Home > Archive > Fortran > December 2005 > high-precision type
You are viewing an archived Text-only version of the thread.
To view this thread in it's original format and/or if you want to reply to
this thread please [click here]
| Author |
high-precision type
|
|
| wakun@wakun.com 2005-12-13, 8:14 am |
| hi there,
I am new to fortran. Someone please tell me the difference of Real*8
and Real*16? Does fortran support quadruple-precision type?
Thanks
| |
| Steven G. Kargl 2005-12-13, 9:57 pm |
| In article <WAKnf.5384$w7.1864@bignews7.bellsouth.net>,
Charles Russell <SPAMworFREEwor@bellsouth.net> writes:
> glen herrmannsfeldt wrote:
>
> Any prospect of this feature appearing in cheap PC hardware in the near
> future? Or of software emulation in gfortran or g95?
gfortran has real(kind=4,8,10,16), which are single, double,
entended, quad precision. But, there must be native hardware
support for the various kinds. On my amd64 FreeBSD system,
I get kind=4, 8, and 10. On an HPUX system, I've seen reports
that all 4 kinds are available.
I doubt that gfortran will have software implementations anytime
soon. Although it's possible to use GMP/MPFR, it would be extremely
slow.
If you need quad precision, check out Alan Miller's big pond.
--
Steve
http://troutmask.apl.washington.edu/~kargl/
| |
| robert.corbett@sun.com 2005-12-15, 3:58 am |
| > Any prospect of this feature appearing in cheap PC hardware
> in the near future?
No. It is possible to build an FPU that implements a slow, but
much faster than software, version of quad using about 15%
more transistors than an FPU that implements just single and
double. The problem is that such an FPU will be slower on
single and double than if quad were not supported. No CPU
vendor would make such a tradeoff. Implementing quad using
a separate FPU would avoid slowing single and double, but then
the number of additional transistors and wires needed makes it
impractical. Cutting yields, and thus profits, for the sake of a
feature that has little use in general-purpose computing.
Games saved floating-point, but games don't need quad.
Bob Corbett
| |
|
| If we're talking games and floating point, we should be talking about
cell chips. Has anybody in the fortran community started thinking about
using them.
Joost
| |
|
| > Has anybody in the fortran community started thinking about using them.
Well, what I can say is that some (no, no, I won't give names :) gfortran
developpers have started thinking about them.
--
FX
|
|
|
|
|