Re: Wishes

[dempson@atlantis.actrix.gen.nz (David Empson)]

In article <49n3nl$c37@apocalypse.dmi.stevens-tech.edu>,
Karl Horster <khorster@attila.stevens-tech.edu> wrote:
> 	Ok, so I used Rsounder for the first time today to see how it 
> worked, and I wanted to convert some files to WAVs for my friends in the 
> PC World.  I chose the fit method and basically watched paint dry.  I 
> have an 8mghz GS, and it was still pretty slow.  What are the chances of 
> something like a mathcoprocessor in the //gs.

There are already two of them.  The "FPE" from Innovative Systems,
which is not available any more, and had reliability problems, and
another card whose name escapes me (are you there, Heinrik?) that is
fully compatible with it.  Both are based around the Motorola 68881 or
68882.

They support a set of patches for the SANE floating point toolset,
which allow GS/OS programs that do floating point calculations to
automatically make use of the card.  It is also possible to use the
card directly, which is significantly faster than going through SANE,
because you avoid the overhead of the toolbox calls.  ORCA/C and
ORCA/Pascal have built-in support for this, provided the program is
specially compiled.n

The question is whether it would make any difference in this case.
Does rSounder make use of floating point calculations to convert WAV
files to the IIgs?  If it is only using integer calculations, then a
maths coprocessor would make no difference.

> I know the biggest drawback of the GS is the slow bus speed

Not really - FPU calculations are relatively slow (some are in the
order of hundreds of microseconds, if not milliseconds).  I don't have
exact figures, but the Apple Numerics Manual says the 68881 offers
speed improvements of 40 to 700 times (averaging 100) on a Mac Plus.
I would expect a greater improvement on the IIgs.  The ORCA/C manual
says that program speed can be about 120 times faster with an FPU.

The SANE toolset takes anything from 0.5 to 100 ms to perform floating
point calculations, depending on the complexity.  Assuming a uniform
improvement of about 100 times, this would result in calculation times
of 5 microseconds to 1 millisecond.

By comparison, an accelerator would only improve the speed of SANE
calculations by up to 3 times (for an 8 MHz accelerator).
-- 
David Empson
dempson@actrix.gen.nz
Snail mail: P.O. Box 27-103, Wellington, New Zealand