[all-commits] [llvm/llvm-project] cb9286: [mlir][xegpu] Fold memref offset in peephole creat...

[Nishant Patel via All-commits]

  Branch: refs/heads/main
  Home:   https://github.com/llvm/llvm-project
  Commit: cb9286833f2b6bac93edcc258289b54b4681037b
      https://github.com/llvm/llvm-project/commit/cb9286833f2b6bac93edcc258289b54b4681037b
  Author: Nishant Patel <nishant.b.patel at intel.com>
  Date:   2026-09-21 (Mon, 21 Sep 2026)

  Changed paths:
    M mlir/lib/Dialect/XeGPU/Transforms/XeGPUPeepHoleOptimizer.cpp
    M mlir/test/Dialect/XeGPU/peephole-optimize.mlir

  Log Message:
  -----------
  [mlir][xegpu] Fold memref offset in peephole create_nd_tdesc rewrite (#224459)

XeGPUCreateNdDescOpPattern in peephole optimization only extracted
strided metadata for dynamic-shape
memrefs, but hasStaticShapeAndStrides() ignores the offset — so a
static-shape memref with a nonzero or dynamic offset (e.g. from a
subview) would silently read from offset 0. This PR changes it to always
extract and fold the offset into the base pointer.


Assisted by Claude



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