Re: f@#$ing MMX emulator

Chris Wedgwood (chris@cybernet.co.nz)
Fri, 19 Jun 1998 15:30:54 +1200


> > have highly specialised and specific purposes. Until a compiler exists
> > that actually makes use of MMX opcodes (which i dread the day this
> > happens), an MMX emulator is COMPLETELY USELESS and just contributes to
>
> The latest egcs can make use of the sparc equivalent of MMX already

But the sparc equivalent seems to be better designed and actually useful for
lots of stuff outside of a few corner cases.

Intel could have done better with MMX by not overlaying shit accross the FPU
registers... use a different register bank and require the the OS sets some
specific bit somewhere to make them valid. It might have even saved a small
amount of silicon.

The argument that Windows blah won't support it doesn't hold, the next
service pack could support this, and they are frequent enough anyhow. Sure,
service packs aren't supposed to add features, but M$ do this anyhow...

Not that any of this matters - MMX emulation is just plain silly. If MMX is
going to be used, its probably used a lot - so emualtion would suck enough
to make you blue in the face.

FPU intructions however, are not necessarily used that often, and having the
odd FPU instruction can still be very useful.

Oh, IMNSHO, putting MMX is userspace is going to make it even slower than in
the kernel by up to two times. It could be a kernel modules, so no bloat and
those who want it could use it.

-Chris

-
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@vger.rutgers.edu