Further improve getDDGridSetup

[gromacs.git] / src / gromacs / mdlib / shake.cpp

/*
 * This file is part of the GROMACS molecular simulation package.
 *
 * Copyright (c) 1991-2000, University of Groningen, The Netherlands.
 * Copyright (c) 2001-2004, The GROMACS development team.
 * Copyright (c) 2013,2014,2015,2017,2018,2019, by the GROMACS development team, led by
 * Mark Abraham, David van der Spoel, Berk Hess, and Erik Lindahl,
 * and including many others, as listed in the AUTHORS file in the
 * top-level source directory and at http://www.gromacs.org.
 *
 * GROMACS is free software; you can redistribute it and/or
 * modify it under the terms of the GNU Lesser General Public License
 * as published by the Free Software Foundation; either version 2.1
 * of the License, or (at your option) any later version.
 *
 * GROMACS is distributed in the hope that it will be useful,
 * but WITHOUT ANY WARRANTY; without even the implied warranty of
 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
 * Lesser General Public License for more details.
 *
 * You should have received a copy of the GNU Lesser General Public
 * License along with GROMACS; if not, see
 * http://www.gnu.org/licenses, or write to the Free Software Foundation,
 * Inc., 51 Franklin Street, Fifth Floor, Boston, MA  02110-1301  USA.
 *
 * If you want to redistribute modifications to GROMACS, please
 * consider that scientific software is very special. Version
 * control is crucial - bugs must be traceable. We will be happy to
 * consider code for inclusion in the official distribution, but
 * derived work must not be called official GROMACS. Details are found
 * in the README & COPYING files - if they are missing, get the
 * official version at http://www.gromacs.org.
 *
 * To help us fund GROMACS development, we humbly ask that you cite
 * the research papers on the package. Check out http://www.gromacs.org.
 */
/*! \internal \file
 * \brief Defines SHAKE code.
 *
 * \author David van der Spoel <david.vanderspoel@icm.uu.se>
 * \author Berk Hess <hess@kth.se>
 * \author Mark Abraham <mark.j.abraham@gmail.com>
 * \ingroup module_mdlib
 */
#include "gmxpre.h"

#include "shake.h"

#include <cmath>

#include <algorithm>

#include "gromacs/domdec/domdec_struct.h"
#include "gromacs/gmxlib/nrnb.h"
#include "gromacs/math/functions.h"
#include "gromacs/math/vec.h"
#include "gromacs/mdlib/constr.h"
#include "gromacs/mdlib/splitter.h"
#include "gromacs/mdtypes/inputrec.h"
#include "gromacs/mdtypes/md_enums.h"
#include "gromacs/mdtypes/mdatom.h"
#include "gromacs/topology/invblock.h"
#include "gromacs/utility/fatalerror.h"
#include "gromacs/utility/smalloc.h"

namespace gmx
{

struct shakedata
{
    rvec *rij;
    real *half_of_reduced_mass;
    real *distance_squared_tolerance;
    real *constraint_distance_squared;
    int   nalloc;
    /* SOR stuff */
    real  delta;
    real  omega;
    real  gamma;
    int   nblocks;        /* The number of SHAKE blocks         */
    int  *sblock;         /* The SHAKE blocks                   */
    int   sblock_nalloc;  /* The allocation size of sblock      */
    /*! \brief Scaled Lagrange multiplier for each constraint.
     *
     * Value is -2 * eta from p. 336 of the paper, divided by the
     * constraint distance. */
    real *scaled_lagrange_multiplier;
    int   lagr_nalloc;    /* The allocation size of scaled_lagrange_multiplier */
};

shakedata *shake_init()
{
    shakedata *d;

    snew(d, 1);

    d->nalloc                      = 0;
    d->rij                         = nullptr;
    d->half_of_reduced_mass        = nullptr;
    d->distance_squared_tolerance  = nullptr;
    d->constraint_distance_squared = nullptr;

    /* SOR initialization */
    d->delta = 0.1;
    d->omega = 1.0;
    d->gamma = 1000000;

    return d;
}

void done_shake(shakedata *d)
{
    sfree(d->rij);
    sfree(d->half_of_reduced_mass);
    sfree(d->distance_squared_tolerance);
    sfree(d->constraint_distance_squared);
    sfree(d->sblock);
    sfree(d->scaled_lagrange_multiplier);
    sfree(d);
}

typedef struct {
    int iatom[3];
    int blocknr;
} t_sortblock;

//! Compares sort blocks.
static int pcomp(const void *p1, const void *p2)
{
    int                db;
    int                min1, min2, max1, max2;
    const t_sortblock *a1 = reinterpret_cast<const t_sortblock*>(p1);
    const t_sortblock *a2 = reinterpret_cast<const t_sortblock*>(p2);

    db = a1->blocknr-a2->blocknr;

    if (db != 0)
    {
        return db;
    }

    min1 = std::min(a1->iatom[1], a1->iatom[2]);
    max1 = std::max(a1->iatom[1], a1->iatom[2]);
    min2 = std::min(a2->iatom[1], a2->iatom[2]);
    max2 = std::max(a2->iatom[1], a2->iatom[2]);

    if (min1 == min2)
    {
        return max1-max2;
    }
    else
    {
        return min1-min2;
    }
}

//! Prints sortblocks
static void pr_sortblock(FILE *fp, const char *title, int nsb, t_sortblock sb[])
{
    int i;

    fprintf(fp, "%s\n", title);
    for (i = 0; (i < nsb); i++)
    {
        fprintf(fp, "i: %5d, iatom: (%5d %5d %5d), blocknr: %5d\n",
                i, sb[i].iatom[0], sb[i].iatom[1], sb[i].iatom[2],
                sb[i].blocknr);
    }
}

//! Reallocates a vector.
static void resizeLagrangianData(shakedata *shaked, int ncons)
{
    if (ncons > shaked->lagr_nalloc)
    {
        shaked->lagr_nalloc = over_alloc_dd(ncons);
        srenew(shaked->scaled_lagrange_multiplier, shaked->lagr_nalloc);
    }
}

void
make_shake_sblock_serial(shakedata *shaked,
                         const t_idef *idef, const t_mdatoms &md)
{
    int          i, j, m, ncons;
    int          bstart, bnr;
    t_blocka     sblocks;
    t_sortblock *sb;
    t_iatom     *iatom;
    int         *inv_sblock;

    /* Since we are processing the local topology,
     * the F_CONSTRNC ilist has been concatenated to the F_CONSTR ilist.
     */
    ncons = idef->il[F_CONSTR].nr/3;

    init_blocka(&sblocks);
    sfree(sblocks.index); // To solve memory leak
    gen_sblocks(nullptr, 0, md.homenr, idef, &sblocks, FALSE);

    /*
       bstart=(idef->nodeid > 0) ? blocks->multinr[idef->nodeid-1] : 0;
       nblocks=blocks->multinr[idef->nodeid] - bstart;
     */
    bstart          = 0;
    shaked->nblocks = sblocks.nr;
    if (debug)
    {
        fprintf(debug, "ncons: %d, bstart: %d, nblocks: %d\n",
                ncons, bstart, shaked->nblocks);
    }

    /* Calculate block number for each atom */
    inv_sblock = make_invblocka(&sblocks, md.nr);

    done_blocka(&sblocks);

    /* Store the block number in temp array and
     * sort the constraints in order of the sblock number
     * and the atom numbers, really sorting a segment of the array!
     */
    iatom = idef->il[F_CONSTR].iatoms;
    snew(sb, ncons);
    for (i = 0; (i < ncons); i++, iatom += 3)
    {
        for (m = 0; (m < 3); m++)
        {
            sb[i].iatom[m] = iatom[m];
        }
        sb[i].blocknr = inv_sblock[iatom[1]];
    }

    /* Now sort the blocks */
    if (debug)
    {
        pr_sortblock(debug, "Before sorting", ncons, sb);
        fprintf(debug, "Going to sort constraints\n");
    }

    std::qsort(sb, ncons, sizeof(*sb), pcomp);

    if (debug)
    {
        pr_sortblock(debug, "After sorting", ncons, sb);
    }

    iatom = idef->il[F_CONSTR].iatoms;
    for (i = 0; (i < ncons); i++, iatom += 3)
    {
        for (m = 0; (m < 3); m++)
        {
            iatom[m] = sb[i].iatom[m];
        }
    }

    j = 0;
    snew(shaked->sblock, shaked->nblocks+1);
    bnr = -2;
    for (i = 0; (i < ncons); i++)
    {
        if (sb[i].blocknr != bnr)
        {
            bnr                 = sb[i].blocknr;
            shaked->sblock[j++] = 3*i;
        }
    }
    /* Last block... */
    shaked->sblock[j++] = 3*ncons;

    if (j != (shaked->nblocks+1))
    {
        fprintf(stderr, "bstart: %d\n", bstart);
        fprintf(stderr, "j: %d, nblocks: %d, ncons: %d\n",
                j, shaked->nblocks, ncons);
        for (i = 0; (i < ncons); i++)
        {
            fprintf(stderr, "i: %5d  sb[i].blocknr: %5d\n", i, sb[i].blocknr);
        }
        for (j = 0; (j <= shaked->nblocks); j++)
        {
            fprintf(stderr, "sblock[%3d]=%5d\n", j, shaked->sblock[j]);
        }
        gmx_fatal(FARGS, "DEATH HORROR: "
                  "sblocks does not match idef->il[F_CONSTR]");
    }
    sfree(sb);
    sfree(inv_sblock);
    resizeLagrangianData(shaked, ncons);
}

// TODO: Check if this code is useful. It might never be called.
void
make_shake_sblock_dd(shakedata          *shaked,
                     const t_ilist      *ilcon,
                     const gmx_domdec_t *dd)
{
    int      ncons, c, cg;
    t_iatom *iatom;

    if (dd->ncg_home+1 > shaked->sblock_nalloc)
    {
        shaked->sblock_nalloc = over_alloc_dd(dd->ncg_home+1);
        srenew(shaked->sblock, shaked->sblock_nalloc);
    }

    ncons           = ilcon->nr/3;
    iatom           = ilcon->iatoms;
    shaked->nblocks = 0;
    cg              = 0;
    for (c = 0; c < ncons; c++)
    {
        if (c == 0 || iatom[1] >= cg + 1)
        {
            shaked->sblock[shaked->nblocks++] = 3*c;
            while (iatom[1] >= cg + 1)
            {
                cg++;
            }
        }
        iatom += 3;
    }
    shaked->sblock[shaked->nblocks] = 3*ncons;
    resizeLagrangianData(shaked, ncons);
}

/*! \brief Inner kernel for SHAKE constraints
 *
 * Original implementation from R.C. van Schaik and W.F. van Gunsteren
 * (ETH Zuerich, June 1992), adapted for GROMACS by David van der
 * Spoel November 1992.
 *
 * The algorithm here is based section five of Ryckaert, Ciccotti and
 * Berendsen, J Comp Phys, 23, 327, 1977.
 *
 * \param[in]    iatom                         Mini-topology of triples of constraint type (unused in this
 *                                             function) and indices of the two atoms involved
 * \param[in]    ncon                          Number of constraints
 * \param[out]   nnit                          Number of iterations performed
 * \param[in]    maxnit                        Maximum number of iterations permitted
 * \param[in]    constraint_distance_squared   The objective value for each constraint
 * \param[inout] positions                     The initial (and final) values of the positions of all atoms
 * \param[in]    initial_displacements         The initial displacements of each constraint
 * \param[in]    half_of_reduced_mass          Half of the reduced mass for each constraint
 * \param[in]    omega                         SHAKE over-relaxation factor (set non-1.0 by
 *                                             using shake-sor=yes in the .mdp, but there is no documentation anywhere)
 * \param[in]    invmass                       Inverse mass of each atom
 * \param[in]    distance_squared_tolerance    Multiplicative tolerance on the difference in the
 *                                             square of the constrained distance (see code)
 * \param[out]   scaled_lagrange_multiplier    Scaled Lagrange multiplier for each constraint (-2 * eta from p. 336
 *                                             of the paper, divided by the constraint distance)
 * \param[out]   nerror                        Zero upon success, returns one more than the index of the
 *                                             problematic constraint if the input was malformed
 *
 * \todo Make SHAKE use better data structures, in particular for iatom. */
void cshake(const int iatom[], int ncon, int *nnit, int maxnit,
            const real constraint_distance_squared[], real positions[],
            const real initial_displacements[], const real half_of_reduced_mass[], real omega,
            const real invmass[], const real distance_squared_tolerance[],
            real scaled_lagrange_multiplier[], int *nerror)
{
    /* default should be increased! MRS 8/4/2009 */
    const real mytol = 1e-10;

    int        ll, i, j, i3, j3, l3;
    int        ix, iy, iz, jx, jy, jz;
    real       r_dot_r_prime;
    real       constraint_distance_squared_ll;
    real       r_prime_squared;
    real       scaled_lagrange_multiplier_ll;
    real       r_prime_x, r_prime_y, r_prime_z, diff, im, jm;
    real       xh, yh, zh, rijx, rijy, rijz;
    int        nit, error, nconv;
    real       iconvf;

    // TODO nconv is used solely as a boolean, so we should write the
    // code like that
    error = 0;
    nconv = 1;
    for (nit = 0; (nit < maxnit) && (nconv != 0) && (error == 0); nit++)
    {
        nconv = 0;
        for (ll = 0; (ll < ncon) && (error == 0); ll++)
        {
            l3    = 3*ll;
            rijx  = initial_displacements[l3+XX];
            rijy  = initial_displacements[l3+YY];
            rijz  = initial_displacements[l3+ZZ];
            i     = iatom[l3+1];
            j     = iatom[l3+2];
            i3    = 3*i;
            j3    = 3*j;
            ix    = i3+XX;
            iy    = i3+YY;
            iz    = i3+ZZ;
            jx    = j3+XX;
            jy    = j3+YY;
            jz    = j3+ZZ;

            /* Compute r prime between atoms i and j, which is the
               displacement *before* this update stage */
            r_prime_x       = positions[ix]-positions[jx];
            r_prime_y       = positions[iy]-positions[jy];
            r_prime_z       = positions[iz]-positions[jz];
            r_prime_squared = (r_prime_x * r_prime_x +
                               r_prime_y * r_prime_y +
                               r_prime_z * r_prime_z);
            constraint_distance_squared_ll = constraint_distance_squared[ll];
            diff    = constraint_distance_squared_ll - r_prime_squared;

            /* iconvf is less than 1 when the error is smaller than a bound */
            iconvf = fabs(diff) * distance_squared_tolerance[ll];

            if (iconvf > 1.0)
            {
                nconv         = static_cast<int>(iconvf);
                r_dot_r_prime = (rijx * r_prime_x +
                                 rijy * r_prime_y +
                                 rijz * r_prime_z);

                if (r_dot_r_prime < constraint_distance_squared_ll * mytol)
                {
                    error = ll+1;
                }
                else
                {
                    /* The next line solves equation 5.6 (neglecting
                       the term in g^2), for g */
                    scaled_lagrange_multiplier_ll   = omega*diff*half_of_reduced_mass[ll]/r_dot_r_prime;
                    scaled_lagrange_multiplier[ll] += scaled_lagrange_multiplier_ll;
                    xh                              = rijx * scaled_lagrange_multiplier_ll;
                    yh                              = rijy * scaled_lagrange_multiplier_ll;
                    zh                              = rijz * scaled_lagrange_multiplier_ll;
                    im                              = invmass[i];
                    jm                              = invmass[j];
                    positions[ix]                  += xh*im;
                    positions[iy]                  += yh*im;
                    positions[iz]                  += zh*im;
                    positions[jx]                  -= xh*jm;
                    positions[jy]                  -= yh*jm;
                    positions[jz]                  -= zh*jm;
                }
            }
        }
    }
    *nnit   = nit;
    *nerror = error;
}

//! Implements RATTLE (ie. SHAKE for velocity verlet integrators)
static void
crattle(const int iatom[], int ncon, int *nnit, int maxnit,
        const real constraint_distance_squared[], real vp[], const real rij[], const real m2[], real omega,
        const real invmass[], const real distance_squared_tolerance[], real scaled_lagrange_multiplier[],
        int *nerror, real invdt)
{
    /*
     *     r.c. van schaik and w.f. van gunsteren
     *     eth zuerich
     *     june 1992
     *     Adapted for use with Gromacs by David van der Spoel november 92 and later.
     *     rattle added by M.R. Shirts, April 2004, from code written by Jay Ponder in TINKER
     *     second part of rattle algorithm
     */

    int          ll, i, j, i3, j3, l3;
    int          ix, iy, iz, jx, jy, jz;
    real         constraint_distance_squared_ll;
    real         vpijd, vx, vy, vz, acor, fac, im, jm;
    real         xh, yh, zh, rijx, rijy, rijz;
    int          nit, error, nconv;
    real         iconvf;

    // TODO nconv is used solely as a boolean, so we should write the
    // code like that
    error = 0;
    nconv = 1;
    for (nit = 0; (nit < maxnit) && (nconv != 0) && (error == 0); nit++)
    {
        nconv = 0;
        for (ll = 0; (ll < ncon) && (error == 0); ll++)
        {
            l3      = 3*ll;
            rijx    = rij[l3+XX];
            rijy    = rij[l3+YY];
            rijz    = rij[l3+ZZ];
            i       = iatom[l3+1];
            j       = iatom[l3+2];
            i3      = 3*i;
            j3      = 3*j;
            ix      = i3+XX;
            iy      = i3+YY;
            iz      = i3+ZZ;
            jx      = j3+XX;
            jy      = j3+YY;
            jz      = j3+ZZ;
            vx      = vp[ix]-vp[jx];
            vy      = vp[iy]-vp[jy];
            vz      = vp[iz]-vp[jz];

            vpijd   = vx*rijx+vy*rijy+vz*rijz;
            constraint_distance_squared_ll = constraint_distance_squared[ll];

            /* iconv is zero when the error is smaller than a bound */
            iconvf   = fabs(vpijd)*(distance_squared_tolerance[ll]/invdt);

            if (iconvf > 1)
            {
                nconv     = static_cast<int>(iconvf);
                fac       = omega*2.0*m2[ll]/constraint_distance_squared_ll;
                acor      = -fac*vpijd;
                scaled_lagrange_multiplier[ll] += acor;
                xh        = rijx*acor;
                yh        = rijy*acor;
                zh        = rijz*acor;

                im        = invmass[i];
                jm        = invmass[j];

                vp[ix] += xh*im;
                vp[iy] += yh*im;
                vp[iz] += zh*im;
                vp[jx] -= xh*jm;
                vp[jy] -= yh*jm;
                vp[jz] -= zh*jm;
            }
        }
    }
    *nnit   = nit;
    *nerror = error;
}

//! Applies SHAKE
static int vec_shakef(FILE *fplog, shakedata *shaked,
                      const real invmass[], int ncon,
                      t_iparams ip[], t_iatom *iatom,
                      real tol, const rvec x[], rvec prime[], real omega,
                      bool bFEP, real lambda, real scaled_lagrange_multiplier[],
                      real invdt, rvec *v,
                      bool bCalcVir, tensor vir_r_m_dr, ConstraintVariable econq)
{
    rvec    *rij;
    real    *half_of_reduced_mass, *distance_squared_tolerance, *constraint_distance_squared;
    int      maxnit = 1000;
    int      nit    = 0, ll, i, j, d, d2, type;
    t_iatom *ia;
    real     L1;
    real     mm    = 0., tmp;
    int      error = 0;
    real     constraint_distance;

    if (ncon > shaked->nalloc)
    {
        shaked->nalloc = over_alloc_dd(ncon);
        srenew(shaked->rij, shaked->nalloc);
        srenew(shaked->half_of_reduced_mass, shaked->nalloc);
        srenew(shaked->distance_squared_tolerance, shaked->nalloc);
        srenew(shaked->constraint_distance_squared, shaked->nalloc);
    }
    rij                          = shaked->rij;
    half_of_reduced_mass         = shaked->half_of_reduced_mass;
    distance_squared_tolerance   = shaked->distance_squared_tolerance;
    constraint_distance_squared  = shaked->constraint_distance_squared;

    L1   = 1.0-lambda;
    ia   = iatom;
    for (ll = 0; (ll < ncon); ll++, ia += 3)
    {
        type  = ia[0];
        i     = ia[1];
        j     = ia[2];

        mm                       = 2.0*(invmass[i]+invmass[j]);
        rij[ll][XX]              = x[i][XX]-x[j][XX];
        rij[ll][YY]              = x[i][YY]-x[j][YY];
        rij[ll][ZZ]              = x[i][ZZ]-x[j][ZZ];
        half_of_reduced_mass[ll] = 1.0/mm;
        if (bFEP)
        {
            constraint_distance = L1*ip[type].constr.dA + lambda*ip[type].constr.dB;
        }
        else
        {
            constraint_distance = ip[type].constr.dA;
        }
        constraint_distance_squared[ll]  = gmx::square(constraint_distance);
        distance_squared_tolerance[ll]   = 0.5/(constraint_distance_squared[ll]*tol);
    }

    switch (econq)
    {
        case ConstraintVariable::Positions:
            cshake(iatom, ncon, &nit, maxnit, constraint_distance_squared, prime[0], rij[0], half_of_reduced_mass, omega, invmass, distance_squared_tolerance, scaled_lagrange_multiplier, &error);
            break;
        case ConstraintVariable::Velocities:
            crattle(iatom, ncon, &nit, maxnit, constraint_distance_squared, prime[0], rij[0], half_of_reduced_mass, omega, invmass, distance_squared_tolerance, scaled_lagrange_multiplier, &error, invdt);
            break;
        default:
            gmx_incons("Unknown constraint quantity for SHAKE");
    }

    if (nit >= maxnit)
    {
        if (fplog)
        {
            fprintf(fplog, "Shake did not converge in %d steps\n", maxnit);
        }
        fprintf(stderr, "Shake did not converge in %d steps\n", maxnit);
        nit = 0;
    }
    else if (error != 0)
    {
        if (fplog)
        {
            fprintf(fplog, "Inner product between old and new vector <= 0.0!\n"
                    "constraint #%d atoms %d and %d\n",
                    error-1, iatom[3*(error-1)+1]+1, iatom[3*(error-1)+2]+1);
        }
        fprintf(stderr, "Inner product between old and new vector <= 0.0!\n"
                "constraint #%d atoms %d and %d\n",
                error-1, iatom[3*(error-1)+1]+1, iatom[3*(error-1)+2]+1);
        nit = 0;
    }

    /* Constraint virial and correct the Lagrange multipliers for the length */

    ia = iatom;

    for (ll = 0; (ll < ncon); ll++, ia += 3)
    {
        type  = ia[0];
        i     = ia[1];
        j     = ia[2];

        if ((econq == ConstraintVariable::Positions) && v != nullptr)
        {
            /* Correct the velocities */
            mm = scaled_lagrange_multiplier[ll]*invmass[i]*invdt;
            for (d = 0; d < DIM; d++)
            {
                v[ia[1]][d] += mm*rij[ll][d];
            }
            mm = scaled_lagrange_multiplier[ll]*invmass[j]*invdt;
            for (d = 0; d < DIM; d++)
            {
                v[ia[2]][d] -= mm*rij[ll][d];
            }
            /* 16 flops */
        }

        /* constraint virial */
        if (bCalcVir)
        {
            mm = scaled_lagrange_multiplier[ll];
            for (d = 0; d < DIM; d++)
            {
                tmp = mm*rij[ll][d];
                for (d2 = 0; d2 < DIM; d2++)
                {
                    vir_r_m_dr[d][d2] -= tmp*rij[ll][d2];
                }
            }
            /* 21 flops */
        }

        /* cshake and crattle produce Lagrange multipliers scaled by
           the reciprocal of the constraint length, so fix that */
        if (bFEP)
        {
            constraint_distance = L1*ip[type].constr.dA + lambda*ip[type].constr.dB;
        }
        else
        {
            constraint_distance = ip[type].constr.dA;
        }
        scaled_lagrange_multiplier[ll] *= constraint_distance;
    }

    return nit;
}

//! Check that constraints are satisfied.
static void check_cons(FILE *log, int nc, const rvec x[], rvec prime[], rvec v[],
                       t_iparams ip[], t_iatom *iatom,
                       const real invmass[], ConstraintVariable econq)
{
    t_iatom *ia;
    int      ai, aj;
    int      i;
    real     d, dp;
    rvec     dx, dv;

    GMX_ASSERT(v, "Input has to be non-null");
    fprintf(log,
            "    i     mi      j     mj      before       after   should be\n");
    ia = iatom;
    for (i = 0; (i < nc); i++, ia += 3)
    {
        ai = ia[1];
        aj = ia[2];
        rvec_sub(x[ai], x[aj], dx);
        d = norm(dx);

        switch (econq)
        {
            case ConstraintVariable::Positions:
                rvec_sub(prime[ai], prime[aj], dx);
                dp = norm(dx);
                fprintf(log, "%5d  %5.2f  %5d  %5.2f  %10.5f  %10.5f  %10.5f\n",
                        ai+1, 1.0/invmass[ai],
                        aj+1, 1.0/invmass[aj], d, dp, ip[ia[0]].constr.dA);
                break;
            case ConstraintVariable::Velocities:
                rvec_sub(v[ai], v[aj], dv);
                d = iprod(dx, dv);
                rvec_sub(prime[ai], prime[aj], dv);
                dp = iprod(dx, dv);
                fprintf(log, "%5d  %5.2f  %5d  %5.2f  %10.5f  %10.5f  %10.5f\n",
                        ai+1, 1.0/invmass[ai],
                        aj+1, 1.0/invmass[aj], d, dp, 0.);
                break;
            default:
                gmx_incons("Unknown constraint quantity for SHAKE");
        }
    }
}

//! Applies SHAKE.
static bool
bshakef(FILE *log, shakedata *shaked,
        const real invmass[],
        const t_idef &idef, const t_inputrec &ir, const rvec x_s[], rvec prime[],
        t_nrnb *nrnb, real lambda, real *dvdlambda,
        real invdt, rvec *v, bool bCalcVir, tensor vir_r_m_dr,
        bool bDumpOnError, ConstraintVariable econq)
{
    t_iatom *iatoms;
    real    *lam, dt_2, dvdl;
    int      i, n0, ncon, blen, type, ll;
    int      tnit = 0, trij = 0;

    ncon = idef.il[F_CONSTR].nr/3;

    for (ll = 0; ll < ncon; ll++)
    {
        shaked->scaled_lagrange_multiplier[ll] = 0;
    }

    // TODO Rewrite this block so that it is obvious that i, iatoms
    // and lam are all iteration variables. Is this easier if the
    // sblock data structure is organized differently?
    iatoms = &(idef.il[F_CONSTR].iatoms[shaked->sblock[0]]);
    lam    = shaked->scaled_lagrange_multiplier;
    for (i = 0; (i < shaked->nblocks); )
    {
        blen  = (shaked->sblock[i+1]-shaked->sblock[i]);
        blen /= 3;
        n0    = vec_shakef(log, shaked, invmass, blen, idef.iparams,
                           iatoms, ir.shake_tol, x_s, prime, shaked->omega,
                           ir.efep != efepNO, lambda, lam, invdt, v, bCalcVir, vir_r_m_dr,
                           econq);

        if (n0 == 0)
        {
            if (bDumpOnError && log)
            {
                {
                    check_cons(log, blen, x_s, prime, v, idef.iparams, iatoms, invmass, econq);
                }
            }
            return FALSE;
        }
        tnit                       += n0*blen;
        trij                       += blen;
        iatoms                     += 3*blen; /* Increment pointer! */
        lam                        += blen;
        i++;
    }
    /* only for position part? */
    if (econq == ConstraintVariable::Positions)
    {
        if (ir.efep != efepNO)
        {
            real bondA, bondB;
            /* TODO This should probably use invdt, so that sd integrator scaling works properly */
            dt_2 = 1/gmx::square(ir.delta_t);
            dvdl = 0;
            for (ll = 0; ll < ncon; ll++)
            {
                type  = idef.il[F_CONSTR].iatoms[3*ll];

                /* Per equations in the manual, dv/dl = -2 \sum_ll lagrangian_ll * r_ll * (d_B - d_A) */
                /* The vector scaled_lagrange_multiplier[ll] contains the value -2 r_ll eta_ll (eta_ll is the
                   estimate of the Langrangian, definition on page 336 of Ryckaert et al 1977),
                   so the pre-factors are already present. */
                bondA = idef.iparams[type].constr.dA;
                bondB = idef.iparams[type].constr.dB;
                dvdl += shaked->scaled_lagrange_multiplier[ll] * dt_2 * (bondB - bondA);
            }
            *dvdlambda += dvdl;
        }
    }
    if (ir.bShakeSOR)
    {
        if (tnit > shaked->gamma)
        {
            shaked->delta *= -0.5;
        }
        shaked->omega += shaked->delta;
        shaked->gamma  = tnit;
    }
    inc_nrnb(nrnb, eNR_SHAKE, tnit);
    inc_nrnb(nrnb, eNR_SHAKE_RIJ, trij);
    if (v)
    {
        inc_nrnb(nrnb, eNR_CONSTR_V, trij*2);
    }
    if (bCalcVir)
    {
        inc_nrnb(nrnb, eNR_CONSTR_VIR, trij);
    }

    return TRUE;
}

bool
constrain_shake(FILE              *log,
                shakedata         *shaked,
                const real         invmass[],
                const t_idef      &idef,
                const t_inputrec  &ir,
                const rvec         x_s[],
                rvec               xprime[],
                rvec               vprime[],
                t_nrnb            *nrnb,
                real               lambda,
                real              *dvdlambda,
                real               invdt,
                rvec              *v,
                bool               bCalcVir,
                tensor             vir_r_m_dr,
                bool               bDumpOnError,
                ConstraintVariable econq)
{
    if (shaked->nblocks == 0)
    {
        return true;
    }
    bool bOK;
    switch (econq)
    {
        case (ConstraintVariable::Positions):
            bOK = bshakef(log, shaked,
                          invmass,
                          idef, ir, x_s, xprime, nrnb,
                          lambda, dvdlambda,
                          invdt, v, bCalcVir, vir_r_m_dr,
                          bDumpOnError, econq);
            break;
        case (ConstraintVariable::Velocities):
            bOK = bshakef(log, shaked,
                          invmass,
                          idef, ir, x_s, vprime, nrnb,
                          lambda, dvdlambda,
                          invdt, nullptr, bCalcVir, vir_r_m_dr,
                          bDumpOnError, econq);
            break;
        default:
            gmx_fatal(FARGS, "Internal error, SHAKE called for constraining something else than coordinates");
    }
    return bOK;
}

}  // namespace gmx