[PyTorch] random_ on floating type with negative values is broken on both CPU and GPU #6338
Description
Reproduced on master and also broken on 0.3.1.
>>> torch.randn(3,3).random_(-1,10).max().item()
9.0
>>> torch.randn(3,3).random_(-1,10).max().item()
1.8446744073709552e+19
>>> torch.randn(3,3).cuda().random_(-1,10).max().item()
4294967296.0
>>> torch.randn(3,3).cuda().random_(-1,10).max().item()
8.0
>>> torch.randn(3,3).cuda().random_(-1,10).max().item()
4294967296.0
>>> torch.randn(3,3).cuda().random_(-1,10).max().item()
4294967296.0
>>> torch.randn(3,3).cuda().random_(-1,10).max().item()
9.0
>>> torch.randn(3,3).cuda().random_(-1,10).max().item()
9.0
>>> torch.randn(3,3).cuda().random_(-1,10).max().item()
8.0
>>> torch.randn(3,3).cuda().random_(-1,10).max().item()
8.0
>>> torch.randn(3,3).cuda().random_(-1,10).max().item()
8.0
>>> torch.randn(3,3).cuda().random_(-1,10).max().item()
4294967296.0
>>> torch.randn(3,3).cuda().random_(-1,10).max().item()
4294967296.0Reported in #6136
Metadata
Metadata
Assignees
Labels
No labels