mirror of
https://github.com/pytorch/pytorch.git
synced 2025-10-20 21:14:14 +08:00
Summary: Move the alloc_trace logic into a separate class, to reduce risk of deadlocks when mixing with CCA's lock. Switch to an std::mutex instead of std::recursive_mutex. Let's us re-use the logic in TraceEntryRingBuffer class for later diffs. Test Plan: CI, resnet run, and FBR model. Differential Revision: D59690408 Pulled By: aaronenyeshi Pull Request resolved: https://github.com/pytorch/pytorch/pull/130741 Approved by: https://github.com/davidberard98