On Mon, Feb 16, 2009 at 3:33 PM, Kirk, Benjamin (JSC-EG)
<[email protected]> wrote:
>
>>> Wait, there's a bug in icc?  I didn't realize we were getting bad
>>> output from vsmoother; I just assumed that you'd turned off vsmoother
>>> optimization because it was excruciatingly slow.  ;-)
>>
>> I'm interested in knowing as well.  I saw that note the other day and
>> just figured that there was actually a subtle bug in vsmoother
>> somewhere that was exposed with higher optimization levels.
>>
>> Ben: Do you know that it's actually a bug in icc?
>
> I recollect seeing this with other codes too, which suggests it is an
> optimization bug in icc.  I've explicitly included John since I bet they
> have seen this at TACC as well.

Yes, although it has mostly been in the context of multi-thousand-line
Fortran files where we see Intel choke and/or optimize forever.  Of
course, at that level the language itself has probably all been
converted to some intermediate compiler language and so C vs. Fortran
distinction is irrelevant.

> John put the --disable-vsmoother in there to work around icc with

I did?  :-)  I think Derek may have been working on getting Larisa's
code ported, so I was just making it a little easier for him to keep
developing without bothering anyone else's compiles.

> optimization, but I use icc so infrequently I was always forgetting the
> option.  Invariably, I'd run ./configure ; make and then get sidetracked -
> coming back much later and icc would still be churning on that file. So I
> put in the compiler pragma.

Seems like a reasonable work-around...  In a compiler release or two
we can try again with full optimization.

-- 
John

------------------------------------------------------------------------------
Open Source Business Conference (OSBC), March 24-25, 2009, San Francisco, CA
-OSBC tackles the biggest issue in open source: Open Sourcing the Enterprise
-Strategies to boost innovation and cut costs with open source participation
-Receive a $600 discount off the registration fee with the source code: SFAD
http://p.sf.net/sfu/XcvMzF8H
_______________________________________________
Libmesh-devel mailing list
[email protected]
https://lists.sourceforge.net/lists/listinfo/libmesh-devel

Reply via email to