CP2K 5.1 / IC 2018 regtests

155 views
Skip to first unread message

drahomirko

unread,
Dec 4, 2017, 7:43:57 AM12/4/17
to cp2k

Sirs,

I am trying to compile CP2K 5.1 with IC/MKL 2018u1. Compilation runs alright, however I am getting number errors and runtime fails during regtesting.  The errors are (perhaps) due to compiler as they read:

Error in `/calcs/cp2k-5.1/regtesting/Linux-x86-64-intel/sopt/../../..//exe/Linux-x86-64-intel/cp2k.sopt': corrupted double-linked list: 0x000000000c9c43d0 ***

*** Error in `/calcs/cp2k-5.1/regtesting/Linux-x86-64-intel/sopt/../../..//exe/Linux-x86-64-intel/cp2k.sopt': double free or corruption (fasttop): 0x000000000c7db370 ***

*** Error in `/calcs/cp2k-5.1/regtesting/Linux-x86-64-intel/sopt/../../..//exe/Linux-x86-64-intel/cp2k.sopt': free(): corrupted unsorted chunks: 

followed by memory backtrace and runtime fail afterwards. 


I am only compiling basic functionality *sopt version at the time to get the compiler settings right.  The makefile I am attaching was working fine with CP2K 4.1 / IC 2017u4

Suggestions are very welcome!

Thank you.

Drahomir


Linux-x86-64-intel.sopt

drahomirko

unread,
Dec 4, 2017, 9:19:37 AM12/4/17
to cp2k

To follow up, I am getting same errors with the CP2K 4.1 that was working correctly with IC 2017, so this issue I suspect is due to compiler. Any ideas how to fix this?

Thanx

Drahomir

Alfio Lazzaro

unread,
Dec 5, 2017, 4:59:54 AM12/5/17
to cp2k
Hi Drahomir,
well, I don't have a clear solution to your problem, therefore I can give you only some suggestions...
First of all, I've checked in the CP2K dashboard and there are not tests based on IC 2018. I see that you are using SOPT version. The corresponding test with IC 2017 on the dashboard is:


(on top of the log you can find the arch file that is used in the test).
I see the the arch file is pretty similar to yours (note that you don't need -D__INTEL and OBJECTS_ARCHITECTURE = machine_intel.o anymore).
A first suggestion would be to use -O0 -g (and remove all other options) and see if the code runs.
It would be also interesting to run under gdb and see where it is crashing.

Alfio

drahomirko

unread,
Dec 5, 2017, 7:10:10 AM12/5/17
to cp2k
Hi Alfio,

thanx for suggestions. I have cp2k 5.1 working with IC 2017 alright. No failed tests and 3 wrong results reported due to numerical accuracy (10^-10) which is alright as well. I only posted *sopt because it was fastest way to check on compiler. Already spent day and half on StackOverflow to get the solution, but all the posts there are in regard to change the code routines. In the meantime I will try to make it work with IC 2018 and in case I am successful I will post solution. 

Alfio Lazzaro

unread,
Dec 5, 2017, 8:01:37 AM12/5/17
to cp2k
Thanks! Actually, I see that there are some discussions on the Fortran group about ICC2018:


ICC2018 is buggy and Intel is working on it...

Alfio
Reply all
Reply to author
Forward
0 new messages