Release 3.0.1 · COMBINE-lab/cuttlefish

Release Notes

Performance release; outputs and interfaces are unchanged.

The weak-super-k-mer label packer's PEXT fast path now dispatches on
runtime CPU-feature detection instead of a compile-time bmi2 gate. Every
distributed 3.0.0 artifact (GitHub binaries, cargo install, bioconda) was
silently running the scalar fallback — forfeiting a measured ~6.5% of the
partition phase that only target-cpu=native builds kept. Any BMI2-capable
CPU (Haswell, 2013+) now gets the fast path regardless of build flags.
Builds now carry portable per-target CPU baselines, mirroring piscem:
x86-64-v3 with AVX2 on x86-64, Neoverse-N1 / Apple-A14 on arm64.
.cargo/config.toml is tracked with these baselines and the release
workflow applies them to the prebuilt binaries; local target-cpu=native
tuning moves to an explicit, gitignored .cargo/config.local.toml.
The prebuilt x86-64 binaries consequently assume x86-64-v3; on older
machines, cargo install cuttlefish-rs-cli still produces a baseline
binary whose runtime dispatch keeps the packer correct everywhere.
cuttlefish build now reports its CPU-capability tier at startup: silent
when the fast paths are compiled in statically, a one-line note on portable
builds (dispatch active, with the RUSTFLAGS recipe for a native rebuild),
and a loud warning on pre-BMI2 CPUs where the scalar fallback runs.

curl --proto '=https' --tlsv1.2 -LsSf https://github.com/COMBINE-lab/cuttlefish/releases/download/v3.0.1/cuttlefish-rs-cli-installer.sh | sh

File	Platform	Checksum
cuttlefish-rs-cli-aarch64-apple-darwin.tar.xz	Apple Silicon macOS	checksum
cuttlefish-rs-cli-x86_64-apple-darwin.tar.xz	Intel macOS	checksum
cuttlefish-rs-cli-aarch64-unknown-linux-gnu.tar.xz	ARM64 Linux	checksum
cuttlefish-rs-cli-x86_64-unknown-linux-gnu.tar.xz	x64 Linux	checksum