2 * This file is part of the GROMACS molecular simulation package.
4 * Copyright (c) 1991-2000, University of Groningen, The Netherlands.
5 * Copyright (c) 2001-2004, The GROMACS development team.
6 * Copyright (c) 2013,2014,2015,2016,2017 by the GROMACS development team.
7 * Copyright (c) 2018,2019,2020, by the GROMACS development team, led by
8 * Mark Abraham, David van der Spoel, Berk Hess, and Erik Lindahl,
9 * and including many others, as listed in the AUTHORS file in the
10 * top-level source directory and at http://www.gromacs.org.
12 * GROMACS is free software; you can redistribute it and/or
13 * modify it under the terms of the GNU Lesser General Public License
14 * as published by the Free Software Foundation; either version 2.1
15 * of the License, or (at your option) any later version.
17 * GROMACS is distributed in the hope that it will be useful,
18 * but WITHOUT ANY WARRANTY; without even the implied warranty of
19 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
20 * Lesser General Public License for more details.
22 * You should have received a copy of the GNU Lesser General Public
23 * License along with GROMACS; if not, see
24 * http://www.gnu.org/licenses, or write to the Free Software Foundation,
25 * Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA.
27 * If you want to redistribute modifications to GROMACS, please
28 * consider that scientific software is very special. Version
29 * control is crucial - bugs must be traceable. We will be happy to
30 * consider code for inclusion in the official distribution, but
31 * derived work must not be called official GROMACS. Details are found
32 * in the README & COPYING files - if they are missing, get the
33 * official version at http://www.gromacs.org.
35 * To help us fund GROMACS development, we humbly ask that you cite
36 * the research papers on the package. Check out http://www.gromacs.org.
48 #include "gromacs/domdec/domdec_struct.h"
49 #include "gromacs/fileio/confio.h"
50 #include "gromacs/gmxlib/network.h"
51 #include "gromacs/gmxlib/nrnb.h"
52 #include "gromacs/listed_forces/disre.h"
53 #include "gromacs/listed_forces/orires.h"
54 #include "gromacs/math/functions.h"
55 #include "gromacs/math/invertmatrix.h"
56 #include "gromacs/math/paddedvector.h"
57 #include "gromacs/math/units.h"
58 #include "gromacs/math/vec.h"
59 #include "gromacs/math/vecdump.h"
60 #include "gromacs/mdlib/boxdeformation.h"
61 #include "gromacs/mdlib/constr.h"
62 #include "gromacs/mdlib/gmx_omp_nthreads.h"
63 #include "gromacs/mdlib/mdatoms.h"
64 #include "gromacs/mdlib/stat.h"
65 #include "gromacs/mdlib/tgroup.h"
66 #include "gromacs/mdtypes/commrec.h"
67 #include "gromacs/mdtypes/group.h"
68 #include "gromacs/mdtypes/inputrec.h"
69 #include "gromacs/mdtypes/md_enums.h"
70 #include "gromacs/mdtypes/mdatom.h"
71 #include "gromacs/mdtypes/state.h"
72 #include "gromacs/pbcutil/boxutilities.h"
73 #include "gromacs/pbcutil/mshift.h"
74 #include "gromacs/pbcutil/pbc.h"
75 #include "gromacs/pulling/pull.h"
76 #include "gromacs/random/tabulatednormaldistribution.h"
77 #include "gromacs/random/threefry.h"
78 #include "gromacs/simd/simd.h"
79 #include "gromacs/timing/wallcycle.h"
80 #include "gromacs/topology/atoms.h"
81 #include "gromacs/utility/exceptions.h"
82 #include "gromacs/utility/fatalerror.h"
83 #include "gromacs/utility/futil.h"
84 #include "gromacs/utility/gmxassert.h"
85 #include "gromacs/utility/gmxomp.h"
86 #include "gromacs/utility/smalloc.h"
88 using namespace gmx; // TODO: Remove when this file is moved into gmx namespace
103 std::vector<real> bd_rf;
105 std::vector<gmx_sd_const_t> sdc;
106 std::vector<gmx_sd_sigma_t> sdsig;
107 /* andersen temperature control stuff */
108 std::vector<bool> randomize_group;
109 std::vector<real> boltzfac;
111 explicit gmx_stochd_t(const t_inputrec* ir);
114 //! pImpled implementation for Update
119 Impl(const t_inputrec* ir, BoxDeformation* boxDeformation);
122 //! stochastic dynamics struct
123 std::unique_ptr<gmx_stochd_t> sd;
124 //! xprime for constraint algorithms
125 PaddedVector<RVec> xp;
126 //! Box deformation handler (or nullptr if inactive).
127 BoxDeformation* deform = nullptr;
130 Update::Update(const t_inputrec* ir, BoxDeformation* boxDeformation) :
131 impl_(new Impl(ir, boxDeformation)){};
135 gmx_stochd_t* Update::sd() const
137 return impl_->sd.get();
140 PaddedVector<RVec>* Update::xp()
145 BoxDeformation* Update::deform() const
147 return impl_->deform;
150 /*! \brief Sets the velocities of virtual sites to zero */
151 static void clearVsiteVelocities(int start, int nrend, const unsigned short* particleType, rvec* gmx_restrict v)
153 for (int a = start; a < nrend; a++)
155 if (particleType[a] == eptVSite)
162 /*! \brief Sets the number of different temperature coupling values */
163 enum class NumTempScaleValues
165 single, //!< Single T-scaling value (either one group or all values =1)
166 multiple //!< Multiple T-scaling values, need to use T-group indices
169 /*! \brief Sets if to apply no or diagonal Parrinello-Rahman pressure scaling
171 * Note that this enum is only used in updateMDLeapfrogSimple(), which does
172 * not handle fully anistropic Parrinello-Rahman scaling, so we only have
173 * options \p no and \p diagonal here and no anistropic option.
175 enum class ApplyParrinelloRahmanVScaling
177 no, //!< Do not apply velocity scaling (not a PR-coupling run or step)
178 diagonal //!< Apply velocity scaling using a diagonal matrix
181 /*! \brief Integrate using leap-frog with T-scaling and optionally diagonal Parrinello-Rahman p-coupling
183 * \tparam numTempScaleValues The number of different T-couple values
184 * \tparam applyPRVScaling Apply Parrinello-Rahman velocity scaling
185 * \param[in] start Index of first atom to update
186 * \param[in] nrend Last atom to update: \p nrend - 1
187 * \param[in] dt The time step
188 * \param[in] dtPressureCouple Time step for pressure coupling
189 * \param[in] invMassPerDim 1/mass per atom and dimension
190 * \param[in] tcstat Temperature coupling information
191 * \param[in] cTC T-coupling group index per atom
192 * \param[in] pRVScaleMatrixDiagonal Parrinello-Rahman v-scale matrix diagonal
193 * \param[in] x Input coordinates
194 * \param[out] xprime Updated coordinates
195 * \param[inout] v Velocities
196 * \param[in] f Forces
198 * We expect this template to get good SIMD acceleration by most compilers,
199 * unlike the more complex general template.
200 * Note that we might get even better SIMD acceleration when we introduce
201 * aligned (and padded) memory, possibly with some hints for the compilers.
203 template<NumTempScaleValues numTempScaleValues, ApplyParrinelloRahmanVScaling applyPRVScaling>
204 static void updateMDLeapfrogSimple(int start,
207 real dtPressureCouple,
208 const rvec* gmx_restrict invMassPerDim,
209 gmx::ArrayRef<const t_grp_tcstat> tcstat,
210 const unsigned short* cTC,
211 const rvec pRVScaleMatrixDiagonal,
212 const rvec* gmx_restrict x,
213 rvec* gmx_restrict xprime,
214 rvec* gmx_restrict v,
215 const rvec* gmx_restrict f)
219 if (numTempScaleValues == NumTempScaleValues::single)
221 lambdaGroup = tcstat[0].lambda;
224 for (int a = start; a < nrend; a++)
226 if (numTempScaleValues == NumTempScaleValues::multiple)
228 lambdaGroup = tcstat[cTC[a]].lambda;
231 for (int d = 0; d < DIM; d++)
233 /* Note that using rvec invMassPerDim results in more efficient
234 * SIMD code, but this increases the cache pressure.
235 * For large systems with PME on the CPU this slows down the
236 * (then already slow) update by 20%. If all data remains in cache,
237 * using rvec is much faster.
239 real vNew = lambdaGroup * v[a][d] + f[a][d] * invMassPerDim[a][d] * dt;
241 if (applyPRVScaling == ApplyParrinelloRahmanVScaling::diagonal)
243 vNew -= dtPressureCouple * pRVScaleMatrixDiagonal[d] * v[a][d];
246 xprime[a][d] = x[a][d] + vNew * dt;
251 #if GMX_SIMD && GMX_SIMD_HAVE_REAL
252 # define GMX_HAVE_SIMD_UPDATE 1
254 # define GMX_HAVE_SIMD_UPDATE 0
257 #if GMX_HAVE_SIMD_UPDATE
259 /*! \brief Load (aligned) the contents of GMX_SIMD_REAL_WIDTH rvec elements sequentially into 3 SIMD registers
261 * The loaded output is:
262 * \p r0: { r[index][XX], r[index][YY], ... }
264 * \p r2: { ..., r[index+GMX_SIMD_REAL_WIDTH-1][YY], r[index+GMX_SIMD_REAL_WIDTH-1][ZZ] }
266 * \param[in] r Real to an rvec array, has to be aligned to SIMD register width
267 * \param[in] index Index of the first rvec triplet of reals to load
268 * \param[out] r0 Pointer to first SIMD register
269 * \param[out] r1 Pointer to second SIMD register
270 * \param[out] r2 Pointer to third SIMD register
272 static inline void simdLoadRvecs(const rvec* r, int index, SimdReal* r0, SimdReal* r1, SimdReal* r2)
274 const real* realPtr = r[index];
276 GMX_ASSERT(isSimdAligned(realPtr), "Pointer should be SIMD aligned");
278 *r0 = simdLoad(realPtr + 0 * GMX_SIMD_REAL_WIDTH);
279 *r1 = simdLoad(realPtr + 1 * GMX_SIMD_REAL_WIDTH);
280 *r2 = simdLoad(realPtr + 2 * GMX_SIMD_REAL_WIDTH);
283 /*! \brief Store (aligned) 3 SIMD registers sequentially to GMX_SIMD_REAL_WIDTH rvec elements
285 * The stored output is:
286 * \p r[index] = { { r0[0], r0[1], ... }
288 * \p r[index+GMX_SIMD_REAL_WIDTH-1] = { ... , r2[GMX_SIMD_REAL_WIDTH-2], r2[GMX_SIMD_REAL_WIDTH-1] }
290 * \param[out] r Pointer to an rvec array, has to be aligned to SIMD register width
291 * \param[in] index Index of the first rvec triplet of reals to store to
292 * \param[in] r0 First SIMD register
293 * \param[in] r1 Second SIMD register
294 * \param[in] r2 Third SIMD register
296 static inline void simdStoreRvecs(rvec* r, int index, SimdReal r0, SimdReal r1, SimdReal r2)
298 real* realPtr = r[index];
300 GMX_ASSERT(isSimdAligned(realPtr), "Pointer should be SIMD aligned");
302 store(realPtr + 0 * GMX_SIMD_REAL_WIDTH, r0);
303 store(realPtr + 1 * GMX_SIMD_REAL_WIDTH, r1);
304 store(realPtr + 2 * GMX_SIMD_REAL_WIDTH, r2);
307 /*! \brief Integrate using leap-frog with single group T-scaling and SIMD
309 * \param[in] start Index of first atom to update
310 * \param[in] nrend Last atom to update: \p nrend - 1
311 * \param[in] dt The time step
312 * \param[in] invMass 1/mass per atom
313 * \param[in] tcstat Temperature coupling information
314 * \param[in] x Input coordinates
315 * \param[out] xprime Updated coordinates
316 * \param[inout] v Velocities
317 * \param[in] f Forces
319 static void updateMDLeapfrogSimpleSimd(int start,
322 const real* gmx_restrict invMass,
323 gmx::ArrayRef<const t_grp_tcstat> tcstat,
324 const rvec* gmx_restrict x,
325 rvec* gmx_restrict xprime,
326 rvec* gmx_restrict v,
327 const rvec* gmx_restrict f)
329 SimdReal timestep(dt);
330 SimdReal lambdaSystem(tcstat[0].lambda);
332 /* We declare variables here, since code is often slower when declaring them inside the loop */
334 /* Note: We should implement a proper PaddedVector, so we don't need this check */
335 GMX_ASSERT(isSimdAligned(invMass), "invMass should be aligned");
337 for (int a = start; a < nrend; a += GMX_SIMD_REAL_WIDTH)
339 SimdReal invMass0, invMass1, invMass2;
340 expandScalarsToTriplets(simdLoad(invMass + a), &invMass0, &invMass1, &invMass2);
344 simdLoadRvecs(v, a, &v0, &v1, &v2);
345 simdLoadRvecs(f, a, &f0, &f1, &f2);
347 v0 = fma(f0 * invMass0, timestep, lambdaSystem * v0);
348 v1 = fma(f1 * invMass1, timestep, lambdaSystem * v1);
349 v2 = fma(f2 * invMass2, timestep, lambdaSystem * v2);
351 simdStoreRvecs(v, a, v0, v1, v2);
354 simdLoadRvecs(x, a, &x0, &x1, &x2);
356 SimdReal xprime0 = fma(v0, timestep, x0);
357 SimdReal xprime1 = fma(v1, timestep, x1);
358 SimdReal xprime2 = fma(v2, timestep, x2);
360 simdStoreRvecs(xprime, a, xprime0, xprime1, xprime2);
364 #endif // GMX_HAVE_SIMD_UPDATE
366 /*! \brief Sets the NEMD acceleration type */
367 enum class AccelerationType
374 /*! \brief Integrate using leap-frog with support for everything
376 * \tparam accelerationType Type of NEMD acceleration
377 * \param[in] start Index of first atom to update
378 * \param[in] nrend Last atom to update: \p nrend - 1
379 * \param[in] doNoseHoover If to apply Nose-Hoover T-coupling
380 * \param[in] dt The time step
381 * \param[in] dtPressureCouple Time step for pressure coupling, is 0 when no pressure
382 * coupling should be applied at this step \param[in] ir The input parameter
383 * record \param[in] md Atom properties \param[in] ekind Kinetic energy
384 * data \param[in] box The box dimensions \param[in] x Input coordinates \param[out]
385 * xprime Updated coordinates \param[inout] v Velocities \param[in]
386 * f Forces \param[in] nh_vxi Nose-Hoover velocity scaling
387 * factors \param[in] M Parrinello-Rahman scaling matrix
389 template<AccelerationType accelerationType>
390 static void updateMDLeapfrogGeneral(int start,
394 real dtPressureCouple,
395 const t_inputrec* ir,
397 const gmx_ekindata_t* ekind,
399 const rvec* gmx_restrict x,
400 rvec* gmx_restrict xprime,
401 rvec* gmx_restrict v,
402 const rvec* gmx_restrict f,
403 const double* gmx_restrict nh_vxi,
406 /* This is a version of the leap-frog integrator that supports
407 * all combinations of T-coupling, P-coupling and NEMD.
408 * Nose-Hoover uses the reversible leap-frog integrator from
409 * Holian et al. Phys Rev E 52(3) : 2338, 1995
412 gmx::ArrayRef<const t_grp_tcstat> tcstat = ekind->tcstat;
413 gmx::ArrayRef<const t_grp_acc> grpstat = ekind->grpstat;
414 const unsigned short* cTC = md->cTC;
415 const unsigned short* cACC = md->cACC;
416 const rvec* accel = ir->opts.acc;
418 const rvec* gmx_restrict invMassPerDim = md->invMassPerDim;
420 /* Initialize group values, changed later when multiple groups are used */
425 real omega_Z = 2 * static_cast<real>(M_PI) / box[ZZ][ZZ];
427 for (int n = start; n < nrend; n++)
433 real lg = tcstat[gt].lambda;
436 real cosineZ, vCosine;
437 #ifdef __INTEL_COMPILER
438 # pragma warning(disable : 280)
440 switch (accelerationType)
442 case AccelerationType::none: copy_rvec(v[n], vRel); break;
443 case AccelerationType::group:
448 /* Avoid scaling the group velocity */
449 rvec_sub(v[n], grpstat[ga].u, vRel);
451 case AccelerationType::cosine:
452 cosineZ = std::cos(x[n][ZZ] * omega_Z);
453 vCosine = cosineZ * ekind->cosacc.vcos;
454 /* Avoid scaling the cosine profile velocity */
455 copy_rvec(v[n], vRel);
462 /* Here we account for multiple time stepping, by increasing
463 * the Nose-Hoover correction by nsttcouple
465 factorNH = 0.5 * ir->nsttcouple * dt * nh_vxi[gt];
468 for (int d = 0; d < DIM; d++)
470 real vNew = (lg * vRel[d]
471 + (f[n][d] * invMassPerDim[n][d] * dt - factorNH * vRel[d]
472 - dtPressureCouple * iprod(M[d], vRel)))
474 switch (accelerationType)
476 case AccelerationType::none: break;
477 case AccelerationType::group:
478 /* Add back the mean velocity and apply acceleration */
479 vNew += grpstat[ga].u[d] + accel[ga][d] * dt;
481 case AccelerationType::cosine:
484 /* Add back the mean velocity and apply acceleration */
485 vNew += vCosine + cosineZ * ekind->cosacc.cos_accel * dt;
490 xprime[n][d] = x[n][d] + vNew * dt;
495 /*! \brief Handles the Leap-frog MD x and v integration */
496 static void do_update_md(int start,
500 const t_inputrec* ir,
502 const gmx_ekindata_t* ekind,
504 const rvec* gmx_restrict x,
505 rvec* gmx_restrict xprime,
506 rvec* gmx_restrict v,
507 const rvec* gmx_restrict f,
508 const double* gmx_restrict nh_vxi,
511 GMX_ASSERT(nrend == start || xprime != x,
512 "For SIMD optimization certain compilers need to have xprime != x");
513 GMX_ASSERT(ir->eI == eiMD,
514 "Leap-frog integrator was called while another integrator was requested");
516 /* Note: Berendsen pressure scaling is handled after do_update_md() */
517 bool doTempCouple = (ir->etc != etcNO && do_per_step(step + ir->nsttcouple - 1, ir->nsttcouple));
518 bool doNoseHoover = (ir->etc == etcNOSEHOOVER && doTempCouple);
519 bool doParrinelloRahman =
520 (ir->epc == epcPARRINELLORAHMAN && do_per_step(step + ir->nstpcouple - 1, ir->nstpcouple));
521 bool doPROffDiagonal = (doParrinelloRahman && (M[YY][XX] != 0 || M[ZZ][XX] != 0 || M[ZZ][YY] != 0));
523 real dtPressureCouple = (doParrinelloRahman ? ir->nstpcouple * dt : 0);
525 /* NEMD (also cosine) acceleration is applied in updateMDLeapFrogGeneral */
526 bool doAcceleration = (ekind->bNEMD || ekind->cosacc.cos_accel != 0);
528 if (doNoseHoover || doPROffDiagonal || doAcceleration)
531 if (!doParrinelloRahman)
533 /* We should not apply PR scaling at this step */
543 updateMDLeapfrogGeneral<AccelerationType::none>(start, nrend, doNoseHoover, dt,
544 dtPressureCouple, ir, md, ekind, box, x,
545 xprime, v, f, nh_vxi, stepM);
547 else if (ekind->bNEMD)
549 updateMDLeapfrogGeneral<AccelerationType::group>(start, nrend, doNoseHoover, dt,
550 dtPressureCouple, ir, md, ekind, box,
551 x, xprime, v, f, nh_vxi, stepM);
555 updateMDLeapfrogGeneral<AccelerationType::cosine>(start, nrend, doNoseHoover, dt,
556 dtPressureCouple, ir, md, ekind, box,
557 x, xprime, v, f, nh_vxi, stepM);
562 /* Use a simple and thus more efficient integration loop. */
563 /* The simple loop does not check for particle type (so it can
564 * be vectorized), which means we need to clear the velocities
565 * of virtual sites in advance, when present. Note that vsite
566 * velocities are computed after update and constraints from
567 * their displacement.
571 /* Note: The overhead of this loop is completely neligible */
572 clearVsiteVelocities(start, nrend, md->ptype, v);
575 /* We determine if we have a single T-coupling lambda value for all
576 * atoms. That allows for better SIMD acceleration in the template.
577 * If we do not do temperature coupling (in the run or this step),
578 * all scaling values are 1, so we effectively have a single value.
580 bool haveSingleTempScaleValue = (!doTempCouple || ekind->ngtc == 1);
582 /* Extract some pointers needed by all cases */
583 const unsigned short* cTC = md->cTC;
584 gmx::ArrayRef<const t_grp_tcstat> tcstat = ekind->tcstat;
585 const rvec* invMassPerDim = md->invMassPerDim;
587 if (doParrinelloRahman)
589 GMX_ASSERT(!doPROffDiagonal,
590 "updateMDLeapfrogSimple only support diagonal Parrinello-Rahman scaling "
594 for (int d = 0; d < DIM; d++)
599 if (haveSingleTempScaleValue)
601 updateMDLeapfrogSimple<NumTempScaleValues::single, ApplyParrinelloRahmanVScaling::diagonal>(
602 start, nrend, dt, dtPressureCouple, invMassPerDim, tcstat, cTC, diagM, x,
607 updateMDLeapfrogSimple<NumTempScaleValues::multiple, ApplyParrinelloRahmanVScaling::diagonal>(
608 start, nrend, dt, dtPressureCouple, invMassPerDim, tcstat, cTC, diagM, x,
614 if (haveSingleTempScaleValue)
616 /* Note that modern compilers are pretty good at vectorizing
617 * updateMDLeapfrogSimple(). But the SIMD version will still
618 * be faster because invMass lowers the cache pressure
619 * compared to invMassPerDim.
621 #if GMX_HAVE_SIMD_UPDATE
622 /* Check if we can use invmass instead of invMassPerDim */
623 if (!md->havePartiallyFrozenAtoms)
625 updateMDLeapfrogSimpleSimd(start, nrend, dt, md->invmass, tcstat, x, xprime, v, f);
630 updateMDLeapfrogSimple<NumTempScaleValues::single, ApplyParrinelloRahmanVScaling::no>(
631 start, nrend, dt, dtPressureCouple, invMassPerDim, tcstat, cTC, nullptr,
637 updateMDLeapfrogSimple<NumTempScaleValues::multiple, ApplyParrinelloRahmanVScaling::no>(
638 start, nrend, dt, dtPressureCouple, invMassPerDim, tcstat, cTC, nullptr, x,
645 static void do_update_vv_vel(int start,
649 const ivec nFreeze[],
650 const real invmass[],
651 const unsigned short ptype[],
652 const unsigned short cFREEZE[],
653 const unsigned short cACC[],
666 g = 0.25 * dt * veta * alpha;
668 mv2 = gmx::series_sinhx(g);
675 for (n = start; n < nrend; n++)
677 real w_dt = invmass[n] * dt;
687 for (d = 0; d < DIM; d++)
689 if ((ptype[n] != eptVSite) && (ptype[n] != eptShell) && !nFreeze[gf][d])
691 v[n][d] = mv1 * (mv1 * v[n][d] + 0.5 * (w_dt * mv2 * f[n][d])) + 0.5 * accel[ga][d] * dt;
699 } /* do_update_vv_vel */
701 static void do_update_vv_pos(int start,
704 const ivec nFreeze[],
705 const unsigned short ptype[],
706 const unsigned short cFREEZE[],
717 /* Would it make more sense if Parrinello-Rahman was put here? */
722 mr2 = gmx::series_sinhx(g);
730 for (n = start; n < nrend; n++)
738 for (d = 0; d < DIM; d++)
740 if ((ptype[n] != eptVSite) && (ptype[n] != eptShell) && !nFreeze[gf][d])
742 xprime[n][d] = mr1 * (mr1 * x[n][d] + mr2 * dt * v[n][d]);
746 xprime[n][d] = x[n][d];
750 } /* do_update_vv_pos */
752 gmx_stochd_t::gmx_stochd_t(const t_inputrec* ir)
754 const t_grpopts* opts = &ir->opts;
755 const int ngtc = opts->ngtc;
761 else if (EI_SD(ir->eI))
766 for (int gt = 0; gt < ngtc; gt++)
768 if (opts->tau_t[gt] > 0)
770 sdc[gt].em = std::exp(-ir->delta_t / opts->tau_t[gt]);
774 /* No friction and noise on this group */
779 else if (ETC_ANDERSEN(ir->etc))
781 randomize_group.resize(ngtc);
782 boltzfac.resize(ngtc);
784 /* for now, assume that all groups, if randomized, are randomized at the same rate, i.e. tau_t is the same. */
785 /* since constraint groups don't necessarily match up with temperature groups! This is checked in readir.c */
787 for (int gt = 0; gt < ngtc; gt++)
789 real reft = std::max<real>(0, opts->ref_t[gt]);
790 if ((opts->tau_t[gt] > 0)
791 && (reft > 0)) /* tau_t or ref_t = 0 means that no randomization is done */
793 randomize_group[gt] = true;
794 boltzfac[gt] = BOLTZ * opts->ref_t[gt];
798 randomize_group[gt] = false;
804 void update_temperature_constants(gmx_stochd_t* sd, const t_inputrec* ir)
808 if (ir->bd_fric != 0)
810 for (int gt = 0; gt < ir->opts.ngtc; gt++)
812 sd->bd_rf[gt] = std::sqrt(2.0 * BOLTZ * ir->opts.ref_t[gt] / (ir->bd_fric * ir->delta_t));
817 for (int gt = 0; gt < ir->opts.ngtc; gt++)
819 sd->bd_rf[gt] = std::sqrt(2.0 * BOLTZ * ir->opts.ref_t[gt]);
825 for (int gt = 0; gt < ir->opts.ngtc; gt++)
827 real kT = BOLTZ * ir->opts.ref_t[gt];
828 /* The mass is accounted for later, since this differs per atom */
829 sd->sdsig[gt].V = std::sqrt(kT * (1 - sd->sdc[gt].em * sd->sdc[gt].em));
834 Update::Impl::Impl(const t_inputrec* ir, BoxDeformation* boxDeformation)
836 sd = std::make_unique<gmx_stochd_t>(ir);
837 update_temperature_constants(sd.get(), ir);
838 xp.resizeWithPadding(0);
839 deform = boxDeformation;
842 void Update::setNumAtoms(int nAtoms)
845 impl_->xp.resizeWithPadding(nAtoms);
848 /*! \brief Sets the SD update type */
849 enum class SDUpdate : int
852 FrictionAndNoiseOnly,
856 /*! \brief SD integrator update
858 * Two phases are required in the general case of a constrained
859 * update, the first phase from the contribution of forces, before
860 * applying constraints, and then a second phase applying the friction
861 * and noise, and then further constraining. For details, see
864 * Without constraints, the two phases can be combined, for
867 * Thus three instantiations of this templated function will be made,
868 * two with only one contribution, and one with both contributions. */
869 template<SDUpdate updateType>
870 static void doSDUpdateGeneral(const gmx_stochd_t& sd,
875 const ivec nFreeze[],
876 const real invmass[],
877 const unsigned short ptype[],
878 const unsigned short cFREEZE[],
879 const unsigned short cACC[],
880 const unsigned short cTC[],
889 // cTC, cACC and cFREEZE can be nullptr any time, but various
890 // instantiations do not make sense with particular pointer
892 if (updateType == SDUpdate::ForcesOnly)
894 GMX_ASSERT(f != nullptr, "SD update with only forces requires forces");
895 GMX_ASSERT(cTC == nullptr, "SD update with only forces cannot handle temperature groups");
897 if (updateType == SDUpdate::FrictionAndNoiseOnly)
899 GMX_ASSERT(f == nullptr, "SD update with only noise cannot handle forces");
900 GMX_ASSERT(cACC == nullptr, "SD update with only noise cannot handle acceleration groups");
902 if (updateType == SDUpdate::Combined)
904 GMX_ASSERT(f != nullptr, "SD update with forces and noise requires forces");
907 // Even 0 bits internal counter gives 2x64 ints (more than enough for three table lookups)
908 gmx::ThreeFry2x64<0> rng(seed, gmx::RandomDomain::UpdateCoordinates);
909 gmx::TabulatedNormalDistribution<real, 14> dist;
911 for (int n = start; n < nrend; n++)
913 int globalAtomIndex = gatindex ? gatindex[n] : n;
914 rng.restart(step, globalAtomIndex);
917 real inverseMass = invmass[n];
918 real invsqrtMass = std::sqrt(inverseMass);
920 int freezeGroup = cFREEZE ? cFREEZE[n] : 0;
921 int accelerationGroup = cACC ? cACC[n] : 0;
922 int temperatureGroup = cTC ? cTC[n] : 0;
924 for (int d = 0; d < DIM; d++)
926 if ((ptype[n] != eptVSite) && (ptype[n] != eptShell) && !nFreeze[freezeGroup][d])
928 if (updateType == SDUpdate::ForcesOnly)
930 real vn = v[n][d] + (inverseMass * f[n][d] + accel[accelerationGroup][d]) * dt;
932 // Simple position update.
933 xprime[n][d] = x[n][d] + v[n][d] * dt;
935 else if (updateType == SDUpdate::FrictionAndNoiseOnly)
938 v[n][d] = (vn * sd.sdc[temperatureGroup].em
939 + invsqrtMass * sd.sdsig[temperatureGroup].V * dist(rng));
940 // The previous phase already updated the
941 // positions with a full v*dt term that must
942 // now be half removed.
943 xprime[n][d] = xprime[n][d] + 0.5 * (v[n][d] - vn) * dt;
947 real vn = v[n][d] + (inverseMass * f[n][d] + accel[accelerationGroup][d]) * dt;
948 v[n][d] = (vn * sd.sdc[temperatureGroup].em
949 + invsqrtMass * sd.sdsig[temperatureGroup].V * dist(rng));
950 // Here we include half of the friction+noise
951 // update of v into the position update.
952 xprime[n][d] = x[n][d] + 0.5 * (vn + v[n][d]) * dt;
957 // When using constraints, the update is split into
958 // two phases, but we only need to zero the update of
959 // virtual, shell or frozen particles in at most one
961 if (updateType != SDUpdate::FrictionAndNoiseOnly)
964 xprime[n][d] = x[n][d];
971 static void do_update_bd(int start,
974 const ivec nFreeze[],
975 const real invmass[],
976 const unsigned short ptype[],
977 const unsigned short cFREEZE[],
978 const unsigned short cTC[],
983 real friction_coefficient,
989 /* note -- these appear to be full step velocities . . . */
994 // Use 1 bit of internal counters to give us 2*2 64-bits values per stream
995 // Each 64-bit value is enough for 4 normal distribution table numbers.
996 gmx::ThreeFry2x64<0> rng(seed, gmx::RandomDomain::UpdateCoordinates);
997 gmx::TabulatedNormalDistribution<real, 14> dist;
999 if (friction_coefficient != 0)
1001 invfr = 1.0 / friction_coefficient;
1004 for (n = start; (n < nrend); n++)
1006 int ng = gatindex ? gatindex[n] : n;
1008 rng.restart(step, ng);
1019 for (d = 0; (d < DIM); d++)
1021 if ((ptype[n] != eptVSite) && (ptype[n] != eptShell) && !nFreeze[gf][d])
1023 if (friction_coefficient != 0)
1025 vn = invfr * f[n][d] + rf[gt] * dist(rng);
1029 /* NOTE: invmass = 2/(mass*friction_constant*dt) */
1030 vn = 0.5 * invmass[n] * f[n][d] * dt
1031 + std::sqrt(0.5 * invmass[n]) * rf[gt] * dist(rng);
1035 xprime[n][d] = x[n][d] + vn * dt;
1040 xprime[n][d] = x[n][d];
1046 static void calc_ke_part_normal(ArrayRef<const RVec> v,
1047 const t_grpopts* opts,
1048 const t_mdatoms* md,
1049 gmx_ekindata_t* ekind,
1051 gmx_bool bEkinAveVel)
1054 gmx::ArrayRef<t_grp_tcstat> tcstat = ekind->tcstat;
1055 gmx::ArrayRef<t_grp_acc> grpstat = ekind->grpstat;
1057 /* three main: VV with AveVel, vv with AveEkin, leap with AveEkin. Leap with AveVel is also
1058 an option, but not supported now.
1059 bEkinAveVel: If TRUE, we sum into ekin, if FALSE, into ekinh.
1062 /* group velocities are calculated in update_ekindata and
1063 * accumulated in acumulate_groups.
1064 * Now the partial global and groups ekin.
1066 for (g = 0; (g < opts->ngtc); g++)
1068 copy_mat(tcstat[g].ekinh, tcstat[g].ekinh_old);
1071 clear_mat(tcstat[g].ekinf);
1072 tcstat[g].ekinscalef_nhc = 1.0; /* need to clear this -- logic is complicated! */
1076 clear_mat(tcstat[g].ekinh);
1079 ekind->dekindl_old = ekind->dekindl;
1080 int nthread = gmx_omp_nthreads_get(emntUpdate);
1082 #pragma omp parallel for num_threads(nthread) schedule(static)
1083 for (int thread = 0; thread < nthread; thread++)
1085 // This OpenMP only loops over arrays and does not call any functions
1086 // or memory allocation. It should not be able to throw, so for now
1087 // we do not need a try/catch wrapper.
1088 int start_t, end_t, n;
1096 start_t = ((thread + 0) * md->homenr) / nthread;
1097 end_t = ((thread + 1) * md->homenr) / nthread;
1099 ekin_sum = ekind->ekin_work[thread];
1100 dekindl_sum = ekind->dekindl_work[thread];
1102 for (gt = 0; gt < opts->ngtc; gt++)
1104 clear_mat(ekin_sum[gt]);
1110 for (n = start_t; n < end_t; n++)
1120 hm = 0.5 * md->massT[n];
1122 for (d = 0; (d < DIM); d++)
1124 v_corrt[d] = v[n][d] - grpstat[ga].u[d];
1126 for (d = 0; (d < DIM); d++)
1128 for (m = 0; (m < DIM); m++)
1130 /* if we're computing a full step velocity, v_corrt[d] has v(t). Otherwise, v(t+dt/2) */
1131 ekin_sum[gt][m][d] += hm * v_corrt[m] * v_corrt[d];
1134 if (md->nMassPerturbed && md->bPerturbed[n])
1136 *dekindl_sum += 0.5 * (md->massB[n] - md->massA[n]) * iprod(v_corrt, v_corrt);
1142 for (int thread = 0; thread < nthread; thread++)
1144 for (g = 0; g < opts->ngtc; g++)
1148 m_add(tcstat[g].ekinf, ekind->ekin_work[thread][g], tcstat[g].ekinf);
1152 m_add(tcstat[g].ekinh, ekind->ekin_work[thread][g], tcstat[g].ekinh);
1156 ekind->dekindl += *ekind->dekindl_work[thread];
1159 inc_nrnb(nrnb, eNR_EKIN, md->homenr);
1162 static void calc_ke_part_visc(const matrix box,
1163 ArrayRef<const RVec> x,
1164 ArrayRef<const RVec> v,
1165 const t_grpopts* opts,
1166 const t_mdatoms* md,
1167 gmx_ekindata_t* ekind,
1169 gmx_bool bEkinAveVel)
1171 int start = 0, homenr = md->homenr;
1172 int g, d, n, m, gt = 0;
1175 gmx::ArrayRef<t_grp_tcstat> tcstat = ekind->tcstat;
1176 t_cos_acc* cosacc = &(ekind->cosacc);
1181 for (g = 0; g < opts->ngtc; g++)
1183 copy_mat(ekind->tcstat[g].ekinh, ekind->tcstat[g].ekinh_old);
1184 clear_mat(ekind->tcstat[g].ekinh);
1186 ekind->dekindl_old = ekind->dekindl;
1188 fac = 2 * M_PI / box[ZZ][ZZ];
1191 for (n = start; n < start + homenr; n++)
1197 hm = 0.5 * md->massT[n];
1199 /* Note that the times of x and v differ by half a step */
1200 /* MRS -- would have to be changed for VV */
1201 cosz = std::cos(fac * x[n][ZZ]);
1202 /* Calculate the amplitude of the new velocity profile */
1203 mvcos += 2 * cosz * md->massT[n] * v[n][XX];
1205 copy_rvec(v[n], v_corrt);
1206 /* Subtract the profile for the kinetic energy */
1207 v_corrt[XX] -= cosz * cosacc->vcos;
1208 for (d = 0; (d < DIM); d++)
1210 for (m = 0; (m < DIM); m++)
1212 /* if we're computing a full step velocity, v_corrt[d] has v(t). Otherwise, v(t+dt/2) */
1215 tcstat[gt].ekinf[m][d] += hm * v_corrt[m] * v_corrt[d];
1219 tcstat[gt].ekinh[m][d] += hm * v_corrt[m] * v_corrt[d];
1223 if (md->nPerturbed && md->bPerturbed[n])
1225 /* The minus sign here might be confusing.
1226 * The kinetic contribution from dH/dl doesn't come from
1227 * d m(l)/2 v^2 / dl, but rather from d p^2/2m(l) / dl,
1228 * where p are the momenta. The difference is only a minus sign.
1230 dekindl -= 0.5 * (md->massB[n] - md->massA[n]) * iprod(v_corrt, v_corrt);
1233 ekind->dekindl = dekindl;
1234 cosacc->mvcos = mvcos;
1236 inc_nrnb(nrnb, eNR_EKIN, homenr);
1239 void calc_ke_part(ArrayRef<const RVec> x,
1240 ArrayRef<const RVec> v,
1242 const t_grpopts* opts,
1243 const t_mdatoms* md,
1244 gmx_ekindata_t* ekind,
1246 gmx_bool bEkinAveVel)
1248 if (ekind->cosacc.cos_accel == 0)
1250 calc_ke_part_normal(v, opts, md, ekind, nrnb, bEkinAveVel);
1254 calc_ke_part_visc(box, x, v, opts, md, ekind, nrnb, bEkinAveVel);
1258 extern void init_ekinstate(ekinstate_t* ekinstate, const t_inputrec* ir)
1260 ekinstate->ekin_n = ir->opts.ngtc;
1261 snew(ekinstate->ekinh, ekinstate->ekin_n);
1262 snew(ekinstate->ekinf, ekinstate->ekin_n);
1263 snew(ekinstate->ekinh_old, ekinstate->ekin_n);
1264 ekinstate->ekinscalef_nhc.resize(ekinstate->ekin_n);
1265 ekinstate->ekinscaleh_nhc.resize(ekinstate->ekin_n);
1266 ekinstate->vscale_nhc.resize(ekinstate->ekin_n);
1267 ekinstate->dekindl = 0;
1268 ekinstate->mvcos = 0;
1269 ekinstate->hasReadEkinState = false;
1272 void update_ekinstate(ekinstate_t* ekinstate, const gmx_ekindata_t* ekind)
1276 for (i = 0; i < ekinstate->ekin_n; i++)
1278 copy_mat(ekind->tcstat[i].ekinh, ekinstate->ekinh[i]);
1279 copy_mat(ekind->tcstat[i].ekinf, ekinstate->ekinf[i]);
1280 copy_mat(ekind->tcstat[i].ekinh_old, ekinstate->ekinh_old[i]);
1281 ekinstate->ekinscalef_nhc[i] = ekind->tcstat[i].ekinscalef_nhc;
1282 ekinstate->ekinscaleh_nhc[i] = ekind->tcstat[i].ekinscaleh_nhc;
1283 ekinstate->vscale_nhc[i] = ekind->tcstat[i].vscale_nhc;
1286 copy_mat(ekind->ekin, ekinstate->ekin_total);
1287 ekinstate->dekindl = ekind->dekindl;
1288 ekinstate->mvcos = ekind->cosacc.mvcos;
1291 void restore_ekinstate_from_state(const t_commrec* cr, gmx_ekindata_t* ekind, const ekinstate_t* ekinstate)
1297 for (i = 0; i < ekinstate->ekin_n; i++)
1299 copy_mat(ekinstate->ekinh[i], ekind->tcstat[i].ekinh);
1300 copy_mat(ekinstate->ekinf[i], ekind->tcstat[i].ekinf);
1301 copy_mat(ekinstate->ekinh_old[i], ekind->tcstat[i].ekinh_old);
1302 ekind->tcstat[i].ekinscalef_nhc = ekinstate->ekinscalef_nhc[i];
1303 ekind->tcstat[i].ekinscaleh_nhc = ekinstate->ekinscaleh_nhc[i];
1304 ekind->tcstat[i].vscale_nhc = ekinstate->vscale_nhc[i];
1307 copy_mat(ekinstate->ekin_total, ekind->ekin);
1309 ekind->dekindl = ekinstate->dekindl;
1310 ekind->cosacc.mvcos = ekinstate->mvcos;
1311 n = ekinstate->ekin_n;
1316 gmx_bcast(sizeof(n), &n, cr);
1317 for (i = 0; i < n; i++)
1319 gmx_bcast(DIM * DIM * sizeof(ekind->tcstat[i].ekinh[0][0]), ekind->tcstat[i].ekinh[0], cr);
1320 gmx_bcast(DIM * DIM * sizeof(ekind->tcstat[i].ekinf[0][0]), ekind->tcstat[i].ekinf[0], cr);
1321 gmx_bcast(DIM * DIM * sizeof(ekind->tcstat[i].ekinh_old[0][0]),
1322 ekind->tcstat[i].ekinh_old[0], cr);
1324 gmx_bcast(sizeof(ekind->tcstat[i].ekinscalef_nhc), &(ekind->tcstat[i].ekinscalef_nhc), cr);
1325 gmx_bcast(sizeof(ekind->tcstat[i].ekinscaleh_nhc), &(ekind->tcstat[i].ekinscaleh_nhc), cr);
1326 gmx_bcast(sizeof(ekind->tcstat[i].vscale_nhc), &(ekind->tcstat[i].vscale_nhc), cr);
1328 gmx_bcast(DIM * DIM * sizeof(ekind->ekin[0][0]), ekind->ekin[0], cr);
1330 gmx_bcast(sizeof(ekind->dekindl), &ekind->dekindl, cr);
1331 gmx_bcast(sizeof(ekind->cosacc.mvcos), &ekind->cosacc.mvcos, cr);
1335 void update_tcouple(int64_t step,
1336 const t_inputrec* inputrec,
1338 gmx_ekindata_t* ekind,
1339 const t_extmass* MassQ,
1340 const t_mdatoms* md)
1343 // This condition was explicitly checked in previous version, but should have never been satisfied
1344 GMX_ASSERT(!(EI_VV(inputrec->eI)
1345 && (inputrecNvtTrotter(inputrec) || inputrecNptTrotter(inputrec)
1346 || inputrecNphTrotter(inputrec))),
1347 "Temperature coupling was requested with velocity verlet and trotter");
1349 bool doTemperatureCoupling = false;
1351 // For VV temperature coupling parameters are updated on the current
1352 // step, for the others - one step before.
1353 if (inputrec->etc == etcNO)
1355 doTemperatureCoupling = false;
1357 else if (EI_VV(inputrec->eI))
1359 doTemperatureCoupling = do_per_step(step, inputrec->nsttcouple);
1363 doTemperatureCoupling = do_per_step(step + inputrec->nsttcouple - 1, inputrec->nsttcouple);
1366 if (doTemperatureCoupling)
1368 real dttc = inputrec->nsttcouple * inputrec->delta_t;
1370 // TODO: berendsen_tcoupl(...), nosehoover_tcoupl(...) and vrescale_tcoupl(...) update
1371 // temperature coupling parameters, which should be reflected in the name of these
1373 switch (inputrec->etc)
1377 berendsen_tcoupl(inputrec, ekind, dttc, state->therm_integral);
1380 nosehoover_tcoupl(&(inputrec->opts), ekind, dttc, state->nosehoover_xi.data(),
1381 state->nosehoover_vxi.data(), MassQ);
1384 vrescale_tcoupl(inputrec, step, ekind, dttc, state->therm_integral.data());
1387 /* rescale in place here */
1388 if (EI_VV(inputrec->eI))
1390 rescale_velocities(ekind, md, 0, md->homenr, state->v.rvec_array());
1395 // Set the T scaling lambda to 1 to have no scaling
1396 // TODO: Do we have to do it on every non-t-couple step?
1397 for (int i = 0; (i < inputrec->opts.ngtc); i++)
1399 ekind->tcstat[i].lambda = 1.0;
1404 void getThreadAtomRange(int numThreads, int threadIndex, int numAtoms, int* startAtom, int* endAtom)
1406 #if GMX_HAVE_SIMD_UPDATE
1407 constexpr int blockSize = GMX_SIMD_REAL_WIDTH;
1409 constexpr int blockSize = 1;
1411 int numBlocks = (numAtoms + blockSize - 1) / blockSize;
1413 *startAtom = ((numBlocks * threadIndex) / numThreads) * blockSize;
1414 *endAtom = ((numBlocks * (threadIndex + 1)) / numThreads) * blockSize;
1415 if (threadIndex == numThreads - 1)
1417 *endAtom = numAtoms;
1421 void update_pcouple_before_coordinates(FILE* fplog,
1423 const t_inputrec* inputrec,
1425 matrix parrinellorahmanMu,
1429 /* Berendsen P-coupling is completely handled after the coordinate update.
1430 * Trotter P-coupling is handled by separate calls to trotter_update().
1432 if (inputrec->epc == epcPARRINELLORAHMAN
1433 && do_per_step(step + inputrec->nstpcouple - 1, inputrec->nstpcouple))
1435 real dtpc = inputrec->nstpcouple * inputrec->delta_t;
1437 parrinellorahman_pcoupl(fplog, step, inputrec, dtpc, state->pres_prev, state->box,
1438 state->box_rel, state->boxv, M, parrinellorahmanMu, bInitStep);
1442 void constrain_velocities(int64_t step,
1443 real* dvdlambda, /* the contribution to be added to the bonded interactions */
1446 gmx::Constraints* constr,
1458 * APPLY CONSTRAINTS:
1465 /* clear out constraints before applying */
1466 clear_mat(vir_part);
1468 /* Constrain the coordinates upd->xp */
1469 constr->apply(do_log, do_ene, step, 1, 1.0, state->x.arrayRefWithPadding(),
1470 state->v.arrayRefWithPadding(), state->v.arrayRefWithPadding().unpaddedArrayRef(),
1471 state->box, state->lambda[efptBONDED], dvdlambda, ArrayRefWithPadding<RVec>(),
1472 bCalcVir ? &vir_con : nullptr, ConstraintVariable::Velocities);
1476 m_add(vir_part, vir_con, vir_part);
1481 void constrain_coordinates(int64_t step,
1482 real* dvdlambda, /* the contribution to be added to the bonded interactions */
1486 gmx::Constraints* constr,
1499 /* clear out constraints before applying */
1500 clear_mat(vir_part);
1502 /* Constrain the coordinates upd->xp */
1503 constr->apply(do_log, do_ene, step, 1, 1.0, state->x.arrayRefWithPadding(),
1504 upd->xp()->arrayRefWithPadding(), ArrayRef<RVec>(), state->box,
1505 state->lambda[efptBONDED], dvdlambda, state->v.arrayRefWithPadding(),
1506 bCalcVir ? &vir_con : nullptr, ConstraintVariable::Positions);
1510 m_add(vir_part, vir_con, vir_part);
1515 void update_sd_second_half(int64_t step,
1516 real* dvdlambda, /* the contribution to be added to the bonded interactions */
1517 const t_inputrec* inputrec, /* input record and box stuff */
1518 const t_mdatoms* md,
1520 const t_commrec* cr,
1522 gmx_wallcycle_t wcycle,
1524 gmx::Constraints* constr,
1532 if (inputrec->eI == eiSD1)
1534 int homenr = md->homenr;
1536 /* Cast delta_t from double to real to make the integrators faster.
1537 * The only reason for having delta_t double is to get accurate values
1538 * for t=delta_t*step when step is larger than float precision.
1539 * For integration dt the accuracy of real suffices, since with
1540 * integral += dt*integrand the increment is nearly always (much) smaller
1541 * than the integral (and the integrand has real precision).
1543 real dt = inputrec->delta_t;
1545 wallcycle_start(wcycle, ewcUPDATE);
1547 int nth = gmx_omp_nthreads_get(emntUpdate);
1549 #pragma omp parallel for num_threads(nth) schedule(static)
1550 for (int th = 0; th < nth; th++)
1554 int start_th, end_th;
1555 getThreadAtomRange(nth, th, homenr, &start_th, &end_th);
1557 doSDUpdateGeneral<SDUpdate::FrictionAndNoiseOnly>(
1558 *upd->sd(), start_th, end_th, dt, inputrec->opts.acc, inputrec->opts.nFreeze,
1559 md->invmass, md->ptype, md->cFREEZE, nullptr, md->cTC, state->x.rvec_array(),
1560 upd->xp()->rvec_array(), state->v.rvec_array(), nullptr, step, inputrec->ld_seed,
1561 DOMAINDECOMP(cr) ? cr->dd->globalAtomIndices.data() : nullptr);
1563 GMX_CATCH_ALL_AND_EXIT_WITH_FATAL_ERROR
1565 inc_nrnb(nrnb, eNR_UPDATE, homenr);
1566 wallcycle_stop(wcycle, ewcUPDATE);
1568 /* Constrain the coordinates upd->xp for half a time step */
1569 constr->apply(do_log, do_ene, step, 1, 0.5, state->x.arrayRefWithPadding(),
1570 upd->xp()->arrayRefWithPadding(), ArrayRef<RVec>(), state->box,
1571 state->lambda[efptBONDED], dvdlambda, state->v.arrayRefWithPadding(), nullptr,
1572 ConstraintVariable::Positions);
1576 void finish_update(const t_inputrec* inputrec, /* input record and box stuff */
1577 const t_mdatoms* md,
1579 const t_graph* graph,
1581 gmx_wallcycle_t wcycle,
1583 const gmx::Constraints* constr)
1585 int homenr = md->homenr;
1587 /* We must always unshift after updating coordinates; if we did not shake
1588 x was shifted in do_force */
1590 /* NOTE Currently we always integrate to a temporary buffer and
1591 * then copy the results back. */
1593 wallcycle_start_nocount(wcycle, ewcUPDATE);
1595 if (md->cFREEZE != nullptr && constr != nullptr)
1597 /* If we have atoms that are frozen along some, but not all
1598 * dimensions, then any constraints will have moved them also along
1599 * the frozen dimensions. To freeze such degrees of freedom
1600 * we copy them back here to later copy them forward. It would
1601 * be more elegant and slightly more efficient to copies zero
1602 * times instead of twice, but the graph case below prevents this.
1604 const ivec* nFreeze = inputrec->opts.nFreeze;
1605 bool partialFreezeAndConstraints = false;
1606 for (int g = 0; g < inputrec->opts.ngfrz; g++)
1608 int numFreezeDim = nFreeze[g][XX] + nFreeze[g][YY] + nFreeze[g][ZZ];
1609 if (numFreezeDim > 0 && numFreezeDim < 3)
1611 partialFreezeAndConstraints = true;
1614 if (partialFreezeAndConstraints)
1616 auto xp = makeArrayRef(*upd->xp()).subArray(0, homenr);
1617 auto x = makeConstArrayRef(state->x).subArray(0, homenr);
1618 for (int i = 0; i < homenr; i++)
1620 int g = md->cFREEZE[i];
1622 for (int d = 0; d < DIM; d++)
1633 if (graph && graph->numNodes() > 0)
1635 unshift_x(graph, state->box, state->x.rvec_array(), upd->xp()->rvec_array());
1636 if (TRICLINIC(state->box))
1638 inc_nrnb(nrnb, eNR_SHIFTX, 2 * graph->numNodes());
1642 inc_nrnb(nrnb, eNR_SHIFTX, graph->numNodes());
1647 auto xp = makeConstArrayRef(*upd->xp()).subArray(0, homenr);
1648 auto x = makeArrayRef(state->x).subArray(0, homenr);
1651 int gmx_unused nth = gmx_omp_nthreads_get(emntUpdate);
1652 #pragma omp parallel for num_threads(nth) schedule(static)
1653 for (int i = 0; i < homenr; i++)
1655 // Trivial statement, does not throw
1659 wallcycle_stop(wcycle, ewcUPDATE);
1661 /* ############# END the update of velocities and positions ######### */
1664 void update_pcouple_after_coordinates(FILE* fplog,
1666 const t_inputrec* inputrec,
1667 const t_mdatoms* md,
1668 const matrix pressure,
1669 const matrix forceVirial,
1670 const matrix constraintVirial,
1671 matrix pressureCouplingMu,
1675 const bool scaleCoordinates)
1678 int homenr = md->homenr;
1680 /* Cast to real for faster code, no loss in precision (see comment above) */
1681 real dt = inputrec->delta_t;
1684 /* now update boxes */
1685 switch (inputrec->epc)
1687 case (epcNO): break;
1688 case (epcBERENDSEN):
1689 if (do_per_step(step, inputrec->nstpcouple))
1691 real dtpc = inputrec->nstpcouple * dt;
1692 berendsen_pcoupl(fplog, step, inputrec, dtpc, pressure, state->box, forceVirial,
1693 constraintVirial, pressureCouplingMu, &state->baros_integral);
1694 berendsen_pscale(inputrec, pressureCouplingMu, state->box, state->box_rel, start,
1695 homenr, state->x.rvec_array(), md->cFREEZE, nrnb, scaleCoordinates);
1698 case (epcPARRINELLORAHMAN):
1699 if (do_per_step(step + inputrec->nstpcouple - 1, inputrec->nstpcouple))
1701 /* The box velocities were updated in do_pr_pcoupl,
1702 * but we dont change the box vectors until we get here
1703 * since we need to be able to shift/unshift above.
1705 real dtpc = inputrec->nstpcouple * dt;
1706 for (int i = 0; i < DIM; i++)
1708 for (int m = 0; m <= i; m++)
1710 state->box[i][m] += dtpc * state->boxv[i][m];
1713 preserve_box_shape(inputrec, state->box_rel, state->box);
1715 /* Scale the coordinates */
1716 if (scaleCoordinates)
1718 auto x = state->x.rvec_array();
1719 for (int n = start; n < start + homenr; n++)
1721 tmvmul_ur0(pressureCouplingMu, x[n], x[n]);
1727 switch (inputrec->epct)
1729 case (epctISOTROPIC):
1730 /* DIM * eta = ln V. so DIM*eta_new = DIM*eta_old + DIM*dt*veta =>
1731 ln V_new = ln V_old + 3*dt*veta => V_new = V_old*exp(3*dt*veta) =>
1732 Side length scales as exp(veta*dt) */
1734 msmul(state->box, std::exp(state->veta * dt), state->box);
1736 /* Relate veta to boxv. veta = d(eta)/dT = (1/DIM)*1/V dV/dT.
1737 o If we assume isotropic scaling, and box length scaling
1738 factor L, then V = L^DIM (det(M)). So dV/dt = DIM
1739 L^(DIM-1) dL/dt det(M), and veta = (1/L) dL/dt. The
1740 determinant of B is L^DIM det(M), and the determinant
1741 of dB/dt is (dL/dT)^DIM det (M). veta will be
1742 (det(dB/dT)/det(B))^(1/3). Then since M =
1743 B_new*(vol_new)^(1/3), dB/dT_new = (veta_new)*B(new). */
1745 msmul(state->box, state->veta, state->boxv);
1755 auto localX = makeArrayRef(state->x).subArray(start, homenr);
1756 upd->deform()->apply(localX, state->box, step);
1760 void update_coords(int64_t step,
1761 const t_inputrec* inputrec, /* input record and box stuff */
1762 const t_mdatoms* md,
1764 gmx::ArrayRefWithPadding<const gmx::RVec> f,
1765 const t_fcdata* fcd,
1766 const gmx_ekindata_t* ekind,
1770 const t_commrec* cr, /* these shouldn't be here -- need to think about it */
1771 const gmx::Constraints* constr)
1773 gmx_bool bDoConstr = (nullptr != constr);
1775 /* Running the velocity half does nothing except for velocity verlet */
1776 if ((UpdatePart == etrtVELOCITY1 || UpdatePart == etrtVELOCITY2) && !EI_VV(inputrec->eI))
1778 gmx_incons("update_coords called for velocity without VV integrator");
1781 int homenr = md->homenr;
1783 /* Cast to real for faster code, no loss in precision (see comment above) */
1784 real dt = inputrec->delta_t;
1786 /* We need to update the NMR restraint history when time averaging is used */
1787 if (state->flags & (1 << estDISRE_RM3TAV))
1789 update_disres_history(fcd, &state->hist);
1791 if (state->flags & (1 << estORIRE_DTAV))
1793 update_orires_history(fcd, &state->hist);
1796 /* ############# START The update of velocities and positions ######### */
1797 int nth = gmx_omp_nthreads_get(emntUpdate);
1799 #pragma omp parallel for num_threads(nth) schedule(static)
1800 for (int th = 0; th < nth; th++)
1804 int start_th, end_th;
1805 getThreadAtomRange(nth, th, homenr, &start_th, &end_th);
1807 const rvec* x_rvec = state->x.rvec_array();
1808 rvec* xp_rvec = upd->xp()->rvec_array();
1809 rvec* v_rvec = state->v.rvec_array();
1810 const rvec* f_rvec = as_rvec_array(f.unpaddedArrayRef().data());
1812 switch (inputrec->eI)
1815 do_update_md(start_th, end_th, step, dt, inputrec, md, ekind, state->box,
1816 x_rvec, xp_rvec, v_rvec, f_rvec, state->nosehoover_vxi.data(), M);
1821 // With constraints, the SD update is done in 2 parts
1822 doSDUpdateGeneral<SDUpdate::ForcesOnly>(
1823 *upd->sd(), start_th, end_th, dt, inputrec->opts.acc, inputrec->opts.nFreeze,
1824 md->invmass, md->ptype, md->cFREEZE, md->cACC, nullptr, x_rvec,
1825 xp_rvec, v_rvec, f_rvec, step, inputrec->ld_seed, nullptr);
1829 doSDUpdateGeneral<SDUpdate::Combined>(
1830 *upd->sd(), start_th, end_th, dt, inputrec->opts.acc,
1831 inputrec->opts.nFreeze, md->invmass, md->ptype, md->cFREEZE, md->cACC,
1832 md->cTC, x_rvec, xp_rvec, v_rvec, f_rvec, step, inputrec->ld_seed,
1833 DOMAINDECOMP(cr) ? cr->dd->globalAtomIndices.data() : nullptr);
1837 do_update_bd(start_th, end_th, dt, inputrec->opts.nFreeze, md->invmass,
1838 md->ptype, md->cFREEZE, md->cTC, x_rvec, xp_rvec, v_rvec, f_rvec,
1839 inputrec->bd_fric, upd->sd()->bd_rf.data(), step, inputrec->ld_seed,
1840 DOMAINDECOMP(cr) ? cr->dd->globalAtomIndices.data() : nullptr);
1845 gmx_bool bExtended = (inputrec->etc == etcNOSEHOOVER || inputrec->epc == epcPARRINELLORAHMAN
1846 || inputrec->epc == epcMTTK);
1848 /* assuming barostat coupled to group 0 */
1849 real alpha = 1.0 + DIM / static_cast<real>(inputrec->opts.nrdf[0]);
1854 do_update_vv_vel(start_th, end_th, dt, inputrec->opts.acc,
1855 inputrec->opts.nFreeze, md->invmass, md->ptype, md->cFREEZE,
1856 md->cACC, v_rvec, f_rvec, bExtended, state->veta, alpha);
1859 do_update_vv_pos(start_th, end_th, dt, inputrec->opts.nFreeze, md->ptype,
1860 md->cFREEZE, x_rvec, xp_rvec, v_rvec, bExtended, state->veta);
1865 default: gmx_fatal(FARGS, "Don't know how to update coordinates");
1868 GMX_CATCH_ALL_AND_EXIT_WITH_FATAL_ERROR
1872 extern gmx_bool update_randomize_velocities(const t_inputrec* ir,
1874 const t_commrec* cr,
1875 const t_mdatoms* md,
1876 gmx::ArrayRef<gmx::RVec> v,
1878 const gmx::Constraints* constr)
1881 real rate = (ir->delta_t) / ir->opts.tau_t[0];
1883 if (ir->etc == etcANDERSEN && constr != nullptr)
1885 /* Currently, Andersen thermostat does not support constrained
1886 systems. Functionality exists in the andersen_tcoupl
1887 function in GROMACS 4.5.7 to allow this combination. That
1888 code could be ported to the current random-number
1889 generation approach, but has not yet been done because of
1890 lack of time and resources. */
1892 "Normal Andersen is currently not supported with constraints, use massive "
1893 "Andersen instead");
1896 /* proceed with andersen if 1) it's fixed probability per
1897 particle andersen or 2) it's massive andersen and it's tau_t/dt */
1898 if ((ir->etc == etcANDERSEN) || do_per_step(step, roundToInt(1.0 / rate)))
1900 andersen_tcoupl(ir, step, cr, md, v, rate, upd->sd()->randomize_group, upd->sd()->boltzfac);