56fd68e7dd
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
331 lines
9.0 KiB
CMake
331 lines
9.0 KiB
CMake
cccl_get_c2h()
|
|
cccl_get_cudatoolkit()
|
|
|
|
## cudax_add_test
|
|
#
|
|
# Add a catch2 test executable and register it with ctest.
|
|
#
|
|
# target_name_var: Variable name to overwrite with the name of the test
|
|
# target. Useful for modifying the test/target after creation.
|
|
# test_name: A unique name for the executable that will be appended to "cudax.test.".
|
|
#
|
|
# Additional arguments will be processed as test sources.
|
|
#
|
|
function(cudax_add_catch2_test target_name_var test_name) # ARGN=test sources
|
|
set(test_target cudax.test.${test_name})
|
|
set(test_sources ${ARGN})
|
|
|
|
cccl_add_executable(${test_target} SOURCES ${ARGN} ADD_CTEST)
|
|
target_include_directories(
|
|
${test_target}
|
|
PRIVATE "${CMAKE_CURRENT_FUNCTION_LIST_DIR}/common"
|
|
)
|
|
target_link_libraries(
|
|
${test_target}
|
|
PRIVATE #
|
|
cudax.compiler_interface
|
|
cccl.c2h.main
|
|
CUDA::cudart_static
|
|
)
|
|
target_compile_options(
|
|
${test_target}
|
|
PRIVATE $<$<COMPILE_LANG_AND_ID:CUDA,NVIDIA>:--extended-lambda>
|
|
)
|
|
|
|
set(${target_name_var} ${test_target} PARENT_SCOPE)
|
|
endfunction()
|
|
|
|
cudax_add_catch2_test(test_target launch
|
|
launch/launch_smoke.cu
|
|
)
|
|
|
|
cudax_add_catch2_test(test_target execution ${cudax_target}
|
|
execution/env.cu
|
|
execution/policies/policies.cu
|
|
execution/policies/get_execution_policy.cu
|
|
execution/test_bulk.cu
|
|
execution/test_concepts.cu
|
|
execution/test_completion_signatures.cu
|
|
execution/test_conditional.cu
|
|
execution/test_continues_on.cu
|
|
execution/test_just.cu
|
|
execution/test_let_value.cu
|
|
execution/test_on.cu
|
|
execution/test_sequence.cu
|
|
execution/test_starts_on.cu
|
|
execution/test_stream_context.cu
|
|
execution/test_task_scheduler.cu
|
|
execution/test_then.cu
|
|
execution/test_trampoline_scheduler.cu
|
|
execution/test_visit.cu
|
|
execution/test_when_all.cu
|
|
execution/test_write_attrs.cu
|
|
execution/test_write_env.cu
|
|
)
|
|
|
|
target_compile_options(
|
|
${test_target}
|
|
PRIVATE $<$<COMPILE_LANG_AND_ID:CUDA,NVIDIA>:-allow-unsupported-compiler>
|
|
)
|
|
|
|
cudax_add_catch2_test(test_target graph
|
|
graph/graph_smoke.cu
|
|
graph/graph_node_ops_smoke.cu
|
|
graph/graph_buffer_smoke.cu
|
|
)
|
|
|
|
cudax_add_catch2_test(test_target stream
|
|
stream/stream_smoke.cu
|
|
)
|
|
|
|
cudax_add_catch2_test(test_target misc
|
|
utility/ensure_current_device.cu
|
|
utility/unstable_unique.cu
|
|
utility/optionally_static.cu
|
|
)
|
|
|
|
cudax_add_catch2_test(test_target containers
|
|
containers/uninitialized_buffer.cu
|
|
)
|
|
|
|
cudax_add_catch2_test(test_target copy_bytes
|
|
copy_bytes/mdspan_d2h_h2d.cu
|
|
copy_bytes/mdspan_d2h_h2d_relaxed.cu
|
|
)
|
|
|
|
cudax_add_catch2_test(test_target fill_bytes
|
|
fill_bytes/fill_bytes_mdspan.cu
|
|
fill_bytes/fill_bytes_mdspan_example.cu
|
|
)
|
|
|
|
cudax_add_catch2_test(test_target copy
|
|
copy/copy.cu
|
|
copy/copy_edge_cases.cu
|
|
copy/copy_vectorize.cu
|
|
copy/copy_vectorize_5d.cu
|
|
copy/copy_llm.cu
|
|
copy/copy_nvmath.cu
|
|
copy/copy_nvmath_transpose.cu
|
|
copy/copy_shared_memory.cu
|
|
)
|
|
|
|
# nvcc 12.0 fails to compile cuco hashers test
|
|
if (
|
|
NOT "${CMAKE_CUDA_COMPILER_ID}" STREQUAL "NVIDIA"
|
|
OR "${CMAKE_CUDA_COMPILER_VERSION}" VERSION_GREATER_EQUAL "12.1"
|
|
)
|
|
cudax_add_catch2_test(test_target cuco.utility.hashers
|
|
cuco/utility/test_hashers.cu
|
|
)
|
|
endif()
|
|
|
|
cudax_add_catch2_test(test_target cuco.fixed_capacity_map.insert_and_contains ${cudax_target}
|
|
cuco/fixed_capacity_map/test_insert_and_contains.cu
|
|
)
|
|
|
|
cudax_add_catch2_test(test_target cuco.fixed_capacity_map.find ${cudax_target}
|
|
cuco/fixed_capacity_map/test_find.cu
|
|
)
|
|
|
|
cudax_add_catch2_test(test_target cuco.fixed_capacity_map.capacity ${cudax_target}
|
|
cuco/fixed_capacity_map/test_capacity.cu
|
|
)
|
|
|
|
cudax_add_catch2_test(test_target cuco.fixed_capacity_map.shared_memory ${cudax_target}
|
|
cuco/fixed_capacity_map/test_shared_memory.cu
|
|
)
|
|
|
|
cudax_add_catch2_test(test_target cuco.fixed_capacity_map.misaligned_storage ${cudax_target}
|
|
cuco/fixed_capacity_map/test_misaligned_storage.cu
|
|
)
|
|
|
|
cudax_add_catch2_test(test_target cuco.fixed_capacity_map.key_sentinel ${cudax_target}
|
|
cuco/fixed_capacity_map/test_key_sentinel.cu
|
|
)
|
|
|
|
cudax_add_catch2_test(test_target cuco.utility.capacity ${cudax_target}
|
|
cuco/utility/test_capacity.cu
|
|
)
|
|
|
|
cudax_add_catch2_test(test_target cuco.hyperloglog ${cudax_target}
|
|
cuco/hyperloglog/test_hyperloglog.cu
|
|
)
|
|
|
|
# cudax_add_cuco_fail_test
|
|
function(cudax_add_cuco_fail_test target_name_var test_name source)
|
|
set(test_target cudax.test.${test_name})
|
|
|
|
cccl_add_executable(
|
|
${test_target}
|
|
SOURCES ${source}
|
|
NO_METATARGETS
|
|
NO_CLANG_TIDY
|
|
)
|
|
target_link_libraries(${test_target} PRIVATE cudax.compiler_interface)
|
|
cccl_add_xfail_compile_target_test(
|
|
${test_target}
|
|
SOURCE_FILE "${source}"
|
|
ERROR_REGEX_LABEL "expected-error"
|
|
)
|
|
|
|
set(${target_name_var} ${test_target} PARENT_SCOPE)
|
|
endfunction()
|
|
|
|
# Expected-failure compile tests (_fail.cu convention)
|
|
cudax_add_cuco_fail_test(test_target cuco.fixed_capacity_map.key_size_fail
|
|
cuco/fixed_capacity_map/test_key_size_fail.cu
|
|
)
|
|
|
|
cudax_add_cuco_fail_test(test_target cuco.fixed_capacity_map.slot_size_fail
|
|
cuco/fixed_capacity_map/test_slot_size_fail.cu
|
|
)
|
|
|
|
cudax_add_catch2_test(test_target green_context
|
|
green_context/green_ctx_smoke.cu
|
|
)
|
|
|
|
cudax_add_catch2_test(test_target kernel
|
|
kernel/kernel_ref.cu
|
|
)
|
|
|
|
cudax_add_catch2_test(test_target library_ref
|
|
library/library_ref.cu
|
|
)
|
|
|
|
cudax_add_catch2_test(test_target library
|
|
library/library.cu
|
|
)
|
|
|
|
cudax_add_catch2_test(test_target algorithm
|
|
algorithm/fill.cu
|
|
algorithm/copy.cu
|
|
)
|
|
|
|
add_subdirectory(multi_gpu)
|
|
|
|
cudax_add_catch2_test(test_target group.mapping.binary_partition
|
|
group/mapping/binary_partition.cu
|
|
)
|
|
cudax_add_catch2_test(test_target group.mapping.composite_mapping
|
|
group/mapping/composite_mapping.cu
|
|
)
|
|
cudax_add_catch2_test(test_target group.mapping.group_as
|
|
group/mapping/group_as.cu
|
|
)
|
|
cudax_add_catch2_test(test_target group.mapping.group_by
|
|
group/mapping/group_by.cu
|
|
)
|
|
cudax_add_catch2_test(test_target group.mapping.identity_mapping
|
|
group/mapping/identity_mapping.cu
|
|
)
|
|
cudax_add_catch2_test(test_target group.mapping.take
|
|
group/mapping/take.cu
|
|
)
|
|
cudax_add_catch2_test(test_target group.synchronizer.lane_synchronizer
|
|
group/synchronizer/lane_synchronizer.cu
|
|
)
|
|
cudax_add_catch2_test(test_target group.synchronizer.barrier_synchronizer
|
|
group/synchronizer/barrier_synchronizer.cu
|
|
)
|
|
cudax_add_catch2_test(test_target group.cooperative_algorithm
|
|
group/cooperative_algorithm.cu
|
|
)
|
|
cudax_add_catch2_test(test_target group.group
|
|
group/group.cu
|
|
)
|
|
cudax_add_catch2_test(test_target group.make_this_group
|
|
group/make_this_group.cu
|
|
)
|
|
cudax_add_catch2_test(test_target group.segmented_algorithm
|
|
group/segmented_algorithm.cu
|
|
)
|
|
cudax_add_catch2_test(test_target group.this_group
|
|
group/this_group.cu
|
|
)
|
|
cudax_add_catch2_test(test_target group.invoke_one
|
|
group/invoke_one.cu
|
|
)
|
|
|
|
cudax_add_catch2_test(test_target coop.any_of.this_thread
|
|
coop/any_of/this_thread.cu
|
|
)
|
|
cudax_add_catch2_test(test_target coop.any_of.this_warp
|
|
coop/any_of/this_warp.cu
|
|
)
|
|
cudax_add_catch2_test(test_target coop.any_of.threads_within_warp
|
|
coop/any_of/threads_within_warp.cu
|
|
)
|
|
|
|
cudax_add_catch2_test(test_target coop.reduce.this_thread
|
|
coop/reduce/this_thread.cu
|
|
)
|
|
cudax_add_catch2_test(test_target coop.reduce.this_warp
|
|
coop/reduce/this_warp.cu
|
|
)
|
|
cudax_add_catch2_test(test_target coop.reduce.this_block
|
|
coop/reduce/this_block.cu
|
|
)
|
|
cudax_add_catch2_test(test_target coop.reduce.this_cluster
|
|
coop/reduce/this_cluster.cu
|
|
)
|
|
cudax_add_catch2_test(test_target coop.reduce.this_grid
|
|
coop/reduce/this_grid.cu
|
|
)
|
|
cudax_add_catch2_test(test_target coop.reduce.threads_within_warp
|
|
coop/reduce/threads_within_warp.cu
|
|
)
|
|
cudax_add_catch2_test(test_target coop.reduce.warps_within_block
|
|
coop/reduce/warps_within_block.cu
|
|
)
|
|
|
|
cudax_add_catch2_test(test_target coop.shuffle.threads_within_warp
|
|
coop/shuffle/threads_within_warp.cu
|
|
)
|
|
|
|
cudax_add_catch2_test(test_target coop.shuffle_down.threads_within_warp
|
|
coop/shuffle_down/threads_within_warp.cu
|
|
)
|
|
|
|
cudax_add_catch2_test(test_target coop.shuffle_up.threads_within_warp
|
|
coop/shuffle_up/threads_within_warp.cu
|
|
)
|
|
|
|
if (cudax_ENABLE_CUFILE)
|
|
cudax_add_catch2_test(test_target cufile.driver_attributes
|
|
cufile/driver_attributes.cu
|
|
)
|
|
target_link_libraries(${test_target} PRIVATE CUDA::cuFile)
|
|
|
|
cudax_add_catch2_test(test_target cufile.driver_register
|
|
cufile/driver_register.cu
|
|
)
|
|
target_link_libraries(${test_target} PRIVATE CUDA::cuFile)
|
|
|
|
cudax_add_catch2_test(test_target cufile.cufile
|
|
cufile/cufile.cu
|
|
)
|
|
target_link_libraries(${test_target} PRIVATE CUDA::cuFile)
|
|
|
|
cudax_add_catch2_test(test_target cufile.cufile_ref
|
|
cufile/cufile_ref.cu
|
|
)
|
|
target_link_libraries(${test_target} PRIVATE CUDA::cuFile)
|
|
|
|
cudax_add_catch2_test(test_target cufile.open_mode
|
|
cufile/open_mode.cu
|
|
)
|
|
target_link_libraries(${test_target} PRIVATE CUDA::cuFile)
|
|
endif()
|
|
|
|
# FIXME: Enable MSVC
|
|
if (cudax_ENABLE_PLACES AND NOT "MSVC" STREQUAL "${CMAKE_CXX_COMPILER_ID}")
|
|
# Places tests are handled separately:
|
|
add_subdirectory(places)
|
|
endif()
|
|
|
|
# FIXME: Enable MSVC
|
|
if (cudax_ENABLE_CUDASTF AND NOT "MSVC" STREQUAL "${CMAKE_CXX_COMPILER_ID}")
|
|
# STF tests are handled separately:
|
|
add_subdirectory(stf)
|
|
endif()
|