Initial Commit for WIP Cuda Execution Space

Author: ElisabethGiem

CMakeLists.txt

@@ -87,8 +87,9 @@ kr_option(KR_ENABLE_HDF5_PARALLEL "use parallel version of HDF5" OFF KR_ENABLE_H
 
 #Exec space options
 kr_option(KR_ENABLE_EXEC_SPACES "enable resilient execution spaces" OFF)
-kr_option(KR_ENABLE_OPENMP_EXEC_SPACE "enable the resilient OpenMP execution space" ON "KR_ENABLE_EXEC_SPACES;KR_OPENMP_DEVICE_ENABLED")
-if(KR_ENABLE_OPENMP_EXEC_SPACE)
+kr_option(KR_ENABLE_OPENMP_EXEC_SPACE "enable resilient OpenMP execution space" OFF "KR_ENABLE_EXEC_SPACES;KR_OPENMP_DEVICE_ENABLED")
+kr_option(KR_ENABLE_CUDA_EXEC_SPACE "enable resilient Cuda execution space" OFF "KR_ENABLE_EXEC_SPACES;KR_CUDA_DEVICE_ENABLED")
+if(KR_ENABLE_EXEC_SPACES)
   set(KR_MODULAR_REDUNDANCY TRIPLE CACHE STRING "choose level of modular redundancy")
   set_property(CACHE KR_MODULAR_REDUNDANCY PROPERTY STRINGS OFF DOUBLE TRIPLE)
   if (KR_MODULAR_REDUNDANCY STREQUAL TRIPLE)
@@ -127,6 +128,9 @@ if (KR_ENABLE_OPENMP_EXEC_SPACE)
   message(STATUS "OpenMP resilient execution spaces are enabled")
 endif()
 
+if (KR_ENABLE_CUDA_EXEC_SPACE)
+  message(STATUS "Cuda resilient execution spaces are enabled")
+endif()
 
 add_subdirectory(src)
 

src/resilience/ErrorHandler.hpp

@@ -42,6 +42,7 @@
 #define INC_RESILIENCE_ERRORHANDLER_HPP
 
 #include <functional>
+#include <Kokkos_Core.hpp>
 
 namespace KokkosResilience {
 /**

src/resilience/exec_space/CMakeLists.txt

@@ -1,3 +1,7 @@
 if (KR_ENABLE_OPENMP_EXEC_SPACE)
   add_subdirectory(openMP)
 endif()
+
+if (KR_ENABLE_CUDA_EXEC_SPACE)
+  add_subdirectory(cuda)
+endif()

src/resilience/exec_space/ExecSpace.hpp

@@ -46,3 +46,10 @@
 #include "openMP/Resilient_OpenMP_Subscriber.hpp"
 #include "openMP/Resilient_HostSpace.hpp"
 #endif
+
+#ifdef KR_ENABLE_CUDA_EXEC_SPACE
+#include "cuda/Resilient_Cuda.hpp"
+#include "cuda/Resilient_Cuda_Parallel_For.hpp"
+#include "cuda/Resilient_Cuda_Subscriber.hpp"
+#include "cuda/Resilient_CudaSpace.hpp"
+#endif

src/resilience/exec_space/openMP/Resilient_OpenMP_Subscriber.hpp

@@ -350,12 +350,16 @@ struct ResilientDuplicatesSubscriber {
 
         // The first copy constructor in a parallel_for for the given view
         if (res.second) {
+          std::cout << "Do we ever enter the first copy constructor? How many times?" << std::endl;
           c.duplicate_count = 0;
         }
-
+        std::cout << "*****In copy constructor, self has now been set to copy c.duplicate_count=" << c.duplicate_count << "****" << std::endl;
         self = c.copy[c.duplicate_count++];
         // Copy all data, every time
         Kokkos::deep_copy(self, other);
+        std::cout << std::endl << "This is the end of the constructor after Kokkos::deep_copy" <<std::endl;
+        std::cout << "c_duplicate_count = "<<c.duplicate_count<< std::endl;
+
       }
     }
   }

tests/CMakeLists.txt

@@ -44,6 +44,12 @@ if (KR_ENABLE_OPENMP_EXEC_SPACE)
                  )
 endif()
 
+if (KR_ENABLE_CUDA_EXEC_SPACE)
+  target_sources(resilience_tests PRIVATE
+	  TestCudaResilientExecution.cpp
+                 )
+endif()
+
 if (KR_WARNINGS_AS_ERRORS)
   if (CMAKE_CXX_COMPILER_ID STREQUAL "GNU" OR CMAKE_CXX_COMPILER_ID STREQUAL "Clang")
     target_compile_options(resilience_tests PRIVATE "-Wall")