Re: FPU avoidance idea
Philip Blundell <firstname.lastname@example.org> writes:
> >How's this for an idea - what if we could cook up a little binary that
> >will do an LD_PRELOAD of the __setfpucw to something harmless?
> That would probably work, yes. I'd managed to delude myself that
> __setfpucw actually got inlined into the .init section of the
> binary, but of course that's not actually the case and it's safely
> tucked away in the library. Doh.
Good. I'll try to do something with that later today. I can probably
just whip up a little static C proggie, and then patch some NOPs over
the evil code.
> The other (somewhat nasty) option that occurred to me just now was
> to include just enough code in the kernel to recognise and ignore
> "wfs" instructions while the emulator is not present, in which case
> probably no binaries would need changing unless they actually tried
> to do real floating point work.
That would be even nicer. Us NetWinder folk will need to wait for
some kernel source first though. (wink, wink, nudge, nudge ;-)