Splines tuning cpu (#239)

* tuning cpu * tuning
CExA-project · Nov 23, 2023 · 0f72c16 · 0f72c16
1 parent c8cd5f0
commit 0f72c16
Show file tree

Hide file tree

Showing 2 changed files with 8 additions and 8 deletions.
diff --git a/benchmarks/splines.cpp b/benchmarks/splines.cpp
@@ -184,14 +184,14 @@ int par_chunks_per_seq_chunk_ref = 1;
 unsigned int preconditionner_max_block_size_ref = 1u;
 #elif defined(KOKKOS_ENABLE_OPENMP)
 std::string chip = "cpu";
-int cols_per_par_chunk_ref = 512;
+int cols_per_par_chunk_ref = 256;
 int par_chunks_per_seq_chunk_ref = Kokkos::OpenMP().concurrency();
-unsigned int preconditionner_max_block_size_ref = 8u;
+unsigned int preconditionner_max_block_size_ref = 32u;
 #elif defined(KOKKOS_ENABLE_SERIAL)
 std::string chip = "cpu";
-int cols_per_par_chunk_ref = 512;
+int cols_per_par_chunk_ref = 256;
 int par_chunks_per_seq_chunk_ref = 1;
-unsigned int preconditionner_max_block_size_ref = 8u;
+unsigned int preconditionner_max_block_size_ref = 32u;
 #endif
 
 BENCHMARK(characteristics_advection)

diff --git a/include/ddc/kernels/splines/matrix_sparse.hpp b/include/ddc/kernels/splines/matrix_sparse.hpp
@@ -67,12 +67,12 @@ class Matrix_Sparse : public Matrix
         } else {
 #ifdef KOKKOS_ENABLE_SERIAL
             if (std::is_same_v<ExecSpace, Kokkos::Serial>) {
-                m_cols_per_par_chunk = 512;
+                m_cols_per_par_chunk = 256;
             }
 #endif
 #ifdef KOKKOS_ENABLE_OPENMP
             if (std::is_same_v<ExecSpace, Kokkos::OpenMP>) {
-                m_cols_per_par_chunk = 512;
+                m_cols_per_par_chunk = 256;
             }
 #endif
 #ifdef KOKKOS_ENABLE_CUDA
@@ -117,12 +117,12 @@ class Matrix_Sparse : public Matrix
         } else {
 #ifdef KOKKOS_ENABLE_SERIAL
             if (std::is_same_v<ExecSpace, Kokkos::Serial>) {
-                m_preconditionner_max_block_size = 8u;
+                m_preconditionner_max_block_size = 32u;
             }
 #endif
 #ifdef KOKKOS_ENABLE_OPENMP
             if (std::is_same_v<ExecSpace, Kokkos::OpenMP>) {
-                m_preconditionner_max_block_size = 8u;
+                m_preconditionner_max_block_size = 32u;
             }
 #endif
 #ifdef KOKKOS_ENABLE_CUDA