test fail with ifort 18.0.1, lots of wrong results

[Alfio Lazzaro]

Hi Yi Wang,
There are other posts in the forum where people are reporting problems with 
ifort 18 (see for example 

https://groups.google.com/d/msg/cp2k/X5M2j6_vOe0/37ITJTTjAQAJ
https://groups.google.com/d/msg/cp2k/LC8nGidw8bU/6uJvwjBsAAAJ

 ). From what I can understand, that version of the compiler is buggy. 
Therefore, I'm surprised that you are reporting that it is working with 
previous versions of CP2K..
Anyway, CP2K is not tested with this version of the intel compiler (we will 
do the tests in the near future). What I can suggest you is to use -O0 for 
compilation and see if the tests pass. Then you can try to increase the 
optimizations (you can also try -fp-model source). Let me know if you can 
find anything useful.

Best regards,

Alfio



Il giorno lunedì 26 febbraio 2018 10:01:22 UTC+1, wan... at gmail.com ha 
scritto:
>
> Hi,
>
> Using the default intel tool chain and my own previous tool chains, I 
> found the ifort 18.0.1 is compatible with cp2k-4.1 but not any later 
> versions.
>
> There would be a lot of wrong results, as well as segmentation faults.
>
> For example, in "Pimd/w512_pint_piglet.inp.out":
>  PINT| Total energy =  : ref = 74.229316490453456 new = 74.1215685331465
>  relative error :   1.45366537e-03 >  numerical tolerance = 1e-11
>
> Some error appear to cause a very long running time, thus I cannot even 
> fully run all the test.
>
> The numerical error appear to be around the same values from cp2k-5.1 to 
> the current truck (18277).
>
> So I guess some modification between 4.1 to 5.1 caused the effects.
>
> Yi Wang
> Nanjing University of Sci & Tech
>
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <https://lists.cp2k.org/archives/cp2k-user/attachments/20180227/f153ba8f/attachment.htm>