Problem with original test:
- Created circular dependencies between queues:
* Queue1: Kernel A → Barrier(waits for signal_2) → Kernel C
* Queue2: Barrier(waits for signal_1) → Kernel B → sets signal_2
- With strict "one kernel at a time" serialization, this created deadlock:
* Queue1 executed Kernel A, then blocked on barrier waiting for signal_2
* Serializer switched to Queue2, but Queue2 was blocked waiting for signal_1
* Neither queue could proceed: Queue1 needed Queue2's Kernel B to complete,
but Queue2 couldn't start until Queue1 finished completely
- Test would hang indefinitely at hsa_signal_wait_relaxed() for signal_2
Solution implemented:
- Reordered packet submission to eliminate circular dependencies
- Ensured signal producers execute before consumers need them:
* Kernel A produces signal_1 before Queue2's barrier needs it
* Kernel B produces signal_2 before Queue1's continuation needs it
- Dependencies now flow forward without cycles, allowing serializer progress
Refactoring changes:
- Extract common functionality into helper functions:
* create_completion_signal() for signal creation
* create_queue() for queue creation
* submit_kernel_packet() for kernel dispatch packets
* submit_barrier_packet() for barrier packets
- Add comprehensive documentation explaining expected execution pattern
- Simplify main() function making the dependency flow more readable
Co-authored-by: Benjamin Welton <bewelton@amd.com>
* Adding GPU index as a parameter for ATT
* Tidy fix
* Using tokenize
* Update tests/rocprofv3/advanced-thread-trace/CMakeLists.txt
Co-authored-by: Indic, Vladimir <Vladimir.Indic@amd.com>
* Update tests/rocprofv3/advanced-thread-trace/CMakeLists.txt
* Adding error logging. Using idx instead of id.
---------
Co-authored-by: Giovanni <gbaraldi@amd.com>
Co-authored-by: Indic, Vladimir <Vladimir.Indic@amd.com>
* use new azure runners
* update core-deb logic in continuous_integration.yml
* more logic changes in continuous_integration.yml
* changes to runs-on conditional in continuous_integration.yml
* update runs-on conditional again on continuous_integration.yml
* utulize matrix system list, add gpu variable and separate runner
* Remove trailing comma from matrix
* Delete .github/workflows/generate_afar.yml
---------
Co-authored-by: Kandula, Venkateshwar reddy <Venkateshwarreddy.Kandula@amd.com>
Co-authored-by: Elwazir, Ammar <Ammar.Elwazir@amd.com>
* adding summary.py to generate tmp <category_region>_summary views
* migrating CSV summary to SDK method of writing CSVs
- Add domain_view to summary.py
- omit the C++ code of writing CSV because it gets revered later anyway
* Add summary subparser and write_sql_view_to_csv function
* adding all <>_summary views generation to summary.py
* add summary_per_rank feature
* add --summary-per-rank
* reconstruct generate_summary_view and create_domain_view
-introduce by_rank
* remove sqr and variance in summary views
* use RocpdImportData instead of connection
* two fixes on summary.py
--modify the generate_summary_view function to return a tuple with view name and sql code
add if_not_exits parameter to generete_summary_view
* Refactor summary.py to allow output path and filename args, and apply time_window
- clean up summary table column headers
- only generate by-rank views if that param is specified
* Add ProcessID to Hostname output and csv, so users can identify the system in the by-rank summaries
* Summary.py, just add hostname to by-rank summaries, instead of creating mapping table
* Summary - migrate csv writer to pandas, for more future flexibility
* Adding a few simple tests for summary.py
* Linting fixes
* add region_categories to summary options
- Automatically retrieve region categories from the database if argument is None
* add backticks for view_names
* fix tests after rebase
* Made code review changes
- fixed whitespace in CMakelists.txt
- adding query.py module & subparser in __main__.py
- refactor summary function to return query
- used query.py to output csv
- used query.py to also output summary to console
- provided new command line options to select summary output to csv or console
* Made fix to jinja template in query.py, as suggested by copilot
* Consolidated output calls to query in export_view function based on feedback
- refactored: helpers, query functions, create view functions
- extended formats to include what query supports (md, html, pdf, json)
- added json format to query, and changed orient=records
- adding jinja2 and reportlab to requirements.txt
* Add version_info for rocpd and roctx
* Add rocpd commandline tool
* Add executable permissions to source/bin/rocpd.py
* Removed rocpd2query, and cleaned up --help examples
---------
Co-authored-by: acanadas <acanadas@amd.com>
Co-authored-by: Jin Tao <jintao12@amd.com>
Co-authored-by: a-canadasruiz <Araceli.CanadasRuiz@amd.com>
Co-authored-by: Jonathan R. Madsen <Jonathan.Madsen@amd.com>
- Fix context tracing domain bitset overflow
- Previous behavior would enable all flags above ROCPROFILER_BUFFER_TRACING_MARKER_CORE_RANGE_API when this domain was enabled.
* Adding test and samples to decoder
* Fix sample
* Formatting
* Fix multi test
* Disable sample
* Fix tests
* Format
* Version fix
* Locking the decoder
* Add atomic
* Review comments
* Format
* Adding readme
* merge conflict and adding PCS+ATT test
* Review comments
* Properly disable PCS test
* Update tests/rocprofv3/advanced-thread-trace/CMakeLists.txt
* Adding back env var test
* Name fix
* Preload sample
* Addressing review comments
* Update docs
---------
Co-authored-by: Giovanni Baraldi <gbaraldi@amd.com>
* expose dimensional info in rocprofiler_counter_info_v1_t.
* add counter_id in dim info.
* address review comments
* format.
* address comments.
* use array of pointers for dimensions_instaces.
* format and comments.
* address comments.
* new line.
* Update counter_defs.yaml
* Update counter_defs.yaml
* Update counter_defs.yaml
* counter_defs.
* format counter defs.
* format counter defs.
* format counter defs.
* show only counters being profiled in metadata.
* Format.
* use config for counters and fix warnings.
* add version for rocprofiler_counter_dimension_info_v1_t struct.
* rename rocprofiler_counter_record_dimension_instance_v1_info_t.
* account device id from pmc for counters metadata.
* move dim structs to counters.h.
* address comments to compare value.
* fix tests.
* Address comments. use pointer of arrays for ABI.
* rebase.
* fix build error.
* use separate metadata::init() for rocprofv3.
* also print not found counters.
* precompute all the perf counters needed to be in metadata.
* Misc.
* format
* Format.
* rocprofiler::sdk::container::c_array
* Address comments.
* source/lib/output/metadata.cpp
* lint.
* add unit test for c_array.
* add unit test and serialization support for c_array container.
* Misc.
* Clean files.
* Format.
* clang-tidy.
* add more checks to c_array.
* misc. typo
* Addr comments.
---------
Co-authored-by: Venkateshwar Reddy Kandula <vkandula@amd.com>
Co-authored-by: Jonathan R. Madsen <Jonathan.Madsen@amd.com>
* Fix null handle
- use .handle=0, not .handle=numeric_limits<>::max()
* Update lib.common.hasher
* Fix ROCPROFILER_CONTEXT_NONE
* Use context operator==
* Update CHANGELOG
* Updated null handle for scratch memory and changed allocation test so that free ops account for null agent
---------
Co-authored-by: Ian Trowbridge <Ian.Trowbridge@amd.com>
* replace azure runners with internal
* change to mi300a for debug
* revert back to mi300
* move some of the load to mi300a
* use mi300a for clang-tidy
---------
Co-authored-by: Kandula, Venkateshwar reddy <Venkateshwarreddy.Kandula@amd.com>
* Arbitrary host-trap sampling skid (doc)
The host-trap PC sampling might introduce a skid of [0, 2]
instructions. We documented this information and provides
some advice to application developers how to find
hot-spots in the profiles generated by host-trap sampling.
add hip table version to 13, API_ID_LAST is not changed from version=12 since no new struct has been added.
Co-authored-by: Venkateshwar Reddy Kandula <vkandula@amd.com>
Cache packet creation in all cases to reduce the number of allocations/
destruction operations made down to KFD. There is a bug that we
encounter after a period of runtime in KFD where allocations fail to be
visable to the GPU (suspect this is a FW issue, similar to other FW
issues they have had along the same lines). This sidesteps that issue in
rocprof (and likely should be done regardless)
Co-authored-by: Benjamin Welton <bewelton@amd.com>
* addressing issues
* doc fix
* test fix
* fix
* fix formatting issue and doc update
* fix column size
* fix
* fix formatting in output
* tests fix
* test fix
* add new line
* add new line
* fix new line
* fixing typo in using-rocprofv3-avail.rst
* Add perfetto support for scratch memory
* Updated tests and docs.
* Update docs data
* Added underflow check
* Record all free events to 0 bytes
* Add format
* Address review comment
* updated tests for scratch memory
* update scratch-memory tests.
* [SWDEV-516561][1/2] Add MARKER_RANGE_EXTENT to capture ROCTX ranges
Range extent to capture all work between roctxpush/pop operations. Entry callback takes place during roxtxpush and exit callback takes place in roctxpop. This is primarily to allow us to keep an ancestor id on the ancestor stack such that all operations that take place within the push/pop context can be annotated as being apart of this range. With the current setup (where push and pop are two separate operations that need to be combined externally), we cannot keep an ancestor id on the stack and thus cannot tie tracing events to particular ranges.
Correlation id information is inherited from the push operation. Ancestor id needs to be added in a future commit that also outputs this ancestor to CSV.
Output:
```
[ctest] {'size': 64, 'kind': 7, 'operation': 1, 'correlation_id': {'internal': 1525, 'external': 0, 'ancestor': 1524}, 'start_timestamp': 2932551479402642, 'end_timestamp': 2932551491178449, 'thread_id': 3254861}
[ctest] {'size': 64, 'kind': 8, 'operation': 2, 'correlation_id': {'internal': 1525, 'external': 0, 'ancestor': 1524}, 'start_timestamp': 2932551479405878, 'end_timestamp': 2932551491181214, 'thread_id': 3254861}
```
Note: Kind 8 = range extent op.
* Merge fix
Revert several changes
source/lib/rocprofiler-sdk/marker/range_marker.*
- separate out range marker implementation for standard marker implementation
Update public API with marker core range
Support marker core range in sdk (source/lib/rocprofiler-sdk)
Transition rocprofiler-sdk-tool and output lib to use marker core range
Misc fixes for tests
Fix logic in lib/output/generate{CSV,Stats}.cpp
Update tests/rocprofv3/tracing-hip-in-libraries (marker validation)
Fix test_otf2_data
* Test fixes
---------
Co-authored-by: Benjamin Welton <bewelton@amd.com>
* Added null check for stream_stack before get_stream_id is called
* Rename function and add check for stream strack before pop
* Removed empty check for stream stack and adding error log for get_stream_id in stream.cpp
Welton, Benjamin
Kandula, Venkateshwar reddy
Bhardwaj, Gopesh
Baraldi, Giovanni
U, Srihari
Radosavljevic, Milan
Rawat, Swati
Trowbridge, Ian
Vaddireddy, Sushma
Bonnell, Jason
Indic, Vladimir
Hui, Young
Madsen, Jonathan
Kuricheti, Mythreya
Gill, Harkirat
Nagaraj, Sriraksha
Wang, Huanran
Tumakaev, Aleksei
Elwazir, Ammar
Roy, Rakesh