2 * This file is part of the GROMACS molecular simulation package.
4 * Copyright (c) 2012,2013,2014, by the GROMACS development team, led by
5 * Mark Abraham, David van der Spoel, Berk Hess, and Erik Lindahl,
6 * and including many others, as listed in the AUTHORS file in the
7 * top-level source directory and at http://www.gromacs.org.
9 * GROMACS is free software; you can redistribute it and/or
10 * modify it under the terms of the GNU Lesser General Public License
11 * as published by the Free Software Foundation; either version 2.1
12 * of the License, or (at your option) any later version.
14 * GROMACS is distributed in the hope that it will be useful,
15 * but WITHOUT ANY WARRANTY; without even the implied warranty of
16 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
17 * Lesser General Public License for more details.
19 * You should have received a copy of the GNU Lesser General Public
20 * License along with GROMACS; if not, see
21 * http://www.gnu.org/licenses, or write to the Free Software Foundation,
22 * Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA.
24 * If you want to redistribute modifications to GROMACS, please
25 * consider that scientific software is very special. Version
26 * control is crucial - bugs must be traceable. We will be happy to
27 * consider code for inclusion in the official distribution, but
28 * derived work must not be called official GROMACS. Details are found
29 * in the README & COPYING files - if they are missing, get the
30 * official version at http://www.gromacs.org.
32 * To help us fund GROMACS development, we humbly ask that you cite
33 * the research papers on the package. Check out http://www.gromacs.org.
36 * Note: this file was generated by the GROMACS avx_256_single kernel generator.
42 #include "../nb_kernel.h"
43 #include "types/simple.h"
44 #include "gromacs/math/vec.h"
47 #include "gromacs/simd/math_x86_avx_256_single.h"
48 #include "kernelutil_x86_avx_256_single.h"
51 * Gromacs nonbonded kernel: nb_kernel_ElecRF_VdwNone_GeomP1P1_VF_avx_256_single
52 * Electrostatics interaction: ReactionField
53 * VdW interaction: None
54 * Geometry: Particle-Particle
55 * Calculate force/pot: PotentialAndForce
58 nb_kernel_ElecRF_VdwNone_GeomP1P1_VF_avx_256_single
59 (t_nblist * gmx_restrict nlist,
60 rvec * gmx_restrict xx,
61 rvec * gmx_restrict ff,
62 t_forcerec * gmx_restrict fr,
63 t_mdatoms * gmx_restrict mdatoms,
64 nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
65 t_nrnb * gmx_restrict nrnb)
67 /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
68 * just 0 for non-waters.
69 * Suffixes A,B,C,D,E,F,G,H refer to j loop unrolling done with AVX, e.g. for the eight different
70 * jnr indices corresponding to data put in the four positions in the SIMD register.
72 int i_shift_offset,i_coord_offset,outeriter,inneriter;
73 int j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
74 int jnrA,jnrB,jnrC,jnrD;
75 int jnrE,jnrF,jnrG,jnrH;
76 int jnrlistA,jnrlistB,jnrlistC,jnrlistD;
77 int jnrlistE,jnrlistF,jnrlistG,jnrlistH;
78 int j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
79 int j_coord_offsetE,j_coord_offsetF,j_coord_offsetG,j_coord_offsetH;
80 int *iinr,*jindex,*jjnr,*shiftidx,*gid;
82 real *shiftvec,*fshift,*x,*f;
83 real *fjptrA,*fjptrB,*fjptrC,*fjptrD,*fjptrE,*fjptrF,*fjptrG,*fjptrH;
85 __m256 tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
86 real * vdwioffsetptr0;
87 __m256 ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
88 int vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D,vdwjidx0E,vdwjidx0F,vdwjidx0G,vdwjidx0H;
89 __m256 jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
90 __m256 dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
91 __m256 velec,felec,velecsum,facel,crf,krf,krf2;
93 __m256 dummy_mask,cutoff_mask;
94 __m256 signbit = _mm256_castsi256_ps( _mm256_set1_epi32(0x80000000) );
95 __m256 one = _mm256_set1_ps(1.0);
96 __m256 two = _mm256_set1_ps(2.0);
102 jindex = nlist->jindex;
104 shiftidx = nlist->shift;
106 shiftvec = fr->shift_vec[0];
107 fshift = fr->fshift[0];
108 facel = _mm256_set1_ps(fr->epsfac);
109 charge = mdatoms->chargeA;
110 krf = _mm256_set1_ps(fr->ic->k_rf);
111 krf2 = _mm256_set1_ps(fr->ic->k_rf*2.0);
112 crf = _mm256_set1_ps(fr->ic->c_rf);
114 /* Avoid stupid compiler warnings */
115 jnrA = jnrB = jnrC = jnrD = jnrE = jnrF = jnrG = jnrH = 0;
128 for(iidx=0;iidx<4*DIM;iidx++)
133 /* Start outer loop over neighborlists */
134 for(iidx=0; iidx<nri; iidx++)
136 /* Load shift vector for this list */
137 i_shift_offset = DIM*shiftidx[iidx];
139 /* Load limits for loop over neighbors */
140 j_index_start = jindex[iidx];
141 j_index_end = jindex[iidx+1];
143 /* Get outer coordinate index */
145 i_coord_offset = DIM*inr;
147 /* Load i particle coords and add shift vector */
148 gmx_mm256_load_shift_and_1rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,&ix0,&iy0,&iz0);
150 fix0 = _mm256_setzero_ps();
151 fiy0 = _mm256_setzero_ps();
152 fiz0 = _mm256_setzero_ps();
154 /* Load parameters for i particles */
155 iq0 = _mm256_mul_ps(facel,_mm256_set1_ps(charge[inr+0]));
157 /* Reset potential sums */
158 velecsum = _mm256_setzero_ps();
160 /* Start inner kernel loop */
161 for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+7]>=0; jidx+=8)
164 /* Get j neighbor index, and coordinate index */
173 j_coord_offsetA = DIM*jnrA;
174 j_coord_offsetB = DIM*jnrB;
175 j_coord_offsetC = DIM*jnrC;
176 j_coord_offsetD = DIM*jnrD;
177 j_coord_offsetE = DIM*jnrE;
178 j_coord_offsetF = DIM*jnrF;
179 j_coord_offsetG = DIM*jnrG;
180 j_coord_offsetH = DIM*jnrH;
182 /* load j atom coordinates */
183 gmx_mm256_load_1rvec_8ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
184 x+j_coord_offsetC,x+j_coord_offsetD,
185 x+j_coord_offsetE,x+j_coord_offsetF,
186 x+j_coord_offsetG,x+j_coord_offsetH,
189 /* Calculate displacement vector */
190 dx00 = _mm256_sub_ps(ix0,jx0);
191 dy00 = _mm256_sub_ps(iy0,jy0);
192 dz00 = _mm256_sub_ps(iz0,jz0);
194 /* Calculate squared distance and things based on it */
195 rsq00 = gmx_mm256_calc_rsq_ps(dx00,dy00,dz00);
197 rinv00 = gmx_mm256_invsqrt_ps(rsq00);
199 rinvsq00 = _mm256_mul_ps(rinv00,rinv00);
201 /* Load parameters for j particles */
202 jq0 = gmx_mm256_load_8real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
203 charge+jnrC+0,charge+jnrD+0,
204 charge+jnrE+0,charge+jnrF+0,
205 charge+jnrG+0,charge+jnrH+0);
207 /**************************
208 * CALCULATE INTERACTIONS *
209 **************************/
211 /* Compute parameters for interactions between i and j atoms */
212 qq00 = _mm256_mul_ps(iq0,jq0);
214 /* REACTION-FIELD ELECTROSTATICS */
215 velec = _mm256_mul_ps(qq00,_mm256_sub_ps(_mm256_add_ps(rinv00,_mm256_mul_ps(krf,rsq00)),crf));
216 felec = _mm256_mul_ps(qq00,_mm256_sub_ps(_mm256_mul_ps(rinv00,rinvsq00),krf2));
218 /* Update potential sum for this i atom from the interaction with this j atom. */
219 velecsum = _mm256_add_ps(velecsum,velec);
223 /* Calculate temporary vectorial force */
224 tx = _mm256_mul_ps(fscal,dx00);
225 ty = _mm256_mul_ps(fscal,dy00);
226 tz = _mm256_mul_ps(fscal,dz00);
228 /* Update vectorial force */
229 fix0 = _mm256_add_ps(fix0,tx);
230 fiy0 = _mm256_add_ps(fiy0,ty);
231 fiz0 = _mm256_add_ps(fiz0,tz);
233 fjptrA = f+j_coord_offsetA;
234 fjptrB = f+j_coord_offsetB;
235 fjptrC = f+j_coord_offsetC;
236 fjptrD = f+j_coord_offsetD;
237 fjptrE = f+j_coord_offsetE;
238 fjptrF = f+j_coord_offsetF;
239 fjptrG = f+j_coord_offsetG;
240 fjptrH = f+j_coord_offsetH;
241 gmx_mm256_decrement_1rvec_8ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjptrE,fjptrF,fjptrG,fjptrH,tx,ty,tz);
243 /* Inner loop uses 32 flops */
249 /* Get j neighbor index, and coordinate index */
250 jnrlistA = jjnr[jidx];
251 jnrlistB = jjnr[jidx+1];
252 jnrlistC = jjnr[jidx+2];
253 jnrlistD = jjnr[jidx+3];
254 jnrlistE = jjnr[jidx+4];
255 jnrlistF = jjnr[jidx+5];
256 jnrlistG = jjnr[jidx+6];
257 jnrlistH = jjnr[jidx+7];
258 /* Sign of each element will be negative for non-real atoms.
259 * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
260 * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
262 dummy_mask = gmx_mm256_set_m128(gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx+4)),_mm_setzero_si128())),
263 gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128())));
265 jnrA = (jnrlistA>=0) ? jnrlistA : 0;
266 jnrB = (jnrlistB>=0) ? jnrlistB : 0;
267 jnrC = (jnrlistC>=0) ? jnrlistC : 0;
268 jnrD = (jnrlistD>=0) ? jnrlistD : 0;
269 jnrE = (jnrlistE>=0) ? jnrlistE : 0;
270 jnrF = (jnrlistF>=0) ? jnrlistF : 0;
271 jnrG = (jnrlistG>=0) ? jnrlistG : 0;
272 jnrH = (jnrlistH>=0) ? jnrlistH : 0;
273 j_coord_offsetA = DIM*jnrA;
274 j_coord_offsetB = DIM*jnrB;
275 j_coord_offsetC = DIM*jnrC;
276 j_coord_offsetD = DIM*jnrD;
277 j_coord_offsetE = DIM*jnrE;
278 j_coord_offsetF = DIM*jnrF;
279 j_coord_offsetG = DIM*jnrG;
280 j_coord_offsetH = DIM*jnrH;
282 /* load j atom coordinates */
283 gmx_mm256_load_1rvec_8ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
284 x+j_coord_offsetC,x+j_coord_offsetD,
285 x+j_coord_offsetE,x+j_coord_offsetF,
286 x+j_coord_offsetG,x+j_coord_offsetH,
289 /* Calculate displacement vector */
290 dx00 = _mm256_sub_ps(ix0,jx0);
291 dy00 = _mm256_sub_ps(iy0,jy0);
292 dz00 = _mm256_sub_ps(iz0,jz0);
294 /* Calculate squared distance and things based on it */
295 rsq00 = gmx_mm256_calc_rsq_ps(dx00,dy00,dz00);
297 rinv00 = gmx_mm256_invsqrt_ps(rsq00);
299 rinvsq00 = _mm256_mul_ps(rinv00,rinv00);
301 /* Load parameters for j particles */
302 jq0 = gmx_mm256_load_8real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
303 charge+jnrC+0,charge+jnrD+0,
304 charge+jnrE+0,charge+jnrF+0,
305 charge+jnrG+0,charge+jnrH+0);
307 /**************************
308 * CALCULATE INTERACTIONS *
309 **************************/
311 /* Compute parameters for interactions between i and j atoms */
312 qq00 = _mm256_mul_ps(iq0,jq0);
314 /* REACTION-FIELD ELECTROSTATICS */
315 velec = _mm256_mul_ps(qq00,_mm256_sub_ps(_mm256_add_ps(rinv00,_mm256_mul_ps(krf,rsq00)),crf));
316 felec = _mm256_mul_ps(qq00,_mm256_sub_ps(_mm256_mul_ps(rinv00,rinvsq00),krf2));
318 /* Update potential sum for this i atom from the interaction with this j atom. */
319 velec = _mm256_andnot_ps(dummy_mask,velec);
320 velecsum = _mm256_add_ps(velecsum,velec);
324 fscal = _mm256_andnot_ps(dummy_mask,fscal);
326 /* Calculate temporary vectorial force */
327 tx = _mm256_mul_ps(fscal,dx00);
328 ty = _mm256_mul_ps(fscal,dy00);
329 tz = _mm256_mul_ps(fscal,dz00);
331 /* Update vectorial force */
332 fix0 = _mm256_add_ps(fix0,tx);
333 fiy0 = _mm256_add_ps(fiy0,ty);
334 fiz0 = _mm256_add_ps(fiz0,tz);
336 fjptrA = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
337 fjptrB = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
338 fjptrC = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
339 fjptrD = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
340 fjptrE = (jnrlistE>=0) ? f+j_coord_offsetE : scratch;
341 fjptrF = (jnrlistF>=0) ? f+j_coord_offsetF : scratch;
342 fjptrG = (jnrlistG>=0) ? f+j_coord_offsetG : scratch;
343 fjptrH = (jnrlistH>=0) ? f+j_coord_offsetH : scratch;
344 gmx_mm256_decrement_1rvec_8ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjptrE,fjptrF,fjptrG,fjptrH,tx,ty,tz);
346 /* Inner loop uses 32 flops */
349 /* End of innermost loop */
351 gmx_mm256_update_iforce_1atom_swizzle_ps(fix0,fiy0,fiz0,
352 f+i_coord_offset,fshift+i_shift_offset);
355 /* Update potential energies */
356 gmx_mm256_update_1pot_ps(velecsum,kernel_data->energygrp_elec+ggid);
358 /* Increment number of inner iterations */
359 inneriter += j_index_end - j_index_start;
361 /* Outer loop uses 8 flops */
364 /* Increment number of outer iterations */
367 /* Update outer/inner flops */
369 inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VF,outeriter*8 + inneriter*32);
372 * Gromacs nonbonded kernel: nb_kernel_ElecRF_VdwNone_GeomP1P1_F_avx_256_single
373 * Electrostatics interaction: ReactionField
374 * VdW interaction: None
375 * Geometry: Particle-Particle
376 * Calculate force/pot: Force
379 nb_kernel_ElecRF_VdwNone_GeomP1P1_F_avx_256_single
380 (t_nblist * gmx_restrict nlist,
381 rvec * gmx_restrict xx,
382 rvec * gmx_restrict ff,
383 t_forcerec * gmx_restrict fr,
384 t_mdatoms * gmx_restrict mdatoms,
385 nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
386 t_nrnb * gmx_restrict nrnb)
388 /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
389 * just 0 for non-waters.
390 * Suffixes A,B,C,D,E,F,G,H refer to j loop unrolling done with AVX, e.g. for the eight different
391 * jnr indices corresponding to data put in the four positions in the SIMD register.
393 int i_shift_offset,i_coord_offset,outeriter,inneriter;
394 int j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
395 int jnrA,jnrB,jnrC,jnrD;
396 int jnrE,jnrF,jnrG,jnrH;
397 int jnrlistA,jnrlistB,jnrlistC,jnrlistD;
398 int jnrlistE,jnrlistF,jnrlistG,jnrlistH;
399 int j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
400 int j_coord_offsetE,j_coord_offsetF,j_coord_offsetG,j_coord_offsetH;
401 int *iinr,*jindex,*jjnr,*shiftidx,*gid;
403 real *shiftvec,*fshift,*x,*f;
404 real *fjptrA,*fjptrB,*fjptrC,*fjptrD,*fjptrE,*fjptrF,*fjptrG,*fjptrH;
406 __m256 tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
407 real * vdwioffsetptr0;
408 __m256 ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
409 int vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D,vdwjidx0E,vdwjidx0F,vdwjidx0G,vdwjidx0H;
410 __m256 jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
411 __m256 dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
412 __m256 velec,felec,velecsum,facel,crf,krf,krf2;
414 __m256 dummy_mask,cutoff_mask;
415 __m256 signbit = _mm256_castsi256_ps( _mm256_set1_epi32(0x80000000) );
416 __m256 one = _mm256_set1_ps(1.0);
417 __m256 two = _mm256_set1_ps(2.0);
423 jindex = nlist->jindex;
425 shiftidx = nlist->shift;
427 shiftvec = fr->shift_vec[0];
428 fshift = fr->fshift[0];
429 facel = _mm256_set1_ps(fr->epsfac);
430 charge = mdatoms->chargeA;
431 krf = _mm256_set1_ps(fr->ic->k_rf);
432 krf2 = _mm256_set1_ps(fr->ic->k_rf*2.0);
433 crf = _mm256_set1_ps(fr->ic->c_rf);
435 /* Avoid stupid compiler warnings */
436 jnrA = jnrB = jnrC = jnrD = jnrE = jnrF = jnrG = jnrH = 0;
449 for(iidx=0;iidx<4*DIM;iidx++)
454 /* Start outer loop over neighborlists */
455 for(iidx=0; iidx<nri; iidx++)
457 /* Load shift vector for this list */
458 i_shift_offset = DIM*shiftidx[iidx];
460 /* Load limits for loop over neighbors */
461 j_index_start = jindex[iidx];
462 j_index_end = jindex[iidx+1];
464 /* Get outer coordinate index */
466 i_coord_offset = DIM*inr;
468 /* Load i particle coords and add shift vector */
469 gmx_mm256_load_shift_and_1rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,&ix0,&iy0,&iz0);
471 fix0 = _mm256_setzero_ps();
472 fiy0 = _mm256_setzero_ps();
473 fiz0 = _mm256_setzero_ps();
475 /* Load parameters for i particles */
476 iq0 = _mm256_mul_ps(facel,_mm256_set1_ps(charge[inr+0]));
478 /* Start inner kernel loop */
479 for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+7]>=0; jidx+=8)
482 /* Get j neighbor index, and coordinate index */
491 j_coord_offsetA = DIM*jnrA;
492 j_coord_offsetB = DIM*jnrB;
493 j_coord_offsetC = DIM*jnrC;
494 j_coord_offsetD = DIM*jnrD;
495 j_coord_offsetE = DIM*jnrE;
496 j_coord_offsetF = DIM*jnrF;
497 j_coord_offsetG = DIM*jnrG;
498 j_coord_offsetH = DIM*jnrH;
500 /* load j atom coordinates */
501 gmx_mm256_load_1rvec_8ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
502 x+j_coord_offsetC,x+j_coord_offsetD,
503 x+j_coord_offsetE,x+j_coord_offsetF,
504 x+j_coord_offsetG,x+j_coord_offsetH,
507 /* Calculate displacement vector */
508 dx00 = _mm256_sub_ps(ix0,jx0);
509 dy00 = _mm256_sub_ps(iy0,jy0);
510 dz00 = _mm256_sub_ps(iz0,jz0);
512 /* Calculate squared distance and things based on it */
513 rsq00 = gmx_mm256_calc_rsq_ps(dx00,dy00,dz00);
515 rinv00 = gmx_mm256_invsqrt_ps(rsq00);
517 rinvsq00 = _mm256_mul_ps(rinv00,rinv00);
519 /* Load parameters for j particles */
520 jq0 = gmx_mm256_load_8real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
521 charge+jnrC+0,charge+jnrD+0,
522 charge+jnrE+0,charge+jnrF+0,
523 charge+jnrG+0,charge+jnrH+0);
525 /**************************
526 * CALCULATE INTERACTIONS *
527 **************************/
529 /* Compute parameters for interactions between i and j atoms */
530 qq00 = _mm256_mul_ps(iq0,jq0);
532 /* REACTION-FIELD ELECTROSTATICS */
533 felec = _mm256_mul_ps(qq00,_mm256_sub_ps(_mm256_mul_ps(rinv00,rinvsq00),krf2));
537 /* Calculate temporary vectorial force */
538 tx = _mm256_mul_ps(fscal,dx00);
539 ty = _mm256_mul_ps(fscal,dy00);
540 tz = _mm256_mul_ps(fscal,dz00);
542 /* Update vectorial force */
543 fix0 = _mm256_add_ps(fix0,tx);
544 fiy0 = _mm256_add_ps(fiy0,ty);
545 fiz0 = _mm256_add_ps(fiz0,tz);
547 fjptrA = f+j_coord_offsetA;
548 fjptrB = f+j_coord_offsetB;
549 fjptrC = f+j_coord_offsetC;
550 fjptrD = f+j_coord_offsetD;
551 fjptrE = f+j_coord_offsetE;
552 fjptrF = f+j_coord_offsetF;
553 fjptrG = f+j_coord_offsetG;
554 fjptrH = f+j_coord_offsetH;
555 gmx_mm256_decrement_1rvec_8ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjptrE,fjptrF,fjptrG,fjptrH,tx,ty,tz);
557 /* Inner loop uses 27 flops */
563 /* Get j neighbor index, and coordinate index */
564 jnrlistA = jjnr[jidx];
565 jnrlistB = jjnr[jidx+1];
566 jnrlistC = jjnr[jidx+2];
567 jnrlistD = jjnr[jidx+3];
568 jnrlistE = jjnr[jidx+4];
569 jnrlistF = jjnr[jidx+5];
570 jnrlistG = jjnr[jidx+6];
571 jnrlistH = jjnr[jidx+7];
572 /* Sign of each element will be negative for non-real atoms.
573 * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
574 * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
576 dummy_mask = gmx_mm256_set_m128(gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx+4)),_mm_setzero_si128())),
577 gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128())));
579 jnrA = (jnrlistA>=0) ? jnrlistA : 0;
580 jnrB = (jnrlistB>=0) ? jnrlistB : 0;
581 jnrC = (jnrlistC>=0) ? jnrlistC : 0;
582 jnrD = (jnrlistD>=0) ? jnrlistD : 0;
583 jnrE = (jnrlistE>=0) ? jnrlistE : 0;
584 jnrF = (jnrlistF>=0) ? jnrlistF : 0;
585 jnrG = (jnrlistG>=0) ? jnrlistG : 0;
586 jnrH = (jnrlistH>=0) ? jnrlistH : 0;
587 j_coord_offsetA = DIM*jnrA;
588 j_coord_offsetB = DIM*jnrB;
589 j_coord_offsetC = DIM*jnrC;
590 j_coord_offsetD = DIM*jnrD;
591 j_coord_offsetE = DIM*jnrE;
592 j_coord_offsetF = DIM*jnrF;
593 j_coord_offsetG = DIM*jnrG;
594 j_coord_offsetH = DIM*jnrH;
596 /* load j atom coordinates */
597 gmx_mm256_load_1rvec_8ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
598 x+j_coord_offsetC,x+j_coord_offsetD,
599 x+j_coord_offsetE,x+j_coord_offsetF,
600 x+j_coord_offsetG,x+j_coord_offsetH,
603 /* Calculate displacement vector */
604 dx00 = _mm256_sub_ps(ix0,jx0);
605 dy00 = _mm256_sub_ps(iy0,jy0);
606 dz00 = _mm256_sub_ps(iz0,jz0);
608 /* Calculate squared distance and things based on it */
609 rsq00 = gmx_mm256_calc_rsq_ps(dx00,dy00,dz00);
611 rinv00 = gmx_mm256_invsqrt_ps(rsq00);
613 rinvsq00 = _mm256_mul_ps(rinv00,rinv00);
615 /* Load parameters for j particles */
616 jq0 = gmx_mm256_load_8real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
617 charge+jnrC+0,charge+jnrD+0,
618 charge+jnrE+0,charge+jnrF+0,
619 charge+jnrG+0,charge+jnrH+0);
621 /**************************
622 * CALCULATE INTERACTIONS *
623 **************************/
625 /* Compute parameters for interactions between i and j atoms */
626 qq00 = _mm256_mul_ps(iq0,jq0);
628 /* REACTION-FIELD ELECTROSTATICS */
629 felec = _mm256_mul_ps(qq00,_mm256_sub_ps(_mm256_mul_ps(rinv00,rinvsq00),krf2));
633 fscal = _mm256_andnot_ps(dummy_mask,fscal);
635 /* Calculate temporary vectorial force */
636 tx = _mm256_mul_ps(fscal,dx00);
637 ty = _mm256_mul_ps(fscal,dy00);
638 tz = _mm256_mul_ps(fscal,dz00);
640 /* Update vectorial force */
641 fix0 = _mm256_add_ps(fix0,tx);
642 fiy0 = _mm256_add_ps(fiy0,ty);
643 fiz0 = _mm256_add_ps(fiz0,tz);
645 fjptrA = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
646 fjptrB = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
647 fjptrC = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
648 fjptrD = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
649 fjptrE = (jnrlistE>=0) ? f+j_coord_offsetE : scratch;
650 fjptrF = (jnrlistF>=0) ? f+j_coord_offsetF : scratch;
651 fjptrG = (jnrlistG>=0) ? f+j_coord_offsetG : scratch;
652 fjptrH = (jnrlistH>=0) ? f+j_coord_offsetH : scratch;
653 gmx_mm256_decrement_1rvec_8ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjptrE,fjptrF,fjptrG,fjptrH,tx,ty,tz);
655 /* Inner loop uses 27 flops */
658 /* End of innermost loop */
660 gmx_mm256_update_iforce_1atom_swizzle_ps(fix0,fiy0,fiz0,
661 f+i_coord_offset,fshift+i_shift_offset);
663 /* Increment number of inner iterations */
664 inneriter += j_index_end - j_index_start;
666 /* Outer loop uses 7 flops */
669 /* Increment number of outer iterations */
672 /* Update outer/inner flops */
674 inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_F,outeriter*7 + inneriter*27);