[Skiboot] [PATCH 09/11] Optimize memset() implementation to do word at a time
stewart at linux.vnet.ibm.com
Mon May 25 13:30:29 AEST 2015
Patrick Williams III <iawillia at us.ibm.com> writes:
> If you compile with -O3, gcc will already do loop unrolling for naive
> implementations. There isn't a good reason to make the code more complex
> when you could just enable -O3 for your libc parts.
I've been thinking about this for a bit... I'm not terribly keen on
different CFLAGS for different parts of the tree.... I'm considering a
dcbz based implementation instead though, especially as Mikey has been
trying to get total executed instruction count down to the absolute
minimum needed (currently down to ~600k from 15M).
More information about the Skiboot