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/sim_util.h"
63 #include "gromacs/mdlib/tgroup.h"
64 #include "gromacs/mdtypes/commrec.h"
65 #include "gromacs/mdtypes/group.h"
66 #include "gromacs/mdtypes/inputrec.h"
67 #include "gromacs/mdtypes/md_enums.h"
68 #include "gromacs/mdtypes/state.h"
69 #include "gromacs/pbcutil/boxutilities.h"
70 #include "gromacs/pbcutil/mshift.h"
71 #include "gromacs/pbcutil/pbc.h"
72 #include "gromacs/pulling/pull.h"
73 #include "gromacs/random/tabulatednormaldistribution.h"
74 #include "gromacs/random/threefry.h"
75 #include "gromacs/simd/simd.h"
76 #include "gromacs/timing/wallcycle.h"
77 #include "gromacs/topology/atoms.h"
78 #include "gromacs/utility/exceptions.h"
79 #include "gromacs/utility/fatalerror.h"
80 #include "gromacs/utility/futil.h"
81 #include "gromacs/utility/gmxassert.h"
82 #include "gromacs/utility/gmxomp.h"
83 #include "gromacs/utility/smalloc.h"
85 using namespace gmx; // TODO: Remove when this file is moved into gmx namespace
87 struct gmx_sd_const_t {
91 struct gmx_sd_sigma_t {
98 std::vector<real> bd_rf;
100 std::vector<gmx_sd_const_t> sdc;
101 std::vector<gmx_sd_sigma_t> sdsig;
102 /* andersen temperature control stuff */
103 std::vector<bool> randomize_group;
104 std::vector<real> boltzfac;
106 explicit gmx_stochd_t(const t_inputrec *ir);
109 //! pImpled implementation for Update
114 Impl(const t_inputrec *ir, BoxDeformation *boxDeformation);
117 //! stochastic dynamics struct
118 std::unique_ptr<gmx_stochd_t> sd;
119 //! xprime for constraint algorithms
120 PaddedVector<RVec> xp;
121 //! Box deformation handler (or nullptr if inactive).
122 BoxDeformation *deform = nullptr;
125 Update::Update(const t_inputrec *ir, BoxDeformation *boxDeformation)
126 : impl_(new Impl(ir, boxDeformation))
132 gmx_stochd_t* Update::sd() const
134 return impl_->sd.get();
137 PaddedVector<RVec> * Update::xp()
142 BoxDeformation * Update::deform() const
144 return impl_->deform;
147 static bool isTemperatureCouplingStep(int64_t step, const t_inputrec *ir)
149 /* We should only couple after a step where energies were determined (for leapfrog versions)
150 or the step energies are determined, for velocity verlet versions */
160 return ir->etc != etcNO &&
161 (ir->nsttcouple == 1 ||
162 do_per_step(step + ir->nsttcouple - offset, ir->nsttcouple));
165 static bool isPressureCouplingStep(int64_t step, const t_inputrec *ir)
167 GMX_ASSERT(ir->epc != epcMTTK, "MTTK pressure coupling is not handled here");
170 if (ir->epc == epcBERENDSEN)
178 /* We should only couple after a step where pressures were determined */
179 return ir->epc != etcNO &&
180 (ir->nstpcouple == 1 ||
181 do_per_step(step + ir->nstpcouple - offset, ir->nstpcouple));
184 /*! \brief Sets the velocities of virtual sites to zero */
185 static void clearVsiteVelocities(int start,
187 const unsigned short *particleType,
188 rvec * gmx_restrict v)
190 for (int a = start; a < nrend; a++)
192 if (particleType[a] == eptVSite)
199 /*! \brief Sets the number of different temperature coupling values */
200 enum class NumTempScaleValues
202 single, //!< Single T-scaling value (either one group or all values =1)
203 multiple //!< Multiple T-scaling values, need to use T-group indices
206 /*! \brief Sets if to apply no or diagonal Parrinello-Rahman pressure scaling
208 * Note that this enum is only used in updateMDLeapfrogSimple(), which does
209 * not handle fully anistropic Parrinello-Rahman scaling, so we only have
210 * options \p no and \p diagonal here and no anistropic option.
212 enum class ApplyParrinelloRahmanVScaling
214 no, //!< Do not apply velocity scaling (not a PR-coupling run or step)
215 diagonal //!< Apply velocity scaling using a diagonal matrix
218 /*! \brief Integrate using leap-frog with T-scaling and optionally diagonal Parrinello-Rahman p-coupling
220 * \tparam numTempScaleValues The number of different T-couple values
221 * \tparam applyPRVScaling Apply Parrinello-Rahman velocity scaling
222 * \param[in] start Index of first atom to update
223 * \param[in] nrend Last atom to update: \p nrend - 1
224 * \param[in] dt The time step
225 * \param[in] dtPressureCouple Time step for pressure coupling
226 * \param[in] invMassPerDim 1/mass per atom and dimension
227 * \param[in] tcstat Temperature coupling information
228 * \param[in] cTC T-coupling group index per atom
229 * \param[in] pRVScaleMatrixDiagonal Parrinello-Rahman v-scale matrix diagonal
230 * \param[in] x Input coordinates
231 * \param[out] xprime Updated coordinates
232 * \param[inout] v Velocities
233 * \param[in] f Forces
235 * We expect this template to get good SIMD acceleration by most compilers,
236 * unlike the more complex general template.
237 * Note that we might get even better SIMD acceleration when we introduce
238 * aligned (and padded) memory, possibly with some hints for the compilers.
240 template<NumTempScaleValues numTempScaleValues,
241 ApplyParrinelloRahmanVScaling applyPRVScaling>
243 updateMDLeapfrogSimple(int start,
246 real dtPressureCouple,
247 const rvec * gmx_restrict invMassPerDim,
248 gmx::ArrayRef<const t_grp_tcstat> tcstat,
249 const unsigned short * cTC,
250 const rvec pRVScaleMatrixDiagonal,
251 const rvec * gmx_restrict x,
252 rvec * gmx_restrict xprime,
253 rvec * gmx_restrict v,
254 const rvec * gmx_restrict f)
258 if (numTempScaleValues == NumTempScaleValues::single)
260 lambdaGroup = tcstat[0].lambda;
263 for (int a = start; a < nrend; a++)
265 if (numTempScaleValues == NumTempScaleValues::multiple)
267 lambdaGroup = tcstat[cTC[a]].lambda;
270 for (int d = 0; d < DIM; d++)
272 /* Note that using rvec invMassPerDim results in more efficient
273 * SIMD code, but this increases the cache pressure.
274 * For large systems with PME on the CPU this slows down the
275 * (then already slow) update by 20%. If all data remains in cache,
276 * using rvec is much faster.
278 real vNew = lambdaGroup*v[a][d] + f[a][d]*invMassPerDim[a][d]*dt;
280 if (applyPRVScaling == ApplyParrinelloRahmanVScaling::diagonal)
282 vNew -= dtPressureCouple*pRVScaleMatrixDiagonal[d]*v[a][d];
285 xprime[a][d] = x[a][d] + vNew*dt;
290 #if GMX_SIMD && GMX_SIMD_HAVE_REAL
291 #define GMX_HAVE_SIMD_UPDATE 1
293 #define GMX_HAVE_SIMD_UPDATE 0
296 #if GMX_HAVE_SIMD_UPDATE
298 /*! \brief Load (aligned) the contents of GMX_SIMD_REAL_WIDTH rvec elements sequentially into 3 SIMD registers
300 * The loaded output is:
301 * \p r0: { r[index][XX], r[index][YY], ... }
303 * \p r2: { ..., r[index+GMX_SIMD_REAL_WIDTH-1][YY], r[index+GMX_SIMD_REAL_WIDTH-1][ZZ] }
305 * \param[in] r Real to an rvec array, has to be aligned to SIMD register width
306 * \param[in] index Index of the first rvec triplet of reals to load
307 * \param[out] r0 Pointer to first SIMD register
308 * \param[out] r1 Pointer to second SIMD register
309 * \param[out] r2 Pointer to third SIMD register
311 static inline void simdLoadRvecs(const rvec *r,
317 const real *realPtr = r[index];
319 GMX_ASSERT(isSimdAligned(realPtr), "Pointer should be SIMD aligned");
321 *r0 = simdLoad(realPtr + 0*GMX_SIMD_REAL_WIDTH);
322 *r1 = simdLoad(realPtr + 1*GMX_SIMD_REAL_WIDTH);
323 *r2 = simdLoad(realPtr + 2*GMX_SIMD_REAL_WIDTH);
326 /*! \brief Store (aligned) 3 SIMD registers sequentially to GMX_SIMD_REAL_WIDTH rvec elements
328 * The stored output is:
329 * \p r[index] = { { r0[0], r0[1], ... }
331 * \p r[index+GMX_SIMD_REAL_WIDTH-1] = { ... , r2[GMX_SIMD_REAL_WIDTH-2], r2[GMX_SIMD_REAL_WIDTH-1] }
333 * \param[out] r Pointer to an rvec array, has to be aligned to SIMD register width
334 * \param[in] index Index of the first rvec triplet of reals to store to
335 * \param[in] r0 First SIMD register
336 * \param[in] r1 Second SIMD register
337 * \param[in] r2 Third SIMD register
339 static inline void simdStoreRvecs(rvec *r,
345 real *realPtr = r[index];
347 GMX_ASSERT(isSimdAligned(realPtr), "Pointer should be SIMD aligned");
349 store(realPtr + 0*GMX_SIMD_REAL_WIDTH, r0);
350 store(realPtr + 1*GMX_SIMD_REAL_WIDTH, r1);
351 store(realPtr + 2*GMX_SIMD_REAL_WIDTH, r2);
354 /*! \brief Integrate using leap-frog with single group T-scaling and SIMD
356 * \param[in] start Index of first atom to update
357 * \param[in] nrend Last atom to update: \p nrend - 1
358 * \param[in] dt The time step
359 * \param[in] invMass 1/mass per atom
360 * \param[in] tcstat Temperature coupling information
361 * \param[in] x Input coordinates
362 * \param[out] xprime Updated coordinates
363 * \param[inout] v Velocities
364 * \param[in] f Forces
367 updateMDLeapfrogSimpleSimd(int start,
370 const real * gmx_restrict invMass,
371 gmx::ArrayRef<const t_grp_tcstat> tcstat,
372 const rvec * gmx_restrict x,
373 rvec * gmx_restrict xprime,
374 rvec * gmx_restrict v,
375 const rvec * gmx_restrict f)
377 SimdReal timestep(dt);
378 SimdReal lambdaSystem(tcstat[0].lambda);
380 /* We declare variables here, since code is often slower when declaring them inside the loop */
382 /* Note: We should implement a proper PaddedVector, so we don't need this check */
383 GMX_ASSERT(isSimdAligned(invMass), "invMass should be aligned");
385 for (int a = start; a < nrend; a += GMX_SIMD_REAL_WIDTH)
387 SimdReal invMass0, invMass1, invMass2;
388 expandScalarsToTriplets(simdLoad(invMass + a),
389 &invMass0, &invMass1, &invMass2);
393 simdLoadRvecs(v, a, &v0, &v1, &v2);
394 simdLoadRvecs(f, a, &f0, &f1, &f2);
396 v0 = fma(f0*invMass0, timestep, lambdaSystem*v0);
397 v1 = fma(f1*invMass1, timestep, lambdaSystem*v1);
398 v2 = fma(f2*invMass2, timestep, lambdaSystem*v2);
400 simdStoreRvecs(v, a, v0, v1, v2);
403 simdLoadRvecs(x, a, &x0, &x1, &x2);
405 SimdReal xprime0 = fma(v0, timestep, x0);
406 SimdReal xprime1 = fma(v1, timestep, x1);
407 SimdReal xprime2 = fma(v2, timestep, x2);
409 simdStoreRvecs(xprime, a, xprime0, xprime1, xprime2);
413 #endif // GMX_HAVE_SIMD_UPDATE
415 /*! \brief Sets the NEMD acceleration type */
416 enum class AccelerationType
421 /*! \brief Integrate using leap-frog with support for everything
423 * \tparam accelerationType Type of NEMD acceleration
424 * \param[in] start Index of first atom to update
425 * \param[in] nrend Last atom to update: \p nrend - 1
426 * \param[in] doNoseHoover If to apply Nose-Hoover T-coupling
427 * \param[in] dt The time step
428 * \param[in] dtPressureCouple Time step for pressure coupling, is 0 when no pressure coupling should be applied at this step
429 * \param[in] ir The input parameter record
430 * \param[in] md Atom properties
431 * \param[in] ekind Kinetic energy data
432 * \param[in] box The box dimensions
433 * \param[in] x Input coordinates
434 * \param[out] xprime Updated coordinates
435 * \param[inout] v Velocities
436 * \param[in] f Forces
437 * \param[in] nh_vxi Nose-Hoover velocity scaling factors
438 * \param[in] M Parrinello-Rahman scaling matrix
440 template<AccelerationType accelerationType>
442 updateMDLeapfrogGeneral(int start,
446 real dtPressureCouple,
447 const t_inputrec * ir,
448 const t_mdatoms * md,
449 const gmx_ekindata_t * ekind,
451 const rvec * gmx_restrict x,
452 rvec * gmx_restrict xprime,
453 rvec * gmx_restrict v,
454 const rvec * gmx_restrict f,
455 const double * gmx_restrict nh_vxi,
458 /* This is a version of the leap-frog integrator that supports
459 * all combinations of T-coupling, P-coupling and NEMD.
460 * Nose-Hoover uses the reversible leap-frog integrator from
461 * Holian et al. Phys Rev E 52(3) : 2338, 1995
464 gmx::ArrayRef<const t_grp_tcstat> tcstat = ekind->tcstat;
465 gmx::ArrayRef<const t_grp_acc> grpstat = ekind->grpstat;
466 const unsigned short * cTC = md->cTC;
467 const unsigned short * cACC = md->cACC;
468 const rvec * accel = ir->opts.acc;
470 const rvec * gmx_restrict invMassPerDim = md->invMassPerDim;
472 /* Initialize group values, changed later when multiple groups are used */
477 real omega_Z = 2*static_cast<real>(M_PI)/box[ZZ][ZZ];
479 for (int n = start; n < nrend; n++)
485 real lg = tcstat[gt].lambda;
488 real cosineZ, vCosine;
489 #ifdef __INTEL_COMPILER
490 #pragma warning( disable : 280 )
492 switch (accelerationType)
494 case AccelerationType::none:
495 copy_rvec(v[n], vRel);
497 case AccelerationType::group:
502 /* Avoid scaling the group velocity */
503 rvec_sub(v[n], grpstat[ga].u, vRel);
505 case AccelerationType::cosine:
506 cosineZ = std::cos(x[n][ZZ]*omega_Z);
507 vCosine = cosineZ*ekind->cosacc.vcos;
508 /* Avoid scaling the cosine profile velocity */
509 copy_rvec(v[n], vRel);
516 /* Here we account for multiple time stepping, by increasing
517 * the Nose-Hoover correction by nsttcouple
519 factorNH = 0.5*ir->nsttcouple*dt*nh_vxi[gt];
522 for (int d = 0; d < DIM; d++)
525 (lg*vRel[d] + (f[n][d]*invMassPerDim[n][d]*dt - factorNH*vRel[d]
526 - dtPressureCouple*iprod(M[d], vRel)))/(1 + factorNH);
527 switch (accelerationType)
529 case AccelerationType::none:
531 case AccelerationType::group:
532 /* Add back the mean velocity and apply acceleration */
533 vNew += grpstat[ga].u[d] + accel[ga][d]*dt;
535 case AccelerationType::cosine:
538 /* Add back the mean velocity and apply acceleration */
539 vNew += vCosine + cosineZ*ekind->cosacc.cos_accel*dt;
544 xprime[n][d] = x[n][d] + vNew*dt;
549 /*! \brief Handles the Leap-frog MD x and v integration */
550 static void do_update_md(int start,
554 const t_inputrec * ir,
555 const t_mdatoms * md,
556 const gmx_ekindata_t * ekind,
558 const rvec * gmx_restrict x,
559 rvec * gmx_restrict xprime,
560 rvec * gmx_restrict v,
561 const rvec * gmx_restrict f,
562 const double * gmx_restrict nh_vxi,
565 GMX_ASSERT(nrend == start || xprime != x, "For SIMD optimization certain compilers need to have xprime != x");
567 /* Note: Berendsen pressure scaling is handled after do_update_md() */
568 bool doTempCouple = isTemperatureCouplingStep(step, ir);
569 bool doNoseHoover = (ir->etc == etcNOSEHOOVER && doTempCouple);
570 bool doParrinelloRahman = (ir->epc == epcPARRINELLORAHMAN && isPressureCouplingStep(step, ir));
571 bool doPROffDiagonal = (doParrinelloRahman && (M[YY][XX] != 0 || M[ZZ][XX] != 0 || M[ZZ][YY] != 0));
573 real dtPressureCouple = (doParrinelloRahman ? ir->nstpcouple*dt : 0);
575 /* NEMD (also cosine) acceleration is applied in updateMDLeapFrogGeneral */
576 bool doAcceleration = (ekind->bNEMD || ekind->cosacc.cos_accel != 0);
578 if (doNoseHoover || doPROffDiagonal || doAcceleration)
581 if (!doParrinelloRahman)
583 /* We should not apply PR scaling at this step */
593 updateMDLeapfrogGeneral<AccelerationType::none>
594 (start, nrend, doNoseHoover, dt, dtPressureCouple,
595 ir, md, ekind, box, x, xprime, v, f, nh_vxi, stepM);
597 else if (ekind->bNEMD)
599 updateMDLeapfrogGeneral<AccelerationType::group>
600 (start, nrend, doNoseHoover, dt, dtPressureCouple,
601 ir, md, ekind, box, x, xprime, v, f, nh_vxi, stepM);
605 updateMDLeapfrogGeneral<AccelerationType::cosine>
606 (start, nrend, doNoseHoover, dt, dtPressureCouple,
607 ir, md, ekind, box, x, xprime, v, f, nh_vxi, stepM);
612 /* Use a simple and thus more efficient integration loop. */
613 /* The simple loop does not check for particle type (so it can
614 * be vectorized), which means we need to clear the velocities
615 * of virtual sites in advance, when present. Note that vsite
616 * velocities are computed after update and constraints from
617 * their displacement.
621 /* Note: The overhead of this loop is completely neligible */
622 clearVsiteVelocities(start, nrend, md->ptype, v);
625 /* We determine if we have a single T-coupling lambda value for all
626 * atoms. That allows for better SIMD acceleration in the template.
627 * If we do not do temperature coupling (in the run or this step),
628 * all scaling values are 1, so we effectively have a single value.
630 bool haveSingleTempScaleValue = (!doTempCouple || ekind->ngtc == 1);
632 /* Extract some pointers needed by all cases */
633 const unsigned short *cTC = md->cTC;
634 gmx::ArrayRef<const t_grp_tcstat> tcstat = ekind->tcstat;
635 const rvec *invMassPerDim = md->invMassPerDim;
637 if (doParrinelloRahman)
639 GMX_ASSERT(!doPROffDiagonal, "updateMDLeapfrogSimple only support diagonal Parrinello-Rahman scaling matrices");
642 for (int d = 0; d < DIM; d++)
647 if (haveSingleTempScaleValue)
649 updateMDLeapfrogSimple
650 <NumTempScaleValues::single,
651 ApplyParrinelloRahmanVScaling::diagonal>
652 (start, nrend, dt, dtPressureCouple,
653 invMassPerDim, tcstat, cTC, diagM, x, xprime, v, f);
657 updateMDLeapfrogSimple
658 <NumTempScaleValues::multiple,
659 ApplyParrinelloRahmanVScaling::diagonal>
660 (start, nrend, dt, dtPressureCouple,
661 invMassPerDim, tcstat, cTC, diagM, x, xprime, v, f);
666 if (haveSingleTempScaleValue)
668 /* Note that modern compilers are pretty good at vectorizing
669 * updateMDLeapfrogSimple(). But the SIMD version will still
670 * be faster because invMass lowers the cache pressure
671 * compared to invMassPerDim.
673 #if GMX_HAVE_SIMD_UPDATE
674 /* Check if we can use invmass instead of invMassPerDim */
675 if (!md->havePartiallyFrozenAtoms)
677 updateMDLeapfrogSimpleSimd
678 (start, nrend, dt, md->invmass, tcstat, x, xprime, v, f);
683 updateMDLeapfrogSimple
684 <NumTempScaleValues::single,
685 ApplyParrinelloRahmanVScaling::no>
686 (start, nrend, dt, dtPressureCouple,
687 invMassPerDim, tcstat, cTC, nullptr, x, xprime, v, f);
692 updateMDLeapfrogSimple
693 <NumTempScaleValues::multiple,
694 ApplyParrinelloRahmanVScaling::no>
695 (start, nrend, dt, dtPressureCouple,
696 invMassPerDim, tcstat, cTC, nullptr, x, xprime, v, f);
702 static void do_update_vv_vel(int start, int nrend, real dt,
703 const rvec accel[], const ivec nFreeze[], const real invmass[],
704 const unsigned short ptype[], const unsigned short cFREEZE[],
705 const unsigned short cACC[], rvec v[], const rvec f[],
706 gmx_bool bExtended, real veta, real alpha)
714 g = 0.25*dt*veta*alpha;
716 mv2 = gmx::series_sinhx(g);
723 for (n = start; n < nrend; n++)
725 real w_dt = invmass[n]*dt;
735 for (d = 0; d < DIM; d++)
737 if ((ptype[n] != eptVSite) && (ptype[n] != eptShell) && !nFreeze[gf][d])
739 v[n][d] = mv1*(mv1*v[n][d] + 0.5*(w_dt*mv2*f[n][d]))+0.5*accel[ga][d]*dt;
747 } /* do_update_vv_vel */
749 static void do_update_vv_pos(int start, int nrend, real dt,
750 const ivec nFreeze[],
751 const unsigned short ptype[], const unsigned short cFREEZE[],
752 const rvec x[], rvec xprime[], const rvec v[],
753 gmx_bool bExtended, real veta)
759 /* Would it make more sense if Parrinello-Rahman was put here? */
764 mr2 = gmx::series_sinhx(g);
772 for (n = start; n < nrend; n++)
780 for (d = 0; d < DIM; d++)
782 if ((ptype[n] != eptVSite) && (ptype[n] != eptShell) && !nFreeze[gf][d])
784 xprime[n][d] = mr1*(mr1*x[n][d]+mr2*dt*v[n][d]);
788 xprime[n][d] = x[n][d];
792 } /* do_update_vv_pos */
794 gmx_stochd_t::gmx_stochd_t(const t_inputrec *ir)
796 const t_grpopts *opts = &ir->opts;
797 const int ngtc = opts->ngtc;
803 else if (EI_SD(ir->eI))
808 for (int gt = 0; gt < ngtc; gt++)
810 if (opts->tau_t[gt] > 0)
812 sdc[gt].em = std::exp(-ir->delta_t/opts->tau_t[gt]);
816 /* No friction and noise on this group */
821 else if (ETC_ANDERSEN(ir->etc))
823 randomize_group.resize(ngtc);
824 boltzfac.resize(ngtc);
826 /* for now, assume that all groups, if randomized, are randomized at the same rate, i.e. tau_t is the same. */
827 /* since constraint groups don't necessarily match up with temperature groups! This is checked in readir.c */
829 for (int gt = 0; gt < ngtc; gt++)
831 real reft = std::max<real>(0, opts->ref_t[gt]);
832 if ((opts->tau_t[gt] > 0) && (reft > 0)) /* tau_t or ref_t = 0 means that no randomization is done */
834 randomize_group[gt] = true;
835 boltzfac[gt] = BOLTZ*opts->ref_t[gt];
839 randomize_group[gt] = false;
845 void update_temperature_constants(gmx_stochd_t *sd, const t_inputrec *ir)
849 if (ir->bd_fric != 0)
851 for (int gt = 0; gt < ir->opts.ngtc; gt++)
853 sd->bd_rf[gt] = std::sqrt(2.0*BOLTZ*ir->opts.ref_t[gt]/(ir->bd_fric*ir->delta_t));
858 for (int gt = 0; gt < ir->opts.ngtc; gt++)
860 sd->bd_rf[gt] = std::sqrt(2.0*BOLTZ*ir->opts.ref_t[gt]);
866 for (int gt = 0; gt < ir->opts.ngtc; gt++)
868 real kT = BOLTZ*ir->opts.ref_t[gt];
869 /* The mass is accounted for later, since this differs per atom */
870 sd->sdsig[gt].V = std::sqrt(kT*(1 - sd->sdc[gt].em * sd->sdc[gt].em));
875 Update::Impl::Impl(const t_inputrec *ir, BoxDeformation *boxDeformation)
877 sd = std::make_unique<gmx_stochd_t>(ir);
878 update_temperature_constants(sd.get(), ir);
879 xp.resizeWithPadding(0);
880 deform = boxDeformation;
883 void Update::setNumAtoms(int nAtoms)
886 impl_->xp.resizeWithPadding(nAtoms);
889 /*! \brief Sets the SD update type */
890 enum class SDUpdate : int
892 ForcesOnly, FrictionAndNoiseOnly, Combined
895 /*! \brief SD integrator update
897 * Two phases are required in the general case of a constrained
898 * update, the first phase from the contribution of forces, before
899 * applying constraints, and then a second phase applying the friction
900 * and noise, and then further constraining. For details, see
903 * Without constraints, the two phases can be combined, for
906 * Thus three instantiations of this templated function will be made,
907 * two with only one contribution, and one with both contributions. */
908 template <SDUpdate updateType>
910 doSDUpdateGeneral(const gmx_stochd_t &sd,
911 int start, int nrend, real dt,
912 const rvec accel[], const ivec nFreeze[],
913 const real invmass[], const unsigned short ptype[],
914 const unsigned short cFREEZE[], const unsigned short cACC[],
915 const unsigned short cTC[],
916 const rvec x[], rvec xprime[], rvec v[], const rvec f[],
917 int64_t step, int seed, const int *gatindex)
919 // cTC, cACC and cFreeze can be nullptr any time, but various
920 // instantiations do not make sense with particular pointer
922 if (updateType == SDUpdate::ForcesOnly)
924 GMX_ASSERT(f != nullptr, "SD update with only forces requires forces");
925 GMX_ASSERT(cTC == nullptr, "SD update with only forces cannot handle temperature groups");
927 if (updateType == SDUpdate::FrictionAndNoiseOnly)
929 GMX_ASSERT(f == nullptr, "SD update with only noise cannot handle forces");
930 GMX_ASSERT(cACC == nullptr, "SD update with only noise cannot handle acceleration groups");
932 if (updateType == SDUpdate::Combined)
934 GMX_ASSERT(f != nullptr, "SD update with forces and noise requires forces");
937 // Even 0 bits internal counter gives 2x64 ints (more than enough for three table lookups)
938 gmx::ThreeFry2x64<0> rng(seed, gmx::RandomDomain::UpdateCoordinates);
939 gmx::TabulatedNormalDistribution<real, 14> dist;
941 for (int n = start; n < nrend; n++)
943 int globalAtomIndex = gatindex ? gatindex[n] : n;
944 rng.restart(step, globalAtomIndex);
947 real inverseMass = invmass[n];
948 real invsqrtMass = std::sqrt(inverseMass);
950 int freezeGroup = cFREEZE ? cFREEZE[n] : 0;
951 int accelerationGroup = cACC ? cACC[n] : 0;
952 int temperatureGroup = cTC ? cTC[n] : 0;
954 for (int d = 0; d < DIM; d++)
956 if ((ptype[n] != eptVSite) && (ptype[n] != eptShell) && !nFreeze[freezeGroup][d])
958 if (updateType == SDUpdate::ForcesOnly)
960 real vn = v[n][d] + (inverseMass*f[n][d] + accel[accelerationGroup][d])*dt;
962 // Simple position update.
963 xprime[n][d] = x[n][d] + v[n][d]*dt;
965 else if (updateType == SDUpdate::FrictionAndNoiseOnly)
968 v[n][d] = (vn*sd.sdc[temperatureGroup].em +
969 invsqrtMass*sd.sdsig[temperatureGroup].V*dist(rng));
970 // The previous phase already updated the
971 // positions with a full v*dt term that must
972 // now be half removed.
973 xprime[n][d] = xprime[n][d] + 0.5*(v[n][d] - vn)*dt;
977 real vn = v[n][d] + (inverseMass*f[n][d] + accel[accelerationGroup][d])*dt;
978 v[n][d] = (vn*sd.sdc[temperatureGroup].em +
979 invsqrtMass*sd.sdsig[temperatureGroup].V*dist(rng));
980 // Here we include half of the friction+noise
981 // update of v into the position update.
982 xprime[n][d] = x[n][d] + 0.5*(vn + v[n][d])*dt;
987 // When using constraints, the update is split into
988 // two phases, but we only need to zero the update of
989 // virtual, shell or frozen particles in at most one
991 if (updateType != SDUpdate::FrictionAndNoiseOnly)
994 xprime[n][d] = x[n][d];
1001 static void do_update_bd(int start, int nrend, real dt,
1002 const ivec nFreeze[],
1003 const real invmass[], const unsigned short ptype[],
1004 const unsigned short cFREEZE[], const unsigned short cTC[],
1005 const rvec x[], rvec xprime[], rvec v[],
1006 const rvec f[], real friction_coefficient,
1007 const real *rf, int64_t step, int seed,
1008 const int* gatindex)
1010 /* note -- these appear to be full step velocities . . . */
1015 // Use 1 bit of internal counters to give us 2*2 64-bits values per stream
1016 // Each 64-bit value is enough for 4 normal distribution table numbers.
1017 gmx::ThreeFry2x64<0> rng(seed, gmx::RandomDomain::UpdateCoordinates);
1018 gmx::TabulatedNormalDistribution<real, 14> dist;
1020 if (friction_coefficient != 0)
1022 invfr = 1.0/friction_coefficient;
1025 for (n = start; (n < nrend); n++)
1027 int ng = gatindex ? gatindex[n] : n;
1029 rng.restart(step, ng);
1040 for (d = 0; (d < DIM); d++)
1042 if ((ptype[n] != eptVSite) && (ptype[n] != eptShell) && !nFreeze[gf][d])
1044 if (friction_coefficient != 0)
1046 vn = invfr*f[n][d] + rf[gt]*dist(rng);
1050 /* NOTE: invmass = 2/(mass*friction_constant*dt) */
1051 vn = 0.5*invmass[n]*f[n][d]*dt
1052 + std::sqrt(0.5*invmass[n])*rf[gt]*dist(rng);
1056 xprime[n][d] = x[n][d]+vn*dt;
1061 xprime[n][d] = x[n][d];
1067 static void calc_ke_part_normal(const rvec v[], const t_grpopts *opts, const t_mdatoms *md,
1068 gmx_ekindata_t *ekind, t_nrnb *nrnb, gmx_bool bEkinAveVel)
1071 gmx::ArrayRef<t_grp_tcstat> tcstat = ekind->tcstat;
1072 gmx::ArrayRef<t_grp_acc> grpstat = ekind->grpstat;
1073 int nthread, thread;
1075 /* three main: VV with AveVel, vv with AveEkin, leap with AveEkin. Leap with AveVel is also
1076 an option, but not supported now.
1077 bEkinAveVel: If TRUE, we sum into ekin, if FALSE, into ekinh.
1080 /* group velocities are calculated in update_ekindata and
1081 * accumulated in acumulate_groups.
1082 * Now the partial global and groups ekin.
1084 for (g = 0; (g < opts->ngtc); g++)
1086 copy_mat(tcstat[g].ekinh, tcstat[g].ekinh_old);
1089 clear_mat(tcstat[g].ekinf);
1090 tcstat[g].ekinscalef_nhc = 1.0; /* need to clear this -- logic is complicated! */
1094 clear_mat(tcstat[g].ekinh);
1097 ekind->dekindl_old = ekind->dekindl;
1098 nthread = gmx_omp_nthreads_get(emntUpdate);
1100 #pragma omp parallel for num_threads(nthread) schedule(static)
1101 for (thread = 0; thread < nthread; thread++)
1103 // This OpenMP only loops over arrays and does not call any functions
1104 // or memory allocation. It should not be able to throw, so for now
1105 // we do not need a try/catch wrapper.
1106 int start_t, end_t, n;
1114 start_t = ((thread+0)*md->homenr)/nthread;
1115 end_t = ((thread+1)*md->homenr)/nthread;
1117 ekin_sum = ekind->ekin_work[thread];
1118 dekindl_sum = ekind->dekindl_work[thread];
1120 for (gt = 0; gt < opts->ngtc; gt++)
1122 clear_mat(ekin_sum[gt]);
1128 for (n = start_t; n < end_t; n++)
1138 hm = 0.5*md->massT[n];
1140 for (d = 0; (d < DIM); d++)
1142 v_corrt[d] = v[n][d] - grpstat[ga].u[d];
1144 for (d = 0; (d < DIM); d++)
1146 for (m = 0; (m < DIM); m++)
1148 /* if we're computing a full step velocity, v_corrt[d] has v(t). Otherwise, v(t+dt/2) */
1149 ekin_sum[gt][m][d] += hm*v_corrt[m]*v_corrt[d];
1152 if (md->nMassPerturbed && md->bPerturbed[n])
1155 0.5*(md->massB[n] - md->massA[n])*iprod(v_corrt, v_corrt);
1161 for (thread = 0; thread < nthread; thread++)
1163 for (g = 0; g < opts->ngtc; g++)
1167 m_add(tcstat[g].ekinf, ekind->ekin_work[thread][g],
1172 m_add(tcstat[g].ekinh, ekind->ekin_work[thread][g],
1177 ekind->dekindl += *ekind->dekindl_work[thread];
1180 inc_nrnb(nrnb, eNR_EKIN, md->homenr);
1183 static void calc_ke_part_visc(const matrix box, const rvec x[], const rvec v[],
1184 const t_grpopts *opts, const t_mdatoms *md,
1185 gmx_ekindata_t *ekind,
1186 t_nrnb *nrnb, gmx_bool bEkinAveVel)
1188 int start = 0, homenr = md->homenr;
1189 int g, d, n, m, gt = 0;
1192 gmx::ArrayRef<t_grp_tcstat> tcstat = ekind->tcstat;
1193 t_cos_acc *cosacc = &(ekind->cosacc);
1198 for (g = 0; g < opts->ngtc; g++)
1200 copy_mat(ekind->tcstat[g].ekinh, ekind->tcstat[g].ekinh_old);
1201 clear_mat(ekind->tcstat[g].ekinh);
1203 ekind->dekindl_old = ekind->dekindl;
1205 fac = 2*M_PI/box[ZZ][ZZ];
1208 for (n = start; n < start+homenr; n++)
1214 hm = 0.5*md->massT[n];
1216 /* Note that the times of x and v differ by half a step */
1217 /* MRS -- would have to be changed for VV */
1218 cosz = std::cos(fac*x[n][ZZ]);
1219 /* Calculate the amplitude of the new velocity profile */
1220 mvcos += 2*cosz*md->massT[n]*v[n][XX];
1222 copy_rvec(v[n], v_corrt);
1223 /* Subtract the profile for the kinetic energy */
1224 v_corrt[XX] -= cosz*cosacc->vcos;
1225 for (d = 0; (d < DIM); d++)
1227 for (m = 0; (m < DIM); m++)
1229 /* if we're computing a full step velocity, v_corrt[d] has v(t). Otherwise, v(t+dt/2) */
1232 tcstat[gt].ekinf[m][d] += hm*v_corrt[m]*v_corrt[d];
1236 tcstat[gt].ekinh[m][d] += hm*v_corrt[m]*v_corrt[d];
1240 if (md->nPerturbed && md->bPerturbed[n])
1242 /* The minus sign here might be confusing.
1243 * The kinetic contribution from dH/dl doesn't come from
1244 * d m(l)/2 v^2 / dl, but rather from d p^2/2m(l) / dl,
1245 * where p are the momenta. The difference is only a minus sign.
1247 dekindl -= 0.5*(md->massB[n] - md->massA[n])*iprod(v_corrt, v_corrt);
1250 ekind->dekindl = dekindl;
1251 cosacc->mvcos = mvcos;
1253 inc_nrnb(nrnb, eNR_EKIN, homenr);
1256 void calc_ke_part(const t_state *state, const t_grpopts *opts, const t_mdatoms *md,
1257 gmx_ekindata_t *ekind, t_nrnb *nrnb, gmx_bool bEkinAveVel)
1259 if (ekind->cosacc.cos_accel == 0)
1261 calc_ke_part_normal(state->v.rvec_array(), opts, md, ekind, nrnb, bEkinAveVel);
1265 calc_ke_part_visc(state->box, state->x.rvec_array(), state->v.rvec_array(), opts, md, ekind, nrnb, bEkinAveVel);
1269 extern void init_ekinstate(ekinstate_t *ekinstate, const t_inputrec *ir)
1271 ekinstate->ekin_n = ir->opts.ngtc;
1272 snew(ekinstate->ekinh, ekinstate->ekin_n);
1273 snew(ekinstate->ekinf, ekinstate->ekin_n);
1274 snew(ekinstate->ekinh_old, ekinstate->ekin_n);
1275 ekinstate->ekinscalef_nhc.resize(ekinstate->ekin_n);
1276 ekinstate->ekinscaleh_nhc.resize(ekinstate->ekin_n);
1277 ekinstate->vscale_nhc.resize(ekinstate->ekin_n);
1278 ekinstate->dekindl = 0;
1279 ekinstate->mvcos = 0;
1282 void update_ekinstate(ekinstate_t *ekinstate, const gmx_ekindata_t *ekind)
1286 for (i = 0; i < ekinstate->ekin_n; i++)
1288 copy_mat(ekind->tcstat[i].ekinh, ekinstate->ekinh[i]);
1289 copy_mat(ekind->tcstat[i].ekinf, ekinstate->ekinf[i]);
1290 copy_mat(ekind->tcstat[i].ekinh_old, ekinstate->ekinh_old[i]);
1291 ekinstate->ekinscalef_nhc[i] = ekind->tcstat[i].ekinscalef_nhc;
1292 ekinstate->ekinscaleh_nhc[i] = ekind->tcstat[i].ekinscaleh_nhc;
1293 ekinstate->vscale_nhc[i] = ekind->tcstat[i].vscale_nhc;
1296 copy_mat(ekind->ekin, ekinstate->ekin_total);
1297 ekinstate->dekindl = ekind->dekindl;
1298 ekinstate->mvcos = ekind->cosacc.mvcos;
1302 void restore_ekinstate_from_state(const t_commrec *cr,
1303 gmx_ekindata_t *ekind, const ekinstate_t *ekinstate)
1309 for (i = 0; i < ekinstate->ekin_n; i++)
1311 copy_mat(ekinstate->ekinh[i], ekind->tcstat[i].ekinh);
1312 copy_mat(ekinstate->ekinf[i], ekind->tcstat[i].ekinf);
1313 copy_mat(ekinstate->ekinh_old[i], ekind->tcstat[i].ekinh_old);
1314 ekind->tcstat[i].ekinscalef_nhc = ekinstate->ekinscalef_nhc[i];
1315 ekind->tcstat[i].ekinscaleh_nhc = ekinstate->ekinscaleh_nhc[i];
1316 ekind->tcstat[i].vscale_nhc = ekinstate->vscale_nhc[i];
1319 copy_mat(ekinstate->ekin_total, ekind->ekin);
1321 ekind->dekindl = ekinstate->dekindl;
1322 ekind->cosacc.mvcos = ekinstate->mvcos;
1323 n = ekinstate->ekin_n;
1328 gmx_bcast(sizeof(n), &n, cr);
1329 for (i = 0; i < n; i++)
1331 gmx_bcast(DIM*DIM*sizeof(ekind->tcstat[i].ekinh[0][0]),
1332 ekind->tcstat[i].ekinh[0], cr);
1333 gmx_bcast(DIM*DIM*sizeof(ekind->tcstat[i].ekinf[0][0]),
1334 ekind->tcstat[i].ekinf[0], cr);
1335 gmx_bcast(DIM*DIM*sizeof(ekind->tcstat[i].ekinh_old[0][0]),
1336 ekind->tcstat[i].ekinh_old[0], cr);
1338 gmx_bcast(sizeof(ekind->tcstat[i].ekinscalef_nhc),
1339 &(ekind->tcstat[i].ekinscalef_nhc), cr);
1340 gmx_bcast(sizeof(ekind->tcstat[i].ekinscaleh_nhc),
1341 &(ekind->tcstat[i].ekinscaleh_nhc), cr);
1342 gmx_bcast(sizeof(ekind->tcstat[i].vscale_nhc),
1343 &(ekind->tcstat[i].vscale_nhc), cr);
1345 gmx_bcast(DIM*DIM*sizeof(ekind->ekin[0][0]),
1346 ekind->ekin[0], cr);
1348 gmx_bcast(sizeof(ekind->dekindl), &ekind->dekindl, cr);
1349 gmx_bcast(sizeof(ekind->cosacc.mvcos), &ekind->cosacc.mvcos, cr);
1353 void update_tcouple(int64_t step,
1354 const t_inputrec *inputrec,
1356 gmx_ekindata_t *ekind,
1357 const t_extmass *MassQ,
1358 const t_mdatoms *md)
1361 bool doTemperatureCoupling = false;
1363 /* if using vv with trotter decomposition methods, we do this elsewhere in the code */
1364 if (!(EI_VV(inputrec->eI) &&
1365 (inputrecNvtTrotter(inputrec) || inputrecNptTrotter(inputrec) || inputrecNphTrotter(inputrec))))
1367 doTemperatureCoupling = isTemperatureCouplingStep(step, inputrec);
1370 if (doTemperatureCoupling)
1372 real dttc = inputrec->nsttcouple*inputrec->delta_t;
1374 switch (inputrec->etc)
1379 berendsen_tcoupl(inputrec, ekind, dttc, state->therm_integral);
1382 nosehoover_tcoupl(&(inputrec->opts), ekind, dttc,
1383 state->nosehoover_xi.data(), state->nosehoover_vxi.data(), MassQ);
1386 vrescale_tcoupl(inputrec, step, ekind, dttc,
1387 state->therm_integral.data());
1390 /* rescale in place here */
1391 if (EI_VV(inputrec->eI))
1393 rescale_velocities(ekind, md, 0, md->homenr, state->v.rvec_array());
1398 /* Set the T scaling lambda to 1 to have no scaling */
1399 for (int i = 0; (i < inputrec->opts.ngtc); i++)
1401 ekind->tcstat[i].lambda = 1.0;
1406 /*! \brief Computes the atom range for a thread to operate on, ensured SIMD aligned ranges
1408 * \param[in] numThreads The number of threads to divide atoms over
1409 * \param[in] threadIndex The thread to get the range for
1410 * \param[in] numAtoms The total number of atoms (on this rank)
1411 * \param[out] startAtom The start of the atom range
1412 * \param[out] endAtom The end of the atom range, note that this is in general not a multiple of the SIMD width
1414 static void getThreadAtomRange(int numThreads, int threadIndex, int numAtoms,
1415 int *startAtom, int *endAtom)
1417 #if GMX_HAVE_SIMD_UPDATE
1418 constexpr int blockSize = GMX_SIMD_REAL_WIDTH;
1420 constexpr int blockSize = 1;
1422 int numBlocks = (numAtoms + blockSize - 1)/blockSize;
1424 *startAtom = ((numBlocks* threadIndex )/numThreads)*blockSize;
1425 *endAtom = ((numBlocks*(threadIndex + 1))/numThreads)*blockSize;
1426 if (threadIndex == numThreads - 1)
1428 *endAtom = numAtoms;
1432 void update_pcouple_before_coordinates(FILE *fplog,
1434 const t_inputrec *inputrec,
1436 matrix parrinellorahmanMu,
1440 /* Berendsen P-coupling is completely handled after the coordinate update.
1441 * Trotter P-coupling is handled by separate calls to trotter_update().
1443 if (inputrec->epc == epcPARRINELLORAHMAN &&
1444 isPressureCouplingStep(step, inputrec))
1446 real dtpc = inputrec->nstpcouple*inputrec->delta_t;
1448 parrinellorahman_pcoupl(fplog, step, inputrec, dtpc, state->pres_prev,
1449 state->box, state->box_rel, state->boxv,
1450 M, parrinellorahmanMu, bInitStep);
1454 void constrain_velocities(int64_t step,
1455 real *dvdlambda, /* the contribution to be added to the bonded interactions */
1458 gmx::Constraints *constr,
1470 * APPLY CONSTRAINTS:
1477 /* clear out constraints before applying */
1478 clear_mat(vir_part);
1480 /* Constrain the coordinates upd->xp */
1481 constr->apply(do_log, do_ene,
1483 state->x.rvec_array(), state->v.rvec_array(), state->v.rvec_array(),
1485 state->lambda[efptBONDED], dvdlambda,
1486 nullptr, bCalcVir ? &vir_con : nullptr, ConstraintVariable::Velocities);
1490 m_add(vir_part, vir_con, vir_part);
1495 void constrain_coordinates(int64_t step,
1496 real *dvdlambda, /* the contribution to be added to the bonded interactions */
1500 gmx::Constraints *constr,
1513 /* clear out constraints before applying */
1514 clear_mat(vir_part);
1516 /* Constrain the coordinates upd->xp */
1517 constr->apply(do_log, do_ene,
1519 state->x.rvec_array(), upd->xp()->rvec_array(), nullptr,
1521 state->lambda[efptBONDED], dvdlambda,
1522 as_rvec_array(state->v.data()), bCalcVir ? &vir_con : nullptr, ConstraintVariable::Positions);
1526 m_add(vir_part, vir_con, vir_part);
1532 update_sd_second_half(int64_t step,
1533 real *dvdlambda, /* the contribution to be added to the bonded interactions */
1534 const t_inputrec *inputrec, /* input record and box stuff */
1535 const t_mdatoms *md,
1537 const t_commrec *cr,
1539 gmx_wallcycle_t wcycle,
1541 gmx::Constraints *constr,
1549 if (inputrec->eI == eiSD1)
1552 int homenr = md->homenr;
1554 /* Cast delta_t from double to real to make the integrators faster.
1555 * The only reason for having delta_t double is to get accurate values
1556 * for t=delta_t*step when step is larger than float precision.
1557 * For integration dt the accuracy of real suffices, since with
1558 * integral += dt*integrand the increment is nearly always (much) smaller
1559 * than the integral (and the integrand has real precision).
1561 real dt = inputrec->delta_t;
1563 wallcycle_start(wcycle, ewcUPDATE);
1565 nth = gmx_omp_nthreads_get(emntUpdate);
1567 #pragma omp parallel for num_threads(nth) schedule(static)
1568 for (th = 0; th < nth; th++)
1572 int start_th, end_th;
1573 getThreadAtomRange(nth, th, homenr, &start_th, &end_th);
1575 doSDUpdateGeneral<SDUpdate::FrictionAndNoiseOnly>
1577 start_th, end_th, dt,
1578 inputrec->opts.acc, inputrec->opts.nFreeze,
1579 md->invmass, md->ptype,
1580 md->cFREEZE, nullptr, md->cTC,
1581 state->x.rvec_array(), upd->xp()->rvec_array(),
1582 state->v.rvec_array(), nullptr,
1583 step, inputrec->ld_seed,
1584 DOMAINDECOMP(cr) ? cr->dd->globalAtomIndices.data() : nullptr);
1586 GMX_CATCH_ALL_AND_EXIT_WITH_FATAL_ERROR;
1588 inc_nrnb(nrnb, eNR_UPDATE, homenr);
1589 wallcycle_stop(wcycle, ewcUPDATE);
1591 /* Constrain the coordinates upd->xp for half a time step */
1592 constr->apply(do_log, do_ene,
1594 state->x.rvec_array(), upd->xp()->rvec_array(), nullptr,
1596 state->lambda[efptBONDED], dvdlambda,
1597 as_rvec_array(state->v.data()), nullptr, ConstraintVariable::Positions);
1601 void finish_update(const t_inputrec *inputrec, /* input record and box stuff */
1602 const t_mdatoms *md,
1604 const t_graph *graph,
1606 gmx_wallcycle_t wcycle,
1608 const gmx::Constraints *constr)
1610 int homenr = md->homenr;
1612 /* We must always unshift after updating coordinates; if we did not shake
1613 x was shifted in do_force */
1615 /* NOTE Currently we always integrate to a temporary buffer and
1616 * then copy the results back. */
1618 wallcycle_start_nocount(wcycle, ewcUPDATE);
1620 if (md->cFREEZE != nullptr && constr != nullptr)
1622 /* If we have atoms that are frozen along some, but not all
1623 * dimensions, then any constraints will have moved them also along
1624 * the frozen dimensions. To freeze such degrees of freedom
1625 * we copy them back here to later copy them forward. It would
1626 * be more elegant and slightly more efficient to copies zero
1627 * times instead of twice, but the graph case below prevents this.
1629 const ivec *nFreeze = inputrec->opts.nFreeze;
1630 bool partialFreezeAndConstraints = false;
1631 for (int g = 0; g < inputrec->opts.ngfrz; g++)
1633 int numFreezeDim = nFreeze[g][XX] + nFreeze[g][YY] + nFreeze[g][ZZ];
1634 if (numFreezeDim > 0 && numFreezeDim < 3)
1636 partialFreezeAndConstraints = true;
1639 if (partialFreezeAndConstraints)
1641 auto xp = makeArrayRef(*upd->xp()).subArray(0, homenr);
1642 auto x = makeConstArrayRef(state->x).subArray(0, homenr);
1643 for (int i = 0; i < homenr; i++)
1645 int g = md->cFREEZE[i];
1647 for (int d = 0; d < DIM; d++)
1658 if (graph && (graph->nnodes > 0))
1660 unshift_x(graph, state->box, state->x.rvec_array(), upd->xp()->rvec_array());
1661 if (TRICLINIC(state->box))
1663 inc_nrnb(nrnb, eNR_SHIFTX, 2*graph->nnodes);
1667 inc_nrnb(nrnb, eNR_SHIFTX, graph->nnodes);
1672 auto xp = makeConstArrayRef(*upd->xp()).subArray(0, homenr);
1673 auto x = makeArrayRef(state->x).subArray(0, homenr);
1674 #ifndef __clang_analyzer__
1675 int gmx_unused nth = gmx_omp_nthreads_get(emntUpdate);
1677 #pragma omp parallel for num_threads(nth) schedule(static)
1678 for (int i = 0; i < homenr; i++)
1680 // Trivial statement, does not throw
1684 wallcycle_stop(wcycle, ewcUPDATE);
1686 /* ############# END the update of velocities and positions ######### */
1689 void update_pcouple_after_coordinates(FILE *fplog,
1691 const t_inputrec *inputrec,
1692 const t_mdatoms *md,
1693 const matrix pressure,
1694 const matrix forceVirial,
1695 const matrix constraintVirial,
1696 const matrix parrinellorahmanMu,
1702 int homenr = md->homenr;
1704 /* Cast to real for faster code, no loss in precision (see comment above) */
1705 real dt = inputrec->delta_t;
1708 /* now update boxes */
1709 switch (inputrec->epc)
1713 case (epcBERENDSEN):
1714 if (isPressureCouplingStep(step, inputrec))
1716 real dtpc = inputrec->nstpcouple*dt;
1718 berendsen_pcoupl(fplog, step, inputrec, dtpc,
1719 pressure, state->box,
1720 forceVirial, constraintVirial,
1721 mu, &state->baros_integral);
1722 berendsen_pscale(inputrec, mu, state->box, state->box_rel,
1723 start, homenr, state->x.rvec_array(),
1727 case (epcPARRINELLORAHMAN):
1728 if (isPressureCouplingStep(step, inputrec))
1730 /* The box velocities were updated in do_pr_pcoupl,
1731 * but we dont change the box vectors until we get here
1732 * since we need to be able to shift/unshift above.
1734 real dtpc = inputrec->nstpcouple*dt;
1735 for (int i = 0; i < DIM; i++)
1737 for (int m = 0; m <= i; m++)
1739 state->box[i][m] += dtpc*state->boxv[i][m];
1742 preserve_box_shape(inputrec, state->box_rel, state->box);
1744 /* Scale the coordinates */
1745 auto x = state->x.rvec_array();
1746 for (int n = start; n < start + homenr; n++)
1748 tmvmul_ur0(parrinellorahmanMu, x[n], x[n]);
1753 switch (inputrec->epct)
1755 case (epctISOTROPIC):
1756 /* DIM * eta = ln V. so DIM*eta_new = DIM*eta_old + DIM*dt*veta =>
1757 ln V_new = ln V_old + 3*dt*veta => V_new = V_old*exp(3*dt*veta) =>
1758 Side length scales as exp(veta*dt) */
1760 msmul(state->box, std::exp(state->veta*dt), state->box);
1762 /* Relate veta to boxv. veta = d(eta)/dT = (1/DIM)*1/V dV/dT.
1763 o If we assume isotropic scaling, and box length scaling
1764 factor L, then V = L^DIM (det(M)). So dV/dt = DIM
1765 L^(DIM-1) dL/dt det(M), and veta = (1/L) dL/dt. The
1766 determinant of B is L^DIM det(M), and the determinant
1767 of dB/dt is (dL/dT)^DIM det (M). veta will be
1768 (det(dB/dT)/det(B))^(1/3). Then since M =
1769 B_new*(vol_new)^(1/3), dB/dT_new = (veta_new)*B(new). */
1771 msmul(state->box, state->veta, state->boxv);
1783 auto localX = makeArrayRef(state->x).subArray(start, homenr);
1784 upd->deform()->apply(localX, state->box, step);
1788 void update_coords(int64_t step,
1789 const t_inputrec *inputrec, /* input record and box stuff */
1790 const t_mdatoms *md,
1792 gmx::ArrayRefWithPadding<gmx::RVec> f,
1793 const t_fcdata *fcd,
1794 const gmx_ekindata_t *ekind,
1798 const t_commrec *cr, /* these shouldn't be here -- need to think about it */
1799 const gmx::Constraints *constr)
1801 gmx_bool bDoConstr = (nullptr != constr);
1803 /* Running the velocity half does nothing except for velocity verlet */
1804 if ((UpdatePart == etrtVELOCITY1 || UpdatePart == etrtVELOCITY2) &&
1805 !EI_VV(inputrec->eI))
1807 gmx_incons("update_coords called for velocity without VV integrator");
1810 int homenr = md->homenr;
1812 /* Cast to real for faster code, no loss in precision (see comment above) */
1813 real dt = inputrec->delta_t;
1815 /* We need to update the NMR restraint history when time averaging is used */
1816 if (state->flags & (1<<estDISRE_RM3TAV))
1818 update_disres_history(fcd, &state->hist);
1820 if (state->flags & (1<<estORIRE_DTAV))
1822 update_orires_history(fcd, &state->hist);
1825 /* ############# START The update of velocities and positions ######### */
1826 int nth = gmx_omp_nthreads_get(emntUpdate);
1828 #pragma omp parallel for num_threads(nth) schedule(static)
1829 for (int th = 0; th < nth; th++)
1833 int start_th, end_th;
1834 getThreadAtomRange(nth, th, homenr, &start_th, &end_th);
1836 const rvec *x_rvec = state->x.rvec_array();
1837 rvec *xp_rvec = upd->xp()->rvec_array();
1838 rvec *v_rvec = state->v.rvec_array();
1839 const rvec *f_rvec = as_rvec_array(f.unpaddedArrayRef().data());
1841 switch (inputrec->eI)
1844 do_update_md(start_th, end_th, step, dt,
1845 inputrec, md, ekind, state->box,
1846 x_rvec, xp_rvec, v_rvec, f_rvec,
1847 state->nosehoover_vxi.data(), M);
1852 // With constraints, the SD update is done in 2 parts
1853 doSDUpdateGeneral<SDUpdate::ForcesOnly>
1855 start_th, end_th, dt,
1856 inputrec->opts.acc, inputrec->opts.nFreeze,
1857 md->invmass, md->ptype,
1858 md->cFREEZE, md->cACC, nullptr,
1859 x_rvec, xp_rvec, v_rvec, f_rvec,
1860 step, inputrec->ld_seed, nullptr);
1864 doSDUpdateGeneral<SDUpdate::Combined>
1866 start_th, end_th, dt,
1867 inputrec->opts.acc, inputrec->opts.nFreeze,
1868 md->invmass, md->ptype,
1869 md->cFREEZE, md->cACC, md->cTC,
1870 x_rvec, xp_rvec, v_rvec, f_rvec,
1871 step, inputrec->ld_seed,
1872 DOMAINDECOMP(cr) ? cr->dd->globalAtomIndices.data() : nullptr);
1876 do_update_bd(start_th, end_th, dt,
1877 inputrec->opts.nFreeze, md->invmass, md->ptype,
1878 md->cFREEZE, md->cTC,
1879 x_rvec, xp_rvec, v_rvec, f_rvec,
1881 upd->sd()->bd_rf.data(),
1882 step, inputrec->ld_seed, DOMAINDECOMP(cr) ? cr->dd->globalAtomIndices.data() : nullptr);
1887 gmx_bool bExtended = (inputrec->etc == etcNOSEHOOVER ||
1888 inputrec->epc == epcPARRINELLORAHMAN ||
1889 inputrec->epc == epcMTTK);
1891 /* assuming barostat coupled to group 0 */
1892 real alpha = 1.0 + DIM/static_cast<real>(inputrec->opts.nrdf[0]);
1897 do_update_vv_vel(start_th, end_th, dt,
1898 inputrec->opts.acc, inputrec->opts.nFreeze,
1899 md->invmass, md->ptype,
1900 md->cFREEZE, md->cACC,
1902 bExtended, state->veta, alpha);
1905 do_update_vv_pos(start_th, end_th, dt,
1906 inputrec->opts.nFreeze,
1907 md->ptype, md->cFREEZE,
1908 x_rvec, xp_rvec, v_rvec,
1909 bExtended, state->veta);
1915 gmx_fatal(FARGS, "Don't know how to update coordinates");
1918 GMX_CATCH_ALL_AND_EXIT_WITH_FATAL_ERROR;
1923 extern gmx_bool update_randomize_velocities(const t_inputrec *ir, int64_t step, const t_commrec *cr,
1924 const t_mdatoms *md,
1925 gmx::ArrayRef<gmx::RVec> v,
1927 const gmx::Constraints *constr)
1930 real rate = (ir->delta_t)/ir->opts.tau_t[0];
1932 if (ir->etc == etcANDERSEN && constr != nullptr)
1934 /* Currently, Andersen thermostat does not support constrained
1935 systems. Functionality exists in the andersen_tcoupl
1936 function in GROMACS 4.5.7 to allow this combination. That
1937 code could be ported to the current random-number
1938 generation approach, but has not yet been done because of
1939 lack of time and resources. */
1940 gmx_fatal(FARGS, "Normal Andersen is currently not supported with constraints, use massive Andersen instead");
1943 /* proceed with andersen if 1) it's fixed probability per
1944 particle andersen or 2) it's massive andersen and it's tau_t/dt */
1945 if ((ir->etc == etcANDERSEN) || do_per_step(step, roundToInt(1.0/rate)))
1947 andersen_tcoupl(ir, step, cr, md, v, rate,
1948 upd->sd()->randomize_group,
1949 upd->sd()->boltzfac);