2 * Note: this file was generated by the Gromacs avx_256_double kernel generator.
4 * This source code is part of
8 * Copyright (c) 2001-2012, The GROMACS Development Team
10 * Gromacs is a library for molecular simulation and trajectory analysis,
11 * written by Erik Lindahl, David van der Spoel, Berk Hess, and others - for
12 * a full list of developers and information, check out http://www.gromacs.org
14 * This program is free software; you can redistribute it and/or modify it under
15 * the terms of the GNU Lesser General Public License as published by the Free
16 * Software Foundation; either version 2 of the License, or (at your option) any
19 * To help fund GROMACS development, we humbly ask that you cite
20 * the papers people have written on it - you can find them on the website.
28 #include "../nb_kernel.h"
29 #include "types/simple.h"
33 #include "gmx_math_x86_avx_256_double.h"
34 #include "kernelutil_x86_avx_256_double.h"
37 * Gromacs nonbonded kernel: nb_kernel_ElecCoul_VdwNone_GeomW4P1_VF_avx_256_double
38 * Electrostatics interaction: Coulomb
39 * VdW interaction: None
40 * Geometry: Water4-Particle
41 * Calculate force/pot: PotentialAndForce
44 nb_kernel_ElecCoul_VdwNone_GeomW4P1_VF_avx_256_double
45 (t_nblist * gmx_restrict nlist,
46 rvec * gmx_restrict xx,
47 rvec * gmx_restrict ff,
48 t_forcerec * gmx_restrict fr,
49 t_mdatoms * gmx_restrict mdatoms,
50 nb_kernel_data_t * gmx_restrict kernel_data,
51 t_nrnb * gmx_restrict nrnb)
53 /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
54 * just 0 for non-waters.
55 * Suffixes A,B,C,D refer to j loop unrolling done with AVX, e.g. for the four different
56 * jnr indices corresponding to data put in the four positions in the SIMD register.
58 int i_shift_offset,i_coord_offset,outeriter,inneriter;
59 int j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
60 int jnrA,jnrB,jnrC,jnrD;
61 int jnrlistA,jnrlistB,jnrlistC,jnrlistD;
62 int jnrlistE,jnrlistF,jnrlistG,jnrlistH;
63 int j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
64 int *iinr,*jindex,*jjnr,*shiftidx,*gid;
66 real *shiftvec,*fshift,*x,*f;
67 real *fjptrA,*fjptrB,*fjptrC,*fjptrD;
69 __m256d tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
70 real * vdwioffsetptr1;
71 __m256d ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
72 real * vdwioffsetptr2;
73 __m256d ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
74 real * vdwioffsetptr3;
75 __m256d ix3,iy3,iz3,fix3,fiy3,fiz3,iq3,isai3;
76 int vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
77 __m256d jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
78 __m256d dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
79 __m256d dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
80 __m256d dx30,dy30,dz30,rsq30,rinv30,rinvsq30,r30,qq30,c6_30,c12_30;
81 __m256d velec,felec,velecsum,facel,crf,krf,krf2;
83 __m256d dummy_mask,cutoff_mask;
84 __m128 tmpmask0,tmpmask1;
85 __m256d signbit = _mm256_castsi256_pd( _mm256_set1_epi32(0x80000000) );
86 __m256d one = _mm256_set1_pd(1.0);
87 __m256d two = _mm256_set1_pd(2.0);
93 jindex = nlist->jindex;
95 shiftidx = nlist->shift;
97 shiftvec = fr->shift_vec[0];
98 fshift = fr->fshift[0];
99 facel = _mm256_set1_pd(fr->epsfac);
100 charge = mdatoms->chargeA;
102 /* Setup water-specific parameters */
103 inr = nlist->iinr[0];
104 iq1 = _mm256_mul_pd(facel,_mm256_set1_pd(charge[inr+1]));
105 iq2 = _mm256_mul_pd(facel,_mm256_set1_pd(charge[inr+2]));
106 iq3 = _mm256_mul_pd(facel,_mm256_set1_pd(charge[inr+3]));
108 /* Avoid stupid compiler warnings */
109 jnrA = jnrB = jnrC = jnrD = 0;
118 for(iidx=0;iidx<4*DIM;iidx++)
123 /* Start outer loop over neighborlists */
124 for(iidx=0; iidx<nri; iidx++)
126 /* Load shift vector for this list */
127 i_shift_offset = DIM*shiftidx[iidx];
129 /* Load limits for loop over neighbors */
130 j_index_start = jindex[iidx];
131 j_index_end = jindex[iidx+1];
133 /* Get outer coordinate index */
135 i_coord_offset = DIM*inr;
137 /* Load i particle coords and add shift vector */
138 gmx_mm256_load_shift_and_3rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset+DIM,
139 &ix1,&iy1,&iz1,&ix2,&iy2,&iz2,&ix3,&iy3,&iz3);
141 fix1 = _mm256_setzero_pd();
142 fiy1 = _mm256_setzero_pd();
143 fiz1 = _mm256_setzero_pd();
144 fix2 = _mm256_setzero_pd();
145 fiy2 = _mm256_setzero_pd();
146 fiz2 = _mm256_setzero_pd();
147 fix3 = _mm256_setzero_pd();
148 fiy3 = _mm256_setzero_pd();
149 fiz3 = _mm256_setzero_pd();
151 /* Reset potential sums */
152 velecsum = _mm256_setzero_pd();
154 /* Start inner kernel loop */
155 for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
158 /* Get j neighbor index, and coordinate index */
163 j_coord_offsetA = DIM*jnrA;
164 j_coord_offsetB = DIM*jnrB;
165 j_coord_offsetC = DIM*jnrC;
166 j_coord_offsetD = DIM*jnrD;
168 /* load j atom coordinates */
169 gmx_mm256_load_1rvec_4ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
170 x+j_coord_offsetC,x+j_coord_offsetD,
173 /* Calculate displacement vector */
174 dx10 = _mm256_sub_pd(ix1,jx0);
175 dy10 = _mm256_sub_pd(iy1,jy0);
176 dz10 = _mm256_sub_pd(iz1,jz0);
177 dx20 = _mm256_sub_pd(ix2,jx0);
178 dy20 = _mm256_sub_pd(iy2,jy0);
179 dz20 = _mm256_sub_pd(iz2,jz0);
180 dx30 = _mm256_sub_pd(ix3,jx0);
181 dy30 = _mm256_sub_pd(iy3,jy0);
182 dz30 = _mm256_sub_pd(iz3,jz0);
184 /* Calculate squared distance and things based on it */
185 rsq10 = gmx_mm256_calc_rsq_pd(dx10,dy10,dz10);
186 rsq20 = gmx_mm256_calc_rsq_pd(dx20,dy20,dz20);
187 rsq30 = gmx_mm256_calc_rsq_pd(dx30,dy30,dz30);
189 rinv10 = gmx_mm256_invsqrt_pd(rsq10);
190 rinv20 = gmx_mm256_invsqrt_pd(rsq20);
191 rinv30 = gmx_mm256_invsqrt_pd(rsq30);
193 rinvsq10 = _mm256_mul_pd(rinv10,rinv10);
194 rinvsq20 = _mm256_mul_pd(rinv20,rinv20);
195 rinvsq30 = _mm256_mul_pd(rinv30,rinv30);
197 /* Load parameters for j particles */
198 jq0 = gmx_mm256_load_4real_swizzle_pd(charge+jnrA+0,charge+jnrB+0,
199 charge+jnrC+0,charge+jnrD+0);
201 fjx0 = _mm256_setzero_pd();
202 fjy0 = _mm256_setzero_pd();
203 fjz0 = _mm256_setzero_pd();
205 /**************************
206 * CALCULATE INTERACTIONS *
207 **************************/
209 /* Compute parameters for interactions between i and j atoms */
210 qq10 = _mm256_mul_pd(iq1,jq0);
212 /* COULOMB ELECTROSTATICS */
213 velec = _mm256_mul_pd(qq10,rinv10);
214 felec = _mm256_mul_pd(velec,rinvsq10);
216 /* Update potential sum for this i atom from the interaction with this j atom. */
217 velecsum = _mm256_add_pd(velecsum,velec);
221 /* Calculate temporary vectorial force */
222 tx = _mm256_mul_pd(fscal,dx10);
223 ty = _mm256_mul_pd(fscal,dy10);
224 tz = _mm256_mul_pd(fscal,dz10);
226 /* Update vectorial force */
227 fix1 = _mm256_add_pd(fix1,tx);
228 fiy1 = _mm256_add_pd(fiy1,ty);
229 fiz1 = _mm256_add_pd(fiz1,tz);
231 fjx0 = _mm256_add_pd(fjx0,tx);
232 fjy0 = _mm256_add_pd(fjy0,ty);
233 fjz0 = _mm256_add_pd(fjz0,tz);
235 /**************************
236 * CALCULATE INTERACTIONS *
237 **************************/
239 /* Compute parameters for interactions between i and j atoms */
240 qq20 = _mm256_mul_pd(iq2,jq0);
242 /* COULOMB ELECTROSTATICS */
243 velec = _mm256_mul_pd(qq20,rinv20);
244 felec = _mm256_mul_pd(velec,rinvsq20);
246 /* Update potential sum for this i atom from the interaction with this j atom. */
247 velecsum = _mm256_add_pd(velecsum,velec);
251 /* Calculate temporary vectorial force */
252 tx = _mm256_mul_pd(fscal,dx20);
253 ty = _mm256_mul_pd(fscal,dy20);
254 tz = _mm256_mul_pd(fscal,dz20);
256 /* Update vectorial force */
257 fix2 = _mm256_add_pd(fix2,tx);
258 fiy2 = _mm256_add_pd(fiy2,ty);
259 fiz2 = _mm256_add_pd(fiz2,tz);
261 fjx0 = _mm256_add_pd(fjx0,tx);
262 fjy0 = _mm256_add_pd(fjy0,ty);
263 fjz0 = _mm256_add_pd(fjz0,tz);
265 /**************************
266 * CALCULATE INTERACTIONS *
267 **************************/
269 /* Compute parameters for interactions between i and j atoms */
270 qq30 = _mm256_mul_pd(iq3,jq0);
272 /* COULOMB ELECTROSTATICS */
273 velec = _mm256_mul_pd(qq30,rinv30);
274 felec = _mm256_mul_pd(velec,rinvsq30);
276 /* Update potential sum for this i atom from the interaction with this j atom. */
277 velecsum = _mm256_add_pd(velecsum,velec);
281 /* Calculate temporary vectorial force */
282 tx = _mm256_mul_pd(fscal,dx30);
283 ty = _mm256_mul_pd(fscal,dy30);
284 tz = _mm256_mul_pd(fscal,dz30);
286 /* Update vectorial force */
287 fix3 = _mm256_add_pd(fix3,tx);
288 fiy3 = _mm256_add_pd(fiy3,ty);
289 fiz3 = _mm256_add_pd(fiz3,tz);
291 fjx0 = _mm256_add_pd(fjx0,tx);
292 fjy0 = _mm256_add_pd(fjy0,ty);
293 fjz0 = _mm256_add_pd(fjz0,tz);
295 fjptrA = f+j_coord_offsetA;
296 fjptrB = f+j_coord_offsetB;
297 fjptrC = f+j_coord_offsetC;
298 fjptrD = f+j_coord_offsetD;
300 gmx_mm256_decrement_1rvec_4ptr_swizzle_pd(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
302 /* Inner loop uses 84 flops */
308 /* Get j neighbor index, and coordinate index */
309 jnrlistA = jjnr[jidx];
310 jnrlistB = jjnr[jidx+1];
311 jnrlistC = jjnr[jidx+2];
312 jnrlistD = jjnr[jidx+3];
313 /* Sign of each element will be negative for non-real atoms.
314 * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
315 * so use it as val = _mm_andnot_pd(mask,val) to clear dummy entries.
317 tmpmask0 = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
319 tmpmask1 = _mm_permute_ps(tmpmask0,_GMX_MM_PERMUTE(3,3,2,2));
320 tmpmask0 = _mm_permute_ps(tmpmask0,_GMX_MM_PERMUTE(1,1,0,0));
321 dummy_mask = _mm256_castps_pd(gmx_mm256_set_m128(tmpmask1,tmpmask0));
323 jnrA = (jnrlistA>=0) ? jnrlistA : 0;
324 jnrB = (jnrlistB>=0) ? jnrlistB : 0;
325 jnrC = (jnrlistC>=0) ? jnrlistC : 0;
326 jnrD = (jnrlistD>=0) ? jnrlistD : 0;
327 j_coord_offsetA = DIM*jnrA;
328 j_coord_offsetB = DIM*jnrB;
329 j_coord_offsetC = DIM*jnrC;
330 j_coord_offsetD = DIM*jnrD;
332 /* load j atom coordinates */
333 gmx_mm256_load_1rvec_4ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
334 x+j_coord_offsetC,x+j_coord_offsetD,
337 /* Calculate displacement vector */
338 dx10 = _mm256_sub_pd(ix1,jx0);
339 dy10 = _mm256_sub_pd(iy1,jy0);
340 dz10 = _mm256_sub_pd(iz1,jz0);
341 dx20 = _mm256_sub_pd(ix2,jx0);
342 dy20 = _mm256_sub_pd(iy2,jy0);
343 dz20 = _mm256_sub_pd(iz2,jz0);
344 dx30 = _mm256_sub_pd(ix3,jx0);
345 dy30 = _mm256_sub_pd(iy3,jy0);
346 dz30 = _mm256_sub_pd(iz3,jz0);
348 /* Calculate squared distance and things based on it */
349 rsq10 = gmx_mm256_calc_rsq_pd(dx10,dy10,dz10);
350 rsq20 = gmx_mm256_calc_rsq_pd(dx20,dy20,dz20);
351 rsq30 = gmx_mm256_calc_rsq_pd(dx30,dy30,dz30);
353 rinv10 = gmx_mm256_invsqrt_pd(rsq10);
354 rinv20 = gmx_mm256_invsqrt_pd(rsq20);
355 rinv30 = gmx_mm256_invsqrt_pd(rsq30);
357 rinvsq10 = _mm256_mul_pd(rinv10,rinv10);
358 rinvsq20 = _mm256_mul_pd(rinv20,rinv20);
359 rinvsq30 = _mm256_mul_pd(rinv30,rinv30);
361 /* Load parameters for j particles */
362 jq0 = gmx_mm256_load_4real_swizzle_pd(charge+jnrA+0,charge+jnrB+0,
363 charge+jnrC+0,charge+jnrD+0);
365 fjx0 = _mm256_setzero_pd();
366 fjy0 = _mm256_setzero_pd();
367 fjz0 = _mm256_setzero_pd();
369 /**************************
370 * CALCULATE INTERACTIONS *
371 **************************/
373 /* Compute parameters for interactions between i and j atoms */
374 qq10 = _mm256_mul_pd(iq1,jq0);
376 /* COULOMB ELECTROSTATICS */
377 velec = _mm256_mul_pd(qq10,rinv10);
378 felec = _mm256_mul_pd(velec,rinvsq10);
380 /* Update potential sum for this i atom from the interaction with this j atom. */
381 velec = _mm256_andnot_pd(dummy_mask,velec);
382 velecsum = _mm256_add_pd(velecsum,velec);
386 fscal = _mm256_andnot_pd(dummy_mask,fscal);
388 /* Calculate temporary vectorial force */
389 tx = _mm256_mul_pd(fscal,dx10);
390 ty = _mm256_mul_pd(fscal,dy10);
391 tz = _mm256_mul_pd(fscal,dz10);
393 /* Update vectorial force */
394 fix1 = _mm256_add_pd(fix1,tx);
395 fiy1 = _mm256_add_pd(fiy1,ty);
396 fiz1 = _mm256_add_pd(fiz1,tz);
398 fjx0 = _mm256_add_pd(fjx0,tx);
399 fjy0 = _mm256_add_pd(fjy0,ty);
400 fjz0 = _mm256_add_pd(fjz0,tz);
402 /**************************
403 * CALCULATE INTERACTIONS *
404 **************************/
406 /* Compute parameters for interactions between i and j atoms */
407 qq20 = _mm256_mul_pd(iq2,jq0);
409 /* COULOMB ELECTROSTATICS */
410 velec = _mm256_mul_pd(qq20,rinv20);
411 felec = _mm256_mul_pd(velec,rinvsq20);
413 /* Update potential sum for this i atom from the interaction with this j atom. */
414 velec = _mm256_andnot_pd(dummy_mask,velec);
415 velecsum = _mm256_add_pd(velecsum,velec);
419 fscal = _mm256_andnot_pd(dummy_mask,fscal);
421 /* Calculate temporary vectorial force */
422 tx = _mm256_mul_pd(fscal,dx20);
423 ty = _mm256_mul_pd(fscal,dy20);
424 tz = _mm256_mul_pd(fscal,dz20);
426 /* Update vectorial force */
427 fix2 = _mm256_add_pd(fix2,tx);
428 fiy2 = _mm256_add_pd(fiy2,ty);
429 fiz2 = _mm256_add_pd(fiz2,tz);
431 fjx0 = _mm256_add_pd(fjx0,tx);
432 fjy0 = _mm256_add_pd(fjy0,ty);
433 fjz0 = _mm256_add_pd(fjz0,tz);
435 /**************************
436 * CALCULATE INTERACTIONS *
437 **************************/
439 /* Compute parameters for interactions between i and j atoms */
440 qq30 = _mm256_mul_pd(iq3,jq0);
442 /* COULOMB ELECTROSTATICS */
443 velec = _mm256_mul_pd(qq30,rinv30);
444 felec = _mm256_mul_pd(velec,rinvsq30);
446 /* Update potential sum for this i atom from the interaction with this j atom. */
447 velec = _mm256_andnot_pd(dummy_mask,velec);
448 velecsum = _mm256_add_pd(velecsum,velec);
452 fscal = _mm256_andnot_pd(dummy_mask,fscal);
454 /* Calculate temporary vectorial force */
455 tx = _mm256_mul_pd(fscal,dx30);
456 ty = _mm256_mul_pd(fscal,dy30);
457 tz = _mm256_mul_pd(fscal,dz30);
459 /* Update vectorial force */
460 fix3 = _mm256_add_pd(fix3,tx);
461 fiy3 = _mm256_add_pd(fiy3,ty);
462 fiz3 = _mm256_add_pd(fiz3,tz);
464 fjx0 = _mm256_add_pd(fjx0,tx);
465 fjy0 = _mm256_add_pd(fjy0,ty);
466 fjz0 = _mm256_add_pd(fjz0,tz);
468 fjptrA = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
469 fjptrB = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
470 fjptrC = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
471 fjptrD = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
473 gmx_mm256_decrement_1rvec_4ptr_swizzle_pd(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
475 /* Inner loop uses 84 flops */
478 /* End of innermost loop */
480 gmx_mm256_update_iforce_3atom_swizzle_pd(fix1,fiy1,fiz1,fix2,fiy2,fiz2,fix3,fiy3,fiz3,
481 f+i_coord_offset+DIM,fshift+i_shift_offset);
484 /* Update potential energies */
485 gmx_mm256_update_1pot_pd(velecsum,kernel_data->energygrp_elec+ggid);
487 /* Increment number of inner iterations */
488 inneriter += j_index_end - j_index_start;
490 /* Outer loop uses 19 flops */
493 /* Increment number of outer iterations */
496 /* Update outer/inner flops */
498 inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_W4_VF,outeriter*19 + inneriter*84);
501 * Gromacs nonbonded kernel: nb_kernel_ElecCoul_VdwNone_GeomW4P1_F_avx_256_double
502 * Electrostatics interaction: Coulomb
503 * VdW interaction: None
504 * Geometry: Water4-Particle
505 * Calculate force/pot: Force
508 nb_kernel_ElecCoul_VdwNone_GeomW4P1_F_avx_256_double
509 (t_nblist * gmx_restrict nlist,
510 rvec * gmx_restrict xx,
511 rvec * gmx_restrict ff,
512 t_forcerec * gmx_restrict fr,
513 t_mdatoms * gmx_restrict mdatoms,
514 nb_kernel_data_t * gmx_restrict kernel_data,
515 t_nrnb * gmx_restrict nrnb)
517 /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
518 * just 0 for non-waters.
519 * Suffixes A,B,C,D refer to j loop unrolling done with AVX, e.g. for the four different
520 * jnr indices corresponding to data put in the four positions in the SIMD register.
522 int i_shift_offset,i_coord_offset,outeriter,inneriter;
523 int j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
524 int jnrA,jnrB,jnrC,jnrD;
525 int jnrlistA,jnrlistB,jnrlistC,jnrlistD;
526 int jnrlistE,jnrlistF,jnrlistG,jnrlistH;
527 int j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
528 int *iinr,*jindex,*jjnr,*shiftidx,*gid;
530 real *shiftvec,*fshift,*x,*f;
531 real *fjptrA,*fjptrB,*fjptrC,*fjptrD;
533 __m256d tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
534 real * vdwioffsetptr1;
535 __m256d ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
536 real * vdwioffsetptr2;
537 __m256d ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
538 real * vdwioffsetptr3;
539 __m256d ix3,iy3,iz3,fix3,fiy3,fiz3,iq3,isai3;
540 int vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
541 __m256d jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
542 __m256d dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
543 __m256d dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
544 __m256d dx30,dy30,dz30,rsq30,rinv30,rinvsq30,r30,qq30,c6_30,c12_30;
545 __m256d velec,felec,velecsum,facel,crf,krf,krf2;
547 __m256d dummy_mask,cutoff_mask;
548 __m128 tmpmask0,tmpmask1;
549 __m256d signbit = _mm256_castsi256_pd( _mm256_set1_epi32(0x80000000) );
550 __m256d one = _mm256_set1_pd(1.0);
551 __m256d two = _mm256_set1_pd(2.0);
557 jindex = nlist->jindex;
559 shiftidx = nlist->shift;
561 shiftvec = fr->shift_vec[0];
562 fshift = fr->fshift[0];
563 facel = _mm256_set1_pd(fr->epsfac);
564 charge = mdatoms->chargeA;
566 /* Setup water-specific parameters */
567 inr = nlist->iinr[0];
568 iq1 = _mm256_mul_pd(facel,_mm256_set1_pd(charge[inr+1]));
569 iq2 = _mm256_mul_pd(facel,_mm256_set1_pd(charge[inr+2]));
570 iq3 = _mm256_mul_pd(facel,_mm256_set1_pd(charge[inr+3]));
572 /* Avoid stupid compiler warnings */
573 jnrA = jnrB = jnrC = jnrD = 0;
582 for(iidx=0;iidx<4*DIM;iidx++)
587 /* Start outer loop over neighborlists */
588 for(iidx=0; iidx<nri; iidx++)
590 /* Load shift vector for this list */
591 i_shift_offset = DIM*shiftidx[iidx];
593 /* Load limits for loop over neighbors */
594 j_index_start = jindex[iidx];
595 j_index_end = jindex[iidx+1];
597 /* Get outer coordinate index */
599 i_coord_offset = DIM*inr;
601 /* Load i particle coords and add shift vector */
602 gmx_mm256_load_shift_and_3rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset+DIM,
603 &ix1,&iy1,&iz1,&ix2,&iy2,&iz2,&ix3,&iy3,&iz3);
605 fix1 = _mm256_setzero_pd();
606 fiy1 = _mm256_setzero_pd();
607 fiz1 = _mm256_setzero_pd();
608 fix2 = _mm256_setzero_pd();
609 fiy2 = _mm256_setzero_pd();
610 fiz2 = _mm256_setzero_pd();
611 fix3 = _mm256_setzero_pd();
612 fiy3 = _mm256_setzero_pd();
613 fiz3 = _mm256_setzero_pd();
615 /* Start inner kernel loop */
616 for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
619 /* Get j neighbor index, and coordinate index */
624 j_coord_offsetA = DIM*jnrA;
625 j_coord_offsetB = DIM*jnrB;
626 j_coord_offsetC = DIM*jnrC;
627 j_coord_offsetD = DIM*jnrD;
629 /* load j atom coordinates */
630 gmx_mm256_load_1rvec_4ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
631 x+j_coord_offsetC,x+j_coord_offsetD,
634 /* Calculate displacement vector */
635 dx10 = _mm256_sub_pd(ix1,jx0);
636 dy10 = _mm256_sub_pd(iy1,jy0);
637 dz10 = _mm256_sub_pd(iz1,jz0);
638 dx20 = _mm256_sub_pd(ix2,jx0);
639 dy20 = _mm256_sub_pd(iy2,jy0);
640 dz20 = _mm256_sub_pd(iz2,jz0);
641 dx30 = _mm256_sub_pd(ix3,jx0);
642 dy30 = _mm256_sub_pd(iy3,jy0);
643 dz30 = _mm256_sub_pd(iz3,jz0);
645 /* Calculate squared distance and things based on it */
646 rsq10 = gmx_mm256_calc_rsq_pd(dx10,dy10,dz10);
647 rsq20 = gmx_mm256_calc_rsq_pd(dx20,dy20,dz20);
648 rsq30 = gmx_mm256_calc_rsq_pd(dx30,dy30,dz30);
650 rinv10 = gmx_mm256_invsqrt_pd(rsq10);
651 rinv20 = gmx_mm256_invsqrt_pd(rsq20);
652 rinv30 = gmx_mm256_invsqrt_pd(rsq30);
654 rinvsq10 = _mm256_mul_pd(rinv10,rinv10);
655 rinvsq20 = _mm256_mul_pd(rinv20,rinv20);
656 rinvsq30 = _mm256_mul_pd(rinv30,rinv30);
658 /* Load parameters for j particles */
659 jq0 = gmx_mm256_load_4real_swizzle_pd(charge+jnrA+0,charge+jnrB+0,
660 charge+jnrC+0,charge+jnrD+0);
662 fjx0 = _mm256_setzero_pd();
663 fjy0 = _mm256_setzero_pd();
664 fjz0 = _mm256_setzero_pd();
666 /**************************
667 * CALCULATE INTERACTIONS *
668 **************************/
670 /* Compute parameters for interactions between i and j atoms */
671 qq10 = _mm256_mul_pd(iq1,jq0);
673 /* COULOMB ELECTROSTATICS */
674 velec = _mm256_mul_pd(qq10,rinv10);
675 felec = _mm256_mul_pd(velec,rinvsq10);
679 /* Calculate temporary vectorial force */
680 tx = _mm256_mul_pd(fscal,dx10);
681 ty = _mm256_mul_pd(fscal,dy10);
682 tz = _mm256_mul_pd(fscal,dz10);
684 /* Update vectorial force */
685 fix1 = _mm256_add_pd(fix1,tx);
686 fiy1 = _mm256_add_pd(fiy1,ty);
687 fiz1 = _mm256_add_pd(fiz1,tz);
689 fjx0 = _mm256_add_pd(fjx0,tx);
690 fjy0 = _mm256_add_pd(fjy0,ty);
691 fjz0 = _mm256_add_pd(fjz0,tz);
693 /**************************
694 * CALCULATE INTERACTIONS *
695 **************************/
697 /* Compute parameters for interactions between i and j atoms */
698 qq20 = _mm256_mul_pd(iq2,jq0);
700 /* COULOMB ELECTROSTATICS */
701 velec = _mm256_mul_pd(qq20,rinv20);
702 felec = _mm256_mul_pd(velec,rinvsq20);
706 /* Calculate temporary vectorial force */
707 tx = _mm256_mul_pd(fscal,dx20);
708 ty = _mm256_mul_pd(fscal,dy20);
709 tz = _mm256_mul_pd(fscal,dz20);
711 /* Update vectorial force */
712 fix2 = _mm256_add_pd(fix2,tx);
713 fiy2 = _mm256_add_pd(fiy2,ty);
714 fiz2 = _mm256_add_pd(fiz2,tz);
716 fjx0 = _mm256_add_pd(fjx0,tx);
717 fjy0 = _mm256_add_pd(fjy0,ty);
718 fjz0 = _mm256_add_pd(fjz0,tz);
720 /**************************
721 * CALCULATE INTERACTIONS *
722 **************************/
724 /* Compute parameters for interactions between i and j atoms */
725 qq30 = _mm256_mul_pd(iq3,jq0);
727 /* COULOMB ELECTROSTATICS */
728 velec = _mm256_mul_pd(qq30,rinv30);
729 felec = _mm256_mul_pd(velec,rinvsq30);
733 /* Calculate temporary vectorial force */
734 tx = _mm256_mul_pd(fscal,dx30);
735 ty = _mm256_mul_pd(fscal,dy30);
736 tz = _mm256_mul_pd(fscal,dz30);
738 /* Update vectorial force */
739 fix3 = _mm256_add_pd(fix3,tx);
740 fiy3 = _mm256_add_pd(fiy3,ty);
741 fiz3 = _mm256_add_pd(fiz3,tz);
743 fjx0 = _mm256_add_pd(fjx0,tx);
744 fjy0 = _mm256_add_pd(fjy0,ty);
745 fjz0 = _mm256_add_pd(fjz0,tz);
747 fjptrA = f+j_coord_offsetA;
748 fjptrB = f+j_coord_offsetB;
749 fjptrC = f+j_coord_offsetC;
750 fjptrD = f+j_coord_offsetD;
752 gmx_mm256_decrement_1rvec_4ptr_swizzle_pd(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
754 /* Inner loop uses 81 flops */
760 /* Get j neighbor index, and coordinate index */
761 jnrlistA = jjnr[jidx];
762 jnrlistB = jjnr[jidx+1];
763 jnrlistC = jjnr[jidx+2];
764 jnrlistD = jjnr[jidx+3];
765 /* Sign of each element will be negative for non-real atoms.
766 * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
767 * so use it as val = _mm_andnot_pd(mask,val) to clear dummy entries.
769 tmpmask0 = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
771 tmpmask1 = _mm_permute_ps(tmpmask0,_GMX_MM_PERMUTE(3,3,2,2));
772 tmpmask0 = _mm_permute_ps(tmpmask0,_GMX_MM_PERMUTE(1,1,0,0));
773 dummy_mask = _mm256_castps_pd(gmx_mm256_set_m128(tmpmask1,tmpmask0));
775 jnrA = (jnrlistA>=0) ? jnrlistA : 0;
776 jnrB = (jnrlistB>=0) ? jnrlistB : 0;
777 jnrC = (jnrlistC>=0) ? jnrlistC : 0;
778 jnrD = (jnrlistD>=0) ? jnrlistD : 0;
779 j_coord_offsetA = DIM*jnrA;
780 j_coord_offsetB = DIM*jnrB;
781 j_coord_offsetC = DIM*jnrC;
782 j_coord_offsetD = DIM*jnrD;
784 /* load j atom coordinates */
785 gmx_mm256_load_1rvec_4ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
786 x+j_coord_offsetC,x+j_coord_offsetD,
789 /* Calculate displacement vector */
790 dx10 = _mm256_sub_pd(ix1,jx0);
791 dy10 = _mm256_sub_pd(iy1,jy0);
792 dz10 = _mm256_sub_pd(iz1,jz0);
793 dx20 = _mm256_sub_pd(ix2,jx0);
794 dy20 = _mm256_sub_pd(iy2,jy0);
795 dz20 = _mm256_sub_pd(iz2,jz0);
796 dx30 = _mm256_sub_pd(ix3,jx0);
797 dy30 = _mm256_sub_pd(iy3,jy0);
798 dz30 = _mm256_sub_pd(iz3,jz0);
800 /* Calculate squared distance and things based on it */
801 rsq10 = gmx_mm256_calc_rsq_pd(dx10,dy10,dz10);
802 rsq20 = gmx_mm256_calc_rsq_pd(dx20,dy20,dz20);
803 rsq30 = gmx_mm256_calc_rsq_pd(dx30,dy30,dz30);
805 rinv10 = gmx_mm256_invsqrt_pd(rsq10);
806 rinv20 = gmx_mm256_invsqrt_pd(rsq20);
807 rinv30 = gmx_mm256_invsqrt_pd(rsq30);
809 rinvsq10 = _mm256_mul_pd(rinv10,rinv10);
810 rinvsq20 = _mm256_mul_pd(rinv20,rinv20);
811 rinvsq30 = _mm256_mul_pd(rinv30,rinv30);
813 /* Load parameters for j particles */
814 jq0 = gmx_mm256_load_4real_swizzle_pd(charge+jnrA+0,charge+jnrB+0,
815 charge+jnrC+0,charge+jnrD+0);
817 fjx0 = _mm256_setzero_pd();
818 fjy0 = _mm256_setzero_pd();
819 fjz0 = _mm256_setzero_pd();
821 /**************************
822 * CALCULATE INTERACTIONS *
823 **************************/
825 /* Compute parameters for interactions between i and j atoms */
826 qq10 = _mm256_mul_pd(iq1,jq0);
828 /* COULOMB ELECTROSTATICS */
829 velec = _mm256_mul_pd(qq10,rinv10);
830 felec = _mm256_mul_pd(velec,rinvsq10);
834 fscal = _mm256_andnot_pd(dummy_mask,fscal);
836 /* Calculate temporary vectorial force */
837 tx = _mm256_mul_pd(fscal,dx10);
838 ty = _mm256_mul_pd(fscal,dy10);
839 tz = _mm256_mul_pd(fscal,dz10);
841 /* Update vectorial force */
842 fix1 = _mm256_add_pd(fix1,tx);
843 fiy1 = _mm256_add_pd(fiy1,ty);
844 fiz1 = _mm256_add_pd(fiz1,tz);
846 fjx0 = _mm256_add_pd(fjx0,tx);
847 fjy0 = _mm256_add_pd(fjy0,ty);
848 fjz0 = _mm256_add_pd(fjz0,tz);
850 /**************************
851 * CALCULATE INTERACTIONS *
852 **************************/
854 /* Compute parameters for interactions between i and j atoms */
855 qq20 = _mm256_mul_pd(iq2,jq0);
857 /* COULOMB ELECTROSTATICS */
858 velec = _mm256_mul_pd(qq20,rinv20);
859 felec = _mm256_mul_pd(velec,rinvsq20);
863 fscal = _mm256_andnot_pd(dummy_mask,fscal);
865 /* Calculate temporary vectorial force */
866 tx = _mm256_mul_pd(fscal,dx20);
867 ty = _mm256_mul_pd(fscal,dy20);
868 tz = _mm256_mul_pd(fscal,dz20);
870 /* Update vectorial force */
871 fix2 = _mm256_add_pd(fix2,tx);
872 fiy2 = _mm256_add_pd(fiy2,ty);
873 fiz2 = _mm256_add_pd(fiz2,tz);
875 fjx0 = _mm256_add_pd(fjx0,tx);
876 fjy0 = _mm256_add_pd(fjy0,ty);
877 fjz0 = _mm256_add_pd(fjz0,tz);
879 /**************************
880 * CALCULATE INTERACTIONS *
881 **************************/
883 /* Compute parameters for interactions between i and j atoms */
884 qq30 = _mm256_mul_pd(iq3,jq0);
886 /* COULOMB ELECTROSTATICS */
887 velec = _mm256_mul_pd(qq30,rinv30);
888 felec = _mm256_mul_pd(velec,rinvsq30);
892 fscal = _mm256_andnot_pd(dummy_mask,fscal);
894 /* Calculate temporary vectorial force */
895 tx = _mm256_mul_pd(fscal,dx30);
896 ty = _mm256_mul_pd(fscal,dy30);
897 tz = _mm256_mul_pd(fscal,dz30);
899 /* Update vectorial force */
900 fix3 = _mm256_add_pd(fix3,tx);
901 fiy3 = _mm256_add_pd(fiy3,ty);
902 fiz3 = _mm256_add_pd(fiz3,tz);
904 fjx0 = _mm256_add_pd(fjx0,tx);
905 fjy0 = _mm256_add_pd(fjy0,ty);
906 fjz0 = _mm256_add_pd(fjz0,tz);
908 fjptrA = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
909 fjptrB = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
910 fjptrC = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
911 fjptrD = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
913 gmx_mm256_decrement_1rvec_4ptr_swizzle_pd(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
915 /* Inner loop uses 81 flops */
918 /* End of innermost loop */
920 gmx_mm256_update_iforce_3atom_swizzle_pd(fix1,fiy1,fiz1,fix2,fiy2,fiz2,fix3,fiy3,fiz3,
921 f+i_coord_offset+DIM,fshift+i_shift_offset);
923 /* Increment number of inner iterations */
924 inneriter += j_index_end - j_index_start;
926 /* Outer loop uses 18 flops */
929 /* Increment number of outer iterations */
932 /* Update outer/inner flops */
934 inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_W4_F,outeriter*18 + inneriter*81);