Cleanup mpi branch after merge/rebase

include/dlaf/init.h

@@ -38,13 +38,13 @@ struct configuration {
   bool print_config = false;
   std::size_t num_np_gpu_streams_per_thread = 3;
   std::size_t num_hp_gpu_streams_per_thread = 3;
-  std::size_t umpire_host_memory_pool_initial_block_bytes = 1 << 30;
-  std::size_t umpire_host_memory_pool_next_block_bytes = 1 << 30;
+  std::size_t umpire_host_memory_pool_initial_block_bytes = 1 << 16;
+  std::size_t umpire_host_memory_pool_next_block_bytes = 1 << 16;
   std::size_t umpire_host_memory_pool_alignment_bytes = 16;
   double umpire_host_memory_pool_coalescing_free_ratio = 1.0;
   double umpire_host_memory_pool_coalescing_reallocation_ratio = 1.0;
-  std::size_t umpire_device_memory_pool_initial_block_bytes = 1 << 30;
-  std::size_t umpire_device_memory_pool_next_block_bytes = 1 << 30;
+  std::size_t umpire_device_memory_pool_initial_block_bytes = 1 << 16;
+  std::size_t umpire_device_memory_pool_next_block_bytes = 1 << 16;
   std::size_t umpire_device_memory_pool_alignment_bytes = 16;
   double umpire_device_memory_pool_coalescing_free_ratio = 1.0;
   double umpire_device_memory_pool_coalescing_reallocation_ratio = 1.0;

include/dlaf/sender/transform_mpi.h

@@ -12,7 +12,6 @@
 #include <type_traits>
 #include <utility>
 
-#include <pika/debugging/print.hpp>
 #include <pika/execution.hpp>
 #include <pika/mpi.hpp>
 
@@ -25,9 +24,6 @@
 
 namespace dlaf::comm::internal {
 
-template <int Level>
-static pika::debug::detail::print_threshold<Level, 0> dla_debug("DLA_MPI");
-
 /// This helper "consumes" a CommunicatorPipelineExclusiveWrapper ensuring that after this call
 /// the one passed as argument gets destroyed. All other types left as they are
 /// by the second overload.
@@ -52,8 +48,6 @@ struct MPICallHelper {
 
   template <typename... Ts>
   auto operator()(Ts&&... ts) -> decltype(std::move(f)(dlaf::common::internal::unwrap(ts)...)) {
-    using namespace pika::debug::detail;
-    PIKA_DETAIL_DP(dla_debug<5>, debug(str<>("MPICallHelper"), pika::debug::print_type<Ts...>(", ")));
     using result_type = decltype(std::move(f)(dlaf::common::internal::unwrap(ts)...));
     if constexpr (std::is_void_v<result_type>) {
       std::move(f)(dlaf::common::internal::unwrap(ts)...);
@@ -74,16 +68,12 @@ MPICallHelper(F&&) -> MPICallHelper<std::decay_t<F>>;
 template <typename F, typename Sender,
           typename = std::enable_if_t<pika::execution::experimental::is_sender_v<Sender>>>
 [[nodiscard]] decltype(auto) transformMPI(F&& f, Sender&& sender) {
-  using dlaf::internal::continues_on;
   namespace ex = pika::execution::experimental;
   namespace mpi = pika::mpi::experimental;
-  namespace mpid = pika::mpi::experimental::detail;
-
-  PIKA_DETAIL_DP(dla_debug<5>, debug(str<>("MPI fn\n")));
-  auto snd1 =
-      std::forward<Sender>(sender) |
-      mpi::transform_mpi(dlaf::common::internal::ConsumeRvalues{MPICallHelper{std::forward<F>(f)}});
-  return ex::make_unique_any_sender(std::move(snd1));
+  namespace dci = dlaf::common::internal;
+  return std::forward<Sender>(sender)                                                  //
+         | mpi::transform_mpi(dci::ConsumeRvalues{MPICallHelper{std::forward<F>(f)}}); //
+         // @todo - check this | ex::drop_operation_state();
 }
 
 template <typename F>

src/init.cpp

@@ -45,6 +45,7 @@ std::ostream& operator<<(std::ostream& os, const configuration& cfg) {
   os << "  umpire_device_memory_pool_coalescing_reallocation_ratio = " << cfg.umpire_device_memory_pool_coalescing_reallocation_ratio << std::endl;
   os << "  num_gpu_blas_handles = " << cfg.num_gpu_blas_handles << std::endl;
   os << "  num_gpu_lapack_handles = " << cfg.num_gpu_lapack_handles << std::endl;
+  os << "  mpi_pool = " << pika::mpi::experimental::get_pool_name() << std::endl;
   // clang-format on
   return os;
 }