?  search:
summary | log | graphiclog1 | graphiclog2 | commit | commitdiff | tree | refs | edit | fork
blame | history | raw | HEAD
Remove nb-parameters from t_forcerec
[gromacs.git] / src / gromacs / gmxlib / nonbonded / nb_kernel_avx_256_single / nb_kernel_ElecEwSh_VdwLJSh_GeomW4P1_avx_256_single.c
blob2d098d311be2a370ea7e94c15e501896f2d4d35a
1 /*
2 * This file is part of the GROMACS molecular simulation package.
3 *
4 * Copyright (c) 2012,2013,2014,2015,2017, 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.
8 *
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.
13 *
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.
18 *
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.
23 *
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.
31 *
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.
34 */
35 /*
36 * Note: this file was generated by the GROMACS avx_256_single kernel generator.
37 */
38 #include "gmxpre.h"
39
40 #include "config.h"
41
42 #include <math.h>
43
44 #include "../nb_kernel.h"
45 #include "gromacs/gmxlib/nrnb.h"
46
47 #include "kernelutil_x86_avx_256_single.h"
48
49 /*
50 * Gromacs nonbonded kernel: nb_kernel_ElecEwSh_VdwLJSh_GeomW4P1_VF_avx_256_single
51 * Electrostatics interaction: Ewald
52 * VdW interaction: LennardJones
53 * Geometry: Water4-Particle
54 * Calculate force/pot: PotentialAndForce
55 */
56 void
57 nb_kernel_ElecEwSh_VdwLJSh_GeomW4P1_VF_avx_256_single
58 (t_nblist * gmx_restrict nlist,
59 rvec * gmx_restrict xx,
60 rvec * gmx_restrict ff,
61 struct t_forcerec * gmx_restrict fr,
62 t_mdatoms * gmx_restrict mdatoms,
63 nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
64 t_nrnb * gmx_restrict nrnb)
65 {
66 /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
67 * just 0 for non-waters.
68 * Suffixes A,B,C,D,E,F,G,H refer to j loop unrolling done with AVX, e.g. for the eight different
69 * jnr indices corresponding to data put in the four positions in the SIMD register.
70 */
71 int i_shift_offset,i_coord_offset,outeriter,inneriter;
72 int j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
73 int jnrA,jnrB,jnrC,jnrD;
74 int jnrE,jnrF,jnrG,jnrH;
75 int jnrlistA,jnrlistB,jnrlistC,jnrlistD;
76 int jnrlistE,jnrlistF,jnrlistG,jnrlistH;
77 int j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
78 int j_coord_offsetE,j_coord_offsetF,j_coord_offsetG,j_coord_offsetH;
79 int *iinr,*jindex,*jjnr,*shiftidx,*gid;
80 real rcutoff_scalar;
81 real *shiftvec,*fshift,*x,*f;
82 real *fjptrA,*fjptrB,*fjptrC,*fjptrD,*fjptrE,*fjptrF,*fjptrG,*fjptrH;
83 real scratch[4*DIM];
84 __m256 tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
85 real * vdwioffsetptr0;
86 __m256 ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
87 real * vdwioffsetptr1;
88 __m256 ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
89 real * vdwioffsetptr2;
90 __m256 ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
91 real * vdwioffsetptr3;
92 __m256 ix3,iy3,iz3,fix3,fiy3,fiz3,iq3,isai3;
93 int vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D,vdwjidx0E,vdwjidx0F,vdwjidx0G,vdwjidx0H;
94 __m256 jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
95 __m256 dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
96 __m256 dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
97 __m256 dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
98 __m256 dx30,dy30,dz30,rsq30,rinv30,rinvsq30,r30,qq30,c6_30,c12_30;
99 __m256 velec,felec,velecsum,facel,crf,krf,krf2;
100 real *charge;
101 int nvdwtype;
102 __m256 rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
103 int *vdwtype;
104 real *vdwparam;
105 __m256 one_sixth = _mm256_set1_ps(1.0/6.0);
106 __m256 one_twelfth = _mm256_set1_ps(1.0/12.0);
107 __m256i ewitab;
108 __m128i ewitab_lo,ewitab_hi;
109 __m256 ewtabscale,eweps,sh_ewald,ewrt,ewtabhalfspace,ewtabF,ewtabFn,ewtabD,ewtabV;
110 __m256 beta,beta2,beta3,zeta2,pmecorrF,pmecorrV,rinv3;
111 real *ewtab;
112 __m256 dummy_mask,cutoff_mask;
113 __m256 signbit = _mm256_castsi256_ps( _mm256_set1_epi32(0x80000000) );
114 __m256 one = _mm256_set1_ps(1.0);
115 __m256 two = _mm256_set1_ps(2.0);
116 x = xx[0];
117 f = ff[0];
118
119 nri = nlist->nri;
120 iinr = nlist->iinr;
121 jindex = nlist->jindex;
122 jjnr = nlist->jjnr;
123 shiftidx = nlist->shift;
124 gid = nlist->gid;
125 shiftvec = fr->shift_vec[0];
126 fshift = fr->fshift[0];
127 facel = _mm256_set1_ps(fr->ic->epsfac);
128 charge = mdatoms->chargeA;
129 nvdwtype = fr->ntype;
130 vdwparam = fr->nbfp;
131 vdwtype = mdatoms->typeA;
132
133 sh_ewald = _mm256_set1_ps(fr->ic->sh_ewald);
134 beta = _mm256_set1_ps(fr->ic->ewaldcoeff_q);
135 beta2 = _mm256_mul_ps(beta,beta);
136 beta3 = _mm256_mul_ps(beta,beta2);
137
138 ewtab = fr->ic->tabq_coul_FDV0;
139 ewtabscale = _mm256_set1_ps(fr->ic->tabq_scale);
140 ewtabhalfspace = _mm256_set1_ps(0.5/fr->ic->tabq_scale);
141
142 /* Setup water-specific parameters */
143 inr = nlist->iinr[0];
144 iq1 = _mm256_mul_ps(facel,_mm256_set1_ps(charge[inr+1]));
145 iq2 = _mm256_mul_ps(facel,_mm256_set1_ps(charge[inr+2]));
146 iq3 = _mm256_mul_ps(facel,_mm256_set1_ps(charge[inr+3]));
147 vdwioffsetptr0 = vdwparam+2*nvdwtype*vdwtype[inr+0];
148
149 /* When we use explicit cutoffs the value must be identical for elec and VdW, so use elec as an arbitrary choice */
150 rcutoff_scalar = fr->ic->rcoulomb;
151 rcutoff = _mm256_set1_ps(rcutoff_scalar);
152 rcutoff2 = _mm256_mul_ps(rcutoff,rcutoff);
153
154 sh_vdw_invrcut6 = _mm256_set1_ps(fr->ic->sh_invrc6);
155 rvdw = _mm256_set1_ps(fr->ic->rvdw);
156
157 /* Avoid stupid compiler warnings */
158 jnrA = jnrB = jnrC = jnrD = jnrE = jnrF = jnrG = jnrH = 0;
159 j_coord_offsetA = 0;
160 j_coord_offsetB = 0;
161 j_coord_offsetC = 0;
162 j_coord_offsetD = 0;
163 j_coord_offsetE = 0;
164 j_coord_offsetF = 0;
165 j_coord_offsetG = 0;
166 j_coord_offsetH = 0;
167
168 outeriter = 0;
169 inneriter = 0;
170
171 for(iidx=0;iidx<4*DIM;iidx++)
172 {
173 scratch[iidx] = 0.0;
174 }
175
176 /* Start outer loop over neighborlists */
177 for(iidx=0; iidx<nri; iidx++)
178 {
179 /* Load shift vector for this list */
180 i_shift_offset = DIM*shiftidx[iidx];
181
182 /* Load limits for loop over neighbors */
183 j_index_start = jindex[iidx];
184 j_index_end = jindex[iidx+1];
185
186 /* Get outer coordinate index */
187 inr = iinr[iidx];
188 i_coord_offset = DIM*inr;
189
190 /* Load i particle coords and add shift vector */
191 gmx_mm256_load_shift_and_4rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,
192 &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2,&ix3,&iy3,&iz3);
193
194 fix0 = _mm256_setzero_ps();
195 fiy0 = _mm256_setzero_ps();
196 fiz0 = _mm256_setzero_ps();
197 fix1 = _mm256_setzero_ps();
198 fiy1 = _mm256_setzero_ps();
199 fiz1 = _mm256_setzero_ps();
200 fix2 = _mm256_setzero_ps();
201 fiy2 = _mm256_setzero_ps();
202 fiz2 = _mm256_setzero_ps();
203 fix3 = _mm256_setzero_ps();
204 fiy3 = _mm256_setzero_ps();
205 fiz3 = _mm256_setzero_ps();
206
207 /* Reset potential sums */
208 velecsum = _mm256_setzero_ps();
209 vvdwsum = _mm256_setzero_ps();
210
211 /* Start inner kernel loop */
212 for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+7]>=0; jidx+=8)
213 {
214
215 /* Get j neighbor index, and coordinate index */
216 jnrA = jjnr[jidx];
217 jnrB = jjnr[jidx+1];
218 jnrC = jjnr[jidx+2];
219 jnrD = jjnr[jidx+3];
220 jnrE = jjnr[jidx+4];
221 jnrF = jjnr[jidx+5];
222 jnrG = jjnr[jidx+6];
223 jnrH = jjnr[jidx+7];
224 j_coord_offsetA = DIM*jnrA;
225 j_coord_offsetB = DIM*jnrB;
226 j_coord_offsetC = DIM*jnrC;
227 j_coord_offsetD = DIM*jnrD;
228 j_coord_offsetE = DIM*jnrE;
229 j_coord_offsetF = DIM*jnrF;
230 j_coord_offsetG = DIM*jnrG;
231 j_coord_offsetH = DIM*jnrH;
232
233 /* load j atom coordinates */
234 gmx_mm256_load_1rvec_8ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
235 x+j_coord_offsetC,x+j_coord_offsetD,
236 x+j_coord_offsetE,x+j_coord_offsetF,
237 x+j_coord_offsetG,x+j_coord_offsetH,
238 &jx0,&jy0,&jz0);
239
240 /* Calculate displacement vector */
241 dx00 = _mm256_sub_ps(ix0,jx0);
242 dy00 = _mm256_sub_ps(iy0,jy0);
243 dz00 = _mm256_sub_ps(iz0,jz0);
244 dx10 = _mm256_sub_ps(ix1,jx0);
245 dy10 = _mm256_sub_ps(iy1,jy0);
246 dz10 = _mm256_sub_ps(iz1,jz0);
247 dx20 = _mm256_sub_ps(ix2,jx0);
248 dy20 = _mm256_sub_ps(iy2,jy0);
249 dz20 = _mm256_sub_ps(iz2,jz0);
250 dx30 = _mm256_sub_ps(ix3,jx0);
251 dy30 = _mm256_sub_ps(iy3,jy0);
252 dz30 = _mm256_sub_ps(iz3,jz0);
253
254 /* Calculate squared distance and things based on it */
255 rsq00 = gmx_mm256_calc_rsq_ps(dx00,dy00,dz00);
256 rsq10 = gmx_mm256_calc_rsq_ps(dx10,dy10,dz10);
257 rsq20 = gmx_mm256_calc_rsq_ps(dx20,dy20,dz20);
258 rsq30 = gmx_mm256_calc_rsq_ps(dx30,dy30,dz30);
259
260 rinv10 = avx256_invsqrt_f(rsq10);
261 rinv20 = avx256_invsqrt_f(rsq20);
262 rinv30 = avx256_invsqrt_f(rsq30);
263
264 rinvsq00 = avx256_inv_f(rsq00);
265 rinvsq10 = _mm256_mul_ps(rinv10,rinv10);
266 rinvsq20 = _mm256_mul_ps(rinv20,rinv20);
267 rinvsq30 = _mm256_mul_ps(rinv30,rinv30);
268
269 /* Load parameters for j particles */
270 jq0 = gmx_mm256_load_8real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
271 charge+jnrC+0,charge+jnrD+0,
272 charge+jnrE+0,charge+jnrF+0,
273 charge+jnrG+0,charge+jnrH+0);
274 vdwjidx0A = 2*vdwtype[jnrA+0];
275 vdwjidx0B = 2*vdwtype[jnrB+0];
276 vdwjidx0C = 2*vdwtype[jnrC+0];
277 vdwjidx0D = 2*vdwtype[jnrD+0];
278 vdwjidx0E = 2*vdwtype[jnrE+0];
279 vdwjidx0F = 2*vdwtype[jnrF+0];
280 vdwjidx0G = 2*vdwtype[jnrG+0];
281 vdwjidx0H = 2*vdwtype[jnrH+0];
282
283 fjx0 = _mm256_setzero_ps();
284 fjy0 = _mm256_setzero_ps();
285 fjz0 = _mm256_setzero_ps();
286
287 /**************************
288 * CALCULATE INTERACTIONS *
289 **************************/
290
291 if (gmx_mm256_any_lt(rsq00,rcutoff2))
292 {
293
294 /* Compute parameters for interactions between i and j atoms */
295 gmx_mm256_load_8pair_swizzle_ps(vdwioffsetptr0+vdwjidx0A,
296 vdwioffsetptr0+vdwjidx0B,
297 vdwioffsetptr0+vdwjidx0C,
298 vdwioffsetptr0+vdwjidx0D,
299 vdwioffsetptr0+vdwjidx0E,
300 vdwioffsetptr0+vdwjidx0F,
301 vdwioffsetptr0+vdwjidx0G,
302 vdwioffsetptr0+vdwjidx0H,
303 &c6_00,&c12_00);
304
305 /* LENNARD-JONES DISPERSION/REPULSION */
306
307 rinvsix = _mm256_mul_ps(_mm256_mul_ps(rinvsq00,rinvsq00),rinvsq00);
308 vvdw6 = _mm256_mul_ps(c6_00,rinvsix);
309 vvdw12 = _mm256_mul_ps(c12_00,_mm256_mul_ps(rinvsix,rinvsix));
310 vvdw = _mm256_sub_ps(_mm256_mul_ps( _mm256_sub_ps(vvdw12 , _mm256_mul_ps(c12_00,_mm256_mul_ps(sh_vdw_invrcut6,sh_vdw_invrcut6))), one_twelfth) ,
311 _mm256_mul_ps( _mm256_sub_ps(vvdw6,_mm256_mul_ps(c6_00,sh_vdw_invrcut6)),one_sixth));
312 fvdw = _mm256_mul_ps(_mm256_sub_ps(vvdw12,vvdw6),rinvsq00);
313
314 cutoff_mask = _mm256_cmp_ps(rsq00,rcutoff2,_CMP_LT_OQ);
315
316 /* Update potential sum for this i atom from the interaction with this j atom. */
317 vvdw = _mm256_and_ps(vvdw,cutoff_mask);
318 vvdwsum = _mm256_add_ps(vvdwsum,vvdw);
319
320 fscal = fvdw;
321
322 fscal = _mm256_and_ps(fscal,cutoff_mask);
323
324 /* Calculate temporary vectorial force */
325 tx = _mm256_mul_ps(fscal,dx00);
326 ty = _mm256_mul_ps(fscal,dy00);
327 tz = _mm256_mul_ps(fscal,dz00);
328
329 /* Update vectorial force */
330 fix0 = _mm256_add_ps(fix0,tx);
331 fiy0 = _mm256_add_ps(fiy0,ty);
332 fiz0 = _mm256_add_ps(fiz0,tz);
333
334 fjx0 = _mm256_add_ps(fjx0,tx);
335 fjy0 = _mm256_add_ps(fjy0,ty);
336 fjz0 = _mm256_add_ps(fjz0,tz);
337
338 }
339
340 /**************************
341 * CALCULATE INTERACTIONS *
342 **************************/
343
344 if (gmx_mm256_any_lt(rsq10,rcutoff2))
345 {
346
347 r10 = _mm256_mul_ps(rsq10,rinv10);
348
349 /* Compute parameters for interactions between i and j atoms */
350 qq10 = _mm256_mul_ps(iq1,jq0);
351
352 /* EWALD ELECTROSTATICS */
353
354 /* Analytical PME correction */
355 zeta2 = _mm256_mul_ps(beta2,rsq10);
356 rinv3 = _mm256_mul_ps(rinvsq10,rinv10);
357 pmecorrF = avx256_pmecorrF_f(zeta2);
358 felec = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
359 felec = _mm256_mul_ps(qq10,felec);
360 pmecorrV = avx256_pmecorrV_f(zeta2);
361 pmecorrV = _mm256_mul_ps(pmecorrV,beta);
362 velec = _mm256_sub_ps(_mm256_sub_ps(rinv10,sh_ewald),pmecorrV);
363 velec = _mm256_mul_ps(qq10,velec);
364
365 cutoff_mask = _mm256_cmp_ps(rsq10,rcutoff2,_CMP_LT_OQ);
366
367 /* Update potential sum for this i atom from the interaction with this j atom. */
368 velec = _mm256_and_ps(velec,cutoff_mask);
369 velecsum = _mm256_add_ps(velecsum,velec);
370
371 fscal = felec;
372
373 fscal = _mm256_and_ps(fscal,cutoff_mask);
374
375 /* Calculate temporary vectorial force */
376 tx = _mm256_mul_ps(fscal,dx10);
377 ty = _mm256_mul_ps(fscal,dy10);
378 tz = _mm256_mul_ps(fscal,dz10);
379
380 /* Update vectorial force */
381 fix1 = _mm256_add_ps(fix1,tx);
382 fiy1 = _mm256_add_ps(fiy1,ty);
383 fiz1 = _mm256_add_ps(fiz1,tz);
384
385 fjx0 = _mm256_add_ps(fjx0,tx);
386 fjy0 = _mm256_add_ps(fjy0,ty);
387 fjz0 = _mm256_add_ps(fjz0,tz);
388
389 }
390
391 /**************************
392 * CALCULATE INTERACTIONS *
393 **************************/
394
395 if (gmx_mm256_any_lt(rsq20,rcutoff2))
396 {
397
398 r20 = _mm256_mul_ps(rsq20,rinv20);
399
400 /* Compute parameters for interactions between i and j atoms */
401 qq20 = _mm256_mul_ps(iq2,jq0);
402
403 /* EWALD ELECTROSTATICS */
404
405 /* Analytical PME correction */
406 zeta2 = _mm256_mul_ps(beta2,rsq20);
407 rinv3 = _mm256_mul_ps(rinvsq20,rinv20);
408 pmecorrF = avx256_pmecorrF_f(zeta2);
409 felec = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
410 felec = _mm256_mul_ps(qq20,felec);
411 pmecorrV = avx256_pmecorrV_f(zeta2);
412 pmecorrV = _mm256_mul_ps(pmecorrV,beta);
413 velec = _mm256_sub_ps(_mm256_sub_ps(rinv20,sh_ewald),pmecorrV);
414 velec = _mm256_mul_ps(qq20,velec);
415
416 cutoff_mask = _mm256_cmp_ps(rsq20,rcutoff2,_CMP_LT_OQ);
417
418 /* Update potential sum for this i atom from the interaction with this j atom. */
419 velec = _mm256_and_ps(velec,cutoff_mask);
420 velecsum = _mm256_add_ps(velecsum,velec);
421
422 fscal = felec;
423
424 fscal = _mm256_and_ps(fscal,cutoff_mask);
425
426 /* Calculate temporary vectorial force */
427 tx = _mm256_mul_ps(fscal,dx20);
428 ty = _mm256_mul_ps(fscal,dy20);
429 tz = _mm256_mul_ps(fscal,dz20);
430
431 /* Update vectorial force */
432 fix2 = _mm256_add_ps(fix2,tx);
433 fiy2 = _mm256_add_ps(fiy2,ty);
434 fiz2 = _mm256_add_ps(fiz2,tz);
435
436 fjx0 = _mm256_add_ps(fjx0,tx);
437 fjy0 = _mm256_add_ps(fjy0,ty);
438 fjz0 = _mm256_add_ps(fjz0,tz);
439
440 }
441
442 /**************************
443 * CALCULATE INTERACTIONS *
444 **************************/
445
446 if (gmx_mm256_any_lt(rsq30,rcutoff2))
447 {
448
449 r30 = _mm256_mul_ps(rsq30,rinv30);
450
451 /* Compute parameters for interactions between i and j atoms */
452 qq30 = _mm256_mul_ps(iq3,jq0);
453
454 /* EWALD ELECTROSTATICS */
455
456 /* Analytical PME correction */
457 zeta2 = _mm256_mul_ps(beta2,rsq30);
458 rinv3 = _mm256_mul_ps(rinvsq30,rinv30);
459 pmecorrF = avx256_pmecorrF_f(zeta2);
460 felec = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
461 felec = _mm256_mul_ps(qq30,felec);
462 pmecorrV = avx256_pmecorrV_f(zeta2);
463 pmecorrV = _mm256_mul_ps(pmecorrV,beta);
464 velec = _mm256_sub_ps(_mm256_sub_ps(rinv30,sh_ewald),pmecorrV);
465 velec = _mm256_mul_ps(qq30,velec);
466
467 cutoff_mask = _mm256_cmp_ps(rsq30,rcutoff2,_CMP_LT_OQ);
468
469 /* Update potential sum for this i atom from the interaction with this j atom. */
470 velec = _mm256_and_ps(velec,cutoff_mask);
471 velecsum = _mm256_add_ps(velecsum,velec);
472
473 fscal = felec;
474
475 fscal = _mm256_and_ps(fscal,cutoff_mask);
476
477 /* Calculate temporary vectorial force */
478 tx = _mm256_mul_ps(fscal,dx30);
479 ty = _mm256_mul_ps(fscal,dy30);
480 tz = _mm256_mul_ps(fscal,dz30);
481
482 /* Update vectorial force */
483 fix3 = _mm256_add_ps(fix3,tx);
484 fiy3 = _mm256_add_ps(fiy3,ty);
485 fiz3 = _mm256_add_ps(fiz3,tz);
486
487 fjx0 = _mm256_add_ps(fjx0,tx);
488 fjy0 = _mm256_add_ps(fjy0,ty);
489 fjz0 = _mm256_add_ps(fjz0,tz);
490
491 }
492
493 fjptrA = f+j_coord_offsetA;
494 fjptrB = f+j_coord_offsetB;
495 fjptrC = f+j_coord_offsetC;
496 fjptrD = f+j_coord_offsetD;
497 fjptrE = f+j_coord_offsetE;
498 fjptrF = f+j_coord_offsetF;
499 fjptrG = f+j_coord_offsetG;
500 fjptrH = f+j_coord_offsetH;
501
502 gmx_mm256_decrement_1rvec_8ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjptrE,fjptrF,fjptrG,fjptrH,fjx0,fjy0,fjz0);
503
504 /* Inner loop uses 371 flops */
505 }
506
507 if(jidx<j_index_end)
508 {
509
510 /* Get j neighbor index, and coordinate index */
511 jnrlistA = jjnr[jidx];
512 jnrlistB = jjnr[jidx+1];
513 jnrlistC = jjnr[jidx+2];
514 jnrlistD = jjnr[jidx+3];
515 jnrlistE = jjnr[jidx+4];
516 jnrlistF = jjnr[jidx+5];
517 jnrlistG = jjnr[jidx+6];
518 jnrlistH = jjnr[jidx+7];
519 /* Sign of each element will be negative for non-real atoms.
520 * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
521 * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
522 */
523 dummy_mask = gmx_mm256_set_m128(gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx+4)),_mm_setzero_si128())),
524 gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128())));
525
526 jnrA = (jnrlistA>=0) ? jnrlistA : 0;
527 jnrB = (jnrlistB>=0) ? jnrlistB : 0;
528 jnrC = (jnrlistC>=0) ? jnrlistC : 0;
529 jnrD = (jnrlistD>=0) ? jnrlistD : 0;
530 jnrE = (jnrlistE>=0) ? jnrlistE : 0;
531 jnrF = (jnrlistF>=0) ? jnrlistF : 0;
532 jnrG = (jnrlistG>=0) ? jnrlistG : 0;
533 jnrH = (jnrlistH>=0) ? jnrlistH : 0;
534 j_coord_offsetA = DIM*jnrA;
535 j_coord_offsetB = DIM*jnrB;
536 j_coord_offsetC = DIM*jnrC;
537 j_coord_offsetD = DIM*jnrD;
538 j_coord_offsetE = DIM*jnrE;
539 j_coord_offsetF = DIM*jnrF;
540 j_coord_offsetG = DIM*jnrG;
541 j_coord_offsetH = DIM*jnrH;
542
543 /* load j atom coordinates */
544 gmx_mm256_load_1rvec_8ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
545 x+j_coord_offsetC,x+j_coord_offsetD,
546 x+j_coord_offsetE,x+j_coord_offsetF,
547 x+j_coord_offsetG,x+j_coord_offsetH,
548 &jx0,&jy0,&jz0);
549
550 /* Calculate displacement vector */
551 dx00 = _mm256_sub_ps(ix0,jx0);
552 dy00 = _mm256_sub_ps(iy0,jy0);
553 dz00 = _mm256_sub_ps(iz0,jz0);
554 dx10 = _mm256_sub_ps(ix1,jx0);
555 dy10 = _mm256_sub_ps(iy1,jy0);
556 dz10 = _mm256_sub_ps(iz1,jz0);
557 dx20 = _mm256_sub_ps(ix2,jx0);
558 dy20 = _mm256_sub_ps(iy2,jy0);
559 dz20 = _mm256_sub_ps(iz2,jz0);
560 dx30 = _mm256_sub_ps(ix3,jx0);
561 dy30 = _mm256_sub_ps(iy3,jy0);
562 dz30 = _mm256_sub_ps(iz3,jz0);
563
564 /* Calculate squared distance and things based on it */
565 rsq00 = gmx_mm256_calc_rsq_ps(dx00,dy00,dz00);
566 rsq10 = gmx_mm256_calc_rsq_ps(dx10,dy10,dz10);
567 rsq20 = gmx_mm256_calc_rsq_ps(dx20,dy20,dz20);
568 rsq30 = gmx_mm256_calc_rsq_ps(dx30,dy30,dz30);
569
570 rinv10 = avx256_invsqrt_f(rsq10);
571 rinv20 = avx256_invsqrt_f(rsq20);
572 rinv30 = avx256_invsqrt_f(rsq30);
573
574 rinvsq00 = avx256_inv_f(rsq00);
575 rinvsq10 = _mm256_mul_ps(rinv10,rinv10);
576 rinvsq20 = _mm256_mul_ps(rinv20,rinv20);
577 rinvsq30 = _mm256_mul_ps(rinv30,rinv30);
578
579 /* Load parameters for j particles */
580 jq0 = gmx_mm256_load_8real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
581 charge+jnrC+0,charge+jnrD+0,
582 charge+jnrE+0,charge+jnrF+0,
583 charge+jnrG+0,charge+jnrH+0);
584 vdwjidx0A = 2*vdwtype[jnrA+0];
585 vdwjidx0B = 2*vdwtype[jnrB+0];
586 vdwjidx0C = 2*vdwtype[jnrC+0];
587 vdwjidx0D = 2*vdwtype[jnrD+0];
588 vdwjidx0E = 2*vdwtype[jnrE+0];
589 vdwjidx0F = 2*vdwtype[jnrF+0];
590 vdwjidx0G = 2*vdwtype[jnrG+0];
591 vdwjidx0H = 2*vdwtype[jnrH+0];
592
593 fjx0 = _mm256_setzero_ps();
594 fjy0 = _mm256_setzero_ps();
595 fjz0 = _mm256_setzero_ps();
596
597 /**************************
598 * CALCULATE INTERACTIONS *
599 **************************/
600
601 if (gmx_mm256_any_lt(rsq00,rcutoff2))
602 {
603
604 /* Compute parameters for interactions between i and j atoms */
605 gmx_mm256_load_8pair_swizzle_ps(vdwioffsetptr0+vdwjidx0A,
606 vdwioffsetptr0+vdwjidx0B,
607 vdwioffsetptr0+vdwjidx0C,
608 vdwioffsetptr0+vdwjidx0D,
609 vdwioffsetptr0+vdwjidx0E,
610 vdwioffsetptr0+vdwjidx0F,
611 vdwioffsetptr0+vdwjidx0G,
612 vdwioffsetptr0+vdwjidx0H,
613 &c6_00,&c12_00);
614
615 /* LENNARD-JONES DISPERSION/REPULSION */
616
617 rinvsix = _mm256_mul_ps(_mm256_mul_ps(rinvsq00,rinvsq00),rinvsq00);
618 vvdw6 = _mm256_mul_ps(c6_00,rinvsix);
619 vvdw12 = _mm256_mul_ps(c12_00,_mm256_mul_ps(rinvsix,rinvsix));
620 vvdw = _mm256_sub_ps(_mm256_mul_ps( _mm256_sub_ps(vvdw12 , _mm256_mul_ps(c12_00,_mm256_mul_ps(sh_vdw_invrcut6,sh_vdw_invrcut6))), one_twelfth) ,
621 _mm256_mul_ps( _mm256_sub_ps(vvdw6,_mm256_mul_ps(c6_00,sh_vdw_invrcut6)),one_sixth));
622 fvdw = _mm256_mul_ps(_mm256_sub_ps(vvdw12,vvdw6),rinvsq00);
623
624 cutoff_mask = _mm256_cmp_ps(rsq00,rcutoff2,_CMP_LT_OQ);
625
626 /* Update potential sum for this i atom from the interaction with this j atom. */
627 vvdw = _mm256_and_ps(vvdw,cutoff_mask);
628 vvdw = _mm256_andnot_ps(dummy_mask,vvdw);
629 vvdwsum = _mm256_add_ps(vvdwsum,vvdw);
630
631 fscal = fvdw;
632
633 fscal = _mm256_and_ps(fscal,cutoff_mask);
634
635 fscal = _mm256_andnot_ps(dummy_mask,fscal);
636
637 /* Calculate temporary vectorial force */
638 tx = _mm256_mul_ps(fscal,dx00);
639 ty = _mm256_mul_ps(fscal,dy00);
640 tz = _mm256_mul_ps(fscal,dz00);
641
642 /* Update vectorial force */
643 fix0 = _mm256_add_ps(fix0,tx);
644 fiy0 = _mm256_add_ps(fiy0,ty);
645 fiz0 = _mm256_add_ps(fiz0,tz);
646
647 fjx0 = _mm256_add_ps(fjx0,tx);
648 fjy0 = _mm256_add_ps(fjy0,ty);
649 fjz0 = _mm256_add_ps(fjz0,tz);
650
651 }
652
653 /**************************
654 * CALCULATE INTERACTIONS *
655 **************************/
656
657 if (gmx_mm256_any_lt(rsq10,rcutoff2))
658 {
659
660 r10 = _mm256_mul_ps(rsq10,rinv10);
661 r10 = _mm256_andnot_ps(dummy_mask,r10);
662
663 /* Compute parameters for interactions between i and j atoms */
664 qq10 = _mm256_mul_ps(iq1,jq0);
665
666 /* EWALD ELECTROSTATICS */
667
668 /* Analytical PME correction */
669 zeta2 = _mm256_mul_ps(beta2,rsq10);
670 rinv3 = _mm256_mul_ps(rinvsq10,rinv10);
671 pmecorrF = avx256_pmecorrF_f(zeta2);
672 felec = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
673 felec = _mm256_mul_ps(qq10,felec);
674 pmecorrV = avx256_pmecorrV_f(zeta2);
675 pmecorrV = _mm256_mul_ps(pmecorrV,beta);
676 velec = _mm256_sub_ps(_mm256_sub_ps(rinv10,sh_ewald),pmecorrV);
677 velec = _mm256_mul_ps(qq10,velec);
678
679 cutoff_mask = _mm256_cmp_ps(rsq10,rcutoff2,_CMP_LT_OQ);
680
681 /* Update potential sum for this i atom from the interaction with this j atom. */
682 velec = _mm256_and_ps(velec,cutoff_mask);
683 velec = _mm256_andnot_ps(dummy_mask,velec);
684 velecsum = _mm256_add_ps(velecsum,velec);
685
686 fscal = felec;
687
688 fscal = _mm256_and_ps(fscal,cutoff_mask);
689
690 fscal = _mm256_andnot_ps(dummy_mask,fscal);
691
692 /* Calculate temporary vectorial force */
693 tx = _mm256_mul_ps(fscal,dx10);
694 ty = _mm256_mul_ps(fscal,dy10);
695 tz = _mm256_mul_ps(fscal,dz10);
696
697 /* Update vectorial force */
698 fix1 = _mm256_add_ps(fix1,tx);
699 fiy1 = _mm256_add_ps(fiy1,ty);
700 fiz1 = _mm256_add_ps(fiz1,tz);
701
702 fjx0 = _mm256_add_ps(fjx0,tx);
703 fjy0 = _mm256_add_ps(fjy0,ty);
704 fjz0 = _mm256_add_ps(fjz0,tz);
705
706 }
707
708 /**************************
709 * CALCULATE INTERACTIONS *
710 **************************/
711
712 if (gmx_mm256_any_lt(rsq20,rcutoff2))
713 {
714
715 r20 = _mm256_mul_ps(rsq20,rinv20);
716 r20 = _mm256_andnot_ps(dummy_mask,r20);
717
718 /* Compute parameters for interactions between i and j atoms */
719 qq20 = _mm256_mul_ps(iq2,jq0);
720
721 /* EWALD ELECTROSTATICS */
722
723 /* Analytical PME correction */
724 zeta2 = _mm256_mul_ps(beta2,rsq20);
725 rinv3 = _mm256_mul_ps(rinvsq20,rinv20);
726 pmecorrF = avx256_pmecorrF_f(zeta2);
727 felec = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
728 felec = _mm256_mul_ps(qq20,felec);
729 pmecorrV = avx256_pmecorrV_f(zeta2);
730 pmecorrV = _mm256_mul_ps(pmecorrV,beta);
731 velec = _mm256_sub_ps(_mm256_sub_ps(rinv20,sh_ewald),pmecorrV);
732 velec = _mm256_mul_ps(qq20,velec);
733
734 cutoff_mask = _mm256_cmp_ps(rsq20,rcutoff2,_CMP_LT_OQ);
735
736 /* Update potential sum for this i atom from the interaction with this j atom. */
737 velec = _mm256_and_ps(velec,cutoff_mask);
738 velec = _mm256_andnot_ps(dummy_mask,velec);
739 velecsum = _mm256_add_ps(velecsum,velec);
740
741 fscal = felec;
742
743 fscal = _mm256_and_ps(fscal,cutoff_mask);
744
745 fscal = _mm256_andnot_ps(dummy_mask,fscal);
746
747 /* Calculate temporary vectorial force */
748 tx = _mm256_mul_ps(fscal,dx20);
749 ty = _mm256_mul_ps(fscal,dy20);
750 tz = _mm256_mul_ps(fscal,dz20);
751
752 /* Update vectorial force */
753 fix2 = _mm256_add_ps(fix2,tx);
754 fiy2 = _mm256_add_ps(fiy2,ty);
755 fiz2 = _mm256_add_ps(fiz2,tz);
756
757 fjx0 = _mm256_add_ps(fjx0,tx);
758 fjy0 = _mm256_add_ps(fjy0,ty);
759 fjz0 = _mm256_add_ps(fjz0,tz);
760
761 }
762
763 /**************************
764 * CALCULATE INTERACTIONS *
765 **************************/
766
767 if (gmx_mm256_any_lt(rsq30,rcutoff2))
768 {
769
770 r30 = _mm256_mul_ps(rsq30,rinv30);
771 r30 = _mm256_andnot_ps(dummy_mask,r30);
772
773 /* Compute parameters for interactions between i and j atoms */
774 qq30 = _mm256_mul_ps(iq3,jq0);
775
776 /* EWALD ELECTROSTATICS */
777
778 /* Analytical PME correction */
779 zeta2 = _mm256_mul_ps(beta2,rsq30);
780 rinv3 = _mm256_mul_ps(rinvsq30,rinv30);
781 pmecorrF = avx256_pmecorrF_f(zeta2);
782 felec = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
783 felec = _mm256_mul_ps(qq30,felec);
784 pmecorrV = avx256_pmecorrV_f(zeta2);
785 pmecorrV = _mm256_mul_ps(pmecorrV,beta);
786 velec = _mm256_sub_ps(_mm256_sub_ps(rinv30,sh_ewald),pmecorrV);
787 velec = _mm256_mul_ps(qq30,velec);
788
789 cutoff_mask = _mm256_cmp_ps(rsq30,rcutoff2,_CMP_LT_OQ);
790
791 /* Update potential sum for this i atom from the interaction with this j atom. */
792 velec = _mm256_and_ps(velec,cutoff_mask);
793 velec = _mm256_andnot_ps(dummy_mask,velec);
794 velecsum = _mm256_add_ps(velecsum,velec);
795
796 fscal = felec;
797
798 fscal = _mm256_and_ps(fscal,cutoff_mask);
799
800 fscal = _mm256_andnot_ps(dummy_mask,fscal);
801
802 /* Calculate temporary vectorial force */
803 tx = _mm256_mul_ps(fscal,dx30);
804 ty = _mm256_mul_ps(fscal,dy30);
805 tz = _mm256_mul_ps(fscal,dz30);
806
807 /* Update vectorial force */
808 fix3 = _mm256_add_ps(fix3,tx);
809 fiy3 = _mm256_add_ps(fiy3,ty);
810 fiz3 = _mm256_add_ps(fiz3,tz);
811
812 fjx0 = _mm256_add_ps(fjx0,tx);
813 fjy0 = _mm256_add_ps(fjy0,ty);
814 fjz0 = _mm256_add_ps(fjz0,tz);
815
816 }
817
818 fjptrA = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
819 fjptrB = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
820 fjptrC = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
821 fjptrD = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
822 fjptrE = (jnrlistE>=0) ? f+j_coord_offsetE : scratch;
823 fjptrF = (jnrlistF>=0) ? f+j_coord_offsetF : scratch;
824 fjptrG = (jnrlistG>=0) ? f+j_coord_offsetG : scratch;
825 fjptrH = (jnrlistH>=0) ? f+j_coord_offsetH : scratch;
826
827 gmx_mm256_decrement_1rvec_8ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjptrE,fjptrF,fjptrG,fjptrH,fjx0,fjy0,fjz0);
828
829 /* Inner loop uses 374 flops */
830 }
831
832 /* End of innermost loop */
833
834 gmx_mm256_update_iforce_4atom_swizzle_ps(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,fix3,fiy3,fiz3,
835 f+i_coord_offset,fshift+i_shift_offset);
836
837 ggid = gid[iidx];
838 /* Update potential energies */
839 gmx_mm256_update_1pot_ps(velecsum,kernel_data->energygrp_elec+ggid);
840 gmx_mm256_update_1pot_ps(vvdwsum,kernel_data->energygrp_vdw+ggid);
841
842 /* Increment number of inner iterations */
843 inneriter += j_index_end - j_index_start;
844
845 /* Outer loop uses 26 flops */
846 }
847
848 /* Increment number of outer iterations */
849 outeriter += nri;
850
851 /* Update outer/inner flops */
852
853 inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W4_VF,outeriter*26 + inneriter*374);
854 }
855 /*
856 * Gromacs nonbonded kernel: nb_kernel_ElecEwSh_VdwLJSh_GeomW4P1_F_avx_256_single
857 * Electrostatics interaction: Ewald
858 * VdW interaction: LennardJones
859 * Geometry: Water4-Particle
860 * Calculate force/pot: Force
861 */
862 void
863 nb_kernel_ElecEwSh_VdwLJSh_GeomW4P1_F_avx_256_single
864 (t_nblist * gmx_restrict nlist,
865 rvec * gmx_restrict xx,
866 rvec * gmx_restrict ff,
867 struct t_forcerec * gmx_restrict fr,
868 t_mdatoms * gmx_restrict mdatoms,
869 nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
870 t_nrnb * gmx_restrict nrnb)
871 {
872 /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
873 * just 0 for non-waters.
874 * Suffixes A,B,C,D,E,F,G,H refer to j loop unrolling done with AVX, e.g. for the eight different
875 * jnr indices corresponding to data put in the four positions in the SIMD register.
876 */
877 int i_shift_offset,i_coord_offset,outeriter,inneriter;
878 int j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
879 int jnrA,jnrB,jnrC,jnrD;
880 int jnrE,jnrF,jnrG,jnrH;
881 int jnrlistA,jnrlistB,jnrlistC,jnrlistD;
882 int jnrlistE,jnrlistF,jnrlistG,jnrlistH;
883 int j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
884 int j_coord_offsetE,j_coord_offsetF,j_coord_offsetG,j_coord_offsetH;
885 int *iinr,*jindex,*jjnr,*shiftidx,*gid;
886 real rcutoff_scalar;
887 real *shiftvec,*fshift,*x,*f;
888 real *fjptrA,*fjptrB,*fjptrC,*fjptrD,*fjptrE,*fjptrF,*fjptrG,*fjptrH;
889 real scratch[4*DIM];
890 __m256 tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
891 real * vdwioffsetptr0;
892 __m256 ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
893 real * vdwioffsetptr1;
894 __m256 ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
895 real * vdwioffsetptr2;
896 __m256 ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
897 real * vdwioffsetptr3;
898 __m256 ix3,iy3,iz3,fix3,fiy3,fiz3,iq3,isai3;
899 int vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D,vdwjidx0E,vdwjidx0F,vdwjidx0G,vdwjidx0H;
900 __m256 jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
901 __m256 dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
902 __m256 dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
903 __m256 dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
904 __m256 dx30,dy30,dz30,rsq30,rinv30,rinvsq30,r30,qq30,c6_30,c12_30;
905 __m256 velec,felec,velecsum,facel,crf,krf,krf2;
906 real *charge;
907 int nvdwtype;
908 __m256 rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
909 int *vdwtype;
910 real *vdwparam;
911 __m256 one_sixth = _mm256_set1_ps(1.0/6.0);
912 __m256 one_twelfth = _mm256_set1_ps(1.0/12.0);
913 __m256i ewitab;
914 __m128i ewitab_lo,ewitab_hi;
915 __m256 ewtabscale,eweps,sh_ewald,ewrt,ewtabhalfspace,ewtabF,ewtabFn,ewtabD,ewtabV;
916 __m256 beta,beta2,beta3,zeta2,pmecorrF,pmecorrV,rinv3;
917 real *ewtab;
918 __m256 dummy_mask,cutoff_mask;
919 __m256 signbit = _mm256_castsi256_ps( _mm256_set1_epi32(0x80000000) );
920 __m256 one = _mm256_set1_ps(1.0);
921 __m256 two = _mm256_set1_ps(2.0);
922 x = xx[0];
923 f = ff[0];
924
925 nri = nlist->nri;
926 iinr = nlist->iinr;
927 jindex = nlist->jindex;
928 jjnr = nlist->jjnr;
929 shiftidx = nlist->shift;
930 gid = nlist->gid;
931 shiftvec = fr->shift_vec[0];
932 fshift = fr->fshift[0];
933 facel = _mm256_set1_ps(fr->ic->epsfac);
934 charge = mdatoms->chargeA;
935 nvdwtype = fr->ntype;
936 vdwparam = fr->nbfp;
937 vdwtype = mdatoms->typeA;
938
939 sh_ewald = _mm256_set1_ps(fr->ic->sh_ewald);
940 beta = _mm256_set1_ps(fr->ic->ewaldcoeff_q);
941 beta2 = _mm256_mul_ps(beta,beta);
942 beta3 = _mm256_mul_ps(beta,beta2);
943
944 ewtab = fr->ic->tabq_coul_F;
945 ewtabscale = _mm256_set1_ps(fr->ic->tabq_scale);
946 ewtabhalfspace = _mm256_set1_ps(0.5/fr->ic->tabq_scale);
947
948 /* Setup water-specific parameters */
949 inr = nlist->iinr[0];
950 iq1 = _mm256_mul_ps(facel,_mm256_set1_ps(charge[inr+1]));
951 iq2 = _mm256_mul_ps(facel,_mm256_set1_ps(charge[inr+2]));
952 iq3 = _mm256_mul_ps(facel,_mm256_set1_ps(charge[inr+3]));
953 vdwioffsetptr0 = vdwparam+2*nvdwtype*vdwtype[inr+0];
954
955 /* When we use explicit cutoffs the value must be identical for elec and VdW, so use elec as an arbitrary choice */
956 rcutoff_scalar = fr->ic->rcoulomb;
957 rcutoff = _mm256_set1_ps(rcutoff_scalar);
958 rcutoff2 = _mm256_mul_ps(rcutoff,rcutoff);
959
960 sh_vdw_invrcut6 = _mm256_set1_ps(fr->ic->sh_invrc6);
961 rvdw = _mm256_set1_ps(fr->ic->rvdw);
962
963 /* Avoid stupid compiler warnings */
964 jnrA = jnrB = jnrC = jnrD = jnrE = jnrF = jnrG = jnrH = 0;
965 j_coord_offsetA = 0;
966 j_coord_offsetB = 0;
967 j_coord_offsetC = 0;
968 j_coord_offsetD = 0;
969 j_coord_offsetE = 0;
970 j_coord_offsetF = 0;
971 j_coord_offsetG = 0;
972 j_coord_offsetH = 0;
973
974 outeriter = 0;
975 inneriter = 0;
976
977 for(iidx=0;iidx<4*DIM;iidx++)
978 {
979 scratch[iidx] = 0.0;
980 }
981
982 /* Start outer loop over neighborlists */
983 for(iidx=0; iidx<nri; iidx++)
984 {
985 /* Load shift vector for this list */
986 i_shift_offset = DIM*shiftidx[iidx];
987
988 /* Load limits for loop over neighbors */
989 j_index_start = jindex[iidx];
990 j_index_end = jindex[iidx+1];
991
992 /* Get outer coordinate index */
993 inr = iinr[iidx];
994 i_coord_offset = DIM*inr;
995
996 /* Load i particle coords and add shift vector */
997 gmx_mm256_load_shift_and_4rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,
998 &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2,&ix3,&iy3,&iz3);
999
1000 fix0 = _mm256_setzero_ps();
1001 fiy0 = _mm256_setzero_ps();
1002 fiz0 = _mm256_setzero_ps();
1003 fix1 = _mm256_setzero_ps();
1004 fiy1 = _mm256_setzero_ps();
1005 fiz1 = _mm256_setzero_ps();
1006 fix2 = _mm256_setzero_ps();
1007 fiy2 = _mm256_setzero_ps();
1008 fiz2 = _mm256_setzero_ps();
1009 fix3 = _mm256_setzero_ps();
1010 fiy3 = _mm256_setzero_ps();
1011 fiz3 = _mm256_setzero_ps();
1012
1013 /* Start inner kernel loop */
1014 for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+7]>=0; jidx+=8)
1015 {
1016
1017 /* Get j neighbor index, and coordinate index */
1018 jnrA = jjnr[jidx];
1019 jnrB = jjnr[jidx+1];
1020 jnrC = jjnr[jidx+2];
1021 jnrD = jjnr[jidx+3];
1022 jnrE = jjnr[jidx+4];
1023 jnrF = jjnr[jidx+5];
1024 jnrG = jjnr[jidx+6];
1025 jnrH = jjnr[jidx+7];
1026 j_coord_offsetA = DIM*jnrA;
1027 j_coord_offsetB = DIM*jnrB;
1028 j_coord_offsetC = DIM*jnrC;
1029 j_coord_offsetD = DIM*jnrD;
1030 j_coord_offsetE = DIM*jnrE;
1031 j_coord_offsetF = DIM*jnrF;
1032 j_coord_offsetG = DIM*jnrG;
1033 j_coord_offsetH = DIM*jnrH;
1034
1035 /* load j atom coordinates */
1036 gmx_mm256_load_1rvec_8ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
1037 x+j_coord_offsetC,x+j_coord_offsetD,
1038 x+j_coord_offsetE,x+j_coord_offsetF,
1039 x+j_coord_offsetG,x+j_coord_offsetH,
1040 &jx0,&jy0,&jz0);
1041
1042 /* Calculate displacement vector */
1043 dx00 = _mm256_sub_ps(ix0,jx0);
1044 dy00 = _mm256_sub_ps(iy0,jy0);
1045 dz00 = _mm256_sub_ps(iz0,jz0);
1046 dx10 = _mm256_sub_ps(ix1,jx0);
1047 dy10 = _mm256_sub_ps(iy1,jy0);
1048 dz10 = _mm256_sub_ps(iz1,jz0);
1049 dx20 = _mm256_sub_ps(ix2,jx0);
1050 dy20 = _mm256_sub_ps(iy2,jy0);
1051 dz20 = _mm256_sub_ps(iz2,jz0);
1052 dx30 = _mm256_sub_ps(ix3,jx0);
1053 dy30 = _mm256_sub_ps(iy3,jy0);
1054 dz30 = _mm256_sub_ps(iz3,jz0);
1055
1056 /* Calculate squared distance and things based on it */
1057 rsq00 = gmx_mm256_calc_rsq_ps(dx00,dy00,dz00);
1058 rsq10 = gmx_mm256_calc_rsq_ps(dx10,dy10,dz10);
1059 rsq20 = gmx_mm256_calc_rsq_ps(dx20,dy20,dz20);
1060 rsq30 = gmx_mm256_calc_rsq_ps(dx30,dy30,dz30);
1061
1062 rinv10 = avx256_invsqrt_f(rsq10);
1063 rinv20 = avx256_invsqrt_f(rsq20);
1064 rinv30 = avx256_invsqrt_f(rsq30);
1065
1066 rinvsq00 = avx256_inv_f(rsq00);
1067 rinvsq10 = _mm256_mul_ps(rinv10,rinv10);
1068 rinvsq20 = _mm256_mul_ps(rinv20,rinv20);
1069 rinvsq30 = _mm256_mul_ps(rinv30,rinv30);
1070
1071 /* Load parameters for j particles */
1072 jq0 = gmx_mm256_load_8real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
1073 charge+jnrC+0,charge+jnrD+0,
1074 charge+jnrE+0,charge+jnrF+0,
1075 charge+jnrG+0,charge+jnrH+0);
1076 vdwjidx0A = 2*vdwtype[jnrA+0];
1077 vdwjidx0B = 2*vdwtype[jnrB+0];
1078 vdwjidx0C = 2*vdwtype[jnrC+0];
1079 vdwjidx0D = 2*vdwtype[jnrD+0];
1080 vdwjidx0E = 2*vdwtype[jnrE+0];
1081 vdwjidx0F = 2*vdwtype[jnrF+0];
1082 vdwjidx0G = 2*vdwtype[jnrG+0];
1083 vdwjidx0H = 2*vdwtype[jnrH+0];
1084
1085 fjx0 = _mm256_setzero_ps();
1086 fjy0 = _mm256_setzero_ps();
1087 fjz0 = _mm256_setzero_ps();
1088
1089 /**************************
1090 * CALCULATE INTERACTIONS *
1091 **************************/
1092
1093 if (gmx_mm256_any_lt(rsq00,rcutoff2))
1094 {
1095
1096 /* Compute parameters for interactions between i and j atoms */
1097 gmx_mm256_load_8pair_swizzle_ps(vdwioffsetptr0+vdwjidx0A,
1098 vdwioffsetptr0+vdwjidx0B,
1099 vdwioffsetptr0+vdwjidx0C,
1100 vdwioffsetptr0+vdwjidx0D,
1101 vdwioffsetptr0+vdwjidx0E,
1102 vdwioffsetptr0+vdwjidx0F,
1103 vdwioffsetptr0+vdwjidx0G,
1104 vdwioffsetptr0+vdwjidx0H,
1105 &c6_00,&c12_00);
1106
1107 /* LENNARD-JONES DISPERSION/REPULSION */
1108
1109 rinvsix = _mm256_mul_ps(_mm256_mul_ps(rinvsq00,rinvsq00),rinvsq00);
1110 fvdw = _mm256_mul_ps(_mm256_sub_ps(_mm256_mul_ps(c12_00,rinvsix),c6_00),_mm256_mul_ps(rinvsix,rinvsq00));
1111
1112 cutoff_mask = _mm256_cmp_ps(rsq00,rcutoff2,_CMP_LT_OQ);
1113
1114 fscal = fvdw;
1115
1116 fscal = _mm256_and_ps(fscal,cutoff_mask);
1117
1118 /* Calculate temporary vectorial force */
1119 tx = _mm256_mul_ps(fscal,dx00);
1120 ty = _mm256_mul_ps(fscal,dy00);
1121 tz = _mm256_mul_ps(fscal,dz00);
1122
1123 /* Update vectorial force */
1124 fix0 = _mm256_add_ps(fix0,tx);
1125 fiy0 = _mm256_add_ps(fiy0,ty);
1126 fiz0 = _mm256_add_ps(fiz0,tz);
1127
1128 fjx0 = _mm256_add_ps(fjx0,tx);
1129 fjy0 = _mm256_add_ps(fjy0,ty);
1130 fjz0 = _mm256_add_ps(fjz0,tz);
1131
1132 }
1133
1134 /**************************
1135 * CALCULATE INTERACTIONS *
1136 **************************/
1137
1138 if (gmx_mm256_any_lt(rsq10,rcutoff2))
1139 {
1140
1141 r10 = _mm256_mul_ps(rsq10,rinv10);
1142
1143 /* Compute parameters for interactions between i and j atoms */
1144 qq10 = _mm256_mul_ps(iq1,jq0);
1145
1146 /* EWALD ELECTROSTATICS */
1147
1148 /* Analytical PME correction */
1149 zeta2 = _mm256_mul_ps(beta2,rsq10);
1150 rinv3 = _mm256_mul_ps(rinvsq10,rinv10);
1151 pmecorrF = avx256_pmecorrF_f(zeta2);
1152 felec = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
1153 felec = _mm256_mul_ps(qq10,felec);
1154
1155 cutoff_mask = _mm256_cmp_ps(rsq10,rcutoff2,_CMP_LT_OQ);
1156
1157 fscal = felec;
1158
1159 fscal = _mm256_and_ps(fscal,cutoff_mask);
1160
1161 /* Calculate temporary vectorial force */
1162 tx = _mm256_mul_ps(fscal,dx10);
1163 ty = _mm256_mul_ps(fscal,dy10);
1164 tz = _mm256_mul_ps(fscal,dz10);
1165
1166 /* Update vectorial force */
1167 fix1 = _mm256_add_ps(fix1,tx);
1168 fiy1 = _mm256_add_ps(fiy1,ty);
1169 fiz1 = _mm256_add_ps(fiz1,tz);
1170
1171 fjx0 = _mm256_add_ps(fjx0,tx);
1172 fjy0 = _mm256_add_ps(fjy0,ty);
1173 fjz0 = _mm256_add_ps(fjz0,tz);
1174
1175 }
1176
1177 /**************************
1178 * CALCULATE INTERACTIONS *
1179 **************************/
1180
1181 if (gmx_mm256_any_lt(rsq20,rcutoff2))
1182 {
1183
1184 r20 = _mm256_mul_ps(rsq20,rinv20);
1185
1186 /* Compute parameters for interactions between i and j atoms */
1187 qq20 = _mm256_mul_ps(iq2,jq0);
1188
1189 /* EWALD ELECTROSTATICS */
1190
1191 /* Analytical PME correction */
1192 zeta2 = _mm256_mul_ps(beta2,rsq20);
1193 rinv3 = _mm256_mul_ps(rinvsq20,rinv20);
1194 pmecorrF = avx256_pmecorrF_f(zeta2);
1195 felec = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
1196 felec = _mm256_mul_ps(qq20,felec);
1197
1198 cutoff_mask = _mm256_cmp_ps(rsq20,rcutoff2,_CMP_LT_OQ);
1199
1200 fscal = felec;
1201
1202 fscal = _mm256_and_ps(fscal,cutoff_mask);
1203
1204 /* Calculate temporary vectorial force */
1205 tx = _mm256_mul_ps(fscal,dx20);
1206 ty = _mm256_mul_ps(fscal,dy20);
1207 tz = _mm256_mul_ps(fscal,dz20);
1208
1209 /* Update vectorial force */
1210 fix2 = _mm256_add_ps(fix2,tx);
1211 fiy2 = _mm256_add_ps(fiy2,ty);
1212 fiz2 = _mm256_add_ps(fiz2,tz);
1213
1214 fjx0 = _mm256_add_ps(fjx0,tx);
1215 fjy0 = _mm256_add_ps(fjy0,ty);
1216 fjz0 = _mm256_add_ps(fjz0,tz);
1217
1218 }
1219
1220 /**************************
1221 * CALCULATE INTERACTIONS *
1222 **************************/
1223
1224 if (gmx_mm256_any_lt(rsq30,rcutoff2))
1225 {
1226
1227 r30 = _mm256_mul_ps(rsq30,rinv30);
1228
1229 /* Compute parameters for interactions between i and j atoms */
1230 qq30 = _mm256_mul_ps(iq3,jq0);
1231
1232 /* EWALD ELECTROSTATICS */
1233
1234 /* Analytical PME correction */
1235 zeta2 = _mm256_mul_ps(beta2,rsq30);
1236 rinv3 = _mm256_mul_ps(rinvsq30,rinv30);
1237 pmecorrF = avx256_pmecorrF_f(zeta2);
1238 felec = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
1239 felec = _mm256_mul_ps(qq30,felec);
1240
1241 cutoff_mask = _mm256_cmp_ps(rsq30,rcutoff2,_CMP_LT_OQ);
1242
1243 fscal = felec;
1244
1245 fscal = _mm256_and_ps(fscal,cutoff_mask);
1246
1247 /* Calculate temporary vectorial force */
1248 tx = _mm256_mul_ps(fscal,dx30);
1249 ty = _mm256_mul_ps(fscal,dy30);
1250 tz = _mm256_mul_ps(fscal,dz30);
1251
1252 /* Update vectorial force */
1253 fix3 = _mm256_add_ps(fix3,tx);
1254 fiy3 = _mm256_add_ps(fiy3,ty);
1255 fiz3 = _mm256_add_ps(fiz3,tz);
1256
1257 fjx0 = _mm256_add_ps(fjx0,tx);
1258 fjy0 = _mm256_add_ps(fjy0,ty);
1259 fjz0 = _mm256_add_ps(fjz0,tz);
1260
1261 }
1262
1263 fjptrA = f+j_coord_offsetA;
1264 fjptrB = f+j_coord_offsetB;
1265 fjptrC = f+j_coord_offsetC;
1266 fjptrD = f+j_coord_offsetD;
1267 fjptrE = f+j_coord_offsetE;
1268 fjptrF = f+j_coord_offsetF;
1269 fjptrG = f+j_coord_offsetG;
1270 fjptrH = f+j_coord_offsetH;
1271
1272 gmx_mm256_decrement_1rvec_8ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjptrE,fjptrF,fjptrG,fjptrH,fjx0,fjy0,fjz0);
1273
1274 /* Inner loop uses 210 flops */
1275 }
1276
1277 if(jidx<j_index_end)
1278 {
1279
1280 /* Get j neighbor index, and coordinate index */
1281 jnrlistA = jjnr[jidx];
1282 jnrlistB = jjnr[jidx+1];
1283 jnrlistC = jjnr[jidx+2];
1284 jnrlistD = jjnr[jidx+3];
1285 jnrlistE = jjnr[jidx+4];
1286 jnrlistF = jjnr[jidx+5];
1287 jnrlistG = jjnr[jidx+6];
1288 jnrlistH = jjnr[jidx+7];
1289 /* Sign of each element will be negative for non-real atoms.
1290 * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
1291 * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
1292 */
1293 dummy_mask = gmx_mm256_set_m128(gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx+4)),_mm_setzero_si128())),
1294 gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128())));
1295
1296 jnrA = (jnrlistA>=0) ? jnrlistA : 0;
1297 jnrB = (jnrlistB>=0) ? jnrlistB : 0;
1298 jnrC = (jnrlistC>=0) ? jnrlistC : 0;
1299 jnrD = (jnrlistD>=0) ? jnrlistD : 0;
1300 jnrE = (jnrlistE>=0) ? jnrlistE : 0;
1301 jnrF = (jnrlistF>=0) ? jnrlistF : 0;
1302 jnrG = (jnrlistG>=0) ? jnrlistG : 0;
1303 jnrH = (jnrlistH>=0) ? jnrlistH : 0;
1304 j_coord_offsetA = DIM*jnrA;
1305 j_coord_offsetB = DIM*jnrB;
1306 j_coord_offsetC = DIM*jnrC;
1307 j_coord_offsetD = DIM*jnrD;
1308 j_coord_offsetE = DIM*jnrE;
1309 j_coord_offsetF = DIM*jnrF;
1310 j_coord_offsetG = DIM*jnrG;
1311 j_coord_offsetH = DIM*jnrH;
1312
1313 /* load j atom coordinates */
1314 gmx_mm256_load_1rvec_8ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
1315 x+j_coord_offsetC,x+j_coord_offsetD,
1316 x+j_coord_offsetE,x+j_coord_offsetF,
1317 x+j_coord_offsetG,x+j_coord_offsetH,
1318 &jx0,&jy0,&jz0);
1319
1320 /* Calculate displacement vector */
1321 dx00 = _mm256_sub_ps(ix0,jx0);
1322 dy00 = _mm256_sub_ps(iy0,jy0);
1323 dz00 = _mm256_sub_ps(iz0,jz0);
1324 dx10 = _mm256_sub_ps(ix1,jx0);
1325 dy10 = _mm256_sub_ps(iy1,jy0);
1326 dz10 = _mm256_sub_ps(iz1,jz0);
1327 dx20 = _mm256_sub_ps(ix2,jx0);
1328 dy20 = _mm256_sub_ps(iy2,jy0);
1329 dz20 = _mm256_sub_ps(iz2,jz0);
1330 dx30 = _mm256_sub_ps(ix3,jx0);
1331 dy30 = _mm256_sub_ps(iy3,jy0);
1332 dz30 = _mm256_sub_ps(iz3,jz0);
1333
1334 /* Calculate squared distance and things based on it */
1335 rsq00 = gmx_mm256_calc_rsq_ps(dx00,dy00,dz00);
1336 rsq10 = gmx_mm256_calc_rsq_ps(dx10,dy10,dz10);
1337 rsq20 = gmx_mm256_calc_rsq_ps(dx20,dy20,dz20);
1338 rsq30 = gmx_mm256_calc_rsq_ps(dx30,dy30,dz30);
1339
1340 rinv10 = avx256_invsqrt_f(rsq10);
1341 rinv20 = avx256_invsqrt_f(rsq20);
1342 rinv30 = avx256_invsqrt_f(rsq30);
1343
1344 rinvsq00 = avx256_inv_f(rsq00);
1345 rinvsq10 = _mm256_mul_ps(rinv10,rinv10);
1346 rinvsq20 = _mm256_mul_ps(rinv20,rinv20);
1347 rinvsq30 = _mm256_mul_ps(rinv30,rinv30);
1348
1349 /* Load parameters for j particles */
1350 jq0 = gmx_mm256_load_8real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
1351 charge+jnrC+0,charge+jnrD+0,
1352 charge+jnrE+0,charge+jnrF+0,
1353 charge+jnrG+0,charge+jnrH+0);
1354 vdwjidx0A = 2*vdwtype[jnrA+0];
1355 vdwjidx0B = 2*vdwtype[jnrB+0];
1356 vdwjidx0C = 2*vdwtype[jnrC+0];
1357 vdwjidx0D = 2*vdwtype[jnrD+0];
1358 vdwjidx0E = 2*vdwtype[jnrE+0];
1359 vdwjidx0F = 2*vdwtype[jnrF+0];
1360 vdwjidx0G = 2*vdwtype[jnrG+0];
1361 vdwjidx0H = 2*vdwtype[jnrH+0];
1362
1363 fjx0 = _mm256_setzero_ps();
1364 fjy0 = _mm256_setzero_ps();
1365 fjz0 = _mm256_setzero_ps();
1366
1367 /**************************
1368 * CALCULATE INTERACTIONS *
1369 **************************/
1370
1371 if (gmx_mm256_any_lt(rsq00,rcutoff2))
1372 {
1373
1374 /* Compute parameters for interactions between i and j atoms */
1375 gmx_mm256_load_8pair_swizzle_ps(vdwioffsetptr0+vdwjidx0A,
1376 vdwioffsetptr0+vdwjidx0B,
1377 vdwioffsetptr0+vdwjidx0C,
1378 vdwioffsetptr0+vdwjidx0D,
1379 vdwioffsetptr0+vdwjidx0E,
1380 vdwioffsetptr0+vdwjidx0F,
1381 vdwioffsetptr0+vdwjidx0G,
1382 vdwioffsetptr0+vdwjidx0H,
1383 &c6_00,&c12_00);
1384
1385 /* LENNARD-JONES DISPERSION/REPULSION */
1386
1387 rinvsix = _mm256_mul_ps(_mm256_mul_ps(rinvsq00,rinvsq00),rinvsq00);
1388 fvdw = _mm256_mul_ps(_mm256_sub_ps(_mm256_mul_ps(c12_00,rinvsix),c6_00),_mm256_mul_ps(rinvsix,rinvsq00));
1389
1390 cutoff_mask = _mm256_cmp_ps(rsq00,rcutoff2,_CMP_LT_OQ);
1391
1392 fscal = fvdw;
1393
1394 fscal = _mm256_and_ps(fscal,cutoff_mask);
1395
1396 fscal = _mm256_andnot_ps(dummy_mask,fscal);
1397
1398 /* Calculate temporary vectorial force */
1399 tx = _mm256_mul_ps(fscal,dx00);
1400 ty = _mm256_mul_ps(fscal,dy00);
1401 tz = _mm256_mul_ps(fscal,dz00);
1402
1403 /* Update vectorial force */
1404 fix0 = _mm256_add_ps(fix0,tx);
1405 fiy0 = _mm256_add_ps(fiy0,ty);
1406 fiz0 = _mm256_add_ps(fiz0,tz);
1407
1408 fjx0 = _mm256_add_ps(fjx0,tx);
1409 fjy0 = _mm256_add_ps(fjy0,ty);
1410 fjz0 = _mm256_add_ps(fjz0,tz);
1411
1412 }
1413
1414 /**************************
1415 * CALCULATE INTERACTIONS *
1416 **************************/
1417
1418 if (gmx_mm256_any_lt(rsq10,rcutoff2))
1419 {
1420
1421 r10 = _mm256_mul_ps(rsq10,rinv10);
1422 r10 = _mm256_andnot_ps(dummy_mask,r10);
1423
1424 /* Compute parameters for interactions between i and j atoms */
1425 qq10 = _mm256_mul_ps(iq1,jq0);
1426
1427 /* EWALD ELECTROSTATICS */
1428
1429 /* Analytical PME correction */
1430 zeta2 = _mm256_mul_ps(beta2,rsq10);
1431 rinv3 = _mm256_mul_ps(rinvsq10,rinv10);
1432 pmecorrF = avx256_pmecorrF_f(zeta2);
1433 felec = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
1434 felec = _mm256_mul_ps(qq10,felec);
1435
1436 cutoff_mask = _mm256_cmp_ps(rsq10,rcutoff2,_CMP_LT_OQ);
1437
1438 fscal = felec;
1439
1440 fscal = _mm256_and_ps(fscal,cutoff_mask);
1441
1442 fscal = _mm256_andnot_ps(dummy_mask,fscal);
1443
1444 /* Calculate temporary vectorial force */
1445 tx = _mm256_mul_ps(fscal,dx10);
1446 ty = _mm256_mul_ps(fscal,dy10);
1447 tz = _mm256_mul_ps(fscal,dz10);
1448
1449 /* Update vectorial force */
1450 fix1 = _mm256_add_ps(fix1,tx);
1451 fiy1 = _mm256_add_ps(fiy1,ty);
1452 fiz1 = _mm256_add_ps(fiz1,tz);
1453
1454 fjx0 = _mm256_add_ps(fjx0,tx);
1455 fjy0 = _mm256_add_ps(fjy0,ty);
1456 fjz0 = _mm256_add_ps(fjz0,tz);
1457
1458 }
1459
1460 /**************************
1461 * CALCULATE INTERACTIONS *
1462 **************************/
1463
1464 if (gmx_mm256_any_lt(rsq20,rcutoff2))
1465 {
1466
1467 r20 = _mm256_mul_ps(rsq20,rinv20);
1468 r20 = _mm256_andnot_ps(dummy_mask,r20);
1469
1470 /* Compute parameters for interactions between i and j atoms */
1471 qq20 = _mm256_mul_ps(iq2,jq0);
1472
1473 /* EWALD ELECTROSTATICS */
1474
1475 /* Analytical PME correction */
1476 zeta2 = _mm256_mul_ps(beta2,rsq20);
1477 rinv3 = _mm256_mul_ps(rinvsq20,rinv20);
1478 pmecorrF = avx256_pmecorrF_f(zeta2);
1479 felec = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
1480 felec = _mm256_mul_ps(qq20,felec);
1481
1482 cutoff_mask = _mm256_cmp_ps(rsq20,rcutoff2,_CMP_LT_OQ);
1483
1484 fscal = felec;
1485
1486 fscal = _mm256_and_ps(fscal,cutoff_mask);
1487
1488 fscal = _mm256_andnot_ps(dummy_mask,fscal);
1489
1490 /* Calculate temporary vectorial force */
1491 tx = _mm256_mul_ps(fscal,dx20);
1492 ty = _mm256_mul_ps(fscal,dy20);
1493 tz = _mm256_mul_ps(fscal,dz20);
1494
1495 /* Update vectorial force */
1496 fix2 = _mm256_add_ps(fix2,tx);
1497 fiy2 = _mm256_add_ps(fiy2,ty);
1498 fiz2 = _mm256_add_ps(fiz2,tz);
1499
1500 fjx0 = _mm256_add_ps(fjx0,tx);
1501 fjy0 = _mm256_add_ps(fjy0,ty);
1502 fjz0 = _mm256_add_ps(fjz0,tz);
1503
1504 }
1505
1506 /**************************
1507 * CALCULATE INTERACTIONS *
1508 **************************/
1509
1510 if (gmx_mm256_any_lt(rsq30,rcutoff2))
1511 {
1512
1513 r30 = _mm256_mul_ps(rsq30,rinv30);
1514 r30 = _mm256_andnot_ps(dummy_mask,r30);
1515
1516 /* Compute parameters for interactions between i and j atoms */
1517 qq30 = _mm256_mul_ps(iq3,jq0);
1518
1519 /* EWALD ELECTROSTATICS */
1520
1521 /* Analytical PME correction */
1522 zeta2 = _mm256_mul_ps(beta2,rsq30);
1523 rinv3 = _mm256_mul_ps(rinvsq30,rinv30);
1524 pmecorrF = avx256_pmecorrF_f(zeta2);
1525 felec = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
1526 felec = _mm256_mul_ps(qq30,felec);
1527
1528 cutoff_mask = _mm256_cmp_ps(rsq30,rcutoff2,_CMP_LT_OQ);
1529
1530 fscal = felec;
1531
1532 fscal = _mm256_and_ps(fscal,cutoff_mask);
1533
1534 fscal = _mm256_andnot_ps(dummy_mask,fscal);
1535
1536 /* Calculate temporary vectorial force */
1537 tx = _mm256_mul_ps(fscal,dx30);
1538 ty = _mm256_mul_ps(fscal,dy30);
1539 tz = _mm256_mul_ps(fscal,dz30);
1540
1541 /* Update vectorial force */
1542 fix3 = _mm256_add_ps(fix3,tx);
1543 fiy3 = _mm256_add_ps(fiy3,ty);
1544 fiz3 = _mm256_add_ps(fiz3,tz);
1545
1546 fjx0 = _mm256_add_ps(fjx0,tx);
1547 fjy0 = _mm256_add_ps(fjy0,ty);
1548 fjz0 = _mm256_add_ps(fjz0,tz);
1549
1550 }
1551
1552 fjptrA = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
1553 fjptrB = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
1554 fjptrC = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
1555 fjptrD = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
1556 fjptrE = (jnrlistE>=0) ? f+j_coord_offsetE : scratch;
1557 fjptrF = (jnrlistF>=0) ? f+j_coord_offsetF : scratch;
1558 fjptrG = (jnrlistG>=0) ? f+j_coord_offsetG : scratch;
1559 fjptrH = (jnrlistH>=0) ? f+j_coord_offsetH : scratch;
1560
1561 gmx_mm256_decrement_1rvec_8ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjptrE,fjptrF,fjptrG,fjptrH,fjx0,fjy0,fjz0);
1562
1563 /* Inner loop uses 213 flops */
1564 }
1565
1566 /* End of innermost loop */
1567
1568 gmx_mm256_update_iforce_4atom_swizzle_ps(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,fix3,fiy3,fiz3,
1569 f+i_coord_offset,fshift+i_shift_offset);
1570
1571 /* Increment number of inner iterations */
1572 inneriter += j_index_end - j_index_start;
1573
1574 /* Outer loop uses 24 flops */
1575 }
1576
1577 /* Increment number of outer iterations */
1578 outeriter += nri;
1579
1580 /* Update outer/inner flops */
1581
1582 inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W4_F,outeriter*24 + inneriter*213);
1583 }
The ultimate molecular dynamics simulation package