On Sat, May 16, 2009 at 08:52:07PM -0500, Rahul Nabar wrote:
> On Sat, May 16, 2009 at 8:37 PM, Greg Lindahl <lind...@pbm.com> wrote:
> > PathScale was always really good on VASP, by the way,
>
> It always amazes me the performance specificity of
> Compiler-CPU-Application combination. Might have to do with that I
> know nothing very much about compiler nor CPU design! :) Almost black
> magic. Just like compiler flags and optimization levels.
Yes compiler flags are a tangle.....
Do look at the "pathopt" tool that Pathscale has.
It is possible to set up a test case that the tool
can then walk through the tangle of compiler flags
and find the best timing and or find cases where a
compiler flag appears to result in a bad (different) answer.
Also with the Pathscale compiler explore the various "keep"
flags. By keeping the intermediate files it is possible to
see the comment-annotations inserted by the compiler.
The annotations are interesting in that the ability
to optimize or not optimize various language structures
is annotated with reasons. By profiling a code and
looking at these annotations in the hot spots very notable
code improvements can result with often modest code
changes.
Other compiler vendors have their tools and tricks too.... By working
with multiple compilers and taking advantage of the strengths of each
compiler important portability and correctness code improvements are
possible.
--
T o m M i t c h e l l
Found me a new hat, now what?
_______________________________________________
Beowulf mailing list, Beowulf@beowulf.org sponsored by Penguin Computing
To change your subscription (digest mode or unsubscribe) visit
http://www.beowulf.org/mailman/listinfo/beowulf
