[Mlir-commits] [mlir] [mlir][xegpu] Convert Vector contraction to XeGPU (PR #122115)

[Adam Siemieniuk]

adam-smnk wrote:

Similarly to the other vector op conversions, I opted to exclude most hardware-specific checks. The lowering largely assumes that the `vector.contract` is already in a form suitable for its target i.e., supported combination of operand types, matrix sizes etc.
Basic checks around K-dim and VNNI factor are present to comply with `xegpu.dpas` op documentation.

@chencha3 The M and N sizes could also be constrained. Although, the dpas verifier doesn't care about them right now, and AFAIK these can technically change a bit depending on target HW. So, as you prefer.

https://github.com/llvm/llvm-project/pull/122115