Intel® SHMEM 1.5.2

The Intel OpenSHMEM team is delighted to announce the release of Intel® SHMEM v1.5.2. Below is a summarized list of features and enhancements that are introduced in this new release.

• Significant stabilization and correctness improvements across runtime, memory management, IPC, and collective operations.
• Build and packaging improvements including split static archives, shared host library support, improved symbol export handling, and refined SYCL host/device compilation behavior.
• Reliability fixes in GPU and IPC paths, including prevention of segmentation faults, safer device buffer handling, and improved cleanup on failure paths.
• Collective operation updates including primary work-group reduce support and rank-0 exscan behavior alignment with validation expectations.
• Extended multi-device and launcher flexibility through explicit device selection support and launcher-independent initialization workflows.
• Expanded hardware support for Intel® Arc™ B-Series (BMG) GPUs.
• Improved CI and test robustness with workflow updates, wrapper and exclusion list refinements, and environment-specific validation fixes.
• Added UID-based MPI initialization path for launcher-independent workflows.
• Added public API support in ishmemx_attr_t: use_uid, nranks, rank, and uid.
• Added ishmemx_uniqueid_t and ishmemx_get_uniqueid to simplify setup.
• Added examples/7_uid_mpcp.cpp demonstrating manual rank launch and UID initialization.
• Added runtime compatibility support for cases where MPI is initialized outside Intel® SHMEM.
• Documentation and examples refreshed to reflect updated runtime and launcher capabilities.