Doxygen comment/explanation for ff_xvmc_init_block().
[libav.git] / libavcodec / mpegvideo_xvmc.c
CommitLineData
302898fc
IK
1/*
2 * XVideo Motion Compensation
3 * Copyright (c) 2003 Ivan Kalvachev
4 *
b78e7197
DB
5 * This file is part of FFmpeg.
6 *
7 * FFmpeg is free software; you can redistribute it and/or
302898fc
IK
8 * modify it under the terms of the GNU Lesser General Public
9 * License as published by the Free Software Foundation; either
b78e7197 10 * version 2.1 of the License, or (at your option) any later version.
302898fc 11 *
b78e7197 12 * FFmpeg is distributed in the hope that it will be useful,
302898fc
IK
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
b78e7197 18 * License along with FFmpeg; if not, write to the Free Software
5509bffa 19 * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
302898fc
IK
20 */
21
2e7b4c84
IK
22#include <limits.h>
23
2e7b4c84
IK
24#include "avcodec.h"
25#include "dsputil.h"
26#include "mpegvideo.h"
27
28#undef NDEBUG
29#include <assert.h>
30
fd949a63 31#include "xvmc.h"
4440bd0d 32#include "xvmc_internal.h"
302898fc 33
8f38ff00
DB
34/**
35 * Initializes the block field of the MpegEncContext pointer passed as
36 * parameter after making sure that the data is not corrupted.
37 */
78f9a878 38void ff_xvmc_init_block(MpegEncContext *s)
148302e7 39{
b71bd0ed 40 struct xvmc_render_state *render = (struct xvmc_render_state*)s->current_picture.data[2];
3ae01928 41 assert(render);
9f00a41c 42 if (!render || render->magic != AV_XVMC_RENDER_MAGIC) {
a579db0c 43 assert(0);
1f7c1d14 44 return; // make sure that this is a render packet
a579db0c 45 }
9f00a41c 46 s->block = (DCTELEM *)(render->data_blocks + render->next_free_data_block_num * 64);
a579db0c
IK
47}
48
78f9a878 49void ff_xvmc_pack_pblocks(MpegEncContext *s, int cbp)
148302e7 50{
b71bd0ed 51 int i, j = 0;
148302e7 52 const int mb_block_count = 4 + (1 << s->chroma_format);
a579db0c 53
148302e7
DB
54 cbp <<= 12-mb_block_count;
55 for (i = 0; i < mb_block_count; i++) {
4ec58e13 56 if (cbp & (1 << 11))
4742762f 57 s->pblocks[i] = (short *)(&s->block[j++]);
4ec58e13 58 else
39dba5e8 59 s->pblocks[i] = NULL;
bb270c08 60 cbp+=cbp;
a579db0c 61 }
2e7b4c84
IK
62}
63
9107f7cd
DB
64/**
65 * This function should be called for every new field and/or frame.
66 * It should be safe to call the function a few times for the same field.
67 */
78f9a878 68int ff_xvmc_field_start(MpegEncContext*s, AVCodecContext *avctx)
148302e7 69{
b71bd0ed 70 struct xvmc_render_state *last, *next, *render = (struct xvmc_render_state*)s->current_picture.data[2];
2e7b4c84 71
3ae01928 72 assert(avctx);
3ae01928 73 assert(render);
9f00a41c 74 if (!render || render->magic != AV_XVMC_RENDER_MAGIC)
a1db3b93 75 return -1; // make sure that this is a render packet
2e7b4c84
IK
76
77 render->picture_structure = s->picture_structure;
9f00a41c 78 render->flags = s->first_field ? 0 : XVMC_SECOND_FIELD;
2e7b4c84 79
cb18fb62 80 if (render->filled_mv_blocks_num) {
9939841f 81 av_log(avctx, AV_LOG_ERROR,
cb18fb62
DB
82 "Rendering surface contains %i unprocessed blocks\n",
83 render->filled_mv_blocks_num);
013cebfb 84 return -1;
9939841f 85 }
2e7b4c84
IK
86
87 render->p_future_surface = NULL;
4ef690f5 88 render->p_past_surface = NULL;
2e7b4c84 89
7e2e870e 90 switch(s->pict_type) {
9701840b 91 case FF_I_TYPE:
1f7c1d14 92 return 0; // no prediction from other frames
9701840b 93 case FF_B_TYPE:
c96da3ed 94 next = (struct xvmc_render_state*)s->next_picture.data[2];
3ae01928
DB
95 assert(next);
96 if (!next)
148302e7 97 return -1;
ea375af8 98 if (next->magic != AV_XVMC_RENDER_MAGIC)
148302e7 99 return -1;
2e7b4c84 100 render->p_future_surface = next->p_surface;
1f7c1d14 101 // no return here, going to set forward prediction
9701840b 102 case FF_P_TYPE:
c96da3ed 103 last = (struct xvmc_render_state*)s->last_picture.data[2];
63a21bc0 104 if (!last)
1f7c1d14 105 last = render; // predict second field from the first
ea375af8 106 if (last->magic != AV_XVMC_RENDER_MAGIC)
148302e7 107 return -1;
2e7b4c84
IK
108 render->p_past_surface = last->p_surface;
109 return 0;
5e5c247a 110 }
2e7b4c84
IK
111
112return -1;
113}
114
0b2eb2e8
DB
115/**
116 * This function should be called for every new field and/or frame.
117 * It should be safe to call the function a few times for the same field.
118 */
78f9a878 119void ff_xvmc_field_end(MpegEncContext *s)
148302e7 120{
b71bd0ed 121 struct xvmc_render_state *render = (struct xvmc_render_state*)s->current_picture.data[2];
3ae01928 122 assert(render);
2e7b4c84 123
148302e7 124 if (render->filled_mv_blocks_num > 0)
99df0aa5 125 ff_draw_horiz_band(s, 0, 0);
2e7b4c84
IK
126}
127
78f9a878 128void ff_xvmc_decode_mb(MpegEncContext *s)
148302e7 129{
7e2e870e
DB
130 XvMCMacroBlock *mv_block;
131 struct xvmc_render_state *render;
132 int i, cbp, blocks_per_mb;
2e7b4c84 133
39dba5e8 134 const int mb_xy = s->mb_y * s->mb_stride + s->mb_x;
2e7b4c84
IK
135
136
148302e7 137 if (s->encoding) {
8ac5c1b2 138 av_log(s->avctx, AV_LOG_ERROR, "XVMC doesn't support encoding!!!\n");
61b76987 139 return;
2e7b4c84
IK
140 }
141
1f7c1d14 142 // from MPV_decode_mb(), update DC predictors for P macroblocks
2e7b4c84
IK
143 if (!s->mb_intra) {
144 s->last_dc[0] =
145 s->last_dc[1] =
146 s->last_dc[2] = 128 << s->intra_dc_precision;
147 }
148
1f7c1d14 149 // MC doesn't skip blocks
160d679c 150 s->mb_skipped = 0;
2e7b4c84
IK
151
152
d15876d3
DB
153 // Do I need to export quant when I could not perform postprocessing?
154 // Anyway, it doesn't hurt.
2e7b4c84
IK
155 s->current_picture.qscale_table[mb_xy] = s->qscale;
156
a1db3b93 157 // start of XVMC-specific code
c96da3ed 158 render = (struct xvmc_render_state*)s->current_picture.data[2];
3ae01928 159 assert(render);
7e2e870e 160 assert(render->magic == AV_XVMC_RENDER_MAGIC);
2e7b4c84 161 assert(render->mv_blocks);
302898fc 162
1f7c1d14 163 // take the next free macroblock
115329f1 164 mv_block = &render->mv_blocks[render->start_mv_blocks_num +
fe4be5db 165 render->filled_mv_blocks_num];
2e7b4c84 166
4ef690f5
DB
167 mv_block->x = s->mb_x;
168 mv_block->y = s->mb_y;
1f7c1d14 169 mv_block->dct_type = s->interlaced_dct; // XVMC_DCT_TYPE_FRAME/FIELD;
4ec58e13 170 if (s->mb_intra) {
1f7c1d14 171 mv_block->macroblock_type = XVMC_MB_TYPE_INTRA; // no MC, all done
4ec58e13 172 } else {
2e7b4c84
IK
173 mv_block->macroblock_type = XVMC_MB_TYPE_PATTERN;
174
148302e7 175 if (s->mv_dir & MV_DIR_FORWARD) {
2d7d8cc7 176 mv_block->macroblock_type |= XVMC_MB_TYPE_MOTION_FORWARD;
0439c09d 177 // PMV[n][dir][xy] = mv[dir][n][xy]
2e7b4c84
IK
178 mv_block->PMV[0][0][0] = s->mv[0][0][0];
179 mv_block->PMV[0][0][1] = s->mv[0][0][1];
180 mv_block->PMV[1][0][0] = s->mv[0][1][0];
181 mv_block->PMV[1][0][1] = s->mv[0][1][1];
182 }
148302e7 183 if (s->mv_dir & MV_DIR_BACKWARD) {
2d7d8cc7 184 mv_block->macroblock_type |= XVMC_MB_TYPE_MOTION_BACKWARD;
2e7b4c84
IK
185 mv_block->PMV[0][1][0] = s->mv[1][0][0];
186 mv_block->PMV[0][1][1] = s->mv[1][0][1];
187 mv_block->PMV[1][1][0] = s->mv[1][1][0];
188 mv_block->PMV[1][1][1] = s->mv[1][1][1];
189 }
190
99df0aa5 191 switch(s->mv_type) {
2e7b4c84
IK
192 case MV_TYPE_16X16:
193 mv_block->motion_type = XVMC_PREDICTION_FRAME;
194 break;
195 case MV_TYPE_16X8:
196 mv_block->motion_type = XVMC_PREDICTION_16x8;
197 break;
198 case MV_TYPE_FIELD:
199 mv_block->motion_type = XVMC_PREDICTION_FIELD;
148302e7
DB
200 if (s->picture_structure == PICT_FRAME) {
201 mv_block->PMV[0][0][1] <<= 1;
202 mv_block->PMV[1][0][1] <<= 1;
203 mv_block->PMV[0][1][1] <<= 1;
204 mv_block->PMV[1][1][1] <<= 1;
2e7b4c84
IK
205 }
206 break;
207 case MV_TYPE_DMV:
208 mv_block->motion_type = XVMC_PREDICTION_DUAL_PRIME;
148302e7 209 if (s->picture_structure == PICT_FRAME) {
2e7b4c84 210
ba585726 211 mv_block->PMV[0][0][0] = s->mv[0][0][0]; // top from top
81189e4f 212 mv_block->PMV[0][0][1] = s->mv[0][0][1] << 1;
2e7b4c84 213
ba585726 214 mv_block->PMV[0][1][0] = s->mv[0][0][0]; // bottom from bottom
81189e4f 215 mv_block->PMV[0][1][1] = s->mv[0][0][1] << 1;
2e7b4c84 216
ba585726 217 mv_block->PMV[1][0][0] = s->mv[0][2][0]; // dmv00, top from bottom
81189e4f 218 mv_block->PMV[1][0][1] = s->mv[0][2][1] << 1; // dmv01
2e7b4c84 219
ba585726 220 mv_block->PMV[1][1][0] = s->mv[0][3][0]; // dmv10, bottom from top
81189e4f 221 mv_block->PMV[1][1][1] = s->mv[0][3][1] << 1; // dmv11
2e7b4c84 222
4ec58e13 223 } else {
ba585726
DB
224 mv_block->PMV[0][1][0] = s->mv[0][2][0]; // dmv00
225 mv_block->PMV[0][1][1] = s->mv[0][2][1]; // dmv01
2e7b4c84
IK
226 }
227 break;
228 default:
229 assert(0);
230 }
231
232 mv_block->motion_vertical_field_select = 0;
233
1f7c1d14 234 // set correct field references
148302e7 235 if (s->mv_type == MV_TYPE_FIELD || s->mv_type == MV_TYPE_16X8) {
90509ec7 236 mv_block->motion_vertical_field_select |= s->field_select[0][0];
7e2e870e
DB
237 mv_block->motion_vertical_field_select |= s->field_select[1][0] << 1;
238 mv_block->motion_vertical_field_select |= s->field_select[0][1] << 2;
239 mv_block->motion_vertical_field_select |= s->field_select[1][1] << 3;
2e7b4c84 240 }
1f7c1d14 241 } // !intra
a1db3b93 242 // time to handle data blocks
2e7b4c84 243 mv_block->index = render->next_free_data_block_num;
5e5c247a 244
2e7b4c84 245 blocks_per_mb = 6;
148302e7 246 if (s->chroma_format >= 2) {
4742762f 247 blocks_per_mb = 4 + (1 << s->chroma_format);
2e7b4c84 248 }
5e5c247a 249
1f7c1d14 250 // calculate cbp
715731a3 251 cbp = 0;
148302e7 252 for (i = 0; i < blocks_per_mb; i++) {
2d7d8cc7 253 cbp += cbp;
148302e7 254 if (s->block_last_index[i] >= 0)
715731a3
IK
255 cbp++;
256 }
115329f1 257
148302e7 258 if (s->flags & CODEC_FLAG_GRAY) {
ba585726 259 if (s->mb_intra) { // intra frames are always full chroma blocks
148302e7 260 for (i = 4; i < blocks_per_mb; i++) {
6e4b41f8 261 memset(s->pblocks[i], 0, sizeof(short)*64); // so we need to clear them
148302e7
DB
262 if (!render->unsigned_intra)
263 s->pblocks[i][0] = 1 << 10;
a579db0c 264 }
4ec58e13 265 } else {
148302e7 266 cbp &= 0xf << (blocks_per_mb - 4);
ba585726 267 blocks_per_mb = 4; // luminance blocks only
715731a3 268 }
a579db0c 269 }
2e7b4c84 270 mv_block->coded_block_pattern = cbp;
148302e7 271 if (cbp == 0)
2e7b4c84
IK
272 mv_block->macroblock_type &= ~XVMC_MB_TYPE_PATTERN;
273
148302e7
DB
274 for (i = 0; i < blocks_per_mb; i++) {
275 if (s->block_last_index[i] >= 0) {
7a74e067 276 // I do not have unsigned_intra MOCO to test, hope it is OK.
eba9cecc 277 if (s->mb_intra && (render->idct || (!render->idct && !render->unsigned_intra)))
148302e7
DB
278 s->pblocks[i][0] -= 1 << 10;
279 if (!render->idct) {
a579db0c 280 s->dsp.idct(s->pblocks[i]);
a1db3b93
DB
281 /* It is unclear if MC hardware requires pixel diff values to be
282 * in the range [-255;255]. TODO: Clipping if such hardware is
283 * ever found. As of now it would only be an unnecessary
284 * slowdown. */
2e7b4c84 285 }
1f7c1d14 286 // copy blocks only if the codec doesn't support pblocks reordering
148302e7 287 if (s->avctx->xvmc_acceleration == 1) {
4742762f 288 memcpy(&render->data_blocks[render->next_free_data_block_num*64],
6e280340 289 s->pblocks[i], sizeof(short)*64);
a579db0c
IK
290 }
291 render->next_free_data_block_num++;
2e7b4c84
IK
292 }
293 }
294 render->filled_mv_blocks_num++;
295
4ef690f5 296 assert(render->filled_mv_blocks_num <= render->total_number_of_mv_blocks);
2e7b4c84 297 assert(render->next_free_data_block_num <= render->total_number_of_data_blocks);
930efaf4
DB
298 /* The above conditions should not be able to fail as long as this function
299 * is used and the following 'if ()' automatically calls a callback to free
300 * blocks. */
2e7b4c84
IK
301
302
148302e7 303 if (render->filled_mv_blocks_num >= render->total_number_of_mv_blocks)
7e2e870e 304 ff_draw_horiz_band(s, 0, 0);
2e7b4c84 305}