[all-commits] [llvm/llvm-project] e436bf: [MLIR][XeGPU] Add sg_map for scatter verification ...

[Artem Kroviakov via All-commits]

  Branch: refs/heads/main
  Home:   https://github.com/llvm/llvm-project
  Commit: e436bf64b50ac7f18218b8168da13176457fa3aa
      https://github.com/llvm/llvm-project/commit/e436bf64b50ac7f18218b8168da13176457fa3aa
  Author: Artem Kroviakov <71938912+akroviakov at users.noreply.github.com>
  Date:   2025-01-30 (Thu, 30 Jan 2025)

  Changed paths:
    M mlir/include/mlir/Dialect/XeGPU/IR/XeGPUOps.td
    M mlir/lib/Dialect/XeGPU/IR/XeGPUOps.cpp
    M mlir/test/Dialect/XeGPU/XeGPUOps.mlir
    M mlir/test/Dialect/XeGPU/invalid.mlir

  Log Message:
  -----------
  [MLIR][XeGPU] Add sg_map for scatter verification (#124300)

This PR adds sg_map to the verification of scatter ops in XeGPU.

The documentation says `chunk_size: indicates the number of continuous
elements accessed for each offset`, it also mentions the fact that
scatter ops are SG-level.
Hence, if an operation is distributed to work-items, a 1-d load means a
work-item reads one element, a 2-d load means a work-item loads
`chunk-size` or second dimension of tdesc elements. The changes in this
PR reflect the documentation with the presence of sg_map attribute
(i.e., distributed case).



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