[SR] Disable aten::where out variant (#73367)
Summary:
Pull Request resolved: https://github.com/pytorch/pytorch/pull/73367
The op is currently bugged w.r.t. a `condition` input that is not the same shape as the others:
```
def forward(self, cond_1d, x, y):
shape = [-1] + [1] * (x.dim() - 1)
cond = cond_1d.view(shape)
return torch.where(cond, x, y).clone()
Condition:
01
00
[ CPUBoolType{2} ]
A:
06 -9
08 -8
[ CPULongType{2,2} ]
B:
-4 05
-5 -2
[ CPULongType{2,2} ]
Actual:
06 05
-5 -2
[ CPULongType{2,2} ]
Expected:
06 -9
-5 -2
[ CPULongType{2,2} ]
```
ghstack-source-id: 149963254
Test Plan: Unit tests exercise broadcasting
Reviewed By: d1jang
Differential Revision: D34454770
fbshipit-source-id: 6ad4c4ca6893d2b87852a17d437437d99ca94ab4
(cherry picked from commit 7135bc40e9fd930c08f5291b7d6b4902ec30005b)
