Hi all,
I traced down the problem and found the difference is triggered by a
1.0e-10 important digit difference through linear algebra library
subroutines between 32bit and 64bit machine. The difference would then
jump to 1.0e-4 in the next iteration. And the sgesvd() can be more
stable and would make the tests pass on 32bit machine.
There is no specific reason for using dsvdc() instead of dgesvd().
Maybe we should just set isvd to 2 to save the energy to look into the
library subroutines.
Best Regards!
Xingping (From Dr. Ray Luo's lab)
> Qin are seeing the same problem on a 32bit Linux, we are looking into it
...
> Ray
> On Fri, Mar 23, 2012 at 9:37 AM, Daniel Roe <daniel.r.roe.gmail.com> wrote:
>> Hi All,
>> I've come across several issues related to PBSA under cygwin.
>> AmberTools:
>> Failure: pbsa_aug_c2 (Li.out, Cl.out, MG.out, Ca.out) Failure:
>> sas2test (Rb.out.5, IM.out.5)
>> All 6 errors have the same format:
>> < The gmres dimension is 185
>> > 10 - ssvdc() or sgesvd() failed in GrToPr!
>> If I switch isvd from 1 to 2 in GrToPr.F90 (line 48) the errors go
away, so unless there is some reason it shouldn't be set at 2 we can
just put an ifdef for CYGWIN and make isvd = 2, otherwise default to 1
(patch for GrToPr.F90 attached).
>> Amber:
>> 3 failures (sander_pbsa_tsr, sander_pbsa_ligand, 02_MMPBSA_Stability)
All 3 exit with similar error messages:
>> SA Bomb in sa_arc()3: Allocates aborted 0 0 5014
0 0
>> I think this might be related to the fixed stack size in cygwin but
since I can't easily change the stack size its tough to test. I'm
looking into it but maybe the pbsa devs have an idea?
>> Failure: sander_pbsa_frc (dbf_2/dadt/force.dat)
>> 3c3
>> < -0.961100E-01 -0.424215E+00 -0.888034E-01 ---
>> > -0.961417E-01 -0.423961E+00 -0.894469E-01
>> 40c40
>> < -0.961100E-01 -0.424215E+00 -0.888034E-01 ---
>> > -0.961417E-01 -0.423961E+00 -0.894469E-01
>> Seems like round-off.
>> Failure: sander_pbsa_frc (dbf_2/dgdc/force.dat)
>> 16c16
>> < 0.620819E-01 -0.187438E+01 0.140104E+01 ---
>> > 0.754525E-01 -0.189257E+01 0.132932E+01
>> 52c52
>> < 0.620819E-01 -0.187438E+01 0.140104E+01 ---
>> > 0.754525E-01 -0.189257E+01 0.132932E+01
>> Some of these differences seem a little too big to be round-off; maybe
the pbsa devs can comment on this.
>> -Dan
>> _______________________________________________
>> AMBER-Developers mailing list
>> AMBER-Developers.ambermd.org
>> http://lists.ambermd.org/mailman/listinfo/amber-developers
> _______________________________________________
> AMBER-Developers mailing list
> AMBER-Developers.ambermd.org
> http://lists.ambermd.org/mailman/listinfo/amber-developers
_______________________________________________
AMBER-Developers mailing list
AMBER-Developers.ambermd.org
http://lists.ambermd.org/mailman/listinfo/amber-developers
Received on Tue Mar 27 2012 - 18:30:03 PDT
