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_VdwNone_GeomW4P1_VF_avx_256_double
54 * Electrostatics interaction: ReactionField
55 * VdW interaction: None
56 * Geometry: Water4-Particle
57 * Calculate force/pot: PotentialAndForce
60 nb_kernel_ElecRFCut_VdwNone_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 * vdwioffsetptr1;
87 __m256d ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
88 real * vdwioffsetptr2;
89 __m256d ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
90 real * vdwioffsetptr3;
91 __m256d ix3,iy3,iz3,fix3,fiy3,fiz3,iq3,isai3;
92 int vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
93 __m256d jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
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;
99 __m256d dummy_mask,cutoff_mask;
100 __m128 tmpmask0,tmpmask1;
101 __m256d signbit = _mm256_castsi256_pd( _mm256_set1_epi32(0x80000000) );
102 __m256d one = _mm256_set1_pd(1.0);
103 __m256d two = _mm256_set1_pd(2.0);
109 jindex = nlist->jindex;
111 shiftidx = nlist->shift;
113 shiftvec = fr->shift_vec[0];
114 fshift = fr->fshift[0];
115 facel = _mm256_set1_pd(fr->epsfac);
116 charge = mdatoms->chargeA;
117 krf = _mm256_set1_pd(fr->ic->k_rf);
118 krf2 = _mm256_set1_pd(fr->ic->k_rf*2.0);
119 crf = _mm256_set1_pd(fr->ic->c_rf);
121 /* Setup water-specific parameters */
122 inr = nlist->iinr[0];
123 iq1 = _mm256_mul_pd(facel,_mm256_set1_pd(charge[inr+1]));
124 iq2 = _mm256_mul_pd(facel,_mm256_set1_pd(charge[inr+2]));
125 iq3 = _mm256_mul_pd(facel,_mm256_set1_pd(charge[inr+3]));
127 /* When we use explicit cutoffs the value must be identical for elec and VdW, so use elec as an arbitrary choice */
128 rcutoff_scalar = fr->rcoulomb;
129 rcutoff = _mm256_set1_pd(rcutoff_scalar);
130 rcutoff2 = _mm256_mul_pd(rcutoff,rcutoff);
132 /* Avoid stupid compiler warnings */
133 jnrA = jnrB = jnrC = jnrD = 0;
142 for(iidx=0;iidx<4*DIM;iidx++)
147 /* Start outer loop over neighborlists */
148 for(iidx=0; iidx<nri; iidx++)
150 /* Load shift vector for this list */
151 i_shift_offset = DIM*shiftidx[iidx];
153 /* Load limits for loop over neighbors */
154 j_index_start = jindex[iidx];
155 j_index_end = jindex[iidx+1];
157 /* Get outer coordinate index */
159 i_coord_offset = DIM*inr;
161 /* Load i particle coords and add shift vector */
162 gmx_mm256_load_shift_and_3rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset+DIM,
163 &ix1,&iy1,&iz1,&ix2,&iy2,&iz2,&ix3,&iy3,&iz3);
165 fix1 = _mm256_setzero_pd();
166 fiy1 = _mm256_setzero_pd();
167 fiz1 = _mm256_setzero_pd();
168 fix2 = _mm256_setzero_pd();
169 fiy2 = _mm256_setzero_pd();
170 fiz2 = _mm256_setzero_pd();
171 fix3 = _mm256_setzero_pd();
172 fiy3 = _mm256_setzero_pd();
173 fiz3 = _mm256_setzero_pd();
175 /* Reset potential sums */
176 velecsum = _mm256_setzero_pd();
178 /* Start inner kernel loop */
179 for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
182 /* Get j neighbor index, and coordinate index */
187 j_coord_offsetA = DIM*jnrA;
188 j_coord_offsetB = DIM*jnrB;
189 j_coord_offsetC = DIM*jnrC;
190 j_coord_offsetD = DIM*jnrD;
192 /* load j atom coordinates */
193 gmx_mm256_load_1rvec_4ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
194 x+j_coord_offsetC,x+j_coord_offsetD,
197 /* Calculate displacement vector */
198 dx10 = _mm256_sub_pd(ix1,jx0);
199 dy10 = _mm256_sub_pd(iy1,jy0);
200 dz10 = _mm256_sub_pd(iz1,jz0);
201 dx20 = _mm256_sub_pd(ix2,jx0);
202 dy20 = _mm256_sub_pd(iy2,jy0);
203 dz20 = _mm256_sub_pd(iz2,jz0);
204 dx30 = _mm256_sub_pd(ix3,jx0);
205 dy30 = _mm256_sub_pd(iy3,jy0);
206 dz30 = _mm256_sub_pd(iz3,jz0);
208 /* Calculate squared distance and things based on it */
209 rsq10 = gmx_mm256_calc_rsq_pd(dx10,dy10,dz10);
210 rsq20 = gmx_mm256_calc_rsq_pd(dx20,dy20,dz20);
211 rsq30 = gmx_mm256_calc_rsq_pd(dx30,dy30,dz30);
213 rinv10 = gmx_mm256_invsqrt_pd(rsq10);
214 rinv20 = gmx_mm256_invsqrt_pd(rsq20);
215 rinv30 = gmx_mm256_invsqrt_pd(rsq30);
217 rinvsq10 = _mm256_mul_pd(rinv10,rinv10);
218 rinvsq20 = _mm256_mul_pd(rinv20,rinv20);
219 rinvsq30 = _mm256_mul_pd(rinv30,rinv30);
221 /* Load parameters for j particles */
222 jq0 = gmx_mm256_load_4real_swizzle_pd(charge+jnrA+0,charge+jnrB+0,
223 charge+jnrC+0,charge+jnrD+0);
225 fjx0 = _mm256_setzero_pd();
226 fjy0 = _mm256_setzero_pd();
227 fjz0 = _mm256_setzero_pd();
229 /**************************
230 * CALCULATE INTERACTIONS *
231 **************************/
233 if (gmx_mm256_any_lt(rsq10,rcutoff2))
236 /* Compute parameters for interactions between i and j atoms */
237 qq10 = _mm256_mul_pd(iq1,jq0);
239 /* REACTION-FIELD ELECTROSTATICS */
240 velec = _mm256_mul_pd(qq10,_mm256_sub_pd(_mm256_add_pd(rinv10,_mm256_mul_pd(krf,rsq10)),crf));
241 felec = _mm256_mul_pd(qq10,_mm256_sub_pd(_mm256_mul_pd(rinv10,rinvsq10),krf2));
243 cutoff_mask = _mm256_cmp_pd(rsq10,rcutoff2,_CMP_LT_OQ);
245 /* Update potential sum for this i atom from the interaction with this j atom. */
246 velec = _mm256_and_pd(velec,cutoff_mask);
247 velecsum = _mm256_add_pd(velecsum,velec);
251 fscal = _mm256_and_pd(fscal,cutoff_mask);
253 /* Calculate temporary vectorial force */
254 tx = _mm256_mul_pd(fscal,dx10);
255 ty = _mm256_mul_pd(fscal,dy10);
256 tz = _mm256_mul_pd(fscal,dz10);
258 /* Update vectorial force */
259 fix1 = _mm256_add_pd(fix1,tx);
260 fiy1 = _mm256_add_pd(fiy1,ty);
261 fiz1 = _mm256_add_pd(fiz1,tz);
263 fjx0 = _mm256_add_pd(fjx0,tx);
264 fjy0 = _mm256_add_pd(fjy0,ty);
265 fjz0 = _mm256_add_pd(fjz0,tz);
269 /**************************
270 * CALCULATE INTERACTIONS *
271 **************************/
273 if (gmx_mm256_any_lt(rsq20,rcutoff2))
276 /* Compute parameters for interactions between i and j atoms */
277 qq20 = _mm256_mul_pd(iq2,jq0);
279 /* REACTION-FIELD ELECTROSTATICS */
280 velec = _mm256_mul_pd(qq20,_mm256_sub_pd(_mm256_add_pd(rinv20,_mm256_mul_pd(krf,rsq20)),crf));
281 felec = _mm256_mul_pd(qq20,_mm256_sub_pd(_mm256_mul_pd(rinv20,rinvsq20),krf2));
283 cutoff_mask = _mm256_cmp_pd(rsq20,rcutoff2,_CMP_LT_OQ);
285 /* Update potential sum for this i atom from the interaction with this j atom. */
286 velec = _mm256_and_pd(velec,cutoff_mask);
287 velecsum = _mm256_add_pd(velecsum,velec);
291 fscal = _mm256_and_pd(fscal,cutoff_mask);
293 /* Calculate temporary vectorial force */
294 tx = _mm256_mul_pd(fscal,dx20);
295 ty = _mm256_mul_pd(fscal,dy20);
296 tz = _mm256_mul_pd(fscal,dz20);
298 /* Update vectorial force */
299 fix2 = _mm256_add_pd(fix2,tx);
300 fiy2 = _mm256_add_pd(fiy2,ty);
301 fiz2 = _mm256_add_pd(fiz2,tz);
303 fjx0 = _mm256_add_pd(fjx0,tx);
304 fjy0 = _mm256_add_pd(fjy0,ty);
305 fjz0 = _mm256_add_pd(fjz0,tz);
309 /**************************
310 * CALCULATE INTERACTIONS *
311 **************************/
313 if (gmx_mm256_any_lt(rsq30,rcutoff2))
316 /* Compute parameters for interactions between i and j atoms */
317 qq30 = _mm256_mul_pd(iq3,jq0);
319 /* REACTION-FIELD ELECTROSTATICS */
320 velec = _mm256_mul_pd(qq30,_mm256_sub_pd(_mm256_add_pd(rinv30,_mm256_mul_pd(krf,rsq30)),crf));
321 felec = _mm256_mul_pd(qq30,_mm256_sub_pd(_mm256_mul_pd(rinv30,rinvsq30),krf2));
323 cutoff_mask = _mm256_cmp_pd(rsq30,rcutoff2,_CMP_LT_OQ);
325 /* Update potential sum for this i atom from the interaction with this j atom. */
326 velec = _mm256_and_pd(velec,cutoff_mask);
327 velecsum = _mm256_add_pd(velecsum,velec);
331 fscal = _mm256_and_pd(fscal,cutoff_mask);
333 /* Calculate temporary vectorial force */
334 tx = _mm256_mul_pd(fscal,dx30);
335 ty = _mm256_mul_pd(fscal,dy30);
336 tz = _mm256_mul_pd(fscal,dz30);
338 /* Update vectorial force */
339 fix3 = _mm256_add_pd(fix3,tx);
340 fiy3 = _mm256_add_pd(fiy3,ty);
341 fiz3 = _mm256_add_pd(fiz3,tz);
343 fjx0 = _mm256_add_pd(fjx0,tx);
344 fjy0 = _mm256_add_pd(fjy0,ty);
345 fjz0 = _mm256_add_pd(fjz0,tz);
349 fjptrA = f+j_coord_offsetA;
350 fjptrB = f+j_coord_offsetB;
351 fjptrC = f+j_coord_offsetC;
352 fjptrD = f+j_coord_offsetD;
354 gmx_mm256_decrement_1rvec_4ptr_swizzle_pd(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
356 /* Inner loop uses 111 flops */
362 /* Get j neighbor index, and coordinate index */
363 jnrlistA = jjnr[jidx];
364 jnrlistB = jjnr[jidx+1];
365 jnrlistC = jjnr[jidx+2];
366 jnrlistD = jjnr[jidx+3];
367 /* Sign of each element will be negative for non-real atoms.
368 * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
369 * so use it as val = _mm_andnot_pd(mask,val) to clear dummy entries.
371 tmpmask0 = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
373 tmpmask1 = _mm_permute_ps(tmpmask0,_GMX_MM_PERMUTE(3,3,2,2));
374 tmpmask0 = _mm_permute_ps(tmpmask0,_GMX_MM_PERMUTE(1,1,0,0));
375 dummy_mask = _mm256_castps_pd(gmx_mm256_set_m128(tmpmask1,tmpmask0));
377 jnrA = (jnrlistA>=0) ? jnrlistA : 0;
378 jnrB = (jnrlistB>=0) ? jnrlistB : 0;
379 jnrC = (jnrlistC>=0) ? jnrlistC : 0;
380 jnrD = (jnrlistD>=0) ? jnrlistD : 0;
381 j_coord_offsetA = DIM*jnrA;
382 j_coord_offsetB = DIM*jnrB;
383 j_coord_offsetC = DIM*jnrC;
384 j_coord_offsetD = DIM*jnrD;
386 /* load j atom coordinates */
387 gmx_mm256_load_1rvec_4ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
388 x+j_coord_offsetC,x+j_coord_offsetD,
391 /* Calculate displacement vector */
392 dx10 = _mm256_sub_pd(ix1,jx0);
393 dy10 = _mm256_sub_pd(iy1,jy0);
394 dz10 = _mm256_sub_pd(iz1,jz0);
395 dx20 = _mm256_sub_pd(ix2,jx0);
396 dy20 = _mm256_sub_pd(iy2,jy0);
397 dz20 = _mm256_sub_pd(iz2,jz0);
398 dx30 = _mm256_sub_pd(ix3,jx0);
399 dy30 = _mm256_sub_pd(iy3,jy0);
400 dz30 = _mm256_sub_pd(iz3,jz0);
402 /* Calculate squared distance and things based on it */
403 rsq10 = gmx_mm256_calc_rsq_pd(dx10,dy10,dz10);
404 rsq20 = gmx_mm256_calc_rsq_pd(dx20,dy20,dz20);
405 rsq30 = gmx_mm256_calc_rsq_pd(dx30,dy30,dz30);
407 rinv10 = gmx_mm256_invsqrt_pd(rsq10);
408 rinv20 = gmx_mm256_invsqrt_pd(rsq20);
409 rinv30 = gmx_mm256_invsqrt_pd(rsq30);
411 rinvsq10 = _mm256_mul_pd(rinv10,rinv10);
412 rinvsq20 = _mm256_mul_pd(rinv20,rinv20);
413 rinvsq30 = _mm256_mul_pd(rinv30,rinv30);
415 /* Load parameters for j particles */
416 jq0 = gmx_mm256_load_4real_swizzle_pd(charge+jnrA+0,charge+jnrB+0,
417 charge+jnrC+0,charge+jnrD+0);
419 fjx0 = _mm256_setzero_pd();
420 fjy0 = _mm256_setzero_pd();
421 fjz0 = _mm256_setzero_pd();
423 /**************************
424 * CALCULATE INTERACTIONS *
425 **************************/
427 if (gmx_mm256_any_lt(rsq10,rcutoff2))
430 /* Compute parameters for interactions between i and j atoms */
431 qq10 = _mm256_mul_pd(iq1,jq0);
433 /* REACTION-FIELD ELECTROSTATICS */
434 velec = _mm256_mul_pd(qq10,_mm256_sub_pd(_mm256_add_pd(rinv10,_mm256_mul_pd(krf,rsq10)),crf));
435 felec = _mm256_mul_pd(qq10,_mm256_sub_pd(_mm256_mul_pd(rinv10,rinvsq10),krf2));
437 cutoff_mask = _mm256_cmp_pd(rsq10,rcutoff2,_CMP_LT_OQ);
439 /* Update potential sum for this i atom from the interaction with this j atom. */
440 velec = _mm256_and_pd(velec,cutoff_mask);
441 velec = _mm256_andnot_pd(dummy_mask,velec);
442 velecsum = _mm256_add_pd(velecsum,velec);
446 fscal = _mm256_and_pd(fscal,cutoff_mask);
448 fscal = _mm256_andnot_pd(dummy_mask,fscal);
450 /* Calculate temporary vectorial force */
451 tx = _mm256_mul_pd(fscal,dx10);
452 ty = _mm256_mul_pd(fscal,dy10);
453 tz = _mm256_mul_pd(fscal,dz10);
455 /* Update vectorial force */
456 fix1 = _mm256_add_pd(fix1,tx);
457 fiy1 = _mm256_add_pd(fiy1,ty);
458 fiz1 = _mm256_add_pd(fiz1,tz);
460 fjx0 = _mm256_add_pd(fjx0,tx);
461 fjy0 = _mm256_add_pd(fjy0,ty);
462 fjz0 = _mm256_add_pd(fjz0,tz);
466 /**************************
467 * CALCULATE INTERACTIONS *
468 **************************/
470 if (gmx_mm256_any_lt(rsq20,rcutoff2))
473 /* Compute parameters for interactions between i and j atoms */
474 qq20 = _mm256_mul_pd(iq2,jq0);
476 /* REACTION-FIELD ELECTROSTATICS */
477 velec = _mm256_mul_pd(qq20,_mm256_sub_pd(_mm256_add_pd(rinv20,_mm256_mul_pd(krf,rsq20)),crf));
478 felec = _mm256_mul_pd(qq20,_mm256_sub_pd(_mm256_mul_pd(rinv20,rinvsq20),krf2));
480 cutoff_mask = _mm256_cmp_pd(rsq20,rcutoff2,_CMP_LT_OQ);
482 /* Update potential sum for this i atom from the interaction with this j atom. */
483 velec = _mm256_and_pd(velec,cutoff_mask);
484 velec = _mm256_andnot_pd(dummy_mask,velec);
485 velecsum = _mm256_add_pd(velecsum,velec);
489 fscal = _mm256_and_pd(fscal,cutoff_mask);
491 fscal = _mm256_andnot_pd(dummy_mask,fscal);
493 /* Calculate temporary vectorial force */
494 tx = _mm256_mul_pd(fscal,dx20);
495 ty = _mm256_mul_pd(fscal,dy20);
496 tz = _mm256_mul_pd(fscal,dz20);
498 /* Update vectorial force */
499 fix2 = _mm256_add_pd(fix2,tx);
500 fiy2 = _mm256_add_pd(fiy2,ty);
501 fiz2 = _mm256_add_pd(fiz2,tz);
503 fjx0 = _mm256_add_pd(fjx0,tx);
504 fjy0 = _mm256_add_pd(fjy0,ty);
505 fjz0 = _mm256_add_pd(fjz0,tz);
509 /**************************
510 * CALCULATE INTERACTIONS *
511 **************************/
513 if (gmx_mm256_any_lt(rsq30,rcutoff2))
516 /* Compute parameters for interactions between i and j atoms */
517 qq30 = _mm256_mul_pd(iq3,jq0);
519 /* REACTION-FIELD ELECTROSTATICS */
520 velec = _mm256_mul_pd(qq30,_mm256_sub_pd(_mm256_add_pd(rinv30,_mm256_mul_pd(krf,rsq30)),crf));
521 felec = _mm256_mul_pd(qq30,_mm256_sub_pd(_mm256_mul_pd(rinv30,rinvsq30),krf2));
523 cutoff_mask = _mm256_cmp_pd(rsq30,rcutoff2,_CMP_LT_OQ);
525 /* Update potential sum for this i atom from the interaction with this j atom. */
526 velec = _mm256_and_pd(velec,cutoff_mask);
527 velec = _mm256_andnot_pd(dummy_mask,velec);
528 velecsum = _mm256_add_pd(velecsum,velec);
532 fscal = _mm256_and_pd(fscal,cutoff_mask);
534 fscal = _mm256_andnot_pd(dummy_mask,fscal);
536 /* Calculate temporary vectorial force */
537 tx = _mm256_mul_pd(fscal,dx30);
538 ty = _mm256_mul_pd(fscal,dy30);
539 tz = _mm256_mul_pd(fscal,dz30);
541 /* Update vectorial force */
542 fix3 = _mm256_add_pd(fix3,tx);
543 fiy3 = _mm256_add_pd(fiy3,ty);
544 fiz3 = _mm256_add_pd(fiz3,tz);
546 fjx0 = _mm256_add_pd(fjx0,tx);
547 fjy0 = _mm256_add_pd(fjy0,ty);
548 fjz0 = _mm256_add_pd(fjz0,tz);
552 fjptrA = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
553 fjptrB = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
554 fjptrC = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
555 fjptrD = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
557 gmx_mm256_decrement_1rvec_4ptr_swizzle_pd(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
559 /* Inner loop uses 111 flops */
562 /* End of innermost loop */
564 gmx_mm256_update_iforce_3atom_swizzle_pd(fix1,fiy1,fiz1,fix2,fiy2,fiz2,fix3,fiy3,fiz3,
565 f+i_coord_offset+DIM,fshift+i_shift_offset);
568 /* Update potential energies */
569 gmx_mm256_update_1pot_pd(velecsum,kernel_data->energygrp_elec+ggid);
571 /* Increment number of inner iterations */
572 inneriter += j_index_end - j_index_start;
574 /* Outer loop uses 19 flops */
577 /* Increment number of outer iterations */
580 /* Update outer/inner flops */
582 inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_W4_VF,outeriter*19 + inneriter*111);
585 * Gromacs nonbonded kernel: nb_kernel_ElecRFCut_VdwNone_GeomW4P1_F_avx_256_double
586 * Electrostatics interaction: ReactionField
587 * VdW interaction: None
588 * Geometry: Water4-Particle
589 * Calculate force/pot: Force
592 nb_kernel_ElecRFCut_VdwNone_GeomW4P1_F_avx_256_double
593 (t_nblist * gmx_restrict nlist,
594 rvec * gmx_restrict xx,
595 rvec * gmx_restrict ff,
596 t_forcerec * gmx_restrict fr,
597 t_mdatoms * gmx_restrict mdatoms,
598 nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
599 t_nrnb * gmx_restrict nrnb)
601 /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
602 * just 0 for non-waters.
603 * Suffixes A,B,C,D refer to j loop unrolling done with AVX, e.g. for the four different
604 * jnr indices corresponding to data put in the four positions in the SIMD register.
606 int i_shift_offset,i_coord_offset,outeriter,inneriter;
607 int j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
608 int jnrA,jnrB,jnrC,jnrD;
609 int jnrlistA,jnrlistB,jnrlistC,jnrlistD;
610 int jnrlistE,jnrlistF,jnrlistG,jnrlistH;
611 int j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
612 int *iinr,*jindex,*jjnr,*shiftidx,*gid;
614 real *shiftvec,*fshift,*x,*f;
615 real *fjptrA,*fjptrB,*fjptrC,*fjptrD;
617 __m256d tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
618 real * vdwioffsetptr1;
619 __m256d ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
620 real * vdwioffsetptr2;
621 __m256d ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
622 real * vdwioffsetptr3;
623 __m256d ix3,iy3,iz3,fix3,fiy3,fiz3,iq3,isai3;
624 int vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
625 __m256d jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
626 __m256d dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
627 __m256d dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
628 __m256d dx30,dy30,dz30,rsq30,rinv30,rinvsq30,r30,qq30,c6_30,c12_30;
629 __m256d velec,felec,velecsum,facel,crf,krf,krf2;
631 __m256d dummy_mask,cutoff_mask;
632 __m128 tmpmask0,tmpmask1;
633 __m256d signbit = _mm256_castsi256_pd( _mm256_set1_epi32(0x80000000) );
634 __m256d one = _mm256_set1_pd(1.0);
635 __m256d two = _mm256_set1_pd(2.0);
641 jindex = nlist->jindex;
643 shiftidx = nlist->shift;
645 shiftvec = fr->shift_vec[0];
646 fshift = fr->fshift[0];
647 facel = _mm256_set1_pd(fr->epsfac);
648 charge = mdatoms->chargeA;
649 krf = _mm256_set1_pd(fr->ic->k_rf);
650 krf2 = _mm256_set1_pd(fr->ic->k_rf*2.0);
651 crf = _mm256_set1_pd(fr->ic->c_rf);
653 /* Setup water-specific parameters */
654 inr = nlist->iinr[0];
655 iq1 = _mm256_mul_pd(facel,_mm256_set1_pd(charge[inr+1]));
656 iq2 = _mm256_mul_pd(facel,_mm256_set1_pd(charge[inr+2]));
657 iq3 = _mm256_mul_pd(facel,_mm256_set1_pd(charge[inr+3]));
659 /* When we use explicit cutoffs the value must be identical for elec and VdW, so use elec as an arbitrary choice */
660 rcutoff_scalar = fr->rcoulomb;
661 rcutoff = _mm256_set1_pd(rcutoff_scalar);
662 rcutoff2 = _mm256_mul_pd(rcutoff,rcutoff);
664 /* Avoid stupid compiler warnings */
665 jnrA = jnrB = jnrC = jnrD = 0;
674 for(iidx=0;iidx<4*DIM;iidx++)
679 /* Start outer loop over neighborlists */
680 for(iidx=0; iidx<nri; iidx++)
682 /* Load shift vector for this list */
683 i_shift_offset = DIM*shiftidx[iidx];
685 /* Load limits for loop over neighbors */
686 j_index_start = jindex[iidx];
687 j_index_end = jindex[iidx+1];
689 /* Get outer coordinate index */
691 i_coord_offset = DIM*inr;
693 /* Load i particle coords and add shift vector */
694 gmx_mm256_load_shift_and_3rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset+DIM,
695 &ix1,&iy1,&iz1,&ix2,&iy2,&iz2,&ix3,&iy3,&iz3);
697 fix1 = _mm256_setzero_pd();
698 fiy1 = _mm256_setzero_pd();
699 fiz1 = _mm256_setzero_pd();
700 fix2 = _mm256_setzero_pd();
701 fiy2 = _mm256_setzero_pd();
702 fiz2 = _mm256_setzero_pd();
703 fix3 = _mm256_setzero_pd();
704 fiy3 = _mm256_setzero_pd();
705 fiz3 = _mm256_setzero_pd();
707 /* Start inner kernel loop */
708 for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
711 /* Get j neighbor index, and coordinate index */
716 j_coord_offsetA = DIM*jnrA;
717 j_coord_offsetB = DIM*jnrB;
718 j_coord_offsetC = DIM*jnrC;
719 j_coord_offsetD = DIM*jnrD;
721 /* load j atom coordinates */
722 gmx_mm256_load_1rvec_4ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
723 x+j_coord_offsetC,x+j_coord_offsetD,
726 /* Calculate displacement vector */
727 dx10 = _mm256_sub_pd(ix1,jx0);
728 dy10 = _mm256_sub_pd(iy1,jy0);
729 dz10 = _mm256_sub_pd(iz1,jz0);
730 dx20 = _mm256_sub_pd(ix2,jx0);
731 dy20 = _mm256_sub_pd(iy2,jy0);
732 dz20 = _mm256_sub_pd(iz2,jz0);
733 dx30 = _mm256_sub_pd(ix3,jx0);
734 dy30 = _mm256_sub_pd(iy3,jy0);
735 dz30 = _mm256_sub_pd(iz3,jz0);
737 /* Calculate squared distance and things based on it */
738 rsq10 = gmx_mm256_calc_rsq_pd(dx10,dy10,dz10);
739 rsq20 = gmx_mm256_calc_rsq_pd(dx20,dy20,dz20);
740 rsq30 = gmx_mm256_calc_rsq_pd(dx30,dy30,dz30);
742 rinv10 = gmx_mm256_invsqrt_pd(rsq10);
743 rinv20 = gmx_mm256_invsqrt_pd(rsq20);
744 rinv30 = gmx_mm256_invsqrt_pd(rsq30);
746 rinvsq10 = _mm256_mul_pd(rinv10,rinv10);
747 rinvsq20 = _mm256_mul_pd(rinv20,rinv20);
748 rinvsq30 = _mm256_mul_pd(rinv30,rinv30);
750 /* Load parameters for j particles */
751 jq0 = gmx_mm256_load_4real_swizzle_pd(charge+jnrA+0,charge+jnrB+0,
752 charge+jnrC+0,charge+jnrD+0);
754 fjx0 = _mm256_setzero_pd();
755 fjy0 = _mm256_setzero_pd();
756 fjz0 = _mm256_setzero_pd();
758 /**************************
759 * CALCULATE INTERACTIONS *
760 **************************/
762 if (gmx_mm256_any_lt(rsq10,rcutoff2))
765 /* Compute parameters for interactions between i and j atoms */
766 qq10 = _mm256_mul_pd(iq1,jq0);
768 /* REACTION-FIELD ELECTROSTATICS */
769 felec = _mm256_mul_pd(qq10,_mm256_sub_pd(_mm256_mul_pd(rinv10,rinvsq10),krf2));
771 cutoff_mask = _mm256_cmp_pd(rsq10,rcutoff2,_CMP_LT_OQ);
775 fscal = _mm256_and_pd(fscal,cutoff_mask);
777 /* Calculate temporary vectorial force */
778 tx = _mm256_mul_pd(fscal,dx10);
779 ty = _mm256_mul_pd(fscal,dy10);
780 tz = _mm256_mul_pd(fscal,dz10);
782 /* Update vectorial force */
783 fix1 = _mm256_add_pd(fix1,tx);
784 fiy1 = _mm256_add_pd(fiy1,ty);
785 fiz1 = _mm256_add_pd(fiz1,tz);
787 fjx0 = _mm256_add_pd(fjx0,tx);
788 fjy0 = _mm256_add_pd(fjy0,ty);
789 fjz0 = _mm256_add_pd(fjz0,tz);
793 /**************************
794 * CALCULATE INTERACTIONS *
795 **************************/
797 if (gmx_mm256_any_lt(rsq20,rcutoff2))
800 /* Compute parameters for interactions between i and j atoms */
801 qq20 = _mm256_mul_pd(iq2,jq0);
803 /* REACTION-FIELD ELECTROSTATICS */
804 felec = _mm256_mul_pd(qq20,_mm256_sub_pd(_mm256_mul_pd(rinv20,rinvsq20),krf2));
806 cutoff_mask = _mm256_cmp_pd(rsq20,rcutoff2,_CMP_LT_OQ);
810 fscal = _mm256_and_pd(fscal,cutoff_mask);
812 /* Calculate temporary vectorial force */
813 tx = _mm256_mul_pd(fscal,dx20);
814 ty = _mm256_mul_pd(fscal,dy20);
815 tz = _mm256_mul_pd(fscal,dz20);
817 /* Update vectorial force */
818 fix2 = _mm256_add_pd(fix2,tx);
819 fiy2 = _mm256_add_pd(fiy2,ty);
820 fiz2 = _mm256_add_pd(fiz2,tz);
822 fjx0 = _mm256_add_pd(fjx0,tx);
823 fjy0 = _mm256_add_pd(fjy0,ty);
824 fjz0 = _mm256_add_pd(fjz0,tz);
828 /**************************
829 * CALCULATE INTERACTIONS *
830 **************************/
832 if (gmx_mm256_any_lt(rsq30,rcutoff2))
835 /* Compute parameters for interactions between i and j atoms */
836 qq30 = _mm256_mul_pd(iq3,jq0);
838 /* REACTION-FIELD ELECTROSTATICS */
839 felec = _mm256_mul_pd(qq30,_mm256_sub_pd(_mm256_mul_pd(rinv30,rinvsq30),krf2));
841 cutoff_mask = _mm256_cmp_pd(rsq30,rcutoff2,_CMP_LT_OQ);
845 fscal = _mm256_and_pd(fscal,cutoff_mask);
847 /* Calculate temporary vectorial force */
848 tx = _mm256_mul_pd(fscal,dx30);
849 ty = _mm256_mul_pd(fscal,dy30);
850 tz = _mm256_mul_pd(fscal,dz30);
852 /* Update vectorial force */
853 fix3 = _mm256_add_pd(fix3,tx);
854 fiy3 = _mm256_add_pd(fiy3,ty);
855 fiz3 = _mm256_add_pd(fiz3,tz);
857 fjx0 = _mm256_add_pd(fjx0,tx);
858 fjy0 = _mm256_add_pd(fjy0,ty);
859 fjz0 = _mm256_add_pd(fjz0,tz);
863 fjptrA = f+j_coord_offsetA;
864 fjptrB = f+j_coord_offsetB;
865 fjptrC = f+j_coord_offsetC;
866 fjptrD = f+j_coord_offsetD;
868 gmx_mm256_decrement_1rvec_4ptr_swizzle_pd(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
870 /* Inner loop uses 93 flops */
876 /* Get j neighbor index, and coordinate index */
877 jnrlistA = jjnr[jidx];
878 jnrlistB = jjnr[jidx+1];
879 jnrlistC = jjnr[jidx+2];
880 jnrlistD = jjnr[jidx+3];
881 /* Sign of each element will be negative for non-real atoms.
882 * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
883 * so use it as val = _mm_andnot_pd(mask,val) to clear dummy entries.
885 tmpmask0 = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
887 tmpmask1 = _mm_permute_ps(tmpmask0,_GMX_MM_PERMUTE(3,3,2,2));
888 tmpmask0 = _mm_permute_ps(tmpmask0,_GMX_MM_PERMUTE(1,1,0,0));
889 dummy_mask = _mm256_castps_pd(gmx_mm256_set_m128(tmpmask1,tmpmask0));
891 jnrA = (jnrlistA>=0) ? jnrlistA : 0;
892 jnrB = (jnrlistB>=0) ? jnrlistB : 0;
893 jnrC = (jnrlistC>=0) ? jnrlistC : 0;
894 jnrD = (jnrlistD>=0) ? jnrlistD : 0;
895 j_coord_offsetA = DIM*jnrA;
896 j_coord_offsetB = DIM*jnrB;
897 j_coord_offsetC = DIM*jnrC;
898 j_coord_offsetD = DIM*jnrD;
900 /* load j atom coordinates */
901 gmx_mm256_load_1rvec_4ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
902 x+j_coord_offsetC,x+j_coord_offsetD,
905 /* Calculate displacement vector */
906 dx10 = _mm256_sub_pd(ix1,jx0);
907 dy10 = _mm256_sub_pd(iy1,jy0);
908 dz10 = _mm256_sub_pd(iz1,jz0);
909 dx20 = _mm256_sub_pd(ix2,jx0);
910 dy20 = _mm256_sub_pd(iy2,jy0);
911 dz20 = _mm256_sub_pd(iz2,jz0);
912 dx30 = _mm256_sub_pd(ix3,jx0);
913 dy30 = _mm256_sub_pd(iy3,jy0);
914 dz30 = _mm256_sub_pd(iz3,jz0);
916 /* Calculate squared distance and things based on it */
917 rsq10 = gmx_mm256_calc_rsq_pd(dx10,dy10,dz10);
918 rsq20 = gmx_mm256_calc_rsq_pd(dx20,dy20,dz20);
919 rsq30 = gmx_mm256_calc_rsq_pd(dx30,dy30,dz30);
921 rinv10 = gmx_mm256_invsqrt_pd(rsq10);
922 rinv20 = gmx_mm256_invsqrt_pd(rsq20);
923 rinv30 = gmx_mm256_invsqrt_pd(rsq30);
925 rinvsq10 = _mm256_mul_pd(rinv10,rinv10);
926 rinvsq20 = _mm256_mul_pd(rinv20,rinv20);
927 rinvsq30 = _mm256_mul_pd(rinv30,rinv30);
929 /* Load parameters for j particles */
930 jq0 = gmx_mm256_load_4real_swizzle_pd(charge+jnrA+0,charge+jnrB+0,
931 charge+jnrC+0,charge+jnrD+0);
933 fjx0 = _mm256_setzero_pd();
934 fjy0 = _mm256_setzero_pd();
935 fjz0 = _mm256_setzero_pd();
937 /**************************
938 * CALCULATE INTERACTIONS *
939 **************************/
941 if (gmx_mm256_any_lt(rsq10,rcutoff2))
944 /* Compute parameters for interactions between i and j atoms */
945 qq10 = _mm256_mul_pd(iq1,jq0);
947 /* REACTION-FIELD ELECTROSTATICS */
948 felec = _mm256_mul_pd(qq10,_mm256_sub_pd(_mm256_mul_pd(rinv10,rinvsq10),krf2));
950 cutoff_mask = _mm256_cmp_pd(rsq10,rcutoff2,_CMP_LT_OQ);
954 fscal = _mm256_and_pd(fscal,cutoff_mask);
956 fscal = _mm256_andnot_pd(dummy_mask,fscal);
958 /* Calculate temporary vectorial force */
959 tx = _mm256_mul_pd(fscal,dx10);
960 ty = _mm256_mul_pd(fscal,dy10);
961 tz = _mm256_mul_pd(fscal,dz10);
963 /* Update vectorial force */
964 fix1 = _mm256_add_pd(fix1,tx);
965 fiy1 = _mm256_add_pd(fiy1,ty);
966 fiz1 = _mm256_add_pd(fiz1,tz);
968 fjx0 = _mm256_add_pd(fjx0,tx);
969 fjy0 = _mm256_add_pd(fjy0,ty);
970 fjz0 = _mm256_add_pd(fjz0,tz);
974 /**************************
975 * CALCULATE INTERACTIONS *
976 **************************/
978 if (gmx_mm256_any_lt(rsq20,rcutoff2))
981 /* Compute parameters for interactions between i and j atoms */
982 qq20 = _mm256_mul_pd(iq2,jq0);
984 /* REACTION-FIELD ELECTROSTATICS */
985 felec = _mm256_mul_pd(qq20,_mm256_sub_pd(_mm256_mul_pd(rinv20,rinvsq20),krf2));
987 cutoff_mask = _mm256_cmp_pd(rsq20,rcutoff2,_CMP_LT_OQ);
991 fscal = _mm256_and_pd(fscal,cutoff_mask);
993 fscal = _mm256_andnot_pd(dummy_mask,fscal);
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);
1011 /**************************
1012 * CALCULATE INTERACTIONS *
1013 **************************/
1015 if (gmx_mm256_any_lt(rsq30,rcutoff2))
1018 /* Compute parameters for interactions between i and j atoms */
1019 qq30 = _mm256_mul_pd(iq3,jq0);
1021 /* REACTION-FIELD ELECTROSTATICS */
1022 felec = _mm256_mul_pd(qq30,_mm256_sub_pd(_mm256_mul_pd(rinv30,rinvsq30),krf2));
1024 cutoff_mask = _mm256_cmp_pd(rsq30,rcutoff2,_CMP_LT_OQ);
1028 fscal = _mm256_and_pd(fscal,cutoff_mask);
1030 fscal = _mm256_andnot_pd(dummy_mask,fscal);
1032 /* Calculate temporary vectorial force */
1033 tx = _mm256_mul_pd(fscal,dx30);
1034 ty = _mm256_mul_pd(fscal,dy30);
1035 tz = _mm256_mul_pd(fscal,dz30);
1037 /* Update vectorial force */
1038 fix3 = _mm256_add_pd(fix3,tx);
1039 fiy3 = _mm256_add_pd(fiy3,ty);
1040 fiz3 = _mm256_add_pd(fiz3,tz);
1042 fjx0 = _mm256_add_pd(fjx0,tx);
1043 fjy0 = _mm256_add_pd(fjy0,ty);
1044 fjz0 = _mm256_add_pd(fjz0,tz);
1048 fjptrA = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
1049 fjptrB = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
1050 fjptrC = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
1051 fjptrD = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
1053 gmx_mm256_decrement_1rvec_4ptr_swizzle_pd(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
1055 /* Inner loop uses 93 flops */
1058 /* End of innermost loop */
1060 gmx_mm256_update_iforce_3atom_swizzle_pd(fix1,fiy1,fiz1,fix2,fiy2,fiz2,fix3,fiy3,fiz3,
1061 f+i_coord_offset+DIM,fshift+i_shift_offset);
1063 /* Increment number of inner iterations */
1064 inneriter += j_index_end - j_index_start;
1066 /* Outer loop uses 18 flops */
1069 /* Increment number of outer iterations */
1072 /* Update outer/inner flops */
1074 inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_W4_F,outeriter*18 + inneriter*93);