This is the mail archive of the fortran@gcc.gnu.org mailing list for the GNU Fortran project.

Index Nav:	[Date Index] [Subject Index] [Author Index] [Thread Index]
Message Nav:	[Date Prev] [Date Next]	[Thread Prev] [Thread Next]
Other format:	[Raw text]

Re: Is real(16) available on x86 platforms?

From: Tim Prince <tprince at myrealbox dot com>
To: Erik Schnetter <schnetter at cct dot lsu dot edu>
Cc: Dominique Dhumieres <dominiq at lps dot ens dot fr>, sgk at troutmask dot apl dot washington dot edu, fortran at gcc dot gnu dot org
Date: Sun, 14 Jan 2007 12:06:04 -0800
Subject: Re: Is real(16) available on x86 platforms?
References: <20070114161612.DE5945BB9A@mailhost.lps.ens.fr> <B7E82124-905C-4B5F-B00E-7AF97E9FF24B@cct.lsu.edu>

Erik Schnetter wrote:

On Jan 14, 2007, at 10:16:12, Dominique Dhumieres wrote:
Your question is sort of wrong.
OK! Is 128-bit floating point type supported on AMD64
running the FX's x86_64-unknown-linux-gnu?
More precisely, I was trying to test a piece of code using
real(16) executing (not working) on a G5 under OSX 10.3 (PPC)
on an AMD64 box under linux where I got the error.
I was a bit surprised because as far as a know real(16) floats
are not supported by the G5 hardware, but a some software
emulation (gmp/mpfr?). I was just expecting the same for AMD64,
why shouldn't it be?
I think real(16) is implemented on PPC by combining two real(8) numbers suitably. That is slower than a hardware-supported implementation, but gives you the accuracy.

Not as much accuracy or range as IEEE standard requires for 128-bit format. More precision but less range than real(10). You could call it a hardware-supported implementation, as it takes advantage of the characteristics of fused multiply-add.

References:
- Re: Is real(16) available on x86 platforms?
  - From: Dominique Dhumieres
- Re: Is real(16) available on x86 platforms?
  - From: Erik Schnetter

Index Nav:	[Date Index] [Subject Index] [Author Index] [Thread Index]
Message Nav:	[Date Prev] [Date Next]	[Thread Prev] [Thread Next]