CCCL (CUDA C++ Core Libraries) provides: - CUB: device/block/warp-level GPU primitives (reduce, scan, sort, topk) - Thrust: high-level parallel algorithms (transform_reduce, sort, scan) - libcudacxx: CUDA C++ standard library (atomics, barriers, memory) - cudax: experimental features (memory resources, allocators) - Tuning policies: per-SM hardware-specific algorithm parameters Competition optimization vectors mapped to CCCL: - Output TPS (83% weight): warp_reduce, block_reduce, device_topk - Input TPS (14% weight): device_scan, block_load, prefetch - Cache TPS (3% weight): prefix caching strategy patterns - Memory (0.9 util): pooled/cached/buddy allocators Source: https://github.com/NVIDIA/cccl (shallow clone, HEAD only) License: Apache-2.0
180 lines
5.6 KiB
CMake
180 lines
5.6 KiB
CMake
# Setup FileCheck if requested and available:
|
|
option(
|
|
THRUST_ENABLE_EXAMPLE_FILECHECK
|
|
"Check example output with the LLVM FileCheck utility."
|
|
OFF
|
|
)
|
|
set(filecheck_data_path "${Thrust_SOURCE_DIR}/internal/test")
|
|
|
|
if (THRUST_ENABLE_EXAMPLE_FILECHECK)
|
|
# TODO this should go into a find module
|
|
find_program(
|
|
THRUST_FILECHECK_EXECUTABLE
|
|
DOC "Path to the LLVM FileCheck utility."
|
|
NAMES
|
|
FileCheck
|
|
FileCheck-3.9
|
|
FileCheck-4.0
|
|
FileCheck-5.0
|
|
FileCheck-6.0
|
|
FileCheck-7
|
|
FileCheck-8
|
|
FileCheck-9
|
|
)
|
|
|
|
if (NOT THRUST_FILECHECK_EXECUTABLE)
|
|
message(
|
|
FATAL_ERROR
|
|
"Could not find the LLVM FileCheck utility. Set THRUST_FILECHECK_EXECUTABLE manually, "
|
|
"or disable THRUST_ENABLE_EXAMPLE_FILECHECK."
|
|
)
|
|
endif()
|
|
|
|
execute_process(
|
|
COMMAND
|
|
"${THRUST_FILECHECK_EXECUTABLE}"
|
|
"${filecheck_data_path}/thrust.smoke.filecheck"
|
|
INPUT_FILE "${Thrust_SOURCE_DIR}/cmake/filecheck_smoke_test"
|
|
RESULT_VARIABLE exit_code
|
|
)
|
|
|
|
if (0 EQUAL exit_code)
|
|
message(STATUS "FileCheck enabled: ${THRUST_FILECHECK_EXECUTABLE}")
|
|
else()
|
|
message(
|
|
FATAL_ERROR
|
|
"The current THRUST_FILECHECK_EXECUTABLE ('${THRUST_FILECHECK_EXECUTABLE}') "
|
|
"does not seem to be a valid FileCheck executable."
|
|
)
|
|
endif()
|
|
endif()
|
|
|
|
## thrust_add_example
|
|
#
|
|
# Add an example executable and register it with ctest.
|
|
#
|
|
# target_name_var: Variable name to overwrite with the name of the example
|
|
# target. Useful for post-processing target information per-backend.
|
|
# example_name: The name of the example minus "<config_prefix>.example." For
|
|
# instance, examples/vector.cu will be "vector", and examples/cuda/copy.cu
|
|
# would be "cuda.copy".
|
|
# example_src: The source file that implements the example.
|
|
# thrust_target: The reference thrust target with configuration information.
|
|
#
|
|
function(
|
|
thrust_add_example
|
|
target_name_var
|
|
example_name
|
|
example_src
|
|
thrust_target
|
|
)
|
|
thrust_get_target_property(config_host ${thrust_target} HOST)
|
|
thrust_get_target_property(config_device ${thrust_target} DEVICE)
|
|
thrust_get_target_property(config_prefix ${thrust_target} PREFIX)
|
|
|
|
# Wrap the .cu file in .cpp for non-CUDA backends
|
|
if ("CUDA" STREQUAL "${config_device}")
|
|
set(real_example_src "${example_src}")
|
|
else()
|
|
thrust_wrap_cu_in_cpp(real_example_src "${example_src}" ${thrust_target})
|
|
endif()
|
|
|
|
# The actual name of the test's target:
|
|
set(example_target ${config_prefix}.example.${example_name})
|
|
set(${target_name_var} ${example_target} PARENT_SCOPE)
|
|
|
|
cccl_add_executable(${example_target} SOURCES "${real_example_src}")
|
|
target_link_libraries(${example_target} PRIVATE ${thrust_target})
|
|
target_include_directories(
|
|
${example_target}
|
|
PRIVATE "${Thrust_SOURCE_DIR}/examples"
|
|
)
|
|
|
|
if ("CUDA" STREQUAL "${config_device}")
|
|
thrust_configure_cuda_target(${example_target} RDC ${THRUST_FORCE_RDC})
|
|
endif()
|
|
|
|
# We do not want to explicitly include `host_device.h` if not needed,
|
|
# so force include the file for non CUDA targets.
|
|
set(gx_msvc "$<CXX_COMPILER_ID:MSVC>")
|
|
set(gx_cxx "$<COMPILE_LANGUAGE:CXX>")
|
|
set(gx_cxx_msvc "$<AND:${gx_cxx},${gx_msvc}>")
|
|
set(gx_cxx_not_msvc "$<AND:${gx_cxx},$<NOT:${gx_msvc}>>")
|
|
target_compile_options(
|
|
${example_target}
|
|
PRIVATE
|
|
"$<${gx_cxx_msvc}:SHELL:/FI include/host_device.h>"
|
|
"$<${gx_cxx_not_msvc}:SHELL:-include include/host_device.h>"
|
|
)
|
|
target_compile_definitions(
|
|
${example_target}
|
|
PRIVATE $<$<COMPILE_LANG_AND_ID:CUDA,NVIDIA>:THRUST_EXAMPLE_DEVICE_SIDE>
|
|
)
|
|
|
|
# Get the name of FileCheck input by stripping out the config name.
|
|
# (e.g. "thrust.cpp.cuda.cpp14.example.xxx" -> "thrust.example.xxx.filecheck")
|
|
string(
|
|
REPLACE
|
|
"${config_prefix}"
|
|
"thrust"
|
|
filecheck_reference_file
|
|
"${example_target}.filecheck"
|
|
)
|
|
|
|
add_test(
|
|
NAME ${example_target}
|
|
# gersemi: off
|
|
COMMAND
|
|
"${CMAKE_COMMAND}"
|
|
"-DEXAMPLE_EXECUTABLE=$<TARGET_FILE:${example_target}>"
|
|
"-DFILECHECK_ENABLED=${THRUST_ENABLE_EXAMPLE_FILECHECK}"
|
|
"-DFILECHECK_EXECUTABLE=${THRUST_FILECHECK_EXECUTABLE}"
|
|
"-DREFERENCE_FILE=${filecheck_data_path}/${filecheck_reference_file}"
|
|
-P "${Thrust_SOURCE_DIR}/cmake/ThrustRunExample.cmake"
|
|
# gersemi: on
|
|
)
|
|
|
|
# Run OMP/TBB tests in serial. Multiple OMP processes will massively
|
|
# oversubscribe the machine with GCC's OMP, and we want to test these with
|
|
# the full CPU available to each unit test.
|
|
set(config_systems ${config_host} ${config_device})
|
|
if (("OMP" IN_LIST config_systems) OR ("TBB" IN_LIST config_systems))
|
|
set_tests_properties(${example_target} PROPERTIES RUN_SERIAL ON)
|
|
endif()
|
|
|
|
# Check for per-example script. Script will be included in the current scope
|
|
# to allow custom property modifications.
|
|
get_filename_component(example_cmake_script "${example_src}" NAME_WLE)
|
|
set(
|
|
example_cmake_script
|
|
"${CMAKE_CURRENT_LIST_DIR}/${example_cmake_script}.cmake"
|
|
)
|
|
# Use a glob so we can detect if this changes:
|
|
file(
|
|
GLOB example_cmake_script
|
|
RELATIVE "${CMAKE_CURRENT_LIST_DIR}"
|
|
CONFIGURE_DEPENDS
|
|
"${example_cmake_script}"
|
|
)
|
|
if (example_cmake_script) # Will be non-empty only if the script exists
|
|
include("${example_cmake_script}")
|
|
endif()
|
|
endfunction()
|
|
|
|
file(
|
|
GLOB example_srcs
|
|
RELATIVE "${CMAKE_CURRENT_LIST_DIR}"
|
|
CONFIGURE_DEPENDS
|
|
*.cu
|
|
*.cpp
|
|
)
|
|
|
|
foreach (thrust_target IN LISTS THRUST_TARGETS)
|
|
foreach (example_src IN LISTS example_srcs)
|
|
get_filename_component(example_name "${example_src}" NAME_WLE)
|
|
thrust_add_example(example_target ${example_name} "${example_src}" ${thrust_target})
|
|
endforeach()
|
|
endforeach()
|
|
|
|
add_subdirectory(cuda)
|