Conference hydra::amiga_v1

Re: .0

>    Microbotics provides a new hardware FPP library that you drop into
>    your LIBS: directory.

They probably replace the IEEE double floating point library.  The 68881
is an IEEE floating point processor.  I am not sure but they probably
do not replace the single precision Fast Floating Point library since it
uses an incompatible number format (I think).

>    Software that has been written to check for
>    the presence of a 68881 will recognize it, and use the new FPP routine.
>
>    The only question I have is, how many packages automatically check
>    for a 68881?

The beauty of replacing one of the standard math libraries in the Amiga
with one that uses the 68881 is that all programs that use that standard
Amiga floating point library do NOT have to check to see if the 68881 is
installed.  The calls to the library just run several times faster.

Remember, libraries on the Amiga are bound to the program at execution
time.  Thus, if you improve how a library does things, all programs that
use that library, be they compiled 1.5 years ago with Amiga C 1.0 from
Commodore, or they be compiled tomorrow with the latest version of
Manx, will benefit.

Having the Manx compiler with native 68881 support would not benefit
the owner of a Starboard with a multifunction card.  The 68881 is
designed to interface to a 68020.  A 68000 or 68010 can not directly
use a 68881: that is you can not directly embed a 68881 instruction
in your 68000 or 68010 code.  (Although, I have heard of a 68010
board that uses a combination of hardware and software hacks to
directly support a 68881.)  However, the 68881 can be indirectly used
by a 68881 by treating the 68881 as a device that the 68000 communicates
with.  Instead of the 68000 program just executing a 68881 instruction,
the 68000 pushes a piece of data at the 68881 which happens to be a
68881 instruction.

This is how the Microbiotics board works.  The user program calls the
usual library routine to multiply to numbers.  The normal library has
been replaced by the Microbiotics library; so the routine that multiplies
two numbers does not do the usual bit hacking in the 68000 to do the
multiply.  Instead it gives a command to the 68881 to multiply the
numbers.  The overhead of the library call and the explicit communication
between the 68000 and the 68881 is slower than if the 68000 could just
"execute" 68881 instruction directly, do it is still faster than doing
the entire process in the 68000.

So, the question becomes: Which of the two floating point libraries on
the Amiga are replaced with Microbiotics library or libraries?  Does the
program you want to speed up use the library that has been replaced
by the Microbiotics version?

    Assuming that Microbotics replaces LIBS:IEEEdoubbas.library, then
    the question really becomes 'which library does the application
    call'.  If it calls for double precision, you're in luck.  If it
    calls for single, then the 68881 will not be used.
    
    What would be the disadvantage of replacing both single and double
    precision libraries?  That way, it seems like all math would be
    speeded up... unless the overhead of calling the 68881 is more time
    consuming than just doing the math in single precision in the 68000.
    
    Very interesting... Thanks again Randy for explaining things more
    clearly.
    
    By the way, does anyone have any idea on how much faster math
    operations are performed?  Say for example computing COS(x) in the
    68881 vs using the 68000 + the math libraries?  I'm assuming here
    (perhaps naively) that the math libraries contain routines for
    evaluating trig and log type functions via an infinite series.
    
    Strong reminder here that I'm just a blockhead mechanical engineer,
    so be patient if I oversimplify things or outright misunderstand
    them.
    

    
    Has anyone priced a 68881?  I believe it is somewhat expensive.
    

    I've seen 68881s for 199.00 mail order.
    John

    Re: 68881 prices
    
    Can you recall the supplier?  I poured over Computer Shopper, but
    couldn't find any sources for moto chips.  There were about 2000
    'Chips 'R' Us' ads for people selling the Intel family.
    
    

    More random info on MicroBotics and 68881 chips...
    
    The MicroBotics Multifunction card has it's own 12 MHz clock.  The
    co-processor does not use the Amiga's clock.
    
    Advanced Computer Products (800) 854-8230 is selling 12 MHz 68881's
    for $219.95, and 16 MHz 68881's for $299.95.
    
    There would be no advantage to getting the 16 MHz version unless
    you planned on someday upgrading to a 14.32 MHz turbo 68020/68881
    card.  In that case, the 12 MHz chip would melt.
    
    My pocket is getting itchy, but my cheapness is winning out so far...
    

    I finally called MB and tried to get some understanding of how their
    68881 hack works.  I spoke with someone named Mike, and what he
    told me was not very encouraging.
    
    1.  There is not a single solitary piece of software available today
    that will benefit from their 68881/math library.  The program would
    have to be written in such a way that 68881 op code is injected
    into the data stream (Mike's words).  MB has sent documentation
    to all major developers on how to check for the presence of their
    libraries & the 68881 chip.
    
    2.  It will probably be quite a while before anyone decides to
    support/upgrade their software to take advantage of the multifunction
    board.
    
    Based on this info, I guess there is not much advantage to running
    out and buying this kit.
    
    

Re: .7

That's funny.  A few weeks ago I saw a few boxes full of multi-function
modules at the Memory Location.  Since all this stuff came up in this
conference, I decided to peek into the box, and see what the documentation
said.

The multi-function card comes with a disk.  On this disk was a replacement
for the standard IEEE double precision Amiga library, a entirely new of
other IEEE functions, and examples of how to call the 68881 directly.

Since they ship a replacement for the standard Amiga library, I would
expect any software that uses that library to be speeded up.

I wonder if Mike really knows what he is talking about, or what the
purpose of the mathieeedoublebas.library they ship is?  Or am I
imaging that I saw that library?

    I saw the same thing you did.  In fact, there are THREE replacement
    libraries.  
    
    I, too wonder if Mike knows what he's talking about.  I tried to
    talk to Redmond Simonsen, but he's never in.
    
    I would certainly hope that this product is more useful than Mike
    indicated.  The credits in the box list Dale Luck and other Amiga
    celebrities as co-authors of the replacement libraries, so I'd like
    to believe that one can actually do something with this kit without
    waiting for the world to re-design itself.
    
    

re: .7 .8 .9

Well, I wasn't following this too closely because I'm not
directly interested, but I vaguely remember skimming an review or
ad in one of the latest Amiga mags.  What I think it said was
that the replacement math libraries were there to translate
library calls into 68881 calls and that existing software could
see some improvement simply by using these replacement libraries.
The big performance improvement would come from accessing the
68881 directly without using the math libraries at all.  Probably
what Mike meant was there wasn't any existing software that would
do that yet.

Maybe?

John

Re: .7, .8, .9. 10

I decided to call Microbiotics and see if they give me a different story.
I talked to Glen.

He said that any program that used the Amiga standard IEEE double precision
library would see a speed up.  I asked him how much of a speed up.  He
said he didn't really know.

I also asked them about their new adapter card for the Starboard 2 so
that it can be used with the Amiga 2000.  The card is going to cost
less than 40 dollars.  (I have been thinking of trading in my 1000 for
a 2000, and I would rather not take a loss on the Starboard 2 I own.)

He said that the adapter board will be out in less than a month.  (I
remembered when they claimed that about the multi-function board: they
were only 400% off.)  I am inclined to believe them since the board
only has to take the 100 pin Zorro bus and turn it into whatever
fraction of the 86 pin raw expansion that a Starboard 2 actually uses.

    Heard a rumor Sunday night on Plink that the MicroBotics SCSI adaptor
    will be a small hardware add-on to the MultiFunction module.
    
    I found this hard to believe, since the MF board is pretty tiny
    and pretty jam-packed.  But then again, my Supra SCSI board has
    approximately nothing on it.
    
    If anyone has an idle few moments, perhaps this could be confirmed
    or anti-confirmed with MB over the phone.
    

    'tis true. Should be on shelves in about 3-4 weeks.
    
    Won't fit if you have a MFU installed.

    Does anyone know if it is a DMA SCSI controller?
    
    just curious,
    
    brad