[gmx-developers] Branches w/working OpenCL support
Mark Abraham
mark.j.abraham at gmail.com
Mon Jun 22 20:41:02 CEST 2015
Hi Mirco,
Thanks very much for the report. The issue with free() was simply me
forgetting to commit matching changes to the code. Roland's fixed some MSVC
issues, and the gmx_wham #include issue is also merged. The one thing we
don't have the ability to test locally is a run with two domains, each on
separate AMD GPU (because we only have one, so far!) so if you have been /
are able to test that case, that would be very welcome.
https://gerrit.gromacs.org/#/c/4314/ is based on all the fixes so far.
Thanks!
Mark
On Fri, Jun 19, 2015 at 11:10 AM Mirco Wahab <
mirco.wahab at chemie.tu-freiberg.de> wrote:
> > Probably not. From the last version (refs/changes/14/4314/26),
> > it appears mdrun would try to detect CPU capabilities at this point:
>
> I did a debug run w/VS2013. The segfault is in
> ocl_pfree(void * h_ptr) called in nbnxn_atomdata_init():
>
> stack trace:
>
> > msvcr120d.dll!_free_dbg_nolock(void * pUserData, int nBlockUse)
> Line 1303 C++
> msvcr120d.dll!_free_dbg(void * pUserData, int nBlockUse) Line
> 1265 C++
> msvcr120d.dll!free(void * pUserData) Line 50 C++
> gmx.exe!ocl_pfree(void * h_ptr) Line 203 C++
> gmx.exe!nbnxn_atomdata_init(_iobuf * fp, nbnxn_atomdata_t * nbat,
> int
> nb_kernel_type, int enbnxninitcombrule, int ntype, const float * nbfp,
> int n_energygroups, int nout, void (void * *, unsigned __int64) * alloc,
> void (void *) * free) Line 701 C
> gmx.exe!init_nb_verlet(_iobuf * fp, nonbonded_verlet_t * *
> nb_verlet,
> int bFEP_NonBonded, const t_inputrec * ir, const t_forcerec * fr, const
> t_commrec * cr, const char * nbpu_opt) Line 2288 C++
> gmx.exe!init_forcerec(_iobuf * fp, output_env * const oenv,
> t_forcerec * fr, t_fcdata * fcd, const t_inputrec * ir, const gmx_mtop_t
> * mtop, const t_commrec * cr, float[3] * box, const char * tabfn, const
> char * tabafn, const char * tabpfn, const char * tabbfn, const char *
> nbpu_opt, int bNoSolvOpt, float print_force) Line 3238 C++
> gmx.exe!mdrunner(gmx_hw_opt_t * hw_opt, _iobuf * fplog, t_commrec *
> cr, int nfile, const t_filenm * fnm, output_env * const oenv, int
> bVerbose, int bCompact, int nstglobalcomm, int * ddxyz, int
> dd_node_order, float rdd, float rconstr, const char * dddlb_opt, float
> dlb_scale, const char * ddcsx, const char * ddcsy, const char * ddcsz,
> const char * nbpu_opt, int nstlist_cmdline, __int64 nsteps_cmdline, int
> nstepout, int resetstep, int nmultisim, int repl_ex_nst, int
> repl_ex_nex, int repl_ex_seed, float pforce, float cpt_period, float
> max_hours, int imdport, unsigned long Flags) Line 1130 C++
> gmx.exe!gmx_mdrun(int argc, char * * argv) Line 542 C++
> gmx.exe!`anonymous namespace'::NoNiceModule::run(int argc, char * *
> argv) Line 157 C++
> gmx.exe!gmx::CommandLineModuleManager::run(int argc, char * * argv)
> Line 554 C++
> gmx.exe!main(int argc, char * * argv) Line 60 C++
> [External Code]
>
> Regards
>
> M.
>
>
> --
> Gromacs Developers mailing list
>
> * Please search the archive at
> http://www.gromacs.org/Support/Mailing_Lists/GMX-developers_List before
> posting!
>
> * Can't post? Read http://www.gromacs.org/Support/Mailing_Lists
>
> * For (un)subscribe requests visit
> https://maillist.sys.kth.se/mailman/listinfo/gromacs.org_gmx-developers
> or send a mail to gmx-developers-request at gromacs.org.
>
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://maillist.sys.kth.se/pipermail/gromacs.org_gmx-developers/attachments/20150622/8920ce39/attachment-0001.html>
More information about the gromacs.org_gmx-developers
mailing list