initial chroma_format changes,xvmc tweaks and codec_cap
[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., 59 Temple Place, Suite 330, Boston, MA  02111-1307  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
59     j=0;
60     for(i=0;i<6;i++){
61         if(cbp & (1<<(5-i)) ){
62            s->pblocks[i] = (short *)(&s->block[(j++)]);
63         }else{
64            s->pblocks[i] = NULL;
65         }
66 //        printf("s->pblocks[%d]=%p ,s->block=%p cbp=%d\n",i,s->pblocks[i],s->block,cbp);
67     }
68     if (s->chroma_format >= 2){
69         if (s->chroma_format == 2){//CHROMA_422
70             for(i=6;i<8;i++){
71                 if(cbp & (1<<(6+7-i)) ){
72                     s->pblocks[i] = (short *)(&s->block[(j++)]);
73                 }else{
74                     s->pblocks[i] = NULL;
75                 }
76             }
77         }else{//CHROMA_444
78             for(i=6; i<12; i++){
79                 if(cbp & (1<<(6+11-i)) ){
80                     s->pblocks[i] = (short *)(&s->block[(j++)]);
81                 }else{
82                     s->pblocks[i] = NULL;
83                 }
84             }
85         }
86     }
87 }
88
89 static int calc_cbp(MpegEncContext *s){
90 /* compute cbp */
91 int  i,cbp = 0;
92     for(i=0; i<4; i++) {
93         if(s->block_last_index[i] >= 0)
94             cbp |= 1 << (5 - i);
95     }
96     if(s->flags & CODEC_FLAG_GRAY)
97          return cbp; //4 block for grayscale one done
98
99
100     for(i=4; i<6; i++) {
101         if(s->block_last_index[i] >= 0)
102             cbp |= 1 << (5 - i);
103     }
104     if(s->chroma_format <  2) return cbp;
105
106
107     if(s->chroma_format == 2){/*CHROMA_422*/
108         for(i=6; i<8; i++) {
109             if(s->block_last_index[i] >= 0)
110                 cbp |= 1 << (6+7 - i);
111         }
112     }else{/*CHROMA_444*/
113         for(i=6; i<12; i++) {
114             if(s->block_last_index[i] >= 0)
115                 cbp |= 1 << (6+11 - i);
116         }
117     }
118     return cbp;
119 }
120
121
122
123 //these functions should be called on every new field or/and frame
124 //They should be safe if they are called few times for same field!
125 int XVMC_field_start(MpegEncContext*s, AVCodecContext *avctx){
126 xvmc_render_state_t * render,* last, * next;
127
128     assert(avctx != NULL);
129
130     render = (xvmc_render_state_t*)s->current_picture.data[2];
131     assert(render != NULL);
132     if( (render == NULL) || (render->magic != MP_XVMC_RENDER_MAGIC) )
133         return -1;//make sure that this is render packet
134
135     render->picture_structure = s->picture_structure;
136     render->flags = (s->first_field)? 0: XVMC_SECOND_FIELD;
137
138 //make sure that all data is drawn by XVMC_end_frame
139     assert(render->filled_mv_blocks_num==0);
140
141     render->p_future_surface = NULL;
142     render->p_past_surface = NULL;
143
144     switch(s->pict_type){
145         case  I_TYPE:
146             return 0;// no prediction from other frames
147         case  B_TYPE:
148             next = (xvmc_render_state_t*)s->next_picture.data[2];
149             assert(next!=NULL);
150             assert(next->state & MP_XVMC_STATE_PREDICTION);
151             if(next == NULL) return -1;
152             if(next->magic != MP_XVMC_RENDER_MAGIC) return -1;
153             render->p_future_surface = next->p_surface;
154             //no return here, going to set forward prediction
155         case  P_TYPE:
156             last = (xvmc_render_state_t*)s->last_picture.data[2];
157             if(last == NULL)// && !s->first_field)
158                 last = render;//predict second field from the first
159             if(last->magic != MP_XVMC_RENDER_MAGIC) return -1;
160             assert(last->state & MP_XVMC_STATE_PREDICTION);
161             render->p_past_surface = last->p_surface;
162             return 0;
163     }
164
165 return -1;
166 }
167
168 void XVMC_field_end(MpegEncContext *s){
169 xvmc_render_state_t * render;
170     render = (xvmc_render_state_t*)s->current_picture.data[2];
171     assert(render != NULL);
172
173     if(render->filled_mv_blocks_num > 0){
174 //        printf("xvmcvideo.c: rendering %d left blocks after last slice!!!\n",render->filled_mv_blocks_num );
175         ff_draw_horiz_band(s,0,0);
176     }
177 }
178
179 void XVMC_decode_mb(MpegEncContext *s){
180 XvMCMacroBlock * mv_block;
181 xvmc_render_state_t * render;
182 int i,cbp,blocks_per_mb;
183
184 const int mb_xy = s->mb_y * s->mb_stride + s->mb_x;
185
186
187     if(s->encoding){
188         av_log(s->avctx, AV_LOG_ERROR, "XVMC doesn't support encoding!!!\n");
189         av_abort();
190     }
191
192    //from MPV_decode_mb(),
193     /* update DC predictors for P macroblocks */
194     if (!s->mb_intra) {
195         s->last_dc[0] =
196         s->last_dc[1] =
197         s->last_dc[2] =  128 << s->intra_dc_precision;
198     }
199
200    //MC doesn't skip blocks
201     s->mb_skiped = 0;
202
203
204    // do I need to export quant when I could not perform postprocessing?
205    // anyway, it doesn't hurrt
206     s->current_picture.qscale_table[mb_xy] = s->qscale;
207
208 //START OF XVMC specific code
209     render = (xvmc_render_state_t*)s->current_picture.data[2];
210     assert(render!=NULL);
211     assert(render->magic==MP_XVMC_RENDER_MAGIC);
212     assert(render->mv_blocks);
213
214     //take the next free macroblock
215     mv_block = &render->mv_blocks[render->start_mv_blocks_num + 
216                                    render->filled_mv_blocks_num ];
217
218 // memset(mv_block,0,sizeof(XvMCMacroBlock));
219
220     mv_block->x = s->mb_x;
221     mv_block->y = s->mb_y;
222     mv_block->dct_type = s->interlaced_dct;//XVMC_DCT_TYPE_FRAME/FIELD;
223 //    mv_block->motion_type = 0;  //zero to silense warnings
224     if(s->mb_intra){
225         mv_block->macroblock_type = XVMC_MB_TYPE_INTRA;//no MC, all done
226     }else{
227         mv_block->macroblock_type = XVMC_MB_TYPE_PATTERN;
228
229         if(s->mv_dir & MV_DIR_FORWARD){
230             mv_block->macroblock_type|= XVMC_MB_TYPE_MOTION_FORWARD;
231             //pmv[n][dir][xy]=mv[dir][n][xy]
232             mv_block->PMV[0][0][0] = s->mv[0][0][0];
233             mv_block->PMV[0][0][1] = s->mv[0][0][1];
234             mv_block->PMV[1][0][0] = s->mv[0][1][0];
235             mv_block->PMV[1][0][1] = s->mv[0][1][1];
236         }
237         if(s->mv_dir & MV_DIR_BACKWARD){
238             mv_block->macroblock_type|=XVMC_MB_TYPE_MOTION_BACKWARD;
239             mv_block->PMV[0][1][0] = s->mv[1][0][0];
240             mv_block->PMV[0][1][1] = s->mv[1][0][1];
241             mv_block->PMV[1][1][0] = s->mv[1][1][0];
242             mv_block->PMV[1][1][1] = s->mv[1][1][1];
243         }
244
245         switch(s->mv_type){
246             case  MV_TYPE_16X16:
247                 mv_block->motion_type = XVMC_PREDICTION_FRAME;
248                 break;
249             case  MV_TYPE_16X8:
250                 mv_block->motion_type = XVMC_PREDICTION_16x8;
251                 break;
252             case  MV_TYPE_FIELD:
253                 mv_block->motion_type = XVMC_PREDICTION_FIELD;
254                 if(s->picture_structure == PICT_FRAME){
255                     mv_block->PMV[0][0][1]<<=1;
256                     mv_block->PMV[1][0][1]<<=1;
257                     mv_block->PMV[0][1][1]<<=1;
258                     mv_block->PMV[1][1][1]<<=1;
259                 }
260                 break;
261             case  MV_TYPE_DMV:
262                 mv_block->motion_type = XVMC_PREDICTION_DUAL_PRIME;
263                 if(s->picture_structure == PICT_FRAME){
264
265                     mv_block->PMV[0][0][0] = s->mv[0][0][0];//top from top
266                     mv_block->PMV[0][0][1] = s->mv[0][0][1]<<1;
267
268                     mv_block->PMV[0][1][0] = s->mv[0][0][0];//bottom from bottom
269                     mv_block->PMV[0][1][1] = s->mv[0][0][1]<<1;
270
271                     mv_block->PMV[1][0][0] = s->mv[0][2][0];//dmv00, top from bottom
272                     mv_block->PMV[1][0][1] = s->mv[0][2][1]<<1;//dmv01
273
274                     mv_block->PMV[1][1][0] = s->mv[0][3][0];//dmv10, bottom from top
275                     mv_block->PMV[1][1][1] = s->mv[0][3][1]<<1;//dmv11
276
277                 }else{
278                     mv_block->PMV[0][1][0] = s->mv[0][2][0];//dmv00
279                     mv_block->PMV[0][1][1] = s->mv[0][2][1];//dmv01
280                 }
281                 break;
282             default:
283                 assert(0);
284         }
285
286         mv_block->motion_vertical_field_select = 0;
287
288 //set correct field referenses
289         if(s->mv_type == MV_TYPE_FIELD || s->mv_type == MV_TYPE_16X8){
290             if( s->field_select[0][0] ) mv_block->motion_vertical_field_select|=1;
291             if( s->field_select[1][0] ) mv_block->motion_vertical_field_select|=2;
292             if( s->field_select[0][1] ) mv_block->motion_vertical_field_select|=4;
293             if( s->field_select[1][1] ) mv_block->motion_vertical_field_select|=8;
294         }
295     }//!intra
296 //time to handle data blocks;
297     mv_block->index = render->next_free_data_block_num;
298
299     blocks_per_mb = 6;
300     if( s->chroma_format >= 2){
301         block_per_mb = 4 + (1 << (s->chroma_format));
302     }
303
304     if(s->flags & CODEC_FLAG_GRAY){
305         if(s->mb_intra){//intra frames are alwasy full chroma block
306             for(i=4; i<blocks_per_mb; i++){
307                 memset(s->pblocks[i],0,sizeof(short)*8*8);//so we need to clear them
308                 if(!render->unsigned_intra)
309                     s->pblocks[i][0] = 1<<10;
310             }
311         }else
312             blocks_per_mb = 4;//Luminance blocks only
313     }
314     cbp = calc_cbp(s);
315     mv_block->coded_block_pattern = cbp;
316     if(cbp == 0)
317         mv_block->macroblock_type &= ~XVMC_MB_TYPE_PATTERN;
318
319     for(i=0; i<blocks_per_mb; i++){
320         if(s->block_last_index[i] >= 0){
321             // i do not have unsigned_intra MOCO to test, hope it is OK
322             if( (s->mb_intra) && ( render->idct || (!render->idct && !render->unsigned_intra)) )
323                 s->pblocks[i][0]-=1<<10;
324             if(!render->idct){
325                 s->dsp.idct(s->pblocks[i]);
326                 //!!TODO!clip!!!
327             }
328 //copy blocks only if the codec doesn't support pblocks reordering
329             if(s->avctx->xvmc_acceleration == 1){
330                 memcpy(&render->data_blocks[(render->next_free_data_block_num)*64],
331                         s->pblocks[i],sizeof(short)*8*8);
332             }else{
333 /*              if(s->pblocks[i] != &render->data_blocks[
334                         (render->next_free_data_block_num)*64]){
335                    printf("ERROR mb(%d,%d) s->pblocks[i]=%p data_block[]=%p\n",
336                    s->mb_x,s->mb_y, s->pblocks[i], 
337                    &render->data_blocks[(render->next_free_data_block_num)*64]);
338                 }*/
339             }
340             render->next_free_data_block_num++;
341         }
342     }
343     render->filled_mv_blocks_num++;
344
345     assert(render->filled_mv_blocks_num <= render->total_number_of_mv_blocks);
346     assert(render->next_free_data_block_num <= render->total_number_of_data_blocks);
347
348
349     if(render->filled_mv_blocks_num >= render->total_number_of_mv_blocks)
350         ff_draw_horiz_band(s,0,0);
351
352 // DumpRenderInfo(render);
353 // DumpMBlockInfo(mv_block);
354
355 }
356
357 #endif