[all-commits] [llvm/llvm-project] 824fc3: [mlir][vector] Constrain broadcast->shape_cast fol...

[Benoit Jacob via All-commits]

  Branch: refs/heads/main
  Home:   https://github.com/llvm/llvm-project
  Commit: 824fc35bd9453e1776f098cc5d7123b1034ea510
      https://github.com/llvm/llvm-project/commit/824fc35bd9453e1776f098cc5d7123b1034ea510
  Author: Benoit Jacob <jacob.benoit.1 at gmail.com>
  Date:   2026-04-07 (Tue, 07 Apr 2026)

  Changed paths:
    M mlir/lib/Dialect/Vector/IR/VectorOps.cpp
    R mlir/test/Dialect/Linalg/transform-op-mmt4d-to-fma.mlir
    M mlir/test/Dialect/Vector/canonicalize/vector-to-shape-cast.mlir
    A mlir/test/Dialect/Vector/vector-multi-reduction-to-fma.mlir

  Log Message:
  -----------
  [mlir][vector] Constrain broadcast->shape_cast folding (#190230)

Fixes https://github.com/llvm/llvm-project/issues/190614.

Do not fold broadcast->shape_cast when that would result in switching
between the two distinct semantic modes of `vector.broadcast`, as
explained in https://github.com/llvm/llvm-project/issues/190614.

This fixes incorrect-result bugs in IREE:
https://github.com/iree-org/iree/issues/23952

---------

Signed-off-by: Benoit Jacob <benoit.jacob at amd.com>



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