UpSampleNearest3d.cpp

#include <ATen/ATen.h>
#include <ATen/NativeFunctions.h>
#include <ATen/native/UpSample.h>
#include <c10/util/irange.h>

namespace at {
namespace meta {

TORCH_META_FUNC(upsample_nearest3d) (
    const Tensor& input,
    IntArrayRef output_size,
    c10::optional<double> scales_d,
    c10::optional<double> scales_h,
    c10::optional<double> scales_w
) {
  auto full_output_size = native::upsample_3d_common_check(input.sizes(), output_size);

  // Allow for empty batch size but not other dimensions
  TORCH_CHECK(
      input.numel() != 0 || c10::multiply_integers(input.sizes().begin() + 1, input.sizes().end()),
      "Non-empty 5D data tensor expected but got a tensor with sizes ",
      input.sizes());

  set_output(full_output_size, input.options().memory_format(input.suggest_memory_format()));
}

TORCH_META_FUNC(_upsample_nearest_exact3d) (
  const Tensor& input,
  IntArrayRef output_size,
  c10::optional<double> scales_d,
  c10::optional<double> scales_h,
  c10::optional<double> scales_w
) {
  auto full_output_size = native::upsample_3d_common_check(input.sizes(), output_size);

  // Allow for empty batch size but not other dimensions
  TORCH_CHECK(
      input.numel() != 0 || c10::multiply_integers(input.sizes().begin() + 1, input.sizes().end()),
      "Non-empty 5D data tensor expected but got a tensor with sizes ",
      input.sizes());

  set_output(full_output_size, input.options().memory_format(input.suggest_memory_format()));
}

TORCH_META_FUNC(upsample_nearest3d_backward) (
    const Tensor& grad_output,
    IntArrayRef output_size,
    IntArrayRef input_size,
    c10::optional<double> scales_d,
    c10::optional<double> scales_h,
    c10::optional<double> scales_w
) {
  auto full_output_size = native::upsample_3d_common_check(input_size, output_size);

  TORCH_CHECK(
      grad_output.dim() == 5,
      "Expected grad_output to be a tensor of dimension 5 but got: dimension ", grad_output.dim());

  for (const auto i : c10::irange(5)) {
    TORCH_CHECK(
        grad_output.size(i) == full_output_size[i],
        "Expected grad_output to have the same shape as output;",
        " output.size(", i, ") = ", full_output_size[i],
        " but got grad_output.size(", i, ") = ", grad_output.size(i));
  }

  set_output(input_size, grad_output.options());
}

TORCH_META_FUNC(_upsample_nearest_exact3d_backward) (
  const Tensor& grad_output,
  IntArrayRef output_size,
  IntArrayRef input_size,
  c10::optional<double> scales_d,
  c10::optional<double> scales_h,
  c10::optional<double> scales_w
) {
  auto full_output_size = native::upsample_3d_common_check(input_size, output_size);

  TORCH_CHECK(
      grad_output.dim() == 5,
      "Expected grad_output to be a tensor of dimension 5 but got: dimension ", grad_output.dim());

  for (const auto i : c10::irange(5)) {
    TORCH_CHECK(
        grad_output.size(i) == full_output_size[i],
        "Expected grad_output to have the same shape as output;",
        " output.size(", i, ") = ", full_output_size[i],
        " but got grad_output.size(", i, ") = ", grad_output.size(i));
  }

  set_output(input_size, grad_output.options());
}

} // namespace meta

namespace native {

TORCH_IMPL_FUNC(upsample_nearest3d_out_cpu) (
    const Tensor& input,
    IntArrayRef output_size,
    c10::optional<double> scales_d,
    c10::optional<double> scales_h,
    c10::optional<double> scales_w,
    const Tensor& output
) {
  upsample_nearest3d_kernel(kCPU, output, input, scales_d, scales_h, scales_w);
}

TORCH_IMPL_FUNC(_upsample_nearest_exact3d_out_cpu) (
    const Tensor& input,
    IntArrayRef output_size,
    c10::optional<double> scales_d,
    c10::optional<double> scales_h,
    c10::optional<double> scales_w,
    const Tensor& output
) {
  _upsample_nearest_exact3d_kernel(kCPU, output, input, scales_d, scales_h, scales_w);
}

TORCH_IMPL_FUNC(upsample_nearest3d_backward_out_cpu) (
    const Tensor& grad_output,
    IntArrayRef output_size,
    IntArrayRef input_size,
    c10::optional<double> scales_d,
    c10::optional<double> scales_h,
    c10::optional<double> scales_w,
    const Tensor& grad_input) {
  grad_input.zero_();
  upsample_nearest3d_backward_kernel(kCPU, grad_input, grad_output, scales_d, scales_h, scales_w);
}

TORCH_IMPL_FUNC(_upsample_nearest_exact3d_backward_out_cpu) (
    const Tensor& grad_output,
    IntArrayRef output_size,
    IntArrayRef input_size,
    c10::optional<double> scales_d,
    c10::optional<double> scales_h,
    c10::optional<double> scales_w,
    const Tensor& grad_input) {
  grad_input.zero_();
  _upsample_nearest_exact3d_backward_kernel(kCPU, grad_input, grad_output, scales_d, scales_h, scales_w);
}

// vec variants

using at::native::upsample::compute_output_size;
using at::native::upsample::get_scale_value;

Tensor upsample_nearest3d_cpu(
    const Tensor& input,
    c10::optional<IntArrayRef> output_size,
    c10::optional<ArrayRef<double>> scale_factors) {
  auto osize = compute_output_size(input.sizes(), output_size, scale_factors);
  auto scale_d = get_scale_value(scale_factors, 0);
  auto scale_h = get_scale_value(scale_factors, 1);
  auto scale_w = get_scale_value(scale_factors, 2);
  return at::upsample_nearest3d(input, osize, scale_d, scale_h, scale_w);
}

Tensor _upsample_nearest_exact3d_cpu(
    const Tensor& input,
    c10::optional<IntArrayRef> output_size,
    c10::optional<ArrayRef<double>> scale_factors) {
  auto osize = compute_output_size(input.sizes(), output_size, scale_factors);
  auto scale_d = get_scale_value(scale_factors, 0);
  auto scale_h = get_scale_value(scale_factors, 1);
  auto scale_w = get_scale_value(scale_factors, 2);
  return at::_upsample_nearest_exact3d(input, osize, scale_d, scale_h, scale_w);
}

// when structured kernels can handle QuantizedCPU, update these overloads to be CompositeExplicitAutograd
Tensor upsample_nearest3d_backward_cpu(
    const Tensor& grad_output,
    c10::optional<IntArrayRef> output_size,
    IntArrayRef input_size,
    c10::optional<ArrayRef<double>> scale_factors) {
  auto osize = compute_output_size(input_size, output_size, scale_factors);
  auto scale_d = get_scale_value(scale_factors, 0);
  auto scale_h = get_scale_value(scale_factors, 1);
  auto scale_w = get_scale_value(scale_factors, 2);
  return at::upsample_nearest3d_backward(grad_output, osize, input_size, scale_d, scale_h, scale_w);
}

Tensor _upsample_nearest_exact3d_backward_cpu(
    const Tensor& grad_output,
    c10::optional<IntArrayRef> output_size,
    IntArrayRef input_size,
    c10::optional<ArrayRef<double>> scale_factors) {
  auto osize = compute_output_size(input_size, output_size, scale_factors);
  auto scale_d = get_scale_value(scale_factors, 0);
  auto scale_h = get_scale_value(scale_factors, 1);
  auto scale_w = get_scale_value(scale_factors, 2);
  return at::_upsample_nearest_exact3d_backward(grad_output, osize, input_size, scale_d, scale_h, scale_w);
}

DEFINE_DISPATCH(upsample_nearest3d_kernel);
DEFINE_DISPATCH(_upsample_nearest_exact3d_kernel);
DEFINE_DISPATCH(upsample_nearest3d_backward_kernel);
DEFINE_DISPATCH(_upsample_nearest_exact3d_backward_kernel);

} // namespace native
} // namespace at