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.
36 * Note: this file was generated by the GROMACS sse2_double kernel generator.
44 #include "../nb_kernel.h"
45 #include "gromacs/legacyheaders/types/simple.h"
46 #include "gromacs/math/vec.h"
47 #include "gromacs/legacyheaders/nrnb.h"
49 #include "gromacs/simd/math_x86_sse2_double.h"
50 #include "kernelutil_x86_sse2_double.h"
53 * Gromacs nonbonded kernel: nb_kernel_ElecRFCut_VdwLJSh_GeomW3P1_VF_sse2_double
54 * Electrostatics interaction: ReactionField
55 * VdW interaction: LennardJones
56 * Geometry: Water3-Particle
57 * Calculate force/pot: PotentialAndForce
60 nb_kernel_ElecRFCut_VdwLJSh_GeomW3P1_VF_sse2_double
61 (t_nblist * gmx_restrict nlist,
62 rvec * gmx_restrict xx,
63 rvec * gmx_restrict ff,
64 t_forcerec * gmx_restrict fr,
65 t_mdatoms * gmx_restrict mdatoms,
66 nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
67 t_nrnb * gmx_restrict nrnb)
69 /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
70 * just 0 for non-waters.
71 * Suffixes A,B refer to j loop unrolling done with SSE double precision, e.g. for the two different
72 * jnr indices corresponding to data put in the four positions in the SIMD register.
74 int i_shift_offset,i_coord_offset,outeriter,inneriter;
75 int j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
77 int j_coord_offsetA,j_coord_offsetB;
78 int *iinr,*jindex,*jjnr,*shiftidx,*gid;
80 real *shiftvec,*fshift,*x,*f;
81 __m128d tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
83 __m128d ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
85 __m128d ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
87 __m128d ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
88 int vdwjidx0A,vdwjidx0B;
89 __m128d jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
90 __m128d dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
91 __m128d dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
92 __m128d dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
93 __m128d velec,felec,velecsum,facel,crf,krf,krf2;
96 __m128d rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
99 __m128d one_sixth = _mm_set1_pd(1.0/6.0);
100 __m128d one_twelfth = _mm_set1_pd(1.0/12.0);
101 __m128d dummy_mask,cutoff_mask;
102 __m128d signbit = gmx_mm_castsi128_pd( _mm_set_epi32(0x80000000,0x00000000,0x80000000,0x00000000) );
103 __m128d one = _mm_set1_pd(1.0);
104 __m128d two = _mm_set1_pd(2.0);
110 jindex = nlist->jindex;
112 shiftidx = nlist->shift;
114 shiftvec = fr->shift_vec[0];
115 fshift = fr->fshift[0];
116 facel = _mm_set1_pd(fr->epsfac);
117 charge = mdatoms->chargeA;
118 krf = _mm_set1_pd(fr->ic->k_rf);
119 krf2 = _mm_set1_pd(fr->ic->k_rf*2.0);
120 crf = _mm_set1_pd(fr->ic->c_rf);
121 nvdwtype = fr->ntype;
123 vdwtype = mdatoms->typeA;
125 /* Setup water-specific parameters */
126 inr = nlist->iinr[0];
127 iq0 = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+0]));
128 iq1 = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+1]));
129 iq2 = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+2]));
130 vdwioffset0 = 2*nvdwtype*vdwtype[inr+0];
132 /* When we use explicit cutoffs the value must be identical for elec and VdW, so use elec as an arbitrary choice */
133 rcutoff_scalar = fr->rcoulomb;
134 rcutoff = _mm_set1_pd(rcutoff_scalar);
135 rcutoff2 = _mm_mul_pd(rcutoff,rcutoff);
137 sh_vdw_invrcut6 = _mm_set1_pd(fr->ic->sh_invrc6);
138 rvdw = _mm_set1_pd(fr->rvdw);
140 /* Avoid stupid compiler warnings */
148 /* Start outer loop over neighborlists */
149 for(iidx=0; iidx<nri; iidx++)
151 /* Load shift vector for this list */
152 i_shift_offset = DIM*shiftidx[iidx];
154 /* Load limits for loop over neighbors */
155 j_index_start = jindex[iidx];
156 j_index_end = jindex[iidx+1];
158 /* Get outer coordinate index */
160 i_coord_offset = DIM*inr;
162 /* Load i particle coords and add shift vector */
163 gmx_mm_load_shift_and_3rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset,
164 &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2);
166 fix0 = _mm_setzero_pd();
167 fiy0 = _mm_setzero_pd();
168 fiz0 = _mm_setzero_pd();
169 fix1 = _mm_setzero_pd();
170 fiy1 = _mm_setzero_pd();
171 fiz1 = _mm_setzero_pd();
172 fix2 = _mm_setzero_pd();
173 fiy2 = _mm_setzero_pd();
174 fiz2 = _mm_setzero_pd();
176 /* Reset potential sums */
177 velecsum = _mm_setzero_pd();
178 vvdwsum = _mm_setzero_pd();
180 /* Start inner kernel loop */
181 for(jidx=j_index_start; jidx<j_index_end-1; jidx+=2)
184 /* Get j neighbor index, and coordinate index */
187 j_coord_offsetA = DIM*jnrA;
188 j_coord_offsetB = DIM*jnrB;
190 /* load j atom coordinates */
191 gmx_mm_load_1rvec_2ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
194 /* Calculate displacement vector */
195 dx00 = _mm_sub_pd(ix0,jx0);
196 dy00 = _mm_sub_pd(iy0,jy0);
197 dz00 = _mm_sub_pd(iz0,jz0);
198 dx10 = _mm_sub_pd(ix1,jx0);
199 dy10 = _mm_sub_pd(iy1,jy0);
200 dz10 = _mm_sub_pd(iz1,jz0);
201 dx20 = _mm_sub_pd(ix2,jx0);
202 dy20 = _mm_sub_pd(iy2,jy0);
203 dz20 = _mm_sub_pd(iz2,jz0);
205 /* Calculate squared distance and things based on it */
206 rsq00 = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
207 rsq10 = gmx_mm_calc_rsq_pd(dx10,dy10,dz10);
208 rsq20 = gmx_mm_calc_rsq_pd(dx20,dy20,dz20);
210 rinv00 = gmx_mm_invsqrt_pd(rsq00);
211 rinv10 = gmx_mm_invsqrt_pd(rsq10);
212 rinv20 = gmx_mm_invsqrt_pd(rsq20);
214 rinvsq00 = _mm_mul_pd(rinv00,rinv00);
215 rinvsq10 = _mm_mul_pd(rinv10,rinv10);
216 rinvsq20 = _mm_mul_pd(rinv20,rinv20);
218 /* Load parameters for j particles */
219 jq0 = gmx_mm_load_2real_swizzle_pd(charge+jnrA+0,charge+jnrB+0);
220 vdwjidx0A = 2*vdwtype[jnrA+0];
221 vdwjidx0B = 2*vdwtype[jnrB+0];
223 fjx0 = _mm_setzero_pd();
224 fjy0 = _mm_setzero_pd();
225 fjz0 = _mm_setzero_pd();
227 /**************************
228 * CALCULATE INTERACTIONS *
229 **************************/
231 if (gmx_mm_any_lt(rsq00,rcutoff2))
234 /* Compute parameters for interactions between i and j atoms */
235 qq00 = _mm_mul_pd(iq0,jq0);
236 gmx_mm_load_2pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,
237 vdwparam+vdwioffset0+vdwjidx0B,&c6_00,&c12_00);
239 /* REACTION-FIELD ELECTROSTATICS */
240 velec = _mm_mul_pd(qq00,_mm_sub_pd(_mm_add_pd(rinv00,_mm_mul_pd(krf,rsq00)),crf));
241 felec = _mm_mul_pd(qq00,_mm_sub_pd(_mm_mul_pd(rinv00,rinvsq00),krf2));
243 /* LENNARD-JONES DISPERSION/REPULSION */
245 rinvsix = _mm_mul_pd(_mm_mul_pd(rinvsq00,rinvsq00),rinvsq00);
246 vvdw6 = _mm_mul_pd(c6_00,rinvsix);
247 vvdw12 = _mm_mul_pd(c12_00,_mm_mul_pd(rinvsix,rinvsix));
248 vvdw = _mm_sub_pd(_mm_mul_pd( _mm_sub_pd(vvdw12 , _mm_mul_pd(c12_00,_mm_mul_pd(sh_vdw_invrcut6,sh_vdw_invrcut6))), one_twelfth) ,
249 _mm_mul_pd( _mm_sub_pd(vvdw6,_mm_mul_pd(c6_00,sh_vdw_invrcut6)),one_sixth));
250 fvdw = _mm_mul_pd(_mm_sub_pd(vvdw12,vvdw6),rinvsq00);
252 cutoff_mask = _mm_cmplt_pd(rsq00,rcutoff2);
254 /* Update potential sum for this i atom from the interaction with this j atom. */
255 velec = _mm_and_pd(velec,cutoff_mask);
256 velecsum = _mm_add_pd(velecsum,velec);
257 vvdw = _mm_and_pd(vvdw,cutoff_mask);
258 vvdwsum = _mm_add_pd(vvdwsum,vvdw);
260 fscal = _mm_add_pd(felec,fvdw);
262 fscal = _mm_and_pd(fscal,cutoff_mask);
264 /* Calculate temporary vectorial force */
265 tx = _mm_mul_pd(fscal,dx00);
266 ty = _mm_mul_pd(fscal,dy00);
267 tz = _mm_mul_pd(fscal,dz00);
269 /* Update vectorial force */
270 fix0 = _mm_add_pd(fix0,tx);
271 fiy0 = _mm_add_pd(fiy0,ty);
272 fiz0 = _mm_add_pd(fiz0,tz);
274 fjx0 = _mm_add_pd(fjx0,tx);
275 fjy0 = _mm_add_pd(fjy0,ty);
276 fjz0 = _mm_add_pd(fjz0,tz);
280 /**************************
281 * CALCULATE INTERACTIONS *
282 **************************/
284 if (gmx_mm_any_lt(rsq10,rcutoff2))
287 /* Compute parameters for interactions between i and j atoms */
288 qq10 = _mm_mul_pd(iq1,jq0);
290 /* REACTION-FIELD ELECTROSTATICS */
291 velec = _mm_mul_pd(qq10,_mm_sub_pd(_mm_add_pd(rinv10,_mm_mul_pd(krf,rsq10)),crf));
292 felec = _mm_mul_pd(qq10,_mm_sub_pd(_mm_mul_pd(rinv10,rinvsq10),krf2));
294 cutoff_mask = _mm_cmplt_pd(rsq10,rcutoff2);
296 /* Update potential sum for this i atom from the interaction with this j atom. */
297 velec = _mm_and_pd(velec,cutoff_mask);
298 velecsum = _mm_add_pd(velecsum,velec);
302 fscal = _mm_and_pd(fscal,cutoff_mask);
304 /* Calculate temporary vectorial force */
305 tx = _mm_mul_pd(fscal,dx10);
306 ty = _mm_mul_pd(fscal,dy10);
307 tz = _mm_mul_pd(fscal,dz10);
309 /* Update vectorial force */
310 fix1 = _mm_add_pd(fix1,tx);
311 fiy1 = _mm_add_pd(fiy1,ty);
312 fiz1 = _mm_add_pd(fiz1,tz);
314 fjx0 = _mm_add_pd(fjx0,tx);
315 fjy0 = _mm_add_pd(fjy0,ty);
316 fjz0 = _mm_add_pd(fjz0,tz);
320 /**************************
321 * CALCULATE INTERACTIONS *
322 **************************/
324 if (gmx_mm_any_lt(rsq20,rcutoff2))
327 /* Compute parameters for interactions between i and j atoms */
328 qq20 = _mm_mul_pd(iq2,jq0);
330 /* REACTION-FIELD ELECTROSTATICS */
331 velec = _mm_mul_pd(qq20,_mm_sub_pd(_mm_add_pd(rinv20,_mm_mul_pd(krf,rsq20)),crf));
332 felec = _mm_mul_pd(qq20,_mm_sub_pd(_mm_mul_pd(rinv20,rinvsq20),krf2));
334 cutoff_mask = _mm_cmplt_pd(rsq20,rcutoff2);
336 /* Update potential sum for this i atom from the interaction with this j atom. */
337 velec = _mm_and_pd(velec,cutoff_mask);
338 velecsum = _mm_add_pd(velecsum,velec);
342 fscal = _mm_and_pd(fscal,cutoff_mask);
344 /* Calculate temporary vectorial force */
345 tx = _mm_mul_pd(fscal,dx20);
346 ty = _mm_mul_pd(fscal,dy20);
347 tz = _mm_mul_pd(fscal,dz20);
349 /* Update vectorial force */
350 fix2 = _mm_add_pd(fix2,tx);
351 fiy2 = _mm_add_pd(fiy2,ty);
352 fiz2 = _mm_add_pd(fiz2,tz);
354 fjx0 = _mm_add_pd(fjx0,tx);
355 fjy0 = _mm_add_pd(fjy0,ty);
356 fjz0 = _mm_add_pd(fjz0,tz);
360 gmx_mm_decrement_1rvec_2ptr_swizzle_pd(f+j_coord_offsetA,f+j_coord_offsetB,fjx0,fjy0,fjz0);
362 /* Inner loop uses 129 flops */
369 j_coord_offsetA = DIM*jnrA;
371 /* load j atom coordinates */
372 gmx_mm_load_1rvec_1ptr_swizzle_pd(x+j_coord_offsetA,
375 /* Calculate displacement vector */
376 dx00 = _mm_sub_pd(ix0,jx0);
377 dy00 = _mm_sub_pd(iy0,jy0);
378 dz00 = _mm_sub_pd(iz0,jz0);
379 dx10 = _mm_sub_pd(ix1,jx0);
380 dy10 = _mm_sub_pd(iy1,jy0);
381 dz10 = _mm_sub_pd(iz1,jz0);
382 dx20 = _mm_sub_pd(ix2,jx0);
383 dy20 = _mm_sub_pd(iy2,jy0);
384 dz20 = _mm_sub_pd(iz2,jz0);
386 /* Calculate squared distance and things based on it */
387 rsq00 = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
388 rsq10 = gmx_mm_calc_rsq_pd(dx10,dy10,dz10);
389 rsq20 = gmx_mm_calc_rsq_pd(dx20,dy20,dz20);
391 rinv00 = gmx_mm_invsqrt_pd(rsq00);
392 rinv10 = gmx_mm_invsqrt_pd(rsq10);
393 rinv20 = gmx_mm_invsqrt_pd(rsq20);
395 rinvsq00 = _mm_mul_pd(rinv00,rinv00);
396 rinvsq10 = _mm_mul_pd(rinv10,rinv10);
397 rinvsq20 = _mm_mul_pd(rinv20,rinv20);
399 /* Load parameters for j particles */
400 jq0 = _mm_load_sd(charge+jnrA+0);
401 vdwjidx0A = 2*vdwtype[jnrA+0];
403 fjx0 = _mm_setzero_pd();
404 fjy0 = _mm_setzero_pd();
405 fjz0 = _mm_setzero_pd();
407 /**************************
408 * CALCULATE INTERACTIONS *
409 **************************/
411 if (gmx_mm_any_lt(rsq00,rcutoff2))
414 /* Compute parameters for interactions between i and j atoms */
415 qq00 = _mm_mul_pd(iq0,jq0);
416 gmx_mm_load_1pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,&c6_00,&c12_00);
418 /* REACTION-FIELD ELECTROSTATICS */
419 velec = _mm_mul_pd(qq00,_mm_sub_pd(_mm_add_pd(rinv00,_mm_mul_pd(krf,rsq00)),crf));
420 felec = _mm_mul_pd(qq00,_mm_sub_pd(_mm_mul_pd(rinv00,rinvsq00),krf2));
422 /* LENNARD-JONES DISPERSION/REPULSION */
424 rinvsix = _mm_mul_pd(_mm_mul_pd(rinvsq00,rinvsq00),rinvsq00);
425 vvdw6 = _mm_mul_pd(c6_00,rinvsix);
426 vvdw12 = _mm_mul_pd(c12_00,_mm_mul_pd(rinvsix,rinvsix));
427 vvdw = _mm_sub_pd(_mm_mul_pd( _mm_sub_pd(vvdw12 , _mm_mul_pd(c12_00,_mm_mul_pd(sh_vdw_invrcut6,sh_vdw_invrcut6))), one_twelfth) ,
428 _mm_mul_pd( _mm_sub_pd(vvdw6,_mm_mul_pd(c6_00,sh_vdw_invrcut6)),one_sixth));
429 fvdw = _mm_mul_pd(_mm_sub_pd(vvdw12,vvdw6),rinvsq00);
431 cutoff_mask = _mm_cmplt_pd(rsq00,rcutoff2);
433 /* Update potential sum for this i atom from the interaction with this j atom. */
434 velec = _mm_and_pd(velec,cutoff_mask);
435 velec = _mm_unpacklo_pd(velec,_mm_setzero_pd());
436 velecsum = _mm_add_pd(velecsum,velec);
437 vvdw = _mm_and_pd(vvdw,cutoff_mask);
438 vvdw = _mm_unpacklo_pd(vvdw,_mm_setzero_pd());
439 vvdwsum = _mm_add_pd(vvdwsum,vvdw);
441 fscal = _mm_add_pd(felec,fvdw);
443 fscal = _mm_and_pd(fscal,cutoff_mask);
445 fscal = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
447 /* Calculate temporary vectorial force */
448 tx = _mm_mul_pd(fscal,dx00);
449 ty = _mm_mul_pd(fscal,dy00);
450 tz = _mm_mul_pd(fscal,dz00);
452 /* Update vectorial force */
453 fix0 = _mm_add_pd(fix0,tx);
454 fiy0 = _mm_add_pd(fiy0,ty);
455 fiz0 = _mm_add_pd(fiz0,tz);
457 fjx0 = _mm_add_pd(fjx0,tx);
458 fjy0 = _mm_add_pd(fjy0,ty);
459 fjz0 = _mm_add_pd(fjz0,tz);
463 /**************************
464 * CALCULATE INTERACTIONS *
465 **************************/
467 if (gmx_mm_any_lt(rsq10,rcutoff2))
470 /* Compute parameters for interactions between i and j atoms */
471 qq10 = _mm_mul_pd(iq1,jq0);
473 /* REACTION-FIELD ELECTROSTATICS */
474 velec = _mm_mul_pd(qq10,_mm_sub_pd(_mm_add_pd(rinv10,_mm_mul_pd(krf,rsq10)),crf));
475 felec = _mm_mul_pd(qq10,_mm_sub_pd(_mm_mul_pd(rinv10,rinvsq10),krf2));
477 cutoff_mask = _mm_cmplt_pd(rsq10,rcutoff2);
479 /* Update potential sum for this i atom from the interaction with this j atom. */
480 velec = _mm_and_pd(velec,cutoff_mask);
481 velec = _mm_unpacklo_pd(velec,_mm_setzero_pd());
482 velecsum = _mm_add_pd(velecsum,velec);
486 fscal = _mm_and_pd(fscal,cutoff_mask);
488 fscal = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
490 /* Calculate temporary vectorial force */
491 tx = _mm_mul_pd(fscal,dx10);
492 ty = _mm_mul_pd(fscal,dy10);
493 tz = _mm_mul_pd(fscal,dz10);
495 /* Update vectorial force */
496 fix1 = _mm_add_pd(fix1,tx);
497 fiy1 = _mm_add_pd(fiy1,ty);
498 fiz1 = _mm_add_pd(fiz1,tz);
500 fjx0 = _mm_add_pd(fjx0,tx);
501 fjy0 = _mm_add_pd(fjy0,ty);
502 fjz0 = _mm_add_pd(fjz0,tz);
506 /**************************
507 * CALCULATE INTERACTIONS *
508 **************************/
510 if (gmx_mm_any_lt(rsq20,rcutoff2))
513 /* Compute parameters for interactions between i and j atoms */
514 qq20 = _mm_mul_pd(iq2,jq0);
516 /* REACTION-FIELD ELECTROSTATICS */
517 velec = _mm_mul_pd(qq20,_mm_sub_pd(_mm_add_pd(rinv20,_mm_mul_pd(krf,rsq20)),crf));
518 felec = _mm_mul_pd(qq20,_mm_sub_pd(_mm_mul_pd(rinv20,rinvsq20),krf2));
520 cutoff_mask = _mm_cmplt_pd(rsq20,rcutoff2);
522 /* Update potential sum for this i atom from the interaction with this j atom. */
523 velec = _mm_and_pd(velec,cutoff_mask);
524 velec = _mm_unpacklo_pd(velec,_mm_setzero_pd());
525 velecsum = _mm_add_pd(velecsum,velec);
529 fscal = _mm_and_pd(fscal,cutoff_mask);
531 fscal = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
533 /* Calculate temporary vectorial force */
534 tx = _mm_mul_pd(fscal,dx20);
535 ty = _mm_mul_pd(fscal,dy20);
536 tz = _mm_mul_pd(fscal,dz20);
538 /* Update vectorial force */
539 fix2 = _mm_add_pd(fix2,tx);
540 fiy2 = _mm_add_pd(fiy2,ty);
541 fiz2 = _mm_add_pd(fiz2,tz);
543 fjx0 = _mm_add_pd(fjx0,tx);
544 fjy0 = _mm_add_pd(fjy0,ty);
545 fjz0 = _mm_add_pd(fjz0,tz);
549 gmx_mm_decrement_1rvec_1ptr_swizzle_pd(f+j_coord_offsetA,fjx0,fjy0,fjz0);
551 /* Inner loop uses 129 flops */
554 /* End of innermost loop */
556 gmx_mm_update_iforce_3atom_swizzle_pd(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,
557 f+i_coord_offset,fshift+i_shift_offset);
560 /* Update potential energies */
561 gmx_mm_update_1pot_pd(velecsum,kernel_data->energygrp_elec+ggid);
562 gmx_mm_update_1pot_pd(vvdwsum,kernel_data->energygrp_vdw+ggid);
564 /* Increment number of inner iterations */
565 inneriter += j_index_end - j_index_start;
567 /* Outer loop uses 20 flops */
570 /* Increment number of outer iterations */
573 /* Update outer/inner flops */
575 inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W3_VF,outeriter*20 + inneriter*129);
578 * Gromacs nonbonded kernel: nb_kernel_ElecRFCut_VdwLJSh_GeomW3P1_F_sse2_double
579 * Electrostatics interaction: ReactionField
580 * VdW interaction: LennardJones
581 * Geometry: Water3-Particle
582 * Calculate force/pot: Force
585 nb_kernel_ElecRFCut_VdwLJSh_GeomW3P1_F_sse2_double
586 (t_nblist * gmx_restrict nlist,
587 rvec * gmx_restrict xx,
588 rvec * gmx_restrict ff,
589 t_forcerec * gmx_restrict fr,
590 t_mdatoms * gmx_restrict mdatoms,
591 nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
592 t_nrnb * gmx_restrict nrnb)
594 /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
595 * just 0 for non-waters.
596 * Suffixes A,B refer to j loop unrolling done with SSE double precision, e.g. for the two different
597 * jnr indices corresponding to data put in the four positions in the SIMD register.
599 int i_shift_offset,i_coord_offset,outeriter,inneriter;
600 int j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
602 int j_coord_offsetA,j_coord_offsetB;
603 int *iinr,*jindex,*jjnr,*shiftidx,*gid;
605 real *shiftvec,*fshift,*x,*f;
606 __m128d tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
608 __m128d ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
610 __m128d ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
612 __m128d ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
613 int vdwjidx0A,vdwjidx0B;
614 __m128d jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
615 __m128d dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
616 __m128d dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
617 __m128d dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
618 __m128d velec,felec,velecsum,facel,crf,krf,krf2;
621 __m128d rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
624 __m128d one_sixth = _mm_set1_pd(1.0/6.0);
625 __m128d one_twelfth = _mm_set1_pd(1.0/12.0);
626 __m128d dummy_mask,cutoff_mask;
627 __m128d signbit = gmx_mm_castsi128_pd( _mm_set_epi32(0x80000000,0x00000000,0x80000000,0x00000000) );
628 __m128d one = _mm_set1_pd(1.0);
629 __m128d two = _mm_set1_pd(2.0);
635 jindex = nlist->jindex;
637 shiftidx = nlist->shift;
639 shiftvec = fr->shift_vec[0];
640 fshift = fr->fshift[0];
641 facel = _mm_set1_pd(fr->epsfac);
642 charge = mdatoms->chargeA;
643 krf = _mm_set1_pd(fr->ic->k_rf);
644 krf2 = _mm_set1_pd(fr->ic->k_rf*2.0);
645 crf = _mm_set1_pd(fr->ic->c_rf);
646 nvdwtype = fr->ntype;
648 vdwtype = mdatoms->typeA;
650 /* Setup water-specific parameters */
651 inr = nlist->iinr[0];
652 iq0 = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+0]));
653 iq1 = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+1]));
654 iq2 = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+2]));
655 vdwioffset0 = 2*nvdwtype*vdwtype[inr+0];
657 /* When we use explicit cutoffs the value must be identical for elec and VdW, so use elec as an arbitrary choice */
658 rcutoff_scalar = fr->rcoulomb;
659 rcutoff = _mm_set1_pd(rcutoff_scalar);
660 rcutoff2 = _mm_mul_pd(rcutoff,rcutoff);
662 sh_vdw_invrcut6 = _mm_set1_pd(fr->ic->sh_invrc6);
663 rvdw = _mm_set1_pd(fr->rvdw);
665 /* Avoid stupid compiler warnings */
673 /* Start outer loop over neighborlists */
674 for(iidx=0; iidx<nri; iidx++)
676 /* Load shift vector for this list */
677 i_shift_offset = DIM*shiftidx[iidx];
679 /* Load limits for loop over neighbors */
680 j_index_start = jindex[iidx];
681 j_index_end = jindex[iidx+1];
683 /* Get outer coordinate index */
685 i_coord_offset = DIM*inr;
687 /* Load i particle coords and add shift vector */
688 gmx_mm_load_shift_and_3rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset,
689 &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2);
691 fix0 = _mm_setzero_pd();
692 fiy0 = _mm_setzero_pd();
693 fiz0 = _mm_setzero_pd();
694 fix1 = _mm_setzero_pd();
695 fiy1 = _mm_setzero_pd();
696 fiz1 = _mm_setzero_pd();
697 fix2 = _mm_setzero_pd();
698 fiy2 = _mm_setzero_pd();
699 fiz2 = _mm_setzero_pd();
701 /* Start inner kernel loop */
702 for(jidx=j_index_start; jidx<j_index_end-1; jidx+=2)
705 /* Get j neighbor index, and coordinate index */
708 j_coord_offsetA = DIM*jnrA;
709 j_coord_offsetB = DIM*jnrB;
711 /* load j atom coordinates */
712 gmx_mm_load_1rvec_2ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
715 /* Calculate displacement vector */
716 dx00 = _mm_sub_pd(ix0,jx0);
717 dy00 = _mm_sub_pd(iy0,jy0);
718 dz00 = _mm_sub_pd(iz0,jz0);
719 dx10 = _mm_sub_pd(ix1,jx0);
720 dy10 = _mm_sub_pd(iy1,jy0);
721 dz10 = _mm_sub_pd(iz1,jz0);
722 dx20 = _mm_sub_pd(ix2,jx0);
723 dy20 = _mm_sub_pd(iy2,jy0);
724 dz20 = _mm_sub_pd(iz2,jz0);
726 /* Calculate squared distance and things based on it */
727 rsq00 = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
728 rsq10 = gmx_mm_calc_rsq_pd(dx10,dy10,dz10);
729 rsq20 = gmx_mm_calc_rsq_pd(dx20,dy20,dz20);
731 rinv00 = gmx_mm_invsqrt_pd(rsq00);
732 rinv10 = gmx_mm_invsqrt_pd(rsq10);
733 rinv20 = gmx_mm_invsqrt_pd(rsq20);
735 rinvsq00 = _mm_mul_pd(rinv00,rinv00);
736 rinvsq10 = _mm_mul_pd(rinv10,rinv10);
737 rinvsq20 = _mm_mul_pd(rinv20,rinv20);
739 /* Load parameters for j particles */
740 jq0 = gmx_mm_load_2real_swizzle_pd(charge+jnrA+0,charge+jnrB+0);
741 vdwjidx0A = 2*vdwtype[jnrA+0];
742 vdwjidx0B = 2*vdwtype[jnrB+0];
744 fjx0 = _mm_setzero_pd();
745 fjy0 = _mm_setzero_pd();
746 fjz0 = _mm_setzero_pd();
748 /**************************
749 * CALCULATE INTERACTIONS *
750 **************************/
752 if (gmx_mm_any_lt(rsq00,rcutoff2))
755 /* Compute parameters for interactions between i and j atoms */
756 qq00 = _mm_mul_pd(iq0,jq0);
757 gmx_mm_load_2pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,
758 vdwparam+vdwioffset0+vdwjidx0B,&c6_00,&c12_00);
760 /* REACTION-FIELD ELECTROSTATICS */
761 felec = _mm_mul_pd(qq00,_mm_sub_pd(_mm_mul_pd(rinv00,rinvsq00),krf2));
763 /* LENNARD-JONES DISPERSION/REPULSION */
765 rinvsix = _mm_mul_pd(_mm_mul_pd(rinvsq00,rinvsq00),rinvsq00);
766 fvdw = _mm_mul_pd(_mm_sub_pd(_mm_mul_pd(c12_00,rinvsix),c6_00),_mm_mul_pd(rinvsix,rinvsq00));
768 cutoff_mask = _mm_cmplt_pd(rsq00,rcutoff2);
770 fscal = _mm_add_pd(felec,fvdw);
772 fscal = _mm_and_pd(fscal,cutoff_mask);
774 /* Calculate temporary vectorial force */
775 tx = _mm_mul_pd(fscal,dx00);
776 ty = _mm_mul_pd(fscal,dy00);
777 tz = _mm_mul_pd(fscal,dz00);
779 /* Update vectorial force */
780 fix0 = _mm_add_pd(fix0,tx);
781 fiy0 = _mm_add_pd(fiy0,ty);
782 fiz0 = _mm_add_pd(fiz0,tz);
784 fjx0 = _mm_add_pd(fjx0,tx);
785 fjy0 = _mm_add_pd(fjy0,ty);
786 fjz0 = _mm_add_pd(fjz0,tz);
790 /**************************
791 * CALCULATE INTERACTIONS *
792 **************************/
794 if (gmx_mm_any_lt(rsq10,rcutoff2))
797 /* Compute parameters for interactions between i and j atoms */
798 qq10 = _mm_mul_pd(iq1,jq0);
800 /* REACTION-FIELD ELECTROSTATICS */
801 felec = _mm_mul_pd(qq10,_mm_sub_pd(_mm_mul_pd(rinv10,rinvsq10),krf2));
803 cutoff_mask = _mm_cmplt_pd(rsq10,rcutoff2);
807 fscal = _mm_and_pd(fscal,cutoff_mask);
809 /* Calculate temporary vectorial force */
810 tx = _mm_mul_pd(fscal,dx10);
811 ty = _mm_mul_pd(fscal,dy10);
812 tz = _mm_mul_pd(fscal,dz10);
814 /* Update vectorial force */
815 fix1 = _mm_add_pd(fix1,tx);
816 fiy1 = _mm_add_pd(fiy1,ty);
817 fiz1 = _mm_add_pd(fiz1,tz);
819 fjx0 = _mm_add_pd(fjx0,tx);
820 fjy0 = _mm_add_pd(fjy0,ty);
821 fjz0 = _mm_add_pd(fjz0,tz);
825 /**************************
826 * CALCULATE INTERACTIONS *
827 **************************/
829 if (gmx_mm_any_lt(rsq20,rcutoff2))
832 /* Compute parameters for interactions between i and j atoms */
833 qq20 = _mm_mul_pd(iq2,jq0);
835 /* REACTION-FIELD ELECTROSTATICS */
836 felec = _mm_mul_pd(qq20,_mm_sub_pd(_mm_mul_pd(rinv20,rinvsq20),krf2));
838 cutoff_mask = _mm_cmplt_pd(rsq20,rcutoff2);
842 fscal = _mm_and_pd(fscal,cutoff_mask);
844 /* Calculate temporary vectorial force */
845 tx = _mm_mul_pd(fscal,dx20);
846 ty = _mm_mul_pd(fscal,dy20);
847 tz = _mm_mul_pd(fscal,dz20);
849 /* Update vectorial force */
850 fix2 = _mm_add_pd(fix2,tx);
851 fiy2 = _mm_add_pd(fiy2,ty);
852 fiz2 = _mm_add_pd(fiz2,tz);
854 fjx0 = _mm_add_pd(fjx0,tx);
855 fjy0 = _mm_add_pd(fjy0,ty);
856 fjz0 = _mm_add_pd(fjz0,tz);
860 gmx_mm_decrement_1rvec_2ptr_swizzle_pd(f+j_coord_offsetA,f+j_coord_offsetB,fjx0,fjy0,fjz0);
862 /* Inner loop uses 100 flops */
869 j_coord_offsetA = DIM*jnrA;
871 /* load j atom coordinates */
872 gmx_mm_load_1rvec_1ptr_swizzle_pd(x+j_coord_offsetA,
875 /* Calculate displacement vector */
876 dx00 = _mm_sub_pd(ix0,jx0);
877 dy00 = _mm_sub_pd(iy0,jy0);
878 dz00 = _mm_sub_pd(iz0,jz0);
879 dx10 = _mm_sub_pd(ix1,jx0);
880 dy10 = _mm_sub_pd(iy1,jy0);
881 dz10 = _mm_sub_pd(iz1,jz0);
882 dx20 = _mm_sub_pd(ix2,jx0);
883 dy20 = _mm_sub_pd(iy2,jy0);
884 dz20 = _mm_sub_pd(iz2,jz0);
886 /* Calculate squared distance and things based on it */
887 rsq00 = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
888 rsq10 = gmx_mm_calc_rsq_pd(dx10,dy10,dz10);
889 rsq20 = gmx_mm_calc_rsq_pd(dx20,dy20,dz20);
891 rinv00 = gmx_mm_invsqrt_pd(rsq00);
892 rinv10 = gmx_mm_invsqrt_pd(rsq10);
893 rinv20 = gmx_mm_invsqrt_pd(rsq20);
895 rinvsq00 = _mm_mul_pd(rinv00,rinv00);
896 rinvsq10 = _mm_mul_pd(rinv10,rinv10);
897 rinvsq20 = _mm_mul_pd(rinv20,rinv20);
899 /* Load parameters for j particles */
900 jq0 = _mm_load_sd(charge+jnrA+0);
901 vdwjidx0A = 2*vdwtype[jnrA+0];
903 fjx0 = _mm_setzero_pd();
904 fjy0 = _mm_setzero_pd();
905 fjz0 = _mm_setzero_pd();
907 /**************************
908 * CALCULATE INTERACTIONS *
909 **************************/
911 if (gmx_mm_any_lt(rsq00,rcutoff2))
914 /* Compute parameters for interactions between i and j atoms */
915 qq00 = _mm_mul_pd(iq0,jq0);
916 gmx_mm_load_1pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,&c6_00,&c12_00);
918 /* REACTION-FIELD ELECTROSTATICS */
919 felec = _mm_mul_pd(qq00,_mm_sub_pd(_mm_mul_pd(rinv00,rinvsq00),krf2));
921 /* LENNARD-JONES DISPERSION/REPULSION */
923 rinvsix = _mm_mul_pd(_mm_mul_pd(rinvsq00,rinvsq00),rinvsq00);
924 fvdw = _mm_mul_pd(_mm_sub_pd(_mm_mul_pd(c12_00,rinvsix),c6_00),_mm_mul_pd(rinvsix,rinvsq00));
926 cutoff_mask = _mm_cmplt_pd(rsq00,rcutoff2);
928 fscal = _mm_add_pd(felec,fvdw);
930 fscal = _mm_and_pd(fscal,cutoff_mask);
932 fscal = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
934 /* Calculate temporary vectorial force */
935 tx = _mm_mul_pd(fscal,dx00);
936 ty = _mm_mul_pd(fscal,dy00);
937 tz = _mm_mul_pd(fscal,dz00);
939 /* Update vectorial force */
940 fix0 = _mm_add_pd(fix0,tx);
941 fiy0 = _mm_add_pd(fiy0,ty);
942 fiz0 = _mm_add_pd(fiz0,tz);
944 fjx0 = _mm_add_pd(fjx0,tx);
945 fjy0 = _mm_add_pd(fjy0,ty);
946 fjz0 = _mm_add_pd(fjz0,tz);
950 /**************************
951 * CALCULATE INTERACTIONS *
952 **************************/
954 if (gmx_mm_any_lt(rsq10,rcutoff2))
957 /* Compute parameters for interactions between i and j atoms */
958 qq10 = _mm_mul_pd(iq1,jq0);
960 /* REACTION-FIELD ELECTROSTATICS */
961 felec = _mm_mul_pd(qq10,_mm_sub_pd(_mm_mul_pd(rinv10,rinvsq10),krf2));
963 cutoff_mask = _mm_cmplt_pd(rsq10,rcutoff2);
967 fscal = _mm_and_pd(fscal,cutoff_mask);
969 fscal = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
971 /* Calculate temporary vectorial force */
972 tx = _mm_mul_pd(fscal,dx10);
973 ty = _mm_mul_pd(fscal,dy10);
974 tz = _mm_mul_pd(fscal,dz10);
976 /* Update vectorial force */
977 fix1 = _mm_add_pd(fix1,tx);
978 fiy1 = _mm_add_pd(fiy1,ty);
979 fiz1 = _mm_add_pd(fiz1,tz);
981 fjx0 = _mm_add_pd(fjx0,tx);
982 fjy0 = _mm_add_pd(fjy0,ty);
983 fjz0 = _mm_add_pd(fjz0,tz);
987 /**************************
988 * CALCULATE INTERACTIONS *
989 **************************/
991 if (gmx_mm_any_lt(rsq20,rcutoff2))
994 /* Compute parameters for interactions between i and j atoms */
995 qq20 = _mm_mul_pd(iq2,jq0);
997 /* REACTION-FIELD ELECTROSTATICS */
998 felec = _mm_mul_pd(qq20,_mm_sub_pd(_mm_mul_pd(rinv20,rinvsq20),krf2));
1000 cutoff_mask = _mm_cmplt_pd(rsq20,rcutoff2);
1004 fscal = _mm_and_pd(fscal,cutoff_mask);
1006 fscal = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
1008 /* Calculate temporary vectorial force */
1009 tx = _mm_mul_pd(fscal,dx20);
1010 ty = _mm_mul_pd(fscal,dy20);
1011 tz = _mm_mul_pd(fscal,dz20);
1013 /* Update vectorial force */
1014 fix2 = _mm_add_pd(fix2,tx);
1015 fiy2 = _mm_add_pd(fiy2,ty);
1016 fiz2 = _mm_add_pd(fiz2,tz);
1018 fjx0 = _mm_add_pd(fjx0,tx);
1019 fjy0 = _mm_add_pd(fjy0,ty);
1020 fjz0 = _mm_add_pd(fjz0,tz);
1024 gmx_mm_decrement_1rvec_1ptr_swizzle_pd(f+j_coord_offsetA,fjx0,fjy0,fjz0);
1026 /* Inner loop uses 100 flops */
1029 /* End of innermost loop */
1031 gmx_mm_update_iforce_3atom_swizzle_pd(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,
1032 f+i_coord_offset,fshift+i_shift_offset);
1034 /* Increment number of inner iterations */
1035 inneriter += j_index_end - j_index_start;
1037 /* Outer loop uses 18 flops */
1040 /* Increment number of outer iterations */
1043 /* Update outer/inner flops */
1045 inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W3_F,outeriter*18 + inneriter*100);