Factorize some code into the new function ff_toupper4().
[ffmpeg.git] / libavcodec / mpegvideo.c
1 /*
2  * The simplest mpeg encoder (well, it was the simplest!)
3  * Copyright (c) 2000,2001 Fabrice Bellard
4  * Copyright (c) 2002-2004 Michael Niedermayer <michaelni@gmx.at>
5  *
6  * 4MV & hq & B-frame encoding stuff by Michael Niedermayer <michaelni@gmx.at>
7  *
8  * This file is part of FFmpeg.
9  *
10  * FFmpeg is free software; you can redistribute it and/or
11  * modify it under the terms of the GNU Lesser General Public
12  * License as published by the Free Software Foundation; either
13  * version 2.1 of the License, or (at your option) any later version.
14  *
15  * FFmpeg is distributed in the hope that it will be useful,
16  * but WITHOUT ANY WARRANTY; without even the implied warranty of
17  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
18  * Lesser General Public License for more details.
19  *
20  * You should have received a copy of the GNU Lesser General Public
21  * License along with FFmpeg; if not, write to the Free Software
22  * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
23  */
24
25 /**
26  * @file
27  * The simplest mpeg encoder (well, it was the simplest!).
28  */
29
30 #include "libavutil/intmath.h"
31 #include "avcodec.h"
32 #include "dsputil.h"
33 #include "internal.h"
34 #include "mpegvideo.h"
35 #include "mpegvideo_common.h"
36 #include "mjpegenc.h"
37 #include "msmpeg4.h"
38 #include "faandct.h"
39 #include "xvmc_internal.h"
40 #include <limits.h>
41
42 //#undef NDEBUG
43 //#include <assert.h>
44
45 static void dct_unquantize_mpeg1_intra_c(MpegEncContext *s,
46                                    DCTELEM *block, int n, int qscale);
47 static void dct_unquantize_mpeg1_inter_c(MpegEncContext *s,
48                                    DCTELEM *block, int n, int qscale);
49 static void dct_unquantize_mpeg2_intra_c(MpegEncContext *s,
50                                    DCTELEM *block, int n, int qscale);
51 static void dct_unquantize_mpeg2_intra_bitexact(MpegEncContext *s,
52                                    DCTELEM *block, int n, int qscale);
53 static void dct_unquantize_mpeg2_inter_c(MpegEncContext *s,
54                                    DCTELEM *block, int n, int qscale);
55 static void dct_unquantize_h263_intra_c(MpegEncContext *s,
56                                   DCTELEM *block, int n, int qscale);
57 static void dct_unquantize_h263_inter_c(MpegEncContext *s,
58                                   DCTELEM *block, int n, int qscale);
59
60
61 /* enable all paranoid tests for rounding, overflows, etc... */
62 //#define PARANOID
63
64 //#define DEBUG
65
66
67 static const uint8_t ff_default_chroma_qscale_table[32]={
68 //  0  1  2  3  4  5  6  7  8  9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31
69     0, 1, 2, 3, 4, 5, 6, 7, 8, 9,10,11,12,13,14,15,16,17,18,19,20,21,22,23,24,25,26,27,28,29,30,31
70 };
71
72 const uint8_t ff_mpeg1_dc_scale_table[128]={
73 //  0  1  2  3  4  5  6  7  8  9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31
74     8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8,
75     8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8,
76     8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8,
77     8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8,
78 };
79
80 static const uint8_t mpeg2_dc_scale_table1[128]={
81 //  0  1  2  3  4  5  6  7  8  9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31
82     4, 4, 4, 4, 4, 4, 4, 4, 4, 4, 4, 4, 4, 4, 4, 4, 4, 4, 4, 4, 4, 4, 4, 4, 4, 4, 4, 4, 4, 4, 4, 4,
83     4, 4, 4, 4, 4, 4, 4, 4, 4, 4, 4, 4, 4, 4, 4, 4, 4, 4, 4, 4, 4, 4, 4, 4, 4, 4, 4, 4, 4, 4, 4, 4,
84     4, 4, 4, 4, 4, 4, 4, 4, 4, 4, 4, 4, 4, 4, 4, 4, 4, 4, 4, 4, 4, 4, 4, 4, 4, 4, 4, 4, 4, 4, 4, 4,
85     4, 4, 4, 4, 4, 4, 4, 4, 4, 4, 4, 4, 4, 4, 4, 4, 4, 4, 4, 4, 4, 4, 4, 4, 4, 4, 4, 4, 4, 4, 4, 4,
86 };
87
88 static const uint8_t mpeg2_dc_scale_table2[128]={
89 //  0  1  2  3  4  5  6  7  8  9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31
90     2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2,
91     2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2,
92     2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2,
93     2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2,
94 };
95
96 static const uint8_t mpeg2_dc_scale_table3[128]={
97 //  0  1  2  3  4  5  6  7  8  9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31
98     1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1,
99     1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1,
100     1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1,
101     1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1,
102 };
103
104 const uint8_t * const ff_mpeg2_dc_scale_table[4]={
105     ff_mpeg1_dc_scale_table,
106     mpeg2_dc_scale_table1,
107     mpeg2_dc_scale_table2,
108     mpeg2_dc_scale_table3,
109 };
110
111 const enum PixelFormat ff_pixfmt_list_420[] = {
112     PIX_FMT_YUV420P,
113     PIX_FMT_NONE
114 };
115
116 const enum PixelFormat ff_hwaccel_pixfmt_list_420[] = {
117     PIX_FMT_DXVA2_VLD,
118     PIX_FMT_VAAPI_VLD,
119     PIX_FMT_YUV420P,
120     PIX_FMT_NONE
121 };
122
123 const uint8_t *ff_find_start_code(const uint8_t * restrict p, const uint8_t *end, uint32_t * restrict state){
124     int i;
125
126     assert(p<=end);
127     if(p>=end)
128         return end;
129
130     for(i=0; i<3; i++){
131         uint32_t tmp= *state << 8;
132         *state= tmp + *(p++);
133         if(tmp == 0x100 || p==end)
134             return p;
135     }
136
137     while(p<end){
138         if     (p[-1] > 1      ) p+= 3;
139         else if(p[-2]          ) p+= 2;
140         else if(p[-3]|(p[-1]-1)) p++;
141         else{
142             p++;
143             break;
144         }
145     }
146
147     p= FFMIN(p, end)-4;
148     *state= AV_RB32(p);
149
150     return p+4;
151 }
152
153 /* init common dct for both encoder and decoder */
154 av_cold int ff_dct_common_init(MpegEncContext *s)
155 {
156     s->dct_unquantize_h263_intra = dct_unquantize_h263_intra_c;
157     s->dct_unquantize_h263_inter = dct_unquantize_h263_inter_c;
158     s->dct_unquantize_mpeg1_intra = dct_unquantize_mpeg1_intra_c;
159     s->dct_unquantize_mpeg1_inter = dct_unquantize_mpeg1_inter_c;
160     s->dct_unquantize_mpeg2_intra = dct_unquantize_mpeg2_intra_c;
161     if(s->flags & CODEC_FLAG_BITEXACT)
162         s->dct_unquantize_mpeg2_intra = dct_unquantize_mpeg2_intra_bitexact;
163     s->dct_unquantize_mpeg2_inter = dct_unquantize_mpeg2_inter_c;
164
165 #if   HAVE_MMX
166     MPV_common_init_mmx(s);
167 #elif ARCH_ALPHA
168     MPV_common_init_axp(s);
169 #elif CONFIG_MLIB
170     MPV_common_init_mlib(s);
171 #elif HAVE_MMI
172     MPV_common_init_mmi(s);
173 #elif ARCH_ARM
174     MPV_common_init_arm(s);
175 #elif HAVE_ALTIVEC
176     MPV_common_init_altivec(s);
177 #elif ARCH_BFIN
178     MPV_common_init_bfin(s);
179 #endif
180
181     /* load & permutate scantables
182        note: only wmv uses different ones
183     */
184     if(s->alternate_scan){
185         ff_init_scantable(s->dsp.idct_permutation, &s->inter_scantable  , ff_alternate_vertical_scan);
186         ff_init_scantable(s->dsp.idct_permutation, &s->intra_scantable  , ff_alternate_vertical_scan);
187     }else{
188         ff_init_scantable(s->dsp.idct_permutation, &s->inter_scantable  , ff_zigzag_direct);
189         ff_init_scantable(s->dsp.idct_permutation, &s->intra_scantable  , ff_zigzag_direct);
190     }
191     ff_init_scantable(s->dsp.idct_permutation, &s->intra_h_scantable, ff_alternate_horizontal_scan);
192     ff_init_scantable(s->dsp.idct_permutation, &s->intra_v_scantable, ff_alternate_vertical_scan);
193
194     return 0;
195 }
196
197 void ff_copy_picture(Picture *dst, Picture *src){
198     *dst = *src;
199     dst->type= FF_BUFFER_TYPE_COPY;
200 }
201
202 /**
203  * Releases a frame buffer
204  */
205 static void free_frame_buffer(MpegEncContext *s, Picture *pic)
206 {
207     s->avctx->release_buffer(s->avctx, (AVFrame*)pic);
208     av_freep(&pic->hwaccel_picture_private);
209 }
210
211 /**
212  * Allocates a frame buffer
213  */
214 static int alloc_frame_buffer(MpegEncContext *s, Picture *pic)
215 {
216     int r;
217
218     if (s->avctx->hwaccel) {
219         assert(!pic->hwaccel_picture_private);
220         if (s->avctx->hwaccel->priv_data_size) {
221             pic->hwaccel_picture_private = av_mallocz(s->avctx->hwaccel->priv_data_size);
222             if (!pic->hwaccel_picture_private) {
223                 av_log(s->avctx, AV_LOG_ERROR, "alloc_frame_buffer() failed (hwaccel private data allocation)\n");
224                 return -1;
225             }
226         }
227     }
228
229     r = s->avctx->get_buffer(s->avctx, (AVFrame*)pic);
230
231     if (r<0 || !pic->age || !pic->type || !pic->data[0]) {
232         av_log(s->avctx, AV_LOG_ERROR, "get_buffer() failed (%d %d %d %p)\n", r, pic->age, pic->type, pic->data[0]);
233         av_freep(&pic->hwaccel_picture_private);
234         return -1;
235     }
236
237     if (s->linesize && (s->linesize != pic->linesize[0] || s->uvlinesize != pic->linesize[1])) {
238         av_log(s->avctx, AV_LOG_ERROR, "get_buffer() failed (stride changed)\n");
239         free_frame_buffer(s, pic);
240         return -1;
241     }
242
243     if (pic->linesize[1] != pic->linesize[2]) {
244         av_log(s->avctx, AV_LOG_ERROR, "get_buffer() failed (uv stride mismatch)\n");
245         free_frame_buffer(s, pic);
246         return -1;
247     }
248
249     return 0;
250 }
251
252 /**
253  * allocates a Picture
254  * The pixels are allocated/set by calling get_buffer() if shared=0
255  */
256 int ff_alloc_picture(MpegEncContext *s, Picture *pic, int shared){
257     const int big_mb_num= s->mb_stride*(s->mb_height+1) + 1; //the +1 is needed so memset(,,stride*height) does not sig11
258     const int mb_array_size= s->mb_stride*s->mb_height;
259     const int b8_array_size= s->b8_stride*s->mb_height*2;
260     const int b4_array_size= s->b4_stride*s->mb_height*4;
261     int i;
262     int r= -1;
263
264     if(shared){
265         assert(pic->data[0]);
266         assert(pic->type == 0 || pic->type == FF_BUFFER_TYPE_SHARED);
267         pic->type= FF_BUFFER_TYPE_SHARED;
268     }else{
269         assert(!pic->data[0]);
270
271         if (alloc_frame_buffer(s, pic) < 0)
272             return -1;
273
274         s->linesize  = pic->linesize[0];
275         s->uvlinesize= pic->linesize[1];
276     }
277
278     if(pic->qscale_table==NULL){
279         if (s->encoding) {
280             FF_ALLOCZ_OR_GOTO(s->avctx, pic->mb_var   , mb_array_size * sizeof(int16_t)  , fail)
281             FF_ALLOCZ_OR_GOTO(s->avctx, pic->mc_mb_var, mb_array_size * sizeof(int16_t)  , fail)
282             FF_ALLOCZ_OR_GOTO(s->avctx, pic->mb_mean  , mb_array_size * sizeof(int8_t )  , fail)
283         }
284
285         FF_ALLOCZ_OR_GOTO(s->avctx, pic->mbskip_table , mb_array_size * sizeof(uint8_t)+2, fail) //the +2 is for the slice end check
286         FF_ALLOCZ_OR_GOTO(s->avctx, pic->qscale_table , mb_array_size * sizeof(uint8_t)  , fail)
287         FF_ALLOCZ_OR_GOTO(s->avctx, pic->mb_type_base , (big_mb_num + s->mb_stride) * sizeof(uint32_t), fail)
288         pic->mb_type= pic->mb_type_base + 2*s->mb_stride+1;
289         if(s->out_format == FMT_H264){
290             for(i=0; i<2; i++){
291                 FF_ALLOCZ_OR_GOTO(s->avctx, pic->motion_val_base[i], 2 * (b4_array_size+4)  * sizeof(int16_t), fail)
292                 pic->motion_val[i]= pic->motion_val_base[i]+4;
293                 FF_ALLOCZ_OR_GOTO(s->avctx, pic->ref_index[i], 4*mb_array_size * sizeof(uint8_t), fail)
294             }
295             pic->motion_subsample_log2= 2;
296         }else if(s->out_format == FMT_H263 || s->encoding || (s->avctx->debug&FF_DEBUG_MV) || (s->avctx->debug_mv)){
297             for(i=0; i<2; i++){
298                 FF_ALLOCZ_OR_GOTO(s->avctx, pic->motion_val_base[i], 2 * (b8_array_size+4) * sizeof(int16_t), fail)
299                 pic->motion_val[i]= pic->motion_val_base[i]+4;
300                 FF_ALLOCZ_OR_GOTO(s->avctx, pic->ref_index[i], 4*mb_array_size * sizeof(uint8_t), fail)
301             }
302             pic->motion_subsample_log2= 3;
303         }
304         if(s->avctx->debug&FF_DEBUG_DCT_COEFF) {
305             FF_ALLOCZ_OR_GOTO(s->avctx, pic->dct_coeff, 64 * mb_array_size * sizeof(DCTELEM)*6, fail)
306         }
307         pic->qstride= s->mb_stride;
308         FF_ALLOCZ_OR_GOTO(s->avctx, pic->pan_scan , 1 * sizeof(AVPanScan), fail)
309     }
310
311     /* It might be nicer if the application would keep track of these
312      * but it would require an API change. */
313     memmove(s->prev_pict_types+1, s->prev_pict_types, PREV_PICT_TYPES_BUFFER_SIZE-1);
314     s->prev_pict_types[0]= s->dropable ? FF_B_TYPE : s->pict_type;
315     if(pic->age < PREV_PICT_TYPES_BUFFER_SIZE && s->prev_pict_types[pic->age] == FF_B_TYPE)
316         pic->age= INT_MAX; // Skipped MBs in B-frames are quite rare in MPEG-1/2 and it is a bit tricky to skip them anyway.
317
318     return 0;
319 fail: //for the FF_ALLOCZ_OR_GOTO macro
320     if(r>=0)
321         free_frame_buffer(s, pic);
322     return -1;
323 }
324
325 /**
326  * deallocates a picture
327  */
328 static void free_picture(MpegEncContext *s, Picture *pic){
329     int i;
330
331     if(pic->data[0] && pic->type!=FF_BUFFER_TYPE_SHARED){
332         free_frame_buffer(s, pic);
333     }
334
335     av_freep(&pic->mb_var);
336     av_freep(&pic->mc_mb_var);
337     av_freep(&pic->mb_mean);
338     av_freep(&pic->mbskip_table);
339     av_freep(&pic->qscale_table);
340     av_freep(&pic->mb_type_base);
341     av_freep(&pic->dct_coeff);
342     av_freep(&pic->pan_scan);
343     pic->mb_type= NULL;
344     for(i=0; i<2; i++){
345         av_freep(&pic->motion_val_base[i]);
346         av_freep(&pic->ref_index[i]);
347     }
348
349     if(pic->type == FF_BUFFER_TYPE_SHARED){
350         for(i=0; i<4; i++){
351             pic->base[i]=
352             pic->data[i]= NULL;
353         }
354         pic->type= 0;
355     }
356 }
357
358 static int init_duplicate_context(MpegEncContext *s, MpegEncContext *base){
359     int i;
360
361     // edge emu needs blocksize + filter length - 1 (=17x17 for halfpel / 21x21 for h264)
362     FF_ALLOCZ_OR_GOTO(s->avctx, s->allocated_edge_emu_buffer, (s->width+64)*2*21*2, fail); //(width + edge + align)*interlaced*MBsize*tolerance
363     s->edge_emu_buffer= s->allocated_edge_emu_buffer + (s->width+64)*2*21;
364
365      //FIXME should be linesize instead of s->width*2 but that is not known before get_buffer()
366     FF_ALLOCZ_OR_GOTO(s->avctx, s->me.scratchpad,  (s->width+64)*4*16*2*sizeof(uint8_t), fail)
367     s->me.temp=         s->me.scratchpad;
368     s->rd_scratchpad=   s->me.scratchpad;
369     s->b_scratchpad=    s->me.scratchpad;
370     s->obmc_scratchpad= s->me.scratchpad + 16;
371     if (s->encoding) {
372         FF_ALLOCZ_OR_GOTO(s->avctx, s->me.map      , ME_MAP_SIZE*sizeof(uint32_t), fail)
373         FF_ALLOCZ_OR_GOTO(s->avctx, s->me.score_map, ME_MAP_SIZE*sizeof(uint32_t), fail)
374         if(s->avctx->noise_reduction){
375             FF_ALLOCZ_OR_GOTO(s->avctx, s->dct_error_sum, 2 * 64 * sizeof(int), fail)
376         }
377     }
378     FF_ALLOCZ_OR_GOTO(s->avctx, s->blocks, 64*12*2 * sizeof(DCTELEM), fail)
379     s->block= s->blocks[0];
380
381     for(i=0;i<12;i++){
382         s->pblocks[i] = &s->block[i];
383     }
384     return 0;
385 fail:
386     return -1; //free() through MPV_common_end()
387 }
388
389 static void free_duplicate_context(MpegEncContext *s){
390     if(s==NULL) return;
391
392     av_freep(&s->allocated_edge_emu_buffer); s->edge_emu_buffer= NULL;
393     av_freep(&s->me.scratchpad);
394     s->me.temp=
395     s->rd_scratchpad=
396     s->b_scratchpad=
397     s->obmc_scratchpad= NULL;
398
399     av_freep(&s->dct_error_sum);
400     av_freep(&s->me.map);
401     av_freep(&s->me.score_map);
402     av_freep(&s->blocks);
403     s->block= NULL;
404 }
405
406 static void backup_duplicate_context(MpegEncContext *bak, MpegEncContext *src){
407 #define COPY(a) bak->a= src->a
408     COPY(allocated_edge_emu_buffer);
409     COPY(edge_emu_buffer);
410     COPY(me.scratchpad);
411     COPY(me.temp);
412     COPY(rd_scratchpad);
413     COPY(b_scratchpad);
414     COPY(obmc_scratchpad);
415     COPY(me.map);
416     COPY(me.score_map);
417     COPY(blocks);
418     COPY(block);
419     COPY(start_mb_y);
420     COPY(end_mb_y);
421     COPY(me.map_generation);
422     COPY(pb);
423     COPY(dct_error_sum);
424     COPY(dct_count[0]);
425     COPY(dct_count[1]);
426 #undef COPY
427 }
428
429 void ff_update_duplicate_context(MpegEncContext *dst, MpegEncContext *src){
430     MpegEncContext bak;
431     int i;
432     //FIXME copy only needed parts
433 //START_TIMER
434     backup_duplicate_context(&bak, dst);
435     memcpy(dst, src, sizeof(MpegEncContext));
436     backup_duplicate_context(dst, &bak);
437     for(i=0;i<12;i++){
438         dst->pblocks[i] = &dst->block[i];
439     }
440 //STOP_TIMER("update_duplicate_context") //about 10k cycles / 0.01 sec for 1000frames on 1ghz with 2 threads
441 }
442
443 /**
444  * sets the given MpegEncContext to common defaults (same for encoding and decoding).
445  * the changed fields will not depend upon the prior state of the MpegEncContext.
446  */
447 void MPV_common_defaults(MpegEncContext *s){
448     s->y_dc_scale_table=
449     s->c_dc_scale_table= ff_mpeg1_dc_scale_table;
450     s->chroma_qscale_table= ff_default_chroma_qscale_table;
451     s->progressive_frame= 1;
452     s->progressive_sequence= 1;
453     s->picture_structure= PICT_FRAME;
454
455     s->coded_picture_number = 0;
456     s->picture_number = 0;
457     s->input_picture_number = 0;
458
459     s->picture_in_gop_number = 0;
460
461     s->f_code = 1;
462     s->b_code = 1;
463 }
464
465 /**
466  * sets the given MpegEncContext to defaults for decoding.
467  * the changed fields will not depend upon the prior state of the MpegEncContext.
468  */
469 void MPV_decode_defaults(MpegEncContext *s){
470     MPV_common_defaults(s);
471 }
472
473 /**
474  * init common structure for both encoder and decoder.
475  * this assumes that some variables like width/height are already set
476  */
477 av_cold int MPV_common_init(MpegEncContext *s)
478 {
479     int y_size, c_size, yc_size, i, mb_array_size, mv_table_size, x, y, threads;
480
481     if(s->codec_id == CODEC_ID_MPEG2VIDEO && !s->progressive_sequence)
482         s->mb_height = (s->height + 31) / 32 * 2;
483     else
484         s->mb_height = (s->height + 15) / 16;
485
486     if(s->avctx->pix_fmt == PIX_FMT_NONE){
487         av_log(s->avctx, AV_LOG_ERROR, "decoding to PIX_FMT_NONE is not supported.\n");
488         return -1;
489     }
490
491     if(s->avctx->thread_count > MAX_THREADS || (s->avctx->thread_count > s->mb_height && s->mb_height)){
492         av_log(s->avctx, AV_LOG_ERROR, "too many threads\n");
493         return -1;
494     }
495
496     if((s->width || s->height) && avcodec_check_dimensions(s->avctx, s->width, s->height))
497         return -1;
498
499     dsputil_init(&s->dsp, s->avctx);
500     ff_dct_common_init(s);
501
502     s->flags= s->avctx->flags;
503     s->flags2= s->avctx->flags2;
504
505     s->mb_width  = (s->width  + 15) / 16;
506     s->mb_stride = s->mb_width + 1;
507     s->b8_stride = s->mb_width*2 + 1;
508     s->b4_stride = s->mb_width*4 + 1;
509     mb_array_size= s->mb_height * s->mb_stride;
510     mv_table_size= (s->mb_height+2) * s->mb_stride + 1;
511
512     /* set chroma shifts */
513     avcodec_get_chroma_sub_sample(s->avctx->pix_fmt,&(s->chroma_x_shift),
514                                                     &(s->chroma_y_shift) );
515
516     /* set default edge pos, will be overriden in decode_header if needed */
517     s->h_edge_pos= s->mb_width*16;
518     s->v_edge_pos= s->mb_height*16;
519
520     s->mb_num = s->mb_width * s->mb_height;
521
522     s->block_wrap[0]=
523     s->block_wrap[1]=
524     s->block_wrap[2]=
525     s->block_wrap[3]= s->b8_stride;
526     s->block_wrap[4]=
527     s->block_wrap[5]= s->mb_stride;
528
529     y_size = s->b8_stride * (2 * s->mb_height + 1);
530     c_size = s->mb_stride * (s->mb_height + 1);
531     yc_size = y_size + 2 * c_size;
532
533     /* convert fourcc to upper case */
534     s->codec_tag = ff_toupper4(s->avctx->codec_tag);
535
536     s->stream_codec_tag = ff_toupper4(s->avctx->stream_codec_tag);
537
538     s->avctx->coded_frame= (AVFrame*)&s->current_picture;
539
540     FF_ALLOCZ_OR_GOTO(s->avctx, s->mb_index2xy, (s->mb_num+1)*sizeof(int), fail) //error ressilience code looks cleaner with this
541     for(y=0; y<s->mb_height; y++){
542         for(x=0; x<s->mb_width; x++){
543             s->mb_index2xy[ x + y*s->mb_width ] = x + y*s->mb_stride;
544         }
545     }
546     s->mb_index2xy[ s->mb_height*s->mb_width ] = (s->mb_height-1)*s->mb_stride + s->mb_width; //FIXME really needed?
547
548     if (s->encoding) {
549         /* Allocate MV tables */
550         FF_ALLOCZ_OR_GOTO(s->avctx, s->p_mv_table_base            , mv_table_size * 2 * sizeof(int16_t), fail)
551         FF_ALLOCZ_OR_GOTO(s->avctx, s->b_forw_mv_table_base       , mv_table_size * 2 * sizeof(int16_t), fail)
552         FF_ALLOCZ_OR_GOTO(s->avctx, s->b_back_mv_table_base       , mv_table_size * 2 * sizeof(int16_t), fail)
553         FF_ALLOCZ_OR_GOTO(s->avctx, s->b_bidir_forw_mv_table_base , mv_table_size * 2 * sizeof(int16_t), fail)
554         FF_ALLOCZ_OR_GOTO(s->avctx, s->b_bidir_back_mv_table_base , mv_table_size * 2 * sizeof(int16_t), fail)
555         FF_ALLOCZ_OR_GOTO(s->avctx, s->b_direct_mv_table_base     , mv_table_size * 2 * sizeof(int16_t), fail)
556         s->p_mv_table           = s->p_mv_table_base            + s->mb_stride + 1;
557         s->b_forw_mv_table      = s->b_forw_mv_table_base       + s->mb_stride + 1;
558         s->b_back_mv_table      = s->b_back_mv_table_base       + s->mb_stride + 1;
559         s->b_bidir_forw_mv_table= s->b_bidir_forw_mv_table_base + s->mb_stride + 1;
560         s->b_bidir_back_mv_table= s->b_bidir_back_mv_table_base + s->mb_stride + 1;
561         s->b_direct_mv_table    = s->b_direct_mv_table_base     + s->mb_stride + 1;
562
563         if(s->msmpeg4_version){
564             FF_ALLOCZ_OR_GOTO(s->avctx, s->ac_stats, 2*2*(MAX_LEVEL+1)*(MAX_RUN+1)*2*sizeof(int), fail);
565         }
566         FF_ALLOCZ_OR_GOTO(s->avctx, s->avctx->stats_out, 256, fail);
567
568         /* Allocate MB type table */
569         FF_ALLOCZ_OR_GOTO(s->avctx, s->mb_type  , mb_array_size * sizeof(uint16_t), fail) //needed for encoding
570
571         FF_ALLOCZ_OR_GOTO(s->avctx, s->lambda_table, mb_array_size * sizeof(int), fail)
572
573         FF_ALLOCZ_OR_GOTO(s->avctx, s->q_intra_matrix  , 64*32   * sizeof(int), fail)
574         FF_ALLOCZ_OR_GOTO(s->avctx, s->q_inter_matrix  , 64*32   * sizeof(int), fail)
575         FF_ALLOCZ_OR_GOTO(s->avctx, s->q_intra_matrix16, 64*32*2 * sizeof(uint16_t), fail)
576         FF_ALLOCZ_OR_GOTO(s->avctx, s->q_inter_matrix16, 64*32*2 * sizeof(uint16_t), fail)
577         FF_ALLOCZ_OR_GOTO(s->avctx, s->input_picture, MAX_PICTURE_COUNT * sizeof(Picture*), fail)
578         FF_ALLOCZ_OR_GOTO(s->avctx, s->reordered_input_picture, MAX_PICTURE_COUNT * sizeof(Picture*), fail)
579
580         if(s->avctx->noise_reduction){
581             FF_ALLOCZ_OR_GOTO(s->avctx, s->dct_offset, 2 * 64 * sizeof(uint16_t), fail)
582         }
583     }
584     FF_ALLOCZ_OR_GOTO(s->avctx, s->picture, MAX_PICTURE_COUNT * sizeof(Picture), fail)
585     for(i = 0; i < MAX_PICTURE_COUNT; i++) {
586         avcodec_get_frame_defaults((AVFrame *)&s->picture[i]);
587     }
588
589     FF_ALLOCZ_OR_GOTO(s->avctx, s->error_status_table, mb_array_size*sizeof(uint8_t), fail)
590
591     if(s->codec_id==CODEC_ID_MPEG4 || (s->flags & CODEC_FLAG_INTERLACED_ME)){
592         /* interlaced direct mode decoding tables */
593             for(i=0; i<2; i++){
594                 int j, k;
595                 for(j=0; j<2; j++){
596                     for(k=0; k<2; k++){
597                         FF_ALLOCZ_OR_GOTO(s->avctx,    s->b_field_mv_table_base[i][j][k], mv_table_size * 2 * sizeof(int16_t), fail)
598                         s->b_field_mv_table[i][j][k] = s->b_field_mv_table_base[i][j][k] + s->mb_stride + 1;
599                     }
600                     FF_ALLOCZ_OR_GOTO(s->avctx, s->b_field_select_table [i][j], mb_array_size * 2 * sizeof(uint8_t), fail)
601                     FF_ALLOCZ_OR_GOTO(s->avctx, s->p_field_mv_table_base[i][j], mv_table_size * 2 * sizeof(int16_t), fail)
602                     s->p_field_mv_table[i][j] = s->p_field_mv_table_base[i][j]+ s->mb_stride + 1;
603                 }
604                 FF_ALLOCZ_OR_GOTO(s->avctx, s->p_field_select_table[i], mb_array_size * 2 * sizeof(uint8_t), fail)
605             }
606     }
607     if (s->out_format == FMT_H263) {
608         /* ac values */
609         FF_ALLOCZ_OR_GOTO(s->avctx, s->ac_val_base, yc_size * sizeof(int16_t) * 16, fail);
610         s->ac_val[0] = s->ac_val_base + s->b8_stride + 1;
611         s->ac_val[1] = s->ac_val_base + y_size + s->mb_stride + 1;
612         s->ac_val[2] = s->ac_val[1] + c_size;
613
614         /* cbp values */
615         FF_ALLOCZ_OR_GOTO(s->avctx, s->coded_block_base, y_size, fail);
616         s->coded_block= s->coded_block_base + s->b8_stride + 1;
617
618         /* cbp, ac_pred, pred_dir */
619         FF_ALLOCZ_OR_GOTO(s->avctx, s->cbp_table     , mb_array_size * sizeof(uint8_t), fail)
620         FF_ALLOCZ_OR_GOTO(s->avctx, s->pred_dir_table, mb_array_size * sizeof(uint8_t), fail)
621     }
622
623     if (s->h263_pred || s->h263_plus || !s->encoding) {
624         /* dc values */
625         //MN: we need these for error resilience of intra-frames
626         FF_ALLOCZ_OR_GOTO(s->avctx, s->dc_val_base, yc_size * sizeof(int16_t), fail);
627         s->dc_val[0] = s->dc_val_base + s->b8_stride + 1;
628         s->dc_val[1] = s->dc_val_base + y_size + s->mb_stride + 1;
629         s->dc_val[2] = s->dc_val[1] + c_size;
630         for(i=0;i<yc_size;i++)
631             s->dc_val_base[i] = 1024;
632     }
633
634     /* which mb is a intra block */
635     FF_ALLOCZ_OR_GOTO(s->avctx, s->mbintra_table, mb_array_size, fail);
636     memset(s->mbintra_table, 1, mb_array_size);
637
638     /* init macroblock skip table */
639     FF_ALLOCZ_OR_GOTO(s->avctx, s->mbskip_table, mb_array_size+2, fail);
640     //Note the +1 is for a quicker mpeg4 slice_end detection
641     FF_ALLOCZ_OR_GOTO(s->avctx, s->prev_pict_types, PREV_PICT_TYPES_BUFFER_SIZE, fail);
642
643     s->parse_context.state= -1;
644     if((s->avctx->debug&(FF_DEBUG_VIS_QP|FF_DEBUG_VIS_MB_TYPE)) || (s->avctx->debug_mv)){
645        s->visualization_buffer[0] = av_malloc((s->mb_width*16 + 2*EDGE_WIDTH) * s->mb_height*16 + 2*EDGE_WIDTH);
646        s->visualization_buffer[1] = av_malloc((s->mb_width*16 + 2*EDGE_WIDTH) * s->mb_height*16 + 2*EDGE_WIDTH);
647        s->visualization_buffer[2] = av_malloc((s->mb_width*16 + 2*EDGE_WIDTH) * s->mb_height*16 + 2*EDGE_WIDTH);
648     }
649
650     s->context_initialized = 1;
651
652     s->thread_context[0]= s;
653     threads = s->avctx->thread_count;
654
655     for(i=1; i<threads; i++){
656         s->thread_context[i]= av_malloc(sizeof(MpegEncContext));
657         memcpy(s->thread_context[i], s, sizeof(MpegEncContext));
658     }
659
660     for(i=0; i<threads; i++){
661         if(init_duplicate_context(s->thread_context[i], s) < 0)
662            goto fail;
663         s->thread_context[i]->start_mb_y= (s->mb_height*(i  ) + s->avctx->thread_count/2) / s->avctx->thread_count;
664         s->thread_context[i]->end_mb_y  = (s->mb_height*(i+1) + s->avctx->thread_count/2) / s->avctx->thread_count;
665     }
666
667     return 0;
668  fail:
669     MPV_common_end(s);
670     return -1;
671 }
672
673 /* init common structure for both encoder and decoder */
674 void MPV_common_end(MpegEncContext *s)
675 {
676     int i, j, k;
677
678     for(i=0; i<s->avctx->thread_count; i++){
679         free_duplicate_context(s->thread_context[i]);
680     }
681     for(i=1; i<s->avctx->thread_count; i++){
682         av_freep(&s->thread_context[i]);
683     }
684
685     av_freep(&s->parse_context.buffer);
686     s->parse_context.buffer_size=0;
687
688     av_freep(&s->mb_type);
689     av_freep(&s->p_mv_table_base);
690     av_freep(&s->b_forw_mv_table_base);
691     av_freep(&s->b_back_mv_table_base);
692     av_freep(&s->b_bidir_forw_mv_table_base);
693     av_freep(&s->b_bidir_back_mv_table_base);
694     av_freep(&s->b_direct_mv_table_base);
695     s->p_mv_table= NULL;
696     s->b_forw_mv_table= NULL;
697     s->b_back_mv_table= NULL;
698     s->b_bidir_forw_mv_table= NULL;
699     s->b_bidir_back_mv_table= NULL;
700     s->b_direct_mv_table= NULL;
701     for(i=0; i<2; i++){
702         for(j=0; j<2; j++){
703             for(k=0; k<2; k++){
704                 av_freep(&s->b_field_mv_table_base[i][j][k]);
705                 s->b_field_mv_table[i][j][k]=NULL;
706             }
707             av_freep(&s->b_field_select_table[i][j]);
708             av_freep(&s->p_field_mv_table_base[i][j]);
709             s->p_field_mv_table[i][j]=NULL;
710         }
711         av_freep(&s->p_field_select_table[i]);
712     }
713
714     av_freep(&s->dc_val_base);
715     av_freep(&s->ac_val_base);
716     av_freep(&s->coded_block_base);
717     av_freep(&s->mbintra_table);
718     av_freep(&s->cbp_table);
719     av_freep(&s->pred_dir_table);
720
721     av_freep(&s->mbskip_table);
722     av_freep(&s->prev_pict_types);
723     av_freep(&s->bitstream_buffer);
724     s->allocated_bitstream_buffer_size=0;
725
726     av_freep(&s->avctx->stats_out);
727     av_freep(&s->ac_stats);
728     av_freep(&s->error_status_table);
729     av_freep(&s->mb_index2xy);
730     av_freep(&s->lambda_table);
731     av_freep(&s->q_intra_matrix);
732     av_freep(&s->q_inter_matrix);
733     av_freep(&s->q_intra_matrix16);
734     av_freep(&s->q_inter_matrix16);
735     av_freep(&s->input_picture);
736     av_freep(&s->reordered_input_picture);
737     av_freep(&s->dct_offset);
738
739     if(s->picture){
740         for(i=0; i<MAX_PICTURE_COUNT; i++){
741             free_picture(s, &s->picture[i]);
742         }
743     }
744     av_freep(&s->picture);
745     s->context_initialized = 0;
746     s->last_picture_ptr=
747     s->next_picture_ptr=
748     s->current_picture_ptr= NULL;
749     s->linesize= s->uvlinesize= 0;
750
751     for(i=0; i<3; i++)
752         av_freep(&s->visualization_buffer[i]);
753
754     avcodec_default_free_buffers(s->avctx);
755 }
756
757 void init_rl(RLTable *rl, uint8_t static_store[2][2*MAX_RUN + MAX_LEVEL + 3])
758 {
759     int8_t max_level[MAX_RUN+1], max_run[MAX_LEVEL+1];
760     uint8_t index_run[MAX_RUN+1];
761     int last, run, level, start, end, i;
762
763     /* If table is static, we can quit if rl->max_level[0] is not NULL */
764     if(static_store && rl->max_level[0])
765         return;
766
767     /* compute max_level[], max_run[] and index_run[] */
768     for(last=0;last<2;last++) {
769         if (last == 0) {
770             start = 0;
771             end = rl->last;
772         } else {
773             start = rl->last;
774             end = rl->n;
775         }
776
777         memset(max_level, 0, MAX_RUN + 1);
778         memset(max_run, 0, MAX_LEVEL + 1);
779         memset(index_run, rl->n, MAX_RUN + 1);
780         for(i=start;i<end;i++) {
781             run = rl->table_run[i];
782             level = rl->table_level[i];
783             if (index_run[run] == rl->n)
784                 index_run[run] = i;
785             if (level > max_level[run])
786                 max_level[run] = level;
787             if (run > max_run[level])
788                 max_run[level] = run;
789         }
790         if(static_store)
791             rl->max_level[last] = static_store[last];
792         else
793             rl->max_level[last] = av_malloc(MAX_RUN + 1);
794         memcpy(rl->max_level[last], max_level, MAX_RUN + 1);
795         if(static_store)
796             rl->max_run[last] = static_store[last] + MAX_RUN + 1;
797         else
798             rl->max_run[last] = av_malloc(MAX_LEVEL + 1);
799         memcpy(rl->max_run[last], max_run, MAX_LEVEL + 1);
800         if(static_store)
801             rl->index_run[last] = static_store[last] + MAX_RUN + MAX_LEVEL + 2;
802         else
803             rl->index_run[last] = av_malloc(MAX_RUN + 1);
804         memcpy(rl->index_run[last], index_run, MAX_RUN + 1);
805     }
806 }
807
808 void init_vlc_rl(RLTable *rl)
809 {
810     int i, q;
811
812     for(q=0; q<32; q++){
813         int qmul= q*2;
814         int qadd= (q-1)|1;
815
816         if(q==0){
817             qmul=1;
818             qadd=0;
819         }
820         for(i=0; i<rl->vlc.table_size; i++){
821             int code= rl->vlc.table[i][0];
822             int len = rl->vlc.table[i][1];
823             int level, run;
824
825             if(len==0){ // illegal code
826                 run= 66;
827                 level= MAX_LEVEL;
828             }else if(len<0){ //more bits needed
829                 run= 0;
830                 level= code;
831             }else{
832                 if(code==rl->n){ //esc
833                     run= 66;
834                     level= 0;
835                 }else{
836                     run=   rl->table_run  [code] + 1;
837                     level= rl->table_level[code] * qmul + qadd;
838                     if(code >= rl->last) run+=192;
839                 }
840             }
841             rl->rl_vlc[q][i].len= len;
842             rl->rl_vlc[q][i].level= level;
843             rl->rl_vlc[q][i].run= run;
844         }
845     }
846 }
847
848 int ff_find_unused_picture(MpegEncContext *s, int shared){
849     int i;
850
851     if(shared){
852         for(i=0; i<MAX_PICTURE_COUNT; i++){
853             if(s->picture[i].data[0]==NULL && s->picture[i].type==0) return i;
854         }
855     }else{
856         for(i=0; i<MAX_PICTURE_COUNT; i++){
857             if(s->picture[i].data[0]==NULL && s->picture[i].type!=0) return i; //FIXME
858         }
859         for(i=0; i<MAX_PICTURE_COUNT; i++){
860             if(s->picture[i].data[0]==NULL) return i;
861         }
862     }
863
864     av_log(s->avctx, AV_LOG_FATAL, "Internal error, picture buffer overflow\n");
865     /* We could return -1, but the codec would crash trying to draw into a
866      * non-existing frame anyway. This is safer than waiting for a random crash.
867      * Also the return of this is never useful, an encoder must only allocate
868      * as much as allowed in the specification. This has no relationship to how
869      * much libavcodec could allocate (and MAX_PICTURE_COUNT is always large
870      * enough for such valid streams).
871      * Plus, a decoder has to check stream validity and remove frames if too
872      * many reference frames are around. Waiting for "OOM" is not correct at
873      * all. Similarly, missing reference frames have to be replaced by
874      * interpolated/MC frames, anything else is a bug in the codec ...
875      */
876     abort();
877     return -1;
878 }
879
880 static void update_noise_reduction(MpegEncContext *s){
881     int intra, i;
882
883     for(intra=0; intra<2; intra++){
884         if(s->dct_count[intra] > (1<<16)){
885             for(i=0; i<64; i++){
886                 s->dct_error_sum[intra][i] >>=1;
887             }
888             s->dct_count[intra] >>= 1;
889         }
890
891         for(i=0; i<64; i++){
892             s->dct_offset[intra][i]= (s->avctx->noise_reduction * s->dct_count[intra] + s->dct_error_sum[intra][i]/2) / (s->dct_error_sum[intra][i]+1);
893         }
894     }
895 }
896
897 /**
898  * generic function for encode/decode called after coding/decoding the header and before a frame is coded/decoded
899  */
900 int MPV_frame_start(MpegEncContext *s, AVCodecContext *avctx)
901 {
902     int i;
903     Picture *pic;
904     s->mb_skipped = 0;
905
906     assert(s->last_picture_ptr==NULL || s->out_format != FMT_H264 || s->codec_id == CODEC_ID_SVQ3);
907
908     /* mark&release old frames */
909     if (s->pict_type != FF_B_TYPE && s->last_picture_ptr && s->last_picture_ptr != s->next_picture_ptr && s->last_picture_ptr->data[0]) {
910       if(s->out_format != FMT_H264 || s->codec_id == CODEC_ID_SVQ3){
911           free_frame_buffer(s, s->last_picture_ptr);
912
913         /* release forgotten pictures */
914         /* if(mpeg124/h263) */
915         if(!s->encoding){
916             for(i=0; i<MAX_PICTURE_COUNT; i++){
917                 if(s->picture[i].data[0] && &s->picture[i] != s->next_picture_ptr && s->picture[i].reference){
918                     av_log(avctx, AV_LOG_ERROR, "releasing zombie picture\n");
919                     free_frame_buffer(s, &s->picture[i]);
920                 }
921             }
922         }
923       }
924     }
925
926     if(!s->encoding){
927         /* release non reference frames */
928         for(i=0; i<MAX_PICTURE_COUNT; i++){
929             if(s->picture[i].data[0] && !s->picture[i].reference /*&& s->picture[i].type!=FF_BUFFER_TYPE_SHARED*/){
930                 free_frame_buffer(s, &s->picture[i]);
931             }
932         }
933
934         if(s->current_picture_ptr && s->current_picture_ptr->data[0]==NULL)
935             pic= s->current_picture_ptr; //we already have a unused image (maybe it was set before reading the header)
936         else{
937             i= ff_find_unused_picture(s, 0);
938             pic= &s->picture[i];
939         }
940
941         pic->reference= 0;
942         if (!s->dropable){
943             if (s->codec_id == CODEC_ID_H264)
944                 pic->reference = s->picture_structure;
945             else if (s->pict_type != FF_B_TYPE)
946                 pic->reference = 3;
947         }
948
949         pic->coded_picture_number= s->coded_picture_number++;
950
951         if(ff_alloc_picture(s, pic, 0) < 0)
952             return -1;
953
954         s->current_picture_ptr= pic;
955         s->current_picture_ptr->top_field_first= s->top_field_first; //FIXME use only the vars from current_pic
956         s->current_picture_ptr->interlaced_frame= !s->progressive_frame && !s->progressive_sequence;
957     }
958
959     s->current_picture_ptr->pict_type= s->pict_type;
960 //    if(s->flags && CODEC_FLAG_QSCALE)
961   //      s->current_picture_ptr->quality= s->new_picture_ptr->quality;
962     s->current_picture_ptr->key_frame= s->pict_type == FF_I_TYPE;
963
964     ff_copy_picture(&s->current_picture, s->current_picture_ptr);
965
966     if (s->pict_type != FF_B_TYPE) {
967         s->last_picture_ptr= s->next_picture_ptr;
968         if(!s->dropable)
969             s->next_picture_ptr= s->current_picture_ptr;
970     }
971 /*    av_log(s->avctx, AV_LOG_DEBUG, "L%p N%p C%p L%p N%p C%p type:%d drop:%d\n", s->last_picture_ptr, s->next_picture_ptr,s->current_picture_ptr,
972         s->last_picture_ptr    ? s->last_picture_ptr->data[0] : NULL,
973         s->next_picture_ptr    ? s->next_picture_ptr->data[0] : NULL,
974         s->current_picture_ptr ? s->current_picture_ptr->data[0] : NULL,
975         s->pict_type, s->dropable);*/
976
977     if(s->codec_id != CODEC_ID_H264){
978         if((s->last_picture_ptr==NULL || s->last_picture_ptr->data[0]==NULL) && s->pict_type!=FF_I_TYPE){
979             av_log(avctx, AV_LOG_ERROR, "warning: first frame is no keyframe\n");
980             /* Allocate a dummy frame */
981             i= ff_find_unused_picture(s, 0);
982             s->last_picture_ptr= &s->picture[i];
983             if(ff_alloc_picture(s, s->last_picture_ptr, 0) < 0)
984                 return -1;
985         }
986         if((s->next_picture_ptr==NULL || s->next_picture_ptr->data[0]==NULL) && s->pict_type==FF_B_TYPE){
987             /* Allocate a dummy frame */
988             i= ff_find_unused_picture(s, 0);
989             s->next_picture_ptr= &s->picture[i];
990             if(ff_alloc_picture(s, s->next_picture_ptr, 0) < 0)
991                 return -1;
992         }
993     }
994
995     if(s->last_picture_ptr) ff_copy_picture(&s->last_picture, s->last_picture_ptr);
996     if(s->next_picture_ptr) ff_copy_picture(&s->next_picture, s->next_picture_ptr);
997
998     assert(s->pict_type == FF_I_TYPE || (s->last_picture_ptr && s->last_picture_ptr->data[0]));
999
1000     if(s->picture_structure!=PICT_FRAME && s->out_format != FMT_H264){
1001         int i;
1002         for(i=0; i<4; i++){
1003             if(s->picture_structure == PICT_BOTTOM_FIELD){
1004                  s->current_picture.data[i] += s->current_picture.linesize[i];
1005             }
1006             s->current_picture.linesize[i] *= 2;
1007             s->last_picture.linesize[i] *=2;
1008             s->next_picture.linesize[i] *=2;
1009         }
1010     }
1011
1012     s->hurry_up= s->avctx->hurry_up;
1013     s->error_recognition= avctx->error_recognition;
1014
1015     /* set dequantizer, we can't do it during init as it might change for mpeg4
1016        and we can't do it in the header decode as init is not called for mpeg4 there yet */
1017     if(s->mpeg_quant || s->codec_id == CODEC_ID_MPEG2VIDEO){
1018         s->dct_unquantize_intra = s->dct_unquantize_mpeg2_intra;
1019         s->dct_unquantize_inter = s->dct_unquantize_mpeg2_inter;
1020     }else if(s->out_format == FMT_H263 || s->out_format == FMT_H261){
1021         s->dct_unquantize_intra = s->dct_unquantize_h263_intra;
1022         s->dct_unquantize_inter = s->dct_unquantize_h263_inter;
1023     }else{
1024         s->dct_unquantize_intra = s->dct_unquantize_mpeg1_intra;
1025         s->dct_unquantize_inter = s->dct_unquantize_mpeg1_inter;
1026     }
1027
1028     if(s->dct_error_sum){
1029         assert(s->avctx->noise_reduction && s->encoding);
1030
1031         update_noise_reduction(s);
1032     }
1033
1034     if(CONFIG_MPEG_XVMC_DECODER && s->avctx->xvmc_acceleration)
1035         return ff_xvmc_field_start(s, avctx);
1036
1037     return 0;
1038 }
1039
1040 /* generic function for encode/decode called after a frame has been coded/decoded */
1041 void MPV_frame_end(MpegEncContext *s)
1042 {
1043     int i;
1044     /* draw edge for correct motion prediction if outside */
1045     //just to make sure that all data is rendered.
1046     if(CONFIG_MPEG_XVMC_DECODER && s->avctx->xvmc_acceleration){
1047         ff_xvmc_field_end(s);
1048     }else if(!s->avctx->hwaccel
1049        && !(s->avctx->codec->capabilities&CODEC_CAP_HWACCEL_VDPAU)
1050        && s->unrestricted_mv
1051        && s->current_picture.reference
1052        && !s->intra_only
1053        && !(s->flags&CODEC_FLAG_EMU_EDGE)) {
1054             s->dsp.draw_edges(s->current_picture.data[0], s->linesize  , s->h_edge_pos   , s->v_edge_pos   , EDGE_WIDTH  );
1055             s->dsp.draw_edges(s->current_picture.data[1], s->uvlinesize, s->h_edge_pos>>1, s->v_edge_pos>>1, EDGE_WIDTH/2);
1056             s->dsp.draw_edges(s->current_picture.data[2], s->uvlinesize, s->h_edge_pos>>1, s->v_edge_pos>>1, EDGE_WIDTH/2);
1057     }
1058     emms_c();
1059
1060     s->last_pict_type    = s->pict_type;
1061     s->last_lambda_for[s->pict_type]= s->current_picture_ptr->quality;
1062     if(s->pict_type!=FF_B_TYPE){
1063         s->last_non_b_pict_type= s->pict_type;
1064     }
1065 #if 0
1066         /* copy back current_picture variables */
1067     for(i=0; i<MAX_PICTURE_COUNT; i++){
1068         if(s->picture[i].data[0] == s->current_picture.data[0]){
1069             s->picture[i]= s->current_picture;
1070             break;
1071         }
1072     }
1073     assert(i<MAX_PICTURE_COUNT);
1074 #endif
1075
1076     if(s->encoding){
1077         /* release non-reference frames */
1078         for(i=0; i<MAX_PICTURE_COUNT; i++){
1079             if(s->picture[i].data[0] && !s->picture[i].reference /*&& s->picture[i].type!=FF_BUFFER_TYPE_SHARED*/){
1080                 free_frame_buffer(s, &s->picture[i]);
1081             }
1082         }
1083     }
1084     // clear copies, to avoid confusion
1085 #if 0
1086     memset(&s->last_picture, 0, sizeof(Picture));
1087     memset(&s->next_picture, 0, sizeof(Picture));
1088     memset(&s->current_picture, 0, sizeof(Picture));
1089 #endif
1090     s->avctx->coded_frame= (AVFrame*)s->current_picture_ptr;
1091 }
1092
1093 /**
1094  * draws an line from (ex, ey) -> (sx, sy).
1095  * @param w width of the image
1096  * @param h height of the image
1097  * @param stride stride/linesize of the image
1098  * @param color color of the arrow
1099  */
1100 static void draw_line(uint8_t *buf, int sx, int sy, int ex, int ey, int w, int h, int stride, int color){
1101     int x, y, fr, f;
1102
1103     sx= av_clip(sx, 0, w-1);
1104     sy= av_clip(sy, 0, h-1);
1105     ex= av_clip(ex, 0, w-1);
1106     ey= av_clip(ey, 0, h-1);
1107
1108     buf[sy*stride + sx]+= color;
1109
1110     if(FFABS(ex - sx) > FFABS(ey - sy)){
1111         if(sx > ex){
1112             FFSWAP(int, sx, ex);
1113             FFSWAP(int, sy, ey);
1114         }
1115         buf+= sx + sy*stride;
1116         ex-= sx;
1117         f= ((ey-sy)<<16)/ex;
1118         for(x= 0; x <= ex; x++){
1119             y = (x*f)>>16;
1120             fr= (x*f)&0xFFFF;
1121             buf[ y   *stride + x]+= (color*(0x10000-fr))>>16;
1122             buf[(y+1)*stride + x]+= (color*         fr )>>16;
1123         }
1124     }else{
1125         if(sy > ey){
1126             FFSWAP(int, sx, ex);
1127             FFSWAP(int, sy, ey);
1128         }
1129         buf+= sx + sy*stride;
1130         ey-= sy;
1131         if(ey) f= ((ex-sx)<<16)/ey;
1132         else   f= 0;
1133         for(y= 0; y <= ey; y++){
1134             x = (y*f)>>16;
1135             fr= (y*f)&0xFFFF;
1136             buf[y*stride + x  ]+= (color*(0x10000-fr))>>16;
1137             buf[y*stride + x+1]+= (color*         fr )>>16;
1138         }
1139     }
1140 }
1141
1142 /**
1143  * draws an arrow from (ex, ey) -> (sx, sy).
1144  * @param w width of the image
1145  * @param h height of the image
1146  * @param stride stride/linesize of the image
1147  * @param color color of the arrow
1148  */
1149 static void draw_arrow(uint8_t *buf, int sx, int sy, int ex, int ey, int w, int h, int stride, int color){
1150     int dx,dy;
1151
1152     sx= av_clip(sx, -100, w+100);
1153     sy= av_clip(sy, -100, h+100);
1154     ex= av_clip(ex, -100, w+100);
1155     ey= av_clip(ey, -100, h+100);
1156
1157     dx= ex - sx;
1158     dy= ey - sy;
1159
1160     if(dx*dx + dy*dy > 3*3){
1161         int rx=  dx + dy;
1162         int ry= -dx + dy;
1163         int length= ff_sqrt((rx*rx + ry*ry)<<8);
1164
1165         //FIXME subpixel accuracy
1166         rx= ROUNDED_DIV(rx*3<<4, length);
1167         ry= ROUNDED_DIV(ry*3<<4, length);
1168
1169         draw_line(buf, sx, sy, sx + rx, sy + ry, w, h, stride, color);
1170         draw_line(buf, sx, sy, sx - ry, sy + rx, w, h, stride, color);
1171     }
1172     draw_line(buf, sx, sy, ex, ey, w, h, stride, color);
1173 }
1174
1175 /**
1176  * prints debuging info for the given picture.
1177  */
1178 void ff_print_debug_info(MpegEncContext *s, AVFrame *pict){
1179
1180     if(s->avctx->hwaccel || !pict || !pict->mb_type) return;
1181
1182     if(s->avctx->debug&(FF_DEBUG_SKIP | FF_DEBUG_QP | FF_DEBUG_MB_TYPE)){
1183         int x,y;
1184
1185         av_log(s->avctx,AV_LOG_DEBUG,"New frame, type: ");
1186         switch (pict->pict_type) {
1187             case FF_I_TYPE: av_log(s->avctx,AV_LOG_DEBUG,"I\n"); break;
1188             case FF_P_TYPE: av_log(s->avctx,AV_LOG_DEBUG,"P\n"); break;
1189             case FF_B_TYPE: av_log(s->avctx,AV_LOG_DEBUG,"B\n"); break;
1190             case FF_S_TYPE: av_log(s->avctx,AV_LOG_DEBUG,"S\n"); break;
1191             case FF_SI_TYPE: av_log(s->avctx,AV_LOG_DEBUG,"SI\n"); break;
1192             case FF_SP_TYPE: av_log(s->avctx,AV_LOG_DEBUG,"SP\n"); break;
1193         }
1194         for(y=0; y<s->mb_height; y++){
1195             for(x=0; x<s->mb_width; x++){
1196                 if(s->avctx->debug&FF_DEBUG_SKIP){
1197                     int count= s->mbskip_table[x + y*s->mb_stride];
1198                     if(count>9) count=9;
1199                     av_log(s->avctx, AV_LOG_DEBUG, "%1d", count);
1200                 }
1201                 if(s->avctx->debug&FF_DEBUG_QP){
1202                     av_log(s->avctx, AV_LOG_DEBUG, "%2d", pict->qscale_table[x + y*s->mb_stride]);
1203                 }
1204                 if(s->avctx->debug&FF_DEBUG_MB_TYPE){
1205                     int mb_type= pict->mb_type[x + y*s->mb_stride];
1206                     //Type & MV direction
1207                     if(IS_PCM(mb_type))
1208                         av_log(s->avctx, AV_LOG_DEBUG, "P");
1209                     else if(IS_INTRA(mb_type) && IS_ACPRED(mb_type))
1210                         av_log(s->avctx, AV_LOG_DEBUG, "A");
1211                     else if(IS_INTRA4x4(mb_type))
1212                         av_log(s->avctx, AV_LOG_DEBUG, "i");
1213                     else if(IS_INTRA16x16(mb_type))
1214                         av_log(s->avctx, AV_LOG_DEBUG, "I");
1215                     else if(IS_DIRECT(mb_type) && IS_SKIP(mb_type))
1216                         av_log(s->avctx, AV_LOG_DEBUG, "d");
1217                     else if(IS_DIRECT(mb_type))
1218                         av_log(s->avctx, AV_LOG_DEBUG, "D");
1219                     else if(IS_GMC(mb_type) && IS_SKIP(mb_type))
1220                         av_log(s->avctx, AV_LOG_DEBUG, "g");
1221                     else if(IS_GMC(mb_type))
1222                         av_log(s->avctx, AV_LOG_DEBUG, "G");
1223                     else if(IS_SKIP(mb_type))
1224                         av_log(s->avctx, AV_LOG_DEBUG, "S");
1225                     else if(!USES_LIST(mb_type, 1))
1226                         av_log(s->avctx, AV_LOG_DEBUG, ">");
1227                     else if(!USES_LIST(mb_type, 0))
1228                         av_log(s->avctx, AV_LOG_DEBUG, "<");
1229                     else{
1230                         assert(USES_LIST(mb_type, 0) && USES_LIST(mb_type, 1));
1231                         av_log(s->avctx, AV_LOG_DEBUG, "X");
1232                     }
1233
1234                     //segmentation
1235                     if(IS_8X8(mb_type))
1236                         av_log(s->avctx, AV_LOG_DEBUG, "+");
1237                     else if(IS_16X8(mb_type))
1238                         av_log(s->avctx, AV_LOG_DEBUG, "-");
1239                     else if(IS_8X16(mb_type))
1240                         av_log(s->avctx, AV_LOG_DEBUG, "|");
1241                     else if(IS_INTRA(mb_type) || IS_16X16(mb_type))
1242                         av_log(s->avctx, AV_LOG_DEBUG, " ");
1243                     else
1244                         av_log(s->avctx, AV_LOG_DEBUG, "?");
1245
1246
1247                     if(IS_INTERLACED(mb_type) && s->codec_id == CODEC_ID_H264)
1248                         av_log(s->avctx, AV_LOG_DEBUG, "=");
1249                     else
1250                         av_log(s->avctx, AV_LOG_DEBUG, " ");
1251                 }
1252 //                av_log(s->avctx, AV_LOG_DEBUG, " ");
1253             }
1254             av_log(s->avctx, AV_LOG_DEBUG, "\n");
1255         }
1256     }
1257
1258     if((s->avctx->debug&(FF_DEBUG_VIS_QP|FF_DEBUG_VIS_MB_TYPE)) || (s->avctx->debug_mv)){
1259         const int shift= 1 + s->quarter_sample;
1260         int mb_y;
1261         uint8_t *ptr;
1262         int i;
1263         int h_chroma_shift, v_chroma_shift, block_height;
1264         const int width = s->avctx->width;
1265         const int height= s->avctx->height;
1266         const int mv_sample_log2= 4 - pict->motion_subsample_log2;
1267         const int mv_stride= (s->mb_width << mv_sample_log2) + (s->codec_id == CODEC_ID_H264 ? 0 : 1);
1268         s->low_delay=0; //needed to see the vectors without trashing the buffers
1269
1270         avcodec_get_chroma_sub_sample(s->avctx->pix_fmt, &h_chroma_shift, &v_chroma_shift);
1271         for(i=0; i<3; i++){
1272             memcpy(s->visualization_buffer[i], pict->data[i], (i==0) ? pict->linesize[i]*height:pict->linesize[i]*height >> v_chroma_shift);
1273             pict->data[i]= s->visualization_buffer[i];
1274         }
1275         pict->type= FF_BUFFER_TYPE_COPY;
1276         ptr= pict->data[0];
1277         block_height = 16>>v_chroma_shift;
1278
1279         for(mb_y=0; mb_y<s->mb_height; mb_y++){
1280             int mb_x;
1281             for(mb_x=0; mb_x<s->mb_width; mb_x++){
1282                 const int mb_index= mb_x + mb_y*s->mb_stride;
1283                 if((s->avctx->debug_mv) && pict->motion_val){
1284                   int type;
1285                   for(type=0; type<3; type++){
1286                     int direction = 0;
1287                     switch (type) {
1288                       case 0: if ((!(s->avctx->debug_mv&FF_DEBUG_VIS_MV_P_FOR)) || (pict->pict_type!=FF_P_TYPE))
1289                                 continue;
1290                               direction = 0;
1291                               break;
1292                       case 1: if ((!(s->avctx->debug_mv&FF_DEBUG_VIS_MV_B_FOR)) || (pict->pict_type!=FF_B_TYPE))
1293                                 continue;
1294                               direction = 0;
1295                               break;
1296                       case 2: if ((!(s->avctx->debug_mv&FF_DEBUG_VIS_MV_B_BACK)) || (pict->pict_type!=FF_B_TYPE))
1297                                 continue;
1298                               direction = 1;
1299                               break;
1300                     }
1301                     if(!USES_LIST(pict->mb_type[mb_index], direction))
1302                         continue;
1303
1304                     if(IS_8X8(pict->mb_type[mb_index])){
1305                       int i;
1306                       for(i=0; i<4; i++){
1307                         int sx= mb_x*16 + 4 + 8*(i&1);
1308                         int sy= mb_y*16 + 4 + 8*(i>>1);
1309                         int xy= (mb_x*2 + (i&1) + (mb_y*2 + (i>>1))*mv_stride) << (mv_sample_log2-1);
1310                         int mx= (pict->motion_val[direction][xy][0]>>shift) + sx;
1311                         int my= (pict->motion_val[direction][xy][1]>>shift) + sy;
1312                         draw_arrow(ptr, sx, sy, mx, my, width, height, s->linesize, 100);
1313                       }
1314                     }else if(IS_16X8(pict->mb_type[mb_index])){
1315                       int i;
1316                       for(i=0; i<2; i++){
1317                         int sx=mb_x*16 + 8;
1318                         int sy=mb_y*16 + 4 + 8*i;
1319                         int xy= (mb_x*2 + (mb_y*2 + i)*mv_stride) << (mv_sample_log2-1);
1320                         int mx=(pict->motion_val[direction][xy][0]>>shift);
1321                         int my=(pict->motion_val[direction][xy][1]>>shift);
1322
1323                         if(IS_INTERLACED(pict->mb_type[mb_index]))
1324                             my*=2;
1325
1326                         draw_arrow(ptr, sx, sy, mx+sx, my+sy, width, height, s->linesize, 100);
1327                       }
1328                     }else if(IS_8X16(pict->mb_type[mb_index])){
1329                       int i;
1330                       for(i=0; i<2; i++){
1331                         int sx=mb_x*16 + 4 + 8*i;
1332                         int sy=mb_y*16 + 8;
1333                         int xy= (mb_x*2 + i + mb_y*2*mv_stride) << (mv_sample_log2-1);
1334                         int mx=(pict->motion_val[direction][xy][0]>>shift);
1335                         int my=(pict->motion_val[direction][xy][1]>>shift);
1336
1337                         if(IS_INTERLACED(pict->mb_type[mb_index]))
1338                             my*=2;
1339
1340                         draw_arrow(ptr, sx, sy, mx+sx, my+sy, width, height, s->linesize, 100);
1341                       }
1342                     }else{
1343                       int sx= mb_x*16 + 8;
1344                       int sy= mb_y*16 + 8;
1345                       int xy= (mb_x + mb_y*mv_stride) << mv_sample_log2;
1346                       int mx= (pict->motion_val[direction][xy][0]>>shift) + sx;
1347                       int my= (pict->motion_val[direction][xy][1]>>shift) + sy;
1348                       draw_arrow(ptr, sx, sy, mx, my, width, height, s->linesize, 100);
1349                     }
1350                   }
1351                 }
1352                 if((s->avctx->debug&FF_DEBUG_VIS_QP) && pict->motion_val){
1353                     uint64_t c= (pict->qscale_table[mb_index]*128/31) * 0x0101010101010101ULL;
1354                     int y;
1355                     for(y=0; y<block_height; y++){
1356                         *(uint64_t*)(pict->data[1] + 8*mb_x + (block_height*mb_y + y)*pict->linesize[1])= c;
1357                         *(uint64_t*)(pict->data[2] + 8*mb_x + (block_height*mb_y + y)*pict->linesize[2])= c;
1358                     }
1359                 }
1360                 if((s->avctx->debug&FF_DEBUG_VIS_MB_TYPE) && pict->motion_val){
1361                     int mb_type= pict->mb_type[mb_index];
1362                     uint64_t u,v;
1363                     int y;
1364 #define COLOR(theta, r)\
1365 u= (int)(128 + r*cos(theta*3.141592/180));\
1366 v= (int)(128 + r*sin(theta*3.141592/180));
1367
1368
1369                     u=v=128;
1370                     if(IS_PCM(mb_type)){
1371                         COLOR(120,48)
1372                     }else if((IS_INTRA(mb_type) && IS_ACPRED(mb_type)) || IS_INTRA16x16(mb_type)){
1373                         COLOR(30,48)
1374                     }else if(IS_INTRA4x4(mb_type)){
1375                         COLOR(90,48)
1376                     }else if(IS_DIRECT(mb_type) && IS_SKIP(mb_type)){
1377 //                        COLOR(120,48)
1378                     }else if(IS_DIRECT(mb_type)){
1379                         COLOR(150,48)
1380                     }else if(IS_GMC(mb_type) && IS_SKIP(mb_type)){
1381                         COLOR(170,48)
1382                     }else if(IS_GMC(mb_type)){
1383                         COLOR(190,48)
1384                     }else if(IS_SKIP(mb_type)){
1385 //                        COLOR(180,48)
1386                     }else if(!USES_LIST(mb_type, 1)){
1387                         COLOR(240,48)
1388                     }else if(!USES_LIST(mb_type, 0)){
1389                         COLOR(0,48)
1390                     }else{
1391                         assert(USES_LIST(mb_type, 0) && USES_LIST(mb_type, 1));
1392                         COLOR(300,48)
1393                     }
1394
1395                     u*= 0x0101010101010101ULL;
1396                     v*= 0x0101010101010101ULL;
1397                     for(y=0; y<block_height; y++){
1398                         *(uint64_t*)(pict->data[1] + 8*mb_x + (block_height*mb_y + y)*pict->linesize[1])= u;
1399                         *(uint64_t*)(pict->data[2] + 8*mb_x + (block_height*mb_y + y)*pict->linesize[2])= v;
1400                     }
1401
1402                     //segmentation
1403                     if(IS_8X8(mb_type) || IS_16X8(mb_type)){
1404                         *(uint64_t*)(pict->data[0] + 16*mb_x + 0 + (16*mb_y + 8)*pict->linesize[0])^= 0x8080808080808080ULL;
1405                         *(uint64_t*)(pict->data[0] + 16*mb_x + 8 + (16*mb_y + 8)*pict->linesize[0])^= 0x8080808080808080ULL;
1406                     }
1407                     if(IS_8X8(mb_type) || IS_8X16(mb_type)){
1408                         for(y=0; y<16; y++)
1409                             pict->data[0][16*mb_x + 8 + (16*mb_y + y)*pict->linesize[0]]^= 0x80;
1410                     }
1411                     if(IS_8X8(mb_type) && mv_sample_log2 >= 2){
1412                         int dm= 1 << (mv_sample_log2-2);
1413                         for(i=0; i<4; i++){
1414                             int sx= mb_x*16 + 8*(i&1);
1415                             int sy= mb_y*16 + 8*(i>>1);
1416                             int xy= (mb_x*2 + (i&1) + (mb_y*2 + (i>>1))*mv_stride) << (mv_sample_log2-1);
1417                             //FIXME bidir
1418                             int32_t *mv = (int32_t*)&pict->motion_val[0][xy];
1419                             if(mv[0] != mv[dm] || mv[dm*mv_stride] != mv[dm*(mv_stride+1)])
1420                                 for(y=0; y<8; y++)
1421                                     pict->data[0][sx + 4 + (sy + y)*pict->linesize[0]]^= 0x80;
1422                             if(mv[0] != mv[dm*mv_stride] || mv[dm] != mv[dm*(mv_stride+1)])
1423                                 *(uint64_t*)(pict->data[0] + sx + (sy + 4)*pict->linesize[0])^= 0x8080808080808080ULL;
1424                         }
1425                     }
1426
1427                     if(IS_INTERLACED(mb_type) && s->codec_id == CODEC_ID_H264){
1428                         // hmm
1429                     }
1430                 }
1431                 s->mbskip_table[mb_index]=0;
1432             }
1433         }
1434     }
1435 }
1436
1437 static inline int hpel_motion_lowres(MpegEncContext *s,
1438                                   uint8_t *dest, uint8_t *src,
1439                                   int field_based, int field_select,
1440                                   int src_x, int src_y,
1441                                   int width, int height, int stride,
1442                                   int h_edge_pos, int v_edge_pos,
1443                                   int w, int h, h264_chroma_mc_func *pix_op,
1444                                   int motion_x, int motion_y)
1445 {
1446     const int lowres= s->avctx->lowres;
1447     const int op_index= FFMIN(lowres, 2);
1448     const int s_mask= (2<<lowres)-1;
1449     int emu=0;
1450     int sx, sy;
1451
1452     if(s->quarter_sample){
1453         motion_x/=2;
1454         motion_y/=2;
1455     }
1456
1457     sx= motion_x & s_mask;
1458     sy= motion_y & s_mask;
1459     src_x += motion_x >> (lowres+1);
1460     src_y += motion_y >> (lowres+1);
1461
1462     src += src_y * stride + src_x;
1463
1464     if(   (unsigned)src_x > h_edge_pos                 - (!!sx) - w
1465        || (unsigned)src_y >(v_edge_pos >> field_based) - (!!sy) - h){
1466         ff_emulated_edge_mc(s->edge_emu_buffer, src, s->linesize, w+1, (h+1)<<field_based,
1467                             src_x, src_y<<field_based, h_edge_pos, v_edge_pos);
1468         src= s->edge_emu_buffer;
1469         emu=1;
1470     }
1471
1472     sx= (sx << 2) >> lowres;
1473     sy= (sy << 2) >> lowres;
1474     if(field_select)
1475         src += s->linesize;
1476     pix_op[op_index](dest, src, stride, h, sx, sy);
1477     return emu;
1478 }
1479
1480 /* apply one mpeg motion vector to the three components */
1481 static av_always_inline void mpeg_motion_lowres(MpegEncContext *s,
1482                                uint8_t *dest_y, uint8_t *dest_cb, uint8_t *dest_cr,
1483                                int field_based, int bottom_field, int field_select,
1484                                uint8_t **ref_picture, h264_chroma_mc_func *pix_op,
1485                                int motion_x, int motion_y, int h, int mb_y)
1486 {
1487     uint8_t *ptr_y, *ptr_cb, *ptr_cr;
1488     int mx, my, src_x, src_y, uvsrc_x, uvsrc_y, uvlinesize, linesize, sx, sy, uvsx, uvsy;
1489     const int lowres= s->avctx->lowres;
1490     const int op_index= FFMIN(lowres, 2);
1491     const int block_s= 8>>lowres;
1492     const int s_mask= (2<<lowres)-1;
1493     const int h_edge_pos = s->h_edge_pos >> lowres;
1494     const int v_edge_pos = s->v_edge_pos >> lowres;
1495     linesize   = s->current_picture.linesize[0] << field_based;
1496     uvlinesize = s->current_picture.linesize[1] << field_based;
1497
1498     if(s->quarter_sample){ //FIXME obviously not perfect but qpel will not work in lowres anyway
1499         motion_x/=2;
1500         motion_y/=2;
1501     }
1502
1503     if(field_based){
1504         motion_y += (bottom_field - field_select)*((1<<lowres)-1);
1505     }
1506
1507     sx= motion_x & s_mask;
1508     sy= motion_y & s_mask;
1509     src_x = s->mb_x*2*block_s               + (motion_x >> (lowres+1));
1510     src_y =(   mb_y*2*block_s>>field_based) + (motion_y >> (lowres+1));
1511
1512     if (s->out_format == FMT_H263) {
1513         uvsx = ((motion_x>>1) & s_mask) | (sx&1);
1514         uvsy = ((motion_y>>1) & s_mask) | (sy&1);
1515         uvsrc_x = src_x>>1;
1516         uvsrc_y = src_y>>1;
1517     }else if(s->out_format == FMT_H261){//even chroma mv's are full pel in H261
1518         mx = motion_x / 4;
1519         my = motion_y / 4;
1520         uvsx = (2*mx) & s_mask;
1521         uvsy = (2*my) & s_mask;
1522         uvsrc_x = s->mb_x*block_s               + (mx >> lowres);
1523         uvsrc_y =    mb_y*block_s               + (my >> lowres);
1524     } else {
1525         mx = motion_x / 2;
1526         my = motion_y / 2;
1527         uvsx = mx & s_mask;
1528         uvsy = my & s_mask;
1529         uvsrc_x = s->mb_x*block_s               + (mx >> (lowres+1));
1530         uvsrc_y =(   mb_y*block_s>>field_based) + (my >> (lowres+1));
1531     }
1532
1533     ptr_y  = ref_picture[0] + src_y * linesize + src_x;
1534     ptr_cb = ref_picture[1] + uvsrc_y * uvlinesize + uvsrc_x;
1535     ptr_cr = ref_picture[2] + uvsrc_y * uvlinesize + uvsrc_x;
1536
1537     if(   (unsigned)src_x > h_edge_pos                 - (!!sx) - 2*block_s
1538        || (unsigned)src_y >(v_edge_pos >> field_based) - (!!sy) - h){
1539             ff_emulated_edge_mc(s->edge_emu_buffer, ptr_y, s->linesize, 17, 17+field_based,
1540                              src_x, src_y<<field_based, h_edge_pos, v_edge_pos);
1541             ptr_y = s->edge_emu_buffer;
1542             if(!CONFIG_GRAY || !(s->flags&CODEC_FLAG_GRAY)){
1543                 uint8_t *uvbuf= s->edge_emu_buffer+18*s->linesize;
1544                 ff_emulated_edge_mc(uvbuf  , ptr_cb, s->uvlinesize, 9, 9+field_based,
1545                                  uvsrc_x, uvsrc_y<<field_based, h_edge_pos>>1, v_edge_pos>>1);
1546                 ff_emulated_edge_mc(uvbuf+16, ptr_cr, s->uvlinesize, 9, 9+field_based,
1547                                  uvsrc_x, uvsrc_y<<field_based, h_edge_pos>>1, v_edge_pos>>1);
1548                 ptr_cb= uvbuf;
1549                 ptr_cr= uvbuf+16;
1550             }
1551     }
1552
1553     if(bottom_field){ //FIXME use this for field pix too instead of the obnoxious hack which changes picture.data
1554         dest_y += s->linesize;
1555         dest_cb+= s->uvlinesize;
1556         dest_cr+= s->uvlinesize;
1557     }
1558
1559     if(field_select){
1560         ptr_y += s->linesize;
1561         ptr_cb+= s->uvlinesize;
1562         ptr_cr+= s->uvlinesize;
1563     }
1564
1565     sx= (sx << 2) >> lowres;
1566     sy= (sy << 2) >> lowres;
1567     pix_op[lowres-1](dest_y, ptr_y, linesize, h, sx, sy);
1568
1569     if(!CONFIG_GRAY || !(s->flags&CODEC_FLAG_GRAY)){
1570         uvsx= (uvsx << 2) >> lowres;
1571         uvsy= (uvsy << 2) >> lowres;
1572         pix_op[op_index](dest_cb, ptr_cb, uvlinesize, h >> s->chroma_y_shift, uvsx, uvsy);
1573         pix_op[op_index](dest_cr, ptr_cr, uvlinesize, h >> s->chroma_y_shift, uvsx, uvsy);
1574     }
1575     //FIXME h261 lowres loop filter
1576 }
1577
1578 static inline void chroma_4mv_motion_lowres(MpegEncContext *s,
1579                                      uint8_t *dest_cb, uint8_t *dest_cr,
1580                                      uint8_t **ref_picture,
1581                                      h264_chroma_mc_func *pix_op,
1582                                      int mx, int my){
1583     const int lowres= s->avctx->lowres;
1584     const int op_index= FFMIN(lowres, 2);
1585     const int block_s= 8>>lowres;
1586     const int s_mask= (2<<lowres)-1;
1587     const int h_edge_pos = s->h_edge_pos >> (lowres+1);
1588     const int v_edge_pos = s->v_edge_pos >> (lowres+1);
1589     int emu=0, src_x, src_y, offset, sx, sy;
1590     uint8_t *ptr;
1591
1592     if(s->quarter_sample){
1593         mx/=2;
1594         my/=2;
1595     }
1596
1597     /* In case of 8X8, we construct a single chroma motion vector
1598        with a special rounding */
1599     mx= ff_h263_round_chroma(mx);
1600     my= ff_h263_round_chroma(my);
1601
1602     sx= mx & s_mask;
1603     sy= my & s_mask;
1604     src_x = s->mb_x*block_s + (mx >> (lowres+1));
1605     src_y = s->mb_y*block_s + (my >> (lowres+1));
1606
1607     offset = src_y * s->uvlinesize + src_x;
1608     ptr = ref_picture[1] + offset;
1609     if(s->flags&CODEC_FLAG_EMU_EDGE){
1610         if(   (unsigned)src_x > h_edge_pos - (!!sx) - block_s
1611            || (unsigned)src_y > v_edge_pos - (!!sy) - block_s){
1612             ff_emulated_edge_mc(s->edge_emu_buffer, ptr, s->uvlinesize, 9, 9, src_x, src_y, h_edge_pos, v_edge_pos);
1613             ptr= s->edge_emu_buffer;
1614             emu=1;
1615         }
1616     }
1617     sx= (sx << 2) >> lowres;
1618     sy= (sy << 2) >> lowres;
1619     pix_op[op_index](dest_cb, ptr, s->uvlinesize, block_s, sx, sy);
1620
1621     ptr = ref_picture[2] + offset;
1622     if(emu){
1623         ff_emulated_edge_mc(s->edge_emu_buffer, ptr, s->uvlinesize, 9, 9, src_x, src_y, h_edge_pos, v_edge_pos);
1624         ptr= s->edge_emu_buffer;
1625     }
1626     pix_op[op_index](dest_cr, ptr, s->uvlinesize, block_s, sx, sy);
1627 }
1628
1629 /**
1630  * motion compensation of a single macroblock
1631  * @param s context
1632  * @param dest_y luma destination pointer
1633  * @param dest_cb chroma cb/u destination pointer
1634  * @param dest_cr chroma cr/v destination pointer
1635  * @param dir direction (0->forward, 1->backward)
1636  * @param ref_picture array[3] of pointers to the 3 planes of the reference picture
1637  * @param pic_op halfpel motion compensation function (average or put normally)
1638  * the motion vectors are taken from s->mv and the MV type from s->mv_type
1639  */
1640 static inline void MPV_motion_lowres(MpegEncContext *s,
1641                               uint8_t *dest_y, uint8_t *dest_cb, uint8_t *dest_cr,
1642                               int dir, uint8_t **ref_picture,
1643                               h264_chroma_mc_func *pix_op)
1644 {
1645     int mx, my;
1646     int mb_x, mb_y, i;
1647     const int lowres= s->avctx->lowres;
1648     const int block_s= 8>>lowres;
1649
1650     mb_x = s->mb_x;
1651     mb_y = s->mb_y;
1652
1653     switch(s->mv_type) {
1654     case MV_TYPE_16X16:
1655         mpeg_motion_lowres(s, dest_y, dest_cb, dest_cr,
1656                     0, 0, 0,
1657                     ref_picture, pix_op,
1658                     s->mv[dir][0][0], s->mv[dir][0][1], 2*block_s, mb_y);
1659         break;
1660     case MV_TYPE_8X8:
1661         mx = 0;
1662         my = 0;
1663             for(i=0;i<4;i++) {
1664                 hpel_motion_lowres(s, dest_y + ((i & 1) + (i >> 1) * s->linesize)*block_s,
1665                             ref_picture[0], 0, 0,
1666                             (2*mb_x + (i & 1))*block_s, (2*mb_y + (i >>1))*block_s,
1667                             s->width, s->height, s->linesize,
1668                             s->h_edge_pos >> lowres, s->v_edge_pos >> lowres,
1669                             block_s, block_s, pix_op,
1670                             s->mv[dir][i][0], s->mv[dir][i][1]);
1671
1672                 mx += s->mv[dir][i][0];
1673                 my += s->mv[dir][i][1];
1674             }
1675
1676         if(!CONFIG_GRAY || !(s->flags&CODEC_FLAG_GRAY))
1677             chroma_4mv_motion_lowres(s, dest_cb, dest_cr, ref_picture, pix_op, mx, my);
1678         break;
1679     case MV_TYPE_FIELD:
1680         if (s->picture_structure == PICT_FRAME) {
1681             /* top field */
1682             mpeg_motion_lowres(s, dest_y, dest_cb, dest_cr,
1683                         1, 0, s->field_select[dir][0],
1684                         ref_picture, pix_op,
1685                         s->mv[dir][0][0], s->mv[dir][0][1], block_s, mb_y);
1686             /* bottom field */
1687             mpeg_motion_lowres(s, dest_y, dest_cb, dest_cr,
1688                         1, 1, s->field_select[dir][1],
1689                         ref_picture, pix_op,
1690                         s->mv[dir][1][0], s->mv[dir][1][1], block_s, mb_y);
1691         } else {
1692             if(s->picture_structure != s->field_select[dir][0] + 1 && s->pict_type != FF_B_TYPE && !s->first_field){
1693                 ref_picture= s->current_picture_ptr->data;
1694             }
1695
1696             mpeg_motion_lowres(s, dest_y, dest_cb, dest_cr,
1697                         0, 0, s->field_select[dir][0],
1698                         ref_picture, pix_op,
1699                         s->mv[dir][0][0], s->mv[dir][0][1], 2*block_s, mb_y>>1);
1700         }
1701         break;
1702     case MV_TYPE_16X8:
1703         for(i=0; i<2; i++){
1704             uint8_t ** ref2picture;
1705
1706             if(s->picture_structure == s->field_select[dir][i] + 1 || s->pict_type == FF_B_TYPE || s->first_field){
1707                 ref2picture= ref_picture;
1708             }else{
1709                 ref2picture= s->current_picture_ptr->data;
1710             }
1711
1712             mpeg_motion_lowres(s, dest_y, dest_cb, dest_cr,
1713                         0, 0, s->field_select[dir][i],
1714                         ref2picture, pix_op,
1715                         s->mv[dir][i][0], s->mv[dir][i][1] + 2*block_s*i, block_s, mb_y>>1);
1716
1717             dest_y += 2*block_s*s->linesize;
1718             dest_cb+= (2*block_s>>s->chroma_y_shift)*s->uvlinesize;
1719             dest_cr+= (2*block_s>>s->chroma_y_shift)*s->uvlinesize;
1720         }
1721         break;
1722     case MV_TYPE_DMV:
1723         if(s->picture_structure == PICT_FRAME){
1724             for(i=0; i<2; i++){
1725                 int j;
1726                 for(j=0; j<2; j++){
1727                     mpeg_motion_lowres(s, dest_y, dest_cb, dest_cr,
1728                                 1, j, j^i,
1729                                 ref_picture, pix_op,
1730                                 s->mv[dir][2*i + j][0], s->mv[dir][2*i + j][1], block_s, mb_y);
1731                 }
1732                 pix_op = s->dsp.avg_h264_chroma_pixels_tab;
1733             }
1734         }else{
1735             for(i=0; i<2; i++){
1736                 mpeg_motion_lowres(s, dest_y, dest_cb, dest_cr,
1737                             0, 0, s->picture_structure != i+1,
1738                             ref_picture, pix_op,
1739                             s->mv[dir][2*i][0],s->mv[dir][2*i][1],2*block_s, mb_y>>1);
1740
1741                 // after put we make avg of the same block
1742                 pix_op = s->dsp.avg_h264_chroma_pixels_tab;
1743
1744                 //opposite parity is always in the same frame if this is second field
1745                 if(!s->first_field){
1746                     ref_picture = s->current_picture_ptr->data;
1747                 }
1748             }
1749         }
1750     break;
1751     default: assert(0);
1752     }
1753 }
1754
1755 /* put block[] to dest[] */
1756 static inline void put_dct(MpegEncContext *s,
1757                            DCTELEM *block, int i, uint8_t *dest, int line_size, int qscale)
1758 {
1759     s->dct_unquantize_intra(s, block, i, qscale);
1760     s->dsp.idct_put (dest, line_size, block);
1761 }
1762
1763 /* add block[] to dest[] */
1764 static inline void add_dct(MpegEncContext *s,
1765                            DCTELEM *block, int i, uint8_t *dest, int line_size)
1766 {
1767     if (s->block_last_index[i] >= 0) {
1768         s->dsp.idct_add (dest, line_size, block);
1769     }
1770 }
1771
1772 static inline void add_dequant_dct(MpegEncContext *s,
1773                            DCTELEM *block, int i, uint8_t *dest, int line_size, int qscale)
1774 {
1775     if (s->block_last_index[i] >= 0) {
1776         s->dct_unquantize_inter(s, block, i, qscale);
1777
1778         s->dsp.idct_add (dest, line_size, block);
1779     }
1780 }
1781
1782 /**
1783  * cleans dc, ac, coded_block for the current non intra MB
1784  */
1785 void ff_clean_intra_table_entries(MpegEncContext *s)
1786 {
1787     int wrap = s->b8_stride;
1788     int xy = s->block_index[0];
1789
1790     s->dc_val[0][xy           ] =
1791     s->dc_val[0][xy + 1       ] =
1792     s->dc_val[0][xy     + wrap] =
1793     s->dc_val[0][xy + 1 + wrap] = 1024;
1794     /* ac pred */
1795     memset(s->ac_val[0][xy       ], 0, 32 * sizeof(int16_t));
1796     memset(s->ac_val[0][xy + wrap], 0, 32 * sizeof(int16_t));
1797     if (s->msmpeg4_version>=3) {
1798         s->coded_block[xy           ] =
1799         s->coded_block[xy + 1       ] =
1800         s->coded_block[xy     + wrap] =
1801         s->coded_block[xy + 1 + wrap] = 0;
1802     }
1803     /* chroma */
1804     wrap = s->mb_stride;
1805     xy = s->mb_x + s->mb_y * wrap;
1806     s->dc_val[1][xy] =
1807     s->dc_val[2][xy] = 1024;
1808     /* ac pred */
1809     memset(s->ac_val[1][xy], 0, 16 * sizeof(int16_t));
1810     memset(s->ac_val[2][xy], 0, 16 * sizeof(int16_t));
1811
1812     s->mbintra_table[xy]= 0;
1813 }
1814
1815 /* generic function called after a macroblock has been parsed by the
1816    decoder or after it has been encoded by the encoder.
1817
1818    Important variables used:
1819    s->mb_intra : true if intra macroblock
1820    s->mv_dir   : motion vector direction
1821    s->mv_type  : motion vector type
1822    s->mv       : motion vector
1823    s->interlaced_dct : true if interlaced dct used (mpeg2)
1824  */
1825 static av_always_inline
1826 void MPV_decode_mb_internal(MpegEncContext *s, DCTELEM block[12][64],
1827                             int lowres_flag, int is_mpeg12)
1828 {
1829     const int mb_xy = s->mb_y * s->mb_stride + s->mb_x;
1830     if(CONFIG_MPEG_XVMC_DECODER && s->avctx->xvmc_acceleration){
1831         ff_xvmc_decode_mb(s);//xvmc uses pblocks
1832         return;
1833     }
1834
1835     if(s->avctx->debug&FF_DEBUG_DCT_COEFF) {
1836        /* save DCT coefficients */
1837        int i,j;
1838        DCTELEM *dct = &s->current_picture.dct_coeff[mb_xy*64*6];
1839        for(i=0; i<6; i++)
1840            for(j=0; j<64; j++)
1841                *dct++ = block[i][s->dsp.idct_permutation[j]];
1842     }
1843
1844     s->current_picture.qscale_table[mb_xy]= s->qscale;
1845
1846     /* update DC predictors for P macroblocks */
1847     if (!s->mb_intra) {
1848         if (!is_mpeg12 && (s->h263_pred || s->h263_aic)) {
1849             if(s->mbintra_table[mb_xy])
1850                 ff_clean_intra_table_entries(s);
1851         } else {
1852             s->last_dc[0] =
1853             s->last_dc[1] =
1854             s->last_dc[2] = 128 << s->intra_dc_precision;
1855         }
1856     }
1857     else if (!is_mpeg12 && (s->h263_pred || s->h263_aic))
1858         s->mbintra_table[mb_xy]=1;
1859
1860     if ((s->flags&CODEC_FLAG_PSNR) || !(s->encoding && (s->intra_only || s->pict_type==FF_B_TYPE) && s->avctx->mb_decision != FF_MB_DECISION_RD)) { //FIXME precalc
1861         uint8_t *dest_y, *dest_cb, *dest_cr;
1862         int dct_linesize, dct_offset;
1863         op_pixels_func (*op_pix)[4];
1864         qpel_mc_func (*op_qpix)[16];
1865         const int linesize= s->current_picture.linesize[0]; //not s->linesize as this would be wrong for field pics
1866         const int uvlinesize= s->current_picture.linesize[1];
1867         const int readable= s->pict_type != FF_B_TYPE || s->encoding || s->avctx->draw_horiz_band || lowres_flag;
1868         const int block_size= lowres_flag ? 8>>s->avctx->lowres : 8;
1869
1870         /* avoid copy if macroblock skipped in last frame too */
1871         /* skip only during decoding as we might trash the buffers during encoding a bit */
1872         if(!s->encoding){
1873             uint8_t *mbskip_ptr = &s->mbskip_table[mb_xy];
1874             const int age= s->current_picture.age;
1875
1876             assert(age);
1877
1878             if (s->mb_skipped) {
1879                 s->mb_skipped= 0;
1880                 assert(s->pict_type!=FF_I_TYPE);
1881
1882                 (*mbskip_ptr) ++; /* indicate that this time we skipped it */
1883                 if(*mbskip_ptr >99) *mbskip_ptr= 99;
1884
1885                 /* if previous was skipped too, then nothing to do !  */
1886                 if (*mbskip_ptr >= age && s->current_picture.reference){
1887                     return;
1888                 }
1889             } else if(!s->current_picture.reference){
1890                 (*mbskip_ptr) ++; /* increase counter so the age can be compared cleanly */
1891                 if(*mbskip_ptr >99) *mbskip_ptr= 99;
1892             } else{
1893                 *mbskip_ptr = 0; /* not skipped */
1894             }
1895         }
1896
1897         dct_linesize = linesize << s->interlaced_dct;
1898         dct_offset =(s->interlaced_dct)? linesize : linesize*block_size;
1899
1900         if(readable){
1901             dest_y=  s->dest[0];
1902             dest_cb= s->dest[1];
1903             dest_cr= s->dest[2];
1904         }else{
1905             dest_y = s->b_scratchpad;
1906             dest_cb= s->b_scratchpad+16*linesize;
1907             dest_cr= s->b_scratchpad+32*linesize;
1908         }
1909
1910         if (!s->mb_intra) {
1911             /* motion handling */
1912             /* decoding or more than one mb_type (MC was already done otherwise) */
1913             if(!s->encoding){
1914                 if(lowres_flag){
1915                     h264_chroma_mc_func *op_pix = s->dsp.put_h264_chroma_pixels_tab;
1916
1917                     if (s->mv_dir & MV_DIR_FORWARD) {
1918                         MPV_motion_lowres(s, dest_y, dest_cb, dest_cr, 0, s->last_picture.data, op_pix);
1919                         op_pix = s->dsp.avg_h264_chroma_pixels_tab;
1920                     }
1921                     if (s->mv_dir & MV_DIR_BACKWARD) {
1922                         MPV_motion_lowres(s, dest_y, dest_cb, dest_cr, 1, s->next_picture.data, op_pix);
1923                     }
1924                 }else{
1925                     op_qpix= s->me.qpel_put;
1926                     if ((!s->no_rounding) || s->pict_type==FF_B_TYPE){
1927                         op_pix = s->dsp.put_pixels_tab;
1928                     }else{
1929                         op_pix = s->dsp.put_no_rnd_pixels_tab;
1930                     }
1931                     if (s->mv_dir & MV_DIR_FORWARD) {
1932                         MPV_motion(s, dest_y, dest_cb, dest_cr, 0, s->last_picture.data, op_pix, op_qpix);
1933                         op_pix = s->dsp.avg_pixels_tab;
1934                         op_qpix= s->me.qpel_avg;
1935                     }
1936                     if (s->mv_dir & MV_DIR_BACKWARD) {
1937                         MPV_motion(s, dest_y, dest_cb, dest_cr, 1, s->next_picture.data, op_pix, op_qpix);
1938                     }
1939                 }
1940             }
1941
1942             /* skip dequant / idct if we are really late ;) */
1943             if(s->hurry_up>1) goto skip_idct;
1944             if(s->avctx->skip_idct){
1945                 if(  (s->avctx->skip_idct >= AVDISCARD_NONREF && s->pict_type == FF_B_TYPE)
1946                    ||(s->avctx->skip_idct >= AVDISCARD_NONKEY && s->pict_type != FF_I_TYPE)
1947                    || s->avctx->skip_idct >= AVDISCARD_ALL)
1948                     goto skip_idct;
1949             }
1950
1951             /* add dct residue */
1952             if(s->encoding || !(   s->h263_msmpeg4 || s->codec_id==CODEC_ID_MPEG1VIDEO || s->codec_id==CODEC_ID_MPEG2VIDEO
1953                                 || (s->codec_id==CODEC_ID_MPEG4 && !s->mpeg_quant))){
1954                 add_dequant_dct(s, block[0], 0, dest_y                          , dct_linesize, s->qscale);
1955                 add_dequant_dct(s, block[1], 1, dest_y              + block_size, dct_linesize, s->qscale);
1956                 add_dequant_dct(s, block[2], 2, dest_y + dct_offset             , dct_linesize, s->qscale);
1957                 add_dequant_dct(s, block[3], 3, dest_y + dct_offset + block_size, dct_linesize, s->qscale);
1958
1959                 if(!CONFIG_GRAY || !(s->flags&CODEC_FLAG_GRAY)){
1960                     if (s->chroma_y_shift){
1961                         add_dequant_dct(s, block[4], 4, dest_cb, uvlinesize, s->chroma_qscale);
1962                         add_dequant_dct(s, block[5], 5, dest_cr, uvlinesize, s->chroma_qscale);
1963                     }else{
1964                         dct_linesize >>= 1;
1965                         dct_offset >>=1;
1966                         add_dequant_dct(s, block[4], 4, dest_cb,              dct_linesize, s->chroma_qscale);
1967                         add_dequant_dct(s, block[5], 5, dest_cr,              dct_linesize, s->chroma_qscale);
1968                         add_dequant_dct(s, block[6], 6, dest_cb + dct_offset, dct_linesize, s->chroma_qscale);
1969                         add_dequant_dct(s, block[7], 7, dest_cr + dct_offset, dct_linesize, s->chroma_qscale);
1970                     }
1971                 }
1972             } else if(is_mpeg12 || (s->codec_id != CODEC_ID_WMV2)){
1973                 add_dct(s, block[0], 0, dest_y                          , dct_linesize);
1974                 add_dct(s, block[1], 1, dest_y              + block_size, dct_linesize);
1975                 add_dct(s, block[2], 2, dest_y + dct_offset             , dct_linesize);
1976                 add_dct(s, block[3], 3, dest_y + dct_offset + block_size, dct_linesize);
1977
1978                 if(!CONFIG_GRAY || !(s->flags&CODEC_FLAG_GRAY)){
1979                     if(s->chroma_y_shift){//Chroma420
1980                         add_dct(s, block[4], 4, dest_cb, uvlinesize);
1981                         add_dct(s, block[5], 5, dest_cr, uvlinesize);
1982                     }else{
1983                         //chroma422
1984                         dct_linesize = uvlinesize << s->interlaced_dct;
1985                         dct_offset =(s->interlaced_dct)? uvlinesize : uvlinesize*8;
1986
1987                         add_dct(s, block[4], 4, dest_cb, dct_linesize);
1988                         add_dct(s, block[5], 5, dest_cr, dct_linesize);
1989                         add_dct(s, block[6], 6, dest_cb+dct_offset, dct_linesize);
1990                         add_dct(s, block[7], 7, dest_cr+dct_offset, dct_linesize);
1991                         if(!s->chroma_x_shift){//Chroma444
1992                             add_dct(s, block[8], 8, dest_cb+8, dct_linesize);
1993                             add_dct(s, block[9], 9, dest_cr+8, dct_linesize);
1994                             add_dct(s, block[10], 10, dest_cb+8+dct_offset, dct_linesize);
1995                             add_dct(s, block[11], 11, dest_cr+8+dct_offset, dct_linesize);
1996                         }
1997                     }
1998                 }//fi gray
1999             }
2000             else if (CONFIG_WMV2_DECODER || CONFIG_WMV2_ENCODER) {
2001                 ff_wmv2_add_mb(s, block, dest_y, dest_cb, dest_cr);
2002             }
2003         } else {
2004             /* dct only in intra block */
2005             if(s->encoding || !(s->codec_id==CODEC_ID_MPEG1VIDEO || s->codec_id==CODEC_ID_MPEG2VIDEO)){
2006                 put_dct(s, block[0], 0, dest_y                          , dct_linesize, s->qscale);
2007                 put_dct(s, block[1], 1, dest_y              + block_size, dct_linesize, s->qscale);
2008                 put_dct(s, block[2], 2, dest_y + dct_offset             , dct_linesize, s->qscale);
2009                 put_dct(s, block[3], 3, dest_y + dct_offset + block_size, dct_linesize, s->qscale);
2010
2011                 if(!CONFIG_GRAY || !(s->flags&CODEC_FLAG_GRAY)){
2012                     if(s->chroma_y_shift){
2013                         put_dct(s, block[4], 4, dest_cb, uvlinesize, s->chroma_qscale);
2014                         put_dct(s, block[5], 5, dest_cr, uvlinesize, s->chroma_qscale);
2015                     }else{
2016                         dct_offset >>=1;
2017                         dct_linesize >>=1;
2018                         put_dct(s, block[4], 4, dest_cb,              dct_linesize, s->chroma_qscale);
2019                         put_dct(s, block[5], 5, dest_cr,              dct_linesize, s->chroma_qscale);
2020                         put_dct(s, block[6], 6, dest_cb + dct_offset, dct_linesize, s->chroma_qscale);
2021                         put_dct(s, block[7], 7, dest_cr + dct_offset, dct_linesize, s->chroma_qscale);
2022                     }
2023                 }
2024             }else{
2025                 s->dsp.idct_put(dest_y                          , dct_linesize, block[0]);
2026                 s->dsp.idct_put(dest_y              + block_size, dct_linesize, block[1]);
2027                 s->dsp.idct_put(dest_y + dct_offset             , dct_linesize, block[2]);
2028                 s->dsp.idct_put(dest_y + dct_offset + block_size, dct_linesize, block[3]);
2029
2030                 if(!CONFIG_GRAY || !(s->flags&CODEC_FLAG_GRAY)){
2031                     if(s->chroma_y_shift){
2032                         s->dsp.idct_put(dest_cb, uvlinesize, block[4]);
2033                         s->dsp.idct_put(dest_cr, uvlinesize, block[5]);
2034                     }else{
2035
2036                         dct_linesize = uvlinesize << s->interlaced_dct;
2037                         dct_offset =(s->interlaced_dct)? uvlinesize : uvlinesize*8;
2038
2039                         s->dsp.idct_put(dest_cb,              dct_linesize, block[4]);
2040                         s->dsp.idct_put(dest_cr,              dct_linesize, block[5]);
2041                         s->dsp.idct_put(dest_cb + dct_offset, dct_linesize, block[6]);
2042                         s->dsp.idct_put(dest_cr + dct_offset, dct_linesize, block[7]);
2043                         if(!s->chroma_x_shift){//Chroma444
2044                             s->dsp.idct_put(dest_cb + 8,              dct_linesize, block[8]);
2045                             s->dsp.idct_put(dest_cr + 8,              dct_linesize, block[9]);
2046                             s->dsp.idct_put(dest_cb + 8 + dct_offset, dct_linesize, block[10]);
2047                             s->dsp.idct_put(dest_cr + 8 + dct_offset, dct_linesize, block[11]);
2048                         }
2049                     }
2050                 }//gray
2051             }
2052         }
2053 skip_idct:
2054         if(!readable){
2055             s->dsp.put_pixels_tab[0][0](s->dest[0], dest_y ,   linesize,16);
2056             s->dsp.put_pixels_tab[s->chroma_x_shift][0](s->dest[1], dest_cb, uvlinesize,16 >> s->chroma_y_shift);
2057             s->dsp.put_pixels_tab[s->chroma_x_shift][0](s->dest[2], dest_cr, uvlinesize,16 >> s->chroma_y_shift);
2058         }
2059     }
2060 }
2061
2062 void MPV_decode_mb(MpegEncContext *s, DCTELEM block[12][64]){
2063 #if !CONFIG_SMALL
2064     if(s->out_format == FMT_MPEG1) {
2065         if(s->avctx->lowres) MPV_decode_mb_internal(s, block, 1, 1);
2066         else                 MPV_decode_mb_internal(s, block, 0, 1);
2067     } else
2068 #endif
2069     if(s->avctx->lowres) MPV_decode_mb_internal(s, block, 1, 0);
2070     else                  MPV_decode_mb_internal(s, block, 0, 0);
2071 }
2072
2073 /**
2074  *
2075  * @param h is the normal height, this will be reduced automatically if needed for the last row
2076  */
2077 void ff_draw_horiz_band(MpegEncContext *s, int y, int h){
2078     if (s->avctx->draw_horiz_band) {
2079         AVFrame *src;
2080         const int field_pic= s->picture_structure != PICT_FRAME;
2081         int offset[4];
2082
2083         h= FFMIN(h, (s->avctx->height>>field_pic) - y);
2084
2085         if(field_pic && !(s->avctx->slice_flags&SLICE_FLAG_ALLOW_FIELD)){
2086             h <<= 1;
2087             y <<= 1;
2088             if(s->first_field) return;
2089         }
2090
2091         if(s->pict_type==FF_B_TYPE || s->low_delay || (s->avctx->slice_flags&SLICE_FLAG_CODED_ORDER))
2092             src= (AVFrame*)s->current_picture_ptr;
2093         else if(s->last_picture_ptr)
2094             src= (AVFrame*)s->last_picture_ptr;
2095         else
2096             return;
2097
2098         if(s->pict_type==FF_B_TYPE && s->picture_structure == PICT_FRAME && s->out_format != FMT_H264){
2099             offset[0]=
2100             offset[1]=
2101             offset[2]=
2102             offset[3]= 0;
2103         }else{
2104             offset[0]= y * s->linesize;
2105             offset[1]=
2106             offset[2]= (y >> s->chroma_y_shift) * s->uvlinesize;
2107             offset[3]= 0;
2108         }
2109
2110         emms_c();
2111
2112         s->avctx->draw_horiz_band(s->avctx, src, offset,
2113                                   y, s->picture_structure, h);
2114     }
2115 }
2116
2117 void ff_init_block_index(MpegEncContext *s){ //FIXME maybe rename
2118     const int linesize= s->current_picture.linesize[0]; //not s->linesize as this would be wrong for field pics
2119     const int uvlinesize= s->current_picture.linesize[1];
2120     const int mb_size= 4 - s->avctx->lowres;
2121
2122     s->block_index[0]= s->b8_stride*(s->mb_y*2    ) - 2 + s->mb_x*2;
2123     s->block_index[1]= s->b8_stride*(s->mb_y*2    ) - 1 + s->mb_x*2;
2124     s->block_index[2]= s->b8_stride*(s->mb_y*2 + 1) - 2 + s->mb_x*2;
2125     s->block_index[3]= s->b8_stride*(s->mb_y*2 + 1) - 1 + s->mb_x*2;
2126     s->block_index[4]= s->mb_stride*(s->mb_y + 1)                + s->b8_stride*s->mb_height*2 + s->mb_x - 1;
2127     s->block_index[5]= s->mb_stride*(s->mb_y + s->mb_height + 2) + s->b8_stride*s->mb_height*2 + s->mb_x - 1;
2128     //block_index is not used by mpeg2, so it is not affected by chroma_format
2129
2130     s->dest[0] = s->current_picture.data[0] + ((s->mb_x - 1) << mb_size);
2131     s->dest[1] = s->current_picture.data[1] + ((s->mb_x - 1) << (mb_size - s->chroma_x_shift));
2132     s->dest[2] = s->current_picture.data[2] + ((s->mb_x - 1) << (mb_size - s->chroma_x_shift));
2133
2134     if(!(s->pict_type==FF_B_TYPE && s->avctx->draw_horiz_band && s->picture_structure==PICT_FRAME))
2135     {
2136         if(s->picture_structure==PICT_FRAME){
2137         s->dest[0] += s->mb_y *   linesize << mb_size;
2138         s->dest[1] += s->mb_y * uvlinesize << (mb_size - s->chroma_y_shift);
2139         s->dest[2] += s->mb_y * uvlinesize << (mb_size - s->chroma_y_shift);
2140         }else{
2141             s->dest[0] += (s->mb_y>>1) *   linesize << mb_size;
2142             s->dest[1] += (s->mb_y>>1) * uvlinesize << (mb_size - s->chroma_y_shift);
2143             s->dest[2] += (s->mb_y>>1) * uvlinesize << (mb_size - s->chroma_y_shift);
2144             assert((s->mb_y&1) == (s->picture_structure == PICT_BOTTOM_FIELD));
2145         }
2146     }
2147 }
2148
2149 void ff_mpeg_flush(AVCodecContext *avctx){
2150     int i;
2151     MpegEncContext *s = avctx->priv_data;
2152
2153     if(s==NULL || s->picture==NULL)
2154         return;
2155
2156     for(i=0; i<MAX_PICTURE_COUNT; i++){
2157        if(s->picture[i].data[0] && (   s->picture[i].type == FF_BUFFER_TYPE_INTERNAL
2158                                     || s->picture[i].type == FF_BUFFER_TYPE_USER))
2159         free_frame_buffer(s, &s->picture[i]);
2160     }
2161     s->current_picture_ptr = s->last_picture_ptr = s->next_picture_ptr = NULL;
2162
2163     s->mb_x= s->mb_y= 0;
2164     s->closed_gop= 0;
2165
2166     s->parse_context.state= -1;
2167     s->parse_context.frame_start_found= 0;
2168     s->parse_context.overread= 0;
2169     s->parse_context.overread_index= 0;
2170     s->parse_context.index= 0;
2171     s->parse_context.last_index= 0;
2172     s->bitstream_buffer_size=0;
2173     s->pp_time=0;
2174 }
2175
2176 static void dct_unquantize_mpeg1_intra_c(MpegEncContext *s,
2177                                    DCTELEM *block, int n, int qscale)
2178 {
2179     int i, level, nCoeffs;
2180     const uint16_t *quant_matrix;
2181
2182     nCoeffs= s->block_last_index[n];
2183
2184     if (n < 4)
2185         block[0] = block[0] * s->y_dc_scale;
2186     else
2187         block[0] = block[0] * s->c_dc_scale;
2188     /* XXX: only mpeg1 */
2189     quant_matrix = s->intra_matrix;
2190     for(i=1;i<=nCoeffs;i++) {
2191         int j= s->intra_scantable.permutated[i];
2192         level = block[j];
2193         if (level) {
2194             if (level < 0) {
2195                 level = -level;
2196                 level = (int)(level * qscale * quant_matrix[j]) >> 3;
2197                 level = (level - 1) | 1;
2198                 level = -level;
2199             } else {
2200                 level = (int)(level * qscale * quant_matrix[j]) >> 3;
2201                 level = (level - 1) | 1;
2202             }
2203             block[j] = level;
2204         }
2205     }
2206 }
2207
2208 static void dct_unquantize_mpeg1_inter_c(MpegEncContext *s,
2209                                    DCTELEM *block, int n, int qscale)
2210 {
2211     int i, level, nCoeffs;
2212     const uint16_t *quant_matrix;
2213
2214     nCoeffs= s->block_last_index[n];
2215
2216     quant_matrix = s->inter_matrix;
2217     for(i=0; i<=nCoeffs; i++) {
2218         int j= s->intra_scantable.permutated[i];
2219         level = block[j];
2220         if (level) {
2221             if (level < 0) {
2222                 level = -level;
2223                 level = (((level << 1) + 1) * qscale *
2224                          ((int) (quant_matrix[j]))) >> 4;
2225                 level = (level - 1) | 1;
2226                 level = -level;
2227             } else {
2228                 level = (((level << 1) + 1) * qscale *
2229                          ((int) (quant_matrix[j]))) >> 4;
2230                 level = (level - 1) | 1;
2231             }
2232             block[j] = level;
2233         }
2234     }
2235 }
2236
2237 static void dct_unquantize_mpeg2_intra_c(MpegEncContext *s,
2238                                    DCTELEM *block, int n, int qscale)
2239 {
2240     int i, level, nCoeffs;
2241     const uint16_t *quant_matrix;
2242
2243     if(s->alternate_scan) nCoeffs= 63;
2244     else nCoeffs= s->block_last_index[n];
2245
2246     if (n < 4)
2247         block[0] = block[0] * s->y_dc_scale;
2248     else
2249         block[0] = block[0] * s->c_dc_scale;
2250     quant_matrix = s->intra_matrix;
2251     for(i=1;i<=nCoeffs;i++) {
2252         int j= s->intra_scantable.permutated[i];
2253         level = block[j];
2254         if (level) {
2255             if (level < 0) {
2256                 level = -level;
2257                 level = (int)(level * qscale * quant_matrix[j]) >> 3;
2258                 level = -level;
2259             } else {
2260                 level = (int)(level * qscale * quant_matrix[j]) >> 3;
2261             }
2262             block[j] = level;
2263         }
2264     }
2265 }
2266
2267 static void dct_unquantize_mpeg2_intra_bitexact(MpegEncContext *s,
2268                                    DCTELEM *block, int n, int qscale)
2269 {
2270     int i, level, nCoeffs;
2271     const uint16_t *quant_matrix;
2272     int sum=-1;
2273
2274     if(s->alternate_scan) nCoeffs= 63;
2275     else nCoeffs= s->block_last_index[n];
2276
2277     if (n < 4)
2278         block[0] = block[0] * s->y_dc_scale;
2279     else
2280         block[0] = block[0] * s->c_dc_scale;
2281     quant_matrix = s->intra_matrix;
2282     for(i=1;i<=nCoeffs;i++) {
2283         int j= s->intra_scantable.permutated[i];
2284         level = block[j];
2285         if (level) {
2286             if (level < 0) {
2287                 level = -level;
2288                 level = (int)(level * qscale * quant_matrix[j]) >> 3;
2289                 level = -level;
2290             } else {
2291                 level = (int)(level * qscale * quant_matrix[j]) >> 3;
2292             }
2293             block[j] = level;
2294             sum+=level;
2295         }
2296     }
2297     block[63]^=sum&1;
2298 }
2299
2300 static void dct_unquantize_mpeg2_inter_c(MpegEncContext *s,
2301                                    DCTELEM *block, int n, int qscale)
2302 {
2303     int i, level, nCoeffs;
2304     const uint16_t *quant_matrix;
2305     int sum=-1;
2306
2307     if(s->alternate_scan) nCoeffs= 63;
2308     else nCoeffs= s->block_last_index[n];
2309
2310     quant_matrix = s->inter_matrix;
2311     for(i=0; i<=nCoeffs; i++) {
2312         int j= s->intra_scantable.permutated[i];
2313         level = block[j];
2314         if (level) {
2315             if (level < 0) {
2316                 level = -level;
2317                 level = (((level << 1) + 1) * qscale *
2318                          ((int) (quant_matrix[j]))) >> 4;
2319                 level = -level;
2320             } else {
2321                 level = (((level << 1) + 1) * qscale *
2322                          ((int) (quant_matrix[j]))) >> 4;
2323             }
2324             block[j] = level;
2325             sum+=level;
2326         }
2327     }
2328     block[63]^=sum&1;
2329 }
2330
2331 static void dct_unquantize_h263_intra_c(MpegEncContext *s,
2332                                   DCTELEM *block, int n, int qscale)
2333 {
2334     int i, level, qmul, qadd;
2335     int nCoeffs;
2336
2337     assert(s->block_last_index[n]>=0);
2338
2339     qmul = qscale << 1;
2340
2341     if (!s->h263_aic) {
2342         if (n < 4)
2343             block[0] = block[0] * s->y_dc_scale;
2344         else
2345             block[0] = block[0] * s->c_dc_scale;
2346         qadd = (qscale - 1) | 1;
2347     }else{
2348         qadd = 0;
2349     }
2350     if(s->ac_pred)
2351         nCoeffs=63;
2352     else
2353         nCoeffs= s->inter_scantable.raster_end[ s->block_last_index[n] ];
2354
2355     for(i=1; i<=nCoeffs; i++) {
2356         level = block[i];
2357         if (level) {
2358             if (level < 0) {
2359                 level = level * qmul - qadd;
2360             } else {
2361                 level = level * qmul + qadd;
2362             }
2363             block[i] = level;
2364         }
2365     }
2366 }
2367
2368 static void dct_unquantize_h263_inter_c(MpegEncContext *s,
2369                                   DCTELEM *block, int n, int qscale)
2370 {
2371     int i, level, qmul, qadd;
2372     int nCoeffs;
2373
2374     assert(s->block_last_index[n]>=0);
2375
2376     qadd = (qscale - 1) | 1;
2377     qmul = qscale << 1;
2378
2379     nCoeffs= s->inter_scantable.raster_end[ s->block_last_index[n] ];
2380
2381     for(i=0; i<=nCoeffs; i++) {
2382         level = block[i];
2383         if (level) {
2384             if (level < 0) {
2385                 level = level * qmul - qadd;
2386             } else {
2387                 level = level * qmul + qadd;
2388             }
2389             block[i] = level;
2390         }
2391     }
2392 }
2393
2394 /**
2395  * set qscale and update qscale dependent variables.
2396  */
2397 void ff_set_qscale(MpegEncContext * s, int qscale)
2398 {
2399     if (qscale < 1)
2400         qscale = 1;
2401     else if (qscale > 31)
2402         qscale = 31;
2403
2404     s->qscale = qscale;
2405     s->chroma_qscale= s->chroma_qscale_table[qscale];
2406
2407     s->y_dc_scale= s->y_dc_scale_table[ qscale ];
2408     s->c_dc_scale= s->c_dc_scale_table[ s->chroma_qscale ];
2409 }