Closed gtank closed 5 years ago
Benchmark from Crostini on a low-end Pixelbook (Intel(R) Core(TM) i5-7Y57 CPU @ 1.20GHz). They're noisy (and there's likely some conflation with other go1.12 changes) but the general picture looks pretty good.
$ benchstat go1.11.5.noasm.bench go1.12rc1.noasm.bench name old time/op new time/op delta IsOnCurve-4 583ns ± 5% 402ns ± 6% -31.03% (p=0.000 n=10+9) Add-4 16.7µs ± 7% 7.0µs ± 9% -58.14% (p=0.000 n=10+10) Double-4 16.4µs ± 2% 6.8µs ± 2% -58.37% (p=0.000 n=8+8) ScalarMult-4 394µs ± 6% 174µs ± 4% -55.91% (p=0.000 n=9+8) FeMul51-4 83.0ns ± 3% 31.3ns ± 3% -62.33% (p=0.000 n=9+8) FeSquare51-4 52.4ns ± 2% 20.5ns ± 2% -60.85% (p=0.000 n=8+8) FeFromBig-4 74.6ns ± 1% 74.0ns ± 3% ~ (p=0.060 n=8+8) FeToBig-4 197ns ± 4% 182ns ± 5% -7.55% (p=0.000 n=9+9)
Benchmark from Crostini on a low-end Pixelbook (Intel(R) Core(TM) i5-7Y57 CPU @ 1.20GHz). They're noisy (and there's likely some conflation with other go1.12 changes) but the general picture looks pretty good.