xref: /llvm-project/offload/test/offloading/ompx_bare_ballot_sync.c (revision 506ca19dc9d6a9f0ad47b82e71525743bbe8cf85)
1 // RUN: %libomptarget-compilexx-run-and-check-generic
2 //
3 // REQUIRES: gpu
4 
5 #include <assert.h>
6 #include <ompx.h>
7 #include <stdint.h>
8 #include <stdio.h>
9 #include <stdlib.h>
10 
11 #pragma omp begin declare variant match(device = {arch(amdgcn)})
12 unsigned get_warp_size() { return __builtin_amdgcn_wavefrontsize(); }
13 #pragma omp end declare variant
14 
15 #pragma omp begin declare variant match(device = {arch(nvptx64)})
16 unsigned get_warp_size() { return __nvvm_read_ptx_sreg_warpsize(); }
17 #pragma omp end declare variant
18 
19 #pragma omp begin declare variant match(device = {kind(cpu)})
20 unsigned get_warp_size() { return 1; }
21 #pragma omp end declare variant
22 
23 int main(int argc, char *argv[]) {
24   const int num_blocks = 1;
25   const int block_size = 256;
26   const int N = num_blocks * block_size;
27   int *res = (int *)malloc(N * sizeof(int));
28 
29 #pragma omp target teams ompx_bare num_teams(num_blocks)                       \
30     thread_limit(block_size) map(from : res[0 : N])
31   {
32     int tid = ompx_thread_id_x();
33     uint64_t mask = ompx_ballot_sync(~0LU, tid & 0x1);
34     if (get_warp_size() == 64)
35       res[tid] = mask == 0xaaaaaaaaaaaaaaaa;
36     else
37       res[tid] = mask == 0xaaaaaaaa;
38   }
39 
40   for (int i = 0; i < N; ++i)
41     assert(res[i]);
42 
43   // CHECK: PASS
44   printf("PASS\n");
45 
46   free(res);
47 
48   return 0;
49 }
50