Skip to content

[mlir][vector] Use vector.broadcast in place of vector.splat #148028

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

Open
wants to merge 1 commit into
base: main
Choose a base branch
from
Open

[mlir][vector] Use vector.broadcast in place of vector.splat #148028

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
1 commit
Select commit
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
2 changes: 1 addition & 1 deletion mlir/lib/Dialect/Arith/Transforms/EmulateUnsupportedFloats.cpp
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -123,7 +123,7 @@ void mlir::arith::populateEmulateUnsupportedFloatsLegality(
vector::FMAOp, vector::OuterProductOp, vector::MatmulOp, vector::ScanOp>(
[&](Operation *op) { return converter.isLegal(op); });
target.addLegalOp<arith::BitcastOp, arith::ExtFOp, arith::TruncFOp,
arith::ConstantOp, vector::SplatOp>();
arith::ConstantOp, vector::SplatOp, vector::BroadcastOp>();
}

void EmulateUnsupportedFloatsPass::runOnOperation() {
Expand Down
2 changes: 1 addition & 1 deletion mlir/lib/Dialect/NVGPU/TransformOps/NVGPUTransformOps.cpp
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -688,7 +688,7 @@ Value MmaSyncBuilder::buildMmaSyncMemRefLoadOperand(

Type elementType = getElementTypeOrSelf(memref.getType());
auto vt = VectorType::get(vectorShape, elementType);
Value res = b.create<vector::SplatOp>(loc, vt, loads[0]);
Value res = b.create<vector::BroadcastOp>(loc, vt, loads[0]);
foreachIndividualVectorElement(
res,
/*applyFn=*/
Expand Down
3 changes: 2 additions & 1 deletion mlir/lib/Dialect/XeGPU/Transforms/XeGPUUnroll.cpp
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -438,7 +438,8 @@ struct UnrollCreateDescOp : public UnrollPattern<xegpu::CreateDescOp> {
// Compute the offset
Value inc = rewriter.create<arith::ConstantIndexOp>(
loc, i * blockedChunkSize);
Value incVec = rewriter.create<vector::SplatOp>(loc, indiceType, inc);
Value incVec =
rewriter.create<vector::BroadcastOp>(loc, indiceType, inc);
Value offsetIndice =
rewriter.create<arith::AddIOp>(loc, indice, incVec);

Expand Down
6 changes: 3 additions & 3 deletions mlir/test/Dialect/NVGPU/transform-matmul-to-nvvm.mlir
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -20,14 +20,14 @@ func.func @matmul_16x8x4xf32_global(
// CHECK: %[[VAL_7:.*]] = affine.apply #[[$div4p8]]()[%[[TIDX]]]
// CHECK: %[[VAL_8:.*]] = affine.apply #[[$mod4]]()[%[[TIDX]]]
// CHECK: %[[VAL_9:.*]] = memref.load %[[VAL_0]][%[[VAL_7]], %[[VAL_8]]] : memref<16x4xf32>
// CHECK: %[[VAL_10:.*]] = vector.splat %[[VAL_6]] : vector<2x1xf32>
// CHECK: %[[VAL_10:.*]] = vector.broadcast %[[VAL_6]] : f32 to vector<2x1xf32>
// CHECK: %[[VAL_11:.*]] = vector.insert %[[VAL_6]], %[[VAL_10]] [0, 0] : f32 into vector<2x1xf32>
// CHECK: %[[LHS:.*]] = vector.insert %[[VAL_9]], %[[VAL_11]] [1, 0] : f32 into vector<2x1xf32>
//
// CHECK: %[[VAL_13:.*]] = affine.apply #[[$mod4]]()[%[[TIDX]]]
// CHECK: %[[VAL_14:.*]] = affine.apply #[[$div4]]()[%[[TIDX]]]
// CHECK: %[[VAL_15:.*]] = memref.load %[[VAL_1]][%[[VAL_13]], %[[VAL_14]]] : memref<4x8xf32>
// CHECK: %[[VAL_16:.*]] = vector.splat %[[VAL_15]] : vector<1x1xf32>
// CHECK: %[[VAL_16:.*]] = vector.broadcast %[[VAL_15]] : f32 to vector<1x1xf32>
// CHECK: %[[RHS:.*]] = vector.insert %[[VAL_15]], %[[VAL_16]] [0, 0] : f32 into vector<1x1xf32>
//
// CHECK: %[[VAL_18:.*]] = affine.apply #[[$div4]]()[%[[TIDX]]]
Expand All @@ -42,7 +42,7 @@ func.func @matmul_16x8x4xf32_global(
// CHECK: %[[VAL_27:.*]] = affine.apply #[[$div4p8]]()[%[[TIDX]]]
// CHECK: %[[VAL_28:.*]] = affine.apply #[[$map4]]()[%[[TIDX]]]
// CHECK: %[[VAL_29:.*]] = memref.load %[[VAL_2]][%[[VAL_27]], %[[VAL_28]]] : memref<16x8xf32>
// CHECK: %[[VAL_30:.*]] = vector.splat %[[VAL_20]] : vector<2x2xf32>
// CHECK: %[[VAL_30:.*]] = vector.broadcast %[[VAL_20]] : f32 to vector<2x2xf32>
// CHECK: %[[VAL_31:.*]] = vector.insert %[[VAL_20]], %[[VAL_30]] [0, 0] : f32 into vector<2x2xf32>
// CHECK: %[[VAL_32:.*]] = vector.insert %[[VAL_23]], %[[VAL_31]] [0, 1] : f32 into vector<2x2xf32>
// CHECK: %[[VAL_33:.*]] = vector.insert %[[VAL_26]], %[[VAL_32]] [1, 0] : f32 into vector<2x2xf32>
Expand Down
Loading