optimized memset for ARM

[Norbert Lange]

Hello,

I wrote some optimized memset for older ARM CPUs, and recently cleaned
up the code and tested it with various march settings (down to armv2).
When I wrote the code, most ARMs only used the normal C-Routine,
obviously you now have code for all variants. Still the current
routines will be slower by a factor of 2-3 on old CPUs (eg. burst
writes if you use store multiple instructions). Newer CPUs likely wont
see much difference, even less if you have a preemptive/multicore OS
running.

Still I`d like to put the a reference to the code out here:
https://github.com/nolange/arm-memroutine
Would be glad if this finds its way into newlib someday. Even if not,
perftests (particularly on barebone OSes), critism and patches are
wecome.

Kind Regards,
Norbert Lange