Forum Message

 

 

We have moved the forum to https://forum.ngsolve.org . This is an archived version of the topics until 05/05/23. All the topics were moved to the new forum and conversations can be continued there. This forum is just kept as legacy to not invalidate old links. If you want to continue a conversation just look for the topic in the new forum.

Notice

The forum is in read only mode.

sometimes, repeatably, continuously allocates memory when calling mat.Inverse

More
2 years 10 months ago #3806 by place
new user on linux.

trying tutorials, all was fine until "navierstokes.py" which always consumes all memory and eventually gets terminated by system oom.

the line causing this is
Code:
inv_stokes = a.mat.Inverse(X.FreeDofs())

if i rem it out.

Code:
# solve Stokes problem for initial conditions: #inv_stokes = a.mat.Inverse(X.FreeDofs()) res = f.vec.CreateVector() res.data = f.vec - a.mat*gfu.vec #gfu.vec.data += inv_stokes * res

it then works, even with another call to Inverse, on another matrix, later on

any ideas why, no messages etc and i can't find anything so am stuck.

BTW this is using the PPA pre-compiled binaries, i am looking at making an appimage without compilation, so i may have something wrong/missing with my installation, but i think this should work.

BTW2 my distro can't install debs and doesn't have anaconda.
More
2 years 10 months ago - 2 years 10 months ago #3820 by place
as a last resort ii tried adding; inverse="sparsecholesky"

and it now works, that is, not trying to use infinite memory.
Last edit: 2 years 10 months ago by place.
More
2 years 10 months ago #3821 by christopher
Hi,
my first guess is that there is something with umfpack, if you set inverse="umfpack" directly does it happen? I cannot reproduce it locally, what version of suitsparse is installed on your system?
Best
Christopher
The following user(s) said Thank You: place
More
2 years 10 months ago - 2 years 10 months ago #3824 by place
hi,

"umfpack" didn't result in any change, however the call is now just never returning, not taking all memory, much more friendly. i updated to a slightly newer version.

haven't heard of "suitsparse", didn't come across a reference in the required packages in the ppa.

i'll check/install and report what happens

seems to be able to use cuda, so gpu acc. on nvidia, nice. i've been hacking on vulkan compute, maybe that's on its way, i have an AMD gpu so this might not do much ATM.

thanks again.
Last edit: 2 years 10 months ago by place. Reason: spelling
More
2 years 10 months ago #3826 by place
AFAICT "libamd2" on Focal (20,04) is build from the code in suitsparse 5.7.1 Feb 20, 2020.
Time to create page: 0.107 seconds