[uClibc] Re: improved memcpy/memset on arm in uClibc-0.9.19
Miles Bader
miles at lsi.nec.co.jp
Thu Sep 18 01:46:50 UTC 2003
Markus Pietrek <maillist at fsforth.de> writes:
> Of course any libc function must behave like the specification. So,
> memcpy() must check itself whether the requirements are fullfilled for
> an optimized approach. But of course this bloats the code, so my
> version is about 4-5 times larger than the standard one :-(
Yup. If you use shared libraries, then the speed gain is probably
worth the space penalty, but if you use static libraries (like me :-(
then it's not so clear.
I've wanted to use the optimized memcpy from the linux kernel for
uclibc too (on the v850), but keep hesitating due to the above point.
I wonder how often memcpy gets called in a typical user program anyway...
-Miles
--
I'd rather be consing.
More information about the uClibc
mailing list