Explicitly cast an implicit conversion from some macro defined type to a double (#48290)
Summary:
Pull Request resolved: https://github.com/pytorch/pytorch/pull/48290
`scalar_t` here is expanded from nested macros to be an input value and
`upper_bound` is templated upon it. Whatever it gives back is unconditionally
cast to a `double` via the fact that it is always passed to
`binary_kernel_reduce_vec` which has a `double` as the fourth argument.
Change it here to be an explicit `static_cast<double>` to do what the compiler
was implicitly doing.
Test Plan: this is an error with -Werror in llvm11. This allows it to build
Reviewed By: ezyang
Differential Revision: D25111258
fbshipit-source-id: 6837afec52821f1f57b8c8f2df2d0eb3fc9b58bd
