* attach: rename librocprofv3-attach
- Renames library to librocprofiler-sdk-rocattach
- ROCAttach library will be formalized and documented in future commit
* Address review comments
- Rename rocprofv3-attach.py to rocprof-attach.py
- Use common filesystem.hpp in rocattach
* Fix component name typo
* Doc fixup
---------
Co-authored-by: JeniferC99 <150404595+JeniferC99@users.noreply.github.com>
* adding ROCpd database merge
* adding ROCpd database merge concatenating all tables
* update merge script
- copy all tables from files
* fix merge format
* Add package submodule, initial POC. Need to refine
* Minor fixes and clean up duplicated code in package.py
* Revamp metadata layout, add wildcard and .rpdb parsing
* Add auto merge & package when > 5 DBs, add examples, don't use auto_merge when using sub-commands merge & package
* - Extend package/yaml inputs to all rocpd modules
- Improve handling more corner cases for bad input files when parsing input parameters (bad yaml files, bad .rpdb folder, folders as input)
- Changed to use UUID in merged filename instead of the time, in auto-merge algorithm
* Minor text fixes for consistancy between modules
* Add more wildcard support and add package, merge tests
* Make changes based on review suggestions
* Move parsing packages into importer.py, simplified adding required params to a function
* fix package test by flattening input list before processing
* Integrate merge.py changes from Jonathan to add name-collision checks, recreating indexes, foreign key check (disabled for now, due to processing time)
* Rework rocpd.<submodule>.{add_args,process_args}
- add_args function returns a functor which accepts input and args
- time_window functor returned from add_args automatically applies time windowing of input
* change merge&package limit to 1, merge should create data views
* Move files by default instead of making copies
- copying can be enabled by passing "copy=True" or --copy cmdline argument
* refactor package to make the logic cleaner, set merge limit back to 5
* Allow automerge-limit param to override limit, change default back to 1. Tests updated to use query, much quicker
* Update --help instructions for package
---------
Co-authored-by: acanadas <acanadas@amd.com>
Co-authored-by: a-canadasruiz <Araceli.CanadasRuiz@amd.com>
Co-authored-by: Young Hui <young.hui@amd.com>
Co-authored-by: Jonathan R. Madsen <jonathanrmadsen@gmail.com>
* attach: rocprofv3-attach py improvements
- Handle error status during detachment
- Add detection and error for changing rocprofv3 configuration on reattachment
- Add and improve console messages during attachment and detachment
- Documentation update pass
* Fix for SWDEV-552584
Two calls to ompt_callback_task_scheduled were issued for the same
prior task. One of them was ompt_task_complete, which causes
internal storage to be release and a pointer zeroed. The other
was ompt_task_early_fulfill, which attempted to reference the
pointer. The callbacks could come in any order as they were
from different threads, thus causing a null pointer
dereference on occasion. The code was changed to do nothing
for the early_fulfill. Additional null pointer checks were
added.
* formatting
* Update ompt.cpp
Co-authored-by: Copilot <175728472+Copilot@users.noreply.github.com>
---------
Co-authored-by: Jonathan R. Madsen <jrmadsen@users.noreply.github.com>
Co-authored-by: Copilot <175728472+Copilot@users.noreply.github.com>
* Fix dimension mismatch for multi-GPU systems with identical architectures
This change addresses an issue where counter dimensions were incorrectly
shared across all GPU agents with the same architecture name, even when
those agents had different hardware configurations (e.g., different CU counts).
Changes:
- Updated getBlockDimensions() to accept agent ID instead of architecture name
- Made dimension cache agent-specific instead of architecture-specific
- Updated set_dimensions() in AST evaluation to use specific agent ID
- Modified all API functions to handle agent-specific dimension lookups
- Updated tests to work with agent-specific dimensions
This fix ensures that dimensions accurately reflect the actual hardware
configuration of each individual GPU agent, preventing dimension mismatches
in multi-GPU systems where GPUs share the same architecture but have
different physical configurations.
Counter ID Representation Changes:
- Modified counter_id encoding to include agent information in bits 37-32
- Agent logical_node_id is encoded as (value + 1) to ensure agent 0 is detectable
- Counter records internally store only 16-bit base metric IDs (bits 15-0)
- Tool reconstructs agent-encoded counter IDs from base metric ID & agent info
- Instance record counter_id field uses bitwise AND mask to extract base metric ID
(counter_id.handle & 0xFFFF) to fit in 16-bit storage
- Output generators (CSV, JSON, Perfetto) use agent-encoded IDs for consistency
- Updated counter_config.cpp and metrics.cpp to extract base metric ID when needed
- All counter lookups now properly handle agent-encoded vs base metric IDs
This ensures counter IDs are consistent between metadata and output records while
maintaining compact storage in instance records.
- Remove unimplemented older API functions
- Remove mentions of reattach API
- Remove details on implementing a process attachment library
- This will return later as a theory of operation
* Changed stream error warning, remove regex search from attach execute test
* Formatting
* Revert accidental change
* Fix stream hang error due to grabbing same lock twice
* Updated add stream code, need to update tests
* Update attachment tests to use streams, threads, and multiple devices
* Update tests and fix stream issues
* Updated error messages to be more explicit, updated json to csv code in conftest to include streams and threads
* Formatting
* Add attachment label to attachment tests and update validation to fix errors
* Fix attach twice conftest
* Disabled thread san tests for attachment since they no longer work with bin file changes
* Updated for comment
* Added null check for getting attach status
* SWDEV-545950 - Add hipStreamCopyAttributes API Implementation
* Add unit test for hipStreamCopyAttributes API
* Add ChangeLog and nvidia mapping for the API
* Update rocprofiler-sdk with new HIP API details
* [rocprofiler-sdk] handle hipStreamCopyAttributes in stream tracing service
- this new HIP function has multiple stream arguments and needs to be skipped because it does not have an explicit create/destroy/set functionality
* Update HIP_RUNTIME_API_TABLE_STEP_VERSION in clr and rocprofiler-sdk
* Resolve merge conflicts
---------
Co-authored-by: Jonathan R. Madsen <jonathanrmadsen@gmail.com>
* Rollup of build changes needed for compat with TheRock.
* When built for a non-default ROCM location, the HIP headers can't be found by a few targets.
* Uses pkg_check for DRM libraries like ROCR-Runtime does (which avoids accidental fallback to system versions).
* Robust fix for nolink targets
* nolink targets essentially exist for include directories
* all nolink targets are automatically added to rocprofiler-sdk-headers with a $<BUILD_INTERFACE:...> generator expression
* Re-add previously used mechanism to find drm libs
---------
Co-authored-by: Marius Brehler <marius.brehler@amd.com>
Co-authored-by: Stella Laurenzo <stellaraccident@gmail.com>
Co-authored-by: Jonathan R. Madsen <jonathanrmadsen@gmail.com>
When users provide an incorrectly formatted YAML file to the -E/--extra-counters
option in rocprofv3, they now receive a clear error message showing:
- What went wrong (invalid YAML format)
- The correct rocprofiler-sdk YAML structure with example
- The actual content that failed to parse
This addresses confusion where users might use the legacy ROCProfiler YAML
format instead of the new rocprofiler-sdk schema format.
* Initial consecutive kernel WIP
* Updated logic after discussion, create context only when needed, change set of captured ids to dispatch_id_t type
* Updated to fix concurrency issues and revert kernel_iterations
* Add captured id in first lock capture
* Updated code to use wlock, added comments, removed some unecessary atomic
* Cleaned up, need to add test
* Add test to check that generated stats csv file is not empty
* Updated test to check if vector-ops kernels are being used
* Fix phase bug
* Updated for comments
* Flattened ATT logic a bit
* Fix incorrect if-statement
* Fix merge conflict
Swati Rawat
Larry Meadows
Marko Crnobrnja Maletić
Mark Meserve
Ammar ELWazir
itrowbri
usrihari123
Young Hui - AMD
Jonathan R. Madsen
Gopesh Bhardwaj
Giovanni Lenzi Baraldi
systems-assistant[bot]
Aleksei Tumakaev
Satyanvesh Dittakavi
Todd tiantuo Li
sluzynsk-amd
Benjamin Welton
Vladimir Indic
Rahul Manocha
Venkateshwar Reddy Kandula
Marius Brehler