2 * This file is part of the GROMACS molecular simulation package.
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.
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_GeomW4P1_VF_avx_256_double
51 * Electrostatics interaction: ReactionField
52 * VdW interaction: CubicSplineTable
53 * Geometry: Water4-Particle
54 * Calculate force/pot: PotentialAndForce
57 nb_kernel_ElecRF_VdwCSTab_GeomW4P1_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 real * vdwioffsetptr3;
90 __m256d ix3,iy3,iz3,fix3,fiy3,fiz3,iq3,isai3;
91 int vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
92 __m256d jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
93 __m256d dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
94 __m256d dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
95 __m256d dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
96 __m256d dx30,dy30,dz30,rsq30,rinv30,rinvsq30,r30,qq30,c6_30,c12_30;
97 __m256d velec,felec,velecsum,facel,crf,krf,krf2;
100 __m256d rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
103 __m256d one_sixth = _mm256_set1_pd(1.0/6.0);
104 __m256d one_twelfth = _mm256_set1_pd(1.0/12.0);
106 __m128i ifour = _mm_set1_epi32(4);
107 __m256d rt,vfeps,vftabscale,Y,F,G,H,Heps,Fp,VV,FF;
109 __m256d dummy_mask,cutoff_mask;
110 __m128 tmpmask0,tmpmask1;
111 __m256d signbit = _mm256_castsi256_pd( _mm256_set1_epi32(0x80000000) );
112 __m256d one = _mm256_set1_pd(1.0);
113 __m256d two = _mm256_set1_pd(2.0);
119 jindex = nlist->jindex;
121 shiftidx = nlist->shift;
123 shiftvec = fr->shift_vec[0];
124 fshift = fr->fshift[0];
125 facel = _mm256_set1_pd(fr->ic->epsfac);
126 charge = mdatoms->chargeA;
127 krf = _mm256_set1_pd(fr->ic->k_rf);
128 krf2 = _mm256_set1_pd(fr->ic->k_rf*2.0);
129 crf = _mm256_set1_pd(fr->ic->c_rf);
130 nvdwtype = fr->ntype;
132 vdwtype = mdatoms->typeA;
134 vftab = kernel_data->table_vdw->data;
135 vftabscale = _mm256_set1_pd(kernel_data->table_vdw->scale);
137 /* Setup water-specific parameters */
138 inr = nlist->iinr[0];
139 iq1 = _mm256_mul_pd(facel,_mm256_set1_pd(charge[inr+1]));
140 iq2 = _mm256_mul_pd(facel,_mm256_set1_pd(charge[inr+2]));
141 iq3 = _mm256_mul_pd(facel,_mm256_set1_pd(charge[inr+3]));
142 vdwioffsetptr0 = vdwparam+2*nvdwtype*vdwtype[inr+0];
144 /* Avoid stupid compiler warnings */
145 jnrA = jnrB = jnrC = jnrD = 0;
154 for(iidx=0;iidx<4*DIM;iidx++)
159 /* Start outer loop over neighborlists */
160 for(iidx=0; iidx<nri; iidx++)
162 /* Load shift vector for this list */
163 i_shift_offset = DIM*shiftidx[iidx];
165 /* Load limits for loop over neighbors */
166 j_index_start = jindex[iidx];
167 j_index_end = jindex[iidx+1];
169 /* Get outer coordinate index */
171 i_coord_offset = DIM*inr;
173 /* Load i particle coords and add shift vector */
174 gmx_mm256_load_shift_and_4rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset,
175 &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2,&ix3,&iy3,&iz3);
177 fix0 = _mm256_setzero_pd();
178 fiy0 = _mm256_setzero_pd();
179 fiz0 = _mm256_setzero_pd();
180 fix1 = _mm256_setzero_pd();
181 fiy1 = _mm256_setzero_pd();
182 fiz1 = _mm256_setzero_pd();
183 fix2 = _mm256_setzero_pd();
184 fiy2 = _mm256_setzero_pd();
185 fiz2 = _mm256_setzero_pd();
186 fix3 = _mm256_setzero_pd();
187 fiy3 = _mm256_setzero_pd();
188 fiz3 = _mm256_setzero_pd();
190 /* Reset potential sums */
191 velecsum = _mm256_setzero_pd();
192 vvdwsum = _mm256_setzero_pd();
194 /* Start inner kernel loop */
195 for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
198 /* Get j neighbor index, and coordinate index */
203 j_coord_offsetA = DIM*jnrA;
204 j_coord_offsetB = DIM*jnrB;
205 j_coord_offsetC = DIM*jnrC;
206 j_coord_offsetD = DIM*jnrD;
208 /* load j atom coordinates */
209 gmx_mm256_load_1rvec_4ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
210 x+j_coord_offsetC,x+j_coord_offsetD,
213 /* Calculate displacement vector */
214 dx00 = _mm256_sub_pd(ix0,jx0);
215 dy00 = _mm256_sub_pd(iy0,jy0);
216 dz00 = _mm256_sub_pd(iz0,jz0);
217 dx10 = _mm256_sub_pd(ix1,jx0);
218 dy10 = _mm256_sub_pd(iy1,jy0);
219 dz10 = _mm256_sub_pd(iz1,jz0);
220 dx20 = _mm256_sub_pd(ix2,jx0);
221 dy20 = _mm256_sub_pd(iy2,jy0);
222 dz20 = _mm256_sub_pd(iz2,jz0);
223 dx30 = _mm256_sub_pd(ix3,jx0);
224 dy30 = _mm256_sub_pd(iy3,jy0);
225 dz30 = _mm256_sub_pd(iz3,jz0);
227 /* Calculate squared distance and things based on it */
228 rsq00 = gmx_mm256_calc_rsq_pd(dx00,dy00,dz00);
229 rsq10 = gmx_mm256_calc_rsq_pd(dx10,dy10,dz10);
230 rsq20 = gmx_mm256_calc_rsq_pd(dx20,dy20,dz20);
231 rsq30 = gmx_mm256_calc_rsq_pd(dx30,dy30,dz30);
233 rinv00 = avx256_invsqrt_d(rsq00);
234 rinv10 = avx256_invsqrt_d(rsq10);
235 rinv20 = avx256_invsqrt_d(rsq20);
236 rinv30 = avx256_invsqrt_d(rsq30);
238 rinvsq10 = _mm256_mul_pd(rinv10,rinv10);
239 rinvsq20 = _mm256_mul_pd(rinv20,rinv20);
240 rinvsq30 = _mm256_mul_pd(rinv30,rinv30);
242 /* Load parameters for j particles */
243 jq0 = gmx_mm256_load_4real_swizzle_pd(charge+jnrA+0,charge+jnrB+0,
244 charge+jnrC+0,charge+jnrD+0);
245 vdwjidx0A = 2*vdwtype[jnrA+0];
246 vdwjidx0B = 2*vdwtype[jnrB+0];
247 vdwjidx0C = 2*vdwtype[jnrC+0];
248 vdwjidx0D = 2*vdwtype[jnrD+0];
250 fjx0 = _mm256_setzero_pd();
251 fjy0 = _mm256_setzero_pd();
252 fjz0 = _mm256_setzero_pd();
254 /**************************
255 * CALCULATE INTERACTIONS *
256 **************************/
258 r00 = _mm256_mul_pd(rsq00,rinv00);
260 /* Compute parameters for interactions between i and j atoms */
261 gmx_mm256_load_4pair_swizzle_pd(vdwioffsetptr0+vdwjidx0A,
262 vdwioffsetptr0+vdwjidx0B,
263 vdwioffsetptr0+vdwjidx0C,
264 vdwioffsetptr0+vdwjidx0D,
267 /* Calculate table index by multiplying r with table scale and truncate to integer */
268 rt = _mm256_mul_pd(r00,vftabscale);
269 vfitab = _mm256_cvttpd_epi32(rt);
270 vfeps = _mm256_sub_pd(rt,_mm256_round_pd(rt, _MM_FROUND_FLOOR));
271 vfitab = _mm_slli_epi32(vfitab,3);
273 /* CUBIC SPLINE TABLE DISPERSION */
274 Y = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
275 F = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
276 G = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,2) );
277 H = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,3) );
278 GMX_MM256_FULLTRANSPOSE4_PD(Y,F,G,H);
279 Heps = _mm256_mul_pd(vfeps,H);
280 Fp = _mm256_add_pd(F,_mm256_mul_pd(vfeps,_mm256_add_pd(G,Heps)));
281 VV = _mm256_add_pd(Y,_mm256_mul_pd(vfeps,Fp));
282 vvdw6 = _mm256_mul_pd(c6_00,VV);
283 FF = _mm256_add_pd(Fp,_mm256_mul_pd(vfeps,_mm256_add_pd(G,_mm256_add_pd(Heps,Heps))));
284 fvdw6 = _mm256_mul_pd(c6_00,FF);
286 /* CUBIC SPLINE TABLE REPULSION */
287 vfitab = _mm_add_epi32(vfitab,ifour);
288 Y = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
289 F = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
290 G = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,2) );
291 H = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,3) );
292 GMX_MM256_FULLTRANSPOSE4_PD(Y,F,G,H);
293 Heps = _mm256_mul_pd(vfeps,H);
294 Fp = _mm256_add_pd(F,_mm256_mul_pd(vfeps,_mm256_add_pd(G,Heps)));
295 VV = _mm256_add_pd(Y,_mm256_mul_pd(vfeps,Fp));
296 vvdw12 = _mm256_mul_pd(c12_00,VV);
297 FF = _mm256_add_pd(Fp,_mm256_mul_pd(vfeps,_mm256_add_pd(G,_mm256_add_pd(Heps,Heps))));
298 fvdw12 = _mm256_mul_pd(c12_00,FF);
299 vvdw = _mm256_add_pd(vvdw12,vvdw6);
300 fvdw = _mm256_xor_pd(signbit,_mm256_mul_pd(_mm256_add_pd(fvdw6,fvdw12),_mm256_mul_pd(vftabscale,rinv00)));
302 /* Update potential sum for this i atom from the interaction with this j atom. */
303 vvdwsum = _mm256_add_pd(vvdwsum,vvdw);
307 /* Calculate temporary vectorial force */
308 tx = _mm256_mul_pd(fscal,dx00);
309 ty = _mm256_mul_pd(fscal,dy00);
310 tz = _mm256_mul_pd(fscal,dz00);
312 /* Update vectorial force */
313 fix0 = _mm256_add_pd(fix0,tx);
314 fiy0 = _mm256_add_pd(fiy0,ty);
315 fiz0 = _mm256_add_pd(fiz0,tz);
317 fjx0 = _mm256_add_pd(fjx0,tx);
318 fjy0 = _mm256_add_pd(fjy0,ty);
319 fjz0 = _mm256_add_pd(fjz0,tz);
321 /**************************
322 * CALCULATE INTERACTIONS *
323 **************************/
325 /* Compute parameters for interactions between i and j atoms */
326 qq10 = _mm256_mul_pd(iq1,jq0);
328 /* REACTION-FIELD ELECTROSTATICS */
329 velec = _mm256_mul_pd(qq10,_mm256_sub_pd(_mm256_add_pd(rinv10,_mm256_mul_pd(krf,rsq10)),crf));
330 felec = _mm256_mul_pd(qq10,_mm256_sub_pd(_mm256_mul_pd(rinv10,rinvsq10),krf2));
332 /* Update potential sum for this i atom from the interaction with this j atom. */
333 velecsum = _mm256_add_pd(velecsum,velec);
337 /* Calculate temporary vectorial force */
338 tx = _mm256_mul_pd(fscal,dx10);
339 ty = _mm256_mul_pd(fscal,dy10);
340 tz = _mm256_mul_pd(fscal,dz10);
342 /* Update vectorial force */
343 fix1 = _mm256_add_pd(fix1,tx);
344 fiy1 = _mm256_add_pd(fiy1,ty);
345 fiz1 = _mm256_add_pd(fiz1,tz);
347 fjx0 = _mm256_add_pd(fjx0,tx);
348 fjy0 = _mm256_add_pd(fjy0,ty);
349 fjz0 = _mm256_add_pd(fjz0,tz);
351 /**************************
352 * CALCULATE INTERACTIONS *
353 **************************/
355 /* Compute parameters for interactions between i and j atoms */
356 qq20 = _mm256_mul_pd(iq2,jq0);
358 /* REACTION-FIELD ELECTROSTATICS */
359 velec = _mm256_mul_pd(qq20,_mm256_sub_pd(_mm256_add_pd(rinv20,_mm256_mul_pd(krf,rsq20)),crf));
360 felec = _mm256_mul_pd(qq20,_mm256_sub_pd(_mm256_mul_pd(rinv20,rinvsq20),krf2));
362 /* Update potential sum for this i atom from the interaction with this j atom. */
363 velecsum = _mm256_add_pd(velecsum,velec);
367 /* Calculate temporary vectorial force */
368 tx = _mm256_mul_pd(fscal,dx20);
369 ty = _mm256_mul_pd(fscal,dy20);
370 tz = _mm256_mul_pd(fscal,dz20);
372 /* Update vectorial force */
373 fix2 = _mm256_add_pd(fix2,tx);
374 fiy2 = _mm256_add_pd(fiy2,ty);
375 fiz2 = _mm256_add_pd(fiz2,tz);
377 fjx0 = _mm256_add_pd(fjx0,tx);
378 fjy0 = _mm256_add_pd(fjy0,ty);
379 fjz0 = _mm256_add_pd(fjz0,tz);
381 /**************************
382 * CALCULATE INTERACTIONS *
383 **************************/
385 /* Compute parameters for interactions between i and j atoms */
386 qq30 = _mm256_mul_pd(iq3,jq0);
388 /* REACTION-FIELD ELECTROSTATICS */
389 velec = _mm256_mul_pd(qq30,_mm256_sub_pd(_mm256_add_pd(rinv30,_mm256_mul_pd(krf,rsq30)),crf));
390 felec = _mm256_mul_pd(qq30,_mm256_sub_pd(_mm256_mul_pd(rinv30,rinvsq30),krf2));
392 /* Update potential sum for this i atom from the interaction with this j atom. */
393 velecsum = _mm256_add_pd(velecsum,velec);
397 /* Calculate temporary vectorial force */
398 tx = _mm256_mul_pd(fscal,dx30);
399 ty = _mm256_mul_pd(fscal,dy30);
400 tz = _mm256_mul_pd(fscal,dz30);
402 /* Update vectorial force */
403 fix3 = _mm256_add_pd(fix3,tx);
404 fiy3 = _mm256_add_pd(fiy3,ty);
405 fiz3 = _mm256_add_pd(fiz3,tz);
407 fjx0 = _mm256_add_pd(fjx0,tx);
408 fjy0 = _mm256_add_pd(fjy0,ty);
409 fjz0 = _mm256_add_pd(fjz0,tz);
411 fjptrA = f+j_coord_offsetA;
412 fjptrB = f+j_coord_offsetB;
413 fjptrC = f+j_coord_offsetC;
414 fjptrD = f+j_coord_offsetD;
416 gmx_mm256_decrement_1rvec_4ptr_swizzle_pd(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
418 /* Inner loop uses 155 flops */
424 /* Get j neighbor index, and coordinate index */
425 jnrlistA = jjnr[jidx];
426 jnrlistB = jjnr[jidx+1];
427 jnrlistC = jjnr[jidx+2];
428 jnrlistD = jjnr[jidx+3];
429 /* Sign of each element will be negative for non-real atoms.
430 * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
431 * so use it as val = _mm_andnot_pd(mask,val) to clear dummy entries.
433 tmpmask0 = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
435 tmpmask1 = _mm_permute_ps(tmpmask0,_GMX_MM_PERMUTE(3,3,2,2));
436 tmpmask0 = _mm_permute_ps(tmpmask0,_GMX_MM_PERMUTE(1,1,0,0));
437 dummy_mask = _mm256_castps_pd(gmx_mm256_set_m128(tmpmask1,tmpmask0));
439 jnrA = (jnrlistA>=0) ? jnrlistA : 0;
440 jnrB = (jnrlistB>=0) ? jnrlistB : 0;
441 jnrC = (jnrlistC>=0) ? jnrlistC : 0;
442 jnrD = (jnrlistD>=0) ? jnrlistD : 0;
443 j_coord_offsetA = DIM*jnrA;
444 j_coord_offsetB = DIM*jnrB;
445 j_coord_offsetC = DIM*jnrC;
446 j_coord_offsetD = DIM*jnrD;
448 /* load j atom coordinates */
449 gmx_mm256_load_1rvec_4ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
450 x+j_coord_offsetC,x+j_coord_offsetD,
453 /* Calculate displacement vector */
454 dx00 = _mm256_sub_pd(ix0,jx0);
455 dy00 = _mm256_sub_pd(iy0,jy0);
456 dz00 = _mm256_sub_pd(iz0,jz0);
457 dx10 = _mm256_sub_pd(ix1,jx0);
458 dy10 = _mm256_sub_pd(iy1,jy0);
459 dz10 = _mm256_sub_pd(iz1,jz0);
460 dx20 = _mm256_sub_pd(ix2,jx0);
461 dy20 = _mm256_sub_pd(iy2,jy0);
462 dz20 = _mm256_sub_pd(iz2,jz0);
463 dx30 = _mm256_sub_pd(ix3,jx0);
464 dy30 = _mm256_sub_pd(iy3,jy0);
465 dz30 = _mm256_sub_pd(iz3,jz0);
467 /* Calculate squared distance and things based on it */
468 rsq00 = gmx_mm256_calc_rsq_pd(dx00,dy00,dz00);
469 rsq10 = gmx_mm256_calc_rsq_pd(dx10,dy10,dz10);
470 rsq20 = gmx_mm256_calc_rsq_pd(dx20,dy20,dz20);
471 rsq30 = gmx_mm256_calc_rsq_pd(dx30,dy30,dz30);
473 rinv00 = avx256_invsqrt_d(rsq00);
474 rinv10 = avx256_invsqrt_d(rsq10);
475 rinv20 = avx256_invsqrt_d(rsq20);
476 rinv30 = avx256_invsqrt_d(rsq30);
478 rinvsq10 = _mm256_mul_pd(rinv10,rinv10);
479 rinvsq20 = _mm256_mul_pd(rinv20,rinv20);
480 rinvsq30 = _mm256_mul_pd(rinv30,rinv30);
482 /* Load parameters for j particles */
483 jq0 = gmx_mm256_load_4real_swizzle_pd(charge+jnrA+0,charge+jnrB+0,
484 charge+jnrC+0,charge+jnrD+0);
485 vdwjidx0A = 2*vdwtype[jnrA+0];
486 vdwjidx0B = 2*vdwtype[jnrB+0];
487 vdwjidx0C = 2*vdwtype[jnrC+0];
488 vdwjidx0D = 2*vdwtype[jnrD+0];
490 fjx0 = _mm256_setzero_pd();
491 fjy0 = _mm256_setzero_pd();
492 fjz0 = _mm256_setzero_pd();
494 /**************************
495 * CALCULATE INTERACTIONS *
496 **************************/
498 r00 = _mm256_mul_pd(rsq00,rinv00);
499 r00 = _mm256_andnot_pd(dummy_mask,r00);
501 /* Compute parameters for interactions between i and j atoms */
502 gmx_mm256_load_4pair_swizzle_pd(vdwioffsetptr0+vdwjidx0A,
503 vdwioffsetptr0+vdwjidx0B,
504 vdwioffsetptr0+vdwjidx0C,
505 vdwioffsetptr0+vdwjidx0D,
508 /* Calculate table index by multiplying r with table scale and truncate to integer */
509 rt = _mm256_mul_pd(r00,vftabscale);
510 vfitab = _mm256_cvttpd_epi32(rt);
511 vfeps = _mm256_sub_pd(rt,_mm256_round_pd(rt, _MM_FROUND_FLOOR));
512 vfitab = _mm_slli_epi32(vfitab,3);
514 /* CUBIC SPLINE TABLE DISPERSION */
515 Y = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
516 F = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
517 G = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,2) );
518 H = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,3) );
519 GMX_MM256_FULLTRANSPOSE4_PD(Y,F,G,H);
520 Heps = _mm256_mul_pd(vfeps,H);
521 Fp = _mm256_add_pd(F,_mm256_mul_pd(vfeps,_mm256_add_pd(G,Heps)));
522 VV = _mm256_add_pd(Y,_mm256_mul_pd(vfeps,Fp));
523 vvdw6 = _mm256_mul_pd(c6_00,VV);
524 FF = _mm256_add_pd(Fp,_mm256_mul_pd(vfeps,_mm256_add_pd(G,_mm256_add_pd(Heps,Heps))));
525 fvdw6 = _mm256_mul_pd(c6_00,FF);
527 /* CUBIC SPLINE TABLE REPULSION */
528 vfitab = _mm_add_epi32(vfitab,ifour);
529 Y = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
530 F = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
531 G = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,2) );
532 H = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,3) );
533 GMX_MM256_FULLTRANSPOSE4_PD(Y,F,G,H);
534 Heps = _mm256_mul_pd(vfeps,H);
535 Fp = _mm256_add_pd(F,_mm256_mul_pd(vfeps,_mm256_add_pd(G,Heps)));
536 VV = _mm256_add_pd(Y,_mm256_mul_pd(vfeps,Fp));
537 vvdw12 = _mm256_mul_pd(c12_00,VV);
538 FF = _mm256_add_pd(Fp,_mm256_mul_pd(vfeps,_mm256_add_pd(G,_mm256_add_pd(Heps,Heps))));
539 fvdw12 = _mm256_mul_pd(c12_00,FF);
540 vvdw = _mm256_add_pd(vvdw12,vvdw6);
541 fvdw = _mm256_xor_pd(signbit,_mm256_mul_pd(_mm256_add_pd(fvdw6,fvdw12),_mm256_mul_pd(vftabscale,rinv00)));
543 /* Update potential sum for this i atom from the interaction with this j atom. */
544 vvdw = _mm256_andnot_pd(dummy_mask,vvdw);
545 vvdwsum = _mm256_add_pd(vvdwsum,vvdw);
549 fscal = _mm256_andnot_pd(dummy_mask,fscal);
551 /* Calculate temporary vectorial force */
552 tx = _mm256_mul_pd(fscal,dx00);
553 ty = _mm256_mul_pd(fscal,dy00);
554 tz = _mm256_mul_pd(fscal,dz00);
556 /* Update vectorial force */
557 fix0 = _mm256_add_pd(fix0,tx);
558 fiy0 = _mm256_add_pd(fiy0,ty);
559 fiz0 = _mm256_add_pd(fiz0,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 qq10 = _mm256_mul_pd(iq1,jq0);
572 /* REACTION-FIELD ELECTROSTATICS */
573 velec = _mm256_mul_pd(qq10,_mm256_sub_pd(_mm256_add_pd(rinv10,_mm256_mul_pd(krf,rsq10)),crf));
574 felec = _mm256_mul_pd(qq10,_mm256_sub_pd(_mm256_mul_pd(rinv10,rinvsq10),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,dx10);
586 ty = _mm256_mul_pd(fscal,dy10);
587 tz = _mm256_mul_pd(fscal,dz10);
589 /* Update vectorial force */
590 fix1 = _mm256_add_pd(fix1,tx);
591 fiy1 = _mm256_add_pd(fiy1,ty);
592 fiz1 = _mm256_add_pd(fiz1,tz);
594 fjx0 = _mm256_add_pd(fjx0,tx);
595 fjy0 = _mm256_add_pd(fjy0,ty);
596 fjz0 = _mm256_add_pd(fjz0,tz);
598 /**************************
599 * CALCULATE INTERACTIONS *
600 **************************/
602 /* Compute parameters for interactions between i and j atoms */
603 qq20 = _mm256_mul_pd(iq2,jq0);
605 /* REACTION-FIELD ELECTROSTATICS */
606 velec = _mm256_mul_pd(qq20,_mm256_sub_pd(_mm256_add_pd(rinv20,_mm256_mul_pd(krf,rsq20)),crf));
607 felec = _mm256_mul_pd(qq20,_mm256_sub_pd(_mm256_mul_pd(rinv20,rinvsq20),krf2));
609 /* Update potential sum for this i atom from the interaction with this j atom. */
610 velec = _mm256_andnot_pd(dummy_mask,velec);
611 velecsum = _mm256_add_pd(velecsum,velec);
615 fscal = _mm256_andnot_pd(dummy_mask,fscal);
617 /* Calculate temporary vectorial force */
618 tx = _mm256_mul_pd(fscal,dx20);
619 ty = _mm256_mul_pd(fscal,dy20);
620 tz = _mm256_mul_pd(fscal,dz20);
622 /* Update vectorial force */
623 fix2 = _mm256_add_pd(fix2,tx);
624 fiy2 = _mm256_add_pd(fiy2,ty);
625 fiz2 = _mm256_add_pd(fiz2,tz);
627 fjx0 = _mm256_add_pd(fjx0,tx);
628 fjy0 = _mm256_add_pd(fjy0,ty);
629 fjz0 = _mm256_add_pd(fjz0,tz);
631 /**************************
632 * CALCULATE INTERACTIONS *
633 **************************/
635 /* Compute parameters for interactions between i and j atoms */
636 qq30 = _mm256_mul_pd(iq3,jq0);
638 /* REACTION-FIELD ELECTROSTATICS */
639 velec = _mm256_mul_pd(qq30,_mm256_sub_pd(_mm256_add_pd(rinv30,_mm256_mul_pd(krf,rsq30)),crf));
640 felec = _mm256_mul_pd(qq30,_mm256_sub_pd(_mm256_mul_pd(rinv30,rinvsq30),krf2));
642 /* Update potential sum for this i atom from the interaction with this j atom. */
643 velec = _mm256_andnot_pd(dummy_mask,velec);
644 velecsum = _mm256_add_pd(velecsum,velec);
648 fscal = _mm256_andnot_pd(dummy_mask,fscal);
650 /* Calculate temporary vectorial force */
651 tx = _mm256_mul_pd(fscal,dx30);
652 ty = _mm256_mul_pd(fscal,dy30);
653 tz = _mm256_mul_pd(fscal,dz30);
655 /* Update vectorial force */
656 fix3 = _mm256_add_pd(fix3,tx);
657 fiy3 = _mm256_add_pd(fiy3,ty);
658 fiz3 = _mm256_add_pd(fiz3,tz);
660 fjx0 = _mm256_add_pd(fjx0,tx);
661 fjy0 = _mm256_add_pd(fjy0,ty);
662 fjz0 = _mm256_add_pd(fjz0,tz);
664 fjptrA = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
665 fjptrB = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
666 fjptrC = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
667 fjptrD = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
669 gmx_mm256_decrement_1rvec_4ptr_swizzle_pd(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
671 /* Inner loop uses 156 flops */
674 /* End of innermost loop */
676 gmx_mm256_update_iforce_4atom_swizzle_pd(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,fix3,fiy3,fiz3,
677 f+i_coord_offset,fshift+i_shift_offset);
680 /* Update potential energies */
681 gmx_mm256_update_1pot_pd(velecsum,kernel_data->energygrp_elec+ggid);
682 gmx_mm256_update_1pot_pd(vvdwsum,kernel_data->energygrp_vdw+ggid);
684 /* Increment number of inner iterations */
685 inneriter += j_index_end - j_index_start;
687 /* Outer loop uses 26 flops */
690 /* Increment number of outer iterations */
693 /* Update outer/inner flops */
695 inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W4_VF,outeriter*26 + inneriter*156);
698 * Gromacs nonbonded kernel: nb_kernel_ElecRF_VdwCSTab_GeomW4P1_F_avx_256_double
699 * Electrostatics interaction: ReactionField
700 * VdW interaction: CubicSplineTable
701 * Geometry: Water4-Particle
702 * Calculate force/pot: Force
705 nb_kernel_ElecRF_VdwCSTab_GeomW4P1_F_avx_256_double
706 (t_nblist * gmx_restrict nlist,
707 rvec * gmx_restrict xx,
708 rvec * gmx_restrict ff,
709 struct t_forcerec * gmx_restrict fr,
710 t_mdatoms * gmx_restrict mdatoms,
711 nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
712 t_nrnb * gmx_restrict nrnb)
714 /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
715 * just 0 for non-waters.
716 * Suffixes A,B,C,D refer to j loop unrolling done with AVX, e.g. for the four different
717 * jnr indices corresponding to data put in the four positions in the SIMD register.
719 int i_shift_offset,i_coord_offset,outeriter,inneriter;
720 int j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
721 int jnrA,jnrB,jnrC,jnrD;
722 int jnrlistA,jnrlistB,jnrlistC,jnrlistD;
723 int jnrlistE,jnrlistF,jnrlistG,jnrlistH;
724 int j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
725 int *iinr,*jindex,*jjnr,*shiftidx,*gid;
727 real *shiftvec,*fshift,*x,*f;
728 real *fjptrA,*fjptrB,*fjptrC,*fjptrD;
730 __m256d tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
731 real * vdwioffsetptr0;
732 __m256d ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
733 real * vdwioffsetptr1;
734 __m256d ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
735 real * vdwioffsetptr2;
736 __m256d ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
737 real * vdwioffsetptr3;
738 __m256d ix3,iy3,iz3,fix3,fiy3,fiz3,iq3,isai3;
739 int vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
740 __m256d jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
741 __m256d dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
742 __m256d dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
743 __m256d dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
744 __m256d dx30,dy30,dz30,rsq30,rinv30,rinvsq30,r30,qq30,c6_30,c12_30;
745 __m256d velec,felec,velecsum,facel,crf,krf,krf2;
748 __m256d rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
751 __m256d one_sixth = _mm256_set1_pd(1.0/6.0);
752 __m256d one_twelfth = _mm256_set1_pd(1.0/12.0);
754 __m128i ifour = _mm_set1_epi32(4);
755 __m256d rt,vfeps,vftabscale,Y,F,G,H,Heps,Fp,VV,FF;
757 __m256d dummy_mask,cutoff_mask;
758 __m128 tmpmask0,tmpmask1;
759 __m256d signbit = _mm256_castsi256_pd( _mm256_set1_epi32(0x80000000) );
760 __m256d one = _mm256_set1_pd(1.0);
761 __m256d two = _mm256_set1_pd(2.0);
767 jindex = nlist->jindex;
769 shiftidx = nlist->shift;
771 shiftvec = fr->shift_vec[0];
772 fshift = fr->fshift[0];
773 facel = _mm256_set1_pd(fr->ic->epsfac);
774 charge = mdatoms->chargeA;
775 krf = _mm256_set1_pd(fr->ic->k_rf);
776 krf2 = _mm256_set1_pd(fr->ic->k_rf*2.0);
777 crf = _mm256_set1_pd(fr->ic->c_rf);
778 nvdwtype = fr->ntype;
780 vdwtype = mdatoms->typeA;
782 vftab = kernel_data->table_vdw->data;
783 vftabscale = _mm256_set1_pd(kernel_data->table_vdw->scale);
785 /* Setup water-specific parameters */
786 inr = nlist->iinr[0];
787 iq1 = _mm256_mul_pd(facel,_mm256_set1_pd(charge[inr+1]));
788 iq2 = _mm256_mul_pd(facel,_mm256_set1_pd(charge[inr+2]));
789 iq3 = _mm256_mul_pd(facel,_mm256_set1_pd(charge[inr+3]));
790 vdwioffsetptr0 = vdwparam+2*nvdwtype*vdwtype[inr+0];
792 /* Avoid stupid compiler warnings */
793 jnrA = jnrB = jnrC = jnrD = 0;
802 for(iidx=0;iidx<4*DIM;iidx++)
807 /* Start outer loop over neighborlists */
808 for(iidx=0; iidx<nri; iidx++)
810 /* Load shift vector for this list */
811 i_shift_offset = DIM*shiftidx[iidx];
813 /* Load limits for loop over neighbors */
814 j_index_start = jindex[iidx];
815 j_index_end = jindex[iidx+1];
817 /* Get outer coordinate index */
819 i_coord_offset = DIM*inr;
821 /* Load i particle coords and add shift vector */
822 gmx_mm256_load_shift_and_4rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset,
823 &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2,&ix3,&iy3,&iz3);
825 fix0 = _mm256_setzero_pd();
826 fiy0 = _mm256_setzero_pd();
827 fiz0 = _mm256_setzero_pd();
828 fix1 = _mm256_setzero_pd();
829 fiy1 = _mm256_setzero_pd();
830 fiz1 = _mm256_setzero_pd();
831 fix2 = _mm256_setzero_pd();
832 fiy2 = _mm256_setzero_pd();
833 fiz2 = _mm256_setzero_pd();
834 fix3 = _mm256_setzero_pd();
835 fiy3 = _mm256_setzero_pd();
836 fiz3 = _mm256_setzero_pd();
838 /* Start inner kernel loop */
839 for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
842 /* Get j neighbor index, and coordinate index */
847 j_coord_offsetA = DIM*jnrA;
848 j_coord_offsetB = DIM*jnrB;
849 j_coord_offsetC = DIM*jnrC;
850 j_coord_offsetD = DIM*jnrD;
852 /* load j atom coordinates */
853 gmx_mm256_load_1rvec_4ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
854 x+j_coord_offsetC,x+j_coord_offsetD,
857 /* Calculate displacement vector */
858 dx00 = _mm256_sub_pd(ix0,jx0);
859 dy00 = _mm256_sub_pd(iy0,jy0);
860 dz00 = _mm256_sub_pd(iz0,jz0);
861 dx10 = _mm256_sub_pd(ix1,jx0);
862 dy10 = _mm256_sub_pd(iy1,jy0);
863 dz10 = _mm256_sub_pd(iz1,jz0);
864 dx20 = _mm256_sub_pd(ix2,jx0);
865 dy20 = _mm256_sub_pd(iy2,jy0);
866 dz20 = _mm256_sub_pd(iz2,jz0);
867 dx30 = _mm256_sub_pd(ix3,jx0);
868 dy30 = _mm256_sub_pd(iy3,jy0);
869 dz30 = _mm256_sub_pd(iz3,jz0);
871 /* Calculate squared distance and things based on it */
872 rsq00 = gmx_mm256_calc_rsq_pd(dx00,dy00,dz00);
873 rsq10 = gmx_mm256_calc_rsq_pd(dx10,dy10,dz10);
874 rsq20 = gmx_mm256_calc_rsq_pd(dx20,dy20,dz20);
875 rsq30 = gmx_mm256_calc_rsq_pd(dx30,dy30,dz30);
877 rinv00 = avx256_invsqrt_d(rsq00);
878 rinv10 = avx256_invsqrt_d(rsq10);
879 rinv20 = avx256_invsqrt_d(rsq20);
880 rinv30 = avx256_invsqrt_d(rsq30);
882 rinvsq10 = _mm256_mul_pd(rinv10,rinv10);
883 rinvsq20 = _mm256_mul_pd(rinv20,rinv20);
884 rinvsq30 = _mm256_mul_pd(rinv30,rinv30);
886 /* Load parameters for j particles */
887 jq0 = gmx_mm256_load_4real_swizzle_pd(charge+jnrA+0,charge+jnrB+0,
888 charge+jnrC+0,charge+jnrD+0);
889 vdwjidx0A = 2*vdwtype[jnrA+0];
890 vdwjidx0B = 2*vdwtype[jnrB+0];
891 vdwjidx0C = 2*vdwtype[jnrC+0];
892 vdwjidx0D = 2*vdwtype[jnrD+0];
894 fjx0 = _mm256_setzero_pd();
895 fjy0 = _mm256_setzero_pd();
896 fjz0 = _mm256_setzero_pd();
898 /**************************
899 * CALCULATE INTERACTIONS *
900 **************************/
902 r00 = _mm256_mul_pd(rsq00,rinv00);
904 /* Compute parameters for interactions between i and j atoms */
905 gmx_mm256_load_4pair_swizzle_pd(vdwioffsetptr0+vdwjidx0A,
906 vdwioffsetptr0+vdwjidx0B,
907 vdwioffsetptr0+vdwjidx0C,
908 vdwioffsetptr0+vdwjidx0D,
911 /* Calculate table index by multiplying r with table scale and truncate to integer */
912 rt = _mm256_mul_pd(r00,vftabscale);
913 vfitab = _mm256_cvttpd_epi32(rt);
914 vfeps = _mm256_sub_pd(rt,_mm256_round_pd(rt, _MM_FROUND_FLOOR));
915 vfitab = _mm_slli_epi32(vfitab,3);
917 /* CUBIC SPLINE TABLE DISPERSION */
918 Y = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
919 F = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
920 G = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,2) );
921 H = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,3) );
922 GMX_MM256_FULLTRANSPOSE4_PD(Y,F,G,H);
923 Heps = _mm256_mul_pd(vfeps,H);
924 Fp = _mm256_add_pd(F,_mm256_mul_pd(vfeps,_mm256_add_pd(G,Heps)));
925 FF = _mm256_add_pd(Fp,_mm256_mul_pd(vfeps,_mm256_add_pd(G,_mm256_add_pd(Heps,Heps))));
926 fvdw6 = _mm256_mul_pd(c6_00,FF);
928 /* CUBIC SPLINE TABLE REPULSION */
929 vfitab = _mm_add_epi32(vfitab,ifour);
930 Y = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
931 F = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
932 G = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,2) );
933 H = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,3) );
934 GMX_MM256_FULLTRANSPOSE4_PD(Y,F,G,H);
935 Heps = _mm256_mul_pd(vfeps,H);
936 Fp = _mm256_add_pd(F,_mm256_mul_pd(vfeps,_mm256_add_pd(G,Heps)));
937 FF = _mm256_add_pd(Fp,_mm256_mul_pd(vfeps,_mm256_add_pd(G,_mm256_add_pd(Heps,Heps))));
938 fvdw12 = _mm256_mul_pd(c12_00,FF);
939 fvdw = _mm256_xor_pd(signbit,_mm256_mul_pd(_mm256_add_pd(fvdw6,fvdw12),_mm256_mul_pd(vftabscale,rinv00)));
943 /* Calculate temporary vectorial force */
944 tx = _mm256_mul_pd(fscal,dx00);
945 ty = _mm256_mul_pd(fscal,dy00);
946 tz = _mm256_mul_pd(fscal,dz00);
948 /* Update vectorial force */
949 fix0 = _mm256_add_pd(fix0,tx);
950 fiy0 = _mm256_add_pd(fiy0,ty);
951 fiz0 = _mm256_add_pd(fiz0,tz);
953 fjx0 = _mm256_add_pd(fjx0,tx);
954 fjy0 = _mm256_add_pd(fjy0,ty);
955 fjz0 = _mm256_add_pd(fjz0,tz);
957 /**************************
958 * CALCULATE INTERACTIONS *
959 **************************/
961 /* Compute parameters for interactions between i and j atoms */
962 qq10 = _mm256_mul_pd(iq1,jq0);
964 /* REACTION-FIELD ELECTROSTATICS */
965 felec = _mm256_mul_pd(qq10,_mm256_sub_pd(_mm256_mul_pd(rinv10,rinvsq10),krf2));
969 /* Calculate temporary vectorial force */
970 tx = _mm256_mul_pd(fscal,dx10);
971 ty = _mm256_mul_pd(fscal,dy10);
972 tz = _mm256_mul_pd(fscal,dz10);
974 /* Update vectorial force */
975 fix1 = _mm256_add_pd(fix1,tx);
976 fiy1 = _mm256_add_pd(fiy1,ty);
977 fiz1 = _mm256_add_pd(fiz1,tz);
979 fjx0 = _mm256_add_pd(fjx0,tx);
980 fjy0 = _mm256_add_pd(fjy0,ty);
981 fjz0 = _mm256_add_pd(fjz0,tz);
983 /**************************
984 * CALCULATE INTERACTIONS *
985 **************************/
987 /* Compute parameters for interactions between i and j atoms */
988 qq20 = _mm256_mul_pd(iq2,jq0);
990 /* REACTION-FIELD ELECTROSTATICS */
991 felec = _mm256_mul_pd(qq20,_mm256_sub_pd(_mm256_mul_pd(rinv20,rinvsq20),krf2));
995 /* Calculate temporary vectorial force */
996 tx = _mm256_mul_pd(fscal,dx20);
997 ty = _mm256_mul_pd(fscal,dy20);
998 tz = _mm256_mul_pd(fscal,dz20);
1000 /* Update vectorial force */
1001 fix2 = _mm256_add_pd(fix2,tx);
1002 fiy2 = _mm256_add_pd(fiy2,ty);
1003 fiz2 = _mm256_add_pd(fiz2,tz);
1005 fjx0 = _mm256_add_pd(fjx0,tx);
1006 fjy0 = _mm256_add_pd(fjy0,ty);
1007 fjz0 = _mm256_add_pd(fjz0,tz);
1009 /**************************
1010 * CALCULATE INTERACTIONS *
1011 **************************/
1013 /* Compute parameters for interactions between i and j atoms */
1014 qq30 = _mm256_mul_pd(iq3,jq0);
1016 /* REACTION-FIELD ELECTROSTATICS */
1017 felec = _mm256_mul_pd(qq30,_mm256_sub_pd(_mm256_mul_pd(rinv30,rinvsq30),krf2));
1021 /* Calculate temporary vectorial force */
1022 tx = _mm256_mul_pd(fscal,dx30);
1023 ty = _mm256_mul_pd(fscal,dy30);
1024 tz = _mm256_mul_pd(fscal,dz30);
1026 /* Update vectorial force */
1027 fix3 = _mm256_add_pd(fix3,tx);
1028 fiy3 = _mm256_add_pd(fiy3,ty);
1029 fiz3 = _mm256_add_pd(fiz3,tz);
1031 fjx0 = _mm256_add_pd(fjx0,tx);
1032 fjy0 = _mm256_add_pd(fjy0,ty);
1033 fjz0 = _mm256_add_pd(fjz0,tz);
1035 fjptrA = f+j_coord_offsetA;
1036 fjptrB = f+j_coord_offsetB;
1037 fjptrC = f+j_coord_offsetC;
1038 fjptrD = f+j_coord_offsetD;
1040 gmx_mm256_decrement_1rvec_4ptr_swizzle_pd(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
1042 /* Inner loop uses 132 flops */
1045 if(jidx<j_index_end)
1048 /* Get j neighbor index, and coordinate index */
1049 jnrlistA = jjnr[jidx];
1050 jnrlistB = jjnr[jidx+1];
1051 jnrlistC = jjnr[jidx+2];
1052 jnrlistD = jjnr[jidx+3];
1053 /* Sign of each element will be negative for non-real atoms.
1054 * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
1055 * so use it as val = _mm_andnot_pd(mask,val) to clear dummy entries.
1057 tmpmask0 = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
1059 tmpmask1 = _mm_permute_ps(tmpmask0,_GMX_MM_PERMUTE(3,3,2,2));
1060 tmpmask0 = _mm_permute_ps(tmpmask0,_GMX_MM_PERMUTE(1,1,0,0));
1061 dummy_mask = _mm256_castps_pd(gmx_mm256_set_m128(tmpmask1,tmpmask0));
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 j_coord_offsetA = DIM*jnrA;
1068 j_coord_offsetB = DIM*jnrB;
1069 j_coord_offsetC = DIM*jnrC;
1070 j_coord_offsetD = DIM*jnrD;
1072 /* load j atom coordinates */
1073 gmx_mm256_load_1rvec_4ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
1074 x+j_coord_offsetC,x+j_coord_offsetD,
1077 /* Calculate displacement vector */
1078 dx00 = _mm256_sub_pd(ix0,jx0);
1079 dy00 = _mm256_sub_pd(iy0,jy0);
1080 dz00 = _mm256_sub_pd(iz0,jz0);
1081 dx10 = _mm256_sub_pd(ix1,jx0);
1082 dy10 = _mm256_sub_pd(iy1,jy0);
1083 dz10 = _mm256_sub_pd(iz1,jz0);
1084 dx20 = _mm256_sub_pd(ix2,jx0);
1085 dy20 = _mm256_sub_pd(iy2,jy0);
1086 dz20 = _mm256_sub_pd(iz2,jz0);
1087 dx30 = _mm256_sub_pd(ix3,jx0);
1088 dy30 = _mm256_sub_pd(iy3,jy0);
1089 dz30 = _mm256_sub_pd(iz3,jz0);
1091 /* Calculate squared distance and things based on it */
1092 rsq00 = gmx_mm256_calc_rsq_pd(dx00,dy00,dz00);
1093 rsq10 = gmx_mm256_calc_rsq_pd(dx10,dy10,dz10);
1094 rsq20 = gmx_mm256_calc_rsq_pd(dx20,dy20,dz20);
1095 rsq30 = gmx_mm256_calc_rsq_pd(dx30,dy30,dz30);
1097 rinv00 = avx256_invsqrt_d(rsq00);
1098 rinv10 = avx256_invsqrt_d(rsq10);
1099 rinv20 = avx256_invsqrt_d(rsq20);
1100 rinv30 = avx256_invsqrt_d(rsq30);
1102 rinvsq10 = _mm256_mul_pd(rinv10,rinv10);
1103 rinvsq20 = _mm256_mul_pd(rinv20,rinv20);
1104 rinvsq30 = _mm256_mul_pd(rinv30,rinv30);
1106 /* Load parameters for j particles */
1107 jq0 = gmx_mm256_load_4real_swizzle_pd(charge+jnrA+0,charge+jnrB+0,
1108 charge+jnrC+0,charge+jnrD+0);
1109 vdwjidx0A = 2*vdwtype[jnrA+0];
1110 vdwjidx0B = 2*vdwtype[jnrB+0];
1111 vdwjidx0C = 2*vdwtype[jnrC+0];
1112 vdwjidx0D = 2*vdwtype[jnrD+0];
1114 fjx0 = _mm256_setzero_pd();
1115 fjy0 = _mm256_setzero_pd();
1116 fjz0 = _mm256_setzero_pd();
1118 /**************************
1119 * CALCULATE INTERACTIONS *
1120 **************************/
1122 r00 = _mm256_mul_pd(rsq00,rinv00);
1123 r00 = _mm256_andnot_pd(dummy_mask,r00);
1125 /* Compute parameters for interactions between i and j atoms */
1126 gmx_mm256_load_4pair_swizzle_pd(vdwioffsetptr0+vdwjidx0A,
1127 vdwioffsetptr0+vdwjidx0B,
1128 vdwioffsetptr0+vdwjidx0C,
1129 vdwioffsetptr0+vdwjidx0D,
1132 /* Calculate table index by multiplying r with table scale and truncate to integer */
1133 rt = _mm256_mul_pd(r00,vftabscale);
1134 vfitab = _mm256_cvttpd_epi32(rt);
1135 vfeps = _mm256_sub_pd(rt,_mm256_round_pd(rt, _MM_FROUND_FLOOR));
1136 vfitab = _mm_slli_epi32(vfitab,3);
1138 /* CUBIC SPLINE TABLE DISPERSION */
1139 Y = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
1140 F = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
1141 G = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,2) );
1142 H = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,3) );
1143 GMX_MM256_FULLTRANSPOSE4_PD(Y,F,G,H);
1144 Heps = _mm256_mul_pd(vfeps,H);
1145 Fp = _mm256_add_pd(F,_mm256_mul_pd(vfeps,_mm256_add_pd(G,Heps)));
1146 FF = _mm256_add_pd(Fp,_mm256_mul_pd(vfeps,_mm256_add_pd(G,_mm256_add_pd(Heps,Heps))));
1147 fvdw6 = _mm256_mul_pd(c6_00,FF);
1149 /* CUBIC SPLINE TABLE REPULSION */
1150 vfitab = _mm_add_epi32(vfitab,ifour);
1151 Y = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
1152 F = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
1153 G = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,2) );
1154 H = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,3) );
1155 GMX_MM256_FULLTRANSPOSE4_PD(Y,F,G,H);
1156 Heps = _mm256_mul_pd(vfeps,H);
1157 Fp = _mm256_add_pd(F,_mm256_mul_pd(vfeps,_mm256_add_pd(G,Heps)));
1158 FF = _mm256_add_pd(Fp,_mm256_mul_pd(vfeps,_mm256_add_pd(G,_mm256_add_pd(Heps,Heps))));
1159 fvdw12 = _mm256_mul_pd(c12_00,FF);
1160 fvdw = _mm256_xor_pd(signbit,_mm256_mul_pd(_mm256_add_pd(fvdw6,fvdw12),_mm256_mul_pd(vftabscale,rinv00)));
1164 fscal = _mm256_andnot_pd(dummy_mask,fscal);
1166 /* Calculate temporary vectorial force */
1167 tx = _mm256_mul_pd(fscal,dx00);
1168 ty = _mm256_mul_pd(fscal,dy00);
1169 tz = _mm256_mul_pd(fscal,dz00);
1171 /* Update vectorial force */
1172 fix0 = _mm256_add_pd(fix0,tx);
1173 fiy0 = _mm256_add_pd(fiy0,ty);
1174 fiz0 = _mm256_add_pd(fiz0,tz);
1176 fjx0 = _mm256_add_pd(fjx0,tx);
1177 fjy0 = _mm256_add_pd(fjy0,ty);
1178 fjz0 = _mm256_add_pd(fjz0,tz);
1180 /**************************
1181 * CALCULATE INTERACTIONS *
1182 **************************/
1184 /* Compute parameters for interactions between i and j atoms */
1185 qq10 = _mm256_mul_pd(iq1,jq0);
1187 /* REACTION-FIELD ELECTROSTATICS */
1188 felec = _mm256_mul_pd(qq10,_mm256_sub_pd(_mm256_mul_pd(rinv10,rinvsq10),krf2));
1192 fscal = _mm256_andnot_pd(dummy_mask,fscal);
1194 /* Calculate temporary vectorial force */
1195 tx = _mm256_mul_pd(fscal,dx10);
1196 ty = _mm256_mul_pd(fscal,dy10);
1197 tz = _mm256_mul_pd(fscal,dz10);
1199 /* Update vectorial force */
1200 fix1 = _mm256_add_pd(fix1,tx);
1201 fiy1 = _mm256_add_pd(fiy1,ty);
1202 fiz1 = _mm256_add_pd(fiz1,tz);
1204 fjx0 = _mm256_add_pd(fjx0,tx);
1205 fjy0 = _mm256_add_pd(fjy0,ty);
1206 fjz0 = _mm256_add_pd(fjz0,tz);
1208 /**************************
1209 * CALCULATE INTERACTIONS *
1210 **************************/
1212 /* Compute parameters for interactions between i and j atoms */
1213 qq20 = _mm256_mul_pd(iq2,jq0);
1215 /* REACTION-FIELD ELECTROSTATICS */
1216 felec = _mm256_mul_pd(qq20,_mm256_sub_pd(_mm256_mul_pd(rinv20,rinvsq20),krf2));
1220 fscal = _mm256_andnot_pd(dummy_mask,fscal);
1222 /* Calculate temporary vectorial force */
1223 tx = _mm256_mul_pd(fscal,dx20);
1224 ty = _mm256_mul_pd(fscal,dy20);
1225 tz = _mm256_mul_pd(fscal,dz20);
1227 /* Update vectorial force */
1228 fix2 = _mm256_add_pd(fix2,tx);
1229 fiy2 = _mm256_add_pd(fiy2,ty);
1230 fiz2 = _mm256_add_pd(fiz2,tz);
1232 fjx0 = _mm256_add_pd(fjx0,tx);
1233 fjy0 = _mm256_add_pd(fjy0,ty);
1234 fjz0 = _mm256_add_pd(fjz0,tz);
1236 /**************************
1237 * CALCULATE INTERACTIONS *
1238 **************************/
1240 /* Compute parameters for interactions between i and j atoms */
1241 qq30 = _mm256_mul_pd(iq3,jq0);
1243 /* REACTION-FIELD ELECTROSTATICS */
1244 felec = _mm256_mul_pd(qq30,_mm256_sub_pd(_mm256_mul_pd(rinv30,rinvsq30),krf2));
1248 fscal = _mm256_andnot_pd(dummy_mask,fscal);
1250 /* Calculate temporary vectorial force */
1251 tx = _mm256_mul_pd(fscal,dx30);
1252 ty = _mm256_mul_pd(fscal,dy30);
1253 tz = _mm256_mul_pd(fscal,dz30);
1255 /* Update vectorial force */
1256 fix3 = _mm256_add_pd(fix3,tx);
1257 fiy3 = _mm256_add_pd(fiy3,ty);
1258 fiz3 = _mm256_add_pd(fiz3,tz);
1260 fjx0 = _mm256_add_pd(fjx0,tx);
1261 fjy0 = _mm256_add_pd(fjy0,ty);
1262 fjz0 = _mm256_add_pd(fjz0,tz);
1264 fjptrA = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
1265 fjptrB = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
1266 fjptrC = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
1267 fjptrD = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
1269 gmx_mm256_decrement_1rvec_4ptr_swizzle_pd(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
1271 /* Inner loop uses 133 flops */
1274 /* End of innermost loop */
1276 gmx_mm256_update_iforce_4atom_swizzle_pd(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,fix3,fiy3,fiz3,
1277 f+i_coord_offset,fshift+i_shift_offset);
1279 /* Increment number of inner iterations */
1280 inneriter += j_index_end - j_index_start;
1282 /* Outer loop uses 24 flops */
1285 /* Increment number of outer iterations */
1288 /* Update outer/inner flops */
1290 inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W4_F,outeriter*24 + inneriter*133);