Merge commit 'cebdedca57d95834a8f8098c7b6a322a1163e26b'
[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/common.h"
41 #include "libavutil/internal.h"
42 #include "libavutil/intreadwrite.h"
43 #include "libavcodec/avcodec.h"
44 #include "libavcodec/dsputil.h"
45
46 #undef fprintf
47 #undef free
48 #undef malloc
49
50 #include "img_format.h"
51 #include "mp_image.h"
52 #include "vf.h"
53 #include "av_helpers.h"
54 #include "libvo/fastmemcpy.h"
55
56 #define XMIN(a,b) ((a) < (b) ? (a) : (b))
57
58 //===========================================================================//
59 static const uint8_t  __attribute__((aligned(8))) dither[8][8]={
60 {  0,  48,  12,  60,   3,  51,  15,  63, },
61 { 32,  16,  44,  28,  35,  19,  47,  31, },
62 {  8,  56,   4,  52,  11,  59,   7,  55, },
63 { 40,  24,  36,  20,  43,  27,  39,  23, },
64 {  2,  50,  14,  62,   1,  49,  13,  61, },
65 { 34,  18,  46,  30,  33,  17,  45,  29, },
66 { 10,  58,   6,  54,   9,  57,   5,  53, },
67 { 42,  26,  38,  22,  41,  25,  37,  21, },
68 };
69
70 static const uint8_t offset[127][2]= {
71 {0,0},
72 {0,0}, {4,4},
73 {0,0}, {2,2}, {6,4}, {4,6},
74 {0,0}, {5,1}, {2,2}, {7,3}, {4,4}, {1,5}, {6,6}, {3,7},
75
76 {0,0}, {4,0}, {1,1}, {5,1}, {3,2}, {7,2}, {2,3}, {6,3},
77 {0,4}, {4,4}, {1,5}, {5,5}, {3,6}, {7,6}, {2,7}, {6,7},
78
79 {0,0}, {0,2}, {0,4}, {0,6}, {1,1}, {1,3}, {1,5}, {1,7},
80 {2,0}, {2,2}, {2,4}, {2,6}, {3,1}, {3,3}, {3,5}, {3,7},
81 {4,0}, {4,2}, {4,4}, {4,6}, {5,1}, {5,3}, {5,5}, {5,7},
82 {6,0}, {6,2}, {6,4}, {6,6}, {7,1}, {7,3}, {7,5}, {7,7},
83
84 {0,0}, {4,4}, {0,4}, {4,0}, {2,2}, {6,6}, {2,6}, {6,2},
85 {0,2}, {4,6}, {0,6}, {4,2}, {2,0}, {6,4}, {2,4}, {6,0},
86 {1,1}, {5,5}, {1,5}, {5,1}, {3,3}, {7,7}, {3,7}, {7,3},
87 {1,3}, {5,7}, {1,7}, {5,3}, {3,1}, {7,5}, {3,5}, {7,1},
88 {0,1}, {4,5}, {0,5}, {4,1}, {2,3}, {6,7}, {2,7}, {6,3},
89 {0,3}, {4,7}, {0,7}, {4,3}, {2,1}, {6,5}, {2,5}, {6,1},
90 {1,0}, {5,4}, {1,4}, {5,0}, {3,2}, {7,6}, {3,6}, {7,2},
91 {1,2}, {5,6}, {1,6}, {5,2}, {3,0}, {7,4}, {3,4}, {7,0},
92 };
93
94 struct vf_priv_s {
95         int log2_count;
96         int qp;
97         int mode;
98         int mpeg2;
99         int temp_stride;
100         uint8_t *src;
101         int16_t *temp;
102         AVCodecContext *avctx;
103         DSPContext dsp;
104         char *non_b_qp;
105 };
106
107 #define SHIFT 22
108
109 static void hardthresh_c(int16_t dst[64], int16_t src[64], int qp, uint8_t *permutation){
110         int i;
111         int bias= 0; //FIXME
112         unsigned int threshold1, threshold2;
113
114         threshold1= qp*((1<<4) - bias) - 1;
115         threshold2= (threshold1<<1);
116
117         memset(dst, 0, 64*sizeof(int16_t));
118         dst[0]= (src[0] + 4)>>3;
119
120         for(i=1; i<64; i++){
121                 int level= src[i];
122                 if(((unsigned)(level+threshold1))>threshold2){
123                         const int j= permutation[i];
124                         dst[j]= (level + 4)>>3;
125                 }
126         }
127 }
128
129 static void softthresh_c(int16_t dst[64], int16_t src[64], int qp, uint8_t *permutation){
130         int i;
131         int bias= 0; //FIXME
132         unsigned int threshold1, threshold2;
133
134         threshold1= qp*((1<<4) - bias) - 1;
135         threshold2= (threshold1<<1);
136
137         memset(dst, 0, 64*sizeof(int16_t));
138         dst[0]= (src[0] + 4)>>3;
139
140         for(i=1; i<64; i++){
141                 int level= src[i];
142                 if(((unsigned)(level+threshold1))>threshold2){
143                         const int j= permutation[i];
144                         if(level>0)
145                                 dst[j]= (level - threshold1 + 4)>>3;
146                         else
147                                 dst[j]= (level + threshold1 + 4)>>3;
148                 }
149         }
150 }
151
152 #if HAVE_MMX
153 static void hardthresh_mmx(int16_t dst[64], int16_t src[64], int qp, uint8_t *permutation){
154         int bias= 0; //FIXME
155         unsigned int threshold1;
156
157         threshold1= qp*((1<<4) - bias) - 1;
158
159         __asm__ volatile(
160 #define REQUANT_CORE(dst0, dst1, dst2, dst3, src0, src1, src2, src3) \
161                 "movq " #src0 ", %%mm0        \n\t"\
162                 "movq " #src1 ", %%mm1        \n\t"\
163                 "movq " #src2 ", %%mm2        \n\t"\
164                 "movq " #src3 ", %%mm3        \n\t"\
165                 "psubw %%mm4, %%mm0        \n\t"\
166                 "psubw %%mm4, %%mm1        \n\t"\
167                 "psubw %%mm4, %%mm2        \n\t"\
168                 "psubw %%mm4, %%mm3        \n\t"\
169                 "paddusw %%mm5, %%mm0        \n\t"\
170                 "paddusw %%mm5, %%mm1        \n\t"\
171                 "paddusw %%mm5, %%mm2        \n\t"\
172                 "paddusw %%mm5, %%mm3        \n\t"\
173                 "paddw %%mm6, %%mm0        \n\t"\
174                 "paddw %%mm6, %%mm1        \n\t"\
175                 "paddw %%mm6, %%mm2        \n\t"\
176                 "paddw %%mm6, %%mm3        \n\t"\
177                 "psubusw %%mm6, %%mm0        \n\t"\
178                 "psubusw %%mm6, %%mm1        \n\t"\
179                 "psubusw %%mm6, %%mm2        \n\t"\
180                 "psubusw %%mm6, %%mm3        \n\t"\
181                 "psraw $3, %%mm0        \n\t"\
182                 "psraw $3, %%mm1        \n\t"\
183                 "psraw $3, %%mm2        \n\t"\
184                 "psraw $3, %%mm3        \n\t"\
185 \
186                 "movq %%mm0, %%mm7        \n\t"\
187                 "punpcklwd %%mm2, %%mm0        \n\t" /*A*/\
188                 "punpckhwd %%mm2, %%mm7        \n\t" /*C*/\
189                 "movq %%mm1, %%mm2        \n\t"\
190                 "punpcklwd %%mm3, %%mm1        \n\t" /*B*/\
191                 "punpckhwd %%mm3, %%mm2        \n\t" /*D*/\
192                 "movq %%mm0, %%mm3        \n\t"\
193                 "punpcklwd %%mm1, %%mm0        \n\t" /*A*/\
194                 "punpckhwd %%mm7, %%mm3        \n\t" /*C*/\
195                 "punpcklwd %%mm2, %%mm7        \n\t" /*B*/\
196                 "punpckhwd %%mm2, %%mm1        \n\t" /*D*/\
197 \
198                 "movq %%mm0, " #dst0 "        \n\t"\
199                 "movq %%mm7, " #dst1 "        \n\t"\
200                 "movq %%mm3, " #dst2 "        \n\t"\
201                 "movq %%mm1, " #dst3 "        \n\t"
202
203                 "movd %2, %%mm4                \n\t"
204                 "movd %3, %%mm5                \n\t"
205                 "movd %4, %%mm6                \n\t"
206                 "packssdw %%mm4, %%mm4        \n\t"
207                 "packssdw %%mm5, %%mm5        \n\t"
208                 "packssdw %%mm6, %%mm6        \n\t"
209                 "packssdw %%mm4, %%mm4        \n\t"
210                 "packssdw %%mm5, %%mm5        \n\t"
211                 "packssdw %%mm6, %%mm6        \n\t"
212                 REQUANT_CORE(  (%1),  8(%1), 16(%1), 24(%1),  (%0), 8(%0), 64(%0), 72(%0))
213                 REQUANT_CORE(32(%1), 40(%1), 48(%1), 56(%1),16(%0),24(%0), 48(%0), 56(%0))
214                 REQUANT_CORE(64(%1), 72(%1), 80(%1), 88(%1),32(%0),40(%0), 96(%0),104(%0))
215                 REQUANT_CORE(96(%1),104(%1),112(%1),120(%1),80(%0),88(%0),112(%0),120(%0))
216                 : : "r" (src), "r" (dst), "g" (threshold1+1), "g" (threshold1+5), "g" (threshold1-4) //FIXME maybe more accurate then needed?
217         );
218         dst[0]= (src[0] + 4)>>3;
219 }
220
221 static void softthresh_mmx(int16_t dst[64], int16_t src[64], int qp, uint8_t *permutation){
222         int bias= 0; //FIXME
223         unsigned int threshold1;
224
225         threshold1= qp*((1<<4) - bias) - 1;
226
227         __asm__ volatile(
228 #undef REQUANT_CORE
229 #define REQUANT_CORE(dst0, dst1, dst2, dst3, src0, src1, src2, src3) \
230                 "movq " #src0 ", %%mm0        \n\t"\
231                 "movq " #src1 ", %%mm1        \n\t"\
232                 "pxor %%mm6, %%mm6        \n\t"\
233                 "pxor %%mm7, %%mm7        \n\t"\
234                 "pcmpgtw %%mm0, %%mm6        \n\t"\
235                 "pcmpgtw %%mm1, %%mm7        \n\t"\
236                 "pxor %%mm6, %%mm0        \n\t"\
237                 "pxor %%mm7, %%mm1        \n\t"\
238                 "psubusw %%mm4, %%mm0        \n\t"\
239                 "psubusw %%mm4, %%mm1        \n\t"\
240                 "pxor %%mm6, %%mm0        \n\t"\
241                 "pxor %%mm7, %%mm1        \n\t"\
242                 "movq " #src2 ", %%mm2        \n\t"\
243                 "movq " #src3 ", %%mm3        \n\t"\
244                 "pxor %%mm6, %%mm6        \n\t"\
245                 "pxor %%mm7, %%mm7        \n\t"\
246                 "pcmpgtw %%mm2, %%mm6        \n\t"\
247                 "pcmpgtw %%mm3, %%mm7        \n\t"\
248                 "pxor %%mm6, %%mm2        \n\t"\
249                 "pxor %%mm7, %%mm3        \n\t"\
250                 "psubusw %%mm4, %%mm2        \n\t"\
251                 "psubusw %%mm4, %%mm3        \n\t"\
252                 "pxor %%mm6, %%mm2        \n\t"\
253                 "pxor %%mm7, %%mm3        \n\t"\
254 \
255                 "paddsw %%mm5, %%mm0        \n\t"\
256                 "paddsw %%mm5, %%mm1        \n\t"\
257                 "paddsw %%mm5, %%mm2        \n\t"\
258                 "paddsw %%mm5, %%mm3        \n\t"\
259                 "psraw $3, %%mm0        \n\t"\
260                 "psraw $3, %%mm1        \n\t"\
261                 "psraw $3, %%mm2        \n\t"\
262                 "psraw $3, %%mm3        \n\t"\
263 \
264                 "movq %%mm0, %%mm7        \n\t"\
265                 "punpcklwd %%mm2, %%mm0        \n\t" /*A*/\
266                 "punpckhwd %%mm2, %%mm7        \n\t" /*C*/\
267                 "movq %%mm1, %%mm2        \n\t"\
268                 "punpcklwd %%mm3, %%mm1        \n\t" /*B*/\
269                 "punpckhwd %%mm3, %%mm2        \n\t" /*D*/\
270                 "movq %%mm0, %%mm3        \n\t"\
271                 "punpcklwd %%mm1, %%mm0        \n\t" /*A*/\
272                 "punpckhwd %%mm7, %%mm3        \n\t" /*C*/\
273                 "punpcklwd %%mm2, %%mm7        \n\t" /*B*/\
274                 "punpckhwd %%mm2, %%mm1        \n\t" /*D*/\
275 \
276                 "movq %%mm0, " #dst0 "        \n\t"\
277                 "movq %%mm7, " #dst1 "        \n\t"\
278                 "movq %%mm3, " #dst2 "        \n\t"\
279                 "movq %%mm1, " #dst3 "        \n\t"
280
281                 "movd %2, %%mm4                \n\t"
282                 "movd %3, %%mm5                \n\t"
283                 "packssdw %%mm4, %%mm4        \n\t"
284                 "packssdw %%mm5, %%mm5        \n\t"
285                 "packssdw %%mm4, %%mm4        \n\t"
286                 "packssdw %%mm5, %%mm5        \n\t"
287                 REQUANT_CORE(  (%1),  8(%1), 16(%1), 24(%1),  (%0), 8(%0), 64(%0), 72(%0))
288                 REQUANT_CORE(32(%1), 40(%1), 48(%1), 56(%1),16(%0),24(%0), 48(%0), 56(%0))
289                 REQUANT_CORE(64(%1), 72(%1), 80(%1), 88(%1),32(%0),40(%0), 96(%0),104(%0))
290                 REQUANT_CORE(96(%1),104(%1),112(%1),120(%1),80(%0),88(%0),112(%0),120(%0))
291                 : : "r" (src), "r" (dst), "g" (threshold1), "rm" (4) //FIXME maybe more accurate then needed?
292         );
293
294         dst[0]= (src[0] + 4)>>3;
295 }
296 #endif
297
298 static inline void add_block(int16_t *dst, int stride, int16_t block[64]){
299         int y;
300
301         for(y=0; y<8; y++){
302                 *(uint32_t*)&dst[0 + y*stride]+= *(uint32_t*)&block[0 + y*8];
303                 *(uint32_t*)&dst[2 + y*stride]+= *(uint32_t*)&block[2 + y*8];
304                 *(uint32_t*)&dst[4 + y*stride]+= *(uint32_t*)&block[4 + y*8];
305                 *(uint32_t*)&dst[6 + y*stride]+= *(uint32_t*)&block[6 + y*8];
306         }
307 }
308
309 static void store_slice_c(uint8_t *dst, int16_t *src, int dst_stride, int src_stride, int width, int height, int log2_scale){
310         int y, x;
311
312 #define STORE(pos) \
313         temp= ((src[x + y*src_stride + pos]<<log2_scale) + d[pos])>>6;\
314         if(temp & 0x100) temp= ~(temp>>31);\
315         dst[x + y*dst_stride + pos]= temp;
316
317         for(y=0; y<height; y++){
318                 const uint8_t *d= dither[y];
319                 for(x=0; x<width; x+=8){
320                         int temp;
321                         STORE(0);
322                         STORE(1);
323                         STORE(2);
324                         STORE(3);
325                         STORE(4);
326                         STORE(5);
327                         STORE(6);
328                         STORE(7);
329                 }
330         }
331 }
332
333 #if HAVE_MMX
334 static void store_slice_mmx(uint8_t *dst, int16_t *src, int dst_stride, int src_stride, int width, int height, int log2_scale){
335         int y;
336
337         for(y=0; y<height; y++){
338                 uint8_t *dst1= dst;
339                 int16_t *src1= src;
340                 __asm__ volatile(
341                         "movq (%3), %%mm3        \n\t"
342                         "movq (%3), %%mm4        \n\t"
343                         "movd %4, %%mm2                \n\t"
344                         "pxor %%mm0, %%mm0        \n\t"
345                         "punpcklbw %%mm0, %%mm3        \n\t"
346                         "punpckhbw %%mm0, %%mm4        \n\t"
347                         "psraw %%mm2, %%mm3        \n\t"
348                         "psraw %%mm2, %%mm4        \n\t"
349                         "movd %5, %%mm2                \n\t"
350                         "1:                        \n\t"
351                         "movq (%0), %%mm0        \n\t"
352                         "movq 8(%0), %%mm1        \n\t"
353                         "paddw %%mm3, %%mm0        \n\t"
354                         "paddw %%mm4, %%mm1        \n\t"
355                         "psraw %%mm2, %%mm0        \n\t"
356                         "psraw %%mm2, %%mm1        \n\t"
357                         "packuswb %%mm1, %%mm0        \n\t"
358                         "movq %%mm0, (%1)         \n\t"
359                         "add $16, %0                \n\t"
360                         "add $8, %1                \n\t"
361                         "cmp %2, %1                \n\t"
362                         " jb 1b                        \n\t"
363                         : "+r" (src1), "+r"(dst1)
364                         : "r"(dst + width), "r"(dither[y]), "g"(log2_scale), "g"(6-log2_scale)
365                 );
366                 src += src_stride;
367                 dst += dst_stride;
368         }
369 //        if(width != mmxw)
370 //                store_slice_c(dst + mmxw, src + mmxw, dst_stride, src_stride, width - mmxw, log2_scale);
371 }
372 #endif
373
374 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;
375
376 static void (*requantize)(int16_t dst[64], int16_t src[64], int qp, uint8_t *permutation)= hardthresh_c;
377
378 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){
379         int x, y, i;
380         const int count= 1<<p->log2_count;
381         const int stride= is_luma ? p->temp_stride : ((width+16+15)&(~15));
382         uint64_t __attribute__((aligned(16))) block_align[32];
383         int16_t *block = (int16_t *)block_align;
384         int16_t *block2= (int16_t *)(block_align+16);
385
386         if (!src || !dst) return; // HACK avoid crash for Y8 colourspace
387         for(y=0; y<height; y++){
388                 int index= 8 + 8*stride + y*stride;
389                 fast_memcpy(p->src + index, src + y*src_stride, width);
390                 for(x=0; x<8; x++){
391                         p->src[index         - x - 1]= p->src[index +         x    ];
392                         p->src[index + width + x    ]= p->src[index + width - x - 1];
393                 }
394         }
395         for(y=0; y<8; y++){
396                 fast_memcpy(p->src + (      7-y)*stride, p->src + (      y+8)*stride, stride);
397                 fast_memcpy(p->src + (height+8+y)*stride, p->src + (height-y+7)*stride, stride);
398         }
399         //FIXME (try edge emu)
400
401         for(y=0; y<height+8; y+=8){
402                 memset(p->temp + (8+y)*stride, 0, 8*stride*sizeof(int16_t));
403                 for(x=0; x<width+8; x+=8){
404                         const int qps= 3 + is_luma;
405                         int qp;
406
407                         if(p->qp)
408                                 qp= p->qp;
409                         else{
410                                 qp= qp_store[ (XMIN(x, width-1)>>qps) + (XMIN(y, height-1)>>qps) * qp_stride];
411                                 qp = FFMAX(1, norm_qscale(qp, p->mpeg2));
412                         }
413                         for(i=0; i<count; i++){
414                                 const int x1= x + offset[i+count-1][0];
415                                 const int y1= y + offset[i+count-1][1];
416                                 const int index= x1 + y1*stride;
417                                 p->dsp.get_pixels(block, p->src + index, stride);
418                                 p->dsp.fdct(block);
419                                 requantize(block2, block, qp, p->dsp.idct_permutation);
420                                 p->dsp.idct(block2);
421                                 add_block(p->temp + index, stride, block2);
422                         }
423                 }
424                 if(y)
425                         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);
426         }
427 #if 0
428         for(y=0; y<height; y++){
429                 for(x=0; x<width; x++){
430                         if((((x>>6) ^ (y>>6)) & 1) == 0)
431                                 dst[x + y*dst_stride]= p->src[8 + 8*stride  + x + y*stride];
432                         if((x&63) == 0 || (y&63)==0)
433                                 dst[x + y*dst_stride] += 128;
434                 }
435         }
436 #endif
437         //FIXME reorder for better caching
438 }
439
440 static int config(struct vf_instance *vf,
441         int width, int height, int d_width, int d_height,
442         unsigned int flags, unsigned int outfmt){
443         int h= (height+16+15)&(~15);
444
445         vf->priv->temp_stride= (width+16+15)&(~15);
446         vf->priv->temp= malloc(vf->priv->temp_stride*h*sizeof(int16_t));
447         vf->priv->src = malloc(vf->priv->temp_stride*h*sizeof(uint8_t));
448
449         return ff_vf_next_config(vf,width,height,d_width,d_height,flags,outfmt);
450 }
451
452 static void get_image(struct vf_instance *vf, mp_image_t *mpi){
453     if(mpi->flags&MP_IMGFLAG_PRESERVE) return; // don't change
454     // ok, we can do pp in-place (or pp disabled):
455     vf->dmpi=ff_vf_get_image(vf->next,mpi->imgfmt,
456         mpi->type, mpi->flags | MP_IMGFLAG_READABLE, mpi->width, mpi->height);
457     mpi->planes[0]=vf->dmpi->planes[0];
458     mpi->stride[0]=vf->dmpi->stride[0];
459     mpi->width=vf->dmpi->width;
460     if(mpi->flags&MP_IMGFLAG_PLANAR){
461         mpi->planes[1]=vf->dmpi->planes[1];
462         mpi->planes[2]=vf->dmpi->planes[2];
463         mpi->stride[1]=vf->dmpi->stride[1];
464         mpi->stride[2]=vf->dmpi->stride[2];
465     }
466     mpi->flags|=MP_IMGFLAG_DIRECT;
467 }
468
469 static int put_image(struct vf_instance *vf, mp_image_t *mpi, double pts){
470         mp_image_t *dmpi;
471
472         if(!(mpi->flags&MP_IMGFLAG_DIRECT)){
473                 // no DR, so get a new image! hope we'll get DR buffer:
474                 dmpi=ff_vf_get_image(vf->next,mpi->imgfmt,
475                     MP_IMGTYPE_TEMP,
476                     MP_IMGFLAG_ACCEPT_STRIDE|MP_IMGFLAG_PREFER_ALIGNED_STRIDE,
477                     mpi->width,mpi->height);
478                 ff_vf_clone_mpi_attributes(dmpi, mpi);
479         }else{
480            dmpi=vf->dmpi;
481         }
482
483         vf->priv->mpeg2= mpi->qscale_type;
484         if(mpi->pict_type != 3 && mpi->qscale && !vf->priv->qp){
485             int w = mpi->qstride;
486             int h = (mpi->h + 15) >> 4;
487             if (!w) {
488                 w = (mpi->w + 15) >> 4;
489                 h = 1;
490             }
491             if(!vf->priv->non_b_qp)
492                 vf->priv->non_b_qp= malloc(w*h);
493             fast_memcpy(vf->priv->non_b_qp, mpi->qscale, w*h);
494         }
495         if(vf->priv->log2_count || !(mpi->flags&MP_IMGFLAG_DIRECT)){
496             char *qp_tab= vf->priv->non_b_qp;
497             if((vf->priv->mode&4) || !qp_tab)
498                 qp_tab= mpi->qscale;
499
500             if(qp_tab || vf->priv->qp){
501                 filter(vf->priv, dmpi->planes[0], mpi->planes[0], dmpi->stride[0], mpi->stride[0], mpi->w, mpi->h, qp_tab, mpi->qstride, 1);
502                 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);
503                 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);
504             }else{
505                 memcpy_pic(dmpi->planes[0], mpi->planes[0], mpi->w, mpi->h, dmpi->stride[0], mpi->stride[0]);
506                 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]);
507                 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]);
508             }
509         }
510
511 #if HAVE_MMX
512         if(ff_gCpuCaps.hasMMX) __asm__ volatile ("emms\n\t");
513 #endif
514 #if HAVE_MMX2
515         if(ff_gCpuCaps.hasMMX2) __asm__ volatile ("sfence\n\t");
516 #endif
517
518         return ff_vf_next_put_image(vf,dmpi, pts);
519 }
520
521 static void uninit(struct vf_instance *vf){
522         if(!vf->priv) return;
523
524         free(vf->priv->temp);
525         vf->priv->temp= NULL;
526         free(vf->priv->src);
527         vf->priv->src= NULL;
528         free(vf->priv->avctx);
529         vf->priv->avctx= NULL;
530         free(vf->priv->non_b_qp);
531         vf->priv->non_b_qp= NULL;
532
533         free(vf->priv);
534         vf->priv=NULL;
535 }
536
537 //===========================================================================//
538 static int query_format(struct vf_instance *vf, unsigned int fmt){
539     switch(fmt){
540         case IMGFMT_YVU9:
541         case IMGFMT_IF09:
542         case IMGFMT_YV12:
543         case IMGFMT_I420:
544         case IMGFMT_IYUV:
545         case IMGFMT_CLPL:
546         case IMGFMT_Y800:
547         case IMGFMT_Y8:
548         case IMGFMT_444P:
549         case IMGFMT_422P:
550         case IMGFMT_411P:
551             return ff_vf_next_query_format(vf,fmt);
552     }
553     return 0;
554 }
555
556 static int control(struct vf_instance *vf, int request, void* data){
557     switch(request){
558     case VFCTRL_QUERY_MAX_PP_LEVEL:
559         return 6;
560     case VFCTRL_SET_PP_LEVEL:
561         vf->priv->log2_count= *((unsigned int*)data);
562         return CONTROL_TRUE;
563     }
564     return ff_vf_next_control(vf,request,data);
565 }
566
567 static int vf_open(vf_instance_t *vf, char *args){
568
569     int log2c=-1;
570
571     vf->config=config;
572     vf->put_image=put_image;
573     vf->get_image=get_image;
574     vf->query_format=query_format;
575     vf->uninit=uninit;
576     vf->control= control;
577     vf->priv=malloc(sizeof(struct vf_priv_s));
578     memset(vf->priv, 0, sizeof(struct vf_priv_s));
579
580     ff_init_avcodec();
581
582     vf->priv->avctx= avcodec_alloc_context3(NULL);
583     ff_dsputil_init(&vf->priv->dsp, vf->priv->avctx);
584
585     vf->priv->log2_count= 3;
586
587     if (args) sscanf(args, "%d:%d:%d", &log2c, &vf->priv->qp, &vf->priv->mode);
588
589     if( log2c >=0 && log2c <=6 )
590         vf->priv->log2_count = log2c;
591
592     if(vf->priv->qp < 0)
593         vf->priv->qp = 0;
594
595     switch(vf->priv->mode&3){
596         default:
597         case 0: requantize= hardthresh_c; break;
598         case 1: requantize= softthresh_c; break;
599     }
600
601 #if HAVE_MMX
602     if(ff_gCpuCaps.hasMMX){
603         store_slice= store_slice_mmx;
604         switch(vf->priv->mode&3){
605             case 0: requantize= hardthresh_mmx; break;
606             case 1: requantize= softthresh_mmx; break;
607         }
608     }
609 #endif
610
611     return 1;
612 }
613
614 const vf_info_t ff_vf_info_spp = {
615     "simple postprocess",
616     "spp",
617     "Michael Niedermayer",
618     "",
619     vf_open,
620     NULL
621 };