330d8983d2
In a nutshell, this moves our libomptarget code to populate the offload subproject. With this commit, users need to enable the new LLVM/Offload subproject as a runtime in their cmake configuration. No further changes are expected for downstream code. Tests and other components still depend on OpenMP and have also not been renamed. The results below are for a build in which OpenMP and Offload are enabled runtimes. In addition to the pure `git mv`, we needed to adjust some CMake files. Nothing is intended to change semantics. ``` ninja check-offload ``` Works with the X86 and AMDGPU offload tests ``` ninja check-openmp ``` Still works but doesn't build offload tests anymore. ``` ls install/lib ``` Shows all expected libraries, incl. - `libomptarget.devicertl.a` - `libomptarget-nvptx-sm_90.bc` - `libomptarget.rtl.amdgpu.so` -> `libomptarget.rtl.amdgpu.so.18git` - `libomptarget.so` -> `libomptarget.so.18git` Fixes: https://github.com/llvm/llvm-project/issues/75124 --------- Co-authored-by: Saiyedul Islam <Saiyedul.Islam@amd.com>
76 lines
2.1 KiB
C
76 lines
2.1 KiB
C
// RUN: %libomptarget-compile-generic -O3 && %libomptarget-run-generic | %fcheck-generic
|
|
|
|
#include <omp.h>
|
|
#include <stdio.h>
|
|
#include <stdlib.h>
|
|
|
|
#define MAX_N 25000
|
|
|
|
void reset_input(double *a, double *a_h, double *b, double *c) {
|
|
for(int i = 0 ; i < MAX_N ; i++) {
|
|
a[i] = a_h[i] = i;
|
|
b[i] = i*2;
|
|
c[i] = i-3;
|
|
}
|
|
}
|
|
|
|
int main(int argc, char *argv[]) {
|
|
double *a = (double *)calloc(MAX_N, sizeof(double));
|
|
double *a_h = (double *)calloc(MAX_N, sizeof(double));
|
|
double *d = (double *)calloc(MAX_N, sizeof(double));
|
|
double *d_h = (double *)calloc(MAX_N, sizeof(double));
|
|
double *b = (double *)calloc(MAX_N, sizeof(double));
|
|
double *c = (double *)calloc(MAX_N, sizeof(double));
|
|
|
|
#pragma omp target enter data map(to:a[:MAX_N],b[:MAX_N],c[:MAX_N],d[:MAX_N])
|
|
|
|
for (int n = 32 ; n < MAX_N ; n+=5000) {
|
|
reset_input(a, a_h, b, c);
|
|
|
|
#pragma omp target update to(a[:n],b[:n],c[:n],d[:n])
|
|
int t = 0;
|
|
for (int tms = 1 ; tms <= 256 ; tms *= 2) { // 8 times
|
|
for (int ths = 32 ; ths <= 1024 ; ths *= 2) { // 6 times
|
|
t++;
|
|
#pragma omp target
|
|
#pragma omp teams num_teams(tms) thread_limit(ths)
|
|
{
|
|
#pragma omp distribute parallel for
|
|
for (int i = 0; i < n; ++i) {
|
|
a[i] += b[i] + c[i];
|
|
}
|
|
#pragma omp distribute parallel for
|
|
for (int i = 0; i < n; ++i) {
|
|
d[i] -= b[i] + c[i];
|
|
}
|
|
}
|
|
} // loop over 'ths'
|
|
} // loop over 'tms'
|
|
|
|
// check results for each 'n'
|
|
for (int times = 0 ; times < t ; times++) {
|
|
for (int i = 0; i < n; ++i) {
|
|
a_h[i] += b[i] + c[i];
|
|
}
|
|
for (int i = 0; i < n; ++i)
|
|
d_h[i] -= b[i] + c[i];
|
|
}
|
|
#pragma omp target update from(a[:n],d[:n])
|
|
|
|
for (int i = 0; i < n; ++i) {
|
|
if (a_h[i] != a[i]) {
|
|
printf("A Error at n = %d, i = %d: host = %f, device = %f\n", n, i, a_h[i], a[i]);
|
|
return 1;
|
|
}
|
|
if (d_h[i] != d[i]) {
|
|
printf("D Error at n = %d, i = %d: host = %lf, device = %lf\n", n, i, d_h[i], d[i]);
|
|
return 1;
|
|
}
|
|
}
|
|
} // loop over 'n'
|
|
|
|
// CHECK: Succeeded
|
|
printf("Succeeded\n");
|
|
return 0;
|
|
}
|