Unite code for handling listed pairs
[alexxy/gromacs.git] / src / gromacs / gmxlib / nonbonded / nonbonded.c
index 2ef53d511d8765b53082d225c699b6a25f4a3eb6..f66f73bb0fcf32f30684a179a9bbbef92e39286d 100644 (file)
  * 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/nonbonded.h"
+
+#include "config.h"
 
 #include <stdio.h>
 #include <stdlib.h>
 
 #include "thread_mpi/threads.h"
 
-#include "typedefs.h"
-#include "txtdump.h"
-#include "ns.h"
-#include "gromacs/math/vec.h"
+#include "gromacs/gmxlib/nonbonded/nb_free_energy.h"
+#include "gromacs/gmxlib/nonbonded/nb_generic.h"
+#include "gromacs/gmxlib/nonbonded/nb_generic_adress.h"
+#include "gromacs/gmxlib/nonbonded/nb_generic_cg.h"
+#include "gromacs/gmxlib/nonbonded/nb_kernel.h"
+#include "gromacs/legacyheaders/force.h"
+#include "gromacs/legacyheaders/macros.h"
+#include "gromacs/legacyheaders/names.h"
+#include "gromacs/legacyheaders/nrnb.h"
+#include "gromacs/legacyheaders/ns.h"
+#include "gromacs/legacyheaders/txtdump.h"
+#include "gromacs/legacyheaders/typedefs.h"
+#include "gromacs/listed-forces/bonded.h"
 #include "gromacs/math/utilities.h"
-#include "macros.h"
-#include "gromacs/utility/cstringutil.h"
-#include "force.h"
-#include "names.h"
-#include "force.h"
-#include "bondf.h"
-#include "nrnb.h"
-#include "nonbonded.h"
+#include "gromacs/math/vec.h"
+#include "gromacs/pbcutil/ishift.h"
+#include "gromacs/pbcutil/mshift.h"
+#include "gromacs/pbcutil/pbc.h"
 #include "gromacs/simd/simd.h"
-
-#include "mshift.h"
+#include "gromacs/utility/cstringutil.h"
 #include "gromacs/utility/fatalerror.h"
 #include "gromacs/utility/smalloc.h"
 
-#include "nb_kernel.h"
-#include "nb_free_energy.h"
-#include "nb_generic.h"
-#include "nb_generic_cg.h"
-#include "nb_generic_adress.h"
-
 /* Different default (c) and SIMD instructions interaction-specific kernels */
-#include "nb_kernel_c/nb_kernel_c.h"
+#include "gromacs/gmxlib/nonbonded/nb_kernel_c/nb_kernel_c.h"
 
 #if (defined GMX_SIMD_X86_SSE2) && !(defined GMX_DOUBLE)
-#    include "nb_kernel_sse2_single/nb_kernel_sse2_single.h"
+#    include "gromacs/gmxlib/nonbonded/nb_kernel_sse2_single/nb_kernel_sse2_single.h"
 #endif
 #if (defined GMX_SIMD_X86_SSE4_1) && !(defined GMX_DOUBLE)
-#    include "nb_kernel_sse4_1_single/nb_kernel_sse4_1_single.h"
+#    include "gromacs/gmxlib/nonbonded/nb_kernel_sse4_1_single/nb_kernel_sse4_1_single.h"
 #endif
 #if (defined GMX_SIMD_X86_AVX_128_FMA) && !(defined GMX_DOUBLE)
-#    include "nb_kernel_avx_128_fma_single/nb_kernel_avx_128_fma_single.h"
+#    include "gromacs/gmxlib/nonbonded/nb_kernel_avx_128_fma_single/nb_kernel_avx_128_fma_single.h"
 #endif
 #if (defined GMX_SIMD_X86_AVX_256_OR_HIGHER) && !(defined GMX_DOUBLE)
-#    include "nb_kernel_avx_256_single/nb_kernel_avx_256_single.h"
+#    include "gromacs/gmxlib/nonbonded/nb_kernel_avx_256_single/nb_kernel_avx_256_single.h"
 #endif
 #if (defined GMX_SIMD_X86_SSE2 && defined GMX_DOUBLE)
-#    include "nb_kernel_sse2_double/nb_kernel_sse2_double.h"
+#    include "gromacs/gmxlib/nonbonded/nb_kernel_sse2_double/nb_kernel_sse2_double.h"
 #endif
 #if (defined GMX_SIMD_X86_SSE4_1 && defined GMX_DOUBLE)
-#    include "nb_kernel_sse4_1_double/nb_kernel_sse4_1_double.h"
+#    include "gromacs/gmxlib/nonbonded/nb_kernel_sse4_1_double/nb_kernel_sse4_1_double.h"
 #endif
 #if (defined GMX_SIMD_X86_AVX_128_FMA && defined GMX_DOUBLE)
-#    include "nb_kernel_avx_128_fma_double/nb_kernel_avx_128_fma_double.h"
+#    include "gromacs/gmxlib/nonbonded/nb_kernel_avx_128_fma_double/nb_kernel_avx_128_fma_double.h"
 #endif
 #if (defined GMX_SIMD_X86_AVX_256_OR_HIGHER && defined GMX_DOUBLE)
-#    include "nb_kernel_avx_256_double/nb_kernel_avx_256_double.h"
+#    include "gromacs/gmxlib/nonbonded/nb_kernel_avx_256_double/nb_kernel_avx_256_double.h"
 #endif
 #if (defined GMX_SIMD_SPARC64_HPC_ACE && defined GMX_DOUBLE)
-#    include "nb_kernel_sparc64_hpc_ace_double/nb_kernel_sparc64_hpc_ace_double.h"
+#    include "gromacs/gmxlib/nonbonded/nb_kernel_sparc64_hpc_ace_double/nb_kernel_sparc64_hpc_ace_double.h"
 #endif
 
 
@@ -163,7 +163,7 @@ gmx_nonbonded_setup(t_forcerec *   fr,
 
 
 void
-gmx_nonbonded_set_kernel_pointers(FILE *log, t_nblist *nl)
+gmx_nonbonded_set_kernel_pointers(FILE *log, t_nblist *nl, gmx_bool bElecAndVdwSwitchDiffers)
 {
     const char *     elec;
     const char *     elec_mod;
@@ -284,9 +284,28 @@ gmx_nonbonded_set_kernel_pointers(FILE *log, t_nblist *nl)
             }
         }
 
-        /* Give up. If this was a water kernel, leave the pointer as NULL, which
-         * will disable water optimization in NS. If it is a particle kernel, set
-         * the pointer to the generic NB kernel.
+        /* For now, the accelerated kernels cannot handle the combination of switch functions for both
+         * electrostatics and VdW that use different switch radius or switch cutoff distances
+         * (both of them enter in the switch function calculation). This would require
+         * us to evaluate two completely separate switch functions for every interaction.
+         * Instead, we disable such kernels by setting the pointer to NULL.
+         * This will cause the generic kernel (which can handle it) to be called instead.
+         *
+         * Note that we typically already enable tabulated coulomb interactions for this case,
+         * so this is mostly a safe-guard to make sure we call the generic kernel if the
+         * tables are disabled.
+         */
+        if ((nl->ielec != GMX_NBKERNEL_ELEC_NONE) && (nl->ielecmod == eintmodPOTSWITCH) &&
+            (nl->ivdw  != GMX_NBKERNEL_VDW_NONE)  && (nl->ivdwmod  == eintmodPOTSWITCH) &&
+            bElecAndVdwSwitchDiffers)
+        {
+            nl->kernelptr_vf = NULL;
+            nl->kernelptr_f  = NULL;
+        }
+
+        /* Give up, pick a generic one instead.
+         * We only do this for particle-particle kernels; by leaving the water-optimized kernel
+         * pointers to NULL, the water optimization will automatically be disabled for this interaction.
          */
         if (nl->kernelptr_vf == NULL && !gmx_strcasecmp_min(geom, "Particle-Particle"))
         {
@@ -298,13 +317,11 @@ gmx_nonbonded_set_kernel_pointers(FILE *log, t_nblist *nl)
                 fprintf(debug,
                         "WARNING - Slow generic NB kernel used for neighborlist with\n"
                         "    Elec: '%s', Modifier: '%s'\n"
-                        "    Vdw:  '%s', Modifier: '%s'\n"
-                        "    Geom: '%s', Other: '%s'\n\n",
-                        elec, elec_mod, vdw, vdw_mod, geom, other);
+                        "    Vdw:  '%s', Modifier: '%s'\n",
+                        elec, elec_mod, vdw, vdw_mod);
             }
         }
     }
-
     return;
 }
 
@@ -417,269 +434,12 @@ void do_nonbonded(t_forcerec *fr,
                     {
                         (*kernelptr)(&(nlist[i]), x, f, fr, mdatoms, &kernel_data, nrnb);
                     }
+                    else
+                    {
+                        gmx_fatal(FARGS, "Non-empty neighborlist does not have any kernel pointer assigned.");
+                    }
                 }
             }
         }
     }
 }
-
-static void
-nb_listed_warning_rlimit(const rvec *x, int ai, int aj, int * global_atom_index, real r, real rlimit)
-{
-    gmx_warning("Listed nonbonded interaction between particles %d and %d\n"
-                "at distance %.3f which is larger than the table limit %.3f nm.\n\n"
-                "This is likely either a 1,4 interaction, or a listed interaction inside\n"
-                "a smaller molecule you are decoupling during a free energy calculation.\n"
-                "Since interactions at distances beyond the table cannot be computed,\n"
-                "they are skipped until they are inside the table limit again. You will\n"
-                "only see this message once, even if it occurs for several interactions.\n\n"
-                "IMPORTANT: This should not happen in a stable simulation, so there is\n"
-                "probably something wrong with your system. Only change the table-extension\n"
-                "distance in the mdp file if you are really sure that is the reason.\n",
-                glatnr(global_atom_index, ai), glatnr(global_atom_index, aj), r, rlimit);
-
-    if (debug)
-    {
-        fprintf(debug,
-                "%8f %8f %8f\n%8f %8f %8f\n1-4 (%d,%d) interaction not within cut-off! r=%g. Ignored\n",
-                x[ai][XX], x[ai][YY], x[ai][ZZ], x[aj][XX], x[aj][YY], x[aj][ZZ],
-                glatnr(global_atom_index, ai), glatnr(global_atom_index, aj), r);
-    }
-}
-
-
-
-/* This might logically belong better in the nb_generic.c module, but it is only
- * used in do_nonbonded_listed(), and we want it to be inlined there to avoid an
- * extra functional call for every single pair listed in the topology.
- */
-static real
-nb_evaluate_single(real r2, real tabscale, real *vftab,
-                   real qq, real c6, real c12, real *velec, real *vvdw)
-{
-    real       rinv, r, rtab, eps, eps2, Y, F, Geps, Heps2, Fp, VVe, FFe, VVd, FFd, VVr, FFr, fscal;
-    int        ntab;
-
-    /* Do the tabulated interactions - first table lookup */
-    rinv             = gmx_invsqrt(r2);
-    r                = r2*rinv;
-    rtab             = r*tabscale;
-    ntab             = rtab;
-    eps              = rtab-ntab;
-    eps2             = eps*eps;
-    ntab             = 12*ntab;
-    /* Electrostatics */
-    Y                = vftab[ntab];
-    F                = vftab[ntab+1];
-    Geps             = eps*vftab[ntab+2];
-    Heps2            = eps2*vftab[ntab+3];
-    Fp               = F+Geps+Heps2;
-    VVe              = Y+eps*Fp;
-    FFe              = Fp+Geps+2.0*Heps2;
-    /* Dispersion */
-    Y                = vftab[ntab+4];
-    F                = vftab[ntab+5];
-    Geps             = eps*vftab[ntab+6];
-    Heps2            = eps2*vftab[ntab+7];
-    Fp               = F+Geps+Heps2;
-    VVd              = Y+eps*Fp;
-    FFd              = Fp+Geps+2.0*Heps2;
-    /* Repulsion */
-    Y                = vftab[ntab+8];
-    F                = vftab[ntab+9];
-    Geps             = eps*vftab[ntab+10];
-    Heps2            = eps2*vftab[ntab+11];
-    Fp               = F+Geps+Heps2;
-    VVr              = Y+eps*Fp;
-    FFr              = Fp+Geps+2.0*Heps2;
-
-    *velec           = qq*VVe;
-    *vvdw            = c6*VVd+c12*VVr;
-
-    fscal            = -(qq*FFe+c6*FFd+c12*FFr)*tabscale*rinv;
-
-    return fscal;
-}
-
-
-real
-do_nonbonded_listed(int ftype, int nbonds,
-                    const t_iatom iatoms[], const t_iparams iparams[],
-                    const rvec x[], rvec f[], rvec fshift[],
-                    const t_pbc *pbc, const t_graph *g,
-                    real *lambda, real *dvdl,
-                    const t_mdatoms *md,
-                    const t_forcerec *fr, gmx_grppairener_t *grppener,
-                    int *global_atom_index)
-{
-    int              ielec, ivdw;
-    real             qq, c6, c12;
-    rvec             dx;
-    ivec             dt;
-    int              i, j, itype, ai, aj, gid;
-    int              fshift_index;
-    real             r2, rinv;
-    real             fscal, velec, vvdw;
-    real *           energygrp_elec;
-    real *           energygrp_vdw;
-    static gmx_bool  warned_rlimit = FALSE;
-    /* Free energy stuff */
-    gmx_bool         bFreeEnergy;
-    real             LFC[2], LFV[2], DLF[2], lfac_coul[2], lfac_vdw[2], dlfac_coul[2], dlfac_vdw[2];
-    real             qqB, c6B, c12B, sigma2_def, sigma2_min;
-
-
-    switch (ftype)
-    {
-        case F_LJ14:
-        case F_LJC14_Q:
-            energygrp_elec = grppener->ener[egCOUL14];
-            energygrp_vdw  = grppener->ener[egLJ14];
-            break;
-        case F_LJC_PAIRS_NB:
-            energygrp_elec = grppener->ener[egCOULSR];
-            energygrp_vdw  = grppener->ener[egLJSR];
-            break;
-        default:
-            energygrp_elec = NULL; /* Keep compiler happy */
-            energygrp_vdw  = NULL; /* Keep compiler happy */
-            gmx_fatal(FARGS, "Unknown function type %d in do_nonbonded14", ftype);
-            break;
-    }
-
-    if (fr->efep != efepNO)
-    {
-        /* Lambda factor for state A=1-lambda and B=lambda */
-        LFC[0] = 1.0 - lambda[efptCOUL];
-        LFV[0] = 1.0 - lambda[efptVDW];
-        LFC[1] = lambda[efptCOUL];
-        LFV[1] = lambda[efptVDW];
-
-        /*derivative of the lambda factor for state A and B */
-        DLF[0] = -1;
-        DLF[1] = 1;
-
-        /* precalculate */
-        sigma2_def = pow(fr->sc_sigma6_def, 1.0/3.0);
-        sigma2_min = pow(fr->sc_sigma6_min, 1.0/3.0);
-
-        for (i = 0; i < 2; i++)
-        {
-            lfac_coul[i]  = (fr->sc_power == 2 ? (1-LFC[i])*(1-LFC[i]) : (1-LFC[i]));
-            dlfac_coul[i] = DLF[i]*fr->sc_power/fr->sc_r_power*(fr->sc_power == 2 ? (1-LFC[i]) : 1);
-            lfac_vdw[i]   = (fr->sc_power == 2 ? (1-LFV[i])*(1-LFV[i]) : (1-LFV[i]));
-            dlfac_vdw[i]  = DLF[i]*fr->sc_power/fr->sc_r_power*(fr->sc_power == 2 ? (1-LFV[i]) : 1);
-        }
-    }
-    else
-    {
-        sigma2_min = sigma2_def = 0;
-    }
-
-    bFreeEnergy = FALSE;
-    for (i = 0; (i < nbonds); )
-    {
-        itype = iatoms[i++];
-        ai    = iatoms[i++];
-        aj    = iatoms[i++];
-        gid   = GID(md->cENER[ai], md->cENER[aj], md->nenergrp);
-
-        /* Get parameters */
-        switch (ftype)
-        {
-            case F_LJ14:
-                bFreeEnergy =
-                    (fr->efep != efepNO &&
-                     ((md->nPerturbed && (md->bPerturbed[ai] || md->bPerturbed[aj])) ||
-                      iparams[itype].lj14.c6A != iparams[itype].lj14.c6B ||
-                      iparams[itype].lj14.c12A != iparams[itype].lj14.c12B));
-                qq               = md->chargeA[ai]*md->chargeA[aj]*fr->epsfac*fr->fudgeQQ;
-                c6               = iparams[itype].lj14.c6A;
-                c12              = iparams[itype].lj14.c12A;
-                break;
-            case F_LJC14_Q:
-                qq               = iparams[itype].ljc14.qi*iparams[itype].ljc14.qj*fr->epsfac*iparams[itype].ljc14.fqq;
-                c6               = iparams[itype].ljc14.c6;
-                c12              = iparams[itype].ljc14.c12;
-                break;
-            case F_LJC_PAIRS_NB:
-                qq               = iparams[itype].ljcnb.qi*iparams[itype].ljcnb.qj*fr->epsfac;
-                c6               = iparams[itype].ljcnb.c6;
-                c12              = iparams[itype].ljcnb.c12;
-                break;
-            default:
-                /* Cannot happen since we called gmx_fatal() above in this case */
-                qq = c6 = c12 = 0; /* Keep compiler happy */
-                break;
-        }
-
-        /* To save flops in the optimized kernels, c6/c12 have 6.0/12.0 derivative prefactors
-         * included in the general nfbp array now. This means the tables are scaled down by the
-         * same factor, so when we use the original c6/c12 parameters from iparams[] they must
-         * be scaled up.
-         */
-        c6  *= 6.0;
-        c12 *= 12.0;
-
-        /* Do we need to apply full periodic boundary conditions? */
-        if (fr->bMolPBC == TRUE)
-        {
-            fshift_index = pbc_dx_aiuc(pbc, x[ai], x[aj], dx);
-        }
-        else
-        {
-            fshift_index = CENTRAL;
-            rvec_sub(x[ai], x[aj], dx);
-        }
-        r2           = norm2(dx);
-
-        if (r2 >= fr->tab14.r*fr->tab14.r)
-        {
-            if (warned_rlimit == FALSE)
-            {
-                nb_listed_warning_rlimit(x, ai, aj, global_atom_index, sqrt(r2), fr->tab14.r);
-                warned_rlimit = TRUE;
-            }
-            continue;
-        }
-
-        if (bFreeEnergy)
-        {
-            /* Currently free energy is only supported for F_LJ14, so no need to check for that if we got here */
-            qqB              = md->chargeB[ai]*md->chargeB[aj]*fr->epsfac*fr->fudgeQQ;
-            c6B              = iparams[itype].lj14.c6B*6.0;
-            c12B             = iparams[itype].lj14.c12B*12.0;
-
-            fscal            = nb_free_energy_evaluate_single(r2, fr->sc_r_power, fr->sc_alphacoul, fr->sc_alphavdw,
-                                                              fr->tab14.scale, fr->tab14.data, qq, c6, c12, qqB, c6B, c12B,
-                                                              LFC, LFV, DLF, lfac_coul, lfac_vdw, dlfac_coul, dlfac_vdw,
-                                                              fr->sc_sigma6_def, fr->sc_sigma6_min, sigma2_def, sigma2_min, &velec, &vvdw, dvdl);
-        }
-        else
-        {
-            /* Evaluate tabulated interaction without free energy */
-            fscal            = nb_evaluate_single(r2, fr->tab14.scale, fr->tab14.data, qq, c6, c12, &velec, &vvdw);
-        }
-
-        energygrp_elec[gid]  += velec;
-        energygrp_vdw[gid]   += vvdw;
-        svmul(fscal, dx, dx);
-
-        /* Add the forces */
-        rvec_inc(f[ai], dx);
-        rvec_dec(f[aj], dx);
-
-        if (g)
-        {
-            /* Correct the shift forces using the graph */
-            ivec_sub(SHIFT_IVEC(g, ai), SHIFT_IVEC(g, aj), dt);
-            fshift_index = IVEC2IS(dt);
-        }
-        if (fshift_index != CENTRAL)
-        {
-            rvec_inc(fshift[fshift_index], dx);
-            rvec_dec(fshift[CENTRAL], dx);
-        }
-    }
-    return 0.0;
-}