Merge branch 'fix-MPIX_CUDA_AWARE_SUPPORT' into 'master'

Add mpi-ext.h for MPIX_CUDA_AWARE_SUPPORT

See merge request walberla/walberla!622

.clang-format

@@ -61,8 +61,6 @@ IncludeCategories:
     Priority:        3
   - Regex:           '^"core/'
     Priority:        4
-  - Regex:           '^"cuda/'
-    Priority:        5
   - Regex:           '^"domain_decomposition/'
     Priority:        6
   - Regex:           '^"executiontree/'
@@ -75,6 +73,8 @@ IncludeCategories:
     Priority:        10
   - Regex:           '^"geometry/'
     Priority:        11
+  - Regex:           '^"gpu/'
+    Priority:        12
   - Regex:           '^"gui/'
     Priority:        12
   - Regex:           '^"lbm/'

src/core/mpi/MPIWrapper.h

@@ -66,6 +66,9 @@ namespace mpistubs {
 #   pragma GCC diagnostic ignored "-Wsign-conversion"
 #endif
 #include <mpi.h>
+#if defined(OPEN_MPI) && OPEN_MPI
+#include <mpi-ext.h>
+#endif
 #if ( defined WALBERLA_CXX_COMPILER_IS_GNU ) || ( defined WALBERLA_CXX_COMPILER_IS_CLANG )
 #   pragma GCC diagnostic pop
 #endif

src/gpu/communication/NonUniformGPUScheme.h

@@ -30,13 +30,13 @@
 
 #include "domain_decomposition/IBlock.h"
 
-#include "stencil/Directions.h"
-
 #include "gpu/ErrorChecking.h"
 #include "gpu/GPUWrapper.h"
 #include "gpu/communication/CustomMemoryBuffer.h"
 #include "gpu/communication/GeneratedNonUniformGPUPackInfo.h"
 
+#include "stencil/Directions.h"
+
 #include <memory>
 #include <thread>
 
@@ -138,7 +138,8 @@ NonUniformGPUScheme< Stencil >::NonUniformGPUScheme(const weak_ptr< StructuredBl
 {
    WALBERLA_MPI_SECTION()
    {
-#if !(defined(MPIX_CUDA_AWARE_SUPPORT) && MPIX_CUDA_AWARE_SUPPORT)
+// Open MPI supports compile time CUDA-aware support check
+#if (defined(OPEN_MPI) && OPEN_MPI) && !(defined(MPIX_CUDA_AWARE_SUPPORT) && MPIX_CUDA_AWARE_SUPPORT)
       WALBERLA_CHECK(!sendDirectlyFromGPU)
 #endif
    }

src/gpu/communication/UniformGPUScheme.impl.h

@@ -19,6 +19,8 @@
 //
 //======================================================================================================================
 
+#include "core/mpi/MPIWrapper.h"
+
 #include "gpu/ParallelStreams.h"
 
 namespace walberla {
@@ -45,7 +47,8 @@ namespace communication {
    {
       WALBERLA_MPI_SECTION()
       {
-#if !(defined(MPIX_CUDA_AWARE_SUPPORT) && MPIX_CUDA_AWARE_SUPPORT)
+// Open MPI supports compile time CUDA-aware support check
+#if (defined(OPEN_MPI) && OPEN_MPI) && !(defined(MPIX_CUDA_AWARE_SUPPORT) && MPIX_CUDA_AWARE_SUPPORT)
          WALBERLA_CHECK(!sendDirectlyFromGPU)
 #endif
       }