File: atomics_gpu_mutex.cpp

package info (click to toggle)
halide 21.0.0-4
  • links: PTS, VCS
  • area: main
  • in suites: forky, sid
  • size: 55,752 kB
  • sloc: cpp: 289,334; ansic: 22,751; python: 7,486; makefile: 4,299; sh: 2,508; java: 1,549; javascript: 282; pascal: 207; xml: 127; asm: 9
file content (32 lines) | stat: -rw-r--r-- 641 bytes parent folder | download | duplicates (4)
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
#include "Halide.h"

using namespace Halide;

int main(int argc, char **argv) {
    int img_size = 10000;

    Func f;
    Var x;
    RDom r(0, img_size);

    f(x) = Tuple(1, 0);
    f(r) = Tuple(f(r)[1] + 1, f(r)[0] + 1);

    f.compute_root();

    RVar ro, ri;
    f.update()
        .atomic()
        .split(r, ro, ri, 8)
        .gpu_blocks(ro)
        .gpu_threads(ri);

    // hist's update will be lowered to mutex locks,
    // and we don't allow GPU blocks on mutex locks since
    // it leads to deadlocks.
    // This should throw an error
    Realization out = f.realize({img_size});

    printf("Success!\n");
    return 0;
}