mirror of
https://git.ffmpeg.org/ffmpeg.git
synced 2024-12-28 02:12:28 +00:00
7bd3b73716
This already fixes a race in the vp9-encparams test. In this test, side data is added to the current frame after having been decoded (and therefore after ff_thread_finish_setup() has been called). Yet the update_thread_context callback called ff_thread_ref_frame() and therefore av_frame_ref() with this frame as source frame and the ensuing read was unsynchronised with adding the side data, i.e. there was a data race. By switching to the ProgressFrame API the implicit av_frame_ref() is removed and the race fixed except if this frame is later reused by a show-existing-frame which uses an explicit av_frame_ref(). The vp9-encparams test does not cover this, so this commit already fixes all the races in this test. This decoder kept multiple references to the same ThreadFrames in the same context and therefore had lots of implicit av_frame_ref() even when decoding single-threaded. This incurred lots of small allocations: When decoding an ordinary 10s video in single-threaded mode the number of allocations reported by Valgrind went down from 57,814 to 20,908; for 10 threads it went down from 84,223 to 21,901. Reviewed-by: Anton Khirnov <anton@khirnov.net> Signed-off-by: Andreas Rheinhardt <andreas.rheinhardt@outlook.com>
365 lines
15 KiB
C
365 lines
15 KiB
C
/*
|
|
* VP9 compatible video decoder
|
|
*
|
|
* Copyright (C) 2013 Ronald S. Bultje <rsbultje gmail com>
|
|
* Copyright (C) 2013 Clément Bœsch <u pkh me>
|
|
*
|
|
* This file is part of FFmpeg.
|
|
*
|
|
* FFmpeg is free software; you can redistribute it and/or
|
|
* modify it under the terms of the GNU Lesser General Public
|
|
* License as published by the Free Software Foundation; either
|
|
* version 2.1 of the License, or (at your option) any later version.
|
|
*
|
|
* FFmpeg is distributed in the hope that it will be useful,
|
|
* 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
|
|
* License along with FFmpeg; if not, write to the Free Software
|
|
* Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
|
|
*/
|
|
|
|
#include "progressframe.h"
|
|
#include "vp89_rac.h"
|
|
#include "vp9data.h"
|
|
#include "vp9dec.h"
|
|
#include "vpx_rac.h"
|
|
|
|
static av_always_inline void clamp_mv(VP9mv *dst, const VP9mv *src,
|
|
VP9TileData *td)
|
|
{
|
|
dst->x = av_clip(src->x, td->min_mv.x, td->max_mv.x);
|
|
dst->y = av_clip(src->y, td->min_mv.y, td->max_mv.y);
|
|
}
|
|
|
|
static void find_ref_mvs(VP9TileData *td,
|
|
VP9mv *pmv, int ref, int z, int idx, int sb)
|
|
{
|
|
static const int8_t mv_ref_blk_off[N_BS_SIZES][8][2] = {
|
|
[BS_64x64] = { { 3, -1 }, { -1, 3 }, { 4, -1 }, { -1, 4 },
|
|
{ -1, -1 }, { 0, -1 }, { -1, 0 }, { 6, -1 } },
|
|
[BS_64x32] = { { 0, -1 }, { -1, 0 }, { 4, -1 }, { -1, 2 },
|
|
{ -1, -1 }, { 0, -3 }, { -3, 0 }, { 2, -1 } },
|
|
[BS_32x64] = { { -1, 0 }, { 0, -1 }, { -1, 4 }, { 2, -1 },
|
|
{ -1, -1 }, { -3, 0 }, { 0, -3 }, { -1, 2 } },
|
|
[BS_32x32] = { { 1, -1 }, { -1, 1 }, { 2, -1 }, { -1, 2 },
|
|
{ -1, -1 }, { 0, -3 }, { -3, 0 }, { -3, -3 } },
|
|
[BS_32x16] = { { 0, -1 }, { -1, 0 }, { 2, -1 }, { -1, -1 },
|
|
{ -1, 1 }, { 0, -3 }, { -3, 0 }, { -3, -3 } },
|
|
[BS_16x32] = { { -1, 0 }, { 0, -1 }, { -1, 2 }, { -1, -1 },
|
|
{ 1, -1 }, { -3, 0 }, { 0, -3 }, { -3, -3 } },
|
|
[BS_16x16] = { { 0, -1 }, { -1, 0 }, { 1, -1 }, { -1, 1 },
|
|
{ -1, -1 }, { 0, -3 }, { -3, 0 }, { -3, -3 } },
|
|
[BS_16x8] = { { 0, -1 }, { -1, 0 }, { 1, -1 }, { -1, -1 },
|
|
{ 0, -2 }, { -2, 0 }, { -2, -1 }, { -1, -2 } },
|
|
[BS_8x16] = { { -1, 0 }, { 0, -1 }, { -1, 1 }, { -1, -1 },
|
|
{ -2, 0 }, { 0, -2 }, { -1, -2 }, { -2, -1 } },
|
|
[BS_8x8] = { { 0, -1 }, { -1, 0 }, { -1, -1 }, { 0, -2 },
|
|
{ -2, 0 }, { -1, -2 }, { -2, -1 }, { -2, -2 } },
|
|
[BS_8x4] = { { 0, -1 }, { -1, 0 }, { -1, -1 }, { 0, -2 },
|
|
{ -2, 0 }, { -1, -2 }, { -2, -1 }, { -2, -2 } },
|
|
[BS_4x8] = { { 0, -1 }, { -1, 0 }, { -1, -1 }, { 0, -2 },
|
|
{ -2, 0 }, { -1, -2 }, { -2, -1 }, { -2, -2 } },
|
|
[BS_4x4] = { { 0, -1 }, { -1, 0 }, { -1, -1 }, { 0, -2 },
|
|
{ -2, 0 }, { -1, -2 }, { -2, -1 }, { -2, -2 } },
|
|
};
|
|
const VP9Context *s = td->s;
|
|
VP9Block *b = td->b;
|
|
int row = td->row, col = td->col, row7 = td->row7;
|
|
const int8_t (*p)[2] = mv_ref_blk_off[b->bs];
|
|
#define INVALID_MV 0x80008000U
|
|
uint32_t mem = INVALID_MV, mem_sub8x8 = INVALID_MV;
|
|
int i;
|
|
|
|
#define RETURN_DIRECT_MV(mv) \
|
|
do { \
|
|
uint32_t m = AV_RN32A(&mv); \
|
|
if (!idx) { \
|
|
AV_WN32A(pmv, m); \
|
|
return; \
|
|
} else if (mem == INVALID_MV) { \
|
|
mem = m; \
|
|
} else if (m != mem) { \
|
|
AV_WN32A(pmv, m); \
|
|
return; \
|
|
} \
|
|
} while (0)
|
|
|
|
if (sb >= 0) {
|
|
if (sb == 2 || sb == 1) {
|
|
RETURN_DIRECT_MV(b->mv[0][z]);
|
|
} else if (sb == 3) {
|
|
RETURN_DIRECT_MV(b->mv[2][z]);
|
|
RETURN_DIRECT_MV(b->mv[1][z]);
|
|
RETURN_DIRECT_MV(b->mv[0][z]);
|
|
}
|
|
|
|
#define RETURN_MV(mv) \
|
|
do { \
|
|
if (sb > 0) { \
|
|
VP9mv tmp; \
|
|
uint32_t m; \
|
|
av_assert2(idx == 1); \
|
|
av_assert2(mem != INVALID_MV); \
|
|
if (mem_sub8x8 == INVALID_MV) { \
|
|
clamp_mv(&tmp, &mv, td); \
|
|
m = AV_RN32A(&tmp); \
|
|
if (m != mem) { \
|
|
AV_WN32A(pmv, m); \
|
|
return; \
|
|
} \
|
|
mem_sub8x8 = AV_RN32A(&mv); \
|
|
} else if (mem_sub8x8 != AV_RN32A(&mv)) { \
|
|
clamp_mv(&tmp, &mv, td); \
|
|
m = AV_RN32A(&tmp); \
|
|
if (m != mem) { \
|
|
AV_WN32A(pmv, m); \
|
|
} else { \
|
|
/* BUG I'm pretty sure this isn't the intention */ \
|
|
AV_WN32A(pmv, 0); \
|
|
} \
|
|
return; \
|
|
} \
|
|
} else { \
|
|
uint32_t m = AV_RN32A(&mv); \
|
|
if (!idx) { \
|
|
clamp_mv(pmv, &mv, td); \
|
|
return; \
|
|
} else if (mem == INVALID_MV) { \
|
|
mem = m; \
|
|
} else if (m != mem) { \
|
|
clamp_mv(pmv, &mv, td); \
|
|
return; \
|
|
} \
|
|
} \
|
|
} while (0)
|
|
|
|
if (row > 0) {
|
|
VP9mvrefPair *mv = &s->s.frames[CUR_FRAME].mv[(row - 1) * s->sb_cols * 8 + col];
|
|
if (mv->ref[0] == ref)
|
|
RETURN_MV(s->above_mv_ctx[2 * col + (sb & 1)][0]);
|
|
else if (mv->ref[1] == ref)
|
|
RETURN_MV(s->above_mv_ctx[2 * col + (sb & 1)][1]);
|
|
}
|
|
if (col > td->tile_col_start) {
|
|
VP9mvrefPair *mv = &s->s.frames[CUR_FRAME].mv[row * s->sb_cols * 8 + col - 1];
|
|
if (mv->ref[0] == ref)
|
|
RETURN_MV(td->left_mv_ctx[2 * row7 + (sb >> 1)][0]);
|
|
else if (mv->ref[1] == ref)
|
|
RETURN_MV(td->left_mv_ctx[2 * row7 + (sb >> 1)][1]);
|
|
}
|
|
i = 2;
|
|
} else {
|
|
i = 0;
|
|
}
|
|
|
|
// previously coded MVs in this neighborhood, using same reference frame
|
|
for (; i < 8; i++) {
|
|
int c = p[i][0] + col, r = p[i][1] + row;
|
|
|
|
if (c >= td->tile_col_start && c < s->cols &&
|
|
r >= 0 && r < s->rows) {
|
|
VP9mvrefPair *mv = &s->s.frames[CUR_FRAME].mv[r * s->sb_cols * 8 + c];
|
|
|
|
if (mv->ref[0] == ref)
|
|
RETURN_MV(mv->mv[0]);
|
|
else if (mv->ref[1] == ref)
|
|
RETURN_MV(mv->mv[1]);
|
|
}
|
|
}
|
|
|
|
// MV at this position in previous frame, using same reference frame
|
|
if (s->s.h.use_last_frame_mvs) {
|
|
VP9mvrefPair *mv = &s->s.frames[REF_FRAME_MVPAIR].mv[row * s->sb_cols * 8 + col];
|
|
|
|
if (!s->s.frames[REF_FRAME_MVPAIR].uses_2pass)
|
|
ff_progress_frame_await(&s->s.frames[REF_FRAME_MVPAIR].tf, row >> 3);
|
|
if (mv->ref[0] == ref)
|
|
RETURN_MV(mv->mv[0]);
|
|
else if (mv->ref[1] == ref)
|
|
RETURN_MV(mv->mv[1]);
|
|
}
|
|
|
|
#define RETURN_SCALE_MV(mv, scale) \
|
|
do { \
|
|
if (scale) { \
|
|
VP9mv mv_temp = { -mv.x, -mv.y }; \
|
|
RETURN_MV(mv_temp); \
|
|
} else { \
|
|
RETURN_MV(mv); \
|
|
} \
|
|
} while (0)
|
|
|
|
// previously coded MVs in this neighborhood, using different reference frame
|
|
for (i = 0; i < 8; i++) {
|
|
int c = p[i][0] + col, r = p[i][1] + row;
|
|
|
|
if (c >= td->tile_col_start && c < s->cols && r >= 0 && r < s->rows) {
|
|
VP9mvrefPair *mv = &s->s.frames[CUR_FRAME].mv[r * s->sb_cols * 8 + c];
|
|
|
|
if (mv->ref[0] != ref && mv->ref[0] >= 0)
|
|
RETURN_SCALE_MV(mv->mv[0],
|
|
s->s.h.signbias[mv->ref[0]] != s->s.h.signbias[ref]);
|
|
if (mv->ref[1] != ref && mv->ref[1] >= 0 &&
|
|
// BUG - libvpx has this condition regardless of whether
|
|
// we used the first ref MV and pre-scaling
|
|
AV_RN32A(&mv->mv[0]) != AV_RN32A(&mv->mv[1])) {
|
|
RETURN_SCALE_MV(mv->mv[1], s->s.h.signbias[mv->ref[1]] != s->s.h.signbias[ref]);
|
|
}
|
|
}
|
|
}
|
|
|
|
// MV at this position in previous frame, using different reference frame
|
|
if (s->s.h.use_last_frame_mvs) {
|
|
VP9mvrefPair *mv = &s->s.frames[REF_FRAME_MVPAIR].mv[row * s->sb_cols * 8 + col];
|
|
|
|
// no need to await_progress, because we already did that above
|
|
if (mv->ref[0] != ref && mv->ref[0] >= 0)
|
|
RETURN_SCALE_MV(mv->mv[0], s->s.h.signbias[mv->ref[0]] != s->s.h.signbias[ref]);
|
|
if (mv->ref[1] != ref && mv->ref[1] >= 0 &&
|
|
// BUG - libvpx has this condition regardless of whether
|
|
// we used the first ref MV and pre-scaling
|
|
AV_RN32A(&mv->mv[0]) != AV_RN32A(&mv->mv[1])) {
|
|
RETURN_SCALE_MV(mv->mv[1], s->s.h.signbias[mv->ref[1]] != s->s.h.signbias[ref]);
|
|
}
|
|
}
|
|
|
|
AV_ZERO32(pmv);
|
|
clamp_mv(pmv, pmv, td);
|
|
#undef INVALID_MV
|
|
#undef RETURN_MV
|
|
#undef RETURN_SCALE_MV
|
|
}
|
|
|
|
static av_always_inline int read_mv_component(VP9TileData *td, int idx, int hp)
|
|
{
|
|
const VP9Context *s = td->s;
|
|
int bit, sign = vpx_rac_get_prob(td->c, s->prob.p.mv_comp[idx].sign);
|
|
int n, c = vp89_rac_get_tree(td->c, ff_vp9_mv_class_tree,
|
|
s->prob.p.mv_comp[idx].classes);
|
|
|
|
td->counts.mv_comp[idx].sign[sign]++;
|
|
td->counts.mv_comp[idx].classes[c]++;
|
|
if (c) {
|
|
int m;
|
|
|
|
for (n = 0, m = 0; m < c; m++) {
|
|
bit = vpx_rac_get_prob(td->c, s->prob.p.mv_comp[idx].bits[m]);
|
|
n |= bit << m;
|
|
td->counts.mv_comp[idx].bits[m][bit]++;
|
|
}
|
|
n <<= 3;
|
|
bit = vp89_rac_get_tree(td->c, ff_vp9_mv_fp_tree,
|
|
s->prob.p.mv_comp[idx].fp);
|
|
n |= bit << 1;
|
|
td->counts.mv_comp[idx].fp[bit]++;
|
|
if (hp) {
|
|
bit = vpx_rac_get_prob(td->c, s->prob.p.mv_comp[idx].hp);
|
|
td->counts.mv_comp[idx].hp[bit]++;
|
|
n |= bit;
|
|
} else {
|
|
n |= 1;
|
|
// bug in libvpx - we count for bw entropy purposes even if the
|
|
// bit wasn't coded
|
|
td->counts.mv_comp[idx].hp[1]++;
|
|
}
|
|
n += 8 << c;
|
|
} else {
|
|
n = vpx_rac_get_prob(td->c, s->prob.p.mv_comp[idx].class0);
|
|
td->counts.mv_comp[idx].class0[n]++;
|
|
bit = vp89_rac_get_tree(td->c, ff_vp9_mv_fp_tree,
|
|
s->prob.p.mv_comp[idx].class0_fp[n]);
|
|
td->counts.mv_comp[idx].class0_fp[n][bit]++;
|
|
n = (n << 3) | (bit << 1);
|
|
if (hp) {
|
|
bit = vpx_rac_get_prob(td->c, s->prob.p.mv_comp[idx].class0_hp);
|
|
td->counts.mv_comp[idx].class0_hp[bit]++;
|
|
n |= bit;
|
|
} else {
|
|
n |= 1;
|
|
// bug in libvpx - we count for bw entropy purposes even if the
|
|
// bit wasn't coded
|
|
td->counts.mv_comp[idx].class0_hp[1]++;
|
|
}
|
|
}
|
|
|
|
return sign ? -(n + 1) : (n + 1);
|
|
}
|
|
|
|
void ff_vp9_fill_mv(VP9TileData *td, VP9mv *mv, int mode, int sb)
|
|
{
|
|
const VP9Context *s = td->s;
|
|
VP9Block *b = td->b;
|
|
|
|
if (mode == ZEROMV) {
|
|
AV_ZERO64(mv);
|
|
} else {
|
|
int hp;
|
|
|
|
// FIXME cache this value and reuse for other subblocks
|
|
find_ref_mvs(td, &mv[0], b->ref[0], 0, mode == NEARMV,
|
|
mode == NEWMV ? -1 : sb);
|
|
// FIXME maybe move this code into find_ref_mvs()
|
|
if ((mode == NEWMV || sb == -1) &&
|
|
!(hp = s->s.h.highprecisionmvs &&
|
|
abs(mv[0].x) < 64 && abs(mv[0].y) < 64)) {
|
|
if (mv[0].y & 1) {
|
|
if (mv[0].y < 0)
|
|
mv[0].y++;
|
|
else
|
|
mv[0].y--;
|
|
}
|
|
if (mv[0].x & 1) {
|
|
if (mv[0].x < 0)
|
|
mv[0].x++;
|
|
else
|
|
mv[0].x--;
|
|
}
|
|
}
|
|
if (mode == NEWMV) {
|
|
enum MVJoint j = vp89_rac_get_tree(td->c, ff_vp9_mv_joint_tree,
|
|
s->prob.p.mv_joint);
|
|
|
|
td->counts.mv_joint[j]++;
|
|
if (j >= MV_JOINT_V)
|
|
mv[0].y += read_mv_component(td, 0, hp);
|
|
if (j & 1)
|
|
mv[0].x += read_mv_component(td, 1, hp);
|
|
}
|
|
|
|
if (b->comp) {
|
|
// FIXME cache this value and reuse for other subblocks
|
|
find_ref_mvs(td, &mv[1], b->ref[1], 1, mode == NEARMV,
|
|
mode == NEWMV ? -1 : sb);
|
|
if ((mode == NEWMV || sb == -1) &&
|
|
!(hp = s->s.h.highprecisionmvs &&
|
|
abs(mv[1].x) < 64 && abs(mv[1].y) < 64)) {
|
|
if (mv[1].y & 1) {
|
|
if (mv[1].y < 0)
|
|
mv[1].y++;
|
|
else
|
|
mv[1].y--;
|
|
}
|
|
if (mv[1].x & 1) {
|
|
if (mv[1].x < 0)
|
|
mv[1].x++;
|
|
else
|
|
mv[1].x--;
|
|
}
|
|
}
|
|
if (mode == NEWMV) {
|
|
enum MVJoint j = vp89_rac_get_tree(td->c, ff_vp9_mv_joint_tree,
|
|
s->prob.p.mv_joint);
|
|
|
|
td->counts.mv_joint[j]++;
|
|
if (j >= MV_JOINT_V)
|
|
mv[1].y += read_mv_component(td, 0, hp);
|
|
if (j & 1)
|
|
mv[1].x += read_mv_component(td, 1, hp);
|
|
}
|
|
}
|
|
}
|
|
}
|