Add OpenCL support to CMake build (8a06a7bc) · Commits · 郑智淋 / lammps

cmake/CMakeLists.txt

+107 −63

Original line number	Diff line number	Diff line
		@@ -491,19 +491,31 @@ if(ENABLE_USER-INTEL)
		endif()

		if(ENABLE_GPU)
		set(GPU_SOURCES_DIR ${LAMMPS_SOURCE_DIR}/GPU)
		set(GPU_SOURCES ${GPU_SOURCES_DIR}/gpu_extra.h
		${GPU_SOURCES_DIR}/fix_gpu.h
		${GPU_SOURCES_DIR}/fix_gpu.cpp)

		set(GPU_API "OpenCL" CACHE STRING "API used by GPU package")
		set_property(CACHE GPU_API PROPERTY STRINGS OpenCL CUDA)

		set(GPU_PREC "SINGLE_DOUBLE" CACHE STRING "LAMMPS GPU precision size")
		set_property(CACHE GPU_PREC PROPERTY STRINGS SINGLE_DOUBLE SINGLE_SINGLE DOUBLE_DOUBLE)

		file(GLOB GPU_LIB_SOURCES ${LAMMPS_LIB_SOURCE_DIR}/gpu/*.cpp)
		file(MAKE_DIRECTORY ${LAMMPS_LIB_BINARY_DIR}/gpu)

		if(GPU_API STREQUAL "CUDA")
		find_package(CUDA REQUIRED)
		find_program(BIN2C bin2c)
		if(NOT BIN2C)
		message(FATAL_ERROR "Couldn't find bin2c, use -DBIN2C helping cmake to find it.")
		endif()
		set(GPU_PREC "SINGLE_DOUBLE" CACHE STRING "Lammps gpu precision size")
		set_property(CACHE GPU_PREC PROPERTY STRINGS SINGLE_DOUBLE SINGLE_SINGLE DOUBLE_DOUBLE)
		option(CUDPP_OPT "Enable CUDPP_OPT" ON)

		set(GPU_ARCH "sm_30" CACHE STRING "LAMMPS GPU CUDA SM architecture")
		set_property(CACHE GPU_ARCH PROPERTY STRINGS sm_10 sm_20 sm_30 sm_60)

		file(GLOB GPU_LIB_SOURCES ${LAMMPS_LIB_SOURCE_DIR}/gpu/*.cpp)
		file(GLOB GPU_LIB_CU ${LAMMPS_LIB_SOURCE_DIR}/gpu/.cu ${CMAKE_SOURCE_DIR}/gpu/.cu)
		list(REMOVE_ITEM GPU_LIB_CU ${LAMMPS_LIB_SOURCE_DIR}/gpu/lal_pppm.cu)

		@@ -521,7 +533,6 @@ if(ENABLE_GPU)
		cuda_compile(GPU_OBJS ${GPU_LIB_CUDPP_CU} OPTIONS $<$<BOOL:${BUILD_SHARED_LIBS}>:-Xcompiler=-fPIC>
		-DUNIX -O3 -Xptxas -v --use_fast_math -DUCL_CUDADR -arch=${GPU_ARCH} -D_${GPU_PREC})

		file(MAKE_DIRECTORY ${LAMMPS_LIB_BINARY_DIR}/gpu)
		foreach(CU_OBJ ${GPU_GEN_OBJS})
		get_filename_component(CU_NAME ${CU_OBJ} NAME_WE)
		string(REGEX REPLACE "^.*_lal_" "" CU_NAME "${CU_NAME}")
		@@ -535,6 +546,7 @@ if(ENABLE_GPU)


		add_library(gpu STATIC ${GPU_LIB_SOURCES} ${GPU_LIB_CUDPP_SOURCES} ${GPU_OBJS})
		target_link_libraries(gpu ${CUDA_LIBRARIES} ${CUDA_CUDA_LIBRARY})
		target_include_directories(gpu PRIVATE ${LAMMPS_LIB_BINARY_DIR}/gpu ${CUDA_INCLUDE_DIRS})
		target_compile_definitions(gpu PRIVATE -D_${GPU_PREC} -DMPI_GERYON -DUCL_NO_EXIT)
		if(CUDPP_OPT)
		@@ -550,13 +562,45 @@ if(ENABLE_GPU)
		target_include_directories(nvc_get_devices PRIVATE ${CUDA_INCLUDE_DIRS})


		# GPU package
		target_link_libraries(gpu ${CUDA_LIBRARIES} ${CUDA_CUDA_LIBRARY})
		elseif(GPU_API STREQUAL "OpenCL")
		find_package(OpenCL REQUIRED)
		if(OpenCL_FOUND)
		set(OCL_TUNE "GENERIC" CACHE STRING "OpenCL Device Tuning")
		set_property(CACHE OCL_TUNE PROPERTY STRINGS INTEL FERMI KEPLER CYPRESS GENERIC)

		set(GPU_SOURCES_DIR ${LAMMPS_SOURCE_DIR}/GPU)
		set(GPU_SOURCES ${GPU_SOURCES_DIR}/gpu_extra.h
		${GPU_SOURCES_DIR}/fix_gpu.h
		${GPU_SOURCES_DIR}/fix_gpu.cpp)
		include(OpenCLUtils)
		set(OCL_COMMON_HEADERS ${LAMMPS_LIB_SOURCE_DIR}/gpu/lal_preprocessor.h ${LAMMPS_LIB_SOURCE_DIR}/gpu/lal_aux_fun1.h)

		file(GLOB GPU_LIB_CU ${LAMMPS_LIB_SOURCE_DIR}/gpu/*.cu)
		list(REMOVE_ITEM GPU_LIB_CU ${LAMMPS_LIB_SOURCE_DIR}/gpu/lal_gayberne.cu ${LAMMPS_LIB_SOURCE_DIR}/gpu/lal_gayberne_lj.cu)

		foreach(GPU_KERNEL ${GPU_LIB_CU})
		get_filename_component(basename ${GPU_KERNEL} NAME_WE)
		string(SUBSTRING ${basename} 4 -1 KERNEL_NAME)
		GenerateOpenCLHeader(${KERNEL_NAME} ${CMAKE_CURRENT_BINARY_DIR}/gpu/${KERNEL_NAME}_cl.h ${OCL_COMMON_HEADERS} ${GPU_KERNEL})
		list(APPEND GPU_LIB_SOURCES ${CMAKE_CURRENT_BINARY_DIR}/gpu/${KERNEL_NAME}_cl.h)
		endforeach()

		GenerateOpenCLHeader(gayberne ${CMAKE_CURRENT_BINARY_DIR}/gpu/gayberne_cl.h ${OCL_COMMON_HEADERS} ${LAMMPS_LIB_SOURCE_DIR}/gpu/lal_ellipsoid_extra.h ${LAMMPS_LIB_SOURCE_DIR}/gpu/lal_gayberne.cu)
		GenerateOpenCLHeader(gayberne_lj ${CMAKE_CURRENT_BINARY_DIR}/gpu/gayberne_lj_cl.h ${OCL_COMMON_HEADERS} ${LAMMPS_LIB_SOURCE_DIR}/gpu/lal_ellipsoid_extra.h ${LAMMPS_LIB_SOURCE_DIR}/gpu/lal_gayberne_lj.cu)
		list(APPEND GPU_LIB_SOURCES ${CMAKE_CURRENT_BINARY_DIR}/gpu/gayberne_cl.h ${CMAKE_CURRENT_BINARY_DIR}/gpu/gayberne_lj_cl.h)

		add_library(gpu STATIC ${GPU_LIB_SOURCES})
		target_link_libraries(gpu ${OpenCL_LIBRARIES})
		target_include_directories(gpu PRIVATE ${CMAKE_CURRENT_BINARY_DIR}/gpu ${OpenCL_INCLUDE_DIRS})
		target_compile_definitions(gpu PRIVATE -D_${GPU_PREC} -DMPI_GERYON -DUCL_NO_EXIT)
		target_compile_definitions(gpu PRIVATE -DUSE_OPENCL)

		list(APPEND LAMMPS_LINK_LIBS gpu)

		add_executable(ocl_get_devices ${LAMMPS_LIB_SOURCE_DIR}/gpu/geryon/ucl_get_devices.cpp)
		target_compile_definitions(ocl_get_devices PRIVATE -DUCL_OPENCL)
		target_link_libraries(ocl_get_devices PRIVATE ${OpenCL_LIBRARIES})
		target_include_directories(ocl_get_devices PRIVATE ${OpenCL_INCLUDE_DIRS})
		endif()
		endif()

		# GPU package
		FindStyleHeaders(${GPU_SOURCES_DIR} FIX_CLASS fix_ FIX)

		set_property(GLOBAL PROPERTY "GPU_SOURCES" "${GPU_SOURCES}")

cmake/Modules/OpenCLUtils.cmake

0 → 100644

+18 −0

Original line number	Diff line number	Diff line
		function(GenerateOpenCLHeader varname outfile files)
		message("Creating ${outfile}...")
		file(WRITE ${outfile} "const char * ${varname} = \n")
		math(EXPR ARG_END "${ARGC}-1")

		foreach(IDX RANGE 2 ${ARG_END})
		list(GET ARGV ${IDX} filename)
		file(READ ${filename} content)
		string(REGEX REPLACE "\\s//[^\n]\n" "" content "${content}")
		string(REGEX REPLACE "\\\\" "\\\\\\\\" content "${content}")
		string(REGEX REPLACE "\"" "\\\\\"" content "${content}")
		string(REGEX REPLACE "([^\n]+)\n" "\"\\1\\\\n\"\n" content "${content}")
		string(REGEX REPLACE "\n+" "\n" content "${content}")
		file(APPEND ${outfile} "${content}")
		endforeach()

		file(APPEND ${outfile} ";\n")
		endfunction(GenerateOpenCLHeader)

Admin message