[inductor] Support alpha= in add/sub #1111
Conversation
| if allow_alpha: | ||
| if alpha is not None and alpha != 1: | ||
| inputs = list(inputs) | ||
| inputs[-1] = mul(inputs[-1], alpha) |
There was a problem hiding this comment.
assert alpha is a scalar?
There was a problem hiding this comment.
mul() already has asserts on its inputs.
There was a problem hiding this comment.
Hmm, I assumed mul would be able to take tensors, where alpha was required to be a scalar. As long as mul asserts, sgtm!
There was a problem hiding this comment.
Often constants get wrapped as tensors in AOT Autograd. So we want this to work for tensors.
There was a problem hiding this comment.
But we still need to check that it's 0d and that it's dtype is correct for the other inputs (e.g. integral inputs and float alpha error out in eager)
jansel
changed the title
|
Why was this added here, rather than simply using the |
|
we can't use aten -> prim decompositions from primTorch because they hardcode broadcasting. |
Fixes #1085