jme@wdl1.UUCP (Joseph M. Earley) (03/15/84)
Mike Brown,
Made two attempts to contact you and did not get a confirmation, I assume my
mail didn't make it.
These are the only responses I received from the floating point question
I put to the net many weeks ago. I couldn't use the information provided.
I suspect the answer lies somewhere in Berkeley. They must have a compiler
that accesses the extended range and extended precision machine instructions.
I have not followed up on that theory but will do so when I am less pressed
by other matters. I have your address and phone number so I will keep you
posted.
Sorry I did not write sooner but I was hoping I could give you something
more substantial than this. Keep the sand out of your boots and stay away
from the girls on the Miracle Mile.
Joe Earley
(Let me know if you receive this.)
P.S. If Kevin Carosso and/or Peter Gross should read this I extend to you my
late, but sincere, thanks for the following responses.
>From fortune!allegra!scgvaxd!kvc Wed Feb 29 01:18:37 1984
>Received: by fortune.UUCP; Wed, 29 Feb 84 01:13 PST
>To: allegra!fortune!wdl1!jme
>Subject: G and H floating...
The G and H floating data types are not supported in hardware by the
780, with or without the floating point accelerator. There is an optional
upgrade kit which WILL give you the microcode, but it will cost you
something on the order of $12,000. You must purchase an extended WCS
and then the new microcode. VMS handles the instructions by providing
software emulation on processors (780's) which do not have the microcode.
(Currently, you specify the emulator when you link your program. Under
VMS V4.0 (maybe out by the summer) it is trapped and handled within
the OS, completely transparent to the executing image)
The VMS emulator works just fine. I use it extensively. Now, if you're
running UNIX, I do not know what you should do. The best solution would
be for UNIX to trap and emulate the instructions transparently (like
VMS V4.0). Otherwise, you may be able to find some sort of handler that
you can link to your program that catches the trap (assuming UNIX allows
the unimplemented instruction fault to be caught and handled by a user
mode routine). If not, you may have to fork over the bucks for the
new WCS and the microcode.
/Kevin Carosso allegra!scgvaxd!engvax!kvc
Hughes Aircraft Co.
>From fortune!amd70!nbires!hao!pag Sat Feb 25 04:24:53 1984
>Received: by fortune.UUCP; Sat, 25 Feb 84 02:48 PST
>Received: by nbires (4.12/3.7)
>id AA22990; Fri, 24 Feb 84 14:31:43 mst
>Date: Fri, 24 Feb 84 13:57:39 mst
>From: nbires!hao!pag (Peter Gross)
>Message-Id: <8402242057.AA01542@hao>
>Received: by hao (4.12/4.7)
>id AA01542; Fri, 24 Feb 84 13:57:39 mst
>To: nbires!amd70!fortune!wdl1!jme
>Subject: Re: Need help on G and H floating pt data types
>News-Path: hplabs!hpda!fortune!wdl1!jme
>References: <169@wdl1.UUCP>
I'm not too sure about the 780 (we have 750's), but you need to
have the WCS (writeable control store) option to get the extended
floating point ops. One of our people looked into it and decided
that you would then have to write assembly language programs to
use them (C and f77 don't generate code for them).
--peter gross
hao!pag