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.
44 #include "../nb_kernel.h"
45 #include "gromacs/legacyheaders/types/simple.h"
46 #include "gromacs/math/vec.h"
47 #include "gromacs/legacyheaders/nrnb.h"
49 #include "gromacs/simd/math_x86_avx_256_double.h"
50 #include "kernelutil_x86_avx_256_double.h"
53 * Gromacs nonbonded kernel: nb_kernel_ElecRF_VdwLJ_GeomW3W3_VF_avx_256_double
54 * Electrostatics interaction: ReactionField
55 * VdW interaction: LennardJones
56 * Geometry: Water3-Water3
57 * Calculate force/pot: PotentialAndForce
60 nb_kernel_ElecRF_VdwLJ_GeomW3W3_VF_avx_256_double
61 (t_nblist * gmx_restrict nlist,
62 rvec * gmx_restrict xx,
63 rvec * gmx_restrict ff,
64 t_forcerec * gmx_restrict fr,
65 t_mdatoms * gmx_restrict mdatoms,
66 nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
67 t_nrnb * gmx_restrict nrnb)
69 /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
70 * just 0 for non-waters.
71 * Suffixes A,B,C,D refer to j loop unrolling done with AVX, e.g. for the four different
72 * jnr indices corresponding to data put in the four positions in the SIMD register.
74 int i_shift_offset,i_coord_offset,outeriter,inneriter;
75 int j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
76 int jnrA,jnrB,jnrC,jnrD;
77 int jnrlistA,jnrlistB,jnrlistC,jnrlistD;
78 int jnrlistE,jnrlistF,jnrlistG,jnrlistH;
79 int j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
80 int *iinr,*jindex,*jjnr,*shiftidx,*gid;
82 real *shiftvec,*fshift,*x,*f;
83 real *fjptrA,*fjptrB,*fjptrC,*fjptrD;
85 __m256d tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
86 real * vdwioffsetptr0;
87 __m256d ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
88 real * vdwioffsetptr1;
89 __m256d ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
90 real * vdwioffsetptr2;
91 __m256d ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
92 int vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
93 __m256d jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
94 int vdwjidx1A,vdwjidx1B,vdwjidx1C,vdwjidx1D;
95 __m256d jx1,jy1,jz1,fjx1,fjy1,fjz1,jq1,isaj1;
96 int vdwjidx2A,vdwjidx2B,vdwjidx2C,vdwjidx2D;
97 __m256d jx2,jy2,jz2,fjx2,fjy2,fjz2,jq2,isaj2;
98 __m256d dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
99 __m256d dx01,dy01,dz01,rsq01,rinv01,rinvsq01,r01,qq01,c6_01,c12_01;
100 __m256d dx02,dy02,dz02,rsq02,rinv02,rinvsq02,r02,qq02,c6_02,c12_02;
101 __m256d dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
102 __m256d dx11,dy11,dz11,rsq11,rinv11,rinvsq11,r11,qq11,c6_11,c12_11;
103 __m256d dx12,dy12,dz12,rsq12,rinv12,rinvsq12,r12,qq12,c6_12,c12_12;
104 __m256d dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
105 __m256d dx21,dy21,dz21,rsq21,rinv21,rinvsq21,r21,qq21,c6_21,c12_21;
106 __m256d dx22,dy22,dz22,rsq22,rinv22,rinvsq22,r22,qq22,c6_22,c12_22;
107 __m256d velec,felec,velecsum,facel,crf,krf,krf2;
110 __m256d rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
113 __m256d one_sixth = _mm256_set1_pd(1.0/6.0);
114 __m256d one_twelfth = _mm256_set1_pd(1.0/12.0);
115 __m256d dummy_mask,cutoff_mask;
116 __m128 tmpmask0,tmpmask1;
117 __m256d signbit = _mm256_castsi256_pd( _mm256_set1_epi32(0x80000000) );
118 __m256d one = _mm256_set1_pd(1.0);
119 __m256d two = _mm256_set1_pd(2.0);
125 jindex = nlist->jindex;
127 shiftidx = nlist->shift;
129 shiftvec = fr->shift_vec[0];
130 fshift = fr->fshift[0];
131 facel = _mm256_set1_pd(fr->epsfac);
132 charge = mdatoms->chargeA;
133 krf = _mm256_set1_pd(fr->ic->k_rf);
134 krf2 = _mm256_set1_pd(fr->ic->k_rf*2.0);
135 crf = _mm256_set1_pd(fr->ic->c_rf);
136 nvdwtype = fr->ntype;
138 vdwtype = mdatoms->typeA;
140 /* Setup water-specific parameters */
141 inr = nlist->iinr[0];
142 iq0 = _mm256_mul_pd(facel,_mm256_set1_pd(charge[inr+0]));
143 iq1 = _mm256_mul_pd(facel,_mm256_set1_pd(charge[inr+1]));
144 iq2 = _mm256_mul_pd(facel,_mm256_set1_pd(charge[inr+2]));
145 vdwioffsetptr0 = vdwparam+2*nvdwtype*vdwtype[inr+0];
147 jq0 = _mm256_set1_pd(charge[inr+0]);
148 jq1 = _mm256_set1_pd(charge[inr+1]);
149 jq2 = _mm256_set1_pd(charge[inr+2]);
150 vdwjidx0A = 2*vdwtype[inr+0];
151 qq00 = _mm256_mul_pd(iq0,jq0);
152 c6_00 = _mm256_set1_pd(vdwioffsetptr0[vdwjidx0A]);
153 c12_00 = _mm256_set1_pd(vdwioffsetptr0[vdwjidx0A+1]);
154 qq01 = _mm256_mul_pd(iq0,jq1);
155 qq02 = _mm256_mul_pd(iq0,jq2);
156 qq10 = _mm256_mul_pd(iq1,jq0);
157 qq11 = _mm256_mul_pd(iq1,jq1);
158 qq12 = _mm256_mul_pd(iq1,jq2);
159 qq20 = _mm256_mul_pd(iq2,jq0);
160 qq21 = _mm256_mul_pd(iq2,jq1);
161 qq22 = _mm256_mul_pd(iq2,jq2);
163 /* Avoid stupid compiler warnings */
164 jnrA = jnrB = jnrC = jnrD = 0;
173 for(iidx=0;iidx<4*DIM;iidx++)
178 /* Start outer loop over neighborlists */
179 for(iidx=0; iidx<nri; iidx++)
181 /* Load shift vector for this list */
182 i_shift_offset = DIM*shiftidx[iidx];
184 /* Load limits for loop over neighbors */
185 j_index_start = jindex[iidx];
186 j_index_end = jindex[iidx+1];
188 /* Get outer coordinate index */
190 i_coord_offset = DIM*inr;
192 /* Load i particle coords and add shift vector */
193 gmx_mm256_load_shift_and_3rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset,
194 &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2);
196 fix0 = _mm256_setzero_pd();
197 fiy0 = _mm256_setzero_pd();
198 fiz0 = _mm256_setzero_pd();
199 fix1 = _mm256_setzero_pd();
200 fiy1 = _mm256_setzero_pd();
201 fiz1 = _mm256_setzero_pd();
202 fix2 = _mm256_setzero_pd();
203 fiy2 = _mm256_setzero_pd();
204 fiz2 = _mm256_setzero_pd();
206 /* Reset potential sums */
207 velecsum = _mm256_setzero_pd();
208 vvdwsum = _mm256_setzero_pd();
210 /* Start inner kernel loop */
211 for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
214 /* Get j neighbor index, and coordinate index */
219 j_coord_offsetA = DIM*jnrA;
220 j_coord_offsetB = DIM*jnrB;
221 j_coord_offsetC = DIM*jnrC;
222 j_coord_offsetD = DIM*jnrD;
224 /* load j atom coordinates */
225 gmx_mm256_load_3rvec_4ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
226 x+j_coord_offsetC,x+j_coord_offsetD,
227 &jx0,&jy0,&jz0,&jx1,&jy1,&jz1,&jx2,&jy2,&jz2);
229 /* Calculate displacement vector */
230 dx00 = _mm256_sub_pd(ix0,jx0);
231 dy00 = _mm256_sub_pd(iy0,jy0);
232 dz00 = _mm256_sub_pd(iz0,jz0);
233 dx01 = _mm256_sub_pd(ix0,jx1);
234 dy01 = _mm256_sub_pd(iy0,jy1);
235 dz01 = _mm256_sub_pd(iz0,jz1);
236 dx02 = _mm256_sub_pd(ix0,jx2);
237 dy02 = _mm256_sub_pd(iy0,jy2);
238 dz02 = _mm256_sub_pd(iz0,jz2);
239 dx10 = _mm256_sub_pd(ix1,jx0);
240 dy10 = _mm256_sub_pd(iy1,jy0);
241 dz10 = _mm256_sub_pd(iz1,jz0);
242 dx11 = _mm256_sub_pd(ix1,jx1);
243 dy11 = _mm256_sub_pd(iy1,jy1);
244 dz11 = _mm256_sub_pd(iz1,jz1);
245 dx12 = _mm256_sub_pd(ix1,jx2);
246 dy12 = _mm256_sub_pd(iy1,jy2);
247 dz12 = _mm256_sub_pd(iz1,jz2);
248 dx20 = _mm256_sub_pd(ix2,jx0);
249 dy20 = _mm256_sub_pd(iy2,jy0);
250 dz20 = _mm256_sub_pd(iz2,jz0);
251 dx21 = _mm256_sub_pd(ix2,jx1);
252 dy21 = _mm256_sub_pd(iy2,jy1);
253 dz21 = _mm256_sub_pd(iz2,jz1);
254 dx22 = _mm256_sub_pd(ix2,jx2);
255 dy22 = _mm256_sub_pd(iy2,jy2);
256 dz22 = _mm256_sub_pd(iz2,jz2);
258 /* Calculate squared distance and things based on it */
259 rsq00 = gmx_mm256_calc_rsq_pd(dx00,dy00,dz00);
260 rsq01 = gmx_mm256_calc_rsq_pd(dx01,dy01,dz01);
261 rsq02 = gmx_mm256_calc_rsq_pd(dx02,dy02,dz02);
262 rsq10 = gmx_mm256_calc_rsq_pd(dx10,dy10,dz10);
263 rsq11 = gmx_mm256_calc_rsq_pd(dx11,dy11,dz11);
264 rsq12 = gmx_mm256_calc_rsq_pd(dx12,dy12,dz12);
265 rsq20 = gmx_mm256_calc_rsq_pd(dx20,dy20,dz20);
266 rsq21 = gmx_mm256_calc_rsq_pd(dx21,dy21,dz21);
267 rsq22 = gmx_mm256_calc_rsq_pd(dx22,dy22,dz22);
269 rinv00 = gmx_mm256_invsqrt_pd(rsq00);
270 rinv01 = gmx_mm256_invsqrt_pd(rsq01);
271 rinv02 = gmx_mm256_invsqrt_pd(rsq02);
272 rinv10 = gmx_mm256_invsqrt_pd(rsq10);
273 rinv11 = gmx_mm256_invsqrt_pd(rsq11);
274 rinv12 = gmx_mm256_invsqrt_pd(rsq12);
275 rinv20 = gmx_mm256_invsqrt_pd(rsq20);
276 rinv21 = gmx_mm256_invsqrt_pd(rsq21);
277 rinv22 = gmx_mm256_invsqrt_pd(rsq22);
279 rinvsq00 = _mm256_mul_pd(rinv00,rinv00);
280 rinvsq01 = _mm256_mul_pd(rinv01,rinv01);
281 rinvsq02 = _mm256_mul_pd(rinv02,rinv02);
282 rinvsq10 = _mm256_mul_pd(rinv10,rinv10);
283 rinvsq11 = _mm256_mul_pd(rinv11,rinv11);
284 rinvsq12 = _mm256_mul_pd(rinv12,rinv12);
285 rinvsq20 = _mm256_mul_pd(rinv20,rinv20);
286 rinvsq21 = _mm256_mul_pd(rinv21,rinv21);
287 rinvsq22 = _mm256_mul_pd(rinv22,rinv22);
289 fjx0 = _mm256_setzero_pd();
290 fjy0 = _mm256_setzero_pd();
291 fjz0 = _mm256_setzero_pd();
292 fjx1 = _mm256_setzero_pd();
293 fjy1 = _mm256_setzero_pd();
294 fjz1 = _mm256_setzero_pd();
295 fjx2 = _mm256_setzero_pd();
296 fjy2 = _mm256_setzero_pd();
297 fjz2 = _mm256_setzero_pd();
299 /**************************
300 * CALCULATE INTERACTIONS *
301 **************************/
303 /* REACTION-FIELD ELECTROSTATICS */
304 velec = _mm256_mul_pd(qq00,_mm256_sub_pd(_mm256_add_pd(rinv00,_mm256_mul_pd(krf,rsq00)),crf));
305 felec = _mm256_mul_pd(qq00,_mm256_sub_pd(_mm256_mul_pd(rinv00,rinvsq00),krf2));
307 /* LENNARD-JONES DISPERSION/REPULSION */
309 rinvsix = _mm256_mul_pd(_mm256_mul_pd(rinvsq00,rinvsq00),rinvsq00);
310 vvdw6 = _mm256_mul_pd(c6_00,rinvsix);
311 vvdw12 = _mm256_mul_pd(c12_00,_mm256_mul_pd(rinvsix,rinvsix));
312 vvdw = _mm256_sub_pd( _mm256_mul_pd(vvdw12,one_twelfth) , _mm256_mul_pd(vvdw6,one_sixth) );
313 fvdw = _mm256_mul_pd(_mm256_sub_pd(vvdw12,vvdw6),rinvsq00);
315 /* Update potential sum for this i atom from the interaction with this j atom. */
316 velecsum = _mm256_add_pd(velecsum,velec);
317 vvdwsum = _mm256_add_pd(vvdwsum,vvdw);
319 fscal = _mm256_add_pd(felec,fvdw);
321 /* Calculate temporary vectorial force */
322 tx = _mm256_mul_pd(fscal,dx00);
323 ty = _mm256_mul_pd(fscal,dy00);
324 tz = _mm256_mul_pd(fscal,dz00);
326 /* Update vectorial force */
327 fix0 = _mm256_add_pd(fix0,tx);
328 fiy0 = _mm256_add_pd(fiy0,ty);
329 fiz0 = _mm256_add_pd(fiz0,tz);
331 fjx0 = _mm256_add_pd(fjx0,tx);
332 fjy0 = _mm256_add_pd(fjy0,ty);
333 fjz0 = _mm256_add_pd(fjz0,tz);
335 /**************************
336 * CALCULATE INTERACTIONS *
337 **************************/
339 /* REACTION-FIELD ELECTROSTATICS */
340 velec = _mm256_mul_pd(qq01,_mm256_sub_pd(_mm256_add_pd(rinv01,_mm256_mul_pd(krf,rsq01)),crf));
341 felec = _mm256_mul_pd(qq01,_mm256_sub_pd(_mm256_mul_pd(rinv01,rinvsq01),krf2));
343 /* Update potential sum for this i atom from the interaction with this j atom. */
344 velecsum = _mm256_add_pd(velecsum,velec);
348 /* Calculate temporary vectorial force */
349 tx = _mm256_mul_pd(fscal,dx01);
350 ty = _mm256_mul_pd(fscal,dy01);
351 tz = _mm256_mul_pd(fscal,dz01);
353 /* Update vectorial force */
354 fix0 = _mm256_add_pd(fix0,tx);
355 fiy0 = _mm256_add_pd(fiy0,ty);
356 fiz0 = _mm256_add_pd(fiz0,tz);
358 fjx1 = _mm256_add_pd(fjx1,tx);
359 fjy1 = _mm256_add_pd(fjy1,ty);
360 fjz1 = _mm256_add_pd(fjz1,tz);
362 /**************************
363 * CALCULATE INTERACTIONS *
364 **************************/
366 /* REACTION-FIELD ELECTROSTATICS */
367 velec = _mm256_mul_pd(qq02,_mm256_sub_pd(_mm256_add_pd(rinv02,_mm256_mul_pd(krf,rsq02)),crf));
368 felec = _mm256_mul_pd(qq02,_mm256_sub_pd(_mm256_mul_pd(rinv02,rinvsq02),krf2));
370 /* Update potential sum for this i atom from the interaction with this j atom. */
371 velecsum = _mm256_add_pd(velecsum,velec);
375 /* Calculate temporary vectorial force */
376 tx = _mm256_mul_pd(fscal,dx02);
377 ty = _mm256_mul_pd(fscal,dy02);
378 tz = _mm256_mul_pd(fscal,dz02);
380 /* Update vectorial force */
381 fix0 = _mm256_add_pd(fix0,tx);
382 fiy0 = _mm256_add_pd(fiy0,ty);
383 fiz0 = _mm256_add_pd(fiz0,tz);
385 fjx2 = _mm256_add_pd(fjx2,tx);
386 fjy2 = _mm256_add_pd(fjy2,ty);
387 fjz2 = _mm256_add_pd(fjz2,tz);
389 /**************************
390 * CALCULATE INTERACTIONS *
391 **************************/
393 /* REACTION-FIELD ELECTROSTATICS */
394 velec = _mm256_mul_pd(qq10,_mm256_sub_pd(_mm256_add_pd(rinv10,_mm256_mul_pd(krf,rsq10)),crf));
395 felec = _mm256_mul_pd(qq10,_mm256_sub_pd(_mm256_mul_pd(rinv10,rinvsq10),krf2));
397 /* Update potential sum for this i atom from the interaction with this j atom. */
398 velecsum = _mm256_add_pd(velecsum,velec);
402 /* Calculate temporary vectorial force */
403 tx = _mm256_mul_pd(fscal,dx10);
404 ty = _mm256_mul_pd(fscal,dy10);
405 tz = _mm256_mul_pd(fscal,dz10);
407 /* Update vectorial force */
408 fix1 = _mm256_add_pd(fix1,tx);
409 fiy1 = _mm256_add_pd(fiy1,ty);
410 fiz1 = _mm256_add_pd(fiz1,tz);
412 fjx0 = _mm256_add_pd(fjx0,tx);
413 fjy0 = _mm256_add_pd(fjy0,ty);
414 fjz0 = _mm256_add_pd(fjz0,tz);
416 /**************************
417 * CALCULATE INTERACTIONS *
418 **************************/
420 /* REACTION-FIELD ELECTROSTATICS */
421 velec = _mm256_mul_pd(qq11,_mm256_sub_pd(_mm256_add_pd(rinv11,_mm256_mul_pd(krf,rsq11)),crf));
422 felec = _mm256_mul_pd(qq11,_mm256_sub_pd(_mm256_mul_pd(rinv11,rinvsq11),krf2));
424 /* Update potential sum for this i atom from the interaction with this j atom. */
425 velecsum = _mm256_add_pd(velecsum,velec);
429 /* Calculate temporary vectorial force */
430 tx = _mm256_mul_pd(fscal,dx11);
431 ty = _mm256_mul_pd(fscal,dy11);
432 tz = _mm256_mul_pd(fscal,dz11);
434 /* Update vectorial force */
435 fix1 = _mm256_add_pd(fix1,tx);
436 fiy1 = _mm256_add_pd(fiy1,ty);
437 fiz1 = _mm256_add_pd(fiz1,tz);
439 fjx1 = _mm256_add_pd(fjx1,tx);
440 fjy1 = _mm256_add_pd(fjy1,ty);
441 fjz1 = _mm256_add_pd(fjz1,tz);
443 /**************************
444 * CALCULATE INTERACTIONS *
445 **************************/
447 /* REACTION-FIELD ELECTROSTATICS */
448 velec = _mm256_mul_pd(qq12,_mm256_sub_pd(_mm256_add_pd(rinv12,_mm256_mul_pd(krf,rsq12)),crf));
449 felec = _mm256_mul_pd(qq12,_mm256_sub_pd(_mm256_mul_pd(rinv12,rinvsq12),krf2));
451 /* Update potential sum for this i atom from the interaction with this j atom. */
452 velecsum = _mm256_add_pd(velecsum,velec);
456 /* Calculate temporary vectorial force */
457 tx = _mm256_mul_pd(fscal,dx12);
458 ty = _mm256_mul_pd(fscal,dy12);
459 tz = _mm256_mul_pd(fscal,dz12);
461 /* Update vectorial force */
462 fix1 = _mm256_add_pd(fix1,tx);
463 fiy1 = _mm256_add_pd(fiy1,ty);
464 fiz1 = _mm256_add_pd(fiz1,tz);
466 fjx2 = _mm256_add_pd(fjx2,tx);
467 fjy2 = _mm256_add_pd(fjy2,ty);
468 fjz2 = _mm256_add_pd(fjz2,tz);
470 /**************************
471 * CALCULATE INTERACTIONS *
472 **************************/
474 /* REACTION-FIELD ELECTROSTATICS */
475 velec = _mm256_mul_pd(qq20,_mm256_sub_pd(_mm256_add_pd(rinv20,_mm256_mul_pd(krf,rsq20)),crf));
476 felec = _mm256_mul_pd(qq20,_mm256_sub_pd(_mm256_mul_pd(rinv20,rinvsq20),krf2));
478 /* Update potential sum for this i atom from the interaction with this j atom. */
479 velecsum = _mm256_add_pd(velecsum,velec);
483 /* Calculate temporary vectorial force */
484 tx = _mm256_mul_pd(fscal,dx20);
485 ty = _mm256_mul_pd(fscal,dy20);
486 tz = _mm256_mul_pd(fscal,dz20);
488 /* Update vectorial force */
489 fix2 = _mm256_add_pd(fix2,tx);
490 fiy2 = _mm256_add_pd(fiy2,ty);
491 fiz2 = _mm256_add_pd(fiz2,tz);
493 fjx0 = _mm256_add_pd(fjx0,tx);
494 fjy0 = _mm256_add_pd(fjy0,ty);
495 fjz0 = _mm256_add_pd(fjz0,tz);
497 /**************************
498 * CALCULATE INTERACTIONS *
499 **************************/
501 /* REACTION-FIELD ELECTROSTATICS */
502 velec = _mm256_mul_pd(qq21,_mm256_sub_pd(_mm256_add_pd(rinv21,_mm256_mul_pd(krf,rsq21)),crf));
503 felec = _mm256_mul_pd(qq21,_mm256_sub_pd(_mm256_mul_pd(rinv21,rinvsq21),krf2));
505 /* Update potential sum for this i atom from the interaction with this j atom. */
506 velecsum = _mm256_add_pd(velecsum,velec);
510 /* Calculate temporary vectorial force */
511 tx = _mm256_mul_pd(fscal,dx21);
512 ty = _mm256_mul_pd(fscal,dy21);
513 tz = _mm256_mul_pd(fscal,dz21);
515 /* Update vectorial force */
516 fix2 = _mm256_add_pd(fix2,tx);
517 fiy2 = _mm256_add_pd(fiy2,ty);
518 fiz2 = _mm256_add_pd(fiz2,tz);
520 fjx1 = _mm256_add_pd(fjx1,tx);
521 fjy1 = _mm256_add_pd(fjy1,ty);
522 fjz1 = _mm256_add_pd(fjz1,tz);
524 /**************************
525 * CALCULATE INTERACTIONS *
526 **************************/
528 /* REACTION-FIELD ELECTROSTATICS */
529 velec = _mm256_mul_pd(qq22,_mm256_sub_pd(_mm256_add_pd(rinv22,_mm256_mul_pd(krf,rsq22)),crf));
530 felec = _mm256_mul_pd(qq22,_mm256_sub_pd(_mm256_mul_pd(rinv22,rinvsq22),krf2));
532 /* Update potential sum for this i atom from the interaction with this j atom. */
533 velecsum = _mm256_add_pd(velecsum,velec);
537 /* Calculate temporary vectorial force */
538 tx = _mm256_mul_pd(fscal,dx22);
539 ty = _mm256_mul_pd(fscal,dy22);
540 tz = _mm256_mul_pd(fscal,dz22);
542 /* Update vectorial force */
543 fix2 = _mm256_add_pd(fix2,tx);
544 fiy2 = _mm256_add_pd(fiy2,ty);
545 fiz2 = _mm256_add_pd(fiz2,tz);
547 fjx2 = _mm256_add_pd(fjx2,tx);
548 fjy2 = _mm256_add_pd(fjy2,ty);
549 fjz2 = _mm256_add_pd(fjz2,tz);
551 fjptrA = f+j_coord_offsetA;
552 fjptrB = f+j_coord_offsetB;
553 fjptrC = f+j_coord_offsetC;
554 fjptrD = f+j_coord_offsetD;
556 gmx_mm256_decrement_3rvec_4ptr_swizzle_pd(fjptrA,fjptrB,fjptrC,fjptrD,
557 fjx0,fjy0,fjz0,fjx1,fjy1,fjz1,fjx2,fjy2,fjz2);
559 /* Inner loop uses 300 flops */
565 /* Get j neighbor index, and coordinate index */
566 jnrlistA = jjnr[jidx];
567 jnrlistB = jjnr[jidx+1];
568 jnrlistC = jjnr[jidx+2];
569 jnrlistD = jjnr[jidx+3];
570 /* Sign of each element will be negative for non-real atoms.
571 * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
572 * so use it as val = _mm_andnot_pd(mask,val) to clear dummy entries.
574 tmpmask0 = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
576 tmpmask1 = _mm_permute_ps(tmpmask0,_GMX_MM_PERMUTE(3,3,2,2));
577 tmpmask0 = _mm_permute_ps(tmpmask0,_GMX_MM_PERMUTE(1,1,0,0));
578 dummy_mask = _mm256_castps_pd(gmx_mm256_set_m128(tmpmask1,tmpmask0));
580 jnrA = (jnrlistA>=0) ? jnrlistA : 0;
581 jnrB = (jnrlistB>=0) ? jnrlistB : 0;
582 jnrC = (jnrlistC>=0) ? jnrlistC : 0;
583 jnrD = (jnrlistD>=0) ? jnrlistD : 0;
584 j_coord_offsetA = DIM*jnrA;
585 j_coord_offsetB = DIM*jnrB;
586 j_coord_offsetC = DIM*jnrC;
587 j_coord_offsetD = DIM*jnrD;
589 /* load j atom coordinates */
590 gmx_mm256_load_3rvec_4ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
591 x+j_coord_offsetC,x+j_coord_offsetD,
592 &jx0,&jy0,&jz0,&jx1,&jy1,&jz1,&jx2,&jy2,&jz2);
594 /* Calculate displacement vector */
595 dx00 = _mm256_sub_pd(ix0,jx0);
596 dy00 = _mm256_sub_pd(iy0,jy0);
597 dz00 = _mm256_sub_pd(iz0,jz0);
598 dx01 = _mm256_sub_pd(ix0,jx1);
599 dy01 = _mm256_sub_pd(iy0,jy1);
600 dz01 = _mm256_sub_pd(iz0,jz1);
601 dx02 = _mm256_sub_pd(ix0,jx2);
602 dy02 = _mm256_sub_pd(iy0,jy2);
603 dz02 = _mm256_sub_pd(iz0,jz2);
604 dx10 = _mm256_sub_pd(ix1,jx0);
605 dy10 = _mm256_sub_pd(iy1,jy0);
606 dz10 = _mm256_sub_pd(iz1,jz0);
607 dx11 = _mm256_sub_pd(ix1,jx1);
608 dy11 = _mm256_sub_pd(iy1,jy1);
609 dz11 = _mm256_sub_pd(iz1,jz1);
610 dx12 = _mm256_sub_pd(ix1,jx2);
611 dy12 = _mm256_sub_pd(iy1,jy2);
612 dz12 = _mm256_sub_pd(iz1,jz2);
613 dx20 = _mm256_sub_pd(ix2,jx0);
614 dy20 = _mm256_sub_pd(iy2,jy0);
615 dz20 = _mm256_sub_pd(iz2,jz0);
616 dx21 = _mm256_sub_pd(ix2,jx1);
617 dy21 = _mm256_sub_pd(iy2,jy1);
618 dz21 = _mm256_sub_pd(iz2,jz1);
619 dx22 = _mm256_sub_pd(ix2,jx2);
620 dy22 = _mm256_sub_pd(iy2,jy2);
621 dz22 = _mm256_sub_pd(iz2,jz2);
623 /* Calculate squared distance and things based on it */
624 rsq00 = gmx_mm256_calc_rsq_pd(dx00,dy00,dz00);
625 rsq01 = gmx_mm256_calc_rsq_pd(dx01,dy01,dz01);
626 rsq02 = gmx_mm256_calc_rsq_pd(dx02,dy02,dz02);
627 rsq10 = gmx_mm256_calc_rsq_pd(dx10,dy10,dz10);
628 rsq11 = gmx_mm256_calc_rsq_pd(dx11,dy11,dz11);
629 rsq12 = gmx_mm256_calc_rsq_pd(dx12,dy12,dz12);
630 rsq20 = gmx_mm256_calc_rsq_pd(dx20,dy20,dz20);
631 rsq21 = gmx_mm256_calc_rsq_pd(dx21,dy21,dz21);
632 rsq22 = gmx_mm256_calc_rsq_pd(dx22,dy22,dz22);
634 rinv00 = gmx_mm256_invsqrt_pd(rsq00);
635 rinv01 = gmx_mm256_invsqrt_pd(rsq01);
636 rinv02 = gmx_mm256_invsqrt_pd(rsq02);
637 rinv10 = gmx_mm256_invsqrt_pd(rsq10);
638 rinv11 = gmx_mm256_invsqrt_pd(rsq11);
639 rinv12 = gmx_mm256_invsqrt_pd(rsq12);
640 rinv20 = gmx_mm256_invsqrt_pd(rsq20);
641 rinv21 = gmx_mm256_invsqrt_pd(rsq21);
642 rinv22 = gmx_mm256_invsqrt_pd(rsq22);
644 rinvsq00 = _mm256_mul_pd(rinv00,rinv00);
645 rinvsq01 = _mm256_mul_pd(rinv01,rinv01);
646 rinvsq02 = _mm256_mul_pd(rinv02,rinv02);
647 rinvsq10 = _mm256_mul_pd(rinv10,rinv10);
648 rinvsq11 = _mm256_mul_pd(rinv11,rinv11);
649 rinvsq12 = _mm256_mul_pd(rinv12,rinv12);
650 rinvsq20 = _mm256_mul_pd(rinv20,rinv20);
651 rinvsq21 = _mm256_mul_pd(rinv21,rinv21);
652 rinvsq22 = _mm256_mul_pd(rinv22,rinv22);
654 fjx0 = _mm256_setzero_pd();
655 fjy0 = _mm256_setzero_pd();
656 fjz0 = _mm256_setzero_pd();
657 fjx1 = _mm256_setzero_pd();
658 fjy1 = _mm256_setzero_pd();
659 fjz1 = _mm256_setzero_pd();
660 fjx2 = _mm256_setzero_pd();
661 fjy2 = _mm256_setzero_pd();
662 fjz2 = _mm256_setzero_pd();
664 /**************************
665 * CALCULATE INTERACTIONS *
666 **************************/
668 /* REACTION-FIELD ELECTROSTATICS */
669 velec = _mm256_mul_pd(qq00,_mm256_sub_pd(_mm256_add_pd(rinv00,_mm256_mul_pd(krf,rsq00)),crf));
670 felec = _mm256_mul_pd(qq00,_mm256_sub_pd(_mm256_mul_pd(rinv00,rinvsq00),krf2));
672 /* LENNARD-JONES DISPERSION/REPULSION */
674 rinvsix = _mm256_mul_pd(_mm256_mul_pd(rinvsq00,rinvsq00),rinvsq00);
675 vvdw6 = _mm256_mul_pd(c6_00,rinvsix);
676 vvdw12 = _mm256_mul_pd(c12_00,_mm256_mul_pd(rinvsix,rinvsix));
677 vvdw = _mm256_sub_pd( _mm256_mul_pd(vvdw12,one_twelfth) , _mm256_mul_pd(vvdw6,one_sixth) );
678 fvdw = _mm256_mul_pd(_mm256_sub_pd(vvdw12,vvdw6),rinvsq00);
680 /* Update potential sum for this i atom from the interaction with this j atom. */
681 velec = _mm256_andnot_pd(dummy_mask,velec);
682 velecsum = _mm256_add_pd(velecsum,velec);
683 vvdw = _mm256_andnot_pd(dummy_mask,vvdw);
684 vvdwsum = _mm256_add_pd(vvdwsum,vvdw);
686 fscal = _mm256_add_pd(felec,fvdw);
688 fscal = _mm256_andnot_pd(dummy_mask,fscal);
690 /* Calculate temporary vectorial force */
691 tx = _mm256_mul_pd(fscal,dx00);
692 ty = _mm256_mul_pd(fscal,dy00);
693 tz = _mm256_mul_pd(fscal,dz00);
695 /* Update vectorial force */
696 fix0 = _mm256_add_pd(fix0,tx);
697 fiy0 = _mm256_add_pd(fiy0,ty);
698 fiz0 = _mm256_add_pd(fiz0,tz);
700 fjx0 = _mm256_add_pd(fjx0,tx);
701 fjy0 = _mm256_add_pd(fjy0,ty);
702 fjz0 = _mm256_add_pd(fjz0,tz);
704 /**************************
705 * CALCULATE INTERACTIONS *
706 **************************/
708 /* REACTION-FIELD ELECTROSTATICS */
709 velec = _mm256_mul_pd(qq01,_mm256_sub_pd(_mm256_add_pd(rinv01,_mm256_mul_pd(krf,rsq01)),crf));
710 felec = _mm256_mul_pd(qq01,_mm256_sub_pd(_mm256_mul_pd(rinv01,rinvsq01),krf2));
712 /* Update potential sum for this i atom from the interaction with this j atom. */
713 velec = _mm256_andnot_pd(dummy_mask,velec);
714 velecsum = _mm256_add_pd(velecsum,velec);
718 fscal = _mm256_andnot_pd(dummy_mask,fscal);
720 /* Calculate temporary vectorial force */
721 tx = _mm256_mul_pd(fscal,dx01);
722 ty = _mm256_mul_pd(fscal,dy01);
723 tz = _mm256_mul_pd(fscal,dz01);
725 /* Update vectorial force */
726 fix0 = _mm256_add_pd(fix0,tx);
727 fiy0 = _mm256_add_pd(fiy0,ty);
728 fiz0 = _mm256_add_pd(fiz0,tz);
730 fjx1 = _mm256_add_pd(fjx1,tx);
731 fjy1 = _mm256_add_pd(fjy1,ty);
732 fjz1 = _mm256_add_pd(fjz1,tz);
734 /**************************
735 * CALCULATE INTERACTIONS *
736 **************************/
738 /* REACTION-FIELD ELECTROSTATICS */
739 velec = _mm256_mul_pd(qq02,_mm256_sub_pd(_mm256_add_pd(rinv02,_mm256_mul_pd(krf,rsq02)),crf));
740 felec = _mm256_mul_pd(qq02,_mm256_sub_pd(_mm256_mul_pd(rinv02,rinvsq02),krf2));
742 /* Update potential sum for this i atom from the interaction with this j atom. */
743 velec = _mm256_andnot_pd(dummy_mask,velec);
744 velecsum = _mm256_add_pd(velecsum,velec);
748 fscal = _mm256_andnot_pd(dummy_mask,fscal);
750 /* Calculate temporary vectorial force */
751 tx = _mm256_mul_pd(fscal,dx02);
752 ty = _mm256_mul_pd(fscal,dy02);
753 tz = _mm256_mul_pd(fscal,dz02);
755 /* Update vectorial force */
756 fix0 = _mm256_add_pd(fix0,tx);
757 fiy0 = _mm256_add_pd(fiy0,ty);
758 fiz0 = _mm256_add_pd(fiz0,tz);
760 fjx2 = _mm256_add_pd(fjx2,tx);
761 fjy2 = _mm256_add_pd(fjy2,ty);
762 fjz2 = _mm256_add_pd(fjz2,tz);
764 /**************************
765 * CALCULATE INTERACTIONS *
766 **************************/
768 /* REACTION-FIELD ELECTROSTATICS */
769 velec = _mm256_mul_pd(qq10,_mm256_sub_pd(_mm256_add_pd(rinv10,_mm256_mul_pd(krf,rsq10)),crf));
770 felec = _mm256_mul_pd(qq10,_mm256_sub_pd(_mm256_mul_pd(rinv10,rinvsq10),krf2));
772 /* Update potential sum for this i atom from the interaction with this j atom. */
773 velec = _mm256_andnot_pd(dummy_mask,velec);
774 velecsum = _mm256_add_pd(velecsum,velec);
778 fscal = _mm256_andnot_pd(dummy_mask,fscal);
780 /* Calculate temporary vectorial force */
781 tx = _mm256_mul_pd(fscal,dx10);
782 ty = _mm256_mul_pd(fscal,dy10);
783 tz = _mm256_mul_pd(fscal,dz10);
785 /* Update vectorial force */
786 fix1 = _mm256_add_pd(fix1,tx);
787 fiy1 = _mm256_add_pd(fiy1,ty);
788 fiz1 = _mm256_add_pd(fiz1,tz);
790 fjx0 = _mm256_add_pd(fjx0,tx);
791 fjy0 = _mm256_add_pd(fjy0,ty);
792 fjz0 = _mm256_add_pd(fjz0,tz);
794 /**************************
795 * CALCULATE INTERACTIONS *
796 **************************/
798 /* REACTION-FIELD ELECTROSTATICS */
799 velec = _mm256_mul_pd(qq11,_mm256_sub_pd(_mm256_add_pd(rinv11,_mm256_mul_pd(krf,rsq11)),crf));
800 felec = _mm256_mul_pd(qq11,_mm256_sub_pd(_mm256_mul_pd(rinv11,rinvsq11),krf2));
802 /* Update potential sum for this i atom from the interaction with this j atom. */
803 velec = _mm256_andnot_pd(dummy_mask,velec);
804 velecsum = _mm256_add_pd(velecsum,velec);
808 fscal = _mm256_andnot_pd(dummy_mask,fscal);
810 /* Calculate temporary vectorial force */
811 tx = _mm256_mul_pd(fscal,dx11);
812 ty = _mm256_mul_pd(fscal,dy11);
813 tz = _mm256_mul_pd(fscal,dz11);
815 /* Update vectorial force */
816 fix1 = _mm256_add_pd(fix1,tx);
817 fiy1 = _mm256_add_pd(fiy1,ty);
818 fiz1 = _mm256_add_pd(fiz1,tz);
820 fjx1 = _mm256_add_pd(fjx1,tx);
821 fjy1 = _mm256_add_pd(fjy1,ty);
822 fjz1 = _mm256_add_pd(fjz1,tz);
824 /**************************
825 * CALCULATE INTERACTIONS *
826 **************************/
828 /* REACTION-FIELD ELECTROSTATICS */
829 velec = _mm256_mul_pd(qq12,_mm256_sub_pd(_mm256_add_pd(rinv12,_mm256_mul_pd(krf,rsq12)),crf));
830 felec = _mm256_mul_pd(qq12,_mm256_sub_pd(_mm256_mul_pd(rinv12,rinvsq12),krf2));
832 /* Update potential sum for this i atom from the interaction with this j atom. */
833 velec = _mm256_andnot_pd(dummy_mask,velec);
834 velecsum = _mm256_add_pd(velecsum,velec);
838 fscal = _mm256_andnot_pd(dummy_mask,fscal);
840 /* Calculate temporary vectorial force */
841 tx = _mm256_mul_pd(fscal,dx12);
842 ty = _mm256_mul_pd(fscal,dy12);
843 tz = _mm256_mul_pd(fscal,dz12);
845 /* Update vectorial force */
846 fix1 = _mm256_add_pd(fix1,tx);
847 fiy1 = _mm256_add_pd(fiy1,ty);
848 fiz1 = _mm256_add_pd(fiz1,tz);
850 fjx2 = _mm256_add_pd(fjx2,tx);
851 fjy2 = _mm256_add_pd(fjy2,ty);
852 fjz2 = _mm256_add_pd(fjz2,tz);
854 /**************************
855 * CALCULATE INTERACTIONS *
856 **************************/
858 /* REACTION-FIELD ELECTROSTATICS */
859 velec = _mm256_mul_pd(qq20,_mm256_sub_pd(_mm256_add_pd(rinv20,_mm256_mul_pd(krf,rsq20)),crf));
860 felec = _mm256_mul_pd(qq20,_mm256_sub_pd(_mm256_mul_pd(rinv20,rinvsq20),krf2));
862 /* Update potential sum for this i atom from the interaction with this j atom. */
863 velec = _mm256_andnot_pd(dummy_mask,velec);
864 velecsum = _mm256_add_pd(velecsum,velec);
868 fscal = _mm256_andnot_pd(dummy_mask,fscal);
870 /* Calculate temporary vectorial force */
871 tx = _mm256_mul_pd(fscal,dx20);
872 ty = _mm256_mul_pd(fscal,dy20);
873 tz = _mm256_mul_pd(fscal,dz20);
875 /* Update vectorial force */
876 fix2 = _mm256_add_pd(fix2,tx);
877 fiy2 = _mm256_add_pd(fiy2,ty);
878 fiz2 = _mm256_add_pd(fiz2,tz);
880 fjx0 = _mm256_add_pd(fjx0,tx);
881 fjy0 = _mm256_add_pd(fjy0,ty);
882 fjz0 = _mm256_add_pd(fjz0,tz);
884 /**************************
885 * CALCULATE INTERACTIONS *
886 **************************/
888 /* REACTION-FIELD ELECTROSTATICS */
889 velec = _mm256_mul_pd(qq21,_mm256_sub_pd(_mm256_add_pd(rinv21,_mm256_mul_pd(krf,rsq21)),crf));
890 felec = _mm256_mul_pd(qq21,_mm256_sub_pd(_mm256_mul_pd(rinv21,rinvsq21),krf2));
892 /* Update potential sum for this i atom from the interaction with this j atom. */
893 velec = _mm256_andnot_pd(dummy_mask,velec);
894 velecsum = _mm256_add_pd(velecsum,velec);
898 fscal = _mm256_andnot_pd(dummy_mask,fscal);
900 /* Calculate temporary vectorial force */
901 tx = _mm256_mul_pd(fscal,dx21);
902 ty = _mm256_mul_pd(fscal,dy21);
903 tz = _mm256_mul_pd(fscal,dz21);
905 /* Update vectorial force */
906 fix2 = _mm256_add_pd(fix2,tx);
907 fiy2 = _mm256_add_pd(fiy2,ty);
908 fiz2 = _mm256_add_pd(fiz2,tz);
910 fjx1 = _mm256_add_pd(fjx1,tx);
911 fjy1 = _mm256_add_pd(fjy1,ty);
912 fjz1 = _mm256_add_pd(fjz1,tz);
914 /**************************
915 * CALCULATE INTERACTIONS *
916 **************************/
918 /* REACTION-FIELD ELECTROSTATICS */
919 velec = _mm256_mul_pd(qq22,_mm256_sub_pd(_mm256_add_pd(rinv22,_mm256_mul_pd(krf,rsq22)),crf));
920 felec = _mm256_mul_pd(qq22,_mm256_sub_pd(_mm256_mul_pd(rinv22,rinvsq22),krf2));
922 /* Update potential sum for this i atom from the interaction with this j atom. */
923 velec = _mm256_andnot_pd(dummy_mask,velec);
924 velecsum = _mm256_add_pd(velecsum,velec);
928 fscal = _mm256_andnot_pd(dummy_mask,fscal);
930 /* Calculate temporary vectorial force */
931 tx = _mm256_mul_pd(fscal,dx22);
932 ty = _mm256_mul_pd(fscal,dy22);
933 tz = _mm256_mul_pd(fscal,dz22);
935 /* Update vectorial force */
936 fix2 = _mm256_add_pd(fix2,tx);
937 fiy2 = _mm256_add_pd(fiy2,ty);
938 fiz2 = _mm256_add_pd(fiz2,tz);
940 fjx2 = _mm256_add_pd(fjx2,tx);
941 fjy2 = _mm256_add_pd(fjy2,ty);
942 fjz2 = _mm256_add_pd(fjz2,tz);
944 fjptrA = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
945 fjptrB = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
946 fjptrC = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
947 fjptrD = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
949 gmx_mm256_decrement_3rvec_4ptr_swizzle_pd(fjptrA,fjptrB,fjptrC,fjptrD,
950 fjx0,fjy0,fjz0,fjx1,fjy1,fjz1,fjx2,fjy2,fjz2);
952 /* Inner loop uses 300 flops */
955 /* End of innermost loop */
957 gmx_mm256_update_iforce_3atom_swizzle_pd(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,
958 f+i_coord_offset,fshift+i_shift_offset);
961 /* Update potential energies */
962 gmx_mm256_update_1pot_pd(velecsum,kernel_data->energygrp_elec+ggid);
963 gmx_mm256_update_1pot_pd(vvdwsum,kernel_data->energygrp_vdw+ggid);
965 /* Increment number of inner iterations */
966 inneriter += j_index_end - j_index_start;
968 /* Outer loop uses 20 flops */
971 /* Increment number of outer iterations */
974 /* Update outer/inner flops */
976 inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W3W3_VF,outeriter*20 + inneriter*300);
979 * Gromacs nonbonded kernel: nb_kernel_ElecRF_VdwLJ_GeomW3W3_F_avx_256_double
980 * Electrostatics interaction: ReactionField
981 * VdW interaction: LennardJones
982 * Geometry: Water3-Water3
983 * Calculate force/pot: Force
986 nb_kernel_ElecRF_VdwLJ_GeomW3W3_F_avx_256_double
987 (t_nblist * gmx_restrict nlist,
988 rvec * gmx_restrict xx,
989 rvec * gmx_restrict ff,
990 t_forcerec * gmx_restrict fr,
991 t_mdatoms * gmx_restrict mdatoms,
992 nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
993 t_nrnb * gmx_restrict nrnb)
995 /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
996 * just 0 for non-waters.
997 * Suffixes A,B,C,D refer to j loop unrolling done with AVX, e.g. for the four different
998 * jnr indices corresponding to data put in the four positions in the SIMD register.
1000 int i_shift_offset,i_coord_offset,outeriter,inneriter;
1001 int j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
1002 int jnrA,jnrB,jnrC,jnrD;
1003 int jnrlistA,jnrlistB,jnrlistC,jnrlistD;
1004 int jnrlistE,jnrlistF,jnrlistG,jnrlistH;
1005 int j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
1006 int *iinr,*jindex,*jjnr,*shiftidx,*gid;
1007 real rcutoff_scalar;
1008 real *shiftvec,*fshift,*x,*f;
1009 real *fjptrA,*fjptrB,*fjptrC,*fjptrD;
1010 real scratch[4*DIM];
1011 __m256d tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
1012 real * vdwioffsetptr0;
1013 __m256d ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
1014 real * vdwioffsetptr1;
1015 __m256d ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
1016 real * vdwioffsetptr2;
1017 __m256d ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
1018 int vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
1019 __m256d jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
1020 int vdwjidx1A,vdwjidx1B,vdwjidx1C,vdwjidx1D;
1021 __m256d jx1,jy1,jz1,fjx1,fjy1,fjz1,jq1,isaj1;
1022 int vdwjidx2A,vdwjidx2B,vdwjidx2C,vdwjidx2D;
1023 __m256d jx2,jy2,jz2,fjx2,fjy2,fjz2,jq2,isaj2;
1024 __m256d dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
1025 __m256d dx01,dy01,dz01,rsq01,rinv01,rinvsq01,r01,qq01,c6_01,c12_01;
1026 __m256d dx02,dy02,dz02,rsq02,rinv02,rinvsq02,r02,qq02,c6_02,c12_02;
1027 __m256d dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
1028 __m256d dx11,dy11,dz11,rsq11,rinv11,rinvsq11,r11,qq11,c6_11,c12_11;
1029 __m256d dx12,dy12,dz12,rsq12,rinv12,rinvsq12,r12,qq12,c6_12,c12_12;
1030 __m256d dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
1031 __m256d dx21,dy21,dz21,rsq21,rinv21,rinvsq21,r21,qq21,c6_21,c12_21;
1032 __m256d dx22,dy22,dz22,rsq22,rinv22,rinvsq22,r22,qq22,c6_22,c12_22;
1033 __m256d velec,felec,velecsum,facel,crf,krf,krf2;
1036 __m256d rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
1039 __m256d one_sixth = _mm256_set1_pd(1.0/6.0);
1040 __m256d one_twelfth = _mm256_set1_pd(1.0/12.0);
1041 __m256d dummy_mask,cutoff_mask;
1042 __m128 tmpmask0,tmpmask1;
1043 __m256d signbit = _mm256_castsi256_pd( _mm256_set1_epi32(0x80000000) );
1044 __m256d one = _mm256_set1_pd(1.0);
1045 __m256d two = _mm256_set1_pd(2.0);
1051 jindex = nlist->jindex;
1053 shiftidx = nlist->shift;
1055 shiftvec = fr->shift_vec[0];
1056 fshift = fr->fshift[0];
1057 facel = _mm256_set1_pd(fr->epsfac);
1058 charge = mdatoms->chargeA;
1059 krf = _mm256_set1_pd(fr->ic->k_rf);
1060 krf2 = _mm256_set1_pd(fr->ic->k_rf*2.0);
1061 crf = _mm256_set1_pd(fr->ic->c_rf);
1062 nvdwtype = fr->ntype;
1063 vdwparam = fr->nbfp;
1064 vdwtype = mdatoms->typeA;
1066 /* Setup water-specific parameters */
1067 inr = nlist->iinr[0];
1068 iq0 = _mm256_mul_pd(facel,_mm256_set1_pd(charge[inr+0]));
1069 iq1 = _mm256_mul_pd(facel,_mm256_set1_pd(charge[inr+1]));
1070 iq2 = _mm256_mul_pd(facel,_mm256_set1_pd(charge[inr+2]));
1071 vdwioffsetptr0 = vdwparam+2*nvdwtype*vdwtype[inr+0];
1073 jq0 = _mm256_set1_pd(charge[inr+0]);
1074 jq1 = _mm256_set1_pd(charge[inr+1]);
1075 jq2 = _mm256_set1_pd(charge[inr+2]);
1076 vdwjidx0A = 2*vdwtype[inr+0];
1077 qq00 = _mm256_mul_pd(iq0,jq0);
1078 c6_00 = _mm256_set1_pd(vdwioffsetptr0[vdwjidx0A]);
1079 c12_00 = _mm256_set1_pd(vdwioffsetptr0[vdwjidx0A+1]);
1080 qq01 = _mm256_mul_pd(iq0,jq1);
1081 qq02 = _mm256_mul_pd(iq0,jq2);
1082 qq10 = _mm256_mul_pd(iq1,jq0);
1083 qq11 = _mm256_mul_pd(iq1,jq1);
1084 qq12 = _mm256_mul_pd(iq1,jq2);
1085 qq20 = _mm256_mul_pd(iq2,jq0);
1086 qq21 = _mm256_mul_pd(iq2,jq1);
1087 qq22 = _mm256_mul_pd(iq2,jq2);
1089 /* Avoid stupid compiler warnings */
1090 jnrA = jnrB = jnrC = jnrD = 0;
1091 j_coord_offsetA = 0;
1092 j_coord_offsetB = 0;
1093 j_coord_offsetC = 0;
1094 j_coord_offsetD = 0;
1099 for(iidx=0;iidx<4*DIM;iidx++)
1101 scratch[iidx] = 0.0;
1104 /* Start outer loop over neighborlists */
1105 for(iidx=0; iidx<nri; iidx++)
1107 /* Load shift vector for this list */
1108 i_shift_offset = DIM*shiftidx[iidx];
1110 /* Load limits for loop over neighbors */
1111 j_index_start = jindex[iidx];
1112 j_index_end = jindex[iidx+1];
1114 /* Get outer coordinate index */
1116 i_coord_offset = DIM*inr;
1118 /* Load i particle coords and add shift vector */
1119 gmx_mm256_load_shift_and_3rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset,
1120 &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2);
1122 fix0 = _mm256_setzero_pd();
1123 fiy0 = _mm256_setzero_pd();
1124 fiz0 = _mm256_setzero_pd();
1125 fix1 = _mm256_setzero_pd();
1126 fiy1 = _mm256_setzero_pd();
1127 fiz1 = _mm256_setzero_pd();
1128 fix2 = _mm256_setzero_pd();
1129 fiy2 = _mm256_setzero_pd();
1130 fiz2 = _mm256_setzero_pd();
1132 /* Start inner kernel loop */
1133 for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
1136 /* Get j neighbor index, and coordinate index */
1138 jnrB = jjnr[jidx+1];
1139 jnrC = jjnr[jidx+2];
1140 jnrD = jjnr[jidx+3];
1141 j_coord_offsetA = DIM*jnrA;
1142 j_coord_offsetB = DIM*jnrB;
1143 j_coord_offsetC = DIM*jnrC;
1144 j_coord_offsetD = DIM*jnrD;
1146 /* load j atom coordinates */
1147 gmx_mm256_load_3rvec_4ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
1148 x+j_coord_offsetC,x+j_coord_offsetD,
1149 &jx0,&jy0,&jz0,&jx1,&jy1,&jz1,&jx2,&jy2,&jz2);
1151 /* Calculate displacement vector */
1152 dx00 = _mm256_sub_pd(ix0,jx0);
1153 dy00 = _mm256_sub_pd(iy0,jy0);
1154 dz00 = _mm256_sub_pd(iz0,jz0);
1155 dx01 = _mm256_sub_pd(ix0,jx1);
1156 dy01 = _mm256_sub_pd(iy0,jy1);
1157 dz01 = _mm256_sub_pd(iz0,jz1);
1158 dx02 = _mm256_sub_pd(ix0,jx2);
1159 dy02 = _mm256_sub_pd(iy0,jy2);
1160 dz02 = _mm256_sub_pd(iz0,jz2);
1161 dx10 = _mm256_sub_pd(ix1,jx0);
1162 dy10 = _mm256_sub_pd(iy1,jy0);
1163 dz10 = _mm256_sub_pd(iz1,jz0);
1164 dx11 = _mm256_sub_pd(ix1,jx1);
1165 dy11 = _mm256_sub_pd(iy1,jy1);
1166 dz11 = _mm256_sub_pd(iz1,jz1);
1167 dx12 = _mm256_sub_pd(ix1,jx2);
1168 dy12 = _mm256_sub_pd(iy1,jy2);
1169 dz12 = _mm256_sub_pd(iz1,jz2);
1170 dx20 = _mm256_sub_pd(ix2,jx0);
1171 dy20 = _mm256_sub_pd(iy2,jy0);
1172 dz20 = _mm256_sub_pd(iz2,jz0);
1173 dx21 = _mm256_sub_pd(ix2,jx1);
1174 dy21 = _mm256_sub_pd(iy2,jy1);
1175 dz21 = _mm256_sub_pd(iz2,jz1);
1176 dx22 = _mm256_sub_pd(ix2,jx2);
1177 dy22 = _mm256_sub_pd(iy2,jy2);
1178 dz22 = _mm256_sub_pd(iz2,jz2);
1180 /* Calculate squared distance and things based on it */
1181 rsq00 = gmx_mm256_calc_rsq_pd(dx00,dy00,dz00);
1182 rsq01 = gmx_mm256_calc_rsq_pd(dx01,dy01,dz01);
1183 rsq02 = gmx_mm256_calc_rsq_pd(dx02,dy02,dz02);
1184 rsq10 = gmx_mm256_calc_rsq_pd(dx10,dy10,dz10);
1185 rsq11 = gmx_mm256_calc_rsq_pd(dx11,dy11,dz11);
1186 rsq12 = gmx_mm256_calc_rsq_pd(dx12,dy12,dz12);
1187 rsq20 = gmx_mm256_calc_rsq_pd(dx20,dy20,dz20);
1188 rsq21 = gmx_mm256_calc_rsq_pd(dx21,dy21,dz21);
1189 rsq22 = gmx_mm256_calc_rsq_pd(dx22,dy22,dz22);
1191 rinv00 = gmx_mm256_invsqrt_pd(rsq00);
1192 rinv01 = gmx_mm256_invsqrt_pd(rsq01);
1193 rinv02 = gmx_mm256_invsqrt_pd(rsq02);
1194 rinv10 = gmx_mm256_invsqrt_pd(rsq10);
1195 rinv11 = gmx_mm256_invsqrt_pd(rsq11);
1196 rinv12 = gmx_mm256_invsqrt_pd(rsq12);
1197 rinv20 = gmx_mm256_invsqrt_pd(rsq20);
1198 rinv21 = gmx_mm256_invsqrt_pd(rsq21);
1199 rinv22 = gmx_mm256_invsqrt_pd(rsq22);
1201 rinvsq00 = _mm256_mul_pd(rinv00,rinv00);
1202 rinvsq01 = _mm256_mul_pd(rinv01,rinv01);
1203 rinvsq02 = _mm256_mul_pd(rinv02,rinv02);
1204 rinvsq10 = _mm256_mul_pd(rinv10,rinv10);
1205 rinvsq11 = _mm256_mul_pd(rinv11,rinv11);
1206 rinvsq12 = _mm256_mul_pd(rinv12,rinv12);
1207 rinvsq20 = _mm256_mul_pd(rinv20,rinv20);
1208 rinvsq21 = _mm256_mul_pd(rinv21,rinv21);
1209 rinvsq22 = _mm256_mul_pd(rinv22,rinv22);
1211 fjx0 = _mm256_setzero_pd();
1212 fjy0 = _mm256_setzero_pd();
1213 fjz0 = _mm256_setzero_pd();
1214 fjx1 = _mm256_setzero_pd();
1215 fjy1 = _mm256_setzero_pd();
1216 fjz1 = _mm256_setzero_pd();
1217 fjx2 = _mm256_setzero_pd();
1218 fjy2 = _mm256_setzero_pd();
1219 fjz2 = _mm256_setzero_pd();
1221 /**************************
1222 * CALCULATE INTERACTIONS *
1223 **************************/
1225 /* REACTION-FIELD ELECTROSTATICS */
1226 felec = _mm256_mul_pd(qq00,_mm256_sub_pd(_mm256_mul_pd(rinv00,rinvsq00),krf2));
1228 /* LENNARD-JONES DISPERSION/REPULSION */
1230 rinvsix = _mm256_mul_pd(_mm256_mul_pd(rinvsq00,rinvsq00),rinvsq00);
1231 fvdw = _mm256_mul_pd(_mm256_sub_pd(_mm256_mul_pd(c12_00,rinvsix),c6_00),_mm256_mul_pd(rinvsix,rinvsq00));
1233 fscal = _mm256_add_pd(felec,fvdw);
1235 /* Calculate temporary vectorial force */
1236 tx = _mm256_mul_pd(fscal,dx00);
1237 ty = _mm256_mul_pd(fscal,dy00);
1238 tz = _mm256_mul_pd(fscal,dz00);
1240 /* Update vectorial force */
1241 fix0 = _mm256_add_pd(fix0,tx);
1242 fiy0 = _mm256_add_pd(fiy0,ty);
1243 fiz0 = _mm256_add_pd(fiz0,tz);
1245 fjx0 = _mm256_add_pd(fjx0,tx);
1246 fjy0 = _mm256_add_pd(fjy0,ty);
1247 fjz0 = _mm256_add_pd(fjz0,tz);
1249 /**************************
1250 * CALCULATE INTERACTIONS *
1251 **************************/
1253 /* REACTION-FIELD ELECTROSTATICS */
1254 felec = _mm256_mul_pd(qq01,_mm256_sub_pd(_mm256_mul_pd(rinv01,rinvsq01),krf2));
1258 /* Calculate temporary vectorial force */
1259 tx = _mm256_mul_pd(fscal,dx01);
1260 ty = _mm256_mul_pd(fscal,dy01);
1261 tz = _mm256_mul_pd(fscal,dz01);
1263 /* Update vectorial force */
1264 fix0 = _mm256_add_pd(fix0,tx);
1265 fiy0 = _mm256_add_pd(fiy0,ty);
1266 fiz0 = _mm256_add_pd(fiz0,tz);
1268 fjx1 = _mm256_add_pd(fjx1,tx);
1269 fjy1 = _mm256_add_pd(fjy1,ty);
1270 fjz1 = _mm256_add_pd(fjz1,tz);
1272 /**************************
1273 * CALCULATE INTERACTIONS *
1274 **************************/
1276 /* REACTION-FIELD ELECTROSTATICS */
1277 felec = _mm256_mul_pd(qq02,_mm256_sub_pd(_mm256_mul_pd(rinv02,rinvsq02),krf2));
1281 /* Calculate temporary vectorial force */
1282 tx = _mm256_mul_pd(fscal,dx02);
1283 ty = _mm256_mul_pd(fscal,dy02);
1284 tz = _mm256_mul_pd(fscal,dz02);
1286 /* Update vectorial force */
1287 fix0 = _mm256_add_pd(fix0,tx);
1288 fiy0 = _mm256_add_pd(fiy0,ty);
1289 fiz0 = _mm256_add_pd(fiz0,tz);
1291 fjx2 = _mm256_add_pd(fjx2,tx);
1292 fjy2 = _mm256_add_pd(fjy2,ty);
1293 fjz2 = _mm256_add_pd(fjz2,tz);
1295 /**************************
1296 * CALCULATE INTERACTIONS *
1297 **************************/
1299 /* REACTION-FIELD ELECTROSTATICS */
1300 felec = _mm256_mul_pd(qq10,_mm256_sub_pd(_mm256_mul_pd(rinv10,rinvsq10),krf2));
1304 /* Calculate temporary vectorial force */
1305 tx = _mm256_mul_pd(fscal,dx10);
1306 ty = _mm256_mul_pd(fscal,dy10);
1307 tz = _mm256_mul_pd(fscal,dz10);
1309 /* Update vectorial force */
1310 fix1 = _mm256_add_pd(fix1,tx);
1311 fiy1 = _mm256_add_pd(fiy1,ty);
1312 fiz1 = _mm256_add_pd(fiz1,tz);
1314 fjx0 = _mm256_add_pd(fjx0,tx);
1315 fjy0 = _mm256_add_pd(fjy0,ty);
1316 fjz0 = _mm256_add_pd(fjz0,tz);
1318 /**************************
1319 * CALCULATE INTERACTIONS *
1320 **************************/
1322 /* REACTION-FIELD ELECTROSTATICS */
1323 felec = _mm256_mul_pd(qq11,_mm256_sub_pd(_mm256_mul_pd(rinv11,rinvsq11),krf2));
1327 /* Calculate temporary vectorial force */
1328 tx = _mm256_mul_pd(fscal,dx11);
1329 ty = _mm256_mul_pd(fscal,dy11);
1330 tz = _mm256_mul_pd(fscal,dz11);
1332 /* Update vectorial force */
1333 fix1 = _mm256_add_pd(fix1,tx);
1334 fiy1 = _mm256_add_pd(fiy1,ty);
1335 fiz1 = _mm256_add_pd(fiz1,tz);
1337 fjx1 = _mm256_add_pd(fjx1,tx);
1338 fjy1 = _mm256_add_pd(fjy1,ty);
1339 fjz1 = _mm256_add_pd(fjz1,tz);
1341 /**************************
1342 * CALCULATE INTERACTIONS *
1343 **************************/
1345 /* REACTION-FIELD ELECTROSTATICS */
1346 felec = _mm256_mul_pd(qq12,_mm256_sub_pd(_mm256_mul_pd(rinv12,rinvsq12),krf2));
1350 /* Calculate temporary vectorial force */
1351 tx = _mm256_mul_pd(fscal,dx12);
1352 ty = _mm256_mul_pd(fscal,dy12);
1353 tz = _mm256_mul_pd(fscal,dz12);
1355 /* Update vectorial force */
1356 fix1 = _mm256_add_pd(fix1,tx);
1357 fiy1 = _mm256_add_pd(fiy1,ty);
1358 fiz1 = _mm256_add_pd(fiz1,tz);
1360 fjx2 = _mm256_add_pd(fjx2,tx);
1361 fjy2 = _mm256_add_pd(fjy2,ty);
1362 fjz2 = _mm256_add_pd(fjz2,tz);
1364 /**************************
1365 * CALCULATE INTERACTIONS *
1366 **************************/
1368 /* REACTION-FIELD ELECTROSTATICS */
1369 felec = _mm256_mul_pd(qq20,_mm256_sub_pd(_mm256_mul_pd(rinv20,rinvsq20),krf2));
1373 /* Calculate temporary vectorial force */
1374 tx = _mm256_mul_pd(fscal,dx20);
1375 ty = _mm256_mul_pd(fscal,dy20);
1376 tz = _mm256_mul_pd(fscal,dz20);
1378 /* Update vectorial force */
1379 fix2 = _mm256_add_pd(fix2,tx);
1380 fiy2 = _mm256_add_pd(fiy2,ty);
1381 fiz2 = _mm256_add_pd(fiz2,tz);
1383 fjx0 = _mm256_add_pd(fjx0,tx);
1384 fjy0 = _mm256_add_pd(fjy0,ty);
1385 fjz0 = _mm256_add_pd(fjz0,tz);
1387 /**************************
1388 * CALCULATE INTERACTIONS *
1389 **************************/
1391 /* REACTION-FIELD ELECTROSTATICS */
1392 felec = _mm256_mul_pd(qq21,_mm256_sub_pd(_mm256_mul_pd(rinv21,rinvsq21),krf2));
1396 /* Calculate temporary vectorial force */
1397 tx = _mm256_mul_pd(fscal,dx21);
1398 ty = _mm256_mul_pd(fscal,dy21);
1399 tz = _mm256_mul_pd(fscal,dz21);
1401 /* Update vectorial force */
1402 fix2 = _mm256_add_pd(fix2,tx);
1403 fiy2 = _mm256_add_pd(fiy2,ty);
1404 fiz2 = _mm256_add_pd(fiz2,tz);
1406 fjx1 = _mm256_add_pd(fjx1,tx);
1407 fjy1 = _mm256_add_pd(fjy1,ty);
1408 fjz1 = _mm256_add_pd(fjz1,tz);
1410 /**************************
1411 * CALCULATE INTERACTIONS *
1412 **************************/
1414 /* REACTION-FIELD ELECTROSTATICS */
1415 felec = _mm256_mul_pd(qq22,_mm256_sub_pd(_mm256_mul_pd(rinv22,rinvsq22),krf2));
1419 /* Calculate temporary vectorial force */
1420 tx = _mm256_mul_pd(fscal,dx22);
1421 ty = _mm256_mul_pd(fscal,dy22);
1422 tz = _mm256_mul_pd(fscal,dz22);
1424 /* Update vectorial force */
1425 fix2 = _mm256_add_pd(fix2,tx);
1426 fiy2 = _mm256_add_pd(fiy2,ty);
1427 fiz2 = _mm256_add_pd(fiz2,tz);
1429 fjx2 = _mm256_add_pd(fjx2,tx);
1430 fjy2 = _mm256_add_pd(fjy2,ty);
1431 fjz2 = _mm256_add_pd(fjz2,tz);
1433 fjptrA = f+j_coord_offsetA;
1434 fjptrB = f+j_coord_offsetB;
1435 fjptrC = f+j_coord_offsetC;
1436 fjptrD = f+j_coord_offsetD;
1438 gmx_mm256_decrement_3rvec_4ptr_swizzle_pd(fjptrA,fjptrB,fjptrC,fjptrD,
1439 fjx0,fjy0,fjz0,fjx1,fjy1,fjz1,fjx2,fjy2,fjz2);
1441 /* Inner loop uses 250 flops */
1444 if(jidx<j_index_end)
1447 /* Get j neighbor index, and coordinate index */
1448 jnrlistA = jjnr[jidx];
1449 jnrlistB = jjnr[jidx+1];
1450 jnrlistC = jjnr[jidx+2];
1451 jnrlistD = jjnr[jidx+3];
1452 /* Sign of each element will be negative for non-real atoms.
1453 * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
1454 * so use it as val = _mm_andnot_pd(mask,val) to clear dummy entries.
1456 tmpmask0 = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
1458 tmpmask1 = _mm_permute_ps(tmpmask0,_GMX_MM_PERMUTE(3,3,2,2));
1459 tmpmask0 = _mm_permute_ps(tmpmask0,_GMX_MM_PERMUTE(1,1,0,0));
1460 dummy_mask = _mm256_castps_pd(gmx_mm256_set_m128(tmpmask1,tmpmask0));
1462 jnrA = (jnrlistA>=0) ? jnrlistA : 0;
1463 jnrB = (jnrlistB>=0) ? jnrlistB : 0;
1464 jnrC = (jnrlistC>=0) ? jnrlistC : 0;
1465 jnrD = (jnrlistD>=0) ? jnrlistD : 0;
1466 j_coord_offsetA = DIM*jnrA;
1467 j_coord_offsetB = DIM*jnrB;
1468 j_coord_offsetC = DIM*jnrC;
1469 j_coord_offsetD = DIM*jnrD;
1471 /* load j atom coordinates */
1472 gmx_mm256_load_3rvec_4ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
1473 x+j_coord_offsetC,x+j_coord_offsetD,
1474 &jx0,&jy0,&jz0,&jx1,&jy1,&jz1,&jx2,&jy2,&jz2);
1476 /* Calculate displacement vector */
1477 dx00 = _mm256_sub_pd(ix0,jx0);
1478 dy00 = _mm256_sub_pd(iy0,jy0);
1479 dz00 = _mm256_sub_pd(iz0,jz0);
1480 dx01 = _mm256_sub_pd(ix0,jx1);
1481 dy01 = _mm256_sub_pd(iy0,jy1);
1482 dz01 = _mm256_sub_pd(iz0,jz1);
1483 dx02 = _mm256_sub_pd(ix0,jx2);
1484 dy02 = _mm256_sub_pd(iy0,jy2);
1485 dz02 = _mm256_sub_pd(iz0,jz2);
1486 dx10 = _mm256_sub_pd(ix1,jx0);
1487 dy10 = _mm256_sub_pd(iy1,jy0);
1488 dz10 = _mm256_sub_pd(iz1,jz0);
1489 dx11 = _mm256_sub_pd(ix1,jx1);
1490 dy11 = _mm256_sub_pd(iy1,jy1);
1491 dz11 = _mm256_sub_pd(iz1,jz1);
1492 dx12 = _mm256_sub_pd(ix1,jx2);
1493 dy12 = _mm256_sub_pd(iy1,jy2);
1494 dz12 = _mm256_sub_pd(iz1,jz2);
1495 dx20 = _mm256_sub_pd(ix2,jx0);
1496 dy20 = _mm256_sub_pd(iy2,jy0);
1497 dz20 = _mm256_sub_pd(iz2,jz0);
1498 dx21 = _mm256_sub_pd(ix2,jx1);
1499 dy21 = _mm256_sub_pd(iy2,jy1);
1500 dz21 = _mm256_sub_pd(iz2,jz1);
1501 dx22 = _mm256_sub_pd(ix2,jx2);
1502 dy22 = _mm256_sub_pd(iy2,jy2);
1503 dz22 = _mm256_sub_pd(iz2,jz2);
1505 /* Calculate squared distance and things based on it */
1506 rsq00 = gmx_mm256_calc_rsq_pd(dx00,dy00,dz00);
1507 rsq01 = gmx_mm256_calc_rsq_pd(dx01,dy01,dz01);
1508 rsq02 = gmx_mm256_calc_rsq_pd(dx02,dy02,dz02);
1509 rsq10 = gmx_mm256_calc_rsq_pd(dx10,dy10,dz10);
1510 rsq11 = gmx_mm256_calc_rsq_pd(dx11,dy11,dz11);
1511 rsq12 = gmx_mm256_calc_rsq_pd(dx12,dy12,dz12);
1512 rsq20 = gmx_mm256_calc_rsq_pd(dx20,dy20,dz20);
1513 rsq21 = gmx_mm256_calc_rsq_pd(dx21,dy21,dz21);
1514 rsq22 = gmx_mm256_calc_rsq_pd(dx22,dy22,dz22);
1516 rinv00 = gmx_mm256_invsqrt_pd(rsq00);
1517 rinv01 = gmx_mm256_invsqrt_pd(rsq01);
1518 rinv02 = gmx_mm256_invsqrt_pd(rsq02);
1519 rinv10 = gmx_mm256_invsqrt_pd(rsq10);
1520 rinv11 = gmx_mm256_invsqrt_pd(rsq11);
1521 rinv12 = gmx_mm256_invsqrt_pd(rsq12);
1522 rinv20 = gmx_mm256_invsqrt_pd(rsq20);
1523 rinv21 = gmx_mm256_invsqrt_pd(rsq21);
1524 rinv22 = gmx_mm256_invsqrt_pd(rsq22);
1526 rinvsq00 = _mm256_mul_pd(rinv00,rinv00);
1527 rinvsq01 = _mm256_mul_pd(rinv01,rinv01);
1528 rinvsq02 = _mm256_mul_pd(rinv02,rinv02);
1529 rinvsq10 = _mm256_mul_pd(rinv10,rinv10);
1530 rinvsq11 = _mm256_mul_pd(rinv11,rinv11);
1531 rinvsq12 = _mm256_mul_pd(rinv12,rinv12);
1532 rinvsq20 = _mm256_mul_pd(rinv20,rinv20);
1533 rinvsq21 = _mm256_mul_pd(rinv21,rinv21);
1534 rinvsq22 = _mm256_mul_pd(rinv22,rinv22);
1536 fjx0 = _mm256_setzero_pd();
1537 fjy0 = _mm256_setzero_pd();
1538 fjz0 = _mm256_setzero_pd();
1539 fjx1 = _mm256_setzero_pd();
1540 fjy1 = _mm256_setzero_pd();
1541 fjz1 = _mm256_setzero_pd();
1542 fjx2 = _mm256_setzero_pd();
1543 fjy2 = _mm256_setzero_pd();
1544 fjz2 = _mm256_setzero_pd();
1546 /**************************
1547 * CALCULATE INTERACTIONS *
1548 **************************/
1550 /* REACTION-FIELD ELECTROSTATICS */
1551 felec = _mm256_mul_pd(qq00,_mm256_sub_pd(_mm256_mul_pd(rinv00,rinvsq00),krf2));
1553 /* LENNARD-JONES DISPERSION/REPULSION */
1555 rinvsix = _mm256_mul_pd(_mm256_mul_pd(rinvsq00,rinvsq00),rinvsq00);
1556 fvdw = _mm256_mul_pd(_mm256_sub_pd(_mm256_mul_pd(c12_00,rinvsix),c6_00),_mm256_mul_pd(rinvsix,rinvsq00));
1558 fscal = _mm256_add_pd(felec,fvdw);
1560 fscal = _mm256_andnot_pd(dummy_mask,fscal);
1562 /* Calculate temporary vectorial force */
1563 tx = _mm256_mul_pd(fscal,dx00);
1564 ty = _mm256_mul_pd(fscal,dy00);
1565 tz = _mm256_mul_pd(fscal,dz00);
1567 /* Update vectorial force */
1568 fix0 = _mm256_add_pd(fix0,tx);
1569 fiy0 = _mm256_add_pd(fiy0,ty);
1570 fiz0 = _mm256_add_pd(fiz0,tz);
1572 fjx0 = _mm256_add_pd(fjx0,tx);
1573 fjy0 = _mm256_add_pd(fjy0,ty);
1574 fjz0 = _mm256_add_pd(fjz0,tz);
1576 /**************************
1577 * CALCULATE INTERACTIONS *
1578 **************************/
1580 /* REACTION-FIELD ELECTROSTATICS */
1581 felec = _mm256_mul_pd(qq01,_mm256_sub_pd(_mm256_mul_pd(rinv01,rinvsq01),krf2));
1585 fscal = _mm256_andnot_pd(dummy_mask,fscal);
1587 /* Calculate temporary vectorial force */
1588 tx = _mm256_mul_pd(fscal,dx01);
1589 ty = _mm256_mul_pd(fscal,dy01);
1590 tz = _mm256_mul_pd(fscal,dz01);
1592 /* Update vectorial force */
1593 fix0 = _mm256_add_pd(fix0,tx);
1594 fiy0 = _mm256_add_pd(fiy0,ty);
1595 fiz0 = _mm256_add_pd(fiz0,tz);
1597 fjx1 = _mm256_add_pd(fjx1,tx);
1598 fjy1 = _mm256_add_pd(fjy1,ty);
1599 fjz1 = _mm256_add_pd(fjz1,tz);
1601 /**************************
1602 * CALCULATE INTERACTIONS *
1603 **************************/
1605 /* REACTION-FIELD ELECTROSTATICS */
1606 felec = _mm256_mul_pd(qq02,_mm256_sub_pd(_mm256_mul_pd(rinv02,rinvsq02),krf2));
1610 fscal = _mm256_andnot_pd(dummy_mask,fscal);
1612 /* Calculate temporary vectorial force */
1613 tx = _mm256_mul_pd(fscal,dx02);
1614 ty = _mm256_mul_pd(fscal,dy02);
1615 tz = _mm256_mul_pd(fscal,dz02);
1617 /* Update vectorial force */
1618 fix0 = _mm256_add_pd(fix0,tx);
1619 fiy0 = _mm256_add_pd(fiy0,ty);
1620 fiz0 = _mm256_add_pd(fiz0,tz);
1622 fjx2 = _mm256_add_pd(fjx2,tx);
1623 fjy2 = _mm256_add_pd(fjy2,ty);
1624 fjz2 = _mm256_add_pd(fjz2,tz);
1626 /**************************
1627 * CALCULATE INTERACTIONS *
1628 **************************/
1630 /* REACTION-FIELD ELECTROSTATICS */
1631 felec = _mm256_mul_pd(qq10,_mm256_sub_pd(_mm256_mul_pd(rinv10,rinvsq10),krf2));
1635 fscal = _mm256_andnot_pd(dummy_mask,fscal);
1637 /* Calculate temporary vectorial force */
1638 tx = _mm256_mul_pd(fscal,dx10);
1639 ty = _mm256_mul_pd(fscal,dy10);
1640 tz = _mm256_mul_pd(fscal,dz10);
1642 /* Update vectorial force */
1643 fix1 = _mm256_add_pd(fix1,tx);
1644 fiy1 = _mm256_add_pd(fiy1,ty);
1645 fiz1 = _mm256_add_pd(fiz1,tz);
1647 fjx0 = _mm256_add_pd(fjx0,tx);
1648 fjy0 = _mm256_add_pd(fjy0,ty);
1649 fjz0 = _mm256_add_pd(fjz0,tz);
1651 /**************************
1652 * CALCULATE INTERACTIONS *
1653 **************************/
1655 /* REACTION-FIELD ELECTROSTATICS */
1656 felec = _mm256_mul_pd(qq11,_mm256_sub_pd(_mm256_mul_pd(rinv11,rinvsq11),krf2));
1660 fscal = _mm256_andnot_pd(dummy_mask,fscal);
1662 /* Calculate temporary vectorial force */
1663 tx = _mm256_mul_pd(fscal,dx11);
1664 ty = _mm256_mul_pd(fscal,dy11);
1665 tz = _mm256_mul_pd(fscal,dz11);
1667 /* Update vectorial force */
1668 fix1 = _mm256_add_pd(fix1,tx);
1669 fiy1 = _mm256_add_pd(fiy1,ty);
1670 fiz1 = _mm256_add_pd(fiz1,tz);
1672 fjx1 = _mm256_add_pd(fjx1,tx);
1673 fjy1 = _mm256_add_pd(fjy1,ty);
1674 fjz1 = _mm256_add_pd(fjz1,tz);
1676 /**************************
1677 * CALCULATE INTERACTIONS *
1678 **************************/
1680 /* REACTION-FIELD ELECTROSTATICS */
1681 felec = _mm256_mul_pd(qq12,_mm256_sub_pd(_mm256_mul_pd(rinv12,rinvsq12),krf2));
1685 fscal = _mm256_andnot_pd(dummy_mask,fscal);
1687 /* Calculate temporary vectorial force */
1688 tx = _mm256_mul_pd(fscal,dx12);
1689 ty = _mm256_mul_pd(fscal,dy12);
1690 tz = _mm256_mul_pd(fscal,dz12);
1692 /* Update vectorial force */
1693 fix1 = _mm256_add_pd(fix1,tx);
1694 fiy1 = _mm256_add_pd(fiy1,ty);
1695 fiz1 = _mm256_add_pd(fiz1,tz);
1697 fjx2 = _mm256_add_pd(fjx2,tx);
1698 fjy2 = _mm256_add_pd(fjy2,ty);
1699 fjz2 = _mm256_add_pd(fjz2,tz);
1701 /**************************
1702 * CALCULATE INTERACTIONS *
1703 **************************/
1705 /* REACTION-FIELD ELECTROSTATICS */
1706 felec = _mm256_mul_pd(qq20,_mm256_sub_pd(_mm256_mul_pd(rinv20,rinvsq20),krf2));
1710 fscal = _mm256_andnot_pd(dummy_mask,fscal);
1712 /* Calculate temporary vectorial force */
1713 tx = _mm256_mul_pd(fscal,dx20);
1714 ty = _mm256_mul_pd(fscal,dy20);
1715 tz = _mm256_mul_pd(fscal,dz20);
1717 /* Update vectorial force */
1718 fix2 = _mm256_add_pd(fix2,tx);
1719 fiy2 = _mm256_add_pd(fiy2,ty);
1720 fiz2 = _mm256_add_pd(fiz2,tz);
1722 fjx0 = _mm256_add_pd(fjx0,tx);
1723 fjy0 = _mm256_add_pd(fjy0,ty);
1724 fjz0 = _mm256_add_pd(fjz0,tz);
1726 /**************************
1727 * CALCULATE INTERACTIONS *
1728 **************************/
1730 /* REACTION-FIELD ELECTROSTATICS */
1731 felec = _mm256_mul_pd(qq21,_mm256_sub_pd(_mm256_mul_pd(rinv21,rinvsq21),krf2));
1735 fscal = _mm256_andnot_pd(dummy_mask,fscal);
1737 /* Calculate temporary vectorial force */
1738 tx = _mm256_mul_pd(fscal,dx21);
1739 ty = _mm256_mul_pd(fscal,dy21);
1740 tz = _mm256_mul_pd(fscal,dz21);
1742 /* Update vectorial force */
1743 fix2 = _mm256_add_pd(fix2,tx);
1744 fiy2 = _mm256_add_pd(fiy2,ty);
1745 fiz2 = _mm256_add_pd(fiz2,tz);
1747 fjx1 = _mm256_add_pd(fjx1,tx);
1748 fjy1 = _mm256_add_pd(fjy1,ty);
1749 fjz1 = _mm256_add_pd(fjz1,tz);
1751 /**************************
1752 * CALCULATE INTERACTIONS *
1753 **************************/
1755 /* REACTION-FIELD ELECTROSTATICS */
1756 felec = _mm256_mul_pd(qq22,_mm256_sub_pd(_mm256_mul_pd(rinv22,rinvsq22),krf2));
1760 fscal = _mm256_andnot_pd(dummy_mask,fscal);
1762 /* Calculate temporary vectorial force */
1763 tx = _mm256_mul_pd(fscal,dx22);
1764 ty = _mm256_mul_pd(fscal,dy22);
1765 tz = _mm256_mul_pd(fscal,dz22);
1767 /* Update vectorial force */
1768 fix2 = _mm256_add_pd(fix2,tx);
1769 fiy2 = _mm256_add_pd(fiy2,ty);
1770 fiz2 = _mm256_add_pd(fiz2,tz);
1772 fjx2 = _mm256_add_pd(fjx2,tx);
1773 fjy2 = _mm256_add_pd(fjy2,ty);
1774 fjz2 = _mm256_add_pd(fjz2,tz);
1776 fjptrA = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
1777 fjptrB = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
1778 fjptrC = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
1779 fjptrD = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
1781 gmx_mm256_decrement_3rvec_4ptr_swizzle_pd(fjptrA,fjptrB,fjptrC,fjptrD,
1782 fjx0,fjy0,fjz0,fjx1,fjy1,fjz1,fjx2,fjy2,fjz2);
1784 /* Inner loop uses 250 flops */
1787 /* End of innermost loop */
1789 gmx_mm256_update_iforce_3atom_swizzle_pd(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,
1790 f+i_coord_offset,fshift+i_shift_offset);
1792 /* Increment number of inner iterations */
1793 inneriter += j_index_end - j_index_start;
1795 /* Outer loop uses 18 flops */
1798 /* Increment number of outer iterations */
1801 /* Update outer/inner flops */
1803 inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W3W3_F,outeriter*18 + inneriter*250);