[Mlir-commits] [mlir] [MLIR] Fuse locations of hoisted / merged constants (PR #74670)

Mehdi Amini llvmlistbot at llvm.org
Wed Dec 6 23:46:44 PST 2023


joker-eph wrote:

> I do agree that fusing with the block location is more controversial than fusing constant op locations together. Perhaps the policy should lie with the constant ops themselves

It's not clear to me why one constant vs another would treat it differently, I think it's best to have a separate PR just so that we can see examples of what we're trying to address exactly with this.

https://github.com/llvm/llvm-project/pull/74670


More information about the Mlir-commits mailing list