DALTON with OpenACC directive

[ankit7540]

Hi all,
Has anybody tried installation of DALTON with OpenACC directive. I tried once but it failed. I removed this directive from the setup line and moved on that time.
Since then I have not tried. I would like to ask other users if they have tried installation with OpenACC. If yes how and some details (if you have time). Any external links are also welcome.
Thanks.

PS-The system I am using has an AMD GPU with 3 GB of GDDR5 RAM.

[taylor]

Have you added code to Dalton yourself that uses OpenACC? I ask because I was not aware of any OpenACC code in Dalton. OpenMP, yes. But as I say I was not aware there was any unless you added it.

You might get more assistance if you posted a little more information, like your setup command line, compiler versions, and a more detailed error than "it failed".

Best regards
Pete
P.S. I am completely out of touch on the AMD hardware. If the path between CPU and GPU has high enough performance you should get some performance improvement using the GPU for matrix operations, which presumably can be done via the AMD-supplied libraries although I have no experience with these.

[taylor]

A further point. You are asking for a total of 20 excited states for N2 in an aug-cc-pVDZ basis. The notion that more than a few states (possibly the lowest in each symmetry) are of any physical meaning is extremely doubtful. I have posted so many times on this I refuse to do it again.

Further you are asking for six states in symmetry 8, which contains Sigma_u^- and Delta_u states. Is this really what you want? Can six such states be of interest?

Best regards
Pete

[bast]

dear Ankit,
currently there are no OpenACC directives in Dalton. The CMake variable
is a leftover from the Dalton/LSDalton development code repository split.
LSDalton can do OpenACC - however I am not sure this functionality has been released yet.
I will now remove the leftover code from Dalton repository.
best wishes,
radovan

[ankit7540]

To Prof. Taylor,
I am sorry, but I did not ask anything about N2. I think your second post here is meant for some other post.

Ankit

[ankit7540]

To Mr. Bast and Prof. Taylor,
Thank you for your kind reply.

I have a question.
In case I do some valid addition to code, and if proper compiler (supporting OpenACC ) is installed, then would it work. Or simply the directive will be neglected ??

Some information:
When I tried -openacc directive, I was using intel's compilers (2015). I expected that if intel compiler supported this directive then even if there is no addition to code, the compilation should be done without any optimization. However the compilation did not go well.

Later I learned that PGI compilers support this directive , but I do not know how to use PGI with DALTON. SO I did not try more.

-Ankit

[bast]

Dear Ankit,
Adding valid OpenACC directives should be understood by the compiler provided you turn the CMake variable on.
Please note that to avoid confusion I have removed the compiler flags in future releases of Dalton so if you port your
code to a later version of Dalton it might stop working. But it is then not difficult to reintroduce the CMake part
of it.
good luck,
radovan

[janus]

- There are no OpenACC directives in Dalton.
- We will release a bunch of GPU-accelerated code in LSDalton in the forthcoming release, but there's nothing in the current release (AFAIR).
- Intel doesn't support OpenACC (and they will never do so).
- As of yet, only PGI and Cray support OpenACC (CAPS did as well, but they went bankrupt). GNU has an early port out in the newest version, but only with support for the OpenACC 1.0 standard. However, they will continue to implement increasingly more, which is awesome.

/ Janus