ce575938334816ebe9fec25b8aed3fd75fa2f4a4
[libav.git] / libavcodec / mpegvideo_xvmc.c
1 /*
2 * XVideo Motion Compensation
3 * Copyright (c) 2003 Ivan Kalvachev
4 *
5 * This file is part of FFmpeg.
6 *
7 * FFmpeg is free software; you can redistribute it and/or
8 * modify it under the terms of the GNU Lesser General Public
9 * License as published by the Free Software Foundation; either
10 * version 2.1 of the License, or (at your option) any later version.
11 *
12 * FFmpeg is distributed in the hope that it will be useful,
13 * but WITHOUT ANY WARRANTY; without even the implied warranty of
14 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
15 * Lesser General Public License for more details.
16 *
17 * You should have received a copy of the GNU Lesser General Public
18 * License along with FFmpeg; if not, write to the Free Software
19 * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
20 */
21
22 #include <limits.h>
23
24 #include "avcodec.h"
25 #include "dsputil.h"
26 #include "mpegvideo.h"
27
28 #undef NDEBUG
29 #include <assert.h>
30
31 #include "xvmc.h"
32 #include "xvmc_internal.h"
33
34 //set s->block
35 void ff_xvmc_init_block(MpegEncContext *s)
36 {
37 struct xvmc_render_state *render;
38 render = (struct xvmc_render_state*)s->current_picture.data[2];
39 assert(render);
40 if (!render || render->magic != AV_XVMC_RENDER_MAGIC) {
41 assert(0);
42 return; // make sure that this is a render packet
43 }
44 s->block = (DCTELEM *)(render->data_blocks + render->next_free_data_block_num * 64);
45 }
46
47 void ff_xvmc_pack_pblocks(MpegEncContext *s, int cbp)
48 {
49 int i, j;
50 const int mb_block_count = 4 + (1 << s->chroma_format);
51
52 j = 0;
53 cbp <<= 12-mb_block_count;
54 for (i = 0; i < mb_block_count; i++) {
55 if (cbp & (1 << 11))
56 s->pblocks[i] = (short *)(&s->block[j++]);
57 else
58 s->pblocks[i] = NULL;
59 cbp+=cbp;
60 }
61 }
62
63 // These functions should be called on every new field and/or frame.
64 // They should be safe if they are called a few times for the same field!
65 int ff_xvmc_field_start(MpegEncContext*s, AVCodecContext *avctx)
66 {
67 struct xvmc_render_state *render, *last, *next;
68
69 assert(avctx);
70
71 render = (struct xvmc_render_state*)s->current_picture.data[2];
72 assert(render);
73 if (!render || render->magic != AV_XVMC_RENDER_MAGIC)
74 return -1; // make sure that this is a render packet
75
76 render->picture_structure = s->picture_structure;
77 render->flags = s->first_field ? 0 : XVMC_SECOND_FIELD;
78
79 if(render->filled_mv_blocks_num){
80 av_log(avctx, AV_LOG_ERROR,
81 "Rendering surface contains %i unprocessed blocks\n",
82 render->filled_mv_blocks_num);
83 assert(0);
84 }
85
86 render->p_future_surface = NULL;
87 render->p_past_surface = NULL;
88
89 switch(s->pict_type) {
90 case FF_I_TYPE:
91 return 0; // no prediction from other frames
92 case FF_B_TYPE:
93 next = (struct xvmc_render_state*)s->next_picture.data[2];
94 assert(next);
95 if (!next)
96 return -1;
97 if (next->magic != AV_XVMC_RENDER_MAGIC)
98 return -1;
99 render->p_future_surface = next->p_surface;
100 // no return here, going to set forward prediction
101 case FF_P_TYPE:
102 last = (struct xvmc_render_state*)s->last_picture.data[2];
103 if (!last)
104 last = render; // predict second field from the first
105 if (last->magic != AV_XVMC_RENDER_MAGIC)
106 return -1;
107 render->p_past_surface = last->p_surface;
108 return 0;
109 }
110
111 return -1;
112 }
113
114 void ff_xvmc_field_end(MpegEncContext *s)
115 {
116 struct xvmc_render_state *render;
117 render = (struct xvmc_render_state*)s->current_picture.data[2];
118 assert(render);
119
120 if (render->filled_mv_blocks_num > 0)
121 ff_draw_horiz_band(s, 0, 0);
122 }
123
124 void ff_xvmc_decode_mb(MpegEncContext *s)
125 {
126 XvMCMacroBlock *mv_block;
127 struct xvmc_render_state *render;
128 int i, cbp, blocks_per_mb;
129
130 const int mb_xy = s->mb_y * s->mb_stride + s->mb_x;
131
132
133 if (s->encoding) {
134 av_log(s->avctx, AV_LOG_ERROR, "XVMC doesn't support encoding!!!\n");
135 return;
136 }
137
138 // from MPV_decode_mb(), update DC predictors for P macroblocks
139 if (!s->mb_intra) {
140 s->last_dc[0] =
141 s->last_dc[1] =
142 s->last_dc[2] = 128 << s->intra_dc_precision;
143 }
144
145 // MC doesn't skip blocks
146 s->mb_skipped = 0;
147
148
149 // Do I need to export quant when I could not perform postprocessing?
150 // Anyway, it doesn't hurt.
151 s->current_picture.qscale_table[mb_xy] = s->qscale;
152
153 // start of XVMC-specific code
154 render = (struct xvmc_render_state*)s->current_picture.data[2];
155 assert(render);
156 assert(render->magic == AV_XVMC_RENDER_MAGIC);
157 assert(render->mv_blocks);
158
159 // take the next free macroblock
160 mv_block = &render->mv_blocks[render->start_mv_blocks_num +
161 render->filled_mv_blocks_num ];
162
163 mv_block->x = s->mb_x;
164 mv_block->y = s->mb_y;
165 mv_block->dct_type = s->interlaced_dct; // XVMC_DCT_TYPE_FRAME/FIELD;
166 if (s->mb_intra) {
167 mv_block->macroblock_type = XVMC_MB_TYPE_INTRA; // no MC, all done
168 } else {
169 mv_block->macroblock_type = XVMC_MB_TYPE_PATTERN;
170
171 if (s->mv_dir & MV_DIR_FORWARD) {
172 mv_block->macroblock_type |= XVMC_MB_TYPE_MOTION_FORWARD;
173 // PMV[n][dir][xy] = mv[dir][n][xy]
174 mv_block->PMV[0][0][0] = s->mv[0][0][0];
175 mv_block->PMV[0][0][1] = s->mv[0][0][1];
176 mv_block->PMV[1][0][0] = s->mv[0][1][0];
177 mv_block->PMV[1][0][1] = s->mv[0][1][1];
178 }
179 if (s->mv_dir & MV_DIR_BACKWARD) {
180 mv_block->macroblock_type |= XVMC_MB_TYPE_MOTION_BACKWARD;
181 mv_block->PMV[0][1][0] = s->mv[1][0][0];
182 mv_block->PMV[0][1][1] = s->mv[1][0][1];
183 mv_block->PMV[1][1][0] = s->mv[1][1][0];
184 mv_block->PMV[1][1][1] = s->mv[1][1][1];
185 }
186
187 switch(s->mv_type) {
188 case MV_TYPE_16X16:
189 mv_block->motion_type = XVMC_PREDICTION_FRAME;
190 break;
191 case MV_TYPE_16X8:
192 mv_block->motion_type = XVMC_PREDICTION_16x8;
193 break;
194 case MV_TYPE_FIELD:
195 mv_block->motion_type = XVMC_PREDICTION_FIELD;
196 if (s->picture_structure == PICT_FRAME) {
197 mv_block->PMV[0][0][1] <<= 1;
198 mv_block->PMV[1][0][1] <<= 1;
199 mv_block->PMV[0][1][1] <<= 1;
200 mv_block->PMV[1][1][1] <<= 1;
201 }
202 break;
203 case MV_TYPE_DMV:
204 mv_block->motion_type = XVMC_PREDICTION_DUAL_PRIME;
205 if (s->picture_structure == PICT_FRAME) {
206
207 mv_block->PMV[0][0][0] = s->mv[0][0][0]; // top from top
208 mv_block->PMV[0][0][1] = s->mv[0][0][1] << 1;
209
210 mv_block->PMV[0][1][0] = s->mv[0][0][0]; // bottom from bottom
211 mv_block->PMV[0][1][1] = s->mv[0][0][1] << 1;
212
213 mv_block->PMV[1][0][0] = s->mv[0][2][0]; // dmv00, top from bottom
214 mv_block->PMV[1][0][1] = s->mv[0][2][1] << 1; // dmv01
215
216 mv_block->PMV[1][1][0] = s->mv[0][3][0]; // dmv10, bottom from top
217 mv_block->PMV[1][1][1] = s->mv[0][3][1] << 1; // dmv11
218
219 } else {
220 mv_block->PMV[0][1][0] = s->mv[0][2][0]; // dmv00
221 mv_block->PMV[0][1][1] = s->mv[0][2][1]; // dmv01
222 }
223 break;
224 default:
225 assert(0);
226 }
227
228 mv_block->motion_vertical_field_select = 0;
229
230 // set correct field references
231 if (s->mv_type == MV_TYPE_FIELD || s->mv_type == MV_TYPE_16X8) {
232 mv_block->motion_vertical_field_select |= s->field_select[0][0];
233 mv_block->motion_vertical_field_select |= s->field_select[1][0] << 1;
234 mv_block->motion_vertical_field_select |= s->field_select[0][1] << 2;
235 mv_block->motion_vertical_field_select |= s->field_select[1][1] << 3;
236 }
237 } // !intra
238 // time to handle data blocks
239 mv_block->index = render->next_free_data_block_num;
240
241 blocks_per_mb = 6;
242 if (s->chroma_format >= 2) {
243 blocks_per_mb = 4 + (1 << s->chroma_format);
244 }
245
246 // calculate cbp
247 cbp = 0;
248 for (i = 0; i < blocks_per_mb; i++) {
249 cbp += cbp;
250 if (s->block_last_index[i] >= 0)
251 cbp++;
252 }
253
254 if (s->flags & CODEC_FLAG_GRAY) {
255 if (s->mb_intra) { // intra frames are always full chroma blocks
256 for (i = 4; i < blocks_per_mb; i++) {
257 memset(s->pblocks[i], 0, sizeof(short)*8*8); // so we need to clear them
258 if (!render->unsigned_intra)
259 s->pblocks[i][0] = 1 << 10;
260 }
261 } else {
262 cbp &= 0xf << (blocks_per_mb - 4);
263 blocks_per_mb = 4; // luminance blocks only
264 }
265 }
266 mv_block->coded_block_pattern = cbp;
267 if (cbp == 0)
268 mv_block->macroblock_type &= ~XVMC_MB_TYPE_PATTERN;
269
270 for (i = 0; i < blocks_per_mb; i++) {
271 if (s->block_last_index[i] >= 0) {
272 // I do not have unsigned_intra MOCO to test, hope it is OK.
273 if (s->mb_intra && (render->idct || (!render->idct && !render->unsigned_intra)))
274 s->pblocks[i][0] -= 1 << 10;
275 if (!render->idct) {
276 s->dsp.idct(s->pblocks[i]);
277 /* It is unclear if MC hardware requires pixel diff values to be
278 * in the range [-255;255]. TODO: Clipping if such hardware is
279 * ever found. As of now it would only be an unnecessary
280 * slowdown. */
281 }
282 // copy blocks only if the codec doesn't support pblocks reordering
283 if (s->avctx->xvmc_acceleration == 1) {
284 memcpy(&render->data_blocks[render->next_free_data_block_num*64],
285 s->pblocks[i], sizeof(short)*64);
286 }
287 render->next_free_data_block_num++;
288 }
289 }
290 render->filled_mv_blocks_num++;
291
292 assert(render->filled_mv_blocks_num <= render->total_number_of_mv_blocks);
293 assert(render->next_free_data_block_num <= render->total_number_of_data_blocks);
294 /*The above conditions should not be able to fail as long as this function is used
295 and following if() automatically call callback to free blocks. */
296
297
298 if (render->filled_mv_blocks_num >= render->total_number_of_mv_blocks)
299 ff_draw_horiz_band(s, 0, 0);
300 }