[AMBER-Developers] Current parallel test failures in the tree

From: Ross Walker <ross.rosswalker.co.uk>
Date: Wed, 24 Mar 2010 12:15:01 -0700

Hi All,

I just ran the serial and parallel tests (mpirun -np 4) and found a few test
failures that we may want comment on:

ifort -V
Intel(R) Fortran Compiler for applications running on Intel(R) 64, Version
10.1 Build 20080801 Package ID: l_fc_p_10.1.018
Copyright (C) 1985-2008 Intel Corporation. All rights reserved.

mpi = mpich2-1.0.7

./configure intel
make
./configure -mpi intel
make parallel

The test_amber_serial.sh gives:

make: Target `test.serial' not remade because of errors.
480 file comparisons passed
3 file comparisons failed
0 tests experienced errors

The 3 failed tests here are just rounding differences so I have updated the
dacdif arguments to avoid this.

The test_amber_parallel.sh (with mpirun -np 4) gives:
make: Target `test.parallel2' not remade because of errors.
350 file comparisons passed
4 file comparisons failed

These test failures are far beyond simple roundoff. The failures are:

1) neb_gb_partial_01.out test/neb-testcases/neb_gb_partial
2) neb_gb_full_01.out test/neb-testcases/neb_gb_full
3) ti_decomp_1.out test/ti_decomp
4) mdout.aladip_ewald_ntb1_qmewald2 test/qmmm_DFTB/aladip_tip3p_ewaldpme

The neb runs are completely hosed and will need to be looked at ASAP. Here
is some example output:

possible FAILURE: check neb_gb_partial_01.out.dif
/home/rcw/cvs_checkouts/amber11/test/neb-testcases/neb_gb_partial
103,104c103,105
< NSTEP = 1 TIME(PS) = 0.00 TEMP(K) = 0.0 PRESS =
0.
< Etot = -31.979 EKtot = 0.001 EPtot =
-31.981

---
> vlimit exceeded for step      0; vmax = **********
>  NSTEP =        1   TIME(PS) =       0.00  TEMP(K) = 11446.4  PRESS =
0.
>  Etot   =       718.645  EKtot   =       750.626  EPtot      =
-31.981
116a118
> vlimit exceeded for step      1; vmax = **********
130a133
> vlimit exceeded for step      2; vmax = **********
144a148
> vlimit exceeded for step      3; vmax = **********
158a163
> vlimit exceeded for step      4; vmax = **********
etc
etc
The ti_decomp_1 also has issues:
possible FAILURE:  check ti_decomp_1.out.dif
/home/rcw/cvs_checkouts/amber11/test/ti_decomp
2c2
<           Amber 10 SANDER                              2008
>           Amber 11 SANDER                              2010
69c69
<      nstlim  =        10, nscm    = 999999999, nrespa  =         1
>      nstlim  =        10, nscm    =         0, nrespa  =         1
19499c19499
<  INTERNAL=        0.
>  INTERNAL=       -3.5243
19500c19500
<  VDWAALS =        0.  EEL     =       19.8810
>  VDWAALS =        0.0238  EEL     =      -21.0686
19503c19503
<  INTERNAL=        0.
>  INTERNAL=       -3.5243
19504c19504
<  VDWAALS =        0.  EEL     =       19.8810
>  VDWAALS =       -0.5234  EEL     =      -21.4545
19508c19508
<  VDWAALS =        0.  EEL     =        0.
>  VDWAALS =        0.5472  EEL     =        0.3859
19521c19521
< TDC      7        0.        0.        9.940487732
> TDC      1        0.        0.000041206       -0.011154263
19522c19522
< TDC     18        0.        0.        9.940487732
> TDC      2        0.        0.000048871       -0.045324314
19523c19523
<                     PRINT DECOMP - SIDECHAIN ENERGIES
> TDC      3        0.        0.000048871       -0.056511261
This is way beyond rounding errors, even with having values printed to way
too much precision for a test case in the output. Who is responsible for
ti_decomp these days? They should probably look at this ASAP.
The final failure, mdout.aladip_ewald_ntb1_qmewald2 is just minor rounding
differences so I have updated the dacdif arguments for this.
All the best
Ross
/\
\/
|\oss Walker
| Assistant Research Professor |
| San Diego Supercomputer Center |
| Tel: +1 858 822 0854 | EMail:- ross.rosswalker.co.uk | 
| http://www.rosswalker.co.uk | http://www.wmd-lab.org/ |
Note: Electronic Mail is not secure, has no guarantee of delivery, may not
be read every day, and should not be used for urgent or sensitive issues.  
_______________________________________________
AMBER-Developers mailing list
AMBER-Developers.ambermd.org
http://lists.ambermd.org/mailman/listinfo/amber-developers

Received on Wed Mar 24 2010 - 12:30:03 PDT