derek.holloway
7502bf64a4
CodeQL / Analyze (actions) (push) Canceled after 0s
CodeQL / Analyze (python) (push) Canceled after 0s
pre-commit / pre-commit (push) Canceled after 0s
Multi-Arch CI / setup (push) Canceled after 0s
Publish build_manylinux_x86_64 images / publish_build_manylinux_x86_64 (push) Canceled after 0s
Setup (Multi-Arch) / setup (push) Canceled after 0s
Multi-Arch CI / Linux::${{ fromJSON(needs.setup.outputs.linux_build_config || '{}').build_variant_label || 'skip' }} (push) Canceled after 0s
Publish TheRock Docker image / build-and-push-image (push) Canceled after 0s
Publish no_rocm_image_ubi10 image / publish_no_rocm_image_ubi10 (push) Canceled after 0s
Multi-Arch CI / Windows::${{ fromJSON(needs.setup.outputs.windows_build_config || '{}').build_variant_label || 'skip' }} (push) Canceled after 0s
Publish TheRock Docker image / build-and-push-image (push) Canceled after 0s
Multi-Arch CI / Manifest Diff (push) Canceled after 0s
TheRock Manifest Diff Report / Generate Manifest Diff Report (push) Canceled after 0s
Multi-Arch CI / CI Summary (push) Canceled after 0s
Publish no_rocm_image_ubuntu24_04 images / publish_no_rocm_image_ubuntu24_04 (push) Canceled after 0s
Publish no_rocm_image_ubuntu24_04_ocl_rt images / publish_no_rocm_image_ubuntu24_04_ocl_rt (push) Canceled after 0s
Publish TheRock Docker image / build-and-push-image (push) Canceled after 0s
Publish TheRock Docker image / build-and-push-image (push) Canceled after 0s
Publish no_rocm_image_ubuntu24_04 images / publish_no_rocm_image_ubuntu24_04_rocgdb (push) Canceled after 0s
Publish no_rocm_image_ubuntu24_04_rocgdb images / publish_no_rocm_image_ubuntu24_04_rocgdb (push) Canceled after 0s
Publish no_rocm_image_ubuntu24_04 images / publish_no_rocm_image_ubuntu24_04_ocl_rt (push) Canceled after 0s
Publish TheRock Docker image / build-and-push-image (push) Canceled after 0s
Unit Tests / Unit Tests :: ubuntu-24.04 (push) Canceled after 0s
Unit Tests / Unit Tests :: windows-2022 (push) Canceled after 0s
Publish no_rocm_image_ubuntu24_04 images / publish_no_rocm_image_ubuntu24_04_media (push) Canceled after 0s
Unit Tests / Unit Tests Summary (push) Canceled after 0s
Publish no_rocm_image_ubuntu24_04 images / publish_no_rocm_image_ubuntu24_04_openmpi (push) Canceled after 0s
1391 lines
45 KiB
CMake
1391 lines
45 KiB
CMake
cmake_minimum_required(VERSION 3.21 FATAL_ERROR)
|
|
|
|
#Add a name for the rocm-systems location
|
|
set(ROCM_SYSTEMS_ROOT "${CMAKE_CURRENT_SOURCE_DIR}/../..")
|
|
|
|
# Set cmake_prefix_path for searching, ROCM_PATH if avail otherwise default to general rocm install path
|
|
set(CMAKE_PREFIX_PATH $ENV{ROCM_PATH} "/opt/rocm/")
|
|
message(STATUS "ROCM_PATH: $ENV{ROCM_PATH}")
|
|
|
|
if(
|
|
CMAKE_SOURCE_DIR STREQUAL CMAKE_BINARY_DIR
|
|
AND CMAKE_CURRENT_SOURCE_DIR STREQUAL CMAKE_SOURCE_DIR
|
|
)
|
|
set(MSG "")
|
|
message(STATUS "Warning! Building from the source directory is not recommended")
|
|
message(STATUS "If unintended, please remove 'CMakeCache.txt' and 'CMakeFiles'")
|
|
message(STATUS "and build from a separate directory")
|
|
message(FATAL_ERROR "In-source build")
|
|
endif()
|
|
|
|
# System info
|
|
cmake_host_system_information(RESULT LOCALHOST QUERY FQDN)
|
|
message(STATUS "Hostname: ${LOCALHOST}")
|
|
|
|
# Versioning info derived from file
|
|
file(READ "${CMAKE_CURRENT_SOURCE_DIR}/VERSION" FULL_VERSION_STRING LIMIT_COUNT 1)
|
|
string(REGEX REPLACE "(\n|\r)" "" FULL_VERSION_STRING "${FULL_VERSION_STRING}")
|
|
set(ROCPROFCOMPUTE_FULL_VERSION "${FULL_VERSION_STRING}")
|
|
string(
|
|
REGEX REPLACE
|
|
"([0-9]+)\.([0-9]+)\.([0-9]+)(.*)"
|
|
"\\1.\\2.\\3"
|
|
ROCPROFCOMPUTE_VERSION
|
|
"${FULL_VERSION_STRING}"
|
|
)
|
|
|
|
# string(REGEX REPLACE "(${ROCPROFCOMPUTE_VERSION})(.*)" "\\2"
|
|
# ROCPROFCOMPUTE_VERSION_TWEAK
|
|
# "${FULL_VERSION_STRING}")
|
|
# string(REGEX REPLACE "^\\." "" ROCPROFCOMPUTE_VERSION_TWEAK
|
|
# "${ROCPROFCOMPUTE_VERSION_TWEAK}")
|
|
|
|
project(
|
|
rocprofiler-compute
|
|
VERSION ${ROCPROFCOMPUTE_VERSION}
|
|
LANGUAGES CXX
|
|
DESCRIPTION
|
|
"A kernel-level profiling tool for machine learning/HPC workloads running on AMD MI GPUs"
|
|
HOMEPAGE_URL
|
|
"https://github.com/ROCm/rocm-systems/tree/develop/projects/rocprofiler-compute"
|
|
)
|
|
|
|
set(PACKAGE_NAME "rocprofiler-compute")
|
|
set(PACKAGE_NAME_UNDERSCORE "rocprofiler_compute")
|
|
set(EXECUTABLE_NAME "rocprof-compute")
|
|
|
|
include(ExternalProject)
|
|
include(GNUInstallDirs)
|
|
include(utils.cmake)
|
|
# Must run before asan_build_dependencies.cmake, whose pthread guard reads the
|
|
# ENABLE_SANITIZER that this module resolves.
|
|
include(cmake/sanitizers.cmake)
|
|
include(${CMAKE_CURRENT_LIST_DIR}/cmake/asan_build_dependencies.cmake)
|
|
|
|
set(BUILD_ENABLE_LINTIAN_OVERRIDES ON CACHE BOOL "Enable/Disable Lintian Overrides")
|
|
set(BUILD_DEBIAN_PKGING_FLAG
|
|
ON
|
|
CACHE BOOL
|
|
"Internal Status Flag to indicate Debian Packaging Build"
|
|
)
|
|
set(COMP_TYPE "main")
|
|
|
|
# version control info
|
|
execute_process(
|
|
COMMAND git log --pretty=format:%h -n 1
|
|
OUTPUT_VARIABLE ROCPROFCOMPUTE_GIT_REV
|
|
OUTPUT_STRIP_TRAILING_WHITESPACE
|
|
)
|
|
message(STATUS "Git revision: ${ROCPROFCOMPUTE_GIT_REV}")
|
|
configure_file(
|
|
${PROJECT_SOURCE_DIR}/cmake/VERSION.sha.in
|
|
${PROJECT_SOURCE_DIR}/VERSION.sha
|
|
@ONLY
|
|
)
|
|
|
|
set(CMAKE_BUILD_TYPE "Release")
|
|
if(CMAKE_INSTALL_PREFIX_INITIALIZED_TO_DEFAULT)
|
|
set(CMAKE_INSTALL_PREFIX "/opt/rocm" CACHE PATH "default install path" FORCE)
|
|
endif()
|
|
message(STATUS "Installation path: ${CMAKE_INSTALL_PREFIX}")
|
|
|
|
# Profile mode requires only Python 3.8+ standard library (no external dependencies)
|
|
# Analyze mode dependencies are checked at runtime via verify_deps()
|
|
# Standalone binary handles its own dependencies via bundled venv
|
|
|
|
option(STANDALONEBINARY "Whether to build standalone binary" OFF)
|
|
|
|
# Only find Python interpreter - no dependency checking
|
|
if(NOT STANDALONEBINARY)
|
|
message(STATUS "Detecting Python interpreter...")
|
|
find_package(Python3 3.8 COMPONENTS Interpreter REQUIRED)
|
|
message(STATUS "Python ${Python3_VERSION} found")
|
|
message(
|
|
STATUS
|
|
"Note: Profile mode uses stdlib only. Analyze mode deps checked at runtime."
|
|
)
|
|
endif()
|
|
|
|
# ----------------------
|
|
# Portable Python command for tests
|
|
# ----------------------
|
|
# For build-time checks, use ${Python3_EXECUTABLE} (absolute path from find_package)
|
|
# For runtime tests (add_test), use portable command that resolves from PATH
|
|
# This allows tests to work when build/install folders are moved to different machines
|
|
set(PYTHON_TEST_COMMAND
|
|
"python3"
|
|
CACHE STRING
|
|
"Python command for running tests (portable)"
|
|
)
|
|
enable_sanitizer_python_launcher(PYTHON_TEST_COMMAND)
|
|
message(STATUS "Python test command: ${PYTHON_TEST_COMMAND}")
|
|
|
|
# ----------------------
|
|
# modulefile creation
|
|
# ----------------------
|
|
|
|
set(MOD_INSTALL_PATH
|
|
"${CMAKE_INSTALL_DATADIR}/${PROJECT_NAME}/modulefiles/${PROJECT_NAME}"
|
|
CACHE STRING
|
|
"Install path for modulefile"
|
|
)
|
|
message(STATUS "Modulefile install path: ${MOD_INSTALL_PATH}")
|
|
|
|
set(moduleFileTemplate "rocprofcompute.lua.in")
|
|
|
|
configure_file(
|
|
${PROJECT_SOURCE_DIR}/cmake/${moduleFileTemplate}
|
|
${PROJECT_BINARY_DIR}/${MOD_INSTALL_PATH}/${ROCPROFCOMPUTE_FULL_VERSION}.lua
|
|
@ONLY
|
|
)
|
|
|
|
# Thera mods
|
|
if(LOCALHOST MATCHES "TheraS01|.*\.thera\.amd\.com|thera-hn")
|
|
list(APPEND CMAKE_MESSAGE_INDENT " ")
|
|
message(STATUS "Using thera-specific modulefile modification")
|
|
file(READ ${PROJECT_SOURCE_DIR}/cmake/modfile.thera.mod mod_additions)
|
|
file(
|
|
APPEND
|
|
${PROJECT_BINARY_DIR}/${MOD_INSTALL_PATH}/${ROCPROFCOMPUTE_FULL_VERSION}.lua
|
|
${mod_additions}
|
|
)
|
|
list(POP_BACK CMAKE_MESSAGE_INDENT)
|
|
endif()
|
|
|
|
# Setup testing collateral
|
|
|
|
option(INSTALL_TESTS "Build test suite" OFF)
|
|
option(ENABLE_TESTS "Enable compilation of testing collateral" OFF)
|
|
set(CMAKE_HIP_FLAGS_RELEASE "-O2")
|
|
if(${ENABLE_TESTS})
|
|
enable_language("C" "HIP")
|
|
add_subdirectory(tests)
|
|
endif()
|
|
message(STATUS "Enable tests compilation: ${ENABLE_TESTS}")
|
|
|
|
enable_testing()
|
|
include(CTest)
|
|
|
|
option(ENABLE_COVERAGE "Enable code coverage" OFF)
|
|
set(COV_OPTION "")
|
|
if(${ENABLE_COVERAGE})
|
|
# Set coverage source path based on test mode
|
|
if(TEST_FROM_INSTALL)
|
|
set(COV_SRC ".")
|
|
else()
|
|
set(COV_SRC "src")
|
|
endif()
|
|
|
|
set(COV_OPTION
|
|
"--cov=${COV_SRC}"
|
|
"--cov-append"
|
|
"--cov-report=term-missing"
|
|
"--cov-report=lcov:tests/coverage.info"
|
|
)
|
|
|
|
set(CTEST_COVERAGE_COMMAND "python3")
|
|
endif()
|
|
message(STATUS "Code coverage: ${ENABLE_COVERAGE}")
|
|
|
|
# CPU threads available for testing
|
|
set(PYTEST_NUMPROCS
|
|
"1"
|
|
CACHE STRING
|
|
"Number of parallel threads to use with CPU-oriented tests"
|
|
)
|
|
message(STATUS "Pytest CPU threadcount: ${PYTEST_NUMPROCS}")
|
|
|
|
# 2 CPU threads available for testing(test-analyze-commands)
|
|
set(PYTEST_NUMPROCS_ANALYSIS
|
|
"4"
|
|
CACHE STRING
|
|
"Number of parallel threads to use with CPU-oriented tests"
|
|
)
|
|
message(STATUS "Pytest CPU threadcount: ${PYTEST_NUMPROCS_ANALYSIS}")
|
|
|
|
# ---------------------------
|
|
# add tests
|
|
# ---------------------------
|
|
|
|
find_package(MPI COMPONENTS CXX)
|
|
if(MPI_CXX_COMPILER)
|
|
message(STATUS "MPI CXX compiler found: ${MPI_CXX_COMPILER}")
|
|
else()
|
|
message(WARNING "MPI CXX compiler not found. MPI tests will be skipped.")
|
|
endif()
|
|
|
|
option(
|
|
TEST_FROM_INSTALL
|
|
"Test from install directory rather than from source directory"
|
|
OFF
|
|
)
|
|
|
|
if(TEST_FROM_INSTALL OR STANDALONEBINARY)
|
|
set(WORKING_DIR_OPTION "")
|
|
else()
|
|
set(WORKING_DIR_OPTION WORKING_DIRECTORY ${PROJECT_SOURCE_DIR})
|
|
endif()
|
|
|
|
if(TEST_FROM_INSTALL)
|
|
# When testing from install, write coverage to install directory
|
|
set(COVERAGE_XML_PATH
|
|
"${CMAKE_INSTALL_PREFIX}/${CMAKE_INSTALL_LIBEXECDIR}/${PACKAGE_NAME}/coverage.xml"
|
|
)
|
|
set(COVERAGE_DATA_PATH
|
|
"${CMAKE_INSTALL_PREFIX}/${CMAKE_INSTALL_LIBEXECDIR}/${PACKAGE_NAME}/.coverage"
|
|
)
|
|
else()
|
|
# When testing from build, write coverage XML to build directory
|
|
# and coverage data to source directory
|
|
set(COVERAGE_XML_PATH "${CMAKE_BINARY_DIR}/coverage.xml")
|
|
set(COVERAGE_DATA_PATH "${PROJECT_SOURCE_DIR}/.coverage")
|
|
endif()
|
|
|
|
set(STANDALONEBINARY_EXTRACT_DIR
|
|
"/tmp"
|
|
CACHE PATH
|
|
"Absolute path to where the standalone binary will extract its contents"
|
|
)
|
|
|
|
if(STANDALONEBINARY)
|
|
set(STANDALONEBINARY_TEST_OPTION "--call-binary")
|
|
else()
|
|
set(STANDALONEBINARY_TEST_OPTION "")
|
|
endif()
|
|
|
|
option(
|
|
SKIP_NATIVE_TOOL_BUILD
|
|
"Skip building native counter collection tool library (for runtime compilation)"
|
|
OFF
|
|
)
|
|
|
|
# -------------------
|
|
# Setup tool library
|
|
# -------------------
|
|
if(NOT SKIP_NATIVE_TOOL_BUILD)
|
|
add_subdirectory(src/lib)
|
|
endif()
|
|
|
|
# ---------------------------
|
|
# Profile mode import guard
|
|
# ---------------------------
|
|
|
|
# Import guard only works when calling main() directly (not in binary subprocess mode)
|
|
if(NOT STANDALONEBINARY)
|
|
add_test(
|
|
NAME test_import_guard
|
|
COMMAND
|
|
${PYTHON_TEST_COMMAND} -m pytest --junitxml=tests/test_import_guard.xml
|
|
${COV_OPTION} tests/test_import_guard.py ${WORKING_DIR_OPTION}
|
|
)
|
|
endif()
|
|
|
|
# ---------------------------
|
|
# profile mode tests
|
|
# ---------------------------
|
|
|
|
add_test(
|
|
NAME test_profile_kernel_execution
|
|
COMMAND
|
|
${PYTHON_TEST_COMMAND} -m pytest ${STANDALONEBINARY_TEST_OPTION} -m
|
|
kernel_execution --junitxml=tests/test_profile_kernel_execution.xml ${COV_OPTION}
|
|
tests/test_profile_general.py ${WORKING_DIR_OPTION}
|
|
)
|
|
|
|
add_test(
|
|
NAME test_profile_dispatch
|
|
COMMAND
|
|
${PYTHON_TEST_COMMAND} -m pytest ${STANDALONEBINARY_TEST_OPTION} -m dispatch
|
|
--junitxml=tests/test_profile_dispatch.xml ${COV_OPTION}
|
|
tests/test_profile_general.py ${WORKING_DIR_OPTION}
|
|
)
|
|
|
|
add_test(
|
|
NAME test_profile_join
|
|
COMMAND
|
|
${PYTHON_TEST_COMMAND} -m pytest ${STANDALONEBINARY_TEST_OPTION} -m join
|
|
--junitxml=tests/test_profile_join.xml ${COV_OPTION} tests/test_profile_general.py
|
|
${WORKING_DIR_OPTION}
|
|
)
|
|
|
|
add_test(
|
|
NAME test_profile_sort
|
|
COMMAND
|
|
${PYTHON_TEST_COMMAND} -m pytest ${STANDALONEBINARY_TEST_OPTION} -m sort
|
|
--junitxml=tests/test_profile_sort.xml ${COV_OPTION} tests/test_profile_general.py
|
|
${WORKING_DIR_OPTION}
|
|
)
|
|
|
|
add_test(
|
|
NAME test_profile_misc
|
|
COMMAND
|
|
${PYTHON_TEST_COMMAND} -m pytest ${STANDALONEBINARY_TEST_OPTION} -m misc
|
|
--junitxml=tests/test_profile_misc.xml ${COV_OPTION} tests/test_profile_general.py
|
|
${WORKING_DIR_OPTION}
|
|
)
|
|
|
|
add_test(
|
|
NAME test_profile_path
|
|
COMMAND
|
|
${PYTHON_TEST_COMMAND} -m pytest ${STANDALONEBINARY_TEST_OPTION} -m path
|
|
--junitxml=tests/test_profile_path.xml ${COV_OPTION} tests/test_profile_general.py
|
|
${WORKING_DIR_OPTION}
|
|
)
|
|
|
|
add_test(
|
|
NAME test_profile_roofline_1
|
|
COMMAND
|
|
${PYTHON_TEST_COMMAND} -m pytest ${STANDALONEBINARY_TEST_OPTION} -m roofline_1
|
|
--junitxml=tests/test_profile_roofline_1.xml ${COV_OPTION}
|
|
tests/test_profile_general.py ${WORKING_DIR_OPTION}
|
|
)
|
|
|
|
add_test(
|
|
NAME test_profile_roofline_2
|
|
COMMAND
|
|
${PYTHON_TEST_COMMAND} -m pytest ${STANDALONEBINARY_TEST_OPTION} -m roofline_2
|
|
--junitxml=tests/test_profile_roofline_2.xml ${COV_OPTION}
|
|
tests/test_profile_general.py ${WORKING_DIR_OPTION}
|
|
)
|
|
|
|
add_test(
|
|
NAME test_profile_section
|
|
COMMAND
|
|
${PYTHON_TEST_COMMAND} -m pytest ${STANDALONEBINARY_TEST_OPTION} -m section
|
|
--junitxml=tests/test_profile_section.xml ${COV_OPTION}
|
|
tests/test_profile_general.py ${WORKING_DIR_OPTION}
|
|
)
|
|
|
|
add_test(
|
|
NAME test_profile_sets_func
|
|
COMMAND
|
|
${PYTHON_TEST_COMMAND} -m pytest ${STANDALONEBINARY_TEST_OPTION} -m sets_func
|
|
--junitxml=tests/test_profile_sets_func.xml ${COV_OPTION}
|
|
tests/test_profile_general.py ${WORKING_DIR_OPTION}
|
|
)
|
|
|
|
add_test(
|
|
NAME test_profile_live_attach_detach
|
|
COMMAND
|
|
${PYTHON_TEST_COMMAND} -m pytest ${STANDALONEBINARY_TEST_OPTION} -s -m
|
|
live_attach_detach --junitxml=tests/test_profile_live_attach_detach.xml
|
|
${COV_OPTION} tests/test_profile_general.py ${WORKING_DIR_OPTION}
|
|
)
|
|
set_tests_properties(
|
|
test_profile_live_attach_detach
|
|
PROPERTIES TIMEOUT 300 LABELS "profile" RESOURCE_GROUPS gpus:1
|
|
)
|
|
|
|
add_test(
|
|
NAME test_profile_iteration_multiplexing_1
|
|
COMMAND
|
|
${PYTHON_TEST_COMMAND} -m pytest ${STANDALONEBINARY_TEST_OPTION} -s -m
|
|
iteration_multiplexing_1
|
|
--junitxml=tests/test_profile_iteration_multiplexing_1.xml ${COV_OPTION}
|
|
tests/test_profile_general.py ${WORKING_DIR_OPTION}
|
|
)
|
|
|
|
add_test(
|
|
NAME test_profile_iteration_multiplexing_2
|
|
COMMAND
|
|
${PYTHON_TEST_COMMAND} -m pytest ${STANDALONEBINARY_TEST_OPTION} -s -m
|
|
iteration_multiplexing_2
|
|
--junitxml=tests/test_profile_iteration_multiplexing_2.xml ${COV_OPTION}
|
|
tests/test_profile_general.py ${WORKING_DIR_OPTION}
|
|
)
|
|
|
|
add_test(
|
|
NAME test_profile_iteration_multiplexing_stochastic
|
|
COMMAND
|
|
${PYTHON_TEST_COMMAND} -m pytest ${STANDALONEBINARY_TEST_OPTION} -s -m
|
|
iteration_multiplexing_stochastic
|
|
--junitxml=tests/test_profile_iteration_multiplexing_stochastic.xml ${COV_OPTION}
|
|
tests/test_profile_general.py ${WORKING_DIR_OPTION}
|
|
)
|
|
|
|
if(MPI_CXX_COMPILER)
|
|
add_test(
|
|
NAME test_profile_multi_rank
|
|
COMMAND
|
|
${PYTHON_TEST_COMMAND} -m pytest -s -m multi_rank
|
|
--junitxml=tests/test_profile_multi_rank.xml ${COV_OPTION}
|
|
tests/test_profile_general.py ${WORKING_DIR_OPTION}
|
|
)
|
|
set_tests_properties(
|
|
test_profile_multi_rank
|
|
PROPERTIES LABELS "profile" RESOURCE_GROUPS gpus:2 TIMEOUT 1800
|
|
)
|
|
endif()
|
|
|
|
add_test(
|
|
NAME test_metric_validation
|
|
COMMAND
|
|
${PYTHON_TEST_COMMAND} -m pytest ${STANDALONEBINARY_TEST_OPTION}
|
|
--junitxml=tests/test_metric_validation.xml ${COV_OPTION}
|
|
tests/test_metric_validation.py ${WORKING_DIR_OPTION}
|
|
)
|
|
|
|
add_test(
|
|
NAME test_native_tool_finder
|
|
COMMAND
|
|
${PYTHON_TEST_COMMAND} -m pytest ${STANDALONEBINARY_TEST_OPTION}
|
|
--junitxml=tests/test_native_tool_finder.xml ${COV_OPTION}
|
|
tests/test_native_tool_finder.py ${WORKING_DIR_OPTION}
|
|
)
|
|
|
|
add_test(
|
|
NAME test_profile_torch_trace
|
|
COMMAND
|
|
${PYTHON_TEST_COMMAND} -m pytest ${STANDALONEBINARY_TEST_OPTION} -m torch_trace
|
|
--junitxml=tests/test_profile_torch_trace.xml ${COV_OPTION}
|
|
tests/test_profile_general.py ${WORKING_DIR_OPTION}
|
|
)
|
|
|
|
add_test(
|
|
NAME test_profile_triton_trace
|
|
COMMAND
|
|
${PYTHON_TEST_COMMAND} -m pytest ${STANDALONEBINARY_TEST_OPTION} -m triton_trace
|
|
--junitxml=tests/test_profile_triton_trace.xml ${COV_OPTION}
|
|
tests/test_profile_general.py ${WORKING_DIR_OPTION}
|
|
)
|
|
|
|
set_tests_properties(
|
|
test_profile_kernel_execution
|
|
test_profile_dispatch
|
|
test_profile_join
|
|
test_profile_sort
|
|
test_profile_misc
|
|
test_profile_path
|
|
test_profile_roofline_1
|
|
test_profile_roofline_2
|
|
test_profile_section
|
|
test_profile_sets_func
|
|
test_profile_iteration_multiplexing_1
|
|
test_profile_iteration_multiplexing_2
|
|
test_profile_iteration_multiplexing_stochastic
|
|
test_profile_torch_trace
|
|
test_profile_triton_trace
|
|
PROPERTIES LABELS "profile" RESOURCE_GROUPS gpus:1 TIMEOUT 1800
|
|
)
|
|
|
|
# ---------------------------
|
|
# analysis command tests
|
|
# ---------------------------
|
|
|
|
add_test(
|
|
NAME test_analyze_commands
|
|
COMMAND
|
|
${PYTHON_TEST_COMMAND} -m pytest ${STANDALONEBINARY_TEST_OPTION} -n
|
|
${PYTEST_NUMPROCS_ANALYSIS} --verbose --junitxml=tests/test_analyze_commands.xml
|
|
${COV_OPTION} tests/test_analyze_commands.py ${WORKING_DIR_OPTION}
|
|
)
|
|
|
|
# ---------------------------
|
|
# analyze workloads tests
|
|
# ---------------------------
|
|
|
|
add_test(
|
|
NAME test_analyze_workloads
|
|
COMMAND
|
|
${PYTHON_TEST_COMMAND} -m pytest ${STANDALONEBINARY_TEST_OPTION} -n
|
|
${PYTEST_NUMPROCS} --junitxml=tests/test_analyze_workloads.xml ${COV_OPTION}
|
|
tests/test_analyze_workloads.py ${WORKING_DIR_OPTION}
|
|
)
|
|
|
|
# ---------------------------
|
|
# TCP counter tests
|
|
# ---------------------------
|
|
|
|
add_test(
|
|
NAME test_L1_cache_counters
|
|
COMMAND
|
|
${PYTHON_TEST_COMMAND} -m pytest ${STANDALONEBINARY_TEST_OPTION} -m L1_cache
|
|
--junitxml=tests/test_L1_cache_counters.xml ${COV_OPTION}
|
|
tests/test_TCP_counters.py ${WORKING_DIR_OPTION}
|
|
)
|
|
|
|
# ---------------------------
|
|
# Spec tests
|
|
# ---------------------------
|
|
|
|
add_test(
|
|
NAME test_num_xcds_spec_class
|
|
COMMAND
|
|
${PYTHON_TEST_COMMAND} -m pytest ${STANDALONEBINARY_TEST_OPTION} -m
|
|
num_xcds_spec_class --junitxml=tests/test_num_xcds_spec_class.xml ${COV_OPTION}
|
|
tests/test_gpu_specs.py ${WORKING_DIR_OPTION}
|
|
)
|
|
|
|
add_test(
|
|
NAME test_num_xcds_cli_output
|
|
COMMAND
|
|
${PYTHON_TEST_COMMAND} -m pytest ${STANDALONEBINARY_TEST_OPTION} -m
|
|
num_xcds_cli_output --junitxml=tests/test_num_xcds_cli_output.xml ${COV_OPTION}
|
|
tests/test_gpu_specs.py ${WORKING_DIR_OPTION}
|
|
)
|
|
|
|
add_test(
|
|
NAME test_mi_gpu_spec
|
|
COMMAND
|
|
${PYTHON_TEST_COMMAND} -m pytest ${STANDALONEBINARY_TEST_OPTION}
|
|
--junitxml=tests/test_mi_gpu_spec.xml ${COV_OPTION} tests/test_mi_gpu_spec.py
|
|
${WORKING_DIR_OPTION}
|
|
)
|
|
|
|
# ---------------------------
|
|
# Utils tests
|
|
# ---------------------------
|
|
|
|
add_test(
|
|
NAME test_utils
|
|
COMMAND
|
|
${PYTHON_TEST_COMMAND} -m pytest ${STANDALONEBINARY_TEST_OPTION}
|
|
--junitxml=tests/test_utils.xml ${COV_OPTION} tests/test_utils.py
|
|
${WORKING_DIR_OPTION}
|
|
)
|
|
|
|
# Memory chart CLI renderers (gfx9 plotille + gfx11 Rich) — no GPU required
|
|
add_test(
|
|
NAME test_mem_chart
|
|
COMMAND
|
|
${PYTHON_TEST_COMMAND} -m pytest ${STANDALONEBINARY_TEST_OPTION}
|
|
--junitxml=tests/test_mem_chart.xml ${COV_OPTION} tests/test_mem_chart.py
|
|
${WORKING_DIR_OPTION}
|
|
)
|
|
|
|
add_test(
|
|
NAME test_roofline_calc
|
|
COMMAND
|
|
${PYTHON_TEST_COMMAND} -m pytest ${STANDALONEBINARY_TEST_OPTION}
|
|
--junitxml=tests/test_roofline_calc.xml ${COV_OPTION} tests/test_roofline_calc.py
|
|
${WORKING_DIR_OPTION}
|
|
)
|
|
|
|
add_test(
|
|
NAME test_roofline_analyze
|
|
COMMAND
|
|
${PYTHON_TEST_COMMAND} -m pytest ${STANDALONEBINARY_TEST_OPTION}
|
|
--junitxml=tests/test_roofline_analyze.xml ${COV_OPTION}
|
|
tests/test_roofline_analyze.py ${WORKING_DIR_OPTION}
|
|
)
|
|
|
|
add_test(
|
|
NAME test_roofline_main
|
|
COMMAND
|
|
${PYTHON_TEST_COMMAND} -m pytest ${STANDALONEBINARY_TEST_OPTION}
|
|
--junitxml=tests/test_roofline_main.xml ${COV_OPTION} tests/test_roofline_main.py
|
|
${WORKING_DIR_OPTION}
|
|
)
|
|
|
|
# -----------------------------------
|
|
# Data imputation tests
|
|
# -----------------------------------
|
|
|
|
add_test(
|
|
NAME test_data_imputation
|
|
COMMAND
|
|
${PYTHON_TEST_COMMAND} -m pytest ${STANDALONEBINARY_TEST_OPTION}
|
|
--junitxml=tests/test_data_imputation.xml ${COV_OPTION}
|
|
tests/test_data_imputation.py ${WORKING_DIR_OPTION}
|
|
)
|
|
|
|
# -----------------------------------
|
|
# Analysis database tests
|
|
# -----------------------------------
|
|
|
|
add_test(
|
|
NAME test_analysis_db
|
|
COMMAND
|
|
${PYTHON_TEST_COMMAND} -m pytest ${STANDALONEBINARY_TEST_OPTION}
|
|
--junitxml=tests/test_analysis_db.xml ${COV_OPTION} tests/test_analysis_db.py
|
|
${WORKING_DIR_OPTION}
|
|
)
|
|
|
|
add_test(
|
|
NAME test_analysis_orm
|
|
COMMAND
|
|
${PYTHON_TEST_COMMAND} -m pytest ${STANDALONEBINARY_TEST_OPTION}
|
|
--junitxml=tests/test_analysis_orm.xml ${COV_OPTION} tests/test_analysis_orm.py
|
|
${WORKING_DIR_OPTION}
|
|
)
|
|
|
|
add_test(
|
|
NAME test_code_object_analysis
|
|
COMMAND
|
|
${PYTHON_TEST_COMMAND} -m pytest ${STANDALONEBINARY_TEST_OPTION}
|
|
--junitxml=tests/test_code_object_analysis.xml ${COV_OPTION}
|
|
tests/test_code_object_analysis.py ${WORKING_DIR_OPTION}
|
|
)
|
|
|
|
# -----------------------------------
|
|
# Metrics unit tests
|
|
# -----------------------------------
|
|
|
|
add_test(
|
|
NAME test_metric_utils
|
|
COMMAND
|
|
${PYTHON_TEST_COMMAND} -m pytest ${STANDALONEBINARY_TEST_OPTION}
|
|
--junitxml=tests/test_metric_utils.xml ${COV_OPTION} tests/test_metric_utils.py
|
|
${WORKING_DIR_OPTION}
|
|
)
|
|
|
|
add_test(
|
|
NAME test_parser
|
|
COMMAND
|
|
${PYTHON_TEST_COMMAND} -m pytest ${STANDALONEBINARY_TEST_OPTION}
|
|
--junitxml=tests/test_parser.xml ${COV_OPTION} tests/test_parser.py
|
|
${WORKING_DIR_OPTION}
|
|
)
|
|
|
|
add_test(
|
|
NAME test_metrics_common
|
|
COMMAND
|
|
${PYTHON_TEST_COMMAND} -m pytest ${STANDALONEBINARY_TEST_OPTION}
|
|
--junitxml=tests/test_metrics_common.xml ${COV_OPTION}
|
|
tests/test_metrics_common.py ${WORKING_DIR_OPTION}
|
|
)
|
|
|
|
# -----------------------------------
|
|
# CLI options unit tests
|
|
# -----------------------------------
|
|
|
|
add_test(
|
|
NAME test_argparser
|
|
COMMAND
|
|
${PYTHON_TEST_COMMAND} -m pytest ${STANDALONEBINARY_TEST_OPTION}
|
|
--junitxml=tests/test_argparser.xml ${COV_OPTION} tests/test_argparser.py
|
|
${WORKING_DIR_OPTION}
|
|
)
|
|
|
|
# ---------------------------
|
|
# PC sampling tests
|
|
# ---------------------------
|
|
|
|
add_test(
|
|
NAME test_pc_sampling
|
|
COMMAND
|
|
${PYTHON_TEST_COMMAND} -m pytest ${STANDALONEBINARY_TEST_OPTION} -s
|
|
--junitxml=tests/test_pc_sampling.xml ${COV_OPTION} tests/test_pc_sampling.py
|
|
${WORKING_DIR_OPTION}
|
|
)
|
|
|
|
add_test(
|
|
NAME test_pc_sampling_analysis
|
|
COMMAND
|
|
${PYTHON_TEST_COMMAND} -m pytest ${STANDALONEBINARY_TEST_OPTION} -s
|
|
--junitxml=tests/test_pc_sampling_analysis.xml ${COV_OPTION}
|
|
tests/test_pc_sampling_analysis.py ${WORKING_DIR_OPTION}
|
|
)
|
|
|
|
set_tests_properties(
|
|
test_pc_sampling
|
|
test_pc_sampling_analysis
|
|
PROPERTIES TIMEOUT 300 LABELS "profile" RESOURCE_GROUPS gpus:1
|
|
)
|
|
|
|
# -----------------------------------
|
|
# TUI component tests
|
|
# -----------------------------------
|
|
|
|
add_test(
|
|
NAME test_tui_components
|
|
COMMAND
|
|
${PYTHON_TEST_COMMAND} -m pytest ${STANDALONEBINARY_TEST_OPTION}
|
|
--junitxml=tests/test_tui_components.xml ${COV_OPTION} -m tui
|
|
tests/test_tui_components.py ${WORKING_DIR_OPTION}
|
|
)
|
|
|
|
# Build the roctx_recordfn pybind11 extension and its gtest.
|
|
if(ENABLE_TESTS)
|
|
add_subdirectory(src/lib/roctx_recordfn)
|
|
endif()
|
|
# -----------------------------------
|
|
# Torch trace tests
|
|
# -----------------------------------
|
|
|
|
add_test(
|
|
NAME test_torch_trace_analysis
|
|
COMMAND
|
|
${PYTHON_TEST_COMMAND} -m pytest ${STANDALONEBINARY_TEST_OPTION}
|
|
--junitxml=tests/test_torch_trace_analysis.xml ${COV_OPTION}
|
|
tests/test_torch_trace_analysis.py ${WORKING_DIR_OPTION}
|
|
)
|
|
|
|
add_test(
|
|
NAME test_torch_trace_coverage
|
|
COMMAND
|
|
${PYTHON_TEST_COMMAND} -m pytest ${STANDALONEBINARY_TEST_OPTION}
|
|
--junitxml=tests/test_torch_trace_coverage.xml ${COV_OPTION}
|
|
tests/test_torch_trace_coverage.py ${WORKING_DIR_OPTION}
|
|
)
|
|
|
|
set_tests_properties(
|
|
test_torch_trace_coverage
|
|
PROPERTIES TIMEOUT 1800 LABELS "profile" RESOURCE_GROUPS gpus:1
|
|
)
|
|
|
|
add_test(
|
|
NAME test_torch_cpp_loader
|
|
COMMAND
|
|
${PYTHON_TEST_COMMAND} -m pytest ${STANDALONEBINARY_TEST_OPTION}
|
|
--junitxml=tests/test_torch_cpp_loader.xml ${COV_OPTION}
|
|
tests/test_torch_cpp_loader.py ${WORKING_DIR_OPTION}
|
|
)
|
|
|
|
add_test(
|
|
NAME test_inject_roctx_package
|
|
COMMAND
|
|
${PYTHON_TEST_COMMAND} -m pytest ${STANDALONEBINARY_TEST_OPTION}
|
|
--junitxml=tests/test_inject_roctx_package.xml ${COV_OPTION}
|
|
tests/test_inject_roctx_package.py ${WORKING_DIR_OPTION}
|
|
)
|
|
|
|
# ---------------------------------------
|
|
# Pure csv compatibility layer for pandas
|
|
# ---------------------------------------
|
|
|
|
add_test(
|
|
NAME test_utils_profile_csv
|
|
COMMAND
|
|
${PYTHON_TEST_COMMAND} -m pytest ${STANDALONEBINARY_TEST_OPTION}
|
|
--junitxml=tests/test_utils_profile_csv.xml ${COV_OPTION}
|
|
tests/test_utils_profile_csv.py ${WORKING_DIR_OPTION}
|
|
)
|
|
|
|
# -----------------------------------
|
|
# Profiler base unit tests
|
|
# -----------------------------------
|
|
|
|
add_test(
|
|
NAME test_profiler_base
|
|
COMMAND
|
|
${PYTHON_TEST_COMMAND} -m pytest ${STANDALONEBINARY_TEST_OPTION}
|
|
--junitxml=tests/test_profiler_base.xml ${COV_OPTION} tests/test_profiler_base.py
|
|
${WORKING_DIR_OPTION}
|
|
)
|
|
|
|
# -----------------------------------
|
|
# PC sampling profiler unit tests
|
|
# -----------------------------------
|
|
|
|
add_test(
|
|
NAME test_pc_sampling_profile
|
|
COMMAND
|
|
${PYTHON_TEST_COMMAND} -m pytest ${STANDALONEBINARY_TEST_OPTION}
|
|
--junitxml=tests/test_pc_sampling_profile.xml ${COV_OPTION}
|
|
tests/test_pc_sampling_profile.py ${WORKING_DIR_OPTION}
|
|
)
|
|
|
|
# -----------------------------------
|
|
# SoC base counter allocation tests
|
|
# -----------------------------------
|
|
|
|
add_test(
|
|
NAME test_soc_base
|
|
COMMAND
|
|
${PYTHON_TEST_COMMAND} -m pytest ${STANDALONEBINARY_TEST_OPTION}
|
|
--junitxml=tests/test_soc_base.xml ${COV_OPTION} tests/test_soc_base.py
|
|
${WORKING_DIR_OPTION}
|
|
)
|
|
|
|
# -----------------------------------
|
|
# TTY table rendering unit tests
|
|
# -----------------------------------
|
|
|
|
add_test(
|
|
NAME test_tty
|
|
COMMAND
|
|
${PYTHON_TEST_COMMAND} -m pytest ${STANDALONEBINARY_TEST_OPTION}
|
|
--junitxml=tests/test_tty.xml ${COV_OPTION} tests/test_tty.py
|
|
${WORKING_DIR_OPTION}
|
|
)
|
|
|
|
if(TARGET test-rocprofiler-compute-tool)
|
|
if(TEST_FROM_INSTALL)
|
|
set(_tool_test_cmd tests/test-rocprofiler-compute-tool)
|
|
else()
|
|
set(_tool_test_cmd $<TARGET_FILE:test-rocprofiler-compute-tool>)
|
|
endif()
|
|
add_test(NAME test-rocprofiler-compute-tool COMMAND ${_tool_test_cmd})
|
|
sanitizer_runtime_env(_san_env OFF)
|
|
if(_san_env)
|
|
set_tests_properties(
|
|
test-rocprofiler-compute-tool
|
|
PROPERTIES ENVIRONMENT "${_san_env}"
|
|
)
|
|
endif()
|
|
endif()
|
|
|
|
if(TARGET test-pc-sampling-collector)
|
|
if(TEST_FROM_INSTALL)
|
|
set(_pcs_test_cmd tests/test-pc-sampling-collector)
|
|
else()
|
|
set(_pcs_test_cmd $<TARGET_FILE:test-pc-sampling-collector>)
|
|
endif()
|
|
add_test(NAME test-pc-sampling-collector COMMAND ${_pcs_test_cmd})
|
|
sanitizer_runtime_env(_san_env OFF)
|
|
if(_san_env)
|
|
set_tests_properties(
|
|
test-pc-sampling-collector
|
|
PROPERTIES ENVIRONMENT "${_san_env}"
|
|
)
|
|
endif()
|
|
endif()
|
|
|
|
# -----------------------------------
|
|
# Generate coverage tests
|
|
# -----------------------------------
|
|
|
|
if(${ENABLE_COVERAGE})
|
|
set(ALL_TEST_NAMES
|
|
test_profile_kernel_execution
|
|
test_profile_dispatch
|
|
test_profile_join
|
|
test_profile_sort
|
|
test_profile_misc
|
|
test_profile_path
|
|
test_profile_roofline_1
|
|
test_profile_roofline_2
|
|
test_profile_section
|
|
test_profile_sets_func
|
|
test_pc_sampling
|
|
test_analyze_commands
|
|
test_analyze_workloads
|
|
test_metric_utils
|
|
test_parser
|
|
test_argparser
|
|
test_L1_cache_counters
|
|
test_num_xcds_spec_class
|
|
test_num_xcds_cli_output
|
|
test_mi_gpu_spec
|
|
test_utils
|
|
test_mem_chart
|
|
test_roofline_calc
|
|
test_roofline_analyze
|
|
test_roofline_main
|
|
test_data_imputation
|
|
test_tui_components
|
|
test_torch_trace_analysis
|
|
test_torch_trace_coverage
|
|
test_profile_torch_trace
|
|
test_profile_triton_trace
|
|
test_torch_cpp_loader
|
|
test_inject_roctx_package
|
|
test_analysis_db
|
|
test_analysis_orm
|
|
test_code_object_analysis
|
|
test_pc_sampling_analysis
|
|
test_pc_sampling_profile
|
|
test_metrics_common
|
|
test_metric_validation
|
|
test_native_tool_finder
|
|
test_profiler_base
|
|
test_soc_base
|
|
test_tty
|
|
test_utils_profile_csv
|
|
test_import_guard
|
|
test_profile_iteration_multiplexing_1
|
|
test_profile_iteration_multiplexing_2
|
|
test_profile_iteration_multiplexing_stochastic
|
|
test_profile_live_attach_detach
|
|
test_profile_multi_rank
|
|
)
|
|
|
|
string(REPLACE ";" ";" ALL_TESTS_STRING "${ALL_TEST_NAMES}")
|
|
|
|
add_test(
|
|
NAME generate_coverage_report
|
|
COMMAND
|
|
${PYTHON_TEST_COMMAND} -m coverage xml -o ${COVERAGE_XML_PATH}
|
|
${WORKING_DIR_OPTION}
|
|
)
|
|
|
|
set_tests_properties(
|
|
generate_coverage_report
|
|
PROPERTIES
|
|
DEPENDS "${ALL_TESTS_STRING}"
|
|
LABELS "coverage"
|
|
ENVIRONMENT "COVERAGE_FILE=${COVERAGE_DATA_PATH}"
|
|
TIMEOUT 600
|
|
)
|
|
endif()
|
|
|
|
# -------------------
|
|
# Test categorization
|
|
# -------------------
|
|
set(_test_categories_yaml "${CMAKE_CURRENT_SOURCE_DIR}/tests/test_categories.yaml")
|
|
include(${ROCM_SYSTEMS_ROOT}/shared/ctest/TestCategories.cmake OPTIONAL)
|
|
|
|
if(COMMAND apply_ctest_category_labels)
|
|
if(EXISTS "${_test_categories_yaml}")
|
|
message(STATUS "YAML-based test categorization")
|
|
apply_ctest_category_labels("${_test_categories_yaml}")
|
|
else()
|
|
message(STATUS "No test categorization (YAML missing)")
|
|
endif()
|
|
else()
|
|
message(STATUS "Test categorization not available (TestCategories.cmake not found)")
|
|
endif()
|
|
|
|
# ---------
|
|
# Install
|
|
# ---------
|
|
|
|
# top-level rocprofiler-compute utility
|
|
install(
|
|
PROGRAMS src/${EXECUTABLE_NAME}
|
|
DESTINATION ${CMAKE_INSTALL_LIBEXECDIR}/${PROJECT_NAME}
|
|
COMPONENT main
|
|
)
|
|
# python dependency requirements
|
|
install(
|
|
FILES requirements.txt
|
|
DESTINATION ${CMAKE_INSTALL_LIBEXECDIR}/${PROJECT_NAME}
|
|
COMPONENT main
|
|
)
|
|
# support files and version info
|
|
install(
|
|
FILES src/argparser.py src/config.py src/rocprof_compute_base.py VERSION VERSION.sha
|
|
DESTINATION ${CMAKE_INSTALL_LIBEXECDIR}/${PROJECT_NAME}
|
|
COMPONENT main
|
|
)
|
|
# src/rocprof_compute_analyze
|
|
install(
|
|
DIRECTORY src/rocprof_compute_analyze
|
|
DESTINATION ${CMAKE_INSTALL_LIBEXECDIR}/${PROJECT_NAME}
|
|
COMPONENT main
|
|
PATTERN src/rocprof_compute_analyze/tests EXCLUDE
|
|
PATTERN "__pycache__" EXCLUDE
|
|
)
|
|
# src/utils
|
|
install(
|
|
DIRECTORY src/utils
|
|
DESTINATION ${CMAKE_INSTALL_LIBEXECDIR}/${PROJECT_NAME}
|
|
COMPONENT main
|
|
PATTERN "rooflines*" EXCLUDE
|
|
PATTERN "__pycache__" EXCLUDE
|
|
)
|
|
# Install the roctx_recordfn sources so the runtime loader can build the
|
|
# extension against the user's PyTorch environment on first use.
|
|
install(
|
|
DIRECTORY src/lib/roctx_recordfn
|
|
DESTINATION ${CMAKE_INSTALL_LIBEXECDIR}/${PROJECT_NAME}/lib
|
|
COMPONENT main
|
|
PATTERN "_build" EXCLUDE
|
|
PATTERN "build" EXCLUDE
|
|
PATTERN "tests" EXCLUDE
|
|
PATTERN "__pycache__" EXCLUDE
|
|
)
|
|
# src/roofline
|
|
install(
|
|
DIRECTORY src/roofline
|
|
DESTINATION ${CMAKE_INSTALL_LIBEXECDIR}/${PROJECT_NAME}
|
|
COMPONENT main
|
|
PATTERN "__pycache__" EXCLUDE
|
|
)
|
|
# src/rocprof_compute_soc
|
|
install(
|
|
DIRECTORY src/rocprof_compute_soc
|
|
DESTINATION ${CMAKE_INSTALL_LIBEXECDIR}/${PROJECT_NAME}
|
|
COMPONENT main
|
|
PATTERN "__pycache__" EXCLUDE
|
|
)
|
|
# src/rocprof_compute_profile
|
|
install(
|
|
DIRECTORY src/rocprof_compute_profile
|
|
DESTINATION ${CMAKE_INSTALL_LIBEXECDIR}/${PROJECT_NAME}
|
|
COMPONENT main
|
|
PATTERN "__pycache__" EXCLUDE
|
|
)
|
|
# src/pc_sampling
|
|
install(
|
|
DIRECTORY src/pc_sampling
|
|
DESTINATION ${CMAKE_INSTALL_LIBEXECDIR}/${PROJECT_NAME}
|
|
COMPONENT main
|
|
PATTERN "__pycache__" EXCLUDE
|
|
)
|
|
# src/rocprof_compute_tui
|
|
install(
|
|
DIRECTORY src/rocprof_compute_tui
|
|
DESTINATION ${CMAKE_INSTALL_LIBEXECDIR}/${PROJECT_NAME}
|
|
COMPONENT main
|
|
PATTERN "__pycache__" EXCLUDE
|
|
)
|
|
|
|
# ==============================================================================
|
|
# Vendored Dependencies Installation
|
|
# ==============================================================================
|
|
# Vendored packages are included as git submodules and installed as pure Python.
|
|
# Each package has: conditional submodule init + unconditional install.
|
|
# See src/vendored/README.md and CONTRIBUTING.md for vendoring guidelines.
|
|
|
|
find_package(Git QUIET)
|
|
|
|
# Vendored package wrapper
|
|
install(
|
|
FILES src/vendored/__init__.py
|
|
DESTINATION ${CMAKE_INSTALL_LIBEXECDIR}/${PROJECT_NAME}/vendored
|
|
COMPONENT main
|
|
)
|
|
|
|
# PyYAML
|
|
if(NOT EXISTS "${PROJECT_SOURCE_DIR}/src/vendored/pyyaml/lib/yaml")
|
|
if(Git_FOUND)
|
|
message(STATUS "Initializing vendored/pyyaml submodule...")
|
|
execute_process(
|
|
COMMAND
|
|
${GIT_EXECUTABLE} submodule update --init --recursive src/vendored/pyyaml
|
|
WORKING_DIRECTORY ${CMAKE_CURRENT_SOURCE_DIR}
|
|
RESULT_VARIABLE GIT_SUBMOD_RESULT
|
|
)
|
|
if(NOT GIT_SUBMOD_RESULT EQUAL "0")
|
|
message(FATAL_ERROR "git submodule update failed for src/vendored/pyyaml")
|
|
endif()
|
|
else()
|
|
message(FATAL_ERROR "vendored/pyyaml not found and git not available")
|
|
endif()
|
|
else()
|
|
message(STATUS "Vendored/pyyaml: OK")
|
|
endif()
|
|
install(
|
|
DIRECTORY src/vendored/pyyaml/lib/yaml
|
|
DESTINATION ${CMAKE_INSTALL_LIBEXECDIR}/${PROJECT_NAME}/vendored/pyyaml/lib
|
|
COMPONENT main
|
|
FILES_MATCHING
|
|
PATTERN "*.py"
|
|
PATTERN "__pycache__" EXCLUDE
|
|
PATTERN "*.pyc" EXCLUDE
|
|
)
|
|
|
|
# samples
|
|
install(
|
|
DIRECTORY sample
|
|
DESTINATION ${CMAKE_INSTALL_DATADIR}/${PROJECT_NAME}
|
|
COMPONENT main
|
|
FILES_MATCHING
|
|
PATTERN "*.hip"
|
|
PATTERN "*.h"
|
|
PATTERN "*.cpp"
|
|
PATTERN "*.py"
|
|
PATTERN "workloads" EXCLUDE
|
|
)
|
|
# modulefile
|
|
install(
|
|
FILES ${PROJECT_BINARY_DIR}/${MOD_INSTALL_PATH}/${ROCPROFCOMPUTE_FULL_VERSION}.lua
|
|
DESTINATION ${MOD_INSTALL_PATH}
|
|
COMPONENT main
|
|
)
|
|
|
|
# install librocprofiler-compute-tool.so or source files
|
|
if(NOT SKIP_NATIVE_TOOL_BUILD)
|
|
install(
|
|
TARGETS rocprofiler-compute-tool
|
|
LIBRARY DESTINATION ${CMAKE_INSTALL_LIBDIR}/rocprofiler-compute COMPONENT main
|
|
)
|
|
else()
|
|
# Install source files for runtime compilation. roctx_recordfn/ is
|
|
# installed by a dedicated rule above and excluded here.
|
|
install(
|
|
DIRECTORY src/lib/
|
|
DESTINATION ${CMAKE_INSTALL_LIBEXECDIR}/${PROJECT_NAME}/lib
|
|
COMPONENT main
|
|
PATTERN "_build" EXCLUDE
|
|
PATTERN "tests" EXCLUDE
|
|
PATTERN "roctx_recordfn" EXCLUDE
|
|
)
|
|
endif()
|
|
|
|
# top-level bin/rocprof-compute launcher (re-execs into libexec entrypoint)
|
|
configure_file(
|
|
${PROJECT_SOURCE_DIR}/cmake/${EXECUTABLE_NAME}.in
|
|
${PROJECT_BINARY_DIR}/${EXECUTABLE_NAME}
|
|
@ONLY
|
|
)
|
|
install(
|
|
PROGRAMS ${PROJECT_BINARY_DIR}/${EXECUTABLE_NAME}
|
|
DESTINATION ${CMAKE_INSTALL_BINDIR}
|
|
COMPONENT main
|
|
)
|
|
|
|
install(
|
|
FILES ${PROJECT_SOURCE_DIR}/LICENSE.md
|
|
DESTINATION ${CMAKE_INSTALL_DOCDIR}
|
|
COMPONENT main
|
|
)
|
|
|
|
# TEST collateral
|
|
if(INSTALL_TESTS)
|
|
install(
|
|
DIRECTORY tests
|
|
DESTINATION ${CMAKE_INSTALL_LIBEXECDIR}/${PROJECT_NAME}
|
|
COMPONENT tests
|
|
FILES_MATCHING
|
|
PATTERN "*.py"
|
|
PATTERN "__pycache__" EXCLUDE
|
|
)
|
|
install(
|
|
PROGRAMS therock_test_runner.py
|
|
DESTINATION ${CMAKE_INSTALL_LIBEXECDIR}/${PROJECT_NAME}
|
|
COMPONENT tests
|
|
)
|
|
install(
|
|
DIRECTORY tests/workloads
|
|
DESTINATION ${CMAKE_INSTALL_LIBEXECDIR}/${PROJECT_NAME}/tests
|
|
COMPONENT tests
|
|
PATTERN "__pycache__" EXCLUDE
|
|
)
|
|
install(
|
|
PROGRAMS
|
|
tests/vcopy
|
|
tests/vmem
|
|
tests/vrandom_access
|
|
tests/vsequential_access
|
|
tests/occupancy
|
|
tests/hip_dynamic_shared
|
|
tests/laplace_eqn
|
|
tests/mat_mul_max
|
|
tests/rocflop
|
|
tests/memcopy
|
|
DESTINATION ${CMAKE_INSTALL_LIBEXECDIR}/${PROJECT_NAME}/tests
|
|
COMPONENT tests
|
|
)
|
|
if(MPI_CXX_COMPILER)
|
|
install(
|
|
PROGRAMS tests/mpi_aware_laplace_eqn
|
|
DESTINATION ${CMAKE_INSTALL_LIBEXECDIR}/${PROJECT_NAME}/tests
|
|
COMPONENT tests
|
|
)
|
|
endif()
|
|
install(
|
|
FILES requirements-test.txt pyproject.toml
|
|
DESTINATION ${CMAKE_INSTALL_LIBEXECDIR}/${PROJECT_NAME}
|
|
COMPONENT tests
|
|
)
|
|
install(
|
|
FILES ${CMAKE_CURRENT_BINARY_DIR}/CTestTestfile.cmake
|
|
COMPONENT tests
|
|
DESTINATION ${CMAKE_INSTALL_LIBEXECDIR}/${PROJECT_NAME}
|
|
)
|
|
endif()
|
|
message(STATUS "Install tests: ${INSTALL_TESTS}")
|
|
|
|
# Standalone binary creation during install
|
|
if(STANDALONEBINARY)
|
|
# We use Nuitka 2.6 which requires exactly Python 3.11.x (any patch version)
|
|
find_package(Python3 3.11 COMPONENTS Interpreter REQUIRED)
|
|
|
|
# Verify exactly 3.11.x (not 3.10.x or 3.12.x)
|
|
if(NOT (Python3_VERSION_MAJOR EQUAL 3 AND Python3_VERSION_MINOR EQUAL 11))
|
|
message(
|
|
FATAL_ERROR
|
|
"Standalone binary requires exactly Python 3.11.x (any patch version). "
|
|
"Found: Python ${Python3_VERSION}"
|
|
)
|
|
endif()
|
|
|
|
install(
|
|
CODE
|
|
"
|
|
message(STATUS \"Building standalone binary...\")
|
|
# Note: Variables defined here use \${VAR} escaping so they're evaluated at install time
|
|
# rather than configure time. Variables from configure time like ${PROJECT_BINARY_DIR}
|
|
# are expanded when CMake processes this file.
|
|
set(VENV_DIR \"${PROJECT_BINARY_DIR}/nuitka-build-venv\")
|
|
# Create virtual environment
|
|
message(STATUS \"Creating virtual environment at \${VENV_DIR}\")
|
|
execute_process(COMMAND ${Python3_EXECUTABLE} -m venv \${VENV_DIR} RESULT_VARIABLE VENV_RESULT)
|
|
if(NOT VENV_RESULT EQUAL 0)
|
|
message(FATAL_ERROR \"Failed to create virtual environment\")
|
|
endif()
|
|
set(VENV_PYTHON \"\${VENV_DIR}/bin/python\")
|
|
# Install requirements
|
|
message(STATUS \"Installing requirements.txt\")
|
|
execute_process(COMMAND \${VENV_PYTHON} -m pip install -r ${PROJECT_SOURCE_DIR}/requirements.txt RESULT_VARIABLE REQUIREMENTS_RESULT)
|
|
if(NOT REQUIREMENTS_RESULT EQUAL 0)
|
|
file(REMOVE_RECURSE \${VENV_DIR})
|
|
message(FATAL_ERROR \"Failed to install requirements.txt\")
|
|
endif()
|
|
# Install nuitka (Note that Nuitka 2.6 will require Python 3.11 interpreter)
|
|
message(STATUS \"Installing nuitka\")
|
|
execute_process(COMMAND \${VENV_PYTHON} -m pip install nuitka==2.6 RESULT_VARIABLE NUITKA_INSTALL_RESULT)
|
|
if(NOT NUITKA_INSTALL_RESULT EQUAL 0)
|
|
file(REMOVE_RECURSE \${VENV_DIR})
|
|
message(FATAL_ERROR \"Failed to install nuitka\")
|
|
endif()
|
|
# Install patchelf 0.17.2.4
|
|
message(STATUS \"Installing patchelf\")
|
|
execute_process(COMMAND \${VENV_PYTHON} -m pip install patchelf==0.17.2.4 RESULT_VARIABLE PATCHELF_INSTALL_RESULT)
|
|
if(NOT PATCHELF_INSTALL_RESULT EQUAL 0)
|
|
file(REMOVE_RECURSE \${VENV_DIR})
|
|
message(FATAL_ERROR \"Failed to install patchelf\")
|
|
endif()
|
|
# Build standalone binary
|
|
message(STATUS \"Building standalone binary with nuitka\")
|
|
execute_process(
|
|
COMMAND \${CMAKE_COMMAND} -E env PATH=\${VENV_DIR}/bin:\$ENV{PATH}
|
|
\${VENV_PYTHON} -m nuitka --mode=onefile --no-deployment-flag=self-execution
|
|
--product-name=${STANDALONEBINARY_EXTRACT_DIR}
|
|
--onefile-tempdir-spec=/{PRODUCT}/rocprof_compute_standalonebinary_{PID}
|
|
--include-data-files=${PROJECT_SOURCE_DIR}/VERSION*=./ --enable-plugin=no-qt
|
|
--include-data-files=src/lib/rocprofiler_compute_tool.cpp=lib/rocprofiler_compute_tool.cpp
|
|
--include-data-files=src/lib/helper.cpp=lib/helper.cpp
|
|
--include-data-files=src/lib/helper.hpp=lib/helper.hpp --include-package=dash_svg
|
|
--include-package-data=dash_svg --include-package=dash_bootstrap_components
|
|
--include-package-data=dash_bootstrap_components --include-package=plotly
|
|
--include-package-data=plotly --include-package=rocprof_compute_analyze
|
|
--include-package-data=rocprof_compute_analyze
|
|
--include-package=rocprof_compute_profile
|
|
--include-package-data=rocprof_compute_profile
|
|
--include-package=rocprof_compute_tui --include-package-data=rocprof_compute_tui
|
|
--include-package=rocprof_compute_soc --include-package-data=rocprof_compute_soc
|
|
--include-package=utils --include-package-data=utils --include-package=hip
|
|
--include-package-data=hip --include-package=dateutil
|
|
--include-package-data=dateutil --include-package=rich
|
|
--include-package-data=rich src/${EXECUTABLE_NAME}
|
|
WORKING_DIRECTORY ${PROJECT_SOURCE_DIR}
|
|
RESULT_VARIABLE NUITKA_RESULT
|
|
)
|
|
if(NOT NUITKA_RESULT EQUAL 0)
|
|
file(REMOVE_RECURSE \${VENV_DIR})
|
|
message(FATAL_ERROR \"Nuitka build failed\")
|
|
endif()
|
|
# Remove library rpath from executable
|
|
message(STATUS \"Removing rpath from binary\")
|
|
execute_process(COMMAND \${VENV_DIR}/bin/patchelf --remove-rpath ${EXECUTABLE_NAME}.bin WORKING_DIRECTORY ${PROJECT_SOURCE_DIR} RESULT_VARIABLE PATCHELF_RESULT)
|
|
if(NOT PATCHELF_RESULT EQUAL 0)
|
|
file(REMOVE_RECURSE \${VENV_DIR})
|
|
message(FATAL_ERROR \"patchelf failed to remove rpath\")
|
|
endif()
|
|
# Clean up virtual environment
|
|
message(STATUS \"Cleaning up virtual environment\")
|
|
file(REMOVE_RECURSE \${VENV_DIR})
|
|
# Install to destination
|
|
file(INSTALL ${PROJECT_SOURCE_DIR}/${EXECUTABLE_NAME}.bin DESTINATION \$ENV{DESTDIR}\${CMAKE_INSTALL_PREFIX}/${CMAKE_INSTALL_LIBEXECDIR}/${PROJECT_NAME})
|
|
# Explicitly set executable permission (required for volume-mounted filesystems)
|
|
execute_process(COMMAND chmod +x \$ENV{DESTDIR}\${CMAKE_INSTALL_PREFIX}/${CMAKE_INSTALL_LIBEXECDIR}/${PROJECT_NAME}/${EXECUTABLE_NAME}.bin)
|
|
# Clean up Nuitka build artifacts from source directory
|
|
message(STATUS \"Cleaning up Nuitka build artifacts\")
|
|
file(REMOVE ${PROJECT_SOURCE_DIR}/${EXECUTABLE_NAME}.bin)
|
|
file(REMOVE_RECURSE ${PROJECT_SOURCE_DIR}/${EXECUTABLE_NAME}.build)
|
|
file(REMOVE_RECURSE ${PROJECT_SOURCE_DIR}/${EXECUTABLE_NAME}.dist)
|
|
file(REMOVE_RECURSE ${PROJECT_SOURCE_DIR}/${EXECUTABLE_NAME}.onefile-build)
|
|
message(STATUS \"Standalone binary installed to \${CMAKE_INSTALL_PREFIX}/${CMAKE_INSTALL_LIBEXECDIR}/${PROJECT_NAME}/${EXECUTABLE_NAME}.bin\")
|
|
"
|
|
COMPONENT main
|
|
)
|
|
endif()
|
|
|
|
# ----------
|
|
# Packaging
|
|
# ----------
|
|
|
|
set(PKG_MAINTAINER_NM "rocprofiler compute support")
|
|
set(PKG_MAINTAINER_EMAIL "dl.omnitools@amd.com")
|
|
message(STATUS "Packaging config...")
|
|
set(CPACK_GENERATOR "DEB" "RPM" CACHE STRING "")
|
|
set(CPACK_PACKAGE_NAME "${PROJECT_NAME}" CACHE STRING "")
|
|
set(CPACK_PACKAGE_CONTACT
|
|
"https://github.com/ROCm/rocm-systems/tree/develop/projects/rocprofiler-compute"
|
|
)
|
|
set(CPACK_PACKAGE_DESCRIPTION_SUMMARY
|
|
"ROCm Compute Profiler: tool for GPU performance profiling"
|
|
)
|
|
set(CPACK_RPM_PACKAGE_DESCRIPTION
|
|
"ROCm Compute Profiler is a performance analysis tool for profiling
|
|
machine learning/HPC workloads running on AMD GPUs."
|
|
)
|
|
set(CPACK_PACKAGE_VENDOR "Advanced Micro Devices, Inc.")
|
|
|
|
# Package versioning
|
|
set(CPACK_PACKAGE_VERSION_MAJOR ${PROJECT_VERSION_MAJOR})
|
|
set(CPACK_PACKAGE_VERSION_MINOR ${PROJECT_VERSION_MINOR})
|
|
set(CPACK_PACKAGE_VERSION_PATCH ${PROJECT_VERSION_PATCH})
|
|
set(CPACK_PACKAGE_VERSION
|
|
"${CPACK_PACKAGE_VERSION_MAJOR}.${CPACK_PACKAGE_VERSION_MINOR}.${CPACK_PACKAGE_VERSION_PATCH}"
|
|
)
|
|
|
|
# RPM package specific variables
|
|
set(CPACK_RPM_PACKAGE_LICENSE "MIT")
|
|
set(CPACK_RPM_COMPONENT_INSTALL ON)
|
|
set(CPACK_RPM_PACKAGE_RELEASE_DIST ON)
|
|
set(CPACK_RPM_FILE_NAME "RPM-DEFAULT")
|
|
set(CPACK_RPM_SPEC_MORE_DEFINE "%undefine __brp_mangle_shebangs")
|
|
|
|
if(DEFINED CPACK_PACKAGING_INSTALL_PREFIX)
|
|
set(CPACK_RPM_EXCLUDE_FROM_AUTO_FILELIST_ADDITION "${CPACK_PACKAGING_INSTALL_PREFIX}")
|
|
endif()
|
|
|
|
# Debian package specific variables
|
|
set(CPACK_DEBIAN_PACKAGE_LICENSE "MIT")
|
|
set(CPACK_DEB_COMPONENT_INSTALL ON)
|
|
set(CPACK_DEBIAN_FILE_NAME "DEB-DEFAULT")
|
|
|
|
# Dependencies
|
|
set(PACKAGE_REQUIRES "rocprofiler" CACHE STRING "Package dependencies")
|
|
set(CPACK_RPM_PACKAGE_REQUIRES ${PACKAGE_REQUIRES})
|
|
set(CPACK_DEBIAN_PACKAGE_DEPENDS ${PACKAGE_REQUIRES})
|
|
|
|
# Handle the project rebranding from omniperf to rocprofiler-compute
|
|
set(OMNIPERF_PACKAGE_NAME "omniperf")
|
|
set(CPACK_RPM_PACKAGE_PROVIDES ${OMNIPERF_PACKAGE_NAME})
|
|
set(CPACK_RPM_PACKAGE_OBSOLETES "${OMNIPERF_PACKAGE_NAME} < 3.0.0")
|
|
set(CPACK_RPM_PACKAGE_CONFLICTS ${OMNIPERF_PACKAGE_NAME})
|
|
|
|
set(CPACK_DEBIAN_PACKAGE_PROVIDES ${OMNIPERF_PACKAGE_NAME})
|
|
set(CPACK_DEBIAN_PACKAGE_REPLACES ${OMNIPERF_PACKAGE_NAME})
|
|
set(CPACK_DEBIAN_PACKAGE_BREAKS ${OMNIPERF_PACKAGE_NAME})
|
|
|
|
# Disable automatic dependency generation
|
|
set(CPACK_RPM_PACKAGE_AUTOREQPROV OFF)
|
|
set(CPACK_RPM_PACKAGE_AUTOREQ OFF)
|
|
set(CPACK_RPM_PACKAGE_AUTOPROV OFF)
|
|
|
|
if(INSTALL_TESTS)
|
|
set(CPACK_RPM_TESTS_PACKAGE_REQUIRES ${CPACK_PACKAGE_NAME})
|
|
set(CPACK_DEBIAN_TESTS_PACKAGE_DEPENDS ${CPACK_PACKAGE_NAME})
|
|
endif()
|
|
|
|
# ----- Check for packaging override -----
|
|
if(DEFINED ENV{ROCM_LIBPATCH_VERSION})
|
|
set(CPACK_PACKAGE_VERSION "${CPACK_PACKAGE_VERSION}.$ENV{ROCM_LIBPATCH_VERSION}")
|
|
endif()
|
|
|
|
if(DEFINED ENV{CPACK_RPM_PACKAGE_RELEASE})
|
|
set(CPACK_RPM_PACKAGE_RELEASE $ENV{CPACK_RPM_PACKAGE_RELEASE})
|
|
else()
|
|
set(CPACK_RPM_PACKAGE_RELEASE "local")
|
|
endif()
|
|
|
|
if(DEFINED ENV{CPACK_DEBIAN_PACKAGE_RELEASE})
|
|
set(CPACK_DEBIAN_PACKAGE_RELEASE $ENV{CPACK_DEBIAN_PACKAGE_RELEASE})
|
|
else()
|
|
set(CPACK_DEBIAN_PACKAGE_RELEASE "local")
|
|
endif()
|
|
|
|
# Log package info
|
|
message(STATUS " Package Name: ${CPACK_PACKAGE_NAME}")
|
|
message(STATUS " Package Version: ${CPACK_PACKAGE_VERSION}")
|
|
message(STATUS " RPM Package Release: ${CPACK_RPM_PACKAGE_RELEASE}")
|
|
message(STATUS " Debian Package Release: ${CPACK_DEBIAN_PACKAGE_RELEASE}")
|
|
message(STATUS " Packaging Install Prefix: ${CPACK_PACKAGING_INSTALL_PREFIX}")
|
|
message(STATUS " Install Tests: ${INSTALL_TESTS}")
|
|
message(STATUS " Package Dependencies: ${PACKAGE_REQUIRES}")
|
|
message(STATUS " CPack Generator: ${CPACK_GENERATOR}")
|
|
|
|
# Source tarball
|
|
set(CPACK_SOURCE_GENERATOR "TGZ")
|
|
set(CPACK_SOURCE_PACKAGE_FILE_NAME ${CMAKE_PROJECT_NAME}-${FULL_VERSION_STRING})
|
|
set(CPACK_SOURCE_IGNORE_FILES
|
|
".*~$"
|
|
\.git/
|
|
\.github
|
|
\.gitmodules
|
|
\.gitignore
|
|
/tests
|
|
/build
|
|
)
|
|
|
|
configure_pkg(${PACKAGE_NAME} ${COMP_TYPE} ${CPACK_PACKAGE_VERSION} ${PKG_MAINTAINER_NM} ${PKG_MAINTAINER_EMAIL})
|
|
|
|
include(CPack)
|