I ran NetBSD on NEC Mobile Pro 800 for couple days now.

I tried to run hbench-OS (a pathced version of lmbench) on that,
and results are surprisingly contradicted.

The libc's bzero reports only 4MB/s bandwidth, while unrolled
write reports 150MB/s. I doubt that there might have something
wrong with libc's assembly version of bzero, even bcopy (pipe has
low bandwidth, too)

I tried to rebuild the libc, but failed, and it's too slow. (I
have no native NetBSD machine around, I did the compilation on
the device)

Can anyone out there help me to build a libc without using the
bcopy.S bzero.S? use libc/string/b{copy,zero}.c instead.

If I am right, this might bring the device's performance up to
couple times faster.

Thanks.

// Shawn