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_VdwLJ_GeomW4P1_VF_avx_256_single
38 * Electrostatics interaction: ReactionField
39 * VdW interaction: LennardJones
40 * Geometry: Water4-Particle
41 * Calculate force/pot: PotentialAndForce
44 nb_kernel_ElecRF_VdwLJ_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 * vdwioffsetptr0;
73 __m256 ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
74 real * vdwioffsetptr1;
75 __m256 ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
76 real * vdwioffsetptr2;
77 __m256 ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
78 real * vdwioffsetptr3;
79 __m256 ix3,iy3,iz3,fix3,fiy3,fiz3,iq3,isai3;
80 int vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D,vdwjidx0E,vdwjidx0F,vdwjidx0G,vdwjidx0H;
81 __m256 jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
82 __m256 dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
83 __m256 dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
84 __m256 dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
85 __m256 dx30,dy30,dz30,rsq30,rinv30,rinvsq30,r30,qq30,c6_30,c12_30;
86 __m256 velec,felec,velecsum,facel,crf,krf,krf2;
89 __m256 rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
92 __m256 one_sixth = _mm256_set1_ps(1.0/6.0);
93 __m256 one_twelfth = _mm256_set1_ps(1.0/12.0);
94 __m256 dummy_mask,cutoff_mask;
95 __m256 signbit = _mm256_castsi256_ps( _mm256_set1_epi32(0x80000000) );
96 __m256 one = _mm256_set1_ps(1.0);
97 __m256 two = _mm256_set1_ps(2.0);
103 jindex = nlist->jindex;
105 shiftidx = nlist->shift;
107 shiftvec = fr->shift_vec[0];
108 fshift = fr->fshift[0];
109 facel = _mm256_set1_ps(fr->epsfac);
110 charge = mdatoms->chargeA;
111 krf = _mm256_set1_ps(fr->ic->k_rf);
112 krf2 = _mm256_set1_ps(fr->ic->k_rf*2.0);
113 crf = _mm256_set1_ps(fr->ic->c_rf);
114 nvdwtype = fr->ntype;
116 vdwtype = mdatoms->typeA;
118 /* Setup water-specific parameters */
119 inr = nlist->iinr[0];
120 iq1 = _mm256_mul_ps(facel,_mm256_set1_ps(charge[inr+1]));
121 iq2 = _mm256_mul_ps(facel,_mm256_set1_ps(charge[inr+2]));
122 iq3 = _mm256_mul_ps(facel,_mm256_set1_ps(charge[inr+3]));
123 vdwioffsetptr0 = vdwparam+2*nvdwtype*vdwtype[inr+0];
125 /* Avoid stupid compiler warnings */
126 jnrA = jnrB = jnrC = jnrD = jnrE = jnrF = jnrG = jnrH = 0;
139 for(iidx=0;iidx<4*DIM;iidx++)
144 /* Start outer loop over neighborlists */
145 for(iidx=0; iidx<nri; iidx++)
147 /* Load shift vector for this list */
148 i_shift_offset = DIM*shiftidx[iidx];
150 /* Load limits for loop over neighbors */
151 j_index_start = jindex[iidx];
152 j_index_end = jindex[iidx+1];
154 /* Get outer coordinate index */
156 i_coord_offset = DIM*inr;
158 /* Load i particle coords and add shift vector */
159 gmx_mm256_load_shift_and_4rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,
160 &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2,&ix3,&iy3,&iz3);
162 fix0 = _mm256_setzero_ps();
163 fiy0 = _mm256_setzero_ps();
164 fiz0 = _mm256_setzero_ps();
165 fix1 = _mm256_setzero_ps();
166 fiy1 = _mm256_setzero_ps();
167 fiz1 = _mm256_setzero_ps();
168 fix2 = _mm256_setzero_ps();
169 fiy2 = _mm256_setzero_ps();
170 fiz2 = _mm256_setzero_ps();
171 fix3 = _mm256_setzero_ps();
172 fiy3 = _mm256_setzero_ps();
173 fiz3 = _mm256_setzero_ps();
175 /* Reset potential sums */
176 velecsum = _mm256_setzero_ps();
177 vvdwsum = _mm256_setzero_ps();
179 /* Start inner kernel loop */
180 for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+7]>=0; jidx+=8)
183 /* Get j neighbor index, and coordinate index */
192 j_coord_offsetA = DIM*jnrA;
193 j_coord_offsetB = DIM*jnrB;
194 j_coord_offsetC = DIM*jnrC;
195 j_coord_offsetD = DIM*jnrD;
196 j_coord_offsetE = DIM*jnrE;
197 j_coord_offsetF = DIM*jnrF;
198 j_coord_offsetG = DIM*jnrG;
199 j_coord_offsetH = DIM*jnrH;
201 /* load j atom coordinates */
202 gmx_mm256_load_1rvec_8ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
203 x+j_coord_offsetC,x+j_coord_offsetD,
204 x+j_coord_offsetE,x+j_coord_offsetF,
205 x+j_coord_offsetG,x+j_coord_offsetH,
208 /* Calculate displacement vector */
209 dx00 = _mm256_sub_ps(ix0,jx0);
210 dy00 = _mm256_sub_ps(iy0,jy0);
211 dz00 = _mm256_sub_ps(iz0,jz0);
212 dx10 = _mm256_sub_ps(ix1,jx0);
213 dy10 = _mm256_sub_ps(iy1,jy0);
214 dz10 = _mm256_sub_ps(iz1,jz0);
215 dx20 = _mm256_sub_ps(ix2,jx0);
216 dy20 = _mm256_sub_ps(iy2,jy0);
217 dz20 = _mm256_sub_ps(iz2,jz0);
218 dx30 = _mm256_sub_ps(ix3,jx0);
219 dy30 = _mm256_sub_ps(iy3,jy0);
220 dz30 = _mm256_sub_ps(iz3,jz0);
222 /* Calculate squared distance and things based on it */
223 rsq00 = gmx_mm256_calc_rsq_ps(dx00,dy00,dz00);
224 rsq10 = gmx_mm256_calc_rsq_ps(dx10,dy10,dz10);
225 rsq20 = gmx_mm256_calc_rsq_ps(dx20,dy20,dz20);
226 rsq30 = gmx_mm256_calc_rsq_ps(dx30,dy30,dz30);
228 rinv10 = gmx_mm256_invsqrt_ps(rsq10);
229 rinv20 = gmx_mm256_invsqrt_ps(rsq20);
230 rinv30 = gmx_mm256_invsqrt_ps(rsq30);
232 rinvsq00 = gmx_mm256_inv_ps(rsq00);
233 rinvsq10 = _mm256_mul_ps(rinv10,rinv10);
234 rinvsq20 = _mm256_mul_ps(rinv20,rinv20);
235 rinvsq30 = _mm256_mul_ps(rinv30,rinv30);
237 /* Load parameters for j particles */
238 jq0 = gmx_mm256_load_8real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
239 charge+jnrC+0,charge+jnrD+0,
240 charge+jnrE+0,charge+jnrF+0,
241 charge+jnrG+0,charge+jnrH+0);
242 vdwjidx0A = 2*vdwtype[jnrA+0];
243 vdwjidx0B = 2*vdwtype[jnrB+0];
244 vdwjidx0C = 2*vdwtype[jnrC+0];
245 vdwjidx0D = 2*vdwtype[jnrD+0];
246 vdwjidx0E = 2*vdwtype[jnrE+0];
247 vdwjidx0F = 2*vdwtype[jnrF+0];
248 vdwjidx0G = 2*vdwtype[jnrG+0];
249 vdwjidx0H = 2*vdwtype[jnrH+0];
251 fjx0 = _mm256_setzero_ps();
252 fjy0 = _mm256_setzero_ps();
253 fjz0 = _mm256_setzero_ps();
255 /**************************
256 * CALCULATE INTERACTIONS *
257 **************************/
259 /* Compute parameters for interactions between i and j atoms */
260 gmx_mm256_load_8pair_swizzle_ps(vdwioffsetptr0+vdwjidx0A,
261 vdwioffsetptr0+vdwjidx0B,
262 vdwioffsetptr0+vdwjidx0C,
263 vdwioffsetptr0+vdwjidx0D,
264 vdwioffsetptr0+vdwjidx0E,
265 vdwioffsetptr0+vdwjidx0F,
266 vdwioffsetptr0+vdwjidx0G,
267 vdwioffsetptr0+vdwjidx0H,
270 /* LENNARD-JONES DISPERSION/REPULSION */
272 rinvsix = _mm256_mul_ps(_mm256_mul_ps(rinvsq00,rinvsq00),rinvsq00);
273 vvdw6 = _mm256_mul_ps(c6_00,rinvsix);
274 vvdw12 = _mm256_mul_ps(c12_00,_mm256_mul_ps(rinvsix,rinvsix));
275 vvdw = _mm256_sub_ps( _mm256_mul_ps(vvdw12,one_twelfth) , _mm256_mul_ps(vvdw6,one_sixth) );
276 fvdw = _mm256_mul_ps(_mm256_sub_ps(vvdw12,vvdw6),rinvsq00);
278 /* Update potential sum for this i atom from the interaction with this j atom. */
279 vvdwsum = _mm256_add_ps(vvdwsum,vvdw);
283 /* Calculate temporary vectorial force */
284 tx = _mm256_mul_ps(fscal,dx00);
285 ty = _mm256_mul_ps(fscal,dy00);
286 tz = _mm256_mul_ps(fscal,dz00);
288 /* Update vectorial force */
289 fix0 = _mm256_add_ps(fix0,tx);
290 fiy0 = _mm256_add_ps(fiy0,ty);
291 fiz0 = _mm256_add_ps(fiz0,tz);
293 fjx0 = _mm256_add_ps(fjx0,tx);
294 fjy0 = _mm256_add_ps(fjy0,ty);
295 fjz0 = _mm256_add_ps(fjz0,tz);
297 /**************************
298 * CALCULATE INTERACTIONS *
299 **************************/
301 /* Compute parameters for interactions between i and j atoms */
302 qq10 = _mm256_mul_ps(iq1,jq0);
304 /* REACTION-FIELD ELECTROSTATICS */
305 velec = _mm256_mul_ps(qq10,_mm256_sub_ps(_mm256_add_ps(rinv10,_mm256_mul_ps(krf,rsq10)),crf));
306 felec = _mm256_mul_ps(qq10,_mm256_sub_ps(_mm256_mul_ps(rinv10,rinvsq10),krf2));
308 /* Update potential sum for this i atom from the interaction with this j atom. */
309 velecsum = _mm256_add_ps(velecsum,velec);
313 /* Calculate temporary vectorial force */
314 tx = _mm256_mul_ps(fscal,dx10);
315 ty = _mm256_mul_ps(fscal,dy10);
316 tz = _mm256_mul_ps(fscal,dz10);
318 /* Update vectorial force */
319 fix1 = _mm256_add_ps(fix1,tx);
320 fiy1 = _mm256_add_ps(fiy1,ty);
321 fiz1 = _mm256_add_ps(fiz1,tz);
323 fjx0 = _mm256_add_ps(fjx0,tx);
324 fjy0 = _mm256_add_ps(fjy0,ty);
325 fjz0 = _mm256_add_ps(fjz0,tz);
327 /**************************
328 * CALCULATE INTERACTIONS *
329 **************************/
331 /* Compute parameters for interactions between i and j atoms */
332 qq20 = _mm256_mul_ps(iq2,jq0);
334 /* REACTION-FIELD ELECTROSTATICS */
335 velec = _mm256_mul_ps(qq20,_mm256_sub_ps(_mm256_add_ps(rinv20,_mm256_mul_ps(krf,rsq20)),crf));
336 felec = _mm256_mul_ps(qq20,_mm256_sub_ps(_mm256_mul_ps(rinv20,rinvsq20),krf2));
338 /* Update potential sum for this i atom from the interaction with this j atom. */
339 velecsum = _mm256_add_ps(velecsum,velec);
343 /* Calculate temporary vectorial force */
344 tx = _mm256_mul_ps(fscal,dx20);
345 ty = _mm256_mul_ps(fscal,dy20);
346 tz = _mm256_mul_ps(fscal,dz20);
348 /* Update vectorial force */
349 fix2 = _mm256_add_ps(fix2,tx);
350 fiy2 = _mm256_add_ps(fiy2,ty);
351 fiz2 = _mm256_add_ps(fiz2,tz);
353 fjx0 = _mm256_add_ps(fjx0,tx);
354 fjy0 = _mm256_add_ps(fjy0,ty);
355 fjz0 = _mm256_add_ps(fjz0,tz);
357 /**************************
358 * CALCULATE INTERACTIONS *
359 **************************/
361 /* Compute parameters for interactions between i and j atoms */
362 qq30 = _mm256_mul_ps(iq3,jq0);
364 /* REACTION-FIELD ELECTROSTATICS */
365 velec = _mm256_mul_ps(qq30,_mm256_sub_ps(_mm256_add_ps(rinv30,_mm256_mul_ps(krf,rsq30)),crf));
366 felec = _mm256_mul_ps(qq30,_mm256_sub_ps(_mm256_mul_ps(rinv30,rinvsq30),krf2));
368 /* Update potential sum for this i atom from the interaction with this j atom. */
369 velecsum = _mm256_add_ps(velecsum,velec);
373 /* Calculate temporary vectorial force */
374 tx = _mm256_mul_ps(fscal,dx30);
375 ty = _mm256_mul_ps(fscal,dy30);
376 tz = _mm256_mul_ps(fscal,dz30);
378 /* Update vectorial force */
379 fix3 = _mm256_add_ps(fix3,tx);
380 fiy3 = _mm256_add_ps(fiy3,ty);
381 fiz3 = _mm256_add_ps(fiz3,tz);
383 fjx0 = _mm256_add_ps(fjx0,tx);
384 fjy0 = _mm256_add_ps(fjy0,ty);
385 fjz0 = _mm256_add_ps(fjz0,tz);
387 fjptrA = f+j_coord_offsetA;
388 fjptrB = f+j_coord_offsetB;
389 fjptrC = f+j_coord_offsetC;
390 fjptrD = f+j_coord_offsetD;
391 fjptrE = f+j_coord_offsetE;
392 fjptrF = f+j_coord_offsetF;
393 fjptrG = f+j_coord_offsetG;
394 fjptrH = f+j_coord_offsetH;
396 gmx_mm256_decrement_1rvec_8ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjptrE,fjptrF,fjptrG,fjptrH,fjx0,fjy0,fjz0);
398 /* Inner loop uses 131 flops */
404 /* Get j neighbor index, and coordinate index */
405 jnrlistA = jjnr[jidx];
406 jnrlistB = jjnr[jidx+1];
407 jnrlistC = jjnr[jidx+2];
408 jnrlistD = jjnr[jidx+3];
409 jnrlistE = jjnr[jidx+4];
410 jnrlistF = jjnr[jidx+5];
411 jnrlistG = jjnr[jidx+6];
412 jnrlistH = jjnr[jidx+7];
413 /* Sign of each element will be negative for non-real atoms.
414 * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
415 * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
417 dummy_mask = gmx_mm256_set_m128(gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx+4)),_mm_setzero_si128())),
418 gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128())));
420 jnrA = (jnrlistA>=0) ? jnrlistA : 0;
421 jnrB = (jnrlistB>=0) ? jnrlistB : 0;
422 jnrC = (jnrlistC>=0) ? jnrlistC : 0;
423 jnrD = (jnrlistD>=0) ? jnrlistD : 0;
424 jnrE = (jnrlistE>=0) ? jnrlistE : 0;
425 jnrF = (jnrlistF>=0) ? jnrlistF : 0;
426 jnrG = (jnrlistG>=0) ? jnrlistG : 0;
427 jnrH = (jnrlistH>=0) ? jnrlistH : 0;
428 j_coord_offsetA = DIM*jnrA;
429 j_coord_offsetB = DIM*jnrB;
430 j_coord_offsetC = DIM*jnrC;
431 j_coord_offsetD = DIM*jnrD;
432 j_coord_offsetE = DIM*jnrE;
433 j_coord_offsetF = DIM*jnrF;
434 j_coord_offsetG = DIM*jnrG;
435 j_coord_offsetH = DIM*jnrH;
437 /* load j atom coordinates */
438 gmx_mm256_load_1rvec_8ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
439 x+j_coord_offsetC,x+j_coord_offsetD,
440 x+j_coord_offsetE,x+j_coord_offsetF,
441 x+j_coord_offsetG,x+j_coord_offsetH,
444 /* Calculate displacement vector */
445 dx00 = _mm256_sub_ps(ix0,jx0);
446 dy00 = _mm256_sub_ps(iy0,jy0);
447 dz00 = _mm256_sub_ps(iz0,jz0);
448 dx10 = _mm256_sub_ps(ix1,jx0);
449 dy10 = _mm256_sub_ps(iy1,jy0);
450 dz10 = _mm256_sub_ps(iz1,jz0);
451 dx20 = _mm256_sub_ps(ix2,jx0);
452 dy20 = _mm256_sub_ps(iy2,jy0);
453 dz20 = _mm256_sub_ps(iz2,jz0);
454 dx30 = _mm256_sub_ps(ix3,jx0);
455 dy30 = _mm256_sub_ps(iy3,jy0);
456 dz30 = _mm256_sub_ps(iz3,jz0);
458 /* Calculate squared distance and things based on it */
459 rsq00 = gmx_mm256_calc_rsq_ps(dx00,dy00,dz00);
460 rsq10 = gmx_mm256_calc_rsq_ps(dx10,dy10,dz10);
461 rsq20 = gmx_mm256_calc_rsq_ps(dx20,dy20,dz20);
462 rsq30 = gmx_mm256_calc_rsq_ps(dx30,dy30,dz30);
464 rinv10 = gmx_mm256_invsqrt_ps(rsq10);
465 rinv20 = gmx_mm256_invsqrt_ps(rsq20);
466 rinv30 = gmx_mm256_invsqrt_ps(rsq30);
468 rinvsq00 = gmx_mm256_inv_ps(rsq00);
469 rinvsq10 = _mm256_mul_ps(rinv10,rinv10);
470 rinvsq20 = _mm256_mul_ps(rinv20,rinv20);
471 rinvsq30 = _mm256_mul_ps(rinv30,rinv30);
473 /* Load parameters for j particles */
474 jq0 = gmx_mm256_load_8real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
475 charge+jnrC+0,charge+jnrD+0,
476 charge+jnrE+0,charge+jnrF+0,
477 charge+jnrG+0,charge+jnrH+0);
478 vdwjidx0A = 2*vdwtype[jnrA+0];
479 vdwjidx0B = 2*vdwtype[jnrB+0];
480 vdwjidx0C = 2*vdwtype[jnrC+0];
481 vdwjidx0D = 2*vdwtype[jnrD+0];
482 vdwjidx0E = 2*vdwtype[jnrE+0];
483 vdwjidx0F = 2*vdwtype[jnrF+0];
484 vdwjidx0G = 2*vdwtype[jnrG+0];
485 vdwjidx0H = 2*vdwtype[jnrH+0];
487 fjx0 = _mm256_setzero_ps();
488 fjy0 = _mm256_setzero_ps();
489 fjz0 = _mm256_setzero_ps();
491 /**************************
492 * CALCULATE INTERACTIONS *
493 **************************/
495 /* Compute parameters for interactions between i and j atoms */
496 gmx_mm256_load_8pair_swizzle_ps(vdwioffsetptr0+vdwjidx0A,
497 vdwioffsetptr0+vdwjidx0B,
498 vdwioffsetptr0+vdwjidx0C,
499 vdwioffsetptr0+vdwjidx0D,
500 vdwioffsetptr0+vdwjidx0E,
501 vdwioffsetptr0+vdwjidx0F,
502 vdwioffsetptr0+vdwjidx0G,
503 vdwioffsetptr0+vdwjidx0H,
506 /* LENNARD-JONES DISPERSION/REPULSION */
508 rinvsix = _mm256_mul_ps(_mm256_mul_ps(rinvsq00,rinvsq00),rinvsq00);
509 vvdw6 = _mm256_mul_ps(c6_00,rinvsix);
510 vvdw12 = _mm256_mul_ps(c12_00,_mm256_mul_ps(rinvsix,rinvsix));
511 vvdw = _mm256_sub_ps( _mm256_mul_ps(vvdw12,one_twelfth) , _mm256_mul_ps(vvdw6,one_sixth) );
512 fvdw = _mm256_mul_ps(_mm256_sub_ps(vvdw12,vvdw6),rinvsq00);
514 /* Update potential sum for this i atom from the interaction with this j atom. */
515 vvdw = _mm256_andnot_ps(dummy_mask,vvdw);
516 vvdwsum = _mm256_add_ps(vvdwsum,vvdw);
520 fscal = _mm256_andnot_ps(dummy_mask,fscal);
522 /* Calculate temporary vectorial force */
523 tx = _mm256_mul_ps(fscal,dx00);
524 ty = _mm256_mul_ps(fscal,dy00);
525 tz = _mm256_mul_ps(fscal,dz00);
527 /* Update vectorial force */
528 fix0 = _mm256_add_ps(fix0,tx);
529 fiy0 = _mm256_add_ps(fiy0,ty);
530 fiz0 = _mm256_add_ps(fiz0,tz);
532 fjx0 = _mm256_add_ps(fjx0,tx);
533 fjy0 = _mm256_add_ps(fjy0,ty);
534 fjz0 = _mm256_add_ps(fjz0,tz);
536 /**************************
537 * CALCULATE INTERACTIONS *
538 **************************/
540 /* Compute parameters for interactions between i and j atoms */
541 qq10 = _mm256_mul_ps(iq1,jq0);
543 /* REACTION-FIELD ELECTROSTATICS */
544 velec = _mm256_mul_ps(qq10,_mm256_sub_ps(_mm256_add_ps(rinv10,_mm256_mul_ps(krf,rsq10)),crf));
545 felec = _mm256_mul_ps(qq10,_mm256_sub_ps(_mm256_mul_ps(rinv10,rinvsq10),krf2));
547 /* Update potential sum for this i atom from the interaction with this j atom. */
548 velec = _mm256_andnot_ps(dummy_mask,velec);
549 velecsum = _mm256_add_ps(velecsum,velec);
553 fscal = _mm256_andnot_ps(dummy_mask,fscal);
555 /* Calculate temporary vectorial force */
556 tx = _mm256_mul_ps(fscal,dx10);
557 ty = _mm256_mul_ps(fscal,dy10);
558 tz = _mm256_mul_ps(fscal,dz10);
560 /* Update vectorial force */
561 fix1 = _mm256_add_ps(fix1,tx);
562 fiy1 = _mm256_add_ps(fiy1,ty);
563 fiz1 = _mm256_add_ps(fiz1,tz);
565 fjx0 = _mm256_add_ps(fjx0,tx);
566 fjy0 = _mm256_add_ps(fjy0,ty);
567 fjz0 = _mm256_add_ps(fjz0,tz);
569 /**************************
570 * CALCULATE INTERACTIONS *
571 **************************/
573 /* Compute parameters for interactions between i and j atoms */
574 qq20 = _mm256_mul_ps(iq2,jq0);
576 /* REACTION-FIELD ELECTROSTATICS */
577 velec = _mm256_mul_ps(qq20,_mm256_sub_ps(_mm256_add_ps(rinv20,_mm256_mul_ps(krf,rsq20)),crf));
578 felec = _mm256_mul_ps(qq20,_mm256_sub_ps(_mm256_mul_ps(rinv20,rinvsq20),krf2));
580 /* Update potential sum for this i atom from the interaction with this j atom. */
581 velec = _mm256_andnot_ps(dummy_mask,velec);
582 velecsum = _mm256_add_ps(velecsum,velec);
586 fscal = _mm256_andnot_ps(dummy_mask,fscal);
588 /* Calculate temporary vectorial force */
589 tx = _mm256_mul_ps(fscal,dx20);
590 ty = _mm256_mul_ps(fscal,dy20);
591 tz = _mm256_mul_ps(fscal,dz20);
593 /* Update vectorial force */
594 fix2 = _mm256_add_ps(fix2,tx);
595 fiy2 = _mm256_add_ps(fiy2,ty);
596 fiz2 = _mm256_add_ps(fiz2,tz);
598 fjx0 = _mm256_add_ps(fjx0,tx);
599 fjy0 = _mm256_add_ps(fjy0,ty);
600 fjz0 = _mm256_add_ps(fjz0,tz);
602 /**************************
603 * CALCULATE INTERACTIONS *
604 **************************/
606 /* Compute parameters for interactions between i and j atoms */
607 qq30 = _mm256_mul_ps(iq3,jq0);
609 /* REACTION-FIELD ELECTROSTATICS */
610 velec = _mm256_mul_ps(qq30,_mm256_sub_ps(_mm256_add_ps(rinv30,_mm256_mul_ps(krf,rsq30)),crf));
611 felec = _mm256_mul_ps(qq30,_mm256_sub_ps(_mm256_mul_ps(rinv30,rinvsq30),krf2));
613 /* Update potential sum for this i atom from the interaction with this j atom. */
614 velec = _mm256_andnot_ps(dummy_mask,velec);
615 velecsum = _mm256_add_ps(velecsum,velec);
619 fscal = _mm256_andnot_ps(dummy_mask,fscal);
621 /* Calculate temporary vectorial force */
622 tx = _mm256_mul_ps(fscal,dx30);
623 ty = _mm256_mul_ps(fscal,dy30);
624 tz = _mm256_mul_ps(fscal,dz30);
626 /* Update vectorial force */
627 fix3 = _mm256_add_ps(fix3,tx);
628 fiy3 = _mm256_add_ps(fiy3,ty);
629 fiz3 = _mm256_add_ps(fiz3,tz);
631 fjx0 = _mm256_add_ps(fjx0,tx);
632 fjy0 = _mm256_add_ps(fjy0,ty);
633 fjz0 = _mm256_add_ps(fjz0,tz);
635 fjptrA = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
636 fjptrB = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
637 fjptrC = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
638 fjptrD = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
639 fjptrE = (jnrlistE>=0) ? f+j_coord_offsetE : scratch;
640 fjptrF = (jnrlistF>=0) ? f+j_coord_offsetF : scratch;
641 fjptrG = (jnrlistG>=0) ? f+j_coord_offsetG : scratch;
642 fjptrH = (jnrlistH>=0) ? f+j_coord_offsetH : scratch;
644 gmx_mm256_decrement_1rvec_8ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjptrE,fjptrF,fjptrG,fjptrH,fjx0,fjy0,fjz0);
646 /* Inner loop uses 131 flops */
649 /* End of innermost loop */
651 gmx_mm256_update_iforce_4atom_swizzle_ps(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,fix3,fiy3,fiz3,
652 f+i_coord_offset,fshift+i_shift_offset);
655 /* Update potential energies */
656 gmx_mm256_update_1pot_ps(velecsum,kernel_data->energygrp_elec+ggid);
657 gmx_mm256_update_1pot_ps(vvdwsum,kernel_data->energygrp_vdw+ggid);
659 /* Increment number of inner iterations */
660 inneriter += j_index_end - j_index_start;
662 /* Outer loop uses 26 flops */
665 /* Increment number of outer iterations */
668 /* Update outer/inner flops */
670 inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W4_VF,outeriter*26 + inneriter*131);
673 * Gromacs nonbonded kernel: nb_kernel_ElecRF_VdwLJ_GeomW4P1_F_avx_256_single
674 * Electrostatics interaction: ReactionField
675 * VdW interaction: LennardJones
676 * Geometry: Water4-Particle
677 * Calculate force/pot: Force
680 nb_kernel_ElecRF_VdwLJ_GeomW4P1_F_avx_256_single
681 (t_nblist * gmx_restrict nlist,
682 rvec * gmx_restrict xx,
683 rvec * gmx_restrict ff,
684 t_forcerec * gmx_restrict fr,
685 t_mdatoms * gmx_restrict mdatoms,
686 nb_kernel_data_t * gmx_restrict kernel_data,
687 t_nrnb * gmx_restrict nrnb)
689 /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
690 * just 0 for non-waters.
691 * Suffixes A,B,C,D,E,F,G,H refer to j loop unrolling done with AVX, e.g. for the eight different
692 * jnr indices corresponding to data put in the four positions in the SIMD register.
694 int i_shift_offset,i_coord_offset,outeriter,inneriter;
695 int j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
696 int jnrA,jnrB,jnrC,jnrD;
697 int jnrE,jnrF,jnrG,jnrH;
698 int jnrlistA,jnrlistB,jnrlistC,jnrlistD;
699 int jnrlistE,jnrlistF,jnrlistG,jnrlistH;
700 int j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
701 int j_coord_offsetE,j_coord_offsetF,j_coord_offsetG,j_coord_offsetH;
702 int *iinr,*jindex,*jjnr,*shiftidx,*gid;
704 real *shiftvec,*fshift,*x,*f;
705 real *fjptrA,*fjptrB,*fjptrC,*fjptrD,*fjptrE,*fjptrF,*fjptrG,*fjptrH;
707 __m256 tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
708 real * vdwioffsetptr0;
709 __m256 ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
710 real * vdwioffsetptr1;
711 __m256 ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
712 real * vdwioffsetptr2;
713 __m256 ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
714 real * vdwioffsetptr3;
715 __m256 ix3,iy3,iz3,fix3,fiy3,fiz3,iq3,isai3;
716 int vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D,vdwjidx0E,vdwjidx0F,vdwjidx0G,vdwjidx0H;
717 __m256 jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
718 __m256 dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
719 __m256 dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
720 __m256 dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
721 __m256 dx30,dy30,dz30,rsq30,rinv30,rinvsq30,r30,qq30,c6_30,c12_30;
722 __m256 velec,felec,velecsum,facel,crf,krf,krf2;
725 __m256 rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
728 __m256 one_sixth = _mm256_set1_ps(1.0/6.0);
729 __m256 one_twelfth = _mm256_set1_ps(1.0/12.0);
730 __m256 dummy_mask,cutoff_mask;
731 __m256 signbit = _mm256_castsi256_ps( _mm256_set1_epi32(0x80000000) );
732 __m256 one = _mm256_set1_ps(1.0);
733 __m256 two = _mm256_set1_ps(2.0);
739 jindex = nlist->jindex;
741 shiftidx = nlist->shift;
743 shiftvec = fr->shift_vec[0];
744 fshift = fr->fshift[0];
745 facel = _mm256_set1_ps(fr->epsfac);
746 charge = mdatoms->chargeA;
747 krf = _mm256_set1_ps(fr->ic->k_rf);
748 krf2 = _mm256_set1_ps(fr->ic->k_rf*2.0);
749 crf = _mm256_set1_ps(fr->ic->c_rf);
750 nvdwtype = fr->ntype;
752 vdwtype = mdatoms->typeA;
754 /* Setup water-specific parameters */
755 inr = nlist->iinr[0];
756 iq1 = _mm256_mul_ps(facel,_mm256_set1_ps(charge[inr+1]));
757 iq2 = _mm256_mul_ps(facel,_mm256_set1_ps(charge[inr+2]));
758 iq3 = _mm256_mul_ps(facel,_mm256_set1_ps(charge[inr+3]));
759 vdwioffsetptr0 = vdwparam+2*nvdwtype*vdwtype[inr+0];
761 /* Avoid stupid compiler warnings */
762 jnrA = jnrB = jnrC = jnrD = jnrE = jnrF = jnrG = jnrH = 0;
775 for(iidx=0;iidx<4*DIM;iidx++)
780 /* Start outer loop over neighborlists */
781 for(iidx=0; iidx<nri; iidx++)
783 /* Load shift vector for this list */
784 i_shift_offset = DIM*shiftidx[iidx];
786 /* Load limits for loop over neighbors */
787 j_index_start = jindex[iidx];
788 j_index_end = jindex[iidx+1];
790 /* Get outer coordinate index */
792 i_coord_offset = DIM*inr;
794 /* Load i particle coords and add shift vector */
795 gmx_mm256_load_shift_and_4rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,
796 &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2,&ix3,&iy3,&iz3);
798 fix0 = _mm256_setzero_ps();
799 fiy0 = _mm256_setzero_ps();
800 fiz0 = _mm256_setzero_ps();
801 fix1 = _mm256_setzero_ps();
802 fiy1 = _mm256_setzero_ps();
803 fiz1 = _mm256_setzero_ps();
804 fix2 = _mm256_setzero_ps();
805 fiy2 = _mm256_setzero_ps();
806 fiz2 = _mm256_setzero_ps();
807 fix3 = _mm256_setzero_ps();
808 fiy3 = _mm256_setzero_ps();
809 fiz3 = _mm256_setzero_ps();
811 /* Start inner kernel loop */
812 for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+7]>=0; jidx+=8)
815 /* Get j neighbor index, and coordinate index */
824 j_coord_offsetA = DIM*jnrA;
825 j_coord_offsetB = DIM*jnrB;
826 j_coord_offsetC = DIM*jnrC;
827 j_coord_offsetD = DIM*jnrD;
828 j_coord_offsetE = DIM*jnrE;
829 j_coord_offsetF = DIM*jnrF;
830 j_coord_offsetG = DIM*jnrG;
831 j_coord_offsetH = DIM*jnrH;
833 /* load j atom coordinates */
834 gmx_mm256_load_1rvec_8ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
835 x+j_coord_offsetC,x+j_coord_offsetD,
836 x+j_coord_offsetE,x+j_coord_offsetF,
837 x+j_coord_offsetG,x+j_coord_offsetH,
840 /* Calculate displacement vector */
841 dx00 = _mm256_sub_ps(ix0,jx0);
842 dy00 = _mm256_sub_ps(iy0,jy0);
843 dz00 = _mm256_sub_ps(iz0,jz0);
844 dx10 = _mm256_sub_ps(ix1,jx0);
845 dy10 = _mm256_sub_ps(iy1,jy0);
846 dz10 = _mm256_sub_ps(iz1,jz0);
847 dx20 = _mm256_sub_ps(ix2,jx0);
848 dy20 = _mm256_sub_ps(iy2,jy0);
849 dz20 = _mm256_sub_ps(iz2,jz0);
850 dx30 = _mm256_sub_ps(ix3,jx0);
851 dy30 = _mm256_sub_ps(iy3,jy0);
852 dz30 = _mm256_sub_ps(iz3,jz0);
854 /* Calculate squared distance and things based on it */
855 rsq00 = gmx_mm256_calc_rsq_ps(dx00,dy00,dz00);
856 rsq10 = gmx_mm256_calc_rsq_ps(dx10,dy10,dz10);
857 rsq20 = gmx_mm256_calc_rsq_ps(dx20,dy20,dz20);
858 rsq30 = gmx_mm256_calc_rsq_ps(dx30,dy30,dz30);
860 rinv10 = gmx_mm256_invsqrt_ps(rsq10);
861 rinv20 = gmx_mm256_invsqrt_ps(rsq20);
862 rinv30 = gmx_mm256_invsqrt_ps(rsq30);
864 rinvsq00 = gmx_mm256_inv_ps(rsq00);
865 rinvsq10 = _mm256_mul_ps(rinv10,rinv10);
866 rinvsq20 = _mm256_mul_ps(rinv20,rinv20);
867 rinvsq30 = _mm256_mul_ps(rinv30,rinv30);
869 /* Load parameters for j particles */
870 jq0 = gmx_mm256_load_8real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
871 charge+jnrC+0,charge+jnrD+0,
872 charge+jnrE+0,charge+jnrF+0,
873 charge+jnrG+0,charge+jnrH+0);
874 vdwjidx0A = 2*vdwtype[jnrA+0];
875 vdwjidx0B = 2*vdwtype[jnrB+0];
876 vdwjidx0C = 2*vdwtype[jnrC+0];
877 vdwjidx0D = 2*vdwtype[jnrD+0];
878 vdwjidx0E = 2*vdwtype[jnrE+0];
879 vdwjidx0F = 2*vdwtype[jnrF+0];
880 vdwjidx0G = 2*vdwtype[jnrG+0];
881 vdwjidx0H = 2*vdwtype[jnrH+0];
883 fjx0 = _mm256_setzero_ps();
884 fjy0 = _mm256_setzero_ps();
885 fjz0 = _mm256_setzero_ps();
887 /**************************
888 * CALCULATE INTERACTIONS *
889 **************************/
891 /* Compute parameters for interactions between i and j atoms */
892 gmx_mm256_load_8pair_swizzle_ps(vdwioffsetptr0+vdwjidx0A,
893 vdwioffsetptr0+vdwjidx0B,
894 vdwioffsetptr0+vdwjidx0C,
895 vdwioffsetptr0+vdwjidx0D,
896 vdwioffsetptr0+vdwjidx0E,
897 vdwioffsetptr0+vdwjidx0F,
898 vdwioffsetptr0+vdwjidx0G,
899 vdwioffsetptr0+vdwjidx0H,
902 /* LENNARD-JONES DISPERSION/REPULSION */
904 rinvsix = _mm256_mul_ps(_mm256_mul_ps(rinvsq00,rinvsq00),rinvsq00);
905 fvdw = _mm256_mul_ps(_mm256_sub_ps(_mm256_mul_ps(c12_00,rinvsix),c6_00),_mm256_mul_ps(rinvsix,rinvsq00));
909 /* Calculate temporary vectorial force */
910 tx = _mm256_mul_ps(fscal,dx00);
911 ty = _mm256_mul_ps(fscal,dy00);
912 tz = _mm256_mul_ps(fscal,dz00);
914 /* Update vectorial force */
915 fix0 = _mm256_add_ps(fix0,tx);
916 fiy0 = _mm256_add_ps(fiy0,ty);
917 fiz0 = _mm256_add_ps(fiz0,tz);
919 fjx0 = _mm256_add_ps(fjx0,tx);
920 fjy0 = _mm256_add_ps(fjy0,ty);
921 fjz0 = _mm256_add_ps(fjz0,tz);
923 /**************************
924 * CALCULATE INTERACTIONS *
925 **************************/
927 /* Compute parameters for interactions between i and j atoms */
928 qq10 = _mm256_mul_ps(iq1,jq0);
930 /* REACTION-FIELD ELECTROSTATICS */
931 felec = _mm256_mul_ps(qq10,_mm256_sub_ps(_mm256_mul_ps(rinv10,rinvsq10),krf2));
935 /* Calculate temporary vectorial force */
936 tx = _mm256_mul_ps(fscal,dx10);
937 ty = _mm256_mul_ps(fscal,dy10);
938 tz = _mm256_mul_ps(fscal,dz10);
940 /* Update vectorial force */
941 fix1 = _mm256_add_ps(fix1,tx);
942 fiy1 = _mm256_add_ps(fiy1,ty);
943 fiz1 = _mm256_add_ps(fiz1,tz);
945 fjx0 = _mm256_add_ps(fjx0,tx);
946 fjy0 = _mm256_add_ps(fjy0,ty);
947 fjz0 = _mm256_add_ps(fjz0,tz);
949 /**************************
950 * CALCULATE INTERACTIONS *
951 **************************/
953 /* Compute parameters for interactions between i and j atoms */
954 qq20 = _mm256_mul_ps(iq2,jq0);
956 /* REACTION-FIELD ELECTROSTATICS */
957 felec = _mm256_mul_ps(qq20,_mm256_sub_ps(_mm256_mul_ps(rinv20,rinvsq20),krf2));
961 /* Calculate temporary vectorial force */
962 tx = _mm256_mul_ps(fscal,dx20);
963 ty = _mm256_mul_ps(fscal,dy20);
964 tz = _mm256_mul_ps(fscal,dz20);
966 /* Update vectorial force */
967 fix2 = _mm256_add_ps(fix2,tx);
968 fiy2 = _mm256_add_ps(fiy2,ty);
969 fiz2 = _mm256_add_ps(fiz2,tz);
971 fjx0 = _mm256_add_ps(fjx0,tx);
972 fjy0 = _mm256_add_ps(fjy0,ty);
973 fjz0 = _mm256_add_ps(fjz0,tz);
975 /**************************
976 * CALCULATE INTERACTIONS *
977 **************************/
979 /* Compute parameters for interactions between i and j atoms */
980 qq30 = _mm256_mul_ps(iq3,jq0);
982 /* REACTION-FIELD ELECTROSTATICS */
983 felec = _mm256_mul_ps(qq30,_mm256_sub_ps(_mm256_mul_ps(rinv30,rinvsq30),krf2));
987 /* Calculate temporary vectorial force */
988 tx = _mm256_mul_ps(fscal,dx30);
989 ty = _mm256_mul_ps(fscal,dy30);
990 tz = _mm256_mul_ps(fscal,dz30);
992 /* Update vectorial force */
993 fix3 = _mm256_add_ps(fix3,tx);
994 fiy3 = _mm256_add_ps(fiy3,ty);
995 fiz3 = _mm256_add_ps(fiz3,tz);
997 fjx0 = _mm256_add_ps(fjx0,tx);
998 fjy0 = _mm256_add_ps(fjy0,ty);
999 fjz0 = _mm256_add_ps(fjz0,tz);
1001 fjptrA = f+j_coord_offsetA;
1002 fjptrB = f+j_coord_offsetB;
1003 fjptrC = f+j_coord_offsetC;
1004 fjptrD = f+j_coord_offsetD;
1005 fjptrE = f+j_coord_offsetE;
1006 fjptrF = f+j_coord_offsetF;
1007 fjptrG = f+j_coord_offsetG;
1008 fjptrH = f+j_coord_offsetH;
1010 gmx_mm256_decrement_1rvec_8ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjptrE,fjptrF,fjptrG,fjptrH,fjx0,fjy0,fjz0);
1012 /* Inner loop uses 111 flops */
1015 if(jidx<j_index_end)
1018 /* Get j neighbor index, and coordinate index */
1019 jnrlistA = jjnr[jidx];
1020 jnrlistB = jjnr[jidx+1];
1021 jnrlistC = jjnr[jidx+2];
1022 jnrlistD = jjnr[jidx+3];
1023 jnrlistE = jjnr[jidx+4];
1024 jnrlistF = jjnr[jidx+5];
1025 jnrlistG = jjnr[jidx+6];
1026 jnrlistH = jjnr[jidx+7];
1027 /* Sign of each element will be negative for non-real atoms.
1028 * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
1029 * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
1031 dummy_mask = gmx_mm256_set_m128(gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx+4)),_mm_setzero_si128())),
1032 gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128())));
1034 jnrA = (jnrlistA>=0) ? jnrlistA : 0;
1035 jnrB = (jnrlistB>=0) ? jnrlistB : 0;
1036 jnrC = (jnrlistC>=0) ? jnrlistC : 0;
1037 jnrD = (jnrlistD>=0) ? jnrlistD : 0;
1038 jnrE = (jnrlistE>=0) ? jnrlistE : 0;
1039 jnrF = (jnrlistF>=0) ? jnrlistF : 0;
1040 jnrG = (jnrlistG>=0) ? jnrlistG : 0;
1041 jnrH = (jnrlistH>=0) ? jnrlistH : 0;
1042 j_coord_offsetA = DIM*jnrA;
1043 j_coord_offsetB = DIM*jnrB;
1044 j_coord_offsetC = DIM*jnrC;
1045 j_coord_offsetD = DIM*jnrD;
1046 j_coord_offsetE = DIM*jnrE;
1047 j_coord_offsetF = DIM*jnrF;
1048 j_coord_offsetG = DIM*jnrG;
1049 j_coord_offsetH = DIM*jnrH;
1051 /* load j atom coordinates */
1052 gmx_mm256_load_1rvec_8ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
1053 x+j_coord_offsetC,x+j_coord_offsetD,
1054 x+j_coord_offsetE,x+j_coord_offsetF,
1055 x+j_coord_offsetG,x+j_coord_offsetH,
1058 /* Calculate displacement vector */
1059 dx00 = _mm256_sub_ps(ix0,jx0);
1060 dy00 = _mm256_sub_ps(iy0,jy0);
1061 dz00 = _mm256_sub_ps(iz0,jz0);
1062 dx10 = _mm256_sub_ps(ix1,jx0);
1063 dy10 = _mm256_sub_ps(iy1,jy0);
1064 dz10 = _mm256_sub_ps(iz1,jz0);
1065 dx20 = _mm256_sub_ps(ix2,jx0);
1066 dy20 = _mm256_sub_ps(iy2,jy0);
1067 dz20 = _mm256_sub_ps(iz2,jz0);
1068 dx30 = _mm256_sub_ps(ix3,jx0);
1069 dy30 = _mm256_sub_ps(iy3,jy0);
1070 dz30 = _mm256_sub_ps(iz3,jz0);
1072 /* Calculate squared distance and things based on it */
1073 rsq00 = gmx_mm256_calc_rsq_ps(dx00,dy00,dz00);
1074 rsq10 = gmx_mm256_calc_rsq_ps(dx10,dy10,dz10);
1075 rsq20 = gmx_mm256_calc_rsq_ps(dx20,dy20,dz20);
1076 rsq30 = gmx_mm256_calc_rsq_ps(dx30,dy30,dz30);
1078 rinv10 = gmx_mm256_invsqrt_ps(rsq10);
1079 rinv20 = gmx_mm256_invsqrt_ps(rsq20);
1080 rinv30 = gmx_mm256_invsqrt_ps(rsq30);
1082 rinvsq00 = gmx_mm256_inv_ps(rsq00);
1083 rinvsq10 = _mm256_mul_ps(rinv10,rinv10);
1084 rinvsq20 = _mm256_mul_ps(rinv20,rinv20);
1085 rinvsq30 = _mm256_mul_ps(rinv30,rinv30);
1087 /* Load parameters for j particles */
1088 jq0 = gmx_mm256_load_8real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
1089 charge+jnrC+0,charge+jnrD+0,
1090 charge+jnrE+0,charge+jnrF+0,
1091 charge+jnrG+0,charge+jnrH+0);
1092 vdwjidx0A = 2*vdwtype[jnrA+0];
1093 vdwjidx0B = 2*vdwtype[jnrB+0];
1094 vdwjidx0C = 2*vdwtype[jnrC+0];
1095 vdwjidx0D = 2*vdwtype[jnrD+0];
1096 vdwjidx0E = 2*vdwtype[jnrE+0];
1097 vdwjidx0F = 2*vdwtype[jnrF+0];
1098 vdwjidx0G = 2*vdwtype[jnrG+0];
1099 vdwjidx0H = 2*vdwtype[jnrH+0];
1101 fjx0 = _mm256_setzero_ps();
1102 fjy0 = _mm256_setzero_ps();
1103 fjz0 = _mm256_setzero_ps();
1105 /**************************
1106 * CALCULATE INTERACTIONS *
1107 **************************/
1109 /* Compute parameters for interactions between i and j atoms */
1110 gmx_mm256_load_8pair_swizzle_ps(vdwioffsetptr0+vdwjidx0A,
1111 vdwioffsetptr0+vdwjidx0B,
1112 vdwioffsetptr0+vdwjidx0C,
1113 vdwioffsetptr0+vdwjidx0D,
1114 vdwioffsetptr0+vdwjidx0E,
1115 vdwioffsetptr0+vdwjidx0F,
1116 vdwioffsetptr0+vdwjidx0G,
1117 vdwioffsetptr0+vdwjidx0H,
1120 /* LENNARD-JONES DISPERSION/REPULSION */
1122 rinvsix = _mm256_mul_ps(_mm256_mul_ps(rinvsq00,rinvsq00),rinvsq00);
1123 fvdw = _mm256_mul_ps(_mm256_sub_ps(_mm256_mul_ps(c12_00,rinvsix),c6_00),_mm256_mul_ps(rinvsix,rinvsq00));
1127 fscal = _mm256_andnot_ps(dummy_mask,fscal);
1129 /* Calculate temporary vectorial force */
1130 tx = _mm256_mul_ps(fscal,dx00);
1131 ty = _mm256_mul_ps(fscal,dy00);
1132 tz = _mm256_mul_ps(fscal,dz00);
1134 /* Update vectorial force */
1135 fix0 = _mm256_add_ps(fix0,tx);
1136 fiy0 = _mm256_add_ps(fiy0,ty);
1137 fiz0 = _mm256_add_ps(fiz0,tz);
1139 fjx0 = _mm256_add_ps(fjx0,tx);
1140 fjy0 = _mm256_add_ps(fjy0,ty);
1141 fjz0 = _mm256_add_ps(fjz0,tz);
1143 /**************************
1144 * CALCULATE INTERACTIONS *
1145 **************************/
1147 /* Compute parameters for interactions between i and j atoms */
1148 qq10 = _mm256_mul_ps(iq1,jq0);
1150 /* REACTION-FIELD ELECTROSTATICS */
1151 felec = _mm256_mul_ps(qq10,_mm256_sub_ps(_mm256_mul_ps(rinv10,rinvsq10),krf2));
1155 fscal = _mm256_andnot_ps(dummy_mask,fscal);
1157 /* Calculate temporary vectorial force */
1158 tx = _mm256_mul_ps(fscal,dx10);
1159 ty = _mm256_mul_ps(fscal,dy10);
1160 tz = _mm256_mul_ps(fscal,dz10);
1162 /* Update vectorial force */
1163 fix1 = _mm256_add_ps(fix1,tx);
1164 fiy1 = _mm256_add_ps(fiy1,ty);
1165 fiz1 = _mm256_add_ps(fiz1,tz);
1167 fjx0 = _mm256_add_ps(fjx0,tx);
1168 fjy0 = _mm256_add_ps(fjy0,ty);
1169 fjz0 = _mm256_add_ps(fjz0,tz);
1171 /**************************
1172 * CALCULATE INTERACTIONS *
1173 **************************/
1175 /* Compute parameters for interactions between i and j atoms */
1176 qq20 = _mm256_mul_ps(iq2,jq0);
1178 /* REACTION-FIELD ELECTROSTATICS */
1179 felec = _mm256_mul_ps(qq20,_mm256_sub_ps(_mm256_mul_ps(rinv20,rinvsq20),krf2));
1183 fscal = _mm256_andnot_ps(dummy_mask,fscal);
1185 /* Calculate temporary vectorial force */
1186 tx = _mm256_mul_ps(fscal,dx20);
1187 ty = _mm256_mul_ps(fscal,dy20);
1188 tz = _mm256_mul_ps(fscal,dz20);
1190 /* Update vectorial force */
1191 fix2 = _mm256_add_ps(fix2,tx);
1192 fiy2 = _mm256_add_ps(fiy2,ty);
1193 fiz2 = _mm256_add_ps(fiz2,tz);
1195 fjx0 = _mm256_add_ps(fjx0,tx);
1196 fjy0 = _mm256_add_ps(fjy0,ty);
1197 fjz0 = _mm256_add_ps(fjz0,tz);
1199 /**************************
1200 * CALCULATE INTERACTIONS *
1201 **************************/
1203 /* Compute parameters for interactions between i and j atoms */
1204 qq30 = _mm256_mul_ps(iq3,jq0);
1206 /* REACTION-FIELD ELECTROSTATICS */
1207 felec = _mm256_mul_ps(qq30,_mm256_sub_ps(_mm256_mul_ps(rinv30,rinvsq30),krf2));
1211 fscal = _mm256_andnot_ps(dummy_mask,fscal);
1213 /* Calculate temporary vectorial force */
1214 tx = _mm256_mul_ps(fscal,dx30);
1215 ty = _mm256_mul_ps(fscal,dy30);
1216 tz = _mm256_mul_ps(fscal,dz30);
1218 /* Update vectorial force */
1219 fix3 = _mm256_add_ps(fix3,tx);
1220 fiy3 = _mm256_add_ps(fiy3,ty);
1221 fiz3 = _mm256_add_ps(fiz3,tz);
1223 fjx0 = _mm256_add_ps(fjx0,tx);
1224 fjy0 = _mm256_add_ps(fjy0,ty);
1225 fjz0 = _mm256_add_ps(fjz0,tz);
1227 fjptrA = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
1228 fjptrB = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
1229 fjptrC = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
1230 fjptrD = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
1231 fjptrE = (jnrlistE>=0) ? f+j_coord_offsetE : scratch;
1232 fjptrF = (jnrlistF>=0) ? f+j_coord_offsetF : scratch;
1233 fjptrG = (jnrlistG>=0) ? f+j_coord_offsetG : scratch;
1234 fjptrH = (jnrlistH>=0) ? f+j_coord_offsetH : scratch;
1236 gmx_mm256_decrement_1rvec_8ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjptrE,fjptrF,fjptrG,fjptrH,fjx0,fjy0,fjz0);
1238 /* Inner loop uses 111 flops */
1241 /* End of innermost loop */
1243 gmx_mm256_update_iforce_4atom_swizzle_ps(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,fix3,fiy3,fiz3,
1244 f+i_coord_offset,fshift+i_shift_offset);
1246 /* Increment number of inner iterations */
1247 inneriter += j_index_end - j_index_start;
1249 /* Outer loop uses 24 flops */
1252 /* Increment number of outer iterations */
1255 /* Update outer/inner flops */
1257 inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W4_F,outeriter*24 + inneriter*111);