See, for example:
|
native::pow_out(result, c10::scalar_to_tensor(base, exp.device()), exp); |
There are several other cases, too, and the pattern is, in general, an antipattern. This may affect type promotion and (see below) impacts performance.
cc @ezyang @gchanan @zou3519 @bdhirsh @jbschlosser @ngimel @VitalyFedyunin
See, for example:
pytorch/aten/src/ATen/native/Pow.cpp
Line 53 in 272f4db
There are several other cases, too, and the pattern is, in general, an antipattern. This may affect type promotion and (see below) impacts performance.
cc @ezyang @gchanan @zou3519 @bdhirsh @jbschlosser @ngimel @VitalyFedyunin