Re: Congrats on G95 LAPACK Success! (fwd)
"Andy Vaught" <[email protected]> Thu, 21 Aug 2003 20:17:32 -0700 (MST)
| Newsgroups | gmane.comp.compilers.g95 |
|---|---|
| Message-ID | <Pine.LNX.4.21.0308212016250.852-100000@mjolnir> |
> Hello all,
>
> My name is Tony Awtrey and I've been working on a US Government
> sponsored study on using Open Source software to reduce life-cycle
> support costs of the simulators used for training. One of the points in
> the report is recommendations on porting older Fortran code used in many
> of the simulators to modern, open platforms.
>
> The last time I researched this area in December of 2002, I tailored a
> set of 15 linear algebra benchmarks written in Fortran 77 to test the
> affect of Fortran to C conversions using 'f2c' (Result: Doubled the SLOC
> count, ruined maintainability and reduced performance 7%-10%). I also
> took the opportunity to test the Intel compilers against the GNU
> compilers to satisfy my own curiosity, but that really isn't going to be
> part of the report we're doing.
>
> When I visited the G95 SourceForge site today I saw you all have been
> very busy since I visited last year. Congrats on the successful LAPACK
> compile tests! I have some other good news as well. My benchmarks were
> scripted and checked into cvs last year, so I checked them out again,
> added a G95 directory and ran your compiler against my tests...
>
> I saved the relevant test data to a PDF here:
>
> http://www.awtrey.com/common/compiler-benchmarks.pdf
>
> Bottom line, your brand new G95 compiler is 7% faster than G77 on the
> same code in the same machine! Here is the link to N.Tajima's FORTRAN
> benchmark that I used in my comparison:
>
> http://serv.apphy.fukui-u.ac.jp/~tajima/bench/source_e.html
>
> Thanks for the hard work and I hope our report can get additional
> funding directed to some of the open source development tool projects
> like yours and the GNU Compiler Collection as a whole.
Thanks for your kind mail. Although I suspect the difference with g77
may be that g77 is doing some very inefficient things as a opposed to g95
doing them efficiently. The version of the backend I am using has a very
broken optimization pass, and I haven't seen much reason to update to a
new version. (It'll eventually happen, though).
Thanks again,
Andy
Oh, it looks like the URL should be:
http://www.awtrey.com/common/compiler-performance.pdf
-------------------------------------------------------
This SF.net email is sponsored by: VM Ware
With VMware you can run multiple operating systems on a single machine.
WITHOUT REBOOTING! Mix Linux / Windows / Novell virtual machines
at the same time. Free trial click here:http://www.vmware.com/wl/offer/358/0