2 * This file is part of the GROMACS molecular simulation package.
4 * Copyright (c) 2012,2013,2014, 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.
38 const nbnxn_ci_t *nbln;
39 const nbnxn_cj_t *l_cj;
44 const real *nbfp0, *nbfp1, *nbfp2 = NULL, *nbfp3 = NULL;
49 gmx_bool do_LJ, half_LJ, do_coul, do_self;
50 int sci, scix, sciy, sciz, sci2;
51 int cjind0, cjind1, cjind;
56 int egps_ishift, egps_imask;
57 int egps_jshift, egps_jmask, egps_jstride;
59 real *vvdwtp[UNROLLI];
63 gmx_simd_real_t shX_S;
64 gmx_simd_real_t shY_S;
65 gmx_simd_real_t shZ_S;
66 gmx_simd_real_t ix_S0, iy_S0, iz_S0;
67 gmx_simd_real_t ix_S1, iy_S1, iz_S1;
68 gmx_simd_real_t ix_S2, iy_S2, iz_S2;
69 gmx_simd_real_t ix_S3, iy_S3, iz_S3;
70 gmx_simd_real_t fix_S0, fiy_S0, fiz_S0;
71 gmx_simd_real_t fix_S1, fiy_S1, fiz_S1;
72 gmx_simd_real_t fix_S2, fiy_S2, fiz_S2;
73 gmx_simd_real_t fix_S3, fiy_S3, fiz_S3;
75 /* We use an i-force SIMD register width of 4 */
76 gmx_simd4_real_t fix_S, fiy_S, fiz_S;
78 /* We use an i-force SIMD register width of 2 */
79 gmx_simd_real_t fix0_S, fiy0_S, fiz0_S;
80 gmx_simd_real_t fix2_S, fiy2_S, fiz2_S;
83 gmx_simd_real_t diagonal_jmi_S;
84 #if UNROLLI == UNROLLJ
85 gmx_simd_bool_t diagonal_mask_S0, diagonal_mask_S1, diagonal_mask_S2, diagonal_mask_S3;
87 gmx_simd_bool_t diagonal_mask0_S0, diagonal_mask0_S1, diagonal_mask0_S2, diagonal_mask0_S3;
88 gmx_simd_bool_t diagonal_mask1_S0, diagonal_mask1_S1, diagonal_mask1_S2, diagonal_mask1_S3;
91 unsigned *exclusion_filter;
92 gmx_exclfilter filter_S0, filter_S1, filter_S2, filter_S3;
94 gmx_simd_real_t zero_S = gmx_simd_set1_r(0.0);
96 gmx_simd_real_t one_S = gmx_simd_set1_r(1.0);
97 gmx_simd_real_t iq_S0 = gmx_simd_setzero_r();
98 gmx_simd_real_t iq_S1 = gmx_simd_setzero_r();
99 gmx_simd_real_t iq_S2 = gmx_simd_setzero_r();
100 gmx_simd_real_t iq_S3 = gmx_simd_setzero_r();
103 gmx_simd_real_t mrc_3_S;
105 gmx_simd_real_t hrc_3_S, moh_rc_S;
110 /* Coulomb table variables */
111 gmx_simd_real_t invtsp_S;
112 const real * tab_coul_F;
114 const real * tab_coul_V;
116 /* Thread-local working buffers for force and potential lookups */
117 int ti0_array[2*GMX_SIMD_REAL_WIDTH], *ti0 = NULL;
118 int ti1_array[2*GMX_SIMD_REAL_WIDTH], *ti1 = NULL;
119 int ti2_array[2*GMX_SIMD_REAL_WIDTH], *ti2 = NULL;
120 int ti3_array[2*GMX_SIMD_REAL_WIDTH], *ti3 = NULL;
122 gmx_simd_real_t mhalfsp_S;
126 #ifdef CALC_COUL_EWALD
127 gmx_simd_real_t beta2_S, beta_S;
130 #if defined CALC_ENERGIES && (defined CALC_COUL_EWALD || defined CALC_COUL_TAB)
131 gmx_simd_real_t sh_ewald_S;
134 #if defined LJ_CUT && defined CALC_ENERGIES
135 gmx_simd_real_t p6_cpot_S, p12_cpot_S;
138 gmx_simd_real_t rswitch_S;
139 gmx_simd_real_t swV3_S, swV4_S, swV5_S;
140 gmx_simd_real_t swF2_S, swF3_S, swF4_S;
142 #ifdef LJ_FORCE_SWITCH
143 gmx_simd_real_t rswitch_S;
144 gmx_simd_real_t p6_fc2_S, p6_fc3_S;
145 gmx_simd_real_t p12_fc2_S, p12_fc3_S;
147 gmx_simd_real_t p6_vc3_S, p6_vc4_S;
148 gmx_simd_real_t p12_vc3_S, p12_vc4_S;
149 gmx_simd_real_t p6_6cpot_S, p12_12cpot_S;
153 real lj_ewaldcoeff2, lj_ewaldcoeff6_6;
154 gmx_simd_real_t mone_S, half_S, lje_c2_S, lje_c6_6_S, lje_vc_S;
160 gmx_simd_real_t hsig_i_S0, seps_i_S0;
161 gmx_simd_real_t hsig_i_S1, seps_i_S1;
162 gmx_simd_real_t hsig_i_S2, seps_i_S2;
163 gmx_simd_real_t hsig_i_S3, seps_i_S3;
166 real pvdw_array[2*UNROLLI*UNROLLJ+3];
167 real *pvdw_c6, *pvdw_c12;
168 gmx_simd_real_t c6_S0, c12_S0;
169 gmx_simd_real_t c6_S1, c12_S1;
170 gmx_simd_real_t c6_S2, c12_S2;
171 gmx_simd_real_t c6_S3, c12_S3;
174 #if defined LJ_COMB_GEOM || defined LJ_EWALD_GEOM
177 gmx_simd_real_t c6s_S0, c12s_S0;
178 gmx_simd_real_t c6s_S1, c12s_S1;
179 gmx_simd_real_t c6s_S2 = gmx_simd_setzero_r();
180 gmx_simd_real_t c12s_S2 = gmx_simd_setzero_r();
181 gmx_simd_real_t c6s_S3 = gmx_simd_setzero_r();
182 gmx_simd_real_t c12s_S3 = gmx_simd_setzero_r();
184 #endif /* LJ_COMB_LB */
186 gmx_simd_real_t vctot_S, Vvdwtot_S;
187 gmx_simd_real_t sixth_S, twelveth_S;
189 gmx_simd_real_t avoid_sing_S;
190 gmx_simd_real_t rc2_S;
191 #ifdef VDW_CUTOFF_CHECK
192 gmx_simd_real_t rcvdw2_S;
201 #if defined LJ_COMB_GEOM || defined LJ_COMB_LB || defined LJ_EWALD_GEOM
204 #if !(defined LJ_COMB_GEOM || defined LJ_COMB_LB)
205 /* No combination rule used */
206 nbfp_ptr = (4 == nbfp_stride) ? nbat->nbfp_s4 : nbat->nbfp;
209 /* Load j-i for the first i */
210 diagonal_jmi_S = gmx_simd_load_r(nbat->simd_4xn_diagonal_j_minus_i);
211 /* Generate all the diagonal masks as comparison results */
212 #if UNROLLI == UNROLLJ
213 diagonal_mask_S0 = gmx_simd_cmplt_r(zero_S, diagonal_jmi_S);
214 diagonal_jmi_S = gmx_simd_sub_r(diagonal_jmi_S, one_S);
215 diagonal_mask_S1 = gmx_simd_cmplt_r(zero_S, diagonal_jmi_S);
216 diagonal_jmi_S = gmx_simd_sub_r(diagonal_jmi_S, one_S);
217 diagonal_mask_S2 = gmx_simd_cmplt_r(zero_S, diagonal_jmi_S);
218 diagonal_jmi_S = gmx_simd_sub_r(diagonal_jmi_S, one_S);
219 diagonal_mask_S3 = gmx_simd_cmplt_r(zero_S, diagonal_jmi_S);
221 #if UNROLLI == 2*UNROLLJ || 2*UNROLLI == UNROLLJ
222 diagonal_mask0_S0 = gmx_simd_cmplt_r(zero_S, diagonal_jmi_S);
223 diagonal_jmi_S = gmx_simd_sub_r(diagonal_jmi_S, one_S);
224 diagonal_mask0_S1 = gmx_simd_cmplt_r(zero_S, diagonal_jmi_S);
225 diagonal_jmi_S = gmx_simd_sub_r(diagonal_jmi_S, one_S);
226 diagonal_mask0_S2 = gmx_simd_cmplt_r(zero_S, diagonal_jmi_S);
227 diagonal_jmi_S = gmx_simd_sub_r(diagonal_jmi_S, one_S);
228 diagonal_mask0_S3 = gmx_simd_cmplt_r(zero_S, diagonal_jmi_S);
229 diagonal_jmi_S = gmx_simd_sub_r(diagonal_jmi_S, one_S);
231 #if UNROLLI == 2*UNROLLJ
232 /* Load j-i for the second half of the j-cluster */
233 diagonal_jmi_S = gmx_simd_load_r(nbat->simd_4xn_diagonal_j_minus_i + UNROLLJ);
236 diagonal_mask1_S0 = gmx_simd_cmplt_r(zero_S, diagonal_jmi_S);
237 diagonal_jmi_S = gmx_simd_sub_r(diagonal_jmi_S, one_S);
238 diagonal_mask1_S1 = gmx_simd_cmplt_r(zero_S, diagonal_jmi_S);
239 diagonal_jmi_S = gmx_simd_sub_r(diagonal_jmi_S, one_S);
240 diagonal_mask1_S2 = gmx_simd_cmplt_r(zero_S, diagonal_jmi_S);
241 diagonal_jmi_S = gmx_simd_sub_r(diagonal_jmi_S, one_S);
242 diagonal_mask1_S3 = gmx_simd_cmplt_r(zero_S, diagonal_jmi_S);
246 /* Load masks for topology exclusion masking. filter_stride is
247 static const, so the conditional will be optimized away. */
248 if (1 == filter_stride)
250 exclusion_filter = nbat->simd_exclusion_filter1;
252 else /* (2 == filter_stride) */
254 exclusion_filter = nbat->simd_exclusion_filter2;
257 /* Here we cast the exclusion filters from unsigned * to int * or real *.
258 * Since we only check bits, the actual value they represent does not
259 * matter, as long as both filter and mask data are treated the same way.
261 filter_S0 = gmx_load_exclusion_filter(exclusion_filter + 0*UNROLLJ*filter_stride);
262 filter_S1 = gmx_load_exclusion_filter(exclusion_filter + 1*UNROLLJ*filter_stride);
263 filter_S2 = gmx_load_exclusion_filter(exclusion_filter + 2*UNROLLJ*filter_stride);
264 filter_S3 = gmx_load_exclusion_filter(exclusion_filter + 3*UNROLLJ*filter_stride);
267 /* Reaction-field constants */
268 mrc_3_S = gmx_simd_set1_r(-2*ic->k_rf);
270 hrc_3_S = gmx_simd_set1_r(ic->k_rf);
271 moh_rc_S = gmx_simd_set1_r(-ic->c_rf);
276 /* Generate aligned table index pointers */
277 ti0 = prepare_table_load_buffer(ti0_array);
278 ti1 = prepare_table_load_buffer(ti1_array);
279 ti2 = prepare_table_load_buffer(ti2_array);
280 ti3 = prepare_table_load_buffer(ti3_array);
282 invtsp_S = gmx_simd_set1_r(ic->tabq_scale);
284 mhalfsp_S = gmx_simd_set1_r(-0.5/ic->tabq_scale);
288 tab_coul_F = ic->tabq_coul_FDV0;
290 tab_coul_F = ic->tabq_coul_F;
291 tab_coul_V = ic->tabq_coul_V;
293 #endif /* CALC_COUL_TAB */
295 #ifdef CALC_COUL_EWALD
296 beta2_S = gmx_simd_set1_r(ic->ewaldcoeff_q*ic->ewaldcoeff_q);
297 beta_S = gmx_simd_set1_r(ic->ewaldcoeff_q);
300 #if (defined CALC_COUL_TAB || defined CALC_COUL_EWALD) && defined CALC_ENERGIES
301 sh_ewald_S = gmx_simd_set1_r(ic->sh_ewald);
304 /* LJ function constants */
305 #if defined CALC_ENERGIES || defined LJ_POT_SWITCH
306 sixth_S = gmx_simd_set1_r(1.0/6.0);
307 twelveth_S = gmx_simd_set1_r(1.0/12.0);
310 #if defined LJ_CUT && defined CALC_ENERGIES
311 /* We shift the potential by cpot, which can be zero */
312 p6_cpot_S = gmx_simd_set1_r(ic->dispersion_shift.cpot);
313 p12_cpot_S = gmx_simd_set1_r(ic->repulsion_shift.cpot);
316 rswitch_S = gmx_simd_set1_r(ic->rvdw_switch);
317 swV3_S = gmx_simd_set1_r(ic->vdw_switch.c3);
318 swV4_S = gmx_simd_set1_r(ic->vdw_switch.c4);
319 swV5_S = gmx_simd_set1_r(ic->vdw_switch.c5);
320 swF2_S = gmx_simd_set1_r(3*ic->vdw_switch.c3);
321 swF3_S = gmx_simd_set1_r(4*ic->vdw_switch.c4);
322 swF4_S = gmx_simd_set1_r(5*ic->vdw_switch.c5);
324 #ifdef LJ_FORCE_SWITCH
325 rswitch_S = gmx_simd_set1_r(ic->rvdw_switch);
326 p6_fc2_S = gmx_simd_set1_r(ic->dispersion_shift.c2);
327 p6_fc3_S = gmx_simd_set1_r(ic->dispersion_shift.c3);
328 p12_fc2_S = gmx_simd_set1_r(ic->repulsion_shift.c2);
329 p12_fc3_S = gmx_simd_set1_r(ic->repulsion_shift.c3);
332 gmx_simd_real_t mthird_S = gmx_simd_set1_r(-1.0/3.0);
333 gmx_simd_real_t mfourth_S = gmx_simd_set1_r(-1.0/4.0);
335 p6_vc3_S = gmx_simd_mul_r(mthird_S, p6_fc2_S);
336 p6_vc4_S = gmx_simd_mul_r(mfourth_S, p6_fc3_S);
337 p6_6cpot_S = gmx_simd_set1_r(ic->dispersion_shift.cpot/6);
338 p12_vc3_S = gmx_simd_mul_r(mthird_S, p12_fc2_S);
339 p12_vc4_S = gmx_simd_mul_r(mfourth_S, p12_fc3_S);
340 p12_12cpot_S = gmx_simd_set1_r(ic->repulsion_shift.cpot/12);
345 mone_S = gmx_simd_set1_r(-1.0);
346 half_S = gmx_simd_set1_r(0.5);
347 lj_ewaldcoeff2 = ic->ewaldcoeff_lj*ic->ewaldcoeff_lj;
348 lj_ewaldcoeff6_6 = lj_ewaldcoeff2*lj_ewaldcoeff2*lj_ewaldcoeff2/6;
349 lje_c2_S = gmx_simd_set1_r(lj_ewaldcoeff2);
350 lje_c6_6_S = gmx_simd_set1_r(lj_ewaldcoeff6_6);
351 /* Determine the grid potential at the cut-off */
352 lje_vc_S = gmx_simd_set1_r(ic->sh_lj_ewald);
355 /* The kernel either supports rcoulomb = rvdw or rcoulomb >= rvdw */
356 rc2_S = gmx_simd_set1_r(ic->rcoulomb*ic->rcoulomb);
357 #ifdef VDW_CUTOFF_CHECK
358 rcvdw2_S = gmx_simd_set1_r(ic->rvdw*ic->rvdw);
361 avoid_sing_S = gmx_simd_set1_r(NBNXN_AVOID_SING_R2_INC);
366 shiftvec = shift_vec[0];
370 pvdw_c6 = gmx_simd_align_real(pvdw_array);
371 pvdw_c12 = pvdw_c6 + UNROLLI*UNROLLJ;
373 for (jp = 0; jp < UNROLLJ; jp++)
375 pvdw_c6 [0*UNROLLJ+jp] = nbat->nbfp[0*2];
376 pvdw_c6 [1*UNROLLJ+jp] = nbat->nbfp[0*2];
377 pvdw_c6 [2*UNROLLJ+jp] = nbat->nbfp[0*2];
378 pvdw_c6 [3*UNROLLJ+jp] = nbat->nbfp[0*2];
380 pvdw_c12[0*UNROLLJ+jp] = nbat->nbfp[0*2+1];
381 pvdw_c12[1*UNROLLJ+jp] = nbat->nbfp[0*2+1];
382 pvdw_c12[2*UNROLLJ+jp] = nbat->nbfp[0*2+1];
383 pvdw_c12[3*UNROLLJ+jp] = nbat->nbfp[0*2+1];
385 c6_S0 = gmx_simd_load_r(pvdw_c6 +0*UNROLLJ);
386 c6_S1 = gmx_simd_load_r(pvdw_c6 +1*UNROLLJ);
387 c6_S2 = gmx_simd_load_r(pvdw_c6 +2*UNROLLJ);
388 c6_S3 = gmx_simd_load_r(pvdw_c6 +3*UNROLLJ);
390 c12_S0 = gmx_simd_load_r(pvdw_c12+0*UNROLLJ);
391 c12_S1 = gmx_simd_load_r(pvdw_c12+1*UNROLLJ);
392 c12_S2 = gmx_simd_load_r(pvdw_c12+2*UNROLLJ);
393 c12_S3 = gmx_simd_load_r(pvdw_c12+3*UNROLLJ);
394 #endif /* FIX_LJ_C */
397 egps_ishift = nbat->neg_2log;
398 egps_imask = (1<<egps_ishift) - 1;
399 egps_jshift = 2*nbat->neg_2log;
400 egps_jmask = (1<<egps_jshift) - 1;
401 egps_jstride = (UNROLLJ>>1)*UNROLLJ;
402 /* Major division is over i-particle energy groups, determine the stride */
403 Vstride_i = nbat->nenergrp*(1<<nbat->neg_2log)*egps_jstride;
409 for (n = 0; n < nbl->nci; n++)
413 ish = (nbln->shift & NBNXN_CI_SHIFT);
415 cjind0 = nbln->cj_ind_start;
416 cjind1 = nbln->cj_ind_end;
418 ci_sh = (ish == CENTRAL ? ci : -1);
420 shX_S = gmx_simd_load1_r(shiftvec+ish3);
421 shY_S = gmx_simd_load1_r(shiftvec+ish3+1);
422 shZ_S = gmx_simd_load1_r(shiftvec+ish3+2);
429 sci = (ci>>1)*STRIDE;
430 scix = sci*DIM + (ci & 1)*(STRIDE>>1);
431 sci2 = sci*2 + (ci & 1)*(STRIDE>>1);
432 sci += (ci & 1)*(STRIDE>>1);
435 /* We have 5 LJ/C combinations, but use only three inner loops,
436 * as the other combinations are unlikely and/or not much faster:
437 * inner half-LJ + C for half-LJ + C / no-LJ + C
438 * inner LJ + C for full-LJ + C
439 * inner LJ for full-LJ + no-C / half-LJ + no-C
441 do_LJ = (nbln->shift & NBNXN_CI_DO_LJ(0));
442 do_coul = (nbln->shift & NBNXN_CI_DO_COUL(0));
443 half_LJ = ((nbln->shift & NBNXN_CI_HALF_LJ(0)) || !do_LJ) && do_coul;
451 egps_i = nbat->energrp[ci];
455 for (ia = 0; ia < UNROLLI; ia++)
457 egp_ia = (egps_i >> (ia*egps_ishift)) & egps_imask;
458 vvdwtp[ia] = Vvdw + egp_ia*Vstride_i;
459 vctp[ia] = Vc + egp_ia*Vstride_i;
466 if (do_self && l_cj[nbln->cj_ind_start].cj == ci_sh)
469 if (do_self && l_cj[nbln->cj_ind_start].cj == (ci_sh<<1))
472 if (do_self && l_cj[nbln->cj_ind_start].cj == (ci_sh>>1))
481 Vc_sub_self = 0.5*ic->c_rf;
485 Vc_sub_self = 0.5*tab_coul_F[2];
487 Vc_sub_self = 0.5*tab_coul_V[0];
490 #ifdef CALC_COUL_EWALD
492 Vc_sub_self = 0.5*ic->ewaldcoeff_q*M_2_SQRTPI;
495 for (ia = 0; ia < UNROLLI; ia++)
501 vctp[ia][((egps_i>>(ia*egps_ishift)) & egps_imask)*egps_jstride]
505 -= facel*qi*qi*Vc_sub_self;
513 for (ia = 0; ia < UNROLLI; ia++)
517 c6_i = nbat->nbfp[nbat->type[sci+ia]*(nbat->ntype + 1)*2]/6;
519 vvdwtp[ia][((egps_i>>(ia*egps_ishift)) & egps_imask)*egps_jstride]
523 += 0.5*c6_i*lj_ewaldcoeff6_6;
526 #endif /* LJ_EWALD_GEOM */
530 /* Load i atom data */
531 sciy = scix + STRIDE;
532 sciz = sciy + STRIDE;
533 ix_S0 = gmx_simd_add_r(gmx_simd_load1_r(x+scix), shX_S);
534 ix_S1 = gmx_simd_add_r(gmx_simd_load1_r(x+scix+1), shX_S);
535 ix_S2 = gmx_simd_add_r(gmx_simd_load1_r(x+scix+2), shX_S);
536 ix_S3 = gmx_simd_add_r(gmx_simd_load1_r(x+scix+3), shX_S);
537 iy_S0 = gmx_simd_add_r(gmx_simd_load1_r(x+sciy), shY_S);
538 iy_S1 = gmx_simd_add_r(gmx_simd_load1_r(x+sciy+1), shY_S);
539 iy_S2 = gmx_simd_add_r(gmx_simd_load1_r(x+sciy+2), shY_S);
540 iy_S3 = gmx_simd_add_r(gmx_simd_load1_r(x+sciy+3), shY_S);
541 iz_S0 = gmx_simd_add_r(gmx_simd_load1_r(x+sciz), shZ_S);
542 iz_S1 = gmx_simd_add_r(gmx_simd_load1_r(x+sciz+1), shZ_S);
543 iz_S2 = gmx_simd_add_r(gmx_simd_load1_r(x+sciz+2), shZ_S);
544 iz_S3 = gmx_simd_add_r(gmx_simd_load1_r(x+sciz+3), shZ_S);
548 iq_S0 = gmx_simd_set1_r(facel*q[sci]);
549 iq_S1 = gmx_simd_set1_r(facel*q[sci+1]);
550 iq_S2 = gmx_simd_set1_r(facel*q[sci+2]);
551 iq_S3 = gmx_simd_set1_r(facel*q[sci+3]);
555 hsig_i_S0 = gmx_simd_load1_r(ljc+sci2+0);
556 hsig_i_S1 = gmx_simd_load1_r(ljc+sci2+1);
557 hsig_i_S2 = gmx_simd_load1_r(ljc+sci2+2);
558 hsig_i_S3 = gmx_simd_load1_r(ljc+sci2+3);
559 seps_i_S0 = gmx_simd_load1_r(ljc+sci2+STRIDE+0);
560 seps_i_S1 = gmx_simd_load1_r(ljc+sci2+STRIDE+1);
561 seps_i_S2 = gmx_simd_load1_r(ljc+sci2+STRIDE+2);
562 seps_i_S3 = gmx_simd_load1_r(ljc+sci2+STRIDE+3);
565 c6s_S0 = gmx_simd_load1_r(ljc+sci2+0);
566 c6s_S1 = gmx_simd_load1_r(ljc+sci2+1);
569 c6s_S2 = gmx_simd_load1_r(ljc+sci2+2);
570 c6s_S3 = gmx_simd_load1_r(ljc+sci2+3);
572 c12s_S0 = gmx_simd_load1_r(ljc+sci2+STRIDE+0);
573 c12s_S1 = gmx_simd_load1_r(ljc+sci2+STRIDE+1);
576 c12s_S2 = gmx_simd_load1_r(ljc+sci2+STRIDE+2);
577 c12s_S3 = gmx_simd_load1_r(ljc+sci2+STRIDE+3);
580 nbfp0 = nbfp_ptr + type[sci ]*nbat->ntype*nbfp_stride;
581 nbfp1 = nbfp_ptr + type[sci+1]*nbat->ntype*nbfp_stride;
584 nbfp2 = nbfp_ptr + type[sci+2]*nbat->ntype*nbfp_stride;
585 nbfp3 = nbfp_ptr + type[sci+3]*nbat->ntype*nbfp_stride;
590 /* We need the geometrically combined C6 for the PME grid correction */
591 c6s_S0 = gmx_simd_load1_r(ljc+sci2+0);
592 c6s_S1 = gmx_simd_load1_r(ljc+sci2+1);
595 c6s_S2 = gmx_simd_load1_r(ljc+sci2+2);
596 c6s_S3 = gmx_simd_load1_r(ljc+sci2+3);
600 /* Zero the potential energy for this list */
601 Vvdwtot_S = gmx_simd_setzero_r();
602 vctot_S = gmx_simd_setzero_r();
604 /* Clear i atom forces */
605 fix_S0 = gmx_simd_setzero_r();
606 fix_S1 = gmx_simd_setzero_r();
607 fix_S2 = gmx_simd_setzero_r();
608 fix_S3 = gmx_simd_setzero_r();
609 fiy_S0 = gmx_simd_setzero_r();
610 fiy_S1 = gmx_simd_setzero_r();
611 fiy_S2 = gmx_simd_setzero_r();
612 fiy_S3 = gmx_simd_setzero_r();
613 fiz_S0 = gmx_simd_setzero_r();
614 fiz_S1 = gmx_simd_setzero_r();
615 fiz_S2 = gmx_simd_setzero_r();
616 fiz_S3 = gmx_simd_setzero_r();
620 /* Currently all kernels use (at least half) LJ */
624 /* Coulomb: all i-atoms, LJ: first half i-atoms */
628 while (cjind < cjind1 && nbl->cj[cjind].excl != NBNXN_INTERACTION_MASK_ALL)
630 #include "nbnxn_kernel_simd_4xn_inner.h"
634 for (; (cjind < cjind1); cjind++)
636 #include "nbnxn_kernel_simd_4xn_inner.h"
643 /* Coulomb: all i-atoms, LJ: all i-atoms */
646 while (cjind < cjind1 && nbl->cj[cjind].excl != NBNXN_INTERACTION_MASK_ALL)
648 #include "nbnxn_kernel_simd_4xn_inner.h"
652 for (; (cjind < cjind1); cjind++)
654 #include "nbnxn_kernel_simd_4xn_inner.h"
660 /* Coulomb: none, LJ: all i-atoms */
662 while (cjind < cjind1 && nbl->cj[cjind].excl != NBNXN_INTERACTION_MASK_ALL)
664 #include "nbnxn_kernel_simd_4xn_inner.h"
668 for (; (cjind < cjind1); cjind++)
670 #include "nbnxn_kernel_simd_4xn_inner.h"
674 ninner += cjind1 - cjind0;
676 /* Add accumulated i-forces to the force array */
678 fix_S = gmx_mm_transpose_sum4_pr(fix_S0, fix_S1, fix_S2, fix_S3);
679 gmx_simd4_store_r(f+scix, gmx_simd4_add_r(fix_S, gmx_simd4_load_r(f+scix)));
681 fiy_S = gmx_mm_transpose_sum4_pr(fiy_S0, fiy_S1, fiy_S2, fiy_S3);
682 gmx_simd4_store_r(f+sciy, gmx_simd4_add_r(fiy_S, gmx_simd4_load_r(f+sciy)));
684 fiz_S = gmx_mm_transpose_sum4_pr(fiz_S0, fiz_S1, fiz_S2, fiz_S3);
685 gmx_simd4_store_r(f+sciz, gmx_simd4_add_r(fiz_S, gmx_simd4_load_r(f+sciz)));
687 #ifdef CALC_SHIFTFORCES
688 fshift[ish3+0] += gmx_simd4_reduce_r(fix_S);
689 fshift[ish3+1] += gmx_simd4_reduce_r(fiy_S);
690 fshift[ish3+2] += gmx_simd4_reduce_r(fiz_S);
693 fix0_S = gmx_mm_transpose_sum2_pr(fix_S0, fix_S1);
694 gmx_simd_store_r(f+scix, gmx_simd_add_r(fix0_S, gmx_simd_load_r(f+scix)));
695 fix2_S = gmx_mm_transpose_sum2_pr(fix_S2, fix_S3);
696 gmx_simd_store_r(f+scix+2, gmx_simd_add_r(fix2_S, gmx_simd_load_r(f+scix+2)));
698 fiy0_S = gmx_mm_transpose_sum2_pr(fiy_S0, fiy_S1);
699 gmx_simd_store_r(f+sciy, gmx_simd_add_r(fiy0_S, gmx_simd_load_r(f+sciy)));
700 fiy2_S = gmx_mm_transpose_sum2_pr(fiy_S2, fiy_S3);
701 gmx_simd_store_r(f+sciy+2, gmx_simd_add_r(fiy2_S, gmx_simd_load_r(f+sciy+2)));
703 fiz0_S = gmx_mm_transpose_sum2_pr(fiz_S0, fiz_S1);
704 gmx_simd_store_r(f+sciz, gmx_simd_add_r(fiz0_S, gmx_simd_load_r(f+sciz)));
705 fiz2_S = gmx_mm_transpose_sum2_pr(fiz_S2, fiz_S3);
706 gmx_simd_store_r(f+sciz+2, gmx_simd_add_r(fiz2_S, gmx_simd_load_r(f+sciz+2)));
708 #ifdef CALC_SHIFTFORCES
709 fshift[ish3+0] += gmx_simd_reduce_r(gmx_simd_add_r(fix0_S, fix2_S));
710 fshift[ish3+1] += gmx_simd_reduce_r(gmx_simd_add_r(fiy0_S, fiy2_S));
711 fshift[ish3+2] += gmx_simd_reduce_r(gmx_simd_add_r(fiz0_S, fiz2_S));
718 *Vc += gmx_simd_reduce_r(vctot_S);
721 *Vvdw += gmx_simd_reduce_r(Vvdwtot_S);
724 /* Outer loop uses 6 flops/iteration */
728 printf("atom pairs %d\n", npair);