This is an archive of the discontinued LLVM Phabricator instance.

Differential D148095

[mlir][Linalg] Allow linalg.copy to be vectorized with masking
ClosedPublic

Authored by nicolasvasilache on Apr 12 2023, 2:34 AM.

Details

Reviewers
aartbik
dcaballe
qcolombet
ftynse
Commits
rG1edfb4b93d33: [mlir][Linalg] Allow linalg.copy to be vectorized with masking

Diff Detail

Event Timeline

nicolasvasilache created this revision.Apr 12 2023, 2:34 AM
Herald added a reviewer: aartbik. · View Herald TranscriptApr 12 2023, 2:34 AM
Herald added a project: Restricted Project. · View Herald Transcript
Herald added subscribers: bviyer, hanchung, Moerafaat and 26 others. · View Herald Transcript
nicolasvasilache requested review of this revision.Apr 12 2023, 2:34 AM
Herald added a reviewer: dcaballe. · View Herald TranscriptApr 12 2023, 2:34 AM
Herald added a project: Restricted Project. · View Herald Transcript
Herald added subscribers: limo1996, stephenneuendorffer. · View Herald Transcript
nicolasvasilache updated this revision to Diff 512729.Apr 12 2023, 2:45 AM

Add test

Herald added a subscriber: • pcwang-thead. · View Herald TranscriptApr 12 2023, 2:45 AM
nicolasvasilache added a reviewer: qcolombet.Apr 12 2023, 2:45 AM
nicolasvasilache updated this revision to Diff 512730.Apr 12 2023, 2:48 AM

Rebase

Harbormaster completed remote builds in B224988: Diff 512730.Apr 12 2023, 3:34 AM
nicolasvasilache added a reviewer: ftynse.Apr 12 2023, 4:44 AM
qcolombet accepted this revision.Apr 12 2023, 4:58 AM
qcolombet added inline comments.
mlir/test/Dialect/Linalg/vectorization.mlir
2743

Could you check for the vector.create_mask?

This revision is now accepted and ready to land.Apr 12 2023, 4:58 AM
nicolasvasilache updated this revision to Diff 512790.Apr 12 2023, 5:27 AM

Check mask

This revision was landed with ongoing or failed builds.Apr 12 2023, 5:35 AM
Closed by commit rG1edfb4b93d33: [mlir][Linalg] Allow linalg.copy to be vectorized with masking (authored by nicolasvasilache). · Explain Why
This revision was automatically updated to reflect the committed changes.
nicolasvasilache added a commit: rG1edfb4b93d33: [mlir][Linalg] Allow linalg.copy to be vectorized with masking.
Harbormaster completed remote builds in B225040: Diff 512790.Apr 12 2023, 5:52 AM

Revision Contents

PathSize
mlir/
lib/
Dialect/
Linalg/
Transforms/
2 lines
test/
Dialect/
Linalg/
20 lines

Diff 512793

mlir/lib/Dialect/Linalg/Transforms/Vectorization.cpp

Show First 20 Lines • Show All 1,276 Lines • ▼ Show 20 Lines if (!reduceOp || !getCombinerOpKind(reduceOp)) {
return failure(); return failure();
} }
} }
return success(); return success();
} }
static LogicalResult vectorizeDynamicLinalgOpPrecondition(linalg::LinalgOp op) { static LogicalResult vectorizeDynamicLinalgOpPrecondition(linalg::LinalgOp op) {
// TODO: Masking only supports dynamic generic ops for now. // TODO: Masking only supports dynamic generic ops for now.
if (!isa<linalg::GenericOp, linalg::FillOp>(op)) if (!isa<linalg::GenericOp, linalg::FillOp, linalg::CopyOp>(op))
return failure(); return failure();
LDBG("Dynamically-shaped op meets vectorization pre-conditions\n"); LDBG("Dynamically-shaped op meets vectorization pre-conditions\n");
return success(); return success();
} }
LogicalResult LogicalResult
mlir::linalg::vectorizeLinalgOpPrecondition(LinalgOp linalgOp, mlir::linalg::vectorizeLinalgOpPrecondition(LinalgOp linalgOp,
▲ Show 20 Lines • Show All 1,577 Lines • Show Last 20 Lines

mlir/test/Dialect/Linalg/vectorization.mlir

Show First 20 Lines • Show All 2,731 Lines • ▼ Show 20 Lines
// CHECK: vector.mask %[[MASK]] { vector.transfer_write %[[BCAST]], {{.*}} {in_bounds = [true, true]} : vector<8x16xf32>, tensor<?x?xf32> } : vector<8x16xi1> // CHECK: vector.mask %[[MASK]] { vector.transfer_write %[[BCAST]], {{.*}} {in_bounds = [true, true]} : vector<8x16xf32>, tensor<?x?xf32> } : vector<8x16xi1>
transform.sequence failures(propagate) { transform.sequence failures(propagate) {
^bb1(%arg1: !pdl.operation): ^bb1(%arg1: !pdl.operation):
%0 = transform.structured.match ops{["linalg.fill"]} in %arg1 : (!pdl.operation) -> !pdl.operation %0 = transform.structured.match ops{["linalg.fill"]} in %arg1 : (!pdl.operation) -> !pdl.operation
transform.structured.masked_vectorize %0 vector_sizes [8, 16] transform.structured.masked_vectorize %0 vector_sizes [8, 16]
} }
// -----
// CHECK-LABEL: func @test_masked_vectorize_linalg_copy
func.func @test_masked_vectorize_linalg_copy(%A : memref<?x?xf32>, %B : memref<?x?xf32>) {
qcolombetUnsubmitted
Not Done
ReplyInline Actions

Could you check for the vector.create_mask?

qcolombet: Could you check for the `vector.create_mask`?
// CHECK: %[[c0:.*]] = arith.constant 0 : index
// CHECK: %[[d0:.*]] = memref.dim %{{.*}}, %[[c0]] : memref<?x?xf32>
// CHECK: %[[c1:.*]] = arith.constant 1 : index
// CHECK: %[[d1:.*]] = memref.dim %{{.*}}, %[[c1]] : memref<?x?xf32>
// CHECK: %[[mask:.*]] = vector.create_mask %[[d0]], %[[d1]] : vector<2x4xi1>
// CHECK: vector.mask %[[mask]] {{.*}} vector.transfer_read %{{.*}} {in_bounds = [true, true]} : memref<?x?xf32>, vector<2x4xf32> } : vector<2x4xi1> -> vector<2x4xf32>
// CHECK: vector.mask %[[mask]] {{.*}} vector.transfer_write %{{.*}} {in_bounds = [true, true]} : vector<2x4xf32>, memref<?x?xf32> } : vector<2x4xi1>
linalg.copy ins(%A : memref<?x?xf32>) outs(%B : memref<?x?xf32>)
return
}
transform.sequence failures(propagate) {
^bb1(%arg1: !pdl.operation):
%0 = transform.structured.match ops{["linalg.copy"]} in %arg1 : (!pdl.operation) -> !pdl.operation
transform.structured.masked_vectorize %0 vector_sizes [2, 4]
}