fds 6.7.8 crashes with fds_local on windows 64 for an immersed boundary case

[bphusted]

When using fds_local -o 1 example_5.fds FDS crashes. Just using fds example_5.fds works.

The error message is given below and an example file is attached.

The problem only appears on a windows pc. The mac os or a linux system do not show the same problem.

Starting FDS ...

MPI Process 0 started on BYG-A201-BPHU

Reading FDS input file ...

Fire Dynamics Simulator

Current Date : June 10, 2022 11:06:08
Revision : FDS6.7.8-0-gfbf3e11ee-release
Revision Date : Tue May 24 18:07:45 2022 -0400
Compiler : Intel ifort 2021.5.0
Compilation Date : Wed 05/25/2022 11:40 AM

MPI Enabled; Number of MPI Processes: 1
OpenMP Disabled

MPI version: 3.1
MPI library version: Intel(R) MPI Library 2021.4 for Windows* OS

Job TITLE : Test specified mass flux bc on sphere and mass conservation.
Job ID string : Example_5

forrtl: severe (157): Program Exception - access violation
Image PC Routine Line Source
impi.dll 00007FFF178ACB65 Unknown Unknown Unknown
impi.dll 00007FFF178B3A43 Unknown Unknown Unknown
impi.dll 00007FFF17B06506 Unknown Unknown Unknown

Example_5.txt

[marcosvanella]

Thank you Bjarne we'll have a look.
Glenn, can you run the case in your windows machine to reproduce the error? Thanks.

[mcgratta]

I'll test it on my computer.

[mcgratta]

This appears to be a problem with the fds_local script. I can run the job successfully like this

mpiexec -n 1 -localonly fds Example_5.fds

For the moment, use this command instead of fds_local.

[gforney]

using the -c option, the command
fds_local -o 1 -c Example_5.fds
generates the following
mpiexec -localonly -n 1 -env I_MPI_FABRICS=shm fds Example_5.fds

if I run the above line after removing -env I_MPI_FABRICS=shm , it works. kevin, do we need the I_MPI_FABRICS entry?

[gforney]

also, if I run using the following (which is what fds_local generates)
mpiexec -localonly -n 1 -env I_MPI_FABRICS=shm fds Example_5.fds
I get the same error

Job ID string : Example_5

forrtl: severe (157): Program Exception - access violation
Image PC Routine Line Source
impi.dll 00007FF8EF7E1711 Unknown Unknown Unknown

[marcosvanella]

Noted the same, yet the problem only shows up when invoking the mkl cluster solver. I'll test this in an isolated piece of code. Might need help with the windows compiling.On Jun 13, 2022 9:44 PM, gforney ***@***.***> wrote: also, if I run using the following (which is what fds_local generates) mpiexec -localonly -n 1 -env I_MPI_FABRICS=shm fds Example_5.fds I get the same error Job ID string : Example_5 forrtl: severe (157): Program Exception - access violation Image PC Routine Line Source impi.dll 00007FF8EF7E1711 Unknown Unknown Unknown —Reply to this email directly, view it on GitHub, or unsubscribe.You are receiving this because you were assigned.Message ID: ***@***.***>

[gforney]

Also I was concerned I did not include the correct Intel files in the bundle but I see the same problem when using fds I built (rather than fds from bunde) and mpiexec from the oneapi I installed.

[mcgratta]

I don't think we need the I_MPI_FABRICS setting. The -localonly option indicates that the job is to run on one computer and I assume that the fabric is set appropriately.

[gforney]

I will take it out when -localonly is used

[gforney]

try you case with nightly bundle
https://drive.google.com/drive/folders/1X-gRYGPGtcewgnNiNBuho3U8zDFVqFsC?usp=sharing

[bphusted]

I tried the nightly build. FDS no longer crashes, but there seems to be a problem with the parameter -o
Regardless of what I set it to, - it seems to use 4 openmp thread each time. Please see below, I asked for two openmp, but got four.

C:\Users\bphu\Documents\summer_school_2022\Examples_Marcos>fds_local -o 2 Example_5.fds

Starting FDS ...

MPI Process 0 started on BYG-A201-BPHU

Reading FDS input file ...

Fire Dynamics Simulator

Current Date : June 14, 2022 19:58:05
Revision : FDS6.7.8-48-g8cfd1be80-nightly
Revision Date : Sun Jun 12 09:18:52 2022 -0400
Compiler :
Compilation Date : Tue 06/14/2022 11:50 AM

MPI Enabled; Number of MPI Processes: 1
OpenMP Enabled; Number of OpenMP Threads: 4

MPI version: 3.1
MPI library version: Intel(R) MPI Library 2021.4 for Windows* OS

Job TITLE : Test specified mass flux bc on sphere and mass conservation.
Job ID string : Example_5

Time Step: 1, Simulation Time: 0.002 s

[gforney]

I'll take a look

[mcgratta]

I have the same problem. The -o option does not give me the right number of OpenMP threads.

[mcgratta]

The -o option is ignored, and whatever OMP_NUM_THREADS has been set is used.

[gforney]

I had typo in the variable used to specify the number of openmp threads , bundle at link below should work now. let me know if you still see a problem. thanks for pointing out the problems
https://drive.google.com/drive/folders/1X-gRYGPGtcewgnNiNBuho3U8zDFVqFsC?usp=sharing

[bphusted]

Thanks. I have tested it, - also with different combinations with -p and -o and I did not find any problems, so the case can be closed