openmp/libomptarget/test/offloading/thread_limit.c

   1 // clang-format off
   2 // RUN: %libomptarget-compile-generic
   3 // RUN: env LIBOMPTARGET_INFO=16 \
   4 // RUN:   %libomptarget-run-generic 2>&1 | %fcheck-generic --check-prefix=DEFAULT
   5
   6 // UNSUPPORTED: nvptx64-nvidia-cuda
   7 // UNSUPPORTED: nvptx64-nvidia-cuda-LTO
   8 // UNSUPPORTED: aarch64-unknown-linux-gnu
   9 // UNSUPPORTED: aarch64-unknown-linux-gnu-LTO
  10 // UNSUPPORTED: x86_64-pc-linux-gnu
  11 // UNSUPPORTED: x86_64-pc-linux-gnu-LTO
  12
  13 int main() {
  14   int n = 1 << 20;
  15   int th = 12;
  16   int te = n / th;
  17 // DEFAULT: 12 (MaxFlatWorkGroupSize:
  18 #pragma omp target
  19 #pragma omp teams loop num_teams(te), thread_limit(th)
  20   for (int i = 0; i < n; i++) {
  21   }
  22
  23 // DEFAULT: 13 (MaxFlatWorkGroupSize:
  24   #pragma omp target
  25   #pragma omp teams distribute parallel for simd num_teams(te), thread_limit(th+1) simdlen(64)
  26   for(int i = 0; i < n; i++) {
  27   }
  28   return 0;
  29 }