?  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_ElecEw_VdwLJ_GeomW3P1_avx_256_single.cpp
blob599d8964abcefd2f47b3a84fd62a1c44fb93bdcc
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_ElecEw_VdwLJ_GeomW3P1_VF_avx_256_single
51 * Electrostatics interaction: Ewald
52 * VdW interaction: LennardJones
53 * Geometry: Water3-Particle
54 * Calculate force/pot: PotentialAndForce
55 */
56 void
57 nb_kernel_ElecEw_VdwLJ_GeomW3P1_VF_avx_256_single
58 (t_nblist * gmx_restrict nlist,
59 rvec * gmx_restrict xx,
60 rvec * gmx_restrict ff,
61 struct t_forcerec * gmx_restrict fr,
62 t_mdatoms * gmx_restrict mdatoms,
63 nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
64 t_nrnb * gmx_restrict nrnb)
65 {
66 /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
67 * just 0 for non-waters.
68 * Suffixes A,B,C,D,E,F,G,H refer to j loop unrolling done with AVX, e.g. for the eight different
69 * jnr indices corresponding to data put in the four positions in the SIMD register.
70 */
71 int i_shift_offset,i_coord_offset,outeriter,inneriter;
72 int j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
73 int jnrA,jnrB,jnrC,jnrD;
74 int jnrE,jnrF,jnrG,jnrH;
75 int jnrlistA,jnrlistB,jnrlistC,jnrlistD;
76 int jnrlistE,jnrlistF,jnrlistG,jnrlistH;
77 int j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
78 int j_coord_offsetE,j_coord_offsetF,j_coord_offsetG,j_coord_offsetH;
79 int *iinr,*jindex,*jjnr,*shiftidx,*gid;
80 real rcutoff_scalar;
81 real *shiftvec,*fshift,*x,*f;
82 real *fjptrA,*fjptrB,*fjptrC,*fjptrD,*fjptrE,*fjptrF,*fjptrG,*fjptrH;
83 real scratch[4*DIM];
84 __m256 tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
85 real * vdwioffsetptr0;
86 __m256 ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
87 real * vdwioffsetptr1;
88 __m256 ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
89 real * vdwioffsetptr2;
90 __m256 ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
91 int vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D,vdwjidx0E,vdwjidx0F,vdwjidx0G,vdwjidx0H;
92 __m256 jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
93 __m256 dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
94 __m256 dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
95 __m256 dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
96 __m256 velec,felec,velecsum,facel,crf,krf,krf2;
97 real *charge;
98 int nvdwtype;
99 __m256 rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
100 int *vdwtype;
101 real *vdwparam;
102 __m256 one_sixth = _mm256_set1_ps(1.0/6.0);
103 __m256 one_twelfth = _mm256_set1_ps(1.0/12.0);
104 __m256i ewitab;
105 __m128i ewitab_lo,ewitab_hi;
106 __m256 ewtabscale,eweps,sh_ewald,ewrt,ewtabhalfspace,ewtabF,ewtabFn,ewtabD,ewtabV;
107 __m256 beta,beta2,beta3,zeta2,pmecorrF,pmecorrV,rinv3;
108 real *ewtab;
109 __m256 dummy_mask,cutoff_mask;
110 __m256 signbit = _mm256_castsi256_ps( _mm256_set1_epi32(0x80000000) );
111 __m256 one = _mm256_set1_ps(1.0);
112 __m256 two = _mm256_set1_ps(2.0);
113 x = xx[0];
114 f = ff[0];
115
116 nri = nlist->nri;
117 iinr = nlist->iinr;
118 jindex = nlist->jindex;
119 jjnr = nlist->jjnr;
120 shiftidx = nlist->shift;
121 gid = nlist->gid;
122 shiftvec = fr->shift_vec[0];
123 fshift = fr->fshift[0];
124 facel = _mm256_set1_ps(fr->ic->epsfac);
125 charge = mdatoms->chargeA;
126 nvdwtype = fr->ntype;
127 vdwparam = fr->nbfp;
128 vdwtype = mdatoms->typeA;
129
130 sh_ewald = _mm256_set1_ps(fr->ic->sh_ewald);
131 beta = _mm256_set1_ps(fr->ic->ewaldcoeff_q);
132 beta2 = _mm256_mul_ps(beta,beta);
133 beta3 = _mm256_mul_ps(beta,beta2);
134
135 ewtab = fr->ic->tabq_coul_FDV0;
136 ewtabscale = _mm256_set1_ps(fr->ic->tabq_scale);
137 ewtabhalfspace = _mm256_set1_ps(0.5/fr->ic->tabq_scale);
138
139 /* Setup water-specific parameters */
140 inr = nlist->iinr[0];
141 iq0 = _mm256_mul_ps(facel,_mm256_set1_ps(charge[inr+0]));
142 iq1 = _mm256_mul_ps(facel,_mm256_set1_ps(charge[inr+1]));
143 iq2 = _mm256_mul_ps(facel,_mm256_set1_ps(charge[inr+2]));
144 vdwioffsetptr0 = vdwparam+2*nvdwtype*vdwtype[inr+0];
145
146 /* Avoid stupid compiler warnings */
147 jnrA = jnrB = jnrC = jnrD = jnrE = jnrF = jnrG = jnrH = 0;
148 j_coord_offsetA = 0;
149 j_coord_offsetB = 0;
150 j_coord_offsetC = 0;
151 j_coord_offsetD = 0;
152 j_coord_offsetE = 0;
153 j_coord_offsetF = 0;
154 j_coord_offsetG = 0;
155 j_coord_offsetH = 0;
156
157 outeriter = 0;
158 inneriter = 0;
159
160 for(iidx=0;iidx<4*DIM;iidx++)
161 {
162 scratch[iidx] = 0.0;
163 }
164
165 /* Start outer loop over neighborlists */
166 for(iidx=0; iidx<nri; iidx++)
167 {
168 /* Load shift vector for this list */
169 i_shift_offset = DIM*shiftidx[iidx];
170
171 /* Load limits for loop over neighbors */
172 j_index_start = jindex[iidx];
173 j_index_end = jindex[iidx+1];
174
175 /* Get outer coordinate index */
176 inr = iinr[iidx];
177 i_coord_offset = DIM*inr;
178
179 /* Load i particle coords and add shift vector */
180 gmx_mm256_load_shift_and_3rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,
181 &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2);
182
183 fix0 = _mm256_setzero_ps();
184 fiy0 = _mm256_setzero_ps();
185 fiz0 = _mm256_setzero_ps();
186 fix1 = _mm256_setzero_ps();
187 fiy1 = _mm256_setzero_ps();
188 fiz1 = _mm256_setzero_ps();
189 fix2 = _mm256_setzero_ps();
190 fiy2 = _mm256_setzero_ps();
191 fiz2 = _mm256_setzero_ps();
192
193 /* Reset potential sums */
194 velecsum = _mm256_setzero_ps();
195 vvdwsum = _mm256_setzero_ps();
196
197 /* Start inner kernel loop */
198 for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+7]>=0; jidx+=8)
199 {
200
201 /* Get j neighbor index, and coordinate index */
202 jnrA = jjnr[jidx];
203 jnrB = jjnr[jidx+1];
204 jnrC = jjnr[jidx+2];
205 jnrD = jjnr[jidx+3];
206 jnrE = jjnr[jidx+4];
207 jnrF = jjnr[jidx+5];
208 jnrG = jjnr[jidx+6];
209 jnrH = jjnr[jidx+7];
210 j_coord_offsetA = DIM*jnrA;
211 j_coord_offsetB = DIM*jnrB;
212 j_coord_offsetC = DIM*jnrC;
213 j_coord_offsetD = DIM*jnrD;
214 j_coord_offsetE = DIM*jnrE;
215 j_coord_offsetF = DIM*jnrF;
216 j_coord_offsetG = DIM*jnrG;
217 j_coord_offsetH = DIM*jnrH;
218
219 /* load j atom coordinates */
220 gmx_mm256_load_1rvec_8ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
221 x+j_coord_offsetC,x+j_coord_offsetD,
222 x+j_coord_offsetE,x+j_coord_offsetF,
223 x+j_coord_offsetG,x+j_coord_offsetH,
224 &jx0,&jy0,&jz0);
225
226 /* Calculate displacement vector */
227 dx00 = _mm256_sub_ps(ix0,jx0);
228 dy00 = _mm256_sub_ps(iy0,jy0);
229 dz00 = _mm256_sub_ps(iz0,jz0);
230 dx10 = _mm256_sub_ps(ix1,jx0);
231 dy10 = _mm256_sub_ps(iy1,jy0);
232 dz10 = _mm256_sub_ps(iz1,jz0);
233 dx20 = _mm256_sub_ps(ix2,jx0);
234 dy20 = _mm256_sub_ps(iy2,jy0);
235 dz20 = _mm256_sub_ps(iz2,jz0);
236
237 /* Calculate squared distance and things based on it */
238 rsq00 = gmx_mm256_calc_rsq_ps(dx00,dy00,dz00);
239 rsq10 = gmx_mm256_calc_rsq_ps(dx10,dy10,dz10);
240 rsq20 = gmx_mm256_calc_rsq_ps(dx20,dy20,dz20);
241
242 rinv00 = avx256_invsqrt_f(rsq00);
243 rinv10 = avx256_invsqrt_f(rsq10);
244 rinv20 = avx256_invsqrt_f(rsq20);
245
246 rinvsq00 = _mm256_mul_ps(rinv00,rinv00);
247 rinvsq10 = _mm256_mul_ps(rinv10,rinv10);
248 rinvsq20 = _mm256_mul_ps(rinv20,rinv20);
249
250 /* Load parameters for j particles */
251 jq0 = gmx_mm256_load_8real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
252 charge+jnrC+0,charge+jnrD+0,
253 charge+jnrE+0,charge+jnrF+0,
254 charge+jnrG+0,charge+jnrH+0);
255 vdwjidx0A = 2*vdwtype[jnrA+0];
256 vdwjidx0B = 2*vdwtype[jnrB+0];
257 vdwjidx0C = 2*vdwtype[jnrC+0];
258 vdwjidx0D = 2*vdwtype[jnrD+0];
259 vdwjidx0E = 2*vdwtype[jnrE+0];
260 vdwjidx0F = 2*vdwtype[jnrF+0];
261 vdwjidx0G = 2*vdwtype[jnrG+0];
262 vdwjidx0H = 2*vdwtype[jnrH+0];
263
264 fjx0 = _mm256_setzero_ps();
265 fjy0 = _mm256_setzero_ps();
266 fjz0 = _mm256_setzero_ps();
267
268 /**************************
269 * CALCULATE INTERACTIONS *
270 **************************/
271
272 r00 = _mm256_mul_ps(rsq00,rinv00);
273
274 /* Compute parameters for interactions between i and j atoms */
275 qq00 = _mm256_mul_ps(iq0,jq0);
276 gmx_mm256_load_8pair_swizzle_ps(vdwioffsetptr0+vdwjidx0A,
277 vdwioffsetptr0+vdwjidx0B,
278 vdwioffsetptr0+vdwjidx0C,
279 vdwioffsetptr0+vdwjidx0D,
280 vdwioffsetptr0+vdwjidx0E,
281 vdwioffsetptr0+vdwjidx0F,
282 vdwioffsetptr0+vdwjidx0G,
283 vdwioffsetptr0+vdwjidx0H,
284 &c6_00,&c12_00);
285
286 /* EWALD ELECTROSTATICS */
287
288 /* Analytical PME correction */
289 zeta2 = _mm256_mul_ps(beta2,rsq00);
290 rinv3 = _mm256_mul_ps(rinvsq00,rinv00);
291 pmecorrF = avx256_pmecorrF_f(zeta2);
292 felec = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
293 felec = _mm256_mul_ps(qq00,felec);
294 pmecorrV = avx256_pmecorrV_f(zeta2);
295 pmecorrV = _mm256_mul_ps(pmecorrV,beta);
296 velec = _mm256_sub_ps(rinv00,pmecorrV);
297 velec = _mm256_mul_ps(qq00,velec);
298
299 /* LENNARD-JONES DISPERSION/REPULSION */
300
301 rinvsix = _mm256_mul_ps(_mm256_mul_ps(rinvsq00,rinvsq00),rinvsq00);
302 vvdw6 = _mm256_mul_ps(c6_00,rinvsix);
303 vvdw12 = _mm256_mul_ps(c12_00,_mm256_mul_ps(rinvsix,rinvsix));
304 vvdw = _mm256_sub_ps( _mm256_mul_ps(vvdw12,one_twelfth) , _mm256_mul_ps(vvdw6,one_sixth) );
305 fvdw = _mm256_mul_ps(_mm256_sub_ps(vvdw12,vvdw6),rinvsq00);
306
307 /* Update potential sum for this i atom from the interaction with this j atom. */
308 velecsum = _mm256_add_ps(velecsum,velec);
309 vvdwsum = _mm256_add_ps(vvdwsum,vvdw);
310
311 fscal = _mm256_add_ps(felec,fvdw);
312
313 /* Calculate temporary vectorial force */
314 tx = _mm256_mul_ps(fscal,dx00);
315 ty = _mm256_mul_ps(fscal,dy00);
316 tz = _mm256_mul_ps(fscal,dz00);
317
318 /* Update vectorial force */
319 fix0 = _mm256_add_ps(fix0,tx);
320 fiy0 = _mm256_add_ps(fiy0,ty);
321 fiz0 = _mm256_add_ps(fiz0,tz);
322
323 fjx0 = _mm256_add_ps(fjx0,tx);
324 fjy0 = _mm256_add_ps(fjy0,ty);
325 fjz0 = _mm256_add_ps(fjz0,tz);
326
327 /**************************
328 * CALCULATE INTERACTIONS *
329 **************************/
330
331 r10 = _mm256_mul_ps(rsq10,rinv10);
332
333 /* Compute parameters for interactions between i and j atoms */
334 qq10 = _mm256_mul_ps(iq1,jq0);
335
336 /* EWALD ELECTROSTATICS */
337
338 /* Analytical PME correction */
339 zeta2 = _mm256_mul_ps(beta2,rsq10);
340 rinv3 = _mm256_mul_ps(rinvsq10,rinv10);
341 pmecorrF = avx256_pmecorrF_f(zeta2);
342 felec = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
343 felec = _mm256_mul_ps(qq10,felec);
344 pmecorrV = avx256_pmecorrV_f(zeta2);
345 pmecorrV = _mm256_mul_ps(pmecorrV,beta);
346 velec = _mm256_sub_ps(rinv10,pmecorrV);
347 velec = _mm256_mul_ps(qq10,velec);
348
349 /* Update potential sum for this i atom from the interaction with this j atom. */
350 velecsum = _mm256_add_ps(velecsum,velec);
351
352 fscal = felec;
353
354 /* Calculate temporary vectorial force */
355 tx = _mm256_mul_ps(fscal,dx10);
356 ty = _mm256_mul_ps(fscal,dy10);
357 tz = _mm256_mul_ps(fscal,dz10);
358
359 /* Update vectorial force */
360 fix1 = _mm256_add_ps(fix1,tx);
361 fiy1 = _mm256_add_ps(fiy1,ty);
362 fiz1 = _mm256_add_ps(fiz1,tz);
363
364 fjx0 = _mm256_add_ps(fjx0,tx);
365 fjy0 = _mm256_add_ps(fjy0,ty);
366 fjz0 = _mm256_add_ps(fjz0,tz);
367
368 /**************************
369 * CALCULATE INTERACTIONS *
370 **************************/
371
372 r20 = _mm256_mul_ps(rsq20,rinv20);
373
374 /* Compute parameters for interactions between i and j atoms */
375 qq20 = _mm256_mul_ps(iq2,jq0);
376
377 /* EWALD ELECTROSTATICS */
378
379 /* Analytical PME correction */
380 zeta2 = _mm256_mul_ps(beta2,rsq20);
381 rinv3 = _mm256_mul_ps(rinvsq20,rinv20);
382 pmecorrF = avx256_pmecorrF_f(zeta2);
383 felec = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
384 felec = _mm256_mul_ps(qq20,felec);
385 pmecorrV = avx256_pmecorrV_f(zeta2);
386 pmecorrV = _mm256_mul_ps(pmecorrV,beta);
387 velec = _mm256_sub_ps(rinv20,pmecorrV);
388 velec = _mm256_mul_ps(qq20,velec);
389
390 /* Update potential sum for this i atom from the interaction with this j atom. */
391 velecsum = _mm256_add_ps(velecsum,velec);
392
393 fscal = felec;
394
395 /* Calculate temporary vectorial force */
396 tx = _mm256_mul_ps(fscal,dx20);
397 ty = _mm256_mul_ps(fscal,dy20);
398 tz = _mm256_mul_ps(fscal,dz20);
399
400 /* Update vectorial force */
401 fix2 = _mm256_add_ps(fix2,tx);
402 fiy2 = _mm256_add_ps(fiy2,ty);
403 fiz2 = _mm256_add_ps(fiz2,tz);
404
405 fjx0 = _mm256_add_ps(fjx0,tx);
406 fjy0 = _mm256_add_ps(fjy0,ty);
407 fjz0 = _mm256_add_ps(fjz0,tz);
408
409 fjptrA = f+j_coord_offsetA;
410 fjptrB = f+j_coord_offsetB;
411 fjptrC = f+j_coord_offsetC;
412 fjptrD = f+j_coord_offsetD;
413 fjptrE = f+j_coord_offsetE;
414 fjptrF = f+j_coord_offsetF;
415 fjptrG = f+j_coord_offsetG;
416 fjptrH = f+j_coord_offsetH;
417
418 gmx_mm256_decrement_1rvec_8ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjptrE,fjptrF,fjptrG,fjptrH,fjx0,fjy0,fjz0);
419
420 /* Inner loop uses 267 flops */
421 }
422
423 if(jidx<j_index_end)
424 {
425
426 /* Get j neighbor index, and coordinate index */
427 jnrlistA = jjnr[jidx];
428 jnrlistB = jjnr[jidx+1];
429 jnrlistC = jjnr[jidx+2];
430 jnrlistD = jjnr[jidx+3];
431 jnrlistE = jjnr[jidx+4];
432 jnrlistF = jjnr[jidx+5];
433 jnrlistG = jjnr[jidx+6];
434 jnrlistH = jjnr[jidx+7];
435 /* Sign of each element will be negative for non-real atoms.
436 * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
437 * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
438 */
439 dummy_mask = gmx_mm256_set_m128(gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx+4)),_mm_setzero_si128())),
440 gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128())));
441
442 jnrA = (jnrlistA>=0) ? jnrlistA : 0;
443 jnrB = (jnrlistB>=0) ? jnrlistB : 0;
444 jnrC = (jnrlistC>=0) ? jnrlistC : 0;
445 jnrD = (jnrlistD>=0) ? jnrlistD : 0;
446 jnrE = (jnrlistE>=0) ? jnrlistE : 0;
447 jnrF = (jnrlistF>=0) ? jnrlistF : 0;
448 jnrG = (jnrlistG>=0) ? jnrlistG : 0;
449 jnrH = (jnrlistH>=0) ? jnrlistH : 0;
450 j_coord_offsetA = DIM*jnrA;
451 j_coord_offsetB = DIM*jnrB;
452 j_coord_offsetC = DIM*jnrC;
453 j_coord_offsetD = DIM*jnrD;
454 j_coord_offsetE = DIM*jnrE;
455 j_coord_offsetF = DIM*jnrF;
456 j_coord_offsetG = DIM*jnrG;
457 j_coord_offsetH = DIM*jnrH;
458
459 /* load j atom coordinates */
460 gmx_mm256_load_1rvec_8ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
461 x+j_coord_offsetC,x+j_coord_offsetD,
462 x+j_coord_offsetE,x+j_coord_offsetF,
463 x+j_coord_offsetG,x+j_coord_offsetH,
464 &jx0,&jy0,&jz0);
465
466 /* Calculate displacement vector */
467 dx00 = _mm256_sub_ps(ix0,jx0);
468 dy00 = _mm256_sub_ps(iy0,jy0);
469 dz00 = _mm256_sub_ps(iz0,jz0);
470 dx10 = _mm256_sub_ps(ix1,jx0);
471 dy10 = _mm256_sub_ps(iy1,jy0);
472 dz10 = _mm256_sub_ps(iz1,jz0);
473 dx20 = _mm256_sub_ps(ix2,jx0);
474 dy20 = _mm256_sub_ps(iy2,jy0);
475 dz20 = _mm256_sub_ps(iz2,jz0);
476
477 /* Calculate squared distance and things based on it */
478 rsq00 = gmx_mm256_calc_rsq_ps(dx00,dy00,dz00);
479 rsq10 = gmx_mm256_calc_rsq_ps(dx10,dy10,dz10);
480 rsq20 = gmx_mm256_calc_rsq_ps(dx20,dy20,dz20);
481
482 rinv00 = avx256_invsqrt_f(rsq00);
483 rinv10 = avx256_invsqrt_f(rsq10);
484 rinv20 = avx256_invsqrt_f(rsq20);
485
486 rinvsq00 = _mm256_mul_ps(rinv00,rinv00);
487 rinvsq10 = _mm256_mul_ps(rinv10,rinv10);
488 rinvsq20 = _mm256_mul_ps(rinv20,rinv20);
489
490 /* Load parameters for j particles */
491 jq0 = gmx_mm256_load_8real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
492 charge+jnrC+0,charge+jnrD+0,
493 charge+jnrE+0,charge+jnrF+0,
494 charge+jnrG+0,charge+jnrH+0);
495 vdwjidx0A = 2*vdwtype[jnrA+0];
496 vdwjidx0B = 2*vdwtype[jnrB+0];
497 vdwjidx0C = 2*vdwtype[jnrC+0];
498 vdwjidx0D = 2*vdwtype[jnrD+0];
499 vdwjidx0E = 2*vdwtype[jnrE+0];
500 vdwjidx0F = 2*vdwtype[jnrF+0];
501 vdwjidx0G = 2*vdwtype[jnrG+0];
502 vdwjidx0H = 2*vdwtype[jnrH+0];
503
504 fjx0 = _mm256_setzero_ps();
505 fjy0 = _mm256_setzero_ps();
506 fjz0 = _mm256_setzero_ps();
507
508 /**************************
509 * CALCULATE INTERACTIONS *
510 **************************/
511
512 r00 = _mm256_mul_ps(rsq00,rinv00);
513 r00 = _mm256_andnot_ps(dummy_mask,r00);
514
515 /* Compute parameters for interactions between i and j atoms */
516 qq00 = _mm256_mul_ps(iq0,jq0);
517 gmx_mm256_load_8pair_swizzle_ps(vdwioffsetptr0+vdwjidx0A,
518 vdwioffsetptr0+vdwjidx0B,
519 vdwioffsetptr0+vdwjidx0C,
520 vdwioffsetptr0+vdwjidx0D,
521 vdwioffsetptr0+vdwjidx0E,
522 vdwioffsetptr0+vdwjidx0F,
523 vdwioffsetptr0+vdwjidx0G,
524 vdwioffsetptr0+vdwjidx0H,
525 &c6_00,&c12_00);
526
527 /* EWALD ELECTROSTATICS */
528
529 /* Analytical PME correction */
530 zeta2 = _mm256_mul_ps(beta2,rsq00);
531 rinv3 = _mm256_mul_ps(rinvsq00,rinv00);
532 pmecorrF = avx256_pmecorrF_f(zeta2);
533 felec = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
534 felec = _mm256_mul_ps(qq00,felec);
535 pmecorrV = avx256_pmecorrV_f(zeta2);
536 pmecorrV = _mm256_mul_ps(pmecorrV,beta);
537 velec = _mm256_sub_ps(rinv00,pmecorrV);
538 velec = _mm256_mul_ps(qq00,velec);
539
540 /* LENNARD-JONES DISPERSION/REPULSION */
541
542 rinvsix = _mm256_mul_ps(_mm256_mul_ps(rinvsq00,rinvsq00),rinvsq00);
543 vvdw6 = _mm256_mul_ps(c6_00,rinvsix);
544 vvdw12 = _mm256_mul_ps(c12_00,_mm256_mul_ps(rinvsix,rinvsix));
545 vvdw = _mm256_sub_ps( _mm256_mul_ps(vvdw12,one_twelfth) , _mm256_mul_ps(vvdw6,one_sixth) );
546 fvdw = _mm256_mul_ps(_mm256_sub_ps(vvdw12,vvdw6),rinvsq00);
547
548 /* Update potential sum for this i atom from the interaction with this j atom. */
549 velec = _mm256_andnot_ps(dummy_mask,velec);
550 velecsum = _mm256_add_ps(velecsum,velec);
551 vvdw = _mm256_andnot_ps(dummy_mask,vvdw);
552 vvdwsum = _mm256_add_ps(vvdwsum,vvdw);
553
554 fscal = _mm256_add_ps(felec,fvdw);
555
556 fscal = _mm256_andnot_ps(dummy_mask,fscal);
557
558 /* Calculate temporary vectorial force */
559 tx = _mm256_mul_ps(fscal,dx00);
560 ty = _mm256_mul_ps(fscal,dy00);
561 tz = _mm256_mul_ps(fscal,dz00);
562
563 /* Update vectorial force */
564 fix0 = _mm256_add_ps(fix0,tx);
565 fiy0 = _mm256_add_ps(fiy0,ty);
566 fiz0 = _mm256_add_ps(fiz0,tz);
567
568 fjx0 = _mm256_add_ps(fjx0,tx);
569 fjy0 = _mm256_add_ps(fjy0,ty);
570 fjz0 = _mm256_add_ps(fjz0,tz);
571
572 /**************************
573 * CALCULATE INTERACTIONS *
574 **************************/
575
576 r10 = _mm256_mul_ps(rsq10,rinv10);
577 r10 = _mm256_andnot_ps(dummy_mask,r10);
578
579 /* Compute parameters for interactions between i and j atoms */
580 qq10 = _mm256_mul_ps(iq1,jq0);
581
582 /* EWALD ELECTROSTATICS */
583
584 /* Analytical PME correction */
585 zeta2 = _mm256_mul_ps(beta2,rsq10);
586 rinv3 = _mm256_mul_ps(rinvsq10,rinv10);
587 pmecorrF = avx256_pmecorrF_f(zeta2);
588 felec = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
589 felec = _mm256_mul_ps(qq10,felec);
590 pmecorrV = avx256_pmecorrV_f(zeta2);
591 pmecorrV = _mm256_mul_ps(pmecorrV,beta);
592 velec = _mm256_sub_ps(rinv10,pmecorrV);
593 velec = _mm256_mul_ps(qq10,velec);
594
595 /* Update potential sum for this i atom from the interaction with this j atom. */
596 velec = _mm256_andnot_ps(dummy_mask,velec);
597 velecsum = _mm256_add_ps(velecsum,velec);
598
599 fscal = felec;
600
601 fscal = _mm256_andnot_ps(dummy_mask,fscal);
602
603 /* Calculate temporary vectorial force */
604 tx = _mm256_mul_ps(fscal,dx10);
605 ty = _mm256_mul_ps(fscal,dy10);
606 tz = _mm256_mul_ps(fscal,dz10);
607
608 /* Update vectorial force */
609 fix1 = _mm256_add_ps(fix1,tx);
610 fiy1 = _mm256_add_ps(fiy1,ty);
611 fiz1 = _mm256_add_ps(fiz1,tz);
612
613 fjx0 = _mm256_add_ps(fjx0,tx);
614 fjy0 = _mm256_add_ps(fjy0,ty);
615 fjz0 = _mm256_add_ps(fjz0,tz);
616
617 /**************************
618 * CALCULATE INTERACTIONS *
619 **************************/
620
621 r20 = _mm256_mul_ps(rsq20,rinv20);
622 r20 = _mm256_andnot_ps(dummy_mask,r20);
623
624 /* Compute parameters for interactions between i and j atoms */
625 qq20 = _mm256_mul_ps(iq2,jq0);
626
627 /* EWALD ELECTROSTATICS */
628
629 /* Analytical PME correction */
630 zeta2 = _mm256_mul_ps(beta2,rsq20);
631 rinv3 = _mm256_mul_ps(rinvsq20,rinv20);
632 pmecorrF = avx256_pmecorrF_f(zeta2);
633 felec = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
634 felec = _mm256_mul_ps(qq20,felec);
635 pmecorrV = avx256_pmecorrV_f(zeta2);
636 pmecorrV = _mm256_mul_ps(pmecorrV,beta);
637 velec = _mm256_sub_ps(rinv20,pmecorrV);
638 velec = _mm256_mul_ps(qq20,velec);
639
640 /* Update potential sum for this i atom from the interaction with this j atom. */
641 velec = _mm256_andnot_ps(dummy_mask,velec);
642 velecsum = _mm256_add_ps(velecsum,velec);
643
644 fscal = felec;
645
646 fscal = _mm256_andnot_ps(dummy_mask,fscal);
647
648 /* Calculate temporary vectorial force */
649 tx = _mm256_mul_ps(fscal,dx20);
650 ty = _mm256_mul_ps(fscal,dy20);
651 tz = _mm256_mul_ps(fscal,dz20);
652
653 /* Update vectorial force */
654 fix2 = _mm256_add_ps(fix2,tx);
655 fiy2 = _mm256_add_ps(fiy2,ty);
656 fiz2 = _mm256_add_ps(fiz2,tz);
657
658 fjx0 = _mm256_add_ps(fjx0,tx);
659 fjy0 = _mm256_add_ps(fjy0,ty);
660 fjz0 = _mm256_add_ps(fjz0,tz);
661
662 fjptrA = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
663 fjptrB = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
664 fjptrC = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
665 fjptrD = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
666 fjptrE = (jnrlistE>=0) ? f+j_coord_offsetE : scratch;
667 fjptrF = (jnrlistF>=0) ? f+j_coord_offsetF : scratch;
668 fjptrG = (jnrlistG>=0) ? f+j_coord_offsetG : scratch;
669 fjptrH = (jnrlistH>=0) ? f+j_coord_offsetH : scratch;
670
671 gmx_mm256_decrement_1rvec_8ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjptrE,fjptrF,fjptrG,fjptrH,fjx0,fjy0,fjz0);
672
673 /* Inner loop uses 270 flops */
674 }
675
676 /* End of innermost loop */
677
678 gmx_mm256_update_iforce_3atom_swizzle_ps(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,
679 f+i_coord_offset,fshift+i_shift_offset);
680
681 ggid = gid[iidx];
682 /* Update potential energies */
683 gmx_mm256_update_1pot_ps(velecsum,kernel_data->energygrp_elec+ggid);
684 gmx_mm256_update_1pot_ps(vvdwsum,kernel_data->energygrp_vdw+ggid);
685
686 /* Increment number of inner iterations */
687 inneriter += j_index_end - j_index_start;
688
689 /* Outer loop uses 20 flops */
690 }
691
692 /* Increment number of outer iterations */
693 outeriter += nri;
694
695 /* Update outer/inner flops */
696
697 inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W3_VF,outeriter*20 + inneriter*270);
698 }
699 /*
700 * Gromacs nonbonded kernel: nb_kernel_ElecEw_VdwLJ_GeomW3P1_F_avx_256_single
701 * Electrostatics interaction: Ewald
702 * VdW interaction: LennardJones
703 * Geometry: Water3-Particle
704 * Calculate force/pot: Force
705 */
706 void
707 nb_kernel_ElecEw_VdwLJ_GeomW3P1_F_avx_256_single
708 (t_nblist * gmx_restrict nlist,
709 rvec * gmx_restrict xx,
710 rvec * gmx_restrict ff,
711 struct t_forcerec * gmx_restrict fr,
712 t_mdatoms * gmx_restrict mdatoms,
713 nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
714 t_nrnb * gmx_restrict nrnb)
715 {
716 /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
717 * just 0 for non-waters.
718 * Suffixes A,B,C,D,E,F,G,H refer to j loop unrolling done with AVX, e.g. for the eight different
719 * jnr indices corresponding to data put in the four positions in the SIMD register.
720 */
721 int i_shift_offset,i_coord_offset,outeriter,inneriter;
722 int j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
723 int jnrA,jnrB,jnrC,jnrD;
724 int jnrE,jnrF,jnrG,jnrH;
725 int jnrlistA,jnrlistB,jnrlistC,jnrlistD;
726 int jnrlistE,jnrlistF,jnrlistG,jnrlistH;
727 int j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
728 int j_coord_offsetE,j_coord_offsetF,j_coord_offsetG,j_coord_offsetH;
729 int *iinr,*jindex,*jjnr,*shiftidx,*gid;
730 real rcutoff_scalar;
731 real *shiftvec,*fshift,*x,*f;
732 real *fjptrA,*fjptrB,*fjptrC,*fjptrD,*fjptrE,*fjptrF,*fjptrG,*fjptrH;
733 real scratch[4*DIM];
734 __m256 tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
735 real * vdwioffsetptr0;
736 __m256 ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
737 real * vdwioffsetptr1;
738 __m256 ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
739 real * vdwioffsetptr2;
740 __m256 ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
741 int vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D,vdwjidx0E,vdwjidx0F,vdwjidx0G,vdwjidx0H;
742 __m256 jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
743 __m256 dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
744 __m256 dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
745 __m256 dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
746 __m256 velec,felec,velecsum,facel,crf,krf,krf2;
747 real *charge;
748 int nvdwtype;
749 __m256 rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
750 int *vdwtype;
751 real *vdwparam;
752 __m256 one_sixth = _mm256_set1_ps(1.0/6.0);
753 __m256 one_twelfth = _mm256_set1_ps(1.0/12.0);
754 __m256i ewitab;
755 __m128i ewitab_lo,ewitab_hi;
756 __m256 ewtabscale,eweps,sh_ewald,ewrt,ewtabhalfspace,ewtabF,ewtabFn,ewtabD,ewtabV;
757 __m256 beta,beta2,beta3,zeta2,pmecorrF,pmecorrV,rinv3;
758 real *ewtab;
759 __m256 dummy_mask,cutoff_mask;
760 __m256 signbit = _mm256_castsi256_ps( _mm256_set1_epi32(0x80000000) );
761 __m256 one = _mm256_set1_ps(1.0);
762 __m256 two = _mm256_set1_ps(2.0);
763 x = xx[0];
764 f = ff[0];
765
766 nri = nlist->nri;
767 iinr = nlist->iinr;
768 jindex = nlist->jindex;
769 jjnr = nlist->jjnr;
770 shiftidx = nlist->shift;
771 gid = nlist->gid;
772 shiftvec = fr->shift_vec[0];
773 fshift = fr->fshift[0];
774 facel = _mm256_set1_ps(fr->ic->epsfac);
775 charge = mdatoms->chargeA;
776 nvdwtype = fr->ntype;
777 vdwparam = fr->nbfp;
778 vdwtype = mdatoms->typeA;
779
780 sh_ewald = _mm256_set1_ps(fr->ic->sh_ewald);
781 beta = _mm256_set1_ps(fr->ic->ewaldcoeff_q);
782 beta2 = _mm256_mul_ps(beta,beta);
783 beta3 = _mm256_mul_ps(beta,beta2);
784
785 ewtab = fr->ic->tabq_coul_F;
786 ewtabscale = _mm256_set1_ps(fr->ic->tabq_scale);
787 ewtabhalfspace = _mm256_set1_ps(0.5/fr->ic->tabq_scale);
788
789 /* Setup water-specific parameters */
790 inr = nlist->iinr[0];
791 iq0 = _mm256_mul_ps(facel,_mm256_set1_ps(charge[inr+0]));
792 iq1 = _mm256_mul_ps(facel,_mm256_set1_ps(charge[inr+1]));
793 iq2 = _mm256_mul_ps(facel,_mm256_set1_ps(charge[inr+2]));
794 vdwioffsetptr0 = vdwparam+2*nvdwtype*vdwtype[inr+0];
795
796 /* Avoid stupid compiler warnings */
797 jnrA = jnrB = jnrC = jnrD = jnrE = jnrF = jnrG = jnrH = 0;
798 j_coord_offsetA = 0;
799 j_coord_offsetB = 0;
800 j_coord_offsetC = 0;
801 j_coord_offsetD = 0;
802 j_coord_offsetE = 0;
803 j_coord_offsetF = 0;
804 j_coord_offsetG = 0;
805 j_coord_offsetH = 0;
806
807 outeriter = 0;
808 inneriter = 0;
809
810 for(iidx=0;iidx<4*DIM;iidx++)
811 {
812 scratch[iidx] = 0.0;
813 }
814
815 /* Start outer loop over neighborlists */
816 for(iidx=0; iidx<nri; iidx++)
817 {
818 /* Load shift vector for this list */
819 i_shift_offset = DIM*shiftidx[iidx];
820
821 /* Load limits for loop over neighbors */
822 j_index_start = jindex[iidx];
823 j_index_end = jindex[iidx+1];
824
825 /* Get outer coordinate index */
826 inr = iinr[iidx];
827 i_coord_offset = DIM*inr;
828
829 /* Load i particle coords and add shift vector */
830 gmx_mm256_load_shift_and_3rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,
831 &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2);
832
833 fix0 = _mm256_setzero_ps();
834 fiy0 = _mm256_setzero_ps();
835 fiz0 = _mm256_setzero_ps();
836 fix1 = _mm256_setzero_ps();
837 fiy1 = _mm256_setzero_ps();
838 fiz1 = _mm256_setzero_ps();
839 fix2 = _mm256_setzero_ps();
840 fiy2 = _mm256_setzero_ps();
841 fiz2 = _mm256_setzero_ps();
842
843 /* Start inner kernel loop */
844 for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+7]>=0; jidx+=8)
845 {
846
847 /* Get j neighbor index, and coordinate index */
848 jnrA = jjnr[jidx];
849 jnrB = jjnr[jidx+1];
850 jnrC = jjnr[jidx+2];
851 jnrD = jjnr[jidx+3];
852 jnrE = jjnr[jidx+4];
853 jnrF = jjnr[jidx+5];
854 jnrG = jjnr[jidx+6];
855 jnrH = jjnr[jidx+7];
856 j_coord_offsetA = DIM*jnrA;
857 j_coord_offsetB = DIM*jnrB;
858 j_coord_offsetC = DIM*jnrC;
859 j_coord_offsetD = DIM*jnrD;
860 j_coord_offsetE = DIM*jnrE;
861 j_coord_offsetF = DIM*jnrF;
862 j_coord_offsetG = DIM*jnrG;
863 j_coord_offsetH = DIM*jnrH;
864
865 /* load j atom coordinates */
866 gmx_mm256_load_1rvec_8ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
867 x+j_coord_offsetC,x+j_coord_offsetD,
868 x+j_coord_offsetE,x+j_coord_offsetF,
869 x+j_coord_offsetG,x+j_coord_offsetH,
870 &jx0,&jy0,&jz0);
871
872 /* Calculate displacement vector */
873 dx00 = _mm256_sub_ps(ix0,jx0);
874 dy00 = _mm256_sub_ps(iy0,jy0);
875 dz00 = _mm256_sub_ps(iz0,jz0);
876 dx10 = _mm256_sub_ps(ix1,jx0);
877 dy10 = _mm256_sub_ps(iy1,jy0);
878 dz10 = _mm256_sub_ps(iz1,jz0);
879 dx20 = _mm256_sub_ps(ix2,jx0);
880 dy20 = _mm256_sub_ps(iy2,jy0);
881 dz20 = _mm256_sub_ps(iz2,jz0);
882
883 /* Calculate squared distance and things based on it */
884 rsq00 = gmx_mm256_calc_rsq_ps(dx00,dy00,dz00);
885 rsq10 = gmx_mm256_calc_rsq_ps(dx10,dy10,dz10);
886 rsq20 = gmx_mm256_calc_rsq_ps(dx20,dy20,dz20);
887
888 rinv00 = avx256_invsqrt_f(rsq00);
889 rinv10 = avx256_invsqrt_f(rsq10);
890 rinv20 = avx256_invsqrt_f(rsq20);
891
892 rinvsq00 = _mm256_mul_ps(rinv00,rinv00);
893 rinvsq10 = _mm256_mul_ps(rinv10,rinv10);
894 rinvsq20 = _mm256_mul_ps(rinv20,rinv20);
895
896 /* Load parameters for j particles */
897 jq0 = gmx_mm256_load_8real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
898 charge+jnrC+0,charge+jnrD+0,
899 charge+jnrE+0,charge+jnrF+0,
900 charge+jnrG+0,charge+jnrH+0);
901 vdwjidx0A = 2*vdwtype[jnrA+0];
902 vdwjidx0B = 2*vdwtype[jnrB+0];
903 vdwjidx0C = 2*vdwtype[jnrC+0];
904 vdwjidx0D = 2*vdwtype[jnrD+0];
905 vdwjidx0E = 2*vdwtype[jnrE+0];
906 vdwjidx0F = 2*vdwtype[jnrF+0];
907 vdwjidx0G = 2*vdwtype[jnrG+0];
908 vdwjidx0H = 2*vdwtype[jnrH+0];
909
910 fjx0 = _mm256_setzero_ps();
911 fjy0 = _mm256_setzero_ps();
912 fjz0 = _mm256_setzero_ps();
913
914 /**************************
915 * CALCULATE INTERACTIONS *
916 **************************/
917
918 r00 = _mm256_mul_ps(rsq00,rinv00);
919
920 /* Compute parameters for interactions between i and j atoms */
921 qq00 = _mm256_mul_ps(iq0,jq0);
922 gmx_mm256_load_8pair_swizzle_ps(vdwioffsetptr0+vdwjidx0A,
923 vdwioffsetptr0+vdwjidx0B,
924 vdwioffsetptr0+vdwjidx0C,
925 vdwioffsetptr0+vdwjidx0D,
926 vdwioffsetptr0+vdwjidx0E,
927 vdwioffsetptr0+vdwjidx0F,
928 vdwioffsetptr0+vdwjidx0G,
929 vdwioffsetptr0+vdwjidx0H,
930 &c6_00,&c12_00);
931
932 /* EWALD ELECTROSTATICS */
933
934 /* Analytical PME correction */
935 zeta2 = _mm256_mul_ps(beta2,rsq00);
936 rinv3 = _mm256_mul_ps(rinvsq00,rinv00);
937 pmecorrF = avx256_pmecorrF_f(zeta2);
938 felec = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
939 felec = _mm256_mul_ps(qq00,felec);
940
941 /* LENNARD-JONES DISPERSION/REPULSION */
942
943 rinvsix = _mm256_mul_ps(_mm256_mul_ps(rinvsq00,rinvsq00),rinvsq00);
944 fvdw = _mm256_mul_ps(_mm256_sub_ps(_mm256_mul_ps(c12_00,rinvsix),c6_00),_mm256_mul_ps(rinvsix,rinvsq00));
945
946 fscal = _mm256_add_ps(felec,fvdw);
947
948 /* Calculate temporary vectorial force */
949 tx = _mm256_mul_ps(fscal,dx00);
950 ty = _mm256_mul_ps(fscal,dy00);
951 tz = _mm256_mul_ps(fscal,dz00);
952
953 /* Update vectorial force */
954 fix0 = _mm256_add_ps(fix0,tx);
955 fiy0 = _mm256_add_ps(fiy0,ty);
956 fiz0 = _mm256_add_ps(fiz0,tz);
957
958 fjx0 = _mm256_add_ps(fjx0,tx);
959 fjy0 = _mm256_add_ps(fjy0,ty);
960 fjz0 = _mm256_add_ps(fjz0,tz);
961
962 /**************************
963 * CALCULATE INTERACTIONS *
964 **************************/
965
966 r10 = _mm256_mul_ps(rsq10,rinv10);
967
968 /* Compute parameters for interactions between i and j atoms */
969 qq10 = _mm256_mul_ps(iq1,jq0);
970
971 /* EWALD ELECTROSTATICS */
972
973 /* Analytical PME correction */
974 zeta2 = _mm256_mul_ps(beta2,rsq10);
975 rinv3 = _mm256_mul_ps(rinvsq10,rinv10);
976 pmecorrF = avx256_pmecorrF_f(zeta2);
977 felec = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
978 felec = _mm256_mul_ps(qq10,felec);
979
980 fscal = felec;
981
982 /* Calculate temporary vectorial force */
983 tx = _mm256_mul_ps(fscal,dx10);
984 ty = _mm256_mul_ps(fscal,dy10);
985 tz = _mm256_mul_ps(fscal,dz10);
986
987 /* Update vectorial force */
988 fix1 = _mm256_add_ps(fix1,tx);
989 fiy1 = _mm256_add_ps(fiy1,ty);
990 fiz1 = _mm256_add_ps(fiz1,tz);
991
992 fjx0 = _mm256_add_ps(fjx0,tx);
993 fjy0 = _mm256_add_ps(fjy0,ty);
994 fjz0 = _mm256_add_ps(fjz0,tz);
995
996 /**************************
997 * CALCULATE INTERACTIONS *
998 **************************/
999
1000 r20 = _mm256_mul_ps(rsq20,rinv20);
1001
1002 /* Compute parameters for interactions between i and j atoms */
1003 qq20 = _mm256_mul_ps(iq2,jq0);
1004
1005 /* EWALD ELECTROSTATICS */
1006
1007 /* Analytical PME correction */
1008 zeta2 = _mm256_mul_ps(beta2,rsq20);
1009 rinv3 = _mm256_mul_ps(rinvsq20,rinv20);
1010 pmecorrF = avx256_pmecorrF_f(zeta2);
1011 felec = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
1012 felec = _mm256_mul_ps(qq20,felec);
1013
1014 fscal = felec;
1015
1016 /* Calculate temporary vectorial force */
1017 tx = _mm256_mul_ps(fscal,dx20);
1018 ty = _mm256_mul_ps(fscal,dy20);
1019 tz = _mm256_mul_ps(fscal,dz20);
1020
1021 /* Update vectorial force */
1022 fix2 = _mm256_add_ps(fix2,tx);
1023 fiy2 = _mm256_add_ps(fiy2,ty);
1024 fiz2 = _mm256_add_ps(fiz2,tz);
1025
1026 fjx0 = _mm256_add_ps(fjx0,tx);
1027 fjy0 = _mm256_add_ps(fjy0,ty);
1028 fjz0 = _mm256_add_ps(fjz0,tz);
1029
1030 fjptrA = f+j_coord_offsetA;
1031 fjptrB = f+j_coord_offsetB;
1032 fjptrC = f+j_coord_offsetC;
1033 fjptrD = f+j_coord_offsetD;
1034 fjptrE = f+j_coord_offsetE;
1035 fjptrF = f+j_coord_offsetF;
1036 fjptrG = f+j_coord_offsetG;
1037 fjptrH = f+j_coord_offsetH;
1038
1039 gmx_mm256_decrement_1rvec_8ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjptrE,fjptrF,fjptrG,fjptrH,fjx0,fjy0,fjz0);
1040
1041 /* Inner loop uses 178 flops */
1042 }
1043
1044 if(jidx<j_index_end)
1045 {
1046
1047 /* Get j neighbor index, and coordinate index */
1048 jnrlistA = jjnr[jidx];
1049 jnrlistB = jjnr[jidx+1];
1050 jnrlistC = jjnr[jidx+2];
1051 jnrlistD = jjnr[jidx+3];
1052 jnrlistE = jjnr[jidx+4];
1053 jnrlistF = jjnr[jidx+5];
1054 jnrlistG = jjnr[jidx+6];
1055 jnrlistH = jjnr[jidx+7];
1056 /* Sign of each element will be negative for non-real atoms.
1057 * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
1058 * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
1059 */
1060 dummy_mask = gmx_mm256_set_m128(gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx+4)),_mm_setzero_si128())),
1061 gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128())));
1062
1063 jnrA = (jnrlistA>=0) ? jnrlistA : 0;
1064 jnrB = (jnrlistB>=0) ? jnrlistB : 0;
1065 jnrC = (jnrlistC>=0) ? jnrlistC : 0;
1066 jnrD = (jnrlistD>=0) ? jnrlistD : 0;
1067 jnrE = (jnrlistE>=0) ? jnrlistE : 0;
1068 jnrF = (jnrlistF>=0) ? jnrlistF : 0;
1069 jnrG = (jnrlistG>=0) ? jnrlistG : 0;
1070 jnrH = (jnrlistH>=0) ? jnrlistH : 0;
1071 j_coord_offsetA = DIM*jnrA;
1072 j_coord_offsetB = DIM*jnrB;
1073 j_coord_offsetC = DIM*jnrC;
1074 j_coord_offsetD = DIM*jnrD;
1075 j_coord_offsetE = DIM*jnrE;
1076 j_coord_offsetF = DIM*jnrF;
1077 j_coord_offsetG = DIM*jnrG;
1078 j_coord_offsetH = DIM*jnrH;
1079
1080 /* load j atom coordinates */
1081 gmx_mm256_load_1rvec_8ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
1082 x+j_coord_offsetC,x+j_coord_offsetD,
1083 x+j_coord_offsetE,x+j_coord_offsetF,
1084 x+j_coord_offsetG,x+j_coord_offsetH,
1085 &jx0,&jy0,&jz0);
1086
1087 /* Calculate displacement vector */
1088 dx00 = _mm256_sub_ps(ix0,jx0);
1089 dy00 = _mm256_sub_ps(iy0,jy0);
1090 dz00 = _mm256_sub_ps(iz0,jz0);
1091 dx10 = _mm256_sub_ps(ix1,jx0);
1092 dy10 = _mm256_sub_ps(iy1,jy0);
1093 dz10 = _mm256_sub_ps(iz1,jz0);
1094 dx20 = _mm256_sub_ps(ix2,jx0);
1095 dy20 = _mm256_sub_ps(iy2,jy0);
1096 dz20 = _mm256_sub_ps(iz2,jz0);
1097
1098 /* Calculate squared distance and things based on it */
1099 rsq00 = gmx_mm256_calc_rsq_ps(dx00,dy00,dz00);
1100 rsq10 = gmx_mm256_calc_rsq_ps(dx10,dy10,dz10);
1101 rsq20 = gmx_mm256_calc_rsq_ps(dx20,dy20,dz20);
1102
1103 rinv00 = avx256_invsqrt_f(rsq00);
1104 rinv10 = avx256_invsqrt_f(rsq10);
1105 rinv20 = avx256_invsqrt_f(rsq20);
1106
1107 rinvsq00 = _mm256_mul_ps(rinv00,rinv00);
1108 rinvsq10 = _mm256_mul_ps(rinv10,rinv10);
1109 rinvsq20 = _mm256_mul_ps(rinv20,rinv20);
1110
1111 /* Load parameters for j particles */
1112 jq0 = gmx_mm256_load_8real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
1113 charge+jnrC+0,charge+jnrD+0,
1114 charge+jnrE+0,charge+jnrF+0,
1115 charge+jnrG+0,charge+jnrH+0);
1116 vdwjidx0A = 2*vdwtype[jnrA+0];
1117 vdwjidx0B = 2*vdwtype[jnrB+0];
1118 vdwjidx0C = 2*vdwtype[jnrC+0];
1119 vdwjidx0D = 2*vdwtype[jnrD+0];
1120 vdwjidx0E = 2*vdwtype[jnrE+0];
1121 vdwjidx0F = 2*vdwtype[jnrF+0];
1122 vdwjidx0G = 2*vdwtype[jnrG+0];
1123 vdwjidx0H = 2*vdwtype[jnrH+0];
1124
1125 fjx0 = _mm256_setzero_ps();
1126 fjy0 = _mm256_setzero_ps();
1127 fjz0 = _mm256_setzero_ps();
1128
1129 /**************************
1130 * CALCULATE INTERACTIONS *
1131 **************************/
1132
1133 r00 = _mm256_mul_ps(rsq00,rinv00);
1134 r00 = _mm256_andnot_ps(dummy_mask,r00);
1135
1136 /* Compute parameters for interactions between i and j atoms */
1137 qq00 = _mm256_mul_ps(iq0,jq0);
1138 gmx_mm256_load_8pair_swizzle_ps(vdwioffsetptr0+vdwjidx0A,
1139 vdwioffsetptr0+vdwjidx0B,
1140 vdwioffsetptr0+vdwjidx0C,
1141 vdwioffsetptr0+vdwjidx0D,
1142 vdwioffsetptr0+vdwjidx0E,
1143 vdwioffsetptr0+vdwjidx0F,
1144 vdwioffsetptr0+vdwjidx0G,
1145 vdwioffsetptr0+vdwjidx0H,
1146 &c6_00,&c12_00);
1147
1148 /* EWALD ELECTROSTATICS */
1149
1150 /* Analytical PME correction */
1151 zeta2 = _mm256_mul_ps(beta2,rsq00);
1152 rinv3 = _mm256_mul_ps(rinvsq00,rinv00);
1153 pmecorrF = avx256_pmecorrF_f(zeta2);
1154 felec = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
1155 felec = _mm256_mul_ps(qq00,felec);
1156
1157 /* LENNARD-JONES DISPERSION/REPULSION */
1158
1159 rinvsix = _mm256_mul_ps(_mm256_mul_ps(rinvsq00,rinvsq00),rinvsq00);
1160 fvdw = _mm256_mul_ps(_mm256_sub_ps(_mm256_mul_ps(c12_00,rinvsix),c6_00),_mm256_mul_ps(rinvsix,rinvsq00));
1161
1162 fscal = _mm256_add_ps(felec,fvdw);
1163
1164 fscal = _mm256_andnot_ps(dummy_mask,fscal);
1165
1166 /* Calculate temporary vectorial force */
1167 tx = _mm256_mul_ps(fscal,dx00);
1168 ty = _mm256_mul_ps(fscal,dy00);
1169 tz = _mm256_mul_ps(fscal,dz00);
1170
1171 /* Update vectorial force */
1172 fix0 = _mm256_add_ps(fix0,tx);
1173 fiy0 = _mm256_add_ps(fiy0,ty);
1174 fiz0 = _mm256_add_ps(fiz0,tz);
1175
1176 fjx0 = _mm256_add_ps(fjx0,tx);
1177 fjy0 = _mm256_add_ps(fjy0,ty);
1178 fjz0 = _mm256_add_ps(fjz0,tz);
1179
1180 /**************************
1181 * CALCULATE INTERACTIONS *
1182 **************************/
1183
1184 r10 = _mm256_mul_ps(rsq10,rinv10);
1185 r10 = _mm256_andnot_ps(dummy_mask,r10);
1186
1187 /* Compute parameters for interactions between i and j atoms */
1188 qq10 = _mm256_mul_ps(iq1,jq0);
1189
1190 /* EWALD ELECTROSTATICS */
1191
1192 /* Analytical PME correction */
1193 zeta2 = _mm256_mul_ps(beta2,rsq10);
1194 rinv3 = _mm256_mul_ps(rinvsq10,rinv10);
1195 pmecorrF = avx256_pmecorrF_f(zeta2);
1196 felec = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
1197 felec = _mm256_mul_ps(qq10,felec);
1198
1199 fscal = felec;
1200
1201 fscal = _mm256_andnot_ps(dummy_mask,fscal);
1202
1203 /* Calculate temporary vectorial force */
1204 tx = _mm256_mul_ps(fscal,dx10);
1205 ty = _mm256_mul_ps(fscal,dy10);
1206 tz = _mm256_mul_ps(fscal,dz10);
1207
1208 /* Update vectorial force */
1209 fix1 = _mm256_add_ps(fix1,tx);
1210 fiy1 = _mm256_add_ps(fiy1,ty);
1211 fiz1 = _mm256_add_ps(fiz1,tz);
1212
1213 fjx0 = _mm256_add_ps(fjx0,tx);
1214 fjy0 = _mm256_add_ps(fjy0,ty);
1215 fjz0 = _mm256_add_ps(fjz0,tz);
1216
1217 /**************************
1218 * CALCULATE INTERACTIONS *
1219 **************************/
1220
1221 r20 = _mm256_mul_ps(rsq20,rinv20);
1222 r20 = _mm256_andnot_ps(dummy_mask,r20);
1223
1224 /* Compute parameters for interactions between i and j atoms */
1225 qq20 = _mm256_mul_ps(iq2,jq0);
1226
1227 /* EWALD ELECTROSTATICS */
1228
1229 /* Analytical PME correction */
1230 zeta2 = _mm256_mul_ps(beta2,rsq20);
1231 rinv3 = _mm256_mul_ps(rinvsq20,rinv20);
1232 pmecorrF = avx256_pmecorrF_f(zeta2);
1233 felec = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
1234 felec = _mm256_mul_ps(qq20,felec);
1235
1236 fscal = felec;
1237
1238 fscal = _mm256_andnot_ps(dummy_mask,fscal);
1239
1240 /* Calculate temporary vectorial force */
1241 tx = _mm256_mul_ps(fscal,dx20);
1242 ty = _mm256_mul_ps(fscal,dy20);
1243 tz = _mm256_mul_ps(fscal,dz20);
1244
1245 /* Update vectorial force */
1246 fix2 = _mm256_add_ps(fix2,tx);
1247 fiy2 = _mm256_add_ps(fiy2,ty);
1248 fiz2 = _mm256_add_ps(fiz2,tz);
1249
1250 fjx0 = _mm256_add_ps(fjx0,tx);
1251 fjy0 = _mm256_add_ps(fjy0,ty);
1252 fjz0 = _mm256_add_ps(fjz0,tz);
1253
1254 fjptrA = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
1255 fjptrB = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
1256 fjptrC = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
1257 fjptrD = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
1258 fjptrE = (jnrlistE>=0) ? f+j_coord_offsetE : scratch;
1259 fjptrF = (jnrlistF>=0) ? f+j_coord_offsetF : scratch;
1260 fjptrG = (jnrlistG>=0) ? f+j_coord_offsetG : scratch;
1261 fjptrH = (jnrlistH>=0) ? f+j_coord_offsetH : scratch;
1262
1263 gmx_mm256_decrement_1rvec_8ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjptrE,fjptrF,fjptrG,fjptrH,fjx0,fjy0,fjz0);
1264
1265 /* Inner loop uses 181 flops */
1266 }
1267
1268 /* End of innermost loop */
1269
1270 gmx_mm256_update_iforce_3atom_swizzle_ps(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,
1271 f+i_coord_offset,fshift+i_shift_offset);
1272
1273 /* Increment number of inner iterations */
1274 inneriter += j_index_end - j_index_start;
1275
1276 /* Outer loop uses 18 flops */
1277 }
1278
1279 /* Increment number of outer iterations */
1280 outeriter += nri;
1281
1282 /* Update outer/inner flops */
1283
1284 inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W3_F,outeriter*18 + inneriter*181);
1285 }
The ultimate molecular dynamics simulation package