* Add put to all pes from all lanes concurrently
* Remove wg_init, use size_t for size params, 64bit data exchange (more
bits for verification masking)
* Rename to flood-test, add put,putnbi,p,get,getnbi,g variants, count time
correctly
* Add flood tester to the testing script
* add to gda test case w/o the _g variant that is not implemented.
[ROCm/rocshmem commit: cca7872bcf]
* Add put to all pes from all lanes concurrently
* Remove wg_init, use size_t for size params, 64bit data exchange (more
bits for verification masking)
* Rename to flood-test, add put,putnbi,p,get,getnbi,g variants, count time
correctly
* Add flood tester to the testing script
* add to gda test case w/o the _g variant that is not implemented.
* Improve native tool discovery and partition detection
- Enhanced native tool path resolution to support CMAKE_INSTALL_LIBDIR variations
(lib, lib64, lib32, etc.) using glob pattern matching
- Extracted path variables to avoid duplication in error messages
- Improved error message clarity by showing exact paths searched for .so and .cpp files
- Simplified code path construction using consistent Path.resolve().parents[x] syntax
- Fixed redundant partition warnings on pre-MI300 GPUs by adding architecture check
- Only query compute/memory partition on MI300+ series (gfx940+)
- Added proper type hints for gpu_arch parameter
- Moved gpu_info extraction after soc_info to ensure gpu_arch is available
- Improved code comments for MI300 series threshold
* Handle gpu arch like a hex string
GCC does not support anonymous structs with members that have non-trivial constructors. This commit changes the header to remove the union when compiling with gcc. This should be a non-breaking change for other compilers.
- use the reduce_psync buffers for synchronization in allreduce, not the
barrier_psync.
- execute a wwg barrier after the allreduce operation. After internal
discussion it was determined that it is required for correctness.
[ROCm/rocshmem commit: 6f512e92a5]
- use the reduce_psync buffers for synchronization in allreduce, not the
barrier_psync.
- execute a wwg barrier after the allreduce operation. After internal
discussion it was determined that it is required for correctness.
Tests were relying on floats for calculating ulp values when validating the output. This is not correct given that the calculations are done using Float16. The fix is to update the test framework to use fp16 ulp instead.
* byteswap<T> returns by value
* replace hand-rolled implementations with Clang __builtin_bswap<N> intrinsics
* new high-level interface endian::to_be, endian::from_be, etc. to indicate conversion direction
[ROCm/rocshmem commit: cf8b72a047]
* byteswap<T> returns by value
* replace hand-rolled implementations with Clang __builtin_bswap<N> intrinsics
* new high-level interface endian::to_be, endian::from_be, etc. to indicate conversion direction
* attach: Formalize ROCAttach API
- Make ROCAttach public with public headers
- Change detach to take a PID
- attach and detach are now reentrant
- Cleanup of states and signal handling in ptrace session
- Fixes mixed up definition of ROCPROF_ATTACH_TOOL_LIBRARY
- ROCPROF_ATTACH_TOOL_LIBRARY now always means the tool library loaded by the attachment target
- ROCPROF_ATTACH_LIBRARY refers to the library used to perform attachment
- Add direct call of rocprof-attach
- Fix python library call of rocprof-attach
- Function now named attach(), changed from main()
* attach: rocprof-compute ROCAttach updates
- Update to new library names
- Correct usage of C lib detach
* attach: add test for rocattach
- Disable ASan, TSan, and UBSan for the new parallel-attach test
- Lower log level for LSan tests, existing behavior from other tests
---------
Co-authored-by: Ammar ELWazir <aelwazir@amd.com>
Motivation
We wish to avoid triggering full Jenkins runs for docs-only PRs, as this takes up testing resources and slows development time. rocm_ci_caller.yml already excludes some docs-only changes, but this can be improved to exclude them along more paths.
Technical Details
The checks that rocm_ci_caller.yml uses to determine if a changed file in a PR is worth a Jenkins run has been increased to exclude more paths and more file suffixes.
JIRA ID
AIROCDOC-78, AIROCDOC-424
Test Plan
Created a test branch users/dsclear/shorten_workflows_test_root with the changes in this PR, branched from develop.
Branched users/dsclear/shorten_workflows_test_bin_3 and users/dsclear/shorten_workflows_test_text_3 from users/dsclear/shorten_workflows_test_root.
Modified users/dsclear/shorten_workflows_test_bin_3 to add two .h files, and submitted a PR into users/dsclear/shorten_workflows_test_root (Test PR, do not merge. Test PR to test Jenkins CI/CD modifications. #2613).
Modified users/dsclear/shorten_workflows_test_text_3 to add a new .txt file, and submitted a PR into users/dsclear/shorten_workflows_test_root (Test PR, do not merge. Test PR to test Jenkins CI/CD modifications (docs only). #2614).
Test Result
The test PR in step 3 caused rocm_ci_caller.yml to attempt to trigger Jenkins, as this is a 'non-docs' change.
The test PR in step 4 had the attempt to trigger Jenkins skipped, as this is a 'docs-only' change.
* Fix the amdgpu version string comparison
The intention behind it was to avoid showing the string if it's not
got information.
Signed-off-by: Mario Limonciello <mario.limonciello@amd.com>
* Display the kernel version in amd-smi output
This is an interesting debugging point, especially in the case of
not having a DKMS package installed.
Signed-off-by: Mario Limonciello <mario.limonciello@amd.com>
Signed-off-by: Mario Limonciello (AMD) <superm1@kernel.org>
* Moving os_kernel_version to static --driver
Signed-off-by: Maisam Arif <Maisam.Arif@amd.com>
---------
Signed-off-by: Mario Limonciello <mario.limonciello@amd.com>
Signed-off-by: Mario Limonciello (AMD) <superm1@kernel.org>
Signed-off-by: Maisam Arif <Maisam.Arif@amd.com>
Co-authored-by: Maisam Arif <Maisam.Arif@amd.com>
Increased threshold from 2100 μs to 3100 µs to accommodate
gpu_metric read time variation across Navi systems.
Signed-off-by: Bindhiya Kanangot Balakrishnan <Bindhiya.KanangotBalakrishnan@amd.com>
Problem:
When TheRock-based PyTorch package is installed along with amdsmi, importing
torch causes a double-free crash on exit (GitHub issue ROCm/TheRock#2269).
Root cause:
Both librocm_smi64.so and libamd_smi.so export the C++ static member
'amd::smi::Device::devInfoTypesStrings'. When libraries are loaded with
RTLD_GLOBAL, the dynamic linker resolves libamd_smi.so's reference to this
symbol to the one in librocm_smi64.so. This causes:
1. librocm_smi64.so registers its destructor for devInfoTypesStrings
2. libamd_smi.so also registers a destructor, but for the SAME address
3. On exit, both destructors run on the same object -> double-free
Fix:
Change devInfoTypesStrings from a class static member to a file-local static
variable. This ensures the symbol has internal linkage and is not exported,
preventing the symbol collision.
Changes:
- rocm_smi_device.h: Remove static member declaration
- rocm_smi_device.cc: Change from 'Device::devInfoTypesStrings' to file-local
'static const std::map<...> devInfoTypesStrings'
- rocm_smi.cc: Remove the global alias to the (now removed) class member
Tested on gfx1151. `import torch` crashed on exit before the fix, and doesn't crash after the fix.
* Add hipDeviceAttributeExpertSchedMode
---------
Co-authored-by: Stefan Sokolovic <stefan.sokolovic2@amd.com>
* Update hipDeviceAttributeExpertSchedMode unit test
* Move check to ROCr from thunk interface
* Revert unrelated whitespace changes
* Revert version bump
---------
Co-authored-by: Stefan Sokolovic <stefan.sokolovic2@amd.com>
## Motivation
- Structured logging with proper log levels (TRACE, DEBUG, INFO, WARNING, ERROR, CRITICAL)
- Better performance through compile-time formatting
- Consistent formatting using fmt library
- Runtime log level control via arguments and environment variables
- Easier maintenance and debugging capabilities
## Technical Details
- Added spdlog as a submodule and integrated it into CMake build system
- Created new `rocprofiler-systems-logger` library wrapping spdlog functionality
- Replaced custom logging macros (`ROCPROFSYS_VERBOSE`, `ROCPROFSYS_DEBUG`, `ROCPROFSYS_FATAL`, `ROCPROFSYS_REQUIRE`, `ROCPROFSYS_CI_THROW`, etc.) with spdlog equivalents (`LOG_DEBUG`, `LOG_WARNING`, `LOG_CRITICAL`, etc.)
- Implemented log level control through command-line arguments and environment variables
- Converted assertion macros to proper error handling with exceptions and std::abort()
* [SWDEV-553434] No direct way to get the BASEBOARD temperature info. Need to iterate all gpus
Signed-off-by: amd-josnarlo <josnarlo.amd.com>
---------
Signed-off-by: amd-josnarlo <josnarlo.amd.com>
Co-authored-by: amd-josnarlo <josnarlo.amd.com>
Add explicit test cases to verify ROCr functionality with interrupts
disabled (HSA_ENABLE_INTERRUPT=0). This ensures compatibility with
virtio, dtif, and WSL configurations which require interrupt-disabled
mode.
Signed-off-by: Horatio Zhang <Hongkun.Zhang@amd.com>
* rocr: Add ProtectMemory API and use it in RemoveAccess
Replace munmap + mmap with mprotect when removing memory access.
This improves performance by 5-10x, ensures atomicity (no race
condition window), and prepares for WSL/DXG compatibility fixes.
Suggested-by: David Yat Sin <David.YatSin@amd.com>
Signed-off-by: Flora Cui <flora.cui@amd.com>
Signed-off-by: Horatio Zhang <Hongkun.Zhang@amd.com>
* rocr: Skip CPU mapping operations on WSL
On WSL, CPU cannot access GPU VRAM due to platform restrictions.
CPU access would fault-in system RAM instead, causing data corruption
and memory leaks. Return HSA_STATUS_ERROR to fail fast rather than
silently creating broken mappings. GPU-to-GPU mappings remain functional.
Signed-off-by: Flora Cui <flora.cui@amd.com>
Signed-off-by: Horatio Zhang <Hongkun.Zhang@amd.com>
* rocr: reduce ifdef linux
v2: Fix IsDXG check logic
Signed-off-by: David Yat Sin <David.YatSin@amd.com>
Signed-off-by: Horatio Zhang <Hongkun.Zhang@amd.com>
---------
Signed-off-by: Horatio Zhang <Hongkun.Zhang@amd.com>
Signed-off-by: David Yat Sin <David.YatSin@amd.com>
Signed-off-by: Flora Cui <flora.cui@amd.com>
* Adding working single node tests
* Revert to old docker sha
* adding back no perf tests
---------
Co-authored-by: Aravind Ravikumar <arravikum@amd.com>
[ROCm/rccl commit: 4b295c9893]
* Adding working single node tests
* Revert to old docker sha
* adding back no perf tests
---------
Co-authored-by: Aravind Ravikumar <arravikum@amd.com>
AMD_LOG_LEVEL_SIZE is being used in a global variable.
This always uses the default value of 2048 because the
HIP runtime doesn't have the opportunity to load
environment variables at the point where global variables
are initialized.
The solution is to use AMD_LOG_LEVEL_SIZE inside
truncate_log_file() function.
## Motivation
<!-- Explain the purpose of this PR and the goals it aims to achieve. -->
Remove Fortran example due to Palamida scan violation.
## Technical Details
<!-- Explain the changes along with any relevant GitHub links. -->
Revert 63713f01e0.
New test to be added later.
Signed-off-by: David Galiffi <David.Galiffi@amd.com>
Sets heavy GitHub CI workflows to not trigger on docs-only changes.
Specifically, sets azure-ci-dispatcher.yml and therock-ci.yml, as well as many rocprofiler workflows, to not trigger when the change consists entirely of docs-only files.
akolliasAMD
German Andryeyev
Aurelien Bouteiller
vedithal-amd
Fábio Mestre
Edgar Gabriel
Kian Cossettini
Omri Mor
Mark Meserve
dsclear-amd
Mario Limonciello
yugang-amd
Bindhiya Kanangot Balakrishnan
Matthias Gehre
Filip Jankovic
Milan Radosavljevic
Joseph Narlo
David Yat Sin
xuchen-amd
Luca Bruni
habajpai-amd
Gopesh Bhardwaj
pghoshamd
hongkzha-amd
Geo Min
Jan Stephan
Adam Pryor
AidanBeltonS
David Galiffi
randyh62