2 * Note: this file was generated by the Gromacs avx_128_fma_double kernel generator.
4 * This source code is part of
8 * Copyright (c) 2001-2012, The GROMACS Development Team
10 * Gromacs is a library for molecular simulation and trajectory analysis,
11 * written by Erik Lindahl, David van der Spoel, Berk Hess, and others - for
12 * a full list of developers and information, check out http://www.gromacs.org
14 * This program is free software; you can redistribute it and/or modify it under
15 * the terms of the GNU Lesser General Public License as published by the Free
16 * Software Foundation; either version 2 of the License, or (at your option) any
19 * To help fund GROMACS development, we humbly ask that you cite
20 * the papers people have written on it - you can find them on the website.
28 #include "../nb_kernel.h"
29 #include "types/simple.h"
33 #include "gmx_math_x86_avx_128_fma_double.h"
34 #include "kernelutil_x86_avx_128_fma_double.h"
37 * Gromacs nonbonded kernel: nb_kernel_ElecCoul_VdwNone_GeomW4P1_VF_avx_128_fma_double
38 * Electrostatics interaction: Coulomb
39 * VdW interaction: None
40 * Geometry: Water4-Particle
41 * Calculate force/pot: PotentialAndForce
44 nb_kernel_ElecCoul_VdwNone_GeomW4P1_VF_avx_128_fma_double
45 (t_nblist * gmx_restrict nlist,
46 rvec * gmx_restrict xx,
47 rvec * gmx_restrict ff,
48 t_forcerec * gmx_restrict fr,
49 t_mdatoms * gmx_restrict mdatoms,
50 nb_kernel_data_t * gmx_restrict kernel_data,
51 t_nrnb * gmx_restrict nrnb)
53 /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
54 * just 0 for non-waters.
55 * Suffixes A,B refer to j loop unrolling done with SSE double precision, e.g. for the two different
56 * jnr indices corresponding to data put in the four positions in the SIMD register.
58 int i_shift_offset,i_coord_offset,outeriter,inneriter;
59 int j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
61 int j_coord_offsetA,j_coord_offsetB;
62 int *iinr,*jindex,*jjnr,*shiftidx,*gid;
64 real *shiftvec,*fshift,*x,*f;
65 __m128d tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
67 __m128d ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
69 __m128d ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
71 __m128d ix3,iy3,iz3,fix3,fiy3,fiz3,iq3,isai3;
72 int vdwjidx0A,vdwjidx0B;
73 __m128d jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
74 __m128d dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
75 __m128d dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
76 __m128d dx30,dy30,dz30,rsq30,rinv30,rinvsq30,r30,qq30,c6_30,c12_30;
77 __m128d velec,felec,velecsum,facel,crf,krf,krf2;
79 __m128d dummy_mask,cutoff_mask;
80 __m128d signbit = gmx_mm_castsi128_pd( _mm_set_epi32(0x80000000,0x00000000,0x80000000,0x00000000) );
81 __m128d one = _mm_set1_pd(1.0);
82 __m128d two = _mm_set1_pd(2.0);
88 jindex = nlist->jindex;
90 shiftidx = nlist->shift;
92 shiftvec = fr->shift_vec[0];
93 fshift = fr->fshift[0];
94 facel = _mm_set1_pd(fr->epsfac);
95 charge = mdatoms->chargeA;
97 /* Setup water-specific parameters */
99 iq1 = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+1]));
100 iq2 = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+2]));
101 iq3 = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+3]));
103 /* Avoid stupid compiler warnings */
111 /* Start outer loop over neighborlists */
112 for(iidx=0; iidx<nri; iidx++)
114 /* Load shift vector for this list */
115 i_shift_offset = DIM*shiftidx[iidx];
117 /* Load limits for loop over neighbors */
118 j_index_start = jindex[iidx];
119 j_index_end = jindex[iidx+1];
121 /* Get outer coordinate index */
123 i_coord_offset = DIM*inr;
125 /* Load i particle coords and add shift vector */
126 gmx_mm_load_shift_and_3rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset+DIM,
127 &ix1,&iy1,&iz1,&ix2,&iy2,&iz2,&ix3,&iy3,&iz3);
129 fix1 = _mm_setzero_pd();
130 fiy1 = _mm_setzero_pd();
131 fiz1 = _mm_setzero_pd();
132 fix2 = _mm_setzero_pd();
133 fiy2 = _mm_setzero_pd();
134 fiz2 = _mm_setzero_pd();
135 fix3 = _mm_setzero_pd();
136 fiy3 = _mm_setzero_pd();
137 fiz3 = _mm_setzero_pd();
139 /* Reset potential sums */
140 velecsum = _mm_setzero_pd();
142 /* Start inner kernel loop */
143 for(jidx=j_index_start; jidx<j_index_end-1; jidx+=2)
146 /* Get j neighbor index, and coordinate index */
149 j_coord_offsetA = DIM*jnrA;
150 j_coord_offsetB = DIM*jnrB;
152 /* load j atom coordinates */
153 gmx_mm_load_1rvec_2ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
156 /* Calculate displacement vector */
157 dx10 = _mm_sub_pd(ix1,jx0);
158 dy10 = _mm_sub_pd(iy1,jy0);
159 dz10 = _mm_sub_pd(iz1,jz0);
160 dx20 = _mm_sub_pd(ix2,jx0);
161 dy20 = _mm_sub_pd(iy2,jy0);
162 dz20 = _mm_sub_pd(iz2,jz0);
163 dx30 = _mm_sub_pd(ix3,jx0);
164 dy30 = _mm_sub_pd(iy3,jy0);
165 dz30 = _mm_sub_pd(iz3,jz0);
167 /* Calculate squared distance and things based on it */
168 rsq10 = gmx_mm_calc_rsq_pd(dx10,dy10,dz10);
169 rsq20 = gmx_mm_calc_rsq_pd(dx20,dy20,dz20);
170 rsq30 = gmx_mm_calc_rsq_pd(dx30,dy30,dz30);
172 rinv10 = gmx_mm_invsqrt_pd(rsq10);
173 rinv20 = gmx_mm_invsqrt_pd(rsq20);
174 rinv30 = gmx_mm_invsqrt_pd(rsq30);
176 rinvsq10 = _mm_mul_pd(rinv10,rinv10);
177 rinvsq20 = _mm_mul_pd(rinv20,rinv20);
178 rinvsq30 = _mm_mul_pd(rinv30,rinv30);
180 /* Load parameters for j particles */
181 jq0 = gmx_mm_load_2real_swizzle_pd(charge+jnrA+0,charge+jnrB+0);
183 fjx0 = _mm_setzero_pd();
184 fjy0 = _mm_setzero_pd();
185 fjz0 = _mm_setzero_pd();
187 /**************************
188 * CALCULATE INTERACTIONS *
189 **************************/
191 /* Compute parameters for interactions between i and j atoms */
192 qq10 = _mm_mul_pd(iq1,jq0);
194 /* COULOMB ELECTROSTATICS */
195 velec = _mm_mul_pd(qq10,rinv10);
196 felec = _mm_mul_pd(velec,rinvsq10);
198 /* Update potential sum for this i atom from the interaction with this j atom. */
199 velecsum = _mm_add_pd(velecsum,velec);
203 /* Update vectorial force */
204 fix1 = _mm_macc_pd(dx10,fscal,fix1);
205 fiy1 = _mm_macc_pd(dy10,fscal,fiy1);
206 fiz1 = _mm_macc_pd(dz10,fscal,fiz1);
208 fjx0 = _mm_macc_pd(dx10,fscal,fjx0);
209 fjy0 = _mm_macc_pd(dy10,fscal,fjy0);
210 fjz0 = _mm_macc_pd(dz10,fscal,fjz0);
212 /**************************
213 * CALCULATE INTERACTIONS *
214 **************************/
216 /* Compute parameters for interactions between i and j atoms */
217 qq20 = _mm_mul_pd(iq2,jq0);
219 /* COULOMB ELECTROSTATICS */
220 velec = _mm_mul_pd(qq20,rinv20);
221 felec = _mm_mul_pd(velec,rinvsq20);
223 /* Update potential sum for this i atom from the interaction with this j atom. */
224 velecsum = _mm_add_pd(velecsum,velec);
228 /* Update vectorial force */
229 fix2 = _mm_macc_pd(dx20,fscal,fix2);
230 fiy2 = _mm_macc_pd(dy20,fscal,fiy2);
231 fiz2 = _mm_macc_pd(dz20,fscal,fiz2);
233 fjx0 = _mm_macc_pd(dx20,fscal,fjx0);
234 fjy0 = _mm_macc_pd(dy20,fscal,fjy0);
235 fjz0 = _mm_macc_pd(dz20,fscal,fjz0);
237 /**************************
238 * CALCULATE INTERACTIONS *
239 **************************/
241 /* Compute parameters for interactions between i and j atoms */
242 qq30 = _mm_mul_pd(iq3,jq0);
244 /* COULOMB ELECTROSTATICS */
245 velec = _mm_mul_pd(qq30,rinv30);
246 felec = _mm_mul_pd(velec,rinvsq30);
248 /* Update potential sum for this i atom from the interaction with this j atom. */
249 velecsum = _mm_add_pd(velecsum,velec);
253 /* Update vectorial force */
254 fix3 = _mm_macc_pd(dx30,fscal,fix3);
255 fiy3 = _mm_macc_pd(dy30,fscal,fiy3);
256 fiz3 = _mm_macc_pd(dz30,fscal,fiz3);
258 fjx0 = _mm_macc_pd(dx30,fscal,fjx0);
259 fjy0 = _mm_macc_pd(dy30,fscal,fjy0);
260 fjz0 = _mm_macc_pd(dz30,fscal,fjz0);
262 gmx_mm_decrement_1rvec_2ptr_swizzle_pd(f+j_coord_offsetA,f+j_coord_offsetB,fjx0,fjy0,fjz0);
264 /* Inner loop uses 96 flops */
271 j_coord_offsetA = DIM*jnrA;
273 /* load j atom coordinates */
274 gmx_mm_load_1rvec_1ptr_swizzle_pd(x+j_coord_offsetA,
277 /* Calculate displacement vector */
278 dx10 = _mm_sub_pd(ix1,jx0);
279 dy10 = _mm_sub_pd(iy1,jy0);
280 dz10 = _mm_sub_pd(iz1,jz0);
281 dx20 = _mm_sub_pd(ix2,jx0);
282 dy20 = _mm_sub_pd(iy2,jy0);
283 dz20 = _mm_sub_pd(iz2,jz0);
284 dx30 = _mm_sub_pd(ix3,jx0);
285 dy30 = _mm_sub_pd(iy3,jy0);
286 dz30 = _mm_sub_pd(iz3,jz0);
288 /* Calculate squared distance and things based on it */
289 rsq10 = gmx_mm_calc_rsq_pd(dx10,dy10,dz10);
290 rsq20 = gmx_mm_calc_rsq_pd(dx20,dy20,dz20);
291 rsq30 = gmx_mm_calc_rsq_pd(dx30,dy30,dz30);
293 rinv10 = gmx_mm_invsqrt_pd(rsq10);
294 rinv20 = gmx_mm_invsqrt_pd(rsq20);
295 rinv30 = gmx_mm_invsqrt_pd(rsq30);
297 rinvsq10 = _mm_mul_pd(rinv10,rinv10);
298 rinvsq20 = _mm_mul_pd(rinv20,rinv20);
299 rinvsq30 = _mm_mul_pd(rinv30,rinv30);
301 /* Load parameters for j particles */
302 jq0 = _mm_load_sd(charge+jnrA+0);
304 fjx0 = _mm_setzero_pd();
305 fjy0 = _mm_setzero_pd();
306 fjz0 = _mm_setzero_pd();
308 /**************************
309 * CALCULATE INTERACTIONS *
310 **************************/
312 /* Compute parameters for interactions between i and j atoms */
313 qq10 = _mm_mul_pd(iq1,jq0);
315 /* COULOMB ELECTROSTATICS */
316 velec = _mm_mul_pd(qq10,rinv10);
317 felec = _mm_mul_pd(velec,rinvsq10);
319 /* Update potential sum for this i atom from the interaction with this j atom. */
320 velec = _mm_unpacklo_pd(velec,_mm_setzero_pd());
321 velecsum = _mm_add_pd(velecsum,velec);
325 fscal = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
327 /* Update vectorial force */
328 fix1 = _mm_macc_pd(dx10,fscal,fix1);
329 fiy1 = _mm_macc_pd(dy10,fscal,fiy1);
330 fiz1 = _mm_macc_pd(dz10,fscal,fiz1);
332 fjx0 = _mm_macc_pd(dx10,fscal,fjx0);
333 fjy0 = _mm_macc_pd(dy10,fscal,fjy0);
334 fjz0 = _mm_macc_pd(dz10,fscal,fjz0);
336 /**************************
337 * CALCULATE INTERACTIONS *
338 **************************/
340 /* Compute parameters for interactions between i and j atoms */
341 qq20 = _mm_mul_pd(iq2,jq0);
343 /* COULOMB ELECTROSTATICS */
344 velec = _mm_mul_pd(qq20,rinv20);
345 felec = _mm_mul_pd(velec,rinvsq20);
347 /* Update potential sum for this i atom from the interaction with this j atom. */
348 velec = _mm_unpacklo_pd(velec,_mm_setzero_pd());
349 velecsum = _mm_add_pd(velecsum,velec);
353 fscal = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
355 /* Update vectorial force */
356 fix2 = _mm_macc_pd(dx20,fscal,fix2);
357 fiy2 = _mm_macc_pd(dy20,fscal,fiy2);
358 fiz2 = _mm_macc_pd(dz20,fscal,fiz2);
360 fjx0 = _mm_macc_pd(dx20,fscal,fjx0);
361 fjy0 = _mm_macc_pd(dy20,fscal,fjy0);
362 fjz0 = _mm_macc_pd(dz20,fscal,fjz0);
364 /**************************
365 * CALCULATE INTERACTIONS *
366 **************************/
368 /* Compute parameters for interactions between i and j atoms */
369 qq30 = _mm_mul_pd(iq3,jq0);
371 /* COULOMB ELECTROSTATICS */
372 velec = _mm_mul_pd(qq30,rinv30);
373 felec = _mm_mul_pd(velec,rinvsq30);
375 /* Update potential sum for this i atom from the interaction with this j atom. */
376 velec = _mm_unpacklo_pd(velec,_mm_setzero_pd());
377 velecsum = _mm_add_pd(velecsum,velec);
381 fscal = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
383 /* Update vectorial force */
384 fix3 = _mm_macc_pd(dx30,fscal,fix3);
385 fiy3 = _mm_macc_pd(dy30,fscal,fiy3);
386 fiz3 = _mm_macc_pd(dz30,fscal,fiz3);
388 fjx0 = _mm_macc_pd(dx30,fscal,fjx0);
389 fjy0 = _mm_macc_pd(dy30,fscal,fjy0);
390 fjz0 = _mm_macc_pd(dz30,fscal,fjz0);
392 gmx_mm_decrement_1rvec_1ptr_swizzle_pd(f+j_coord_offsetA,fjx0,fjy0,fjz0);
394 /* Inner loop uses 96 flops */
397 /* End of innermost loop */
399 gmx_mm_update_iforce_3atom_swizzle_pd(fix1,fiy1,fiz1,fix2,fiy2,fiz2,fix3,fiy3,fiz3,
400 f+i_coord_offset+DIM,fshift+i_shift_offset);
403 /* Update potential energies */
404 gmx_mm_update_1pot_pd(velecsum,kernel_data->energygrp_elec+ggid);
406 /* Increment number of inner iterations */
407 inneriter += j_index_end - j_index_start;
409 /* Outer loop uses 19 flops */
412 /* Increment number of outer iterations */
415 /* Update outer/inner flops */
417 inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_W4_VF,outeriter*19 + inneriter*96);
420 * Gromacs nonbonded kernel: nb_kernel_ElecCoul_VdwNone_GeomW4P1_F_avx_128_fma_double
421 * Electrostatics interaction: Coulomb
422 * VdW interaction: None
423 * Geometry: Water4-Particle
424 * Calculate force/pot: Force
427 nb_kernel_ElecCoul_VdwNone_GeomW4P1_F_avx_128_fma_double
428 (t_nblist * gmx_restrict nlist,
429 rvec * gmx_restrict xx,
430 rvec * gmx_restrict ff,
431 t_forcerec * gmx_restrict fr,
432 t_mdatoms * gmx_restrict mdatoms,
433 nb_kernel_data_t * gmx_restrict kernel_data,
434 t_nrnb * gmx_restrict nrnb)
436 /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
437 * just 0 for non-waters.
438 * Suffixes A,B refer to j loop unrolling done with SSE double precision, e.g. for the two different
439 * jnr indices corresponding to data put in the four positions in the SIMD register.
441 int i_shift_offset,i_coord_offset,outeriter,inneriter;
442 int j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
444 int j_coord_offsetA,j_coord_offsetB;
445 int *iinr,*jindex,*jjnr,*shiftidx,*gid;
447 real *shiftvec,*fshift,*x,*f;
448 __m128d tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
450 __m128d ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
452 __m128d ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
454 __m128d ix3,iy3,iz3,fix3,fiy3,fiz3,iq3,isai3;
455 int vdwjidx0A,vdwjidx0B;
456 __m128d jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
457 __m128d dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
458 __m128d dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
459 __m128d dx30,dy30,dz30,rsq30,rinv30,rinvsq30,r30,qq30,c6_30,c12_30;
460 __m128d velec,felec,velecsum,facel,crf,krf,krf2;
462 __m128d dummy_mask,cutoff_mask;
463 __m128d signbit = gmx_mm_castsi128_pd( _mm_set_epi32(0x80000000,0x00000000,0x80000000,0x00000000) );
464 __m128d one = _mm_set1_pd(1.0);
465 __m128d two = _mm_set1_pd(2.0);
471 jindex = nlist->jindex;
473 shiftidx = nlist->shift;
475 shiftvec = fr->shift_vec[0];
476 fshift = fr->fshift[0];
477 facel = _mm_set1_pd(fr->epsfac);
478 charge = mdatoms->chargeA;
480 /* Setup water-specific parameters */
481 inr = nlist->iinr[0];
482 iq1 = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+1]));
483 iq2 = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+2]));
484 iq3 = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+3]));
486 /* Avoid stupid compiler warnings */
494 /* Start outer loop over neighborlists */
495 for(iidx=0; iidx<nri; iidx++)
497 /* Load shift vector for this list */
498 i_shift_offset = DIM*shiftidx[iidx];
500 /* Load limits for loop over neighbors */
501 j_index_start = jindex[iidx];
502 j_index_end = jindex[iidx+1];
504 /* Get outer coordinate index */
506 i_coord_offset = DIM*inr;
508 /* Load i particle coords and add shift vector */
509 gmx_mm_load_shift_and_3rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset+DIM,
510 &ix1,&iy1,&iz1,&ix2,&iy2,&iz2,&ix3,&iy3,&iz3);
512 fix1 = _mm_setzero_pd();
513 fiy1 = _mm_setzero_pd();
514 fiz1 = _mm_setzero_pd();
515 fix2 = _mm_setzero_pd();
516 fiy2 = _mm_setzero_pd();
517 fiz2 = _mm_setzero_pd();
518 fix3 = _mm_setzero_pd();
519 fiy3 = _mm_setzero_pd();
520 fiz3 = _mm_setzero_pd();
522 /* Start inner kernel loop */
523 for(jidx=j_index_start; jidx<j_index_end-1; jidx+=2)
526 /* Get j neighbor index, and coordinate index */
529 j_coord_offsetA = DIM*jnrA;
530 j_coord_offsetB = DIM*jnrB;
532 /* load j atom coordinates */
533 gmx_mm_load_1rvec_2ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
536 /* Calculate displacement vector */
537 dx10 = _mm_sub_pd(ix1,jx0);
538 dy10 = _mm_sub_pd(iy1,jy0);
539 dz10 = _mm_sub_pd(iz1,jz0);
540 dx20 = _mm_sub_pd(ix2,jx0);
541 dy20 = _mm_sub_pd(iy2,jy0);
542 dz20 = _mm_sub_pd(iz2,jz0);
543 dx30 = _mm_sub_pd(ix3,jx0);
544 dy30 = _mm_sub_pd(iy3,jy0);
545 dz30 = _mm_sub_pd(iz3,jz0);
547 /* Calculate squared distance and things based on it */
548 rsq10 = gmx_mm_calc_rsq_pd(dx10,dy10,dz10);
549 rsq20 = gmx_mm_calc_rsq_pd(dx20,dy20,dz20);
550 rsq30 = gmx_mm_calc_rsq_pd(dx30,dy30,dz30);
552 rinv10 = gmx_mm_invsqrt_pd(rsq10);
553 rinv20 = gmx_mm_invsqrt_pd(rsq20);
554 rinv30 = gmx_mm_invsqrt_pd(rsq30);
556 rinvsq10 = _mm_mul_pd(rinv10,rinv10);
557 rinvsq20 = _mm_mul_pd(rinv20,rinv20);
558 rinvsq30 = _mm_mul_pd(rinv30,rinv30);
560 /* Load parameters for j particles */
561 jq0 = gmx_mm_load_2real_swizzle_pd(charge+jnrA+0,charge+jnrB+0);
563 fjx0 = _mm_setzero_pd();
564 fjy0 = _mm_setzero_pd();
565 fjz0 = _mm_setzero_pd();
567 /**************************
568 * CALCULATE INTERACTIONS *
569 **************************/
571 /* Compute parameters for interactions between i and j atoms */
572 qq10 = _mm_mul_pd(iq1,jq0);
574 /* COULOMB ELECTROSTATICS */
575 velec = _mm_mul_pd(qq10,rinv10);
576 felec = _mm_mul_pd(velec,rinvsq10);
580 /* Update vectorial force */
581 fix1 = _mm_macc_pd(dx10,fscal,fix1);
582 fiy1 = _mm_macc_pd(dy10,fscal,fiy1);
583 fiz1 = _mm_macc_pd(dz10,fscal,fiz1);
585 fjx0 = _mm_macc_pd(dx10,fscal,fjx0);
586 fjy0 = _mm_macc_pd(dy10,fscal,fjy0);
587 fjz0 = _mm_macc_pd(dz10,fscal,fjz0);
589 /**************************
590 * CALCULATE INTERACTIONS *
591 **************************/
593 /* Compute parameters for interactions between i and j atoms */
594 qq20 = _mm_mul_pd(iq2,jq0);
596 /* COULOMB ELECTROSTATICS */
597 velec = _mm_mul_pd(qq20,rinv20);
598 felec = _mm_mul_pd(velec,rinvsq20);
602 /* Update vectorial force */
603 fix2 = _mm_macc_pd(dx20,fscal,fix2);
604 fiy2 = _mm_macc_pd(dy20,fscal,fiy2);
605 fiz2 = _mm_macc_pd(dz20,fscal,fiz2);
607 fjx0 = _mm_macc_pd(dx20,fscal,fjx0);
608 fjy0 = _mm_macc_pd(dy20,fscal,fjy0);
609 fjz0 = _mm_macc_pd(dz20,fscal,fjz0);
611 /**************************
612 * CALCULATE INTERACTIONS *
613 **************************/
615 /* Compute parameters for interactions between i and j atoms */
616 qq30 = _mm_mul_pd(iq3,jq0);
618 /* COULOMB ELECTROSTATICS */
619 velec = _mm_mul_pd(qq30,rinv30);
620 felec = _mm_mul_pd(velec,rinvsq30);
624 /* Update vectorial force */
625 fix3 = _mm_macc_pd(dx30,fscal,fix3);
626 fiy3 = _mm_macc_pd(dy30,fscal,fiy3);
627 fiz3 = _mm_macc_pd(dz30,fscal,fiz3);
629 fjx0 = _mm_macc_pd(dx30,fscal,fjx0);
630 fjy0 = _mm_macc_pd(dy30,fscal,fjy0);
631 fjz0 = _mm_macc_pd(dz30,fscal,fjz0);
633 gmx_mm_decrement_1rvec_2ptr_swizzle_pd(f+j_coord_offsetA,f+j_coord_offsetB,fjx0,fjy0,fjz0);
635 /* Inner loop uses 93 flops */
642 j_coord_offsetA = DIM*jnrA;
644 /* load j atom coordinates */
645 gmx_mm_load_1rvec_1ptr_swizzle_pd(x+j_coord_offsetA,
648 /* Calculate displacement vector */
649 dx10 = _mm_sub_pd(ix1,jx0);
650 dy10 = _mm_sub_pd(iy1,jy0);
651 dz10 = _mm_sub_pd(iz1,jz0);
652 dx20 = _mm_sub_pd(ix2,jx0);
653 dy20 = _mm_sub_pd(iy2,jy0);
654 dz20 = _mm_sub_pd(iz2,jz0);
655 dx30 = _mm_sub_pd(ix3,jx0);
656 dy30 = _mm_sub_pd(iy3,jy0);
657 dz30 = _mm_sub_pd(iz3,jz0);
659 /* Calculate squared distance and things based on it */
660 rsq10 = gmx_mm_calc_rsq_pd(dx10,dy10,dz10);
661 rsq20 = gmx_mm_calc_rsq_pd(dx20,dy20,dz20);
662 rsq30 = gmx_mm_calc_rsq_pd(dx30,dy30,dz30);
664 rinv10 = gmx_mm_invsqrt_pd(rsq10);
665 rinv20 = gmx_mm_invsqrt_pd(rsq20);
666 rinv30 = gmx_mm_invsqrt_pd(rsq30);
668 rinvsq10 = _mm_mul_pd(rinv10,rinv10);
669 rinvsq20 = _mm_mul_pd(rinv20,rinv20);
670 rinvsq30 = _mm_mul_pd(rinv30,rinv30);
672 /* Load parameters for j particles */
673 jq0 = _mm_load_sd(charge+jnrA+0);
675 fjx0 = _mm_setzero_pd();
676 fjy0 = _mm_setzero_pd();
677 fjz0 = _mm_setzero_pd();
679 /**************************
680 * CALCULATE INTERACTIONS *
681 **************************/
683 /* Compute parameters for interactions between i and j atoms */
684 qq10 = _mm_mul_pd(iq1,jq0);
686 /* COULOMB ELECTROSTATICS */
687 velec = _mm_mul_pd(qq10,rinv10);
688 felec = _mm_mul_pd(velec,rinvsq10);
692 fscal = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
694 /* Update vectorial force */
695 fix1 = _mm_macc_pd(dx10,fscal,fix1);
696 fiy1 = _mm_macc_pd(dy10,fscal,fiy1);
697 fiz1 = _mm_macc_pd(dz10,fscal,fiz1);
699 fjx0 = _mm_macc_pd(dx10,fscal,fjx0);
700 fjy0 = _mm_macc_pd(dy10,fscal,fjy0);
701 fjz0 = _mm_macc_pd(dz10,fscal,fjz0);
703 /**************************
704 * CALCULATE INTERACTIONS *
705 **************************/
707 /* Compute parameters for interactions between i and j atoms */
708 qq20 = _mm_mul_pd(iq2,jq0);
710 /* COULOMB ELECTROSTATICS */
711 velec = _mm_mul_pd(qq20,rinv20);
712 felec = _mm_mul_pd(velec,rinvsq20);
716 fscal = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
718 /* Update vectorial force */
719 fix2 = _mm_macc_pd(dx20,fscal,fix2);
720 fiy2 = _mm_macc_pd(dy20,fscal,fiy2);
721 fiz2 = _mm_macc_pd(dz20,fscal,fiz2);
723 fjx0 = _mm_macc_pd(dx20,fscal,fjx0);
724 fjy0 = _mm_macc_pd(dy20,fscal,fjy0);
725 fjz0 = _mm_macc_pd(dz20,fscal,fjz0);
727 /**************************
728 * CALCULATE INTERACTIONS *
729 **************************/
731 /* Compute parameters for interactions between i and j atoms */
732 qq30 = _mm_mul_pd(iq3,jq0);
734 /* COULOMB ELECTROSTATICS */
735 velec = _mm_mul_pd(qq30,rinv30);
736 felec = _mm_mul_pd(velec,rinvsq30);
740 fscal = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
742 /* Update vectorial force */
743 fix3 = _mm_macc_pd(dx30,fscal,fix3);
744 fiy3 = _mm_macc_pd(dy30,fscal,fiy3);
745 fiz3 = _mm_macc_pd(dz30,fscal,fiz3);
747 fjx0 = _mm_macc_pd(dx30,fscal,fjx0);
748 fjy0 = _mm_macc_pd(dy30,fscal,fjy0);
749 fjz0 = _mm_macc_pd(dz30,fscal,fjz0);
751 gmx_mm_decrement_1rvec_1ptr_swizzle_pd(f+j_coord_offsetA,fjx0,fjy0,fjz0);
753 /* Inner loop uses 93 flops */
756 /* End of innermost loop */
758 gmx_mm_update_iforce_3atom_swizzle_pd(fix1,fiy1,fiz1,fix2,fiy2,fiz2,fix3,fiy3,fiz3,
759 f+i_coord_offset+DIM,fshift+i_shift_offset);
761 /* Increment number of inner iterations */
762 inneriter += j_index_end - j_index_start;
764 /* Outer loop uses 18 flops */
767 /* Increment number of outer iterations */
770 /* Update outer/inner flops */
772 inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_W4_F,outeriter*18 + inneriter*93);