PyTorch applies missing std::move calls via clang-tidy (PR #189583)
Clang-tidy's new rule caught 100+ missed moves; only one false positive.
Skylion007 submitted Pull Request #189583 to the PyTorch repository, applying missing `std::move` calls discovered by clang-tidy's new `performance-use-std-move` rule. The rule, running on nightly builds, scans for places where variables are passed to functions or returned but not moved, causing avoidable copies. The author double-checked results with Claude and Codex, reporting only a single false positive in the entire PyTorch codebase.
The check is deliberately conservative by default, minimizing risk of breaking changes. The patch touches numerous files across the PyTorch framework, optimizing internal APIs and reducing CPU overhead. This is part of ongoing efforts to squeeze performance from PyTorch's C++ backend, especially relevant for production inference and training loops where every microsecond counts.
- New clang-tidy rule `performance-use-std-move` identifies missing move operations in C++ code
- Only one false positive found across the entire PyTorch codebase after verification with Claude and Codex
- Patch reduces unnecessary copies, improving runtime performance especially in hot paths
Why It Matters
Small C++ optimizations like these compound to make PyTorch faster for production AI workloads.