Summary:
There is a module called `2to3` which you can target for future specifically to remove these, the directory of `caffe2` has the most redundant imports:
```2to3 -f future -w caffe2```
Pull Request resolved: https://github.com/pytorch/pytorch/pull/45033
Reviewed By: seemethere
Differential Revision: D23808648
Pulled By: bugra
fbshipit-source-id: 38971900f0fe43ab44a9168e57f2307580d36a38
Summary:
Changes in PR https://github.com/pytorch/pytorch/issues/39759 broke HIP caffe2.
hipify for caffe2 renames CUDA to HIP; torch does not.
If caffe2 calls into torch, it needs to use CUDA-named functions.
CC ezyang xw285cornell sunway513 houseroad dzhulgakov
Pull Request resolved: https://github.com/pytorch/pytorch/pull/39801
Differential Revision: D21982493
Pulled By: xw285cornell
fbshipit-source-id: 8e88e0fb80c71f0342e23ef0214a42d5542bdc70
Summary:
THCAllocator functionality is pretty obscure and it's hard to get it working with HIP because of how Caffe2/PyTorch rules are set up (see https://github.com/pytorch/pytorch/issues/39801). Let's just disable the test.
Pull Request resolved: https://github.com/pytorch/pytorch/pull/39843
Reviewed By: zou3519
Differential Revision: D21998687
Pulled By: dzhulgakov
fbshipit-source-id: cd12ba30cdfee658b98393ed3a72e83f4ecf1c9c
Summary:
Pull Request resolved: https://github.com/pytorch/pytorch/pull/39759
Caffe2 has a mode where it uses PT's caching allocator. Somehow we were not calling the initialization explicitly.
Now, I have no idea why it worked before. Probably worth to run a bisect separately.
Reviewed By: houseroad
Differential Revision: D21962331
fbshipit-source-id: f16ad6b27a67dbe0bda93939cca8c94620d22a09
