hubert@feyrer.de said:
> Can you post some numbers for comparison, and for curiosity, before
> and after? 

What I've changed was just getcontext(), which is not what bytebench
tests. So one would have to build a libc with more or all syscalls
using the old int0x80.
Perhaps after the 22nd - I'll be AFK until then...

Just to get a number, I've run the syscall part on the (2x1.8GHz)
amd64 box:
$ /usr/pkg/libexec/bytebench/syscall 10
2211452 loops

That's not impressive, my 1GHz P3 is much faster...
Otoh, to be fair, it should be noted that the amd64 fast syscall
path has still some room for optimization - there are more registers
saved/restored than necessary. And it is an SMP thing, so we get
all the locking overhead.
I've done that change right now to avoid binary compatibility
issues later when the 2.0 release process has begun.

best regards
Matthias