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_GeomW3W3_VF_avx_256_single
38 * Electrostatics interaction: ReactionField
39 * VdW interaction: LennardJones
40 * Geometry: Water3-Water3
41 * Calculate force/pot: PotentialAndForce
44 nb_kernel_ElecRF_VdwLJ_GeomW3W3_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 int vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D,vdwjidx0E,vdwjidx0F,vdwjidx0G,vdwjidx0H;
79 __m256 jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
80 int vdwjidx1A,vdwjidx1B,vdwjidx1C,vdwjidx1D,vdwjidx1E,vdwjidx1F,vdwjidx1G,vdwjidx1H;
81 __m256 jx1,jy1,jz1,fjx1,fjy1,fjz1,jq1,isaj1;
82 int vdwjidx2A,vdwjidx2B,vdwjidx2C,vdwjidx2D,vdwjidx2E,vdwjidx2F,vdwjidx2G,vdwjidx2H;
83 __m256 jx2,jy2,jz2,fjx2,fjy2,fjz2,jq2,isaj2;
84 __m256 dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
85 __m256 dx01,dy01,dz01,rsq01,rinv01,rinvsq01,r01,qq01,c6_01,c12_01;
86 __m256 dx02,dy02,dz02,rsq02,rinv02,rinvsq02,r02,qq02,c6_02,c12_02;
87 __m256 dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
88 __m256 dx11,dy11,dz11,rsq11,rinv11,rinvsq11,r11,qq11,c6_11,c12_11;
89 __m256 dx12,dy12,dz12,rsq12,rinv12,rinvsq12,r12,qq12,c6_12,c12_12;
90 __m256 dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
91 __m256 dx21,dy21,dz21,rsq21,rinv21,rinvsq21,r21,qq21,c6_21,c12_21;
92 __m256 dx22,dy22,dz22,rsq22,rinv22,rinvsq22,r22,qq22,c6_22,c12_22;
93 __m256 velec,felec,velecsum,facel,crf,krf,krf2;
96 __m256 rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
99 __m256 one_sixth = _mm256_set1_ps(1.0/6.0);
100 __m256 one_twelfth = _mm256_set1_ps(1.0/12.0);
101 __m256 dummy_mask,cutoff_mask;
102 __m256 signbit = _mm256_castsi256_ps( _mm256_set1_epi32(0x80000000) );
103 __m256 one = _mm256_set1_ps(1.0);
104 __m256 two = _mm256_set1_ps(2.0);
110 jindex = nlist->jindex;
112 shiftidx = nlist->shift;
114 shiftvec = fr->shift_vec[0];
115 fshift = fr->fshift[0];
116 facel = _mm256_set1_ps(fr->epsfac);
117 charge = mdatoms->chargeA;
118 krf = _mm256_set1_ps(fr->ic->k_rf);
119 krf2 = _mm256_set1_ps(fr->ic->k_rf*2.0);
120 crf = _mm256_set1_ps(fr->ic->c_rf);
121 nvdwtype = fr->ntype;
123 vdwtype = mdatoms->typeA;
125 /* Setup water-specific parameters */
126 inr = nlist->iinr[0];
127 iq0 = _mm256_mul_ps(facel,_mm256_set1_ps(charge[inr+0]));
128 iq1 = _mm256_mul_ps(facel,_mm256_set1_ps(charge[inr+1]));
129 iq2 = _mm256_mul_ps(facel,_mm256_set1_ps(charge[inr+2]));
130 vdwioffsetptr0 = vdwparam+2*nvdwtype*vdwtype[inr+0];
132 jq0 = _mm256_set1_ps(charge[inr+0]);
133 jq1 = _mm256_set1_ps(charge[inr+1]);
134 jq2 = _mm256_set1_ps(charge[inr+2]);
135 vdwjidx0A = 2*vdwtype[inr+0];
136 qq00 = _mm256_mul_ps(iq0,jq0);
137 c6_00 = _mm256_set1_ps(vdwioffsetptr0[vdwjidx0A]);
138 c12_00 = _mm256_set1_ps(vdwioffsetptr0[vdwjidx0A+1]);
139 qq01 = _mm256_mul_ps(iq0,jq1);
140 qq02 = _mm256_mul_ps(iq0,jq2);
141 qq10 = _mm256_mul_ps(iq1,jq0);
142 qq11 = _mm256_mul_ps(iq1,jq1);
143 qq12 = _mm256_mul_ps(iq1,jq2);
144 qq20 = _mm256_mul_ps(iq2,jq0);
145 qq21 = _mm256_mul_ps(iq2,jq1);
146 qq22 = _mm256_mul_ps(iq2,jq2);
148 /* Avoid stupid compiler warnings */
149 jnrA = jnrB = jnrC = jnrD = jnrE = jnrF = jnrG = jnrH = 0;
162 for(iidx=0;iidx<4*DIM;iidx++)
167 /* Start outer loop over neighborlists */
168 for(iidx=0; iidx<nri; iidx++)
170 /* Load shift vector for this list */
171 i_shift_offset = DIM*shiftidx[iidx];
173 /* Load limits for loop over neighbors */
174 j_index_start = jindex[iidx];
175 j_index_end = jindex[iidx+1];
177 /* Get outer coordinate index */
179 i_coord_offset = DIM*inr;
181 /* Load i particle coords and add shift vector */
182 gmx_mm256_load_shift_and_3rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,
183 &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2);
185 fix0 = _mm256_setzero_ps();
186 fiy0 = _mm256_setzero_ps();
187 fiz0 = _mm256_setzero_ps();
188 fix1 = _mm256_setzero_ps();
189 fiy1 = _mm256_setzero_ps();
190 fiz1 = _mm256_setzero_ps();
191 fix2 = _mm256_setzero_ps();
192 fiy2 = _mm256_setzero_ps();
193 fiz2 = _mm256_setzero_ps();
195 /* Reset potential sums */
196 velecsum = _mm256_setzero_ps();
197 vvdwsum = _mm256_setzero_ps();
199 /* Start inner kernel loop */
200 for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+7]>=0; jidx+=8)
203 /* Get j neighbor index, and coordinate index */
212 j_coord_offsetA = DIM*jnrA;
213 j_coord_offsetB = DIM*jnrB;
214 j_coord_offsetC = DIM*jnrC;
215 j_coord_offsetD = DIM*jnrD;
216 j_coord_offsetE = DIM*jnrE;
217 j_coord_offsetF = DIM*jnrF;
218 j_coord_offsetG = DIM*jnrG;
219 j_coord_offsetH = DIM*jnrH;
221 /* load j atom coordinates */
222 gmx_mm256_load_3rvec_8ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
223 x+j_coord_offsetC,x+j_coord_offsetD,
224 x+j_coord_offsetE,x+j_coord_offsetF,
225 x+j_coord_offsetG,x+j_coord_offsetH,
226 &jx0,&jy0,&jz0,&jx1,&jy1,&jz1,&jx2,&jy2,&jz2);
228 /* Calculate displacement vector */
229 dx00 = _mm256_sub_ps(ix0,jx0);
230 dy00 = _mm256_sub_ps(iy0,jy0);
231 dz00 = _mm256_sub_ps(iz0,jz0);
232 dx01 = _mm256_sub_ps(ix0,jx1);
233 dy01 = _mm256_sub_ps(iy0,jy1);
234 dz01 = _mm256_sub_ps(iz0,jz1);
235 dx02 = _mm256_sub_ps(ix0,jx2);
236 dy02 = _mm256_sub_ps(iy0,jy2);
237 dz02 = _mm256_sub_ps(iz0,jz2);
238 dx10 = _mm256_sub_ps(ix1,jx0);
239 dy10 = _mm256_sub_ps(iy1,jy0);
240 dz10 = _mm256_sub_ps(iz1,jz0);
241 dx11 = _mm256_sub_ps(ix1,jx1);
242 dy11 = _mm256_sub_ps(iy1,jy1);
243 dz11 = _mm256_sub_ps(iz1,jz1);
244 dx12 = _mm256_sub_ps(ix1,jx2);
245 dy12 = _mm256_sub_ps(iy1,jy2);
246 dz12 = _mm256_sub_ps(iz1,jz2);
247 dx20 = _mm256_sub_ps(ix2,jx0);
248 dy20 = _mm256_sub_ps(iy2,jy0);
249 dz20 = _mm256_sub_ps(iz2,jz0);
250 dx21 = _mm256_sub_ps(ix2,jx1);
251 dy21 = _mm256_sub_ps(iy2,jy1);
252 dz21 = _mm256_sub_ps(iz2,jz1);
253 dx22 = _mm256_sub_ps(ix2,jx2);
254 dy22 = _mm256_sub_ps(iy2,jy2);
255 dz22 = _mm256_sub_ps(iz2,jz2);
257 /* Calculate squared distance and things based on it */
258 rsq00 = gmx_mm256_calc_rsq_ps(dx00,dy00,dz00);
259 rsq01 = gmx_mm256_calc_rsq_ps(dx01,dy01,dz01);
260 rsq02 = gmx_mm256_calc_rsq_ps(dx02,dy02,dz02);
261 rsq10 = gmx_mm256_calc_rsq_ps(dx10,dy10,dz10);
262 rsq11 = gmx_mm256_calc_rsq_ps(dx11,dy11,dz11);
263 rsq12 = gmx_mm256_calc_rsq_ps(dx12,dy12,dz12);
264 rsq20 = gmx_mm256_calc_rsq_ps(dx20,dy20,dz20);
265 rsq21 = gmx_mm256_calc_rsq_ps(dx21,dy21,dz21);
266 rsq22 = gmx_mm256_calc_rsq_ps(dx22,dy22,dz22);
268 rinv00 = gmx_mm256_invsqrt_ps(rsq00);
269 rinv01 = gmx_mm256_invsqrt_ps(rsq01);
270 rinv02 = gmx_mm256_invsqrt_ps(rsq02);
271 rinv10 = gmx_mm256_invsqrt_ps(rsq10);
272 rinv11 = gmx_mm256_invsqrt_ps(rsq11);
273 rinv12 = gmx_mm256_invsqrt_ps(rsq12);
274 rinv20 = gmx_mm256_invsqrt_ps(rsq20);
275 rinv21 = gmx_mm256_invsqrt_ps(rsq21);
276 rinv22 = gmx_mm256_invsqrt_ps(rsq22);
278 rinvsq00 = _mm256_mul_ps(rinv00,rinv00);
279 rinvsq01 = _mm256_mul_ps(rinv01,rinv01);
280 rinvsq02 = _mm256_mul_ps(rinv02,rinv02);
281 rinvsq10 = _mm256_mul_ps(rinv10,rinv10);
282 rinvsq11 = _mm256_mul_ps(rinv11,rinv11);
283 rinvsq12 = _mm256_mul_ps(rinv12,rinv12);
284 rinvsq20 = _mm256_mul_ps(rinv20,rinv20);
285 rinvsq21 = _mm256_mul_ps(rinv21,rinv21);
286 rinvsq22 = _mm256_mul_ps(rinv22,rinv22);
288 fjx0 = _mm256_setzero_ps();
289 fjy0 = _mm256_setzero_ps();
290 fjz0 = _mm256_setzero_ps();
291 fjx1 = _mm256_setzero_ps();
292 fjy1 = _mm256_setzero_ps();
293 fjz1 = _mm256_setzero_ps();
294 fjx2 = _mm256_setzero_ps();
295 fjy2 = _mm256_setzero_ps();
296 fjz2 = _mm256_setzero_ps();
298 /**************************
299 * CALCULATE INTERACTIONS *
300 **************************/
302 /* REACTION-FIELD ELECTROSTATICS */
303 velec = _mm256_mul_ps(qq00,_mm256_sub_ps(_mm256_add_ps(rinv00,_mm256_mul_ps(krf,rsq00)),crf));
304 felec = _mm256_mul_ps(qq00,_mm256_sub_ps(_mm256_mul_ps(rinv00,rinvsq00),krf2));
306 /* LENNARD-JONES DISPERSION/REPULSION */
308 rinvsix = _mm256_mul_ps(_mm256_mul_ps(rinvsq00,rinvsq00),rinvsq00);
309 vvdw6 = _mm256_mul_ps(c6_00,rinvsix);
310 vvdw12 = _mm256_mul_ps(c12_00,_mm256_mul_ps(rinvsix,rinvsix));
311 vvdw = _mm256_sub_ps( _mm256_mul_ps(vvdw12,one_twelfth) , _mm256_mul_ps(vvdw6,one_sixth) );
312 fvdw = _mm256_mul_ps(_mm256_sub_ps(vvdw12,vvdw6),rinvsq00);
314 /* Update potential sum for this i atom from the interaction with this j atom. */
315 velecsum = _mm256_add_ps(velecsum,velec);
316 vvdwsum = _mm256_add_ps(vvdwsum,vvdw);
318 fscal = _mm256_add_ps(felec,fvdw);
320 /* Calculate temporary vectorial force */
321 tx = _mm256_mul_ps(fscal,dx00);
322 ty = _mm256_mul_ps(fscal,dy00);
323 tz = _mm256_mul_ps(fscal,dz00);
325 /* Update vectorial force */
326 fix0 = _mm256_add_ps(fix0,tx);
327 fiy0 = _mm256_add_ps(fiy0,ty);
328 fiz0 = _mm256_add_ps(fiz0,tz);
330 fjx0 = _mm256_add_ps(fjx0,tx);
331 fjy0 = _mm256_add_ps(fjy0,ty);
332 fjz0 = _mm256_add_ps(fjz0,tz);
334 /**************************
335 * CALCULATE INTERACTIONS *
336 **************************/
338 /* REACTION-FIELD ELECTROSTATICS */
339 velec = _mm256_mul_ps(qq01,_mm256_sub_ps(_mm256_add_ps(rinv01,_mm256_mul_ps(krf,rsq01)),crf));
340 felec = _mm256_mul_ps(qq01,_mm256_sub_ps(_mm256_mul_ps(rinv01,rinvsq01),krf2));
342 /* Update potential sum for this i atom from the interaction with this j atom. */
343 velecsum = _mm256_add_ps(velecsum,velec);
347 /* Calculate temporary vectorial force */
348 tx = _mm256_mul_ps(fscal,dx01);
349 ty = _mm256_mul_ps(fscal,dy01);
350 tz = _mm256_mul_ps(fscal,dz01);
352 /* Update vectorial force */
353 fix0 = _mm256_add_ps(fix0,tx);
354 fiy0 = _mm256_add_ps(fiy0,ty);
355 fiz0 = _mm256_add_ps(fiz0,tz);
357 fjx1 = _mm256_add_ps(fjx1,tx);
358 fjy1 = _mm256_add_ps(fjy1,ty);
359 fjz1 = _mm256_add_ps(fjz1,tz);
361 /**************************
362 * CALCULATE INTERACTIONS *
363 **************************/
365 /* REACTION-FIELD ELECTROSTATICS */
366 velec = _mm256_mul_ps(qq02,_mm256_sub_ps(_mm256_add_ps(rinv02,_mm256_mul_ps(krf,rsq02)),crf));
367 felec = _mm256_mul_ps(qq02,_mm256_sub_ps(_mm256_mul_ps(rinv02,rinvsq02),krf2));
369 /* Update potential sum for this i atom from the interaction with this j atom. */
370 velecsum = _mm256_add_ps(velecsum,velec);
374 /* Calculate temporary vectorial force */
375 tx = _mm256_mul_ps(fscal,dx02);
376 ty = _mm256_mul_ps(fscal,dy02);
377 tz = _mm256_mul_ps(fscal,dz02);
379 /* Update vectorial force */
380 fix0 = _mm256_add_ps(fix0,tx);
381 fiy0 = _mm256_add_ps(fiy0,ty);
382 fiz0 = _mm256_add_ps(fiz0,tz);
384 fjx2 = _mm256_add_ps(fjx2,tx);
385 fjy2 = _mm256_add_ps(fjy2,ty);
386 fjz2 = _mm256_add_ps(fjz2,tz);
388 /**************************
389 * CALCULATE INTERACTIONS *
390 **************************/
392 /* REACTION-FIELD ELECTROSTATICS */
393 velec = _mm256_mul_ps(qq10,_mm256_sub_ps(_mm256_add_ps(rinv10,_mm256_mul_ps(krf,rsq10)),crf));
394 felec = _mm256_mul_ps(qq10,_mm256_sub_ps(_mm256_mul_ps(rinv10,rinvsq10),krf2));
396 /* Update potential sum for this i atom from the interaction with this j atom. */
397 velecsum = _mm256_add_ps(velecsum,velec);
401 /* Calculate temporary vectorial force */
402 tx = _mm256_mul_ps(fscal,dx10);
403 ty = _mm256_mul_ps(fscal,dy10);
404 tz = _mm256_mul_ps(fscal,dz10);
406 /* Update vectorial force */
407 fix1 = _mm256_add_ps(fix1,tx);
408 fiy1 = _mm256_add_ps(fiy1,ty);
409 fiz1 = _mm256_add_ps(fiz1,tz);
411 fjx0 = _mm256_add_ps(fjx0,tx);
412 fjy0 = _mm256_add_ps(fjy0,ty);
413 fjz0 = _mm256_add_ps(fjz0,tz);
415 /**************************
416 * CALCULATE INTERACTIONS *
417 **************************/
419 /* REACTION-FIELD ELECTROSTATICS */
420 velec = _mm256_mul_ps(qq11,_mm256_sub_ps(_mm256_add_ps(rinv11,_mm256_mul_ps(krf,rsq11)),crf));
421 felec = _mm256_mul_ps(qq11,_mm256_sub_ps(_mm256_mul_ps(rinv11,rinvsq11),krf2));
423 /* Update potential sum for this i atom from the interaction with this j atom. */
424 velecsum = _mm256_add_ps(velecsum,velec);
428 /* Calculate temporary vectorial force */
429 tx = _mm256_mul_ps(fscal,dx11);
430 ty = _mm256_mul_ps(fscal,dy11);
431 tz = _mm256_mul_ps(fscal,dz11);
433 /* Update vectorial force */
434 fix1 = _mm256_add_ps(fix1,tx);
435 fiy1 = _mm256_add_ps(fiy1,ty);
436 fiz1 = _mm256_add_ps(fiz1,tz);
438 fjx1 = _mm256_add_ps(fjx1,tx);
439 fjy1 = _mm256_add_ps(fjy1,ty);
440 fjz1 = _mm256_add_ps(fjz1,tz);
442 /**************************
443 * CALCULATE INTERACTIONS *
444 **************************/
446 /* REACTION-FIELD ELECTROSTATICS */
447 velec = _mm256_mul_ps(qq12,_mm256_sub_ps(_mm256_add_ps(rinv12,_mm256_mul_ps(krf,rsq12)),crf));
448 felec = _mm256_mul_ps(qq12,_mm256_sub_ps(_mm256_mul_ps(rinv12,rinvsq12),krf2));
450 /* Update potential sum for this i atom from the interaction with this j atom. */
451 velecsum = _mm256_add_ps(velecsum,velec);
455 /* Calculate temporary vectorial force */
456 tx = _mm256_mul_ps(fscal,dx12);
457 ty = _mm256_mul_ps(fscal,dy12);
458 tz = _mm256_mul_ps(fscal,dz12);
460 /* Update vectorial force */
461 fix1 = _mm256_add_ps(fix1,tx);
462 fiy1 = _mm256_add_ps(fiy1,ty);
463 fiz1 = _mm256_add_ps(fiz1,tz);
465 fjx2 = _mm256_add_ps(fjx2,tx);
466 fjy2 = _mm256_add_ps(fjy2,ty);
467 fjz2 = _mm256_add_ps(fjz2,tz);
469 /**************************
470 * CALCULATE INTERACTIONS *
471 **************************/
473 /* REACTION-FIELD ELECTROSTATICS */
474 velec = _mm256_mul_ps(qq20,_mm256_sub_ps(_mm256_add_ps(rinv20,_mm256_mul_ps(krf,rsq20)),crf));
475 felec = _mm256_mul_ps(qq20,_mm256_sub_ps(_mm256_mul_ps(rinv20,rinvsq20),krf2));
477 /* Update potential sum for this i atom from the interaction with this j atom. */
478 velecsum = _mm256_add_ps(velecsum,velec);
482 /* Calculate temporary vectorial force */
483 tx = _mm256_mul_ps(fscal,dx20);
484 ty = _mm256_mul_ps(fscal,dy20);
485 tz = _mm256_mul_ps(fscal,dz20);
487 /* Update vectorial force */
488 fix2 = _mm256_add_ps(fix2,tx);
489 fiy2 = _mm256_add_ps(fiy2,ty);
490 fiz2 = _mm256_add_ps(fiz2,tz);
492 fjx0 = _mm256_add_ps(fjx0,tx);
493 fjy0 = _mm256_add_ps(fjy0,ty);
494 fjz0 = _mm256_add_ps(fjz0,tz);
496 /**************************
497 * CALCULATE INTERACTIONS *
498 **************************/
500 /* REACTION-FIELD ELECTROSTATICS */
501 velec = _mm256_mul_ps(qq21,_mm256_sub_ps(_mm256_add_ps(rinv21,_mm256_mul_ps(krf,rsq21)),crf));
502 felec = _mm256_mul_ps(qq21,_mm256_sub_ps(_mm256_mul_ps(rinv21,rinvsq21),krf2));
504 /* Update potential sum for this i atom from the interaction with this j atom. */
505 velecsum = _mm256_add_ps(velecsum,velec);
509 /* Calculate temporary vectorial force */
510 tx = _mm256_mul_ps(fscal,dx21);
511 ty = _mm256_mul_ps(fscal,dy21);
512 tz = _mm256_mul_ps(fscal,dz21);
514 /* Update vectorial force */
515 fix2 = _mm256_add_ps(fix2,tx);
516 fiy2 = _mm256_add_ps(fiy2,ty);
517 fiz2 = _mm256_add_ps(fiz2,tz);
519 fjx1 = _mm256_add_ps(fjx1,tx);
520 fjy1 = _mm256_add_ps(fjy1,ty);
521 fjz1 = _mm256_add_ps(fjz1,tz);
523 /**************************
524 * CALCULATE INTERACTIONS *
525 **************************/
527 /* REACTION-FIELD ELECTROSTATICS */
528 velec = _mm256_mul_ps(qq22,_mm256_sub_ps(_mm256_add_ps(rinv22,_mm256_mul_ps(krf,rsq22)),crf));
529 felec = _mm256_mul_ps(qq22,_mm256_sub_ps(_mm256_mul_ps(rinv22,rinvsq22),krf2));
531 /* Update potential sum for this i atom from the interaction with this j atom. */
532 velecsum = _mm256_add_ps(velecsum,velec);
536 /* Calculate temporary vectorial force */
537 tx = _mm256_mul_ps(fscal,dx22);
538 ty = _mm256_mul_ps(fscal,dy22);
539 tz = _mm256_mul_ps(fscal,dz22);
541 /* Update vectorial force */
542 fix2 = _mm256_add_ps(fix2,tx);
543 fiy2 = _mm256_add_ps(fiy2,ty);
544 fiz2 = _mm256_add_ps(fiz2,tz);
546 fjx2 = _mm256_add_ps(fjx2,tx);
547 fjy2 = _mm256_add_ps(fjy2,ty);
548 fjz2 = _mm256_add_ps(fjz2,tz);
550 fjptrA = f+j_coord_offsetA;
551 fjptrB = f+j_coord_offsetB;
552 fjptrC = f+j_coord_offsetC;
553 fjptrD = f+j_coord_offsetD;
554 fjptrE = f+j_coord_offsetE;
555 fjptrF = f+j_coord_offsetF;
556 fjptrG = f+j_coord_offsetG;
557 fjptrH = f+j_coord_offsetH;
559 gmx_mm256_decrement_3rvec_8ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjptrE,fjptrF,fjptrG,fjptrH,
560 fjx0,fjy0,fjz0,fjx1,fjy1,fjz1,fjx2,fjy2,fjz2);
562 /* Inner loop uses 300 flops */
568 /* Get j neighbor index, and coordinate index */
569 jnrlistA = jjnr[jidx];
570 jnrlistB = jjnr[jidx+1];
571 jnrlistC = jjnr[jidx+2];
572 jnrlistD = jjnr[jidx+3];
573 jnrlistE = jjnr[jidx+4];
574 jnrlistF = jjnr[jidx+5];
575 jnrlistG = jjnr[jidx+6];
576 jnrlistH = jjnr[jidx+7];
577 /* Sign of each element will be negative for non-real atoms.
578 * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
579 * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
581 dummy_mask = gmx_mm256_set_m128(gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx+4)),_mm_setzero_si128())),
582 gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128())));
584 jnrA = (jnrlistA>=0) ? jnrlistA : 0;
585 jnrB = (jnrlistB>=0) ? jnrlistB : 0;
586 jnrC = (jnrlistC>=0) ? jnrlistC : 0;
587 jnrD = (jnrlistD>=0) ? jnrlistD : 0;
588 jnrE = (jnrlistE>=0) ? jnrlistE : 0;
589 jnrF = (jnrlistF>=0) ? jnrlistF : 0;
590 jnrG = (jnrlistG>=0) ? jnrlistG : 0;
591 jnrH = (jnrlistH>=0) ? jnrlistH : 0;
592 j_coord_offsetA = DIM*jnrA;
593 j_coord_offsetB = DIM*jnrB;
594 j_coord_offsetC = DIM*jnrC;
595 j_coord_offsetD = DIM*jnrD;
596 j_coord_offsetE = DIM*jnrE;
597 j_coord_offsetF = DIM*jnrF;
598 j_coord_offsetG = DIM*jnrG;
599 j_coord_offsetH = DIM*jnrH;
601 /* load j atom coordinates */
602 gmx_mm256_load_3rvec_8ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
603 x+j_coord_offsetC,x+j_coord_offsetD,
604 x+j_coord_offsetE,x+j_coord_offsetF,
605 x+j_coord_offsetG,x+j_coord_offsetH,
606 &jx0,&jy0,&jz0,&jx1,&jy1,&jz1,&jx2,&jy2,&jz2);
608 /* Calculate displacement vector */
609 dx00 = _mm256_sub_ps(ix0,jx0);
610 dy00 = _mm256_sub_ps(iy0,jy0);
611 dz00 = _mm256_sub_ps(iz0,jz0);
612 dx01 = _mm256_sub_ps(ix0,jx1);
613 dy01 = _mm256_sub_ps(iy0,jy1);
614 dz01 = _mm256_sub_ps(iz0,jz1);
615 dx02 = _mm256_sub_ps(ix0,jx2);
616 dy02 = _mm256_sub_ps(iy0,jy2);
617 dz02 = _mm256_sub_ps(iz0,jz2);
618 dx10 = _mm256_sub_ps(ix1,jx0);
619 dy10 = _mm256_sub_ps(iy1,jy0);
620 dz10 = _mm256_sub_ps(iz1,jz0);
621 dx11 = _mm256_sub_ps(ix1,jx1);
622 dy11 = _mm256_sub_ps(iy1,jy1);
623 dz11 = _mm256_sub_ps(iz1,jz1);
624 dx12 = _mm256_sub_ps(ix1,jx2);
625 dy12 = _mm256_sub_ps(iy1,jy2);
626 dz12 = _mm256_sub_ps(iz1,jz2);
627 dx20 = _mm256_sub_ps(ix2,jx0);
628 dy20 = _mm256_sub_ps(iy2,jy0);
629 dz20 = _mm256_sub_ps(iz2,jz0);
630 dx21 = _mm256_sub_ps(ix2,jx1);
631 dy21 = _mm256_sub_ps(iy2,jy1);
632 dz21 = _mm256_sub_ps(iz2,jz1);
633 dx22 = _mm256_sub_ps(ix2,jx2);
634 dy22 = _mm256_sub_ps(iy2,jy2);
635 dz22 = _mm256_sub_ps(iz2,jz2);
637 /* Calculate squared distance and things based on it */
638 rsq00 = gmx_mm256_calc_rsq_ps(dx00,dy00,dz00);
639 rsq01 = gmx_mm256_calc_rsq_ps(dx01,dy01,dz01);
640 rsq02 = gmx_mm256_calc_rsq_ps(dx02,dy02,dz02);
641 rsq10 = gmx_mm256_calc_rsq_ps(dx10,dy10,dz10);
642 rsq11 = gmx_mm256_calc_rsq_ps(dx11,dy11,dz11);
643 rsq12 = gmx_mm256_calc_rsq_ps(dx12,dy12,dz12);
644 rsq20 = gmx_mm256_calc_rsq_ps(dx20,dy20,dz20);
645 rsq21 = gmx_mm256_calc_rsq_ps(dx21,dy21,dz21);
646 rsq22 = gmx_mm256_calc_rsq_ps(dx22,dy22,dz22);
648 rinv00 = gmx_mm256_invsqrt_ps(rsq00);
649 rinv01 = gmx_mm256_invsqrt_ps(rsq01);
650 rinv02 = gmx_mm256_invsqrt_ps(rsq02);
651 rinv10 = gmx_mm256_invsqrt_ps(rsq10);
652 rinv11 = gmx_mm256_invsqrt_ps(rsq11);
653 rinv12 = gmx_mm256_invsqrt_ps(rsq12);
654 rinv20 = gmx_mm256_invsqrt_ps(rsq20);
655 rinv21 = gmx_mm256_invsqrt_ps(rsq21);
656 rinv22 = gmx_mm256_invsqrt_ps(rsq22);
658 rinvsq00 = _mm256_mul_ps(rinv00,rinv00);
659 rinvsq01 = _mm256_mul_ps(rinv01,rinv01);
660 rinvsq02 = _mm256_mul_ps(rinv02,rinv02);
661 rinvsq10 = _mm256_mul_ps(rinv10,rinv10);
662 rinvsq11 = _mm256_mul_ps(rinv11,rinv11);
663 rinvsq12 = _mm256_mul_ps(rinv12,rinv12);
664 rinvsq20 = _mm256_mul_ps(rinv20,rinv20);
665 rinvsq21 = _mm256_mul_ps(rinv21,rinv21);
666 rinvsq22 = _mm256_mul_ps(rinv22,rinv22);
668 fjx0 = _mm256_setzero_ps();
669 fjy0 = _mm256_setzero_ps();
670 fjz0 = _mm256_setzero_ps();
671 fjx1 = _mm256_setzero_ps();
672 fjy1 = _mm256_setzero_ps();
673 fjz1 = _mm256_setzero_ps();
674 fjx2 = _mm256_setzero_ps();
675 fjy2 = _mm256_setzero_ps();
676 fjz2 = _mm256_setzero_ps();
678 /**************************
679 * CALCULATE INTERACTIONS *
680 **************************/
682 /* REACTION-FIELD ELECTROSTATICS */
683 velec = _mm256_mul_ps(qq00,_mm256_sub_ps(_mm256_add_ps(rinv00,_mm256_mul_ps(krf,rsq00)),crf));
684 felec = _mm256_mul_ps(qq00,_mm256_sub_ps(_mm256_mul_ps(rinv00,rinvsq00),krf2));
686 /* LENNARD-JONES DISPERSION/REPULSION */
688 rinvsix = _mm256_mul_ps(_mm256_mul_ps(rinvsq00,rinvsq00),rinvsq00);
689 vvdw6 = _mm256_mul_ps(c6_00,rinvsix);
690 vvdw12 = _mm256_mul_ps(c12_00,_mm256_mul_ps(rinvsix,rinvsix));
691 vvdw = _mm256_sub_ps( _mm256_mul_ps(vvdw12,one_twelfth) , _mm256_mul_ps(vvdw6,one_sixth) );
692 fvdw = _mm256_mul_ps(_mm256_sub_ps(vvdw12,vvdw6),rinvsq00);
694 /* Update potential sum for this i atom from the interaction with this j atom. */
695 velec = _mm256_andnot_ps(dummy_mask,velec);
696 velecsum = _mm256_add_ps(velecsum,velec);
697 vvdw = _mm256_andnot_ps(dummy_mask,vvdw);
698 vvdwsum = _mm256_add_ps(vvdwsum,vvdw);
700 fscal = _mm256_add_ps(felec,fvdw);
702 fscal = _mm256_andnot_ps(dummy_mask,fscal);
704 /* Calculate temporary vectorial force */
705 tx = _mm256_mul_ps(fscal,dx00);
706 ty = _mm256_mul_ps(fscal,dy00);
707 tz = _mm256_mul_ps(fscal,dz00);
709 /* Update vectorial force */
710 fix0 = _mm256_add_ps(fix0,tx);
711 fiy0 = _mm256_add_ps(fiy0,ty);
712 fiz0 = _mm256_add_ps(fiz0,tz);
714 fjx0 = _mm256_add_ps(fjx0,tx);
715 fjy0 = _mm256_add_ps(fjy0,ty);
716 fjz0 = _mm256_add_ps(fjz0,tz);
718 /**************************
719 * CALCULATE INTERACTIONS *
720 **************************/
722 /* REACTION-FIELD ELECTROSTATICS */
723 velec = _mm256_mul_ps(qq01,_mm256_sub_ps(_mm256_add_ps(rinv01,_mm256_mul_ps(krf,rsq01)),crf));
724 felec = _mm256_mul_ps(qq01,_mm256_sub_ps(_mm256_mul_ps(rinv01,rinvsq01),krf2));
726 /* Update potential sum for this i atom from the interaction with this j atom. */
727 velec = _mm256_andnot_ps(dummy_mask,velec);
728 velecsum = _mm256_add_ps(velecsum,velec);
732 fscal = _mm256_andnot_ps(dummy_mask,fscal);
734 /* Calculate temporary vectorial force */
735 tx = _mm256_mul_ps(fscal,dx01);
736 ty = _mm256_mul_ps(fscal,dy01);
737 tz = _mm256_mul_ps(fscal,dz01);
739 /* Update vectorial force */
740 fix0 = _mm256_add_ps(fix0,tx);
741 fiy0 = _mm256_add_ps(fiy0,ty);
742 fiz0 = _mm256_add_ps(fiz0,tz);
744 fjx1 = _mm256_add_ps(fjx1,tx);
745 fjy1 = _mm256_add_ps(fjy1,ty);
746 fjz1 = _mm256_add_ps(fjz1,tz);
748 /**************************
749 * CALCULATE INTERACTIONS *
750 **************************/
752 /* REACTION-FIELD ELECTROSTATICS */
753 velec = _mm256_mul_ps(qq02,_mm256_sub_ps(_mm256_add_ps(rinv02,_mm256_mul_ps(krf,rsq02)),crf));
754 felec = _mm256_mul_ps(qq02,_mm256_sub_ps(_mm256_mul_ps(rinv02,rinvsq02),krf2));
756 /* Update potential sum for this i atom from the interaction with this j atom. */
757 velec = _mm256_andnot_ps(dummy_mask,velec);
758 velecsum = _mm256_add_ps(velecsum,velec);
762 fscal = _mm256_andnot_ps(dummy_mask,fscal);
764 /* Calculate temporary vectorial force */
765 tx = _mm256_mul_ps(fscal,dx02);
766 ty = _mm256_mul_ps(fscal,dy02);
767 tz = _mm256_mul_ps(fscal,dz02);
769 /* Update vectorial force */
770 fix0 = _mm256_add_ps(fix0,tx);
771 fiy0 = _mm256_add_ps(fiy0,ty);
772 fiz0 = _mm256_add_ps(fiz0,tz);
774 fjx2 = _mm256_add_ps(fjx2,tx);
775 fjy2 = _mm256_add_ps(fjy2,ty);
776 fjz2 = _mm256_add_ps(fjz2,tz);
778 /**************************
779 * CALCULATE INTERACTIONS *
780 **************************/
782 /* REACTION-FIELD ELECTROSTATICS */
783 velec = _mm256_mul_ps(qq10,_mm256_sub_ps(_mm256_add_ps(rinv10,_mm256_mul_ps(krf,rsq10)),crf));
784 felec = _mm256_mul_ps(qq10,_mm256_sub_ps(_mm256_mul_ps(rinv10,rinvsq10),krf2));
786 /* Update potential sum for this i atom from the interaction with this j atom. */
787 velec = _mm256_andnot_ps(dummy_mask,velec);
788 velecsum = _mm256_add_ps(velecsum,velec);
792 fscal = _mm256_andnot_ps(dummy_mask,fscal);
794 /* Calculate temporary vectorial force */
795 tx = _mm256_mul_ps(fscal,dx10);
796 ty = _mm256_mul_ps(fscal,dy10);
797 tz = _mm256_mul_ps(fscal,dz10);
799 /* Update vectorial force */
800 fix1 = _mm256_add_ps(fix1,tx);
801 fiy1 = _mm256_add_ps(fiy1,ty);
802 fiz1 = _mm256_add_ps(fiz1,tz);
804 fjx0 = _mm256_add_ps(fjx0,tx);
805 fjy0 = _mm256_add_ps(fjy0,ty);
806 fjz0 = _mm256_add_ps(fjz0,tz);
808 /**************************
809 * CALCULATE INTERACTIONS *
810 **************************/
812 /* REACTION-FIELD ELECTROSTATICS */
813 velec = _mm256_mul_ps(qq11,_mm256_sub_ps(_mm256_add_ps(rinv11,_mm256_mul_ps(krf,rsq11)),crf));
814 felec = _mm256_mul_ps(qq11,_mm256_sub_ps(_mm256_mul_ps(rinv11,rinvsq11),krf2));
816 /* Update potential sum for this i atom from the interaction with this j atom. */
817 velec = _mm256_andnot_ps(dummy_mask,velec);
818 velecsum = _mm256_add_ps(velecsum,velec);
822 fscal = _mm256_andnot_ps(dummy_mask,fscal);
824 /* Calculate temporary vectorial force */
825 tx = _mm256_mul_ps(fscal,dx11);
826 ty = _mm256_mul_ps(fscal,dy11);
827 tz = _mm256_mul_ps(fscal,dz11);
829 /* Update vectorial force */
830 fix1 = _mm256_add_ps(fix1,tx);
831 fiy1 = _mm256_add_ps(fiy1,ty);
832 fiz1 = _mm256_add_ps(fiz1,tz);
834 fjx1 = _mm256_add_ps(fjx1,tx);
835 fjy1 = _mm256_add_ps(fjy1,ty);
836 fjz1 = _mm256_add_ps(fjz1,tz);
838 /**************************
839 * CALCULATE INTERACTIONS *
840 **************************/
842 /* REACTION-FIELD ELECTROSTATICS */
843 velec = _mm256_mul_ps(qq12,_mm256_sub_ps(_mm256_add_ps(rinv12,_mm256_mul_ps(krf,rsq12)),crf));
844 felec = _mm256_mul_ps(qq12,_mm256_sub_ps(_mm256_mul_ps(rinv12,rinvsq12),krf2));
846 /* Update potential sum for this i atom from the interaction with this j atom. */
847 velec = _mm256_andnot_ps(dummy_mask,velec);
848 velecsum = _mm256_add_ps(velecsum,velec);
852 fscal = _mm256_andnot_ps(dummy_mask,fscal);
854 /* Calculate temporary vectorial force */
855 tx = _mm256_mul_ps(fscal,dx12);
856 ty = _mm256_mul_ps(fscal,dy12);
857 tz = _mm256_mul_ps(fscal,dz12);
859 /* Update vectorial force */
860 fix1 = _mm256_add_ps(fix1,tx);
861 fiy1 = _mm256_add_ps(fiy1,ty);
862 fiz1 = _mm256_add_ps(fiz1,tz);
864 fjx2 = _mm256_add_ps(fjx2,tx);
865 fjy2 = _mm256_add_ps(fjy2,ty);
866 fjz2 = _mm256_add_ps(fjz2,tz);
868 /**************************
869 * CALCULATE INTERACTIONS *
870 **************************/
872 /* REACTION-FIELD ELECTROSTATICS */
873 velec = _mm256_mul_ps(qq20,_mm256_sub_ps(_mm256_add_ps(rinv20,_mm256_mul_ps(krf,rsq20)),crf));
874 felec = _mm256_mul_ps(qq20,_mm256_sub_ps(_mm256_mul_ps(rinv20,rinvsq20),krf2));
876 /* Update potential sum for this i atom from the interaction with this j atom. */
877 velec = _mm256_andnot_ps(dummy_mask,velec);
878 velecsum = _mm256_add_ps(velecsum,velec);
882 fscal = _mm256_andnot_ps(dummy_mask,fscal);
884 /* Calculate temporary vectorial force */
885 tx = _mm256_mul_ps(fscal,dx20);
886 ty = _mm256_mul_ps(fscal,dy20);
887 tz = _mm256_mul_ps(fscal,dz20);
889 /* Update vectorial force */
890 fix2 = _mm256_add_ps(fix2,tx);
891 fiy2 = _mm256_add_ps(fiy2,ty);
892 fiz2 = _mm256_add_ps(fiz2,tz);
894 fjx0 = _mm256_add_ps(fjx0,tx);
895 fjy0 = _mm256_add_ps(fjy0,ty);
896 fjz0 = _mm256_add_ps(fjz0,tz);
898 /**************************
899 * CALCULATE INTERACTIONS *
900 **************************/
902 /* REACTION-FIELD ELECTROSTATICS */
903 velec = _mm256_mul_ps(qq21,_mm256_sub_ps(_mm256_add_ps(rinv21,_mm256_mul_ps(krf,rsq21)),crf));
904 felec = _mm256_mul_ps(qq21,_mm256_sub_ps(_mm256_mul_ps(rinv21,rinvsq21),krf2));
906 /* Update potential sum for this i atom from the interaction with this j atom. */
907 velec = _mm256_andnot_ps(dummy_mask,velec);
908 velecsum = _mm256_add_ps(velecsum,velec);
912 fscal = _mm256_andnot_ps(dummy_mask,fscal);
914 /* Calculate temporary vectorial force */
915 tx = _mm256_mul_ps(fscal,dx21);
916 ty = _mm256_mul_ps(fscal,dy21);
917 tz = _mm256_mul_ps(fscal,dz21);
919 /* Update vectorial force */
920 fix2 = _mm256_add_ps(fix2,tx);
921 fiy2 = _mm256_add_ps(fiy2,ty);
922 fiz2 = _mm256_add_ps(fiz2,tz);
924 fjx1 = _mm256_add_ps(fjx1,tx);
925 fjy1 = _mm256_add_ps(fjy1,ty);
926 fjz1 = _mm256_add_ps(fjz1,tz);
928 /**************************
929 * CALCULATE INTERACTIONS *
930 **************************/
932 /* REACTION-FIELD ELECTROSTATICS */
933 velec = _mm256_mul_ps(qq22,_mm256_sub_ps(_mm256_add_ps(rinv22,_mm256_mul_ps(krf,rsq22)),crf));
934 felec = _mm256_mul_ps(qq22,_mm256_sub_ps(_mm256_mul_ps(rinv22,rinvsq22),krf2));
936 /* Update potential sum for this i atom from the interaction with this j atom. */
937 velec = _mm256_andnot_ps(dummy_mask,velec);
938 velecsum = _mm256_add_ps(velecsum,velec);
942 fscal = _mm256_andnot_ps(dummy_mask,fscal);
944 /* Calculate temporary vectorial force */
945 tx = _mm256_mul_ps(fscal,dx22);
946 ty = _mm256_mul_ps(fscal,dy22);
947 tz = _mm256_mul_ps(fscal,dz22);
949 /* Update vectorial force */
950 fix2 = _mm256_add_ps(fix2,tx);
951 fiy2 = _mm256_add_ps(fiy2,ty);
952 fiz2 = _mm256_add_ps(fiz2,tz);
954 fjx2 = _mm256_add_ps(fjx2,tx);
955 fjy2 = _mm256_add_ps(fjy2,ty);
956 fjz2 = _mm256_add_ps(fjz2,tz);
958 fjptrA = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
959 fjptrB = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
960 fjptrC = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
961 fjptrD = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
962 fjptrE = (jnrlistE>=0) ? f+j_coord_offsetE : scratch;
963 fjptrF = (jnrlistF>=0) ? f+j_coord_offsetF : scratch;
964 fjptrG = (jnrlistG>=0) ? f+j_coord_offsetG : scratch;
965 fjptrH = (jnrlistH>=0) ? f+j_coord_offsetH : scratch;
967 gmx_mm256_decrement_3rvec_8ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjptrE,fjptrF,fjptrG,fjptrH,
968 fjx0,fjy0,fjz0,fjx1,fjy1,fjz1,fjx2,fjy2,fjz2);
970 /* Inner loop uses 300 flops */
973 /* End of innermost loop */
975 gmx_mm256_update_iforce_3atom_swizzle_ps(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,
976 f+i_coord_offset,fshift+i_shift_offset);
979 /* Update potential energies */
980 gmx_mm256_update_1pot_ps(velecsum,kernel_data->energygrp_elec+ggid);
981 gmx_mm256_update_1pot_ps(vvdwsum,kernel_data->energygrp_vdw+ggid);
983 /* Increment number of inner iterations */
984 inneriter += j_index_end - j_index_start;
986 /* Outer loop uses 20 flops */
989 /* Increment number of outer iterations */
992 /* Update outer/inner flops */
994 inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W3W3_VF,outeriter*20 + inneriter*300);
997 * Gromacs nonbonded kernel: nb_kernel_ElecRF_VdwLJ_GeomW3W3_F_avx_256_single
998 * Electrostatics interaction: ReactionField
999 * VdW interaction: LennardJones
1000 * Geometry: Water3-Water3
1001 * Calculate force/pot: Force
1004 nb_kernel_ElecRF_VdwLJ_GeomW3W3_F_avx_256_single
1005 (t_nblist * gmx_restrict nlist,
1006 rvec * gmx_restrict xx,
1007 rvec * gmx_restrict ff,
1008 t_forcerec * gmx_restrict fr,
1009 t_mdatoms * gmx_restrict mdatoms,
1010 nb_kernel_data_t * gmx_restrict kernel_data,
1011 t_nrnb * gmx_restrict nrnb)
1013 /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
1014 * just 0 for non-waters.
1015 * Suffixes A,B,C,D,E,F,G,H refer to j loop unrolling done with AVX, e.g. for the eight different
1016 * jnr indices corresponding to data put in the four positions in the SIMD register.
1018 int i_shift_offset,i_coord_offset,outeriter,inneriter;
1019 int j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
1020 int jnrA,jnrB,jnrC,jnrD;
1021 int jnrE,jnrF,jnrG,jnrH;
1022 int jnrlistA,jnrlistB,jnrlistC,jnrlistD;
1023 int jnrlistE,jnrlistF,jnrlistG,jnrlistH;
1024 int j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
1025 int j_coord_offsetE,j_coord_offsetF,j_coord_offsetG,j_coord_offsetH;
1026 int *iinr,*jindex,*jjnr,*shiftidx,*gid;
1027 real rcutoff_scalar;
1028 real *shiftvec,*fshift,*x,*f;
1029 real *fjptrA,*fjptrB,*fjptrC,*fjptrD,*fjptrE,*fjptrF,*fjptrG,*fjptrH;
1030 real scratch[4*DIM];
1031 __m256 tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
1032 real * vdwioffsetptr0;
1033 __m256 ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
1034 real * vdwioffsetptr1;
1035 __m256 ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
1036 real * vdwioffsetptr2;
1037 __m256 ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
1038 int vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D,vdwjidx0E,vdwjidx0F,vdwjidx0G,vdwjidx0H;
1039 __m256 jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
1040 int vdwjidx1A,vdwjidx1B,vdwjidx1C,vdwjidx1D,vdwjidx1E,vdwjidx1F,vdwjidx1G,vdwjidx1H;
1041 __m256 jx1,jy1,jz1,fjx1,fjy1,fjz1,jq1,isaj1;
1042 int vdwjidx2A,vdwjidx2B,vdwjidx2C,vdwjidx2D,vdwjidx2E,vdwjidx2F,vdwjidx2G,vdwjidx2H;
1043 __m256 jx2,jy2,jz2,fjx2,fjy2,fjz2,jq2,isaj2;
1044 __m256 dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
1045 __m256 dx01,dy01,dz01,rsq01,rinv01,rinvsq01,r01,qq01,c6_01,c12_01;
1046 __m256 dx02,dy02,dz02,rsq02,rinv02,rinvsq02,r02,qq02,c6_02,c12_02;
1047 __m256 dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
1048 __m256 dx11,dy11,dz11,rsq11,rinv11,rinvsq11,r11,qq11,c6_11,c12_11;
1049 __m256 dx12,dy12,dz12,rsq12,rinv12,rinvsq12,r12,qq12,c6_12,c12_12;
1050 __m256 dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
1051 __m256 dx21,dy21,dz21,rsq21,rinv21,rinvsq21,r21,qq21,c6_21,c12_21;
1052 __m256 dx22,dy22,dz22,rsq22,rinv22,rinvsq22,r22,qq22,c6_22,c12_22;
1053 __m256 velec,felec,velecsum,facel,crf,krf,krf2;
1056 __m256 rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
1059 __m256 one_sixth = _mm256_set1_ps(1.0/6.0);
1060 __m256 one_twelfth = _mm256_set1_ps(1.0/12.0);
1061 __m256 dummy_mask,cutoff_mask;
1062 __m256 signbit = _mm256_castsi256_ps( _mm256_set1_epi32(0x80000000) );
1063 __m256 one = _mm256_set1_ps(1.0);
1064 __m256 two = _mm256_set1_ps(2.0);
1070 jindex = nlist->jindex;
1072 shiftidx = nlist->shift;
1074 shiftvec = fr->shift_vec[0];
1075 fshift = fr->fshift[0];
1076 facel = _mm256_set1_ps(fr->epsfac);
1077 charge = mdatoms->chargeA;
1078 krf = _mm256_set1_ps(fr->ic->k_rf);
1079 krf2 = _mm256_set1_ps(fr->ic->k_rf*2.0);
1080 crf = _mm256_set1_ps(fr->ic->c_rf);
1081 nvdwtype = fr->ntype;
1082 vdwparam = fr->nbfp;
1083 vdwtype = mdatoms->typeA;
1085 /* Setup water-specific parameters */
1086 inr = nlist->iinr[0];
1087 iq0 = _mm256_mul_ps(facel,_mm256_set1_ps(charge[inr+0]));
1088 iq1 = _mm256_mul_ps(facel,_mm256_set1_ps(charge[inr+1]));
1089 iq2 = _mm256_mul_ps(facel,_mm256_set1_ps(charge[inr+2]));
1090 vdwioffsetptr0 = vdwparam+2*nvdwtype*vdwtype[inr+0];
1092 jq0 = _mm256_set1_ps(charge[inr+0]);
1093 jq1 = _mm256_set1_ps(charge[inr+1]);
1094 jq2 = _mm256_set1_ps(charge[inr+2]);
1095 vdwjidx0A = 2*vdwtype[inr+0];
1096 qq00 = _mm256_mul_ps(iq0,jq0);
1097 c6_00 = _mm256_set1_ps(vdwioffsetptr0[vdwjidx0A]);
1098 c12_00 = _mm256_set1_ps(vdwioffsetptr0[vdwjidx0A+1]);
1099 qq01 = _mm256_mul_ps(iq0,jq1);
1100 qq02 = _mm256_mul_ps(iq0,jq2);
1101 qq10 = _mm256_mul_ps(iq1,jq0);
1102 qq11 = _mm256_mul_ps(iq1,jq1);
1103 qq12 = _mm256_mul_ps(iq1,jq2);
1104 qq20 = _mm256_mul_ps(iq2,jq0);
1105 qq21 = _mm256_mul_ps(iq2,jq1);
1106 qq22 = _mm256_mul_ps(iq2,jq2);
1108 /* Avoid stupid compiler warnings */
1109 jnrA = jnrB = jnrC = jnrD = jnrE = jnrF = jnrG = jnrH = 0;
1110 j_coord_offsetA = 0;
1111 j_coord_offsetB = 0;
1112 j_coord_offsetC = 0;
1113 j_coord_offsetD = 0;
1114 j_coord_offsetE = 0;
1115 j_coord_offsetF = 0;
1116 j_coord_offsetG = 0;
1117 j_coord_offsetH = 0;
1122 for(iidx=0;iidx<4*DIM;iidx++)
1124 scratch[iidx] = 0.0;
1127 /* Start outer loop over neighborlists */
1128 for(iidx=0; iidx<nri; iidx++)
1130 /* Load shift vector for this list */
1131 i_shift_offset = DIM*shiftidx[iidx];
1133 /* Load limits for loop over neighbors */
1134 j_index_start = jindex[iidx];
1135 j_index_end = jindex[iidx+1];
1137 /* Get outer coordinate index */
1139 i_coord_offset = DIM*inr;
1141 /* Load i particle coords and add shift vector */
1142 gmx_mm256_load_shift_and_3rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,
1143 &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2);
1145 fix0 = _mm256_setzero_ps();
1146 fiy0 = _mm256_setzero_ps();
1147 fiz0 = _mm256_setzero_ps();
1148 fix1 = _mm256_setzero_ps();
1149 fiy1 = _mm256_setzero_ps();
1150 fiz1 = _mm256_setzero_ps();
1151 fix2 = _mm256_setzero_ps();
1152 fiy2 = _mm256_setzero_ps();
1153 fiz2 = _mm256_setzero_ps();
1155 /* Start inner kernel loop */
1156 for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+7]>=0; jidx+=8)
1159 /* Get j neighbor index, and coordinate index */
1161 jnrB = jjnr[jidx+1];
1162 jnrC = jjnr[jidx+2];
1163 jnrD = jjnr[jidx+3];
1164 jnrE = jjnr[jidx+4];
1165 jnrF = jjnr[jidx+5];
1166 jnrG = jjnr[jidx+6];
1167 jnrH = jjnr[jidx+7];
1168 j_coord_offsetA = DIM*jnrA;
1169 j_coord_offsetB = DIM*jnrB;
1170 j_coord_offsetC = DIM*jnrC;
1171 j_coord_offsetD = DIM*jnrD;
1172 j_coord_offsetE = DIM*jnrE;
1173 j_coord_offsetF = DIM*jnrF;
1174 j_coord_offsetG = DIM*jnrG;
1175 j_coord_offsetH = DIM*jnrH;
1177 /* load j atom coordinates */
1178 gmx_mm256_load_3rvec_8ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
1179 x+j_coord_offsetC,x+j_coord_offsetD,
1180 x+j_coord_offsetE,x+j_coord_offsetF,
1181 x+j_coord_offsetG,x+j_coord_offsetH,
1182 &jx0,&jy0,&jz0,&jx1,&jy1,&jz1,&jx2,&jy2,&jz2);
1184 /* Calculate displacement vector */
1185 dx00 = _mm256_sub_ps(ix0,jx0);
1186 dy00 = _mm256_sub_ps(iy0,jy0);
1187 dz00 = _mm256_sub_ps(iz0,jz0);
1188 dx01 = _mm256_sub_ps(ix0,jx1);
1189 dy01 = _mm256_sub_ps(iy0,jy1);
1190 dz01 = _mm256_sub_ps(iz0,jz1);
1191 dx02 = _mm256_sub_ps(ix0,jx2);
1192 dy02 = _mm256_sub_ps(iy0,jy2);
1193 dz02 = _mm256_sub_ps(iz0,jz2);
1194 dx10 = _mm256_sub_ps(ix1,jx0);
1195 dy10 = _mm256_sub_ps(iy1,jy0);
1196 dz10 = _mm256_sub_ps(iz1,jz0);
1197 dx11 = _mm256_sub_ps(ix1,jx1);
1198 dy11 = _mm256_sub_ps(iy1,jy1);
1199 dz11 = _mm256_sub_ps(iz1,jz1);
1200 dx12 = _mm256_sub_ps(ix1,jx2);
1201 dy12 = _mm256_sub_ps(iy1,jy2);
1202 dz12 = _mm256_sub_ps(iz1,jz2);
1203 dx20 = _mm256_sub_ps(ix2,jx0);
1204 dy20 = _mm256_sub_ps(iy2,jy0);
1205 dz20 = _mm256_sub_ps(iz2,jz0);
1206 dx21 = _mm256_sub_ps(ix2,jx1);
1207 dy21 = _mm256_sub_ps(iy2,jy1);
1208 dz21 = _mm256_sub_ps(iz2,jz1);
1209 dx22 = _mm256_sub_ps(ix2,jx2);
1210 dy22 = _mm256_sub_ps(iy2,jy2);
1211 dz22 = _mm256_sub_ps(iz2,jz2);
1213 /* Calculate squared distance and things based on it */
1214 rsq00 = gmx_mm256_calc_rsq_ps(dx00,dy00,dz00);
1215 rsq01 = gmx_mm256_calc_rsq_ps(dx01,dy01,dz01);
1216 rsq02 = gmx_mm256_calc_rsq_ps(dx02,dy02,dz02);
1217 rsq10 = gmx_mm256_calc_rsq_ps(dx10,dy10,dz10);
1218 rsq11 = gmx_mm256_calc_rsq_ps(dx11,dy11,dz11);
1219 rsq12 = gmx_mm256_calc_rsq_ps(dx12,dy12,dz12);
1220 rsq20 = gmx_mm256_calc_rsq_ps(dx20,dy20,dz20);
1221 rsq21 = gmx_mm256_calc_rsq_ps(dx21,dy21,dz21);
1222 rsq22 = gmx_mm256_calc_rsq_ps(dx22,dy22,dz22);
1224 rinv00 = gmx_mm256_invsqrt_ps(rsq00);
1225 rinv01 = gmx_mm256_invsqrt_ps(rsq01);
1226 rinv02 = gmx_mm256_invsqrt_ps(rsq02);
1227 rinv10 = gmx_mm256_invsqrt_ps(rsq10);
1228 rinv11 = gmx_mm256_invsqrt_ps(rsq11);
1229 rinv12 = gmx_mm256_invsqrt_ps(rsq12);
1230 rinv20 = gmx_mm256_invsqrt_ps(rsq20);
1231 rinv21 = gmx_mm256_invsqrt_ps(rsq21);
1232 rinv22 = gmx_mm256_invsqrt_ps(rsq22);
1234 rinvsq00 = _mm256_mul_ps(rinv00,rinv00);
1235 rinvsq01 = _mm256_mul_ps(rinv01,rinv01);
1236 rinvsq02 = _mm256_mul_ps(rinv02,rinv02);
1237 rinvsq10 = _mm256_mul_ps(rinv10,rinv10);
1238 rinvsq11 = _mm256_mul_ps(rinv11,rinv11);
1239 rinvsq12 = _mm256_mul_ps(rinv12,rinv12);
1240 rinvsq20 = _mm256_mul_ps(rinv20,rinv20);
1241 rinvsq21 = _mm256_mul_ps(rinv21,rinv21);
1242 rinvsq22 = _mm256_mul_ps(rinv22,rinv22);
1244 fjx0 = _mm256_setzero_ps();
1245 fjy0 = _mm256_setzero_ps();
1246 fjz0 = _mm256_setzero_ps();
1247 fjx1 = _mm256_setzero_ps();
1248 fjy1 = _mm256_setzero_ps();
1249 fjz1 = _mm256_setzero_ps();
1250 fjx2 = _mm256_setzero_ps();
1251 fjy2 = _mm256_setzero_ps();
1252 fjz2 = _mm256_setzero_ps();
1254 /**************************
1255 * CALCULATE INTERACTIONS *
1256 **************************/
1258 /* REACTION-FIELD ELECTROSTATICS */
1259 felec = _mm256_mul_ps(qq00,_mm256_sub_ps(_mm256_mul_ps(rinv00,rinvsq00),krf2));
1261 /* LENNARD-JONES DISPERSION/REPULSION */
1263 rinvsix = _mm256_mul_ps(_mm256_mul_ps(rinvsq00,rinvsq00),rinvsq00);
1264 fvdw = _mm256_mul_ps(_mm256_sub_ps(_mm256_mul_ps(c12_00,rinvsix),c6_00),_mm256_mul_ps(rinvsix,rinvsq00));
1266 fscal = _mm256_add_ps(felec,fvdw);
1268 /* Calculate temporary vectorial force */
1269 tx = _mm256_mul_ps(fscal,dx00);
1270 ty = _mm256_mul_ps(fscal,dy00);
1271 tz = _mm256_mul_ps(fscal,dz00);
1273 /* Update vectorial force */
1274 fix0 = _mm256_add_ps(fix0,tx);
1275 fiy0 = _mm256_add_ps(fiy0,ty);
1276 fiz0 = _mm256_add_ps(fiz0,tz);
1278 fjx0 = _mm256_add_ps(fjx0,tx);
1279 fjy0 = _mm256_add_ps(fjy0,ty);
1280 fjz0 = _mm256_add_ps(fjz0,tz);
1282 /**************************
1283 * CALCULATE INTERACTIONS *
1284 **************************/
1286 /* REACTION-FIELD ELECTROSTATICS */
1287 felec = _mm256_mul_ps(qq01,_mm256_sub_ps(_mm256_mul_ps(rinv01,rinvsq01),krf2));
1291 /* Calculate temporary vectorial force */
1292 tx = _mm256_mul_ps(fscal,dx01);
1293 ty = _mm256_mul_ps(fscal,dy01);
1294 tz = _mm256_mul_ps(fscal,dz01);
1296 /* Update vectorial force */
1297 fix0 = _mm256_add_ps(fix0,tx);
1298 fiy0 = _mm256_add_ps(fiy0,ty);
1299 fiz0 = _mm256_add_ps(fiz0,tz);
1301 fjx1 = _mm256_add_ps(fjx1,tx);
1302 fjy1 = _mm256_add_ps(fjy1,ty);
1303 fjz1 = _mm256_add_ps(fjz1,tz);
1305 /**************************
1306 * CALCULATE INTERACTIONS *
1307 **************************/
1309 /* REACTION-FIELD ELECTROSTATICS */
1310 felec = _mm256_mul_ps(qq02,_mm256_sub_ps(_mm256_mul_ps(rinv02,rinvsq02),krf2));
1314 /* Calculate temporary vectorial force */
1315 tx = _mm256_mul_ps(fscal,dx02);
1316 ty = _mm256_mul_ps(fscal,dy02);
1317 tz = _mm256_mul_ps(fscal,dz02);
1319 /* Update vectorial force */
1320 fix0 = _mm256_add_ps(fix0,tx);
1321 fiy0 = _mm256_add_ps(fiy0,ty);
1322 fiz0 = _mm256_add_ps(fiz0,tz);
1324 fjx2 = _mm256_add_ps(fjx2,tx);
1325 fjy2 = _mm256_add_ps(fjy2,ty);
1326 fjz2 = _mm256_add_ps(fjz2,tz);
1328 /**************************
1329 * CALCULATE INTERACTIONS *
1330 **************************/
1332 /* REACTION-FIELD ELECTROSTATICS */
1333 felec = _mm256_mul_ps(qq10,_mm256_sub_ps(_mm256_mul_ps(rinv10,rinvsq10),krf2));
1337 /* Calculate temporary vectorial force */
1338 tx = _mm256_mul_ps(fscal,dx10);
1339 ty = _mm256_mul_ps(fscal,dy10);
1340 tz = _mm256_mul_ps(fscal,dz10);
1342 /* Update vectorial force */
1343 fix1 = _mm256_add_ps(fix1,tx);
1344 fiy1 = _mm256_add_ps(fiy1,ty);
1345 fiz1 = _mm256_add_ps(fiz1,tz);
1347 fjx0 = _mm256_add_ps(fjx0,tx);
1348 fjy0 = _mm256_add_ps(fjy0,ty);
1349 fjz0 = _mm256_add_ps(fjz0,tz);
1351 /**************************
1352 * CALCULATE INTERACTIONS *
1353 **************************/
1355 /* REACTION-FIELD ELECTROSTATICS */
1356 felec = _mm256_mul_ps(qq11,_mm256_sub_ps(_mm256_mul_ps(rinv11,rinvsq11),krf2));
1360 /* Calculate temporary vectorial force */
1361 tx = _mm256_mul_ps(fscal,dx11);
1362 ty = _mm256_mul_ps(fscal,dy11);
1363 tz = _mm256_mul_ps(fscal,dz11);
1365 /* Update vectorial force */
1366 fix1 = _mm256_add_ps(fix1,tx);
1367 fiy1 = _mm256_add_ps(fiy1,ty);
1368 fiz1 = _mm256_add_ps(fiz1,tz);
1370 fjx1 = _mm256_add_ps(fjx1,tx);
1371 fjy1 = _mm256_add_ps(fjy1,ty);
1372 fjz1 = _mm256_add_ps(fjz1,tz);
1374 /**************************
1375 * CALCULATE INTERACTIONS *
1376 **************************/
1378 /* REACTION-FIELD ELECTROSTATICS */
1379 felec = _mm256_mul_ps(qq12,_mm256_sub_ps(_mm256_mul_ps(rinv12,rinvsq12),krf2));
1383 /* Calculate temporary vectorial force */
1384 tx = _mm256_mul_ps(fscal,dx12);
1385 ty = _mm256_mul_ps(fscal,dy12);
1386 tz = _mm256_mul_ps(fscal,dz12);
1388 /* Update vectorial force */
1389 fix1 = _mm256_add_ps(fix1,tx);
1390 fiy1 = _mm256_add_ps(fiy1,ty);
1391 fiz1 = _mm256_add_ps(fiz1,tz);
1393 fjx2 = _mm256_add_ps(fjx2,tx);
1394 fjy2 = _mm256_add_ps(fjy2,ty);
1395 fjz2 = _mm256_add_ps(fjz2,tz);
1397 /**************************
1398 * CALCULATE INTERACTIONS *
1399 **************************/
1401 /* REACTION-FIELD ELECTROSTATICS */
1402 felec = _mm256_mul_ps(qq20,_mm256_sub_ps(_mm256_mul_ps(rinv20,rinvsq20),krf2));
1406 /* Calculate temporary vectorial force */
1407 tx = _mm256_mul_ps(fscal,dx20);
1408 ty = _mm256_mul_ps(fscal,dy20);
1409 tz = _mm256_mul_ps(fscal,dz20);
1411 /* Update vectorial force */
1412 fix2 = _mm256_add_ps(fix2,tx);
1413 fiy2 = _mm256_add_ps(fiy2,ty);
1414 fiz2 = _mm256_add_ps(fiz2,tz);
1416 fjx0 = _mm256_add_ps(fjx0,tx);
1417 fjy0 = _mm256_add_ps(fjy0,ty);
1418 fjz0 = _mm256_add_ps(fjz0,tz);
1420 /**************************
1421 * CALCULATE INTERACTIONS *
1422 **************************/
1424 /* REACTION-FIELD ELECTROSTATICS */
1425 felec = _mm256_mul_ps(qq21,_mm256_sub_ps(_mm256_mul_ps(rinv21,rinvsq21),krf2));
1429 /* Calculate temporary vectorial force */
1430 tx = _mm256_mul_ps(fscal,dx21);
1431 ty = _mm256_mul_ps(fscal,dy21);
1432 tz = _mm256_mul_ps(fscal,dz21);
1434 /* Update vectorial force */
1435 fix2 = _mm256_add_ps(fix2,tx);
1436 fiy2 = _mm256_add_ps(fiy2,ty);
1437 fiz2 = _mm256_add_ps(fiz2,tz);
1439 fjx1 = _mm256_add_ps(fjx1,tx);
1440 fjy1 = _mm256_add_ps(fjy1,ty);
1441 fjz1 = _mm256_add_ps(fjz1,tz);
1443 /**************************
1444 * CALCULATE INTERACTIONS *
1445 **************************/
1447 /* REACTION-FIELD ELECTROSTATICS */
1448 felec = _mm256_mul_ps(qq22,_mm256_sub_ps(_mm256_mul_ps(rinv22,rinvsq22),krf2));
1452 /* Calculate temporary vectorial force */
1453 tx = _mm256_mul_ps(fscal,dx22);
1454 ty = _mm256_mul_ps(fscal,dy22);
1455 tz = _mm256_mul_ps(fscal,dz22);
1457 /* Update vectorial force */
1458 fix2 = _mm256_add_ps(fix2,tx);
1459 fiy2 = _mm256_add_ps(fiy2,ty);
1460 fiz2 = _mm256_add_ps(fiz2,tz);
1462 fjx2 = _mm256_add_ps(fjx2,tx);
1463 fjy2 = _mm256_add_ps(fjy2,ty);
1464 fjz2 = _mm256_add_ps(fjz2,tz);
1466 fjptrA = f+j_coord_offsetA;
1467 fjptrB = f+j_coord_offsetB;
1468 fjptrC = f+j_coord_offsetC;
1469 fjptrD = f+j_coord_offsetD;
1470 fjptrE = f+j_coord_offsetE;
1471 fjptrF = f+j_coord_offsetF;
1472 fjptrG = f+j_coord_offsetG;
1473 fjptrH = f+j_coord_offsetH;
1475 gmx_mm256_decrement_3rvec_8ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjptrE,fjptrF,fjptrG,fjptrH,
1476 fjx0,fjy0,fjz0,fjx1,fjy1,fjz1,fjx2,fjy2,fjz2);
1478 /* Inner loop uses 250 flops */
1481 if(jidx<j_index_end)
1484 /* Get j neighbor index, and coordinate index */
1485 jnrlistA = jjnr[jidx];
1486 jnrlistB = jjnr[jidx+1];
1487 jnrlistC = jjnr[jidx+2];
1488 jnrlistD = jjnr[jidx+3];
1489 jnrlistE = jjnr[jidx+4];
1490 jnrlistF = jjnr[jidx+5];
1491 jnrlistG = jjnr[jidx+6];
1492 jnrlistH = jjnr[jidx+7];
1493 /* Sign of each element will be negative for non-real atoms.
1494 * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
1495 * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
1497 dummy_mask = gmx_mm256_set_m128(gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx+4)),_mm_setzero_si128())),
1498 gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128())));
1500 jnrA = (jnrlistA>=0) ? jnrlistA : 0;
1501 jnrB = (jnrlistB>=0) ? jnrlistB : 0;
1502 jnrC = (jnrlistC>=0) ? jnrlistC : 0;
1503 jnrD = (jnrlistD>=0) ? jnrlistD : 0;
1504 jnrE = (jnrlistE>=0) ? jnrlistE : 0;
1505 jnrF = (jnrlistF>=0) ? jnrlistF : 0;
1506 jnrG = (jnrlistG>=0) ? jnrlistG : 0;
1507 jnrH = (jnrlistH>=0) ? jnrlistH : 0;
1508 j_coord_offsetA = DIM*jnrA;
1509 j_coord_offsetB = DIM*jnrB;
1510 j_coord_offsetC = DIM*jnrC;
1511 j_coord_offsetD = DIM*jnrD;
1512 j_coord_offsetE = DIM*jnrE;
1513 j_coord_offsetF = DIM*jnrF;
1514 j_coord_offsetG = DIM*jnrG;
1515 j_coord_offsetH = DIM*jnrH;
1517 /* load j atom coordinates */
1518 gmx_mm256_load_3rvec_8ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
1519 x+j_coord_offsetC,x+j_coord_offsetD,
1520 x+j_coord_offsetE,x+j_coord_offsetF,
1521 x+j_coord_offsetG,x+j_coord_offsetH,
1522 &jx0,&jy0,&jz0,&jx1,&jy1,&jz1,&jx2,&jy2,&jz2);
1524 /* Calculate displacement vector */
1525 dx00 = _mm256_sub_ps(ix0,jx0);
1526 dy00 = _mm256_sub_ps(iy0,jy0);
1527 dz00 = _mm256_sub_ps(iz0,jz0);
1528 dx01 = _mm256_sub_ps(ix0,jx1);
1529 dy01 = _mm256_sub_ps(iy0,jy1);
1530 dz01 = _mm256_sub_ps(iz0,jz1);
1531 dx02 = _mm256_sub_ps(ix0,jx2);
1532 dy02 = _mm256_sub_ps(iy0,jy2);
1533 dz02 = _mm256_sub_ps(iz0,jz2);
1534 dx10 = _mm256_sub_ps(ix1,jx0);
1535 dy10 = _mm256_sub_ps(iy1,jy0);
1536 dz10 = _mm256_sub_ps(iz1,jz0);
1537 dx11 = _mm256_sub_ps(ix1,jx1);
1538 dy11 = _mm256_sub_ps(iy1,jy1);
1539 dz11 = _mm256_sub_ps(iz1,jz1);
1540 dx12 = _mm256_sub_ps(ix1,jx2);
1541 dy12 = _mm256_sub_ps(iy1,jy2);
1542 dz12 = _mm256_sub_ps(iz1,jz2);
1543 dx20 = _mm256_sub_ps(ix2,jx0);
1544 dy20 = _mm256_sub_ps(iy2,jy0);
1545 dz20 = _mm256_sub_ps(iz2,jz0);
1546 dx21 = _mm256_sub_ps(ix2,jx1);
1547 dy21 = _mm256_sub_ps(iy2,jy1);
1548 dz21 = _mm256_sub_ps(iz2,jz1);
1549 dx22 = _mm256_sub_ps(ix2,jx2);
1550 dy22 = _mm256_sub_ps(iy2,jy2);
1551 dz22 = _mm256_sub_ps(iz2,jz2);
1553 /* Calculate squared distance and things based on it */
1554 rsq00 = gmx_mm256_calc_rsq_ps(dx00,dy00,dz00);
1555 rsq01 = gmx_mm256_calc_rsq_ps(dx01,dy01,dz01);
1556 rsq02 = gmx_mm256_calc_rsq_ps(dx02,dy02,dz02);
1557 rsq10 = gmx_mm256_calc_rsq_ps(dx10,dy10,dz10);
1558 rsq11 = gmx_mm256_calc_rsq_ps(dx11,dy11,dz11);
1559 rsq12 = gmx_mm256_calc_rsq_ps(dx12,dy12,dz12);
1560 rsq20 = gmx_mm256_calc_rsq_ps(dx20,dy20,dz20);
1561 rsq21 = gmx_mm256_calc_rsq_ps(dx21,dy21,dz21);
1562 rsq22 = gmx_mm256_calc_rsq_ps(dx22,dy22,dz22);
1564 rinv00 = gmx_mm256_invsqrt_ps(rsq00);
1565 rinv01 = gmx_mm256_invsqrt_ps(rsq01);
1566 rinv02 = gmx_mm256_invsqrt_ps(rsq02);
1567 rinv10 = gmx_mm256_invsqrt_ps(rsq10);
1568 rinv11 = gmx_mm256_invsqrt_ps(rsq11);
1569 rinv12 = gmx_mm256_invsqrt_ps(rsq12);
1570 rinv20 = gmx_mm256_invsqrt_ps(rsq20);
1571 rinv21 = gmx_mm256_invsqrt_ps(rsq21);
1572 rinv22 = gmx_mm256_invsqrt_ps(rsq22);
1574 rinvsq00 = _mm256_mul_ps(rinv00,rinv00);
1575 rinvsq01 = _mm256_mul_ps(rinv01,rinv01);
1576 rinvsq02 = _mm256_mul_ps(rinv02,rinv02);
1577 rinvsq10 = _mm256_mul_ps(rinv10,rinv10);
1578 rinvsq11 = _mm256_mul_ps(rinv11,rinv11);
1579 rinvsq12 = _mm256_mul_ps(rinv12,rinv12);
1580 rinvsq20 = _mm256_mul_ps(rinv20,rinv20);
1581 rinvsq21 = _mm256_mul_ps(rinv21,rinv21);
1582 rinvsq22 = _mm256_mul_ps(rinv22,rinv22);
1584 fjx0 = _mm256_setzero_ps();
1585 fjy0 = _mm256_setzero_ps();
1586 fjz0 = _mm256_setzero_ps();
1587 fjx1 = _mm256_setzero_ps();
1588 fjy1 = _mm256_setzero_ps();
1589 fjz1 = _mm256_setzero_ps();
1590 fjx2 = _mm256_setzero_ps();
1591 fjy2 = _mm256_setzero_ps();
1592 fjz2 = _mm256_setzero_ps();
1594 /**************************
1595 * CALCULATE INTERACTIONS *
1596 **************************/
1598 /* REACTION-FIELD ELECTROSTATICS */
1599 felec = _mm256_mul_ps(qq00,_mm256_sub_ps(_mm256_mul_ps(rinv00,rinvsq00),krf2));
1601 /* LENNARD-JONES DISPERSION/REPULSION */
1603 rinvsix = _mm256_mul_ps(_mm256_mul_ps(rinvsq00,rinvsq00),rinvsq00);
1604 fvdw = _mm256_mul_ps(_mm256_sub_ps(_mm256_mul_ps(c12_00,rinvsix),c6_00),_mm256_mul_ps(rinvsix,rinvsq00));
1606 fscal = _mm256_add_ps(felec,fvdw);
1608 fscal = _mm256_andnot_ps(dummy_mask,fscal);
1610 /* Calculate temporary vectorial force */
1611 tx = _mm256_mul_ps(fscal,dx00);
1612 ty = _mm256_mul_ps(fscal,dy00);
1613 tz = _mm256_mul_ps(fscal,dz00);
1615 /* Update vectorial force */
1616 fix0 = _mm256_add_ps(fix0,tx);
1617 fiy0 = _mm256_add_ps(fiy0,ty);
1618 fiz0 = _mm256_add_ps(fiz0,tz);
1620 fjx0 = _mm256_add_ps(fjx0,tx);
1621 fjy0 = _mm256_add_ps(fjy0,ty);
1622 fjz0 = _mm256_add_ps(fjz0,tz);
1624 /**************************
1625 * CALCULATE INTERACTIONS *
1626 **************************/
1628 /* REACTION-FIELD ELECTROSTATICS */
1629 felec = _mm256_mul_ps(qq01,_mm256_sub_ps(_mm256_mul_ps(rinv01,rinvsq01),krf2));
1633 fscal = _mm256_andnot_ps(dummy_mask,fscal);
1635 /* Calculate temporary vectorial force */
1636 tx = _mm256_mul_ps(fscal,dx01);
1637 ty = _mm256_mul_ps(fscal,dy01);
1638 tz = _mm256_mul_ps(fscal,dz01);
1640 /* Update vectorial force */
1641 fix0 = _mm256_add_ps(fix0,tx);
1642 fiy0 = _mm256_add_ps(fiy0,ty);
1643 fiz0 = _mm256_add_ps(fiz0,tz);
1645 fjx1 = _mm256_add_ps(fjx1,tx);
1646 fjy1 = _mm256_add_ps(fjy1,ty);
1647 fjz1 = _mm256_add_ps(fjz1,tz);
1649 /**************************
1650 * CALCULATE INTERACTIONS *
1651 **************************/
1653 /* REACTION-FIELD ELECTROSTATICS */
1654 felec = _mm256_mul_ps(qq02,_mm256_sub_ps(_mm256_mul_ps(rinv02,rinvsq02),krf2));
1658 fscal = _mm256_andnot_ps(dummy_mask,fscal);
1660 /* Calculate temporary vectorial force */
1661 tx = _mm256_mul_ps(fscal,dx02);
1662 ty = _mm256_mul_ps(fscal,dy02);
1663 tz = _mm256_mul_ps(fscal,dz02);
1665 /* Update vectorial force */
1666 fix0 = _mm256_add_ps(fix0,tx);
1667 fiy0 = _mm256_add_ps(fiy0,ty);
1668 fiz0 = _mm256_add_ps(fiz0,tz);
1670 fjx2 = _mm256_add_ps(fjx2,tx);
1671 fjy2 = _mm256_add_ps(fjy2,ty);
1672 fjz2 = _mm256_add_ps(fjz2,tz);
1674 /**************************
1675 * CALCULATE INTERACTIONS *
1676 **************************/
1678 /* REACTION-FIELD ELECTROSTATICS */
1679 felec = _mm256_mul_ps(qq10,_mm256_sub_ps(_mm256_mul_ps(rinv10,rinvsq10),krf2));
1683 fscal = _mm256_andnot_ps(dummy_mask,fscal);
1685 /* Calculate temporary vectorial force */
1686 tx = _mm256_mul_ps(fscal,dx10);
1687 ty = _mm256_mul_ps(fscal,dy10);
1688 tz = _mm256_mul_ps(fscal,dz10);
1690 /* Update vectorial force */
1691 fix1 = _mm256_add_ps(fix1,tx);
1692 fiy1 = _mm256_add_ps(fiy1,ty);
1693 fiz1 = _mm256_add_ps(fiz1,tz);
1695 fjx0 = _mm256_add_ps(fjx0,tx);
1696 fjy0 = _mm256_add_ps(fjy0,ty);
1697 fjz0 = _mm256_add_ps(fjz0,tz);
1699 /**************************
1700 * CALCULATE INTERACTIONS *
1701 **************************/
1703 /* REACTION-FIELD ELECTROSTATICS */
1704 felec = _mm256_mul_ps(qq11,_mm256_sub_ps(_mm256_mul_ps(rinv11,rinvsq11),krf2));
1708 fscal = _mm256_andnot_ps(dummy_mask,fscal);
1710 /* Calculate temporary vectorial force */
1711 tx = _mm256_mul_ps(fscal,dx11);
1712 ty = _mm256_mul_ps(fscal,dy11);
1713 tz = _mm256_mul_ps(fscal,dz11);
1715 /* Update vectorial force */
1716 fix1 = _mm256_add_ps(fix1,tx);
1717 fiy1 = _mm256_add_ps(fiy1,ty);
1718 fiz1 = _mm256_add_ps(fiz1,tz);
1720 fjx1 = _mm256_add_ps(fjx1,tx);
1721 fjy1 = _mm256_add_ps(fjy1,ty);
1722 fjz1 = _mm256_add_ps(fjz1,tz);
1724 /**************************
1725 * CALCULATE INTERACTIONS *
1726 **************************/
1728 /* REACTION-FIELD ELECTROSTATICS */
1729 felec = _mm256_mul_ps(qq12,_mm256_sub_ps(_mm256_mul_ps(rinv12,rinvsq12),krf2));
1733 fscal = _mm256_andnot_ps(dummy_mask,fscal);
1735 /* Calculate temporary vectorial force */
1736 tx = _mm256_mul_ps(fscal,dx12);
1737 ty = _mm256_mul_ps(fscal,dy12);
1738 tz = _mm256_mul_ps(fscal,dz12);
1740 /* Update vectorial force */
1741 fix1 = _mm256_add_ps(fix1,tx);
1742 fiy1 = _mm256_add_ps(fiy1,ty);
1743 fiz1 = _mm256_add_ps(fiz1,tz);
1745 fjx2 = _mm256_add_ps(fjx2,tx);
1746 fjy2 = _mm256_add_ps(fjy2,ty);
1747 fjz2 = _mm256_add_ps(fjz2,tz);
1749 /**************************
1750 * CALCULATE INTERACTIONS *
1751 **************************/
1753 /* REACTION-FIELD ELECTROSTATICS */
1754 felec = _mm256_mul_ps(qq20,_mm256_sub_ps(_mm256_mul_ps(rinv20,rinvsq20),krf2));
1758 fscal = _mm256_andnot_ps(dummy_mask,fscal);
1760 /* Calculate temporary vectorial force */
1761 tx = _mm256_mul_ps(fscal,dx20);
1762 ty = _mm256_mul_ps(fscal,dy20);
1763 tz = _mm256_mul_ps(fscal,dz20);
1765 /* Update vectorial force */
1766 fix2 = _mm256_add_ps(fix2,tx);
1767 fiy2 = _mm256_add_ps(fiy2,ty);
1768 fiz2 = _mm256_add_ps(fiz2,tz);
1770 fjx0 = _mm256_add_ps(fjx0,tx);
1771 fjy0 = _mm256_add_ps(fjy0,ty);
1772 fjz0 = _mm256_add_ps(fjz0,tz);
1774 /**************************
1775 * CALCULATE INTERACTIONS *
1776 **************************/
1778 /* REACTION-FIELD ELECTROSTATICS */
1779 felec = _mm256_mul_ps(qq21,_mm256_sub_ps(_mm256_mul_ps(rinv21,rinvsq21),krf2));
1783 fscal = _mm256_andnot_ps(dummy_mask,fscal);
1785 /* Calculate temporary vectorial force */
1786 tx = _mm256_mul_ps(fscal,dx21);
1787 ty = _mm256_mul_ps(fscal,dy21);
1788 tz = _mm256_mul_ps(fscal,dz21);
1790 /* Update vectorial force */
1791 fix2 = _mm256_add_ps(fix2,tx);
1792 fiy2 = _mm256_add_ps(fiy2,ty);
1793 fiz2 = _mm256_add_ps(fiz2,tz);
1795 fjx1 = _mm256_add_ps(fjx1,tx);
1796 fjy1 = _mm256_add_ps(fjy1,ty);
1797 fjz1 = _mm256_add_ps(fjz1,tz);
1799 /**************************
1800 * CALCULATE INTERACTIONS *
1801 **************************/
1803 /* REACTION-FIELD ELECTROSTATICS */
1804 felec = _mm256_mul_ps(qq22,_mm256_sub_ps(_mm256_mul_ps(rinv22,rinvsq22),krf2));
1808 fscal = _mm256_andnot_ps(dummy_mask,fscal);
1810 /* Calculate temporary vectorial force */
1811 tx = _mm256_mul_ps(fscal,dx22);
1812 ty = _mm256_mul_ps(fscal,dy22);
1813 tz = _mm256_mul_ps(fscal,dz22);
1815 /* Update vectorial force */
1816 fix2 = _mm256_add_ps(fix2,tx);
1817 fiy2 = _mm256_add_ps(fiy2,ty);
1818 fiz2 = _mm256_add_ps(fiz2,tz);
1820 fjx2 = _mm256_add_ps(fjx2,tx);
1821 fjy2 = _mm256_add_ps(fjy2,ty);
1822 fjz2 = _mm256_add_ps(fjz2,tz);
1824 fjptrA = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
1825 fjptrB = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
1826 fjptrC = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
1827 fjptrD = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
1828 fjptrE = (jnrlistE>=0) ? f+j_coord_offsetE : scratch;
1829 fjptrF = (jnrlistF>=0) ? f+j_coord_offsetF : scratch;
1830 fjptrG = (jnrlistG>=0) ? f+j_coord_offsetG : scratch;
1831 fjptrH = (jnrlistH>=0) ? f+j_coord_offsetH : scratch;
1833 gmx_mm256_decrement_3rvec_8ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjptrE,fjptrF,fjptrG,fjptrH,
1834 fjx0,fjy0,fjz0,fjx1,fjy1,fjz1,fjx2,fjy2,fjz2);
1836 /* Inner loop uses 250 flops */
1839 /* End of innermost loop */
1841 gmx_mm256_update_iforce_3atom_swizzle_ps(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,
1842 f+i_coord_offset,fshift+i_shift_offset);
1844 /* Increment number of inner iterations */
1845 inneriter += j_index_end - j_index_start;
1847 /* Outer loop uses 18 flops */
1850 /* Increment number of outer iterations */
1853 /* Update outer/inner flops */
1855 inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W3W3_F,outeriter*18 + inneriter*250);