[Mlir-commits] [mlir] [mlir][x86vector] AVX Convert/Broadcast F16 to F32 instructions (PR #137917)

[Adam Siemieniuk]

adam-smnk wrote:

AFAIK, these intrinsics are pretty much equivalent to the bf16 versions with the only difference being expected f16 type, right?

It could be worth taking a moment to generalize the existing bf16 ops to cover both cases. The input memref might need to be constrained to supported data types (still any rank or unranked) to help with intrinsic name generation. That'd also help in guiding users toward correct usage.

https://github.com/llvm/llvm-project/pull/137917