Skip to content

Commit

Permalink
Fix Kokkos_ENABLE_CUDA_LAMBDA for Trilinos
Browse files Browse the repository at this point in the history
  • Loading branch information
masterleinad committed May 23, 2023
1 parent 382f0be commit 85b014b
Showing 1 changed file with 8 additions and 1 deletion.
9 changes: 8 additions & 1 deletion cmake/kokkos_enable_options.cmake
Original file line number Diff line number Diff line change
Expand Up @@ -34,7 +34,14 @@ KOKKOS_ENABLE_OPTION(CUDA_LDG_INTRINSIC OFF "Whether to use CUDA LDG intrinsic
# bogus warning, but also exports the Kokkos_ENABLE_CUDA_LAMBDA variable and
# sets it to ON. This if-clause is a crutch that delays the refactoring of the
# way we declare all options until after we get rid of TriBITS.
KOKKOS_ENABLE_OPTION(CUDA_LAMBDA ${Kokkos_ENABLE_CUDA} "Whether to allow lambda expressions on the device with NVCC **DEPRECATED**")
IF (Trilinos_ENABLE_Kokkos AND TPL_ENABLE_CUDA)
SET(CUDA_LAMBDA_DEFAULT ON)
ELSEIF (KOKKOS_ENABLE_CUDA)
SET(CUDA_LAMBDA_DEFAULT ON)
ELSE()
SET(CUDA_LAMBDA_DEFAULT OFF)
ENDIF()
KOKKOS_ENABLE_OPTION(CUDA_LAMBDA ${CUDA_LAMBDA_DEFAULT} "Whether to allow lambda expressions on the device with NVCC **DEPRECATED**")

# As of 08/12/2021 CudaMallocAsync causes issues if UCX is used as MPI communication layer.
KOKKOS_ENABLE_OPTION(IMPL_CUDA_MALLOC_ASYNC OFF "Whether to enable CudaMallocAsync (requires CUDA Toolkit 11.2)")
Expand Down

0 comments on commit 85b014b

Please sign in to comment.