[Mlir-commits] [mlir] [mlir][OpenACC] Support static and dynamic multi-rank OpenACC array reduction accumulators (PR #210853)

[llvmlistbot at llvm.org]

khaki3 wrote:

>would it be worth carrying that decision forward explicitly (a small side-table keyed on the storage value, or an attribute on acc.reduction_accumulate_array) rather than reconstructing it through casts/stores/loads? 

Yes, you are right. This PR does not touch the classification logic anymore. It only adds rank-N initialization and indexing. I agree that we should carry the storage decision explicitly, but that should be separate work.

This PR is ready.

https://github.com/llvm/llvm-project/pull/210853