blob: b469c4e26f2f9f0f373ea482902cbca285fc8a5b
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 | #include <X11/extensions/XvMC.h> |
24 | |
25 | #include "avcodec.h" |
26 | #include "mpegutils.h" |
27 | #include "mpegvideo.h" |
28 | |
29 | #undef NDEBUG |
30 | #include <assert.h> |
31 | |
32 | #include "xvmc.h" |
33 | #include "xvmc_internal.h" |
34 | #include "version.h" |
35 | |
36 | /** |
37 | * Initialize the block field of the MpegEncContext pointer passed as |
38 | * parameter after making sure that the data is not corrupted. |
39 | * In order to implement something like direct rendering instead of decoding |
40 | * coefficients in s->blocks and then copying them, copy them directly |
41 | * into the data_blocks array provided by xvmc. |
42 | */ |
43 | void ff_xvmc_init_block(MpegEncContext *s) |
44 | { |
45 | struct xvmc_pix_fmt *render = (struct xvmc_pix_fmt*)s->current_picture.f->data[2]; |
46 | assert(render && render->xvmc_id == AV_XVMC_ID); |
47 | |
48 | s->block = (int16_t (*)[64])(render->data_blocks + render->next_free_data_block_num * 64); |
49 | } |
50 | |
51 | static void exchange_uv(MpegEncContext *s) |
52 | { |
53 | int16_t (*tmp)[64]; |
54 | |
55 | tmp = s->pblocks[4]; |
56 | s->pblocks[4] = s->pblocks[5]; |
57 | s->pblocks[5] = tmp; |
58 | } |
59 | |
60 | /** |
61 | * Fill individual block pointers, so there are no gaps in the data_block array |
62 | * in case not all blocks in the macroblock are coded. |
63 | */ |
64 | void ff_xvmc_pack_pblocks(MpegEncContext *s, int cbp) |
65 | { |
66 | int i, j = 0; |
67 | const int mb_block_count = 4 + (1 << s->chroma_format); |
68 | |
69 | cbp <<= 12-mb_block_count; |
70 | for (i = 0; i < mb_block_count; i++) { |
71 | if (cbp & (1 << 11)) |
72 | s->pblocks[i] = &s->block[j++]; |
73 | else |
74 | s->pblocks[i] = NULL; |
75 | cbp += cbp; |
76 | } |
77 | if (s->swap_uv) { |
78 | exchange_uv(s); |
79 | } |
80 | } |
81 | |
82 | /** |
83 | * Find and store the surfaces that are used as reference frames. |
84 | * This function should be called for every new field and/or frame. |
85 | * It should be safe to call the function a few times for the same field. |
86 | */ |
87 | static int ff_xvmc_field_start(AVCodecContext *avctx, const uint8_t *buf, uint32_t buf_size) |
88 | { |
89 | struct MpegEncContext *s = avctx->priv_data; |
90 | struct xvmc_pix_fmt *last, *next, *render = (struct xvmc_pix_fmt*)s->current_picture.f->data[2]; |
91 | const int mb_block_count = 4 + (1 << s->chroma_format); |
92 | |
93 | assert(avctx); |
94 | if (!render || render->xvmc_id != AV_XVMC_ID || |
95 | !render->data_blocks || !render->mv_blocks || |
96 | (unsigned int)render->allocated_mv_blocks > INT_MAX/(64*6) || |
97 | (unsigned int)render->allocated_data_blocks > INT_MAX/64 || |
98 | !render->p_surface) { |
99 | av_log(avctx, AV_LOG_ERROR, |
100 | "Render token doesn't look as expected.\n"); |
101 | return -1; // make sure that this is a render packet |
102 | } |
103 | |
104 | if (render->filled_mv_blocks_num) { |
105 | av_log(avctx, AV_LOG_ERROR, |
106 | "Rendering surface contains %i unprocessed blocks.\n", |
107 | render->filled_mv_blocks_num); |
108 | return -1; |
109 | } |
110 | if (render->allocated_mv_blocks < 1 || |
111 | render->allocated_data_blocks < render->allocated_mv_blocks*mb_block_count || |
112 | render->start_mv_blocks_num >= render->allocated_mv_blocks || |
113 | render->next_free_data_block_num > |
114 | render->allocated_data_blocks - |
115 | mb_block_count*(render->allocated_mv_blocks-render->start_mv_blocks_num)) { |
116 | av_log(avctx, AV_LOG_ERROR, |
117 | "Rendering surface doesn't provide enough block structures to work with.\n"); |
118 | return -1; |
119 | } |
120 | |
121 | render->picture_structure = s->picture_structure; |
122 | render->flags = s->first_field ? 0 : XVMC_SECOND_FIELD; |
123 | render->p_future_surface = NULL; |
124 | render->p_past_surface = NULL; |
125 | |
126 | switch(s->pict_type) { |
127 | case AV_PICTURE_TYPE_I: |
128 | return 0; // no prediction from other frames |
129 | case AV_PICTURE_TYPE_B: |
130 | next = (struct xvmc_pix_fmt*)s->next_picture.f->data[2]; |
131 | if (!next) |
132 | return -1; |
133 | if (next->xvmc_id != AV_XVMC_ID) |
134 | return -1; |
135 | render->p_future_surface = next->p_surface; |
136 | // no return here, going to set forward prediction |
137 | case AV_PICTURE_TYPE_P: |
138 | last = (struct xvmc_pix_fmt*)s->last_picture.f->data[2]; |
139 | if (!last) |
140 | last = render; // predict second field from the first |
141 | if (last->xvmc_id != AV_XVMC_ID) |
142 | return -1; |
143 | render->p_past_surface = last->p_surface; |
144 | return 0; |
145 | } |
146 | |
147 | return -1; |
148 | } |
149 | |
150 | /** |
151 | * Complete frame/field rendering by passing any remaining blocks. |
152 | * Normally ff_draw_horiz_band() is called for each slice, however, |
153 | * some leftover blocks, for example from error_resilience(), may remain. |
154 | * It should be safe to call the function a few times for the same field. |
155 | */ |
156 | static int ff_xvmc_field_end(AVCodecContext *avctx) |
157 | { |
158 | struct MpegEncContext *s = avctx->priv_data; |
159 | struct xvmc_pix_fmt *render = (struct xvmc_pix_fmt*)s->current_picture.f->data[2]; |
160 | assert(render); |
161 | |
162 | if (render->filled_mv_blocks_num > 0) |
163 | ff_mpeg_draw_horiz_band(s, 0, 0); |
164 | return 0; |
165 | } |
166 | |
167 | /** |
168 | * Synthesize the data needed by XvMC to render one macroblock of data. |
169 | * Fill all relevant fields, if necessary do IDCT. |
170 | */ |
171 | static void ff_xvmc_decode_mb(struct MpegEncContext *s) |
172 | { |
173 | XvMCMacroBlock *mv_block; |
174 | struct xvmc_pix_fmt *render; |
175 | int i, cbp, blocks_per_mb; |
176 | |
177 | const int mb_xy = s->mb_y * s->mb_stride + s->mb_x; |
178 | |
179 | |
180 | if (s->encoding) { |
181 | av_log(s->avctx, AV_LOG_ERROR, "XVMC doesn't support encoding!!!\n"); |
182 | return; |
183 | } |
184 | |
185 | // from ff_mpv_decode_mb(), update DC predictors for P macroblocks |
186 | if (!s->mb_intra) { |
187 | s->last_dc[0] = |
188 | s->last_dc[1] = |
189 | s->last_dc[2] = 128 << s->intra_dc_precision; |
190 | } |
191 | |
192 | // MC doesn't skip blocks |
193 | s->mb_skipped = 0; |
194 | |
195 | |
196 | // Do I need to export quant when I could not perform postprocessing? |
197 | // Anyway, it doesn't hurt. |
198 | s->current_picture.qscale_table[mb_xy] = s->qscale; |
199 | |
200 | // start of XVMC-specific code |
201 | render = (struct xvmc_pix_fmt*)s->current_picture.f->data[2]; |
202 | assert(render); |
203 | assert(render->xvmc_id == AV_XVMC_ID); |
204 | assert(render->mv_blocks); |
205 | |
206 | // take the next free macroblock |
207 | mv_block = &render->mv_blocks[render->start_mv_blocks_num + |
208 | render->filled_mv_blocks_num]; |
209 | |
210 | mv_block->x = s->mb_x; |
211 | mv_block->y = s->mb_y; |
212 | mv_block->dct_type = s->interlaced_dct; // XVMC_DCT_TYPE_FRAME/FIELD; |
213 | if (s->mb_intra) { |
214 | mv_block->macroblock_type = XVMC_MB_TYPE_INTRA; // no MC, all done |
215 | } else { |
216 | mv_block->macroblock_type = XVMC_MB_TYPE_PATTERN; |
217 | |
218 | if (s->mv_dir & MV_DIR_FORWARD) { |
219 | mv_block->macroblock_type |= XVMC_MB_TYPE_MOTION_FORWARD; |
220 | // PMV[n][dir][xy] = mv[dir][n][xy] |
221 | mv_block->PMV[0][0][0] = s->mv[0][0][0]; |
222 | mv_block->PMV[0][0][1] = s->mv[0][0][1]; |
223 | mv_block->PMV[1][0][0] = s->mv[0][1][0]; |
224 | mv_block->PMV[1][0][1] = s->mv[0][1][1]; |
225 | } |
226 | if (s->mv_dir & MV_DIR_BACKWARD) { |
227 | mv_block->macroblock_type |= XVMC_MB_TYPE_MOTION_BACKWARD; |
228 | mv_block->PMV[0][1][0] = s->mv[1][0][0]; |
229 | mv_block->PMV[0][1][1] = s->mv[1][0][1]; |
230 | mv_block->PMV[1][1][0] = s->mv[1][1][0]; |
231 | mv_block->PMV[1][1][1] = s->mv[1][1][1]; |
232 | } |
233 | |
234 | switch(s->mv_type) { |
235 | case MV_TYPE_16X16: |
236 | mv_block->motion_type = XVMC_PREDICTION_FRAME; |
237 | break; |
238 | case MV_TYPE_16X8: |
239 | mv_block->motion_type = XVMC_PREDICTION_16x8; |
240 | break; |
241 | case MV_TYPE_FIELD: |
242 | mv_block->motion_type = XVMC_PREDICTION_FIELD; |
243 | if (s->picture_structure == PICT_FRAME) { |
244 | mv_block->PMV[0][0][1] <<= 1; |
245 | mv_block->PMV[1][0][1] <<= 1; |
246 | mv_block->PMV[0][1][1] <<= 1; |
247 | mv_block->PMV[1][1][1] <<= 1; |
248 | } |
249 | break; |
250 | case MV_TYPE_DMV: |
251 | mv_block->motion_type = XVMC_PREDICTION_DUAL_PRIME; |
252 | if (s->picture_structure == PICT_FRAME) { |
253 | |
254 | mv_block->PMV[0][0][0] = s->mv[0][0][0]; // top from top |
255 | mv_block->PMV[0][0][1] = s->mv[0][0][1] << 1; |
256 | |
257 | mv_block->PMV[0][1][0] = s->mv[0][0][0]; // bottom from bottom |
258 | mv_block->PMV[0][1][1] = s->mv[0][0][1] << 1; |
259 | |
260 | mv_block->PMV[1][0][0] = s->mv[0][2][0]; // dmv00, top from bottom |
261 | mv_block->PMV[1][0][1] = s->mv[0][2][1] << 1; // dmv01 |
262 | |
263 | mv_block->PMV[1][1][0] = s->mv[0][3][0]; // dmv10, bottom from top |
264 | mv_block->PMV[1][1][1] = s->mv[0][3][1] << 1; // dmv11 |
265 | |
266 | } else { |
267 | mv_block->PMV[0][1][0] = s->mv[0][2][0]; // dmv00 |
268 | mv_block->PMV[0][1][1] = s->mv[0][2][1]; // dmv01 |
269 | } |
270 | break; |
271 | default: |
272 | assert(0); |
273 | } |
274 | |
275 | mv_block->motion_vertical_field_select = 0; |
276 | |
277 | // set correct field references |
278 | if (s->mv_type == MV_TYPE_FIELD || s->mv_type == MV_TYPE_16X8) { |
279 | mv_block->motion_vertical_field_select |= s->field_select[0][0]; |
280 | mv_block->motion_vertical_field_select |= s->field_select[1][0] << 1; |
281 | mv_block->motion_vertical_field_select |= s->field_select[0][1] << 2; |
282 | mv_block->motion_vertical_field_select |= s->field_select[1][1] << 3; |
283 | } |
284 | } // !intra |
285 | // time to handle data blocks |
286 | mv_block->index = render->next_free_data_block_num; |
287 | |
288 | blocks_per_mb = 6; |
289 | if (s->chroma_format >= 2) { |
290 | blocks_per_mb = 4 + (1 << s->chroma_format); |
291 | } |
292 | |
293 | // calculate cbp |
294 | cbp = 0; |
295 | for (i = 0; i < blocks_per_mb; i++) { |
296 | cbp += cbp; |
297 | if (s->block_last_index[i] >= 0) |
298 | cbp++; |
299 | } |
300 | |
301 | if (s->avctx->flags & AV_CODEC_FLAG_GRAY) { |
302 | if (s->mb_intra) { // intra frames are always full chroma blocks |
303 | for (i = 4; i < blocks_per_mb; i++) { |
304 | memset(s->pblocks[i], 0, sizeof(*s->pblocks[i])); // so we need to clear them |
305 | if (!render->unsigned_intra) |
306 | *s->pblocks[i][0] = 1 << 10; |
307 | } |
308 | } else { |
309 | cbp &= 0xf << (blocks_per_mb - 4); |
310 | blocks_per_mb = 4; // luminance blocks only |
311 | } |
312 | } |
313 | mv_block->coded_block_pattern = cbp; |
314 | if (cbp == 0) |
315 | mv_block->macroblock_type &= ~XVMC_MB_TYPE_PATTERN; |
316 | |
317 | for (i = 0; i < blocks_per_mb; i++) { |
318 | if (s->block_last_index[i] >= 0) { |
319 | // I do not have unsigned_intra MOCO to test, hope it is OK. |
320 | if (s->mb_intra && (render->idct || !render->unsigned_intra)) |
321 | *s->pblocks[i][0] -= 1 << 10; |
322 | if (!render->idct) { |
323 | s->idsp.idct(*s->pblocks[i]); |
324 | /* It is unclear if MC hardware requires pixel diff values to be |
325 | * in the range [-255;255]. TODO: Clipping if such hardware is |
326 | * ever found. As of now it would only be an unnecessary |
327 | * slowdown. */ |
328 | } |
329 | // copy blocks only if the codec doesn't support pblocks reordering |
330 | if (!s->pack_pblocks) { |
331 | memcpy(&render->data_blocks[render->next_free_data_block_num*64], |
332 | s->pblocks[i], sizeof(*s->pblocks[i])); |
333 | } |
334 | render->next_free_data_block_num++; |
335 | } |
336 | } |
337 | render->filled_mv_blocks_num++; |
338 | |
339 | assert(render->filled_mv_blocks_num <= render->allocated_mv_blocks); |
340 | assert(render->next_free_data_block_num <= render->allocated_data_blocks); |
341 | /* The above conditions should not be able to fail as long as this function |
342 | * is used and the following 'if ()' automatically calls a callback to free |
343 | * blocks. */ |
344 | |
345 | |
346 | if (render->filled_mv_blocks_num == render->allocated_mv_blocks) |
347 | ff_mpeg_draw_horiz_band(s, 0, 0); |
348 | } |
349 | |
350 | #if CONFIG_MPEG1_XVMC_HWACCEL |
351 | AVHWAccel ff_mpeg1_xvmc_hwaccel = { |
352 | .name = "mpeg1_xvmc", |
353 | .type = AVMEDIA_TYPE_VIDEO, |
354 | .id = AV_CODEC_ID_MPEG1VIDEO, |
355 | .pix_fmt = AV_PIX_FMT_XVMC, |
356 | .start_frame = ff_xvmc_field_start, |
357 | .end_frame = ff_xvmc_field_end, |
358 | .decode_slice = NULL, |
359 | .decode_mb = ff_xvmc_decode_mb, |
360 | .priv_data_size = 0, |
361 | }; |
362 | #endif |
363 | |
364 | #if CONFIG_MPEG2_XVMC_HWACCEL |
365 | AVHWAccel ff_mpeg2_xvmc_hwaccel = { |
366 | .name = "mpeg2_xvmc", |
367 | .type = AVMEDIA_TYPE_VIDEO, |
368 | .id = AV_CODEC_ID_MPEG2VIDEO, |
369 | .pix_fmt = AV_PIX_FMT_XVMC, |
370 | .start_frame = ff_xvmc_field_start, |
371 | .end_frame = ff_xvmc_field_end, |
372 | .decode_slice = NULL, |
373 | .decode_mb = ff_xvmc_decode_mb, |
374 | .priv_data_size = 0, |
375 | }; |
376 | #endif |
377 |