vf_mp: remove boxblur wrapper
[ffmpeg.git] / libavfilter / libmpcodecs / vf_spp.c
1 /*
2  * Copyright (C) 2003 Michael Niedermayer <michaelni@gmx.at>
3  *
4  * This file is part of MPlayer.
5  *
6  * MPlayer is free software; you can redistribute it and/or modify
7  * it under the terms of the GNU General Public License as published by
8  * the Free Software Foundation; either version 2 of the License, or
9  * (at your option) any later version.
10  *
11  * MPlayer is distributed in the hope that it will be useful,
12  * but WITHOUT ANY WARRANTY; without even the implied warranty of
13  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
14  * GNU General Public License for more details.
15  *
16  * You should have received a copy of the GNU General Public License along
17  * with MPlayer; if not, write to the Free Software Foundation, Inc.,
18  * 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA.
19  */
20
21 /*
22  * This implementation is based on an algorithm described in
23  * "Aria Nosratinia Embedded Post-Processing for
24  * Enhancement of Compressed Images (1999)"
25  * (http://citeseer.nj.nec.com/nosratinia99embedded.html)
26  */
27
28
29 #include <stdio.h>
30 #include <stdlib.h>
31 #include <string.h>
32 #include <inttypes.h>
33 #include <math.h>
34
35 #include "config.h"
36
37 #include "mp_msg.h"
38 #include "cpudetect.h"
39
40 #include "libavutil/internal.h"
41 #include "libavutil/intreadwrite.h"
42 #include "libavcodec/avcodec.h"
43 #include "libavcodec/dsputil.h"
44
45 #undef fprintf
46 #undef free
47 #undef malloc
48
49 #include "img_format.h"
50 #include "mp_image.h"
51 #include "vf.h"
52 #include "vd_ffmpeg.h"
53 #include "libvo/fastmemcpy.h"
54
55 #define XMIN(a,b) ((a) < (b) ? (a) : (b))
56
57 //===========================================================================//
58 static const uint8_t  __attribute__((aligned(8))) dither[8][8]={
59 {  0,  48,  12,  60,   3,  51,  15,  63, },
60 { 32,  16,  44,  28,  35,  19,  47,  31, },
61 {  8,  56,   4,  52,  11,  59,   7,  55, },
62 { 40,  24,  36,  20,  43,  27,  39,  23, },
63 {  2,  50,  14,  62,   1,  49,  13,  61, },
64 { 34,  18,  46,  30,  33,  17,  45,  29, },
65 { 10,  58,   6,  54,   9,  57,   5,  53, },
66 { 42,  26,  38,  22,  41,  25,  37,  21, },
67 };
68
69 static const uint8_t offset[127][2]= {
70 {0,0},
71 {0,0}, {4,4},
72 {0,0}, {2,2}, {6,4}, {4,6},
73 {0,0}, {5,1}, {2,2}, {7,3}, {4,4}, {1,5}, {6,6}, {3,7},
74
75 {0,0}, {4,0}, {1,1}, {5,1}, {3,2}, {7,2}, {2,3}, {6,3},
76 {0,4}, {4,4}, {1,5}, {5,5}, {3,6}, {7,6}, {2,7}, {6,7},
77
78 {0,0}, {0,2}, {0,4}, {0,6}, {1,1}, {1,3}, {1,5}, {1,7},
79 {2,0}, {2,2}, {2,4}, {2,6}, {3,1}, {3,3}, {3,5}, {3,7},
80 {4,0}, {4,2}, {4,4}, {4,6}, {5,1}, {5,3}, {5,5}, {5,7},
81 {6,0}, {6,2}, {6,4}, {6,6}, {7,1}, {7,3}, {7,5}, {7,7},
82
83 {0,0}, {4,4}, {0,4}, {4,0}, {2,2}, {6,6}, {2,6}, {6,2},
84 {0,2}, {4,6}, {0,6}, {4,2}, {2,0}, {6,4}, {2,4}, {6,0},
85 {1,1}, {5,5}, {1,5}, {5,1}, {3,3}, {7,7}, {3,7}, {7,3},
86 {1,3}, {5,7}, {1,7}, {5,3}, {3,1}, {7,5}, {3,5}, {7,1},
87 {0,1}, {4,5}, {0,5}, {4,1}, {2,3}, {6,7}, {2,7}, {6,3},
88 {0,3}, {4,7}, {0,7}, {4,3}, {2,1}, {6,5}, {2,5}, {6,1},
89 {1,0}, {5,4}, {1,4}, {5,0}, {3,2}, {7,6}, {3,6}, {7,2},
90 {1,2}, {5,6}, {1,6}, {5,2}, {3,0}, {7,4}, {3,4}, {7,0},
91 };
92
93 struct vf_priv_s {
94         int log2_count;
95         int qp;
96         int mode;
97         int mpeg2;
98         int temp_stride;
99         uint8_t *src;
100         int16_t *temp;
101         AVCodecContext *avctx;
102         DSPContext dsp;
103         char *non_b_qp;
104 };
105
106 #define SHIFT 22
107
108 static void hardthresh_c(DCTELEM dst[64], DCTELEM src[64], int qp, uint8_t *permutation){
109         int i;
110         int bias= 0; //FIXME
111         unsigned int threshold1, threshold2;
112
113         threshold1= qp*((1<<4) - bias) - 1;
114         threshold2= (threshold1<<1);
115
116         memset(dst, 0, 64*sizeof(DCTELEM));
117         dst[0]= (src[0] + 4)>>3;
118
119         for(i=1; i<64; i++){
120                 int level= src[i];
121                 if(((unsigned)(level+threshold1))>threshold2){
122                         const int j= permutation[i];
123                         dst[j]= (level + 4)>>3;
124                 }
125         }
126 }
127
128 static void softthresh_c(DCTELEM dst[64], DCTELEM src[64], int qp, uint8_t *permutation){
129         int i;
130         int bias= 0; //FIXME
131         unsigned int threshold1, threshold2;
132
133         threshold1= qp*((1<<4) - bias) - 1;
134         threshold2= (threshold1<<1);
135
136         memset(dst, 0, 64*sizeof(DCTELEM));
137         dst[0]= (src[0] + 4)>>3;
138
139         for(i=1; i<64; i++){
140                 int level= src[i];
141                 if(((unsigned)(level+threshold1))>threshold2){
142                         const int j= permutation[i];
143                         if(level>0)
144                                 dst[j]= (level - threshold1 + 4)>>3;
145                         else
146                                 dst[j]= (level + threshold1 + 4)>>3;
147                 }
148         }
149 }
150
151 #if HAVE_MMX
152 static void hardthresh_mmx(DCTELEM dst[64], DCTELEM src[64], int qp, uint8_t *permutation){
153         int bias= 0; //FIXME
154         unsigned int threshold1;
155
156         threshold1= qp*((1<<4) - bias) - 1;
157
158         __asm__ volatile(
159 #define REQUANT_CORE(dst0, dst1, dst2, dst3, src0, src1, src2, src3) \
160                 "movq " #src0 ", %%mm0        \n\t"\
161                 "movq " #src1 ", %%mm1        \n\t"\
162                 "movq " #src2 ", %%mm2        \n\t"\
163                 "movq " #src3 ", %%mm3        \n\t"\
164                 "psubw %%mm4, %%mm0        \n\t"\
165                 "psubw %%mm4, %%mm1        \n\t"\
166                 "psubw %%mm4, %%mm2        \n\t"\
167                 "psubw %%mm4, %%mm3        \n\t"\
168                 "paddusw %%mm5, %%mm0        \n\t"\
169                 "paddusw %%mm5, %%mm1        \n\t"\
170                 "paddusw %%mm5, %%mm2        \n\t"\
171                 "paddusw %%mm5, %%mm3        \n\t"\
172                 "paddw %%mm6, %%mm0        \n\t"\
173                 "paddw %%mm6, %%mm1        \n\t"\
174                 "paddw %%mm6, %%mm2        \n\t"\
175                 "paddw %%mm6, %%mm3        \n\t"\
176                 "psubusw %%mm6, %%mm0        \n\t"\
177                 "psubusw %%mm6, %%mm1        \n\t"\
178                 "psubusw %%mm6, %%mm2        \n\t"\
179                 "psubusw %%mm6, %%mm3        \n\t"\
180                 "psraw $3, %%mm0        \n\t"\
181                 "psraw $3, %%mm1        \n\t"\
182                 "psraw $3, %%mm2        \n\t"\
183                 "psraw $3, %%mm3        \n\t"\
184 \
185                 "movq %%mm0, %%mm7        \n\t"\
186                 "punpcklwd %%mm2, %%mm0        \n\t" /*A*/\
187                 "punpckhwd %%mm2, %%mm7        \n\t" /*C*/\
188                 "movq %%mm1, %%mm2        \n\t"\
189                 "punpcklwd %%mm3, %%mm1        \n\t" /*B*/\
190                 "punpckhwd %%mm3, %%mm2        \n\t" /*D*/\
191                 "movq %%mm0, %%mm3        \n\t"\
192                 "punpcklwd %%mm1, %%mm0        \n\t" /*A*/\
193                 "punpckhwd %%mm7, %%mm3        \n\t" /*C*/\
194                 "punpcklwd %%mm2, %%mm7        \n\t" /*B*/\
195                 "punpckhwd %%mm2, %%mm1        \n\t" /*D*/\
196 \
197                 "movq %%mm0, " #dst0 "        \n\t"\
198                 "movq %%mm7, " #dst1 "        \n\t"\
199                 "movq %%mm3, " #dst2 "        \n\t"\
200                 "movq %%mm1, " #dst3 "        \n\t"
201
202                 "movd %2, %%mm4                \n\t"
203                 "movd %3, %%mm5                \n\t"
204                 "movd %4, %%mm6                \n\t"
205                 "packssdw %%mm4, %%mm4        \n\t"
206                 "packssdw %%mm5, %%mm5        \n\t"
207                 "packssdw %%mm6, %%mm6        \n\t"
208                 "packssdw %%mm4, %%mm4        \n\t"
209                 "packssdw %%mm5, %%mm5        \n\t"
210                 "packssdw %%mm6, %%mm6        \n\t"
211                 REQUANT_CORE(  (%1),  8(%1), 16(%1), 24(%1),  (%0), 8(%0), 64(%0), 72(%0))
212                 REQUANT_CORE(32(%1), 40(%1), 48(%1), 56(%1),16(%0),24(%0), 48(%0), 56(%0))
213                 REQUANT_CORE(64(%1), 72(%1), 80(%1), 88(%1),32(%0),40(%0), 96(%0),104(%0))
214                 REQUANT_CORE(96(%1),104(%1),112(%1),120(%1),80(%0),88(%0),112(%0),120(%0))
215                 : : "r" (src), "r" (dst), "g" (threshold1+1), "g" (threshold1+5), "g" (threshold1-4) //FIXME maybe more accurate then needed?
216         );
217         dst[0]= (src[0] + 4)>>3;
218 }
219
220 static void softthresh_mmx(DCTELEM dst[64], DCTELEM src[64], int qp, uint8_t *permutation){
221         int bias= 0; //FIXME
222         unsigned int threshold1;
223
224         threshold1= qp*((1<<4) - bias) - 1;
225
226         __asm__ volatile(
227 #undef REQUANT_CORE
228 #define REQUANT_CORE(dst0, dst1, dst2, dst3, src0, src1, src2, src3) \
229                 "movq " #src0 ", %%mm0        \n\t"\
230                 "movq " #src1 ", %%mm1        \n\t"\
231                 "pxor %%mm6, %%mm6        \n\t"\
232                 "pxor %%mm7, %%mm7        \n\t"\
233                 "pcmpgtw %%mm0, %%mm6        \n\t"\
234                 "pcmpgtw %%mm1, %%mm7        \n\t"\
235                 "pxor %%mm6, %%mm0        \n\t"\
236                 "pxor %%mm7, %%mm1        \n\t"\
237                 "psubusw %%mm4, %%mm0        \n\t"\
238                 "psubusw %%mm4, %%mm1        \n\t"\
239                 "pxor %%mm6, %%mm0        \n\t"\
240                 "pxor %%mm7, %%mm1        \n\t"\
241                 "movq " #src2 ", %%mm2        \n\t"\
242                 "movq " #src3 ", %%mm3        \n\t"\
243                 "pxor %%mm6, %%mm6        \n\t"\
244                 "pxor %%mm7, %%mm7        \n\t"\
245                 "pcmpgtw %%mm2, %%mm6        \n\t"\
246                 "pcmpgtw %%mm3, %%mm7        \n\t"\
247                 "pxor %%mm6, %%mm2        \n\t"\
248                 "pxor %%mm7, %%mm3        \n\t"\
249                 "psubusw %%mm4, %%mm2        \n\t"\
250                 "psubusw %%mm4, %%mm3        \n\t"\
251                 "pxor %%mm6, %%mm2        \n\t"\
252                 "pxor %%mm7, %%mm3        \n\t"\
253 \
254                 "paddsw %%mm5, %%mm0        \n\t"\
255                 "paddsw %%mm5, %%mm1        \n\t"\
256                 "paddsw %%mm5, %%mm2        \n\t"\
257                 "paddsw %%mm5, %%mm3        \n\t"\
258                 "psraw $3, %%mm0        \n\t"\
259                 "psraw $3, %%mm1        \n\t"\
260                 "psraw $3, %%mm2        \n\t"\
261                 "psraw $3, %%mm3        \n\t"\
262 \
263                 "movq %%mm0, %%mm7        \n\t"\
264                 "punpcklwd %%mm2, %%mm0        \n\t" /*A*/\
265                 "punpckhwd %%mm2, %%mm7        \n\t" /*C*/\
266                 "movq %%mm1, %%mm2        \n\t"\
267                 "punpcklwd %%mm3, %%mm1        \n\t" /*B*/\
268                 "punpckhwd %%mm3, %%mm2        \n\t" /*D*/\
269                 "movq %%mm0, %%mm3        \n\t"\
270                 "punpcklwd %%mm1, %%mm0        \n\t" /*A*/\
271                 "punpckhwd %%mm7, %%mm3        \n\t" /*C*/\
272                 "punpcklwd %%mm2, %%mm7        \n\t" /*B*/\
273                 "punpckhwd %%mm2, %%mm1        \n\t" /*D*/\
274 \
275                 "movq %%mm0, " #dst0 "        \n\t"\
276                 "movq %%mm7, " #dst1 "        \n\t"\
277                 "movq %%mm3, " #dst2 "        \n\t"\
278                 "movq %%mm1, " #dst3 "        \n\t"
279
280                 "movd %2, %%mm4                \n\t"
281                 "movd %3, %%mm5                \n\t"
282                 "packssdw %%mm4, %%mm4        \n\t"
283                 "packssdw %%mm5, %%mm5        \n\t"
284                 "packssdw %%mm4, %%mm4        \n\t"
285                 "packssdw %%mm5, %%mm5        \n\t"
286                 REQUANT_CORE(  (%1),  8(%1), 16(%1), 24(%1),  (%0), 8(%0), 64(%0), 72(%0))
287                 REQUANT_CORE(32(%1), 40(%1), 48(%1), 56(%1),16(%0),24(%0), 48(%0), 56(%0))
288                 REQUANT_CORE(64(%1), 72(%1), 80(%1), 88(%1),32(%0),40(%0), 96(%0),104(%0))
289                 REQUANT_CORE(96(%1),104(%1),112(%1),120(%1),80(%0),88(%0),112(%0),120(%0))
290                 : : "r" (src), "r" (dst), "g" (threshold1), "rm" (4) //FIXME maybe more accurate then needed?
291         );
292
293         dst[0]= (src[0] + 4)>>3;
294 }
295 #endif
296
297 static inline void add_block(int16_t *dst, int stride, DCTELEM block[64]){
298         int y;
299
300         for(y=0; y<8; y++){
301                 *(uint32_t*)&dst[0 + y*stride]+= *(uint32_t*)&block[0 + y*8];
302                 *(uint32_t*)&dst[2 + y*stride]+= *(uint32_t*)&block[2 + y*8];
303                 *(uint32_t*)&dst[4 + y*stride]+= *(uint32_t*)&block[4 + y*8];
304                 *(uint32_t*)&dst[6 + y*stride]+= *(uint32_t*)&block[6 + y*8];
305         }
306 }
307
308 static void store_slice_c(uint8_t *dst, int16_t *src, int dst_stride, int src_stride, int width, int height, int log2_scale){
309         int y, x;
310
311 #define STORE(pos) \
312         temp= ((src[x + y*src_stride + pos]<<log2_scale) + d[pos])>>6;\
313         if(temp & 0x100) temp= ~(temp>>31);\
314         dst[x + y*dst_stride + pos]= temp;
315
316         for(y=0; y<height; y++){
317                 const uint8_t *d= dither[y];
318                 for(x=0; x<width; x+=8){
319                         int temp;
320                         STORE(0);
321                         STORE(1);
322                         STORE(2);
323                         STORE(3);
324                         STORE(4);
325                         STORE(5);
326                         STORE(6);
327                         STORE(7);
328                 }
329         }
330 }
331
332 #if HAVE_MMX
333 static void store_slice_mmx(uint8_t *dst, int16_t *src, int dst_stride, int src_stride, int width, int height, int log2_scale){
334         int y;
335
336         for(y=0; y<height; y++){
337                 uint8_t *dst1= dst;
338                 int16_t *src1= src;
339                 __asm__ volatile(
340                         "movq (%3), %%mm3        \n\t"
341                         "movq (%3), %%mm4        \n\t"
342                         "movd %4, %%mm2                \n\t"
343                         "pxor %%mm0, %%mm0        \n\t"
344                         "punpcklbw %%mm0, %%mm3        \n\t"
345                         "punpckhbw %%mm0, %%mm4        \n\t"
346                         "psraw %%mm2, %%mm3        \n\t"
347                         "psraw %%mm2, %%mm4        \n\t"
348                         "movd %5, %%mm2                \n\t"
349                         "1:                        \n\t"
350                         "movq (%0), %%mm0        \n\t"
351                         "movq 8(%0), %%mm1        \n\t"
352                         "paddw %%mm3, %%mm0        \n\t"
353                         "paddw %%mm4, %%mm1        \n\t"
354                         "psraw %%mm2, %%mm0        \n\t"
355                         "psraw %%mm2, %%mm1        \n\t"
356                         "packuswb %%mm1, %%mm0        \n\t"
357                         "movq %%mm0, (%1)         \n\t"
358                         "add $16, %0                \n\t"
359                         "add $8, %1                \n\t"
360                         "cmp %2, %1                \n\t"
361                         " jb 1b                        \n\t"
362                         : "+r" (src1), "+r"(dst1)
363                         : "r"(dst + width), "r"(dither[y]), "g"(log2_scale), "g"(6-log2_scale)
364                 );
365                 src += src_stride;
366                 dst += dst_stride;
367         }
368 //        if(width != mmxw)
369 //                store_slice_c(dst + mmxw, src + mmxw, dst_stride, src_stride, width - mmxw, log2_scale);
370 }
371 #endif
372
373 static void (*store_slice)(uint8_t *dst, int16_t *src, int dst_stride, int src_stride, int width, int height, int log2_scale)= store_slice_c;
374
375 static void (*requantize)(DCTELEM dst[64], DCTELEM src[64], int qp, uint8_t *permutation)= hardthresh_c;
376
377 static void filter(struct vf_priv_s *p, uint8_t *dst, uint8_t *src, int dst_stride, int src_stride, int width, int height, uint8_t *qp_store, int qp_stride, int is_luma){
378         int x, y, i;
379         const int count= 1<<p->log2_count;
380         const int stride= is_luma ? p->temp_stride : ((width+16+15)&(~15));
381         uint64_t __attribute__((aligned(16))) block_align[32];
382         DCTELEM *block = (DCTELEM *)block_align;
383         DCTELEM *block2= (DCTELEM *)(block_align+16);
384
385         if (!src || !dst) return; // HACK avoid crash for Y8 colourspace
386         for(y=0; y<height; y++){
387                 int index= 8 + 8*stride + y*stride;
388                 fast_memcpy(p->src + index, src + y*src_stride, width);
389                 for(x=0; x<8; x++){
390                         p->src[index         - x - 1]= p->src[index +         x    ];
391                         p->src[index + width + x    ]= p->src[index + width - x - 1];
392                 }
393         }
394         for(y=0; y<8; y++){
395                 fast_memcpy(p->src + (      7-y)*stride, p->src + (      y+8)*stride, stride);
396                 fast_memcpy(p->src + (height+8+y)*stride, p->src + (height-y+7)*stride, stride);
397         }
398         //FIXME (try edge emu)
399
400         for(y=0; y<height+8; y+=8){
401                 memset(p->temp + (8+y)*stride, 0, 8*stride*sizeof(int16_t));
402                 for(x=0; x<width+8; x+=8){
403                         const int qps= 3 + is_luma;
404                         int qp;
405
406                         if(p->qp)
407                                 qp= p->qp;
408                         else{
409                                 qp= qp_store[ (XMIN(x, width-1)>>qps) + (XMIN(y, height-1)>>qps) * qp_stride];
410                                 qp = FFMAX(1, norm_qscale(qp, p->mpeg2));
411                         }
412                         for(i=0; i<count; i++){
413                                 const int x1= x + offset[i+count-1][0];
414                                 const int y1= y + offset[i+count-1][1];
415                                 const int index= x1 + y1*stride;
416                                 p->dsp.get_pixels(block, p->src + index, stride);
417                                 p->dsp.fdct(block);
418                                 requantize(block2, block, qp, p->dsp.idct_permutation);
419                                 p->dsp.idct(block2);
420                                 add_block(p->temp + index, stride, block2);
421                         }
422                 }
423                 if(y)
424                         store_slice(dst + (y-8)*dst_stride, p->temp + 8 + y*stride, dst_stride, stride, width, XMIN(8, height+8-y), 6-p->log2_count);
425         }
426 #if 0
427         for(y=0; y<height; y++){
428                 for(x=0; x<width; x++){
429                         if((((x>>6) ^ (y>>6)) & 1) == 0)
430                                 dst[x + y*dst_stride]= p->src[8 + 8*stride  + x + y*stride];
431                         if((x&63) == 0 || (y&63)==0)
432                                 dst[x + y*dst_stride] += 128;
433                 }
434         }
435 #endif
436         //FIXME reorder for better caching
437 }
438
439 static int config(struct vf_instance *vf,
440         int width, int height, int d_width, int d_height,
441         unsigned int flags, unsigned int outfmt){
442         int h= (height+16+15)&(~15);
443
444         vf->priv->temp_stride= (width+16+15)&(~15);
445         vf->priv->temp= malloc(vf->priv->temp_stride*h*sizeof(int16_t));
446         vf->priv->src = malloc(vf->priv->temp_stride*h*sizeof(uint8_t));
447
448         return vf_next_config(vf,width,height,d_width,d_height,flags,outfmt);
449 }
450
451 static void get_image(struct vf_instance *vf, mp_image_t *mpi){
452     if(mpi->flags&MP_IMGFLAG_PRESERVE) return; // don't change
453     // ok, we can do pp in-place (or pp disabled):
454     vf->dmpi=vf_get_image(vf->next,mpi->imgfmt,
455         mpi->type, mpi->flags | MP_IMGFLAG_READABLE, mpi->width, mpi->height);
456     mpi->planes[0]=vf->dmpi->planes[0];
457     mpi->stride[0]=vf->dmpi->stride[0];
458     mpi->width=vf->dmpi->width;
459     if(mpi->flags&MP_IMGFLAG_PLANAR){
460         mpi->planes[1]=vf->dmpi->planes[1];
461         mpi->planes[2]=vf->dmpi->planes[2];
462         mpi->stride[1]=vf->dmpi->stride[1];
463         mpi->stride[2]=vf->dmpi->stride[2];
464     }
465     mpi->flags|=MP_IMGFLAG_DIRECT;
466 }
467
468 static int put_image(struct vf_instance *vf, mp_image_t *mpi, double pts){
469         mp_image_t *dmpi;
470
471         if(!(mpi->flags&MP_IMGFLAG_DIRECT)){
472                 // no DR, so get a new image! hope we'll get DR buffer:
473                 dmpi=vf_get_image(vf->next,mpi->imgfmt,
474                     MP_IMGTYPE_TEMP,
475                     MP_IMGFLAG_ACCEPT_STRIDE|MP_IMGFLAG_PREFER_ALIGNED_STRIDE,
476                     mpi->width,mpi->height);
477                 vf_clone_mpi_attributes(dmpi, mpi);
478         }else{
479            dmpi=vf->dmpi;
480         }
481
482         vf->priv->mpeg2= mpi->qscale_type;
483         if(mpi->pict_type != 3 && mpi->qscale && !vf->priv->qp){
484             int w = mpi->qstride;
485             int h = (mpi->h + 15) >> 4;
486             if (!w) {
487                 w = (mpi->w + 15) >> 4;
488                 h = 1;
489             }
490             if(!vf->priv->non_b_qp)
491                 vf->priv->non_b_qp= malloc(w*h);
492             fast_memcpy(vf->priv->non_b_qp, mpi->qscale, w*h);
493         }
494         if(vf->priv->log2_count || !(mpi->flags&MP_IMGFLAG_DIRECT)){
495             char *qp_tab= vf->priv->non_b_qp;
496             if((vf->priv->mode&4) || !qp_tab)
497                 qp_tab= mpi->qscale;
498
499             if(qp_tab || vf->priv->qp){
500                 filter(vf->priv, dmpi->planes[0], mpi->planes[0], dmpi->stride[0], mpi->stride[0], mpi->w, mpi->h, qp_tab, mpi->qstride, 1);
501                 filter(vf->priv, dmpi->planes[1], mpi->planes[1], dmpi->stride[1], mpi->stride[1], mpi->w>>mpi->chroma_x_shift, mpi->h>>mpi->chroma_y_shift, qp_tab, mpi->qstride, 0);
502                 filter(vf->priv, dmpi->planes[2], mpi->planes[2], dmpi->stride[2], mpi->stride[2], mpi->w>>mpi->chroma_x_shift, mpi->h>>mpi->chroma_y_shift, qp_tab, mpi->qstride, 0);
503             }else{
504                 memcpy_pic(dmpi->planes[0], mpi->planes[0], mpi->w, mpi->h, dmpi->stride[0], mpi->stride[0]);
505                 memcpy_pic(dmpi->planes[1], mpi->planes[1], mpi->w>>mpi->chroma_x_shift, mpi->h>>mpi->chroma_y_shift, dmpi->stride[1], mpi->stride[1]);
506                 memcpy_pic(dmpi->planes[2], mpi->planes[2], mpi->w>>mpi->chroma_x_shift, mpi->h>>mpi->chroma_y_shift, dmpi->stride[2], mpi->stride[2]);
507             }
508         }
509
510 #if HAVE_MMX
511         if(gCpuCaps.hasMMX) __asm__ volatile ("emms\n\t");
512 #endif
513 #if HAVE_MMX2
514         if(gCpuCaps.hasMMX2) __asm__ volatile ("sfence\n\t");
515 #endif
516
517         return vf_next_put_image(vf,dmpi, pts);
518 }
519
520 static void uninit(struct vf_instance *vf){
521         if(!vf->priv) return;
522
523         free(vf->priv->temp);
524         vf->priv->temp= NULL;
525         free(vf->priv->src);
526         vf->priv->src= NULL;
527         free(vf->priv->avctx);
528         vf->priv->avctx= NULL;
529         free(vf->priv->non_b_qp);
530         vf->priv->non_b_qp= NULL;
531
532         free(vf->priv);
533         vf->priv=NULL;
534 }
535
536 //===========================================================================//
537 static int query_format(struct vf_instance *vf, unsigned int fmt){
538     switch(fmt){
539         case IMGFMT_YVU9:
540         case IMGFMT_IF09:
541         case IMGFMT_YV12:
542         case IMGFMT_I420:
543         case IMGFMT_IYUV:
544         case IMGFMT_CLPL:
545         case IMGFMT_Y800:
546         case IMGFMT_Y8:
547         case IMGFMT_444P:
548         case IMGFMT_422P:
549         case IMGFMT_411P:
550             return vf_next_query_format(vf,fmt);
551     }
552     return 0;
553 }
554
555 static int control(struct vf_instance *vf, int request, void* data){
556     switch(request){
557     case VFCTRL_QUERY_MAX_PP_LEVEL:
558         return 6;
559     case VFCTRL_SET_PP_LEVEL:
560         vf->priv->log2_count= *((unsigned int*)data);
561         return CONTROL_TRUE;
562     }
563     return vf_next_control(vf,request,data);
564 }
565
566 static int vf_open(vf_instance_t *vf, char *args){
567
568     int log2c=-1;
569
570     vf->config=config;
571     vf->put_image=put_image;
572     vf->get_image=get_image;
573     vf->query_format=query_format;
574     vf->uninit=uninit;
575     vf->control= control;
576     vf->priv=malloc(sizeof(struct vf_priv_s));
577     memset(vf->priv, 0, sizeof(struct vf_priv_s));
578
579     init_avcodec();
580
581     vf->priv->avctx= avcodec_alloc_context();
582     dsputil_init(&vf->priv->dsp, vf->priv->avctx);
583
584     vf->priv->log2_count= 3;
585
586     if (args) sscanf(args, "%d:%d:%d", &log2c, &vf->priv->qp, &vf->priv->mode);
587
588     if( log2c >=0 && log2c <=6 )
589         vf->priv->log2_count = log2c;
590
591     if(vf->priv->qp < 0)
592         vf->priv->qp = 0;
593
594     switch(vf->priv->mode&3){
595         default:
596         case 0: requantize= hardthresh_c; break;
597         case 1: requantize= softthresh_c; break;
598     }
599
600 #if HAVE_MMX
601     if(gCpuCaps.hasMMX){
602         store_slice= store_slice_mmx;
603         switch(vf->priv->mode&3){
604             case 0: requantize= hardthresh_mmx; break;
605             case 1: requantize= softthresh_mmx; break;
606         }
607     }
608 #endif
609
610     return 1;
611 }
612
613 const vf_info_t vf_info_spp = {
614     "simple postprocess",
615     "spp",
616     "Michael Niedermayer",
617     "",
618     vf_open,
619     NULL
620 };