18;\ | |
330 int Y2=1<<18;\ | |
331 int U=1<<18;\ | |
332 int V=1<<18;\ | |
333 type *r, *b, *g;\ | |
334 const int i2= 2*i;\ | |
335 \ | |
336 for(j=0; j<lumFilterSize; j++)\ | |
337 {\ | |
338 Y1 += lumSrc[j][i2] * lumFilter[j];\ | |
339 Y2 += lumSrc[j][i2+1] * lumFilter[j];\ | |
340 }\ | |
341 for(j=0; j<chrFilterSize; j++)\ | |
342 {\ | |
343 U += chrSrc[j][i] * chrFilter[j];\ | |
344 V += chrSrc[j][i+2048] * chrFilter[j];\ | |
345 }\ | |
346 Y1>>=19;\ | |
347 Y2>>=19;\ | |
348 U >>=19;\ | |
349 V >>=19;\ | |
350 if((Y1|Y2|U|V)&256)\ | |
351 {\ | |
352 if(Y1>255) Y1=255;\ | |
353 else if(Y1<0)Y1=0;\ | |
354 if(Y2>255) Y2=255;\ | |
355 else if(Y2<0)Y2=0;\ | |
356 if(U>255) U=255;\ | |
357 else if(U<0) U=0;\ | |
358 if(V>255) V=255;\ | |
359 else if(V<0) V=0;\ | |
360 } | |
361 | |
362 #define YSCALE_YUV_2_RGBX_C(type) \ | |
363 YSCALE_YUV_2_PACKEDX_C(type)\ | |
364 r = c->table_rV[V];\ | |
365 g = c->table_gU[U] + c->table_gV[V];\ | |
366 b = c->table_bU[U];\ | |
367 | |
368 #define YSCALE_YUV_2_PACKED2_C \ | |
369 for(i=0; i<(dstW>>1); i++){\ | |
370 const int i2= 2*i;\ | |
371 int Y1= (buf0[i2 ]*yalpha1+buf1[i2 ]*yalpha)>>19;\ | |
372 int Y2= (buf0[i2+1]*yalpha1+buf1[i2+1]*yalpha)>>19;\ | |
373 int U= (uvbuf0[i ]*uvalpha1+uvbuf1[i ]*uvalpha)>>19;\ | |
374 int V= (uvbuf0[i+2048]*uvalpha1+uvbuf1[i+2048]*uvalpha)>>19;\ | |
375 | |
376 #define YSCALE_YUV_2_RGB2_C(type) \ | |
377 YSCALE_YUV_2_PACKED2_C\ | |
378 type *r, *b, *g;\ | |
379 r = c->table_rV[V];\ | |
380 g = c->table_gU[U] + c->table_gV[V];\ | |
381 b = c->table_bU[U];\ | |
382 | |
383 #define YSCALE_YUV_2_PACKED1_C \ | |
384 for(i=0; i<(dstW>>1); i++){\ | |
385 const int i2= 2*i;\ | |
386 int Y1= buf0[i2 ]>>7;\ | |
387 int Y2= buf0[i2+1]>>7;\ | |
388 int U= (uvbuf1[i ])>>7;\ | |
389 int V= (uvbuf1[i+2048])>>7;\ | |
390 | |
391 #define YSCALE_YUV_2_RGB1_C(type) \ | |
392 YSCALE_YUV_2_PACKED1_C\ | |
393 type *r, *b, *g;\ | |
394 r = c->table_rV[V];\ | |
395 g = c->table_gU[U] + c->table_gV[V];\ | |
396 b = c->table_bU[U];\ | |
397 | |
398 #define YSCALE_YUV_2_PACKED1B_C \ | |
399 for(i=0; i<(dstW>>1); i++){\ | |
400 const int i2= 2*i;\ | |
401 int Y1= buf0[i2 ]>>7;\ | |
402 int Y2= buf0[i2+1]>>7;\ | |
403 int U= (uvbuf0[i ] + uvbuf1[i ])>>8;\ | |
404 int V= (uvbuf0[i+2048] + uvbuf1[i+2048])>>8;\ | |
405 | |
406 #define YSCALE_YUV_2_RGB1B_C(type) \ | |
407 YSCALE_YUV_2_PACKED1B_C\ | |
408 type *r, *b, *g;\ | |
409 r = c->table_rV[V];\ | |
410 g = c->table_gU[U] + c->table_gV[V];\ | |
411 b = c->table_bU[U];\ | |
412 | |
413 #define YSCALE_YUV_2_ANYRGB_C(func, func2)\ | |
414 switch(c->dstFormat)\ | |
415 {\ | |
416 case IMGFMT_BGR32:\ | |
417 case IMGFMT_RGB32:\ | |
418 func(uint32_t)\ | |
419 ((uint32_t*)dest)[i2+0]= r[Y1] + g[Y1] + b[Y1];\ | |
420 ((uint32_t*)dest)[i2+1]= r[Y2] + g[Y2] + b[Y2];\ | |
421 } \ | |
422 break;\ | |
423 case IMGFMT_RGB24:\ | |
424 func(uint8_t)\ | |
425 ((uint8_t*)dest)[0]= r[Y1];\ | |
426 ((uint8_t*)dest)[1]= g[Y1];\ | |
427 ((uint8_t*)dest)[2]= b[Y1];\ | |
428 ((uint8_t*)dest)[3]= r[Y2];\ | |
429 ((uint8_t*)dest)[4]= g[Y2];\ | |
430 ((uint8_t*)dest)[5]= b[Y2];\ | |
431 dest+=6;\ | |
432 }\ | |
433 break;\ | |
434 case IMGFMT_BGR24:\ | |
435 func(uint8_t)\ | |
436 ((uint8_t*)dest)[0]= b[Y1];\ | |
437 ((uint8_t*)dest)[1]= g[Y1];\ | |
438 ((uint8_t*)dest)[2]= r[Y1];\ | |
439 ((uint8_t*)dest)[3]= b[Y2];\ | |
440 ((uint8_t*)dest)[4]= g[Y2];\ | |
441 ((uint8_t*)dest)[5]= r[Y2];\ | |
442 dest+=6;\ | |
443 }\ | |
444 break;\ | |
445 case IMGFMT_RGB16:\ | |
446 case IMGFMT_BGR16:\ | |
447 {\ | |
448 const int dr1= dither_2x2_8[y&1 ][0];\ | |
449 const int dg1= dither_2x2_4[y&1 ][0];\ | |
450 const int db1= dither_2x2_8[(y&1)^1][0];\ | |
451 const int dr2= dither_2x2_8[y&1 ][1];\ | |
452 const int dg2= dither_2x2_4[y&1 ][1];\ | |
453 const int db2= dither_2x2_8[(y&1)^1][1];\ | |
454 func(uint16_t)\ | |
455 ((uint16_t*)dest)[i2+0]= r[Y1+dr1] + g[Y1+dg1] + b[Y1+db1];\ | |
456 ((uint16_t*)dest)[i2+1]= r[Y2+dr2] + g[Y2+dg2] + b[Y2+db2];\ | |
457 }\ | |
458 }\ | |
459 break;\ | |
460 case IMGFMT_RGB15:\ | |
461 case IMGFMT_BGR15:\ | |
462 {\ | |
463 const int dr1= dither_2x2_8[y&1 ][0];\ | |
464 const int dg1= dither_2x2_8[y&1 ][1];\ | |
465 const int db1= dither_2x2_8[(y&1)^1][0];\ | |
466 const int dr2= dither_2x2_8[y&1 ][1];\ | |
467 const int dg2= dither_2x2_8[y&1 ][0];\ | |
468 const int db2= dither_2x2_8[(y&1)^1][1];\ | |
469 func(uint16_t)\ | |
470 ((uint16_t*)dest)[i2+0]= r[Y1+dr1] + g[Y1+dg1] + b[Y1+db1];\ | |
471 ((uint16_t*)dest)[i2+1]= r[Y2+dr2] + g[Y2+dg2] + b[Y2+db2];\ | |
472 }\ | |
473 }\ | |
474 break;\ | |
475 case IMGFMT_RGB8:\ | |
476 case IMGFMT_BGR8:\ | |
477 {\ | |
478 const uint8_t * const d64= dither_8x8_73[y&7];\ | |
479 const uint8_t * const d32= dither_8x8_32[y&7];\ | |
480 func(uint8_t)\ | |
481 ((uint8_t*)dest)[i2+0]= r[Y1+d32[(i2+0)&7]] + g[Y1+d32[(i2+0)&7]] + b[Y1+d64[(i2+0)&7]];\ | |
482 ((uint8_t*)dest)[i2+1]= r[Y2+d32[(i2+1)&7]] + g[Y2+d32[(i2+1)&7]] + b[Y2+d64[(i2+1)&7]];\ | |
483 }\ | |
484 }\ | |
485 break;\ | |
486 case IMGFMT_RGB4:\ | |
487 case IMGFMT_BGR4:\ | |
488 {\ | |
489 const uint8_t * const d64= dither_8x8_73 [y&7];\ | |
490 const uint8_t * const d128=dither_8x8_220[y&7];\ | |
491 func(uint8_t)\ | |
492 ((uint8_t*)dest)[i]= r[Y1+d128[(i2+0)&7]] + g[Y1+d64[(i2+0)&7]] + b[Y1+d128[(i2+0)&7]]\ | |
493 + ((r[Y2+d128[(i2+1)&7]] + g[Y2+d64[(i2+1)&7]] + b[Y2+d128[(i2+1)&7]])<<4);\ | |
494 }\ | |
495 }\ | |
496 break;\ | |
497 case IMGFMT_RG4B:\ | |
498 case IMGFMT_BG4B:\ | |
499 {\ | |
500 const uint8_t * const d64= dither_8x8_73 [y&7];\ | |
501 const uint8_t * const d128=dither_8x8_220[y&7];\ | |
502 func(uint8_t)\ | |
503 ((uint8_t*)dest)[i2+0]= r[Y1+d128[(i2+0)&7]] + g[Y1+d64[(i2+0)&7]] + b[Y1+d128[(i2+0)&7]];\ | |
504 ((uint8_t*)dest)[i2+1]= r[Y2+d128[(i2+1)&7]] + g[Y2+d64[(i2+1)&7]] + b[Y2+d128[(i2+1)&7]];\ | |
505 }\ | |
506 }\ | |
507 break;\ | |
508 case IMGFMT_RGB1:\ | |
509 case IMGFMT_BGR1:\ | |
510 {\ | |
511 const uint8_t * const d128=dither_8x8_220[y&7];\ | |
512 uint8_t *g= c->table_gU[128] + c->table_gV[128];\ | |
513 for(i=0; i<dstW-7; i+=8){\ | |
514 int acc;\ | |
515 acc = g[((buf0[i ]*yalpha1+buf1[i ]*yalpha)>>19) + d128[0]];\ | |
516 acc+= acc + g[((buf0[i+1]*yalpha1+buf1[i+1]*yalpha)>>19) + d128[1]];\ | |
517 acc+= acc + g[((buf0[i+2]*yalpha1+buf1[i+2]*yalpha)>>19) + d128[2]];\ | |
518 acc+= acc + g[((buf0[i+3]*yalpha1+buf1[i+3]*yalpha)>>19) + d128[3]];\ | |
519 acc+= acc + g[((buf0[i+4]*yalpha1+buf1[i+4]*yalpha)>>19) + d128[4]];\ | |
520 acc+= acc + g[((buf0[i+5]*yalpha1+buf1[i+5]*yalpha)>>19) + d128[5]];\ | |
521 acc+= acc + g[((buf0[i+6]*yalpha1+buf1[i+6]*yalpha)>>19) + d128[6]];\ | |
522 acc+= acc + g[((buf0[i+7]*yalpha1+buf1[i+7]*yalpha)>>19) + d128[7]];\ | |
523 ((uint8_t*)dest)[0]= acc;\ | |
524 dest++;\ | |
525 }\ | |
526 \ | |
527 /*\ | |
528 ((uint8_t*)dest)-= dstW>>4;\ | |
529 {\ | |
530 int acc=0;\ | |
531 int left=0;\ | |
532 static int top[1024];\ | |
533 static int last_new[1024][1024];\ | |
534 static int last_in3[1024][1024];\ | |
535 static int drift[1024][1024];\ | |
536 int topLeft=0;\ | |
537 int shift=0;\ | |
538 int count=0;\ | |
539 const uint8_t * const d128=dither_8x8_220[y&7];\ | |
540 int error_new=0;\ | |
541 int error_in3=0;\ | |
542 int f=0;\ | |
543 \ | |
544 for(i=dstW>>1; i<dstW; i++){\ | |
545 int in= ((buf0[i ]*yalpha1+buf1[i ]*yalpha)>>19);\ | |
546 int in2 = (76309 * (in - 16) + 32768) >> 16;\ | |
547 int in3 = (in2 < 0) ? 0 : ((in2 > 255) ? 255 : in2);\ | |
548 int old= (left*7 + topLeft + top[i]*5 + top[i+1]*3)/20 + in3\ | |
549 + (last_new[y][i] - in3)*f/256;\ | |
550 int new= old> 128 ? 255 : 0;\ | |
551 \ | |
552 error_new+= ABS(last_new[y][i] - new);\ | |
553 error_in3+= ABS(last_in3[y][i] - in3);\ | |
554 f= error_new - error_in3*4;\ | |
555 if(f<0) f=0;\ | |
556 if(f>256) f=256;\ | |
557 \ | |
558 topLeft= top[i];\ | |
559 left= top[i]= old - new;\ | |
560 last_new[y][i]= new;\ | |
561 last_in3[y][i]= in3;\ | |
562 \ | |
563 acc+= acc + (new&1);\ | |
564 if((i&7)==6){\ | |
565 ((uint8_t*)dest)[0]= acc;\ | |
566 ((uint8_t*)dest)++;\ | |
567 }\ | |
568 }\ | |
569 }\ | |
570 */\ | |
571 }\ | |
572 break;\ | |
573 case IMGFMT_YUY2:\ | |
574 func2\ | |
575 ((uint8_t*)dest)[2*i2+0]= Y1;\ | |
576 ((uint8_t*)dest)[2*i2+1]= U;\ | |
577 ((uint8_t*)dest)[2*i2+2]= Y2;\ | |
578 ((uint8_t*)dest)[2*i2+3]= V;\ | |
579 } \ | |
580 break;\ | |
581 case IMGFMT_UYVY:\ | |
582 func2\ | |
583 ((uint8_t*)dest)[2*i2+0]= U;\ | |
584 ((uint8_t*)dest)[2*i2+1]= Y1;\ | |
585 ((uint8_t*)dest)[2*i2+2]= V;\ | |
586 ((uint8_t*)dest)[2*i2+3]= Y2;\ | |
587 } \ | |
588 break;\ | |
589 }\ | |
590 | |
591 | |
592 static inline void yuv2packedXinC(SwsContext *c, int16_t *lumFilter, int16_t **lumSrc, int lumFilterSize, | |
593 int16_t *chrFilter, int16_t **chrSrc, int chrFilterSize, | |
594 uint8_t *dest, int dstW, int y) | |
595 { | |
596 int i; | |
597 switch(c->dstFormat) | |
598 { | |
599 case IMGFMT_RGB32: | |
600 case IMGFMT_BGR32: | |
601 YSCALE_YUV_2_RGBX_C(uint32_t) | |
602 ((uint32_t*)dest)[i2+0]= r[Y1] + g[Y1] + b[Y1]; | |
603 ((uint32_t*)dest)[i2+1]= r[Y2] + g[Y2] + b[Y2]; | |
604 } | |
605 break; | |
606 case IMGFMT_RGB24: | |
607 YSCALE_YUV_2_RGBX_C(uint8_t) | |
608 ((uint8_t*)dest)[0]= r[Y1]; | |
609 ((uint8_t*)dest)[1]= g[Y1]; | |
610 ((uint8_t*)dest)[2]= b[Y1]; | |
611 ((uint8_t*)dest)[3]= r[Y2]; | |
612 ((uint8_t*)dest)[4]= g[Y2]; | |
613 ((uint8_t*)dest)[5]= b[Y2]; | |
614 dest+=6; | |
615 } | |
616 break; | |
617 case IMGFMT_BGR24: | |
618 YSCALE_YUV_2_RGBX_C(uint8_t) | |
619 ((uint8_t*)dest)[0]= b[Y1]; | |
620 ((uint8_t*)dest)[1]= g[Y1]; | |
621 ((uint8_t*)dest)[2]= r[Y1]; | |
622 ((uint8_t*)dest)[3]= b[Y2]; | |
623 ((uint8_t*)dest)[4]= g[Y2]; | |
624 ((uint8_t*)dest)[5]= r[Y2]; | |
625 dest+=6; | |
626 } | |
627 break; | |
628 case IMGFMT_RGB16: | |
629 case IMGFMT_BGR16: | |
630 { | |
631 const int dr1= dither_2x2_8[y&1 ][0]; | |
632 const int dg1= dither_2x2_4[y&1 ][0]; | |
633 const int db1= dither_2x2_8[(y&1)^1][0]; | |
634 const int dr2= dither_2x2_8[y&1 ][1]; | |
635 const int dg2= dither_2x2_4[y&1 ][1]; | |
636 const int db2= dither_2x2_8[(y&1)^1][1]; | |
637 YSCALE_YUV_2_RGBX_C(uint16_t) | |
638 ((uint16_t*)dest)[i2+0]= r[Y1+dr1] + g[Y1+dg1] + b[Y1+db1]; | |
639 ((uint16_t*)dest)[i2+1]= r[Y2+dr2] + g[Y2+dg2] + b[Y2+db2]; | |
640 } | |
641 } | |
642 break; | |
643 case IMGFMT_RGB15: | |
644 case IMGFMT_BGR15: | |
645 { | |
646 const int dr1= dither_2x2_8[y&1 ][0]; | |
647 const int dg1= dither_2x2_8[y&1 ][1]; | |
648 const int db1= dither_2x2_8[(y&1)^1][0]; | |
649 const int dr2= dither_2x2_8[y&1 ][1]; | |
650 const int dg2= dither_2x2_8[y&1 ][0]; | |
651 const int db2= dither_2x2_8[(y&1)^1][1]; | |
652 YSCALE_YUV_2_RGBX_C(uint16_t) | |
653 ((uint16_t*)dest)[i2+0]= r[Y1+dr1] + g[Y1+dg1] + b[Y1+db1]; | |
654 ((uint16_t*)dest)[i2+1]= r[Y2+dr2] + g[Y2+dg2] + b[Y2+db2]; | |
655 } | |
656 } | |
657 break; | |
658 case IMGFMT_RGB8: | |
659 case IMGFMT_BGR8: | |
660 { | |
661 const uint8_t * const d64= dither_8x8_73[y&7]; | |
662 const uint8_t * const d32= dither_8x8_32[y&7]; | |
663 YSCALE_YUV_2_RGBX_C(uint8_t) | |
664 ((uint8_t*)dest)[i2+0]= r[Y1+d32[(i2+0)&7]] + g[Y1+d32[(i2+0)&7]] + b[Y1+d64[(i2+0)&7]]; | |
665 ((uint8_t*)dest)[i2+1]= r[Y2+d32[(i2+1)&7]] + g[Y2+d32[(i2+1)&7]] + b[Y2+d64[(i2+1)&7]]; | |
666 } | |
667 } | |
668 break; | |
669 case IMGFMT_RGB4: | |
670 case IMGFMT_BGR4: | |
671 { | |
672 const uint8_t * const d64= dither_8x8_73 [y&7]; | |
673 const uint8_t * const d128=dither_8x8_220[y&7]; | |
674 YSCALE_YUV_2_RGBX_C(uint8_t) | |
675 ((uint8_t*)dest)[i]= r[Y1+d128[(i2+0)&7]] + g[Y1+d64[(i2+0)&7]] + b[Y1+d128[(i2+0)&7]] | |
676 +((r[Y2+d128[(i2+1)&7]] + g[Y2+d64[(i2+1)&7]] + b[Y2+d128[(i2+1)&7]])<<4); | |
677 } | |
678 } | |
679 break; | |
680 case IMGFMT_RG4B: | |
681 case IMGFMT_BG4B: | |
682 { | |
683 const uint8_t * const d64= dither_8x8_73 [y&7]; | |
684 const uint8_t * const d128=dither_8x8_220[y&7]; | |
685 YSCALE_YUV_2_RGBX_C(uint8_t) | |
686 ((uint8_t*)dest)[i2+0]= r[Y1+d128[(i2+0)&7]] + g[Y1+d64[(i2+0)&7]] + b[Y1+d128[(i2+0)&7]]; | |
687 ((uint8_t*)dest)[i2+1]= r[Y2+d128[(i2+1)&7]] + g[Y2+d64[(i2+1)&7]] + b[Y2+d128[(i2+1)&7]]; | |
688 } | |
689 } | |
690 break; | |
691 case IMGFMT_RGB1: | |
692 case IMGFMT_BGR1: | |
693 { | |
694 const uint8_t * const d128=dither_8x8_220[y&7]; | |
695 uint8_t *g= c->table_gU[128] + c->table_gV[128]; | |
696 int acc=0; | |
697 for(i=0; i<dstW-1; i+=2){ | |
698 int j; | |
699 int Y1=1<<18; | |
700 int Y2=1<<18; | |
701 | |
702 for(j=0; j<lumFilterSize; j++) | |
703 { | |
704 Y1 += lumSrc[j][i] * lumFilter[j]; | |
705 Y2 += lumSrc[j][i+1] * lumFilter[j]; | |
706 } | |
707 Y1>>=19; | |
708 Y2>>=19; | |
709 if((Y1|Y2)&256) | |
710 { | |
711 if(Y1>255) Y1=255; | |
712 else if(Y1<0)Y1=0; | |
713 if(Y2>255) Y2=255; | |
714 else if(Y2<0)Y2=0; | |
715 } | |
716 acc+= acc + g[Y1+d128[(i+0)&7]]; | |
717 acc+= acc + g[Y2+d128[(i+1)&7]]; | |
718 if((i&7)==6){ | |
719 ((uint8_t*)dest)[0]= acc; | |
720 dest++; | |
721 } | |
722 } | |
723 } | |
724 break; | |
725 case IMGFMT_YUY2: | |
726 YSCALE_YUV_2_PACKEDX_C(void) | |
727 ((uint8_t*)dest)[2*i2+0]= Y1; | |
728 ((uint8_t*)dest)[2*i2+1]= U; | |
729 ((uint8_t*)dest)[2*i2+2]= Y2; | |
730 ((uint8_t*)dest)[2*i2+3]= V; | |
731 } | |
732 break; | |
733 case IMGFMT_UYVY: | |
734 YSCALE_YUV_2_PACKEDX_C(void) | |
735 ((uint8_t*)dest)[2*i2+0]= U; | |
736 ((uint8_t*)dest)[2*i2+1]= Y1; | |
737 ((uint8_t*)dest)[2*i2+2]= V; | |
738 ((uint8_t*)dest)[2*i2+3]= Y2; | |
739 } | |
740 break; | |
741 } | |
742 } | |
743 | |
744 | |
745 //Note: we have C, X86, MMX, MMX2, 3DNOW version therse no 3DNOW+MMX2 one | |
746 //Plain C versions | |
747 #if !defined (HAVE_MMX) || defined (RUNTIME_CPUDETECT) | |
748 #define COMPILE_C | |
749 #endif | |
750 | |
751 #ifdef ARCH_POWERPC | |
752 #if defined (HAVE_ALTIVEC) || defined (RUNTIME_CPUDETECT) | |
753 #define COMPILE_ALTIVEC | |
754 #endif //HAVE_ALTIVEC | |
755 #endif //ARCH_POWERPC | |
756 | |
757 #if defined(ARCH_X86) || defined(ARCH_X86_64) | |
758 | |
759 #if (defined (HAVE_MMX) && !defined (HAVE_3DNOW) && !defined (HAVE_MMX2)) || defined (RUNTIME_CPUDETECT) | |
760 #define COMPILE_MMX | |
761 #endif | |
762 | |
763 #if defined (HAVE_MMX2) || defined (RUNTIME_CPUDETECT) | |
764 #define COMPILE_MMX2 | |
765 #endif | |
766 | |
767 #if (defined (HAVE_3DNOW) && !defined (HAVE_MMX2)) || defined (RUNTIME_CPUDETECT) | |
768 #define COMPILE_3DNOW | |
769 #endif | |
770 #endif //ARCH_X86 || ARCH_X86_64 | |
771 | |
772 #undef HAVE_MMX | |
773 #undef HAVE_MMX2 | |
774 #undef HAVE_3DNOW | |
775 | |
776 #ifdef COMPILE_C | |
777 #undef HAVE_MMX | |
778 #undef HAVE_MMX2 | |
779 #undef HAVE_3DNOW | |
780 #undef HAVE_ALTIVEC | |
781 #define RENAME(a) a ## _C | |
782 #include "swscale_template.c" | |
783 #endif | |
784 | |
785 #ifdef ARCH_POWERPC | |
786 #ifdef COMPILE_ALTIVEC | |
787 #undef RENAME | |
788 #define HAVE_ALTIVEC | |
789 #define RENAME(a) a ## _altivec | |
790 #include "swscale_template.c" | |
791 #endif | |
792 #endif //ARCH_POWERPC | |
793 | |
794 #if defined(ARCH_X86) || defined(ARCH_X86_64) | |
795 | |
796 //X86 versions | |
797 /* | |
798 #undef RENAME | |
799 #undef HAVE_MMX | |
800 #undef HAVE_MMX2 | |
801 #undef HAVE_3DNOW | |
802 #define ARCH_X86 | |
803 #define RENAME(a) a ## _X86 | |
804 #include "swscale_template.c" | |
805 */ | |
806 //MMX versions | |
807 #ifdef COMPILE_MMX | |
808 #undef RENAME | |
809 #define HAVE_MMX | |
810 #undef HAVE_MMX2 | |
811 #undef HAVE_3DNOW | |
812 #define RENAME(a) a ## _MMX | |
813 #include "swscale_template.c" | |
814 #endif | |
815 | |
816 //MMX2 versions | |
817 #ifdef COMPILE_MMX2 | |
818 #undef RENAME | |
819 #define HAVE_MMX | |
820 #define HAVE_MMX2 | |
821 #undef HAVE_3DNOW | |
822 #define RENAME(a) a ## _MMX2 | |
823 #include "swscale_template.c" | |
824 #endif | |
825 | |
826 //3DNOW versions | |
827 #ifdef COMPILE_3DNOW | |
828 #undef RENAME | |
829 #define HAVE_MMX | |
830 #undef HAVE_MMX2 | |
831 #define HAVE_3DNOW | |
832 #define RENAME(a) a ## _3DNow | |
833 #include "swscale_template.c" | |
834 #endif | |
835 | |
836 #endif //ARCH_X86 || ARCH_X86_64 | |
837 | |
838 // minor note: the HAVE_xyz is messed up after that line so don't use it | |
839 | |
840 static double getSplineCoeff(double a, double b, double c, double d, double dist) | |
841 { | |
842 // printf("%f %f %f %f %f\n", a,b,c,d,dist); | |
843 if(dist<=1.0) return ((d*dist + c)*dist + b)*dist +a; | |
844 else return getSplineCoeff( 0.0, | |
845 b+ 2.0*c + 3.0*d, | |
846 c + 3.0*d, | |
847 -b- 3.0*c - 6.0*d, | |
848 dist-1.0); | |
849 } | |
850 | |
19172
bae6c99a99cc
vertical scaler with accurate rounding, some people on doom9 can see +-1 errors
michael
parents:
19169
diff
changeset
|
851 static inline int initFilter(int16_t **outFilter, int16_t **filterPos, int *outFilterSize, int xInc, |
18861 | 852 int srcW, int dstW, int filterAlign, int one, int flags, |
853 SwsVector *srcFilter, SwsVector *dstFilter, double param[2]) | |
854 { | |
855 int i; | |
856 int filterSize; | |
857 int filter2Size; | |
858 int minFilterSize; | |
859 double *filter=NULL; | |
860 double *filter2=NULL; | |
861 #if defined(ARCH_X86) || defined(ARCH_X86_64) | |
862 if(flags & SWS_CPU_CAPS_MMX) | |
863 asm volatile("emms\n\t"::: "memory"); //FIXME this shouldnt be required but it IS (even for non mmx versions) | |
864 #endif | |
865 | |
866 // Note the +1 is for the MMXscaler which reads over the end | |
19168
624173ae90b9
Missing part of the malloc -> av_malloc patch: memalign must be replaced as well!
reimar
parents:
19143
diff
changeset
|
867 *filterPos = av_malloc((dstW+1)*sizeof(int16_t)); |
18861 | 868 |
869 if(ABS(xInc - 0x10000) <10) // unscaled | |
870 { | |
871 int i; | |
872 filterSize= 1; | |
19168
624173ae90b9
Missing part of the malloc -> av_malloc patch: memalign must be replaced as well!
reimar
parents:
19143
diff
changeset
|
873 filter= av_malloc(dstW*sizeof(double)*filterSize); |
18861 | 874 for(i=0; i<dstW*filterSize; i++) filter[i]=0; |
875 | |
876 for(i=0; i<dstW; i++) | |
877 { | |
878 filter[i*filterSize]=1; | |
879 (*filterPos)[i]=i; | |
880 } | |
881 | |
882 } | |
883 else if(flags&SWS_POINT) // lame looking point sampling mode | |
884 { | |
885 int i; | |
886 int xDstInSrc; | |
887 filterSize= 1; | |
19168
624173ae90b9
Missing part of the malloc -> av_malloc patch: memalign must be replaced as well!
reimar
parents:
19143
diff
changeset
|
888 filter= av_malloc(dstW*sizeof(double)*filterSize); |
18861 | 889 |
890 xDstInSrc= xInc/2 - 0x8000; | |
891 for(i=0; i<dstW; i++) | |
892 { | |
893 int xx= (xDstInSrc - ((filterSize-1)<<15) + (1<<15))>>16; | |
894 | |
895 (*filterPos)[i]= xx; | |
896 filter[i]= 1.0; | |
897 xDstInSrc+= xInc; | |
898 } | |
899 } | |
900 else if((xInc <= (1<<16) && (flags&SWS_AREA)) || (flags&SWS_FAST_BILINEAR)) // bilinear upscale | |
901 { | |
902 int i; | |
903 int xDstInSrc; | |
904 if (flags&SWS_BICUBIC) filterSize= 4; | |
905 else if(flags&SWS_X ) filterSize= 4; | |
906 else filterSize= 2; // SWS_BILINEAR / SWS_AREA | |
19168
624173ae90b9
Missing part of the malloc -> av_malloc patch: memalign must be replaced as well!
reimar
parents:
19143
diff
changeset
|
907 filter= av_malloc(dstW*sizeof(double)*filterSize); |
18861 | 908 |
909 xDstInSrc= xInc/2 - 0x8000; | |
910 for(i=0; i<dstW; i++) | |
911 { | |
912 int xx= (xDstInSrc - ((filterSize-1)<<15) + (1<<15))>>16; | |
913 int j; | |
914 | |
915 (*filterPos)[i]= xx; | |
916 //Bilinear upscale / linear interpolate / Area averaging | |
917 for(j=0; j<filterSize; j++) | |
918 { | |
919 double d= ABS((xx<<16) - xDstInSrc)/(double)(1<<16); | |
920 double coeff= 1.0 - d; | |
921 if(coeff<0) coeff=0; | |
922 filter[i*filterSize + j]= coeff; | |
923 xx++; | |
924 } | |
925 xDstInSrc+= xInc; | |
926 } | |
927 } | |
928 else | |
929 { | |
930 double xDstInSrc; | |
931 double sizeFactor, filterSizeInSrc; | |
932 const double xInc1= (double)xInc / (double)(1<<16); | |
933 | |
934 if (flags&SWS_BICUBIC) sizeFactor= 4.0; | |
935 else if(flags&SWS_X) sizeFactor= 8.0; | |
936 else if(flags&SWS_AREA) sizeFactor= 1.0; //downscale only, for upscale it is bilinear | |
937 else if(flags&SWS_GAUSS) sizeFactor= 8.0; // infinite ;) | |
938 else if(flags&SWS_LANCZOS) sizeFactor= param[0] != SWS_PARAM_DEFAULT ? 2.0*param[0] : 6.0; | |
939 else if(flags&SWS_SINC) sizeFactor= 20.0; // infinite ;) | |
940 else if(flags&SWS_SPLINE) sizeFactor= 20.0; // infinite ;) | |
941 else if(flags&SWS_BILINEAR) sizeFactor= 2.0; | |
942 else { | |
943 sizeFactor= 0.0; //GCC warning killer | |
944 ASSERT(0) | |
945 } | |
946 | |
947 if(xInc1 <= 1.0) filterSizeInSrc= sizeFactor; // upscale | |
948 else filterSizeInSrc= sizeFactor*srcW / (double)dstW; | |
949 | |
950 filterSize= (int)ceil(1 + filterSizeInSrc); // will be reduced later if possible | |
951 if(filterSize > srcW-2) filterSize=srcW-2; | |
952 | |
19168
624173ae90b9
Missing part of the malloc -> av_malloc patch: memalign must be replaced as well!
reimar
parents:
19143
diff
changeset
|
953 filter= av_malloc(dstW*sizeof(double)*filterSize); |
18861 | 954 |
955 xDstInSrc= xInc1 / 2.0 - 0.5; | |
956 for(i=0; i<dstW; i++) | |
957 { | |
958 int xx= (int)(xDstInSrc - (filterSize-1)*0.5 + 0.5); | |
959 int j; | |
960 (*filterPos)[i]= xx; | |
961 for(j=0; j<filterSize; j++) | |
962 { | |
963 double d= ABS(xx - xDstInSrc)/filterSizeInSrc*sizeFactor; | |
964 double coeff; | |
965 if(flags & SWS_BICUBIC) | |
966 { | |
967 double B= param[0] != SWS_PARAM_DEFAULT ? param[0] : 0.0; | |
968 double C= param[1] != SWS_PARAM_DEFAULT ? param[1] : 0.6; | |
969 | |
970 if(d<1.0) | |
971 coeff = (12-9*B-6*C)*d*d*d + (-18+12*B+6*C)*d*d + 6-2*B; | |
972 else if(d<2.0) | |
973 coeff = (-B-6*C)*d*d*d + (6*B+30*C)*d*d + (-12*B-48*C)*d +8*B+24*C; | |
974 else | |
975 coeff=0.0; | |
976 } | |
977 /* else if(flags & SWS_X) | |
978 { | |
979 double p= param ? param*0.01 : 0.3; | |
980 coeff = d ? sin(d*PI)/(d*PI) : 1.0; | |
981 coeff*= pow(2.0, - p*d*d); | |
982 }*/ | |
983 else if(flags & SWS_X) | |
984 { | |
985 double A= param[0] != SWS_PARAM_DEFAULT ? param[0] : 1.0; | |
986 | |
987 if(d<1.0) | |
988 coeff = cos(d*PI); | |
989 else | |
990 coeff=-1.0; | |
991 if(coeff<0.0) coeff= -pow(-coeff, A); | |
992 else coeff= pow( coeff, A); | |
993 coeff= coeff*0.5 + 0.5; | |
994 } | |
995 else if(flags & SWS_AREA) | |
996 { | |
997 double srcPixelSize= 1.0/xInc1; | |
998 if(d + srcPixelSize/2 < 0.5) coeff= 1.0; | |
999 else if(d - srcPixelSize/2 < 0.5) coeff= (0.5-d)/srcPixelSize + 0.5; | |
1000 else coeff=0.0; | |
1001 } | |
1002 else if(flags & SWS_GAUSS) | |
1003 { | |
1004 double p= param[0] != SWS_PARAM_DEFAULT ? param[0] : 3.0; | |
1005 coeff = pow(2.0, - p*d*d); | |
1006 } | |
1007 else if(flags & SWS_SINC) | |
1008 { | |
1009 coeff = d ? sin(d*PI)/(d*PI) : 1.0; | |
1010 } | |
1011 else if(flags & SWS_LANCZOS) | |
1012 { | |
1013 double p= param[0] != SWS_PARAM_DEFAULT ? param[0] : 3.0; | |
1014 coeff = d ? sin(d*PI)*sin(d*PI/p)/(d*d*PI*PI/p) : 1.0; | |
1015 if(d>p) coeff=0; | |
1016 } | |
1017 else if(flags & SWS_BILINEAR) | |
1018 { | |
1019 coeff= 1.0 - d; | |
1020 if(coeff<0) coeff=0; | |
1021 } | |
1022 else if(flags & SWS_SPLINE) | |
1023 { | |
1024 double p=-2.196152422706632; | |
1025 coeff = getSplineCoeff(1.0, 0.0, p, -p-1.0, d); | |
1026 } | |
1027 else { | |
1028 coeff= 0.0; //GCC warning killer | |
1029 ASSERT(0) | |
1030 } | |
1031 | |
1032 filter[i*filterSize + j]= coeff; | |
1033 xx++; | |
1034 } | |
1035 xDstInSrc+= xInc1; | |
1036 } | |
1037 } | |
1038 | |
1039 /* apply src & dst Filter to filter -> filter2 | |
19143
c4dac777b44c
Use libavutil in libswscale, and allow it to be built out of the mplayer tree
lucabe
parents:
18861
diff
changeset
|
1040 av_free(filter); |
18861 | 1041 */ |
1042 ASSERT(filterSize>0) | |
1043 filter2Size= filterSize; | |
1044 if(srcFilter) filter2Size+= srcFilter->length - 1; | |
1045 if(dstFilter) filter2Size+= dstFilter->length - 1; | |
1046 ASSERT(filter2Size>0) | |
19168
624173ae90b9
Missing part of the malloc -> av_malloc patch: memalign must be replaced as well!
reimar
parents:
19143
diff
changeset
|
1047 filter2= av_malloc(filter2Size*dstW*sizeof(double)); |
18861 | 1048 |
1049 for(i=0; i<dstW; i++) | |
1050 { | |
1051 int j; | |
1052 SwsVector scaleFilter; | |
1053 SwsVector *outVec; | |
1054 | |
1055 scaleFilter.coeff= filter + i*filterSize; | |
1056 scaleFilter.length= filterSize; | |
1057 | |
1058 if(srcFilter) outVec= sws_getConvVec(srcFilter, &scaleFilter); | |
1059 else outVec= &scaleFilter; | |
1060 | |
1061 ASSERT(outVec->length == filter2Size) | |
1062 //FIXME dstFilter | |
1063 | |
1064 for(j=0; j<outVec->length; j++) | |
1065 { | |
1066 filter2[i*filter2Size + j]= outVec->coeff[j]; | |
1067 } | |
1068 | |
1069 (*filterPos)[i]+= (filterSize-1)/2 - (filter2Size-1)/2; | |
1070 | |
1071 if(outVec != &scaleFilter) sws_freeVec(outVec); | |
1072 } | |
19143
c4dac777b44c
Use libavutil in libswscale, and allow it to be built out of the mplayer tree
lucabe
parents:
18861
diff
changeset
|
1073 av_free(filter); filter=NULL; |
18861 | 1074 |
1075 /* try to reduce the filter-size (step1 find size and shift left) */ | |
1076 // Assume its near normalized (*0.5 or *2.0 is ok but * 0.001 is not) | |
1077 minFilterSize= 0; | |
1078 for(i=dstW-1; i>=0; i--) | |
1079 { | |
1080 int min= filter2Size; | |
1081 int j; | |
1082 double cutOff=0.0; | |
1083 | |
1084 /* get rid off near zero elements on the left by shifting left */ | |
1085 for(j=0; j<filter2Size; j++) | |
1086 { | |
1087 int k; | |
1088 cutOff += ABS(filter2[i*filter2Size]); | |
1089 | |
1090 if(cutOff > SWS_MAX_REDUCE_CUTOFF) break; | |
1091 | |
1092 /* preserve Monotonicity because the core can't handle the filter otherwise */ | |
1093 if(i<dstW-1 && (*filterPos)[i] >= (*filterPos)[i+1]) break; | |
1094 | |
1095 // Move filter coeffs left | |
1096 for(k=1; k<filter2Size; k++) | |
1097 filter2[i*filter2Size + k - 1]= filter2[i*filter2Size + k]; | |
1098 filter2[i*filter2Size + k - 1]= 0.0; | |
1099 (*filterPos)[i]++; | |
1100 } | |
1101 | |
1102 cutOff=0.0; | |
1103 /* count near zeros on the right */ | |
1104 for(j=filter2Size-1; j>0; j--) | |
1105 { | |
1106 cutOff += ABS(filter2[i*filter2Size + j]); | |
1107 | |
1108 if(cutOff > SWS_MAX_REDUCE_CUTOFF) break; | |
1109 min--; | |
1110 } | |
1111 | |
1112 if(min>minFilterSize) minFilterSize= min; | |
1113 } | |
1114 | |
1115 if (flags & SWS_CPU_CAPS_ALTIVEC) { | |
1116 // we can handle the special case 4, | |
1117 // so we don't want to go to the full 8 | |
1118 if (minFilterSize < 5) | |
1119 filterAlign = 4; | |
1120 | |
1121 // we really don't want to waste our time | |
1122 // doing useless computation, so fall-back on | |
1123 // the scalar C code for very small filter. | |
1124 // vectorizing is worth it only if you have | |
1125 // decent-sized vector. | |
1126 if (minFilterSize < 3) | |
1127 filterAlign = 1; | |
1128 } | |
1129 | |
19172
bae6c99a99cc
vertical scaler with accurate rounding, some people on doom9 can see +-1 errors
michael
parents:
19169
diff
changeset
|
1130 if (flags & SWS_CPU_CAPS_MMX) { |
bae6c99a99cc
vertical scaler with accurate rounding, some people on doom9 can see +-1 errors
michael
parents:
19169
diff
changeset
|
1131 // special case for unscaled vertical filtering |
bae6c99a99cc
vertical scaler with accurate rounding, some people on doom9 can see +-1 errors
michael
parents:
19169
diff
changeset
|
1132 if(minFilterSize == 1 && filterAlign == 2) |
bae6c99a99cc
vertical scaler with accurate rounding, some people on doom9 can see +-1 errors
michael
parents:
19169
diff
changeset
|
1133 filterAlign= 1; |
bae6c99a99cc
vertical scaler with accurate rounding, some people on doom9 can see +-1 errors
michael
parents:
19169
diff
changeset
|
1134 } |
bae6c99a99cc
vertical scaler with accurate rounding, some people on doom9 can see +-1 errors
michael
parents:
19169
diff
changeset
|
1135 |
18861 | 1136 ASSERT(minFilterSize > 0) |
1137 filterSize= (minFilterSize +(filterAlign-1)) & (~(filterAlign-1)); | |
1138 ASSERT(filterSize > 0) | |
19168
624173ae90b9
Missing part of the malloc -> av_malloc patch: memalign must be replaced as well!
reimar
parents:
19143
diff
changeset
|
1139 filter= av_malloc(filterSize*dstW*sizeof(double)); |
19172
bae6c99a99cc
vertical scaler with accurate rounding, some people on doom9 can see +-1 errors
michael
parents:
19169
diff
changeset
|
1140 if(filterSize >= MAX_FILTER_SIZE) |
bae6c99a99cc
vertical scaler with accurate rounding, some people on doom9 can see +-1 errors
michael
parents:
19169
diff
changeset
|
1141 return -1; |
18861 | 1142 *outFilterSize= filterSize; |
1143 | |
1144 if(flags&SWS_PRINT_INFO) | |
1145 MSG_V("SwScaler: reducing / aligning filtersize %d -> %d\n", filter2Size, filterSize); | |
1146 /* try to reduce the filter-size (step2 reduce it) */ | |
1147 for(i=0; i<dstW; i++) | |
1148 { | |
1149 int j; | |
1150 | |
1151 for(j=0; j<filterSize; j++) | |
1152 { | |
1153 if(j>=filter2Size) filter[i*filterSize + j]= 0.0; | |
1154 else filter[i*filterSize + j]= filter2[i*filter2Size + j]; | |
1155 } | |
1156 } | |
19143
c4dac777b44c
Use libavutil in libswscale, and allow it to be built out of the mplayer tree
lucabe
parents:
18861
diff
changeset
|
1157 av_free(filter2); filter2=NULL; |
18861 | 1158 |
1159 | |
1160 //FIXME try to align filterpos if possible | |
1161 | |
1162 //fix borders | |
1163 for(i=0; i<dstW; i++) | |
1164 { | |
1165 int j; | |
1166 if((*filterPos)[i] < 0) | |
1167 { | |
1168 // Move filter coeffs left to compensate for filterPos | |
1169 for(j=1; j<filterSize; j++) | |
1170 { | |
1171 int left= MAX(j + (*filterPos)[i], 0); | |
1172 filter[i*filterSize + left] += filter[i*filterSize + j]; | |
1173 filter[i*filterSize + j]=0; | |
1174 } | |
1175 (*filterPos)[i]= 0; | |
1176 } | |
1177 | |
1178 if((*filterPos)[i] + filterSize > srcW) | |
1179 { | |
1180 int shift= (*filterPos)[i] + filterSize - srcW; | |
1181 // Move filter coeffs right to compensate for filterPos | |
1182 for(j=filterSize-2; j>=0; j--) | |
1183 { | |
1184 int right= MIN(j + shift, filterSize-1); | |
1185 filter[i*filterSize +right] += filter[i*filterSize +j]; | |
1186 filter[i*filterSize +j]=0; | |
1187 } | |
1188 (*filterPos)[i]= srcW - filterSize; | |
1189 } | |
1190 } | |
1191 | |
1192 // Note the +1 is for the MMXscaler which reads over the end | |
1193 /* align at 16 for AltiVec (needed by hScale_altivec_real) */ | |
19168
624173ae90b9
Missing part of the malloc -> av_malloc patch: memalign must be replaced as well!
reimar
parents:
19143
diff
changeset
|
1194 *outFilter= av_malloc(*outFilterSize*(dstW+1)*sizeof(int16_t)); |
18861 | 1195 memset(*outFilter, 0, *outFilterSize*(dstW+1)*sizeof(int16_t)); |
1196 | |
1197 /* Normalize & Store in outFilter */ | |
1198 for(i=0; i<dstW; i++) | |
1199 { | |
1200 int j; | |
1201 double error=0; | |
1202 double sum=0; | |
1203 double scale= one; | |
1204 | |
1205 for(j=0; j<filterSize; j++) | |
1206 { | |
1207 sum+= filter[i*filterSize + j]; | |
1208 } | |
1209 scale/= sum; | |
1210 for(j=0; j<*outFilterSize; j++) | |
1211 { | |
1212 double v= filter[i*filterSize + j]*scale + error; | |
1213 int intV= floor(v + 0.5); | |
1214 (*outFilter)[i*(*outFilterSize) + j]= intV; | |
1215 error = v - intV; | |
1216 } | |
1217 } | |
1218 | |
1219 (*filterPos)[dstW]= (*filterPos)[dstW-1]; // the MMX scaler will read over the end | |
1220 for(i=0; i<*outFilterSize; i++) | |
1221 { | |
1222 int j= dstW*(*outFilterSize); | |
1223 (*outFilter)[j + i]= (*outFilter)[j + i - (*outFilterSize)]; | |
1224 } | |
1225 | |
19143
c4dac777b44c
Use libavutil in libswscale, and allow it to be built out of the mplayer tree
lucabe
parents:
18861
diff
changeset
|
1226 av_free(filter); |
19172
bae6c99a99cc
vertical scaler with accurate rounding, some people on doom9 can see +-1 errors
michael
parents:
19169
diff
changeset
|
1227 return 0; |
18861 | 1228 } |
1229 | |
1230 #if defined(ARCH_X86) || defined(ARCH_X86_64) | |
1231 static void initMMX2HScaler(int dstW, int xInc, uint8_t *funnyCode, int16_t *filter, int32_t *filterPos, int numSplits) | |
1232 { | |
1233 uint8_t *fragmentA; | |
1234 long imm8OfPShufW1A; | |
1235 long imm8OfPShufW2A; | |
1236 long fragmentLengthA; | |
1237 uint8_t *fragmentB; | |
1238 long imm8OfPShufW1B; | |
1239 long imm8OfPShufW2B; | |
1240 long fragmentLengthB; | |
1241 int fragmentPos; | |
1242 | |
1243 int xpos, i; | |
1244 | |
1245 // create an optimized horizontal scaling routine | |
1246 | |
1247 //code fragment | |
1248 | |
1249 asm volatile( | |
1250 "jmp 9f \n\t" | |
1251 // Begin | |
1252 "0: \n\t" | |
1253 "movq (%%"REG_d", %%"REG_a"), %%mm3\n\t" | |
1254 "movd (%%"REG_c", %%"REG_S"), %%mm0\n\t" | |
1255 "movd 1(%%"REG_c", %%"REG_S"), %%mm1\n\t" | |
1256 "punpcklbw %%mm7, %%mm1 \n\t" | |
1257 "punpcklbw %%mm7, %%mm0 \n\t" | |
1258 "pshufw $0xFF, %%mm1, %%mm1 \n\t" | |
1259 "1: \n\t" | |
1260 "pshufw $0xFF, %%mm0, %%mm0 \n\t" | |
1261 "2: \n\t" | |
1262 "psubw %%mm1, %%mm0 \n\t" | |
1263 "movl 8(%%"REG_b", %%"REG_a"), %%esi\n\t" | |
1264 "pmullw %%mm3, %%mm0 \n\t" | |
1265 "psllw $7, %%mm1 \n\t" | |
1266 "paddw %%mm1, %%mm0 \n\t" | |
1267 | |
1268 "movq %%mm0, (%%"REG_D", %%"REG_a")\n\t" | |
1269 | |
1270 "add $8, %%"REG_a" \n\t" | |
1271 // End | |
1272 "9: \n\t" | |
1273 // "int $3\n\t" | |
1274 "lea 0b, %0 \n\t" | |
1275 "lea 1b, %1 \n\t" | |
1276 "lea 2b, %2 \n\t" | |
1277 "dec %1 \n\t" | |
1278 "dec %2 \n\t" | |
1279 "sub %0, %1 \n\t" | |
1280 "sub %0, %2 \n\t" | |
1281 "lea 9b, %3 \n\t" | |
1282 "sub %0, %3 \n\t" | |
1283 | |
1284 | |
1285 :"=r" (fragmentA), "=r" (imm8OfPShufW1A), "=r" (imm8OfPShufW2A), | |
1286 "=r" (fragmentLengthA) | |
1287 ); | |
1288 | |
1289 asm volatile( | |
1290 "jmp 9f \n\t" | |
1291 // Begin | |
1292 "0: \n\t" | |
1293 "movq (%%"REG_d", %%"REG_a"), %%mm3\n\t" | |
1294 "movd (%%"REG_c", %%"REG_S"), %%mm0\n\t" | |
1295 "punpcklbw %%mm7, %%mm0 \n\t" | |
1296 "pshufw $0xFF, %%mm0, %%mm1 \n\t" | |
1297 "1: \n\t" | |
1298 "pshufw $0xFF, %%mm0, %%mm0 \n\t" | |
1299 "2: \n\t" | |
1300 "psubw %%mm1, %%mm0 \n\t" | |
1301 "movl 8(%%"REG_b", %%"REG_a"), %%esi\n\t" | |
1302 "pmullw %%mm3, %%mm0 \n\t" | |
1303 "psllw $7, %%mm1 \n\t" | |
1304 "paddw %%mm1, %%mm0 \n\t" | |
1305 | |
1306 "movq %%mm0, (%%"REG_D", %%"REG_a")\n\t" | |
1307 | |
1308 "add $8, %%"REG_a" \n\t" | |
1309 // End | |
1310 "9: \n\t" | |
1311 // "int $3\n\t" | |
1312 "lea 0b, %0 \n\t" | |
1313 "lea 1b, %1 \n\t" | |
1314 "lea 2b, %2 \n\t" | |
1315 "dec %1 \n\t" | |
1316 "dec %2 \n\t" | |
1317 "sub %0, %1 \n\t" | |
1318 "sub %0, %2 \n\t" | |
1319 "lea 9b, %3 \n\t" | |
1320 "sub %0, %3 \n\t" | |
1321 | |
1322 | |
1323 :"=r" (fragmentB), "=r" (imm8OfPShufW1B), "=r" (imm8OfPShufW2B), | |
1324 "=r" (fragmentLengthB) | |
1325 ); | |
1326 | |
1327 xpos= 0; //lumXInc/2 - 0x8000; // difference between pixel centers | |
1328 fragmentPos=0; | |
1329 | |
1330 for(i=0; i<dstW/numSplits; i++) | |
1331 { | |
1332 int xx=xpos>>16; | |
1333 | |
1334 if((i&3) == 0) | |
1335 { | |
1336 int a=0; | |
1337 int b=((xpos+xInc)>>16) - xx; | |
1338 int c=((xpos+xInc*2)>>16) - xx; | |
1339 int d=((xpos+xInc*3)>>16) - xx; | |
1340 | |
1341 filter[i ] = (( xpos & 0xFFFF) ^ 0xFFFF)>>9; | |
1342 filter[i+1] = (((xpos+xInc ) & 0xFFFF) ^ 0xFFFF)>>9; | |
1343 filter[i+2] = (((xpos+xInc*2) & 0xFFFF) ^ 0xFFFF)>>9; | |
1344 filter[i+3] = (((xpos+xInc*3) & 0xFFFF) ^ 0xFFFF)>>9; | |
1345 filterPos[i/2]= xx; | |
1346 | |
1347 if(d+1<4) | |
1348 { | |
1349 int maxShift= 3-(d+1); | |
1350 int shift=0; | |
1351 | |
1352 memcpy(funnyCode + fragmentPos, fragmentB, fragmentLengthB); | |
1353 | |
1354 funnyCode[fragmentPos + imm8OfPShufW1B]= | |
1355 (a+1) | ((b+1)<<2) | ((c+1)<<4) | ((d+1)<<6); | |
1356 funnyCode[fragmentPos + imm8OfPShufW2B]= | |
1357 a | (b<<2) | (c<<4) | (d<<6); | |
1358 | |
1359 if(i+3>=dstW) shift=maxShift; //avoid overread | |
1360 else if((filterPos[i/2]&3) <= maxShift) shift=filterPos[i/2]&3; //Align | |
1361 | |
1362 if(shift && i>=shift) | |
1363 { | |
1364 funnyCode[fragmentPos + imm8OfPShufW1B]+= 0x55*shift; | |
1365 funnyCode[fragmentPos + imm8OfPShufW2B]+= 0x55*shift; | |
1366 filterPos[i/2]-=shift; | |
1367 } | |
1368 | |
1369 fragmentPos+= fragmentLengthB; | |
1370 } | |
1371 else | |
1372 { | |
1373 int maxShift= 3-d; | |
1374 int shift=0; | |
1375 | |
1376 memcpy(funnyCode + fragmentPos, fragmentA, fragmentLengthA); | |
1377 | |
1378 funnyCode[fragmentPos + imm8OfPShufW1A]= | |
1379 funnyCode[fragmentPos + imm8OfPShufW2A]= | |
1380 a | (b<<2) | (c<<4) | (d<<6); | |
1381 | |
1382 if(i+4>=dstW) shift=maxShift; //avoid overread | |
1383 else if((filterPos[i/2]&3) <= maxShift) shift=filterPos[i/2]&3; //partial align | |
1384 | |
1385 if(shift && i>=shift) | |
1386 { | |
1387 funnyCode[fragmentPos + imm8OfPShufW1A]+= 0x55*shift; | |
1388 funnyCode[fragmentPos + imm8OfPShufW2A]+= 0x55*shift; | |
1389 filterPos[i/2]-=shift; | |
1390 } | |
1391 | |
1392 fragmentPos+= fragmentLengthA; | |
1393 } | |
1394 | |
1395 funnyCode[fragmentPos]= RET; | |
1396 } | |
1397 xpos+=xInc; | |
1398 } | |
1399 filterPos[i/2]= xpos>>16; // needed to jump to the next part | |
1400 } | |
1401 #endif // ARCH_X86 || ARCH_X86_64 | |
1402 | |
1403 static void globalInit(void){ | |
1404 // generating tables: | |
1405 int i; | |
1406 for(i=0; i<768; i++){ | |
1407 int c= MIN(MAX(i-256, 0), 255); | |
1408 clip_table[i]=c; | |
1409 } | |
1410 } | |
1411 | |
1412 static SwsFunc getSwsFunc(int flags){ | |
1413 | |
1414 #ifdef RUNTIME_CPUDETECT | |
1415 #if defined(ARCH_X86) || defined(ARCH_X86_64) | |
1416 // ordered per speed fasterst first | |
1417 if(flags & SWS_CPU_CAPS_MMX2) | |
1418 return swScale_MMX2; | |
1419 else if(flags & SWS_CPU_CAPS_3DNOW) | |
1420 return swScale_3DNow; | |
1421 else if(flags & SWS_CPU_CAPS_MMX) | |
1422 return swScale_MMX; | |
1423 else | |
1424 return swScale_C; | |
1425 | |
1426 #else | |
1427 #ifdef ARCH_POWERPC | |
1428 if(flags & SWS_CPU_CAPS_ALTIVEC) | |
1429 return swScale_altivec; | |
1430 else | |
1431 return swScale_C; | |
1432 #endif | |
1433 return swScale_C; | |
1434 #endif | |
1435 #else //RUNTIME_CPUDETECT | |
1436 #ifdef HAVE_MMX2 | |
1437 return swScale_MMX2; | |
1438 #elif defined (HAVE_3DNOW) | |
1439 return swScale_3DNow; | |
1440 #elif defined (HAVE_MMX) | |
1441 return swScale_MMX; | |
1442 #elif defined (HAVE_ALTIVEC) | |
1443 return swScale_altivec; | |
1444 #else | |
1445 return swScale_C; | |
1446 #endif | |
1447 #endif //!RUNTIME_CPUDETECT | |
1448 } | |
1449 | |
1450 static int PlanarToNV12Wrapper(SwsContext *c, uint8_t* src[], int srcStride[], int srcSliceY, | |
1451 int srcSliceH, uint8_t* dstParam[], int dstStride[]){ | |
1452 uint8_t *dst=dstParam[0] + dstStride[0]*srcSliceY; | |
1453 /* Copy Y plane */ | |
1454 if(dstStride[0]==srcStride[0] && srcStride[0] > 0) | |
1455 memcpy(dst, src[0], srcSliceH*dstStride[0]); | |
1456 else | |
1457 { | |
1458 int i; | |
1459 uint8_t *srcPtr= src[0]; | |
1460 uint8_t *dstPtr= dst; | |
1461 for(i=0; i<srcSliceH; i++) | |
1462 { | |
1463 memcpy(dstPtr, srcPtr, c->srcW); | |
1464 srcPtr+= srcStride[0]; | |
1465 dstPtr+= dstStride[0]; | |
1466 } | |
1467 } | |
1468 dst = dstParam[1] + dstStride[1]*srcSliceY/2; | |
1469 if (c->dstFormat == IMGFMT_NV12) | |
1470 interleaveBytes( src[1],src[2],dst,c->srcW/2,srcSliceH/2,srcStride[1],srcStride[2],dstStride[0] ); | |
1471 else | |
1472 interleaveBytes( src[2],src[1],dst,c->srcW/2,srcSliceH/2,srcStride[2],srcStride[1],dstStride[0] ); | |
1473 | |
1474 return srcSliceH; | |
1475 } | |
1476 | |
1477 static int PlanarToYuy2Wrapper(SwsContext *c, uint8_t* src[], int srcStride[], int srcSliceY, | |
1478 int srcSliceH, uint8_t* dstParam[], int dstStride[]){ | |
1479 uint8_t *dst=dstParam[0] + dstStride[0]*srcSliceY; | |
1480 | |
1481 yv12toyuy2( src[0],src[1],src[2],dst,c->srcW,srcSliceH,srcStride[0],srcStride[1],dstStride[0] ); | |
1482 | |
1483 return srcSliceH; | |
1484 } | |
1485 | |
1486 static int PlanarToUyvyWrapper(SwsContext *c, uint8_t* src[], int srcStride[], int srcSliceY, | |
1487 int srcSliceH, uint8_t* dstParam[], int dstStride[]){ | |
1488 uint8_t *dst=dstParam[0] + dstStride[0]*srcSliceY; | |
1489 | |
1490 yv12touyvy( src[0],src[1],src[2],dst,c->srcW,srcSliceH,srcStride[0],srcStride[1],dstStride[0] ); | |
1491 | |
1492 return srcSliceH; | |
1493 } | |
1494 | |
1495 /* {RGB,BGR}{15,16,24,32} -> {RGB,BGR}{15,16,24,32} */ | |
1496 static int rgb2rgbWrapper(SwsContext *c, uint8_t* src[], int srcStride[], int srcSliceY, | |
1497 int srcSliceH, uint8_t* dst[], int dstStride[]){ | |
1498 const int srcFormat= c->srcFormat; | |
1499 const int dstFormat= c->dstFormat; | |
1500 const int srcBpp= ((srcFormat&0xFF) + 7)>>3; | |
1501 const int dstBpp= ((dstFormat&0xFF) + 7)>>3; | |
1502 const int srcId= (srcFormat&0xFF)>>2; // 1:0, 4:1, 8:2, 15:3, 16:4, 24:6, 32:8 | |
1503 const int dstId= (dstFormat&0xFF)>>2; | |
1504 void (*conv)(const uint8_t *src, uint8_t *dst, long src_size)=NULL; | |
1505 | |
1506 /* BGR -> BGR */ | |
1507 if( (isBGR(srcFormat) && isBGR(dstFormat)) | |
1508 || (isRGB(srcFormat) && isRGB(dstFormat))){ | |
1509 switch(srcId | (dstId<<4)){ | |
1510 case 0x34: conv= rgb16to15; break; | |
1511 case 0x36: conv= rgb24to15; break; | |
1512 case 0x38: conv= rgb32to15; break; | |
1513 case 0x43: conv= rgb15to16; break; | |
1514 case 0x46: conv= rgb24to16; break; | |
1515 case 0x48: conv= rgb32to16; break; | |
1516 case 0x63: conv= rgb15to24; break; | |
1517 case 0x64: conv= rgb16to24; break; | |
1518 case 0x68: conv= rgb32to24; break; | |
1519 case 0x83: conv= rgb15to32; break; | |
1520 case 0x84: conv= rgb16to32; break; | |
1521 case 0x86: conv= rgb24to32; break; | |
1522 default: MSG_ERR("swScaler: internal error %s -> %s converter\n", | |
19143
c4dac777b44c
Use libavutil in libswscale, and allow it to be built out of the mplayer tree
lucabe
parents:
18861
diff
changeset
|
1523 sws_format_name(srcFormat), sws_format_name(dstFormat)); break; |
18861 | 1524 } |
1525 }else if( (isBGR(srcFormat) && isRGB(dstFormat)) | |
1526 || (isRGB(srcFormat) && isBGR(dstFormat))){ | |
1527 switch(srcId | (dstId<<4)){ | |
1528 case 0x33: conv= rgb15tobgr15; break; | |
1529 case 0x34: conv= rgb16tobgr15; break; | |
1530 case 0x36: conv= rgb24tobgr15; break; | |
1531 case 0x38: conv= rgb32tobgr15; break; | |
1532 case 0x43: conv= rgb15tobgr16; break; | |
1533 case 0x44: conv= rgb16tobgr16; break; | |
1534 case 0x46: conv= rgb24tobgr16; break; | |
1535 case 0x48: conv= rgb32tobgr16; break; | |
1536 case 0x63: conv= rgb15tobgr24; break; | |
1537 case 0x64: conv= rgb16tobgr24; break; | |
1538 case 0x66: conv= rgb24tobgr24; break; | |
1539 case 0x68: conv= rgb32tobgr24; break; | |
1540 case 0x83: conv= rgb15tobgr32; break; | |
1541 case 0x84: conv= rgb16tobgr32; break; | |
1542 case 0x86: conv= rgb24tobgr32; break; | |
1543 case 0x88: conv= rgb32tobgr32; break; | |
1544 default: MSG_ERR("swScaler: internal error %s -> %s converter\n", | |
19143
c4dac777b44c
Use libavutil in libswscale, and allow it to be built out of the mplayer tree
lucabe
parents:
18861
diff
changeset
|
1545 sws_format_name(srcFormat), sws_format_name(dstFormat)); break; |
18861 | 1546 } |
1547 }else{ | |
1548 MSG_ERR("swScaler: internal error %s -> %s converter\n", | |
19143
c4dac777b44c
Use libavutil in libswscale, and allow it to be built out of the mplayer tree
lucabe
parents:
18861
diff
changeset
|
1549 sws_format_name(srcFormat), sws_format_name(dstFormat)); |
18861 | 1550 } |
1551 | |
1552 if(dstStride[0]*srcBpp == srcStride[0]*dstBpp) | |
1553 conv(src[0], dst[0] + dstStride[0]*srcSliceY, srcSliceH*srcStride[0]); | |
1554 else | |
1555 { | |
1556 int i; | |
1557 uint8_t *srcPtr= src[0]; | |
1558 uint8_t *dstPtr= dst[0] + dstStride[0]*srcSliceY; | |
1559 | |
1560 for(i=0; i<srcSliceH; i++) | |
1561 { | |
1562 conv(srcPtr, dstPtr, c->srcW*srcBpp); | |
1563 srcPtr+= srcStride[0]; | |
1564 dstPtr+= dstStride[0]; | |
1565 } | |
1566 } | |
1567 return srcSliceH; | |
1568 } | |
1569 | |
1570 static int bgr24toyv12Wrapper(SwsContext *c, uint8_t* src[], int srcStride[], int srcSliceY, | |
1571 int srcSliceH, uint8_t* dst[], int dstStride[]){ | |
1572 | |
1573 rgb24toyv12( | |
1574 src[0], | |
1575 dst[0]+ srcSliceY *dstStride[0], | |
1576 dst[1]+(srcSliceY>>1)*dstStride[1], | |
1577 dst[2]+(srcSliceY>>1)*dstStride[2], | |
1578 c->srcW, srcSliceH, | |
1579 dstStride[0], dstStride[1], srcStride[0]); | |
1580 return srcSliceH; | |
1581 } | |
1582 | |
1583 static int yvu9toyv12Wrapper(SwsContext *c, uint8_t* src[], int srcStride[], int srcSliceY, | |
1584 int srcSliceH, uint8_t* dst[], int dstStride[]){ | |
1585 int i; | |
1586 | |
1587 /* copy Y */ | |
1588 if(srcStride[0]==dstStride[0] && srcStride[0] > 0) | |
1589 memcpy(dst[0]+ srcSliceY*dstStride[0], src[0], srcStride[0]*srcSliceH); | |
1590 else{ | |
1591 uint8_t *srcPtr= src[0]; | |
1592 uint8_t *dstPtr= dst[0] + dstStride[0]*srcSliceY; | |
1593 | |
1594 for(i=0; i<srcSliceH; i++) | |
1595 { | |
1596 memcpy(dstPtr, srcPtr, c->srcW); | |
1597 srcPtr+= srcStride[0]; | |
1598 dstPtr+= dstStride[0]; | |
1599 } | |
1600 } | |
1601 | |
1602 if(c->dstFormat==IMGFMT_YV12){ | |
1603 planar2x(src[1], dst[1], c->chrSrcW, c->chrSrcH, srcStride[1], dstStride[1]); | |
1604 planar2x(src[2], dst[2], c->chrSrcW, c->chrSrcH, srcStride[2], dstStride[2]); | |
1605 }else{ | |
1606 planar2x(src[1], dst[2], c->chrSrcW, c->chrSrcH, srcStride[1], dstStride[2]); | |
1607 planar2x(src[2], dst[1], c->chrSrcW, c->chrSrcH, srcStride[2], dstStride[1]); | |
1608 } | |
1609 return srcSliceH; | |
1610 } | |
1611 | |
1612 /** | |
1613 * bring pointers in YUV order instead of YVU | |
1614 */ | |
1615 static inline void sws_orderYUV(int format, uint8_t * sortedP[], int sortedStride[], uint8_t * p[], int stride[]){ | |
1616 if(format == IMGFMT_YV12 || format == IMGFMT_YVU9 | |
1617 || format == IMGFMT_444P || format == IMGFMT_422P || format == IMGFMT_411P){ | |
1618 sortedP[0]= p[0]; | |
1619 sortedP[1]= p[2]; | |
1620 sortedP[2]= p[1]; | |
1621 sortedStride[0]= stride[0]; | |
1622 sortedStride[1]= stride[2]; | |
1623 sortedStride[2]= stride[1]; | |
1624 } | |
1625 else if(isPacked(format) || isGray(format) || format == IMGFMT_Y8) | |
1626 { | |
1627 sortedP[0]= p[0]; | |
1628 sortedP[1]= | |
1629 sortedP[2]= NULL; | |
1630 sortedStride[0]= stride[0]; | |
1631 sortedStride[1]= | |
1632 sortedStride[2]= 0; | |
1633 } | |
1634 else if(format == IMGFMT_I420 || format == IMGFMT_IYUV) | |
1635 { | |
1636 sortedP[0]= p[0]; | |
1637 sortedP[1]= p[1]; | |
1638 sortedP[2]= p[2]; | |
1639 sortedStride[0]= stride[0]; | |
1640 sortedStride[1]= stride[1]; | |
1641 sortedStride[2]= stride[2]; | |
1642 } | |
1643 else if(format == IMGFMT_NV12 || format == IMGFMT_NV21) | |
1644 { | |
1645 sortedP[0]= p[0]; | |
1646 sortedP[1]= p[1]; | |
1647 sortedP[2]= NULL; | |
1648 sortedStride[0]= stride[0]; | |
1649 sortedStride[1]= stride[1]; | |
1650 sortedStride[2]= 0; | |
1651 }else{ | |
1652 MSG_ERR("internal error in orderYUV\n"); | |
1653 } | |
1654 } | |
1655 | |
1656 /* unscaled copy like stuff (assumes nearly identical formats) */ | |
1657 static int simpleCopy(SwsContext *c, uint8_t* src[], int srcStride[], int srcSliceY, | |
1658 int srcSliceH, uint8_t* dst[], int dstStride[]){ | |
1659 | |
1660 if(isPacked(c->srcFormat)) | |
1661 { | |
1662 if(dstStride[0]==srcStride[0] && srcStride[0] > 0) | |
1663 memcpy(dst[0] + dstStride[0]*srcSliceY, src[0], srcSliceH*dstStride[0]); | |
1664 else | |
1665 { | |
1666 int i; | |
1667 uint8_t *srcPtr= src[0]; | |
1668 uint8_t *dstPtr= dst[0] + dstStride[0]*srcSliceY; | |
1669 int length=0; | |
1670 | |
1671 /* universal length finder */ | |
1672 while(length+c->srcW <= ABS(dstStride[0]) | |
1673 && length+c->srcW <= ABS(srcStride[0])) length+= c->srcW; | |
1674 ASSERT(length!=0); | |
1675 | |
1676 for(i=0; i<srcSliceH; i++) | |
1677 { | |
1678 memcpy(dstPtr, srcPtr, length); | |
1679 srcPtr+= srcStride[0]; | |
1680 dstPtr+= dstStride[0]; | |
1681 } | |
1682 } | |
1683 } | |
1684 else | |
1685 { /* Planar YUV or gray */ | |
1686 int plane; | |
1687 for(plane=0; plane<3; plane++) | |
1688 { | |
1689 int length= plane==0 ? c->srcW : -((-c->srcW )>>c->chrDstHSubSample); | |
1690 int y= plane==0 ? srcSliceY: -((-srcSliceY)>>c->chrDstVSubSample); | |
1691 int height= plane==0 ? srcSliceH: -((-srcSliceH)>>c->chrDstVSubSample); | |
1692 | |
1693 if((isGray(c->srcFormat) || isGray(c->dstFormat)) && plane>0) | |
1694 { | |
1695 if(!isGray(c->dstFormat)) | |
1696 memset(dst[plane], 128, dstStride[plane]*height); | |
1697 } | |
1698 else | |
1699 { | |
1700 if(dstStride[plane]==srcStride[plane] && srcStride[plane] > 0) | |
1701 memcpy(dst[plane] + dstStride[plane]*y, src[plane], height*dstStride[plane]); | |
1702 else | |
1703 { | |
1704 int i; | |
1705 uint8_t *srcPtr= src[plane]; | |
1706 uint8_t *dstPtr= dst[plane] + dstStride[plane]*y; | |
1707 for(i=0; i<height; i++) | |
1708 { | |
1709 memcpy(dstPtr, srcPtr, length); | |
1710 srcPtr+= srcStride[plane]; | |
1711 dstPtr+= dstStride[plane]; | |
1712 } | |
1713 } | |
1714 } | |
1715 } | |
1716 } | |
1717 return srcSliceH; | |
1718 } | |
1719 | |
1720 static int remove_dup_fourcc(int fourcc) | |
1721 { | |
1722 switch(fourcc) | |
1723 { | |
1724 case IMGFMT_I420: | |
1725 case IMGFMT_IYUV: return IMGFMT_YV12; | |
1726 case IMGFMT_Y8 : return IMGFMT_Y800; | |
1727 case IMGFMT_IF09: return IMGFMT_YVU9; | |
1728 default: return fourcc; | |
1729 } | |
1730 } | |
1731 | |
1732 static void getSubSampleFactors(int *h, int *v, int format){ | |
1733 switch(format){ | |
1734 case IMGFMT_UYVY: | |
1735 case IMGFMT_YUY2: | |
1736 *h=1; | |
1737 *v=0; | |
1738 break; | |
1739 case IMGFMT_YV12: | |
1740 case IMGFMT_Y800: //FIXME remove after different subsamplings are fully implemented | |
1741 case IMGFMT_NV12: | |
1742 case IMGFMT_NV21: | |
1743 *h=1; | |
1744 *v=1; | |
1745 break; | |
1746 case IMGFMT_YVU9: | |
1747 *h=2; | |
1748 *v=2; | |
1749 break; | |
1750 case IMGFMT_444P: | |
1751 *h=0; | |
1752 *v=0; | |
1753 break; | |
1754 case IMGFMT_422P: | |
1755 *h=1; | |
1756 *v=0; | |
1757 break; | |
1758 case IMGFMT_411P: | |
1759 *h=2; | |
1760 *v=0; | |
1761 break; | |
1762 default: | |
1763 *h=0; | |
1764 *v=0; | |
1765 break; | |
1766 } | |
1767 } | |
1768 | |
1769 static uint16_t roundToInt16(int64_t f){ | |
1770 int r= (f + (1<<15))>>16; | |
1771 if(r<-0x7FFF) return 0x8000; | |
1772 else if(r> 0x7FFF) return 0x7FFF; | |
1773 else return r; | |
1774 } | |
1775 | |
1776 /** | |
1777 * @param inv_table the yuv2rgb coeffs, normally Inverse_Table_6_9[x] | |
1778 * @param fullRange if 1 then the luma range is 0..255 if 0 its 16..235 | |
1779 * @return -1 if not supported | |
1780 */ | |
1781 int sws_setColorspaceDetails(SwsContext *c, const int inv_table[4], int srcRange, const int table[4], int dstRange, int brightness, int contrast, int saturation){ | |
1782 int64_t crv = inv_table[0]; | |
1783 int64_t cbu = inv_table[1]; | |
1784 int64_t cgu = -inv_table[2]; | |
1785 int64_t cgv = -inv_table[3]; | |
1786 int64_t cy = 1<<16; | |
1787 int64_t oy = 0; | |
1788 | |
1789 if(isYUV(c->dstFormat) || isGray(c->dstFormat)) return -1; | |
1790 memcpy(c->srcColorspaceTable, inv_table, sizeof(int)*4); | |
1791 memcpy(c->dstColorspaceTable, table, sizeof(int)*4); | |
1792 | |
1793 c->brightness= brightness; | |
1794 c->contrast = contrast; | |
1795 c->saturation= saturation; | |
1796 c->srcRange = srcRange; | |
1797 c->dstRange = dstRange; | |
1798 | |
1799 c->uOffset= 0x0400040004000400LL; | |
1800 c->vOffset= 0x0400040004000400LL; | |
1801 | |
1802 if(!srcRange){ | |
1803 cy= (cy*255) / 219; | |
1804 oy= 16<<16; | |
1805 } | |
1806 | |
1807 cy = (cy *contrast )>>16; | |
1808 crv= (crv*contrast * saturation)>>32; | |
1809 cbu= (cbu*contrast * saturation)>>32; | |
1810 cgu= (cgu*contrast * saturation)>>32; | |
1811 cgv= (cgv*contrast * saturation)>>32; | |
1812 | |
1813 oy -= 256*brightness; | |
1814 | |
1815 c->yCoeff= roundToInt16(cy *8192) * 0x0001000100010001ULL; | |
1816 c->vrCoeff= roundToInt16(crv*8192) * 0x0001000100010001ULL; | |
1817 c->ubCoeff= roundToInt16(cbu*8192) * 0x0001000100010001ULL; | |
1818 c->vgCoeff= roundToInt16(cgv*8192) * 0x0001000100010001ULL; | |
1819 c->ugCoeff= roundToInt16(cgu*8192) * 0x0001000100010001ULL; | |
1820 c->yOffset= roundToInt16(oy * 8) * 0x0001000100010001ULL; | |
1821 | |
1822 yuv2rgb_c_init_tables(c, inv_table, srcRange, brightness, contrast, saturation); | |
1823 //FIXME factorize | |
1824 | |
1825 #ifdef COMPILE_ALTIVEC | |
1826 if (c->flags & SWS_CPU_CAPS_ALTIVEC) | |
1827 yuv2rgb_altivec_init_tables (c, inv_table, brightness, contrast, saturation); | |
1828 #endif | |
1829 return 0; | |
1830 } | |
1831 | |
1832 /** | |
1833 * @return -1 if not supported | |
1834 */ | |
1835 int sws_getColorspaceDetails(SwsContext *c, int **inv_table, int *srcRange, int **table, int *dstRange, int *brightness, int *contrast, int *saturation){ | |
1836 if(isYUV(c->dstFormat) || isGray(c->dstFormat)) return -1; | |
1837 | |
1838 *inv_table = c->srcColorspaceTable; | |
1839 *table = c->dstColorspaceTable; | |
1840 *srcRange = c->srcRange; | |
1841 *dstRange = c->dstRange; | |
1842 *brightness= c->brightness; | |
1843 *contrast = c->contrast; | |
1844 *saturation= c->saturation; | |
1845 | |
1846 return 0; | |
1847 } | |
1848 | |
1849 SwsContext *sws_getContext(int srcW, int srcH, int origSrcFormat, int dstW, int dstH, int origDstFormat, int flags, | |
1850 SwsFilter *srcFilter, SwsFilter *dstFilter, double *param){ | |
1851 | |
1852 SwsContext *c; | |
1853 int i; | |
1854 int usesVFilter, usesHFilter; | |
1855 int unscaled, needsDither; | |
1856 int srcFormat, dstFormat; | |
1857 SwsFilter dummyFilter= {NULL, NULL, NULL, NULL}; | |
1858 #if defined(ARCH_X86) || defined(ARCH_X86_64) | |
1859 if(flags & SWS_CPU_CAPS_MMX) | |
1860 asm volatile("emms\n\t"::: "memory"); | |
1861 #endif | |
1862 | |
1863 #ifndef RUNTIME_CPUDETECT //ensure that the flags match the compiled variant if cpudetect is off | |
1864 flags &= ~(SWS_CPU_CAPS_MMX|SWS_CPU_CAPS_MMX2|SWS_CPU_CAPS_3DNOW|SWS_CPU_CAPS_ALTIVEC); | |
1865 #ifdef HAVE_MMX2 | |
1866 flags |= SWS_CPU_CAPS_MMX|SWS_CPU_CAPS_MMX2; | |
1867 #elif defined (HAVE_3DNOW) | |
1868 flags |= SWS_CPU_CAPS_MMX|SWS_CPU_CAPS_3DNOW; | |
1869 #elif defined (HAVE_MMX) | |
1870 flags |= SWS_CPU_CAPS_MMX; | |
1871 #elif defined (HAVE_ALTIVEC) | |
1872 flags |= SWS_CPU_CAPS_ALTIVEC; | |
1873 #endif | |
1874 #endif | |
1875 if(clip_table[512] != 255) globalInit(); | |
1876 if(rgb15to16 == NULL) sws_rgb2rgb_init(flags); | |
1877 | |
1878 /* avoid duplicate Formats, so we don't need to check to much */ | |
1879 srcFormat = remove_dup_fourcc(origSrcFormat); | |
1880 dstFormat = remove_dup_fourcc(origDstFormat); | |
1881 | |
1882 unscaled = (srcW == dstW && srcH == dstH); | |
1883 needsDither= (isBGR(dstFormat) || isRGB(dstFormat)) | |
1884 && (dstFormat&0xFF)<24 | |
1885 && ((dstFormat&0xFF)<(srcFormat&0xFF) || (!(isRGB(srcFormat) || isBGR(srcFormat)))); | |
1886 | |
1887 if(!isSupportedIn(srcFormat)) | |
1888 { | |
19143
c4dac777b44c
Use libavutil in libswscale, and allow it to be built out of the mplayer tree
lucabe
parents:
18861
diff
changeset
|
1889 MSG_ERR("swScaler: %s is not supported as input format\n", sws_format_name(srcFormat)); |
18861 | 1890 return NULL; |
1891 } | |
1892 if(!isSupportedOut(dstFormat)) | |
1893 { | |
19143
c4dac777b44c
Use libavutil in libswscale, and allow it to be built out of the mplayer tree
lucabe
parents:
18861
diff
changeset
|
1894 MSG_ERR("swScaler: %s is not supported as output format\n", sws_format_name(dstFormat)); |
18861 | 1895 return NULL; |
1896 } | |
1897 | |
1898 /* sanity check */ | |
1899 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 | |
1900 { | |
1901 MSG_ERR("swScaler: %dx%d -> %dx%d is invalid scaling dimension\n", | |
1902 srcW, srcH, dstW, dstH); | |
1903 return NULL; | |
1904 } | |
1905 | |
1906 if(!dstFilter) dstFilter= &dummyFilter; | |
1907 if(!srcFilter) srcFilter= &dummyFilter; | |
1908 | |
19168
624173ae90b9
Missing part of the malloc -> av_malloc patch: memalign must be replaced as well!
reimar
parents:
19143
diff
changeset
|
1909 c= av_malloc(sizeof(SwsContext)); |
18861 | 1910 memset(c, 0, sizeof(SwsContext)); |
1911 | |
1912 c->srcW= srcW; | |
1913 c->srcH= srcH; | |
1914 c->dstW= dstW; | |
1915 c->dstH= dstH; | |
1916 c->lumXInc= ((srcW<<16) + (dstW>>1))/dstW; | |
1917 c->lumYInc= ((srcH<<16) + (dstH>>1))/dstH; | |
1918 c->flags= flags; | |
1919 c->dstFormat= dstFormat; | |
1920 c->srcFormat= srcFormat; | |
1921 c->origDstFormat= origDstFormat; | |
1922 c->origSrcFormat= origSrcFormat; | |
1923 c->vRounder= 4* 0x0001000100010001ULL; | |
1924 | |
1925 usesHFilter= usesVFilter= 0; | |
1926 if(dstFilter->lumV!=NULL && dstFilter->lumV->length>1) usesVFilter=1; | |
1927 if(dstFilter->lumH!=NULL && dstFilter->lumH->length>1) usesHFilter=1; | |
1928 if(dstFilter->chrV!=NULL && dstFilter->chrV->length>1) usesVFilter=1; | |
1929 if(dstFilter->chrH!=NULL && dstFilter->chrH->length>1) usesHFilter=1; | |
1930 if(srcFilter->lumV!=NULL && srcFilter->lumV->length>1) usesVFilter=1; | |
1931 if(srcFilter->lumH!=NULL && srcFilter->lumH->length>1) usesHFilter=1; | |
1932 if(srcFilter->chrV!=NULL && srcFilter->chrV->length>1) usesVFilter=1; | |
1933 if(srcFilter->chrH!=NULL && srcFilter->chrH->length>1) usesHFilter=1; | |
1934 | |
1935 getSubSampleFactors(&c->chrSrcHSubSample, &c->chrSrcVSubSample, srcFormat); | |
1936 getSubSampleFactors(&c->chrDstHSubSample, &c->chrDstVSubSample, dstFormat); | |
1937 | |
1938 // reuse chroma for 2 pixles rgb/bgr unless user wants full chroma interpolation | |
1939 if((isBGR(dstFormat) || isRGB(dstFormat)) && !(flags&SWS_FULL_CHR_H_INT)) c->chrDstHSubSample=1; | |
1940 | |
1941 // drop some chroma lines if the user wants it | |
1942 c->vChrDrop= (flags&SWS_SRC_V_CHR_DROP_MASK)>>SWS_SRC_V_CHR_DROP_SHIFT; | |
1943 c->chrSrcVSubSample+= c->vChrDrop; | |
1944 | |
1945 // drop every 2. pixel for chroma calculation unless user wants full chroma | |
1946 if((isBGR(srcFormat) || isRGB(srcFormat)) && !(flags&SWS_FULL_CHR_H_INP)) | |
1947 c->chrSrcHSubSample=1; | |
1948 | |
1949 if(param){ | |
1950 c->param[0] = param[0]; | |
1951 c->param[1] = param[1]; | |
1952 }else{ | |
1953 c->param[0] = | |
1954 c->param[1] = SWS_PARAM_DEFAULT; | |
1955 } | |
1956 | |
1957 c->chrIntHSubSample= c->chrDstHSubSample; | |
1958 c->chrIntVSubSample= c->chrSrcVSubSample; | |
1959 | |
1960 // note the -((-x)>>y) is so that we allways round toward +inf | |
1961 c->chrSrcW= -((-srcW) >> c->chrSrcHSubSample); | |
1962 c->chrSrcH= -((-srcH) >> c->chrSrcVSubSample); | |
1963 c->chrDstW= -((-dstW) >> c->chrDstHSubSample); | |
1964 c->chrDstH= -((-dstH) >> c->chrDstVSubSample); | |
1965 | |
1966 sws_setColorspaceDetails(c, Inverse_Table_6_9[SWS_CS_DEFAULT], 0, Inverse_Table_6_9[SWS_CS_DEFAULT] /* FIXME*/, 0, 0, 1<<16, 1<<16); | |
1967 | |
1968 /* unscaled special Cases */ | |
1969 if(unscaled && !usesHFilter && !usesVFilter) | |
1970 { | |
1971 /* yv12_to_nv12 */ | |
1972 if(srcFormat == IMGFMT_YV12 && (dstFormat == IMGFMT_NV12 || dstFormat == IMGFMT_NV21)) | |
1973 { | |
1974 c->swScale= PlanarToNV12Wrapper; | |
1975 } | |
1976 /* yuv2bgr */ | |
1977 if((srcFormat==IMGFMT_YV12 || srcFormat==IMGFMT_422P) && (isBGR(dstFormat) || isRGB(dstFormat))) | |
1978 { | |
1979 c->swScale= yuv2rgb_get_func_ptr(c); | |
1980 } | |
1981 | |
1982 if( srcFormat==IMGFMT_YVU9 && dstFormat==IMGFMT_YV12 ) | |
1983 { | |
1984 c->swScale= yvu9toyv12Wrapper; | |
1985 } | |
1986 | |
1987 /* bgr24toYV12 */ | |
1988 if(srcFormat==IMGFMT_BGR24 && dstFormat==IMGFMT_YV12) | |
1989 c->swScale= bgr24toyv12Wrapper; | |
1990 | |
1991 /* rgb/bgr -> rgb/bgr (no dither needed forms) */ | |
1992 if( (isBGR(srcFormat) || isRGB(srcFormat)) | |
1993 && (isBGR(dstFormat) || isRGB(dstFormat)) | |
1994 && !needsDither) | |
1995 c->swScale= rgb2rgbWrapper; | |
1996 | |
1997 /* LQ converters if -sws 0 or -sws 4*/ | |
1998 if(c->flags&(SWS_FAST_BILINEAR|SWS_POINT)){ | |
1999 /* rgb/bgr -> rgb/bgr (dither needed forms) */ | |
2000 if( (isBGR(srcFormat) || isRGB(srcFormat)) | |
2001 && (isBGR(dstFormat) || isRGB(dstFormat)) | |
2002 && needsDither) | |
2003 c->swScale= rgb2rgbWrapper; | |
2004 | |
2005 /* yv12_to_yuy2 */ | |
2006 if(srcFormat == IMGFMT_YV12 && | |
2007 (dstFormat == IMGFMT_YUY2 || dstFormat == IMGFMT_UYVY)) | |
2008 { | |
2009 if (dstFormat == IMGFMT_YUY2) | |
2010 c->swScale= PlanarToYuy2Wrapper; | |
2011 else | |
2012 c->swScale= PlanarToUyvyWrapper; | |
2013 } | |
2014 } | |
2015 | |
2016 #ifdef COMPILE_ALTIVEC | |
2017 if ((c->flags & SWS_CPU_CAPS_ALTIVEC) && | |
2018 ((srcFormat == IMGFMT_YV12 && | |
2019 (dstFormat == IMGFMT_YUY2 || dstFormat == IMGFMT_UYVY)))) { | |
2020 // unscaled YV12 -> packed YUV, we want speed | |
2021 if (dstFormat == IMGFMT_YUY2) | |
2022 c->swScale= yv12toyuy2_unscaled_altivec; | |
2023 else | |
2024 c->swScale= yv12touyvy_unscaled_altivec; | |
2025 } | |
2026 #endif | |
2027 | |
2028 /* simple copy */ | |
2029 if( srcFormat == dstFormat | |
2030 || (isPlanarYUV(srcFormat) && isGray(dstFormat)) | |
2031 || (isPlanarYUV(dstFormat) && isGray(srcFormat)) | |
2032 ) | |
2033 { | |
2034 c->swScale= simpleCopy; | |
2035 } | |
2036 | |
2037 if(c->swScale){ | |
2038 if(flags&SWS_PRINT_INFO) | |
2039 MSG_INFO("SwScaler: using unscaled %s -> %s special converter\n", | |
19143
c4dac777b44c
Use libavutil in libswscale, and allow it to be built out of the mplayer tree
lucabe
parents:
18861
diff
changeset
|
2040 sws_format_name(srcFormat), sws_format_name(dstFormat)); |
18861 | 2041 return c; |
2042 } | |
2043 } | |
2044 | |
2045 if(flags & SWS_CPU_CAPS_MMX2) | |
2046 { | |
2047 c->canMMX2BeUsed= (dstW >=srcW && (dstW&31)==0 && (srcW&15)==0) ? 1 : 0; | |
2048 if(!c->canMMX2BeUsed && dstW >=srcW && (srcW&15)==0 && (flags&SWS_FAST_BILINEAR)) | |
2049 { | |
2050 if(flags&SWS_PRINT_INFO) | |
2051 MSG_INFO("SwScaler: output Width is not a multiple of 32 -> no MMX2 scaler\n"); | |
2052 } | |
2053 if(usesHFilter) c->canMMX2BeUsed=0; | |
2054 } | |
2055 else | |
2056 c->canMMX2BeUsed=0; | |
2057 | |
2058 c->chrXInc= ((c->chrSrcW<<16) + (c->chrDstW>>1))/c->chrDstW; | |
2059 c->chrYInc= ((c->chrSrcH<<16) + (c->chrDstH>>1))/c->chrDstH; | |
2060 | |
2061 // match pixel 0 of the src to pixel 0 of dst and match pixel n-2 of src to pixel n-2 of dst | |
2062 // but only for the FAST_BILINEAR mode otherwise do correct scaling | |
2063 // n-2 is the last chrominance sample available | |
2064 // this is not perfect, but noone shuld notice the difference, the more correct variant | |
2065 // would be like the vertical one, but that would require some special code for the | |
2066 // first and last pixel | |
2067 if(flags&SWS_FAST_BILINEAR) | |
2068 { | |
2069 if(c->canMMX2BeUsed) | |
2070 { | |
2071 c->lumXInc+= 20; | |
2072 c->chrXInc+= 20; | |
2073 } | |
2074 //we don't use the x86asm scaler if mmx is available | |
2075 else if(flags & SWS_CPU_CAPS_MMX) | |
2076 { | |
2077 c->lumXInc = ((srcW-2)<<16)/(dstW-2) - 20; | |
2078 c->chrXInc = ((c->chrSrcW-2)<<16)/(c->chrDstW-2) - 20; | |
2079 } | |
2080 } | |
2081 | |
2082 /* precalculate horizontal scaler filter coefficients */ | |
2083 { | |
2084 const int filterAlign= | |
2085 (flags & SWS_CPU_CAPS_MMX) ? 4 : | |
2086 (flags & SWS_CPU_CAPS_ALTIVEC) ? 8 : | |
2087 1; | |
2088 | |
2089 initFilter(&c->hLumFilter, &c->hLumFilterPos, &c->hLumFilterSize, c->lumXInc, | |
2090 srcW , dstW, filterAlign, 1<<14, | |
2091 (flags&SWS_BICUBLIN) ? (flags|SWS_BICUBIC) : flags, | |
2092 srcFilter->lumH, dstFilter->lumH, c->param); | |
2093 initFilter(&c->hChrFilter, &c->hChrFilterPos, &c->hChrFilterSize, c->chrXInc, | |
2094 c->chrSrcW, c->chrDstW, filterAlign, 1<<14, | |
2095 (flags&SWS_BICUBLIN) ? (flags|SWS_BILINEAR) : flags, | |
2096 srcFilter->chrH, dstFilter->chrH, c->param); | |
2097 | |
2098 #if defined(ARCH_X86) || defined(ARCH_X86_64) | |
2099 // can't downscale !!! | |
2100 if(c->canMMX2BeUsed && (flags & SWS_FAST_BILINEAR)) | |
2101 { | |
2102 #define MAX_FUNNY_CODE_SIZE 10000 | |
2103 #ifdef MAP_ANONYMOUS | |
2104 c->funnyYCode = (uint8_t*)mmap(NULL, MAX_FUNNY_CODE_SIZE, PROT_EXEC | PROT_READ | PROT_WRITE, MAP_PRIVATE | MAP_ANONYMOUS, 0, 0); | |
2105 c->funnyUVCode = (uint8_t*)mmap(NULL, MAX_FUNNY_CODE_SIZE, PROT_EXEC | PROT_READ | PROT_WRITE, MAP_PRIVATE | MAP_ANONYMOUS, 0, 0); | |
2106 #else | |
19168
624173ae90b9
Missing part of the malloc -> av_malloc patch: memalign must be replaced as well!
reimar
parents:
19143
diff
changeset
|
2107 c->funnyYCode = av_malloc(MAX_FUNNY_CODE_SIZE); |
624173ae90b9
Missing part of the malloc -> av_malloc patch: memalign must be replaced as well!
reimar
parents:
19143
diff
changeset
|
2108 c->funnyUVCode = av_malloc(MAX_FUNNY_CODE_SIZE); |
18861 | 2109 #endif |
2110 | |
19168
624173ae90b9
Missing part of the malloc -> av_malloc patch: memalign must be replaced as well!
reimar
parents:
19143
diff
changeset
|
2111 c->lumMmx2Filter = av_malloc((dstW /8+8)*sizeof(int16_t)); |
624173ae90b9
Missing part of the malloc -> av_malloc patch: memalign must be replaced as well!
reimar
parents:
19143
diff
changeset
|
2112 c->chrMmx2Filter = av_malloc((c->chrDstW /4+8)*sizeof(int16_t)); |
624173ae90b9
Missing part of the malloc -> av_malloc patch: memalign must be replaced as well!
reimar
parents:
19143
diff
changeset
|
2113 c->lumMmx2FilterPos= av_malloc((dstW /2/8+8)*sizeof(int32_t)); |
624173ae90b9
Missing part of the malloc -> av_malloc patch: memalign must be replaced as well!
reimar
parents:
19143
diff
changeset
|
2114 c->chrMmx2FilterPos= av_malloc((c->chrDstW/2/4+8)*sizeof(int32_t)); |
18861 | 2115 |
2116 initMMX2HScaler( dstW, c->lumXInc, c->funnyYCode , c->lumMmx2Filter, c->lumMmx2FilterPos, 8); | |
2117 initMMX2HScaler(c->chrDstW, c->chrXInc, c->funnyUVCode, c->chrMmx2Filter, c->chrMmx2FilterPos, 4); | |
2118 } | |
2119 #endif | |
2120 } // Init Horizontal stuff | |
2121 | |
2122 | |
2123 | |
2124 /* precalculate vertical scaler filter coefficients */ | |
2125 { | |
2126 const int filterAlign= | |
19172
bae6c99a99cc
vertical scaler with accurate rounding, some people on doom9 can see +-1 errors
michael
parents:
19169
diff
changeset
|
2127 (flags & SWS_CPU_CAPS_MMX) && (flags & SWS_ACCURATE_RND) ? 2 : |
18861 | 2128 (flags & SWS_CPU_CAPS_ALTIVEC) ? 8 : |
2129 1; | |
2130 | |
2131 initFilter(&c->vLumFilter, &c->vLumFilterPos, &c->vLumFilterSize, c->lumYInc, | |
2132 srcH , dstH, filterAlign, (1<<12)-4, | |
2133 (flags&SWS_BICUBLIN) ? (flags|SWS_BICUBIC) : flags, | |
2134 srcFilter->lumV, dstFilter->lumV, c->param); | |
2135 initFilter(&c->vChrFilter, &c->vChrFilterPos, &c->vChrFilterSize, c->chrYInc, | |
2136 c->chrSrcH, c->chrDstH, filterAlign, (1<<12)-4, | |
2137 (flags&SWS_BICUBLIN) ? (flags|SWS_BILINEAR) : flags, | |
2138 srcFilter->chrV, dstFilter->chrV, c->param); | |
2139 | |
2140 #ifdef HAVE_ALTIVEC | |
19168
624173ae90b9
Missing part of the malloc -> av_malloc patch: memalign must be replaced as well!
reimar
parents:
19143
diff
changeset
|
2141 c->vYCoeffsBank = av_malloc(sizeof (vector signed short)*c->vLumFilterSize*c->dstH); |
19169
70ea0a8d3b4a
Fix typo introduced in the memalign->av_malloc conversion (there is no
pacman
parents:
19168
diff
changeset
|
2142 c->vCCoeffsBank = av_malloc(sizeof (vector signed short)*c->vChrFilterSize*c->chrDstH); |
18861 | 2143 |
2144 for (i=0;i<c->vLumFilterSize*c->dstH;i++) { | |
2145 int j; | |
2146 short *p = (short *)&c->vYCoeffsBank[i]; | |
2147 for (j=0;j<8;j++) | |
2148 p[j] = c->vLumFilter[i]; | |
2149 } | |
2150 | |
2151 for (i=0;i<c->vChrFilterSize*c->chrDstH;i++) { | |
2152 int j; | |
2153 short *p = (short *)&c->vCCoeffsBank[i]; | |
2154 for (j=0;j<8;j++) | |
2155 p[j] = c->vChrFilter[i]; | |
2156 } | |
2157 #endif | |
2158 } | |
2159 | |
2160 // Calculate Buffer Sizes so that they won't run out while handling these damn slices | |
2161 c->vLumBufSize= c->vLumFilterSize; | |
2162 c->vChrBufSize= c->vChrFilterSize; | |
2163 for(i=0; i<dstH; i++) | |
2164 { | |
2165 int chrI= i*c->chrDstH / dstH; | |
2166 int nextSlice= MAX(c->vLumFilterPos[i ] + c->vLumFilterSize - 1, | |
2167 ((c->vChrFilterPos[chrI] + c->vChrFilterSize - 1)<<c->chrSrcVSubSample)); | |
2168 | |
2169 nextSlice>>= c->chrSrcVSubSample; | |
2170 nextSlice<<= c->chrSrcVSubSample; | |
2171 if(c->vLumFilterPos[i ] + c->vLumBufSize < nextSlice) | |
2172 c->vLumBufSize= nextSlice - c->vLumFilterPos[i ]; | |
2173 if(c->vChrFilterPos[chrI] + c->vChrBufSize < (nextSlice>>c->chrSrcVSubSample)) | |
2174 c->vChrBufSize= (nextSlice>>c->chrSrcVSubSample) - c->vChrFilterPos[chrI]; | |
2175 } | |
2176 | |
2177 // allocate pixbufs (we use dynamic allocation because otherwise we would need to | |
19168
624173ae90b9
Missing part of the malloc -> av_malloc patch: memalign must be replaced as well!
reimar
parents:
19143
diff
changeset
|
2178 c->lumPixBuf= av_malloc(c->vLumBufSize*2*sizeof(int16_t*)); |
624173ae90b9
Missing part of the malloc -> av_malloc patch: memalign must be replaced as well!
reimar
parents:
19143
diff
changeset
|
2179 c->chrPixBuf= av_malloc(c->vChrBufSize*2*sizeof(int16_t*)); |
18861 | 2180 //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) |
2181 /* align at 16 bytes for AltiVec */ | |
2182 for(i=0; i<c->vLumBufSize; i++) | |
19168
624173ae90b9
Missing part of the malloc -> av_malloc patch: memalign must be replaced as well!
reimar
parents:
19143
diff
changeset
|
2183 c->lumPixBuf[i]= c->lumPixBuf[i+c->vLumBufSize]= av_malloc(4000); |
18861 | 2184 for(i=0; i<c->vChrBufSize; i++) |
19168
624173ae90b9
Missing part of the malloc -> av_malloc patch: memalign must be replaced as well!
reimar
parents:
19143
diff
changeset
|
2185 c->chrPixBuf[i]= c->chrPixBuf[i+c->vChrBufSize]= av_malloc(8000); |
18861 | 2186 |
2187 //try to avoid drawing green stuff between the right end and the stride end | |
2188 for(i=0; i<c->vLumBufSize; i++) memset(c->lumPixBuf[i], 0, 4000); | |
2189 for(i=0; i<c->vChrBufSize; i++) memset(c->chrPixBuf[i], 64, 8000); | |
2190 | |
2191 ASSERT(c->chrDstH <= dstH) | |
2192 | |
2193 if(flags&SWS_PRINT_INFO) | |
2194 { | |
2195 #ifdef DITHER1XBPP | |
2196 char *dither= " dithered"; | |
2197 #else | |
2198 char *dither= ""; | |
2199 #endif | |
2200 if(flags&SWS_FAST_BILINEAR) | |
2201 MSG_INFO("\nSwScaler: FAST_BILINEAR scaler, "); | |
2202 else if(flags&SWS_BILINEAR) | |
2203 MSG_INFO("\nSwScaler: BILINEAR scaler, "); | |
2204 else if(flags&SWS_BICUBIC) | |
2205 MSG_INFO("\nSwScaler: BICUBIC scaler, "); | |
2206 else if(flags&SWS_X) | |
2207 MSG_INFO("\nSwScaler: Experimental scaler, "); | |
2208 else if(flags&SWS_POINT) | |
2209 MSG_INFO("\nSwScaler: Nearest Neighbor / POINT scaler, "); | |
2210 else if(flags&SWS_AREA) | |
2211 MSG_INFO("\nSwScaler: Area Averageing scaler, "); | |
2212 else if(flags&SWS_BICUBLIN) | |
2213 MSG_INFO("\nSwScaler: luma BICUBIC / chroma BILINEAR scaler, "); | |
2214 else if(flags&SWS_GAUSS) | |
2215 MSG_INFO("\nSwScaler: Gaussian scaler, "); | |
2216 else if(flags&SWS_SINC) | |
2217 MSG_INFO("\nSwScaler: Sinc scaler, "); | |
2218 else if(flags&SWS_LANCZOS) | |
2219 MSG_INFO("\nSwScaler: Lanczos scaler, "); | |
2220 else if(flags&SWS_SPLINE) | |
2221 MSG_INFO("\nSwScaler: Bicubic spline scaler, "); | |
2222 else | |
2223 MSG_INFO("\nSwScaler: ehh flags invalid?! "); | |
2224 | |
2225 if(dstFormat==IMGFMT_BGR15 || dstFormat==IMGFMT_BGR16) | |
2226 MSG_INFO("from %s to%s %s ", | |
19143
c4dac777b44c
Use libavutil in libswscale, and allow it to be built out of the mplayer tree
lucabe
parents:
18861
diff
changeset
|
2227 sws_format_name(srcFormat), dither, sws_format_name(dstFormat)); |
18861 | 2228 else |
2229 MSG_INFO("from %s to %s ", | |
19143
c4dac777b44c
Use libavutil in libswscale, and allow it to be built out of the mplayer tree
lucabe
parents:
18861
diff
changeset
|
2230 sws_format_name(srcFormat), sws_format_name(dstFormat)); |
18861 | 2231 |
2232 if(flags & SWS_CPU_CAPS_MMX2) | |
2233 MSG_INFO("using MMX2\n"); | |
2234 else if(flags & SWS_CPU_CAPS_3DNOW) | |
2235 MSG_INFO("using 3DNOW\n"); | |
2236 else if(flags & SWS_CPU_CAPS_MMX) | |
2237 MSG_INFO("using MMX\n"); | |
2238 else if(flags & SWS_CPU_CAPS_ALTIVEC) | |
2239 MSG_INFO("using AltiVec\n"); | |
2240 else | |
2241 MSG_INFO("using C\n"); | |
2242 } | |
2243 | |
2244 if(flags & SWS_PRINT_INFO) | |
2245 { | |
2246 if(flags & SWS_CPU_CAPS_MMX) | |
2247 { | |
2248 if(c->canMMX2BeUsed && (flags&SWS_FAST_BILINEAR)) | |
2249 MSG_V("SwScaler: using FAST_BILINEAR MMX2 scaler for horizontal scaling\n"); | |
2250 else | |
2251 { | |
2252 if(c->hLumFilterSize==4) | |
2253 MSG_V("SwScaler: using 4-tap MMX scaler for horizontal luminance scaling\n"); | |
2254 else if(c->hLumFilterSize==8) | |
2255 MSG_V("SwScaler: using 8-tap MMX scaler for horizontal luminance scaling\n"); | |
2256 else | |
2257 MSG_V("SwScaler: using n-tap MMX scaler for horizontal luminance scaling\n"); | |
2258 | |
2259 if(c->hChrFilterSize==4) | |
2260 MSG_V("SwScaler: using 4-tap MMX scaler for horizontal chrominance scaling\n"); | |
2261 else if(c->hChrFilterSize==8) | |
2262 MSG_V("SwScaler: using 8-tap MMX scaler for horizontal chrominance scaling\n"); | |
2263 else | |
2264 MSG_V("SwScaler: using n-tap MMX scaler for horizontal chrominance scaling\n"); | |
2265 } | |
2266 } | |
2267 else | |
2268 { | |
2269 #if defined(ARCH_X86) || defined(ARCH_X86_64) | |
2270 MSG_V("SwScaler: using X86-Asm scaler for horizontal scaling\n"); | |
2271 #else | |
2272 if(flags & SWS_FAST_BILINEAR) | |
2273 MSG_V("SwScaler: using FAST_BILINEAR C scaler for horizontal scaling\n"); | |
2274 else | |
2275 MSG_V("SwScaler: using C scaler for horizontal scaling\n"); | |
2276 #endif | |
2277 } | |
2278 if(isPlanarYUV(dstFormat)) | |
2279 { | |
2280 if(c->vLumFilterSize==1) | |
2281 MSG_V("SwScaler: using 1-tap %s \"scaler\" for vertical scaling (YV12 like)\n", (flags & SWS_CPU_CAPS_MMX) ? "MMX" : "C"); | |
2282 else | |
2283 MSG_V("SwScaler: using n-tap %s scaler for vertical scaling (YV12 like)\n", (flags & SWS_CPU_CAPS_MMX) ? "MMX" : "C"); | |
2284 } | |
2285 else | |
2286 { | |
2287 if(c->vLumFilterSize==1 && c->vChrFilterSize==2) | |
2288 MSG_V("SwScaler: using 1-tap %s \"scaler\" for vertical luminance scaling (BGR)\n" | |
2289 "SwScaler: 2-tap scaler for vertical chrominance scaling (BGR)\n",(flags & SWS_CPU_CAPS_MMX) ? "MMX" : "C"); | |
2290 else if(c->vLumFilterSize==2 && c->vChrFilterSize==2) | |
2291 MSG_V("SwScaler: using 2-tap linear %s scaler for vertical scaling (BGR)\n", (flags & SWS_CPU_CAPS_MMX) ? "MMX" : "C"); | |
2292 else | |
2293 MSG_V("SwScaler: using n-tap %s scaler for vertical scaling (BGR)\n", (flags & SWS_CPU_CAPS_MMX) ? "MMX" : "C"); | |
2294 } | |
2295 | |
2296 if(dstFormat==IMGFMT_BGR24) | |
2297 MSG_V("SwScaler: using %s YV12->BGR24 Converter\n", | |
2298 (flags & SWS_CPU_CAPS_MMX2) ? "MMX2" : ((flags & SWS_CPU_CAPS_MMX) ? "MMX" : "C")); | |
2299 else if(dstFormat==IMGFMT_BGR32) | |
2300 MSG_V("SwScaler: using %s YV12->BGR32 Converter\n", (flags & SWS_CPU_CAPS_MMX) ? "MMX" : "C"); | |
2301 else if(dstFormat==IMGFMT_BGR16) | |
2302 MSG_V("SwScaler: using %s YV12->BGR16 Converter\n", (flags & SWS_CPU_CAPS_MMX) ? "MMX" : "C"); | |
2303 else if(dstFormat==IMGFMT_BGR15) | |
2304 MSG_V("SwScaler: using %s YV12->BGR15 Converter\n", (flags & SWS_CPU_CAPS_MMX) ? "MMX" : "C"); | |
2305 | |
2306 MSG_V("SwScaler: %dx%d -> %dx%d\n", srcW, srcH, dstW, dstH); | |
2307 } | |
2308 if(flags & SWS_PRINT_INFO) | |
2309 { | |
2310 MSG_DBG2("SwScaler:Lum srcW=%d srcH=%d dstW=%d dstH=%d xInc=%d yInc=%d\n", | |
2311 c->srcW, c->srcH, c->dstW, c->dstH, c->lumXInc, c->lumYInc); | |
2312 MSG_DBG2("SwScaler:Chr srcW=%d srcH=%d dstW=%d dstH=%d xInc=%d yInc=%d\n", | |
2313 c->chrSrcW, c->chrSrcH, c->chrDstW, c->chrDstH, c->chrXInc, c->chrYInc); | |
2314 } | |
2315 | |
2316 c->swScale= getSwsFunc(flags); | |
2317 return c; | |
2318 } | |
2319 | |
2320 /** | |
2321 * swscale warper, so we don't need to export the SwsContext. | |
2322 * assumes planar YUV to be in YUV order instead of YVU | |
2323 */ | |
2324 int sws_scale_ordered(SwsContext *c, uint8_t* src[], int srcStride[], int srcSliceY, | |
2325 int srcSliceH, uint8_t* dst[], int dstStride[]){ | |
2326 if (c->sliceDir == 0 && srcSliceY != 0 && srcSliceY + srcSliceH != c->srcH) { | |
2327 MSG_ERR("swScaler: slices start in the middle!\n"); | |
2328 return 0; | |
2329 } | |
2330 if (c->sliceDir == 0) { | |
2331 if (srcSliceY == 0) c->sliceDir = 1; else c->sliceDir = -1; | |
2332 } | |
2333 | |
2334 // copy strides, so they can safely be modified | |
2335 if (c->sliceDir == 1) { | |
2336 // slices go from top to bottom | |
2337 int srcStride2[3]= {srcStride[0], srcStride[1], srcStride[2]}; | |
2338 int dstStride2[3]= {dstStride[0], dstStride[1], dstStride[2]}; | |
2339 return c->swScale(c, src, srcStride2, srcSliceY, srcSliceH, dst, dstStride2); | |
2340 } else { | |
2341 // slices go from bottom to top => we flip the image internally | |
2342 uint8_t* src2[3]= {src[0] + (srcSliceH-1)*srcStride[0], | |
2343 src[1] + ((srcSliceH>>c->chrSrcVSubSample)-1)*srcStride[1], | |
2344 src[2] + ((srcSliceH>>c->chrSrcVSubSample)-1)*srcStride[2] | |
2345 }; | |
2346 uint8_t* dst2[3]= {dst[0] + (c->dstH-1)*dstStride[0], | |
2347 dst[1] + ((c->dstH>>c->chrDstVSubSample)-1)*dstStride[1], | |
2348 dst[2] + ((c->dstH>>c->chrDstVSubSample)-1)*dstStride[2]}; | |
2349 int srcStride2[3]= {-srcStride[0], -srcStride[1], -srcStride[2]}; | |
2350 int dstStride2[3]= {-dstStride[0], -dstStride[1], -dstStride[2]}; | |
2351 | |
2352 return c->swScale(c, src2, srcStride2, c->srcH-srcSliceY-srcSliceH, srcSliceH, dst2, dstStride2); | |
2353 } | |
2354 } | |
2355 | |
2356 /** | |
2357 * swscale warper, so we don't need to export the SwsContext | |
2358 */ | |
2359 int sws_scale(SwsContext *c, uint8_t* srcParam[], int srcStrideParam[], int srcSliceY, | |
2360 int srcSliceH, uint8_t* dstParam[], int dstStrideParam[]){ | |
2361 int srcStride[3]; | |
2362 int dstStride[3]; | |
2363 uint8_t *src[3]; | |
2364 uint8_t *dst[3]; | |
2365 sws_orderYUV(c->origSrcFormat, src, srcStride, srcParam, srcStrideParam); | |
2366 sws_orderYUV(c->origDstFormat, dst, dstStride, dstParam, dstStrideParam); | |
2367 //printf("sws: slice %d %d\n", srcSliceY, srcSliceH); | |
2368 | |
2369 return c->swScale(c, src, srcStride, srcSliceY, srcSliceH, dst, dstStride); | |
2370 } | |
2371 | |
2372 SwsFilter *sws_getDefaultFilter(float lumaGBlur, float chromaGBlur, | |
2373 float lumaSharpen, float chromaSharpen, | |
2374 float chromaHShift, float chromaVShift, | |
2375 int verbose) | |
2376 { | |
19143
c4dac777b44c
Use libavutil in libswscale, and allow it to be built out of the mplayer tree
lucabe
parents:
18861
diff
changeset
|
2377 SwsFilter *filter= av_malloc(sizeof(SwsFilter)); |
18861 | 2378 |
2379 if(lumaGBlur!=0.0){ | |
2380 filter->lumH= sws_getGaussianVec(lumaGBlur, 3.0); | |
2381 filter->lumV= sws_getGaussianVec(lumaGBlur, 3.0); | |
2382 }else{ | |
2383 filter->lumH= sws_getIdentityVec(); | |
2384 filter->lumV= sws_getIdentityVec(); | |
2385 } | |
2386 | |
2387 if(chromaGBlur!=0.0){ | |
2388 filter->chrH= sws_getGaussianVec(chromaGBlur, 3.0); | |
2389 filter->chrV= sws_getGaussianVec(chromaGBlur, 3.0); | |
2390 }else{ | |
2391 filter->chrH= sws_getIdentityVec(); | |
2392 filter->chrV= sws_getIdentityVec(); | |
2393 } | |
2394 | |
2395 if(chromaSharpen!=0.0){ | |
2396 SwsVector *id= sws_getIdentityVec(); | |
2397 sws_scaleVec(filter->chrH, -chromaSharpen); | |
2398 sws_scaleVec(filter->chrV, -chromaSharpen); | |
2399 sws_addVec(filter->chrH, id); | |
2400 sws_addVec(filter->chrV, id); | |
2401 sws_freeVec(id); | |
2402 } | |
2403 | |
2404 if(lumaSharpen!=0.0){ | |
2405 SwsVector *id= sws_getIdentityVec(); | |
2406 sws_scaleVec(filter->lumH, -lumaSharpen); | |
2407 sws_scaleVec(filter->lumV, -lumaSharpen); | |
2408 sws_addVec(filter->lumH, id); | |
2409 sws_addVec(filter->lumV, id); | |
2410 sws_freeVec(id); | |
2411 } | |
2412 | |
2413 if(chromaHShift != 0.0) | |
2414 sws_shiftVec(filter->chrH, (int)(chromaHShift+0.5)); | |
2415 | |
2416 if(chromaVShift != 0.0) | |
2417 sws_shiftVec(filter->chrV, (int)(chromaVShift+0.5)); | |
2418 | |
2419 sws_normalizeVec(filter->chrH, 1.0); | |
2420 sws_normalizeVec(filter->chrV, 1.0); | |
2421 sws_normalizeVec(filter->lumH, 1.0); | |
2422 sws_normalizeVec(filter->lumV, 1.0); | |
2423 | |
2424 if(verbose) sws_printVec(filter->chrH); | |
2425 if(verbose) sws_printVec(filter->lumH); | |
2426 | |
2427 return filter; | |
2428 } | |
2429 | |
2430 /** | |
2431 * returns a normalized gaussian curve used to filter stuff | |
2432 * quality=3 is high quality, lowwer is lowwer quality | |
2433 */ | |
2434 SwsVector *sws_getGaussianVec(double variance, double quality){ | |
2435 const int length= (int)(variance*quality + 0.5) | 1; | |
2436 int i; | |
19168
624173ae90b9
Missing part of the malloc -> av_malloc patch: memalign must be replaced as well!
reimar
parents:
19143
diff
changeset
|
2437 double *coeff= av_malloc(length*sizeof(double)); |
18861 | 2438 double middle= (length-1)*0.5; |
19143
c4dac777b44c
Use libavutil in libswscale, and allow it to be built out of the mplayer tree
lucabe
parents:
18861
diff
changeset
|
2439 SwsVector *vec= av_malloc(sizeof(SwsVector)); |
18861 | 2440 |
2441 vec->coeff= coeff; | |
2442 vec->length= length; | |
2443 | |
2444 for(i=0; i<length; i++) | |
2445 { | |
2446 double dist= i-middle; | |
2447 coeff[i]= exp( -dist*dist/(2*variance*variance) ) / sqrt(2*variance*PI); | |
2448 } | |
2449 | |
2450 sws_normalizeVec(vec, 1.0); | |
2451 | |
2452 return vec; | |
2453 } | |
2454 | |
2455 SwsVector *sws_getConstVec(double c, int length){ | |
2456 int i; | |
19168
624173ae90b9
Missing part of the malloc -> av_malloc patch: memalign must be replaced as well!
reimar
parents:
19143
diff
changeset
|
2457 double *coeff= av_malloc(length*sizeof(double)); |
19143
c4dac777b44c
Use libavutil in libswscale, and allow it to be built out of the mplayer tree
lucabe
parents:
18861
diff
changeset
|
2458 SwsVector *vec= av_malloc(sizeof(SwsVector)); |
18861 | 2459 |
2460 vec->coeff= coeff; | |
2461 vec->length= length; | |
2462 | |
2463 for(i=0; i<length; i++) | |
2464 coeff[i]= c; | |
2465 | |
2466 return vec; | |
2467 } | |
2468 | |
2469 | |
2470 SwsVector *sws_getIdentityVec(void){ | |
2471 return sws_getConstVec(1.0, 1); | |
2472 } | |
2473 | |
2474 double sws_dcVec(SwsVector *a){ | |
2475 int i; | |
2476 double sum=0; | |
2477 | |
2478 for(i=0; i<a->length; i++) | |
2479 sum+= a->coeff[i]; | |
2480 | |
2481 return sum; | |
2482 } | |
2483 | |
2484 void sws_scaleVec(SwsVector *a, double scalar){ | |
2485 int i; | |
2486 | |
2487 for(i=0; i<a->length; i++) | |
2488 a->coeff[i]*= scalar; | |
2489 } | |
2490 | |
2491 void sws_normalizeVec(SwsVector *a, double height){ | |
2492 sws_scaleVec(a, height/sws_dcVec(a)); | |
2493 } | |
2494 | |
2495 static SwsVector *sws_getConvVec(SwsVector *a, SwsVector *b){ | |
2496 int length= a->length + b->length - 1; | |
19168
624173ae90b9
Missing part of the malloc -> av_malloc patch: memalign must be replaced as well!
reimar
parents:
19143
diff
changeset
|
2497 double *coeff= av_malloc(length*sizeof(double)); |
18861 | 2498 int i, j; |
19143
c4dac777b44c
Use libavutil in libswscale, and allow it to be built out of the mplayer tree
lucabe
parents:
18861
diff
changeset
|
2499 SwsVector *vec= av_malloc(sizeof(SwsVector)); |
18861 | 2500 |
2501 vec->coeff= coeff; | |
2502 vec->length= length; | |
2503 | |
2504 for(i=0; i<length; i++) coeff[i]= 0.0; | |
2505 | |
2506 for(i=0; i<a->length; i++) | |
2507 { | |
2508 for(j=0; j<b->length; j++) | |
2509 { | |
2510 coeff[i+j]+= a->coeff[i]*b->coeff[j]; | |
2511 } | |
2512 } | |
2513 | |
2514 return vec; | |
2515 } | |
2516 | |
2517 static SwsVector *sws_sumVec(SwsVector *a, SwsVector *b){ | |
2518 int length= MAX(a->length, b->length); | |
19168
624173ae90b9
Missing part of the malloc -> av_malloc patch: memalign must be replaced as well!
reimar
parents:
19143
diff
changeset
|
2519 double *coeff= av_malloc(length*sizeof(double)); |
18861 | 2520 int i; |
19143
c4dac777b44c
Use libavutil in libswscale, and allow it to be built out of the mplayer tree
lucabe
parents:
18861
diff
changeset
|
2521 SwsVector *vec= av_malloc(sizeof(SwsVector)); |
18861 | 2522 |
2523 vec->coeff= coeff; | |
2524 vec->length= length; | |
2525 | |
2526 for(i=0; i<length; i++) coeff[i]= 0.0; | |
2527 | |
2528 for(i=0; i<a->length; i++) coeff[i + (length-1)/2 - (a->length-1)/2]+= a->coeff[i]; | |
2529 for(i=0; i<b->length; i++) coeff[i + (length-1)/2 - (b->length-1)/2]+= b->coeff[i]; | |
2530 | |
2531 return vec; | |
2532 } | |
2533 | |
2534 static SwsVector *sws_diffVec(SwsVector *a, SwsVector *b){ | |
2535 int length= MAX(a->length, b->length); | |
19168
624173ae90b9
Missing part of the malloc -> av_malloc patch: memalign must be replaced as well!
reimar
parents:
19143
diff
changeset
|
2536 double *coeff= av_malloc(length*sizeof(double)); |
18861 | 2537 int i; |
19143
c4dac777b44c
Use libavutil in libswscale, and allow it to be built out of the mplayer tree
lucabe
parents:
18861
diff
changeset
|
2538 SwsVector *vec= av_malloc(sizeof(SwsVector)); |
18861 | 2539 |
2540 vec->coeff= coeff; | |
2541 vec->length= length; | |
2542 | |
2543 for(i=0; i<length; i++) coeff[i]= 0.0; | |
2544 | |
2545 for(i=0; i<a->length; i++) coeff[i + (length-1)/2 - (a->length-1)/2]+= a->coeff[i]; | |
2546 for(i=0; i<b->length; i++) coeff[i + (length-1)/2 - (b->length-1)/2]-= b->coeff[i]; | |
2547 | |
2548 return vec; | |
2549 } | |
2550 | |
2551 /* shift left / or right if "shift" is negative */ | |
2552 static SwsVector *sws_getShiftedVec(SwsVector *a, int shift){ | |
2553 int length= a->length + ABS(shift)*2; | |
19168
624173ae90b9
Missing part of the malloc -> av_malloc patch: memalign must be replaced as well!
reimar
parents:
19143
diff
changeset
|
2554 double *coeff= av_malloc(length*sizeof(double)); |
18861 | 2555 int i; |
19143
c4dac777b44c
Use libavutil in libswscale, and allow it to be built out of the mplayer tree
lucabe
parents:
18861
diff
changeset
|
2556 SwsVector *vec= av_malloc(sizeof(SwsVector)); |
18861 | 2557 |
2558 vec->coeff= coeff; | |
2559 vec->length= length; | |
2560 | |
2561 for(i=0; i<length; i++) coeff[i]= 0.0; | |
2562 | |
2563 for(i=0; i<a->length; i++) | |
2564 { | |
2565 coeff[i + (length-1)/2 - (a->length-1)/2 - shift]= a->coeff[i]; | |
2566 } | |
2567 | |
2568 return vec; | |
2569 } | |
2570 | |
2571 void sws_shiftVec(SwsVector *a, int shift){ | |
2572 SwsVector *shifted= sws_getShiftedVec(a, shift); | |
19143
c4dac777b44c
Use libavutil in libswscale, and allow it to be built out of the mplayer tree
lucabe
parents:
18861
diff
changeset
|
2573 av_free(a->coeff); |
18861 | 2574 a->coeff= shifted->coeff; |
2575 a->length= shifted->length; | |
19143
c4dac777b44c
Use libavutil in libswscale, and allow it to be built out of the mplayer tree
lucabe
parents:
18861
diff
changeset
|
2576 av_free(shifted); |
18861 | 2577 } |
2578 | |
2579 void sws_addVec(SwsVector *a, SwsVector *b){ | |
2580 SwsVector *sum= sws_sumVec(a, b); | |
19143
c4dac777b44c
Use libavutil in libswscale, and allow it to be built out of the mplayer tree
lucabe
parents:
18861
diff
changeset
|
2581 av_free(a->coeff); |
18861 | 2582 a->coeff= sum->coeff; |
2583 a->length= sum->length; | |
19143
c4dac777b44c
Use libavutil in libswscale, and allow it to be built out of the mplayer tree
lucabe
parents:
18861
diff
changeset
|
2584 av_free(sum); |
18861 | 2585 } |
2586 | |
2587 void sws_subVec(SwsVector *a, SwsVector *b){ | |
2588 SwsVector *diff= sws_diffVec(a, b); | |
19143
c4dac777b44c
Use libavutil in libswscale, and allow it to be built out of the mplayer tree
lucabe
parents:
18861
diff
changeset
|
2589 av_free(a->coeff); |
18861 | 2590 a->coeff= diff->coeff; |
2591 a->length= diff->length; | |
19143
c4dac777b44c
Use libavutil in libswscale, and allow it to be built out of the mplayer tree
lucabe
parents:
18861
diff
changeset
|
2592 av_free(diff); |
18861 | 2593 } |
2594 | |
2595 void sws_convVec(SwsVector *a, SwsVector *b){ | |
2596 SwsVector *conv= sws_getConvVec(a, b); | |
19143
c4dac777b44c
Use libavutil in libswscale, and allow it to be built out of the mplayer tree
lucabe
parents:
18861
diff
changeset
|
2597 av_free(a->coeff); |
18861 | 2598 a->coeff= conv->coeff; |
2599 a->length= conv->length; | |
19143
c4dac777b44c
Use libavutil in libswscale, and allow it to be built out of the mplayer tree
lucabe
parents:
18861
diff
changeset
|
2600 av_free(conv); |
18861 | 2601 } |
2602 | |
2603 SwsVector *sws_cloneVec(SwsVector *a){ | |
19168
624173ae90b9
Missing part of the malloc -> av_malloc patch: memalign must be replaced as well!
reimar
parents:
19143
diff
changeset
|
2604 double *coeff= av_malloc(a->length*sizeof(double)); |
18861 | 2605 int i; |
19143
c4dac777b44c
Use libavutil in libswscale, and allow it to be built out of the mplayer tree
lucabe
parents:
18861
diff
changeset
|
2606 SwsVector *vec= av_malloc(sizeof(SwsVector)); |
18861 | 2607 |
2608 vec->coeff= coeff; | |
2609 vec->length= a->length; | |
2610 | |
2611 for(i=0; i<a->length; i++) coeff[i]= a->coeff[i]; | |
2612 | |
2613 return vec; | |
2614 } | |
2615 | |
2616 void sws_printVec(SwsVector *a){ | |
2617 int i; | |
2618 double max=0; | |
2619 double min=0; | |
2620 double range; | |
2621 | |
2622 for(i=0; i<a->length; i++) | |
2623 if(a->coeff[i]>max) max= a->coeff[i]; | |
2624 | |
2625 for(i=0; i<a->length; i++) | |
2626 if(a->coeff[i]<min) min= a->coeff[i]; | |
2627 | |
2628 range= max - min; | |
2629 | |
2630 for(i=0; i<a->length; i++) | |
2631 { | |
2632 int x= (int)((a->coeff[i]-min)*60.0/range +0.5); | |
2633 MSG_DBG2("%1.3f ", a->coeff[i]); | |
2634 for(;x>0; x--) MSG_DBG2(" "); | |
2635 MSG_DBG2("|\n"); | |
2636 } | |
2637 } | |
2638 | |
2639 void sws_freeVec(SwsVector *a){ | |
2640 if(!a) return; | |
19143
c4dac777b44c
Use libavutil in libswscale, and allow it to be built out of the mplayer tree
lucabe
parents:
18861
diff
changeset
|
2641 av_free(a->coeff); |
18861 | 2642 a->coeff=NULL; |
2643 a->length=0; | |
19143
c4dac777b44c
Use libavutil in libswscale, and allow it to be built out of the mplayer tree
lucabe
parents:
18861
diff
changeset
|
2644 av_free(a); |
18861 | 2645 } |
2646 | |
2647 void sws_freeFilter(SwsFilter *filter){ | |
2648 if(!filter) return; | |
2649 | |
2650 if(filter->lumH) sws_freeVec(filter->lumH); | |
2651 if(filter->lumV) sws_freeVec(filter->lumV); | |
2652 if(filter->chrH) sws_freeVec(filter->chrH); | |
2653 if(filter->chrV) sws_freeVec(filter->chrV); | |
19143
c4dac777b44c
Use libavutil in libswscale, and allow it to be built out of the mplayer tree
lucabe
parents:
18861
diff
changeset
|
2654 av_free(filter); |
18861 | 2655 } |
2656 | |
2657 | |
2658 void sws_freeContext(SwsContext *c){ | |
2659 int i; | |
2660 if(!c) return; | |
2661 | |
2662 if(c->lumPixBuf) | |
2663 { | |
2664 for(i=0; i<c->vLumBufSize; i++) | |
2665 { | |
19143
c4dac777b44c
Use libavutil in libswscale, and allow it to be built out of the mplayer tree
lucabe
parents:
18861
diff
changeset
|
2666 av_free(c->lumPixBuf[i]); |
18861 | 2667 c->lumPixBuf[i]=NULL; |
2668 } | |
19143
c4dac777b44c
Use libavutil in libswscale, and allow it to be built out of the mplayer tree
lucabe
parents:
18861
diff
changeset
|
2669 av_free(c->lumPixBuf); |
18861 | 2670 c->lumPixBuf=NULL; |
2671 } | |
2672 | |
2673 if(c->chrPixBuf) | |
2674 { | |
2675 for(i=0; i<c->vChrBufSize; i++) | |
2676 { | |
19143
c4dac777b44c
Use libavutil in libswscale, and allow it to be built out of the mplayer tree
lucabe
parents:
18861
diff
changeset
|
2677 av_free(c->chrPixBuf[i]); |
18861 | 2678 c->chrPixBuf[i]=NULL; |
2679 } | |
19143
c4dac777b44c
Use libavutil in libswscale, and allow it to be built out of the mplayer tree
lucabe
parents:
18861
diff
changeset
|
2680 av_free(c->chrPixBuf); |
18861 | 2681 c->chrPixBuf=NULL; |
2682 } | |
2683 | |
19143
c4dac777b44c
Use libavutil in libswscale, and allow it to be built out of the mplayer tree
lucabe
parents:
18861
diff
changeset
|
2684 av_free(c->vLumFilter); |
18861 | 2685 c->vLumFilter = NULL; |
19143
c4dac777b44c
Use libavutil in libswscale, and allow it to be built out of the mplayer tree
lucabe
parents:
18861
diff
changeset
|
2686 av_free(c->vChrFilter); |
18861 | 2687 c->vChrFilter = NULL; |
19143
c4dac777b44c
Use libavutil in libswscale, and allow it to be built out of the mplayer tree
lucabe
parents:
18861
diff
changeset
|
2688 av_free(c->hLumFilter); |
18861 | 2689 c->hLumFilter = NULL; |
19143
c4dac777b44c
Use libavutil in libswscale, and allow it to be built out of the mplayer tree
lucabe
parents:
18861
diff
changeset
|
2690 av_free(c->hChrFilter); |
18861 | 2691 c->hChrFilter = NULL; |
2692 #ifdef HAVE_ALTIVEC | |
19143
c4dac777b44c
Use libavutil in libswscale, and allow it to be built out of the mplayer tree
lucabe
parents:
18861
diff
changeset
|
2693 av_free(c->vYCoeffsBank); |
18861 | 2694 c->vYCoeffsBank = NULL; |
19143
c4dac777b44c
Use libavutil in libswscale, and allow it to be built out of the mplayer tree
lucabe
parents:
18861
diff
changeset
|
2695 av_free(c->vCCoeffsBank); |
18861 | 2696 c->vCCoeffsBank = NULL; |
2697 #endif | |
2698 | |
19143
c4dac777b44c
Use libavutil in libswscale, and allow it to be built out of the mplayer tree
lucabe
parents:
18861
diff
changeset
|
2699 av_free(c->vLumFilterPos); |
18861 | 2700 c->vLumFilterPos = NULL; |
19143
c4dac777b44c
Use libavutil in libswscale, and allow it to be built out of the mplayer tree
lucabe
parents:
18861
diff
changeset
|
2701 av_free(c->vChrFilterPos); |
18861 | 2702 c->vChrFilterPos = NULL; |
19143
c4dac777b44c
Use libavutil in libswscale, and allow it to be built out of the mplayer tree
lucabe
parents:
18861
diff
changeset
|
2703 av_free(c->hLumFilterPos); |
18861 | 2704 c->hLumFilterPos = NULL; |
19143
c4dac777b44c
Use libavutil in libswscale, and allow it to be built out of the mplayer tree
lucabe
parents:
18861
diff
changeset
|
2705 av_free(c->hChrFilterPos); |
18861 | 2706 c->hChrFilterPos = NULL; |
2707 | |
2708 #if defined(ARCH_X86) || defined(ARCH_X86_64) | |
2709 #ifdef MAP_ANONYMOUS | |
2710 if(c->funnyYCode) munmap(c->funnyYCode, MAX_FUNNY_CODE_SIZE); | |
2711 if(c->funnyUVCode) munmap(c->funnyUVCode, MAX_FUNNY_CODE_SIZE); | |
2712 #else | |
19143
c4dac777b44c
Use libavutil in libswscale, and allow it to be built out of the mplayer tree
lucabe
parents:
18861
diff
changeset
|
2713 av_free(c->funnyYCode); |
c4dac777b44c
Use libavutil in libswscale, and allow it to be built out of the mplayer tree
lucabe
parents:
18861
diff
changeset
|
2714 av_free(c->funnyUVCode); |
18861 | 2715 #endif |
2716 c->funnyYCode=NULL; | |
2717 c->funnyUVCode=NULL; | |
2718 #endif | |
2719 | |
19143
c4dac777b44c
Use libavutil in libswscale, and allow it to be built out of the mplayer tree
lucabe
parents:
18861
diff
changeset
|
2720 av_free(c->lumMmx2Filter); |
18861 | 2721 c->lumMmx2Filter=NULL; |
19143
c4dac777b44c
Use libavutil in libswscale, and allow it to be built out of the mplayer tree
lucabe
parents:
18861
diff
changeset
|
2722 av_free(c->chrMmx2Filter); |
18861 | 2723 c->chrMmx2Filter=NULL; |
19143
c4dac777b44c
Use libavutil in libswscale, and allow it to be built out of the mplayer tree
lucabe
parents:
18861
diff
changeset
|
2724 av_free(c->lumMmx2FilterPos); |
18861 | 2725 c->lumMmx2FilterPos=NULL; |
19143
c4dac777b44c
Use libavutil in libswscale, and allow it to be built out of the mplayer tree
lucabe
parents:
18861
diff
changeset
|
2726 av_free(c->chrMmx2FilterPos); |
18861 | 2727 c->chrMmx2FilterPos=NULL; |
19143
c4dac777b44c
Use libavutil in libswscale, and allow it to be built out of the mplayer tree
lucabe
parents:
18861
diff
changeset
|
2728 av_free(c->yuvTable); |
18861 | 2729 c->yuvTable=NULL; |
2730 | |
19143
c4dac777b44c
Use libavutil in libswscale, and allow it to be built out of the mplayer tree
lucabe
parents:
18861
diff
changeset
|
2731 av_free(c); |
18861 | 2732 } |
2733 |