kSpatialDim Class — pytorch Architecture

Source Code

aten/src/ATen/native/quantized/cpu/qconv_dynamic.cpp lines 25–63

template <int kSpatialDim>
at::Tensor PackedConvWeight<kSpatialDim>::apply_dynamic(
    const at::Tensor& input,
    bool reduce_range) {
  TORCH_CHECK(
      fbgemm::fbgemmSupportedCPU(), "Your CPU does not support FBGEMM.");

  float x_min, x_max;
  fbgemm::FindMinMax(
      /*m=*/input.data_ptr<float>(),
      /*min=*/&x_min,
      /*max=*/&x_max,
      /*len=*/input.numel());

  // Input tensor is quantized as 8-bit unsigned values
  static constexpr int precision = 8;
  static constexpr bool is_signed = false;

  // Calculate scale and zero point for quantization of input tensor
  auto q_params = quant_utils::ChooseQuantizationParams(
      /*min=*/x_min,
      /*max=*/x_max,
      /*qmin=*/is_signed ? -(1 << (precision - 1)) : 0,
      /*qmax=*/
      is_signed ? ((1 << (precision - 1)) - 1) : (1 << precision) - 1,
      /*preserve_sparsity=*/false,
      /*force_scale_power_of_two=*/false,
      /*reduce_range=*/reduce_range);

  // Quantize input
  at::Tensor q_input = at::quantize_per_tensor(
      input, q_params.scale, q_params.zero_point, c10::kQUInt8);

  at::Tensor out =
      apply_impl<false>(q_input, q_params.scale, q_params.zero_point);

  return at::dequantize(out); // TODO: optimized kernel that outputs fp32 so
                              // this step isn't necessary
}