2 * This file is part of the GROMACS molecular simulation package.
4 * Copyright (c) 1991-2000, University of Groningen, The Netherlands.
5 * Copyright (c) 2001-2004, The GROMACS development team.
6 * Copyright (c) 2013,2014,2015,2016,2017,2018,2019, by the GROMACS development team, led by
7 * Mark Abraham, David van der Spoel, Berk Hess, and Erik Lindahl,
8 * and including many others, as listed in the AUTHORS file in the
9 * top-level source directory and at http://www.gromacs.org.
11 * GROMACS is free software; you can redistribute it and/or
12 * modify it under the terms of the GNU Lesser General Public License
13 * as published by the Free Software Foundation; either version 2.1
14 * of the License, or (at your option) any later version.
16 * GROMACS is distributed in the hope that it will be useful,
17 * but WITHOUT ANY WARRANTY; without even the implied warranty of
18 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
19 * Lesser General Public License for more details.
21 * You should have received a copy of the GNU Lesser General Public
22 * License along with GROMACS; if not, see
23 * http://www.gnu.org/licenses, or write to the Free Software Foundation,
24 * Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA.
26 * If you want to redistribute modifications to GROMACS, please
27 * consider that scientific software is very special. Version
28 * control is crucial - bugs must be traceable. We will be happy to
29 * consider code for inclusion in the official distribution, but
30 * derived work must not be called official GROMACS. Details are found
31 * in the README & COPYING files - if they are missing, get the
32 * official version at http://www.gromacs.org.
34 * To help us fund GROMACS development, we humbly ask that you cite
35 * the research papers on the package. Check out http://www.gromacs.org.
47 #include "gromacs/domdec/domdec_struct.h"
48 #include "gromacs/fileio/confio.h"
49 #include "gromacs/gmxlib/network.h"
50 #include "gromacs/gmxlib/nrnb.h"
51 #include "gromacs/listed_forces/disre.h"
52 #include "gromacs/listed_forces/orires.h"
53 #include "gromacs/math/functions.h"
54 #include "gromacs/math/invertmatrix.h"
55 #include "gromacs/math/paddedvector.h"
56 #include "gromacs/math/units.h"
57 #include "gromacs/math/vec.h"
58 #include "gromacs/math/vecdump.h"
59 #include "gromacs/mdlib/boxdeformation.h"
60 #include "gromacs/mdlib/constr.h"
61 #include "gromacs/mdlib/gmx_omp_nthreads.h"
62 #include "gromacs/mdlib/mdatoms.h"
63 #include "gromacs/mdlib/stat.h"
64 #include "gromacs/mdlib/tgroup.h"
65 #include "gromacs/mdtypes/commrec.h"
66 #include "gromacs/mdtypes/group.h"
67 #include "gromacs/mdtypes/inputrec.h"
68 #include "gromacs/mdtypes/md_enums.h"
69 #include "gromacs/mdtypes/state.h"
70 #include "gromacs/pbcutil/boxutilities.h"
71 #include "gromacs/pbcutil/mshift.h"
72 #include "gromacs/pbcutil/pbc.h"
73 #include "gromacs/pulling/pull.h"
74 #include "gromacs/random/tabulatednormaldistribution.h"
75 #include "gromacs/random/threefry.h"
76 #include "gromacs/simd/simd.h"
77 #include "gromacs/timing/wallcycle.h"
78 #include "gromacs/topology/atoms.h"
79 #include "gromacs/utility/exceptions.h"
80 #include "gromacs/utility/fatalerror.h"
81 #include "gromacs/utility/futil.h"
82 #include "gromacs/utility/gmxassert.h"
83 #include "gromacs/utility/gmxomp.h"
84 #include "gromacs/utility/smalloc.h"
86 using namespace gmx; // TODO: Remove when this file is moved into gmx namespace
88 struct gmx_sd_const_t {
92 struct gmx_sd_sigma_t {
99 std::vector<real> bd_rf;
101 std::vector<gmx_sd_const_t> sdc;
102 std::vector<gmx_sd_sigma_t> sdsig;
103 /* andersen temperature control stuff */
104 std::vector<bool> randomize_group;
105 std::vector<real> boltzfac;
107 explicit gmx_stochd_t(const t_inputrec *ir);
110 //! pImpled implementation for Update
115 Impl(const t_inputrec *ir, BoxDeformation *boxDeformation);
118 //! stochastic dynamics struct
119 std::unique_ptr<gmx_stochd_t> sd;
120 //! xprime for constraint algorithms
121 PaddedVector<RVec> xp;
122 //! Box deformation handler (or nullptr if inactive).
123 BoxDeformation *deform = nullptr;
126 Update::Update(const t_inputrec *ir, BoxDeformation *boxDeformation)
127 : impl_(new Impl(ir, boxDeformation))
133 gmx_stochd_t* Update::sd() const
135 return impl_->sd.get();
138 PaddedVector<RVec> * Update::xp()
143 BoxDeformation * Update::deform() const
145 return impl_->deform;
148 static bool isPressureCouplingStep(int64_t step, const t_inputrec *ir)
150 GMX_ASSERT(ir->epc != epcMTTK, "MTTK pressure coupling is not handled here");
153 if (ir->epc == epcBERENDSEN)
161 /* We should only couple after a step where pressures were determined */
162 return ir->epc != etcNO &&
163 (ir->nstpcouple == 1 ||
164 do_per_step(step + ir->nstpcouple - offset, ir->nstpcouple));
167 /*! \brief Sets the velocities of virtual sites to zero */
168 static void clearVsiteVelocities(int start,
170 const unsigned short *particleType,
171 rvec * gmx_restrict v)
173 for (int a = start; a < nrend; a++)
175 if (particleType[a] == eptVSite)
182 /*! \brief Sets the number of different temperature coupling values */
183 enum class NumTempScaleValues
185 single, //!< Single T-scaling value (either one group or all values =1)
186 multiple //!< Multiple T-scaling values, need to use T-group indices
189 /*! \brief Sets if to apply no or diagonal Parrinello-Rahman pressure scaling
191 * Note that this enum is only used in updateMDLeapfrogSimple(), which does
192 * not handle fully anistropic Parrinello-Rahman scaling, so we only have
193 * options \p no and \p diagonal here and no anistropic option.
195 enum class ApplyParrinelloRahmanVScaling
197 no, //!< Do not apply velocity scaling (not a PR-coupling run or step)
198 diagonal //!< Apply velocity scaling using a diagonal matrix
201 /*! \brief Integrate using leap-frog with T-scaling and optionally diagonal Parrinello-Rahman p-coupling
203 * \tparam numTempScaleValues The number of different T-couple values
204 * \tparam applyPRVScaling Apply Parrinello-Rahman velocity scaling
205 * \param[in] start Index of first atom to update
206 * \param[in] nrend Last atom to update: \p nrend - 1
207 * \param[in] dt The time step
208 * \param[in] dtPressureCouple Time step for pressure coupling
209 * \param[in] invMassPerDim 1/mass per atom and dimension
210 * \param[in] tcstat Temperature coupling information
211 * \param[in] cTC T-coupling group index per atom
212 * \param[in] pRVScaleMatrixDiagonal Parrinello-Rahman v-scale matrix diagonal
213 * \param[in] x Input coordinates
214 * \param[out] xprime Updated coordinates
215 * \param[inout] v Velocities
216 * \param[in] f Forces
218 * We expect this template to get good SIMD acceleration by most compilers,
219 * unlike the more complex general template.
220 * Note that we might get even better SIMD acceleration when we introduce
221 * aligned (and padded) memory, possibly with some hints for the compilers.
223 template<NumTempScaleValues numTempScaleValues,
224 ApplyParrinelloRahmanVScaling applyPRVScaling>
226 updateMDLeapfrogSimple(int start,
229 real dtPressureCouple,
230 const rvec * gmx_restrict invMassPerDim,
231 gmx::ArrayRef<const t_grp_tcstat> tcstat,
232 const unsigned short * cTC,
233 const rvec pRVScaleMatrixDiagonal,
234 const rvec * gmx_restrict x,
235 rvec * gmx_restrict xprime,
236 rvec * gmx_restrict v,
237 const rvec * gmx_restrict f)
241 if (numTempScaleValues == NumTempScaleValues::single)
243 lambdaGroup = tcstat[0].lambda;
246 for (int a = start; a < nrend; a++)
248 if (numTempScaleValues == NumTempScaleValues::multiple)
250 lambdaGroup = tcstat[cTC[a]].lambda;
253 for (int d = 0; d < DIM; d++)
255 /* Note that using rvec invMassPerDim results in more efficient
256 * SIMD code, but this increases the cache pressure.
257 * For large systems with PME on the CPU this slows down the
258 * (then already slow) update by 20%. If all data remains in cache,
259 * using rvec is much faster.
261 real vNew = lambdaGroup*v[a][d] + f[a][d]*invMassPerDim[a][d]*dt;
263 if (applyPRVScaling == ApplyParrinelloRahmanVScaling::diagonal)
265 vNew -= dtPressureCouple*pRVScaleMatrixDiagonal[d]*v[a][d];
268 xprime[a][d] = x[a][d] + vNew*dt;
273 #if GMX_SIMD && GMX_SIMD_HAVE_REAL
274 #define GMX_HAVE_SIMD_UPDATE 1
276 #define GMX_HAVE_SIMD_UPDATE 0
279 #if GMX_HAVE_SIMD_UPDATE
281 /*! \brief Load (aligned) the contents of GMX_SIMD_REAL_WIDTH rvec elements sequentially into 3 SIMD registers
283 * The loaded output is:
284 * \p r0: { r[index][XX], r[index][YY], ... }
286 * \p r2: { ..., r[index+GMX_SIMD_REAL_WIDTH-1][YY], r[index+GMX_SIMD_REAL_WIDTH-1][ZZ] }
288 * \param[in] r Real to an rvec array, has to be aligned to SIMD register width
289 * \param[in] index Index of the first rvec triplet of reals to load
290 * \param[out] r0 Pointer to first SIMD register
291 * \param[out] r1 Pointer to second SIMD register
292 * \param[out] r2 Pointer to third SIMD register
294 static inline void simdLoadRvecs(const rvec *r,
300 const real *realPtr = r[index];
302 GMX_ASSERT(isSimdAligned(realPtr), "Pointer should be SIMD aligned");
304 *r0 = simdLoad(realPtr + 0*GMX_SIMD_REAL_WIDTH);
305 *r1 = simdLoad(realPtr + 1*GMX_SIMD_REAL_WIDTH);
306 *r2 = simdLoad(realPtr + 2*GMX_SIMD_REAL_WIDTH);
309 /*! \brief Store (aligned) 3 SIMD registers sequentially to GMX_SIMD_REAL_WIDTH rvec elements
311 * The stored output is:
312 * \p r[index] = { { r0[0], r0[1], ... }
314 * \p r[index+GMX_SIMD_REAL_WIDTH-1] = { ... , r2[GMX_SIMD_REAL_WIDTH-2], r2[GMX_SIMD_REAL_WIDTH-1] }
316 * \param[out] r Pointer to an rvec array, has to be aligned to SIMD register width
317 * \param[in] index Index of the first rvec triplet of reals to store to
318 * \param[in] r0 First SIMD register
319 * \param[in] r1 Second SIMD register
320 * \param[in] r2 Third SIMD register
322 static inline void simdStoreRvecs(rvec *r,
328 real *realPtr = r[index];
330 GMX_ASSERT(isSimdAligned(realPtr), "Pointer should be SIMD aligned");
332 store(realPtr + 0*GMX_SIMD_REAL_WIDTH, r0);
333 store(realPtr + 1*GMX_SIMD_REAL_WIDTH, r1);
334 store(realPtr + 2*GMX_SIMD_REAL_WIDTH, r2);
337 /*! \brief Integrate using leap-frog with single group T-scaling and SIMD
339 * \param[in] start Index of first atom to update
340 * \param[in] nrend Last atom to update: \p nrend - 1
341 * \param[in] dt The time step
342 * \param[in] invMass 1/mass per atom
343 * \param[in] tcstat Temperature coupling information
344 * \param[in] x Input coordinates
345 * \param[out] xprime Updated coordinates
346 * \param[inout] v Velocities
347 * \param[in] f Forces
350 updateMDLeapfrogSimpleSimd(int start,
353 const real * gmx_restrict invMass,
354 gmx::ArrayRef<const t_grp_tcstat> tcstat,
355 const rvec * gmx_restrict x,
356 rvec * gmx_restrict xprime,
357 rvec * gmx_restrict v,
358 const rvec * gmx_restrict f)
360 SimdReal timestep(dt);
361 SimdReal lambdaSystem(tcstat[0].lambda);
363 /* We declare variables here, since code is often slower when declaring them inside the loop */
365 /* Note: We should implement a proper PaddedVector, so we don't need this check */
366 GMX_ASSERT(isSimdAligned(invMass), "invMass should be aligned");
368 for (int a = start; a < nrend; a += GMX_SIMD_REAL_WIDTH)
370 SimdReal invMass0, invMass1, invMass2;
371 expandScalarsToTriplets(simdLoad(invMass + a),
372 &invMass0, &invMass1, &invMass2);
376 simdLoadRvecs(v, a, &v0, &v1, &v2);
377 simdLoadRvecs(f, a, &f0, &f1, &f2);
379 v0 = fma(f0*invMass0, timestep, lambdaSystem*v0);
380 v1 = fma(f1*invMass1, timestep, lambdaSystem*v1);
381 v2 = fma(f2*invMass2, timestep, lambdaSystem*v2);
383 simdStoreRvecs(v, a, v0, v1, v2);
386 simdLoadRvecs(x, a, &x0, &x1, &x2);
388 SimdReal xprime0 = fma(v0, timestep, x0);
389 SimdReal xprime1 = fma(v1, timestep, x1);
390 SimdReal xprime2 = fma(v2, timestep, x2);
392 simdStoreRvecs(xprime, a, xprime0, xprime1, xprime2);
396 #endif // GMX_HAVE_SIMD_UPDATE
398 /*! \brief Sets the NEMD acceleration type */
399 enum class AccelerationType
404 /*! \brief Integrate using leap-frog with support for everything
406 * \tparam accelerationType Type of NEMD acceleration
407 * \param[in] start Index of first atom to update
408 * \param[in] nrend Last atom to update: \p nrend - 1
409 * \param[in] doNoseHoover If to apply Nose-Hoover T-coupling
410 * \param[in] dt The time step
411 * \param[in] dtPressureCouple Time step for pressure coupling, is 0 when no pressure coupling should be applied at this step
412 * \param[in] ir The input parameter record
413 * \param[in] md Atom properties
414 * \param[in] ekind Kinetic energy data
415 * \param[in] box The box dimensions
416 * \param[in] x Input coordinates
417 * \param[out] xprime Updated coordinates
418 * \param[inout] v Velocities
419 * \param[in] f Forces
420 * \param[in] nh_vxi Nose-Hoover velocity scaling factors
421 * \param[in] M Parrinello-Rahman scaling matrix
423 template<AccelerationType accelerationType>
425 updateMDLeapfrogGeneral(int start,
429 real dtPressureCouple,
430 const t_inputrec * ir,
431 const t_mdatoms * md,
432 const gmx_ekindata_t * ekind,
434 const rvec * gmx_restrict x,
435 rvec * gmx_restrict xprime,
436 rvec * gmx_restrict v,
437 const rvec * gmx_restrict f,
438 const double * gmx_restrict nh_vxi,
441 /* This is a version of the leap-frog integrator that supports
442 * all combinations of T-coupling, P-coupling and NEMD.
443 * Nose-Hoover uses the reversible leap-frog integrator from
444 * Holian et al. Phys Rev E 52(3) : 2338, 1995
447 gmx::ArrayRef<const t_grp_tcstat> tcstat = ekind->tcstat;
448 gmx::ArrayRef<const t_grp_acc> grpstat = ekind->grpstat;
449 const unsigned short * cTC = md->cTC;
450 const unsigned short * cACC = md->cACC;
451 const rvec * accel = ir->opts.acc;
453 const rvec * gmx_restrict invMassPerDim = md->invMassPerDim;
455 /* Initialize group values, changed later when multiple groups are used */
460 real omega_Z = 2*static_cast<real>(M_PI)/box[ZZ][ZZ];
462 for (int n = start; n < nrend; n++)
468 real lg = tcstat[gt].lambda;
471 real cosineZ, vCosine;
472 #ifdef __INTEL_COMPILER
473 #pragma warning( disable : 280 )
475 switch (accelerationType)
477 case AccelerationType::none:
478 copy_rvec(v[n], vRel);
480 case AccelerationType::group:
485 /* Avoid scaling the group velocity */
486 rvec_sub(v[n], grpstat[ga].u, vRel);
488 case AccelerationType::cosine:
489 cosineZ = std::cos(x[n][ZZ]*omega_Z);
490 vCosine = cosineZ*ekind->cosacc.vcos;
491 /* Avoid scaling the cosine profile velocity */
492 copy_rvec(v[n], vRel);
499 /* Here we account for multiple time stepping, by increasing
500 * the Nose-Hoover correction by nsttcouple
502 factorNH = 0.5*ir->nsttcouple*dt*nh_vxi[gt];
505 for (int d = 0; d < DIM; d++)
508 (lg*vRel[d] + (f[n][d]*invMassPerDim[n][d]*dt - factorNH*vRel[d]
509 - dtPressureCouple*iprod(M[d], vRel)))/(1 + factorNH);
510 switch (accelerationType)
512 case AccelerationType::none:
514 case AccelerationType::group:
515 /* Add back the mean velocity and apply acceleration */
516 vNew += grpstat[ga].u[d] + accel[ga][d]*dt;
518 case AccelerationType::cosine:
521 /* Add back the mean velocity and apply acceleration */
522 vNew += vCosine + cosineZ*ekind->cosacc.cos_accel*dt;
527 xprime[n][d] = x[n][d] + vNew*dt;
532 /*! \brief Handles the Leap-frog MD x and v integration */
533 static void do_update_md(int start,
537 const t_inputrec * ir,
538 const t_mdatoms * md,
539 const gmx_ekindata_t * ekind,
541 const rvec * gmx_restrict x,
542 rvec * gmx_restrict xprime,
543 rvec * gmx_restrict v,
544 const rvec * gmx_restrict f,
545 const double * gmx_restrict nh_vxi,
548 GMX_ASSERT(nrend == start || xprime != x, "For SIMD optimization certain compilers need to have xprime != x");
549 GMX_ASSERT(ir->eI == eiMD, "Leap-frog integrator was called while another integrator was requested");
551 /* Note: Berendsen pressure scaling is handled after do_update_md() */
552 bool doTempCouple = (ir->etc != etcNO && do_per_step(step + ir->nsttcouple - 1, ir->nsttcouple));
553 bool doNoseHoover = (ir->etc == etcNOSEHOOVER && doTempCouple);
554 bool doParrinelloRahman = (ir->epc == epcPARRINELLORAHMAN && isPressureCouplingStep(step, ir));
555 bool doPROffDiagonal = (doParrinelloRahman && (M[YY][XX] != 0 || M[ZZ][XX] != 0 || M[ZZ][YY] != 0));
557 real dtPressureCouple = (doParrinelloRahman ? ir->nstpcouple*dt : 0);
559 /* NEMD (also cosine) acceleration is applied in updateMDLeapFrogGeneral */
560 bool doAcceleration = (ekind->bNEMD || ekind->cosacc.cos_accel != 0);
562 if (doNoseHoover || doPROffDiagonal || doAcceleration)
565 if (!doParrinelloRahman)
567 /* We should not apply PR scaling at this step */
577 updateMDLeapfrogGeneral<AccelerationType::none>
578 (start, nrend, doNoseHoover, dt, dtPressureCouple,
579 ir, md, ekind, box, x, xprime, v, f, nh_vxi, stepM);
581 else if (ekind->bNEMD)
583 updateMDLeapfrogGeneral<AccelerationType::group>
584 (start, nrend, doNoseHoover, dt, dtPressureCouple,
585 ir, md, ekind, box, x, xprime, v, f, nh_vxi, stepM);
589 updateMDLeapfrogGeneral<AccelerationType::cosine>
590 (start, nrend, doNoseHoover, dt, dtPressureCouple,
591 ir, md, ekind, box, x, xprime, v, f, nh_vxi, stepM);
596 /* Use a simple and thus more efficient integration loop. */
597 /* The simple loop does not check for particle type (so it can
598 * be vectorized), which means we need to clear the velocities
599 * of virtual sites in advance, when present. Note that vsite
600 * velocities are computed after update and constraints from
601 * their displacement.
605 /* Note: The overhead of this loop is completely neligible */
606 clearVsiteVelocities(start, nrend, md->ptype, v);
609 /* We determine if we have a single T-coupling lambda value for all
610 * atoms. That allows for better SIMD acceleration in the template.
611 * If we do not do temperature coupling (in the run or this step),
612 * all scaling values are 1, so we effectively have a single value.
614 bool haveSingleTempScaleValue = (!doTempCouple || ekind->ngtc == 1);
616 /* Extract some pointers needed by all cases */
617 const unsigned short *cTC = md->cTC;
618 gmx::ArrayRef<const t_grp_tcstat> tcstat = ekind->tcstat;
619 const rvec *invMassPerDim = md->invMassPerDim;
621 if (doParrinelloRahman)
623 GMX_ASSERT(!doPROffDiagonal, "updateMDLeapfrogSimple only support diagonal Parrinello-Rahman scaling matrices");
626 for (int d = 0; d < DIM; d++)
631 if (haveSingleTempScaleValue)
633 updateMDLeapfrogSimple
634 <NumTempScaleValues::single,
635 ApplyParrinelloRahmanVScaling::diagonal>
636 (start, nrend, dt, dtPressureCouple,
637 invMassPerDim, tcstat, cTC, diagM, x, xprime, v, f);
641 updateMDLeapfrogSimple
642 <NumTempScaleValues::multiple,
643 ApplyParrinelloRahmanVScaling::diagonal>
644 (start, nrend, dt, dtPressureCouple,
645 invMassPerDim, tcstat, cTC, diagM, x, xprime, v, f);
650 if (haveSingleTempScaleValue)
652 /* Note that modern compilers are pretty good at vectorizing
653 * updateMDLeapfrogSimple(). But the SIMD version will still
654 * be faster because invMass lowers the cache pressure
655 * compared to invMassPerDim.
657 #if GMX_HAVE_SIMD_UPDATE
658 /* Check if we can use invmass instead of invMassPerDim */
659 if (!md->havePartiallyFrozenAtoms)
661 updateMDLeapfrogSimpleSimd
662 (start, nrend, dt, md->invmass, tcstat, x, xprime, v, f);
667 updateMDLeapfrogSimple
668 <NumTempScaleValues::single,
669 ApplyParrinelloRahmanVScaling::no>
670 (start, nrend, dt, dtPressureCouple,
671 invMassPerDim, tcstat, cTC, nullptr, x, xprime, v, f);
676 updateMDLeapfrogSimple
677 <NumTempScaleValues::multiple,
678 ApplyParrinelloRahmanVScaling::no>
679 (start, nrend, dt, dtPressureCouple,
680 invMassPerDim, tcstat, cTC, nullptr, x, xprime, v, f);
686 static void do_update_vv_vel(int start, int nrend, real dt,
687 const rvec accel[], const ivec nFreeze[], const real invmass[],
688 const unsigned short ptype[], const unsigned short cFREEZE[],
689 const unsigned short cACC[], rvec v[], const rvec f[],
690 gmx_bool bExtended, real veta, real alpha)
698 g = 0.25*dt*veta*alpha;
700 mv2 = gmx::series_sinhx(g);
707 for (n = start; n < nrend; n++)
709 real w_dt = invmass[n]*dt;
719 for (d = 0; d < DIM; d++)
721 if ((ptype[n] != eptVSite) && (ptype[n] != eptShell) && !nFreeze[gf][d])
723 v[n][d] = mv1*(mv1*v[n][d] + 0.5*(w_dt*mv2*f[n][d]))+0.5*accel[ga][d]*dt;
731 } /* do_update_vv_vel */
733 static void do_update_vv_pos(int start, int nrend, real dt,
734 const ivec nFreeze[],
735 const unsigned short ptype[], const unsigned short cFREEZE[],
736 const rvec x[], rvec xprime[], const rvec v[],
737 gmx_bool bExtended, real veta)
743 /* Would it make more sense if Parrinello-Rahman was put here? */
748 mr2 = gmx::series_sinhx(g);
756 for (n = start; n < nrend; n++)
764 for (d = 0; d < DIM; d++)
766 if ((ptype[n] != eptVSite) && (ptype[n] != eptShell) && !nFreeze[gf][d])
768 xprime[n][d] = mr1*(mr1*x[n][d]+mr2*dt*v[n][d]);
772 xprime[n][d] = x[n][d];
776 } /* do_update_vv_pos */
778 gmx_stochd_t::gmx_stochd_t(const t_inputrec *ir)
780 const t_grpopts *opts = &ir->opts;
781 const int ngtc = opts->ngtc;
787 else if (EI_SD(ir->eI))
792 for (int gt = 0; gt < ngtc; gt++)
794 if (opts->tau_t[gt] > 0)
796 sdc[gt].em = std::exp(-ir->delta_t/opts->tau_t[gt]);
800 /* No friction and noise on this group */
805 else if (ETC_ANDERSEN(ir->etc))
807 randomize_group.resize(ngtc);
808 boltzfac.resize(ngtc);
810 /* for now, assume that all groups, if randomized, are randomized at the same rate, i.e. tau_t is the same. */
811 /* since constraint groups don't necessarily match up with temperature groups! This is checked in readir.c */
813 for (int gt = 0; gt < ngtc; gt++)
815 real reft = std::max<real>(0, opts->ref_t[gt]);
816 if ((opts->tau_t[gt] > 0) && (reft > 0)) /* tau_t or ref_t = 0 means that no randomization is done */
818 randomize_group[gt] = true;
819 boltzfac[gt] = BOLTZ*opts->ref_t[gt];
823 randomize_group[gt] = false;
829 void update_temperature_constants(gmx_stochd_t *sd, const t_inputrec *ir)
833 if (ir->bd_fric != 0)
835 for (int gt = 0; gt < ir->opts.ngtc; gt++)
837 sd->bd_rf[gt] = std::sqrt(2.0*BOLTZ*ir->opts.ref_t[gt]/(ir->bd_fric*ir->delta_t));
842 for (int gt = 0; gt < ir->opts.ngtc; gt++)
844 sd->bd_rf[gt] = std::sqrt(2.0*BOLTZ*ir->opts.ref_t[gt]);
850 for (int gt = 0; gt < ir->opts.ngtc; gt++)
852 real kT = BOLTZ*ir->opts.ref_t[gt];
853 /* The mass is accounted for later, since this differs per atom */
854 sd->sdsig[gt].V = std::sqrt(kT*(1 - sd->sdc[gt].em * sd->sdc[gt].em));
859 Update::Impl::Impl(const t_inputrec *ir, BoxDeformation *boxDeformation)
861 sd = std::make_unique<gmx_stochd_t>(ir);
862 update_temperature_constants(sd.get(), ir);
863 xp.resizeWithPadding(0);
864 deform = boxDeformation;
867 void Update::setNumAtoms(int nAtoms)
870 impl_->xp.resizeWithPadding(nAtoms);
873 /*! \brief Sets the SD update type */
874 enum class SDUpdate : int
876 ForcesOnly, FrictionAndNoiseOnly, Combined
879 /*! \brief SD integrator update
881 * Two phases are required in the general case of a constrained
882 * update, the first phase from the contribution of forces, before
883 * applying constraints, and then a second phase applying the friction
884 * and noise, and then further constraining. For details, see
887 * Without constraints, the two phases can be combined, for
890 * Thus three instantiations of this templated function will be made,
891 * two with only one contribution, and one with both contributions. */
892 template <SDUpdate updateType>
894 doSDUpdateGeneral(const gmx_stochd_t &sd,
895 int start, int nrend, real dt,
896 const rvec accel[], const ivec nFreeze[],
897 const real invmass[], const unsigned short ptype[],
898 const unsigned short cFREEZE[], const unsigned short cACC[],
899 const unsigned short cTC[],
900 const rvec x[], rvec xprime[], rvec v[], const rvec f[],
901 int64_t step, int seed, const int *gatindex)
903 // cTC, cACC and cFREEZE can be nullptr any time, but various
904 // instantiations do not make sense with particular pointer
906 if (updateType == SDUpdate::ForcesOnly)
908 GMX_ASSERT(f != nullptr, "SD update with only forces requires forces");
909 GMX_ASSERT(cTC == nullptr, "SD update with only forces cannot handle temperature groups");
911 if (updateType == SDUpdate::FrictionAndNoiseOnly)
913 GMX_ASSERT(f == nullptr, "SD update with only noise cannot handle forces");
914 GMX_ASSERT(cACC == nullptr, "SD update with only noise cannot handle acceleration groups");
916 if (updateType == SDUpdate::Combined)
918 GMX_ASSERT(f != nullptr, "SD update with forces and noise requires forces");
921 // Even 0 bits internal counter gives 2x64 ints (more than enough for three table lookups)
922 gmx::ThreeFry2x64<0> rng(seed, gmx::RandomDomain::UpdateCoordinates);
923 gmx::TabulatedNormalDistribution<real, 14> dist;
925 for (int n = start; n < nrend; n++)
927 int globalAtomIndex = gatindex ? gatindex[n] : n;
928 rng.restart(step, globalAtomIndex);
931 real inverseMass = invmass[n];
932 real invsqrtMass = std::sqrt(inverseMass);
934 int freezeGroup = cFREEZE ? cFREEZE[n] : 0;
935 int accelerationGroup = cACC ? cACC[n] : 0;
936 int temperatureGroup = cTC ? cTC[n] : 0;
938 for (int d = 0; d < DIM; d++)
940 if ((ptype[n] != eptVSite) && (ptype[n] != eptShell) && !nFreeze[freezeGroup][d])
942 if (updateType == SDUpdate::ForcesOnly)
944 real vn = v[n][d] + (inverseMass*f[n][d] + accel[accelerationGroup][d])*dt;
946 // Simple position update.
947 xprime[n][d] = x[n][d] + v[n][d]*dt;
949 else if (updateType == SDUpdate::FrictionAndNoiseOnly)
952 v[n][d] = (vn*sd.sdc[temperatureGroup].em +
953 invsqrtMass*sd.sdsig[temperatureGroup].V*dist(rng));
954 // The previous phase already updated the
955 // positions with a full v*dt term that must
956 // now be half removed.
957 xprime[n][d] = xprime[n][d] + 0.5*(v[n][d] - vn)*dt;
961 real vn = v[n][d] + (inverseMass*f[n][d] + accel[accelerationGroup][d])*dt;
962 v[n][d] = (vn*sd.sdc[temperatureGroup].em +
963 invsqrtMass*sd.sdsig[temperatureGroup].V*dist(rng));
964 // Here we include half of the friction+noise
965 // update of v into the position update.
966 xprime[n][d] = x[n][d] + 0.5*(vn + v[n][d])*dt;
971 // When using constraints, the update is split into
972 // two phases, but we only need to zero the update of
973 // virtual, shell or frozen particles in at most one
975 if (updateType != SDUpdate::FrictionAndNoiseOnly)
978 xprime[n][d] = x[n][d];
985 static void do_update_bd(int start, int nrend, real dt,
986 const ivec nFreeze[],
987 const real invmass[], const unsigned short ptype[],
988 const unsigned short cFREEZE[], const unsigned short cTC[],
989 const rvec x[], rvec xprime[], rvec v[],
990 const rvec f[], real friction_coefficient,
991 const real *rf, int64_t step, int seed,
994 /* note -- these appear to be full step velocities . . . */
999 // Use 1 bit of internal counters to give us 2*2 64-bits values per stream
1000 // Each 64-bit value is enough for 4 normal distribution table numbers.
1001 gmx::ThreeFry2x64<0> rng(seed, gmx::RandomDomain::UpdateCoordinates);
1002 gmx::TabulatedNormalDistribution<real, 14> dist;
1004 if (friction_coefficient != 0)
1006 invfr = 1.0/friction_coefficient;
1009 for (n = start; (n < nrend); n++)
1011 int ng = gatindex ? gatindex[n] : n;
1013 rng.restart(step, ng);
1024 for (d = 0; (d < DIM); d++)
1026 if ((ptype[n] != eptVSite) && (ptype[n] != eptShell) && !nFreeze[gf][d])
1028 if (friction_coefficient != 0)
1030 vn = invfr*f[n][d] + rf[gt]*dist(rng);
1034 /* NOTE: invmass = 2/(mass*friction_constant*dt) */
1035 vn = 0.5*invmass[n]*f[n][d]*dt
1036 + std::sqrt(0.5*invmass[n])*rf[gt]*dist(rng);
1040 xprime[n][d] = x[n][d]+vn*dt;
1045 xprime[n][d] = x[n][d];
1051 static void calc_ke_part_normal(const rvec v[], const t_grpopts *opts, const t_mdatoms *md,
1052 gmx_ekindata_t *ekind, t_nrnb *nrnb, gmx_bool bEkinAveVel)
1055 gmx::ArrayRef<t_grp_tcstat> tcstat = ekind->tcstat;
1056 gmx::ArrayRef<t_grp_acc> grpstat = ekind->grpstat;
1057 int nthread, thread;
1059 /* three main: VV with AveVel, vv with AveEkin, leap with AveEkin. Leap with AveVel is also
1060 an option, but not supported now.
1061 bEkinAveVel: If TRUE, we sum into ekin, if FALSE, into ekinh.
1064 /* group velocities are calculated in update_ekindata and
1065 * accumulated in acumulate_groups.
1066 * Now the partial global and groups ekin.
1068 for (g = 0; (g < opts->ngtc); g++)
1070 copy_mat(tcstat[g].ekinh, tcstat[g].ekinh_old);
1073 clear_mat(tcstat[g].ekinf);
1074 tcstat[g].ekinscalef_nhc = 1.0; /* need to clear this -- logic is complicated! */
1078 clear_mat(tcstat[g].ekinh);
1081 ekind->dekindl_old = ekind->dekindl;
1082 nthread = gmx_omp_nthreads_get(emntUpdate);
1084 #pragma omp parallel for num_threads(nthread) schedule(static)
1085 for (thread = 0; thread < nthread; thread++)
1087 // This OpenMP only loops over arrays and does not call any functions
1088 // or memory allocation. It should not be able to throw, so for now
1089 // we do not need a try/catch wrapper.
1090 int start_t, end_t, n;
1098 start_t = ((thread+0)*md->homenr)/nthread;
1099 end_t = ((thread+1)*md->homenr)/nthread;
1101 ekin_sum = ekind->ekin_work[thread];
1102 dekindl_sum = ekind->dekindl_work[thread];
1104 for (gt = 0; gt < opts->ngtc; gt++)
1106 clear_mat(ekin_sum[gt]);
1112 for (n = start_t; n < end_t; n++)
1122 hm = 0.5*md->massT[n];
1124 for (d = 0; (d < DIM); d++)
1126 v_corrt[d] = v[n][d] - grpstat[ga].u[d];
1128 for (d = 0; (d < DIM); d++)
1130 for (m = 0; (m < DIM); m++)
1132 /* if we're computing a full step velocity, v_corrt[d] has v(t). Otherwise, v(t+dt/2) */
1133 ekin_sum[gt][m][d] += hm*v_corrt[m]*v_corrt[d];
1136 if (md->nMassPerturbed && md->bPerturbed[n])
1139 0.5*(md->massB[n] - md->massA[n])*iprod(v_corrt, v_corrt);
1145 for (thread = 0; thread < nthread; thread++)
1147 for (g = 0; g < opts->ngtc; g++)
1151 m_add(tcstat[g].ekinf, ekind->ekin_work[thread][g],
1156 m_add(tcstat[g].ekinh, ekind->ekin_work[thread][g],
1161 ekind->dekindl += *ekind->dekindl_work[thread];
1164 inc_nrnb(nrnb, eNR_EKIN, md->homenr);
1167 static void calc_ke_part_visc(const matrix box, const rvec x[], const rvec v[],
1168 const t_grpopts *opts, const t_mdatoms *md,
1169 gmx_ekindata_t *ekind,
1170 t_nrnb *nrnb, gmx_bool bEkinAveVel)
1172 int start = 0, homenr = md->homenr;
1173 int g, d, n, m, gt = 0;
1176 gmx::ArrayRef<t_grp_tcstat> tcstat = ekind->tcstat;
1177 t_cos_acc *cosacc = &(ekind->cosacc);
1182 for (g = 0; g < opts->ngtc; g++)
1184 copy_mat(ekind->tcstat[g].ekinh, ekind->tcstat[g].ekinh_old);
1185 clear_mat(ekind->tcstat[g].ekinh);
1187 ekind->dekindl_old = ekind->dekindl;
1189 fac = 2*M_PI/box[ZZ][ZZ];
1192 for (n = start; n < start+homenr; n++)
1198 hm = 0.5*md->massT[n];
1200 /* Note that the times of x and v differ by half a step */
1201 /* MRS -- would have to be changed for VV */
1202 cosz = std::cos(fac*x[n][ZZ]);
1203 /* Calculate the amplitude of the new velocity profile */
1204 mvcos += 2*cosz*md->massT[n]*v[n][XX];
1206 copy_rvec(v[n], v_corrt);
1207 /* Subtract the profile for the kinetic energy */
1208 v_corrt[XX] -= cosz*cosacc->vcos;
1209 for (d = 0; (d < DIM); d++)
1211 for (m = 0; (m < DIM); m++)
1213 /* if we're computing a full step velocity, v_corrt[d] has v(t). Otherwise, v(t+dt/2) */
1216 tcstat[gt].ekinf[m][d] += hm*v_corrt[m]*v_corrt[d];
1220 tcstat[gt].ekinh[m][d] += hm*v_corrt[m]*v_corrt[d];
1224 if (md->nPerturbed && md->bPerturbed[n])
1226 /* The minus sign here might be confusing.
1227 * The kinetic contribution from dH/dl doesn't come from
1228 * d m(l)/2 v^2 / dl, but rather from d p^2/2m(l) / dl,
1229 * where p are the momenta. The difference is only a minus sign.
1231 dekindl -= 0.5*(md->massB[n] - md->massA[n])*iprod(v_corrt, v_corrt);
1234 ekind->dekindl = dekindl;
1235 cosacc->mvcos = mvcos;
1237 inc_nrnb(nrnb, eNR_EKIN, homenr);
1240 void calc_ke_part(const t_state *state, const t_grpopts *opts, const t_mdatoms *md,
1241 gmx_ekindata_t *ekind, t_nrnb *nrnb, gmx_bool bEkinAveVel)
1243 if (ekind->cosacc.cos_accel == 0)
1245 calc_ke_part_normal(state->v.rvec_array(), opts, md, ekind, nrnb, bEkinAveVel);
1249 calc_ke_part_visc(state->box, state->x.rvec_array(), state->v.rvec_array(), opts, md, ekind, nrnb, bEkinAveVel);
1253 extern void init_ekinstate(ekinstate_t *ekinstate, const t_inputrec *ir)
1255 ekinstate->ekin_n = ir->opts.ngtc;
1256 snew(ekinstate->ekinh, ekinstate->ekin_n);
1257 snew(ekinstate->ekinf, ekinstate->ekin_n);
1258 snew(ekinstate->ekinh_old, ekinstate->ekin_n);
1259 ekinstate->ekinscalef_nhc.resize(ekinstate->ekin_n);
1260 ekinstate->ekinscaleh_nhc.resize(ekinstate->ekin_n);
1261 ekinstate->vscale_nhc.resize(ekinstate->ekin_n);
1262 ekinstate->dekindl = 0;
1263 ekinstate->mvcos = 0;
1266 void update_ekinstate(ekinstate_t *ekinstate, const gmx_ekindata_t *ekind)
1270 for (i = 0; i < ekinstate->ekin_n; i++)
1272 copy_mat(ekind->tcstat[i].ekinh, ekinstate->ekinh[i]);
1273 copy_mat(ekind->tcstat[i].ekinf, ekinstate->ekinf[i]);
1274 copy_mat(ekind->tcstat[i].ekinh_old, ekinstate->ekinh_old[i]);
1275 ekinstate->ekinscalef_nhc[i] = ekind->tcstat[i].ekinscalef_nhc;
1276 ekinstate->ekinscaleh_nhc[i] = ekind->tcstat[i].ekinscaleh_nhc;
1277 ekinstate->vscale_nhc[i] = ekind->tcstat[i].vscale_nhc;
1280 copy_mat(ekind->ekin, ekinstate->ekin_total);
1281 ekinstate->dekindl = ekind->dekindl;
1282 ekinstate->mvcos = ekind->cosacc.mvcos;
1286 void restore_ekinstate_from_state(const t_commrec *cr,
1287 gmx_ekindata_t *ekind, const ekinstate_t *ekinstate)
1293 for (i = 0; i < ekinstate->ekin_n; i++)
1295 copy_mat(ekinstate->ekinh[i], ekind->tcstat[i].ekinh);
1296 copy_mat(ekinstate->ekinf[i], ekind->tcstat[i].ekinf);
1297 copy_mat(ekinstate->ekinh_old[i], ekind->tcstat[i].ekinh_old);
1298 ekind->tcstat[i].ekinscalef_nhc = ekinstate->ekinscalef_nhc[i];
1299 ekind->tcstat[i].ekinscaleh_nhc = ekinstate->ekinscaleh_nhc[i];
1300 ekind->tcstat[i].vscale_nhc = ekinstate->vscale_nhc[i];
1303 copy_mat(ekinstate->ekin_total, ekind->ekin);
1305 ekind->dekindl = ekinstate->dekindl;
1306 ekind->cosacc.mvcos = ekinstate->mvcos;
1307 n = ekinstate->ekin_n;
1312 gmx_bcast(sizeof(n), &n, cr);
1313 for (i = 0; i < n; i++)
1315 gmx_bcast(DIM*DIM*sizeof(ekind->tcstat[i].ekinh[0][0]),
1316 ekind->tcstat[i].ekinh[0], cr);
1317 gmx_bcast(DIM*DIM*sizeof(ekind->tcstat[i].ekinf[0][0]),
1318 ekind->tcstat[i].ekinf[0], cr);
1319 gmx_bcast(DIM*DIM*sizeof(ekind->tcstat[i].ekinh_old[0][0]),
1320 ekind->tcstat[i].ekinh_old[0], cr);
1322 gmx_bcast(sizeof(ekind->tcstat[i].ekinscalef_nhc),
1323 &(ekind->tcstat[i].ekinscalef_nhc), cr);
1324 gmx_bcast(sizeof(ekind->tcstat[i].ekinscaleh_nhc),
1325 &(ekind->tcstat[i].ekinscaleh_nhc), cr);
1326 gmx_bcast(sizeof(ekind->tcstat[i].vscale_nhc),
1327 &(ekind->tcstat[i].vscale_nhc), cr);
1329 gmx_bcast(DIM*DIM*sizeof(ekind->ekin[0][0]),
1330 ekind->ekin[0], cr);
1332 gmx_bcast(sizeof(ekind->dekindl), &ekind->dekindl, cr);
1333 gmx_bcast(sizeof(ekind->cosacc.mvcos), &ekind->cosacc.mvcos, cr);
1337 void update_tcouple(int64_t step,
1338 const t_inputrec *inputrec,
1340 gmx_ekindata_t *ekind,
1341 const t_extmass *MassQ,
1342 const t_mdatoms *md)
1345 // This condition was explicitly checked in previous version, but should have never been satisfied
1346 GMX_ASSERT(!(EI_VV(inputrec->eI) &&
1347 (inputrecNvtTrotter(inputrec) || inputrecNptTrotter(inputrec) || inputrecNphTrotter(inputrec))),
1348 "Temperature coupling was requested with velocity verlet and trotter");
1350 bool doTemperatureCoupling = false;
1352 // For VV temperature coupling parameters are updated on the current
1353 // step, for the others - one step before.
1354 if (inputrec->etc == etcNO)
1356 doTemperatureCoupling = false;
1358 else if (EI_VV(inputrec->eI))
1360 doTemperatureCoupling = do_per_step(step, inputrec->nsttcouple);
1364 doTemperatureCoupling = do_per_step(step + inputrec->nsttcouple - 1, inputrec->nsttcouple);
1367 if (doTemperatureCoupling)
1369 real dttc = inputrec->nsttcouple*inputrec->delta_t;
1371 //TODO: berendsen_tcoupl(...), nosehoover_tcoupl(...) and vrescale_tcoupl(...) update
1372 // temperature coupling parameters, which should be reflected in the name of these
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 // TODO: Do we have to do it on every non-t-couple step?
1400 for (int i = 0; (i < inputrec->opts.ngtc); i++)
1402 ekind->tcstat[i].lambda = 1.0;
1407 /*! \brief Computes the atom range for a thread to operate on, ensured SIMD aligned ranges
1409 * \param[in] numThreads The number of threads to divide atoms over
1410 * \param[in] threadIndex The thread to get the range for
1411 * \param[in] numAtoms The total number of atoms (on this rank)
1412 * \param[out] startAtom The start of the atom range
1413 * \param[out] endAtom The end of the atom range, note that this is in general not a multiple of the SIMD width
1415 static void getThreadAtomRange(int numThreads, int threadIndex, int numAtoms,
1416 int *startAtom, int *endAtom)
1418 #if GMX_HAVE_SIMD_UPDATE
1419 constexpr int blockSize = GMX_SIMD_REAL_WIDTH;
1421 constexpr int blockSize = 1;
1423 int numBlocks = (numAtoms + blockSize - 1)/blockSize;
1425 *startAtom = ((numBlocks* threadIndex )/numThreads)*blockSize;
1426 *endAtom = ((numBlocks*(threadIndex + 1))/numThreads)*blockSize;
1427 if (threadIndex == numThreads - 1)
1429 *endAtom = numAtoms;
1433 void update_pcouple_before_coordinates(FILE *fplog,
1435 const t_inputrec *inputrec,
1437 matrix parrinellorahmanMu,
1441 /* Berendsen P-coupling is completely handled after the coordinate update.
1442 * Trotter P-coupling is handled by separate calls to trotter_update().
1444 if (inputrec->epc == epcPARRINELLORAHMAN &&
1445 isPressureCouplingStep(step, inputrec))
1447 real dtpc = inputrec->nstpcouple*inputrec->delta_t;
1449 parrinellorahman_pcoupl(fplog, step, inputrec, dtpc, state->pres_prev,
1450 state->box, state->box_rel, state->boxv,
1451 M, parrinellorahmanMu, bInitStep);
1455 void constrain_velocities(int64_t step,
1456 real *dvdlambda, /* the contribution to be added to the bonded interactions */
1459 gmx::Constraints *constr,
1471 * APPLY CONSTRAINTS:
1478 /* clear out constraints before applying */
1479 clear_mat(vir_part);
1481 /* Constrain the coordinates upd->xp */
1482 constr->apply(do_log, do_ene,
1484 state->x.rvec_array(), state->v.rvec_array(), state->v.rvec_array(),
1486 state->lambda[efptBONDED], dvdlambda,
1487 nullptr, bCalcVir ? &vir_con : nullptr, ConstraintVariable::Velocities);
1491 m_add(vir_part, vir_con, vir_part);
1496 void constrain_coordinates(int64_t step,
1497 real *dvdlambda, /* the contribution to be added to the bonded interactions */
1501 gmx::Constraints *constr,
1514 /* clear out constraints before applying */
1515 clear_mat(vir_part);
1517 /* Constrain the coordinates upd->xp */
1518 constr->apply(do_log, do_ene,
1520 state->x.rvec_array(), upd->xp()->rvec_array(), nullptr,
1522 state->lambda[efptBONDED], dvdlambda,
1523 as_rvec_array(state->v.data()), bCalcVir ? &vir_con : nullptr, ConstraintVariable::Positions);
1527 m_add(vir_part, vir_con, vir_part);
1533 update_sd_second_half(int64_t step,
1534 real *dvdlambda, /* the contribution to be added to the bonded interactions */
1535 const t_inputrec *inputrec, /* input record and box stuff */
1536 const t_mdatoms *md,
1538 const t_commrec *cr,
1540 gmx_wallcycle_t wcycle,
1542 gmx::Constraints *constr,
1550 if (inputrec->eI == eiSD1)
1553 int homenr = md->homenr;
1555 /* Cast delta_t from double to real to make the integrators faster.
1556 * The only reason for having delta_t double is to get accurate values
1557 * for t=delta_t*step when step is larger than float precision.
1558 * For integration dt the accuracy of real suffices, since with
1559 * integral += dt*integrand the increment is nearly always (much) smaller
1560 * than the integral (and the integrand has real precision).
1562 real dt = inputrec->delta_t;
1564 wallcycle_start(wcycle, ewcUPDATE);
1566 nth = gmx_omp_nthreads_get(emntUpdate);
1568 #pragma omp parallel for num_threads(nth) schedule(static)
1569 for (th = 0; th < nth; th++)
1573 int start_th, end_th;
1574 getThreadAtomRange(nth, th, homenr, &start_th, &end_th);
1576 doSDUpdateGeneral<SDUpdate::FrictionAndNoiseOnly>
1578 start_th, end_th, dt,
1579 inputrec->opts.acc, inputrec->opts.nFreeze,
1580 md->invmass, md->ptype,
1581 md->cFREEZE, nullptr, md->cTC,
1582 state->x.rvec_array(), upd->xp()->rvec_array(),
1583 state->v.rvec_array(), nullptr,
1584 step, inputrec->ld_seed,
1585 DOMAINDECOMP(cr) ? cr->dd->globalAtomIndices.data() : nullptr);
1587 GMX_CATCH_ALL_AND_EXIT_WITH_FATAL_ERROR;
1589 inc_nrnb(nrnb, eNR_UPDATE, homenr);
1590 wallcycle_stop(wcycle, ewcUPDATE);
1592 /* Constrain the coordinates upd->xp for half a time step */
1593 constr->apply(do_log, do_ene,
1595 state->x.rvec_array(), upd->xp()->rvec_array(), nullptr,
1597 state->lambda[efptBONDED], dvdlambda,
1598 as_rvec_array(state->v.data()), nullptr, ConstraintVariable::Positions);
1602 void finish_update(const t_inputrec *inputrec, /* input record and box stuff */
1603 const t_mdatoms *md,
1605 const t_graph *graph,
1607 gmx_wallcycle_t wcycle,
1609 const gmx::Constraints *constr)
1611 int homenr = md->homenr;
1613 /* We must always unshift after updating coordinates; if we did not shake
1614 x was shifted in do_force */
1616 /* NOTE Currently we always integrate to a temporary buffer and
1617 * then copy the results back. */
1619 wallcycle_start_nocount(wcycle, ewcUPDATE);
1621 if (md->cFREEZE != nullptr && constr != nullptr)
1623 /* If we have atoms that are frozen along some, but not all
1624 * dimensions, then any constraints will have moved them also along
1625 * the frozen dimensions. To freeze such degrees of freedom
1626 * we copy them back here to later copy them forward. It would
1627 * be more elegant and slightly more efficient to copies zero
1628 * times instead of twice, but the graph case below prevents this.
1630 const ivec *nFreeze = inputrec->opts.nFreeze;
1631 bool partialFreezeAndConstraints = false;
1632 for (int g = 0; g < inputrec->opts.ngfrz; g++)
1634 int numFreezeDim = nFreeze[g][XX] + nFreeze[g][YY] + nFreeze[g][ZZ];
1635 if (numFreezeDim > 0 && numFreezeDim < 3)
1637 partialFreezeAndConstraints = true;
1640 if (partialFreezeAndConstraints)
1642 auto xp = makeArrayRef(*upd->xp()).subArray(0, homenr);
1643 auto x = makeConstArrayRef(state->x).subArray(0, homenr);
1644 for (int i = 0; i < homenr; i++)
1646 int g = md->cFREEZE[i];
1648 for (int d = 0; d < DIM; d++)
1659 if (graph && (graph->nnodes > 0))
1661 unshift_x(graph, state->box, state->x.rvec_array(), upd->xp()->rvec_array());
1662 if (TRICLINIC(state->box))
1664 inc_nrnb(nrnb, eNR_SHIFTX, 2*graph->nnodes);
1668 inc_nrnb(nrnb, eNR_SHIFTX, graph->nnodes);
1673 auto xp = makeConstArrayRef(*upd->xp()).subArray(0, homenr);
1674 auto x = makeArrayRef(state->x).subArray(0, homenr);
1675 #ifndef __clang_analyzer__
1676 int gmx_unused nth = gmx_omp_nthreads_get(emntUpdate);
1678 #pragma omp parallel for num_threads(nth) schedule(static)
1679 for (int i = 0; i < homenr; i++)
1681 // Trivial statement, does not throw
1685 wallcycle_stop(wcycle, ewcUPDATE);
1687 /* ############# END the update of velocities and positions ######### */
1690 void update_pcouple_after_coordinates(FILE *fplog,
1692 const t_inputrec *inputrec,
1693 const t_mdatoms *md,
1694 const matrix pressure,
1695 const matrix forceVirial,
1696 const matrix constraintVirial,
1697 const matrix parrinellorahmanMu,
1703 int homenr = md->homenr;
1705 /* Cast to real for faster code, no loss in precision (see comment above) */
1706 real dt = inputrec->delta_t;
1709 /* now update boxes */
1710 switch (inputrec->epc)
1714 case (epcBERENDSEN):
1715 if (isPressureCouplingStep(step, inputrec))
1717 real dtpc = inputrec->nstpcouple*dt;
1719 berendsen_pcoupl(fplog, step, inputrec, dtpc,
1720 pressure, state->box,
1721 forceVirial, constraintVirial,
1722 mu, &state->baros_integral);
1723 berendsen_pscale(inputrec, mu, state->box, state->box_rel,
1724 start, homenr, state->x.rvec_array(),
1728 case (epcPARRINELLORAHMAN):
1729 if (isPressureCouplingStep(step, inputrec))
1731 /* The box velocities were updated in do_pr_pcoupl,
1732 * but we dont change the box vectors until we get here
1733 * since we need to be able to shift/unshift above.
1735 real dtpc = inputrec->nstpcouple*dt;
1736 for (int i = 0; i < DIM; i++)
1738 for (int m = 0; m <= i; m++)
1740 state->box[i][m] += dtpc*state->boxv[i][m];
1743 preserve_box_shape(inputrec, state->box_rel, state->box);
1745 /* Scale the coordinates */
1746 auto x = state->x.rvec_array();
1747 for (int n = start; n < start + homenr; n++)
1749 tmvmul_ur0(parrinellorahmanMu, x[n], x[n]);
1754 switch (inputrec->epct)
1756 case (epctISOTROPIC):
1757 /* DIM * eta = ln V. so DIM*eta_new = DIM*eta_old + DIM*dt*veta =>
1758 ln V_new = ln V_old + 3*dt*veta => V_new = V_old*exp(3*dt*veta) =>
1759 Side length scales as exp(veta*dt) */
1761 msmul(state->box, std::exp(state->veta*dt), state->box);
1763 /* Relate veta to boxv. veta = d(eta)/dT = (1/DIM)*1/V dV/dT.
1764 o If we assume isotropic scaling, and box length scaling
1765 factor L, then V = L^DIM (det(M)). So dV/dt = DIM
1766 L^(DIM-1) dL/dt det(M), and veta = (1/L) dL/dt. The
1767 determinant of B is L^DIM det(M), and the determinant
1768 of dB/dt is (dL/dT)^DIM det (M). veta will be
1769 (det(dB/dT)/det(B))^(1/3). Then since M =
1770 B_new*(vol_new)^(1/3), dB/dT_new = (veta_new)*B(new). */
1772 msmul(state->box, state->veta, state->boxv);
1784 auto localX = makeArrayRef(state->x).subArray(start, homenr);
1785 upd->deform()->apply(localX, state->box, step);
1789 void update_coords(int64_t step,
1790 const t_inputrec *inputrec, /* input record and box stuff */
1791 const t_mdatoms *md,
1793 gmx::ArrayRefWithPadding<gmx::RVec> f,
1794 const t_fcdata *fcd,
1795 const gmx_ekindata_t *ekind,
1799 const t_commrec *cr, /* these shouldn't be here -- need to think about it */
1800 const gmx::Constraints *constr)
1802 gmx_bool bDoConstr = (nullptr != constr);
1804 /* Running the velocity half does nothing except for velocity verlet */
1805 if ((UpdatePart == etrtVELOCITY1 || UpdatePart == etrtVELOCITY2) &&
1806 !EI_VV(inputrec->eI))
1808 gmx_incons("update_coords called for velocity without VV integrator");
1811 int homenr = md->homenr;
1813 /* Cast to real for faster code, no loss in precision (see comment above) */
1814 real dt = inputrec->delta_t;
1816 /* We need to update the NMR restraint history when time averaging is used */
1817 if (state->flags & (1<<estDISRE_RM3TAV))
1819 update_disres_history(fcd, &state->hist);
1821 if (state->flags & (1<<estORIRE_DTAV))
1823 update_orires_history(fcd, &state->hist);
1826 /* ############# START The update of velocities and positions ######### */
1827 int nth = gmx_omp_nthreads_get(emntUpdate);
1829 #pragma omp parallel for num_threads(nth) schedule(static)
1830 for (int th = 0; th < nth; th++)
1834 int start_th, end_th;
1835 getThreadAtomRange(nth, th, homenr, &start_th, &end_th);
1837 const rvec *x_rvec = state->x.rvec_array();
1838 rvec *xp_rvec = upd->xp()->rvec_array();
1839 rvec *v_rvec = state->v.rvec_array();
1840 const rvec *f_rvec = as_rvec_array(f.unpaddedArrayRef().data());
1842 switch (inputrec->eI)
1845 do_update_md(start_th, end_th, step, dt,
1846 inputrec, md, ekind, state->box,
1847 x_rvec, xp_rvec, v_rvec, f_rvec,
1848 state->nosehoover_vxi.data(), M);
1853 // With constraints, the SD update is done in 2 parts
1854 doSDUpdateGeneral<SDUpdate::ForcesOnly>
1856 start_th, end_th, dt,
1857 inputrec->opts.acc, inputrec->opts.nFreeze,
1858 md->invmass, md->ptype,
1859 md->cFREEZE, md->cACC, nullptr,
1860 x_rvec, xp_rvec, v_rvec, f_rvec,
1861 step, inputrec->ld_seed, nullptr);
1865 doSDUpdateGeneral<SDUpdate::Combined>
1867 start_th, end_th, dt,
1868 inputrec->opts.acc, inputrec->opts.nFreeze,
1869 md->invmass, md->ptype,
1870 md->cFREEZE, md->cACC, md->cTC,
1871 x_rvec, xp_rvec, v_rvec, f_rvec,
1872 step, inputrec->ld_seed,
1873 DOMAINDECOMP(cr) ? cr->dd->globalAtomIndices.data() : nullptr);
1877 do_update_bd(start_th, end_th, dt,
1878 inputrec->opts.nFreeze, md->invmass, md->ptype,
1879 md->cFREEZE, md->cTC,
1880 x_rvec, xp_rvec, v_rvec, f_rvec,
1882 upd->sd()->bd_rf.data(),
1883 step, inputrec->ld_seed, DOMAINDECOMP(cr) ? cr->dd->globalAtomIndices.data() : nullptr);
1888 gmx_bool bExtended = (inputrec->etc == etcNOSEHOOVER ||
1889 inputrec->epc == epcPARRINELLORAHMAN ||
1890 inputrec->epc == epcMTTK);
1892 /* assuming barostat coupled to group 0 */
1893 real alpha = 1.0 + DIM/static_cast<real>(inputrec->opts.nrdf[0]);
1898 do_update_vv_vel(start_th, end_th, dt,
1899 inputrec->opts.acc, inputrec->opts.nFreeze,
1900 md->invmass, md->ptype,
1901 md->cFREEZE, md->cACC,
1903 bExtended, state->veta, alpha);
1906 do_update_vv_pos(start_th, end_th, dt,
1907 inputrec->opts.nFreeze,
1908 md->ptype, md->cFREEZE,
1909 x_rvec, xp_rvec, v_rvec,
1910 bExtended, state->veta);
1916 gmx_fatal(FARGS, "Don't know how to update coordinates");
1919 GMX_CATCH_ALL_AND_EXIT_WITH_FATAL_ERROR;
1924 extern gmx_bool update_randomize_velocities(const t_inputrec *ir, int64_t step, const t_commrec *cr,
1925 const t_mdatoms *md,
1926 gmx::ArrayRef<gmx::RVec> v,
1928 const gmx::Constraints *constr)
1931 real rate = (ir->delta_t)/ir->opts.tau_t[0];
1933 if (ir->etc == etcANDERSEN && constr != nullptr)
1935 /* Currently, Andersen thermostat does not support constrained
1936 systems. Functionality exists in the andersen_tcoupl
1937 function in GROMACS 4.5.7 to allow this combination. That
1938 code could be ported to the current random-number
1939 generation approach, but has not yet been done because of
1940 lack of time and resources. */
1941 gmx_fatal(FARGS, "Normal Andersen is currently not supported with constraints, use massive Andersen instead");
1944 /* proceed with andersen if 1) it's fixed probability per
1945 particle andersen or 2) it's massive andersen and it's tau_t/dt */
1946 if ((ir->etc == etcANDERSEN) || do_per_step(step, roundToInt(1.0/rate)))
1948 andersen_tcoupl(ir, step, cr, md, v, rate,
1949 upd->sd()->randomize_group,
1950 upd->sd()->boltzfac);