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_single 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_single.h"
48 #include "kernelutil_x86_avx_256_single.h"
51 * Gromacs nonbonded kernel: nb_kernel_ElecRFCut_VdwNone_GeomW4P1_VF_avx_256_single
52 * Electrostatics interaction: ReactionField
53 * VdW interaction: None
54 * Geometry: Water4-Particle
55 * Calculate force/pot: PotentialAndForce
58 nb_kernel_ElecRFCut_VdwNone_GeomW4P1_VF_avx_256_single
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,E,F,G,H refer to j loop unrolling done with AVX, e.g. for the eight 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 jnrE,jnrF,jnrG,jnrH;
76 int jnrlistA,jnrlistB,jnrlistC,jnrlistD;
77 int jnrlistE,jnrlistF,jnrlistG,jnrlistH;
78 int j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
79 int j_coord_offsetE,j_coord_offsetF,j_coord_offsetG,j_coord_offsetH;
80 int *iinr,*jindex,*jjnr,*shiftidx,*gid;
82 real *shiftvec,*fshift,*x,*f;
83 real *fjptrA,*fjptrB,*fjptrC,*fjptrD,*fjptrE,*fjptrF,*fjptrG,*fjptrH;
85 __m256 tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
86 real * vdwioffsetptr1;
87 __m256 ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
88 real * vdwioffsetptr2;
89 __m256 ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
90 real * vdwioffsetptr3;
91 __m256 ix3,iy3,iz3,fix3,fiy3,fiz3,iq3,isai3;
92 int vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D,vdwjidx0E,vdwjidx0F,vdwjidx0G,vdwjidx0H;
93 __m256 jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
94 __m256 dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
95 __m256 dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
96 __m256 dx30,dy30,dz30,rsq30,rinv30,rinvsq30,r30,qq30,c6_30,c12_30;
97 __m256 velec,felec,velecsum,facel,crf,krf,krf2;
99 __m256 dummy_mask,cutoff_mask;
100 __m256 signbit = _mm256_castsi256_ps( _mm256_set1_epi32(0x80000000) );
101 __m256 one = _mm256_set1_ps(1.0);
102 __m256 two = _mm256_set1_ps(2.0);
108 jindex = nlist->jindex;
110 shiftidx = nlist->shift;
112 shiftvec = fr->shift_vec[0];
113 fshift = fr->fshift[0];
114 facel = _mm256_set1_ps(fr->epsfac);
115 charge = mdatoms->chargeA;
116 krf = _mm256_set1_ps(fr->ic->k_rf);
117 krf2 = _mm256_set1_ps(fr->ic->k_rf*2.0);
118 crf = _mm256_set1_ps(fr->ic->c_rf);
120 /* Setup water-specific parameters */
121 inr = nlist->iinr[0];
122 iq1 = _mm256_mul_ps(facel,_mm256_set1_ps(charge[inr+1]));
123 iq2 = _mm256_mul_ps(facel,_mm256_set1_ps(charge[inr+2]));
124 iq3 = _mm256_mul_ps(facel,_mm256_set1_ps(charge[inr+3]));
126 /* When we use explicit cutoffs the value must be identical for elec and VdW, so use elec as an arbitrary choice */
127 rcutoff_scalar = fr->rcoulomb;
128 rcutoff = _mm256_set1_ps(rcutoff_scalar);
129 rcutoff2 = _mm256_mul_ps(rcutoff,rcutoff);
131 /* Avoid stupid compiler warnings */
132 jnrA = jnrB = jnrC = jnrD = jnrE = jnrF = jnrG = jnrH = 0;
145 for(iidx=0;iidx<4*DIM;iidx++)
150 /* Start outer loop over neighborlists */
151 for(iidx=0; iidx<nri; iidx++)
153 /* Load shift vector for this list */
154 i_shift_offset = DIM*shiftidx[iidx];
156 /* Load limits for loop over neighbors */
157 j_index_start = jindex[iidx];
158 j_index_end = jindex[iidx+1];
160 /* Get outer coordinate index */
162 i_coord_offset = DIM*inr;
164 /* Load i particle coords and add shift vector */
165 gmx_mm256_load_shift_and_3rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset+DIM,
166 &ix1,&iy1,&iz1,&ix2,&iy2,&iz2,&ix3,&iy3,&iz3);
168 fix1 = _mm256_setzero_ps();
169 fiy1 = _mm256_setzero_ps();
170 fiz1 = _mm256_setzero_ps();
171 fix2 = _mm256_setzero_ps();
172 fiy2 = _mm256_setzero_ps();
173 fiz2 = _mm256_setzero_ps();
174 fix3 = _mm256_setzero_ps();
175 fiy3 = _mm256_setzero_ps();
176 fiz3 = _mm256_setzero_ps();
178 /* Reset potential sums */
179 velecsum = _mm256_setzero_ps();
181 /* Start inner kernel loop */
182 for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+7]>=0; jidx+=8)
185 /* Get j neighbor index, and coordinate index */
194 j_coord_offsetA = DIM*jnrA;
195 j_coord_offsetB = DIM*jnrB;
196 j_coord_offsetC = DIM*jnrC;
197 j_coord_offsetD = DIM*jnrD;
198 j_coord_offsetE = DIM*jnrE;
199 j_coord_offsetF = DIM*jnrF;
200 j_coord_offsetG = DIM*jnrG;
201 j_coord_offsetH = DIM*jnrH;
203 /* load j atom coordinates */
204 gmx_mm256_load_1rvec_8ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
205 x+j_coord_offsetC,x+j_coord_offsetD,
206 x+j_coord_offsetE,x+j_coord_offsetF,
207 x+j_coord_offsetG,x+j_coord_offsetH,
210 /* Calculate displacement vector */
211 dx10 = _mm256_sub_ps(ix1,jx0);
212 dy10 = _mm256_sub_ps(iy1,jy0);
213 dz10 = _mm256_sub_ps(iz1,jz0);
214 dx20 = _mm256_sub_ps(ix2,jx0);
215 dy20 = _mm256_sub_ps(iy2,jy0);
216 dz20 = _mm256_sub_ps(iz2,jz0);
217 dx30 = _mm256_sub_ps(ix3,jx0);
218 dy30 = _mm256_sub_ps(iy3,jy0);
219 dz30 = _mm256_sub_ps(iz3,jz0);
221 /* Calculate squared distance and things based on it */
222 rsq10 = gmx_mm256_calc_rsq_ps(dx10,dy10,dz10);
223 rsq20 = gmx_mm256_calc_rsq_ps(dx20,dy20,dz20);
224 rsq30 = gmx_mm256_calc_rsq_ps(dx30,dy30,dz30);
226 rinv10 = gmx_mm256_invsqrt_ps(rsq10);
227 rinv20 = gmx_mm256_invsqrt_ps(rsq20);
228 rinv30 = gmx_mm256_invsqrt_ps(rsq30);
230 rinvsq10 = _mm256_mul_ps(rinv10,rinv10);
231 rinvsq20 = _mm256_mul_ps(rinv20,rinv20);
232 rinvsq30 = _mm256_mul_ps(rinv30,rinv30);
234 /* Load parameters for j particles */
235 jq0 = gmx_mm256_load_8real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
236 charge+jnrC+0,charge+jnrD+0,
237 charge+jnrE+0,charge+jnrF+0,
238 charge+jnrG+0,charge+jnrH+0);
240 fjx0 = _mm256_setzero_ps();
241 fjy0 = _mm256_setzero_ps();
242 fjz0 = _mm256_setzero_ps();
244 /**************************
245 * CALCULATE INTERACTIONS *
246 **************************/
248 if (gmx_mm256_any_lt(rsq10,rcutoff2))
251 /* Compute parameters for interactions between i and j atoms */
252 qq10 = _mm256_mul_ps(iq1,jq0);
254 /* REACTION-FIELD ELECTROSTATICS */
255 velec = _mm256_mul_ps(qq10,_mm256_sub_ps(_mm256_add_ps(rinv10,_mm256_mul_ps(krf,rsq10)),crf));
256 felec = _mm256_mul_ps(qq10,_mm256_sub_ps(_mm256_mul_ps(rinv10,rinvsq10),krf2));
258 cutoff_mask = _mm256_cmp_ps(rsq10,rcutoff2,_CMP_LT_OQ);
260 /* Update potential sum for this i atom from the interaction with this j atom. */
261 velec = _mm256_and_ps(velec,cutoff_mask);
262 velecsum = _mm256_add_ps(velecsum,velec);
266 fscal = _mm256_and_ps(fscal,cutoff_mask);
268 /* Calculate temporary vectorial force */
269 tx = _mm256_mul_ps(fscal,dx10);
270 ty = _mm256_mul_ps(fscal,dy10);
271 tz = _mm256_mul_ps(fscal,dz10);
273 /* Update vectorial force */
274 fix1 = _mm256_add_ps(fix1,tx);
275 fiy1 = _mm256_add_ps(fiy1,ty);
276 fiz1 = _mm256_add_ps(fiz1,tz);
278 fjx0 = _mm256_add_ps(fjx0,tx);
279 fjy0 = _mm256_add_ps(fjy0,ty);
280 fjz0 = _mm256_add_ps(fjz0,tz);
284 /**************************
285 * CALCULATE INTERACTIONS *
286 **************************/
288 if (gmx_mm256_any_lt(rsq20,rcutoff2))
291 /* Compute parameters for interactions between i and j atoms */
292 qq20 = _mm256_mul_ps(iq2,jq0);
294 /* REACTION-FIELD ELECTROSTATICS */
295 velec = _mm256_mul_ps(qq20,_mm256_sub_ps(_mm256_add_ps(rinv20,_mm256_mul_ps(krf,rsq20)),crf));
296 felec = _mm256_mul_ps(qq20,_mm256_sub_ps(_mm256_mul_ps(rinv20,rinvsq20),krf2));
298 cutoff_mask = _mm256_cmp_ps(rsq20,rcutoff2,_CMP_LT_OQ);
300 /* Update potential sum for this i atom from the interaction with this j atom. */
301 velec = _mm256_and_ps(velec,cutoff_mask);
302 velecsum = _mm256_add_ps(velecsum,velec);
306 fscal = _mm256_and_ps(fscal,cutoff_mask);
308 /* Calculate temporary vectorial force */
309 tx = _mm256_mul_ps(fscal,dx20);
310 ty = _mm256_mul_ps(fscal,dy20);
311 tz = _mm256_mul_ps(fscal,dz20);
313 /* Update vectorial force */
314 fix2 = _mm256_add_ps(fix2,tx);
315 fiy2 = _mm256_add_ps(fiy2,ty);
316 fiz2 = _mm256_add_ps(fiz2,tz);
318 fjx0 = _mm256_add_ps(fjx0,tx);
319 fjy0 = _mm256_add_ps(fjy0,ty);
320 fjz0 = _mm256_add_ps(fjz0,tz);
324 /**************************
325 * CALCULATE INTERACTIONS *
326 **************************/
328 if (gmx_mm256_any_lt(rsq30,rcutoff2))
331 /* Compute parameters for interactions between i and j atoms */
332 qq30 = _mm256_mul_ps(iq3,jq0);
334 /* REACTION-FIELD ELECTROSTATICS */
335 velec = _mm256_mul_ps(qq30,_mm256_sub_ps(_mm256_add_ps(rinv30,_mm256_mul_ps(krf,rsq30)),crf));
336 felec = _mm256_mul_ps(qq30,_mm256_sub_ps(_mm256_mul_ps(rinv30,rinvsq30),krf2));
338 cutoff_mask = _mm256_cmp_ps(rsq30,rcutoff2,_CMP_LT_OQ);
340 /* Update potential sum for this i atom from the interaction with this j atom. */
341 velec = _mm256_and_ps(velec,cutoff_mask);
342 velecsum = _mm256_add_ps(velecsum,velec);
346 fscal = _mm256_and_ps(fscal,cutoff_mask);
348 /* Calculate temporary vectorial force */
349 tx = _mm256_mul_ps(fscal,dx30);
350 ty = _mm256_mul_ps(fscal,dy30);
351 tz = _mm256_mul_ps(fscal,dz30);
353 /* Update vectorial force */
354 fix3 = _mm256_add_ps(fix3,tx);
355 fiy3 = _mm256_add_ps(fiy3,ty);
356 fiz3 = _mm256_add_ps(fiz3,tz);
358 fjx0 = _mm256_add_ps(fjx0,tx);
359 fjy0 = _mm256_add_ps(fjy0,ty);
360 fjz0 = _mm256_add_ps(fjz0,tz);
364 fjptrA = f+j_coord_offsetA;
365 fjptrB = f+j_coord_offsetB;
366 fjptrC = f+j_coord_offsetC;
367 fjptrD = f+j_coord_offsetD;
368 fjptrE = f+j_coord_offsetE;
369 fjptrF = f+j_coord_offsetF;
370 fjptrG = f+j_coord_offsetG;
371 fjptrH = f+j_coord_offsetH;
373 gmx_mm256_decrement_1rvec_8ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjptrE,fjptrF,fjptrG,fjptrH,fjx0,fjy0,fjz0);
375 /* Inner loop uses 111 flops */
381 /* Get j neighbor index, and coordinate index */
382 jnrlistA = jjnr[jidx];
383 jnrlistB = jjnr[jidx+1];
384 jnrlistC = jjnr[jidx+2];
385 jnrlistD = jjnr[jidx+3];
386 jnrlistE = jjnr[jidx+4];
387 jnrlistF = jjnr[jidx+5];
388 jnrlistG = jjnr[jidx+6];
389 jnrlistH = jjnr[jidx+7];
390 /* Sign of each element will be negative for non-real atoms.
391 * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
392 * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
394 dummy_mask = gmx_mm256_set_m128(gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx+4)),_mm_setzero_si128())),
395 gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128())));
397 jnrA = (jnrlistA>=0) ? jnrlistA : 0;
398 jnrB = (jnrlistB>=0) ? jnrlistB : 0;
399 jnrC = (jnrlistC>=0) ? jnrlistC : 0;
400 jnrD = (jnrlistD>=0) ? jnrlistD : 0;
401 jnrE = (jnrlistE>=0) ? jnrlistE : 0;
402 jnrF = (jnrlistF>=0) ? jnrlistF : 0;
403 jnrG = (jnrlistG>=0) ? jnrlistG : 0;
404 jnrH = (jnrlistH>=0) ? jnrlistH : 0;
405 j_coord_offsetA = DIM*jnrA;
406 j_coord_offsetB = DIM*jnrB;
407 j_coord_offsetC = DIM*jnrC;
408 j_coord_offsetD = DIM*jnrD;
409 j_coord_offsetE = DIM*jnrE;
410 j_coord_offsetF = DIM*jnrF;
411 j_coord_offsetG = DIM*jnrG;
412 j_coord_offsetH = DIM*jnrH;
414 /* load j atom coordinates */
415 gmx_mm256_load_1rvec_8ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
416 x+j_coord_offsetC,x+j_coord_offsetD,
417 x+j_coord_offsetE,x+j_coord_offsetF,
418 x+j_coord_offsetG,x+j_coord_offsetH,
421 /* Calculate displacement vector */
422 dx10 = _mm256_sub_ps(ix1,jx0);
423 dy10 = _mm256_sub_ps(iy1,jy0);
424 dz10 = _mm256_sub_ps(iz1,jz0);
425 dx20 = _mm256_sub_ps(ix2,jx0);
426 dy20 = _mm256_sub_ps(iy2,jy0);
427 dz20 = _mm256_sub_ps(iz2,jz0);
428 dx30 = _mm256_sub_ps(ix3,jx0);
429 dy30 = _mm256_sub_ps(iy3,jy0);
430 dz30 = _mm256_sub_ps(iz3,jz0);
432 /* Calculate squared distance and things based on it */
433 rsq10 = gmx_mm256_calc_rsq_ps(dx10,dy10,dz10);
434 rsq20 = gmx_mm256_calc_rsq_ps(dx20,dy20,dz20);
435 rsq30 = gmx_mm256_calc_rsq_ps(dx30,dy30,dz30);
437 rinv10 = gmx_mm256_invsqrt_ps(rsq10);
438 rinv20 = gmx_mm256_invsqrt_ps(rsq20);
439 rinv30 = gmx_mm256_invsqrt_ps(rsq30);
441 rinvsq10 = _mm256_mul_ps(rinv10,rinv10);
442 rinvsq20 = _mm256_mul_ps(rinv20,rinv20);
443 rinvsq30 = _mm256_mul_ps(rinv30,rinv30);
445 /* Load parameters for j particles */
446 jq0 = gmx_mm256_load_8real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
447 charge+jnrC+0,charge+jnrD+0,
448 charge+jnrE+0,charge+jnrF+0,
449 charge+jnrG+0,charge+jnrH+0);
451 fjx0 = _mm256_setzero_ps();
452 fjy0 = _mm256_setzero_ps();
453 fjz0 = _mm256_setzero_ps();
455 /**************************
456 * CALCULATE INTERACTIONS *
457 **************************/
459 if (gmx_mm256_any_lt(rsq10,rcutoff2))
462 /* Compute parameters for interactions between i and j atoms */
463 qq10 = _mm256_mul_ps(iq1,jq0);
465 /* REACTION-FIELD ELECTROSTATICS */
466 velec = _mm256_mul_ps(qq10,_mm256_sub_ps(_mm256_add_ps(rinv10,_mm256_mul_ps(krf,rsq10)),crf));
467 felec = _mm256_mul_ps(qq10,_mm256_sub_ps(_mm256_mul_ps(rinv10,rinvsq10),krf2));
469 cutoff_mask = _mm256_cmp_ps(rsq10,rcutoff2,_CMP_LT_OQ);
471 /* Update potential sum for this i atom from the interaction with this j atom. */
472 velec = _mm256_and_ps(velec,cutoff_mask);
473 velec = _mm256_andnot_ps(dummy_mask,velec);
474 velecsum = _mm256_add_ps(velecsum,velec);
478 fscal = _mm256_and_ps(fscal,cutoff_mask);
480 fscal = _mm256_andnot_ps(dummy_mask,fscal);
482 /* Calculate temporary vectorial force */
483 tx = _mm256_mul_ps(fscal,dx10);
484 ty = _mm256_mul_ps(fscal,dy10);
485 tz = _mm256_mul_ps(fscal,dz10);
487 /* Update vectorial force */
488 fix1 = _mm256_add_ps(fix1,tx);
489 fiy1 = _mm256_add_ps(fiy1,ty);
490 fiz1 = _mm256_add_ps(fiz1,tz);
492 fjx0 = _mm256_add_ps(fjx0,tx);
493 fjy0 = _mm256_add_ps(fjy0,ty);
494 fjz0 = _mm256_add_ps(fjz0,tz);
498 /**************************
499 * CALCULATE INTERACTIONS *
500 **************************/
502 if (gmx_mm256_any_lt(rsq20,rcutoff2))
505 /* Compute parameters for interactions between i and j atoms */
506 qq20 = _mm256_mul_ps(iq2,jq0);
508 /* REACTION-FIELD ELECTROSTATICS */
509 velec = _mm256_mul_ps(qq20,_mm256_sub_ps(_mm256_add_ps(rinv20,_mm256_mul_ps(krf,rsq20)),crf));
510 felec = _mm256_mul_ps(qq20,_mm256_sub_ps(_mm256_mul_ps(rinv20,rinvsq20),krf2));
512 cutoff_mask = _mm256_cmp_ps(rsq20,rcutoff2,_CMP_LT_OQ);
514 /* Update potential sum for this i atom from the interaction with this j atom. */
515 velec = _mm256_and_ps(velec,cutoff_mask);
516 velec = _mm256_andnot_ps(dummy_mask,velec);
517 velecsum = _mm256_add_ps(velecsum,velec);
521 fscal = _mm256_and_ps(fscal,cutoff_mask);
523 fscal = _mm256_andnot_ps(dummy_mask,fscal);
525 /* Calculate temporary vectorial force */
526 tx = _mm256_mul_ps(fscal,dx20);
527 ty = _mm256_mul_ps(fscal,dy20);
528 tz = _mm256_mul_ps(fscal,dz20);
530 /* Update vectorial force */
531 fix2 = _mm256_add_ps(fix2,tx);
532 fiy2 = _mm256_add_ps(fiy2,ty);
533 fiz2 = _mm256_add_ps(fiz2,tz);
535 fjx0 = _mm256_add_ps(fjx0,tx);
536 fjy0 = _mm256_add_ps(fjy0,ty);
537 fjz0 = _mm256_add_ps(fjz0,tz);
541 /**************************
542 * CALCULATE INTERACTIONS *
543 **************************/
545 if (gmx_mm256_any_lt(rsq30,rcutoff2))
548 /* Compute parameters for interactions between i and j atoms */
549 qq30 = _mm256_mul_ps(iq3,jq0);
551 /* REACTION-FIELD ELECTROSTATICS */
552 velec = _mm256_mul_ps(qq30,_mm256_sub_ps(_mm256_add_ps(rinv30,_mm256_mul_ps(krf,rsq30)),crf));
553 felec = _mm256_mul_ps(qq30,_mm256_sub_ps(_mm256_mul_ps(rinv30,rinvsq30),krf2));
555 cutoff_mask = _mm256_cmp_ps(rsq30,rcutoff2,_CMP_LT_OQ);
557 /* Update potential sum for this i atom from the interaction with this j atom. */
558 velec = _mm256_and_ps(velec,cutoff_mask);
559 velec = _mm256_andnot_ps(dummy_mask,velec);
560 velecsum = _mm256_add_ps(velecsum,velec);
564 fscal = _mm256_and_ps(fscal,cutoff_mask);
566 fscal = _mm256_andnot_ps(dummy_mask,fscal);
568 /* Calculate temporary vectorial force */
569 tx = _mm256_mul_ps(fscal,dx30);
570 ty = _mm256_mul_ps(fscal,dy30);
571 tz = _mm256_mul_ps(fscal,dz30);
573 /* Update vectorial force */
574 fix3 = _mm256_add_ps(fix3,tx);
575 fiy3 = _mm256_add_ps(fiy3,ty);
576 fiz3 = _mm256_add_ps(fiz3,tz);
578 fjx0 = _mm256_add_ps(fjx0,tx);
579 fjy0 = _mm256_add_ps(fjy0,ty);
580 fjz0 = _mm256_add_ps(fjz0,tz);
584 fjptrA = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
585 fjptrB = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
586 fjptrC = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
587 fjptrD = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
588 fjptrE = (jnrlistE>=0) ? f+j_coord_offsetE : scratch;
589 fjptrF = (jnrlistF>=0) ? f+j_coord_offsetF : scratch;
590 fjptrG = (jnrlistG>=0) ? f+j_coord_offsetG : scratch;
591 fjptrH = (jnrlistH>=0) ? f+j_coord_offsetH : scratch;
593 gmx_mm256_decrement_1rvec_8ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjptrE,fjptrF,fjptrG,fjptrH,fjx0,fjy0,fjz0);
595 /* Inner loop uses 111 flops */
598 /* End of innermost loop */
600 gmx_mm256_update_iforce_3atom_swizzle_ps(fix1,fiy1,fiz1,fix2,fiy2,fiz2,fix3,fiy3,fiz3,
601 f+i_coord_offset+DIM,fshift+i_shift_offset);
604 /* Update potential energies */
605 gmx_mm256_update_1pot_ps(velecsum,kernel_data->energygrp_elec+ggid);
607 /* Increment number of inner iterations */
608 inneriter += j_index_end - j_index_start;
610 /* Outer loop uses 19 flops */
613 /* Increment number of outer iterations */
616 /* Update outer/inner flops */
618 inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_W4_VF,outeriter*19 + inneriter*111);
621 * Gromacs nonbonded kernel: nb_kernel_ElecRFCut_VdwNone_GeomW4P1_F_avx_256_single
622 * Electrostatics interaction: ReactionField
623 * VdW interaction: None
624 * Geometry: Water4-Particle
625 * Calculate force/pot: Force
628 nb_kernel_ElecRFCut_VdwNone_GeomW4P1_F_avx_256_single
629 (t_nblist * gmx_restrict nlist,
630 rvec * gmx_restrict xx,
631 rvec * gmx_restrict ff,
632 t_forcerec * gmx_restrict fr,
633 t_mdatoms * gmx_restrict mdatoms,
634 nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
635 t_nrnb * gmx_restrict nrnb)
637 /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
638 * just 0 for non-waters.
639 * Suffixes A,B,C,D,E,F,G,H refer to j loop unrolling done with AVX, e.g. for the eight different
640 * jnr indices corresponding to data put in the four positions in the SIMD register.
642 int i_shift_offset,i_coord_offset,outeriter,inneriter;
643 int j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
644 int jnrA,jnrB,jnrC,jnrD;
645 int jnrE,jnrF,jnrG,jnrH;
646 int jnrlistA,jnrlistB,jnrlistC,jnrlistD;
647 int jnrlistE,jnrlistF,jnrlistG,jnrlistH;
648 int j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
649 int j_coord_offsetE,j_coord_offsetF,j_coord_offsetG,j_coord_offsetH;
650 int *iinr,*jindex,*jjnr,*shiftidx,*gid;
652 real *shiftvec,*fshift,*x,*f;
653 real *fjptrA,*fjptrB,*fjptrC,*fjptrD,*fjptrE,*fjptrF,*fjptrG,*fjptrH;
655 __m256 tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
656 real * vdwioffsetptr1;
657 __m256 ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
658 real * vdwioffsetptr2;
659 __m256 ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
660 real * vdwioffsetptr3;
661 __m256 ix3,iy3,iz3,fix3,fiy3,fiz3,iq3,isai3;
662 int vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D,vdwjidx0E,vdwjidx0F,vdwjidx0G,vdwjidx0H;
663 __m256 jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
664 __m256 dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
665 __m256 dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
666 __m256 dx30,dy30,dz30,rsq30,rinv30,rinvsq30,r30,qq30,c6_30,c12_30;
667 __m256 velec,felec,velecsum,facel,crf,krf,krf2;
669 __m256 dummy_mask,cutoff_mask;
670 __m256 signbit = _mm256_castsi256_ps( _mm256_set1_epi32(0x80000000) );
671 __m256 one = _mm256_set1_ps(1.0);
672 __m256 two = _mm256_set1_ps(2.0);
678 jindex = nlist->jindex;
680 shiftidx = nlist->shift;
682 shiftvec = fr->shift_vec[0];
683 fshift = fr->fshift[0];
684 facel = _mm256_set1_ps(fr->epsfac);
685 charge = mdatoms->chargeA;
686 krf = _mm256_set1_ps(fr->ic->k_rf);
687 krf2 = _mm256_set1_ps(fr->ic->k_rf*2.0);
688 crf = _mm256_set1_ps(fr->ic->c_rf);
690 /* Setup water-specific parameters */
691 inr = nlist->iinr[0];
692 iq1 = _mm256_mul_ps(facel,_mm256_set1_ps(charge[inr+1]));
693 iq2 = _mm256_mul_ps(facel,_mm256_set1_ps(charge[inr+2]));
694 iq3 = _mm256_mul_ps(facel,_mm256_set1_ps(charge[inr+3]));
696 /* When we use explicit cutoffs the value must be identical for elec and VdW, so use elec as an arbitrary choice */
697 rcutoff_scalar = fr->rcoulomb;
698 rcutoff = _mm256_set1_ps(rcutoff_scalar);
699 rcutoff2 = _mm256_mul_ps(rcutoff,rcutoff);
701 /* Avoid stupid compiler warnings */
702 jnrA = jnrB = jnrC = jnrD = jnrE = jnrF = jnrG = jnrH = 0;
715 for(iidx=0;iidx<4*DIM;iidx++)
720 /* Start outer loop over neighborlists */
721 for(iidx=0; iidx<nri; iidx++)
723 /* Load shift vector for this list */
724 i_shift_offset = DIM*shiftidx[iidx];
726 /* Load limits for loop over neighbors */
727 j_index_start = jindex[iidx];
728 j_index_end = jindex[iidx+1];
730 /* Get outer coordinate index */
732 i_coord_offset = DIM*inr;
734 /* Load i particle coords and add shift vector */
735 gmx_mm256_load_shift_and_3rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset+DIM,
736 &ix1,&iy1,&iz1,&ix2,&iy2,&iz2,&ix3,&iy3,&iz3);
738 fix1 = _mm256_setzero_ps();
739 fiy1 = _mm256_setzero_ps();
740 fiz1 = _mm256_setzero_ps();
741 fix2 = _mm256_setzero_ps();
742 fiy2 = _mm256_setzero_ps();
743 fiz2 = _mm256_setzero_ps();
744 fix3 = _mm256_setzero_ps();
745 fiy3 = _mm256_setzero_ps();
746 fiz3 = _mm256_setzero_ps();
748 /* Start inner kernel loop */
749 for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+7]>=0; jidx+=8)
752 /* Get j neighbor index, and coordinate index */
761 j_coord_offsetA = DIM*jnrA;
762 j_coord_offsetB = DIM*jnrB;
763 j_coord_offsetC = DIM*jnrC;
764 j_coord_offsetD = DIM*jnrD;
765 j_coord_offsetE = DIM*jnrE;
766 j_coord_offsetF = DIM*jnrF;
767 j_coord_offsetG = DIM*jnrG;
768 j_coord_offsetH = DIM*jnrH;
770 /* load j atom coordinates */
771 gmx_mm256_load_1rvec_8ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
772 x+j_coord_offsetC,x+j_coord_offsetD,
773 x+j_coord_offsetE,x+j_coord_offsetF,
774 x+j_coord_offsetG,x+j_coord_offsetH,
777 /* Calculate displacement vector */
778 dx10 = _mm256_sub_ps(ix1,jx0);
779 dy10 = _mm256_sub_ps(iy1,jy0);
780 dz10 = _mm256_sub_ps(iz1,jz0);
781 dx20 = _mm256_sub_ps(ix2,jx0);
782 dy20 = _mm256_sub_ps(iy2,jy0);
783 dz20 = _mm256_sub_ps(iz2,jz0);
784 dx30 = _mm256_sub_ps(ix3,jx0);
785 dy30 = _mm256_sub_ps(iy3,jy0);
786 dz30 = _mm256_sub_ps(iz3,jz0);
788 /* Calculate squared distance and things based on it */
789 rsq10 = gmx_mm256_calc_rsq_ps(dx10,dy10,dz10);
790 rsq20 = gmx_mm256_calc_rsq_ps(dx20,dy20,dz20);
791 rsq30 = gmx_mm256_calc_rsq_ps(dx30,dy30,dz30);
793 rinv10 = gmx_mm256_invsqrt_ps(rsq10);
794 rinv20 = gmx_mm256_invsqrt_ps(rsq20);
795 rinv30 = gmx_mm256_invsqrt_ps(rsq30);
797 rinvsq10 = _mm256_mul_ps(rinv10,rinv10);
798 rinvsq20 = _mm256_mul_ps(rinv20,rinv20);
799 rinvsq30 = _mm256_mul_ps(rinv30,rinv30);
801 /* Load parameters for j particles */
802 jq0 = gmx_mm256_load_8real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
803 charge+jnrC+0,charge+jnrD+0,
804 charge+jnrE+0,charge+jnrF+0,
805 charge+jnrG+0,charge+jnrH+0);
807 fjx0 = _mm256_setzero_ps();
808 fjy0 = _mm256_setzero_ps();
809 fjz0 = _mm256_setzero_ps();
811 /**************************
812 * CALCULATE INTERACTIONS *
813 **************************/
815 if (gmx_mm256_any_lt(rsq10,rcutoff2))
818 /* Compute parameters for interactions between i and j atoms */
819 qq10 = _mm256_mul_ps(iq1,jq0);
821 /* REACTION-FIELD ELECTROSTATICS */
822 felec = _mm256_mul_ps(qq10,_mm256_sub_ps(_mm256_mul_ps(rinv10,rinvsq10),krf2));
824 cutoff_mask = _mm256_cmp_ps(rsq10,rcutoff2,_CMP_LT_OQ);
828 fscal = _mm256_and_ps(fscal,cutoff_mask);
830 /* Calculate temporary vectorial force */
831 tx = _mm256_mul_ps(fscal,dx10);
832 ty = _mm256_mul_ps(fscal,dy10);
833 tz = _mm256_mul_ps(fscal,dz10);
835 /* Update vectorial force */
836 fix1 = _mm256_add_ps(fix1,tx);
837 fiy1 = _mm256_add_ps(fiy1,ty);
838 fiz1 = _mm256_add_ps(fiz1,tz);
840 fjx0 = _mm256_add_ps(fjx0,tx);
841 fjy0 = _mm256_add_ps(fjy0,ty);
842 fjz0 = _mm256_add_ps(fjz0,tz);
846 /**************************
847 * CALCULATE INTERACTIONS *
848 **************************/
850 if (gmx_mm256_any_lt(rsq20,rcutoff2))
853 /* Compute parameters for interactions between i and j atoms */
854 qq20 = _mm256_mul_ps(iq2,jq0);
856 /* REACTION-FIELD ELECTROSTATICS */
857 felec = _mm256_mul_ps(qq20,_mm256_sub_ps(_mm256_mul_ps(rinv20,rinvsq20),krf2));
859 cutoff_mask = _mm256_cmp_ps(rsq20,rcutoff2,_CMP_LT_OQ);
863 fscal = _mm256_and_ps(fscal,cutoff_mask);
865 /* Calculate temporary vectorial force */
866 tx = _mm256_mul_ps(fscal,dx20);
867 ty = _mm256_mul_ps(fscal,dy20);
868 tz = _mm256_mul_ps(fscal,dz20);
870 /* Update vectorial force */
871 fix2 = _mm256_add_ps(fix2,tx);
872 fiy2 = _mm256_add_ps(fiy2,ty);
873 fiz2 = _mm256_add_ps(fiz2,tz);
875 fjx0 = _mm256_add_ps(fjx0,tx);
876 fjy0 = _mm256_add_ps(fjy0,ty);
877 fjz0 = _mm256_add_ps(fjz0,tz);
881 /**************************
882 * CALCULATE INTERACTIONS *
883 **************************/
885 if (gmx_mm256_any_lt(rsq30,rcutoff2))
888 /* Compute parameters for interactions between i and j atoms */
889 qq30 = _mm256_mul_ps(iq3,jq0);
891 /* REACTION-FIELD ELECTROSTATICS */
892 felec = _mm256_mul_ps(qq30,_mm256_sub_ps(_mm256_mul_ps(rinv30,rinvsq30),krf2));
894 cutoff_mask = _mm256_cmp_ps(rsq30,rcutoff2,_CMP_LT_OQ);
898 fscal = _mm256_and_ps(fscal,cutoff_mask);
900 /* Calculate temporary vectorial force */
901 tx = _mm256_mul_ps(fscal,dx30);
902 ty = _mm256_mul_ps(fscal,dy30);
903 tz = _mm256_mul_ps(fscal,dz30);
905 /* Update vectorial force */
906 fix3 = _mm256_add_ps(fix3,tx);
907 fiy3 = _mm256_add_ps(fiy3,ty);
908 fiz3 = _mm256_add_ps(fiz3,tz);
910 fjx0 = _mm256_add_ps(fjx0,tx);
911 fjy0 = _mm256_add_ps(fjy0,ty);
912 fjz0 = _mm256_add_ps(fjz0,tz);
916 fjptrA = f+j_coord_offsetA;
917 fjptrB = f+j_coord_offsetB;
918 fjptrC = f+j_coord_offsetC;
919 fjptrD = f+j_coord_offsetD;
920 fjptrE = f+j_coord_offsetE;
921 fjptrF = f+j_coord_offsetF;
922 fjptrG = f+j_coord_offsetG;
923 fjptrH = f+j_coord_offsetH;
925 gmx_mm256_decrement_1rvec_8ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjptrE,fjptrF,fjptrG,fjptrH,fjx0,fjy0,fjz0);
927 /* Inner loop uses 93 flops */
933 /* Get j neighbor index, and coordinate index */
934 jnrlistA = jjnr[jidx];
935 jnrlistB = jjnr[jidx+1];
936 jnrlistC = jjnr[jidx+2];
937 jnrlistD = jjnr[jidx+3];
938 jnrlistE = jjnr[jidx+4];
939 jnrlistF = jjnr[jidx+5];
940 jnrlistG = jjnr[jidx+6];
941 jnrlistH = jjnr[jidx+7];
942 /* Sign of each element will be negative for non-real atoms.
943 * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
944 * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
946 dummy_mask = gmx_mm256_set_m128(gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx+4)),_mm_setzero_si128())),
947 gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128())));
949 jnrA = (jnrlistA>=0) ? jnrlistA : 0;
950 jnrB = (jnrlistB>=0) ? jnrlistB : 0;
951 jnrC = (jnrlistC>=0) ? jnrlistC : 0;
952 jnrD = (jnrlistD>=0) ? jnrlistD : 0;
953 jnrE = (jnrlistE>=0) ? jnrlistE : 0;
954 jnrF = (jnrlistF>=0) ? jnrlistF : 0;
955 jnrG = (jnrlistG>=0) ? jnrlistG : 0;
956 jnrH = (jnrlistH>=0) ? jnrlistH : 0;
957 j_coord_offsetA = DIM*jnrA;
958 j_coord_offsetB = DIM*jnrB;
959 j_coord_offsetC = DIM*jnrC;
960 j_coord_offsetD = DIM*jnrD;
961 j_coord_offsetE = DIM*jnrE;
962 j_coord_offsetF = DIM*jnrF;
963 j_coord_offsetG = DIM*jnrG;
964 j_coord_offsetH = DIM*jnrH;
966 /* load j atom coordinates */
967 gmx_mm256_load_1rvec_8ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
968 x+j_coord_offsetC,x+j_coord_offsetD,
969 x+j_coord_offsetE,x+j_coord_offsetF,
970 x+j_coord_offsetG,x+j_coord_offsetH,
973 /* Calculate displacement vector */
974 dx10 = _mm256_sub_ps(ix1,jx0);
975 dy10 = _mm256_sub_ps(iy1,jy0);
976 dz10 = _mm256_sub_ps(iz1,jz0);
977 dx20 = _mm256_sub_ps(ix2,jx0);
978 dy20 = _mm256_sub_ps(iy2,jy0);
979 dz20 = _mm256_sub_ps(iz2,jz0);
980 dx30 = _mm256_sub_ps(ix3,jx0);
981 dy30 = _mm256_sub_ps(iy3,jy0);
982 dz30 = _mm256_sub_ps(iz3,jz0);
984 /* Calculate squared distance and things based on it */
985 rsq10 = gmx_mm256_calc_rsq_ps(dx10,dy10,dz10);
986 rsq20 = gmx_mm256_calc_rsq_ps(dx20,dy20,dz20);
987 rsq30 = gmx_mm256_calc_rsq_ps(dx30,dy30,dz30);
989 rinv10 = gmx_mm256_invsqrt_ps(rsq10);
990 rinv20 = gmx_mm256_invsqrt_ps(rsq20);
991 rinv30 = gmx_mm256_invsqrt_ps(rsq30);
993 rinvsq10 = _mm256_mul_ps(rinv10,rinv10);
994 rinvsq20 = _mm256_mul_ps(rinv20,rinv20);
995 rinvsq30 = _mm256_mul_ps(rinv30,rinv30);
997 /* Load parameters for j particles */
998 jq0 = gmx_mm256_load_8real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
999 charge+jnrC+0,charge+jnrD+0,
1000 charge+jnrE+0,charge+jnrF+0,
1001 charge+jnrG+0,charge+jnrH+0);
1003 fjx0 = _mm256_setzero_ps();
1004 fjy0 = _mm256_setzero_ps();
1005 fjz0 = _mm256_setzero_ps();
1007 /**************************
1008 * CALCULATE INTERACTIONS *
1009 **************************/
1011 if (gmx_mm256_any_lt(rsq10,rcutoff2))
1014 /* Compute parameters for interactions between i and j atoms */
1015 qq10 = _mm256_mul_ps(iq1,jq0);
1017 /* REACTION-FIELD ELECTROSTATICS */
1018 felec = _mm256_mul_ps(qq10,_mm256_sub_ps(_mm256_mul_ps(rinv10,rinvsq10),krf2));
1020 cutoff_mask = _mm256_cmp_ps(rsq10,rcutoff2,_CMP_LT_OQ);
1024 fscal = _mm256_and_ps(fscal,cutoff_mask);
1026 fscal = _mm256_andnot_ps(dummy_mask,fscal);
1028 /* Calculate temporary vectorial force */
1029 tx = _mm256_mul_ps(fscal,dx10);
1030 ty = _mm256_mul_ps(fscal,dy10);
1031 tz = _mm256_mul_ps(fscal,dz10);
1033 /* Update vectorial force */
1034 fix1 = _mm256_add_ps(fix1,tx);
1035 fiy1 = _mm256_add_ps(fiy1,ty);
1036 fiz1 = _mm256_add_ps(fiz1,tz);
1038 fjx0 = _mm256_add_ps(fjx0,tx);
1039 fjy0 = _mm256_add_ps(fjy0,ty);
1040 fjz0 = _mm256_add_ps(fjz0,tz);
1044 /**************************
1045 * CALCULATE INTERACTIONS *
1046 **************************/
1048 if (gmx_mm256_any_lt(rsq20,rcutoff2))
1051 /* Compute parameters for interactions between i and j atoms */
1052 qq20 = _mm256_mul_ps(iq2,jq0);
1054 /* REACTION-FIELD ELECTROSTATICS */
1055 felec = _mm256_mul_ps(qq20,_mm256_sub_ps(_mm256_mul_ps(rinv20,rinvsq20),krf2));
1057 cutoff_mask = _mm256_cmp_ps(rsq20,rcutoff2,_CMP_LT_OQ);
1061 fscal = _mm256_and_ps(fscal,cutoff_mask);
1063 fscal = _mm256_andnot_ps(dummy_mask,fscal);
1065 /* Calculate temporary vectorial force */
1066 tx = _mm256_mul_ps(fscal,dx20);
1067 ty = _mm256_mul_ps(fscal,dy20);
1068 tz = _mm256_mul_ps(fscal,dz20);
1070 /* Update vectorial force */
1071 fix2 = _mm256_add_ps(fix2,tx);
1072 fiy2 = _mm256_add_ps(fiy2,ty);
1073 fiz2 = _mm256_add_ps(fiz2,tz);
1075 fjx0 = _mm256_add_ps(fjx0,tx);
1076 fjy0 = _mm256_add_ps(fjy0,ty);
1077 fjz0 = _mm256_add_ps(fjz0,tz);
1081 /**************************
1082 * CALCULATE INTERACTIONS *
1083 **************************/
1085 if (gmx_mm256_any_lt(rsq30,rcutoff2))
1088 /* Compute parameters for interactions between i and j atoms */
1089 qq30 = _mm256_mul_ps(iq3,jq0);
1091 /* REACTION-FIELD ELECTROSTATICS */
1092 felec = _mm256_mul_ps(qq30,_mm256_sub_ps(_mm256_mul_ps(rinv30,rinvsq30),krf2));
1094 cutoff_mask = _mm256_cmp_ps(rsq30,rcutoff2,_CMP_LT_OQ);
1098 fscal = _mm256_and_ps(fscal,cutoff_mask);
1100 fscal = _mm256_andnot_ps(dummy_mask,fscal);
1102 /* Calculate temporary vectorial force */
1103 tx = _mm256_mul_ps(fscal,dx30);
1104 ty = _mm256_mul_ps(fscal,dy30);
1105 tz = _mm256_mul_ps(fscal,dz30);
1107 /* Update vectorial force */
1108 fix3 = _mm256_add_ps(fix3,tx);
1109 fiy3 = _mm256_add_ps(fiy3,ty);
1110 fiz3 = _mm256_add_ps(fiz3,tz);
1112 fjx0 = _mm256_add_ps(fjx0,tx);
1113 fjy0 = _mm256_add_ps(fjy0,ty);
1114 fjz0 = _mm256_add_ps(fjz0,tz);
1118 fjptrA = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
1119 fjptrB = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
1120 fjptrC = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
1121 fjptrD = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
1122 fjptrE = (jnrlistE>=0) ? f+j_coord_offsetE : scratch;
1123 fjptrF = (jnrlistF>=0) ? f+j_coord_offsetF : scratch;
1124 fjptrG = (jnrlistG>=0) ? f+j_coord_offsetG : scratch;
1125 fjptrH = (jnrlistH>=0) ? f+j_coord_offsetH : scratch;
1127 gmx_mm256_decrement_1rvec_8ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjptrE,fjptrF,fjptrG,fjptrH,fjx0,fjy0,fjz0);
1129 /* Inner loop uses 93 flops */
1132 /* End of innermost loop */
1134 gmx_mm256_update_iforce_3atom_swizzle_ps(fix1,fiy1,fiz1,fix2,fiy2,fiz2,fix3,fiy3,fiz3,
1135 f+i_coord_offset+DIM,fshift+i_shift_offset);
1137 /* Increment number of inner iterations */
1138 inneriter += j_index_end - j_index_start;
1140 /* Outer loop uses 18 flops */
1143 /* Increment number of outer iterations */
1146 /* Update outer/inner flops */
1148 inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_W4_F,outeriter*18 + inneriter*93);