perf: widening 32x32 multiply in the vector kernels, extracted into VectorMath - #8

Merged

unsafePtr merged 20 commits into

perf/widening-multiply-32

Aug 7, 2026

Commits on Jul 27, 2026

refactor: move vector kernels into VectorMath
unsafePtr
committed
74ba40a
perf: use widening 32x32 multiply in vector kernels
unsafePtr
committed
9efcd13
test: cover VectorMath kernels directly
unsafePtr
committed
0b82f2a
test: cover fast-path rejection of values too large for 32/64 bytes
unsafePtr
committed
210d6d0
refactor: fast decode paths return bool, drop misleading Try prefix
unsafePtr
committed
f64e22d
refactor: mark stateless Base58 helpers static
unsafePtr
committed
aaa6951
ci: add arm64 codegen probe for the vector kernels
unsafePtr
committed
175c4fb
ci: fix arm64 codegen probe filter and fail on empty dump
unsafePtr
committed
b2b9aec
perf: use uzp1 and umull for the widening multiply on arm64
unsafePtr
committed
36f4555
perf: narrow with xtn instead of uzp1 in the arm64 widening multiply
unsafePtr
committed
d1ef6ef
ci: probe arm64 on neoverse and apple silicon
unsafePtr
committed
dcd9bdc
ci: add cross-platform benchmark comparison against the branch baseline
unsafePtr
committed
cd141ac
chore: drop the net9 benchmark jobs
unsafePtr
committed
548fb9c
Revert "ci: add cross-platform benchmark comparison against the branch baseline"
unsafePtr
committed
8f70bea

Commits on Aug 6, 2026