Use GPU preprocessor option to compile for GPU

Libensemble · Jan 11, 2023 · bd74796 · bd74796
1 parent f2331fa
commit bd74796
Show file tree

Hide file tree

Showing 2 changed files with 15 additions and 13 deletions.
diff --git a/libensemble/tests/scaling_tests/forces/forces_app/build_forces.sh b/libensemble/tests/scaling_tests/forces/forces_app/build_forces.sh
@@ -33,14 +33,14 @@ mpicc -O3 -o forces.x forces.c -lm
 # Need to toggle to OpenMP target directive in forces.c.
 
 # xl
-# xlc_r -O3 -qsmp=omp -qoffload -o forces.x forces.c
+# xlc_r -WF,-DGPU -O3 -qsmp=omp -qoffload -o forces.x forces.c
 
 # Nvidia (nvc) compiler with mpicc and on Cray system with target (Perlmutter)
-# mpicc -O3 -fopenmp -mp=gpu -o forces.x forces.c
-# cc -O3 -fopenmp -mp=gpu -target-accel=nvidia80 -o forces.x forces.c
+# mpicc -DGPU -O3 -fopenmp -mp=gpu -o forces.x forces.c
+# cc -DGPU -O3 -fopenmp -mp=gpu -target-accel=nvidia80 -o forces.x forces.c
 
 # Spock/Crusher (AMD ROCm compiler)
-# cc -I${ROCM_PATH}/include -L${ROCM_PATH}/lib -lamdhip64 -fopenmp -O3 -o forces.x forces.c
+# cc -DGPU -I${ROCM_PATH}/include -L${ROCM_PATH}/lib -lamdhip64 -fopenmp -O3 -o forces.x forces.c
 
 # Intel oneAPI (Clang based) Compiler (JIT compiled for device)
-# mpiicx -O3 -fiopenmp -fopenmp-targets=spir64 -o forces.x forces.c
+# mpiicx -DGPU -O3 -fiopenmp -fopenmp-targets=spir64 -o forces.x forces.c
diff --git a/libensemble/tests/scaling_tests/forces/forces_app/forces.c b/libensemble/tests/scaling_tests/forces/forces_app/forces.c
@@ -9,7 +9,10 @@
     Particle force arrays are allreduced across ranks.
 
     Sept 2019:
-    Added OpenMP options for CPU and GPU. Toggle in forces_naive function.
+    Added OpenMP options for CPU and GPU.
+
+    Jan 2022:
+    Use GPU preprocessor option to compile for GPU (e.g. -DGPU).
 
     Run executable on N procs:
 
@@ -138,18 +141,17 @@ double forces_naive(int n, int lower, int upper, particle* parr) {
     double ret = 0.0;
     double dx, dy, dz, r, force;
 
+#ifdef GPU
     // For GPU/Accelerators
-    /*
     #pragma omp target teams distribute parallel for \
                 map(to: lower, upper, n) map(tofrom: parr[0:n]) \
-                reduction(+: ret) //thread_limit(128) //*/
-
-    // For CPU
-    //*
+                reduction(+: ret)
+#else
+    // Use OpenMP threads on CPU
     #pragma omp parallel for default(none) shared(n, lower, upper, parr) \
                              private(i, j, dx, dy, dz, r, force) \
-                             reduction(+:ret)  //*/
-
+                             reduction(+:ret)
+#endif
     for(i=lower; i<upper; i++) {
         for(j=0; j<n; j++){