Ke Wen d444384003 [SymmMem] Tiled reduce (#162243) ... Added op: `tile_reduce(Tensor input, Tensor(a!) out, int root, str group_name)` For now supports only: - NVSHMEM backed symmetric tensor; - 2D tensor and tile; - torch.float. Testing on right-bottom quandrant: ``` rank 0: tensor([[0., 0., 0., 0., 0., 0., 0., 0.], [0., 0., 0., 0., 0., 0., 0., 0.], [0., 0., 0., 0., 0., 0., 0., 0.], [0., 0., 0., 0., 0., 0., 0., 0.], [0., 0., 0., 0., 1., 1., 1., 1.], [0., 0., 0., 0., 1., 1., 1., 1.], [0., 0., 0., 0., 1., 1., 1., 1.], [0., 0., 0., 0., 1., 1., 1., 1.]], device='cuda:0') PASSED ``` Pull Request resolved: https://github.com/pytorch/pytorch/pull/162243 Approved by: https://github.com/ngimel		2025-10-08 02:03:04 +00:00
..
ddp	[BE][CI] bump ruff to 0.8.4 (#143753)	2024-12-24 12:24:10 +00:00
bench_nvshmem_tile_reduce.py	[SymmMem] Tiled reduce (#162243)	2025-10-08 02:03:04 +00:00