mirror of
https://github.com/zebrajr/pytorch.git
synced 2026-01-15 12:15:51 +00:00
This PR fixes: - `torch.nonzero` for large tensors on ROCm. It was malfunctioning due to a known hip compiler problem with `::min` for int64_t arguments. Fixed by expliced typing to `std::min<int64_t>` - using `torch.ops.aten.miopen_batch_norm` instead of `torch.ops.aten.cudnn_batch_norm` on ROCm Fixed tests: - Fixes #168878. - Fixes #168879. - Fixes #168553. - Fixes #168554. Pull Request resolved: https://github.com/pytorch/pytorch/pull/169827 Approved by: https://github.com/jeffdaily, https://github.com/mlazos, https://github.com/cyyever Co-authored-by: Jeff Daily <jeff.daily@amd.com>