Decompose aten.fmod into aten.mul,sub,div etc.

[srinathava]

As titled, create a new decomposition for aten.fmod.Tensor to aten.div, aten.trunc, aten.mul and aten.sub. Note that we only use aten.trunc for floating point operations. This further gets decomposed to aten.where etc. by other existing decompositions.

This decomposition now makes TOSA pass for a simple model with aten.fmod while it makes stablehlo fail. For now, we disallow this decomposition for stablehlo

[sjain-stanford]

LGTM

[vivekkhandelwal1]

@srinathava Since this commit adds the decomposition for the AtenFmodTensorOp, the existing lowering for the same should be removed from here:

torch-mlir/lib/Conversion/TorchToLinalg/Uncategorized.cpp

Lines 1285 to 1307 in 0474082

	if (auto fmod = dyn_cast<AtenFmodTensorOp>(op)) {
	Type newResultType =
	cast<RankedTensorType>(converter->convertType(fmod.getType()))
	.getElementType();
	Value self = convertScalarToDtype(b, loc, payloadArgs[0], newResultType);
	Value other = convertScalarToDtype(b, loc, payloadArgs[1], newResultType);
	Value result;
	if (isa<mlir::FloatType>(newResultType)) {
	Value n = b.create<arith::DivFOp>(loc, self, other);
	n = b.create<math::TruncOp>(loc, n);
	Value n_y = b.create<arith::MulFOp>(loc, n, other);
	result = b.create<arith::SubFOp>(loc, self, n_y);
	} else if (isa<mlir::IntegerType>(newResultType)) {
	Value n = b.create<arith::DivSIOp>(loc, self, other);
	Value n_y = b.create<arith::MulIOp>(loc, n, other);
	result = b.create<arith::SubIOp>(loc, self, n_y);
	} else {
	fmod.emitError("Unsupported type encountered for AtenFmodTensorOp.");
	}
	return result;
	}

[srinathava]

@srinathava Since this commit adds the decomposition for the AtenFmodTensorOp, the existing lowering for the same should be removed from here:

torch-mlir/lib/Conversion/TorchToLinalg/Uncategorized.cpp

Lines 1285 to 1307 in 0474082

	if (auto fmod = dyn_cast<AtenFmodTensorOp>(op)) {
	Type newResultType =
	cast<RankedTensorType>(converter->convertType(fmod.getType()))
	.getElementType();
	Value self = convertScalarToDtype(b, loc, payloadArgs[0], newResultType);
	Value other = convertScalarToDtype(b, loc, payloadArgs[1], newResultType);
	Value result;
	if (isa<mlir::FloatType>(newResultType)) {
	Value n = b.create<arith::DivFOp>(loc, self, other);
	n = b.create<math::TruncOp>(loc, n);
	Value n_y = b.create<arith::MulFOp>(loc, n, other);
	result = b.create<arith::SubFOp>(loc, self, n_y);
	} else if (isa<mlir::IntegerType>(newResultType)) {
	Value n = b.create<arith::DivSIOp>(loc, self, other);
	Value n_y = b.create<arith::MulIOp>(loc, n, other);
	result = b.create<arith::SubIOp>(loc, self, n_y);
	} else {
	fmod.emitError("Unsupported type encountered for AtenFmodTensorOp.");
	}
	return result;
	}

@vivekkhandelwal1, thanks for pointing that out. I'll send out a follow-up PR for the cleanup shortly.