is there a bug in roi_pooling.cu?

[conansherry]

template<typename Dtype>
__global__ void ROIPoolForwardKernel(const int count, const Dtype* bottom_data,
                                     const float spatial_scale, const int channels,
                                     const int height, const int width,
                                     const int pooled_height, const int pooled_width,
                                     const Dtype* bottom_rois, Dtype* top_data,
                                     Dtype* argmax_data) {
  for (int index = (blockIdx.x + blockIdx.y * gridDim.x) * blockDim.x + threadIdx.x;
       index < count;
       index += blockDim.x * gridDim.x * gridDim.y) {
    // (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;

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

    if (roi_batch_ind < 0) {
      top_data[index] = 0;
      argmax_data[index] = 0;
      continue;
    }

    int roi_start_w = round(bottom_rois[1] * spatial_scale);
    int roi_start_h = round(bottom_rois[2] * spatial_scale);
    int roi_end_w = round(bottom_rois[3] * spatial_scale);
    int roi_end_h = round(bottom_rois[4] * spatial_scale);

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

    int hstart = static_cast<int>(floor(static_cast<Dtype>(ph)
                                        * bin_size_h));
    int wstart = static_cast<int>(floor(static_cast<Dtype>(pw)
                                        * bin_size_w));
    int hend = static_cast<int>(ceil(static_cast<Dtype>(ph + 1)
                                     * bin_size_h));
    int wend = static_cast<int>(ceil(static_cast<Dtype>(pw + 1)
                                     * bin_size_w));

    // Add roi offsets and clip to input boundaries
    hstart = min(max(hstart + roi_start_h, 0), height);
    hend = min(max(hend + roi_start_h, 0), height);
    wstart = min(max(wstart + roi_start_w, 0), width);
    wend = min(max(wend + roi_start_w, 0), width);
    bool is_empty = (hend <= hstart) || (wend <= wstart);

    // Define an empty pooling region to be zero
    Dtype maxval = is_empty ? 0 : -FLT_MAX;
    // If nothing is pooled, argmax = -1 causes nothing to be backprop'd
    int maxidx = -1;
    bottom_data += (roi_batch_ind * channels + c) * height * width;
    for (int h = hstart; h < hend; ++h) {
      for (int w = wstart; w < wend; ++w) {
        int bottom_index = h * width + w;
        if (bottom_data[bottom_index] > maxval) {
          maxval = bottom_data[bottom_index];
          maxidx = bottom_index;
        }
      }
    }
    top_data[index] = maxval;
    argmax_data[index] = (Dtype)maxidx;
  }
}

if blockDim.x * gridDim.x * gridDim.y + (blockIdx.x + blockIdx.y * gridDim.x) * blockDim.x + threadIdx.x < count:
bottom_rois += n * 5;
bottom_data += (roi_batch_ind * channels + c) * height * width;
will be assigned a wrong value?

[conansherry]

template<typename Dtype>
inline void ROIPoolForward(const Tensor<gpu, 4, Dtype> &out,
                           const Tensor<gpu, 4, Dtype> &data,
                           const Tensor<gpu, 2, Dtype> &bbox,
                           const Tensor<gpu, 4, Dtype> &max_idx,
                           const float spatial_scale) {
  const Dtype *bottom_data = data.dptr_;
  const Dtype *bottom_rois = bbox.dptr_;
  Dtype *top_data = out.dptr_;
  Dtype *argmax_data = max_idx.dptr_;
  const int count = out.shape_.Size();
  const int channels = data.size(1);
  const int height = data.size(2);
  const int width = data.size(3);
  const int pooled_height = out.size(2);
  const int pooled_width = out.size(3);
  const int gridSize = (count + kMaxThreadsPerBlock - 1) / kMaxThreadsPerBlock;
  dim3 dimGrid(kMaxGridDim, (gridSize + kMaxGridDim - 1) / kMaxGridDim);
  dim3 dimBlock(kMaxThreadsPerBlock);
  CheckLaunchParam(dimGrid, dimBlock, "ROIPooling Forward");
  cudaStream_t stream = Stream<gpu>::GetStream(out.stream_);
  ROIPoolForwardKernel<Dtype><<<dimGrid, dimBlock, 0, stream>>>(
      count, bottom_data, spatial_scale, channels, height, width,
      pooled_height, pooled_width, bottom_rois, top_data, argmax_data);
}

theory:
floor((m + n - 1) / n) >= m / n (m, n is interger and greater than 0)

blockDim.x * gridDim.x * gridDim.y = 
512 * floor((floor((count + 1023) / 1024) + 511) / 512) * 1024 >= 
512 * floor((count / 1024 + 511) / 512) * 1024 >= 
512 * (count / 1024 / 512) * 1024 >=
count

so it isn't a bug. only if count >> threads in gpu will cause error.

[szha]

@apache/mxnet-committers: This issue has been inactive for the past 90 days. It has no label and needs triage.

For general "how-to" questions, our user forum (and Chinese version) is a good place to get help.

[lanking520]

@conansherry Hi, are you facing the same issues in the most recent update?

[lanking520]

@sandeep-krishnamurthy please add CUDA, C++ to this topic