llvm
/
torch-mlir
mirror of https://github.com/llvm/torch-mlir
1
0
Fork 0
Code Issues Packages Projects Releases Wiki Activity

Update to LLVM 3157f03a349cfc852cdd994675eaa9652caa2e3a (#2060) 

New requirement to explicitly cast for interfaces https://reviews.llvm.org/D148493
pull/2063/head snapshot-20230426.820
Eric Kunze 2023-04-25 08:52:46 -07:00 committed by GitHub
parent 6b19c5e582
commit 6a833e1922
No known key found for this signature in database
GPG Key ID: 4AEE18F83AFDEB23
9 changed files with 14 additions and 14 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

2
externals/llvm-external-projects/torch-mlir-dialects/lib/Dialect/TMTensor/Transforms/Bufferize.cpp vendored
View File

@ -87,7 +87,7 @@ static TMTensorOp createTMTensorOpOnBuffers(ConversionPatternRewriter &rewriter,
ValueRange outputs) {
SmallVector<Value, 8> newOperands = inputs;
newOperands.append(outputs.begin(), outputs.end());
return tmtensorOp.clone(rewriter, tmtensorOp->getLoc(), {}, newOperands);
return cast<TMTensorOp>(tmtensorOp.clone(rewriter, tmtensorOp->getLoc(), {}, newOperands));
}
/// Generic conversion pattern that matches any TMTensorOp. This avoids template

2
externals/llvm-project vendored

@ -1 +1 @@
Subproject commit 3abae1c88416858cf2e9a7ed9417bc52033933b4
Subproject commit 3157f03a349cfc852cdd994675eaa9652caa2e3a

2
externals/mlir-hlo vendored

@ -1 +1 @@
Subproject commit 70ebc73a28767b772e09e0dd03cf3e47457435a2
Subproject commit a9ac343738945be7744e304b72704128e48aa06f

4
lib/Conversion/TorchToLinalg/Pooling.cpp
View File

@ -80,7 +80,7 @@ static LogicalResult createPoolingOp(
highPaddingIncludingNC[2] += strideInts[0];
highPaddingIncludingNC[3] += strideInts[1];
}
Value initValue = rewriter.create<arith::ConstantOp>(loc, initValueAttr);
Value initValue = rewriter.create<arith::ConstantOp>(loc, cast<TypedAttr>(initValueAttr));
paddedInput = torch_to_linalg::getPaddedTensor(
op, rewriter, self, lowPaddingIncludingNC, highPaddingIncludingNC,
initValue);
@ -154,7 +154,7 @@ public:
return rewriter.notifyMatchFailure(op, "invalid pooling parameters");
Type elementType = self.getType().cast<RankedTensorType>().getElementType();
auto smallestFPValueAttr = rewriter.getFloatAttr(
TypedAttr smallestFPValueAttr = rewriter.getFloatAttr(
elementType,
APFloat::getLargest(
elementType.cast<mlir::FloatType>().getFloatSemantics(),

4
lib/Conversion/TorchToLinalg/Reduction.cpp
View File

@ -277,7 +277,7 @@ static Value createLinalgPayloadForReduceOp(OpBuilder &b, Location loc,
Value result = payloadArgs[1];
Value self = convertScalarToDtype(b, loc, elem, resultElementType);
auto abs = b.create<math::AbsFOp>(loc, self);
Attribute twoAttr = b.getFloatAttr(resultElementType, 2.0);
TypedAttr twoAttr = b.getFloatAttr(resultElementType, 2.0);
auto ord = b.create<arith::ConstantOp>(loc, twoAttr);
auto pow = b.create<math::PowFOp>(loc, abs, ord);
return b.create<arith::AddFOp>(loc, pow, result);
@ -403,7 +403,7 @@ private:
return rewriter.notifyMatchFailure(op, "unimplemented: ord = +/- inf");
// Raise each summed value to the inverse of the order of the norm.
Attribute oneAttr = rewriter.getFloatAttr(elemType, 1.0);
TypedAttr oneAttr = rewriter.getFloatAttr(elemType, 1.0);
auto oneValue = rewriter.create<arith::ConstantOp>(loc, oneAttr);
auto inverseOrdValue =
rewriter.create<arith::DivFOp>(loc, oneValue, ordValue);

8
lib/Conversion/TorchToTMTensor/TorchToTMTensor.cpp
View File

@ -56,7 +56,7 @@ using namespace mlir::torch::TMTensor;
// that these patterns become mostly mechanical associations of
// "aten.foo -> linalg.foo".
static Attribute getNumericLimit(PatternRewriter &rewriter, Type elementType,
static TypedAttr getNumericLimit(PatternRewriter &rewriter, Type elementType,
bool getMin = true) {
auto bitWidth = elementType.getIntOrFloatBitWidth();
if (llvm::isa<mlir::IntegerType>(elementType)) {
@ -1184,7 +1184,7 @@ public:
if (reduceEnum == torch_upstream::ReductionType::MEAN) {
SmallVector<Value> selfShape =
getTensorSizes(rewriter, loc, adaptor.getSelf());
Attribute initAttr;
TypedAttr initAttr;
if (llvm::isa<mlir::FloatType>(srcType.getElementType())) {
initAttr = rewriter.getFloatAttr(srcType.getElementType(), 1);
} else if (llvm::isa<mlir::IntegerType>(srcType.getElementType())) {
@ -1220,13 +1220,13 @@ public:
} else if (reduceEnum == torch_upstream::ReductionType::MAX) {
// Set the values in the input tensor to the smallest element of that
// type
auto minAttr = getNumericLimit(rewriter, srcType.getElementType(),
TypedAttr minAttr = getNumericLimit(rewriter, srcType.getElementType(),
/*getMin=*/true);
normalizationValue = rewriter.create<arith::ConstantOp>(loc, minAttr);
} else if (reduceEnum == torch_upstream::ReductionType::MIN) {
// Set the values in the input tensor to the largest element of that
// type
auto maxAttr = getNumericLimit(rewriter, srcType.getElementType(),
TypedAttr maxAttr = getNumericLimit(rewriter, srcType.getElementType(),
/*getMin=*/false);
normalizationValue = rewriter.create<arith::ConstantOp>(loc, maxAttr);
}

2
lib/Conversion/Utils/Utils.cpp
View File

@ -197,7 +197,7 @@ Value getTensorSize(OpBuilder &b, Location loc, Value tensor) {
// Creates a constant of type `elemType` with value `val`.
Value getConstant(OpBuilder &b, Location loc, int64_t val, Type elemType) {
Attribute attr = {};
TypedAttr attr = {};
if (elemType.isa<mlir::FloatType>())
attr = b.getFloatAttr(elemType, val);
if (elemType.isa<mlir::IndexType>())

2
lib/Dialect/TorchConversion/IR/TorchConversionDialect.cpp
View File

@ -73,5 +73,5 @@ Operation *TorchConversionDialect::materializeConstant(OpBuilder &builder,
value.cast<IntegerAttr>());
}
return builder.create<arith::ConstantOp>(loc, value, type);
return arith::ConstantOp::materialize(builder, value, type, loc);
}

2
lib/Dialect/TorchConversion/Transforms/VerifyLinalgOnTensorsBackendContract.cpp
View File

@ -77,7 +77,7 @@ class VerifyLinalgOnTensorsBackendContractPass
// Tensor operations should go through linalg and the tensor dialect.
target.addDynamicallyLegalDialect<linalg::LinalgDialect>(opHasLegalTypes);
target.addDynamicallyLegalDialect<tensor::TensorDialect>(opHasLegalTypes);
target.addDynamicallyLegalDialect<AffineDialect>(opHasLegalTypes);
target.addDynamicallyLegalDialect<affine::AffineDialect>(opHasLegalTypes);
target.addDynamicallyLegalDialect<cf::ControlFlowDialect>(opHasLegalTypes);
target.addDynamicallyLegalDialect<TMTensorDialect>(opHasLegalTypes);
target.addDynamicallyLegalDialect<scf::SCFDialect>(opHasLegalTypes);