Re: [AMBER-Developers] Quick and CUDA in AmberTools

From: Gerald Monard <>
Date: Tue, 23 Feb 2016 20:14:45 +0100

On 02/23/2016 07:33 PM, Ross Walker wrote:
>> On Feb 23, 2016, at 8:57 AM, Gerald Monard <> wrote:
>> Hello,
>> I was going to send to all of you some details about the current status
>> of Quick in the next few days, but you were faster...
>> So, for some details:
>> On 02/23/2016 04:29 PM, Ross Walker wrote:
>>> Hi Gerald,
>>> I see commits for Quick in master. This looks cool. Note will it be built as part of the standard ./configure -cuda build?
>> Yes and no. For the moment, the AmberTools/src/Makefile contains a line
>> for building quick.cuda but it is commented out. My problem right now is
>> that one of the cuda files needs ~1.5 hour to compile! I was going to
>> request some help from Kennie's group and yours to see if we can speed
>> up compilation.
>> For the moment, only the serial and MPI versions of quick are built by
>> default. If you want quick.cuda, you need to "cd AmberTools/src/quick &&
>> make".
> Yeap - the slowest compiler on the planet strikes again. I can get this escalated directly to the right people at NVIDIA which will probably be quickest way of addressing it. Which file is it specifically? I'd need to put together a standalone repo - i.e. the minimum set files of files they would need to reproduce the problem and then I can get us some help.

The file is I can provide you with a tar file containing
the minimum required files. FYI, the compilation time is not that
surprising since "nvcc -E | wc -l" gives something like
130k lines. This is because the file contains many #include. I tried to
use nvcc -dc to do separate compilations, but there are also many
__inline__ and I haven't found any working solution yet.

>>> One thing to note is that pmemd.cuda from AMBER 16 onwards is going to need a minimum of NVCC v7.5 so it would probably be best if we just restrict the whole of AMBER + AmberTools to require a minimum of NVCC 7.5 for simplicity.
>> OK. I guess you worried about the cuda-5.0... I currently writing tests
>> and I'm checking that I can have the same results for cuda > 5. I don't
>> mind going directly to 7.5, that will save me some work.
> Cool - so if I change the configure script to test for 7.5 and later or quit and request one upgrades NVCC there would be no objects yes?

That should be OK for me, although I haven't tested 7.5 yet.

Btw, my compilation time problem comes also by the fact that ./configure
-cuda builds a config.h that requires nvcc to compile for many
architectures (5 gencode if I count correctly). Reducing this number
could also help since nvcc recompiles the same for each architecture
(and it's not parallel!). Maybe through configure?

All the best,


> All the best
> Ross
> /\
> \/
> |\oss Walker
> ---------------------------------------------------------
> | Associate Research Professor |
> | San Diego Supercomputer Center |
> | Adjunct Associate Professor |
> | Dept. of Chemistry and Biochemistry |
> | University of California San Diego |
> | NVIDIA Fellow |
> | | |
> | Tel: +1 858 822 0854 | EMail:- |
> ---------------------------------------------------------
> Note: Electronic Mail is not secure, has no guarantee of delivery, may not be read every day, and should not be used for urgent or sensitive issues.
> _______________________________________________
> AMBER-Developers mailing list

  Prof. Gerald MONARD
  SRSMC, Université de Lorraine, CNRS
  Boulevard des Aiguillettes B.P. 70239
  F-54506 Vandoeuvre-les-Nancy, FRANCE
  e-mail :
  tel.   : +33 (0)383.684.381
  fax    : +33 (0)383.684.371
  web    :
AMBER-Developers mailing list
Received on Tue Feb 23 2016 - 11:30:03 PST
Custom Search