Add hand vectorised avx512 binary operators - Githubissues

marian-nmt / marian-dev

Fast Neural Machine Translation in C++ - development repository

https://marian-nmt.github.io

Other

257 stars 126 forks source link

Add hand vectorised avx512 binary operators #961

Open XapaJIaMnu opened 2 years ago

XapaJIaMnu commented 2 years ago

Description

This pull requests adds avx512 vectorisation for binary operators

List of changes:

AVX512 vectorisations
Small optimisations for avx2 and sse cases

Added dependencies: none

How to test

Please run this on one of your servers with avx512 support to see if you actually get some performance benefits from it

Checklist

[x] I have tested the code manually
[x] I have run regression tests
[x] I have read and followed CONTRIBUTING.md
[x] I have updated CHANGELOG.md

I have attached the output of the failed regression tests: regtest.tar.gz