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 sse4_1_double kernel generator.
42 #include "../nb_kernel.h"
43 #include "types/simple.h"
44 #include "gromacs/math/vec.h"
47 #include "gromacs/simd/math_x86_sse4_1_double.h"
48 #include "kernelutil_x86_sse4_1_double.h"
51 * Gromacs nonbonded kernel: nb_kernel_ElecRF_VdwLJ_GeomW4P1_VF_sse4_1_double
52 * Electrostatics interaction: ReactionField
53 * VdW interaction: LennardJones
54 * Geometry: Water4-Particle
55 * Calculate force/pot: PotentialAndForce
58 nb_kernel_ElecRF_VdwLJ_GeomW4P1_VF_sse4_1_double
59 (t_nblist * gmx_restrict nlist,
60 rvec * gmx_restrict xx,
61 rvec * gmx_restrict ff,
62 t_forcerec * gmx_restrict fr,
63 t_mdatoms * gmx_restrict mdatoms,
64 nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
65 t_nrnb * gmx_restrict nrnb)
67 /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
68 * just 0 for non-waters.
69 * Suffixes A,B refer to j loop unrolling done with SSE double precision, e.g. for the two different
70 * jnr indices corresponding to data put in the four positions in the SIMD register.
72 int i_shift_offset,i_coord_offset,outeriter,inneriter;
73 int j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
75 int j_coord_offsetA,j_coord_offsetB;
76 int *iinr,*jindex,*jjnr,*shiftidx,*gid;
78 real *shiftvec,*fshift,*x,*f;
79 __m128d tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
81 __m128d ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
83 __m128d ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
85 __m128d ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
87 __m128d ix3,iy3,iz3,fix3,fiy3,fiz3,iq3,isai3;
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 dx30,dy30,dz30,rsq30,rinv30,rinvsq30,r30,qq30,c6_30,c12_30;
94 __m128d velec,felec,velecsum,facel,crf,krf,krf2;
97 __m128d rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
100 __m128d one_sixth = _mm_set1_pd(1.0/6.0);
101 __m128d one_twelfth = _mm_set1_pd(1.0/12.0);
102 __m128d dummy_mask,cutoff_mask;
103 __m128d signbit = gmx_mm_castsi128_pd( _mm_set_epi32(0x80000000,0x00000000,0x80000000,0x00000000) );
104 __m128d one = _mm_set1_pd(1.0);
105 __m128d two = _mm_set1_pd(2.0);
111 jindex = nlist->jindex;
113 shiftidx = nlist->shift;
115 shiftvec = fr->shift_vec[0];
116 fshift = fr->fshift[0];
117 facel = _mm_set1_pd(fr->epsfac);
118 charge = mdatoms->chargeA;
119 krf = _mm_set1_pd(fr->ic->k_rf);
120 krf2 = _mm_set1_pd(fr->ic->k_rf*2.0);
121 crf = _mm_set1_pd(fr->ic->c_rf);
122 nvdwtype = fr->ntype;
124 vdwtype = mdatoms->typeA;
126 /* Setup water-specific parameters */
127 inr = nlist->iinr[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 iq3 = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+3]));
131 vdwioffset0 = 2*nvdwtype*vdwtype[inr+0];
133 /* Avoid stupid compiler warnings */
141 /* Start outer loop over neighborlists */
142 for(iidx=0; iidx<nri; iidx++)
144 /* Load shift vector for this list */
145 i_shift_offset = DIM*shiftidx[iidx];
147 /* Load limits for loop over neighbors */
148 j_index_start = jindex[iidx];
149 j_index_end = jindex[iidx+1];
151 /* Get outer coordinate index */
153 i_coord_offset = DIM*inr;
155 /* Load i particle coords and add shift vector */
156 gmx_mm_load_shift_and_4rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset,
157 &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2,&ix3,&iy3,&iz3);
159 fix0 = _mm_setzero_pd();
160 fiy0 = _mm_setzero_pd();
161 fiz0 = _mm_setzero_pd();
162 fix1 = _mm_setzero_pd();
163 fiy1 = _mm_setzero_pd();
164 fiz1 = _mm_setzero_pd();
165 fix2 = _mm_setzero_pd();
166 fiy2 = _mm_setzero_pd();
167 fiz2 = _mm_setzero_pd();
168 fix3 = _mm_setzero_pd();
169 fiy3 = _mm_setzero_pd();
170 fiz3 = _mm_setzero_pd();
172 /* Reset potential sums */
173 velecsum = _mm_setzero_pd();
174 vvdwsum = _mm_setzero_pd();
176 /* Start inner kernel loop */
177 for(jidx=j_index_start; jidx<j_index_end-1; jidx+=2)
180 /* Get j neighbor index, and coordinate index */
183 j_coord_offsetA = DIM*jnrA;
184 j_coord_offsetB = DIM*jnrB;
186 /* load j atom coordinates */
187 gmx_mm_load_1rvec_2ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
190 /* Calculate displacement vector */
191 dx00 = _mm_sub_pd(ix0,jx0);
192 dy00 = _mm_sub_pd(iy0,jy0);
193 dz00 = _mm_sub_pd(iz0,jz0);
194 dx10 = _mm_sub_pd(ix1,jx0);
195 dy10 = _mm_sub_pd(iy1,jy0);
196 dz10 = _mm_sub_pd(iz1,jz0);
197 dx20 = _mm_sub_pd(ix2,jx0);
198 dy20 = _mm_sub_pd(iy2,jy0);
199 dz20 = _mm_sub_pd(iz2,jz0);
200 dx30 = _mm_sub_pd(ix3,jx0);
201 dy30 = _mm_sub_pd(iy3,jy0);
202 dz30 = _mm_sub_pd(iz3,jz0);
204 /* Calculate squared distance and things based on it */
205 rsq00 = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
206 rsq10 = gmx_mm_calc_rsq_pd(dx10,dy10,dz10);
207 rsq20 = gmx_mm_calc_rsq_pd(dx20,dy20,dz20);
208 rsq30 = gmx_mm_calc_rsq_pd(dx30,dy30,dz30);
210 rinv10 = gmx_mm_invsqrt_pd(rsq10);
211 rinv20 = gmx_mm_invsqrt_pd(rsq20);
212 rinv30 = gmx_mm_invsqrt_pd(rsq30);
214 rinvsq00 = gmx_mm_inv_pd(rsq00);
215 rinvsq10 = _mm_mul_pd(rinv10,rinv10);
216 rinvsq20 = _mm_mul_pd(rinv20,rinv20);
217 rinvsq30 = _mm_mul_pd(rinv30,rinv30);
219 /* Load parameters for j particles */
220 jq0 = gmx_mm_load_2real_swizzle_pd(charge+jnrA+0,charge+jnrB+0);
221 vdwjidx0A = 2*vdwtype[jnrA+0];
222 vdwjidx0B = 2*vdwtype[jnrB+0];
224 fjx0 = _mm_setzero_pd();
225 fjy0 = _mm_setzero_pd();
226 fjz0 = _mm_setzero_pd();
228 /**************************
229 * CALCULATE INTERACTIONS *
230 **************************/
232 /* Compute parameters for interactions between i and j atoms */
233 gmx_mm_load_2pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,
234 vdwparam+vdwioffset0+vdwjidx0B,&c6_00,&c12_00);
236 /* LENNARD-JONES DISPERSION/REPULSION */
238 rinvsix = _mm_mul_pd(_mm_mul_pd(rinvsq00,rinvsq00),rinvsq00);
239 vvdw6 = _mm_mul_pd(c6_00,rinvsix);
240 vvdw12 = _mm_mul_pd(c12_00,_mm_mul_pd(rinvsix,rinvsix));
241 vvdw = _mm_sub_pd( _mm_mul_pd(vvdw12,one_twelfth) , _mm_mul_pd(vvdw6,one_sixth) );
242 fvdw = _mm_mul_pd(_mm_sub_pd(vvdw12,vvdw6),rinvsq00);
244 /* Update potential sum for this i atom from the interaction with this j atom. */
245 vvdwsum = _mm_add_pd(vvdwsum,vvdw);
249 /* Calculate temporary vectorial force */
250 tx = _mm_mul_pd(fscal,dx00);
251 ty = _mm_mul_pd(fscal,dy00);
252 tz = _mm_mul_pd(fscal,dz00);
254 /* Update vectorial force */
255 fix0 = _mm_add_pd(fix0,tx);
256 fiy0 = _mm_add_pd(fiy0,ty);
257 fiz0 = _mm_add_pd(fiz0,tz);
259 fjx0 = _mm_add_pd(fjx0,tx);
260 fjy0 = _mm_add_pd(fjy0,ty);
261 fjz0 = _mm_add_pd(fjz0,tz);
263 /**************************
264 * CALCULATE INTERACTIONS *
265 **************************/
267 /* Compute parameters for interactions between i and j atoms */
268 qq10 = _mm_mul_pd(iq1,jq0);
270 /* REACTION-FIELD ELECTROSTATICS */
271 velec = _mm_mul_pd(qq10,_mm_sub_pd(_mm_add_pd(rinv10,_mm_mul_pd(krf,rsq10)),crf));
272 felec = _mm_mul_pd(qq10,_mm_sub_pd(_mm_mul_pd(rinv10,rinvsq10),krf2));
274 /* Update potential sum for this i atom from the interaction with this j atom. */
275 velecsum = _mm_add_pd(velecsum,velec);
279 /* Calculate temporary vectorial force */
280 tx = _mm_mul_pd(fscal,dx10);
281 ty = _mm_mul_pd(fscal,dy10);
282 tz = _mm_mul_pd(fscal,dz10);
284 /* Update vectorial force */
285 fix1 = _mm_add_pd(fix1,tx);
286 fiy1 = _mm_add_pd(fiy1,ty);
287 fiz1 = _mm_add_pd(fiz1,tz);
289 fjx0 = _mm_add_pd(fjx0,tx);
290 fjy0 = _mm_add_pd(fjy0,ty);
291 fjz0 = _mm_add_pd(fjz0,tz);
293 /**************************
294 * CALCULATE INTERACTIONS *
295 **************************/
297 /* Compute parameters for interactions between i and j atoms */
298 qq20 = _mm_mul_pd(iq2,jq0);
300 /* REACTION-FIELD ELECTROSTATICS */
301 velec = _mm_mul_pd(qq20,_mm_sub_pd(_mm_add_pd(rinv20,_mm_mul_pd(krf,rsq20)),crf));
302 felec = _mm_mul_pd(qq20,_mm_sub_pd(_mm_mul_pd(rinv20,rinvsq20),krf2));
304 /* Update potential sum for this i atom from the interaction with this j atom. */
305 velecsum = _mm_add_pd(velecsum,velec);
309 /* Calculate temporary vectorial force */
310 tx = _mm_mul_pd(fscal,dx20);
311 ty = _mm_mul_pd(fscal,dy20);
312 tz = _mm_mul_pd(fscal,dz20);
314 /* Update vectorial force */
315 fix2 = _mm_add_pd(fix2,tx);
316 fiy2 = _mm_add_pd(fiy2,ty);
317 fiz2 = _mm_add_pd(fiz2,tz);
319 fjx0 = _mm_add_pd(fjx0,tx);
320 fjy0 = _mm_add_pd(fjy0,ty);
321 fjz0 = _mm_add_pd(fjz0,tz);
323 /**************************
324 * CALCULATE INTERACTIONS *
325 **************************/
327 /* Compute parameters for interactions between i and j atoms */
328 qq30 = _mm_mul_pd(iq3,jq0);
330 /* REACTION-FIELD ELECTROSTATICS */
331 velec = _mm_mul_pd(qq30,_mm_sub_pd(_mm_add_pd(rinv30,_mm_mul_pd(krf,rsq30)),crf));
332 felec = _mm_mul_pd(qq30,_mm_sub_pd(_mm_mul_pd(rinv30,rinvsq30),krf2));
334 /* Update potential sum for this i atom from the interaction with this j atom. */
335 velecsum = _mm_add_pd(velecsum,velec);
339 /* Calculate temporary vectorial force */
340 tx = _mm_mul_pd(fscal,dx30);
341 ty = _mm_mul_pd(fscal,dy30);
342 tz = _mm_mul_pd(fscal,dz30);
344 /* Update vectorial force */
345 fix3 = _mm_add_pd(fix3,tx);
346 fiy3 = _mm_add_pd(fiy3,ty);
347 fiz3 = _mm_add_pd(fiz3,tz);
349 fjx0 = _mm_add_pd(fjx0,tx);
350 fjy0 = _mm_add_pd(fjy0,ty);
351 fjz0 = _mm_add_pd(fjz0,tz);
353 gmx_mm_decrement_1rvec_2ptr_swizzle_pd(f+j_coord_offsetA,f+j_coord_offsetB,fjx0,fjy0,fjz0);
355 /* Inner loop uses 131 flops */
362 j_coord_offsetA = DIM*jnrA;
364 /* load j atom coordinates */
365 gmx_mm_load_1rvec_1ptr_swizzle_pd(x+j_coord_offsetA,
368 /* Calculate displacement vector */
369 dx00 = _mm_sub_pd(ix0,jx0);
370 dy00 = _mm_sub_pd(iy0,jy0);
371 dz00 = _mm_sub_pd(iz0,jz0);
372 dx10 = _mm_sub_pd(ix1,jx0);
373 dy10 = _mm_sub_pd(iy1,jy0);
374 dz10 = _mm_sub_pd(iz1,jz0);
375 dx20 = _mm_sub_pd(ix2,jx0);
376 dy20 = _mm_sub_pd(iy2,jy0);
377 dz20 = _mm_sub_pd(iz2,jz0);
378 dx30 = _mm_sub_pd(ix3,jx0);
379 dy30 = _mm_sub_pd(iy3,jy0);
380 dz30 = _mm_sub_pd(iz3,jz0);
382 /* Calculate squared distance and things based on it */
383 rsq00 = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
384 rsq10 = gmx_mm_calc_rsq_pd(dx10,dy10,dz10);
385 rsq20 = gmx_mm_calc_rsq_pd(dx20,dy20,dz20);
386 rsq30 = gmx_mm_calc_rsq_pd(dx30,dy30,dz30);
388 rinv10 = gmx_mm_invsqrt_pd(rsq10);
389 rinv20 = gmx_mm_invsqrt_pd(rsq20);
390 rinv30 = gmx_mm_invsqrt_pd(rsq30);
392 rinvsq00 = gmx_mm_inv_pd(rsq00);
393 rinvsq10 = _mm_mul_pd(rinv10,rinv10);
394 rinvsq20 = _mm_mul_pd(rinv20,rinv20);
395 rinvsq30 = _mm_mul_pd(rinv30,rinv30);
397 /* Load parameters for j particles */
398 jq0 = _mm_load_sd(charge+jnrA+0);
399 vdwjidx0A = 2*vdwtype[jnrA+0];
401 fjx0 = _mm_setzero_pd();
402 fjy0 = _mm_setzero_pd();
403 fjz0 = _mm_setzero_pd();
405 /**************************
406 * CALCULATE INTERACTIONS *
407 **************************/
409 /* Compute parameters for interactions between i and j atoms */
410 gmx_mm_load_1pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,&c6_00,&c12_00);
412 /* LENNARD-JONES DISPERSION/REPULSION */
414 rinvsix = _mm_mul_pd(_mm_mul_pd(rinvsq00,rinvsq00),rinvsq00);
415 vvdw6 = _mm_mul_pd(c6_00,rinvsix);
416 vvdw12 = _mm_mul_pd(c12_00,_mm_mul_pd(rinvsix,rinvsix));
417 vvdw = _mm_sub_pd( _mm_mul_pd(vvdw12,one_twelfth) , _mm_mul_pd(vvdw6,one_sixth) );
418 fvdw = _mm_mul_pd(_mm_sub_pd(vvdw12,vvdw6),rinvsq00);
420 /* Update potential sum for this i atom from the interaction with this j atom. */
421 vvdw = _mm_unpacklo_pd(vvdw,_mm_setzero_pd());
422 vvdwsum = _mm_add_pd(vvdwsum,vvdw);
426 fscal = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
428 /* Calculate temporary vectorial force */
429 tx = _mm_mul_pd(fscal,dx00);
430 ty = _mm_mul_pd(fscal,dy00);
431 tz = _mm_mul_pd(fscal,dz00);
433 /* Update vectorial force */
434 fix0 = _mm_add_pd(fix0,tx);
435 fiy0 = _mm_add_pd(fiy0,ty);
436 fiz0 = _mm_add_pd(fiz0,tz);
438 fjx0 = _mm_add_pd(fjx0,tx);
439 fjy0 = _mm_add_pd(fjy0,ty);
440 fjz0 = _mm_add_pd(fjz0,tz);
442 /**************************
443 * CALCULATE INTERACTIONS *
444 **************************/
446 /* Compute parameters for interactions between i and j atoms */
447 qq10 = _mm_mul_pd(iq1,jq0);
449 /* REACTION-FIELD ELECTROSTATICS */
450 velec = _mm_mul_pd(qq10,_mm_sub_pd(_mm_add_pd(rinv10,_mm_mul_pd(krf,rsq10)),crf));
451 felec = _mm_mul_pd(qq10,_mm_sub_pd(_mm_mul_pd(rinv10,rinvsq10),krf2));
453 /* Update potential sum for this i atom from the interaction with this j atom. */
454 velec = _mm_unpacklo_pd(velec,_mm_setzero_pd());
455 velecsum = _mm_add_pd(velecsum,velec);
459 fscal = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
461 /* Calculate temporary vectorial force */
462 tx = _mm_mul_pd(fscal,dx10);
463 ty = _mm_mul_pd(fscal,dy10);
464 tz = _mm_mul_pd(fscal,dz10);
466 /* Update vectorial force */
467 fix1 = _mm_add_pd(fix1,tx);
468 fiy1 = _mm_add_pd(fiy1,ty);
469 fiz1 = _mm_add_pd(fiz1,tz);
471 fjx0 = _mm_add_pd(fjx0,tx);
472 fjy0 = _mm_add_pd(fjy0,ty);
473 fjz0 = _mm_add_pd(fjz0,tz);
475 /**************************
476 * CALCULATE INTERACTIONS *
477 **************************/
479 /* Compute parameters for interactions between i and j atoms */
480 qq20 = _mm_mul_pd(iq2,jq0);
482 /* REACTION-FIELD ELECTROSTATICS */
483 velec = _mm_mul_pd(qq20,_mm_sub_pd(_mm_add_pd(rinv20,_mm_mul_pd(krf,rsq20)),crf));
484 felec = _mm_mul_pd(qq20,_mm_sub_pd(_mm_mul_pd(rinv20,rinvsq20),krf2));
486 /* Update potential sum for this i atom from the interaction with this j atom. */
487 velec = _mm_unpacklo_pd(velec,_mm_setzero_pd());
488 velecsum = _mm_add_pd(velecsum,velec);
492 fscal = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
494 /* Calculate temporary vectorial force */
495 tx = _mm_mul_pd(fscal,dx20);
496 ty = _mm_mul_pd(fscal,dy20);
497 tz = _mm_mul_pd(fscal,dz20);
499 /* Update vectorial force */
500 fix2 = _mm_add_pd(fix2,tx);
501 fiy2 = _mm_add_pd(fiy2,ty);
502 fiz2 = _mm_add_pd(fiz2,tz);
504 fjx0 = _mm_add_pd(fjx0,tx);
505 fjy0 = _mm_add_pd(fjy0,ty);
506 fjz0 = _mm_add_pd(fjz0,tz);
508 /**************************
509 * CALCULATE INTERACTIONS *
510 **************************/
512 /* Compute parameters for interactions between i and j atoms */
513 qq30 = _mm_mul_pd(iq3,jq0);
515 /* REACTION-FIELD ELECTROSTATICS */
516 velec = _mm_mul_pd(qq30,_mm_sub_pd(_mm_add_pd(rinv30,_mm_mul_pd(krf,rsq30)),crf));
517 felec = _mm_mul_pd(qq30,_mm_sub_pd(_mm_mul_pd(rinv30,rinvsq30),krf2));
519 /* Update potential sum for this i atom from the interaction with this j atom. */
520 velec = _mm_unpacklo_pd(velec,_mm_setzero_pd());
521 velecsum = _mm_add_pd(velecsum,velec);
525 fscal = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
527 /* Calculate temporary vectorial force */
528 tx = _mm_mul_pd(fscal,dx30);
529 ty = _mm_mul_pd(fscal,dy30);
530 tz = _mm_mul_pd(fscal,dz30);
532 /* Update vectorial force */
533 fix3 = _mm_add_pd(fix3,tx);
534 fiy3 = _mm_add_pd(fiy3,ty);
535 fiz3 = _mm_add_pd(fiz3,tz);
537 fjx0 = _mm_add_pd(fjx0,tx);
538 fjy0 = _mm_add_pd(fjy0,ty);
539 fjz0 = _mm_add_pd(fjz0,tz);
541 gmx_mm_decrement_1rvec_1ptr_swizzle_pd(f+j_coord_offsetA,fjx0,fjy0,fjz0);
543 /* Inner loop uses 131 flops */
546 /* End of innermost loop */
548 gmx_mm_update_iforce_4atom_swizzle_pd(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,fix3,fiy3,fiz3,
549 f+i_coord_offset,fshift+i_shift_offset);
552 /* Update potential energies */
553 gmx_mm_update_1pot_pd(velecsum,kernel_data->energygrp_elec+ggid);
554 gmx_mm_update_1pot_pd(vvdwsum,kernel_data->energygrp_vdw+ggid);
556 /* Increment number of inner iterations */
557 inneriter += j_index_end - j_index_start;
559 /* Outer loop uses 26 flops */
562 /* Increment number of outer iterations */
565 /* Update outer/inner flops */
567 inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W4_VF,outeriter*26 + inneriter*131);
570 * Gromacs nonbonded kernel: nb_kernel_ElecRF_VdwLJ_GeomW4P1_F_sse4_1_double
571 * Electrostatics interaction: ReactionField
572 * VdW interaction: LennardJones
573 * Geometry: Water4-Particle
574 * Calculate force/pot: Force
577 nb_kernel_ElecRF_VdwLJ_GeomW4P1_F_sse4_1_double
578 (t_nblist * gmx_restrict nlist,
579 rvec * gmx_restrict xx,
580 rvec * gmx_restrict ff,
581 t_forcerec * gmx_restrict fr,
582 t_mdatoms * gmx_restrict mdatoms,
583 nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
584 t_nrnb * gmx_restrict nrnb)
586 /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
587 * just 0 for non-waters.
588 * Suffixes A,B refer to j loop unrolling done with SSE double precision, e.g. for the two different
589 * jnr indices corresponding to data put in the four positions in the SIMD register.
591 int i_shift_offset,i_coord_offset,outeriter,inneriter;
592 int j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
594 int j_coord_offsetA,j_coord_offsetB;
595 int *iinr,*jindex,*jjnr,*shiftidx,*gid;
597 real *shiftvec,*fshift,*x,*f;
598 __m128d tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
600 __m128d ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
602 __m128d ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
604 __m128d ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
606 __m128d ix3,iy3,iz3,fix3,fiy3,fiz3,iq3,isai3;
607 int vdwjidx0A,vdwjidx0B;
608 __m128d jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
609 __m128d dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
610 __m128d dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
611 __m128d dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
612 __m128d dx30,dy30,dz30,rsq30,rinv30,rinvsq30,r30,qq30,c6_30,c12_30;
613 __m128d velec,felec,velecsum,facel,crf,krf,krf2;
616 __m128d rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
619 __m128d one_sixth = _mm_set1_pd(1.0/6.0);
620 __m128d one_twelfth = _mm_set1_pd(1.0/12.0);
621 __m128d dummy_mask,cutoff_mask;
622 __m128d signbit = gmx_mm_castsi128_pd( _mm_set_epi32(0x80000000,0x00000000,0x80000000,0x00000000) );
623 __m128d one = _mm_set1_pd(1.0);
624 __m128d two = _mm_set1_pd(2.0);
630 jindex = nlist->jindex;
632 shiftidx = nlist->shift;
634 shiftvec = fr->shift_vec[0];
635 fshift = fr->fshift[0];
636 facel = _mm_set1_pd(fr->epsfac);
637 charge = mdatoms->chargeA;
638 krf = _mm_set1_pd(fr->ic->k_rf);
639 krf2 = _mm_set1_pd(fr->ic->k_rf*2.0);
640 crf = _mm_set1_pd(fr->ic->c_rf);
641 nvdwtype = fr->ntype;
643 vdwtype = mdatoms->typeA;
645 /* Setup water-specific parameters */
646 inr = nlist->iinr[0];
647 iq1 = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+1]));
648 iq2 = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+2]));
649 iq3 = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+3]));
650 vdwioffset0 = 2*nvdwtype*vdwtype[inr+0];
652 /* Avoid stupid compiler warnings */
660 /* Start outer loop over neighborlists */
661 for(iidx=0; iidx<nri; iidx++)
663 /* Load shift vector for this list */
664 i_shift_offset = DIM*shiftidx[iidx];
666 /* Load limits for loop over neighbors */
667 j_index_start = jindex[iidx];
668 j_index_end = jindex[iidx+1];
670 /* Get outer coordinate index */
672 i_coord_offset = DIM*inr;
674 /* Load i particle coords and add shift vector */
675 gmx_mm_load_shift_and_4rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset,
676 &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2,&ix3,&iy3,&iz3);
678 fix0 = _mm_setzero_pd();
679 fiy0 = _mm_setzero_pd();
680 fiz0 = _mm_setzero_pd();
681 fix1 = _mm_setzero_pd();
682 fiy1 = _mm_setzero_pd();
683 fiz1 = _mm_setzero_pd();
684 fix2 = _mm_setzero_pd();
685 fiy2 = _mm_setzero_pd();
686 fiz2 = _mm_setzero_pd();
687 fix3 = _mm_setzero_pd();
688 fiy3 = _mm_setzero_pd();
689 fiz3 = _mm_setzero_pd();
691 /* Start inner kernel loop */
692 for(jidx=j_index_start; jidx<j_index_end-1; jidx+=2)
695 /* Get j neighbor index, and coordinate index */
698 j_coord_offsetA = DIM*jnrA;
699 j_coord_offsetB = DIM*jnrB;
701 /* load j atom coordinates */
702 gmx_mm_load_1rvec_2ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
705 /* Calculate displacement vector */
706 dx00 = _mm_sub_pd(ix0,jx0);
707 dy00 = _mm_sub_pd(iy0,jy0);
708 dz00 = _mm_sub_pd(iz0,jz0);
709 dx10 = _mm_sub_pd(ix1,jx0);
710 dy10 = _mm_sub_pd(iy1,jy0);
711 dz10 = _mm_sub_pd(iz1,jz0);
712 dx20 = _mm_sub_pd(ix2,jx0);
713 dy20 = _mm_sub_pd(iy2,jy0);
714 dz20 = _mm_sub_pd(iz2,jz0);
715 dx30 = _mm_sub_pd(ix3,jx0);
716 dy30 = _mm_sub_pd(iy3,jy0);
717 dz30 = _mm_sub_pd(iz3,jz0);
719 /* Calculate squared distance and things based on it */
720 rsq00 = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
721 rsq10 = gmx_mm_calc_rsq_pd(dx10,dy10,dz10);
722 rsq20 = gmx_mm_calc_rsq_pd(dx20,dy20,dz20);
723 rsq30 = gmx_mm_calc_rsq_pd(dx30,dy30,dz30);
725 rinv10 = gmx_mm_invsqrt_pd(rsq10);
726 rinv20 = gmx_mm_invsqrt_pd(rsq20);
727 rinv30 = gmx_mm_invsqrt_pd(rsq30);
729 rinvsq00 = gmx_mm_inv_pd(rsq00);
730 rinvsq10 = _mm_mul_pd(rinv10,rinv10);
731 rinvsq20 = _mm_mul_pd(rinv20,rinv20);
732 rinvsq30 = _mm_mul_pd(rinv30,rinv30);
734 /* Load parameters for j particles */
735 jq0 = gmx_mm_load_2real_swizzle_pd(charge+jnrA+0,charge+jnrB+0);
736 vdwjidx0A = 2*vdwtype[jnrA+0];
737 vdwjidx0B = 2*vdwtype[jnrB+0];
739 fjx0 = _mm_setzero_pd();
740 fjy0 = _mm_setzero_pd();
741 fjz0 = _mm_setzero_pd();
743 /**************************
744 * CALCULATE INTERACTIONS *
745 **************************/
747 /* Compute parameters for interactions between i and j atoms */
748 gmx_mm_load_2pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,
749 vdwparam+vdwioffset0+vdwjidx0B,&c6_00,&c12_00);
751 /* LENNARD-JONES DISPERSION/REPULSION */
753 rinvsix = _mm_mul_pd(_mm_mul_pd(rinvsq00,rinvsq00),rinvsq00);
754 fvdw = _mm_mul_pd(_mm_sub_pd(_mm_mul_pd(c12_00,rinvsix),c6_00),_mm_mul_pd(rinvsix,rinvsq00));
758 /* Calculate temporary vectorial force */
759 tx = _mm_mul_pd(fscal,dx00);
760 ty = _mm_mul_pd(fscal,dy00);
761 tz = _mm_mul_pd(fscal,dz00);
763 /* Update vectorial force */
764 fix0 = _mm_add_pd(fix0,tx);
765 fiy0 = _mm_add_pd(fiy0,ty);
766 fiz0 = _mm_add_pd(fiz0,tz);
768 fjx0 = _mm_add_pd(fjx0,tx);
769 fjy0 = _mm_add_pd(fjy0,ty);
770 fjz0 = _mm_add_pd(fjz0,tz);
772 /**************************
773 * CALCULATE INTERACTIONS *
774 **************************/
776 /* Compute parameters for interactions between i and j atoms */
777 qq10 = _mm_mul_pd(iq1,jq0);
779 /* REACTION-FIELD ELECTROSTATICS */
780 felec = _mm_mul_pd(qq10,_mm_sub_pd(_mm_mul_pd(rinv10,rinvsq10),krf2));
784 /* Calculate temporary vectorial force */
785 tx = _mm_mul_pd(fscal,dx10);
786 ty = _mm_mul_pd(fscal,dy10);
787 tz = _mm_mul_pd(fscal,dz10);
789 /* Update vectorial force */
790 fix1 = _mm_add_pd(fix1,tx);
791 fiy1 = _mm_add_pd(fiy1,ty);
792 fiz1 = _mm_add_pd(fiz1,tz);
794 fjx0 = _mm_add_pd(fjx0,tx);
795 fjy0 = _mm_add_pd(fjy0,ty);
796 fjz0 = _mm_add_pd(fjz0,tz);
798 /**************************
799 * CALCULATE INTERACTIONS *
800 **************************/
802 /* Compute parameters for interactions between i and j atoms */
803 qq20 = _mm_mul_pd(iq2,jq0);
805 /* REACTION-FIELD ELECTROSTATICS */
806 felec = _mm_mul_pd(qq20,_mm_sub_pd(_mm_mul_pd(rinv20,rinvsq20),krf2));
810 /* Calculate temporary vectorial force */
811 tx = _mm_mul_pd(fscal,dx20);
812 ty = _mm_mul_pd(fscal,dy20);
813 tz = _mm_mul_pd(fscal,dz20);
815 /* Update vectorial force */
816 fix2 = _mm_add_pd(fix2,tx);
817 fiy2 = _mm_add_pd(fiy2,ty);
818 fiz2 = _mm_add_pd(fiz2,tz);
820 fjx0 = _mm_add_pd(fjx0,tx);
821 fjy0 = _mm_add_pd(fjy0,ty);
822 fjz0 = _mm_add_pd(fjz0,tz);
824 /**************************
825 * CALCULATE INTERACTIONS *
826 **************************/
828 /* Compute parameters for interactions between i and j atoms */
829 qq30 = _mm_mul_pd(iq3,jq0);
831 /* REACTION-FIELD ELECTROSTATICS */
832 felec = _mm_mul_pd(qq30,_mm_sub_pd(_mm_mul_pd(rinv30,rinvsq30),krf2));
836 /* Calculate temporary vectorial force */
837 tx = _mm_mul_pd(fscal,dx30);
838 ty = _mm_mul_pd(fscal,dy30);
839 tz = _mm_mul_pd(fscal,dz30);
841 /* Update vectorial force */
842 fix3 = _mm_add_pd(fix3,tx);
843 fiy3 = _mm_add_pd(fiy3,ty);
844 fiz3 = _mm_add_pd(fiz3,tz);
846 fjx0 = _mm_add_pd(fjx0,tx);
847 fjy0 = _mm_add_pd(fjy0,ty);
848 fjz0 = _mm_add_pd(fjz0,tz);
850 gmx_mm_decrement_1rvec_2ptr_swizzle_pd(f+j_coord_offsetA,f+j_coord_offsetB,fjx0,fjy0,fjz0);
852 /* Inner loop uses 111 flops */
859 j_coord_offsetA = DIM*jnrA;
861 /* load j atom coordinates */
862 gmx_mm_load_1rvec_1ptr_swizzle_pd(x+j_coord_offsetA,
865 /* Calculate displacement vector */
866 dx00 = _mm_sub_pd(ix0,jx0);
867 dy00 = _mm_sub_pd(iy0,jy0);
868 dz00 = _mm_sub_pd(iz0,jz0);
869 dx10 = _mm_sub_pd(ix1,jx0);
870 dy10 = _mm_sub_pd(iy1,jy0);
871 dz10 = _mm_sub_pd(iz1,jz0);
872 dx20 = _mm_sub_pd(ix2,jx0);
873 dy20 = _mm_sub_pd(iy2,jy0);
874 dz20 = _mm_sub_pd(iz2,jz0);
875 dx30 = _mm_sub_pd(ix3,jx0);
876 dy30 = _mm_sub_pd(iy3,jy0);
877 dz30 = _mm_sub_pd(iz3,jz0);
879 /* Calculate squared distance and things based on it */
880 rsq00 = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
881 rsq10 = gmx_mm_calc_rsq_pd(dx10,dy10,dz10);
882 rsq20 = gmx_mm_calc_rsq_pd(dx20,dy20,dz20);
883 rsq30 = gmx_mm_calc_rsq_pd(dx30,dy30,dz30);
885 rinv10 = gmx_mm_invsqrt_pd(rsq10);
886 rinv20 = gmx_mm_invsqrt_pd(rsq20);
887 rinv30 = gmx_mm_invsqrt_pd(rsq30);
889 rinvsq00 = gmx_mm_inv_pd(rsq00);
890 rinvsq10 = _mm_mul_pd(rinv10,rinv10);
891 rinvsq20 = _mm_mul_pd(rinv20,rinv20);
892 rinvsq30 = _mm_mul_pd(rinv30,rinv30);
894 /* Load parameters for j particles */
895 jq0 = _mm_load_sd(charge+jnrA+0);
896 vdwjidx0A = 2*vdwtype[jnrA+0];
898 fjx0 = _mm_setzero_pd();
899 fjy0 = _mm_setzero_pd();
900 fjz0 = _mm_setzero_pd();
902 /**************************
903 * CALCULATE INTERACTIONS *
904 **************************/
906 /* Compute parameters for interactions between i and j atoms */
907 gmx_mm_load_1pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,&c6_00,&c12_00);
909 /* LENNARD-JONES DISPERSION/REPULSION */
911 rinvsix = _mm_mul_pd(_mm_mul_pd(rinvsq00,rinvsq00),rinvsq00);
912 fvdw = _mm_mul_pd(_mm_sub_pd(_mm_mul_pd(c12_00,rinvsix),c6_00),_mm_mul_pd(rinvsix,rinvsq00));
916 fscal = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
918 /* Calculate temporary vectorial force */
919 tx = _mm_mul_pd(fscal,dx00);
920 ty = _mm_mul_pd(fscal,dy00);
921 tz = _mm_mul_pd(fscal,dz00);
923 /* Update vectorial force */
924 fix0 = _mm_add_pd(fix0,tx);
925 fiy0 = _mm_add_pd(fiy0,ty);
926 fiz0 = _mm_add_pd(fiz0,tz);
928 fjx0 = _mm_add_pd(fjx0,tx);
929 fjy0 = _mm_add_pd(fjy0,ty);
930 fjz0 = _mm_add_pd(fjz0,tz);
932 /**************************
933 * CALCULATE INTERACTIONS *
934 **************************/
936 /* Compute parameters for interactions between i and j atoms */
937 qq10 = _mm_mul_pd(iq1,jq0);
939 /* REACTION-FIELD ELECTROSTATICS */
940 felec = _mm_mul_pd(qq10,_mm_sub_pd(_mm_mul_pd(rinv10,rinvsq10),krf2));
944 fscal = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
946 /* Calculate temporary vectorial force */
947 tx = _mm_mul_pd(fscal,dx10);
948 ty = _mm_mul_pd(fscal,dy10);
949 tz = _mm_mul_pd(fscal,dz10);
951 /* Update vectorial force */
952 fix1 = _mm_add_pd(fix1,tx);
953 fiy1 = _mm_add_pd(fiy1,ty);
954 fiz1 = _mm_add_pd(fiz1,tz);
956 fjx0 = _mm_add_pd(fjx0,tx);
957 fjy0 = _mm_add_pd(fjy0,ty);
958 fjz0 = _mm_add_pd(fjz0,tz);
960 /**************************
961 * CALCULATE INTERACTIONS *
962 **************************/
964 /* Compute parameters for interactions between i and j atoms */
965 qq20 = _mm_mul_pd(iq2,jq0);
967 /* REACTION-FIELD ELECTROSTATICS */
968 felec = _mm_mul_pd(qq20,_mm_sub_pd(_mm_mul_pd(rinv20,rinvsq20),krf2));
972 fscal = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
974 /* Calculate temporary vectorial force */
975 tx = _mm_mul_pd(fscal,dx20);
976 ty = _mm_mul_pd(fscal,dy20);
977 tz = _mm_mul_pd(fscal,dz20);
979 /* Update vectorial force */
980 fix2 = _mm_add_pd(fix2,tx);
981 fiy2 = _mm_add_pd(fiy2,ty);
982 fiz2 = _mm_add_pd(fiz2,tz);
984 fjx0 = _mm_add_pd(fjx0,tx);
985 fjy0 = _mm_add_pd(fjy0,ty);
986 fjz0 = _mm_add_pd(fjz0,tz);
988 /**************************
989 * CALCULATE INTERACTIONS *
990 **************************/
992 /* Compute parameters for interactions between i and j atoms */
993 qq30 = _mm_mul_pd(iq3,jq0);
995 /* REACTION-FIELD ELECTROSTATICS */
996 felec = _mm_mul_pd(qq30,_mm_sub_pd(_mm_mul_pd(rinv30,rinvsq30),krf2));
1000 fscal = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
1002 /* Calculate temporary vectorial force */
1003 tx = _mm_mul_pd(fscal,dx30);
1004 ty = _mm_mul_pd(fscal,dy30);
1005 tz = _mm_mul_pd(fscal,dz30);
1007 /* Update vectorial force */
1008 fix3 = _mm_add_pd(fix3,tx);
1009 fiy3 = _mm_add_pd(fiy3,ty);
1010 fiz3 = _mm_add_pd(fiz3,tz);
1012 fjx0 = _mm_add_pd(fjx0,tx);
1013 fjy0 = _mm_add_pd(fjy0,ty);
1014 fjz0 = _mm_add_pd(fjz0,tz);
1016 gmx_mm_decrement_1rvec_1ptr_swizzle_pd(f+j_coord_offsetA,fjx0,fjy0,fjz0);
1018 /* Inner loop uses 111 flops */
1021 /* End of innermost loop */
1023 gmx_mm_update_iforce_4atom_swizzle_pd(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,fix3,fiy3,fiz3,
1024 f+i_coord_offset,fshift+i_shift_offset);
1026 /* Increment number of inner iterations */
1027 inneriter += j_index_end - j_index_start;
1029 /* Outer loop uses 24 flops */
1032 /* Increment number of outer iterations */
1035 /* Update outer/inner flops */
1037 inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W4_F,outeriter*24 + inneriter*111);