Eliminate tensor copies from compute_common_type_ in TensorIterator. (#30018)
Summary:
This requires refactoring at::native::result_type to operate as a
state machine, processing the input types one at a time. There may
be other places in the code base that could benefit from adopting
this approach as well.
Pull Request resolved: https://github.com/pytorch/pytorch/pull/30018
Differential Revision: D18606427
Pulled By: resistor
fbshipit-source-id: f6b779326bdb746508690cf7ca6de777adc66244
