An ASAN run of the release build revealed some elements of
the supported_isas static map were still using stack data. This
change makes it use heap data so it will persist.
Change-Id: Ie51887e88b9e2dec27acfc97ea45a6219fea971c
* When hipMemset3dAsync is captured, a 3d extent can set be as a parameter (depth > 1). That worked on nvidia, but on amd wrong portion of array was filled because when creating Memset3D command, extent dimensions were used to create pitchedPtr, instead of original array width and height.
* Also, when capturing hipMemset3dAsync, nvidia allows any of the extent dimension to be 0, and in that case, no work should be done.
Change-Id: I46a605bf9ae801cd3348e98d528c21263a8eefce
1. Fix LDSSize type to be uint32_t.
2. Prevent clWaitForEvents running on complete events whose
HostQueue have been destructed.
Change-Id: I829e915f56b37db2ba76bb876c9656166534f154
- Create bins each with its own map and lock. This would help cases
where the hash of a VA is differnet than ther one which falls in
different bin, and there is no lock contention
- Use STL shared mutexes, that way we can unique_lock for map updates
vs simple reads which can use shared_lock
Change-Id: I118818be65c6373700f5e511045babb6a398938a
SDMA queue resources are limited when all SDMA copies are bottle necked
into 2 engines. Callers will not be able to make the best decisions
to allocate queue resources fairly so have ROCr fallback to old round
robin behaviour dictated by KFD.
Change-Id: I93d52297976d74e20129c5eb1dcfbfa5aa5067a7
Migrates profiler_serializer class in QueueController to have an instance per-agent instead of one globally. Other changes in this commit are to allow for maps of the queues associated with each agent to be passed to profiler_serializer when it is turned on/off. Existing test cases cover whether or not the kernels are serialized (multistream app). New test case added to show that this serialization only occurs on a per device level with a kernel launched on one device waiting for a value to be set on the other.
[ROCm/rocprofiler-sdk commit: 4a5b1d98c2]
Migrates profiler_serializer class in QueueController to have an instance per-agent instead of one globally. Other changes in this commit are to allow for maps of the queues associated with each agent to be passed to profiler_serializer when it is turned on/off. Existing test cases cover whether or not the kernels are serialized (multistream app). New test case added to show that this serialization only occurs on a per device level with a kernel launched on one device waiting for a value to be set on the other.
Add an atomic counter to track the outstanding HSA handlers.
Wait on CPU for the callbacks if the number exceeds the value
in DEBUG_HIP_BLOCK_SYNC env variable.
Change-Id: I95dc8c4bf0258c7e59411b7504220709ed6898c5
* Update tests/bin/transpose/transpose.cpp
- add hipMemGetInfo call to display the available vs. total memory on the GPU
* Update tests/rocprofv3/summary/validate.py
- Updated test_summary_display_data after addition of hipMemGetInfo to transpose test exe
* Tweak code coverage comment uploading
- create unique orphan branch per PR
- reduce quality of PNG files (85 -> 70)
* Revert some of code coverage comment uploading
- remove creation of unique orphan branch per PR
* Tweak code coverage comment uploading
- create unique orphan branch per PR
[ROCm/rocprofiler-sdk commit: 5e1643cf81]
* Update tests/bin/transpose/transpose.cpp
- add hipMemGetInfo call to display the available vs. total memory on the GPU
* Update tests/rocprofv3/summary/validate.py
- Updated test_summary_display_data after addition of hipMemGetInfo to transpose test exe
* Tweak code coverage comment uploading
- create unique orphan branch per PR
- reduce quality of PNG files (85 -> 70)
* Revert some of code coverage comment uploading
- remove creation of unique orphan branch per PR
* Tweak code coverage comment uploading
- create unique orphan branch per PR
- Add the new path to avoid WaitAny() calls in AsyncEventsLoopp() with
HSA_WAIT_ANY_DEBUG key. The new path is selected by default.
The optimizaiton combines all logic of WaitAny() in a single processing loop
and avoids extra memory allocations or ref counting. Also it won't spin
on the CPU if all events are busy.
Change-Id: I197ce60d0d023fbb672f700d6e87702686f1f55a
On GPUs where EOP is handled in asic, the read_dispatch_id is not always
updated after each packet. Look for the first dispatch packet that needs
scratch memory before allocating scratch.
Change-Id: Ibf4b4b485f99bf2fabfe48e9609ca99111fdafbe
* Change all rocprofiler-X target names to rocprofiler-sdk-X
* Update rocprofiler-sdk-config.cmake
- fix install tree target names
- simplify logic for using find w/ components and find w/o components
* Update rocprofiler-sdk-roctx-config.cmake
- simplify logic for using find w/ components and find w/o components
* Update samples/intercept_table/CMakeLists.txt
- demonstrate/test use of `find_package(rocprofiler-sdk ... COMPONENTS ...)`
Vladana Stojiljkovic
Yiannis Papadopoulos
Chris Freehill
Alex Xie
Tao Sang
Rahul Manocha
srawat
Gopesh Bhardwaj
Istvan Kiss
Saleel Kudchadker
Jonathan R. Madsen
Giovanni Lenzi Baraldi
Jonathan Kim
Ajay
Rakesh Roy
Manjunath P Jakaraddi
Gopesh
Ben Richard
Benjamin Welton
German Andryeyev
Jaydeep Patel
David Yat Sin