Hwo to use atomic_min in warp

Omniverse Isaac Sim
,
1

Isaac Sim Version

4.5.0

@wp.kernel
def bin_bbox_process(bbox_corners: wp.array(ndim=3, dtype=wp.float32),
                     x_offset: wp.float32,
                     y_offset: wp.float32,
                     x_res: wp.float32,
                     y_res: wp.float32,
                    #  bbox_corners_bin_idx: wp.array(ndim=2, dtype=wp.vec2ui),
                    aligned_bbox_corners_idx: wp.array(ndim=2, dtype=wp.uint32)
                    ):
    i, j = wp.tid()
    bbox_corner_spher = cartesian_to_spherical(wp.vec3(bbox_corners[i,j,0],
                                                       bbox_corners[i,j,1],
                                                       bbox_corners[i,j,1]))
    
    x_bin_idx = wp.uint32((bbox_corner_spher[0] - x_offset) / x_res)
    y_bin_idx = wp.uint32((bbox_corner_spher[1] - y_offset) / y_res)

   # This line gives an error.
    wp.atomic_min(aligned_bbox_corners_idx, wp.uint32(0), wp.uint32(0), x_bin_idx)

Gives an kernel compilation error saying

no instance of overloaded function "wp::atomic_min" matches the argument list
            argument types are: (wp::uint32 *, wp::uint32)
          detected during instantiation of "T wp::atomic_min(const A<T> &, int, int, T) [with A=wp::array_t, T=wp::uint32]"

The weird thing is If replace atomic_min with atomic_add, this kernel can run without any problem.

But atomic_min and atomic_add has exactly same syntax, please correct if I am wrong.

Thanks

2

An update to this problem, I have used wp.float32 or wp.int32 for the value and array entries in this function,
now this kernel can be compiled without any problem.

Is this a bug that this function only accepts unsigned data type?

4

Thank you for reporting this. We have shared it with our team and will provide an update once we have more information.

5

Hi @HaoyuMa, thanks for reporting this! We have a fix for Warp that is in progress and that should be part of the next 1.7.2 release coming at the end of the month.

6

Just to expand on what Chris said, we previously only exposed wp.atomic_min() on signed integer types (int). We will expose atomic operations on unsigned integers in the upcoming 1.7.2 or 1.8.0 release.

7

This topic was automatically closed 14 days after the last reply. New replies are no longer allowed.