float2::x and float2::y may not be the same as float on ROCm (#35593)
Summary:
This causes ambiguity and can be triggered sometimes (e.g., by https://github.com/pytorch/pytorch/issues/35217). Explicitly convert them to float.
error: conditional expression is ambiguous; 'const
hip_impl::Scalar_accessor<float, Native_vec_, 0>' can be converted to
'float' and vice versa
Pull Request resolved: https://github.com/pytorch/pytorch/pull/35593
Differential Revision: D20735663
Pulled By: ezyang
fbshipit-source-id: ae6a38a08e59821bae13eb0b9f9bdf21a008d5c0
