Speedup in recent versions (FDS 6.7.9)

104 views
Skip to first unread message

Ed Wynn

unread,
Sep 6, 2022, 2:59:11 PM9/6/22
to FDS and Smokeview Discussions
Hi All.

I thought I would mention that I have seen good reductions in runtime, moving from 6.7.7 to 6.7.9 -- specifically the (main) version without OpenMP. In a real-world case, the speedup was more than the 20% suggested in the 6.7.8 Release Notes -- more like 30%.

And, while I have that positive thing to say, I also thank the developers (and other contributors) for the FDS/Smokeview package in general. It is reliable and does a good job. Thanks!

Ed

Kevin McGrattan

unread,
Sep 6, 2022, 5:10:33 PM9/6/22
to fds...@googlegroups.com
Yes, OpenMP directives within the source code degrade loop optimization, meaning that the compiler cannot fully optimize the many 3D loops within the code if there is a chance that the end user might invoke more than one OpenMP thread. So we decided to completely turn off OpenMP unless the end user calls for it. 

I rarely use OpenMP. I get much more speed up running multiple meshes on multiple cores. 

Franco Marra

unread,
Sep 13, 2022, 7:13:59 AM9/13/22
to FDS and Smokeview Discussions
Dear Kevin,

maybe it could be useful to not suppress the date and time output on the CHID.out file when MPI processes are more than 32, as reported in the User Guide. As an example, the output line could be like:

Time Step:  17200  August  6, 2022  11:42:46, Simulation Time:    119.57 s, Step Size: 0.00595 s

This would be not too verbose. Just a suggestion.

I always thank you so much for your wonderful job.

Best regards,

Franco

Kevin McGrattan

unread,
Sep 13, 2022, 9:06:14 AM9/13/22
to fds...@googlegroups.com
There is a file called CHID_steps.csv that is output automatically and contains this information.

Franco Marra

unread,
Sep 13, 2022, 12:32:59 PM9/13/22
to FDS and Smokeview Discussions
Dear Kevin,

I am very sorry!
This is indicated near the end of the User Guide, I have not read this part carefully enough. My double mistake (the first was not to think you must have already provided a mechanism for this!).

Thanks for your kind reply.

Best regards,

Franco

Kevin McGrattan

unread,
Sep 13, 2022, 1:30:01 PM9/13/22
to fds...@googlegroups.com
I did not do this. This was a contribution from another user who liked the short run summary file. I don't even remember where we documented this file. In any event, there are only a few .csv files that get created in an FDS run, and this is one of them.
Reply all
Reply to author
Forward
0 new messages