Re: [AMBER] ambertools 1.4, tests

From: Ben Roberts <roberts.qtp.ufl.edu>
Date: Mon, 3 May 2010 09:53:38 -0400

Hi MJ and Alan,

I thought I'd chime in here since I'm wrestling with the same things.

I think -g might do more than just turn off optimisations. It also often mucks about with the memory layout. Sigh - I hate Heisenbugs.

Like Alan, I upgraded to 11.1.088, and I'm running OS X 10.6.3. I see the same bugs in AmberTools that he did.

I'm willing to help diagnose the problem - time permitting. So we can coordinate our efforts, what might be most helpful for me to try first?

Ben

On 3/5/2010, at 6:56 a.m., Alan wrote:

> HI Mengjuei!
>
> Great it seems you're on the right way. I don't know much compiling, but
> when using '-g' (for debug), it also disable optimisations I guess. Would be
> the optimisations flag in the end that generates the differences in tests?
>
> Another question. Do you have pmemd.MPI working with openmpi and intel
> compiler on Mac?
>
> Cheers,
>
> Alan
>
> On Mon, May 3, 2010 at 11:02, Mengjuei Hsieh <mjhsieh.gmail.com> wrote:
>
>>> I can confirm that with Mac OS X 10.6.3/Intel 11.1.084 compiler
>>> (intel64), EPB indeed shows all zeros in that test (nab/pbdmp),
>>> accompanied by 3 other minor nab test (numerical) fails. So far I have
>>> no clue why.
>>
>> I have found a couple of bugs in my C code, but even if I fixed those,
>> I'd still need to use "icc -g" in order to pass the test...
>>
>> Sincerely,
>> --
>> Mengjuei
>>


_______________________________________________
AMBER mailing list
AMBER.ambermd.org
http://lists.ambermd.org/mailman/listinfo/amber
Received on Mon May 03 2010 - 07:00:03 PDT
Custom Search