2 * This file is part of the GROMACS molecular simulation package.
4 * Copyright (c) 2012,2013, 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 avx_256_double kernel generator.
44 #include "../nb_kernel.h"
45 #include "types/simple.h"
49 #include "gromacs/simd/math_x86_avx_256_double.h"
50 #include "kernelutil_x86_avx_256_double.h"
53 * Gromacs nonbonded kernel: nb_kernel_ElecRFCut_VdwNone_GeomW3W3_VF_avx_256_double
54 * Electrostatics interaction: ReactionField
55 * VdW interaction: None
56 * Geometry: Water3-Water3
57 * Calculate force/pot: PotentialAndForce
60 nb_kernel_ElecRFCut_VdwNone_GeomW3W3_VF_avx_256_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,C,D refer to j loop unrolling done with AVX, e.g. for the four 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;
76 int jnrA,jnrB,jnrC,jnrD;
77 int jnrlistA,jnrlistB,jnrlistC,jnrlistD;
78 int jnrlistE,jnrlistF,jnrlistG,jnrlistH;
79 int j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
80 int *iinr,*jindex,*jjnr,*shiftidx,*gid;
82 real *shiftvec,*fshift,*x,*f;
83 real *fjptrA,*fjptrB,*fjptrC,*fjptrD;
85 __m256d tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
86 real * vdwioffsetptr0;
87 __m256d ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
88 real * vdwioffsetptr1;
89 __m256d ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
90 real * vdwioffsetptr2;
91 __m256d ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
92 int vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
93 __m256d jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
94 int vdwjidx1A,vdwjidx1B,vdwjidx1C,vdwjidx1D;
95 __m256d jx1,jy1,jz1,fjx1,fjy1,fjz1,jq1,isaj1;
96 int vdwjidx2A,vdwjidx2B,vdwjidx2C,vdwjidx2D;
97 __m256d jx2,jy2,jz2,fjx2,fjy2,fjz2,jq2,isaj2;
98 __m256d dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
99 __m256d dx01,dy01,dz01,rsq01,rinv01,rinvsq01,r01,qq01,c6_01,c12_01;
100 __m256d dx02,dy02,dz02,rsq02,rinv02,rinvsq02,r02,qq02,c6_02,c12_02;
101 __m256d dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
102 __m256d dx11,dy11,dz11,rsq11,rinv11,rinvsq11,r11,qq11,c6_11,c12_11;
103 __m256d dx12,dy12,dz12,rsq12,rinv12,rinvsq12,r12,qq12,c6_12,c12_12;
104 __m256d dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
105 __m256d dx21,dy21,dz21,rsq21,rinv21,rinvsq21,r21,qq21,c6_21,c12_21;
106 __m256d dx22,dy22,dz22,rsq22,rinv22,rinvsq22,r22,qq22,c6_22,c12_22;
107 __m256d velec,felec,velecsum,facel,crf,krf,krf2;
109 __m256d dummy_mask,cutoff_mask;
110 __m128 tmpmask0,tmpmask1;
111 __m256d signbit = _mm256_castsi256_pd( _mm256_set1_epi32(0x80000000) );
112 __m256d one = _mm256_set1_pd(1.0);
113 __m256d two = _mm256_set1_pd(2.0);
119 jindex = nlist->jindex;
121 shiftidx = nlist->shift;
123 shiftvec = fr->shift_vec[0];
124 fshift = fr->fshift[0];
125 facel = _mm256_set1_pd(fr->epsfac);
126 charge = mdatoms->chargeA;
127 krf = _mm256_set1_pd(fr->ic->k_rf);
128 krf2 = _mm256_set1_pd(fr->ic->k_rf*2.0);
129 crf = _mm256_set1_pd(fr->ic->c_rf);
131 /* Setup water-specific parameters */
132 inr = nlist->iinr[0];
133 iq0 = _mm256_mul_pd(facel,_mm256_set1_pd(charge[inr+0]));
134 iq1 = _mm256_mul_pd(facel,_mm256_set1_pd(charge[inr+1]));
135 iq2 = _mm256_mul_pd(facel,_mm256_set1_pd(charge[inr+2]));
137 jq0 = _mm256_set1_pd(charge[inr+0]);
138 jq1 = _mm256_set1_pd(charge[inr+1]);
139 jq2 = _mm256_set1_pd(charge[inr+2]);
140 qq00 = _mm256_mul_pd(iq0,jq0);
141 qq01 = _mm256_mul_pd(iq0,jq1);
142 qq02 = _mm256_mul_pd(iq0,jq2);
143 qq10 = _mm256_mul_pd(iq1,jq0);
144 qq11 = _mm256_mul_pd(iq1,jq1);
145 qq12 = _mm256_mul_pd(iq1,jq2);
146 qq20 = _mm256_mul_pd(iq2,jq0);
147 qq21 = _mm256_mul_pd(iq2,jq1);
148 qq22 = _mm256_mul_pd(iq2,jq2);
150 /* When we use explicit cutoffs the value must be identical for elec and VdW, so use elec as an arbitrary choice */
151 rcutoff_scalar = fr->rcoulomb;
152 rcutoff = _mm256_set1_pd(rcutoff_scalar);
153 rcutoff2 = _mm256_mul_pd(rcutoff,rcutoff);
155 /* Avoid stupid compiler warnings */
156 jnrA = jnrB = jnrC = jnrD = 0;
165 for(iidx=0;iidx<4*DIM;iidx++)
170 /* Start outer loop over neighborlists */
171 for(iidx=0; iidx<nri; iidx++)
173 /* Load shift vector for this list */
174 i_shift_offset = DIM*shiftidx[iidx];
176 /* Load limits for loop over neighbors */
177 j_index_start = jindex[iidx];
178 j_index_end = jindex[iidx+1];
180 /* Get outer coordinate index */
182 i_coord_offset = DIM*inr;
184 /* Load i particle coords and add shift vector */
185 gmx_mm256_load_shift_and_3rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset,
186 &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2);
188 fix0 = _mm256_setzero_pd();
189 fiy0 = _mm256_setzero_pd();
190 fiz0 = _mm256_setzero_pd();
191 fix1 = _mm256_setzero_pd();
192 fiy1 = _mm256_setzero_pd();
193 fiz1 = _mm256_setzero_pd();
194 fix2 = _mm256_setzero_pd();
195 fiy2 = _mm256_setzero_pd();
196 fiz2 = _mm256_setzero_pd();
198 /* Reset potential sums */
199 velecsum = _mm256_setzero_pd();
201 /* Start inner kernel loop */
202 for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
205 /* Get j neighbor index, and coordinate index */
210 j_coord_offsetA = DIM*jnrA;
211 j_coord_offsetB = DIM*jnrB;
212 j_coord_offsetC = DIM*jnrC;
213 j_coord_offsetD = DIM*jnrD;
215 /* load j atom coordinates */
216 gmx_mm256_load_3rvec_4ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
217 x+j_coord_offsetC,x+j_coord_offsetD,
218 &jx0,&jy0,&jz0,&jx1,&jy1,&jz1,&jx2,&jy2,&jz2);
220 /* Calculate displacement vector */
221 dx00 = _mm256_sub_pd(ix0,jx0);
222 dy00 = _mm256_sub_pd(iy0,jy0);
223 dz00 = _mm256_sub_pd(iz0,jz0);
224 dx01 = _mm256_sub_pd(ix0,jx1);
225 dy01 = _mm256_sub_pd(iy0,jy1);
226 dz01 = _mm256_sub_pd(iz0,jz1);
227 dx02 = _mm256_sub_pd(ix0,jx2);
228 dy02 = _mm256_sub_pd(iy0,jy2);
229 dz02 = _mm256_sub_pd(iz0,jz2);
230 dx10 = _mm256_sub_pd(ix1,jx0);
231 dy10 = _mm256_sub_pd(iy1,jy0);
232 dz10 = _mm256_sub_pd(iz1,jz0);
233 dx11 = _mm256_sub_pd(ix1,jx1);
234 dy11 = _mm256_sub_pd(iy1,jy1);
235 dz11 = _mm256_sub_pd(iz1,jz1);
236 dx12 = _mm256_sub_pd(ix1,jx2);
237 dy12 = _mm256_sub_pd(iy1,jy2);
238 dz12 = _mm256_sub_pd(iz1,jz2);
239 dx20 = _mm256_sub_pd(ix2,jx0);
240 dy20 = _mm256_sub_pd(iy2,jy0);
241 dz20 = _mm256_sub_pd(iz2,jz0);
242 dx21 = _mm256_sub_pd(ix2,jx1);
243 dy21 = _mm256_sub_pd(iy2,jy1);
244 dz21 = _mm256_sub_pd(iz2,jz1);
245 dx22 = _mm256_sub_pd(ix2,jx2);
246 dy22 = _mm256_sub_pd(iy2,jy2);
247 dz22 = _mm256_sub_pd(iz2,jz2);
249 /* Calculate squared distance and things based on it */
250 rsq00 = gmx_mm256_calc_rsq_pd(dx00,dy00,dz00);
251 rsq01 = gmx_mm256_calc_rsq_pd(dx01,dy01,dz01);
252 rsq02 = gmx_mm256_calc_rsq_pd(dx02,dy02,dz02);
253 rsq10 = gmx_mm256_calc_rsq_pd(dx10,dy10,dz10);
254 rsq11 = gmx_mm256_calc_rsq_pd(dx11,dy11,dz11);
255 rsq12 = gmx_mm256_calc_rsq_pd(dx12,dy12,dz12);
256 rsq20 = gmx_mm256_calc_rsq_pd(dx20,dy20,dz20);
257 rsq21 = gmx_mm256_calc_rsq_pd(dx21,dy21,dz21);
258 rsq22 = gmx_mm256_calc_rsq_pd(dx22,dy22,dz22);
260 rinv00 = gmx_mm256_invsqrt_pd(rsq00);
261 rinv01 = gmx_mm256_invsqrt_pd(rsq01);
262 rinv02 = gmx_mm256_invsqrt_pd(rsq02);
263 rinv10 = gmx_mm256_invsqrt_pd(rsq10);
264 rinv11 = gmx_mm256_invsqrt_pd(rsq11);
265 rinv12 = gmx_mm256_invsqrt_pd(rsq12);
266 rinv20 = gmx_mm256_invsqrt_pd(rsq20);
267 rinv21 = gmx_mm256_invsqrt_pd(rsq21);
268 rinv22 = gmx_mm256_invsqrt_pd(rsq22);
270 rinvsq00 = _mm256_mul_pd(rinv00,rinv00);
271 rinvsq01 = _mm256_mul_pd(rinv01,rinv01);
272 rinvsq02 = _mm256_mul_pd(rinv02,rinv02);
273 rinvsq10 = _mm256_mul_pd(rinv10,rinv10);
274 rinvsq11 = _mm256_mul_pd(rinv11,rinv11);
275 rinvsq12 = _mm256_mul_pd(rinv12,rinv12);
276 rinvsq20 = _mm256_mul_pd(rinv20,rinv20);
277 rinvsq21 = _mm256_mul_pd(rinv21,rinv21);
278 rinvsq22 = _mm256_mul_pd(rinv22,rinv22);
280 fjx0 = _mm256_setzero_pd();
281 fjy0 = _mm256_setzero_pd();
282 fjz0 = _mm256_setzero_pd();
283 fjx1 = _mm256_setzero_pd();
284 fjy1 = _mm256_setzero_pd();
285 fjz1 = _mm256_setzero_pd();
286 fjx2 = _mm256_setzero_pd();
287 fjy2 = _mm256_setzero_pd();
288 fjz2 = _mm256_setzero_pd();
290 /**************************
291 * CALCULATE INTERACTIONS *
292 **************************/
294 if (gmx_mm256_any_lt(rsq00,rcutoff2))
297 /* REACTION-FIELD ELECTROSTATICS */
298 velec = _mm256_mul_pd(qq00,_mm256_sub_pd(_mm256_add_pd(rinv00,_mm256_mul_pd(krf,rsq00)),crf));
299 felec = _mm256_mul_pd(qq00,_mm256_sub_pd(_mm256_mul_pd(rinv00,rinvsq00),krf2));
301 cutoff_mask = _mm256_cmp_pd(rsq00,rcutoff2,_CMP_LT_OQ);
303 /* Update potential sum for this i atom from the interaction with this j atom. */
304 velec = _mm256_and_pd(velec,cutoff_mask);
305 velecsum = _mm256_add_pd(velecsum,velec);
309 fscal = _mm256_and_pd(fscal,cutoff_mask);
311 /* Calculate temporary vectorial force */
312 tx = _mm256_mul_pd(fscal,dx00);
313 ty = _mm256_mul_pd(fscal,dy00);
314 tz = _mm256_mul_pd(fscal,dz00);
316 /* Update vectorial force */
317 fix0 = _mm256_add_pd(fix0,tx);
318 fiy0 = _mm256_add_pd(fiy0,ty);
319 fiz0 = _mm256_add_pd(fiz0,tz);
321 fjx0 = _mm256_add_pd(fjx0,tx);
322 fjy0 = _mm256_add_pd(fjy0,ty);
323 fjz0 = _mm256_add_pd(fjz0,tz);
327 /**************************
328 * CALCULATE INTERACTIONS *
329 **************************/
331 if (gmx_mm256_any_lt(rsq01,rcutoff2))
334 /* REACTION-FIELD ELECTROSTATICS */
335 velec = _mm256_mul_pd(qq01,_mm256_sub_pd(_mm256_add_pd(rinv01,_mm256_mul_pd(krf,rsq01)),crf));
336 felec = _mm256_mul_pd(qq01,_mm256_sub_pd(_mm256_mul_pd(rinv01,rinvsq01),krf2));
338 cutoff_mask = _mm256_cmp_pd(rsq01,rcutoff2,_CMP_LT_OQ);
340 /* Update potential sum for this i atom from the interaction with this j atom. */
341 velec = _mm256_and_pd(velec,cutoff_mask);
342 velecsum = _mm256_add_pd(velecsum,velec);
346 fscal = _mm256_and_pd(fscal,cutoff_mask);
348 /* Calculate temporary vectorial force */
349 tx = _mm256_mul_pd(fscal,dx01);
350 ty = _mm256_mul_pd(fscal,dy01);
351 tz = _mm256_mul_pd(fscal,dz01);
353 /* Update vectorial force */
354 fix0 = _mm256_add_pd(fix0,tx);
355 fiy0 = _mm256_add_pd(fiy0,ty);
356 fiz0 = _mm256_add_pd(fiz0,tz);
358 fjx1 = _mm256_add_pd(fjx1,tx);
359 fjy1 = _mm256_add_pd(fjy1,ty);
360 fjz1 = _mm256_add_pd(fjz1,tz);
364 /**************************
365 * CALCULATE INTERACTIONS *
366 **************************/
368 if (gmx_mm256_any_lt(rsq02,rcutoff2))
371 /* REACTION-FIELD ELECTROSTATICS */
372 velec = _mm256_mul_pd(qq02,_mm256_sub_pd(_mm256_add_pd(rinv02,_mm256_mul_pd(krf,rsq02)),crf));
373 felec = _mm256_mul_pd(qq02,_mm256_sub_pd(_mm256_mul_pd(rinv02,rinvsq02),krf2));
375 cutoff_mask = _mm256_cmp_pd(rsq02,rcutoff2,_CMP_LT_OQ);
377 /* Update potential sum for this i atom from the interaction with this j atom. */
378 velec = _mm256_and_pd(velec,cutoff_mask);
379 velecsum = _mm256_add_pd(velecsum,velec);
383 fscal = _mm256_and_pd(fscal,cutoff_mask);
385 /* Calculate temporary vectorial force */
386 tx = _mm256_mul_pd(fscal,dx02);
387 ty = _mm256_mul_pd(fscal,dy02);
388 tz = _mm256_mul_pd(fscal,dz02);
390 /* Update vectorial force */
391 fix0 = _mm256_add_pd(fix0,tx);
392 fiy0 = _mm256_add_pd(fiy0,ty);
393 fiz0 = _mm256_add_pd(fiz0,tz);
395 fjx2 = _mm256_add_pd(fjx2,tx);
396 fjy2 = _mm256_add_pd(fjy2,ty);
397 fjz2 = _mm256_add_pd(fjz2,tz);
401 /**************************
402 * CALCULATE INTERACTIONS *
403 **************************/
405 if (gmx_mm256_any_lt(rsq10,rcutoff2))
408 /* REACTION-FIELD ELECTROSTATICS */
409 velec = _mm256_mul_pd(qq10,_mm256_sub_pd(_mm256_add_pd(rinv10,_mm256_mul_pd(krf,rsq10)),crf));
410 felec = _mm256_mul_pd(qq10,_mm256_sub_pd(_mm256_mul_pd(rinv10,rinvsq10),krf2));
412 cutoff_mask = _mm256_cmp_pd(rsq10,rcutoff2,_CMP_LT_OQ);
414 /* Update potential sum for this i atom from the interaction with this j atom. */
415 velec = _mm256_and_pd(velec,cutoff_mask);
416 velecsum = _mm256_add_pd(velecsum,velec);
420 fscal = _mm256_and_pd(fscal,cutoff_mask);
422 /* Calculate temporary vectorial force */
423 tx = _mm256_mul_pd(fscal,dx10);
424 ty = _mm256_mul_pd(fscal,dy10);
425 tz = _mm256_mul_pd(fscal,dz10);
427 /* Update vectorial force */
428 fix1 = _mm256_add_pd(fix1,tx);
429 fiy1 = _mm256_add_pd(fiy1,ty);
430 fiz1 = _mm256_add_pd(fiz1,tz);
432 fjx0 = _mm256_add_pd(fjx0,tx);
433 fjy0 = _mm256_add_pd(fjy0,ty);
434 fjz0 = _mm256_add_pd(fjz0,tz);
438 /**************************
439 * CALCULATE INTERACTIONS *
440 **************************/
442 if (gmx_mm256_any_lt(rsq11,rcutoff2))
445 /* REACTION-FIELD ELECTROSTATICS */
446 velec = _mm256_mul_pd(qq11,_mm256_sub_pd(_mm256_add_pd(rinv11,_mm256_mul_pd(krf,rsq11)),crf));
447 felec = _mm256_mul_pd(qq11,_mm256_sub_pd(_mm256_mul_pd(rinv11,rinvsq11),krf2));
449 cutoff_mask = _mm256_cmp_pd(rsq11,rcutoff2,_CMP_LT_OQ);
451 /* Update potential sum for this i atom from the interaction with this j atom. */
452 velec = _mm256_and_pd(velec,cutoff_mask);
453 velecsum = _mm256_add_pd(velecsum,velec);
457 fscal = _mm256_and_pd(fscal,cutoff_mask);
459 /* Calculate temporary vectorial force */
460 tx = _mm256_mul_pd(fscal,dx11);
461 ty = _mm256_mul_pd(fscal,dy11);
462 tz = _mm256_mul_pd(fscal,dz11);
464 /* Update vectorial force */
465 fix1 = _mm256_add_pd(fix1,tx);
466 fiy1 = _mm256_add_pd(fiy1,ty);
467 fiz1 = _mm256_add_pd(fiz1,tz);
469 fjx1 = _mm256_add_pd(fjx1,tx);
470 fjy1 = _mm256_add_pd(fjy1,ty);
471 fjz1 = _mm256_add_pd(fjz1,tz);
475 /**************************
476 * CALCULATE INTERACTIONS *
477 **************************/
479 if (gmx_mm256_any_lt(rsq12,rcutoff2))
482 /* REACTION-FIELD ELECTROSTATICS */
483 velec = _mm256_mul_pd(qq12,_mm256_sub_pd(_mm256_add_pd(rinv12,_mm256_mul_pd(krf,rsq12)),crf));
484 felec = _mm256_mul_pd(qq12,_mm256_sub_pd(_mm256_mul_pd(rinv12,rinvsq12),krf2));
486 cutoff_mask = _mm256_cmp_pd(rsq12,rcutoff2,_CMP_LT_OQ);
488 /* Update potential sum for this i atom from the interaction with this j atom. */
489 velec = _mm256_and_pd(velec,cutoff_mask);
490 velecsum = _mm256_add_pd(velecsum,velec);
494 fscal = _mm256_and_pd(fscal,cutoff_mask);
496 /* Calculate temporary vectorial force */
497 tx = _mm256_mul_pd(fscal,dx12);
498 ty = _mm256_mul_pd(fscal,dy12);
499 tz = _mm256_mul_pd(fscal,dz12);
501 /* Update vectorial force */
502 fix1 = _mm256_add_pd(fix1,tx);
503 fiy1 = _mm256_add_pd(fiy1,ty);
504 fiz1 = _mm256_add_pd(fiz1,tz);
506 fjx2 = _mm256_add_pd(fjx2,tx);
507 fjy2 = _mm256_add_pd(fjy2,ty);
508 fjz2 = _mm256_add_pd(fjz2,tz);
512 /**************************
513 * CALCULATE INTERACTIONS *
514 **************************/
516 if (gmx_mm256_any_lt(rsq20,rcutoff2))
519 /* REACTION-FIELD ELECTROSTATICS */
520 velec = _mm256_mul_pd(qq20,_mm256_sub_pd(_mm256_add_pd(rinv20,_mm256_mul_pd(krf,rsq20)),crf));
521 felec = _mm256_mul_pd(qq20,_mm256_sub_pd(_mm256_mul_pd(rinv20,rinvsq20),krf2));
523 cutoff_mask = _mm256_cmp_pd(rsq20,rcutoff2,_CMP_LT_OQ);
525 /* Update potential sum for this i atom from the interaction with this j atom. */
526 velec = _mm256_and_pd(velec,cutoff_mask);
527 velecsum = _mm256_add_pd(velecsum,velec);
531 fscal = _mm256_and_pd(fscal,cutoff_mask);
533 /* Calculate temporary vectorial force */
534 tx = _mm256_mul_pd(fscal,dx20);
535 ty = _mm256_mul_pd(fscal,dy20);
536 tz = _mm256_mul_pd(fscal,dz20);
538 /* Update vectorial force */
539 fix2 = _mm256_add_pd(fix2,tx);
540 fiy2 = _mm256_add_pd(fiy2,ty);
541 fiz2 = _mm256_add_pd(fiz2,tz);
543 fjx0 = _mm256_add_pd(fjx0,tx);
544 fjy0 = _mm256_add_pd(fjy0,ty);
545 fjz0 = _mm256_add_pd(fjz0,tz);
549 /**************************
550 * CALCULATE INTERACTIONS *
551 **************************/
553 if (gmx_mm256_any_lt(rsq21,rcutoff2))
556 /* REACTION-FIELD ELECTROSTATICS */
557 velec = _mm256_mul_pd(qq21,_mm256_sub_pd(_mm256_add_pd(rinv21,_mm256_mul_pd(krf,rsq21)),crf));
558 felec = _mm256_mul_pd(qq21,_mm256_sub_pd(_mm256_mul_pd(rinv21,rinvsq21),krf2));
560 cutoff_mask = _mm256_cmp_pd(rsq21,rcutoff2,_CMP_LT_OQ);
562 /* Update potential sum for this i atom from the interaction with this j atom. */
563 velec = _mm256_and_pd(velec,cutoff_mask);
564 velecsum = _mm256_add_pd(velecsum,velec);
568 fscal = _mm256_and_pd(fscal,cutoff_mask);
570 /* Calculate temporary vectorial force */
571 tx = _mm256_mul_pd(fscal,dx21);
572 ty = _mm256_mul_pd(fscal,dy21);
573 tz = _mm256_mul_pd(fscal,dz21);
575 /* Update vectorial force */
576 fix2 = _mm256_add_pd(fix2,tx);
577 fiy2 = _mm256_add_pd(fiy2,ty);
578 fiz2 = _mm256_add_pd(fiz2,tz);
580 fjx1 = _mm256_add_pd(fjx1,tx);
581 fjy1 = _mm256_add_pd(fjy1,ty);
582 fjz1 = _mm256_add_pd(fjz1,tz);
586 /**************************
587 * CALCULATE INTERACTIONS *
588 **************************/
590 if (gmx_mm256_any_lt(rsq22,rcutoff2))
593 /* REACTION-FIELD ELECTROSTATICS */
594 velec = _mm256_mul_pd(qq22,_mm256_sub_pd(_mm256_add_pd(rinv22,_mm256_mul_pd(krf,rsq22)),crf));
595 felec = _mm256_mul_pd(qq22,_mm256_sub_pd(_mm256_mul_pd(rinv22,rinvsq22),krf2));
597 cutoff_mask = _mm256_cmp_pd(rsq22,rcutoff2,_CMP_LT_OQ);
599 /* Update potential sum for this i atom from the interaction with this j atom. */
600 velec = _mm256_and_pd(velec,cutoff_mask);
601 velecsum = _mm256_add_pd(velecsum,velec);
605 fscal = _mm256_and_pd(fscal,cutoff_mask);
607 /* Calculate temporary vectorial force */
608 tx = _mm256_mul_pd(fscal,dx22);
609 ty = _mm256_mul_pd(fscal,dy22);
610 tz = _mm256_mul_pd(fscal,dz22);
612 /* Update vectorial force */
613 fix2 = _mm256_add_pd(fix2,tx);
614 fiy2 = _mm256_add_pd(fiy2,ty);
615 fiz2 = _mm256_add_pd(fiz2,tz);
617 fjx2 = _mm256_add_pd(fjx2,tx);
618 fjy2 = _mm256_add_pd(fjy2,ty);
619 fjz2 = _mm256_add_pd(fjz2,tz);
623 fjptrA = f+j_coord_offsetA;
624 fjptrB = f+j_coord_offsetB;
625 fjptrC = f+j_coord_offsetC;
626 fjptrD = f+j_coord_offsetD;
628 gmx_mm256_decrement_3rvec_4ptr_swizzle_pd(fjptrA,fjptrB,fjptrC,fjptrD,
629 fjx0,fjy0,fjz0,fjx1,fjy1,fjz1,fjx2,fjy2,fjz2);
631 /* Inner loop uses 324 flops */
637 /* Get j neighbor index, and coordinate index */
638 jnrlistA = jjnr[jidx];
639 jnrlistB = jjnr[jidx+1];
640 jnrlistC = jjnr[jidx+2];
641 jnrlistD = jjnr[jidx+3];
642 /* Sign of each element will be negative for non-real atoms.
643 * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
644 * so use it as val = _mm_andnot_pd(mask,val) to clear dummy entries.
646 tmpmask0 = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
648 tmpmask1 = _mm_permute_ps(tmpmask0,_GMX_MM_PERMUTE(3,3,2,2));
649 tmpmask0 = _mm_permute_ps(tmpmask0,_GMX_MM_PERMUTE(1,1,0,0));
650 dummy_mask = _mm256_castps_pd(gmx_mm256_set_m128(tmpmask1,tmpmask0));
652 jnrA = (jnrlistA>=0) ? jnrlistA : 0;
653 jnrB = (jnrlistB>=0) ? jnrlistB : 0;
654 jnrC = (jnrlistC>=0) ? jnrlistC : 0;
655 jnrD = (jnrlistD>=0) ? jnrlistD : 0;
656 j_coord_offsetA = DIM*jnrA;
657 j_coord_offsetB = DIM*jnrB;
658 j_coord_offsetC = DIM*jnrC;
659 j_coord_offsetD = DIM*jnrD;
661 /* load j atom coordinates */
662 gmx_mm256_load_3rvec_4ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
663 x+j_coord_offsetC,x+j_coord_offsetD,
664 &jx0,&jy0,&jz0,&jx1,&jy1,&jz1,&jx2,&jy2,&jz2);
666 /* Calculate displacement vector */
667 dx00 = _mm256_sub_pd(ix0,jx0);
668 dy00 = _mm256_sub_pd(iy0,jy0);
669 dz00 = _mm256_sub_pd(iz0,jz0);
670 dx01 = _mm256_sub_pd(ix0,jx1);
671 dy01 = _mm256_sub_pd(iy0,jy1);
672 dz01 = _mm256_sub_pd(iz0,jz1);
673 dx02 = _mm256_sub_pd(ix0,jx2);
674 dy02 = _mm256_sub_pd(iy0,jy2);
675 dz02 = _mm256_sub_pd(iz0,jz2);
676 dx10 = _mm256_sub_pd(ix1,jx0);
677 dy10 = _mm256_sub_pd(iy1,jy0);
678 dz10 = _mm256_sub_pd(iz1,jz0);
679 dx11 = _mm256_sub_pd(ix1,jx1);
680 dy11 = _mm256_sub_pd(iy1,jy1);
681 dz11 = _mm256_sub_pd(iz1,jz1);
682 dx12 = _mm256_sub_pd(ix1,jx2);
683 dy12 = _mm256_sub_pd(iy1,jy2);
684 dz12 = _mm256_sub_pd(iz1,jz2);
685 dx20 = _mm256_sub_pd(ix2,jx0);
686 dy20 = _mm256_sub_pd(iy2,jy0);
687 dz20 = _mm256_sub_pd(iz2,jz0);
688 dx21 = _mm256_sub_pd(ix2,jx1);
689 dy21 = _mm256_sub_pd(iy2,jy1);
690 dz21 = _mm256_sub_pd(iz2,jz1);
691 dx22 = _mm256_sub_pd(ix2,jx2);
692 dy22 = _mm256_sub_pd(iy2,jy2);
693 dz22 = _mm256_sub_pd(iz2,jz2);
695 /* Calculate squared distance and things based on it */
696 rsq00 = gmx_mm256_calc_rsq_pd(dx00,dy00,dz00);
697 rsq01 = gmx_mm256_calc_rsq_pd(dx01,dy01,dz01);
698 rsq02 = gmx_mm256_calc_rsq_pd(dx02,dy02,dz02);
699 rsq10 = gmx_mm256_calc_rsq_pd(dx10,dy10,dz10);
700 rsq11 = gmx_mm256_calc_rsq_pd(dx11,dy11,dz11);
701 rsq12 = gmx_mm256_calc_rsq_pd(dx12,dy12,dz12);
702 rsq20 = gmx_mm256_calc_rsq_pd(dx20,dy20,dz20);
703 rsq21 = gmx_mm256_calc_rsq_pd(dx21,dy21,dz21);
704 rsq22 = gmx_mm256_calc_rsq_pd(dx22,dy22,dz22);
706 rinv00 = gmx_mm256_invsqrt_pd(rsq00);
707 rinv01 = gmx_mm256_invsqrt_pd(rsq01);
708 rinv02 = gmx_mm256_invsqrt_pd(rsq02);
709 rinv10 = gmx_mm256_invsqrt_pd(rsq10);
710 rinv11 = gmx_mm256_invsqrt_pd(rsq11);
711 rinv12 = gmx_mm256_invsqrt_pd(rsq12);
712 rinv20 = gmx_mm256_invsqrt_pd(rsq20);
713 rinv21 = gmx_mm256_invsqrt_pd(rsq21);
714 rinv22 = gmx_mm256_invsqrt_pd(rsq22);
716 rinvsq00 = _mm256_mul_pd(rinv00,rinv00);
717 rinvsq01 = _mm256_mul_pd(rinv01,rinv01);
718 rinvsq02 = _mm256_mul_pd(rinv02,rinv02);
719 rinvsq10 = _mm256_mul_pd(rinv10,rinv10);
720 rinvsq11 = _mm256_mul_pd(rinv11,rinv11);
721 rinvsq12 = _mm256_mul_pd(rinv12,rinv12);
722 rinvsq20 = _mm256_mul_pd(rinv20,rinv20);
723 rinvsq21 = _mm256_mul_pd(rinv21,rinv21);
724 rinvsq22 = _mm256_mul_pd(rinv22,rinv22);
726 fjx0 = _mm256_setzero_pd();
727 fjy0 = _mm256_setzero_pd();
728 fjz0 = _mm256_setzero_pd();
729 fjx1 = _mm256_setzero_pd();
730 fjy1 = _mm256_setzero_pd();
731 fjz1 = _mm256_setzero_pd();
732 fjx2 = _mm256_setzero_pd();
733 fjy2 = _mm256_setzero_pd();
734 fjz2 = _mm256_setzero_pd();
736 /**************************
737 * CALCULATE INTERACTIONS *
738 **************************/
740 if (gmx_mm256_any_lt(rsq00,rcutoff2))
743 /* REACTION-FIELD ELECTROSTATICS */
744 velec = _mm256_mul_pd(qq00,_mm256_sub_pd(_mm256_add_pd(rinv00,_mm256_mul_pd(krf,rsq00)),crf));
745 felec = _mm256_mul_pd(qq00,_mm256_sub_pd(_mm256_mul_pd(rinv00,rinvsq00),krf2));
747 cutoff_mask = _mm256_cmp_pd(rsq00,rcutoff2,_CMP_LT_OQ);
749 /* Update potential sum for this i atom from the interaction with this j atom. */
750 velec = _mm256_and_pd(velec,cutoff_mask);
751 velec = _mm256_andnot_pd(dummy_mask,velec);
752 velecsum = _mm256_add_pd(velecsum,velec);
756 fscal = _mm256_and_pd(fscal,cutoff_mask);
758 fscal = _mm256_andnot_pd(dummy_mask,fscal);
760 /* Calculate temporary vectorial force */
761 tx = _mm256_mul_pd(fscal,dx00);
762 ty = _mm256_mul_pd(fscal,dy00);
763 tz = _mm256_mul_pd(fscal,dz00);
765 /* Update vectorial force */
766 fix0 = _mm256_add_pd(fix0,tx);
767 fiy0 = _mm256_add_pd(fiy0,ty);
768 fiz0 = _mm256_add_pd(fiz0,tz);
770 fjx0 = _mm256_add_pd(fjx0,tx);
771 fjy0 = _mm256_add_pd(fjy0,ty);
772 fjz0 = _mm256_add_pd(fjz0,tz);
776 /**************************
777 * CALCULATE INTERACTIONS *
778 **************************/
780 if (gmx_mm256_any_lt(rsq01,rcutoff2))
783 /* REACTION-FIELD ELECTROSTATICS */
784 velec = _mm256_mul_pd(qq01,_mm256_sub_pd(_mm256_add_pd(rinv01,_mm256_mul_pd(krf,rsq01)),crf));
785 felec = _mm256_mul_pd(qq01,_mm256_sub_pd(_mm256_mul_pd(rinv01,rinvsq01),krf2));
787 cutoff_mask = _mm256_cmp_pd(rsq01,rcutoff2,_CMP_LT_OQ);
789 /* Update potential sum for this i atom from the interaction with this j atom. */
790 velec = _mm256_and_pd(velec,cutoff_mask);
791 velec = _mm256_andnot_pd(dummy_mask,velec);
792 velecsum = _mm256_add_pd(velecsum,velec);
796 fscal = _mm256_and_pd(fscal,cutoff_mask);
798 fscal = _mm256_andnot_pd(dummy_mask,fscal);
800 /* Calculate temporary vectorial force */
801 tx = _mm256_mul_pd(fscal,dx01);
802 ty = _mm256_mul_pd(fscal,dy01);
803 tz = _mm256_mul_pd(fscal,dz01);
805 /* Update vectorial force */
806 fix0 = _mm256_add_pd(fix0,tx);
807 fiy0 = _mm256_add_pd(fiy0,ty);
808 fiz0 = _mm256_add_pd(fiz0,tz);
810 fjx1 = _mm256_add_pd(fjx1,tx);
811 fjy1 = _mm256_add_pd(fjy1,ty);
812 fjz1 = _mm256_add_pd(fjz1,tz);
816 /**************************
817 * CALCULATE INTERACTIONS *
818 **************************/
820 if (gmx_mm256_any_lt(rsq02,rcutoff2))
823 /* REACTION-FIELD ELECTROSTATICS */
824 velec = _mm256_mul_pd(qq02,_mm256_sub_pd(_mm256_add_pd(rinv02,_mm256_mul_pd(krf,rsq02)),crf));
825 felec = _mm256_mul_pd(qq02,_mm256_sub_pd(_mm256_mul_pd(rinv02,rinvsq02),krf2));
827 cutoff_mask = _mm256_cmp_pd(rsq02,rcutoff2,_CMP_LT_OQ);
829 /* Update potential sum for this i atom from the interaction with this j atom. */
830 velec = _mm256_and_pd(velec,cutoff_mask);
831 velec = _mm256_andnot_pd(dummy_mask,velec);
832 velecsum = _mm256_add_pd(velecsum,velec);
836 fscal = _mm256_and_pd(fscal,cutoff_mask);
838 fscal = _mm256_andnot_pd(dummy_mask,fscal);
840 /* Calculate temporary vectorial force */
841 tx = _mm256_mul_pd(fscal,dx02);
842 ty = _mm256_mul_pd(fscal,dy02);
843 tz = _mm256_mul_pd(fscal,dz02);
845 /* Update vectorial force */
846 fix0 = _mm256_add_pd(fix0,tx);
847 fiy0 = _mm256_add_pd(fiy0,ty);
848 fiz0 = _mm256_add_pd(fiz0,tz);
850 fjx2 = _mm256_add_pd(fjx2,tx);
851 fjy2 = _mm256_add_pd(fjy2,ty);
852 fjz2 = _mm256_add_pd(fjz2,tz);
856 /**************************
857 * CALCULATE INTERACTIONS *
858 **************************/
860 if (gmx_mm256_any_lt(rsq10,rcutoff2))
863 /* REACTION-FIELD ELECTROSTATICS */
864 velec = _mm256_mul_pd(qq10,_mm256_sub_pd(_mm256_add_pd(rinv10,_mm256_mul_pd(krf,rsq10)),crf));
865 felec = _mm256_mul_pd(qq10,_mm256_sub_pd(_mm256_mul_pd(rinv10,rinvsq10),krf2));
867 cutoff_mask = _mm256_cmp_pd(rsq10,rcutoff2,_CMP_LT_OQ);
869 /* Update potential sum for this i atom from the interaction with this j atom. */
870 velec = _mm256_and_pd(velec,cutoff_mask);
871 velec = _mm256_andnot_pd(dummy_mask,velec);
872 velecsum = _mm256_add_pd(velecsum,velec);
876 fscal = _mm256_and_pd(fscal,cutoff_mask);
878 fscal = _mm256_andnot_pd(dummy_mask,fscal);
880 /* Calculate temporary vectorial force */
881 tx = _mm256_mul_pd(fscal,dx10);
882 ty = _mm256_mul_pd(fscal,dy10);
883 tz = _mm256_mul_pd(fscal,dz10);
885 /* Update vectorial force */
886 fix1 = _mm256_add_pd(fix1,tx);
887 fiy1 = _mm256_add_pd(fiy1,ty);
888 fiz1 = _mm256_add_pd(fiz1,tz);
890 fjx0 = _mm256_add_pd(fjx0,tx);
891 fjy0 = _mm256_add_pd(fjy0,ty);
892 fjz0 = _mm256_add_pd(fjz0,tz);
896 /**************************
897 * CALCULATE INTERACTIONS *
898 **************************/
900 if (gmx_mm256_any_lt(rsq11,rcutoff2))
903 /* REACTION-FIELD ELECTROSTATICS */
904 velec = _mm256_mul_pd(qq11,_mm256_sub_pd(_mm256_add_pd(rinv11,_mm256_mul_pd(krf,rsq11)),crf));
905 felec = _mm256_mul_pd(qq11,_mm256_sub_pd(_mm256_mul_pd(rinv11,rinvsq11),krf2));
907 cutoff_mask = _mm256_cmp_pd(rsq11,rcutoff2,_CMP_LT_OQ);
909 /* Update potential sum for this i atom from the interaction with this j atom. */
910 velec = _mm256_and_pd(velec,cutoff_mask);
911 velec = _mm256_andnot_pd(dummy_mask,velec);
912 velecsum = _mm256_add_pd(velecsum,velec);
916 fscal = _mm256_and_pd(fscal,cutoff_mask);
918 fscal = _mm256_andnot_pd(dummy_mask,fscal);
920 /* Calculate temporary vectorial force */
921 tx = _mm256_mul_pd(fscal,dx11);
922 ty = _mm256_mul_pd(fscal,dy11);
923 tz = _mm256_mul_pd(fscal,dz11);
925 /* Update vectorial force */
926 fix1 = _mm256_add_pd(fix1,tx);
927 fiy1 = _mm256_add_pd(fiy1,ty);
928 fiz1 = _mm256_add_pd(fiz1,tz);
930 fjx1 = _mm256_add_pd(fjx1,tx);
931 fjy1 = _mm256_add_pd(fjy1,ty);
932 fjz1 = _mm256_add_pd(fjz1,tz);
936 /**************************
937 * CALCULATE INTERACTIONS *
938 **************************/
940 if (gmx_mm256_any_lt(rsq12,rcutoff2))
943 /* REACTION-FIELD ELECTROSTATICS */
944 velec = _mm256_mul_pd(qq12,_mm256_sub_pd(_mm256_add_pd(rinv12,_mm256_mul_pd(krf,rsq12)),crf));
945 felec = _mm256_mul_pd(qq12,_mm256_sub_pd(_mm256_mul_pd(rinv12,rinvsq12),krf2));
947 cutoff_mask = _mm256_cmp_pd(rsq12,rcutoff2,_CMP_LT_OQ);
949 /* Update potential sum for this i atom from the interaction with this j atom. */
950 velec = _mm256_and_pd(velec,cutoff_mask);
951 velec = _mm256_andnot_pd(dummy_mask,velec);
952 velecsum = _mm256_add_pd(velecsum,velec);
956 fscal = _mm256_and_pd(fscal,cutoff_mask);
958 fscal = _mm256_andnot_pd(dummy_mask,fscal);
960 /* Calculate temporary vectorial force */
961 tx = _mm256_mul_pd(fscal,dx12);
962 ty = _mm256_mul_pd(fscal,dy12);
963 tz = _mm256_mul_pd(fscal,dz12);
965 /* Update vectorial force */
966 fix1 = _mm256_add_pd(fix1,tx);
967 fiy1 = _mm256_add_pd(fiy1,ty);
968 fiz1 = _mm256_add_pd(fiz1,tz);
970 fjx2 = _mm256_add_pd(fjx2,tx);
971 fjy2 = _mm256_add_pd(fjy2,ty);
972 fjz2 = _mm256_add_pd(fjz2,tz);
976 /**************************
977 * CALCULATE INTERACTIONS *
978 **************************/
980 if (gmx_mm256_any_lt(rsq20,rcutoff2))
983 /* REACTION-FIELD ELECTROSTATICS */
984 velec = _mm256_mul_pd(qq20,_mm256_sub_pd(_mm256_add_pd(rinv20,_mm256_mul_pd(krf,rsq20)),crf));
985 felec = _mm256_mul_pd(qq20,_mm256_sub_pd(_mm256_mul_pd(rinv20,rinvsq20),krf2));
987 cutoff_mask = _mm256_cmp_pd(rsq20,rcutoff2,_CMP_LT_OQ);
989 /* Update potential sum for this i atom from the interaction with this j atom. */
990 velec = _mm256_and_pd(velec,cutoff_mask);
991 velec = _mm256_andnot_pd(dummy_mask,velec);
992 velecsum = _mm256_add_pd(velecsum,velec);
996 fscal = _mm256_and_pd(fscal,cutoff_mask);
998 fscal = _mm256_andnot_pd(dummy_mask,fscal);
1000 /* Calculate temporary vectorial force */
1001 tx = _mm256_mul_pd(fscal,dx20);
1002 ty = _mm256_mul_pd(fscal,dy20);
1003 tz = _mm256_mul_pd(fscal,dz20);
1005 /* Update vectorial force */
1006 fix2 = _mm256_add_pd(fix2,tx);
1007 fiy2 = _mm256_add_pd(fiy2,ty);
1008 fiz2 = _mm256_add_pd(fiz2,tz);
1010 fjx0 = _mm256_add_pd(fjx0,tx);
1011 fjy0 = _mm256_add_pd(fjy0,ty);
1012 fjz0 = _mm256_add_pd(fjz0,tz);
1016 /**************************
1017 * CALCULATE INTERACTIONS *
1018 **************************/
1020 if (gmx_mm256_any_lt(rsq21,rcutoff2))
1023 /* REACTION-FIELD ELECTROSTATICS */
1024 velec = _mm256_mul_pd(qq21,_mm256_sub_pd(_mm256_add_pd(rinv21,_mm256_mul_pd(krf,rsq21)),crf));
1025 felec = _mm256_mul_pd(qq21,_mm256_sub_pd(_mm256_mul_pd(rinv21,rinvsq21),krf2));
1027 cutoff_mask = _mm256_cmp_pd(rsq21,rcutoff2,_CMP_LT_OQ);
1029 /* Update potential sum for this i atom from the interaction with this j atom. */
1030 velec = _mm256_and_pd(velec,cutoff_mask);
1031 velec = _mm256_andnot_pd(dummy_mask,velec);
1032 velecsum = _mm256_add_pd(velecsum,velec);
1036 fscal = _mm256_and_pd(fscal,cutoff_mask);
1038 fscal = _mm256_andnot_pd(dummy_mask,fscal);
1040 /* Calculate temporary vectorial force */
1041 tx = _mm256_mul_pd(fscal,dx21);
1042 ty = _mm256_mul_pd(fscal,dy21);
1043 tz = _mm256_mul_pd(fscal,dz21);
1045 /* Update vectorial force */
1046 fix2 = _mm256_add_pd(fix2,tx);
1047 fiy2 = _mm256_add_pd(fiy2,ty);
1048 fiz2 = _mm256_add_pd(fiz2,tz);
1050 fjx1 = _mm256_add_pd(fjx1,tx);
1051 fjy1 = _mm256_add_pd(fjy1,ty);
1052 fjz1 = _mm256_add_pd(fjz1,tz);
1056 /**************************
1057 * CALCULATE INTERACTIONS *
1058 **************************/
1060 if (gmx_mm256_any_lt(rsq22,rcutoff2))
1063 /* REACTION-FIELD ELECTROSTATICS */
1064 velec = _mm256_mul_pd(qq22,_mm256_sub_pd(_mm256_add_pd(rinv22,_mm256_mul_pd(krf,rsq22)),crf));
1065 felec = _mm256_mul_pd(qq22,_mm256_sub_pd(_mm256_mul_pd(rinv22,rinvsq22),krf2));
1067 cutoff_mask = _mm256_cmp_pd(rsq22,rcutoff2,_CMP_LT_OQ);
1069 /* Update potential sum for this i atom from the interaction with this j atom. */
1070 velec = _mm256_and_pd(velec,cutoff_mask);
1071 velec = _mm256_andnot_pd(dummy_mask,velec);
1072 velecsum = _mm256_add_pd(velecsum,velec);
1076 fscal = _mm256_and_pd(fscal,cutoff_mask);
1078 fscal = _mm256_andnot_pd(dummy_mask,fscal);
1080 /* Calculate temporary vectorial force */
1081 tx = _mm256_mul_pd(fscal,dx22);
1082 ty = _mm256_mul_pd(fscal,dy22);
1083 tz = _mm256_mul_pd(fscal,dz22);
1085 /* Update vectorial force */
1086 fix2 = _mm256_add_pd(fix2,tx);
1087 fiy2 = _mm256_add_pd(fiy2,ty);
1088 fiz2 = _mm256_add_pd(fiz2,tz);
1090 fjx2 = _mm256_add_pd(fjx2,tx);
1091 fjy2 = _mm256_add_pd(fjy2,ty);
1092 fjz2 = _mm256_add_pd(fjz2,tz);
1096 fjptrA = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
1097 fjptrB = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
1098 fjptrC = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
1099 fjptrD = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
1101 gmx_mm256_decrement_3rvec_4ptr_swizzle_pd(fjptrA,fjptrB,fjptrC,fjptrD,
1102 fjx0,fjy0,fjz0,fjx1,fjy1,fjz1,fjx2,fjy2,fjz2);
1104 /* Inner loop uses 324 flops */
1107 /* End of innermost loop */
1109 gmx_mm256_update_iforce_3atom_swizzle_pd(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,
1110 f+i_coord_offset,fshift+i_shift_offset);
1113 /* Update potential energies */
1114 gmx_mm256_update_1pot_pd(velecsum,kernel_data->energygrp_elec+ggid);
1116 /* Increment number of inner iterations */
1117 inneriter += j_index_end - j_index_start;
1119 /* Outer loop uses 19 flops */
1122 /* Increment number of outer iterations */
1125 /* Update outer/inner flops */
1127 inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_W3W3_VF,outeriter*19 + inneriter*324);
1130 * Gromacs nonbonded kernel: nb_kernel_ElecRFCut_VdwNone_GeomW3W3_F_avx_256_double
1131 * Electrostatics interaction: ReactionField
1132 * VdW interaction: None
1133 * Geometry: Water3-Water3
1134 * Calculate force/pot: Force
1137 nb_kernel_ElecRFCut_VdwNone_GeomW3W3_F_avx_256_double
1138 (t_nblist * gmx_restrict nlist,
1139 rvec * gmx_restrict xx,
1140 rvec * gmx_restrict ff,
1141 t_forcerec * gmx_restrict fr,
1142 t_mdatoms * gmx_restrict mdatoms,
1143 nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
1144 t_nrnb * gmx_restrict nrnb)
1146 /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
1147 * just 0 for non-waters.
1148 * Suffixes A,B,C,D refer to j loop unrolling done with AVX, e.g. for the four different
1149 * jnr indices corresponding to data put in the four positions in the SIMD register.
1151 int i_shift_offset,i_coord_offset,outeriter,inneriter;
1152 int j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
1153 int jnrA,jnrB,jnrC,jnrD;
1154 int jnrlistA,jnrlistB,jnrlistC,jnrlistD;
1155 int jnrlistE,jnrlistF,jnrlistG,jnrlistH;
1156 int j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
1157 int *iinr,*jindex,*jjnr,*shiftidx,*gid;
1158 real rcutoff_scalar;
1159 real *shiftvec,*fshift,*x,*f;
1160 real *fjptrA,*fjptrB,*fjptrC,*fjptrD;
1161 real scratch[4*DIM];
1162 __m256d tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
1163 real * vdwioffsetptr0;
1164 __m256d ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
1165 real * vdwioffsetptr1;
1166 __m256d ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
1167 real * vdwioffsetptr2;
1168 __m256d ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
1169 int vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
1170 __m256d jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
1171 int vdwjidx1A,vdwjidx1B,vdwjidx1C,vdwjidx1D;
1172 __m256d jx1,jy1,jz1,fjx1,fjy1,fjz1,jq1,isaj1;
1173 int vdwjidx2A,vdwjidx2B,vdwjidx2C,vdwjidx2D;
1174 __m256d jx2,jy2,jz2,fjx2,fjy2,fjz2,jq2,isaj2;
1175 __m256d dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
1176 __m256d dx01,dy01,dz01,rsq01,rinv01,rinvsq01,r01,qq01,c6_01,c12_01;
1177 __m256d dx02,dy02,dz02,rsq02,rinv02,rinvsq02,r02,qq02,c6_02,c12_02;
1178 __m256d dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
1179 __m256d dx11,dy11,dz11,rsq11,rinv11,rinvsq11,r11,qq11,c6_11,c12_11;
1180 __m256d dx12,dy12,dz12,rsq12,rinv12,rinvsq12,r12,qq12,c6_12,c12_12;
1181 __m256d dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
1182 __m256d dx21,dy21,dz21,rsq21,rinv21,rinvsq21,r21,qq21,c6_21,c12_21;
1183 __m256d dx22,dy22,dz22,rsq22,rinv22,rinvsq22,r22,qq22,c6_22,c12_22;
1184 __m256d velec,felec,velecsum,facel,crf,krf,krf2;
1186 __m256d dummy_mask,cutoff_mask;
1187 __m128 tmpmask0,tmpmask1;
1188 __m256d signbit = _mm256_castsi256_pd( _mm256_set1_epi32(0x80000000) );
1189 __m256d one = _mm256_set1_pd(1.0);
1190 __m256d two = _mm256_set1_pd(2.0);
1196 jindex = nlist->jindex;
1198 shiftidx = nlist->shift;
1200 shiftvec = fr->shift_vec[0];
1201 fshift = fr->fshift[0];
1202 facel = _mm256_set1_pd(fr->epsfac);
1203 charge = mdatoms->chargeA;
1204 krf = _mm256_set1_pd(fr->ic->k_rf);
1205 krf2 = _mm256_set1_pd(fr->ic->k_rf*2.0);
1206 crf = _mm256_set1_pd(fr->ic->c_rf);
1208 /* Setup water-specific parameters */
1209 inr = nlist->iinr[0];
1210 iq0 = _mm256_mul_pd(facel,_mm256_set1_pd(charge[inr+0]));
1211 iq1 = _mm256_mul_pd(facel,_mm256_set1_pd(charge[inr+1]));
1212 iq2 = _mm256_mul_pd(facel,_mm256_set1_pd(charge[inr+2]));
1214 jq0 = _mm256_set1_pd(charge[inr+0]);
1215 jq1 = _mm256_set1_pd(charge[inr+1]);
1216 jq2 = _mm256_set1_pd(charge[inr+2]);
1217 qq00 = _mm256_mul_pd(iq0,jq0);
1218 qq01 = _mm256_mul_pd(iq0,jq1);
1219 qq02 = _mm256_mul_pd(iq0,jq2);
1220 qq10 = _mm256_mul_pd(iq1,jq0);
1221 qq11 = _mm256_mul_pd(iq1,jq1);
1222 qq12 = _mm256_mul_pd(iq1,jq2);
1223 qq20 = _mm256_mul_pd(iq2,jq0);
1224 qq21 = _mm256_mul_pd(iq2,jq1);
1225 qq22 = _mm256_mul_pd(iq2,jq2);
1227 /* When we use explicit cutoffs the value must be identical for elec and VdW, so use elec as an arbitrary choice */
1228 rcutoff_scalar = fr->rcoulomb;
1229 rcutoff = _mm256_set1_pd(rcutoff_scalar);
1230 rcutoff2 = _mm256_mul_pd(rcutoff,rcutoff);
1232 /* Avoid stupid compiler warnings */
1233 jnrA = jnrB = jnrC = jnrD = 0;
1234 j_coord_offsetA = 0;
1235 j_coord_offsetB = 0;
1236 j_coord_offsetC = 0;
1237 j_coord_offsetD = 0;
1242 for(iidx=0;iidx<4*DIM;iidx++)
1244 scratch[iidx] = 0.0;
1247 /* Start outer loop over neighborlists */
1248 for(iidx=0; iidx<nri; iidx++)
1250 /* Load shift vector for this list */
1251 i_shift_offset = DIM*shiftidx[iidx];
1253 /* Load limits for loop over neighbors */
1254 j_index_start = jindex[iidx];
1255 j_index_end = jindex[iidx+1];
1257 /* Get outer coordinate index */
1259 i_coord_offset = DIM*inr;
1261 /* Load i particle coords and add shift vector */
1262 gmx_mm256_load_shift_and_3rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset,
1263 &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2);
1265 fix0 = _mm256_setzero_pd();
1266 fiy0 = _mm256_setzero_pd();
1267 fiz0 = _mm256_setzero_pd();
1268 fix1 = _mm256_setzero_pd();
1269 fiy1 = _mm256_setzero_pd();
1270 fiz1 = _mm256_setzero_pd();
1271 fix2 = _mm256_setzero_pd();
1272 fiy2 = _mm256_setzero_pd();
1273 fiz2 = _mm256_setzero_pd();
1275 /* Start inner kernel loop */
1276 for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
1279 /* Get j neighbor index, and coordinate index */
1281 jnrB = jjnr[jidx+1];
1282 jnrC = jjnr[jidx+2];
1283 jnrD = jjnr[jidx+3];
1284 j_coord_offsetA = DIM*jnrA;
1285 j_coord_offsetB = DIM*jnrB;
1286 j_coord_offsetC = DIM*jnrC;
1287 j_coord_offsetD = DIM*jnrD;
1289 /* load j atom coordinates */
1290 gmx_mm256_load_3rvec_4ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
1291 x+j_coord_offsetC,x+j_coord_offsetD,
1292 &jx0,&jy0,&jz0,&jx1,&jy1,&jz1,&jx2,&jy2,&jz2);
1294 /* Calculate displacement vector */
1295 dx00 = _mm256_sub_pd(ix0,jx0);
1296 dy00 = _mm256_sub_pd(iy0,jy0);
1297 dz00 = _mm256_sub_pd(iz0,jz0);
1298 dx01 = _mm256_sub_pd(ix0,jx1);
1299 dy01 = _mm256_sub_pd(iy0,jy1);
1300 dz01 = _mm256_sub_pd(iz0,jz1);
1301 dx02 = _mm256_sub_pd(ix0,jx2);
1302 dy02 = _mm256_sub_pd(iy0,jy2);
1303 dz02 = _mm256_sub_pd(iz0,jz2);
1304 dx10 = _mm256_sub_pd(ix1,jx0);
1305 dy10 = _mm256_sub_pd(iy1,jy0);
1306 dz10 = _mm256_sub_pd(iz1,jz0);
1307 dx11 = _mm256_sub_pd(ix1,jx1);
1308 dy11 = _mm256_sub_pd(iy1,jy1);
1309 dz11 = _mm256_sub_pd(iz1,jz1);
1310 dx12 = _mm256_sub_pd(ix1,jx2);
1311 dy12 = _mm256_sub_pd(iy1,jy2);
1312 dz12 = _mm256_sub_pd(iz1,jz2);
1313 dx20 = _mm256_sub_pd(ix2,jx0);
1314 dy20 = _mm256_sub_pd(iy2,jy0);
1315 dz20 = _mm256_sub_pd(iz2,jz0);
1316 dx21 = _mm256_sub_pd(ix2,jx1);
1317 dy21 = _mm256_sub_pd(iy2,jy1);
1318 dz21 = _mm256_sub_pd(iz2,jz1);
1319 dx22 = _mm256_sub_pd(ix2,jx2);
1320 dy22 = _mm256_sub_pd(iy2,jy2);
1321 dz22 = _mm256_sub_pd(iz2,jz2);
1323 /* Calculate squared distance and things based on it */
1324 rsq00 = gmx_mm256_calc_rsq_pd(dx00,dy00,dz00);
1325 rsq01 = gmx_mm256_calc_rsq_pd(dx01,dy01,dz01);
1326 rsq02 = gmx_mm256_calc_rsq_pd(dx02,dy02,dz02);
1327 rsq10 = gmx_mm256_calc_rsq_pd(dx10,dy10,dz10);
1328 rsq11 = gmx_mm256_calc_rsq_pd(dx11,dy11,dz11);
1329 rsq12 = gmx_mm256_calc_rsq_pd(dx12,dy12,dz12);
1330 rsq20 = gmx_mm256_calc_rsq_pd(dx20,dy20,dz20);
1331 rsq21 = gmx_mm256_calc_rsq_pd(dx21,dy21,dz21);
1332 rsq22 = gmx_mm256_calc_rsq_pd(dx22,dy22,dz22);
1334 rinv00 = gmx_mm256_invsqrt_pd(rsq00);
1335 rinv01 = gmx_mm256_invsqrt_pd(rsq01);
1336 rinv02 = gmx_mm256_invsqrt_pd(rsq02);
1337 rinv10 = gmx_mm256_invsqrt_pd(rsq10);
1338 rinv11 = gmx_mm256_invsqrt_pd(rsq11);
1339 rinv12 = gmx_mm256_invsqrt_pd(rsq12);
1340 rinv20 = gmx_mm256_invsqrt_pd(rsq20);
1341 rinv21 = gmx_mm256_invsqrt_pd(rsq21);
1342 rinv22 = gmx_mm256_invsqrt_pd(rsq22);
1344 rinvsq00 = _mm256_mul_pd(rinv00,rinv00);
1345 rinvsq01 = _mm256_mul_pd(rinv01,rinv01);
1346 rinvsq02 = _mm256_mul_pd(rinv02,rinv02);
1347 rinvsq10 = _mm256_mul_pd(rinv10,rinv10);
1348 rinvsq11 = _mm256_mul_pd(rinv11,rinv11);
1349 rinvsq12 = _mm256_mul_pd(rinv12,rinv12);
1350 rinvsq20 = _mm256_mul_pd(rinv20,rinv20);
1351 rinvsq21 = _mm256_mul_pd(rinv21,rinv21);
1352 rinvsq22 = _mm256_mul_pd(rinv22,rinv22);
1354 fjx0 = _mm256_setzero_pd();
1355 fjy0 = _mm256_setzero_pd();
1356 fjz0 = _mm256_setzero_pd();
1357 fjx1 = _mm256_setzero_pd();
1358 fjy1 = _mm256_setzero_pd();
1359 fjz1 = _mm256_setzero_pd();
1360 fjx2 = _mm256_setzero_pd();
1361 fjy2 = _mm256_setzero_pd();
1362 fjz2 = _mm256_setzero_pd();
1364 /**************************
1365 * CALCULATE INTERACTIONS *
1366 **************************/
1368 if (gmx_mm256_any_lt(rsq00,rcutoff2))
1371 /* REACTION-FIELD ELECTROSTATICS */
1372 felec = _mm256_mul_pd(qq00,_mm256_sub_pd(_mm256_mul_pd(rinv00,rinvsq00),krf2));
1374 cutoff_mask = _mm256_cmp_pd(rsq00,rcutoff2,_CMP_LT_OQ);
1378 fscal = _mm256_and_pd(fscal,cutoff_mask);
1380 /* Calculate temporary vectorial force */
1381 tx = _mm256_mul_pd(fscal,dx00);
1382 ty = _mm256_mul_pd(fscal,dy00);
1383 tz = _mm256_mul_pd(fscal,dz00);
1385 /* Update vectorial force */
1386 fix0 = _mm256_add_pd(fix0,tx);
1387 fiy0 = _mm256_add_pd(fiy0,ty);
1388 fiz0 = _mm256_add_pd(fiz0,tz);
1390 fjx0 = _mm256_add_pd(fjx0,tx);
1391 fjy0 = _mm256_add_pd(fjy0,ty);
1392 fjz0 = _mm256_add_pd(fjz0,tz);
1396 /**************************
1397 * CALCULATE INTERACTIONS *
1398 **************************/
1400 if (gmx_mm256_any_lt(rsq01,rcutoff2))
1403 /* REACTION-FIELD ELECTROSTATICS */
1404 felec = _mm256_mul_pd(qq01,_mm256_sub_pd(_mm256_mul_pd(rinv01,rinvsq01),krf2));
1406 cutoff_mask = _mm256_cmp_pd(rsq01,rcutoff2,_CMP_LT_OQ);
1410 fscal = _mm256_and_pd(fscal,cutoff_mask);
1412 /* Calculate temporary vectorial force */
1413 tx = _mm256_mul_pd(fscal,dx01);
1414 ty = _mm256_mul_pd(fscal,dy01);
1415 tz = _mm256_mul_pd(fscal,dz01);
1417 /* Update vectorial force */
1418 fix0 = _mm256_add_pd(fix0,tx);
1419 fiy0 = _mm256_add_pd(fiy0,ty);
1420 fiz0 = _mm256_add_pd(fiz0,tz);
1422 fjx1 = _mm256_add_pd(fjx1,tx);
1423 fjy1 = _mm256_add_pd(fjy1,ty);
1424 fjz1 = _mm256_add_pd(fjz1,tz);
1428 /**************************
1429 * CALCULATE INTERACTIONS *
1430 **************************/
1432 if (gmx_mm256_any_lt(rsq02,rcutoff2))
1435 /* REACTION-FIELD ELECTROSTATICS */
1436 felec = _mm256_mul_pd(qq02,_mm256_sub_pd(_mm256_mul_pd(rinv02,rinvsq02),krf2));
1438 cutoff_mask = _mm256_cmp_pd(rsq02,rcutoff2,_CMP_LT_OQ);
1442 fscal = _mm256_and_pd(fscal,cutoff_mask);
1444 /* Calculate temporary vectorial force */
1445 tx = _mm256_mul_pd(fscal,dx02);
1446 ty = _mm256_mul_pd(fscal,dy02);
1447 tz = _mm256_mul_pd(fscal,dz02);
1449 /* Update vectorial force */
1450 fix0 = _mm256_add_pd(fix0,tx);
1451 fiy0 = _mm256_add_pd(fiy0,ty);
1452 fiz0 = _mm256_add_pd(fiz0,tz);
1454 fjx2 = _mm256_add_pd(fjx2,tx);
1455 fjy2 = _mm256_add_pd(fjy2,ty);
1456 fjz2 = _mm256_add_pd(fjz2,tz);
1460 /**************************
1461 * CALCULATE INTERACTIONS *
1462 **************************/
1464 if (gmx_mm256_any_lt(rsq10,rcutoff2))
1467 /* REACTION-FIELD ELECTROSTATICS */
1468 felec = _mm256_mul_pd(qq10,_mm256_sub_pd(_mm256_mul_pd(rinv10,rinvsq10),krf2));
1470 cutoff_mask = _mm256_cmp_pd(rsq10,rcutoff2,_CMP_LT_OQ);
1474 fscal = _mm256_and_pd(fscal,cutoff_mask);
1476 /* Calculate temporary vectorial force */
1477 tx = _mm256_mul_pd(fscal,dx10);
1478 ty = _mm256_mul_pd(fscal,dy10);
1479 tz = _mm256_mul_pd(fscal,dz10);
1481 /* Update vectorial force */
1482 fix1 = _mm256_add_pd(fix1,tx);
1483 fiy1 = _mm256_add_pd(fiy1,ty);
1484 fiz1 = _mm256_add_pd(fiz1,tz);
1486 fjx0 = _mm256_add_pd(fjx0,tx);
1487 fjy0 = _mm256_add_pd(fjy0,ty);
1488 fjz0 = _mm256_add_pd(fjz0,tz);
1492 /**************************
1493 * CALCULATE INTERACTIONS *
1494 **************************/
1496 if (gmx_mm256_any_lt(rsq11,rcutoff2))
1499 /* REACTION-FIELD ELECTROSTATICS */
1500 felec = _mm256_mul_pd(qq11,_mm256_sub_pd(_mm256_mul_pd(rinv11,rinvsq11),krf2));
1502 cutoff_mask = _mm256_cmp_pd(rsq11,rcutoff2,_CMP_LT_OQ);
1506 fscal = _mm256_and_pd(fscal,cutoff_mask);
1508 /* Calculate temporary vectorial force */
1509 tx = _mm256_mul_pd(fscal,dx11);
1510 ty = _mm256_mul_pd(fscal,dy11);
1511 tz = _mm256_mul_pd(fscal,dz11);
1513 /* Update vectorial force */
1514 fix1 = _mm256_add_pd(fix1,tx);
1515 fiy1 = _mm256_add_pd(fiy1,ty);
1516 fiz1 = _mm256_add_pd(fiz1,tz);
1518 fjx1 = _mm256_add_pd(fjx1,tx);
1519 fjy1 = _mm256_add_pd(fjy1,ty);
1520 fjz1 = _mm256_add_pd(fjz1,tz);
1524 /**************************
1525 * CALCULATE INTERACTIONS *
1526 **************************/
1528 if (gmx_mm256_any_lt(rsq12,rcutoff2))
1531 /* REACTION-FIELD ELECTROSTATICS */
1532 felec = _mm256_mul_pd(qq12,_mm256_sub_pd(_mm256_mul_pd(rinv12,rinvsq12),krf2));
1534 cutoff_mask = _mm256_cmp_pd(rsq12,rcutoff2,_CMP_LT_OQ);
1538 fscal = _mm256_and_pd(fscal,cutoff_mask);
1540 /* Calculate temporary vectorial force */
1541 tx = _mm256_mul_pd(fscal,dx12);
1542 ty = _mm256_mul_pd(fscal,dy12);
1543 tz = _mm256_mul_pd(fscal,dz12);
1545 /* Update vectorial force */
1546 fix1 = _mm256_add_pd(fix1,tx);
1547 fiy1 = _mm256_add_pd(fiy1,ty);
1548 fiz1 = _mm256_add_pd(fiz1,tz);
1550 fjx2 = _mm256_add_pd(fjx2,tx);
1551 fjy2 = _mm256_add_pd(fjy2,ty);
1552 fjz2 = _mm256_add_pd(fjz2,tz);
1556 /**************************
1557 * CALCULATE INTERACTIONS *
1558 **************************/
1560 if (gmx_mm256_any_lt(rsq20,rcutoff2))
1563 /* REACTION-FIELD ELECTROSTATICS */
1564 felec = _mm256_mul_pd(qq20,_mm256_sub_pd(_mm256_mul_pd(rinv20,rinvsq20),krf2));
1566 cutoff_mask = _mm256_cmp_pd(rsq20,rcutoff2,_CMP_LT_OQ);
1570 fscal = _mm256_and_pd(fscal,cutoff_mask);
1572 /* Calculate temporary vectorial force */
1573 tx = _mm256_mul_pd(fscal,dx20);
1574 ty = _mm256_mul_pd(fscal,dy20);
1575 tz = _mm256_mul_pd(fscal,dz20);
1577 /* Update vectorial force */
1578 fix2 = _mm256_add_pd(fix2,tx);
1579 fiy2 = _mm256_add_pd(fiy2,ty);
1580 fiz2 = _mm256_add_pd(fiz2,tz);
1582 fjx0 = _mm256_add_pd(fjx0,tx);
1583 fjy0 = _mm256_add_pd(fjy0,ty);
1584 fjz0 = _mm256_add_pd(fjz0,tz);
1588 /**************************
1589 * CALCULATE INTERACTIONS *
1590 **************************/
1592 if (gmx_mm256_any_lt(rsq21,rcutoff2))
1595 /* REACTION-FIELD ELECTROSTATICS */
1596 felec = _mm256_mul_pd(qq21,_mm256_sub_pd(_mm256_mul_pd(rinv21,rinvsq21),krf2));
1598 cutoff_mask = _mm256_cmp_pd(rsq21,rcutoff2,_CMP_LT_OQ);
1602 fscal = _mm256_and_pd(fscal,cutoff_mask);
1604 /* Calculate temporary vectorial force */
1605 tx = _mm256_mul_pd(fscal,dx21);
1606 ty = _mm256_mul_pd(fscal,dy21);
1607 tz = _mm256_mul_pd(fscal,dz21);
1609 /* Update vectorial force */
1610 fix2 = _mm256_add_pd(fix2,tx);
1611 fiy2 = _mm256_add_pd(fiy2,ty);
1612 fiz2 = _mm256_add_pd(fiz2,tz);
1614 fjx1 = _mm256_add_pd(fjx1,tx);
1615 fjy1 = _mm256_add_pd(fjy1,ty);
1616 fjz1 = _mm256_add_pd(fjz1,tz);
1620 /**************************
1621 * CALCULATE INTERACTIONS *
1622 **************************/
1624 if (gmx_mm256_any_lt(rsq22,rcutoff2))
1627 /* REACTION-FIELD ELECTROSTATICS */
1628 felec = _mm256_mul_pd(qq22,_mm256_sub_pd(_mm256_mul_pd(rinv22,rinvsq22),krf2));
1630 cutoff_mask = _mm256_cmp_pd(rsq22,rcutoff2,_CMP_LT_OQ);
1634 fscal = _mm256_and_pd(fscal,cutoff_mask);
1636 /* Calculate temporary vectorial force */
1637 tx = _mm256_mul_pd(fscal,dx22);
1638 ty = _mm256_mul_pd(fscal,dy22);
1639 tz = _mm256_mul_pd(fscal,dz22);
1641 /* Update vectorial force */
1642 fix2 = _mm256_add_pd(fix2,tx);
1643 fiy2 = _mm256_add_pd(fiy2,ty);
1644 fiz2 = _mm256_add_pd(fiz2,tz);
1646 fjx2 = _mm256_add_pd(fjx2,tx);
1647 fjy2 = _mm256_add_pd(fjy2,ty);
1648 fjz2 = _mm256_add_pd(fjz2,tz);
1652 fjptrA = f+j_coord_offsetA;
1653 fjptrB = f+j_coord_offsetB;
1654 fjptrC = f+j_coord_offsetC;
1655 fjptrD = f+j_coord_offsetD;
1657 gmx_mm256_decrement_3rvec_4ptr_swizzle_pd(fjptrA,fjptrB,fjptrC,fjptrD,
1658 fjx0,fjy0,fjz0,fjx1,fjy1,fjz1,fjx2,fjy2,fjz2);
1660 /* Inner loop uses 270 flops */
1663 if(jidx<j_index_end)
1666 /* Get j neighbor index, and coordinate index */
1667 jnrlistA = jjnr[jidx];
1668 jnrlistB = jjnr[jidx+1];
1669 jnrlistC = jjnr[jidx+2];
1670 jnrlistD = jjnr[jidx+3];
1671 /* Sign of each element will be negative for non-real atoms.
1672 * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
1673 * so use it as val = _mm_andnot_pd(mask,val) to clear dummy entries.
1675 tmpmask0 = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
1677 tmpmask1 = _mm_permute_ps(tmpmask0,_GMX_MM_PERMUTE(3,3,2,2));
1678 tmpmask0 = _mm_permute_ps(tmpmask0,_GMX_MM_PERMUTE(1,1,0,0));
1679 dummy_mask = _mm256_castps_pd(gmx_mm256_set_m128(tmpmask1,tmpmask0));
1681 jnrA = (jnrlistA>=0) ? jnrlistA : 0;
1682 jnrB = (jnrlistB>=0) ? jnrlistB : 0;
1683 jnrC = (jnrlistC>=0) ? jnrlistC : 0;
1684 jnrD = (jnrlistD>=0) ? jnrlistD : 0;
1685 j_coord_offsetA = DIM*jnrA;
1686 j_coord_offsetB = DIM*jnrB;
1687 j_coord_offsetC = DIM*jnrC;
1688 j_coord_offsetD = DIM*jnrD;
1690 /* load j atom coordinates */
1691 gmx_mm256_load_3rvec_4ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
1692 x+j_coord_offsetC,x+j_coord_offsetD,
1693 &jx0,&jy0,&jz0,&jx1,&jy1,&jz1,&jx2,&jy2,&jz2);
1695 /* Calculate displacement vector */
1696 dx00 = _mm256_sub_pd(ix0,jx0);
1697 dy00 = _mm256_sub_pd(iy0,jy0);
1698 dz00 = _mm256_sub_pd(iz0,jz0);
1699 dx01 = _mm256_sub_pd(ix0,jx1);
1700 dy01 = _mm256_sub_pd(iy0,jy1);
1701 dz01 = _mm256_sub_pd(iz0,jz1);
1702 dx02 = _mm256_sub_pd(ix0,jx2);
1703 dy02 = _mm256_sub_pd(iy0,jy2);
1704 dz02 = _mm256_sub_pd(iz0,jz2);
1705 dx10 = _mm256_sub_pd(ix1,jx0);
1706 dy10 = _mm256_sub_pd(iy1,jy0);
1707 dz10 = _mm256_sub_pd(iz1,jz0);
1708 dx11 = _mm256_sub_pd(ix1,jx1);
1709 dy11 = _mm256_sub_pd(iy1,jy1);
1710 dz11 = _mm256_sub_pd(iz1,jz1);
1711 dx12 = _mm256_sub_pd(ix1,jx2);
1712 dy12 = _mm256_sub_pd(iy1,jy2);
1713 dz12 = _mm256_sub_pd(iz1,jz2);
1714 dx20 = _mm256_sub_pd(ix2,jx0);
1715 dy20 = _mm256_sub_pd(iy2,jy0);
1716 dz20 = _mm256_sub_pd(iz2,jz0);
1717 dx21 = _mm256_sub_pd(ix2,jx1);
1718 dy21 = _mm256_sub_pd(iy2,jy1);
1719 dz21 = _mm256_sub_pd(iz2,jz1);
1720 dx22 = _mm256_sub_pd(ix2,jx2);
1721 dy22 = _mm256_sub_pd(iy2,jy2);
1722 dz22 = _mm256_sub_pd(iz2,jz2);
1724 /* Calculate squared distance and things based on it */
1725 rsq00 = gmx_mm256_calc_rsq_pd(dx00,dy00,dz00);
1726 rsq01 = gmx_mm256_calc_rsq_pd(dx01,dy01,dz01);
1727 rsq02 = gmx_mm256_calc_rsq_pd(dx02,dy02,dz02);
1728 rsq10 = gmx_mm256_calc_rsq_pd(dx10,dy10,dz10);
1729 rsq11 = gmx_mm256_calc_rsq_pd(dx11,dy11,dz11);
1730 rsq12 = gmx_mm256_calc_rsq_pd(dx12,dy12,dz12);
1731 rsq20 = gmx_mm256_calc_rsq_pd(dx20,dy20,dz20);
1732 rsq21 = gmx_mm256_calc_rsq_pd(dx21,dy21,dz21);
1733 rsq22 = gmx_mm256_calc_rsq_pd(dx22,dy22,dz22);
1735 rinv00 = gmx_mm256_invsqrt_pd(rsq00);
1736 rinv01 = gmx_mm256_invsqrt_pd(rsq01);
1737 rinv02 = gmx_mm256_invsqrt_pd(rsq02);
1738 rinv10 = gmx_mm256_invsqrt_pd(rsq10);
1739 rinv11 = gmx_mm256_invsqrt_pd(rsq11);
1740 rinv12 = gmx_mm256_invsqrt_pd(rsq12);
1741 rinv20 = gmx_mm256_invsqrt_pd(rsq20);
1742 rinv21 = gmx_mm256_invsqrt_pd(rsq21);
1743 rinv22 = gmx_mm256_invsqrt_pd(rsq22);
1745 rinvsq00 = _mm256_mul_pd(rinv00,rinv00);
1746 rinvsq01 = _mm256_mul_pd(rinv01,rinv01);
1747 rinvsq02 = _mm256_mul_pd(rinv02,rinv02);
1748 rinvsq10 = _mm256_mul_pd(rinv10,rinv10);
1749 rinvsq11 = _mm256_mul_pd(rinv11,rinv11);
1750 rinvsq12 = _mm256_mul_pd(rinv12,rinv12);
1751 rinvsq20 = _mm256_mul_pd(rinv20,rinv20);
1752 rinvsq21 = _mm256_mul_pd(rinv21,rinv21);
1753 rinvsq22 = _mm256_mul_pd(rinv22,rinv22);
1755 fjx0 = _mm256_setzero_pd();
1756 fjy0 = _mm256_setzero_pd();
1757 fjz0 = _mm256_setzero_pd();
1758 fjx1 = _mm256_setzero_pd();
1759 fjy1 = _mm256_setzero_pd();
1760 fjz1 = _mm256_setzero_pd();
1761 fjx2 = _mm256_setzero_pd();
1762 fjy2 = _mm256_setzero_pd();
1763 fjz2 = _mm256_setzero_pd();
1765 /**************************
1766 * CALCULATE INTERACTIONS *
1767 **************************/
1769 if (gmx_mm256_any_lt(rsq00,rcutoff2))
1772 /* REACTION-FIELD ELECTROSTATICS */
1773 felec = _mm256_mul_pd(qq00,_mm256_sub_pd(_mm256_mul_pd(rinv00,rinvsq00),krf2));
1775 cutoff_mask = _mm256_cmp_pd(rsq00,rcutoff2,_CMP_LT_OQ);
1779 fscal = _mm256_and_pd(fscal,cutoff_mask);
1781 fscal = _mm256_andnot_pd(dummy_mask,fscal);
1783 /* Calculate temporary vectorial force */
1784 tx = _mm256_mul_pd(fscal,dx00);
1785 ty = _mm256_mul_pd(fscal,dy00);
1786 tz = _mm256_mul_pd(fscal,dz00);
1788 /* Update vectorial force */
1789 fix0 = _mm256_add_pd(fix0,tx);
1790 fiy0 = _mm256_add_pd(fiy0,ty);
1791 fiz0 = _mm256_add_pd(fiz0,tz);
1793 fjx0 = _mm256_add_pd(fjx0,tx);
1794 fjy0 = _mm256_add_pd(fjy0,ty);
1795 fjz0 = _mm256_add_pd(fjz0,tz);
1799 /**************************
1800 * CALCULATE INTERACTIONS *
1801 **************************/
1803 if (gmx_mm256_any_lt(rsq01,rcutoff2))
1806 /* REACTION-FIELD ELECTROSTATICS */
1807 felec = _mm256_mul_pd(qq01,_mm256_sub_pd(_mm256_mul_pd(rinv01,rinvsq01),krf2));
1809 cutoff_mask = _mm256_cmp_pd(rsq01,rcutoff2,_CMP_LT_OQ);
1813 fscal = _mm256_and_pd(fscal,cutoff_mask);
1815 fscal = _mm256_andnot_pd(dummy_mask,fscal);
1817 /* Calculate temporary vectorial force */
1818 tx = _mm256_mul_pd(fscal,dx01);
1819 ty = _mm256_mul_pd(fscal,dy01);
1820 tz = _mm256_mul_pd(fscal,dz01);
1822 /* Update vectorial force */
1823 fix0 = _mm256_add_pd(fix0,tx);
1824 fiy0 = _mm256_add_pd(fiy0,ty);
1825 fiz0 = _mm256_add_pd(fiz0,tz);
1827 fjx1 = _mm256_add_pd(fjx1,tx);
1828 fjy1 = _mm256_add_pd(fjy1,ty);
1829 fjz1 = _mm256_add_pd(fjz1,tz);
1833 /**************************
1834 * CALCULATE INTERACTIONS *
1835 **************************/
1837 if (gmx_mm256_any_lt(rsq02,rcutoff2))
1840 /* REACTION-FIELD ELECTROSTATICS */
1841 felec = _mm256_mul_pd(qq02,_mm256_sub_pd(_mm256_mul_pd(rinv02,rinvsq02),krf2));
1843 cutoff_mask = _mm256_cmp_pd(rsq02,rcutoff2,_CMP_LT_OQ);
1847 fscal = _mm256_and_pd(fscal,cutoff_mask);
1849 fscal = _mm256_andnot_pd(dummy_mask,fscal);
1851 /* Calculate temporary vectorial force */
1852 tx = _mm256_mul_pd(fscal,dx02);
1853 ty = _mm256_mul_pd(fscal,dy02);
1854 tz = _mm256_mul_pd(fscal,dz02);
1856 /* Update vectorial force */
1857 fix0 = _mm256_add_pd(fix0,tx);
1858 fiy0 = _mm256_add_pd(fiy0,ty);
1859 fiz0 = _mm256_add_pd(fiz0,tz);
1861 fjx2 = _mm256_add_pd(fjx2,tx);
1862 fjy2 = _mm256_add_pd(fjy2,ty);
1863 fjz2 = _mm256_add_pd(fjz2,tz);
1867 /**************************
1868 * CALCULATE INTERACTIONS *
1869 **************************/
1871 if (gmx_mm256_any_lt(rsq10,rcutoff2))
1874 /* REACTION-FIELD ELECTROSTATICS */
1875 felec = _mm256_mul_pd(qq10,_mm256_sub_pd(_mm256_mul_pd(rinv10,rinvsq10),krf2));
1877 cutoff_mask = _mm256_cmp_pd(rsq10,rcutoff2,_CMP_LT_OQ);
1881 fscal = _mm256_and_pd(fscal,cutoff_mask);
1883 fscal = _mm256_andnot_pd(dummy_mask,fscal);
1885 /* Calculate temporary vectorial force */
1886 tx = _mm256_mul_pd(fscal,dx10);
1887 ty = _mm256_mul_pd(fscal,dy10);
1888 tz = _mm256_mul_pd(fscal,dz10);
1890 /* Update vectorial force */
1891 fix1 = _mm256_add_pd(fix1,tx);
1892 fiy1 = _mm256_add_pd(fiy1,ty);
1893 fiz1 = _mm256_add_pd(fiz1,tz);
1895 fjx0 = _mm256_add_pd(fjx0,tx);
1896 fjy0 = _mm256_add_pd(fjy0,ty);
1897 fjz0 = _mm256_add_pd(fjz0,tz);
1901 /**************************
1902 * CALCULATE INTERACTIONS *
1903 **************************/
1905 if (gmx_mm256_any_lt(rsq11,rcutoff2))
1908 /* REACTION-FIELD ELECTROSTATICS */
1909 felec = _mm256_mul_pd(qq11,_mm256_sub_pd(_mm256_mul_pd(rinv11,rinvsq11),krf2));
1911 cutoff_mask = _mm256_cmp_pd(rsq11,rcutoff2,_CMP_LT_OQ);
1915 fscal = _mm256_and_pd(fscal,cutoff_mask);
1917 fscal = _mm256_andnot_pd(dummy_mask,fscal);
1919 /* Calculate temporary vectorial force */
1920 tx = _mm256_mul_pd(fscal,dx11);
1921 ty = _mm256_mul_pd(fscal,dy11);
1922 tz = _mm256_mul_pd(fscal,dz11);
1924 /* Update vectorial force */
1925 fix1 = _mm256_add_pd(fix1,tx);
1926 fiy1 = _mm256_add_pd(fiy1,ty);
1927 fiz1 = _mm256_add_pd(fiz1,tz);
1929 fjx1 = _mm256_add_pd(fjx1,tx);
1930 fjy1 = _mm256_add_pd(fjy1,ty);
1931 fjz1 = _mm256_add_pd(fjz1,tz);
1935 /**************************
1936 * CALCULATE INTERACTIONS *
1937 **************************/
1939 if (gmx_mm256_any_lt(rsq12,rcutoff2))
1942 /* REACTION-FIELD ELECTROSTATICS */
1943 felec = _mm256_mul_pd(qq12,_mm256_sub_pd(_mm256_mul_pd(rinv12,rinvsq12),krf2));
1945 cutoff_mask = _mm256_cmp_pd(rsq12,rcutoff2,_CMP_LT_OQ);
1949 fscal = _mm256_and_pd(fscal,cutoff_mask);
1951 fscal = _mm256_andnot_pd(dummy_mask,fscal);
1953 /* Calculate temporary vectorial force */
1954 tx = _mm256_mul_pd(fscal,dx12);
1955 ty = _mm256_mul_pd(fscal,dy12);
1956 tz = _mm256_mul_pd(fscal,dz12);
1958 /* Update vectorial force */
1959 fix1 = _mm256_add_pd(fix1,tx);
1960 fiy1 = _mm256_add_pd(fiy1,ty);
1961 fiz1 = _mm256_add_pd(fiz1,tz);
1963 fjx2 = _mm256_add_pd(fjx2,tx);
1964 fjy2 = _mm256_add_pd(fjy2,ty);
1965 fjz2 = _mm256_add_pd(fjz2,tz);
1969 /**************************
1970 * CALCULATE INTERACTIONS *
1971 **************************/
1973 if (gmx_mm256_any_lt(rsq20,rcutoff2))
1976 /* REACTION-FIELD ELECTROSTATICS */
1977 felec = _mm256_mul_pd(qq20,_mm256_sub_pd(_mm256_mul_pd(rinv20,rinvsq20),krf2));
1979 cutoff_mask = _mm256_cmp_pd(rsq20,rcutoff2,_CMP_LT_OQ);
1983 fscal = _mm256_and_pd(fscal,cutoff_mask);
1985 fscal = _mm256_andnot_pd(dummy_mask,fscal);
1987 /* Calculate temporary vectorial force */
1988 tx = _mm256_mul_pd(fscal,dx20);
1989 ty = _mm256_mul_pd(fscal,dy20);
1990 tz = _mm256_mul_pd(fscal,dz20);
1992 /* Update vectorial force */
1993 fix2 = _mm256_add_pd(fix2,tx);
1994 fiy2 = _mm256_add_pd(fiy2,ty);
1995 fiz2 = _mm256_add_pd(fiz2,tz);
1997 fjx0 = _mm256_add_pd(fjx0,tx);
1998 fjy0 = _mm256_add_pd(fjy0,ty);
1999 fjz0 = _mm256_add_pd(fjz0,tz);
2003 /**************************
2004 * CALCULATE INTERACTIONS *
2005 **************************/
2007 if (gmx_mm256_any_lt(rsq21,rcutoff2))
2010 /* REACTION-FIELD ELECTROSTATICS */
2011 felec = _mm256_mul_pd(qq21,_mm256_sub_pd(_mm256_mul_pd(rinv21,rinvsq21),krf2));
2013 cutoff_mask = _mm256_cmp_pd(rsq21,rcutoff2,_CMP_LT_OQ);
2017 fscal = _mm256_and_pd(fscal,cutoff_mask);
2019 fscal = _mm256_andnot_pd(dummy_mask,fscal);
2021 /* Calculate temporary vectorial force */
2022 tx = _mm256_mul_pd(fscal,dx21);
2023 ty = _mm256_mul_pd(fscal,dy21);
2024 tz = _mm256_mul_pd(fscal,dz21);
2026 /* Update vectorial force */
2027 fix2 = _mm256_add_pd(fix2,tx);
2028 fiy2 = _mm256_add_pd(fiy2,ty);
2029 fiz2 = _mm256_add_pd(fiz2,tz);
2031 fjx1 = _mm256_add_pd(fjx1,tx);
2032 fjy1 = _mm256_add_pd(fjy1,ty);
2033 fjz1 = _mm256_add_pd(fjz1,tz);
2037 /**************************
2038 * CALCULATE INTERACTIONS *
2039 **************************/
2041 if (gmx_mm256_any_lt(rsq22,rcutoff2))
2044 /* REACTION-FIELD ELECTROSTATICS */
2045 felec = _mm256_mul_pd(qq22,_mm256_sub_pd(_mm256_mul_pd(rinv22,rinvsq22),krf2));
2047 cutoff_mask = _mm256_cmp_pd(rsq22,rcutoff2,_CMP_LT_OQ);
2051 fscal = _mm256_and_pd(fscal,cutoff_mask);
2053 fscal = _mm256_andnot_pd(dummy_mask,fscal);
2055 /* Calculate temporary vectorial force */
2056 tx = _mm256_mul_pd(fscal,dx22);
2057 ty = _mm256_mul_pd(fscal,dy22);
2058 tz = _mm256_mul_pd(fscal,dz22);
2060 /* Update vectorial force */
2061 fix2 = _mm256_add_pd(fix2,tx);
2062 fiy2 = _mm256_add_pd(fiy2,ty);
2063 fiz2 = _mm256_add_pd(fiz2,tz);
2065 fjx2 = _mm256_add_pd(fjx2,tx);
2066 fjy2 = _mm256_add_pd(fjy2,ty);
2067 fjz2 = _mm256_add_pd(fjz2,tz);
2071 fjptrA = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
2072 fjptrB = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
2073 fjptrC = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
2074 fjptrD = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
2076 gmx_mm256_decrement_3rvec_4ptr_swizzle_pd(fjptrA,fjptrB,fjptrC,fjptrD,
2077 fjx0,fjy0,fjz0,fjx1,fjy1,fjz1,fjx2,fjy2,fjz2);
2079 /* Inner loop uses 270 flops */
2082 /* End of innermost loop */
2084 gmx_mm256_update_iforce_3atom_swizzle_pd(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,
2085 f+i_coord_offset,fshift+i_shift_offset);
2087 /* Increment number of inner iterations */
2088 inneriter += j_index_end - j_index_start;
2090 /* Outer loop uses 18 flops */
2093 /* Increment number of outer iterations */
2096 /* Update outer/inner flops */
2098 inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_W3W3_F,outeriter*18 + inneriter*270);