[all-commits] [llvm/llvm-project] ebeb87: [MLIR][XeGPU] Restrict XeGPUPropagateLayout analys...

[Tuomas Kärnä via All-commits]

  Branch: refs/heads/main
  Home:   https://github.com/llvm/llvm-project
  Commit: ebeb871f9a5fd31f088b7fb952c99cc4da17eeb1
      https://github.com/llvm/llvm-project/commit/ebeb871f9a5fd31f088b7fb952c99cc4da17eeb1
  Author: Tuomas Kärnä <tuomas.karna at intel.com>
  Date:   2026-09-09 (Wed, 09 Sep 2026)

  Changed paths:
    M mlir/lib/Dialect/XeGPU/Transforms/XeGPUPropagateLayout.cpp

  Log Message:
  -----------
  [MLIR][XeGPU] Restrict XeGPUPropagateLayout analysis scope (#221793)

This PR prevents concurrent access to traversal and mutation of the
shared IR by restricting the walk scope to determine program order
within each gpu.module.

Previously, although the pass is scoped to a single gpu.module, this
walk climbs all the way up to the enclosing top-level builtin.module and
traverses the entire module — including sibling gpu.modules. When there
is multiple gpu.modules, it causes concurrent traversal and mutation of
the shared IR and eventually leads to a segfault.

Assisted-by: Claude



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