QLinearPackWeightInt8 Class — pytorch Architecture

Source Code

aten/src/ATen/native/quantized/cpu/qlinear_prepack.cpp lines 585–614

class QLinearPackWeightInt8 final {
 public:
  static c10::intrusive_ptr<LinearPackedParamsBase> run(
      at::Tensor weight,
      std::optional<Tensor> bias) {
    auto& ctx = at::globalContext();

#ifdef USE_FBGEMM
    if (ctx.qEngine() == at::QEngine::FBGEMM ||
        ctx.qEngine() == at::QEngine::X86) {
      return PackedLinearWeight::prepack(std::move(weight), std::move(bias));
    }
#endif
#ifdef USE_PYTORCH_QNNPACK
    if (ctx.qEngine() == at::QEngine::QNNPACK) {
      return PackedLinearWeightsQnnp::prepack(
          std::move(weight), std::move(bias));
    }
#endif
#if AT_MKLDNN_ENABLED()
    if (ctx.qEngine() == at::QEngine::ONEDNN) {
      return PackedLinearWeightsOnednn::prepack(std::move(weight), std::move(bias));
    }
#endif // #if AT_MKLDNN_ENABLED()
    TORCH_CHECK(
        false,
        "Didn't find engine for operation quantized::linear_prepack ",
        toString(ctx.qEngine()));
  }
};