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, 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.
45 #include "gromacs/math/utilities.h"
53 #include "gmx_fatal.h"
59 #include "nonbonded.h"
61 #include "gromacs/simd/simd.h"
62 #include "gromacs/simd/simd_math.h"
63 #include "gromacs/simd/vector_operations.h"
65 /* Find a better place for this? */
66 const int cmap_coeff_matrix[] = {
67 1, 0, -3, 2, 0, 0, 0, 0, -3, 0, 9, -6, 2, 0, -6, 4,
68 0, 0, 0, 0, 0, 0, 0, 0, 3, 0, -9, 6, -2, 0, 6, -4,
69 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 9, -6, 0, 0, -6, 4,
70 0, 0, 3, -2, 0, 0, 0, 0, 0, 0, -9, 6, 0, 0, 6, -4,
71 0, 0, 0, 0, 1, 0, -3, 2, -2, 0, 6, -4, 1, 0, -3, 2,
72 0, 0, 0, 0, 0, 0, 0, 0, -1, 0, 3, -2, 1, 0, -3, 2,
73 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, -3, 2, 0, 0, 3, -2,
74 0, 0, 0, 0, 0, 0, 3, -2, 0, 0, -6, 4, 0, 0, 3, -2,
75 0, 1, -2, 1, 0, 0, 0, 0, 0, -3, 6, -3, 0, 2, -4, 2,
76 0, 0, 0, 0, 0, 0, 0, 0, 0, 3, -6, 3, 0, -2, 4, -2,
77 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, -3, 3, 0, 0, 2, -2,
78 0, 0, -1, 1, 0, 0, 0, 0, 0, 0, 3, -3, 0, 0, -2, 2,
79 0, 0, 0, 0, 0, 1, -2, 1, 0, -2, 4, -2, 0, 1, -2, 1,
80 0, 0, 0, 0, 0, 0, 0, 0, 0, -1, 2, -1, 0, 1, -2, 1,
81 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 1, -1, 0, 0, -1, 1,
82 0, 0, 0, 0, 0, 0, -1, 1, 0, 0, 2, -2, 0, 0, -1, 1
87 int glatnr(int *global_atom_index, int i)
91 if (global_atom_index == NULL)
97 atnr = global_atom_index[i] + 1;
103 static int pbc_rvec_sub(const t_pbc *pbc, const rvec xi, const rvec xj, rvec dx)
107 return pbc_dx_aiuc(pbc, xi, xj, dx);
111 rvec_sub(xi, xj, dx);
116 #ifdef GMX_SIMD_HAVE_REAL
118 /* SIMD PBC data structure, containing 1/boxdiag and the box vectors */
120 gmx_simd_real_t inv_bzz;
121 gmx_simd_real_t inv_byy;
122 gmx_simd_real_t inv_bxx;
131 /* Set the SIMD pbc data from a normal t_pbc struct */
132 static void set_pbc_simd(const t_pbc *pbc, pbc_simd_t *pbc_simd)
137 /* Setting inv_bdiag to 0 effectively turns off PBC */
138 clear_rvec(inv_bdiag);
141 for (d = 0; d < pbc->ndim_ePBC; d++)
143 inv_bdiag[d] = 1.0/pbc->box[d][d];
147 pbc_simd->inv_bzz = gmx_simd_set1_r(inv_bdiag[ZZ]);
148 pbc_simd->inv_byy = gmx_simd_set1_r(inv_bdiag[YY]);
149 pbc_simd->inv_bxx = gmx_simd_set1_r(inv_bdiag[XX]);
153 pbc_simd->bzx = gmx_simd_set1_r(pbc->box[ZZ][XX]);
154 pbc_simd->bzy = gmx_simd_set1_r(pbc->box[ZZ][YY]);
155 pbc_simd->bzz = gmx_simd_set1_r(pbc->box[ZZ][ZZ]);
156 pbc_simd->byx = gmx_simd_set1_r(pbc->box[YY][XX]);
157 pbc_simd->byy = gmx_simd_set1_r(pbc->box[YY][YY]);
158 pbc_simd->bxx = gmx_simd_set1_r(pbc->box[XX][XX]);
162 pbc_simd->bzx = gmx_simd_setzero_r();
163 pbc_simd->bzy = gmx_simd_setzero_r();
164 pbc_simd->bzz = gmx_simd_setzero_r();
165 pbc_simd->byx = gmx_simd_setzero_r();
166 pbc_simd->byy = gmx_simd_setzero_r();
167 pbc_simd->bxx = gmx_simd_setzero_r();
171 /* Correct distance vector *dx,*dy,*dz for PBC using SIMD */
172 static gmx_inline void
173 pbc_dx_simd(gmx_simd_real_t *dx, gmx_simd_real_t *dy, gmx_simd_real_t *dz,
174 const pbc_simd_t *pbc)
178 sh = gmx_simd_round_r(gmx_simd_mul_r(*dz, pbc->inv_bzz));
179 *dx = gmx_simd_fnmadd_r(sh, pbc->bzx, *dx);
180 *dy = gmx_simd_fnmadd_r(sh, pbc->bzy, *dy);
181 *dz = gmx_simd_fnmadd_r(sh, pbc->bzz, *dz);
183 sh = gmx_simd_round_r(gmx_simd_mul_r(*dy, pbc->inv_byy));
184 *dx = gmx_simd_fnmadd_r(sh, pbc->byx, *dx);
185 *dy = gmx_simd_fnmadd_r(sh, pbc->byy, *dy);
187 sh = gmx_simd_round_r(gmx_simd_mul_r(*dx, pbc->inv_bxx));
188 *dx = gmx_simd_fnmadd_r(sh, pbc->bxx, *dx);
191 #endif /* GMX_SIMD_HAVE_REAL */
194 * Morse potential bond by Frank Everdij
196 * Three parameters needed:
198 * b0 = equilibrium distance in nm
199 * be = beta in nm^-1 (actually, it's nu_e*Sqrt(2*pi*pi*mu/D_e))
200 * cb = well depth in kJ/mol
202 * Note: the potential is referenced to be +cb at infinite separation
203 * and zero at the equilibrium distance!
206 real morse_bonds(int nbonds,
207 const t_iatom forceatoms[], const t_iparams forceparams[],
208 const rvec x[], rvec f[], rvec fshift[],
209 const t_pbc *pbc, const t_graph *g,
210 real lambda, real *dvdlambda,
211 const t_mdatoms gmx_unused *md, t_fcdata gmx_unused *fcd,
212 int gmx_unused *global_atom_index)
214 const real one = 1.0;
215 const real two = 2.0;
216 real dr, dr2, temp, omtemp, cbomtemp, fbond, vbond, fij, vtot;
217 real b0, be, cb, b0A, beA, cbA, b0B, beB, cbB, L1;
219 int i, m, ki, type, ai, aj;
223 for (i = 0; (i < nbonds); )
225 type = forceatoms[i++];
226 ai = forceatoms[i++];
227 aj = forceatoms[i++];
229 b0A = forceparams[type].morse.b0A;
230 beA = forceparams[type].morse.betaA;
231 cbA = forceparams[type].morse.cbA;
233 b0B = forceparams[type].morse.b0B;
234 beB = forceparams[type].morse.betaB;
235 cbB = forceparams[type].morse.cbB;
237 L1 = one-lambda; /* 1 */
238 b0 = L1*b0A + lambda*b0B; /* 3 */
239 be = L1*beA + lambda*beB; /* 3 */
240 cb = L1*cbA + lambda*cbB; /* 3 */
242 ki = pbc_rvec_sub(pbc, x[ai], x[aj], dx); /* 3 */
243 dr2 = iprod(dx, dx); /* 5 */
244 dr = dr2*gmx_invsqrt(dr2); /* 10 */
245 temp = exp(-be*(dr-b0)); /* 12 */
249 /* bonds are constrainted. This may _not_ include bond constraints if they are lambda dependent */
250 *dvdlambda += cbB-cbA;
254 omtemp = one-temp; /* 1 */
255 cbomtemp = cb*omtemp; /* 1 */
256 vbond = cbomtemp*omtemp; /* 1 */
257 fbond = -two*be*temp*cbomtemp*gmx_invsqrt(dr2); /* 9 */
258 vtot += vbond; /* 1 */
260 *dvdlambda += (cbB - cbA) * omtemp * omtemp - (2-2*omtemp)*omtemp * cb * ((b0B-b0A)*be - (beB-beA)*(dr-b0)); /* 15 */
264 ivec_sub(SHIFT_IVEC(g, ai), SHIFT_IVEC(g, aj), dt);
268 for (m = 0; (m < DIM); m++) /* 15 */
273 fshift[ki][m] += fij;
274 fshift[CENTRAL][m] -= fij;
280 real cubic_bonds(int nbonds,
281 const t_iatom forceatoms[], const t_iparams forceparams[],
282 const rvec x[], rvec f[], rvec fshift[],
283 const t_pbc *pbc, const t_graph *g,
284 real gmx_unused lambda, real gmx_unused *dvdlambda,
285 const t_mdatoms gmx_unused *md, t_fcdata gmx_unused *fcd,
286 int gmx_unused *global_atom_index)
288 const real three = 3.0;
289 const real two = 2.0;
291 real dr, dr2, dist, kdist, kdist2, fbond, vbond, fij, vtot;
293 int i, m, ki, type, ai, aj;
297 for (i = 0; (i < nbonds); )
299 type = forceatoms[i++];
300 ai = forceatoms[i++];
301 aj = forceatoms[i++];
303 b0 = forceparams[type].cubic.b0;
304 kb = forceparams[type].cubic.kb;
305 kcub = forceparams[type].cubic.kcub;
307 ki = pbc_rvec_sub(pbc, x[ai], x[aj], dx); /* 3 */
308 dr2 = iprod(dx, dx); /* 5 */
315 dr = dr2*gmx_invsqrt(dr2); /* 10 */
320 vbond = kdist2 + kcub*kdist2*dist;
321 fbond = -(two*kdist + three*kdist2*kcub)/dr;
323 vtot += vbond; /* 21 */
327 ivec_sub(SHIFT_IVEC(g, ai), SHIFT_IVEC(g, aj), dt);
330 for (m = 0; (m < DIM); m++) /* 15 */
335 fshift[ki][m] += fij;
336 fshift[CENTRAL][m] -= fij;
342 real FENE_bonds(int nbonds,
343 const t_iatom forceatoms[], const t_iparams forceparams[],
344 const rvec x[], rvec f[], rvec fshift[],
345 const t_pbc *pbc, const t_graph *g,
346 real gmx_unused lambda, real gmx_unused *dvdlambda,
347 const t_mdatoms gmx_unused *md, t_fcdata gmx_unused *fcd,
348 int *global_atom_index)
350 const real half = 0.5;
351 const real one = 1.0;
353 real dr, dr2, bm2, omdr2obm2, fbond, vbond, fij, vtot;
355 int i, m, ki, type, ai, aj;
359 for (i = 0; (i < nbonds); )
361 type = forceatoms[i++];
362 ai = forceatoms[i++];
363 aj = forceatoms[i++];
365 bm = forceparams[type].fene.bm;
366 kb = forceparams[type].fene.kb;
368 ki = pbc_rvec_sub(pbc, x[ai], x[aj], dx); /* 3 */
369 dr2 = iprod(dx, dx); /* 5 */
381 "r^2 (%f) >= bm^2 (%f) in FENE bond between atoms %d and %d",
383 glatnr(global_atom_index, ai),
384 glatnr(global_atom_index, aj));
387 omdr2obm2 = one - dr2/bm2;
389 vbond = -half*kb*bm2*log(omdr2obm2);
390 fbond = -kb/omdr2obm2;
392 vtot += vbond; /* 35 */
396 ivec_sub(SHIFT_IVEC(g, ai), SHIFT_IVEC(g, aj), dt);
399 for (m = 0; (m < DIM); m++) /* 15 */
404 fshift[ki][m] += fij;
405 fshift[CENTRAL][m] -= fij;
411 real harmonic(real kA, real kB, real xA, real xB, real x, real lambda,
414 const real half = 0.5;
415 real L1, kk, x0, dx, dx2;
416 real v, f, dvdlambda;
419 kk = L1*kA+lambda*kB;
420 x0 = L1*xA+lambda*xB;
427 dvdlambda = half*(kB-kA)*dx2 + (xA-xB)*kk*dx;
434 /* That was 19 flops */
438 real bonds(int nbonds,
439 const t_iatom forceatoms[], const t_iparams forceparams[],
440 const rvec x[], rvec f[], rvec fshift[],
441 const t_pbc *pbc, const t_graph *g,
442 real lambda, real *dvdlambda,
443 const t_mdatoms gmx_unused *md, t_fcdata gmx_unused *fcd,
444 int gmx_unused *global_atom_index)
446 int i, m, ki, ai, aj, type;
447 real dr, dr2, fbond, vbond, fij, vtot;
452 for (i = 0; (i < nbonds); )
454 type = forceatoms[i++];
455 ai = forceatoms[i++];
456 aj = forceatoms[i++];
458 ki = pbc_rvec_sub(pbc, x[ai], x[aj], dx); /* 3 */
459 dr2 = iprod(dx, dx); /* 5 */
460 dr = dr2*gmx_invsqrt(dr2); /* 10 */
462 *dvdlambda += harmonic(forceparams[type].harmonic.krA,
463 forceparams[type].harmonic.krB,
464 forceparams[type].harmonic.rA,
465 forceparams[type].harmonic.rB,
466 dr, lambda, &vbond, &fbond); /* 19 */
474 vtot += vbond; /* 1*/
475 fbond *= gmx_invsqrt(dr2); /* 6 */
479 fprintf(debug, "BONDS: dr = %10g vbond = %10g fbond = %10g\n",
485 ivec_sub(SHIFT_IVEC(g, ai), SHIFT_IVEC(g, aj), dt);
488 for (m = 0; (m < DIM); m++) /* 15 */
493 fshift[ki][m] += fij;
494 fshift[CENTRAL][m] -= fij;
500 real restraint_bonds(int nbonds,
501 const t_iatom forceatoms[], const t_iparams forceparams[],
502 const rvec x[], rvec f[], rvec fshift[],
503 const t_pbc *pbc, const t_graph *g,
504 real lambda, real *dvdlambda,
505 const t_mdatoms gmx_unused *md, t_fcdata gmx_unused *fcd,
506 int gmx_unused *global_atom_index)
508 int i, m, ki, ai, aj, type;
509 real dr, dr2, fbond, vbond, fij, vtot;
511 real low, dlow, up1, dup1, up2, dup2, k, dk;
519 for (i = 0; (i < nbonds); )
521 type = forceatoms[i++];
522 ai = forceatoms[i++];
523 aj = forceatoms[i++];
525 ki = pbc_rvec_sub(pbc, x[ai], x[aj], dx); /* 3 */
526 dr2 = iprod(dx, dx); /* 5 */
527 dr = dr2*gmx_invsqrt(dr2); /* 10 */
529 low = L1*forceparams[type].restraint.lowA + lambda*forceparams[type].restraint.lowB;
530 dlow = -forceparams[type].restraint.lowA + forceparams[type].restraint.lowB;
531 up1 = L1*forceparams[type].restraint.up1A + lambda*forceparams[type].restraint.up1B;
532 dup1 = -forceparams[type].restraint.up1A + forceparams[type].restraint.up1B;
533 up2 = L1*forceparams[type].restraint.up2A + lambda*forceparams[type].restraint.up2B;
534 dup2 = -forceparams[type].restraint.up2A + forceparams[type].restraint.up2B;
535 k = L1*forceparams[type].restraint.kA + lambda*forceparams[type].restraint.kB;
536 dk = -forceparams[type].restraint.kA + forceparams[type].restraint.kB;
545 *dvdlambda += 0.5*dk*drh2 - k*dlow*drh;
558 *dvdlambda += 0.5*dk*drh2 - k*dup1*drh;
563 vbond = k*(up2 - up1)*(0.5*(up2 - up1) + drh);
564 fbond = -k*(up2 - up1);
565 *dvdlambda += dk*(up2 - up1)*(0.5*(up2 - up1) + drh)
566 + k*(dup2 - dup1)*(up2 - up1 + drh)
567 - k*(up2 - up1)*dup2;
575 vtot += vbond; /* 1*/
576 fbond *= gmx_invsqrt(dr2); /* 6 */
580 fprintf(debug, "BONDS: dr = %10g vbond = %10g fbond = %10g\n",
586 ivec_sub(SHIFT_IVEC(g, ai), SHIFT_IVEC(g, aj), dt);
589 for (m = 0; (m < DIM); m++) /* 15 */
594 fshift[ki][m] += fij;
595 fshift[CENTRAL][m] -= fij;
602 real polarize(int nbonds,
603 const t_iatom forceatoms[], const t_iparams forceparams[],
604 const rvec x[], rvec f[], rvec fshift[],
605 const t_pbc *pbc, const t_graph *g,
606 real lambda, real *dvdlambda,
607 const t_mdatoms *md, t_fcdata gmx_unused *fcd,
608 int gmx_unused *global_atom_index)
610 int i, m, ki, ai, aj, type;
611 real dr, dr2, fbond, vbond, fij, vtot, ksh;
616 for (i = 0; (i < nbonds); )
618 type = forceatoms[i++];
619 ai = forceatoms[i++];
620 aj = forceatoms[i++];
621 ksh = sqr(md->chargeA[aj])*ONE_4PI_EPS0/forceparams[type].polarize.alpha;
624 fprintf(debug, "POL: local ai = %d aj = %d ksh = %.3f\n", ai, aj, ksh);
627 ki = pbc_rvec_sub(pbc, x[ai], x[aj], dx); /* 3 */
628 dr2 = iprod(dx, dx); /* 5 */
629 dr = dr2*gmx_invsqrt(dr2); /* 10 */
631 *dvdlambda += harmonic(ksh, ksh, 0, 0, dr, lambda, &vbond, &fbond); /* 19 */
638 vtot += vbond; /* 1*/
639 fbond *= gmx_invsqrt(dr2); /* 6 */
643 ivec_sub(SHIFT_IVEC(g, ai), SHIFT_IVEC(g, aj), dt);
646 for (m = 0; (m < DIM); m++) /* 15 */
651 fshift[ki][m] += fij;
652 fshift[CENTRAL][m] -= fij;
658 real anharm_polarize(int nbonds,
659 const t_iatom forceatoms[], const t_iparams forceparams[],
660 const rvec x[], rvec f[], rvec fshift[],
661 const t_pbc *pbc, const t_graph *g,
662 real lambda, real *dvdlambda,
663 const t_mdatoms *md, t_fcdata gmx_unused *fcd,
664 int gmx_unused *global_atom_index)
666 int i, m, ki, ai, aj, type;
667 real dr, dr2, fbond, vbond, fij, vtot, ksh, khyp, drcut, ddr, ddr3;
672 for (i = 0; (i < nbonds); )
674 type = forceatoms[i++];
675 ai = forceatoms[i++];
676 aj = forceatoms[i++];
677 ksh = sqr(md->chargeA[aj])*ONE_4PI_EPS0/forceparams[type].anharm_polarize.alpha; /* 7*/
678 khyp = forceparams[type].anharm_polarize.khyp;
679 drcut = forceparams[type].anharm_polarize.drcut;
682 fprintf(debug, "POL: local ai = %d aj = %d ksh = %.3f\n", ai, aj, ksh);
685 ki = pbc_rvec_sub(pbc, x[ai], x[aj], dx); /* 3 */
686 dr2 = iprod(dx, dx); /* 5 */
687 dr = dr2*gmx_invsqrt(dr2); /* 10 */
689 *dvdlambda += harmonic(ksh, ksh, 0, 0, dr, lambda, &vbond, &fbond); /* 19 */
700 vbond += khyp*ddr*ddr3;
701 fbond -= 4*khyp*ddr3;
703 fbond *= gmx_invsqrt(dr2); /* 6 */
704 vtot += vbond; /* 1*/
708 ivec_sub(SHIFT_IVEC(g, ai), SHIFT_IVEC(g, aj), dt);
711 for (m = 0; (m < DIM); m++) /* 15 */
716 fshift[ki][m] += fij;
717 fshift[CENTRAL][m] -= fij;
723 real water_pol(int nbonds,
724 const t_iatom forceatoms[], const t_iparams forceparams[],
725 const rvec x[], rvec f[], rvec gmx_unused fshift[],
726 const t_pbc gmx_unused *pbc, const t_graph gmx_unused *g,
727 real gmx_unused lambda, real gmx_unused *dvdlambda,
728 const t_mdatoms gmx_unused *md, t_fcdata gmx_unused *fcd,
729 int gmx_unused *global_atom_index)
731 /* This routine implements anisotropic polarizibility for water, through
732 * a shell connected to a dummy with spring constant that differ in the
733 * three spatial dimensions in the molecular frame.
735 int i, m, aO, aH1, aH2, aD, aS, type, type0;
736 rvec dOH1, dOH2, dHH, dOD, dDS, nW, kk, dx, kdx, proj;
740 real vtot, fij, r_HH, r_OD, r_nW, tx, ty, tz, qS;
745 type0 = forceatoms[0];
747 qS = md->chargeA[aS];
748 kk[XX] = sqr(qS)*ONE_4PI_EPS0/forceparams[type0].wpol.al_x;
749 kk[YY] = sqr(qS)*ONE_4PI_EPS0/forceparams[type0].wpol.al_y;
750 kk[ZZ] = sqr(qS)*ONE_4PI_EPS0/forceparams[type0].wpol.al_z;
751 r_HH = 1.0/forceparams[type0].wpol.rHH;
752 r_OD = 1.0/forceparams[type0].wpol.rOD;
755 fprintf(debug, "WPOL: qS = %10.5f aS = %5d\n", qS, aS);
756 fprintf(debug, "WPOL: kk = %10.3f %10.3f %10.3f\n",
757 kk[XX], kk[YY], kk[ZZ]);
758 fprintf(debug, "WPOL: rOH = %10.3f rHH = %10.3f rOD = %10.3f\n",
759 forceparams[type0].wpol.rOH,
760 forceparams[type0].wpol.rHH,
761 forceparams[type0].wpol.rOD);
763 for (i = 0; (i < nbonds); i += 6)
765 type = forceatoms[i];
768 gmx_fatal(FARGS, "Sorry, type = %d, type0 = %d, file = %s, line = %d",
769 type, type0, __FILE__, __LINE__);
771 aO = forceatoms[i+1];
772 aH1 = forceatoms[i+2];
773 aH2 = forceatoms[i+3];
774 aD = forceatoms[i+4];
775 aS = forceatoms[i+5];
777 /* Compute vectors describing the water frame */
778 rvec_sub(x[aH1], x[aO], dOH1);
779 rvec_sub(x[aH2], x[aO], dOH2);
780 rvec_sub(x[aH2], x[aH1], dHH);
781 rvec_sub(x[aD], x[aO], dOD);
782 rvec_sub(x[aS], x[aD], dDS);
783 cprod(dOH1, dOH2, nW);
785 /* Compute inverse length of normal vector
786 * (this one could be precomputed, but I'm too lazy now)
788 r_nW = gmx_invsqrt(iprod(nW, nW));
789 /* This is for precision, but does not make a big difference,
792 r_OD = gmx_invsqrt(iprod(dOD, dOD));
794 /* Normalize the vectors in the water frame */
796 svmul(r_HH, dHH, dHH);
797 svmul(r_OD, dOD, dOD);
799 /* Compute displacement of shell along components of the vector */
800 dx[ZZ] = iprod(dDS, dOD);
801 /* Compute projection on the XY plane: dDS - dx[ZZ]*dOD */
802 for (m = 0; (m < DIM); m++)
804 proj[m] = dDS[m]-dx[ZZ]*dOD[m];
807 /*dx[XX] = iprod(dDS,nW);
808 dx[YY] = iprod(dDS,dHH);*/
809 dx[XX] = iprod(proj, nW);
810 for (m = 0; (m < DIM); m++)
812 proj[m] -= dx[XX]*nW[m];
814 dx[YY] = iprod(proj, dHH);
819 fprintf(debug, "WPOL: dx2=%10g dy2=%10g dz2=%10g sum=%10g dDS^2=%10g\n",
820 sqr(dx[XX]), sqr(dx[YY]), sqr(dx[ZZ]), iprod(dx, dx), iprod(dDS, dDS));
821 fprintf(debug, "WPOL: dHH=(%10g,%10g,%10g)\n", dHH[XX], dHH[YY], dHH[ZZ]);
822 fprintf(debug, "WPOL: dOD=(%10g,%10g,%10g), 1/r_OD = %10g\n",
823 dOD[XX], dOD[YY], dOD[ZZ], 1/r_OD);
824 fprintf(debug, "WPOL: nW =(%10g,%10g,%10g), 1/r_nW = %10g\n",
825 nW[XX], nW[YY], nW[ZZ], 1/r_nW);
826 fprintf(debug, "WPOL: dx =%10g, dy =%10g, dz =%10g\n",
827 dx[XX], dx[YY], dx[ZZ]);
828 fprintf(debug, "WPOL: dDSx=%10g, dDSy=%10g, dDSz=%10g\n",
829 dDS[XX], dDS[YY], dDS[ZZ]);
832 /* Now compute the forces and energy */
833 kdx[XX] = kk[XX]*dx[XX];
834 kdx[YY] = kk[YY]*dx[YY];
835 kdx[ZZ] = kk[ZZ]*dx[ZZ];
836 vtot += iprod(dx, kdx);
837 for (m = 0; (m < DIM); m++)
839 /* This is a tensor operation but written out for speed */
853 fprintf(debug, "WPOL: vwpol=%g\n", 0.5*iprod(dx, kdx));
854 fprintf(debug, "WPOL: df = (%10g, %10g, %10g)\n", df[XX], df[YY], df[ZZ]);
862 static real do_1_thole(const rvec xi, const rvec xj, rvec fi, rvec fj,
863 const t_pbc *pbc, real qq,
864 rvec fshift[], real afac)
867 real r12sq, r12_1, r12n, r12bar, v0, v1, fscal, ebar, fff;
870 t = pbc_rvec_sub(pbc, xi, xj, r12); /* 3 */
872 r12sq = iprod(r12, r12); /* 5 */
873 r12_1 = gmx_invsqrt(r12sq); /* 5 */
874 r12bar = afac/r12_1; /* 5 */
875 v0 = qq*ONE_4PI_EPS0*r12_1; /* 2 */
876 ebar = exp(-r12bar); /* 5 */
877 v1 = (1-(1+0.5*r12bar)*ebar); /* 4 */
878 fscal = ((v0*r12_1)*v1 - v0*0.5*afac*ebar*(r12bar+1))*r12_1; /* 9 */
881 fprintf(debug, "THOLE: v0 = %.3f v1 = %.3f r12= % .3f r12bar = %.3f fscal = %.3f ebar = %.3f\n", v0, v1, 1/r12_1, r12bar, fscal, ebar);
884 for (m = 0; (m < DIM); m++)
890 fshift[CENTRAL][m] -= fff;
893 return v0*v1; /* 1 */
897 real thole_pol(int nbonds,
898 const t_iatom forceatoms[], const t_iparams forceparams[],
899 const rvec x[], rvec f[], rvec fshift[],
900 const t_pbc *pbc, const t_graph gmx_unused *g,
901 real gmx_unused lambda, real gmx_unused *dvdlambda,
902 const t_mdatoms *md, t_fcdata gmx_unused *fcd,
903 int gmx_unused *global_atom_index)
905 /* Interaction between two pairs of particles with opposite charge */
906 int i, type, a1, da1, a2, da2;
907 real q1, q2, qq, a, al1, al2, afac;
910 for (i = 0; (i < nbonds); )
912 type = forceatoms[i++];
913 a1 = forceatoms[i++];
914 da1 = forceatoms[i++];
915 a2 = forceatoms[i++];
916 da2 = forceatoms[i++];
917 q1 = md->chargeA[da1];
918 q2 = md->chargeA[da2];
919 a = forceparams[type].thole.a;
920 al1 = forceparams[type].thole.alpha1;
921 al2 = forceparams[type].thole.alpha2;
923 afac = a*pow(al1*al2, -1.0/6.0);
924 V += do_1_thole(x[a1], x[a2], f[a1], f[a2], pbc, qq, fshift, afac);
925 V += do_1_thole(x[da1], x[a2], f[da1], f[a2], pbc, -qq, fshift, afac);
926 V += do_1_thole(x[a1], x[da2], f[a1], f[da2], pbc, -qq, fshift, afac);
927 V += do_1_thole(x[da1], x[da2], f[da1], f[da2], pbc, qq, fshift, afac);
933 real bond_angle(const rvec xi, const rvec xj, const rvec xk, const t_pbc *pbc,
934 rvec r_ij, rvec r_kj, real *costh,
936 /* Return value is the angle between the bonds i-j and j-k */
941 *t1 = pbc_rvec_sub(pbc, xi, xj, r_ij); /* 3 */
942 *t2 = pbc_rvec_sub(pbc, xk, xj, r_kj); /* 3 */
944 *costh = cos_angle(r_ij, r_kj); /* 25 */
945 th = acos(*costh); /* 10 */
950 real angles(int nbonds,
951 const t_iatom forceatoms[], const t_iparams forceparams[],
952 const rvec x[], rvec f[], rvec fshift[],
953 const t_pbc *pbc, const t_graph *g,
954 real lambda, real *dvdlambda,
955 const t_mdatoms gmx_unused *md, t_fcdata gmx_unused *fcd,
956 int gmx_unused *global_atom_index)
958 int i, ai, aj, ak, t1, t2, type;
960 real cos_theta, cos_theta2, theta, dVdt, va, vtot;
961 ivec jt, dt_ij, dt_kj;
964 for (i = 0; i < nbonds; )
966 type = forceatoms[i++];
967 ai = forceatoms[i++];
968 aj = forceatoms[i++];
969 ak = forceatoms[i++];
971 theta = bond_angle(x[ai], x[aj], x[ak], pbc,
972 r_ij, r_kj, &cos_theta, &t1, &t2); /* 41 */
974 *dvdlambda += harmonic(forceparams[type].harmonic.krA,
975 forceparams[type].harmonic.krB,
976 forceparams[type].harmonic.rA*DEG2RAD,
977 forceparams[type].harmonic.rB*DEG2RAD,
978 theta, lambda, &va, &dVdt); /* 21 */
981 cos_theta2 = sqr(cos_theta);
991 st = dVdt*gmx_invsqrt(1 - cos_theta2); /* 12 */
992 sth = st*cos_theta; /* 1 */
996 fprintf(debug, "ANGLES: theta = %10g vth = %10g dV/dtheta = %10g\n",
997 theta*RAD2DEG, va, dVdt);
1000 nrij2 = iprod(r_ij, r_ij); /* 5 */
1001 nrkj2 = iprod(r_kj, r_kj); /* 5 */
1003 nrij_1 = gmx_invsqrt(nrij2); /* 10 */
1004 nrkj_1 = gmx_invsqrt(nrkj2); /* 10 */
1006 cik = st*nrij_1*nrkj_1; /* 2 */
1007 cii = sth*nrij_1*nrij_1; /* 2 */
1008 ckk = sth*nrkj_1*nrkj_1; /* 2 */
1010 for (m = 0; m < DIM; m++)
1012 f_i[m] = -(cik*r_kj[m] - cii*r_ij[m]);
1013 f_k[m] = -(cik*r_ij[m] - ckk*r_kj[m]);
1014 f_j[m] = -f_i[m] - f_k[m];
1021 copy_ivec(SHIFT_IVEC(g, aj), jt);
1023 ivec_sub(SHIFT_IVEC(g, ai), jt, dt_ij);
1024 ivec_sub(SHIFT_IVEC(g, ak), jt, dt_kj);
1025 t1 = IVEC2IS(dt_ij);
1026 t2 = IVEC2IS(dt_kj);
1028 rvec_inc(fshift[t1], f_i);
1029 rvec_inc(fshift[CENTRAL], f_j);
1030 rvec_inc(fshift[t2], f_k);
1037 #ifdef GMX_SIMD_HAVE_REAL
1039 /* As angles, but using SIMD to calculate many dihedrals at once.
1040 * This routines does not calculate energies and shift forces.
1042 static gmx_inline void
1043 angles_noener_simd(int nbonds,
1044 const t_iatom forceatoms[], const t_iparams forceparams[],
1045 const rvec x[], rvec f[],
1046 const t_pbc *pbc, const t_graph gmx_unused *g,
1047 real gmx_unused lambda,
1048 const t_mdatoms gmx_unused *md, t_fcdata gmx_unused *fcd,
1049 int gmx_unused *global_atom_index)
1053 int type, ai[GMX_SIMD_REAL_WIDTH], aj[GMX_SIMD_REAL_WIDTH];
1054 int ak[GMX_SIMD_REAL_WIDTH];
1055 real coeff_array[2*GMX_SIMD_REAL_WIDTH+GMX_SIMD_REAL_WIDTH], *coeff;
1056 real dr_array[2*DIM*GMX_SIMD_REAL_WIDTH+GMX_SIMD_REAL_WIDTH], *dr;
1057 real f_buf_array[6*GMX_SIMD_REAL_WIDTH+GMX_SIMD_REAL_WIDTH], *f_buf;
1058 gmx_simd_real_t k_S, theta0_S;
1059 gmx_simd_real_t rijx_S, rijy_S, rijz_S;
1060 gmx_simd_real_t rkjx_S, rkjy_S, rkjz_S;
1061 gmx_simd_real_t one_S;
1062 gmx_simd_real_t min_one_plus_eps_S;
1063 gmx_simd_real_t rij_rkj_S;
1064 gmx_simd_real_t nrij2_S, nrij_1_S;
1065 gmx_simd_real_t nrkj2_S, nrkj_1_S;
1066 gmx_simd_real_t cos_S, invsin_S;
1067 gmx_simd_real_t theta_S;
1068 gmx_simd_real_t st_S, sth_S;
1069 gmx_simd_real_t cik_S, cii_S, ckk_S;
1070 gmx_simd_real_t f_ix_S, f_iy_S, f_iz_S;
1071 gmx_simd_real_t f_kx_S, f_ky_S, f_kz_S;
1072 pbc_simd_t pbc_simd;
1074 /* Ensure register memory alignment */
1075 coeff = gmx_simd_align_r(coeff_array);
1076 dr = gmx_simd_align_r(dr_array);
1077 f_buf = gmx_simd_align_r(f_buf_array);
1079 set_pbc_simd(pbc, &pbc_simd);
1081 one_S = gmx_simd_set1_r(1.0);
1083 /* The smallest number > -1 */
1084 min_one_plus_eps_S = gmx_simd_set1_r(-1.0 + 2*GMX_REAL_EPS);
1086 /* nbonds is the number of angles times nfa1, here we step GMX_SIMD_REAL_WIDTH angles */
1087 for (i = 0; (i < nbonds); i += GMX_SIMD_REAL_WIDTH*nfa1)
1089 /* Collect atoms for GMX_SIMD_REAL_WIDTH angles.
1090 * iu indexes into forceatoms, we should not let iu go beyond nbonds.
1093 for (s = 0; s < GMX_SIMD_REAL_WIDTH; s++)
1095 type = forceatoms[iu];
1096 ai[s] = forceatoms[iu+1];
1097 aj[s] = forceatoms[iu+2];
1098 ak[s] = forceatoms[iu+3];
1100 coeff[s] = forceparams[type].harmonic.krA;
1101 coeff[GMX_SIMD_REAL_WIDTH+s] = forceparams[type].harmonic.rA*DEG2RAD;
1103 /* If you can't use pbc_dx_simd below for PBC, e.g. because
1104 * you can't round in SIMD, use pbc_rvec_sub here.
1106 /* Store the non PBC corrected distances packed and aligned */
1107 for (m = 0; m < DIM; m++)
1109 dr[s + m *GMX_SIMD_REAL_WIDTH] = x[ai[s]][m] - x[aj[s]][m];
1110 dr[s + (DIM+m)*GMX_SIMD_REAL_WIDTH] = x[ak[s]][m] - x[aj[s]][m];
1113 /* At the end fill the arrays with identical entries */
1114 if (iu + nfa1 < nbonds)
1120 k_S = gmx_simd_load_r(coeff);
1121 theta0_S = gmx_simd_load_r(coeff+GMX_SIMD_REAL_WIDTH);
1123 rijx_S = gmx_simd_load_r(dr + 0*GMX_SIMD_REAL_WIDTH);
1124 rijy_S = gmx_simd_load_r(dr + 1*GMX_SIMD_REAL_WIDTH);
1125 rijz_S = gmx_simd_load_r(dr + 2*GMX_SIMD_REAL_WIDTH);
1126 rkjx_S = gmx_simd_load_r(dr + 3*GMX_SIMD_REAL_WIDTH);
1127 rkjy_S = gmx_simd_load_r(dr + 4*GMX_SIMD_REAL_WIDTH);
1128 rkjz_S = gmx_simd_load_r(dr + 5*GMX_SIMD_REAL_WIDTH);
1130 pbc_dx_simd(&rijx_S, &rijy_S, &rijz_S, &pbc_simd);
1131 pbc_dx_simd(&rkjx_S, &rkjy_S, &rkjz_S, &pbc_simd);
1133 rij_rkj_S = gmx_simd_iprod_r(rijx_S, rijy_S, rijz_S,
1134 rkjx_S, rkjy_S, rkjz_S);
1136 nrij2_S = gmx_simd_norm2_r(rijx_S, rijy_S, rijz_S);
1137 nrkj2_S = gmx_simd_norm2_r(rkjx_S, rkjy_S, rkjz_S);
1139 nrij_1_S = gmx_simd_invsqrt_r(nrij2_S);
1140 nrkj_1_S = gmx_simd_invsqrt_r(nrkj2_S);
1142 cos_S = gmx_simd_mul_r(rij_rkj_S, gmx_simd_mul_r(nrij_1_S, nrkj_1_S));
1144 /* To allow for 180 degrees, we take the max of cos and -1 + 1bit,
1145 * so we can safely get the 1/sin from 1/sqrt(1 - cos^2).
1146 * This also ensures that rounding errors would cause the argument
1147 * of gmx_simd_acos_r to be < -1.
1148 * Note that we do not take precautions for cos(0)=1, so the outer
1149 * atoms in an angle should not be on top of each other.
1151 cos_S = gmx_simd_max_r(cos_S, min_one_plus_eps_S);
1153 theta_S = gmx_simd_acos_r(cos_S);
1155 invsin_S = gmx_simd_invsqrt_r(gmx_simd_sub_r(one_S, gmx_simd_mul_r(cos_S, cos_S)));
1157 st_S = gmx_simd_mul_r(gmx_simd_mul_r(k_S, gmx_simd_sub_r(theta0_S, theta_S)),
1159 sth_S = gmx_simd_mul_r(st_S, cos_S);
1161 cik_S = gmx_simd_mul_r(st_S, gmx_simd_mul_r(nrij_1_S, nrkj_1_S));
1162 cii_S = gmx_simd_mul_r(sth_S, gmx_simd_mul_r(nrij_1_S, nrij_1_S));
1163 ckk_S = gmx_simd_mul_r(sth_S, gmx_simd_mul_r(nrkj_1_S, nrkj_1_S));
1165 f_ix_S = gmx_simd_mul_r(cii_S, rijx_S);
1166 f_ix_S = gmx_simd_fnmadd_r(cik_S, rkjx_S, f_ix_S);
1167 f_iy_S = gmx_simd_mul_r(cii_S, rijy_S);
1168 f_iy_S = gmx_simd_fnmadd_r(cik_S, rkjy_S, f_iy_S);
1169 f_iz_S = gmx_simd_mul_r(cii_S, rijz_S);
1170 f_iz_S = gmx_simd_fnmadd_r(cik_S, rkjz_S, f_iz_S);
1171 f_kx_S = gmx_simd_mul_r(ckk_S, rkjx_S);
1172 f_kx_S = gmx_simd_fnmadd_r(cik_S, rijx_S, f_kx_S);
1173 f_ky_S = gmx_simd_mul_r(ckk_S, rkjy_S);
1174 f_ky_S = gmx_simd_fnmadd_r(cik_S, rijy_S, f_ky_S);
1175 f_kz_S = gmx_simd_mul_r(ckk_S, rkjz_S);
1176 f_kz_S = gmx_simd_fnmadd_r(cik_S, rijz_S, f_kz_S);
1178 gmx_simd_store_r(f_buf + 0*GMX_SIMD_REAL_WIDTH, f_ix_S);
1179 gmx_simd_store_r(f_buf + 1*GMX_SIMD_REAL_WIDTH, f_iy_S);
1180 gmx_simd_store_r(f_buf + 2*GMX_SIMD_REAL_WIDTH, f_iz_S);
1181 gmx_simd_store_r(f_buf + 3*GMX_SIMD_REAL_WIDTH, f_kx_S);
1182 gmx_simd_store_r(f_buf + 4*GMX_SIMD_REAL_WIDTH, f_ky_S);
1183 gmx_simd_store_r(f_buf + 5*GMX_SIMD_REAL_WIDTH, f_kz_S);
1189 for (m = 0; m < DIM; m++)
1191 f[ai[s]][m] += f_buf[s + m*GMX_SIMD_REAL_WIDTH];
1192 f[aj[s]][m] -= f_buf[s + m*GMX_SIMD_REAL_WIDTH] + f_buf[s + (DIM+m)*GMX_SIMD_REAL_WIDTH];
1193 f[ak[s]][m] += f_buf[s + (DIM+m)*GMX_SIMD_REAL_WIDTH];
1198 while (s < GMX_SIMD_REAL_WIDTH && iu < nbonds);
1202 #endif /* GMX_SIMD_HAVE_REAL */
1204 real linear_angles(int nbonds,
1205 const t_iatom forceatoms[], const t_iparams forceparams[],
1206 const rvec x[], rvec f[], rvec fshift[],
1207 const t_pbc *pbc, const t_graph *g,
1208 real lambda, real *dvdlambda,
1209 const t_mdatoms gmx_unused *md, t_fcdata gmx_unused *fcd,
1210 int gmx_unused *global_atom_index)
1212 int i, m, ai, aj, ak, t1, t2, type;
1214 real L1, kA, kB, aA, aB, dr, dr2, va, vtot, a, b, klin;
1215 ivec jt, dt_ij, dt_kj;
1216 rvec r_ij, r_kj, r_ik, dx;
1220 for (i = 0; (i < nbonds); )
1222 type = forceatoms[i++];
1223 ai = forceatoms[i++];
1224 aj = forceatoms[i++];
1225 ak = forceatoms[i++];
1227 kA = forceparams[type].linangle.klinA;
1228 kB = forceparams[type].linangle.klinB;
1229 klin = L1*kA + lambda*kB;
1231 aA = forceparams[type].linangle.aA;
1232 aB = forceparams[type].linangle.aB;
1233 a = L1*aA+lambda*aB;
1236 t1 = pbc_rvec_sub(pbc, x[ai], x[aj], r_ij);
1237 t2 = pbc_rvec_sub(pbc, x[ak], x[aj], r_kj);
1238 rvec_sub(r_ij, r_kj, r_ik);
1241 for (m = 0; (m < DIM); m++)
1243 dr = -a * r_ij[m] - b * r_kj[m];
1248 f_j[m] = -(f_i[m]+f_k[m]);
1254 *dvdlambda += 0.5*(kB-kA)*dr2 + klin*(aB-aA)*iprod(dx, r_ik);
1260 copy_ivec(SHIFT_IVEC(g, aj), jt);
1262 ivec_sub(SHIFT_IVEC(g, ai), jt, dt_ij);
1263 ivec_sub(SHIFT_IVEC(g, ak), jt, dt_kj);
1264 t1 = IVEC2IS(dt_ij);
1265 t2 = IVEC2IS(dt_kj);
1267 rvec_inc(fshift[t1], f_i);
1268 rvec_inc(fshift[CENTRAL], f_j);
1269 rvec_inc(fshift[t2], f_k);
1274 real urey_bradley(int nbonds,
1275 const t_iatom forceatoms[], const t_iparams forceparams[],
1276 const rvec x[], rvec f[], rvec fshift[],
1277 const t_pbc *pbc, const t_graph *g,
1278 real lambda, real *dvdlambda,
1279 const t_mdatoms gmx_unused *md, t_fcdata gmx_unused *fcd,
1280 int gmx_unused *global_atom_index)
1282 int i, m, ai, aj, ak, t1, t2, type, ki;
1283 rvec r_ij, r_kj, r_ik;
1284 real cos_theta, cos_theta2, theta;
1285 real dVdt, va, vtot, dr, dr2, vbond, fbond, fik;
1286 real kthA, th0A, kUBA, r13A, kthB, th0B, kUBB, r13B;
1287 ivec jt, dt_ij, dt_kj, dt_ik;
1290 for (i = 0; (i < nbonds); )
1292 type = forceatoms[i++];
1293 ai = forceatoms[i++];
1294 aj = forceatoms[i++];
1295 ak = forceatoms[i++];
1296 th0A = forceparams[type].u_b.thetaA*DEG2RAD;
1297 kthA = forceparams[type].u_b.kthetaA;
1298 r13A = forceparams[type].u_b.r13A;
1299 kUBA = forceparams[type].u_b.kUBA;
1300 th0B = forceparams[type].u_b.thetaB*DEG2RAD;
1301 kthB = forceparams[type].u_b.kthetaB;
1302 r13B = forceparams[type].u_b.r13B;
1303 kUBB = forceparams[type].u_b.kUBB;
1305 theta = bond_angle(x[ai], x[aj], x[ak], pbc,
1306 r_ij, r_kj, &cos_theta, &t1, &t2); /* 41 */
1308 *dvdlambda += harmonic(kthA, kthB, th0A, th0B, theta, lambda, &va, &dVdt); /* 21 */
1311 ki = pbc_rvec_sub(pbc, x[ai], x[ak], r_ik); /* 3 */
1312 dr2 = iprod(r_ik, r_ik); /* 5 */
1313 dr = dr2*gmx_invsqrt(dr2); /* 10 */
1315 *dvdlambda += harmonic(kUBA, kUBB, r13A, r13B, dr, lambda, &vbond, &fbond); /* 19 */
1317 cos_theta2 = sqr(cos_theta); /* 1 */
1325 st = dVdt*gmx_invsqrt(1 - cos_theta2); /* 12 */
1326 sth = st*cos_theta; /* 1 */
1330 fprintf(debug, "ANGLES: theta = %10g vth = %10g dV/dtheta = %10g\n",
1331 theta*RAD2DEG, va, dVdt);
1334 nrkj2 = iprod(r_kj, r_kj); /* 5 */
1335 nrij2 = iprod(r_ij, r_ij);
1337 cik = st*gmx_invsqrt(nrkj2*nrij2); /* 12 */
1338 cii = sth/nrij2; /* 10 */
1339 ckk = sth/nrkj2; /* 10 */
1341 for (m = 0; (m < DIM); m++) /* 39 */
1343 f_i[m] = -(cik*r_kj[m]-cii*r_ij[m]);
1344 f_k[m] = -(cik*r_ij[m]-ckk*r_kj[m]);
1345 f_j[m] = -f_i[m]-f_k[m];
1352 copy_ivec(SHIFT_IVEC(g, aj), jt);
1354 ivec_sub(SHIFT_IVEC(g, ai), jt, dt_ij);
1355 ivec_sub(SHIFT_IVEC(g, ak), jt, dt_kj);
1356 t1 = IVEC2IS(dt_ij);
1357 t2 = IVEC2IS(dt_kj);
1359 rvec_inc(fshift[t1], f_i);
1360 rvec_inc(fshift[CENTRAL], f_j);
1361 rvec_inc(fshift[t2], f_k);
1363 /* Time for the bond calculations */
1369 vtot += vbond; /* 1*/
1370 fbond *= gmx_invsqrt(dr2); /* 6 */
1374 ivec_sub(SHIFT_IVEC(g, ai), SHIFT_IVEC(g, ak), dt_ik);
1375 ki = IVEC2IS(dt_ik);
1377 for (m = 0; (m < DIM); m++) /* 15 */
1379 fik = fbond*r_ik[m];
1382 fshift[ki][m] += fik;
1383 fshift[CENTRAL][m] -= fik;
1389 real quartic_angles(int nbonds,
1390 const t_iatom forceatoms[], const t_iparams forceparams[],
1391 const rvec x[], rvec f[], rvec fshift[],
1392 const t_pbc *pbc, const t_graph *g,
1393 real gmx_unused lambda, real gmx_unused *dvdlambda,
1394 const t_mdatoms gmx_unused *md, t_fcdata gmx_unused *fcd,
1395 int gmx_unused *global_atom_index)
1397 int i, j, ai, aj, ak, t1, t2, type;
1399 real cos_theta, cos_theta2, theta, dt, dVdt, va, dtp, c, vtot;
1400 ivec jt, dt_ij, dt_kj;
1403 for (i = 0; (i < nbonds); )
1405 type = forceatoms[i++];
1406 ai = forceatoms[i++];
1407 aj = forceatoms[i++];
1408 ak = forceatoms[i++];
1410 theta = bond_angle(x[ai], x[aj], x[ak], pbc,
1411 r_ij, r_kj, &cos_theta, &t1, &t2); /* 41 */
1413 dt = theta - forceparams[type].qangle.theta*DEG2RAD; /* 2 */
1416 va = forceparams[type].qangle.c[0];
1418 for (j = 1; j <= 4; j++)
1420 c = forceparams[type].qangle.c[j];
1429 cos_theta2 = sqr(cos_theta); /* 1 */
1438 st = dVdt*gmx_invsqrt(1 - cos_theta2); /* 12 */
1439 sth = st*cos_theta; /* 1 */
1443 fprintf(debug, "ANGLES: theta = %10g vth = %10g dV/dtheta = %10g\n",
1444 theta*RAD2DEG, va, dVdt);
1447 nrkj2 = iprod(r_kj, r_kj); /* 5 */
1448 nrij2 = iprod(r_ij, r_ij);
1450 cik = st*gmx_invsqrt(nrkj2*nrij2); /* 12 */
1451 cii = sth/nrij2; /* 10 */
1452 ckk = sth/nrkj2; /* 10 */
1454 for (m = 0; (m < DIM); m++) /* 39 */
1456 f_i[m] = -(cik*r_kj[m]-cii*r_ij[m]);
1457 f_k[m] = -(cik*r_ij[m]-ckk*r_kj[m]);
1458 f_j[m] = -f_i[m]-f_k[m];
1465 copy_ivec(SHIFT_IVEC(g, aj), jt);
1467 ivec_sub(SHIFT_IVEC(g, ai), jt, dt_ij);
1468 ivec_sub(SHIFT_IVEC(g, ak), jt, dt_kj);
1469 t1 = IVEC2IS(dt_ij);
1470 t2 = IVEC2IS(dt_kj);
1472 rvec_inc(fshift[t1], f_i);
1473 rvec_inc(fshift[CENTRAL], f_j);
1474 rvec_inc(fshift[t2], f_k);
1480 real dih_angle(const rvec xi, const rvec xj, const rvec xk, const rvec xl,
1482 rvec r_ij, rvec r_kj, rvec r_kl, rvec m, rvec n,
1483 real *sign, int *t1, int *t2, int *t3)
1487 *t1 = pbc_rvec_sub(pbc, xi, xj, r_ij); /* 3 */
1488 *t2 = pbc_rvec_sub(pbc, xk, xj, r_kj); /* 3 */
1489 *t3 = pbc_rvec_sub(pbc, xk, xl, r_kl); /* 3 */
1491 cprod(r_ij, r_kj, m); /* 9 */
1492 cprod(r_kj, r_kl, n); /* 9 */
1493 phi = gmx_angle(m, n); /* 49 (assuming 25 for atan2) */
1494 ipr = iprod(r_ij, n); /* 5 */
1495 (*sign) = (ipr < 0.0) ? -1.0 : 1.0;
1496 phi = (*sign)*phi; /* 1 */
1502 #ifdef GMX_SIMD_HAVE_REAL
1504 /* As dih_angle above, but calculates 4 dihedral angles at once using SIMD,
1505 * also calculates the pre-factor required for the dihedral force update.
1506 * Note that bv and buf should be register aligned.
1508 static gmx_inline void
1509 dih_angle_simd(const rvec *x,
1510 const int *ai, const int *aj, const int *ak, const int *al,
1511 const pbc_simd_t *pbc,
1513 gmx_simd_real_t *phi_S,
1514 gmx_simd_real_t *mx_S, gmx_simd_real_t *my_S, gmx_simd_real_t *mz_S,
1515 gmx_simd_real_t *nx_S, gmx_simd_real_t *ny_S, gmx_simd_real_t *nz_S,
1516 gmx_simd_real_t *nrkj_m2_S,
1517 gmx_simd_real_t *nrkj_n2_S,
1522 gmx_simd_real_t rijx_S, rijy_S, rijz_S;
1523 gmx_simd_real_t rkjx_S, rkjy_S, rkjz_S;
1524 gmx_simd_real_t rklx_S, rkly_S, rklz_S;
1525 gmx_simd_real_t cx_S, cy_S, cz_S;
1526 gmx_simd_real_t cn_S;
1527 gmx_simd_real_t s_S;
1528 gmx_simd_real_t ipr_S;
1529 gmx_simd_real_t iprm_S, iprn_S;
1530 gmx_simd_real_t nrkj2_S, nrkj_1_S, nrkj_2_S, nrkj_S;
1531 gmx_simd_real_t toler_S;
1532 gmx_simd_real_t p_S, q_S;
1533 gmx_simd_real_t nrkj2_min_S;
1534 gmx_simd_real_t real_eps_S;
1536 /* Used to avoid division by zero.
1537 * We take into acount that we multiply the result by real_eps_S.
1539 nrkj2_min_S = gmx_simd_set1_r(GMX_REAL_MIN/(2*GMX_REAL_EPS));
1541 /* The value of the last significant bit (GMX_REAL_EPS is half of that) */
1542 real_eps_S = gmx_simd_set1_r(2*GMX_REAL_EPS);
1544 for (s = 0; s < GMX_SIMD_REAL_WIDTH; s++)
1546 /* If you can't use pbc_dx_simd below for PBC, e.g. because
1547 * you can't round in SIMD, use pbc_rvec_sub here.
1549 for (m = 0; m < DIM; m++)
1551 dr[s + (0*DIM + m)*GMX_SIMD_REAL_WIDTH] = x[ai[s]][m] - x[aj[s]][m];
1552 dr[s + (1*DIM + m)*GMX_SIMD_REAL_WIDTH] = x[ak[s]][m] - x[aj[s]][m];
1553 dr[s + (2*DIM + m)*GMX_SIMD_REAL_WIDTH] = x[ak[s]][m] - x[al[s]][m];
1557 rijx_S = gmx_simd_load_r(dr + 0*GMX_SIMD_REAL_WIDTH);
1558 rijy_S = gmx_simd_load_r(dr + 1*GMX_SIMD_REAL_WIDTH);
1559 rijz_S = gmx_simd_load_r(dr + 2*GMX_SIMD_REAL_WIDTH);
1560 rkjx_S = gmx_simd_load_r(dr + 3*GMX_SIMD_REAL_WIDTH);
1561 rkjy_S = gmx_simd_load_r(dr + 4*GMX_SIMD_REAL_WIDTH);
1562 rkjz_S = gmx_simd_load_r(dr + 5*GMX_SIMD_REAL_WIDTH);
1563 rklx_S = gmx_simd_load_r(dr + 6*GMX_SIMD_REAL_WIDTH);
1564 rkly_S = gmx_simd_load_r(dr + 7*GMX_SIMD_REAL_WIDTH);
1565 rklz_S = gmx_simd_load_r(dr + 8*GMX_SIMD_REAL_WIDTH);
1567 pbc_dx_simd(&rijx_S, &rijy_S, &rijz_S, pbc);
1568 pbc_dx_simd(&rkjx_S, &rkjy_S, &rkjz_S, pbc);
1569 pbc_dx_simd(&rklx_S, &rkly_S, &rklz_S, pbc);
1571 gmx_simd_cprod_r(rijx_S, rijy_S, rijz_S,
1572 rkjx_S, rkjy_S, rkjz_S,
1575 gmx_simd_cprod_r(rkjx_S, rkjy_S, rkjz_S,
1576 rklx_S, rkly_S, rklz_S,
1579 gmx_simd_cprod_r(*mx_S, *my_S, *mz_S,
1580 *nx_S, *ny_S, *nz_S,
1581 &cx_S, &cy_S, &cz_S);
1583 cn_S = gmx_simd_sqrt_r(gmx_simd_norm2_r(cx_S, cy_S, cz_S));
1585 s_S = gmx_simd_iprod_r(*mx_S, *my_S, *mz_S, *nx_S, *ny_S, *nz_S);
1587 /* Determine the dihedral angle, the sign might need correction */
1588 *phi_S = gmx_simd_atan2_r(cn_S, s_S);
1590 ipr_S = gmx_simd_iprod_r(rijx_S, rijy_S, rijz_S,
1591 *nx_S, *ny_S, *nz_S);
1593 iprm_S = gmx_simd_norm2_r(*mx_S, *my_S, *mz_S);
1594 iprn_S = gmx_simd_norm2_r(*nx_S, *ny_S, *nz_S);
1596 nrkj2_S = gmx_simd_norm2_r(rkjx_S, rkjy_S, rkjz_S);
1598 /* Avoid division by zero. When zero, the result is multiplied by 0
1599 * anyhow, so the 3 max below do not affect the final result.
1601 nrkj2_S = gmx_simd_max_r(nrkj2_S, nrkj2_min_S);
1602 nrkj_1_S = gmx_simd_invsqrt_r(nrkj2_S);
1603 nrkj_2_S = gmx_simd_mul_r(nrkj_1_S, nrkj_1_S);
1604 nrkj_S = gmx_simd_mul_r(nrkj2_S, nrkj_1_S);
1606 toler_S = gmx_simd_mul_r(nrkj2_S, real_eps_S);
1608 /* Here the plain-C code uses a conditional, but we can't do that in SIMD.
1609 * So we take a max with the tolerance instead. Since we multiply with
1610 * m or n later, the max does not affect the results.
1612 iprm_S = gmx_simd_max_r(iprm_S, toler_S);
1613 iprn_S = gmx_simd_max_r(iprn_S, toler_S);
1614 *nrkj_m2_S = gmx_simd_mul_r(nrkj_S, gmx_simd_inv_r(iprm_S));
1615 *nrkj_n2_S = gmx_simd_mul_r(nrkj_S, gmx_simd_inv_r(iprn_S));
1617 /* Set sign of phi_S with the sign of ipr_S; phi_S is currently positive */
1618 *phi_S = gmx_simd_xor_sign_r(*phi_S, ipr_S);
1619 p_S = gmx_simd_iprod_r(rijx_S, rijy_S, rijz_S,
1620 rkjx_S, rkjy_S, rkjz_S);
1621 p_S = gmx_simd_mul_r(p_S, nrkj_2_S);
1623 q_S = gmx_simd_iprod_r(rklx_S, rkly_S, rklz_S,
1624 rkjx_S, rkjy_S, rkjz_S);
1625 q_S = gmx_simd_mul_r(q_S, nrkj_2_S);
1627 gmx_simd_store_r(p, p_S);
1628 gmx_simd_store_r(q, q_S);
1631 #endif /* GMX_SIMD_HAVE_REAL */
1634 void do_dih_fup(int i, int j, int k, int l, real ddphi,
1635 rvec r_ij, rvec r_kj, rvec r_kl,
1636 rvec m, rvec n, rvec f[], rvec fshift[],
1637 const t_pbc *pbc, const t_graph *g,
1638 const rvec x[], int t1, int t2, int t3)
1641 rvec f_i, f_j, f_k, f_l;
1642 rvec uvec, vvec, svec, dx_jl;
1643 real iprm, iprn, nrkj, nrkj2, nrkj_1, nrkj_2;
1644 real a, b, p, q, toler;
1645 ivec jt, dt_ij, dt_kj, dt_lj;
1647 iprm = iprod(m, m); /* 5 */
1648 iprn = iprod(n, n); /* 5 */
1649 nrkj2 = iprod(r_kj, r_kj); /* 5 */
1650 toler = nrkj2*GMX_REAL_EPS;
1651 if ((iprm > toler) && (iprn > toler))
1653 nrkj_1 = gmx_invsqrt(nrkj2); /* 10 */
1654 nrkj_2 = nrkj_1*nrkj_1; /* 1 */
1655 nrkj = nrkj2*nrkj_1; /* 1 */
1656 a = -ddphi*nrkj/iprm; /* 11 */
1657 svmul(a, m, f_i); /* 3 */
1658 b = ddphi*nrkj/iprn; /* 11 */
1659 svmul(b, n, f_l); /* 3 */
1660 p = iprod(r_ij, r_kj); /* 5 */
1661 p *= nrkj_2; /* 1 */
1662 q = iprod(r_kl, r_kj); /* 5 */
1663 q *= nrkj_2; /* 1 */
1664 svmul(p, f_i, uvec); /* 3 */
1665 svmul(q, f_l, vvec); /* 3 */
1666 rvec_sub(uvec, vvec, svec); /* 3 */
1667 rvec_sub(f_i, svec, f_j); /* 3 */
1668 rvec_add(f_l, svec, f_k); /* 3 */
1669 rvec_inc(f[i], f_i); /* 3 */
1670 rvec_dec(f[j], f_j); /* 3 */
1671 rvec_dec(f[k], f_k); /* 3 */
1672 rvec_inc(f[l], f_l); /* 3 */
1676 copy_ivec(SHIFT_IVEC(g, j), jt);
1677 ivec_sub(SHIFT_IVEC(g, i), jt, dt_ij);
1678 ivec_sub(SHIFT_IVEC(g, k), jt, dt_kj);
1679 ivec_sub(SHIFT_IVEC(g, l), jt, dt_lj);
1680 t1 = IVEC2IS(dt_ij);
1681 t2 = IVEC2IS(dt_kj);
1682 t3 = IVEC2IS(dt_lj);
1686 t3 = pbc_rvec_sub(pbc, x[l], x[j], dx_jl);
1693 rvec_inc(fshift[t1], f_i);
1694 rvec_dec(fshift[CENTRAL], f_j);
1695 rvec_dec(fshift[t2], f_k);
1696 rvec_inc(fshift[t3], f_l);
1701 /* As do_dih_fup above, but without shift forces */
1703 do_dih_fup_noshiftf(int i, int j, int k, int l, real ddphi,
1704 rvec r_ij, rvec r_kj, rvec r_kl,
1705 rvec m, rvec n, rvec f[])
1707 rvec f_i, f_j, f_k, f_l;
1708 rvec uvec, vvec, svec, dx_jl;
1709 real iprm, iprn, nrkj, nrkj2, nrkj_1, nrkj_2;
1710 real a, b, p, q, toler;
1711 ivec jt, dt_ij, dt_kj, dt_lj;
1713 iprm = iprod(m, m); /* 5 */
1714 iprn = iprod(n, n); /* 5 */
1715 nrkj2 = iprod(r_kj, r_kj); /* 5 */
1716 toler = nrkj2*GMX_REAL_EPS;
1717 if ((iprm > toler) && (iprn > toler))
1719 nrkj_1 = gmx_invsqrt(nrkj2); /* 10 */
1720 nrkj_2 = nrkj_1*nrkj_1; /* 1 */
1721 nrkj = nrkj2*nrkj_1; /* 1 */
1722 a = -ddphi*nrkj/iprm; /* 11 */
1723 svmul(a, m, f_i); /* 3 */
1724 b = ddphi*nrkj/iprn; /* 11 */
1725 svmul(b, n, f_l); /* 3 */
1726 p = iprod(r_ij, r_kj); /* 5 */
1727 p *= nrkj_2; /* 1 */
1728 q = iprod(r_kl, r_kj); /* 5 */
1729 q *= nrkj_2; /* 1 */
1730 svmul(p, f_i, uvec); /* 3 */
1731 svmul(q, f_l, vvec); /* 3 */
1732 rvec_sub(uvec, vvec, svec); /* 3 */
1733 rvec_sub(f_i, svec, f_j); /* 3 */
1734 rvec_add(f_l, svec, f_k); /* 3 */
1735 rvec_inc(f[i], f_i); /* 3 */
1736 rvec_dec(f[j], f_j); /* 3 */
1737 rvec_dec(f[k], f_k); /* 3 */
1738 rvec_inc(f[l], f_l); /* 3 */
1742 /* As do_dih_fup_noshiftf above, but with pre-calculated pre-factors */
1743 static gmx_inline void
1744 do_dih_fup_noshiftf_precalc(int i, int j, int k, int l,
1746 real f_i_x, real f_i_y, real f_i_z,
1747 real mf_l_x, real mf_l_y, real mf_l_z,
1750 rvec f_i, f_j, f_k, f_l;
1751 rvec uvec, vvec, svec;
1759 svmul(p, f_i, uvec);
1760 svmul(q, f_l, vvec);
1761 rvec_sub(uvec, vvec, svec);
1762 rvec_sub(f_i, svec, f_j);
1763 rvec_add(f_l, svec, f_k);
1764 rvec_inc(f[i], f_i);
1765 rvec_dec(f[j], f_j);
1766 rvec_dec(f[k], f_k);
1767 rvec_inc(f[l], f_l);
1771 real dopdihs(real cpA, real cpB, real phiA, real phiB, int mult,
1772 real phi, real lambda, real *V, real *F)
1774 real v, dvdlambda, mdphi, v1, sdphi, ddphi;
1775 real L1 = 1.0 - lambda;
1776 real ph0 = (L1*phiA + lambda*phiB)*DEG2RAD;
1777 real dph0 = (phiB - phiA)*DEG2RAD;
1778 real cp = L1*cpA + lambda*cpB;
1780 mdphi = mult*phi - ph0;
1782 ddphi = -cp*mult*sdphi;
1783 v1 = 1.0 + cos(mdphi);
1786 dvdlambda = (cpB - cpA)*v1 + cp*dph0*sdphi;
1793 /* That was 40 flops */
1797 dopdihs_noener(real cpA, real cpB, real phiA, real phiB, int mult,
1798 real phi, real lambda, real *F)
1800 real mdphi, sdphi, ddphi;
1801 real L1 = 1.0 - lambda;
1802 real ph0 = (L1*phiA + lambda*phiB)*DEG2RAD;
1803 real cp = L1*cpA + lambda*cpB;
1805 mdphi = mult*phi - ph0;
1807 ddphi = -cp*mult*sdphi;
1811 /* That was 20 flops */
1815 dopdihs_mdphi(real cpA, real cpB, real phiA, real phiB, int mult,
1816 real phi, real lambda, real *cp, real *mdphi)
1818 real L1 = 1.0 - lambda;
1819 real ph0 = (L1*phiA + lambda*phiB)*DEG2RAD;
1821 *cp = L1*cpA + lambda*cpB;
1823 *mdphi = mult*phi - ph0;
1826 static real dopdihs_min(real cpA, real cpB, real phiA, real phiB, int mult,
1827 real phi, real lambda, real *V, real *F)
1828 /* similar to dopdihs, except for a minus sign *
1829 * and a different treatment of mult/phi0 */
1831 real v, dvdlambda, mdphi, v1, sdphi, ddphi;
1832 real L1 = 1.0 - lambda;
1833 real ph0 = (L1*phiA + lambda*phiB)*DEG2RAD;
1834 real dph0 = (phiB - phiA)*DEG2RAD;
1835 real cp = L1*cpA + lambda*cpB;
1837 mdphi = mult*(phi-ph0);
1839 ddphi = cp*mult*sdphi;
1840 v1 = 1.0-cos(mdphi);
1843 dvdlambda = (cpB-cpA)*v1 + cp*dph0*sdphi;
1850 /* That was 40 flops */
1853 real pdihs(int nbonds,
1854 const t_iatom forceatoms[], const t_iparams forceparams[],
1855 const rvec x[], rvec f[], rvec fshift[],
1856 const t_pbc *pbc, const t_graph *g,
1857 real lambda, real *dvdlambda,
1858 const t_mdatoms gmx_unused *md, t_fcdata gmx_unused *fcd,
1859 int gmx_unused *global_atom_index)
1861 int i, type, ai, aj, ak, al;
1863 rvec r_ij, r_kj, r_kl, m, n;
1864 real phi, sign, ddphi, vpd, vtot;
1868 for (i = 0; (i < nbonds); )
1870 type = forceatoms[i++];
1871 ai = forceatoms[i++];
1872 aj = forceatoms[i++];
1873 ak = forceatoms[i++];
1874 al = forceatoms[i++];
1876 phi = dih_angle(x[ai], x[aj], x[ak], x[al], pbc, r_ij, r_kj, r_kl, m, n,
1877 &sign, &t1, &t2, &t3); /* 84 */
1878 *dvdlambda += dopdihs(forceparams[type].pdihs.cpA,
1879 forceparams[type].pdihs.cpB,
1880 forceparams[type].pdihs.phiA,
1881 forceparams[type].pdihs.phiB,
1882 forceparams[type].pdihs.mult,
1883 phi, lambda, &vpd, &ddphi);
1886 do_dih_fup(ai, aj, ak, al, ddphi, r_ij, r_kj, r_kl, m, n,
1887 f, fshift, pbc, g, x, t1, t2, t3); /* 112 */
1890 fprintf(debug, "pdih: (%d,%d,%d,%d) phi=%g\n",
1891 ai, aj, ak, al, phi);
1898 void make_dp_periodic(real *dp) /* 1 flop? */
1900 /* dp cannot be outside (-pi,pi) */
1905 else if (*dp < -M_PI)
1912 /* As pdihs above, but without calculating energies and shift forces */
1914 pdihs_noener(int nbonds,
1915 const t_iatom forceatoms[], const t_iparams forceparams[],
1916 const rvec x[], rvec f[],
1917 const t_pbc gmx_unused *pbc, const t_graph gmx_unused *g,
1919 const t_mdatoms gmx_unused *md, t_fcdata gmx_unused *fcd,
1920 int gmx_unused *global_atom_index)
1922 int i, type, ai, aj, ak, al;
1924 rvec r_ij, r_kj, r_kl, m, n;
1925 real phi, sign, ddphi_tot, ddphi;
1927 for (i = 0; (i < nbonds); )
1929 ai = forceatoms[i+1];
1930 aj = forceatoms[i+2];
1931 ak = forceatoms[i+3];
1932 al = forceatoms[i+4];
1934 phi = dih_angle(x[ai], x[aj], x[ak], x[al], pbc, r_ij, r_kj, r_kl, m, n,
1935 &sign, &t1, &t2, &t3);
1939 /* Loop over dihedrals working on the same atoms,
1940 * so we avoid recalculating angles and force distributions.
1944 type = forceatoms[i];
1945 dopdihs_noener(forceparams[type].pdihs.cpA,
1946 forceparams[type].pdihs.cpB,
1947 forceparams[type].pdihs.phiA,
1948 forceparams[type].pdihs.phiB,
1949 forceparams[type].pdihs.mult,
1950 phi, lambda, &ddphi);
1955 while (i < nbonds &&
1956 forceatoms[i+1] == ai &&
1957 forceatoms[i+2] == aj &&
1958 forceatoms[i+3] == ak &&
1959 forceatoms[i+4] == al);
1961 do_dih_fup_noshiftf(ai, aj, ak, al, ddphi_tot, r_ij, r_kj, r_kl, m, n, f);
1966 #ifdef GMX_SIMD_HAVE_REAL
1968 /* As pdihs_noner above, but using SIMD to calculate many dihedrals at once */
1970 pdihs_noener_simd(int nbonds,
1971 const t_iatom forceatoms[], const t_iparams forceparams[],
1972 const rvec x[], rvec f[],
1973 const t_pbc *pbc, const t_graph gmx_unused *g,
1974 real gmx_unused lambda,
1975 const t_mdatoms gmx_unused *md, t_fcdata gmx_unused *fcd,
1976 int gmx_unused *global_atom_index)
1980 int type, ai[GMX_SIMD_REAL_WIDTH], aj[GMX_SIMD_REAL_WIDTH], ak[GMX_SIMD_REAL_WIDTH], al[GMX_SIMD_REAL_WIDTH];
1981 int t1[GMX_SIMD_REAL_WIDTH], t2[GMX_SIMD_REAL_WIDTH], t3[GMX_SIMD_REAL_WIDTH];
1983 real dr_array[3*DIM*GMX_SIMD_REAL_WIDTH+GMX_SIMD_REAL_WIDTH], *dr;
1984 real buf_array[7*GMX_SIMD_REAL_WIDTH+GMX_SIMD_REAL_WIDTH], *buf;
1985 real *cp, *phi0, *mult, *phi, *p, *q, *sf_i, *msf_l;
1986 gmx_simd_real_t phi0_S, phi_S;
1987 gmx_simd_real_t mx_S, my_S, mz_S;
1988 gmx_simd_real_t nx_S, ny_S, nz_S;
1989 gmx_simd_real_t nrkj_m2_S, nrkj_n2_S;
1990 gmx_simd_real_t cp_S, mdphi_S, mult_S;
1991 gmx_simd_real_t sin_S, cos_S;
1992 gmx_simd_real_t mddphi_S;
1993 gmx_simd_real_t sf_i_S, msf_l_S;
1994 pbc_simd_t pbc_simd;
1996 /* Ensure SIMD register alignment */
1997 dr = gmx_simd_align_r(dr_array);
1998 buf = gmx_simd_align_r(buf_array);
2000 /* Extract aligned pointer for parameters and variables */
2001 cp = buf + 0*GMX_SIMD_REAL_WIDTH;
2002 phi0 = buf + 1*GMX_SIMD_REAL_WIDTH;
2003 mult = buf + 2*GMX_SIMD_REAL_WIDTH;
2004 p = buf + 3*GMX_SIMD_REAL_WIDTH;
2005 q = buf + 4*GMX_SIMD_REAL_WIDTH;
2006 sf_i = buf + 5*GMX_SIMD_REAL_WIDTH;
2007 msf_l = buf + 6*GMX_SIMD_REAL_WIDTH;
2009 set_pbc_simd(pbc, &pbc_simd);
2011 /* nbonds is the number of dihedrals times nfa1, here we step GMX_SIMD_REAL_WIDTH dihs */
2012 for (i = 0; (i < nbonds); i += GMX_SIMD_REAL_WIDTH*nfa1)
2014 /* Collect atoms quadruplets for GMX_SIMD_REAL_WIDTH dihedrals.
2015 * iu indexes into forceatoms, we should not let iu go beyond nbonds.
2018 for (s = 0; s < GMX_SIMD_REAL_WIDTH; s++)
2020 type = forceatoms[iu];
2021 ai[s] = forceatoms[iu+1];
2022 aj[s] = forceatoms[iu+2];
2023 ak[s] = forceatoms[iu+3];
2024 al[s] = forceatoms[iu+4];
2026 cp[s] = forceparams[type].pdihs.cpA;
2027 phi0[s] = forceparams[type].pdihs.phiA*DEG2RAD;
2028 mult[s] = forceparams[type].pdihs.mult;
2030 /* At the end fill the arrays with identical entries */
2031 if (iu + nfa1 < nbonds)
2037 /* Caclulate GMX_SIMD_REAL_WIDTH dihedral angles at once */
2038 dih_angle_simd(x, ai, aj, ak, al, &pbc_simd,
2041 &mx_S, &my_S, &mz_S,
2042 &nx_S, &ny_S, &nz_S,
2047 cp_S = gmx_simd_load_r(cp);
2048 phi0_S = gmx_simd_load_r(phi0);
2049 mult_S = gmx_simd_load_r(mult);
2051 mdphi_S = gmx_simd_sub_r(gmx_simd_mul_r(mult_S, phi_S), phi0_S);
2053 /* Calculate GMX_SIMD_REAL_WIDTH sines at once */
2054 gmx_simd_sincos_r(mdphi_S, &sin_S, &cos_S);
2055 mddphi_S = gmx_simd_mul_r(gmx_simd_mul_r(cp_S, mult_S), sin_S);
2056 sf_i_S = gmx_simd_mul_r(mddphi_S, nrkj_m2_S);
2057 msf_l_S = gmx_simd_mul_r(mddphi_S, nrkj_n2_S);
2059 /* After this m?_S will contain f[i] */
2060 mx_S = gmx_simd_mul_r(sf_i_S, mx_S);
2061 my_S = gmx_simd_mul_r(sf_i_S, my_S);
2062 mz_S = gmx_simd_mul_r(sf_i_S, mz_S);
2064 /* After this m?_S will contain -f[l] */
2065 nx_S = gmx_simd_mul_r(msf_l_S, nx_S);
2066 ny_S = gmx_simd_mul_r(msf_l_S, ny_S);
2067 nz_S = gmx_simd_mul_r(msf_l_S, nz_S);
2069 gmx_simd_store_r(dr + 0*GMX_SIMD_REAL_WIDTH, mx_S);
2070 gmx_simd_store_r(dr + 1*GMX_SIMD_REAL_WIDTH, my_S);
2071 gmx_simd_store_r(dr + 2*GMX_SIMD_REAL_WIDTH, mz_S);
2072 gmx_simd_store_r(dr + 3*GMX_SIMD_REAL_WIDTH, nx_S);
2073 gmx_simd_store_r(dr + 4*GMX_SIMD_REAL_WIDTH, ny_S);
2074 gmx_simd_store_r(dr + 5*GMX_SIMD_REAL_WIDTH, nz_S);
2080 do_dih_fup_noshiftf_precalc(ai[s], aj[s], ak[s], al[s],
2082 dr[ XX *GMX_SIMD_REAL_WIDTH+s],
2083 dr[ YY *GMX_SIMD_REAL_WIDTH+s],
2084 dr[ ZZ *GMX_SIMD_REAL_WIDTH+s],
2085 dr[(DIM+XX)*GMX_SIMD_REAL_WIDTH+s],
2086 dr[(DIM+YY)*GMX_SIMD_REAL_WIDTH+s],
2087 dr[(DIM+ZZ)*GMX_SIMD_REAL_WIDTH+s],
2092 while (s < GMX_SIMD_REAL_WIDTH && iu < nbonds);
2096 #endif /* GMX_SIMD_HAVE_REAL */
2099 real idihs(int nbonds,
2100 const t_iatom forceatoms[], const t_iparams forceparams[],
2101 const rvec x[], rvec f[], rvec fshift[],
2102 const t_pbc *pbc, const t_graph *g,
2103 real lambda, real *dvdlambda,
2104 const t_mdatoms gmx_unused *md, t_fcdata gmx_unused *fcd,
2105 int gmx_unused *global_atom_index)
2107 int i, type, ai, aj, ak, al;
2109 real phi, phi0, dphi0, ddphi, sign, vtot;
2110 rvec r_ij, r_kj, r_kl, m, n;
2111 real L1, kk, dp, dp2, kA, kB, pA, pB, dvdl_term;
2116 for (i = 0; (i < nbonds); )
2118 type = forceatoms[i++];
2119 ai = forceatoms[i++];
2120 aj = forceatoms[i++];
2121 ak = forceatoms[i++];
2122 al = forceatoms[i++];
2124 phi = dih_angle(x[ai], x[aj], x[ak], x[al], pbc, r_ij, r_kj, r_kl, m, n,
2125 &sign, &t1, &t2, &t3); /* 84 */
2127 /* phi can jump if phi0 is close to Pi/-Pi, which will cause huge
2128 * force changes if we just apply a normal harmonic.
2129 * Instead, we first calculate phi-phi0 and take it modulo (-Pi,Pi).
2130 * This means we will never have the periodicity problem, unless
2131 * the dihedral is Pi away from phiO, which is very unlikely due to
2134 kA = forceparams[type].harmonic.krA;
2135 kB = forceparams[type].harmonic.krB;
2136 pA = forceparams[type].harmonic.rA;
2137 pB = forceparams[type].harmonic.rB;
2139 kk = L1*kA + lambda*kB;
2140 phi0 = (L1*pA + lambda*pB)*DEG2RAD;
2141 dphi0 = (pB - pA)*DEG2RAD;
2145 make_dp_periodic(&dp);
2152 dvdl_term += 0.5*(kB - kA)*dp2 - kk*dphi0*dp;
2154 do_dih_fup(ai, aj, ak, al, (real)(-ddphi), r_ij, r_kj, r_kl, m, n,
2155 f, fshift, pbc, g, x, t1, t2, t3); /* 112 */
2160 fprintf(debug, "idih: (%d,%d,%d,%d) phi=%g\n",
2161 ai, aj, ak, al, phi);
2166 *dvdlambda += dvdl_term;
2171 /*! \brief returns dx, rdist, and dpdl for functions posres() and fbposres()
2173 static void posres_dx(const rvec x, const rvec pos0A, const rvec pos0B,
2174 const rvec comA_sc, const rvec comB_sc,
2176 t_pbc *pbc, int refcoord_scaling, int npbcdim,
2177 rvec dx, rvec rdist, rvec dpdl)
2180 real posA, posB, L1, ref = 0.;
2185 for (m = 0; m < DIM; m++)
2191 switch (refcoord_scaling)
2195 rdist[m] = L1*posA + lambda*posB;
2196 dpdl[m] = posB - posA;
2199 /* Box relative coordinates are stored for dimensions with pbc */
2200 posA *= pbc->box[m][m];
2201 posB *= pbc->box[m][m];
2202 for (d = m+1; d < npbcdim; d++)
2204 posA += pos0A[d]*pbc->box[d][m];
2205 posB += pos0B[d]*pbc->box[d][m];
2207 ref = L1*posA + lambda*posB;
2209 dpdl[m] = posB - posA;
2212 ref = L1*comA_sc[m] + lambda*comB_sc[m];
2213 rdist[m] = L1*posA + lambda*posB;
2214 dpdl[m] = comB_sc[m] - comA_sc[m] + posB - posA;
2217 gmx_fatal(FARGS, "No such scaling method implemented");
2222 ref = L1*posA + lambda*posB;
2224 dpdl[m] = posB - posA;
2227 /* We do pbc_dx with ref+rdist,
2228 * since with only ref we can be up to half a box vector wrong.
2230 pos[m] = ref + rdist[m];
2235 pbc_dx(pbc, x, pos, dx);
2239 rvec_sub(x, pos, dx);
2243 /*! \brief Adds forces of flat-bottomed positions restraints to f[]
2244 * and fixes vir_diag. Returns the flat-bottomed potential. */
2245 real fbposres(int nbonds,
2246 const t_iatom forceatoms[], const t_iparams forceparams[],
2247 const rvec x[], rvec f[], rvec vir_diag,
2249 int refcoord_scaling, int ePBC, rvec com)
2250 /* compute flat-bottomed positions restraints */
2252 int i, ai, m, d, type, npbcdim = 0, fbdim;
2253 const t_iparams *pr;
2255 real ref = 0, dr, dr2, rpot, rfb, rfb2, fact, invdr;
2256 rvec com_sc, rdist, pos, dx, dpdl, fm;
2259 npbcdim = ePBC2npbcdim(ePBC);
2261 if (refcoord_scaling == erscCOM)
2264 for (m = 0; m < npbcdim; m++)
2266 for (d = m; d < npbcdim; d++)
2268 com_sc[m] += com[d]*pbc->box[d][m];
2274 for (i = 0; (i < nbonds); )
2276 type = forceatoms[i++];
2277 ai = forceatoms[i++];
2278 pr = &forceparams[type];
2280 /* same calculation as for normal posres, but with identical A and B states, and lambda==0 */
2281 posres_dx(x[ai], forceparams[type].fbposres.pos0, forceparams[type].fbposres.pos0,
2282 com_sc, com_sc, 0.0,
2283 pbc, refcoord_scaling, npbcdim,
2289 kk = pr->fbposres.k;
2290 rfb = pr->fbposres.r;
2293 /* with rfb<0, push particle out of the sphere/cylinder/layer */
2301 switch (pr->fbposres.geom)
2303 case efbposresSPHERE:
2304 /* spherical flat-bottom posres */
2307 ( (dr2 > rfb2 && bInvert == FALSE ) || (dr2 < rfb2 && bInvert == TRUE ) )
2311 v = 0.5*kk*sqr(dr - rfb);
2312 fact = -kk*(dr-rfb)/dr; /* Force pointing to the center pos0 */
2313 svmul(fact, dx, fm);
2316 case efbposresCYLINDER:
2317 /* cylidrical flat-bottom posres in x-y plane. fm[ZZ] = 0. */
2318 dr2 = sqr(dx[XX])+sqr(dx[YY]);
2320 ( (dr2 > rfb2 && bInvert == FALSE ) || (dr2 < rfb2 && bInvert == TRUE ) )
2325 v = 0.5*kk*sqr(dr - rfb);
2326 fm[XX] = -kk*(dr-rfb)*dx[XX]*invdr; /* Force pointing to the center */
2327 fm[YY] = -kk*(dr-rfb)*dx[YY]*invdr;
2330 case efbposresX: /* fbdim=XX */
2331 case efbposresY: /* fbdim=YY */
2332 case efbposresZ: /* fbdim=ZZ */
2333 /* 1D flat-bottom potential */
2334 fbdim = pr->fbposres.geom - efbposresX;
2336 if ( ( dr > rfb && bInvert == FALSE ) || ( 0 < dr && dr < rfb && bInvert == TRUE ) )
2338 v = 0.5*kk*sqr(dr - rfb);
2339 fm[fbdim] = -kk*(dr - rfb);
2341 else if ( (dr < (-rfb) && bInvert == FALSE ) || ( (-rfb) < dr && dr < 0 && bInvert == TRUE ))
2343 v = 0.5*kk*sqr(dr + rfb);
2344 fm[fbdim] = -kk*(dr + rfb);
2351 for (m = 0; (m < DIM); m++)
2354 /* Here we correct for the pbc_dx which included rdist */
2355 vir_diag[m] -= 0.5*(dx[m] + rdist[m])*fm[m];
2363 real posres(int nbonds,
2364 const t_iatom forceatoms[], const t_iparams forceparams[],
2365 const rvec x[], rvec f[], rvec vir_diag,
2367 real lambda, real *dvdlambda,
2368 int refcoord_scaling, int ePBC, rvec comA, rvec comB)
2370 int i, ai, m, d, type, ki, npbcdim = 0;
2371 const t_iparams *pr;
2374 real posA, posB, ref = 0;
2375 rvec comA_sc, comB_sc, rdist, dpdl, pos, dx;
2376 gmx_bool bForceValid = TRUE;
2378 if ((f == NULL) || (vir_diag == NULL)) /* should both be null together! */
2380 bForceValid = FALSE;
2383 npbcdim = ePBC2npbcdim(ePBC);
2385 if (refcoord_scaling == erscCOM)
2387 clear_rvec(comA_sc);
2388 clear_rvec(comB_sc);
2389 for (m = 0; m < npbcdim; m++)
2391 for (d = m; d < npbcdim; d++)
2393 comA_sc[m] += comA[d]*pbc->box[d][m];
2394 comB_sc[m] += comB[d]*pbc->box[d][m];
2402 for (i = 0; (i < nbonds); )
2404 type = forceatoms[i++];
2405 ai = forceatoms[i++];
2406 pr = &forceparams[type];
2408 /* return dx, rdist, and dpdl */
2409 posres_dx(x[ai], forceparams[type].posres.pos0A, forceparams[type].posres.pos0B,
2410 comA_sc, comB_sc, lambda,
2411 pbc, refcoord_scaling, npbcdim,
2414 for (m = 0; (m < DIM); m++)
2416 kk = L1*pr->posres.fcA[m] + lambda*pr->posres.fcB[m];
2418 vtot += 0.5*kk*dx[m]*dx[m];
2420 0.5*(pr->posres.fcB[m] - pr->posres.fcA[m])*dx[m]*dx[m]
2423 /* Here we correct for the pbc_dx which included rdist */
2427 vir_diag[m] -= 0.5*(dx[m] + rdist[m])*fm;
2435 static real low_angres(int nbonds,
2436 const t_iatom forceatoms[], const t_iparams forceparams[],
2437 const rvec x[], rvec f[], rvec fshift[],
2438 const t_pbc *pbc, const t_graph *g,
2439 real lambda, real *dvdlambda,
2442 int i, m, type, ai, aj, ak, al;
2444 real phi, cos_phi, cos_phi2, vid, vtot, dVdphi;
2445 rvec r_ij, r_kl, f_i, f_k = {0, 0, 0};
2446 real st, sth, nrij2, nrkl2, c, cij, ckl;
2449 t2 = 0; /* avoid warning with gcc-3.3. It is never used uninitialized */
2452 ak = al = 0; /* to avoid warnings */
2453 for (i = 0; i < nbonds; )
2455 type = forceatoms[i++];
2456 ai = forceatoms[i++];
2457 aj = forceatoms[i++];
2458 t1 = pbc_rvec_sub(pbc, x[aj], x[ai], r_ij); /* 3 */
2461 ak = forceatoms[i++];
2462 al = forceatoms[i++];
2463 t2 = pbc_rvec_sub(pbc, x[al], x[ak], r_kl); /* 3 */
2472 cos_phi = cos_angle(r_ij, r_kl); /* 25 */
2473 phi = acos(cos_phi); /* 10 */
2475 *dvdlambda += dopdihs_min(forceparams[type].pdihs.cpA,
2476 forceparams[type].pdihs.cpB,
2477 forceparams[type].pdihs.phiA,
2478 forceparams[type].pdihs.phiB,
2479 forceparams[type].pdihs.mult,
2480 phi, lambda, &vid, &dVdphi); /* 40 */
2484 cos_phi2 = sqr(cos_phi); /* 1 */
2487 st = -dVdphi*gmx_invsqrt(1 - cos_phi2); /* 12 */
2488 sth = st*cos_phi; /* 1 */
2489 nrij2 = iprod(r_ij, r_ij); /* 5 */
2490 nrkl2 = iprod(r_kl, r_kl); /* 5 */
2492 c = st*gmx_invsqrt(nrij2*nrkl2); /* 11 */
2493 cij = sth/nrij2; /* 10 */
2494 ckl = sth/nrkl2; /* 10 */
2496 for (m = 0; m < DIM; m++) /* 18+18 */
2498 f_i[m] = (c*r_kl[m]-cij*r_ij[m]);
2503 f_k[m] = (c*r_ij[m]-ckl*r_kl[m]);
2511 ivec_sub(SHIFT_IVEC(g, ai), SHIFT_IVEC(g, aj), dt);
2514 rvec_inc(fshift[t1], f_i);
2515 rvec_dec(fshift[CENTRAL], f_i);
2520 ivec_sub(SHIFT_IVEC(g, ak), SHIFT_IVEC(g, al), dt);
2523 rvec_inc(fshift[t2], f_k);
2524 rvec_dec(fshift[CENTRAL], f_k);
2529 return vtot; /* 184 / 157 (bZAxis) total */
2532 real angres(int nbonds,
2533 const t_iatom forceatoms[], const t_iparams forceparams[],
2534 const rvec x[], rvec f[], rvec fshift[],
2535 const t_pbc *pbc, const t_graph *g,
2536 real lambda, real *dvdlambda,
2537 const t_mdatoms gmx_unused *md, t_fcdata gmx_unused *fcd,
2538 int gmx_unused *global_atom_index)
2540 return low_angres(nbonds, forceatoms, forceparams, x, f, fshift, pbc, g,
2541 lambda, dvdlambda, FALSE);
2544 real angresz(int nbonds,
2545 const t_iatom forceatoms[], const t_iparams forceparams[],
2546 const rvec x[], rvec f[], rvec fshift[],
2547 const t_pbc *pbc, const t_graph *g,
2548 real lambda, real *dvdlambda,
2549 const t_mdatoms gmx_unused *md, t_fcdata gmx_unused *fcd,
2550 int gmx_unused *global_atom_index)
2552 return low_angres(nbonds, forceatoms, forceparams, x, f, fshift, pbc, g,
2553 lambda, dvdlambda, TRUE);
2556 real dihres(int nbonds,
2557 const t_iatom forceatoms[], const t_iparams forceparams[],
2558 const rvec x[], rvec f[], rvec fshift[],
2559 const t_pbc *pbc, const t_graph *g,
2560 real lambda, real *dvdlambda,
2561 const t_mdatoms gmx_unused *md, t_fcdata gmx_unused *fcd,
2562 int gmx_unused *global_atom_index)
2565 int ai, aj, ak, al, i, k, type, t1, t2, t3;
2566 real phi0A, phi0B, dphiA, dphiB, kfacA, kfacB, phi0, dphi, kfac;
2567 real phi, ddphi, ddp, ddp2, dp, sign, d2r, fc, L1;
2568 rvec r_ij, r_kj, r_kl, m, n;
2575 for (i = 0; (i < nbonds); )
2577 type = forceatoms[i++];
2578 ai = forceatoms[i++];
2579 aj = forceatoms[i++];
2580 ak = forceatoms[i++];
2581 al = forceatoms[i++];
2583 phi0A = forceparams[type].dihres.phiA*d2r;
2584 dphiA = forceparams[type].dihres.dphiA*d2r;
2585 kfacA = forceparams[type].dihres.kfacA;
2587 phi0B = forceparams[type].dihres.phiB*d2r;
2588 dphiB = forceparams[type].dihres.dphiB*d2r;
2589 kfacB = forceparams[type].dihres.kfacB;
2591 phi0 = L1*phi0A + lambda*phi0B;
2592 dphi = L1*dphiA + lambda*dphiB;
2593 kfac = L1*kfacA + lambda*kfacB;
2595 phi = dih_angle(x[ai], x[aj], x[ak], x[al], pbc, r_ij, r_kj, r_kl, m, n,
2596 &sign, &t1, &t2, &t3);
2601 fprintf(debug, "dihres[%d]: %d %d %d %d : phi=%f, dphi=%f, kfac=%f\n",
2602 k++, ai, aj, ak, al, phi0, dphi, kfac);
2604 /* phi can jump if phi0 is close to Pi/-Pi, which will cause huge
2605 * force changes if we just apply a normal harmonic.
2606 * Instead, we first calculate phi-phi0 and take it modulo (-Pi,Pi).
2607 * This means we will never have the periodicity problem, unless
2608 * the dihedral is Pi away from phiO, which is very unlikely due to
2612 make_dp_periodic(&dp);
2618 else if (dp < -dphi)
2630 vtot += 0.5*kfac*ddp2;
2633 *dvdlambda += 0.5*(kfacB - kfacA)*ddp2;
2634 /* lambda dependence from changing restraint distances */
2637 *dvdlambda -= kfac*ddp*((dphiB - dphiA)+(phi0B - phi0A));
2641 *dvdlambda += kfac*ddp*((dphiB - dphiA)-(phi0B - phi0A));
2643 do_dih_fup(ai, aj, ak, al, ddphi, r_ij, r_kj, r_kl, m, n,
2644 f, fshift, pbc, g, x, t1, t2, t3); /* 112 */
2651 real unimplemented(int gmx_unused nbonds,
2652 const t_iatom gmx_unused forceatoms[], const t_iparams gmx_unused forceparams[],
2653 const rvec gmx_unused x[], rvec gmx_unused f[], rvec gmx_unused fshift[],
2654 const t_pbc gmx_unused *pbc, const t_graph gmx_unused *g,
2655 real gmx_unused lambda, real gmx_unused *dvdlambda,
2656 const t_mdatoms gmx_unused *md, t_fcdata gmx_unused *fcd,
2657 int gmx_unused *global_atom_index)
2659 gmx_impl("*** you are using a not implemented function");
2661 return 0.0; /* To make the compiler happy */
2664 real rbdihs(int nbonds,
2665 const t_iatom forceatoms[], const t_iparams forceparams[],
2666 const rvec x[], rvec f[], rvec fshift[],
2667 const t_pbc *pbc, const t_graph *g,
2668 real lambda, real *dvdlambda,
2669 const t_mdatoms gmx_unused *md, t_fcdata gmx_unused *fcd,
2670 int gmx_unused *global_atom_index)
2672 const real c0 = 0.0, c1 = 1.0, c2 = 2.0, c3 = 3.0, c4 = 4.0, c5 = 5.0;
2673 int type, ai, aj, ak, al, i, j;
2675 rvec r_ij, r_kj, r_kl, m, n;
2676 real parmA[NR_RBDIHS];
2677 real parmB[NR_RBDIHS];
2678 real parm[NR_RBDIHS];
2679 real cos_phi, phi, rbp, rbpBA;
2680 real v, sign, ddphi, sin_phi;
2682 real L1 = 1.0-lambda;
2686 for (i = 0; (i < nbonds); )
2688 type = forceatoms[i++];
2689 ai = forceatoms[i++];
2690 aj = forceatoms[i++];
2691 ak = forceatoms[i++];
2692 al = forceatoms[i++];
2694 phi = dih_angle(x[ai], x[aj], x[ak], x[al], pbc, r_ij, r_kj, r_kl, m, n,
2695 &sign, &t1, &t2, &t3); /* 84 */
2697 /* Change to polymer convention */
2704 phi -= M_PI; /* 1 */
2708 /* Beware of accuracy loss, cannot use 1-sqrt(cos^2) ! */
2711 for (j = 0; (j < NR_RBDIHS); j++)
2713 parmA[j] = forceparams[type].rbdihs.rbcA[j];
2714 parmB[j] = forceparams[type].rbdihs.rbcB[j];
2715 parm[j] = L1*parmA[j]+lambda*parmB[j];
2717 /* Calculate cosine powers */
2718 /* Calculate the energy */
2719 /* Calculate the derivative */
2722 dvdl_term += (parmB[0]-parmA[0]);
2727 rbpBA = parmB[1]-parmA[1];
2728 ddphi += rbp*cosfac;
2731 dvdl_term += cosfac*rbpBA;
2733 rbpBA = parmB[2]-parmA[2];
2734 ddphi += c2*rbp*cosfac;
2737 dvdl_term += cosfac*rbpBA;
2739 rbpBA = parmB[3]-parmA[3];
2740 ddphi += c3*rbp*cosfac;
2743 dvdl_term += cosfac*rbpBA;
2745 rbpBA = parmB[4]-parmA[4];
2746 ddphi += c4*rbp*cosfac;
2749 dvdl_term += cosfac*rbpBA;
2751 rbpBA = parmB[5]-parmA[5];
2752 ddphi += c5*rbp*cosfac;
2755 dvdl_term += cosfac*rbpBA;
2757 ddphi = -ddphi*sin_phi; /* 11 */
2759 do_dih_fup(ai, aj, ak, al, ddphi, r_ij, r_kj, r_kl, m, n,
2760 f, fshift, pbc, g, x, t1, t2, t3); /* 112 */
2763 *dvdlambda += dvdl_term;
2768 int cmap_setup_grid_index(int ip, int grid_spacing, int *ipm1, int *ipp1, int *ipp2)
2774 ip = ip + grid_spacing - 1;
2776 else if (ip > grid_spacing)
2778 ip = ip - grid_spacing - 1;
2787 im1 = grid_spacing - 1;
2789 else if (ip == grid_spacing-2)
2793 else if (ip == grid_spacing-1)
2807 real cmap_dihs(int nbonds,
2808 const t_iatom forceatoms[], const t_iparams forceparams[],
2809 const gmx_cmap_t *cmap_grid,
2810 const rvec x[], rvec f[], rvec fshift[],
2811 const t_pbc *pbc, const t_graph *g,
2812 real gmx_unused lambda, real gmx_unused *dvdlambda,
2813 const t_mdatoms gmx_unused *md, t_fcdata gmx_unused *fcd,
2814 int gmx_unused *global_atom_index)
2816 int i, j, k, n, idx;
2817 int ai, aj, ak, al, am;
2818 int a1i, a1j, a1k, a1l, a2i, a2j, a2k, a2l;
2820 int t11, t21, t31, t12, t22, t32;
2821 int iphi1, ip1m1, ip1p1, ip1p2;
2822 int iphi2, ip2m1, ip2p1, ip2p2;
2824 int pos1, pos2, pos3, pos4, tmp;
2826 real ty[4], ty1[4], ty2[4], ty12[4], tc[16], tx[16];
2827 real phi1, psi1, cos_phi1, sin_phi1, sign1, xphi1;
2828 real phi2, psi2, cos_phi2, sin_phi2, sign2, xphi2;
2829 real dx, xx, tt, tu, e, df1, df2, ddf1, ddf2, ddf12, vtot;
2830 real ra21, rb21, rg21, rg1, rgr1, ra2r1, rb2r1, rabr1;
2831 real ra22, rb22, rg22, rg2, rgr2, ra2r2, rb2r2, rabr2;
2832 real fg1, hg1, fga1, hgb1, gaa1, gbb1;
2833 real fg2, hg2, fga2, hgb2, gaa2, gbb2;
2836 rvec r1_ij, r1_kj, r1_kl, m1, n1;
2837 rvec r2_ij, r2_kj, r2_kl, m2, n2;
2838 rvec f1_i, f1_j, f1_k, f1_l;
2839 rvec f2_i, f2_j, f2_k, f2_l;
2840 rvec a1, b1, a2, b2;
2841 rvec f1, g1, h1, f2, g2, h2;
2842 rvec dtf1, dtg1, dth1, dtf2, dtg2, dth2;
2843 ivec jt1, dt1_ij, dt1_kj, dt1_lj;
2844 ivec jt2, dt2_ij, dt2_kj, dt2_lj;
2848 int loop_index[4][4] = {
2855 /* Total CMAP energy */
2858 for (n = 0; n < nbonds; )
2860 /* Five atoms are involved in the two torsions */
2861 type = forceatoms[n++];
2862 ai = forceatoms[n++];
2863 aj = forceatoms[n++];
2864 ak = forceatoms[n++];
2865 al = forceatoms[n++];
2866 am = forceatoms[n++];
2868 /* Which CMAP type is this */
2869 cmapA = forceparams[type].cmap.cmapA;
2870 cmapd = cmap_grid->cmapdata[cmapA].cmap;
2878 phi1 = dih_angle(x[a1i], x[a1j], x[a1k], x[a1l], pbc, r1_ij, r1_kj, r1_kl, m1, n1,
2879 &sign1, &t11, &t21, &t31); /* 84 */
2881 cos_phi1 = cos(phi1);
2883 a1[0] = r1_ij[1]*r1_kj[2]-r1_ij[2]*r1_kj[1];
2884 a1[1] = r1_ij[2]*r1_kj[0]-r1_ij[0]*r1_kj[2];
2885 a1[2] = r1_ij[0]*r1_kj[1]-r1_ij[1]*r1_kj[0]; /* 9 */
2887 b1[0] = r1_kl[1]*r1_kj[2]-r1_kl[2]*r1_kj[1];
2888 b1[1] = r1_kl[2]*r1_kj[0]-r1_kl[0]*r1_kj[2];
2889 b1[2] = r1_kl[0]*r1_kj[1]-r1_kl[1]*r1_kj[0]; /* 9 */
2891 tmp = pbc_rvec_sub(pbc, x[a1l], x[a1k], h1);
2893 ra21 = iprod(a1, a1); /* 5 */
2894 rb21 = iprod(b1, b1); /* 5 */
2895 rg21 = iprod(r1_kj, r1_kj); /* 5 */
2901 rabr1 = sqrt(ra2r1*rb2r1);
2903 sin_phi1 = rg1 * rabr1 * iprod(a1, h1) * (-1);
2905 if (cos_phi1 < -0.5 || cos_phi1 > 0.5)
2907 phi1 = asin(sin_phi1);
2917 phi1 = -M_PI - phi1;
2923 phi1 = acos(cos_phi1);
2931 xphi1 = phi1 + M_PI; /* 1 */
2933 /* Second torsion */
2939 phi2 = dih_angle(x[a2i], x[a2j], x[a2k], x[a2l], pbc, r2_ij, r2_kj, r2_kl, m2, n2,
2940 &sign2, &t12, &t22, &t32); /* 84 */
2942 cos_phi2 = cos(phi2);
2944 a2[0] = r2_ij[1]*r2_kj[2]-r2_ij[2]*r2_kj[1];
2945 a2[1] = r2_ij[2]*r2_kj[0]-r2_ij[0]*r2_kj[2];
2946 a2[2] = r2_ij[0]*r2_kj[1]-r2_ij[1]*r2_kj[0]; /* 9 */
2948 b2[0] = r2_kl[1]*r2_kj[2]-r2_kl[2]*r2_kj[1];
2949 b2[1] = r2_kl[2]*r2_kj[0]-r2_kl[0]*r2_kj[2];
2950 b2[2] = r2_kl[0]*r2_kj[1]-r2_kl[1]*r2_kj[0]; /* 9 */
2952 tmp = pbc_rvec_sub(pbc, x[a2l], x[a2k], h2);
2954 ra22 = iprod(a2, a2); /* 5 */
2955 rb22 = iprod(b2, b2); /* 5 */
2956 rg22 = iprod(r2_kj, r2_kj); /* 5 */
2962 rabr2 = sqrt(ra2r2*rb2r2);
2964 sin_phi2 = rg2 * rabr2 * iprod(a2, h2) * (-1);
2966 if (cos_phi2 < -0.5 || cos_phi2 > 0.5)
2968 phi2 = asin(sin_phi2);
2978 phi2 = -M_PI - phi2;
2984 phi2 = acos(cos_phi2);
2992 xphi2 = phi2 + M_PI; /* 1 */
2994 /* Range mangling */
2997 xphi1 = xphi1 + 2*M_PI;
2999 else if (xphi1 >= 2*M_PI)
3001 xphi1 = xphi1 - 2*M_PI;
3006 xphi2 = xphi2 + 2*M_PI;
3008 else if (xphi2 >= 2*M_PI)
3010 xphi2 = xphi2 - 2*M_PI;
3013 /* Number of grid points */
3014 dx = 2*M_PI / cmap_grid->grid_spacing;
3016 /* Where on the grid are we */
3017 iphi1 = (int)(xphi1/dx);
3018 iphi2 = (int)(xphi2/dx);
3020 iphi1 = cmap_setup_grid_index(iphi1, cmap_grid->grid_spacing, &ip1m1, &ip1p1, &ip1p2);
3021 iphi2 = cmap_setup_grid_index(iphi2, cmap_grid->grid_spacing, &ip2m1, &ip2p1, &ip2p2);
3023 pos1 = iphi1*cmap_grid->grid_spacing+iphi2;
3024 pos2 = ip1p1*cmap_grid->grid_spacing+iphi2;
3025 pos3 = ip1p1*cmap_grid->grid_spacing+ip2p1;
3026 pos4 = iphi1*cmap_grid->grid_spacing+ip2p1;
3028 ty[0] = cmapd[pos1*4];
3029 ty[1] = cmapd[pos2*4];
3030 ty[2] = cmapd[pos3*4];
3031 ty[3] = cmapd[pos4*4];
3033 ty1[0] = cmapd[pos1*4+1];
3034 ty1[1] = cmapd[pos2*4+1];
3035 ty1[2] = cmapd[pos3*4+1];
3036 ty1[3] = cmapd[pos4*4+1];
3038 ty2[0] = cmapd[pos1*4+2];
3039 ty2[1] = cmapd[pos2*4+2];
3040 ty2[2] = cmapd[pos3*4+2];
3041 ty2[3] = cmapd[pos4*4+2];
3043 ty12[0] = cmapd[pos1*4+3];
3044 ty12[1] = cmapd[pos2*4+3];
3045 ty12[2] = cmapd[pos3*4+3];
3046 ty12[3] = cmapd[pos4*4+3];
3048 /* Switch to degrees */
3049 dx = 360.0 / cmap_grid->grid_spacing;
3050 xphi1 = xphi1 * RAD2DEG;
3051 xphi2 = xphi2 * RAD2DEG;
3053 for (i = 0; i < 4; i++) /* 16 */
3056 tx[i+4] = ty1[i]*dx;
3057 tx[i+8] = ty2[i]*dx;
3058 tx[i+12] = ty12[i]*dx*dx;
3062 for (i = 0; i < 4; i++) /* 1056 */
3064 for (j = 0; j < 4; j++)
3067 for (k = 0; k < 16; k++)
3069 xx = xx + cmap_coeff_matrix[k*16+idx]*tx[k];
3077 tt = (xphi1-iphi1*dx)/dx;
3078 tu = (xphi2-iphi2*dx)/dx;
3087 for (i = 3; i >= 0; i--)
3089 l1 = loop_index[i][3];
3090 l2 = loop_index[i][2];
3091 l3 = loop_index[i][1];
3093 e = tt * e + ((tc[i*4+3]*tu+tc[i*4+2])*tu + tc[i*4+1])*tu+tc[i*4];
3094 df1 = tu * df1 + (3.0*tc[l1]*tt+2.0*tc[l2])*tt+tc[l3];
3095 df2 = tt * df2 + (3.0*tc[i*4+3]*tu+2.0*tc[i*4+2])*tu+tc[i*4+1];
3096 ddf1 = tu * ddf1 + 2.0*3.0*tc[l1]*tt+2.0*tc[l2];
3097 ddf2 = tt * ddf2 + 2.0*3.0*tc[4*i+3]*tu+2.0*tc[4*i+2];
3100 ddf12 = tc[5] + 2.0*tc[9]*tt + 3.0*tc[13]*tt*tt + 2.0*tu*(tc[6]+2.0*tc[10]*tt+3.0*tc[14]*tt*tt) +
3101 3.0*tu*tu*(tc[7]+2.0*tc[11]*tt+3.0*tc[15]*tt*tt);
3106 ddf1 = ddf1 * fac * fac;
3107 ddf2 = ddf2 * fac * fac;
3108 ddf12 = ddf12 * fac * fac;
3113 /* Do forces - first torsion */
3114 fg1 = iprod(r1_ij, r1_kj);
3115 hg1 = iprod(r1_kl, r1_kj);
3116 fga1 = fg1*ra2r1*rgr1;
3117 hgb1 = hg1*rb2r1*rgr1;
3121 for (i = 0; i < DIM; i++)
3123 dtf1[i] = gaa1 * a1[i];
3124 dtg1[i] = fga1 * a1[i] - hgb1 * b1[i];
3125 dth1[i] = gbb1 * b1[i];
3127 f1[i] = df1 * dtf1[i];
3128 g1[i] = df1 * dtg1[i];
3129 h1[i] = df1 * dth1[i];
3132 f1_j[i] = -f1[i] - g1[i];
3133 f1_k[i] = h1[i] + g1[i];
3136 f[a1i][i] = f[a1i][i] + f1_i[i];
3137 f[a1j][i] = f[a1j][i] + f1_j[i]; /* - f1[i] - g1[i] */
3138 f[a1k][i] = f[a1k][i] + f1_k[i]; /* h1[i] + g1[i] */
3139 f[a1l][i] = f[a1l][i] + f1_l[i]; /* h1[i] */
3142 /* Do forces - second torsion */
3143 fg2 = iprod(r2_ij, r2_kj);
3144 hg2 = iprod(r2_kl, r2_kj);
3145 fga2 = fg2*ra2r2*rgr2;
3146 hgb2 = hg2*rb2r2*rgr2;
3150 for (i = 0; i < DIM; i++)
3152 dtf2[i] = gaa2 * a2[i];
3153 dtg2[i] = fga2 * a2[i] - hgb2 * b2[i];
3154 dth2[i] = gbb2 * b2[i];
3156 f2[i] = df2 * dtf2[i];
3157 g2[i] = df2 * dtg2[i];
3158 h2[i] = df2 * dth2[i];
3161 f2_j[i] = -f2[i] - g2[i];
3162 f2_k[i] = h2[i] + g2[i];
3165 f[a2i][i] = f[a2i][i] + f2_i[i]; /* f2[i] */
3166 f[a2j][i] = f[a2j][i] + f2_j[i]; /* - f2[i] - g2[i] */
3167 f[a2k][i] = f[a2k][i] + f2_k[i]; /* h2[i] + g2[i] */
3168 f[a2l][i] = f[a2l][i] + f2_l[i]; /* - h2[i] */
3174 copy_ivec(SHIFT_IVEC(g, a1j), jt1);
3175 ivec_sub(SHIFT_IVEC(g, a1i), jt1, dt1_ij);
3176 ivec_sub(SHIFT_IVEC(g, a1k), jt1, dt1_kj);
3177 ivec_sub(SHIFT_IVEC(g, a1l), jt1, dt1_lj);
3178 t11 = IVEC2IS(dt1_ij);
3179 t21 = IVEC2IS(dt1_kj);
3180 t31 = IVEC2IS(dt1_lj);
3182 copy_ivec(SHIFT_IVEC(g, a2j), jt2);
3183 ivec_sub(SHIFT_IVEC(g, a2i), jt2, dt2_ij);
3184 ivec_sub(SHIFT_IVEC(g, a2k), jt2, dt2_kj);
3185 ivec_sub(SHIFT_IVEC(g, a2l), jt2, dt2_lj);
3186 t12 = IVEC2IS(dt2_ij);
3187 t22 = IVEC2IS(dt2_kj);
3188 t32 = IVEC2IS(dt2_lj);
3192 t31 = pbc_rvec_sub(pbc, x[a1l], x[a1j], h1);
3193 t32 = pbc_rvec_sub(pbc, x[a2l], x[a2j], h2);
3201 rvec_inc(fshift[t11], f1_i);
3202 rvec_inc(fshift[CENTRAL], f1_j);
3203 rvec_inc(fshift[t21], f1_k);
3204 rvec_inc(fshift[t31], f1_l);
3206 rvec_inc(fshift[t21], f2_i);
3207 rvec_inc(fshift[CENTRAL], f2_j);
3208 rvec_inc(fshift[t22], f2_k);
3209 rvec_inc(fshift[t32], f2_l);
3216 /***********************************************************
3218 * G R O M O S 9 6 F U N C T I O N S
3220 ***********************************************************/
3221 real g96harmonic(real kA, real kB, real xA, real xB, real x, real lambda,
3224 const real half = 0.5;
3225 real L1, kk, x0, dx, dx2;
3226 real v, f, dvdlambda;
3229 kk = L1*kA+lambda*kB;
3230 x0 = L1*xA+lambda*xB;
3237 dvdlambda = half*(kB-kA)*dx2 + (xA-xB)*kk*dx;
3244 /* That was 21 flops */
3247 real g96bonds(int nbonds,
3248 const t_iatom forceatoms[], const t_iparams forceparams[],
3249 const rvec x[], rvec f[], rvec fshift[],
3250 const t_pbc *pbc, const t_graph *g,
3251 real lambda, real *dvdlambda,
3252 const t_mdatoms gmx_unused *md, t_fcdata gmx_unused *fcd,
3253 int gmx_unused *global_atom_index)
3255 int i, m, ki, ai, aj, type;
3256 real dr2, fbond, vbond, fij, vtot;
3261 for (i = 0; (i < nbonds); )
3263 type = forceatoms[i++];
3264 ai = forceatoms[i++];
3265 aj = forceatoms[i++];
3267 ki = pbc_rvec_sub(pbc, x[ai], x[aj], dx); /* 3 */
3268 dr2 = iprod(dx, dx); /* 5 */
3270 *dvdlambda += g96harmonic(forceparams[type].harmonic.krA,
3271 forceparams[type].harmonic.krB,
3272 forceparams[type].harmonic.rA,
3273 forceparams[type].harmonic.rB,
3274 dr2, lambda, &vbond, &fbond);
3276 vtot += 0.5*vbond; /* 1*/
3280 fprintf(debug, "G96-BONDS: dr = %10g vbond = %10g fbond = %10g\n",
3281 sqrt(dr2), vbond, fbond);
3287 ivec_sub(SHIFT_IVEC(g, ai), SHIFT_IVEC(g, aj), dt);
3290 for (m = 0; (m < DIM); m++) /* 15 */
3295 fshift[ki][m] += fij;
3296 fshift[CENTRAL][m] -= fij;
3302 real g96bond_angle(const rvec xi, const rvec xj, const rvec xk, const t_pbc *pbc,
3303 rvec r_ij, rvec r_kj,
3305 /* Return value is the angle between the bonds i-j and j-k */
3309 *t1 = pbc_rvec_sub(pbc, xi, xj, r_ij); /* 3 */
3310 *t2 = pbc_rvec_sub(pbc, xk, xj, r_kj); /* 3 */
3312 costh = cos_angle(r_ij, r_kj); /* 25 */
3317 real g96angles(int nbonds,
3318 const t_iatom forceatoms[], const t_iparams forceparams[],
3319 const rvec x[], rvec f[], rvec fshift[],
3320 const t_pbc *pbc, const t_graph *g,
3321 real lambda, real *dvdlambda,
3322 const t_mdatoms gmx_unused *md, t_fcdata gmx_unused *fcd,
3323 int gmx_unused *global_atom_index)
3325 int i, ai, aj, ak, type, m, t1, t2;
3327 real cos_theta, dVdt, va, vtot;
3328 real rij_1, rij_2, rkj_1, rkj_2, rijrkj_1;
3330 ivec jt, dt_ij, dt_kj;
3333 for (i = 0; (i < nbonds); )
3335 type = forceatoms[i++];
3336 ai = forceatoms[i++];
3337 aj = forceatoms[i++];
3338 ak = forceatoms[i++];
3340 cos_theta = g96bond_angle(x[ai], x[aj], x[ak], pbc, r_ij, r_kj, &t1, &t2);
3342 *dvdlambda += g96harmonic(forceparams[type].harmonic.krA,
3343 forceparams[type].harmonic.krB,
3344 forceparams[type].harmonic.rA,
3345 forceparams[type].harmonic.rB,
3346 cos_theta, lambda, &va, &dVdt);
3349 rij_1 = gmx_invsqrt(iprod(r_ij, r_ij));
3350 rkj_1 = gmx_invsqrt(iprod(r_kj, r_kj));
3351 rij_2 = rij_1*rij_1;
3352 rkj_2 = rkj_1*rkj_1;
3353 rijrkj_1 = rij_1*rkj_1; /* 23 */
3358 fprintf(debug, "G96ANGLES: costheta = %10g vth = %10g dV/dct = %10g\n",
3359 cos_theta, va, dVdt);
3362 for (m = 0; (m < DIM); m++) /* 42 */
3364 f_i[m] = dVdt*(r_kj[m]*rijrkj_1 - r_ij[m]*rij_2*cos_theta);
3365 f_k[m] = dVdt*(r_ij[m]*rijrkj_1 - r_kj[m]*rkj_2*cos_theta);
3366 f_j[m] = -f_i[m]-f_k[m];
3374 copy_ivec(SHIFT_IVEC(g, aj), jt);
3376 ivec_sub(SHIFT_IVEC(g, ai), jt, dt_ij);
3377 ivec_sub(SHIFT_IVEC(g, ak), jt, dt_kj);
3378 t1 = IVEC2IS(dt_ij);
3379 t2 = IVEC2IS(dt_kj);
3381 rvec_inc(fshift[t1], f_i);
3382 rvec_inc(fshift[CENTRAL], f_j);
3383 rvec_inc(fshift[t2], f_k); /* 9 */
3389 real cross_bond_bond(int nbonds,
3390 const t_iatom forceatoms[], const t_iparams forceparams[],
3391 const rvec x[], rvec f[], rvec fshift[],
3392 const t_pbc *pbc, const t_graph *g,
3393 real gmx_unused lambda, real gmx_unused *dvdlambda,
3394 const t_mdatoms gmx_unused *md, t_fcdata gmx_unused *fcd,
3395 int gmx_unused *global_atom_index)
3397 /* Potential from Lawrence and Skimmer, Chem. Phys. Lett. 372 (2003)
3400 int i, ai, aj, ak, type, m, t1, t2;
3402 real vtot, vrr, s1, s2, r1, r2, r1e, r2e, krr;
3404 ivec jt, dt_ij, dt_kj;
3407 for (i = 0; (i < nbonds); )
3409 type = forceatoms[i++];
3410 ai = forceatoms[i++];
3411 aj = forceatoms[i++];
3412 ak = forceatoms[i++];
3413 r1e = forceparams[type].cross_bb.r1e;
3414 r2e = forceparams[type].cross_bb.r2e;
3415 krr = forceparams[type].cross_bb.krr;
3417 /* Compute distance vectors ... */
3418 t1 = pbc_rvec_sub(pbc, x[ai], x[aj], r_ij);
3419 t2 = pbc_rvec_sub(pbc, x[ak], x[aj], r_kj);
3421 /* ... and their lengths */
3425 /* Deviations from ideality */
3429 /* Energy (can be negative!) */
3434 svmul(-krr*s2/r1, r_ij, f_i);
3435 svmul(-krr*s1/r2, r_kj, f_k);
3437 for (m = 0; (m < DIM); m++) /* 12 */
3439 f_j[m] = -f_i[m] - f_k[m];
3448 copy_ivec(SHIFT_IVEC(g, aj), jt);
3450 ivec_sub(SHIFT_IVEC(g, ai), jt, dt_ij);
3451 ivec_sub(SHIFT_IVEC(g, ak), jt, dt_kj);
3452 t1 = IVEC2IS(dt_ij);
3453 t2 = IVEC2IS(dt_kj);
3455 rvec_inc(fshift[t1], f_i);
3456 rvec_inc(fshift[CENTRAL], f_j);
3457 rvec_inc(fshift[t2], f_k); /* 9 */
3463 real cross_bond_angle(int nbonds,
3464 const t_iatom forceatoms[], const t_iparams forceparams[],
3465 const rvec x[], rvec f[], rvec fshift[],
3466 const t_pbc *pbc, const t_graph *g,
3467 real gmx_unused lambda, real gmx_unused *dvdlambda,
3468 const t_mdatoms gmx_unused *md, t_fcdata gmx_unused *fcd,
3469 int gmx_unused *global_atom_index)
3471 /* Potential from Lawrence and Skimmer, Chem. Phys. Lett. 372 (2003)
3474 int i, ai, aj, ak, type, m, t1, t2, t3;
3475 rvec r_ij, r_kj, r_ik;
3476 real vtot, vrt, s1, s2, s3, r1, r2, r3, r1e, r2e, r3e, krt, k1, k2, k3;
3478 ivec jt, dt_ij, dt_kj;
3481 for (i = 0; (i < nbonds); )
3483 type = forceatoms[i++];
3484 ai = forceatoms[i++];
3485 aj = forceatoms[i++];
3486 ak = forceatoms[i++];
3487 r1e = forceparams[type].cross_ba.r1e;
3488 r2e = forceparams[type].cross_ba.r2e;
3489 r3e = forceparams[type].cross_ba.r3e;
3490 krt = forceparams[type].cross_ba.krt;
3492 /* Compute distance vectors ... */
3493 t1 = pbc_rvec_sub(pbc, x[ai], x[aj], r_ij);
3494 t2 = pbc_rvec_sub(pbc, x[ak], x[aj], r_kj);
3495 t3 = pbc_rvec_sub(pbc, x[ai], x[ak], r_ik);
3497 /* ... and their lengths */
3502 /* Deviations from ideality */
3507 /* Energy (can be negative!) */
3508 vrt = krt*s3*(s1+s2);
3514 k3 = -krt*(s1+s2)/r3;
3515 for (m = 0; (m < DIM); m++)
3517 f_i[m] = k1*r_ij[m] + k3*r_ik[m];
3518 f_k[m] = k2*r_kj[m] - k3*r_ik[m];
3519 f_j[m] = -f_i[m] - f_k[m];
3522 for (m = 0; (m < DIM); m++) /* 12 */
3532 copy_ivec(SHIFT_IVEC(g, aj), jt);
3534 ivec_sub(SHIFT_IVEC(g, ai), jt, dt_ij);
3535 ivec_sub(SHIFT_IVEC(g, ak), jt, dt_kj);
3536 t1 = IVEC2IS(dt_ij);
3537 t2 = IVEC2IS(dt_kj);
3539 rvec_inc(fshift[t1], f_i);
3540 rvec_inc(fshift[CENTRAL], f_j);
3541 rvec_inc(fshift[t2], f_k); /* 9 */
3547 static real bonded_tab(const char *type, int table_nr,
3548 const bondedtable_t *table, real kA, real kB, real r,
3549 real lambda, real *V, real *F)
3551 real k, tabscale, *VFtab, rt, eps, eps2, Yt, Ft, Geps, Heps2, Fp, VV, FF;
3553 real v, f, dvdlambda;
3555 k = (1.0 - lambda)*kA + lambda*kB;
3557 tabscale = table->scale;
3558 VFtab = table->data;
3564 gmx_fatal(FARGS, "A tabulated %s interaction table number %d is out of the table range: r %f, between table indices %d and %d, table length %d",
3565 type, table_nr, r, n0, n0+1, table->n);
3572 Geps = VFtab[nnn+2]*eps;
3573 Heps2 = VFtab[nnn+3]*eps2;
3574 Fp = Ft + Geps + Heps2;
3576 FF = Fp + Geps + 2.0*Heps2;
3578 *F = -k*FF*tabscale;
3580 dvdlambda = (kB - kA)*VV;
3584 /* That was 22 flops */
3587 real tab_bonds(int nbonds,
3588 const t_iatom forceatoms[], const t_iparams forceparams[],
3589 const rvec x[], rvec f[], rvec fshift[],
3590 const t_pbc *pbc, const t_graph *g,
3591 real lambda, real *dvdlambda,
3592 const t_mdatoms gmx_unused *md, t_fcdata *fcd,
3593 int gmx_unused *global_atom_index)
3595 int i, m, ki, ai, aj, type, table;
3596 real dr, dr2, fbond, vbond, fij, vtot;
3601 for (i = 0; (i < nbonds); )
3603 type = forceatoms[i++];
3604 ai = forceatoms[i++];
3605 aj = forceatoms[i++];
3607 ki = pbc_rvec_sub(pbc, x[ai], x[aj], dx); /* 3 */
3608 dr2 = iprod(dx, dx); /* 5 */
3609 dr = dr2*gmx_invsqrt(dr2); /* 10 */
3611 table = forceparams[type].tab.table;
3613 *dvdlambda += bonded_tab("bond", table,
3614 &fcd->bondtab[table],
3615 forceparams[type].tab.kA,
3616 forceparams[type].tab.kB,
3617 dr, lambda, &vbond, &fbond); /* 22 */
3625 vtot += vbond; /* 1*/
3626 fbond *= gmx_invsqrt(dr2); /* 6 */
3630 fprintf(debug, "TABBONDS: dr = %10g vbond = %10g fbond = %10g\n",
3636 ivec_sub(SHIFT_IVEC(g, ai), SHIFT_IVEC(g, aj), dt);
3639 for (m = 0; (m < DIM); m++) /* 15 */
3644 fshift[ki][m] += fij;
3645 fshift[CENTRAL][m] -= fij;
3651 real tab_angles(int nbonds,
3652 const t_iatom forceatoms[], const t_iparams forceparams[],
3653 const rvec x[], rvec f[], rvec fshift[],
3654 const t_pbc *pbc, const t_graph *g,
3655 real lambda, real *dvdlambda,
3656 const t_mdatoms gmx_unused *md, t_fcdata *fcd,
3657 int gmx_unused *global_atom_index)
3659 int i, ai, aj, ak, t1, t2, type, table;
3661 real cos_theta, cos_theta2, theta, dVdt, va, vtot;
3662 ivec jt, dt_ij, dt_kj;
3665 for (i = 0; (i < nbonds); )
3667 type = forceatoms[i++];
3668 ai = forceatoms[i++];
3669 aj = forceatoms[i++];
3670 ak = forceatoms[i++];
3672 theta = bond_angle(x[ai], x[aj], x[ak], pbc,
3673 r_ij, r_kj, &cos_theta, &t1, &t2); /* 41 */
3675 table = forceparams[type].tab.table;
3677 *dvdlambda += bonded_tab("angle", table,
3678 &fcd->angletab[table],
3679 forceparams[type].tab.kA,
3680 forceparams[type].tab.kB,
3681 theta, lambda, &va, &dVdt); /* 22 */
3684 cos_theta2 = sqr(cos_theta); /* 1 */
3693 st = dVdt*gmx_invsqrt(1 - cos_theta2); /* 12 */
3694 sth = st*cos_theta; /* 1 */
3698 fprintf(debug, "ANGLES: theta = %10g vth = %10g dV/dtheta = %10g\n",
3699 theta*RAD2DEG, va, dVdt);
3702 nrkj2 = iprod(r_kj, r_kj); /* 5 */
3703 nrij2 = iprod(r_ij, r_ij);
3705 cik = st*gmx_invsqrt(nrkj2*nrij2); /* 12 */
3706 cii = sth/nrij2; /* 10 */
3707 ckk = sth/nrkj2; /* 10 */
3709 for (m = 0; (m < DIM); m++) /* 39 */
3711 f_i[m] = -(cik*r_kj[m]-cii*r_ij[m]);
3712 f_k[m] = -(cik*r_ij[m]-ckk*r_kj[m]);
3713 f_j[m] = -f_i[m]-f_k[m];
3720 copy_ivec(SHIFT_IVEC(g, aj), jt);
3722 ivec_sub(SHIFT_IVEC(g, ai), jt, dt_ij);
3723 ivec_sub(SHIFT_IVEC(g, ak), jt, dt_kj);
3724 t1 = IVEC2IS(dt_ij);
3725 t2 = IVEC2IS(dt_kj);
3727 rvec_inc(fshift[t1], f_i);
3728 rvec_inc(fshift[CENTRAL], f_j);
3729 rvec_inc(fshift[t2], f_k);
3735 real tab_dihs(int nbonds,
3736 const t_iatom forceatoms[], const t_iparams forceparams[],
3737 const rvec x[], rvec f[], rvec fshift[],
3738 const t_pbc *pbc, const t_graph *g,
3739 real lambda, real *dvdlambda,
3740 const t_mdatoms gmx_unused *md, t_fcdata *fcd,
3741 int gmx_unused *global_atom_index)
3743 int i, type, ai, aj, ak, al, table;
3745 rvec r_ij, r_kj, r_kl, m, n;
3746 real phi, sign, ddphi, vpd, vtot;
3749 for (i = 0; (i < nbonds); )
3751 type = forceatoms[i++];
3752 ai = forceatoms[i++];
3753 aj = forceatoms[i++];
3754 ak = forceatoms[i++];
3755 al = forceatoms[i++];
3757 phi = dih_angle(x[ai], x[aj], x[ak], x[al], pbc, r_ij, r_kj, r_kl, m, n,
3758 &sign, &t1, &t2, &t3); /* 84 */
3760 table = forceparams[type].tab.table;
3762 /* Hopefully phi+M_PI never results in values < 0 */
3763 *dvdlambda += bonded_tab("dihedral", table,
3764 &fcd->dihtab[table],
3765 forceparams[type].tab.kA,
3766 forceparams[type].tab.kB,
3767 phi+M_PI, lambda, &vpd, &ddphi);
3770 do_dih_fup(ai, aj, ak, al, -ddphi, r_ij, r_kj, r_kl, m, n,
3771 f, fshift, pbc, g, x, t1, t2, t3); /* 112 */
3774 fprintf(debug, "pdih: (%d,%d,%d,%d) phi=%g\n",
3775 ai, aj, ak, al, phi);
3782 /* Return if this is a potential calculated in bondfree.c,
3783 * i.e. an interaction that actually calculates a potential and
3784 * works on multiple atoms (not e.g. a connection or a position restraint).
3786 static gmx_inline gmx_bool ftype_is_bonded_potential(int ftype)
3789 (interaction_function[ftype].flags & IF_BOND) &&
3790 !(ftype == F_CONNBONDS || ftype == F_POSRES || ftype == F_FBPOSRES) &&
3791 (ftype < F_GB12 || ftype > F_GB14);
3794 static void divide_bondeds_over_threads(t_idef *idef, int nthreads)
3801 idef->nthreads = nthreads;
3803 if (F_NRE*(nthreads+1) > idef->il_thread_division_nalloc)
3805 idef->il_thread_division_nalloc = F_NRE*(nthreads+1);
3806 snew(idef->il_thread_division, idef->il_thread_division_nalloc);
3809 for (ftype = 0; ftype < F_NRE; ftype++)
3811 if (ftype_is_bonded_potential(ftype))
3813 nat1 = interaction_function[ftype].nratoms + 1;
3815 for (t = 0; t <= nthreads; t++)
3817 /* Divide the interactions equally over the threads.
3818 * When the different types of bonded interactions
3819 * are distributed roughly equally over the threads,
3820 * this should lead to well localized output into
3821 * the force buffer on each thread.
3822 * If this is not the case, a more advanced scheme
3823 * (not implemented yet) will do better.
3825 il_nr_thread = (((idef->il[ftype].nr/nat1)*t)/nthreads)*nat1;
3827 /* Ensure that distance restraint pairs with the same label
3828 * end up on the same thread.
3829 * This is slighlty tricky code, since the next for iteration
3830 * may have an initial il_nr_thread lower than the final value
3831 * in the previous iteration, but this will anyhow be increased
3832 * to the approriate value again by this while loop.
3834 while (ftype == F_DISRES &&
3836 il_nr_thread < idef->il[ftype].nr &&
3837 idef->iparams[idef->il[ftype].iatoms[il_nr_thread]].disres.label ==
3838 idef->iparams[idef->il[ftype].iatoms[il_nr_thread-nat1]].disres.label)
3840 il_nr_thread += nat1;
3843 idef->il_thread_division[ftype*(nthreads+1)+t] = il_nr_thread;
3850 calc_bonded_reduction_mask(const t_idef *idef,
3855 int ftype, nb, nat1, nb0, nb1, i, a;
3859 for (ftype = 0; ftype < F_NRE; ftype++)
3861 if (ftype_is_bonded_potential(ftype))
3863 nb = idef->il[ftype].nr;
3866 nat1 = interaction_function[ftype].nratoms + 1;
3868 /* Divide this interaction equally over the threads.
3869 * This is not stored: should match division in calc_bonds.
3871 nb0 = idef->il_thread_division[ftype*(nt+1)+t];
3872 nb1 = idef->il_thread_division[ftype*(nt+1)+t+1];
3874 for (i = nb0; i < nb1; i += nat1)
3876 for (a = 1; a < nat1; a++)
3878 mask |= (1U << (idef->il[ftype].iatoms[i+a]>>shift));
3888 void setup_bonded_threading(t_forcerec *fr, t_idef *idef)
3890 #define MAX_BLOCK_BITS 32
3894 assert(fr->nthreads >= 1);
3896 /* Divide the bonded interaction over the threads */
3897 divide_bondeds_over_threads(idef, fr->nthreads);
3899 if (fr->nthreads == 1)
3906 /* We divide the force array in a maximum of 32 blocks.
3907 * Minimum force block reduction size is 2^6=64.
3910 while (fr->natoms_force > (int)(MAX_BLOCK_BITS*(1U<<fr->red_ashift)))
3916 fprintf(debug, "bonded force buffer block atom shift %d bits\n",
3920 /* Determine to which blocks each thread's bonded force calculation
3921 * contributes. Store this is a mask for each thread.
3923 #pragma omp parallel for num_threads(fr->nthreads) schedule(static)
3924 for (t = 1; t < fr->nthreads; t++)
3926 fr->f_t[t].red_mask =
3927 calc_bonded_reduction_mask(idef, fr->red_ashift, t, fr->nthreads);
3930 /* Determine the maximum number of blocks we need to reduce over */
3933 for (t = 0; t < fr->nthreads; t++)
3936 for (b = 0; b < MAX_BLOCK_BITS; b++)
3938 if (fr->f_t[t].red_mask & (1U<<b))
3940 fr->red_nblock = max(fr->red_nblock, b+1);
3946 fprintf(debug, "thread %d flags %x count %d\n",
3947 t, fr->f_t[t].red_mask, c);
3953 fprintf(debug, "Number of blocks to reduce: %d of size %d\n",
3954 fr->red_nblock, 1<<fr->red_ashift);
3955 fprintf(debug, "Reduction density %.2f density/#thread %.2f\n",
3956 ctot*(1<<fr->red_ashift)/(double)fr->natoms_force,
3957 ctot*(1<<fr->red_ashift)/(double)(fr->natoms_force*fr->nthreads));
3961 static void zero_thread_forces(f_thread_t *f_t, int n,
3962 int nblock, int blocksize)
3964 int b, a0, a1, a, i, j;
3966 if (n > f_t->f_nalloc)
3968 f_t->f_nalloc = over_alloc_large(n);
3969 srenew(f_t->f, f_t->f_nalloc);
3972 if (f_t->red_mask != 0)
3974 for (b = 0; b < nblock; b++)
3976 if (f_t->red_mask && (1U<<b))
3979 a1 = min((b+1)*blocksize, n);
3980 for (a = a0; a < a1; a++)
3982 clear_rvec(f_t->f[a]);
3987 for (i = 0; i < SHIFTS; i++)
3989 clear_rvec(f_t->fshift[i]);
3991 for (i = 0; i < F_NRE; i++)
3995 for (i = 0; i < egNR; i++)
3997 for (j = 0; j < f_t->grpp.nener; j++)
3999 f_t->grpp.ener[i][j] = 0;
4002 for (i = 0; i < efptNR; i++)
4008 static void reduce_thread_force_buffer(int n, rvec *f,
4009 int nthreads, f_thread_t *f_t,
4010 int nblock, int block_size)
4012 /* The max thread number is arbitrary,
4013 * we used a fixed number to avoid memory management.
4014 * Using more than 16 threads is probably never useful performance wise.
4016 #define MAX_BONDED_THREADS 256
4019 if (nthreads > MAX_BONDED_THREADS)
4021 gmx_fatal(FARGS, "Can not reduce bonded forces on more than %d threads",
4022 MAX_BONDED_THREADS);
4025 /* This reduction can run on any number of threads,
4026 * independently of nthreads.
4028 #pragma omp parallel for num_threads(nthreads) schedule(static)
4029 for (b = 0; b < nblock; b++)
4031 rvec *fp[MAX_BONDED_THREADS];
4035 /* Determine which threads contribute to this block */
4037 for (ft = 1; ft < nthreads; ft++)
4039 if (f_t[ft].red_mask & (1U<<b))
4041 fp[nfb++] = f_t[ft].f;
4046 /* Reduce force buffers for threads that contribute */
4048 a1 = (b+1)*block_size;
4050 for (a = a0; a < a1; a++)
4052 for (fb = 0; fb < nfb; fb++)
4054 rvec_inc(f[a], fp[fb][a]);
4061 static void reduce_thread_forces(int n, rvec *f, rvec *fshift,
4062 real *ener, gmx_grppairener_t *grpp, real *dvdl,
4063 int nthreads, f_thread_t *f_t,
4064 int nblock, int block_size,
4065 gmx_bool bCalcEnerVir,
4070 /* Reduce the bonded force buffer */
4071 reduce_thread_force_buffer(n, f, nthreads, f_t, nblock, block_size);
4074 /* When necessary, reduce energy and virial using one thread only */
4079 for (i = 0; i < SHIFTS; i++)
4081 for (t = 1; t < nthreads; t++)
4083 rvec_inc(fshift[i], f_t[t].fshift[i]);
4086 for (i = 0; i < F_NRE; i++)
4088 for (t = 1; t < nthreads; t++)
4090 ener[i] += f_t[t].ener[i];
4093 for (i = 0; i < egNR; i++)
4095 for (j = 0; j < f_t[1].grpp.nener; j++)
4097 for (t = 1; t < nthreads; t++)
4100 grpp->ener[i][j] += f_t[t].grpp.ener[i][j];
4106 for (i = 0; i < efptNR; i++)
4109 for (t = 1; t < nthreads; t++)
4111 dvdl[i] += f_t[t].dvdl[i];
4118 static real calc_one_bond(FILE *fplog, int thread,
4119 int ftype, const t_idef *idef,
4120 rvec x[], rvec f[], rvec fshift[],
4122 const t_pbc *pbc, const t_graph *g,
4123 gmx_grppairener_t *grpp,
4125 real *lambda, real *dvdl,
4126 const t_mdatoms *md, t_fcdata *fcd,
4127 gmx_bool bCalcEnerVir,
4128 int *global_atom_index, gmx_bool bPrintSepPot)
4130 int nat1, nbonds, efptFTYPE;
4135 if (IS_RESTRAINT_TYPE(ftype))
4137 efptFTYPE = efptRESTRAINT;
4141 efptFTYPE = efptBONDED;
4144 nat1 = interaction_function[ftype].nratoms + 1;
4145 nbonds = idef->il[ftype].nr/nat1;
4146 iatoms = idef->il[ftype].iatoms;
4148 nb0 = idef->il_thread_division[ftype*(idef->nthreads+1)+thread];
4149 nbn = idef->il_thread_division[ftype*(idef->nthreads+1)+thread+1] - nb0;
4151 if (!IS_LISTED_LJ_C(ftype))
4153 if (ftype == F_CMAP)
4155 v = cmap_dihs(nbn, iatoms+nb0,
4156 idef->iparams, &idef->cmap_grid,
4157 (const rvec*)x, f, fshift,
4158 pbc, g, lambda[efptFTYPE], &(dvdl[efptFTYPE]),
4159 md, fcd, global_atom_index);
4161 #ifdef GMX_SIMD_HAVE_REAL
4162 else if (ftype == F_ANGLES &&
4163 !bCalcEnerVir && fr->efep == efepNO)
4165 /* No energies, shift forces, dvdl */
4166 angles_noener_simd(nbn, idef->il[ftype].iatoms+nb0,
4169 pbc, g, lambda[efptFTYPE], md, fcd,
4174 else if (ftype == F_PDIHS &&
4175 !bCalcEnerVir && fr->efep == efepNO)
4177 /* No energies, shift forces, dvdl */
4178 #ifdef GMX_SIMD_HAVE_REAL
4183 (nbn, idef->il[ftype].iatoms+nb0,
4186 pbc, g, lambda[efptFTYPE], md, fcd,
4192 v = interaction_function[ftype].ifunc(nbn, iatoms+nb0,
4194 (const rvec*)x, f, fshift,
4195 pbc, g, lambda[efptFTYPE], &(dvdl[efptFTYPE]),
4196 md, fcd, global_atom_index);
4200 fprintf(fplog, " %-23s #%4d V %12.5e dVdl %12.5e\n",
4201 interaction_function[ftype].longname,
4202 nbonds, v, lambda[efptFTYPE]);
4207 v = do_nonbonded_listed(ftype, nbn, iatoms+nb0, idef->iparams, (const rvec*)x, f, fshift,
4208 pbc, g, lambda, dvdl, md, fr, grpp, global_atom_index);
4212 fprintf(fplog, " %-5s + %-15s #%4d dVdl %12.5e\n",
4213 interaction_function[ftype].longname,
4214 interaction_function[F_LJ14].longname, nbonds, dvdl[efptVDW]);
4215 fprintf(fplog, " %-5s + %-15s #%4d dVdl %12.5e\n",
4216 interaction_function[ftype].longname,
4217 interaction_function[F_COUL14].longname, nbonds, dvdl[efptCOUL]);
4223 inc_nrnb(nrnb, interaction_function[ftype].nrnb_ind, nbonds);
4229 void calc_bonds(FILE *fplog, const gmx_multisim_t *ms,
4231 rvec x[], history_t *hist,
4232 rvec f[], t_forcerec *fr,
4233 const t_pbc *pbc, const t_graph *g,
4234 gmx_enerdata_t *enerd, t_nrnb *nrnb,
4236 const t_mdatoms *md,
4237 t_fcdata *fcd, int *global_atom_index,
4238 t_atomtypes gmx_unused *atype, gmx_genborn_t gmx_unused *born,
4240 gmx_bool bPrintSepPot, gmx_int64_t step)
4242 gmx_bool bCalcEnerVir;
4244 real v, dvdl[efptNR], dvdl_dum[efptNR]; /* The dummy array is to have a place to store the dhdl at other values
4245 of lambda, which will be thrown away in the end*/
4246 const t_pbc *pbc_null;
4250 assert(fr->nthreads == idef->nthreads);
4252 bCalcEnerVir = (force_flags & (GMX_FORCE_VIRIAL | GMX_FORCE_ENERGY));
4254 for (i = 0; i < efptNR; i++)
4268 fprintf(fplog, "Step %s: bonded V and dVdl for this node\n",
4269 gmx_step_str(step, buf));
4275 p_graph(debug, "Bondage is fun", g);
4279 /* Do pre force calculation stuff which might require communication */
4280 if (idef->il[F_ORIRES].nr)
4282 enerd->term[F_ORIRESDEV] =
4283 calc_orires_dev(ms, idef->il[F_ORIRES].nr,
4284 idef->il[F_ORIRES].iatoms,
4285 idef->iparams, md, (const rvec*)x,
4286 pbc_null, fcd, hist);
4288 if (idef->il[F_DISRES].nr)
4290 calc_disres_R_6(idef->il[F_DISRES].nr,
4291 idef->il[F_DISRES].iatoms,
4292 idef->iparams, (const rvec*)x, pbc_null,
4295 if (fcd->disres.nsystems > 1)
4297 gmx_sum_sim(2*fcd->disres.nres, fcd->disres.Rt_6, ms);
4302 #pragma omp parallel for num_threads(fr->nthreads) schedule(static)
4303 for (thread = 0; thread < fr->nthreads; thread++)
4310 gmx_grppairener_t *grpp;
4315 fshift = fr->fshift;
4317 grpp = &enerd->grpp;
4322 zero_thread_forces(&fr->f_t[thread], fr->natoms_force,
4323 fr->red_nblock, 1<<fr->red_ashift);
4325 ft = fr->f_t[thread].f;
4326 fshift = fr->f_t[thread].fshift;
4327 epot = fr->f_t[thread].ener;
4328 grpp = &fr->f_t[thread].grpp;
4329 dvdlt = fr->f_t[thread].dvdl;
4331 /* Loop over all bonded force types to calculate the bonded forces */
4332 for (ftype = 0; (ftype < F_NRE); ftype++)
4334 if (idef->il[ftype].nr > 0 && ftype_is_bonded_potential(ftype))
4336 v = calc_one_bond(fplog, thread, ftype, idef, x,
4337 ft, fshift, fr, pbc_null, g, grpp,
4338 nrnb, lambda, dvdlt,
4339 md, fcd, bCalcEnerVir,
4340 global_atom_index, bPrintSepPot);
4345 if (fr->nthreads > 1)
4347 reduce_thread_forces(fr->natoms_force, f, fr->fshift,
4348 enerd->term, &enerd->grpp, dvdl,
4349 fr->nthreads, fr->f_t,
4350 fr->red_nblock, 1<<fr->red_ashift,
4352 force_flags & GMX_FORCE_DHDL);
4354 if (force_flags & GMX_FORCE_DHDL)
4356 for (i = 0; i < efptNR; i++)
4358 enerd->dvdl_nonlin[i] += dvdl[i];
4362 /* Copy the sum of violations for the distance restraints from fcd */
4365 enerd->term[F_DISRESVIOL] = fcd->disres.sumviol;
4370 void calc_bonds_lambda(FILE *fplog,
4374 const t_pbc *pbc, const t_graph *g,
4375 gmx_grppairener_t *grpp, real *epot, t_nrnb *nrnb,
4377 const t_mdatoms *md,
4379 int *global_atom_index)
4381 int i, ftype, nr_nonperturbed, nr;
4383 real dvdl_dum[efptNR];
4385 const t_pbc *pbc_null;
4397 /* Copy the whole idef, so we can modify the contents locally */
4399 idef_fe.nthreads = 1;
4400 snew(idef_fe.il_thread_division, F_NRE*(idef_fe.nthreads+1));
4402 /* We already have the forces, so we use temp buffers here */
4403 snew(f, fr->natoms_force);
4404 snew(fshift, SHIFTS);
4406 /* Loop over all bonded force types to calculate the bonded energies */
4407 for (ftype = 0; (ftype < F_NRE); ftype++)
4409 if (ftype_is_bonded_potential(ftype))
4411 /* Set the work range of thread 0 to the perturbed bondeds only */
4412 nr_nonperturbed = idef->il[ftype].nr_nonperturbed;
4413 nr = idef->il[ftype].nr;
4414 idef_fe.il_thread_division[ftype*2+0] = nr_nonperturbed;
4415 idef_fe.il_thread_division[ftype*2+1] = nr;
4417 /* This is only to get the flop count correct */
4418 idef_fe.il[ftype].nr = nr - nr_nonperturbed;
4420 if (nr - nr_nonperturbed > 0)
4422 v = calc_one_bond(fplog, 0, ftype, &idef_fe,
4423 x, f, fshift, fr, pbc_null, g,
4424 grpp, nrnb, lambda, dvdl_dum,
4426 global_atom_index, FALSE);
4435 sfree(idef_fe.il_thread_division);