Madsen, Jonathan
00c46fd5e5
* Ability to select alternative compiler per file
Implementation of ompt interface to rocprofiler SDK. task_create and task_schedule are not supported.
Misc updates
Update OpenMP target sample
- samples/ompt -> samples/openmp_target
- fix sample test of openmp-target
- reorganize files
Rework OpenMP implementation
Minor OpenMP implementation cleanup
Rename samples/openmp_target CMake targets
Add tests/bin/openmp
- OpenMP target test app in tests/bin/openmp/target
Format samples/openmp_target CMakeLists.txt
Misc lib/rocprofiler-sdk/openmp cleanup
- fix includes
- convert_arg
Update openmp.def.cpp
- tweak includes
- remove lots of temporary variables
Update samples
- common::get_callback_id_names() -> common::get_callback_tracing_names()
- add kernel dispatch, memory copy, scratch memory buffered tracing to openmp target sample
Fix code object operation names
- add "CODE_OBJECT_" prefix
Update include/rocprofiler-sdk/openmp/api_id.h
- remove spurious comment
Miscellaneous openmp updates
- similar API for openmp_begin and openmp_end
- move implementations of ompt callbacks to openmp.cpp
- ompt_{thread_begin,thread_end,parallel_begin,parallel_end}_callbacks are openmp_events
[SWDEV-484495] Fix int truncation in CSV output (#1098)
CSV output truncates doubles to ints when it shouldn't. Derived metrics
are (mostly) doubles and lose precision (or become worthless) if treated
as an int. Converted these to double to match the format we return from
rocprof-sdk.
Co-authored-by: Benjamin Welton <ben@amd.com>
Update limit for max counter records in rocprof-tool (#1073)
A fixed sized std::array is used to store counter records in rocprofiler SDK. This limit was breached in SWDEV-484742. Upping the limit to 512 to be less likely to reach this limit again.
adding proxy ompt_data_t * arguments
fixes for proxy pointers
- Implement proxy ompt_data_t* pointers for clients
- Add ompt_data_t* arguments back to callback API
- Modify openmp sample to illustrate use of proxy pointers
formatting
SWDEV-467350: Skipping tool counter iteration for unsupported hardware (#1083)
Fixing some accumulate metrics (#1089)
* Fixing some accumulate metrics
* Fixing some more accumulate metrics
---------
Co-authored-by: Benjamin Welton <bewelton@amd.com>
updating rocprofv3 help options (#1113)
* updating rocprofv3 help options
* updating CHANGELOG
Fixing installed pacakge tests in CI (#1119)
* Fixing installed pacakge tests in CI
* Formatted rocprofv3.py with black formatter
SWDEV-488948: PC Sampling - Correlation class to provide some thread safety. Adding multithread tests. (#1112)
* SWDEV-488948: PC Sampling - Correlation class to provide some thread safety. Adding multithread tests.
* Update source/lib/rocprofiler-sdk/pc_sampling/parser/correlation.hpp
Co-authored-by: Vladimir Indic <139573562+vlaindic@users.noreply.github.com>
* Update source/lib/rocprofiler-sdk/pc_sampling/parser/correlation.hpp
Co-authored-by: Vladimir Indic <139573562+vlaindic@users.noreply.github.com>
* Adding backlog for codeobj changes
* Formatting
* Update source/lib/rocprofiler-sdk/pc_sampling/code_object.hpp
Co-authored-by: Vladimir Indic <139573562+vlaindic@users.noreply.github.com>
* Update source/lib/rocprofiler-sdk/pc_sampling/code_object.hpp
Co-authored-by: Vladimir Indic <139573562+vlaindic@users.noreply.github.com>
---------
Co-authored-by: Vladimir Indic <139573562+vlaindic@users.noreply.github.com>
SWDEV-487621: Fixes for metric definitions (#1118)
* Fixes for metric definitions
* Removing gfx8
* Update changelog
* Fixing unit tests
* Small fixes
* Fix for write size
Fix PSDB change (#1120)
Reverts change to `source/include/rocprofiler-sdk/callback_tracing.h`
from commit 9b2ece76c3
clang-18 build fix for RCCL (#1123)
Removes ambiguity on const usage, which clang-18 complains about
(preventing build with warn error).
mem copy direction field update (#1124)
Adding Node-id for debugging with log level trace (#1090)
fix botched rebase
Per Jonathan to remove -rdynamic warning so CI will continue
pedantic formatting
Correct the package name of rocprofiler-sdk (#1126)
* Correct the package name of rocprofiler-sdk
ROCM VERSION(for ex: 60300) was missing in the package name.
Added the same
* Use cmake cache string while setting the variable for ROCm Version
* correct the cmake-format
---------
Co-authored-by: Ranjith Ramakrishnan <Ranjith.Ramakrishnan@amd.com>
Fixing kokkosp tool library packaging (#1121)
* Fixing kokkosp tool library packaging
* Update source/lib/rocprofiler-sdk-tool/kokkosp/CMakeLists.txt
Co-authored-by: github-actions[bot] <41898282+github-actions[bot]@users.noreply.github.com>
* Update CMakeLists.txt
* Update CMakeLists.txt
* Component Requirement in CPack
* Adding package dependency
* Update CMakeLists.txt
* Update rocprofiler_config_packaging.cmake
* Fix rocprofiler-sdk-tool-kokkosp BUILD/INSTALL RPATH
- CMAKE_INSTALL_LIBDIR doesn't help
* Add BUILD/INSTALL RPATH to rocprofv3-trigger-list-metrics
- fixes packaging issues
* Update packaging
- core depends on rocprofiler-sdk-roctx
- add CPACK_DEBIAN_PACKAGE_SHLIBDEPS_PRIVATE_DIRS to resolve inter-package dependencies
* Fix package depends version format
* Improve tests/rocprofv3/summary/validate logging
* Update CI workflow
- prioritize roctx package in Install Packages step
* Remove setting <package-name>_VERSION in config.cmake.in
- this is automatically handled by existence of <package-name>-config-version.cmake
* Update rocprofiler-sdk-config.cmake
- relax find_package versioning requirements to same major and minor version
* Update rocprofiler-sdk-config.cmake
- relax find_package versioning requirements (remove EXACT, specify range)
* Tweak CI workflow
* Update perfetto_reader.py
- better handle failure to load trace processor
* Misc cleanup for config packaging
* Update config packaging
* Update config packaging
* Revert perfetto for core-rpm packages
* Revert perfetto for core-rpm packages
- perfetto < 0.9.0
* Tweak tests/rocprofv3/summary/validate.py
- reorder some checks
---------
Co-authored-by: Ammar Elwazir <aelwazir@useocpm2m-387-013.amd.com>
Co-authored-by: github-actions[bot] <41898282+github-actions[bot]@users.noreply.github.com>
Co-authored-by: Jonathan R. Madsen <jonathanrmadsen@gmail.com>
Clang Warning Fixes (#1131)
Builds prevented on clang-18
Adding start and end timestamp columns in csv (#1128)
* Adding start and end timestamp columns in csv
* Adding assert check for the counter timestamps
---------
Co-authored-by: Gopesh Bhardwaj <gopesh.bhardwaj@amd.com>
rocprofv3: docs and help menu updates (#1129)
* doc updates
* Correcting ROCtx information
* Making ROCTx string consistent
* missing occurence
Renamed agent profiling service to device counting service (#1132)
* Renamed agent profiling service to device counting service
Name more aptly represents what agent profiling did (device wide
counter collection). Conversion of existing user code can be
performed by the following find/sed command:
find . -type f -exec sed -i 's/rocprofiler_agent_profile_callback_t/rocprofiler_device_counting_service_callback_t/g; s/rocprofiler_configure_agent_profile_counting_service/rocprofiler_configure_device_counting_service/g; s/agent_profile.h/device_counting_service.h/g; s/rocprofiler_sample_agent_profile_counting_service/rocprofiler_sample_device_counting_service/g' {} +
* Converted dispatch profile to dispatch counting service
* Debug for functioal counters test
* Minor changes for CI
* Minor fix
* More fixes for CI
* Update evaluate_ast.cpp
---------
Co-authored-by: Benjamin Welton <ben@amd.com>
Testing updated RPM dockers (#1136)
* Testing updated RPM dockers
* Trying to fix PSDB for test package dependency
Agent Profiling Fixes for Broken/Improper API Usage (#1122)
Prevent's multiple setups of agent profiling on the same agent.
Fixes agent read context to only read agents that were setup.
Prevent copy of agent profiling internal data struct and reset
hsa_signal on move to prevent inadvertant delete.
Simplifying PR template (#1139)
Implementation of ompt interface to rocprofiler SDK. task_create and task_schedule are not supported.
Fixing installed pacakge tests in CI (#1119)
* Fixing installed pacakge tests in CI
* Formatted rocprofv3.py with black formatter
Fix PSDB change (#1120)
Reverts change to `source/include/rocprofiler-sdk/callback_tracing.h`
from commit 9b2ece76c3
delete unused files
added arguments to some OMPT buffter records
* Fix cmake issues
Remove rocprofiler_ompt_finalize_tool
- a public API function is not necessary: should just finalize rocprofiler-sdk
Fix duplicate ROCPROFILER_{BUFFER,CALLBACK}_TRACING_KIND_STRING
Add lib/rocprofiler-sdk/ompt.hpp
- declares rocprofiler::sdk::finalize_ompt
Remove change to tests/rocprofv3/summary/conftest.py
Add set_fini_status(1) back to registration.cpp
Deleted uneeded files
Incoporate OpenMP code and sample
Fix merge issues with amd-staging
Add push_correlation_id for OpenMP tasking; improve debugability
fixup bad merge
* Suppress OpenMP data race
* Fix openmp_target sample
* Enum and struct name changes + source code reorg
- remove mix of ompt and openmp
- opted for ompt
- changes made for consistency
- ompt_api -> ompt
- openmp_api -> ompt
- OPENMP -> OMPT
* Update tests and more renaming
- dest_device_num -> dst_device_num
- src_addr -> src_address
- dest_addr -> dst_address
- remove info_type::begin
- require OMP_TARGET_OFFLOAD
* Update openmp-target test/sample env and labels
* Formatting
* Tweaks to cmake for openmp target
- Disable for thread sanitizers due to preloading issue
* OpenMP target cmake updates
- remove gfx1010 (fails on mi300)
- OPENMP_GPU_TARGETS
* Remove device_unload and target_map_emi support
- these are never supported by AMD OpenMP compilers
* Update CI workflow
- exclude openmp-target tests from navi3 and vega20
---------
Co-authored-by: Larry Meadows <Lawrence.Meadows@amd.com>
Co-authored-by: Jonathan R. Madsen <jonathanrmadsen@gmail.com>
ROCprofiler-SDK: Application Profiling, Tracing, and Performance Analysis
Note
rocprofiler-sdk is currently considered a beta version and is subject to change in future releases
Overview
ROCProfiler-SDK is AMD’s new and improved tooling infrastructure, providing a hardware-specific low-level performance analysis interface for profiling and tracing GPU compute applications. To see what's changed Click Here
Note
The published documentation is available at ROCprofiler-SDK documentation in an organized, easy-to-read format, with search and a table of contents. The documentation source files reside in the
rocprofiler-sdk/source/docsfolder of this repository. As with all ROCm projects, the documentation is open source. For more information on contributing to the documentation, see Contribute to ROCm documentation.
GPU Metrics
- GPU hardware counters
- HIP API tracing
- HIP kernel tracing
- HSA API tracing
- HSA operation tracing
- Marker(ROCTx) tracing
- PC Sampling (Beta)
Tool Support
rocprofv3 is the command line tool built using the rocprofiler-sdk library and shipped with the ROCm stack. To see details on the command line options of rocprofv3, please see rocprofv3 user guide Click Here
Documentation
We make use of doxygen to generate API documentation automatically. The generated document can be found in the following path:
<ROCM_PATH>/share/html/rocprofiler-sdk
ROCM_PATH by default is /opt/rocm It can be set by the user in different locations if needed.
Build and Installation
git clone https://git@github.com:ROCm/rocprofiler-sdk.git rocprofiler-sdk-source
cmake \
-B rocprofiler-sdk-build \
-D ROCPROFILER_BUILD_TESTS=ON \
-D ROCPROFILER_BUILD_SAMPLES=ON \
-D CMAKE_INSTALL_PREFIX=/opt/rocm \
rocprofiler-sdk-source
cmake --build rocprofiler-sdk-build --target all --parallel 8
To install ROCprofiler, run:
cmake --build rocprofiler-sdk-build --target install
Please see the detailed section on build and installation here: Click Here
Support
Please report in the Github Issues.
Limitations
-
Individual XCC mode is not supported.
-
By default, PC sampling API is disabled. To use PC sampling. Setting the
ROCPROFILER_PC_SAMPLING_BETA_ENABLEDenvironment variable grants access to the PC Sampling experimental beta feature. This feature is still under development and may not be completely stable.-
Risk Acknowledgment: By activating this environment variable, you acknowledge and accept the following potential risks:
- Hardware Freeze: This beta feature could cause your hardware to freeze unexpectedly.
- Need for Cold Restart: In the event of a hardware freeze, you may need to perform a cold restart (turning the hardware off and on) to restore normal operations. Please use this beta feature cautiously. It may affect your system's stability and performance. Proceed at your own risk.
-
At this point, We do not recommend stress-testing the beta implementation.
-
Correlation IDs provided by the PC sampling service are verified only for HIP API calls.
-
Timestamps in PC sampling records might not be 100% accurate.
-
Using PC sampling on multi-threaded applications might fail with
HSA_STATUS_ERROR_EXCEPTION.Furthermore, if three or more threads launch operations to the same agent, and if PC sampling is enabled, theHSA_STATUS_ERROR_EXCEPTIONmight appear.
-
-
Navi3x requires a stable power state for counter collection. Currently, this state needs to be set by the user. To do so, set "power_dpm_force_performance_level" to be writeable for non-root users, then set performance level to profile_standard:
sudo chmod 777 /sys/class/drm/card0/device/power_dpm_force_performance_level echo profile_standard >> /sys/class/drm/card0/device/power_dpm_force_performance_levelRecommended: "profile_standard" for counter collection and "auto" for all other profiling. Use rocm-smi to verify the current power state. For multiGPU systems (includes integrated graphics), replace "card0" by the desired card.
Warning
The latest mainline version of AQLprofile can be found at https://repo.radeon.com/rocm/misc/aqlprofile/. However, it's important to note that updates to the public AQLProfile may not occur as frequently as updates to the rocprofiler-sdk. This discrepancy could lead to a potential mismatch between the AQLprofile binary and the rocprofiler-sdk source.