- Add parallel_find (#90, @gasche)
- Update CI (#93, @Sudha247)
- Optimisation to work-stealing (#96, @art-w)
- Improve docs presentation (#99, @metanivek)
- Property based tests (#100, jmid)
- Task: avoid double handler installation (#101, @gasche & @clef-men)
- Fix a benign data-race in Chan reported by ocaml-tsan (#103, @art-w)
- Dune, opam, and GitHub Actions fixes (#105, @MisterDA)
- domain local await support (#107, @polytypic)
- Windows run on GitHub Actions (#110, @Sudha247)
- Adjust PBTs based on recommended_domain_count (#112, @jmid)
- Test condition tweaks (#113, @jmid)
This release includes:
- Bug fix for
parallel_for_reduce
on empty loops. - Make Chan.t and Task.promise injective #69
- Add lockfree dependency #70
- CI fixes (#73, #76)
- Breaking change: Rename
num_additional_domains
tonum_domains
for setup_pool - Documentation updates (#80, #81, #82)
Includes Effect.eff -> Effect.t change from OCaml trunk. (#65)
This release fixes compatibility with OCaml 5.00.0+trunk in #61. Breaks compatibility with older Multicore variants 4.12.0+domains and 4.12.0+domains+effects
This release includes:
- Usage of effect handlers for task creation. This introduces a breaking change; all computations need to be enclosed in a Task.run function. See #51.
- Multi_channel uses a per-channel domain-local key, removing the global key. #50
- Bug fixes in parallel_scan. #60
Corresponding updates for breaking changes introduced in ocaml-multicore/ocaml-multicore#704
- Updated with the new interface Domain.cpu_relax
- Domain.timer_ticks replaced with Mirage clock.
- #45 adds support for named pools. This is a breaking change with setup_pool taking an optional name parameter and an extra unit parameter.
- A minor bug fix in parallel_for_reduce.
This release includes:
- A breaking change for Task pools where the num_domains argument has been renamed num_additional_domains to clear up potential confusion; see #31.
- A new work-stealing scheduler for Task pools using domain local Chase Lev deques #29; this can improve performance significantly for some workloads.
- A removal of closure allocation in Chan #28.
- A move to using the Mutex & Condition modules for the implementation of Chan #24.
- Various documentation and packaging improvements (#21, #27, #30, #32).
Updates to:
- parallel_for to use new task distribution algorithm and allow default chunk_size (#16)
- parallel_for_reduce to use new task distribution algorithm and allow default chunk_size parameter (#18)
recv_poll
made non-allocating- Addition of parallel_scan #5
- New Tasks library with support for async/await parallelism and parallel for loops.
- Adds support for non-blocking Chan.send_poll and Chan.recv_poll.
Thanks to @gasche for API design discussions.
Initial release