Why pytorch call cuda math api in two different way, like ::aten or std::aten

1

why pytorch call cuda match api in two different way, like ::aten or std::aten.
In op function, the ::aten is called. pytorch/aten/src/ATen/native/cuda/UnaryGeometricAtanKernel.cu at main · pytorch/pytorch · GitHub
and in foreach op function, std::aten is called.
pytorch/aten/src/ATen/native/cuda/ForeachUnaryOp.cu at main · pytorch/pytorch · GitHub

@ezyang @albanD

2

@crcrpar Hi, do you know anything about this. Thx

1 Like