2 * Note: this file was generated by the Gromacs avx_128_fma_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_128_fma_single.h"
34 #include "kernelutil_x86_avx_128_fma_single.h"
37 * Gromacs nonbonded kernel: nb_kernel_ElecEwSh_VdwNone_GeomP1P1_VF_avx_128_fma_single
38 * Electrostatics interaction: Ewald
39 * VdW interaction: None
40 * Geometry: Particle-Particle
41 * Calculate force/pot: PotentialAndForce
44 nb_kernel_ElecEwSh_VdwNone_GeomP1P1_VF_avx_128_fma_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 refer to j loop unrolling done with AVX_128, 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 j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
63 int *iinr,*jindex,*jjnr,*shiftidx,*gid;
65 real *shiftvec,*fshift,*x,*f;
66 real *fjptrA,*fjptrB,*fjptrC,*fjptrD;
68 __m128 fscal,rcutoff,rcutoff2,jidxall;
70 __m128 ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
71 int vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
72 __m128 jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
73 __m128 dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
74 __m128 velec,felec,velecsum,facel,crf,krf,krf2;
77 __m128 ewtabscale,eweps,twoeweps,sh_ewald,ewrt,ewtabhalfspace,ewtabF,ewtabFn,ewtabD,ewtabV;
78 __m128 beta,beta2,beta3,zeta2,pmecorrF,pmecorrV,rinv3;
80 __m128 dummy_mask,cutoff_mask;
81 __m128 signbit = _mm_castsi128_ps( _mm_set1_epi32(0x80000000) );
82 __m128 one = _mm_set1_ps(1.0);
83 __m128 two = _mm_set1_ps(2.0);
89 jindex = nlist->jindex;
91 shiftidx = nlist->shift;
93 shiftvec = fr->shift_vec[0];
94 fshift = fr->fshift[0];
95 facel = _mm_set1_ps(fr->epsfac);
96 charge = mdatoms->chargeA;
98 sh_ewald = _mm_set1_ps(fr->ic->sh_ewald);
99 beta = _mm_set1_ps(fr->ic->ewaldcoeff);
100 beta2 = _mm_mul_ps(beta,beta);
101 beta3 = _mm_mul_ps(beta,beta2);
102 ewtab = fr->ic->tabq_coul_FDV0;
103 ewtabscale = _mm_set1_ps(fr->ic->tabq_scale);
104 ewtabhalfspace = _mm_set1_ps(0.5/fr->ic->tabq_scale);
106 /* When we use explicit cutoffs the value must be identical for elec and VdW, so use elec as an arbitrary choice */
107 rcutoff_scalar = fr->rcoulomb;
108 rcutoff = _mm_set1_ps(rcutoff_scalar);
109 rcutoff2 = _mm_mul_ps(rcutoff,rcutoff);
111 /* Avoid stupid compiler warnings */
112 jnrA = jnrB = jnrC = jnrD = 0;
121 for(iidx=0;iidx<4*DIM;iidx++)
126 /* Start outer loop over neighborlists */
127 for(iidx=0; iidx<nri; iidx++)
129 /* Load shift vector for this list */
130 i_shift_offset = DIM*shiftidx[iidx];
132 /* Load limits for loop over neighbors */
133 j_index_start = jindex[iidx];
134 j_index_end = jindex[iidx+1];
136 /* Get outer coordinate index */
138 i_coord_offset = DIM*inr;
140 /* Load i particle coords and add shift vector */
141 gmx_mm_load_shift_and_1rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,&ix0,&iy0,&iz0);
143 fix0 = _mm_setzero_ps();
144 fiy0 = _mm_setzero_ps();
145 fiz0 = _mm_setzero_ps();
147 /* Load parameters for i particles */
148 iq0 = _mm_mul_ps(facel,_mm_load1_ps(charge+inr+0));
150 /* Reset potential sums */
151 velecsum = _mm_setzero_ps();
153 /* Start inner kernel loop */
154 for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
157 /* Get j neighbor index, and coordinate index */
162 j_coord_offsetA = DIM*jnrA;
163 j_coord_offsetB = DIM*jnrB;
164 j_coord_offsetC = DIM*jnrC;
165 j_coord_offsetD = DIM*jnrD;
167 /* load j atom coordinates */
168 gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
169 x+j_coord_offsetC,x+j_coord_offsetD,
172 /* Calculate displacement vector */
173 dx00 = _mm_sub_ps(ix0,jx0);
174 dy00 = _mm_sub_ps(iy0,jy0);
175 dz00 = _mm_sub_ps(iz0,jz0);
177 /* Calculate squared distance and things based on it */
178 rsq00 = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
180 rinv00 = gmx_mm_invsqrt_ps(rsq00);
182 rinvsq00 = _mm_mul_ps(rinv00,rinv00);
184 /* Load parameters for j particles */
185 jq0 = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
186 charge+jnrC+0,charge+jnrD+0);
188 /**************************
189 * CALCULATE INTERACTIONS *
190 **************************/
192 if (gmx_mm_any_lt(rsq00,rcutoff2))
195 r00 = _mm_mul_ps(rsq00,rinv00);
197 /* Compute parameters for interactions between i and j atoms */
198 qq00 = _mm_mul_ps(iq0,jq0);
200 /* EWALD ELECTROSTATICS */
202 /* Analytical PME correction */
203 zeta2 = _mm_mul_ps(beta2,rsq00);
204 rinv3 = _mm_mul_ps(rinvsq00,rinv00);
205 pmecorrF = gmx_mm_pmecorrF_ps(zeta2);
206 felec = _mm_macc_ps(pmecorrF,beta3,rinv3);
207 felec = _mm_mul_ps(qq00,felec);
208 pmecorrV = gmx_mm_pmecorrV_ps(zeta2);
209 velec = _mm_nmacc_ps(pmecorrV,beta,_mm_sub_ps(rinv00,sh_ewald));
210 velec = _mm_mul_ps(qq00,velec);
212 cutoff_mask = _mm_cmplt_ps(rsq00,rcutoff2);
214 /* Update potential sum for this i atom from the interaction with this j atom. */
215 velec = _mm_and_ps(velec,cutoff_mask);
216 velecsum = _mm_add_ps(velecsum,velec);
220 fscal = _mm_and_ps(fscal,cutoff_mask);
222 /* Update vectorial force */
223 fix0 = _mm_macc_ps(dx00,fscal,fix0);
224 fiy0 = _mm_macc_ps(dy00,fscal,fiy0);
225 fiz0 = _mm_macc_ps(dz00,fscal,fiz0);
227 fjptrA = f+j_coord_offsetA;
228 fjptrB = f+j_coord_offsetB;
229 fjptrC = f+j_coord_offsetC;
230 fjptrD = f+j_coord_offsetD;
231 gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,
232 _mm_mul_ps(dx00,fscal),
233 _mm_mul_ps(dy00,fscal),
234 _mm_mul_ps(dz00,fscal));
238 /* Inner loop uses 33 flops */
244 /* Get j neighbor index, and coordinate index */
245 jnrlistA = jjnr[jidx];
246 jnrlistB = jjnr[jidx+1];
247 jnrlistC = jjnr[jidx+2];
248 jnrlistD = jjnr[jidx+3];
249 /* Sign of each element will be negative for non-real atoms.
250 * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
251 * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
253 dummy_mask = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
254 jnrA = (jnrlistA>=0) ? jnrlistA : 0;
255 jnrB = (jnrlistB>=0) ? jnrlistB : 0;
256 jnrC = (jnrlistC>=0) ? jnrlistC : 0;
257 jnrD = (jnrlistD>=0) ? jnrlistD : 0;
258 j_coord_offsetA = DIM*jnrA;
259 j_coord_offsetB = DIM*jnrB;
260 j_coord_offsetC = DIM*jnrC;
261 j_coord_offsetD = DIM*jnrD;
263 /* load j atom coordinates */
264 gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
265 x+j_coord_offsetC,x+j_coord_offsetD,
268 /* Calculate displacement vector */
269 dx00 = _mm_sub_ps(ix0,jx0);
270 dy00 = _mm_sub_ps(iy0,jy0);
271 dz00 = _mm_sub_ps(iz0,jz0);
273 /* Calculate squared distance and things based on it */
274 rsq00 = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
276 rinv00 = gmx_mm_invsqrt_ps(rsq00);
278 rinvsq00 = _mm_mul_ps(rinv00,rinv00);
280 /* Load parameters for j particles */
281 jq0 = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
282 charge+jnrC+0,charge+jnrD+0);
284 /**************************
285 * CALCULATE INTERACTIONS *
286 **************************/
288 if (gmx_mm_any_lt(rsq00,rcutoff2))
291 r00 = _mm_mul_ps(rsq00,rinv00);
292 r00 = _mm_andnot_ps(dummy_mask,r00);
294 /* Compute parameters for interactions between i and j atoms */
295 qq00 = _mm_mul_ps(iq0,jq0);
297 /* EWALD ELECTROSTATICS */
299 /* Analytical PME correction */
300 zeta2 = _mm_mul_ps(beta2,rsq00);
301 rinv3 = _mm_mul_ps(rinvsq00,rinv00);
302 pmecorrF = gmx_mm_pmecorrF_ps(zeta2);
303 felec = _mm_macc_ps(pmecorrF,beta3,rinv3);
304 felec = _mm_mul_ps(qq00,felec);
305 pmecorrV = gmx_mm_pmecorrV_ps(zeta2);
306 velec = _mm_nmacc_ps(pmecorrV,beta,_mm_sub_ps(rinv00,sh_ewald));
307 velec = _mm_mul_ps(qq00,velec);
309 cutoff_mask = _mm_cmplt_ps(rsq00,rcutoff2);
311 /* Update potential sum for this i atom from the interaction with this j atom. */
312 velec = _mm_and_ps(velec,cutoff_mask);
313 velec = _mm_andnot_ps(dummy_mask,velec);
314 velecsum = _mm_add_ps(velecsum,velec);
318 fscal = _mm_and_ps(fscal,cutoff_mask);
320 fscal = _mm_andnot_ps(dummy_mask,fscal);
322 /* Update vectorial force */
323 fix0 = _mm_macc_ps(dx00,fscal,fix0);
324 fiy0 = _mm_macc_ps(dy00,fscal,fiy0);
325 fiz0 = _mm_macc_ps(dz00,fscal,fiz0);
327 fjptrA = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
328 fjptrB = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
329 fjptrC = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
330 fjptrD = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
331 gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,
332 _mm_mul_ps(dx00,fscal),
333 _mm_mul_ps(dy00,fscal),
334 _mm_mul_ps(dz00,fscal));
338 /* Inner loop uses 34 flops */
341 /* End of innermost loop */
343 gmx_mm_update_iforce_1atom_swizzle_ps(fix0,fiy0,fiz0,
344 f+i_coord_offset,fshift+i_shift_offset);
347 /* Update potential energies */
348 gmx_mm_update_1pot_ps(velecsum,kernel_data->energygrp_elec+ggid);
350 /* Increment number of inner iterations */
351 inneriter += j_index_end - j_index_start;
353 /* Outer loop uses 8 flops */
356 /* Increment number of outer iterations */
359 /* Update outer/inner flops */
361 inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VF,outeriter*8 + inneriter*34);
364 * Gromacs nonbonded kernel: nb_kernel_ElecEwSh_VdwNone_GeomP1P1_F_avx_128_fma_single
365 * Electrostatics interaction: Ewald
366 * VdW interaction: None
367 * Geometry: Particle-Particle
368 * Calculate force/pot: Force
371 nb_kernel_ElecEwSh_VdwNone_GeomP1P1_F_avx_128_fma_single
372 (t_nblist * gmx_restrict nlist,
373 rvec * gmx_restrict xx,
374 rvec * gmx_restrict ff,
375 t_forcerec * gmx_restrict fr,
376 t_mdatoms * gmx_restrict mdatoms,
377 nb_kernel_data_t * gmx_restrict kernel_data,
378 t_nrnb * gmx_restrict nrnb)
380 /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
381 * just 0 for non-waters.
382 * Suffixes A,B,C,D refer to j loop unrolling done with AVX_128, e.g. for the four different
383 * jnr indices corresponding to data put in the four positions in the SIMD register.
385 int i_shift_offset,i_coord_offset,outeriter,inneriter;
386 int j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
387 int jnrA,jnrB,jnrC,jnrD;
388 int jnrlistA,jnrlistB,jnrlistC,jnrlistD;
389 int j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
390 int *iinr,*jindex,*jjnr,*shiftidx,*gid;
392 real *shiftvec,*fshift,*x,*f;
393 real *fjptrA,*fjptrB,*fjptrC,*fjptrD;
395 __m128 fscal,rcutoff,rcutoff2,jidxall;
397 __m128 ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
398 int vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
399 __m128 jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
400 __m128 dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
401 __m128 velec,felec,velecsum,facel,crf,krf,krf2;
404 __m128 ewtabscale,eweps,twoeweps,sh_ewald,ewrt,ewtabhalfspace,ewtabF,ewtabFn,ewtabD,ewtabV;
405 __m128 beta,beta2,beta3,zeta2,pmecorrF,pmecorrV,rinv3;
407 __m128 dummy_mask,cutoff_mask;
408 __m128 signbit = _mm_castsi128_ps( _mm_set1_epi32(0x80000000) );
409 __m128 one = _mm_set1_ps(1.0);
410 __m128 two = _mm_set1_ps(2.0);
416 jindex = nlist->jindex;
418 shiftidx = nlist->shift;
420 shiftvec = fr->shift_vec[0];
421 fshift = fr->fshift[0];
422 facel = _mm_set1_ps(fr->epsfac);
423 charge = mdatoms->chargeA;
425 sh_ewald = _mm_set1_ps(fr->ic->sh_ewald);
426 beta = _mm_set1_ps(fr->ic->ewaldcoeff);
427 beta2 = _mm_mul_ps(beta,beta);
428 beta3 = _mm_mul_ps(beta,beta2);
429 ewtab = fr->ic->tabq_coul_F;
430 ewtabscale = _mm_set1_ps(fr->ic->tabq_scale);
431 ewtabhalfspace = _mm_set1_ps(0.5/fr->ic->tabq_scale);
433 /* When we use explicit cutoffs the value must be identical for elec and VdW, so use elec as an arbitrary choice */
434 rcutoff_scalar = fr->rcoulomb;
435 rcutoff = _mm_set1_ps(rcutoff_scalar);
436 rcutoff2 = _mm_mul_ps(rcutoff,rcutoff);
438 /* Avoid stupid compiler warnings */
439 jnrA = jnrB = jnrC = jnrD = 0;
448 for(iidx=0;iidx<4*DIM;iidx++)
453 /* Start outer loop over neighborlists */
454 for(iidx=0; iidx<nri; iidx++)
456 /* Load shift vector for this list */
457 i_shift_offset = DIM*shiftidx[iidx];
459 /* Load limits for loop over neighbors */
460 j_index_start = jindex[iidx];
461 j_index_end = jindex[iidx+1];
463 /* Get outer coordinate index */
465 i_coord_offset = DIM*inr;
467 /* Load i particle coords and add shift vector */
468 gmx_mm_load_shift_and_1rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,&ix0,&iy0,&iz0);
470 fix0 = _mm_setzero_ps();
471 fiy0 = _mm_setzero_ps();
472 fiz0 = _mm_setzero_ps();
474 /* Load parameters for i particles */
475 iq0 = _mm_mul_ps(facel,_mm_load1_ps(charge+inr+0));
477 /* Start inner kernel loop */
478 for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
481 /* Get j neighbor index, and coordinate index */
486 j_coord_offsetA = DIM*jnrA;
487 j_coord_offsetB = DIM*jnrB;
488 j_coord_offsetC = DIM*jnrC;
489 j_coord_offsetD = DIM*jnrD;
491 /* load j atom coordinates */
492 gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
493 x+j_coord_offsetC,x+j_coord_offsetD,
496 /* Calculate displacement vector */
497 dx00 = _mm_sub_ps(ix0,jx0);
498 dy00 = _mm_sub_ps(iy0,jy0);
499 dz00 = _mm_sub_ps(iz0,jz0);
501 /* Calculate squared distance and things based on it */
502 rsq00 = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
504 rinv00 = gmx_mm_invsqrt_ps(rsq00);
506 rinvsq00 = _mm_mul_ps(rinv00,rinv00);
508 /* Load parameters for j particles */
509 jq0 = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
510 charge+jnrC+0,charge+jnrD+0);
512 /**************************
513 * CALCULATE INTERACTIONS *
514 **************************/
516 if (gmx_mm_any_lt(rsq00,rcutoff2))
519 r00 = _mm_mul_ps(rsq00,rinv00);
521 /* Compute parameters for interactions between i and j atoms */
522 qq00 = _mm_mul_ps(iq0,jq0);
524 /* EWALD ELECTROSTATICS */
526 /* Analytical PME correction */
527 zeta2 = _mm_mul_ps(beta2,rsq00);
528 rinv3 = _mm_mul_ps(rinvsq00,rinv00);
529 pmecorrF = gmx_mm_pmecorrF_ps(zeta2);
530 felec = _mm_macc_ps(pmecorrF,beta3,rinv3);
531 felec = _mm_mul_ps(qq00,felec);
533 cutoff_mask = _mm_cmplt_ps(rsq00,rcutoff2);
537 fscal = _mm_and_ps(fscal,cutoff_mask);
539 /* Update vectorial force */
540 fix0 = _mm_macc_ps(dx00,fscal,fix0);
541 fiy0 = _mm_macc_ps(dy00,fscal,fiy0);
542 fiz0 = _mm_macc_ps(dz00,fscal,fiz0);
544 fjptrA = f+j_coord_offsetA;
545 fjptrB = f+j_coord_offsetB;
546 fjptrC = f+j_coord_offsetC;
547 fjptrD = f+j_coord_offsetD;
548 gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,
549 _mm_mul_ps(dx00,fscal),
550 _mm_mul_ps(dy00,fscal),
551 _mm_mul_ps(dz00,fscal));
555 /* Inner loop uses 31 flops */
561 /* Get j neighbor index, and coordinate index */
562 jnrlistA = jjnr[jidx];
563 jnrlistB = jjnr[jidx+1];
564 jnrlistC = jjnr[jidx+2];
565 jnrlistD = jjnr[jidx+3];
566 /* Sign of each element will be negative for non-real atoms.
567 * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
568 * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
570 dummy_mask = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
571 jnrA = (jnrlistA>=0) ? jnrlistA : 0;
572 jnrB = (jnrlistB>=0) ? jnrlistB : 0;
573 jnrC = (jnrlistC>=0) ? jnrlistC : 0;
574 jnrD = (jnrlistD>=0) ? jnrlistD : 0;
575 j_coord_offsetA = DIM*jnrA;
576 j_coord_offsetB = DIM*jnrB;
577 j_coord_offsetC = DIM*jnrC;
578 j_coord_offsetD = DIM*jnrD;
580 /* load j atom coordinates */
581 gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
582 x+j_coord_offsetC,x+j_coord_offsetD,
585 /* Calculate displacement vector */
586 dx00 = _mm_sub_ps(ix0,jx0);
587 dy00 = _mm_sub_ps(iy0,jy0);
588 dz00 = _mm_sub_ps(iz0,jz0);
590 /* Calculate squared distance and things based on it */
591 rsq00 = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
593 rinv00 = gmx_mm_invsqrt_ps(rsq00);
595 rinvsq00 = _mm_mul_ps(rinv00,rinv00);
597 /* Load parameters for j particles */
598 jq0 = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
599 charge+jnrC+0,charge+jnrD+0);
601 /**************************
602 * CALCULATE INTERACTIONS *
603 **************************/
605 if (gmx_mm_any_lt(rsq00,rcutoff2))
608 r00 = _mm_mul_ps(rsq00,rinv00);
609 r00 = _mm_andnot_ps(dummy_mask,r00);
611 /* Compute parameters for interactions between i and j atoms */
612 qq00 = _mm_mul_ps(iq0,jq0);
614 /* EWALD ELECTROSTATICS */
616 /* Analytical PME correction */
617 zeta2 = _mm_mul_ps(beta2,rsq00);
618 rinv3 = _mm_mul_ps(rinvsq00,rinv00);
619 pmecorrF = gmx_mm_pmecorrF_ps(zeta2);
620 felec = _mm_macc_ps(pmecorrF,beta3,rinv3);
621 felec = _mm_mul_ps(qq00,felec);
623 cutoff_mask = _mm_cmplt_ps(rsq00,rcutoff2);
627 fscal = _mm_and_ps(fscal,cutoff_mask);
629 fscal = _mm_andnot_ps(dummy_mask,fscal);
631 /* Update vectorial force */
632 fix0 = _mm_macc_ps(dx00,fscal,fix0);
633 fiy0 = _mm_macc_ps(dy00,fscal,fiy0);
634 fiz0 = _mm_macc_ps(dz00,fscal,fiz0);
636 fjptrA = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
637 fjptrB = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
638 fjptrC = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
639 fjptrD = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
640 gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,
641 _mm_mul_ps(dx00,fscal),
642 _mm_mul_ps(dy00,fscal),
643 _mm_mul_ps(dz00,fscal));
647 /* Inner loop uses 32 flops */
650 /* End of innermost loop */
652 gmx_mm_update_iforce_1atom_swizzle_ps(fix0,fiy0,fiz0,
653 f+i_coord_offset,fshift+i_shift_offset);
655 /* Increment number of inner iterations */
656 inneriter += j_index_end - j_index_start;
658 /* Outer loop uses 7 flops */
661 /* Increment number of outer iterations */
664 /* Update outer/inner flops */
666 inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_F,outeriter*7 + inneriter*32);