2 * This file is part of the GROMACS molecular simulation package.
4 * Copyright (c) 2012,2013,2014,2015,2016,2017,2018,2019, by the GROMACS development team, led by
5 * Mark Abraham, David van der Spoel, Berk Hess, and Erik Lindahl,
6 * and including many others, as listed in the AUTHORS file in the
7 * top-level source directory and at http://www.gromacs.org.
9 * GROMACS is free software; you can redistribute it and/or
10 * modify it under the terms of the GNU Lesser General Public License
11 * as published by the Free Software Foundation; either version 2.1
12 * of the License, or (at your option) any later version.
14 * GROMACS is distributed in the hope that it will be useful,
15 * but WITHOUT ANY WARRANTY; without even the implied warranty of
16 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
17 * Lesser General Public License for more details.
19 * You should have received a copy of the GNU Lesser General Public
20 * License along with GROMACS; if not, see
21 * http://www.gnu.org/licenses, or write to the Free Software Foundation,
22 * Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA.
24 * If you want to redistribute modifications to GROMACS, please
25 * consider that scientific software is very special. Version
26 * control is crucial - bugs must be traceable. We will be happy to
27 * consider code for inclusion in the official distribution, but
28 * derived work must not be called official GROMACS. Details are found
29 * in the README & COPYING files - if they are missing, get the
30 * official version at http://www.gromacs.org.
32 * To help us fund GROMACS development, we humbly ask that you cite
33 * the research papers on the package. Check out http://www.gromacs.org.
40 /* Unpack pointers for output */
41 real* f = out->f.data();
42 real* fshift = out->fshift.data();
45 real* Vvdw = out->VSvdw.data();
46 real* Vc = out->VSc.data();
48 real* Vvdw = out->Vvdw.data();
49 real* Vc = out->Vc.data();
53 const nbnxn_cj_t* l_cj;
56 gmx_bool do_LJ, half_LJ, do_coul;
57 int cjind0, cjind1, cjind;
61 int egps_ishift, egps_imask;
62 int egps_jshift, egps_jmask, egps_jstride;
64 real* vvdwtp[UNROLLI];
71 SimdReal ix_S0, iy_S0, iz_S0;
72 SimdReal ix_S2, iy_S2, iz_S2;
73 SimdReal fix_S0, fiy_S0, fiz_S0;
74 SimdReal fix_S2, fiy_S2, fiz_S2;
76 SimdReal diagonal_jmi_S;
77 #if UNROLLI == UNROLLJ
78 SimdBool diagonal_mask_S0, diagonal_mask_S2;
80 SimdBool diagonal_mask0_S0, diagonal_mask0_S2;
81 SimdBool diagonal_mask1_S0, diagonal_mask1_S2;
84 SimdBitMask filter_S0, filter_S2;
89 SimdReal iq_S0 = setZero();
90 SimdReal iq_S2 = setZero();
95 SimdReal hrc_3_S, moh_rc_S;
100 /* Coulomb table variables */
102 const real* tab_coul_F;
103 # if defined CALC_ENERGIES && !defined TAB_FDV0
104 const real* tab_coul_V;
107 # ifdef CALC_ENERGIES
112 #ifdef CALC_COUL_EWALD
113 SimdReal beta2_S, beta_S;
116 #if defined CALC_ENERGIES && (defined CALC_COUL_EWALD || defined CALC_COUL_TAB)
120 #if defined LJ_CUT && defined CALC_ENERGIES
121 SimdReal p6_cpot_S, p12_cpot_S;
125 SimdReal swV3_S, swV4_S, swV5_S;
126 SimdReal swF2_S, swF3_S, swF4_S;
128 #ifdef LJ_FORCE_SWITCH
130 SimdReal p6_fc2_S, p6_fc3_S;
131 SimdReal p12_fc2_S, p12_fc3_S;
132 # ifdef CALC_ENERGIES
133 SimdReal p6_vc3_S, p6_vc4_S;
134 SimdReal p12_vc3_S, p12_vc4_S;
135 SimdReal p6_6cpot_S, p12_12cpot_S;
139 real lj_ewaldcoeff2, lj_ewaldcoeff6_6;
140 SimdReal mone_S, half_S, lje_c2_S, lje_c6_6_S;
144 SimdReal hsig_i_S0, seps_i_S0;
145 SimdReal hsig_i_S2, seps_i_S2;
148 alignas(GMX_SIMD_ALIGNMENT) real pvdw_c6[2 * UNROLLI * UNROLLJ];
149 real* pvdw_c12 = pvdw_c6 + UNROLLI * UNROLLJ;
151 #endif /* LJ_COMB_LB */
155 #ifdef VDW_CUTOFF_CHECK
165 const nbnxn_atomdata_t::Params& nbatParams = nbat->params();
167 #if defined LJ_COMB_GEOM || defined LJ_COMB_LB || defined LJ_EWALD_GEOM
168 const real* gmx_restrict ljc = nbatParams.lj_comb.data();
170 #if !(defined LJ_COMB_GEOM || defined LJ_COMB_LB || defined FIX_LJ_C)
171 /* No combination rule used */
172 const real* gmx_restrict nbfp_ptr = nbatParams.nbfp_aligned.data();
173 const int* gmx_restrict type = nbatParams.type.data();
176 /* Load j-i for the first i */
177 diagonal_jmi_S = load<SimdReal>(nbat->simdMasks.diagonal_2xnn_j_minus_i.data());
178 /* Generate all the diagonal masks as comparison results */
179 #if UNROLLI == UNROLLJ
180 diagonal_mask_S0 = (zero_S < diagonal_jmi_S);
181 diagonal_jmi_S = diagonal_jmi_S - one_S;
182 diagonal_jmi_S = diagonal_jmi_S - one_S;
183 diagonal_mask_S2 = (zero_S < diagonal_jmi_S);
185 # if 2 * UNROLLI == UNROLLJ
186 diagonal_mask0_S0 = (zero_S < diagonal_jmi_S);
187 diagonal_jmi_S = diagonal_jmi_S - one_S;
188 diagonal_jmi_S = diagonal_jmi_S - one_S;
189 diagonal_mask0_S2 = (zero_S < diagonal_jmi_S);
190 diagonal_jmi_S = diagonal_jmi_S - one_S;
191 diagonal_jmi_S = diagonal_jmi_S - one_S;
192 diagonal_mask1_S0 = (zero_S < diagonal_jmi_S);
193 diagonal_jmi_S = diagonal_jmi_S - one_S;
194 diagonal_jmi_S = diagonal_jmi_S - one_S;
195 diagonal_mask1_S2 = (zero_S < diagonal_jmi_S);
199 /* Load masks for topology exclusion masking. filter_stride is
200 static const, so the conditional will be optimized away. */
201 #if GMX_DOUBLE && !GMX_SIMD_HAVE_INT32_LOGICAL
202 const std::uint64_t* gmx_restrict exclusion_filter = nbat->simdMasks.exclusion_filter64.data();
204 const std::uint32_t* gmx_restrict exclusion_filter = nbat->simdMasks.exclusion_filter.data();
207 /* Here we cast the exclusion filters from unsigned * to int * or real *.
208 * Since we only check bits, the actual value they represent does not
209 * matter, as long as both filter and mask data are treated the same way.
211 #if GMX_SIMD_HAVE_INT32_LOGICAL
212 filter_S0 = load<SimdBitMask>(reinterpret_cast<const int*>(exclusion_filter + 0 * UNROLLJ));
213 filter_S2 = load<SimdBitMask>(reinterpret_cast<const int*>(exclusion_filter + 2 * UNROLLJ));
215 filter_S0 = load<SimdBitMask>(reinterpret_cast<const real*>(exclusion_filter + 0 * UNROLLJ));
216 filter_S2 = load<SimdBitMask>(reinterpret_cast<const real*>(exclusion_filter + 2 * UNROLLJ));
220 /* Reaction-field constants */
221 mrc_3_S = SimdReal(-2 * ic->k_rf);
222 # ifdef CALC_ENERGIES
223 hrc_3_S = SimdReal(ic->k_rf);
224 moh_rc_S = SimdReal(-ic->c_rf);
230 invtsp_S = SimdReal(ic->coulombEwaldTables->scale);
231 # ifdef CALC_ENERGIES
232 mhalfsp_S = SimdReal(-0.5_real / ic->coulombEwaldTables->scale);
236 tab_coul_F = ic->coulombEwaldTables->tableFDV0.data();
238 tab_coul_F = ic->coulombEwaldTables->tableF.data();
239 # ifdef CALC_ENERGIES
240 tab_coul_V = ic->coulombEwaldTables->tableV.data();
243 #endif /* CALC_COUL_TAB */
245 #ifdef CALC_COUL_EWALD
246 beta2_S = SimdReal(ic->ewaldcoeff_q * ic->ewaldcoeff_q);
247 beta_S = SimdReal(ic->ewaldcoeff_q);
250 #if (defined CALC_COUL_TAB || defined CALC_COUL_EWALD) && defined CALC_ENERGIES
251 sh_ewald_S = SimdReal(ic->sh_ewald);
254 /* LJ function constants */
255 #if defined CALC_ENERGIES || defined LJ_POT_SWITCH
256 SimdReal sixth_S = SimdReal(1.0 / 6.0);
257 SimdReal twelveth_S = SimdReal(1.0 / 12.0);
260 #if defined LJ_CUT && defined CALC_ENERGIES
261 /* We shift the potential by cpot, which can be zero */
262 p6_cpot_S = SimdReal(ic->dispersion_shift.cpot);
263 p12_cpot_S = SimdReal(ic->repulsion_shift.cpot);
266 rswitch_S = SimdReal(ic->rvdw_switch);
267 swV3_S = SimdReal(ic->vdw_switch.c3);
268 swV4_S = SimdReal(ic->vdw_switch.c4);
269 swV5_S = SimdReal(ic->vdw_switch.c5);
270 swF2_S = SimdReal(3 * ic->vdw_switch.c3);
271 swF3_S = SimdReal(4 * ic->vdw_switch.c4);
272 swF4_S = SimdReal(5 * ic->vdw_switch.c5);
274 #ifdef LJ_FORCE_SWITCH
275 rswitch_S = SimdReal(ic->rvdw_switch);
276 p6_fc2_S = SimdReal(ic->dispersion_shift.c2);
277 p6_fc3_S = SimdReal(ic->dispersion_shift.c3);
278 p12_fc2_S = SimdReal(ic->repulsion_shift.c2);
279 p12_fc3_S = SimdReal(ic->repulsion_shift.c3);
280 # ifdef CALC_ENERGIES
282 SimdReal mthird_S = SimdReal(-1.0 / 3.0);
283 SimdReal mfourth_S = SimdReal(-1.0 / 4.0);
285 p6_vc3_S = mthird_S * p6_fc2_S;
286 p6_vc4_S = mfourth_S * p6_fc3_S;
287 p6_6cpot_S = SimdReal(ic->dispersion_shift.cpot / 6);
288 p12_vc3_S = mthird_S * p12_fc2_S;
289 p12_vc4_S = mfourth_S * p12_fc3_S;
290 p12_12cpot_S = SimdReal(ic->repulsion_shift.cpot / 12);
295 mone_S = SimdReal(-1.0);
296 half_S = SimdReal(0.5);
297 lj_ewaldcoeff2 = ic->ewaldcoeff_lj * ic->ewaldcoeff_lj;
298 lj_ewaldcoeff6_6 = lj_ewaldcoeff2 * lj_ewaldcoeff2 * lj_ewaldcoeff2 / 6;
299 lje_c2_S = SimdReal(lj_ewaldcoeff2);
300 lje_c6_6_S = SimdReal(lj_ewaldcoeff6_6);
301 # ifdef CALC_ENERGIES
302 /* Determine the grid potential at the cut-off */
303 SimdReal lje_vc_S = SimdReal(ic->sh_lj_ewald);
307 /* The kernel either supports rcoulomb = rvdw or rcoulomb >= rvdw */
308 rc2_S = SimdReal(ic->rcoulomb * ic->rcoulomb);
309 #ifdef VDW_CUTOFF_CHECK
310 rcvdw2_S = SimdReal(ic->rvdw * ic->rvdw);
313 minRsq_S = SimdReal(NBNXN_MIN_RSQ);
315 const real* gmx_restrict q = nbatParams.q.data();
316 const real facel = ic->epsfac;
317 const real* gmx_restrict shiftvec = shift_vec[0];
318 const real* gmx_restrict x = nbat->x().data();
322 for (jp = 0; jp < UNROLLJ; jp++)
324 pvdw_c6[0 * UNROLLJ + jp] = nbat->nbfp[0 * 2];
325 pvdw_c6[1 * UNROLLJ + jp] = nbat->nbfp[0 * 2];
326 pvdw_c6[2 * UNROLLJ + jp] = nbat->nbfp[0 * 2];
327 pvdw_c6[3 * UNROLLJ + jp] = nbat->nbfp[0 * 2];
329 pvdw_c12[0 * UNROLLJ + jp] = nbat->nbfp[0 * 2 + 1];
330 pvdw_c12[1 * UNROLLJ + jp] = nbat->nbfp[0 * 2 + 1];
331 pvdw_c12[2 * UNROLLJ + jp] = nbat->nbfp[0 * 2 + 1];
332 pvdw_c12[3 * UNROLLJ + jp] = nbat->nbfp[0 * 2 + 1];
334 SimdReal c6_S0 = load<SimdReal>(pvdw_c6 + 0 * UNROLLJ);
335 SimdReal c6_S1 = load<SimdReal>(pvdw_c6 + 1 * UNROLLJ);
336 SimdReal c6_S2 = load<SimdReal>(pvdw_c6 + 2 * UNROLLJ);
337 SimdReal c6_S3 = load<SimdReal>(pvdw_c6 + 3 * UNROLLJ);
339 SimdReal c12_S0 = load<SimdReal>(pvdw_c12 + 0 * UNROLLJ);
340 SimdReal c12_S1 = load<SimdReal>(pvdw_c12 + 1 * UNROLLJ);
341 SimdReal c12_S2 = load<SimdReal>(pvdw_c12 + 2 * UNROLLJ);
342 SimdReal c12_S3 = load<SimdReal>(pvdw_c12 + 3 * UNROLLJ);
343 #endif /* FIX_LJ_C */
346 egps_ishift = nbatParams.neg_2log;
347 egps_imask = (1 << egps_ishift) - 1;
348 egps_jshift = 2 * nbatParams.neg_2log;
349 egps_jmask = (1 << egps_jshift) - 1;
350 egps_jstride = (UNROLLJ >> 1) * UNROLLJ;
351 /* Major division is over i-particle energy groups, determine the stride */
352 Vstride_i = nbatParams.nenergrp * (1 << nbatParams.neg_2log) * egps_jstride;
355 l_cj = nbl->cj.data();
358 for (const nbnxn_ci_t& ciEntry : nbl->ci)
360 ish = (ciEntry.shift & NBNXN_CI_SHIFT);
362 cjind0 = ciEntry.cj_ind_start;
363 cjind1 = ciEntry.cj_ind_end;
365 ci_sh = (ish == CENTRAL ? ci : -1);
367 shX_S = SimdReal(shiftvec[ish3]);
368 shY_S = SimdReal(shiftvec[ish3 + 1]);
369 shZ_S = SimdReal(shiftvec[ish3 + 2]);
372 int sci = ci * STRIDE;
373 int scix = sci * DIM;
374 # if defined LJ_COMB_LB || defined LJ_COMB_GEOM || defined LJ_EWALD_GEOM
378 int sci = (ci >> 1) * STRIDE;
379 int scix = sci * DIM + (ci & 1) * (STRIDE >> 1);
380 # if defined LJ_COMB_LB || defined LJ_COMB_GEOM || defined LJ_EWALD_GEOM
381 int sci2 = sci * 2 + (ci & 1) * (STRIDE >> 1);
383 sci += (ci & 1) * (STRIDE >> 1);
386 /* We have 5 LJ/C combinations, but use only three inner loops,
387 * as the other combinations are unlikely and/or not much faster:
388 * inner half-LJ + C for half-LJ + C / no-LJ + C
389 * inner LJ + C for full-LJ + C
390 * inner LJ for full-LJ + no-C / half-LJ + no-C
392 do_LJ = ((ciEntry.shift & NBNXN_CI_DO_LJ(0)) != 0);
393 do_coul = ((ciEntry.shift & NBNXN_CI_DO_COUL(0)) != 0);
394 half_LJ = (((ciEntry.shift & NBNXN_CI_HALF_LJ(0)) != 0) || !do_LJ) && do_coul;
397 egps_i = nbatParams.energrp[ci];
401 for (ia = 0; ia < UNROLLI; ia++)
403 egp_ia = (egps_i >> (ia * egps_ishift)) & egps_imask;
404 vvdwtp[ia] = Vvdw + egp_ia * Vstride_i;
405 vctp[ia] = Vc + egp_ia * Vstride_i;
411 # ifdef LJ_EWALD_GEOM
412 gmx_bool do_self = TRUE;
414 gmx_bool do_self = do_coul;
417 if (do_self && l_cj[ciEntry.cj_ind_start].cj == ci_sh)
420 if (do_self && l_cj[ciEntry.cj_ind_start].cj == (ci_sh >> 1))
429 Vc_sub_self = 0.5 * ic->c_rf;
431 # ifdef CALC_COUL_TAB
433 Vc_sub_self = 0.5 * tab_coul_F[2];
435 Vc_sub_self = 0.5 * tab_coul_V[0];
438 # ifdef CALC_COUL_EWALD
440 Vc_sub_self = 0.5 * ic->ewaldcoeff_q * M_2_SQRTPI;
443 for (ia = 0; ia < UNROLLI; ia++)
448 # ifdef ENERGY_GROUPS
449 vctp[ia][((egps_i >> (ia * egps_ishift)) & egps_imask) * egps_jstride]
453 -= facel * qi * qi * Vc_sub_self;
457 # ifdef LJ_EWALD_GEOM
461 for (ia = 0; ia < UNROLLI; ia++)
465 c6_i = nbatParams.nbfp[nbatParams.type[sci + ia] * (nbatParams.numTypes + 1) * 2]
467 # ifdef ENERGY_GROUPS
468 vvdwtp[ia][((egps_i >> (ia * egps_ishift)) & egps_imask) * egps_jstride]
472 += 0.5 * c6_i * lj_ewaldcoeff6_6;
475 # endif /* LJ_EWALD */
479 /* Load i atom data */
480 int sciy = scix + STRIDE;
481 int sciz = sciy + STRIDE;
482 ix_S0 = loadU1DualHsimd(x + scix);
483 ix_S2 = loadU1DualHsimd(x + scix + 2);
484 iy_S0 = loadU1DualHsimd(x + sciy);
485 iy_S2 = loadU1DualHsimd(x + sciy + 2);
486 iz_S0 = loadU1DualHsimd(x + sciz);
487 iz_S2 = loadU1DualHsimd(x + sciz + 2);
488 ix_S0 = ix_S0 + shX_S;
489 ix_S2 = ix_S2 + shX_S;
490 iy_S0 = iy_S0 + shY_S;
491 iy_S2 = iy_S2 + shY_S;
492 iz_S0 = iz_S0 + shZ_S;
493 iz_S2 = iz_S2 + shZ_S;
499 facel_S = SimdReal(facel);
501 iq_S0 = loadU1DualHsimd(q + sci);
502 iq_S2 = loadU1DualHsimd(q + sci + 2);
503 iq_S0 = facel_S * iq_S0;
504 iq_S2 = facel_S * iq_S2;
508 hsig_i_S0 = loadU1DualHsimd(ljc + sci2);
509 hsig_i_S2 = loadU1DualHsimd(ljc + sci2 + 2);
510 seps_i_S0 = loadU1DualHsimd(ljc + sci2 + STRIDE);
511 seps_i_S2 = loadU1DualHsimd(ljc + sci2 + STRIDE + 2);
514 SimdReal c6s_S0, c12s_S0;
515 SimdReal c6s_S2, c12s_S2;
517 c6s_S0 = loadU1DualHsimd(ljc + sci2);
521 c6s_S2 = loadU1DualHsimd(ljc + sci2 + 2);
523 c12s_S0 = loadU1DualHsimd(ljc + sci2 + STRIDE);
526 c12s_S2 = loadU1DualHsimd(ljc + sci2 + STRIDE + 2);
528 # elif !defined LJ_COMB_LB && !defined FIX_LJ_C
529 const int numTypes = nbatParams.numTypes;
530 const real* nbfp0 = nbfp_ptr + type[sci] * numTypes * c_simdBestPairAlignment;
531 const real* nbfp1 = nbfp_ptr + type[sci + 1] * numTypes * c_simdBestPairAlignment;
532 const real *nbfp2 = nullptr, *nbfp3 = nullptr;
535 nbfp2 = nbfp_ptr + type[sci + 2] * numTypes * c_simdBestPairAlignment;
536 nbfp3 = nbfp_ptr + type[sci + 3] * numTypes * c_simdBestPairAlignment;
541 /* We need the geometrically combined C6 for the PME grid correction */
542 SimdReal c6s_S0, c6s_S2;
543 c6s_S0 = loadU1DualHsimd(ljc + sci2);
546 c6s_S2 = loadU1DualHsimd(ljc + sci2 + 2);
550 /* Zero the potential energy for this list */
552 SimdReal Vvdwtot_S = setZero();
553 SimdReal vctot_S = setZero();
556 /* Clear i atom forces */
566 /* Currently all kernels use (at least half) LJ */
570 /* Coulomb: all i-atoms, LJ: first half i-atoms */
574 while (cjind < cjind1 && nbl->cj[cjind].excl != NBNXN_INTERACTION_MASK_ALL)
576 #include "kernel_inner.h"
580 for (; (cjind < cjind1); cjind++)
582 #include "kernel_inner.h"
589 /* Coulomb: all i-atoms, LJ: all i-atoms */
592 while (cjind < cjind1 && nbl->cj[cjind].excl != NBNXN_INTERACTION_MASK_ALL)
594 #include "kernel_inner.h"
598 for (; (cjind < cjind1); cjind++)
600 #include "kernel_inner.h"
606 /* Coulomb: none, LJ: all i-atoms */
608 while (cjind < cjind1 && nbl->cj[cjind].excl != NBNXN_INTERACTION_MASK_ALL)
610 #include "kernel_inner.h"
614 for (; (cjind < cjind1); cjind++)
616 #include "kernel_inner.h"
620 ninner += cjind1 - cjind0;
622 /* Add accumulated i-forces to the force array */
623 real fShiftX = reduceIncr4ReturnSumHsimd(f + scix, fix_S0, fix_S2);
624 real fShiftY = reduceIncr4ReturnSumHsimd(f + sciy, fiy_S0, fiy_S2);
625 real fShiftZ = reduceIncr4ReturnSumHsimd(f + sciz, fiz_S0, fiz_S2);
627 #ifdef CALC_SHIFTFORCES
628 fshift[ish3 + 0] += fShiftX;
629 fshift[ish3 + 1] += fShiftY;
630 fshift[ish3 + 2] += fShiftZ;
636 *Vc += reduce(vctot_S);
638 *Vvdw += reduce(Vvdwtot_S);
641 /* Outer loop uses 6 flops/iteration */
645 printf("atom pairs %d\n", npair);