Jonathan R. Madsen
f167317524
* Public C++ header files (source/include/rocprofiler-sdk/cxx)
* Update samples/api_buffered_tracing
- scratch memory and page migration
- README
* Update samples/api_buffered_tracing
- page migration component in sample
* Update tests/page-migration/validate.py
- fix checks for page migration operation names
* Update tests/page-migration/validate.py
- fix get_allocated_pages
* Update scratch memory and page migration validations
* Fix include/rocprofiler-sdk/cxx installation
* Rework include/rocprofiler-sdk/cxx
- Improve name_info to support const char*, string_view, string
* Update samples/api_{buffered,callback}_tracing
* External correlation ID request sample
- includes correlation ID retirement demo
* Update samples/api_buffered_tracing/README.md
* Update lib/rocprofiler-sdk/hsa/queue.cpp
- generate correlation ID for kernel launch if one doesn't exist
* Remove priority check from tool libraries (samples/tests)
- if(priority > 0) return nullptr check in rocprofiler_configure has proliferated beyond its intended use
* Apply suggestions from code review
[ROCm/rocprofiler-sdk commit: de13d2ac5d]
529 B
529 B
API Buffer Tracing Sample
Services
- Code object callback tracing for mapping kernel IDs to kernel names
- HSA API (Core, AMD Ext)
- HIP API (Runtime)
- Kernel dispatch
- Memory copy
- Page Migration
- Scratch Memory
Properties
- Buffer size of 4096 bytes which is automatically flushed once >= 87.5% of buffer is filled (3584 bytes)
- Creation of dedicated thread for buffer callback delivery
- Push external correlation IDs once per thread (value is thread ID)
- Receives notifications for internal thread creation