make sure that all communicators that have been created during the
runtime of the application are correctly freed again, to avoid memory
leaks.
[ROCm/rocshmem commit: bce329abd6]
When compiling with -O0, some compilers generate a xchg instruction for
the __atomic_store(...) built-in. Using xchg on MMIO memory is
undefined-behavior and may be ignored on certain CPUs.
When compiling with -O0, some compilers generate a xchg instruction for
the __atomic_store(...) built-in. Using xchg on MMIO memory is
undefined-behavior and may be ignored on certain CPUs.
[ROCm/ROCR-Runtime commit: f011a9506d]
Fix profiler_v2 compatibility layer
* Removing trafficBytes in profiler_v3 breaks casting to ncclProfilerEventDescr_v2_t
in the compatibility layer for profiler_v2 interface. This patch fixes the issue
by making the conversion between the two descriptors explicit.
[ROCm/rccl commit: 8171af656b]
Fix profiler_v2 compatibility layer
* Removing trafficBytes in profiler_v3 breaks casting to ncclProfilerEventDescr_v2_t
in the compatibility layer for profiler_v2 interface. This patch fixes the issue
by making the conversion between the two descriptors explicit.
feat: Report PLDM Bundle from SMC to IB
Code changes related to the following:
* APIs
* CLI
* Unit tests
Signed-off-by: Oliveira, Daniel <daniel.oliveira@amd.com>
Change-Id: I35ccf01eb612ca80e3ae6b72039085c18c989222
feat: Report PLDM Bundle from SMC to IB
Code changes related to the following:
* APIs
* CLI
* Unit tests
Signed-off-by: Oliveira, Daniel <daniel.oliveira@amd.com>
Change-Id: I35ccf01eb612ca80e3ae6b72039085c18c989222
[ROCm/amdsmi commit: fe9b6eeb49]
- ROCr now reports preferred engine for copy status. We can leverage
this for max bandwidth for inter-GPU copies
- Cleanup logging
[ROCm/clr commit: 1b0ea080e4]
* SWDEV-532676 - Pass hip-path instead of rocm-path
Commit 7b7537c1703db0cc53ee1f36089406b555ae9a9e in llvm-project enforces that no hip-device-lib-path is forwarded by hipcc. To ensure that runtime is found, pass hip-path instead.
* SWDEV-532676 - Fix comment
Co-authored-by: Lambert, Jacob <Jacob.Lambert@amd.com>
---------
Co-authored-by: Lambert, Jacob <Jacob.Lambert@amd.com>
[ROCm/hip-tests commit: 65158bc9f1]
* SWDEV-532676 - Pass hip-path instead of rocm-path
Commit 7b7537c1703db0cc53ee1f36089406b555ae9a9e in llvm-project enforces that no hip-device-lib-path is forwarded by hipcc. To ensure that runtime is found, pass hip-path instead.
* SWDEV-532676 - Fix comment
Co-authored-by: Lambert, Jacob <Jacob.Lambert@amd.com>
---------
Co-authored-by: Lambert, Jacob <Jacob.Lambert@amd.com>
* Updated kernel rename service to use internal correaltion IDs for external correlation IDs and kernel rename values
* Updated for review comments
* Changed if-condition in generateJSON.cpp to check if string view is empty before calling get_entry
[ROCm/rocprofiler-sdk commit: afb27f3f1a]
* Updated kernel rename service to use internal correaltion IDs for external correlation IDs and kernel rename values
* Updated for review comments
* Changed if-condition in generateJSON.cpp to check if string view is empty before calling get_entry
* [rocprofv3] Add rocpd output support (part 1: prelude)
- git submodules for sqlite3, GOTCHA, and pybind11
- HIP stream data
- rocprofiler_query_intercept_table_name(...)
- serialization load
- rocprofiler::sdk::get_perfetto_category(KindT)
- rocprofiler::sdk::parse::strip
- common library updates
- md5sum
- hasher
- simple_timer
- static_tl_object
- get_process_start_time_ns(pid_t)
- output library updates
- node_info
- file_generator (generator is now virtual base class)
- stream info updates
* Added submodules
* Code review updates
* Minor unused-but-set-X warning fixes
* Update CI
- install libsqlite3-dev package
* Update CI
- install libsqlite3-dev package
* Fix static thread-local object memory leak
- also fix signal handler chaining
* Remove URL from comment
* Remove page migration exception
* Enable ROCPROFILER_BUILD_SQLITE3 by default
- try find_package(SQLite3) first and then build when ROCPROFILER_BUILD_SQLITE3=ON
* Fix gotcha installation
- make install of target optional
* Validate tracing + counter collection dispatch data
- i.e. correlation ids, thread ids, timestamps
* Make find_package(SQLite3) optional
- ROCm CI does not have SQLite3 dev package installed and cannot build from source (missing tclsh)
* Fixes to tracing + counter collection test
* get_process_start_time_ns update
- original implementation did not work
* Fix pytest-packages test_perfetto_data for counter collection
- erroneous failure when used with same PMC + multiple agents
* cmake policy: option() honors normal variables
- for GOTCHA submodule
* Improve samples/api_buffered_tracing stability
- reduce likelihood of sporadic exception throw
* Update gotcha submodule
---------
Co-authored-by: Jonathan R. Madsen <jonathanrmadsen@gmail.com>
[ROCm/rocprofiler-sdk commit: 7166b1ab58]
* [rocprofv3] Add rocpd output support (part 1: prelude)
- git submodules for sqlite3, GOTCHA, and pybind11
- HIP stream data
- rocprofiler_query_intercept_table_name(...)
- serialization load
- rocprofiler::sdk::get_perfetto_category(KindT)
- rocprofiler::sdk::parse::strip
- common library updates
- md5sum
- hasher
- simple_timer
- static_tl_object
- get_process_start_time_ns(pid_t)
- output library updates
- node_info
- file_generator (generator is now virtual base class)
- stream info updates
* Added submodules
* Code review updates
* Minor unused-but-set-X warning fixes
* Update CI
- install libsqlite3-dev package
* Update CI
- install libsqlite3-dev package
* Fix static thread-local object memory leak
- also fix signal handler chaining
* Remove URL from comment
* Remove page migration exception
* Enable ROCPROFILER_BUILD_SQLITE3 by default
- try find_package(SQLite3) first and then build when ROCPROFILER_BUILD_SQLITE3=ON
* Fix gotcha installation
- make install of target optional
* Validate tracing + counter collection dispatch data
- i.e. correlation ids, thread ids, timestamps
* Make find_package(SQLite3) optional
- ROCm CI does not have SQLite3 dev package installed and cannot build from source (missing tclsh)
* Fixes to tracing + counter collection test
* get_process_start_time_ns update
- original implementation did not work
* Fix pytest-packages test_perfetto_data for counter collection
- erroneous failure when used with same PMC + multiple agents
* cmake policy: option() honors normal variables
- for GOTCHA submodule
* Improve samples/api_buffered_tracing stability
- reduce likelihood of sporadic exception throw
* Update gotcha submodule
---------
Co-authored-by: Jonathan R. Madsen <jonathanrmadsen@gmail.com>
Jayaprakash, Karthik
Assiouras, Ioannis
Yiannis Papadopoulos
Andryeyev, German
jamessiddeley-amd
Edgar Gabriel
David Yat Sin
Giuseppe Congiu
Maisam Arif
Daniel Oliveira
AL Musaffar, Yazen
Mewar, Deepak
Madsen, Jonathan
Kudchadker, Saleel
Deshpande, Chinmay Diwakar
isaki001
Arm Patinyasakdikul
Dittakavi, Satyanvesh
Trowbridge, Ian