quantize_val_arm Class — pytorch Architecture

Source Code

aten/src/ATen/native/quantized/AffineQuantizerBase.cpp lines 74–86

template <typename T>
T quantize_val_arm(
    const float scale,
    const int32_t zero_point,
    const float value) {
  constexpr int32_t qmin = std::numeric_limits<T>::min();
  constexpr int32_t qmax = std::numeric_limits<T>::max();
  float inv_scale = 1.0f / scale;
  auto r = zero_point + static_cast<int32_t>(std::nearbyint(value * inv_scale));
  r = std::max(r, qmin);
  r = std::min(r, qmax);
  return static_cast<T>(r);
}