Re: [AMBER-Developers] Volunteers needed for some Amber18 release issues

From: Charles Lin <clin92.ucsd.edu>
Date: Mon, 19 Mar 2018 18:44:47 +0000

Lastly, the energy bug is related to indexing the wrong buffers.


Around line 2716

else

  buf_ctr=5

  ti_kin_ene(1, ti_eke) = eke - reduce_buf_out(buf_ctr - 5)

  ti_kin_ene(1, ti_eke) = eke - reduce_buf_out(buf_ctr - 4)


The mpi version expects

  ti_kin_ene(1, ti_eke) = eke - reduce_buf_out(buf_ctr + 1)

  ti_kin_ene(1, ti_eke) = eke - reduce_buf_out(buf_ctr + 2)

so the easy solution is to #ifdef an mpi version to use those buffers, and otherwise use the serial version. I have a feeling the buffers might be pretty broken for the other types of energies though for MPI. Same with the previous bug I showed use an ifdef for MPI and else for serial.

(I was doing my tests on test/pmemdTI/softcore/short_md).

Anyways that's all the bandwidth I have today to work on this. If anyone wants to find a more robust solution feel free.



Charlie Lin
PhD Candidate
Walker Lab
GlaxoSmithKline & University of California, San Diego
________________________________
From: Charles Lin
Sent: Monday, March 19, 2018 2:17:02 PM
To: david.case.rutgers.edu; AMBER Developers Mailing List
Subject: Re: [AMBER-Developers] Volunteers needed for some Amber18 release issues


So I did a little bit of debugging on problem 3.


The softcore TI region is an issue where the buffers themselves are being flipped I believe this has something to do with the reverse iteration around lines 2716ish (you're basically looking for code involving ti_sc_eke)


If you go around line 2870

flip

ti_ene(1,si_kin_ene) = ti_kin_ene(1,ti_sc_eke)

ti_ene(2,si_kin_ene) = ti_kin_ene(2,ti_sc_eke)

to

ti_ene(1,si_kin_ene) = ti_kin_ene(2,ti_sc_eke)

ti_ene(2,si_kin_ene) = ti_kin_ene(1,ti_sc_eke)

(I believe you also have to do this for the not langevin case)


This is more of a hack than a fix if someone wants to find a more robust solution.



Now there's also the issue with region 2 temperatures that was a separate issue, which I'm looking into.


Charlie
________________________________
From: Cruzeiro,Vinicius Wilian D <vwcruzeiro.ufl.edu>
Sent: Monday, March 19, 2018 9:35:38 AM
To: david.case.rutgers.edu; AMBER Developers Mailing List
Subject: Re: [AMBER-Developers] Volunteers needed for some Amber18 release issues

Hi,

I can help with problem 1.

Best,


Vinícius Wilian D Cruzeiro

PhD Candidate
Department of Chemistry, Physical Chemistry Division
University of Florida, United States

Voice: +1(352)846-1633<tel:+1(352)846-1633>

On Mar 19, 2018, at 8:32 AM, David A Case <david.case.rutgers.edu<mailto:david.case.rutgers.edu>> wrote:

Hi everyone:

We need volunteers to address the following issues. If you are willing/able
to help out, please reply to the list, to avoid duplication of effort.

1. If --skip-python is selected at configure time, the test suite still
tries to run python-related tests. (Tests need to consult config.h
to see if python has been disabled.)

2. MacOS specific: the mmpbsa_py/07_Comprehensive test fails (serial
mode). This is actually a long-standing problem. Someone needs to run
the test by hand, saving intermediate outputs, and try to isolate what
is going on. There may be one frame that is handled badly, or simply
something odd in how results are accumulated. Doesn't depend on which
version of gnu compilers is used, but never seems to show up on Linux.

3. Many pmemdTI CPU tests are failing in parallel (both Linux and Mac).
There is an issue where (for softcore) the save file looks correct but
the current output is printing zero for the temperature. Other examples
look like more than just roundoff errors.

4. We are getting segfaults in sander ips tests, especially in parallel.
I'll contact Xiongwu about this, but it would help if someone could
compile with -debug set, fire up a debugger (do the serial version
first), and provide better location info for the segfault.

Note that item 3 above is crucial: we need to have TI working in pmemd!

...thanks!....dac


_______________________________________________
AMBER-Developers mailing list
AMBER-Developers.ambermd.org<mailto:AMBER-Developers.ambermd.org>
https://urldefense.proofpoint.com/v2/url?u=http-3A__lists.ambermd.org_mailman_listinfo_amber-2Ddevelopers&d=DwICAg&c=pZJPUDQ3SB9JplYbifm4nt2lEVG5pWx2KikqINpWlZM&r=LIQu8OlVNKmzfbMg9_5FnKrt9-DrdQBJXyFyocKAWXc&m=6ss3A-p2WMts5xeSNn5cvMN9lSZDhkMUmAw2ZjybX4Q&s=byBIgkHM19OHQLAK1g86o-r2MxiJr_9jDqwlRsJUNws&e=
_______________________________________________
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 Mon Mar 19 2018 - 12:00:04 PDT
Custom Search