}

  LogicalResult matchAndRewrite(TF::PartitionedCallOp op,
                                PatternRewriter& rewriter) const override {
    const auto f_attr = mlir::dyn_cast<FlatSymbolRefAttr>(op.getFAttr());
    StringRef function_name = f_attr.getValue();
    if (!function_name.starts_with(kDequantizeFunctionName)) return failure();

    llvm::SmallVector<Operation*> users(op->getUsers().begin(),

    auto *new_ewise_op =
        rewriter.create(op->getLoc(), op->getName().getIdentifier(),
                        {tpose_arg1.getOperand(0), tpose_arg2.getOperand(0)},
                        new_out_type, op->getAttrs());

    // Apply original tranpose to output of ewise op.
    auto out_tpose_op = rewriter.create<TFL::TransposeOp>(
        new_ewise_op->getLoc(), op->getResult(0).getType(),

    if operation_id in operation_ids:
        message = (
            f"Duplicate Operation ID {operation_id} for function "
            + f"{route.endpoint.__name__}"
        )
        file_name = getattr(route.endpoint, "__globals__", {}).get("__file__")
        if file_name:
            message += f" at {file_name}"
        warnings.warn(message, stacklevel=1)
    operation_ids.add(operation_id)

    }
    OpBuilder builder(call);
    auto new_call = builder.create<CallOp>(
        call.getLoc(), info.decomposed_callee.getFunctionType().getResults(),
        new_operands, call->getAttrs());
    new_call->setAttr(
        "f", SymbolRefAttr::get(
                 builder.getContext(),
                 const_cast<func::FuncOp&>(info.decomposed_callee).getName()));

  EXPECT_TRUE(isa<TF::XlaCallModuleOp>(lifted_op));
  EXPECT_EQ(
      mlir::cast<StringAttr>(lifted_op->getAttr("_original_entry_function")),
      "composite_dot_general_fn_1");
  EXPECT_EQ(
      mlir::cast<ArrayAttr>(lifted_dot_general_op->getAttr("precision_config")),
      builder_.getArrayAttr(SmallVector<Attribute>(
          1, mlir::stablehlo::PrecisionAttr::get(

      devices.push_back(builder.getNamedAttr(name, metadata));
    }
  }

  op->setAttr(kDevicesAttr, builder.getDictionaryAttr(devices));
}

mlir::LogicalResult GetDevicesFromOp(mlir::Operation* op,
                                     mlir::TF::RuntimeDevices* devices) {
  auto devices_attr = op->getAttr(kDevicesAttr);
  if (!devices_attr) return mlir::success();

    if (auto device = res->getAttrOfType<StringAttr>("device")) {
      if (!device.getValue().empty())
        result_id->setAttr("device", device);
      else
        result_id->setAttr("device", compile_device_op);
    } else if (compile_device_op) {
      result_id->setAttr("device", compile_device_op);
    }
    res.getOutput().replaceAllUsesWith(compile_op->getResult(0));
  }

        rewriter.getDenseI64ArrayAttr(kOihwToHwioPermutation));

    // [b, 0, 1, f]x[0, 1, i, o]->[b, 0, 1, f]
    const auto new_dimension_nums = rewriter.getAttr<ConvDimensionNumbersAttr>(
        /*inputBatchDimension=*/0, /*inputFeatureDimension=*/3,
        /*inputSpatialDimensions=*/SmallVector<int64_t>{1, 2},
        /*kernelInputFeatureDimension=*/2, /*kernelOutputFeatureDimension=*/3,

  for (const auto& attr_def : op_reg_data->op_def.attr()) {
    registered_attrs.insert(attr_def.name());
  }
  // Attributes are not in the registered attributes set will be ignored.
  for (auto& attr : inst->getAttrs()) {
    if (registered_attrs.find(attr.getName()) == registered_attrs.end()) {
      attrs_to_ignore->insert(
          absl::string_view(attr.getName().data(), attr.getName().size()));
    }
  }

            << min_num_elements_for_weights_ << " elements).";
        call_op->removeAttr(kQuantTraitAttrName);
      }
    }

    StringRef function_name =
        mlir::cast<FlatSymbolRefAttr>(call_op.getFAttr()).getValue();
    if ((quantization_method_ == tensorflow::quantization::QuantizationMethod::
                                     METHOD_DYNAMIC_RANGE_INT8) &&
        (function_name.contains("batch_matmul") ||