[all-commits] [llvm/llvm-project] 44583e: [mlir][x86] Fix - multiple issues / F8 support for...

[Arun Thangamani via All-commits]

  Branch: refs/heads/main
  Home:   https://github.com/llvm/llvm-project
  Commit: 44583eca36b6e4e11cc171bc05d9eb0c5b716e1d
      https://github.com/llvm/llvm-project/commit/44583eca36b6e4e11cc171bc05d9eb0c5b716e1d
  Author: Arun Thangamani <arun.thangamani at intel.com>
  Date:   2026-05-26 (Tue, 26 May 2026)

  Changed paths:
    M mlir/lib/Dialect/X86/Transforms/VectorContractToAMXDotProduct.cpp
    M mlir/lib/Dialect/X86/Utils/X86Utils.cpp
    M mlir/test/Dialect/X86/AMX/vector-contract-to-tiled-dp.mlir

  Log Message:
  -----------
  [mlir][x86] Fix - multiple issues / F8 support for AMX dot-product lowering (#196984)

This patch fixes issues or support additional patterns for AMX
`dot-product` lowering.

1. Fix issue related to write-back to `C` matrix,
2. Supports additional lowering pattern where the cache tile sizes are:
32,32,32,
3. Online packing - loop peeling is now based on `step` size,
4. Extends support for `f8` lowering (`mx-fp8` lowering will be
supported after vector.contract has `mx` support).



To unsubscribe from these emails, change your notification settings at https://github.com/llvm/llvm-project/settings/notifications