On Mac OS 10.6.2 with the Intel compilers (11.1.084) and their MKL libraries:
AmberTools serial
=================
Everything passes (except for ptraj_trajectory, which is not run; see Dan's note on a different branch of this thread).
make: Target `test.serial' not remade because of errors.
283 file comparisons passed
0 file comparisons failed
0 tests experienced errors
Amber serial
============
403 file comparisons passed
1 file comparison failed
0 tests experienced errors
qmmm_DFTB tests are skipped on the grounds of "DFTB SLKO files not found". Also, the tip5p test with PMEMD is missing these lines of output:
EKCMT = 0.2373 VIRIAL = 15.5606 VOLUME = 0.0001
Density = 0.0000
Everything else runs smoothly, once a symlink is made to the chamber directory.
AmberTools parallel
===================
I second Dan's observation about numprocs.awk (which I resolved by linking to amber11/test/numprocs.awk) and testing the rism tests.
I saw this remark while rebuilding with rismmpi for the sake of testing:
_amber_rism_interface.f(69): warning #6380: The structure length is not a multiple of its largest element; could create misalignments for arrays of this type. [RISMPRM_T]
type rismprm_t
-------^
Then, once the compilation was finished, with RISM, every single NAB test failed, with missing symbols:
"_mpi_gather_", referenced from:
_rism3d_opendx_mp_writedx_ in libnab.a(rism3d_opendx.o)
"_mpi_gatherv_", referenced from:
_rism3d_opendx_mp_writedx_ in libnab.a(rism3d_opendx.o)
"_mpi_bcast_", referenced from:
_rism_mpi_bcast_ in libnab.a(amber_rism_interface.o)
_rism_mpi_bcast_ in libnab.a(amber_rism_interface.o)
...rhubarb rhubarb...
_rism3d_solu_c_mp_rism3d_solu_mpi_clone_ in libnab.a(rism3d_solu_c.o)
"_mpi_allreduce_", referenced from:
_rism3d_c_mp_rism3d_pmv_ in libnab.a(rism3d_c.o)
_rism_util_mp_corr_drift_ in libnab.a(rism_util.o)
_rism_util_mp_checksum_ in libnab.a(rism_util.o)
_fce_c_mp_fce_update_ in libnab.a(fce_c.o)
_mdiis_blas_c_mp_mdiis_blas_advance_ in libnab.a(mdiis_blas_c.o)
_mdiis_orig_c_mp_mdiis_orig_advance_ in libnab.a(mdiis_orig_c.o)
_mdiis_orig_c_mp_mdiis_orig_advance_ in libnab.a(mdiis_orig_c.o)
"_mpi_reduce_", referenced from:
_rism_force_ in libnab.a(amber_rism_interface.o)
ld: symbol(s) not found
cc failed!
Ignoring the rism tests, with 1, 2 or 4 CPUs, I get:
make: Target `test.parallel2' not remade because of errors.
27 file comparisons passed
0 file comparisons failed
2 tests experienced errors
where the two errors are the aforementioned RISM tests not being skipped.
With 8 CPUs, I get:
make: Target `test.parallel2' not remade because of errors.
20 file comparisons passed
7 file comparisons failed
1 test experienced an error
The 1 error was, again, a RISM test. (The other RISM test was skipped, as 8 is an inappropriate number of CPUs for it.)
The seven file comparisons that failed are all in ptraj_comprehensive. dist_end_to_end.list is missing the last four entries, 97 to 100 - does it work by some kind of modulus of the number of processors? Ditto for watershell.list, phi, psi and omega. Also, the output trajectories - test.mdcrd and trajectory_test.mdcrd - are missing their ends, perhaps for the same reason.
Amber parallel
==============
* When running Make Clean before building in parallel, Make attempts to clean the directories sqm, rism, pbsa, blas, lapack and arpack - but because they don't exist (except in tools) and aren't preceded by a minus sign, it throws up its hands and quits.
* The test Run.dhfr.noboxinfo, though it passes, produces an MPI error. I did some investigation, and found that the reason for this is that the test, as written, generates an error condition (it seems it checks for the handling of the "no box" situation). I've prepared a patch file for the run script (attached), which is designed to print out a "this is expected" message to reassure users.
--- Run.dhfr.noboxinfo.bak 2010-03-30 20:30:05.000000000 -0400
+++ Run.dhfr.noboxinfo 2010-03-30 20:41:46.000000000 -0400
.. -21,6 +21,12 ..
set output = mdout.dhfr.noboxinfo
+echo ""
+echo "Note: When this test is run in parallel, an MPI exit error is normal."
+echo "This is because it tests bad input handling, and sander's correct"
+echo "behaviour is to exit with an error code."
+echo ""
+
touch dummy
$DO_PARALLEL $TESTsander -O -i gbin -c md12noboxinfo.x -o $output < dummy
* On any number of CPUs, with the solvent cap test I put in rdparm.f, the test Run.pb_ivcap0 now fails. (Was I over-zealous by taking out PB as well as GB?) Also, this failure causes test.sander.BASIC to stop. Which is strange; I had thought "make -k" would be implied.
* On any number of CPUs, the qmmm_DFTB tests are skipped because the DFTB SLKO files are not found.
* On 1 CPU, rem_gb_2rep/Run.rem fails (more groups than number of processors). Likewise for rem_wat/Run.rem, rem_hybrid/Run.rem, sodium/Run.sodium and evb/malon_bond_umb_GAFF/Run.evb.
* On 1 CPU, the PMEMD tests fail, requiring at least 2. More accurately, the first test fails, and the others aren't even tried. Which is strange; I had thought "make -k" would be implied.
* On 2 CPUs, the test evb/malon_dbonds_umb_dg_UFF_3DG_qi_full_2D-PMF/Run.evb segfaults, as does evb/malon_dbonds_umb_dg_UFF_3DG_qi_full_corrF/Run.evb.
* On 4 CPUs, the following tests failed:
- ti_decomp/Run.ti_decomp (differences are attached)
- neb-testcases/neb_gb_partial/Run.neb_gb_partial (cat: illegal option -- a)
- neb-testcases/neb_gb_full/Run.neb_gb_full (cat: illegal option -- a)
- neb-testcases/neb_explicit/Run.neb_explicit (cat: illegal option -- a)
* On 8 CPUs, the following tests failed:
- ti_decomp/Run.ti_decomp (differences are attached)
- neb-testcases/neb_gb_partial/Run.neb_gb_partial (cat: illegal option -- a)
- neb-testcases/neb_gb_full/Run.neb_gb_full (cat: illegal option -- a)
- neb-testcases/neb_explicit/Run.neb_explicit (cat: illegal option -- a)
- ncsu/smd2 ("must have more residues than processors")
- qmmm2/xcrd_build_test/Run.ortho_qmewald0 ("Non bond list overflow! Check MAXPR in locmem.f"), which stopped the QM/MM tests (surprisingly, again, considering -k)
1 CPU:
176 file comparisons passed
4 file comparisons failed
4 tests experienced errors
2 CPUs:
344 file comparisons passed
1 file comparison failed
3 tests experienced errors
4 CPUs:
300 file comparisons passed
26 file comparisons failed
1 test experienced an error
8 CPUs:
236 file comparisons passed
29 file comparisons failed
2 tests experienced errors
Cheers,
Ben
On 28/3/2010, at 10:52 p.m., case wrote:
> Hi everyone:
>
> I have placed the latest release candidates here:
>
> http://ambermd.org/downloads/AmberTools.28mar10.tar.bz2
> http://ambermd.org/downloads/Amber11.28mar10.tar.bz2
>
> Notes:
>
> 1. AmberTools can stand on its own, but Amber11 requires you to first download
> and untar AmberTools. Note: you don't need to *compile* AmberTools, but you
> do need to untar and configure it.
>
> If you don't want to read the instructions: go to amber11/AmberTools/src
> and run configure from there. This will create a config.h file that is used
> by both AmberTools and Amber11.
>
> 2. I left out the 2_million_atom and cuda test suites: these take up 200
> Mbytes, and won't be used by most people. I'm thinking of distributing
> these in a separate tar file. Many people have to download Amber over pretty
> slow internet connections.
>
> 3. The code in CVS no longer corresponds to the release candidates, but the
> differences are few. I still encourage people to update the documentation,
> and commits can still be made in the amber11/doc subdirectory.
>
> ...thx...dac
>
> _______________________________________________
> 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 30 2010 - 19:00:03 PDT
