[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]
Re: [f-cpu] Re: Floating-Point
> On Tue, Aug 21, 2001 at 12:09:37AM -0400, nicO wrote:
> The manual says:
>
> 00 32-bit
> 01 64-bit
> 1x unassigned
Shouldn't we have SRs (yeah I know) to define the sizes the same as for the
int unit? Then the software can choose the FP widths that it wants. Default
might be:
00 32bit
01 64 bit
10 80 bit (optional and I don't like binary numbers that have more than one 1
in them and if the registers are 128 bit anyway, why not just bump up the
resolution?)
11 128 bit (optional, but better than 80bit ;-)
That's four more SRs, but they would only have to be state-saved if they were
being used.
Alternatively, just use the Int SRs for the float width.
(we waste the 8 value - assuming we are using the defaults abd assuming
no-one wants 8-bit floats (well someone might!) - and 16 bit floats are of
dubious value at best, we would also have to re-define int types away if we
wanted >64 bit floats and 80bit floats would be out (but I don't mind that -
not if 128bit is avaliable) ) It depends on how many int-widths are likely to
be used by the a program segment using floats, I guess. Is a mix adding up
to four enough?
--------------------------------------------------------
Glenn Alexander - The man with no surname and a silly hat.
(B.Teach, B.Ed Major IT Education, University of Wollongong Australia)
(Now avaliable in China!)
http://members.ozemail.com.au/~glenalec (last update: 2001.07.29)
I use GNU/Linux: http://www.linux.org
from Debian: http://www.debian.org
and KDE : http://www.kde.org
--------------------------------------------------------
Fight software piracy. Use GNU! [ http://www.gnu.org ]
--------------------------------------------------------
The above message was bought to you by 'sigrot'
*************************************************************
To unsubscribe, send an e-mail to majordomo@seul.org with
unsubscribe f-cpu in the body. http://f-cpu.seul.org/