2 * This file is part of the GROMACS molecular simulation package.
4 * Copyright (c) 2012,2013, by the GROMACS development team, led by
5 * Mark Abraham, David van der Spoel, Berk Hess, and Erik Lindahl,
6 * and including many others, as listed in the AUTHORS file in the
7 * top-level source directory and at http://www.gromacs.org.
9 * GROMACS is free software; you can redistribute it and/or
10 * modify it under the terms of the GNU Lesser General Public License
11 * as published by the Free Software Foundation; either version 2.1
12 * of the License, or (at your option) any later version.
14 * GROMACS is distributed in the hope that it will be useful,
15 * but WITHOUT ANY WARRANTY; without even the implied warranty of
16 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
17 * Lesser General Public License for more details.
19 * You should have received a copy of the GNU Lesser General Public
20 * License along with GROMACS; if not, see
21 * http://www.gnu.org/licenses, or write to the Free Software Foundation,
22 * Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA.
24 * If you want to redistribute modifications to GROMACS, please
25 * consider that scientific software is very special. Version
26 * control is crucial - bugs must be traceable. We will be happy to
27 * consider code for inclusion in the official distribution, but
28 * derived work must not be called official GROMACS. Details are found
29 * in the README & COPYING files - if they are missing, get the
30 * official version at http://www.gromacs.org.
32 * To help us fund GROMACS development, we humbly ask that you cite
33 * the research papers on the package. Check out http://www.gromacs.org.
36 * Note: this file was generated by the GROMACS avx_256_single kernel generator.
44 #include "../nb_kernel.h"
45 #include "types/simple.h"
49 #include "gromacs/simd/math_x86_avx_256_single.h"
50 #include "kernelutil_x86_avx_256_single.h"
53 * Gromacs nonbonded kernel: nb_kernel_ElecRFCut_VdwLJSw_GeomW3P1_VF_avx_256_single
54 * Electrostatics interaction: ReactionField
55 * VdW interaction: LennardJones
56 * Geometry: Water3-Particle
57 * Calculate force/pot: PotentialAndForce
60 nb_kernel_ElecRFCut_VdwLJSw_GeomW3P1_VF_avx_256_single
61 (t_nblist * gmx_restrict nlist,
62 rvec * gmx_restrict xx,
63 rvec * gmx_restrict ff,
64 t_forcerec * gmx_restrict fr,
65 t_mdatoms * gmx_restrict mdatoms,
66 nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
67 t_nrnb * gmx_restrict nrnb)
69 /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
70 * just 0 for non-waters.
71 * Suffixes A,B,C,D,E,F,G,H refer to j loop unrolling done with AVX, e.g. for the eight different
72 * jnr indices corresponding to data put in the four positions in the SIMD register.
74 int i_shift_offset,i_coord_offset,outeriter,inneriter;
75 int j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
76 int jnrA,jnrB,jnrC,jnrD;
77 int jnrE,jnrF,jnrG,jnrH;
78 int jnrlistA,jnrlistB,jnrlistC,jnrlistD;
79 int jnrlistE,jnrlistF,jnrlistG,jnrlistH;
80 int j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
81 int j_coord_offsetE,j_coord_offsetF,j_coord_offsetG,j_coord_offsetH;
82 int *iinr,*jindex,*jjnr,*shiftidx,*gid;
84 real *shiftvec,*fshift,*x,*f;
85 real *fjptrA,*fjptrB,*fjptrC,*fjptrD,*fjptrE,*fjptrF,*fjptrG,*fjptrH;
87 __m256 tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
88 real * vdwioffsetptr0;
89 __m256 ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
90 real * vdwioffsetptr1;
91 __m256 ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
92 real * vdwioffsetptr2;
93 __m256 ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
94 int vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D,vdwjidx0E,vdwjidx0F,vdwjidx0G,vdwjidx0H;
95 __m256 jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
96 __m256 dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
97 __m256 dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
98 __m256 dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
99 __m256 velec,felec,velecsum,facel,crf,krf,krf2;
102 __m256 rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
105 __m256 one_sixth = _mm256_set1_ps(1.0/6.0);
106 __m256 one_twelfth = _mm256_set1_ps(1.0/12.0);
107 __m256 rswitch,swV3,swV4,swV5,swF2,swF3,swF4,d,d2,sw,dsw;
108 real rswitch_scalar,d_scalar;
109 __m256 dummy_mask,cutoff_mask;
110 __m256 signbit = _mm256_castsi256_ps( _mm256_set1_epi32(0x80000000) );
111 __m256 one = _mm256_set1_ps(1.0);
112 __m256 two = _mm256_set1_ps(2.0);
118 jindex = nlist->jindex;
120 shiftidx = nlist->shift;
122 shiftvec = fr->shift_vec[0];
123 fshift = fr->fshift[0];
124 facel = _mm256_set1_ps(fr->epsfac);
125 charge = mdatoms->chargeA;
126 krf = _mm256_set1_ps(fr->ic->k_rf);
127 krf2 = _mm256_set1_ps(fr->ic->k_rf*2.0);
128 crf = _mm256_set1_ps(fr->ic->c_rf);
129 nvdwtype = fr->ntype;
131 vdwtype = mdatoms->typeA;
133 /* Setup water-specific parameters */
134 inr = nlist->iinr[0];
135 iq0 = _mm256_mul_ps(facel,_mm256_set1_ps(charge[inr+0]));
136 iq1 = _mm256_mul_ps(facel,_mm256_set1_ps(charge[inr+1]));
137 iq2 = _mm256_mul_ps(facel,_mm256_set1_ps(charge[inr+2]));
138 vdwioffsetptr0 = vdwparam+2*nvdwtype*vdwtype[inr+0];
140 /* When we use explicit cutoffs the value must be identical for elec and VdW, so use elec as an arbitrary choice */
141 rcutoff_scalar = fr->rcoulomb;
142 rcutoff = _mm256_set1_ps(rcutoff_scalar);
143 rcutoff2 = _mm256_mul_ps(rcutoff,rcutoff);
145 rswitch_scalar = fr->rvdw_switch;
146 rswitch = _mm256_set1_ps(rswitch_scalar);
147 /* Setup switch parameters */
148 d_scalar = rcutoff_scalar-rswitch_scalar;
149 d = _mm256_set1_ps(d_scalar);
150 swV3 = _mm256_set1_ps(-10.0/(d_scalar*d_scalar*d_scalar));
151 swV4 = _mm256_set1_ps( 15.0/(d_scalar*d_scalar*d_scalar*d_scalar));
152 swV5 = _mm256_set1_ps( -6.0/(d_scalar*d_scalar*d_scalar*d_scalar*d_scalar));
153 swF2 = _mm256_set1_ps(-30.0/(d_scalar*d_scalar*d_scalar));
154 swF3 = _mm256_set1_ps( 60.0/(d_scalar*d_scalar*d_scalar*d_scalar));
155 swF4 = _mm256_set1_ps(-30.0/(d_scalar*d_scalar*d_scalar*d_scalar*d_scalar));
157 /* Avoid stupid compiler warnings */
158 jnrA = jnrB = jnrC = jnrD = jnrE = jnrF = jnrG = jnrH = 0;
171 for(iidx=0;iidx<4*DIM;iidx++)
176 /* Start outer loop over neighborlists */
177 for(iidx=0; iidx<nri; iidx++)
179 /* Load shift vector for this list */
180 i_shift_offset = DIM*shiftidx[iidx];
182 /* Load limits for loop over neighbors */
183 j_index_start = jindex[iidx];
184 j_index_end = jindex[iidx+1];
186 /* Get outer coordinate index */
188 i_coord_offset = DIM*inr;
190 /* Load i particle coords and add shift vector */
191 gmx_mm256_load_shift_and_3rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,
192 &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2);
194 fix0 = _mm256_setzero_ps();
195 fiy0 = _mm256_setzero_ps();
196 fiz0 = _mm256_setzero_ps();
197 fix1 = _mm256_setzero_ps();
198 fiy1 = _mm256_setzero_ps();
199 fiz1 = _mm256_setzero_ps();
200 fix2 = _mm256_setzero_ps();
201 fiy2 = _mm256_setzero_ps();
202 fiz2 = _mm256_setzero_ps();
204 /* Reset potential sums */
205 velecsum = _mm256_setzero_ps();
206 vvdwsum = _mm256_setzero_ps();
208 /* Start inner kernel loop */
209 for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+7]>=0; jidx+=8)
212 /* Get j neighbor index, and coordinate index */
221 j_coord_offsetA = DIM*jnrA;
222 j_coord_offsetB = DIM*jnrB;
223 j_coord_offsetC = DIM*jnrC;
224 j_coord_offsetD = DIM*jnrD;
225 j_coord_offsetE = DIM*jnrE;
226 j_coord_offsetF = DIM*jnrF;
227 j_coord_offsetG = DIM*jnrG;
228 j_coord_offsetH = DIM*jnrH;
230 /* load j atom coordinates */
231 gmx_mm256_load_1rvec_8ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
232 x+j_coord_offsetC,x+j_coord_offsetD,
233 x+j_coord_offsetE,x+j_coord_offsetF,
234 x+j_coord_offsetG,x+j_coord_offsetH,
237 /* Calculate displacement vector */
238 dx00 = _mm256_sub_ps(ix0,jx0);
239 dy00 = _mm256_sub_ps(iy0,jy0);
240 dz00 = _mm256_sub_ps(iz0,jz0);
241 dx10 = _mm256_sub_ps(ix1,jx0);
242 dy10 = _mm256_sub_ps(iy1,jy0);
243 dz10 = _mm256_sub_ps(iz1,jz0);
244 dx20 = _mm256_sub_ps(ix2,jx0);
245 dy20 = _mm256_sub_ps(iy2,jy0);
246 dz20 = _mm256_sub_ps(iz2,jz0);
248 /* Calculate squared distance and things based on it */
249 rsq00 = gmx_mm256_calc_rsq_ps(dx00,dy00,dz00);
250 rsq10 = gmx_mm256_calc_rsq_ps(dx10,dy10,dz10);
251 rsq20 = gmx_mm256_calc_rsq_ps(dx20,dy20,dz20);
253 rinv00 = gmx_mm256_invsqrt_ps(rsq00);
254 rinv10 = gmx_mm256_invsqrt_ps(rsq10);
255 rinv20 = gmx_mm256_invsqrt_ps(rsq20);
257 rinvsq00 = _mm256_mul_ps(rinv00,rinv00);
258 rinvsq10 = _mm256_mul_ps(rinv10,rinv10);
259 rinvsq20 = _mm256_mul_ps(rinv20,rinv20);
261 /* Load parameters for j particles */
262 jq0 = gmx_mm256_load_8real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
263 charge+jnrC+0,charge+jnrD+0,
264 charge+jnrE+0,charge+jnrF+0,
265 charge+jnrG+0,charge+jnrH+0);
266 vdwjidx0A = 2*vdwtype[jnrA+0];
267 vdwjidx0B = 2*vdwtype[jnrB+0];
268 vdwjidx0C = 2*vdwtype[jnrC+0];
269 vdwjidx0D = 2*vdwtype[jnrD+0];
270 vdwjidx0E = 2*vdwtype[jnrE+0];
271 vdwjidx0F = 2*vdwtype[jnrF+0];
272 vdwjidx0G = 2*vdwtype[jnrG+0];
273 vdwjidx0H = 2*vdwtype[jnrH+0];
275 fjx0 = _mm256_setzero_ps();
276 fjy0 = _mm256_setzero_ps();
277 fjz0 = _mm256_setzero_ps();
279 /**************************
280 * CALCULATE INTERACTIONS *
281 **************************/
283 if (gmx_mm256_any_lt(rsq00,rcutoff2))
286 r00 = _mm256_mul_ps(rsq00,rinv00);
288 /* Compute parameters for interactions between i and j atoms */
289 qq00 = _mm256_mul_ps(iq0,jq0);
290 gmx_mm256_load_8pair_swizzle_ps(vdwioffsetptr0+vdwjidx0A,
291 vdwioffsetptr0+vdwjidx0B,
292 vdwioffsetptr0+vdwjidx0C,
293 vdwioffsetptr0+vdwjidx0D,
294 vdwioffsetptr0+vdwjidx0E,
295 vdwioffsetptr0+vdwjidx0F,
296 vdwioffsetptr0+vdwjidx0G,
297 vdwioffsetptr0+vdwjidx0H,
300 /* REACTION-FIELD ELECTROSTATICS */
301 velec = _mm256_mul_ps(qq00,_mm256_sub_ps(_mm256_add_ps(rinv00,_mm256_mul_ps(krf,rsq00)),crf));
302 felec = _mm256_mul_ps(qq00,_mm256_sub_ps(_mm256_mul_ps(rinv00,rinvsq00),krf2));
304 /* LENNARD-JONES DISPERSION/REPULSION */
306 rinvsix = _mm256_mul_ps(_mm256_mul_ps(rinvsq00,rinvsq00),rinvsq00);
307 vvdw6 = _mm256_mul_ps(c6_00,rinvsix);
308 vvdw12 = _mm256_mul_ps(c12_00,_mm256_mul_ps(rinvsix,rinvsix));
309 vvdw = _mm256_sub_ps( _mm256_mul_ps(vvdw12,one_twelfth) , _mm256_mul_ps(vvdw6,one_sixth) );
310 fvdw = _mm256_mul_ps(_mm256_sub_ps(vvdw12,vvdw6),rinvsq00);
312 d = _mm256_sub_ps(r00,rswitch);
313 d = _mm256_max_ps(d,_mm256_setzero_ps());
314 d2 = _mm256_mul_ps(d,d);
315 sw = _mm256_add_ps(one,_mm256_mul_ps(d2,_mm256_mul_ps(d,_mm256_add_ps(swV3,_mm256_mul_ps(d,_mm256_add_ps(swV4,_mm256_mul_ps(d,swV5)))))));
317 dsw = _mm256_mul_ps(d2,_mm256_add_ps(swF2,_mm256_mul_ps(d,_mm256_add_ps(swF3,_mm256_mul_ps(d,swF4)))));
319 /* Evaluate switch function */
320 /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
321 fvdw = _mm256_sub_ps( _mm256_mul_ps(fvdw,sw) , _mm256_mul_ps(rinv00,_mm256_mul_ps(vvdw,dsw)) );
322 vvdw = _mm256_mul_ps(vvdw,sw);
323 cutoff_mask = _mm256_cmp_ps(rsq00,rcutoff2,_CMP_LT_OQ);
325 /* Update potential sum for this i atom from the interaction with this j atom. */
326 velec = _mm256_and_ps(velec,cutoff_mask);
327 velecsum = _mm256_add_ps(velecsum,velec);
328 vvdw = _mm256_and_ps(vvdw,cutoff_mask);
329 vvdwsum = _mm256_add_ps(vvdwsum,vvdw);
331 fscal = _mm256_add_ps(felec,fvdw);
333 fscal = _mm256_and_ps(fscal,cutoff_mask);
335 /* Calculate temporary vectorial force */
336 tx = _mm256_mul_ps(fscal,dx00);
337 ty = _mm256_mul_ps(fscal,dy00);
338 tz = _mm256_mul_ps(fscal,dz00);
340 /* Update vectorial force */
341 fix0 = _mm256_add_ps(fix0,tx);
342 fiy0 = _mm256_add_ps(fiy0,ty);
343 fiz0 = _mm256_add_ps(fiz0,tz);
345 fjx0 = _mm256_add_ps(fjx0,tx);
346 fjy0 = _mm256_add_ps(fjy0,ty);
347 fjz0 = _mm256_add_ps(fjz0,tz);
351 /**************************
352 * CALCULATE INTERACTIONS *
353 **************************/
355 if (gmx_mm256_any_lt(rsq10,rcutoff2))
358 /* Compute parameters for interactions between i and j atoms */
359 qq10 = _mm256_mul_ps(iq1,jq0);
361 /* REACTION-FIELD ELECTROSTATICS */
362 velec = _mm256_mul_ps(qq10,_mm256_sub_ps(_mm256_add_ps(rinv10,_mm256_mul_ps(krf,rsq10)),crf));
363 felec = _mm256_mul_ps(qq10,_mm256_sub_ps(_mm256_mul_ps(rinv10,rinvsq10),krf2));
365 cutoff_mask = _mm256_cmp_ps(rsq10,rcutoff2,_CMP_LT_OQ);
367 /* Update potential sum for this i atom from the interaction with this j atom. */
368 velec = _mm256_and_ps(velec,cutoff_mask);
369 velecsum = _mm256_add_ps(velecsum,velec);
373 fscal = _mm256_and_ps(fscal,cutoff_mask);
375 /* Calculate temporary vectorial force */
376 tx = _mm256_mul_ps(fscal,dx10);
377 ty = _mm256_mul_ps(fscal,dy10);
378 tz = _mm256_mul_ps(fscal,dz10);
380 /* Update vectorial force */
381 fix1 = _mm256_add_ps(fix1,tx);
382 fiy1 = _mm256_add_ps(fiy1,ty);
383 fiz1 = _mm256_add_ps(fiz1,tz);
385 fjx0 = _mm256_add_ps(fjx0,tx);
386 fjy0 = _mm256_add_ps(fjy0,ty);
387 fjz0 = _mm256_add_ps(fjz0,tz);
391 /**************************
392 * CALCULATE INTERACTIONS *
393 **************************/
395 if (gmx_mm256_any_lt(rsq20,rcutoff2))
398 /* Compute parameters for interactions between i and j atoms */
399 qq20 = _mm256_mul_ps(iq2,jq0);
401 /* REACTION-FIELD ELECTROSTATICS */
402 velec = _mm256_mul_ps(qq20,_mm256_sub_ps(_mm256_add_ps(rinv20,_mm256_mul_ps(krf,rsq20)),crf));
403 felec = _mm256_mul_ps(qq20,_mm256_sub_ps(_mm256_mul_ps(rinv20,rinvsq20),krf2));
405 cutoff_mask = _mm256_cmp_ps(rsq20,rcutoff2,_CMP_LT_OQ);
407 /* Update potential sum for this i atom from the interaction with this j atom. */
408 velec = _mm256_and_ps(velec,cutoff_mask);
409 velecsum = _mm256_add_ps(velecsum,velec);
413 fscal = _mm256_and_ps(fscal,cutoff_mask);
415 /* Calculate temporary vectorial force */
416 tx = _mm256_mul_ps(fscal,dx20);
417 ty = _mm256_mul_ps(fscal,dy20);
418 tz = _mm256_mul_ps(fscal,dz20);
420 /* Update vectorial force */
421 fix2 = _mm256_add_ps(fix2,tx);
422 fiy2 = _mm256_add_ps(fiy2,ty);
423 fiz2 = _mm256_add_ps(fiz2,tz);
425 fjx0 = _mm256_add_ps(fjx0,tx);
426 fjy0 = _mm256_add_ps(fjy0,ty);
427 fjz0 = _mm256_add_ps(fjz0,tz);
431 fjptrA = f+j_coord_offsetA;
432 fjptrB = f+j_coord_offsetB;
433 fjptrC = f+j_coord_offsetC;
434 fjptrD = f+j_coord_offsetD;
435 fjptrE = f+j_coord_offsetE;
436 fjptrF = f+j_coord_offsetF;
437 fjptrG = f+j_coord_offsetG;
438 fjptrH = f+j_coord_offsetH;
440 gmx_mm256_decrement_1rvec_8ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjptrE,fjptrF,fjptrG,fjptrH,fjx0,fjy0,fjz0);
442 /* Inner loop uses 145 flops */
448 /* Get j neighbor index, and coordinate index */
449 jnrlistA = jjnr[jidx];
450 jnrlistB = jjnr[jidx+1];
451 jnrlistC = jjnr[jidx+2];
452 jnrlistD = jjnr[jidx+3];
453 jnrlistE = jjnr[jidx+4];
454 jnrlistF = jjnr[jidx+5];
455 jnrlistG = jjnr[jidx+6];
456 jnrlistH = jjnr[jidx+7];
457 /* Sign of each element will be negative for non-real atoms.
458 * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
459 * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
461 dummy_mask = gmx_mm256_set_m128(gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx+4)),_mm_setzero_si128())),
462 gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128())));
464 jnrA = (jnrlistA>=0) ? jnrlistA : 0;
465 jnrB = (jnrlistB>=0) ? jnrlistB : 0;
466 jnrC = (jnrlistC>=0) ? jnrlistC : 0;
467 jnrD = (jnrlistD>=0) ? jnrlistD : 0;
468 jnrE = (jnrlistE>=0) ? jnrlistE : 0;
469 jnrF = (jnrlistF>=0) ? jnrlistF : 0;
470 jnrG = (jnrlistG>=0) ? jnrlistG : 0;
471 jnrH = (jnrlistH>=0) ? jnrlistH : 0;
472 j_coord_offsetA = DIM*jnrA;
473 j_coord_offsetB = DIM*jnrB;
474 j_coord_offsetC = DIM*jnrC;
475 j_coord_offsetD = DIM*jnrD;
476 j_coord_offsetE = DIM*jnrE;
477 j_coord_offsetF = DIM*jnrF;
478 j_coord_offsetG = DIM*jnrG;
479 j_coord_offsetH = DIM*jnrH;
481 /* load j atom coordinates */
482 gmx_mm256_load_1rvec_8ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
483 x+j_coord_offsetC,x+j_coord_offsetD,
484 x+j_coord_offsetE,x+j_coord_offsetF,
485 x+j_coord_offsetG,x+j_coord_offsetH,
488 /* Calculate displacement vector */
489 dx00 = _mm256_sub_ps(ix0,jx0);
490 dy00 = _mm256_sub_ps(iy0,jy0);
491 dz00 = _mm256_sub_ps(iz0,jz0);
492 dx10 = _mm256_sub_ps(ix1,jx0);
493 dy10 = _mm256_sub_ps(iy1,jy0);
494 dz10 = _mm256_sub_ps(iz1,jz0);
495 dx20 = _mm256_sub_ps(ix2,jx0);
496 dy20 = _mm256_sub_ps(iy2,jy0);
497 dz20 = _mm256_sub_ps(iz2,jz0);
499 /* Calculate squared distance and things based on it */
500 rsq00 = gmx_mm256_calc_rsq_ps(dx00,dy00,dz00);
501 rsq10 = gmx_mm256_calc_rsq_ps(dx10,dy10,dz10);
502 rsq20 = gmx_mm256_calc_rsq_ps(dx20,dy20,dz20);
504 rinv00 = gmx_mm256_invsqrt_ps(rsq00);
505 rinv10 = gmx_mm256_invsqrt_ps(rsq10);
506 rinv20 = gmx_mm256_invsqrt_ps(rsq20);
508 rinvsq00 = _mm256_mul_ps(rinv00,rinv00);
509 rinvsq10 = _mm256_mul_ps(rinv10,rinv10);
510 rinvsq20 = _mm256_mul_ps(rinv20,rinv20);
512 /* Load parameters for j particles */
513 jq0 = gmx_mm256_load_8real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
514 charge+jnrC+0,charge+jnrD+0,
515 charge+jnrE+0,charge+jnrF+0,
516 charge+jnrG+0,charge+jnrH+0);
517 vdwjidx0A = 2*vdwtype[jnrA+0];
518 vdwjidx0B = 2*vdwtype[jnrB+0];
519 vdwjidx0C = 2*vdwtype[jnrC+0];
520 vdwjidx0D = 2*vdwtype[jnrD+0];
521 vdwjidx0E = 2*vdwtype[jnrE+0];
522 vdwjidx0F = 2*vdwtype[jnrF+0];
523 vdwjidx0G = 2*vdwtype[jnrG+0];
524 vdwjidx0H = 2*vdwtype[jnrH+0];
526 fjx0 = _mm256_setzero_ps();
527 fjy0 = _mm256_setzero_ps();
528 fjz0 = _mm256_setzero_ps();
530 /**************************
531 * CALCULATE INTERACTIONS *
532 **************************/
534 if (gmx_mm256_any_lt(rsq00,rcutoff2))
537 r00 = _mm256_mul_ps(rsq00,rinv00);
538 r00 = _mm256_andnot_ps(dummy_mask,r00);
540 /* Compute parameters for interactions between i and j atoms */
541 qq00 = _mm256_mul_ps(iq0,jq0);
542 gmx_mm256_load_8pair_swizzle_ps(vdwioffsetptr0+vdwjidx0A,
543 vdwioffsetptr0+vdwjidx0B,
544 vdwioffsetptr0+vdwjidx0C,
545 vdwioffsetptr0+vdwjidx0D,
546 vdwioffsetptr0+vdwjidx0E,
547 vdwioffsetptr0+vdwjidx0F,
548 vdwioffsetptr0+vdwjidx0G,
549 vdwioffsetptr0+vdwjidx0H,
552 /* REACTION-FIELD ELECTROSTATICS */
553 velec = _mm256_mul_ps(qq00,_mm256_sub_ps(_mm256_add_ps(rinv00,_mm256_mul_ps(krf,rsq00)),crf));
554 felec = _mm256_mul_ps(qq00,_mm256_sub_ps(_mm256_mul_ps(rinv00,rinvsq00),krf2));
556 /* LENNARD-JONES DISPERSION/REPULSION */
558 rinvsix = _mm256_mul_ps(_mm256_mul_ps(rinvsq00,rinvsq00),rinvsq00);
559 vvdw6 = _mm256_mul_ps(c6_00,rinvsix);
560 vvdw12 = _mm256_mul_ps(c12_00,_mm256_mul_ps(rinvsix,rinvsix));
561 vvdw = _mm256_sub_ps( _mm256_mul_ps(vvdw12,one_twelfth) , _mm256_mul_ps(vvdw6,one_sixth) );
562 fvdw = _mm256_mul_ps(_mm256_sub_ps(vvdw12,vvdw6),rinvsq00);
564 d = _mm256_sub_ps(r00,rswitch);
565 d = _mm256_max_ps(d,_mm256_setzero_ps());
566 d2 = _mm256_mul_ps(d,d);
567 sw = _mm256_add_ps(one,_mm256_mul_ps(d2,_mm256_mul_ps(d,_mm256_add_ps(swV3,_mm256_mul_ps(d,_mm256_add_ps(swV4,_mm256_mul_ps(d,swV5)))))));
569 dsw = _mm256_mul_ps(d2,_mm256_add_ps(swF2,_mm256_mul_ps(d,_mm256_add_ps(swF3,_mm256_mul_ps(d,swF4)))));
571 /* Evaluate switch function */
572 /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
573 fvdw = _mm256_sub_ps( _mm256_mul_ps(fvdw,sw) , _mm256_mul_ps(rinv00,_mm256_mul_ps(vvdw,dsw)) );
574 vvdw = _mm256_mul_ps(vvdw,sw);
575 cutoff_mask = _mm256_cmp_ps(rsq00,rcutoff2,_CMP_LT_OQ);
577 /* Update potential sum for this i atom from the interaction with this j atom. */
578 velec = _mm256_and_ps(velec,cutoff_mask);
579 velec = _mm256_andnot_ps(dummy_mask,velec);
580 velecsum = _mm256_add_ps(velecsum,velec);
581 vvdw = _mm256_and_ps(vvdw,cutoff_mask);
582 vvdw = _mm256_andnot_ps(dummy_mask,vvdw);
583 vvdwsum = _mm256_add_ps(vvdwsum,vvdw);
585 fscal = _mm256_add_ps(felec,fvdw);
587 fscal = _mm256_and_ps(fscal,cutoff_mask);
589 fscal = _mm256_andnot_ps(dummy_mask,fscal);
591 /* Calculate temporary vectorial force */
592 tx = _mm256_mul_ps(fscal,dx00);
593 ty = _mm256_mul_ps(fscal,dy00);
594 tz = _mm256_mul_ps(fscal,dz00);
596 /* Update vectorial force */
597 fix0 = _mm256_add_ps(fix0,tx);
598 fiy0 = _mm256_add_ps(fiy0,ty);
599 fiz0 = _mm256_add_ps(fiz0,tz);
601 fjx0 = _mm256_add_ps(fjx0,tx);
602 fjy0 = _mm256_add_ps(fjy0,ty);
603 fjz0 = _mm256_add_ps(fjz0,tz);
607 /**************************
608 * CALCULATE INTERACTIONS *
609 **************************/
611 if (gmx_mm256_any_lt(rsq10,rcutoff2))
614 /* Compute parameters for interactions between i and j atoms */
615 qq10 = _mm256_mul_ps(iq1,jq0);
617 /* REACTION-FIELD ELECTROSTATICS */
618 velec = _mm256_mul_ps(qq10,_mm256_sub_ps(_mm256_add_ps(rinv10,_mm256_mul_ps(krf,rsq10)),crf));
619 felec = _mm256_mul_ps(qq10,_mm256_sub_ps(_mm256_mul_ps(rinv10,rinvsq10),krf2));
621 cutoff_mask = _mm256_cmp_ps(rsq10,rcutoff2,_CMP_LT_OQ);
623 /* Update potential sum for this i atom from the interaction with this j atom. */
624 velec = _mm256_and_ps(velec,cutoff_mask);
625 velec = _mm256_andnot_ps(dummy_mask,velec);
626 velecsum = _mm256_add_ps(velecsum,velec);
630 fscal = _mm256_and_ps(fscal,cutoff_mask);
632 fscal = _mm256_andnot_ps(dummy_mask,fscal);
634 /* Calculate temporary vectorial force */
635 tx = _mm256_mul_ps(fscal,dx10);
636 ty = _mm256_mul_ps(fscal,dy10);
637 tz = _mm256_mul_ps(fscal,dz10);
639 /* Update vectorial force */
640 fix1 = _mm256_add_ps(fix1,tx);
641 fiy1 = _mm256_add_ps(fiy1,ty);
642 fiz1 = _mm256_add_ps(fiz1,tz);
644 fjx0 = _mm256_add_ps(fjx0,tx);
645 fjy0 = _mm256_add_ps(fjy0,ty);
646 fjz0 = _mm256_add_ps(fjz0,tz);
650 /**************************
651 * CALCULATE INTERACTIONS *
652 **************************/
654 if (gmx_mm256_any_lt(rsq20,rcutoff2))
657 /* Compute parameters for interactions between i and j atoms */
658 qq20 = _mm256_mul_ps(iq2,jq0);
660 /* REACTION-FIELD ELECTROSTATICS */
661 velec = _mm256_mul_ps(qq20,_mm256_sub_ps(_mm256_add_ps(rinv20,_mm256_mul_ps(krf,rsq20)),crf));
662 felec = _mm256_mul_ps(qq20,_mm256_sub_ps(_mm256_mul_ps(rinv20,rinvsq20),krf2));
664 cutoff_mask = _mm256_cmp_ps(rsq20,rcutoff2,_CMP_LT_OQ);
666 /* Update potential sum for this i atom from the interaction with this j atom. */
667 velec = _mm256_and_ps(velec,cutoff_mask);
668 velec = _mm256_andnot_ps(dummy_mask,velec);
669 velecsum = _mm256_add_ps(velecsum,velec);
673 fscal = _mm256_and_ps(fscal,cutoff_mask);
675 fscal = _mm256_andnot_ps(dummy_mask,fscal);
677 /* Calculate temporary vectorial force */
678 tx = _mm256_mul_ps(fscal,dx20);
679 ty = _mm256_mul_ps(fscal,dy20);
680 tz = _mm256_mul_ps(fscal,dz20);
682 /* Update vectorial force */
683 fix2 = _mm256_add_ps(fix2,tx);
684 fiy2 = _mm256_add_ps(fiy2,ty);
685 fiz2 = _mm256_add_ps(fiz2,tz);
687 fjx0 = _mm256_add_ps(fjx0,tx);
688 fjy0 = _mm256_add_ps(fjy0,ty);
689 fjz0 = _mm256_add_ps(fjz0,tz);
693 fjptrA = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
694 fjptrB = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
695 fjptrC = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
696 fjptrD = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
697 fjptrE = (jnrlistE>=0) ? f+j_coord_offsetE : scratch;
698 fjptrF = (jnrlistF>=0) ? f+j_coord_offsetF : scratch;
699 fjptrG = (jnrlistG>=0) ? f+j_coord_offsetG : scratch;
700 fjptrH = (jnrlistH>=0) ? f+j_coord_offsetH : scratch;
702 gmx_mm256_decrement_1rvec_8ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjptrE,fjptrF,fjptrG,fjptrH,fjx0,fjy0,fjz0);
704 /* Inner loop uses 146 flops */
707 /* End of innermost loop */
709 gmx_mm256_update_iforce_3atom_swizzle_ps(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,
710 f+i_coord_offset,fshift+i_shift_offset);
713 /* Update potential energies */
714 gmx_mm256_update_1pot_ps(velecsum,kernel_data->energygrp_elec+ggid);
715 gmx_mm256_update_1pot_ps(vvdwsum,kernel_data->energygrp_vdw+ggid);
717 /* Increment number of inner iterations */
718 inneriter += j_index_end - j_index_start;
720 /* Outer loop uses 20 flops */
723 /* Increment number of outer iterations */
726 /* Update outer/inner flops */
728 inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W3_VF,outeriter*20 + inneriter*146);
731 * Gromacs nonbonded kernel: nb_kernel_ElecRFCut_VdwLJSw_GeomW3P1_F_avx_256_single
732 * Electrostatics interaction: ReactionField
733 * VdW interaction: LennardJones
734 * Geometry: Water3-Particle
735 * Calculate force/pot: Force
738 nb_kernel_ElecRFCut_VdwLJSw_GeomW3P1_F_avx_256_single
739 (t_nblist * gmx_restrict nlist,
740 rvec * gmx_restrict xx,
741 rvec * gmx_restrict ff,
742 t_forcerec * gmx_restrict fr,
743 t_mdatoms * gmx_restrict mdatoms,
744 nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
745 t_nrnb * gmx_restrict nrnb)
747 /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
748 * just 0 for non-waters.
749 * Suffixes A,B,C,D,E,F,G,H refer to j loop unrolling done with AVX, e.g. for the eight different
750 * jnr indices corresponding to data put in the four positions in the SIMD register.
752 int i_shift_offset,i_coord_offset,outeriter,inneriter;
753 int j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
754 int jnrA,jnrB,jnrC,jnrD;
755 int jnrE,jnrF,jnrG,jnrH;
756 int jnrlistA,jnrlistB,jnrlistC,jnrlistD;
757 int jnrlistE,jnrlistF,jnrlistG,jnrlistH;
758 int j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
759 int j_coord_offsetE,j_coord_offsetF,j_coord_offsetG,j_coord_offsetH;
760 int *iinr,*jindex,*jjnr,*shiftidx,*gid;
762 real *shiftvec,*fshift,*x,*f;
763 real *fjptrA,*fjptrB,*fjptrC,*fjptrD,*fjptrE,*fjptrF,*fjptrG,*fjptrH;
765 __m256 tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
766 real * vdwioffsetptr0;
767 __m256 ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
768 real * vdwioffsetptr1;
769 __m256 ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
770 real * vdwioffsetptr2;
771 __m256 ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
772 int vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D,vdwjidx0E,vdwjidx0F,vdwjidx0G,vdwjidx0H;
773 __m256 jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
774 __m256 dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
775 __m256 dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
776 __m256 dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
777 __m256 velec,felec,velecsum,facel,crf,krf,krf2;
780 __m256 rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
783 __m256 one_sixth = _mm256_set1_ps(1.0/6.0);
784 __m256 one_twelfth = _mm256_set1_ps(1.0/12.0);
785 __m256 rswitch,swV3,swV4,swV5,swF2,swF3,swF4,d,d2,sw,dsw;
786 real rswitch_scalar,d_scalar;
787 __m256 dummy_mask,cutoff_mask;
788 __m256 signbit = _mm256_castsi256_ps( _mm256_set1_epi32(0x80000000) );
789 __m256 one = _mm256_set1_ps(1.0);
790 __m256 two = _mm256_set1_ps(2.0);
796 jindex = nlist->jindex;
798 shiftidx = nlist->shift;
800 shiftvec = fr->shift_vec[0];
801 fshift = fr->fshift[0];
802 facel = _mm256_set1_ps(fr->epsfac);
803 charge = mdatoms->chargeA;
804 krf = _mm256_set1_ps(fr->ic->k_rf);
805 krf2 = _mm256_set1_ps(fr->ic->k_rf*2.0);
806 crf = _mm256_set1_ps(fr->ic->c_rf);
807 nvdwtype = fr->ntype;
809 vdwtype = mdatoms->typeA;
811 /* Setup water-specific parameters */
812 inr = nlist->iinr[0];
813 iq0 = _mm256_mul_ps(facel,_mm256_set1_ps(charge[inr+0]));
814 iq1 = _mm256_mul_ps(facel,_mm256_set1_ps(charge[inr+1]));
815 iq2 = _mm256_mul_ps(facel,_mm256_set1_ps(charge[inr+2]));
816 vdwioffsetptr0 = vdwparam+2*nvdwtype*vdwtype[inr+0];
818 /* When we use explicit cutoffs the value must be identical for elec and VdW, so use elec as an arbitrary choice */
819 rcutoff_scalar = fr->rcoulomb;
820 rcutoff = _mm256_set1_ps(rcutoff_scalar);
821 rcutoff2 = _mm256_mul_ps(rcutoff,rcutoff);
823 rswitch_scalar = fr->rvdw_switch;
824 rswitch = _mm256_set1_ps(rswitch_scalar);
825 /* Setup switch parameters */
826 d_scalar = rcutoff_scalar-rswitch_scalar;
827 d = _mm256_set1_ps(d_scalar);
828 swV3 = _mm256_set1_ps(-10.0/(d_scalar*d_scalar*d_scalar));
829 swV4 = _mm256_set1_ps( 15.0/(d_scalar*d_scalar*d_scalar*d_scalar));
830 swV5 = _mm256_set1_ps( -6.0/(d_scalar*d_scalar*d_scalar*d_scalar*d_scalar));
831 swF2 = _mm256_set1_ps(-30.0/(d_scalar*d_scalar*d_scalar));
832 swF3 = _mm256_set1_ps( 60.0/(d_scalar*d_scalar*d_scalar*d_scalar));
833 swF4 = _mm256_set1_ps(-30.0/(d_scalar*d_scalar*d_scalar*d_scalar*d_scalar));
835 /* Avoid stupid compiler warnings */
836 jnrA = jnrB = jnrC = jnrD = jnrE = jnrF = jnrG = jnrH = 0;
849 for(iidx=0;iidx<4*DIM;iidx++)
854 /* Start outer loop over neighborlists */
855 for(iidx=0; iidx<nri; iidx++)
857 /* Load shift vector for this list */
858 i_shift_offset = DIM*shiftidx[iidx];
860 /* Load limits for loop over neighbors */
861 j_index_start = jindex[iidx];
862 j_index_end = jindex[iidx+1];
864 /* Get outer coordinate index */
866 i_coord_offset = DIM*inr;
868 /* Load i particle coords and add shift vector */
869 gmx_mm256_load_shift_and_3rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,
870 &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2);
872 fix0 = _mm256_setzero_ps();
873 fiy0 = _mm256_setzero_ps();
874 fiz0 = _mm256_setzero_ps();
875 fix1 = _mm256_setzero_ps();
876 fiy1 = _mm256_setzero_ps();
877 fiz1 = _mm256_setzero_ps();
878 fix2 = _mm256_setzero_ps();
879 fiy2 = _mm256_setzero_ps();
880 fiz2 = _mm256_setzero_ps();
882 /* Start inner kernel loop */
883 for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+7]>=0; jidx+=8)
886 /* Get j neighbor index, and coordinate index */
895 j_coord_offsetA = DIM*jnrA;
896 j_coord_offsetB = DIM*jnrB;
897 j_coord_offsetC = DIM*jnrC;
898 j_coord_offsetD = DIM*jnrD;
899 j_coord_offsetE = DIM*jnrE;
900 j_coord_offsetF = DIM*jnrF;
901 j_coord_offsetG = DIM*jnrG;
902 j_coord_offsetH = DIM*jnrH;
904 /* load j atom coordinates */
905 gmx_mm256_load_1rvec_8ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
906 x+j_coord_offsetC,x+j_coord_offsetD,
907 x+j_coord_offsetE,x+j_coord_offsetF,
908 x+j_coord_offsetG,x+j_coord_offsetH,
911 /* Calculate displacement vector */
912 dx00 = _mm256_sub_ps(ix0,jx0);
913 dy00 = _mm256_sub_ps(iy0,jy0);
914 dz00 = _mm256_sub_ps(iz0,jz0);
915 dx10 = _mm256_sub_ps(ix1,jx0);
916 dy10 = _mm256_sub_ps(iy1,jy0);
917 dz10 = _mm256_sub_ps(iz1,jz0);
918 dx20 = _mm256_sub_ps(ix2,jx0);
919 dy20 = _mm256_sub_ps(iy2,jy0);
920 dz20 = _mm256_sub_ps(iz2,jz0);
922 /* Calculate squared distance and things based on it */
923 rsq00 = gmx_mm256_calc_rsq_ps(dx00,dy00,dz00);
924 rsq10 = gmx_mm256_calc_rsq_ps(dx10,dy10,dz10);
925 rsq20 = gmx_mm256_calc_rsq_ps(dx20,dy20,dz20);
927 rinv00 = gmx_mm256_invsqrt_ps(rsq00);
928 rinv10 = gmx_mm256_invsqrt_ps(rsq10);
929 rinv20 = gmx_mm256_invsqrt_ps(rsq20);
931 rinvsq00 = _mm256_mul_ps(rinv00,rinv00);
932 rinvsq10 = _mm256_mul_ps(rinv10,rinv10);
933 rinvsq20 = _mm256_mul_ps(rinv20,rinv20);
935 /* Load parameters for j particles */
936 jq0 = gmx_mm256_load_8real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
937 charge+jnrC+0,charge+jnrD+0,
938 charge+jnrE+0,charge+jnrF+0,
939 charge+jnrG+0,charge+jnrH+0);
940 vdwjidx0A = 2*vdwtype[jnrA+0];
941 vdwjidx0B = 2*vdwtype[jnrB+0];
942 vdwjidx0C = 2*vdwtype[jnrC+0];
943 vdwjidx0D = 2*vdwtype[jnrD+0];
944 vdwjidx0E = 2*vdwtype[jnrE+0];
945 vdwjidx0F = 2*vdwtype[jnrF+0];
946 vdwjidx0G = 2*vdwtype[jnrG+0];
947 vdwjidx0H = 2*vdwtype[jnrH+0];
949 fjx0 = _mm256_setzero_ps();
950 fjy0 = _mm256_setzero_ps();
951 fjz0 = _mm256_setzero_ps();
953 /**************************
954 * CALCULATE INTERACTIONS *
955 **************************/
957 if (gmx_mm256_any_lt(rsq00,rcutoff2))
960 r00 = _mm256_mul_ps(rsq00,rinv00);
962 /* Compute parameters for interactions between i and j atoms */
963 qq00 = _mm256_mul_ps(iq0,jq0);
964 gmx_mm256_load_8pair_swizzle_ps(vdwioffsetptr0+vdwjidx0A,
965 vdwioffsetptr0+vdwjidx0B,
966 vdwioffsetptr0+vdwjidx0C,
967 vdwioffsetptr0+vdwjidx0D,
968 vdwioffsetptr0+vdwjidx0E,
969 vdwioffsetptr0+vdwjidx0F,
970 vdwioffsetptr0+vdwjidx0G,
971 vdwioffsetptr0+vdwjidx0H,
974 /* REACTION-FIELD ELECTROSTATICS */
975 felec = _mm256_mul_ps(qq00,_mm256_sub_ps(_mm256_mul_ps(rinv00,rinvsq00),krf2));
977 /* LENNARD-JONES DISPERSION/REPULSION */
979 rinvsix = _mm256_mul_ps(_mm256_mul_ps(rinvsq00,rinvsq00),rinvsq00);
980 vvdw6 = _mm256_mul_ps(c6_00,rinvsix);
981 vvdw12 = _mm256_mul_ps(c12_00,_mm256_mul_ps(rinvsix,rinvsix));
982 vvdw = _mm256_sub_ps( _mm256_mul_ps(vvdw12,one_twelfth) , _mm256_mul_ps(vvdw6,one_sixth) );
983 fvdw = _mm256_mul_ps(_mm256_sub_ps(vvdw12,vvdw6),rinvsq00);
985 d = _mm256_sub_ps(r00,rswitch);
986 d = _mm256_max_ps(d,_mm256_setzero_ps());
987 d2 = _mm256_mul_ps(d,d);
988 sw = _mm256_add_ps(one,_mm256_mul_ps(d2,_mm256_mul_ps(d,_mm256_add_ps(swV3,_mm256_mul_ps(d,_mm256_add_ps(swV4,_mm256_mul_ps(d,swV5)))))));
990 dsw = _mm256_mul_ps(d2,_mm256_add_ps(swF2,_mm256_mul_ps(d,_mm256_add_ps(swF3,_mm256_mul_ps(d,swF4)))));
992 /* Evaluate switch function */
993 /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
994 fvdw = _mm256_sub_ps( _mm256_mul_ps(fvdw,sw) , _mm256_mul_ps(rinv00,_mm256_mul_ps(vvdw,dsw)) );
995 cutoff_mask = _mm256_cmp_ps(rsq00,rcutoff2,_CMP_LT_OQ);
997 fscal = _mm256_add_ps(felec,fvdw);
999 fscal = _mm256_and_ps(fscal,cutoff_mask);
1001 /* Calculate temporary vectorial force */
1002 tx = _mm256_mul_ps(fscal,dx00);
1003 ty = _mm256_mul_ps(fscal,dy00);
1004 tz = _mm256_mul_ps(fscal,dz00);
1006 /* Update vectorial force */
1007 fix0 = _mm256_add_ps(fix0,tx);
1008 fiy0 = _mm256_add_ps(fiy0,ty);
1009 fiz0 = _mm256_add_ps(fiz0,tz);
1011 fjx0 = _mm256_add_ps(fjx0,tx);
1012 fjy0 = _mm256_add_ps(fjy0,ty);
1013 fjz0 = _mm256_add_ps(fjz0,tz);
1017 /**************************
1018 * CALCULATE INTERACTIONS *
1019 **************************/
1021 if (gmx_mm256_any_lt(rsq10,rcutoff2))
1024 /* Compute parameters for interactions between i and j atoms */
1025 qq10 = _mm256_mul_ps(iq1,jq0);
1027 /* REACTION-FIELD ELECTROSTATICS */
1028 felec = _mm256_mul_ps(qq10,_mm256_sub_ps(_mm256_mul_ps(rinv10,rinvsq10),krf2));
1030 cutoff_mask = _mm256_cmp_ps(rsq10,rcutoff2,_CMP_LT_OQ);
1034 fscal = _mm256_and_ps(fscal,cutoff_mask);
1036 /* Calculate temporary vectorial force */
1037 tx = _mm256_mul_ps(fscal,dx10);
1038 ty = _mm256_mul_ps(fscal,dy10);
1039 tz = _mm256_mul_ps(fscal,dz10);
1041 /* Update vectorial force */
1042 fix1 = _mm256_add_ps(fix1,tx);
1043 fiy1 = _mm256_add_ps(fiy1,ty);
1044 fiz1 = _mm256_add_ps(fiz1,tz);
1046 fjx0 = _mm256_add_ps(fjx0,tx);
1047 fjy0 = _mm256_add_ps(fjy0,ty);
1048 fjz0 = _mm256_add_ps(fjz0,tz);
1052 /**************************
1053 * CALCULATE INTERACTIONS *
1054 **************************/
1056 if (gmx_mm256_any_lt(rsq20,rcutoff2))
1059 /* Compute parameters for interactions between i and j atoms */
1060 qq20 = _mm256_mul_ps(iq2,jq0);
1062 /* REACTION-FIELD ELECTROSTATICS */
1063 felec = _mm256_mul_ps(qq20,_mm256_sub_ps(_mm256_mul_ps(rinv20,rinvsq20),krf2));
1065 cutoff_mask = _mm256_cmp_ps(rsq20,rcutoff2,_CMP_LT_OQ);
1069 fscal = _mm256_and_ps(fscal,cutoff_mask);
1071 /* Calculate temporary vectorial force */
1072 tx = _mm256_mul_ps(fscal,dx20);
1073 ty = _mm256_mul_ps(fscal,dy20);
1074 tz = _mm256_mul_ps(fscal,dz20);
1076 /* Update vectorial force */
1077 fix2 = _mm256_add_ps(fix2,tx);
1078 fiy2 = _mm256_add_ps(fiy2,ty);
1079 fiz2 = _mm256_add_ps(fiz2,tz);
1081 fjx0 = _mm256_add_ps(fjx0,tx);
1082 fjy0 = _mm256_add_ps(fjy0,ty);
1083 fjz0 = _mm256_add_ps(fjz0,tz);
1087 fjptrA = f+j_coord_offsetA;
1088 fjptrB = f+j_coord_offsetB;
1089 fjptrC = f+j_coord_offsetC;
1090 fjptrD = f+j_coord_offsetD;
1091 fjptrE = f+j_coord_offsetE;
1092 fjptrF = f+j_coord_offsetF;
1093 fjptrG = f+j_coord_offsetG;
1094 fjptrH = f+j_coord_offsetH;
1096 gmx_mm256_decrement_1rvec_8ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjptrE,fjptrF,fjptrG,fjptrH,fjx0,fjy0,fjz0);
1098 /* Inner loop uses 124 flops */
1101 if(jidx<j_index_end)
1104 /* Get j neighbor index, and coordinate index */
1105 jnrlistA = jjnr[jidx];
1106 jnrlistB = jjnr[jidx+1];
1107 jnrlistC = jjnr[jidx+2];
1108 jnrlistD = jjnr[jidx+3];
1109 jnrlistE = jjnr[jidx+4];
1110 jnrlistF = jjnr[jidx+5];
1111 jnrlistG = jjnr[jidx+6];
1112 jnrlistH = jjnr[jidx+7];
1113 /* Sign of each element will be negative for non-real atoms.
1114 * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
1115 * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
1117 dummy_mask = gmx_mm256_set_m128(gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx+4)),_mm_setzero_si128())),
1118 gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128())));
1120 jnrA = (jnrlistA>=0) ? jnrlistA : 0;
1121 jnrB = (jnrlistB>=0) ? jnrlistB : 0;
1122 jnrC = (jnrlistC>=0) ? jnrlistC : 0;
1123 jnrD = (jnrlistD>=0) ? jnrlistD : 0;
1124 jnrE = (jnrlistE>=0) ? jnrlistE : 0;
1125 jnrF = (jnrlistF>=0) ? jnrlistF : 0;
1126 jnrG = (jnrlistG>=0) ? jnrlistG : 0;
1127 jnrH = (jnrlistH>=0) ? jnrlistH : 0;
1128 j_coord_offsetA = DIM*jnrA;
1129 j_coord_offsetB = DIM*jnrB;
1130 j_coord_offsetC = DIM*jnrC;
1131 j_coord_offsetD = DIM*jnrD;
1132 j_coord_offsetE = DIM*jnrE;
1133 j_coord_offsetF = DIM*jnrF;
1134 j_coord_offsetG = DIM*jnrG;
1135 j_coord_offsetH = DIM*jnrH;
1137 /* load j atom coordinates */
1138 gmx_mm256_load_1rvec_8ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
1139 x+j_coord_offsetC,x+j_coord_offsetD,
1140 x+j_coord_offsetE,x+j_coord_offsetF,
1141 x+j_coord_offsetG,x+j_coord_offsetH,
1144 /* Calculate displacement vector */
1145 dx00 = _mm256_sub_ps(ix0,jx0);
1146 dy00 = _mm256_sub_ps(iy0,jy0);
1147 dz00 = _mm256_sub_ps(iz0,jz0);
1148 dx10 = _mm256_sub_ps(ix1,jx0);
1149 dy10 = _mm256_sub_ps(iy1,jy0);
1150 dz10 = _mm256_sub_ps(iz1,jz0);
1151 dx20 = _mm256_sub_ps(ix2,jx0);
1152 dy20 = _mm256_sub_ps(iy2,jy0);
1153 dz20 = _mm256_sub_ps(iz2,jz0);
1155 /* Calculate squared distance and things based on it */
1156 rsq00 = gmx_mm256_calc_rsq_ps(dx00,dy00,dz00);
1157 rsq10 = gmx_mm256_calc_rsq_ps(dx10,dy10,dz10);
1158 rsq20 = gmx_mm256_calc_rsq_ps(dx20,dy20,dz20);
1160 rinv00 = gmx_mm256_invsqrt_ps(rsq00);
1161 rinv10 = gmx_mm256_invsqrt_ps(rsq10);
1162 rinv20 = gmx_mm256_invsqrt_ps(rsq20);
1164 rinvsq00 = _mm256_mul_ps(rinv00,rinv00);
1165 rinvsq10 = _mm256_mul_ps(rinv10,rinv10);
1166 rinvsq20 = _mm256_mul_ps(rinv20,rinv20);
1168 /* Load parameters for j particles */
1169 jq0 = gmx_mm256_load_8real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
1170 charge+jnrC+0,charge+jnrD+0,
1171 charge+jnrE+0,charge+jnrF+0,
1172 charge+jnrG+0,charge+jnrH+0);
1173 vdwjidx0A = 2*vdwtype[jnrA+0];
1174 vdwjidx0B = 2*vdwtype[jnrB+0];
1175 vdwjidx0C = 2*vdwtype[jnrC+0];
1176 vdwjidx0D = 2*vdwtype[jnrD+0];
1177 vdwjidx0E = 2*vdwtype[jnrE+0];
1178 vdwjidx0F = 2*vdwtype[jnrF+0];
1179 vdwjidx0G = 2*vdwtype[jnrG+0];
1180 vdwjidx0H = 2*vdwtype[jnrH+0];
1182 fjx0 = _mm256_setzero_ps();
1183 fjy0 = _mm256_setzero_ps();
1184 fjz0 = _mm256_setzero_ps();
1186 /**************************
1187 * CALCULATE INTERACTIONS *
1188 **************************/
1190 if (gmx_mm256_any_lt(rsq00,rcutoff2))
1193 r00 = _mm256_mul_ps(rsq00,rinv00);
1194 r00 = _mm256_andnot_ps(dummy_mask,r00);
1196 /* Compute parameters for interactions between i and j atoms */
1197 qq00 = _mm256_mul_ps(iq0,jq0);
1198 gmx_mm256_load_8pair_swizzle_ps(vdwioffsetptr0+vdwjidx0A,
1199 vdwioffsetptr0+vdwjidx0B,
1200 vdwioffsetptr0+vdwjidx0C,
1201 vdwioffsetptr0+vdwjidx0D,
1202 vdwioffsetptr0+vdwjidx0E,
1203 vdwioffsetptr0+vdwjidx0F,
1204 vdwioffsetptr0+vdwjidx0G,
1205 vdwioffsetptr0+vdwjidx0H,
1208 /* REACTION-FIELD ELECTROSTATICS */
1209 felec = _mm256_mul_ps(qq00,_mm256_sub_ps(_mm256_mul_ps(rinv00,rinvsq00),krf2));
1211 /* LENNARD-JONES DISPERSION/REPULSION */
1213 rinvsix = _mm256_mul_ps(_mm256_mul_ps(rinvsq00,rinvsq00),rinvsq00);
1214 vvdw6 = _mm256_mul_ps(c6_00,rinvsix);
1215 vvdw12 = _mm256_mul_ps(c12_00,_mm256_mul_ps(rinvsix,rinvsix));
1216 vvdw = _mm256_sub_ps( _mm256_mul_ps(vvdw12,one_twelfth) , _mm256_mul_ps(vvdw6,one_sixth) );
1217 fvdw = _mm256_mul_ps(_mm256_sub_ps(vvdw12,vvdw6),rinvsq00);
1219 d = _mm256_sub_ps(r00,rswitch);
1220 d = _mm256_max_ps(d,_mm256_setzero_ps());
1221 d2 = _mm256_mul_ps(d,d);
1222 sw = _mm256_add_ps(one,_mm256_mul_ps(d2,_mm256_mul_ps(d,_mm256_add_ps(swV3,_mm256_mul_ps(d,_mm256_add_ps(swV4,_mm256_mul_ps(d,swV5)))))));
1224 dsw = _mm256_mul_ps(d2,_mm256_add_ps(swF2,_mm256_mul_ps(d,_mm256_add_ps(swF3,_mm256_mul_ps(d,swF4)))));
1226 /* Evaluate switch function */
1227 /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
1228 fvdw = _mm256_sub_ps( _mm256_mul_ps(fvdw,sw) , _mm256_mul_ps(rinv00,_mm256_mul_ps(vvdw,dsw)) );
1229 cutoff_mask = _mm256_cmp_ps(rsq00,rcutoff2,_CMP_LT_OQ);
1231 fscal = _mm256_add_ps(felec,fvdw);
1233 fscal = _mm256_and_ps(fscal,cutoff_mask);
1235 fscal = _mm256_andnot_ps(dummy_mask,fscal);
1237 /* Calculate temporary vectorial force */
1238 tx = _mm256_mul_ps(fscal,dx00);
1239 ty = _mm256_mul_ps(fscal,dy00);
1240 tz = _mm256_mul_ps(fscal,dz00);
1242 /* Update vectorial force */
1243 fix0 = _mm256_add_ps(fix0,tx);
1244 fiy0 = _mm256_add_ps(fiy0,ty);
1245 fiz0 = _mm256_add_ps(fiz0,tz);
1247 fjx0 = _mm256_add_ps(fjx0,tx);
1248 fjy0 = _mm256_add_ps(fjy0,ty);
1249 fjz0 = _mm256_add_ps(fjz0,tz);
1253 /**************************
1254 * CALCULATE INTERACTIONS *
1255 **************************/
1257 if (gmx_mm256_any_lt(rsq10,rcutoff2))
1260 /* Compute parameters for interactions between i and j atoms */
1261 qq10 = _mm256_mul_ps(iq1,jq0);
1263 /* REACTION-FIELD ELECTROSTATICS */
1264 felec = _mm256_mul_ps(qq10,_mm256_sub_ps(_mm256_mul_ps(rinv10,rinvsq10),krf2));
1266 cutoff_mask = _mm256_cmp_ps(rsq10,rcutoff2,_CMP_LT_OQ);
1270 fscal = _mm256_and_ps(fscal,cutoff_mask);
1272 fscal = _mm256_andnot_ps(dummy_mask,fscal);
1274 /* Calculate temporary vectorial force */
1275 tx = _mm256_mul_ps(fscal,dx10);
1276 ty = _mm256_mul_ps(fscal,dy10);
1277 tz = _mm256_mul_ps(fscal,dz10);
1279 /* Update vectorial force */
1280 fix1 = _mm256_add_ps(fix1,tx);
1281 fiy1 = _mm256_add_ps(fiy1,ty);
1282 fiz1 = _mm256_add_ps(fiz1,tz);
1284 fjx0 = _mm256_add_ps(fjx0,tx);
1285 fjy0 = _mm256_add_ps(fjy0,ty);
1286 fjz0 = _mm256_add_ps(fjz0,tz);
1290 /**************************
1291 * CALCULATE INTERACTIONS *
1292 **************************/
1294 if (gmx_mm256_any_lt(rsq20,rcutoff2))
1297 /* Compute parameters for interactions between i and j atoms */
1298 qq20 = _mm256_mul_ps(iq2,jq0);
1300 /* REACTION-FIELD ELECTROSTATICS */
1301 felec = _mm256_mul_ps(qq20,_mm256_sub_ps(_mm256_mul_ps(rinv20,rinvsq20),krf2));
1303 cutoff_mask = _mm256_cmp_ps(rsq20,rcutoff2,_CMP_LT_OQ);
1307 fscal = _mm256_and_ps(fscal,cutoff_mask);
1309 fscal = _mm256_andnot_ps(dummy_mask,fscal);
1311 /* Calculate temporary vectorial force */
1312 tx = _mm256_mul_ps(fscal,dx20);
1313 ty = _mm256_mul_ps(fscal,dy20);
1314 tz = _mm256_mul_ps(fscal,dz20);
1316 /* Update vectorial force */
1317 fix2 = _mm256_add_ps(fix2,tx);
1318 fiy2 = _mm256_add_ps(fiy2,ty);
1319 fiz2 = _mm256_add_ps(fiz2,tz);
1321 fjx0 = _mm256_add_ps(fjx0,tx);
1322 fjy0 = _mm256_add_ps(fjy0,ty);
1323 fjz0 = _mm256_add_ps(fjz0,tz);
1327 fjptrA = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
1328 fjptrB = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
1329 fjptrC = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
1330 fjptrD = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
1331 fjptrE = (jnrlistE>=0) ? f+j_coord_offsetE : scratch;
1332 fjptrF = (jnrlistF>=0) ? f+j_coord_offsetF : scratch;
1333 fjptrG = (jnrlistG>=0) ? f+j_coord_offsetG : scratch;
1334 fjptrH = (jnrlistH>=0) ? f+j_coord_offsetH : scratch;
1336 gmx_mm256_decrement_1rvec_8ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjptrE,fjptrF,fjptrG,fjptrH,fjx0,fjy0,fjz0);
1338 /* Inner loop uses 125 flops */
1341 /* End of innermost loop */
1343 gmx_mm256_update_iforce_3atom_swizzle_ps(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,
1344 f+i_coord_offset,fshift+i_shift_offset);
1346 /* Increment number of inner iterations */
1347 inneriter += j_index_end - j_index_start;
1349 /* Outer loop uses 18 flops */
1352 /* Increment number of outer iterations */
1355 /* Update outer/inner flops */
1357 inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W3_F,outeriter*18 + inneriter*125);