2 * This file is part of the GROMACS molecular simulation package.
4 * Copyright (c) 2012,2013,2014,2015,2017,2018, by the GROMACS development team, led by
5 * Mark Abraham, David van der Spoel, Berk Hess, and Erik Lindahl,
6 * and including many others, as listed in the AUTHORS file in the
7 * top-level source directory and at http://www.gromacs.org.
9 * GROMACS is free software; you can redistribute it and/or
10 * modify it under the terms of the GNU Lesser General Public License
11 * as published by the Free Software Foundation; either version 2.1
12 * of the License, or (at your option) any later version.
14 * GROMACS is distributed in the hope that it will be useful,
15 * but WITHOUT ANY WARRANTY; without even the implied warranty of
16 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
17 * Lesser General Public License for more details.
19 * You should have received a copy of the GNU Lesser General Public
20 * License along with GROMACS; if not, see
21 * http://www.gnu.org/licenses, or write to the Free Software Foundation,
22 * Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA.
24 * If you want to redistribute modifications to GROMACS, please
25 * consider that scientific software is very special. Version
26 * control is crucial - bugs must be traceable. We will be happy to
27 * consider code for inclusion in the official distribution, but
28 * derived work must not be called official GROMACS. Details are found
29 * in the README & COPYING files - if they are missing, get the
30 * official version at http://www.gromacs.org.
32 * To help us fund GROMACS development, we humbly ask that you cite
33 * the research papers on the package. Check out http://www.gromacs.org.
36 * Note: this file was generated by the GROMACS avx_256_double kernel generator.
44 #include "../nb_kernel.h"
45 #include "gromacs/gmxlib/nrnb.h"
47 #include "kernelutil_x86_avx_256_double.h"
50 * Gromacs nonbonded kernel: nb_kernel_ElecCoul_VdwNone_GeomW3P1_VF_avx_256_double
51 * Electrostatics interaction: Coulomb
52 * VdW interaction: None
53 * Geometry: Water3-Particle
54 * Calculate force/pot: PotentialAndForce
57 nb_kernel_ElecCoul_VdwNone_GeomW3P1_VF_avx_256_double
58 (t_nblist * gmx_restrict nlist,
59 rvec * gmx_restrict xx,
60 rvec * gmx_restrict ff,
61 struct t_forcerec * gmx_restrict fr,
62 t_mdatoms * gmx_restrict mdatoms,
63 nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
64 t_nrnb * gmx_restrict nrnb)
66 /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
67 * just 0 for non-waters.
68 * Suffixes A,B,C,D refer to j loop unrolling done with AVX, e.g. for the four different
69 * jnr indices corresponding to data put in the four positions in the SIMD register.
71 int i_shift_offset,i_coord_offset,outeriter,inneriter;
72 int j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
73 int jnrA,jnrB,jnrC,jnrD;
74 int jnrlistA,jnrlistB,jnrlistC,jnrlistD;
75 int jnrlistE,jnrlistF,jnrlistG,jnrlistH;
76 int j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
77 int *iinr,*jindex,*jjnr,*shiftidx,*gid;
79 real *shiftvec,*fshift,*x,*f;
80 real *fjptrA,*fjptrB,*fjptrC,*fjptrD;
82 __m256d tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
83 real * vdwioffsetptr0;
84 __m256d ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
85 real * vdwioffsetptr1;
86 __m256d ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
87 real * vdwioffsetptr2;
88 __m256d ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
89 int vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
90 __m256d jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
91 __m256d dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
92 __m256d dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
93 __m256d dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
94 __m256d velec,felec,velecsum,facel,crf,krf,krf2;
96 __m256d dummy_mask,cutoff_mask;
97 __m128 tmpmask0,tmpmask1;
98 __m256d signbit = _mm256_castsi256_pd( _mm256_set1_epi32(0x80000000) );
99 __m256d one = _mm256_set1_pd(1.0);
100 __m256d two = _mm256_set1_pd(2.0);
106 jindex = nlist->jindex;
108 shiftidx = nlist->shift;
110 shiftvec = fr->shift_vec[0];
111 fshift = fr->fshift[0];
112 facel = _mm256_set1_pd(fr->ic->epsfac);
113 charge = mdatoms->chargeA;
115 /* Setup water-specific parameters */
116 inr = nlist->iinr[0];
117 iq0 = _mm256_mul_pd(facel,_mm256_set1_pd(charge[inr+0]));
118 iq1 = _mm256_mul_pd(facel,_mm256_set1_pd(charge[inr+1]));
119 iq2 = _mm256_mul_pd(facel,_mm256_set1_pd(charge[inr+2]));
121 /* Avoid stupid compiler warnings */
122 jnrA = jnrB = jnrC = jnrD = 0;
131 for(iidx=0;iidx<4*DIM;iidx++)
136 /* Start outer loop over neighborlists */
137 for(iidx=0; iidx<nri; iidx++)
139 /* Load shift vector for this list */
140 i_shift_offset = DIM*shiftidx[iidx];
142 /* Load limits for loop over neighbors */
143 j_index_start = jindex[iidx];
144 j_index_end = jindex[iidx+1];
146 /* Get outer coordinate index */
148 i_coord_offset = DIM*inr;
150 /* Load i particle coords and add shift vector */
151 gmx_mm256_load_shift_and_3rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset,
152 &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2);
154 fix0 = _mm256_setzero_pd();
155 fiy0 = _mm256_setzero_pd();
156 fiz0 = _mm256_setzero_pd();
157 fix1 = _mm256_setzero_pd();
158 fiy1 = _mm256_setzero_pd();
159 fiz1 = _mm256_setzero_pd();
160 fix2 = _mm256_setzero_pd();
161 fiy2 = _mm256_setzero_pd();
162 fiz2 = _mm256_setzero_pd();
164 /* Reset potential sums */
165 velecsum = _mm256_setzero_pd();
167 /* Start inner kernel loop */
168 for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
171 /* Get j neighbor index, and coordinate index */
176 j_coord_offsetA = DIM*jnrA;
177 j_coord_offsetB = DIM*jnrB;
178 j_coord_offsetC = DIM*jnrC;
179 j_coord_offsetD = DIM*jnrD;
181 /* load j atom coordinates */
182 gmx_mm256_load_1rvec_4ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
183 x+j_coord_offsetC,x+j_coord_offsetD,
186 /* Calculate displacement vector */
187 dx00 = _mm256_sub_pd(ix0,jx0);
188 dy00 = _mm256_sub_pd(iy0,jy0);
189 dz00 = _mm256_sub_pd(iz0,jz0);
190 dx10 = _mm256_sub_pd(ix1,jx0);
191 dy10 = _mm256_sub_pd(iy1,jy0);
192 dz10 = _mm256_sub_pd(iz1,jz0);
193 dx20 = _mm256_sub_pd(ix2,jx0);
194 dy20 = _mm256_sub_pd(iy2,jy0);
195 dz20 = _mm256_sub_pd(iz2,jz0);
197 /* Calculate squared distance and things based on it */
198 rsq00 = gmx_mm256_calc_rsq_pd(dx00,dy00,dz00);
199 rsq10 = gmx_mm256_calc_rsq_pd(dx10,dy10,dz10);
200 rsq20 = gmx_mm256_calc_rsq_pd(dx20,dy20,dz20);
202 rinv00 = avx256_invsqrt_d(rsq00);
203 rinv10 = avx256_invsqrt_d(rsq10);
204 rinv20 = avx256_invsqrt_d(rsq20);
206 rinvsq00 = _mm256_mul_pd(rinv00,rinv00);
207 rinvsq10 = _mm256_mul_pd(rinv10,rinv10);
208 rinvsq20 = _mm256_mul_pd(rinv20,rinv20);
210 /* Load parameters for j particles */
211 jq0 = gmx_mm256_load_4real_swizzle_pd(charge+jnrA+0,charge+jnrB+0,
212 charge+jnrC+0,charge+jnrD+0);
214 fjx0 = _mm256_setzero_pd();
215 fjy0 = _mm256_setzero_pd();
216 fjz0 = _mm256_setzero_pd();
218 /**************************
219 * CALCULATE INTERACTIONS *
220 **************************/
222 /* Compute parameters for interactions between i and j atoms */
223 qq00 = _mm256_mul_pd(iq0,jq0);
225 /* COULOMB ELECTROSTATICS */
226 velec = _mm256_mul_pd(qq00,rinv00);
227 felec = _mm256_mul_pd(velec,rinvsq00);
229 /* Update potential sum for this i atom from the interaction with this j atom. */
230 velecsum = _mm256_add_pd(velecsum,velec);
234 /* Calculate temporary vectorial force */
235 tx = _mm256_mul_pd(fscal,dx00);
236 ty = _mm256_mul_pd(fscal,dy00);
237 tz = _mm256_mul_pd(fscal,dz00);
239 /* Update vectorial force */
240 fix0 = _mm256_add_pd(fix0,tx);
241 fiy0 = _mm256_add_pd(fiy0,ty);
242 fiz0 = _mm256_add_pd(fiz0,tz);
244 fjx0 = _mm256_add_pd(fjx0,tx);
245 fjy0 = _mm256_add_pd(fjy0,ty);
246 fjz0 = _mm256_add_pd(fjz0,tz);
248 /**************************
249 * CALCULATE INTERACTIONS *
250 **************************/
252 /* Compute parameters for interactions between i and j atoms */
253 qq10 = _mm256_mul_pd(iq1,jq0);
255 /* COULOMB ELECTROSTATICS */
256 velec = _mm256_mul_pd(qq10,rinv10);
257 felec = _mm256_mul_pd(velec,rinvsq10);
259 /* Update potential sum for this i atom from the interaction with this j atom. */
260 velecsum = _mm256_add_pd(velecsum,velec);
264 /* Calculate temporary vectorial force */
265 tx = _mm256_mul_pd(fscal,dx10);
266 ty = _mm256_mul_pd(fscal,dy10);
267 tz = _mm256_mul_pd(fscal,dz10);
269 /* Update vectorial force */
270 fix1 = _mm256_add_pd(fix1,tx);
271 fiy1 = _mm256_add_pd(fiy1,ty);
272 fiz1 = _mm256_add_pd(fiz1,tz);
274 fjx0 = _mm256_add_pd(fjx0,tx);
275 fjy0 = _mm256_add_pd(fjy0,ty);
276 fjz0 = _mm256_add_pd(fjz0,tz);
278 /**************************
279 * CALCULATE INTERACTIONS *
280 **************************/
282 /* Compute parameters for interactions between i and j atoms */
283 qq20 = _mm256_mul_pd(iq2,jq0);
285 /* COULOMB ELECTROSTATICS */
286 velec = _mm256_mul_pd(qq20,rinv20);
287 felec = _mm256_mul_pd(velec,rinvsq20);
289 /* Update potential sum for this i atom from the interaction with this j atom. */
290 velecsum = _mm256_add_pd(velecsum,velec);
294 /* Calculate temporary vectorial force */
295 tx = _mm256_mul_pd(fscal,dx20);
296 ty = _mm256_mul_pd(fscal,dy20);
297 tz = _mm256_mul_pd(fscal,dz20);
299 /* Update vectorial force */
300 fix2 = _mm256_add_pd(fix2,tx);
301 fiy2 = _mm256_add_pd(fiy2,ty);
302 fiz2 = _mm256_add_pd(fiz2,tz);
304 fjx0 = _mm256_add_pd(fjx0,tx);
305 fjy0 = _mm256_add_pd(fjy0,ty);
306 fjz0 = _mm256_add_pd(fjz0,tz);
308 fjptrA = f+j_coord_offsetA;
309 fjptrB = f+j_coord_offsetB;
310 fjptrC = f+j_coord_offsetC;
311 fjptrD = f+j_coord_offsetD;
313 gmx_mm256_decrement_1rvec_4ptr_swizzle_pd(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
315 /* Inner loop uses 84 flops */
321 /* Get j neighbor index, and coordinate index */
322 jnrlistA = jjnr[jidx];
323 jnrlistB = jjnr[jidx+1];
324 jnrlistC = jjnr[jidx+2];
325 jnrlistD = jjnr[jidx+3];
326 /* Sign of each element will be negative for non-real atoms.
327 * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
328 * so use it as val = _mm_andnot_pd(mask,val) to clear dummy entries.
330 tmpmask0 = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
332 tmpmask1 = _mm_permute_ps(tmpmask0,_GMX_MM_PERMUTE(3,3,2,2));
333 tmpmask0 = _mm_permute_ps(tmpmask0,_GMX_MM_PERMUTE(1,1,0,0));
334 dummy_mask = _mm256_castps_pd(gmx_mm256_set_m128(tmpmask1,tmpmask0));
336 jnrA = (jnrlistA>=0) ? jnrlistA : 0;
337 jnrB = (jnrlistB>=0) ? jnrlistB : 0;
338 jnrC = (jnrlistC>=0) ? jnrlistC : 0;
339 jnrD = (jnrlistD>=0) ? jnrlistD : 0;
340 j_coord_offsetA = DIM*jnrA;
341 j_coord_offsetB = DIM*jnrB;
342 j_coord_offsetC = DIM*jnrC;
343 j_coord_offsetD = DIM*jnrD;
345 /* load j atom coordinates */
346 gmx_mm256_load_1rvec_4ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
347 x+j_coord_offsetC,x+j_coord_offsetD,
350 /* Calculate displacement vector */
351 dx00 = _mm256_sub_pd(ix0,jx0);
352 dy00 = _mm256_sub_pd(iy0,jy0);
353 dz00 = _mm256_sub_pd(iz0,jz0);
354 dx10 = _mm256_sub_pd(ix1,jx0);
355 dy10 = _mm256_sub_pd(iy1,jy0);
356 dz10 = _mm256_sub_pd(iz1,jz0);
357 dx20 = _mm256_sub_pd(ix2,jx0);
358 dy20 = _mm256_sub_pd(iy2,jy0);
359 dz20 = _mm256_sub_pd(iz2,jz0);
361 /* Calculate squared distance and things based on it */
362 rsq00 = gmx_mm256_calc_rsq_pd(dx00,dy00,dz00);
363 rsq10 = gmx_mm256_calc_rsq_pd(dx10,dy10,dz10);
364 rsq20 = gmx_mm256_calc_rsq_pd(dx20,dy20,dz20);
366 rinv00 = avx256_invsqrt_d(rsq00);
367 rinv10 = avx256_invsqrt_d(rsq10);
368 rinv20 = avx256_invsqrt_d(rsq20);
370 rinvsq00 = _mm256_mul_pd(rinv00,rinv00);
371 rinvsq10 = _mm256_mul_pd(rinv10,rinv10);
372 rinvsq20 = _mm256_mul_pd(rinv20,rinv20);
374 /* Load parameters for j particles */
375 jq0 = gmx_mm256_load_4real_swizzle_pd(charge+jnrA+0,charge+jnrB+0,
376 charge+jnrC+0,charge+jnrD+0);
378 fjx0 = _mm256_setzero_pd();
379 fjy0 = _mm256_setzero_pd();
380 fjz0 = _mm256_setzero_pd();
382 /**************************
383 * CALCULATE INTERACTIONS *
384 **************************/
386 /* Compute parameters for interactions between i and j atoms */
387 qq00 = _mm256_mul_pd(iq0,jq0);
389 /* COULOMB ELECTROSTATICS */
390 velec = _mm256_mul_pd(qq00,rinv00);
391 felec = _mm256_mul_pd(velec,rinvsq00);
393 /* Update potential sum for this i atom from the interaction with this j atom. */
394 velec = _mm256_andnot_pd(dummy_mask,velec);
395 velecsum = _mm256_add_pd(velecsum,velec);
399 fscal = _mm256_andnot_pd(dummy_mask,fscal);
401 /* Calculate temporary vectorial force */
402 tx = _mm256_mul_pd(fscal,dx00);
403 ty = _mm256_mul_pd(fscal,dy00);
404 tz = _mm256_mul_pd(fscal,dz00);
406 /* Update vectorial force */
407 fix0 = _mm256_add_pd(fix0,tx);
408 fiy0 = _mm256_add_pd(fiy0,ty);
409 fiz0 = _mm256_add_pd(fiz0,tz);
411 fjx0 = _mm256_add_pd(fjx0,tx);
412 fjy0 = _mm256_add_pd(fjy0,ty);
413 fjz0 = _mm256_add_pd(fjz0,tz);
415 /**************************
416 * CALCULATE INTERACTIONS *
417 **************************/
419 /* Compute parameters for interactions between i and j atoms */
420 qq10 = _mm256_mul_pd(iq1,jq0);
422 /* COULOMB ELECTROSTATICS */
423 velec = _mm256_mul_pd(qq10,rinv10);
424 felec = _mm256_mul_pd(velec,rinvsq10);
426 /* Update potential sum for this i atom from the interaction with this j atom. */
427 velec = _mm256_andnot_pd(dummy_mask,velec);
428 velecsum = _mm256_add_pd(velecsum,velec);
432 fscal = _mm256_andnot_pd(dummy_mask,fscal);
434 /* Calculate temporary vectorial force */
435 tx = _mm256_mul_pd(fscal,dx10);
436 ty = _mm256_mul_pd(fscal,dy10);
437 tz = _mm256_mul_pd(fscal,dz10);
439 /* Update vectorial force */
440 fix1 = _mm256_add_pd(fix1,tx);
441 fiy1 = _mm256_add_pd(fiy1,ty);
442 fiz1 = _mm256_add_pd(fiz1,tz);
444 fjx0 = _mm256_add_pd(fjx0,tx);
445 fjy0 = _mm256_add_pd(fjy0,ty);
446 fjz0 = _mm256_add_pd(fjz0,tz);
448 /**************************
449 * CALCULATE INTERACTIONS *
450 **************************/
452 /* Compute parameters for interactions between i and j atoms */
453 qq20 = _mm256_mul_pd(iq2,jq0);
455 /* COULOMB ELECTROSTATICS */
456 velec = _mm256_mul_pd(qq20,rinv20);
457 felec = _mm256_mul_pd(velec,rinvsq20);
459 /* Update potential sum for this i atom from the interaction with this j atom. */
460 velec = _mm256_andnot_pd(dummy_mask,velec);
461 velecsum = _mm256_add_pd(velecsum,velec);
465 fscal = _mm256_andnot_pd(dummy_mask,fscal);
467 /* Calculate temporary vectorial force */
468 tx = _mm256_mul_pd(fscal,dx20);
469 ty = _mm256_mul_pd(fscal,dy20);
470 tz = _mm256_mul_pd(fscal,dz20);
472 /* Update vectorial force */
473 fix2 = _mm256_add_pd(fix2,tx);
474 fiy2 = _mm256_add_pd(fiy2,ty);
475 fiz2 = _mm256_add_pd(fiz2,tz);
477 fjx0 = _mm256_add_pd(fjx0,tx);
478 fjy0 = _mm256_add_pd(fjy0,ty);
479 fjz0 = _mm256_add_pd(fjz0,tz);
481 fjptrA = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
482 fjptrB = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
483 fjptrC = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
484 fjptrD = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
486 gmx_mm256_decrement_1rvec_4ptr_swizzle_pd(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
488 /* Inner loop uses 84 flops */
491 /* End of innermost loop */
493 gmx_mm256_update_iforce_3atom_swizzle_pd(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,
494 f+i_coord_offset,fshift+i_shift_offset);
497 /* Update potential energies */
498 gmx_mm256_update_1pot_pd(velecsum,kernel_data->energygrp_elec+ggid);
500 /* Increment number of inner iterations */
501 inneriter += j_index_end - j_index_start;
503 /* Outer loop uses 19 flops */
506 /* Increment number of outer iterations */
509 /* Update outer/inner flops */
511 inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_W3_VF,outeriter*19 + inneriter*84);
514 * Gromacs nonbonded kernel: nb_kernel_ElecCoul_VdwNone_GeomW3P1_F_avx_256_double
515 * Electrostatics interaction: Coulomb
516 * VdW interaction: None
517 * Geometry: Water3-Particle
518 * Calculate force/pot: Force
521 nb_kernel_ElecCoul_VdwNone_GeomW3P1_F_avx_256_double
522 (t_nblist * gmx_restrict nlist,
523 rvec * gmx_restrict xx,
524 rvec * gmx_restrict ff,
525 struct t_forcerec * gmx_restrict fr,
526 t_mdatoms * gmx_restrict mdatoms,
527 nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
528 t_nrnb * gmx_restrict nrnb)
530 /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
531 * just 0 for non-waters.
532 * Suffixes A,B,C,D refer to j loop unrolling done with AVX, e.g. for the four different
533 * jnr indices corresponding to data put in the four positions in the SIMD register.
535 int i_shift_offset,i_coord_offset,outeriter,inneriter;
536 int j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
537 int jnrA,jnrB,jnrC,jnrD;
538 int jnrlistA,jnrlistB,jnrlistC,jnrlistD;
539 int jnrlistE,jnrlistF,jnrlistG,jnrlistH;
540 int j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
541 int *iinr,*jindex,*jjnr,*shiftidx,*gid;
543 real *shiftvec,*fshift,*x,*f;
544 real *fjptrA,*fjptrB,*fjptrC,*fjptrD;
546 __m256d tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
547 real * vdwioffsetptr0;
548 __m256d ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
549 real * vdwioffsetptr1;
550 __m256d ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
551 real * vdwioffsetptr2;
552 __m256d ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
553 int vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
554 __m256d jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
555 __m256d dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
556 __m256d dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
557 __m256d dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
558 __m256d velec,felec,velecsum,facel,crf,krf,krf2;
560 __m256d dummy_mask,cutoff_mask;
561 __m128 tmpmask0,tmpmask1;
562 __m256d signbit = _mm256_castsi256_pd( _mm256_set1_epi32(0x80000000) );
563 __m256d one = _mm256_set1_pd(1.0);
564 __m256d two = _mm256_set1_pd(2.0);
570 jindex = nlist->jindex;
572 shiftidx = nlist->shift;
574 shiftvec = fr->shift_vec[0];
575 fshift = fr->fshift[0];
576 facel = _mm256_set1_pd(fr->ic->epsfac);
577 charge = mdatoms->chargeA;
579 /* Setup water-specific parameters */
580 inr = nlist->iinr[0];
581 iq0 = _mm256_mul_pd(facel,_mm256_set1_pd(charge[inr+0]));
582 iq1 = _mm256_mul_pd(facel,_mm256_set1_pd(charge[inr+1]));
583 iq2 = _mm256_mul_pd(facel,_mm256_set1_pd(charge[inr+2]));
585 /* Avoid stupid compiler warnings */
586 jnrA = jnrB = jnrC = jnrD = 0;
595 for(iidx=0;iidx<4*DIM;iidx++)
600 /* Start outer loop over neighborlists */
601 for(iidx=0; iidx<nri; iidx++)
603 /* Load shift vector for this list */
604 i_shift_offset = DIM*shiftidx[iidx];
606 /* Load limits for loop over neighbors */
607 j_index_start = jindex[iidx];
608 j_index_end = jindex[iidx+1];
610 /* Get outer coordinate index */
612 i_coord_offset = DIM*inr;
614 /* Load i particle coords and add shift vector */
615 gmx_mm256_load_shift_and_3rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset,
616 &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2);
618 fix0 = _mm256_setzero_pd();
619 fiy0 = _mm256_setzero_pd();
620 fiz0 = _mm256_setzero_pd();
621 fix1 = _mm256_setzero_pd();
622 fiy1 = _mm256_setzero_pd();
623 fiz1 = _mm256_setzero_pd();
624 fix2 = _mm256_setzero_pd();
625 fiy2 = _mm256_setzero_pd();
626 fiz2 = _mm256_setzero_pd();
628 /* Start inner kernel loop */
629 for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
632 /* Get j neighbor index, and coordinate index */
637 j_coord_offsetA = DIM*jnrA;
638 j_coord_offsetB = DIM*jnrB;
639 j_coord_offsetC = DIM*jnrC;
640 j_coord_offsetD = DIM*jnrD;
642 /* load j atom coordinates */
643 gmx_mm256_load_1rvec_4ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
644 x+j_coord_offsetC,x+j_coord_offsetD,
647 /* Calculate displacement vector */
648 dx00 = _mm256_sub_pd(ix0,jx0);
649 dy00 = _mm256_sub_pd(iy0,jy0);
650 dz00 = _mm256_sub_pd(iz0,jz0);
651 dx10 = _mm256_sub_pd(ix1,jx0);
652 dy10 = _mm256_sub_pd(iy1,jy0);
653 dz10 = _mm256_sub_pd(iz1,jz0);
654 dx20 = _mm256_sub_pd(ix2,jx0);
655 dy20 = _mm256_sub_pd(iy2,jy0);
656 dz20 = _mm256_sub_pd(iz2,jz0);
658 /* Calculate squared distance and things based on it */
659 rsq00 = gmx_mm256_calc_rsq_pd(dx00,dy00,dz00);
660 rsq10 = gmx_mm256_calc_rsq_pd(dx10,dy10,dz10);
661 rsq20 = gmx_mm256_calc_rsq_pd(dx20,dy20,dz20);
663 rinv00 = avx256_invsqrt_d(rsq00);
664 rinv10 = avx256_invsqrt_d(rsq10);
665 rinv20 = avx256_invsqrt_d(rsq20);
667 rinvsq00 = _mm256_mul_pd(rinv00,rinv00);
668 rinvsq10 = _mm256_mul_pd(rinv10,rinv10);
669 rinvsq20 = _mm256_mul_pd(rinv20,rinv20);
671 /* Load parameters for j particles */
672 jq0 = gmx_mm256_load_4real_swizzle_pd(charge+jnrA+0,charge+jnrB+0,
673 charge+jnrC+0,charge+jnrD+0);
675 fjx0 = _mm256_setzero_pd();
676 fjy0 = _mm256_setzero_pd();
677 fjz0 = _mm256_setzero_pd();
679 /**************************
680 * CALCULATE INTERACTIONS *
681 **************************/
683 /* Compute parameters for interactions between i and j atoms */
684 qq00 = _mm256_mul_pd(iq0,jq0);
686 /* COULOMB ELECTROSTATICS */
687 velec = _mm256_mul_pd(qq00,rinv00);
688 felec = _mm256_mul_pd(velec,rinvsq00);
692 /* Calculate temporary vectorial force */
693 tx = _mm256_mul_pd(fscal,dx00);
694 ty = _mm256_mul_pd(fscal,dy00);
695 tz = _mm256_mul_pd(fscal,dz00);
697 /* Update vectorial force */
698 fix0 = _mm256_add_pd(fix0,tx);
699 fiy0 = _mm256_add_pd(fiy0,ty);
700 fiz0 = _mm256_add_pd(fiz0,tz);
702 fjx0 = _mm256_add_pd(fjx0,tx);
703 fjy0 = _mm256_add_pd(fjy0,ty);
704 fjz0 = _mm256_add_pd(fjz0,tz);
706 /**************************
707 * CALCULATE INTERACTIONS *
708 **************************/
710 /* Compute parameters for interactions between i and j atoms */
711 qq10 = _mm256_mul_pd(iq1,jq0);
713 /* COULOMB ELECTROSTATICS */
714 velec = _mm256_mul_pd(qq10,rinv10);
715 felec = _mm256_mul_pd(velec,rinvsq10);
719 /* Calculate temporary vectorial force */
720 tx = _mm256_mul_pd(fscal,dx10);
721 ty = _mm256_mul_pd(fscal,dy10);
722 tz = _mm256_mul_pd(fscal,dz10);
724 /* Update vectorial force */
725 fix1 = _mm256_add_pd(fix1,tx);
726 fiy1 = _mm256_add_pd(fiy1,ty);
727 fiz1 = _mm256_add_pd(fiz1,tz);
729 fjx0 = _mm256_add_pd(fjx0,tx);
730 fjy0 = _mm256_add_pd(fjy0,ty);
731 fjz0 = _mm256_add_pd(fjz0,tz);
733 /**************************
734 * CALCULATE INTERACTIONS *
735 **************************/
737 /* Compute parameters for interactions between i and j atoms */
738 qq20 = _mm256_mul_pd(iq2,jq0);
740 /* COULOMB ELECTROSTATICS */
741 velec = _mm256_mul_pd(qq20,rinv20);
742 felec = _mm256_mul_pd(velec,rinvsq20);
746 /* Calculate temporary vectorial force */
747 tx = _mm256_mul_pd(fscal,dx20);
748 ty = _mm256_mul_pd(fscal,dy20);
749 tz = _mm256_mul_pd(fscal,dz20);
751 /* Update vectorial force */
752 fix2 = _mm256_add_pd(fix2,tx);
753 fiy2 = _mm256_add_pd(fiy2,ty);
754 fiz2 = _mm256_add_pd(fiz2,tz);
756 fjx0 = _mm256_add_pd(fjx0,tx);
757 fjy0 = _mm256_add_pd(fjy0,ty);
758 fjz0 = _mm256_add_pd(fjz0,tz);
760 fjptrA = f+j_coord_offsetA;
761 fjptrB = f+j_coord_offsetB;
762 fjptrC = f+j_coord_offsetC;
763 fjptrD = f+j_coord_offsetD;
765 gmx_mm256_decrement_1rvec_4ptr_swizzle_pd(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
767 /* Inner loop uses 81 flops */
773 /* Get j neighbor index, and coordinate index */
774 jnrlistA = jjnr[jidx];
775 jnrlistB = jjnr[jidx+1];
776 jnrlistC = jjnr[jidx+2];
777 jnrlistD = jjnr[jidx+3];
778 /* Sign of each element will be negative for non-real atoms.
779 * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
780 * so use it as val = _mm_andnot_pd(mask,val) to clear dummy entries.
782 tmpmask0 = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
784 tmpmask1 = _mm_permute_ps(tmpmask0,_GMX_MM_PERMUTE(3,3,2,2));
785 tmpmask0 = _mm_permute_ps(tmpmask0,_GMX_MM_PERMUTE(1,1,0,0));
786 dummy_mask = _mm256_castps_pd(gmx_mm256_set_m128(tmpmask1,tmpmask0));
788 jnrA = (jnrlistA>=0) ? jnrlistA : 0;
789 jnrB = (jnrlistB>=0) ? jnrlistB : 0;
790 jnrC = (jnrlistC>=0) ? jnrlistC : 0;
791 jnrD = (jnrlistD>=0) ? jnrlistD : 0;
792 j_coord_offsetA = DIM*jnrA;
793 j_coord_offsetB = DIM*jnrB;
794 j_coord_offsetC = DIM*jnrC;
795 j_coord_offsetD = DIM*jnrD;
797 /* load j atom coordinates */
798 gmx_mm256_load_1rvec_4ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
799 x+j_coord_offsetC,x+j_coord_offsetD,
802 /* Calculate displacement vector */
803 dx00 = _mm256_sub_pd(ix0,jx0);
804 dy00 = _mm256_sub_pd(iy0,jy0);
805 dz00 = _mm256_sub_pd(iz0,jz0);
806 dx10 = _mm256_sub_pd(ix1,jx0);
807 dy10 = _mm256_sub_pd(iy1,jy0);
808 dz10 = _mm256_sub_pd(iz1,jz0);
809 dx20 = _mm256_sub_pd(ix2,jx0);
810 dy20 = _mm256_sub_pd(iy2,jy0);
811 dz20 = _mm256_sub_pd(iz2,jz0);
813 /* Calculate squared distance and things based on it */
814 rsq00 = gmx_mm256_calc_rsq_pd(dx00,dy00,dz00);
815 rsq10 = gmx_mm256_calc_rsq_pd(dx10,dy10,dz10);
816 rsq20 = gmx_mm256_calc_rsq_pd(dx20,dy20,dz20);
818 rinv00 = avx256_invsqrt_d(rsq00);
819 rinv10 = avx256_invsqrt_d(rsq10);
820 rinv20 = avx256_invsqrt_d(rsq20);
822 rinvsq00 = _mm256_mul_pd(rinv00,rinv00);
823 rinvsq10 = _mm256_mul_pd(rinv10,rinv10);
824 rinvsq20 = _mm256_mul_pd(rinv20,rinv20);
826 /* Load parameters for j particles */
827 jq0 = gmx_mm256_load_4real_swizzle_pd(charge+jnrA+0,charge+jnrB+0,
828 charge+jnrC+0,charge+jnrD+0);
830 fjx0 = _mm256_setzero_pd();
831 fjy0 = _mm256_setzero_pd();
832 fjz0 = _mm256_setzero_pd();
834 /**************************
835 * CALCULATE INTERACTIONS *
836 **************************/
838 /* Compute parameters for interactions between i and j atoms */
839 qq00 = _mm256_mul_pd(iq0,jq0);
841 /* COULOMB ELECTROSTATICS */
842 velec = _mm256_mul_pd(qq00,rinv00);
843 felec = _mm256_mul_pd(velec,rinvsq00);
847 fscal = _mm256_andnot_pd(dummy_mask,fscal);
849 /* Calculate temporary vectorial force */
850 tx = _mm256_mul_pd(fscal,dx00);
851 ty = _mm256_mul_pd(fscal,dy00);
852 tz = _mm256_mul_pd(fscal,dz00);
854 /* Update vectorial force */
855 fix0 = _mm256_add_pd(fix0,tx);
856 fiy0 = _mm256_add_pd(fiy0,ty);
857 fiz0 = _mm256_add_pd(fiz0,tz);
859 fjx0 = _mm256_add_pd(fjx0,tx);
860 fjy0 = _mm256_add_pd(fjy0,ty);
861 fjz0 = _mm256_add_pd(fjz0,tz);
863 /**************************
864 * CALCULATE INTERACTIONS *
865 **************************/
867 /* Compute parameters for interactions between i and j atoms */
868 qq10 = _mm256_mul_pd(iq1,jq0);
870 /* COULOMB ELECTROSTATICS */
871 velec = _mm256_mul_pd(qq10,rinv10);
872 felec = _mm256_mul_pd(velec,rinvsq10);
876 fscal = _mm256_andnot_pd(dummy_mask,fscal);
878 /* Calculate temporary vectorial force */
879 tx = _mm256_mul_pd(fscal,dx10);
880 ty = _mm256_mul_pd(fscal,dy10);
881 tz = _mm256_mul_pd(fscal,dz10);
883 /* Update vectorial force */
884 fix1 = _mm256_add_pd(fix1,tx);
885 fiy1 = _mm256_add_pd(fiy1,ty);
886 fiz1 = _mm256_add_pd(fiz1,tz);
888 fjx0 = _mm256_add_pd(fjx0,tx);
889 fjy0 = _mm256_add_pd(fjy0,ty);
890 fjz0 = _mm256_add_pd(fjz0,tz);
892 /**************************
893 * CALCULATE INTERACTIONS *
894 **************************/
896 /* Compute parameters for interactions between i and j atoms */
897 qq20 = _mm256_mul_pd(iq2,jq0);
899 /* COULOMB ELECTROSTATICS */
900 velec = _mm256_mul_pd(qq20,rinv20);
901 felec = _mm256_mul_pd(velec,rinvsq20);
905 fscal = _mm256_andnot_pd(dummy_mask,fscal);
907 /* Calculate temporary vectorial force */
908 tx = _mm256_mul_pd(fscal,dx20);
909 ty = _mm256_mul_pd(fscal,dy20);
910 tz = _mm256_mul_pd(fscal,dz20);
912 /* Update vectorial force */
913 fix2 = _mm256_add_pd(fix2,tx);
914 fiy2 = _mm256_add_pd(fiy2,ty);
915 fiz2 = _mm256_add_pd(fiz2,tz);
917 fjx0 = _mm256_add_pd(fjx0,tx);
918 fjy0 = _mm256_add_pd(fjy0,ty);
919 fjz0 = _mm256_add_pd(fjz0,tz);
921 fjptrA = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
922 fjptrB = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
923 fjptrC = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
924 fjptrD = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
926 gmx_mm256_decrement_1rvec_4ptr_swizzle_pd(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
928 /* Inner loop uses 81 flops */
931 /* End of innermost loop */
933 gmx_mm256_update_iforce_3atom_swizzle_pd(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,
934 f+i_coord_offset,fshift+i_shift_offset);
936 /* Increment number of inner iterations */
937 inneriter += j_index_end - j_index_start;
939 /* Outer loop uses 18 flops */
942 /* Increment number of outer iterations */
945 /* Update outer/inner flops */
947 inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_W3_F,outeriter*18 + inneriter*81);