1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
|
/*
* Copyright © 2018, VideoLAN and dav1d authors
* Copyright © 2018, Two Orioles, LLC
* All rights reserved.
*
* Redistribution and use in source and binary forms, with or without
* modification, are permitted provided that the following conditions are met:
*
* 1. Redistributions of source code must retain the above copyright notice, this
* list of conditions and the following disclaimer.
*
* 2. Redistributions in binary form must reproduce the above copyright notice,
* this list of conditions and the following disclaimer in the documentation
* and/or other materials provided with the distribution.
*
* THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS" AND
* ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED
* WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE
* DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER OR CONTRIBUTORS BE LIABLE FOR
* ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES
* (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES;
* LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND
* ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
* (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS
* SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
*/
#include "config.h"
#include "src/thread_task.h"
void *dav1d_frame_task(void *const data) {
Dav1dFrameContext *const f = data;
pthread_mutex_lock(&f->frame_thread.td.lock);
for (;;) {
while (!f->n_tile_data && !f->frame_thread.die) {
pthread_cond_wait(&f->frame_thread.td.cond,
&f->frame_thread.td.lock);
}
if (f->frame_thread.die) break;
pthread_mutex_unlock(&f->frame_thread.td.lock);
dav1d_decode_frame(f);
pthread_mutex_lock(&f->frame_thread.td.lock);
f->n_tile_data = 0;
pthread_cond_signal(&f->frame_thread.td.cond);
}
pthread_mutex_unlock(&f->frame_thread.td.lock);
return NULL;
}
void *dav1d_tile_task(void *const data) {
Dav1dTileContext *const t = data;
struct FrameTileThreadData *const fttd = t->tile_thread.fttd;
const Dav1dFrameContext *const f = t->f;
const int tile_thread_idx = t - f->tc;
const uint64_t mask = 1ULL << tile_thread_idx;
for (;;) {
pthread_mutex_lock(&fttd->lock);
fttd->available |= mask;
int did_signal = 0;
while (!fttd->tasks_left && !t->tile_thread.die) {
if (!did_signal) {
did_signal = 1;
pthread_cond_signal(&fttd->icond);
}
pthread_cond_wait(&fttd->cond, &fttd->lock);
}
if (t->tile_thread.die) {
pthread_mutex_unlock(&fttd->lock);
break;
}
fttd->available &= ~mask;
const int task_idx = fttd->num_tasks - fttd->tasks_left--;
pthread_mutex_unlock(&fttd->lock);
if (f->frame_thread.pass == 1 || f->n_tc >= f->frame_hdr.tiling.cols) {
// we can (or in fact, if >, we need to) do full tile decoding.
// loopfilter happens in the main thread
Dav1dTileState *const ts = t->ts = &f->ts[task_idx];
for (t->by = ts->tiling.row_start; t->by < ts->tiling.row_end;
t->by += f->sb_step)
{
dav1d_decode_tile_sbrow(t);
// signal progress
pthread_mutex_lock(&ts->tile_thread.lock);
atomic_store(&ts->progress, 1 + (t->by >> f->sb_shift));
pthread_cond_signal(&ts->tile_thread.cond);
pthread_mutex_unlock(&ts->tile_thread.lock);
}
} else {
const int sby = f->tile_thread.task_idx_to_sby_and_tile_idx[task_idx][0];
const int tile_idx = f->tile_thread.task_idx_to_sby_and_tile_idx[task_idx][1];
Dav1dTileState *const ts = &f->ts[tile_idx];
// the interleaved decoding can sometimes cause dependency issues
// if one part of the frame decodes signifcantly faster than others.
// Ideally, we'd "skip" tile_sbrows where dependencies are missing,
// and resume them later as dependencies are met. This also would
// solve the broadcast() below and allow us to use signal(). However,
// for now, we use linear dependency tracking because it's simpler.
if (atomic_load(&ts->progress) < sby) {
pthread_mutex_lock(&ts->tile_thread.lock);
while (atomic_load(&ts->progress) < sby)
pthread_cond_wait(&ts->tile_thread.cond,
&ts->tile_thread.lock);
pthread_mutex_unlock(&ts->tile_thread.lock);
}
// we need to interleave sbrow decoding for all tile cols in a
// tile row, since otherwise subsequent threads will be blocked
// waiting for the post-filter to complete
t->ts = ts;
t->by = sby << f->sb_shift;
dav1d_decode_tile_sbrow(t);
// signal progress
pthread_mutex_lock(&ts->tile_thread.lock);
atomic_store(&ts->progress, 1 + sby);
pthread_cond_broadcast(&ts->tile_thread.cond);
pthread_mutex_unlock(&ts->tile_thread.lock);
}
}
return NULL;
}
|
