[all-commits] [llvm/llvm-project] 1dd00d: [mlir][Vector] Fix a propagation bug with broadcast

[qcolombet via All-commits]

  Branch: refs/heads/main
  Home:   https://github.com/llvm/llvm-project
  Commit: 1dd00d39037b14e06555a79a397ee1e85d787db9
      https://github.com/llvm/llvm-project/commit/1dd00d39037b14e06555a79a397ee1e85d787db9
  Author: Quentin Colombet <quentin.colombet at gmail.com>
  Date:   2023-06-06 (Tue, 06 Jun 2023)

  Changed paths:
    M mlir/lib/Dialect/Vector/Transforms/VectorDistribute.cpp
    M mlir/test/Dialect/Vector/vector-warp-distribute.mlir

  Log Message:
  -----------
  [mlir][Vector] Fix a propagation bug with broadcast

In the vector distribute patterns, we used to move
`vector.broadcast`s out of `vector.warp_execute_on_lane0`s
irrespectively of how they were defined.

This could create broadcast operations with invalid semantic.
E.g.,
```
%r = warop ...[32] ... -> vector<1x2xf32> {
  %val = broadcast %in : vector<64xf32> to vetor<1x64xf32>
  vector.yield %val : vector<1x64xf32>
}
```
=>
```
%r = warop ...[32] ... -> vector<64xf32> {
  vector.yield %in : vector<64xf32>
}
// Broadcasting to a narrower type!
broadcast %r : vector<64xf32> to vector<1x2xf32>
```

The root issue is we are trying to broadcast something that is not the same
for each thread, so there is actually nothing to propagate here.

The fix checks that the broadcast we want to create actually makes sense.

Differential Revision: https://reviews.llvm.org/D152154