| 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(size: 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(format: "PASS\n"); |
| 45 | |
| 46 | free(ptr: res); |
| 47 | |
| 48 | return 0; |
| 49 | } |
| 50 |