zlib-ng / zlib-ng / 28027510757

94%

push

github

Add Adler32 ARM NEON DotProd variant

Introduce a dotprod-accelerated Adler-32 using vdotq_u32 for both the
byte sum (s1) and the position-weighted sum (s2). Four independent
accumulator sets break the dependency chains between successive dotprod
instructions, allowing the pipeline to fill without stalling on results
from the prior iteration.

On Apple M3, the dotprod path is roughly 49% faster than the standard
NEON implementation for inputs above 32 bytes.

Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>

16584 of 18023 branches covered (92.02%)

Branch coverage included in aggregate %.

61 of 61 new or added lines in 6 files covered. (100.0%)

12819 of 13330 relevant lines covered (96.17%)

387396248.14 hits per line