2 * This file is part of the GROMACS molecular simulation package.
4 * Copyright (c) 1991-2000, University of Groningen, The Netherlands.
5 * Copyright (c) 2001-2004, The GROMACS development team.
6 * Copyright (c) 2013,2014,2015,2016,2017 by the GROMACS development team.
7 * Copyright (c) 2018,2019,2020,2021, by the GROMACS development team, led by
8 * Mark Abraham, David van der Spoel, Berk Hess, and Erik Lindahl,
9 * and including many others, as listed in the AUTHORS file in the
10 * top-level source directory and at http://www.gromacs.org.
12 * GROMACS is free software; you can redistribute it and/or
13 * modify it under the terms of the GNU Lesser General Public License
14 * as published by the Free Software Foundation; either version 2.1
15 * of the License, or (at your option) any later version.
17 * GROMACS is distributed in the hope that it will be useful,
18 * but WITHOUT ANY WARRANTY; without even the implied warranty of
19 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
20 * Lesser General Public License for more details.
22 * You should have received a copy of the GNU Lesser General Public
23 * License along with GROMACS; if not, see
24 * http://www.gnu.org/licenses, or write to the Free Software Foundation,
25 * Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA.
27 * If you want to redistribute modifications to GROMACS, please
28 * consider that scientific software is very special. Version
29 * control is crucial - bugs must be traceable. We will be happy to
30 * consider code for inclusion in the official distribution, but
31 * derived work must not be called official GROMACS. Details are found
32 * in the README & COPYING files - if they are missing, get the
33 * official version at http://www.gromacs.org.
35 * To help us fund GROMACS development, we humbly ask that you cite
36 * the research papers on the package. Check out http://www.gromacs.org.
39 * \brief Defines LINCS code.
41 * \author Berk Hess <hess@kth.se>
42 * \author Mark Abraham <mark.j.abraham@gmail.com>
43 * \ingroup module_mdlib
59 #include "gromacs/domdec/domdec.h"
60 #include "gromacs/domdec/domdec_struct.h"
61 #include "gromacs/gmxlib/nrnb.h"
62 #include "gromacs/math/functions.h"
63 #include "gromacs/math/paddedvector.h"
64 #include "gromacs/math/units.h"
65 #include "gromacs/math/vec.h"
66 #include "gromacs/mdlib/constr.h"
67 #include "gromacs/mdlib/gmx_omp_nthreads.h"
68 #include "gromacs/mdrunutility/multisim.h"
69 #include "gromacs/mdtypes/commrec.h"
70 #include "gromacs/mdtypes/inputrec.h"
71 #include "gromacs/mdtypes/md_enums.h"
72 #include "gromacs/mdtypes/observablesreducer.h"
73 #include "gromacs/pbcutil/pbc.h"
74 #include "gromacs/pbcutil/pbc_simd.h"
75 #include "gromacs/simd/simd.h"
76 #include "gromacs/simd/simd_math.h"
77 #include "gromacs/simd/vector_operations.h"
78 #include "gromacs/topology/mtop_util.h"
79 #include "gromacs/utility/alignedallocator.h"
80 #include "gromacs/utility/arrayref.h"
81 #include "gromacs/utility/basedefinitions.h"
82 #include "gromacs/utility/bitmask.h"
83 #include "gromacs/utility/cstringutil.h"
84 #include "gromacs/utility/exceptions.h"
85 #include "gromacs/utility/fatalerror.h"
86 #include "gromacs/utility/gmxomp.h"
87 #include "gromacs/utility/listoflists.h"
88 #include "gromacs/utility/pleasecite.h"
93 //! Indices of the two atoms involved in a single constraint
96 //! \brief Constructor, does not initialize to catch bugs and faster construction
105 //! Unit of work within LINCS.
108 //! First constraint for this task.
110 //! b1-1 is the last constraint for this task.
112 //! The number of constraints in triangles.
114 //! The list of triangle constraints.
115 std::vector<int> triangle;
116 //! The bits tell if the matrix element should be used.
117 std::vector<int> tri_bits;
118 //! Constraint index for updating atom data.
119 std::vector<int> ind;
120 //! Constraint index for updating atom data.
121 std::vector<int> ind_r;
122 //! Temporary variable for virial calculation.
123 tensor vir_r_m_dr = { { 0 } };
124 //! Temporary variable for lambda derivative.
128 /*! \brief Data for LINCS algorithm.
133 //! The global number of constraints.
135 //! The global number of flexible constraints.
137 //! The global number of constraints in triangles.
138 int ncg_triangle = 0;
139 //! The number of iterations.
141 //! The order of the matrix expansion.
143 //! The maximum number of constraints connected to a single atom.
146 //! The number of real constraints.
148 //! The number of constraints including padding for SIMD.
150 //! The number of constraint connections.
152 //! The FE lambda value used for filling blc and blmf.
154 //! mapping from topology to LINCS constraints.
155 std::vector<int> con_index;
156 //! The reference distance in topology A.
157 std::vector<real, AlignedAllocator<real>> bllen0;
158 //! The reference distance in top B - the r.d. in top A.
159 std::vector<real, AlignedAllocator<real>> ddist;
160 //! The atom pairs involved in the constraints.
161 std::vector<AtomPair> atoms;
162 //! 1/sqrt(invmass1 invmass2).
163 std::vector<real, AlignedAllocator<real>> blc;
164 //! As blc, but with all masses 1.
165 std::vector<real, AlignedAllocator<real>> blc1;
166 //! Index into blbnb and blmf.
167 std::vector<int> blnr;
168 //! List of constraint connections.
169 std::vector<int> blbnb;
170 //! The local number of constraints in triangles.
172 //! The number of constraint connections in triangles.
173 int ncc_triangle = 0;
174 //! Communicate before each LINCS interation.
175 bool bCommIter = false;
176 //! Matrix of mass factors for constraint connections.
177 std::vector<real> blmf;
178 //! As blmf, but with all masses 1.
179 std::vector<real> blmf1;
180 //! The reference bond length.
181 std::vector<real, AlignedAllocator<real>> bllen;
182 //! The local atom count per constraint, can be NULL.
183 std::vector<int> nlocat;
185 /*! \brief The number of tasks used for LINCS work.
187 * \todo This is mostly used to loop over \c task, which would
188 * be nicer to do with range-based for loops, but the thread
189 * index is used for constructing bit masks and organizing the
190 * virial output buffer, so other things need to change,
193 /*! \brief LINCS thread division */
194 std::vector<Task> task;
195 //! Atom flags for thread parallelization.
196 std::vector<gmx_bitmask_t> atf;
197 //! Are the LINCS tasks interdependent?
198 bool bTaskDep = false;
199 //! Are there triangle constraints that cross task borders?
200 bool bTaskDepTri = false;
201 //! Arrays for temporary storage in the LINCS algorithm.
203 PaddedVector<gmx::RVec> tmpv;
204 std::vector<real> tmpncc;
205 std::vector<real, AlignedAllocator<real>> tmp1;
206 std::vector<real, AlignedAllocator<real>> tmp2;
207 std::vector<real, AlignedAllocator<real>> tmp3;
208 std::vector<real, AlignedAllocator<real>> tmp4;
210 //! The Lagrange multipliers times -1.
211 std::vector<real, AlignedAllocator<real>> mlambda;
212 /*! \brief Callback used after constraining to require reduction
213 * of values later used to compute the constraint RMS relative
214 * deviation, so the latter can be output. */
215 std::optional<ObservablesReducerBuilder::CallbackToRequireReduction> callbackToRequireReduction;
216 /*! \brief View used for reducing the components of the global
217 * relative RMS constraint deviation.
219 * Can be written any time, but that is only useful when followed
220 * by a call of the callbackToRequireReduction. Useful to read
221 * only from the callback that the ObservablesReducer will later
222 * make after reduction. */
223 ArrayRef<double> rmsdReductionBuffer;
224 /*! \brief The value of the constraint RMS deviation after it has
227 * When DD is active, filled by the ObservablesReducer, otherwise
228 * filled directly here. */
229 std::optional<double> constraintRmsDeviation;
232 /*! \brief Define simd_width for memory allocation used for SIMD code */
233 #if GMX_SIMD_HAVE_REAL
234 static const int simd_width = GMX_SIMD_REAL_WIDTH;
236 static const int simd_width = 1;
239 real lincs_rmsd(const Lincs* lincsd)
241 if (lincsd->constraintRmsDeviation.has_value())
243 return real(lincsd->constraintRmsDeviation.value());
251 /*! \brief Do a set of nrec LINCS matrix multiplications.
253 * This function will return with up to date thread-local
254 * constraint data, without an OpenMP barrier.
256 static void lincs_matrix_expand(const Lincs& lincsd,
258 gmx::ArrayRef<const real> blcc,
259 gmx::ArrayRef<real> rhs1,
260 gmx::ArrayRef<real> rhs2,
261 gmx::ArrayRef<real> sol)
263 gmx::ArrayRef<const int> blnr = lincsd.blnr;
264 gmx::ArrayRef<const int> blbnb = lincsd.blbnb;
266 const int b0 = li_task.b0;
267 const int b1 = li_task.b1;
268 const int nrec = lincsd.nOrder;
270 for (int rec = 0; rec < nrec; rec++)
276 for (int b = b0; b < b1; b++)
282 for (n = blnr[b]; n < blnr[b + 1]; n++)
284 mvb = mvb + blcc[n] * rhs1[blbnb[n]];
287 sol[b] = sol[b] + mvb;
290 std::swap(rhs1, rhs2);
291 } /* nrec*(ncons+2*nrtot) flops */
293 if (lincsd.ntriangle > 0)
295 /* Perform an extra nrec recursions for only the constraints
296 * involved in rigid triangles.
297 * In this way their accuracy should come close to those of the other
298 * constraints, since traingles of constraints can produce eigenvalues
299 * around 0.7, while the effective eigenvalue for bond constraints
300 * is around 0.4 (and 0.7*0.7=0.5).
305 /* We need a barrier here, since other threads might still be
306 * reading the contents of rhs1 and/o rhs2.
307 * We could avoid this barrier by introducing two extra rhs
308 * arrays for the triangle constraints only.
313 /* Constraints involved in a triangle are ensured to be in the same
314 * LINCS task. This means no barriers are required during the extra
315 * iterations for the triangle constraints.
317 gmx::ArrayRef<const int> triangle = li_task.triangle;
318 gmx::ArrayRef<const int> tri_bits = li_task.tri_bits;
320 for (int rec = 0; rec < nrec; rec++)
322 for (int tb = 0; tb < li_task.ntriangle; tb++)
324 int b, bits, nr0, nr1, n;
332 for (n = nr0; n < nr1; n++)
334 if (bits & (1 << (n - nr0)))
336 mvb = mvb + blcc[n] * rhs1[blbnb[n]];
340 sol[b] = sol[b] + mvb;
343 std::swap(rhs1, rhs2);
344 } /* nrec*(ntriangle + ncc_triangle*2) flops */
346 if (lincsd.bTaskDepTri)
348 /* The constraints triangles are decoupled from each other,
349 * but constraints in one triangle cross thread task borders.
350 * We could probably avoid this with more advanced setup code.
357 //! Update atomic coordinates when an index is not required.
358 static void lincs_update_atoms_noind(int ncons,
359 gmx::ArrayRef<const AtomPair> atoms,
361 gmx::ArrayRef<const real> fac,
362 gmx::ArrayRef<const gmx::RVec> r,
363 gmx::ArrayRef<const real> invmass,
366 if (!invmass.empty())
368 for (int b = 0; b < ncons; b++)
370 int i = atoms[b].index1;
371 int j = atoms[b].index2;
372 real mvb = preFactor * fac[b];
373 real im1 = invmass[i];
374 real im2 = invmass[j];
375 real tmp0 = r[b][0] * mvb;
376 real tmp1 = r[b][1] * mvb;
377 real tmp2 = r[b][2] * mvb;
378 x[i][0] -= tmp0 * im1;
379 x[i][1] -= tmp1 * im1;
380 x[i][2] -= tmp2 * im1;
381 x[j][0] += tmp0 * im2;
382 x[j][1] += tmp1 * im2;
383 x[j][2] += tmp2 * im2;
384 } /* 16 ncons flops */
388 for (int b = 0; b < ncons; b++)
390 int i = atoms[b].index1;
391 int j = atoms[b].index2;
392 real mvb = preFactor * fac[b];
393 real tmp0 = r[b][0] * mvb;
394 real tmp1 = r[b][1] * mvb;
395 real tmp2 = r[b][2] * mvb;
406 //! Update atomic coordinates when an index is required.
407 static void lincs_update_atoms_ind(gmx::ArrayRef<const int> ind,
408 gmx::ArrayRef<const AtomPair> atoms,
410 gmx::ArrayRef<const real> fac,
411 gmx::ArrayRef<const gmx::RVec> r,
412 gmx::ArrayRef<const real> invmass,
415 if (!invmass.empty())
419 int i = atoms[b].index1;
420 int j = atoms[b].index2;
421 real mvb = preFactor * fac[b];
422 real im1 = invmass[i];
423 real im2 = invmass[j];
424 real tmp0 = r[b][0] * mvb;
425 real tmp1 = r[b][1] * mvb;
426 real tmp2 = r[b][2] * mvb;
427 x[i][0] -= tmp0 * im1;
428 x[i][1] -= tmp1 * im1;
429 x[i][2] -= tmp2 * im1;
430 x[j][0] += tmp0 * im2;
431 x[j][1] += tmp1 * im2;
432 x[j][2] += tmp2 * im2;
433 } /* 16 ncons flops */
439 int i = atoms[b].index1;
440 int j = atoms[b].index2;
441 real mvb = preFactor * fac[b];
442 real tmp0 = r[b][0] * mvb;
443 real tmp1 = r[b][1] * mvb;
444 real tmp2 = r[b][2] * mvb;
451 } /* 16 ncons flops */
455 //! Update coordinates for atoms.
456 static void lincs_update_atoms(Lincs* li,
459 gmx::ArrayRef<const real> fac,
460 gmx::ArrayRef<const gmx::RVec> r,
461 gmx::ArrayRef<const real> invmass,
466 /* Single thread, we simply update for all constraints */
467 lincs_update_atoms_noind(li->nc_real, li->atoms, preFactor, fac, r, invmass, x);
471 /* Update the atom vector components for our thread local
472 * constraints that only access our local atom range.
473 * This can be done without a barrier.
475 lincs_update_atoms_ind(li->task[th].ind, li->atoms, preFactor, fac, r, invmass, x);
477 if (!li->task[li->ntask].ind.empty())
479 /* Update the constraints that operate on atoms
480 * in multiple thread atom blocks on the master thread.
485 lincs_update_atoms_ind(li->task[li->ntask].ind, li->atoms, preFactor, fac, r, invmass, x);
491 #if GMX_SIMD_HAVE_REAL
492 //! Helper function so that we can run TSAN with SIMD support (where implemented).
494 static inline void gmx_simdcall gatherLoadUTransposeTSANSafe(const real* base,
495 const std::int32_t* offset,
500 # if (CMAKE_BUILD_TYPE == CMAKE_BUILD_TYPE_TSAN) && GMX_SIMD_X86_AVX2_256
501 // This function is only implemented in this case
502 gatherLoadUTransposeSafe<align>(base, offset, v0, v1, v2);
504 gatherLoadUTranspose<align>(base, offset, v0, v1, v2);
508 /*! \brief Calculate the constraint distance vectors r to project on from x.
510 * Determine the right-hand side of the matrix equation using quantity f.
511 * This function only differs from calc_dr_x_xp_simd below in that
512 * no constraint length is subtracted and no PBC is used for f. */
513 static void gmx_simdcall calc_dr_x_f_simd(int b0,
515 gmx::ArrayRef<const AtomPair> atoms,
516 const rvec* gmx_restrict x,
517 const rvec* gmx_restrict f,
518 const real* gmx_restrict blc,
519 const real* pbc_simd,
520 rvec* gmx_restrict r,
521 real* gmx_restrict rhs,
522 real* gmx_restrict sol)
524 assert(b0 % GMX_SIMD_REAL_WIDTH == 0);
526 alignas(GMX_SIMD_ALIGNMENT) std::int32_t offset2[GMX_SIMD_REAL_WIDTH];
528 for (int i = 0; i < GMX_SIMD_REAL_WIDTH; i++)
533 for (int bs = b0; bs < b1; bs += GMX_SIMD_REAL_WIDTH)
535 SimdReal x0_S, y0_S, z0_S;
536 SimdReal x1_S, y1_S, z1_S;
537 SimdReal rx_S, ry_S, rz_S, n2_S, il_S;
538 SimdReal fx_S, fy_S, fz_S, ip_S, rhs_S;
539 alignas(GMX_SIMD_ALIGNMENT) std::int32_t offset0[GMX_SIMD_REAL_WIDTH];
540 alignas(GMX_SIMD_ALIGNMENT) std::int32_t offset1[GMX_SIMD_REAL_WIDTH];
542 for (int i = 0; i < GMX_SIMD_REAL_WIDTH; i++)
544 offset0[i] = atoms[bs + i].index1;
545 offset1[i] = atoms[bs + i].index2;
548 gatherLoadUTransposeTSANSafe<3>(reinterpret_cast<const real*>(x), offset0, &x0_S, &y0_S, &z0_S);
549 gatherLoadUTransposeTSANSafe<3>(reinterpret_cast<const real*>(x), offset1, &x1_S, &y1_S, &z1_S);
554 pbc_correct_dx_simd(&rx_S, &ry_S, &rz_S, pbc_simd);
556 n2_S = norm2(rx_S, ry_S, rz_S);
557 il_S = invsqrt(n2_S);
563 transposeScatterStoreU<3>(reinterpret_cast<real*>(r + bs), offset2, rx_S, ry_S, rz_S);
565 gatherLoadUTransposeTSANSafe<3>(reinterpret_cast<const real*>(f), offset0, &x0_S, &y0_S, &z0_S);
566 gatherLoadUTransposeTSANSafe<3>(reinterpret_cast<const real*>(f), offset1, &x1_S, &y1_S, &z1_S);
571 ip_S = iprod(rx_S, ry_S, rz_S, fx_S, fy_S, fz_S);
573 rhs_S = load<SimdReal>(blc + bs) * ip_S;
575 store(rhs + bs, rhs_S);
576 store(sol + bs, rhs_S);
579 #endif // GMX_SIMD_HAVE_REAL
581 /*! \brief LINCS projection, works on derivatives of the coordinates. */
582 static void do_lincsp(ArrayRefWithPadding<const RVec> xPadded,
583 ArrayRefWithPadding<RVec> fPadded,
588 ArrayRef<const real> invmass,
589 ConstraintVariable econq,
594 const int b0 = lincsd->task[th].b0;
595 const int b1 = lincsd->task[th].b1;
597 gmx::ArrayRef<const AtomPair> atoms = lincsd->atoms;
598 gmx::ArrayRef<gmx::RVec> r = lincsd->tmpv;
599 gmx::ArrayRef<const int> blnr = lincsd->blnr;
600 gmx::ArrayRef<const int> blbnb = lincsd->blbnb;
602 gmx::ArrayRef<const real> blc;
603 gmx::ArrayRef<const real> blmf;
604 if (econq != ConstraintVariable::Force)
606 /* Use mass-weighted parameters */
612 /* Use non mass-weighted parameters */
614 blmf = lincsd->blmf1;
616 gmx::ArrayRef<real> blcc = lincsd->tmpncc;
617 gmx::ArrayRef<real> rhs1 = lincsd->tmp1;
618 gmx::ArrayRef<real> rhs2 = lincsd->tmp2;
619 gmx::ArrayRef<real> sol = lincsd->tmp3;
621 const rvec* x = as_rvec_array(xPadded.paddedArrayRef().data());
622 rvec* f = as_rvec_array(fPadded.paddedArrayRef().data());
624 #if GMX_SIMD_HAVE_REAL
625 /* This SIMD code does the same as the plain-C code after the #else.
626 * The only difference is that we always call pbc code, as with SIMD
627 * the overhead of pbc computation (when not needed) is small.
629 alignas(GMX_SIMD_ALIGNMENT) real pbc_simd[9 * GMX_SIMD_REAL_WIDTH];
631 /* Convert the pbc struct for SIMD */
632 set_pbc_simd(pbc, pbc_simd);
634 /* Compute normalized x i-j vectors, store in r.
635 * Compute the inner product of r and xp i-j and store in rhs1.
638 b0, b1, atoms, x, f, blc.data(), pbc_simd, as_rvec_array(r.data()), rhs1.data(), sol.data());
640 #else // GMX_SIMD_HAVE_REAL
642 /* Compute normalized i-j vectors */
645 for (int b = b0; b < b1; b++)
649 pbc_dx_aiuc(pbc, x[atoms[b].index1], x[atoms[b].index2], dx);
655 for (int b = b0; b < b1; b++)
659 rvec_sub(x[atoms[b].index1], x[atoms[b].index2], dx);
661 } /* 16 ncons flops */
664 for (int b = b0; b < b1; b++)
666 int i = atoms[b].index1;
667 int j = atoms[b].index2;
669 * (r[b][0] * (f[i][0] - f[j][0]) + r[b][1] * (f[i][1] - f[j][1])
670 + r[b][2] * (f[i][2] - f[j][2]));
676 #endif // GMX_SIMD_HAVE_REAL
678 if (lincsd->bTaskDep)
680 /* We need a barrier, since the matrix construction below
681 * can access entries in r of other threads.
686 /* Construct the (sparse) LINCS matrix */
687 for (int b = b0; b < b1; b++)
689 for (int n = blnr[b]; n < blnr[b + 1]; n++)
691 blcc[n] = blmf[n] * ::iprod(r[b], r[blbnb[n]]);
694 /* Together: 23*ncons + 6*nrtot flops */
696 lincs_matrix_expand(*lincsd, lincsd->task[th], blcc, rhs1, rhs2, sol);
697 /* nrec*(ncons+2*nrtot) flops */
699 if (econq == ConstraintVariable::Deriv_FlexCon)
701 /* We only want to constraint the flexible constraints,
702 * so we mask out the normal ones by setting sol to 0.
704 for (int b = b0; b < b1; b++)
706 if (!(lincsd->bllen0[b] == 0 && lincsd->ddist[b] == 0))
713 /* We multiply sol by blc, so we can use lincs_update_atoms for OpenMP */
714 for (int b = b0; b < b1; b++)
719 /* When constraining forces, we should not use mass weighting,
720 * so we pass invmass=NULL, which results in the use of 1 for all atoms.
722 lincs_update_atoms(lincsd,
727 (econq != ConstraintVariable::Force) ? invmass : gmx::ArrayRef<real>(),
728 as_rvec_array(fp.data()));
733 for (int b = b0; b < b1; b++)
735 dhdlambda -= sol[b] * lincsd->ddist[b];
738 lincsd->task[th].dhdlambda = dhdlambda;
743 /* Constraint virial,
744 * determines sum r_bond x delta f,
745 * where delta f is the constraint correction
746 * of the quantity that is being constrained.
748 for (int b = b0; b < b1; b++)
750 const real mvb = lincsd->bllen[b] * sol[b];
751 for (int i = 0; i < DIM; i++)
753 const real tmp1 = mvb * r[b][i];
754 for (int j = 0; j < DIM; j++)
756 rmdf[i][j] += tmp1 * r[b][j];
759 } /* 23 ncons flops */
763 #if GMX_SIMD_HAVE_REAL
765 /*! \brief Calculate the constraint distance vectors r to project on from x.
767 * Determine the right-hand side of the matrix equation using coordinates xp. */
768 static void gmx_simdcall calc_dr_x_xp_simd(int b0,
770 gmx::ArrayRef<const AtomPair> atoms,
771 const rvec* gmx_restrict x,
772 const rvec* gmx_restrict xp,
773 const real* gmx_restrict bllen,
774 const real* gmx_restrict blc,
775 const real* pbc_simd,
776 rvec* gmx_restrict r,
777 real* gmx_restrict rhs,
778 real* gmx_restrict sol)
780 assert(b0 % GMX_SIMD_REAL_WIDTH == 0);
781 alignas(GMX_SIMD_ALIGNMENT) std::int32_t offset2[GMX_SIMD_REAL_WIDTH];
783 for (int i = 0; i < GMX_SIMD_REAL_WIDTH; i++)
788 for (int bs = b0; bs < b1; bs += GMX_SIMD_REAL_WIDTH)
790 SimdReal x0_S, y0_S, z0_S;
791 SimdReal x1_S, y1_S, z1_S;
792 SimdReal rx_S, ry_S, rz_S, n2_S, il_S;
793 SimdReal rxp_S, ryp_S, rzp_S, ip_S, rhs_S;
794 alignas(GMX_SIMD_ALIGNMENT) std::int32_t offset0[GMX_SIMD_REAL_WIDTH];
795 alignas(GMX_SIMD_ALIGNMENT) std::int32_t offset1[GMX_SIMD_REAL_WIDTH];
797 for (int i = 0; i < GMX_SIMD_REAL_WIDTH; i++)
799 offset0[i] = atoms[bs + i].index1;
800 offset1[i] = atoms[bs + i].index2;
803 gatherLoadUTransposeTSANSafe<3>(reinterpret_cast<const real*>(x), offset0, &x0_S, &y0_S, &z0_S);
804 gatherLoadUTransposeTSANSafe<3>(reinterpret_cast<const real*>(x), offset1, &x1_S, &y1_S, &z1_S);
809 pbc_correct_dx_simd(&rx_S, &ry_S, &rz_S, pbc_simd);
811 n2_S = norm2(rx_S, ry_S, rz_S);
812 il_S = invsqrt(n2_S);
818 transposeScatterStoreU<3>(reinterpret_cast<real*>(r + bs), offset2, rx_S, ry_S, rz_S);
820 gatherLoadUTransposeTSANSafe<3>(reinterpret_cast<const real*>(xp), offset0, &x0_S, &y0_S, &z0_S);
821 gatherLoadUTransposeTSANSafe<3>(reinterpret_cast<const real*>(xp), offset1, &x1_S, &y1_S, &z1_S);
827 pbc_correct_dx_simd(&rxp_S, &ryp_S, &rzp_S, pbc_simd);
829 ip_S = iprod(rx_S, ry_S, rz_S, rxp_S, ryp_S, rzp_S);
831 rhs_S = load<SimdReal>(blc + bs) * (ip_S - load<SimdReal>(bllen + bs));
833 store(rhs + bs, rhs_S);
834 store(sol + bs, rhs_S);
837 #endif // GMX_SIMD_HAVE_REAL
839 /*! \brief Determine the distances and right-hand side for the next iteration. */
840 gmx_unused static void calc_dist_iter(int b0,
842 gmx::ArrayRef<const AtomPair> atoms,
843 const rvec* gmx_restrict xp,
844 const real* gmx_restrict bllen,
845 const real* gmx_restrict blc,
848 real* gmx_restrict rhs,
849 real* gmx_restrict sol,
852 for (int b = b0; b < b1; b++)
858 pbc_dx_aiuc(pbc, xp[atoms[b].index1], xp[atoms[b].index2], dx);
862 rvec_sub(xp[atoms[b].index1], xp[atoms[b].index2], dx);
864 real len2 = len * len;
865 real dlen2 = 2 * len2 - ::norm2(dx);
866 if (dlen2 < wfac * len2)
868 /* not race free - see detailed comment in caller */
874 mvb = blc[b] * (len - dlen2 * gmx::invsqrt(dlen2));
882 } /* 20*ncons flops */
885 #if GMX_SIMD_HAVE_REAL
886 /*! \brief As calc_dist_iter(), but using SIMD intrinsics. */
887 static void gmx_simdcall calc_dist_iter_simd(int b0,
889 gmx::ArrayRef<const AtomPair> atoms,
890 const rvec* gmx_restrict x,
891 const real* gmx_restrict bllen,
892 const real* gmx_restrict blc,
893 const real* pbc_simd,
895 real* gmx_restrict rhs,
896 real* gmx_restrict sol,
899 SimdReal min_S(GMX_REAL_MIN);
901 SimdReal wfac_S(wfac);
904 assert(b0 % GMX_SIMD_REAL_WIDTH == 0);
906 /* Initialize all to FALSE */
907 warn_S = (two_S < setZero());
909 for (int bs = b0; bs < b1; bs += GMX_SIMD_REAL_WIDTH)
911 SimdReal x0_S, y0_S, z0_S;
912 SimdReal x1_S, y1_S, z1_S;
913 SimdReal rx_S, ry_S, rz_S, n2_S;
914 SimdReal len_S, len2_S, dlen2_S, lc_S, blc_S;
915 alignas(GMX_SIMD_ALIGNMENT) std::int32_t offset0[GMX_SIMD_REAL_WIDTH];
916 alignas(GMX_SIMD_ALIGNMENT) std::int32_t offset1[GMX_SIMD_REAL_WIDTH];
918 for (int i = 0; i < GMX_SIMD_REAL_WIDTH; i++)
920 offset0[i] = atoms[bs + i].index1;
921 offset1[i] = atoms[bs + i].index2;
924 gatherLoadUTransposeTSANSafe<3>(reinterpret_cast<const real*>(x), offset0, &x0_S, &y0_S, &z0_S);
925 gatherLoadUTransposeTSANSafe<3>(reinterpret_cast<const real*>(x), offset1, &x1_S, &y1_S, &z1_S);
931 pbc_correct_dx_simd(&rx_S, &ry_S, &rz_S, pbc_simd);
933 n2_S = norm2(rx_S, ry_S, rz_S);
935 len_S = load<SimdReal>(bllen + bs);
936 len2_S = len_S * len_S;
938 dlen2_S = fms(two_S, len2_S, n2_S);
940 warn_S = warn_S || (dlen2_S < (wfac_S * len2_S));
942 /* Avoid 1/0 by taking the max with REAL_MIN.
943 * Note: when dlen2 is close to zero (90 degree constraint rotation),
944 * the accuracy of the algorithm is no longer relevant.
946 dlen2_S = max(dlen2_S, min_S);
948 lc_S = fnma(dlen2_S, invsqrt(dlen2_S), len_S);
950 blc_S = load<SimdReal>(blc + bs);
954 store(rhs + bs, lc_S);
955 store(sol + bs, lc_S);
963 #endif // GMX_SIMD_HAVE_REAL
965 //! Implements LINCS constraining.
966 static void do_lincs(ArrayRefWithPadding<const RVec> xPadded,
967 ArrayRefWithPadding<RVec> xpPadded,
972 ArrayRef<const real> invmass,
982 const rvec* x = as_rvec_array(xPadded.paddedArrayRef().data());
983 rvec* xp = as_rvec_array(xpPadded.paddedArrayRef().data());
984 rvec* gmx_restrict v = as_rvec_array(vRef.data());
986 const int b0 = lincsd->task[th].b0;
987 const int b1 = lincsd->task[th].b1;
989 gmx::ArrayRef<const AtomPair> atoms = lincsd->atoms;
990 gmx::ArrayRef<gmx::RVec> r = lincsd->tmpv;
991 gmx::ArrayRef<const int> blnr = lincsd->blnr;
992 gmx::ArrayRef<const int> blbnb = lincsd->blbnb;
993 gmx::ArrayRef<const real> blc = lincsd->blc;
994 gmx::ArrayRef<const real> blmf = lincsd->blmf;
995 gmx::ArrayRef<const real> bllen = lincsd->bllen;
996 gmx::ArrayRef<real> blcc = lincsd->tmpncc;
997 gmx::ArrayRef<real> rhs1 = lincsd->tmp1;
998 gmx::ArrayRef<real> rhs2 = lincsd->tmp2;
999 gmx::ArrayRef<real> sol = lincsd->tmp3;
1000 gmx::ArrayRef<real> blc_sol = lincsd->tmp4;
1001 gmx::ArrayRef<real> mlambda = lincsd->mlambda;
1002 gmx::ArrayRef<const int> nlocat = lincsd->nlocat;
1004 #if GMX_SIMD_HAVE_REAL
1006 /* This SIMD code does the same as the plain-C code after the #else.
1007 * The only difference is that we always call pbc code, as with SIMD
1008 * the overhead of pbc computation (when not needed) is small.
1010 alignas(GMX_SIMD_ALIGNMENT) real pbc_simd[9 * GMX_SIMD_REAL_WIDTH];
1012 /* Convert the pbc struct for SIMD */
1013 set_pbc_simd(pbc, pbc_simd);
1015 /* Compute normalized x i-j vectors, store in r.
1016 * Compute the inner product of r and xp i-j and store in rhs1.
1019 b0, b1, atoms, x, xp, bllen.data(), blc.data(), pbc_simd, as_rvec_array(r.data()), rhs1.data(), sol.data());
1021 #else // GMX_SIMD_HAVE_REAL
1025 /* Compute normalized i-j vectors */
1026 for (int b = b0; b < b1; b++)
1029 pbc_dx_aiuc(pbc, x[atoms[b].index1], x[atoms[b].index2], dx);
1032 pbc_dx_aiuc(pbc, xp[atoms[b].index1], xp[atoms[b].index2], dx);
1033 real mvb = blc[b] * (::iprod(r[b], dx) - bllen[b]);
1040 /* Compute normalized i-j vectors */
1041 for (int b = b0; b < b1; b++)
1043 int i = atoms[b].index1;
1044 int j = atoms[b].index2;
1045 real tmp0 = x[i][0] - x[j][0];
1046 real tmp1 = x[i][1] - x[j][1];
1047 real tmp2 = x[i][2] - x[j][2];
1048 real rlen = gmx::invsqrt(tmp0 * tmp0 + tmp1 * tmp1 + tmp2 * tmp2);
1049 r[b][0] = rlen * tmp0;
1050 r[b][1] = rlen * tmp1;
1051 r[b][2] = rlen * tmp2;
1052 /* 16 ncons flops */
1055 * (r[b][0] * (xp[i][0] - xp[j][0]) + r[b][1] * (xp[i][1] - xp[j][1])
1056 + r[b][2] * (xp[i][2] - xp[j][2]) - bllen[b]);
1061 /* Together: 26*ncons + 6*nrtot flops */
1064 #endif // GMX_SIMD_HAVE_REAL
1066 if (lincsd->bTaskDep)
1068 /* We need a barrier, since the matrix construction below
1069 * can access entries in r of other threads.
1074 /* Construct the (sparse) LINCS matrix */
1075 for (int b = b0; b < b1; b++)
1077 for (int n = blnr[b]; n < blnr[b + 1]; n++)
1079 blcc[n] = blmf[n] * gmx::dot(r[b], r[blbnb[n]]);
1082 /* Together: 26*ncons + 6*nrtot flops */
1084 lincs_matrix_expand(*lincsd, lincsd->task[th], blcc, rhs1, rhs2, sol);
1085 /* nrec*(ncons+2*nrtot) flops */
1087 #if GMX_SIMD_HAVE_REAL
1088 for (int b = b0; b < b1; b += GMX_SIMD_REAL_WIDTH)
1090 SimdReal t1 = load<SimdReal>(blc.data() + b);
1091 SimdReal t2 = load<SimdReal>(sol.data() + b);
1092 store(mlambda.data() + b, t1 * t2);
1095 for (int b = b0; b < b1; b++)
1097 mlambda[b] = blc[b] * sol[b];
1099 #endif // GMX_SIMD_HAVE_REAL
1101 /* Update the coordinates */
1102 lincs_update_atoms(lincsd, th, 1.0, mlambda, r, invmass, xp);
1105 ******** Correction for centripetal effects ********
1110 wfac = std::cos(gmx::c_deg2Rad * wangle);
1113 for (int iter = 0; iter < lincsd->nIter; iter++)
1115 if ((lincsd->bCommIter && haveDDAtomOrdering(*cr) && cr->dd->constraints))
1120 /* Communicate the corrected non-local coordinates */
1121 if (haveDDAtomOrdering(*cr))
1123 dd_move_x_constraints(cr->dd, box, xpPadded.unpaddedArrayRef(), ArrayRef<RVec>(), FALSE);
1128 else if (lincsd->bTaskDep)
1133 #if GMX_SIMD_HAVE_REAL
1134 calc_dist_iter_simd(
1135 b0, b1, atoms, xp, bllen.data(), blc.data(), pbc_simd, wfac, rhs1.data(), sol.data(), bWarn);
1137 calc_dist_iter(b0, b1, atoms, xp, bllen.data(), blc.data(), pbc, wfac, rhs1.data(), sol.data(), bWarn);
1138 /* 20*ncons flops */
1139 #endif // GMX_SIMD_HAVE_REAL
1141 lincs_matrix_expand(*lincsd, lincsd->task[th], blcc, rhs1, rhs2, sol);
1142 /* nrec*(ncons+2*nrtot) flops */
1144 #if GMX_SIMD_HAVE_REAL
1145 for (int b = b0; b < b1; b += GMX_SIMD_REAL_WIDTH)
1147 SimdReal t1 = load<SimdReal>(blc.data() + b);
1148 SimdReal t2 = load<SimdReal>(sol.data() + b);
1149 SimdReal mvb = t1 * t2;
1150 store(blc_sol.data() + b, mvb);
1151 store(mlambda.data() + b, load<SimdReal>(mlambda.data() + b) + mvb);
1154 for (int b = b0; b < b1; b++)
1156 real mvb = blc[b] * sol[b];
1160 #endif // GMX_SIMD_HAVE_REAL
1162 /* Update the coordinates */
1163 lincs_update_atoms(lincsd, th, 1.0, blc_sol, r, invmass, xp);
1165 /* nit*ncons*(37+9*nrec) flops */
1169 /* Update the velocities */
1170 lincs_update_atoms(lincsd, th, invdt, mlambda, r, invmass, v);
1171 /* 16 ncons flops */
1174 if (!nlocat.empty() && (bCalcDHDL || bCalcVir))
1176 if (lincsd->bTaskDep)
1178 /* In lincs_update_atoms threads might cross-read mlambda */
1182 /* Only account for local atoms */
1183 for (int b = b0; b < b1; b++)
1185 mlambda[b] *= 0.5 * nlocat[b];
1192 for (int b = b0; b < b1; b++)
1194 /* Note that this this is dhdl*dt^2, the dt^2 factor is corrected
1195 * later after the contributions are reduced over the threads.
1197 dhdl -= lincsd->mlambda[b] * lincsd->ddist[b];
1199 lincsd->task[th].dhdlambda = dhdl;
1204 /* Constraint virial */
1205 for (int b = b0; b < b1; b++)
1207 real tmp0 = -bllen[b] * mlambda[b];
1208 for (int i = 0; i < DIM; i++)
1210 real tmp1 = tmp0 * r[b][i];
1211 for (int j = 0; j < DIM; j++)
1213 vir_r_m_dr[i][j] -= tmp1 * r[b][j];
1216 } /* 22 ncons flops */
1220 * 26*ncons + 6*nrtot + nrec*(ncons+2*nrtot)
1221 * + nit * (20*ncons + nrec*(ncons+2*nrtot) + 17 ncons)
1223 * (26+nrec)*ncons + (6+2*nrec)*nrtot
1224 * + nit * ((37+nrec)*ncons + 2*nrec*nrtot)
1226 * (63+nrec)*ncons + (6+4*nrec)*nrtot
1230 /*! \brief Sets the elements in the LINCS matrix for task task. */
1231 static void set_lincs_matrix_task(Lincs* li,
1233 ArrayRef<const real> invmass,
1235 int* nCrossTaskTriangles)
1237 /* Construct the coupling coefficient matrix blmf */
1238 li_task->ntriangle = 0;
1240 *nCrossTaskTriangles = 0;
1241 for (int i = li_task->b0; i < li_task->b1; i++)
1243 const int a1 = li->atoms[i].index1;
1244 const int a2 = li->atoms[i].index2;
1245 for (int n = li->blnr[i]; n < li->blnr[i + 1]; n++)
1247 const int k = li->blbnb[n];
1249 /* If we are using multiple, independent tasks for LINCS,
1250 * the calls to check_assign_connected should have
1251 * put all connected constraints in our task.
1253 assert(li->bTaskDep || (k >= li_task->b0 && k < li_task->b1));
1256 if (a1 == li->atoms[k].index1 || a2 == li->atoms[k].index2)
1266 if (a1 == li->atoms[k].index1 || a1 == li->atoms[k].index2)
1276 li->blmf[n] = sign * invmass[center] * li->blc[i] * li->blc[k];
1277 li->blmf1[n] = sign * 0.5;
1278 if (li->ncg_triangle > 0)
1280 /* Look for constraint triangles */
1281 for (int nk = li->blnr[k]; nk < li->blnr[k + 1]; nk++)
1283 const int kk = li->blbnb[nk];
1284 if (kk != i && kk != k && (li->atoms[kk].index1 == end || li->atoms[kk].index2 == end))
1286 /* Check if the constraints in this triangle actually
1287 * belong to a different task. We still assign them
1288 * here, since it's convenient for the triangle
1289 * iterations, but we then need an extra barrier.
1291 if (k < li_task->b0 || k >= li_task->b1 || kk < li_task->b0 || kk >= li_task->b1)
1293 (*nCrossTaskTriangles)++;
1296 if (li_task->ntriangle == 0 || li_task->triangle[li_task->ntriangle - 1] < i)
1298 /* Add this constraint to the triangle list */
1299 li_task->triangle[li_task->ntriangle] = i;
1300 li_task->tri_bits[li_task->ntriangle] = 0;
1301 li_task->ntriangle++;
1302 if (li->blnr[i + 1] - li->blnr[i]
1303 > static_cast<int>(sizeof(li_task->tri_bits[0]) * 8 - 1))
1306 "A constraint is connected to %d constraints, this is "
1307 "more than the %zu allowed for constraints participating "
1309 li->blnr[i + 1] - li->blnr[i],
1310 sizeof(li_task->tri_bits[0]) * 8 - 1);
1313 li_task->tri_bits[li_task->ntriangle - 1] |= (1 << (n - li->blnr[i]));
1322 /*! \brief Sets the elements in the LINCS matrix. */
1323 static void set_lincs_matrix(Lincs* li, ArrayRef<const real> invmass, real lambda)
1325 const real invsqrt2 = 0.7071067811865475244;
1327 for (int i = 0; (i < li->nc); i++)
1329 const int a1 = li->atoms[i].index1;
1330 const int a2 = li->atoms[i].index2;
1331 li->blc[i] = gmx::invsqrt(invmass[a1] + invmass[a2]);
1332 li->blc1[i] = invsqrt2;
1335 /* Construct the coupling coefficient matrix blmf */
1336 int ntriangle = 0, ncc_triangle = 0, nCrossTaskTriangles = 0;
1337 #pragma omp parallel for reduction(+: ntriangle, ncc_triangle, nCrossTaskTriangles) num_threads(li->ntask) schedule(static)
1338 for (int th = 0; th < li->ntask; th++)
1342 set_lincs_matrix_task(li, &li->task[th], invmass, &ncc_triangle, &nCrossTaskTriangles);
1343 ntriangle += li->task[th].ntriangle;
1345 GMX_CATCH_ALL_AND_EXIT_WITH_FATAL_ERROR
1347 li->ntriangle = ntriangle;
1348 li->ncc_triangle = ncc_triangle;
1349 li->bTaskDepTri = (nCrossTaskTriangles > 0);
1353 fprintf(debug, "The %d constraints participate in %d triangles\n", li->nc, li->ntriangle);
1354 fprintf(debug, "There are %d constraint couplings, of which %d in triangles\n", li->ncc, li->ncc_triangle);
1355 if (li->ntriangle > 0 && li->ntask > 1)
1358 "%d constraint triangles contain constraints assigned to different tasks\n",
1359 nCrossTaskTriangles);
1364 * so we know with which lambda value the masses have been set.
1366 li->matlam = lambda;
1369 //! Finds all triangles of atoms that share constraints to a central atom.
1370 static int count_triangle_constraints(const InteractionLists& ilist, const ListOfLists<int>& at2con)
1372 const int ncon1 = ilist[F_CONSTR].size() / 3;
1373 const int ncon_tot = ncon1 + ilist[F_CONSTRNC].size() / 3;
1375 gmx::ArrayRef<const int> ia1 = ilist[F_CONSTR].iatoms;
1376 gmx::ArrayRef<const int> ia2 = ilist[F_CONSTRNC].iatoms;
1378 int ncon_triangle = 0;
1379 for (int c0 = 0; c0 < ncon_tot; c0++)
1381 bool bTriangle = FALSE;
1382 const int* iap = constr_iatomptr(ia1, ia2, c0);
1383 const int a00 = iap[1];
1384 const int a01 = iap[2];
1385 for (const int c1 : at2con[a01])
1389 const int* iap = constr_iatomptr(ia1, ia2, c1);
1390 const int a10 = iap[1];
1391 const int a11 = iap[2];
1401 for (const int c2 : at2con[ac1])
1403 if (c2 != c0 && c2 != c1)
1405 const int* iap = constr_iatomptr(ia1, ia2, c2);
1406 const int a20 = iap[1];
1407 const int a21 = iap[2];
1408 if (a20 == a00 || a21 == a00)
1422 return ncon_triangle;
1425 //! Finds sequences of sequential constraints.
1426 static bool more_than_two_sequential_constraints(const InteractionLists& ilist, const ListOfLists<int>& at2con)
1428 const int ncon1 = ilist[F_CONSTR].size() / 3;
1429 const int ncon_tot = ncon1 + ilist[F_CONSTRNC].size() / 3;
1431 gmx::ArrayRef<const int> ia1 = ilist[F_CONSTR].iatoms;
1432 gmx::ArrayRef<const int> ia2 = ilist[F_CONSTRNC].iatoms;
1434 for (int c = 0; c < ncon_tot; c++)
1436 const int* iap = constr_iatomptr(ia1, ia2, c);
1437 const int a1 = iap[1];
1438 const int a2 = iap[2];
1439 /* Check if this constraint has constraints connected at both atoms */
1440 if (at2con[a1].ssize() > 1 && at2con[a2].ssize() > 1)
1449 Lincs* init_lincs(FILE* fplog,
1450 const gmx_mtop_t& mtop,
1451 int nflexcon_global,
1452 ArrayRef<const ListOfLists<int>> atomToConstraintsPerMolType,
1456 ObservablesReducerBuilder* observablesReducerBuilder)
1458 // TODO this should become a unique_ptr
1460 bool bMoreThanTwoSeq;
1464 fprintf(fplog, "\nInitializing%s LINear Constraint Solver\n", bPLINCS ? " Parallel" : "");
1469 li->ncg = gmx_mtop_ftype_count(mtop, F_CONSTR) + gmx_mtop_ftype_count(mtop, F_CONSTRNC);
1470 li->ncg_flex = nflexcon_global;
1473 li->nOrder = nProjOrder;
1475 li->max_connect = 0;
1476 for (size_t mt = 0; mt < mtop.moltype.size(); mt++)
1478 const auto& at2con = atomToConstraintsPerMolType[mt];
1479 for (int a = 0; a < mtop.moltype[mt].atoms.nr; a++)
1481 li->max_connect = std::max(li->max_connect, int(at2con[a].ssize()));
1485 li->ncg_triangle = 0;
1486 bMoreThanTwoSeq = FALSE;
1487 for (const gmx_molblock_t& molb : mtop.molblock)
1489 const gmx_moltype_t& molt = mtop.moltype[molb.type];
1490 const auto& at2con = atomToConstraintsPerMolType[molb.type];
1492 li->ncg_triangle += molb.nmol * count_triangle_constraints(molt.ilist, at2con);
1494 if (!bMoreThanTwoSeq && more_than_two_sequential_constraints(molt.ilist, at2con))
1496 bMoreThanTwoSeq = TRUE;
1500 /* Check if we need to communicate not only before LINCS,
1501 * but also before each iteration.
1502 * The check for only two sequential constraints is only
1503 * useful for the common case of H-bond only constraints.
1504 * With more effort we could also make it useful for small
1505 * molecules with nr. sequential constraints <= nOrder-1.
1507 li->bCommIter = (bPLINCS && (li->nOrder < 1 || bMoreThanTwoSeq));
1509 if (debug && bPLINCS)
1511 fprintf(debug, "PLINCS communication before each iteration: %d\n", static_cast<int>(li->bCommIter));
1514 /* LINCS can run on any number of threads.
1515 * Currently the number is fixed for the whole simulation,
1516 * but it could be set in set_lincs().
1517 * The current constraint to task assignment code can create independent
1518 * tasks only when not more than two constraints are connected sequentially.
1520 li->ntask = gmx_omp_nthreads_get(ModuleMultiThread::Lincs);
1521 li->bTaskDep = (li->ntask > 1 && bMoreThanTwoSeq);
1524 fprintf(debug, "LINCS: using %d threads, tasks are %sdependent\n", li->ntask, li->bTaskDep ? "" : "in");
1532 /* Allocate an extra elements for "task-overlap" constraints */
1533 li->task.resize(li->ntask + 1);
1536 if (bPLINCS || li->ncg_triangle > 0)
1538 please_cite(fplog, "Hess2008a");
1542 please_cite(fplog, "Hess97a");
1547 fprintf(fplog, "The number of constraints is %d\n", li->ncg);
1551 "There are constraints between atoms in different decomposition domains,\n"
1552 "will communicate selected coordinates each lincs iteration\n");
1554 if (li->ncg_triangle > 0)
1557 "%d constraints are involved in constraint triangles,\n"
1558 "will apply an additional matrix expansion of order %d for couplings\n"
1559 "between constraints inside triangles\n",
1565 if (observablesReducerBuilder)
1567 ObservablesReducerBuilder::CallbackFromBuilder callbackFromBuilder =
1568 [li](ObservablesReducerBuilder::CallbackToRequireReduction c, gmx::ArrayRef<double> v) {
1569 li->callbackToRequireReduction = std::move(c);
1570 li->rmsdReductionBuffer = v;
1573 // Make the callback that runs afer reduction.
1574 ObservablesReducerBuilder::CallbackAfterReduction callbackAfterReduction = [li](gmx::Step /*step*/) {
1575 if (li->rmsdReductionBuffer[0] > 0)
1577 li->constraintRmsDeviation =
1578 std::sqrt(li->rmsdReductionBuffer[1] / li->rmsdReductionBuffer[0]);
1582 const int reductionValuesRequired = 2;
1583 observablesReducerBuilder->addSubscriber(reductionValuesRequired,
1584 std::move(callbackFromBuilder),
1585 std::move(callbackAfterReduction));
1591 void done_lincs(Lincs* li)
1596 /*! \brief Sets up the work division over the threads. */
1597 static void lincs_thread_setup(Lincs* li, int natoms)
1599 li->atf.resize(natoms);
1601 gmx::ArrayRef<gmx_bitmask_t> atf = li->atf;
1603 /* Clear the atom flags */
1604 for (gmx_bitmask_t& mask : atf)
1606 bitmask_clear(&mask);
1609 if (li->ntask > BITMASK_SIZE)
1611 gmx_fatal(FARGS, "More than %d threads is not supported for LINCS.", BITMASK_SIZE);
1614 for (int th = 0; th < li->ntask; th++)
1616 const Task* li_task = &li->task[th];
1618 /* For each atom set a flag for constraints from each */
1619 for (int b = li_task->b0; b < li_task->b1; b++)
1621 bitmask_set_bit(&atf[li->atoms[b].index1], th);
1622 bitmask_set_bit(&atf[li->atoms[b].index2], th);
1626 #pragma omp parallel for num_threads(li->ntask) schedule(static)
1627 for (int th = 0; th < li->ntask; th++)
1635 li_task = &li->task[th];
1637 bitmask_init_low_bits(&mask, th);
1639 li_task->ind.clear();
1640 li_task->ind_r.clear();
1641 for (b = li_task->b0; b < li_task->b1; b++)
1643 /* We let the constraint with the lowest thread index
1644 * operate on atoms with constraints from multiple threads.
1646 if (bitmask_is_disjoint(atf[li->atoms[b].index1], mask)
1647 && bitmask_is_disjoint(atf[li->atoms[b].index2], mask))
1649 /* Add the constraint to the local atom update index */
1650 li_task->ind.push_back(b);
1654 /* Add the constraint to the rest block */
1655 li_task->ind_r.push_back(b);
1659 GMX_CATCH_ALL_AND_EXIT_WITH_FATAL_ERROR
1662 /* We need to copy all constraints which have not be assigned
1663 * to a thread to a separate list which will be handled by one thread.
1665 Task* li_m = &li->task[li->ntask];
1668 for (int th = 0; th < li->ntask; th++)
1670 const Task& li_task = li->task[th];
1672 for (int ind_r : li_task.ind_r)
1674 li_m->ind.push_back(ind_r);
1679 fprintf(debug, "LINCS thread %d: %zu constraints\n", th, li_task.ind.size());
1685 fprintf(debug, "LINCS thread r: %zu constraints\n", li_m->ind.size());
1689 //! Assign a constraint.
1690 static void assign_constraint(Lincs* li,
1691 int constraint_index,
1696 const ListOfLists<int>& at2con)
1702 /* Make an mapping of local topology constraint index to LINCS index */
1703 li->con_index[constraint_index] = con;
1705 li->bllen0[con] = lenA;
1706 li->ddist[con] = lenB - lenA;
1707 /* Set the length to the topology A length */
1708 li->bllen[con] = lenA;
1709 li->atoms[con].index1 = a1;
1710 li->atoms[con].index2 = a2;
1712 /* Make space in the constraint connection matrix for constraints
1713 * connected to both end of the current constraint.
1715 li->ncc += at2con[a1].ssize() - 1 + at2con[a2].ssize() - 1;
1717 li->blnr[con + 1] = li->ncc;
1719 /* Increase the constraint count */
1723 /*! \brief Check if constraint with topology index constraint_index is connected
1724 * to other constraints, and if so add those connected constraints to our task. */
1725 static void check_assign_connected(Lincs* li,
1726 gmx::ArrayRef<const int> iatom,
1727 const InteractionDefinitions& idef,
1731 const ListOfLists<int>& at2con)
1733 /* Currently this function only supports constraint groups
1734 * in which all constraints share at least one atom
1735 * (e.g. H-bond constraints).
1736 * Check both ends of the current constraint for
1737 * connected constraints. We need to assign those
1740 for (int end = 0; end < 2; end++)
1742 const int a = (end == 0 ? a1 : a2);
1744 for (const int cc : at2con[a])
1746 /* Check if constraint cc has not yet been assigned */
1747 if (li->con_index[cc] == -1)
1749 const int type = iatom[cc * 3];
1750 const real lenA = idef.iparams[type].constr.dA;
1751 const real lenB = idef.iparams[type].constr.dB;
1753 if (bDynamics || lenA != 0 || lenB != 0)
1755 assign_constraint(li, cc, iatom[3 * cc + 1], iatom[3 * cc + 2], lenA, lenB, at2con);
1762 /*! \brief Check if constraint with topology index constraint_index is involved
1763 * in a constraint triangle, and if so add the other two constraints
1764 * in the triangle to our task. */
1765 static void check_assign_triangle(Lincs* li,
1766 gmx::ArrayRef<const int> iatom,
1767 const InteractionDefinitions& idef,
1769 int constraint_index,
1772 const ListOfLists<int>& at2con)
1774 int nca, cc[32], ca[32];
1775 int c_triangle[2] = { -1, -1 };
1778 for (const int c : at2con[a1])
1780 if (c != constraint_index)
1784 aa1 = iatom[c * 3 + 1];
1785 aa2 = iatom[c * 3 + 2];
1801 for (const int c : at2con[a2])
1803 if (c != constraint_index)
1807 aa1 = iatom[c * 3 + 1];
1808 aa2 = iatom[c * 3 + 2];
1811 for (i = 0; i < nca; i++)
1815 c_triangle[0] = cc[i];
1822 for (i = 0; i < nca; i++)
1826 c_triangle[0] = cc[i];
1834 if (c_triangle[0] >= 0)
1838 for (end = 0; end < 2; end++)
1840 /* Check if constraint c_triangle[end] has not yet been assigned */
1841 if (li->con_index[c_triangle[end]] == -1)
1846 i = c_triangle[end] * 3;
1848 lenA = idef.iparams[type].constr.dA;
1849 lenB = idef.iparams[type].constr.dB;
1851 if (bDynamics || lenA != 0 || lenB != 0)
1853 assign_constraint(li, c_triangle[end], iatom[i + 1], iatom[i + 2], lenA, lenB, at2con);
1860 //! Sets matrix indices.
1861 static void set_matrix_indices(Lincs* li, const Task& li_task, const ListOfLists<int>& at2con, bool bSortMatrix)
1863 for (int b = li_task.b0; b < li_task.b1; b++)
1865 const int a1 = li->atoms[b].index1;
1866 const int a2 = li->atoms[b].index2;
1868 int i = li->blnr[b];
1869 for (const int constraint : at2con[a1])
1871 const int concon = li->con_index[constraint];
1874 li->blbnb[i++] = concon;
1877 for (const int constraint : at2con[a2])
1879 const int concon = li->con_index[constraint];
1882 li->blbnb[i++] = concon;
1888 /* Order the blbnb matrix to optimize memory access */
1889 std::sort(li->blbnb.begin() + li->blnr[b], li->blbnb.begin() + li->blnr[b + 1]);
1894 void set_lincs(const InteractionDefinitions& idef,
1896 ArrayRef<const real> invmass,
1899 const t_commrec* cr,
1905 /* Zero the thread index ranges.
1906 * Otherwise without local constraints we could return with old ranges.
1908 for (int i = 0; i < li->ntask; i++)
1912 li->task[i].ind.clear();
1916 li->task[li->ntask].ind.clear();
1919 /* This is the local topology, so there are only F_CONSTR constraints */
1920 if (idef.il[F_CONSTR].empty())
1922 /* There are no constraints,
1923 * we do not need to fill any data structures.
1930 fprintf(debug, "Building the LINCS connectivity\n");
1934 if (haveDDAtomOrdering(*cr))
1936 if (cr->dd->constraints)
1940 dd_get_constraint_range(*cr->dd, &start, &natoms);
1944 natoms = dd_numHomeAtoms(*cr->dd);
1952 const ListOfLists<int> at2con =
1953 make_at2con(natoms, idef.il, idef.iparams, flexibleConstraintTreatment(bDynamics));
1955 const int ncon_tot = idef.il[F_CONSTR].size() / 3;
1957 /* Ensure we have enough padding for aligned loads for each thread */
1958 const int numEntries = ncon_tot + li->ntask * simd_width;
1959 li->con_index.resize(numEntries);
1960 li->bllen0.resize(numEntries);
1961 li->ddist.resize(numEntries);
1962 li->atoms.resize(numEntries);
1963 li->blc.resize(numEntries);
1964 li->blc1.resize(numEntries);
1965 li->blnr.resize(numEntries + 1);
1966 li->bllen.resize(numEntries);
1967 li->tmpv.resizeWithPadding(numEntries);
1968 if (haveDDAtomOrdering(*cr))
1970 li->nlocat.resize(numEntries);
1972 li->tmp1.resize(numEntries);
1973 li->tmp2.resize(numEntries);
1974 li->tmp3.resize(numEntries);
1975 li->tmp4.resize(numEntries);
1976 li->mlambda.resize(numEntries);
1978 gmx::ArrayRef<const int> iatom = idef.il[F_CONSTR].iatoms;
1980 li->blnr[0] = li->ncc;
1982 /* Assign the constraints for li->ntask LINCS tasks.
1983 * We target a uniform distribution of constraints over the tasks.
1984 * Note that when flexible constraints are present, but are removed here
1985 * (e.g. because we are doing EM) we get imbalance, but since that doesn't
1986 * happen during normal MD, that's ok.
1989 /* Determine the number of constraints we need to assign here */
1990 int ncon_assign = ncon_tot;
1993 /* With energy minimization, flexible constraints are ignored
1994 * (and thus minimized, as they should be).
1996 ncon_assign -= countFlexibleConstraints(idef.il, idef.iparams);
1999 /* Set the target constraint count per task to exactly uniform,
2000 * this might be overridden below.
2002 int ncon_target = (ncon_assign + li->ntask - 1) / li->ntask;
2004 /* Mark all constraints as unassigned by setting their index to -1 */
2005 for (int con = 0; con < ncon_tot; con++)
2007 li->con_index[con] = -1;
2011 for (int th = 0; th < li->ntask; th++)
2015 li_task = &li->task[th];
2017 #if GMX_SIMD_HAVE_REAL
2018 /* With indepedent tasks we likely have H-bond constraints or constraint
2019 * pairs. The connected constraints will be pulled into the task, so the
2020 * constraints per task will often exceed ncon_target.
2021 * Triangle constraints can also increase the count, but there are
2022 * relatively few of those, so we usually expect to get ncon_target.
2026 /* We round ncon_target to a multiple of GMX_SIMD_WIDTH,
2027 * since otherwise a lot of operations can be wasted.
2028 * There are several ways to round here, we choose the one
2029 * that alternates block sizes, which helps with Intel HT.
2031 ncon_target = ((ncon_assign * (th + 1)) / li->ntask - li->nc_real + GMX_SIMD_REAL_WIDTH - 1)
2032 & ~(GMX_SIMD_REAL_WIDTH - 1);
2034 #endif // GMX_SIMD==2 && GMX_SIMD_HAVE_REAL
2036 /* Continue filling the arrays where we left off with the previous task,
2037 * including padding for SIMD.
2039 li_task->b0 = li->nc;
2041 gmx::ArrayRef<const t_iparams> iparams = idef.iparams;
2043 while (con < ncon_tot && li->nc - li_task->b0 < ncon_target)
2045 if (li->con_index[con] == -1)
2050 type = iatom[3 * con];
2051 a1 = iatom[3 * con + 1];
2052 a2 = iatom[3 * con + 2];
2053 lenA = iparams[type].constr.dA;
2054 lenB = iparams[type].constr.dB;
2055 /* Skip the flexible constraints when not doing dynamics */
2056 if (bDynamics || lenA != 0 || lenB != 0)
2058 assign_constraint(li, con, a1, a2, lenA, lenB, at2con);
2060 if (li->ntask > 1 && !li->bTaskDep)
2062 /* We can generate independent tasks. Check if we
2063 * need to assign connected constraints to our task.
2065 check_assign_connected(li, iatom, idef, bDynamics, a1, a2, at2con);
2067 if (li->ntask > 1 && li->ncg_triangle > 0)
2069 /* Ensure constraints in one triangle are assigned
2072 check_assign_triangle(li, iatom, idef, bDynamics, con, a1, a2, at2con);
2080 li_task->b1 = li->nc;
2084 /* Copy the last atom pair indices and lengths for constraints
2085 * up to a multiple of simd_width, such that we can do all
2086 * SIMD operations without having to worry about end effects.
2090 li->nc = ((li_task->b1 + simd_width - 1) / simd_width) * simd_width;
2091 last = li_task->b1 - 1;
2092 for (i = li_task->b1; i < li->nc; i++)
2094 li->atoms[i] = li->atoms[last];
2095 li->bllen0[i] = li->bllen0[last];
2096 li->ddist[i] = li->ddist[last];
2097 li->bllen[i] = li->bllen[last];
2098 li->blnr[i + 1] = li->blnr[last + 1];
2102 /* Keep track of how many constraints we assigned */
2103 li->nc_real += li_task->b1 - li_task->b0;
2107 fprintf(debug, "LINCS task %d constraints %d - %d\n", th, li_task->b0, li_task->b1);
2111 assert(li->nc_real == ncon_assign);
2115 /* Without DD we order the blbnb matrix to optimize memory access.
2116 * With DD the overhead of sorting is more than the gain during access.
2118 bSortMatrix = !haveDDAtomOrdering(*cr);
2120 li->blbnb.resize(li->ncc);
2122 #pragma omp parallel for num_threads(li->ntask) schedule(static)
2123 for (int th = 0; th < li->ntask; th++)
2127 Task& li_task = li->task[th];
2129 if (li->ncg_triangle > 0)
2131 /* This is allocating too much, but it is difficult to improve */
2132 li_task.triangle.resize(li_task.b1 - li_task.b0);
2133 li_task.tri_bits.resize(li_task.b1 - li_task.b0);
2136 set_matrix_indices(li, li_task, at2con, bSortMatrix);
2138 GMX_CATCH_ALL_AND_EXIT_WITH_FATAL_ERROR
2141 if (cr->dd == nullptr)
2143 /* Since the matrix is static, we should free some memory */
2144 li->blbnb.resize(li->ncc);
2147 li->blmf.resize(li->ncc);
2148 li->blmf1.resize(li->ncc);
2149 li->tmpncc.resize(li->ncc);
2151 gmx::ArrayRef<const int> nlocat_dd = dd_constraints_nlocalatoms(cr->dd);
2152 if (!nlocat_dd.empty())
2154 /* Convert nlocat from local topology to LINCS constraint indexing */
2155 for (con = 0; con < ncon_tot; con++)
2157 li->nlocat[li->con_index[con]] = nlocat_dd[con];
2167 fprintf(debug, "Number of constraints is %d, padded %d, couplings %d\n", li->nc_real, li->nc, li->ncc);
2172 lincs_thread_setup(li, numAtoms);
2175 set_lincs_matrix(li, invmass, lambda);
2178 //! Issues a warning when LINCS constraints cannot be satisfied.
2179 static void lincs_warning(gmx_domdec_t* dd,
2180 ArrayRef<const RVec> x,
2181 ArrayRef<const RVec> xprime,
2184 gmx::ArrayRef<const AtomPair> atoms,
2185 gmx::ArrayRef<const real> bllen,
2190 real wfac = std::cos(gmx::c_deg2Rad * wangle);
2193 "bonds that rotated more than %g degrees:\n"
2194 " atom 1 atom 2 angle previous, current, constraint length\n",
2197 for (int b = 0; b < ncons; b++)
2199 const int i = atoms[b].index1;
2200 const int j = atoms[b].index2;
2205 pbc_dx_aiuc(pbc, x[i], x[j], v0);
2206 pbc_dx_aiuc(pbc, xprime[i], xprime[j], v1);
2210 rvec_sub(x[i], x[j], v0);
2211 rvec_sub(xprime[i], xprime[j], v1);
2215 real cosine = ::iprod(v0, v1) / (d0 * d1);
2219 " %6d %6d %5.1f %8.4f %8.4f %8.4f\n",
2222 gmx::c_rad2Deg * std::acos(cosine),
2226 if (!std::isfinite(d1))
2228 gmx_fatal(FARGS, "Bond length not finite.");
2234 if (*warncount > maxwarn)
2236 too_many_constraint_warnings(ConstraintAlgorithm::Lincs, *warncount);
2240 //! Status information about how well LINCS satisified the constraints in this domain
2241 struct LincsDeviations
2243 //! The maximum over all bonds in this domain of the relative deviation in bond lengths
2244 real maxDeviation = 0;
2245 //! Sum over all bonds in this domain of the squared relative deviation
2246 real sumSquaredDeviation = 0;
2247 //! Index of bond with max deviation
2248 int indexOfMaxDeviation = -1;
2249 //! Number of bonds constrained in this domain
2250 int numConstraints = 0;
2253 //! Determine how well the constraints have been satisfied.
2254 static LincsDeviations makeLincsDeviations(const Lincs& lincsd, ArrayRef<const RVec> x, const t_pbc* pbc)
2256 LincsDeviations result;
2257 const ArrayRef<const AtomPair> atoms = lincsd.atoms;
2258 const ArrayRef<const real> bllen = lincsd.bllen;
2259 const ArrayRef<const int> nlocat = lincsd.nlocat;
2261 for (int task = 0; task < lincsd.ntask; task++)
2263 for (int b = lincsd.task[task].b0; b < lincsd.task[task].b1; b++)
2268 pbc_dx_aiuc(pbc, x[atoms[b].index1], x[atoms[b].index2], dx);
2272 rvec_sub(x[atoms[b].index1], x[atoms[b].index2], dx);
2274 real r2 = ::norm2(dx);
2275 real len = r2 * gmx::invsqrt(r2);
2276 real d = std::abs(len / bllen[b] - 1.0_real);
2277 if (d > result.maxDeviation && (nlocat.empty() || nlocat[b]))
2279 result.maxDeviation = d;
2280 result.indexOfMaxDeviation = b;
2284 result.sumSquaredDeviation += d * d;
2285 result.numConstraints++;
2289 result.sumSquaredDeviation += nlocat[b] * d * d;
2290 result.numConstraints += nlocat[b];
2295 if (!nlocat.empty())
2297 result.numConstraints /= 2;
2298 result.sumSquaredDeviation *= 0.5;
2303 bool constrain_lincs(bool computeRmsd,
2304 const t_inputrec& ir,
2307 ArrayRef<const real> invmass,
2308 const t_commrec* cr,
2309 const gmx_multisim_t* ms,
2310 ArrayRefWithPadding<const RVec> xPadded,
2311 ArrayRefWithPadding<RVec> xprimePadded,
2312 ArrayRef<RVec> min_proj,
2315 const bool hasMassPerturbed,
2322 ConstraintVariable econq,
2329 /* This boolean should be set by a flag passed to this routine.
2330 * We can also easily check if any constraint length is changed,
2331 * if not dH/dlambda=0 and we can also set the boolean to FALSE.
2333 bool bCalcDHDL = (ir.efep != FreeEnergyPerturbationType::No && dvdlambda != nullptr);
2335 if (lincsd->nc == 0 && cr->dd == nullptr)
2340 ArrayRef<const RVec> x = xPadded.unpaddedArrayRef();
2341 ArrayRef<RVec> xprime = xprimePadded.unpaddedArrayRef();
2343 if (econq == ConstraintVariable::Positions)
2345 /* We can't use bCalcDHDL here, since NULL can be passed for dvdlambda
2346 * also with efep!=fepNO.
2348 if (ir.efep != FreeEnergyPerturbationType::No)
2350 if (hasMassPerturbed && lincsd->matlam != lambda)
2352 set_lincs_matrix(lincsd, invmass, lambda);
2355 for (int i = 0; i < lincsd->nc; i++)
2357 lincsd->bllen[i] = lincsd->bllen0[i] + lambda * lincsd->ddist[i];
2361 if (lincsd->ncg_flex)
2363 /* Set the flexible constraint lengths to the old lengths */
2366 for (int i = 0; i < lincsd->nc; i++)
2368 if (lincsd->bllen[i] == 0)
2371 pbc_dx_aiuc(pbc, x[lincsd->atoms[i].index1], x[lincsd->atoms[i].index2], dx);
2372 lincsd->bllen[i] = norm(dx);
2378 for (int i = 0; i < lincsd->nc; i++)
2380 if (lincsd->bllen[i] == 0)
2382 lincsd->bllen[i] = std::sqrt(
2383 distance2(x[lincsd->atoms[i].index1], x[lincsd->atoms[i].index2]));
2389 const bool printDebugOutput = ((debug != nullptr) && lincsd->nc > 0);
2390 if (printDebugOutput)
2392 LincsDeviations deviations = makeLincsDeviations(*lincsd, xprime, pbc);
2393 fprintf(debug, " Rel. Constraint Deviation: RMS MAX between atoms\n");
2395 " Before LINCS %.6f %.6f %6d %6d\n",
2396 std::sqrt(deviations.sumSquaredDeviation / deviations.numConstraints),
2397 deviations.maxDeviation,
2398 ddglatnr(cr->dd, lincsd->atoms[deviations.indexOfMaxDeviation].index1),
2399 ddglatnr(cr->dd, lincsd->atoms[deviations.indexOfMaxDeviation].index2));
2402 /* This bWarn var can be updated by multiple threads
2403 * at the same time. But as we only need to detect
2404 * if a warning occurred or not, this is not an issue.
2408 /* The OpenMP parallel region of constrain_lincs for coords */
2409 #pragma omp parallel num_threads(lincsd->ntask)
2413 int th = gmx_omp_get_thread_num();
2415 clear_mat(lincsd->task[th].vir_r_m_dr);
2431 th == 0 ? vir_r_m_dr : lincsd->task[th].vir_r_m_dr);
2433 GMX_CATCH_ALL_AND_EXIT_WITH_FATAL_ERROR
2436 if (computeRmsd || printDebugOutput || bWarn)
2438 LincsDeviations deviations = makeLincsDeviations(*lincsd, xprime, pbc);
2442 if (lincsd->callbackToRequireReduction.has_value())
2444 // This is reduced across domains in compute_globals and
2445 // reported to the log file.
2446 lincsd->rmsdReductionBuffer[0] = deviations.numConstraints;
2447 lincsd->rmsdReductionBuffer[1] = deviations.sumSquaredDeviation;
2449 // Call the ObservablesReducer via the callback it
2450 // gave us for the purpose.
2451 ObservablesReducerStatus status =
2452 lincsd->callbackToRequireReduction.value()(ReductionRequirement::Soon);
2453 GMX_RELEASE_ASSERT(status == ObservablesReducerStatus::ReadyToReduce,
2454 "The LINCS RMSD is computed after observables have been "
2455 "reduced, please reorder them.");
2459 // Compute the deviation directly
2460 lincsd->constraintRmsDeviation =
2461 std::sqrt(deviations.sumSquaredDeviation / deviations.numConstraints);
2464 if (printDebugOutput)
2467 " After LINCS %.6f %.6f %6d %6d\n\n",
2468 std::sqrt(deviations.sumSquaredDeviation / deviations.numConstraints),
2469 deviations.maxDeviation,
2470 ddglatnr(cr->dd, lincsd->atoms[deviations.indexOfMaxDeviation].index1),
2471 ddglatnr(cr->dd, lincsd->atoms[deviations.indexOfMaxDeviation].index2));
2476 if (maxwarn < INT_MAX)
2478 std::string simMesg;
2481 simMesg += gmx::formatString(" in simulation %d", ms->simulationIndex_);
2485 ", time %g (ps) LINCS WARNING%s\n"
2486 "relative constraint deviation after LINCS:\n"
2487 "rms %.6f, max %.6f (between atoms %d and %d)\n",
2489 ir.init_t + step * ir.delta_t,
2491 std::sqrt(deviations.sumSquaredDeviation / deviations.numConstraints),
2492 deviations.maxDeviation,
2493 ddglatnr(cr->dd, lincsd->atoms[deviations.indexOfMaxDeviation].index1),
2494 ddglatnr(cr->dd, lincsd->atoms[deviations.indexOfMaxDeviation].index2));
2497 cr->dd, x, xprime, pbc, lincsd->nc, lincsd->atoms, lincsd->bllen, ir.LincsWarnAngle, maxwarn, warncount);
2499 bOK = (deviations.maxDeviation < 0.5);
2503 if (lincsd->ncg_flex)
2505 for (int i = 0; (i < lincsd->nc); i++)
2507 if (lincsd->bllen0[i] == 0 && lincsd->ddist[i] == 0)
2509 lincsd->bllen[i] = 0;
2516 /* The OpenMP parallel region of constrain_lincs for derivatives */
2517 #pragma omp parallel num_threads(lincsd->ntask)
2521 int th = gmx_omp_get_thread_num();
2533 th == 0 ? vir_r_m_dr : lincsd->task[th].vir_r_m_dr);
2535 GMX_CATCH_ALL_AND_EXIT_WITH_FATAL_ERROR
2541 /* Reduce the dH/dlambda contributions over the threads */
2546 for (th = 0; th < lincsd->ntask; th++)
2548 dhdlambda += lincsd->task[th].dhdlambda;
2550 if (econq == ConstraintVariable::Positions)
2552 /* dhdlambda contains dH/dlambda*dt^2, correct for this */
2553 /* TODO This should probably use invdt, so that sd integrator scaling works properly */
2554 dhdlambda /= ir.delta_t * ir.delta_t;
2556 *dvdlambda += dhdlambda;
2559 if (bCalcVir && lincsd->ntask > 1)
2561 for (int i = 1; i < lincsd->ntask; i++)
2563 m_add(vir_r_m_dr, lincsd->task[i].vir_r_m_dr, vir_r_m_dr);
2567 /* count assuming nit=1 */
2568 inc_nrnb(nrnb, eNR_LINCS, lincsd->nc_real);
2569 inc_nrnb(nrnb, eNR_LINCSMAT, (2 + lincsd->nOrder) * lincsd->ncc);
2570 if (lincsd->ntriangle > 0)
2572 inc_nrnb(nrnb, eNR_LINCSMAT, lincsd->nOrder * lincsd->ncc_triangle);
2576 inc_nrnb(nrnb, eNR_CONSTR_V, lincsd->nc_real * 2);
2580 inc_nrnb(nrnb, eNR_CONSTR_VIR, lincsd->nc_real);