pytorch/torch/csrc/distributed/c10d/Functional.hpp
PyTorch MergeBot e7f1e306df Revert "[c10d][Partial-Graph Overlap] Support calling .wait_tensor() within compiled region on output tensor of eager async_op=True collective (#137763)"
This reverts commit 362ca54f03.

Reverted https://github.com/pytorch/pytorch/pull/137763 on behalf of https://github.com/wdvr due to this change is breaking our prod training pipeline (verified with bisect) by increasing memory consumption 4x and causing OOM ([comment](https://github.com/pytorch/pytorch/pull/137763#issuecomment-2435962833))
2024-10-24 17:46:09 +00:00

11 lines
212 B
C++

#pragma once
#include <torch/csrc/distributed/c10d/Work.hpp>
namespace c10d {
C10_EXPORT void register_work(
const at::Tensor& tensor,
const c10::intrusive_ptr<c10d::Work>& work);
} // namespace c10d