Merge pull request #3517 from anagainaru/set-cuda-architecture

Set the CMAKE GPU architectures to what is used for Kokkos
ornladios · Mar 2, 2023 · 5c49319 · 5c49319
2 parents 08055ac + e6a0702
commit 5c49319
Show file tree

Hide file tree

Showing 3 changed files with 12 additions and 3 deletions.
diff --git a/CMakeLists.txt b/CMakeLists.txt
@@ -173,11 +173,19 @@ if(ADIOS2_HAVE_CUDA OR ADIOS2_HAVE_Kokkos_CUDA)
     set(CMAKE_CUDA_STANDARD 11)
     set(CMAKE_CUDA_STANDARD_REQUIRED TRUE)
     if(NOT DEFINED CMAKE_CUDA_ARCHITECTURES)
-        # Mininum common non-deprecated architecture
-        set(CMAKE_CUDA_ARCHITECTURES 52)
+        if(DEFINED Kokkos_CUDA_ARCHITECTURES)
+            set(CMAKE_CUDA_ARCHITECTURES Kokkos_CUDA_ARCHITECTURES)
+        else()
+            # Mininum common non-deprecated architecture
+            set(CMAKE_CUDA_ARCHITECTURES 52)
+        endif()
     endif()
 endif()
 
+if(NOT DEFINED CMAKE_HIP_ARCHITECTURES AND DEFINED Kokkos_HIP_ARCHITECTURES)
+    set(CMAKE_HIP_ARCHITECTURES Kokkos_HIP_ARCHITECTURES)
+endif()
+
 if(ADIOS2_HAVE_MPI)
   if(MPIEXEC_MAX_NUMPROCS LESS 4 AND "$ENV{OMPI_MCA_rmaps_base_oversubscribe}")
     message(STATUS "OpenMPI oversubscribe detected: raising MPIEXEC_MAX_NUMPROCS to 4 for testing")

diff --git a/docs/user_guide/source/engines/sst.rst b/docs/user_guide/source/engines/sst.rst
@@ -291,6 +291,7 @@ BeginStep timeouts) and writer-side rules (like queue limit behavior) apply.
  DataTransport                   string              **default varies by platform**, UCX, MPI, RDMA, WAN
  WANDataTransport                string              **sockets**, enet, ib
  ControlTransport                string              **TCP**, Scalable
+ MarshalMethod                   string              **BP5**, BP, FFS
  NetworkInterface                string              **NULL**
  ControlInterface                string              **NULL**
  DataInterface                   string              **NULL**

diff --git a/source/adios2/toolkit/sst/sst_data.h b/source/adios2/toolkit/sst/sst_data.h
@@ -54,7 +54,7 @@ typedef struct _SstStats
 } * SstStats;
 
 #define SST_FOREACH_PARAMETER_TYPE_4ARGS(MACRO)                                \
-    MACRO(MarshalMethod, MarshalMethod, size_t, SstMarshalBP)                  \
+    MACRO(MarshalMethod, MarshalMethod, size_t, SstMarshalBP5)                 \
     MACRO(verbose, Int, int, 0)                                                \
     MACRO(RegistrationMethod, RegMethod, size_t, 0)                            \
     MACRO(StepDistributionMode, StepDistributionMode, size_t, StepsAllToAll)   \