[Mlir-commits] [mlir] [mlir][linalg] Scaled contraction op (PR #203958)

[Adam Siemieniuk]

adam-smnk wrote:

Rebased and updated:
- no special case for tiling - tiling's general semi-affine indexing map validation is sufficient to ensure correct `scaled_contract` tiling, this keeps behavior consistent w.r.t. named op and `linalg.generic`, tiling is using common infrastructure in line with the rest of structured ops
- body uses `arith.scaling_*` when possible - data type combinations that don't match these ops use manually built scaling computation as a fallback
- refined test cases

https://github.com/llvm/llvm-project/pull/203958