input_ = fused_func_op_.getArgument(0);
  bias_ = fused_func_op_.getArgument(2);

  weight_ = fused_func_op_.getArgument(1);
  weight_type_ = mlir::cast<RankedTensorType>(weight_.getType());

  if (weight_type_.getRank() != 2) {
    return fused_func_op_.emitError() << "The weight tensor was not of rank 2";
  }

  if (weight_type_.getDimSize(1) % num_gates_ != 0) {

                         METHOD_STATIC_RANGE_WEIGHT_ONLY_INT8,
                     "weight_only", "Post-training weight-only quantizaiton"))};

  Option<bool> enable_per_channel_quantization_{
      *this, "enable-per-channel-quantization", llvm::cl::init(false),
      llvm::cl::desc("Whether enable per-channel quantized weights.")};
};

// Apply constant transformations for the op_set.

		ret += edge.Weight
	}
	return ret
}

type edgeList []*Edge

func (el edgeList) Len() int {
	return len(el)
}

func (el edgeList) Less(i, j int) bool {
	if el[i].Weight != el[j].Weight {
		return abs64(el[i].Weight) > abs64(el[j].Weight)
	}

	from1 := el[i].Src.Info.PrintableName()
	from2 := el[j].Src.Info.PrintableName()
	if from1 != from2 {
		return from1 < from2
	}

    $constant_operand_indices are the indices of the inputs that are constant to the TPU program (e.g. weights in inference), the rest of the inputs are input tensors.
    constant_operand_indices is sorted in ascending order.
    $operands_with_static_shape are indices of operands that are tagged with a maximum static shape.

      enable_per_channel_quantized_weight_;
  // Change this to user-given bit width once we have custom configuration.
  options.bit_width_ = 8;

  // Insert quantization parameters for weights for ops with `weight_only_ptq`
  // attribute.
  pm.addNestedPass<func::FuncOp>(createInsertWeightParamPass());

  // PrepareQuantizePass uses SymbolTable to fetch relevant GEMM ops for

              input.getDefiningOp())) {
        // Tensors with derived scale are biases, and handled in propagation.
        if (tensor_property.use_derived_scale) continue;
        // For weights, use quantization scale inferred from the values.
        if (failed(processConstantOp(op, input.getDefiningOp(), index,
                                     tensor_property, rewriter))) {
          return failure();

                     "Uses TF Uniform Quantized ops"))};

  Option<int64_t> min_num_elements_for_weights_{
      *this, "min-num-elements-for-weights", llvm::cl::init(0),
      llvm::cl::desc("The minimum required number of elements in a weight "
                     "array to apply quantization.")};

  Option<QuantMethod> quantization_method_{
      *this, "quantization-method",

		},
		// Consider Affinity, Anti Affinity and symmetry together.
		// for Affinity, the weights are:                8,  0,  0,  0
		// for Anti Affinity, the weights are:           0, -5,  0,  0
		// for Affinity symmetry, the weights are:       0,  0,  8,  0
		// for Anti Affinity symmetry, the weights are:  0,  0,  0, -5
		{

    LOG(INFO) << quantized_tensor->name()->str() << " "
              << float_tensor->name()->str();
    if (ExpectEqualTensor(quantized_tensor, float_tensor)) {
      if (quantized && quantized_tensor->name()->str().find("weights")) {
        // If tensor is quantized, data type and buffer contents can be
        // different between float and quantized tensors. So do those tests
        // separately in the test body without checking them here.

    $constant_operand_indices are the indices of the inputs that are constant to the TPU program (e.g. weights in inference), the rest of the inputs are input tensors.
    constant_operand_indices is sorted in ascending order.
    $operands_with_static_shape are indices of operands that are tagged with a maximum static shape.