[Mlir-commits] [mlir] Add Vector bitwidth target to Linearize Vectorizable and Constant Ops (PR #83314)

[Diego Caballero]

dcaballe wrote:

Hey @Hardcode84! Just to provide a bit more context, the goal here is to add a bit more control to the flattening. The ultimate state should allow us to (optionally) unroll up to the first multiple of the provided vector bitwidth, and leave the rest of the unrolling to the passes intended for that. Hopefully that makes sense to you! We are also adding the same functionality to the vector transfer read/write counterpart: https://github.com/llvm/llvm-project/pull/81966

https://github.com/llvm/llvm-project/pull/83314