1c5ff27e5fd6de14b4ffca0a8857c76c79643df5
[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 Libav.
9  *
10  * Libav 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  * Libav 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 Libav; 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 "libavutil/imgutils.h"
32 #include "avcodec.h"
33 #include "dsputil.h"
34 #include "internal.h"
35 #include "mpegvideo.h"
36 #include "mpegvideo_common.h"
37 #include "mjpegenc.h"
38 #include "msmpeg4.h"
39 #include "faandct.h"
40 #include "xvmc_internal.h"
41 #include <limits.h>
42
43 //#undef NDEBUG
44 //#include <assert.h>
45
46 static void dct_unquantize_mpeg1_intra_c(MpegEncContext *s,
47                                    DCTELEM *block, int n, int qscale);
48 static void dct_unquantize_mpeg1_inter_c(MpegEncContext *s,
49                                    DCTELEM *block, int n, int qscale);
50 static void dct_unquantize_mpeg2_intra_c(MpegEncContext *s,
51                                    DCTELEM *block, int n, int qscale);
52 static void dct_unquantize_mpeg2_intra_bitexact(MpegEncContext *s,
53                                    DCTELEM *block, int n, int qscale);
54 static void dct_unquantize_mpeg2_inter_c(MpegEncContext *s,
55                                    DCTELEM *block, int n, int qscale);
56 static void dct_unquantize_h263_intra_c(MpegEncContext *s,
57                                   DCTELEM *block, int n, int qscale);
58 static void dct_unquantize_h263_inter_c(MpegEncContext *s,
59                                   DCTELEM *block, int n, int qscale);
60
61
62 /* enable all paranoid tests for rounding, overflows, etc... */
63 //#define PARANOID
64
65 //#define DEBUG
66
67
68 static const uint8_t ff_default_chroma_qscale_table[32]={
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     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
71 };
72
73 const uint8_t ff_mpeg1_dc_scale_table[128]={
74 //  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
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     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,
79 };
80
81 static const uint8_t mpeg2_dc_scale_table1[128]={
82 //  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
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     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,
87 };
88
89 static const uint8_t mpeg2_dc_scale_table2[128]={
90 //  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
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     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,
95 };
96
97 static const uint8_t mpeg2_dc_scale_table3[128]={
98 //  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
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     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,
103 };
104
105 const uint8_t * const ff_mpeg2_dc_scale_table[4]={
106     ff_mpeg1_dc_scale_table,
107     mpeg2_dc_scale_table1,
108     mpeg2_dc_scale_table2,
109     mpeg2_dc_scale_table3,
110 };
111
112 const enum PixelFormat ff_pixfmt_list_420[] = {
113     PIX_FMT_YUV420P,
114     PIX_FMT_NONE
115 };
116
117 const enum PixelFormat ff_hwaccel_pixfmt_list_420[] = {
118     PIX_FMT_DXVA2_VLD,
119     PIX_FMT_VAAPI_VLD,
120     PIX_FMT_YUV420P,
121     PIX_FMT_NONE
122 };
123
124 const uint8_t *ff_find_start_code(const uint8_t * restrict p, const uint8_t *end, uint32_t * restrict state){
125     int i;
126
127     assert(p<=end);
128     if(p>=end)
129         return end;
130
131     for(i=0; i<3; i++){
132         uint32_t tmp= *state << 8;
133         *state= tmp + *(p++);
134         if(tmp == 0x100 || p==end)
135             return p;
136     }
137
138     while(p<end){
139         if     (p[-1] > 1      ) p+= 3;
140         else if(p[-2]          ) p+= 2;
141         else if(p[-3]|(p[-1]-1)) p++;
142         else{
143             p++;
144             break;
145         }
146     }
147
148     p= FFMIN(p, end)-4;
149     *state= AV_RB32(p);
150
151     return p+4;
152 }
153
154 /* init common dct for both encoder and decoder */
155 av_cold int ff_dct_common_init(MpegEncContext *s)
156 {
157     s->dct_unquantize_h263_intra = dct_unquantize_h263_intra_c;
158     s->dct_unquantize_h263_inter = dct_unquantize_h263_inter_c;
159     s->dct_unquantize_mpeg1_intra = dct_unquantize_mpeg1_intra_c;
160     s->dct_unquantize_mpeg1_inter = dct_unquantize_mpeg1_inter_c;
161     s->dct_unquantize_mpeg2_intra = dct_unquantize_mpeg2_intra_c;
162     if(s->flags & CODEC_FLAG_BITEXACT)
163         s->dct_unquantize_mpeg2_intra = dct_unquantize_mpeg2_intra_bitexact;
164     s->dct_unquantize_mpeg2_inter = dct_unquantize_mpeg2_inter_c;
165
166 #if   HAVE_MMX
167     MPV_common_init_mmx(s);
168 #elif ARCH_ALPHA
169     MPV_common_init_axp(s);
170 #elif CONFIG_MLIB
171     MPV_common_init_mlib(s);
172 #elif HAVE_MMI
173     MPV_common_init_mmi(s);
174 #elif ARCH_ARM
175     MPV_common_init_arm(s);
176 #elif HAVE_ALTIVEC
177     MPV_common_init_altivec(s);
178 #elif ARCH_BFIN
179     MPV_common_init_bfin(s);
180 #endif
181
182     /* load & permutate scantables
183        note: only wmv uses different ones
184     */
185     if(s->alternate_scan){
186         ff_init_scantable(s->dsp.idct_permutation, &s->inter_scantable  , ff_alternate_vertical_scan);
187         ff_init_scantable(s->dsp.idct_permutation, &s->intra_scantable  , ff_alternate_vertical_scan);
188     }else{
189         ff_init_scantable(s->dsp.idct_permutation, &s->inter_scantable  , ff_zigzag_direct);
190         ff_init_scantable(s->dsp.idct_permutation, &s->intra_scantable  , ff_zigzag_direct);
191     }
192     ff_init_scantable(s->dsp.idct_permutation, &s->intra_h_scantable, ff_alternate_horizontal_scan);
193     ff_init_scantable(s->dsp.idct_permutation, &s->intra_v_scantable, ff_alternate_vertical_scan);
194
195     return 0;
196 }
197
198 void ff_copy_picture(Picture *dst, Picture *src){
199     *dst = *src;
200     dst->type= FF_BUFFER_TYPE_COPY;
201 }
202
203 /**
204  * Release a frame buffer
205  */
206 static void free_frame_buffer(MpegEncContext *s, Picture *pic)
207 {
208     s->avctx->release_buffer(s->avctx, (AVFrame*)pic);
209     av_freep(&pic->hwaccel_picture_private);
210 }
211
212 /**
213  * Allocate a frame buffer
214  */
215 static int alloc_frame_buffer(MpegEncContext *s, Picture *pic)
216 {
217     int r;
218
219     if (s->avctx->hwaccel) {
220         assert(!pic->hwaccel_picture_private);
221         if (s->avctx->hwaccel->priv_data_size) {
222             pic->hwaccel_picture_private = av_mallocz(s->avctx->hwaccel->priv_data_size);
223             if (!pic->hwaccel_picture_private) {
224                 av_log(s->avctx, AV_LOG_ERROR, "alloc_frame_buffer() failed (hwaccel private data allocation)\n");
225                 return -1;
226             }
227         }
228     }
229
230     r = s->avctx->get_buffer(s->avctx, (AVFrame*)pic);
231
232     if (r<0 || !pic->age || !pic->type || !pic->data[0]) {
233         av_log(s->avctx, AV_LOG_ERROR, "get_buffer() failed (%d %d %d %p)\n", r, pic->age, pic->type, pic->data[0]);
234         av_freep(&pic->hwaccel_picture_private);
235         return -1;
236     }
237
238     if (s->linesize && (s->linesize != pic->linesize[0] || s->uvlinesize != pic->linesize[1])) {
239         av_log(s->avctx, AV_LOG_ERROR, "get_buffer() failed (stride changed)\n");
240         free_frame_buffer(s, pic);
241         return -1;
242     }
243
244     if (pic->linesize[1] != pic->linesize[2]) {
245         av_log(s->avctx, AV_LOG_ERROR, "get_buffer() failed (uv stride mismatch)\n");
246         free_frame_buffer(s, pic);
247         return -1;
248     }
249
250     return 0;
251 }
252
253 /**
254  * allocates a Picture
255  * The pixels are allocated/set by calling get_buffer() if shared=0
256  */
257 int ff_alloc_picture(MpegEncContext *s, Picture *pic, int shared){
258     const int big_mb_num= s->mb_stride*(s->mb_height+1) + 1; //the +1 is needed so memset(,,stride*height) does not sig11
259     const int mb_array_size= s->mb_stride*s->mb_height;
260     const int b8_array_size= s->b8_stride*s->mb_height*2;
261     const int b4_array_size= s->b4_stride*s->mb_height*4;
262     int i;
263     int r= -1;
264
265     if(shared){
266         assert(pic->data[0]);
267         assert(pic->type == 0 || pic->type == FF_BUFFER_TYPE_SHARED);
268         pic->type= FF_BUFFER_TYPE_SHARED;
269     }else{
270         assert(!pic->data[0]);
271
272         if (alloc_frame_buffer(s, pic) < 0)
273             return -1;
274
275         s->linesize  = pic->linesize[0];
276         s->uvlinesize= pic->linesize[1];
277     }
278
279     if(pic->qscale_table==NULL){
280         if (s->encoding) {
281             FF_ALLOCZ_OR_GOTO(s->avctx, pic->mb_var   , mb_array_size * sizeof(int16_t)  , fail)
282             FF_ALLOCZ_OR_GOTO(s->avctx, pic->mc_mb_var, mb_array_size * sizeof(int16_t)  , fail)
283             FF_ALLOCZ_OR_GOTO(s->avctx, pic->mb_mean  , mb_array_size * sizeof(int8_t )  , fail)
284         }
285
286         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
287         FF_ALLOCZ_OR_GOTO(s->avctx, pic->qscale_table , mb_array_size * sizeof(uint8_t)  , fail)
288         FF_ALLOCZ_OR_GOTO(s->avctx, pic->mb_type_base , (big_mb_num + s->mb_stride) * sizeof(uint32_t), fail)
289         pic->mb_type= pic->mb_type_base + 2*s->mb_stride+1;
290         if(s->out_format == FMT_H264){
291             for(i=0; i<2; i++){
292                 FF_ALLOCZ_OR_GOTO(s->avctx, pic->motion_val_base[i], 2 * (b4_array_size+4)  * sizeof(int16_t), fail)
293                 pic->motion_val[i]= pic->motion_val_base[i]+4;
294                 FF_ALLOCZ_OR_GOTO(s->avctx, pic->ref_index[i], 4*mb_array_size * sizeof(uint8_t), fail)
295             }
296             pic->motion_subsample_log2= 2;
297         }else if(s->out_format == FMT_H263 || s->encoding || (s->avctx->debug&FF_DEBUG_MV) || (s->avctx->debug_mv)){
298             for(i=0; i<2; i++){
299                 FF_ALLOCZ_OR_GOTO(s->avctx, pic->motion_val_base[i], 2 * (b8_array_size+4) * sizeof(int16_t), fail)
300                 pic->motion_val[i]= pic->motion_val_base[i]+4;
301                 FF_ALLOCZ_OR_GOTO(s->avctx, pic->ref_index[i], 4*mb_array_size * sizeof(uint8_t), fail)
302             }
303             pic->motion_subsample_log2= 3;
304         }
305         if(s->avctx->debug&FF_DEBUG_DCT_COEFF) {
306             FF_ALLOCZ_OR_GOTO(s->avctx, pic->dct_coeff, 64 * mb_array_size * sizeof(DCTELEM)*6, fail)
307         }
308         pic->qstride= s->mb_stride;
309         FF_ALLOCZ_OR_GOTO(s->avctx, pic->pan_scan , 1 * sizeof(AVPanScan), fail)
310     }
311
312     /* It might be nicer if the application would keep track of these
313      * but it would require an API change. */
314     memmove(s->prev_pict_types+1, s->prev_pict_types, PREV_PICT_TYPES_BUFFER_SIZE-1);
315     s->prev_pict_types[0]= s->dropable ? FF_B_TYPE : s->pict_type;
316     if(pic->age < PREV_PICT_TYPES_BUFFER_SIZE && s->prev_pict_types[pic->age] == FF_B_TYPE)
317         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.
318
319     return 0;
320 fail: //for the FF_ALLOCZ_OR_GOTO macro
321     if(r>=0)
322         free_frame_buffer(s, pic);
323     return -1;
324 }
325
326 /**
327  * deallocates a picture
328  */
329 static void free_picture(MpegEncContext *s, Picture *pic){
330     int i;
331
332     if(pic->data[0] && pic->type!=FF_BUFFER_TYPE_SHARED){
333         free_frame_buffer(s, pic);
334     }
335
336     av_freep(&pic->mb_var);
337     av_freep(&pic->mc_mb_var);
338     av_freep(&pic->mb_mean);
339     av_freep(&pic->mbskip_table);
340     av_freep(&pic->qscale_table);
341     av_freep(&pic->mb_type_base);
342     av_freep(&pic->dct_coeff);
343     av_freep(&pic->pan_scan);
344     pic->mb_type= NULL;
345     for(i=0; i<2; i++){
346         av_freep(&pic->motion_val_base[i]);
347         av_freep(&pic->ref_index[i]);
348     }
349
350     if(pic->type == FF_BUFFER_TYPE_SHARED){
351         for(i=0; i<4; i++){
352             pic->base[i]=
353             pic->data[i]= NULL;
354         }
355         pic->type= 0;
356     }
357 }
358
359 static int init_duplicate_context(MpegEncContext *s, MpegEncContext *base){
360     int y_size = s->b8_stride * (2 * s->mb_height + 1);
361     int c_size = s->mb_stride * (s->mb_height + 1);
362     int yc_size = y_size + 2 * c_size;
363     int i;
364
365     // edge emu needs blocksize + filter length - 1 (=17x17 for halfpel / 21x21 for h264)
366     FF_ALLOCZ_OR_GOTO(s->avctx, s->allocated_edge_emu_buffer, (s->width+64)*2*21*2, fail); //(width + edge + align)*interlaced*MBsize*tolerance
367     s->edge_emu_buffer= s->allocated_edge_emu_buffer + (s->width+64)*2*21;
368
369      //FIXME should be linesize instead of s->width*2 but that is not known before get_buffer()
370     FF_ALLOCZ_OR_GOTO(s->avctx, s->me.scratchpad,  (s->width+64)*4*16*2*sizeof(uint8_t), fail)
371     s->me.temp=         s->me.scratchpad;
372     s->rd_scratchpad=   s->me.scratchpad;
373     s->b_scratchpad=    s->me.scratchpad;
374     s->obmc_scratchpad= s->me.scratchpad + 16;
375     if (s->encoding) {
376         FF_ALLOCZ_OR_GOTO(s->avctx, s->me.map      , ME_MAP_SIZE*sizeof(uint32_t), fail)
377         FF_ALLOCZ_OR_GOTO(s->avctx, s->me.score_map, ME_MAP_SIZE*sizeof(uint32_t), fail)
378         if(s->avctx->noise_reduction){
379             FF_ALLOCZ_OR_GOTO(s->avctx, s->dct_error_sum, 2 * 64 * sizeof(int), fail)
380         }
381     }
382     FF_ALLOCZ_OR_GOTO(s->avctx, s->blocks, 64*12*2 * sizeof(DCTELEM), fail)
383     s->block= s->blocks[0];
384
385     for(i=0;i<12;i++){
386         s->pblocks[i] = &s->block[i];
387     }
388
389     if (s->out_format == FMT_H263) {
390         /* ac values */
391         FF_ALLOCZ_OR_GOTO(s->avctx, s->ac_val_base, yc_size * sizeof(int16_t) * 16, fail);
392         s->ac_val[0] = s->ac_val_base + s->b8_stride + 1;
393         s->ac_val[1] = s->ac_val_base + y_size + s->mb_stride + 1;
394         s->ac_val[2] = s->ac_val[1] + c_size;
395     }
396
397     return 0;
398 fail:
399     return -1; //free() through MPV_common_end()
400 }
401
402 static void free_duplicate_context(MpegEncContext *s){
403     if(s==NULL) return;
404
405     av_freep(&s->allocated_edge_emu_buffer); s->edge_emu_buffer= NULL;
406     av_freep(&s->me.scratchpad);
407     s->me.temp=
408     s->rd_scratchpad=
409     s->b_scratchpad=
410     s->obmc_scratchpad= NULL;
411
412     av_freep(&s->dct_error_sum);
413     av_freep(&s->me.map);
414     av_freep(&s->me.score_map);
415     av_freep(&s->blocks);
416     av_freep(&s->ac_val_base);
417     s->block= NULL;
418 }
419
420 static void backup_duplicate_context(MpegEncContext *bak, MpegEncContext *src){
421 #define COPY(a) bak->a= src->a
422     COPY(allocated_edge_emu_buffer);
423     COPY(edge_emu_buffer);
424     COPY(me.scratchpad);
425     COPY(me.temp);
426     COPY(rd_scratchpad);
427     COPY(b_scratchpad);
428     COPY(obmc_scratchpad);
429     COPY(me.map);
430     COPY(me.score_map);
431     COPY(blocks);
432     COPY(block);
433     COPY(start_mb_y);
434     COPY(end_mb_y);
435     COPY(me.map_generation);
436     COPY(pb);
437     COPY(dct_error_sum);
438     COPY(dct_count[0]);
439     COPY(dct_count[1]);
440     COPY(ac_val_base);
441     COPY(ac_val[0]);
442     COPY(ac_val[1]);
443     COPY(ac_val[2]);
444 #undef COPY
445 }
446
447 void ff_update_duplicate_context(MpegEncContext *dst, MpegEncContext *src){
448     MpegEncContext bak;
449     int i;
450     //FIXME copy only needed parts
451 //START_TIMER
452     backup_duplicate_context(&bak, dst);
453     memcpy(dst, src, sizeof(MpegEncContext));
454     backup_duplicate_context(dst, &bak);
455     for(i=0;i<12;i++){
456         dst->pblocks[i] = &dst->block[i];
457     }
458 //STOP_TIMER("update_duplicate_context") //about 10k cycles / 0.01 sec for 1000frames on 1ghz with 2 threads
459 }
460
461 /**
462  * sets the given MpegEncContext to common defaults (same for encoding and decoding).
463  * the changed fields will not depend upon the prior state of the MpegEncContext.
464  */
465 void MPV_common_defaults(MpegEncContext *s){
466     s->y_dc_scale_table=
467     s->c_dc_scale_table= ff_mpeg1_dc_scale_table;
468     s->chroma_qscale_table= ff_default_chroma_qscale_table;
469     s->progressive_frame= 1;
470     s->progressive_sequence= 1;
471     s->picture_structure= PICT_FRAME;
472
473     s->coded_picture_number = 0;
474     s->picture_number = 0;
475     s->input_picture_number = 0;
476
477     s->picture_in_gop_number = 0;
478
479     s->f_code = 1;
480     s->b_code = 1;
481 }
482
483 /**
484  * sets the given MpegEncContext to defaults for decoding.
485  * the changed fields will not depend upon the prior state of the MpegEncContext.
486  */
487 void MPV_decode_defaults(MpegEncContext *s){
488     MPV_common_defaults(s);
489 }
490
491 /**
492  * init common structure for both encoder and decoder.
493  * this assumes that some variables like width/height are already set
494  */
495 av_cold int MPV_common_init(MpegEncContext *s)
496 {
497     int y_size, c_size, yc_size, i, mb_array_size, mv_table_size, x, y, threads;
498
499     if(s->codec_id == CODEC_ID_MPEG2VIDEO && !s->progressive_sequence)
500         s->mb_height = (s->height + 31) / 32 * 2;
501     else if (s->codec_id != CODEC_ID_H264)
502         s->mb_height = (s->height + 15) / 16;
503
504     if(s->avctx->pix_fmt == PIX_FMT_NONE){
505         av_log(s->avctx, AV_LOG_ERROR, "decoding to PIX_FMT_NONE is not supported.\n");
506         return -1;
507     }
508
509     if(s->avctx->thread_count > MAX_THREADS || (s->avctx->thread_count > s->mb_height && s->mb_height)){
510         av_log(s->avctx, AV_LOG_ERROR, "too many threads\n");
511         return -1;
512     }
513
514     if((s->width || s->height) && av_image_check_size(s->width, s->height, 0, s->avctx))
515         return -1;
516
517     dsputil_init(&s->dsp, s->avctx);
518     ff_dct_common_init(s);
519
520     s->flags= s->avctx->flags;
521     s->flags2= s->avctx->flags2;
522
523     s->mb_width  = (s->width  + 15) / 16;
524     s->mb_stride = s->mb_width + 1;
525     s->b8_stride = s->mb_width*2 + 1;
526     s->b4_stride = s->mb_width*4 + 1;
527     mb_array_size= s->mb_height * s->mb_stride;
528     mv_table_size= (s->mb_height+2) * s->mb_stride + 1;
529
530     /* set chroma shifts */
531     avcodec_get_chroma_sub_sample(s->avctx->pix_fmt,&(s->chroma_x_shift),
532                                                     &(s->chroma_y_shift) );
533
534     /* set default edge pos, will be overriden in decode_header if needed */
535     s->h_edge_pos= s->mb_width*16;
536     s->v_edge_pos= s->mb_height*16;
537
538     s->mb_num = s->mb_width * s->mb_height;
539
540     s->block_wrap[0]=
541     s->block_wrap[1]=
542     s->block_wrap[2]=
543     s->block_wrap[3]= s->b8_stride;
544     s->block_wrap[4]=
545     s->block_wrap[5]= s->mb_stride;
546
547     y_size = s->b8_stride * (2 * s->mb_height + 1);
548     c_size = s->mb_stride * (s->mb_height + 1);
549     yc_size = y_size + 2 * c_size;
550
551     /* convert fourcc to upper case */
552     s->codec_tag = ff_toupper4(s->avctx->codec_tag);
553
554     s->stream_codec_tag = ff_toupper4(s->avctx->stream_codec_tag);
555
556     s->avctx->coded_frame= (AVFrame*)&s->current_picture;
557
558     FF_ALLOCZ_OR_GOTO(s->avctx, s->mb_index2xy, (s->mb_num+1)*sizeof(int), fail) //error ressilience code looks cleaner with this
559     for(y=0; y<s->mb_height; y++){
560         for(x=0; x<s->mb_width; x++){
561             s->mb_index2xy[ x + y*s->mb_width ] = x + y*s->mb_stride;
562         }
563     }
564     s->mb_index2xy[ s->mb_height*s->mb_width ] = (s->mb_height-1)*s->mb_stride + s->mb_width; //FIXME really needed?
565
566     if (s->encoding) {
567         /* Allocate MV tables */
568         FF_ALLOCZ_OR_GOTO(s->avctx, s->p_mv_table_base            , mv_table_size * 2 * sizeof(int16_t), fail)
569         FF_ALLOCZ_OR_GOTO(s->avctx, s->b_forw_mv_table_base       , mv_table_size * 2 * sizeof(int16_t), fail)
570         FF_ALLOCZ_OR_GOTO(s->avctx, s->b_back_mv_table_base       , mv_table_size * 2 * sizeof(int16_t), fail)
571         FF_ALLOCZ_OR_GOTO(s->avctx, s->b_bidir_forw_mv_table_base , mv_table_size * 2 * sizeof(int16_t), fail)
572         FF_ALLOCZ_OR_GOTO(s->avctx, s->b_bidir_back_mv_table_base , mv_table_size * 2 * sizeof(int16_t), fail)
573         FF_ALLOCZ_OR_GOTO(s->avctx, s->b_direct_mv_table_base     , mv_table_size * 2 * sizeof(int16_t), fail)
574         s->p_mv_table           = s->p_mv_table_base            + s->mb_stride + 1;
575         s->b_forw_mv_table      = s->b_forw_mv_table_base       + s->mb_stride + 1;
576         s->b_back_mv_table      = s->b_back_mv_table_base       + s->mb_stride + 1;
577         s->b_bidir_forw_mv_table= s->b_bidir_forw_mv_table_base + s->mb_stride + 1;
578         s->b_bidir_back_mv_table= s->b_bidir_back_mv_table_base + s->mb_stride + 1;
579         s->b_direct_mv_table    = s->b_direct_mv_table_base     + s->mb_stride + 1;
580
581         if(s->msmpeg4_version){
582             FF_ALLOCZ_OR_GOTO(s->avctx, s->ac_stats, 2*2*(MAX_LEVEL+1)*(MAX_RUN+1)*2*sizeof(int), fail);
583         }
584         FF_ALLOCZ_OR_GOTO(s->avctx, s->avctx->stats_out, 256, fail);
585
586         /* Allocate MB type table */
587         FF_ALLOCZ_OR_GOTO(s->avctx, s->mb_type  , mb_array_size * sizeof(uint16_t), fail) //needed for encoding
588
589         FF_ALLOCZ_OR_GOTO(s->avctx, s->lambda_table, mb_array_size * sizeof(int), fail)
590
591         FF_ALLOCZ_OR_GOTO(s->avctx, s->q_intra_matrix  , 64*32   * sizeof(int), fail)
592         FF_ALLOCZ_OR_GOTO(s->avctx, s->q_inter_matrix  , 64*32   * sizeof(int), fail)
593         FF_ALLOCZ_OR_GOTO(s->avctx, s->q_intra_matrix16, 64*32*2 * sizeof(uint16_t), fail)
594         FF_ALLOCZ_OR_GOTO(s->avctx, s->q_inter_matrix16, 64*32*2 * sizeof(uint16_t), fail)
595         FF_ALLOCZ_OR_GOTO(s->avctx, s->input_picture, MAX_PICTURE_COUNT * sizeof(Picture*), fail)
596         FF_ALLOCZ_OR_GOTO(s->avctx, s->reordered_input_picture, MAX_PICTURE_COUNT * sizeof(Picture*), fail)
597
598         if(s->avctx->noise_reduction){
599             FF_ALLOCZ_OR_GOTO(s->avctx, s->dct_offset, 2 * 64 * sizeof(uint16_t), fail)
600         }
601     }
602     FF_ALLOCZ_OR_GOTO(s->avctx, s->picture, MAX_PICTURE_COUNT * sizeof(Picture), fail)
603     for(i = 0; i < MAX_PICTURE_COUNT; i++) {
604         avcodec_get_frame_defaults((AVFrame *)&s->picture[i]);
605     }
606
607     FF_ALLOCZ_OR_GOTO(s->avctx, s->error_status_table, mb_array_size*sizeof(uint8_t), fail)
608
609     if(s->codec_id==CODEC_ID_MPEG4 || (s->flags & CODEC_FLAG_INTERLACED_ME)){
610         /* interlaced direct mode decoding tables */
611             for(i=0; i<2; i++){
612                 int j, k;
613                 for(j=0; j<2; j++){
614                     for(k=0; k<2; k++){
615                         FF_ALLOCZ_OR_GOTO(s->avctx,    s->b_field_mv_table_base[i][j][k], mv_table_size * 2 * sizeof(int16_t), fail)
616                         s->b_field_mv_table[i][j][k] = s->b_field_mv_table_base[i][j][k] + s->mb_stride + 1;
617                     }
618                     FF_ALLOCZ_OR_GOTO(s->avctx, s->b_field_select_table [i][j], mb_array_size * 2 * sizeof(uint8_t), fail)
619                     FF_ALLOCZ_OR_GOTO(s->avctx, s->p_field_mv_table_base[i][j], mv_table_size * 2 * sizeof(int16_t), fail)
620                     s->p_field_mv_table[i][j] = s->p_field_mv_table_base[i][j]+ s->mb_stride + 1;
621                 }
622                 FF_ALLOCZ_OR_GOTO(s->avctx, s->p_field_select_table[i], mb_array_size * 2 * sizeof(uint8_t), fail)
623             }
624     }
625     if (s->out_format == FMT_H263) {
626         /* cbp values */
627         FF_ALLOCZ_OR_GOTO(s->avctx, s->coded_block_base, y_size, fail);
628         s->coded_block= s->coded_block_base + s->b8_stride + 1;
629
630         /* cbp, ac_pred, pred_dir */
631         FF_ALLOCZ_OR_GOTO(s->avctx, s->cbp_table     , mb_array_size * sizeof(uint8_t), fail)
632         FF_ALLOCZ_OR_GOTO(s->avctx, s->pred_dir_table, mb_array_size * sizeof(uint8_t), fail)
633     }
634
635     if (s->h263_pred || s->h263_plus || !s->encoding) {
636         /* dc values */
637         //MN: we need these for error resilience of intra-frames
638         FF_ALLOCZ_OR_GOTO(s->avctx, s->dc_val_base, yc_size * sizeof(int16_t), fail);
639         s->dc_val[0] = s->dc_val_base + s->b8_stride + 1;
640         s->dc_val[1] = s->dc_val_base + y_size + s->mb_stride + 1;
641         s->dc_val[2] = s->dc_val[1] + c_size;
642         for(i=0;i<yc_size;i++)
643             s->dc_val_base[i] = 1024;
644     }
645
646     /* which mb is a intra block */
647     FF_ALLOCZ_OR_GOTO(s->avctx, s->mbintra_table, mb_array_size, fail);
648     memset(s->mbintra_table, 1, mb_array_size);
649
650     /* init macroblock skip table */
651     FF_ALLOCZ_OR_GOTO(s->avctx, s->mbskip_table, mb_array_size+2, fail);
652     //Note the +1 is for a quicker mpeg4 slice_end detection
653     FF_ALLOCZ_OR_GOTO(s->avctx, s->prev_pict_types, PREV_PICT_TYPES_BUFFER_SIZE, fail);
654
655     s->parse_context.state= -1;
656     if((s->avctx->debug&(FF_DEBUG_VIS_QP|FF_DEBUG_VIS_MB_TYPE)) || (s->avctx->debug_mv)){
657        s->visualization_buffer[0] = av_malloc((s->mb_width*16 + 2*EDGE_WIDTH) * s->mb_height*16 + 2*EDGE_WIDTH);
658        s->visualization_buffer[1] = av_malloc((s->mb_width*16 + 2*EDGE_WIDTH) * s->mb_height*16 + 2*EDGE_WIDTH);
659        s->visualization_buffer[2] = av_malloc((s->mb_width*16 + 2*EDGE_WIDTH) * s->mb_height*16 + 2*EDGE_WIDTH);
660     }
661
662     s->context_initialized = 1;
663
664     s->thread_context[0]= s;
665     threads = s->avctx->thread_count;
666
667     for(i=1; i<threads; i++){
668         s->thread_context[i]= av_malloc(sizeof(MpegEncContext));
669         memcpy(s->thread_context[i], s, sizeof(MpegEncContext));
670     }
671
672     for(i=0; i<threads; i++){
673         if(init_duplicate_context(s->thread_context[i], s) < 0)
674            goto fail;
675         s->thread_context[i]->start_mb_y= (s->mb_height*(i  ) + s->avctx->thread_count/2) / s->avctx->thread_count;
676         s->thread_context[i]->end_mb_y  = (s->mb_height*(i+1) + s->avctx->thread_count/2) / s->avctx->thread_count;
677     }
678
679     return 0;
680  fail:
681     MPV_common_end(s);
682     return -1;
683 }
684
685 /* init common structure for both encoder and decoder */
686 void MPV_common_end(MpegEncContext *s)
687 {
688     int i, j, k;
689
690     for(i=0; i<s->avctx->thread_count; i++){
691         free_duplicate_context(s->thread_context[i]);
692     }
693     for(i=1; i<s->avctx->thread_count; i++){
694         av_freep(&s->thread_context[i]);
695     }
696
697     av_freep(&s->parse_context.buffer);
698     s->parse_context.buffer_size=0;
699
700     av_freep(&s->mb_type);
701     av_freep(&s->p_mv_table_base);
702     av_freep(&s->b_forw_mv_table_base);
703     av_freep(&s->b_back_mv_table_base);
704     av_freep(&s->b_bidir_forw_mv_table_base);
705     av_freep(&s->b_bidir_back_mv_table_base);
706     av_freep(&s->b_direct_mv_table_base);
707     s->p_mv_table= NULL;
708     s->b_forw_mv_table= NULL;
709     s->b_back_mv_table= NULL;
710     s->b_bidir_forw_mv_table= NULL;
711     s->b_bidir_back_mv_table= NULL;
712     s->b_direct_mv_table= NULL;
713     for(i=0; i<2; i++){
714         for(j=0; j<2; j++){
715             for(k=0; k<2; k++){
716                 av_freep(&s->b_field_mv_table_base[i][j][k]);
717                 s->b_field_mv_table[i][j][k]=NULL;
718             }
719             av_freep(&s->b_field_select_table[i][j]);
720             av_freep(&s->p_field_mv_table_base[i][j]);
721             s->p_field_mv_table[i][j]=NULL;
722         }
723         av_freep(&s->p_field_select_table[i]);
724     }
725
726     av_freep(&s->dc_val_base);
727     av_freep(&s->coded_block_base);
728     av_freep(&s->mbintra_table);
729     av_freep(&s->cbp_table);
730     av_freep(&s->pred_dir_table);
731
732     av_freep(&s->mbskip_table);
733     av_freep(&s->prev_pict_types);
734     av_freep(&s->bitstream_buffer);
735     s->allocated_bitstream_buffer_size=0;
736
737     av_freep(&s->avctx->stats_out);
738     av_freep(&s->ac_stats);
739     av_freep(&s->error_status_table);
740     av_freep(&s->mb_index2xy);
741     av_freep(&s->lambda_table);
742     av_freep(&s->q_intra_matrix);
743     av_freep(&s->q_inter_matrix);
744     av_freep(&s->q_intra_matrix16);
745     av_freep(&s->q_inter_matrix16);
746     av_freep(&s->input_picture);
747     av_freep(&s->reordered_input_picture);
748     av_freep(&s->dct_offset);
749
750     if(s->picture){
751         for(i=0; i<MAX_PICTURE_COUNT; i++){
752             free_picture(s, &s->picture[i]);
753         }
754     }
755     av_freep(&s->picture);
756     s->context_initialized = 0;
757     s->last_picture_ptr=
758     s->next_picture_ptr=
759     s->current_picture_ptr= NULL;
760     s->linesize= s->uvlinesize= 0;
761
762     for(i=0; i<3; i++)
763         av_freep(&s->visualization_buffer[i]);
764
765     avcodec_default_free_buffers(s->avctx);
766 }
767
768 void init_rl(RLTable *rl, uint8_t static_store[2][2*MAX_RUN + MAX_LEVEL + 3])
769 {
770     int8_t max_level[MAX_RUN+1], max_run[MAX_LEVEL+1];
771     uint8_t index_run[MAX_RUN+1];
772     int last, run, level, start, end, i;
773
774     /* If table is static, we can quit if rl->max_level[0] is not NULL */
775     if(static_store && rl->max_level[0])
776         return;
777
778     /* compute max_level[], max_run[] and index_run[] */
779     for(last=0;last<2;last++) {
780         if (last == 0) {
781             start = 0;
782             end = rl->last;
783         } else {
784             start = rl->last;
785             end = rl->n;
786         }
787
788         memset(max_level, 0, MAX_RUN + 1);
789         memset(max_run, 0, MAX_LEVEL + 1);
790         memset(index_run, rl->n, MAX_RUN + 1);
791         for(i=start;i<end;i++) {
792             run = rl->table_run[i];
793             level = rl->table_level[i];
794             if (index_run[run] == rl->n)
795                 index_run[run] = i;
796             if (level > max_level[run])
797                 max_level[run] = level;
798             if (run > max_run[level])
799                 max_run[level] = run;
800         }
801         if(static_store)
802             rl->max_level[last] = static_store[last];
803         else
804             rl->max_level[last] = av_malloc(MAX_RUN + 1);
805         memcpy(rl->max_level[last], max_level, MAX_RUN + 1);
806         if(static_store)
807             rl->max_run[last] = static_store[last] + MAX_RUN + 1;
808         else
809             rl->max_run[last] = av_malloc(MAX_LEVEL + 1);
810         memcpy(rl->max_run[last], max_run, MAX_LEVEL + 1);
811         if(static_store)
812             rl->index_run[last] = static_store[last] + MAX_RUN + MAX_LEVEL + 2;
813         else
814             rl->index_run[last] = av_malloc(MAX_RUN + 1);
815         memcpy(rl->index_run[last], index_run, MAX_RUN + 1);
816     }
817 }
818
819 void init_vlc_rl(RLTable *rl)
820 {
821     int i, q;
822
823     for(q=0; q<32; q++){
824         int qmul= q*2;
825         int qadd= (q-1)|1;
826
827         if(q==0){
828             qmul=1;
829             qadd=0;
830         }
831         for(i=0; i<rl->vlc.table_size; i++){
832             int code= rl->vlc.table[i][0];
833             int len = rl->vlc.table[i][1];
834             int level, run;
835
836             if(len==0){ // illegal code
837                 run= 66;
838                 level= MAX_LEVEL;
839             }else if(len<0){ //more bits needed
840                 run= 0;
841                 level= code;
842             }else{
843                 if(code==rl->n){ //esc
844                     run= 66;
845                     level= 0;
846                 }else{
847                     run=   rl->table_run  [code] + 1;
848                     level= rl->table_level[code] * qmul + qadd;
849                     if(code >= rl->last) run+=192;
850                 }
851             }
852             rl->rl_vlc[q][i].len= len;
853             rl->rl_vlc[q][i].level= level;
854             rl->rl_vlc[q][i].run= run;
855         }
856     }
857 }
858
859 int ff_find_unused_picture(MpegEncContext *s, int shared){
860     int i;
861
862     if(shared){
863         for(i=0; i<MAX_PICTURE_COUNT; i++){
864             if(s->picture[i].data[0]==NULL && s->picture[i].type==0) return i;
865         }
866     }else{
867         for(i=0; i<MAX_PICTURE_COUNT; i++){
868             if(s->picture[i].data[0]==NULL && s->picture[i].type!=0) return i; //FIXME
869         }
870         for(i=0; i<MAX_PICTURE_COUNT; i++){
871             if(s->picture[i].data[0]==NULL) return i;
872         }
873     }
874
875     av_log(s->avctx, AV_LOG_FATAL, "Internal error, picture buffer overflow\n");
876     /* We could return -1, but the codec would crash trying to draw into a
877      * non-existing frame anyway. This is safer than waiting for a random crash.
878      * Also the return of this is never useful, an encoder must only allocate
879      * as much as allowed in the specification. This has no relationship to how
880      * much libavcodec could allocate (and MAX_PICTURE_COUNT is always large
881      * enough for such valid streams).
882      * Plus, a decoder has to check stream validity and remove frames if too
883      * many reference frames are around. Waiting for "OOM" is not correct at
884      * all. Similarly, missing reference frames have to be replaced by
885      * interpolated/MC frames, anything else is a bug in the codec ...
886      */
887     abort();
888     return -1;
889 }
890
891 static void update_noise_reduction(MpegEncContext *s){
892     int intra, i;
893
894     for(intra=0; intra<2; intra++){
895         if(s->dct_count[intra] > (1<<16)){
896             for(i=0; i<64; i++){
897                 s->dct_error_sum[intra][i] >>=1;
898             }
899             s->dct_count[intra] >>= 1;
900         }
901
902         for(i=0; i<64; i++){
903             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);
904         }
905     }
906 }
907
908 /**
909  * generic function for encode/decode called after coding/decoding the header and before a frame is coded/decoded
910  */
911 int MPV_frame_start(MpegEncContext *s, AVCodecContext *avctx)
912 {
913     int i;
914     Picture *pic;
915     s->mb_skipped = 0;
916
917     assert(s->last_picture_ptr==NULL || s->out_format != FMT_H264 || s->codec_id == CODEC_ID_SVQ3);
918
919     /* mark&release old frames */
920     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]) {
921       if(s->out_format != FMT_H264 || s->codec_id == CODEC_ID_SVQ3){
922           free_frame_buffer(s, s->last_picture_ptr);
923
924         /* release forgotten pictures */
925         /* if(mpeg124/h263) */
926         if(!s->encoding){
927             for(i=0; i<MAX_PICTURE_COUNT; i++){
928                 if(s->picture[i].data[0] && &s->picture[i] != s->next_picture_ptr && s->picture[i].reference){
929                     av_log(avctx, AV_LOG_ERROR, "releasing zombie picture\n");
930                     free_frame_buffer(s, &s->picture[i]);
931                 }
932             }
933         }
934       }
935     }
936
937     if(!s->encoding){
938         /* release non reference frames */
939         for(i=0; i<MAX_PICTURE_COUNT; i++){
940             if(s->picture[i].data[0] && !s->picture[i].reference /*&& s->picture[i].type!=FF_BUFFER_TYPE_SHARED*/){
941                 free_frame_buffer(s, &s->picture[i]);
942             }
943         }
944
945         if(s->current_picture_ptr && s->current_picture_ptr->data[0]==NULL)
946             pic= s->current_picture_ptr; //we already have a unused image (maybe it was set before reading the header)
947         else{
948             i= ff_find_unused_picture(s, 0);
949             pic= &s->picture[i];
950         }
951
952         pic->reference= 0;
953         if (!s->dropable){
954             if (s->codec_id == CODEC_ID_H264)
955                 pic->reference = s->picture_structure;
956             else if (s->pict_type != FF_B_TYPE)
957                 pic->reference = 3;
958         }
959
960         pic->coded_picture_number= s->coded_picture_number++;
961
962         if(ff_alloc_picture(s, pic, 0) < 0)
963             return -1;
964
965         s->current_picture_ptr= pic;
966         //FIXME use only the vars from current_pic
967         s->current_picture_ptr->top_field_first= s->top_field_first;
968         if(s->codec_id == CODEC_ID_MPEG1VIDEO || s->codec_id == CODEC_ID_MPEG2VIDEO) {
969             if(s->picture_structure != PICT_FRAME)
970                 s->current_picture_ptr->top_field_first= (s->picture_structure == PICT_TOP_FIELD) == s->first_field;
971         }
972         s->current_picture_ptr->interlaced_frame= !s->progressive_frame && !s->progressive_sequence;
973     }
974
975     s->current_picture_ptr->pict_type= s->pict_type;
976 //    if(s->flags && CODEC_FLAG_QSCALE)
977   //      s->current_picture_ptr->quality= s->new_picture_ptr->quality;
978     s->current_picture_ptr->key_frame= s->pict_type == FF_I_TYPE;
979
980     ff_copy_picture(&s->current_picture, s->current_picture_ptr);
981
982     if (s->pict_type != FF_B_TYPE) {
983         s->last_picture_ptr= s->next_picture_ptr;
984         if(!s->dropable)
985             s->next_picture_ptr= s->current_picture_ptr;
986     }
987 /*    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,
988         s->last_picture_ptr    ? s->last_picture_ptr->data[0] : NULL,
989         s->next_picture_ptr    ? s->next_picture_ptr->data[0] : NULL,
990         s->current_picture_ptr ? s->current_picture_ptr->data[0] : NULL,
991         s->pict_type, s->dropable);*/
992
993     if(s->codec_id != CODEC_ID_H264){
994         if((s->last_picture_ptr==NULL || s->last_picture_ptr->data[0]==NULL) && s->pict_type!=FF_I_TYPE){
995             av_log(avctx, AV_LOG_ERROR, "warning: first frame is no keyframe\n");
996             /* Allocate a dummy frame */
997             i= ff_find_unused_picture(s, 0);
998             s->last_picture_ptr= &s->picture[i];
999             if(ff_alloc_picture(s, s->last_picture_ptr, 0) < 0)
1000                 return -1;
1001         }
1002         if((s->next_picture_ptr==NULL || s->next_picture_ptr->data[0]==NULL) && s->pict_type==FF_B_TYPE){
1003             /* Allocate a dummy frame */
1004             i= ff_find_unused_picture(s, 0);
1005             s->next_picture_ptr= &s->picture[i];
1006             if(ff_alloc_picture(s, s->next_picture_ptr, 0) < 0)
1007                 return -1;
1008         }
1009     }
1010
1011     if(s->last_picture_ptr) ff_copy_picture(&s->last_picture, s->last_picture_ptr);
1012     if(s->next_picture_ptr) ff_copy_picture(&s->next_picture, s->next_picture_ptr);
1013
1014     assert(s->pict_type == FF_I_TYPE || (s->last_picture_ptr && s->last_picture_ptr->data[0]));
1015
1016     if(s->picture_structure!=PICT_FRAME && s->out_format != FMT_H264){
1017         int i;
1018         for(i=0; i<4; i++){
1019             if(s->picture_structure == PICT_BOTTOM_FIELD){
1020                  s->current_picture.data[i] += s->current_picture.linesize[i];
1021             }
1022             s->current_picture.linesize[i] *= 2;
1023             s->last_picture.linesize[i] *=2;
1024             s->next_picture.linesize[i] *=2;
1025         }
1026     }
1027
1028     s->hurry_up= s->avctx->hurry_up;
1029     s->error_recognition= avctx->error_recognition;
1030
1031     /* set dequantizer, we can't do it during init as it might change for mpeg4
1032        and we can't do it in the header decode as init is not called for mpeg4 there yet */
1033     if(s->mpeg_quant || s->codec_id == CODEC_ID_MPEG2VIDEO){
1034         s->dct_unquantize_intra = s->dct_unquantize_mpeg2_intra;
1035         s->dct_unquantize_inter = s->dct_unquantize_mpeg2_inter;
1036     }else if(s->out_format == FMT_H263 || s->out_format == FMT_H261){
1037         s->dct_unquantize_intra = s->dct_unquantize_h263_intra;
1038         s->dct_unquantize_inter = s->dct_unquantize_h263_inter;
1039     }else{
1040         s->dct_unquantize_intra = s->dct_unquantize_mpeg1_intra;
1041         s->dct_unquantize_inter = s->dct_unquantize_mpeg1_inter;
1042     }
1043
1044     if(s->dct_error_sum){
1045         assert(s->avctx->noise_reduction && s->encoding);
1046
1047         update_noise_reduction(s);
1048     }
1049
1050     if(CONFIG_MPEG_XVMC_DECODER && s->avctx->xvmc_acceleration)
1051         return ff_xvmc_field_start(s, avctx);
1052
1053     return 0;
1054 }
1055
1056 /* generic function for encode/decode called after a frame has been coded/decoded */
1057 void MPV_frame_end(MpegEncContext *s)
1058 {
1059     int i;
1060     /* draw edge for correct motion prediction if outside */
1061     //just to make sure that all data is rendered.
1062     if(CONFIG_MPEG_XVMC_DECODER && s->avctx->xvmc_acceleration){
1063         ff_xvmc_field_end(s);
1064     }else if(!s->avctx->hwaccel
1065        && !(s->avctx->codec->capabilities&CODEC_CAP_HWACCEL_VDPAU)
1066        && s->unrestricted_mv
1067        && s->current_picture.reference
1068        && !s->intra_only
1069        && !(s->flags&CODEC_FLAG_EMU_EDGE)) {
1070             s->dsp.draw_edges(s->current_picture.data[0], s->linesize  ,
1071                               s->h_edge_pos   , s->v_edge_pos   ,
1072                               EDGE_WIDTH  , EDGE_TOP | EDGE_BOTTOM);
1073             s->dsp.draw_edges(s->current_picture.data[1], s->uvlinesize,
1074                               s->h_edge_pos>>1, s->v_edge_pos>>1,
1075                               EDGE_WIDTH/2, EDGE_TOP | EDGE_BOTTOM);
1076             s->dsp.draw_edges(s->current_picture.data[2], s->uvlinesize,
1077                               s->h_edge_pos>>1, s->v_edge_pos>>1,
1078                               EDGE_WIDTH/2, EDGE_TOP | EDGE_BOTTOM);
1079     }
1080     emms_c();
1081
1082     s->last_pict_type    = s->pict_type;
1083     s->last_lambda_for[s->pict_type]= s->current_picture_ptr->quality;
1084     if(s->pict_type!=FF_B_TYPE){
1085         s->last_non_b_pict_type= s->pict_type;
1086     }
1087 #if 0
1088         /* copy back current_picture variables */
1089     for(i=0; i<MAX_PICTURE_COUNT; i++){
1090         if(s->picture[i].data[0] == s->current_picture.data[0]){
1091             s->picture[i]= s->current_picture;
1092             break;
1093         }
1094     }
1095     assert(i<MAX_PICTURE_COUNT);
1096 #endif
1097
1098     if(s->encoding){
1099         /* release non-reference frames */
1100         for(i=0; i<MAX_PICTURE_COUNT; i++){
1101             if(s->picture[i].data[0] && !s->picture[i].reference /*&& s->picture[i].type!=FF_BUFFER_TYPE_SHARED*/){
1102                 free_frame_buffer(s, &s->picture[i]);
1103             }
1104         }
1105     }
1106     // clear copies, to avoid confusion
1107 #if 0
1108     memset(&s->last_picture, 0, sizeof(Picture));
1109     memset(&s->next_picture, 0, sizeof(Picture));
1110     memset(&s->current_picture, 0, sizeof(Picture));
1111 #endif
1112     s->avctx->coded_frame= (AVFrame*)s->current_picture_ptr;
1113 }
1114
1115 /**
1116  * draws an line from (ex, ey) -> (sx, sy).
1117  * @param w width of the image
1118  * @param h height of the image
1119  * @param stride stride/linesize of the image
1120  * @param color color of the arrow
1121  */
1122 static void draw_line(uint8_t *buf, int sx, int sy, int ex, int ey, int w, int h, int stride, int color){
1123     int x, y, fr, f;
1124
1125     sx= av_clip(sx, 0, w-1);
1126     sy= av_clip(sy, 0, h-1);
1127     ex= av_clip(ex, 0, w-1);
1128     ey= av_clip(ey, 0, h-1);
1129
1130     buf[sy*stride + sx]+= color;
1131
1132     if(FFABS(ex - sx) > FFABS(ey - sy)){
1133         if(sx > ex){
1134             FFSWAP(int, sx, ex);
1135             FFSWAP(int, sy, ey);
1136         }
1137         buf+= sx + sy*stride;
1138         ex-= sx;
1139         f= ((ey-sy)<<16)/ex;
1140         for(x= 0; x <= ex; x++){
1141             y = (x*f)>>16;
1142             fr= (x*f)&0xFFFF;
1143             buf[ y   *stride + x]+= (color*(0x10000-fr))>>16;
1144             buf[(y+1)*stride + x]+= (color*         fr )>>16;
1145         }
1146     }else{
1147         if(sy > ey){
1148             FFSWAP(int, sx, ex);
1149             FFSWAP(int, sy, ey);
1150         }
1151         buf+= sx + sy*stride;
1152         ey-= sy;
1153         if(ey) f= ((ex-sx)<<16)/ey;
1154         else   f= 0;
1155         for(y= 0; y <= ey; y++){
1156             x = (y*f)>>16;
1157             fr= (y*f)&0xFFFF;
1158             buf[y*stride + x  ]+= (color*(0x10000-fr))>>16;
1159             buf[y*stride + x+1]+= (color*         fr )>>16;
1160         }
1161     }
1162 }
1163
1164 /**
1165  * draws an arrow from (ex, ey) -> (sx, sy).
1166  * @param w width of the image
1167  * @param h height of the image
1168  * @param stride stride/linesize of the image
1169  * @param color color of the arrow
1170  */
1171 static void draw_arrow(uint8_t *buf, int sx, int sy, int ex, int ey, int w, int h, int stride, int color){
1172     int dx,dy;
1173
1174     sx= av_clip(sx, -100, w+100);
1175     sy= av_clip(sy, -100, h+100);
1176     ex= av_clip(ex, -100, w+100);
1177     ey= av_clip(ey, -100, h+100);
1178
1179     dx= ex - sx;
1180     dy= ey - sy;
1181
1182     if(dx*dx + dy*dy > 3*3){
1183         int rx=  dx + dy;
1184         int ry= -dx + dy;
1185         int length= ff_sqrt((rx*rx + ry*ry)<<8);
1186
1187         //FIXME subpixel accuracy
1188         rx= ROUNDED_DIV(rx*3<<4, length);
1189         ry= ROUNDED_DIV(ry*3<<4, length);
1190
1191         draw_line(buf, sx, sy, sx + rx, sy + ry, w, h, stride, color);
1192         draw_line(buf, sx, sy, sx - ry, sy + rx, w, h, stride, color);
1193     }
1194     draw_line(buf, sx, sy, ex, ey, w, h, stride, color);
1195 }
1196
1197 /**
1198  * prints debuging info for the given picture.
1199  */
1200 void ff_print_debug_info(MpegEncContext *s, AVFrame *pict){
1201
1202     if(s->avctx->hwaccel || !pict || !pict->mb_type) return;
1203
1204     if(s->avctx->debug&(FF_DEBUG_SKIP | FF_DEBUG_QP | FF_DEBUG_MB_TYPE)){
1205         int x,y;
1206
1207         av_log(s->avctx,AV_LOG_DEBUG,"New frame, type: ");
1208         switch (pict->pict_type) {
1209             case FF_I_TYPE: av_log(s->avctx,AV_LOG_DEBUG,"I\n"); break;
1210             case FF_P_TYPE: av_log(s->avctx,AV_LOG_DEBUG,"P\n"); break;
1211             case FF_B_TYPE: av_log(s->avctx,AV_LOG_DEBUG,"B\n"); break;
1212             case FF_S_TYPE: av_log(s->avctx,AV_LOG_DEBUG,"S\n"); break;
1213             case FF_SI_TYPE: av_log(s->avctx,AV_LOG_DEBUG,"SI\n"); break;
1214             case FF_SP_TYPE: av_log(s->avctx,AV_LOG_DEBUG,"SP\n"); break;
1215         }
1216         for(y=0; y<s->mb_height; y++){
1217             for(x=0; x<s->mb_width; x++){
1218                 if(s->avctx->debug&FF_DEBUG_SKIP){
1219                     int count= s->mbskip_table[x + y*s->mb_stride];
1220                     if(count>9) count=9;
1221                     av_log(s->avctx, AV_LOG_DEBUG, "%1d", count);
1222                 }
1223                 if(s->avctx->debug&FF_DEBUG_QP){
1224                     av_log(s->avctx, AV_LOG_DEBUG, "%2d", pict->qscale_table[x + y*s->mb_stride]);
1225                 }
1226                 if(s->avctx->debug&FF_DEBUG_MB_TYPE){
1227                     int mb_type= pict->mb_type[x + y*s->mb_stride];
1228                     //Type & MV direction
1229                     if(IS_PCM(mb_type))
1230                         av_log(s->avctx, AV_LOG_DEBUG, "P");
1231                     else if(IS_INTRA(mb_type) && IS_ACPRED(mb_type))
1232                         av_log(s->avctx, AV_LOG_DEBUG, "A");
1233                     else if(IS_INTRA4x4(mb_type))
1234                         av_log(s->avctx, AV_LOG_DEBUG, "i");
1235                     else if(IS_INTRA16x16(mb_type))
1236                         av_log(s->avctx, AV_LOG_DEBUG, "I");
1237                     else if(IS_DIRECT(mb_type) && IS_SKIP(mb_type))
1238                         av_log(s->avctx, AV_LOG_DEBUG, "d");
1239                     else if(IS_DIRECT(mb_type))
1240                         av_log(s->avctx, AV_LOG_DEBUG, "D");
1241                     else if(IS_GMC(mb_type) && IS_SKIP(mb_type))
1242                         av_log(s->avctx, AV_LOG_DEBUG, "g");
1243                     else if(IS_GMC(mb_type))
1244                         av_log(s->avctx, AV_LOG_DEBUG, "G");
1245                     else if(IS_SKIP(mb_type))
1246                         av_log(s->avctx, AV_LOG_DEBUG, "S");
1247                     else if(!USES_LIST(mb_type, 1))
1248                         av_log(s->avctx, AV_LOG_DEBUG, ">");
1249                     else if(!USES_LIST(mb_type, 0))
1250                         av_log(s->avctx, AV_LOG_DEBUG, "<");
1251                     else{
1252                         assert(USES_LIST(mb_type, 0) && USES_LIST(mb_type, 1));
1253                         av_log(s->avctx, AV_LOG_DEBUG, "X");
1254                     }
1255
1256                     //segmentation
1257                     if(IS_8X8(mb_type))
1258                         av_log(s->avctx, AV_LOG_DEBUG, "+");
1259                     else if(IS_16X8(mb_type))
1260                         av_log(s->avctx, AV_LOG_DEBUG, "-");
1261                     else if(IS_8X16(mb_type))
1262                         av_log(s->avctx, AV_LOG_DEBUG, "|");
1263                     else if(IS_INTRA(mb_type) || IS_16X16(mb_type))
1264                         av_log(s->avctx, AV_LOG_DEBUG, " ");
1265                     else
1266                         av_log(s->avctx, AV_LOG_DEBUG, "?");
1267
1268
1269                     if(IS_INTERLACED(mb_type))
1270                         av_log(s->avctx, AV_LOG_DEBUG, "=");
1271                     else
1272                         av_log(s->avctx, AV_LOG_DEBUG, " ");
1273                 }
1274 //                av_log(s->avctx, AV_LOG_DEBUG, " ");
1275             }
1276             av_log(s->avctx, AV_LOG_DEBUG, "\n");
1277         }
1278     }
1279
1280     if((s->avctx->debug&(FF_DEBUG_VIS_QP|FF_DEBUG_VIS_MB_TYPE)) || (s->avctx->debug_mv)){
1281         const int shift= 1 + s->quarter_sample;
1282         int mb_y;
1283         uint8_t *ptr;
1284         int i;
1285         int h_chroma_shift, v_chroma_shift, block_height;
1286         const int width = s->avctx->width;
1287         const int height= s->avctx->height;
1288         const int mv_sample_log2= 4 - pict->motion_subsample_log2;
1289         const int mv_stride= (s->mb_width << mv_sample_log2) + (s->codec_id == CODEC_ID_H264 ? 0 : 1);
1290         s->low_delay=0; //needed to see the vectors without trashing the buffers
1291
1292         avcodec_get_chroma_sub_sample(s->avctx->pix_fmt, &h_chroma_shift, &v_chroma_shift);
1293         for(i=0; i<3; i++){
1294             memcpy(s->visualization_buffer[i], pict->data[i], (i==0) ? pict->linesize[i]*height:pict->linesize[i]*height >> v_chroma_shift);
1295             pict->data[i]= s->visualization_buffer[i];
1296         }
1297         pict->type= FF_BUFFER_TYPE_COPY;
1298         ptr= pict->data[0];
1299         block_height = 16>>v_chroma_shift;
1300
1301         for(mb_y=0; mb_y<s->mb_height; mb_y++){
1302             int mb_x;
1303             for(mb_x=0; mb_x<s->mb_width; mb_x++){
1304                 const int mb_index= mb_x + mb_y*s->mb_stride;
1305                 if((s->avctx->debug_mv) && pict->motion_val){
1306                   int type;
1307                   for(type=0; type<3; type++){
1308                     int direction = 0;
1309                     switch (type) {
1310                       case 0: if ((!(s->avctx->debug_mv&FF_DEBUG_VIS_MV_P_FOR)) || (pict->pict_type!=FF_P_TYPE))
1311                                 continue;
1312                               direction = 0;
1313                               break;
1314                       case 1: if ((!(s->avctx->debug_mv&FF_DEBUG_VIS_MV_B_FOR)) || (pict->pict_type!=FF_B_TYPE))
1315                                 continue;
1316                               direction = 0;
1317                               break;
1318                       case 2: if ((!(s->avctx->debug_mv&FF_DEBUG_VIS_MV_B_BACK)) || (pict->pict_type!=FF_B_TYPE))
1319                                 continue;
1320                               direction = 1;
1321                               break;
1322                     }
1323                     if(!USES_LIST(pict->mb_type[mb_index], direction))
1324                         continue;
1325
1326                     if(IS_8X8(pict->mb_type[mb_index])){
1327                       int i;
1328                       for(i=0; i<4; i++){
1329                         int sx= mb_x*16 + 4 + 8*(i&1);
1330                         int sy= mb_y*16 + 4 + 8*(i>>1);
1331                         int xy= (mb_x*2 + (i&1) + (mb_y*2 + (i>>1))*mv_stride) << (mv_sample_log2-1);
1332                         int mx= (pict->motion_val[direction][xy][0]>>shift) + sx;
1333                         int my= (pict->motion_val[direction][xy][1]>>shift) + sy;
1334                         draw_arrow(ptr, sx, sy, mx, my, width, height, s->linesize, 100);
1335                       }
1336                     }else if(IS_16X8(pict->mb_type[mb_index])){
1337                       int i;
1338                       for(i=0; i<2; i++){
1339                         int sx=mb_x*16 + 8;
1340                         int sy=mb_y*16 + 4 + 8*i;
1341                         int xy= (mb_x*2 + (mb_y*2 + i)*mv_stride) << (mv_sample_log2-1);
1342                         int mx=(pict->motion_val[direction][xy][0]>>shift);
1343                         int my=(pict->motion_val[direction][xy][1]>>shift);
1344
1345                         if(IS_INTERLACED(pict->mb_type[mb_index]))
1346                             my*=2;
1347
1348                         draw_arrow(ptr, sx, sy, mx+sx, my+sy, width, height, s->linesize, 100);
1349                       }
1350                     }else if(IS_8X16(pict->mb_type[mb_index])){
1351                       int i;
1352                       for(i=0; i<2; i++){
1353                         int sx=mb_x*16 + 4 + 8*i;
1354                         int sy=mb_y*16 + 8;
1355                         int xy= (mb_x*2 + i + mb_y*2*mv_stride) << (mv_sample_log2-1);
1356                         int mx=(pict->motion_val[direction][xy][0]>>shift);
1357                         int my=(pict->motion_val[direction][xy][1]>>shift);
1358
1359                         if(IS_INTERLACED(pict->mb_type[mb_index]))
1360                             my*=2;
1361
1362                         draw_arrow(ptr, sx, sy, mx+sx, my+sy, width, height, s->linesize, 100);
1363                       }
1364                     }else{
1365                       int sx= mb_x*16 + 8;
1366                       int sy= mb_y*16 + 8;
1367                       int xy= (mb_x + mb_y*mv_stride) << mv_sample_log2;
1368                       int mx= (pict->motion_val[direction][xy][0]>>shift) + sx;
1369                       int my= (pict->motion_val[direction][xy][1]>>shift) + sy;
1370                       draw_arrow(ptr, sx, sy, mx, my, width, height, s->linesize, 100);
1371                     }
1372                   }
1373                 }
1374                 if((s->avctx->debug&FF_DEBUG_VIS_QP) && pict->motion_val){
1375                     uint64_t c= (pict->qscale_table[mb_index]*128/31) * 0x0101010101010101ULL;
1376                     int y;
1377                     for(y=0; y<block_height; y++){
1378                         *(uint64_t*)(pict->data[1] + 8*mb_x + (block_height*mb_y + y)*pict->linesize[1])= c;
1379                         *(uint64_t*)(pict->data[2] + 8*mb_x + (block_height*mb_y + y)*pict->linesize[2])= c;
1380                     }
1381                 }
1382                 if((s->avctx->debug&FF_DEBUG_VIS_MB_TYPE) && pict->motion_val){
1383                     int mb_type= pict->mb_type[mb_index];
1384                     uint64_t u,v;
1385                     int y;
1386 #define COLOR(theta, r)\
1387 u= (int)(128 + r*cos(theta*3.141592/180));\
1388 v= (int)(128 + r*sin(theta*3.141592/180));
1389
1390
1391                     u=v=128;
1392                     if(IS_PCM(mb_type)){
1393                         COLOR(120,48)
1394                     }else if((IS_INTRA(mb_type) && IS_ACPRED(mb_type)) || IS_INTRA16x16(mb_type)){
1395                         COLOR(30,48)
1396                     }else if(IS_INTRA4x4(mb_type)){
1397                         COLOR(90,48)
1398                     }else if(IS_DIRECT(mb_type) && IS_SKIP(mb_type)){
1399 //                        COLOR(120,48)
1400                     }else if(IS_DIRECT(mb_type)){
1401                         COLOR(150,48)
1402                     }else if(IS_GMC(mb_type) && IS_SKIP(mb_type)){
1403                         COLOR(170,48)
1404                     }else if(IS_GMC(mb_type)){
1405                         COLOR(190,48)
1406                     }else if(IS_SKIP(mb_type)){
1407 //                        COLOR(180,48)
1408                     }else if(!USES_LIST(mb_type, 1)){
1409                         COLOR(240,48)
1410                     }else if(!USES_LIST(mb_type, 0)){
1411                         COLOR(0,48)
1412                     }else{
1413                         assert(USES_LIST(mb_type, 0) && USES_LIST(mb_type, 1));
1414                         COLOR(300,48)
1415                     }
1416
1417                     u*= 0x0101010101010101ULL;
1418                     v*= 0x0101010101010101ULL;
1419                     for(y=0; y<block_height; y++){
1420                         *(uint64_t*)(pict->data[1] + 8*mb_x + (block_height*mb_y + y)*pict->linesize[1])= u;
1421                         *(uint64_t*)(pict->data[2] + 8*mb_x + (block_height*mb_y + y)*pict->linesize[2])= v;
1422                     }
1423
1424                     //segmentation
1425                     if(IS_8X8(mb_type) || IS_16X8(mb_type)){
1426                         *(uint64_t*)(pict->data[0] + 16*mb_x + 0 + (16*mb_y + 8)*pict->linesize[0])^= 0x8080808080808080ULL;
1427                         *(uint64_t*)(pict->data[0] + 16*mb_x + 8 + (16*mb_y + 8)*pict->linesize[0])^= 0x8080808080808080ULL;
1428                     }
1429                     if(IS_8X8(mb_type) || IS_8X16(mb_type)){
1430                         for(y=0; y<16; y++)
1431                             pict->data[0][16*mb_x + 8 + (16*mb_y + y)*pict->linesize[0]]^= 0x80;
1432                     }
1433                     if(IS_8X8(mb_type) && mv_sample_log2 >= 2){
1434                         int dm= 1 << (mv_sample_log2-2);
1435                         for(i=0; i<4; i++){
1436                             int sx= mb_x*16 + 8*(i&1);
1437                             int sy= mb_y*16 + 8*(i>>1);
1438                             int xy= (mb_x*2 + (i&1) + (mb_y*2 + (i>>1))*mv_stride) << (mv_sample_log2-1);
1439                             //FIXME bidir
1440                             int32_t *mv = (int32_t*)&pict->motion_val[0][xy];
1441                             if(mv[0] != mv[dm] || mv[dm*mv_stride] != mv[dm*(mv_stride+1)])
1442                                 for(y=0; y<8; y++)
1443                                     pict->data[0][sx + 4 + (sy + y)*pict->linesize[0]]^= 0x80;
1444                             if(mv[0] != mv[dm*mv_stride] || mv[dm] != mv[dm*(mv_stride+1)])
1445                                 *(uint64_t*)(pict->data[0] + sx + (sy + 4)*pict->linesize[0])^= 0x8080808080808080ULL;
1446                         }
1447                     }
1448
1449                     if(IS_INTERLACED(mb_type) && s->codec_id == CODEC_ID_H264){
1450                         // hmm
1451                     }
1452                 }
1453                 s->mbskip_table[mb_index]=0;
1454             }
1455         }
1456     }
1457 }
1458
1459 static inline int hpel_motion_lowres(MpegEncContext *s,
1460                                   uint8_t *dest, uint8_t *src,
1461                                   int field_based, int field_select,
1462                                   int src_x, int src_y,
1463                                   int width, int height, int stride,
1464                                   int h_edge_pos, int v_edge_pos,
1465                                   int w, int h, h264_chroma_mc_func *pix_op,
1466                                   int motion_x, int motion_y)
1467 {
1468     const int lowres= s->avctx->lowres;
1469     const int op_index= FFMIN(lowres, 2);
1470     const int s_mask= (2<<lowres)-1;
1471     int emu=0;
1472     int sx, sy;
1473
1474     if(s->quarter_sample){
1475         motion_x/=2;
1476         motion_y/=2;
1477     }
1478
1479     sx= motion_x & s_mask;
1480     sy= motion_y & s_mask;
1481     src_x += motion_x >> (lowres+1);
1482     src_y += motion_y >> (lowres+1);
1483
1484     src += src_y * stride + src_x;
1485
1486     if(   (unsigned)src_x > h_edge_pos                 - (!!sx) - w
1487        || (unsigned)src_y >(v_edge_pos >> field_based) - (!!sy) - h){
1488         s->dsp.emulated_edge_mc(s->edge_emu_buffer, src, s->linesize, w+1, (h+1)<<field_based,
1489                             src_x, src_y<<field_based, h_edge_pos, v_edge_pos);
1490         src= s->edge_emu_buffer;
1491         emu=1;
1492     }
1493
1494     sx= (sx << 2) >> lowres;
1495     sy= (sy << 2) >> lowres;
1496     if(field_select)
1497         src += s->linesize;
1498     pix_op[op_index](dest, src, stride, h, sx, sy);
1499     return emu;
1500 }
1501
1502 /* apply one mpeg motion vector to the three components */
1503 static av_always_inline void mpeg_motion_lowres(MpegEncContext *s,
1504                                uint8_t *dest_y, uint8_t *dest_cb, uint8_t *dest_cr,
1505                                int field_based, int bottom_field, int field_select,
1506                                uint8_t **ref_picture, h264_chroma_mc_func *pix_op,
1507                                int motion_x, int motion_y, int h, int mb_y)
1508 {
1509     uint8_t *ptr_y, *ptr_cb, *ptr_cr;
1510     int mx, my, src_x, src_y, uvsrc_x, uvsrc_y, uvlinesize, linesize, sx, sy, uvsx, uvsy;
1511     const int lowres= s->avctx->lowres;
1512     const int op_index= FFMIN(lowres, 2);
1513     const int block_s= 8>>lowres;
1514     const int s_mask= (2<<lowres)-1;
1515     const int h_edge_pos = s->h_edge_pos >> lowres;
1516     const int v_edge_pos = s->v_edge_pos >> lowres;
1517     linesize   = s->current_picture.linesize[0] << field_based;
1518     uvlinesize = s->current_picture.linesize[1] << field_based;
1519
1520     if(s->quarter_sample){ //FIXME obviously not perfect but qpel will not work in lowres anyway
1521         motion_x/=2;
1522         motion_y/=2;
1523     }
1524
1525     if(field_based){
1526         motion_y += (bottom_field - field_select)*((1<<lowres)-1);
1527     }
1528
1529     sx= motion_x & s_mask;
1530     sy= motion_y & s_mask;
1531     src_x = s->mb_x*2*block_s               + (motion_x >> (lowres+1));
1532     src_y =(   mb_y*2*block_s>>field_based) + (motion_y >> (lowres+1));
1533
1534     if (s->out_format == FMT_H263) {
1535         uvsx = ((motion_x>>1) & s_mask) | (sx&1);
1536         uvsy = ((motion_y>>1) & s_mask) | (sy&1);
1537         uvsrc_x = src_x>>1;
1538         uvsrc_y = src_y>>1;
1539     }else if(s->out_format == FMT_H261){//even chroma mv's are full pel in H261
1540         mx = motion_x / 4;
1541         my = motion_y / 4;
1542         uvsx = (2*mx) & s_mask;
1543         uvsy = (2*my) & s_mask;
1544         uvsrc_x = s->mb_x*block_s               + (mx >> lowres);
1545         uvsrc_y =    mb_y*block_s               + (my >> lowres);
1546     } else {
1547         mx = motion_x / 2;
1548         my = motion_y / 2;
1549         uvsx = mx & s_mask;
1550         uvsy = my & s_mask;
1551         uvsrc_x = s->mb_x*block_s               + (mx >> (lowres+1));
1552         uvsrc_y =(   mb_y*block_s>>field_based) + (my >> (lowres+1));
1553     }
1554
1555     ptr_y  = ref_picture[0] + src_y * linesize + src_x;
1556     ptr_cb = ref_picture[1] + uvsrc_y * uvlinesize + uvsrc_x;
1557     ptr_cr = ref_picture[2] + uvsrc_y * uvlinesize + uvsrc_x;
1558
1559     if(   (unsigned)src_x > h_edge_pos                 - (!!sx) - 2*block_s
1560        || (unsigned)src_y >(v_edge_pos >> field_based) - (!!sy) - h){
1561             s->dsp.emulated_edge_mc(s->edge_emu_buffer, ptr_y, s->linesize, 17, 17+field_based,
1562                              src_x, src_y<<field_based, h_edge_pos, v_edge_pos);
1563             ptr_y = s->edge_emu_buffer;
1564             if(!CONFIG_GRAY || !(s->flags&CODEC_FLAG_GRAY)){
1565                 uint8_t *uvbuf= s->edge_emu_buffer+18*s->linesize;
1566                 s->dsp.emulated_edge_mc(uvbuf  , ptr_cb, s->uvlinesize, 9, 9+field_based,
1567                                  uvsrc_x, uvsrc_y<<field_based, h_edge_pos>>1, v_edge_pos>>1);
1568                 s->dsp.emulated_edge_mc(uvbuf+16, ptr_cr, s->uvlinesize, 9, 9+field_based,
1569                                  uvsrc_x, uvsrc_y<<field_based, h_edge_pos>>1, v_edge_pos>>1);
1570                 ptr_cb= uvbuf;
1571                 ptr_cr= uvbuf+16;
1572             }
1573     }
1574
1575     if(bottom_field){ //FIXME use this for field pix too instead of the obnoxious hack which changes picture.data
1576         dest_y += s->linesize;
1577         dest_cb+= s->uvlinesize;
1578         dest_cr+= s->uvlinesize;
1579     }
1580
1581     if(field_select){
1582         ptr_y += s->linesize;
1583         ptr_cb+= s->uvlinesize;
1584         ptr_cr+= s->uvlinesize;
1585     }
1586
1587     sx= (sx << 2) >> lowres;
1588     sy= (sy << 2) >> lowres;
1589     pix_op[lowres-1](dest_y, ptr_y, linesize, h, sx, sy);
1590
1591     if(!CONFIG_GRAY || !(s->flags&CODEC_FLAG_GRAY)){
1592         uvsx= (uvsx << 2) >> lowres;
1593         uvsy= (uvsy << 2) >> lowres;
1594         pix_op[op_index](dest_cb, ptr_cb, uvlinesize, h >> s->chroma_y_shift, uvsx, uvsy);
1595         pix_op[op_index](dest_cr, ptr_cr, uvlinesize, h >> s->chroma_y_shift, uvsx, uvsy);
1596     }
1597     //FIXME h261 lowres loop filter
1598 }
1599
1600 static inline void chroma_4mv_motion_lowres(MpegEncContext *s,
1601                                      uint8_t *dest_cb, uint8_t *dest_cr,
1602                                      uint8_t **ref_picture,
1603                                      h264_chroma_mc_func *pix_op,
1604                                      int mx, int my){
1605     const int lowres= s->avctx->lowres;
1606     const int op_index= FFMIN(lowres, 2);
1607     const int block_s= 8>>lowres;
1608     const int s_mask= (2<<lowres)-1;
1609     const int h_edge_pos = s->h_edge_pos >> (lowres+1);
1610     const int v_edge_pos = s->v_edge_pos >> (lowres+1);
1611     int emu=0, src_x, src_y, offset, sx, sy;
1612     uint8_t *ptr;
1613
1614     if(s->quarter_sample){
1615         mx/=2;
1616         my/=2;
1617     }
1618
1619     /* In case of 8X8, we construct a single chroma motion vector
1620        with a special rounding */
1621     mx= ff_h263_round_chroma(mx);
1622     my= ff_h263_round_chroma(my);
1623
1624     sx= mx & s_mask;
1625     sy= my & s_mask;
1626     src_x = s->mb_x*block_s + (mx >> (lowres+1));
1627     src_y = s->mb_y*block_s + (my >> (lowres+1));
1628
1629     offset = src_y * s->uvlinesize + src_x;
1630     ptr = ref_picture[1] + offset;
1631     if(s->flags&CODEC_FLAG_EMU_EDGE){
1632         if(   (unsigned)src_x > h_edge_pos - (!!sx) - block_s
1633            || (unsigned)src_y > v_edge_pos - (!!sy) - block_s){
1634             s->dsp.emulated_edge_mc(s->edge_emu_buffer, ptr, s->uvlinesize, 9, 9, src_x, src_y, h_edge_pos, v_edge_pos);
1635             ptr= s->edge_emu_buffer;
1636             emu=1;
1637         }
1638     }
1639     sx= (sx << 2) >> lowres;
1640     sy= (sy << 2) >> lowres;
1641     pix_op[op_index](dest_cb, ptr, s->uvlinesize, block_s, sx, sy);
1642
1643     ptr = ref_picture[2] + offset;
1644     if(emu){
1645         s->dsp.emulated_edge_mc(s->edge_emu_buffer, ptr, s->uvlinesize, 9, 9, src_x, src_y, h_edge_pos, v_edge_pos);
1646         ptr= s->edge_emu_buffer;
1647     }
1648     pix_op[op_index](dest_cr, ptr, s->uvlinesize, block_s, sx, sy);
1649 }
1650
1651 /**
1652  * motion compensation of a single macroblock
1653  * @param s context
1654  * @param dest_y luma destination pointer
1655  * @param dest_cb chroma cb/u destination pointer
1656  * @param dest_cr chroma cr/v destination pointer
1657  * @param dir direction (0->forward, 1->backward)
1658  * @param ref_picture array[3] of pointers to the 3 planes of the reference picture
1659  * @param pix_op halfpel motion compensation function (average or put normally)
1660  * the motion vectors are taken from s->mv and the MV type from s->mv_type
1661  */
1662 static inline void MPV_motion_lowres(MpegEncContext *s,
1663                               uint8_t *dest_y, uint8_t *dest_cb, uint8_t *dest_cr,
1664                               int dir, uint8_t **ref_picture,
1665                               h264_chroma_mc_func *pix_op)
1666 {
1667     int mx, my;
1668     int mb_x, mb_y, i;
1669     const int lowres= s->avctx->lowres;
1670     const int block_s= 8>>lowres;
1671
1672     mb_x = s->mb_x;
1673     mb_y = s->mb_y;
1674
1675     switch(s->mv_type) {
1676     case MV_TYPE_16X16:
1677         mpeg_motion_lowres(s, dest_y, dest_cb, dest_cr,
1678                     0, 0, 0,
1679                     ref_picture, pix_op,
1680                     s->mv[dir][0][0], s->mv[dir][0][1], 2*block_s, mb_y);
1681         break;
1682     case MV_TYPE_8X8:
1683         mx = 0;
1684         my = 0;
1685             for(i=0;i<4;i++) {
1686                 hpel_motion_lowres(s, dest_y + ((i & 1) + (i >> 1) * s->linesize)*block_s,
1687                             ref_picture[0], 0, 0,
1688                             (2*mb_x + (i & 1))*block_s, (2*mb_y + (i >>1))*block_s,
1689                             s->width, s->height, s->linesize,
1690                             s->h_edge_pos >> lowres, s->v_edge_pos >> lowres,
1691                             block_s, block_s, pix_op,
1692                             s->mv[dir][i][0], s->mv[dir][i][1]);
1693
1694                 mx += s->mv[dir][i][0];
1695                 my += s->mv[dir][i][1];
1696             }
1697
1698         if(!CONFIG_GRAY || !(s->flags&CODEC_FLAG_GRAY))
1699             chroma_4mv_motion_lowres(s, dest_cb, dest_cr, ref_picture, pix_op, mx, my);
1700         break;
1701     case MV_TYPE_FIELD:
1702         if (s->picture_structure == PICT_FRAME) {
1703             /* top field */
1704             mpeg_motion_lowres(s, dest_y, dest_cb, dest_cr,
1705                         1, 0, s->field_select[dir][0],
1706                         ref_picture, pix_op,
1707                         s->mv[dir][0][0], s->mv[dir][0][1], block_s, mb_y);
1708             /* bottom field */
1709             mpeg_motion_lowres(s, dest_y, dest_cb, dest_cr,
1710                         1, 1, s->field_select[dir][1],
1711                         ref_picture, pix_op,
1712                         s->mv[dir][1][0], s->mv[dir][1][1], block_s, mb_y);
1713         } else {
1714             if(s->picture_structure != s->field_select[dir][0] + 1 && s->pict_type != FF_B_TYPE && !s->first_field){
1715                 ref_picture= s->current_picture_ptr->data;
1716             }
1717
1718             mpeg_motion_lowres(s, dest_y, dest_cb, dest_cr,
1719                         0, 0, s->field_select[dir][0],
1720                         ref_picture, pix_op,
1721                         s->mv[dir][0][0], s->mv[dir][0][1], 2*block_s, mb_y>>1);
1722         }
1723         break;
1724     case MV_TYPE_16X8:
1725         for(i=0; i<2; i++){
1726             uint8_t ** ref2picture;
1727
1728             if(s->picture_structure == s->field_select[dir][i] + 1 || s->pict_type == FF_B_TYPE || s->first_field){
1729                 ref2picture= ref_picture;
1730             }else{
1731                 ref2picture= s->current_picture_ptr->data;
1732             }
1733
1734             mpeg_motion_lowres(s, dest_y, dest_cb, dest_cr,
1735                         0, 0, s->field_select[dir][i],
1736                         ref2picture, pix_op,
1737                         s->mv[dir][i][0], s->mv[dir][i][1] + 2*block_s*i, block_s, mb_y>>1);
1738
1739             dest_y += 2*block_s*s->linesize;
1740             dest_cb+= (2*block_s>>s->chroma_y_shift)*s->uvlinesize;
1741             dest_cr+= (2*block_s>>s->chroma_y_shift)*s->uvlinesize;
1742         }
1743         break;
1744     case MV_TYPE_DMV:
1745         if(s->picture_structure == PICT_FRAME){
1746             for(i=0; i<2; i++){
1747                 int j;
1748                 for(j=0; j<2; j++){
1749                     mpeg_motion_lowres(s, dest_y, dest_cb, dest_cr,
1750                                 1, j, j^i,
1751                                 ref_picture, pix_op,
1752                                 s->mv[dir][2*i + j][0], s->mv[dir][2*i + j][1], block_s, mb_y);
1753                 }
1754                 pix_op = s->dsp.avg_h264_chroma_pixels_tab;
1755             }
1756         }else{
1757             for(i=0; i<2; i++){
1758                 mpeg_motion_lowres(s, dest_y, dest_cb, dest_cr,
1759                             0, 0, s->picture_structure != i+1,
1760                             ref_picture, pix_op,
1761                             s->mv[dir][2*i][0],s->mv[dir][2*i][1],2*block_s, mb_y>>1);
1762
1763                 // after put we make avg of the same block
1764                 pix_op = s->dsp.avg_h264_chroma_pixels_tab;
1765
1766                 //opposite parity is always in the same frame if this is second field
1767                 if(!s->first_field){
1768                     ref_picture = s->current_picture_ptr->data;
1769                 }
1770             }
1771         }
1772     break;
1773     default: assert(0);
1774     }
1775 }
1776
1777 /* put block[] to dest[] */
1778 static inline void put_dct(MpegEncContext *s,
1779                            DCTELEM *block, int i, uint8_t *dest, int line_size, int qscale)
1780 {
1781     s->dct_unquantize_intra(s, block, i, qscale);
1782     s->dsp.idct_put (dest, line_size, block);
1783 }
1784
1785 /* add block[] to dest[] */
1786 static inline void add_dct(MpegEncContext *s,
1787                            DCTELEM *block, int i, uint8_t *dest, int line_size)
1788 {
1789     if (s->block_last_index[i] >= 0) {
1790         s->dsp.idct_add (dest, line_size, block);
1791     }
1792 }
1793
1794 static inline void add_dequant_dct(MpegEncContext *s,
1795                            DCTELEM *block, int i, uint8_t *dest, int line_size, int qscale)
1796 {
1797     if (s->block_last_index[i] >= 0) {
1798         s->dct_unquantize_inter(s, block, i, qscale);
1799
1800         s->dsp.idct_add (dest, line_size, block);
1801     }
1802 }
1803
1804 /**
1805  * cleans dc, ac, coded_block for the current non intra MB
1806  */
1807 void ff_clean_intra_table_entries(MpegEncContext *s)
1808 {
1809     int wrap = s->b8_stride;
1810     int xy = s->block_index[0];
1811
1812     s->dc_val[0][xy           ] =
1813     s->dc_val[0][xy + 1       ] =
1814     s->dc_val[0][xy     + wrap] =
1815     s->dc_val[0][xy + 1 + wrap] = 1024;
1816     /* ac pred */
1817     memset(s->ac_val[0][xy       ], 0, 32 * sizeof(int16_t));
1818     memset(s->ac_val[0][xy + wrap], 0, 32 * sizeof(int16_t));
1819     if (s->msmpeg4_version>=3) {
1820         s->coded_block[xy           ] =
1821         s->coded_block[xy + 1       ] =
1822         s->coded_block[xy     + wrap] =
1823         s->coded_block[xy + 1 + wrap] = 0;
1824     }
1825     /* chroma */
1826     wrap = s->mb_stride;
1827     xy = s->mb_x + s->mb_y * wrap;
1828     s->dc_val[1][xy] =
1829     s->dc_val[2][xy] = 1024;
1830     /* ac pred */
1831     memset(s->ac_val[1][xy], 0, 16 * sizeof(int16_t));
1832     memset(s->ac_val[2][xy], 0, 16 * sizeof(int16_t));
1833
1834     s->mbintra_table[xy]= 0;
1835 }
1836
1837 /* generic function called after a macroblock has been parsed by the
1838    decoder or after it has been encoded by the encoder.
1839
1840    Important variables used:
1841    s->mb_intra : true if intra macroblock
1842    s->mv_dir   : motion vector direction
1843    s->mv_type  : motion vector type
1844    s->mv       : motion vector
1845    s->interlaced_dct : true if interlaced dct used (mpeg2)
1846  */
1847 static av_always_inline
1848 void MPV_decode_mb_internal(MpegEncContext *s, DCTELEM block[12][64],
1849                             int lowres_flag, int is_mpeg12)
1850 {
1851     const int mb_xy = s->mb_y * s->mb_stride + s->mb_x;
1852     if(CONFIG_MPEG_XVMC_DECODER && s->avctx->xvmc_acceleration){
1853         ff_xvmc_decode_mb(s);//xvmc uses pblocks
1854         return;
1855     }
1856
1857     if(s->avctx->debug&FF_DEBUG_DCT_COEFF) {
1858        /* save DCT coefficients */
1859        int i,j;
1860        DCTELEM *dct = &s->current_picture.dct_coeff[mb_xy*64*6];
1861        for(i=0; i<6; i++)
1862            for(j=0; j<64; j++)
1863                *dct++ = block[i][s->dsp.idct_permutation[j]];
1864     }
1865
1866     s->current_picture.qscale_table[mb_xy]= s->qscale;
1867
1868     /* update DC predictors for P macroblocks */
1869     if (!s->mb_intra) {
1870         if (!is_mpeg12 && (s->h263_pred || s->h263_aic)) {
1871             if(s->mbintra_table[mb_xy])
1872                 ff_clean_intra_table_entries(s);
1873         } else {
1874             s->last_dc[0] =
1875             s->last_dc[1] =
1876             s->last_dc[2] = 128 << s->intra_dc_precision;
1877         }
1878     }
1879     else if (!is_mpeg12 && (s->h263_pred || s->h263_aic))
1880         s->mbintra_table[mb_xy]=1;
1881
1882     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
1883         uint8_t *dest_y, *dest_cb, *dest_cr;
1884         int dct_linesize, dct_offset;
1885         op_pixels_func (*op_pix)[4];
1886         qpel_mc_func (*op_qpix)[16];
1887         const int linesize= s->current_picture.linesize[0]; //not s->linesize as this would be wrong for field pics
1888         const int uvlinesize= s->current_picture.linesize[1];
1889         const int readable= s->pict_type != FF_B_TYPE || s->encoding || s->avctx->draw_horiz_band || lowres_flag;
1890         const int block_size= lowres_flag ? 8>>s->avctx->lowres : 8;
1891
1892         /* avoid copy if macroblock skipped in last frame too */
1893         /* skip only during decoding as we might trash the buffers during encoding a bit */
1894         if(!s->encoding){
1895             uint8_t *mbskip_ptr = &s->mbskip_table[mb_xy];
1896             const int age= s->current_picture.age;
1897
1898             assert(age);
1899
1900             if (s->mb_skipped) {
1901                 s->mb_skipped= 0;
1902                 assert(s->pict_type!=FF_I_TYPE);
1903
1904                 (*mbskip_ptr) ++; /* indicate that this time we skipped it */
1905                 if(*mbskip_ptr >99) *mbskip_ptr= 99;
1906
1907                 /* if previous was skipped too, then nothing to do !  */
1908                 if (*mbskip_ptr >= age && s->current_picture.reference){
1909                     return;
1910                 }
1911             } else if(!s->current_picture.reference){
1912                 (*mbskip_ptr) ++; /* increase counter so the age can be compared cleanly */
1913                 if(*mbskip_ptr >99) *mbskip_ptr= 99;
1914             } else{
1915                 *mbskip_ptr = 0; /* not skipped */
1916             }
1917         }
1918
1919         dct_linesize = linesize << s->interlaced_dct;
1920         dct_offset =(s->interlaced_dct)? linesize : linesize*block_size;
1921
1922         if(readable){
1923             dest_y=  s->dest[0];
1924             dest_cb= s->dest[1];
1925             dest_cr= s->dest[2];
1926         }else{
1927             dest_y = s->b_scratchpad;
1928             dest_cb= s->b_scratchpad+16*linesize;
1929             dest_cr= s->b_scratchpad+32*linesize;
1930         }
1931
1932         if (!s->mb_intra) {
1933             /* motion handling */
1934             /* decoding or more than one mb_type (MC was already done otherwise) */
1935             if(!s->encoding){
1936                 if(lowres_flag){
1937                     h264_chroma_mc_func *op_pix = s->dsp.put_h264_chroma_pixels_tab;
1938
1939                     if (s->mv_dir & MV_DIR_FORWARD) {
1940                         MPV_motion_lowres(s, dest_y, dest_cb, dest_cr, 0, s->last_picture.data, op_pix);
1941                         op_pix = s->dsp.avg_h264_chroma_pixels_tab;
1942                     }
1943                     if (s->mv_dir & MV_DIR_BACKWARD) {
1944                         MPV_motion_lowres(s, dest_y, dest_cb, dest_cr, 1, s->next_picture.data, op_pix);
1945                     }
1946                 }else{
1947                     op_qpix= s->me.qpel_put;
1948                     if ((!s->no_rounding) || s->pict_type==FF_B_TYPE){
1949                         op_pix = s->dsp.put_pixels_tab;
1950                     }else{
1951                         op_pix = s->dsp.put_no_rnd_pixels_tab;
1952                     }
1953                     if (s->mv_dir & MV_DIR_FORWARD) {
1954                         MPV_motion(s, dest_y, dest_cb, dest_cr, 0, s->last_picture.data, op_pix, op_qpix);
1955                         op_pix = s->dsp.avg_pixels_tab;
1956                         op_qpix= s->me.qpel_avg;
1957                     }
1958                     if (s->mv_dir & MV_DIR_BACKWARD) {
1959                         MPV_motion(s, dest_y, dest_cb, dest_cr, 1, s->next_picture.data, op_pix, op_qpix);
1960                     }
1961                 }
1962             }
1963
1964             /* skip dequant / idct if we are really late ;) */
1965             if(s->hurry_up>1) goto skip_idct;
1966             if(s->avctx->skip_idct){
1967                 if(  (s->avctx->skip_idct >= AVDISCARD_NONREF && s->pict_type == FF_B_TYPE)
1968                    ||(s->avctx->skip_idct >= AVDISCARD_NONKEY && s->pict_type != FF_I_TYPE)
1969                    || s->avctx->skip_idct >= AVDISCARD_ALL)
1970                     goto skip_idct;
1971             }
1972
1973             /* add dct residue */
1974             if(s->encoding || !(   s->h263_msmpeg4 || s->codec_id==CODEC_ID_MPEG1VIDEO || s->codec_id==CODEC_ID_MPEG2VIDEO
1975                                 || (s->codec_id==CODEC_ID_MPEG4 && !s->mpeg_quant))){
1976                 add_dequant_dct(s, block[0], 0, dest_y                          , dct_linesize, s->qscale);
1977                 add_dequant_dct(s, block[1], 1, dest_y              + block_size, dct_linesize, s->qscale);
1978                 add_dequant_dct(s, block[2], 2, dest_y + dct_offset             , dct_linesize, s->qscale);
1979                 add_dequant_dct(s, block[3], 3, dest_y + dct_offset + block_size, dct_linesize, s->qscale);
1980
1981                 if(!CONFIG_GRAY || !(s->flags&CODEC_FLAG_GRAY)){
1982                     if (s->chroma_y_shift){
1983                         add_dequant_dct(s, block[4], 4, dest_cb, uvlinesize, s->chroma_qscale);
1984                         add_dequant_dct(s, block[5], 5, dest_cr, uvlinesize, s->chroma_qscale);
1985                     }else{
1986                         dct_linesize >>= 1;
1987                         dct_offset >>=1;
1988                         add_dequant_dct(s, block[4], 4, dest_cb,              dct_linesize, s->chroma_qscale);
1989                         add_dequant_dct(s, block[5], 5, dest_cr,              dct_linesize, s->chroma_qscale);
1990                         add_dequant_dct(s, block[6], 6, dest_cb + dct_offset, dct_linesize, s->chroma_qscale);
1991                         add_dequant_dct(s, block[7], 7, dest_cr + dct_offset, dct_linesize, s->chroma_qscale);
1992                     }
1993                 }
1994             } else if(is_mpeg12 || (s->codec_id != CODEC_ID_WMV2)){
1995                 add_dct(s, block[0], 0, dest_y                          , dct_linesize);
1996                 add_dct(s, block[1], 1, dest_y              + block_size, dct_linesize);
1997                 add_dct(s, block[2], 2, dest_y + dct_offset             , dct_linesize);
1998                 add_dct(s, block[3], 3, dest_y + dct_offset + block_size, dct_linesize);
1999
2000                 if(!CONFIG_GRAY || !(s->flags&CODEC_FLAG_GRAY)){
2001                     if(s->chroma_y_shift){//Chroma420
2002                         add_dct(s, block[4], 4, dest_cb, uvlinesize);
2003                         add_dct(s, block[5], 5, dest_cr, uvlinesize);
2004                     }else{
2005                         //chroma422
2006                         dct_linesize = uvlinesize << s->interlaced_dct;
2007                         dct_offset =(s->interlaced_dct)? uvlinesize : uvlinesize*8;
2008
2009                         add_dct(s, block[4], 4, dest_cb, dct_linesize);
2010                         add_dct(s, block[5], 5, dest_cr, dct_linesize);
2011                         add_dct(s, block[6], 6, dest_cb+dct_offset, dct_linesize);
2012                         add_dct(s, block[7], 7, dest_cr+dct_offset, dct_linesize);
2013                         if(!s->chroma_x_shift){//Chroma444
2014                             add_dct(s, block[8], 8, dest_cb+8, dct_linesize);
2015                             add_dct(s, block[9], 9, dest_cr+8, dct_linesize);
2016                             add_dct(s, block[10], 10, dest_cb+8+dct_offset, dct_linesize);
2017                             add_dct(s, block[11], 11, dest_cr+8+dct_offset, dct_linesize);
2018                         }
2019                     }
2020                 }//fi gray
2021             }
2022             else if (CONFIG_WMV2_DECODER || CONFIG_WMV2_ENCODER) {
2023                 ff_wmv2_add_mb(s, block, dest_y, dest_cb, dest_cr);
2024             }
2025         } else {
2026             /* dct only in intra block */
2027             if(s->encoding || !(s->codec_id==CODEC_ID_MPEG1VIDEO || s->codec_id==CODEC_ID_MPEG2VIDEO)){
2028                 put_dct(s, block[0], 0, dest_y                          , dct_linesize, s->qscale);
2029                 put_dct(s, block[1], 1, dest_y              + block_size, dct_linesize, s->qscale);
2030                 put_dct(s, block[2], 2, dest_y + dct_offset             , dct_linesize, s->qscale);
2031                 put_dct(s, block[3], 3, dest_y + dct_offset + block_size, dct_linesize, s->qscale);
2032
2033                 if(!CONFIG_GRAY || !(s->flags&CODEC_FLAG_GRAY)){
2034                     if(s->chroma_y_shift){
2035                         put_dct(s, block[4], 4, dest_cb, uvlinesize, s->chroma_qscale);
2036                         put_dct(s, block[5], 5, dest_cr, uvlinesize, s->chroma_qscale);
2037                     }else{
2038                         dct_offset >>=1;
2039                         dct_linesize >>=1;
2040                         put_dct(s, block[4], 4, dest_cb,              dct_linesize, s->chroma_qscale);
2041                         put_dct(s, block[5], 5, dest_cr,              dct_linesize, s->chroma_qscale);
2042                         put_dct(s, block[6], 6, dest_cb + dct_offset, dct_linesize, s->chroma_qscale);
2043                         put_dct(s, block[7], 7, dest_cr + dct_offset, dct_linesize, s->chroma_qscale);
2044                     }
2045                 }
2046             }else{
2047                 s->dsp.idct_put(dest_y                          , dct_linesize, block[0]);
2048                 s->dsp.idct_put(dest_y              + block_size, dct_linesize, block[1]);
2049                 s->dsp.idct_put(dest_y + dct_offset             , dct_linesize, block[2]);
2050                 s->dsp.idct_put(dest_y + dct_offset + block_size, dct_linesize, block[3]);
2051
2052                 if(!CONFIG_GRAY || !(s->flags&CODEC_FLAG_GRAY)){
2053                     if(s->chroma_y_shift){
2054                         s->dsp.idct_put(dest_cb, uvlinesize, block[4]);
2055                         s->dsp.idct_put(dest_cr, uvlinesize, block[5]);
2056                     }else{
2057
2058                         dct_linesize = uvlinesize << s->interlaced_dct;
2059                         dct_offset =(s->interlaced_dct)? uvlinesize : uvlinesize*8;
2060
2061                         s->dsp.idct_put(dest_cb,              dct_linesize, block[4]);
2062                         s->dsp.idct_put(dest_cr,              dct_linesize, block[5]);
2063                         s->dsp.idct_put(dest_cb + dct_offset, dct_linesize, block[6]);
2064                         s->dsp.idct_put(dest_cr + dct_offset, dct_linesize, block[7]);
2065                         if(!s->chroma_x_shift){//Chroma444
2066                             s->dsp.idct_put(dest_cb + 8,              dct_linesize, block[8]);
2067                             s->dsp.idct_put(dest_cr + 8,              dct_linesize, block[9]);
2068                             s->dsp.idct_put(dest_cb + 8 + dct_offset, dct_linesize, block[10]);
2069                             s->dsp.idct_put(dest_cr + 8 + dct_offset, dct_linesize, block[11]);
2070                         }
2071                     }
2072                 }//gray
2073             }
2074         }
2075 skip_idct:
2076         if(!readable){
2077             s->dsp.put_pixels_tab[0][0](s->dest[0], dest_y ,   linesize,16);
2078             s->dsp.put_pixels_tab[s->chroma_x_shift][0](s->dest[1], dest_cb, uvlinesize,16 >> s->chroma_y_shift);
2079             s->dsp.put_pixels_tab[s->chroma_x_shift][0](s->dest[2], dest_cr, uvlinesize,16 >> s->chroma_y_shift);
2080         }
2081     }
2082 }
2083
2084 void MPV_decode_mb(MpegEncContext *s, DCTELEM block[12][64]){
2085 #if !CONFIG_SMALL
2086     if(s->out_format == FMT_MPEG1) {
2087         if(s->avctx->lowres) MPV_decode_mb_internal(s, block, 1, 1);
2088         else                 MPV_decode_mb_internal(s, block, 0, 1);
2089     } else
2090 #endif
2091     if(s->avctx->lowres) MPV_decode_mb_internal(s, block, 1, 0);
2092     else                  MPV_decode_mb_internal(s, block, 0, 0);
2093 }
2094
2095 /**
2096  *
2097  * @param h is the normal height, this will be reduced automatically if needed for the last row
2098  */
2099 void ff_draw_horiz_band(MpegEncContext *s, int y, int h){
2100     if (s->avctx->draw_horiz_band) {
2101         AVFrame *src;
2102         const int field_pic= s->picture_structure != PICT_FRAME;
2103         int offset[4];
2104
2105         h= FFMIN(h, (s->avctx->height>>field_pic) - y);
2106
2107         if(field_pic && !(s->avctx->slice_flags&SLICE_FLAG_ALLOW_FIELD)){
2108             h <<= 1;
2109             y <<= 1;
2110             if(s->first_field) return;
2111         }
2112
2113         if(s->pict_type==FF_B_TYPE || s->low_delay || (s->avctx->slice_flags&SLICE_FLAG_CODED_ORDER))
2114             src= (AVFrame*)s->current_picture_ptr;
2115         else if(s->last_picture_ptr)
2116             src= (AVFrame*)s->last_picture_ptr;
2117         else
2118             return;
2119
2120         if(s->pict_type==FF_B_TYPE && s->picture_structure == PICT_FRAME && s->out_format != FMT_H264){
2121             offset[0]=
2122             offset[1]=
2123             offset[2]=
2124             offset[3]= 0;
2125         }else{
2126             offset[0]= y * s->linesize;
2127             offset[1]=
2128             offset[2]= (y >> s->chroma_y_shift) * s->uvlinesize;
2129             offset[3]= 0;
2130         }
2131
2132         emms_c();
2133
2134         s->avctx->draw_horiz_band(s->avctx, src, offset,
2135                                   y, s->picture_structure, h);
2136     }
2137 }
2138
2139 void ff_init_block_index(MpegEncContext *s){ //FIXME maybe rename
2140     const int linesize= s->current_picture.linesize[0]; //not s->linesize as this would be wrong for field pics
2141     const int uvlinesize= s->current_picture.linesize[1];
2142     const int mb_size= 4 - s->avctx->lowres;
2143
2144     s->block_index[0]= s->b8_stride*(s->mb_y*2    ) - 2 + s->mb_x*2;
2145     s->block_index[1]= s->b8_stride*(s->mb_y*2    ) - 1 + s->mb_x*2;
2146     s->block_index[2]= s->b8_stride*(s->mb_y*2 + 1) - 2 + s->mb_x*2;
2147     s->block_index[3]= s->b8_stride*(s->mb_y*2 + 1) - 1 + s->mb_x*2;
2148     s->block_index[4]= s->mb_stride*(s->mb_y + 1)                + s->b8_stride*s->mb_height*2 + s->mb_x - 1;
2149     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;
2150     //block_index is not used by mpeg2, so it is not affected by chroma_format
2151
2152     s->dest[0] = s->current_picture.data[0] + ((s->mb_x - 1) << mb_size);
2153     s->dest[1] = s->current_picture.data[1] + ((s->mb_x - 1) << (mb_size - s->chroma_x_shift));
2154     s->dest[2] = s->current_picture.data[2] + ((s->mb_x - 1) << (mb_size - s->chroma_x_shift));
2155
2156     if(!(s->pict_type==FF_B_TYPE && s->avctx->draw_horiz_band && s->picture_structure==PICT_FRAME))
2157     {
2158         if(s->picture_structure==PICT_FRAME){
2159         s->dest[0] += s->mb_y *   linesize << mb_size;
2160         s->dest[1] += s->mb_y * uvlinesize << (mb_size - s->chroma_y_shift);
2161         s->dest[2] += s->mb_y * uvlinesize << (mb_size - s->chroma_y_shift);
2162         }else{
2163             s->dest[0] += (s->mb_y>>1) *   linesize << mb_size;
2164             s->dest[1] += (s->mb_y>>1) * uvlinesize << (mb_size - s->chroma_y_shift);
2165             s->dest[2] += (s->mb_y>>1) * uvlinesize << (mb_size - s->chroma_y_shift);
2166             assert((s->mb_y&1) == (s->picture_structure == PICT_BOTTOM_FIELD));
2167         }
2168     }
2169 }
2170
2171 void ff_mpeg_flush(AVCodecContext *avctx){
2172     int i;
2173     MpegEncContext *s = avctx->priv_data;
2174
2175     if(s==NULL || s->picture==NULL)
2176         return;
2177
2178     for(i=0; i<MAX_PICTURE_COUNT; i++){
2179        if(s->picture[i].data[0] && (   s->picture[i].type == FF_BUFFER_TYPE_INTERNAL
2180                                     || s->picture[i].type == FF_BUFFER_TYPE_USER))
2181         free_frame_buffer(s, &s->picture[i]);
2182     }
2183     s->current_picture_ptr = s->last_picture_ptr = s->next_picture_ptr = NULL;
2184
2185     s->mb_x= s->mb_y= 0;
2186     s->closed_gop= 0;
2187
2188     s->parse_context.state= -1;
2189     s->parse_context.frame_start_found= 0;
2190     s->parse_context.overread= 0;
2191     s->parse_context.overread_index= 0;
2192     s->parse_context.index= 0;
2193     s->parse_context.last_index= 0;
2194     s->bitstream_buffer_size=0;
2195     s->pp_time=0;
2196 }
2197
2198 static void dct_unquantize_mpeg1_intra_c(MpegEncContext *s,
2199                                    DCTELEM *block, int n, int qscale)
2200 {
2201     int i, level, nCoeffs;
2202     const uint16_t *quant_matrix;
2203
2204     nCoeffs= s->block_last_index[n];
2205
2206     if (n < 4)
2207         block[0] = block[0] * s->y_dc_scale;
2208     else
2209         block[0] = block[0] * s->c_dc_scale;
2210     /* XXX: only mpeg1 */
2211     quant_matrix = s->intra_matrix;
2212     for(i=1;i<=nCoeffs;i++) {
2213         int j= s->intra_scantable.permutated[i];
2214         level = block[j];
2215         if (level) {
2216             if (level < 0) {
2217                 level = -level;
2218                 level = (int)(level * qscale * quant_matrix[j]) >> 3;
2219                 level = (level - 1) | 1;
2220                 level = -level;
2221             } else {
2222                 level = (int)(level * qscale * quant_matrix[j]) >> 3;
2223                 level = (level - 1) | 1;
2224             }
2225             block[j] = level;
2226         }
2227     }
2228 }
2229
2230 static void dct_unquantize_mpeg1_inter_c(MpegEncContext *s,
2231                                    DCTELEM *block, int n, int qscale)
2232 {
2233     int i, level, nCoeffs;
2234     const uint16_t *quant_matrix;
2235
2236     nCoeffs= s->block_last_index[n];
2237
2238     quant_matrix = s->inter_matrix;
2239     for(i=0; i<=nCoeffs; i++) {
2240         int j= s->intra_scantable.permutated[i];
2241         level = block[j];
2242         if (level) {
2243             if (level < 0) {
2244                 level = -level;
2245                 level = (((level << 1) + 1) * qscale *
2246                          ((int) (quant_matrix[j]))) >> 4;
2247                 level = (level - 1) | 1;
2248                 level = -level;
2249             } else {
2250                 level = (((level << 1) + 1) * qscale *
2251                          ((int) (quant_matrix[j]))) >> 4;
2252                 level = (level - 1) | 1;
2253             }
2254             block[j] = level;
2255         }
2256     }
2257 }
2258
2259 static void dct_unquantize_mpeg2_intra_c(MpegEncContext *s,
2260                                    DCTELEM *block, int n, int qscale)
2261 {
2262     int i, level, nCoeffs;
2263     const uint16_t *quant_matrix;
2264
2265     if(s->alternate_scan) nCoeffs= 63;
2266     else nCoeffs= s->block_last_index[n];
2267
2268     if (n < 4)
2269         block[0] = block[0] * s->y_dc_scale;
2270     else
2271         block[0] = block[0] * s->c_dc_scale;
2272     quant_matrix = s->intra_matrix;
2273     for(i=1;i<=nCoeffs;i++) {
2274         int j= s->intra_scantable.permutated[i];
2275         level = block[j];
2276         if (level) {
2277             if (level < 0) {
2278                 level = -level;
2279                 level = (int)(level * qscale * quant_matrix[j]) >> 3;
2280                 level = -level;
2281             } else {
2282                 level = (int)(level * qscale * quant_matrix[j]) >> 3;
2283             }
2284             block[j] = level;
2285         }
2286     }
2287 }
2288
2289 static void dct_unquantize_mpeg2_intra_bitexact(MpegEncContext *s,
2290                                    DCTELEM *block, int n, int qscale)
2291 {
2292     int i, level, nCoeffs;
2293     const uint16_t *quant_matrix;
2294     int sum=-1;
2295
2296     if(s->alternate_scan) nCoeffs= 63;
2297     else nCoeffs= s->block_last_index[n];
2298
2299     if (n < 4)
2300         block[0] = block[0] * s->y_dc_scale;
2301     else
2302         block[0] = block[0] * s->c_dc_scale;
2303     quant_matrix = s->intra_matrix;
2304     for(i=1;i<=nCoeffs;i++) {
2305         int j= s->intra_scantable.permutated[i];
2306         level = block[j];
2307         if (level) {
2308             if (level < 0) {
2309                 level = -level;
2310                 level = (int)(level * qscale * quant_matrix[j]) >> 3;
2311                 level = -level;
2312             } else {
2313                 level = (int)(level * qscale * quant_matrix[j]) >> 3;
2314             }
2315             block[j] = level;
2316             sum+=level;
2317         }
2318     }
2319     block[63]^=sum&1;
2320 }
2321
2322 static void dct_unquantize_mpeg2_inter_c(MpegEncContext *s,
2323                                    DCTELEM *block, int n, int qscale)
2324 {
2325     int i, level, nCoeffs;
2326     const uint16_t *quant_matrix;
2327     int sum=-1;
2328
2329     if(s->alternate_scan) nCoeffs= 63;
2330     else nCoeffs= s->block_last_index[n];
2331
2332     quant_matrix = s->inter_matrix;
2333     for(i=0; i<=nCoeffs; i++) {
2334         int j= s->intra_scantable.permutated[i];
2335         level = block[j];
2336         if (level) {
2337             if (level < 0) {
2338                 level = -level;
2339                 level = (((level << 1) + 1) * qscale *
2340                          ((int) (quant_matrix[j]))) >> 4;
2341                 level = -level;
2342             } else {
2343                 level = (((level << 1) + 1) * qscale *
2344                          ((int) (quant_matrix[j]))) >> 4;
2345             }
2346             block[j] = level;
2347             sum+=level;
2348         }
2349     }
2350     block[63]^=sum&1;
2351 }
2352
2353 static void dct_unquantize_h263_intra_c(MpegEncContext *s,
2354                                   DCTELEM *block, int n, int qscale)
2355 {
2356     int i, level, qmul, qadd;
2357     int nCoeffs;
2358
2359     assert(s->block_last_index[n]>=0);
2360
2361     qmul = qscale << 1;
2362
2363     if (!s->h263_aic) {
2364         if (n < 4)
2365             block[0] = block[0] * s->y_dc_scale;
2366         else
2367             block[0] = block[0] * s->c_dc_scale;
2368         qadd = (qscale - 1) | 1;
2369     }else{
2370         qadd = 0;
2371     }
2372     if(s->ac_pred)
2373         nCoeffs=63;
2374     else
2375         nCoeffs= s->inter_scantable.raster_end[ s->block_last_index[n] ];
2376
2377     for(i=1; i<=nCoeffs; i++) {
2378         level = block[i];
2379         if (level) {
2380             if (level < 0) {
2381                 level = level * qmul - qadd;
2382             } else {
2383                 level = level * qmul + qadd;
2384             }
2385             block[i] = level;
2386         }
2387     }
2388 }
2389
2390 static void dct_unquantize_h263_inter_c(MpegEncContext *s,
2391                                   DCTELEM *block, int n, int qscale)
2392 {
2393     int i, level, qmul, qadd;
2394     int nCoeffs;
2395
2396     assert(s->block_last_index[n]>=0);
2397
2398     qadd = (qscale - 1) | 1;
2399     qmul = qscale << 1;
2400
2401     nCoeffs= s->inter_scantable.raster_end[ s->block_last_index[n] ];
2402
2403     for(i=0; i<=nCoeffs; i++) {
2404         level = block[i];
2405         if (level) {
2406             if (level < 0) {
2407                 level = level * qmul - qadd;
2408             } else {
2409                 level = level * qmul + qadd;
2410             }
2411             block[i] = level;
2412         }
2413     }
2414 }
2415
2416 /**
2417  * set qscale and update qscale dependent variables.
2418  */
2419 void ff_set_qscale(MpegEncContext * s, int qscale)
2420 {
2421     if (qscale < 1)
2422         qscale = 1;
2423     else if (qscale > 31)
2424         qscale = 31;
2425
2426     s->qscale = qscale;
2427     s->chroma_qscale= s->chroma_qscale_table[qscale];
2428
2429     s->y_dc_scale= s->y_dc_scale_table[ qscale ];
2430     s->c_dc_scale= s->c_dc_scale_table[ s->chroma_qscale ];
2431 }