Skip to content

[mlir][arith.constant]Fix element type of the dense attributes in target attributes to be consistent with result type in LLVM::detail::oneToOneRewrite() #149787

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 4 commits into
base: main
Choose a base branch
from
Open

[mlir][arith.constant]Fix element type of the dense attributes in target attributes to be consistent with result type in LLVM::detail::oneToOneRewrite() #149787

Show file tree
Hide file tree
Changes from 1 commit
Commits
Show all changes
4 commits
Select commit Hold shift + click to select a range
d4e6a3f
Fix element type of target attributes in oneToOneRewrite when convert…
Jul 21, 2025
891ecaa
Merge branch 'main' of github.com:MengmSun/llvm-project into mlir/fix…
Jul 30, 2025
38b4227
Spread elements type conversion to all valid type attr
Jul 30, 2025
0af167a
Merge branch 'main' into mlir/fix/arith_constant_target_attr
MengmSun Jul 30, 2025
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
27 changes: 26 additions & 1 deletion mlir/lib/Conversion/LLVMCommon/Pattern.cpp
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -331,10 +331,35 @@ LogicalResult LLVM::detail::oneToOneRewrite(
return failure();
}

// If the targetAttrs contains DenseElementsAttr,
// and the element type of the DenseElementsAttr and result type is
// inconsistent after the conversion of result types, we need to convert the
// element type of the DenseElementsAttr to the target type by creating a new
// DenseElementsAttr with the converted element type, and use the new
// DenseElementsAttr to replace the old one in the targetAttrs
SmallVector<NamedAttribute> convertedAttrs;
for (auto attr : targetAttrs) {
if (auto denseAttr = dyn_cast<DenseElementsAttr>(attr.getValue())) {
VectorType vectorType = dyn_cast<VectorType>(denseAttr.getType());
if (vectorType) {
auto convertedElementType =
typeConverter.convertType(vectorType.getElementType());
VectorType convertedVectorType =
VectorType::get(vectorType.getShape(), convertedElementType,
vectorType.getScalableDims());
convertedAttrs.emplace_back(
attr.getName(), DenseElementsAttr::getFromRawBuffer(
convertedVectorType, denseAttr.getRawData()));
}
} else {
convertedAttrs.push_back(attr);
Copy link
Contributor

@krzysz00 krzysz00 Jul 21, 2025

Choose a reason for hiding this comment

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

Shouldn't this also be done for scalar attributes? Ex. a index constant becoming an i64 constant?

}
}

// Create the operation through state since we don't know its C++ type.
Operation *newOp =
rewriter.create(op->getLoc(), rewriter.getStringAttr(targetOp), operands,
resultTypes, targetAttrs);
resultTypes, convertedAttrs);

setNativeProperties(newOp, overflowFlags);

Expand Down
17 changes: 16 additions & 1 deletion mlir/test/Conversion/ArithToLLVM/arith-to-llvm.mlir
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -428,7 +428,7 @@ func.func @fcmp(f32, f32) -> () {

// CHECK-LABEL: @index_vector
func.func @index_vector(%arg0: vector<4xindex>) {
// CHECK: %[[CST:.*]] = llvm.mlir.constant(dense<[0, 1, 2, 3]> : vector<4xindex>) : vector<4xi64>
// CHECK: %[[CST:.*]] = llvm.mlir.constant(dense<[0, 1, 2, 3]> : vector<4xi64>) : vector<4xi64>
%0 = arith.constant dense<[0, 1, 2, 3]> : vector<4xindex>
// CHECK: %[[V:.*]] = llvm.add %{{.*}}, %[[CST]] : vector<4xi64>
%1 = arith.addi %arg0, %0 : vector<4xindex>
Expand All @@ -437,6 +437,21 @@ func.func @index_vector(%arg0: vector<4xindex>) {

// -----

// CHECK-LABEL: @f8E4M3FN_vector
func.func @f8E4M3FN_vector() -> vector<4xf8E4M3FN> {
// CHECK: %[[CST0:.*]] = llvm.mlir.constant(dense<0> : vector<4xi8>) : vector<4xi8>
%0 = arith.constant dense<0.000000e+00> : vector<4xf8E4M3FN>
// CHECK: %[[CST1:.*]] = llvm.mlir.constant(dense<[56, 64, 68, 72]> : vector<4xi8>) : vector<4xi8>
%1 = arith.constant dense<[1.0, 2.0, 3.0, 4.0]> : vector<4xf8E4M3FN>
// CHECK: %[[V:.*]] = llvm.mlir.constant(dense<[56, 64, 68, 72]> : vector<4xi8>) : vector<4xi8>
%2 = arith.addf %0, %1 : vector<4xf8E4M3FN>
// CHECK: %[[RES:.*]] = builtin.unrealized_conversion_cast %[[V]] : vector<4xi8> to vector<4xf8E4M3FN>
// CHECK-NEXT: return %[[RES]] : vector<4xf8E4M3FN>
func.return %2 : vector<4xf8E4M3FN>
}

// -----

// CHECK-LABEL: @bitcast_1d
func.func @bitcast_1d(%arg0: vector<2xf32>) {
// CHECK: llvm.bitcast %{{.*}} : vector<2xf32> to vector<2xi32>
Expand Down
Loading