optionally (use_lpc=2) support Cholesky factorization for finding the lpc coeficients
[ffmpeg.git] / libavcodec / xvmcvideo.c
1 /*
2  * XVideo Motion Compensation
3  * Copyright (c) 2003 Ivan Kalvachev
4  *
5  * This library is free software; you can redistribute it and/or
6  * modify it under the terms of the GNU Lesser General Public
7  * License as published by the Free Software Foundation; either
8  * version 2 of the License, or (at your option) any later version.
9  *
10  * This library is distributed in the hope that it will be useful,
11  * but WITHOUT ANY WARRANTY; without even the implied warranty of
12  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
13  * Lesser General Public License for more details.
14  *
15  * You should have received a copy of the GNU Lesser General Public
16  * License along with this library; if not, write to the Free Software
17  * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
18  */
19
20 #include <limits.h>
21
22 //avcodec include
23 #include "avcodec.h"
24 #include "dsputil.h"
25 #include "mpegvideo.h"
26
27 #undef NDEBUG
28 #include <assert.h>
29
30 #ifdef USE_FASTMEMCPY
31 #include "fastmemcpy.h"
32 #endif
33
34 #ifdef HAVE_XVMC
35
36 //X11 includes are in the xvmc_render.h
37 //by replacing it with none-X one
38 //XvMC emulation could be performed
39
40 #include "xvmc_render.h"
41
42 //#include "xvmc_debug.h"
43
44 //set s->block
45 inline void XVMC_init_block(MpegEncContext *s){
46 xvmc_render_state_t * render;
47     render = (xvmc_render_state_t*)s->current_picture.data[2];
48     assert(render != NULL);
49     if( (render == NULL) || (render->magic != MP_XVMC_RENDER_MAGIC) ){
50         assert(0);
51         return;//make sure that this is render packet
52     }
53     s->block =(DCTELEM *)(render->data_blocks+(render->next_free_data_block_num)*64);
54 }
55
56 void XVMC_pack_pblocks(MpegEncContext *s, int cbp){
57 int i,j;
58 const int mb_block_count = 4+(1<<s->chroma_format);
59
60     j=0;
61     cbp<<= 12-mb_block_count;
62     for(i=0; i<mb_block_count; i++){
63         if(cbp & (1<<11)) {
64            s->pblocks[i] = (short *)(&s->block[(j++)]);
65         }else{
66            s->pblocks[i] = NULL;
67         }
68         cbp+=cbp;
69 //        printf("s->pblocks[%d]=%p ,s->block=%p cbp=%d\n",i,s->pblocks[i],s->block,cbp);
70     }
71 }
72
73 //these functions should be called on every new field or/and frame
74 //They should be safe if they are called few times for same field!
75 int XVMC_field_start(MpegEncContext*s, AVCodecContext *avctx){
76 xvmc_render_state_t * render,* last, * next;
77
78     assert(avctx != NULL);
79
80     render = (xvmc_render_state_t*)s->current_picture.data[2];
81     assert(render != NULL);
82     if( (render == NULL) || (render->magic != MP_XVMC_RENDER_MAGIC) )
83         return -1;//make sure that this is render packet
84
85     render->picture_structure = s->picture_structure;
86     render->flags = (s->first_field)? 0: XVMC_SECOND_FIELD;
87
88 //make sure that all data is drawn by XVMC_end_frame
89     assert(render->filled_mv_blocks_num==0);
90
91     render->p_future_surface = NULL;
92     render->p_past_surface = NULL;
93
94     switch(s->pict_type){
95         case  I_TYPE:
96             return 0;// no prediction from other frames
97         case  B_TYPE:
98             next = (xvmc_render_state_t*)s->next_picture.data[2];
99             assert(next!=NULL);
100             assert(next->state & MP_XVMC_STATE_PREDICTION);
101             if(next == NULL) return -1;
102             if(next->magic != MP_XVMC_RENDER_MAGIC) return -1;
103             render->p_future_surface = next->p_surface;
104             //no return here, going to set forward prediction
105         case  P_TYPE:
106             last = (xvmc_render_state_t*)s->last_picture.data[2];
107             if(last == NULL)// && !s->first_field)
108                 last = render;//predict second field from the first
109             if(last->magic != MP_XVMC_RENDER_MAGIC) return -1;
110             assert(last->state & MP_XVMC_STATE_PREDICTION);
111             render->p_past_surface = last->p_surface;
112             return 0;
113     }
114
115 return -1;
116 }
117
118 void XVMC_field_end(MpegEncContext *s){
119 xvmc_render_state_t * render;
120     render = (xvmc_render_state_t*)s->current_picture.data[2];
121     assert(render != NULL);
122
123     if(render->filled_mv_blocks_num > 0){
124 //        printf("xvmcvideo.c: rendering %d left blocks after last slice!!!\n",render->filled_mv_blocks_num );
125         ff_draw_horiz_band(s,0,0);
126     }
127 }
128
129 void XVMC_decode_mb(MpegEncContext *s){
130 XvMCMacroBlock * mv_block;
131 xvmc_render_state_t * render;
132 int i,cbp,blocks_per_mb;
133
134 const int mb_xy = s->mb_y * s->mb_stride + s->mb_x;
135
136
137     if(s->encoding){
138         av_log(s->avctx, AV_LOG_ERROR, "XVMC doesn't support encoding!!!\n");
139         return -1;
140     }
141
142    //from MPV_decode_mb(),
143     /* update DC predictors for P macroblocks */
144     if (!s->mb_intra) {
145         s->last_dc[0] =
146         s->last_dc[1] =
147         s->last_dc[2] =  128 << s->intra_dc_precision;
148     }
149
150    //MC doesn't skip blocks
151     s->mb_skipped = 0;
152
153
154    // do I need to export quant when I could not perform postprocessing?
155    // anyway, it doesn't hurrt
156     s->current_picture.qscale_table[mb_xy] = s->qscale;
157
158 //START OF XVMC specific code
159     render = (xvmc_render_state_t*)s->current_picture.data[2];
160     assert(render!=NULL);
161     assert(render->magic==MP_XVMC_RENDER_MAGIC);
162     assert(render->mv_blocks);
163
164     //take the next free macroblock
165     mv_block = &render->mv_blocks[render->start_mv_blocks_num +
166                                    render->filled_mv_blocks_num ];
167
168 // memset(mv_block,0,sizeof(XvMCMacroBlock));
169
170     mv_block->x = s->mb_x;
171     mv_block->y = s->mb_y;
172     mv_block->dct_type = s->interlaced_dct;//XVMC_DCT_TYPE_FRAME/FIELD;
173 //    mv_block->motion_type = 0;  //zero to silense warnings
174     if(s->mb_intra){
175         mv_block->macroblock_type = XVMC_MB_TYPE_INTRA;//no MC, all done
176     }else{
177         mv_block->macroblock_type = XVMC_MB_TYPE_PATTERN;
178
179         if(s->mv_dir & MV_DIR_FORWARD){
180             mv_block->macroblock_type|= XVMC_MB_TYPE_MOTION_FORWARD;
181             //pmv[n][dir][xy]=mv[dir][n][xy]
182             mv_block->PMV[0][0][0] = s->mv[0][0][0];
183             mv_block->PMV[0][0][1] = s->mv[0][0][1];
184             mv_block->PMV[1][0][0] = s->mv[0][1][0];
185             mv_block->PMV[1][0][1] = s->mv[0][1][1];
186         }
187         if(s->mv_dir & MV_DIR_BACKWARD){
188             mv_block->macroblock_type|=XVMC_MB_TYPE_MOTION_BACKWARD;
189             mv_block->PMV[0][1][0] = s->mv[1][0][0];
190             mv_block->PMV[0][1][1] = s->mv[1][0][1];
191             mv_block->PMV[1][1][0] = s->mv[1][1][0];
192             mv_block->PMV[1][1][1] = s->mv[1][1][1];
193         }
194
195         switch(s->mv_type){
196             case  MV_TYPE_16X16:
197                 mv_block->motion_type = XVMC_PREDICTION_FRAME;
198                 break;
199             case  MV_TYPE_16X8:
200                 mv_block->motion_type = XVMC_PREDICTION_16x8;
201                 break;
202             case  MV_TYPE_FIELD:
203                 mv_block->motion_type = XVMC_PREDICTION_FIELD;
204                 if(s->picture_structure == PICT_FRAME){
205                     mv_block->PMV[0][0][1]<<=1;
206                     mv_block->PMV[1][0][1]<<=1;
207                     mv_block->PMV[0][1][1]<<=1;
208                     mv_block->PMV[1][1][1]<<=1;
209                 }
210                 break;
211             case  MV_TYPE_DMV:
212                 mv_block->motion_type = XVMC_PREDICTION_DUAL_PRIME;
213                 if(s->picture_structure == PICT_FRAME){
214
215                     mv_block->PMV[0][0][0] = s->mv[0][0][0];//top from top
216                     mv_block->PMV[0][0][1] = s->mv[0][0][1]<<1;
217
218                     mv_block->PMV[0][1][0] = s->mv[0][0][0];//bottom from bottom
219                     mv_block->PMV[0][1][1] = s->mv[0][0][1]<<1;
220
221                     mv_block->PMV[1][0][0] = s->mv[0][2][0];//dmv00, top from bottom
222                     mv_block->PMV[1][0][1] = s->mv[0][2][1]<<1;//dmv01
223
224                     mv_block->PMV[1][1][0] = s->mv[0][3][0];//dmv10, bottom from top
225                     mv_block->PMV[1][1][1] = s->mv[0][3][1]<<1;//dmv11
226
227                 }else{
228                     mv_block->PMV[0][1][0] = s->mv[0][2][0];//dmv00
229                     mv_block->PMV[0][1][1] = s->mv[0][2][1];//dmv01
230                 }
231                 break;
232             default:
233                 assert(0);
234         }
235
236         mv_block->motion_vertical_field_select = 0;
237
238 //set correct field referenses
239         if(s->mv_type == MV_TYPE_FIELD || s->mv_type == MV_TYPE_16X8){
240             if( s->field_select[0][0] ) mv_block->motion_vertical_field_select|=1;
241             if( s->field_select[1][0] ) mv_block->motion_vertical_field_select|=2;
242             if( s->field_select[0][1] ) mv_block->motion_vertical_field_select|=4;
243             if( s->field_select[1][1] ) mv_block->motion_vertical_field_select|=8;
244         }
245     }//!intra
246 //time to handle data blocks;
247     mv_block->index = render->next_free_data_block_num;
248
249     blocks_per_mb = 6;
250     if( s->chroma_format >= 2){
251         blocks_per_mb = 4 + (1 << (s->chroma_format));
252     }
253
254 //  calculate cbp
255     cbp = 0;
256     for(i=0; i<blocks_per_mb; i++) {
257         cbp+= cbp;
258         if(s->block_last_index[i] >= 0)
259             cbp++;
260     }
261
262     if(s->flags & CODEC_FLAG_GRAY){
263         if(s->mb_intra){//intra frames are alwasy full chroma block
264             for(i=4; i<blocks_per_mb; i++){
265                 memset(s->pblocks[i],0,sizeof(short)*8*8);//so we need to clear them
266                 if(!render->unsigned_intra)
267                     s->pblocks[i][0] = 1<<10;
268             }
269         }else{
270             cbp&= 0xf << (blocks_per_mb - 4);
271             blocks_per_mb = 4;//Luminance blocks only
272         }
273     }
274     mv_block->coded_block_pattern = cbp;
275     if(cbp == 0)
276         mv_block->macroblock_type &= ~XVMC_MB_TYPE_PATTERN;
277
278     for(i=0; i<blocks_per_mb; i++){
279         if(s->block_last_index[i] >= 0){
280             // i do not have unsigned_intra MOCO to test, hope it is OK
281             if( (s->mb_intra) && ( render->idct || (!render->idct && !render->unsigned_intra)) )
282                 s->pblocks[i][0]-=1<<10;
283             if(!render->idct){
284                 s->dsp.idct(s->pblocks[i]);
285                 //!!TODO!clip!!!
286             }
287 //copy blocks only if the codec doesn't support pblocks reordering
288             if(s->avctx->xvmc_acceleration == 1){
289                 memcpy(&render->data_blocks[(render->next_free_data_block_num)*64],
290                         s->pblocks[i],sizeof(short)*8*8);
291             }else{
292 /*              if(s->pblocks[i] != &render->data_blocks[
293                         (render->next_free_data_block_num)*64]){
294                    printf("ERROR mb(%d,%d) s->pblocks[i]=%p data_block[]=%p\n",
295                    s->mb_x,s->mb_y, s->pblocks[i],
296                    &render->data_blocks[(render->next_free_data_block_num)*64]);
297                 }*/
298             }
299             render->next_free_data_block_num++;
300         }
301     }
302     render->filled_mv_blocks_num++;
303
304     assert(render->filled_mv_blocks_num <= render->total_number_of_mv_blocks);
305     assert(render->next_free_data_block_num <= render->total_number_of_data_blocks);
306
307
308     if(render->filled_mv_blocks_num >= render->total_number_of_mv_blocks)
309         ff_draw_horiz_band(s,0,0);
310
311 // DumpRenderInfo(render);
312 // DumpMBlockInfo(mv_block);
313
314 }
315
316 #endif