2 * This file is part of the GROMACS molecular simulation package.
4 * Copyright (c) 2012,2013,2014, 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.
42 #include "../nb_kernel.h"
43 #include "gromacs/legacyheaders/types/simple.h"
44 #include "gromacs/math/vec.h"
45 #include "gromacs/legacyheaders/nrnb.h"
47 #include "gromacs/simd/math_x86_avx_256_double.h"
48 #include "kernelutil_x86_avx_256_double.h"
51 * Gromacs nonbonded kernel: nb_kernel_ElecRFCut_VdwNone_GeomW4P1_VF_avx_256_double
52 * Electrostatics interaction: ReactionField
53 * VdW interaction: None
54 * Geometry: Water4-Particle
55 * Calculate force/pot: PotentialAndForce
58 nb_kernel_ElecRFCut_VdwNone_GeomW4P1_VF_avx_256_double
59 (t_nblist * gmx_restrict nlist,
60 rvec * gmx_restrict xx,
61 rvec * gmx_restrict ff,
62 t_forcerec * gmx_restrict fr,
63 t_mdatoms * gmx_restrict mdatoms,
64 nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
65 t_nrnb * gmx_restrict nrnb)
67 /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
68 * just 0 for non-waters.
69 * Suffixes A,B,C,D refer to j loop unrolling done with AVX, e.g. for the four different
70 * jnr indices corresponding to data put in the four positions in the SIMD register.
72 int i_shift_offset,i_coord_offset,outeriter,inneriter;
73 int j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
74 int jnrA,jnrB,jnrC,jnrD;
75 int jnrlistA,jnrlistB,jnrlistC,jnrlistD;
76 int jnrlistE,jnrlistF,jnrlistG,jnrlistH;
77 int j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
78 int *iinr,*jindex,*jjnr,*shiftidx,*gid;
80 real *shiftvec,*fshift,*x,*f;
81 real *fjptrA,*fjptrB,*fjptrC,*fjptrD;
83 __m256d tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
84 real * vdwioffsetptr1;
85 __m256d ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
86 real * vdwioffsetptr2;
87 __m256d ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
88 real * vdwioffsetptr3;
89 __m256d ix3,iy3,iz3,fix3,fiy3,fiz3,iq3,isai3;
90 int vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
91 __m256d jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
92 __m256d dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
93 __m256d dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
94 __m256d dx30,dy30,dz30,rsq30,rinv30,rinvsq30,r30,qq30,c6_30,c12_30;
95 __m256d velec,felec,velecsum,facel,crf,krf,krf2;
97 __m256d dummy_mask,cutoff_mask;
98 __m128 tmpmask0,tmpmask1;
99 __m256d signbit = _mm256_castsi256_pd( _mm256_set1_epi32(0x80000000) );
100 __m256d one = _mm256_set1_pd(1.0);
101 __m256d two = _mm256_set1_pd(2.0);
107 jindex = nlist->jindex;
109 shiftidx = nlist->shift;
111 shiftvec = fr->shift_vec[0];
112 fshift = fr->fshift[0];
113 facel = _mm256_set1_pd(fr->epsfac);
114 charge = mdatoms->chargeA;
115 krf = _mm256_set1_pd(fr->ic->k_rf);
116 krf2 = _mm256_set1_pd(fr->ic->k_rf*2.0);
117 crf = _mm256_set1_pd(fr->ic->c_rf);
119 /* Setup water-specific parameters */
120 inr = nlist->iinr[0];
121 iq1 = _mm256_mul_pd(facel,_mm256_set1_pd(charge[inr+1]));
122 iq2 = _mm256_mul_pd(facel,_mm256_set1_pd(charge[inr+2]));
123 iq3 = _mm256_mul_pd(facel,_mm256_set1_pd(charge[inr+3]));
125 /* When we use explicit cutoffs the value must be identical for elec and VdW, so use elec as an arbitrary choice */
126 rcutoff_scalar = fr->rcoulomb;
127 rcutoff = _mm256_set1_pd(rcutoff_scalar);
128 rcutoff2 = _mm256_mul_pd(rcutoff,rcutoff);
130 /* Avoid stupid compiler warnings */
131 jnrA = jnrB = jnrC = jnrD = 0;
140 for(iidx=0;iidx<4*DIM;iidx++)
145 /* Start outer loop over neighborlists */
146 for(iidx=0; iidx<nri; iidx++)
148 /* Load shift vector for this list */
149 i_shift_offset = DIM*shiftidx[iidx];
151 /* Load limits for loop over neighbors */
152 j_index_start = jindex[iidx];
153 j_index_end = jindex[iidx+1];
155 /* Get outer coordinate index */
157 i_coord_offset = DIM*inr;
159 /* Load i particle coords and add shift vector */
160 gmx_mm256_load_shift_and_3rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset+DIM,
161 &ix1,&iy1,&iz1,&ix2,&iy2,&iz2,&ix3,&iy3,&iz3);
163 fix1 = _mm256_setzero_pd();
164 fiy1 = _mm256_setzero_pd();
165 fiz1 = _mm256_setzero_pd();
166 fix2 = _mm256_setzero_pd();
167 fiy2 = _mm256_setzero_pd();
168 fiz2 = _mm256_setzero_pd();
169 fix3 = _mm256_setzero_pd();
170 fiy3 = _mm256_setzero_pd();
171 fiz3 = _mm256_setzero_pd();
173 /* Reset potential sums */
174 velecsum = _mm256_setzero_pd();
176 /* Start inner kernel loop */
177 for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
180 /* Get j neighbor index, and coordinate index */
185 j_coord_offsetA = DIM*jnrA;
186 j_coord_offsetB = DIM*jnrB;
187 j_coord_offsetC = DIM*jnrC;
188 j_coord_offsetD = DIM*jnrD;
190 /* load j atom coordinates */
191 gmx_mm256_load_1rvec_4ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
192 x+j_coord_offsetC,x+j_coord_offsetD,
195 /* Calculate displacement vector */
196 dx10 = _mm256_sub_pd(ix1,jx0);
197 dy10 = _mm256_sub_pd(iy1,jy0);
198 dz10 = _mm256_sub_pd(iz1,jz0);
199 dx20 = _mm256_sub_pd(ix2,jx0);
200 dy20 = _mm256_sub_pd(iy2,jy0);
201 dz20 = _mm256_sub_pd(iz2,jz0);
202 dx30 = _mm256_sub_pd(ix3,jx0);
203 dy30 = _mm256_sub_pd(iy3,jy0);
204 dz30 = _mm256_sub_pd(iz3,jz0);
206 /* Calculate squared distance and things based on it */
207 rsq10 = gmx_mm256_calc_rsq_pd(dx10,dy10,dz10);
208 rsq20 = gmx_mm256_calc_rsq_pd(dx20,dy20,dz20);
209 rsq30 = gmx_mm256_calc_rsq_pd(dx30,dy30,dz30);
211 rinv10 = gmx_mm256_invsqrt_pd(rsq10);
212 rinv20 = gmx_mm256_invsqrt_pd(rsq20);
213 rinv30 = gmx_mm256_invsqrt_pd(rsq30);
215 rinvsq10 = _mm256_mul_pd(rinv10,rinv10);
216 rinvsq20 = _mm256_mul_pd(rinv20,rinv20);
217 rinvsq30 = _mm256_mul_pd(rinv30,rinv30);
219 /* Load parameters for j particles */
220 jq0 = gmx_mm256_load_4real_swizzle_pd(charge+jnrA+0,charge+jnrB+0,
221 charge+jnrC+0,charge+jnrD+0);
223 fjx0 = _mm256_setzero_pd();
224 fjy0 = _mm256_setzero_pd();
225 fjz0 = _mm256_setzero_pd();
227 /**************************
228 * CALCULATE INTERACTIONS *
229 **************************/
231 if (gmx_mm256_any_lt(rsq10,rcutoff2))
234 /* Compute parameters for interactions between i and j atoms */
235 qq10 = _mm256_mul_pd(iq1,jq0);
237 /* REACTION-FIELD ELECTROSTATICS */
238 velec = _mm256_mul_pd(qq10,_mm256_sub_pd(_mm256_add_pd(rinv10,_mm256_mul_pd(krf,rsq10)),crf));
239 felec = _mm256_mul_pd(qq10,_mm256_sub_pd(_mm256_mul_pd(rinv10,rinvsq10),krf2));
241 cutoff_mask = _mm256_cmp_pd(rsq10,rcutoff2,_CMP_LT_OQ);
243 /* Update potential sum for this i atom from the interaction with this j atom. */
244 velec = _mm256_and_pd(velec,cutoff_mask);
245 velecsum = _mm256_add_pd(velecsum,velec);
249 fscal = _mm256_and_pd(fscal,cutoff_mask);
251 /* Calculate temporary vectorial force */
252 tx = _mm256_mul_pd(fscal,dx10);
253 ty = _mm256_mul_pd(fscal,dy10);
254 tz = _mm256_mul_pd(fscal,dz10);
256 /* Update vectorial force */
257 fix1 = _mm256_add_pd(fix1,tx);
258 fiy1 = _mm256_add_pd(fiy1,ty);
259 fiz1 = _mm256_add_pd(fiz1,tz);
261 fjx0 = _mm256_add_pd(fjx0,tx);
262 fjy0 = _mm256_add_pd(fjy0,ty);
263 fjz0 = _mm256_add_pd(fjz0,tz);
267 /**************************
268 * CALCULATE INTERACTIONS *
269 **************************/
271 if (gmx_mm256_any_lt(rsq20,rcutoff2))
274 /* Compute parameters for interactions between i and j atoms */
275 qq20 = _mm256_mul_pd(iq2,jq0);
277 /* REACTION-FIELD ELECTROSTATICS */
278 velec = _mm256_mul_pd(qq20,_mm256_sub_pd(_mm256_add_pd(rinv20,_mm256_mul_pd(krf,rsq20)),crf));
279 felec = _mm256_mul_pd(qq20,_mm256_sub_pd(_mm256_mul_pd(rinv20,rinvsq20),krf2));
281 cutoff_mask = _mm256_cmp_pd(rsq20,rcutoff2,_CMP_LT_OQ);
283 /* Update potential sum for this i atom from the interaction with this j atom. */
284 velec = _mm256_and_pd(velec,cutoff_mask);
285 velecsum = _mm256_add_pd(velecsum,velec);
289 fscal = _mm256_and_pd(fscal,cutoff_mask);
291 /* Calculate temporary vectorial force */
292 tx = _mm256_mul_pd(fscal,dx20);
293 ty = _mm256_mul_pd(fscal,dy20);
294 tz = _mm256_mul_pd(fscal,dz20);
296 /* Update vectorial force */
297 fix2 = _mm256_add_pd(fix2,tx);
298 fiy2 = _mm256_add_pd(fiy2,ty);
299 fiz2 = _mm256_add_pd(fiz2,tz);
301 fjx0 = _mm256_add_pd(fjx0,tx);
302 fjy0 = _mm256_add_pd(fjy0,ty);
303 fjz0 = _mm256_add_pd(fjz0,tz);
307 /**************************
308 * CALCULATE INTERACTIONS *
309 **************************/
311 if (gmx_mm256_any_lt(rsq30,rcutoff2))
314 /* Compute parameters for interactions between i and j atoms */
315 qq30 = _mm256_mul_pd(iq3,jq0);
317 /* REACTION-FIELD ELECTROSTATICS */
318 velec = _mm256_mul_pd(qq30,_mm256_sub_pd(_mm256_add_pd(rinv30,_mm256_mul_pd(krf,rsq30)),crf));
319 felec = _mm256_mul_pd(qq30,_mm256_sub_pd(_mm256_mul_pd(rinv30,rinvsq30),krf2));
321 cutoff_mask = _mm256_cmp_pd(rsq30,rcutoff2,_CMP_LT_OQ);
323 /* Update potential sum for this i atom from the interaction with this j atom. */
324 velec = _mm256_and_pd(velec,cutoff_mask);
325 velecsum = _mm256_add_pd(velecsum,velec);
329 fscal = _mm256_and_pd(fscal,cutoff_mask);
331 /* Calculate temporary vectorial force */
332 tx = _mm256_mul_pd(fscal,dx30);
333 ty = _mm256_mul_pd(fscal,dy30);
334 tz = _mm256_mul_pd(fscal,dz30);
336 /* Update vectorial force */
337 fix3 = _mm256_add_pd(fix3,tx);
338 fiy3 = _mm256_add_pd(fiy3,ty);
339 fiz3 = _mm256_add_pd(fiz3,tz);
341 fjx0 = _mm256_add_pd(fjx0,tx);
342 fjy0 = _mm256_add_pd(fjy0,ty);
343 fjz0 = _mm256_add_pd(fjz0,tz);
347 fjptrA = f+j_coord_offsetA;
348 fjptrB = f+j_coord_offsetB;
349 fjptrC = f+j_coord_offsetC;
350 fjptrD = f+j_coord_offsetD;
352 gmx_mm256_decrement_1rvec_4ptr_swizzle_pd(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
354 /* Inner loop uses 111 flops */
360 /* Get j neighbor index, and coordinate index */
361 jnrlistA = jjnr[jidx];
362 jnrlistB = jjnr[jidx+1];
363 jnrlistC = jjnr[jidx+2];
364 jnrlistD = jjnr[jidx+3];
365 /* Sign of each element will be negative for non-real atoms.
366 * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
367 * so use it as val = _mm_andnot_pd(mask,val) to clear dummy entries.
369 tmpmask0 = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
371 tmpmask1 = _mm_permute_ps(tmpmask0,_GMX_MM_PERMUTE(3,3,2,2));
372 tmpmask0 = _mm_permute_ps(tmpmask0,_GMX_MM_PERMUTE(1,1,0,0));
373 dummy_mask = _mm256_castps_pd(gmx_mm256_set_m128(tmpmask1,tmpmask0));
375 jnrA = (jnrlistA>=0) ? jnrlistA : 0;
376 jnrB = (jnrlistB>=0) ? jnrlistB : 0;
377 jnrC = (jnrlistC>=0) ? jnrlistC : 0;
378 jnrD = (jnrlistD>=0) ? jnrlistD : 0;
379 j_coord_offsetA = DIM*jnrA;
380 j_coord_offsetB = DIM*jnrB;
381 j_coord_offsetC = DIM*jnrC;
382 j_coord_offsetD = DIM*jnrD;
384 /* load j atom coordinates */
385 gmx_mm256_load_1rvec_4ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
386 x+j_coord_offsetC,x+j_coord_offsetD,
389 /* Calculate displacement vector */
390 dx10 = _mm256_sub_pd(ix1,jx0);
391 dy10 = _mm256_sub_pd(iy1,jy0);
392 dz10 = _mm256_sub_pd(iz1,jz0);
393 dx20 = _mm256_sub_pd(ix2,jx0);
394 dy20 = _mm256_sub_pd(iy2,jy0);
395 dz20 = _mm256_sub_pd(iz2,jz0);
396 dx30 = _mm256_sub_pd(ix3,jx0);
397 dy30 = _mm256_sub_pd(iy3,jy0);
398 dz30 = _mm256_sub_pd(iz3,jz0);
400 /* Calculate squared distance and things based on it */
401 rsq10 = gmx_mm256_calc_rsq_pd(dx10,dy10,dz10);
402 rsq20 = gmx_mm256_calc_rsq_pd(dx20,dy20,dz20);
403 rsq30 = gmx_mm256_calc_rsq_pd(dx30,dy30,dz30);
405 rinv10 = gmx_mm256_invsqrt_pd(rsq10);
406 rinv20 = gmx_mm256_invsqrt_pd(rsq20);
407 rinv30 = gmx_mm256_invsqrt_pd(rsq30);
409 rinvsq10 = _mm256_mul_pd(rinv10,rinv10);
410 rinvsq20 = _mm256_mul_pd(rinv20,rinv20);
411 rinvsq30 = _mm256_mul_pd(rinv30,rinv30);
413 /* Load parameters for j particles */
414 jq0 = gmx_mm256_load_4real_swizzle_pd(charge+jnrA+0,charge+jnrB+0,
415 charge+jnrC+0,charge+jnrD+0);
417 fjx0 = _mm256_setzero_pd();
418 fjy0 = _mm256_setzero_pd();
419 fjz0 = _mm256_setzero_pd();
421 /**************************
422 * CALCULATE INTERACTIONS *
423 **************************/
425 if (gmx_mm256_any_lt(rsq10,rcutoff2))
428 /* Compute parameters for interactions between i and j atoms */
429 qq10 = _mm256_mul_pd(iq1,jq0);
431 /* REACTION-FIELD ELECTROSTATICS */
432 velec = _mm256_mul_pd(qq10,_mm256_sub_pd(_mm256_add_pd(rinv10,_mm256_mul_pd(krf,rsq10)),crf));
433 felec = _mm256_mul_pd(qq10,_mm256_sub_pd(_mm256_mul_pd(rinv10,rinvsq10),krf2));
435 cutoff_mask = _mm256_cmp_pd(rsq10,rcutoff2,_CMP_LT_OQ);
437 /* Update potential sum for this i atom from the interaction with this j atom. */
438 velec = _mm256_and_pd(velec,cutoff_mask);
439 velec = _mm256_andnot_pd(dummy_mask,velec);
440 velecsum = _mm256_add_pd(velecsum,velec);
444 fscal = _mm256_and_pd(fscal,cutoff_mask);
446 fscal = _mm256_andnot_pd(dummy_mask,fscal);
448 /* Calculate temporary vectorial force */
449 tx = _mm256_mul_pd(fscal,dx10);
450 ty = _mm256_mul_pd(fscal,dy10);
451 tz = _mm256_mul_pd(fscal,dz10);
453 /* Update vectorial force */
454 fix1 = _mm256_add_pd(fix1,tx);
455 fiy1 = _mm256_add_pd(fiy1,ty);
456 fiz1 = _mm256_add_pd(fiz1,tz);
458 fjx0 = _mm256_add_pd(fjx0,tx);
459 fjy0 = _mm256_add_pd(fjy0,ty);
460 fjz0 = _mm256_add_pd(fjz0,tz);
464 /**************************
465 * CALCULATE INTERACTIONS *
466 **************************/
468 if (gmx_mm256_any_lt(rsq20,rcutoff2))
471 /* Compute parameters for interactions between i and j atoms */
472 qq20 = _mm256_mul_pd(iq2,jq0);
474 /* REACTION-FIELD ELECTROSTATICS */
475 velec = _mm256_mul_pd(qq20,_mm256_sub_pd(_mm256_add_pd(rinv20,_mm256_mul_pd(krf,rsq20)),crf));
476 felec = _mm256_mul_pd(qq20,_mm256_sub_pd(_mm256_mul_pd(rinv20,rinvsq20),krf2));
478 cutoff_mask = _mm256_cmp_pd(rsq20,rcutoff2,_CMP_LT_OQ);
480 /* Update potential sum for this i atom from the interaction with this j atom. */
481 velec = _mm256_and_pd(velec,cutoff_mask);
482 velec = _mm256_andnot_pd(dummy_mask,velec);
483 velecsum = _mm256_add_pd(velecsum,velec);
487 fscal = _mm256_and_pd(fscal,cutoff_mask);
489 fscal = _mm256_andnot_pd(dummy_mask,fscal);
491 /* Calculate temporary vectorial force */
492 tx = _mm256_mul_pd(fscal,dx20);
493 ty = _mm256_mul_pd(fscal,dy20);
494 tz = _mm256_mul_pd(fscal,dz20);
496 /* Update vectorial force */
497 fix2 = _mm256_add_pd(fix2,tx);
498 fiy2 = _mm256_add_pd(fiy2,ty);
499 fiz2 = _mm256_add_pd(fiz2,tz);
501 fjx0 = _mm256_add_pd(fjx0,tx);
502 fjy0 = _mm256_add_pd(fjy0,ty);
503 fjz0 = _mm256_add_pd(fjz0,tz);
507 /**************************
508 * CALCULATE INTERACTIONS *
509 **************************/
511 if (gmx_mm256_any_lt(rsq30,rcutoff2))
514 /* Compute parameters for interactions between i and j atoms */
515 qq30 = _mm256_mul_pd(iq3,jq0);
517 /* REACTION-FIELD ELECTROSTATICS */
518 velec = _mm256_mul_pd(qq30,_mm256_sub_pd(_mm256_add_pd(rinv30,_mm256_mul_pd(krf,rsq30)),crf));
519 felec = _mm256_mul_pd(qq30,_mm256_sub_pd(_mm256_mul_pd(rinv30,rinvsq30),krf2));
521 cutoff_mask = _mm256_cmp_pd(rsq30,rcutoff2,_CMP_LT_OQ);
523 /* Update potential sum for this i atom from the interaction with this j atom. */
524 velec = _mm256_and_pd(velec,cutoff_mask);
525 velec = _mm256_andnot_pd(dummy_mask,velec);
526 velecsum = _mm256_add_pd(velecsum,velec);
530 fscal = _mm256_and_pd(fscal,cutoff_mask);
532 fscal = _mm256_andnot_pd(dummy_mask,fscal);
534 /* Calculate temporary vectorial force */
535 tx = _mm256_mul_pd(fscal,dx30);
536 ty = _mm256_mul_pd(fscal,dy30);
537 tz = _mm256_mul_pd(fscal,dz30);
539 /* Update vectorial force */
540 fix3 = _mm256_add_pd(fix3,tx);
541 fiy3 = _mm256_add_pd(fiy3,ty);
542 fiz3 = _mm256_add_pd(fiz3,tz);
544 fjx0 = _mm256_add_pd(fjx0,tx);
545 fjy0 = _mm256_add_pd(fjy0,ty);
546 fjz0 = _mm256_add_pd(fjz0,tz);
550 fjptrA = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
551 fjptrB = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
552 fjptrC = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
553 fjptrD = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
555 gmx_mm256_decrement_1rvec_4ptr_swizzle_pd(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
557 /* Inner loop uses 111 flops */
560 /* End of innermost loop */
562 gmx_mm256_update_iforce_3atom_swizzle_pd(fix1,fiy1,fiz1,fix2,fiy2,fiz2,fix3,fiy3,fiz3,
563 f+i_coord_offset+DIM,fshift+i_shift_offset);
566 /* Update potential energies */
567 gmx_mm256_update_1pot_pd(velecsum,kernel_data->energygrp_elec+ggid);
569 /* Increment number of inner iterations */
570 inneriter += j_index_end - j_index_start;
572 /* Outer loop uses 19 flops */
575 /* Increment number of outer iterations */
578 /* Update outer/inner flops */
580 inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_W4_VF,outeriter*19 + inneriter*111);
583 * Gromacs nonbonded kernel: nb_kernel_ElecRFCut_VdwNone_GeomW4P1_F_avx_256_double
584 * Electrostatics interaction: ReactionField
585 * VdW interaction: None
586 * Geometry: Water4-Particle
587 * Calculate force/pot: Force
590 nb_kernel_ElecRFCut_VdwNone_GeomW4P1_F_avx_256_double
591 (t_nblist * gmx_restrict nlist,
592 rvec * gmx_restrict xx,
593 rvec * gmx_restrict ff,
594 t_forcerec * gmx_restrict fr,
595 t_mdatoms * gmx_restrict mdatoms,
596 nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
597 t_nrnb * gmx_restrict nrnb)
599 /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
600 * just 0 for non-waters.
601 * Suffixes A,B,C,D refer to j loop unrolling done with AVX, e.g. for the four different
602 * jnr indices corresponding to data put in the four positions in the SIMD register.
604 int i_shift_offset,i_coord_offset,outeriter,inneriter;
605 int j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
606 int jnrA,jnrB,jnrC,jnrD;
607 int jnrlistA,jnrlistB,jnrlistC,jnrlistD;
608 int jnrlistE,jnrlistF,jnrlistG,jnrlistH;
609 int j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
610 int *iinr,*jindex,*jjnr,*shiftidx,*gid;
612 real *shiftvec,*fshift,*x,*f;
613 real *fjptrA,*fjptrB,*fjptrC,*fjptrD;
615 __m256d tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
616 real * vdwioffsetptr1;
617 __m256d ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
618 real * vdwioffsetptr2;
619 __m256d ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
620 real * vdwioffsetptr3;
621 __m256d ix3,iy3,iz3,fix3,fiy3,fiz3,iq3,isai3;
622 int vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
623 __m256d jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
624 __m256d dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
625 __m256d dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
626 __m256d dx30,dy30,dz30,rsq30,rinv30,rinvsq30,r30,qq30,c6_30,c12_30;
627 __m256d velec,felec,velecsum,facel,crf,krf,krf2;
629 __m256d dummy_mask,cutoff_mask;
630 __m128 tmpmask0,tmpmask1;
631 __m256d signbit = _mm256_castsi256_pd( _mm256_set1_epi32(0x80000000) );
632 __m256d one = _mm256_set1_pd(1.0);
633 __m256d two = _mm256_set1_pd(2.0);
639 jindex = nlist->jindex;
641 shiftidx = nlist->shift;
643 shiftvec = fr->shift_vec[0];
644 fshift = fr->fshift[0];
645 facel = _mm256_set1_pd(fr->epsfac);
646 charge = mdatoms->chargeA;
647 krf = _mm256_set1_pd(fr->ic->k_rf);
648 krf2 = _mm256_set1_pd(fr->ic->k_rf*2.0);
649 crf = _mm256_set1_pd(fr->ic->c_rf);
651 /* Setup water-specific parameters */
652 inr = nlist->iinr[0];
653 iq1 = _mm256_mul_pd(facel,_mm256_set1_pd(charge[inr+1]));
654 iq2 = _mm256_mul_pd(facel,_mm256_set1_pd(charge[inr+2]));
655 iq3 = _mm256_mul_pd(facel,_mm256_set1_pd(charge[inr+3]));
657 /* When we use explicit cutoffs the value must be identical for elec and VdW, so use elec as an arbitrary choice */
658 rcutoff_scalar = fr->rcoulomb;
659 rcutoff = _mm256_set1_pd(rcutoff_scalar);
660 rcutoff2 = _mm256_mul_pd(rcutoff,rcutoff);
662 /* Avoid stupid compiler warnings */
663 jnrA = jnrB = jnrC = jnrD = 0;
672 for(iidx=0;iidx<4*DIM;iidx++)
677 /* Start outer loop over neighborlists */
678 for(iidx=0; iidx<nri; iidx++)
680 /* Load shift vector for this list */
681 i_shift_offset = DIM*shiftidx[iidx];
683 /* Load limits for loop over neighbors */
684 j_index_start = jindex[iidx];
685 j_index_end = jindex[iidx+1];
687 /* Get outer coordinate index */
689 i_coord_offset = DIM*inr;
691 /* Load i particle coords and add shift vector */
692 gmx_mm256_load_shift_and_3rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset+DIM,
693 &ix1,&iy1,&iz1,&ix2,&iy2,&iz2,&ix3,&iy3,&iz3);
695 fix1 = _mm256_setzero_pd();
696 fiy1 = _mm256_setzero_pd();
697 fiz1 = _mm256_setzero_pd();
698 fix2 = _mm256_setzero_pd();
699 fiy2 = _mm256_setzero_pd();
700 fiz2 = _mm256_setzero_pd();
701 fix3 = _mm256_setzero_pd();
702 fiy3 = _mm256_setzero_pd();
703 fiz3 = _mm256_setzero_pd();
705 /* Start inner kernel loop */
706 for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
709 /* Get j neighbor index, and coordinate index */
714 j_coord_offsetA = DIM*jnrA;
715 j_coord_offsetB = DIM*jnrB;
716 j_coord_offsetC = DIM*jnrC;
717 j_coord_offsetD = DIM*jnrD;
719 /* load j atom coordinates */
720 gmx_mm256_load_1rvec_4ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
721 x+j_coord_offsetC,x+j_coord_offsetD,
724 /* Calculate displacement vector */
725 dx10 = _mm256_sub_pd(ix1,jx0);
726 dy10 = _mm256_sub_pd(iy1,jy0);
727 dz10 = _mm256_sub_pd(iz1,jz0);
728 dx20 = _mm256_sub_pd(ix2,jx0);
729 dy20 = _mm256_sub_pd(iy2,jy0);
730 dz20 = _mm256_sub_pd(iz2,jz0);
731 dx30 = _mm256_sub_pd(ix3,jx0);
732 dy30 = _mm256_sub_pd(iy3,jy0);
733 dz30 = _mm256_sub_pd(iz3,jz0);
735 /* Calculate squared distance and things based on it */
736 rsq10 = gmx_mm256_calc_rsq_pd(dx10,dy10,dz10);
737 rsq20 = gmx_mm256_calc_rsq_pd(dx20,dy20,dz20);
738 rsq30 = gmx_mm256_calc_rsq_pd(dx30,dy30,dz30);
740 rinv10 = gmx_mm256_invsqrt_pd(rsq10);
741 rinv20 = gmx_mm256_invsqrt_pd(rsq20);
742 rinv30 = gmx_mm256_invsqrt_pd(rsq30);
744 rinvsq10 = _mm256_mul_pd(rinv10,rinv10);
745 rinvsq20 = _mm256_mul_pd(rinv20,rinv20);
746 rinvsq30 = _mm256_mul_pd(rinv30,rinv30);
748 /* Load parameters for j particles */
749 jq0 = gmx_mm256_load_4real_swizzle_pd(charge+jnrA+0,charge+jnrB+0,
750 charge+jnrC+0,charge+jnrD+0);
752 fjx0 = _mm256_setzero_pd();
753 fjy0 = _mm256_setzero_pd();
754 fjz0 = _mm256_setzero_pd();
756 /**************************
757 * CALCULATE INTERACTIONS *
758 **************************/
760 if (gmx_mm256_any_lt(rsq10,rcutoff2))
763 /* Compute parameters for interactions between i and j atoms */
764 qq10 = _mm256_mul_pd(iq1,jq0);
766 /* REACTION-FIELD ELECTROSTATICS */
767 felec = _mm256_mul_pd(qq10,_mm256_sub_pd(_mm256_mul_pd(rinv10,rinvsq10),krf2));
769 cutoff_mask = _mm256_cmp_pd(rsq10,rcutoff2,_CMP_LT_OQ);
773 fscal = _mm256_and_pd(fscal,cutoff_mask);
775 /* Calculate temporary vectorial force */
776 tx = _mm256_mul_pd(fscal,dx10);
777 ty = _mm256_mul_pd(fscal,dy10);
778 tz = _mm256_mul_pd(fscal,dz10);
780 /* Update vectorial force */
781 fix1 = _mm256_add_pd(fix1,tx);
782 fiy1 = _mm256_add_pd(fiy1,ty);
783 fiz1 = _mm256_add_pd(fiz1,tz);
785 fjx0 = _mm256_add_pd(fjx0,tx);
786 fjy0 = _mm256_add_pd(fjy0,ty);
787 fjz0 = _mm256_add_pd(fjz0,tz);
791 /**************************
792 * CALCULATE INTERACTIONS *
793 **************************/
795 if (gmx_mm256_any_lt(rsq20,rcutoff2))
798 /* Compute parameters for interactions between i and j atoms */
799 qq20 = _mm256_mul_pd(iq2,jq0);
801 /* REACTION-FIELD ELECTROSTATICS */
802 felec = _mm256_mul_pd(qq20,_mm256_sub_pd(_mm256_mul_pd(rinv20,rinvsq20),krf2));
804 cutoff_mask = _mm256_cmp_pd(rsq20,rcutoff2,_CMP_LT_OQ);
808 fscal = _mm256_and_pd(fscal,cutoff_mask);
810 /* Calculate temporary vectorial force */
811 tx = _mm256_mul_pd(fscal,dx20);
812 ty = _mm256_mul_pd(fscal,dy20);
813 tz = _mm256_mul_pd(fscal,dz20);
815 /* Update vectorial force */
816 fix2 = _mm256_add_pd(fix2,tx);
817 fiy2 = _mm256_add_pd(fiy2,ty);
818 fiz2 = _mm256_add_pd(fiz2,tz);
820 fjx0 = _mm256_add_pd(fjx0,tx);
821 fjy0 = _mm256_add_pd(fjy0,ty);
822 fjz0 = _mm256_add_pd(fjz0,tz);
826 /**************************
827 * CALCULATE INTERACTIONS *
828 **************************/
830 if (gmx_mm256_any_lt(rsq30,rcutoff2))
833 /* Compute parameters for interactions between i and j atoms */
834 qq30 = _mm256_mul_pd(iq3,jq0);
836 /* REACTION-FIELD ELECTROSTATICS */
837 felec = _mm256_mul_pd(qq30,_mm256_sub_pd(_mm256_mul_pd(rinv30,rinvsq30),krf2));
839 cutoff_mask = _mm256_cmp_pd(rsq30,rcutoff2,_CMP_LT_OQ);
843 fscal = _mm256_and_pd(fscal,cutoff_mask);
845 /* Calculate temporary vectorial force */
846 tx = _mm256_mul_pd(fscal,dx30);
847 ty = _mm256_mul_pd(fscal,dy30);
848 tz = _mm256_mul_pd(fscal,dz30);
850 /* Update vectorial force */
851 fix3 = _mm256_add_pd(fix3,tx);
852 fiy3 = _mm256_add_pd(fiy3,ty);
853 fiz3 = _mm256_add_pd(fiz3,tz);
855 fjx0 = _mm256_add_pd(fjx0,tx);
856 fjy0 = _mm256_add_pd(fjy0,ty);
857 fjz0 = _mm256_add_pd(fjz0,tz);
861 fjptrA = f+j_coord_offsetA;
862 fjptrB = f+j_coord_offsetB;
863 fjptrC = f+j_coord_offsetC;
864 fjptrD = f+j_coord_offsetD;
866 gmx_mm256_decrement_1rvec_4ptr_swizzle_pd(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
868 /* Inner loop uses 93 flops */
874 /* Get j neighbor index, and coordinate index */
875 jnrlistA = jjnr[jidx];
876 jnrlistB = jjnr[jidx+1];
877 jnrlistC = jjnr[jidx+2];
878 jnrlistD = jjnr[jidx+3];
879 /* Sign of each element will be negative for non-real atoms.
880 * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
881 * so use it as val = _mm_andnot_pd(mask,val) to clear dummy entries.
883 tmpmask0 = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
885 tmpmask1 = _mm_permute_ps(tmpmask0,_GMX_MM_PERMUTE(3,3,2,2));
886 tmpmask0 = _mm_permute_ps(tmpmask0,_GMX_MM_PERMUTE(1,1,0,0));
887 dummy_mask = _mm256_castps_pd(gmx_mm256_set_m128(tmpmask1,tmpmask0));
889 jnrA = (jnrlistA>=0) ? jnrlistA : 0;
890 jnrB = (jnrlistB>=0) ? jnrlistB : 0;
891 jnrC = (jnrlistC>=0) ? jnrlistC : 0;
892 jnrD = (jnrlistD>=0) ? jnrlistD : 0;
893 j_coord_offsetA = DIM*jnrA;
894 j_coord_offsetB = DIM*jnrB;
895 j_coord_offsetC = DIM*jnrC;
896 j_coord_offsetD = DIM*jnrD;
898 /* load j atom coordinates */
899 gmx_mm256_load_1rvec_4ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
900 x+j_coord_offsetC,x+j_coord_offsetD,
903 /* Calculate displacement vector */
904 dx10 = _mm256_sub_pd(ix1,jx0);
905 dy10 = _mm256_sub_pd(iy1,jy0);
906 dz10 = _mm256_sub_pd(iz1,jz0);
907 dx20 = _mm256_sub_pd(ix2,jx0);
908 dy20 = _mm256_sub_pd(iy2,jy0);
909 dz20 = _mm256_sub_pd(iz2,jz0);
910 dx30 = _mm256_sub_pd(ix3,jx0);
911 dy30 = _mm256_sub_pd(iy3,jy0);
912 dz30 = _mm256_sub_pd(iz3,jz0);
914 /* Calculate squared distance and things based on it */
915 rsq10 = gmx_mm256_calc_rsq_pd(dx10,dy10,dz10);
916 rsq20 = gmx_mm256_calc_rsq_pd(dx20,dy20,dz20);
917 rsq30 = gmx_mm256_calc_rsq_pd(dx30,dy30,dz30);
919 rinv10 = gmx_mm256_invsqrt_pd(rsq10);
920 rinv20 = gmx_mm256_invsqrt_pd(rsq20);
921 rinv30 = gmx_mm256_invsqrt_pd(rsq30);
923 rinvsq10 = _mm256_mul_pd(rinv10,rinv10);
924 rinvsq20 = _mm256_mul_pd(rinv20,rinv20);
925 rinvsq30 = _mm256_mul_pd(rinv30,rinv30);
927 /* Load parameters for j particles */
928 jq0 = gmx_mm256_load_4real_swizzle_pd(charge+jnrA+0,charge+jnrB+0,
929 charge+jnrC+0,charge+jnrD+0);
931 fjx0 = _mm256_setzero_pd();
932 fjy0 = _mm256_setzero_pd();
933 fjz0 = _mm256_setzero_pd();
935 /**************************
936 * CALCULATE INTERACTIONS *
937 **************************/
939 if (gmx_mm256_any_lt(rsq10,rcutoff2))
942 /* Compute parameters for interactions between i and j atoms */
943 qq10 = _mm256_mul_pd(iq1,jq0);
945 /* REACTION-FIELD ELECTROSTATICS */
946 felec = _mm256_mul_pd(qq10,_mm256_sub_pd(_mm256_mul_pd(rinv10,rinvsq10),krf2));
948 cutoff_mask = _mm256_cmp_pd(rsq10,rcutoff2,_CMP_LT_OQ);
952 fscal = _mm256_and_pd(fscal,cutoff_mask);
954 fscal = _mm256_andnot_pd(dummy_mask,fscal);
956 /* Calculate temporary vectorial force */
957 tx = _mm256_mul_pd(fscal,dx10);
958 ty = _mm256_mul_pd(fscal,dy10);
959 tz = _mm256_mul_pd(fscal,dz10);
961 /* Update vectorial force */
962 fix1 = _mm256_add_pd(fix1,tx);
963 fiy1 = _mm256_add_pd(fiy1,ty);
964 fiz1 = _mm256_add_pd(fiz1,tz);
966 fjx0 = _mm256_add_pd(fjx0,tx);
967 fjy0 = _mm256_add_pd(fjy0,ty);
968 fjz0 = _mm256_add_pd(fjz0,tz);
972 /**************************
973 * CALCULATE INTERACTIONS *
974 **************************/
976 if (gmx_mm256_any_lt(rsq20,rcutoff2))
979 /* Compute parameters for interactions between i and j atoms */
980 qq20 = _mm256_mul_pd(iq2,jq0);
982 /* REACTION-FIELD ELECTROSTATICS */
983 felec = _mm256_mul_pd(qq20,_mm256_sub_pd(_mm256_mul_pd(rinv20,rinvsq20),krf2));
985 cutoff_mask = _mm256_cmp_pd(rsq20,rcutoff2,_CMP_LT_OQ);
989 fscal = _mm256_and_pd(fscal,cutoff_mask);
991 fscal = _mm256_andnot_pd(dummy_mask,fscal);
993 /* Calculate temporary vectorial force */
994 tx = _mm256_mul_pd(fscal,dx20);
995 ty = _mm256_mul_pd(fscal,dy20);
996 tz = _mm256_mul_pd(fscal,dz20);
998 /* Update vectorial force */
999 fix2 = _mm256_add_pd(fix2,tx);
1000 fiy2 = _mm256_add_pd(fiy2,ty);
1001 fiz2 = _mm256_add_pd(fiz2,tz);
1003 fjx0 = _mm256_add_pd(fjx0,tx);
1004 fjy0 = _mm256_add_pd(fjy0,ty);
1005 fjz0 = _mm256_add_pd(fjz0,tz);
1009 /**************************
1010 * CALCULATE INTERACTIONS *
1011 **************************/
1013 if (gmx_mm256_any_lt(rsq30,rcutoff2))
1016 /* Compute parameters for interactions between i and j atoms */
1017 qq30 = _mm256_mul_pd(iq3,jq0);
1019 /* REACTION-FIELD ELECTROSTATICS */
1020 felec = _mm256_mul_pd(qq30,_mm256_sub_pd(_mm256_mul_pd(rinv30,rinvsq30),krf2));
1022 cutoff_mask = _mm256_cmp_pd(rsq30,rcutoff2,_CMP_LT_OQ);
1026 fscal = _mm256_and_pd(fscal,cutoff_mask);
1028 fscal = _mm256_andnot_pd(dummy_mask,fscal);
1030 /* Calculate temporary vectorial force */
1031 tx = _mm256_mul_pd(fscal,dx30);
1032 ty = _mm256_mul_pd(fscal,dy30);
1033 tz = _mm256_mul_pd(fscal,dz30);
1035 /* Update vectorial force */
1036 fix3 = _mm256_add_pd(fix3,tx);
1037 fiy3 = _mm256_add_pd(fiy3,ty);
1038 fiz3 = _mm256_add_pd(fiz3,tz);
1040 fjx0 = _mm256_add_pd(fjx0,tx);
1041 fjy0 = _mm256_add_pd(fjy0,ty);
1042 fjz0 = _mm256_add_pd(fjz0,tz);
1046 fjptrA = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
1047 fjptrB = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
1048 fjptrC = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
1049 fjptrD = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
1051 gmx_mm256_decrement_1rvec_4ptr_swizzle_pd(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
1053 /* Inner loop uses 93 flops */
1056 /* End of innermost loop */
1058 gmx_mm256_update_iforce_3atom_swizzle_pd(fix1,fiy1,fiz1,fix2,fiy2,fiz2,fix3,fiy3,fiz3,
1059 f+i_coord_offset+DIM,fshift+i_shift_offset);
1061 /* Increment number of inner iterations */
1062 inneriter += j_index_end - j_index_start;
1064 /* Outer loop uses 18 flops */
1067 /* Increment number of outer iterations */
1070 /* Update outer/inner flops */
1072 inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_W4_F,outeriter*18 + inneriter*93);