c722ef4874
Summary: This is only really meaningful for the NVPTX target. Not all build environments support host LTO and these are redundant tests, just clean this up and make it run faster.
55 lines
1.1 KiB
C++
55 lines
1.1 KiB
C++
// clang-format off
|
|
// RUN: %libomptarget-compilexx-generic && %libomptarget-run-generic 2>&1 | %fcheck-generic
|
|
// clang-format on
|
|
|
|
// UNSUPPORTED: aarch64-unknown-linux-gnu
|
|
// UNSUPPORTED: x86_64-unknown-linux-gnu
|
|
// UNSUPPORTED: s390x-ibm-linux-gnu
|
|
|
|
// REQUIRES: amdgcn-amd-amdhsa
|
|
|
|
#include <omp.h>
|
|
#include <stdio.h>
|
|
|
|
#define N 100
|
|
|
|
bool schedule(int lb, int ub, int stride, int chunk) {
|
|
int i;
|
|
|
|
int result[N];
|
|
for (i = 0; i < N; i++) {
|
|
result[i] = 0;
|
|
}
|
|
|
|
#pragma omp target parallel for schedule(dynamic, chunk) \
|
|
map(tofrom : result[ : N])
|
|
for (i = lb; i < ub; i += stride) {
|
|
result[i] += i;
|
|
}
|
|
|
|
int value = 0;
|
|
bool success = true;
|
|
for (i = 0; i < N; i += stride) {
|
|
if (value != result[i]) {
|
|
printf("ERROR: result[%d] = %d instead of %d\n", i, result[i], value);
|
|
success = false;
|
|
break;
|
|
}
|
|
value += stride;
|
|
}
|
|
|
|
return success;
|
|
}
|
|
|
|
int main() {
|
|
// CHECK: SUCCESS CHUNK SIZE 1
|
|
if (schedule(0, N, 5, 1))
|
|
printf("SUCCESS CHUNK SIZE 1\n");
|
|
|
|
// CHECK: SUCCESS CHUNK SIZE 3
|
|
if (schedule(0, N, 5, 3))
|
|
printf("SUCCESS CHUNK SIZE 3\n");
|
|
|
|
return 0;
|
|
}
|