invalid device ordinal

[LucasColas]

Everytime I try to run the code I get the same error :
point cloud cnt = 1553
{-0.519505, -3.18713e+301, -0.271626}
scale = 1.09354e+304
register init point cloud
allocated memory for the world space bounding box ...
calling builder...
done building tree, took 2.8ms
KDT Upper: (0 0.00291451 0)
KDT Lower: (0 -0.5 0)
Ps, Ns read
Build Tree
Build Done
DataprePointCloud
[Debug] grid reso: 400
[Debug] max depth: 10
[Debug] normal kw: 10
[Debug] use radii: 0
[Debug] offset 1: 0.05
[Debug] offset 2: 0.01
begin generate random normals
generate normals end
igl_octree_leaves.size = 2
_ = 0 0 0
_(0) = 0
_.rows = 1, _cols = 3
O_W.rows = 153, O_W.cols = 1
leaves_center.rows = 2
leaves_idx.rows = 2
smooth normals to Octree leaves
Eigen_Ns shape = 1553, 3
Ps shape = 1553, 3
igl_octree_leaves.size = 2
allocated memory for the world space bounding box ...
calling builder...
done building tree, took 3.3ms
KDT Upper: (0.000245564 0.00266894 0.000245564)
KDT Lower: (0.000245564 -0.499754 0.000245564)
Build Tree 2
Build Tree 2 Done
igl_octree_leaves.size = 2
done searching, took 125.9us
res_x = 69, res_y = 400, res_z = 69
GV shape = 1904400, 3
FWN Pre
FWN Pre Done
Res X: 69 Res Y: 400 Res Z: 69
GPU Point Cloud Size: 2
node_num: 72
Prepare Time = 0
Gauss Map Normal Initialized
iter = 0
GWN Done
GWN Done
GWN time = 0
terminate called after throwing an instance of 'thrust::THRUST_200700_520_NS::system::system_error'
what(): after reduction step 1: cudaErrorInvalidDevice: invalid device ordinal
Abandon (core dumped)

If I do a nvidia-smi or nvcc it finds my GPU / nvcc. Is the error related to the fact that somes of those values are giant :
{-0.519505, -3.18713e+301, -0.271626}
scale = 1.09354e+304
?
Why is it. The point clouds are taken from iSPR : https://github.com/houfei0801/ipsr/tree/master/data