CalculiX and PaStiX solver Windows version

22

Dropbox still works.

PS. These packages are already available in mingw64: petsc and slepc


23

Hi,

This looks really encouraging but is is very case specific ?

Would you be able to publish your benchmark case as I’m trying the exe’s and nothing seems to make any perceivable difference in runtimes between these and SPOOLES from http://www.dhondt.de/calculix_2.17.1_4win.zip.

Thanks

Paul

24

My very simple benchmark:

I may add that more complicated models containing commands such as: *PLASTIC, *PRE-TENSION, *CONTACT (~10), significantly sped up, the time needed to calculate them decreased from 9h to approximately 3h (PARDISO vs PaStiX).

1 Like
25

Hi, I tried running the Sandwich test example from M. Kraska with this version but it does not converge, while standard CCX windows version does. Anyone else has similar issues? Link to the example files: Sandwichtest.zip - Google Drive

26

Try with: set PASTIX_MIXED_PRECISION=0

27

That fixed the issue. Many thanks. Calculation time almost halved in my computer. Good job.
Are this settings documented or at least roughly explained anywhere?

28

See chapter 4 from: Thesis Peter Wauligmann: PaStiX + CalculiX

29

I added a pastix packages to msys2:

https://packages.msys2.org/base/mingw-w64-pastix

This is a base release and currently does not work with CalculiX 2.17.

After a little modification works with ccx, but sometimes incorrectly:

  1. Sometimes when there is a decrease in the calculation step.
  2. When using MIXED PRECISION, it does not work at all, “nan” error
30

There is an update of exodus extras for ccx 2.17, I attach a link to version 2.17 + exodus + pastix (for windows)

https://www.openaircraft.com/calculix_extras.html

2 Likes
31

Thank you so much @rafal.brzegowy !!

32

First of all, thank you very much for sharing.

  1. I have obtained the following warning during testing the program:
EXODUS: Warning: This code was compiled with exodus version 8.11,
          but was linked with exodus library version 8.12
          This is probably an error in the build process of this code.

Is this OK?

  1. What environmental variable settings are preferable for this version of CCX+PasTiX? I use the following set:
rem ==== CalculiX settings ========
set CCX_NPROC_RESULTS=%NUMBER_OF_PROCESSORS%   
set CCX_NPROC_STIFFNESS=%NUMBER_OF_PROCESSORS%
set OMP_NUM_THREADS=%NUMBER_OF_PROCESSORS%

rem ==== PasTiX settings ===========
set OPENBLAS_NUM_THREADS=1
set PASTIX_MIXED_PRECISION=1
set PASTIX_ORDERING=0
set PASTIX_SCHEDULER=0
1 Like
33

This Exodus version allow to get results in EXO format compatible con Paraview? How could we change the format of the results to exo and not .frd?

1 Like
34 
ccx.exe -i inp_file_name -o exo

mentioned at the CalculiX-extra web page

1 Like
35

  1. I think is OK, but I will rebuild this version of ccx in a few days. This ccx is extremely large due to a bug in starpu: Significant difference in starpu package size: 193MB 1.3.8-2 vs. 6.27MB 1.3.8-1 · Issue #9360 · msys2/MINGW-packages · GitHub

  2. OK

2 Likes
36

New version, without bug in starpu (smaller size):

4 Likes
37

ccxv
I got this message.

1 Like
38

hi,

i have no trouble downloading by virus detection, run on Windows 10 64bit latest updates with built in Defender & Malwarebytes protections.

also i can confrm CaculiX PaStix ran faster than MKL Pardiso on AMD processor, the ratio is about 5/6 respectively. tested on multipart contact analysis with elastic material (attached), quadratic tetrahedral element meshing, setup and running inside PrePoMax.

best.

MultipartContactFrictionAnalysisElastic_CCX_PaStix
MultipartContactFrictionAnalysisElastic_CCX_PaStix1005×795 134 KB

*edited/updates

in case of bolt contact with oversize holes, PaStix solver is more handle easely to convergence than MKL Pardiso. the ratio is about 2/3 respectively.

MultipartBearingContactFrictionAnalysisElastic_CCX_PaStix
MultipartBearingContactFrictionAnalysisElastic_CCX_PaStix1005×795 71.2 KB

2 Likes
39

Another day, It could be downloaded.
Then another program, I have to turn off real-time protection to download.

40

Hi !

Firstly, thank you so much for your job on this solver. I tested it and it work perfectly !

But I have a problem with large mesh (1million elements). the process seems to go well (processors work at 100%) but I receive a job failed message in the end.
Did you experiment the same issue with large mesh ?

Best regards,

1 Like
41

Me too. 2,000,000 node problem works with MKL Pardiso 500,000 works and 1,000,000 doesn’t with Pastix.

1 Like