NASP/cuda-samples
52
0
Fork 0
mirror of https://github.com/NVIDIA/cuda-samples.git synced 2025-11-04 15:47:50 +08:00
Code Issues Packages Projects Releases Wiki Activity

Refactor CMakeLists.txt under 6_Performance

Browse Source
This commit is contained in:
Rob Armstrong 2024-12-16 14:52:10 -08:00
parent 00999c1789
commit cbfab74480
5 changed files with 95 additions and 27 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

24
Samples/6_Performance/LargeKernelParameter/CMakeLists.txt
View File

@ -1,11 +1,25 @@
cmake_minimum_required(VERSION 3.20)
list(APPEND CMAKE_MODULE_PATH "${CMAKE_SOURCE_DIR}/cmake/Modules")
project(LargeKernelParameter LANGUAGES C CXX CUDA)
find_package(CUDAToolkit REQUIRED)
set(CMAKE_CUDA_ARCHITECTURES 70 75 80 86 89 90)
if(CMAKE_BUILD_TYPE STREQUAL "Debug")
# set(CMAKE_CUDA_FLAGS "${CMAKE_CUDA_FLAGS} -G") # enable cuda-gdb (expensive)
endif()
# Include directories and libraries
include_directories(../../../Common)
# Source file
set(SRC_FILES
LargeKernelParameter.cu
)
# Add target for LargeKernelParameter
add_executable(LargeKernelParameter ${SRC_FILES})
add_executable(LargeKernelParameter LargeKernelParameter.cu)
target_compile_options(LargeKernelParameter PRIVATE $<$<COMPILE_LANGUAGE:CUDA>:--extended-lambda>)
target_compile_features(LargeKernelParameter PRIVATE cxx_std_17 cuda_std_17)
set_target_properties(LargeKernelParameter PROPERTIES CUDA_SEPARABLE_COMPILATION ON)

26
Samples/6_Performance/UnifiedMemoryPerf/CMakeLists.txt
View File

@ -1,15 +1,27 @@
cmake_minimum_required(VERSION 3.20)
list(APPEND CMAKE_MODULE_PATH "${CMAKE_SOURCE_DIR}/cmake/Modules")
project(UnifiedMemoryPerf LANGUAGES C CXX CUDA)
find_package(CUDAToolkit REQUIRED)
set(CMAKE_CUDA_ARCHITECTURES "native")
if(CMAKE_BUILD_TYPE STREQUAL "Debug")
# set(CMAKE_CUDA_FLAGS "${CMAKE_CUDA_FLAGS} -G") # enable cuda-gdb (expensive)
endif()
# Include directories and libraries
include_directories(../../../Common)
# Source file
set(SRC_FILES
helperFunctions.cpp
matrixMultiplyPerf.cu
commonKernels.cu
)
# Add target for UnifiedMemoryPerf
add_executable(UnifiedMemoryPerf ${SRC_FILES})
add_executable(UnifiedMemoryPerf helperFunctions.cpp matrixMultiplyPerf.cu commonKernels.cu)
target_compile_options(UnifiedMemoryPerf PRIVATE $<$<COMPILE_LANGUAGE:CUDA>:--extended-lambda>)
target_compile_features(UnifiedMemoryPerf PRIVATE cxx_std_17 cuda_std_17)
set_target_properties(UnifiedMemoryPerf PROPERTIES CUDA_SEPARABLE_COMPILATION ON)
target_include_directories(UnifiedMemoryPerf PRIVATE

24
Samples/6_Performance/alignedTypes/CMakeLists.txt
View File

@ -1,11 +1,25 @@
cmake_minimum_required(VERSION 3.20)
list(APPEND CMAKE_MODULE_PATH "${CMAKE_SOURCE_DIR}/cmake/Modules")
project(alignedTypes LANGUAGES C CXX CUDA)
find_package(CUDAToolkit REQUIRED)
set(CMAKE_CUDA_ARCHITECTURES "native")
if(CMAKE_BUILD_TYPE STREQUAL "Debug")
# set(CMAKE_CUDA_FLAGS "${CMAKE_CUDA_FLAGS} -G") # enable cuda-gdb (expensive)
endif()
# Include directories and libraries
include_directories(../../../Common)
# Source file
set(SRC_FILES
alignedTypes.cu
)
# Add target for alignedTypes
add_executable(alignedTypes ${SRC_FILES})
add_executable(alignedTypes alignedTypes.cu)
target_compile_options(alignedTypes PRIVATE $<$<COMPILE_LANGUAGE:CUDA>:--extended-lambda>)
target_compile_features(alignedTypes PRIVATE cxx_std_17 cuda_std_17)
set_target_properties(alignedTypes PROPERTIES CUDA_SEPARABLE_COMPILATION ON)

24
Samples/6_Performance/cudaGraphsPerfScaling/CMakeLists.txt
View File

@ -1,11 +1,25 @@
cmake_minimum_required(VERSION 3.20)
list(APPEND CMAKE_MODULE_PATH "${CMAKE_SOURCE_DIR}/cmake/Modules")
project(cudaGraphsPerfScaling LANGUAGES C CXX CUDA)
find_package(CUDAToolkit REQUIRED)
set(CMAKE_CUDA_ARCHITECTURES "native")
if(CMAKE_BUILD_TYPE STREQUAL "Debug")
# set(CMAKE_CUDA_FLAGS "${CMAKE_CUDA_FLAGS} -G") # enable cuda-gdb (expensive)
endif()
# Include directories and libraries
include_directories(../../../Common)
# Source file
set(SRC_FILES
cudaGraphPerfScaling.cu
)
# Add target for cudaGraphsPerfScaling
add_executable(cudaGraphsPerfScaling ${SRC_FILES})
add_executable(cudaGraphsPerfScaling cudaGraphPerfScaling.cu)
target_compile_options(cudaGraphsPerfScaling PRIVATE $<$<COMPILE_LANGUAGE:CUDA>:--extended-lambda>)
target_compile_features(cudaGraphsPerfScaling PRIVATE cxx_std_17 cuda_std_17)
set_target_properties(cudaGraphsPerfScaling PROPERTIES CUDA_SEPARABLE_COMPILATION ON)

24
Samples/6_Performance/transpose/CMakeLists.txt
View File

@ -1,11 +1,25 @@
cmake_minimum_required(VERSION 3.20)
list(APPEND CMAKE_MODULE_PATH "${CMAKE_SOURCE_DIR}/cmake/Modules")
project(transpose LANGUAGES C CXX CUDA)
find_package(CUDAToolkit REQUIRED)
set(CMAKE_CUDA_ARCHITECTURES "native")
if(CMAKE_BUILD_TYPE STREQUAL "Debug")
# set(CMAKE_CUDA_FLAGS "${CMAKE_CUDA_FLAGS} -G") # enable cuda-gdb (expensive)
endif()
# Include directories and libraries
include_directories(../../../Common)
# Source file
set(SRC_FILES
transpose.cu
)
# Add target for transpose
add_executable(transpose ${SRC_FILES})
add_executable(transpose transpose.cu)
target_compile_options(transpose PRIVATE $<$<COMPILE_LANGUAGE:CUDA>:--extended-lambda>)
target_compile_features(transpose PRIVATE cxx_std_17 cuda_std_17)
set_target_properties(transpose PROPERTIES CUDA_SEPARABLE_COMPILATION ON)