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/state.h"
71 #include "gromacs/pbcutil/boxutilities.h"
72 #include "gromacs/pbcutil/mshift.h"
73 #include "gromacs/pbcutil/pbc.h"
74 #include "gromacs/pulling/pull.h"
75 #include "gromacs/random/tabulatednormaldistribution.h"
76 #include "gromacs/random/threefry.h"
77 #include "gromacs/simd/simd.h"
78 #include "gromacs/timing/wallcycle.h"
79 #include "gromacs/topology/atoms.h"
80 #include "gromacs/utility/exceptions.h"
81 #include "gromacs/utility/fatalerror.h"
82 #include "gromacs/utility/futil.h"
83 #include "gromacs/utility/gmxassert.h"
84 #include "gromacs/utility/gmxomp.h"
85 #include "gromacs/utility/smalloc.h"
87 using namespace gmx; // TODO: Remove when this file is moved into gmx namespace
102 std::vector<real> bd_rf;
104 std::vector<gmx_sd_const_t> sdc;
105 std::vector<gmx_sd_sigma_t> sdsig;
106 /* andersen temperature control stuff */
107 std::vector<bool> randomize_group;
108 std::vector<real> boltzfac;
110 explicit gmx_stochd_t(const t_inputrec* ir);
113 //! pImpled implementation for Update
118 Impl(const t_inputrec* ir, BoxDeformation* boxDeformation);
121 //! stochastic dynamics struct
122 std::unique_ptr<gmx_stochd_t> sd;
123 //! xprime for constraint algorithms
124 PaddedVector<RVec> xp;
125 //! Box deformation handler (or nullptr if inactive).
126 BoxDeformation* deform = nullptr;
129 Update::Update(const t_inputrec* ir, BoxDeformation* boxDeformation) :
130 impl_(new Impl(ir, boxDeformation)){};
134 gmx_stochd_t* Update::sd() const
136 return impl_->sd.get();
139 PaddedVector<RVec>* Update::xp()
144 BoxDeformation* Update::deform() const
146 return impl_->deform;
149 /*! \brief Sets the velocities of virtual sites to zero */
150 static void clearVsiteVelocities(int start, int nrend, const unsigned short* particleType, rvec* gmx_restrict v)
152 for (int a = start; a < nrend; a++)
154 if (particleType[a] == eptVSite)
161 /*! \brief Sets the number of different temperature coupling values */
162 enum class NumTempScaleValues
164 single, //!< Single T-scaling value (either one group or all values =1)
165 multiple //!< Multiple T-scaling values, need to use T-group indices
168 /*! \brief Sets if to apply no or diagonal Parrinello-Rahman pressure scaling
170 * Note that this enum is only used in updateMDLeapfrogSimple(), which does
171 * not handle fully anistropic Parrinello-Rahman scaling, so we only have
172 * options \p no and \p diagonal here and no anistropic option.
174 enum class ApplyParrinelloRahmanVScaling
176 no, //!< Do not apply velocity scaling (not a PR-coupling run or step)
177 diagonal //!< Apply velocity scaling using a diagonal matrix
180 /*! \brief Integrate using leap-frog with T-scaling and optionally diagonal Parrinello-Rahman p-coupling
182 * \tparam numTempScaleValues The number of different T-couple values
183 * \tparam applyPRVScaling Apply Parrinello-Rahman velocity scaling
184 * \param[in] start Index of first atom to update
185 * \param[in] nrend Last atom to update: \p nrend - 1
186 * \param[in] dt The time step
187 * \param[in] dtPressureCouple Time step for pressure coupling
188 * \param[in] invMassPerDim 1/mass per atom and dimension
189 * \param[in] tcstat Temperature coupling information
190 * \param[in] cTC T-coupling group index per atom
191 * \param[in] pRVScaleMatrixDiagonal Parrinello-Rahman v-scale matrix diagonal
192 * \param[in] x Input coordinates
193 * \param[out] xprime Updated coordinates
194 * \param[inout] v Velocities
195 * \param[in] f Forces
197 * We expect this template to get good SIMD acceleration by most compilers,
198 * unlike the more complex general template.
199 * Note that we might get even better SIMD acceleration when we introduce
200 * aligned (and padded) memory, possibly with some hints for the compilers.
202 template<NumTempScaleValues numTempScaleValues, ApplyParrinelloRahmanVScaling applyPRVScaling>
203 static void updateMDLeapfrogSimple(int start,
206 real dtPressureCouple,
207 const rvec* gmx_restrict invMassPerDim,
208 gmx::ArrayRef<const t_grp_tcstat> tcstat,
209 const unsigned short* cTC,
210 const rvec pRVScaleMatrixDiagonal,
211 const rvec* gmx_restrict x,
212 rvec* gmx_restrict xprime,
213 rvec* gmx_restrict v,
214 const rvec* gmx_restrict f)
218 if (numTempScaleValues == NumTempScaleValues::single)
220 lambdaGroup = tcstat[0].lambda;
223 for (int a = start; a < nrend; a++)
225 if (numTempScaleValues == NumTempScaleValues::multiple)
227 lambdaGroup = tcstat[cTC[a]].lambda;
230 for (int d = 0; d < DIM; d++)
232 /* Note that using rvec invMassPerDim results in more efficient
233 * SIMD code, but this increases the cache pressure.
234 * For large systems with PME on the CPU this slows down the
235 * (then already slow) update by 20%. If all data remains in cache,
236 * using rvec is much faster.
238 real vNew = lambdaGroup * v[a][d] + f[a][d] * invMassPerDim[a][d] * dt;
240 if (applyPRVScaling == ApplyParrinelloRahmanVScaling::diagonal)
242 vNew -= dtPressureCouple * pRVScaleMatrixDiagonal[d] * v[a][d];
245 xprime[a][d] = x[a][d] + vNew * dt;
250 #if GMX_SIMD && GMX_SIMD_HAVE_REAL
251 # define GMX_HAVE_SIMD_UPDATE 1
253 # define GMX_HAVE_SIMD_UPDATE 0
256 #if GMX_HAVE_SIMD_UPDATE
258 /*! \brief Load (aligned) the contents of GMX_SIMD_REAL_WIDTH rvec elements sequentially into 3 SIMD registers
260 * The loaded output is:
261 * \p r0: { r[index][XX], r[index][YY], ... }
263 * \p r2: { ..., r[index+GMX_SIMD_REAL_WIDTH-1][YY], r[index+GMX_SIMD_REAL_WIDTH-1][ZZ] }
265 * \param[in] r Real to an rvec array, has to be aligned to SIMD register width
266 * \param[in] index Index of the first rvec triplet of reals to load
267 * \param[out] r0 Pointer to first SIMD register
268 * \param[out] r1 Pointer to second SIMD register
269 * \param[out] r2 Pointer to third SIMD register
271 static inline void simdLoadRvecs(const rvec* r, int index, SimdReal* r0, SimdReal* r1, SimdReal* r2)
273 const real* realPtr = r[index];
275 GMX_ASSERT(isSimdAligned(realPtr), "Pointer should be SIMD aligned");
277 *r0 = simdLoad(realPtr + 0 * GMX_SIMD_REAL_WIDTH);
278 *r1 = simdLoad(realPtr + 1 * GMX_SIMD_REAL_WIDTH);
279 *r2 = simdLoad(realPtr + 2 * GMX_SIMD_REAL_WIDTH);
282 /*! \brief Store (aligned) 3 SIMD registers sequentially to GMX_SIMD_REAL_WIDTH rvec elements
284 * The stored output is:
285 * \p r[index] = { { r0[0], r0[1], ... }
287 * \p r[index+GMX_SIMD_REAL_WIDTH-1] = { ... , r2[GMX_SIMD_REAL_WIDTH-2], r2[GMX_SIMD_REAL_WIDTH-1] }
289 * \param[out] r Pointer to an rvec array, has to be aligned to SIMD register width
290 * \param[in] index Index of the first rvec triplet of reals to store to
291 * \param[in] r0 First SIMD register
292 * \param[in] r1 Second SIMD register
293 * \param[in] r2 Third SIMD register
295 static inline void simdStoreRvecs(rvec* r, int index, SimdReal r0, SimdReal r1, SimdReal r2)
297 real* realPtr = r[index];
299 GMX_ASSERT(isSimdAligned(realPtr), "Pointer should be SIMD aligned");
301 store(realPtr + 0 * GMX_SIMD_REAL_WIDTH, r0);
302 store(realPtr + 1 * GMX_SIMD_REAL_WIDTH, r1);
303 store(realPtr + 2 * GMX_SIMD_REAL_WIDTH, r2);
306 /*! \brief Integrate using leap-frog with single group T-scaling and SIMD
308 * \param[in] start Index of first atom to update
309 * \param[in] nrend Last atom to update: \p nrend - 1
310 * \param[in] dt The time step
311 * \param[in] invMass 1/mass per atom
312 * \param[in] tcstat Temperature coupling information
313 * \param[in] x Input coordinates
314 * \param[out] xprime Updated coordinates
315 * \param[inout] v Velocities
316 * \param[in] f Forces
318 static void updateMDLeapfrogSimpleSimd(int start,
321 const real* gmx_restrict invMass,
322 gmx::ArrayRef<const t_grp_tcstat> tcstat,
323 const rvec* gmx_restrict x,
324 rvec* gmx_restrict xprime,
325 rvec* gmx_restrict v,
326 const rvec* gmx_restrict f)
328 SimdReal timestep(dt);
329 SimdReal lambdaSystem(tcstat[0].lambda);
331 /* We declare variables here, since code is often slower when declaring them inside the loop */
333 /* Note: We should implement a proper PaddedVector, so we don't need this check */
334 GMX_ASSERT(isSimdAligned(invMass), "invMass should be aligned");
336 for (int a = start; a < nrend; a += GMX_SIMD_REAL_WIDTH)
338 SimdReal invMass0, invMass1, invMass2;
339 expandScalarsToTriplets(simdLoad(invMass + a), &invMass0, &invMass1, &invMass2);
343 simdLoadRvecs(v, a, &v0, &v1, &v2);
344 simdLoadRvecs(f, a, &f0, &f1, &f2);
346 v0 = fma(f0 * invMass0, timestep, lambdaSystem * v0);
347 v1 = fma(f1 * invMass1, timestep, lambdaSystem * v1);
348 v2 = fma(f2 * invMass2, timestep, lambdaSystem * v2);
350 simdStoreRvecs(v, a, v0, v1, v2);
353 simdLoadRvecs(x, a, &x0, &x1, &x2);
355 SimdReal xprime0 = fma(v0, timestep, x0);
356 SimdReal xprime1 = fma(v1, timestep, x1);
357 SimdReal xprime2 = fma(v2, timestep, x2);
359 simdStoreRvecs(xprime, a, xprime0, xprime1, xprime2);
363 #endif // GMX_HAVE_SIMD_UPDATE
365 /*! \brief Sets the NEMD acceleration type */
366 enum class AccelerationType
373 /*! \brief Integrate using leap-frog with support for everything
375 * \tparam accelerationType Type of NEMD acceleration
376 * \param[in] start Index of first atom to update
377 * \param[in] nrend Last atom to update: \p nrend - 1
378 * \param[in] doNoseHoover If to apply Nose-Hoover T-coupling
379 * \param[in] dt The time step
380 * \param[in] dtPressureCouple Time step for pressure coupling, is 0 when no pressure
381 * coupling should be applied at this step \param[in] ir The input parameter
382 * record \param[in] md Atom properties \param[in] ekind Kinetic energy
383 * data \param[in] box The box dimensions \param[in] x Input coordinates \param[out]
384 * xprime Updated coordinates \param[inout] v Velocities \param[in]
385 * f Forces \param[in] nh_vxi Nose-Hoover velocity scaling
386 * factors \param[in] M Parrinello-Rahman scaling matrix
388 template<AccelerationType accelerationType>
389 static void updateMDLeapfrogGeneral(int start,
393 real dtPressureCouple,
394 const t_inputrec* ir,
396 const gmx_ekindata_t* ekind,
398 const rvec* gmx_restrict x,
399 rvec* gmx_restrict xprime,
400 rvec* gmx_restrict v,
401 const rvec* gmx_restrict f,
402 const double* gmx_restrict nh_vxi,
405 /* This is a version of the leap-frog integrator that supports
406 * all combinations of T-coupling, P-coupling and NEMD.
407 * Nose-Hoover uses the reversible leap-frog integrator from
408 * Holian et al. Phys Rev E 52(3) : 2338, 1995
411 gmx::ArrayRef<const t_grp_tcstat> tcstat = ekind->tcstat;
412 gmx::ArrayRef<const t_grp_acc> grpstat = ekind->grpstat;
413 const unsigned short* cTC = md->cTC;
414 const unsigned short* cACC = md->cACC;
415 const rvec* accel = ir->opts.acc;
417 const rvec* gmx_restrict invMassPerDim = md->invMassPerDim;
419 /* Initialize group values, changed later when multiple groups are used */
424 real omega_Z = 2 * static_cast<real>(M_PI) / box[ZZ][ZZ];
426 for (int n = start; n < nrend; n++)
432 real lg = tcstat[gt].lambda;
435 real cosineZ, vCosine;
436 #ifdef __INTEL_COMPILER
437 # pragma warning(disable : 280)
439 switch (accelerationType)
441 case AccelerationType::none: copy_rvec(v[n], vRel); break;
442 case AccelerationType::group:
447 /* Avoid scaling the group velocity */
448 rvec_sub(v[n], grpstat[ga].u, vRel);
450 case AccelerationType::cosine:
451 cosineZ = std::cos(x[n][ZZ] * omega_Z);
452 vCosine = cosineZ * ekind->cosacc.vcos;
453 /* Avoid scaling the cosine profile velocity */
454 copy_rvec(v[n], vRel);
461 /* Here we account for multiple time stepping, by increasing
462 * the Nose-Hoover correction by nsttcouple
464 factorNH = 0.5 * ir->nsttcouple * dt * nh_vxi[gt];
467 for (int d = 0; d < DIM; d++)
469 real vNew = (lg * vRel[d]
470 + (f[n][d] * invMassPerDim[n][d] * dt - factorNH * vRel[d]
471 - dtPressureCouple * iprod(M[d], vRel)))
473 switch (accelerationType)
475 case AccelerationType::none: break;
476 case AccelerationType::group:
477 /* Add back the mean velocity and apply acceleration */
478 vNew += grpstat[ga].u[d] + accel[ga][d] * dt;
480 case AccelerationType::cosine:
483 /* Add back the mean velocity and apply acceleration */
484 vNew += vCosine + cosineZ * ekind->cosacc.cos_accel * dt;
489 xprime[n][d] = x[n][d] + vNew * dt;
494 /*! \brief Handles the Leap-frog MD x and v integration */
495 static void do_update_md(int start,
499 const t_inputrec* ir,
501 const gmx_ekindata_t* ekind,
503 const rvec* gmx_restrict x,
504 rvec* gmx_restrict xprime,
505 rvec* gmx_restrict v,
506 const rvec* gmx_restrict f,
507 const double* gmx_restrict nh_vxi,
510 GMX_ASSERT(nrend == start || xprime != x,
511 "For SIMD optimization certain compilers need to have xprime != x");
512 GMX_ASSERT(ir->eI == eiMD,
513 "Leap-frog integrator was called while another integrator was requested");
515 /* Note: Berendsen pressure scaling is handled after do_update_md() */
516 bool doTempCouple = (ir->etc != etcNO && do_per_step(step + ir->nsttcouple - 1, ir->nsttcouple));
517 bool doNoseHoover = (ir->etc == etcNOSEHOOVER && doTempCouple);
518 bool doParrinelloRahman =
519 (ir->epc == epcPARRINELLORAHMAN && do_per_step(step + ir->nstpcouple - 1, ir->nstpcouple));
520 bool doPROffDiagonal = (doParrinelloRahman && (M[YY][XX] != 0 || M[ZZ][XX] != 0 || M[ZZ][YY] != 0));
522 real dtPressureCouple = (doParrinelloRahman ? ir->nstpcouple * dt : 0);
524 /* NEMD (also cosine) acceleration is applied in updateMDLeapFrogGeneral */
525 bool doAcceleration = (ekind->bNEMD || ekind->cosacc.cos_accel != 0);
527 if (doNoseHoover || doPROffDiagonal || doAcceleration)
530 if (!doParrinelloRahman)
532 /* We should not apply PR scaling at this step */
542 updateMDLeapfrogGeneral<AccelerationType::none>(start, nrend, doNoseHoover, dt,
543 dtPressureCouple, ir, md, ekind, box, x,
544 xprime, v, f, nh_vxi, stepM);
546 else if (ekind->bNEMD)
548 updateMDLeapfrogGeneral<AccelerationType::group>(start, nrend, doNoseHoover, dt,
549 dtPressureCouple, ir, md, ekind, box,
550 x, xprime, v, f, nh_vxi, stepM);
554 updateMDLeapfrogGeneral<AccelerationType::cosine>(start, nrend, doNoseHoover, dt,
555 dtPressureCouple, ir, md, ekind, box,
556 x, xprime, v, f, nh_vxi, stepM);
561 /* Use a simple and thus more efficient integration loop. */
562 /* The simple loop does not check for particle type (so it can
563 * be vectorized), which means we need to clear the velocities
564 * of virtual sites in advance, when present. Note that vsite
565 * velocities are computed after update and constraints from
566 * their displacement.
570 /* Note: The overhead of this loop is completely neligible */
571 clearVsiteVelocities(start, nrend, md->ptype, v);
574 /* We determine if we have a single T-coupling lambda value for all
575 * atoms. That allows for better SIMD acceleration in the template.
576 * If we do not do temperature coupling (in the run or this step),
577 * all scaling values are 1, so we effectively have a single value.
579 bool haveSingleTempScaleValue = (!doTempCouple || ekind->ngtc == 1);
581 /* Extract some pointers needed by all cases */
582 const unsigned short* cTC = md->cTC;
583 gmx::ArrayRef<const t_grp_tcstat> tcstat = ekind->tcstat;
584 const rvec* invMassPerDim = md->invMassPerDim;
586 if (doParrinelloRahman)
588 GMX_ASSERT(!doPROffDiagonal,
589 "updateMDLeapfrogSimple only support diagonal Parrinello-Rahman scaling "
593 for (int d = 0; d < DIM; d++)
598 if (haveSingleTempScaleValue)
600 updateMDLeapfrogSimple<NumTempScaleValues::single, ApplyParrinelloRahmanVScaling::diagonal>(
601 start, nrend, dt, dtPressureCouple, invMassPerDim, tcstat, cTC, diagM, x,
606 updateMDLeapfrogSimple<NumTempScaleValues::multiple, ApplyParrinelloRahmanVScaling::diagonal>(
607 start, nrend, dt, dtPressureCouple, invMassPerDim, tcstat, cTC, diagM, x,
613 if (haveSingleTempScaleValue)
615 /* Note that modern compilers are pretty good at vectorizing
616 * updateMDLeapfrogSimple(). But the SIMD version will still
617 * be faster because invMass lowers the cache pressure
618 * compared to invMassPerDim.
620 #if GMX_HAVE_SIMD_UPDATE
621 /* Check if we can use invmass instead of invMassPerDim */
622 if (!md->havePartiallyFrozenAtoms)
624 updateMDLeapfrogSimpleSimd(start, nrend, dt, md->invmass, tcstat, x, xprime, v, f);
629 updateMDLeapfrogSimple<NumTempScaleValues::single, ApplyParrinelloRahmanVScaling::no>(
630 start, nrend, dt, dtPressureCouple, invMassPerDim, tcstat, cTC, nullptr,
636 updateMDLeapfrogSimple<NumTempScaleValues::multiple, ApplyParrinelloRahmanVScaling::no>(
637 start, nrend, dt, dtPressureCouple, invMassPerDim, tcstat, cTC, nullptr, x,
644 static void do_update_vv_vel(int start,
648 const ivec nFreeze[],
649 const real invmass[],
650 const unsigned short ptype[],
651 const unsigned short cFREEZE[],
652 const unsigned short cACC[],
665 g = 0.25 * dt * veta * alpha;
667 mv2 = gmx::series_sinhx(g);
674 for (n = start; n < nrend; n++)
676 real w_dt = invmass[n] * dt;
686 for (d = 0; d < DIM; d++)
688 if ((ptype[n] != eptVSite) && (ptype[n] != eptShell) && !nFreeze[gf][d])
690 v[n][d] = mv1 * (mv1 * v[n][d] + 0.5 * (w_dt * mv2 * f[n][d])) + 0.5 * accel[ga][d] * dt;
698 } /* do_update_vv_vel */
700 static void do_update_vv_pos(int start,
703 const ivec nFreeze[],
704 const unsigned short ptype[],
705 const unsigned short cFREEZE[],
716 /* Would it make more sense if Parrinello-Rahman was put here? */
721 mr2 = gmx::series_sinhx(g);
729 for (n = start; n < nrend; n++)
737 for (d = 0; d < DIM; d++)
739 if ((ptype[n] != eptVSite) && (ptype[n] != eptShell) && !nFreeze[gf][d])
741 xprime[n][d] = mr1 * (mr1 * x[n][d] + mr2 * dt * v[n][d]);
745 xprime[n][d] = x[n][d];
749 } /* do_update_vv_pos */
751 gmx_stochd_t::gmx_stochd_t(const t_inputrec* ir)
753 const t_grpopts* opts = &ir->opts;
754 const int ngtc = opts->ngtc;
760 else if (EI_SD(ir->eI))
765 for (int gt = 0; gt < ngtc; gt++)
767 if (opts->tau_t[gt] > 0)
769 sdc[gt].em = std::exp(-ir->delta_t / opts->tau_t[gt]);
773 /* No friction and noise on this group */
778 else if (ETC_ANDERSEN(ir->etc))
780 randomize_group.resize(ngtc);
781 boltzfac.resize(ngtc);
783 /* for now, assume that all groups, if randomized, are randomized at the same rate, i.e. tau_t is the same. */
784 /* since constraint groups don't necessarily match up with temperature groups! This is checked in readir.c */
786 for (int gt = 0; gt < ngtc; gt++)
788 real reft = std::max<real>(0, opts->ref_t[gt]);
789 if ((opts->tau_t[gt] > 0)
790 && (reft > 0)) /* tau_t or ref_t = 0 means that no randomization is done */
792 randomize_group[gt] = true;
793 boltzfac[gt] = BOLTZ * opts->ref_t[gt];
797 randomize_group[gt] = false;
803 void update_temperature_constants(gmx_stochd_t* sd, const t_inputrec* ir)
807 if (ir->bd_fric != 0)
809 for (int gt = 0; gt < ir->opts.ngtc; gt++)
811 sd->bd_rf[gt] = std::sqrt(2.0 * BOLTZ * ir->opts.ref_t[gt] / (ir->bd_fric * ir->delta_t));
816 for (int gt = 0; gt < ir->opts.ngtc; gt++)
818 sd->bd_rf[gt] = std::sqrt(2.0 * BOLTZ * ir->opts.ref_t[gt]);
824 for (int gt = 0; gt < ir->opts.ngtc; gt++)
826 real kT = BOLTZ * ir->opts.ref_t[gt];
827 /* The mass is accounted for later, since this differs per atom */
828 sd->sdsig[gt].V = std::sqrt(kT * (1 - sd->sdc[gt].em * sd->sdc[gt].em));
833 Update::Impl::Impl(const t_inputrec* ir, BoxDeformation* boxDeformation)
835 sd = std::make_unique<gmx_stochd_t>(ir);
836 update_temperature_constants(sd.get(), ir);
837 xp.resizeWithPadding(0);
838 deform = boxDeformation;
841 void Update::setNumAtoms(int nAtoms)
844 impl_->xp.resizeWithPadding(nAtoms);
847 /*! \brief Sets the SD update type */
848 enum class SDUpdate : int
851 FrictionAndNoiseOnly,
855 /*! \brief SD integrator update
857 * Two phases are required in the general case of a constrained
858 * update, the first phase from the contribution of forces, before
859 * applying constraints, and then a second phase applying the friction
860 * and noise, and then further constraining. For details, see
863 * Without constraints, the two phases can be combined, for
866 * Thus three instantiations of this templated function will be made,
867 * two with only one contribution, and one with both contributions. */
868 template<SDUpdate updateType>
869 static void doSDUpdateGeneral(const gmx_stochd_t& sd,
874 const ivec nFreeze[],
875 const real invmass[],
876 const unsigned short ptype[],
877 const unsigned short cFREEZE[],
878 const unsigned short cACC[],
879 const unsigned short cTC[],
888 // cTC, cACC and cFREEZE can be nullptr any time, but various
889 // instantiations do not make sense with particular pointer
891 if (updateType == SDUpdate::ForcesOnly)
893 GMX_ASSERT(f != nullptr, "SD update with only forces requires forces");
894 GMX_ASSERT(cTC == nullptr, "SD update with only forces cannot handle temperature groups");
896 if (updateType == SDUpdate::FrictionAndNoiseOnly)
898 GMX_ASSERT(f == nullptr, "SD update with only noise cannot handle forces");
899 GMX_ASSERT(cACC == nullptr, "SD update with only noise cannot handle acceleration groups");
901 if (updateType == SDUpdate::Combined)
903 GMX_ASSERT(f != nullptr, "SD update with forces and noise requires forces");
906 // Even 0 bits internal counter gives 2x64 ints (more than enough for three table lookups)
907 gmx::ThreeFry2x64<0> rng(seed, gmx::RandomDomain::UpdateCoordinates);
908 gmx::TabulatedNormalDistribution<real, 14> dist;
910 for (int n = start; n < nrend; n++)
912 int globalAtomIndex = gatindex ? gatindex[n] : n;
913 rng.restart(step, globalAtomIndex);
916 real inverseMass = invmass[n];
917 real invsqrtMass = std::sqrt(inverseMass);
919 int freezeGroup = cFREEZE ? cFREEZE[n] : 0;
920 int accelerationGroup = cACC ? cACC[n] : 0;
921 int temperatureGroup = cTC ? cTC[n] : 0;
923 for (int d = 0; d < DIM; d++)
925 if ((ptype[n] != eptVSite) && (ptype[n] != eptShell) && !nFreeze[freezeGroup][d])
927 if (updateType == SDUpdate::ForcesOnly)
929 real vn = v[n][d] + (inverseMass * f[n][d] + accel[accelerationGroup][d]) * dt;
931 // Simple position update.
932 xprime[n][d] = x[n][d] + v[n][d] * dt;
934 else if (updateType == SDUpdate::FrictionAndNoiseOnly)
937 v[n][d] = (vn * sd.sdc[temperatureGroup].em
938 + invsqrtMass * sd.sdsig[temperatureGroup].V * dist(rng));
939 // The previous phase already updated the
940 // positions with a full v*dt term that must
941 // now be half removed.
942 xprime[n][d] = xprime[n][d] + 0.5 * (v[n][d] - vn) * dt;
946 real vn = v[n][d] + (inverseMass * f[n][d] + accel[accelerationGroup][d]) * dt;
947 v[n][d] = (vn * sd.sdc[temperatureGroup].em
948 + invsqrtMass * sd.sdsig[temperatureGroup].V * dist(rng));
949 // Here we include half of the friction+noise
950 // update of v into the position update.
951 xprime[n][d] = x[n][d] + 0.5 * (vn + v[n][d]) * dt;
956 // When using constraints, the update is split into
957 // two phases, but we only need to zero the update of
958 // virtual, shell or frozen particles in at most one
960 if (updateType != SDUpdate::FrictionAndNoiseOnly)
963 xprime[n][d] = x[n][d];
970 static void do_update_bd(int start,
973 const ivec nFreeze[],
974 const real invmass[],
975 const unsigned short ptype[],
976 const unsigned short cFREEZE[],
977 const unsigned short cTC[],
982 real friction_coefficient,
988 /* note -- these appear to be full step velocities . . . */
993 // Use 1 bit of internal counters to give us 2*2 64-bits values per stream
994 // Each 64-bit value is enough for 4 normal distribution table numbers.
995 gmx::ThreeFry2x64<0> rng(seed, gmx::RandomDomain::UpdateCoordinates);
996 gmx::TabulatedNormalDistribution<real, 14> dist;
998 if (friction_coefficient != 0)
1000 invfr = 1.0 / friction_coefficient;
1003 for (n = start; (n < nrend); n++)
1005 int ng = gatindex ? gatindex[n] : n;
1007 rng.restart(step, ng);
1018 for (d = 0; (d < DIM); d++)
1020 if ((ptype[n] != eptVSite) && (ptype[n] != eptShell) && !nFreeze[gf][d])
1022 if (friction_coefficient != 0)
1024 vn = invfr * f[n][d] + rf[gt] * dist(rng);
1028 /* NOTE: invmass = 2/(mass*friction_constant*dt) */
1029 vn = 0.5 * invmass[n] * f[n][d] * dt
1030 + std::sqrt(0.5 * invmass[n]) * rf[gt] * dist(rng);
1034 xprime[n][d] = x[n][d] + vn * dt;
1039 xprime[n][d] = x[n][d];
1045 static void calc_ke_part_normal(const rvec v[],
1046 const t_grpopts* opts,
1047 const t_mdatoms* md,
1048 gmx_ekindata_t* ekind,
1050 gmx_bool bEkinAveVel)
1053 gmx::ArrayRef<t_grp_tcstat> tcstat = ekind->tcstat;
1054 gmx::ArrayRef<t_grp_acc> grpstat = ekind->grpstat;
1056 /* three main: VV with AveVel, vv with AveEkin, leap with AveEkin. Leap with AveVel is also
1057 an option, but not supported now.
1058 bEkinAveVel: If TRUE, we sum into ekin, if FALSE, into ekinh.
1061 /* group velocities are calculated in update_ekindata and
1062 * accumulated in acumulate_groups.
1063 * Now the partial global and groups ekin.
1065 for (g = 0; (g < opts->ngtc); g++)
1067 copy_mat(tcstat[g].ekinh, tcstat[g].ekinh_old);
1070 clear_mat(tcstat[g].ekinf);
1071 tcstat[g].ekinscalef_nhc = 1.0; /* need to clear this -- logic is complicated! */
1075 clear_mat(tcstat[g].ekinh);
1078 ekind->dekindl_old = ekind->dekindl;
1079 int nthread = gmx_omp_nthreads_get(emntUpdate);
1081 #pragma omp parallel for num_threads(nthread) schedule(static)
1082 for (int thread = 0; thread < nthread; thread++)
1084 // This OpenMP only loops over arrays and does not call any functions
1085 // or memory allocation. It should not be able to throw, so for now
1086 // we do not need a try/catch wrapper.
1087 int start_t, end_t, n;
1095 start_t = ((thread + 0) * md->homenr) / nthread;
1096 end_t = ((thread + 1) * md->homenr) / nthread;
1098 ekin_sum = ekind->ekin_work[thread];
1099 dekindl_sum = ekind->dekindl_work[thread];
1101 for (gt = 0; gt < opts->ngtc; gt++)
1103 clear_mat(ekin_sum[gt]);
1109 for (n = start_t; n < end_t; n++)
1119 hm = 0.5 * md->massT[n];
1121 for (d = 0; (d < DIM); d++)
1123 v_corrt[d] = v[n][d] - grpstat[ga].u[d];
1125 for (d = 0; (d < DIM); d++)
1127 for (m = 0; (m < DIM); m++)
1129 /* if we're computing a full step velocity, v_corrt[d] has v(t). Otherwise, v(t+dt/2) */
1130 ekin_sum[gt][m][d] += hm * v_corrt[m] * v_corrt[d];
1133 if (md->nMassPerturbed && md->bPerturbed[n])
1135 *dekindl_sum += 0.5 * (md->massB[n] - md->massA[n]) * iprod(v_corrt, v_corrt);
1141 for (int thread = 0; thread < nthread; thread++)
1143 for (g = 0; g < opts->ngtc; g++)
1147 m_add(tcstat[g].ekinf, ekind->ekin_work[thread][g], tcstat[g].ekinf);
1151 m_add(tcstat[g].ekinh, ekind->ekin_work[thread][g], tcstat[g].ekinh);
1155 ekind->dekindl += *ekind->dekindl_work[thread];
1158 inc_nrnb(nrnb, eNR_EKIN, md->homenr);
1161 static void calc_ke_part_visc(const matrix box,
1164 const t_grpopts* opts,
1165 const t_mdatoms* md,
1166 gmx_ekindata_t* ekind,
1168 gmx_bool bEkinAveVel)
1170 int start = 0, homenr = md->homenr;
1171 int g, d, n, m, gt = 0;
1174 gmx::ArrayRef<t_grp_tcstat> tcstat = ekind->tcstat;
1175 t_cos_acc* cosacc = &(ekind->cosacc);
1180 for (g = 0; g < opts->ngtc; g++)
1182 copy_mat(ekind->tcstat[g].ekinh, ekind->tcstat[g].ekinh_old);
1183 clear_mat(ekind->tcstat[g].ekinh);
1185 ekind->dekindl_old = ekind->dekindl;
1187 fac = 2 * M_PI / box[ZZ][ZZ];
1190 for (n = start; n < start + homenr; n++)
1196 hm = 0.5 * md->massT[n];
1198 /* Note that the times of x and v differ by half a step */
1199 /* MRS -- would have to be changed for VV */
1200 cosz = std::cos(fac * x[n][ZZ]);
1201 /* Calculate the amplitude of the new velocity profile */
1202 mvcos += 2 * cosz * md->massT[n] * v[n][XX];
1204 copy_rvec(v[n], v_corrt);
1205 /* Subtract the profile for the kinetic energy */
1206 v_corrt[XX] -= cosz * cosacc->vcos;
1207 for (d = 0; (d < DIM); d++)
1209 for (m = 0; (m < DIM); m++)
1211 /* if we're computing a full step velocity, v_corrt[d] has v(t). Otherwise, v(t+dt/2) */
1214 tcstat[gt].ekinf[m][d] += hm * v_corrt[m] * v_corrt[d];
1218 tcstat[gt].ekinh[m][d] += hm * v_corrt[m] * v_corrt[d];
1222 if (md->nPerturbed && md->bPerturbed[n])
1224 /* The minus sign here might be confusing.
1225 * The kinetic contribution from dH/dl doesn't come from
1226 * d m(l)/2 v^2 / dl, but rather from d p^2/2m(l) / dl,
1227 * where p are the momenta. The difference is only a minus sign.
1229 dekindl -= 0.5 * (md->massB[n] - md->massA[n]) * iprod(v_corrt, v_corrt);
1232 ekind->dekindl = dekindl;
1233 cosacc->mvcos = mvcos;
1235 inc_nrnb(nrnb, eNR_EKIN, homenr);
1238 void calc_ke_part(const rvec* x,
1241 const t_grpopts* opts,
1242 const t_mdatoms* md,
1243 gmx_ekindata_t* ekind,
1245 gmx_bool bEkinAveVel)
1247 if (ekind->cosacc.cos_accel == 0)
1249 calc_ke_part_normal(v, opts, md, ekind, nrnb, bEkinAveVel);
1253 calc_ke_part_visc(box, x, v, opts, md, ekind, nrnb, bEkinAveVel);
1257 extern void init_ekinstate(ekinstate_t* ekinstate, const t_inputrec* ir)
1259 ekinstate->ekin_n = ir->opts.ngtc;
1260 snew(ekinstate->ekinh, ekinstate->ekin_n);
1261 snew(ekinstate->ekinf, ekinstate->ekin_n);
1262 snew(ekinstate->ekinh_old, ekinstate->ekin_n);
1263 ekinstate->ekinscalef_nhc.resize(ekinstate->ekin_n);
1264 ekinstate->ekinscaleh_nhc.resize(ekinstate->ekin_n);
1265 ekinstate->vscale_nhc.resize(ekinstate->ekin_n);
1266 ekinstate->dekindl = 0;
1267 ekinstate->mvcos = 0;
1268 ekinstate->hasReadEkinState = false;
1271 void update_ekinstate(ekinstate_t* ekinstate, const gmx_ekindata_t* ekind)
1275 for (i = 0; i < ekinstate->ekin_n; i++)
1277 copy_mat(ekind->tcstat[i].ekinh, ekinstate->ekinh[i]);
1278 copy_mat(ekind->tcstat[i].ekinf, ekinstate->ekinf[i]);
1279 copy_mat(ekind->tcstat[i].ekinh_old, ekinstate->ekinh_old[i]);
1280 ekinstate->ekinscalef_nhc[i] = ekind->tcstat[i].ekinscalef_nhc;
1281 ekinstate->ekinscaleh_nhc[i] = ekind->tcstat[i].ekinscaleh_nhc;
1282 ekinstate->vscale_nhc[i] = ekind->tcstat[i].vscale_nhc;
1285 copy_mat(ekind->ekin, ekinstate->ekin_total);
1286 ekinstate->dekindl = ekind->dekindl;
1287 ekinstate->mvcos = ekind->cosacc.mvcos;
1290 void restore_ekinstate_from_state(const t_commrec* cr, gmx_ekindata_t* ekind, const ekinstate_t* ekinstate)
1296 for (i = 0; i < ekinstate->ekin_n; i++)
1298 copy_mat(ekinstate->ekinh[i], ekind->tcstat[i].ekinh);
1299 copy_mat(ekinstate->ekinf[i], ekind->tcstat[i].ekinf);
1300 copy_mat(ekinstate->ekinh_old[i], ekind->tcstat[i].ekinh_old);
1301 ekind->tcstat[i].ekinscalef_nhc = ekinstate->ekinscalef_nhc[i];
1302 ekind->tcstat[i].ekinscaleh_nhc = ekinstate->ekinscaleh_nhc[i];
1303 ekind->tcstat[i].vscale_nhc = ekinstate->vscale_nhc[i];
1306 copy_mat(ekinstate->ekin_total, ekind->ekin);
1308 ekind->dekindl = ekinstate->dekindl;
1309 ekind->cosacc.mvcos = ekinstate->mvcos;
1310 n = ekinstate->ekin_n;
1315 gmx_bcast(sizeof(n), &n, cr);
1316 for (i = 0; i < n; i++)
1318 gmx_bcast(DIM * DIM * sizeof(ekind->tcstat[i].ekinh[0][0]), ekind->tcstat[i].ekinh[0], cr);
1319 gmx_bcast(DIM * DIM * sizeof(ekind->tcstat[i].ekinf[0][0]), ekind->tcstat[i].ekinf[0], cr);
1320 gmx_bcast(DIM * DIM * sizeof(ekind->tcstat[i].ekinh_old[0][0]),
1321 ekind->tcstat[i].ekinh_old[0], cr);
1323 gmx_bcast(sizeof(ekind->tcstat[i].ekinscalef_nhc), &(ekind->tcstat[i].ekinscalef_nhc), cr);
1324 gmx_bcast(sizeof(ekind->tcstat[i].ekinscaleh_nhc), &(ekind->tcstat[i].ekinscaleh_nhc), cr);
1325 gmx_bcast(sizeof(ekind->tcstat[i].vscale_nhc), &(ekind->tcstat[i].vscale_nhc), cr);
1327 gmx_bcast(DIM * DIM * sizeof(ekind->ekin[0][0]), ekind->ekin[0], cr);
1329 gmx_bcast(sizeof(ekind->dekindl), &ekind->dekindl, cr);
1330 gmx_bcast(sizeof(ekind->cosacc.mvcos), &ekind->cosacc.mvcos, cr);
1334 void update_tcouple(int64_t step,
1335 const t_inputrec* inputrec,
1337 gmx_ekindata_t* ekind,
1338 const t_extmass* MassQ,
1339 const t_mdatoms* md)
1342 // This condition was explicitly checked in previous version, but should have never been satisfied
1343 GMX_ASSERT(!(EI_VV(inputrec->eI)
1344 && (inputrecNvtTrotter(inputrec) || inputrecNptTrotter(inputrec)
1345 || inputrecNphTrotter(inputrec))),
1346 "Temperature coupling was requested with velocity verlet and trotter");
1348 bool doTemperatureCoupling = false;
1350 // For VV temperature coupling parameters are updated on the current
1351 // step, for the others - one step before.
1352 if (inputrec->etc == etcNO)
1354 doTemperatureCoupling = false;
1356 else if (EI_VV(inputrec->eI))
1358 doTemperatureCoupling = do_per_step(step, inputrec->nsttcouple);
1362 doTemperatureCoupling = do_per_step(step + inputrec->nsttcouple - 1, inputrec->nsttcouple);
1365 if (doTemperatureCoupling)
1367 real dttc = inputrec->nsttcouple * inputrec->delta_t;
1369 // TODO: berendsen_tcoupl(...), nosehoover_tcoupl(...) and vrescale_tcoupl(...) update
1370 // temperature coupling parameters, which should be reflected in the name of these
1372 switch (inputrec->etc)
1376 berendsen_tcoupl(inputrec, ekind, dttc, state->therm_integral);
1379 nosehoover_tcoupl(&(inputrec->opts), ekind, dttc, state->nosehoover_xi.data(),
1380 state->nosehoover_vxi.data(), MassQ);
1383 vrescale_tcoupl(inputrec, step, ekind, dttc, state->therm_integral.data());
1386 /* rescale in place here */
1387 if (EI_VV(inputrec->eI))
1389 rescale_velocities(ekind, md, 0, md->homenr, state->v.rvec_array());
1394 // Set the T scaling lambda to 1 to have no scaling
1395 // TODO: Do we have to do it on every non-t-couple step?
1396 for (int i = 0; (i < inputrec->opts.ngtc); i++)
1398 ekind->tcstat[i].lambda = 1.0;
1403 void getThreadAtomRange(int numThreads, int threadIndex, int numAtoms, int* startAtom, int* endAtom)
1405 #if GMX_HAVE_SIMD_UPDATE
1406 constexpr int blockSize = GMX_SIMD_REAL_WIDTH;
1408 constexpr int blockSize = 1;
1410 int numBlocks = (numAtoms + blockSize - 1) / blockSize;
1412 *startAtom = ((numBlocks * threadIndex) / numThreads) * blockSize;
1413 *endAtom = ((numBlocks * (threadIndex + 1)) / numThreads) * blockSize;
1414 if (threadIndex == numThreads - 1)
1416 *endAtom = numAtoms;
1420 void update_pcouple_before_coordinates(FILE* fplog,
1422 const t_inputrec* inputrec,
1424 matrix parrinellorahmanMu,
1428 /* Berendsen P-coupling is completely handled after the coordinate update.
1429 * Trotter P-coupling is handled by separate calls to trotter_update().
1431 if (inputrec->epc == epcPARRINELLORAHMAN
1432 && do_per_step(step + inputrec->nstpcouple - 1, inputrec->nstpcouple))
1434 real dtpc = inputrec->nstpcouple * inputrec->delta_t;
1436 parrinellorahman_pcoupl(fplog, step, inputrec, dtpc, state->pres_prev, state->box,
1437 state->box_rel, state->boxv, M, parrinellorahmanMu, bInitStep);
1441 void constrain_velocities(int64_t step,
1442 real* dvdlambda, /* the contribution to be added to the bonded interactions */
1445 gmx::Constraints* constr,
1457 * APPLY CONSTRAINTS:
1464 /* clear out constraints before applying */
1465 clear_mat(vir_part);
1467 /* Constrain the coordinates upd->xp */
1468 constr->apply(do_log, do_ene, step, 1, 1.0, state->x.arrayRefWithPadding(),
1469 state->v.arrayRefWithPadding(), state->v.arrayRefWithPadding().unpaddedArrayRef(),
1470 state->box, state->lambda[efptBONDED], dvdlambda, ArrayRefWithPadding<RVec>(),
1471 bCalcVir ? &vir_con : nullptr, ConstraintVariable::Velocities);
1475 m_add(vir_part, vir_con, vir_part);
1480 void constrain_coordinates(int64_t step,
1481 real* dvdlambda, /* the contribution to be added to the bonded interactions */
1485 gmx::Constraints* constr,
1498 /* clear out constraints before applying */
1499 clear_mat(vir_part);
1501 /* Constrain the coordinates upd->xp */
1502 constr->apply(do_log, do_ene, step, 1, 1.0, state->x.arrayRefWithPadding(),
1503 upd->xp()->arrayRefWithPadding(), ArrayRef<RVec>(), state->box,
1504 state->lambda[efptBONDED], dvdlambda, state->v.arrayRefWithPadding(),
1505 bCalcVir ? &vir_con : nullptr, ConstraintVariable::Positions);
1509 m_add(vir_part, vir_con, vir_part);
1514 void update_sd_second_half(int64_t step,
1515 real* dvdlambda, /* the contribution to be added to the bonded interactions */
1516 const t_inputrec* inputrec, /* input record and box stuff */
1517 const t_mdatoms* md,
1519 const t_commrec* cr,
1521 gmx_wallcycle_t wcycle,
1523 gmx::Constraints* constr,
1531 if (inputrec->eI == eiSD1)
1533 int homenr = md->homenr;
1535 /* Cast delta_t from double to real to make the integrators faster.
1536 * The only reason for having delta_t double is to get accurate values
1537 * for t=delta_t*step when step is larger than float precision.
1538 * For integration dt the accuracy of real suffices, since with
1539 * integral += dt*integrand the increment is nearly always (much) smaller
1540 * than the integral (and the integrand has real precision).
1542 real dt = inputrec->delta_t;
1544 wallcycle_start(wcycle, ewcUPDATE);
1546 int nth = gmx_omp_nthreads_get(emntUpdate);
1548 #pragma omp parallel for num_threads(nth) schedule(static)
1549 for (int th = 0; th < nth; th++)
1553 int start_th, end_th;
1554 getThreadAtomRange(nth, th, homenr, &start_th, &end_th);
1556 doSDUpdateGeneral<SDUpdate::FrictionAndNoiseOnly>(
1557 *upd->sd(), start_th, end_th, dt, inputrec->opts.acc, inputrec->opts.nFreeze,
1558 md->invmass, md->ptype, md->cFREEZE, nullptr, md->cTC, state->x.rvec_array(),
1559 upd->xp()->rvec_array(), state->v.rvec_array(), nullptr, step, inputrec->ld_seed,
1560 DOMAINDECOMP(cr) ? cr->dd->globalAtomIndices.data() : nullptr);
1562 GMX_CATCH_ALL_AND_EXIT_WITH_FATAL_ERROR
1564 inc_nrnb(nrnb, eNR_UPDATE, homenr);
1565 wallcycle_stop(wcycle, ewcUPDATE);
1567 /* Constrain the coordinates upd->xp for half a time step */
1568 constr->apply(do_log, do_ene, step, 1, 0.5, state->x.arrayRefWithPadding(),
1569 upd->xp()->arrayRefWithPadding(), ArrayRef<RVec>(), state->box,
1570 state->lambda[efptBONDED], dvdlambda, state->v.arrayRefWithPadding(), nullptr,
1571 ConstraintVariable::Positions);
1575 void finish_update(const t_inputrec* inputrec, /* input record and box stuff */
1576 const t_mdatoms* md,
1578 const t_graph* graph,
1580 gmx_wallcycle_t wcycle,
1582 const gmx::Constraints* constr)
1584 int homenr = md->homenr;
1586 /* We must always unshift after updating coordinates; if we did not shake
1587 x was shifted in do_force */
1589 /* NOTE Currently we always integrate to a temporary buffer and
1590 * then copy the results back. */
1592 wallcycle_start_nocount(wcycle, ewcUPDATE);
1594 if (md->cFREEZE != nullptr && constr != nullptr)
1596 /* If we have atoms that are frozen along some, but not all
1597 * dimensions, then any constraints will have moved them also along
1598 * the frozen dimensions. To freeze such degrees of freedom
1599 * we copy them back here to later copy them forward. It would
1600 * be more elegant and slightly more efficient to copies zero
1601 * times instead of twice, but the graph case below prevents this.
1603 const ivec* nFreeze = inputrec->opts.nFreeze;
1604 bool partialFreezeAndConstraints = false;
1605 for (int g = 0; g < inputrec->opts.ngfrz; g++)
1607 int numFreezeDim = nFreeze[g][XX] + nFreeze[g][YY] + nFreeze[g][ZZ];
1608 if (numFreezeDim > 0 && numFreezeDim < 3)
1610 partialFreezeAndConstraints = true;
1613 if (partialFreezeAndConstraints)
1615 auto xp = makeArrayRef(*upd->xp()).subArray(0, homenr);
1616 auto x = makeConstArrayRef(state->x).subArray(0, homenr);
1617 for (int i = 0; i < homenr; i++)
1619 int g = md->cFREEZE[i];
1621 for (int d = 0; d < DIM; d++)
1632 if (graph && (graph->nnodes > 0))
1634 unshift_x(graph, state->box, state->x.rvec_array(), upd->xp()->rvec_array());
1635 if (TRICLINIC(state->box))
1637 inc_nrnb(nrnb, eNR_SHIFTX, 2 * graph->nnodes);
1641 inc_nrnb(nrnb, eNR_SHIFTX, graph->nnodes);
1646 auto xp = makeConstArrayRef(*upd->xp()).subArray(0, homenr);
1647 auto x = makeArrayRef(state->x).subArray(0, homenr);
1650 int gmx_unused nth = gmx_omp_nthreads_get(emntUpdate);
1651 #pragma omp parallel for num_threads(nth) schedule(static)
1652 for (int i = 0; i < homenr; i++)
1654 // Trivial statement, does not throw
1658 wallcycle_stop(wcycle, ewcUPDATE);
1660 /* ############# END the update of velocities and positions ######### */
1663 void update_pcouple_after_coordinates(FILE* fplog,
1665 const t_inputrec* inputrec,
1666 const t_mdatoms* md,
1667 const matrix pressure,
1668 const matrix forceVirial,
1669 const matrix constraintVirial,
1670 matrix pressureCouplingMu,
1674 const bool scaleCoordinates)
1677 int homenr = md->homenr;
1679 /* Cast to real for faster code, no loss in precision (see comment above) */
1680 real dt = inputrec->delta_t;
1683 /* now update boxes */
1684 switch (inputrec->epc)
1686 case (epcNO): break;
1687 case (epcBERENDSEN):
1688 if (do_per_step(step, inputrec->nstpcouple))
1690 real dtpc = inputrec->nstpcouple * dt;
1691 berendsen_pcoupl(fplog, step, inputrec, dtpc, pressure, state->box, forceVirial,
1692 constraintVirial, pressureCouplingMu, &state->baros_integral);
1693 berendsen_pscale(inputrec, pressureCouplingMu, state->box, state->box_rel, start,
1694 homenr, state->x.rvec_array(), md->cFREEZE, nrnb, scaleCoordinates);
1697 case (epcPARRINELLORAHMAN):
1698 if (do_per_step(step + inputrec->nstpcouple - 1, inputrec->nstpcouple))
1700 /* The box velocities were updated in do_pr_pcoupl,
1701 * but we dont change the box vectors until we get here
1702 * since we need to be able to shift/unshift above.
1704 real dtpc = inputrec->nstpcouple * dt;
1705 for (int i = 0; i < DIM; i++)
1707 for (int m = 0; m <= i; m++)
1709 state->box[i][m] += dtpc * state->boxv[i][m];
1712 preserve_box_shape(inputrec, state->box_rel, state->box);
1714 /* Scale the coordinates */
1715 if (scaleCoordinates)
1717 auto x = state->x.rvec_array();
1718 for (int n = start; n < start + homenr; n++)
1720 tmvmul_ur0(pressureCouplingMu, x[n], x[n]);
1726 switch (inputrec->epct)
1728 case (epctISOTROPIC):
1729 /* DIM * eta = ln V. so DIM*eta_new = DIM*eta_old + DIM*dt*veta =>
1730 ln V_new = ln V_old + 3*dt*veta => V_new = V_old*exp(3*dt*veta) =>
1731 Side length scales as exp(veta*dt) */
1733 msmul(state->box, std::exp(state->veta * dt), state->box);
1735 /* Relate veta to boxv. veta = d(eta)/dT = (1/DIM)*1/V dV/dT.
1736 o If we assume isotropic scaling, and box length scaling
1737 factor L, then V = L^DIM (det(M)). So dV/dt = DIM
1738 L^(DIM-1) dL/dt det(M), and veta = (1/L) dL/dt. The
1739 determinant of B is L^DIM det(M), and the determinant
1740 of dB/dt is (dL/dT)^DIM det (M). veta will be
1741 (det(dB/dT)/det(B))^(1/3). Then since M =
1742 B_new*(vol_new)^(1/3), dB/dT_new = (veta_new)*B(new). */
1744 msmul(state->box, state->veta, state->boxv);
1754 auto localX = makeArrayRef(state->x).subArray(start, homenr);
1755 upd->deform()->apply(localX, state->box, step);
1759 void update_coords(int64_t step,
1760 const t_inputrec* inputrec, /* input record and box stuff */
1761 const t_mdatoms* md,
1763 gmx::ArrayRefWithPadding<const gmx::RVec> f,
1764 const t_fcdata* fcd,
1765 const gmx_ekindata_t* ekind,
1769 const t_commrec* cr, /* these shouldn't be here -- need to think about it */
1770 const gmx::Constraints* constr)
1772 gmx_bool bDoConstr = (nullptr != constr);
1774 /* Running the velocity half does nothing except for velocity verlet */
1775 if ((UpdatePart == etrtVELOCITY1 || UpdatePart == etrtVELOCITY2) && !EI_VV(inputrec->eI))
1777 gmx_incons("update_coords called for velocity without VV integrator");
1780 int homenr = md->homenr;
1782 /* Cast to real for faster code, no loss in precision (see comment above) */
1783 real dt = inputrec->delta_t;
1785 /* We need to update the NMR restraint history when time averaging is used */
1786 if (state->flags & (1 << estDISRE_RM3TAV))
1788 update_disres_history(fcd, &state->hist);
1790 if (state->flags & (1 << estORIRE_DTAV))
1792 update_orires_history(fcd, &state->hist);
1795 /* ############# START The update of velocities and positions ######### */
1796 int nth = gmx_omp_nthreads_get(emntUpdate);
1798 #pragma omp parallel for num_threads(nth) schedule(static)
1799 for (int th = 0; th < nth; th++)
1803 int start_th, end_th;
1804 getThreadAtomRange(nth, th, homenr, &start_th, &end_th);
1806 const rvec* x_rvec = state->x.rvec_array();
1807 rvec* xp_rvec = upd->xp()->rvec_array();
1808 rvec* v_rvec = state->v.rvec_array();
1809 const rvec* f_rvec = as_rvec_array(f.unpaddedArrayRef().data());
1811 switch (inputrec->eI)
1814 do_update_md(start_th, end_th, step, dt, inputrec, md, ekind, state->box,
1815 x_rvec, xp_rvec, v_rvec, f_rvec, state->nosehoover_vxi.data(), M);
1820 // With constraints, the SD update is done in 2 parts
1821 doSDUpdateGeneral<SDUpdate::ForcesOnly>(
1822 *upd->sd(), start_th, end_th, dt, inputrec->opts.acc, inputrec->opts.nFreeze,
1823 md->invmass, md->ptype, md->cFREEZE, md->cACC, nullptr, x_rvec,
1824 xp_rvec, v_rvec, f_rvec, step, inputrec->ld_seed, nullptr);
1828 doSDUpdateGeneral<SDUpdate::Combined>(
1829 *upd->sd(), start_th, end_th, dt, inputrec->opts.acc,
1830 inputrec->opts.nFreeze, md->invmass, md->ptype, md->cFREEZE, md->cACC,
1831 md->cTC, x_rvec, xp_rvec, v_rvec, f_rvec, step, inputrec->ld_seed,
1832 DOMAINDECOMP(cr) ? cr->dd->globalAtomIndices.data() : nullptr);
1836 do_update_bd(start_th, end_th, dt, inputrec->opts.nFreeze, md->invmass,
1837 md->ptype, md->cFREEZE, md->cTC, x_rvec, xp_rvec, v_rvec, f_rvec,
1838 inputrec->bd_fric, upd->sd()->bd_rf.data(), step, inputrec->ld_seed,
1839 DOMAINDECOMP(cr) ? cr->dd->globalAtomIndices.data() : nullptr);
1844 gmx_bool bExtended = (inputrec->etc == etcNOSEHOOVER || inputrec->epc == epcPARRINELLORAHMAN
1845 || inputrec->epc == epcMTTK);
1847 /* assuming barostat coupled to group 0 */
1848 real alpha = 1.0 + DIM / static_cast<real>(inputrec->opts.nrdf[0]);
1853 do_update_vv_vel(start_th, end_th, dt, inputrec->opts.acc,
1854 inputrec->opts.nFreeze, md->invmass, md->ptype, md->cFREEZE,
1855 md->cACC, v_rvec, f_rvec, bExtended, state->veta, alpha);
1858 do_update_vv_pos(start_th, end_th, dt, inputrec->opts.nFreeze, md->ptype,
1859 md->cFREEZE, x_rvec, xp_rvec, v_rvec, bExtended, state->veta);
1864 default: gmx_fatal(FARGS, "Don't know how to update coordinates");
1867 GMX_CATCH_ALL_AND_EXIT_WITH_FATAL_ERROR
1871 extern gmx_bool update_randomize_velocities(const t_inputrec* ir,
1873 const t_commrec* cr,
1874 const t_mdatoms* md,
1875 gmx::ArrayRef<gmx::RVec> v,
1877 const gmx::Constraints* constr)
1880 real rate = (ir->delta_t) / ir->opts.tau_t[0];
1882 if (ir->etc == etcANDERSEN && constr != nullptr)
1884 /* Currently, Andersen thermostat does not support constrained
1885 systems. Functionality exists in the andersen_tcoupl
1886 function in GROMACS 4.5.7 to allow this combination. That
1887 code could be ported to the current random-number
1888 generation approach, but has not yet been done because of
1889 lack of time and resources. */
1891 "Normal Andersen is currently not supported with constraints, use massive "
1892 "Andersen instead");
1895 /* proceed with andersen if 1) it's fixed probability per
1896 particle andersen or 2) it's massive andersen and it's tau_t/dt */
1897 if ((ir->etc == etcANDERSEN) || do_per_step(step, roundToInt(1.0 / rate)))
1899 andersen_tcoupl(ir, step, cr, md, v, rate, upd->sd()->randomize_group, upd->sd()->boltzfac);