Use SIMD

[Sc00bz]

Your benchmark numbers are too low because you don't use SIMD such as SSE2, AVX, AVX2, and AVX512. Your CPU "2019 Macbook Pro (2.6 GHz 6-Core Intel Core i7)" is likely a i7-8850H. If I didn't mess up because that's the only CPU I could find that matches those stats. A i7-8850H has AVX2 so your numbers should be about 7.6x higher with a correct implementation.

Note "When running on a single processor it took around 80 seconds to generate all of the proximity keys. When running on all CPUs (12) it takes around 20 seconds." means you messed up. First you should only run 6 threads because similar work loads when run on a hyperthreading CPU slows it down vs using correct core count. But likely your threading code is poor because 2/3 speed should not happen. It should be more like at least 80% speed with using all hyperthreading "cores" (ie 80/6/.8 or 16.7 seconds unless there is heavy rounding going on).

Oh right the average number of new cases for the last week is 82.1k vs 50k per day. Also it's 14 vs 10 days. So my guess for your laptop is 24 seconds for a single thread and 4.0 seconds for 6 threads. Ignoring 82.1k vs 50k and 14 vs 10 days it's 11 seconds for a single thread and 1.8 seconds for 6 threads. This all assumes that the current SHA256 code is optimal, but if not then it's less.

[fewstera]

Thank you for the response, I wasn't aware of SIMD, this is the response I was hoping for by publishing this repo.

My processor is an 'Intel(R) Core(TM) i7-9750H CPU @ 2.60GHz', which appears to have AVX2 support. For a more accurate representation, I should probably be running these tests on a mobile rather than my laptop.

I tried a drop-in SIMD sha256 implementation (https://github.com/minio/sha256-simd), but this made a negligible difference, I believe this may be due to the HMAC function still being from the go standard library implementation.