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) 2013,2014,2015,2016,2017,2018,2019, by the GROMACS development team, led by
7 * Mark Abraham, David van der Spoel, Berk Hess, and Erik Lindahl,
8 * and including many others, as listed in the AUTHORS file in the
9 * top-level source directory and at http://www.gromacs.org.
11 * GROMACS is free software; you can redistribute it and/or
12 * modify it under the terms of the GNU Lesser General Public License
13 * as published by the Free Software Foundation; either version 2.1
14 * of the License, or (at your option) any later version.
16 * GROMACS is distributed in the hope that it will be useful,
17 * but WITHOUT ANY WARRANTY; without even the implied warranty of
18 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
19 * Lesser General Public License for more details.
21 * You should have received a copy of the GNU Lesser General Public
22 * License along with GROMACS; if not, see
23 * http://www.gnu.org/licenses, or write to the Free Software Foundation,
24 * Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA.
26 * If you want to redistribute modifications to GROMACS, please
27 * consider that scientific software is very special. Version
28 * control is crucial - bugs must be traceable. We will be happy to
29 * consider code for inclusion in the official distribution, but
30 * derived work must not be called official GROMACS. Details are found
31 * in the README & COPYING files - if they are missing, get the
32 * official version at http://www.gromacs.org.
34 * To help us fund GROMACS development, we humbly ask that you cite
35 * the research papers on the package. Check out http://www.gromacs.org.
47 #include "gromacs/domdec/domdec_struct.h"
48 #include "gromacs/fileio/confio.h"
49 #include "gromacs/gmxlib/network.h"
50 #include "gromacs/gmxlib/nrnb.h"
51 #include "gromacs/listed_forces/disre.h"
52 #include "gromacs/listed_forces/orires.h"
53 #include "gromacs/math/functions.h"
54 #include "gromacs/math/invertmatrix.h"
55 #include "gromacs/math/paddedvector.h"
56 #include "gromacs/math/units.h"
57 #include "gromacs/math/vec.h"
58 #include "gromacs/math/vecdump.h"
59 #include "gromacs/mdlib/boxdeformation.h"
60 #include "gromacs/mdlib/constr.h"
61 #include "gromacs/mdlib/gmx_omp_nthreads.h"
62 #include "gromacs/mdlib/mdatoms.h"
63 #include "gromacs/mdlib/stat.h"
64 #include "gromacs/mdlib/tgroup.h"
65 #include "gromacs/mdtypes/commrec.h"
66 #include "gromacs/mdtypes/group.h"
67 #include "gromacs/mdtypes/inputrec.h"
68 #include "gromacs/mdtypes/md_enums.h"
69 #include "gromacs/mdtypes/state.h"
70 #include "gromacs/pbcutil/boxutilities.h"
71 #include "gromacs/pbcutil/mshift.h"
72 #include "gromacs/pbcutil/pbc.h"
73 #include "gromacs/pulling/pull.h"
74 #include "gromacs/random/tabulatednormaldistribution.h"
75 #include "gromacs/random/threefry.h"
76 #include "gromacs/simd/simd.h"
77 #include "gromacs/timing/wallcycle.h"
78 #include "gromacs/topology/atoms.h"
79 #include "gromacs/utility/exceptions.h"
80 #include "gromacs/utility/fatalerror.h"
81 #include "gromacs/utility/futil.h"
82 #include "gromacs/utility/gmxassert.h"
83 #include "gromacs/utility/gmxomp.h"
84 #include "gromacs/utility/smalloc.h"
86 using namespace gmx; // TODO: Remove when this file is moved into gmx namespace
101 std::vector<real> bd_rf;
103 std::vector<gmx_sd_const_t> sdc;
104 std::vector<gmx_sd_sigma_t> sdsig;
105 /* andersen temperature control stuff */
106 std::vector<bool> randomize_group;
107 std::vector<real> boltzfac;
109 explicit gmx_stochd_t(const t_inputrec* ir);
112 //! pImpled implementation for Update
117 Impl(const t_inputrec* ir, BoxDeformation* boxDeformation);
120 //! stochastic dynamics struct
121 std::unique_ptr<gmx_stochd_t> sd;
122 //! xprime for constraint algorithms
123 PaddedVector<RVec> xp;
124 //! Box deformation handler (or nullptr if inactive).
125 BoxDeformation* deform = nullptr;
128 Update::Update(const t_inputrec* ir, BoxDeformation* boxDeformation) :
129 impl_(new Impl(ir, boxDeformation)){};
133 gmx_stochd_t* Update::sd() const
135 return impl_->sd.get();
138 PaddedVector<RVec>* Update::xp()
143 BoxDeformation* Update::deform() const
145 return impl_->deform;
148 /*! \brief Sets the velocities of virtual sites to zero */
149 static void clearVsiteVelocities(int start, int nrend, const unsigned short* particleType, rvec* gmx_restrict v)
151 for (int a = start; a < nrend; a++)
153 if (particleType[a] == eptVSite)
160 /*! \brief Sets the number of different temperature coupling values */
161 enum class NumTempScaleValues
163 single, //!< Single T-scaling value (either one group or all values =1)
164 multiple //!< Multiple T-scaling values, need to use T-group indices
167 /*! \brief Sets if to apply no or diagonal Parrinello-Rahman pressure scaling
169 * Note that this enum is only used in updateMDLeapfrogSimple(), which does
170 * not handle fully anistropic Parrinello-Rahman scaling, so we only have
171 * options \p no and \p diagonal here and no anistropic option.
173 enum class ApplyParrinelloRahmanVScaling
175 no, //!< Do not apply velocity scaling (not a PR-coupling run or step)
176 diagonal //!< Apply velocity scaling using a diagonal matrix
179 /*! \brief Integrate using leap-frog with T-scaling and optionally diagonal Parrinello-Rahman p-coupling
181 * \tparam numTempScaleValues The number of different T-couple values
182 * \tparam applyPRVScaling Apply Parrinello-Rahman velocity scaling
183 * \param[in] start Index of first atom to update
184 * \param[in] nrend Last atom to update: \p nrend - 1
185 * \param[in] dt The time step
186 * \param[in] dtPressureCouple Time step for pressure coupling
187 * \param[in] invMassPerDim 1/mass per atom and dimension
188 * \param[in] tcstat Temperature coupling information
189 * \param[in] cTC T-coupling group index per atom
190 * \param[in] pRVScaleMatrixDiagonal Parrinello-Rahman v-scale matrix diagonal
191 * \param[in] x Input coordinates
192 * \param[out] xprime Updated coordinates
193 * \param[inout] v Velocities
194 * \param[in] f Forces
196 * We expect this template to get good SIMD acceleration by most compilers,
197 * unlike the more complex general template.
198 * Note that we might get even better SIMD acceleration when we introduce
199 * aligned (and padded) memory, possibly with some hints for the compilers.
201 template<NumTempScaleValues numTempScaleValues, ApplyParrinelloRahmanVScaling applyPRVScaling>
202 static void updateMDLeapfrogSimple(int start,
205 real dtPressureCouple,
206 const rvec* gmx_restrict invMassPerDim,
207 gmx::ArrayRef<const t_grp_tcstat> tcstat,
208 const unsigned short* cTC,
209 const rvec pRVScaleMatrixDiagonal,
210 const rvec* gmx_restrict x,
211 rvec* gmx_restrict xprime,
212 rvec* gmx_restrict v,
213 const rvec* gmx_restrict f)
217 if (numTempScaleValues == NumTempScaleValues::single)
219 lambdaGroup = tcstat[0].lambda;
222 for (int a = start; a < nrend; a++)
224 if (numTempScaleValues == NumTempScaleValues::multiple)
226 lambdaGroup = tcstat[cTC[a]].lambda;
229 for (int d = 0; d < DIM; d++)
231 /* Note that using rvec invMassPerDim results in more efficient
232 * SIMD code, but this increases the cache pressure.
233 * For large systems with PME on the CPU this slows down the
234 * (then already slow) update by 20%. If all data remains in cache,
235 * using rvec is much faster.
237 real vNew = lambdaGroup * v[a][d] + f[a][d] * invMassPerDim[a][d] * dt;
239 if (applyPRVScaling == ApplyParrinelloRahmanVScaling::diagonal)
241 vNew -= dtPressureCouple * pRVScaleMatrixDiagonal[d] * v[a][d];
244 xprime[a][d] = x[a][d] + vNew * dt;
249 #if GMX_SIMD && GMX_SIMD_HAVE_REAL
250 # define GMX_HAVE_SIMD_UPDATE 1
252 # define GMX_HAVE_SIMD_UPDATE 0
255 #if GMX_HAVE_SIMD_UPDATE
257 /*! \brief Load (aligned) the contents of GMX_SIMD_REAL_WIDTH rvec elements sequentially into 3 SIMD registers
259 * The loaded output is:
260 * \p r0: { r[index][XX], r[index][YY], ... }
262 * \p r2: { ..., r[index+GMX_SIMD_REAL_WIDTH-1][YY], r[index+GMX_SIMD_REAL_WIDTH-1][ZZ] }
264 * \param[in] r Real to an rvec array, has to be aligned to SIMD register width
265 * \param[in] index Index of the first rvec triplet of reals to load
266 * \param[out] r0 Pointer to first SIMD register
267 * \param[out] r1 Pointer to second SIMD register
268 * \param[out] r2 Pointer to third SIMD register
270 static inline void simdLoadRvecs(const rvec* r, int index, SimdReal* r0, SimdReal* r1, SimdReal* r2)
272 const real* realPtr = r[index];
274 GMX_ASSERT(isSimdAligned(realPtr), "Pointer should be SIMD aligned");
276 *r0 = simdLoad(realPtr + 0 * GMX_SIMD_REAL_WIDTH);
277 *r1 = simdLoad(realPtr + 1 * GMX_SIMD_REAL_WIDTH);
278 *r2 = simdLoad(realPtr + 2 * GMX_SIMD_REAL_WIDTH);
281 /*! \brief Store (aligned) 3 SIMD registers sequentially to GMX_SIMD_REAL_WIDTH rvec elements
283 * The stored output is:
284 * \p r[index] = { { r0[0], r0[1], ... }
286 * \p r[index+GMX_SIMD_REAL_WIDTH-1] = { ... , r2[GMX_SIMD_REAL_WIDTH-2], r2[GMX_SIMD_REAL_WIDTH-1] }
288 * \param[out] r Pointer to an rvec array, has to be aligned to SIMD register width
289 * \param[in] index Index of the first rvec triplet of reals to store to
290 * \param[in] r0 First SIMD register
291 * \param[in] r1 Second SIMD register
292 * \param[in] r2 Third SIMD register
294 static inline void simdStoreRvecs(rvec* r, int index, SimdReal r0, SimdReal r1, SimdReal r2)
296 real* realPtr = r[index];
298 GMX_ASSERT(isSimdAligned(realPtr), "Pointer should be SIMD aligned");
300 store(realPtr + 0 * GMX_SIMD_REAL_WIDTH, r0);
301 store(realPtr + 1 * GMX_SIMD_REAL_WIDTH, r1);
302 store(realPtr + 2 * GMX_SIMD_REAL_WIDTH, r2);
305 /*! \brief Integrate using leap-frog with single group T-scaling and SIMD
307 * \param[in] start Index of first atom to update
308 * \param[in] nrend Last atom to update: \p nrend - 1
309 * \param[in] dt The time step
310 * \param[in] invMass 1/mass per atom
311 * \param[in] tcstat Temperature coupling information
312 * \param[in] x Input coordinates
313 * \param[out] xprime Updated coordinates
314 * \param[inout] v Velocities
315 * \param[in] f Forces
317 static void updateMDLeapfrogSimpleSimd(int start,
320 const real* gmx_restrict invMass,
321 gmx::ArrayRef<const t_grp_tcstat> tcstat,
322 const rvec* gmx_restrict x,
323 rvec* gmx_restrict xprime,
324 rvec* gmx_restrict v,
325 const rvec* gmx_restrict f)
327 SimdReal timestep(dt);
328 SimdReal lambdaSystem(tcstat[0].lambda);
330 /* We declare variables here, since code is often slower when declaring them inside the loop */
332 /* Note: We should implement a proper PaddedVector, so we don't need this check */
333 GMX_ASSERT(isSimdAligned(invMass), "invMass should be aligned");
335 for (int a = start; a < nrend; a += GMX_SIMD_REAL_WIDTH)
337 SimdReal invMass0, invMass1, invMass2;
338 expandScalarsToTriplets(simdLoad(invMass + a), &invMass0, &invMass1, &invMass2);
342 simdLoadRvecs(v, a, &v0, &v1, &v2);
343 simdLoadRvecs(f, a, &f0, &f1, &f2);
345 v0 = fma(f0 * invMass0, timestep, lambdaSystem * v0);
346 v1 = fma(f1 * invMass1, timestep, lambdaSystem * v1);
347 v2 = fma(f2 * invMass2, timestep, lambdaSystem * v2);
349 simdStoreRvecs(v, a, v0, v1, v2);
352 simdLoadRvecs(x, a, &x0, &x1, &x2);
354 SimdReal xprime0 = fma(v0, timestep, x0);
355 SimdReal xprime1 = fma(v1, timestep, x1);
356 SimdReal xprime2 = fma(v2, timestep, x2);
358 simdStoreRvecs(xprime, a, xprime0, xprime1, xprime2);
362 #endif // GMX_HAVE_SIMD_UPDATE
364 /*! \brief Sets the NEMD acceleration type */
365 enum class AccelerationType
372 /*! \brief Integrate using leap-frog with support for everything
374 * \tparam accelerationType Type of NEMD acceleration
375 * \param[in] start Index of first atom to update
376 * \param[in] nrend Last atom to update: \p nrend - 1
377 * \param[in] doNoseHoover If to apply Nose-Hoover T-coupling
378 * \param[in] dt The time step
379 * \param[in] dtPressureCouple Time step for pressure coupling, is 0 when no pressure
380 * coupling should be applied at this step \param[in] ir The input parameter
381 * record \param[in] md Atom properties \param[in] ekind Kinetic energy
382 * data \param[in] box The box dimensions \param[in] x Input coordinates \param[out]
383 * xprime Updated coordinates \param[inout] v Velocities \param[in]
384 * f Forces \param[in] nh_vxi Nose-Hoover velocity scaling
385 * factors \param[in] M Parrinello-Rahman scaling matrix
387 template<AccelerationType accelerationType>
388 static void updateMDLeapfrogGeneral(int start,
392 real dtPressureCouple,
393 const t_inputrec* ir,
395 const gmx_ekindata_t* ekind,
397 const rvec* gmx_restrict x,
398 rvec* gmx_restrict xprime,
399 rvec* gmx_restrict v,
400 const rvec* gmx_restrict f,
401 const double* gmx_restrict nh_vxi,
404 /* This is a version of the leap-frog integrator that supports
405 * all combinations of T-coupling, P-coupling and NEMD.
406 * Nose-Hoover uses the reversible leap-frog integrator from
407 * Holian et al. Phys Rev E 52(3) : 2338, 1995
410 gmx::ArrayRef<const t_grp_tcstat> tcstat = ekind->tcstat;
411 gmx::ArrayRef<const t_grp_acc> grpstat = ekind->grpstat;
412 const unsigned short* cTC = md->cTC;
413 const unsigned short* cACC = md->cACC;
414 const rvec* accel = ir->opts.acc;
416 const rvec* gmx_restrict invMassPerDim = md->invMassPerDim;
418 /* Initialize group values, changed later when multiple groups are used */
423 real omega_Z = 2 * static_cast<real>(M_PI) / box[ZZ][ZZ];
425 for (int n = start; n < nrend; n++)
431 real lg = tcstat[gt].lambda;
434 real cosineZ, vCosine;
435 #ifdef __INTEL_COMPILER
436 # pragma warning(disable : 280)
438 switch (accelerationType)
440 case AccelerationType::none: copy_rvec(v[n], vRel); break;
441 case AccelerationType::group:
446 /* Avoid scaling the group velocity */
447 rvec_sub(v[n], grpstat[ga].u, vRel);
449 case AccelerationType::cosine:
450 cosineZ = std::cos(x[n][ZZ] * omega_Z);
451 vCosine = cosineZ * ekind->cosacc.vcos;
452 /* Avoid scaling the cosine profile velocity */
453 copy_rvec(v[n], vRel);
460 /* Here we account for multiple time stepping, by increasing
461 * the Nose-Hoover correction by nsttcouple
463 factorNH = 0.5 * ir->nsttcouple * dt * nh_vxi[gt];
466 for (int d = 0; d < DIM; d++)
468 real vNew = (lg * vRel[d]
469 + (f[n][d] * invMassPerDim[n][d] * dt - factorNH * vRel[d]
470 - dtPressureCouple * iprod(M[d], vRel)))
472 switch (accelerationType)
474 case AccelerationType::none: break;
475 case AccelerationType::group:
476 /* Add back the mean velocity and apply acceleration */
477 vNew += grpstat[ga].u[d] + accel[ga][d] * dt;
479 case AccelerationType::cosine:
482 /* Add back the mean velocity and apply acceleration */
483 vNew += vCosine + cosineZ * ekind->cosacc.cos_accel * dt;
488 xprime[n][d] = x[n][d] + vNew * dt;
493 /*! \brief Handles the Leap-frog MD x and v integration */
494 static void do_update_md(int start,
498 const t_inputrec* ir,
500 const gmx_ekindata_t* ekind,
502 const rvec* gmx_restrict x,
503 rvec* gmx_restrict xprime,
504 rvec* gmx_restrict v,
505 const rvec* gmx_restrict f,
506 const double* gmx_restrict nh_vxi,
509 GMX_ASSERT(nrend == start || xprime != x,
510 "For SIMD optimization certain compilers need to have xprime != x");
511 GMX_ASSERT(ir->eI == eiMD,
512 "Leap-frog integrator was called while another integrator was requested");
514 /* Note: Berendsen pressure scaling is handled after do_update_md() */
515 bool doTempCouple = (ir->etc != etcNO && do_per_step(step + ir->nsttcouple - 1, ir->nsttcouple));
516 bool doNoseHoover = (ir->etc == etcNOSEHOOVER && doTempCouple);
517 bool doParrinelloRahman =
518 (ir->epc == epcPARRINELLORAHMAN && do_per_step(step + ir->nstpcouple - 1, ir->nstpcouple));
519 bool doPROffDiagonal = (doParrinelloRahman && (M[YY][XX] != 0 || M[ZZ][XX] != 0 || M[ZZ][YY] != 0));
521 real dtPressureCouple = (doParrinelloRahman ? ir->nstpcouple * dt : 0);
523 /* NEMD (also cosine) acceleration is applied in updateMDLeapFrogGeneral */
524 bool doAcceleration = (ekind->bNEMD || ekind->cosacc.cos_accel != 0);
526 if (doNoseHoover || doPROffDiagonal || doAcceleration)
529 if (!doParrinelloRahman)
531 /* We should not apply PR scaling at this step */
541 updateMDLeapfrogGeneral<AccelerationType::none>(start, nrend, doNoseHoover, dt,
542 dtPressureCouple, ir, md, ekind, box, x,
543 xprime, v, f, nh_vxi, stepM);
545 else if (ekind->bNEMD)
547 updateMDLeapfrogGeneral<AccelerationType::group>(start, nrend, doNoseHoover, dt,
548 dtPressureCouple, ir, md, ekind, box,
549 x, xprime, v, f, nh_vxi, stepM);
553 updateMDLeapfrogGeneral<AccelerationType::cosine>(start, nrend, doNoseHoover, dt,
554 dtPressureCouple, ir, md, ekind, box,
555 x, xprime, v, f, nh_vxi, stepM);
560 /* Use a simple and thus more efficient integration loop. */
561 /* The simple loop does not check for particle type (so it can
562 * be vectorized), which means we need to clear the velocities
563 * of virtual sites in advance, when present. Note that vsite
564 * velocities are computed after update and constraints from
565 * their displacement.
569 /* Note: The overhead of this loop is completely neligible */
570 clearVsiteVelocities(start, nrend, md->ptype, v);
573 /* We determine if we have a single T-coupling lambda value for all
574 * atoms. That allows for better SIMD acceleration in the template.
575 * If we do not do temperature coupling (in the run or this step),
576 * all scaling values are 1, so we effectively have a single value.
578 bool haveSingleTempScaleValue = (!doTempCouple || ekind->ngtc == 1);
580 /* Extract some pointers needed by all cases */
581 const unsigned short* cTC = md->cTC;
582 gmx::ArrayRef<const t_grp_tcstat> tcstat = ekind->tcstat;
583 const rvec* invMassPerDim = md->invMassPerDim;
585 if (doParrinelloRahman)
587 GMX_ASSERT(!doPROffDiagonal,
588 "updateMDLeapfrogSimple only support diagonal Parrinello-Rahman scaling "
592 for (int d = 0; d < DIM; d++)
597 if (haveSingleTempScaleValue)
599 updateMDLeapfrogSimple<NumTempScaleValues::single, ApplyParrinelloRahmanVScaling::diagonal>(
600 start, nrend, dt, dtPressureCouple, invMassPerDim, tcstat, cTC, diagM, x,
605 updateMDLeapfrogSimple<NumTempScaleValues::multiple, ApplyParrinelloRahmanVScaling::diagonal>(
606 start, nrend, dt, dtPressureCouple, invMassPerDim, tcstat, cTC, diagM, x,
612 if (haveSingleTempScaleValue)
614 /* Note that modern compilers are pretty good at vectorizing
615 * updateMDLeapfrogSimple(). But the SIMD version will still
616 * be faster because invMass lowers the cache pressure
617 * compared to invMassPerDim.
619 #if GMX_HAVE_SIMD_UPDATE
620 /* Check if we can use invmass instead of invMassPerDim */
621 if (!md->havePartiallyFrozenAtoms)
623 updateMDLeapfrogSimpleSimd(start, nrend, dt, md->invmass, tcstat, x, xprime, v, f);
628 updateMDLeapfrogSimple<NumTempScaleValues::single, ApplyParrinelloRahmanVScaling::no>(
629 start, nrend, dt, dtPressureCouple, invMassPerDim, tcstat, cTC, nullptr,
635 updateMDLeapfrogSimple<NumTempScaleValues::multiple, ApplyParrinelloRahmanVScaling::no>(
636 start, nrend, dt, dtPressureCouple, invMassPerDim, tcstat, cTC, nullptr, x,
643 static void do_update_vv_vel(int start,
647 const ivec nFreeze[],
648 const real invmass[],
649 const unsigned short ptype[],
650 const unsigned short cFREEZE[],
651 const unsigned short cACC[],
664 g = 0.25 * dt * veta * alpha;
666 mv2 = gmx::series_sinhx(g);
673 for (n = start; n < nrend; n++)
675 real w_dt = invmass[n] * dt;
685 for (d = 0; d < DIM; d++)
687 if ((ptype[n] != eptVSite) && (ptype[n] != eptShell) && !nFreeze[gf][d])
689 v[n][d] = mv1 * (mv1 * v[n][d] + 0.5 * (w_dt * mv2 * f[n][d])) + 0.5 * accel[ga][d] * dt;
697 } /* do_update_vv_vel */
699 static void do_update_vv_pos(int start,
702 const ivec nFreeze[],
703 const unsigned short ptype[],
704 const unsigned short cFREEZE[],
715 /* Would it make more sense if Parrinello-Rahman was put here? */
720 mr2 = gmx::series_sinhx(g);
728 for (n = start; n < nrend; n++)
736 for (d = 0; d < DIM; d++)
738 if ((ptype[n] != eptVSite) && (ptype[n] != eptShell) && !nFreeze[gf][d])
740 xprime[n][d] = mr1 * (mr1 * x[n][d] + mr2 * dt * v[n][d]);
744 xprime[n][d] = x[n][d];
748 } /* do_update_vv_pos */
750 gmx_stochd_t::gmx_stochd_t(const t_inputrec* ir)
752 const t_grpopts* opts = &ir->opts;
753 const int ngtc = opts->ngtc;
759 else if (EI_SD(ir->eI))
764 for (int gt = 0; gt < ngtc; gt++)
766 if (opts->tau_t[gt] > 0)
768 sdc[gt].em = std::exp(-ir->delta_t / opts->tau_t[gt]);
772 /* No friction and noise on this group */
777 else if (ETC_ANDERSEN(ir->etc))
779 randomize_group.resize(ngtc);
780 boltzfac.resize(ngtc);
782 /* for now, assume that all groups, if randomized, are randomized at the same rate, i.e. tau_t is the same. */
783 /* since constraint groups don't necessarily match up with temperature groups! This is checked in readir.c */
785 for (int gt = 0; gt < ngtc; gt++)
787 real reft = std::max<real>(0, opts->ref_t[gt]);
788 if ((opts->tau_t[gt] > 0)
789 && (reft > 0)) /* tau_t or ref_t = 0 means that no randomization is done */
791 randomize_group[gt] = true;
792 boltzfac[gt] = BOLTZ * opts->ref_t[gt];
796 randomize_group[gt] = false;
802 void update_temperature_constants(gmx_stochd_t* sd, const t_inputrec* ir)
806 if (ir->bd_fric != 0)
808 for (int gt = 0; gt < ir->opts.ngtc; gt++)
810 sd->bd_rf[gt] = std::sqrt(2.0 * BOLTZ * ir->opts.ref_t[gt] / (ir->bd_fric * ir->delta_t));
815 for (int gt = 0; gt < ir->opts.ngtc; gt++)
817 sd->bd_rf[gt] = std::sqrt(2.0 * BOLTZ * ir->opts.ref_t[gt]);
823 for (int gt = 0; gt < ir->opts.ngtc; gt++)
825 real kT = BOLTZ * ir->opts.ref_t[gt];
826 /* The mass is accounted for later, since this differs per atom */
827 sd->sdsig[gt].V = std::sqrt(kT * (1 - sd->sdc[gt].em * sd->sdc[gt].em));
832 Update::Impl::Impl(const t_inputrec* ir, BoxDeformation* boxDeformation)
834 sd = std::make_unique<gmx_stochd_t>(ir);
835 update_temperature_constants(sd.get(), ir);
836 xp.resizeWithPadding(0);
837 deform = boxDeformation;
840 void Update::setNumAtoms(int nAtoms)
843 impl_->xp.resizeWithPadding(nAtoms);
846 /*! \brief Sets the SD update type */
847 enum class SDUpdate : int
850 FrictionAndNoiseOnly,
854 /*! \brief SD integrator update
856 * Two phases are required in the general case of a constrained
857 * update, the first phase from the contribution of forces, before
858 * applying constraints, and then a second phase applying the friction
859 * and noise, and then further constraining. For details, see
862 * Without constraints, the two phases can be combined, for
865 * Thus three instantiations of this templated function will be made,
866 * two with only one contribution, and one with both contributions. */
867 template<SDUpdate updateType>
868 static void doSDUpdateGeneral(const gmx_stochd_t& sd,
873 const ivec nFreeze[],
874 const real invmass[],
875 const unsigned short ptype[],
876 const unsigned short cFREEZE[],
877 const unsigned short cACC[],
878 const unsigned short cTC[],
887 // cTC, cACC and cFREEZE can be nullptr any time, but various
888 // instantiations do not make sense with particular pointer
890 if (updateType == SDUpdate::ForcesOnly)
892 GMX_ASSERT(f != nullptr, "SD update with only forces requires forces");
893 GMX_ASSERT(cTC == nullptr, "SD update with only forces cannot handle temperature groups");
895 if (updateType == SDUpdate::FrictionAndNoiseOnly)
897 GMX_ASSERT(f == nullptr, "SD update with only noise cannot handle forces");
898 GMX_ASSERT(cACC == nullptr, "SD update with only noise cannot handle acceleration groups");
900 if (updateType == SDUpdate::Combined)
902 GMX_ASSERT(f != nullptr, "SD update with forces and noise requires forces");
905 // Even 0 bits internal counter gives 2x64 ints (more than enough for three table lookups)
906 gmx::ThreeFry2x64<0> rng(seed, gmx::RandomDomain::UpdateCoordinates);
907 gmx::TabulatedNormalDistribution<real, 14> dist;
909 for (int n = start; n < nrend; n++)
911 int globalAtomIndex = gatindex ? gatindex[n] : n;
912 rng.restart(step, globalAtomIndex);
915 real inverseMass = invmass[n];
916 real invsqrtMass = std::sqrt(inverseMass);
918 int freezeGroup = cFREEZE ? cFREEZE[n] : 0;
919 int accelerationGroup = cACC ? cACC[n] : 0;
920 int temperatureGroup = cTC ? cTC[n] : 0;
922 for (int d = 0; d < DIM; d++)
924 if ((ptype[n] != eptVSite) && (ptype[n] != eptShell) && !nFreeze[freezeGroup][d])
926 if (updateType == SDUpdate::ForcesOnly)
928 real vn = v[n][d] + (inverseMass * f[n][d] + accel[accelerationGroup][d]) * dt;
930 // Simple position update.
931 xprime[n][d] = x[n][d] + v[n][d] * dt;
933 else if (updateType == SDUpdate::FrictionAndNoiseOnly)
936 v[n][d] = (vn * sd.sdc[temperatureGroup].em
937 + invsqrtMass * sd.sdsig[temperatureGroup].V * dist(rng));
938 // The previous phase already updated the
939 // positions with a full v*dt term that must
940 // now be half removed.
941 xprime[n][d] = xprime[n][d] + 0.5 * (v[n][d] - vn) * dt;
945 real vn = v[n][d] + (inverseMass * f[n][d] + accel[accelerationGroup][d]) * dt;
946 v[n][d] = (vn * sd.sdc[temperatureGroup].em
947 + invsqrtMass * sd.sdsig[temperatureGroup].V * dist(rng));
948 // Here we include half of the friction+noise
949 // update of v into the position update.
950 xprime[n][d] = x[n][d] + 0.5 * (vn + v[n][d]) * dt;
955 // When using constraints, the update is split into
956 // two phases, but we only need to zero the update of
957 // virtual, shell or frozen particles in at most one
959 if (updateType != SDUpdate::FrictionAndNoiseOnly)
962 xprime[n][d] = x[n][d];
969 static void do_update_bd(int start,
972 const ivec nFreeze[],
973 const real invmass[],
974 const unsigned short ptype[],
975 const unsigned short cFREEZE[],
976 const unsigned short cTC[],
981 real friction_coefficient,
987 /* note -- these appear to be full step velocities . . . */
992 // Use 1 bit of internal counters to give us 2*2 64-bits values per stream
993 // Each 64-bit value is enough for 4 normal distribution table numbers.
994 gmx::ThreeFry2x64<0> rng(seed, gmx::RandomDomain::UpdateCoordinates);
995 gmx::TabulatedNormalDistribution<real, 14> dist;
997 if (friction_coefficient != 0)
999 invfr = 1.0 / friction_coefficient;
1002 for (n = start; (n < nrend); n++)
1004 int ng = gatindex ? gatindex[n] : n;
1006 rng.restart(step, ng);
1017 for (d = 0; (d < DIM); d++)
1019 if ((ptype[n] != eptVSite) && (ptype[n] != eptShell) && !nFreeze[gf][d])
1021 if (friction_coefficient != 0)
1023 vn = invfr * f[n][d] + rf[gt] * dist(rng);
1027 /* NOTE: invmass = 2/(mass*friction_constant*dt) */
1028 vn = 0.5 * invmass[n] * f[n][d] * dt
1029 + std::sqrt(0.5 * invmass[n]) * rf[gt] * dist(rng);
1033 xprime[n][d] = x[n][d] + vn * dt;
1038 xprime[n][d] = x[n][d];
1044 static void calc_ke_part_normal(const rvec v[],
1045 const t_grpopts* opts,
1046 const t_mdatoms* md,
1047 gmx_ekindata_t* ekind,
1049 gmx_bool bEkinAveVel)
1052 gmx::ArrayRef<t_grp_tcstat> tcstat = ekind->tcstat;
1053 gmx::ArrayRef<t_grp_acc> grpstat = ekind->grpstat;
1055 /* three main: VV with AveVel, vv with AveEkin, leap with AveEkin. Leap with AveVel is also
1056 an option, but not supported now.
1057 bEkinAveVel: If TRUE, we sum into ekin, if FALSE, into ekinh.
1060 /* group velocities are calculated in update_ekindata and
1061 * accumulated in acumulate_groups.
1062 * Now the partial global and groups ekin.
1064 for (g = 0; (g < opts->ngtc); g++)
1066 copy_mat(tcstat[g].ekinh, tcstat[g].ekinh_old);
1069 clear_mat(tcstat[g].ekinf);
1070 tcstat[g].ekinscalef_nhc = 1.0; /* need to clear this -- logic is complicated! */
1074 clear_mat(tcstat[g].ekinh);
1077 ekind->dekindl_old = ekind->dekindl;
1078 int nthread = gmx_omp_nthreads_get(emntUpdate);
1080 #pragma omp parallel for num_threads(nthread) schedule(static)
1081 for (int thread = 0; thread < nthread; thread++)
1083 // This OpenMP only loops over arrays and does not call any functions
1084 // or memory allocation. It should not be able to throw, so for now
1085 // we do not need a try/catch wrapper.
1086 int start_t, end_t, n;
1094 start_t = ((thread + 0) * md->homenr) / nthread;
1095 end_t = ((thread + 1) * md->homenr) / nthread;
1097 ekin_sum = ekind->ekin_work[thread];
1098 dekindl_sum = ekind->dekindl_work[thread];
1100 for (gt = 0; gt < opts->ngtc; gt++)
1102 clear_mat(ekin_sum[gt]);
1108 for (n = start_t; n < end_t; n++)
1118 hm = 0.5 * md->massT[n];
1120 for (d = 0; (d < DIM); d++)
1122 v_corrt[d] = v[n][d] - grpstat[ga].u[d];
1124 for (d = 0; (d < DIM); d++)
1126 for (m = 0; (m < DIM); m++)
1128 /* if we're computing a full step velocity, v_corrt[d] has v(t). Otherwise, v(t+dt/2) */
1129 ekin_sum[gt][m][d] += hm * v_corrt[m] * v_corrt[d];
1132 if (md->nMassPerturbed && md->bPerturbed[n])
1134 *dekindl_sum += 0.5 * (md->massB[n] - md->massA[n]) * iprod(v_corrt, v_corrt);
1140 for (int thread = 0; thread < nthread; thread++)
1142 for (g = 0; g < opts->ngtc; g++)
1146 m_add(tcstat[g].ekinf, ekind->ekin_work[thread][g], tcstat[g].ekinf);
1150 m_add(tcstat[g].ekinh, ekind->ekin_work[thread][g], tcstat[g].ekinh);
1154 ekind->dekindl += *ekind->dekindl_work[thread];
1157 inc_nrnb(nrnb, eNR_EKIN, md->homenr);
1160 static void calc_ke_part_visc(const matrix box,
1163 const t_grpopts* opts,
1164 const t_mdatoms* md,
1165 gmx_ekindata_t* ekind,
1167 gmx_bool bEkinAveVel)
1169 int start = 0, homenr = md->homenr;
1170 int g, d, n, m, gt = 0;
1173 gmx::ArrayRef<t_grp_tcstat> tcstat = ekind->tcstat;
1174 t_cos_acc* cosacc = &(ekind->cosacc);
1179 for (g = 0; g < opts->ngtc; g++)
1181 copy_mat(ekind->tcstat[g].ekinh, ekind->tcstat[g].ekinh_old);
1182 clear_mat(ekind->tcstat[g].ekinh);
1184 ekind->dekindl_old = ekind->dekindl;
1186 fac = 2 * M_PI / box[ZZ][ZZ];
1189 for (n = start; n < start + homenr; n++)
1195 hm = 0.5 * md->massT[n];
1197 /* Note that the times of x and v differ by half a step */
1198 /* MRS -- would have to be changed for VV */
1199 cosz = std::cos(fac * x[n][ZZ]);
1200 /* Calculate the amplitude of the new velocity profile */
1201 mvcos += 2 * cosz * md->massT[n] * v[n][XX];
1203 copy_rvec(v[n], v_corrt);
1204 /* Subtract the profile for the kinetic energy */
1205 v_corrt[XX] -= cosz * cosacc->vcos;
1206 for (d = 0; (d < DIM); d++)
1208 for (m = 0; (m < DIM); m++)
1210 /* if we're computing a full step velocity, v_corrt[d] has v(t). Otherwise, v(t+dt/2) */
1213 tcstat[gt].ekinf[m][d] += hm * v_corrt[m] * v_corrt[d];
1217 tcstat[gt].ekinh[m][d] += hm * v_corrt[m] * v_corrt[d];
1221 if (md->nPerturbed && md->bPerturbed[n])
1223 /* The minus sign here might be confusing.
1224 * The kinetic contribution from dH/dl doesn't come from
1225 * d m(l)/2 v^2 / dl, but rather from d p^2/2m(l) / dl,
1226 * where p are the momenta. The difference is only a minus sign.
1228 dekindl -= 0.5 * (md->massB[n] - md->massA[n]) * iprod(v_corrt, v_corrt);
1231 ekind->dekindl = dekindl;
1232 cosacc->mvcos = mvcos;
1234 inc_nrnb(nrnb, eNR_EKIN, homenr);
1237 void calc_ke_part(const rvec* x,
1240 const t_grpopts* opts,
1241 const t_mdatoms* md,
1242 gmx_ekindata_t* ekind,
1244 gmx_bool bEkinAveVel)
1246 if (ekind->cosacc.cos_accel == 0)
1248 calc_ke_part_normal(v, opts, md, ekind, nrnb, bEkinAveVel);
1252 calc_ke_part_visc(box, x, v, opts, md, ekind, nrnb, bEkinAveVel);
1256 extern void init_ekinstate(ekinstate_t* ekinstate, const t_inputrec* ir)
1258 ekinstate->ekin_n = ir->opts.ngtc;
1259 snew(ekinstate->ekinh, ekinstate->ekin_n);
1260 snew(ekinstate->ekinf, ekinstate->ekin_n);
1261 snew(ekinstate->ekinh_old, ekinstate->ekin_n);
1262 ekinstate->ekinscalef_nhc.resize(ekinstate->ekin_n);
1263 ekinstate->ekinscaleh_nhc.resize(ekinstate->ekin_n);
1264 ekinstate->vscale_nhc.resize(ekinstate->ekin_n);
1265 ekinstate->dekindl = 0;
1266 ekinstate->mvcos = 0;
1267 ekinstate->hasReadEkinState = false;
1270 void update_ekinstate(ekinstate_t* ekinstate, const gmx_ekindata_t* ekind)
1274 for (i = 0; i < ekinstate->ekin_n; i++)
1276 copy_mat(ekind->tcstat[i].ekinh, ekinstate->ekinh[i]);
1277 copy_mat(ekind->tcstat[i].ekinf, ekinstate->ekinf[i]);
1278 copy_mat(ekind->tcstat[i].ekinh_old, ekinstate->ekinh_old[i]);
1279 ekinstate->ekinscalef_nhc[i] = ekind->tcstat[i].ekinscalef_nhc;
1280 ekinstate->ekinscaleh_nhc[i] = ekind->tcstat[i].ekinscaleh_nhc;
1281 ekinstate->vscale_nhc[i] = ekind->tcstat[i].vscale_nhc;
1284 copy_mat(ekind->ekin, ekinstate->ekin_total);
1285 ekinstate->dekindl = ekind->dekindl;
1286 ekinstate->mvcos = ekind->cosacc.mvcos;
1289 void restore_ekinstate_from_state(const t_commrec* cr, gmx_ekindata_t* ekind, const ekinstate_t* ekinstate)
1295 for (i = 0; i < ekinstate->ekin_n; i++)
1297 copy_mat(ekinstate->ekinh[i], ekind->tcstat[i].ekinh);
1298 copy_mat(ekinstate->ekinf[i], ekind->tcstat[i].ekinf);
1299 copy_mat(ekinstate->ekinh_old[i], ekind->tcstat[i].ekinh_old);
1300 ekind->tcstat[i].ekinscalef_nhc = ekinstate->ekinscalef_nhc[i];
1301 ekind->tcstat[i].ekinscaleh_nhc = ekinstate->ekinscaleh_nhc[i];
1302 ekind->tcstat[i].vscale_nhc = ekinstate->vscale_nhc[i];
1305 copy_mat(ekinstate->ekin_total, ekind->ekin);
1307 ekind->dekindl = ekinstate->dekindl;
1308 ekind->cosacc.mvcos = ekinstate->mvcos;
1309 n = ekinstate->ekin_n;
1314 gmx_bcast(sizeof(n), &n, cr);
1315 for (i = 0; i < n; i++)
1317 gmx_bcast(DIM * DIM * sizeof(ekind->tcstat[i].ekinh[0][0]), ekind->tcstat[i].ekinh[0], cr);
1318 gmx_bcast(DIM * DIM * sizeof(ekind->tcstat[i].ekinf[0][0]), ekind->tcstat[i].ekinf[0], cr);
1319 gmx_bcast(DIM * DIM * sizeof(ekind->tcstat[i].ekinh_old[0][0]),
1320 ekind->tcstat[i].ekinh_old[0], cr);
1322 gmx_bcast(sizeof(ekind->tcstat[i].ekinscalef_nhc), &(ekind->tcstat[i].ekinscalef_nhc), cr);
1323 gmx_bcast(sizeof(ekind->tcstat[i].ekinscaleh_nhc), &(ekind->tcstat[i].ekinscaleh_nhc), cr);
1324 gmx_bcast(sizeof(ekind->tcstat[i].vscale_nhc), &(ekind->tcstat[i].vscale_nhc), cr);
1326 gmx_bcast(DIM * DIM * sizeof(ekind->ekin[0][0]), ekind->ekin[0], cr);
1328 gmx_bcast(sizeof(ekind->dekindl), &ekind->dekindl, cr);
1329 gmx_bcast(sizeof(ekind->cosacc.mvcos), &ekind->cosacc.mvcos, cr);
1333 void update_tcouple(int64_t step,
1334 const t_inputrec* inputrec,
1336 gmx_ekindata_t* ekind,
1337 const t_extmass* MassQ,
1338 const t_mdatoms* md)
1341 // This condition was explicitly checked in previous version, but should have never been satisfied
1342 GMX_ASSERT(!(EI_VV(inputrec->eI)
1343 && (inputrecNvtTrotter(inputrec) || inputrecNptTrotter(inputrec)
1344 || inputrecNphTrotter(inputrec))),
1345 "Temperature coupling was requested with velocity verlet and trotter");
1347 bool doTemperatureCoupling = false;
1349 // For VV temperature coupling parameters are updated on the current
1350 // step, for the others - one step before.
1351 if (inputrec->etc == etcNO)
1353 doTemperatureCoupling = false;
1355 else if (EI_VV(inputrec->eI))
1357 doTemperatureCoupling = do_per_step(step, inputrec->nsttcouple);
1361 doTemperatureCoupling = do_per_step(step + inputrec->nsttcouple - 1, inputrec->nsttcouple);
1364 if (doTemperatureCoupling)
1366 real dttc = inputrec->nsttcouple * inputrec->delta_t;
1368 // TODO: berendsen_tcoupl(...), nosehoover_tcoupl(...) and vrescale_tcoupl(...) update
1369 // temperature coupling parameters, which should be reflected in the name of these
1371 switch (inputrec->etc)
1375 berendsen_tcoupl(inputrec, ekind, dttc, state->therm_integral);
1378 nosehoover_tcoupl(&(inputrec->opts), ekind, dttc, state->nosehoover_xi.data(),
1379 state->nosehoover_vxi.data(), MassQ);
1382 vrescale_tcoupl(inputrec, step, ekind, dttc, state->therm_integral.data());
1385 /* rescale in place here */
1386 if (EI_VV(inputrec->eI))
1388 rescale_velocities(ekind, md, 0, md->homenr, state->v.rvec_array());
1393 // Set the T scaling lambda to 1 to have no scaling
1394 // TODO: Do we have to do it on every non-t-couple step?
1395 for (int i = 0; (i < inputrec->opts.ngtc); i++)
1397 ekind->tcstat[i].lambda = 1.0;
1402 void getThreadAtomRange(int numThreads, int threadIndex, int numAtoms, int* startAtom, int* endAtom)
1404 #if GMX_HAVE_SIMD_UPDATE
1405 constexpr int blockSize = GMX_SIMD_REAL_WIDTH;
1407 constexpr int blockSize = 1;
1409 int numBlocks = (numAtoms + blockSize - 1) / blockSize;
1411 *startAtom = ((numBlocks * threadIndex) / numThreads) * blockSize;
1412 *endAtom = ((numBlocks * (threadIndex + 1)) / numThreads) * blockSize;
1413 if (threadIndex == numThreads - 1)
1415 *endAtom = numAtoms;
1419 void update_pcouple_before_coordinates(FILE* fplog,
1421 const t_inputrec* inputrec,
1423 matrix parrinellorahmanMu,
1427 /* Berendsen P-coupling is completely handled after the coordinate update.
1428 * Trotter P-coupling is handled by separate calls to trotter_update().
1430 if (inputrec->epc == epcPARRINELLORAHMAN
1431 && do_per_step(step + inputrec->nstpcouple - 1, inputrec->nstpcouple))
1433 real dtpc = inputrec->nstpcouple * inputrec->delta_t;
1435 parrinellorahman_pcoupl(fplog, step, inputrec, dtpc, state->pres_prev, state->box,
1436 state->box_rel, state->boxv, M, parrinellorahmanMu, bInitStep);
1440 void constrain_velocities(int64_t step,
1441 real* dvdlambda, /* the contribution to be added to the bonded interactions */
1444 gmx::Constraints* constr,
1456 * APPLY CONSTRAINTS:
1463 /* clear out constraints before applying */
1464 clear_mat(vir_part);
1466 /* Constrain the coordinates upd->xp */
1467 constr->apply(do_log, do_ene, step, 1, 1.0, state->x.rvec_array(), state->v.rvec_array(),
1468 state->v.rvec_array(), state->box, state->lambda[efptBONDED], dvdlambda,
1469 nullptr, bCalcVir ? &vir_con : nullptr, ConstraintVariable::Velocities);
1473 m_add(vir_part, vir_con, vir_part);
1478 void constrain_coordinates(int64_t step,
1479 real* dvdlambda, /* the contribution to be added to the bonded interactions */
1483 gmx::Constraints* constr,
1496 /* clear out constraints before applying */
1497 clear_mat(vir_part);
1499 /* Constrain the coordinates upd->xp */
1500 constr->apply(do_log, do_ene, step, 1, 1.0, state->x.rvec_array(), upd->xp()->rvec_array(),
1501 nullptr, state->box, state->lambda[efptBONDED], dvdlambda,
1502 as_rvec_array(state->v.data()), bCalcVir ? &vir_con : nullptr,
1503 ConstraintVariable::Positions);
1507 m_add(vir_part, vir_con, vir_part);
1512 void update_sd_second_half(int64_t step,
1513 real* dvdlambda, /* the contribution to be added to the bonded interactions */
1514 const t_inputrec* inputrec, /* input record and box stuff */
1515 const t_mdatoms* md,
1517 const t_commrec* cr,
1519 gmx_wallcycle_t wcycle,
1521 gmx::Constraints* constr,
1529 if (inputrec->eI == eiSD1)
1531 int homenr = md->homenr;
1533 /* Cast delta_t from double to real to make the integrators faster.
1534 * The only reason for having delta_t double is to get accurate values
1535 * for t=delta_t*step when step is larger than float precision.
1536 * For integration dt the accuracy of real suffices, since with
1537 * integral += dt*integrand the increment is nearly always (much) smaller
1538 * than the integral (and the integrand has real precision).
1540 real dt = inputrec->delta_t;
1542 wallcycle_start(wcycle, ewcUPDATE);
1544 int nth = gmx_omp_nthreads_get(emntUpdate);
1546 #pragma omp parallel for num_threads(nth) schedule(static)
1547 for (int th = 0; th < nth; th++)
1551 int start_th, end_th;
1552 getThreadAtomRange(nth, th, homenr, &start_th, &end_th);
1554 doSDUpdateGeneral<SDUpdate::FrictionAndNoiseOnly>(
1555 *upd->sd(), start_th, end_th, dt, inputrec->opts.acc, inputrec->opts.nFreeze,
1556 md->invmass, md->ptype, md->cFREEZE, nullptr, md->cTC, state->x.rvec_array(),
1557 upd->xp()->rvec_array(), state->v.rvec_array(), nullptr, step, inputrec->ld_seed,
1558 DOMAINDECOMP(cr) ? cr->dd->globalAtomIndices.data() : nullptr);
1560 GMX_CATCH_ALL_AND_EXIT_WITH_FATAL_ERROR
1562 inc_nrnb(nrnb, eNR_UPDATE, homenr);
1563 wallcycle_stop(wcycle, ewcUPDATE);
1565 /* Constrain the coordinates upd->xp for half a time step */
1566 constr->apply(do_log, do_ene, step, 1, 0.5, state->x.rvec_array(), upd->xp()->rvec_array(),
1567 nullptr, state->box, state->lambda[efptBONDED], dvdlambda,
1568 as_rvec_array(state->v.data()), nullptr, ConstraintVariable::Positions);
1572 void finish_update(const t_inputrec* inputrec, /* input record and box stuff */
1573 const t_mdatoms* md,
1575 const t_graph* graph,
1577 gmx_wallcycle_t wcycle,
1579 const gmx::Constraints* constr)
1581 int homenr = md->homenr;
1583 /* We must always unshift after updating coordinates; if we did not shake
1584 x was shifted in do_force */
1586 /* NOTE Currently we always integrate to a temporary buffer and
1587 * then copy the results back. */
1589 wallcycle_start_nocount(wcycle, ewcUPDATE);
1591 if (md->cFREEZE != nullptr && constr != nullptr)
1593 /* If we have atoms that are frozen along some, but not all
1594 * dimensions, then any constraints will have moved them also along
1595 * the frozen dimensions. To freeze such degrees of freedom
1596 * we copy them back here to later copy them forward. It would
1597 * be more elegant and slightly more efficient to copies zero
1598 * times instead of twice, but the graph case below prevents this.
1600 const ivec* nFreeze = inputrec->opts.nFreeze;
1601 bool partialFreezeAndConstraints = false;
1602 for (int g = 0; g < inputrec->opts.ngfrz; g++)
1604 int numFreezeDim = nFreeze[g][XX] + nFreeze[g][YY] + nFreeze[g][ZZ];
1605 if (numFreezeDim > 0 && numFreezeDim < 3)
1607 partialFreezeAndConstraints = true;
1610 if (partialFreezeAndConstraints)
1612 auto xp = makeArrayRef(*upd->xp()).subArray(0, homenr);
1613 auto x = makeConstArrayRef(state->x).subArray(0, homenr);
1614 for (int i = 0; i < homenr; i++)
1616 int g = md->cFREEZE[i];
1618 for (int d = 0; d < DIM; d++)
1629 if (graph && (graph->nnodes > 0))
1631 unshift_x(graph, state->box, state->x.rvec_array(), upd->xp()->rvec_array());
1632 if (TRICLINIC(state->box))
1634 inc_nrnb(nrnb, eNR_SHIFTX, 2 * graph->nnodes);
1638 inc_nrnb(nrnb, eNR_SHIFTX, graph->nnodes);
1643 auto xp = makeConstArrayRef(*upd->xp()).subArray(0, homenr);
1644 auto x = makeArrayRef(state->x).subArray(0, homenr);
1647 int gmx_unused nth = gmx_omp_nthreads_get(emntUpdate);
1648 #pragma omp parallel for num_threads(nth) schedule(static)
1649 for (int i = 0; i < homenr; i++)
1651 // Trivial statement, does not throw
1655 wallcycle_stop(wcycle, ewcUPDATE);
1657 /* ############# END the update of velocities and positions ######### */
1660 void update_pcouple_after_coordinates(FILE* fplog,
1662 const t_inputrec* inputrec,
1663 const t_mdatoms* md,
1664 const matrix pressure,
1665 const matrix forceVirial,
1666 const matrix constraintVirial,
1667 matrix pressureCouplingMu,
1671 const bool scaleCoordinates)
1674 int homenr = md->homenr;
1676 /* Cast to real for faster code, no loss in precision (see comment above) */
1677 real dt = inputrec->delta_t;
1680 /* now update boxes */
1681 switch (inputrec->epc)
1683 case (epcNO): break;
1684 case (epcBERENDSEN):
1685 if (do_per_step(step, inputrec->nstpcouple))
1687 real dtpc = inputrec->nstpcouple * dt;
1688 berendsen_pcoupl(fplog, step, inputrec, dtpc, pressure, state->box, forceVirial,
1689 constraintVirial, pressureCouplingMu, &state->baros_integral);
1690 berendsen_pscale(inputrec, pressureCouplingMu, state->box, state->box_rel, start,
1691 homenr, state->x.rvec_array(), md->cFREEZE, nrnb, scaleCoordinates);
1694 case (epcPARRINELLORAHMAN):
1695 if (do_per_step(step + inputrec->nstpcouple - 1, inputrec->nstpcouple))
1697 /* The box velocities were updated in do_pr_pcoupl,
1698 * but we dont change the box vectors until we get here
1699 * since we need to be able to shift/unshift above.
1701 real dtpc = inputrec->nstpcouple * dt;
1702 for (int i = 0; i < DIM; i++)
1704 for (int m = 0; m <= i; m++)
1706 state->box[i][m] += dtpc * state->boxv[i][m];
1709 preserve_box_shape(inputrec, state->box_rel, state->box);
1711 /* Scale the coordinates */
1712 if (scaleCoordinates)
1714 auto x = state->x.rvec_array();
1715 for (int n = start; n < start + homenr; n++)
1717 tmvmul_ur0(pressureCouplingMu, x[n], x[n]);
1723 switch (inputrec->epct)
1725 case (epctISOTROPIC):
1726 /* DIM * eta = ln V. so DIM*eta_new = DIM*eta_old + DIM*dt*veta =>
1727 ln V_new = ln V_old + 3*dt*veta => V_new = V_old*exp(3*dt*veta) =>
1728 Side length scales as exp(veta*dt) */
1730 msmul(state->box, std::exp(state->veta * dt), state->box);
1732 /* Relate veta to boxv. veta = d(eta)/dT = (1/DIM)*1/V dV/dT.
1733 o If we assume isotropic scaling, and box length scaling
1734 factor L, then V = L^DIM (det(M)). So dV/dt = DIM
1735 L^(DIM-1) dL/dt det(M), and veta = (1/L) dL/dt. The
1736 determinant of B is L^DIM det(M), and the determinant
1737 of dB/dt is (dL/dT)^DIM det (M). veta will be
1738 (det(dB/dT)/det(B))^(1/3). Then since M =
1739 B_new*(vol_new)^(1/3), dB/dT_new = (veta_new)*B(new). */
1741 msmul(state->box, state->veta, state->boxv);
1751 auto localX = makeArrayRef(state->x).subArray(start, homenr);
1752 upd->deform()->apply(localX, state->box, step);
1756 void update_coords(int64_t step,
1757 const t_inputrec* inputrec, /* input record and box stuff */
1758 const t_mdatoms* md,
1760 gmx::ArrayRefWithPadding<const gmx::RVec> f,
1761 const t_fcdata* fcd,
1762 const gmx_ekindata_t* ekind,
1766 const t_commrec* cr, /* these shouldn't be here -- need to think about it */
1767 const gmx::Constraints* constr)
1769 gmx_bool bDoConstr = (nullptr != constr);
1771 /* Running the velocity half does nothing except for velocity verlet */
1772 if ((UpdatePart == etrtVELOCITY1 || UpdatePart == etrtVELOCITY2) && !EI_VV(inputrec->eI))
1774 gmx_incons("update_coords called for velocity without VV integrator");
1777 int homenr = md->homenr;
1779 /* Cast to real for faster code, no loss in precision (see comment above) */
1780 real dt = inputrec->delta_t;
1782 /* We need to update the NMR restraint history when time averaging is used */
1783 if (state->flags & (1 << estDISRE_RM3TAV))
1785 update_disres_history(fcd, &state->hist);
1787 if (state->flags & (1 << estORIRE_DTAV))
1789 update_orires_history(fcd, &state->hist);
1792 /* ############# START The update of velocities and positions ######### */
1793 int nth = gmx_omp_nthreads_get(emntUpdate);
1795 #pragma omp parallel for num_threads(nth) schedule(static)
1796 for (int th = 0; th < nth; th++)
1800 int start_th, end_th;
1801 getThreadAtomRange(nth, th, homenr, &start_th, &end_th);
1803 const rvec* x_rvec = state->x.rvec_array();
1804 rvec* xp_rvec = upd->xp()->rvec_array();
1805 rvec* v_rvec = state->v.rvec_array();
1806 const rvec* f_rvec = as_rvec_array(f.unpaddedArrayRef().data());
1808 switch (inputrec->eI)
1811 do_update_md(start_th, end_th, step, dt, inputrec, md, ekind, state->box,
1812 x_rvec, xp_rvec, v_rvec, f_rvec, state->nosehoover_vxi.data(), M);
1817 // With constraints, the SD update is done in 2 parts
1818 doSDUpdateGeneral<SDUpdate::ForcesOnly>(
1819 *upd->sd(), start_th, end_th, dt, inputrec->opts.acc, inputrec->opts.nFreeze,
1820 md->invmass, md->ptype, md->cFREEZE, md->cACC, nullptr, x_rvec,
1821 xp_rvec, v_rvec, f_rvec, step, inputrec->ld_seed, nullptr);
1825 doSDUpdateGeneral<SDUpdate::Combined>(
1826 *upd->sd(), start_th, end_th, dt, inputrec->opts.acc,
1827 inputrec->opts.nFreeze, md->invmass, md->ptype, md->cFREEZE, md->cACC,
1828 md->cTC, x_rvec, xp_rvec, v_rvec, f_rvec, step, inputrec->ld_seed,
1829 DOMAINDECOMP(cr) ? cr->dd->globalAtomIndices.data() : nullptr);
1833 do_update_bd(start_th, end_th, dt, inputrec->opts.nFreeze, md->invmass,
1834 md->ptype, md->cFREEZE, md->cTC, x_rvec, xp_rvec, v_rvec, f_rvec,
1835 inputrec->bd_fric, upd->sd()->bd_rf.data(), step, inputrec->ld_seed,
1836 DOMAINDECOMP(cr) ? cr->dd->globalAtomIndices.data() : nullptr);
1841 gmx_bool bExtended = (inputrec->etc == etcNOSEHOOVER || inputrec->epc == epcPARRINELLORAHMAN
1842 || inputrec->epc == epcMTTK);
1844 /* assuming barostat coupled to group 0 */
1845 real alpha = 1.0 + DIM / static_cast<real>(inputrec->opts.nrdf[0]);
1850 do_update_vv_vel(start_th, end_th, dt, inputrec->opts.acc,
1851 inputrec->opts.nFreeze, md->invmass, md->ptype, md->cFREEZE,
1852 md->cACC, v_rvec, f_rvec, bExtended, state->veta, alpha);
1855 do_update_vv_pos(start_th, end_th, dt, inputrec->opts.nFreeze, md->ptype,
1856 md->cFREEZE, x_rvec, xp_rvec, v_rvec, bExtended, state->veta);
1861 default: gmx_fatal(FARGS, "Don't know how to update coordinates");
1864 GMX_CATCH_ALL_AND_EXIT_WITH_FATAL_ERROR
1868 extern gmx_bool update_randomize_velocities(const t_inputrec* ir,
1870 const t_commrec* cr,
1871 const t_mdatoms* md,
1872 gmx::ArrayRef<gmx::RVec> v,
1874 const gmx::Constraints* constr)
1877 real rate = (ir->delta_t) / ir->opts.tau_t[0];
1879 if (ir->etc == etcANDERSEN && constr != nullptr)
1881 /* Currently, Andersen thermostat does not support constrained
1882 systems. Functionality exists in the andersen_tcoupl
1883 function in GROMACS 4.5.7 to allow this combination. That
1884 code could be ported to the current random-number
1885 generation approach, but has not yet been done because of
1886 lack of time and resources. */
1888 "Normal Andersen is currently not supported with constraints, use massive "
1889 "Andersen instead");
1892 /* proceed with andersen if 1) it's fixed probability per
1893 particle andersen or 2) it's massive andersen and it's tau_t/dt */
1894 if ((ir->etc == etcANDERSEN) || do_per_step(step, roundToInt(1.0 / rate)))
1896 andersen_tcoupl(ir, step, cr, md, v, rate, upd->sd()->randomize_group, upd->sd()->boltzfac);