// Copyright (c) Facebook, Inc. and its affiliates. All Rights Reserved.
#include <ATen/ATen.h>
#include <ATen/cuda/CUDAContext.h>

#include <THC/THC.h>
#include <THC/THCAtomics.cuh>
#include <THC/THCDeviceUtils.cuh>

// TODO make it in a common file
#define CUDA_1D_KERNEL_LOOP(i, n)                            \
  for (int i = blockIdx.x * blockDim.x + threadIdx.x; i < n; \
       i += blockDim.x * gridDim.x)


template <typename U, typename T>
__device__ T bilinear_interpolate(const U* bottom_data,
    const int height, const int width,
    T y, T x,
    const int index /* index for debug only*/) {

  // deal with cases that inverse elements are out of feature map boundary
  if (y < -1.0 || y > height || x < -1.0 || x > width) {
    //empty
    return 0;
  }

  if (y <= 0) y = 0;
  if (x <= 0) x = 0;

  int y_low = (int) y;
  int x_low = (int) x;
  int y_high;
  int x_high;

  if (y_low >= height - 1) {
    y_high = y_low = height - 1;
    y = (T) y_low;
  } else {
    y_high = y_low + 1;
  }

  if (x_low >= width - 1) {
    x_high = x_low = width - 1;
    x = (T) x_low;
  } else {
    x_high = x_low + 1;
  }

  T ly = y - y_low;
  T lx = x - x_low;
  T hy = 1. - ly, hx = 1. - lx;
  // do bilinear interpolation
  T v1 = bottom_data[y_low * width + x_low];
  T v2 = bottom_data[y_low * width + x_high];
  T v3 = bottom_data[y_high * width + x_low];
  T v4 = bottom_data[y_high * width + x_high];
  T w1 = hy * hx, w2 = hy * lx, w3 = ly * hx, w4 = ly * lx;

  T val = (w1 * v1 + w2 * v2 + w3 * v3 + w4 * v4);

  return val;
}

// rois in math type (float). This is because ROIs come in as float. 
// TODO: Change other blocks producing ROI to support half type as well
template <typename U, typename T>
__global__ void RoIAlignForward(const int nthreads, const U* bottom_data,
    const T spatial_scale, const int channels,
    const int height, const int width,
    const int pooled_height, const int pooled_width,
    const int sampling_ratio,
    const T* bottom_rois, U* top_data) {
  CUDA_1D_KERNEL_LOOP(index, nthreads) {
    // (n, c, ph, pw) is an element in the pooled output
    int pw = index % pooled_width;
    int ph = (index / pooled_width) % pooled_height;
    int c = (index / pooled_width / pooled_height) % channels;
    int n = index / pooled_width / pooled_height / channels;

    const T* offset_bottom_rois = bottom_rois + n * 5;
    int roi_batch_ind = offset_bottom_rois[0];

    // Do not using rounding; this implementation detail is critical
    T roi_start_w = offset_bottom_rois[1] * spatial_scale;
    T roi_start_h = offset_bottom_rois[2] * spatial_scale;
    T roi_end_w = offset_bottom_rois[3] * spatial_scale;
    T roi_end_h = offset_bottom_rois[4] * spatial_scale;
    // T roi_start_w = round(offset_bottom_rois[1] * spatial_scale);
    // T roi_start_h = round(offset_bottom_rois[2] * spatial_scale);
    // T roi_end_w = round(offset_bottom_rois[3] * spatial_scale);
    // T roi_end_h = round(offset_bottom_rois[4] * spatial_scale);

    // Force malformed ROIs to be 1x1
    T roi_width = max(roi_end_w - roi_start_w, (T)1.);
    T roi_height = max(roi_end_h - roi_start_h, (T)1.);
    T bin_size_h = static_cast<T>(roi_height) / static_cast<T>(pooled_height);
    T bin_size_w = static_cast<T>(roi_width) / static_cast<T>(pooled_width);

    const U* offset_bottom_data = bottom_data + (roi_batch_ind * channels + c) * height * width;

    // We use roi_bin_grid to sample the grid and mimic integral
    int roi_bin_grid_h = (sampling_ratio > 0) ? sampling_ratio : ceil(roi_height / pooled_height); // e.g., = 2
    int roi_bin_grid_w = (sampling_ratio > 0) ? sampling_ratio : ceil(roi_width / pooled_width);

    // We do average (integral) pooling inside a bin
    const T count = roi_bin_grid_h * roi_bin_grid_w; // e.g. = 4

    T output_val = 0.;
    for (int iy = 0; iy < roi_bin_grid_h; iy ++) // e.g., iy = 0, 1
    {
      const T y = roi_start_h + ph * bin_size_h + static_cast<T>(iy + .5f) * bin_size_h / static_cast<T>(roi_bin_grid_h); // e.g., 0.5, 1.5
      for (int ix = 0; ix < roi_bin_grid_w; ix ++)
      {
        const T x = roi_start_w + pw * bin_size_w + static_cast<T>(ix + .5f) * bin_size_w / static_cast<T>(roi_bin_grid_w);

        T val = bilinear_interpolate(offset_bottom_data, height, width, y, x, index);
        output_val += val;
      }
    }
    output_val /= count;

    top_data[index] = output_val;
  }
}

template <typename U, typename T>
__device__ T bilinear_interpolate_nhwc(const U* bottom_data,
    const int height, const int width, const int channels,
    T y, T x,
    const int index /* index for debug only*/) {

  // deal with cases that inverse elements are out of feature map boundary
  if (y < -1.0 || y > height || x < -1.0 || x > width) {
    //empty
    return 0;
  }

  if (y <= 0) y = 0;
  if (x <= 0) x = 0;

  int y_low = (int) y;
  int x_low = (int) x;
  int y_high;
  int x_high;

  if (y_low >= height - 1) {
    y_high = y_low = height - 1;
    y = (T) y_low;
  } else {
    y_high = y_low + 1;
  }

  if (x_low >= width - 1) {
    x_high = x_low = width - 1;
    x = (T) x_low;
  } else {
    x_high = x_low + 1;
  }

  T ly = y - y_low;
  T lx = x - x_low;
  T hy = 1. - ly, hx = 1. - lx;
  // do bilinear interpolation
  T v1 = bottom_data[channels * (y_low * width + x_low)];
  T v2 = bottom_data[channels * (y_low * width + x_high)];
  T v3 = bottom_data[channels * (y_high * width + x_low)];
  T v4 = bottom_data[channels * (y_high * width + x_high)];
  T w1 = hy * hx, w2 = hy * lx, w3 = ly * hx, w4 = ly * lx;

  T val = (w1 * v1 + w2 * v2 + w3 * v3 + w4 * v4);

  return val;
}

// rois in math type (float). This is because ROIs come in as float. 
// TODO: Change other blocks producing ROI to support half type as well
template <typename U, typename T>
__global__ void RoIAlignForwardNHWC(const int nthreads, const U* bottom_data,
    const T spatial_scale, const int channels,
    const int height, const int width,
    const int pooled_height, const int pooled_width,
    const int sampling_ratio,
    const T* bottom_rois, U* top_data) {
  CUDA_1D_KERNEL_LOOP(index, nthreads) {
    // (n, c, ph, pw) is an element in the pooled output
    int c = index % channels;
    int pw = (index / channels) % pooled_width;
    int ph = (index / channels / pooled_width) % pooled_height;
    int n = index / pooled_width / pooled_height / channels;

    const T* offset_bottom_rois = bottom_rois + n * 5;
    int roi_batch_ind = offset_bottom_rois[0];

    // Do not using rounding; this implementation detail is critical
    T roi_start_w = offset_bottom_rois[1] * spatial_scale;
    T roi_start_h = offset_bottom_rois[2] * spatial_scale;
    T roi_end_w = offset_bottom_rois[3] * spatial_scale;
    T roi_end_h = offset_bottom_rois[4] * spatial_scale;
    // T roi_start_w = round(offset_bottom_rois[1] * spatial_scale);
    // T roi_start_h = round(offset_bottom_rois[2] * spatial_scale);
    // T roi_end_w = round(offset_bottom_rois[3] * spatial_scale);
    // T roi_end_h = round(offset_bottom_rois[4] * spatial_scale);

    // Force malformed ROIs to be 1x1
    T roi_width = max(roi_end_w - roi_start_w, (T)1.);
    T roi_height = max(roi_end_h - roi_start_h, (T)1.);
    T bin_size_h = static_cast<T>(roi_height) / static_cast<T>(pooled_height);
    T bin_size_w = static_cast<T>(roi_width) / static_cast<T>(pooled_width);

    const U* offset_bottom_data = bottom_data + (roi_batch_ind * channels * height * width + c);

    // We use roi_bin_grid to sample the grid and mimic integral
    int roi_bin_grid_h = (sampling_ratio > 0) ? sampling_ratio : ceil(roi_height / pooled_height); // e.g., = 2
    int roi_bin_grid_w = (sampling_ratio > 0) ? sampling_ratio : ceil(roi_width / pooled_width);

    // We do average (integral) pooling inside a bin
    const T count = roi_bin_grid_h * roi_bin_grid_w; // e.g. = 4

    T output_val = 0.;
    for (int iy = 0; iy < roi_bin_grid_h; iy ++) // e.g., iy = 0, 1
    {
      const T y = roi_start_h + ph * bin_size_h + static_cast<T>(iy + .5f) * bin_size_h / static_cast<T>(roi_bin_grid_h); // e.g., 0.5, 1.5
      for (int ix = 0; ix < roi_bin_grid_w; ix ++)
      {
        const T x = roi_start_w + pw * bin_size_w + static_cast<T>(ix + .5f) * bin_size_w / static_cast<T>(roi_bin_grid_w);

        T val = bilinear_interpolate_nhwc(offset_bottom_data, height, width, channels, y, x, index);
        output_val += val;
      }
    }
    output_val /= count;

    top_data[index] = output_val;
  }
}

template <typename T>
__device__ void bilinear_interpolate_gradient(
    const int height, const int width,
    T y, T x,
    T & w1, T & w2, T & w3, T & w4,
    int & x_low, int & x_high, int & y_low, int & y_high,
    const int index /* index for debug only*/) {

  // deal with cases that inverse elements are out of feature map boundary
  if (y < -1.0 || y > height || x < -1.0 || x > width) {
    //empty
    w1 = w2 = w3 = w4 = 0.;
    x_low = x_high = y_low = y_high = -1;
    return;
  }

  if (y <= 0) y = 0;
  if (x <= 0) x = 0;

  y_low = (int) y;
  x_low = (int) x;

  if (y_low >= height - 1) {
    y_high = y_low = height - 1;
    y = (T) y_low;
  } else {
    y_high = y_low + 1;
  }

  if (x_low >= width - 1) {
    x_high = x_low = width - 1;
    x = (T) x_low;
  } else {
    x_high = x_low + 1;
  }

  T ly = y - y_low;
  T lx = x - x_low;
  T hy = 1. - ly, hx = 1. - lx;

  // reference in forward
  // T v1 = bottom_data[y_low * width + x_low];
  // T v2 = bottom_data[y_low * width + x_high];
  // T v3 = bottom_data[y_high * width + x_low];
  // T v4 = bottom_data[y_high * width + x_high];
  // T val = (w1 * v1 + w2 * v2 + w3 * v3 + w4 * v4);

  w1 = hy * hx, w2 = hy * lx, w3 = ly * hx, w4 = ly * lx;

  return;
}

template <typename U, typename T>
__global__ void RoIAlignBackwardFeature(const int nthreads, const U* top_diff,
    const int num_rois, const T spatial_scale,
    const int channels, const int height, const int width,
    const int pooled_height, const int pooled_width,
    const int sampling_ratio,
    U* bottom_diff,
    const T* bottom_rois) {
  CUDA_1D_KERNEL_LOOP(index, nthreads) {
    // (n, c, ph, pw) is an element in the pooled output
    int pw = index % pooled_width;
    int ph = (index / pooled_width) % pooled_height;
    int c = (index / pooled_width / pooled_height) % channels;
    int n = index / pooled_width / pooled_height / channels;

    const T* offset_bottom_rois = bottom_rois + n * 5;
    int roi_batch_ind = offset_bottom_rois[0];

    // Do not using rounding; this implementation detail is critical
    T roi_start_w = offset_bottom_rois[1] * spatial_scale;
    T roi_start_h = offset_bottom_rois[2] * spatial_scale;
    T roi_end_w = offset_bottom_rois[3] * spatial_scale;
    T roi_end_h = offset_bottom_rois[4] * spatial_scale;
    // T roi_start_w = round(offset_bottom_rois[1] * spatial_scale);
    // T roi_start_h = round(offset_bottom_rois[2] * spatial_scale);
    // T roi_end_w = round(offset_bottom_rois[3] * spatial_scale);
    // T roi_end_h = round(offset_bottom_rois[4] * spatial_scale);

    // Force malformed ROIs to be 1x1
    T roi_width = max(roi_end_w - roi_start_w, (T)1.);
    T roi_height = max(roi_end_h - roi_start_h, (T)1.);
    T bin_size_h = static_cast<T>(roi_height) / static_cast<T>(pooled_height);
    T bin_size_w = static_cast<T>(roi_width) / static_cast<T>(pooled_width);

    U* offset_bottom_diff = bottom_diff + (roi_batch_ind * channels + c) * height * width;

    int top_offset    = (n * channels + c) * pooled_height * pooled_width;
    const U* offset_top_diff = top_diff + top_offset;
    const T top_diff_this_bin = offset_top_diff[ph * pooled_width + pw];

    // We use roi_bin_grid to sample the grid and mimic integral
    int roi_bin_grid_h = (sampling_ratio > 0) ? sampling_ratio : ceil(roi_height / pooled_height); // e.g., = 2
    int roi_bin_grid_w = (sampling_ratio > 0) ? sampling_ratio : ceil(roi_width / pooled_width);

    // We do average (integral) pooling inside a bin
    const T count = roi_bin_grid_h * roi_bin_grid_w; // e.g. = 4

    for (int iy = 0; iy < roi_bin_grid_h; iy ++) // e.g., iy = 0, 1
    {
      const T y = roi_start_h + ph * bin_size_h + static_cast<T>(iy + .5f) * bin_size_h / static_cast<T>(roi_bin_grid_h); // e.g., 0.5, 1.5
      for (int ix = 0; ix < roi_bin_grid_w; ix ++)
      {
        const T x = roi_start_w + pw * bin_size_w + static_cast<T>(ix + .5f) * bin_size_w / static_cast<T>(roi_bin_grid_w);

        T w1, w2, w3, w4;
        int x_low, x_high, y_low, y_high;

        bilinear_interpolate_gradient(height, width, y, x,
            w1, w2, w3, w4,
            x_low, x_high, y_low, y_high,
            index);

        T g1 = top_diff_this_bin * w1 / count;
        T g2 = top_diff_this_bin * w2 / count;
        T g3 = top_diff_this_bin * w3 / count;
        T g4 = top_diff_this_bin * w4 / count;

        if (x_low >= 0 && x_high >= 0 && y_low >= 0 && y_high >= 0)
        {
          atomicAdd(offset_bottom_diff + y_low * width + x_low, static_cast<T>(g1));
          atomicAdd(offset_bottom_diff + y_low * width + x_high, static_cast<T>(g2));
          atomicAdd(offset_bottom_diff + y_high * width + x_low, static_cast<T>(g3));
          atomicAdd(offset_bottom_diff + y_high * width + x_high, static_cast<T>(g4));
        } // if
      } // ix
    } // iy
  } // CUDA_1D_KERNEL_LOOP
} // RoIAlignBackward

template <typename U, typename T>
__global__ void RoIAlignBackwardFeatureNHWC(const int nthreads, const U* top_diff,
    const int num_rois, const T spatial_scale,
    const int channels, const int height, const int width,
    const int pooled_height, const int pooled_width,
    const int sampling_ratio,
    U* bottom_diff,
    const T* bottom_rois) {
  CUDA_1D_KERNEL_LOOP(index, nthreads) {
    // (n, c, ph, pw) is an element in the pooled output
    int c = index % channels;
    int pw = (index / channels) % pooled_width;
    int ph = (index / channels / pooled_width) % pooled_height;
    int n = index / pooled_width / pooled_height / channels;

    const T* offset_bottom_rois = bottom_rois + n * 5;
    int roi_batch_ind = offset_bottom_rois[0];

    // Do not using rounding; this implementation detail is critical
    T roi_start_w = offset_bottom_rois[1] * spatial_scale;
    T roi_start_h = offset_bottom_rois[2] * spatial_scale;
    T roi_end_w = offset_bottom_rois[3] * spatial_scale;
    T roi_end_h = offset_bottom_rois[4] * spatial_scale;
    // T roi_start_w = round(offset_bottom_rois[1] * spatial_scale);
    // T roi_start_h = round(offset_bottom_rois[2] * spatial_scale);
    // T roi_end_w = round(offset_bottom_rois[3] * spatial_scale);
    // T roi_end_h = round(offset_bottom_rois[4] * spatial_scale);

    // Force malformed ROIs to be 1x1
    T roi_width = max(roi_end_w - roi_start_w, (T)1.);
    T roi_height = max(roi_end_h - roi_start_h, (T)1.);
    T bin_size_h = static_cast<T>(roi_height) / static_cast<T>(pooled_height);
    T bin_size_w = static_cast<T>(roi_width) / static_cast<T>(pooled_width);

    U* offset_bottom_diff = bottom_diff + (roi_batch_ind * channels * height * width + c);

    int top_offset    = n * channels * pooled_height * pooled_width + c;
    const U* offset_top_diff = top_diff + top_offset;
    const T top_diff_this_bin = offset_top_diff[channels * (ph * pooled_width + pw)];

    // We use roi_bin_grid to sample the grid and mimic integral
    int roi_bin_grid_h = (sampling_ratio > 0) ? sampling_ratio : ceil(roi_height / pooled_height); // e.g., = 2
    int roi_bin_grid_w = (sampling_ratio > 0) ? sampling_ratio : ceil(roi_width / pooled_width);

    // We do average (integral) pooling inside a bin
    const T count = roi_bin_grid_h * roi_bin_grid_w; // e.g. = 4

    for (int iy = 0; iy < roi_bin_grid_h; iy ++) // e.g., iy = 0, 1
    {
      const T y = roi_start_h + ph * bin_size_h + static_cast<T>(iy + .5f) * bin_size_h / static_cast<T>(roi_bin_grid_h); // e.g., 0.5, 1.5
      for (int ix = 0; ix < roi_bin_grid_w; ix ++)
      {
        const T x = roi_start_w + pw * bin_size_w + static_cast<T>(ix + .5f) * bin_size_w / static_cast<T>(roi_bin_grid_w);

        T w1, w2, w3, w4;
        int x_low, x_high, y_low, y_high;

        bilinear_interpolate_gradient(height, width, y, x,
            w1, w2, w3, w4,
            x_low, x_high, y_low, y_high,
            index);

        T g1 = top_diff_this_bin * w1 / count;
        T g2 = top_diff_this_bin * w2 / count;
        T g3 = top_diff_this_bin * w3 / count;
        T g4 = top_diff_this_bin * w4 / count;

        if (x_low >= 0 && x_high >= 0 && y_low >= 0 && y_high >= 0)
        {
          atomicAdd(offset_bottom_diff + channels * (y_low * width + x_low), static_cast<T>(g1));
          atomicAdd(offset_bottom_diff + channels * (y_low * width + x_high), static_cast<T>(g2));
          atomicAdd(offset_bottom_diff + channels * (y_high * width + x_low), static_cast<T>(g3));
          atomicAdd(offset_bottom_diff + channels * (y_high * width + x_high), static_cast<T>(g4));
        } // if
      } // ix
    } // iy
  } // CUDA_1D_KERNEL_LOOP
} // RoIAlignBackward


at::Tensor ROIAlign_forward_cuda(const at::Tensor& input,
                                 const at::Tensor& rois,
                                 const float spatial_scale,
                                 const int pooled_height,
                                 const int pooled_width,
                                 const int sampling_ratio,
				 const bool is_nhwc) {
  AT_ASSERTM(input.is_cuda(), "input must be a CUDA tensor");
  AT_ASSERTM(rois.is_cuda(), "rois must be a CUDA tensor");

  auto num_rois = rois.size(0);
  auto channels = is_nhwc ? input.size(3) : input.size(1);
  auto height   = is_nhwc ? input.size(1) : input.size(2);
  auto width    = is_nhwc ? input.size(2) : input.size(3);

  auto output = is_nhwc ? at::empty({num_rois, pooled_height, pooled_width, channels}, input.options()) : 
	  at::empty({num_rois, channels, pooled_height, pooled_width}, input.options());
  auto output_size = num_rois * pooled_height * pooled_width * channels;
  cudaStream_t stream = at::cuda::getCurrentCUDAStream();

  if (output.numel() == 0) {
    THCudaCheck(cudaGetLastError());
    return output;
  }

  int gridSize;
  int blockSize;
  cudaOccupancyMaxPotentialBlockSize(&gridSize,
                                     &blockSize,
                                     (void*) RoIAlignForward<float, float>,
                                     0,  // dynamic memory
                                     0); // maximum utilized threads   

  dim3 grid(gridSize);
  dim3 block(blockSize);
  
  //TODO: Math type is hard coded to float assuming double is not used, if needed, add a case for double as well. 
  //In case of double, it should be <double, double>, not <double, float>
  //TODO: ROIs come in as float, fix other blocks so they come in as same type as input. 
  if (!is_nhwc){
    AT_DISPATCH_FLOATING_TYPES_AND_HALF(input.scalar_type(), "ROIAlign_forward", [&] {
      RoIAlignForward<scalar_t, float><<<grid, block, 0, stream>>>(
           output_size,
           input.contiguous().data_ptr<scalar_t>(),
           spatial_scale,
           channels,
           height,
           width,
           pooled_height,
           pooled_width,
           sampling_ratio,
           rois.contiguous().data_ptr<float>(),
           output.data_ptr<scalar_t>());
    });
  }
  else{
    AT_DISPATCH_FLOATING_TYPES_AND_HALF(input.scalar_type(), "ROIAlign_forward", [&] {
      RoIAlignForwardNHWC<scalar_t, float><<<grid, block, 0, stream>>>(
           output_size,
           input.contiguous().data_ptr<scalar_t>(),
           spatial_scale,
           channels,
           height,
           width,
           pooled_height,
           pooled_width,
           sampling_ratio,
           rois.contiguous().data_ptr<float>(),
           output.data_ptr<scalar_t>());
    });
  }
  THCudaCheck(cudaGetLastError());
  return output;
}

// TODO remove the dependency on input and use instead its sizes -> save memory
// NHWC + layout transposes are faster than NCHW, so just keep the NHWC implementation for backward pass
at::Tensor ROIAlign_backward_cuda(const at::Tensor& grad,
                                  const at::Tensor& rois,
                                  const float spatial_scale,
                                  const int pooled_height,
                                  const int pooled_width,
                                  const int batch_size,
                                  const int channels,
                                  const int height,
                                  const int width,
                                  const int sampling_ratio,
				  const bool is_nhwc) {
  AT_ASSERTM(grad.is_cuda(), "grad must be a CUDA tensor");
  AT_ASSERTM(rois.is_cuda(), "rois must be a CUDA tensor");

  auto num_rois = rois.size(0);
  auto grad_input = is_nhwc ? at::zeros({batch_size, height, width, channels}, grad.options()) : 
	  at::zeros({batch_size, channels, height, width}, grad.options());


  cudaStream_t stream = at::cuda::getCurrentCUDAStream();

  // handle possibly empty gradients
  if (grad.numel() == 0) {
    THCudaCheck(cudaGetLastError());
    return grad_input;
  }

  int gridSize;
  int blockSize;
  cudaOccupancyMaxPotentialBlockSize(&gridSize,
                                     &blockSize,
                                     (void*) RoIAlignBackwardFeature<float, float>,
                                     0,  // dynamic memory
                                     0); // maximum utilized threads   

  dim3 grid(gridSize);
  dim3 block(blockSize);
  
  //TODO: Math type is hard coded to float assuming double is not used, if needed, add a case for double as well. 
  //In case of double, it should be <double, double>, not <double, float>
  //TODO: ROIs come in as float, fix other blocks so they come in as same type as input. 
  if (!is_nhwc){
    AT_DISPATCH_FLOATING_TYPES_AND_HALF(grad.scalar_type(), "ROIAlign_backward", [&] {
      RoIAlignBackwardFeature<scalar_t, float><<<grid, block, 0, stream>>>(
           grad.numel(),
           grad.contiguous().data_ptr<scalar_t>(),
           num_rois,
           spatial_scale,
           channels,
           height,
           width,
           pooled_height,
           pooled_width,
           sampling_ratio,
           grad_input.data_ptr<scalar_t>(),
           rois.contiguous().data_ptr<float>());
    });
  }
  else{
    AT_DISPATCH_FLOATING_TYPES_AND_HALF(grad.scalar_type(), "ROIAlign_backward", [&] {
      RoIAlignBackwardFeatureNHWC<scalar_t, float><<<grid, block, 0, stream>>>(
           grad.numel(),
           grad.contiguous().data_ptr<scalar_t>(),
           num_rois,
           spatial_scale,
           channels,
           height,
           width,
           pooled_height,
           pooled_width,
           sampling_ratio,
           grad_input.data_ptr<scalar_t>(),
           rois.contiguous().data_ptr<float>());
    });
  }
  THCudaCheck(cudaGetLastError());
  return grad_input;
}