302898fc |
/*
* XVideo Motion Compensation
* Copyright (c) 2003 Ivan Kalvachev
* |
b78e7197 |
* This file is part of FFmpeg.
*
* FFmpeg is free software; you can redistribute it and/or |
302898fc |
* modify it under the terms of the GNU Lesser General Public
* License as published by the Free Software Foundation; either |
b78e7197 |
* version 2.1 of the License, or (at your option) any later version. |
302898fc |
* |
b78e7197 |
* FFmpeg is distributed in the hope that it will be useful, |
302898fc |
* but WITHOUT ANY WARRANTY; without even the implied warranty of
* MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
* Lesser General Public License for more details.
*
* You should have received a copy of the GNU Lesser General Public |
b78e7197 |
* License along with FFmpeg; if not, write to the Free Software |
5509bffa |
* Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA |
302898fc |
*/
|
2e7b4c84 |
#include <limits.h> |
2213f431 |
#include <X11/extensions/XvMC.h> |
2e7b4c84 |
#include "avcodec.h" |
e0c16e4e |
#include "mpegutils.h" |
2e7b4c84 |
#include "mpegvideo.h"
#undef NDEBUG
#include <assert.h>
|
fd949a63 |
#include "xvmc.h" |
4440bd0d |
#include "xvmc_internal.h" |
19e30a58 |
#include "version.h"
|
8f38ff00 |
/** |
49bd8e4b |
* Initialize the block field of the MpegEncContext pointer passed as |
8f38ff00 |
* parameter after making sure that the data is not corrupted. |
d758a5cb |
* In order to implement something like direct rendering instead of decoding
* coefficients in s->blocks and then copying them, copy them directly |
3d43e42f |
* into the data_blocks array provided by xvmc. |
8f38ff00 |
*/ |
78f9a878 |
void ff_xvmc_init_block(MpegEncContext *s) |
148302e7 |
{ |
f6774f90 |
struct xvmc_pix_fmt *render = (struct xvmc_pix_fmt*)s->current_picture.f->data[2]; |
4e30fa6e |
assert(render && render->xvmc_id == AV_XVMC_ID); |
d2d600b7 |
|
88bd7fdc |
s->block = (int16_t (*)[64])(render->data_blocks + render->next_free_data_block_num * 64); |
a579db0c |
}
|
1c63aed2 |
static void exchange_uv(MpegEncContext *s)
{
int16_t (*tmp)[64];
tmp = s->pblocks[4];
s->pblocks[4] = s->pblocks[5];
s->pblocks[5] = tmp;
}
|
3d43e42f |
/** |
49bd8e4b |
* Fill individual block pointers, so there are no gaps in the data_block array |
63864cb8 |
* in case not all blocks in the macroblock are coded. |
3d43e42f |
*/ |
78f9a878 |
void ff_xvmc_pack_pblocks(MpegEncContext *s, int cbp) |
148302e7 |
{ |
b71bd0ed |
int i, j = 0; |
148302e7 |
const int mb_block_count = 4 + (1 << s->chroma_format); |
a579db0c |
|
148302e7 |
cbp <<= 12-mb_block_count;
for (i = 0; i < mb_block_count; i++) { |
4ec58e13 |
if (cbp & (1 << 11)) |
21effaa4 |
s->pblocks[i] = &s->block[j++]; |
4ec58e13 |
else |
39dba5e8 |
s->pblocks[i] = NULL; |
42dfc2bf |
cbp += cbp; |
a579db0c |
} |
1c63aed2 |
if (s->swap_uv) {
exchange_uv(s);
} |
2e7b4c84 |
}
|
9107f7cd |
/** |
49bd8e4b |
* Find and store the surfaces that are used as reference frames. |
9107f7cd |
* This function should be called for every new field and/or frame.
* It should be safe to call the function a few times for the same field.
*/ |
1c63aed2 |
static int ff_xvmc_field_start(AVCodecContext *avctx, const uint8_t *buf, uint32_t buf_size) |
148302e7 |
{ |
1c63aed2 |
struct MpegEncContext *s = avctx->priv_data; |
f6774f90 |
struct xvmc_pix_fmt *last, *next, *render = (struct xvmc_pix_fmt*)s->current_picture.f->data[2]; |
d2d600b7 |
const int mb_block_count = 4 + (1 << s->chroma_format); |
2e7b4c84 |
|
3ae01928 |
assert(avctx); |
4e30fa6e |
if (!render || render->xvmc_id != AV_XVMC_ID || |
859bb3cf |
!render->data_blocks || !render->mv_blocks ||
(unsigned int)render->allocated_mv_blocks > INT_MAX/(64*6) ||
(unsigned int)render->allocated_data_blocks > INT_MAX/64 ||
!render->p_surface) { |
d2d600b7 |
av_log(avctx, AV_LOG_ERROR,
"Render token doesn't look as expected.\n"); |
a1db3b93 |
return -1; // make sure that this is a render packet |
d2d600b7 |
} |
2e7b4c84 |
|
cb18fb62 |
if (render->filled_mv_blocks_num) { |
9939841f |
av_log(avctx, AV_LOG_ERROR, |
d2d600b7 |
"Rendering surface contains %i unprocessed blocks.\n", |
cb18fb62 |
render->filled_mv_blocks_num); |
013cebfb |
return -1; |
9939841f |
} |
cad1928f |
if (render->allocated_mv_blocks < 1 || |
649c171a |
render->allocated_data_blocks < render->allocated_mv_blocks*mb_block_count ||
render->start_mv_blocks_num >= render->allocated_mv_blocks ||
render->next_free_data_block_num >
render->allocated_data_blocks -
mb_block_count*(render->allocated_mv_blocks-render->start_mv_blocks_num)) { |
d2d600b7 |
av_log(avctx, AV_LOG_ERROR,
"Rendering surface doesn't provide enough block structures to work with.\n");
return -1;
} |
2e7b4c84 |
|
d76c5ed5 |
render->picture_structure = s->picture_structure;
render->flags = s->first_field ? 0 : XVMC_SECOND_FIELD; |
9e0a8a36 |
render->p_future_surface = NULL;
render->p_past_surface = NULL; |
2e7b4c84 |
|
7e2e870e |
switch(s->pict_type) { |
ce5e49b0 |
case AV_PICTURE_TYPE_I: |
1f7c1d14 |
return 0; // no prediction from other frames |
ce5e49b0 |
case AV_PICTURE_TYPE_B: |
f6774f90 |
next = (struct xvmc_pix_fmt*)s->next_picture.f->data[2]; |
3ae01928 |
if (!next) |
148302e7 |
return -1; |
4e30fa6e |
if (next->xvmc_id != AV_XVMC_ID) |
148302e7 |
return -1; |
2e7b4c84 |
render->p_future_surface = next->p_surface; |
1f7c1d14 |
// no return here, going to set forward prediction |
ce5e49b0 |
case AV_PICTURE_TYPE_P: |
f6774f90 |
last = (struct xvmc_pix_fmt*)s->last_picture.f->data[2]; |
63a21bc0 |
if (!last) |
1f7c1d14 |
last = render; // predict second field from the first |
4e30fa6e |
if (last->xvmc_id != AV_XVMC_ID) |
148302e7 |
return -1; |
2e7b4c84 |
render->p_past_surface = last->p_surface;
return 0; |
5e5c247a |
} |
2e7b4c84 |
return -1;
}
|
0b2eb2e8 |
/** |
49bd8e4b |
* Complete frame/field rendering by passing any remaining blocks. |
d758a5cb |
* Normally ff_draw_horiz_band() is called for each slice, however,
* some leftover blocks, for example from error_resilience(), may remain. |
0b2eb2e8 |
* It should be safe to call the function a few times for the same field.
*/ |
1c63aed2 |
static int ff_xvmc_field_end(AVCodecContext *avctx) |
148302e7 |
{ |
1c63aed2 |
struct MpegEncContext *s = avctx->priv_data; |
f6774f90 |
struct xvmc_pix_fmt *render = (struct xvmc_pix_fmt*)s->current_picture.f->data[2]; |
3ae01928 |
assert(render); |
2e7b4c84 |
|
148302e7 |
if (render->filled_mv_blocks_num > 0) |
1d0feb5d |
ff_mpeg_draw_horiz_band(s, 0, 0); |
1c63aed2 |
return 0; |
2e7b4c84 |
}
|
3d43e42f |
/** |
49bd8e4b |
* Synthesize the data needed by XvMC to render one macroblock of data.
* Fill all relevant fields, if necessary do IDCT. |
3d43e42f |
*/ |
1c63aed2 |
static void ff_xvmc_decode_mb(struct MpegEncContext *s) |
148302e7 |
{ |
7e2e870e |
XvMCMacroBlock *mv_block; |
f602b044 |
struct xvmc_pix_fmt *render; |
7e2e870e |
int i, cbp, blocks_per_mb; |
2e7b4c84 |
|
39dba5e8 |
const int mb_xy = s->mb_y * s->mb_stride + s->mb_x; |
2e7b4c84 |
|
148302e7 |
if (s->encoding) { |
8ac5c1b2 |
av_log(s->avctx, AV_LOG_ERROR, "XVMC doesn't support encoding!!!\n"); |
61b76987 |
return; |
2e7b4c84 |
}
|
835f798c |
// from ff_mpv_decode_mb(), update DC predictors for P macroblocks |
2e7b4c84 |
if (!s->mb_intra) {
s->last_dc[0] =
s->last_dc[1] =
s->last_dc[2] = 128 << s->intra_dc_precision;
}
|
1f7c1d14 |
// MC doesn't skip blocks |
160d679c |
s->mb_skipped = 0; |
2e7b4c84 |
|
d15876d3 |
// Do I need to export quant when I could not perform postprocessing?
// Anyway, it doesn't hurt. |
759001c5 |
s->current_picture.qscale_table[mb_xy] = s->qscale; |
2e7b4c84 |
|
a1db3b93 |
// start of XVMC-specific code |
f6774f90 |
render = (struct xvmc_pix_fmt*)s->current_picture.f->data[2]; |
3ae01928 |
assert(render); |
4e30fa6e |
assert(render->xvmc_id == AV_XVMC_ID); |
2e7b4c84 |
assert(render->mv_blocks); |
302898fc |
|
1f7c1d14 |
// take the next free macroblock |
115329f1 |
mv_block = &render->mv_blocks[render->start_mv_blocks_num + |
fe4be5db |
render->filled_mv_blocks_num]; |
2e7b4c84 |
|
4ef690f5 |
mv_block->x = s->mb_x;
mv_block->y = s->mb_y; |
1f7c1d14 |
mv_block->dct_type = s->interlaced_dct; // XVMC_DCT_TYPE_FRAME/FIELD; |
4ec58e13 |
if (s->mb_intra) { |
1f7c1d14 |
mv_block->macroblock_type = XVMC_MB_TYPE_INTRA; // no MC, all done |
4ec58e13 |
} else { |
2e7b4c84 |
mv_block->macroblock_type = XVMC_MB_TYPE_PATTERN;
|
148302e7 |
if (s->mv_dir & MV_DIR_FORWARD) { |
2d7d8cc7 |
mv_block->macroblock_type |= XVMC_MB_TYPE_MOTION_FORWARD; |
0439c09d |
// PMV[n][dir][xy] = mv[dir][n][xy] |
2e7b4c84 |
mv_block->PMV[0][0][0] = s->mv[0][0][0];
mv_block->PMV[0][0][1] = s->mv[0][0][1];
mv_block->PMV[1][0][0] = s->mv[0][1][0];
mv_block->PMV[1][0][1] = s->mv[0][1][1];
} |
148302e7 |
if (s->mv_dir & MV_DIR_BACKWARD) { |
2d7d8cc7 |
mv_block->macroblock_type |= XVMC_MB_TYPE_MOTION_BACKWARD; |
2e7b4c84 |
mv_block->PMV[0][1][0] = s->mv[1][0][0];
mv_block->PMV[0][1][1] = s->mv[1][0][1];
mv_block->PMV[1][1][0] = s->mv[1][1][0];
mv_block->PMV[1][1][1] = s->mv[1][1][1];
}
|
99df0aa5 |
switch(s->mv_type) { |
2e7b4c84 |
case MV_TYPE_16X16:
mv_block->motion_type = XVMC_PREDICTION_FRAME;
break;
case MV_TYPE_16X8:
mv_block->motion_type = XVMC_PREDICTION_16x8;
break;
case MV_TYPE_FIELD:
mv_block->motion_type = XVMC_PREDICTION_FIELD; |
148302e7 |
if (s->picture_structure == PICT_FRAME) {
mv_block->PMV[0][0][1] <<= 1;
mv_block->PMV[1][0][1] <<= 1;
mv_block->PMV[0][1][1] <<= 1;
mv_block->PMV[1][1][1] <<= 1; |
2e7b4c84 |
}
break;
case MV_TYPE_DMV:
mv_block->motion_type = XVMC_PREDICTION_DUAL_PRIME; |
148302e7 |
if (s->picture_structure == PICT_FRAME) { |
2e7b4c84 |
|
ba585726 |
mv_block->PMV[0][0][0] = s->mv[0][0][0]; // top from top |
81189e4f |
mv_block->PMV[0][0][1] = s->mv[0][0][1] << 1; |
2e7b4c84 |
|
ba585726 |
mv_block->PMV[0][1][0] = s->mv[0][0][0]; // bottom from bottom |
81189e4f |
mv_block->PMV[0][1][1] = s->mv[0][0][1] << 1; |
2e7b4c84 |
|
ba585726 |
mv_block->PMV[1][0][0] = s->mv[0][2][0]; // dmv00, top from bottom |
81189e4f |
mv_block->PMV[1][0][1] = s->mv[0][2][1] << 1; // dmv01 |
2e7b4c84 |
|
ba585726 |
mv_block->PMV[1][1][0] = s->mv[0][3][0]; // dmv10, bottom from top |
81189e4f |
mv_block->PMV[1][1][1] = s->mv[0][3][1] << 1; // dmv11 |
2e7b4c84 |
|
4ec58e13 |
} else { |
ba585726 |
mv_block->PMV[0][1][0] = s->mv[0][2][0]; // dmv00
mv_block->PMV[0][1][1] = s->mv[0][2][1]; // dmv01 |
2e7b4c84 |
}
break;
default:
assert(0);
}
mv_block->motion_vertical_field_select = 0;
|
1f7c1d14 |
// set correct field references |
148302e7 |
if (s->mv_type == MV_TYPE_FIELD || s->mv_type == MV_TYPE_16X8) { |
90509ec7 |
mv_block->motion_vertical_field_select |= s->field_select[0][0]; |
7e2e870e |
mv_block->motion_vertical_field_select |= s->field_select[1][0] << 1;
mv_block->motion_vertical_field_select |= s->field_select[0][1] << 2;
mv_block->motion_vertical_field_select |= s->field_select[1][1] << 3; |
2e7b4c84 |
} |
1f7c1d14 |
} // !intra |
a1db3b93 |
// time to handle data blocks |
2e7b4c84 |
mv_block->index = render->next_free_data_block_num; |
5e5c247a |
|
2e7b4c84 |
blocks_per_mb = 6; |
148302e7 |
if (s->chroma_format >= 2) { |
4742762f |
blocks_per_mb = 4 + (1 << s->chroma_format); |
2e7b4c84 |
} |
5e5c247a |
|
1f7c1d14 |
// calculate cbp |
715731a3 |
cbp = 0; |
148302e7 |
for (i = 0; i < blocks_per_mb; i++) { |
2d7d8cc7 |
cbp += cbp; |
148302e7 |
if (s->block_last_index[i] >= 0) |
715731a3 |
cbp++;
} |
115329f1 |
|
7c6eb0a1 |
if (s->avctx->flags & AV_CODEC_FLAG_GRAY) { |
ba585726 |
if (s->mb_intra) { // intra frames are always full chroma blocks |
148302e7 |
for (i = 4; i < blocks_per_mb; i++) { |
21effaa4 |
memset(s->pblocks[i], 0, sizeof(*s->pblocks[i])); // so we need to clear them |
148302e7 |
if (!render->unsigned_intra) |
21effaa4 |
*s->pblocks[i][0] = 1 << 10; |
a579db0c |
} |
4ec58e13 |
} else { |
148302e7 |
cbp &= 0xf << (blocks_per_mb - 4); |
ba585726 |
blocks_per_mb = 4; // luminance blocks only |
715731a3 |
} |
a579db0c |
} |
2e7b4c84 |
mv_block->coded_block_pattern = cbp; |
148302e7 |
if (cbp == 0) |
2e7b4c84 |
mv_block->macroblock_type &= ~XVMC_MB_TYPE_PATTERN;
|
148302e7 |
for (i = 0; i < blocks_per_mb; i++) {
if (s->block_last_index[i] >= 0) { |
7a74e067 |
// I do not have unsigned_intra MOCO to test, hope it is OK. |
fe655f67 |
if (s->mb_intra && (render->idct || !render->unsigned_intra)) |
21effaa4 |
*s->pblocks[i][0] -= 1 << 10; |
148302e7 |
if (!render->idct) { |
e3fcb143 |
s->idsp.idct(*s->pblocks[i]); |
a1db3b93 |
/* It is unclear if MC hardware requires pixel diff values to be
* in the range [-255;255]. TODO: Clipping if such hardware is
* ever found. As of now it would only be an unnecessary
* slowdown. */ |
2e7b4c84 |
} |
1f7c1d14 |
// copy blocks only if the codec doesn't support pblocks reordering |
1c63aed2 |
if (!s->pack_pblocks) { |
4742762f |
memcpy(&render->data_blocks[render->next_free_data_block_num*64], |
21effaa4 |
s->pblocks[i], sizeof(*s->pblocks[i])); |
a579db0c |
}
render->next_free_data_block_num++; |
2e7b4c84 |
}
}
render->filled_mv_blocks_num++;
|
bd0eddb7 |
assert(render->filled_mv_blocks_num <= render->allocated_mv_blocks);
assert(render->next_free_data_block_num <= render->allocated_data_blocks);
/* The above conditions should not be able to fail as long as this function
* is used and the following 'if ()' automatically calls a callback to free
* blocks. */ |
2e7b4c84 |
|
cad1928f |
if (render->filled_mv_blocks_num == render->allocated_mv_blocks) |
1d0feb5d |
ff_mpeg_draw_horiz_band(s, 0, 0); |
2e7b4c84 |
} |
19e30a58 |
|
1c63aed2 |
#if CONFIG_MPEG1_XVMC_HWACCEL
AVHWAccel ff_mpeg1_xvmc_hwaccel = {
.name = "mpeg1_xvmc",
.type = AVMEDIA_TYPE_VIDEO,
.id = AV_CODEC_ID_MPEG1VIDEO,
.pix_fmt = AV_PIX_FMT_XVMC,
.start_frame = ff_xvmc_field_start,
.end_frame = ff_xvmc_field_end,
.decode_slice = NULL,
.decode_mb = ff_xvmc_decode_mb,
.priv_data_size = 0,
};
#endif
#if CONFIG_MPEG2_XVMC_HWACCEL
AVHWAccel ff_mpeg2_xvmc_hwaccel = {
.name = "mpeg2_xvmc",
.type = AVMEDIA_TYPE_VIDEO,
.id = AV_CODEC_ID_MPEG2VIDEO,
.pix_fmt = AV_PIX_FMT_XVMC,
.start_frame = ff_xvmc_field_start,
.end_frame = ff_xvmc_field_end,
.decode_slice = NULL,
.decode_mb = ff_xvmc_decode_mb,
.priv_data_size = 0,
};
#endif |