Maybe try (test PaStiX): set PASTIX_MIXED_PRECISION=0
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
and separately: set PASTIX_SCHEDULER=1
Previous links have expired so I am posting new ones, there are two versions in the archive:
Requires a PARDISO (mkl) library
Does not require PARDISO libraries (PARDISO cannot be used)
Both of these versions, as part of the tests, have the option to choose (can be added to: cmdStartup.bat from bConverged ):
set PASTIX_ORDERING=0
0 - Scotch, 1 - Metis
and:
set PASTIX_SCHEDULER=1
0 - Static, 1 - StarPU, 3 - Sequential, 2 - parsec (not working yet)
my patch of pastix.c:
// Set best PaStiX …
Info(?):
opened 01:05PM - 18 Sep 20 UTC
closed 08:46AM - 04 Nov 20 UTC
Hi @3rav ,
I will try to make a version without the need of cuda (at least a co… mpile time option).
You don't necessarily need parsec, it should work using StarPU or Static scheduling (didn't try this route for quite some time), but the performance will probably suffer.
_Originally posted by @Kabbone in https://github.com/Dhondtguido/PaStiX4CalculiX/issues/1#issuecomment-666996876_
Using StarPU instead of Static scheduling causes a very big drop in performance (Total PaStiX Time: 154.12 vs 9.49).
Please see attachments:
[StarPU_time.txt](https://github.com/Dhondtguido/PaStiX4CalculiX/files/5245812/StarPU_time.txt)
[Static_time.txt](https://github.com/Dhondtguido/PaStiX4CalculiX/files/5245813/Static_time.txt)
“Looking at the top of the thread, the best performance should be with dynamic (with the fix included in 6.3.1) or StarPU if your matrix is big enough. For small ones, static should probably be better.”
2 Likes