[Beowulf] g77 limits...

Robert G. Brown rgb at phy.duke.edu
Thu Feb 23 10:19:05 PST 2006


On Wed, 22 Feb 2006, Greg Lindahl wrote:

> On Wed, Feb 22, 2006 at 05:45:06PM -0500, Mark Hahn wrote:
>
>> these days, gcc does pretty well compared to commercial compilers, if you
>> take care to -march, -msse2 and other tweaks.  I can't see why this
>> pretty respectable code quality for c/c++/etc wouldn't also apply to
>> fortran.  even the ada and java people seem to be almost happy with
>> gcc's code-gen...
>
> Er, it depends. gcc is a pretty good compiler for integer, but it's
> way behind the curve for numerical code and peak options. For Fortran,
> there's also the issue that performance-sensitive use of intrinsics
> really needs them expanded and optimized in-line. I don't know if
> g95 or gfortran have implemented these yet, but I do see people showing
> relatively poor performance for those cases.
>
> -- greg
> (disclaimer: we sell compilers, but then again, we compare against
> gcc in a regular basis...)

Yeah, yours was the first compiler I suggested.  And I suspect/hope that
we end up getting at least one copy of your whole suite to build
against.  The software the project will distribute should certainly be
ABLE to build against gcc and g++ and g-fort*-whatever, but for people
who plan to use it in expensive high end resources, it is just silly to
give up even a very few percent in overall performance on many many
systems if it can be had for a few hundred dollars.

    rgb

-- 
Robert G. Brown	                       http://www.phy.duke.edu/~rgb/
Duke University Dept. of Physics, Box 90305
Durham, N.C. 27708-0305
Phone: 1-919-660-2567  Fax: 919-660-2525     email:rgb at phy.duke.edu





More information about the Beowulf mailing list