V4.5.0 Performance improvements, Bug fixes, add hpa_hgemm

amcamd released this 12 Sep 14:55

· 3670 commits to master since this release

Features

add support for vega20
add hpa_hgemm assembly and source
tuning for sgemm and hgemm
bug fixes for sgemm and hgemm small sizes
use SGPR for alpha and beta

Assets 2