bbc87a52737cc54633674938a71d7e7cc8d6c8ac
[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
33 //set s->block
34 void ff_xvmc_init_block(MpegEncContext *s)
35 {
36 struct xvmc_render_state *render;
37 render = (struct xvmc_render_state*)s->current_picture.data[2];
38 assert(render);
39 if (!render || render->magic != AV_XVMC_RENDER_MAGIC) {
40 assert(0);
41 return; // make sure that this is a render packet
42 }
43 s->block = (DCTELEM *)(render->data_blocks + render->next_free_data_block_num * 64);
44 }
45
46 void ff_xvmc_pack_pblocks(MpegEncContext *s, int cbp)
47 {
48 int i, j;
49 const int mb_block_count = 4 + (1 << s->chroma_format);
50
51 j = 0;
52 cbp <<= 12-mb_block_count;
53 for (i = 0; i < mb_block_count; i++) {
54 if (cbp & (1 << 11))
55 s->pblocks[i] = (short *)(&s->block[j++]);
56 else
57 s->pblocks[i] = NULL;
58 cbp+=cbp;
59 }
60 }
61
62 // These functions should be called on every new field and/or frame.
63 // They should be safe if they are called a few times for the same field!
64 int ff_xvmc_field_start(MpegEncContext*s, AVCodecContext *avctx)
65 {
66 struct xvmc_render_state *render, *last, *next;
67
68 assert(avctx);
69
70 render = (struct xvmc_render_state*)s->current_picture.data[2];
71 assert(render);
72 if (!render || render->magic != AV_XVMC_RENDER_MAGIC)
73 return -1; // make sure that this is a render packet
74
75 render->picture_structure = s->picture_structure;
76 render->flags = s->first_field ? 0 : XVMC_SECOND_FIELD;
77
78 assert(render->filled_mv_blocks_num == 0);
79
80 render->p_future_surface = NULL;
81 render->p_past_surface = NULL;
82
83 switch(s->pict_type) {
84 case FF_I_TYPE:
85 return 0; // no prediction from other frames
86 case FF_B_TYPE:
87 next = (struct xvmc_render_state*)s->next_picture.data[2];
88 assert(next);
89 if (!next)
90 return -1;
91 if (next->magic != AV_XVMC_RENDER_MAGIC)
92 return -1;
93 render->p_future_surface = next->p_surface;
94 // no return here, going to set forward prediction
95 case FF_P_TYPE:
96 last = (struct xvmc_render_state*)s->last_picture.data[2];
97 if (!last)
98 last = render; // predict second field from the first
99 if (last->magic != AV_XVMC_RENDER_MAGIC)
100 return -1;
101 render->p_past_surface = last->p_surface;
102 return 0;
103 }
104
105 return -1;
106 }
107
108 void ff_xvmc_field_end(MpegEncContext *s)
109 {
110 struct xvmc_render_state *render;
111 render = (struct xvmc_render_state*)s->current_picture.data[2];
112 assert(render);
113
114 if (render->filled_mv_blocks_num > 0)
115 ff_draw_horiz_band(s,0,0);
116 }
117
118 void ff_xvmc_decode_mb(MpegEncContext *s)
119 {
120 XvMCMacroBlock *mv_block;
121 struct xvmc_render_state *render;
122 int i, cbp, blocks_per_mb;
123
124 const int mb_xy = s->mb_y * s->mb_stride + s->mb_x;
125
126
127 if (s->encoding) {
128 av_log(s->avctx, AV_LOG_ERROR, "XVMC doesn't support encoding!!!\n");
129 return;
130 }
131
132 // from MPV_decode_mb(), update DC predictors for P macroblocks
133 if (!s->mb_intra) {
134 s->last_dc[0] =
135 s->last_dc[1] =
136 s->last_dc[2] = 128 << s->intra_dc_precision;
137 }
138
139 // MC doesn't skip blocks
140 s->mb_skipped = 0;
141
142
143 // Do I need to export quant when I could not perform postprocessing?
144 // Anyway, it doesn't hurt.
145 s->current_picture.qscale_table[mb_xy] = s->qscale;
146
147 // start of XVMC-specific code
148 render = (struct xvmc_render_state*)s->current_picture.data[2];
149 assert(render);
150 assert(render->magic == AV_XVMC_RENDER_MAGIC);
151 assert(render->mv_blocks);
152
153 // take the next free macroblock
154 mv_block = &render->mv_blocks[render->start_mv_blocks_num +
155 render->filled_mv_blocks_num ];
156
157 mv_block->x = s->mb_x;
158 mv_block->y = s->mb_y;
159 mv_block->dct_type = s->interlaced_dct; // XVMC_DCT_TYPE_FRAME/FIELD;
160 if (s->mb_intra) {
161 mv_block->macroblock_type = XVMC_MB_TYPE_INTRA; // no MC, all done
162 } else {
163 mv_block->macroblock_type = XVMC_MB_TYPE_PATTERN;
164
165 if (s->mv_dir & MV_DIR_FORWARD) {
166 mv_block->macroblock_type |= XVMC_MB_TYPE_MOTION_FORWARD;
167 // PMV[n][dir][xy] = mv[dir][n][xy]
168 mv_block->PMV[0][0][0] = s->mv[0][0][0];
169 mv_block->PMV[0][0][1] = s->mv[0][0][1];
170 mv_block->PMV[1][0][0] = s->mv[0][1][0];
171 mv_block->PMV[1][0][1] = s->mv[0][1][1];
172 }
173 if (s->mv_dir & MV_DIR_BACKWARD) {
174 mv_block->macroblock_type |= XVMC_MB_TYPE_MOTION_BACKWARD;
175 mv_block->PMV[0][1][0] = s->mv[1][0][0];
176 mv_block->PMV[0][1][1] = s->mv[1][0][1];
177 mv_block->PMV[1][1][0] = s->mv[1][1][0];
178 mv_block->PMV[1][1][1] = s->mv[1][1][1];
179 }
180
181 switch(s->mv_type){
182 case MV_TYPE_16X16:
183 mv_block->motion_type = XVMC_PREDICTION_FRAME;
184 break;
185 case MV_TYPE_16X8:
186 mv_block->motion_type = XVMC_PREDICTION_16x8;
187 break;
188 case MV_TYPE_FIELD:
189 mv_block->motion_type = XVMC_PREDICTION_FIELD;
190 if (s->picture_structure == PICT_FRAME) {
191 mv_block->PMV[0][0][1] <<= 1;
192 mv_block->PMV[1][0][1] <<= 1;
193 mv_block->PMV[0][1][1] <<= 1;
194 mv_block->PMV[1][1][1] <<= 1;
195 }
196 break;
197 case MV_TYPE_DMV:
198 mv_block->motion_type = XVMC_PREDICTION_DUAL_PRIME;
199 if (s->picture_structure == PICT_FRAME) {
200
201 mv_block->PMV[0][0][0] = s->mv[0][0][0]; // top from top
202 mv_block->PMV[0][0][1] = s->mv[0][0][1]<<1;
203
204 mv_block->PMV[0][1][0] = s->mv[0][0][0]; // bottom from bottom
205 mv_block->PMV[0][1][1] = s->mv[0][0][1]<<1;
206
207 mv_block->PMV[1][0][0] = s->mv[0][2][0]; // dmv00, top from bottom
208 mv_block->PMV[1][0][1] = s->mv[0][2][1]<<1; // dmv01
209
210 mv_block->PMV[1][1][0] = s->mv[0][3][0]; // dmv10, bottom from top
211 mv_block->PMV[1][1][1] = s->mv[0][3][1]<<1; // dmv11
212
213 } else {
214 mv_block->PMV[0][1][0] = s->mv[0][2][0]; // dmv00
215 mv_block->PMV[0][1][1] = s->mv[0][2][1]; // dmv01
216 }
217 break;
218 default:
219 assert(0);
220 }
221
222 mv_block->motion_vertical_field_select = 0;
223
224 // set correct field references
225 if (s->mv_type == MV_TYPE_FIELD || s->mv_type == MV_TYPE_16X8) {
226 mv_block->motion_vertical_field_select |= s->field_select[0][0];
227 mv_block->motion_vertical_field_select |= s->field_select[1][0] << 1;
228 mv_block->motion_vertical_field_select |= s->field_select[0][1] << 2;
229 mv_block->motion_vertical_field_select |= s->field_select[1][1] << 3;
230 }
231 } // !intra
232 // time to handle data blocks
233 mv_block->index = render->next_free_data_block_num;
234
235 blocks_per_mb = 6;
236 if (s->chroma_format >= 2) {
237 blocks_per_mb = 4 + (1 << s->chroma_format);
238 }
239
240 // calculate cbp
241 cbp = 0;
242 for (i = 0; i < blocks_per_mb; i++) {
243 cbp += cbp;
244 if (s->block_last_index[i] >= 0)
245 cbp++;
246 }
247
248 if (s->flags & CODEC_FLAG_GRAY) {
249 if (s->mb_intra) { // intra frames are always full chroma blocks
250 for (i = 4; i < blocks_per_mb; i++) {
251 memset(s->pblocks[i], 0, sizeof(short)*8*8); // so we need to clear them
252 if (!render->unsigned_intra)
253 s->pblocks[i][0] = 1 << 10;
254 }
255 } else {
256 cbp &= 0xf << (blocks_per_mb - 4);
257 blocks_per_mb = 4; // luminance blocks only
258 }
259 }
260 mv_block->coded_block_pattern = cbp;
261 if (cbp == 0)
262 mv_block->macroblock_type &= ~XVMC_MB_TYPE_PATTERN;
263
264 for (i = 0; i < blocks_per_mb; i++) {
265 if (s->block_last_index[i] >= 0) {
266 // I do not have unsigned_intra MOCO to test, hope it is OK.
267 if (s->mb_intra && (render->idct || (!render->idct && !render->unsigned_intra)))
268 s->pblocks[i][0] -= 1 << 10;
269 if (!render->idct) {
270 s->dsp.idct(s->pblocks[i]);
271 /* It is unclear if MC hardware requires pixel diff values to be
272 * in the range [-255;255]. TODO: Clipping if such hardware is
273 * ever found. As of now it would only be an unnecessary
274 * slowdown. */
275 }
276 // copy blocks only if the codec doesn't support pblocks reordering
277 if (s->avctx->xvmc_acceleration == 1) {
278 memcpy(&render->data_blocks[render->next_free_data_block_num*64],
279 s->pblocks[i],sizeof(short)*8*8);
280 }
281 render->next_free_data_block_num++;
282 }
283 }
284 render->filled_mv_blocks_num++;
285
286 assert(render->filled_mv_blocks_num <= render->total_number_of_mv_blocks);
287 assert(render->next_free_data_block_num <= render->total_number_of_data_blocks);
288
289
290 if (render->filled_mv_blocks_num >= render->total_number_of_mv_blocks)
291 ff_draw_horiz_band(s, 0, 0);
292 }