search

ARudik / feelpp

Automatically exported from code.google.com/p/feelpp
0 stars 0 forks source link

memory leak in backend->solve() #7

Open GoogleCodeExporter opened 9 years ago

GoogleCodeExporter commented 9 years ago 
What steps will reproduce the problem?
1. destroy and re-create the backend at each time iteration

memory grows after each delete/construct

VD: create a testcase to reproduce

Original issue reported on code.google.com by christop...@feelpp.org on 12 Mar 2012 at 4:12

GoogleCodeExporter commented 9 years ago

Original comment by christop...@feelpp.org on 15 Mar 2012 at 10:06

GoogleCodeExporter commented 9 years ago 
Test case created and reproduce the problem in :
fsi/applications/levelset/laplacian

Original comment by vincent....@gmail.com on 15 Mar 2012 at 3:16

GoogleCodeExporter commented 9 years ago 
Could you put it in the testsuite for example testsuite/feeldiscr ?
that way this test will run and show up in the CDash.

thanks a lot for your help. I will try to have a look next week

Original comment by christop...@feelpp.org on 15 Mar 2012 at 3:38

GoogleCodeExporter commented 9 years ago

Original comment by christop...@feelpp.org on 18 Mar 2012 at 2:51

GoogleCodeExporter commented 9 years ago 
I found indeed a memory leak in the preconditioner framework. After checking 
with tcmalloc there does not seem to be a memory leak anymore in 
test_solve_leak.cpp with the backends. I have done up to 2000 iterations 
without a change in the memory consumption.

However note that I have commented out the codes before and after the backend 
calls.
I am now testing when reverting this change and see if there are memory leaks 
elsewhere.

Original comment by christop...@feelpp.org on 3 Apr 2012 at 8:25

GoogleCodeExporter commented 9 years ago 
Thanks for the fix !

It looks better but I still have a small increase of the memory.
It is more obvious with umfpack preconditioner but I can see it without.

for example if I run the test with :
./test_solve_leak --hsize=0.04 --pc-factor-mat-solver-package-type=umfpack
Some "solve" call with recreated backend do not free all the memory.
It doesn't happen evey iteration ...

(in my code with a big matrix, the memory leak is still obvious)

Original comment by vincent....@gmail.com on 5 Apr 2012 at 5:09

GoogleCodeExporter commented 9 years ago

Original comment by christop...@feelpp.org on 8 May 2012 at 9:59

GoogleCodeExporter commented 9 years ago 
is it still leaking ?

Original comment by christop...@feelpp.org on 7 Dec 2012 at 8:49