Glibc Preparing Another Round Of Optimizations To Benefit AVX-512 CPUs

Just last week I wrote about Glibc preparing more optimizations for AVX-512 CPUs with implementing more C library functions in EVEX512 enhanced vector extension versions. This week there is more EVEX512 function work.

Sunil Pandey has implemented EVEX512 versions of the strchrnul, strchr, and wcschr character in string functions. This follows the EVEX512 versions of memchr, rawmemchr, and wmemchr that came last week and other AVX-512/EVEX512 work to land in the GNU C library over the past few years.

For AVX512VL + AVX512BW CPUs these optimized functions can take 30% less CPU cycles than the standard AVX paths. The timing though of this AVX-512 work is nice with Zen 4 bringing AVX-512 to AMD CPUs in an efficient manner.

The latest EVEX512 function work for Glibc can be found via this commit. These latest Glibc performance optimizations and other enhancements will be part of Glibc 2.37 that is expected to be released in February.