Draft: #48: vt and magistrate support

CMakeLists.txt

@@ -37,7 +37,12 @@ set_property(TARGET resilience PROPERTY CXX_STANDARD ${Kokkos_CXX_STANDARD})
 target_link_libraries(resilience PUBLIC Kokkos::kokkos)
 
 option(KR_ENABLE_VELOC "use VeloC backend for automatic checkpointing" ON)
-option(KR_ENABLE_STDFILE "use StdFile backend for automatic checkpointing" ON)
+option(KR_ENABLE_STDFILE "use StdFile backend for automatic checkpointing" OFF)
+
+option(KR_ENABLE_MAGISTRATE "use Magistrate for serializing and deserializing" OFF)
+option(KR_ENABLE_RESILIENT_EXEC "enable resilient execution spaces" OFF)
+
+option(KR_ENABLE_VT "use VT for backend coordination" OFF)
 
 include(CMakeDependentOption)
 
@@ -55,6 +60,14 @@ if (KR_ENABLE_VELOC)
   endif()
 endif()
 
+if (KR_ENABLE_VT)
+  find_package(vt REQUIRED)
+  target_link_libraries(resilience PUBLIC vt::runtime::vt)
+  target_compile_definitions(resilience PUBLIC KR_ENABLE_VT)
+
+  set(KR_ENABLE_MAGISTRATE ON)
+endif()
+
 # StdFile backend
 if (KR_ENABLE_STDFILE)
   target_compile_definitions(resilience PUBLIC KR_ENABLE_STDFILE)
@@ -70,6 +83,12 @@ if (KR_ENABLE_TRACING)
   target_compile_definitions(resilience PUBLIC KR_ENABLE_TRACING)
 endif()
 
+if (KR_ENABLE_MAGISTRATE)
+    find_package(checkpoint REQUIRED)
+    target_link_libraries(resilience PUBLIC vt::lib::checkpoint)
+    target_compile_definitions(resilience PUBLIC KR_ENABLE_MAGISTRATE)
+endif()
+
 option( KR_ENABLE_STDIO "use stdio for manual checkpoint" OFF )
 option( KR_ENABLE_HDF5 "add HDF5 support" OFF )
 option( KR_ENABLE_HDF5_PARALLEL "use parallel version of HDF5" OFF )

cmake/resilienceConfig.cmake.in

@@ -9,7 +9,7 @@ SET(KR_ENABLE_HDF5 @KR_ENABLE_HDF5@)
 SET(KR_ENABLE_VELOC @KR_ENABLE_VELOC@)
 
 # VeloC needs to add a cmake config...
-LIST(APPEND CMAKE_MODULE_PATH "${CMAKE_CURRENT_LIST_DIR}/Modules/")
+LIST(APPEND CMAKE_MODULE_PATH "${CMAKE_CURRENT_LIST_DIR}/../Modules/" "${CMAKE_CURRENT_LIST_DIR}/../cmake/Modules/")
 message(STATUS "Module path: ${CMAKE_MODULE_PATH}")
 
 find_dependency(Kokkos REQUIRED NO_CMAKE_PACKAGE_REGISTRY HINTS @Kokkos_DIR@)
@@ -25,5 +25,11 @@ if (@KR_ENABLE_HDF5@)
    find_dependency(HDF5 REQUIRED)
 endif()
 
+if (@KR_ENABLE_MAGISTRATE@)
+  set(CHECKPOINT_DIR @CHECKPOINT_DIR@)
+   find_dependency(checkpoint REQUIRED)
+   set(KR_ENABLE_MAGISTRATE @KR_ENABLE_MAGISTRATE@)
+endif()
+
 set(Boost_DIR @Boost_DIR@)
 find_dependency(Boost REQUIRED)

examples/SimpleCheckpoint.cpp

@@ -46,9 +46,9 @@
 
 #include <mpi.h>
 #include <Kokkos_Core.hpp>
-#include <resilience/Context.hpp>
-#include <resilience/veloc/VelocBackend.hpp>
-#include <resilience/AutomaticCheckpoint.hpp>
+#include <resilience/Resilience.hpp>
+
+using chkpt_view = Kokkos::Experimental::SubscribableViewHooks<KokkosResilience::DynamicViewHooksSubscriber>;
 
 int
 main( int argc, char **argv )
@@ -60,15 +60,24 @@ main( int argc, char **argv )
     auto ctx = KokkosResilience::make_context( MPI_COMM_WORLD, "config.json" );
 
     int  dim0 = 5, dim1 = 5;
-    auto view = Kokkos::View< double ** >( "test_view", dim0, dim1 );
+    auto view = Kokkos::View< double **, chkpt_view>( "test_view", dim0, dim1 );
 
     KokkosResilience::checkpoint( *ctx, "test_checkpoint", 0, [view, dim0, dim1]() {
       Kokkos::parallel_for( dim0, KOKKOS_LAMBDA( int i ) {
         for ( int j = 0; j < dim1; ++j )
           view( i, j ) = 3.0;
       } );
-    } );
+    });
 
+    for(int i = 0; i < dim0; i++){
+      for(int j = 0; j < dim1; j++){
+        if(view(i,j) != 3.0) {
+          fprintf(stderr, "Error: view(%d,%d) = %f, not %f\n", i, j, view(i,j), 3.0);
+          exit(1);
+        }
+      }
+    }
+    printf("Success!\n");
   }
   Kokkos::finalize();
 

examples/SimpleFileCheckpoint.cpp

@@ -46,27 +46,42 @@
 #endif
 
 #include <Kokkos_Core.hpp>
-#include <resilience/Context.hpp>
-#include <resilience/stdfile/StdFileBackend.hpp>
-#include <resilience/AutomaticCheckpoint.hpp>
+#include <resilience/Resilience.hpp>
+#include <mpi.h>
+
+using chkpt_view = Kokkos::Experimental::SubscribableViewHooks<KokkosResilience::DynamicViewHooksSubscriber>;
 
 int
 main( int argc, char **argv )
 {
+  MPI_Init( &argc, &argv );
+
   Kokkos::initialize( argc, argv );
   {
-    auto ctx = KokkosResilience::make_context( "checkpoint.data", "config_file.json" );
+    auto ctx = KokkosResilience::make_context( MPI_COMM_WORLD, "config_file.json" );
 
     int  dim0 = 5, dim1 = 5;
-    auto view = Kokkos::View< double ** >( "test_view", dim0, dim1 );
+    auto view = Kokkos::View< double **, chkpt_view>( "test_view", dim0, dim1 );
 
     KokkosResilience::checkpoint( *ctx, "test_checkpoint", 0, [view, dim0, dim1]() {
       Kokkos::parallel_for( dim0, KOKKOS_LAMBDA( int i ) {
         for ( int j = 0; j < dim1; ++j )
           view( i, j ) = 3.0;
       } );
-    } );
+    }, [](int){return true;} );
+
+    for(int i = 0; i < dim0; i++){
+      for(int j = 0; j < dim1; j++){
+        if(view(i,j) != 3.0) {
+          fprintf(stderr, "Error: view(%d,%d) = %f, not %f\n", i, j, view(i,j), 3.0);
+          exit(1);
+        }
+      }
+    }
+    printf("Success!\n");
 
   }
   Kokkos::finalize();
+
+  MPI_Finalize();
 }

examples/benchmark_multiviews.cpp

@@ -137,7 +137,7 @@ int main(int argc, char *argv[]) {
       wtime = MPI_Wtime();
       std::size_t i = 1 + KokkosResilience::latest_version(*ctx, "test_kokkos");
 
-      while(i < nsteps) {
+      while(i < nsteps ) {
 
          KokkosResilience::checkpoint(*ctx, "test_kokkos", i, [=]() {   // Nic, tell me what should I put for []/
 

examples/config_file.json

@@ -2,11 +2,12 @@
   "backend": "stdfile",
   "backends": {
     "stdfile": {
-      "config": "file_test.cfg"
+      "directory": "./stdfile_chkpts/",
+      "filename_prefix": "simple_"
     }
   },
   "filter": {
     "type": "time",
     "interval": 10
   }
-}
+}