2 * Note: this file was generated by the Gromacs avx_256_single 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_single.h"
34 #include "kernelutil_x86_avx_256_single.h"
37 * Gromacs nonbonded kernel: nb_kernel_ElecEw_VdwNone_GeomW4P1_VF_avx_256_single
38 * Electrostatics interaction: Ewald
39 * VdW interaction: None
40 * Geometry: Water4-Particle
41 * Calculate force/pot: PotentialAndForce
44 nb_kernel_ElecEw_VdwNone_GeomW4P1_VF_avx_256_single
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,E,F,G,H refer to j loop unrolling done with AVX, e.g. for the eight 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 jnrE,jnrF,jnrG,jnrH;
62 int jnrlistA,jnrlistB,jnrlistC,jnrlistD;
63 int jnrlistE,jnrlistF,jnrlistG,jnrlistH;
64 int j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
65 int j_coord_offsetE,j_coord_offsetF,j_coord_offsetG,j_coord_offsetH;
66 int *iinr,*jindex,*jjnr,*shiftidx,*gid;
68 real *shiftvec,*fshift,*x,*f;
69 real *fjptrA,*fjptrB,*fjptrC,*fjptrD,*fjptrE,*fjptrF,*fjptrG,*fjptrH;
71 __m256 tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
72 real * vdwioffsetptr1;
73 __m256 ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
74 real * vdwioffsetptr2;
75 __m256 ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
76 real * vdwioffsetptr3;
77 __m256 ix3,iy3,iz3,fix3,fiy3,fiz3,iq3,isai3;
78 int vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D,vdwjidx0E,vdwjidx0F,vdwjidx0G,vdwjidx0H;
79 __m256 jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
80 __m256 dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
81 __m256 dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
82 __m256 dx30,dy30,dz30,rsq30,rinv30,rinvsq30,r30,qq30,c6_30,c12_30;
83 __m256 velec,felec,velecsum,facel,crf,krf,krf2;
86 __m128i ewitab_lo,ewitab_hi;
87 __m256 ewtabscale,eweps,sh_ewald,ewrt,ewtabhalfspace,ewtabF,ewtabFn,ewtabD,ewtabV;
88 __m256 beta,beta2,beta3,zeta2,pmecorrF,pmecorrV,rinv3;
90 __m256 dummy_mask,cutoff_mask;
91 __m256 signbit = _mm256_castsi256_ps( _mm256_set1_epi32(0x80000000) );
92 __m256 one = _mm256_set1_ps(1.0);
93 __m256 two = _mm256_set1_ps(2.0);
99 jindex = nlist->jindex;
101 shiftidx = nlist->shift;
103 shiftvec = fr->shift_vec[0];
104 fshift = fr->fshift[0];
105 facel = _mm256_set1_ps(fr->epsfac);
106 charge = mdatoms->chargeA;
108 sh_ewald = _mm256_set1_ps(fr->ic->sh_ewald);
109 beta = _mm256_set1_ps(fr->ic->ewaldcoeff);
110 beta2 = _mm256_mul_ps(beta,beta);
111 beta3 = _mm256_mul_ps(beta,beta2);
113 ewtab = fr->ic->tabq_coul_FDV0;
114 ewtabscale = _mm256_set1_ps(fr->ic->tabq_scale);
115 ewtabhalfspace = _mm256_set1_ps(0.5/fr->ic->tabq_scale);
117 /* Setup water-specific parameters */
118 inr = nlist->iinr[0];
119 iq1 = _mm256_mul_ps(facel,_mm256_set1_ps(charge[inr+1]));
120 iq2 = _mm256_mul_ps(facel,_mm256_set1_ps(charge[inr+2]));
121 iq3 = _mm256_mul_ps(facel,_mm256_set1_ps(charge[inr+3]));
123 /* Avoid stupid compiler warnings */
124 jnrA = jnrB = jnrC = jnrD = jnrE = jnrF = jnrG = jnrH = 0;
137 for(iidx=0;iidx<4*DIM;iidx++)
142 /* Start outer loop over neighborlists */
143 for(iidx=0; iidx<nri; iidx++)
145 /* Load shift vector for this list */
146 i_shift_offset = DIM*shiftidx[iidx];
148 /* Load limits for loop over neighbors */
149 j_index_start = jindex[iidx];
150 j_index_end = jindex[iidx+1];
152 /* Get outer coordinate index */
154 i_coord_offset = DIM*inr;
156 /* Load i particle coords and add shift vector */
157 gmx_mm256_load_shift_and_3rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset+DIM,
158 &ix1,&iy1,&iz1,&ix2,&iy2,&iz2,&ix3,&iy3,&iz3);
160 fix1 = _mm256_setzero_ps();
161 fiy1 = _mm256_setzero_ps();
162 fiz1 = _mm256_setzero_ps();
163 fix2 = _mm256_setzero_ps();
164 fiy2 = _mm256_setzero_ps();
165 fiz2 = _mm256_setzero_ps();
166 fix3 = _mm256_setzero_ps();
167 fiy3 = _mm256_setzero_ps();
168 fiz3 = _mm256_setzero_ps();
170 /* Reset potential sums */
171 velecsum = _mm256_setzero_ps();
173 /* Start inner kernel loop */
174 for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+7]>=0; jidx+=8)
177 /* Get j neighbor index, and coordinate index */
186 j_coord_offsetA = DIM*jnrA;
187 j_coord_offsetB = DIM*jnrB;
188 j_coord_offsetC = DIM*jnrC;
189 j_coord_offsetD = DIM*jnrD;
190 j_coord_offsetE = DIM*jnrE;
191 j_coord_offsetF = DIM*jnrF;
192 j_coord_offsetG = DIM*jnrG;
193 j_coord_offsetH = DIM*jnrH;
195 /* load j atom coordinates */
196 gmx_mm256_load_1rvec_8ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
197 x+j_coord_offsetC,x+j_coord_offsetD,
198 x+j_coord_offsetE,x+j_coord_offsetF,
199 x+j_coord_offsetG,x+j_coord_offsetH,
202 /* Calculate displacement vector */
203 dx10 = _mm256_sub_ps(ix1,jx0);
204 dy10 = _mm256_sub_ps(iy1,jy0);
205 dz10 = _mm256_sub_ps(iz1,jz0);
206 dx20 = _mm256_sub_ps(ix2,jx0);
207 dy20 = _mm256_sub_ps(iy2,jy0);
208 dz20 = _mm256_sub_ps(iz2,jz0);
209 dx30 = _mm256_sub_ps(ix3,jx0);
210 dy30 = _mm256_sub_ps(iy3,jy0);
211 dz30 = _mm256_sub_ps(iz3,jz0);
213 /* Calculate squared distance and things based on it */
214 rsq10 = gmx_mm256_calc_rsq_ps(dx10,dy10,dz10);
215 rsq20 = gmx_mm256_calc_rsq_ps(dx20,dy20,dz20);
216 rsq30 = gmx_mm256_calc_rsq_ps(dx30,dy30,dz30);
218 rinv10 = gmx_mm256_invsqrt_ps(rsq10);
219 rinv20 = gmx_mm256_invsqrt_ps(rsq20);
220 rinv30 = gmx_mm256_invsqrt_ps(rsq30);
222 rinvsq10 = _mm256_mul_ps(rinv10,rinv10);
223 rinvsq20 = _mm256_mul_ps(rinv20,rinv20);
224 rinvsq30 = _mm256_mul_ps(rinv30,rinv30);
226 /* Load parameters for j particles */
227 jq0 = gmx_mm256_load_8real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
228 charge+jnrC+0,charge+jnrD+0,
229 charge+jnrE+0,charge+jnrF+0,
230 charge+jnrG+0,charge+jnrH+0);
232 fjx0 = _mm256_setzero_ps();
233 fjy0 = _mm256_setzero_ps();
234 fjz0 = _mm256_setzero_ps();
236 /**************************
237 * CALCULATE INTERACTIONS *
238 **************************/
240 r10 = _mm256_mul_ps(rsq10,rinv10);
242 /* Compute parameters for interactions between i and j atoms */
243 qq10 = _mm256_mul_ps(iq1,jq0);
245 /* EWALD ELECTROSTATICS */
247 /* Analytical PME correction */
248 zeta2 = _mm256_mul_ps(beta2,rsq10);
249 rinv3 = _mm256_mul_ps(rinvsq10,rinv10);
250 pmecorrF = gmx_mm256_pmecorrF_ps(zeta2);
251 felec = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
252 felec = _mm256_mul_ps(qq10,felec);
253 pmecorrV = gmx_mm256_pmecorrV_ps(zeta2);
254 pmecorrV = _mm256_mul_ps(pmecorrV,beta);
255 velec = _mm256_sub_ps(rinv10,pmecorrV);
256 velec = _mm256_mul_ps(qq10,velec);
258 /* Update potential sum for this i atom from the interaction with this j atom. */
259 velecsum = _mm256_add_ps(velecsum,velec);
263 /* Calculate temporary vectorial force */
264 tx = _mm256_mul_ps(fscal,dx10);
265 ty = _mm256_mul_ps(fscal,dy10);
266 tz = _mm256_mul_ps(fscal,dz10);
268 /* Update vectorial force */
269 fix1 = _mm256_add_ps(fix1,tx);
270 fiy1 = _mm256_add_ps(fiy1,ty);
271 fiz1 = _mm256_add_ps(fiz1,tz);
273 fjx0 = _mm256_add_ps(fjx0,tx);
274 fjy0 = _mm256_add_ps(fjy0,ty);
275 fjz0 = _mm256_add_ps(fjz0,tz);
277 /**************************
278 * CALCULATE INTERACTIONS *
279 **************************/
281 r20 = _mm256_mul_ps(rsq20,rinv20);
283 /* Compute parameters for interactions between i and j atoms */
284 qq20 = _mm256_mul_ps(iq2,jq0);
286 /* EWALD ELECTROSTATICS */
288 /* Analytical PME correction */
289 zeta2 = _mm256_mul_ps(beta2,rsq20);
290 rinv3 = _mm256_mul_ps(rinvsq20,rinv20);
291 pmecorrF = gmx_mm256_pmecorrF_ps(zeta2);
292 felec = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
293 felec = _mm256_mul_ps(qq20,felec);
294 pmecorrV = gmx_mm256_pmecorrV_ps(zeta2);
295 pmecorrV = _mm256_mul_ps(pmecorrV,beta);
296 velec = _mm256_sub_ps(rinv20,pmecorrV);
297 velec = _mm256_mul_ps(qq20,velec);
299 /* Update potential sum for this i atom from the interaction with this j atom. */
300 velecsum = _mm256_add_ps(velecsum,velec);
304 /* Calculate temporary vectorial force */
305 tx = _mm256_mul_ps(fscal,dx20);
306 ty = _mm256_mul_ps(fscal,dy20);
307 tz = _mm256_mul_ps(fscal,dz20);
309 /* Update vectorial force */
310 fix2 = _mm256_add_ps(fix2,tx);
311 fiy2 = _mm256_add_ps(fiy2,ty);
312 fiz2 = _mm256_add_ps(fiz2,tz);
314 fjx0 = _mm256_add_ps(fjx0,tx);
315 fjy0 = _mm256_add_ps(fjy0,ty);
316 fjz0 = _mm256_add_ps(fjz0,tz);
318 /**************************
319 * CALCULATE INTERACTIONS *
320 **************************/
322 r30 = _mm256_mul_ps(rsq30,rinv30);
324 /* Compute parameters for interactions between i and j atoms */
325 qq30 = _mm256_mul_ps(iq3,jq0);
327 /* EWALD ELECTROSTATICS */
329 /* Analytical PME correction */
330 zeta2 = _mm256_mul_ps(beta2,rsq30);
331 rinv3 = _mm256_mul_ps(rinvsq30,rinv30);
332 pmecorrF = gmx_mm256_pmecorrF_ps(zeta2);
333 felec = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
334 felec = _mm256_mul_ps(qq30,felec);
335 pmecorrV = gmx_mm256_pmecorrV_ps(zeta2);
336 pmecorrV = _mm256_mul_ps(pmecorrV,beta);
337 velec = _mm256_sub_ps(rinv30,pmecorrV);
338 velec = _mm256_mul_ps(qq30,velec);
340 /* Update potential sum for this i atom from the interaction with this j atom. */
341 velecsum = _mm256_add_ps(velecsum,velec);
345 /* Calculate temporary vectorial force */
346 tx = _mm256_mul_ps(fscal,dx30);
347 ty = _mm256_mul_ps(fscal,dy30);
348 tz = _mm256_mul_ps(fscal,dz30);
350 /* Update vectorial force */
351 fix3 = _mm256_add_ps(fix3,tx);
352 fiy3 = _mm256_add_ps(fiy3,ty);
353 fiz3 = _mm256_add_ps(fiz3,tz);
355 fjx0 = _mm256_add_ps(fjx0,tx);
356 fjy0 = _mm256_add_ps(fjy0,ty);
357 fjz0 = _mm256_add_ps(fjz0,tz);
359 fjptrA = f+j_coord_offsetA;
360 fjptrB = f+j_coord_offsetB;
361 fjptrC = f+j_coord_offsetC;
362 fjptrD = f+j_coord_offsetD;
363 fjptrE = f+j_coord_offsetE;
364 fjptrF = f+j_coord_offsetF;
365 fjptrG = f+j_coord_offsetG;
366 fjptrH = f+j_coord_offsetH;
368 gmx_mm256_decrement_1rvec_8ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjptrE,fjptrF,fjptrG,fjptrH,fjx0,fjy0,fjz0);
370 /* Inner loop uses 255 flops */
376 /* Get j neighbor index, and coordinate index */
377 jnrlistA = jjnr[jidx];
378 jnrlistB = jjnr[jidx+1];
379 jnrlistC = jjnr[jidx+2];
380 jnrlistD = jjnr[jidx+3];
381 jnrlistE = jjnr[jidx+4];
382 jnrlistF = jjnr[jidx+5];
383 jnrlistG = jjnr[jidx+6];
384 jnrlistH = jjnr[jidx+7];
385 /* Sign of each element will be negative for non-real atoms.
386 * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
387 * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
389 dummy_mask = gmx_mm256_set_m128(gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx+4)),_mm_setzero_si128())),
390 gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128())));
392 jnrA = (jnrlistA>=0) ? jnrlistA : 0;
393 jnrB = (jnrlistB>=0) ? jnrlistB : 0;
394 jnrC = (jnrlistC>=0) ? jnrlistC : 0;
395 jnrD = (jnrlistD>=0) ? jnrlistD : 0;
396 jnrE = (jnrlistE>=0) ? jnrlistE : 0;
397 jnrF = (jnrlistF>=0) ? jnrlistF : 0;
398 jnrG = (jnrlistG>=0) ? jnrlistG : 0;
399 jnrH = (jnrlistH>=0) ? jnrlistH : 0;
400 j_coord_offsetA = DIM*jnrA;
401 j_coord_offsetB = DIM*jnrB;
402 j_coord_offsetC = DIM*jnrC;
403 j_coord_offsetD = DIM*jnrD;
404 j_coord_offsetE = DIM*jnrE;
405 j_coord_offsetF = DIM*jnrF;
406 j_coord_offsetG = DIM*jnrG;
407 j_coord_offsetH = DIM*jnrH;
409 /* load j atom coordinates */
410 gmx_mm256_load_1rvec_8ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
411 x+j_coord_offsetC,x+j_coord_offsetD,
412 x+j_coord_offsetE,x+j_coord_offsetF,
413 x+j_coord_offsetG,x+j_coord_offsetH,
416 /* Calculate displacement vector */
417 dx10 = _mm256_sub_ps(ix1,jx0);
418 dy10 = _mm256_sub_ps(iy1,jy0);
419 dz10 = _mm256_sub_ps(iz1,jz0);
420 dx20 = _mm256_sub_ps(ix2,jx0);
421 dy20 = _mm256_sub_ps(iy2,jy0);
422 dz20 = _mm256_sub_ps(iz2,jz0);
423 dx30 = _mm256_sub_ps(ix3,jx0);
424 dy30 = _mm256_sub_ps(iy3,jy0);
425 dz30 = _mm256_sub_ps(iz3,jz0);
427 /* Calculate squared distance and things based on it */
428 rsq10 = gmx_mm256_calc_rsq_ps(dx10,dy10,dz10);
429 rsq20 = gmx_mm256_calc_rsq_ps(dx20,dy20,dz20);
430 rsq30 = gmx_mm256_calc_rsq_ps(dx30,dy30,dz30);
432 rinv10 = gmx_mm256_invsqrt_ps(rsq10);
433 rinv20 = gmx_mm256_invsqrt_ps(rsq20);
434 rinv30 = gmx_mm256_invsqrt_ps(rsq30);
436 rinvsq10 = _mm256_mul_ps(rinv10,rinv10);
437 rinvsq20 = _mm256_mul_ps(rinv20,rinv20);
438 rinvsq30 = _mm256_mul_ps(rinv30,rinv30);
440 /* Load parameters for j particles */
441 jq0 = gmx_mm256_load_8real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
442 charge+jnrC+0,charge+jnrD+0,
443 charge+jnrE+0,charge+jnrF+0,
444 charge+jnrG+0,charge+jnrH+0);
446 fjx0 = _mm256_setzero_ps();
447 fjy0 = _mm256_setzero_ps();
448 fjz0 = _mm256_setzero_ps();
450 /**************************
451 * CALCULATE INTERACTIONS *
452 **************************/
454 r10 = _mm256_mul_ps(rsq10,rinv10);
455 r10 = _mm256_andnot_ps(dummy_mask,r10);
457 /* Compute parameters for interactions between i and j atoms */
458 qq10 = _mm256_mul_ps(iq1,jq0);
460 /* EWALD ELECTROSTATICS */
462 /* Analytical PME correction */
463 zeta2 = _mm256_mul_ps(beta2,rsq10);
464 rinv3 = _mm256_mul_ps(rinvsq10,rinv10);
465 pmecorrF = gmx_mm256_pmecorrF_ps(zeta2);
466 felec = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
467 felec = _mm256_mul_ps(qq10,felec);
468 pmecorrV = gmx_mm256_pmecorrV_ps(zeta2);
469 pmecorrV = _mm256_mul_ps(pmecorrV,beta);
470 velec = _mm256_sub_ps(rinv10,pmecorrV);
471 velec = _mm256_mul_ps(qq10,velec);
473 /* Update potential sum for this i atom from the interaction with this j atom. */
474 velec = _mm256_andnot_ps(dummy_mask,velec);
475 velecsum = _mm256_add_ps(velecsum,velec);
479 fscal = _mm256_andnot_ps(dummy_mask,fscal);
481 /* Calculate temporary vectorial force */
482 tx = _mm256_mul_ps(fscal,dx10);
483 ty = _mm256_mul_ps(fscal,dy10);
484 tz = _mm256_mul_ps(fscal,dz10);
486 /* Update vectorial force */
487 fix1 = _mm256_add_ps(fix1,tx);
488 fiy1 = _mm256_add_ps(fiy1,ty);
489 fiz1 = _mm256_add_ps(fiz1,tz);
491 fjx0 = _mm256_add_ps(fjx0,tx);
492 fjy0 = _mm256_add_ps(fjy0,ty);
493 fjz0 = _mm256_add_ps(fjz0,tz);
495 /**************************
496 * CALCULATE INTERACTIONS *
497 **************************/
499 r20 = _mm256_mul_ps(rsq20,rinv20);
500 r20 = _mm256_andnot_ps(dummy_mask,r20);
502 /* Compute parameters for interactions between i and j atoms */
503 qq20 = _mm256_mul_ps(iq2,jq0);
505 /* EWALD ELECTROSTATICS */
507 /* Analytical PME correction */
508 zeta2 = _mm256_mul_ps(beta2,rsq20);
509 rinv3 = _mm256_mul_ps(rinvsq20,rinv20);
510 pmecorrF = gmx_mm256_pmecorrF_ps(zeta2);
511 felec = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
512 felec = _mm256_mul_ps(qq20,felec);
513 pmecorrV = gmx_mm256_pmecorrV_ps(zeta2);
514 pmecorrV = _mm256_mul_ps(pmecorrV,beta);
515 velec = _mm256_sub_ps(rinv20,pmecorrV);
516 velec = _mm256_mul_ps(qq20,velec);
518 /* Update potential sum for this i atom from the interaction with this j atom. */
519 velec = _mm256_andnot_ps(dummy_mask,velec);
520 velecsum = _mm256_add_ps(velecsum,velec);
524 fscal = _mm256_andnot_ps(dummy_mask,fscal);
526 /* Calculate temporary vectorial force */
527 tx = _mm256_mul_ps(fscal,dx20);
528 ty = _mm256_mul_ps(fscal,dy20);
529 tz = _mm256_mul_ps(fscal,dz20);
531 /* Update vectorial force */
532 fix2 = _mm256_add_ps(fix2,tx);
533 fiy2 = _mm256_add_ps(fiy2,ty);
534 fiz2 = _mm256_add_ps(fiz2,tz);
536 fjx0 = _mm256_add_ps(fjx0,tx);
537 fjy0 = _mm256_add_ps(fjy0,ty);
538 fjz0 = _mm256_add_ps(fjz0,tz);
540 /**************************
541 * CALCULATE INTERACTIONS *
542 **************************/
544 r30 = _mm256_mul_ps(rsq30,rinv30);
545 r30 = _mm256_andnot_ps(dummy_mask,r30);
547 /* Compute parameters for interactions between i and j atoms */
548 qq30 = _mm256_mul_ps(iq3,jq0);
550 /* EWALD ELECTROSTATICS */
552 /* Analytical PME correction */
553 zeta2 = _mm256_mul_ps(beta2,rsq30);
554 rinv3 = _mm256_mul_ps(rinvsq30,rinv30);
555 pmecorrF = gmx_mm256_pmecorrF_ps(zeta2);
556 felec = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
557 felec = _mm256_mul_ps(qq30,felec);
558 pmecorrV = gmx_mm256_pmecorrV_ps(zeta2);
559 pmecorrV = _mm256_mul_ps(pmecorrV,beta);
560 velec = _mm256_sub_ps(rinv30,pmecorrV);
561 velec = _mm256_mul_ps(qq30,velec);
563 /* Update potential sum for this i atom from the interaction with this j atom. */
564 velec = _mm256_andnot_ps(dummy_mask,velec);
565 velecsum = _mm256_add_ps(velecsum,velec);
569 fscal = _mm256_andnot_ps(dummy_mask,fscal);
571 /* Calculate temporary vectorial force */
572 tx = _mm256_mul_ps(fscal,dx30);
573 ty = _mm256_mul_ps(fscal,dy30);
574 tz = _mm256_mul_ps(fscal,dz30);
576 /* Update vectorial force */
577 fix3 = _mm256_add_ps(fix3,tx);
578 fiy3 = _mm256_add_ps(fiy3,ty);
579 fiz3 = _mm256_add_ps(fiz3,tz);
581 fjx0 = _mm256_add_ps(fjx0,tx);
582 fjy0 = _mm256_add_ps(fjy0,ty);
583 fjz0 = _mm256_add_ps(fjz0,tz);
585 fjptrA = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
586 fjptrB = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
587 fjptrC = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
588 fjptrD = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
589 fjptrE = (jnrlistE>=0) ? f+j_coord_offsetE : scratch;
590 fjptrF = (jnrlistF>=0) ? f+j_coord_offsetF : scratch;
591 fjptrG = (jnrlistG>=0) ? f+j_coord_offsetG : scratch;
592 fjptrH = (jnrlistH>=0) ? f+j_coord_offsetH : scratch;
594 gmx_mm256_decrement_1rvec_8ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjptrE,fjptrF,fjptrG,fjptrH,fjx0,fjy0,fjz0);
596 /* Inner loop uses 258 flops */
599 /* End of innermost loop */
601 gmx_mm256_update_iforce_3atom_swizzle_ps(fix1,fiy1,fiz1,fix2,fiy2,fiz2,fix3,fiy3,fiz3,
602 f+i_coord_offset+DIM,fshift+i_shift_offset);
605 /* Update potential energies */
606 gmx_mm256_update_1pot_ps(velecsum,kernel_data->energygrp_elec+ggid);
608 /* Increment number of inner iterations */
609 inneriter += j_index_end - j_index_start;
611 /* Outer loop uses 19 flops */
614 /* Increment number of outer iterations */
617 /* Update outer/inner flops */
619 inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_W4_VF,outeriter*19 + inneriter*258);
622 * Gromacs nonbonded kernel: nb_kernel_ElecEw_VdwNone_GeomW4P1_F_avx_256_single
623 * Electrostatics interaction: Ewald
624 * VdW interaction: None
625 * Geometry: Water4-Particle
626 * Calculate force/pot: Force
629 nb_kernel_ElecEw_VdwNone_GeomW4P1_F_avx_256_single
630 (t_nblist * gmx_restrict nlist,
631 rvec * gmx_restrict xx,
632 rvec * gmx_restrict ff,
633 t_forcerec * gmx_restrict fr,
634 t_mdatoms * gmx_restrict mdatoms,
635 nb_kernel_data_t * gmx_restrict kernel_data,
636 t_nrnb * gmx_restrict nrnb)
638 /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
639 * just 0 for non-waters.
640 * Suffixes A,B,C,D,E,F,G,H refer to j loop unrolling done with AVX, e.g. for the eight different
641 * jnr indices corresponding to data put in the four positions in the SIMD register.
643 int i_shift_offset,i_coord_offset,outeriter,inneriter;
644 int j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
645 int jnrA,jnrB,jnrC,jnrD;
646 int jnrE,jnrF,jnrG,jnrH;
647 int jnrlistA,jnrlistB,jnrlistC,jnrlistD;
648 int jnrlistE,jnrlistF,jnrlistG,jnrlistH;
649 int j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
650 int j_coord_offsetE,j_coord_offsetF,j_coord_offsetG,j_coord_offsetH;
651 int *iinr,*jindex,*jjnr,*shiftidx,*gid;
653 real *shiftvec,*fshift,*x,*f;
654 real *fjptrA,*fjptrB,*fjptrC,*fjptrD,*fjptrE,*fjptrF,*fjptrG,*fjptrH;
656 __m256 tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
657 real * vdwioffsetptr1;
658 __m256 ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
659 real * vdwioffsetptr2;
660 __m256 ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
661 real * vdwioffsetptr3;
662 __m256 ix3,iy3,iz3,fix3,fiy3,fiz3,iq3,isai3;
663 int vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D,vdwjidx0E,vdwjidx0F,vdwjidx0G,vdwjidx0H;
664 __m256 jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
665 __m256 dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
666 __m256 dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
667 __m256 dx30,dy30,dz30,rsq30,rinv30,rinvsq30,r30,qq30,c6_30,c12_30;
668 __m256 velec,felec,velecsum,facel,crf,krf,krf2;
671 __m128i ewitab_lo,ewitab_hi;
672 __m256 ewtabscale,eweps,sh_ewald,ewrt,ewtabhalfspace,ewtabF,ewtabFn,ewtabD,ewtabV;
673 __m256 beta,beta2,beta3,zeta2,pmecorrF,pmecorrV,rinv3;
675 __m256 dummy_mask,cutoff_mask;
676 __m256 signbit = _mm256_castsi256_ps( _mm256_set1_epi32(0x80000000) );
677 __m256 one = _mm256_set1_ps(1.0);
678 __m256 two = _mm256_set1_ps(2.0);
684 jindex = nlist->jindex;
686 shiftidx = nlist->shift;
688 shiftvec = fr->shift_vec[0];
689 fshift = fr->fshift[0];
690 facel = _mm256_set1_ps(fr->epsfac);
691 charge = mdatoms->chargeA;
693 sh_ewald = _mm256_set1_ps(fr->ic->sh_ewald);
694 beta = _mm256_set1_ps(fr->ic->ewaldcoeff);
695 beta2 = _mm256_mul_ps(beta,beta);
696 beta3 = _mm256_mul_ps(beta,beta2);
698 ewtab = fr->ic->tabq_coul_F;
699 ewtabscale = _mm256_set1_ps(fr->ic->tabq_scale);
700 ewtabhalfspace = _mm256_set1_ps(0.5/fr->ic->tabq_scale);
702 /* Setup water-specific parameters */
703 inr = nlist->iinr[0];
704 iq1 = _mm256_mul_ps(facel,_mm256_set1_ps(charge[inr+1]));
705 iq2 = _mm256_mul_ps(facel,_mm256_set1_ps(charge[inr+2]));
706 iq3 = _mm256_mul_ps(facel,_mm256_set1_ps(charge[inr+3]));
708 /* Avoid stupid compiler warnings */
709 jnrA = jnrB = jnrC = jnrD = jnrE = jnrF = jnrG = jnrH = 0;
722 for(iidx=0;iidx<4*DIM;iidx++)
727 /* Start outer loop over neighborlists */
728 for(iidx=0; iidx<nri; iidx++)
730 /* Load shift vector for this list */
731 i_shift_offset = DIM*shiftidx[iidx];
733 /* Load limits for loop over neighbors */
734 j_index_start = jindex[iidx];
735 j_index_end = jindex[iidx+1];
737 /* Get outer coordinate index */
739 i_coord_offset = DIM*inr;
741 /* Load i particle coords and add shift vector */
742 gmx_mm256_load_shift_and_3rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset+DIM,
743 &ix1,&iy1,&iz1,&ix2,&iy2,&iz2,&ix3,&iy3,&iz3);
745 fix1 = _mm256_setzero_ps();
746 fiy1 = _mm256_setzero_ps();
747 fiz1 = _mm256_setzero_ps();
748 fix2 = _mm256_setzero_ps();
749 fiy2 = _mm256_setzero_ps();
750 fiz2 = _mm256_setzero_ps();
751 fix3 = _mm256_setzero_ps();
752 fiy3 = _mm256_setzero_ps();
753 fiz3 = _mm256_setzero_ps();
755 /* Start inner kernel loop */
756 for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+7]>=0; jidx+=8)
759 /* Get j neighbor index, and coordinate index */
768 j_coord_offsetA = DIM*jnrA;
769 j_coord_offsetB = DIM*jnrB;
770 j_coord_offsetC = DIM*jnrC;
771 j_coord_offsetD = DIM*jnrD;
772 j_coord_offsetE = DIM*jnrE;
773 j_coord_offsetF = DIM*jnrF;
774 j_coord_offsetG = DIM*jnrG;
775 j_coord_offsetH = DIM*jnrH;
777 /* load j atom coordinates */
778 gmx_mm256_load_1rvec_8ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
779 x+j_coord_offsetC,x+j_coord_offsetD,
780 x+j_coord_offsetE,x+j_coord_offsetF,
781 x+j_coord_offsetG,x+j_coord_offsetH,
784 /* Calculate displacement vector */
785 dx10 = _mm256_sub_ps(ix1,jx0);
786 dy10 = _mm256_sub_ps(iy1,jy0);
787 dz10 = _mm256_sub_ps(iz1,jz0);
788 dx20 = _mm256_sub_ps(ix2,jx0);
789 dy20 = _mm256_sub_ps(iy2,jy0);
790 dz20 = _mm256_sub_ps(iz2,jz0);
791 dx30 = _mm256_sub_ps(ix3,jx0);
792 dy30 = _mm256_sub_ps(iy3,jy0);
793 dz30 = _mm256_sub_ps(iz3,jz0);
795 /* Calculate squared distance and things based on it */
796 rsq10 = gmx_mm256_calc_rsq_ps(dx10,dy10,dz10);
797 rsq20 = gmx_mm256_calc_rsq_ps(dx20,dy20,dz20);
798 rsq30 = gmx_mm256_calc_rsq_ps(dx30,dy30,dz30);
800 rinv10 = gmx_mm256_invsqrt_ps(rsq10);
801 rinv20 = gmx_mm256_invsqrt_ps(rsq20);
802 rinv30 = gmx_mm256_invsqrt_ps(rsq30);
804 rinvsq10 = _mm256_mul_ps(rinv10,rinv10);
805 rinvsq20 = _mm256_mul_ps(rinv20,rinv20);
806 rinvsq30 = _mm256_mul_ps(rinv30,rinv30);
808 /* Load parameters for j particles */
809 jq0 = gmx_mm256_load_8real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
810 charge+jnrC+0,charge+jnrD+0,
811 charge+jnrE+0,charge+jnrF+0,
812 charge+jnrG+0,charge+jnrH+0);
814 fjx0 = _mm256_setzero_ps();
815 fjy0 = _mm256_setzero_ps();
816 fjz0 = _mm256_setzero_ps();
818 /**************************
819 * CALCULATE INTERACTIONS *
820 **************************/
822 r10 = _mm256_mul_ps(rsq10,rinv10);
824 /* Compute parameters for interactions between i and j atoms */
825 qq10 = _mm256_mul_ps(iq1,jq0);
827 /* EWALD ELECTROSTATICS */
829 /* Analytical PME correction */
830 zeta2 = _mm256_mul_ps(beta2,rsq10);
831 rinv3 = _mm256_mul_ps(rinvsq10,rinv10);
832 pmecorrF = gmx_mm256_pmecorrF_ps(zeta2);
833 felec = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
834 felec = _mm256_mul_ps(qq10,felec);
838 /* Calculate temporary vectorial force */
839 tx = _mm256_mul_ps(fscal,dx10);
840 ty = _mm256_mul_ps(fscal,dy10);
841 tz = _mm256_mul_ps(fscal,dz10);
843 /* Update vectorial force */
844 fix1 = _mm256_add_ps(fix1,tx);
845 fiy1 = _mm256_add_ps(fiy1,ty);
846 fiz1 = _mm256_add_ps(fiz1,tz);
848 fjx0 = _mm256_add_ps(fjx0,tx);
849 fjy0 = _mm256_add_ps(fjy0,ty);
850 fjz0 = _mm256_add_ps(fjz0,tz);
852 /**************************
853 * CALCULATE INTERACTIONS *
854 **************************/
856 r20 = _mm256_mul_ps(rsq20,rinv20);
858 /* Compute parameters for interactions between i and j atoms */
859 qq20 = _mm256_mul_ps(iq2,jq0);
861 /* EWALD ELECTROSTATICS */
863 /* Analytical PME correction */
864 zeta2 = _mm256_mul_ps(beta2,rsq20);
865 rinv3 = _mm256_mul_ps(rinvsq20,rinv20);
866 pmecorrF = gmx_mm256_pmecorrF_ps(zeta2);
867 felec = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
868 felec = _mm256_mul_ps(qq20,felec);
872 /* Calculate temporary vectorial force */
873 tx = _mm256_mul_ps(fscal,dx20);
874 ty = _mm256_mul_ps(fscal,dy20);
875 tz = _mm256_mul_ps(fscal,dz20);
877 /* Update vectorial force */
878 fix2 = _mm256_add_ps(fix2,tx);
879 fiy2 = _mm256_add_ps(fiy2,ty);
880 fiz2 = _mm256_add_ps(fiz2,tz);
882 fjx0 = _mm256_add_ps(fjx0,tx);
883 fjy0 = _mm256_add_ps(fjy0,ty);
884 fjz0 = _mm256_add_ps(fjz0,tz);
886 /**************************
887 * CALCULATE INTERACTIONS *
888 **************************/
890 r30 = _mm256_mul_ps(rsq30,rinv30);
892 /* Compute parameters for interactions between i and j atoms */
893 qq30 = _mm256_mul_ps(iq3,jq0);
895 /* EWALD ELECTROSTATICS */
897 /* Analytical PME correction */
898 zeta2 = _mm256_mul_ps(beta2,rsq30);
899 rinv3 = _mm256_mul_ps(rinvsq30,rinv30);
900 pmecorrF = gmx_mm256_pmecorrF_ps(zeta2);
901 felec = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
902 felec = _mm256_mul_ps(qq30,felec);
906 /* Calculate temporary vectorial force */
907 tx = _mm256_mul_ps(fscal,dx30);
908 ty = _mm256_mul_ps(fscal,dy30);
909 tz = _mm256_mul_ps(fscal,dz30);
911 /* Update vectorial force */
912 fix3 = _mm256_add_ps(fix3,tx);
913 fiy3 = _mm256_add_ps(fiy3,ty);
914 fiz3 = _mm256_add_ps(fiz3,tz);
916 fjx0 = _mm256_add_ps(fjx0,tx);
917 fjy0 = _mm256_add_ps(fjy0,ty);
918 fjz0 = _mm256_add_ps(fjz0,tz);
920 fjptrA = f+j_coord_offsetA;
921 fjptrB = f+j_coord_offsetB;
922 fjptrC = f+j_coord_offsetC;
923 fjptrD = f+j_coord_offsetD;
924 fjptrE = f+j_coord_offsetE;
925 fjptrF = f+j_coord_offsetF;
926 fjptrG = f+j_coord_offsetG;
927 fjptrH = f+j_coord_offsetH;
929 gmx_mm256_decrement_1rvec_8ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjptrE,fjptrF,fjptrG,fjptrH,fjx0,fjy0,fjz0);
931 /* Inner loop uses 171 flops */
937 /* Get j neighbor index, and coordinate index */
938 jnrlistA = jjnr[jidx];
939 jnrlistB = jjnr[jidx+1];
940 jnrlistC = jjnr[jidx+2];
941 jnrlistD = jjnr[jidx+3];
942 jnrlistE = jjnr[jidx+4];
943 jnrlistF = jjnr[jidx+5];
944 jnrlistG = jjnr[jidx+6];
945 jnrlistH = jjnr[jidx+7];
946 /* Sign of each element will be negative for non-real atoms.
947 * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
948 * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
950 dummy_mask = gmx_mm256_set_m128(gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx+4)),_mm_setzero_si128())),
951 gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128())));
953 jnrA = (jnrlistA>=0) ? jnrlistA : 0;
954 jnrB = (jnrlistB>=0) ? jnrlistB : 0;
955 jnrC = (jnrlistC>=0) ? jnrlistC : 0;
956 jnrD = (jnrlistD>=0) ? jnrlistD : 0;
957 jnrE = (jnrlistE>=0) ? jnrlistE : 0;
958 jnrF = (jnrlistF>=0) ? jnrlistF : 0;
959 jnrG = (jnrlistG>=0) ? jnrlistG : 0;
960 jnrH = (jnrlistH>=0) ? jnrlistH : 0;
961 j_coord_offsetA = DIM*jnrA;
962 j_coord_offsetB = DIM*jnrB;
963 j_coord_offsetC = DIM*jnrC;
964 j_coord_offsetD = DIM*jnrD;
965 j_coord_offsetE = DIM*jnrE;
966 j_coord_offsetF = DIM*jnrF;
967 j_coord_offsetG = DIM*jnrG;
968 j_coord_offsetH = DIM*jnrH;
970 /* load j atom coordinates */
971 gmx_mm256_load_1rvec_8ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
972 x+j_coord_offsetC,x+j_coord_offsetD,
973 x+j_coord_offsetE,x+j_coord_offsetF,
974 x+j_coord_offsetG,x+j_coord_offsetH,
977 /* Calculate displacement vector */
978 dx10 = _mm256_sub_ps(ix1,jx0);
979 dy10 = _mm256_sub_ps(iy1,jy0);
980 dz10 = _mm256_sub_ps(iz1,jz0);
981 dx20 = _mm256_sub_ps(ix2,jx0);
982 dy20 = _mm256_sub_ps(iy2,jy0);
983 dz20 = _mm256_sub_ps(iz2,jz0);
984 dx30 = _mm256_sub_ps(ix3,jx0);
985 dy30 = _mm256_sub_ps(iy3,jy0);
986 dz30 = _mm256_sub_ps(iz3,jz0);
988 /* Calculate squared distance and things based on it */
989 rsq10 = gmx_mm256_calc_rsq_ps(dx10,dy10,dz10);
990 rsq20 = gmx_mm256_calc_rsq_ps(dx20,dy20,dz20);
991 rsq30 = gmx_mm256_calc_rsq_ps(dx30,dy30,dz30);
993 rinv10 = gmx_mm256_invsqrt_ps(rsq10);
994 rinv20 = gmx_mm256_invsqrt_ps(rsq20);
995 rinv30 = gmx_mm256_invsqrt_ps(rsq30);
997 rinvsq10 = _mm256_mul_ps(rinv10,rinv10);
998 rinvsq20 = _mm256_mul_ps(rinv20,rinv20);
999 rinvsq30 = _mm256_mul_ps(rinv30,rinv30);
1001 /* Load parameters for j particles */
1002 jq0 = gmx_mm256_load_8real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
1003 charge+jnrC+0,charge+jnrD+0,
1004 charge+jnrE+0,charge+jnrF+0,
1005 charge+jnrG+0,charge+jnrH+0);
1007 fjx0 = _mm256_setzero_ps();
1008 fjy0 = _mm256_setzero_ps();
1009 fjz0 = _mm256_setzero_ps();
1011 /**************************
1012 * CALCULATE INTERACTIONS *
1013 **************************/
1015 r10 = _mm256_mul_ps(rsq10,rinv10);
1016 r10 = _mm256_andnot_ps(dummy_mask,r10);
1018 /* Compute parameters for interactions between i and j atoms */
1019 qq10 = _mm256_mul_ps(iq1,jq0);
1021 /* EWALD ELECTROSTATICS */
1023 /* Analytical PME correction */
1024 zeta2 = _mm256_mul_ps(beta2,rsq10);
1025 rinv3 = _mm256_mul_ps(rinvsq10,rinv10);
1026 pmecorrF = gmx_mm256_pmecorrF_ps(zeta2);
1027 felec = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
1028 felec = _mm256_mul_ps(qq10,felec);
1032 fscal = _mm256_andnot_ps(dummy_mask,fscal);
1034 /* Calculate temporary vectorial force */
1035 tx = _mm256_mul_ps(fscal,dx10);
1036 ty = _mm256_mul_ps(fscal,dy10);
1037 tz = _mm256_mul_ps(fscal,dz10);
1039 /* Update vectorial force */
1040 fix1 = _mm256_add_ps(fix1,tx);
1041 fiy1 = _mm256_add_ps(fiy1,ty);
1042 fiz1 = _mm256_add_ps(fiz1,tz);
1044 fjx0 = _mm256_add_ps(fjx0,tx);
1045 fjy0 = _mm256_add_ps(fjy0,ty);
1046 fjz0 = _mm256_add_ps(fjz0,tz);
1048 /**************************
1049 * CALCULATE INTERACTIONS *
1050 **************************/
1052 r20 = _mm256_mul_ps(rsq20,rinv20);
1053 r20 = _mm256_andnot_ps(dummy_mask,r20);
1055 /* Compute parameters for interactions between i and j atoms */
1056 qq20 = _mm256_mul_ps(iq2,jq0);
1058 /* EWALD ELECTROSTATICS */
1060 /* Analytical PME correction */
1061 zeta2 = _mm256_mul_ps(beta2,rsq20);
1062 rinv3 = _mm256_mul_ps(rinvsq20,rinv20);
1063 pmecorrF = gmx_mm256_pmecorrF_ps(zeta2);
1064 felec = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
1065 felec = _mm256_mul_ps(qq20,felec);
1069 fscal = _mm256_andnot_ps(dummy_mask,fscal);
1071 /* Calculate temporary vectorial force */
1072 tx = _mm256_mul_ps(fscal,dx20);
1073 ty = _mm256_mul_ps(fscal,dy20);
1074 tz = _mm256_mul_ps(fscal,dz20);
1076 /* Update vectorial force */
1077 fix2 = _mm256_add_ps(fix2,tx);
1078 fiy2 = _mm256_add_ps(fiy2,ty);
1079 fiz2 = _mm256_add_ps(fiz2,tz);
1081 fjx0 = _mm256_add_ps(fjx0,tx);
1082 fjy0 = _mm256_add_ps(fjy0,ty);
1083 fjz0 = _mm256_add_ps(fjz0,tz);
1085 /**************************
1086 * CALCULATE INTERACTIONS *
1087 **************************/
1089 r30 = _mm256_mul_ps(rsq30,rinv30);
1090 r30 = _mm256_andnot_ps(dummy_mask,r30);
1092 /* Compute parameters for interactions between i and j atoms */
1093 qq30 = _mm256_mul_ps(iq3,jq0);
1095 /* EWALD ELECTROSTATICS */
1097 /* Analytical PME correction */
1098 zeta2 = _mm256_mul_ps(beta2,rsq30);
1099 rinv3 = _mm256_mul_ps(rinvsq30,rinv30);
1100 pmecorrF = gmx_mm256_pmecorrF_ps(zeta2);
1101 felec = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
1102 felec = _mm256_mul_ps(qq30,felec);
1106 fscal = _mm256_andnot_ps(dummy_mask,fscal);
1108 /* Calculate temporary vectorial force */
1109 tx = _mm256_mul_ps(fscal,dx30);
1110 ty = _mm256_mul_ps(fscal,dy30);
1111 tz = _mm256_mul_ps(fscal,dz30);
1113 /* Update vectorial force */
1114 fix3 = _mm256_add_ps(fix3,tx);
1115 fiy3 = _mm256_add_ps(fiy3,ty);
1116 fiz3 = _mm256_add_ps(fiz3,tz);
1118 fjx0 = _mm256_add_ps(fjx0,tx);
1119 fjy0 = _mm256_add_ps(fjy0,ty);
1120 fjz0 = _mm256_add_ps(fjz0,tz);
1122 fjptrA = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
1123 fjptrB = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
1124 fjptrC = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
1125 fjptrD = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
1126 fjptrE = (jnrlistE>=0) ? f+j_coord_offsetE : scratch;
1127 fjptrF = (jnrlistF>=0) ? f+j_coord_offsetF : scratch;
1128 fjptrG = (jnrlistG>=0) ? f+j_coord_offsetG : scratch;
1129 fjptrH = (jnrlistH>=0) ? f+j_coord_offsetH : scratch;
1131 gmx_mm256_decrement_1rvec_8ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjptrE,fjptrF,fjptrG,fjptrH,fjx0,fjy0,fjz0);
1133 /* Inner loop uses 174 flops */
1136 /* End of innermost loop */
1138 gmx_mm256_update_iforce_3atom_swizzle_ps(fix1,fiy1,fiz1,fix2,fiy2,fiz2,fix3,fiy3,fiz3,
1139 f+i_coord_offset+DIM,fshift+i_shift_offset);
1141 /* Increment number of inner iterations */
1142 inneriter += j_index_end - j_index_start;
1144 /* Outer loop uses 18 flops */
1147 /* Increment number of outer iterations */
1150 /* Update outer/inner flops */
1152 inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_W4_F,outeriter*18 + inneriter*174);