Advice on paralllel builds

Problems with Dalton installation? Find answers or ask for help here
Post Reply
mikep219
Posts: 9
Joined: 06 Feb 2020, 17:00
First name(s): Mike
Last name(s): Pacey
Affiliation: Lancaster University
Country: United Kingdom

Advice on paralllel builds

Post by mikep219 » 06 Feb 2020, 17:20

[Standard caveat: I'm an HPC Sysadmin, not a Chemist]

One of our local Chemists has asked me to build and install Dalton on our cluster. I'm using Intel compilers with the following build options:

./setup --mpi --mkl=sequential

and some of the ctest checks are failing with the error:

----
WARNING: CC is not MPI parallelized!
For parallelization speedup, e.g. use parallel MKL

ERROR: CC not implemented for parallel calculations
----

If I'm reading that correctly, that's an expected error for this build type? Is there a minimum set of different parallel build types that ensure each type
of model being run gets the most out of parallelism? (I'm seeing MPI, OpenMP and MKL parallelism as options, and I'm wary that trying to build all three into a single binary might not be best approach)

taylor
Posts: 595
Joined: 15 Oct 2013, 05:37
First name(s): Peter
Middle name(s): Robert
Last name(s): Taylor
Affiliation: Tianjin University
Country: China

Re: Advice on paralllel builds

Post by taylor » 07 Feb 2020, 00:50

As far as I know there is no OpenMP parallelism in Dalton, so that possibility is not relevant. Quite a lot of the code for calculating molecular properties is MPI parallelized and exploiting that will give a speed-up for some calculations. However, there is, as the output indicates, no MPI functionality in the coupled-cluster (CC) part of the code. On the other hand, the program makes considerable use of matrix operations, so use of the threaded MKL can give performance benefits. This applies to the CC part as well as the other parts of the code.

Best regards
Pete
P.S. Sorry, I cannot help with your phosphorescence question.

mikep219
Posts: 9
Joined: 06 Feb 2020, 17:00
First name(s): Mike
Last name(s): Pacey
Affiliation: Lancaster University
Country: United Kingdom

Re: Advice on paralllel builds

Post by mikep219 » 07 Feb 2020, 15:53

Thanks for the advice. Looking at the help output for the setup script, I'm seeing an option "--omp" for OpenMP - I'm also seeing several omp pragma lines in the C++ source code files. Maybe this is a new feature for Dalton?

User avatar
magnus
Posts: 523
Joined: 27 Jun 2013, 16:32
First name(s): Jógvan Magnus
Middle name(s): Haugaard
Last name(s): Olsen
Affiliation: Aarhus University
Country: Denmark

Re: Advice on paralllel builds

Post by magnus » 07 Feb 2020, 16:02

It is mostly a leftover from a few years back when the Dalton and LSDalton (which supports OpenMP) codes shared the same repository and setup files. These will be cleaned for the upcoming release. Also the omp pragmas that you see are in parts of the code that are not in use and will likely be removed soon.

mikep219
Posts: 9
Joined: 06 Feb 2020, 17:00
First name(s): Mike
Last name(s): Pacey
Affiliation: Lancaster University
Country: United Kingdom

Re: Advice on paralllel builds

Post by mikep219 » 07 Feb 2020, 16:18

Thanks for explaining - that cuts down the number of test builds I need to try!

Post Reply

Who is online

Users browsing this forum: No registered users and 3 guests