Drop Clang+CUDA workaround

fnrizzi · Nov 3, 2023 · 1e1ed13 · 1e1ed13
1 parent 6fc7a49
commit 1e1ed13
Show file tree

Hide file tree

Showing 4 changed files with 0 additions and 12 deletions.
diff --git a/Makefile.kokkos b/Makefile.kokkos
@@ -687,10 +687,6 @@ ifeq ($(KOKKOS_INTERNAL_USE_CUDA), 1)
     endif
   endif
 
-  ifeq ($(KOKKOS_INTERNAL_COMPILER_CLANG), 1)
-    tmp := $(call kokkos_append_header,"$H""define KOKKOS_IMPL_CUDA_CLANG_WORKAROUND")
-  endif
-
   ifeq ($(KOKKOS_INTERNAL_CUDA_DISABLE_MALLOC_ASYNC), 0)
     tmp := $(call kokkos_append_header,"$H""define KOKKOS_ENABLE_IMPL_CUDA_MALLOC_ASYNC")
   else

diff --git a/cmake/KokkosCore_config.h.in b/cmake/KokkosCore_config.h.in
@@ -59,7 +59,6 @@
 #cmakedefine KOKKOS_ENABLE_HBWSPACE
 #cmakedefine KOKKOS_ENABLE_LIBDL
 #cmakedefine KOKKOS_ENABLE_LIBQUADMATH
-#cmakedefine KOKKOS_IMPL_CUDA_CLANG_WORKAROUND
 #cmakedefine KOKKOS_ENABLE_ONEDPL
 
 #cmakedefine KOKKOS_ARCH_SSE42

diff --git a/cmake/kokkos_arch.cmake b/cmake/kokkos_arch.cmake
@@ -191,9 +191,6 @@ IF (KOKKOS_CXX_COMPILER_ID STREQUAL Clang)
   ELSEIF(CUDAToolkit_BIN_DIR)
     GLOBAL_APPEND(KOKKOS_CUDA_OPTIONS --cuda-path=${CUDAToolkit_BIN_DIR}/..)
   ENDIF()
-  IF (KOKKOS_ENABLE_CUDA)
-     SET(KOKKOS_IMPL_CUDA_CLANG_WORKAROUND ON CACHE BOOL "enable CUDA Clang workarounds" FORCE)
-  ENDIF()
 ELSEIF (KOKKOS_CXX_COMPILER_ID STREQUAL NVHPC)
   SET(CUDA_ARCH_FLAG "-gpu")
   GLOBAL_APPEND(KOKKOS_CUDA_OPTIONS -cuda)

diff --git a/core/unit_test/TestTeamVector.hpp b/core/unit_test/TestTeamVector.hpp
@@ -1012,7 +1012,6 @@ struct checkScan {
 };
 }  // namespace VectorScanReducer
 
-#if !defined(KOKKOS_IMPL_CUDA_CLANG_WORKAROUND)
 TEST(TEST_CATEGORY, team_vector) {
   ASSERT_TRUE((TestTeamVector::Test<TEST_EXECSPACE>(0)));
   ASSERT_TRUE((TestTeamVector::Test<TEST_EXECSPACE>(1)));
@@ -1028,9 +1027,7 @@ TEST(TEST_CATEGORY, team_vector) {
   ASSERT_TRUE((TestTeamVector::Test<TEST_EXECSPACE>(11)));
   ASSERT_TRUE((TestTeamVector::Test<TEST_EXECSPACE>(12)));
 }
-#endif
 
-#if !defined(KOKKOS_IMPL_CUDA_CLANG_WORKAROUND)
 TEST(TEST_CATEGORY, triple_nested_parallelism) {
 // With KOKKOS_ENABLE_DEBUG enabled, the functor uses too many registers to run
 // with a team size of 32 on GPUs, 16 is the max possible (at least on a K80
@@ -1055,7 +1052,6 @@ TEST(TEST_CATEGORY, triple_nested_parallelism) {
   TestTripleNestedReduce<double, TEST_EXECSPACE>(8192, 2048, 16, 16);
   TestTripleNestedReduce<double, TEST_EXECSPACE>(8192, 2048, 7, 16);
 }
-#endif
 
 TEST(TEST_CATEGORY, parallel_scan_with_reducers) {
   using T = double;