Skip to content

[mlir][VectorOps] Add fold vector.shuffle -> vector.interleave (4/4) #80968

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 6, 2024
Merged

[mlir][VectorOps] Add fold vector.shuffle -> vector.interleave (4/4) #80968

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
d45d47e
[mlir][VectorOps] Add fold vector.shuffle -> vector.interleave
MacDue Feb 7, 2024
6574611
Fixups
MacDue Mar 6, 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
42 changes: 41 additions & 1 deletion mlir/lib/Dialect/Vector/IR/VectorOps.cpp
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -2479,11 +2479,51 @@ class ShuffleSplat final : public OpRewritePattern<ShuffleOp> {
}
};

/// Pattern to rewrite a fixed-size interleave via vector.shuffle to
/// vector.interleave.
class ShuffleInterleave : public OpRewritePattern<ShuffleOp> {
public:
using OpRewritePattern::OpRewritePattern;

LogicalResult matchAndRewrite(ShuffleOp op,
PatternRewriter &rewriter) const override {
VectorType resultType = op.getResultVectorType();
if (resultType.isScalable())
return rewriter.notifyMatchFailure(
op, "ShuffleOp can't represent a scalable interleave");
Comment on lines +2491 to +2493
Copy link
Contributor

Choose a reason for hiding this comment

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

this case shouldn't be allowed by the verifier so I would replace this with an assert.

Copy link
Member Author

Choose a reason for hiding this comment

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

There's currently no checks on vector.shuffle for scalable vectors 😔

Copy link
Contributor

Choose a reason for hiding this comment

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

it would be better to add that to the verified


if (resultType.getRank() != 1)
return rewriter.notifyMatchFailure(
op, "ShuffleOp can't represent an n-D interleave");

VectorType sourceType = op.getV1VectorType();
if (sourceType != op.getV2VectorType() ||
Comment on lines +2499 to +2500
Copy link
Contributor

Choose a reason for hiding this comment

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

Can shuffle inputs have different types?

Copy link
Member Author

Choose a reason for hiding this comment

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

I don't think so. From the shuffle docs:

the two operands must have the same element type as the result

Copy link
Contributor

Choose a reason for hiding this comment

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

I thought their shapes should also match. Anyways...

sourceType.getNumElements() * 2 != resultType.getNumElements()) {
return rewriter.notifyMatchFailure(
op, "ShuffleOp types don't match an interleave");
}

ArrayAttr shuffleMask = op.getMask();
int64_t resultVectorSize = resultType.getNumElements();
for (int i = 0, e = resultVectorSize / 2; i < e; ++i) {
int64_t maskValueA = cast<IntegerAttr>(shuffleMask[i * 2]).getInt();
int64_t maskValueB = cast<IntegerAttr>(shuffleMask[(i * 2) + 1]).getInt();
if (maskValueA != i || maskValueB != (resultVectorSize / 2) + i)
return rewriter.notifyMatchFailure(op,
"ShuffleOp mask not interleaving");
}

rewriter.replaceOpWithNewOp<InterleaveOp>(op, op.getV1(), op.getV2());
return success();
}
};

} // namespace

void ShuffleOp::getCanonicalizationPatterns(RewritePatternSet &results,
MLIRContext *context) {
results.add<ShuffleSplat, Canonicalize0DShuffleOp>(context);
results.add<ShuffleSplat, ShuffleInterleave, Canonicalize0DShuffleOp>(
context);
}

//===----------------------------------------------------------------------===//
Expand Down
23 changes: 23 additions & 0 deletions mlir/test/Dialect/Vector/canonicalize.mlir
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -2567,3 +2567,26 @@ func.func @load_store_forwarding_rank_mismatch(%v0: vector<4x1x1xf32>, %arg0: te
tensor<4x4x4xf32>, vector<1x100x4x5xf32>
return %r : vector<1x100x4x5xf32>
}

// -----

// CHECK-LABEL: func.func @rank_0_shuffle_to_interleave(
// CHECK-SAME: %[[LHS:.*]]: vector<f64>, %[[RHS:.*]]: vector<f64>)
func.func @rank_0_shuffle_to_interleave(%arg0: vector<f64>, %arg1: vector<f64>) -> vector<2xf64>
{
// CHECK: %[[ZIP:.*]] = vector.interleave %[[LHS]], %[[RHS]] : vector<f64>
// CHECK: return %[[ZIP]]
%0 = vector.shuffle %arg0, %arg1 [0, 1] : vector<f64>, vector<f64>
return %0 : vector<2xf64>
}

// -----

// CHECK-LABEL: func.func @rank_1_shuffle_to_interleave(
// CHECK-SAME: %[[LHS:.*]]: vector<6xi32>, %[[RHS:.*]]: vector<6xi32>)
func.func @rank_1_shuffle_to_interleave(%arg0: vector<6xi32>, %arg1: vector<6xi32>) -> vector<12xi32> {
// CHECK: %[[ZIP:.*]] = vector.interleave %[[LHS]], %[[RHS]] : vector<6xi32>
// CHECK: return %[[ZIP]]
%0 = vector.shuffle %arg0, %arg1 [0, 6, 1, 7, 2, 8, 3, 9, 4, 10, 5, 11] : vector<6xi32>, vector<6xi32>
return %0 : vector<12xi32>
}