2 * Note: this file was generated by the Gromacs avx_256_double kernel generator.
4 * This source code is part of
8 * Copyright (c) 2001-2012, The GROMACS Development Team
10 * Gromacs is a library for molecular simulation and trajectory analysis,
11 * written by Erik Lindahl, David van der Spoel, Berk Hess, and others - for
12 * a full list of developers and information, check out http://www.gromacs.org
14 * This program is free software; you can redistribute it and/or modify it under
15 * the terms of the GNU Lesser General Public License as published by the Free
16 * Software Foundation; either version 2 of the License, or (at your option) any
19 * To help fund GROMACS development, we humbly ask that you cite
20 * the papers people have written on it - you can find them on the website.
28 #include "../nb_kernel.h"
29 #include "types/simple.h"
33 #include "gmx_math_x86_avx_256_double.h"
34 #include "kernelutil_x86_avx_256_double.h"
37 * Gromacs nonbonded kernel: nb_kernel_ElecRFCut_VdwLJSh_GeomW4P1_VF_avx_256_double
38 * Electrostatics interaction: ReactionField
39 * VdW interaction: LennardJones
40 * Geometry: Water4-Particle
41 * Calculate force/pot: PotentialAndForce
44 nb_kernel_ElecRFCut_VdwLJSh_GeomW4P1_VF_avx_256_double
45 (t_nblist * gmx_restrict nlist,
46 rvec * gmx_restrict xx,
47 rvec * gmx_restrict ff,
48 t_forcerec * gmx_restrict fr,
49 t_mdatoms * gmx_restrict mdatoms,
50 nb_kernel_data_t * gmx_restrict kernel_data,
51 t_nrnb * gmx_restrict nrnb)
53 /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
54 * just 0 for non-waters.
55 * Suffixes A,B,C,D refer to j loop unrolling done with AVX, e.g. for the four different
56 * jnr indices corresponding to data put in the four positions in the SIMD register.
58 int i_shift_offset,i_coord_offset,outeriter,inneriter;
59 int j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
60 int jnrA,jnrB,jnrC,jnrD;
61 int jnrlistA,jnrlistB,jnrlistC,jnrlistD;
62 int jnrlistE,jnrlistF,jnrlistG,jnrlistH;
63 int j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
64 int *iinr,*jindex,*jjnr,*shiftidx,*gid;
66 real *shiftvec,*fshift,*x,*f;
67 real *fjptrA,*fjptrB,*fjptrC,*fjptrD;
69 __m256d tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
70 real * vdwioffsetptr0;
71 __m256d ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
72 real * vdwioffsetptr1;
73 __m256d ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
74 real * vdwioffsetptr2;
75 __m256d ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
76 real * vdwioffsetptr3;
77 __m256d ix3,iy3,iz3,fix3,fiy3,fiz3,iq3,isai3;
78 int vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
79 __m256d jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
80 __m256d dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
81 __m256d dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
82 __m256d dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
83 __m256d dx30,dy30,dz30,rsq30,rinv30,rinvsq30,r30,qq30,c6_30,c12_30;
84 __m256d velec,felec,velecsum,facel,crf,krf,krf2;
87 __m256d rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
90 __m256d one_sixth = _mm256_set1_pd(1.0/6.0);
91 __m256d one_twelfth = _mm256_set1_pd(1.0/12.0);
92 __m256d dummy_mask,cutoff_mask;
93 __m128 tmpmask0,tmpmask1;
94 __m256d signbit = _mm256_castsi256_pd( _mm256_set1_epi32(0x80000000) );
95 __m256d one = _mm256_set1_pd(1.0);
96 __m256d two = _mm256_set1_pd(2.0);
102 jindex = nlist->jindex;
104 shiftidx = nlist->shift;
106 shiftvec = fr->shift_vec[0];
107 fshift = fr->fshift[0];
108 facel = _mm256_set1_pd(fr->epsfac);
109 charge = mdatoms->chargeA;
110 krf = _mm256_set1_pd(fr->ic->k_rf);
111 krf2 = _mm256_set1_pd(fr->ic->k_rf*2.0);
112 crf = _mm256_set1_pd(fr->ic->c_rf);
113 nvdwtype = fr->ntype;
115 vdwtype = mdatoms->typeA;
117 /* Setup water-specific parameters */
118 inr = nlist->iinr[0];
119 iq1 = _mm256_mul_pd(facel,_mm256_set1_pd(charge[inr+1]));
120 iq2 = _mm256_mul_pd(facel,_mm256_set1_pd(charge[inr+2]));
121 iq3 = _mm256_mul_pd(facel,_mm256_set1_pd(charge[inr+3]));
122 vdwioffsetptr0 = vdwparam+2*nvdwtype*vdwtype[inr+0];
124 /* When we use explicit cutoffs the value must be identical for elec and VdW, so use elec as an arbitrary choice */
125 rcutoff_scalar = fr->rcoulomb;
126 rcutoff = _mm256_set1_pd(rcutoff_scalar);
127 rcutoff2 = _mm256_mul_pd(rcutoff,rcutoff);
129 sh_vdw_invrcut6 = _mm256_set1_pd(fr->ic->sh_invrc6);
130 rvdw = _mm256_set1_pd(fr->rvdw);
132 /* Avoid stupid compiler warnings */
133 jnrA = jnrB = jnrC = jnrD = 0;
142 for(iidx=0;iidx<4*DIM;iidx++)
147 /* Start outer loop over neighborlists */
148 for(iidx=0; iidx<nri; iidx++)
150 /* Load shift vector for this list */
151 i_shift_offset = DIM*shiftidx[iidx];
153 /* Load limits for loop over neighbors */
154 j_index_start = jindex[iidx];
155 j_index_end = jindex[iidx+1];
157 /* Get outer coordinate index */
159 i_coord_offset = DIM*inr;
161 /* Load i particle coords and add shift vector */
162 gmx_mm256_load_shift_and_4rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset,
163 &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2,&ix3,&iy3,&iz3);
165 fix0 = _mm256_setzero_pd();
166 fiy0 = _mm256_setzero_pd();
167 fiz0 = _mm256_setzero_pd();
168 fix1 = _mm256_setzero_pd();
169 fiy1 = _mm256_setzero_pd();
170 fiz1 = _mm256_setzero_pd();
171 fix2 = _mm256_setzero_pd();
172 fiy2 = _mm256_setzero_pd();
173 fiz2 = _mm256_setzero_pd();
174 fix3 = _mm256_setzero_pd();
175 fiy3 = _mm256_setzero_pd();
176 fiz3 = _mm256_setzero_pd();
178 /* Reset potential sums */
179 velecsum = _mm256_setzero_pd();
180 vvdwsum = _mm256_setzero_pd();
182 /* Start inner kernel loop */
183 for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
186 /* Get j neighbor index, and coordinate index */
191 j_coord_offsetA = DIM*jnrA;
192 j_coord_offsetB = DIM*jnrB;
193 j_coord_offsetC = DIM*jnrC;
194 j_coord_offsetD = DIM*jnrD;
196 /* load j atom coordinates */
197 gmx_mm256_load_1rvec_4ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
198 x+j_coord_offsetC,x+j_coord_offsetD,
201 /* Calculate displacement vector */
202 dx00 = _mm256_sub_pd(ix0,jx0);
203 dy00 = _mm256_sub_pd(iy0,jy0);
204 dz00 = _mm256_sub_pd(iz0,jz0);
205 dx10 = _mm256_sub_pd(ix1,jx0);
206 dy10 = _mm256_sub_pd(iy1,jy0);
207 dz10 = _mm256_sub_pd(iz1,jz0);
208 dx20 = _mm256_sub_pd(ix2,jx0);
209 dy20 = _mm256_sub_pd(iy2,jy0);
210 dz20 = _mm256_sub_pd(iz2,jz0);
211 dx30 = _mm256_sub_pd(ix3,jx0);
212 dy30 = _mm256_sub_pd(iy3,jy0);
213 dz30 = _mm256_sub_pd(iz3,jz0);
215 /* Calculate squared distance and things based on it */
216 rsq00 = gmx_mm256_calc_rsq_pd(dx00,dy00,dz00);
217 rsq10 = gmx_mm256_calc_rsq_pd(dx10,dy10,dz10);
218 rsq20 = gmx_mm256_calc_rsq_pd(dx20,dy20,dz20);
219 rsq30 = gmx_mm256_calc_rsq_pd(dx30,dy30,dz30);
221 rinv10 = gmx_mm256_invsqrt_pd(rsq10);
222 rinv20 = gmx_mm256_invsqrt_pd(rsq20);
223 rinv30 = gmx_mm256_invsqrt_pd(rsq30);
225 rinvsq00 = gmx_mm256_inv_pd(rsq00);
226 rinvsq10 = _mm256_mul_pd(rinv10,rinv10);
227 rinvsq20 = _mm256_mul_pd(rinv20,rinv20);
228 rinvsq30 = _mm256_mul_pd(rinv30,rinv30);
230 /* Load parameters for j particles */
231 jq0 = gmx_mm256_load_4real_swizzle_pd(charge+jnrA+0,charge+jnrB+0,
232 charge+jnrC+0,charge+jnrD+0);
233 vdwjidx0A = 2*vdwtype[jnrA+0];
234 vdwjidx0B = 2*vdwtype[jnrB+0];
235 vdwjidx0C = 2*vdwtype[jnrC+0];
236 vdwjidx0D = 2*vdwtype[jnrD+0];
238 fjx0 = _mm256_setzero_pd();
239 fjy0 = _mm256_setzero_pd();
240 fjz0 = _mm256_setzero_pd();
242 /**************************
243 * CALCULATE INTERACTIONS *
244 **************************/
246 if (gmx_mm256_any_lt(rsq00,rcutoff2))
249 /* Compute parameters for interactions between i and j atoms */
250 gmx_mm256_load_4pair_swizzle_pd(vdwioffsetptr0+vdwjidx0A,
251 vdwioffsetptr0+vdwjidx0B,
252 vdwioffsetptr0+vdwjidx0C,
253 vdwioffsetptr0+vdwjidx0D,
256 /* LENNARD-JONES DISPERSION/REPULSION */
258 rinvsix = _mm256_mul_pd(_mm256_mul_pd(rinvsq00,rinvsq00),rinvsq00);
259 vvdw6 = _mm256_mul_pd(c6_00,rinvsix);
260 vvdw12 = _mm256_mul_pd(c12_00,_mm256_mul_pd(rinvsix,rinvsix));
261 vvdw = _mm256_sub_pd(_mm256_mul_pd( _mm256_sub_pd(vvdw12 , _mm256_mul_pd(c12_00,_mm256_mul_pd(sh_vdw_invrcut6,sh_vdw_invrcut6))), one_twelfth) ,
262 _mm256_mul_pd( _mm256_sub_pd(vvdw6,_mm256_mul_pd(c6_00,sh_vdw_invrcut6)),one_sixth));
263 fvdw = _mm256_mul_pd(_mm256_sub_pd(vvdw12,vvdw6),rinvsq00);
265 cutoff_mask = _mm256_cmp_pd(rsq00,rcutoff2,_CMP_LT_OQ);
267 /* Update potential sum for this i atom from the interaction with this j atom. */
268 vvdw = _mm256_and_pd(vvdw,cutoff_mask);
269 vvdwsum = _mm256_add_pd(vvdwsum,vvdw);
273 fscal = _mm256_and_pd(fscal,cutoff_mask);
275 /* Calculate temporary vectorial force */
276 tx = _mm256_mul_pd(fscal,dx00);
277 ty = _mm256_mul_pd(fscal,dy00);
278 tz = _mm256_mul_pd(fscal,dz00);
280 /* Update vectorial force */
281 fix0 = _mm256_add_pd(fix0,tx);
282 fiy0 = _mm256_add_pd(fiy0,ty);
283 fiz0 = _mm256_add_pd(fiz0,tz);
285 fjx0 = _mm256_add_pd(fjx0,tx);
286 fjy0 = _mm256_add_pd(fjy0,ty);
287 fjz0 = _mm256_add_pd(fjz0,tz);
291 /**************************
292 * CALCULATE INTERACTIONS *
293 **************************/
295 if (gmx_mm256_any_lt(rsq10,rcutoff2))
298 /* Compute parameters for interactions between i and j atoms */
299 qq10 = _mm256_mul_pd(iq1,jq0);
301 /* REACTION-FIELD ELECTROSTATICS */
302 velec = _mm256_mul_pd(qq10,_mm256_sub_pd(_mm256_add_pd(rinv10,_mm256_mul_pd(krf,rsq10)),crf));
303 felec = _mm256_mul_pd(qq10,_mm256_sub_pd(_mm256_mul_pd(rinv10,rinvsq10),krf2));
305 cutoff_mask = _mm256_cmp_pd(rsq10,rcutoff2,_CMP_LT_OQ);
307 /* Update potential sum for this i atom from the interaction with this j atom. */
308 velec = _mm256_and_pd(velec,cutoff_mask);
309 velecsum = _mm256_add_pd(velecsum,velec);
313 fscal = _mm256_and_pd(fscal,cutoff_mask);
315 /* Calculate temporary vectorial force */
316 tx = _mm256_mul_pd(fscal,dx10);
317 ty = _mm256_mul_pd(fscal,dy10);
318 tz = _mm256_mul_pd(fscal,dz10);
320 /* Update vectorial force */
321 fix1 = _mm256_add_pd(fix1,tx);
322 fiy1 = _mm256_add_pd(fiy1,ty);
323 fiz1 = _mm256_add_pd(fiz1,tz);
325 fjx0 = _mm256_add_pd(fjx0,tx);
326 fjy0 = _mm256_add_pd(fjy0,ty);
327 fjz0 = _mm256_add_pd(fjz0,tz);
331 /**************************
332 * CALCULATE INTERACTIONS *
333 **************************/
335 if (gmx_mm256_any_lt(rsq20,rcutoff2))
338 /* Compute parameters for interactions between i and j atoms */
339 qq20 = _mm256_mul_pd(iq2,jq0);
341 /* REACTION-FIELD ELECTROSTATICS */
342 velec = _mm256_mul_pd(qq20,_mm256_sub_pd(_mm256_add_pd(rinv20,_mm256_mul_pd(krf,rsq20)),crf));
343 felec = _mm256_mul_pd(qq20,_mm256_sub_pd(_mm256_mul_pd(rinv20,rinvsq20),krf2));
345 cutoff_mask = _mm256_cmp_pd(rsq20,rcutoff2,_CMP_LT_OQ);
347 /* Update potential sum for this i atom from the interaction with this j atom. */
348 velec = _mm256_and_pd(velec,cutoff_mask);
349 velecsum = _mm256_add_pd(velecsum,velec);
353 fscal = _mm256_and_pd(fscal,cutoff_mask);
355 /* Calculate temporary vectorial force */
356 tx = _mm256_mul_pd(fscal,dx20);
357 ty = _mm256_mul_pd(fscal,dy20);
358 tz = _mm256_mul_pd(fscal,dz20);
360 /* Update vectorial force */
361 fix2 = _mm256_add_pd(fix2,tx);
362 fiy2 = _mm256_add_pd(fiy2,ty);
363 fiz2 = _mm256_add_pd(fiz2,tz);
365 fjx0 = _mm256_add_pd(fjx0,tx);
366 fjy0 = _mm256_add_pd(fjy0,ty);
367 fjz0 = _mm256_add_pd(fjz0,tz);
371 /**************************
372 * CALCULATE INTERACTIONS *
373 **************************/
375 if (gmx_mm256_any_lt(rsq30,rcutoff2))
378 /* Compute parameters for interactions between i and j atoms */
379 qq30 = _mm256_mul_pd(iq3,jq0);
381 /* REACTION-FIELD ELECTROSTATICS */
382 velec = _mm256_mul_pd(qq30,_mm256_sub_pd(_mm256_add_pd(rinv30,_mm256_mul_pd(krf,rsq30)),crf));
383 felec = _mm256_mul_pd(qq30,_mm256_sub_pd(_mm256_mul_pd(rinv30,rinvsq30),krf2));
385 cutoff_mask = _mm256_cmp_pd(rsq30,rcutoff2,_CMP_LT_OQ);
387 /* Update potential sum for this i atom from the interaction with this j atom. */
388 velec = _mm256_and_pd(velec,cutoff_mask);
389 velecsum = _mm256_add_pd(velecsum,velec);
393 fscal = _mm256_and_pd(fscal,cutoff_mask);
395 /* Calculate temporary vectorial force */
396 tx = _mm256_mul_pd(fscal,dx30);
397 ty = _mm256_mul_pd(fscal,dy30);
398 tz = _mm256_mul_pd(fscal,dz30);
400 /* Update vectorial force */
401 fix3 = _mm256_add_pd(fix3,tx);
402 fiy3 = _mm256_add_pd(fiy3,ty);
403 fiz3 = _mm256_add_pd(fiz3,tz);
405 fjx0 = _mm256_add_pd(fjx0,tx);
406 fjy0 = _mm256_add_pd(fjy0,ty);
407 fjz0 = _mm256_add_pd(fjz0,tz);
411 fjptrA = f+j_coord_offsetA;
412 fjptrB = f+j_coord_offsetB;
413 fjptrC = f+j_coord_offsetC;
414 fjptrD = f+j_coord_offsetD;
416 gmx_mm256_decrement_1rvec_4ptr_swizzle_pd(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
418 /* Inner loop uses 152 flops */
424 /* Get j neighbor index, and coordinate index */
425 jnrlistA = jjnr[jidx];
426 jnrlistB = jjnr[jidx+1];
427 jnrlistC = jjnr[jidx+2];
428 jnrlistD = jjnr[jidx+3];
429 /* Sign of each element will be negative for non-real atoms.
430 * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
431 * so use it as val = _mm_andnot_pd(mask,val) to clear dummy entries.
433 tmpmask0 = gmx_mm_castsi128_pd(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
435 tmpmask1 = _mm_permute_ps(tmpmask0,_GMX_MM_PERMUTE(3,3,2,2));
436 tmpmask0 = _mm_permute_ps(tmpmask0,_GMX_MM_PERMUTE(1,1,0,0));
437 dummy_mask = _mm256_castps_pd(gmx_mm256_set_m128(tmpmask1,tmpmask0));
439 jnrA = (jnrlistA>=0) ? jnrlistA : 0;
440 jnrB = (jnrlistB>=0) ? jnrlistB : 0;
441 jnrC = (jnrlistC>=0) ? jnrlistC : 0;
442 jnrD = (jnrlistD>=0) ? jnrlistD : 0;
443 j_coord_offsetA = DIM*jnrA;
444 j_coord_offsetB = DIM*jnrB;
445 j_coord_offsetC = DIM*jnrC;
446 j_coord_offsetD = DIM*jnrD;
448 /* load j atom coordinates */
449 gmx_mm256_load_1rvec_4ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
450 x+j_coord_offsetC,x+j_coord_offsetD,
453 /* Calculate displacement vector */
454 dx00 = _mm256_sub_pd(ix0,jx0);
455 dy00 = _mm256_sub_pd(iy0,jy0);
456 dz00 = _mm256_sub_pd(iz0,jz0);
457 dx10 = _mm256_sub_pd(ix1,jx0);
458 dy10 = _mm256_sub_pd(iy1,jy0);
459 dz10 = _mm256_sub_pd(iz1,jz0);
460 dx20 = _mm256_sub_pd(ix2,jx0);
461 dy20 = _mm256_sub_pd(iy2,jy0);
462 dz20 = _mm256_sub_pd(iz2,jz0);
463 dx30 = _mm256_sub_pd(ix3,jx0);
464 dy30 = _mm256_sub_pd(iy3,jy0);
465 dz30 = _mm256_sub_pd(iz3,jz0);
467 /* Calculate squared distance and things based on it */
468 rsq00 = gmx_mm256_calc_rsq_pd(dx00,dy00,dz00);
469 rsq10 = gmx_mm256_calc_rsq_pd(dx10,dy10,dz10);
470 rsq20 = gmx_mm256_calc_rsq_pd(dx20,dy20,dz20);
471 rsq30 = gmx_mm256_calc_rsq_pd(dx30,dy30,dz30);
473 rinv10 = gmx_mm256_invsqrt_pd(rsq10);
474 rinv20 = gmx_mm256_invsqrt_pd(rsq20);
475 rinv30 = gmx_mm256_invsqrt_pd(rsq30);
477 rinvsq00 = gmx_mm256_inv_pd(rsq00);
478 rinvsq10 = _mm256_mul_pd(rinv10,rinv10);
479 rinvsq20 = _mm256_mul_pd(rinv20,rinv20);
480 rinvsq30 = _mm256_mul_pd(rinv30,rinv30);
482 /* Load parameters for j particles */
483 jq0 = gmx_mm256_load_4real_swizzle_pd(charge+jnrA+0,charge+jnrB+0,
484 charge+jnrC+0,charge+jnrD+0);
485 vdwjidx0A = 2*vdwtype[jnrA+0];
486 vdwjidx0B = 2*vdwtype[jnrB+0];
487 vdwjidx0C = 2*vdwtype[jnrC+0];
488 vdwjidx0D = 2*vdwtype[jnrD+0];
490 fjx0 = _mm256_setzero_pd();
491 fjy0 = _mm256_setzero_pd();
492 fjz0 = _mm256_setzero_pd();
494 /**************************
495 * CALCULATE INTERACTIONS *
496 **************************/
498 if (gmx_mm256_any_lt(rsq00,rcutoff2))
501 /* Compute parameters for interactions between i and j atoms */
502 gmx_mm256_load_4pair_swizzle_pd(vdwioffsetptr0+vdwjidx0A,
503 vdwioffsetptr0+vdwjidx0B,
504 vdwioffsetptr0+vdwjidx0C,
505 vdwioffsetptr0+vdwjidx0D,
508 /* LENNARD-JONES DISPERSION/REPULSION */
510 rinvsix = _mm256_mul_pd(_mm256_mul_pd(rinvsq00,rinvsq00),rinvsq00);
511 vvdw6 = _mm256_mul_pd(c6_00,rinvsix);
512 vvdw12 = _mm256_mul_pd(c12_00,_mm256_mul_pd(rinvsix,rinvsix));
513 vvdw = _mm256_sub_pd(_mm256_mul_pd( _mm256_sub_pd(vvdw12 , _mm256_mul_pd(c12_00,_mm256_mul_pd(sh_vdw_invrcut6,sh_vdw_invrcut6))), one_twelfth) ,
514 _mm256_mul_pd( _mm256_sub_pd(vvdw6,_mm256_mul_pd(c6_00,sh_vdw_invrcut6)),one_sixth));
515 fvdw = _mm256_mul_pd(_mm256_sub_pd(vvdw12,vvdw6),rinvsq00);
517 cutoff_mask = _mm256_cmp_pd(rsq00,rcutoff2,_CMP_LT_OQ);
519 /* Update potential sum for this i atom from the interaction with this j atom. */
520 vvdw = _mm256_and_pd(vvdw,cutoff_mask);
521 vvdw = _mm256_andnot_pd(dummy_mask,vvdw);
522 vvdwsum = _mm256_add_pd(vvdwsum,vvdw);
526 fscal = _mm256_and_pd(fscal,cutoff_mask);
528 fscal = _mm256_andnot_pd(dummy_mask,fscal);
530 /* Calculate temporary vectorial force */
531 tx = _mm256_mul_pd(fscal,dx00);
532 ty = _mm256_mul_pd(fscal,dy00);
533 tz = _mm256_mul_pd(fscal,dz00);
535 /* Update vectorial force */
536 fix0 = _mm256_add_pd(fix0,tx);
537 fiy0 = _mm256_add_pd(fiy0,ty);
538 fiz0 = _mm256_add_pd(fiz0,tz);
540 fjx0 = _mm256_add_pd(fjx0,tx);
541 fjy0 = _mm256_add_pd(fjy0,ty);
542 fjz0 = _mm256_add_pd(fjz0,tz);
546 /**************************
547 * CALCULATE INTERACTIONS *
548 **************************/
550 if (gmx_mm256_any_lt(rsq10,rcutoff2))
553 /* Compute parameters for interactions between i and j atoms */
554 qq10 = _mm256_mul_pd(iq1,jq0);
556 /* REACTION-FIELD ELECTROSTATICS */
557 velec = _mm256_mul_pd(qq10,_mm256_sub_pd(_mm256_add_pd(rinv10,_mm256_mul_pd(krf,rsq10)),crf));
558 felec = _mm256_mul_pd(qq10,_mm256_sub_pd(_mm256_mul_pd(rinv10,rinvsq10),krf2));
560 cutoff_mask = _mm256_cmp_pd(rsq10,rcutoff2,_CMP_LT_OQ);
562 /* Update potential sum for this i atom from the interaction with this j atom. */
563 velec = _mm256_and_pd(velec,cutoff_mask);
564 velec = _mm256_andnot_pd(dummy_mask,velec);
565 velecsum = _mm256_add_pd(velecsum,velec);
569 fscal = _mm256_and_pd(fscal,cutoff_mask);
571 fscal = _mm256_andnot_pd(dummy_mask,fscal);
573 /* Calculate temporary vectorial force */
574 tx = _mm256_mul_pd(fscal,dx10);
575 ty = _mm256_mul_pd(fscal,dy10);
576 tz = _mm256_mul_pd(fscal,dz10);
578 /* Update vectorial force */
579 fix1 = _mm256_add_pd(fix1,tx);
580 fiy1 = _mm256_add_pd(fiy1,ty);
581 fiz1 = _mm256_add_pd(fiz1,tz);
583 fjx0 = _mm256_add_pd(fjx0,tx);
584 fjy0 = _mm256_add_pd(fjy0,ty);
585 fjz0 = _mm256_add_pd(fjz0,tz);
589 /**************************
590 * CALCULATE INTERACTIONS *
591 **************************/
593 if (gmx_mm256_any_lt(rsq20,rcutoff2))
596 /* Compute parameters for interactions between i and j atoms */
597 qq20 = _mm256_mul_pd(iq2,jq0);
599 /* REACTION-FIELD ELECTROSTATICS */
600 velec = _mm256_mul_pd(qq20,_mm256_sub_pd(_mm256_add_pd(rinv20,_mm256_mul_pd(krf,rsq20)),crf));
601 felec = _mm256_mul_pd(qq20,_mm256_sub_pd(_mm256_mul_pd(rinv20,rinvsq20),krf2));
603 cutoff_mask = _mm256_cmp_pd(rsq20,rcutoff2,_CMP_LT_OQ);
605 /* Update potential sum for this i atom from the interaction with this j atom. */
606 velec = _mm256_and_pd(velec,cutoff_mask);
607 velec = _mm256_andnot_pd(dummy_mask,velec);
608 velecsum = _mm256_add_pd(velecsum,velec);
612 fscal = _mm256_and_pd(fscal,cutoff_mask);
614 fscal = _mm256_andnot_pd(dummy_mask,fscal);
616 /* Calculate temporary vectorial force */
617 tx = _mm256_mul_pd(fscal,dx20);
618 ty = _mm256_mul_pd(fscal,dy20);
619 tz = _mm256_mul_pd(fscal,dz20);
621 /* Update vectorial force */
622 fix2 = _mm256_add_pd(fix2,tx);
623 fiy2 = _mm256_add_pd(fiy2,ty);
624 fiz2 = _mm256_add_pd(fiz2,tz);
626 fjx0 = _mm256_add_pd(fjx0,tx);
627 fjy0 = _mm256_add_pd(fjy0,ty);
628 fjz0 = _mm256_add_pd(fjz0,tz);
632 /**************************
633 * CALCULATE INTERACTIONS *
634 **************************/
636 if (gmx_mm256_any_lt(rsq30,rcutoff2))
639 /* Compute parameters for interactions between i and j atoms */
640 qq30 = _mm256_mul_pd(iq3,jq0);
642 /* REACTION-FIELD ELECTROSTATICS */
643 velec = _mm256_mul_pd(qq30,_mm256_sub_pd(_mm256_add_pd(rinv30,_mm256_mul_pd(krf,rsq30)),crf));
644 felec = _mm256_mul_pd(qq30,_mm256_sub_pd(_mm256_mul_pd(rinv30,rinvsq30),krf2));
646 cutoff_mask = _mm256_cmp_pd(rsq30,rcutoff2,_CMP_LT_OQ);
648 /* Update potential sum for this i atom from the interaction with this j atom. */
649 velec = _mm256_and_pd(velec,cutoff_mask);
650 velec = _mm256_andnot_pd(dummy_mask,velec);
651 velecsum = _mm256_add_pd(velecsum,velec);
655 fscal = _mm256_and_pd(fscal,cutoff_mask);
657 fscal = _mm256_andnot_pd(dummy_mask,fscal);
659 /* Calculate temporary vectorial force */
660 tx = _mm256_mul_pd(fscal,dx30);
661 ty = _mm256_mul_pd(fscal,dy30);
662 tz = _mm256_mul_pd(fscal,dz30);
664 /* Update vectorial force */
665 fix3 = _mm256_add_pd(fix3,tx);
666 fiy3 = _mm256_add_pd(fiy3,ty);
667 fiz3 = _mm256_add_pd(fiz3,tz);
669 fjx0 = _mm256_add_pd(fjx0,tx);
670 fjy0 = _mm256_add_pd(fjy0,ty);
671 fjz0 = _mm256_add_pd(fjz0,tz);
675 fjptrA = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
676 fjptrB = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
677 fjptrC = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
678 fjptrD = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
680 gmx_mm256_decrement_1rvec_4ptr_swizzle_pd(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
682 /* Inner loop uses 152 flops */
685 /* End of innermost loop */
687 gmx_mm256_update_iforce_4atom_swizzle_pd(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,fix3,fiy3,fiz3,
688 f+i_coord_offset,fshift+i_shift_offset);
691 /* Update potential energies */
692 gmx_mm256_update_1pot_pd(velecsum,kernel_data->energygrp_elec+ggid);
693 gmx_mm256_update_1pot_pd(vvdwsum,kernel_data->energygrp_vdw+ggid);
695 /* Increment number of inner iterations */
696 inneriter += j_index_end - j_index_start;
698 /* Outer loop uses 26 flops */
701 /* Increment number of outer iterations */
704 /* Update outer/inner flops */
706 inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W4_VF,outeriter*26 + inneriter*152);
709 * Gromacs nonbonded kernel: nb_kernel_ElecRFCut_VdwLJSh_GeomW4P1_F_avx_256_double
710 * Electrostatics interaction: ReactionField
711 * VdW interaction: LennardJones
712 * Geometry: Water4-Particle
713 * Calculate force/pot: Force
716 nb_kernel_ElecRFCut_VdwLJSh_GeomW4P1_F_avx_256_double
717 (t_nblist * gmx_restrict nlist,
718 rvec * gmx_restrict xx,
719 rvec * gmx_restrict ff,
720 t_forcerec * gmx_restrict fr,
721 t_mdatoms * gmx_restrict mdatoms,
722 nb_kernel_data_t * gmx_restrict kernel_data,
723 t_nrnb * gmx_restrict nrnb)
725 /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
726 * just 0 for non-waters.
727 * Suffixes A,B,C,D refer to j loop unrolling done with AVX, e.g. for the four different
728 * jnr indices corresponding to data put in the four positions in the SIMD register.
730 int i_shift_offset,i_coord_offset,outeriter,inneriter;
731 int j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
732 int jnrA,jnrB,jnrC,jnrD;
733 int jnrlistA,jnrlistB,jnrlistC,jnrlistD;
734 int jnrlistE,jnrlistF,jnrlistG,jnrlistH;
735 int j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
736 int *iinr,*jindex,*jjnr,*shiftidx,*gid;
738 real *shiftvec,*fshift,*x,*f;
739 real *fjptrA,*fjptrB,*fjptrC,*fjptrD;
741 __m256d tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
742 real * vdwioffsetptr0;
743 __m256d ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
744 real * vdwioffsetptr1;
745 __m256d ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
746 real * vdwioffsetptr2;
747 __m256d ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
748 real * vdwioffsetptr3;
749 __m256d ix3,iy3,iz3,fix3,fiy3,fiz3,iq3,isai3;
750 int vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
751 __m256d jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
752 __m256d dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
753 __m256d dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
754 __m256d dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
755 __m256d dx30,dy30,dz30,rsq30,rinv30,rinvsq30,r30,qq30,c6_30,c12_30;
756 __m256d velec,felec,velecsum,facel,crf,krf,krf2;
759 __m256d rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
762 __m256d one_sixth = _mm256_set1_pd(1.0/6.0);
763 __m256d one_twelfth = _mm256_set1_pd(1.0/12.0);
764 __m256d dummy_mask,cutoff_mask;
765 __m128 tmpmask0,tmpmask1;
766 __m256d signbit = _mm256_castsi256_pd( _mm256_set1_epi32(0x80000000) );
767 __m256d one = _mm256_set1_pd(1.0);
768 __m256d two = _mm256_set1_pd(2.0);
774 jindex = nlist->jindex;
776 shiftidx = nlist->shift;
778 shiftvec = fr->shift_vec[0];
779 fshift = fr->fshift[0];
780 facel = _mm256_set1_pd(fr->epsfac);
781 charge = mdatoms->chargeA;
782 krf = _mm256_set1_pd(fr->ic->k_rf);
783 krf2 = _mm256_set1_pd(fr->ic->k_rf*2.0);
784 crf = _mm256_set1_pd(fr->ic->c_rf);
785 nvdwtype = fr->ntype;
787 vdwtype = mdatoms->typeA;
789 /* Setup water-specific parameters */
790 inr = nlist->iinr[0];
791 iq1 = _mm256_mul_pd(facel,_mm256_set1_pd(charge[inr+1]));
792 iq2 = _mm256_mul_pd(facel,_mm256_set1_pd(charge[inr+2]));
793 iq3 = _mm256_mul_pd(facel,_mm256_set1_pd(charge[inr+3]));
794 vdwioffsetptr0 = vdwparam+2*nvdwtype*vdwtype[inr+0];
796 /* When we use explicit cutoffs the value must be identical for elec and VdW, so use elec as an arbitrary choice */
797 rcutoff_scalar = fr->rcoulomb;
798 rcutoff = _mm256_set1_pd(rcutoff_scalar);
799 rcutoff2 = _mm256_mul_pd(rcutoff,rcutoff);
801 sh_vdw_invrcut6 = _mm256_set1_pd(fr->ic->sh_invrc6);
802 rvdw = _mm256_set1_pd(fr->rvdw);
804 /* Avoid stupid compiler warnings */
805 jnrA = jnrB = jnrC = jnrD = 0;
814 for(iidx=0;iidx<4*DIM;iidx++)
819 /* Start outer loop over neighborlists */
820 for(iidx=0; iidx<nri; iidx++)
822 /* Load shift vector for this list */
823 i_shift_offset = DIM*shiftidx[iidx];
825 /* Load limits for loop over neighbors */
826 j_index_start = jindex[iidx];
827 j_index_end = jindex[iidx+1];
829 /* Get outer coordinate index */
831 i_coord_offset = DIM*inr;
833 /* Load i particle coords and add shift vector */
834 gmx_mm256_load_shift_and_4rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset,
835 &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2,&ix3,&iy3,&iz3);
837 fix0 = _mm256_setzero_pd();
838 fiy0 = _mm256_setzero_pd();
839 fiz0 = _mm256_setzero_pd();
840 fix1 = _mm256_setzero_pd();
841 fiy1 = _mm256_setzero_pd();
842 fiz1 = _mm256_setzero_pd();
843 fix2 = _mm256_setzero_pd();
844 fiy2 = _mm256_setzero_pd();
845 fiz2 = _mm256_setzero_pd();
846 fix3 = _mm256_setzero_pd();
847 fiy3 = _mm256_setzero_pd();
848 fiz3 = _mm256_setzero_pd();
850 /* Start inner kernel loop */
851 for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
854 /* Get j neighbor index, and coordinate index */
859 j_coord_offsetA = DIM*jnrA;
860 j_coord_offsetB = DIM*jnrB;
861 j_coord_offsetC = DIM*jnrC;
862 j_coord_offsetD = DIM*jnrD;
864 /* load j atom coordinates */
865 gmx_mm256_load_1rvec_4ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
866 x+j_coord_offsetC,x+j_coord_offsetD,
869 /* Calculate displacement vector */
870 dx00 = _mm256_sub_pd(ix0,jx0);
871 dy00 = _mm256_sub_pd(iy0,jy0);
872 dz00 = _mm256_sub_pd(iz0,jz0);
873 dx10 = _mm256_sub_pd(ix1,jx0);
874 dy10 = _mm256_sub_pd(iy1,jy0);
875 dz10 = _mm256_sub_pd(iz1,jz0);
876 dx20 = _mm256_sub_pd(ix2,jx0);
877 dy20 = _mm256_sub_pd(iy2,jy0);
878 dz20 = _mm256_sub_pd(iz2,jz0);
879 dx30 = _mm256_sub_pd(ix3,jx0);
880 dy30 = _mm256_sub_pd(iy3,jy0);
881 dz30 = _mm256_sub_pd(iz3,jz0);
883 /* Calculate squared distance and things based on it */
884 rsq00 = gmx_mm256_calc_rsq_pd(dx00,dy00,dz00);
885 rsq10 = gmx_mm256_calc_rsq_pd(dx10,dy10,dz10);
886 rsq20 = gmx_mm256_calc_rsq_pd(dx20,dy20,dz20);
887 rsq30 = gmx_mm256_calc_rsq_pd(dx30,dy30,dz30);
889 rinv10 = gmx_mm256_invsqrt_pd(rsq10);
890 rinv20 = gmx_mm256_invsqrt_pd(rsq20);
891 rinv30 = gmx_mm256_invsqrt_pd(rsq30);
893 rinvsq00 = gmx_mm256_inv_pd(rsq00);
894 rinvsq10 = _mm256_mul_pd(rinv10,rinv10);
895 rinvsq20 = _mm256_mul_pd(rinv20,rinv20);
896 rinvsq30 = _mm256_mul_pd(rinv30,rinv30);
898 /* Load parameters for j particles */
899 jq0 = gmx_mm256_load_4real_swizzle_pd(charge+jnrA+0,charge+jnrB+0,
900 charge+jnrC+0,charge+jnrD+0);
901 vdwjidx0A = 2*vdwtype[jnrA+0];
902 vdwjidx0B = 2*vdwtype[jnrB+0];
903 vdwjidx0C = 2*vdwtype[jnrC+0];
904 vdwjidx0D = 2*vdwtype[jnrD+0];
906 fjx0 = _mm256_setzero_pd();
907 fjy0 = _mm256_setzero_pd();
908 fjz0 = _mm256_setzero_pd();
910 /**************************
911 * CALCULATE INTERACTIONS *
912 **************************/
914 if (gmx_mm256_any_lt(rsq00,rcutoff2))
917 /* Compute parameters for interactions between i and j atoms */
918 gmx_mm256_load_4pair_swizzle_pd(vdwioffsetptr0+vdwjidx0A,
919 vdwioffsetptr0+vdwjidx0B,
920 vdwioffsetptr0+vdwjidx0C,
921 vdwioffsetptr0+vdwjidx0D,
924 /* LENNARD-JONES DISPERSION/REPULSION */
926 rinvsix = _mm256_mul_pd(_mm256_mul_pd(rinvsq00,rinvsq00),rinvsq00);
927 fvdw = _mm256_mul_pd(_mm256_sub_pd(_mm256_mul_pd(c12_00,rinvsix),c6_00),_mm256_mul_pd(rinvsix,rinvsq00));
929 cutoff_mask = _mm256_cmp_pd(rsq00,rcutoff2,_CMP_LT_OQ);
933 fscal = _mm256_and_pd(fscal,cutoff_mask);
935 /* Calculate temporary vectorial force */
936 tx = _mm256_mul_pd(fscal,dx00);
937 ty = _mm256_mul_pd(fscal,dy00);
938 tz = _mm256_mul_pd(fscal,dz00);
940 /* Update vectorial force */
941 fix0 = _mm256_add_pd(fix0,tx);
942 fiy0 = _mm256_add_pd(fiy0,ty);
943 fiz0 = _mm256_add_pd(fiz0,tz);
945 fjx0 = _mm256_add_pd(fjx0,tx);
946 fjy0 = _mm256_add_pd(fjy0,ty);
947 fjz0 = _mm256_add_pd(fjz0,tz);
951 /**************************
952 * CALCULATE INTERACTIONS *
953 **************************/
955 if (gmx_mm256_any_lt(rsq10,rcutoff2))
958 /* Compute parameters for interactions between i and j atoms */
959 qq10 = _mm256_mul_pd(iq1,jq0);
961 /* REACTION-FIELD ELECTROSTATICS */
962 felec = _mm256_mul_pd(qq10,_mm256_sub_pd(_mm256_mul_pd(rinv10,rinvsq10),krf2));
964 cutoff_mask = _mm256_cmp_pd(rsq10,rcutoff2,_CMP_LT_OQ);
968 fscal = _mm256_and_pd(fscal,cutoff_mask);
970 /* Calculate temporary vectorial force */
971 tx = _mm256_mul_pd(fscal,dx10);
972 ty = _mm256_mul_pd(fscal,dy10);
973 tz = _mm256_mul_pd(fscal,dz10);
975 /* Update vectorial force */
976 fix1 = _mm256_add_pd(fix1,tx);
977 fiy1 = _mm256_add_pd(fiy1,ty);
978 fiz1 = _mm256_add_pd(fiz1,tz);
980 fjx0 = _mm256_add_pd(fjx0,tx);
981 fjy0 = _mm256_add_pd(fjy0,ty);
982 fjz0 = _mm256_add_pd(fjz0,tz);
986 /**************************
987 * CALCULATE INTERACTIONS *
988 **************************/
990 if (gmx_mm256_any_lt(rsq20,rcutoff2))
993 /* Compute parameters for interactions between i and j atoms */
994 qq20 = _mm256_mul_pd(iq2,jq0);
996 /* REACTION-FIELD ELECTROSTATICS */
997 felec = _mm256_mul_pd(qq20,_mm256_sub_pd(_mm256_mul_pd(rinv20,rinvsq20),krf2));
999 cutoff_mask = _mm256_cmp_pd(rsq20,rcutoff2,_CMP_LT_OQ);
1003 fscal = _mm256_and_pd(fscal,cutoff_mask);
1005 /* Calculate temporary vectorial force */
1006 tx = _mm256_mul_pd(fscal,dx20);
1007 ty = _mm256_mul_pd(fscal,dy20);
1008 tz = _mm256_mul_pd(fscal,dz20);
1010 /* Update vectorial force */
1011 fix2 = _mm256_add_pd(fix2,tx);
1012 fiy2 = _mm256_add_pd(fiy2,ty);
1013 fiz2 = _mm256_add_pd(fiz2,tz);
1015 fjx0 = _mm256_add_pd(fjx0,tx);
1016 fjy0 = _mm256_add_pd(fjy0,ty);
1017 fjz0 = _mm256_add_pd(fjz0,tz);
1021 /**************************
1022 * CALCULATE INTERACTIONS *
1023 **************************/
1025 if (gmx_mm256_any_lt(rsq30,rcutoff2))
1028 /* Compute parameters for interactions between i and j atoms */
1029 qq30 = _mm256_mul_pd(iq3,jq0);
1031 /* REACTION-FIELD ELECTROSTATICS */
1032 felec = _mm256_mul_pd(qq30,_mm256_sub_pd(_mm256_mul_pd(rinv30,rinvsq30),krf2));
1034 cutoff_mask = _mm256_cmp_pd(rsq30,rcutoff2,_CMP_LT_OQ);
1038 fscal = _mm256_and_pd(fscal,cutoff_mask);
1040 /* Calculate temporary vectorial force */
1041 tx = _mm256_mul_pd(fscal,dx30);
1042 ty = _mm256_mul_pd(fscal,dy30);
1043 tz = _mm256_mul_pd(fscal,dz30);
1045 /* Update vectorial force */
1046 fix3 = _mm256_add_pd(fix3,tx);
1047 fiy3 = _mm256_add_pd(fiy3,ty);
1048 fiz3 = _mm256_add_pd(fiz3,tz);
1050 fjx0 = _mm256_add_pd(fjx0,tx);
1051 fjy0 = _mm256_add_pd(fjy0,ty);
1052 fjz0 = _mm256_add_pd(fjz0,tz);
1056 fjptrA = f+j_coord_offsetA;
1057 fjptrB = f+j_coord_offsetB;
1058 fjptrC = f+j_coord_offsetC;
1059 fjptrD = f+j_coord_offsetD;
1061 gmx_mm256_decrement_1rvec_4ptr_swizzle_pd(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
1063 /* Inner loop uses 123 flops */
1066 if(jidx<j_index_end)
1069 /* Get j neighbor index, and coordinate index */
1070 jnrlistA = jjnr[jidx];
1071 jnrlistB = jjnr[jidx+1];
1072 jnrlistC = jjnr[jidx+2];
1073 jnrlistD = jjnr[jidx+3];
1074 /* Sign of each element will be negative for non-real atoms.
1075 * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
1076 * so use it as val = _mm_andnot_pd(mask,val) to clear dummy entries.
1078 tmpmask0 = gmx_mm_castsi128_pd(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
1080 tmpmask1 = _mm_permute_ps(tmpmask0,_GMX_MM_PERMUTE(3,3,2,2));
1081 tmpmask0 = _mm_permute_ps(tmpmask0,_GMX_MM_PERMUTE(1,1,0,0));
1082 dummy_mask = _mm256_castps_pd(gmx_mm256_set_m128(tmpmask1,tmpmask0));
1084 jnrA = (jnrlistA>=0) ? jnrlistA : 0;
1085 jnrB = (jnrlistB>=0) ? jnrlistB : 0;
1086 jnrC = (jnrlistC>=0) ? jnrlistC : 0;
1087 jnrD = (jnrlistD>=0) ? jnrlistD : 0;
1088 j_coord_offsetA = DIM*jnrA;
1089 j_coord_offsetB = DIM*jnrB;
1090 j_coord_offsetC = DIM*jnrC;
1091 j_coord_offsetD = DIM*jnrD;
1093 /* load j atom coordinates */
1094 gmx_mm256_load_1rvec_4ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
1095 x+j_coord_offsetC,x+j_coord_offsetD,
1098 /* Calculate displacement vector */
1099 dx00 = _mm256_sub_pd(ix0,jx0);
1100 dy00 = _mm256_sub_pd(iy0,jy0);
1101 dz00 = _mm256_sub_pd(iz0,jz0);
1102 dx10 = _mm256_sub_pd(ix1,jx0);
1103 dy10 = _mm256_sub_pd(iy1,jy0);
1104 dz10 = _mm256_sub_pd(iz1,jz0);
1105 dx20 = _mm256_sub_pd(ix2,jx0);
1106 dy20 = _mm256_sub_pd(iy2,jy0);
1107 dz20 = _mm256_sub_pd(iz2,jz0);
1108 dx30 = _mm256_sub_pd(ix3,jx0);
1109 dy30 = _mm256_sub_pd(iy3,jy0);
1110 dz30 = _mm256_sub_pd(iz3,jz0);
1112 /* Calculate squared distance and things based on it */
1113 rsq00 = gmx_mm256_calc_rsq_pd(dx00,dy00,dz00);
1114 rsq10 = gmx_mm256_calc_rsq_pd(dx10,dy10,dz10);
1115 rsq20 = gmx_mm256_calc_rsq_pd(dx20,dy20,dz20);
1116 rsq30 = gmx_mm256_calc_rsq_pd(dx30,dy30,dz30);
1118 rinv10 = gmx_mm256_invsqrt_pd(rsq10);
1119 rinv20 = gmx_mm256_invsqrt_pd(rsq20);
1120 rinv30 = gmx_mm256_invsqrt_pd(rsq30);
1122 rinvsq00 = gmx_mm256_inv_pd(rsq00);
1123 rinvsq10 = _mm256_mul_pd(rinv10,rinv10);
1124 rinvsq20 = _mm256_mul_pd(rinv20,rinv20);
1125 rinvsq30 = _mm256_mul_pd(rinv30,rinv30);
1127 /* Load parameters for j particles */
1128 jq0 = gmx_mm256_load_4real_swizzle_pd(charge+jnrA+0,charge+jnrB+0,
1129 charge+jnrC+0,charge+jnrD+0);
1130 vdwjidx0A = 2*vdwtype[jnrA+0];
1131 vdwjidx0B = 2*vdwtype[jnrB+0];
1132 vdwjidx0C = 2*vdwtype[jnrC+0];
1133 vdwjidx0D = 2*vdwtype[jnrD+0];
1135 fjx0 = _mm256_setzero_pd();
1136 fjy0 = _mm256_setzero_pd();
1137 fjz0 = _mm256_setzero_pd();
1139 /**************************
1140 * CALCULATE INTERACTIONS *
1141 **************************/
1143 if (gmx_mm256_any_lt(rsq00,rcutoff2))
1146 /* Compute parameters for interactions between i and j atoms */
1147 gmx_mm256_load_4pair_swizzle_pd(vdwioffsetptr0+vdwjidx0A,
1148 vdwioffsetptr0+vdwjidx0B,
1149 vdwioffsetptr0+vdwjidx0C,
1150 vdwioffsetptr0+vdwjidx0D,
1153 /* LENNARD-JONES DISPERSION/REPULSION */
1155 rinvsix = _mm256_mul_pd(_mm256_mul_pd(rinvsq00,rinvsq00),rinvsq00);
1156 fvdw = _mm256_mul_pd(_mm256_sub_pd(_mm256_mul_pd(c12_00,rinvsix),c6_00),_mm256_mul_pd(rinvsix,rinvsq00));
1158 cutoff_mask = _mm256_cmp_pd(rsq00,rcutoff2,_CMP_LT_OQ);
1162 fscal = _mm256_and_pd(fscal,cutoff_mask);
1164 fscal = _mm256_andnot_pd(dummy_mask,fscal);
1166 /* Calculate temporary vectorial force */
1167 tx = _mm256_mul_pd(fscal,dx00);
1168 ty = _mm256_mul_pd(fscal,dy00);
1169 tz = _mm256_mul_pd(fscal,dz00);
1171 /* Update vectorial force */
1172 fix0 = _mm256_add_pd(fix0,tx);
1173 fiy0 = _mm256_add_pd(fiy0,ty);
1174 fiz0 = _mm256_add_pd(fiz0,tz);
1176 fjx0 = _mm256_add_pd(fjx0,tx);
1177 fjy0 = _mm256_add_pd(fjy0,ty);
1178 fjz0 = _mm256_add_pd(fjz0,tz);
1182 /**************************
1183 * CALCULATE INTERACTIONS *
1184 **************************/
1186 if (gmx_mm256_any_lt(rsq10,rcutoff2))
1189 /* Compute parameters for interactions between i and j atoms */
1190 qq10 = _mm256_mul_pd(iq1,jq0);
1192 /* REACTION-FIELD ELECTROSTATICS */
1193 felec = _mm256_mul_pd(qq10,_mm256_sub_pd(_mm256_mul_pd(rinv10,rinvsq10),krf2));
1195 cutoff_mask = _mm256_cmp_pd(rsq10,rcutoff2,_CMP_LT_OQ);
1199 fscal = _mm256_and_pd(fscal,cutoff_mask);
1201 fscal = _mm256_andnot_pd(dummy_mask,fscal);
1203 /* Calculate temporary vectorial force */
1204 tx = _mm256_mul_pd(fscal,dx10);
1205 ty = _mm256_mul_pd(fscal,dy10);
1206 tz = _mm256_mul_pd(fscal,dz10);
1208 /* Update vectorial force */
1209 fix1 = _mm256_add_pd(fix1,tx);
1210 fiy1 = _mm256_add_pd(fiy1,ty);
1211 fiz1 = _mm256_add_pd(fiz1,tz);
1213 fjx0 = _mm256_add_pd(fjx0,tx);
1214 fjy0 = _mm256_add_pd(fjy0,ty);
1215 fjz0 = _mm256_add_pd(fjz0,tz);
1219 /**************************
1220 * CALCULATE INTERACTIONS *
1221 **************************/
1223 if (gmx_mm256_any_lt(rsq20,rcutoff2))
1226 /* Compute parameters for interactions between i and j atoms */
1227 qq20 = _mm256_mul_pd(iq2,jq0);
1229 /* REACTION-FIELD ELECTROSTATICS */
1230 felec = _mm256_mul_pd(qq20,_mm256_sub_pd(_mm256_mul_pd(rinv20,rinvsq20),krf2));
1232 cutoff_mask = _mm256_cmp_pd(rsq20,rcutoff2,_CMP_LT_OQ);
1236 fscal = _mm256_and_pd(fscal,cutoff_mask);
1238 fscal = _mm256_andnot_pd(dummy_mask,fscal);
1240 /* Calculate temporary vectorial force */
1241 tx = _mm256_mul_pd(fscal,dx20);
1242 ty = _mm256_mul_pd(fscal,dy20);
1243 tz = _mm256_mul_pd(fscal,dz20);
1245 /* Update vectorial force */
1246 fix2 = _mm256_add_pd(fix2,tx);
1247 fiy2 = _mm256_add_pd(fiy2,ty);
1248 fiz2 = _mm256_add_pd(fiz2,tz);
1250 fjx0 = _mm256_add_pd(fjx0,tx);
1251 fjy0 = _mm256_add_pd(fjy0,ty);
1252 fjz0 = _mm256_add_pd(fjz0,tz);
1256 /**************************
1257 * CALCULATE INTERACTIONS *
1258 **************************/
1260 if (gmx_mm256_any_lt(rsq30,rcutoff2))
1263 /* Compute parameters for interactions between i and j atoms */
1264 qq30 = _mm256_mul_pd(iq3,jq0);
1266 /* REACTION-FIELD ELECTROSTATICS */
1267 felec = _mm256_mul_pd(qq30,_mm256_sub_pd(_mm256_mul_pd(rinv30,rinvsq30),krf2));
1269 cutoff_mask = _mm256_cmp_pd(rsq30,rcutoff2,_CMP_LT_OQ);
1273 fscal = _mm256_and_pd(fscal,cutoff_mask);
1275 fscal = _mm256_andnot_pd(dummy_mask,fscal);
1277 /* Calculate temporary vectorial force */
1278 tx = _mm256_mul_pd(fscal,dx30);
1279 ty = _mm256_mul_pd(fscal,dy30);
1280 tz = _mm256_mul_pd(fscal,dz30);
1282 /* Update vectorial force */
1283 fix3 = _mm256_add_pd(fix3,tx);
1284 fiy3 = _mm256_add_pd(fiy3,ty);
1285 fiz3 = _mm256_add_pd(fiz3,tz);
1287 fjx0 = _mm256_add_pd(fjx0,tx);
1288 fjy0 = _mm256_add_pd(fjy0,ty);
1289 fjz0 = _mm256_add_pd(fjz0,tz);
1293 fjptrA = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
1294 fjptrB = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
1295 fjptrC = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
1296 fjptrD = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
1298 gmx_mm256_decrement_1rvec_4ptr_swizzle_pd(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
1300 /* Inner loop uses 123 flops */
1303 /* End of innermost loop */
1305 gmx_mm256_update_iforce_4atom_swizzle_pd(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,fix3,fiy3,fiz3,
1306 f+i_coord_offset,fshift+i_shift_offset);
1308 /* Increment number of inner iterations */
1309 inneriter += j_index_end - j_index_start;
1311 /* Outer loop uses 24 flops */
1314 /* Increment number of outer iterations */
1317 /* Update outer/inner flops */
1319 inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W4_F,outeriter*24 + inneriter*123);