Dheemanth 5443602d89
CUDA 13.4 samples update - v13.4-public
Release 13.4 of the CUDA Samples supported by CUDA Toolkit 13.4.
See Changelog for more information.
2026-09-09 17:07:08 -05:00

88 lines
2.8 KiB
CMake

cmake_minimum_required(VERSION 3.20)
set(SAMPLE_DISALLOW_ARCHS 75)
include("${CMAKE_CURRENT_SOURCE_DIR}/../../../cmake/DetectCudaArch.cmake")
if(SAMPLE_SKIP_BUILD)
return()
endif()
list(APPEND CMAKE_MODULE_PATH "${CMAKE_CURRENT_SOURCE_DIR}/../../../cmake/Modules")
project(tileMatmulAutotuner LANGUAGES C CXX CUDA)
# Tile samples need C++20. If the host or CUDA compiler can't provide it,
# skip this sample instead of aborting configure.
if(NOT "cxx_std_20" IN_LIST CMAKE_CXX_COMPILE_FEATURES
OR (CMAKE_CXX_COMPILER_ID STREQUAL "GNU" AND CMAKE_CXX_COMPILER_VERSION VERSION_LESS 11)
OR NOT "cuda_std_20" IN_LIST CMAKE_CUDA_COMPILE_FEATURES)
message(STATUS "Compiler lacks required C++20 support - skipping ${PROJECT_NAME}")
return()
endif()
find_package(CUDAToolkit REQUIRED)
get_filename_component(CUDA_TOOLKIT_BIN_DIR "${CMAKE_CUDA_COMPILER}" DIRECTORY)
find_program(TILEIRAS_EXECUTABLE tileiras
HINTS "${CUDA_TOOLKIT_BIN_DIR}"
REQUIRED
)
set(CMAKE_CUDA_FLAGS "${CMAKE_CUDA_FLAGS} --enable-tile")
if(ENABLE_CUDA_DEBUG)
set(CMAKE_CUDA_FLAGS "${CMAKE_CUDA_FLAGS} -G")
else()
set(CMAKE_CUDA_FLAGS "${CMAKE_CUDA_FLAGS} -lineinfo") # add line information to all builds for debug tools (exclusive to -G option)
endif()
# This sample is not supported on QNX (needs tileiras/nvcc at runtime).
if(CMAKE_SYSTEM_NAME STREQUAL "QNX")
message(STATUS "Will not build sample ${PROJECT_NAME} - not supported on QNX")
return()
endif()
# Include directories and libraries
include_directories(../../../Common ../Benchmark_Common)
# Source files
add_executable(tileMatmulAutotuner
matmul_autotuner.cpp
autotuner_search_space.conf
)
target_compile_features(tileMatmulAutotuner PRIVATE cxx_std_20 cuda_std_20)
target_include_directories(tileMatmulAutotuner PRIVATE ${CUDAToolkit_INCLUDE_DIRS})
list(GET CUDAToolkit_INCLUDE_DIRS 0 CUDA_INCLUDE_DIR)
file(TO_CMAKE_PATH "${CUDA_INCLUDE_DIR}" CUDA_INCLUDE_DIR_FOR_DEFINE)
file(TO_CMAKE_PATH "${TILEIRAS_EXECUTABLE}" TILEIRAS_EXECUTABLE_FOR_DEFINE)
file(TO_CMAKE_PATH "${CMAKE_CUDA_COMPILER}" NVCC_EXECUTABLE_FOR_DEFINE)
target_compile_definitions(tileMatmulAutotuner PRIVATE
CUDA_INCLUDE_PATH="${CUDA_INCLUDE_DIR_FOR_DEFINE}"
NVCC_PATH="${NVCC_EXECUTABLE_FOR_DEFINE}"
TILEIRAS_PATH="${TILEIRAS_EXECUTABLE_FOR_DEFINE}"
)
target_link_libraries(tileMatmulAutotuner PRIVATE
CUDA::cuda_driver
CUDA::cudart
CUDA::nvrtc
)
add_custom_command(TARGET tileMatmulAutotuner POST_BUILD
COMMAND ${CMAKE_COMMAND} -E copy_if_different
${CMAKE_CURRENT_SOURCE_DIR}/matmul.cu
${CMAKE_CURRENT_BINARY_DIR}
COMMAND ${CMAKE_COMMAND} -E copy_if_different
${CMAKE_CURRENT_SOURCE_DIR}/autotuner_search_space.conf
${CMAKE_CURRENT_BINARY_DIR}
)
# Include installation configuration
include(${CMAKE_CURRENT_SOURCE_DIR}/../../../cmake/InstallSamples.cmake)
setup_samples_install()