K&R whitespace cosmetics
[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{
095edd3e 40 struct xvmc_pixfmt_render *render = (struct xvmc_pixfmt_render*)s->current_picture.data[2];
d2d600b7
IK
41 assert(render && render->magic_id == AV_XVMC_RENDER_MAGIC);
42
9f00a41c 43 s->block = (DCTELEM *)(render->data_blocks + render->next_free_data_block_num * 64);
a579db0c
IK
44}
45
78f9a878 46void ff_xvmc_pack_pblocks(MpegEncContext *s, int cbp)
148302e7 47{
b71bd0ed 48 int i, j = 0;
148302e7 49 const int mb_block_count = 4 + (1 << s->chroma_format);
a579db0c 50
148302e7
DB
51 cbp <<= 12-mb_block_count;
52 for (i = 0; i < mb_block_count; i++) {
4ec58e13 53 if (cbp & (1 << 11))
4742762f 54 s->pblocks[i] = (short *)(&s->block[j++]);
4ec58e13 55 else
39dba5e8 56 s->pblocks[i] = NULL;
bb270c08 57 cbp+=cbp;
a579db0c 58 }
2e7b4c84
IK
59}
60
9107f7cd
DB
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 */
78f9a878 65int ff_xvmc_field_start(MpegEncContext*s, AVCodecContext *avctx)
148302e7 66{
095edd3e 67 struct xvmc_pixfmt_render *last, *next, *render = (struct xvmc_pixfmt_render*)s->current_picture.data[2];
d2d600b7 68 const int mb_block_count = 4 + (1 << s->chroma_format);
2e7b4c84 69
3ae01928 70 assert(avctx);
d2d600b7 71 if (!render || render->magic_id != AV_XVMC_RENDER_MAGIC ||
9043202e 72 !render->data_blocks || !render->mv_blocks) {
d2d600b7
IK
73 av_log(avctx, AV_LOG_ERROR,
74 "Render token doesn't look as expected.\n");
a1db3b93 75 return -1; // make sure that this is a render packet
d2d600b7 76 }
2e7b4c84 77
cb18fb62 78 if (render->filled_mv_blocks_num) {
9939841f 79 av_log(avctx, AV_LOG_ERROR,
d2d600b7 80 "Rendering surface contains %i unprocessed blocks.\n",
cb18fb62 81 render->filled_mv_blocks_num);
013cebfb 82 return -1;
9939841f 83 }
d2d600b7
IK
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 }
2e7b4c84 90
d76c5ed5
IK
91 render->picture_structure = s->picture_structure;
92 render->flags = s->first_field ? 0 : XVMC_SECOND_FIELD;
9e0a8a36
IK
93 render->p_future_surface = NULL;
94 render->p_past_surface = NULL;
2e7b4c84 95
7e2e870e 96 switch(s->pict_type) {
9701840b 97 case FF_I_TYPE:
1f7c1d14 98 return 0; // no prediction from other frames
9701840b 99 case FF_B_TYPE:
095edd3e 100 next = (struct xvmc_pixfmt_render*)s->next_picture.data[2];
3ae01928 101 if (!next)
148302e7 102 return -1;
a0723a4d 103 if (next->magic_id != AV_XVMC_RENDER_MAGIC)
148302e7 104 return -1;
2e7b4c84 105 render->p_future_surface = next->p_surface;
1f7c1d14 106 // no return here, going to set forward prediction
9701840b 107 case FF_P_TYPE:
095edd3e 108 last = (struct xvmc_pixfmt_render*)s->last_picture.data[2];
63a21bc0 109 if (!last)
1f7c1d14 110 last = render; // predict second field from the first
a0723a4d 111 if (last->magic_id != AV_XVMC_RENDER_MAGIC)
148302e7 112 return -1;
2e7b4c84
IK
113 render->p_past_surface = last->p_surface;
114 return 0;
5e5c247a 115 }
2e7b4c84
IK
116
117return -1;
118}
119
0b2eb2e8
DB
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 */
78f9a878 124void ff_xvmc_field_end(MpegEncContext *s)
148302e7 125{
095edd3e 126 struct xvmc_pixfmt_render *render = (struct xvmc_pixfmt_render*)s->current_picture.data[2];
3ae01928 127 assert(render);
2e7b4c84 128
148302e7 129 if (render->filled_mv_blocks_num > 0)
99df0aa5 130 ff_draw_horiz_band(s, 0, 0);
2e7b4c84
IK
131}
132
78f9a878 133void ff_xvmc_decode_mb(MpegEncContext *s)
148302e7 134{
7e2e870e 135 XvMCMacroBlock *mv_block;
095edd3e 136 struct xvmc_pixfmt_render *render;
7e2e870e 137 int i, cbp, blocks_per_mb;
2e7b4c84 138
39dba5e8 139 const int mb_xy = s->mb_y * s->mb_stride + s->mb_x;
2e7b4c84
IK
140
141
148302e7 142 if (s->encoding) {
8ac5c1b2 143 av_log(s->avctx, AV_LOG_ERROR, "XVMC doesn't support encoding!!!\n");
61b76987 144 return;
2e7b4c84
IK
145 }
146
1f7c1d14 147 // from MPV_decode_mb(), update DC predictors for P macroblocks
2e7b4c84
IK
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
1f7c1d14 154 // MC doesn't skip blocks
160d679c 155 s->mb_skipped = 0;
2e7b4c84
IK
156
157
d15876d3
DB
158 // Do I need to export quant when I could not perform postprocessing?
159 // Anyway, it doesn't hurt.
2e7b4c84
IK
160 s->current_picture.qscale_table[mb_xy] = s->qscale;
161
a1db3b93 162 // start of XVMC-specific code
095edd3e 163 render = (struct xvmc_pixfmt_render*)s->current_picture.data[2];
3ae01928 164 assert(render);
a0723a4d 165 assert(render->magic_id == AV_XVMC_RENDER_MAGIC);
2e7b4c84 166 assert(render->mv_blocks);
302898fc 167
1f7c1d14 168 // take the next free macroblock
115329f1 169 mv_block = &render->mv_blocks[render->start_mv_blocks_num +
fe4be5db 170 render->filled_mv_blocks_num];
2e7b4c84 171
4ef690f5
DB
172 mv_block->x = s->mb_x;
173 mv_block->y = s->mb_y;
1f7c1d14 174 mv_block->dct_type = s->interlaced_dct; // XVMC_DCT_TYPE_FRAME/FIELD;
4ec58e13 175 if (s->mb_intra) {
1f7c1d14 176 mv_block->macroblock_type = XVMC_MB_TYPE_INTRA; // no MC, all done
4ec58e13 177 } else {
2e7b4c84
IK
178 mv_block->macroblock_type = XVMC_MB_TYPE_PATTERN;
179
148302e7 180 if (s->mv_dir & MV_DIR_FORWARD) {
2d7d8cc7 181 mv_block->macroblock_type |= XVMC_MB_TYPE_MOTION_FORWARD;
0439c09d 182 // PMV[n][dir][xy] = mv[dir][n][xy]
2e7b4c84
IK
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 }
148302e7 188 if (s->mv_dir & MV_DIR_BACKWARD) {
2d7d8cc7 189 mv_block->macroblock_type |= XVMC_MB_TYPE_MOTION_BACKWARD;
2e7b4c84
IK
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
99df0aa5 196 switch(s->mv_type) {
2e7b4c84
IK
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;
148302e7
DB
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;
2e7b4c84
IK
210 }
211 break;
212 case MV_TYPE_DMV:
213 mv_block->motion_type = XVMC_PREDICTION_DUAL_PRIME;
148302e7 214 if (s->picture_structure == PICT_FRAME) {
2e7b4c84 215
ba585726 216 mv_block->PMV[0][0][0] = s->mv[0][0][0]; // top from top
81189e4f 217 mv_block->PMV[0][0][1] = s->mv[0][0][1] << 1;
2e7b4c84 218
ba585726 219 mv_block->PMV[0][1][0] = s->mv[0][0][0]; // bottom from bottom
81189e4f 220 mv_block->PMV[0][1][1] = s->mv[0][0][1] << 1;
2e7b4c84 221
ba585726 222 mv_block->PMV[1][0][0] = s->mv[0][2][0]; // dmv00, top from bottom
81189e4f 223 mv_block->PMV[1][0][1] = s->mv[0][2][1] << 1; // dmv01
2e7b4c84 224
ba585726 225 mv_block->PMV[1][1][0] = s->mv[0][3][0]; // dmv10, bottom from top
81189e4f 226 mv_block->PMV[1][1][1] = s->mv[0][3][1] << 1; // dmv11
2e7b4c84 227
4ec58e13 228 } else {
ba585726
DB
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
2e7b4c84
IK
231 }
232 break;
233 default:
234 assert(0);
235 }
236
237 mv_block->motion_vertical_field_select = 0;
238
1f7c1d14 239 // set correct field references
148302e7 240 if (s->mv_type == MV_TYPE_FIELD || s->mv_type == MV_TYPE_16X8) {
90509ec7 241 mv_block->motion_vertical_field_select |= s->field_select[0][0];
7e2e870e
DB
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;
2e7b4c84 245 }
1f7c1d14 246 } // !intra
a1db3b93 247 // time to handle data blocks
2e7b4c84 248 mv_block->index = render->next_free_data_block_num;
5e5c247a 249
2e7b4c84 250 blocks_per_mb = 6;
148302e7 251 if (s->chroma_format >= 2) {
4742762f 252 blocks_per_mb = 4 + (1 << s->chroma_format);
2e7b4c84 253 }
5e5c247a 254
1f7c1d14 255 // calculate cbp
715731a3 256 cbp = 0;
148302e7 257 for (i = 0; i < blocks_per_mb; i++) {
2d7d8cc7 258 cbp += cbp;
148302e7 259 if (s->block_last_index[i] >= 0)
715731a3
IK
260 cbp++;
261 }
115329f1 262
148302e7 263 if (s->flags & CODEC_FLAG_GRAY) {
ba585726 264 if (s->mb_intra) { // intra frames are always full chroma blocks
148302e7 265 for (i = 4; i < blocks_per_mb; i++) {
6e4b41f8 266 memset(s->pblocks[i], 0, sizeof(short)*64); // so we need to clear them
148302e7
DB
267 if (!render->unsigned_intra)
268 s->pblocks[i][0] = 1 << 10;
a579db0c 269 }
4ec58e13 270 } else {
148302e7 271 cbp &= 0xf << (blocks_per_mb - 4);
ba585726 272 blocks_per_mb = 4; // luminance blocks only
715731a3 273 }
a579db0c 274 }
2e7b4c84 275 mv_block->coded_block_pattern = cbp;
148302e7 276 if (cbp == 0)
2e7b4c84
IK
277 mv_block->macroblock_type &= ~XVMC_MB_TYPE_PATTERN;
278
148302e7
DB
279 for (i = 0; i < blocks_per_mb; i++) {
280 if (s->block_last_index[i] >= 0) {
7a74e067 281 // I do not have unsigned_intra MOCO to test, hope it is OK.
eba9cecc 282 if (s->mb_intra && (render->idct || (!render->idct && !render->unsigned_intra)))
148302e7
DB
283 s->pblocks[i][0] -= 1 << 10;
284 if (!render->idct) {
a579db0c 285 s->dsp.idct(s->pblocks[i]);
a1db3b93
DB
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. */
2e7b4c84 290 }
1f7c1d14 291 // copy blocks only if the codec doesn't support pblocks reordering
148302e7 292 if (s->avctx->xvmc_acceleration == 1) {
4742762f 293 memcpy(&render->data_blocks[render->next_free_data_block_num*64],
6e280340 294 s->pblocks[i], sizeof(short)*64);
a579db0c
IK
295 }
296 render->next_free_data_block_num++;
2e7b4c84
IK
297 }
298 }
299 render->filled_mv_blocks_num++;
300
1a90cf3c 301 assert(render->filled_mv_blocks_num <= render->total_number_of_mv_blocks);
2e7b4c84 302 assert(render->next_free_data_block_num <= render->total_number_of_data_blocks);
1a90cf3c 303 /* The above conditions should not be able to fail as long as this function
930efaf4
DB
304 * is used and the following 'if ()' automatically calls a callback to free
305 * blocks. */
2e7b4c84
IK
306
307
afb9342a 308 if (render->filled_mv_blocks_num == render->total_number_of_mv_blocks)
7e2e870e 309 ff_draw_horiz_band(s, 0, 0);
2e7b4c84 310}