?  search:
summary | log | graphiclog1 | graphiclog2 | commit | commitdiff | tree | refs | edit | fork
blame | history | raw | HEAD
Fix segmentation fault in minimize
[gromacs.git] / src / gromacs / gmxlib / nonbonded / nb_kernel_avx_256_single / nb_kernel_ElecEwSh_VdwLJEwSh_GeomW4W4_avx_256_single.cpp
blobef15fbe7507fee1ff823c595fc16a61b3e68a7bf
1 /*
2 * This file is part of the GROMACS molecular simulation package.
3 *
4 * Copyright (c) 2012,2013,2014,2015,2017,2018, 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_VdwLJEwSh_GeomW4W4_VF_avx_256_single
51 * Electrostatics interaction: Ewald
52 * VdW interaction: LJEwald
53 * Geometry: Water4-Water4
54 * Calculate force/pot: PotentialAndForce
55 */
56 void
57 nb_kernel_ElecEwSh_VdwLJEwSh_GeomW4W4_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 real * vdwgridioffsetptr0;
87 __m256 ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
88 real * vdwioffsetptr1;
89 real * vdwgridioffsetptr1;
90 __m256 ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
91 real * vdwioffsetptr2;
92 real * vdwgridioffsetptr2;
93 __m256 ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
94 real * vdwioffsetptr3;
95 real * vdwgridioffsetptr3;
96 __m256 ix3,iy3,iz3,fix3,fiy3,fiz3,iq3,isai3;
97 int vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D,vdwjidx0E,vdwjidx0F,vdwjidx0G,vdwjidx0H;
98 __m256 jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
99 int vdwjidx1A,vdwjidx1B,vdwjidx1C,vdwjidx1D,vdwjidx1E,vdwjidx1F,vdwjidx1G,vdwjidx1H;
100 __m256 jx1,jy1,jz1,fjx1,fjy1,fjz1,jq1,isaj1;
101 int vdwjidx2A,vdwjidx2B,vdwjidx2C,vdwjidx2D,vdwjidx2E,vdwjidx2F,vdwjidx2G,vdwjidx2H;
102 __m256 jx2,jy2,jz2,fjx2,fjy2,fjz2,jq2,isaj2;
103 int vdwjidx3A,vdwjidx3B,vdwjidx3C,vdwjidx3D,vdwjidx3E,vdwjidx3F,vdwjidx3G,vdwjidx3H;
104 __m256 jx3,jy3,jz3,fjx3,fjy3,fjz3,jq3,isaj3;
105 __m256 dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
106 __m256 dx11,dy11,dz11,rsq11,rinv11,rinvsq11,r11,qq11,c6_11,c12_11;
107 __m256 dx12,dy12,dz12,rsq12,rinv12,rinvsq12,r12,qq12,c6_12,c12_12;
108 __m256 dx13,dy13,dz13,rsq13,rinv13,rinvsq13,r13,qq13,c6_13,c12_13;
109 __m256 dx21,dy21,dz21,rsq21,rinv21,rinvsq21,r21,qq21,c6_21,c12_21;
110 __m256 dx22,dy22,dz22,rsq22,rinv22,rinvsq22,r22,qq22,c6_22,c12_22;
111 __m256 dx23,dy23,dz23,rsq23,rinv23,rinvsq23,r23,qq23,c6_23,c12_23;
112 __m256 dx31,dy31,dz31,rsq31,rinv31,rinvsq31,r31,qq31,c6_31,c12_31;
113 __m256 dx32,dy32,dz32,rsq32,rinv32,rinvsq32,r32,qq32,c6_32,c12_32;
114 __m256 dx33,dy33,dz33,rsq33,rinv33,rinvsq33,r33,qq33,c6_33,c12_33;
115 __m256 velec,felec,velecsum,facel,crf,krf,krf2;
116 real *charge;
117 int nvdwtype;
118 __m256 rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
119 int *vdwtype;
120 real *vdwparam;
121 __m256 one_sixth = _mm256_set1_ps(1.0/6.0);
122 __m256 one_twelfth = _mm256_set1_ps(1.0/12.0);
123 __m256 c6grid_00;
124 __m256 c6grid_11;
125 __m256 c6grid_12;
126 __m256 c6grid_13;
127 __m256 c6grid_21;
128 __m256 c6grid_22;
129 __m256 c6grid_23;
130 __m256 c6grid_31;
131 __m256 c6grid_32;
132 __m256 c6grid_33;
133 real *vdwgridparam;
134 __m256 ewclj,ewclj2,ewclj6,ewcljrsq,poly,exponent,f6A,f6B,sh_lj_ewald;
135 __m256 one_half = _mm256_set1_ps(0.5);
136 __m256 minus_one = _mm256_set1_ps(-1.0);
137 __m256i ewitab;
138 __m128i ewitab_lo,ewitab_hi;
139 __m256 ewtabscale,eweps,sh_ewald,ewrt,ewtabhalfspace,ewtabF,ewtabFn,ewtabD,ewtabV;
140 __m256 beta,beta2,beta3,zeta2,pmecorrF,pmecorrV,rinv3;
141 real *ewtab;
142 __m256 dummy_mask,cutoff_mask;
143 __m256 signbit = _mm256_castsi256_ps( _mm256_set1_epi32(0x80000000) );
144 __m256 one = _mm256_set1_ps(1.0);
145 __m256 two = _mm256_set1_ps(2.0);
146 x = xx[0];
147 f = ff[0];
148
149 nri = nlist->nri;
150 iinr = nlist->iinr;
151 jindex = nlist->jindex;
152 jjnr = nlist->jjnr;
153 shiftidx = nlist->shift;
154 gid = nlist->gid;
155 shiftvec = fr->shift_vec[0];
156 fshift = fr->fshift[0];
157 facel = _mm256_set1_ps(fr->ic->epsfac);
158 charge = mdatoms->chargeA;
159 nvdwtype = fr->ntype;
160 vdwparam = fr->nbfp;
161 vdwtype = mdatoms->typeA;
162 vdwgridparam = fr->ljpme_c6grid;
163 sh_lj_ewald = _mm256_set1_ps(fr->ic->sh_lj_ewald);
164 ewclj = _mm256_set1_ps(fr->ic->ewaldcoeff_lj);
165 ewclj2 = _mm256_mul_ps(minus_one,_mm256_mul_ps(ewclj,ewclj));
166
167 sh_ewald = _mm256_set1_ps(fr->ic->sh_ewald);
168 beta = _mm256_set1_ps(fr->ic->ewaldcoeff_q);
169 beta2 = _mm256_mul_ps(beta,beta);
170 beta3 = _mm256_mul_ps(beta,beta2);
171
172 ewtab = fr->ic->tabq_coul_FDV0;
173 ewtabscale = _mm256_set1_ps(fr->ic->tabq_scale);
174 ewtabhalfspace = _mm256_set1_ps(0.5/fr->ic->tabq_scale);
175
176 /* Setup water-specific parameters */
177 inr = nlist->iinr[0];
178 iq1 = _mm256_mul_ps(facel,_mm256_set1_ps(charge[inr+1]));
179 iq2 = _mm256_mul_ps(facel,_mm256_set1_ps(charge[inr+2]));
180 iq3 = _mm256_mul_ps(facel,_mm256_set1_ps(charge[inr+3]));
181 vdwioffsetptr0 = vdwparam+2*nvdwtype*vdwtype[inr+0];
182 vdwgridioffsetptr0 = vdwgridparam+2*nvdwtype*vdwtype[inr+0];
183
184 jq1 = _mm256_set1_ps(charge[inr+1]);
185 jq2 = _mm256_set1_ps(charge[inr+2]);
186 jq3 = _mm256_set1_ps(charge[inr+3]);
187 vdwjidx0A = 2*vdwtype[inr+0];
188 c6_00 = _mm256_set1_ps(vdwioffsetptr0[vdwjidx0A]);
189 c12_00 = _mm256_set1_ps(vdwioffsetptr0[vdwjidx0A+1]);
190 c6grid_00 = _mm256_set1_ps(vdwgridioffsetptr0[vdwjidx0A]);
191 qq11 = _mm256_mul_ps(iq1,jq1);
192 qq12 = _mm256_mul_ps(iq1,jq2);
193 qq13 = _mm256_mul_ps(iq1,jq3);
194 qq21 = _mm256_mul_ps(iq2,jq1);
195 qq22 = _mm256_mul_ps(iq2,jq2);
196 qq23 = _mm256_mul_ps(iq2,jq3);
197 qq31 = _mm256_mul_ps(iq3,jq1);
198 qq32 = _mm256_mul_ps(iq3,jq2);
199 qq33 = _mm256_mul_ps(iq3,jq3);
200
201 /* When we use explicit cutoffs the value must be identical for elec and VdW, so use elec as an arbitrary choice */
202 rcutoff_scalar = fr->ic->rcoulomb;
203 rcutoff = _mm256_set1_ps(rcutoff_scalar);
204 rcutoff2 = _mm256_mul_ps(rcutoff,rcutoff);
205
206 sh_vdw_invrcut6 = _mm256_set1_ps(fr->ic->sh_invrc6);
207 rvdw = _mm256_set1_ps(fr->ic->rvdw);
208
209 /* Avoid stupid compiler warnings */
210 jnrA = jnrB = jnrC = jnrD = jnrE = jnrF = jnrG = jnrH = 0;
211 j_coord_offsetA = 0;
212 j_coord_offsetB = 0;
213 j_coord_offsetC = 0;
214 j_coord_offsetD = 0;
215 j_coord_offsetE = 0;
216 j_coord_offsetF = 0;
217 j_coord_offsetG = 0;
218 j_coord_offsetH = 0;
219
220 outeriter = 0;
221 inneriter = 0;
222
223 for(iidx=0;iidx<4*DIM;iidx++)
224 {
225 scratch[iidx] = 0.0;
226 }
227
228 /* Start outer loop over neighborlists */
229 for(iidx=0; iidx<nri; iidx++)
230 {
231 /* Load shift vector for this list */
232 i_shift_offset = DIM*shiftidx[iidx];
233
234 /* Load limits for loop over neighbors */
235 j_index_start = jindex[iidx];
236 j_index_end = jindex[iidx+1];
237
238 /* Get outer coordinate index */
239 inr = iinr[iidx];
240 i_coord_offset = DIM*inr;
241
242 /* Load i particle coords and add shift vector */
243 gmx_mm256_load_shift_and_4rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,
244 &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2,&ix3,&iy3,&iz3);
245
246 fix0 = _mm256_setzero_ps();
247 fiy0 = _mm256_setzero_ps();
248 fiz0 = _mm256_setzero_ps();
249 fix1 = _mm256_setzero_ps();
250 fiy1 = _mm256_setzero_ps();
251 fiz1 = _mm256_setzero_ps();
252 fix2 = _mm256_setzero_ps();
253 fiy2 = _mm256_setzero_ps();
254 fiz2 = _mm256_setzero_ps();
255 fix3 = _mm256_setzero_ps();
256 fiy3 = _mm256_setzero_ps();
257 fiz3 = _mm256_setzero_ps();
258
259 /* Reset potential sums */
260 velecsum = _mm256_setzero_ps();
261 vvdwsum = _mm256_setzero_ps();
262
263 /* Start inner kernel loop */
264 for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+7]>=0; jidx+=8)
265 {
266
267 /* Get j neighbor index, and coordinate index */
268 jnrA = jjnr[jidx];
269 jnrB = jjnr[jidx+1];
270 jnrC = jjnr[jidx+2];
271 jnrD = jjnr[jidx+3];
272 jnrE = jjnr[jidx+4];
273 jnrF = jjnr[jidx+5];
274 jnrG = jjnr[jidx+6];
275 jnrH = jjnr[jidx+7];
276 j_coord_offsetA = DIM*jnrA;
277 j_coord_offsetB = DIM*jnrB;
278 j_coord_offsetC = DIM*jnrC;
279 j_coord_offsetD = DIM*jnrD;
280 j_coord_offsetE = DIM*jnrE;
281 j_coord_offsetF = DIM*jnrF;
282 j_coord_offsetG = DIM*jnrG;
283 j_coord_offsetH = DIM*jnrH;
284
285 /* load j atom coordinates */
286 gmx_mm256_load_4rvec_8ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
287 x+j_coord_offsetC,x+j_coord_offsetD,
288 x+j_coord_offsetE,x+j_coord_offsetF,
289 x+j_coord_offsetG,x+j_coord_offsetH,
290 &jx0,&jy0,&jz0,&jx1,&jy1,&jz1,&jx2,
291 &jy2,&jz2,&jx3,&jy3,&jz3);
292
293 /* Calculate displacement vector */
294 dx00 = _mm256_sub_ps(ix0,jx0);
295 dy00 = _mm256_sub_ps(iy0,jy0);
296 dz00 = _mm256_sub_ps(iz0,jz0);
297 dx11 = _mm256_sub_ps(ix1,jx1);
298 dy11 = _mm256_sub_ps(iy1,jy1);
299 dz11 = _mm256_sub_ps(iz1,jz1);
300 dx12 = _mm256_sub_ps(ix1,jx2);
301 dy12 = _mm256_sub_ps(iy1,jy2);
302 dz12 = _mm256_sub_ps(iz1,jz2);
303 dx13 = _mm256_sub_ps(ix1,jx3);
304 dy13 = _mm256_sub_ps(iy1,jy3);
305 dz13 = _mm256_sub_ps(iz1,jz3);
306 dx21 = _mm256_sub_ps(ix2,jx1);
307 dy21 = _mm256_sub_ps(iy2,jy1);
308 dz21 = _mm256_sub_ps(iz2,jz1);
309 dx22 = _mm256_sub_ps(ix2,jx2);
310 dy22 = _mm256_sub_ps(iy2,jy2);
311 dz22 = _mm256_sub_ps(iz2,jz2);
312 dx23 = _mm256_sub_ps(ix2,jx3);
313 dy23 = _mm256_sub_ps(iy2,jy3);
314 dz23 = _mm256_sub_ps(iz2,jz3);
315 dx31 = _mm256_sub_ps(ix3,jx1);
316 dy31 = _mm256_sub_ps(iy3,jy1);
317 dz31 = _mm256_sub_ps(iz3,jz1);
318 dx32 = _mm256_sub_ps(ix3,jx2);
319 dy32 = _mm256_sub_ps(iy3,jy2);
320 dz32 = _mm256_sub_ps(iz3,jz2);
321 dx33 = _mm256_sub_ps(ix3,jx3);
322 dy33 = _mm256_sub_ps(iy3,jy3);
323 dz33 = _mm256_sub_ps(iz3,jz3);
324
325 /* Calculate squared distance and things based on it */
326 rsq00 = gmx_mm256_calc_rsq_ps(dx00,dy00,dz00);
327 rsq11 = gmx_mm256_calc_rsq_ps(dx11,dy11,dz11);
328 rsq12 = gmx_mm256_calc_rsq_ps(dx12,dy12,dz12);
329 rsq13 = gmx_mm256_calc_rsq_ps(dx13,dy13,dz13);
330 rsq21 = gmx_mm256_calc_rsq_ps(dx21,dy21,dz21);
331 rsq22 = gmx_mm256_calc_rsq_ps(dx22,dy22,dz22);
332 rsq23 = gmx_mm256_calc_rsq_ps(dx23,dy23,dz23);
333 rsq31 = gmx_mm256_calc_rsq_ps(dx31,dy31,dz31);
334 rsq32 = gmx_mm256_calc_rsq_ps(dx32,dy32,dz32);
335 rsq33 = gmx_mm256_calc_rsq_ps(dx33,dy33,dz33);
336
337 rinv00 = avx256_invsqrt_f(rsq00);
338 rinv11 = avx256_invsqrt_f(rsq11);
339 rinv12 = avx256_invsqrt_f(rsq12);
340 rinv13 = avx256_invsqrt_f(rsq13);
341 rinv21 = avx256_invsqrt_f(rsq21);
342 rinv22 = avx256_invsqrt_f(rsq22);
343 rinv23 = avx256_invsqrt_f(rsq23);
344 rinv31 = avx256_invsqrt_f(rsq31);
345 rinv32 = avx256_invsqrt_f(rsq32);
346 rinv33 = avx256_invsqrt_f(rsq33);
347
348 rinvsq00 = _mm256_mul_ps(rinv00,rinv00);
349 rinvsq11 = _mm256_mul_ps(rinv11,rinv11);
350 rinvsq12 = _mm256_mul_ps(rinv12,rinv12);
351 rinvsq13 = _mm256_mul_ps(rinv13,rinv13);
352 rinvsq21 = _mm256_mul_ps(rinv21,rinv21);
353 rinvsq22 = _mm256_mul_ps(rinv22,rinv22);
354 rinvsq23 = _mm256_mul_ps(rinv23,rinv23);
355 rinvsq31 = _mm256_mul_ps(rinv31,rinv31);
356 rinvsq32 = _mm256_mul_ps(rinv32,rinv32);
357 rinvsq33 = _mm256_mul_ps(rinv33,rinv33);
358
359 fjx0 = _mm256_setzero_ps();
360 fjy0 = _mm256_setzero_ps();
361 fjz0 = _mm256_setzero_ps();
362 fjx1 = _mm256_setzero_ps();
363 fjy1 = _mm256_setzero_ps();
364 fjz1 = _mm256_setzero_ps();
365 fjx2 = _mm256_setzero_ps();
366 fjy2 = _mm256_setzero_ps();
367 fjz2 = _mm256_setzero_ps();
368 fjx3 = _mm256_setzero_ps();
369 fjy3 = _mm256_setzero_ps();
370 fjz3 = _mm256_setzero_ps();
371
372 /**************************
373 * CALCULATE INTERACTIONS *
374 **************************/
375
376 if (gmx_mm256_any_lt(rsq00,rcutoff2))
377 {
378
379 r00 = _mm256_mul_ps(rsq00,rinv00);
380
381 /* Analytical LJ-PME */
382 rinvsix = _mm256_mul_ps(_mm256_mul_ps(rinvsq00,rinvsq00),rinvsq00);
383 ewcljrsq = _mm256_mul_ps(ewclj2,rsq00);
384 ewclj6 = _mm256_mul_ps(ewclj2,_mm256_mul_ps(ewclj2,ewclj2));
385 exponent = avx256_exp_f(ewcljrsq);
386 /* poly = exp(-(beta*r)^2) * (1 + (beta*r)^2 + (beta*r)^4 /2) */
387 poly = _mm256_mul_ps(exponent,_mm256_add_ps(_mm256_sub_ps(one,ewcljrsq),_mm256_mul_ps(_mm256_mul_ps(ewcljrsq,ewcljrsq),one_half)));
388 /* vvdw6 = [C6 - C6grid * (1-poly)]/r6 */
389 vvdw6 = _mm256_mul_ps(_mm256_sub_ps(c6_00,_mm256_mul_ps(c6grid_00,_mm256_sub_ps(one,poly))),rinvsix);
390 vvdw12 = _mm256_mul_ps(c12_00,_mm256_mul_ps(rinvsix,rinvsix));
391 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) ,
392 _mm256_mul_ps( _mm256_sub_ps(vvdw6,_mm256_add_ps(_mm256_mul_ps(c6_00,sh_vdw_invrcut6),_mm256_mul_ps(c6grid_00,sh_lj_ewald))),one_sixth));
393 /* fvdw = vvdw12/r - (vvdw6/r + (C6grid * exponent * beta^6)/r) */
394 fvdw = _mm256_mul_ps(_mm256_sub_ps(vvdw12,_mm256_sub_ps(vvdw6,_mm256_mul_ps(_mm256_mul_ps(c6grid_00,one_sixth),_mm256_mul_ps(exponent,ewclj6)))),rinvsq00);
395
396 cutoff_mask = _mm256_cmp_ps(rsq00,rcutoff2,_CMP_LT_OQ);
397
398 /* Update potential sum for this i atom from the interaction with this j atom. */
399 vvdw = _mm256_and_ps(vvdw,cutoff_mask);
400 vvdwsum = _mm256_add_ps(vvdwsum,vvdw);
401
402 fscal = fvdw;
403
404 fscal = _mm256_and_ps(fscal,cutoff_mask);
405
406 /* Calculate temporary vectorial force */
407 tx = _mm256_mul_ps(fscal,dx00);
408 ty = _mm256_mul_ps(fscal,dy00);
409 tz = _mm256_mul_ps(fscal,dz00);
410
411 /* Update vectorial force */
412 fix0 = _mm256_add_ps(fix0,tx);
413 fiy0 = _mm256_add_ps(fiy0,ty);
414 fiz0 = _mm256_add_ps(fiz0,tz);
415
416 fjx0 = _mm256_add_ps(fjx0,tx);
417 fjy0 = _mm256_add_ps(fjy0,ty);
418 fjz0 = _mm256_add_ps(fjz0,tz);
419
420 }
421
422 /**************************
423 * CALCULATE INTERACTIONS *
424 **************************/
425
426 if (gmx_mm256_any_lt(rsq11,rcutoff2))
427 {
428
429 r11 = _mm256_mul_ps(rsq11,rinv11);
430
431 /* EWALD ELECTROSTATICS */
432
433 /* Analytical PME correction */
434 zeta2 = _mm256_mul_ps(beta2,rsq11);
435 rinv3 = _mm256_mul_ps(rinvsq11,rinv11);
436 pmecorrF = avx256_pmecorrF_f(zeta2);
437 felec = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
438 felec = _mm256_mul_ps(qq11,felec);
439 pmecorrV = avx256_pmecorrV_f(zeta2);
440 pmecorrV = _mm256_mul_ps(pmecorrV,beta);
441 velec = _mm256_sub_ps(_mm256_sub_ps(rinv11,sh_ewald),pmecorrV);
442 velec = _mm256_mul_ps(qq11,velec);
443
444 cutoff_mask = _mm256_cmp_ps(rsq11,rcutoff2,_CMP_LT_OQ);
445
446 /* Update potential sum for this i atom from the interaction with this j atom. */
447 velec = _mm256_and_ps(velec,cutoff_mask);
448 velecsum = _mm256_add_ps(velecsum,velec);
449
450 fscal = felec;
451
452 fscal = _mm256_and_ps(fscal,cutoff_mask);
453
454 /* Calculate temporary vectorial force */
455 tx = _mm256_mul_ps(fscal,dx11);
456 ty = _mm256_mul_ps(fscal,dy11);
457 tz = _mm256_mul_ps(fscal,dz11);
458
459 /* Update vectorial force */
460 fix1 = _mm256_add_ps(fix1,tx);
461 fiy1 = _mm256_add_ps(fiy1,ty);
462 fiz1 = _mm256_add_ps(fiz1,tz);
463
464 fjx1 = _mm256_add_ps(fjx1,tx);
465 fjy1 = _mm256_add_ps(fjy1,ty);
466 fjz1 = _mm256_add_ps(fjz1,tz);
467
468 }
469
470 /**************************
471 * CALCULATE INTERACTIONS *
472 **************************/
473
474 if (gmx_mm256_any_lt(rsq12,rcutoff2))
475 {
476
477 r12 = _mm256_mul_ps(rsq12,rinv12);
478
479 /* EWALD ELECTROSTATICS */
480
481 /* Analytical PME correction */
482 zeta2 = _mm256_mul_ps(beta2,rsq12);
483 rinv3 = _mm256_mul_ps(rinvsq12,rinv12);
484 pmecorrF = avx256_pmecorrF_f(zeta2);
485 felec = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
486 felec = _mm256_mul_ps(qq12,felec);
487 pmecorrV = avx256_pmecorrV_f(zeta2);
488 pmecorrV = _mm256_mul_ps(pmecorrV,beta);
489 velec = _mm256_sub_ps(_mm256_sub_ps(rinv12,sh_ewald),pmecorrV);
490 velec = _mm256_mul_ps(qq12,velec);
491
492 cutoff_mask = _mm256_cmp_ps(rsq12,rcutoff2,_CMP_LT_OQ);
493
494 /* Update potential sum for this i atom from the interaction with this j atom. */
495 velec = _mm256_and_ps(velec,cutoff_mask);
496 velecsum = _mm256_add_ps(velecsum,velec);
497
498 fscal = felec;
499
500 fscal = _mm256_and_ps(fscal,cutoff_mask);
501
502 /* Calculate temporary vectorial force */
503 tx = _mm256_mul_ps(fscal,dx12);
504 ty = _mm256_mul_ps(fscal,dy12);
505 tz = _mm256_mul_ps(fscal,dz12);
506
507 /* Update vectorial force */
508 fix1 = _mm256_add_ps(fix1,tx);
509 fiy1 = _mm256_add_ps(fiy1,ty);
510 fiz1 = _mm256_add_ps(fiz1,tz);
511
512 fjx2 = _mm256_add_ps(fjx2,tx);
513 fjy2 = _mm256_add_ps(fjy2,ty);
514 fjz2 = _mm256_add_ps(fjz2,tz);
515
516 }
517
518 /**************************
519 * CALCULATE INTERACTIONS *
520 **************************/
521
522 if (gmx_mm256_any_lt(rsq13,rcutoff2))
523 {
524
525 r13 = _mm256_mul_ps(rsq13,rinv13);
526
527 /* EWALD ELECTROSTATICS */
528
529 /* Analytical PME correction */
530 zeta2 = _mm256_mul_ps(beta2,rsq13);
531 rinv3 = _mm256_mul_ps(rinvsq13,rinv13);
532 pmecorrF = avx256_pmecorrF_f(zeta2);
533 felec = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
534 felec = _mm256_mul_ps(qq13,felec);
535 pmecorrV = avx256_pmecorrV_f(zeta2);
536 pmecorrV = _mm256_mul_ps(pmecorrV,beta);
537 velec = _mm256_sub_ps(_mm256_sub_ps(rinv13,sh_ewald),pmecorrV);
538 velec = _mm256_mul_ps(qq13,velec);
539
540 cutoff_mask = _mm256_cmp_ps(rsq13,rcutoff2,_CMP_LT_OQ);
541
542 /* Update potential sum for this i atom from the interaction with this j atom. */
543 velec = _mm256_and_ps(velec,cutoff_mask);
544 velecsum = _mm256_add_ps(velecsum,velec);
545
546 fscal = felec;
547
548 fscal = _mm256_and_ps(fscal,cutoff_mask);
549
550 /* Calculate temporary vectorial force */
551 tx = _mm256_mul_ps(fscal,dx13);
552 ty = _mm256_mul_ps(fscal,dy13);
553 tz = _mm256_mul_ps(fscal,dz13);
554
555 /* Update vectorial force */
556 fix1 = _mm256_add_ps(fix1,tx);
557 fiy1 = _mm256_add_ps(fiy1,ty);
558 fiz1 = _mm256_add_ps(fiz1,tz);
559
560 fjx3 = _mm256_add_ps(fjx3,tx);
561 fjy3 = _mm256_add_ps(fjy3,ty);
562 fjz3 = _mm256_add_ps(fjz3,tz);
563
564 }
565
566 /**************************
567 * CALCULATE INTERACTIONS *
568 **************************/
569
570 if (gmx_mm256_any_lt(rsq21,rcutoff2))
571 {
572
573 r21 = _mm256_mul_ps(rsq21,rinv21);
574
575 /* EWALD ELECTROSTATICS */
576
577 /* Analytical PME correction */
578 zeta2 = _mm256_mul_ps(beta2,rsq21);
579 rinv3 = _mm256_mul_ps(rinvsq21,rinv21);
580 pmecorrF = avx256_pmecorrF_f(zeta2);
581 felec = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
582 felec = _mm256_mul_ps(qq21,felec);
583 pmecorrV = avx256_pmecorrV_f(zeta2);
584 pmecorrV = _mm256_mul_ps(pmecorrV,beta);
585 velec = _mm256_sub_ps(_mm256_sub_ps(rinv21,sh_ewald),pmecorrV);
586 velec = _mm256_mul_ps(qq21,velec);
587
588 cutoff_mask = _mm256_cmp_ps(rsq21,rcutoff2,_CMP_LT_OQ);
589
590 /* Update potential sum for this i atom from the interaction with this j atom. */
591 velec = _mm256_and_ps(velec,cutoff_mask);
592 velecsum = _mm256_add_ps(velecsum,velec);
593
594 fscal = felec;
595
596 fscal = _mm256_and_ps(fscal,cutoff_mask);
597
598 /* Calculate temporary vectorial force */
599 tx = _mm256_mul_ps(fscal,dx21);
600 ty = _mm256_mul_ps(fscal,dy21);
601 tz = _mm256_mul_ps(fscal,dz21);
602
603 /* Update vectorial force */
604 fix2 = _mm256_add_ps(fix2,tx);
605 fiy2 = _mm256_add_ps(fiy2,ty);
606 fiz2 = _mm256_add_ps(fiz2,tz);
607
608 fjx1 = _mm256_add_ps(fjx1,tx);
609 fjy1 = _mm256_add_ps(fjy1,ty);
610 fjz1 = _mm256_add_ps(fjz1,tz);
611
612 }
613
614 /**************************
615 * CALCULATE INTERACTIONS *
616 **************************/
617
618 if (gmx_mm256_any_lt(rsq22,rcutoff2))
619 {
620
621 r22 = _mm256_mul_ps(rsq22,rinv22);
622
623 /* EWALD ELECTROSTATICS */
624
625 /* Analytical PME correction */
626 zeta2 = _mm256_mul_ps(beta2,rsq22);
627 rinv3 = _mm256_mul_ps(rinvsq22,rinv22);
628 pmecorrF = avx256_pmecorrF_f(zeta2);
629 felec = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
630 felec = _mm256_mul_ps(qq22,felec);
631 pmecorrV = avx256_pmecorrV_f(zeta2);
632 pmecorrV = _mm256_mul_ps(pmecorrV,beta);
633 velec = _mm256_sub_ps(_mm256_sub_ps(rinv22,sh_ewald),pmecorrV);
634 velec = _mm256_mul_ps(qq22,velec);
635
636 cutoff_mask = _mm256_cmp_ps(rsq22,rcutoff2,_CMP_LT_OQ);
637
638 /* Update potential sum for this i atom from the interaction with this j atom. */
639 velec = _mm256_and_ps(velec,cutoff_mask);
640 velecsum = _mm256_add_ps(velecsum,velec);
641
642 fscal = felec;
643
644 fscal = _mm256_and_ps(fscal,cutoff_mask);
645
646 /* Calculate temporary vectorial force */
647 tx = _mm256_mul_ps(fscal,dx22);
648 ty = _mm256_mul_ps(fscal,dy22);
649 tz = _mm256_mul_ps(fscal,dz22);
650
651 /* Update vectorial force */
652 fix2 = _mm256_add_ps(fix2,tx);
653 fiy2 = _mm256_add_ps(fiy2,ty);
654 fiz2 = _mm256_add_ps(fiz2,tz);
655
656 fjx2 = _mm256_add_ps(fjx2,tx);
657 fjy2 = _mm256_add_ps(fjy2,ty);
658 fjz2 = _mm256_add_ps(fjz2,tz);
659
660 }
661
662 /**************************
663 * CALCULATE INTERACTIONS *
664 **************************/
665
666 if (gmx_mm256_any_lt(rsq23,rcutoff2))
667 {
668
669 r23 = _mm256_mul_ps(rsq23,rinv23);
670
671 /* EWALD ELECTROSTATICS */
672
673 /* Analytical PME correction */
674 zeta2 = _mm256_mul_ps(beta2,rsq23);
675 rinv3 = _mm256_mul_ps(rinvsq23,rinv23);
676 pmecorrF = avx256_pmecorrF_f(zeta2);
677 felec = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
678 felec = _mm256_mul_ps(qq23,felec);
679 pmecorrV = avx256_pmecorrV_f(zeta2);
680 pmecorrV = _mm256_mul_ps(pmecorrV,beta);
681 velec = _mm256_sub_ps(_mm256_sub_ps(rinv23,sh_ewald),pmecorrV);
682 velec = _mm256_mul_ps(qq23,velec);
683
684 cutoff_mask = _mm256_cmp_ps(rsq23,rcutoff2,_CMP_LT_OQ);
685
686 /* Update potential sum for this i atom from the interaction with this j atom. */
687 velec = _mm256_and_ps(velec,cutoff_mask);
688 velecsum = _mm256_add_ps(velecsum,velec);
689
690 fscal = felec;
691
692 fscal = _mm256_and_ps(fscal,cutoff_mask);
693
694 /* Calculate temporary vectorial force */
695 tx = _mm256_mul_ps(fscal,dx23);
696 ty = _mm256_mul_ps(fscal,dy23);
697 tz = _mm256_mul_ps(fscal,dz23);
698
699 /* Update vectorial force */
700 fix2 = _mm256_add_ps(fix2,tx);
701 fiy2 = _mm256_add_ps(fiy2,ty);
702 fiz2 = _mm256_add_ps(fiz2,tz);
703
704 fjx3 = _mm256_add_ps(fjx3,tx);
705 fjy3 = _mm256_add_ps(fjy3,ty);
706 fjz3 = _mm256_add_ps(fjz3,tz);
707
708 }
709
710 /**************************
711 * CALCULATE INTERACTIONS *
712 **************************/
713
714 if (gmx_mm256_any_lt(rsq31,rcutoff2))
715 {
716
717 r31 = _mm256_mul_ps(rsq31,rinv31);
718
719 /* EWALD ELECTROSTATICS */
720
721 /* Analytical PME correction */
722 zeta2 = _mm256_mul_ps(beta2,rsq31);
723 rinv3 = _mm256_mul_ps(rinvsq31,rinv31);
724 pmecorrF = avx256_pmecorrF_f(zeta2);
725 felec = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
726 felec = _mm256_mul_ps(qq31,felec);
727 pmecorrV = avx256_pmecorrV_f(zeta2);
728 pmecorrV = _mm256_mul_ps(pmecorrV,beta);
729 velec = _mm256_sub_ps(_mm256_sub_ps(rinv31,sh_ewald),pmecorrV);
730 velec = _mm256_mul_ps(qq31,velec);
731
732 cutoff_mask = _mm256_cmp_ps(rsq31,rcutoff2,_CMP_LT_OQ);
733
734 /* Update potential sum for this i atom from the interaction with this j atom. */
735 velec = _mm256_and_ps(velec,cutoff_mask);
736 velecsum = _mm256_add_ps(velecsum,velec);
737
738 fscal = felec;
739
740 fscal = _mm256_and_ps(fscal,cutoff_mask);
741
742 /* Calculate temporary vectorial force */
743 tx = _mm256_mul_ps(fscal,dx31);
744 ty = _mm256_mul_ps(fscal,dy31);
745 tz = _mm256_mul_ps(fscal,dz31);
746
747 /* Update vectorial force */
748 fix3 = _mm256_add_ps(fix3,tx);
749 fiy3 = _mm256_add_ps(fiy3,ty);
750 fiz3 = _mm256_add_ps(fiz3,tz);
751
752 fjx1 = _mm256_add_ps(fjx1,tx);
753 fjy1 = _mm256_add_ps(fjy1,ty);
754 fjz1 = _mm256_add_ps(fjz1,tz);
755
756 }
757
758 /**************************
759 * CALCULATE INTERACTIONS *
760 **************************/
761
762 if (gmx_mm256_any_lt(rsq32,rcutoff2))
763 {
764
765 r32 = _mm256_mul_ps(rsq32,rinv32);
766
767 /* EWALD ELECTROSTATICS */
768
769 /* Analytical PME correction */
770 zeta2 = _mm256_mul_ps(beta2,rsq32);
771 rinv3 = _mm256_mul_ps(rinvsq32,rinv32);
772 pmecorrF = avx256_pmecorrF_f(zeta2);
773 felec = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
774 felec = _mm256_mul_ps(qq32,felec);
775 pmecorrV = avx256_pmecorrV_f(zeta2);
776 pmecorrV = _mm256_mul_ps(pmecorrV,beta);
777 velec = _mm256_sub_ps(_mm256_sub_ps(rinv32,sh_ewald),pmecorrV);
778 velec = _mm256_mul_ps(qq32,velec);
779
780 cutoff_mask = _mm256_cmp_ps(rsq32,rcutoff2,_CMP_LT_OQ);
781
782 /* Update potential sum for this i atom from the interaction with this j atom. */
783 velec = _mm256_and_ps(velec,cutoff_mask);
784 velecsum = _mm256_add_ps(velecsum,velec);
785
786 fscal = felec;
787
788 fscal = _mm256_and_ps(fscal,cutoff_mask);
789
790 /* Calculate temporary vectorial force */
791 tx = _mm256_mul_ps(fscal,dx32);
792 ty = _mm256_mul_ps(fscal,dy32);
793 tz = _mm256_mul_ps(fscal,dz32);
794
795 /* Update vectorial force */
796 fix3 = _mm256_add_ps(fix3,tx);
797 fiy3 = _mm256_add_ps(fiy3,ty);
798 fiz3 = _mm256_add_ps(fiz3,tz);
799
800 fjx2 = _mm256_add_ps(fjx2,tx);
801 fjy2 = _mm256_add_ps(fjy2,ty);
802 fjz2 = _mm256_add_ps(fjz2,tz);
803
804 }
805
806 /**************************
807 * CALCULATE INTERACTIONS *
808 **************************/
809
810 if (gmx_mm256_any_lt(rsq33,rcutoff2))
811 {
812
813 r33 = _mm256_mul_ps(rsq33,rinv33);
814
815 /* EWALD ELECTROSTATICS */
816
817 /* Analytical PME correction */
818 zeta2 = _mm256_mul_ps(beta2,rsq33);
819 rinv3 = _mm256_mul_ps(rinvsq33,rinv33);
820 pmecorrF = avx256_pmecorrF_f(zeta2);
821 felec = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
822 felec = _mm256_mul_ps(qq33,felec);
823 pmecorrV = avx256_pmecorrV_f(zeta2);
824 pmecorrV = _mm256_mul_ps(pmecorrV,beta);
825 velec = _mm256_sub_ps(_mm256_sub_ps(rinv33,sh_ewald),pmecorrV);
826 velec = _mm256_mul_ps(qq33,velec);
827
828 cutoff_mask = _mm256_cmp_ps(rsq33,rcutoff2,_CMP_LT_OQ);
829
830 /* Update potential sum for this i atom from the interaction with this j atom. */
831 velec = _mm256_and_ps(velec,cutoff_mask);
832 velecsum = _mm256_add_ps(velecsum,velec);
833
834 fscal = felec;
835
836 fscal = _mm256_and_ps(fscal,cutoff_mask);
837
838 /* Calculate temporary vectorial force */
839 tx = _mm256_mul_ps(fscal,dx33);
840 ty = _mm256_mul_ps(fscal,dy33);
841 tz = _mm256_mul_ps(fscal,dz33);
842
843 /* Update vectorial force */
844 fix3 = _mm256_add_ps(fix3,tx);
845 fiy3 = _mm256_add_ps(fiy3,ty);
846 fiz3 = _mm256_add_ps(fiz3,tz);
847
848 fjx3 = _mm256_add_ps(fjx3,tx);
849 fjy3 = _mm256_add_ps(fjy3,ty);
850 fjz3 = _mm256_add_ps(fjz3,tz);
851
852 }
853
854 fjptrA = f+j_coord_offsetA;
855 fjptrB = f+j_coord_offsetB;
856 fjptrC = f+j_coord_offsetC;
857 fjptrD = f+j_coord_offsetD;
858 fjptrE = f+j_coord_offsetE;
859 fjptrF = f+j_coord_offsetF;
860 fjptrG = f+j_coord_offsetG;
861 fjptrH = f+j_coord_offsetH;
862
863 gmx_mm256_decrement_4rvec_8ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjptrE,fjptrF,fjptrG,fjptrH,
864 fjx0,fjy0,fjz0,fjx1,fjy1,fjz1,
865 fjx2,fjy2,fjz2,fjx3,fjy3,fjz3);
866
867 /* Inner loop uses 1046 flops */
868 }
869
870 if(jidx<j_index_end)
871 {
872
873 /* Get j neighbor index, and coordinate index */
874 jnrlistA = jjnr[jidx];
875 jnrlistB = jjnr[jidx+1];
876 jnrlistC = jjnr[jidx+2];
877 jnrlistD = jjnr[jidx+3];
878 jnrlistE = jjnr[jidx+4];
879 jnrlistF = jjnr[jidx+5];
880 jnrlistG = jjnr[jidx+6];
881 jnrlistH = jjnr[jidx+7];
882 /* Sign of each element will be negative for non-real atoms.
883 * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
884 * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
885 */
886 dummy_mask = gmx_mm256_set_m128(gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx+4)),_mm_setzero_si128())),
887 gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128())));
888
889 jnrA = (jnrlistA>=0) ? jnrlistA : 0;
890 jnrB = (jnrlistB>=0) ? jnrlistB : 0;
891 jnrC = (jnrlistC>=0) ? jnrlistC : 0;
892 jnrD = (jnrlistD>=0) ? jnrlistD : 0;
893 jnrE = (jnrlistE>=0) ? jnrlistE : 0;
894 jnrF = (jnrlistF>=0) ? jnrlistF : 0;
895 jnrG = (jnrlistG>=0) ? jnrlistG : 0;
896 jnrH = (jnrlistH>=0) ? jnrlistH : 0;
897 j_coord_offsetA = DIM*jnrA;
898 j_coord_offsetB = DIM*jnrB;
899 j_coord_offsetC = DIM*jnrC;
900 j_coord_offsetD = DIM*jnrD;
901 j_coord_offsetE = DIM*jnrE;
902 j_coord_offsetF = DIM*jnrF;
903 j_coord_offsetG = DIM*jnrG;
904 j_coord_offsetH = DIM*jnrH;
905
906 /* load j atom coordinates */
907 gmx_mm256_load_4rvec_8ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
908 x+j_coord_offsetC,x+j_coord_offsetD,
909 x+j_coord_offsetE,x+j_coord_offsetF,
910 x+j_coord_offsetG,x+j_coord_offsetH,
911 &jx0,&jy0,&jz0,&jx1,&jy1,&jz1,&jx2,
912 &jy2,&jz2,&jx3,&jy3,&jz3);
913
914 /* Calculate displacement vector */
915 dx00 = _mm256_sub_ps(ix0,jx0);
916 dy00 = _mm256_sub_ps(iy0,jy0);
917 dz00 = _mm256_sub_ps(iz0,jz0);
918 dx11 = _mm256_sub_ps(ix1,jx1);
919 dy11 = _mm256_sub_ps(iy1,jy1);
920 dz11 = _mm256_sub_ps(iz1,jz1);
921 dx12 = _mm256_sub_ps(ix1,jx2);
922 dy12 = _mm256_sub_ps(iy1,jy2);
923 dz12 = _mm256_sub_ps(iz1,jz2);
924 dx13 = _mm256_sub_ps(ix1,jx3);
925 dy13 = _mm256_sub_ps(iy1,jy3);
926 dz13 = _mm256_sub_ps(iz1,jz3);
927 dx21 = _mm256_sub_ps(ix2,jx1);
928 dy21 = _mm256_sub_ps(iy2,jy1);
929 dz21 = _mm256_sub_ps(iz2,jz1);
930 dx22 = _mm256_sub_ps(ix2,jx2);
931 dy22 = _mm256_sub_ps(iy2,jy2);
932 dz22 = _mm256_sub_ps(iz2,jz2);
933 dx23 = _mm256_sub_ps(ix2,jx3);
934 dy23 = _mm256_sub_ps(iy2,jy3);
935 dz23 = _mm256_sub_ps(iz2,jz3);
936 dx31 = _mm256_sub_ps(ix3,jx1);
937 dy31 = _mm256_sub_ps(iy3,jy1);
938 dz31 = _mm256_sub_ps(iz3,jz1);
939 dx32 = _mm256_sub_ps(ix3,jx2);
940 dy32 = _mm256_sub_ps(iy3,jy2);
941 dz32 = _mm256_sub_ps(iz3,jz2);
942 dx33 = _mm256_sub_ps(ix3,jx3);
943 dy33 = _mm256_sub_ps(iy3,jy3);
944 dz33 = _mm256_sub_ps(iz3,jz3);
945
946 /* Calculate squared distance and things based on it */
947 rsq00 = gmx_mm256_calc_rsq_ps(dx00,dy00,dz00);
948 rsq11 = gmx_mm256_calc_rsq_ps(dx11,dy11,dz11);
949 rsq12 = gmx_mm256_calc_rsq_ps(dx12,dy12,dz12);
950 rsq13 = gmx_mm256_calc_rsq_ps(dx13,dy13,dz13);
951 rsq21 = gmx_mm256_calc_rsq_ps(dx21,dy21,dz21);
952 rsq22 = gmx_mm256_calc_rsq_ps(dx22,dy22,dz22);
953 rsq23 = gmx_mm256_calc_rsq_ps(dx23,dy23,dz23);
954 rsq31 = gmx_mm256_calc_rsq_ps(dx31,dy31,dz31);
955 rsq32 = gmx_mm256_calc_rsq_ps(dx32,dy32,dz32);
956 rsq33 = gmx_mm256_calc_rsq_ps(dx33,dy33,dz33);
957
958 rinv00 = avx256_invsqrt_f(rsq00);
959 rinv11 = avx256_invsqrt_f(rsq11);
960 rinv12 = avx256_invsqrt_f(rsq12);
961 rinv13 = avx256_invsqrt_f(rsq13);
962 rinv21 = avx256_invsqrt_f(rsq21);
963 rinv22 = avx256_invsqrt_f(rsq22);
964 rinv23 = avx256_invsqrt_f(rsq23);
965 rinv31 = avx256_invsqrt_f(rsq31);
966 rinv32 = avx256_invsqrt_f(rsq32);
967 rinv33 = avx256_invsqrt_f(rsq33);
968
969 rinvsq00 = _mm256_mul_ps(rinv00,rinv00);
970 rinvsq11 = _mm256_mul_ps(rinv11,rinv11);
971 rinvsq12 = _mm256_mul_ps(rinv12,rinv12);
972 rinvsq13 = _mm256_mul_ps(rinv13,rinv13);
973 rinvsq21 = _mm256_mul_ps(rinv21,rinv21);
974 rinvsq22 = _mm256_mul_ps(rinv22,rinv22);
975 rinvsq23 = _mm256_mul_ps(rinv23,rinv23);
976 rinvsq31 = _mm256_mul_ps(rinv31,rinv31);
977 rinvsq32 = _mm256_mul_ps(rinv32,rinv32);
978 rinvsq33 = _mm256_mul_ps(rinv33,rinv33);
979
980 fjx0 = _mm256_setzero_ps();
981 fjy0 = _mm256_setzero_ps();
982 fjz0 = _mm256_setzero_ps();
983 fjx1 = _mm256_setzero_ps();
984 fjy1 = _mm256_setzero_ps();
985 fjz1 = _mm256_setzero_ps();
986 fjx2 = _mm256_setzero_ps();
987 fjy2 = _mm256_setzero_ps();
988 fjz2 = _mm256_setzero_ps();
989 fjx3 = _mm256_setzero_ps();
990 fjy3 = _mm256_setzero_ps();
991 fjz3 = _mm256_setzero_ps();
992
993 /**************************
994 * CALCULATE INTERACTIONS *
995 **************************/
996
997 if (gmx_mm256_any_lt(rsq00,rcutoff2))
998 {
999
1000 r00 = _mm256_mul_ps(rsq00,rinv00);
1001 r00 = _mm256_andnot_ps(dummy_mask,r00);
1002
1003 /* Analytical LJ-PME */
1004 rinvsix = _mm256_mul_ps(_mm256_mul_ps(rinvsq00,rinvsq00),rinvsq00);
1005 ewcljrsq = _mm256_mul_ps(ewclj2,rsq00);
1006 ewclj6 = _mm256_mul_ps(ewclj2,_mm256_mul_ps(ewclj2,ewclj2));
1007 exponent = avx256_exp_f(ewcljrsq);
1008 /* poly = exp(-(beta*r)^2) * (1 + (beta*r)^2 + (beta*r)^4 /2) */
1009 poly = _mm256_mul_ps(exponent,_mm256_add_ps(_mm256_sub_ps(one,ewcljrsq),_mm256_mul_ps(_mm256_mul_ps(ewcljrsq,ewcljrsq),one_half)));
1010 /* vvdw6 = [C6 - C6grid * (1-poly)]/r6 */
1011 vvdw6 = _mm256_mul_ps(_mm256_sub_ps(c6_00,_mm256_mul_ps(c6grid_00,_mm256_sub_ps(one,poly))),rinvsix);
1012 vvdw12 = _mm256_mul_ps(c12_00,_mm256_mul_ps(rinvsix,rinvsix));
1013 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) ,
1014 _mm256_mul_ps( _mm256_sub_ps(vvdw6,_mm256_add_ps(_mm256_mul_ps(c6_00,sh_vdw_invrcut6),_mm256_mul_ps(c6grid_00,sh_lj_ewald))),one_sixth));
1015 /* fvdw = vvdw12/r - (vvdw6/r + (C6grid * exponent * beta^6)/r) */
1016 fvdw = _mm256_mul_ps(_mm256_sub_ps(vvdw12,_mm256_sub_ps(vvdw6,_mm256_mul_ps(_mm256_mul_ps(c6grid_00,one_sixth),_mm256_mul_ps(exponent,ewclj6)))),rinvsq00);
1017
1018 cutoff_mask = _mm256_cmp_ps(rsq00,rcutoff2,_CMP_LT_OQ);
1019
1020 /* Update potential sum for this i atom from the interaction with this j atom. */
1021 vvdw = _mm256_and_ps(vvdw,cutoff_mask);
1022 vvdw = _mm256_andnot_ps(dummy_mask,vvdw);
1023 vvdwsum = _mm256_add_ps(vvdwsum,vvdw);
1024
1025 fscal = fvdw;
1026
1027 fscal = _mm256_and_ps(fscal,cutoff_mask);
1028
1029 fscal = _mm256_andnot_ps(dummy_mask,fscal);
1030
1031 /* Calculate temporary vectorial force */
1032 tx = _mm256_mul_ps(fscal,dx00);
1033 ty = _mm256_mul_ps(fscal,dy00);
1034 tz = _mm256_mul_ps(fscal,dz00);
1035
1036 /* Update vectorial force */
1037 fix0 = _mm256_add_ps(fix0,tx);
1038 fiy0 = _mm256_add_ps(fiy0,ty);
1039 fiz0 = _mm256_add_ps(fiz0,tz);
1040
1041 fjx0 = _mm256_add_ps(fjx0,tx);
1042 fjy0 = _mm256_add_ps(fjy0,ty);
1043 fjz0 = _mm256_add_ps(fjz0,tz);
1044
1045 }
1046
1047 /**************************
1048 * CALCULATE INTERACTIONS *
1049 **************************/
1050
1051 if (gmx_mm256_any_lt(rsq11,rcutoff2))
1052 {
1053
1054 r11 = _mm256_mul_ps(rsq11,rinv11);
1055 r11 = _mm256_andnot_ps(dummy_mask,r11);
1056
1057 /* EWALD ELECTROSTATICS */
1058
1059 /* Analytical PME correction */
1060 zeta2 = _mm256_mul_ps(beta2,rsq11);
1061 rinv3 = _mm256_mul_ps(rinvsq11,rinv11);
1062 pmecorrF = avx256_pmecorrF_f(zeta2);
1063 felec = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
1064 felec = _mm256_mul_ps(qq11,felec);
1065 pmecorrV = avx256_pmecorrV_f(zeta2);
1066 pmecorrV = _mm256_mul_ps(pmecorrV,beta);
1067 velec = _mm256_sub_ps(_mm256_sub_ps(rinv11,sh_ewald),pmecorrV);
1068 velec = _mm256_mul_ps(qq11,velec);
1069
1070 cutoff_mask = _mm256_cmp_ps(rsq11,rcutoff2,_CMP_LT_OQ);
1071
1072 /* Update potential sum for this i atom from the interaction with this j atom. */
1073 velec = _mm256_and_ps(velec,cutoff_mask);
1074 velec = _mm256_andnot_ps(dummy_mask,velec);
1075 velecsum = _mm256_add_ps(velecsum,velec);
1076
1077 fscal = felec;
1078
1079 fscal = _mm256_and_ps(fscal,cutoff_mask);
1080
1081 fscal = _mm256_andnot_ps(dummy_mask,fscal);
1082
1083 /* Calculate temporary vectorial force */
1084 tx = _mm256_mul_ps(fscal,dx11);
1085 ty = _mm256_mul_ps(fscal,dy11);
1086 tz = _mm256_mul_ps(fscal,dz11);
1087
1088 /* Update vectorial force */
1089 fix1 = _mm256_add_ps(fix1,tx);
1090 fiy1 = _mm256_add_ps(fiy1,ty);
1091 fiz1 = _mm256_add_ps(fiz1,tz);
1092
1093 fjx1 = _mm256_add_ps(fjx1,tx);
1094 fjy1 = _mm256_add_ps(fjy1,ty);
1095 fjz1 = _mm256_add_ps(fjz1,tz);
1096
1097 }
1098
1099 /**************************
1100 * CALCULATE INTERACTIONS *
1101 **************************/
1102
1103 if (gmx_mm256_any_lt(rsq12,rcutoff2))
1104 {
1105
1106 r12 = _mm256_mul_ps(rsq12,rinv12);
1107 r12 = _mm256_andnot_ps(dummy_mask,r12);
1108
1109 /* EWALD ELECTROSTATICS */
1110
1111 /* Analytical PME correction */
1112 zeta2 = _mm256_mul_ps(beta2,rsq12);
1113 rinv3 = _mm256_mul_ps(rinvsq12,rinv12);
1114 pmecorrF = avx256_pmecorrF_f(zeta2);
1115 felec = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
1116 felec = _mm256_mul_ps(qq12,felec);
1117 pmecorrV = avx256_pmecorrV_f(zeta2);
1118 pmecorrV = _mm256_mul_ps(pmecorrV,beta);
1119 velec = _mm256_sub_ps(_mm256_sub_ps(rinv12,sh_ewald),pmecorrV);
1120 velec = _mm256_mul_ps(qq12,velec);
1121
1122 cutoff_mask = _mm256_cmp_ps(rsq12,rcutoff2,_CMP_LT_OQ);
1123
1124 /* Update potential sum for this i atom from the interaction with this j atom. */
1125 velec = _mm256_and_ps(velec,cutoff_mask);
1126 velec = _mm256_andnot_ps(dummy_mask,velec);
1127 velecsum = _mm256_add_ps(velecsum,velec);
1128
1129 fscal = felec;
1130
1131 fscal = _mm256_and_ps(fscal,cutoff_mask);
1132
1133 fscal = _mm256_andnot_ps(dummy_mask,fscal);
1134
1135 /* Calculate temporary vectorial force */
1136 tx = _mm256_mul_ps(fscal,dx12);
1137 ty = _mm256_mul_ps(fscal,dy12);
1138 tz = _mm256_mul_ps(fscal,dz12);
1139
1140 /* Update vectorial force */
1141 fix1 = _mm256_add_ps(fix1,tx);
1142 fiy1 = _mm256_add_ps(fiy1,ty);
1143 fiz1 = _mm256_add_ps(fiz1,tz);
1144
1145 fjx2 = _mm256_add_ps(fjx2,tx);
1146 fjy2 = _mm256_add_ps(fjy2,ty);
1147 fjz2 = _mm256_add_ps(fjz2,tz);
1148
1149 }
1150
1151 /**************************
1152 * CALCULATE INTERACTIONS *
1153 **************************/
1154
1155 if (gmx_mm256_any_lt(rsq13,rcutoff2))
1156 {
1157
1158 r13 = _mm256_mul_ps(rsq13,rinv13);
1159 r13 = _mm256_andnot_ps(dummy_mask,r13);
1160
1161 /* EWALD ELECTROSTATICS */
1162
1163 /* Analytical PME correction */
1164 zeta2 = _mm256_mul_ps(beta2,rsq13);
1165 rinv3 = _mm256_mul_ps(rinvsq13,rinv13);
1166 pmecorrF = avx256_pmecorrF_f(zeta2);
1167 felec = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
1168 felec = _mm256_mul_ps(qq13,felec);
1169 pmecorrV = avx256_pmecorrV_f(zeta2);
1170 pmecorrV = _mm256_mul_ps(pmecorrV,beta);
1171 velec = _mm256_sub_ps(_mm256_sub_ps(rinv13,sh_ewald),pmecorrV);
1172 velec = _mm256_mul_ps(qq13,velec);
1173
1174 cutoff_mask = _mm256_cmp_ps(rsq13,rcutoff2,_CMP_LT_OQ);
1175
1176 /* Update potential sum for this i atom from the interaction with this j atom. */
1177 velec = _mm256_and_ps(velec,cutoff_mask);
1178 velec = _mm256_andnot_ps(dummy_mask,velec);
1179 velecsum = _mm256_add_ps(velecsum,velec);
1180
1181 fscal = felec;
1182
1183 fscal = _mm256_and_ps(fscal,cutoff_mask);
1184
1185 fscal = _mm256_andnot_ps(dummy_mask,fscal);
1186
1187 /* Calculate temporary vectorial force */
1188 tx = _mm256_mul_ps(fscal,dx13);
1189 ty = _mm256_mul_ps(fscal,dy13);
1190 tz = _mm256_mul_ps(fscal,dz13);
1191
1192 /* Update vectorial force */
1193 fix1 = _mm256_add_ps(fix1,tx);
1194 fiy1 = _mm256_add_ps(fiy1,ty);
1195 fiz1 = _mm256_add_ps(fiz1,tz);
1196
1197 fjx3 = _mm256_add_ps(fjx3,tx);
1198 fjy3 = _mm256_add_ps(fjy3,ty);
1199 fjz3 = _mm256_add_ps(fjz3,tz);
1200
1201 }
1202
1203 /**************************
1204 * CALCULATE INTERACTIONS *
1205 **************************/
1206
1207 if (gmx_mm256_any_lt(rsq21,rcutoff2))
1208 {
1209
1210 r21 = _mm256_mul_ps(rsq21,rinv21);
1211 r21 = _mm256_andnot_ps(dummy_mask,r21);
1212
1213 /* EWALD ELECTROSTATICS */
1214
1215 /* Analytical PME correction */
1216 zeta2 = _mm256_mul_ps(beta2,rsq21);
1217 rinv3 = _mm256_mul_ps(rinvsq21,rinv21);
1218 pmecorrF = avx256_pmecorrF_f(zeta2);
1219 felec = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
1220 felec = _mm256_mul_ps(qq21,felec);
1221 pmecorrV = avx256_pmecorrV_f(zeta2);
1222 pmecorrV = _mm256_mul_ps(pmecorrV,beta);
1223 velec = _mm256_sub_ps(_mm256_sub_ps(rinv21,sh_ewald),pmecorrV);
1224 velec = _mm256_mul_ps(qq21,velec);
1225
1226 cutoff_mask = _mm256_cmp_ps(rsq21,rcutoff2,_CMP_LT_OQ);
1227
1228 /* Update potential sum for this i atom from the interaction with this j atom. */
1229 velec = _mm256_and_ps(velec,cutoff_mask);
1230 velec = _mm256_andnot_ps(dummy_mask,velec);
1231 velecsum = _mm256_add_ps(velecsum,velec);
1232
1233 fscal = felec;
1234
1235 fscal = _mm256_and_ps(fscal,cutoff_mask);
1236
1237 fscal = _mm256_andnot_ps(dummy_mask,fscal);
1238
1239 /* Calculate temporary vectorial force */
1240 tx = _mm256_mul_ps(fscal,dx21);
1241 ty = _mm256_mul_ps(fscal,dy21);
1242 tz = _mm256_mul_ps(fscal,dz21);
1243
1244 /* Update vectorial force */
1245 fix2 = _mm256_add_ps(fix2,tx);
1246 fiy2 = _mm256_add_ps(fiy2,ty);
1247 fiz2 = _mm256_add_ps(fiz2,tz);
1248
1249 fjx1 = _mm256_add_ps(fjx1,tx);
1250 fjy1 = _mm256_add_ps(fjy1,ty);
1251 fjz1 = _mm256_add_ps(fjz1,tz);
1252
1253 }
1254
1255 /**************************
1256 * CALCULATE INTERACTIONS *
1257 **************************/
1258
1259 if (gmx_mm256_any_lt(rsq22,rcutoff2))
1260 {
1261
1262 r22 = _mm256_mul_ps(rsq22,rinv22);
1263 r22 = _mm256_andnot_ps(dummy_mask,r22);
1264
1265 /* EWALD ELECTROSTATICS */
1266
1267 /* Analytical PME correction */
1268 zeta2 = _mm256_mul_ps(beta2,rsq22);
1269 rinv3 = _mm256_mul_ps(rinvsq22,rinv22);
1270 pmecorrF = avx256_pmecorrF_f(zeta2);
1271 felec = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
1272 felec = _mm256_mul_ps(qq22,felec);
1273 pmecorrV = avx256_pmecorrV_f(zeta2);
1274 pmecorrV = _mm256_mul_ps(pmecorrV,beta);
1275 velec = _mm256_sub_ps(_mm256_sub_ps(rinv22,sh_ewald),pmecorrV);
1276 velec = _mm256_mul_ps(qq22,velec);
1277
1278 cutoff_mask = _mm256_cmp_ps(rsq22,rcutoff2,_CMP_LT_OQ);
1279
1280 /* Update potential sum for this i atom from the interaction with this j atom. */
1281 velec = _mm256_and_ps(velec,cutoff_mask);
1282 velec = _mm256_andnot_ps(dummy_mask,velec);
1283 velecsum = _mm256_add_ps(velecsum,velec);
1284
1285 fscal = felec;
1286
1287 fscal = _mm256_and_ps(fscal,cutoff_mask);
1288
1289 fscal = _mm256_andnot_ps(dummy_mask,fscal);
1290
1291 /* Calculate temporary vectorial force */
1292 tx = _mm256_mul_ps(fscal,dx22);
1293 ty = _mm256_mul_ps(fscal,dy22);
1294 tz = _mm256_mul_ps(fscal,dz22);
1295
1296 /* Update vectorial force */
1297 fix2 = _mm256_add_ps(fix2,tx);
1298 fiy2 = _mm256_add_ps(fiy2,ty);
1299 fiz2 = _mm256_add_ps(fiz2,tz);
1300
1301 fjx2 = _mm256_add_ps(fjx2,tx);
1302 fjy2 = _mm256_add_ps(fjy2,ty);
1303 fjz2 = _mm256_add_ps(fjz2,tz);
1304
1305 }
1306
1307 /**************************
1308 * CALCULATE INTERACTIONS *
1309 **************************/
1310
1311 if (gmx_mm256_any_lt(rsq23,rcutoff2))
1312 {
1313
1314 r23 = _mm256_mul_ps(rsq23,rinv23);
1315 r23 = _mm256_andnot_ps(dummy_mask,r23);
1316
1317 /* EWALD ELECTROSTATICS */
1318
1319 /* Analytical PME correction */
1320 zeta2 = _mm256_mul_ps(beta2,rsq23);
1321 rinv3 = _mm256_mul_ps(rinvsq23,rinv23);
1322 pmecorrF = avx256_pmecorrF_f(zeta2);
1323 felec = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
1324 felec = _mm256_mul_ps(qq23,felec);
1325 pmecorrV = avx256_pmecorrV_f(zeta2);
1326 pmecorrV = _mm256_mul_ps(pmecorrV,beta);
1327 velec = _mm256_sub_ps(_mm256_sub_ps(rinv23,sh_ewald),pmecorrV);
1328 velec = _mm256_mul_ps(qq23,velec);
1329
1330 cutoff_mask = _mm256_cmp_ps(rsq23,rcutoff2,_CMP_LT_OQ);
1331
1332 /* Update potential sum for this i atom from the interaction with this j atom. */
1333 velec = _mm256_and_ps(velec,cutoff_mask);
1334 velec = _mm256_andnot_ps(dummy_mask,velec);
1335 velecsum = _mm256_add_ps(velecsum,velec);
1336
1337 fscal = felec;
1338
1339 fscal = _mm256_and_ps(fscal,cutoff_mask);
1340
1341 fscal = _mm256_andnot_ps(dummy_mask,fscal);
1342
1343 /* Calculate temporary vectorial force */
1344 tx = _mm256_mul_ps(fscal,dx23);
1345 ty = _mm256_mul_ps(fscal,dy23);
1346 tz = _mm256_mul_ps(fscal,dz23);
1347
1348 /* Update vectorial force */
1349 fix2 = _mm256_add_ps(fix2,tx);
1350 fiy2 = _mm256_add_ps(fiy2,ty);
1351 fiz2 = _mm256_add_ps(fiz2,tz);
1352
1353 fjx3 = _mm256_add_ps(fjx3,tx);
1354 fjy3 = _mm256_add_ps(fjy3,ty);
1355 fjz3 = _mm256_add_ps(fjz3,tz);
1356
1357 }
1358
1359 /**************************
1360 * CALCULATE INTERACTIONS *
1361 **************************/
1362
1363 if (gmx_mm256_any_lt(rsq31,rcutoff2))
1364 {
1365
1366 r31 = _mm256_mul_ps(rsq31,rinv31);
1367 r31 = _mm256_andnot_ps(dummy_mask,r31);
1368
1369 /* EWALD ELECTROSTATICS */
1370
1371 /* Analytical PME correction */
1372 zeta2 = _mm256_mul_ps(beta2,rsq31);
1373 rinv3 = _mm256_mul_ps(rinvsq31,rinv31);
1374 pmecorrF = avx256_pmecorrF_f(zeta2);
1375 felec = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
1376 felec = _mm256_mul_ps(qq31,felec);
1377 pmecorrV = avx256_pmecorrV_f(zeta2);
1378 pmecorrV = _mm256_mul_ps(pmecorrV,beta);
1379 velec = _mm256_sub_ps(_mm256_sub_ps(rinv31,sh_ewald),pmecorrV);
1380 velec = _mm256_mul_ps(qq31,velec);
1381
1382 cutoff_mask = _mm256_cmp_ps(rsq31,rcutoff2,_CMP_LT_OQ);
1383
1384 /* Update potential sum for this i atom from the interaction with this j atom. */
1385 velec = _mm256_and_ps(velec,cutoff_mask);
1386 velec = _mm256_andnot_ps(dummy_mask,velec);
1387 velecsum = _mm256_add_ps(velecsum,velec);
1388
1389 fscal = felec;
1390
1391 fscal = _mm256_and_ps(fscal,cutoff_mask);
1392
1393 fscal = _mm256_andnot_ps(dummy_mask,fscal);
1394
1395 /* Calculate temporary vectorial force */
1396 tx = _mm256_mul_ps(fscal,dx31);
1397 ty = _mm256_mul_ps(fscal,dy31);
1398 tz = _mm256_mul_ps(fscal,dz31);
1399
1400 /* Update vectorial force */
1401 fix3 = _mm256_add_ps(fix3,tx);
1402 fiy3 = _mm256_add_ps(fiy3,ty);
1403 fiz3 = _mm256_add_ps(fiz3,tz);
1404
1405 fjx1 = _mm256_add_ps(fjx1,tx);
1406 fjy1 = _mm256_add_ps(fjy1,ty);
1407 fjz1 = _mm256_add_ps(fjz1,tz);
1408
1409 }
1410
1411 /**************************
1412 * CALCULATE INTERACTIONS *
1413 **************************/
1414
1415 if (gmx_mm256_any_lt(rsq32,rcutoff2))
1416 {
1417
1418 r32 = _mm256_mul_ps(rsq32,rinv32);
1419 r32 = _mm256_andnot_ps(dummy_mask,r32);
1420
1421 /* EWALD ELECTROSTATICS */
1422
1423 /* Analytical PME correction */
1424 zeta2 = _mm256_mul_ps(beta2,rsq32);
1425 rinv3 = _mm256_mul_ps(rinvsq32,rinv32);
1426 pmecorrF = avx256_pmecorrF_f(zeta2);
1427 felec = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
1428 felec = _mm256_mul_ps(qq32,felec);
1429 pmecorrV = avx256_pmecorrV_f(zeta2);
1430 pmecorrV = _mm256_mul_ps(pmecorrV,beta);
1431 velec = _mm256_sub_ps(_mm256_sub_ps(rinv32,sh_ewald),pmecorrV);
1432 velec = _mm256_mul_ps(qq32,velec);
1433
1434 cutoff_mask = _mm256_cmp_ps(rsq32,rcutoff2,_CMP_LT_OQ);
1435
1436 /* Update potential sum for this i atom from the interaction with this j atom. */
1437 velec = _mm256_and_ps(velec,cutoff_mask);
1438 velec = _mm256_andnot_ps(dummy_mask,velec);
1439 velecsum = _mm256_add_ps(velecsum,velec);
1440
1441 fscal = felec;
1442
1443 fscal = _mm256_and_ps(fscal,cutoff_mask);
1444
1445 fscal = _mm256_andnot_ps(dummy_mask,fscal);
1446
1447 /* Calculate temporary vectorial force */
1448 tx = _mm256_mul_ps(fscal,dx32);
1449 ty = _mm256_mul_ps(fscal,dy32);
1450 tz = _mm256_mul_ps(fscal,dz32);
1451
1452 /* Update vectorial force */
1453 fix3 = _mm256_add_ps(fix3,tx);
1454 fiy3 = _mm256_add_ps(fiy3,ty);
1455 fiz3 = _mm256_add_ps(fiz3,tz);
1456
1457 fjx2 = _mm256_add_ps(fjx2,tx);
1458 fjy2 = _mm256_add_ps(fjy2,ty);
1459 fjz2 = _mm256_add_ps(fjz2,tz);
1460
1461 }
1462
1463 /**************************
1464 * CALCULATE INTERACTIONS *
1465 **************************/
1466
1467 if (gmx_mm256_any_lt(rsq33,rcutoff2))
1468 {
1469
1470 r33 = _mm256_mul_ps(rsq33,rinv33);
1471 r33 = _mm256_andnot_ps(dummy_mask,r33);
1472
1473 /* EWALD ELECTROSTATICS */
1474
1475 /* Analytical PME correction */
1476 zeta2 = _mm256_mul_ps(beta2,rsq33);
1477 rinv3 = _mm256_mul_ps(rinvsq33,rinv33);
1478 pmecorrF = avx256_pmecorrF_f(zeta2);
1479 felec = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
1480 felec = _mm256_mul_ps(qq33,felec);
1481 pmecorrV = avx256_pmecorrV_f(zeta2);
1482 pmecorrV = _mm256_mul_ps(pmecorrV,beta);
1483 velec = _mm256_sub_ps(_mm256_sub_ps(rinv33,sh_ewald),pmecorrV);
1484 velec = _mm256_mul_ps(qq33,velec);
1485
1486 cutoff_mask = _mm256_cmp_ps(rsq33,rcutoff2,_CMP_LT_OQ);
1487
1488 /* Update potential sum for this i atom from the interaction with this j atom. */
1489 velec = _mm256_and_ps(velec,cutoff_mask);
1490 velec = _mm256_andnot_ps(dummy_mask,velec);
1491 velecsum = _mm256_add_ps(velecsum,velec);
1492
1493 fscal = felec;
1494
1495 fscal = _mm256_and_ps(fscal,cutoff_mask);
1496
1497 fscal = _mm256_andnot_ps(dummy_mask,fscal);
1498
1499 /* Calculate temporary vectorial force */
1500 tx = _mm256_mul_ps(fscal,dx33);
1501 ty = _mm256_mul_ps(fscal,dy33);
1502 tz = _mm256_mul_ps(fscal,dz33);
1503
1504 /* Update vectorial force */
1505 fix3 = _mm256_add_ps(fix3,tx);
1506 fiy3 = _mm256_add_ps(fiy3,ty);
1507 fiz3 = _mm256_add_ps(fiz3,tz);
1508
1509 fjx3 = _mm256_add_ps(fjx3,tx);
1510 fjy3 = _mm256_add_ps(fjy3,ty);
1511 fjz3 = _mm256_add_ps(fjz3,tz);
1512
1513 }
1514
1515 fjptrA = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
1516 fjptrB = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
1517 fjptrC = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
1518 fjptrD = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
1519 fjptrE = (jnrlistE>=0) ? f+j_coord_offsetE : scratch;
1520 fjptrF = (jnrlistF>=0) ? f+j_coord_offsetF : scratch;
1521 fjptrG = (jnrlistG>=0) ? f+j_coord_offsetG : scratch;
1522 fjptrH = (jnrlistH>=0) ? f+j_coord_offsetH : scratch;
1523
1524 gmx_mm256_decrement_4rvec_8ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjptrE,fjptrF,fjptrG,fjptrH,
1525 fjx0,fjy0,fjz0,fjx1,fjy1,fjz1,
1526 fjx2,fjy2,fjz2,fjx3,fjy3,fjz3);
1527
1528 /* Inner loop uses 1056 flops */
1529 }
1530
1531 /* End of innermost loop */
1532
1533 gmx_mm256_update_iforce_4atom_swizzle_ps(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,fix3,fiy3,fiz3,
1534 f+i_coord_offset,fshift+i_shift_offset);
1535
1536 ggid = gid[iidx];
1537 /* Update potential energies */
1538 gmx_mm256_update_1pot_ps(velecsum,kernel_data->energygrp_elec+ggid);
1539 gmx_mm256_update_1pot_ps(vvdwsum,kernel_data->energygrp_vdw+ggid);
1540
1541 /* Increment number of inner iterations */
1542 inneriter += j_index_end - j_index_start;
1543
1544 /* Outer loop uses 26 flops */
1545 }
1546
1547 /* Increment number of outer iterations */
1548 outeriter += nri;
1549
1550 /* Update outer/inner flops */
1551
1552 inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W4W4_VF,outeriter*26 + inneriter*1056);
1553 }
1554 /*
1555 * Gromacs nonbonded kernel: nb_kernel_ElecEwSh_VdwLJEwSh_GeomW4W4_F_avx_256_single
1556 * Electrostatics interaction: Ewald
1557 * VdW interaction: LJEwald
1558 * Geometry: Water4-Water4
1559 * Calculate force/pot: Force
1560 */
1561 void
1562 nb_kernel_ElecEwSh_VdwLJEwSh_GeomW4W4_F_avx_256_single
1563 (t_nblist * gmx_restrict nlist,
1564 rvec * gmx_restrict xx,
1565 rvec * gmx_restrict ff,
1566 struct t_forcerec * gmx_restrict fr,
1567 t_mdatoms * gmx_restrict mdatoms,
1568 nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
1569 t_nrnb * gmx_restrict nrnb)
1570 {
1571 /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
1572 * just 0 for non-waters.
1573 * Suffixes A,B,C,D,E,F,G,H refer to j loop unrolling done with AVX, e.g. for the eight different
1574 * jnr indices corresponding to data put in the four positions in the SIMD register.
1575 */
1576 int i_shift_offset,i_coord_offset,outeriter,inneriter;
1577 int j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
1578 int jnrA,jnrB,jnrC,jnrD;
1579 int jnrE,jnrF,jnrG,jnrH;
1580 int jnrlistA,jnrlistB,jnrlistC,jnrlistD;
1581 int jnrlistE,jnrlistF,jnrlistG,jnrlistH;
1582 int j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
1583 int j_coord_offsetE,j_coord_offsetF,j_coord_offsetG,j_coord_offsetH;
1584 int *iinr,*jindex,*jjnr,*shiftidx,*gid;
1585 real rcutoff_scalar;
1586 real *shiftvec,*fshift,*x,*f;
1587 real *fjptrA,*fjptrB,*fjptrC,*fjptrD,*fjptrE,*fjptrF,*fjptrG,*fjptrH;
1588 real scratch[4*DIM];
1589 __m256 tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
1590 real * vdwioffsetptr0;
1591 real * vdwgridioffsetptr0;
1592 __m256 ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
1593 real * vdwioffsetptr1;
1594 real * vdwgridioffsetptr1;
1595 __m256 ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
1596 real * vdwioffsetptr2;
1597 real * vdwgridioffsetptr2;
1598 __m256 ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
1599 real * vdwioffsetptr3;
1600 real * vdwgridioffsetptr3;
1601 __m256 ix3,iy3,iz3,fix3,fiy3,fiz3,iq3,isai3;
1602 int vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D,vdwjidx0E,vdwjidx0F,vdwjidx0G,vdwjidx0H;
1603 __m256 jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
1604 int vdwjidx1A,vdwjidx1B,vdwjidx1C,vdwjidx1D,vdwjidx1E,vdwjidx1F,vdwjidx1G,vdwjidx1H;
1605 __m256 jx1,jy1,jz1,fjx1,fjy1,fjz1,jq1,isaj1;
1606 int vdwjidx2A,vdwjidx2B,vdwjidx2C,vdwjidx2D,vdwjidx2E,vdwjidx2F,vdwjidx2G,vdwjidx2H;
1607 __m256 jx2,jy2,jz2,fjx2,fjy2,fjz2,jq2,isaj2;
1608 int vdwjidx3A,vdwjidx3B,vdwjidx3C,vdwjidx3D,vdwjidx3E,vdwjidx3F,vdwjidx3G,vdwjidx3H;
1609 __m256 jx3,jy3,jz3,fjx3,fjy3,fjz3,jq3,isaj3;
1610 __m256 dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
1611 __m256 dx11,dy11,dz11,rsq11,rinv11,rinvsq11,r11,qq11,c6_11,c12_11;
1612 __m256 dx12,dy12,dz12,rsq12,rinv12,rinvsq12,r12,qq12,c6_12,c12_12;
1613 __m256 dx13,dy13,dz13,rsq13,rinv13,rinvsq13,r13,qq13,c6_13,c12_13;
1614 __m256 dx21,dy21,dz21,rsq21,rinv21,rinvsq21,r21,qq21,c6_21,c12_21;
1615 __m256 dx22,dy22,dz22,rsq22,rinv22,rinvsq22,r22,qq22,c6_22,c12_22;
1616 __m256 dx23,dy23,dz23,rsq23,rinv23,rinvsq23,r23,qq23,c6_23,c12_23;
1617 __m256 dx31,dy31,dz31,rsq31,rinv31,rinvsq31,r31,qq31,c6_31,c12_31;
1618 __m256 dx32,dy32,dz32,rsq32,rinv32,rinvsq32,r32,qq32,c6_32,c12_32;
1619 __m256 dx33,dy33,dz33,rsq33,rinv33,rinvsq33,r33,qq33,c6_33,c12_33;
1620 __m256 velec,felec,velecsum,facel,crf,krf,krf2;
1621 real *charge;
1622 int nvdwtype;
1623 __m256 rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
1624 int *vdwtype;
1625 real *vdwparam;
1626 __m256 one_sixth = _mm256_set1_ps(1.0/6.0);
1627 __m256 one_twelfth = _mm256_set1_ps(1.0/12.0);
1628 __m256 c6grid_00;
1629 __m256 c6grid_11;
1630 __m256 c6grid_12;
1631 __m256 c6grid_13;
1632 __m256 c6grid_21;
1633 __m256 c6grid_22;
1634 __m256 c6grid_23;
1635 __m256 c6grid_31;
1636 __m256 c6grid_32;
1637 __m256 c6grid_33;
1638 real *vdwgridparam;
1639 __m256 ewclj,ewclj2,ewclj6,ewcljrsq,poly,exponent,f6A,f6B,sh_lj_ewald;
1640 __m256 one_half = _mm256_set1_ps(0.5);
1641 __m256 minus_one = _mm256_set1_ps(-1.0);
1642 __m256i ewitab;
1643 __m128i ewitab_lo,ewitab_hi;
1644 __m256 ewtabscale,eweps,sh_ewald,ewrt,ewtabhalfspace,ewtabF,ewtabFn,ewtabD,ewtabV;
1645 __m256 beta,beta2,beta3,zeta2,pmecorrF,pmecorrV,rinv3;
1646 real *ewtab;
1647 __m256 dummy_mask,cutoff_mask;
1648 __m256 signbit = _mm256_castsi256_ps( _mm256_set1_epi32(0x80000000) );
1649 __m256 one = _mm256_set1_ps(1.0);
1650 __m256 two = _mm256_set1_ps(2.0);
1651 x = xx[0];
1652 f = ff[0];
1653
1654 nri = nlist->nri;
1655 iinr = nlist->iinr;
1656 jindex = nlist->jindex;
1657 jjnr = nlist->jjnr;
1658 shiftidx = nlist->shift;
1659 gid = nlist->gid;
1660 shiftvec = fr->shift_vec[0];
1661 fshift = fr->fshift[0];
1662 facel = _mm256_set1_ps(fr->ic->epsfac);
1663 charge = mdatoms->chargeA;
1664 nvdwtype = fr->ntype;
1665 vdwparam = fr->nbfp;
1666 vdwtype = mdatoms->typeA;
1667 vdwgridparam = fr->ljpme_c6grid;
1668 sh_lj_ewald = _mm256_set1_ps(fr->ic->sh_lj_ewald);
1669 ewclj = _mm256_set1_ps(fr->ic->ewaldcoeff_lj);
1670 ewclj2 = _mm256_mul_ps(minus_one,_mm256_mul_ps(ewclj,ewclj));
1671
1672 sh_ewald = _mm256_set1_ps(fr->ic->sh_ewald);
1673 beta = _mm256_set1_ps(fr->ic->ewaldcoeff_q);
1674 beta2 = _mm256_mul_ps(beta,beta);
1675 beta3 = _mm256_mul_ps(beta,beta2);
1676
1677 ewtab = fr->ic->tabq_coul_F;
1678 ewtabscale = _mm256_set1_ps(fr->ic->tabq_scale);
1679 ewtabhalfspace = _mm256_set1_ps(0.5/fr->ic->tabq_scale);
1680
1681 /* Setup water-specific parameters */
1682 inr = nlist->iinr[0];
1683 iq1 = _mm256_mul_ps(facel,_mm256_set1_ps(charge[inr+1]));
1684 iq2 = _mm256_mul_ps(facel,_mm256_set1_ps(charge[inr+2]));
1685 iq3 = _mm256_mul_ps(facel,_mm256_set1_ps(charge[inr+3]));
1686 vdwioffsetptr0 = vdwparam+2*nvdwtype*vdwtype[inr+0];
1687 vdwgridioffsetptr0 = vdwgridparam+2*nvdwtype*vdwtype[inr+0];
1688
1689 jq1 = _mm256_set1_ps(charge[inr+1]);
1690 jq2 = _mm256_set1_ps(charge[inr+2]);
1691 jq3 = _mm256_set1_ps(charge[inr+3]);
1692 vdwjidx0A = 2*vdwtype[inr+0];
1693 c6_00 = _mm256_set1_ps(vdwioffsetptr0[vdwjidx0A]);
1694 c12_00 = _mm256_set1_ps(vdwioffsetptr0[vdwjidx0A+1]);
1695 c6grid_00 = _mm256_set1_ps(vdwgridioffsetptr0[vdwjidx0A]);
1696 qq11 = _mm256_mul_ps(iq1,jq1);
1697 qq12 = _mm256_mul_ps(iq1,jq2);
1698 qq13 = _mm256_mul_ps(iq1,jq3);
1699 qq21 = _mm256_mul_ps(iq2,jq1);
1700 qq22 = _mm256_mul_ps(iq2,jq2);
1701 qq23 = _mm256_mul_ps(iq2,jq3);
1702 qq31 = _mm256_mul_ps(iq3,jq1);
1703 qq32 = _mm256_mul_ps(iq3,jq2);
1704 qq33 = _mm256_mul_ps(iq3,jq3);
1705
1706 /* When we use explicit cutoffs the value must be identical for elec and VdW, so use elec as an arbitrary choice */
1707 rcutoff_scalar = fr->ic->rcoulomb;
1708 rcutoff = _mm256_set1_ps(rcutoff_scalar);
1709 rcutoff2 = _mm256_mul_ps(rcutoff,rcutoff);
1710
1711 sh_vdw_invrcut6 = _mm256_set1_ps(fr->ic->sh_invrc6);
1712 rvdw = _mm256_set1_ps(fr->ic->rvdw);
1713
1714 /* Avoid stupid compiler warnings */
1715 jnrA = jnrB = jnrC = jnrD = jnrE = jnrF = jnrG = jnrH = 0;
1716 j_coord_offsetA = 0;
1717 j_coord_offsetB = 0;
1718 j_coord_offsetC = 0;
1719 j_coord_offsetD = 0;
1720 j_coord_offsetE = 0;
1721 j_coord_offsetF = 0;
1722 j_coord_offsetG = 0;
1723 j_coord_offsetH = 0;
1724
1725 outeriter = 0;
1726 inneriter = 0;
1727
1728 for(iidx=0;iidx<4*DIM;iidx++)
1729 {
1730 scratch[iidx] = 0.0;
1731 }
1732
1733 /* Start outer loop over neighborlists */
1734 for(iidx=0; iidx<nri; iidx++)
1735 {
1736 /* Load shift vector for this list */
1737 i_shift_offset = DIM*shiftidx[iidx];
1738
1739 /* Load limits for loop over neighbors */
1740 j_index_start = jindex[iidx];
1741 j_index_end = jindex[iidx+1];
1742
1743 /* Get outer coordinate index */
1744 inr = iinr[iidx];
1745 i_coord_offset = DIM*inr;
1746
1747 /* Load i particle coords and add shift vector */
1748 gmx_mm256_load_shift_and_4rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,
1749 &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2,&ix3,&iy3,&iz3);
1750
1751 fix0 = _mm256_setzero_ps();
1752 fiy0 = _mm256_setzero_ps();
1753 fiz0 = _mm256_setzero_ps();
1754 fix1 = _mm256_setzero_ps();
1755 fiy1 = _mm256_setzero_ps();
1756 fiz1 = _mm256_setzero_ps();
1757 fix2 = _mm256_setzero_ps();
1758 fiy2 = _mm256_setzero_ps();
1759 fiz2 = _mm256_setzero_ps();
1760 fix3 = _mm256_setzero_ps();
1761 fiy3 = _mm256_setzero_ps();
1762 fiz3 = _mm256_setzero_ps();
1763
1764 /* Start inner kernel loop */
1765 for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+7]>=0; jidx+=8)
1766 {
1767
1768 /* Get j neighbor index, and coordinate index */
1769 jnrA = jjnr[jidx];
1770 jnrB = jjnr[jidx+1];
1771 jnrC = jjnr[jidx+2];
1772 jnrD = jjnr[jidx+3];
1773 jnrE = jjnr[jidx+4];
1774 jnrF = jjnr[jidx+5];
1775 jnrG = jjnr[jidx+6];
1776 jnrH = jjnr[jidx+7];
1777 j_coord_offsetA = DIM*jnrA;
1778 j_coord_offsetB = DIM*jnrB;
1779 j_coord_offsetC = DIM*jnrC;
1780 j_coord_offsetD = DIM*jnrD;
1781 j_coord_offsetE = DIM*jnrE;
1782 j_coord_offsetF = DIM*jnrF;
1783 j_coord_offsetG = DIM*jnrG;
1784 j_coord_offsetH = DIM*jnrH;
1785
1786 /* load j atom coordinates */
1787 gmx_mm256_load_4rvec_8ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
1788 x+j_coord_offsetC,x+j_coord_offsetD,
1789 x+j_coord_offsetE,x+j_coord_offsetF,
1790 x+j_coord_offsetG,x+j_coord_offsetH,
1791 &jx0,&jy0,&jz0,&jx1,&jy1,&jz1,&jx2,
1792 &jy2,&jz2,&jx3,&jy3,&jz3);
1793
1794 /* Calculate displacement vector */
1795 dx00 = _mm256_sub_ps(ix0,jx0);
1796 dy00 = _mm256_sub_ps(iy0,jy0);
1797 dz00 = _mm256_sub_ps(iz0,jz0);
1798 dx11 = _mm256_sub_ps(ix1,jx1);
1799 dy11 = _mm256_sub_ps(iy1,jy1);
1800 dz11 = _mm256_sub_ps(iz1,jz1);
1801 dx12 = _mm256_sub_ps(ix1,jx2);
1802 dy12 = _mm256_sub_ps(iy1,jy2);
1803 dz12 = _mm256_sub_ps(iz1,jz2);
1804 dx13 = _mm256_sub_ps(ix1,jx3);
1805 dy13 = _mm256_sub_ps(iy1,jy3);
1806 dz13 = _mm256_sub_ps(iz1,jz3);
1807 dx21 = _mm256_sub_ps(ix2,jx1);
1808 dy21 = _mm256_sub_ps(iy2,jy1);
1809 dz21 = _mm256_sub_ps(iz2,jz1);
1810 dx22 = _mm256_sub_ps(ix2,jx2);
1811 dy22 = _mm256_sub_ps(iy2,jy2);
1812 dz22 = _mm256_sub_ps(iz2,jz2);
1813 dx23 = _mm256_sub_ps(ix2,jx3);
1814 dy23 = _mm256_sub_ps(iy2,jy3);
1815 dz23 = _mm256_sub_ps(iz2,jz3);
1816 dx31 = _mm256_sub_ps(ix3,jx1);
1817 dy31 = _mm256_sub_ps(iy3,jy1);
1818 dz31 = _mm256_sub_ps(iz3,jz1);
1819 dx32 = _mm256_sub_ps(ix3,jx2);
1820 dy32 = _mm256_sub_ps(iy3,jy2);
1821 dz32 = _mm256_sub_ps(iz3,jz2);
1822 dx33 = _mm256_sub_ps(ix3,jx3);
1823 dy33 = _mm256_sub_ps(iy3,jy3);
1824 dz33 = _mm256_sub_ps(iz3,jz3);
1825
1826 /* Calculate squared distance and things based on it */
1827 rsq00 = gmx_mm256_calc_rsq_ps(dx00,dy00,dz00);
1828 rsq11 = gmx_mm256_calc_rsq_ps(dx11,dy11,dz11);
1829 rsq12 = gmx_mm256_calc_rsq_ps(dx12,dy12,dz12);
1830 rsq13 = gmx_mm256_calc_rsq_ps(dx13,dy13,dz13);
1831 rsq21 = gmx_mm256_calc_rsq_ps(dx21,dy21,dz21);
1832 rsq22 = gmx_mm256_calc_rsq_ps(dx22,dy22,dz22);
1833 rsq23 = gmx_mm256_calc_rsq_ps(dx23,dy23,dz23);
1834 rsq31 = gmx_mm256_calc_rsq_ps(dx31,dy31,dz31);
1835 rsq32 = gmx_mm256_calc_rsq_ps(dx32,dy32,dz32);
1836 rsq33 = gmx_mm256_calc_rsq_ps(dx33,dy33,dz33);
1837
1838 rinv00 = avx256_invsqrt_f(rsq00);
1839 rinv11 = avx256_invsqrt_f(rsq11);
1840 rinv12 = avx256_invsqrt_f(rsq12);
1841 rinv13 = avx256_invsqrt_f(rsq13);
1842 rinv21 = avx256_invsqrt_f(rsq21);
1843 rinv22 = avx256_invsqrt_f(rsq22);
1844 rinv23 = avx256_invsqrt_f(rsq23);
1845 rinv31 = avx256_invsqrt_f(rsq31);
1846 rinv32 = avx256_invsqrt_f(rsq32);
1847 rinv33 = avx256_invsqrt_f(rsq33);
1848
1849 rinvsq00 = _mm256_mul_ps(rinv00,rinv00);
1850 rinvsq11 = _mm256_mul_ps(rinv11,rinv11);
1851 rinvsq12 = _mm256_mul_ps(rinv12,rinv12);
1852 rinvsq13 = _mm256_mul_ps(rinv13,rinv13);
1853 rinvsq21 = _mm256_mul_ps(rinv21,rinv21);
1854 rinvsq22 = _mm256_mul_ps(rinv22,rinv22);
1855 rinvsq23 = _mm256_mul_ps(rinv23,rinv23);
1856 rinvsq31 = _mm256_mul_ps(rinv31,rinv31);
1857 rinvsq32 = _mm256_mul_ps(rinv32,rinv32);
1858 rinvsq33 = _mm256_mul_ps(rinv33,rinv33);
1859
1860 fjx0 = _mm256_setzero_ps();
1861 fjy0 = _mm256_setzero_ps();
1862 fjz0 = _mm256_setzero_ps();
1863 fjx1 = _mm256_setzero_ps();
1864 fjy1 = _mm256_setzero_ps();
1865 fjz1 = _mm256_setzero_ps();
1866 fjx2 = _mm256_setzero_ps();
1867 fjy2 = _mm256_setzero_ps();
1868 fjz2 = _mm256_setzero_ps();
1869 fjx3 = _mm256_setzero_ps();
1870 fjy3 = _mm256_setzero_ps();
1871 fjz3 = _mm256_setzero_ps();
1872
1873 /**************************
1874 * CALCULATE INTERACTIONS *
1875 **************************/
1876
1877 if (gmx_mm256_any_lt(rsq00,rcutoff2))
1878 {
1879
1880 r00 = _mm256_mul_ps(rsq00,rinv00);
1881
1882 /* Analytical LJ-PME */
1883 rinvsix = _mm256_mul_ps(_mm256_mul_ps(rinvsq00,rinvsq00),rinvsq00);
1884 ewcljrsq = _mm256_mul_ps(ewclj2,rsq00);
1885 ewclj6 = _mm256_mul_ps(ewclj2,_mm256_mul_ps(ewclj2,ewclj2));
1886 exponent = avx256_exp_f(ewcljrsq);
1887 /* poly = exp(-(beta*r)^2) * (1 + (beta*r)^2 + (beta*r)^4 /2) */
1888 poly = _mm256_mul_ps(exponent,_mm256_add_ps(_mm256_sub_ps(one,ewcljrsq),_mm256_mul_ps(_mm256_mul_ps(ewcljrsq,ewcljrsq),one_half)));
1889 /* f6A = 6 * C6grid * (1 - poly) */
1890 f6A = _mm256_mul_ps(c6grid_00,_mm256_sub_ps(one,poly));
1891 /* f6B = C6grid * exponent * beta^6 */
1892 f6B = _mm256_mul_ps(_mm256_mul_ps(c6grid_00,one_sixth),_mm256_mul_ps(exponent,ewclj6));
1893 /* fvdw = 12*C12/r13 - ((6*C6 - f6A)/r6 + f6B)/r */
1894 fvdw = _mm256_mul_ps(_mm256_add_ps(_mm256_mul_ps(_mm256_sub_ps(_mm256_mul_ps(c12_00,rinvsix),_mm256_sub_ps(c6_00,f6A)),rinvsix),f6B),rinvsq00);
1895
1896 cutoff_mask = _mm256_cmp_ps(rsq00,rcutoff2,_CMP_LT_OQ);
1897
1898 fscal = fvdw;
1899
1900 fscal = _mm256_and_ps(fscal,cutoff_mask);
1901
1902 /* Calculate temporary vectorial force */
1903 tx = _mm256_mul_ps(fscal,dx00);
1904 ty = _mm256_mul_ps(fscal,dy00);
1905 tz = _mm256_mul_ps(fscal,dz00);
1906
1907 /* Update vectorial force */
1908 fix0 = _mm256_add_ps(fix0,tx);
1909 fiy0 = _mm256_add_ps(fiy0,ty);
1910 fiz0 = _mm256_add_ps(fiz0,tz);
1911
1912 fjx0 = _mm256_add_ps(fjx0,tx);
1913 fjy0 = _mm256_add_ps(fjy0,ty);
1914 fjz0 = _mm256_add_ps(fjz0,tz);
1915
1916 }
1917
1918 /**************************
1919 * CALCULATE INTERACTIONS *
1920 **************************/
1921
1922 if (gmx_mm256_any_lt(rsq11,rcutoff2))
1923 {
1924
1925 r11 = _mm256_mul_ps(rsq11,rinv11);
1926
1927 /* EWALD ELECTROSTATICS */
1928
1929 /* Analytical PME correction */
1930 zeta2 = _mm256_mul_ps(beta2,rsq11);
1931 rinv3 = _mm256_mul_ps(rinvsq11,rinv11);
1932 pmecorrF = avx256_pmecorrF_f(zeta2);
1933 felec = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
1934 felec = _mm256_mul_ps(qq11,felec);
1935
1936 cutoff_mask = _mm256_cmp_ps(rsq11,rcutoff2,_CMP_LT_OQ);
1937
1938 fscal = felec;
1939
1940 fscal = _mm256_and_ps(fscal,cutoff_mask);
1941
1942 /* Calculate temporary vectorial force */
1943 tx = _mm256_mul_ps(fscal,dx11);
1944 ty = _mm256_mul_ps(fscal,dy11);
1945 tz = _mm256_mul_ps(fscal,dz11);
1946
1947 /* Update vectorial force */
1948 fix1 = _mm256_add_ps(fix1,tx);
1949 fiy1 = _mm256_add_ps(fiy1,ty);
1950 fiz1 = _mm256_add_ps(fiz1,tz);
1951
1952 fjx1 = _mm256_add_ps(fjx1,tx);
1953 fjy1 = _mm256_add_ps(fjy1,ty);
1954 fjz1 = _mm256_add_ps(fjz1,tz);
1955
1956 }
1957
1958 /**************************
1959 * CALCULATE INTERACTIONS *
1960 **************************/
1961
1962 if (gmx_mm256_any_lt(rsq12,rcutoff2))
1963 {
1964
1965 r12 = _mm256_mul_ps(rsq12,rinv12);
1966
1967 /* EWALD ELECTROSTATICS */
1968
1969 /* Analytical PME correction */
1970 zeta2 = _mm256_mul_ps(beta2,rsq12);
1971 rinv3 = _mm256_mul_ps(rinvsq12,rinv12);
1972 pmecorrF = avx256_pmecorrF_f(zeta2);
1973 felec = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
1974 felec = _mm256_mul_ps(qq12,felec);
1975
1976 cutoff_mask = _mm256_cmp_ps(rsq12,rcutoff2,_CMP_LT_OQ);
1977
1978 fscal = felec;
1979
1980 fscal = _mm256_and_ps(fscal,cutoff_mask);
1981
1982 /* Calculate temporary vectorial force */
1983 tx = _mm256_mul_ps(fscal,dx12);
1984 ty = _mm256_mul_ps(fscal,dy12);
1985 tz = _mm256_mul_ps(fscal,dz12);
1986
1987 /* Update vectorial force */
1988 fix1 = _mm256_add_ps(fix1,tx);
1989 fiy1 = _mm256_add_ps(fiy1,ty);
1990 fiz1 = _mm256_add_ps(fiz1,tz);
1991
1992 fjx2 = _mm256_add_ps(fjx2,tx);
1993 fjy2 = _mm256_add_ps(fjy2,ty);
1994 fjz2 = _mm256_add_ps(fjz2,tz);
1995
1996 }
1997
1998 /**************************
1999 * CALCULATE INTERACTIONS *
2000 **************************/
2001
2002 if (gmx_mm256_any_lt(rsq13,rcutoff2))
2003 {
2004
2005 r13 = _mm256_mul_ps(rsq13,rinv13);
2006
2007 /* EWALD ELECTROSTATICS */
2008
2009 /* Analytical PME correction */
2010 zeta2 = _mm256_mul_ps(beta2,rsq13);
2011 rinv3 = _mm256_mul_ps(rinvsq13,rinv13);
2012 pmecorrF = avx256_pmecorrF_f(zeta2);
2013 felec = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
2014 felec = _mm256_mul_ps(qq13,felec);
2015
2016 cutoff_mask = _mm256_cmp_ps(rsq13,rcutoff2,_CMP_LT_OQ);
2017
2018 fscal = felec;
2019
2020 fscal = _mm256_and_ps(fscal,cutoff_mask);
2021
2022 /* Calculate temporary vectorial force */
2023 tx = _mm256_mul_ps(fscal,dx13);
2024 ty = _mm256_mul_ps(fscal,dy13);
2025 tz = _mm256_mul_ps(fscal,dz13);
2026
2027 /* Update vectorial force */
2028 fix1 = _mm256_add_ps(fix1,tx);
2029 fiy1 = _mm256_add_ps(fiy1,ty);
2030 fiz1 = _mm256_add_ps(fiz1,tz);
2031
2032 fjx3 = _mm256_add_ps(fjx3,tx);
2033 fjy3 = _mm256_add_ps(fjy3,ty);
2034 fjz3 = _mm256_add_ps(fjz3,tz);
2035
2036 }
2037
2038 /**************************
2039 * CALCULATE INTERACTIONS *
2040 **************************/
2041
2042 if (gmx_mm256_any_lt(rsq21,rcutoff2))
2043 {
2044
2045 r21 = _mm256_mul_ps(rsq21,rinv21);
2046
2047 /* EWALD ELECTROSTATICS */
2048
2049 /* Analytical PME correction */
2050 zeta2 = _mm256_mul_ps(beta2,rsq21);
2051 rinv3 = _mm256_mul_ps(rinvsq21,rinv21);
2052 pmecorrF = avx256_pmecorrF_f(zeta2);
2053 felec = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
2054 felec = _mm256_mul_ps(qq21,felec);
2055
2056 cutoff_mask = _mm256_cmp_ps(rsq21,rcutoff2,_CMP_LT_OQ);
2057
2058 fscal = felec;
2059
2060 fscal = _mm256_and_ps(fscal,cutoff_mask);
2061
2062 /* Calculate temporary vectorial force */
2063 tx = _mm256_mul_ps(fscal,dx21);
2064 ty = _mm256_mul_ps(fscal,dy21);
2065 tz = _mm256_mul_ps(fscal,dz21);
2066
2067 /* Update vectorial force */
2068 fix2 = _mm256_add_ps(fix2,tx);
2069 fiy2 = _mm256_add_ps(fiy2,ty);
2070 fiz2 = _mm256_add_ps(fiz2,tz);
2071
2072 fjx1 = _mm256_add_ps(fjx1,tx);
2073 fjy1 = _mm256_add_ps(fjy1,ty);
2074 fjz1 = _mm256_add_ps(fjz1,tz);
2075
2076 }
2077
2078 /**************************
2079 * CALCULATE INTERACTIONS *
2080 **************************/
2081
2082 if (gmx_mm256_any_lt(rsq22,rcutoff2))
2083 {
2084
2085 r22 = _mm256_mul_ps(rsq22,rinv22);
2086
2087 /* EWALD ELECTROSTATICS */
2088
2089 /* Analytical PME correction */
2090 zeta2 = _mm256_mul_ps(beta2,rsq22);
2091 rinv3 = _mm256_mul_ps(rinvsq22,rinv22);
2092 pmecorrF = avx256_pmecorrF_f(zeta2);
2093 felec = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
2094 felec = _mm256_mul_ps(qq22,felec);
2095
2096 cutoff_mask = _mm256_cmp_ps(rsq22,rcutoff2,_CMP_LT_OQ);
2097
2098 fscal = felec;
2099
2100 fscal = _mm256_and_ps(fscal,cutoff_mask);
2101
2102 /* Calculate temporary vectorial force */
2103 tx = _mm256_mul_ps(fscal,dx22);
2104 ty = _mm256_mul_ps(fscal,dy22);
2105 tz = _mm256_mul_ps(fscal,dz22);
2106
2107 /* Update vectorial force */
2108 fix2 = _mm256_add_ps(fix2,tx);
2109 fiy2 = _mm256_add_ps(fiy2,ty);
2110 fiz2 = _mm256_add_ps(fiz2,tz);
2111
2112 fjx2 = _mm256_add_ps(fjx2,tx);
2113 fjy2 = _mm256_add_ps(fjy2,ty);
2114 fjz2 = _mm256_add_ps(fjz2,tz);
2115
2116 }
2117
2118 /**************************
2119 * CALCULATE INTERACTIONS *
2120 **************************/
2121
2122 if (gmx_mm256_any_lt(rsq23,rcutoff2))
2123 {
2124
2125 r23 = _mm256_mul_ps(rsq23,rinv23);
2126
2127 /* EWALD ELECTROSTATICS */
2128
2129 /* Analytical PME correction */
2130 zeta2 = _mm256_mul_ps(beta2,rsq23);
2131 rinv3 = _mm256_mul_ps(rinvsq23,rinv23);
2132 pmecorrF = avx256_pmecorrF_f(zeta2);
2133 felec = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
2134 felec = _mm256_mul_ps(qq23,felec);
2135
2136 cutoff_mask = _mm256_cmp_ps(rsq23,rcutoff2,_CMP_LT_OQ);
2137
2138 fscal = felec;
2139
2140 fscal = _mm256_and_ps(fscal,cutoff_mask);
2141
2142 /* Calculate temporary vectorial force */
2143 tx = _mm256_mul_ps(fscal,dx23);
2144 ty = _mm256_mul_ps(fscal,dy23);
2145 tz = _mm256_mul_ps(fscal,dz23);
2146
2147 /* Update vectorial force */
2148 fix2 = _mm256_add_ps(fix2,tx);
2149 fiy2 = _mm256_add_ps(fiy2,ty);
2150 fiz2 = _mm256_add_ps(fiz2,tz);
2151
2152 fjx3 = _mm256_add_ps(fjx3,tx);
2153 fjy3 = _mm256_add_ps(fjy3,ty);
2154 fjz3 = _mm256_add_ps(fjz3,tz);
2155
2156 }
2157
2158 /**************************
2159 * CALCULATE INTERACTIONS *
2160 **************************/
2161
2162 if (gmx_mm256_any_lt(rsq31,rcutoff2))
2163 {
2164
2165 r31 = _mm256_mul_ps(rsq31,rinv31);
2166
2167 /* EWALD ELECTROSTATICS */
2168
2169 /* Analytical PME correction */
2170 zeta2 = _mm256_mul_ps(beta2,rsq31);
2171 rinv3 = _mm256_mul_ps(rinvsq31,rinv31);
2172 pmecorrF = avx256_pmecorrF_f(zeta2);
2173 felec = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
2174 felec = _mm256_mul_ps(qq31,felec);
2175
2176 cutoff_mask = _mm256_cmp_ps(rsq31,rcutoff2,_CMP_LT_OQ);
2177
2178 fscal = felec;
2179
2180 fscal = _mm256_and_ps(fscal,cutoff_mask);
2181
2182 /* Calculate temporary vectorial force */
2183 tx = _mm256_mul_ps(fscal,dx31);
2184 ty = _mm256_mul_ps(fscal,dy31);
2185 tz = _mm256_mul_ps(fscal,dz31);
2186
2187 /* Update vectorial force */
2188 fix3 = _mm256_add_ps(fix3,tx);
2189 fiy3 = _mm256_add_ps(fiy3,ty);
2190 fiz3 = _mm256_add_ps(fiz3,tz);
2191
2192 fjx1 = _mm256_add_ps(fjx1,tx);
2193 fjy1 = _mm256_add_ps(fjy1,ty);
2194 fjz1 = _mm256_add_ps(fjz1,tz);
2195
2196 }
2197
2198 /**************************
2199 * CALCULATE INTERACTIONS *
2200 **************************/
2201
2202 if (gmx_mm256_any_lt(rsq32,rcutoff2))
2203 {
2204
2205 r32 = _mm256_mul_ps(rsq32,rinv32);
2206
2207 /* EWALD ELECTROSTATICS */
2208
2209 /* Analytical PME correction */
2210 zeta2 = _mm256_mul_ps(beta2,rsq32);
2211 rinv3 = _mm256_mul_ps(rinvsq32,rinv32);
2212 pmecorrF = avx256_pmecorrF_f(zeta2);
2213 felec = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
2214 felec = _mm256_mul_ps(qq32,felec);
2215
2216 cutoff_mask = _mm256_cmp_ps(rsq32,rcutoff2,_CMP_LT_OQ);
2217
2218 fscal = felec;
2219
2220 fscal = _mm256_and_ps(fscal,cutoff_mask);
2221
2222 /* Calculate temporary vectorial force */
2223 tx = _mm256_mul_ps(fscal,dx32);
2224 ty = _mm256_mul_ps(fscal,dy32);
2225 tz = _mm256_mul_ps(fscal,dz32);
2226
2227 /* Update vectorial force */
2228 fix3 = _mm256_add_ps(fix3,tx);
2229 fiy3 = _mm256_add_ps(fiy3,ty);
2230 fiz3 = _mm256_add_ps(fiz3,tz);
2231
2232 fjx2 = _mm256_add_ps(fjx2,tx);
2233 fjy2 = _mm256_add_ps(fjy2,ty);
2234 fjz2 = _mm256_add_ps(fjz2,tz);
2235
2236 }
2237
2238 /**************************
2239 * CALCULATE INTERACTIONS *
2240 **************************/
2241
2242 if (gmx_mm256_any_lt(rsq33,rcutoff2))
2243 {
2244
2245 r33 = _mm256_mul_ps(rsq33,rinv33);
2246
2247 /* EWALD ELECTROSTATICS */
2248
2249 /* Analytical PME correction */
2250 zeta2 = _mm256_mul_ps(beta2,rsq33);
2251 rinv3 = _mm256_mul_ps(rinvsq33,rinv33);
2252 pmecorrF = avx256_pmecorrF_f(zeta2);
2253 felec = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
2254 felec = _mm256_mul_ps(qq33,felec);
2255
2256 cutoff_mask = _mm256_cmp_ps(rsq33,rcutoff2,_CMP_LT_OQ);
2257
2258 fscal = felec;
2259
2260 fscal = _mm256_and_ps(fscal,cutoff_mask);
2261
2262 /* Calculate temporary vectorial force */
2263 tx = _mm256_mul_ps(fscal,dx33);
2264 ty = _mm256_mul_ps(fscal,dy33);
2265 tz = _mm256_mul_ps(fscal,dz33);
2266
2267 /* Update vectorial force */
2268 fix3 = _mm256_add_ps(fix3,tx);
2269 fiy3 = _mm256_add_ps(fiy3,ty);
2270 fiz3 = _mm256_add_ps(fiz3,tz);
2271
2272 fjx3 = _mm256_add_ps(fjx3,tx);
2273 fjy3 = _mm256_add_ps(fjy3,ty);
2274 fjz3 = _mm256_add_ps(fjz3,tz);
2275
2276 }
2277
2278 fjptrA = f+j_coord_offsetA;
2279 fjptrB = f+j_coord_offsetB;
2280 fjptrC = f+j_coord_offsetC;
2281 fjptrD = f+j_coord_offsetD;
2282 fjptrE = f+j_coord_offsetE;
2283 fjptrF = f+j_coord_offsetF;
2284 fjptrG = f+j_coord_offsetG;
2285 fjptrH = f+j_coord_offsetH;
2286
2287 gmx_mm256_decrement_4rvec_8ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjptrE,fjptrF,fjptrG,fjptrH,
2288 fjx0,fjy0,fjz0,fjx1,fjy1,fjz1,
2289 fjx2,fjy2,fjz2,fjx3,fjy3,fjz3);
2290
2291 /* Inner loop uses 583 flops */
2292 }
2293
2294 if(jidx<j_index_end)
2295 {
2296
2297 /* Get j neighbor index, and coordinate index */
2298 jnrlistA = jjnr[jidx];
2299 jnrlistB = jjnr[jidx+1];
2300 jnrlistC = jjnr[jidx+2];
2301 jnrlistD = jjnr[jidx+3];
2302 jnrlistE = jjnr[jidx+4];
2303 jnrlistF = jjnr[jidx+5];
2304 jnrlistG = jjnr[jidx+6];
2305 jnrlistH = jjnr[jidx+7];
2306 /* Sign of each element will be negative for non-real atoms.
2307 * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
2308 * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
2309 */
2310 dummy_mask = gmx_mm256_set_m128(gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx+4)),_mm_setzero_si128())),
2311 gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128())));
2312
2313 jnrA = (jnrlistA>=0) ? jnrlistA : 0;
2314 jnrB = (jnrlistB>=0) ? jnrlistB : 0;
2315 jnrC = (jnrlistC>=0) ? jnrlistC : 0;
2316 jnrD = (jnrlistD>=0) ? jnrlistD : 0;
2317 jnrE = (jnrlistE>=0) ? jnrlistE : 0;
2318 jnrF = (jnrlistF>=0) ? jnrlistF : 0;
2319 jnrG = (jnrlistG>=0) ? jnrlistG : 0;
2320 jnrH = (jnrlistH>=0) ? jnrlistH : 0;
2321 j_coord_offsetA = DIM*jnrA;
2322 j_coord_offsetB = DIM*jnrB;
2323 j_coord_offsetC = DIM*jnrC;
2324 j_coord_offsetD = DIM*jnrD;
2325 j_coord_offsetE = DIM*jnrE;
2326 j_coord_offsetF = DIM*jnrF;
2327 j_coord_offsetG = DIM*jnrG;
2328 j_coord_offsetH = DIM*jnrH;
2329
2330 /* load j atom coordinates */
2331 gmx_mm256_load_4rvec_8ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
2332 x+j_coord_offsetC,x+j_coord_offsetD,
2333 x+j_coord_offsetE,x+j_coord_offsetF,
2334 x+j_coord_offsetG,x+j_coord_offsetH,
2335 &jx0,&jy0,&jz0,&jx1,&jy1,&jz1,&jx2,
2336 &jy2,&jz2,&jx3,&jy3,&jz3);
2337
2338 /* Calculate displacement vector */
2339 dx00 = _mm256_sub_ps(ix0,jx0);
2340 dy00 = _mm256_sub_ps(iy0,jy0);
2341 dz00 = _mm256_sub_ps(iz0,jz0);
2342 dx11 = _mm256_sub_ps(ix1,jx1);
2343 dy11 = _mm256_sub_ps(iy1,jy1);
2344 dz11 = _mm256_sub_ps(iz1,jz1);
2345 dx12 = _mm256_sub_ps(ix1,jx2);
2346 dy12 = _mm256_sub_ps(iy1,jy2);
2347 dz12 = _mm256_sub_ps(iz1,jz2);
2348 dx13 = _mm256_sub_ps(ix1,jx3);
2349 dy13 = _mm256_sub_ps(iy1,jy3);
2350 dz13 = _mm256_sub_ps(iz1,jz3);
2351 dx21 = _mm256_sub_ps(ix2,jx1);
2352 dy21 = _mm256_sub_ps(iy2,jy1);
2353 dz21 = _mm256_sub_ps(iz2,jz1);
2354 dx22 = _mm256_sub_ps(ix2,jx2);
2355 dy22 = _mm256_sub_ps(iy2,jy2);
2356 dz22 = _mm256_sub_ps(iz2,jz2);
2357 dx23 = _mm256_sub_ps(ix2,jx3);
2358 dy23 = _mm256_sub_ps(iy2,jy3);
2359 dz23 = _mm256_sub_ps(iz2,jz3);
2360 dx31 = _mm256_sub_ps(ix3,jx1);
2361 dy31 = _mm256_sub_ps(iy3,jy1);
2362 dz31 = _mm256_sub_ps(iz3,jz1);
2363 dx32 = _mm256_sub_ps(ix3,jx2);
2364 dy32 = _mm256_sub_ps(iy3,jy2);
2365 dz32 = _mm256_sub_ps(iz3,jz2);
2366 dx33 = _mm256_sub_ps(ix3,jx3);
2367 dy33 = _mm256_sub_ps(iy3,jy3);
2368 dz33 = _mm256_sub_ps(iz3,jz3);
2369
2370 /* Calculate squared distance and things based on it */
2371 rsq00 = gmx_mm256_calc_rsq_ps(dx00,dy00,dz00);
2372 rsq11 = gmx_mm256_calc_rsq_ps(dx11,dy11,dz11);
2373 rsq12 = gmx_mm256_calc_rsq_ps(dx12,dy12,dz12);
2374 rsq13 = gmx_mm256_calc_rsq_ps(dx13,dy13,dz13);
2375 rsq21 = gmx_mm256_calc_rsq_ps(dx21,dy21,dz21);
2376 rsq22 = gmx_mm256_calc_rsq_ps(dx22,dy22,dz22);
2377 rsq23 = gmx_mm256_calc_rsq_ps(dx23,dy23,dz23);
2378 rsq31 = gmx_mm256_calc_rsq_ps(dx31,dy31,dz31);
2379 rsq32 = gmx_mm256_calc_rsq_ps(dx32,dy32,dz32);
2380 rsq33 = gmx_mm256_calc_rsq_ps(dx33,dy33,dz33);
2381
2382 rinv00 = avx256_invsqrt_f(rsq00);
2383 rinv11 = avx256_invsqrt_f(rsq11);
2384 rinv12 = avx256_invsqrt_f(rsq12);
2385 rinv13 = avx256_invsqrt_f(rsq13);
2386 rinv21 = avx256_invsqrt_f(rsq21);
2387 rinv22 = avx256_invsqrt_f(rsq22);
2388 rinv23 = avx256_invsqrt_f(rsq23);
2389 rinv31 = avx256_invsqrt_f(rsq31);
2390 rinv32 = avx256_invsqrt_f(rsq32);
2391 rinv33 = avx256_invsqrt_f(rsq33);
2392
2393 rinvsq00 = _mm256_mul_ps(rinv00,rinv00);
2394 rinvsq11 = _mm256_mul_ps(rinv11,rinv11);
2395 rinvsq12 = _mm256_mul_ps(rinv12,rinv12);
2396 rinvsq13 = _mm256_mul_ps(rinv13,rinv13);
2397 rinvsq21 = _mm256_mul_ps(rinv21,rinv21);
2398 rinvsq22 = _mm256_mul_ps(rinv22,rinv22);
2399 rinvsq23 = _mm256_mul_ps(rinv23,rinv23);
2400 rinvsq31 = _mm256_mul_ps(rinv31,rinv31);
2401 rinvsq32 = _mm256_mul_ps(rinv32,rinv32);
2402 rinvsq33 = _mm256_mul_ps(rinv33,rinv33);
2403
2404 fjx0 = _mm256_setzero_ps();
2405 fjy0 = _mm256_setzero_ps();
2406 fjz0 = _mm256_setzero_ps();
2407 fjx1 = _mm256_setzero_ps();
2408 fjy1 = _mm256_setzero_ps();
2409 fjz1 = _mm256_setzero_ps();
2410 fjx2 = _mm256_setzero_ps();
2411 fjy2 = _mm256_setzero_ps();
2412 fjz2 = _mm256_setzero_ps();
2413 fjx3 = _mm256_setzero_ps();
2414 fjy3 = _mm256_setzero_ps();
2415 fjz3 = _mm256_setzero_ps();
2416
2417 /**************************
2418 * CALCULATE INTERACTIONS *
2419 **************************/
2420
2421 if (gmx_mm256_any_lt(rsq00,rcutoff2))
2422 {
2423
2424 r00 = _mm256_mul_ps(rsq00,rinv00);
2425 r00 = _mm256_andnot_ps(dummy_mask,r00);
2426
2427 /* Analytical LJ-PME */
2428 rinvsix = _mm256_mul_ps(_mm256_mul_ps(rinvsq00,rinvsq00),rinvsq00);
2429 ewcljrsq = _mm256_mul_ps(ewclj2,rsq00);
2430 ewclj6 = _mm256_mul_ps(ewclj2,_mm256_mul_ps(ewclj2,ewclj2));
2431 exponent = avx256_exp_f(ewcljrsq);
2432 /* poly = exp(-(beta*r)^2) * (1 + (beta*r)^2 + (beta*r)^4 /2) */
2433 poly = _mm256_mul_ps(exponent,_mm256_add_ps(_mm256_sub_ps(one,ewcljrsq),_mm256_mul_ps(_mm256_mul_ps(ewcljrsq,ewcljrsq),one_half)));
2434 /* f6A = 6 * C6grid * (1 - poly) */
2435 f6A = _mm256_mul_ps(c6grid_00,_mm256_sub_ps(one,poly));
2436 /* f6B = C6grid * exponent * beta^6 */
2437 f6B = _mm256_mul_ps(_mm256_mul_ps(c6grid_00,one_sixth),_mm256_mul_ps(exponent,ewclj6));
2438 /* fvdw = 12*C12/r13 - ((6*C6 - f6A)/r6 + f6B)/r */
2439 fvdw = _mm256_mul_ps(_mm256_add_ps(_mm256_mul_ps(_mm256_sub_ps(_mm256_mul_ps(c12_00,rinvsix),_mm256_sub_ps(c6_00,f6A)),rinvsix),f6B),rinvsq00);
2440
2441 cutoff_mask = _mm256_cmp_ps(rsq00,rcutoff2,_CMP_LT_OQ);
2442
2443 fscal = fvdw;
2444
2445 fscal = _mm256_and_ps(fscal,cutoff_mask);
2446
2447 fscal = _mm256_andnot_ps(dummy_mask,fscal);
2448
2449 /* Calculate temporary vectorial force */
2450 tx = _mm256_mul_ps(fscal,dx00);
2451 ty = _mm256_mul_ps(fscal,dy00);
2452 tz = _mm256_mul_ps(fscal,dz00);
2453
2454 /* Update vectorial force */
2455 fix0 = _mm256_add_ps(fix0,tx);
2456 fiy0 = _mm256_add_ps(fiy0,ty);
2457 fiz0 = _mm256_add_ps(fiz0,tz);
2458
2459 fjx0 = _mm256_add_ps(fjx0,tx);
2460 fjy0 = _mm256_add_ps(fjy0,ty);
2461 fjz0 = _mm256_add_ps(fjz0,tz);
2462
2463 }
2464
2465 /**************************
2466 * CALCULATE INTERACTIONS *
2467 **************************/
2468
2469 if (gmx_mm256_any_lt(rsq11,rcutoff2))
2470 {
2471
2472 r11 = _mm256_mul_ps(rsq11,rinv11);
2473 r11 = _mm256_andnot_ps(dummy_mask,r11);
2474
2475 /* EWALD ELECTROSTATICS */
2476
2477 /* Analytical PME correction */
2478 zeta2 = _mm256_mul_ps(beta2,rsq11);
2479 rinv3 = _mm256_mul_ps(rinvsq11,rinv11);
2480 pmecorrF = avx256_pmecorrF_f(zeta2);
2481 felec = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
2482 felec = _mm256_mul_ps(qq11,felec);
2483
2484 cutoff_mask = _mm256_cmp_ps(rsq11,rcutoff2,_CMP_LT_OQ);
2485
2486 fscal = felec;
2487
2488 fscal = _mm256_and_ps(fscal,cutoff_mask);
2489
2490 fscal = _mm256_andnot_ps(dummy_mask,fscal);
2491
2492 /* Calculate temporary vectorial force */
2493 tx = _mm256_mul_ps(fscal,dx11);
2494 ty = _mm256_mul_ps(fscal,dy11);
2495 tz = _mm256_mul_ps(fscal,dz11);
2496
2497 /* Update vectorial force */
2498 fix1 = _mm256_add_ps(fix1,tx);
2499 fiy1 = _mm256_add_ps(fiy1,ty);
2500 fiz1 = _mm256_add_ps(fiz1,tz);
2501
2502 fjx1 = _mm256_add_ps(fjx1,tx);
2503 fjy1 = _mm256_add_ps(fjy1,ty);
2504 fjz1 = _mm256_add_ps(fjz1,tz);
2505
2506 }
2507
2508 /**************************
2509 * CALCULATE INTERACTIONS *
2510 **************************/
2511
2512 if (gmx_mm256_any_lt(rsq12,rcutoff2))
2513 {
2514
2515 r12 = _mm256_mul_ps(rsq12,rinv12);
2516 r12 = _mm256_andnot_ps(dummy_mask,r12);
2517
2518 /* EWALD ELECTROSTATICS */
2519
2520 /* Analytical PME correction */
2521 zeta2 = _mm256_mul_ps(beta2,rsq12);
2522 rinv3 = _mm256_mul_ps(rinvsq12,rinv12);
2523 pmecorrF = avx256_pmecorrF_f(zeta2);
2524 felec = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
2525 felec = _mm256_mul_ps(qq12,felec);
2526
2527 cutoff_mask = _mm256_cmp_ps(rsq12,rcutoff2,_CMP_LT_OQ);
2528
2529 fscal = felec;
2530
2531 fscal = _mm256_and_ps(fscal,cutoff_mask);
2532
2533 fscal = _mm256_andnot_ps(dummy_mask,fscal);
2534
2535 /* Calculate temporary vectorial force */
2536 tx = _mm256_mul_ps(fscal,dx12);
2537 ty = _mm256_mul_ps(fscal,dy12);
2538 tz = _mm256_mul_ps(fscal,dz12);
2539
2540 /* Update vectorial force */
2541 fix1 = _mm256_add_ps(fix1,tx);
2542 fiy1 = _mm256_add_ps(fiy1,ty);
2543 fiz1 = _mm256_add_ps(fiz1,tz);
2544
2545 fjx2 = _mm256_add_ps(fjx2,tx);
2546 fjy2 = _mm256_add_ps(fjy2,ty);
2547 fjz2 = _mm256_add_ps(fjz2,tz);
2548
2549 }
2550
2551 /**************************
2552 * CALCULATE INTERACTIONS *
2553 **************************/
2554
2555 if (gmx_mm256_any_lt(rsq13,rcutoff2))
2556 {
2557
2558 r13 = _mm256_mul_ps(rsq13,rinv13);
2559 r13 = _mm256_andnot_ps(dummy_mask,r13);
2560
2561 /* EWALD ELECTROSTATICS */
2562
2563 /* Analytical PME correction */
2564 zeta2 = _mm256_mul_ps(beta2,rsq13);
2565 rinv3 = _mm256_mul_ps(rinvsq13,rinv13);
2566 pmecorrF = avx256_pmecorrF_f(zeta2);
2567 felec = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
2568 felec = _mm256_mul_ps(qq13,felec);
2569
2570 cutoff_mask = _mm256_cmp_ps(rsq13,rcutoff2,_CMP_LT_OQ);
2571
2572 fscal = felec;
2573
2574 fscal = _mm256_and_ps(fscal,cutoff_mask);
2575
2576 fscal = _mm256_andnot_ps(dummy_mask,fscal);
2577
2578 /* Calculate temporary vectorial force */
2579 tx = _mm256_mul_ps(fscal,dx13);
2580 ty = _mm256_mul_ps(fscal,dy13);
2581 tz = _mm256_mul_ps(fscal,dz13);
2582
2583 /* Update vectorial force */
2584 fix1 = _mm256_add_ps(fix1,tx);
2585 fiy1 = _mm256_add_ps(fiy1,ty);
2586 fiz1 = _mm256_add_ps(fiz1,tz);
2587
2588 fjx3 = _mm256_add_ps(fjx3,tx);
2589 fjy3 = _mm256_add_ps(fjy3,ty);
2590 fjz3 = _mm256_add_ps(fjz3,tz);
2591
2592 }
2593
2594 /**************************
2595 * CALCULATE INTERACTIONS *
2596 **************************/
2597
2598 if (gmx_mm256_any_lt(rsq21,rcutoff2))
2599 {
2600
2601 r21 = _mm256_mul_ps(rsq21,rinv21);
2602 r21 = _mm256_andnot_ps(dummy_mask,r21);
2603
2604 /* EWALD ELECTROSTATICS */
2605
2606 /* Analytical PME correction */
2607 zeta2 = _mm256_mul_ps(beta2,rsq21);
2608 rinv3 = _mm256_mul_ps(rinvsq21,rinv21);
2609 pmecorrF = avx256_pmecorrF_f(zeta2);
2610 felec = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
2611 felec = _mm256_mul_ps(qq21,felec);
2612
2613 cutoff_mask = _mm256_cmp_ps(rsq21,rcutoff2,_CMP_LT_OQ);
2614
2615 fscal = felec;
2616
2617 fscal = _mm256_and_ps(fscal,cutoff_mask);
2618
2619 fscal = _mm256_andnot_ps(dummy_mask,fscal);
2620
2621 /* Calculate temporary vectorial force */
2622 tx = _mm256_mul_ps(fscal,dx21);
2623 ty = _mm256_mul_ps(fscal,dy21);
2624 tz = _mm256_mul_ps(fscal,dz21);
2625
2626 /* Update vectorial force */
2627 fix2 = _mm256_add_ps(fix2,tx);
2628 fiy2 = _mm256_add_ps(fiy2,ty);
2629 fiz2 = _mm256_add_ps(fiz2,tz);
2630
2631 fjx1 = _mm256_add_ps(fjx1,tx);
2632 fjy1 = _mm256_add_ps(fjy1,ty);
2633 fjz1 = _mm256_add_ps(fjz1,tz);
2634
2635 }
2636
2637 /**************************
2638 * CALCULATE INTERACTIONS *
2639 **************************/
2640
2641 if (gmx_mm256_any_lt(rsq22,rcutoff2))
2642 {
2643
2644 r22 = _mm256_mul_ps(rsq22,rinv22);
2645 r22 = _mm256_andnot_ps(dummy_mask,r22);
2646
2647 /* EWALD ELECTROSTATICS */
2648
2649 /* Analytical PME correction */
2650 zeta2 = _mm256_mul_ps(beta2,rsq22);
2651 rinv3 = _mm256_mul_ps(rinvsq22,rinv22);
2652 pmecorrF = avx256_pmecorrF_f(zeta2);
2653 felec = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
2654 felec = _mm256_mul_ps(qq22,felec);
2655
2656 cutoff_mask = _mm256_cmp_ps(rsq22,rcutoff2,_CMP_LT_OQ);
2657
2658 fscal = felec;
2659
2660 fscal = _mm256_and_ps(fscal,cutoff_mask);
2661
2662 fscal = _mm256_andnot_ps(dummy_mask,fscal);
2663
2664 /* Calculate temporary vectorial force */
2665 tx = _mm256_mul_ps(fscal,dx22);
2666 ty = _mm256_mul_ps(fscal,dy22);
2667 tz = _mm256_mul_ps(fscal,dz22);
2668
2669 /* Update vectorial force */
2670 fix2 = _mm256_add_ps(fix2,tx);
2671 fiy2 = _mm256_add_ps(fiy2,ty);
2672 fiz2 = _mm256_add_ps(fiz2,tz);
2673
2674 fjx2 = _mm256_add_ps(fjx2,tx);
2675 fjy2 = _mm256_add_ps(fjy2,ty);
2676 fjz2 = _mm256_add_ps(fjz2,tz);
2677
2678 }
2679
2680 /**************************
2681 * CALCULATE INTERACTIONS *
2682 **************************/
2683
2684 if (gmx_mm256_any_lt(rsq23,rcutoff2))
2685 {
2686
2687 r23 = _mm256_mul_ps(rsq23,rinv23);
2688 r23 = _mm256_andnot_ps(dummy_mask,r23);
2689
2690 /* EWALD ELECTROSTATICS */
2691
2692 /* Analytical PME correction */
2693 zeta2 = _mm256_mul_ps(beta2,rsq23);
2694 rinv3 = _mm256_mul_ps(rinvsq23,rinv23);
2695 pmecorrF = avx256_pmecorrF_f(zeta2);
2696 felec = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
2697 felec = _mm256_mul_ps(qq23,felec);
2698
2699 cutoff_mask = _mm256_cmp_ps(rsq23,rcutoff2,_CMP_LT_OQ);
2700
2701 fscal = felec;
2702
2703 fscal = _mm256_and_ps(fscal,cutoff_mask);
2704
2705 fscal = _mm256_andnot_ps(dummy_mask,fscal);
2706
2707 /* Calculate temporary vectorial force */
2708 tx = _mm256_mul_ps(fscal,dx23);
2709 ty = _mm256_mul_ps(fscal,dy23);
2710 tz = _mm256_mul_ps(fscal,dz23);
2711
2712 /* Update vectorial force */
2713 fix2 = _mm256_add_ps(fix2,tx);
2714 fiy2 = _mm256_add_ps(fiy2,ty);
2715 fiz2 = _mm256_add_ps(fiz2,tz);
2716
2717 fjx3 = _mm256_add_ps(fjx3,tx);
2718 fjy3 = _mm256_add_ps(fjy3,ty);
2719 fjz3 = _mm256_add_ps(fjz3,tz);
2720
2721 }
2722
2723 /**************************
2724 * CALCULATE INTERACTIONS *
2725 **************************/
2726
2727 if (gmx_mm256_any_lt(rsq31,rcutoff2))
2728 {
2729
2730 r31 = _mm256_mul_ps(rsq31,rinv31);
2731 r31 = _mm256_andnot_ps(dummy_mask,r31);
2732
2733 /* EWALD ELECTROSTATICS */
2734
2735 /* Analytical PME correction */
2736 zeta2 = _mm256_mul_ps(beta2,rsq31);
2737 rinv3 = _mm256_mul_ps(rinvsq31,rinv31);
2738 pmecorrF = avx256_pmecorrF_f(zeta2);
2739 felec = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
2740 felec = _mm256_mul_ps(qq31,felec);
2741
2742 cutoff_mask = _mm256_cmp_ps(rsq31,rcutoff2,_CMP_LT_OQ);
2743
2744 fscal = felec;
2745
2746 fscal = _mm256_and_ps(fscal,cutoff_mask);
2747
2748 fscal = _mm256_andnot_ps(dummy_mask,fscal);
2749
2750 /* Calculate temporary vectorial force */
2751 tx = _mm256_mul_ps(fscal,dx31);
2752 ty = _mm256_mul_ps(fscal,dy31);
2753 tz = _mm256_mul_ps(fscal,dz31);
2754
2755 /* Update vectorial force */
2756 fix3 = _mm256_add_ps(fix3,tx);
2757 fiy3 = _mm256_add_ps(fiy3,ty);
2758 fiz3 = _mm256_add_ps(fiz3,tz);
2759
2760 fjx1 = _mm256_add_ps(fjx1,tx);
2761 fjy1 = _mm256_add_ps(fjy1,ty);
2762 fjz1 = _mm256_add_ps(fjz1,tz);
2763
2764 }
2765
2766 /**************************
2767 * CALCULATE INTERACTIONS *
2768 **************************/
2769
2770 if (gmx_mm256_any_lt(rsq32,rcutoff2))
2771 {
2772
2773 r32 = _mm256_mul_ps(rsq32,rinv32);
2774 r32 = _mm256_andnot_ps(dummy_mask,r32);
2775
2776 /* EWALD ELECTROSTATICS */
2777
2778 /* Analytical PME correction */
2779 zeta2 = _mm256_mul_ps(beta2,rsq32);
2780 rinv3 = _mm256_mul_ps(rinvsq32,rinv32);
2781 pmecorrF = avx256_pmecorrF_f(zeta2);
2782 felec = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
2783 felec = _mm256_mul_ps(qq32,felec);
2784
2785 cutoff_mask = _mm256_cmp_ps(rsq32,rcutoff2,_CMP_LT_OQ);
2786
2787 fscal = felec;
2788
2789 fscal = _mm256_and_ps(fscal,cutoff_mask);
2790
2791 fscal = _mm256_andnot_ps(dummy_mask,fscal);
2792
2793 /* Calculate temporary vectorial force */
2794 tx = _mm256_mul_ps(fscal,dx32);
2795 ty = _mm256_mul_ps(fscal,dy32);
2796 tz = _mm256_mul_ps(fscal,dz32);
2797
2798 /* Update vectorial force */
2799 fix3 = _mm256_add_ps(fix3,tx);
2800 fiy3 = _mm256_add_ps(fiy3,ty);
2801 fiz3 = _mm256_add_ps(fiz3,tz);
2802
2803 fjx2 = _mm256_add_ps(fjx2,tx);
2804 fjy2 = _mm256_add_ps(fjy2,ty);
2805 fjz2 = _mm256_add_ps(fjz2,tz);
2806
2807 }
2808
2809 /**************************
2810 * CALCULATE INTERACTIONS *
2811 **************************/
2812
2813 if (gmx_mm256_any_lt(rsq33,rcutoff2))
2814 {
2815
2816 r33 = _mm256_mul_ps(rsq33,rinv33);
2817 r33 = _mm256_andnot_ps(dummy_mask,r33);
2818
2819 /* EWALD ELECTROSTATICS */
2820
2821 /* Analytical PME correction */
2822 zeta2 = _mm256_mul_ps(beta2,rsq33);
2823 rinv3 = _mm256_mul_ps(rinvsq33,rinv33);
2824 pmecorrF = avx256_pmecorrF_f(zeta2);
2825 felec = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
2826 felec = _mm256_mul_ps(qq33,felec);
2827
2828 cutoff_mask = _mm256_cmp_ps(rsq33,rcutoff2,_CMP_LT_OQ);
2829
2830 fscal = felec;
2831
2832 fscal = _mm256_and_ps(fscal,cutoff_mask);
2833
2834 fscal = _mm256_andnot_ps(dummy_mask,fscal);
2835
2836 /* Calculate temporary vectorial force */
2837 tx = _mm256_mul_ps(fscal,dx33);
2838 ty = _mm256_mul_ps(fscal,dy33);
2839 tz = _mm256_mul_ps(fscal,dz33);
2840
2841 /* Update vectorial force */
2842 fix3 = _mm256_add_ps(fix3,tx);
2843 fiy3 = _mm256_add_ps(fiy3,ty);
2844 fiz3 = _mm256_add_ps(fiz3,tz);
2845
2846 fjx3 = _mm256_add_ps(fjx3,tx);
2847 fjy3 = _mm256_add_ps(fjy3,ty);
2848 fjz3 = _mm256_add_ps(fjz3,tz);
2849
2850 }
2851
2852 fjptrA = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
2853 fjptrB = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
2854 fjptrC = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
2855 fjptrD = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
2856 fjptrE = (jnrlistE>=0) ? f+j_coord_offsetE : scratch;
2857 fjptrF = (jnrlistF>=0) ? f+j_coord_offsetF : scratch;
2858 fjptrG = (jnrlistG>=0) ? f+j_coord_offsetG : scratch;
2859 fjptrH = (jnrlistH>=0) ? f+j_coord_offsetH : scratch;
2860
2861 gmx_mm256_decrement_4rvec_8ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjptrE,fjptrF,fjptrG,fjptrH,
2862 fjx0,fjy0,fjz0,fjx1,fjy1,fjz1,
2863 fjx2,fjy2,fjz2,fjx3,fjy3,fjz3);
2864
2865 /* Inner loop uses 593 flops */
2866 }
2867
2868 /* End of innermost loop */
2869
2870 gmx_mm256_update_iforce_4atom_swizzle_ps(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,fix3,fiy3,fiz3,
2871 f+i_coord_offset,fshift+i_shift_offset);
2872
2873 /* Increment number of inner iterations */
2874 inneriter += j_index_end - j_index_start;
2875
2876 /* Outer loop uses 24 flops */
2877 }
2878
2879 /* Increment number of outer iterations */
2880 outeriter += nri;
2881
2882 /* Update outer/inner flops */
2883
2884 inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W4W4_F,outeriter*24 + inneriter*593);
2885 }
The ultimate molecular dynamics simulation package