Update to deal.II 9.2 - Githubissues

kronbichler / adaflo

adaflo, an adaptive finite element solver for incompressible fluid flow and two-phase flow

Other

24 stars 14 forks source link

Update to deal.II 9.2 #3

Closed peterrum closed 4 years ago

peterrum commented 4 years ago

I'll run the tests later.

peterrum commented 4 years ago

@kronbichler I have updated the code to deal.II 9.2. My experiences were:

running test out-of-source is not possible (I had to manually copy the input files to the build folder)
running tests takes quite long (I have not waited till completion and instead run everything in paralle with 40 processes and even so flow_past_cylinder and flow_past_square_cylinder took very long)
if I am doing something wrong in the last point and the tests are indeed much faster I would suggest to introduce CI
in some of the tests there are segmentation faults during calling the final destructors - I am not sure about the reason, e.g., mpirun -np 40 spurious_currents spurious_currents_ls.prm
there is not .clang-format file so that I could not format the changes

FYI @mschreter

kronbichler commented 4 years ago

Let us merge this now; I will add a suitable clang-format file later and look into the tests.

kronbichler commented 4 years ago

I did commits to fix those:

running test out-of-source is not possible (I had to manually copy the input files to the build folder) there is not .clang-format file so that I could not format the changes

Regarding the tests: These are known to be slow in debug mode but reasonably fast in release mode. That said, there were some spurious commits in the test folder resulting in too big computations. More of them could surely be somewhat reduced but I did not get around to doing it yet. The more annoying issue is that I get failures due to iteration counts of the iterative solver to be one or two off; that would require some more work. I have to check the MPI issue locally with debug mode later.

kronbichler commented 4 years ago

Another todo is that some tests, like the rising_bubble.cc, run with threads as well, variable set to -1. This leads to very poor performance on machines with many cores and paralle tests (I just verified that locally).

peterrum commented 4 years ago

Thanks @kronbichler!