Work around NVHPC 23.x issues

Works around long double usage in device code being treated stricter, and now needing a workaround similar to NVCC for implementation of traits and constants.
nliber · May 5, 2023 · 60b982a · 60b982a
1 parent ea134de
commit 60b982a
Show file tree

Hide file tree

Showing 2 changed files with 10 additions and 2 deletions.
diff --git a/core/unit_test/TestMathematicalConstants.hpp b/core/unit_test/TestMathematicalConstants.hpp
@@ -63,7 +63,7 @@ struct TestMathematicalConstants {
 
   KOKKOS_FUNCTION void use_on_device() const {
 #if defined(KOKKOS_COMPILER_NVCC) || defined(KOKKOS_ENABLE_OPENMPTARGET) || \
-    defined(KOKKOS_ENABLE_OPENACC)
+    defined(KOKKOS_ENABLE_OPENACC) || defined(KOKKOS_COMPILER_NVHPC)
     take_by_value(Trait::value);
 #else
     (void)take_address_of(Trait::value);

diff --git a/core/unit_test/TestNumericTraits.hpp b/core/unit_test/TestNumericTraits.hpp
@@ -40,7 +40,14 @@ struct extrema {
 
   DEFINE_EXTREMA(float, -FLT_MAX, FLT_MAX);
   DEFINE_EXTREMA(double, -DBL_MAX, DBL_MAX);
+
+// FIXME_NVHPC: with 23.3 using long double in KOKKOS_FUNCTION is hard error
+#if !defined(KOKKOS_ENABLE_CUDA) && !defined(KOKKOS_COMPILER_NVHPC)
   DEFINE_EXTREMA(long double, -LDBL_MAX, LDBL_MAX);
+#else
+  static long double min(long double) { return -LDBL_MAX; }
+  static long double max(long double) { return LDBL_MAX; }
+#endif
 
 #undef DEFINE_EXTREMA
 };
@@ -163,7 +170,8 @@ struct TestNumericTraits {
   }
 
   KOKKOS_FUNCTION void use_on_device() const {
-#if defined(KOKKOS_COMPILER_NVCC) || defined(KOKKOS_ENABLE_OPENMPTARGET)
+#if defined(KOKKOS_COMPILER_NVCC) || defined(KOKKOS_COMPILER_NVHPC) || \
+    defined(KOKKOS_ENABLE_OPENMPTARGET) || defined(KOKKOS_ENABLE_OPENACC)
     take_by_value(trait<T>::value);
 #else
     (void)take_address_of(trait<T>::value);