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_ElecRF_VdwNone_GeomP1P1_VF_avx_256_single
38 * Electrostatics interaction: ReactionField
39 * VdW interaction: None
40 * Geometry: Particle-Particle
41 * Calculate force/pot: PotentialAndForce
44 nb_kernel_ElecRF_VdwNone_GeomP1P1_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 * vdwioffsetptr0;
73 __m256 ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
74 int vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D,vdwjidx0E,vdwjidx0F,vdwjidx0G,vdwjidx0H;
75 __m256 jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
76 __m256 dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
77 __m256 velec,felec,velecsum,facel,crf,krf,krf2;
79 __m256 dummy_mask,cutoff_mask;
80 __m256 signbit = _mm256_castsi256_ps( _mm256_set1_epi32(0x80000000) );
81 __m256 one = _mm256_set1_ps(1.0);
82 __m256 two = _mm256_set1_ps(2.0);
88 jindex = nlist->jindex;
90 shiftidx = nlist->shift;
92 shiftvec = fr->shift_vec[0];
93 fshift = fr->fshift[0];
94 facel = _mm256_set1_ps(fr->epsfac);
95 charge = mdatoms->chargeA;
96 krf = _mm256_set1_ps(fr->ic->k_rf);
97 krf2 = _mm256_set1_ps(fr->ic->k_rf*2.0);
98 crf = _mm256_set1_ps(fr->ic->c_rf);
100 /* Avoid stupid compiler warnings */
101 jnrA = jnrB = jnrC = jnrD = jnrE = jnrF = jnrG = jnrH = 0;
114 for(iidx=0;iidx<4*DIM;iidx++)
119 /* Start outer loop over neighborlists */
120 for(iidx=0; iidx<nri; iidx++)
122 /* Load shift vector for this list */
123 i_shift_offset = DIM*shiftidx[iidx];
125 /* Load limits for loop over neighbors */
126 j_index_start = jindex[iidx];
127 j_index_end = jindex[iidx+1];
129 /* Get outer coordinate index */
131 i_coord_offset = DIM*inr;
133 /* Load i particle coords and add shift vector */
134 gmx_mm256_load_shift_and_1rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,&ix0,&iy0,&iz0);
136 fix0 = _mm256_setzero_ps();
137 fiy0 = _mm256_setzero_ps();
138 fiz0 = _mm256_setzero_ps();
140 /* Load parameters for i particles */
141 iq0 = _mm256_mul_ps(facel,_mm256_set1_ps(charge[inr+0]));
143 /* Reset potential sums */
144 velecsum = _mm256_setzero_ps();
146 /* Start inner kernel loop */
147 for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+7]>=0; jidx+=8)
150 /* Get j neighbor index, and coordinate index */
159 j_coord_offsetA = DIM*jnrA;
160 j_coord_offsetB = DIM*jnrB;
161 j_coord_offsetC = DIM*jnrC;
162 j_coord_offsetD = DIM*jnrD;
163 j_coord_offsetE = DIM*jnrE;
164 j_coord_offsetF = DIM*jnrF;
165 j_coord_offsetG = DIM*jnrG;
166 j_coord_offsetH = DIM*jnrH;
168 /* load j atom coordinates */
169 gmx_mm256_load_1rvec_8ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
170 x+j_coord_offsetC,x+j_coord_offsetD,
171 x+j_coord_offsetE,x+j_coord_offsetF,
172 x+j_coord_offsetG,x+j_coord_offsetH,
175 /* Calculate displacement vector */
176 dx00 = _mm256_sub_ps(ix0,jx0);
177 dy00 = _mm256_sub_ps(iy0,jy0);
178 dz00 = _mm256_sub_ps(iz0,jz0);
180 /* Calculate squared distance and things based on it */
181 rsq00 = gmx_mm256_calc_rsq_ps(dx00,dy00,dz00);
183 rinv00 = gmx_mm256_invsqrt_ps(rsq00);
185 rinvsq00 = _mm256_mul_ps(rinv00,rinv00);
187 /* Load parameters for j particles */
188 jq0 = gmx_mm256_load_8real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
189 charge+jnrC+0,charge+jnrD+0,
190 charge+jnrE+0,charge+jnrF+0,
191 charge+jnrG+0,charge+jnrH+0);
193 /**************************
194 * CALCULATE INTERACTIONS *
195 **************************/
197 /* Compute parameters for interactions between i and j atoms */
198 qq00 = _mm256_mul_ps(iq0,jq0);
200 /* REACTION-FIELD ELECTROSTATICS */
201 velec = _mm256_mul_ps(qq00,_mm256_sub_ps(_mm256_add_ps(rinv00,_mm256_mul_ps(krf,rsq00)),crf));
202 felec = _mm256_mul_ps(qq00,_mm256_sub_ps(_mm256_mul_ps(rinv00,rinvsq00),krf2));
204 /* Update potential sum for this i atom from the interaction with this j atom. */
205 velecsum = _mm256_add_ps(velecsum,velec);
209 /* Calculate temporary vectorial force */
210 tx = _mm256_mul_ps(fscal,dx00);
211 ty = _mm256_mul_ps(fscal,dy00);
212 tz = _mm256_mul_ps(fscal,dz00);
214 /* Update vectorial force */
215 fix0 = _mm256_add_ps(fix0,tx);
216 fiy0 = _mm256_add_ps(fiy0,ty);
217 fiz0 = _mm256_add_ps(fiz0,tz);
219 fjptrA = f+j_coord_offsetA;
220 fjptrB = f+j_coord_offsetB;
221 fjptrC = f+j_coord_offsetC;
222 fjptrD = f+j_coord_offsetD;
223 fjptrE = f+j_coord_offsetE;
224 fjptrF = f+j_coord_offsetF;
225 fjptrG = f+j_coord_offsetG;
226 fjptrH = f+j_coord_offsetH;
227 gmx_mm256_decrement_1rvec_8ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjptrE,fjptrF,fjptrG,fjptrH,tx,ty,tz);
229 /* Inner loop uses 32 flops */
235 /* Get j neighbor index, and coordinate index */
236 jnrlistA = jjnr[jidx];
237 jnrlistB = jjnr[jidx+1];
238 jnrlistC = jjnr[jidx+2];
239 jnrlistD = jjnr[jidx+3];
240 jnrlistE = jjnr[jidx+4];
241 jnrlistF = jjnr[jidx+5];
242 jnrlistG = jjnr[jidx+6];
243 jnrlistH = jjnr[jidx+7];
244 /* Sign of each element will be negative for non-real atoms.
245 * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
246 * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
248 dummy_mask = gmx_mm256_set_m128(gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx+4)),_mm_setzero_si128())),
249 gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128())));
251 jnrA = (jnrlistA>=0) ? jnrlistA : 0;
252 jnrB = (jnrlistB>=0) ? jnrlistB : 0;
253 jnrC = (jnrlistC>=0) ? jnrlistC : 0;
254 jnrD = (jnrlistD>=0) ? jnrlistD : 0;
255 jnrE = (jnrlistE>=0) ? jnrlistE : 0;
256 jnrF = (jnrlistF>=0) ? jnrlistF : 0;
257 jnrG = (jnrlistG>=0) ? jnrlistG : 0;
258 jnrH = (jnrlistH>=0) ? jnrlistH : 0;
259 j_coord_offsetA = DIM*jnrA;
260 j_coord_offsetB = DIM*jnrB;
261 j_coord_offsetC = DIM*jnrC;
262 j_coord_offsetD = DIM*jnrD;
263 j_coord_offsetE = DIM*jnrE;
264 j_coord_offsetF = DIM*jnrF;
265 j_coord_offsetG = DIM*jnrG;
266 j_coord_offsetH = DIM*jnrH;
268 /* load j atom coordinates */
269 gmx_mm256_load_1rvec_8ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
270 x+j_coord_offsetC,x+j_coord_offsetD,
271 x+j_coord_offsetE,x+j_coord_offsetF,
272 x+j_coord_offsetG,x+j_coord_offsetH,
275 /* Calculate displacement vector */
276 dx00 = _mm256_sub_ps(ix0,jx0);
277 dy00 = _mm256_sub_ps(iy0,jy0);
278 dz00 = _mm256_sub_ps(iz0,jz0);
280 /* Calculate squared distance and things based on it */
281 rsq00 = gmx_mm256_calc_rsq_ps(dx00,dy00,dz00);
283 rinv00 = gmx_mm256_invsqrt_ps(rsq00);
285 rinvsq00 = _mm256_mul_ps(rinv00,rinv00);
287 /* Load parameters for j particles */
288 jq0 = gmx_mm256_load_8real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
289 charge+jnrC+0,charge+jnrD+0,
290 charge+jnrE+0,charge+jnrF+0,
291 charge+jnrG+0,charge+jnrH+0);
293 /**************************
294 * CALCULATE INTERACTIONS *
295 **************************/
297 /* Compute parameters for interactions between i and j atoms */
298 qq00 = _mm256_mul_ps(iq0,jq0);
300 /* REACTION-FIELD ELECTROSTATICS */
301 velec = _mm256_mul_ps(qq00,_mm256_sub_ps(_mm256_add_ps(rinv00,_mm256_mul_ps(krf,rsq00)),crf));
302 felec = _mm256_mul_ps(qq00,_mm256_sub_ps(_mm256_mul_ps(rinv00,rinvsq00),krf2));
304 /* Update potential sum for this i atom from the interaction with this j atom. */
305 velec = _mm256_andnot_ps(dummy_mask,velec);
306 velecsum = _mm256_add_ps(velecsum,velec);
310 fscal = _mm256_andnot_ps(dummy_mask,fscal);
312 /* Calculate temporary vectorial force */
313 tx = _mm256_mul_ps(fscal,dx00);
314 ty = _mm256_mul_ps(fscal,dy00);
315 tz = _mm256_mul_ps(fscal,dz00);
317 /* Update vectorial force */
318 fix0 = _mm256_add_ps(fix0,tx);
319 fiy0 = _mm256_add_ps(fiy0,ty);
320 fiz0 = _mm256_add_ps(fiz0,tz);
322 fjptrA = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
323 fjptrB = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
324 fjptrC = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
325 fjptrD = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
326 fjptrE = (jnrlistE>=0) ? f+j_coord_offsetE : scratch;
327 fjptrF = (jnrlistF>=0) ? f+j_coord_offsetF : scratch;
328 fjptrG = (jnrlistG>=0) ? f+j_coord_offsetG : scratch;
329 fjptrH = (jnrlistH>=0) ? f+j_coord_offsetH : scratch;
330 gmx_mm256_decrement_1rvec_8ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjptrE,fjptrF,fjptrG,fjptrH,tx,ty,tz);
332 /* Inner loop uses 32 flops */
335 /* End of innermost loop */
337 gmx_mm256_update_iforce_1atom_swizzle_ps(fix0,fiy0,fiz0,
338 f+i_coord_offset,fshift+i_shift_offset);
341 /* Update potential energies */
342 gmx_mm256_update_1pot_ps(velecsum,kernel_data->energygrp_elec+ggid);
344 /* Increment number of inner iterations */
345 inneriter += j_index_end - j_index_start;
347 /* Outer loop uses 8 flops */
350 /* Increment number of outer iterations */
353 /* Update outer/inner flops */
355 inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VF,outeriter*8 + inneriter*32);
358 * Gromacs nonbonded kernel: nb_kernel_ElecRF_VdwNone_GeomP1P1_F_avx_256_single
359 * Electrostatics interaction: ReactionField
360 * VdW interaction: None
361 * Geometry: Particle-Particle
362 * Calculate force/pot: Force
365 nb_kernel_ElecRF_VdwNone_GeomP1P1_F_avx_256_single
366 (t_nblist * gmx_restrict nlist,
367 rvec * gmx_restrict xx,
368 rvec * gmx_restrict ff,
369 t_forcerec * gmx_restrict fr,
370 t_mdatoms * gmx_restrict mdatoms,
371 nb_kernel_data_t * gmx_restrict kernel_data,
372 t_nrnb * gmx_restrict nrnb)
374 /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
375 * just 0 for non-waters.
376 * Suffixes A,B,C,D,E,F,G,H refer to j loop unrolling done with AVX, e.g. for the eight different
377 * jnr indices corresponding to data put in the four positions in the SIMD register.
379 int i_shift_offset,i_coord_offset,outeriter,inneriter;
380 int j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
381 int jnrA,jnrB,jnrC,jnrD;
382 int jnrE,jnrF,jnrG,jnrH;
383 int jnrlistA,jnrlistB,jnrlistC,jnrlistD;
384 int jnrlistE,jnrlistF,jnrlistG,jnrlistH;
385 int j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
386 int j_coord_offsetE,j_coord_offsetF,j_coord_offsetG,j_coord_offsetH;
387 int *iinr,*jindex,*jjnr,*shiftidx,*gid;
389 real *shiftvec,*fshift,*x,*f;
390 real *fjptrA,*fjptrB,*fjptrC,*fjptrD,*fjptrE,*fjptrF,*fjptrG,*fjptrH;
392 __m256 tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
393 real * vdwioffsetptr0;
394 __m256 ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
395 int vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D,vdwjidx0E,vdwjidx0F,vdwjidx0G,vdwjidx0H;
396 __m256 jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
397 __m256 dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
398 __m256 velec,felec,velecsum,facel,crf,krf,krf2;
400 __m256 dummy_mask,cutoff_mask;
401 __m256 signbit = _mm256_castsi256_ps( _mm256_set1_epi32(0x80000000) );
402 __m256 one = _mm256_set1_ps(1.0);
403 __m256 two = _mm256_set1_ps(2.0);
409 jindex = nlist->jindex;
411 shiftidx = nlist->shift;
413 shiftvec = fr->shift_vec[0];
414 fshift = fr->fshift[0];
415 facel = _mm256_set1_ps(fr->epsfac);
416 charge = mdatoms->chargeA;
417 krf = _mm256_set1_ps(fr->ic->k_rf);
418 krf2 = _mm256_set1_ps(fr->ic->k_rf*2.0);
419 crf = _mm256_set1_ps(fr->ic->c_rf);
421 /* Avoid stupid compiler warnings */
422 jnrA = jnrB = jnrC = jnrD = jnrE = jnrF = jnrG = jnrH = 0;
435 for(iidx=0;iidx<4*DIM;iidx++)
440 /* Start outer loop over neighborlists */
441 for(iidx=0; iidx<nri; iidx++)
443 /* Load shift vector for this list */
444 i_shift_offset = DIM*shiftidx[iidx];
446 /* Load limits for loop over neighbors */
447 j_index_start = jindex[iidx];
448 j_index_end = jindex[iidx+1];
450 /* Get outer coordinate index */
452 i_coord_offset = DIM*inr;
454 /* Load i particle coords and add shift vector */
455 gmx_mm256_load_shift_and_1rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,&ix0,&iy0,&iz0);
457 fix0 = _mm256_setzero_ps();
458 fiy0 = _mm256_setzero_ps();
459 fiz0 = _mm256_setzero_ps();
461 /* Load parameters for i particles */
462 iq0 = _mm256_mul_ps(facel,_mm256_set1_ps(charge[inr+0]));
464 /* Start inner kernel loop */
465 for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+7]>=0; jidx+=8)
468 /* Get j neighbor index, and coordinate index */
477 j_coord_offsetA = DIM*jnrA;
478 j_coord_offsetB = DIM*jnrB;
479 j_coord_offsetC = DIM*jnrC;
480 j_coord_offsetD = DIM*jnrD;
481 j_coord_offsetE = DIM*jnrE;
482 j_coord_offsetF = DIM*jnrF;
483 j_coord_offsetG = DIM*jnrG;
484 j_coord_offsetH = DIM*jnrH;
486 /* load j atom coordinates */
487 gmx_mm256_load_1rvec_8ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
488 x+j_coord_offsetC,x+j_coord_offsetD,
489 x+j_coord_offsetE,x+j_coord_offsetF,
490 x+j_coord_offsetG,x+j_coord_offsetH,
493 /* Calculate displacement vector */
494 dx00 = _mm256_sub_ps(ix0,jx0);
495 dy00 = _mm256_sub_ps(iy0,jy0);
496 dz00 = _mm256_sub_ps(iz0,jz0);
498 /* Calculate squared distance and things based on it */
499 rsq00 = gmx_mm256_calc_rsq_ps(dx00,dy00,dz00);
501 rinv00 = gmx_mm256_invsqrt_ps(rsq00);
503 rinvsq00 = _mm256_mul_ps(rinv00,rinv00);
505 /* Load parameters for j particles */
506 jq0 = gmx_mm256_load_8real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
507 charge+jnrC+0,charge+jnrD+0,
508 charge+jnrE+0,charge+jnrF+0,
509 charge+jnrG+0,charge+jnrH+0);
511 /**************************
512 * CALCULATE INTERACTIONS *
513 **************************/
515 /* Compute parameters for interactions between i and j atoms */
516 qq00 = _mm256_mul_ps(iq0,jq0);
518 /* REACTION-FIELD ELECTROSTATICS */
519 felec = _mm256_mul_ps(qq00,_mm256_sub_ps(_mm256_mul_ps(rinv00,rinvsq00),krf2));
523 /* Calculate temporary vectorial force */
524 tx = _mm256_mul_ps(fscal,dx00);
525 ty = _mm256_mul_ps(fscal,dy00);
526 tz = _mm256_mul_ps(fscal,dz00);
528 /* Update vectorial force */
529 fix0 = _mm256_add_ps(fix0,tx);
530 fiy0 = _mm256_add_ps(fiy0,ty);
531 fiz0 = _mm256_add_ps(fiz0,tz);
533 fjptrA = f+j_coord_offsetA;
534 fjptrB = f+j_coord_offsetB;
535 fjptrC = f+j_coord_offsetC;
536 fjptrD = f+j_coord_offsetD;
537 fjptrE = f+j_coord_offsetE;
538 fjptrF = f+j_coord_offsetF;
539 fjptrG = f+j_coord_offsetG;
540 fjptrH = f+j_coord_offsetH;
541 gmx_mm256_decrement_1rvec_8ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjptrE,fjptrF,fjptrG,fjptrH,tx,ty,tz);
543 /* Inner loop uses 27 flops */
549 /* Get j neighbor index, and coordinate index */
550 jnrlistA = jjnr[jidx];
551 jnrlistB = jjnr[jidx+1];
552 jnrlistC = jjnr[jidx+2];
553 jnrlistD = jjnr[jidx+3];
554 jnrlistE = jjnr[jidx+4];
555 jnrlistF = jjnr[jidx+5];
556 jnrlistG = jjnr[jidx+6];
557 jnrlistH = jjnr[jidx+7];
558 /* Sign of each element will be negative for non-real atoms.
559 * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
560 * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
562 dummy_mask = gmx_mm256_set_m128(gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx+4)),_mm_setzero_si128())),
563 gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128())));
565 jnrA = (jnrlistA>=0) ? jnrlistA : 0;
566 jnrB = (jnrlistB>=0) ? jnrlistB : 0;
567 jnrC = (jnrlistC>=0) ? jnrlistC : 0;
568 jnrD = (jnrlistD>=0) ? jnrlistD : 0;
569 jnrE = (jnrlistE>=0) ? jnrlistE : 0;
570 jnrF = (jnrlistF>=0) ? jnrlistF : 0;
571 jnrG = (jnrlistG>=0) ? jnrlistG : 0;
572 jnrH = (jnrlistH>=0) ? jnrlistH : 0;
573 j_coord_offsetA = DIM*jnrA;
574 j_coord_offsetB = DIM*jnrB;
575 j_coord_offsetC = DIM*jnrC;
576 j_coord_offsetD = DIM*jnrD;
577 j_coord_offsetE = DIM*jnrE;
578 j_coord_offsetF = DIM*jnrF;
579 j_coord_offsetG = DIM*jnrG;
580 j_coord_offsetH = DIM*jnrH;
582 /* load j atom coordinates */
583 gmx_mm256_load_1rvec_8ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
584 x+j_coord_offsetC,x+j_coord_offsetD,
585 x+j_coord_offsetE,x+j_coord_offsetF,
586 x+j_coord_offsetG,x+j_coord_offsetH,
589 /* Calculate displacement vector */
590 dx00 = _mm256_sub_ps(ix0,jx0);
591 dy00 = _mm256_sub_ps(iy0,jy0);
592 dz00 = _mm256_sub_ps(iz0,jz0);
594 /* Calculate squared distance and things based on it */
595 rsq00 = gmx_mm256_calc_rsq_ps(dx00,dy00,dz00);
597 rinv00 = gmx_mm256_invsqrt_ps(rsq00);
599 rinvsq00 = _mm256_mul_ps(rinv00,rinv00);
601 /* Load parameters for j particles */
602 jq0 = gmx_mm256_load_8real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
603 charge+jnrC+0,charge+jnrD+0,
604 charge+jnrE+0,charge+jnrF+0,
605 charge+jnrG+0,charge+jnrH+0);
607 /**************************
608 * CALCULATE INTERACTIONS *
609 **************************/
611 /* Compute parameters for interactions between i and j atoms */
612 qq00 = _mm256_mul_ps(iq0,jq0);
614 /* REACTION-FIELD ELECTROSTATICS */
615 felec = _mm256_mul_ps(qq00,_mm256_sub_ps(_mm256_mul_ps(rinv00,rinvsq00),krf2));
619 fscal = _mm256_andnot_ps(dummy_mask,fscal);
621 /* Calculate temporary vectorial force */
622 tx = _mm256_mul_ps(fscal,dx00);
623 ty = _mm256_mul_ps(fscal,dy00);
624 tz = _mm256_mul_ps(fscal,dz00);
626 /* Update vectorial force */
627 fix0 = _mm256_add_ps(fix0,tx);
628 fiy0 = _mm256_add_ps(fiy0,ty);
629 fiz0 = _mm256_add_ps(fiz0,tz);
631 fjptrA = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
632 fjptrB = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
633 fjptrC = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
634 fjptrD = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
635 fjptrE = (jnrlistE>=0) ? f+j_coord_offsetE : scratch;
636 fjptrF = (jnrlistF>=0) ? f+j_coord_offsetF : scratch;
637 fjptrG = (jnrlistG>=0) ? f+j_coord_offsetG : scratch;
638 fjptrH = (jnrlistH>=0) ? f+j_coord_offsetH : scratch;
639 gmx_mm256_decrement_1rvec_8ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjptrE,fjptrF,fjptrG,fjptrH,tx,ty,tz);
641 /* Inner loop uses 27 flops */
644 /* End of innermost loop */
646 gmx_mm256_update_iforce_1atom_swizzle_ps(fix0,fiy0,fiz0,
647 f+i_coord_offset,fshift+i_shift_offset);
649 /* Increment number of inner iterations */
650 inneriter += j_index_end - j_index_start;
652 /* Outer loop uses 7 flops */
655 /* Increment number of outer iterations */
658 /* Update outer/inner flops */
660 inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_F,outeriter*7 + inneriter*27);