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
|
#include "Halide.h"
using namespace Halide;
Var x, y;
void check(Func f) {
Buffer<int> out = f.realize({256, 256});
out.for_each_element([&](int x, int y) {
if (out(x, y) != x + y) {
printf("out(%d, %d) = %d instead of %d\n", x, y, out(x, y), x + y);
exit(1);
}
});
}
void make_pipeline(Func &A, Func &B) {
A(x, y) = x + y;
B(x, y) = A(x, y);
}
int main(int argc, char **argv) {
if (get_jit_target_from_environment().arch == Target::WebAssembly) {
printf("[SKIP] WebAssembly does not support async() yet.\n");
return 0;
}
if (get_jit_target_from_environment().has_feature(Target::OpenGLCompute)) {
printf("Skipping test for OpenGLCompute as it does not support copy_to_host/device() yet"
" (halide_buffer_copy is unimplemented in that backend).\n");
return 0;
}
// Make a list of extern pipeline stages (just copies) all async
// and connected by double buffers, then try various nestings of
// them. This is a stress test of the async extern storage folding
// logic.
// Basic double-buffered A->B, with no extern stages
{
Func A, B;
make_pipeline(A, B);
A.store_root().compute_at(B, y).fold_storage(y, 2).async();
check(B);
}
// Inject a copy stage between them
{
Func A, B;
make_pipeline(A, B);
A.store_root().compute_at(B, y).fold_storage(y, 2).async();
A.in().store_root().compute_at(B, y).fold_storage(y, 2).async().copy_to_host();
check(B);
}
// Inject a copy stage between them, but nest the first stage into it
{
Func A, B;
make_pipeline(A, B);
A.store_root().compute_at(A.in(), Var::outermost()).fold_storage(y, 2).async();
A.in().store_root().compute_at(B, y).fold_storage(y, 2).async().copy_to_host();
check(B);
}
// Two copy stages, flat
{
Func A, B;
make_pipeline(A, B);
A.store_root().compute_at(B, y).fold_storage(y, 2).async();
A.in().store_root().compute_at(B, y).fold_storage(y, 2).copy_to_host().async();
A.in().in().store_root().compute_at(B, y).fold_storage(y, 2).copy_to_host().async();
check(B);
}
// Two copy stages, each stage nested inside the outermost var of the next
{
Func A, B;
make_pipeline(A, B);
A.store_root().compute_at(A.in(), Var::outermost()).fold_storage(y, 2).async();
A.in().store_root().compute_at(A.in().in(), Var::outermost()).fold_storage(y, 2).copy_to_host().async();
A.in().in().store_root().compute_at(B, y).fold_storage(y, 2).copy_to_host().async();
check(B);
}
if (get_jit_target_from_environment().has_gpu_feature()) {
// Two copy stages, to the device and back, flat
{
Func A, B;
make_pipeline(A, B);
A.store_root().compute_at(B, y).fold_storage(y, 2).async();
A.in().store_root().compute_at(B, y).fold_storage(y, 2).copy_to_device().async();
A.in().in().store_root().compute_at(B, y).fold_storage(y, 2).copy_to_host().async();
check(B);
}
// Two copy stages, to the device and back, each stage nested inside the outermost var of the next
{
Func A, B;
make_pipeline(A, B);
A.store_root().compute_at(A.in(), Var::outermost()).fold_storage(y, 2).async();
A.in().store_root().compute_at(A.in().in(), Var::outermost()).fold_storage(y, 2).copy_to_device().async();
A.in().in().store_root().compute_at(B, y).fold_storage(y, 2).copy_to_host().async();
check(B);
}
// The same, but make one of the copy stages non-extern to force a shared host-dev allocation
{
Func A, B;
make_pipeline(A, B);
A.store_root().compute_at(B, y).fold_storage(y, 2).async();
A.in().store_root().compute_at(B, y).fold_storage(y, 2).async();
A.in().in().store_root().compute_at(B, y).fold_storage(y, 2).copy_to_host().async();
check(B);
}
{
Func A, B;
make_pipeline(A, B);
A.store_root().compute_at(A.in(), Var::outermost()).fold_storage(y, 2).async();
A.in().store_root().compute_at(A.in().in(), Var::outermost()).fold_storage(y, 2).async();
A.in().in().store_root().compute_at(B, y).fold_storage(y, 2).copy_to_host().async();
check(B);
}
{
Func A, B;
make_pipeline(A, B);
A.store_root().compute_at(B, y).fold_storage(y, 2).async();
A.in().store_root().compute_at(B, y).fold_storage(y, 2).copy_to_device().async();
A.in().in().store_root().compute_at(B, y).fold_storage(y, 2).async();
check(B);
}
{
Func A, B;
make_pipeline(A, B);
A.store_root().compute_at(A.in(), Var::outermost()).fold_storage(y, 2).async();
A.in().store_root().compute_at(A.in().in(), Var::outermost()).fold_storage(y, 2).copy_to_device().async();
A.in().in().store_root().compute_at(B, y).fold_storage(y, 2).async();
check(B);
}
}
printf("Success!\n");
return 0;
}
|
