[gmx-developers] setting GMX_NB_GENERIC fails regression tests

[Mark Abraham]

Hi,

Today I tried GROMACS 4.0.5 on a version of the regression tests that I 
checked out from git last week. Using optimized loops, the kernel tests 
passed. With environment variable GMX_NB_GENERIC set to 1, most kernel 
tests failed. The 15 working tests were kernel0[123]0 and 
kernel[123][0123]0 - i.e. all those without solvent optimization. All 
failures were found in checkpot.out, and all were with the long-ranged 
Coulomb component (and thus the total potential too). Various patterns 
existed:

[mxa224 at xe kernel]$ grep Coul */checkpot.out
kernel101/checkpot.out:Coulomb (LR)     step   0:      -66.0101,  step 
  0:     -1279.84
kernel102/checkpot.out:Coulomb (LR)     step   0:      -7582.15,  step 
  0:  1.08249e+06
kernel103/checkpot.out:Coulomb (LR)     step   0:       63.6915,  step 
  0:      -71.726
kernel104/checkpot.out:Coulomb (LR)     step   0:      -6869.62,  step 
  0:            0
kernel111/checkpot.out:Coulomb (LR)     step   0:       110.958,  step 
  0:      7178.53
kernel112/checkpot.out:Coulomb (LR)     step   0:      -7582.15,  step 
  0:  1.08249e+06
kernel113/checkpot.out:Coulomb (LR)     step   0:       41.6662,  step 
  0:      106.611
kernel114/checkpot.out:Coulomb (LR)     step   0:      -6869.62,  step 
  0:            0
kernel121/checkpot.out:Coulomb (LR)     step   0:       130.947,  step 
  0:       6066.3
kernel122/checkpot.out:Coulomb (LR)     step   0:      -7582.16,  step 
  0:  1.08249e+06
kernel123/checkpot.out:Coulomb (LR)     step   0:       41.6662,  step 
  0:      106.611
kernel124/checkpot.out:Coulomb (LR)     step   0:      -6869.61,  step 
  0:            0
kernel131/checkpot.out:Coulomb (LR)     step   0:       113.182,  step 
  0:      6625.11
kernel132/checkpot.out:Coulomb (LR)     step   0:      -7582.16,  step 
  0:  1.08249e+06
kernel133/checkpot.out:Coulomb (LR)     step   0:       41.6662,  step 
  0:      106.611
kernel134/checkpot.out:Coulomb (LR)     step   0:      -6869.62,  step 
  0:            0
kernel201/checkpot.out:Coulomb (LR)     step   0:      -67.4894,  step 
  0:     -447.931
kernel202/checkpot.out:Coulomb (LR)     step   0:      -7495.99,  step 
  0:       185610
kernel203/checkpot.out:Coulomb (LR)     step   0:       95.6825,  step 
  0:     -16.3288
kernel204/checkpot.out:Coulomb (LR)     step   0:       -6775.9,  step 
  0:            0
kernel211/checkpot.out:Coulomb (LR)     step   0:       46.4688,  step 
  0:      815.426
kernel212/checkpot.out:Coulomb (LR)     step   0:      -7495.99,  step 
  0:       185610
kernel213/checkpot.out:Coulomb (LR)     step   0:      -23.7805,  step 
  0:      49.7022
kernel214/checkpot.out:Coulomb (LR)     step   0:       -6775.9,  step 
  0:            0
kernel221/checkpot.out:Coulomb (LR)     step   0:       59.1389,  step 
  0:      759.341
kernel222/checkpot.out:Coulomb (LR)     step   0:         -7496,  step 
  0:       185610
kernel223/checkpot.out:Coulomb (LR)     step   0:      -23.7809,  step 
  0:      49.7022
kernel224/checkpot.out:Coulomb (LR)     step   0:      -6775.91,  step 
  0:            0
kernel231/checkpot.out:Coulomb (LR)     step   0:       46.4262,  step 
  0:      825.525
kernel232/checkpot.out:Coulomb (LR)     step   0:         -7496,  step 
  0:       185610
kernel233/checkpot.out:Coulomb (LR)     step   0:      -23.7805,  step 
  0:      49.7022
kernel234/checkpot.out:Coulomb (LR)     step   0:       -6775.9,  step 
  0:            0
kernel301/checkpot.out:Coulomb (LR)     step   0:       120.178,  step 
  0:     -958.108
kernel302/checkpot.out:Coulomb (LR)     step   0:      -9930.02,  step 
  0:       845348
kernel303/checkpot.out:Coulomb (LR)     step   0:       43.5326,  step 
  0:     -230.578
kernel304/checkpot.out:Coulomb (LR)     step   0:      -12752.2,  step 
  0:            0
kernel311/checkpot.out:Coulomb (LR)     step   0:       469.031,  step 
  0:      5266.61
kernel312/checkpot.out:Coulomb (LR)     step   0:      -9930.02,  step 
  0:       845348
kernel313/checkpot.out:Coulomb (LR)     step   0:       37.0683,  step 
  0:      152.617
kernel314/checkpot.out:Coulomb (LR)     step   0:      -12752.2,  step 
  0:            0
kernel321/checkpot.out:Coulomb (LR)     step   0:       562.748,  step 
  0:      4894.47
kernel322/checkpot.out:Coulomb (LR)     step   0:      -9930.01,  step 
  0:       845348
kernel323/checkpot.out:Coulomb (LR)     step   0:       37.0682,  step 
  0:      152.617
kernel324/checkpot.out:Coulomb (LR)     step   0:      -12752.2,  step 
  0:            0
kernel331/checkpot.out:Coulomb (LR)     step   0:        549.84,  step 
  0:      5218.29
kernel332/checkpot.out:Coulomb (LR)     step   0:      -9930.02,  step 
  0:       845348
kernel333/checkpot.out:Coulomb (LR)     step   0:       37.0683,  step 
  0:      152.617
kernel334/checkpot.out:Coulomb (LR)     step   0:      -12752.2,  step 
  0:            0

Is this occurring because the solvent optimization means charge 
parameters are not being looked up correctly? If so, how do we make 
GMX_NB_GENERIC trigger correct behaviour?

Mark