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
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
|
#include "Halide.h"
using namespace Halide;
int multi_thread_type_test(MemoryType memory_type) {
Func f1("f1"), f2("f2"), f3("f3"), f4("f4"), f5("f5"), f6("f6");
Var x, y, z;
f1(x, y, z) = cast<uint8_t>(1);
f2(x, y, z) = cast<uint32_t>(f1(x + 1, y, z) + f1(x, y + 1, z));
f3(x, y, z) = cast<uint16_t>(f2(x + 1, y, z) + f2(x, y + 1, z));
f4(x, y, z) = cast<uint16_t>(f3(x + 1, y, z) + f3(x, y + 1, z));
f5(x, y, z) = cast<uint32_t>(f4(x + 1, y, z) + f4(x, y + 1, z));
f6(x, y, z) = cast<uint8_t>(f5(x + 1, y, z) + f5(x, y + 1, z));
Var thread_x, thread_y;
f6.compute_root().gpu_tile(x, y, thread_x, thread_y, 1, 1);
f5.compute_at(f6, x).gpu_threads(x, y).store_in(memory_type);
f4.compute_at(f6, x).gpu_threads(x, y).store_in(memory_type);
f3.compute_at(f6, x).gpu_threads(x, y).store_in(memory_type);
f2.compute_at(f6, x).gpu_threads(x, y).store_in(memory_type);
f1.compute_at(f6, x).gpu_threads(x, y).store_in(memory_type);
const int size_x = 200;
const int size_y = 200;
const int size_z = 4;
Buffer<uint8_t> out = f6.realize({size_x, size_y, size_z});
uint8_t correct = 32;
for (int z = 0; z < size_z; z++) {
for (int y = 0; y < size_y; y++) {
for (int x = 0; x < size_x; x++) {
if (out(x, y, z) != correct) {
printf("out(%d, %d, %d) = %d instead of %d\n",
x, y, z, out(x, y, z), correct);
return 1;
}
}
}
}
printf("OK\n");
return 0;
}
int pyramid_test(MemoryType memory_type) {
const int levels = 10;
const int size_x = 100;
const int size_y = 100;
Var x, y, z, xo, xi, yo, yi, thread_x, thread_y;
std::vector<Func> funcs(levels);
funcs[0](x, y) = 1;
for (int i = 1; i < levels; ++i) {
funcs[i](x, y) = funcs[i - 1](2 * x, y);
}
funcs[levels - 1]
.compute_root()
.gpu_tile(x, y, thread_x, thread_y, 3, 4);
for (int i = levels - 2; i >= 0; --i) {
funcs[i]
.compute_at(funcs[levels - 1], x)
.split(x, xo, xi, 1 << (levels - i - 1))
.gpu_threads(xo, y)
.store_in(memory_type);
}
Buffer<int> out = funcs[levels - 1].realize({size_x, size_y});
int correct = 1;
for (int y = 0; y < size_y; y++) {
for (int x = 0; x < size_x; x++) {
if (out(x, y) != correct) {
printf("out(%d, %d) = %d instead of %d\n",
x, y, out(x, y), correct);
return 1;
}
}
}
printf("OK\n");
return 0;
}
int inverted_pyramid_test(MemoryType memory_type) {
const int levels = 6;
const int size_x = 8 * 16 * 4;
const int size_y = 8 * 16 * 4;
Var x, y, z, yo, yi, yii, xo, xi, xii, thread_x, thread_y;
std::vector<Func> funcs(levels);
funcs[0](x, y) = 1;
for (int i = 1; i < levels; ++i) {
funcs[i](x, y) = funcs[i - 1](x / 2, y);
}
funcs[levels - 1]
.compute_root()
.tile(x, y, xi, yi, 64, 64)
.gpu_blocks(x, y)
.tile(xi, yi, xii, yii, 16, 16)
.gpu_threads(xi, yi);
for (int i = levels - 2; i >= 0; --i) {
funcs[i]
.compute_at(funcs[levels - 1], x)
.tile(x, y, xi, yi, 4, 4)
.gpu_threads(xi, yi)
.store_in(memory_type);
}
funcs[levels - 1]
.bound(x, 0, size_x)
.bound(y, 0, size_y);
Buffer<int> out = funcs[levels - 1].realize({size_x, size_y});
int correct = 1;
for (int y = 0; y < size_y; y++) {
for (int x = 0; x < size_x; x++) {
if (out(x, y) != correct) {
printf("out(%d, %d) = %d instead of %d\n",
x, y, out(x, y), correct);
return 1;
}
}
}
printf("OK\n");
return 0;
}
int dynamic_shared_test(MemoryType memory_type) {
Func f1, f2, f3, f4;
Var x, xo, xi, thread_xo;
f1(x) = x;
f2(x) = f1(x) + f1(2 * x);
f3(x) = f2(x) + f2(2 * x);
f4(x) = f3(x) + f3(2 * x);
f4.split(x, xo, xi, 16).gpu_tile(xo, thread_xo, 16);
f3.compute_at(f4, xo).split(x, xo, xi, 16).gpu_threads(xi).store_in(memory_type);
f2.compute_at(f4, xo).split(x, xo, xi, 16).gpu_threads(xi).store_in(memory_type);
f1.compute_at(f4, xo).split(x, xo, xi, 16).gpu_threads(xi).store_in(memory_type);
// The amount of shared memory required varies with x
Buffer<int> out = f4.realize({500});
for (int x = 0; x < out.width(); x++) {
int correct = 27 * x;
if (out(x) != correct) {
printf("out(%d) = %d instead of %d\n",
x, out(x), correct);
return 1;
}
}
printf("OK\n");
return 0;
}
int main(int argc, char **argv) {
Target t = get_jit_target_from_environment();
if (!t.has_gpu_feature()) {
printf("[SKIP] No GPU target enabled.\n");
return 0;
}
for (auto memory_type : {MemoryType::GPUShared, MemoryType::Heap}) {
printf("Running multi thread type test\n");
if (multi_thread_type_test(memory_type) != 0) {
return 1;
}
printf("Running pyramid test\n");
if (pyramid_test(memory_type) != 0) {
return 1;
}
printf("Running inverted pyramid test\n");
if (inverted_pyramid_test(memory_type) != 0) {
return 1;
}
printf("Running dynamic shared test\n");
if (t.has_feature(Target::Vulkan) && ((t.os == Target::IOS) || t.os == Target::OSX)) {
printf("Skipping test for Vulkan on iOS/OSX (MoltenVK doesn't support dynamic sizes for shared memory)!\n");
} else {
if (dynamic_shared_test(memory_type) != 0) {
return 1;
}
}
}
printf("Success!\n");
return 0;
}
|
