Skip to content

[x86_64] Assembly version of SHA1Transform added to hash package

J. Gareth "Kit" Moreton requested to merge CuriousKit/optimisations:sha1-x86_64 into main

Summary

This merge request builds on !485 (merged) to provide an assembly version of the SHA-1 hash algorithm in the hash package.

What is the current bug behavior?

N/A

What is the behavior after applying this patch?

New assembly version of the SHA-1 algorithm that should be significantly faster than the Pascal fallback, as well as providing cleaner expansion options for other platform-specific implementations.

Relevant logs and/or screenshots

Running a timing test under x86_64-win64 - before (Pascal):

SHA1(Hello):             473 ns/call
SHA1(World x 1000):      26792 ns/call

After (x86_64 assembly language):

SHA1(Hello):             223 ns/call
SHA1(World x 1000):      8119 ns/call

Additional notes

Despite attempts to speed things up with SIMD instructions, instruction rearranging, utilising more registers for some of the data expansion, and merging rounds together (the last 6 rounds do calculate some of the data expansion 2 rounds at a time, using the full 64-bit general purpose registers), performance was overall worse due to additional overhead.

Merge request reports