2 * This file is part of the GROMACS molecular simulation package.
4 * Copyright (c) 2012,2013,2014, by the GROMACS development team, led by
5 * Mark Abraham, David van der Spoel, Berk Hess, and Erik Lindahl,
6 * and including many others, as listed in the AUTHORS file in the
7 * top-level source directory and at http://www.gromacs.org.
9 * GROMACS is free software; you can redistribute it and/or
10 * modify it under the terms of the GNU Lesser General Public License
11 * as published by the Free Software Foundation; either version 2.1
12 * of the License, or (at your option) any later version.
14 * GROMACS is distributed in the hope that it will be useful,
15 * but WITHOUT ANY WARRANTY; without even the implied warranty of
16 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
17 * Lesser General Public License for more details.
19 * You should have received a copy of the GNU Lesser General Public
20 * License along with GROMACS; if not, see
21 * http://www.gnu.org/licenses, or write to the Free Software Foundation,
22 * Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA.
24 * If you want to redistribute modifications to GROMACS, please
25 * consider that scientific software is very special. Version
26 * control is crucial - bugs must be traceable. We will be happy to
27 * consider code for inclusion in the official distribution, but
28 * derived work must not be called official GROMACS. Details are found
29 * in the README & COPYING files - if they are missing, get the
30 * official version at http://www.gromacs.org.
32 * To help us fund GROMACS development, we humbly ask that you cite
33 * the research papers on the package. Check out http://www.gromacs.org.
36 * Note: this file was generated by the GROMACS avx_256_double kernel generator.
42 #include "../nb_kernel.h"
43 #include "gromacs/legacyheaders/types/simple.h"
44 #include "gromacs/math/vec.h"
45 #include "gromacs/legacyheaders/nrnb.h"
47 #include "gromacs/simd/math_x86_avx_256_double.h"
48 #include "kernelutil_x86_avx_256_double.h"
51 * Gromacs nonbonded kernel: nb_kernel_ElecRFCut_VdwNone_GeomW3W3_VF_avx_256_double
52 * Electrostatics interaction: ReactionField
53 * VdW interaction: None
54 * Geometry: Water3-Water3
55 * Calculate force/pot: PotentialAndForce
58 nb_kernel_ElecRFCut_VdwNone_GeomW3W3_VF_avx_256_double
59 (t_nblist * gmx_restrict nlist,
60 rvec * gmx_restrict xx,
61 rvec * gmx_restrict ff,
62 t_forcerec * gmx_restrict fr,
63 t_mdatoms * gmx_restrict mdatoms,
64 nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
65 t_nrnb * gmx_restrict nrnb)
67 /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
68 * just 0 for non-waters.
69 * Suffixes A,B,C,D refer to j loop unrolling done with AVX, e.g. for the four different
70 * jnr indices corresponding to data put in the four positions in the SIMD register.
72 int i_shift_offset,i_coord_offset,outeriter,inneriter;
73 int j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
74 int jnrA,jnrB,jnrC,jnrD;
75 int jnrlistA,jnrlistB,jnrlistC,jnrlistD;
76 int jnrlistE,jnrlistF,jnrlistG,jnrlistH;
77 int j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
78 int *iinr,*jindex,*jjnr,*shiftidx,*gid;
80 real *shiftvec,*fshift,*x,*f;
81 real *fjptrA,*fjptrB,*fjptrC,*fjptrD;
83 __m256d tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
84 real * vdwioffsetptr0;
85 __m256d ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
86 real * vdwioffsetptr1;
87 __m256d ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
88 real * vdwioffsetptr2;
89 __m256d ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
90 int vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
91 __m256d jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
92 int vdwjidx1A,vdwjidx1B,vdwjidx1C,vdwjidx1D;
93 __m256d jx1,jy1,jz1,fjx1,fjy1,fjz1,jq1,isaj1;
94 int vdwjidx2A,vdwjidx2B,vdwjidx2C,vdwjidx2D;
95 __m256d jx2,jy2,jz2,fjx2,fjy2,fjz2,jq2,isaj2;
96 __m256d dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
97 __m256d dx01,dy01,dz01,rsq01,rinv01,rinvsq01,r01,qq01,c6_01,c12_01;
98 __m256d dx02,dy02,dz02,rsq02,rinv02,rinvsq02,r02,qq02,c6_02,c12_02;
99 __m256d dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
100 __m256d dx11,dy11,dz11,rsq11,rinv11,rinvsq11,r11,qq11,c6_11,c12_11;
101 __m256d dx12,dy12,dz12,rsq12,rinv12,rinvsq12,r12,qq12,c6_12,c12_12;
102 __m256d dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
103 __m256d dx21,dy21,dz21,rsq21,rinv21,rinvsq21,r21,qq21,c6_21,c12_21;
104 __m256d dx22,dy22,dz22,rsq22,rinv22,rinvsq22,r22,qq22,c6_22,c12_22;
105 __m256d velec,felec,velecsum,facel,crf,krf,krf2;
107 __m256d dummy_mask,cutoff_mask;
108 __m128 tmpmask0,tmpmask1;
109 __m256d signbit = _mm256_castsi256_pd( _mm256_set1_epi32(0x80000000) );
110 __m256d one = _mm256_set1_pd(1.0);
111 __m256d two = _mm256_set1_pd(2.0);
117 jindex = nlist->jindex;
119 shiftidx = nlist->shift;
121 shiftvec = fr->shift_vec[0];
122 fshift = fr->fshift[0];
123 facel = _mm256_set1_pd(fr->epsfac);
124 charge = mdatoms->chargeA;
125 krf = _mm256_set1_pd(fr->ic->k_rf);
126 krf2 = _mm256_set1_pd(fr->ic->k_rf*2.0);
127 crf = _mm256_set1_pd(fr->ic->c_rf);
129 /* Setup water-specific parameters */
130 inr = nlist->iinr[0];
131 iq0 = _mm256_mul_pd(facel,_mm256_set1_pd(charge[inr+0]));
132 iq1 = _mm256_mul_pd(facel,_mm256_set1_pd(charge[inr+1]));
133 iq2 = _mm256_mul_pd(facel,_mm256_set1_pd(charge[inr+2]));
135 jq0 = _mm256_set1_pd(charge[inr+0]);
136 jq1 = _mm256_set1_pd(charge[inr+1]);
137 jq2 = _mm256_set1_pd(charge[inr+2]);
138 qq00 = _mm256_mul_pd(iq0,jq0);
139 qq01 = _mm256_mul_pd(iq0,jq1);
140 qq02 = _mm256_mul_pd(iq0,jq2);
141 qq10 = _mm256_mul_pd(iq1,jq0);
142 qq11 = _mm256_mul_pd(iq1,jq1);
143 qq12 = _mm256_mul_pd(iq1,jq2);
144 qq20 = _mm256_mul_pd(iq2,jq0);
145 qq21 = _mm256_mul_pd(iq2,jq1);
146 qq22 = _mm256_mul_pd(iq2,jq2);
148 /* When we use explicit cutoffs the value must be identical for elec and VdW, so use elec as an arbitrary choice */
149 rcutoff_scalar = fr->rcoulomb;
150 rcutoff = _mm256_set1_pd(rcutoff_scalar);
151 rcutoff2 = _mm256_mul_pd(rcutoff,rcutoff);
153 /* Avoid stupid compiler warnings */
154 jnrA = jnrB = jnrC = jnrD = 0;
163 for(iidx=0;iidx<4*DIM;iidx++)
168 /* Start outer loop over neighborlists */
169 for(iidx=0; iidx<nri; iidx++)
171 /* Load shift vector for this list */
172 i_shift_offset = DIM*shiftidx[iidx];
174 /* Load limits for loop over neighbors */
175 j_index_start = jindex[iidx];
176 j_index_end = jindex[iidx+1];
178 /* Get outer coordinate index */
180 i_coord_offset = DIM*inr;
182 /* Load i particle coords and add shift vector */
183 gmx_mm256_load_shift_and_3rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset,
184 &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2);
186 fix0 = _mm256_setzero_pd();
187 fiy0 = _mm256_setzero_pd();
188 fiz0 = _mm256_setzero_pd();
189 fix1 = _mm256_setzero_pd();
190 fiy1 = _mm256_setzero_pd();
191 fiz1 = _mm256_setzero_pd();
192 fix2 = _mm256_setzero_pd();
193 fiy2 = _mm256_setzero_pd();
194 fiz2 = _mm256_setzero_pd();
196 /* Reset potential sums */
197 velecsum = _mm256_setzero_pd();
199 /* Start inner kernel loop */
200 for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
203 /* Get j neighbor index, and coordinate index */
208 j_coord_offsetA = DIM*jnrA;
209 j_coord_offsetB = DIM*jnrB;
210 j_coord_offsetC = DIM*jnrC;
211 j_coord_offsetD = DIM*jnrD;
213 /* load j atom coordinates */
214 gmx_mm256_load_3rvec_4ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
215 x+j_coord_offsetC,x+j_coord_offsetD,
216 &jx0,&jy0,&jz0,&jx1,&jy1,&jz1,&jx2,&jy2,&jz2);
218 /* Calculate displacement vector */
219 dx00 = _mm256_sub_pd(ix0,jx0);
220 dy00 = _mm256_sub_pd(iy0,jy0);
221 dz00 = _mm256_sub_pd(iz0,jz0);
222 dx01 = _mm256_sub_pd(ix0,jx1);
223 dy01 = _mm256_sub_pd(iy0,jy1);
224 dz01 = _mm256_sub_pd(iz0,jz1);
225 dx02 = _mm256_sub_pd(ix0,jx2);
226 dy02 = _mm256_sub_pd(iy0,jy2);
227 dz02 = _mm256_sub_pd(iz0,jz2);
228 dx10 = _mm256_sub_pd(ix1,jx0);
229 dy10 = _mm256_sub_pd(iy1,jy0);
230 dz10 = _mm256_sub_pd(iz1,jz0);
231 dx11 = _mm256_sub_pd(ix1,jx1);
232 dy11 = _mm256_sub_pd(iy1,jy1);
233 dz11 = _mm256_sub_pd(iz1,jz1);
234 dx12 = _mm256_sub_pd(ix1,jx2);
235 dy12 = _mm256_sub_pd(iy1,jy2);
236 dz12 = _mm256_sub_pd(iz1,jz2);
237 dx20 = _mm256_sub_pd(ix2,jx0);
238 dy20 = _mm256_sub_pd(iy2,jy0);
239 dz20 = _mm256_sub_pd(iz2,jz0);
240 dx21 = _mm256_sub_pd(ix2,jx1);
241 dy21 = _mm256_sub_pd(iy2,jy1);
242 dz21 = _mm256_sub_pd(iz2,jz1);
243 dx22 = _mm256_sub_pd(ix2,jx2);
244 dy22 = _mm256_sub_pd(iy2,jy2);
245 dz22 = _mm256_sub_pd(iz2,jz2);
247 /* Calculate squared distance and things based on it */
248 rsq00 = gmx_mm256_calc_rsq_pd(dx00,dy00,dz00);
249 rsq01 = gmx_mm256_calc_rsq_pd(dx01,dy01,dz01);
250 rsq02 = gmx_mm256_calc_rsq_pd(dx02,dy02,dz02);
251 rsq10 = gmx_mm256_calc_rsq_pd(dx10,dy10,dz10);
252 rsq11 = gmx_mm256_calc_rsq_pd(dx11,dy11,dz11);
253 rsq12 = gmx_mm256_calc_rsq_pd(dx12,dy12,dz12);
254 rsq20 = gmx_mm256_calc_rsq_pd(dx20,dy20,dz20);
255 rsq21 = gmx_mm256_calc_rsq_pd(dx21,dy21,dz21);
256 rsq22 = gmx_mm256_calc_rsq_pd(dx22,dy22,dz22);
258 rinv00 = gmx_mm256_invsqrt_pd(rsq00);
259 rinv01 = gmx_mm256_invsqrt_pd(rsq01);
260 rinv02 = gmx_mm256_invsqrt_pd(rsq02);
261 rinv10 = gmx_mm256_invsqrt_pd(rsq10);
262 rinv11 = gmx_mm256_invsqrt_pd(rsq11);
263 rinv12 = gmx_mm256_invsqrt_pd(rsq12);
264 rinv20 = gmx_mm256_invsqrt_pd(rsq20);
265 rinv21 = gmx_mm256_invsqrt_pd(rsq21);
266 rinv22 = gmx_mm256_invsqrt_pd(rsq22);
268 rinvsq00 = _mm256_mul_pd(rinv00,rinv00);
269 rinvsq01 = _mm256_mul_pd(rinv01,rinv01);
270 rinvsq02 = _mm256_mul_pd(rinv02,rinv02);
271 rinvsq10 = _mm256_mul_pd(rinv10,rinv10);
272 rinvsq11 = _mm256_mul_pd(rinv11,rinv11);
273 rinvsq12 = _mm256_mul_pd(rinv12,rinv12);
274 rinvsq20 = _mm256_mul_pd(rinv20,rinv20);
275 rinvsq21 = _mm256_mul_pd(rinv21,rinv21);
276 rinvsq22 = _mm256_mul_pd(rinv22,rinv22);
278 fjx0 = _mm256_setzero_pd();
279 fjy0 = _mm256_setzero_pd();
280 fjz0 = _mm256_setzero_pd();
281 fjx1 = _mm256_setzero_pd();
282 fjy1 = _mm256_setzero_pd();
283 fjz1 = _mm256_setzero_pd();
284 fjx2 = _mm256_setzero_pd();
285 fjy2 = _mm256_setzero_pd();
286 fjz2 = _mm256_setzero_pd();
288 /**************************
289 * CALCULATE INTERACTIONS *
290 **************************/
292 if (gmx_mm256_any_lt(rsq00,rcutoff2))
295 /* REACTION-FIELD ELECTROSTATICS */
296 velec = _mm256_mul_pd(qq00,_mm256_sub_pd(_mm256_add_pd(rinv00,_mm256_mul_pd(krf,rsq00)),crf));
297 felec = _mm256_mul_pd(qq00,_mm256_sub_pd(_mm256_mul_pd(rinv00,rinvsq00),krf2));
299 cutoff_mask = _mm256_cmp_pd(rsq00,rcutoff2,_CMP_LT_OQ);
301 /* Update potential sum for this i atom from the interaction with this j atom. */
302 velec = _mm256_and_pd(velec,cutoff_mask);
303 velecsum = _mm256_add_pd(velecsum,velec);
307 fscal = _mm256_and_pd(fscal,cutoff_mask);
309 /* Calculate temporary vectorial force */
310 tx = _mm256_mul_pd(fscal,dx00);
311 ty = _mm256_mul_pd(fscal,dy00);
312 tz = _mm256_mul_pd(fscal,dz00);
314 /* Update vectorial force */
315 fix0 = _mm256_add_pd(fix0,tx);
316 fiy0 = _mm256_add_pd(fiy0,ty);
317 fiz0 = _mm256_add_pd(fiz0,tz);
319 fjx0 = _mm256_add_pd(fjx0,tx);
320 fjy0 = _mm256_add_pd(fjy0,ty);
321 fjz0 = _mm256_add_pd(fjz0,tz);
325 /**************************
326 * CALCULATE INTERACTIONS *
327 **************************/
329 if (gmx_mm256_any_lt(rsq01,rcutoff2))
332 /* REACTION-FIELD ELECTROSTATICS */
333 velec = _mm256_mul_pd(qq01,_mm256_sub_pd(_mm256_add_pd(rinv01,_mm256_mul_pd(krf,rsq01)),crf));
334 felec = _mm256_mul_pd(qq01,_mm256_sub_pd(_mm256_mul_pd(rinv01,rinvsq01),krf2));
336 cutoff_mask = _mm256_cmp_pd(rsq01,rcutoff2,_CMP_LT_OQ);
338 /* Update potential sum for this i atom from the interaction with this j atom. */
339 velec = _mm256_and_pd(velec,cutoff_mask);
340 velecsum = _mm256_add_pd(velecsum,velec);
344 fscal = _mm256_and_pd(fscal,cutoff_mask);
346 /* Calculate temporary vectorial force */
347 tx = _mm256_mul_pd(fscal,dx01);
348 ty = _mm256_mul_pd(fscal,dy01);
349 tz = _mm256_mul_pd(fscal,dz01);
351 /* Update vectorial force */
352 fix0 = _mm256_add_pd(fix0,tx);
353 fiy0 = _mm256_add_pd(fiy0,ty);
354 fiz0 = _mm256_add_pd(fiz0,tz);
356 fjx1 = _mm256_add_pd(fjx1,tx);
357 fjy1 = _mm256_add_pd(fjy1,ty);
358 fjz1 = _mm256_add_pd(fjz1,tz);
362 /**************************
363 * CALCULATE INTERACTIONS *
364 **************************/
366 if (gmx_mm256_any_lt(rsq02,rcutoff2))
369 /* REACTION-FIELD ELECTROSTATICS */
370 velec = _mm256_mul_pd(qq02,_mm256_sub_pd(_mm256_add_pd(rinv02,_mm256_mul_pd(krf,rsq02)),crf));
371 felec = _mm256_mul_pd(qq02,_mm256_sub_pd(_mm256_mul_pd(rinv02,rinvsq02),krf2));
373 cutoff_mask = _mm256_cmp_pd(rsq02,rcutoff2,_CMP_LT_OQ);
375 /* Update potential sum for this i atom from the interaction with this j atom. */
376 velec = _mm256_and_pd(velec,cutoff_mask);
377 velecsum = _mm256_add_pd(velecsum,velec);
381 fscal = _mm256_and_pd(fscal,cutoff_mask);
383 /* Calculate temporary vectorial force */
384 tx = _mm256_mul_pd(fscal,dx02);
385 ty = _mm256_mul_pd(fscal,dy02);
386 tz = _mm256_mul_pd(fscal,dz02);
388 /* Update vectorial force */
389 fix0 = _mm256_add_pd(fix0,tx);
390 fiy0 = _mm256_add_pd(fiy0,ty);
391 fiz0 = _mm256_add_pd(fiz0,tz);
393 fjx2 = _mm256_add_pd(fjx2,tx);
394 fjy2 = _mm256_add_pd(fjy2,ty);
395 fjz2 = _mm256_add_pd(fjz2,tz);
399 /**************************
400 * CALCULATE INTERACTIONS *
401 **************************/
403 if (gmx_mm256_any_lt(rsq10,rcutoff2))
406 /* REACTION-FIELD ELECTROSTATICS */
407 velec = _mm256_mul_pd(qq10,_mm256_sub_pd(_mm256_add_pd(rinv10,_mm256_mul_pd(krf,rsq10)),crf));
408 felec = _mm256_mul_pd(qq10,_mm256_sub_pd(_mm256_mul_pd(rinv10,rinvsq10),krf2));
410 cutoff_mask = _mm256_cmp_pd(rsq10,rcutoff2,_CMP_LT_OQ);
412 /* Update potential sum for this i atom from the interaction with this j atom. */
413 velec = _mm256_and_pd(velec,cutoff_mask);
414 velecsum = _mm256_add_pd(velecsum,velec);
418 fscal = _mm256_and_pd(fscal,cutoff_mask);
420 /* Calculate temporary vectorial force */
421 tx = _mm256_mul_pd(fscal,dx10);
422 ty = _mm256_mul_pd(fscal,dy10);
423 tz = _mm256_mul_pd(fscal,dz10);
425 /* Update vectorial force */
426 fix1 = _mm256_add_pd(fix1,tx);
427 fiy1 = _mm256_add_pd(fiy1,ty);
428 fiz1 = _mm256_add_pd(fiz1,tz);
430 fjx0 = _mm256_add_pd(fjx0,tx);
431 fjy0 = _mm256_add_pd(fjy0,ty);
432 fjz0 = _mm256_add_pd(fjz0,tz);
436 /**************************
437 * CALCULATE INTERACTIONS *
438 **************************/
440 if (gmx_mm256_any_lt(rsq11,rcutoff2))
443 /* REACTION-FIELD ELECTROSTATICS */
444 velec = _mm256_mul_pd(qq11,_mm256_sub_pd(_mm256_add_pd(rinv11,_mm256_mul_pd(krf,rsq11)),crf));
445 felec = _mm256_mul_pd(qq11,_mm256_sub_pd(_mm256_mul_pd(rinv11,rinvsq11),krf2));
447 cutoff_mask = _mm256_cmp_pd(rsq11,rcutoff2,_CMP_LT_OQ);
449 /* Update potential sum for this i atom from the interaction with this j atom. */
450 velec = _mm256_and_pd(velec,cutoff_mask);
451 velecsum = _mm256_add_pd(velecsum,velec);
455 fscal = _mm256_and_pd(fscal,cutoff_mask);
457 /* Calculate temporary vectorial force */
458 tx = _mm256_mul_pd(fscal,dx11);
459 ty = _mm256_mul_pd(fscal,dy11);
460 tz = _mm256_mul_pd(fscal,dz11);
462 /* Update vectorial force */
463 fix1 = _mm256_add_pd(fix1,tx);
464 fiy1 = _mm256_add_pd(fiy1,ty);
465 fiz1 = _mm256_add_pd(fiz1,tz);
467 fjx1 = _mm256_add_pd(fjx1,tx);
468 fjy1 = _mm256_add_pd(fjy1,ty);
469 fjz1 = _mm256_add_pd(fjz1,tz);
473 /**************************
474 * CALCULATE INTERACTIONS *
475 **************************/
477 if (gmx_mm256_any_lt(rsq12,rcutoff2))
480 /* REACTION-FIELD ELECTROSTATICS */
481 velec = _mm256_mul_pd(qq12,_mm256_sub_pd(_mm256_add_pd(rinv12,_mm256_mul_pd(krf,rsq12)),crf));
482 felec = _mm256_mul_pd(qq12,_mm256_sub_pd(_mm256_mul_pd(rinv12,rinvsq12),krf2));
484 cutoff_mask = _mm256_cmp_pd(rsq12,rcutoff2,_CMP_LT_OQ);
486 /* Update potential sum for this i atom from the interaction with this j atom. */
487 velec = _mm256_and_pd(velec,cutoff_mask);
488 velecsum = _mm256_add_pd(velecsum,velec);
492 fscal = _mm256_and_pd(fscal,cutoff_mask);
494 /* Calculate temporary vectorial force */
495 tx = _mm256_mul_pd(fscal,dx12);
496 ty = _mm256_mul_pd(fscal,dy12);
497 tz = _mm256_mul_pd(fscal,dz12);
499 /* Update vectorial force */
500 fix1 = _mm256_add_pd(fix1,tx);
501 fiy1 = _mm256_add_pd(fiy1,ty);
502 fiz1 = _mm256_add_pd(fiz1,tz);
504 fjx2 = _mm256_add_pd(fjx2,tx);
505 fjy2 = _mm256_add_pd(fjy2,ty);
506 fjz2 = _mm256_add_pd(fjz2,tz);
510 /**************************
511 * CALCULATE INTERACTIONS *
512 **************************/
514 if (gmx_mm256_any_lt(rsq20,rcutoff2))
517 /* REACTION-FIELD ELECTROSTATICS */
518 velec = _mm256_mul_pd(qq20,_mm256_sub_pd(_mm256_add_pd(rinv20,_mm256_mul_pd(krf,rsq20)),crf));
519 felec = _mm256_mul_pd(qq20,_mm256_sub_pd(_mm256_mul_pd(rinv20,rinvsq20),krf2));
521 cutoff_mask = _mm256_cmp_pd(rsq20,rcutoff2,_CMP_LT_OQ);
523 /* Update potential sum for this i atom from the interaction with this j atom. */
524 velec = _mm256_and_pd(velec,cutoff_mask);
525 velecsum = _mm256_add_pd(velecsum,velec);
529 fscal = _mm256_and_pd(fscal,cutoff_mask);
531 /* Calculate temporary vectorial force */
532 tx = _mm256_mul_pd(fscal,dx20);
533 ty = _mm256_mul_pd(fscal,dy20);
534 tz = _mm256_mul_pd(fscal,dz20);
536 /* Update vectorial force */
537 fix2 = _mm256_add_pd(fix2,tx);
538 fiy2 = _mm256_add_pd(fiy2,ty);
539 fiz2 = _mm256_add_pd(fiz2,tz);
541 fjx0 = _mm256_add_pd(fjx0,tx);
542 fjy0 = _mm256_add_pd(fjy0,ty);
543 fjz0 = _mm256_add_pd(fjz0,tz);
547 /**************************
548 * CALCULATE INTERACTIONS *
549 **************************/
551 if (gmx_mm256_any_lt(rsq21,rcutoff2))
554 /* REACTION-FIELD ELECTROSTATICS */
555 velec = _mm256_mul_pd(qq21,_mm256_sub_pd(_mm256_add_pd(rinv21,_mm256_mul_pd(krf,rsq21)),crf));
556 felec = _mm256_mul_pd(qq21,_mm256_sub_pd(_mm256_mul_pd(rinv21,rinvsq21),krf2));
558 cutoff_mask = _mm256_cmp_pd(rsq21,rcutoff2,_CMP_LT_OQ);
560 /* Update potential sum for this i atom from the interaction with this j atom. */
561 velec = _mm256_and_pd(velec,cutoff_mask);
562 velecsum = _mm256_add_pd(velecsum,velec);
566 fscal = _mm256_and_pd(fscal,cutoff_mask);
568 /* Calculate temporary vectorial force */
569 tx = _mm256_mul_pd(fscal,dx21);
570 ty = _mm256_mul_pd(fscal,dy21);
571 tz = _mm256_mul_pd(fscal,dz21);
573 /* Update vectorial force */
574 fix2 = _mm256_add_pd(fix2,tx);
575 fiy2 = _mm256_add_pd(fiy2,ty);
576 fiz2 = _mm256_add_pd(fiz2,tz);
578 fjx1 = _mm256_add_pd(fjx1,tx);
579 fjy1 = _mm256_add_pd(fjy1,ty);
580 fjz1 = _mm256_add_pd(fjz1,tz);
584 /**************************
585 * CALCULATE INTERACTIONS *
586 **************************/
588 if (gmx_mm256_any_lt(rsq22,rcutoff2))
591 /* REACTION-FIELD ELECTROSTATICS */
592 velec = _mm256_mul_pd(qq22,_mm256_sub_pd(_mm256_add_pd(rinv22,_mm256_mul_pd(krf,rsq22)),crf));
593 felec = _mm256_mul_pd(qq22,_mm256_sub_pd(_mm256_mul_pd(rinv22,rinvsq22),krf2));
595 cutoff_mask = _mm256_cmp_pd(rsq22,rcutoff2,_CMP_LT_OQ);
597 /* Update potential sum for this i atom from the interaction with this j atom. */
598 velec = _mm256_and_pd(velec,cutoff_mask);
599 velecsum = _mm256_add_pd(velecsum,velec);
603 fscal = _mm256_and_pd(fscal,cutoff_mask);
605 /* Calculate temporary vectorial force */
606 tx = _mm256_mul_pd(fscal,dx22);
607 ty = _mm256_mul_pd(fscal,dy22);
608 tz = _mm256_mul_pd(fscal,dz22);
610 /* Update vectorial force */
611 fix2 = _mm256_add_pd(fix2,tx);
612 fiy2 = _mm256_add_pd(fiy2,ty);
613 fiz2 = _mm256_add_pd(fiz2,tz);
615 fjx2 = _mm256_add_pd(fjx2,tx);
616 fjy2 = _mm256_add_pd(fjy2,ty);
617 fjz2 = _mm256_add_pd(fjz2,tz);
621 fjptrA = f+j_coord_offsetA;
622 fjptrB = f+j_coord_offsetB;
623 fjptrC = f+j_coord_offsetC;
624 fjptrD = f+j_coord_offsetD;
626 gmx_mm256_decrement_3rvec_4ptr_swizzle_pd(fjptrA,fjptrB,fjptrC,fjptrD,
627 fjx0,fjy0,fjz0,fjx1,fjy1,fjz1,fjx2,fjy2,fjz2);
629 /* Inner loop uses 324 flops */
635 /* Get j neighbor index, and coordinate index */
636 jnrlistA = jjnr[jidx];
637 jnrlistB = jjnr[jidx+1];
638 jnrlistC = jjnr[jidx+2];
639 jnrlistD = jjnr[jidx+3];
640 /* Sign of each element will be negative for non-real atoms.
641 * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
642 * so use it as val = _mm_andnot_pd(mask,val) to clear dummy entries.
644 tmpmask0 = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
646 tmpmask1 = _mm_permute_ps(tmpmask0,_GMX_MM_PERMUTE(3,3,2,2));
647 tmpmask0 = _mm_permute_ps(tmpmask0,_GMX_MM_PERMUTE(1,1,0,0));
648 dummy_mask = _mm256_castps_pd(gmx_mm256_set_m128(tmpmask1,tmpmask0));
650 jnrA = (jnrlistA>=0) ? jnrlistA : 0;
651 jnrB = (jnrlistB>=0) ? jnrlistB : 0;
652 jnrC = (jnrlistC>=0) ? jnrlistC : 0;
653 jnrD = (jnrlistD>=0) ? jnrlistD : 0;
654 j_coord_offsetA = DIM*jnrA;
655 j_coord_offsetB = DIM*jnrB;
656 j_coord_offsetC = DIM*jnrC;
657 j_coord_offsetD = DIM*jnrD;
659 /* load j atom coordinates */
660 gmx_mm256_load_3rvec_4ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
661 x+j_coord_offsetC,x+j_coord_offsetD,
662 &jx0,&jy0,&jz0,&jx1,&jy1,&jz1,&jx2,&jy2,&jz2);
664 /* Calculate displacement vector */
665 dx00 = _mm256_sub_pd(ix0,jx0);
666 dy00 = _mm256_sub_pd(iy0,jy0);
667 dz00 = _mm256_sub_pd(iz0,jz0);
668 dx01 = _mm256_sub_pd(ix0,jx1);
669 dy01 = _mm256_sub_pd(iy0,jy1);
670 dz01 = _mm256_sub_pd(iz0,jz1);
671 dx02 = _mm256_sub_pd(ix0,jx2);
672 dy02 = _mm256_sub_pd(iy0,jy2);
673 dz02 = _mm256_sub_pd(iz0,jz2);
674 dx10 = _mm256_sub_pd(ix1,jx0);
675 dy10 = _mm256_sub_pd(iy1,jy0);
676 dz10 = _mm256_sub_pd(iz1,jz0);
677 dx11 = _mm256_sub_pd(ix1,jx1);
678 dy11 = _mm256_sub_pd(iy1,jy1);
679 dz11 = _mm256_sub_pd(iz1,jz1);
680 dx12 = _mm256_sub_pd(ix1,jx2);
681 dy12 = _mm256_sub_pd(iy1,jy2);
682 dz12 = _mm256_sub_pd(iz1,jz2);
683 dx20 = _mm256_sub_pd(ix2,jx0);
684 dy20 = _mm256_sub_pd(iy2,jy0);
685 dz20 = _mm256_sub_pd(iz2,jz0);
686 dx21 = _mm256_sub_pd(ix2,jx1);
687 dy21 = _mm256_sub_pd(iy2,jy1);
688 dz21 = _mm256_sub_pd(iz2,jz1);
689 dx22 = _mm256_sub_pd(ix2,jx2);
690 dy22 = _mm256_sub_pd(iy2,jy2);
691 dz22 = _mm256_sub_pd(iz2,jz2);
693 /* Calculate squared distance and things based on it */
694 rsq00 = gmx_mm256_calc_rsq_pd(dx00,dy00,dz00);
695 rsq01 = gmx_mm256_calc_rsq_pd(dx01,dy01,dz01);
696 rsq02 = gmx_mm256_calc_rsq_pd(dx02,dy02,dz02);
697 rsq10 = gmx_mm256_calc_rsq_pd(dx10,dy10,dz10);
698 rsq11 = gmx_mm256_calc_rsq_pd(dx11,dy11,dz11);
699 rsq12 = gmx_mm256_calc_rsq_pd(dx12,dy12,dz12);
700 rsq20 = gmx_mm256_calc_rsq_pd(dx20,dy20,dz20);
701 rsq21 = gmx_mm256_calc_rsq_pd(dx21,dy21,dz21);
702 rsq22 = gmx_mm256_calc_rsq_pd(dx22,dy22,dz22);
704 rinv00 = gmx_mm256_invsqrt_pd(rsq00);
705 rinv01 = gmx_mm256_invsqrt_pd(rsq01);
706 rinv02 = gmx_mm256_invsqrt_pd(rsq02);
707 rinv10 = gmx_mm256_invsqrt_pd(rsq10);
708 rinv11 = gmx_mm256_invsqrt_pd(rsq11);
709 rinv12 = gmx_mm256_invsqrt_pd(rsq12);
710 rinv20 = gmx_mm256_invsqrt_pd(rsq20);
711 rinv21 = gmx_mm256_invsqrt_pd(rsq21);
712 rinv22 = gmx_mm256_invsqrt_pd(rsq22);
714 rinvsq00 = _mm256_mul_pd(rinv00,rinv00);
715 rinvsq01 = _mm256_mul_pd(rinv01,rinv01);
716 rinvsq02 = _mm256_mul_pd(rinv02,rinv02);
717 rinvsq10 = _mm256_mul_pd(rinv10,rinv10);
718 rinvsq11 = _mm256_mul_pd(rinv11,rinv11);
719 rinvsq12 = _mm256_mul_pd(rinv12,rinv12);
720 rinvsq20 = _mm256_mul_pd(rinv20,rinv20);
721 rinvsq21 = _mm256_mul_pd(rinv21,rinv21);
722 rinvsq22 = _mm256_mul_pd(rinv22,rinv22);
724 fjx0 = _mm256_setzero_pd();
725 fjy0 = _mm256_setzero_pd();
726 fjz0 = _mm256_setzero_pd();
727 fjx1 = _mm256_setzero_pd();
728 fjy1 = _mm256_setzero_pd();
729 fjz1 = _mm256_setzero_pd();
730 fjx2 = _mm256_setzero_pd();
731 fjy2 = _mm256_setzero_pd();
732 fjz2 = _mm256_setzero_pd();
734 /**************************
735 * CALCULATE INTERACTIONS *
736 **************************/
738 if (gmx_mm256_any_lt(rsq00,rcutoff2))
741 /* REACTION-FIELD ELECTROSTATICS */
742 velec = _mm256_mul_pd(qq00,_mm256_sub_pd(_mm256_add_pd(rinv00,_mm256_mul_pd(krf,rsq00)),crf));
743 felec = _mm256_mul_pd(qq00,_mm256_sub_pd(_mm256_mul_pd(rinv00,rinvsq00),krf2));
745 cutoff_mask = _mm256_cmp_pd(rsq00,rcutoff2,_CMP_LT_OQ);
747 /* Update potential sum for this i atom from the interaction with this j atom. */
748 velec = _mm256_and_pd(velec,cutoff_mask);
749 velec = _mm256_andnot_pd(dummy_mask,velec);
750 velecsum = _mm256_add_pd(velecsum,velec);
754 fscal = _mm256_and_pd(fscal,cutoff_mask);
756 fscal = _mm256_andnot_pd(dummy_mask,fscal);
758 /* Calculate temporary vectorial force */
759 tx = _mm256_mul_pd(fscal,dx00);
760 ty = _mm256_mul_pd(fscal,dy00);
761 tz = _mm256_mul_pd(fscal,dz00);
763 /* Update vectorial force */
764 fix0 = _mm256_add_pd(fix0,tx);
765 fiy0 = _mm256_add_pd(fiy0,ty);
766 fiz0 = _mm256_add_pd(fiz0,tz);
768 fjx0 = _mm256_add_pd(fjx0,tx);
769 fjy0 = _mm256_add_pd(fjy0,ty);
770 fjz0 = _mm256_add_pd(fjz0,tz);
774 /**************************
775 * CALCULATE INTERACTIONS *
776 **************************/
778 if (gmx_mm256_any_lt(rsq01,rcutoff2))
781 /* REACTION-FIELD ELECTROSTATICS */
782 velec = _mm256_mul_pd(qq01,_mm256_sub_pd(_mm256_add_pd(rinv01,_mm256_mul_pd(krf,rsq01)),crf));
783 felec = _mm256_mul_pd(qq01,_mm256_sub_pd(_mm256_mul_pd(rinv01,rinvsq01),krf2));
785 cutoff_mask = _mm256_cmp_pd(rsq01,rcutoff2,_CMP_LT_OQ);
787 /* Update potential sum for this i atom from the interaction with this j atom. */
788 velec = _mm256_and_pd(velec,cutoff_mask);
789 velec = _mm256_andnot_pd(dummy_mask,velec);
790 velecsum = _mm256_add_pd(velecsum,velec);
794 fscal = _mm256_and_pd(fscal,cutoff_mask);
796 fscal = _mm256_andnot_pd(dummy_mask,fscal);
798 /* Calculate temporary vectorial force */
799 tx = _mm256_mul_pd(fscal,dx01);
800 ty = _mm256_mul_pd(fscal,dy01);
801 tz = _mm256_mul_pd(fscal,dz01);
803 /* Update vectorial force */
804 fix0 = _mm256_add_pd(fix0,tx);
805 fiy0 = _mm256_add_pd(fiy0,ty);
806 fiz0 = _mm256_add_pd(fiz0,tz);
808 fjx1 = _mm256_add_pd(fjx1,tx);
809 fjy1 = _mm256_add_pd(fjy1,ty);
810 fjz1 = _mm256_add_pd(fjz1,tz);
814 /**************************
815 * CALCULATE INTERACTIONS *
816 **************************/
818 if (gmx_mm256_any_lt(rsq02,rcutoff2))
821 /* REACTION-FIELD ELECTROSTATICS */
822 velec = _mm256_mul_pd(qq02,_mm256_sub_pd(_mm256_add_pd(rinv02,_mm256_mul_pd(krf,rsq02)),crf));
823 felec = _mm256_mul_pd(qq02,_mm256_sub_pd(_mm256_mul_pd(rinv02,rinvsq02),krf2));
825 cutoff_mask = _mm256_cmp_pd(rsq02,rcutoff2,_CMP_LT_OQ);
827 /* Update potential sum for this i atom from the interaction with this j atom. */
828 velec = _mm256_and_pd(velec,cutoff_mask);
829 velec = _mm256_andnot_pd(dummy_mask,velec);
830 velecsum = _mm256_add_pd(velecsum,velec);
834 fscal = _mm256_and_pd(fscal,cutoff_mask);
836 fscal = _mm256_andnot_pd(dummy_mask,fscal);
838 /* Calculate temporary vectorial force */
839 tx = _mm256_mul_pd(fscal,dx02);
840 ty = _mm256_mul_pd(fscal,dy02);
841 tz = _mm256_mul_pd(fscal,dz02);
843 /* Update vectorial force */
844 fix0 = _mm256_add_pd(fix0,tx);
845 fiy0 = _mm256_add_pd(fiy0,ty);
846 fiz0 = _mm256_add_pd(fiz0,tz);
848 fjx2 = _mm256_add_pd(fjx2,tx);
849 fjy2 = _mm256_add_pd(fjy2,ty);
850 fjz2 = _mm256_add_pd(fjz2,tz);
854 /**************************
855 * CALCULATE INTERACTIONS *
856 **************************/
858 if (gmx_mm256_any_lt(rsq10,rcutoff2))
861 /* REACTION-FIELD ELECTROSTATICS */
862 velec = _mm256_mul_pd(qq10,_mm256_sub_pd(_mm256_add_pd(rinv10,_mm256_mul_pd(krf,rsq10)),crf));
863 felec = _mm256_mul_pd(qq10,_mm256_sub_pd(_mm256_mul_pd(rinv10,rinvsq10),krf2));
865 cutoff_mask = _mm256_cmp_pd(rsq10,rcutoff2,_CMP_LT_OQ);
867 /* Update potential sum for this i atom from the interaction with this j atom. */
868 velec = _mm256_and_pd(velec,cutoff_mask);
869 velec = _mm256_andnot_pd(dummy_mask,velec);
870 velecsum = _mm256_add_pd(velecsum,velec);
874 fscal = _mm256_and_pd(fscal,cutoff_mask);
876 fscal = _mm256_andnot_pd(dummy_mask,fscal);
878 /* Calculate temporary vectorial force */
879 tx = _mm256_mul_pd(fscal,dx10);
880 ty = _mm256_mul_pd(fscal,dy10);
881 tz = _mm256_mul_pd(fscal,dz10);
883 /* Update vectorial force */
884 fix1 = _mm256_add_pd(fix1,tx);
885 fiy1 = _mm256_add_pd(fiy1,ty);
886 fiz1 = _mm256_add_pd(fiz1,tz);
888 fjx0 = _mm256_add_pd(fjx0,tx);
889 fjy0 = _mm256_add_pd(fjy0,ty);
890 fjz0 = _mm256_add_pd(fjz0,tz);
894 /**************************
895 * CALCULATE INTERACTIONS *
896 **************************/
898 if (gmx_mm256_any_lt(rsq11,rcutoff2))
901 /* REACTION-FIELD ELECTROSTATICS */
902 velec = _mm256_mul_pd(qq11,_mm256_sub_pd(_mm256_add_pd(rinv11,_mm256_mul_pd(krf,rsq11)),crf));
903 felec = _mm256_mul_pd(qq11,_mm256_sub_pd(_mm256_mul_pd(rinv11,rinvsq11),krf2));
905 cutoff_mask = _mm256_cmp_pd(rsq11,rcutoff2,_CMP_LT_OQ);
907 /* Update potential sum for this i atom from the interaction with this j atom. */
908 velec = _mm256_and_pd(velec,cutoff_mask);
909 velec = _mm256_andnot_pd(dummy_mask,velec);
910 velecsum = _mm256_add_pd(velecsum,velec);
914 fscal = _mm256_and_pd(fscal,cutoff_mask);
916 fscal = _mm256_andnot_pd(dummy_mask,fscal);
918 /* Calculate temporary vectorial force */
919 tx = _mm256_mul_pd(fscal,dx11);
920 ty = _mm256_mul_pd(fscal,dy11);
921 tz = _mm256_mul_pd(fscal,dz11);
923 /* Update vectorial force */
924 fix1 = _mm256_add_pd(fix1,tx);
925 fiy1 = _mm256_add_pd(fiy1,ty);
926 fiz1 = _mm256_add_pd(fiz1,tz);
928 fjx1 = _mm256_add_pd(fjx1,tx);
929 fjy1 = _mm256_add_pd(fjy1,ty);
930 fjz1 = _mm256_add_pd(fjz1,tz);
934 /**************************
935 * CALCULATE INTERACTIONS *
936 **************************/
938 if (gmx_mm256_any_lt(rsq12,rcutoff2))
941 /* REACTION-FIELD ELECTROSTATICS */
942 velec = _mm256_mul_pd(qq12,_mm256_sub_pd(_mm256_add_pd(rinv12,_mm256_mul_pd(krf,rsq12)),crf));
943 felec = _mm256_mul_pd(qq12,_mm256_sub_pd(_mm256_mul_pd(rinv12,rinvsq12),krf2));
945 cutoff_mask = _mm256_cmp_pd(rsq12,rcutoff2,_CMP_LT_OQ);
947 /* Update potential sum for this i atom from the interaction with this j atom. */
948 velec = _mm256_and_pd(velec,cutoff_mask);
949 velec = _mm256_andnot_pd(dummy_mask,velec);
950 velecsum = _mm256_add_pd(velecsum,velec);
954 fscal = _mm256_and_pd(fscal,cutoff_mask);
956 fscal = _mm256_andnot_pd(dummy_mask,fscal);
958 /* Calculate temporary vectorial force */
959 tx = _mm256_mul_pd(fscal,dx12);
960 ty = _mm256_mul_pd(fscal,dy12);
961 tz = _mm256_mul_pd(fscal,dz12);
963 /* Update vectorial force */
964 fix1 = _mm256_add_pd(fix1,tx);
965 fiy1 = _mm256_add_pd(fiy1,ty);
966 fiz1 = _mm256_add_pd(fiz1,tz);
968 fjx2 = _mm256_add_pd(fjx2,tx);
969 fjy2 = _mm256_add_pd(fjy2,ty);
970 fjz2 = _mm256_add_pd(fjz2,tz);
974 /**************************
975 * CALCULATE INTERACTIONS *
976 **************************/
978 if (gmx_mm256_any_lt(rsq20,rcutoff2))
981 /* REACTION-FIELD ELECTROSTATICS */
982 velec = _mm256_mul_pd(qq20,_mm256_sub_pd(_mm256_add_pd(rinv20,_mm256_mul_pd(krf,rsq20)),crf));
983 felec = _mm256_mul_pd(qq20,_mm256_sub_pd(_mm256_mul_pd(rinv20,rinvsq20),krf2));
985 cutoff_mask = _mm256_cmp_pd(rsq20,rcutoff2,_CMP_LT_OQ);
987 /* Update potential sum for this i atom from the interaction with this j atom. */
988 velec = _mm256_and_pd(velec,cutoff_mask);
989 velec = _mm256_andnot_pd(dummy_mask,velec);
990 velecsum = _mm256_add_pd(velecsum,velec);
994 fscal = _mm256_and_pd(fscal,cutoff_mask);
996 fscal = _mm256_andnot_pd(dummy_mask,fscal);
998 /* Calculate temporary vectorial force */
999 tx = _mm256_mul_pd(fscal,dx20);
1000 ty = _mm256_mul_pd(fscal,dy20);
1001 tz = _mm256_mul_pd(fscal,dz20);
1003 /* Update vectorial force */
1004 fix2 = _mm256_add_pd(fix2,tx);
1005 fiy2 = _mm256_add_pd(fiy2,ty);
1006 fiz2 = _mm256_add_pd(fiz2,tz);
1008 fjx0 = _mm256_add_pd(fjx0,tx);
1009 fjy0 = _mm256_add_pd(fjy0,ty);
1010 fjz0 = _mm256_add_pd(fjz0,tz);
1014 /**************************
1015 * CALCULATE INTERACTIONS *
1016 **************************/
1018 if (gmx_mm256_any_lt(rsq21,rcutoff2))
1021 /* REACTION-FIELD ELECTROSTATICS */
1022 velec = _mm256_mul_pd(qq21,_mm256_sub_pd(_mm256_add_pd(rinv21,_mm256_mul_pd(krf,rsq21)),crf));
1023 felec = _mm256_mul_pd(qq21,_mm256_sub_pd(_mm256_mul_pd(rinv21,rinvsq21),krf2));
1025 cutoff_mask = _mm256_cmp_pd(rsq21,rcutoff2,_CMP_LT_OQ);
1027 /* Update potential sum for this i atom from the interaction with this j atom. */
1028 velec = _mm256_and_pd(velec,cutoff_mask);
1029 velec = _mm256_andnot_pd(dummy_mask,velec);
1030 velecsum = _mm256_add_pd(velecsum,velec);
1034 fscal = _mm256_and_pd(fscal,cutoff_mask);
1036 fscal = _mm256_andnot_pd(dummy_mask,fscal);
1038 /* Calculate temporary vectorial force */
1039 tx = _mm256_mul_pd(fscal,dx21);
1040 ty = _mm256_mul_pd(fscal,dy21);
1041 tz = _mm256_mul_pd(fscal,dz21);
1043 /* Update vectorial force */
1044 fix2 = _mm256_add_pd(fix2,tx);
1045 fiy2 = _mm256_add_pd(fiy2,ty);
1046 fiz2 = _mm256_add_pd(fiz2,tz);
1048 fjx1 = _mm256_add_pd(fjx1,tx);
1049 fjy1 = _mm256_add_pd(fjy1,ty);
1050 fjz1 = _mm256_add_pd(fjz1,tz);
1054 /**************************
1055 * CALCULATE INTERACTIONS *
1056 **************************/
1058 if (gmx_mm256_any_lt(rsq22,rcutoff2))
1061 /* REACTION-FIELD ELECTROSTATICS */
1062 velec = _mm256_mul_pd(qq22,_mm256_sub_pd(_mm256_add_pd(rinv22,_mm256_mul_pd(krf,rsq22)),crf));
1063 felec = _mm256_mul_pd(qq22,_mm256_sub_pd(_mm256_mul_pd(rinv22,rinvsq22),krf2));
1065 cutoff_mask = _mm256_cmp_pd(rsq22,rcutoff2,_CMP_LT_OQ);
1067 /* Update potential sum for this i atom from the interaction with this j atom. */
1068 velec = _mm256_and_pd(velec,cutoff_mask);
1069 velec = _mm256_andnot_pd(dummy_mask,velec);
1070 velecsum = _mm256_add_pd(velecsum,velec);
1074 fscal = _mm256_and_pd(fscal,cutoff_mask);
1076 fscal = _mm256_andnot_pd(dummy_mask,fscal);
1078 /* Calculate temporary vectorial force */
1079 tx = _mm256_mul_pd(fscal,dx22);
1080 ty = _mm256_mul_pd(fscal,dy22);
1081 tz = _mm256_mul_pd(fscal,dz22);
1083 /* Update vectorial force */
1084 fix2 = _mm256_add_pd(fix2,tx);
1085 fiy2 = _mm256_add_pd(fiy2,ty);
1086 fiz2 = _mm256_add_pd(fiz2,tz);
1088 fjx2 = _mm256_add_pd(fjx2,tx);
1089 fjy2 = _mm256_add_pd(fjy2,ty);
1090 fjz2 = _mm256_add_pd(fjz2,tz);
1094 fjptrA = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
1095 fjptrB = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
1096 fjptrC = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
1097 fjptrD = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
1099 gmx_mm256_decrement_3rvec_4ptr_swizzle_pd(fjptrA,fjptrB,fjptrC,fjptrD,
1100 fjx0,fjy0,fjz0,fjx1,fjy1,fjz1,fjx2,fjy2,fjz2);
1102 /* Inner loop uses 324 flops */
1105 /* End of innermost loop */
1107 gmx_mm256_update_iforce_3atom_swizzle_pd(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,
1108 f+i_coord_offset,fshift+i_shift_offset);
1111 /* Update potential energies */
1112 gmx_mm256_update_1pot_pd(velecsum,kernel_data->energygrp_elec+ggid);
1114 /* Increment number of inner iterations */
1115 inneriter += j_index_end - j_index_start;
1117 /* Outer loop uses 19 flops */
1120 /* Increment number of outer iterations */
1123 /* Update outer/inner flops */
1125 inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_W3W3_VF,outeriter*19 + inneriter*324);
1128 * Gromacs nonbonded kernel: nb_kernel_ElecRFCut_VdwNone_GeomW3W3_F_avx_256_double
1129 * Electrostatics interaction: ReactionField
1130 * VdW interaction: None
1131 * Geometry: Water3-Water3
1132 * Calculate force/pot: Force
1135 nb_kernel_ElecRFCut_VdwNone_GeomW3W3_F_avx_256_double
1136 (t_nblist * gmx_restrict nlist,
1137 rvec * gmx_restrict xx,
1138 rvec * gmx_restrict ff,
1139 t_forcerec * gmx_restrict fr,
1140 t_mdatoms * gmx_restrict mdatoms,
1141 nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
1142 t_nrnb * gmx_restrict nrnb)
1144 /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
1145 * just 0 for non-waters.
1146 * Suffixes A,B,C,D refer to j loop unrolling done with AVX, e.g. for the four different
1147 * jnr indices corresponding to data put in the four positions in the SIMD register.
1149 int i_shift_offset,i_coord_offset,outeriter,inneriter;
1150 int j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
1151 int jnrA,jnrB,jnrC,jnrD;
1152 int jnrlistA,jnrlistB,jnrlistC,jnrlistD;
1153 int jnrlistE,jnrlistF,jnrlistG,jnrlistH;
1154 int j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
1155 int *iinr,*jindex,*jjnr,*shiftidx,*gid;
1156 real rcutoff_scalar;
1157 real *shiftvec,*fshift,*x,*f;
1158 real *fjptrA,*fjptrB,*fjptrC,*fjptrD;
1159 real scratch[4*DIM];
1160 __m256d tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
1161 real * vdwioffsetptr0;
1162 __m256d ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
1163 real * vdwioffsetptr1;
1164 __m256d ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
1165 real * vdwioffsetptr2;
1166 __m256d ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
1167 int vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
1168 __m256d jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
1169 int vdwjidx1A,vdwjidx1B,vdwjidx1C,vdwjidx1D;
1170 __m256d jx1,jy1,jz1,fjx1,fjy1,fjz1,jq1,isaj1;
1171 int vdwjidx2A,vdwjidx2B,vdwjidx2C,vdwjidx2D;
1172 __m256d jx2,jy2,jz2,fjx2,fjy2,fjz2,jq2,isaj2;
1173 __m256d dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
1174 __m256d dx01,dy01,dz01,rsq01,rinv01,rinvsq01,r01,qq01,c6_01,c12_01;
1175 __m256d dx02,dy02,dz02,rsq02,rinv02,rinvsq02,r02,qq02,c6_02,c12_02;
1176 __m256d dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
1177 __m256d dx11,dy11,dz11,rsq11,rinv11,rinvsq11,r11,qq11,c6_11,c12_11;
1178 __m256d dx12,dy12,dz12,rsq12,rinv12,rinvsq12,r12,qq12,c6_12,c12_12;
1179 __m256d dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
1180 __m256d dx21,dy21,dz21,rsq21,rinv21,rinvsq21,r21,qq21,c6_21,c12_21;
1181 __m256d dx22,dy22,dz22,rsq22,rinv22,rinvsq22,r22,qq22,c6_22,c12_22;
1182 __m256d velec,felec,velecsum,facel,crf,krf,krf2;
1184 __m256d dummy_mask,cutoff_mask;
1185 __m128 tmpmask0,tmpmask1;
1186 __m256d signbit = _mm256_castsi256_pd( _mm256_set1_epi32(0x80000000) );
1187 __m256d one = _mm256_set1_pd(1.0);
1188 __m256d two = _mm256_set1_pd(2.0);
1194 jindex = nlist->jindex;
1196 shiftidx = nlist->shift;
1198 shiftvec = fr->shift_vec[0];
1199 fshift = fr->fshift[0];
1200 facel = _mm256_set1_pd(fr->epsfac);
1201 charge = mdatoms->chargeA;
1202 krf = _mm256_set1_pd(fr->ic->k_rf);
1203 krf2 = _mm256_set1_pd(fr->ic->k_rf*2.0);
1204 crf = _mm256_set1_pd(fr->ic->c_rf);
1206 /* Setup water-specific parameters */
1207 inr = nlist->iinr[0];
1208 iq0 = _mm256_mul_pd(facel,_mm256_set1_pd(charge[inr+0]));
1209 iq1 = _mm256_mul_pd(facel,_mm256_set1_pd(charge[inr+1]));
1210 iq2 = _mm256_mul_pd(facel,_mm256_set1_pd(charge[inr+2]));
1212 jq0 = _mm256_set1_pd(charge[inr+0]);
1213 jq1 = _mm256_set1_pd(charge[inr+1]);
1214 jq2 = _mm256_set1_pd(charge[inr+2]);
1215 qq00 = _mm256_mul_pd(iq0,jq0);
1216 qq01 = _mm256_mul_pd(iq0,jq1);
1217 qq02 = _mm256_mul_pd(iq0,jq2);
1218 qq10 = _mm256_mul_pd(iq1,jq0);
1219 qq11 = _mm256_mul_pd(iq1,jq1);
1220 qq12 = _mm256_mul_pd(iq1,jq2);
1221 qq20 = _mm256_mul_pd(iq2,jq0);
1222 qq21 = _mm256_mul_pd(iq2,jq1);
1223 qq22 = _mm256_mul_pd(iq2,jq2);
1225 /* When we use explicit cutoffs the value must be identical for elec and VdW, so use elec as an arbitrary choice */
1226 rcutoff_scalar = fr->rcoulomb;
1227 rcutoff = _mm256_set1_pd(rcutoff_scalar);
1228 rcutoff2 = _mm256_mul_pd(rcutoff,rcutoff);
1230 /* Avoid stupid compiler warnings */
1231 jnrA = jnrB = jnrC = jnrD = 0;
1232 j_coord_offsetA = 0;
1233 j_coord_offsetB = 0;
1234 j_coord_offsetC = 0;
1235 j_coord_offsetD = 0;
1240 for(iidx=0;iidx<4*DIM;iidx++)
1242 scratch[iidx] = 0.0;
1245 /* Start outer loop over neighborlists */
1246 for(iidx=0; iidx<nri; iidx++)
1248 /* Load shift vector for this list */
1249 i_shift_offset = DIM*shiftidx[iidx];
1251 /* Load limits for loop over neighbors */
1252 j_index_start = jindex[iidx];
1253 j_index_end = jindex[iidx+1];
1255 /* Get outer coordinate index */
1257 i_coord_offset = DIM*inr;
1259 /* Load i particle coords and add shift vector */
1260 gmx_mm256_load_shift_and_3rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset,
1261 &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2);
1263 fix0 = _mm256_setzero_pd();
1264 fiy0 = _mm256_setzero_pd();
1265 fiz0 = _mm256_setzero_pd();
1266 fix1 = _mm256_setzero_pd();
1267 fiy1 = _mm256_setzero_pd();
1268 fiz1 = _mm256_setzero_pd();
1269 fix2 = _mm256_setzero_pd();
1270 fiy2 = _mm256_setzero_pd();
1271 fiz2 = _mm256_setzero_pd();
1273 /* Start inner kernel loop */
1274 for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
1277 /* Get j neighbor index, and coordinate index */
1279 jnrB = jjnr[jidx+1];
1280 jnrC = jjnr[jidx+2];
1281 jnrD = jjnr[jidx+3];
1282 j_coord_offsetA = DIM*jnrA;
1283 j_coord_offsetB = DIM*jnrB;
1284 j_coord_offsetC = DIM*jnrC;
1285 j_coord_offsetD = DIM*jnrD;
1287 /* load j atom coordinates */
1288 gmx_mm256_load_3rvec_4ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
1289 x+j_coord_offsetC,x+j_coord_offsetD,
1290 &jx0,&jy0,&jz0,&jx1,&jy1,&jz1,&jx2,&jy2,&jz2);
1292 /* Calculate displacement vector */
1293 dx00 = _mm256_sub_pd(ix0,jx0);
1294 dy00 = _mm256_sub_pd(iy0,jy0);
1295 dz00 = _mm256_sub_pd(iz0,jz0);
1296 dx01 = _mm256_sub_pd(ix0,jx1);
1297 dy01 = _mm256_sub_pd(iy0,jy1);
1298 dz01 = _mm256_sub_pd(iz0,jz1);
1299 dx02 = _mm256_sub_pd(ix0,jx2);
1300 dy02 = _mm256_sub_pd(iy0,jy2);
1301 dz02 = _mm256_sub_pd(iz0,jz2);
1302 dx10 = _mm256_sub_pd(ix1,jx0);
1303 dy10 = _mm256_sub_pd(iy1,jy0);
1304 dz10 = _mm256_sub_pd(iz1,jz0);
1305 dx11 = _mm256_sub_pd(ix1,jx1);
1306 dy11 = _mm256_sub_pd(iy1,jy1);
1307 dz11 = _mm256_sub_pd(iz1,jz1);
1308 dx12 = _mm256_sub_pd(ix1,jx2);
1309 dy12 = _mm256_sub_pd(iy1,jy2);
1310 dz12 = _mm256_sub_pd(iz1,jz2);
1311 dx20 = _mm256_sub_pd(ix2,jx0);
1312 dy20 = _mm256_sub_pd(iy2,jy0);
1313 dz20 = _mm256_sub_pd(iz2,jz0);
1314 dx21 = _mm256_sub_pd(ix2,jx1);
1315 dy21 = _mm256_sub_pd(iy2,jy1);
1316 dz21 = _mm256_sub_pd(iz2,jz1);
1317 dx22 = _mm256_sub_pd(ix2,jx2);
1318 dy22 = _mm256_sub_pd(iy2,jy2);
1319 dz22 = _mm256_sub_pd(iz2,jz2);
1321 /* Calculate squared distance and things based on it */
1322 rsq00 = gmx_mm256_calc_rsq_pd(dx00,dy00,dz00);
1323 rsq01 = gmx_mm256_calc_rsq_pd(dx01,dy01,dz01);
1324 rsq02 = gmx_mm256_calc_rsq_pd(dx02,dy02,dz02);
1325 rsq10 = gmx_mm256_calc_rsq_pd(dx10,dy10,dz10);
1326 rsq11 = gmx_mm256_calc_rsq_pd(dx11,dy11,dz11);
1327 rsq12 = gmx_mm256_calc_rsq_pd(dx12,dy12,dz12);
1328 rsq20 = gmx_mm256_calc_rsq_pd(dx20,dy20,dz20);
1329 rsq21 = gmx_mm256_calc_rsq_pd(dx21,dy21,dz21);
1330 rsq22 = gmx_mm256_calc_rsq_pd(dx22,dy22,dz22);
1332 rinv00 = gmx_mm256_invsqrt_pd(rsq00);
1333 rinv01 = gmx_mm256_invsqrt_pd(rsq01);
1334 rinv02 = gmx_mm256_invsqrt_pd(rsq02);
1335 rinv10 = gmx_mm256_invsqrt_pd(rsq10);
1336 rinv11 = gmx_mm256_invsqrt_pd(rsq11);
1337 rinv12 = gmx_mm256_invsqrt_pd(rsq12);
1338 rinv20 = gmx_mm256_invsqrt_pd(rsq20);
1339 rinv21 = gmx_mm256_invsqrt_pd(rsq21);
1340 rinv22 = gmx_mm256_invsqrt_pd(rsq22);
1342 rinvsq00 = _mm256_mul_pd(rinv00,rinv00);
1343 rinvsq01 = _mm256_mul_pd(rinv01,rinv01);
1344 rinvsq02 = _mm256_mul_pd(rinv02,rinv02);
1345 rinvsq10 = _mm256_mul_pd(rinv10,rinv10);
1346 rinvsq11 = _mm256_mul_pd(rinv11,rinv11);
1347 rinvsq12 = _mm256_mul_pd(rinv12,rinv12);
1348 rinvsq20 = _mm256_mul_pd(rinv20,rinv20);
1349 rinvsq21 = _mm256_mul_pd(rinv21,rinv21);
1350 rinvsq22 = _mm256_mul_pd(rinv22,rinv22);
1352 fjx0 = _mm256_setzero_pd();
1353 fjy0 = _mm256_setzero_pd();
1354 fjz0 = _mm256_setzero_pd();
1355 fjx1 = _mm256_setzero_pd();
1356 fjy1 = _mm256_setzero_pd();
1357 fjz1 = _mm256_setzero_pd();
1358 fjx2 = _mm256_setzero_pd();
1359 fjy2 = _mm256_setzero_pd();
1360 fjz2 = _mm256_setzero_pd();
1362 /**************************
1363 * CALCULATE INTERACTIONS *
1364 **************************/
1366 if (gmx_mm256_any_lt(rsq00,rcutoff2))
1369 /* REACTION-FIELD ELECTROSTATICS */
1370 felec = _mm256_mul_pd(qq00,_mm256_sub_pd(_mm256_mul_pd(rinv00,rinvsq00),krf2));
1372 cutoff_mask = _mm256_cmp_pd(rsq00,rcutoff2,_CMP_LT_OQ);
1376 fscal = _mm256_and_pd(fscal,cutoff_mask);
1378 /* Calculate temporary vectorial force */
1379 tx = _mm256_mul_pd(fscal,dx00);
1380 ty = _mm256_mul_pd(fscal,dy00);
1381 tz = _mm256_mul_pd(fscal,dz00);
1383 /* Update vectorial force */
1384 fix0 = _mm256_add_pd(fix0,tx);
1385 fiy0 = _mm256_add_pd(fiy0,ty);
1386 fiz0 = _mm256_add_pd(fiz0,tz);
1388 fjx0 = _mm256_add_pd(fjx0,tx);
1389 fjy0 = _mm256_add_pd(fjy0,ty);
1390 fjz0 = _mm256_add_pd(fjz0,tz);
1394 /**************************
1395 * CALCULATE INTERACTIONS *
1396 **************************/
1398 if (gmx_mm256_any_lt(rsq01,rcutoff2))
1401 /* REACTION-FIELD ELECTROSTATICS */
1402 felec = _mm256_mul_pd(qq01,_mm256_sub_pd(_mm256_mul_pd(rinv01,rinvsq01),krf2));
1404 cutoff_mask = _mm256_cmp_pd(rsq01,rcutoff2,_CMP_LT_OQ);
1408 fscal = _mm256_and_pd(fscal,cutoff_mask);
1410 /* Calculate temporary vectorial force */
1411 tx = _mm256_mul_pd(fscal,dx01);
1412 ty = _mm256_mul_pd(fscal,dy01);
1413 tz = _mm256_mul_pd(fscal,dz01);
1415 /* Update vectorial force */
1416 fix0 = _mm256_add_pd(fix0,tx);
1417 fiy0 = _mm256_add_pd(fiy0,ty);
1418 fiz0 = _mm256_add_pd(fiz0,tz);
1420 fjx1 = _mm256_add_pd(fjx1,tx);
1421 fjy1 = _mm256_add_pd(fjy1,ty);
1422 fjz1 = _mm256_add_pd(fjz1,tz);
1426 /**************************
1427 * CALCULATE INTERACTIONS *
1428 **************************/
1430 if (gmx_mm256_any_lt(rsq02,rcutoff2))
1433 /* REACTION-FIELD ELECTROSTATICS */
1434 felec = _mm256_mul_pd(qq02,_mm256_sub_pd(_mm256_mul_pd(rinv02,rinvsq02),krf2));
1436 cutoff_mask = _mm256_cmp_pd(rsq02,rcutoff2,_CMP_LT_OQ);
1440 fscal = _mm256_and_pd(fscal,cutoff_mask);
1442 /* Calculate temporary vectorial force */
1443 tx = _mm256_mul_pd(fscal,dx02);
1444 ty = _mm256_mul_pd(fscal,dy02);
1445 tz = _mm256_mul_pd(fscal,dz02);
1447 /* Update vectorial force */
1448 fix0 = _mm256_add_pd(fix0,tx);
1449 fiy0 = _mm256_add_pd(fiy0,ty);
1450 fiz0 = _mm256_add_pd(fiz0,tz);
1452 fjx2 = _mm256_add_pd(fjx2,tx);
1453 fjy2 = _mm256_add_pd(fjy2,ty);
1454 fjz2 = _mm256_add_pd(fjz2,tz);
1458 /**************************
1459 * CALCULATE INTERACTIONS *
1460 **************************/
1462 if (gmx_mm256_any_lt(rsq10,rcutoff2))
1465 /* REACTION-FIELD ELECTROSTATICS */
1466 felec = _mm256_mul_pd(qq10,_mm256_sub_pd(_mm256_mul_pd(rinv10,rinvsq10),krf2));
1468 cutoff_mask = _mm256_cmp_pd(rsq10,rcutoff2,_CMP_LT_OQ);
1472 fscal = _mm256_and_pd(fscal,cutoff_mask);
1474 /* Calculate temporary vectorial force */
1475 tx = _mm256_mul_pd(fscal,dx10);
1476 ty = _mm256_mul_pd(fscal,dy10);
1477 tz = _mm256_mul_pd(fscal,dz10);
1479 /* Update vectorial force */
1480 fix1 = _mm256_add_pd(fix1,tx);
1481 fiy1 = _mm256_add_pd(fiy1,ty);
1482 fiz1 = _mm256_add_pd(fiz1,tz);
1484 fjx0 = _mm256_add_pd(fjx0,tx);
1485 fjy0 = _mm256_add_pd(fjy0,ty);
1486 fjz0 = _mm256_add_pd(fjz0,tz);
1490 /**************************
1491 * CALCULATE INTERACTIONS *
1492 **************************/
1494 if (gmx_mm256_any_lt(rsq11,rcutoff2))
1497 /* REACTION-FIELD ELECTROSTATICS */
1498 felec = _mm256_mul_pd(qq11,_mm256_sub_pd(_mm256_mul_pd(rinv11,rinvsq11),krf2));
1500 cutoff_mask = _mm256_cmp_pd(rsq11,rcutoff2,_CMP_LT_OQ);
1504 fscal = _mm256_and_pd(fscal,cutoff_mask);
1506 /* Calculate temporary vectorial force */
1507 tx = _mm256_mul_pd(fscal,dx11);
1508 ty = _mm256_mul_pd(fscal,dy11);
1509 tz = _mm256_mul_pd(fscal,dz11);
1511 /* Update vectorial force */
1512 fix1 = _mm256_add_pd(fix1,tx);
1513 fiy1 = _mm256_add_pd(fiy1,ty);
1514 fiz1 = _mm256_add_pd(fiz1,tz);
1516 fjx1 = _mm256_add_pd(fjx1,tx);
1517 fjy1 = _mm256_add_pd(fjy1,ty);
1518 fjz1 = _mm256_add_pd(fjz1,tz);
1522 /**************************
1523 * CALCULATE INTERACTIONS *
1524 **************************/
1526 if (gmx_mm256_any_lt(rsq12,rcutoff2))
1529 /* REACTION-FIELD ELECTROSTATICS */
1530 felec = _mm256_mul_pd(qq12,_mm256_sub_pd(_mm256_mul_pd(rinv12,rinvsq12),krf2));
1532 cutoff_mask = _mm256_cmp_pd(rsq12,rcutoff2,_CMP_LT_OQ);
1536 fscal = _mm256_and_pd(fscal,cutoff_mask);
1538 /* Calculate temporary vectorial force */
1539 tx = _mm256_mul_pd(fscal,dx12);
1540 ty = _mm256_mul_pd(fscal,dy12);
1541 tz = _mm256_mul_pd(fscal,dz12);
1543 /* Update vectorial force */
1544 fix1 = _mm256_add_pd(fix1,tx);
1545 fiy1 = _mm256_add_pd(fiy1,ty);
1546 fiz1 = _mm256_add_pd(fiz1,tz);
1548 fjx2 = _mm256_add_pd(fjx2,tx);
1549 fjy2 = _mm256_add_pd(fjy2,ty);
1550 fjz2 = _mm256_add_pd(fjz2,tz);
1554 /**************************
1555 * CALCULATE INTERACTIONS *
1556 **************************/
1558 if (gmx_mm256_any_lt(rsq20,rcutoff2))
1561 /* REACTION-FIELD ELECTROSTATICS */
1562 felec = _mm256_mul_pd(qq20,_mm256_sub_pd(_mm256_mul_pd(rinv20,rinvsq20),krf2));
1564 cutoff_mask = _mm256_cmp_pd(rsq20,rcutoff2,_CMP_LT_OQ);
1568 fscal = _mm256_and_pd(fscal,cutoff_mask);
1570 /* Calculate temporary vectorial force */
1571 tx = _mm256_mul_pd(fscal,dx20);
1572 ty = _mm256_mul_pd(fscal,dy20);
1573 tz = _mm256_mul_pd(fscal,dz20);
1575 /* Update vectorial force */
1576 fix2 = _mm256_add_pd(fix2,tx);
1577 fiy2 = _mm256_add_pd(fiy2,ty);
1578 fiz2 = _mm256_add_pd(fiz2,tz);
1580 fjx0 = _mm256_add_pd(fjx0,tx);
1581 fjy0 = _mm256_add_pd(fjy0,ty);
1582 fjz0 = _mm256_add_pd(fjz0,tz);
1586 /**************************
1587 * CALCULATE INTERACTIONS *
1588 **************************/
1590 if (gmx_mm256_any_lt(rsq21,rcutoff2))
1593 /* REACTION-FIELD ELECTROSTATICS */
1594 felec = _mm256_mul_pd(qq21,_mm256_sub_pd(_mm256_mul_pd(rinv21,rinvsq21),krf2));
1596 cutoff_mask = _mm256_cmp_pd(rsq21,rcutoff2,_CMP_LT_OQ);
1600 fscal = _mm256_and_pd(fscal,cutoff_mask);
1602 /* Calculate temporary vectorial force */
1603 tx = _mm256_mul_pd(fscal,dx21);
1604 ty = _mm256_mul_pd(fscal,dy21);
1605 tz = _mm256_mul_pd(fscal,dz21);
1607 /* Update vectorial force */
1608 fix2 = _mm256_add_pd(fix2,tx);
1609 fiy2 = _mm256_add_pd(fiy2,ty);
1610 fiz2 = _mm256_add_pd(fiz2,tz);
1612 fjx1 = _mm256_add_pd(fjx1,tx);
1613 fjy1 = _mm256_add_pd(fjy1,ty);
1614 fjz1 = _mm256_add_pd(fjz1,tz);
1618 /**************************
1619 * CALCULATE INTERACTIONS *
1620 **************************/
1622 if (gmx_mm256_any_lt(rsq22,rcutoff2))
1625 /* REACTION-FIELD ELECTROSTATICS */
1626 felec = _mm256_mul_pd(qq22,_mm256_sub_pd(_mm256_mul_pd(rinv22,rinvsq22),krf2));
1628 cutoff_mask = _mm256_cmp_pd(rsq22,rcutoff2,_CMP_LT_OQ);
1632 fscal = _mm256_and_pd(fscal,cutoff_mask);
1634 /* Calculate temporary vectorial force */
1635 tx = _mm256_mul_pd(fscal,dx22);
1636 ty = _mm256_mul_pd(fscal,dy22);
1637 tz = _mm256_mul_pd(fscal,dz22);
1639 /* Update vectorial force */
1640 fix2 = _mm256_add_pd(fix2,tx);
1641 fiy2 = _mm256_add_pd(fiy2,ty);
1642 fiz2 = _mm256_add_pd(fiz2,tz);
1644 fjx2 = _mm256_add_pd(fjx2,tx);
1645 fjy2 = _mm256_add_pd(fjy2,ty);
1646 fjz2 = _mm256_add_pd(fjz2,tz);
1650 fjptrA = f+j_coord_offsetA;
1651 fjptrB = f+j_coord_offsetB;
1652 fjptrC = f+j_coord_offsetC;
1653 fjptrD = f+j_coord_offsetD;
1655 gmx_mm256_decrement_3rvec_4ptr_swizzle_pd(fjptrA,fjptrB,fjptrC,fjptrD,
1656 fjx0,fjy0,fjz0,fjx1,fjy1,fjz1,fjx2,fjy2,fjz2);
1658 /* Inner loop uses 270 flops */
1661 if(jidx<j_index_end)
1664 /* Get j neighbor index, and coordinate index */
1665 jnrlistA = jjnr[jidx];
1666 jnrlistB = jjnr[jidx+1];
1667 jnrlistC = jjnr[jidx+2];
1668 jnrlistD = jjnr[jidx+3];
1669 /* Sign of each element will be negative for non-real atoms.
1670 * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
1671 * so use it as val = _mm_andnot_pd(mask,val) to clear dummy entries.
1673 tmpmask0 = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
1675 tmpmask1 = _mm_permute_ps(tmpmask0,_GMX_MM_PERMUTE(3,3,2,2));
1676 tmpmask0 = _mm_permute_ps(tmpmask0,_GMX_MM_PERMUTE(1,1,0,0));
1677 dummy_mask = _mm256_castps_pd(gmx_mm256_set_m128(tmpmask1,tmpmask0));
1679 jnrA = (jnrlistA>=0) ? jnrlistA : 0;
1680 jnrB = (jnrlistB>=0) ? jnrlistB : 0;
1681 jnrC = (jnrlistC>=0) ? jnrlistC : 0;
1682 jnrD = (jnrlistD>=0) ? jnrlistD : 0;
1683 j_coord_offsetA = DIM*jnrA;
1684 j_coord_offsetB = DIM*jnrB;
1685 j_coord_offsetC = DIM*jnrC;
1686 j_coord_offsetD = DIM*jnrD;
1688 /* load j atom coordinates */
1689 gmx_mm256_load_3rvec_4ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
1690 x+j_coord_offsetC,x+j_coord_offsetD,
1691 &jx0,&jy0,&jz0,&jx1,&jy1,&jz1,&jx2,&jy2,&jz2);
1693 /* Calculate displacement vector */
1694 dx00 = _mm256_sub_pd(ix0,jx0);
1695 dy00 = _mm256_sub_pd(iy0,jy0);
1696 dz00 = _mm256_sub_pd(iz0,jz0);
1697 dx01 = _mm256_sub_pd(ix0,jx1);
1698 dy01 = _mm256_sub_pd(iy0,jy1);
1699 dz01 = _mm256_sub_pd(iz0,jz1);
1700 dx02 = _mm256_sub_pd(ix0,jx2);
1701 dy02 = _mm256_sub_pd(iy0,jy2);
1702 dz02 = _mm256_sub_pd(iz0,jz2);
1703 dx10 = _mm256_sub_pd(ix1,jx0);
1704 dy10 = _mm256_sub_pd(iy1,jy0);
1705 dz10 = _mm256_sub_pd(iz1,jz0);
1706 dx11 = _mm256_sub_pd(ix1,jx1);
1707 dy11 = _mm256_sub_pd(iy1,jy1);
1708 dz11 = _mm256_sub_pd(iz1,jz1);
1709 dx12 = _mm256_sub_pd(ix1,jx2);
1710 dy12 = _mm256_sub_pd(iy1,jy2);
1711 dz12 = _mm256_sub_pd(iz1,jz2);
1712 dx20 = _mm256_sub_pd(ix2,jx0);
1713 dy20 = _mm256_sub_pd(iy2,jy0);
1714 dz20 = _mm256_sub_pd(iz2,jz0);
1715 dx21 = _mm256_sub_pd(ix2,jx1);
1716 dy21 = _mm256_sub_pd(iy2,jy1);
1717 dz21 = _mm256_sub_pd(iz2,jz1);
1718 dx22 = _mm256_sub_pd(ix2,jx2);
1719 dy22 = _mm256_sub_pd(iy2,jy2);
1720 dz22 = _mm256_sub_pd(iz2,jz2);
1722 /* Calculate squared distance and things based on it */
1723 rsq00 = gmx_mm256_calc_rsq_pd(dx00,dy00,dz00);
1724 rsq01 = gmx_mm256_calc_rsq_pd(dx01,dy01,dz01);
1725 rsq02 = gmx_mm256_calc_rsq_pd(dx02,dy02,dz02);
1726 rsq10 = gmx_mm256_calc_rsq_pd(dx10,dy10,dz10);
1727 rsq11 = gmx_mm256_calc_rsq_pd(dx11,dy11,dz11);
1728 rsq12 = gmx_mm256_calc_rsq_pd(dx12,dy12,dz12);
1729 rsq20 = gmx_mm256_calc_rsq_pd(dx20,dy20,dz20);
1730 rsq21 = gmx_mm256_calc_rsq_pd(dx21,dy21,dz21);
1731 rsq22 = gmx_mm256_calc_rsq_pd(dx22,dy22,dz22);
1733 rinv00 = gmx_mm256_invsqrt_pd(rsq00);
1734 rinv01 = gmx_mm256_invsqrt_pd(rsq01);
1735 rinv02 = gmx_mm256_invsqrt_pd(rsq02);
1736 rinv10 = gmx_mm256_invsqrt_pd(rsq10);
1737 rinv11 = gmx_mm256_invsqrt_pd(rsq11);
1738 rinv12 = gmx_mm256_invsqrt_pd(rsq12);
1739 rinv20 = gmx_mm256_invsqrt_pd(rsq20);
1740 rinv21 = gmx_mm256_invsqrt_pd(rsq21);
1741 rinv22 = gmx_mm256_invsqrt_pd(rsq22);
1743 rinvsq00 = _mm256_mul_pd(rinv00,rinv00);
1744 rinvsq01 = _mm256_mul_pd(rinv01,rinv01);
1745 rinvsq02 = _mm256_mul_pd(rinv02,rinv02);
1746 rinvsq10 = _mm256_mul_pd(rinv10,rinv10);
1747 rinvsq11 = _mm256_mul_pd(rinv11,rinv11);
1748 rinvsq12 = _mm256_mul_pd(rinv12,rinv12);
1749 rinvsq20 = _mm256_mul_pd(rinv20,rinv20);
1750 rinvsq21 = _mm256_mul_pd(rinv21,rinv21);
1751 rinvsq22 = _mm256_mul_pd(rinv22,rinv22);
1753 fjx0 = _mm256_setzero_pd();
1754 fjy0 = _mm256_setzero_pd();
1755 fjz0 = _mm256_setzero_pd();
1756 fjx1 = _mm256_setzero_pd();
1757 fjy1 = _mm256_setzero_pd();
1758 fjz1 = _mm256_setzero_pd();
1759 fjx2 = _mm256_setzero_pd();
1760 fjy2 = _mm256_setzero_pd();
1761 fjz2 = _mm256_setzero_pd();
1763 /**************************
1764 * CALCULATE INTERACTIONS *
1765 **************************/
1767 if (gmx_mm256_any_lt(rsq00,rcutoff2))
1770 /* REACTION-FIELD ELECTROSTATICS */
1771 felec = _mm256_mul_pd(qq00,_mm256_sub_pd(_mm256_mul_pd(rinv00,rinvsq00),krf2));
1773 cutoff_mask = _mm256_cmp_pd(rsq00,rcutoff2,_CMP_LT_OQ);
1777 fscal = _mm256_and_pd(fscal,cutoff_mask);
1779 fscal = _mm256_andnot_pd(dummy_mask,fscal);
1781 /* Calculate temporary vectorial force */
1782 tx = _mm256_mul_pd(fscal,dx00);
1783 ty = _mm256_mul_pd(fscal,dy00);
1784 tz = _mm256_mul_pd(fscal,dz00);
1786 /* Update vectorial force */
1787 fix0 = _mm256_add_pd(fix0,tx);
1788 fiy0 = _mm256_add_pd(fiy0,ty);
1789 fiz0 = _mm256_add_pd(fiz0,tz);
1791 fjx0 = _mm256_add_pd(fjx0,tx);
1792 fjy0 = _mm256_add_pd(fjy0,ty);
1793 fjz0 = _mm256_add_pd(fjz0,tz);
1797 /**************************
1798 * CALCULATE INTERACTIONS *
1799 **************************/
1801 if (gmx_mm256_any_lt(rsq01,rcutoff2))
1804 /* REACTION-FIELD ELECTROSTATICS */
1805 felec = _mm256_mul_pd(qq01,_mm256_sub_pd(_mm256_mul_pd(rinv01,rinvsq01),krf2));
1807 cutoff_mask = _mm256_cmp_pd(rsq01,rcutoff2,_CMP_LT_OQ);
1811 fscal = _mm256_and_pd(fscal,cutoff_mask);
1813 fscal = _mm256_andnot_pd(dummy_mask,fscal);
1815 /* Calculate temporary vectorial force */
1816 tx = _mm256_mul_pd(fscal,dx01);
1817 ty = _mm256_mul_pd(fscal,dy01);
1818 tz = _mm256_mul_pd(fscal,dz01);
1820 /* Update vectorial force */
1821 fix0 = _mm256_add_pd(fix0,tx);
1822 fiy0 = _mm256_add_pd(fiy0,ty);
1823 fiz0 = _mm256_add_pd(fiz0,tz);
1825 fjx1 = _mm256_add_pd(fjx1,tx);
1826 fjy1 = _mm256_add_pd(fjy1,ty);
1827 fjz1 = _mm256_add_pd(fjz1,tz);
1831 /**************************
1832 * CALCULATE INTERACTIONS *
1833 **************************/
1835 if (gmx_mm256_any_lt(rsq02,rcutoff2))
1838 /* REACTION-FIELD ELECTROSTATICS */
1839 felec = _mm256_mul_pd(qq02,_mm256_sub_pd(_mm256_mul_pd(rinv02,rinvsq02),krf2));
1841 cutoff_mask = _mm256_cmp_pd(rsq02,rcutoff2,_CMP_LT_OQ);
1845 fscal = _mm256_and_pd(fscal,cutoff_mask);
1847 fscal = _mm256_andnot_pd(dummy_mask,fscal);
1849 /* Calculate temporary vectorial force */
1850 tx = _mm256_mul_pd(fscal,dx02);
1851 ty = _mm256_mul_pd(fscal,dy02);
1852 tz = _mm256_mul_pd(fscal,dz02);
1854 /* Update vectorial force */
1855 fix0 = _mm256_add_pd(fix0,tx);
1856 fiy0 = _mm256_add_pd(fiy0,ty);
1857 fiz0 = _mm256_add_pd(fiz0,tz);
1859 fjx2 = _mm256_add_pd(fjx2,tx);
1860 fjy2 = _mm256_add_pd(fjy2,ty);
1861 fjz2 = _mm256_add_pd(fjz2,tz);
1865 /**************************
1866 * CALCULATE INTERACTIONS *
1867 **************************/
1869 if (gmx_mm256_any_lt(rsq10,rcutoff2))
1872 /* REACTION-FIELD ELECTROSTATICS */
1873 felec = _mm256_mul_pd(qq10,_mm256_sub_pd(_mm256_mul_pd(rinv10,rinvsq10),krf2));
1875 cutoff_mask = _mm256_cmp_pd(rsq10,rcutoff2,_CMP_LT_OQ);
1879 fscal = _mm256_and_pd(fscal,cutoff_mask);
1881 fscal = _mm256_andnot_pd(dummy_mask,fscal);
1883 /* Calculate temporary vectorial force */
1884 tx = _mm256_mul_pd(fscal,dx10);
1885 ty = _mm256_mul_pd(fscal,dy10);
1886 tz = _mm256_mul_pd(fscal,dz10);
1888 /* Update vectorial force */
1889 fix1 = _mm256_add_pd(fix1,tx);
1890 fiy1 = _mm256_add_pd(fiy1,ty);
1891 fiz1 = _mm256_add_pd(fiz1,tz);
1893 fjx0 = _mm256_add_pd(fjx0,tx);
1894 fjy0 = _mm256_add_pd(fjy0,ty);
1895 fjz0 = _mm256_add_pd(fjz0,tz);
1899 /**************************
1900 * CALCULATE INTERACTIONS *
1901 **************************/
1903 if (gmx_mm256_any_lt(rsq11,rcutoff2))
1906 /* REACTION-FIELD ELECTROSTATICS */
1907 felec = _mm256_mul_pd(qq11,_mm256_sub_pd(_mm256_mul_pd(rinv11,rinvsq11),krf2));
1909 cutoff_mask = _mm256_cmp_pd(rsq11,rcutoff2,_CMP_LT_OQ);
1913 fscal = _mm256_and_pd(fscal,cutoff_mask);
1915 fscal = _mm256_andnot_pd(dummy_mask,fscal);
1917 /* Calculate temporary vectorial force */
1918 tx = _mm256_mul_pd(fscal,dx11);
1919 ty = _mm256_mul_pd(fscal,dy11);
1920 tz = _mm256_mul_pd(fscal,dz11);
1922 /* Update vectorial force */
1923 fix1 = _mm256_add_pd(fix1,tx);
1924 fiy1 = _mm256_add_pd(fiy1,ty);
1925 fiz1 = _mm256_add_pd(fiz1,tz);
1927 fjx1 = _mm256_add_pd(fjx1,tx);
1928 fjy1 = _mm256_add_pd(fjy1,ty);
1929 fjz1 = _mm256_add_pd(fjz1,tz);
1933 /**************************
1934 * CALCULATE INTERACTIONS *
1935 **************************/
1937 if (gmx_mm256_any_lt(rsq12,rcutoff2))
1940 /* REACTION-FIELD ELECTROSTATICS */
1941 felec = _mm256_mul_pd(qq12,_mm256_sub_pd(_mm256_mul_pd(rinv12,rinvsq12),krf2));
1943 cutoff_mask = _mm256_cmp_pd(rsq12,rcutoff2,_CMP_LT_OQ);
1947 fscal = _mm256_and_pd(fscal,cutoff_mask);
1949 fscal = _mm256_andnot_pd(dummy_mask,fscal);
1951 /* Calculate temporary vectorial force */
1952 tx = _mm256_mul_pd(fscal,dx12);
1953 ty = _mm256_mul_pd(fscal,dy12);
1954 tz = _mm256_mul_pd(fscal,dz12);
1956 /* Update vectorial force */
1957 fix1 = _mm256_add_pd(fix1,tx);
1958 fiy1 = _mm256_add_pd(fiy1,ty);
1959 fiz1 = _mm256_add_pd(fiz1,tz);
1961 fjx2 = _mm256_add_pd(fjx2,tx);
1962 fjy2 = _mm256_add_pd(fjy2,ty);
1963 fjz2 = _mm256_add_pd(fjz2,tz);
1967 /**************************
1968 * CALCULATE INTERACTIONS *
1969 **************************/
1971 if (gmx_mm256_any_lt(rsq20,rcutoff2))
1974 /* REACTION-FIELD ELECTROSTATICS */
1975 felec = _mm256_mul_pd(qq20,_mm256_sub_pd(_mm256_mul_pd(rinv20,rinvsq20),krf2));
1977 cutoff_mask = _mm256_cmp_pd(rsq20,rcutoff2,_CMP_LT_OQ);
1981 fscal = _mm256_and_pd(fscal,cutoff_mask);
1983 fscal = _mm256_andnot_pd(dummy_mask,fscal);
1985 /* Calculate temporary vectorial force */
1986 tx = _mm256_mul_pd(fscal,dx20);
1987 ty = _mm256_mul_pd(fscal,dy20);
1988 tz = _mm256_mul_pd(fscal,dz20);
1990 /* Update vectorial force */
1991 fix2 = _mm256_add_pd(fix2,tx);
1992 fiy2 = _mm256_add_pd(fiy2,ty);
1993 fiz2 = _mm256_add_pd(fiz2,tz);
1995 fjx0 = _mm256_add_pd(fjx0,tx);
1996 fjy0 = _mm256_add_pd(fjy0,ty);
1997 fjz0 = _mm256_add_pd(fjz0,tz);
2001 /**************************
2002 * CALCULATE INTERACTIONS *
2003 **************************/
2005 if (gmx_mm256_any_lt(rsq21,rcutoff2))
2008 /* REACTION-FIELD ELECTROSTATICS */
2009 felec = _mm256_mul_pd(qq21,_mm256_sub_pd(_mm256_mul_pd(rinv21,rinvsq21),krf2));
2011 cutoff_mask = _mm256_cmp_pd(rsq21,rcutoff2,_CMP_LT_OQ);
2015 fscal = _mm256_and_pd(fscal,cutoff_mask);
2017 fscal = _mm256_andnot_pd(dummy_mask,fscal);
2019 /* Calculate temporary vectorial force */
2020 tx = _mm256_mul_pd(fscal,dx21);
2021 ty = _mm256_mul_pd(fscal,dy21);
2022 tz = _mm256_mul_pd(fscal,dz21);
2024 /* Update vectorial force */
2025 fix2 = _mm256_add_pd(fix2,tx);
2026 fiy2 = _mm256_add_pd(fiy2,ty);
2027 fiz2 = _mm256_add_pd(fiz2,tz);
2029 fjx1 = _mm256_add_pd(fjx1,tx);
2030 fjy1 = _mm256_add_pd(fjy1,ty);
2031 fjz1 = _mm256_add_pd(fjz1,tz);
2035 /**************************
2036 * CALCULATE INTERACTIONS *
2037 **************************/
2039 if (gmx_mm256_any_lt(rsq22,rcutoff2))
2042 /* REACTION-FIELD ELECTROSTATICS */
2043 felec = _mm256_mul_pd(qq22,_mm256_sub_pd(_mm256_mul_pd(rinv22,rinvsq22),krf2));
2045 cutoff_mask = _mm256_cmp_pd(rsq22,rcutoff2,_CMP_LT_OQ);
2049 fscal = _mm256_and_pd(fscal,cutoff_mask);
2051 fscal = _mm256_andnot_pd(dummy_mask,fscal);
2053 /* Calculate temporary vectorial force */
2054 tx = _mm256_mul_pd(fscal,dx22);
2055 ty = _mm256_mul_pd(fscal,dy22);
2056 tz = _mm256_mul_pd(fscal,dz22);
2058 /* Update vectorial force */
2059 fix2 = _mm256_add_pd(fix2,tx);
2060 fiy2 = _mm256_add_pd(fiy2,ty);
2061 fiz2 = _mm256_add_pd(fiz2,tz);
2063 fjx2 = _mm256_add_pd(fjx2,tx);
2064 fjy2 = _mm256_add_pd(fjy2,ty);
2065 fjz2 = _mm256_add_pd(fjz2,tz);
2069 fjptrA = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
2070 fjptrB = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
2071 fjptrC = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
2072 fjptrD = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
2074 gmx_mm256_decrement_3rvec_4ptr_swizzle_pd(fjptrA,fjptrB,fjptrC,fjptrD,
2075 fjx0,fjy0,fjz0,fjx1,fjy1,fjz1,fjx2,fjy2,fjz2);
2077 /* Inner loop uses 270 flops */
2080 /* End of innermost loop */
2082 gmx_mm256_update_iforce_3atom_swizzle_pd(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,
2083 f+i_coord_offset,fshift+i_shift_offset);
2085 /* Increment number of inner iterations */
2086 inneriter += j_index_end - j_index_start;
2088 /* Outer loop uses 18 flops */
2091 /* Increment number of outer iterations */
2094 /* Update outer/inner flops */
2096 inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_W3W3_F,outeriter*18 + inneriter*270);