Re: [AMBER-Developers] CUDA build fails for PBSA

From: Ray Luo <rluo.uci.edu>
Date: Mon, 13 Mar 2017 11:46:55 -0700

Cool! Ruxi is testing it ... I think we can at least make sure it
works under gfortran, clang, and intel compilers before release. We
don't have access to cray machines so can't test it.

Ray
--
Ray Luo, Ph.D.
Professor of Structural Biology/Biochemistry/Biophysics,
Chemical Physics, Biomedical Engineering, and Chemical Engineering
Department of Molecular Biology and Biochemistry
University of California, Irvine, CA 92697-3900
On Mon, Mar 13, 2017 at 6:16 AM, Daniel Roe <daniel.r.roe.gmail.com> wrote:
> I just pushed a potential fix. Seems to work well for me. Let me know
> if anyone sees something break.
>
> -Dan
>
> On Sat, Mar 11, 2017 at 4:03 PM, Ruxi Qi <ruxiq.uci.edu> wrote:
>> Hi Jason,
>>
>> I'll test pmemd CUDA flags with PBSA.
>>
>> Best,
>>
>> Ruxi
>>
>>
>> On 3/11/17 12:49, Jason Swails wrote:
>>> I saw the same thing as Dan.  I needed to add -lstdc++ to the linker line
>>> to resolve this.
>>>
>>> It seems either your gfortran build links the C++ libraries on its own or
>>> it doesn't.  If it does, then everything will work.  If it doesn't, you get
>>> the error unless you explicitly link libstdc++.  We saw exactly the same
>>> problem with pmemd.cuda in the early days and spent a fair bit of effort
>>> iterating over the available compilers and options and such before we
>>> finally got something that works.  Look at line 1406 of configure2 and
>>> you'll see fc_cxx_link_flag being added to the CUDA library.  Note that not
>>> all compilers use libstdc++, though (which is why fc_cxx_link_flag exists
>>> and changes based on your compiler choice).
>>>
>>> I *highly* recommend specifying flags in configure rather than hardcoding
>>> them in the Makefile.  In fact, I would recommend just trying the pmemd
>>> CUDA flags and see if that works.  If not, you may need to add another
>>> variable for PBSA's CUDA libraries.
>>>
>>> All the best,
>>> Jason
>>>
>>> On Sat, Mar 11, 2017 at 3:20 PM, Ray Luo <rluo.uci.edu> wrote:
>>>
>>>>     From Ruxi ...
>>>>     It's strange our gnu 4.8.2 with cuda-7.5 on gpu-1-1 passed the tests,
>>>> even I
>>>>     didn't use -lstdc++.
>>>>     Best,
>>>>     Ruxi
>>>>
>>>>     On Mar 11, 2017 11:59 AM, Ray Luo <rluo.uci.edu> wrote:
>>>>
>>>>     Thanks for spotting this! Will fix it soon.
>>>>     Ray
>>>> _______________________________________________
>>>> AMBER-Developers mailing list
>>>> AMBER-Developers.ambermd.org
>>>> http://lists.ambermd.org/mailman/listinfo/amber-developers
>>>>
>>>
>>>
>>
>>
>> _______________________________________________
>> AMBER-Developers mailing list
>> AMBER-Developers.ambermd.org
>> http://lists.ambermd.org/mailman/listinfo/amber-developers
>
>
>
> --
> -------------------------
> Daniel R. Roe
> Laboratory of Computational Biology
> National Institutes of Health, NHLBI
> 5635 Fishers Ln, Rm T900
> Rockville MD, 20852
> https://www.lobos.nih.gov/lcb
>
> _______________________________________________
> AMBER-Developers mailing list
> AMBER-Developers.ambermd.org
> http://lists.ambermd.org/mailman/listinfo/amber-developers
_______________________________________________
AMBER-Developers mailing list
AMBER-Developers.ambermd.org
http://lists.ambermd.org/mailman/listinfo/amber-developers
Received on Mon Mar 13 2017 - 12:00:03 PDT
Custom Search