Incorrect values compared to torch sum

[zuodexin]

Hello. It seems that scatter_sum is not precise compared to torch implementation. Is this a bug? I encountered this problem in my personal project which is sensitive to numerical precision and it caused my program to produce incorrect results .

Python 3.8.10 (default, May 26 2023, 14:05:08) 
[GCC 9.4.0] on linux
Type "help", "copyright", "credits" or "license" for more information.
>>> import torch_scatter
>>> import torch
>>> a = torch.randn(1,7,6)*100000
>>> ii = torch.zeros(7).long()
>>> ss = scatter_sum(a, ii, dim=1, dim_size=1)
>>> ss
tensor([[[-186314.9688, -331904.1875,  159179.0938,  189606.7188,   91803.9062,
            92107.0312]]])
>>> ss_target = a.sum(dim=1)
>>> ss_target
tensor([[-186314.9688, -331904.1875,  159179.0938,  189606.7188,   91803.9141,
           92107.0000]])
>>> ss==ss_target
tensor([[[ True,  True,  True,  True, False, False]]])

version
torch 1.10.0+cu113
torch-scatter 2.1.2

[github-actions]

This issue had no activity for 6 months. It will be closed in 2 weeks unless there is some new activity. Is this issue already resolved?