301b9ad6e8e43aec4c34dec70bff123eaace96fa
[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 /**
35 * Initializes the block field of the MpegEncContext pointer passed as
36 * parameter after making sure that the data is not corrupted.
37 */
38 void ff_xvmc_init_block(MpegEncContext *s)
39 {
40 struct xvmc_pixfmt_render *render = (struct xvmc_pixfmt_render*)s->current_picture.data[2];
41 assert(render && render->magic_id == AV_XVMC_RENDER_MAGIC);
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 = 0;
49 const int mb_block_count = 4 + (1 << s->chroma_format);
50
51 cbp <<= 12-mb_block_count;
52 for (i = 0; i < mb_block_count; i++) {
53 if (cbp & (1 << 11))
54 s->pblocks[i] = (short *)(&s->block[j++]);
55 else
56 s->pblocks[i] = NULL;
57 cbp+=cbp;
58 }
59 }
60
61 /**
62 * This function should be called for every new field and/or frame.
63 * It should be safe to call the function a few times for the same field.
64 */
65 int ff_xvmc_field_start(MpegEncContext*s, AVCodecContext *avctx)
66 {
67 struct xvmc_pixfmt_render *last, *next, *render = (struct xvmc_pixfmt_render*)s->current_picture.data[2];
68 const int mb_block_count = 4 + (1 << s->chroma_format);
69
70 assert(avctx);
71 if (!render || render->magic_id != AV_XVMC_RENDER_MAGIC ||
72 !render->data_blocks || !render->mv_blocks){
73 av_log(avctx, AV_LOG_ERROR,
74 "Render token doesn't look as expected.\n");
75 return -1; // make sure that this is a render packet
76 }
77
78 if (render->filled_mv_blocks_num) {
79 av_log(avctx, AV_LOG_ERROR,
80 "Rendering surface contains %i unprocessed blocks.\n",
81 render->filled_mv_blocks_num);
82 return -1;
83 }
84 if (render->total_number_of_mv_blocks < 1 ||
85 render->total_number_of_data_blocks < mb_block_count) {
86 av_log(avctx, AV_LOG_ERROR,
87 "Rendering surface doesn't provide enough block structures to work with.\n");
88 return -1;
89 }
90
91 render->picture_structure = s->picture_structure;
92 render->flags = s->first_field ? 0 : XVMC_SECOND_FIELD;
93 render->p_future_surface = NULL;
94 render->p_past_surface = NULL;
95
96 switch(s->pict_type) {
97 case FF_I_TYPE:
98 return 0; // no prediction from other frames
99 case FF_B_TYPE:
100 next = (struct xvmc_pixfmt_render*)s->next_picture.data[2];
101 if (!next)
102 return -1;
103 if (next->magic_id != AV_XVMC_RENDER_MAGIC)
104 return -1;
105 render->p_future_surface = next->p_surface;
106 // no return here, going to set forward prediction
107 case FF_P_TYPE:
108 last = (struct xvmc_pixfmt_render*)s->last_picture.data[2];
109 if (!last)
110 last = render; // predict second field from the first
111 if (last->magic_id != AV_XVMC_RENDER_MAGIC)
112 return -1;
113 render->p_past_surface = last->p_surface;
114 return 0;
115 }
116
117 return -1;
118 }
119
120 /**
121 * This function should be called for every new field and/or frame.
122 * It should be safe to call the function a few times for the same field.
123 */
124 void ff_xvmc_field_end(MpegEncContext *s)
125 {
126 struct xvmc_pixfmt_render *render = (struct xvmc_pixfmt_render*)s->current_picture.data[2];
127 assert(render);
128
129 if (render->filled_mv_blocks_num > 0)
130 ff_draw_horiz_band(s, 0, 0);
131 }
132
133 void ff_xvmc_decode_mb(MpegEncContext *s)
134 {
135 XvMCMacroBlock *mv_block;
136 struct xvmc_pixfmt_render *render;
137 int i, cbp, blocks_per_mb;
138
139 const int mb_xy = s->mb_y * s->mb_stride + s->mb_x;
140
141
142 if (s->encoding) {
143 av_log(s->avctx, AV_LOG_ERROR, "XVMC doesn't support encoding!!!\n");
144 return;
145 }
146
147 // from MPV_decode_mb(), update DC predictors for P macroblocks
148 if (!s->mb_intra) {
149 s->last_dc[0] =
150 s->last_dc[1] =
151 s->last_dc[2] = 128 << s->intra_dc_precision;
152 }
153
154 // MC doesn't skip blocks
155 s->mb_skipped = 0;
156
157
158 // Do I need to export quant when I could not perform postprocessing?
159 // Anyway, it doesn't hurt.
160 s->current_picture.qscale_table[mb_xy] = s->qscale;
161
162 // start of XVMC-specific code
163 render = (struct xvmc_pixfmt_render*)s->current_picture.data[2];
164 assert(render);
165 assert(render->magic_id == AV_XVMC_RENDER_MAGIC);
166 assert(render->mv_blocks);
167
168 // take the next free macroblock
169 mv_block = &render->mv_blocks[render->start_mv_blocks_num +
170 render->filled_mv_blocks_num];
171
172 mv_block->x = s->mb_x;
173 mv_block->y = s->mb_y;
174 mv_block->dct_type = s->interlaced_dct; // XVMC_DCT_TYPE_FRAME/FIELD;
175 if (s->mb_intra) {
176 mv_block->macroblock_type = XVMC_MB_TYPE_INTRA; // no MC, all done
177 } else {
178 mv_block->macroblock_type = XVMC_MB_TYPE_PATTERN;
179
180 if (s->mv_dir & MV_DIR_FORWARD) {
181 mv_block->macroblock_type |= XVMC_MB_TYPE_MOTION_FORWARD;
182 // PMV[n][dir][xy] = mv[dir][n][xy]
183 mv_block->PMV[0][0][0] = s->mv[0][0][0];
184 mv_block->PMV[0][0][1] = s->mv[0][0][1];
185 mv_block->PMV[1][0][0] = s->mv[0][1][0];
186 mv_block->PMV[1][0][1] = s->mv[0][1][1];
187 }
188 if (s->mv_dir & MV_DIR_BACKWARD) {
189 mv_block->macroblock_type |= XVMC_MB_TYPE_MOTION_BACKWARD;
190 mv_block->PMV[0][1][0] = s->mv[1][0][0];
191 mv_block->PMV[0][1][1] = s->mv[1][0][1];
192 mv_block->PMV[1][1][0] = s->mv[1][1][0];
193 mv_block->PMV[1][1][1] = s->mv[1][1][1];
194 }
195
196 switch(s->mv_type) {
197 case MV_TYPE_16X16:
198 mv_block->motion_type = XVMC_PREDICTION_FRAME;
199 break;
200 case MV_TYPE_16X8:
201 mv_block->motion_type = XVMC_PREDICTION_16x8;
202 break;
203 case MV_TYPE_FIELD:
204 mv_block->motion_type = XVMC_PREDICTION_FIELD;
205 if (s->picture_structure == PICT_FRAME) {
206 mv_block->PMV[0][0][1] <<= 1;
207 mv_block->PMV[1][0][1] <<= 1;
208 mv_block->PMV[0][1][1] <<= 1;
209 mv_block->PMV[1][1][1] <<= 1;
210 }
211 break;
212 case MV_TYPE_DMV:
213 mv_block->motion_type = XVMC_PREDICTION_DUAL_PRIME;
214 if (s->picture_structure == PICT_FRAME) {
215
216 mv_block->PMV[0][0][0] = s->mv[0][0][0]; // top from top
217 mv_block->PMV[0][0][1] = s->mv[0][0][1] << 1;
218
219 mv_block->PMV[0][1][0] = s->mv[0][0][0]; // bottom from bottom
220 mv_block->PMV[0][1][1] = s->mv[0][0][1] << 1;
221
222 mv_block->PMV[1][0][0] = s->mv[0][2][0]; // dmv00, top from bottom
223 mv_block->PMV[1][0][1] = s->mv[0][2][1] << 1; // dmv01
224
225 mv_block->PMV[1][1][0] = s->mv[0][3][0]; // dmv10, bottom from top
226 mv_block->PMV[1][1][1] = s->mv[0][3][1] << 1; // dmv11
227
228 } else {
229 mv_block->PMV[0][1][0] = s->mv[0][2][0]; // dmv00
230 mv_block->PMV[0][1][1] = s->mv[0][2][1]; // dmv01
231 }
232 break;
233 default:
234 assert(0);
235 }
236
237 mv_block->motion_vertical_field_select = 0;
238
239 // set correct field references
240 if (s->mv_type == MV_TYPE_FIELD || s->mv_type == MV_TYPE_16X8) {
241 mv_block->motion_vertical_field_select |= s->field_select[0][0];
242 mv_block->motion_vertical_field_select |= s->field_select[1][0] << 1;
243 mv_block->motion_vertical_field_select |= s->field_select[0][1] << 2;
244 mv_block->motion_vertical_field_select |= s->field_select[1][1] << 3;
245 }
246 } // !intra
247 // time to handle data blocks
248 mv_block->index = render->next_free_data_block_num;
249
250 blocks_per_mb = 6;
251 if (s->chroma_format >= 2) {
252 blocks_per_mb = 4 + (1 << s->chroma_format);
253 }
254
255 // calculate cbp
256 cbp = 0;
257 for (i = 0; i < blocks_per_mb; i++) {
258 cbp += cbp;
259 if (s->block_last_index[i] >= 0)
260 cbp++;
261 }
262
263 if (s->flags & CODEC_FLAG_GRAY) {
264 if (s->mb_intra) { // intra frames are always full chroma blocks
265 for (i = 4; i < blocks_per_mb; i++) {
266 memset(s->pblocks[i], 0, sizeof(short)*64); // so we need to clear them
267 if (!render->unsigned_intra)
268 s->pblocks[i][0] = 1 << 10;
269 }
270 } else {
271 cbp &= 0xf << (blocks_per_mb - 4);
272 blocks_per_mb = 4; // luminance blocks only
273 }
274 }
275 mv_block->coded_block_pattern = cbp;
276 if (cbp == 0)
277 mv_block->macroblock_type &= ~XVMC_MB_TYPE_PATTERN;
278
279 for (i = 0; i < blocks_per_mb; i++) {
280 if (s->block_last_index[i] >= 0) {
281 // I do not have unsigned_intra MOCO to test, hope it is OK.
282 if (s->mb_intra && (render->idct || (!render->idct && !render->unsigned_intra)))
283 s->pblocks[i][0] -= 1 << 10;
284 if (!render->idct) {
285 s->dsp.idct(s->pblocks[i]);
286 /* It is unclear if MC hardware requires pixel diff values to be
287 * in the range [-255;255]. TODO: Clipping if such hardware is
288 * ever found. As of now it would only be an unnecessary
289 * slowdown. */
290 }
291 // copy blocks only if the codec doesn't support pblocks reordering
292 if (s->avctx->xvmc_acceleration == 1) {
293 memcpy(&render->data_blocks[render->next_free_data_block_num*64],
294 s->pblocks[i], sizeof(short)*64);
295 }
296 render->next_free_data_block_num++;
297 }
298 }
299 render->filled_mv_blocks_num++;
300
301 assert(render->filled_mv_blocks_num <= render->total_number_of_mv_blocks);
302 assert(render->next_free_data_block_num <= render->total_number_of_data_blocks);
303 /* The above conditions should not be able to fail as long as this function
304 * is used and the following 'if ()' automatically calls a callback to free
305 * 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 }