Re: fpu emulation

On Wed, 6 Sep 2000, Andy Valencia wrote:

> The interesting part is how to do it in a "microkernel" sort of way. It'd
> no doubt be easy enough to pull the FPU emulation from a monolithic kernel,
> and paste it into the VSTa kernel. But... ick. You'd have to review its
> design carefully anyway to handle the preemptive, SMP, and multi-threaded
> issues. But even then, why bloat a kernel which has avoided bloat for so
> long?
>
> So instead, the question is how to attach it to the FPU-using process's
> address space, and intercept math faults? In the latest release, the raw
> event handling system calls of VSTa have been wrapped in an event registry.
> POSIX signal emulation uses that, and it seems like a nice place for FPU
> emulation to attach, too. But to have this automatically present means that
> each process (at least each using floating point) has to "arm" this handler.
> A clever way of doing this without an impact on integer-only programs would
> be nice.
>
> I've always got about this far in my thinking, and then I look at how few SX
> chips are left in the world. :->

well, no doubt SX is a dying kind.. but as terminal I think a 486sx-50
would do just fine or perhaps a mc68030-33 without FPU...

but still something like this might be needed in the future perhaps for
MMX emulation on non-MMX processors (even thought I don't think that will
ever be needed considering the small increase in speed mmx gives... but
there is/will be similar extensions...)

but btw, I also had problems with the newest release to start some
 like vi, emacs, less and more. I haven't checked if it will work on some
a bit more recent computer, but will very soon.
And when I tried to do anything in / (a FAT-16 filesystem mounted there)
the dos driver died and said: rw.c/165: pack_error: null in extesion.
all other directories on the fat drive works all right with long filenames
and all (yippie)

/Erik
Received on Wed Sep 6 13:16:00 2000