__builtin_memcpy() slower than memcpy/bcopy (and on linux it is the opposite) ?

[Luigi Rizzo]

On Wed, Jan 23, 2013 at 11:29 AM, Artem Belevich <art at freebsd.org> wrote:

> On Wed, Jan 23, 2013 at 8:32 AM, Luigi Rizzo <rizzo at iet.unipi.it> wrote:
> > Probably our compiler folks have some ideas on this...
> >
> > When doing netmap i found that on FreeBSD memcpy/bcopy was expensive,
> > __builtin_memcpy() was even worse, and so i ended up writing
> > my custom routine, (called pkt_copy() in the program below).
> > This happens with gcc 4.2.1, clang, gcc 4.6.4
>
> The program does not seem to have pkt_copy. It does have fast_bcopy.
> Is that the one you meant by pkt_copy?
>
>
sorry for the confusion, i did some last-minute name changes.

pkt_copy() is the name of the C function,
./testloop -m fastcopy is the name you need to use to run pkt_copy()

cheers
luigi