Missing UInt support fo getScalarTypeForType

[renxida]

torch-mlir/lib/Dialect/Torch/Utils/Utils.cpp

Lines 47 to 100 in b3942ff

	torch_upstream::ScalarType Torch::getScalarTypeForType(Type type) {
	if (isa<Float32Type>(type))
	return torch_upstream::ScalarType::Float;
	if (isa<Float64Type>(type))
	return torch_upstream::ScalarType::Double;
	if (type.isSignedInteger(64))
	return torch_upstream::ScalarType::Long;
	if (type.isSignedInteger(32))
	return torch_upstream::ScalarType::Int;
	if (type.isSignedInteger(16))
	return torch_upstream::ScalarType::Short;
	if (type.isSignlessInteger(1))
	return torch_upstream::ScalarType::Bool;
	if (type.isBF16())
	return torch_upstream::ScalarType::BFloat16;
	if (type.isF16())
	return torch_upstream::ScalarType::Half;
	if (type.isUnsignedInteger(8))
	return torch_upstream::ScalarType::Byte;
	if (type.isSignedInteger(8))
	return torch_upstream::ScalarType::Char;
	if (isa<QUInt8Type>(type))
	return torch_upstream::ScalarType::QUInt8;
	if (isa<QInt8Type>(type))
	return torch_upstream::ScalarType::QInt8;
	if (isa<QInt16Type>(type))
	return torch_upstream::ScalarType::QInt16;
	if (isa<QInt32Type>(type))
	return torch_upstream::ScalarType::QInt32;
	if (isa<ComplexType>(type)) {
	mlir::Type complexElemType = cast<ComplexType>(type).getElementType();
	if (complexElemType.isF16())
	return torch_upstream::ScalarType::ComplexHalf;
	if (complexElemType.isF32())
	return torch_upstream::ScalarType::ComplexFloat;
	if (complexElemType.isF64())
	return torch_upstream::ScalarType::ComplexDouble;
	}
	if (isa<Float8E5M2Type>(type))
	return torch_upstream::ScalarType::Float8_e5m2;
	if (isa<Float8E4M3FNType>(type))
	return torch_upstream::ScalarType::Float8_e4m3fn;
	if (isa<Float8E5M2FNUZType>(type))
	return torch_upstream::ScalarType::Float8_e5m2fnuz;
	if (isa<Float8E4M3FNUZType>(type))
	return torch_upstream::ScalarType::Float8_e4m3fnuz;
	llvm::report_fatal_error("unhandled type for getScalarTypeForType");
	}
	Type Torch::getTypeForTorchType(
	MLIRContext *context, Type type,
	mlir::IntegerType::SignednessSemantics signedness) {
	if (isa<Torch::IntType>(type))
	return IntegerType::get(context, 64, signedness);
	if (isa<Torch::FloatType>(type))

As seen here, we don't have support for getScalarTypeForType. Also, the error message not very informative: it just says unhandled type for getScalarTypeForType without specifying what the unsupported type is.

Implementing this also requires modifying this ScalarType enum in TorchUpstream.h:

torch-mlir/include/torch-mlir/Dialect/Torch/Utils/TorchUpstream.h

Lines 88 to 98 in b3942ff

	#define AT_FORALL_SCALAR_TYPES_WITH_COMPLEX_AND_QINTS(_) \
	_(uint8_t, Byte) /* 0 */ \
	_(int8_t, Char) /* 1 */ \
	_(int16_t, Short) /* 2 */ \
	_(int, Int) /* 3 */ \
	_(int64_t, Long) /* 4 */ \
	_(at::Half, Half) /* 5 */ \
	_(float, Float) /* 6 */ \
	_(double, Double) /* 7 */ \
	_(c10::complex<c10::Half>, ComplexHalf) /* 8 */ \
	_(c10::complex<float>, ComplexFloat) /* 9 */ \

which is not up to date compared to the ScalarType.h in the pytorch/pytorch repo:
https://github.com/pytorch/pytorch/blob/main/c10/core/ScalarType.h#L84-L93

Related issue where this prevents supporting certain onnx dtypes: #3255