mirror of
https://github.com/pytorch/pytorch.git
synced 2025-10-20 21:14:14 +08:00
95198f8299
resolves: https://github.com/pytorch/pytorch/issues/138721 Summary: Delete the uses of deleted nodes. The double for-loop is icky here, but N should be pretty small and removing it requires refactoring the datastructures involved, which is a bigger endeavor. Test Plan: Normal test coverage should be sufficient. There were a couple of spots in the scheduler code that didn't check users being deleted, so I'll run a perf test to see what impact that has, and to make sure N^2 doesn't affect compile times. Perf: https://hud.pytorch.org/benchmark/compilers?dashboard=torchinductor&startTime=Tue%2C%2029%20Oct%202024%2017%3A41%3A36%20GMT&stopTime=Tue%2C%2005%20Nov%202024%2018%3A41%3A36%20GMT&granularity=hour&suite=torchbench&mode=inference&dtype=bfloat16&deviceName=cuda%20(a100)&lBranch=exclamaforte/prune-deleted-users&lCommit=5cb1aa6f7d8a52acdae0c7cf36b8c2d536d7f0d1&rBranch=main&rCommit=f4ee5a243dbb31e6310e5632b1c87898b299df2c off of nov4 nightly Pull Request resolved: https://github.com/pytorch/pytorch/pull/139447 Approved by: https://github.com/eellison
Note [TH abstraction violation] ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~ TH/THC provide some hpp headers, which are proper C++ headers rather than C headers. These headers serve double duty as *internal implementation detail* headers, whose contents should largely not be used by external clients. Ideally, we would not install these headers at all; instead, you should use public functions (in headers like `THTensor.h`, NOT `THTensor.hpp`) to manipulate these structs. However, there are a few places in torch/csrc where we violate this abstraction. They are marked with a pointer to this note. Each of those sites will have to be refactored when we refactor the guts of THTensor and related structures.