aten/src/ATen/native/UpSampleLinear1d.cpp

// Adapted from interp.cpp from Caffe util by Pauline Luc
// Originally developed by George Papandreou

#include <ATen/ATen.h>
#include <ATen/NativeFunctions.h>
#include <ATen/native/UpSample.h>

namespace at {
namespace native {
namespace {

template <typename scalar_t>
static void upsample_linear1d_out_frame(
    scalar_t* odata,
    scalar_t* idata,
    int64_t input_width,
    int64_t output_width,
    int64_t nbatch,
    int64_t channels,
    bool align_corners,
    c10::optional<double> scales) {
  channels = channels * nbatch;

  // special case: just copy
  if (input_width == output_width) {
    for (int64_t w2 = 0; w2 < output_width; ++w2) {
      const int64_t w1 = w2;
      const scalar_t* pos1 = &idata[w1];
      scalar_t* pos2 = &odata[w2];

      for (int64_t c = 0; c < channels; ++c) {
        pos2[0] = pos1[0];
        pos1 += input_width;
        pos2 += output_width;
      }
    }
    return;
  }
  const scalar_t rwidth = area_pixel_compute_scale<scalar_t>(
      input_width, output_width, align_corners, scales);

  for (int64_t w2 = 0; w2 < output_width; ++w2) {
    const scalar_t w1r = area_pixel_compute_source_index<scalar_t>(
        rwidth, w2, align_corners, /*cubic=*/false);

    const int64_t w1 = w1r;
    const int64_t w1p = (w1 < input_width - 1) ? 1 : 0;
    const scalar_t w1lambda = w1r - w1;
    const scalar_t w0lambda = static_cast<scalar_t>(1.) - w1lambda;
    const scalar_t* pos1 = &idata[w1];
    // index w2 is interpolated by idata[w1] and (itself or idata[w1 + 1])
    scalar_t* pos2 = &odata[w2];

    for (int64_t c = 0; c < channels; ++c) {
      pos2[0] = w0lambda * pos1[0] + w1lambda * pos1[w1p];
      pos1 += input_width;
      pos2 += output_width;
    }
  }
}

template <typename scalar_t>
static void upsample_linear1d_backward_out_frame(
    scalar_t* odata,
    scalar_t* idata,
    int64_t input_width,
    int64_t output_width,
    int64_t nbatch,
    int64_t channels,
    bool align_corners,
    c10::optional<double> scales) {
  channels = nbatch * channels;

  // special case: same-size matching grids
  if (input_width == output_width) {
    for (int64_t w2 = 0; w2 < output_width; ++w2) {
      const int64_t w1 = w2;
      scalar_t* pos1 = &idata[w1];
      const scalar_t* pos2 = &odata[w2];

      for (int64_t c = 0; c < channels; ++c) {
        pos1[0] += pos2[0];
        pos1 += input_width;
        pos2 += output_width;
      }
    }
    return;
  }
  const scalar_t rwidth = area_pixel_compute_scale<scalar_t>(
      input_width, output_width, align_corners, scales);

  for (int64_t w2 = 0; w2 < output_width; ++w2) {
    const scalar_t w1r = area_pixel_compute_source_index<scalar_t>(
        rwidth, w2, align_corners, /*cubic=*/false);

    const int64_t w1 = w1r;
    const int64_t w1p = (w1 < input_width - 1) ? 1 : 0;
    const scalar_t w1lambda = w1r - w1;
    const scalar_t w0lambda = static_cast<scalar_t>(1.) - w1lambda;
    scalar_t* pos1 = &idata[w1];
    const scalar_t* pos2 = &odata[w2];

    for (int64_t c = 0; c < channels; ++c) {
      pos1[0] += w0lambda * pos2[0];
      pos1[w1p] += w1lambda * pos2[0];
      pos1 += input_width;
      pos2 += output_width;
    }
  }
}

static void upsample_linear1d_out_cpu_template(
    Tensor& output,
    const Tensor& input_,
    IntArrayRef output_size,
    bool align_corners,
    c10::optional<double> scales) {
  TORCH_CHECK(
      output_size.size() == 1,
      "It is expected output_size equals to 1, but got size ",
      output_size.size());

  int64_t output_width = output_size[0];

  int64_t nbatch = input_.size(0);
  int64_t channels = input_.size(1);
  int64_t input_width = input_.size(2);

  upsample_1d_shape_check(
      input_,
      Tensor(),
      nbatch,
      channels,
      input_width,
      output_width);

  auto input = input_.contiguous();

  output.resize_({nbatch, channels, output_width});
  output.zero_();

  AT_ASSERT(input_width > 0 && output_width > 0);

  AT_DISPATCH_FLOATING_TYPES_AND_HALF(input.scalar_type(), "upsample_linear1d", [&] {
    auto* idata = input.data_ptr<scalar_t>();
    auto* odata = output.data_ptr<scalar_t>();

    upsample_linear1d_out_frame<scalar_t>(
        odata,
        idata,
        input_width,
        output_width,
        nbatch,
        channels,
        align_corners,
        scales);
  });
}

static void upsample_linear1d_backward_out_cpu_template(
    Tensor& grad_input,
    const Tensor& grad_output_,
    IntArrayRef output_size,
    IntArrayRef input_size,
    bool align_corners,
    c10::optional<double> scales) {
  TORCH_CHECK(
      output_size.size() == 1,
      "It is expected output_size equals to 1, but got size ",
      output_size.size());

  TORCH_CHECK(
      input_size.size() == 3,
      "It is expected input_size equals to 3, but got size ",
      input_size.size());

  int64_t output_width = output_size[0];

  int64_t nbatch = input_size[0];
  int64_t channels = input_size[1];
  int64_t input_width = input_size[2];

  upsample_1d_shape_check(
      Tensor(),
      grad_output_,
      nbatch,
      channels,
      input_width,
      output_width);

  auto grad_output = grad_output_.contiguous();

  grad_input.resize_({nbatch, channels, input_width});
  grad_input.zero_();

  AT_DISPATCH_FLOATING_TYPES_AND_HALF(
      grad_output.scalar_type(), "upsample_linear1d_backward", [&] {
        scalar_t* idata = grad_input.data_ptr<scalar_t>();
        scalar_t* odata = grad_output.data_ptr<scalar_t>();

        upsample_linear1d_backward_out_frame<scalar_t>(
            odata,
            idata,
            input_width,
            output_width,
            nbatch,
            channels,
            align_corners,
            scales);
      });
}
} // namespace

Tensor& upsample_linear1d_out_cpu(
    Tensor& output,
    const Tensor& input,
    IntArrayRef output_size,
    bool align_corners,
    c10::optional<double> scales) {
  upsample_linear1d_out_cpu_template(output, input, output_size, align_corners, scales);
  return output;
}

Tensor upsample_linear1d_cpu(
    const Tensor& input,
    IntArrayRef output_size,
    bool align_corners,
    c10::optional<double> scales) {
  auto output = at::empty({0}, input.options());
  upsample_linear1d_out_cpu_template(output, input, output_size, align_corners, scales);
  return output;
}

Tensor& upsample_linear1d_backward_out_cpu(
    Tensor& grad_input,
    const Tensor& grad_output,
    IntArrayRef output_size,
    IntArrayRef input_size,
    bool align_corners,
    c10::optional<double> scales) {
  upsample_linear1d_backward_out_cpu_template(
      grad_input, grad_output, output_size, input_size, align_corners, scales);
  return grad_input;
}

Tensor upsample_linear1d_backward_cpu(
    const Tensor& grad_output,
    IntArrayRef output_size,
    IntArrayRef input_size,
    bool align_corners,
    c10::optional<double> scales) {
  auto grad_input = at::zeros(input_size, grad_output.options());
  upsample_linear1d_backward_out_cpu_template(
      grad_input, grad_output, output_size, input_size, align_corners, scales);
  return grad_input;
}

} // namespace native
} // namespace at