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_ElecRF_VdwNone_GeomW3P1_VF_avx_128_fma_double
38 * Electrostatics interaction: ReactionField
39 * VdW interaction: None
40 * Geometry: Water3-Particle
41 * Calculate force/pot: PotentialAndForce
44 nb_kernel_ElecRF_VdwNone_GeomW3P1_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 ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
69 __m128d ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
71 __m128d ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
72 int vdwjidx0A,vdwjidx0B;
73 __m128d jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
74 __m128d dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
75 __m128d dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
76 __m128d dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
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;
96 krf = _mm_set1_pd(fr->ic->k_rf);
97 krf2 = _mm_set1_pd(fr->ic->k_rf*2.0);
98 crf = _mm_set1_pd(fr->ic->c_rf);
100 /* Setup water-specific parameters */
101 inr = nlist->iinr[0];
102 iq0 = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+0]));
103 iq1 = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+1]));
104 iq2 = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+2]));
106 /* Avoid stupid compiler warnings */
114 /* Start outer loop over neighborlists */
115 for(iidx=0; iidx<nri; iidx++)
117 /* Load shift vector for this list */
118 i_shift_offset = DIM*shiftidx[iidx];
120 /* Load limits for loop over neighbors */
121 j_index_start = jindex[iidx];
122 j_index_end = jindex[iidx+1];
124 /* Get outer coordinate index */
126 i_coord_offset = DIM*inr;
128 /* Load i particle coords and add shift vector */
129 gmx_mm_load_shift_and_3rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset,
130 &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2);
132 fix0 = _mm_setzero_pd();
133 fiy0 = _mm_setzero_pd();
134 fiz0 = _mm_setzero_pd();
135 fix1 = _mm_setzero_pd();
136 fiy1 = _mm_setzero_pd();
137 fiz1 = _mm_setzero_pd();
138 fix2 = _mm_setzero_pd();
139 fiy2 = _mm_setzero_pd();
140 fiz2 = _mm_setzero_pd();
142 /* Reset potential sums */
143 velecsum = _mm_setzero_pd();
145 /* Start inner kernel loop */
146 for(jidx=j_index_start; jidx<j_index_end-1; jidx+=2)
149 /* Get j neighbor index, and coordinate index */
152 j_coord_offsetA = DIM*jnrA;
153 j_coord_offsetB = DIM*jnrB;
155 /* load j atom coordinates */
156 gmx_mm_load_1rvec_2ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
159 /* Calculate displacement vector */
160 dx00 = _mm_sub_pd(ix0,jx0);
161 dy00 = _mm_sub_pd(iy0,jy0);
162 dz00 = _mm_sub_pd(iz0,jz0);
163 dx10 = _mm_sub_pd(ix1,jx0);
164 dy10 = _mm_sub_pd(iy1,jy0);
165 dz10 = _mm_sub_pd(iz1,jz0);
166 dx20 = _mm_sub_pd(ix2,jx0);
167 dy20 = _mm_sub_pd(iy2,jy0);
168 dz20 = _mm_sub_pd(iz2,jz0);
170 /* Calculate squared distance and things based on it */
171 rsq00 = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
172 rsq10 = gmx_mm_calc_rsq_pd(dx10,dy10,dz10);
173 rsq20 = gmx_mm_calc_rsq_pd(dx20,dy20,dz20);
175 rinv00 = gmx_mm_invsqrt_pd(rsq00);
176 rinv10 = gmx_mm_invsqrt_pd(rsq10);
177 rinv20 = gmx_mm_invsqrt_pd(rsq20);
179 rinvsq00 = _mm_mul_pd(rinv00,rinv00);
180 rinvsq10 = _mm_mul_pd(rinv10,rinv10);
181 rinvsq20 = _mm_mul_pd(rinv20,rinv20);
183 /* Load parameters for j particles */
184 jq0 = gmx_mm_load_2real_swizzle_pd(charge+jnrA+0,charge+jnrB+0);
186 fjx0 = _mm_setzero_pd();
187 fjy0 = _mm_setzero_pd();
188 fjz0 = _mm_setzero_pd();
190 /**************************
191 * CALCULATE INTERACTIONS *
192 **************************/
194 /* Compute parameters for interactions between i and j atoms */
195 qq00 = _mm_mul_pd(iq0,jq0);
197 /* REACTION-FIELD ELECTROSTATICS */
198 velec = _mm_mul_pd(qq00,_mm_sub_pd(_mm_macc_pd(krf,rsq00,rinv00),crf));
199 felec = _mm_mul_pd(qq00,_mm_msub_pd(rinv00,rinvsq00,krf2));
201 /* Update potential sum for this i atom from the interaction with this j atom. */
202 velecsum = _mm_add_pd(velecsum,velec);
206 /* Update vectorial force */
207 fix0 = _mm_macc_pd(dx00,fscal,fix0);
208 fiy0 = _mm_macc_pd(dy00,fscal,fiy0);
209 fiz0 = _mm_macc_pd(dz00,fscal,fiz0);
211 fjx0 = _mm_macc_pd(dx00,fscal,fjx0);
212 fjy0 = _mm_macc_pd(dy00,fscal,fjy0);
213 fjz0 = _mm_macc_pd(dz00,fscal,fjz0);
215 /**************************
216 * CALCULATE INTERACTIONS *
217 **************************/
219 /* Compute parameters for interactions between i and j atoms */
220 qq10 = _mm_mul_pd(iq1,jq0);
222 /* REACTION-FIELD ELECTROSTATICS */
223 velec = _mm_mul_pd(qq10,_mm_sub_pd(_mm_macc_pd(krf,rsq10,rinv10),crf));
224 felec = _mm_mul_pd(qq10,_mm_msub_pd(rinv10,rinvsq10,krf2));
226 /* Update potential sum for this i atom from the interaction with this j atom. */
227 velecsum = _mm_add_pd(velecsum,velec);
231 /* Update vectorial force */
232 fix1 = _mm_macc_pd(dx10,fscal,fix1);
233 fiy1 = _mm_macc_pd(dy10,fscal,fiy1);
234 fiz1 = _mm_macc_pd(dz10,fscal,fiz1);
236 fjx0 = _mm_macc_pd(dx10,fscal,fjx0);
237 fjy0 = _mm_macc_pd(dy10,fscal,fjy0);
238 fjz0 = _mm_macc_pd(dz10,fscal,fjz0);
240 /**************************
241 * CALCULATE INTERACTIONS *
242 **************************/
244 /* Compute parameters for interactions between i and j atoms */
245 qq20 = _mm_mul_pd(iq2,jq0);
247 /* REACTION-FIELD ELECTROSTATICS */
248 velec = _mm_mul_pd(qq20,_mm_sub_pd(_mm_macc_pd(krf,rsq20,rinv20),crf));
249 felec = _mm_mul_pd(qq20,_mm_msub_pd(rinv20,rinvsq20,krf2));
251 /* Update potential sum for this i atom from the interaction with this j atom. */
252 velecsum = _mm_add_pd(velecsum,velec);
256 /* Update vectorial force */
257 fix2 = _mm_macc_pd(dx20,fscal,fix2);
258 fiy2 = _mm_macc_pd(dy20,fscal,fiy2);
259 fiz2 = _mm_macc_pd(dz20,fscal,fiz2);
261 fjx0 = _mm_macc_pd(dx20,fscal,fjx0);
262 fjy0 = _mm_macc_pd(dy20,fscal,fjy0);
263 fjz0 = _mm_macc_pd(dz20,fscal,fjz0);
265 gmx_mm_decrement_1rvec_2ptr_swizzle_pd(f+j_coord_offsetA,f+j_coord_offsetB,fjx0,fjy0,fjz0);
267 /* Inner loop uses 108 flops */
274 j_coord_offsetA = DIM*jnrA;
276 /* load j atom coordinates */
277 gmx_mm_load_1rvec_1ptr_swizzle_pd(x+j_coord_offsetA,
280 /* Calculate displacement vector */
281 dx00 = _mm_sub_pd(ix0,jx0);
282 dy00 = _mm_sub_pd(iy0,jy0);
283 dz00 = _mm_sub_pd(iz0,jz0);
284 dx10 = _mm_sub_pd(ix1,jx0);
285 dy10 = _mm_sub_pd(iy1,jy0);
286 dz10 = _mm_sub_pd(iz1,jz0);
287 dx20 = _mm_sub_pd(ix2,jx0);
288 dy20 = _mm_sub_pd(iy2,jy0);
289 dz20 = _mm_sub_pd(iz2,jz0);
291 /* Calculate squared distance and things based on it */
292 rsq00 = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
293 rsq10 = gmx_mm_calc_rsq_pd(dx10,dy10,dz10);
294 rsq20 = gmx_mm_calc_rsq_pd(dx20,dy20,dz20);
296 rinv00 = gmx_mm_invsqrt_pd(rsq00);
297 rinv10 = gmx_mm_invsqrt_pd(rsq10);
298 rinv20 = gmx_mm_invsqrt_pd(rsq20);
300 rinvsq00 = _mm_mul_pd(rinv00,rinv00);
301 rinvsq10 = _mm_mul_pd(rinv10,rinv10);
302 rinvsq20 = _mm_mul_pd(rinv20,rinv20);
304 /* Load parameters for j particles */
305 jq0 = _mm_load_sd(charge+jnrA+0);
307 fjx0 = _mm_setzero_pd();
308 fjy0 = _mm_setzero_pd();
309 fjz0 = _mm_setzero_pd();
311 /**************************
312 * CALCULATE INTERACTIONS *
313 **************************/
315 /* Compute parameters for interactions between i and j atoms */
316 qq00 = _mm_mul_pd(iq0,jq0);
318 /* REACTION-FIELD ELECTROSTATICS */
319 velec = _mm_mul_pd(qq00,_mm_sub_pd(_mm_macc_pd(krf,rsq00,rinv00),crf));
320 felec = _mm_mul_pd(qq00,_mm_msub_pd(rinv00,rinvsq00,krf2));
322 /* Update potential sum for this i atom from the interaction with this j atom. */
323 velec = _mm_unpacklo_pd(velec,_mm_setzero_pd());
324 velecsum = _mm_add_pd(velecsum,velec);
328 fscal = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
330 /* Update vectorial force */
331 fix0 = _mm_macc_pd(dx00,fscal,fix0);
332 fiy0 = _mm_macc_pd(dy00,fscal,fiy0);
333 fiz0 = _mm_macc_pd(dz00,fscal,fiz0);
335 fjx0 = _mm_macc_pd(dx00,fscal,fjx0);
336 fjy0 = _mm_macc_pd(dy00,fscal,fjy0);
337 fjz0 = _mm_macc_pd(dz00,fscal,fjz0);
339 /**************************
340 * CALCULATE INTERACTIONS *
341 **************************/
343 /* Compute parameters for interactions between i and j atoms */
344 qq10 = _mm_mul_pd(iq1,jq0);
346 /* REACTION-FIELD ELECTROSTATICS */
347 velec = _mm_mul_pd(qq10,_mm_sub_pd(_mm_macc_pd(krf,rsq10,rinv10),crf));
348 felec = _mm_mul_pd(qq10,_mm_msub_pd(rinv10,rinvsq10,krf2));
350 /* Update potential sum for this i atom from the interaction with this j atom. */
351 velec = _mm_unpacklo_pd(velec,_mm_setzero_pd());
352 velecsum = _mm_add_pd(velecsum,velec);
356 fscal = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
358 /* Update vectorial force */
359 fix1 = _mm_macc_pd(dx10,fscal,fix1);
360 fiy1 = _mm_macc_pd(dy10,fscal,fiy1);
361 fiz1 = _mm_macc_pd(dz10,fscal,fiz1);
363 fjx0 = _mm_macc_pd(dx10,fscal,fjx0);
364 fjy0 = _mm_macc_pd(dy10,fscal,fjy0);
365 fjz0 = _mm_macc_pd(dz10,fscal,fjz0);
367 /**************************
368 * CALCULATE INTERACTIONS *
369 **************************/
371 /* Compute parameters for interactions between i and j atoms */
372 qq20 = _mm_mul_pd(iq2,jq0);
374 /* REACTION-FIELD ELECTROSTATICS */
375 velec = _mm_mul_pd(qq20,_mm_sub_pd(_mm_macc_pd(krf,rsq20,rinv20),crf));
376 felec = _mm_mul_pd(qq20,_mm_msub_pd(rinv20,rinvsq20,krf2));
378 /* Update potential sum for this i atom from the interaction with this j atom. */
379 velec = _mm_unpacklo_pd(velec,_mm_setzero_pd());
380 velecsum = _mm_add_pd(velecsum,velec);
384 fscal = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
386 /* Update vectorial force */
387 fix2 = _mm_macc_pd(dx20,fscal,fix2);
388 fiy2 = _mm_macc_pd(dy20,fscal,fiy2);
389 fiz2 = _mm_macc_pd(dz20,fscal,fiz2);
391 fjx0 = _mm_macc_pd(dx20,fscal,fjx0);
392 fjy0 = _mm_macc_pd(dy20,fscal,fjy0);
393 fjz0 = _mm_macc_pd(dz20,fscal,fjz0);
395 gmx_mm_decrement_1rvec_1ptr_swizzle_pd(f+j_coord_offsetA,fjx0,fjy0,fjz0);
397 /* Inner loop uses 108 flops */
400 /* End of innermost loop */
402 gmx_mm_update_iforce_3atom_swizzle_pd(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,
403 f+i_coord_offset,fshift+i_shift_offset);
406 /* Update potential energies */
407 gmx_mm_update_1pot_pd(velecsum,kernel_data->energygrp_elec+ggid);
409 /* Increment number of inner iterations */
410 inneriter += j_index_end - j_index_start;
412 /* Outer loop uses 19 flops */
415 /* Increment number of outer iterations */
418 /* Update outer/inner flops */
420 inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_W3_VF,outeriter*19 + inneriter*108);
423 * Gromacs nonbonded kernel: nb_kernel_ElecRF_VdwNone_GeomW3P1_F_avx_128_fma_double
424 * Electrostatics interaction: ReactionField
425 * VdW interaction: None
426 * Geometry: Water3-Particle
427 * Calculate force/pot: Force
430 nb_kernel_ElecRF_VdwNone_GeomW3P1_F_avx_128_fma_double
431 (t_nblist * gmx_restrict nlist,
432 rvec * gmx_restrict xx,
433 rvec * gmx_restrict ff,
434 t_forcerec * gmx_restrict fr,
435 t_mdatoms * gmx_restrict mdatoms,
436 nb_kernel_data_t * gmx_restrict kernel_data,
437 t_nrnb * gmx_restrict nrnb)
439 /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
440 * just 0 for non-waters.
441 * Suffixes A,B refer to j loop unrolling done with SSE double precision, e.g. for the two different
442 * jnr indices corresponding to data put in the four positions in the SIMD register.
444 int i_shift_offset,i_coord_offset,outeriter,inneriter;
445 int j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
447 int j_coord_offsetA,j_coord_offsetB;
448 int *iinr,*jindex,*jjnr,*shiftidx,*gid;
450 real *shiftvec,*fshift,*x,*f;
451 __m128d tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
453 __m128d ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
455 __m128d ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
457 __m128d ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
458 int vdwjidx0A,vdwjidx0B;
459 __m128d jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
460 __m128d dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
461 __m128d dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
462 __m128d dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
463 __m128d velec,felec,velecsum,facel,crf,krf,krf2;
465 __m128d dummy_mask,cutoff_mask;
466 __m128d signbit = gmx_mm_castsi128_pd( _mm_set_epi32(0x80000000,0x00000000,0x80000000,0x00000000) );
467 __m128d one = _mm_set1_pd(1.0);
468 __m128d two = _mm_set1_pd(2.0);
474 jindex = nlist->jindex;
476 shiftidx = nlist->shift;
478 shiftvec = fr->shift_vec[0];
479 fshift = fr->fshift[0];
480 facel = _mm_set1_pd(fr->epsfac);
481 charge = mdatoms->chargeA;
482 krf = _mm_set1_pd(fr->ic->k_rf);
483 krf2 = _mm_set1_pd(fr->ic->k_rf*2.0);
484 crf = _mm_set1_pd(fr->ic->c_rf);
486 /* Setup water-specific parameters */
487 inr = nlist->iinr[0];
488 iq0 = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+0]));
489 iq1 = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+1]));
490 iq2 = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+2]));
492 /* Avoid stupid compiler warnings */
500 /* Start outer loop over neighborlists */
501 for(iidx=0; iidx<nri; iidx++)
503 /* Load shift vector for this list */
504 i_shift_offset = DIM*shiftidx[iidx];
506 /* Load limits for loop over neighbors */
507 j_index_start = jindex[iidx];
508 j_index_end = jindex[iidx+1];
510 /* Get outer coordinate index */
512 i_coord_offset = DIM*inr;
514 /* Load i particle coords and add shift vector */
515 gmx_mm_load_shift_and_3rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset,
516 &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2);
518 fix0 = _mm_setzero_pd();
519 fiy0 = _mm_setzero_pd();
520 fiz0 = _mm_setzero_pd();
521 fix1 = _mm_setzero_pd();
522 fiy1 = _mm_setzero_pd();
523 fiz1 = _mm_setzero_pd();
524 fix2 = _mm_setzero_pd();
525 fiy2 = _mm_setzero_pd();
526 fiz2 = _mm_setzero_pd();
528 /* Start inner kernel loop */
529 for(jidx=j_index_start; jidx<j_index_end-1; jidx+=2)
532 /* Get j neighbor index, and coordinate index */
535 j_coord_offsetA = DIM*jnrA;
536 j_coord_offsetB = DIM*jnrB;
538 /* load j atom coordinates */
539 gmx_mm_load_1rvec_2ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
542 /* Calculate displacement vector */
543 dx00 = _mm_sub_pd(ix0,jx0);
544 dy00 = _mm_sub_pd(iy0,jy0);
545 dz00 = _mm_sub_pd(iz0,jz0);
546 dx10 = _mm_sub_pd(ix1,jx0);
547 dy10 = _mm_sub_pd(iy1,jy0);
548 dz10 = _mm_sub_pd(iz1,jz0);
549 dx20 = _mm_sub_pd(ix2,jx0);
550 dy20 = _mm_sub_pd(iy2,jy0);
551 dz20 = _mm_sub_pd(iz2,jz0);
553 /* Calculate squared distance and things based on it */
554 rsq00 = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
555 rsq10 = gmx_mm_calc_rsq_pd(dx10,dy10,dz10);
556 rsq20 = gmx_mm_calc_rsq_pd(dx20,dy20,dz20);
558 rinv00 = gmx_mm_invsqrt_pd(rsq00);
559 rinv10 = gmx_mm_invsqrt_pd(rsq10);
560 rinv20 = gmx_mm_invsqrt_pd(rsq20);
562 rinvsq00 = _mm_mul_pd(rinv00,rinv00);
563 rinvsq10 = _mm_mul_pd(rinv10,rinv10);
564 rinvsq20 = _mm_mul_pd(rinv20,rinv20);
566 /* Load parameters for j particles */
567 jq0 = gmx_mm_load_2real_swizzle_pd(charge+jnrA+0,charge+jnrB+0);
569 fjx0 = _mm_setzero_pd();
570 fjy0 = _mm_setzero_pd();
571 fjz0 = _mm_setzero_pd();
573 /**************************
574 * CALCULATE INTERACTIONS *
575 **************************/
577 /* Compute parameters for interactions between i and j atoms */
578 qq00 = _mm_mul_pd(iq0,jq0);
580 /* REACTION-FIELD ELECTROSTATICS */
581 felec = _mm_mul_pd(qq00,_mm_msub_pd(rinv00,rinvsq00,krf2));
585 /* Update vectorial force */
586 fix0 = _mm_macc_pd(dx00,fscal,fix0);
587 fiy0 = _mm_macc_pd(dy00,fscal,fiy0);
588 fiz0 = _mm_macc_pd(dz00,fscal,fiz0);
590 fjx0 = _mm_macc_pd(dx00,fscal,fjx0);
591 fjy0 = _mm_macc_pd(dy00,fscal,fjy0);
592 fjz0 = _mm_macc_pd(dz00,fscal,fjz0);
594 /**************************
595 * CALCULATE INTERACTIONS *
596 **************************/
598 /* Compute parameters for interactions between i and j atoms */
599 qq10 = _mm_mul_pd(iq1,jq0);
601 /* REACTION-FIELD ELECTROSTATICS */
602 felec = _mm_mul_pd(qq10,_mm_msub_pd(rinv10,rinvsq10,krf2));
606 /* Update vectorial force */
607 fix1 = _mm_macc_pd(dx10,fscal,fix1);
608 fiy1 = _mm_macc_pd(dy10,fscal,fiy1);
609 fiz1 = _mm_macc_pd(dz10,fscal,fiz1);
611 fjx0 = _mm_macc_pd(dx10,fscal,fjx0);
612 fjy0 = _mm_macc_pd(dy10,fscal,fjy0);
613 fjz0 = _mm_macc_pd(dz10,fscal,fjz0);
615 /**************************
616 * CALCULATE INTERACTIONS *
617 **************************/
619 /* Compute parameters for interactions between i and j atoms */
620 qq20 = _mm_mul_pd(iq2,jq0);
622 /* REACTION-FIELD ELECTROSTATICS */
623 felec = _mm_mul_pd(qq20,_mm_msub_pd(rinv20,rinvsq20,krf2));
627 /* Update vectorial force */
628 fix2 = _mm_macc_pd(dx20,fscal,fix2);
629 fiy2 = _mm_macc_pd(dy20,fscal,fiy2);
630 fiz2 = _mm_macc_pd(dz20,fscal,fiz2);
632 fjx0 = _mm_macc_pd(dx20,fscal,fjx0);
633 fjy0 = _mm_macc_pd(dy20,fscal,fjy0);
634 fjz0 = _mm_macc_pd(dz20,fscal,fjz0);
636 gmx_mm_decrement_1rvec_2ptr_swizzle_pd(f+j_coord_offsetA,f+j_coord_offsetB,fjx0,fjy0,fjz0);
638 /* Inner loop uses 93 flops */
645 j_coord_offsetA = DIM*jnrA;
647 /* load j atom coordinates */
648 gmx_mm_load_1rvec_1ptr_swizzle_pd(x+j_coord_offsetA,
651 /* Calculate displacement vector */
652 dx00 = _mm_sub_pd(ix0,jx0);
653 dy00 = _mm_sub_pd(iy0,jy0);
654 dz00 = _mm_sub_pd(iz0,jz0);
655 dx10 = _mm_sub_pd(ix1,jx0);
656 dy10 = _mm_sub_pd(iy1,jy0);
657 dz10 = _mm_sub_pd(iz1,jz0);
658 dx20 = _mm_sub_pd(ix2,jx0);
659 dy20 = _mm_sub_pd(iy2,jy0);
660 dz20 = _mm_sub_pd(iz2,jz0);
662 /* Calculate squared distance and things based on it */
663 rsq00 = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
664 rsq10 = gmx_mm_calc_rsq_pd(dx10,dy10,dz10);
665 rsq20 = gmx_mm_calc_rsq_pd(dx20,dy20,dz20);
667 rinv00 = gmx_mm_invsqrt_pd(rsq00);
668 rinv10 = gmx_mm_invsqrt_pd(rsq10);
669 rinv20 = gmx_mm_invsqrt_pd(rsq20);
671 rinvsq00 = _mm_mul_pd(rinv00,rinv00);
672 rinvsq10 = _mm_mul_pd(rinv10,rinv10);
673 rinvsq20 = _mm_mul_pd(rinv20,rinv20);
675 /* Load parameters for j particles */
676 jq0 = _mm_load_sd(charge+jnrA+0);
678 fjx0 = _mm_setzero_pd();
679 fjy0 = _mm_setzero_pd();
680 fjz0 = _mm_setzero_pd();
682 /**************************
683 * CALCULATE INTERACTIONS *
684 **************************/
686 /* Compute parameters for interactions between i and j atoms */
687 qq00 = _mm_mul_pd(iq0,jq0);
689 /* REACTION-FIELD ELECTROSTATICS */
690 felec = _mm_mul_pd(qq00,_mm_msub_pd(rinv00,rinvsq00,krf2));
694 fscal = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
696 /* Update vectorial force */
697 fix0 = _mm_macc_pd(dx00,fscal,fix0);
698 fiy0 = _mm_macc_pd(dy00,fscal,fiy0);
699 fiz0 = _mm_macc_pd(dz00,fscal,fiz0);
701 fjx0 = _mm_macc_pd(dx00,fscal,fjx0);
702 fjy0 = _mm_macc_pd(dy00,fscal,fjy0);
703 fjz0 = _mm_macc_pd(dz00,fscal,fjz0);
705 /**************************
706 * CALCULATE INTERACTIONS *
707 **************************/
709 /* Compute parameters for interactions between i and j atoms */
710 qq10 = _mm_mul_pd(iq1,jq0);
712 /* REACTION-FIELD ELECTROSTATICS */
713 felec = _mm_mul_pd(qq10,_mm_msub_pd(rinv10,rinvsq10,krf2));
717 fscal = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
719 /* Update vectorial force */
720 fix1 = _mm_macc_pd(dx10,fscal,fix1);
721 fiy1 = _mm_macc_pd(dy10,fscal,fiy1);
722 fiz1 = _mm_macc_pd(dz10,fscal,fiz1);
724 fjx0 = _mm_macc_pd(dx10,fscal,fjx0);
725 fjy0 = _mm_macc_pd(dy10,fscal,fjy0);
726 fjz0 = _mm_macc_pd(dz10,fscal,fjz0);
728 /**************************
729 * CALCULATE INTERACTIONS *
730 **************************/
732 /* Compute parameters for interactions between i and j atoms */
733 qq20 = _mm_mul_pd(iq2,jq0);
735 /* REACTION-FIELD ELECTROSTATICS */
736 felec = _mm_mul_pd(qq20,_mm_msub_pd(rinv20,rinvsq20,krf2));
740 fscal = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
742 /* Update vectorial force */
743 fix2 = _mm_macc_pd(dx20,fscal,fix2);
744 fiy2 = _mm_macc_pd(dy20,fscal,fiy2);
745 fiz2 = _mm_macc_pd(dz20,fscal,fiz2);
747 fjx0 = _mm_macc_pd(dx20,fscal,fjx0);
748 fjy0 = _mm_macc_pd(dy20,fscal,fjy0);
749 fjz0 = _mm_macc_pd(dz20,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(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,
759 f+i_coord_offset,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_W3_F,outeriter*18 + inneriter*93);