Benchmark

back to home

I write and optimize high-performance arithmetic software. Here are benchmarks from my implementations.


Invert_2p255m19 - SafeGCD / Bernstein-Yang Inverse

Explore the implementation on GitHub ↗

Inversion modulo 2255 − 19 on Raspberry Pi 400 / Cortex-A72.

3917.22 cycles per inversion, 13.84% fewer cycles than the earlier implementation's 4546.24 cycles measured in the same Cortex-A72 benchmark run.

Warm-cache repeated calls with paired empty-call subtraction; the result is the median of six per-run medians. See the measurement record.

Inversion modulo 2255 − 19 on Raspberry Pi 5 / Cortex-A76.

3279.70 cycles per inversion, 10.79% fewer cycles than the 3676.26-cycle baseline (see below).

(commit date)

Inversion modulo 2255 − 19 on Raspberry Pi 5 / Cortex-A76.

3676.26 cycles per inversion, approximately 8.1% fewer cycles using about 4000 cycles as the reference baseline (see Accelerating and Verifying Constant-Time Modular Inversion).