[Wien] ifx OMP Problem in lapw0 caused segmentation fault during runtime

[Fecher, Gerhard]

Happy new year,
as mentioned last year I received a segmentation fault from lapw0 during runtime.
There was no error during compilation when using ifx with the options:
current:FOPT:-free -w -O3 -xHost -fp-model=precise -DINTEL_VML -traceback -assume buffered_io -I$(MKLROOT)

The segmentation fault disapeared after changing lapw0. In paricular, I commented the !$omp directives after
the older comment in line 1649:
! jdoumont 30/7/20 omp statements cause floating invalid/segmentation error (big case) during runtime???
lines 1649-1651 changed to:
!! !$omp parallel do reduction(+:rhopw00,cwk,cvout) &
!! !$omp& private(jrj1,arg,BESpw,bes,lm1,L,M,LMMULT1,YKA,TC) &
!! !$omp& schedule(guided)

Note the error did not appear when compiled with ifort, however, 
it seems that the problem was arising already earlier, see in the fiel changes in SRC_lapw0:
* 2 new omp loops (in lapw0.F) caused issues for me: the loops where SHPBES is calculated;
  I'm not sure what exactly the issue is, but multiple subroutines are called
  inside the loop. Some of them using shared variables from modules.
  Maybe some data is shared that should be private?
  I commented out the omp satements and added a comment. with my name 'jdoumont'

Actally the other mentoned problems by 'jdoumont (e.g.: lines 1369 ff.) had no effect.

I did not check further, what the problem is in detail,
however, lapw0 is running already without the changes suggested by Laurence in charge.f
Maybe the errors appear only in a particular environment.


Some additional notes:
I also tested succesfully -xAVX2 instead of xHost
>From my test there is, compared to ifort, no loss of speed when using ifx with the above given compiler switches
I will perform some benchmarks after our Cluster is updated, actually I use only an older desktop computer
 
Ciao
Gerhard

DEEP THOUGHT in D. Adams; Hitchhikers Guide to the Galaxy:
"I think the problem, to be quite honest with you,
is that you have never actually known what the question is."

====================================
Dr. Gerhard H. Fecher
Institut of Physics
Johannes Gutenberg - University
55099 Mainz
________________________________________