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
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
|
/* SPDX-License-Identifier: GPL-2.0-or-later
* Copyright 2011 Blender Foundation. */
#include "COM_ExecutionSystem.h"
#include "COM_Debug.h"
#include "COM_ExecutionGroup.h"
#include "COM_FullFrameExecutionModel.h"
#include "COM_NodeOperation.h"
#include "COM_NodeOperationBuilder.h"
#include "COM_TiledExecutionModel.h"
#include "COM_WorkPackage.h"
#include "COM_WorkScheduler.h"
#ifdef WITH_CXX_GUARDEDALLOC
# include "MEM_guardedalloc.h"
#endif
namespace blender::compositor {
ExecutionSystem::ExecutionSystem(RenderData *rd,
Scene *scene,
bNodeTree *editingtree,
bool rendering,
bool fastcalculation,
const char *view_name)
{
num_work_threads_ = WorkScheduler::get_num_cpu_threads();
context_.set_view_name(view_name);
context_.set_scene(scene);
context_.set_bnodetree(editingtree);
context_.set_preview_hash(editingtree->previews);
context_.set_fast_calculation(fastcalculation);
/* initialize the CompositorContext */
if (rendering) {
context_.set_quality((eCompositorQuality)editingtree->render_quality);
}
else {
context_.set_quality((eCompositorQuality)editingtree->edit_quality);
}
context_.set_rendering(rendering);
context_.setHasActiveOpenCLDevices(WorkScheduler::has_gpu_devices() &&
(editingtree->flag & NTREE_COM_OPENCL));
context_.set_render_data(rd);
BLI_mutex_init(&work_mutex_);
BLI_condition_init(&work_finished_cond_);
{
NodeOperationBuilder builder(&context_, editingtree, this);
builder.convert_to_operations(this);
}
switch (context_.get_execution_model()) {
case eExecutionModel::Tiled:
execution_model_ = new TiledExecutionModel(context_, operations_, groups_);
break;
case eExecutionModel::FullFrame:
execution_model_ = new FullFrameExecutionModel(context_, active_buffers_, operations_);
break;
default:
BLI_assert_msg(0, "Non implemented execution model");
break;
}
}
ExecutionSystem::~ExecutionSystem()
{
BLI_condition_end(&work_finished_cond_);
BLI_mutex_end(&work_mutex_);
delete execution_model_;
for (NodeOperation *operation : operations_) {
delete operation;
}
operations_.clear();
for (ExecutionGroup *group : groups_) {
delete group;
}
groups_.clear();
}
void ExecutionSystem::set_operations(const Vector<NodeOperation *> &operations,
const Vector<ExecutionGroup *> &groups)
{
operations_ = operations;
groups_ = groups;
}
void ExecutionSystem::execute()
{
DebugInfo::execute_started(this);
for (NodeOperation *op : operations_) {
op->init_data();
}
execution_model_->execute(*this);
}
void ExecutionSystem::execute_work(const rcti &work_rect,
std::function<void(const rcti &split_rect)> work_func)
{
if (is_breaked()) {
return;
}
/* Split work vertically to maximize continuous memory. */
const int work_height = BLI_rcti_size_y(&work_rect);
const int num_sub_works = MIN2(num_work_threads_, work_height);
const int split_height = num_sub_works == 0 ? 0 : work_height / num_sub_works;
int remaining_height = work_height - split_height * num_sub_works;
Vector<WorkPackage> sub_works(num_sub_works);
int sub_work_y = work_rect.ymin;
int num_sub_works_finished = 0;
for (int i = 0; i < num_sub_works; i++) {
int sub_work_height = split_height;
/* Distribute remaining height between sub-works. */
if (remaining_height > 0) {
sub_work_height++;
remaining_height--;
}
WorkPackage &sub_work = sub_works[i];
sub_work.type = eWorkPackageType::CustomFunction;
sub_work.execute_fn = [=, &work_func, &work_rect]() {
if (is_breaked()) {
return;
}
rcti split_rect;
BLI_rcti_init(
&split_rect, work_rect.xmin, work_rect.xmax, sub_work_y, sub_work_y + sub_work_height);
work_func(split_rect);
};
sub_work.executed_fn = [&]() {
BLI_mutex_lock(&work_mutex_);
num_sub_works_finished++;
if (num_sub_works_finished == num_sub_works) {
BLI_condition_notify_one(&work_finished_cond_);
}
BLI_mutex_unlock(&work_mutex_);
};
WorkScheduler::schedule(&sub_work);
sub_work_y += sub_work_height;
}
BLI_assert(sub_work_y == work_rect.ymax);
WorkScheduler::finish();
/* Ensure all sub-works finished.
* TODO: This a workaround for WorkScheduler::finish() not waiting all works on queue threading
* model. Sync code should be removed once it's fixed. */
BLI_mutex_lock(&work_mutex_);
if (num_sub_works_finished < num_sub_works) {
BLI_condition_wait(&work_finished_cond_, &work_mutex_);
}
BLI_mutex_unlock(&work_mutex_);
}
bool ExecutionSystem::is_breaked() const
{
const bNodeTree *btree = context_.get_bnodetree();
return btree->test_break(btree->tbh);
}
} // namespace blender::compositor
|
