remove compilation flag use_kokkos and vector version (#265)

* [remove] vector macro in datatype

* [remove] compiler option in cmake

* [remove] use_kokkos flag removed from code

Author: ABandet

CMakeLists.txt

@@ -5,11 +5,6 @@ cmake_minimum_required( VERSION 3.22.1 )
 #-----------------------------------------------------
 set(CMAKE_VERBOSE_MAKEFILE OFF)
 
-#--------------------------------------------------------
-# Check cmake option conflicts
-#--------------------------------------------------------
-include(cmake/CheckOptions.cmake)
-
 #--------------------------------------------------------
 # Configure compiler/project
 #--------------------------------------------------------
@@ -22,7 +17,6 @@ include(cmake/ProxyConfig.cmake)    # This must go *before* other configurations
 #--------------------------------------------------------
 include(cmake/PywrapConfig.cmake)   # This must go *before* kokkos and adios configurations
 include(cmake/MPIConfig.cmake)      # This must go *before* adios configuration
-include(cmake/KokkosConfig.cmake)
 include(cmake/ADIOS2Config.cmake)
 include(cmake/GoogleTestConfig.cmake)
 include(cmake/GoogleBenchConfig.cmake)

cmake/CheckOptions.cmake

@@ -2,9 +2,9 @@
 # Kokkos related functions
 #-------------------------------------------------------------------
 
+find_package(Kokkos REQUIRED)
+
 function(target_link_kokkos_if_enabled target_name)
-  if (USE_KOKKOS)
-    target_link_libraries(${target_name} PUBLIC Kokkos::kokkos)
-    target_compile_definitions(${target_name} PUBLIC USE_KOKKOS)
-  endif()
+  target_link_libraries(${target_name} PUBLIC Kokkos::kokkos)
+  target_compile_definitions(${target_name} PUBLIC USE_KOKKOS)
 endfunction()

cmake/KokkosConfig.cmake

@@ -7,12 +7,7 @@ option(COMPILE_SEM "Compile Spectral Elements Method simulation" ON)
 option(COMPILE_FD "Compile finite elements simulation" ON)
 
 # Programming models
-option(USE_VECTOR "Use vectors." OFF)
-option(USE_KOKKOS "Use KOKKOS to parallelise loops" OFF)
-option(USE_KOKKOS_TEAMS "use hierarchical parallelism in Kokkos" OFF)
-option(ENABLE_CUDA "Enable cuda compilation" OFF)
 option(USE_MPI "Enable MPI compilation" OFF)
-option(ENABLE_HIP "Enable hip compilation" OFF)
 
 # Python wrapping
 option(ENABLE_PYWRAP "Enable python binding compilation with pybind11" OFF)

cmake/KokkosUtils.cmake

@@ -12,13 +12,6 @@ function(print_configuration_summary)
   message(STATUS "  COMPILE_FD:           ${COMPILE_FD}")
   message(STATUS "")
 
-  message(STATUS "Programming Models:")
-  message(STATUS "  USE_VECTOR:           ${USE_VECTOR}")
-  message(STATUS "  USE_KOKKOS:           ${USE_KOKKOS}")
-  message(STATUS "  ENABLE_CUDA:          ${ENABLE_CUDA}")
-  message(STATUS "  USE_KOKKOS_TEAMS:     ${USE_KOKKOS_TEAMS}")
-  message(STATUS "")
-
   message(STATUS "Python Wrapping:")
   message(STATUS "  ENABLE_PYWRAP:        ${ENABLE_PYWRAP}")
   message(STATUS "")

cmake/ProxyConfig.cmake

@@ -8,27 +8,10 @@
 #define IDX3_eta1(i, j, k) \
   ((nz + 2) * (ny + 2) * ((i) + 1) + (nz + 2) * ((j) + 1) + ((k) + 1))
 
-#if defined(USE_KOKKOS)
 #define LOOP3DHEAD(x3, y3, z3, x4, y4, z4) Kokkos::parallel_for(Kokkos::MDRangePolicy<Kokkos::Rank<3>>({z3,x3,y3},{z4,x4,y4}),KOKKOS_LAMBDA(int k,int i,int j) {
-#else
-#define LOOP3DHEAD(x3, y3, z3, x4, y4, z4) \
-  for (int i = x3; i < x4; ++i)            \
-  {                                        \
-    for (int j = y3; j < y4; ++j)          \
-    {                                      \
-      for (int k = z3; k < z4; ++k)        \
-      {
-#endif
 
-#if defined(USE_KOKKOS)
 #define LOOP3DEND \
   });
-#else
-#define LOOP3DEND \
-  }               \
-  }               \
-  }
-#endif
 
 #define CREATEVIEWINNER              \
   vectorReal coefx = myModels.coefx; \
@@ -47,9 +30,5 @@
 #define CREATEVIEWSPONGE vectorReal spongeArray = myModels.spongeArray;
 #define PN_Global pnGlobal
 
-#if defined(USE_KOKKOS)
 #define FDFENCE Kokkos::fence();
-#else
-#define FDFENCE
-#endif
 #endif  // FUNTIDES_DISCRETIZATION_FD_KERNELS_INCLUDE_FD_MACROS_H_

cmake/ProxyUtils.cmake

@@ -19,9 +19,7 @@
 #include <iostream>
 #include <memory>
 
-#ifdef USE_KOKKOS
 #include <Kokkos_Core.hpp>
-#endif
 
 #include "fd_options.h"
 #include "fd_proxy.h"
@@ -37,7 +35,6 @@ using std::chrono::time_point;
 // Global start time for total execution timing
 time_point<system_clock> g_start_init_time;
 
-#ifdef USE_KOKKOS
 /**
  * @brief RAII wrapper for Kokkos initialization/finalization.
  */
@@ -56,7 +53,6 @@ class KokkosScope
   KokkosScope(KokkosScope&&) = delete;
   KokkosScope& operator=(KokkosScope&&) = delete;
 };
-#endif
 
 /**
  * @brief Converts nanoseconds duration to seconds as a double.
@@ -120,11 +116,9 @@ void Compute(fdtd::FdtdProxy& fd_sim)
  */
 void ConfigureParallelEnvironment()
 {
-#ifdef USE_KOKKOS
   // Configure OpenMP thread binding for optimal performance
   setenv("OMP_PROC_BIND", "spread", 0);  // Don't override if already set
   setenv("OMP_PLACES", "threads", 0);
-#endif
 }
 
 }  // namespace
@@ -147,11 +141,9 @@ int main(int argc, char* argv[])
   // Configure parallel execution environment
   ConfigureParallelEnvironment();
 
-#ifdef USE_KOKKOS
   // RAII: Kokkos will be automatically finalized when this object goes out of
   // scope
   KokkosScope kokkos_scope(argc, argv);
-#endif
 
   try
   {

src/discretization/fd/kernels/include/fd_macros.h

@@ -73,13 +73,10 @@ int main(int argc, char *argv[])
   MPI_Comm_size(MPI_COMM_WORLD, &size);
 #endif
 
-#ifdef USE_KOKKOS
   setenv("OMP_PROC_BIND", "spread", 1);
   setenv("OMP_PLACES", "threads", 1);
   Kokkos::initialize(argc, argv);
   {
-#endif
-
     cxxopts::Options options("SEM Proxy", "Runs the SEM simulation.");
     options.allow_unrecognised_options();  // lets Kokkos flags pass
 
@@ -114,10 +111,8 @@ int main(int argc, char *argv[])
     SEMproxy semsim(opt);
 
     compute_loop(semsim);
-#ifdef USE_KOKKOS
   }
   Kokkos::finalize();
-#endif
 
 #ifdef USE_MPI
   MPI_Barrier(MPI_COMM_WORLD);

src/main/fd/src/main.cc

@@ -328,21 +328,9 @@ void SEMproxy::run()
 
     for (int i = 0; i < pnAtReceiver.extent(0); i++)
     {
-#ifdef USE_KOKKOS
       auto subview = Kokkos::subview(pnAtReceiver, i, Kokkos::ALL());
       vectorReal subset("receiver_save", num_sample_);
       Kokkos::deep_copy(subset, subview);
-#else
-      auto& subview = pnAtReceiver;
-      vectorReal subset(subview.extent(0) * subview.extent(1));
-      for (size_t k = 0; k < subview.extent(0); ++k)
-      {
-        for (size_t j = 0; j < subview.extent(1); ++j)
-        {
-          subset[k * subview.extent(1) + j] = subview(k, j);
-        }
-      }
-#endif
       io_ctrl_->saveReceiver(subset, src_coord_);
     }
 
@@ -461,21 +449,9 @@ void SEMproxy::run()
 
     for (int i = 0; i < uxnAtReceiver.extent(0); i++)
     {
-#ifdef USE_KOKKOS
       auto subview = Kokkos::subview(uxnAtReceiver, i, Kokkos::ALL());
       vectorReal subset("receiver_save", num_sample_);
       Kokkos::deep_copy(subset, subview);
-#else
-      auto& subview = pnAtReceiver;
-      vectorReal subset(subview.extent(0) * subview.extent(1));
-      for (size_t k = 0; k < subview.extent(0); ++k)
-      {
-        for (size_t j = 0; j < subview.extent(1); ++j)
-        {
-          subset[k * subview.extent(1) + j] = subview(k, j);
-        }
-      }
-#endif  // USE_KOKKOS
       io_ctrl_->saveReceiver(subset, src_coord_);
     }
 
@@ -887,7 +863,6 @@ void SEMproxy::init_source()
 
 void SEMproxy::saveSnapshot(int timestep, VECTOR_REAL_VIEW data) const
 {
-#ifdef USE_KOKKOS
   Kokkos::fence();
   auto nb_nodes = data.extent(0);
 
@@ -896,10 +871,6 @@ void SEMproxy::saveSnapshot(int timestep, VECTOR_REAL_VIEW data) const
   Kokkos::parallel_for(
       "copy_column", nb_nodes, KOKKOS_LAMBDA(int i) { subset(i) = data(i); });
   Kokkos::fence();
-#else
-  auto& subset = data;
-#endif  // USE_KOKKOS
-
   io_ctrl_->saveSnapshot(subset, timestep);
 }