Enable MPI tracing for Fortran (#185)

- Move the MPI gotcha functionality from Timemory to the repo.
- Add the PMPI Fortran MPI functions to the existing mpi gotcha handle.

[ROCm/rocprofiler-systems commit: 4fcd8cc]

Author: sputhala-amd

projects/rocprofiler-systems/cmake/Packages.cmake

@@ -210,13 +210,12 @@ set(_ROCPROFSYS_MPI_HEADERS_ALLOW_MPICH ${MPI_HEADERS_ALLOW_MPICH})
 if(ROCPROFSYS_USE_MPI)
     find_package(MPI ${rocprofiler_systems_FIND_QUIETLY} REQUIRED)
     target_link_libraries(rocprofiler-systems-mpi INTERFACE MPI::MPI_C MPI::MPI_CXX)
-    rocprofiler_systems_target_compile_definitions(
-        rocprofiler-systems-mpi INTERFACE TIMEMORY_USE_MPI=1 ROCPROFSYS_USE_MPI)
+    rocprofiler_systems_target_compile_definitions(rocprofiler-systems-mpi
+                                                   INTERFACE ROCPROFSYS_USE_MPI)
 elseif(ROCPROFSYS_USE_MPI_HEADERS)
     find_package(MPI-Headers ${rocprofiler_systems_FIND_QUIETLY} REQUIRED)
-    rocprofiler_systems_target_compile_definitions(
-        rocprofiler-systems-mpi INTERFACE TIMEMORY_USE_MPI_HEADERS=1
-                                          ROCPROFSYS_USE_MPI_HEADERS)
+    rocprofiler_systems_target_compile_definitions(rocprofiler-systems-mpi
+                                                   INTERFACE ROCPROFSYS_USE_MPI_HEADERS)
     target_link_libraries(rocprofiler-systems-mpi INTERFACE MPI::MPI_HEADERS)
 endif()
 
@@ -543,9 +542,6 @@ set(TIMEMORY_QUIET_CONFIG
     CACHE BOOL "Make timemory configuration quieter")
 
 # timemory feature settings
-set(TIMEMORY_USE_MPI
-    ${ROCPROFSYS_USE_MPI}
-    CACHE BOOL "Enable MPI support in timemory" FORCE)
 set(TIMEMORY_USE_GOTCHA
     ON
     CACHE BOOL "Enable GOTCHA support in timemory")

projects/rocprofiler-systems/examples/transpose/CMakeLists.txt

@@ -41,7 +41,7 @@ if((NOT CMAKE_CXX_COMPILER_IS_HIPCC OR (NOT CMAKE_CXX_COMPILER_ID MATCHES "Clang
     return()
 endif()
 
-option(TRANSPOSE_USE_MPI "Enable MPI support in transpose exe" ${TIMEMORY_USE_MPI})
+option(TRANSPOSE_USE_MPI "Enable MPI support in transpose exe" ${ROCPROFSYS_USE_MPI})
 
 find_package(Threads REQUIRED)
 if(TRANSPOSE_USE_MPI)

projects/rocprofiler-systems/source/bin/rocprof-sys-instrument/rocprof-sys-instrument.cpp

@@ -1711,8 +1711,10 @@ main(int argc, char** argv)
                                          { "rocprofsys_user_stop_thread_trace" });
 #if ROCPROFSYS_USE_MPI > 0 || ROCPROFSYS_USE_MPI_HEADERS > 0
     // if any of the below MPI functions are found, enable MPI support
-    for(const auto* itr : { "MPI_Init", "MPI_Init_thread", "MPI_Finalize",
-                            "MPI_Comm_rank", "MPI_Comm_size" })
+    for(const auto* itr :
+        { "MPI_Init", "MPI_Init_thread", "MPI_Finalize", "MPI_Comm_rank", "MPI_Comm_size",
+          "MPI_INIT", "mpi_init", "mpi_init_", "mpi_init__", "MPI_INIT_THREAD",
+          "mpi_init_thread", "mpi_init_thread_", "mpi_init_thread__" })
     {
         if(find_function(app_image, itr) != nullptr)
         {

projects/rocprofiler-systems/source/lib/core/CMakeLists.txt

@@ -31,6 +31,7 @@ set(core_headers
     ${CMAKE_CURRENT_LIST_DIR}/exception.hpp
     ${CMAKE_CURRENT_LIST_DIR}/gpu.hpp
     ${CMAKE_CURRENT_LIST_DIR}/locking.hpp
+    ${CMAKE_CURRENT_LIST_DIR}/mpi.hpp
     ${CMAKE_CURRENT_LIST_DIR}/mproc.hpp
     ${CMAKE_CURRENT_LIST_DIR}/perf.hpp
     ${CMAKE_CURRENT_LIST_DIR}/perfetto.hpp

projects/rocprofiler-systems/source/lib/core/config.cpp

@@ -1046,7 +1046,7 @@ configure_settings(bool _init)
     settings::use_output_suffix() = _config->get<bool>("ROCPROFSYS_USE_PID");
     if(settings::use_output_suffix())
         settings::default_process_suffix() = process::get_id();
-#if !defined(TIMEMORY_USE_MPI) && defined(TIMEMORY_USE_MPI_HEADERS)
+#if !defined(ROCPROFSYS_USE_MPI) && defined(ROCPROFSYS_USE_MPI_HEADERS)
     if(tim::dmp::is_initialized()) settings::default_process_suffix() = tim::dmp::rank();
 #endif
 
@@ -1367,7 +1367,7 @@ configure_disabled_settings(const std::shared_ptr<settings>& _config)
         _config->find(itr)->second->set_hidden(true);
 #endif
 
-#if !defined(TIMEMORY_USE_MPI) || TIMEMORY_USE_MPI == 0
+#if !defined(ROCPROFSYS_USE_MPI) || ROCPROFSYS_USE_MPI == 0
     _config->disable("ROCPROFSYS_PERFETTO_COMBINE_TRACES");
     _config->disable("ROCPROFSYS_COLLAPSE_PROCESSES");
     _config->find("ROCPROFSYS_PERFETTO_COMBINE_TRACES")->second->set_hidden(true);
@@ -1991,7 +1991,7 @@ get_perfetto_buffer_size()
 bool
 get_perfetto_combined_traces()
 {
-#if defined(TIMEMORY_USE_MPI) && TIMEMORY_USE_MPI > 0
+#if defined(ROCPROFSYS_USE_MPI) && ROCPROFSYS_USE_MPI > 0
     static auto _v = get_config()->find("ROCPROFSYS_PERFETTO_COMBINE_TRACES");
     return static_cast<tim::tsettings<bool>&>(*_v->second).get();
 #else

projects/rocprofiler-systems/source/lib/core/debug.hpp

@@ -159,9 +159,9 @@ as_hex<void*>(void*, size_t);
 #endif
 
 #if !defined(ROCPROFSYS_DEBUG_PROCESS_IDENTIFIER)
-#    if defined(TIMEMORY_USE_MPI)
+#    if defined(ROCPROFSYS_USE_MPI)
 #        define ROCPROFSYS_DEBUG_PROCESS_IDENTIFIER static_cast<int>(::tim::dmp::rank())
-#    elif defined(TIMEMORY_USE_MPI_HEADERS)
+#    elif defined(ROCPROFSYS_USE_MPI_HEADERS)
 #        define ROCPROFSYS_DEBUG_PROCESS_IDENTIFIER                                      \
             (::tim::dmp::is_initialized()) ? static_cast<int>(::tim::dmp::rank())        \
                                            : static_cast<int>(::tim::process::get_id())