Add an option to use the mimalloc memory allocator

.gitignore

@@ -1,7 +1,10 @@
 CMakeSettings.json
+compile_commands.json
 build
+build_*
 install
+.cache/
 .vs/
+.vscode/
 .idea/
 /tags
-

.gitmodules

@@ -12,4 +12,7 @@
 	url = https://github.com/catchorg/Catch2.git
 [submodule "vendor/libenvpp"]
 	path = vendor/libenvpp
-	url = https://github.com/ph3at/libenvpp.git
+	url = https://github.com/ph3at/libenvpp.git
+[submodule "vendor/mimalloc"]
+	path = vendor/mimalloc
+	url = https://github.com/microsoft/mimalloc.git

CMakeLists.txt

@@ -9,6 +9,9 @@ message(VERBOSE "Celerity version is ${CELERITY_VERSION_MAJOR}.${CELERITY_VERSIO
 
 project(celerity_runtime VERSION ${Celerity_VERSION} LANGUAGES CXX)
 set_property(GLOBAL PROPERTY USE_FOLDERS ON)
+set(CMAKE_EXPORT_COMPILE_COMMANDS ON)
+
+option(CELERITY_USE_MIMALLOC "Use the mimalloc memory allocator" ON)
 
 set(CELERITY_CMAKE_DIR "${PROJECT_SOURCE_DIR}/cmake")
 set(CMAKE_MODULE_PATH "${CMAKE_MODULE_PATH}" "${CELERITY_CMAKE_DIR}")
@@ -62,7 +65,7 @@ else()
       "Please choose one implementation using -DCELERITY_SYCL_IMPL=hipSYCL|ComputeCpp|DPC++.")
   else()
     set(CELERITY_SYCL_IMPL "${AVAILABLE_SYCL_IMPLS}")
-    message(STATUS "Automatically chooosing ${CELERITY_SYCL_IMPL} because it is the only SYCL implementation available")
+    message(STATUS "Automatically choosing ${CELERITY_SYCL_IMPL} because it is the only SYCL implementation available")
   endif()
 endif()
 
@@ -184,6 +187,15 @@ list(APPEND CMAKE_MODULE_PATH ${catch2_SOURCE_DIR}/extras)
 set(LIBENVPP_INSTALL ON CACHE BOOL "" FORCE)
 fetch_content_from_submodule(libenvpp vendor/libenvpp)
 
+if(CELERITY_USE_MIMALLOC)
+  set(MI_OVERRIDE ON CACHE BOOL "" FORCE)
+  set(MI_BUILD_SHARED ON CACHE BOOL "" FORCE)
+  set(MI_BUILD_STATIC OFF CACHE BOOL "" FORCE)
+  set(MI_BUILD_OBJECT OFF CACHE BOOL "" FORCE)
+  set(MI_BUILD_TESTS OFF CACHE BOOL "" FORCE)
+  fetch_content_from_submodule(mimalloc vendor/mimalloc)
+endif()
+
 configure_file(include/version.h.in include/version.h @ONLY)
 
 # Add includes to library so they show up in IDEs
@@ -258,6 +270,10 @@ if(WIN32 AND CELERITY_SYCL_IMPL STREQUAL "DPC++")
   )
 endif()
 
+if(CELERITY_USE_MIMALLOC)
+  target_link_libraries(celerity_runtime PUBLIC mimalloc)
+endif()
+
 target_link_libraries(celerity_runtime PUBLIC
   Threads::Threads
   MPI::MPI_CXX
@@ -287,6 +303,7 @@ target_compile_definitions(celerity_runtime PUBLIC
   $<BUILD_INTERFACE:
     $<$<CONFIG:Debug>:CELERITY_DETAIL_ENABLE_DEBUG>
   >
+  CELERITY_USE_MIMALLOC=$<BOOL:${CELERITY_USE_MIMALLOC}>
   CELERITY_FEATURE_SCALAR_REDUCTIONS=$<BOOL:${CELERITY_FEATURE_SCALAR_REDUCTIONS}>
   CELERITY_FEATURE_SIMPLE_SCALAR_REDUCTIONS=$<BOOL:${CELERITY_FEATURE_SIMPLE_SCALAR_REDUCTIONS}>
   CELERITY_FEATURE_LOCAL_ACCESSOR=$<BOOL:${CELERITY_FEATURE_LOCAL_ACCESSOR}>

cmake/celerity-config.cmake.in

@@ -23,6 +23,7 @@ find_dependency(fmt REQUIRED)
 find_dependency(spdlog REQUIRED)
 find_dependency(small_vector REQUIRED)
 find_dependency(libenvpp REQUIRED)
+find_dependency(mimalloc REQUIRED)
 if(@CELERITY_ENABLE_CUDA_BACKEND@)
   if(CMAKE_VERSION VERSION_GREATER_EQUAL "3.17")
     find_dependency(CUDAToolkit REQUIRED)

src/runtime.cc

@@ -12,6 +12,11 @@
 
 #include <mpi.h>
 
+#if CELERITY_USE_MIMALLOC
+// override default new/delete operators to use the mimalloc memory allocator
+#include <mimalloc-new-delete.h>
+#endif
+
 #include "affinity.h"
 #include "buffer.h"
 #include "buffer_manager.h"
@@ -79,6 +84,14 @@ namespace detail {
 #endif
 	}
 
+	static const char* get_mimalloc_string() {
+#if CELERITY_USE_MIMALLOC
+		return "using mimalloc";
+#else
+		return "using the default allocator";
+#endif
+	}
+
 	static std::string get_sycl_version() {
 #if defined(__COMPUTECPP__)
 		return fmt::format("ComputeCpp {}.{}.{}", COMPUTECPP_VERSION_MAJOR, COMPUTECPP_VERSION_MINOR, COMPUTECPP_VERSION_PATCH);
@@ -145,8 +158,8 @@ namespace detail {
 			m_task_mngr->register_task_callback([this](const task* tsk) { m_schdlr->notify_task_created(tsk); });
 		}
 
-		CELERITY_INFO(
-		    "Celerity runtime version {} running on {}. PID = {}, build type = {}", get_version_string(), get_sycl_version(), get_pid(), get_build_type());
+		CELERITY_INFO("Celerity runtime version {} running on {}. PID = {}, build type = {}, {}", get_version_string(), get_sycl_version(), get_pid(),
+		    get_build_type(), get_mimalloc_string());
 		m_d_queue->init(*m_cfg, user_device_or_selector);
 	}