absl::StrCat("Converting ", debugString(type), " to DataType"));
    }
  } else if (auto complex_type = mlir::dyn_cast<mlir::ComplexType>(type)) {
    auto etype = complex_type.getElementType();
    if (etype.isF32()) {
      *dtype = DT_COMPLEX64;
      return absl::OkStatus();
    } else if (etype.isF64()) {
      *dtype = DT_COMPLEX128;
      return absl::OkStatus();
    }

def QuantizeByQuantizedType : NativeCodeCall<"quant::Quantize($0, $1.getValue())">;
def F32ElementsAttr : ElementsAttrBase<
  CPred<"$_self.cast<ElementsAttr>().getShapedType().getElementType().isF32()">, "float constant tensor">;

// Squash tfl.dequantize and tfl.quantize pairs.
// TODO(fengliuai): Compare the scale of input and output. This can also be
// squashed to a requantize op if the scales are different.

                .put(ModelActionRole.Discover, AddProjectionsAction.of(subject, descriptor,
                    ModelMapModelProjection.managed(schema.getType(), schema.getElementType(), ChildNodeInitializerStrategyAccessors.fromPrivateData())
                ))
                .put(ModelActionRole.Create, DirectNodeInputUsingModelAction.of(subject, descriptor,

    if (!output_type || !output_type.getElementType().isF32()) return failure();

    auto input_type = mlir::dyn_cast<ShapedType>(input_dequant.getType());
    // TODO(renjieliu): support UniformQuantizedPerAxisType.
    auto q_type = mlir::dyn_cast_or_null<quant::UniformQuantizedType>(
        input_type.getElementType());
    if (!q_type) return failure();

        }
        if (type.equals(Type.DOUBLE_TYPE)) {
            return ClassName.DOUBLE;
        }
        if (type.getSort() == Type.ARRAY) {
            return ArrayTypeName.of(typeName(type.getElementType()));
        }
        return className(type);
    }

    public static ClassName className(Type type) {
        return ClassName.bestGuess(type.getClassName().replace("$", "."));
    }

  output_shape[3] = composite_result_shape[1];

  auto input_type = mlir::cast<ShapedType>(old_op->getOperand(0).getType());

  return RankedTensorType::get(output_shape, input_type.getElementType());
}
}  // namespace odml

  // doesn't require quantization.
  auto tensor_type = mlir::dyn_cast<TensorType>(value.getType());
  if (!tensor_type) {
    // There are none type values.
    return;
  }
  if (!tensor_type.getElementType().isF32()) return;

  // If the result is consumed by a quantize op, it has been quantized.
  if (value.hasOneUse() &&
      llvm::isa<TFL::QuantizeOp>(*value.getUsers().begin()))
    return;

    if (!splat_elements_attr) {
      return;
    }
    if (splat_elements_attr.getNumElements() == 1) {
      return;
    }
    auto element_type = splat_elements_attr.getType().getElementType();
    if (mlir::isa<ComplexType>(element_type) ||
        mlir::isa<quant::QuantizedType>(element_type)) {
      return;
    }
    op_builder->setInsertionPoint(const_op);

        if (parameterInfo.getKind() == ParameterKindInfo.VARARG_METHOD_PARAMETER && renderVararg) {
            return CodeBlock.of("$T...", typeName(parameterInfo.getParameterType().getElementType()));
        }
        return CodeBlock.of("$T", typeName(parameterInfo.getParameterType()));
    }

                defaultArtifacts = null;
                changingValue.handle(previousArtifacts);
            }
        }

        @Override
        public Class<? extends PublishArtifact> getElementType() {
            return PublishArtifact.class;
        }

        @Override
        public int size() {
            if (artifacts == null) {
                return 0;
            }