2 * Note: this file was generated by the Gromacs avx_256_double kernel generator.
4 * This source code is part of
8 * Copyright (c) 2001-2012, The GROMACS Development Team
10 * Gromacs is a library for molecular simulation and trajectory analysis,
11 * written by Erik Lindahl, David van der Spoel, Berk Hess, and others - for
12 * a full list of developers and information, check out http://www.gromacs.org
14 * This program is free software; you can redistribute it and/or modify it under
15 * the terms of the GNU Lesser General Public License as published by the Free
16 * Software Foundation; either version 2 of the License, or (at your option) any
19 * To help fund GROMACS development, we humbly ask that you cite
20 * the papers people have written on it - you can find them on the website.
28 #include "../nb_kernel.h"
29 #include "types/simple.h"
33 #include "gmx_math_x86_avx_256_double.h"
34 #include "kernelutil_x86_avx_256_double.h"
37 * Gromacs nonbonded kernel: nb_kernel_ElecRF_VdwNone_GeomW4P1_VF_avx_256_double
38 * Electrostatics interaction: ReactionField
39 * VdW interaction: None
40 * Geometry: Water4-Particle
41 * Calculate force/pot: PotentialAndForce
44 nb_kernel_ElecRF_VdwNone_GeomW4P1_VF_avx_256_double
45 (t_nblist * gmx_restrict nlist,
46 rvec * gmx_restrict xx,
47 rvec * gmx_restrict ff,
48 t_forcerec * gmx_restrict fr,
49 t_mdatoms * gmx_restrict mdatoms,
50 nb_kernel_data_t * gmx_restrict kernel_data,
51 t_nrnb * gmx_restrict nrnb)
53 /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
54 * just 0 for non-waters.
55 * Suffixes A,B,C,D refer to j loop unrolling done with AVX, e.g. for the four different
56 * jnr indices corresponding to data put in the four positions in the SIMD register.
58 int i_shift_offset,i_coord_offset,outeriter,inneriter;
59 int j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
60 int jnrA,jnrB,jnrC,jnrD;
61 int jnrlistA,jnrlistB,jnrlistC,jnrlistD;
62 int jnrlistE,jnrlistF,jnrlistG,jnrlistH;
63 int j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
64 int *iinr,*jindex,*jjnr,*shiftidx,*gid;
66 real *shiftvec,*fshift,*x,*f;
67 real *fjptrA,*fjptrB,*fjptrC,*fjptrD;
69 __m256d tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
70 real * vdwioffsetptr1;
71 __m256d ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
72 real * vdwioffsetptr2;
73 __m256d ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
74 real * vdwioffsetptr3;
75 __m256d ix3,iy3,iz3,fix3,fiy3,fiz3,iq3,isai3;
76 int vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
77 __m256d jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
78 __m256d dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
79 __m256d dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
80 __m256d dx30,dy30,dz30,rsq30,rinv30,rinvsq30,r30,qq30,c6_30,c12_30;
81 __m256d velec,felec,velecsum,facel,crf,krf,krf2;
83 __m256d dummy_mask,cutoff_mask;
84 __m128 tmpmask0,tmpmask1;
85 __m256d signbit = _mm256_castsi256_pd( _mm256_set1_epi32(0x80000000) );
86 __m256d one = _mm256_set1_pd(1.0);
87 __m256d two = _mm256_set1_pd(2.0);
93 jindex = nlist->jindex;
95 shiftidx = nlist->shift;
97 shiftvec = fr->shift_vec[0];
98 fshift = fr->fshift[0];
99 facel = _mm256_set1_pd(fr->epsfac);
100 charge = mdatoms->chargeA;
101 krf = _mm256_set1_pd(fr->ic->k_rf);
102 krf2 = _mm256_set1_pd(fr->ic->k_rf*2.0);
103 crf = _mm256_set1_pd(fr->ic->c_rf);
105 /* Setup water-specific parameters */
106 inr = nlist->iinr[0];
107 iq1 = _mm256_mul_pd(facel,_mm256_set1_pd(charge[inr+1]));
108 iq2 = _mm256_mul_pd(facel,_mm256_set1_pd(charge[inr+2]));
109 iq3 = _mm256_mul_pd(facel,_mm256_set1_pd(charge[inr+3]));
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_mm256_load_shift_and_3rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset+DIM,
142 &ix1,&iy1,&iz1,&ix2,&iy2,&iz2,&ix3,&iy3,&iz3);
144 fix1 = _mm256_setzero_pd();
145 fiy1 = _mm256_setzero_pd();
146 fiz1 = _mm256_setzero_pd();
147 fix2 = _mm256_setzero_pd();
148 fiy2 = _mm256_setzero_pd();
149 fiz2 = _mm256_setzero_pd();
150 fix3 = _mm256_setzero_pd();
151 fiy3 = _mm256_setzero_pd();
152 fiz3 = _mm256_setzero_pd();
154 /* Reset potential sums */
155 velecsum = _mm256_setzero_pd();
157 /* Start inner kernel loop */
158 for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
161 /* Get j neighbor index, and coordinate index */
166 j_coord_offsetA = DIM*jnrA;
167 j_coord_offsetB = DIM*jnrB;
168 j_coord_offsetC = DIM*jnrC;
169 j_coord_offsetD = DIM*jnrD;
171 /* load j atom coordinates */
172 gmx_mm256_load_1rvec_4ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
173 x+j_coord_offsetC,x+j_coord_offsetD,
176 /* Calculate displacement vector */
177 dx10 = _mm256_sub_pd(ix1,jx0);
178 dy10 = _mm256_sub_pd(iy1,jy0);
179 dz10 = _mm256_sub_pd(iz1,jz0);
180 dx20 = _mm256_sub_pd(ix2,jx0);
181 dy20 = _mm256_sub_pd(iy2,jy0);
182 dz20 = _mm256_sub_pd(iz2,jz0);
183 dx30 = _mm256_sub_pd(ix3,jx0);
184 dy30 = _mm256_sub_pd(iy3,jy0);
185 dz30 = _mm256_sub_pd(iz3,jz0);
187 /* Calculate squared distance and things based on it */
188 rsq10 = gmx_mm256_calc_rsq_pd(dx10,dy10,dz10);
189 rsq20 = gmx_mm256_calc_rsq_pd(dx20,dy20,dz20);
190 rsq30 = gmx_mm256_calc_rsq_pd(dx30,dy30,dz30);
192 rinv10 = gmx_mm256_invsqrt_pd(rsq10);
193 rinv20 = gmx_mm256_invsqrt_pd(rsq20);
194 rinv30 = gmx_mm256_invsqrt_pd(rsq30);
196 rinvsq10 = _mm256_mul_pd(rinv10,rinv10);
197 rinvsq20 = _mm256_mul_pd(rinv20,rinv20);
198 rinvsq30 = _mm256_mul_pd(rinv30,rinv30);
200 /* Load parameters for j particles */
201 jq0 = gmx_mm256_load_4real_swizzle_pd(charge+jnrA+0,charge+jnrB+0,
202 charge+jnrC+0,charge+jnrD+0);
204 fjx0 = _mm256_setzero_pd();
205 fjy0 = _mm256_setzero_pd();
206 fjz0 = _mm256_setzero_pd();
208 /**************************
209 * CALCULATE INTERACTIONS *
210 **************************/
212 /* Compute parameters for interactions between i and j atoms */
213 qq10 = _mm256_mul_pd(iq1,jq0);
215 /* REACTION-FIELD ELECTROSTATICS */
216 velec = _mm256_mul_pd(qq10,_mm256_sub_pd(_mm256_add_pd(rinv10,_mm256_mul_pd(krf,rsq10)),crf));
217 felec = _mm256_mul_pd(qq10,_mm256_sub_pd(_mm256_mul_pd(rinv10,rinvsq10),krf2));
219 /* Update potential sum for this i atom from the interaction with this j atom. */
220 velecsum = _mm256_add_pd(velecsum,velec);
224 /* Calculate temporary vectorial force */
225 tx = _mm256_mul_pd(fscal,dx10);
226 ty = _mm256_mul_pd(fscal,dy10);
227 tz = _mm256_mul_pd(fscal,dz10);
229 /* Update vectorial force */
230 fix1 = _mm256_add_pd(fix1,tx);
231 fiy1 = _mm256_add_pd(fiy1,ty);
232 fiz1 = _mm256_add_pd(fiz1,tz);
234 fjx0 = _mm256_add_pd(fjx0,tx);
235 fjy0 = _mm256_add_pd(fjy0,ty);
236 fjz0 = _mm256_add_pd(fjz0,tz);
238 /**************************
239 * CALCULATE INTERACTIONS *
240 **************************/
242 /* Compute parameters for interactions between i and j atoms */
243 qq20 = _mm256_mul_pd(iq2,jq0);
245 /* REACTION-FIELD ELECTROSTATICS */
246 velec = _mm256_mul_pd(qq20,_mm256_sub_pd(_mm256_add_pd(rinv20,_mm256_mul_pd(krf,rsq20)),crf));
247 felec = _mm256_mul_pd(qq20,_mm256_sub_pd(_mm256_mul_pd(rinv20,rinvsq20),krf2));
249 /* Update potential sum for this i atom from the interaction with this j atom. */
250 velecsum = _mm256_add_pd(velecsum,velec);
254 /* Calculate temporary vectorial force */
255 tx = _mm256_mul_pd(fscal,dx20);
256 ty = _mm256_mul_pd(fscal,dy20);
257 tz = _mm256_mul_pd(fscal,dz20);
259 /* Update vectorial force */
260 fix2 = _mm256_add_pd(fix2,tx);
261 fiy2 = _mm256_add_pd(fiy2,ty);
262 fiz2 = _mm256_add_pd(fiz2,tz);
264 fjx0 = _mm256_add_pd(fjx0,tx);
265 fjy0 = _mm256_add_pd(fjy0,ty);
266 fjz0 = _mm256_add_pd(fjz0,tz);
268 /**************************
269 * CALCULATE INTERACTIONS *
270 **************************/
272 /* Compute parameters for interactions between i and j atoms */
273 qq30 = _mm256_mul_pd(iq3,jq0);
275 /* REACTION-FIELD ELECTROSTATICS */
276 velec = _mm256_mul_pd(qq30,_mm256_sub_pd(_mm256_add_pd(rinv30,_mm256_mul_pd(krf,rsq30)),crf));
277 felec = _mm256_mul_pd(qq30,_mm256_sub_pd(_mm256_mul_pd(rinv30,rinvsq30),krf2));
279 /* Update potential sum for this i atom from the interaction with this j atom. */
280 velecsum = _mm256_add_pd(velecsum,velec);
284 /* Calculate temporary vectorial force */
285 tx = _mm256_mul_pd(fscal,dx30);
286 ty = _mm256_mul_pd(fscal,dy30);
287 tz = _mm256_mul_pd(fscal,dz30);
289 /* Update vectorial force */
290 fix3 = _mm256_add_pd(fix3,tx);
291 fiy3 = _mm256_add_pd(fiy3,ty);
292 fiz3 = _mm256_add_pd(fiz3,tz);
294 fjx0 = _mm256_add_pd(fjx0,tx);
295 fjy0 = _mm256_add_pd(fjy0,ty);
296 fjz0 = _mm256_add_pd(fjz0,tz);
298 fjptrA = f+j_coord_offsetA;
299 fjptrB = f+j_coord_offsetB;
300 fjptrC = f+j_coord_offsetC;
301 fjptrD = f+j_coord_offsetD;
303 gmx_mm256_decrement_1rvec_4ptr_swizzle_pd(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
305 /* Inner loop uses 99 flops */
311 /* Get j neighbor index, and coordinate index */
312 jnrlistA = jjnr[jidx];
313 jnrlistB = jjnr[jidx+1];
314 jnrlistC = jjnr[jidx+2];
315 jnrlistD = jjnr[jidx+3];
316 /* Sign of each element will be negative for non-real atoms.
317 * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
318 * so use it as val = _mm_andnot_pd(mask,val) to clear dummy entries.
320 tmpmask0 = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
322 tmpmask1 = _mm_permute_ps(tmpmask0,_GMX_MM_PERMUTE(3,3,2,2));
323 tmpmask0 = _mm_permute_ps(tmpmask0,_GMX_MM_PERMUTE(1,1,0,0));
324 dummy_mask = _mm256_castps_pd(gmx_mm256_set_m128(tmpmask1,tmpmask0));
326 jnrA = (jnrlistA>=0) ? jnrlistA : 0;
327 jnrB = (jnrlistB>=0) ? jnrlistB : 0;
328 jnrC = (jnrlistC>=0) ? jnrlistC : 0;
329 jnrD = (jnrlistD>=0) ? jnrlistD : 0;
330 j_coord_offsetA = DIM*jnrA;
331 j_coord_offsetB = DIM*jnrB;
332 j_coord_offsetC = DIM*jnrC;
333 j_coord_offsetD = DIM*jnrD;
335 /* load j atom coordinates */
336 gmx_mm256_load_1rvec_4ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
337 x+j_coord_offsetC,x+j_coord_offsetD,
340 /* Calculate displacement vector */
341 dx10 = _mm256_sub_pd(ix1,jx0);
342 dy10 = _mm256_sub_pd(iy1,jy0);
343 dz10 = _mm256_sub_pd(iz1,jz0);
344 dx20 = _mm256_sub_pd(ix2,jx0);
345 dy20 = _mm256_sub_pd(iy2,jy0);
346 dz20 = _mm256_sub_pd(iz2,jz0);
347 dx30 = _mm256_sub_pd(ix3,jx0);
348 dy30 = _mm256_sub_pd(iy3,jy0);
349 dz30 = _mm256_sub_pd(iz3,jz0);
351 /* Calculate squared distance and things based on it */
352 rsq10 = gmx_mm256_calc_rsq_pd(dx10,dy10,dz10);
353 rsq20 = gmx_mm256_calc_rsq_pd(dx20,dy20,dz20);
354 rsq30 = gmx_mm256_calc_rsq_pd(dx30,dy30,dz30);
356 rinv10 = gmx_mm256_invsqrt_pd(rsq10);
357 rinv20 = gmx_mm256_invsqrt_pd(rsq20);
358 rinv30 = gmx_mm256_invsqrt_pd(rsq30);
360 rinvsq10 = _mm256_mul_pd(rinv10,rinv10);
361 rinvsq20 = _mm256_mul_pd(rinv20,rinv20);
362 rinvsq30 = _mm256_mul_pd(rinv30,rinv30);
364 /* Load parameters for j particles */
365 jq0 = gmx_mm256_load_4real_swizzle_pd(charge+jnrA+0,charge+jnrB+0,
366 charge+jnrC+0,charge+jnrD+0);
368 fjx0 = _mm256_setzero_pd();
369 fjy0 = _mm256_setzero_pd();
370 fjz0 = _mm256_setzero_pd();
372 /**************************
373 * CALCULATE INTERACTIONS *
374 **************************/
376 /* Compute parameters for interactions between i and j atoms */
377 qq10 = _mm256_mul_pd(iq1,jq0);
379 /* REACTION-FIELD ELECTROSTATICS */
380 velec = _mm256_mul_pd(qq10,_mm256_sub_pd(_mm256_add_pd(rinv10,_mm256_mul_pd(krf,rsq10)),crf));
381 felec = _mm256_mul_pd(qq10,_mm256_sub_pd(_mm256_mul_pd(rinv10,rinvsq10),krf2));
383 /* Update potential sum for this i atom from the interaction with this j atom. */
384 velec = _mm256_andnot_pd(dummy_mask,velec);
385 velecsum = _mm256_add_pd(velecsum,velec);
389 fscal = _mm256_andnot_pd(dummy_mask,fscal);
391 /* Calculate temporary vectorial force */
392 tx = _mm256_mul_pd(fscal,dx10);
393 ty = _mm256_mul_pd(fscal,dy10);
394 tz = _mm256_mul_pd(fscal,dz10);
396 /* Update vectorial force */
397 fix1 = _mm256_add_pd(fix1,tx);
398 fiy1 = _mm256_add_pd(fiy1,ty);
399 fiz1 = _mm256_add_pd(fiz1,tz);
401 fjx0 = _mm256_add_pd(fjx0,tx);
402 fjy0 = _mm256_add_pd(fjy0,ty);
403 fjz0 = _mm256_add_pd(fjz0,tz);
405 /**************************
406 * CALCULATE INTERACTIONS *
407 **************************/
409 /* Compute parameters for interactions between i and j atoms */
410 qq20 = _mm256_mul_pd(iq2,jq0);
412 /* REACTION-FIELD ELECTROSTATICS */
413 velec = _mm256_mul_pd(qq20,_mm256_sub_pd(_mm256_add_pd(rinv20,_mm256_mul_pd(krf,rsq20)),crf));
414 felec = _mm256_mul_pd(qq20,_mm256_sub_pd(_mm256_mul_pd(rinv20,rinvsq20),krf2));
416 /* Update potential sum for this i atom from the interaction with this j atom. */
417 velec = _mm256_andnot_pd(dummy_mask,velec);
418 velecsum = _mm256_add_pd(velecsum,velec);
422 fscal = _mm256_andnot_pd(dummy_mask,fscal);
424 /* Calculate temporary vectorial force */
425 tx = _mm256_mul_pd(fscal,dx20);
426 ty = _mm256_mul_pd(fscal,dy20);
427 tz = _mm256_mul_pd(fscal,dz20);
429 /* Update vectorial force */
430 fix2 = _mm256_add_pd(fix2,tx);
431 fiy2 = _mm256_add_pd(fiy2,ty);
432 fiz2 = _mm256_add_pd(fiz2,tz);
434 fjx0 = _mm256_add_pd(fjx0,tx);
435 fjy0 = _mm256_add_pd(fjy0,ty);
436 fjz0 = _mm256_add_pd(fjz0,tz);
438 /**************************
439 * CALCULATE INTERACTIONS *
440 **************************/
442 /* Compute parameters for interactions between i and j atoms */
443 qq30 = _mm256_mul_pd(iq3,jq0);
445 /* REACTION-FIELD ELECTROSTATICS */
446 velec = _mm256_mul_pd(qq30,_mm256_sub_pd(_mm256_add_pd(rinv30,_mm256_mul_pd(krf,rsq30)),crf));
447 felec = _mm256_mul_pd(qq30,_mm256_sub_pd(_mm256_mul_pd(rinv30,rinvsq30),krf2));
449 /* Update potential sum for this i atom from the interaction with this j atom. */
450 velec = _mm256_andnot_pd(dummy_mask,velec);
451 velecsum = _mm256_add_pd(velecsum,velec);
455 fscal = _mm256_andnot_pd(dummy_mask,fscal);
457 /* Calculate temporary vectorial force */
458 tx = _mm256_mul_pd(fscal,dx30);
459 ty = _mm256_mul_pd(fscal,dy30);
460 tz = _mm256_mul_pd(fscal,dz30);
462 /* Update vectorial force */
463 fix3 = _mm256_add_pd(fix3,tx);
464 fiy3 = _mm256_add_pd(fiy3,ty);
465 fiz3 = _mm256_add_pd(fiz3,tz);
467 fjx0 = _mm256_add_pd(fjx0,tx);
468 fjy0 = _mm256_add_pd(fjy0,ty);
469 fjz0 = _mm256_add_pd(fjz0,tz);
471 fjptrA = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
472 fjptrB = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
473 fjptrC = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
474 fjptrD = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
476 gmx_mm256_decrement_1rvec_4ptr_swizzle_pd(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
478 /* Inner loop uses 99 flops */
481 /* End of innermost loop */
483 gmx_mm256_update_iforce_3atom_swizzle_pd(fix1,fiy1,fiz1,fix2,fiy2,fiz2,fix3,fiy3,fiz3,
484 f+i_coord_offset+DIM,fshift+i_shift_offset);
487 /* Update potential energies */
488 gmx_mm256_update_1pot_pd(velecsum,kernel_data->energygrp_elec+ggid);
490 /* Increment number of inner iterations */
491 inneriter += j_index_end - j_index_start;
493 /* Outer loop uses 19 flops */
496 /* Increment number of outer iterations */
499 /* Update outer/inner flops */
501 inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_W4_VF,outeriter*19 + inneriter*99);
504 * Gromacs nonbonded kernel: nb_kernel_ElecRF_VdwNone_GeomW4P1_F_avx_256_double
505 * Electrostatics interaction: ReactionField
506 * VdW interaction: None
507 * Geometry: Water4-Particle
508 * Calculate force/pot: Force
511 nb_kernel_ElecRF_VdwNone_GeomW4P1_F_avx_256_double
512 (t_nblist * gmx_restrict nlist,
513 rvec * gmx_restrict xx,
514 rvec * gmx_restrict ff,
515 t_forcerec * gmx_restrict fr,
516 t_mdatoms * gmx_restrict mdatoms,
517 nb_kernel_data_t * gmx_restrict kernel_data,
518 t_nrnb * gmx_restrict nrnb)
520 /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
521 * just 0 for non-waters.
522 * Suffixes A,B,C,D refer to j loop unrolling done with AVX, e.g. for the four different
523 * jnr indices corresponding to data put in the four positions in the SIMD register.
525 int i_shift_offset,i_coord_offset,outeriter,inneriter;
526 int j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
527 int jnrA,jnrB,jnrC,jnrD;
528 int jnrlistA,jnrlistB,jnrlistC,jnrlistD;
529 int jnrlistE,jnrlistF,jnrlistG,jnrlistH;
530 int j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
531 int *iinr,*jindex,*jjnr,*shiftidx,*gid;
533 real *shiftvec,*fshift,*x,*f;
534 real *fjptrA,*fjptrB,*fjptrC,*fjptrD;
536 __m256d tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
537 real * vdwioffsetptr1;
538 __m256d ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
539 real * vdwioffsetptr2;
540 __m256d ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
541 real * vdwioffsetptr3;
542 __m256d ix3,iy3,iz3,fix3,fiy3,fiz3,iq3,isai3;
543 int vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
544 __m256d jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
545 __m256d dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
546 __m256d dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
547 __m256d dx30,dy30,dz30,rsq30,rinv30,rinvsq30,r30,qq30,c6_30,c12_30;
548 __m256d velec,felec,velecsum,facel,crf,krf,krf2;
550 __m256d dummy_mask,cutoff_mask;
551 __m128 tmpmask0,tmpmask1;
552 __m256d signbit = _mm256_castsi256_pd( _mm256_set1_epi32(0x80000000) );
553 __m256d one = _mm256_set1_pd(1.0);
554 __m256d two = _mm256_set1_pd(2.0);
560 jindex = nlist->jindex;
562 shiftidx = nlist->shift;
564 shiftvec = fr->shift_vec[0];
565 fshift = fr->fshift[0];
566 facel = _mm256_set1_pd(fr->epsfac);
567 charge = mdatoms->chargeA;
568 krf = _mm256_set1_pd(fr->ic->k_rf);
569 krf2 = _mm256_set1_pd(fr->ic->k_rf*2.0);
570 crf = _mm256_set1_pd(fr->ic->c_rf);
572 /* Setup water-specific parameters */
573 inr = nlist->iinr[0];
574 iq1 = _mm256_mul_pd(facel,_mm256_set1_pd(charge[inr+1]));
575 iq2 = _mm256_mul_pd(facel,_mm256_set1_pd(charge[inr+2]));
576 iq3 = _mm256_mul_pd(facel,_mm256_set1_pd(charge[inr+3]));
578 /* Avoid stupid compiler warnings */
579 jnrA = jnrB = jnrC = jnrD = 0;
588 for(iidx=0;iidx<4*DIM;iidx++)
593 /* Start outer loop over neighborlists */
594 for(iidx=0; iidx<nri; iidx++)
596 /* Load shift vector for this list */
597 i_shift_offset = DIM*shiftidx[iidx];
599 /* Load limits for loop over neighbors */
600 j_index_start = jindex[iidx];
601 j_index_end = jindex[iidx+1];
603 /* Get outer coordinate index */
605 i_coord_offset = DIM*inr;
607 /* Load i particle coords and add shift vector */
608 gmx_mm256_load_shift_and_3rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset+DIM,
609 &ix1,&iy1,&iz1,&ix2,&iy2,&iz2,&ix3,&iy3,&iz3);
611 fix1 = _mm256_setzero_pd();
612 fiy1 = _mm256_setzero_pd();
613 fiz1 = _mm256_setzero_pd();
614 fix2 = _mm256_setzero_pd();
615 fiy2 = _mm256_setzero_pd();
616 fiz2 = _mm256_setzero_pd();
617 fix3 = _mm256_setzero_pd();
618 fiy3 = _mm256_setzero_pd();
619 fiz3 = _mm256_setzero_pd();
621 /* Start inner kernel loop */
622 for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
625 /* Get j neighbor index, and coordinate index */
630 j_coord_offsetA = DIM*jnrA;
631 j_coord_offsetB = DIM*jnrB;
632 j_coord_offsetC = DIM*jnrC;
633 j_coord_offsetD = DIM*jnrD;
635 /* load j atom coordinates */
636 gmx_mm256_load_1rvec_4ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
637 x+j_coord_offsetC,x+j_coord_offsetD,
640 /* Calculate displacement vector */
641 dx10 = _mm256_sub_pd(ix1,jx0);
642 dy10 = _mm256_sub_pd(iy1,jy0);
643 dz10 = _mm256_sub_pd(iz1,jz0);
644 dx20 = _mm256_sub_pd(ix2,jx0);
645 dy20 = _mm256_sub_pd(iy2,jy0);
646 dz20 = _mm256_sub_pd(iz2,jz0);
647 dx30 = _mm256_sub_pd(ix3,jx0);
648 dy30 = _mm256_sub_pd(iy3,jy0);
649 dz30 = _mm256_sub_pd(iz3,jz0);
651 /* Calculate squared distance and things based on it */
652 rsq10 = gmx_mm256_calc_rsq_pd(dx10,dy10,dz10);
653 rsq20 = gmx_mm256_calc_rsq_pd(dx20,dy20,dz20);
654 rsq30 = gmx_mm256_calc_rsq_pd(dx30,dy30,dz30);
656 rinv10 = gmx_mm256_invsqrt_pd(rsq10);
657 rinv20 = gmx_mm256_invsqrt_pd(rsq20);
658 rinv30 = gmx_mm256_invsqrt_pd(rsq30);
660 rinvsq10 = _mm256_mul_pd(rinv10,rinv10);
661 rinvsq20 = _mm256_mul_pd(rinv20,rinv20);
662 rinvsq30 = _mm256_mul_pd(rinv30,rinv30);
664 /* Load parameters for j particles */
665 jq0 = gmx_mm256_load_4real_swizzle_pd(charge+jnrA+0,charge+jnrB+0,
666 charge+jnrC+0,charge+jnrD+0);
668 fjx0 = _mm256_setzero_pd();
669 fjy0 = _mm256_setzero_pd();
670 fjz0 = _mm256_setzero_pd();
672 /**************************
673 * CALCULATE INTERACTIONS *
674 **************************/
676 /* Compute parameters for interactions between i and j atoms */
677 qq10 = _mm256_mul_pd(iq1,jq0);
679 /* REACTION-FIELD ELECTROSTATICS */
680 felec = _mm256_mul_pd(qq10,_mm256_sub_pd(_mm256_mul_pd(rinv10,rinvsq10),krf2));
684 /* Calculate temporary vectorial force */
685 tx = _mm256_mul_pd(fscal,dx10);
686 ty = _mm256_mul_pd(fscal,dy10);
687 tz = _mm256_mul_pd(fscal,dz10);
689 /* Update vectorial force */
690 fix1 = _mm256_add_pd(fix1,tx);
691 fiy1 = _mm256_add_pd(fiy1,ty);
692 fiz1 = _mm256_add_pd(fiz1,tz);
694 fjx0 = _mm256_add_pd(fjx0,tx);
695 fjy0 = _mm256_add_pd(fjy0,ty);
696 fjz0 = _mm256_add_pd(fjz0,tz);
698 /**************************
699 * CALCULATE INTERACTIONS *
700 **************************/
702 /* Compute parameters for interactions between i and j atoms */
703 qq20 = _mm256_mul_pd(iq2,jq0);
705 /* REACTION-FIELD ELECTROSTATICS */
706 felec = _mm256_mul_pd(qq20,_mm256_sub_pd(_mm256_mul_pd(rinv20,rinvsq20),krf2));
710 /* Calculate temporary vectorial force */
711 tx = _mm256_mul_pd(fscal,dx20);
712 ty = _mm256_mul_pd(fscal,dy20);
713 tz = _mm256_mul_pd(fscal,dz20);
715 /* Update vectorial force */
716 fix2 = _mm256_add_pd(fix2,tx);
717 fiy2 = _mm256_add_pd(fiy2,ty);
718 fiz2 = _mm256_add_pd(fiz2,tz);
720 fjx0 = _mm256_add_pd(fjx0,tx);
721 fjy0 = _mm256_add_pd(fjy0,ty);
722 fjz0 = _mm256_add_pd(fjz0,tz);
724 /**************************
725 * CALCULATE INTERACTIONS *
726 **************************/
728 /* Compute parameters for interactions between i and j atoms */
729 qq30 = _mm256_mul_pd(iq3,jq0);
731 /* REACTION-FIELD ELECTROSTATICS */
732 felec = _mm256_mul_pd(qq30,_mm256_sub_pd(_mm256_mul_pd(rinv30,rinvsq30),krf2));
736 /* Calculate temporary vectorial force */
737 tx = _mm256_mul_pd(fscal,dx30);
738 ty = _mm256_mul_pd(fscal,dy30);
739 tz = _mm256_mul_pd(fscal,dz30);
741 /* Update vectorial force */
742 fix3 = _mm256_add_pd(fix3,tx);
743 fiy3 = _mm256_add_pd(fiy3,ty);
744 fiz3 = _mm256_add_pd(fiz3,tz);
746 fjx0 = _mm256_add_pd(fjx0,tx);
747 fjy0 = _mm256_add_pd(fjy0,ty);
748 fjz0 = _mm256_add_pd(fjz0,tz);
750 fjptrA = f+j_coord_offsetA;
751 fjptrB = f+j_coord_offsetB;
752 fjptrC = f+j_coord_offsetC;
753 fjptrD = f+j_coord_offsetD;
755 gmx_mm256_decrement_1rvec_4ptr_swizzle_pd(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
757 /* Inner loop uses 84 flops */
763 /* Get j neighbor index, and coordinate index */
764 jnrlistA = jjnr[jidx];
765 jnrlistB = jjnr[jidx+1];
766 jnrlistC = jjnr[jidx+2];
767 jnrlistD = jjnr[jidx+3];
768 /* Sign of each element will be negative for non-real atoms.
769 * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
770 * so use it as val = _mm_andnot_pd(mask,val) to clear dummy entries.
772 tmpmask0 = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
774 tmpmask1 = _mm_permute_ps(tmpmask0,_GMX_MM_PERMUTE(3,3,2,2));
775 tmpmask0 = _mm_permute_ps(tmpmask0,_GMX_MM_PERMUTE(1,1,0,0));
776 dummy_mask = _mm256_castps_pd(gmx_mm256_set_m128(tmpmask1,tmpmask0));
778 jnrA = (jnrlistA>=0) ? jnrlistA : 0;
779 jnrB = (jnrlistB>=0) ? jnrlistB : 0;
780 jnrC = (jnrlistC>=0) ? jnrlistC : 0;
781 jnrD = (jnrlistD>=0) ? jnrlistD : 0;
782 j_coord_offsetA = DIM*jnrA;
783 j_coord_offsetB = DIM*jnrB;
784 j_coord_offsetC = DIM*jnrC;
785 j_coord_offsetD = DIM*jnrD;
787 /* load j atom coordinates */
788 gmx_mm256_load_1rvec_4ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
789 x+j_coord_offsetC,x+j_coord_offsetD,
792 /* Calculate displacement vector */
793 dx10 = _mm256_sub_pd(ix1,jx0);
794 dy10 = _mm256_sub_pd(iy1,jy0);
795 dz10 = _mm256_sub_pd(iz1,jz0);
796 dx20 = _mm256_sub_pd(ix2,jx0);
797 dy20 = _mm256_sub_pd(iy2,jy0);
798 dz20 = _mm256_sub_pd(iz2,jz0);
799 dx30 = _mm256_sub_pd(ix3,jx0);
800 dy30 = _mm256_sub_pd(iy3,jy0);
801 dz30 = _mm256_sub_pd(iz3,jz0);
803 /* Calculate squared distance and things based on it */
804 rsq10 = gmx_mm256_calc_rsq_pd(dx10,dy10,dz10);
805 rsq20 = gmx_mm256_calc_rsq_pd(dx20,dy20,dz20);
806 rsq30 = gmx_mm256_calc_rsq_pd(dx30,dy30,dz30);
808 rinv10 = gmx_mm256_invsqrt_pd(rsq10);
809 rinv20 = gmx_mm256_invsqrt_pd(rsq20);
810 rinv30 = gmx_mm256_invsqrt_pd(rsq30);
812 rinvsq10 = _mm256_mul_pd(rinv10,rinv10);
813 rinvsq20 = _mm256_mul_pd(rinv20,rinv20);
814 rinvsq30 = _mm256_mul_pd(rinv30,rinv30);
816 /* Load parameters for j particles */
817 jq0 = gmx_mm256_load_4real_swizzle_pd(charge+jnrA+0,charge+jnrB+0,
818 charge+jnrC+0,charge+jnrD+0);
820 fjx0 = _mm256_setzero_pd();
821 fjy0 = _mm256_setzero_pd();
822 fjz0 = _mm256_setzero_pd();
824 /**************************
825 * CALCULATE INTERACTIONS *
826 **************************/
828 /* Compute parameters for interactions between i and j atoms */
829 qq10 = _mm256_mul_pd(iq1,jq0);
831 /* REACTION-FIELD ELECTROSTATICS */
832 felec = _mm256_mul_pd(qq10,_mm256_sub_pd(_mm256_mul_pd(rinv10,rinvsq10),krf2));
836 fscal = _mm256_andnot_pd(dummy_mask,fscal);
838 /* Calculate temporary vectorial force */
839 tx = _mm256_mul_pd(fscal,dx10);
840 ty = _mm256_mul_pd(fscal,dy10);
841 tz = _mm256_mul_pd(fscal,dz10);
843 /* Update vectorial force */
844 fix1 = _mm256_add_pd(fix1,tx);
845 fiy1 = _mm256_add_pd(fiy1,ty);
846 fiz1 = _mm256_add_pd(fiz1,tz);
848 fjx0 = _mm256_add_pd(fjx0,tx);
849 fjy0 = _mm256_add_pd(fjy0,ty);
850 fjz0 = _mm256_add_pd(fjz0,tz);
852 /**************************
853 * CALCULATE INTERACTIONS *
854 **************************/
856 /* Compute parameters for interactions between i and j atoms */
857 qq20 = _mm256_mul_pd(iq2,jq0);
859 /* REACTION-FIELD ELECTROSTATICS */
860 felec = _mm256_mul_pd(qq20,_mm256_sub_pd(_mm256_mul_pd(rinv20,rinvsq20),krf2));
864 fscal = _mm256_andnot_pd(dummy_mask,fscal);
866 /* Calculate temporary vectorial force */
867 tx = _mm256_mul_pd(fscal,dx20);
868 ty = _mm256_mul_pd(fscal,dy20);
869 tz = _mm256_mul_pd(fscal,dz20);
871 /* Update vectorial force */
872 fix2 = _mm256_add_pd(fix2,tx);
873 fiy2 = _mm256_add_pd(fiy2,ty);
874 fiz2 = _mm256_add_pd(fiz2,tz);
876 fjx0 = _mm256_add_pd(fjx0,tx);
877 fjy0 = _mm256_add_pd(fjy0,ty);
878 fjz0 = _mm256_add_pd(fjz0,tz);
880 /**************************
881 * CALCULATE INTERACTIONS *
882 **************************/
884 /* Compute parameters for interactions between i and j atoms */
885 qq30 = _mm256_mul_pd(iq3,jq0);
887 /* REACTION-FIELD ELECTROSTATICS */
888 felec = _mm256_mul_pd(qq30,_mm256_sub_pd(_mm256_mul_pd(rinv30,rinvsq30),krf2));
892 fscal = _mm256_andnot_pd(dummy_mask,fscal);
894 /* Calculate temporary vectorial force */
895 tx = _mm256_mul_pd(fscal,dx30);
896 ty = _mm256_mul_pd(fscal,dy30);
897 tz = _mm256_mul_pd(fscal,dz30);
899 /* Update vectorial force */
900 fix3 = _mm256_add_pd(fix3,tx);
901 fiy3 = _mm256_add_pd(fiy3,ty);
902 fiz3 = _mm256_add_pd(fiz3,tz);
904 fjx0 = _mm256_add_pd(fjx0,tx);
905 fjy0 = _mm256_add_pd(fjy0,ty);
906 fjz0 = _mm256_add_pd(fjz0,tz);
908 fjptrA = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
909 fjptrB = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
910 fjptrC = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
911 fjptrD = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
913 gmx_mm256_decrement_1rvec_4ptr_swizzle_pd(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
915 /* Inner loop uses 84 flops */
918 /* End of innermost loop */
920 gmx_mm256_update_iforce_3atom_swizzle_pd(fix1,fiy1,fiz1,fix2,fiy2,fiz2,fix3,fiy3,fiz3,
921 f+i_coord_offset+DIM,fshift+i_shift_offset);
923 /* Increment number of inner iterations */
924 inneriter += j_index_end - j_index_start;
926 /* Outer loop uses 18 flops */
929 /* Increment number of outer iterations */
932 /* Update outer/inner flops */
934 inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_W4_F,outeriter*18 + inneriter*84);