* To help us fund GROMACS development, we humbly ask that you cite
* the research papers on the package. Check out http://www.gromacs.org.
*/
-#ifdef HAVE_CONFIG_H
-#include <config.h>
-#endif
+#include "gmxpre.h"
+
+#include "gromacs/legacyheaders/sim_util.h"
+
+#include "config.h"
#include <assert.h>
#include <math.h>
#include <stdio.h>
#include <string.h>
-#ifdef HAVE_SYS_TIME_H
-#include <sys/time.h>
-#endif
-
-#include "typedefs.h"
-#include "gromacs/utility/cstringutil.h"
-#include "gromacs/utility/smalloc.h"
-#include "names.h"
-#include "txtdump.h"
-#include "pbc.h"
-#include "chargegroup.h"
-#include "vec.h"
-#include "nrnb.h"
-#include "mshift.h"
-#include "mdrun.h"
-#include "sim_util.h"
-#include "update.h"
-#include "physics.h"
-#include "main.h"
-#include "mdatoms.h"
-#include "force.h"
-#include "bondf.h"
-#include "pme.h"
-#include "disre.h"
-#include "orires.h"
-#include "network.h"
-#include "calcmu.h"
-#include "constr.h"
-#include "xvgr.h"
-#include "copyrite.h"
-#include "domdec.h"
-#include "genborn.h"
-#include "nbnxn_atomdata.h"
-#include "nbnxn_search.h"
-#include "nbnxn_kernels/nbnxn_kernel_ref.h"
-#include "nbnxn_kernels/simd_4xn/nbnxn_kernel_simd_4xn.h"
-#include "nbnxn_kernels/simd_2xnn/nbnxn_kernel_simd_2xnn.h"
-#include "nbnxn_kernels/nbnxn_kernel_gpu_ref.h"
-#include "nonbonded.h"
-#include "../gmxlib/nonbonded/nb_kernel.h"
-#include "../gmxlib/nonbonded/nb_free_energy.h"
-#include "gromacs/timing/wallcycle.h"
-#include "gromacs/timing/walltime_accounting.h"
-#include "gromacs/utility/gmxmpi.h"
#include "gromacs/essentialdynamics/edsam.h"
+#include "gromacs/ewald/pme.h"
+#include "gromacs/gmxlib/nonbonded/nb_free_energy.h"
+#include "gromacs/gmxlib/nonbonded/nb_kernel.h"
+#include "gromacs/imd/imd.h"
+#include "gromacs/legacyheaders/calcmu.h"
+#include "gromacs/legacyheaders/chargegroup.h"
+#include "gromacs/legacyheaders/constr.h"
+#include "gromacs/legacyheaders/copyrite.h"
+#include "gromacs/legacyheaders/disre.h"
+#include "gromacs/legacyheaders/domdec.h"
+#include "gromacs/legacyheaders/force.h"
+#include "gromacs/legacyheaders/genborn.h"
+#include "gromacs/legacyheaders/gmx_omp_nthreads.h"
+#include "gromacs/legacyheaders/mdatoms.h"
+#include "gromacs/legacyheaders/mdrun.h"
+#include "gromacs/legacyheaders/names.h"
+#include "gromacs/legacyheaders/network.h"
+#include "gromacs/legacyheaders/nonbonded.h"
+#include "gromacs/legacyheaders/nrnb.h"
+#include "gromacs/legacyheaders/orires.h"
+#include "gromacs/legacyheaders/qmmm.h"
+#include "gromacs/legacyheaders/txtdump.h"
+#include "gromacs/legacyheaders/typedefs.h"
+#include "gromacs/legacyheaders/update.h"
+#include "gromacs/legacyheaders/types/commrec.h"
+#include "gromacs/listed-forces/bonded.h"
+#include "gromacs/math/units.h"
+#include "gromacs/math/vec.h"
+#include "gromacs/mdlib/nb_verlet.h"
+#include "gromacs/mdlib/nbnxn_atomdata.h"
+#include "gromacs/mdlib/nbnxn_search.h"
+#include "gromacs/mdlib/nbnxn_cuda/nbnxn_cuda.h"
+#include "gromacs/mdlib/nbnxn_cuda/nbnxn_cuda_data_mgmt.h"
+#include "gromacs/mdlib/nbnxn_kernels/nbnxn_kernel_gpu_ref.h"
+#include "gromacs/mdlib/nbnxn_kernels/nbnxn_kernel_ref.h"
+#include "gromacs/mdlib/nbnxn_kernels/simd_2xnn/nbnxn_kernel_simd_2xnn.h"
+#include "gromacs/mdlib/nbnxn_kernels/simd_4xn/nbnxn_kernel_simd_4xn.h"
+#include "gromacs/pbcutil/ishift.h"
+#include "gromacs/pbcutil/mshift.h"
+#include "gromacs/pbcutil/pbc.h"
#include "gromacs/pulling/pull.h"
#include "gromacs/pulling/pull_rotation.h"
-#include "gromacs/imd/imd.h"
-#include "adress.h"
-#include "qmmm.h"
-
-#include "gmx_omp_nthreads.h"
+#include "gromacs/timing/wallcycle.h"
+#include "gromacs/timing/walltime_accounting.h"
+#include "gromacs/utility/cstringutil.h"
+#include "gromacs/utility/gmxmpi.h"
+#include "gromacs/utility/smalloc.h"
+#include "gromacs/utility/sysinfo.h"
-#include "nbnxn_cuda_data_mgmt.h"
-#include "nbnxn_cuda/nbnxn_cuda.h"
+#include "adress.h"
void print_time(FILE *out,
gmx_walltime_accounting_t walltime_accounting,
tensor vir_part, t_graph *graph, matrix box,
t_nrnb *nrnb, const t_forcerec *fr, int ePBC)
{
- int i, j;
- tensor virtest;
+ int i;
/* The short-range virial from surrounding boxes */
clear_mat(vir_part);
}
}
-static void posres_wrapper(FILE *fplog,
- int flags,
- gmx_bool bSepDVDL,
+static void posres_wrapper(int flags,
t_inputrec *ir,
t_nrnb *nrnb,
gmx_localtop_t *top,
ir->ePBC == epbcNONE ? NULL : &pbc,
lambda[efptRESTRAINT], &dvdl,
fr->rc_scaling, fr->ePBC, fr->posres_com, fr->posres_comB);
- if (bSepDVDL)
- {
- gmx_print_sepdvdl(fplog, interaction_function[F_POSRES].longname, v, dvdl);
- }
enerd->term[F_POSRES] += v;
/* If just the force constant changes, the FEP term is linear,
* but if k changes, it is not.
{
for (i = 0; i < enerd->n_lambda; i++)
{
- real dvdl_dum, lambda_dum;
+ real lambda_dum;
lambda_dum = (i == 0 ? lambda[efptRESTRAINT] : ir->fepvals->all_lambda[efptRESTRAINT][i-1]);
v = posres(top->idef.il[F_POSRES].nr, top->idef.il[F_POSRES].iatoms,
inc_nrnb(nrnb, eNR_FBPOSRES, top->idef.il[F_FBPOSRES].nr/2);
}
-static void pull_potential_wrapper(FILE *fplog,
- gmx_bool bSepDVDL,
- t_commrec *cr,
+static void pull_potential_wrapper(t_commrec *cr,
t_inputrec *ir,
matrix box, rvec x[],
rvec f[],
enerd->term[F_COM_PULL] +=
pull_potential(ir->ePull, ir->pull, mdatoms, &pbc,
cr, t, lambda[efptRESTRAINT], x, f, vir_force, &dvdl);
- if (bSepDVDL)
- {
- gmx_print_sepdvdl(fplog, "Com pull", enerd->term[F_COM_PULL], dvdl);
- }
enerd->dvdl_lin[efptRESTRAINT] += dvdl;
wallcycle_stop(wcycle, ewcPULLPOT);
}
-static void pme_receive_force_ener(FILE *fplog,
- gmx_bool bSepDVDL,
- t_commrec *cr,
+static void pme_receive_force_ener(t_commrec *cr,
gmx_wallcycle_t wcycle,
gmx_enerdata_t *enerd,
t_forcerec *fr)
{
- real e_q, e_lj, v, dvdl_q, dvdl_lj;
+ real e_q, e_lj, dvdl_q, dvdl_lj;
float cycles_ppdpme, cycles_seppme;
cycles_ppdpme = wallcycle_stop(wcycle, ewcPPDURINGPME);
gmx_pme_receive_f(cr, fr->f_novirsum, fr->vir_el_recip, &e_q,
fr->vir_lj_recip, &e_lj, &dvdl_q, &dvdl_lj,
&cycles_seppme);
- if (bSepDVDL)
- {
- gmx_print_sepdvdl(fplog, "Electrostatic PME mesh", e_q, dvdl_q);
- gmx_print_sepdvdl(fplog, "Lennard-Jones PME mesh", e_lj, dvdl_lj);
- }
enerd->term[F_COUL_RECIP] += e_q;
enerd->term[F_LJ_RECIP] += e_lj;
enerd->dvdl_lin[efptCOUL] += dvdl_q;
t_nrnb *nrnb,
gmx_wallcycle_t wcycle)
{
- int nnbl, kernel_type, enr_nbnxn_kernel_ljc, enr_nbnxn_kernel_lj;
- char *env;
+ int enr_nbnxn_kernel_ljc, enr_nbnxn_kernel_lj;
nonbonded_verlet_group_t *nbvg;
gmx_bool bCUDA;
wallcycle_sub_stop(wcycle, ewcsNONBONDED);
}
+gmx_bool use_GPU(const nonbonded_verlet_t *nbv)
+{
+ return nbv != NULL && nbv->bUseGPU;
+}
+
void do_force_cutsVERLET(FILE *fplog, t_commrec *cr,
t_inputrec *inputrec,
gmx_int64_t step, t_nrnb *nrnb, gmx_wallcycle_t wcycle,
gmx_bool bBornRadii,
int flags)
{
- int cg0, cg1, i, j;
+ int cg1, i, j;
int start, homenr;
- int nb_kernel_type;
double mu[2*DIM];
- gmx_bool bSepDVDL, bStateChanged, bNS, bFillGrid, bCalcCGCM, bBS;
+ gmx_bool bStateChanged, bNS, bFillGrid, bCalcCGCM;
gmx_bool bDoLongRange, bDoForces, bSepLRF, bUseGPU, bUseOrEmulGPU;
gmx_bool bDiffKernels = FALSE;
- matrix boxs;
rvec vzero, box_diag;
- real e, v, dvdl;
float cycles_pme, cycles_force, cycles_wait_gpu;
nonbonded_verlet_t *nbv;
cycles_force = 0;
cycles_wait_gpu = 0;
nbv = fr->nbv;
- nb_kernel_type = fr->nbv->grp[0].kernel_type;
start = 0;
homenr = mdatoms->homenr;
- bSepDVDL = (fr->bSepDVDL && do_per_step(step, inputrec->nstlog));
-
clear_mat(vir_force);
- cg0 = 0;
if (DOMAINDECOMP(cr))
{
cg1 = cr->dd->ncg_tot;
#ifdef GMX_MPI
if (!(cr->duty & DUTY_PME))
{
+ gmx_bool bBS;
+ matrix boxs;
+
/* Send particle coordinates to the pme nodes.
* Since this is only implemented for domain decomposition
* and domain decomposition does not use the graph,
}
/* We calculate the non-bonded forces, when done on the CPU, here.
- * We do this before calling do_force_lowlevel, as in there bondeds
- * forces are calculated before PME, which does communication.
- * With this order, non-bonded and bonded force calculation imbalance
- * can be balanced out by the domain decomposition load balancing.
+ * We do this before calling do_force_lowlevel, because in that
+ * function, the listed forces are calculated before PME, which
+ * does communication. With this order, non-bonded and listed
+ * force calculation imbalance can be balanced out by the domain
+ * decomposition load balancing.
*/
if (!bUseOrEmulGPU)
update_QMMMrec(cr, fr, x, mdatoms, box, top);
}
- if ((flags & GMX_FORCE_BONDED) && top->idef.il[F_POSRES].nr > 0)
+ if ((flags & GMX_FORCE_LISTED) && top->idef.il[F_POSRES].nr > 0)
{
- posres_wrapper(fplog, flags, bSepDVDL, inputrec, nrnb, top, box, x,
+ posres_wrapper(flags, inputrec, nrnb, top, box, x,
enerd, lambda, fr);
}
- if ((flags & GMX_FORCE_BONDED) && top->idef.il[F_FBPOSRES].nr > 0)
+ if ((flags & GMX_FORCE_LISTED) && top->idef.il[F_FBPOSRES].nr > 0)
{
fbposres_wrapper(inputrec, nrnb, top, box, x, enerd, fr);
}
/* Compute the bonded and non-bonded energies and optionally forces */
- do_force_lowlevel(fplog, step, fr, inputrec, &(top->idef),
+ do_force_lowlevel(fr, inputrec, &(top->idef),
cr, nrnb, wcycle, mdatoms,
x, hist, f, bSepLRF ? fr->f_twin : f, enerd, fcd, top, fr->born,
- &(top->atomtypes), bBornRadii, box,
+ bBornRadii, box,
inputrec->fepvals, lambda, graph, &(top->excls), fr->mu_tot,
flags, &cycles_pme);
/* Since the COM pulling is always done mass-weighted, no forces are
* applied to vsites and this call can be done after vsite spreading.
*/
- pull_potential_wrapper(fplog, bSepDVDL, cr, inputrec, box, x,
+ pull_potential_wrapper(cr, inputrec, box, x,
f, vir_force, mdatoms, enerd, lambda, t,
wcycle);
}
/* In case of node-splitting, the PP nodes receive the long-range
* forces, virial and energy from the PME nodes here.
*/
- pme_receive_force_ener(fplog, bSepDVDL, cr, wcycle, enerd, fr);
+ pme_receive_force_ener(cr, wcycle, enerd, fr);
}
if (bDoForces)
int cg0, cg1, i, j;
int start, homenr;
double mu[2*DIM];
- gmx_bool bSepDVDL, bStateChanged, bNS, bFillGrid, bCalcCGCM, bBS;
- gmx_bool bDoLongRangeNS, bDoForces, bDoPotential, bSepLRF;
+ gmx_bool bStateChanged, bNS, bFillGrid, bCalcCGCM;
+ gmx_bool bDoLongRangeNS, bDoForces, bSepLRF;
gmx_bool bDoAdressWF;
- matrix boxs;
- rvec vzero, box_diag;
- real e, v, dvdlambda[efptNR];
t_pbc pbc;
float cycles_pme, cycles_force;
start = 0;
homenr = mdatoms->homenr;
- bSepDVDL = (fr->bSepDVDL && do_per_step(step, inputrec->nstlog));
-
clear_mat(vir_force);
cg0 = 0;
bFillGrid = (bNS && bStateChanged);
bCalcCGCM = (bFillGrid && !DOMAINDECOMP(cr));
bDoForces = (flags & GMX_FORCE_FORCES);
- bDoPotential = (flags & GMX_FORCE_ENERGY);
bSepLRF = ((inputrec->nstcalclr > 1) && bDoForces &&
(flags & GMX_FORCE_SEPLRF) && (flags & GMX_FORCE_DO_LR));
#ifdef GMX_MPI
if (!(cr->duty & DUTY_PME))
{
+ gmx_bool bBS;
+ matrix boxs;
+
/* Send particle coordinates to the pme nodes.
* Since this is only implemented for domain decomposition
* and domain decomposition does not use the graph,
update_QMMMrec(cr, fr, x, mdatoms, box, top);
}
- if ((flags & GMX_FORCE_BONDED) && top->idef.il[F_POSRES].nr > 0)
+ if ((flags & GMX_FORCE_LISTED) && top->idef.il[F_POSRES].nr > 0)
{
- posres_wrapper(fplog, flags, bSepDVDL, inputrec, nrnb, top, box, x,
+ posres_wrapper(flags, inputrec, nrnb, top, box, x,
enerd, lambda, fr);
}
- if ((flags & GMX_FORCE_BONDED) && top->idef.il[F_FBPOSRES].nr > 0)
+ if ((flags & GMX_FORCE_LISTED) && top->idef.il[F_FBPOSRES].nr > 0)
{
fbposres_wrapper(inputrec, nrnb, top, box, x, enerd, fr);
}
/* Compute the bonded and non-bonded energies and optionally forces */
- do_force_lowlevel(fplog, step, fr, inputrec, &(top->idef),
+ do_force_lowlevel(fr, inputrec, &(top->idef),
cr, nrnb, wcycle, mdatoms,
x, hist, f, bSepLRF ? fr->f_twin : f, enerd, fcd, top, fr->born,
- &(top->atomtypes), bBornRadii, box,
+ bBornRadii, box,
inputrec->fepvals, lambda,
graph, &(top->excls), fr->mu_tot,
flags,
if (inputrec->ePull == epullUMBRELLA || inputrec->ePull == epullCONST_F)
{
- pull_potential_wrapper(fplog, bSepDVDL, cr, inputrec, box, x,
+ pull_potential_wrapper(cr, inputrec, box, x,
f, vir_force, mdatoms, enerd, lambda, t,
wcycle);
}
/* In case of node-splitting, the PP nodes receive the long-range
* forces, virial and energy from the PME nodes here.
*/
- pme_receive_force_ener(fplog, bSepDVDL, cr, wcycle, enerd, fr);
+ pme_receive_force_ener(cr, wcycle, enerd, fr);
}
if (bDoForces)
double invscale, invscale2, invscale3;
double r, ea, eb, ec, pa, pb, pc, pd;
double y0, f, g, h;
- int ri, offset, tabfactor;
+ int ri, offset;
+ double tabfactor;
invscale = 1.0/scale;
invscale2 = invscale*invscale;
void calc_enervirdiff(FILE *fplog, int eDispCorr, t_forcerec *fr)
{
- double eners[2], virs[2], enersum, virsum, y0, f, g, h;
- double r0, r1, r, rc3, rc9, ea, eb, ec, pa, pb, pc, pd;
- double invscale, invscale2, invscale3;
- int ri0, ri1, ri, i, offstart, offset;
- real scale, *vdwtab, tabfactor, tmp;
+ double eners[2], virs[2], enersum, virsum;
+ double r0, rc3, rc9;
+ int ri0, ri1, i;
+ real scale, *vdwtab;
fr->enershiftsix = 0;
fr->enershifttwelve = 0;
vdwtab = fr->nblists[0].table_vdw.data;
/* Round the cut-offs to exact table values for precision */
- ri0 = floor(fr->rvdw_switch*scale);
- ri1 = ceil(fr->rvdw*scale);
+ ri0 = static_cast<int>(floor(fr->rvdw_switch*scale));
+ ri1 = static_cast<int>(ceil(fr->rvdw*scale));
/* The code below has some support for handling force-switching, i.e.
* when the force (instead of potential) is switched over a limited
ri0 = (fr->vdw_modifier == eintmodPOTSHIFT) ? ri1 : ri0;
r0 = ri0/scale;
- r1 = ri1/scale;
rc3 = r0*r0*r0;
rc9 = rc3*rc3*rc3;
}
}
-void calc_dispcorr(FILE *fplog, t_inputrec *ir, t_forcerec *fr,
- gmx_int64_t step, int natoms,
+void calc_dispcorr(t_inputrec *ir, t_forcerec *fr,
+ int natoms,
matrix box, real lambda, tensor pres, tensor virial,
real *prescorr, real *enercorr, real *dvdlcorr)
{
}
}
- if (fr->bSepDVDL && do_per_step(step, ir->nstlog))
- {
- gmx_print_sepdvdl(fplog, "Dispersion correction", *enercorr, dvdlambda);
- }
if (fr->efep != efepNO)
{
*dvdlcorr += dvdlambda;
t_inputrec *inputrec,
t_nrnb nrnb[], gmx_wallcycle_t wcycle,
gmx_walltime_accounting_t walltime_accounting,
- wallclock_gpu_t *gputimes,
+ nonbonded_verlet_t *nbv,
gmx_bool bWriteStat)
{
- int i, j;
t_nrnb *nrnb_tot = NULL;
- real delta_t;
- double nbfs, mflop;
+ double delta_t = 0;
+ double nbfs = 0, mflop = 0;
double elapsed_time,
elapsed_time_over_all_ranks,
elapsed_time_over_all_threads,
if (SIMMASTER(cr))
{
+ wallclock_gpu_t* gputimes = use_GPU(nbv) ?
+ nbnxn_cuda_get_timings(nbv->cu_nbv) : NULL;
wallcycle_print(fplog, cr->nnodes, cr->npmenodes,
elapsed_time_over_all_ranks,
wcycle, gputimes);
{
delta_t = inputrec->delta_t;
}
- else
- {
- delta_t = 0;
- }
if (fplog)
{
gmx_bool *bSimAnn, t_vcm **vcm, unsigned long Flags,
gmx_wallcycle_t wcycle)
{
- int i, j, n;
- real tmpt, mod;
+ int i;
/* Initial values */
*t = *t0 = ir->init_t;
biod.pnpi.spb.ru / alexxy / gromacs.git / blobdiff