Merge branch 'llvm' into 'main' (c93a18db) · Commits · Cianciosa, Mark / graph_framework

CMakeLists.txt

+108 −0

Original line number	Diff line number	Diff line
		@@ -16,6 +16,9 @@ set (CMAKE_MODULE_PATH ${CMAKE_MODULE_PATH};${CMAKE_CURRENT_SOURCE_DIR}/cmake)
		#-------------------------------------------------------------------------------
		# Setup build types
		#-------------------------------------------------------------------------------
		if (NOT CMAKE_BUILD_TYPE)
		set (CMAKE_BUILD_TYPE Release CACHE STRING "Choose the type of build." FORCE)
		endif ()
		set_property (CACHE CMAKE_BUILD_TYPE PROPERTY STRINGS
		Debug
		Release
		@@ -117,6 +120,111 @@ target_link_options (sanitizer
		$<$<AND:$<CONFIG:Sanitized>,$<PLATFORM_ID:Darwin>>:-fsanitize-trap=all>
		)

		#-------------------------------------------------------------------------------
		# Setup access method.
		#-------------------------------------------------------------------------------
		option (USE_SSH "Use ssh to access git repos" OFF)
		if (${USE_SSH})
		set (URL_PROTO "git@")
		set (URL_SEP ":")
		else ()
		set (URL_PROTO "https://")
		set (URL_SEP "/")
		endif ()

		#-------------------------------------------------------------------------------
		# Define a macro to register new projects.
		#-------------------------------------------------------------------------------
		include (FetchContent)
		find_package (Git)

		function (register_project reg_name dir url default_tag sub_dir)
		set (BUILD_TAG_${dir} ${default_tag} CACHE STRING "Name of the tag to checkout.")

		# Anything after the tag is assumed to be a project setting.
		foreach (i RANGE 6 ${ARGC} 3)
		math (EXPR key "${i} - 1" OUTPUT_FORMAT DECIMAL)
		math (EXPR type "${i} + 1" OUTPUT_FORMAT DECIMAL)
		set (${ARGV${key}} ${ARGV${i}} CACHE ${ARGV${type}} "" FORCE)
		endforeach ()

		FetchContent_Declare (
		${reg_name}
		GIT_REPOSITORY ${url}
		GIT_TAG origin/${BUILD_TAG_${dir}}
		SOURCE_DIR ${CMAKE_CURRENT_SOURCE_DIR}/${dir}
		SOURCE_SUBDIR ${sub_dir}
		)

		FetchContent_MakeAvailable (${reg_name})

		if (${GIT_FOUND})
		# By default cmake clones projects in a headless state. After the repo is
		# cloned checkout the requested tag so repo is in a working state.
		execute_process (
		COMMAND ${GIT_EXECUTABLE} checkout ${BUILD_TAG_${dir}}
		WORKING_DIRECTORY ${${reg_name}_SOURCE_DIR}
		)

		# Add a taraget to pull the latest version before building. Note dependency is
		# registered in the sub project CMakeList.txt. Not sure how this should handle
		# multiple targets in a project yet. Name must match the subproject pull_
		# dependency.
		add_custom_target (
		pull_${reg_name}
		ALL
		COMMAND ${GIT_EXECUTABLE} pull
		WORKING_DIRECTORY ${${reg_name}_SOURCE_DIR}
		)
		endif ()
		endfunction ()

		register_project (llvm
		LLVM
		${URL_PROTO}github.com${URL_SEP}llvm/llvm-project.git
		main
		llvm
		LLVM_BUILD_LLVM_DYLIB ON BOOL
		LLVM_TARGETS_TO_BUILD Native STRING
		)

		#-------------------------------------------------------------------------------
		# Define an llvm dependency target
		#-------------------------------------------------------------------------------
		FetchContent_GetProperties (
		llvm
		BINARY_DIR llvm_binary_dir
		SOURCE_DIR llvm_source_dir
		)

		# Do not build llvm until pull command is finished.
		add_dependencies (LLVMHello pull_llvm)
		add_dependencies (LLVMHello_exports pull_llvm)
		add_dependencies (LLVMDemangle pull_llvm)
		add_dependencies (Remarks_exports pull_llvm)
		add_dependencies (LLVMSupportBlake3 pull_llvm)
		add_dependencies (BugpointPasses_exports pull_llvm)
		add_dependencies (LTO_exports pull_llvm)
		add_dependencies (llvm-PerfectShuffle pull_llvm)
		add_dependencies (SecondLib pull_llvm)
		add_dependencies (DynamicLibraryLib pull_llvm)
		add_dependencies (PipSqueak pull_llvm)
		add_dependencies (obj.LLVMTableGenCommon pull_llvm)
		add_dependencies (llvm_vcsrevision_h pull_llvm)
		add_dependencies (benchmark pull_llvm)
		add_dependencies (benchmark_main pull_llvm)

		add_library (llvm_dep INTERFACE)
		target_include_directories (llvm_dep
		INTERFACE
		$<BUILD_INTERFACE:${llvm_source_dir}/llvm/include>
		$<BUILD_INTERFACE:${llvm_binary_dir}/include>
		)
		target_link_libraries (llvm_dep
		INTERFACE
		LLVM
		)

		#-------------------------------------------------------------------------------
		# Setup targets
		#-------------------------------------------------------------------------------

graph_framework.xcodeproj/project.pbxproj

+18 −2

Original line number	Diff line number	Diff line
		@@ -1349,7 +1349,10 @@
		GCC_WARN_UNUSED_FUNCTION = YES;
		GCC_WARN_UNUSED_VARIABLE = YES;
		HEADER_SEARCH_PATHS = /usr/local/include;
		LIBRARY_SEARCH_PATHS = /usr/local/lib;
		LIBRARY_SEARCH_PATHS = (
		/usr/local/lib,
		"build/_deps/llvm-build/lib",
		);
		MACOSX_DEPLOYMENT_TARGET = 14.0;
		MTL_ENABLE_DEBUG_INFO = INCLUDE_SOURCE;
		MTL_FAST_MATH = YES;
		@@ -1360,8 +1363,13 @@
		"-ld_classic",
		"-rpath",
		/usr/local/lib,
		"-lLLVM",
		);
		SDKROOT = macosx;
		USER_HEADER_SEARCH_PATHS = (
		LLVM/llvm/include,
		"build/_deps/llvm-build/include",
		);
		};
		name = Debug;
		};
		@@ -1421,7 +1429,10 @@
		GCC_WARN_UNUSED_FUNCTION = YES;
		GCC_WARN_UNUSED_VARIABLE = YES;
		HEADER_SEARCH_PATHS = /usr/local/include;
		LIBRARY_SEARCH_PATHS = /usr/local/lib;
		LIBRARY_SEARCH_PATHS = (
		/usr/local/lib,
		"build/_deps/llvm-build/lib",
		);
		MACOSX_DEPLOYMENT_TARGET = 14.0;
		MTL_ENABLE_DEBUG_INFO = NO;
		MTL_FAST_MATH = YES;
		@@ -1432,8 +1443,13 @@
		"-ld_classic",
		"-rpath",
		/usr/local/lib,
		"-lLLVM",
		);
		SDKROOT = macosx;
		USER_HEADER_SEARCH_PATHS = (
		LLVM/llvm/include,
		"build/_deps/llvm-build/include",
		);
		};
		name = Release;
		};

graph_framework/CMakeLists.txt

+4 −2

Original line number	Diff line number	Diff line
		@@ -13,13 +13,14 @@ target_compile_definitions (rays
		INTERFACE
		$<$<PLATFORM_ID:Darwin>:CXX="${CMAKE_CXX_COMPILER} -isysroot ${CMAKE_OSX_SYSROOT} -I${CMAKE_CURRENT_SOURCE_DIR} -std=gnu++2a">
		$<$<PLATFORM_ID:Darwin>:CXX_FLAGS="-g -fsanitize=undefined -fsanitize=float-divide-by-zero -fsanitize-trap=all">
		$<$<PLATFORM_ID:Linux>:CXX="${CMAKE_CXX_COMPILER} -I${CMAKE_CURRENT_SOURCE_DIR} -std=gnu++2a">
		$<$<PLATFORM_ID:Linux>:CXX_FLAGS="-g -fsanitize=undefined -fsanitize=float-divide-by-zero">
		$<$<PLATFORM_ID:Linux>:CXX="${CMAKE_CXX_COMPILER} -I${CMAKE_CURRENT_SOURCE_DIR} -std=gnu++2a -fPIC -fno-use-cxa-atexit">
		$<$<PLATFORM_ID:Linux>:CXX_FLAGS="-g"> #-fsanitize=undefined -fsanitize=float-divide-by-zero">
		EFIT_FILE="${CMAKE_CURRENT_SOURCE_DIR}/../graph_tests/efit.nc"
		VMEC_FILE="${CMAKE_CURRENT_SOURCE_DIR}/../graph_tests/vmec.nc"
		$<$<BOOL:${USE_CUDA}>:HEADER_DIR="$<BUILD_INTERFACE:${CMAKE_CURRENT_SOURCE_DIR}>">
		$<$<BOOL:${USE_STATIC}>:STATIC>
		)

		target_include_directories (rays
		INTERFACE
		$<BUILD_INTERFACE:${CMAKE_CURRENT_SOURCE_DIR}>
		@@ -31,6 +32,7 @@ target_link_libraries (rays
		NetCDF::NetCDF
		$<$<PLATFORM_ID:Linux>:${CMAKE_DL_LIBS}>
		$<$<PLATFORM_ID:Linux>:pthread>
		llvm_dep
		)
		target_precompile_headers (rays
		INTERFACE

graph_framework/absorption.hpp

+6 −5

Original line number	Diff line number	Diff line
		@@ -143,11 +143,12 @@ namespace absorption {
		inputs.push_back(graph::variable_cast(this->t));
		inputs.push_back(graph::variable_cast(this->w));

		dispersion::dispersion_interface<DISPERSION_FUNCTION> D(w,
		auto D = DISPERSION_FUNCTION().D(w,
		kx + kx_amp,
		ky + ky_amp,
		kz + kz_amp,
		x, y, z, t, eq);

		solver::newton(work, {kamp}, inputs, {D.get_d()});

		inputs = {

graph_framework/cpu_context.hpp

+43 −32

Original line number	Diff line number	Diff line
		@@ -13,12 +13,14 @@
		#include <cstring>
		#include <thread>

		#include <dlfcn.h>
		#include <unistd.h>
		#include <sys/wait.h>

		#include "node.hpp"

		#include "llvm/Support/TargetSelect.h"
		#include "llvm/ExecutionEngine/Orc/LLJIT.h"

		namespace gpu {
		//------------------------------------------------------------------------------
		/// @brief Split a string by the space delimiter.
		@@ -71,10 +73,10 @@ namespace gpu {
		template<jit::float_scalar T, bool SAFE_MATH=false>
		class cpu_context {
		private:
		/// Library name.
		std::string library_name;
		/// Code object name.
		std::string object_name;
		/// Handle for the dynamic library.
		void *lib_handle;
		std::unique_ptr<llvm::orc::LLJIT> jit;
		/// Argument map.
		std::map<graph::leaf_node<T, SAFE_MATH> *, std::vector<T>> kernel_arguments;
		/// Host buffer map.
		@@ -104,22 +106,22 @@ namespace gpu {
		///
		/// @params[in] index Concurrent index. Not used.
		//------------------------------------------------------------------------------
		cpu_context(const size_t index) {}
		cpu_context(const size_t index) {
		llvm::InitializeNativeTarget();
		}

		//------------------------------------------------------------------------------
		/// @brief Destruct a cpu context.
		//------------------------------------------------------------------------------
		~cpu_context() {
		dlclose(lib_handle);

		if (!library_name.empty()) {
		// A new instance of the class can be created before the library is deleted.
		if (!object_name.empty()) {
		// A new instance of the class can be created before the object is deleted.
		// Wait for the fork to finish before the destructor exits. This could cause a
		// problem where the new library gets deleted before the context tries to load
		// problem where the new object gets deleted before the context tries to load
		// it.
		auto pid = fork();
		if (pid == 0) {
		execlp("rm", "rm", library_name.c_str(), NULL);
		execlp("rm", "rm", object_name.c_str(), NULL);
		exit(0);
		}
		int error = 0;
		@@ -155,22 +157,18 @@ namespace gpu {
		temp_stream.str(std::string());
		temp_stream.clear();

		temp_stream << "./" << filename << ".so";
		library_name = temp_stream.str();
		temp_stream << "temp_" << thread_id << ".o";
		object_name = temp_stream.str();

		temp_stream.str(std::string());
		temp_stream.clear();
		#ifdef __APPLE__
		temp_stream << CXX << " -dynamiclib -flat_namespace ";
		#else
		temp_stream << CXX << " -fPIC -shared ";
		#endif
		temp_stream << CXX << " ";
		#ifndef NDEBUG
		temp_stream << CXX_FLAGS << " ";
		#else
		temp_stream << "-O3 ";
		#endif
		temp_stream << filename << " -o " << library_name;
		temp_stream << "-c " << filename;

		if (jit::verbose) {
		std::cout << "CPU info." << std::endl;
		@@ -190,28 +188,39 @@ namespace gpu {
		if (error) {
		std::cerr << "Failed to compile cpu kernel. Check source code in "
		<< filename << std::endl;
		exit(error);
		exit(-1);
		}

		#ifdef NDEBUG
		if (fork() == 0) {
		pid = fork();
		if (pid == 0) {
		execlp("rm", "rm", filename.c_str(), NULL);
		exit(0);
		}
		#endif

		lib_handle = dlopen(library_name.c_str(), RTLD_LAZY);
		if (!lib_handle) {
		std::cout << "Failed to load library. " << library_name
		<< std::endl;
		std::cout << dlerror() << std::endl;
		exit(1);
		auto object = llvm::MemoryBuffer::getFileAsStream(object_name);

		auto jit_try = llvm::orc::LLJITBuilder().create();
		if (auto jiterror = jit_try.takeError()) {
		std::cerr << "Failed to build JIT : " << toString(std::move(jiterror)) << std::endl;
		exit(-1);
		}
		jit = std::move(jit_try.get());

		jit->addObjectFile(std::move(object.get()));

		if (jit::verbose) {
		std::cout << " Library name : " << library_name << std::endl;
		std::cout << " Library handle : " << reinterpret_cast<size_t> (lib_handle) << std::endl;
		std::cout << " Object name : " << object_name << std::endl;
		}

		#ifdef NDEBUG
		waitpid(pid, &error, 0);
		if (error) {
		std::cerr << "Failed to remove " << filename << std::endl;
		exit(-1);
		}
		#endif
		}

		//------------------------------------------------------------------------------
		@@ -227,7 +236,9 @@ namespace gpu {
		graph::input_nodes<T, SAFE_MATH> inputs,
		graph::output_nodes<T, SAFE_MATH> outputs,
		const size_t num_rays) {
		void *kernel = dlsym(lib_handle, kernel_name.c_str());
		auto entry = std::move(jit->lookup(kernel_name)).get();
		auto kernel = entry.toPtr<void()(std::map<size_t, T > &)> ();

		if (!kernel) {
		std::cerr << "Failed to load function. " << kernel_name
		<< std::endl;
		@@ -258,7 +269,7 @@ namespace gpu {
		}

		return [kernel, buffers] () mutable {
		((void ()(std::map<size_t, T > &))kernel)(buffers);
		kernel(buffers);
		};
		}

		@@ -392,7 +403,7 @@ namespace gpu {
		source_buffer << std::endl;
		source_buffer << "extern \"C\" void " << name << "(" << std::endl;

		source_buffer << " std::map<size_t, ";
		source_buffer << " map<size_t, ";
		jit::add_type<T> (source_buffer);
		source_buffer << " *> &args) {" << std::endl;