Skip to content

[mlir][transform] Fix failure in flattening already flattened linalg ops #86037

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Merged
merged 2 commits into from
Mar 21, 2024
Merged

[mlir][transform] Fix failure in flattening already flattened linalg ops #86037

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
2 commits
Select commit Hold shift + click to select a range
127b998
Fix failure in flattening already flattened linalg ops
srcarroll Mar 20, 2024
ddd177d
Merge branch 'main' into fix-elementwise-flattening
srcarroll Mar 21, 2024
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
15 changes: 10 additions & 5 deletions mlir/lib/Dialect/Linalg/TransformOps/LinalgTransformOps.cpp
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -3269,15 +3269,20 @@ DiagnosedSilenceableFailure transform::FlattenElementwiseLinalgOp::applyToOne(
transform::ApplyToEachResultList &results,
transform::TransformState &state) {
rewriter.setInsertionPoint(target);
if (target.getNumLoops() <= 1)
if (!isElementwise(target)) {
failed(rewriter.notifyMatchFailure(
target, "only elementwise flattening is supported"));
return emitDefaultSilenceableFailure(target);
Comment on lines +3273 to +3275
Copy link
Member

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Instead of sending the error message to the non-existent rewrite driver, this should rather emit it as a silenceable failure! The default failure messages produced by emitDefaultSilenceableFailure are extremely unhelpful and must only be used as last resort.

I also haven't seen the usage of failed as a way to suppress the compiler warning about LogicalResult being unused. The common idiom is to use the C-style cast to void.

Copy link
Contributor Author

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Sorry, I'm new to the transform framework and still learning how things work. I was just following another implementation in this same file for the message. I'll submit another PR to fix.

About tagging you. I made an RFC for this per your suggestion with a mention to the original PR and Mahesh was the only one that took any action so I just went with him again for this. I'll be sure to include you in the future.

Copy link
Member

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

No big deal, the point of review is also for contributors to learn things.

Copy link
Contributor Author

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

appreciate it. thanks for teaching me. :)

}
// If rank <= 1, do nothing
if (target.getNumLoops() <= 1) {
results.push_back(target);
return DiagnosedSilenceableFailure::success();
}
ReassociationIndices reassociation(target.getNumLoops());
std::iota(reassociation.begin(), reassociation.end(), 0);
auto maybeFlattened =
(isElementwise(target))
? collapseOpIterationDims(target, reassociation, rewriter)
: FailureOr<CollapseResult>(rewriter.notifyMatchFailure(
target, "only elementwise flattening is supported"));
collapseOpIterationDims(target, reassociation, rewriter);
if (failed(maybeFlattened))
return emitDefaultSilenceableFailure(target);
results.push_back(maybeFlattened->collapsedOp);
Expand Down
21 changes: 21 additions & 0 deletions mlir/test/Dialect/Linalg/flatten-elementwise.mlir
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -67,6 +67,27 @@ module attributes {transform.with_named_sequence} {

// -----

// CHECK-LABEL: func.func @map_already_flat(
// CHECK-SAME: %[[ARG0:[a-zA-Z0-9_]*]]: memref<32xf32>
// CHECK-SAME: %[[ARG1:[a-zA-Z0-9_]*]]: memref<32xf32>
// CHECK-SAME: %[[ARG2:[a-zA-Z0-9_]*]]: memref<32xf32>
// CHECK-NEXT: linalg.map { arith.addf } ins(%[[ARG0]], %[[ARG1]] : memref<32xf32>, memref<32xf32>) outs(%[[ARG2]] : memref<32xf32>)
func.func @map_already_flat(%arg0: memref<32xf32>, %arg1: memref<32xf32>, %arg2: memref<32xf32>) {
linalg.map {arith.addf} ins(%arg0, %arg1: memref<32xf32>, memref<32xf32>) outs(%arg2: memref<32xf32>)
return
}

module attributes {transform.with_named_sequence} {
transform.named_sequence @__transform_main(%arg1: !transform.any_op {transform.readonly}) {
%0 = transform.structured.match interface{LinalgOp} in %arg1 : (!transform.any_op) -> !transform.any_op
%flattened = transform.structured.flatten_elementwise %0
: (!transform.any_op) -> !transform.any_op
transform.yield
}
}

// -----

// CHECK: #[[$MAP0:.*]] = affine_map<(d0) -> (d0)>
// CHECK-LABEL: func.func @generic
// CHECK-SAME: %[[ARG0:[a-zA-Z0-9_]*]]: memref<32x7xf32>
Expand Down