[Nek5000-users] testing suite

nek5000-users at lists.mcs.anl.gov nek5000-users at lists.mcs.anl.gov
Fri Aug 24 08:58:09 CDT 2012


Hi Neks

I have a few questions about test suite in nekton (in nek5_svn/tests/).
Nek5000 is a part of CRESTA project and I need simple method for testing
new versions of the code on different machines. Tests in nek5_svn/tests/
are perfect, but I've got some problems:

1) genmap work perfect with pgi compiler, but with gnu compilers returns
segmentation fault for conj_ht test. It's due to not initialized key
array in subroutine spec_bis_conn. It would be good to add
-finit-local-zero flag for gfortran in maketools

2) tests in 2d_eigtest work fine for pgi, but fail for gfortran, since
the difference in the lambda error is bigger than 10^-6. It happens for
N=3 and 4 only; eig1.err:
GNU:
  2   19 0  5.7831860E+00  2.1065324E-02  1.6329577E-01 lambda
  3   34 0  5.7831860E+00  6.8588903E-04  1.3936160E-01 lambda
  4   61 0  5.7831860E+00  6.4127192E-07  2.6444060E-02 lambda
  5   84 0  5.7831860E+00  4.7561843E-08  3.3011994E-02 lambda
  6  121 0  5.7831860E+00  3.2211656E-10  1.1637760E-02 lambda
  7  178 0  5.7831860E+00  2.8581582E-12  1.6504072E-03 lambda
  8  243 0  5.7831860E+00  5.4534155E-13  4.8188495E-04 lambda
  9  314 0  5.7831860E+00  8.6153307E-14  2.9147672E-05 lambda
 10  400 0  5.7831860E+00  3.2018832E-12  6.8264261E-06 lambda
 11  478 0  5.7831860E+00  7.3994144E-12  4.4899419E-07 lambda
 12  614 0  5.7831860E+00  6.7972294E-12  4.0358872E-08 lambda
 13  715 0  5.7831860E+00  2.7814195E-11  2.0542709E-09 lambda
 14  829 0  5.7831860E+00  1.8316904E-11  1.1108244E-10 lambda
 15  883 0  5.7831860E+00  1.1938006E-11  3.0416639E-12 lambda
 16  900 0  5.7831860E+00  1.6848745E-12  5.3768924E-12 lambda
PGI:
  2   19 1  5.7831860E+00  2.1065324E-02  1.6329577E-01 lambda
  3   37 1  5.7831860E+00  6.8588903E-04  1.4437432E-01 lambda
  4   61 1  5.7831860E+00  6.4127192E-07  2.8477333E-02 lambda
  5   89 1  5.7831860E+00  4.7561947E-08  3.3011994E-02 lambda
  6  119 1  5.7831860E+00  3.2096015E-10  1.1637760E-02 lambda
  7  178 1  5.7831860E+00  2.7320368E-12  1.6504072E-03 lambda
  8  235 1  5.7831860E+00  1.0373924E-12  4.8188495E-04 lambda
  9  313 1  5.7831860E+00  5.7545080E-12  2.9147672E-05 lambda
 10  400 1  5.7831860E+00  1.0382806E-12  6.8264260E-06 lambda
 11  477 1  5.7831860E+00  2.2977176E-12  4.4899373E-07 lambda
 12  614 1  5.7831860E+00  2.7853275E-12  4.0358850E-08 lambda
 13  713 1  5.7831860E+00  1.4867219E-11  2.0542157E-09 lambda
 14  826 1  5.7831860E+00  1.4868107E-11  1.1256349E-10 lambda
 15  884 1  5.7831860E+00  1.1343815E-11  3.8686936E-12 lambda
 16  900 1  5.7831860E+00  5.1991300E-11  1.1914564E-11 lambda

Have you got the same result? Should the tolerance in Analysis.py be
bigger in this case? Moreover this test was failing for me with 

[Example 2d_eig/INT2: Serial-iter/err]...I couldn't find all the
requested value in the log file...
[Example 2d_eig/INT2: Serial-iter/err]... 2   ,  3   ,  4   ,  10  ,
were not found...
Example 2d_eig/INT2: Serial-iter/err : F 

due to single list 
value = [[' 2   ',19,2,6],
         [' 2   ',1.6329577E-01,1e-06,2],
         [' 3   ',37,2,6],
         [' 3   ',1.4437432E-01,1e-06,2],
         [' 4   ',61,2,6],
         [' 4   ',2.8477333E-02,1e-06,2],
         [' 10  ',400,2,6],
         [' 10  ',6.8264260E-06,1e-06,2]]

in Analysis.py. To get rid of this I had to divide it to 2 lists 
value = [[' 2   ',19,2,6],
         [' 3   ',37,2,6],
         [' 4   ',61,2,6],
         [' 10  ',400,2,6]]

value = [[' 2   ',1.6329577E-01,1e-06,2],
         [' 3   ',1.4437432E-01,1e-06,2],
         [' 4   ',2.8477333E-02,1e-06,2],
         [' 10  ',6.8264260E-06,1e-06,2]]

3) mhd-gpf (case without fast diagonalization method lfdm=0) fails both
for pgi and gnu compilers. It seems pressure does not converge giving
nan for gfortran. Cases with lfdm=1 give no errors. 

4) I found most of moab tests commented in Analysis.py. Does it mean
moab is not fully implemented yet?

5) I found testing scripts tuned for single environment (./ in the PATH,
moab and matlab locations hidden in the script), so I've modified them
slightly to make them easier to run for others. Is it possible to add my
version of scripts to the repository?

Regards

Adam





More information about the Nek5000-users mailing list