2 * This file is part of the GROMACS molecular simulation package.
4 * Copyright (c) 2012,2013, 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 "types/simple.h"
49 #include "gromacs/simd/math_x86_avx_256_double.h"
50 #include "kernelutil_x86_avx_256_double.h"
53 * Gromacs nonbonded kernel: nb_kernel_ElecRFCut_VdwLJSw_GeomW4P1_VF_avx_256_double
54 * Electrostatics interaction: ReactionField
55 * VdW interaction: LennardJones
56 * Geometry: Water4-Particle
57 * Calculate force/pot: PotentialAndForce
60 nb_kernel_ElecRFCut_VdwLJSw_GeomW4P1_VF_avx_256_double
61 (t_nblist * gmx_restrict nlist,
62 rvec * gmx_restrict xx,
63 rvec * gmx_restrict ff,
64 t_forcerec * gmx_restrict fr,
65 t_mdatoms * gmx_restrict mdatoms,
66 nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
67 t_nrnb * gmx_restrict nrnb)
69 /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
70 * just 0 for non-waters.
71 * Suffixes A,B,C,D refer to j loop unrolling done with AVX, e.g. for the four different
72 * jnr indices corresponding to data put in the four positions in the SIMD register.
74 int i_shift_offset,i_coord_offset,outeriter,inneriter;
75 int j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
76 int jnrA,jnrB,jnrC,jnrD;
77 int jnrlistA,jnrlistB,jnrlistC,jnrlistD;
78 int jnrlistE,jnrlistF,jnrlistG,jnrlistH;
79 int j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
80 int *iinr,*jindex,*jjnr,*shiftidx,*gid;
82 real *shiftvec,*fshift,*x,*f;
83 real *fjptrA,*fjptrB,*fjptrC,*fjptrD;
85 __m256d tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
86 real * vdwioffsetptr0;
87 __m256d ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
88 real * vdwioffsetptr1;
89 __m256d ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
90 real * vdwioffsetptr2;
91 __m256d ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
92 real * vdwioffsetptr3;
93 __m256d ix3,iy3,iz3,fix3,fiy3,fiz3,iq3,isai3;
94 int vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
95 __m256d jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
96 __m256d dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
97 __m256d dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
98 __m256d dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
99 __m256d dx30,dy30,dz30,rsq30,rinv30,rinvsq30,r30,qq30,c6_30,c12_30;
100 __m256d velec,felec,velecsum,facel,crf,krf,krf2;
103 __m256d rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
106 __m256d one_sixth = _mm256_set1_pd(1.0/6.0);
107 __m256d one_twelfth = _mm256_set1_pd(1.0/12.0);
108 __m256d rswitch,swV3,swV4,swV5,swF2,swF3,swF4,d,d2,sw,dsw;
109 real rswitch_scalar,d_scalar;
110 __m256d dummy_mask,cutoff_mask;
111 __m128 tmpmask0,tmpmask1;
112 __m256d signbit = _mm256_castsi256_pd( _mm256_set1_epi32(0x80000000) );
113 __m256d one = _mm256_set1_pd(1.0);
114 __m256d two = _mm256_set1_pd(2.0);
120 jindex = nlist->jindex;
122 shiftidx = nlist->shift;
124 shiftvec = fr->shift_vec[0];
125 fshift = fr->fshift[0];
126 facel = _mm256_set1_pd(fr->epsfac);
127 charge = mdatoms->chargeA;
128 krf = _mm256_set1_pd(fr->ic->k_rf);
129 krf2 = _mm256_set1_pd(fr->ic->k_rf*2.0);
130 crf = _mm256_set1_pd(fr->ic->c_rf);
131 nvdwtype = fr->ntype;
133 vdwtype = mdatoms->typeA;
135 /* Setup water-specific parameters */
136 inr = nlist->iinr[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 iq3 = _mm256_mul_pd(facel,_mm256_set1_pd(charge[inr+3]));
140 vdwioffsetptr0 = vdwparam+2*nvdwtype*vdwtype[inr+0];
142 /* When we use explicit cutoffs the value must be identical for elec and VdW, so use elec as an arbitrary choice */
143 rcutoff_scalar = fr->rcoulomb;
144 rcutoff = _mm256_set1_pd(rcutoff_scalar);
145 rcutoff2 = _mm256_mul_pd(rcutoff,rcutoff);
147 rswitch_scalar = fr->rvdw_switch;
148 rswitch = _mm256_set1_pd(rswitch_scalar);
149 /* Setup switch parameters */
150 d_scalar = rcutoff_scalar-rswitch_scalar;
151 d = _mm256_set1_pd(d_scalar);
152 swV3 = _mm256_set1_pd(-10.0/(d_scalar*d_scalar*d_scalar));
153 swV4 = _mm256_set1_pd( 15.0/(d_scalar*d_scalar*d_scalar*d_scalar));
154 swV5 = _mm256_set1_pd( -6.0/(d_scalar*d_scalar*d_scalar*d_scalar*d_scalar));
155 swF2 = _mm256_set1_pd(-30.0/(d_scalar*d_scalar*d_scalar));
156 swF3 = _mm256_set1_pd( 60.0/(d_scalar*d_scalar*d_scalar*d_scalar));
157 swF4 = _mm256_set1_pd(-30.0/(d_scalar*d_scalar*d_scalar*d_scalar*d_scalar));
159 /* Avoid stupid compiler warnings */
160 jnrA = jnrB = jnrC = jnrD = 0;
169 for(iidx=0;iidx<4*DIM;iidx++)
174 /* Start outer loop over neighborlists */
175 for(iidx=0; iidx<nri; iidx++)
177 /* Load shift vector for this list */
178 i_shift_offset = DIM*shiftidx[iidx];
180 /* Load limits for loop over neighbors */
181 j_index_start = jindex[iidx];
182 j_index_end = jindex[iidx+1];
184 /* Get outer coordinate index */
186 i_coord_offset = DIM*inr;
188 /* Load i particle coords and add shift vector */
189 gmx_mm256_load_shift_and_4rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset,
190 &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2,&ix3,&iy3,&iz3);
192 fix0 = _mm256_setzero_pd();
193 fiy0 = _mm256_setzero_pd();
194 fiz0 = _mm256_setzero_pd();
195 fix1 = _mm256_setzero_pd();
196 fiy1 = _mm256_setzero_pd();
197 fiz1 = _mm256_setzero_pd();
198 fix2 = _mm256_setzero_pd();
199 fiy2 = _mm256_setzero_pd();
200 fiz2 = _mm256_setzero_pd();
201 fix3 = _mm256_setzero_pd();
202 fiy3 = _mm256_setzero_pd();
203 fiz3 = _mm256_setzero_pd();
205 /* Reset potential sums */
206 velecsum = _mm256_setzero_pd();
207 vvdwsum = _mm256_setzero_pd();
209 /* Start inner kernel loop */
210 for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
213 /* Get j neighbor index, and coordinate index */
218 j_coord_offsetA = DIM*jnrA;
219 j_coord_offsetB = DIM*jnrB;
220 j_coord_offsetC = DIM*jnrC;
221 j_coord_offsetD = DIM*jnrD;
223 /* load j atom coordinates */
224 gmx_mm256_load_1rvec_4ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
225 x+j_coord_offsetC,x+j_coord_offsetD,
228 /* Calculate displacement vector */
229 dx00 = _mm256_sub_pd(ix0,jx0);
230 dy00 = _mm256_sub_pd(iy0,jy0);
231 dz00 = _mm256_sub_pd(iz0,jz0);
232 dx10 = _mm256_sub_pd(ix1,jx0);
233 dy10 = _mm256_sub_pd(iy1,jy0);
234 dz10 = _mm256_sub_pd(iz1,jz0);
235 dx20 = _mm256_sub_pd(ix2,jx0);
236 dy20 = _mm256_sub_pd(iy2,jy0);
237 dz20 = _mm256_sub_pd(iz2,jz0);
238 dx30 = _mm256_sub_pd(ix3,jx0);
239 dy30 = _mm256_sub_pd(iy3,jy0);
240 dz30 = _mm256_sub_pd(iz3,jz0);
242 /* Calculate squared distance and things based on it */
243 rsq00 = gmx_mm256_calc_rsq_pd(dx00,dy00,dz00);
244 rsq10 = gmx_mm256_calc_rsq_pd(dx10,dy10,dz10);
245 rsq20 = gmx_mm256_calc_rsq_pd(dx20,dy20,dz20);
246 rsq30 = gmx_mm256_calc_rsq_pd(dx30,dy30,dz30);
248 rinv00 = gmx_mm256_invsqrt_pd(rsq00);
249 rinv10 = gmx_mm256_invsqrt_pd(rsq10);
250 rinv20 = gmx_mm256_invsqrt_pd(rsq20);
251 rinv30 = gmx_mm256_invsqrt_pd(rsq30);
253 rinvsq00 = _mm256_mul_pd(rinv00,rinv00);
254 rinvsq10 = _mm256_mul_pd(rinv10,rinv10);
255 rinvsq20 = _mm256_mul_pd(rinv20,rinv20);
256 rinvsq30 = _mm256_mul_pd(rinv30,rinv30);
258 /* Load parameters for j particles */
259 jq0 = gmx_mm256_load_4real_swizzle_pd(charge+jnrA+0,charge+jnrB+0,
260 charge+jnrC+0,charge+jnrD+0);
261 vdwjidx0A = 2*vdwtype[jnrA+0];
262 vdwjidx0B = 2*vdwtype[jnrB+0];
263 vdwjidx0C = 2*vdwtype[jnrC+0];
264 vdwjidx0D = 2*vdwtype[jnrD+0];
266 fjx0 = _mm256_setzero_pd();
267 fjy0 = _mm256_setzero_pd();
268 fjz0 = _mm256_setzero_pd();
270 /**************************
271 * CALCULATE INTERACTIONS *
272 **************************/
274 if (gmx_mm256_any_lt(rsq00,rcutoff2))
277 r00 = _mm256_mul_pd(rsq00,rinv00);
279 /* Compute parameters for interactions between i and j atoms */
280 gmx_mm256_load_4pair_swizzle_pd(vdwioffsetptr0+vdwjidx0A,
281 vdwioffsetptr0+vdwjidx0B,
282 vdwioffsetptr0+vdwjidx0C,
283 vdwioffsetptr0+vdwjidx0D,
286 /* LENNARD-JONES DISPERSION/REPULSION */
288 rinvsix = _mm256_mul_pd(_mm256_mul_pd(rinvsq00,rinvsq00),rinvsq00);
289 vvdw6 = _mm256_mul_pd(c6_00,rinvsix);
290 vvdw12 = _mm256_mul_pd(c12_00,_mm256_mul_pd(rinvsix,rinvsix));
291 vvdw = _mm256_sub_pd( _mm256_mul_pd(vvdw12,one_twelfth) , _mm256_mul_pd(vvdw6,one_sixth) );
292 fvdw = _mm256_mul_pd(_mm256_sub_pd(vvdw12,vvdw6),rinvsq00);
294 d = _mm256_sub_pd(r00,rswitch);
295 d = _mm256_max_pd(d,_mm256_setzero_pd());
296 d2 = _mm256_mul_pd(d,d);
297 sw = _mm256_add_pd(one,_mm256_mul_pd(d2,_mm256_mul_pd(d,_mm256_add_pd(swV3,_mm256_mul_pd(d,_mm256_add_pd(swV4,_mm256_mul_pd(d,swV5)))))));
299 dsw = _mm256_mul_pd(d2,_mm256_add_pd(swF2,_mm256_mul_pd(d,_mm256_add_pd(swF3,_mm256_mul_pd(d,swF4)))));
301 /* Evaluate switch function */
302 /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
303 fvdw = _mm256_sub_pd( _mm256_mul_pd(fvdw,sw) , _mm256_mul_pd(rinv00,_mm256_mul_pd(vvdw,dsw)) );
304 vvdw = _mm256_mul_pd(vvdw,sw);
305 cutoff_mask = _mm256_cmp_pd(rsq00,rcutoff2,_CMP_LT_OQ);
307 /* Update potential sum for this i atom from the interaction with this j atom. */
308 vvdw = _mm256_and_pd(vvdw,cutoff_mask);
309 vvdwsum = _mm256_add_pd(vvdwsum,vvdw);
313 fscal = _mm256_and_pd(fscal,cutoff_mask);
315 /* Calculate temporary vectorial force */
316 tx = _mm256_mul_pd(fscal,dx00);
317 ty = _mm256_mul_pd(fscal,dy00);
318 tz = _mm256_mul_pd(fscal,dz00);
320 /* Update vectorial force */
321 fix0 = _mm256_add_pd(fix0,tx);
322 fiy0 = _mm256_add_pd(fiy0,ty);
323 fiz0 = _mm256_add_pd(fiz0,tz);
325 fjx0 = _mm256_add_pd(fjx0,tx);
326 fjy0 = _mm256_add_pd(fjy0,ty);
327 fjz0 = _mm256_add_pd(fjz0,tz);
331 /**************************
332 * CALCULATE INTERACTIONS *
333 **************************/
335 if (gmx_mm256_any_lt(rsq10,rcutoff2))
338 /* Compute parameters for interactions between i and j atoms */
339 qq10 = _mm256_mul_pd(iq1,jq0);
341 /* REACTION-FIELD ELECTROSTATICS */
342 velec = _mm256_mul_pd(qq10,_mm256_sub_pd(_mm256_add_pd(rinv10,_mm256_mul_pd(krf,rsq10)),crf));
343 felec = _mm256_mul_pd(qq10,_mm256_sub_pd(_mm256_mul_pd(rinv10,rinvsq10),krf2));
345 cutoff_mask = _mm256_cmp_pd(rsq10,rcutoff2,_CMP_LT_OQ);
347 /* Update potential sum for this i atom from the interaction with this j atom. */
348 velec = _mm256_and_pd(velec,cutoff_mask);
349 velecsum = _mm256_add_pd(velecsum,velec);
353 fscal = _mm256_and_pd(fscal,cutoff_mask);
355 /* Calculate temporary vectorial force */
356 tx = _mm256_mul_pd(fscal,dx10);
357 ty = _mm256_mul_pd(fscal,dy10);
358 tz = _mm256_mul_pd(fscal,dz10);
360 /* Update vectorial force */
361 fix1 = _mm256_add_pd(fix1,tx);
362 fiy1 = _mm256_add_pd(fiy1,ty);
363 fiz1 = _mm256_add_pd(fiz1,tz);
365 fjx0 = _mm256_add_pd(fjx0,tx);
366 fjy0 = _mm256_add_pd(fjy0,ty);
367 fjz0 = _mm256_add_pd(fjz0,tz);
371 /**************************
372 * CALCULATE INTERACTIONS *
373 **************************/
375 if (gmx_mm256_any_lt(rsq20,rcutoff2))
378 /* Compute parameters for interactions between i and j atoms */
379 qq20 = _mm256_mul_pd(iq2,jq0);
381 /* REACTION-FIELD ELECTROSTATICS */
382 velec = _mm256_mul_pd(qq20,_mm256_sub_pd(_mm256_add_pd(rinv20,_mm256_mul_pd(krf,rsq20)),crf));
383 felec = _mm256_mul_pd(qq20,_mm256_sub_pd(_mm256_mul_pd(rinv20,rinvsq20),krf2));
385 cutoff_mask = _mm256_cmp_pd(rsq20,rcutoff2,_CMP_LT_OQ);
387 /* Update potential sum for this i atom from the interaction with this j atom. */
388 velec = _mm256_and_pd(velec,cutoff_mask);
389 velecsum = _mm256_add_pd(velecsum,velec);
393 fscal = _mm256_and_pd(fscal,cutoff_mask);
395 /* Calculate temporary vectorial force */
396 tx = _mm256_mul_pd(fscal,dx20);
397 ty = _mm256_mul_pd(fscal,dy20);
398 tz = _mm256_mul_pd(fscal,dz20);
400 /* Update vectorial force */
401 fix2 = _mm256_add_pd(fix2,tx);
402 fiy2 = _mm256_add_pd(fiy2,ty);
403 fiz2 = _mm256_add_pd(fiz2,tz);
405 fjx0 = _mm256_add_pd(fjx0,tx);
406 fjy0 = _mm256_add_pd(fjy0,ty);
407 fjz0 = _mm256_add_pd(fjz0,tz);
411 /**************************
412 * CALCULATE INTERACTIONS *
413 **************************/
415 if (gmx_mm256_any_lt(rsq30,rcutoff2))
418 /* Compute parameters for interactions between i and j atoms */
419 qq30 = _mm256_mul_pd(iq3,jq0);
421 /* REACTION-FIELD ELECTROSTATICS */
422 velec = _mm256_mul_pd(qq30,_mm256_sub_pd(_mm256_add_pd(rinv30,_mm256_mul_pd(krf,rsq30)),crf));
423 felec = _mm256_mul_pd(qq30,_mm256_sub_pd(_mm256_mul_pd(rinv30,rinvsq30),krf2));
425 cutoff_mask = _mm256_cmp_pd(rsq30,rcutoff2,_CMP_LT_OQ);
427 /* Update potential sum for this i atom from the interaction with this j atom. */
428 velec = _mm256_and_pd(velec,cutoff_mask);
429 velecsum = _mm256_add_pd(velecsum,velec);
433 fscal = _mm256_and_pd(fscal,cutoff_mask);
435 /* Calculate temporary vectorial force */
436 tx = _mm256_mul_pd(fscal,dx30);
437 ty = _mm256_mul_pd(fscal,dy30);
438 tz = _mm256_mul_pd(fscal,dz30);
440 /* Update vectorial force */
441 fix3 = _mm256_add_pd(fix3,tx);
442 fiy3 = _mm256_add_pd(fiy3,ty);
443 fiz3 = _mm256_add_pd(fiz3,tz);
445 fjx0 = _mm256_add_pd(fjx0,tx);
446 fjy0 = _mm256_add_pd(fjy0,ty);
447 fjz0 = _mm256_add_pd(fjz0,tz);
451 fjptrA = f+j_coord_offsetA;
452 fjptrB = f+j_coord_offsetB;
453 fjptrC = f+j_coord_offsetC;
454 fjptrD = f+j_coord_offsetD;
456 gmx_mm256_decrement_1rvec_4ptr_swizzle_pd(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
458 /* Inner loop uses 170 flops */
464 /* Get j neighbor index, and coordinate index */
465 jnrlistA = jjnr[jidx];
466 jnrlistB = jjnr[jidx+1];
467 jnrlistC = jjnr[jidx+2];
468 jnrlistD = jjnr[jidx+3];
469 /* Sign of each element will be negative for non-real atoms.
470 * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
471 * so use it as val = _mm_andnot_pd(mask,val) to clear dummy entries.
473 tmpmask0 = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
475 tmpmask1 = _mm_permute_ps(tmpmask0,_GMX_MM_PERMUTE(3,3,2,2));
476 tmpmask0 = _mm_permute_ps(tmpmask0,_GMX_MM_PERMUTE(1,1,0,0));
477 dummy_mask = _mm256_castps_pd(gmx_mm256_set_m128(tmpmask1,tmpmask0));
479 jnrA = (jnrlistA>=0) ? jnrlistA : 0;
480 jnrB = (jnrlistB>=0) ? jnrlistB : 0;
481 jnrC = (jnrlistC>=0) ? jnrlistC : 0;
482 jnrD = (jnrlistD>=0) ? jnrlistD : 0;
483 j_coord_offsetA = DIM*jnrA;
484 j_coord_offsetB = DIM*jnrB;
485 j_coord_offsetC = DIM*jnrC;
486 j_coord_offsetD = DIM*jnrD;
488 /* load j atom coordinates */
489 gmx_mm256_load_1rvec_4ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
490 x+j_coord_offsetC,x+j_coord_offsetD,
493 /* Calculate displacement vector */
494 dx00 = _mm256_sub_pd(ix0,jx0);
495 dy00 = _mm256_sub_pd(iy0,jy0);
496 dz00 = _mm256_sub_pd(iz0,jz0);
497 dx10 = _mm256_sub_pd(ix1,jx0);
498 dy10 = _mm256_sub_pd(iy1,jy0);
499 dz10 = _mm256_sub_pd(iz1,jz0);
500 dx20 = _mm256_sub_pd(ix2,jx0);
501 dy20 = _mm256_sub_pd(iy2,jy0);
502 dz20 = _mm256_sub_pd(iz2,jz0);
503 dx30 = _mm256_sub_pd(ix3,jx0);
504 dy30 = _mm256_sub_pd(iy3,jy0);
505 dz30 = _mm256_sub_pd(iz3,jz0);
507 /* Calculate squared distance and things based on it */
508 rsq00 = gmx_mm256_calc_rsq_pd(dx00,dy00,dz00);
509 rsq10 = gmx_mm256_calc_rsq_pd(dx10,dy10,dz10);
510 rsq20 = gmx_mm256_calc_rsq_pd(dx20,dy20,dz20);
511 rsq30 = gmx_mm256_calc_rsq_pd(dx30,dy30,dz30);
513 rinv00 = gmx_mm256_invsqrt_pd(rsq00);
514 rinv10 = gmx_mm256_invsqrt_pd(rsq10);
515 rinv20 = gmx_mm256_invsqrt_pd(rsq20);
516 rinv30 = gmx_mm256_invsqrt_pd(rsq30);
518 rinvsq00 = _mm256_mul_pd(rinv00,rinv00);
519 rinvsq10 = _mm256_mul_pd(rinv10,rinv10);
520 rinvsq20 = _mm256_mul_pd(rinv20,rinv20);
521 rinvsq30 = _mm256_mul_pd(rinv30,rinv30);
523 /* Load parameters for j particles */
524 jq0 = gmx_mm256_load_4real_swizzle_pd(charge+jnrA+0,charge+jnrB+0,
525 charge+jnrC+0,charge+jnrD+0);
526 vdwjidx0A = 2*vdwtype[jnrA+0];
527 vdwjidx0B = 2*vdwtype[jnrB+0];
528 vdwjidx0C = 2*vdwtype[jnrC+0];
529 vdwjidx0D = 2*vdwtype[jnrD+0];
531 fjx0 = _mm256_setzero_pd();
532 fjy0 = _mm256_setzero_pd();
533 fjz0 = _mm256_setzero_pd();
535 /**************************
536 * CALCULATE INTERACTIONS *
537 **************************/
539 if (gmx_mm256_any_lt(rsq00,rcutoff2))
542 r00 = _mm256_mul_pd(rsq00,rinv00);
543 r00 = _mm256_andnot_pd(dummy_mask,r00);
545 /* Compute parameters for interactions between i and j atoms */
546 gmx_mm256_load_4pair_swizzle_pd(vdwioffsetptr0+vdwjidx0A,
547 vdwioffsetptr0+vdwjidx0B,
548 vdwioffsetptr0+vdwjidx0C,
549 vdwioffsetptr0+vdwjidx0D,
552 /* LENNARD-JONES DISPERSION/REPULSION */
554 rinvsix = _mm256_mul_pd(_mm256_mul_pd(rinvsq00,rinvsq00),rinvsq00);
555 vvdw6 = _mm256_mul_pd(c6_00,rinvsix);
556 vvdw12 = _mm256_mul_pd(c12_00,_mm256_mul_pd(rinvsix,rinvsix));
557 vvdw = _mm256_sub_pd( _mm256_mul_pd(vvdw12,one_twelfth) , _mm256_mul_pd(vvdw6,one_sixth) );
558 fvdw = _mm256_mul_pd(_mm256_sub_pd(vvdw12,vvdw6),rinvsq00);
560 d = _mm256_sub_pd(r00,rswitch);
561 d = _mm256_max_pd(d,_mm256_setzero_pd());
562 d2 = _mm256_mul_pd(d,d);
563 sw = _mm256_add_pd(one,_mm256_mul_pd(d2,_mm256_mul_pd(d,_mm256_add_pd(swV3,_mm256_mul_pd(d,_mm256_add_pd(swV4,_mm256_mul_pd(d,swV5)))))));
565 dsw = _mm256_mul_pd(d2,_mm256_add_pd(swF2,_mm256_mul_pd(d,_mm256_add_pd(swF3,_mm256_mul_pd(d,swF4)))));
567 /* Evaluate switch function */
568 /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
569 fvdw = _mm256_sub_pd( _mm256_mul_pd(fvdw,sw) , _mm256_mul_pd(rinv00,_mm256_mul_pd(vvdw,dsw)) );
570 vvdw = _mm256_mul_pd(vvdw,sw);
571 cutoff_mask = _mm256_cmp_pd(rsq00,rcutoff2,_CMP_LT_OQ);
573 /* Update potential sum for this i atom from the interaction with this j atom. */
574 vvdw = _mm256_and_pd(vvdw,cutoff_mask);
575 vvdw = _mm256_andnot_pd(dummy_mask,vvdw);
576 vvdwsum = _mm256_add_pd(vvdwsum,vvdw);
580 fscal = _mm256_and_pd(fscal,cutoff_mask);
582 fscal = _mm256_andnot_pd(dummy_mask,fscal);
584 /* Calculate temporary vectorial force */
585 tx = _mm256_mul_pd(fscal,dx00);
586 ty = _mm256_mul_pd(fscal,dy00);
587 tz = _mm256_mul_pd(fscal,dz00);
589 /* Update vectorial force */
590 fix0 = _mm256_add_pd(fix0,tx);
591 fiy0 = _mm256_add_pd(fiy0,ty);
592 fiz0 = _mm256_add_pd(fiz0,tz);
594 fjx0 = _mm256_add_pd(fjx0,tx);
595 fjy0 = _mm256_add_pd(fjy0,ty);
596 fjz0 = _mm256_add_pd(fjz0,tz);
600 /**************************
601 * CALCULATE INTERACTIONS *
602 **************************/
604 if (gmx_mm256_any_lt(rsq10,rcutoff2))
607 /* Compute parameters for interactions between i and j atoms */
608 qq10 = _mm256_mul_pd(iq1,jq0);
610 /* REACTION-FIELD ELECTROSTATICS */
611 velec = _mm256_mul_pd(qq10,_mm256_sub_pd(_mm256_add_pd(rinv10,_mm256_mul_pd(krf,rsq10)),crf));
612 felec = _mm256_mul_pd(qq10,_mm256_sub_pd(_mm256_mul_pd(rinv10,rinvsq10),krf2));
614 cutoff_mask = _mm256_cmp_pd(rsq10,rcutoff2,_CMP_LT_OQ);
616 /* Update potential sum for this i atom from the interaction with this j atom. */
617 velec = _mm256_and_pd(velec,cutoff_mask);
618 velec = _mm256_andnot_pd(dummy_mask,velec);
619 velecsum = _mm256_add_pd(velecsum,velec);
623 fscal = _mm256_and_pd(fscal,cutoff_mask);
625 fscal = _mm256_andnot_pd(dummy_mask,fscal);
627 /* Calculate temporary vectorial force */
628 tx = _mm256_mul_pd(fscal,dx10);
629 ty = _mm256_mul_pd(fscal,dy10);
630 tz = _mm256_mul_pd(fscal,dz10);
632 /* Update vectorial force */
633 fix1 = _mm256_add_pd(fix1,tx);
634 fiy1 = _mm256_add_pd(fiy1,ty);
635 fiz1 = _mm256_add_pd(fiz1,tz);
637 fjx0 = _mm256_add_pd(fjx0,tx);
638 fjy0 = _mm256_add_pd(fjy0,ty);
639 fjz0 = _mm256_add_pd(fjz0,tz);
643 /**************************
644 * CALCULATE INTERACTIONS *
645 **************************/
647 if (gmx_mm256_any_lt(rsq20,rcutoff2))
650 /* Compute parameters for interactions between i and j atoms */
651 qq20 = _mm256_mul_pd(iq2,jq0);
653 /* REACTION-FIELD ELECTROSTATICS */
654 velec = _mm256_mul_pd(qq20,_mm256_sub_pd(_mm256_add_pd(rinv20,_mm256_mul_pd(krf,rsq20)),crf));
655 felec = _mm256_mul_pd(qq20,_mm256_sub_pd(_mm256_mul_pd(rinv20,rinvsq20),krf2));
657 cutoff_mask = _mm256_cmp_pd(rsq20,rcutoff2,_CMP_LT_OQ);
659 /* Update potential sum for this i atom from the interaction with this j atom. */
660 velec = _mm256_and_pd(velec,cutoff_mask);
661 velec = _mm256_andnot_pd(dummy_mask,velec);
662 velecsum = _mm256_add_pd(velecsum,velec);
666 fscal = _mm256_and_pd(fscal,cutoff_mask);
668 fscal = _mm256_andnot_pd(dummy_mask,fscal);
670 /* Calculate temporary vectorial force */
671 tx = _mm256_mul_pd(fscal,dx20);
672 ty = _mm256_mul_pd(fscal,dy20);
673 tz = _mm256_mul_pd(fscal,dz20);
675 /* Update vectorial force */
676 fix2 = _mm256_add_pd(fix2,tx);
677 fiy2 = _mm256_add_pd(fiy2,ty);
678 fiz2 = _mm256_add_pd(fiz2,tz);
680 fjx0 = _mm256_add_pd(fjx0,tx);
681 fjy0 = _mm256_add_pd(fjy0,ty);
682 fjz0 = _mm256_add_pd(fjz0,tz);
686 /**************************
687 * CALCULATE INTERACTIONS *
688 **************************/
690 if (gmx_mm256_any_lt(rsq30,rcutoff2))
693 /* Compute parameters for interactions between i and j atoms */
694 qq30 = _mm256_mul_pd(iq3,jq0);
696 /* REACTION-FIELD ELECTROSTATICS */
697 velec = _mm256_mul_pd(qq30,_mm256_sub_pd(_mm256_add_pd(rinv30,_mm256_mul_pd(krf,rsq30)),crf));
698 felec = _mm256_mul_pd(qq30,_mm256_sub_pd(_mm256_mul_pd(rinv30,rinvsq30),krf2));
700 cutoff_mask = _mm256_cmp_pd(rsq30,rcutoff2,_CMP_LT_OQ);
702 /* Update potential sum for this i atom from the interaction with this j atom. */
703 velec = _mm256_and_pd(velec,cutoff_mask);
704 velec = _mm256_andnot_pd(dummy_mask,velec);
705 velecsum = _mm256_add_pd(velecsum,velec);
709 fscal = _mm256_and_pd(fscal,cutoff_mask);
711 fscal = _mm256_andnot_pd(dummy_mask,fscal);
713 /* Calculate temporary vectorial force */
714 tx = _mm256_mul_pd(fscal,dx30);
715 ty = _mm256_mul_pd(fscal,dy30);
716 tz = _mm256_mul_pd(fscal,dz30);
718 /* Update vectorial force */
719 fix3 = _mm256_add_pd(fix3,tx);
720 fiy3 = _mm256_add_pd(fiy3,ty);
721 fiz3 = _mm256_add_pd(fiz3,tz);
723 fjx0 = _mm256_add_pd(fjx0,tx);
724 fjy0 = _mm256_add_pd(fjy0,ty);
725 fjz0 = _mm256_add_pd(fjz0,tz);
729 fjptrA = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
730 fjptrB = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
731 fjptrC = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
732 fjptrD = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
734 gmx_mm256_decrement_1rvec_4ptr_swizzle_pd(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
736 /* Inner loop uses 171 flops */
739 /* End of innermost loop */
741 gmx_mm256_update_iforce_4atom_swizzle_pd(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,fix3,fiy3,fiz3,
742 f+i_coord_offset,fshift+i_shift_offset);
745 /* Update potential energies */
746 gmx_mm256_update_1pot_pd(velecsum,kernel_data->energygrp_elec+ggid);
747 gmx_mm256_update_1pot_pd(vvdwsum,kernel_data->energygrp_vdw+ggid);
749 /* Increment number of inner iterations */
750 inneriter += j_index_end - j_index_start;
752 /* Outer loop uses 26 flops */
755 /* Increment number of outer iterations */
758 /* Update outer/inner flops */
760 inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W4_VF,outeriter*26 + inneriter*171);
763 * Gromacs nonbonded kernel: nb_kernel_ElecRFCut_VdwLJSw_GeomW4P1_F_avx_256_double
764 * Electrostatics interaction: ReactionField
765 * VdW interaction: LennardJones
766 * Geometry: Water4-Particle
767 * Calculate force/pot: Force
770 nb_kernel_ElecRFCut_VdwLJSw_GeomW4P1_F_avx_256_double
771 (t_nblist * gmx_restrict nlist,
772 rvec * gmx_restrict xx,
773 rvec * gmx_restrict ff,
774 t_forcerec * gmx_restrict fr,
775 t_mdatoms * gmx_restrict mdatoms,
776 nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
777 t_nrnb * gmx_restrict nrnb)
779 /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
780 * just 0 for non-waters.
781 * Suffixes A,B,C,D refer to j loop unrolling done with AVX, e.g. for the four different
782 * jnr indices corresponding to data put in the four positions in the SIMD register.
784 int i_shift_offset,i_coord_offset,outeriter,inneriter;
785 int j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
786 int jnrA,jnrB,jnrC,jnrD;
787 int jnrlistA,jnrlistB,jnrlistC,jnrlistD;
788 int jnrlistE,jnrlistF,jnrlistG,jnrlistH;
789 int j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
790 int *iinr,*jindex,*jjnr,*shiftidx,*gid;
792 real *shiftvec,*fshift,*x,*f;
793 real *fjptrA,*fjptrB,*fjptrC,*fjptrD;
795 __m256d tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
796 real * vdwioffsetptr0;
797 __m256d ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
798 real * vdwioffsetptr1;
799 __m256d ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
800 real * vdwioffsetptr2;
801 __m256d ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
802 real * vdwioffsetptr3;
803 __m256d ix3,iy3,iz3,fix3,fiy3,fiz3,iq3,isai3;
804 int vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
805 __m256d jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
806 __m256d dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
807 __m256d dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
808 __m256d dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
809 __m256d dx30,dy30,dz30,rsq30,rinv30,rinvsq30,r30,qq30,c6_30,c12_30;
810 __m256d velec,felec,velecsum,facel,crf,krf,krf2;
813 __m256d rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
816 __m256d one_sixth = _mm256_set1_pd(1.0/6.0);
817 __m256d one_twelfth = _mm256_set1_pd(1.0/12.0);
818 __m256d rswitch,swV3,swV4,swV5,swF2,swF3,swF4,d,d2,sw,dsw;
819 real rswitch_scalar,d_scalar;
820 __m256d dummy_mask,cutoff_mask;
821 __m128 tmpmask0,tmpmask1;
822 __m256d signbit = _mm256_castsi256_pd( _mm256_set1_epi32(0x80000000) );
823 __m256d one = _mm256_set1_pd(1.0);
824 __m256d two = _mm256_set1_pd(2.0);
830 jindex = nlist->jindex;
832 shiftidx = nlist->shift;
834 shiftvec = fr->shift_vec[0];
835 fshift = fr->fshift[0];
836 facel = _mm256_set1_pd(fr->epsfac);
837 charge = mdatoms->chargeA;
838 krf = _mm256_set1_pd(fr->ic->k_rf);
839 krf2 = _mm256_set1_pd(fr->ic->k_rf*2.0);
840 crf = _mm256_set1_pd(fr->ic->c_rf);
841 nvdwtype = fr->ntype;
843 vdwtype = mdatoms->typeA;
845 /* Setup water-specific parameters */
846 inr = nlist->iinr[0];
847 iq1 = _mm256_mul_pd(facel,_mm256_set1_pd(charge[inr+1]));
848 iq2 = _mm256_mul_pd(facel,_mm256_set1_pd(charge[inr+2]));
849 iq3 = _mm256_mul_pd(facel,_mm256_set1_pd(charge[inr+3]));
850 vdwioffsetptr0 = vdwparam+2*nvdwtype*vdwtype[inr+0];
852 /* When we use explicit cutoffs the value must be identical for elec and VdW, so use elec as an arbitrary choice */
853 rcutoff_scalar = fr->rcoulomb;
854 rcutoff = _mm256_set1_pd(rcutoff_scalar);
855 rcutoff2 = _mm256_mul_pd(rcutoff,rcutoff);
857 rswitch_scalar = fr->rvdw_switch;
858 rswitch = _mm256_set1_pd(rswitch_scalar);
859 /* Setup switch parameters */
860 d_scalar = rcutoff_scalar-rswitch_scalar;
861 d = _mm256_set1_pd(d_scalar);
862 swV3 = _mm256_set1_pd(-10.0/(d_scalar*d_scalar*d_scalar));
863 swV4 = _mm256_set1_pd( 15.0/(d_scalar*d_scalar*d_scalar*d_scalar));
864 swV5 = _mm256_set1_pd( -6.0/(d_scalar*d_scalar*d_scalar*d_scalar*d_scalar));
865 swF2 = _mm256_set1_pd(-30.0/(d_scalar*d_scalar*d_scalar));
866 swF3 = _mm256_set1_pd( 60.0/(d_scalar*d_scalar*d_scalar*d_scalar));
867 swF4 = _mm256_set1_pd(-30.0/(d_scalar*d_scalar*d_scalar*d_scalar*d_scalar));
869 /* Avoid stupid compiler warnings */
870 jnrA = jnrB = jnrC = jnrD = 0;
879 for(iidx=0;iidx<4*DIM;iidx++)
884 /* Start outer loop over neighborlists */
885 for(iidx=0; iidx<nri; iidx++)
887 /* Load shift vector for this list */
888 i_shift_offset = DIM*shiftidx[iidx];
890 /* Load limits for loop over neighbors */
891 j_index_start = jindex[iidx];
892 j_index_end = jindex[iidx+1];
894 /* Get outer coordinate index */
896 i_coord_offset = DIM*inr;
898 /* Load i particle coords and add shift vector */
899 gmx_mm256_load_shift_and_4rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset,
900 &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2,&ix3,&iy3,&iz3);
902 fix0 = _mm256_setzero_pd();
903 fiy0 = _mm256_setzero_pd();
904 fiz0 = _mm256_setzero_pd();
905 fix1 = _mm256_setzero_pd();
906 fiy1 = _mm256_setzero_pd();
907 fiz1 = _mm256_setzero_pd();
908 fix2 = _mm256_setzero_pd();
909 fiy2 = _mm256_setzero_pd();
910 fiz2 = _mm256_setzero_pd();
911 fix3 = _mm256_setzero_pd();
912 fiy3 = _mm256_setzero_pd();
913 fiz3 = _mm256_setzero_pd();
915 /* Start inner kernel loop */
916 for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
919 /* Get j neighbor index, and coordinate index */
924 j_coord_offsetA = DIM*jnrA;
925 j_coord_offsetB = DIM*jnrB;
926 j_coord_offsetC = DIM*jnrC;
927 j_coord_offsetD = DIM*jnrD;
929 /* load j atom coordinates */
930 gmx_mm256_load_1rvec_4ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
931 x+j_coord_offsetC,x+j_coord_offsetD,
934 /* Calculate displacement vector */
935 dx00 = _mm256_sub_pd(ix0,jx0);
936 dy00 = _mm256_sub_pd(iy0,jy0);
937 dz00 = _mm256_sub_pd(iz0,jz0);
938 dx10 = _mm256_sub_pd(ix1,jx0);
939 dy10 = _mm256_sub_pd(iy1,jy0);
940 dz10 = _mm256_sub_pd(iz1,jz0);
941 dx20 = _mm256_sub_pd(ix2,jx0);
942 dy20 = _mm256_sub_pd(iy2,jy0);
943 dz20 = _mm256_sub_pd(iz2,jz0);
944 dx30 = _mm256_sub_pd(ix3,jx0);
945 dy30 = _mm256_sub_pd(iy3,jy0);
946 dz30 = _mm256_sub_pd(iz3,jz0);
948 /* Calculate squared distance and things based on it */
949 rsq00 = gmx_mm256_calc_rsq_pd(dx00,dy00,dz00);
950 rsq10 = gmx_mm256_calc_rsq_pd(dx10,dy10,dz10);
951 rsq20 = gmx_mm256_calc_rsq_pd(dx20,dy20,dz20);
952 rsq30 = gmx_mm256_calc_rsq_pd(dx30,dy30,dz30);
954 rinv00 = gmx_mm256_invsqrt_pd(rsq00);
955 rinv10 = gmx_mm256_invsqrt_pd(rsq10);
956 rinv20 = gmx_mm256_invsqrt_pd(rsq20);
957 rinv30 = gmx_mm256_invsqrt_pd(rsq30);
959 rinvsq00 = _mm256_mul_pd(rinv00,rinv00);
960 rinvsq10 = _mm256_mul_pd(rinv10,rinv10);
961 rinvsq20 = _mm256_mul_pd(rinv20,rinv20);
962 rinvsq30 = _mm256_mul_pd(rinv30,rinv30);
964 /* Load parameters for j particles */
965 jq0 = gmx_mm256_load_4real_swizzle_pd(charge+jnrA+0,charge+jnrB+0,
966 charge+jnrC+0,charge+jnrD+0);
967 vdwjidx0A = 2*vdwtype[jnrA+0];
968 vdwjidx0B = 2*vdwtype[jnrB+0];
969 vdwjidx0C = 2*vdwtype[jnrC+0];
970 vdwjidx0D = 2*vdwtype[jnrD+0];
972 fjx0 = _mm256_setzero_pd();
973 fjy0 = _mm256_setzero_pd();
974 fjz0 = _mm256_setzero_pd();
976 /**************************
977 * CALCULATE INTERACTIONS *
978 **************************/
980 if (gmx_mm256_any_lt(rsq00,rcutoff2))
983 r00 = _mm256_mul_pd(rsq00,rinv00);
985 /* Compute parameters for interactions between i and j atoms */
986 gmx_mm256_load_4pair_swizzle_pd(vdwioffsetptr0+vdwjidx0A,
987 vdwioffsetptr0+vdwjidx0B,
988 vdwioffsetptr0+vdwjidx0C,
989 vdwioffsetptr0+vdwjidx0D,
992 /* LENNARD-JONES DISPERSION/REPULSION */
994 rinvsix = _mm256_mul_pd(_mm256_mul_pd(rinvsq00,rinvsq00),rinvsq00);
995 vvdw6 = _mm256_mul_pd(c6_00,rinvsix);
996 vvdw12 = _mm256_mul_pd(c12_00,_mm256_mul_pd(rinvsix,rinvsix));
997 vvdw = _mm256_sub_pd( _mm256_mul_pd(vvdw12,one_twelfth) , _mm256_mul_pd(vvdw6,one_sixth) );
998 fvdw = _mm256_mul_pd(_mm256_sub_pd(vvdw12,vvdw6),rinvsq00);
1000 d = _mm256_sub_pd(r00,rswitch);
1001 d = _mm256_max_pd(d,_mm256_setzero_pd());
1002 d2 = _mm256_mul_pd(d,d);
1003 sw = _mm256_add_pd(one,_mm256_mul_pd(d2,_mm256_mul_pd(d,_mm256_add_pd(swV3,_mm256_mul_pd(d,_mm256_add_pd(swV4,_mm256_mul_pd(d,swV5)))))));
1005 dsw = _mm256_mul_pd(d2,_mm256_add_pd(swF2,_mm256_mul_pd(d,_mm256_add_pd(swF3,_mm256_mul_pd(d,swF4)))));
1007 /* Evaluate switch function */
1008 /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
1009 fvdw = _mm256_sub_pd( _mm256_mul_pd(fvdw,sw) , _mm256_mul_pd(rinv00,_mm256_mul_pd(vvdw,dsw)) );
1010 cutoff_mask = _mm256_cmp_pd(rsq00,rcutoff2,_CMP_LT_OQ);
1014 fscal = _mm256_and_pd(fscal,cutoff_mask);
1016 /* Calculate temporary vectorial force */
1017 tx = _mm256_mul_pd(fscal,dx00);
1018 ty = _mm256_mul_pd(fscal,dy00);
1019 tz = _mm256_mul_pd(fscal,dz00);
1021 /* Update vectorial force */
1022 fix0 = _mm256_add_pd(fix0,tx);
1023 fiy0 = _mm256_add_pd(fiy0,ty);
1024 fiz0 = _mm256_add_pd(fiz0,tz);
1026 fjx0 = _mm256_add_pd(fjx0,tx);
1027 fjy0 = _mm256_add_pd(fjy0,ty);
1028 fjz0 = _mm256_add_pd(fjz0,tz);
1032 /**************************
1033 * CALCULATE INTERACTIONS *
1034 **************************/
1036 if (gmx_mm256_any_lt(rsq10,rcutoff2))
1039 /* Compute parameters for interactions between i and j atoms */
1040 qq10 = _mm256_mul_pd(iq1,jq0);
1042 /* REACTION-FIELD ELECTROSTATICS */
1043 felec = _mm256_mul_pd(qq10,_mm256_sub_pd(_mm256_mul_pd(rinv10,rinvsq10),krf2));
1045 cutoff_mask = _mm256_cmp_pd(rsq10,rcutoff2,_CMP_LT_OQ);
1049 fscal = _mm256_and_pd(fscal,cutoff_mask);
1051 /* Calculate temporary vectorial force */
1052 tx = _mm256_mul_pd(fscal,dx10);
1053 ty = _mm256_mul_pd(fscal,dy10);
1054 tz = _mm256_mul_pd(fscal,dz10);
1056 /* Update vectorial force */
1057 fix1 = _mm256_add_pd(fix1,tx);
1058 fiy1 = _mm256_add_pd(fiy1,ty);
1059 fiz1 = _mm256_add_pd(fiz1,tz);
1061 fjx0 = _mm256_add_pd(fjx0,tx);
1062 fjy0 = _mm256_add_pd(fjy0,ty);
1063 fjz0 = _mm256_add_pd(fjz0,tz);
1067 /**************************
1068 * CALCULATE INTERACTIONS *
1069 **************************/
1071 if (gmx_mm256_any_lt(rsq20,rcutoff2))
1074 /* Compute parameters for interactions between i and j atoms */
1075 qq20 = _mm256_mul_pd(iq2,jq0);
1077 /* REACTION-FIELD ELECTROSTATICS */
1078 felec = _mm256_mul_pd(qq20,_mm256_sub_pd(_mm256_mul_pd(rinv20,rinvsq20),krf2));
1080 cutoff_mask = _mm256_cmp_pd(rsq20,rcutoff2,_CMP_LT_OQ);
1084 fscal = _mm256_and_pd(fscal,cutoff_mask);
1086 /* Calculate temporary vectorial force */
1087 tx = _mm256_mul_pd(fscal,dx20);
1088 ty = _mm256_mul_pd(fscal,dy20);
1089 tz = _mm256_mul_pd(fscal,dz20);
1091 /* Update vectorial force */
1092 fix2 = _mm256_add_pd(fix2,tx);
1093 fiy2 = _mm256_add_pd(fiy2,ty);
1094 fiz2 = _mm256_add_pd(fiz2,tz);
1096 fjx0 = _mm256_add_pd(fjx0,tx);
1097 fjy0 = _mm256_add_pd(fjy0,ty);
1098 fjz0 = _mm256_add_pd(fjz0,tz);
1102 /**************************
1103 * CALCULATE INTERACTIONS *
1104 **************************/
1106 if (gmx_mm256_any_lt(rsq30,rcutoff2))
1109 /* Compute parameters for interactions between i and j atoms */
1110 qq30 = _mm256_mul_pd(iq3,jq0);
1112 /* REACTION-FIELD ELECTROSTATICS */
1113 felec = _mm256_mul_pd(qq30,_mm256_sub_pd(_mm256_mul_pd(rinv30,rinvsq30),krf2));
1115 cutoff_mask = _mm256_cmp_pd(rsq30,rcutoff2,_CMP_LT_OQ);
1119 fscal = _mm256_and_pd(fscal,cutoff_mask);
1121 /* Calculate temporary vectorial force */
1122 tx = _mm256_mul_pd(fscal,dx30);
1123 ty = _mm256_mul_pd(fscal,dy30);
1124 tz = _mm256_mul_pd(fscal,dz30);
1126 /* Update vectorial force */
1127 fix3 = _mm256_add_pd(fix3,tx);
1128 fiy3 = _mm256_add_pd(fiy3,ty);
1129 fiz3 = _mm256_add_pd(fiz3,tz);
1131 fjx0 = _mm256_add_pd(fjx0,tx);
1132 fjy0 = _mm256_add_pd(fjy0,ty);
1133 fjz0 = _mm256_add_pd(fjz0,tz);
1137 fjptrA = f+j_coord_offsetA;
1138 fjptrB = f+j_coord_offsetB;
1139 fjptrC = f+j_coord_offsetC;
1140 fjptrD = f+j_coord_offsetD;
1142 gmx_mm256_decrement_1rvec_4ptr_swizzle_pd(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
1144 /* Inner loop uses 149 flops */
1147 if(jidx<j_index_end)
1150 /* Get j neighbor index, and coordinate index */
1151 jnrlistA = jjnr[jidx];
1152 jnrlistB = jjnr[jidx+1];
1153 jnrlistC = jjnr[jidx+2];
1154 jnrlistD = jjnr[jidx+3];
1155 /* Sign of each element will be negative for non-real atoms.
1156 * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
1157 * so use it as val = _mm_andnot_pd(mask,val) to clear dummy entries.
1159 tmpmask0 = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
1161 tmpmask1 = _mm_permute_ps(tmpmask0,_GMX_MM_PERMUTE(3,3,2,2));
1162 tmpmask0 = _mm_permute_ps(tmpmask0,_GMX_MM_PERMUTE(1,1,0,0));
1163 dummy_mask = _mm256_castps_pd(gmx_mm256_set_m128(tmpmask1,tmpmask0));
1165 jnrA = (jnrlistA>=0) ? jnrlistA : 0;
1166 jnrB = (jnrlistB>=0) ? jnrlistB : 0;
1167 jnrC = (jnrlistC>=0) ? jnrlistC : 0;
1168 jnrD = (jnrlistD>=0) ? jnrlistD : 0;
1169 j_coord_offsetA = DIM*jnrA;
1170 j_coord_offsetB = DIM*jnrB;
1171 j_coord_offsetC = DIM*jnrC;
1172 j_coord_offsetD = DIM*jnrD;
1174 /* load j atom coordinates */
1175 gmx_mm256_load_1rvec_4ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
1176 x+j_coord_offsetC,x+j_coord_offsetD,
1179 /* Calculate displacement vector */
1180 dx00 = _mm256_sub_pd(ix0,jx0);
1181 dy00 = _mm256_sub_pd(iy0,jy0);
1182 dz00 = _mm256_sub_pd(iz0,jz0);
1183 dx10 = _mm256_sub_pd(ix1,jx0);
1184 dy10 = _mm256_sub_pd(iy1,jy0);
1185 dz10 = _mm256_sub_pd(iz1,jz0);
1186 dx20 = _mm256_sub_pd(ix2,jx0);
1187 dy20 = _mm256_sub_pd(iy2,jy0);
1188 dz20 = _mm256_sub_pd(iz2,jz0);
1189 dx30 = _mm256_sub_pd(ix3,jx0);
1190 dy30 = _mm256_sub_pd(iy3,jy0);
1191 dz30 = _mm256_sub_pd(iz3,jz0);
1193 /* Calculate squared distance and things based on it */
1194 rsq00 = gmx_mm256_calc_rsq_pd(dx00,dy00,dz00);
1195 rsq10 = gmx_mm256_calc_rsq_pd(dx10,dy10,dz10);
1196 rsq20 = gmx_mm256_calc_rsq_pd(dx20,dy20,dz20);
1197 rsq30 = gmx_mm256_calc_rsq_pd(dx30,dy30,dz30);
1199 rinv00 = gmx_mm256_invsqrt_pd(rsq00);
1200 rinv10 = gmx_mm256_invsqrt_pd(rsq10);
1201 rinv20 = gmx_mm256_invsqrt_pd(rsq20);
1202 rinv30 = gmx_mm256_invsqrt_pd(rsq30);
1204 rinvsq00 = _mm256_mul_pd(rinv00,rinv00);
1205 rinvsq10 = _mm256_mul_pd(rinv10,rinv10);
1206 rinvsq20 = _mm256_mul_pd(rinv20,rinv20);
1207 rinvsq30 = _mm256_mul_pd(rinv30,rinv30);
1209 /* Load parameters for j particles */
1210 jq0 = gmx_mm256_load_4real_swizzle_pd(charge+jnrA+0,charge+jnrB+0,
1211 charge+jnrC+0,charge+jnrD+0);
1212 vdwjidx0A = 2*vdwtype[jnrA+0];
1213 vdwjidx0B = 2*vdwtype[jnrB+0];
1214 vdwjidx0C = 2*vdwtype[jnrC+0];
1215 vdwjidx0D = 2*vdwtype[jnrD+0];
1217 fjx0 = _mm256_setzero_pd();
1218 fjy0 = _mm256_setzero_pd();
1219 fjz0 = _mm256_setzero_pd();
1221 /**************************
1222 * CALCULATE INTERACTIONS *
1223 **************************/
1225 if (gmx_mm256_any_lt(rsq00,rcutoff2))
1228 r00 = _mm256_mul_pd(rsq00,rinv00);
1229 r00 = _mm256_andnot_pd(dummy_mask,r00);
1231 /* Compute parameters for interactions between i and j atoms */
1232 gmx_mm256_load_4pair_swizzle_pd(vdwioffsetptr0+vdwjidx0A,
1233 vdwioffsetptr0+vdwjidx0B,
1234 vdwioffsetptr0+vdwjidx0C,
1235 vdwioffsetptr0+vdwjidx0D,
1238 /* LENNARD-JONES DISPERSION/REPULSION */
1240 rinvsix = _mm256_mul_pd(_mm256_mul_pd(rinvsq00,rinvsq00),rinvsq00);
1241 vvdw6 = _mm256_mul_pd(c6_00,rinvsix);
1242 vvdw12 = _mm256_mul_pd(c12_00,_mm256_mul_pd(rinvsix,rinvsix));
1243 vvdw = _mm256_sub_pd( _mm256_mul_pd(vvdw12,one_twelfth) , _mm256_mul_pd(vvdw6,one_sixth) );
1244 fvdw = _mm256_mul_pd(_mm256_sub_pd(vvdw12,vvdw6),rinvsq00);
1246 d = _mm256_sub_pd(r00,rswitch);
1247 d = _mm256_max_pd(d,_mm256_setzero_pd());
1248 d2 = _mm256_mul_pd(d,d);
1249 sw = _mm256_add_pd(one,_mm256_mul_pd(d2,_mm256_mul_pd(d,_mm256_add_pd(swV3,_mm256_mul_pd(d,_mm256_add_pd(swV4,_mm256_mul_pd(d,swV5)))))));
1251 dsw = _mm256_mul_pd(d2,_mm256_add_pd(swF2,_mm256_mul_pd(d,_mm256_add_pd(swF3,_mm256_mul_pd(d,swF4)))));
1253 /* Evaluate switch function */
1254 /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
1255 fvdw = _mm256_sub_pd( _mm256_mul_pd(fvdw,sw) , _mm256_mul_pd(rinv00,_mm256_mul_pd(vvdw,dsw)) );
1256 cutoff_mask = _mm256_cmp_pd(rsq00,rcutoff2,_CMP_LT_OQ);
1260 fscal = _mm256_and_pd(fscal,cutoff_mask);
1262 fscal = _mm256_andnot_pd(dummy_mask,fscal);
1264 /* Calculate temporary vectorial force */
1265 tx = _mm256_mul_pd(fscal,dx00);
1266 ty = _mm256_mul_pd(fscal,dy00);
1267 tz = _mm256_mul_pd(fscal,dz00);
1269 /* Update vectorial force */
1270 fix0 = _mm256_add_pd(fix0,tx);
1271 fiy0 = _mm256_add_pd(fiy0,ty);
1272 fiz0 = _mm256_add_pd(fiz0,tz);
1274 fjx0 = _mm256_add_pd(fjx0,tx);
1275 fjy0 = _mm256_add_pd(fjy0,ty);
1276 fjz0 = _mm256_add_pd(fjz0,tz);
1280 /**************************
1281 * CALCULATE INTERACTIONS *
1282 **************************/
1284 if (gmx_mm256_any_lt(rsq10,rcutoff2))
1287 /* Compute parameters for interactions between i and j atoms */
1288 qq10 = _mm256_mul_pd(iq1,jq0);
1290 /* REACTION-FIELD ELECTROSTATICS */
1291 felec = _mm256_mul_pd(qq10,_mm256_sub_pd(_mm256_mul_pd(rinv10,rinvsq10),krf2));
1293 cutoff_mask = _mm256_cmp_pd(rsq10,rcutoff2,_CMP_LT_OQ);
1297 fscal = _mm256_and_pd(fscal,cutoff_mask);
1299 fscal = _mm256_andnot_pd(dummy_mask,fscal);
1301 /* Calculate temporary vectorial force */
1302 tx = _mm256_mul_pd(fscal,dx10);
1303 ty = _mm256_mul_pd(fscal,dy10);
1304 tz = _mm256_mul_pd(fscal,dz10);
1306 /* Update vectorial force */
1307 fix1 = _mm256_add_pd(fix1,tx);
1308 fiy1 = _mm256_add_pd(fiy1,ty);
1309 fiz1 = _mm256_add_pd(fiz1,tz);
1311 fjx0 = _mm256_add_pd(fjx0,tx);
1312 fjy0 = _mm256_add_pd(fjy0,ty);
1313 fjz0 = _mm256_add_pd(fjz0,tz);
1317 /**************************
1318 * CALCULATE INTERACTIONS *
1319 **************************/
1321 if (gmx_mm256_any_lt(rsq20,rcutoff2))
1324 /* Compute parameters for interactions between i and j atoms */
1325 qq20 = _mm256_mul_pd(iq2,jq0);
1327 /* REACTION-FIELD ELECTROSTATICS */
1328 felec = _mm256_mul_pd(qq20,_mm256_sub_pd(_mm256_mul_pd(rinv20,rinvsq20),krf2));
1330 cutoff_mask = _mm256_cmp_pd(rsq20,rcutoff2,_CMP_LT_OQ);
1334 fscal = _mm256_and_pd(fscal,cutoff_mask);
1336 fscal = _mm256_andnot_pd(dummy_mask,fscal);
1338 /* Calculate temporary vectorial force */
1339 tx = _mm256_mul_pd(fscal,dx20);
1340 ty = _mm256_mul_pd(fscal,dy20);
1341 tz = _mm256_mul_pd(fscal,dz20);
1343 /* Update vectorial force */
1344 fix2 = _mm256_add_pd(fix2,tx);
1345 fiy2 = _mm256_add_pd(fiy2,ty);
1346 fiz2 = _mm256_add_pd(fiz2,tz);
1348 fjx0 = _mm256_add_pd(fjx0,tx);
1349 fjy0 = _mm256_add_pd(fjy0,ty);
1350 fjz0 = _mm256_add_pd(fjz0,tz);
1354 /**************************
1355 * CALCULATE INTERACTIONS *
1356 **************************/
1358 if (gmx_mm256_any_lt(rsq30,rcutoff2))
1361 /* Compute parameters for interactions between i and j atoms */
1362 qq30 = _mm256_mul_pd(iq3,jq0);
1364 /* REACTION-FIELD ELECTROSTATICS */
1365 felec = _mm256_mul_pd(qq30,_mm256_sub_pd(_mm256_mul_pd(rinv30,rinvsq30),krf2));
1367 cutoff_mask = _mm256_cmp_pd(rsq30,rcutoff2,_CMP_LT_OQ);
1371 fscal = _mm256_and_pd(fscal,cutoff_mask);
1373 fscal = _mm256_andnot_pd(dummy_mask,fscal);
1375 /* Calculate temporary vectorial force */
1376 tx = _mm256_mul_pd(fscal,dx30);
1377 ty = _mm256_mul_pd(fscal,dy30);
1378 tz = _mm256_mul_pd(fscal,dz30);
1380 /* Update vectorial force */
1381 fix3 = _mm256_add_pd(fix3,tx);
1382 fiy3 = _mm256_add_pd(fiy3,ty);
1383 fiz3 = _mm256_add_pd(fiz3,tz);
1385 fjx0 = _mm256_add_pd(fjx0,tx);
1386 fjy0 = _mm256_add_pd(fjy0,ty);
1387 fjz0 = _mm256_add_pd(fjz0,tz);
1391 fjptrA = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
1392 fjptrB = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
1393 fjptrC = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
1394 fjptrD = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
1396 gmx_mm256_decrement_1rvec_4ptr_swizzle_pd(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
1398 /* Inner loop uses 150 flops */
1401 /* End of innermost loop */
1403 gmx_mm256_update_iforce_4atom_swizzle_pd(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,fix3,fiy3,fiz3,
1404 f+i_coord_offset,fshift+i_shift_offset);
1406 /* Increment number of inner iterations */
1407 inneriter += j_index_end - j_index_start;
1409 /* Outer loop uses 24 flops */
1412 /* Increment number of outer iterations */
1415 /* Update outer/inner flops */
1417 inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W4_F,outeriter*24 + inneriter*150);