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
|
#include "Halide.h"
using namespace Halide;
int main() {
Target target = get_jit_target_from_environment();
if (!target.has_gpu_feature()) {
printf("[SKIP] No GPU target enabled.\n");
return 0;
}
Var x, y, xi, yi;
ImageParam table(Int(32), 1);
Func f, g, h;
// It's possible to have a buffer simultaneously read on CPU and
// GPU if a load from it gets lifted into a predicate used by skip
// stages. This tests that path.
f(x, y) = x * 2 + y + table(x);
g(x, y) = x + y * 2 + table(y);
h(x, y) = select(table(0) == 0, f(x, y), g(x, y));
f.compute_root().gpu_tile(x, y, xi, yi, 8, 8);
g.compute_root().gpu_tile(x, y, xi, yi, 8, 8);
h.compute_root().gpu_tile(x, y, xi, yi, 8, 8);
Buffer<int32_t> t(32);
t.fill(17);
t(0) = 0;
table.set(t);
Buffer<int32_t> result1 = h.realize({20, 20});
t(0) = 1;
table.set(t);
Buffer<int32_t> result2 = h.realize({20, 20});
for (int y = 0; y < 20; y++) {
for (int x = 0; x < 20; x++) {
int c1 = x * 2 + y + (x == 0 ? 0 : 17);
int c2 = x + y * 2 + (y == 0 ? 1 : 17);
if (result1(x, y) != c1) {
printf("result1(%d, %d) = %d instead of %d\n",
x, y, result1(x, y), c1);
return 1;
}
if (result2(x, y) != c2) {
printf("result2(%d, %d) = %d instead of %d\n",
x, y, result2(x, y), c2);
return 1;
}
}
}
printf("Success!\n");
return 0;
}
|
