Files
lammps/lib/kokkos/kokkos_5731.diff
2023-01-20 07:40:52 -07:00

47 lines
1.7 KiB
Diff

diff --git a/lib/kokkos/core/src/Cuda/Kokkos_Cuda_BlockSize_Deduction.hpp b/lib/kokkos/core/src/Cuda/Kokkos_Cuda_BlockSize_Deduction.hpp
index 30b6958a67..b94f053272 100644
--- a/lib/kokkos/core/src/Cuda/Kokkos_Cuda_BlockSize_Deduction.hpp
+++ b/lib/kokkos/core/src/Cuda/Kokkos_Cuda_BlockSize_Deduction.hpp
@@ -207,41 +207,6 @@ int cuda_get_opt_block_size(const CudaInternal* cuda_instance,
LaunchBounds{});
}
-// NOTE these number can be obtained several ways:
-// * One option is to download the CUDA Occupancy Calculator spreadsheet, select
-// "Compute Capability" first and check what is the smallest "Shared Memory
-// Size Config" that is available. The "Shared Memory Per Multiprocessor" in
-// bytes is then to be found below in the summary.
-// * Another option would be to look for the information in the "Tuning
-// Guide(s)" of the CUDA Toolkit Documentation for each GPU architecture, in
-// the "Shared Memory" section (more tedious)
-inline size_t get_shmem_per_sm_prefer_l1(cudaDeviceProp const& properties) {
- int const compute_capability = properties.major * 10 + properties.minor;
- return [compute_capability]() {
- switch (compute_capability) {
- case 30:
- case 32:
- case 35: return 16;
- case 37: return 80;
- case 50:
- case 53:
- case 60:
- case 62: return 64;
- case 52:
- case 61: return 96;
- case 70:
- case 80:
- case 86:
- case 90: return 8;
- case 75: return 32;
- default:
- Kokkos::Impl::throw_runtime_exception(
- "Unknown device in cuda block size deduction");
- }
- return 0;
- }() * 1024;
-}
-
} // namespace Impl
} // namespace Kokkos