You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
- Finalization Control: Ensures proper finalization of classes when references to the OPAL utility layer reach zero. A new smoke test has been added to validate this behavior.
18
+
- Test and CI Additions: A smoke test was introduced for session management, also integrated into macOS GitHub CI to ensure stability across platforms.
19
+
- Recognize requests from the same instance rather than same communicator, fixes Waitall/Waitany/Testall/Testany.
20
+
- Multiple Init/Finalize Handling: Improvements to handling session multiple initializations and finalizations, addressing issues with parameter management and session destruction.
21
+
22
+
- CUDA and NVIDIA HPC Compiler Compatibility:
23
+
- Stream-ordered Allocations: Adjustments for handling CUDA stream-ordered allocations and assigning primary device context in `opal/cuda` (Akshay Venkatesh).
24
+
- nvfortran Compatibility: Configure script updated to handle NVIDIA's `nvfortran` compiler, enabling proper compilation with `-fPIC` flags for shared libraries (Gilles Gouaillardet).
25
+
- CUDA Memory Address Management: Improvements for handling virtual memory manager (VMM) pointers in CUDA applications to prevent runtime issues.
26
+
27
+
- MPI Communication and Resource Management:
28
+
- Info Duplication: Enhanced `MPI_Info_dup` function to ensure consistent key management across duplicated info objects, supporting better memory and resource handling.
29
+
- OFI/Initialization Improvements: Refactored code for OFI common components to support session-based initialization models per MPI 4 standards.
30
+
- Resource Leak Fixes: Addressed memory leaks and reference count issues in several components to improve memory management and stability in high-demand environments.
31
+
32
+
- Detailed Locking Protocol: Modified default file-locking protocols in UFS component to ensure data consistency, especially when using data-sieving operations, which require broader locking.
33
+
34
+
- MPI_T Interface Enhancements: Modified MCA parameter settings to allow runtime tuning of collectives through the MPI_T interface, adding flexibility for dynamic rules in collective operations.
35
+
- Corrected various documentation typos and clarified parameter settings in several API references, improving readability and usability for end users.
0 commit comments