https://github.com/MattPD commented:
FWIW, I looked specifically at the explicit-task `in_reduction` MLIR to LLVM IR translation, and that part looks correct to me. The runtime lookup is emitted inside the outlined task body, so it runs on the executing thread's gtid. The `in_reduction` block arguments are remapped to the per-task storage returned by `__kmpc_task_reduction_get_th_data`, and they are disjoint from the privatization block arguments. The non-default-address-space handling matches the approach in #199670, with the default-address-space path unchanged. I didn't review the broader design beyond that path, so I'll leave the overall sign-off to the other reviewers. https://github.com/llvm/llvm-project/pull/202611 _______________________________________________ llvm-branch-commits mailing list [email protected] https://lists.llvm.org/cgi-bin/mailman/listinfo/llvm-branch-commits
