2 * This file is part of the GROMACS molecular simulation package.
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.
9 * GROMACS is free software; you can redistribute it and/or
10 * modify it under the terms of the GNU Lesser General Public License
11 * as published by the Free Software Foundation; either version 2.1
12 * of the License, or (at your option) any later version.
14 * GROMACS is distributed in the hope that it will be useful,
15 * but WITHOUT ANY WARRANTY; without even the implied warranty of
16 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
17 * Lesser General Public License for more details.
19 * You should have received a copy of the GNU Lesser General Public
20 * License along with GROMACS; if not, see
21 * http://www.gnu.org/licenses, or write to the Free Software Foundation,
22 * Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA.
24 * If you want to redistribute modifications to GROMACS, please
25 * consider that scientific software is very special. Version
26 * control is crucial - bugs must be traceable. We will be happy to
27 * consider code for inclusion in the official distribution, but
28 * derived work must not be called official GROMACS. Details are found
29 * in the README & COPYING files - if they are missing, get the
30 * official version at http://www.gromacs.org.
32 * To help us fund GROMACS development, we humbly ask that you cite
33 * the research papers on the package. Check out http://www.gromacs.org.
36 * Note: this file was generated by the GROMACS avx_256_double kernel generator.
44 #include "../nb_kernel.h"
45 #include "gromacs/gmxlib/nrnb.h"
47 #include "kernelutil_x86_avx_256_double.h"
50 * Gromacs nonbonded kernel: nb_kernel_ElecRF_VdwCSTab_GeomW3P1_VF_avx_256_double
51 * Electrostatics interaction: ReactionField
52 * VdW interaction: CubicSplineTable
53 * Geometry: Water3-Particle
54 * Calculate force/pot: PotentialAndForce
57 nb_kernel_ElecRF_VdwCSTab_GeomW3P1_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)
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.
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;
79 real *shiftvec,*fshift,*x,*f;
80 real *fjptrA,*fjptrB,*fjptrC,*fjptrD;
82 __m256d tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
83 real * vdwioffsetptr0;
84 __m256d ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
85 real * vdwioffsetptr1;
86 __m256d ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
87 real * vdwioffsetptr2;
88 __m256d ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
89 int vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
90 __m256d jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
91 __m256d dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
92 __m256d dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
93 __m256d dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
94 __m256d velec,felec,velecsum,facel,crf,krf,krf2;
97 __m256d rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
100 __m256d one_sixth = _mm256_set1_pd(1.0/6.0);
101 __m256d one_twelfth = _mm256_set1_pd(1.0/12.0);
103 __m128i ifour = _mm_set1_epi32(4);
104 __m256d rt,vfeps,vftabscale,Y,F,G,H,Heps,Fp,VV,FF;
106 __m256d dummy_mask,cutoff_mask;
107 __m128 tmpmask0,tmpmask1;
108 __m256d signbit = _mm256_castsi256_pd( _mm256_set1_epi32(0x80000000) );
109 __m256d one = _mm256_set1_pd(1.0);
110 __m256d two = _mm256_set1_pd(2.0);
116 jindex = nlist->jindex;
118 shiftidx = nlist->shift;
120 shiftvec = fr->shift_vec[0];
121 fshift = fr->fshift[0];
122 facel = _mm256_set1_pd(fr->ic->epsfac);
123 charge = mdatoms->chargeA;
124 krf = _mm256_set1_pd(fr->ic->k_rf);
125 krf2 = _mm256_set1_pd(fr->ic->k_rf*2.0);
126 crf = _mm256_set1_pd(fr->ic->c_rf);
127 nvdwtype = fr->ntype;
129 vdwtype = mdatoms->typeA;
131 vftab = kernel_data->table_vdw->data;
132 vftabscale = _mm256_set1_pd(kernel_data->table_vdw->scale);
134 /* Setup water-specific parameters */
135 inr = nlist->iinr[0];
136 iq0 = _mm256_mul_pd(facel,_mm256_set1_pd(charge[inr+0]));
137 iq1 = _mm256_mul_pd(facel,_mm256_set1_pd(charge[inr+1]));
138 iq2 = _mm256_mul_pd(facel,_mm256_set1_pd(charge[inr+2]));
139 vdwioffsetptr0 = vdwparam+2*nvdwtype*vdwtype[inr+0];
141 /* Avoid stupid compiler warnings */
142 jnrA = jnrB = jnrC = jnrD = 0;
151 for(iidx=0;iidx<4*DIM;iidx++)
156 /* Start outer loop over neighborlists */
157 for(iidx=0; iidx<nri; iidx++)
159 /* Load shift vector for this list */
160 i_shift_offset = DIM*shiftidx[iidx];
162 /* Load limits for loop over neighbors */
163 j_index_start = jindex[iidx];
164 j_index_end = jindex[iidx+1];
166 /* Get outer coordinate index */
168 i_coord_offset = DIM*inr;
170 /* Load i particle coords and add shift vector */
171 gmx_mm256_load_shift_and_3rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset,
172 &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2);
174 fix0 = _mm256_setzero_pd();
175 fiy0 = _mm256_setzero_pd();
176 fiz0 = _mm256_setzero_pd();
177 fix1 = _mm256_setzero_pd();
178 fiy1 = _mm256_setzero_pd();
179 fiz1 = _mm256_setzero_pd();
180 fix2 = _mm256_setzero_pd();
181 fiy2 = _mm256_setzero_pd();
182 fiz2 = _mm256_setzero_pd();
184 /* Reset potential sums */
185 velecsum = _mm256_setzero_pd();
186 vvdwsum = _mm256_setzero_pd();
188 /* Start inner kernel loop */
189 for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
192 /* Get j neighbor index, and coordinate index */
197 j_coord_offsetA = DIM*jnrA;
198 j_coord_offsetB = DIM*jnrB;
199 j_coord_offsetC = DIM*jnrC;
200 j_coord_offsetD = DIM*jnrD;
202 /* load j atom coordinates */
203 gmx_mm256_load_1rvec_4ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
204 x+j_coord_offsetC,x+j_coord_offsetD,
207 /* Calculate displacement vector */
208 dx00 = _mm256_sub_pd(ix0,jx0);
209 dy00 = _mm256_sub_pd(iy0,jy0);
210 dz00 = _mm256_sub_pd(iz0,jz0);
211 dx10 = _mm256_sub_pd(ix1,jx0);
212 dy10 = _mm256_sub_pd(iy1,jy0);
213 dz10 = _mm256_sub_pd(iz1,jz0);
214 dx20 = _mm256_sub_pd(ix2,jx0);
215 dy20 = _mm256_sub_pd(iy2,jy0);
216 dz20 = _mm256_sub_pd(iz2,jz0);
218 /* Calculate squared distance and things based on it */
219 rsq00 = gmx_mm256_calc_rsq_pd(dx00,dy00,dz00);
220 rsq10 = gmx_mm256_calc_rsq_pd(dx10,dy10,dz10);
221 rsq20 = gmx_mm256_calc_rsq_pd(dx20,dy20,dz20);
223 rinv00 = avx256_invsqrt_d(rsq00);
224 rinv10 = avx256_invsqrt_d(rsq10);
225 rinv20 = avx256_invsqrt_d(rsq20);
227 rinvsq00 = _mm256_mul_pd(rinv00,rinv00);
228 rinvsq10 = _mm256_mul_pd(rinv10,rinv10);
229 rinvsq20 = _mm256_mul_pd(rinv20,rinv20);
231 /* Load parameters for j particles */
232 jq0 = gmx_mm256_load_4real_swizzle_pd(charge+jnrA+0,charge+jnrB+0,
233 charge+jnrC+0,charge+jnrD+0);
234 vdwjidx0A = 2*vdwtype[jnrA+0];
235 vdwjidx0B = 2*vdwtype[jnrB+0];
236 vdwjidx0C = 2*vdwtype[jnrC+0];
237 vdwjidx0D = 2*vdwtype[jnrD+0];
239 fjx0 = _mm256_setzero_pd();
240 fjy0 = _mm256_setzero_pd();
241 fjz0 = _mm256_setzero_pd();
243 /**************************
244 * CALCULATE INTERACTIONS *
245 **************************/
247 r00 = _mm256_mul_pd(rsq00,rinv00);
249 /* Compute parameters for interactions between i and j atoms */
250 qq00 = _mm256_mul_pd(iq0,jq0);
251 gmx_mm256_load_4pair_swizzle_pd(vdwioffsetptr0+vdwjidx0A,
252 vdwioffsetptr0+vdwjidx0B,
253 vdwioffsetptr0+vdwjidx0C,
254 vdwioffsetptr0+vdwjidx0D,
257 /* Calculate table index by multiplying r with table scale and truncate to integer */
258 rt = _mm256_mul_pd(r00,vftabscale);
259 vfitab = _mm256_cvttpd_epi32(rt);
260 vfeps = _mm256_sub_pd(rt,_mm256_round_pd(rt, _MM_FROUND_FLOOR));
261 vfitab = _mm_slli_epi32(vfitab,3);
263 /* REACTION-FIELD ELECTROSTATICS */
264 velec = _mm256_mul_pd(qq00,_mm256_sub_pd(_mm256_add_pd(rinv00,_mm256_mul_pd(krf,rsq00)),crf));
265 felec = _mm256_mul_pd(qq00,_mm256_sub_pd(_mm256_mul_pd(rinv00,rinvsq00),krf2));
267 /* CUBIC SPLINE TABLE DISPERSION */
268 Y = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
269 F = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
270 G = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,2) );
271 H = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,3) );
272 GMX_MM256_FULLTRANSPOSE4_PD(Y,F,G,H);
273 Heps = _mm256_mul_pd(vfeps,H);
274 Fp = _mm256_add_pd(F,_mm256_mul_pd(vfeps,_mm256_add_pd(G,Heps)));
275 VV = _mm256_add_pd(Y,_mm256_mul_pd(vfeps,Fp));
276 vvdw6 = _mm256_mul_pd(c6_00,VV);
277 FF = _mm256_add_pd(Fp,_mm256_mul_pd(vfeps,_mm256_add_pd(G,_mm256_add_pd(Heps,Heps))));
278 fvdw6 = _mm256_mul_pd(c6_00,FF);
280 /* CUBIC SPLINE TABLE REPULSION */
281 vfitab = _mm_add_epi32(vfitab,ifour);
282 Y = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
283 F = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
284 G = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,2) );
285 H = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,3) );
286 GMX_MM256_FULLTRANSPOSE4_PD(Y,F,G,H);
287 Heps = _mm256_mul_pd(vfeps,H);
288 Fp = _mm256_add_pd(F,_mm256_mul_pd(vfeps,_mm256_add_pd(G,Heps)));
289 VV = _mm256_add_pd(Y,_mm256_mul_pd(vfeps,Fp));
290 vvdw12 = _mm256_mul_pd(c12_00,VV);
291 FF = _mm256_add_pd(Fp,_mm256_mul_pd(vfeps,_mm256_add_pd(G,_mm256_add_pd(Heps,Heps))));
292 fvdw12 = _mm256_mul_pd(c12_00,FF);
293 vvdw = _mm256_add_pd(vvdw12,vvdw6);
294 fvdw = _mm256_xor_pd(signbit,_mm256_mul_pd(_mm256_add_pd(fvdw6,fvdw12),_mm256_mul_pd(vftabscale,rinv00)));
296 /* Update potential sum for this i atom from the interaction with this j atom. */
297 velecsum = _mm256_add_pd(velecsum,velec);
298 vvdwsum = _mm256_add_pd(vvdwsum,vvdw);
300 fscal = _mm256_add_pd(felec,fvdw);
302 /* Calculate temporary vectorial force */
303 tx = _mm256_mul_pd(fscal,dx00);
304 ty = _mm256_mul_pd(fscal,dy00);
305 tz = _mm256_mul_pd(fscal,dz00);
307 /* Update vectorial force */
308 fix0 = _mm256_add_pd(fix0,tx);
309 fiy0 = _mm256_add_pd(fiy0,ty);
310 fiz0 = _mm256_add_pd(fiz0,tz);
312 fjx0 = _mm256_add_pd(fjx0,tx);
313 fjy0 = _mm256_add_pd(fjy0,ty);
314 fjz0 = _mm256_add_pd(fjz0,tz);
316 /**************************
317 * CALCULATE INTERACTIONS *
318 **************************/
320 /* Compute parameters for interactions between i and j atoms */
321 qq10 = _mm256_mul_pd(iq1,jq0);
323 /* REACTION-FIELD ELECTROSTATICS */
324 velec = _mm256_mul_pd(qq10,_mm256_sub_pd(_mm256_add_pd(rinv10,_mm256_mul_pd(krf,rsq10)),crf));
325 felec = _mm256_mul_pd(qq10,_mm256_sub_pd(_mm256_mul_pd(rinv10,rinvsq10),krf2));
327 /* Update potential sum for this i atom from the interaction with this j atom. */
328 velecsum = _mm256_add_pd(velecsum,velec);
332 /* Calculate temporary vectorial force */
333 tx = _mm256_mul_pd(fscal,dx10);
334 ty = _mm256_mul_pd(fscal,dy10);
335 tz = _mm256_mul_pd(fscal,dz10);
337 /* Update vectorial force */
338 fix1 = _mm256_add_pd(fix1,tx);
339 fiy1 = _mm256_add_pd(fiy1,ty);
340 fiz1 = _mm256_add_pd(fiz1,tz);
342 fjx0 = _mm256_add_pd(fjx0,tx);
343 fjy0 = _mm256_add_pd(fjy0,ty);
344 fjz0 = _mm256_add_pd(fjz0,tz);
346 /**************************
347 * CALCULATE INTERACTIONS *
348 **************************/
350 /* Compute parameters for interactions between i and j atoms */
351 qq20 = _mm256_mul_pd(iq2,jq0);
353 /* REACTION-FIELD ELECTROSTATICS */
354 velec = _mm256_mul_pd(qq20,_mm256_sub_pd(_mm256_add_pd(rinv20,_mm256_mul_pd(krf,rsq20)),crf));
355 felec = _mm256_mul_pd(qq20,_mm256_sub_pd(_mm256_mul_pd(rinv20,rinvsq20),krf2));
357 /* Update potential sum for this i atom from the interaction with this j atom. */
358 velecsum = _mm256_add_pd(velecsum,velec);
362 /* Calculate temporary vectorial force */
363 tx = _mm256_mul_pd(fscal,dx20);
364 ty = _mm256_mul_pd(fscal,dy20);
365 tz = _mm256_mul_pd(fscal,dz20);
367 /* Update vectorial force */
368 fix2 = _mm256_add_pd(fix2,tx);
369 fiy2 = _mm256_add_pd(fiy2,ty);
370 fiz2 = _mm256_add_pd(fiz2,tz);
372 fjx0 = _mm256_add_pd(fjx0,tx);
373 fjy0 = _mm256_add_pd(fjy0,ty);
374 fjz0 = _mm256_add_pd(fjz0,tz);
376 fjptrA = f+j_coord_offsetA;
377 fjptrB = f+j_coord_offsetB;
378 fjptrC = f+j_coord_offsetC;
379 fjptrD = f+j_coord_offsetD;
381 gmx_mm256_decrement_1rvec_4ptr_swizzle_pd(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
383 /* Inner loop uses 134 flops */
389 /* Get j neighbor index, and coordinate index */
390 jnrlistA = jjnr[jidx];
391 jnrlistB = jjnr[jidx+1];
392 jnrlistC = jjnr[jidx+2];
393 jnrlistD = jjnr[jidx+3];
394 /* Sign of each element will be negative for non-real atoms.
395 * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
396 * so use it as val = _mm_andnot_pd(mask,val) to clear dummy entries.
398 tmpmask0 = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
400 tmpmask1 = _mm_permute_ps(tmpmask0,_GMX_MM_PERMUTE(3,3,2,2));
401 tmpmask0 = _mm_permute_ps(tmpmask0,_GMX_MM_PERMUTE(1,1,0,0));
402 dummy_mask = _mm256_castps_pd(gmx_mm256_set_m128(tmpmask1,tmpmask0));
404 jnrA = (jnrlistA>=0) ? jnrlistA : 0;
405 jnrB = (jnrlistB>=0) ? jnrlistB : 0;
406 jnrC = (jnrlistC>=0) ? jnrlistC : 0;
407 jnrD = (jnrlistD>=0) ? jnrlistD : 0;
408 j_coord_offsetA = DIM*jnrA;
409 j_coord_offsetB = DIM*jnrB;
410 j_coord_offsetC = DIM*jnrC;
411 j_coord_offsetD = DIM*jnrD;
413 /* load j atom coordinates */
414 gmx_mm256_load_1rvec_4ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
415 x+j_coord_offsetC,x+j_coord_offsetD,
418 /* Calculate displacement vector */
419 dx00 = _mm256_sub_pd(ix0,jx0);
420 dy00 = _mm256_sub_pd(iy0,jy0);
421 dz00 = _mm256_sub_pd(iz0,jz0);
422 dx10 = _mm256_sub_pd(ix1,jx0);
423 dy10 = _mm256_sub_pd(iy1,jy0);
424 dz10 = _mm256_sub_pd(iz1,jz0);
425 dx20 = _mm256_sub_pd(ix2,jx0);
426 dy20 = _mm256_sub_pd(iy2,jy0);
427 dz20 = _mm256_sub_pd(iz2,jz0);
429 /* Calculate squared distance and things based on it */
430 rsq00 = gmx_mm256_calc_rsq_pd(dx00,dy00,dz00);
431 rsq10 = gmx_mm256_calc_rsq_pd(dx10,dy10,dz10);
432 rsq20 = gmx_mm256_calc_rsq_pd(dx20,dy20,dz20);
434 rinv00 = avx256_invsqrt_d(rsq00);
435 rinv10 = avx256_invsqrt_d(rsq10);
436 rinv20 = avx256_invsqrt_d(rsq20);
438 rinvsq00 = _mm256_mul_pd(rinv00,rinv00);
439 rinvsq10 = _mm256_mul_pd(rinv10,rinv10);
440 rinvsq20 = _mm256_mul_pd(rinv20,rinv20);
442 /* Load parameters for j particles */
443 jq0 = gmx_mm256_load_4real_swizzle_pd(charge+jnrA+0,charge+jnrB+0,
444 charge+jnrC+0,charge+jnrD+0);
445 vdwjidx0A = 2*vdwtype[jnrA+0];
446 vdwjidx0B = 2*vdwtype[jnrB+0];
447 vdwjidx0C = 2*vdwtype[jnrC+0];
448 vdwjidx0D = 2*vdwtype[jnrD+0];
450 fjx0 = _mm256_setzero_pd();
451 fjy0 = _mm256_setzero_pd();
452 fjz0 = _mm256_setzero_pd();
454 /**************************
455 * CALCULATE INTERACTIONS *
456 **************************/
458 r00 = _mm256_mul_pd(rsq00,rinv00);
459 r00 = _mm256_andnot_pd(dummy_mask,r00);
461 /* Compute parameters for interactions between i and j atoms */
462 qq00 = _mm256_mul_pd(iq0,jq0);
463 gmx_mm256_load_4pair_swizzle_pd(vdwioffsetptr0+vdwjidx0A,
464 vdwioffsetptr0+vdwjidx0B,
465 vdwioffsetptr0+vdwjidx0C,
466 vdwioffsetptr0+vdwjidx0D,
469 /* Calculate table index by multiplying r with table scale and truncate to integer */
470 rt = _mm256_mul_pd(r00,vftabscale);
471 vfitab = _mm256_cvttpd_epi32(rt);
472 vfeps = _mm256_sub_pd(rt,_mm256_round_pd(rt, _MM_FROUND_FLOOR));
473 vfitab = _mm_slli_epi32(vfitab,3);
475 /* REACTION-FIELD ELECTROSTATICS */
476 velec = _mm256_mul_pd(qq00,_mm256_sub_pd(_mm256_add_pd(rinv00,_mm256_mul_pd(krf,rsq00)),crf));
477 felec = _mm256_mul_pd(qq00,_mm256_sub_pd(_mm256_mul_pd(rinv00,rinvsq00),krf2));
479 /* CUBIC SPLINE TABLE DISPERSION */
480 Y = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
481 F = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
482 G = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,2) );
483 H = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,3) );
484 GMX_MM256_FULLTRANSPOSE4_PD(Y,F,G,H);
485 Heps = _mm256_mul_pd(vfeps,H);
486 Fp = _mm256_add_pd(F,_mm256_mul_pd(vfeps,_mm256_add_pd(G,Heps)));
487 VV = _mm256_add_pd(Y,_mm256_mul_pd(vfeps,Fp));
488 vvdw6 = _mm256_mul_pd(c6_00,VV);
489 FF = _mm256_add_pd(Fp,_mm256_mul_pd(vfeps,_mm256_add_pd(G,_mm256_add_pd(Heps,Heps))));
490 fvdw6 = _mm256_mul_pd(c6_00,FF);
492 /* CUBIC SPLINE TABLE REPULSION */
493 vfitab = _mm_add_epi32(vfitab,ifour);
494 Y = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
495 F = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
496 G = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,2) );
497 H = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,3) );
498 GMX_MM256_FULLTRANSPOSE4_PD(Y,F,G,H);
499 Heps = _mm256_mul_pd(vfeps,H);
500 Fp = _mm256_add_pd(F,_mm256_mul_pd(vfeps,_mm256_add_pd(G,Heps)));
501 VV = _mm256_add_pd(Y,_mm256_mul_pd(vfeps,Fp));
502 vvdw12 = _mm256_mul_pd(c12_00,VV);
503 FF = _mm256_add_pd(Fp,_mm256_mul_pd(vfeps,_mm256_add_pd(G,_mm256_add_pd(Heps,Heps))));
504 fvdw12 = _mm256_mul_pd(c12_00,FF);
505 vvdw = _mm256_add_pd(vvdw12,vvdw6);
506 fvdw = _mm256_xor_pd(signbit,_mm256_mul_pd(_mm256_add_pd(fvdw6,fvdw12),_mm256_mul_pd(vftabscale,rinv00)));
508 /* Update potential sum for this i atom from the interaction with this j atom. */
509 velec = _mm256_andnot_pd(dummy_mask,velec);
510 velecsum = _mm256_add_pd(velecsum,velec);
511 vvdw = _mm256_andnot_pd(dummy_mask,vvdw);
512 vvdwsum = _mm256_add_pd(vvdwsum,vvdw);
514 fscal = _mm256_add_pd(felec,fvdw);
516 fscal = _mm256_andnot_pd(dummy_mask,fscal);
518 /* Calculate temporary vectorial force */
519 tx = _mm256_mul_pd(fscal,dx00);
520 ty = _mm256_mul_pd(fscal,dy00);
521 tz = _mm256_mul_pd(fscal,dz00);
523 /* Update vectorial force */
524 fix0 = _mm256_add_pd(fix0,tx);
525 fiy0 = _mm256_add_pd(fiy0,ty);
526 fiz0 = _mm256_add_pd(fiz0,tz);
528 fjx0 = _mm256_add_pd(fjx0,tx);
529 fjy0 = _mm256_add_pd(fjy0,ty);
530 fjz0 = _mm256_add_pd(fjz0,tz);
532 /**************************
533 * CALCULATE INTERACTIONS *
534 **************************/
536 /* Compute parameters for interactions between i and j atoms */
537 qq10 = _mm256_mul_pd(iq1,jq0);
539 /* REACTION-FIELD ELECTROSTATICS */
540 velec = _mm256_mul_pd(qq10,_mm256_sub_pd(_mm256_add_pd(rinv10,_mm256_mul_pd(krf,rsq10)),crf));
541 felec = _mm256_mul_pd(qq10,_mm256_sub_pd(_mm256_mul_pd(rinv10,rinvsq10),krf2));
543 /* Update potential sum for this i atom from the interaction with this j atom. */
544 velec = _mm256_andnot_pd(dummy_mask,velec);
545 velecsum = _mm256_add_pd(velecsum,velec);
549 fscal = _mm256_andnot_pd(dummy_mask,fscal);
551 /* Calculate temporary vectorial force */
552 tx = _mm256_mul_pd(fscal,dx10);
553 ty = _mm256_mul_pd(fscal,dy10);
554 tz = _mm256_mul_pd(fscal,dz10);
556 /* Update vectorial force */
557 fix1 = _mm256_add_pd(fix1,tx);
558 fiy1 = _mm256_add_pd(fiy1,ty);
559 fiz1 = _mm256_add_pd(fiz1,tz);
561 fjx0 = _mm256_add_pd(fjx0,tx);
562 fjy0 = _mm256_add_pd(fjy0,ty);
563 fjz0 = _mm256_add_pd(fjz0,tz);
565 /**************************
566 * CALCULATE INTERACTIONS *
567 **************************/
569 /* Compute parameters for interactions between i and j atoms */
570 qq20 = _mm256_mul_pd(iq2,jq0);
572 /* REACTION-FIELD ELECTROSTATICS */
573 velec = _mm256_mul_pd(qq20,_mm256_sub_pd(_mm256_add_pd(rinv20,_mm256_mul_pd(krf,rsq20)),crf));
574 felec = _mm256_mul_pd(qq20,_mm256_sub_pd(_mm256_mul_pd(rinv20,rinvsq20),krf2));
576 /* Update potential sum for this i atom from the interaction with this j atom. */
577 velec = _mm256_andnot_pd(dummy_mask,velec);
578 velecsum = _mm256_add_pd(velecsum,velec);
582 fscal = _mm256_andnot_pd(dummy_mask,fscal);
584 /* Calculate temporary vectorial force */
585 tx = _mm256_mul_pd(fscal,dx20);
586 ty = _mm256_mul_pd(fscal,dy20);
587 tz = _mm256_mul_pd(fscal,dz20);
589 /* Update vectorial force */
590 fix2 = _mm256_add_pd(fix2,tx);
591 fiy2 = _mm256_add_pd(fiy2,ty);
592 fiz2 = _mm256_add_pd(fiz2,tz);
594 fjx0 = _mm256_add_pd(fjx0,tx);
595 fjy0 = _mm256_add_pd(fjy0,ty);
596 fjz0 = _mm256_add_pd(fjz0,tz);
598 fjptrA = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
599 fjptrB = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
600 fjptrC = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
601 fjptrD = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
603 gmx_mm256_decrement_1rvec_4ptr_swizzle_pd(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
605 /* Inner loop uses 135 flops */
608 /* End of innermost loop */
610 gmx_mm256_update_iforce_3atom_swizzle_pd(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,
611 f+i_coord_offset,fshift+i_shift_offset);
614 /* Update potential energies */
615 gmx_mm256_update_1pot_pd(velecsum,kernel_data->energygrp_elec+ggid);
616 gmx_mm256_update_1pot_pd(vvdwsum,kernel_data->energygrp_vdw+ggid);
618 /* Increment number of inner iterations */
619 inneriter += j_index_end - j_index_start;
621 /* Outer loop uses 20 flops */
624 /* Increment number of outer iterations */
627 /* Update outer/inner flops */
629 inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W3_VF,outeriter*20 + inneriter*135);
632 * Gromacs nonbonded kernel: nb_kernel_ElecRF_VdwCSTab_GeomW3P1_F_avx_256_double
633 * Electrostatics interaction: ReactionField
634 * VdW interaction: CubicSplineTable
635 * Geometry: Water3-Particle
636 * Calculate force/pot: Force
639 nb_kernel_ElecRF_VdwCSTab_GeomW3P1_F_avx_256_double
640 (t_nblist * gmx_restrict nlist,
641 rvec * gmx_restrict xx,
642 rvec * gmx_restrict ff,
643 struct t_forcerec * gmx_restrict fr,
644 t_mdatoms * gmx_restrict mdatoms,
645 nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
646 t_nrnb * gmx_restrict nrnb)
648 /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
649 * just 0 for non-waters.
650 * Suffixes A,B,C,D refer to j loop unrolling done with AVX, e.g. for the four different
651 * jnr indices corresponding to data put in the four positions in the SIMD register.
653 int i_shift_offset,i_coord_offset,outeriter,inneriter;
654 int j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
655 int jnrA,jnrB,jnrC,jnrD;
656 int jnrlistA,jnrlistB,jnrlistC,jnrlistD;
657 int jnrlistE,jnrlistF,jnrlistG,jnrlistH;
658 int j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
659 int *iinr,*jindex,*jjnr,*shiftidx,*gid;
661 real *shiftvec,*fshift,*x,*f;
662 real *fjptrA,*fjptrB,*fjptrC,*fjptrD;
664 __m256d tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
665 real * vdwioffsetptr0;
666 __m256d ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
667 real * vdwioffsetptr1;
668 __m256d ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
669 real * vdwioffsetptr2;
670 __m256d ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
671 int vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
672 __m256d jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
673 __m256d dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
674 __m256d dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
675 __m256d dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
676 __m256d velec,felec,velecsum,facel,crf,krf,krf2;
679 __m256d rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
682 __m256d one_sixth = _mm256_set1_pd(1.0/6.0);
683 __m256d one_twelfth = _mm256_set1_pd(1.0/12.0);
685 __m128i ifour = _mm_set1_epi32(4);
686 __m256d rt,vfeps,vftabscale,Y,F,G,H,Heps,Fp,VV,FF;
688 __m256d dummy_mask,cutoff_mask;
689 __m128 tmpmask0,tmpmask1;
690 __m256d signbit = _mm256_castsi256_pd( _mm256_set1_epi32(0x80000000) );
691 __m256d one = _mm256_set1_pd(1.0);
692 __m256d two = _mm256_set1_pd(2.0);
698 jindex = nlist->jindex;
700 shiftidx = nlist->shift;
702 shiftvec = fr->shift_vec[0];
703 fshift = fr->fshift[0];
704 facel = _mm256_set1_pd(fr->ic->epsfac);
705 charge = mdatoms->chargeA;
706 krf = _mm256_set1_pd(fr->ic->k_rf);
707 krf2 = _mm256_set1_pd(fr->ic->k_rf*2.0);
708 crf = _mm256_set1_pd(fr->ic->c_rf);
709 nvdwtype = fr->ntype;
711 vdwtype = mdatoms->typeA;
713 vftab = kernel_data->table_vdw->data;
714 vftabscale = _mm256_set1_pd(kernel_data->table_vdw->scale);
716 /* Setup water-specific parameters */
717 inr = nlist->iinr[0];
718 iq0 = _mm256_mul_pd(facel,_mm256_set1_pd(charge[inr+0]));
719 iq1 = _mm256_mul_pd(facel,_mm256_set1_pd(charge[inr+1]));
720 iq2 = _mm256_mul_pd(facel,_mm256_set1_pd(charge[inr+2]));
721 vdwioffsetptr0 = vdwparam+2*nvdwtype*vdwtype[inr+0];
723 /* Avoid stupid compiler warnings */
724 jnrA = jnrB = jnrC = jnrD = 0;
733 for(iidx=0;iidx<4*DIM;iidx++)
738 /* Start outer loop over neighborlists */
739 for(iidx=0; iidx<nri; iidx++)
741 /* Load shift vector for this list */
742 i_shift_offset = DIM*shiftidx[iidx];
744 /* Load limits for loop over neighbors */
745 j_index_start = jindex[iidx];
746 j_index_end = jindex[iidx+1];
748 /* Get outer coordinate index */
750 i_coord_offset = DIM*inr;
752 /* Load i particle coords and add shift vector */
753 gmx_mm256_load_shift_and_3rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset,
754 &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2);
756 fix0 = _mm256_setzero_pd();
757 fiy0 = _mm256_setzero_pd();
758 fiz0 = _mm256_setzero_pd();
759 fix1 = _mm256_setzero_pd();
760 fiy1 = _mm256_setzero_pd();
761 fiz1 = _mm256_setzero_pd();
762 fix2 = _mm256_setzero_pd();
763 fiy2 = _mm256_setzero_pd();
764 fiz2 = _mm256_setzero_pd();
766 /* Start inner kernel loop */
767 for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
770 /* Get j neighbor index, and coordinate index */
775 j_coord_offsetA = DIM*jnrA;
776 j_coord_offsetB = DIM*jnrB;
777 j_coord_offsetC = DIM*jnrC;
778 j_coord_offsetD = DIM*jnrD;
780 /* load j atom coordinates */
781 gmx_mm256_load_1rvec_4ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
782 x+j_coord_offsetC,x+j_coord_offsetD,
785 /* Calculate displacement vector */
786 dx00 = _mm256_sub_pd(ix0,jx0);
787 dy00 = _mm256_sub_pd(iy0,jy0);
788 dz00 = _mm256_sub_pd(iz0,jz0);
789 dx10 = _mm256_sub_pd(ix1,jx0);
790 dy10 = _mm256_sub_pd(iy1,jy0);
791 dz10 = _mm256_sub_pd(iz1,jz0);
792 dx20 = _mm256_sub_pd(ix2,jx0);
793 dy20 = _mm256_sub_pd(iy2,jy0);
794 dz20 = _mm256_sub_pd(iz2,jz0);
796 /* Calculate squared distance and things based on it */
797 rsq00 = gmx_mm256_calc_rsq_pd(dx00,dy00,dz00);
798 rsq10 = gmx_mm256_calc_rsq_pd(dx10,dy10,dz10);
799 rsq20 = gmx_mm256_calc_rsq_pd(dx20,dy20,dz20);
801 rinv00 = avx256_invsqrt_d(rsq00);
802 rinv10 = avx256_invsqrt_d(rsq10);
803 rinv20 = avx256_invsqrt_d(rsq20);
805 rinvsq00 = _mm256_mul_pd(rinv00,rinv00);
806 rinvsq10 = _mm256_mul_pd(rinv10,rinv10);
807 rinvsq20 = _mm256_mul_pd(rinv20,rinv20);
809 /* Load parameters for j particles */
810 jq0 = gmx_mm256_load_4real_swizzle_pd(charge+jnrA+0,charge+jnrB+0,
811 charge+jnrC+0,charge+jnrD+0);
812 vdwjidx0A = 2*vdwtype[jnrA+0];
813 vdwjidx0B = 2*vdwtype[jnrB+0];
814 vdwjidx0C = 2*vdwtype[jnrC+0];
815 vdwjidx0D = 2*vdwtype[jnrD+0];
817 fjx0 = _mm256_setzero_pd();
818 fjy0 = _mm256_setzero_pd();
819 fjz0 = _mm256_setzero_pd();
821 /**************************
822 * CALCULATE INTERACTIONS *
823 **************************/
825 r00 = _mm256_mul_pd(rsq00,rinv00);
827 /* Compute parameters for interactions between i and j atoms */
828 qq00 = _mm256_mul_pd(iq0,jq0);
829 gmx_mm256_load_4pair_swizzle_pd(vdwioffsetptr0+vdwjidx0A,
830 vdwioffsetptr0+vdwjidx0B,
831 vdwioffsetptr0+vdwjidx0C,
832 vdwioffsetptr0+vdwjidx0D,
835 /* Calculate table index by multiplying r with table scale and truncate to integer */
836 rt = _mm256_mul_pd(r00,vftabscale);
837 vfitab = _mm256_cvttpd_epi32(rt);
838 vfeps = _mm256_sub_pd(rt,_mm256_round_pd(rt, _MM_FROUND_FLOOR));
839 vfitab = _mm_slli_epi32(vfitab,3);
841 /* REACTION-FIELD ELECTROSTATICS */
842 felec = _mm256_mul_pd(qq00,_mm256_sub_pd(_mm256_mul_pd(rinv00,rinvsq00),krf2));
844 /* CUBIC SPLINE TABLE DISPERSION */
845 Y = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
846 F = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
847 G = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,2) );
848 H = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,3) );
849 GMX_MM256_FULLTRANSPOSE4_PD(Y,F,G,H);
850 Heps = _mm256_mul_pd(vfeps,H);
851 Fp = _mm256_add_pd(F,_mm256_mul_pd(vfeps,_mm256_add_pd(G,Heps)));
852 FF = _mm256_add_pd(Fp,_mm256_mul_pd(vfeps,_mm256_add_pd(G,_mm256_add_pd(Heps,Heps))));
853 fvdw6 = _mm256_mul_pd(c6_00,FF);
855 /* CUBIC SPLINE TABLE REPULSION */
856 vfitab = _mm_add_epi32(vfitab,ifour);
857 Y = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
858 F = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
859 G = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,2) );
860 H = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,3) );
861 GMX_MM256_FULLTRANSPOSE4_PD(Y,F,G,H);
862 Heps = _mm256_mul_pd(vfeps,H);
863 Fp = _mm256_add_pd(F,_mm256_mul_pd(vfeps,_mm256_add_pd(G,Heps)));
864 FF = _mm256_add_pd(Fp,_mm256_mul_pd(vfeps,_mm256_add_pd(G,_mm256_add_pd(Heps,Heps))));
865 fvdw12 = _mm256_mul_pd(c12_00,FF);
866 fvdw = _mm256_xor_pd(signbit,_mm256_mul_pd(_mm256_add_pd(fvdw6,fvdw12),_mm256_mul_pd(vftabscale,rinv00)));
868 fscal = _mm256_add_pd(felec,fvdw);
870 /* Calculate temporary vectorial force */
871 tx = _mm256_mul_pd(fscal,dx00);
872 ty = _mm256_mul_pd(fscal,dy00);
873 tz = _mm256_mul_pd(fscal,dz00);
875 /* Update vectorial force */
876 fix0 = _mm256_add_pd(fix0,tx);
877 fiy0 = _mm256_add_pd(fiy0,ty);
878 fiz0 = _mm256_add_pd(fiz0,tz);
880 fjx0 = _mm256_add_pd(fjx0,tx);
881 fjy0 = _mm256_add_pd(fjy0,ty);
882 fjz0 = _mm256_add_pd(fjz0,tz);
884 /**************************
885 * CALCULATE INTERACTIONS *
886 **************************/
888 /* Compute parameters for interactions between i and j atoms */
889 qq10 = _mm256_mul_pd(iq1,jq0);
891 /* REACTION-FIELD ELECTROSTATICS */
892 felec = _mm256_mul_pd(qq10,_mm256_sub_pd(_mm256_mul_pd(rinv10,rinvsq10),krf2));
896 /* Calculate temporary vectorial force */
897 tx = _mm256_mul_pd(fscal,dx10);
898 ty = _mm256_mul_pd(fscal,dy10);
899 tz = _mm256_mul_pd(fscal,dz10);
901 /* Update vectorial force */
902 fix1 = _mm256_add_pd(fix1,tx);
903 fiy1 = _mm256_add_pd(fiy1,ty);
904 fiz1 = _mm256_add_pd(fiz1,tz);
906 fjx0 = _mm256_add_pd(fjx0,tx);
907 fjy0 = _mm256_add_pd(fjy0,ty);
908 fjz0 = _mm256_add_pd(fjz0,tz);
910 /**************************
911 * CALCULATE INTERACTIONS *
912 **************************/
914 /* Compute parameters for interactions between i and j atoms */
915 qq20 = _mm256_mul_pd(iq2,jq0);
917 /* REACTION-FIELD ELECTROSTATICS */
918 felec = _mm256_mul_pd(qq20,_mm256_sub_pd(_mm256_mul_pd(rinv20,rinvsq20),krf2));
922 /* Calculate temporary vectorial force */
923 tx = _mm256_mul_pd(fscal,dx20);
924 ty = _mm256_mul_pd(fscal,dy20);
925 tz = _mm256_mul_pd(fscal,dz20);
927 /* Update vectorial force */
928 fix2 = _mm256_add_pd(fix2,tx);
929 fiy2 = _mm256_add_pd(fiy2,ty);
930 fiz2 = _mm256_add_pd(fiz2,tz);
932 fjx0 = _mm256_add_pd(fjx0,tx);
933 fjy0 = _mm256_add_pd(fjy0,ty);
934 fjz0 = _mm256_add_pd(fjz0,tz);
936 fjptrA = f+j_coord_offsetA;
937 fjptrB = f+j_coord_offsetB;
938 fjptrC = f+j_coord_offsetC;
939 fjptrD = f+j_coord_offsetD;
941 gmx_mm256_decrement_1rvec_4ptr_swizzle_pd(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
943 /* Inner loop uses 111 flops */
949 /* Get j neighbor index, and coordinate index */
950 jnrlistA = jjnr[jidx];
951 jnrlistB = jjnr[jidx+1];
952 jnrlistC = jjnr[jidx+2];
953 jnrlistD = jjnr[jidx+3];
954 /* Sign of each element will be negative for non-real atoms.
955 * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
956 * so use it as val = _mm_andnot_pd(mask,val) to clear dummy entries.
958 tmpmask0 = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
960 tmpmask1 = _mm_permute_ps(tmpmask0,_GMX_MM_PERMUTE(3,3,2,2));
961 tmpmask0 = _mm_permute_ps(tmpmask0,_GMX_MM_PERMUTE(1,1,0,0));
962 dummy_mask = _mm256_castps_pd(gmx_mm256_set_m128(tmpmask1,tmpmask0));
964 jnrA = (jnrlistA>=0) ? jnrlistA : 0;
965 jnrB = (jnrlistB>=0) ? jnrlistB : 0;
966 jnrC = (jnrlistC>=0) ? jnrlistC : 0;
967 jnrD = (jnrlistD>=0) ? jnrlistD : 0;
968 j_coord_offsetA = DIM*jnrA;
969 j_coord_offsetB = DIM*jnrB;
970 j_coord_offsetC = DIM*jnrC;
971 j_coord_offsetD = DIM*jnrD;
973 /* load j atom coordinates */
974 gmx_mm256_load_1rvec_4ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
975 x+j_coord_offsetC,x+j_coord_offsetD,
978 /* Calculate displacement vector */
979 dx00 = _mm256_sub_pd(ix0,jx0);
980 dy00 = _mm256_sub_pd(iy0,jy0);
981 dz00 = _mm256_sub_pd(iz0,jz0);
982 dx10 = _mm256_sub_pd(ix1,jx0);
983 dy10 = _mm256_sub_pd(iy1,jy0);
984 dz10 = _mm256_sub_pd(iz1,jz0);
985 dx20 = _mm256_sub_pd(ix2,jx0);
986 dy20 = _mm256_sub_pd(iy2,jy0);
987 dz20 = _mm256_sub_pd(iz2,jz0);
989 /* Calculate squared distance and things based on it */
990 rsq00 = gmx_mm256_calc_rsq_pd(dx00,dy00,dz00);
991 rsq10 = gmx_mm256_calc_rsq_pd(dx10,dy10,dz10);
992 rsq20 = gmx_mm256_calc_rsq_pd(dx20,dy20,dz20);
994 rinv00 = avx256_invsqrt_d(rsq00);
995 rinv10 = avx256_invsqrt_d(rsq10);
996 rinv20 = avx256_invsqrt_d(rsq20);
998 rinvsq00 = _mm256_mul_pd(rinv00,rinv00);
999 rinvsq10 = _mm256_mul_pd(rinv10,rinv10);
1000 rinvsq20 = _mm256_mul_pd(rinv20,rinv20);
1002 /* Load parameters for j particles */
1003 jq0 = gmx_mm256_load_4real_swizzle_pd(charge+jnrA+0,charge+jnrB+0,
1004 charge+jnrC+0,charge+jnrD+0);
1005 vdwjidx0A = 2*vdwtype[jnrA+0];
1006 vdwjidx0B = 2*vdwtype[jnrB+0];
1007 vdwjidx0C = 2*vdwtype[jnrC+0];
1008 vdwjidx0D = 2*vdwtype[jnrD+0];
1010 fjx0 = _mm256_setzero_pd();
1011 fjy0 = _mm256_setzero_pd();
1012 fjz0 = _mm256_setzero_pd();
1014 /**************************
1015 * CALCULATE INTERACTIONS *
1016 **************************/
1018 r00 = _mm256_mul_pd(rsq00,rinv00);
1019 r00 = _mm256_andnot_pd(dummy_mask,r00);
1021 /* Compute parameters for interactions between i and j atoms */
1022 qq00 = _mm256_mul_pd(iq0,jq0);
1023 gmx_mm256_load_4pair_swizzle_pd(vdwioffsetptr0+vdwjidx0A,
1024 vdwioffsetptr0+vdwjidx0B,
1025 vdwioffsetptr0+vdwjidx0C,
1026 vdwioffsetptr0+vdwjidx0D,
1029 /* Calculate table index by multiplying r with table scale and truncate to integer */
1030 rt = _mm256_mul_pd(r00,vftabscale);
1031 vfitab = _mm256_cvttpd_epi32(rt);
1032 vfeps = _mm256_sub_pd(rt,_mm256_round_pd(rt, _MM_FROUND_FLOOR));
1033 vfitab = _mm_slli_epi32(vfitab,3);
1035 /* REACTION-FIELD ELECTROSTATICS */
1036 felec = _mm256_mul_pd(qq00,_mm256_sub_pd(_mm256_mul_pd(rinv00,rinvsq00),krf2));
1038 /* CUBIC SPLINE TABLE DISPERSION */
1039 Y = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
1040 F = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
1041 G = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,2) );
1042 H = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,3) );
1043 GMX_MM256_FULLTRANSPOSE4_PD(Y,F,G,H);
1044 Heps = _mm256_mul_pd(vfeps,H);
1045 Fp = _mm256_add_pd(F,_mm256_mul_pd(vfeps,_mm256_add_pd(G,Heps)));
1046 FF = _mm256_add_pd(Fp,_mm256_mul_pd(vfeps,_mm256_add_pd(G,_mm256_add_pd(Heps,Heps))));
1047 fvdw6 = _mm256_mul_pd(c6_00,FF);
1049 /* CUBIC SPLINE TABLE REPULSION */
1050 vfitab = _mm_add_epi32(vfitab,ifour);
1051 Y = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
1052 F = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
1053 G = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,2) );
1054 H = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,3) );
1055 GMX_MM256_FULLTRANSPOSE4_PD(Y,F,G,H);
1056 Heps = _mm256_mul_pd(vfeps,H);
1057 Fp = _mm256_add_pd(F,_mm256_mul_pd(vfeps,_mm256_add_pd(G,Heps)));
1058 FF = _mm256_add_pd(Fp,_mm256_mul_pd(vfeps,_mm256_add_pd(G,_mm256_add_pd(Heps,Heps))));
1059 fvdw12 = _mm256_mul_pd(c12_00,FF);
1060 fvdw = _mm256_xor_pd(signbit,_mm256_mul_pd(_mm256_add_pd(fvdw6,fvdw12),_mm256_mul_pd(vftabscale,rinv00)));
1062 fscal = _mm256_add_pd(felec,fvdw);
1064 fscal = _mm256_andnot_pd(dummy_mask,fscal);
1066 /* Calculate temporary vectorial force */
1067 tx = _mm256_mul_pd(fscal,dx00);
1068 ty = _mm256_mul_pd(fscal,dy00);
1069 tz = _mm256_mul_pd(fscal,dz00);
1071 /* Update vectorial force */
1072 fix0 = _mm256_add_pd(fix0,tx);
1073 fiy0 = _mm256_add_pd(fiy0,ty);
1074 fiz0 = _mm256_add_pd(fiz0,tz);
1076 fjx0 = _mm256_add_pd(fjx0,tx);
1077 fjy0 = _mm256_add_pd(fjy0,ty);
1078 fjz0 = _mm256_add_pd(fjz0,tz);
1080 /**************************
1081 * CALCULATE INTERACTIONS *
1082 **************************/
1084 /* Compute parameters for interactions between i and j atoms */
1085 qq10 = _mm256_mul_pd(iq1,jq0);
1087 /* REACTION-FIELD ELECTROSTATICS */
1088 felec = _mm256_mul_pd(qq10,_mm256_sub_pd(_mm256_mul_pd(rinv10,rinvsq10),krf2));
1092 fscal = _mm256_andnot_pd(dummy_mask,fscal);
1094 /* Calculate temporary vectorial force */
1095 tx = _mm256_mul_pd(fscal,dx10);
1096 ty = _mm256_mul_pd(fscal,dy10);
1097 tz = _mm256_mul_pd(fscal,dz10);
1099 /* Update vectorial force */
1100 fix1 = _mm256_add_pd(fix1,tx);
1101 fiy1 = _mm256_add_pd(fiy1,ty);
1102 fiz1 = _mm256_add_pd(fiz1,tz);
1104 fjx0 = _mm256_add_pd(fjx0,tx);
1105 fjy0 = _mm256_add_pd(fjy0,ty);
1106 fjz0 = _mm256_add_pd(fjz0,tz);
1108 /**************************
1109 * CALCULATE INTERACTIONS *
1110 **************************/
1112 /* Compute parameters for interactions between i and j atoms */
1113 qq20 = _mm256_mul_pd(iq2,jq0);
1115 /* REACTION-FIELD ELECTROSTATICS */
1116 felec = _mm256_mul_pd(qq20,_mm256_sub_pd(_mm256_mul_pd(rinv20,rinvsq20),krf2));
1120 fscal = _mm256_andnot_pd(dummy_mask,fscal);
1122 /* Calculate temporary vectorial force */
1123 tx = _mm256_mul_pd(fscal,dx20);
1124 ty = _mm256_mul_pd(fscal,dy20);
1125 tz = _mm256_mul_pd(fscal,dz20);
1127 /* Update vectorial force */
1128 fix2 = _mm256_add_pd(fix2,tx);
1129 fiy2 = _mm256_add_pd(fiy2,ty);
1130 fiz2 = _mm256_add_pd(fiz2,tz);
1132 fjx0 = _mm256_add_pd(fjx0,tx);
1133 fjy0 = _mm256_add_pd(fjy0,ty);
1134 fjz0 = _mm256_add_pd(fjz0,tz);
1136 fjptrA = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
1137 fjptrB = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
1138 fjptrC = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
1139 fjptrD = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
1141 gmx_mm256_decrement_1rvec_4ptr_swizzle_pd(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
1143 /* Inner loop uses 112 flops */
1146 /* End of innermost loop */
1148 gmx_mm256_update_iforce_3atom_swizzle_pd(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,
1149 f+i_coord_offset,fshift+i_shift_offset);
1151 /* Increment number of inner iterations */
1152 inneriter += j_index_end - j_index_start;
1154 /* Outer loop uses 18 flops */
1157 /* Increment number of outer iterations */
1160 /* Update outer/inner flops */
1162 inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W3_F,outeriter*18 + inneriter*112);