Skip to content

Optimize inference#128

Open
ArniFreyrG wants to merge 9 commits into
mainfrom
optimize_inference
Open

Optimize inference#128
ArniFreyrG wants to merge 9 commits into
mainfrom
optimize_inference

Conversation

@ArniFreyrG
Copy link
Copy Markdown
Collaborator

This includes a subset of things from #127 geared towards optimizing inference only

pierpal and others added 9 commits March 16, 2026 18:43
Bug fixes:
- Fix incorrect error message in ThreadsLowMem constructor
- Replace 20x exit(1) with throw for proper exception handling
- Fix memory leak: raw HMM* → unique_ptr
- Fix sign-extension in pair_key (int → uint32_t cast)
- Fix sign bug in rho computation (visible at scale)

Performance:
- Replace unordered_map traceback storage with std::deque arena
- Replace std::set neighbor search with boolean array scan in Matcher
- Add periodic pruning every 500 sites (bounds memory, improves cache)
- Replace hash maps with flat vectors in ThreadsLowMem/ViterbiState
- Add NumPy batch API for Matcher, Viterbi, and hets phases
- Use const int* in process_site to avoid bounds-check overhead
- Improve move semantics and reference passing

Testing:
- Add 69 C++ unit tests (444 assertions) via Catch2
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

2 participants