[all-commits] [llvm/llvm-project] dd5357: [mlir][XeGPU][Transform] Add XeGPU contiguity anal...

Md Abdullah Shahneous Bari via All-commits all-commits at lists.llvm.org
Fri Jun 26 09:15:43 PDT 2026


  Branch: refs/heads/main
  Home:   https://github.com/llvm/llvm-project
  Commit: dd5357d38d6b73e3a687bcc5ea8cb3a858fb3fea
      https://github.com/llvm/llvm-project/commit/dd5357d38d6b73e3a687bcc5ea8cb3a858fb3fea
  Author: Md Abdullah Shahneous Bari <md.abdullah.shahneous.bari at intel.com>
  Date:   2026-06-26 (Fri, 26 Jun 2026)

  Changed paths:
    M mlir/include/mlir/Dialect/XeGPU/IR/XeGPUOps.td
    M mlir/include/mlir/Dialect/XeGPU/Transforms/Transforms.h
    M mlir/lib/Conversion/VectorToXeGPU/VectorToXeGPU.cpp
    M mlir/lib/Dialect/XeGPU/IR/XeGPUOps.cpp
    M mlir/lib/Dialect/XeGPU/Transforms/CMakeLists.txt
    A mlir/lib/Dialect/XeGPU/Transforms/XeGPUContiguityAnalysis.cpp
    M mlir/lib/Dialect/XeGPU/Transforms/XeGPUSgToLaneDistribute.cpp
    M mlir/lib/Dialect/XeGPU/Transforms/XeGPUUnroll.cpp
    M mlir/lib/Dialect/XeGPU/Transforms/XeGPUWgToSgDistribute.cpp
    A mlir/test/Dialect/XeGPU/contiguity-analysis.mlir
    M mlir/test/Dialect/XeGPU/invalid.mlir
    M mlir/test/Dialect/XeGPU/ops.mlir
    A mlir/test/Dialect/XeGPU/test-xegpu-coalesce-gather-scatter.mlir
    M mlir/test/lib/Dialect/XeGPU/TestXeGPUTransforms.cpp

  Log Message:
  -----------
  [mlir][XeGPU][Transform] Add XeGPU contiguity analysis. (#201684)

Add an AxisInfo-based (borrows the idea from Triton Axis Info analysis)
dataflow analysis that computes, for each
`xegpu.load` / `xegpu.store` gather/scatter, how many elements are
contiguous
along the innermost offsets dimension, and stamps that count as a
`contiguity` **operation attribute** (`OptionalAttr<I64Attr>`) on the
op.

`contiguity` is a target-independent property of the offsets, not a
request tied to any optimization — a consumer is free to use or ignore
it. The
analysis performs no rewrite. Turning the property into a concrete
`lane_layout` / `lane_data` split (which needs the subgroup size) and
the
actual memory-message rewrite are consumer concerns, handled by later
layout-propagation steps (subsequent PRs) or, for testing, by the apply
helper
in the test pass (`-test-xegpu-coalesce-gather-scatter`).

---------

Co-authored-by: Claude Opus 4.8 <noreply at anthropic.com>



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


More information about the All-commits mailing list