?  search:
summary | log | graphiclog1 | graphiclog2 | commit | commitdiff | tree | refs | edit | fork
blame | history | raw | HEAD
Remove nb-parameters from t_forcerec
[gromacs.git] / src / gromacs / gmxlib / nonbonded / nb_kernel_avx_256_double / nb_kernel_ElecRFCut_VdwLJSh_GeomP1P1_avx_256_double.c
blob373c6c8fe0fb9faa931d85e5b4ccc4e6f0a97486
1 /*
2 * This file is part of the GROMACS molecular simulation package.
3 *
4 * Copyright (c) 2012,2013,2014,2015,2017, by the GROMACS development team, led by
5 * Mark Abraham, David van der Spoel, Berk Hess, and Erik Lindahl,
6 * and including many others, as listed in the AUTHORS file in the
7 * top-level source directory and at http://www.gromacs.org.
8 *
9 * GROMACS is free software; you can redistribute it and/or
10 * modify it under the terms of the GNU Lesser General Public License
11 * as published by the Free Software Foundation; either version 2.1
12 * of the License, or (at your option) any later version.
13 *
14 * GROMACS is distributed in the hope that it will be useful,
15 * but WITHOUT ANY WARRANTY; without even the implied warranty of
16 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
17 * Lesser General Public License for more details.
18 *
19 * You should have received a copy of the GNU Lesser General Public
20 * License along with GROMACS; if not, see
21 * http://www.gnu.org/licenses, or write to the Free Software Foundation,
22 * Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA.
23 *
24 * If you want to redistribute modifications to GROMACS, please
25 * consider that scientific software is very special. Version
26 * control is crucial - bugs must be traceable. We will be happy to
27 * consider code for inclusion in the official distribution, but
28 * derived work must not be called official GROMACS. Details are found
29 * in the README & COPYING files - if they are missing, get the
30 * official version at http://www.gromacs.org.
31 *
32 * To help us fund GROMACS development, we humbly ask that you cite
33 * the research papers on the package. Check out http://www.gromacs.org.
34 */
35 /*
36 * Note: this file was generated by the GROMACS avx_256_double 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_double.h"
48
49 /*
50 * Gromacs nonbonded kernel: nb_kernel_ElecRFCut_VdwLJSh_GeomP1P1_VF_avx_256_double
51 * Electrostatics interaction: ReactionField
52 * VdW interaction: LennardJones
53 * Geometry: Particle-Particle
54 * Calculate force/pot: PotentialAndForce
55 */
56 void
57 nb_kernel_ElecRFCut_VdwLJSh_GeomP1P1_VF_avx_256_double
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 refer to j loop unrolling done with AVX, e.g. for the four 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 jnrlistA,jnrlistB,jnrlistC,jnrlistD;
75 int jnrlistE,jnrlistF,jnrlistG,jnrlistH;
76 int j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
77 int *iinr,*jindex,*jjnr,*shiftidx,*gid;
78 real rcutoff_scalar;
79 real *shiftvec,*fshift,*x,*f;
80 real *fjptrA,*fjptrB,*fjptrC,*fjptrD;
81 real scratch[4*DIM];
82 __m256d tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
83 real * vdwioffsetptr0;
84 __m256d ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
85 int vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
86 __m256d jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
87 __m256d dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
88 __m256d velec,felec,velecsum,facel,crf,krf,krf2;
89 real *charge;
90 int nvdwtype;
91 __m256d rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
92 int *vdwtype;
93 real *vdwparam;
94 __m256d one_sixth = _mm256_set1_pd(1.0/6.0);
95 __m256d one_twelfth = _mm256_set1_pd(1.0/12.0);
96 __m256d dummy_mask,cutoff_mask;
97 __m128 tmpmask0,tmpmask1;
98 __m256d signbit = _mm256_castsi256_pd( _mm256_set1_epi32(0x80000000) );
99 __m256d one = _mm256_set1_pd(1.0);
100 __m256d two = _mm256_set1_pd(2.0);
101 x = xx[0];
102 f = ff[0];
103
104 nri = nlist->nri;
105 iinr = nlist->iinr;
106 jindex = nlist->jindex;
107 jjnr = nlist->jjnr;
108 shiftidx = nlist->shift;
109 gid = nlist->gid;
110 shiftvec = fr->shift_vec[0];
111 fshift = fr->fshift[0];
112 facel = _mm256_set1_pd(fr->ic->epsfac);
113 charge = mdatoms->chargeA;
114 krf = _mm256_set1_pd(fr->ic->k_rf);
115 krf2 = _mm256_set1_pd(fr->ic->k_rf*2.0);
116 crf = _mm256_set1_pd(fr->ic->c_rf);
117 nvdwtype = fr->ntype;
118 vdwparam = fr->nbfp;
119 vdwtype = mdatoms->typeA;
120
121 /* When we use explicit cutoffs the value must be identical for elec and VdW, so use elec as an arbitrary choice */
122 rcutoff_scalar = fr->ic->rcoulomb;
123 rcutoff = _mm256_set1_pd(rcutoff_scalar);
124 rcutoff2 = _mm256_mul_pd(rcutoff,rcutoff);
125
126 sh_vdw_invrcut6 = _mm256_set1_pd(fr->ic->sh_invrc6);
127 rvdw = _mm256_set1_pd(fr->ic->rvdw);
128
129 /* Avoid stupid compiler warnings */
130 jnrA = jnrB = jnrC = jnrD = 0;
131 j_coord_offsetA = 0;
132 j_coord_offsetB = 0;
133 j_coord_offsetC = 0;
134 j_coord_offsetD = 0;
135
136 outeriter = 0;
137 inneriter = 0;
138
139 for(iidx=0;iidx<4*DIM;iidx++)
140 {
141 scratch[iidx] = 0.0;
142 }
143
144 /* Start outer loop over neighborlists */
145 for(iidx=0; iidx<nri; iidx++)
146 {
147 /* Load shift vector for this list */
148 i_shift_offset = DIM*shiftidx[iidx];
149
150 /* Load limits for loop over neighbors */
151 j_index_start = jindex[iidx];
152 j_index_end = jindex[iidx+1];
153
154 /* Get outer coordinate index */
155 inr = iinr[iidx];
156 i_coord_offset = DIM*inr;
157
158 /* Load i particle coords and add shift vector */
159 gmx_mm256_load_shift_and_1rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset,&ix0,&iy0,&iz0);
160
161 fix0 = _mm256_setzero_pd();
162 fiy0 = _mm256_setzero_pd();
163 fiz0 = _mm256_setzero_pd();
164
165 /* Load parameters for i particles */
166 iq0 = _mm256_mul_pd(facel,_mm256_set1_pd(charge[inr+0]));
167 vdwioffsetptr0 = vdwparam+2*nvdwtype*vdwtype[inr+0];
168
169 /* Reset potential sums */
170 velecsum = _mm256_setzero_pd();
171 vvdwsum = _mm256_setzero_pd();
172
173 /* Start inner kernel loop */
174 for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
175 {
176
177 /* Get j neighbor index, and coordinate index */
178 jnrA = jjnr[jidx];
179 jnrB = jjnr[jidx+1];
180 jnrC = jjnr[jidx+2];
181 jnrD = jjnr[jidx+3];
182 j_coord_offsetA = DIM*jnrA;
183 j_coord_offsetB = DIM*jnrB;
184 j_coord_offsetC = DIM*jnrC;
185 j_coord_offsetD = DIM*jnrD;
186
187 /* load j atom coordinates */
188 gmx_mm256_load_1rvec_4ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
189 x+j_coord_offsetC,x+j_coord_offsetD,
190 &jx0,&jy0,&jz0);
191
192 /* Calculate displacement vector */
193 dx00 = _mm256_sub_pd(ix0,jx0);
194 dy00 = _mm256_sub_pd(iy0,jy0);
195 dz00 = _mm256_sub_pd(iz0,jz0);
196
197 /* Calculate squared distance and things based on it */
198 rsq00 = gmx_mm256_calc_rsq_pd(dx00,dy00,dz00);
199
200 rinv00 = avx256_invsqrt_d(rsq00);
201
202 rinvsq00 = _mm256_mul_pd(rinv00,rinv00);
203
204 /* Load parameters for j particles */
205 jq0 = gmx_mm256_load_4real_swizzle_pd(charge+jnrA+0,charge+jnrB+0,
206 charge+jnrC+0,charge+jnrD+0);
207 vdwjidx0A = 2*vdwtype[jnrA+0];
208 vdwjidx0B = 2*vdwtype[jnrB+0];
209 vdwjidx0C = 2*vdwtype[jnrC+0];
210 vdwjidx0D = 2*vdwtype[jnrD+0];
211
212 /**************************
213 * CALCULATE INTERACTIONS *
214 **************************/
215
216 if (gmx_mm256_any_lt(rsq00,rcutoff2))
217 {
218
219 /* Compute parameters for interactions between i and j atoms */
220 qq00 = _mm256_mul_pd(iq0,jq0);
221 gmx_mm256_load_4pair_swizzle_pd(vdwioffsetptr0+vdwjidx0A,
222 vdwioffsetptr0+vdwjidx0B,
223 vdwioffsetptr0+vdwjidx0C,
224 vdwioffsetptr0+vdwjidx0D,
225 &c6_00,&c12_00);
226
227 /* REACTION-FIELD ELECTROSTATICS */
228 velec = _mm256_mul_pd(qq00,_mm256_sub_pd(_mm256_add_pd(rinv00,_mm256_mul_pd(krf,rsq00)),crf));
229 felec = _mm256_mul_pd(qq00,_mm256_sub_pd(_mm256_mul_pd(rinv00,rinvsq00),krf2));
230
231 /* LENNARD-JONES DISPERSION/REPULSION */
232
233 rinvsix = _mm256_mul_pd(_mm256_mul_pd(rinvsq00,rinvsq00),rinvsq00);
234 vvdw6 = _mm256_mul_pd(c6_00,rinvsix);
235 vvdw12 = _mm256_mul_pd(c12_00,_mm256_mul_pd(rinvsix,rinvsix));
236 vvdw = _mm256_sub_pd(_mm256_mul_pd( _mm256_sub_pd(vvdw12 , _mm256_mul_pd(c12_00,_mm256_mul_pd(sh_vdw_invrcut6,sh_vdw_invrcut6))), one_twelfth) ,
237 _mm256_mul_pd( _mm256_sub_pd(vvdw6,_mm256_mul_pd(c6_00,sh_vdw_invrcut6)),one_sixth));
238 fvdw = _mm256_mul_pd(_mm256_sub_pd(vvdw12,vvdw6),rinvsq00);
239
240 cutoff_mask = _mm256_cmp_pd(rsq00,rcutoff2,_CMP_LT_OQ);
241
242 /* Update potential sum for this i atom from the interaction with this j atom. */
243 velec = _mm256_and_pd(velec,cutoff_mask);
244 velecsum = _mm256_add_pd(velecsum,velec);
245 vvdw = _mm256_and_pd(vvdw,cutoff_mask);
246 vvdwsum = _mm256_add_pd(vvdwsum,vvdw);
247
248 fscal = _mm256_add_pd(felec,fvdw);
249
250 fscal = _mm256_and_pd(fscal,cutoff_mask);
251
252 /* Calculate temporary vectorial force */
253 tx = _mm256_mul_pd(fscal,dx00);
254 ty = _mm256_mul_pd(fscal,dy00);
255 tz = _mm256_mul_pd(fscal,dz00);
256
257 /* Update vectorial force */
258 fix0 = _mm256_add_pd(fix0,tx);
259 fiy0 = _mm256_add_pd(fiy0,ty);
260 fiz0 = _mm256_add_pd(fiz0,tz);
261
262 fjptrA = f+j_coord_offsetA;
263 fjptrB = f+j_coord_offsetB;
264 fjptrC = f+j_coord_offsetC;
265 fjptrD = f+j_coord_offsetD;
266 gmx_mm256_decrement_1rvec_4ptr_swizzle_pd(fjptrA,fjptrB,fjptrC,fjptrD,tx,ty,tz);
267
268 }
269
270 /* Inner loop uses 54 flops */
271 }
272
273 if(jidx<j_index_end)
274 {
275
276 /* Get j neighbor index, and coordinate index */
277 jnrlistA = jjnr[jidx];
278 jnrlistB = jjnr[jidx+1];
279 jnrlistC = jjnr[jidx+2];
280 jnrlistD = jjnr[jidx+3];
281 /* Sign of each element will be negative for non-real atoms.
282 * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
283 * so use it as val = _mm_andnot_pd(mask,val) to clear dummy entries.
284 */
285 tmpmask0 = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
286
287 tmpmask1 = _mm_permute_ps(tmpmask0,_GMX_MM_PERMUTE(3,3,2,2));
288 tmpmask0 = _mm_permute_ps(tmpmask0,_GMX_MM_PERMUTE(1,1,0,0));
289 dummy_mask = _mm256_castps_pd(gmx_mm256_set_m128(tmpmask1,tmpmask0));
290
291 jnrA = (jnrlistA>=0) ? jnrlistA : 0;
292 jnrB = (jnrlistB>=0) ? jnrlistB : 0;
293 jnrC = (jnrlistC>=0) ? jnrlistC : 0;
294 jnrD = (jnrlistD>=0) ? jnrlistD : 0;
295 j_coord_offsetA = DIM*jnrA;
296 j_coord_offsetB = DIM*jnrB;
297 j_coord_offsetC = DIM*jnrC;
298 j_coord_offsetD = DIM*jnrD;
299
300 /* load j atom coordinates */
301 gmx_mm256_load_1rvec_4ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
302 x+j_coord_offsetC,x+j_coord_offsetD,
303 &jx0,&jy0,&jz0);
304
305 /* Calculate displacement vector */
306 dx00 = _mm256_sub_pd(ix0,jx0);
307 dy00 = _mm256_sub_pd(iy0,jy0);
308 dz00 = _mm256_sub_pd(iz0,jz0);
309
310 /* Calculate squared distance and things based on it */
311 rsq00 = gmx_mm256_calc_rsq_pd(dx00,dy00,dz00);
312
313 rinv00 = avx256_invsqrt_d(rsq00);
314
315 rinvsq00 = _mm256_mul_pd(rinv00,rinv00);
316
317 /* Load parameters for j particles */
318 jq0 = gmx_mm256_load_4real_swizzle_pd(charge+jnrA+0,charge+jnrB+0,
319 charge+jnrC+0,charge+jnrD+0);
320 vdwjidx0A = 2*vdwtype[jnrA+0];
321 vdwjidx0B = 2*vdwtype[jnrB+0];
322 vdwjidx0C = 2*vdwtype[jnrC+0];
323 vdwjidx0D = 2*vdwtype[jnrD+0];
324
325 /**************************
326 * CALCULATE INTERACTIONS *
327 **************************/
328
329 if (gmx_mm256_any_lt(rsq00,rcutoff2))
330 {
331
332 /* Compute parameters for interactions between i and j atoms */
333 qq00 = _mm256_mul_pd(iq0,jq0);
334 gmx_mm256_load_4pair_swizzle_pd(vdwioffsetptr0+vdwjidx0A,
335 vdwioffsetptr0+vdwjidx0B,
336 vdwioffsetptr0+vdwjidx0C,
337 vdwioffsetptr0+vdwjidx0D,
338 &c6_00,&c12_00);
339
340 /* REACTION-FIELD ELECTROSTATICS */
341 velec = _mm256_mul_pd(qq00,_mm256_sub_pd(_mm256_add_pd(rinv00,_mm256_mul_pd(krf,rsq00)),crf));
342 felec = _mm256_mul_pd(qq00,_mm256_sub_pd(_mm256_mul_pd(rinv00,rinvsq00),krf2));
343
344 /* LENNARD-JONES DISPERSION/REPULSION */
345
346 rinvsix = _mm256_mul_pd(_mm256_mul_pd(rinvsq00,rinvsq00),rinvsq00);
347 vvdw6 = _mm256_mul_pd(c6_00,rinvsix);
348 vvdw12 = _mm256_mul_pd(c12_00,_mm256_mul_pd(rinvsix,rinvsix));
349 vvdw = _mm256_sub_pd(_mm256_mul_pd( _mm256_sub_pd(vvdw12 , _mm256_mul_pd(c12_00,_mm256_mul_pd(sh_vdw_invrcut6,sh_vdw_invrcut6))), one_twelfth) ,
350 _mm256_mul_pd( _mm256_sub_pd(vvdw6,_mm256_mul_pd(c6_00,sh_vdw_invrcut6)),one_sixth));
351 fvdw = _mm256_mul_pd(_mm256_sub_pd(vvdw12,vvdw6),rinvsq00);
352
353 cutoff_mask = _mm256_cmp_pd(rsq00,rcutoff2,_CMP_LT_OQ);
354
355 /* Update potential sum for this i atom from the interaction with this j atom. */
356 velec = _mm256_and_pd(velec,cutoff_mask);
357 velec = _mm256_andnot_pd(dummy_mask,velec);
358 velecsum = _mm256_add_pd(velecsum,velec);
359 vvdw = _mm256_and_pd(vvdw,cutoff_mask);
360 vvdw = _mm256_andnot_pd(dummy_mask,vvdw);
361 vvdwsum = _mm256_add_pd(vvdwsum,vvdw);
362
363 fscal = _mm256_add_pd(felec,fvdw);
364
365 fscal = _mm256_and_pd(fscal,cutoff_mask);
366
367 fscal = _mm256_andnot_pd(dummy_mask,fscal);
368
369 /* Calculate temporary vectorial force */
370 tx = _mm256_mul_pd(fscal,dx00);
371 ty = _mm256_mul_pd(fscal,dy00);
372 tz = _mm256_mul_pd(fscal,dz00);
373
374 /* Update vectorial force */
375 fix0 = _mm256_add_pd(fix0,tx);
376 fiy0 = _mm256_add_pd(fiy0,ty);
377 fiz0 = _mm256_add_pd(fiz0,tz);
378
379 fjptrA = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
380 fjptrB = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
381 fjptrC = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
382 fjptrD = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
383 gmx_mm256_decrement_1rvec_4ptr_swizzle_pd(fjptrA,fjptrB,fjptrC,fjptrD,tx,ty,tz);
384
385 }
386
387 /* Inner loop uses 54 flops */
388 }
389
390 /* End of innermost loop */
391
392 gmx_mm256_update_iforce_1atom_swizzle_pd(fix0,fiy0,fiz0,
393 f+i_coord_offset,fshift+i_shift_offset);
394
395 ggid = gid[iidx];
396 /* Update potential energies */
397 gmx_mm256_update_1pot_pd(velecsum,kernel_data->energygrp_elec+ggid);
398 gmx_mm256_update_1pot_pd(vvdwsum,kernel_data->energygrp_vdw+ggid);
399
400 /* Increment number of inner iterations */
401 inneriter += j_index_end - j_index_start;
402
403 /* Outer loop uses 9 flops */
404 }
405
406 /* Increment number of outer iterations */
407 outeriter += nri;
408
409 /* Update outer/inner flops */
410
411 inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_VF,outeriter*9 + inneriter*54);
412 }
413 /*
414 * Gromacs nonbonded kernel: nb_kernel_ElecRFCut_VdwLJSh_GeomP1P1_F_avx_256_double
415 * Electrostatics interaction: ReactionField
416 * VdW interaction: LennardJones
417 * Geometry: Particle-Particle
418 * Calculate force/pot: Force
419 */
420 void
421 nb_kernel_ElecRFCut_VdwLJSh_GeomP1P1_F_avx_256_double
422 (t_nblist * gmx_restrict nlist,
423 rvec * gmx_restrict xx,
424 rvec * gmx_restrict ff,
425 struct t_forcerec * gmx_restrict fr,
426 t_mdatoms * gmx_restrict mdatoms,
427 nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
428 t_nrnb * gmx_restrict nrnb)
429 {
430 /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
431 * just 0 for non-waters.
432 * Suffixes A,B,C,D refer to j loop unrolling done with AVX, e.g. for the four different
433 * jnr indices corresponding to data put in the four positions in the SIMD register.
434 */
435 int i_shift_offset,i_coord_offset,outeriter,inneriter;
436 int j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
437 int jnrA,jnrB,jnrC,jnrD;
438 int jnrlistA,jnrlistB,jnrlistC,jnrlistD;
439 int jnrlistE,jnrlistF,jnrlistG,jnrlistH;
440 int j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
441 int *iinr,*jindex,*jjnr,*shiftidx,*gid;
442 real rcutoff_scalar;
443 real *shiftvec,*fshift,*x,*f;
444 real *fjptrA,*fjptrB,*fjptrC,*fjptrD;
445 real scratch[4*DIM];
446 __m256d tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
447 real * vdwioffsetptr0;
448 __m256d ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
449 int vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
450 __m256d jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
451 __m256d dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
452 __m256d velec,felec,velecsum,facel,crf,krf,krf2;
453 real *charge;
454 int nvdwtype;
455 __m256d rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
456 int *vdwtype;
457 real *vdwparam;
458 __m256d one_sixth = _mm256_set1_pd(1.0/6.0);
459 __m256d one_twelfth = _mm256_set1_pd(1.0/12.0);
460 __m256d dummy_mask,cutoff_mask;
461 __m128 tmpmask0,tmpmask1;
462 __m256d signbit = _mm256_castsi256_pd( _mm256_set1_epi32(0x80000000) );
463 __m256d one = _mm256_set1_pd(1.0);
464 __m256d two = _mm256_set1_pd(2.0);
465 x = xx[0];
466 f = ff[0];
467
468 nri = nlist->nri;
469 iinr = nlist->iinr;
470 jindex = nlist->jindex;
471 jjnr = nlist->jjnr;
472 shiftidx = nlist->shift;
473 gid = nlist->gid;
474 shiftvec = fr->shift_vec[0];
475 fshift = fr->fshift[0];
476 facel = _mm256_set1_pd(fr->ic->epsfac);
477 charge = mdatoms->chargeA;
478 krf = _mm256_set1_pd(fr->ic->k_rf);
479 krf2 = _mm256_set1_pd(fr->ic->k_rf*2.0);
480 crf = _mm256_set1_pd(fr->ic->c_rf);
481 nvdwtype = fr->ntype;
482 vdwparam = fr->nbfp;
483 vdwtype = mdatoms->typeA;
484
485 /* When we use explicit cutoffs the value must be identical for elec and VdW, so use elec as an arbitrary choice */
486 rcutoff_scalar = fr->ic->rcoulomb;
487 rcutoff = _mm256_set1_pd(rcutoff_scalar);
488 rcutoff2 = _mm256_mul_pd(rcutoff,rcutoff);
489
490 sh_vdw_invrcut6 = _mm256_set1_pd(fr->ic->sh_invrc6);
491 rvdw = _mm256_set1_pd(fr->ic->rvdw);
492
493 /* Avoid stupid compiler warnings */
494 jnrA = jnrB = jnrC = jnrD = 0;
495 j_coord_offsetA = 0;
496 j_coord_offsetB = 0;
497 j_coord_offsetC = 0;
498 j_coord_offsetD = 0;
499
500 outeriter = 0;
501 inneriter = 0;
502
503 for(iidx=0;iidx<4*DIM;iidx++)
504 {
505 scratch[iidx] = 0.0;
506 }
507
508 /* Start outer loop over neighborlists */
509 for(iidx=0; iidx<nri; iidx++)
510 {
511 /* Load shift vector for this list */
512 i_shift_offset = DIM*shiftidx[iidx];
513
514 /* Load limits for loop over neighbors */
515 j_index_start = jindex[iidx];
516 j_index_end = jindex[iidx+1];
517
518 /* Get outer coordinate index */
519 inr = iinr[iidx];
520 i_coord_offset = DIM*inr;
521
522 /* Load i particle coords and add shift vector */
523 gmx_mm256_load_shift_and_1rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset,&ix0,&iy0,&iz0);
524
525 fix0 = _mm256_setzero_pd();
526 fiy0 = _mm256_setzero_pd();
527 fiz0 = _mm256_setzero_pd();
528
529 /* Load parameters for i particles */
530 iq0 = _mm256_mul_pd(facel,_mm256_set1_pd(charge[inr+0]));
531 vdwioffsetptr0 = vdwparam+2*nvdwtype*vdwtype[inr+0];
532
533 /* Start inner kernel loop */
534 for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
535 {
536
537 /* Get j neighbor index, and coordinate index */
538 jnrA = jjnr[jidx];
539 jnrB = jjnr[jidx+1];
540 jnrC = jjnr[jidx+2];
541 jnrD = jjnr[jidx+3];
542 j_coord_offsetA = DIM*jnrA;
543 j_coord_offsetB = DIM*jnrB;
544 j_coord_offsetC = DIM*jnrC;
545 j_coord_offsetD = DIM*jnrD;
546
547 /* load j atom coordinates */
548 gmx_mm256_load_1rvec_4ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
549 x+j_coord_offsetC,x+j_coord_offsetD,
550 &jx0,&jy0,&jz0);
551
552 /* Calculate displacement vector */
553 dx00 = _mm256_sub_pd(ix0,jx0);
554 dy00 = _mm256_sub_pd(iy0,jy0);
555 dz00 = _mm256_sub_pd(iz0,jz0);
556
557 /* Calculate squared distance and things based on it */
558 rsq00 = gmx_mm256_calc_rsq_pd(dx00,dy00,dz00);
559
560 rinv00 = avx256_invsqrt_d(rsq00);
561
562 rinvsq00 = _mm256_mul_pd(rinv00,rinv00);
563
564 /* Load parameters for j particles */
565 jq0 = gmx_mm256_load_4real_swizzle_pd(charge+jnrA+0,charge+jnrB+0,
566 charge+jnrC+0,charge+jnrD+0);
567 vdwjidx0A = 2*vdwtype[jnrA+0];
568 vdwjidx0B = 2*vdwtype[jnrB+0];
569 vdwjidx0C = 2*vdwtype[jnrC+0];
570 vdwjidx0D = 2*vdwtype[jnrD+0];
571
572 /**************************
573 * CALCULATE INTERACTIONS *
574 **************************/
575
576 if (gmx_mm256_any_lt(rsq00,rcutoff2))
577 {
578
579 /* Compute parameters for interactions between i and j atoms */
580 qq00 = _mm256_mul_pd(iq0,jq0);
581 gmx_mm256_load_4pair_swizzle_pd(vdwioffsetptr0+vdwjidx0A,
582 vdwioffsetptr0+vdwjidx0B,
583 vdwioffsetptr0+vdwjidx0C,
584 vdwioffsetptr0+vdwjidx0D,
585 &c6_00,&c12_00);
586
587 /* REACTION-FIELD ELECTROSTATICS */
588 felec = _mm256_mul_pd(qq00,_mm256_sub_pd(_mm256_mul_pd(rinv00,rinvsq00),krf2));
589
590 /* LENNARD-JONES DISPERSION/REPULSION */
591
592 rinvsix = _mm256_mul_pd(_mm256_mul_pd(rinvsq00,rinvsq00),rinvsq00);
593 fvdw = _mm256_mul_pd(_mm256_sub_pd(_mm256_mul_pd(c12_00,rinvsix),c6_00),_mm256_mul_pd(rinvsix,rinvsq00));
594
595 cutoff_mask = _mm256_cmp_pd(rsq00,rcutoff2,_CMP_LT_OQ);
596
597 fscal = _mm256_add_pd(felec,fvdw);
598
599 fscal = _mm256_and_pd(fscal,cutoff_mask);
600
601 /* Calculate temporary vectorial force */
602 tx = _mm256_mul_pd(fscal,dx00);
603 ty = _mm256_mul_pd(fscal,dy00);
604 tz = _mm256_mul_pd(fscal,dz00);
605
606 /* Update vectorial force */
607 fix0 = _mm256_add_pd(fix0,tx);
608 fiy0 = _mm256_add_pd(fiy0,ty);
609 fiz0 = _mm256_add_pd(fiz0,tz);
610
611 fjptrA = f+j_coord_offsetA;
612 fjptrB = f+j_coord_offsetB;
613 fjptrC = f+j_coord_offsetC;
614 fjptrD = f+j_coord_offsetD;
615 gmx_mm256_decrement_1rvec_4ptr_swizzle_pd(fjptrA,fjptrB,fjptrC,fjptrD,tx,ty,tz);
616
617 }
618
619 /* Inner loop uses 37 flops */
620 }
621
622 if(jidx<j_index_end)
623 {
624
625 /* Get j neighbor index, and coordinate index */
626 jnrlistA = jjnr[jidx];
627 jnrlistB = jjnr[jidx+1];
628 jnrlistC = jjnr[jidx+2];
629 jnrlistD = jjnr[jidx+3];
630 /* Sign of each element will be negative for non-real atoms.
631 * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
632 * so use it as val = _mm_andnot_pd(mask,val) to clear dummy entries.
633 */
634 tmpmask0 = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
635
636 tmpmask1 = _mm_permute_ps(tmpmask0,_GMX_MM_PERMUTE(3,3,2,2));
637 tmpmask0 = _mm_permute_ps(tmpmask0,_GMX_MM_PERMUTE(1,1,0,0));
638 dummy_mask = _mm256_castps_pd(gmx_mm256_set_m128(tmpmask1,tmpmask0));
639
640 jnrA = (jnrlistA>=0) ? jnrlistA : 0;
641 jnrB = (jnrlistB>=0) ? jnrlistB : 0;
642 jnrC = (jnrlistC>=0) ? jnrlistC : 0;
643 jnrD = (jnrlistD>=0) ? jnrlistD : 0;
644 j_coord_offsetA = DIM*jnrA;
645 j_coord_offsetB = DIM*jnrB;
646 j_coord_offsetC = DIM*jnrC;
647 j_coord_offsetD = DIM*jnrD;
648
649 /* load j atom coordinates */
650 gmx_mm256_load_1rvec_4ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
651 x+j_coord_offsetC,x+j_coord_offsetD,
652 &jx0,&jy0,&jz0);
653
654 /* Calculate displacement vector */
655 dx00 = _mm256_sub_pd(ix0,jx0);
656 dy00 = _mm256_sub_pd(iy0,jy0);
657 dz00 = _mm256_sub_pd(iz0,jz0);
658
659 /* Calculate squared distance and things based on it */
660 rsq00 = gmx_mm256_calc_rsq_pd(dx00,dy00,dz00);
661
662 rinv00 = avx256_invsqrt_d(rsq00);
663
664 rinvsq00 = _mm256_mul_pd(rinv00,rinv00);
665
666 /* Load parameters for j particles */
667 jq0 = gmx_mm256_load_4real_swizzle_pd(charge+jnrA+0,charge+jnrB+0,
668 charge+jnrC+0,charge+jnrD+0);
669 vdwjidx0A = 2*vdwtype[jnrA+0];
670 vdwjidx0B = 2*vdwtype[jnrB+0];
671 vdwjidx0C = 2*vdwtype[jnrC+0];
672 vdwjidx0D = 2*vdwtype[jnrD+0];
673
674 /**************************
675 * CALCULATE INTERACTIONS *
676 **************************/
677
678 if (gmx_mm256_any_lt(rsq00,rcutoff2))
679 {
680
681 /* Compute parameters for interactions between i and j atoms */
682 qq00 = _mm256_mul_pd(iq0,jq0);
683 gmx_mm256_load_4pair_swizzle_pd(vdwioffsetptr0+vdwjidx0A,
684 vdwioffsetptr0+vdwjidx0B,
685 vdwioffsetptr0+vdwjidx0C,
686 vdwioffsetptr0+vdwjidx0D,
687 &c6_00,&c12_00);
688
689 /* REACTION-FIELD ELECTROSTATICS */
690 felec = _mm256_mul_pd(qq00,_mm256_sub_pd(_mm256_mul_pd(rinv00,rinvsq00),krf2));
691
692 /* LENNARD-JONES DISPERSION/REPULSION */
693
694 rinvsix = _mm256_mul_pd(_mm256_mul_pd(rinvsq00,rinvsq00),rinvsq00);
695 fvdw = _mm256_mul_pd(_mm256_sub_pd(_mm256_mul_pd(c12_00,rinvsix),c6_00),_mm256_mul_pd(rinvsix,rinvsq00));
696
697 cutoff_mask = _mm256_cmp_pd(rsq00,rcutoff2,_CMP_LT_OQ);
698
699 fscal = _mm256_add_pd(felec,fvdw);
700
701 fscal = _mm256_and_pd(fscal,cutoff_mask);
702
703 fscal = _mm256_andnot_pd(dummy_mask,fscal);
704
705 /* Calculate temporary vectorial force */
706 tx = _mm256_mul_pd(fscal,dx00);
707 ty = _mm256_mul_pd(fscal,dy00);
708 tz = _mm256_mul_pd(fscal,dz00);
709
710 /* Update vectorial force */
711 fix0 = _mm256_add_pd(fix0,tx);
712 fiy0 = _mm256_add_pd(fiy0,ty);
713 fiz0 = _mm256_add_pd(fiz0,tz);
714
715 fjptrA = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
716 fjptrB = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
717 fjptrC = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
718 fjptrD = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
719 gmx_mm256_decrement_1rvec_4ptr_swizzle_pd(fjptrA,fjptrB,fjptrC,fjptrD,tx,ty,tz);
720
721 }
722
723 /* Inner loop uses 37 flops */
724 }
725
726 /* End of innermost loop */
727
728 gmx_mm256_update_iforce_1atom_swizzle_pd(fix0,fiy0,fiz0,
729 f+i_coord_offset,fshift+i_shift_offset);
730
731 /* Increment number of inner iterations */
732 inneriter += j_index_end - j_index_start;
733
734 /* Outer loop uses 7 flops */
735 }
736
737 /* Increment number of outer iterations */
738 outeriter += nri;
739
740 /* Update outer/inner flops */
741
742 inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_F,outeriter*7 + inneriter*37);
743 }
The ultimate molecular dynamics simulation package