230b437d05
### 1. ElementType deduction for pointer-based array sections Problem: Pointer-based array sections were previously ignored during `ElementType` deduction, leading to incorrect assumptions about array item types. This often resulted in out-of-bounds access, as seen in the assertion failure: ``` Assertion `idx < size()' failed. llvm-project/llvm/include/llvm/ADT/SmallVector.h:292: reference llvm::SmallVectorTemplateCommon<llvm::Value *>::operatorsize_type [T = llvm::Value *] ``` Fix: Added a check in clang/lib/CodeGen/CGOpenMPRuntime.cpp to ensure `ElementType` is correctly detected for cases involving non-contiguous updates with a base pointer. Impact: Resolves failures in OpenMP_VV (formerly sollve_vv) and other offload/clang-OpenMP tests: All tests under: https://github.com/OpenMP-Validation-and-Verification/OpenMP_VV/tree/master/tests/5.0/target_update test_target_update_mapper_from_discontiguous.c test_target_update_mapper_to_discontiguous.c test_target_update_to_discontiguous.c test_target_update_from_discontiguous.c ### 2. Zero-dimension propagation in struct member mappings Problem: A zero-dimension entry for struct members introduced inconsistencies in complex mapping logic within OMPIRBuilder.cpp. Placeholder zeros propagated to emitNonContiguousDescriptor(), breaking reverse indexing logic and corrupting IR: Loops assume `Dims[I] >= 1`. When `Dims[I] == 0`: Reverse indexing still stores pointers to uninitialized allocas or mismatched slots. Runtime interprets `ArgSizes[I]` (derived from `Dims[I])` as dimensionality, causing size/offset calculations to collapse to zero → results in `size=0` async copy and plugin interface errors. Fix: Prepend a synthetic dimension of size 1 instead of appending a zero, preserving correctness in `targetDataUpdate()` for non-contiguous updates. Impact: Added dedicated test cases that previously failed on main.
The LLVM/Offload Subproject
The Offload subproject aims at providing tooling, runtimes, and APIs that allow users to execute code on accelerators or other "co-processors" that may or may not match the architecture of their "host". In the long run, all kinds of targets are in scope of this effort, including but not limited to: CPUs, GPUs, FPGAs, AI/ML accelerators, distributed resources, etc.
For OpenMP offload users, the project is ready and fully usable. The final API design is still under development. More content will show up here and on our webpage soon. In the meantime, people are encouraged to participate in our meetings (see below) and check our development board as well as the discussions on Discourse.
Meetings
Every second Wednesday, 7:00 - 8:00am PT, starting Jan 24, 2024. Alternates with the OpenMP in LLVM meeting. invite.ics Meeting Minutes and Agenda