QLinearPackWeightInt8Onednn Class — pytorch Architecture

Source Code

aten/src/ATen/native/quantized/cpu/qlinear_prepack.cpp lines 681–694

class QLinearPackWeightInt8Onednn final {
 public:
  static at::Tensor run(
    // NOLINTNEXTLINE(performance-unnecessary-value-param)
    [[maybe_unused]] at::Tensor weight, // Not QTensor
    // NOLINTNEXTLINE(performance-unnecessary-value-param)
    [[maybe_unused]] std::optional<torch::List<int64_t>> input_shape) {
#if AT_MKLDNN_ENABLED()
    return pack_weight_to_onednn_tensor(weight, input_shape);
#else
    TORCH_CHECK(false, "Unimplemented as onednn is not available.");
#endif
  }
};