[Mlir-commits] [mlir] [MLIR][OpenMP] Group clause operands into structures (PR #86797)

Sergio Afonso llvmlistbot at llvm.org
Mon Apr 8 06:14:01 PDT 2024


skatrak wrote:

> So, to clarify, is the patch providing a set of new builders (build functions) that can accept clause operands in any order? If so can you change the title to reflect this?

I'll try to explain myself a bit better: The main purpose of this patch is to define a set of structures holding MLIR operands associated with each clause and operation. It also happens to add a first use for these structures, which is the definition new op builders that take a single instance of one of these structures. The new builders prevent having to check the order of arguments since they become named fields of the single structure instance passed in. Would you suggest updating this PR name to "Group clause operands into structures and use them to define simplified op builders" or something like that?

> 
> How does this affect the ones that use block arguments?

It doesn't, because these clause structures are directly forwarded to operation arguments and attributes of the corresponding MLIR operation, and entry block arguments are instead part of the region(s) attached to the operation and not the `mlir::Operation` itself. Since these regions are created separately from the operation itself, they are (at least initially) not stored in these structures.


https://github.com/llvm/llvm-project/pull/86797


More information about the Mlir-commits mailing list