GCMC issues with GFortran and openMP (CASS V1.2)

  • ozzie787
  • ozzie787's Avatar Topic Author
  • Offline
  • New Member
  • New Member
More
7 years 9 months ago #172 by ozzie787
Hello forum,

I have been running the Nitrogen GCMC example to test Cassandra 1.2 and noticed an issue with reproducing the result provided with the example when using a GFortran compiled version of the code with openMP.

While both Intel and serial GFortran versions reproduce the result, GFortran with openMP is miles off (See attached graph plot.jpg). The graph is from an Ubuntu system but this is also seen on OS X where deletion only seems to occur (See plot2.png).




Running this example with CASS V1.1 does not display this abnormality with any compiler/openMP combination.

For now, I am just highlighting what I have observed in case it is not an isolated incident.

Regards,

Stephen

OS's/Compiler(s):
Ubuntu 15.04: gfortran 4.9.2, ifort 13.0.0 20120731
OS X 10.11.5: gfortran (Homebrew gcc 6.1.0 --without-multilib)

Please Log in to join the conversation.

  • ozzie787
  • ozzie787's Avatar Topic Author
  • Offline
  • New Member
  • New Member
More
7 years 9 months ago #173 by ozzie787
Apologies the forum software won't let me upload the image files of the graphs as allegedly a couple of hundred kb is too big, but if they are wanted they exist and are available upon request

Please Log in to join the conversation.

More
7 years 9 months ago #187 by ryangmullen
Hi Stephen,

A charge variable was not correctly allocated for openMP in the Ewald calculations. This was causing all translations, rotations, and insertions to be rejected. You can download the corrected code now.

Ryan

Please Log in to join the conversation.

Time to create page: 0.102 seconds