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

[Ivan Butygin]

Hardcode84 wrote:

> Out of curiosity, is this for CPU codegen? When do we need it? I thought that we want to get rid of shape_cast. Instead, we want the mermef.collapse_shape version? Is it for "scalar loads/stores" + "fully utilization for vector computation"?

I was the one who added this transform initially, in our case we have a vectors which are logically 2D, but eventually lowered to LLVM/SPIR-V intrinsics which only support 1D types so we operate them as 2D and flatten as the last step.

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