src/gromacs/gmxlib/nonbonded/nb_kernel_avx_256_single/nb_kernel_ElecRFCut_VdwLJSh_GeomW3P1_avx_256_single.c

   1 /*
   2  * This file is part of the GROMACS molecular simulation package.
   3  *
   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.
   8  *
   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.
  13  *
  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.
  18  *
  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.
  23  *
  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.
  31  *
  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.
  34  */
  35 /*
  36  * Note: this file was generated by the GROMACS avx_256_single kernel generator.
  37  */
  38 #ifdef HAVE_CONFIG_H
  39 #include <config.h>
  40 #endif
  41
  42 #include <math.h>
  43
  44 #include "../nb_kernel.h"
  45 #include "types/simple.h"
  46 #include "vec.h"
  47 #include "nrnb.h"
  48
  49 #include "gromacs/simd/math_x86_avx_256_single.h"
  50 #include "kernelutil_x86_avx_256_single.h"
  51
  52 /*
  53  * Gromacs nonbonded kernel:   nb_kernel_ElecRFCut_VdwLJSh_GeomW3P1_VF_avx_256_single
  54  * Electrostatics interaction: ReactionField
  55  * VdW interaction:            LennardJones
  56  * Geometry:                   Water3-Particle
  57  * Calculate force/pot:        PotentialAndForce
  58  */
  59 void
  60 nb_kernel_ElecRFCut_VdwLJSh_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)
  68 {
  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.
  73      */
  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;
  83     real             rcutoff_scalar;
  84     real             *shiftvec,*fshift,*x,*f;
  85     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD,*fjptrE,*fjptrF,*fjptrG,*fjptrH;
  86     real             scratch[4*DIM];
  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;
 100     real             *charge;
 101     int              nvdwtype;
 102     __m256           rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
 103     int              *vdwtype;
 104     real             *vdwparam;
 105     __m256           one_sixth   = _mm256_set1_ps(1.0/6.0);
 106     __m256           one_twelfth = _mm256_set1_ps(1.0/12.0);
 107     __m256           dummy_mask,cutoff_mask;
 108     __m256           signbit = _mm256_castsi256_ps( _mm256_set1_epi32(0x80000000) );
 109     __m256           one     = _mm256_set1_ps(1.0);
 110     __m256           two     = _mm256_set1_ps(2.0);
 111     x                = xx[0];
 112     f                = ff[0];
 113
 114     nri              = nlist->nri;
 115     iinr             = nlist->iinr;
 116     jindex           = nlist->jindex;
 117     jjnr             = nlist->jjnr;
 118     shiftidx         = nlist->shift;
 119     gid              = nlist->gid;
 120     shiftvec         = fr->shift_vec[0];
 121     fshift           = fr->fshift[0];
 122     facel            = _mm256_set1_ps(fr->epsfac);
 123     charge           = mdatoms->chargeA;
 124     krf              = _mm256_set1_ps(fr->ic->k_rf);
 125     krf2             = _mm256_set1_ps(fr->ic->k_rf*2.0);
 126     crf              = _mm256_set1_ps(fr->ic->c_rf);
 127     nvdwtype         = fr->ntype;
 128     vdwparam         = fr->nbfp;
 129     vdwtype          = mdatoms->typeA;
 130
 131     /* Setup water-specific parameters */
 132     inr              = nlist->iinr[0];
 133     iq0              = _mm256_mul_ps(facel,_mm256_set1_ps(charge[inr+0]));
 134     iq1              = _mm256_mul_ps(facel,_mm256_set1_ps(charge[inr+1]));
 135     iq2              = _mm256_mul_ps(facel,_mm256_set1_ps(charge[inr+2]));
 136     vdwioffsetptr0   = vdwparam+2*nvdwtype*vdwtype[inr+0];
 137
 138     /* When we use explicit cutoffs the value must be identical for elec and VdW, so use elec as an arbitrary choice */
 139     rcutoff_scalar   = fr->rcoulomb;
 140     rcutoff          = _mm256_set1_ps(rcutoff_scalar);
 141     rcutoff2         = _mm256_mul_ps(rcutoff,rcutoff);
 142
 143     sh_vdw_invrcut6  = _mm256_set1_ps(fr->ic->sh_invrc6);
 144     rvdw             = _mm256_set1_ps(fr->rvdw);
 145
 146     /* Avoid stupid compiler warnings */
 147     jnrA = jnrB = jnrC = jnrD = jnrE = jnrF = jnrG = jnrH = 0;
 148     j_coord_offsetA = 0;
 149     j_coord_offsetB = 0;
 150     j_coord_offsetC = 0;
 151     j_coord_offsetD = 0;
 152     j_coord_offsetE = 0;
 153     j_coord_offsetF = 0;
 154     j_coord_offsetG = 0;
 155     j_coord_offsetH = 0;
 156
 157     outeriter        = 0;
 158     inneriter        = 0;
 159
 160     for(iidx=0;iidx<4*DIM;iidx++)
 161     {
 162         scratch[iidx] = 0.0;
 163     }
 164
 165     /* Start outer loop over neighborlists */
 166     for(iidx=0; iidx<nri; iidx++)
 167     {
 168         /* Load shift vector for this list */
 169         i_shift_offset   = DIM*shiftidx[iidx];
 170
 171         /* Load limits for loop over neighbors */
 172         j_index_start    = jindex[iidx];
 173         j_index_end      = jindex[iidx+1];
 174
 175         /* Get outer coordinate index */
 176         inr              = iinr[iidx];
 177         i_coord_offset   = DIM*inr;
 178
 179         /* Load i particle coords and add shift vector */
 180         gmx_mm256_load_shift_and_3rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,
 181                                                     &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2);
 182
 183         fix0             = _mm256_setzero_ps();
 184         fiy0             = _mm256_setzero_ps();
 185         fiz0             = _mm256_setzero_ps();
 186         fix1             = _mm256_setzero_ps();
 187         fiy1             = _mm256_setzero_ps();
 188         fiz1             = _mm256_setzero_ps();
 189         fix2             = _mm256_setzero_ps();
 190         fiy2             = _mm256_setzero_ps();
 191         fiz2             = _mm256_setzero_ps();
 192
 193         /* Reset potential sums */
 194         velecsum         = _mm256_setzero_ps();
 195         vvdwsum          = _mm256_setzero_ps();
 196
 197         /* Start inner kernel loop */
 198         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+7]>=0; jidx+=8)
 199         {
 200
 201             /* Get j neighbor index, and coordinate index */
 202             jnrA             = jjnr[jidx];
 203             jnrB             = jjnr[jidx+1];
 204             jnrC             = jjnr[jidx+2];
 205             jnrD             = jjnr[jidx+3];
 206             jnrE             = jjnr[jidx+4];
 207             jnrF             = jjnr[jidx+5];
 208             jnrG             = jjnr[jidx+6];
 209             jnrH             = jjnr[jidx+7];
 210             j_coord_offsetA  = DIM*jnrA;
 211             j_coord_offsetB  = DIM*jnrB;
 212             j_coord_offsetC  = DIM*jnrC;
 213             j_coord_offsetD  = DIM*jnrD;
 214             j_coord_offsetE  = DIM*jnrE;
 215             j_coord_offsetF  = DIM*jnrF;
 216             j_coord_offsetG  = DIM*jnrG;
 217             j_coord_offsetH  = DIM*jnrH;
 218
 219             /* load j atom coordinates */
 220             gmx_mm256_load_1rvec_8ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 221                                                  x+j_coord_offsetC,x+j_coord_offsetD,
 222                                                  x+j_coord_offsetE,x+j_coord_offsetF,
 223                                                  x+j_coord_offsetG,x+j_coord_offsetH,
 224                                                  &jx0,&jy0,&jz0);
 225
 226             /* Calculate displacement vector */
 227             dx00             = _mm256_sub_ps(ix0,jx0);
 228             dy00             = _mm256_sub_ps(iy0,jy0);
 229             dz00             = _mm256_sub_ps(iz0,jz0);
 230             dx10             = _mm256_sub_ps(ix1,jx0);
 231             dy10             = _mm256_sub_ps(iy1,jy0);
 232             dz10             = _mm256_sub_ps(iz1,jz0);
 233             dx20             = _mm256_sub_ps(ix2,jx0);
 234             dy20             = _mm256_sub_ps(iy2,jy0);
 235             dz20             = _mm256_sub_ps(iz2,jz0);
 236
 237             /* Calculate squared distance and things based on it */
 238             rsq00            = gmx_mm256_calc_rsq_ps(dx00,dy00,dz00);
 239             rsq10            = gmx_mm256_calc_rsq_ps(dx10,dy10,dz10);
 240             rsq20            = gmx_mm256_calc_rsq_ps(dx20,dy20,dz20);
 241
 242             rinv00           = gmx_mm256_invsqrt_ps(rsq00);
 243             rinv10           = gmx_mm256_invsqrt_ps(rsq10);
 244             rinv20           = gmx_mm256_invsqrt_ps(rsq20);
 245
 246             rinvsq00         = _mm256_mul_ps(rinv00,rinv00);
 247             rinvsq10         = _mm256_mul_ps(rinv10,rinv10);
 248             rinvsq20         = _mm256_mul_ps(rinv20,rinv20);
 249
 250             /* Load parameters for j particles */
 251             jq0              = gmx_mm256_load_8real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 252                                                                  charge+jnrC+0,charge+jnrD+0,
 253                                                                  charge+jnrE+0,charge+jnrF+0,
 254                                                                  charge+jnrG+0,charge+jnrH+0);
 255             vdwjidx0A        = 2*vdwtype[jnrA+0];
 256             vdwjidx0B        = 2*vdwtype[jnrB+0];
 257             vdwjidx0C        = 2*vdwtype[jnrC+0];
 258             vdwjidx0D        = 2*vdwtype[jnrD+0];
 259             vdwjidx0E        = 2*vdwtype[jnrE+0];
 260             vdwjidx0F        = 2*vdwtype[jnrF+0];
 261             vdwjidx0G        = 2*vdwtype[jnrG+0];
 262             vdwjidx0H        = 2*vdwtype[jnrH+0];
 263
 264             fjx0             = _mm256_setzero_ps();
 265             fjy0             = _mm256_setzero_ps();
 266             fjz0             = _mm256_setzero_ps();
 267
 268             /**************************
 269              * CALCULATE INTERACTIONS *
 270              **************************/
 271
 272             if (gmx_mm256_any_lt(rsq00,rcutoff2))
 273             {
 274
 275             /* Compute parameters for interactions between i and j atoms */
 276             qq00             = _mm256_mul_ps(iq0,jq0);
 277             gmx_mm256_load_8pair_swizzle_ps(vdwioffsetptr0+vdwjidx0A,
 278                                             vdwioffsetptr0+vdwjidx0B,
 279                                             vdwioffsetptr0+vdwjidx0C,
 280                                             vdwioffsetptr0+vdwjidx0D,
 281                                             vdwioffsetptr0+vdwjidx0E,
 282                                             vdwioffsetptr0+vdwjidx0F,
 283                                             vdwioffsetptr0+vdwjidx0G,
 284                                             vdwioffsetptr0+vdwjidx0H,
 285                                             &c6_00,&c12_00);
 286
 287             /* REACTION-FIELD ELECTROSTATICS */
 288             velec            = _mm256_mul_ps(qq00,_mm256_sub_ps(_mm256_add_ps(rinv00,_mm256_mul_ps(krf,rsq00)),crf));
 289             felec            = _mm256_mul_ps(qq00,_mm256_sub_ps(_mm256_mul_ps(rinv00,rinvsq00),krf2));
 290
 291             /* LENNARD-JONES DISPERSION/REPULSION */
 292
 293             rinvsix          = _mm256_mul_ps(_mm256_mul_ps(rinvsq00,rinvsq00),rinvsq00);
 294             vvdw6            = _mm256_mul_ps(c6_00,rinvsix);
 295             vvdw12           = _mm256_mul_ps(c12_00,_mm256_mul_ps(rinvsix,rinvsix));
 296             vvdw             = _mm256_sub_ps(_mm256_mul_ps( _mm256_sub_ps(vvdw12 , _mm256_mul_ps(c12_00,_mm256_mul_ps(sh_vdw_invrcut6,sh_vdw_invrcut6))), one_twelfth) ,
 297                                           _mm256_mul_ps( _mm256_sub_ps(vvdw6,_mm256_mul_ps(c6_00,sh_vdw_invrcut6)),one_sixth));
 298             fvdw             = _mm256_mul_ps(_mm256_sub_ps(vvdw12,vvdw6),rinvsq00);
 299
 300             cutoff_mask      = _mm256_cmp_ps(rsq00,rcutoff2,_CMP_LT_OQ);
 301
 302             /* Update potential sum for this i atom from the interaction with this j atom. */
 303             velec            = _mm256_and_ps(velec,cutoff_mask);
 304             velecsum         = _mm256_add_ps(velecsum,velec);
 305             vvdw             = _mm256_and_ps(vvdw,cutoff_mask);
 306             vvdwsum          = _mm256_add_ps(vvdwsum,vvdw);
 307
 308             fscal            = _mm256_add_ps(felec,fvdw);
 309
 310             fscal            = _mm256_and_ps(fscal,cutoff_mask);
 311
 312             /* Calculate temporary vectorial force */
 313             tx               = _mm256_mul_ps(fscal,dx00);
 314             ty               = _mm256_mul_ps(fscal,dy00);
 315             tz               = _mm256_mul_ps(fscal,dz00);
 316
 317             /* Update vectorial force */
 318             fix0             = _mm256_add_ps(fix0,tx);
 319             fiy0             = _mm256_add_ps(fiy0,ty);
 320             fiz0             = _mm256_add_ps(fiz0,tz);
 321
 322             fjx0             = _mm256_add_ps(fjx0,tx);
 323             fjy0             = _mm256_add_ps(fjy0,ty);
 324             fjz0             = _mm256_add_ps(fjz0,tz);
 325
 326             }
 327
 328             /**************************
 329              * CALCULATE INTERACTIONS *
 330              **************************/
 331
 332             if (gmx_mm256_any_lt(rsq10,rcutoff2))
 333             {
 334
 335             /* Compute parameters for interactions between i and j atoms */
 336             qq10             = _mm256_mul_ps(iq1,jq0);
 337
 338             /* REACTION-FIELD ELECTROSTATICS */
 339             velec            = _mm256_mul_ps(qq10,_mm256_sub_ps(_mm256_add_ps(rinv10,_mm256_mul_ps(krf,rsq10)),crf));
 340             felec            = _mm256_mul_ps(qq10,_mm256_sub_ps(_mm256_mul_ps(rinv10,rinvsq10),krf2));
 341
 342             cutoff_mask      = _mm256_cmp_ps(rsq10,rcutoff2,_CMP_LT_OQ);
 343
 344             /* Update potential sum for this i atom from the interaction with this j atom. */
 345             velec            = _mm256_and_ps(velec,cutoff_mask);
 346             velecsum         = _mm256_add_ps(velecsum,velec);
 347
 348             fscal            = felec;
 349
 350             fscal            = _mm256_and_ps(fscal,cutoff_mask);
 351
 352             /* Calculate temporary vectorial force */
 353             tx               = _mm256_mul_ps(fscal,dx10);
 354             ty               = _mm256_mul_ps(fscal,dy10);
 355             tz               = _mm256_mul_ps(fscal,dz10);
 356
 357             /* Update vectorial force */
 358             fix1             = _mm256_add_ps(fix1,tx);
 359             fiy1             = _mm256_add_ps(fiy1,ty);
 360             fiz1             = _mm256_add_ps(fiz1,tz);
 361
 362             fjx0             = _mm256_add_ps(fjx0,tx);
 363             fjy0             = _mm256_add_ps(fjy0,ty);
 364             fjz0             = _mm256_add_ps(fjz0,tz);
 365
 366             }
 367
 368             /**************************
 369              * CALCULATE INTERACTIONS *
 370              **************************/
 371
 372             if (gmx_mm256_any_lt(rsq20,rcutoff2))
 373             {
 374
 375             /* Compute parameters for interactions between i and j atoms */
 376             qq20             = _mm256_mul_ps(iq2,jq0);
 377
 378             /* REACTION-FIELD ELECTROSTATICS */
 379             velec            = _mm256_mul_ps(qq20,_mm256_sub_ps(_mm256_add_ps(rinv20,_mm256_mul_ps(krf,rsq20)),crf));
 380             felec            = _mm256_mul_ps(qq20,_mm256_sub_ps(_mm256_mul_ps(rinv20,rinvsq20),krf2));
 381
 382             cutoff_mask      = _mm256_cmp_ps(rsq20,rcutoff2,_CMP_LT_OQ);
 383
 384             /* Update potential sum for this i atom from the interaction with this j atom. */
 385             velec            = _mm256_and_ps(velec,cutoff_mask);
 386             velecsum         = _mm256_add_ps(velecsum,velec);
 387
 388             fscal            = felec;
 389
 390             fscal            = _mm256_and_ps(fscal,cutoff_mask);
 391
 392             /* Calculate temporary vectorial force */
 393             tx               = _mm256_mul_ps(fscal,dx20);
 394             ty               = _mm256_mul_ps(fscal,dy20);
 395             tz               = _mm256_mul_ps(fscal,dz20);
 396
 397             /* Update vectorial force */
 398             fix2             = _mm256_add_ps(fix2,tx);
 399             fiy2             = _mm256_add_ps(fiy2,ty);
 400             fiz2             = _mm256_add_ps(fiz2,tz);
 401
 402             fjx0             = _mm256_add_ps(fjx0,tx);
 403             fjy0             = _mm256_add_ps(fjy0,ty);
 404             fjz0             = _mm256_add_ps(fjz0,tz);
 405
 406             }
 407
 408             fjptrA             = f+j_coord_offsetA;
 409             fjptrB             = f+j_coord_offsetB;
 410             fjptrC             = f+j_coord_offsetC;
 411             fjptrD             = f+j_coord_offsetD;
 412             fjptrE             = f+j_coord_offsetE;
 413             fjptrF             = f+j_coord_offsetF;
 414             fjptrG             = f+j_coord_offsetG;
 415             fjptrH             = f+j_coord_offsetH;
 416
 417             gmx_mm256_decrement_1rvec_8ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjptrE,fjptrF,fjptrG,fjptrH,fjx0,fjy0,fjz0);
 418
 419             /* Inner loop uses 129 flops */
 420         }
 421
 422         if(jidx<j_index_end)
 423         {
 424
 425             /* Get j neighbor index, and coordinate index */
 426             jnrlistA         = jjnr[jidx];
 427             jnrlistB         = jjnr[jidx+1];
 428             jnrlistC         = jjnr[jidx+2];
 429             jnrlistD         = jjnr[jidx+3];
 430             jnrlistE         = jjnr[jidx+4];
 431             jnrlistF         = jjnr[jidx+5];
 432             jnrlistG         = jjnr[jidx+6];
 433             jnrlistH         = jjnr[jidx+7];
 434             /* Sign of each element will be negative for non-real atoms.
 435              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 436              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
 437              */
 438             dummy_mask = gmx_mm256_set_m128(gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx+4)),_mm_setzero_si128())),
 439                                             gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128())));
 440
 441             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
 442             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
 443             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
 444             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
 445             jnrE       = (jnrlistE>=0) ? jnrlistE : 0;
 446             jnrF       = (jnrlistF>=0) ? jnrlistF : 0;
 447             jnrG       = (jnrlistG>=0) ? jnrlistG : 0;
 448             jnrH       = (jnrlistH>=0) ? jnrlistH : 0;
 449             j_coord_offsetA  = DIM*jnrA;
 450             j_coord_offsetB  = DIM*jnrB;
 451             j_coord_offsetC  = DIM*jnrC;
 452             j_coord_offsetD  = DIM*jnrD;
 453             j_coord_offsetE  = DIM*jnrE;
 454             j_coord_offsetF  = DIM*jnrF;
 455             j_coord_offsetG  = DIM*jnrG;
 456             j_coord_offsetH  = DIM*jnrH;
 457
 458             /* load j atom coordinates */
 459             gmx_mm256_load_1rvec_8ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 460                                                  x+j_coord_offsetC,x+j_coord_offsetD,
 461                                                  x+j_coord_offsetE,x+j_coord_offsetF,
 462                                                  x+j_coord_offsetG,x+j_coord_offsetH,
 463                                                  &jx0,&jy0,&jz0);
 464
 465             /* Calculate displacement vector */
 466             dx00             = _mm256_sub_ps(ix0,jx0);
 467             dy00             = _mm256_sub_ps(iy0,jy0);
 468             dz00             = _mm256_sub_ps(iz0,jz0);
 469             dx10             = _mm256_sub_ps(ix1,jx0);
 470             dy10             = _mm256_sub_ps(iy1,jy0);
 471             dz10             = _mm256_sub_ps(iz1,jz0);
 472             dx20             = _mm256_sub_ps(ix2,jx0);
 473             dy20             = _mm256_sub_ps(iy2,jy0);
 474             dz20             = _mm256_sub_ps(iz2,jz0);
 475
 476             /* Calculate squared distance and things based on it */
 477             rsq00            = gmx_mm256_calc_rsq_ps(dx00,dy00,dz00);
 478             rsq10            = gmx_mm256_calc_rsq_ps(dx10,dy10,dz10);
 479             rsq20            = gmx_mm256_calc_rsq_ps(dx20,dy20,dz20);
 480
 481             rinv00           = gmx_mm256_invsqrt_ps(rsq00);
 482             rinv10           = gmx_mm256_invsqrt_ps(rsq10);
 483             rinv20           = gmx_mm256_invsqrt_ps(rsq20);
 484
 485             rinvsq00         = _mm256_mul_ps(rinv00,rinv00);
 486             rinvsq10         = _mm256_mul_ps(rinv10,rinv10);
 487             rinvsq20         = _mm256_mul_ps(rinv20,rinv20);
 488
 489             /* Load parameters for j particles */
 490             jq0              = gmx_mm256_load_8real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 491                                                                  charge+jnrC+0,charge+jnrD+0,
 492                                                                  charge+jnrE+0,charge+jnrF+0,
 493                                                                  charge+jnrG+0,charge+jnrH+0);
 494             vdwjidx0A        = 2*vdwtype[jnrA+0];
 495             vdwjidx0B        = 2*vdwtype[jnrB+0];
 496             vdwjidx0C        = 2*vdwtype[jnrC+0];
 497             vdwjidx0D        = 2*vdwtype[jnrD+0];
 498             vdwjidx0E        = 2*vdwtype[jnrE+0];
 499             vdwjidx0F        = 2*vdwtype[jnrF+0];
 500             vdwjidx0G        = 2*vdwtype[jnrG+0];
 501             vdwjidx0H        = 2*vdwtype[jnrH+0];
 502
 503             fjx0             = _mm256_setzero_ps();
 504             fjy0             = _mm256_setzero_ps();
 505             fjz0             = _mm256_setzero_ps();
 506
 507             /**************************
 508              * CALCULATE INTERACTIONS *
 509              **************************/
 510
 511             if (gmx_mm256_any_lt(rsq00,rcutoff2))
 512             {
 513
 514             /* Compute parameters for interactions between i and j atoms */
 515             qq00             = _mm256_mul_ps(iq0,jq0);
 516             gmx_mm256_load_8pair_swizzle_ps(vdwioffsetptr0+vdwjidx0A,
 517                                             vdwioffsetptr0+vdwjidx0B,
 518                                             vdwioffsetptr0+vdwjidx0C,
 519                                             vdwioffsetptr0+vdwjidx0D,
 520                                             vdwioffsetptr0+vdwjidx0E,
 521                                             vdwioffsetptr0+vdwjidx0F,
 522                                             vdwioffsetptr0+vdwjidx0G,
 523                                             vdwioffsetptr0+vdwjidx0H,
 524                                             &c6_00,&c12_00);
 525
 526             /* REACTION-FIELD ELECTROSTATICS */
 527             velec            = _mm256_mul_ps(qq00,_mm256_sub_ps(_mm256_add_ps(rinv00,_mm256_mul_ps(krf,rsq00)),crf));
 528             felec            = _mm256_mul_ps(qq00,_mm256_sub_ps(_mm256_mul_ps(rinv00,rinvsq00),krf2));
 529
 530             /* LENNARD-JONES DISPERSION/REPULSION */
 531
 532             rinvsix          = _mm256_mul_ps(_mm256_mul_ps(rinvsq00,rinvsq00),rinvsq00);
 533             vvdw6            = _mm256_mul_ps(c6_00,rinvsix);
 534             vvdw12           = _mm256_mul_ps(c12_00,_mm256_mul_ps(rinvsix,rinvsix));
 535             vvdw             = _mm256_sub_ps(_mm256_mul_ps( _mm256_sub_ps(vvdw12 , _mm256_mul_ps(c12_00,_mm256_mul_ps(sh_vdw_invrcut6,sh_vdw_invrcut6))), one_twelfth) ,
 536                                           _mm256_mul_ps( _mm256_sub_ps(vvdw6,_mm256_mul_ps(c6_00,sh_vdw_invrcut6)),one_sixth));
 537             fvdw             = _mm256_mul_ps(_mm256_sub_ps(vvdw12,vvdw6),rinvsq00);
 538
 539             cutoff_mask      = _mm256_cmp_ps(rsq00,rcutoff2,_CMP_LT_OQ);
 540
 541             /* Update potential sum for this i atom from the interaction with this j atom. */
 542             velec            = _mm256_and_ps(velec,cutoff_mask);
 543             velec            = _mm256_andnot_ps(dummy_mask,velec);
 544             velecsum         = _mm256_add_ps(velecsum,velec);
 545             vvdw             = _mm256_and_ps(vvdw,cutoff_mask);
 546             vvdw             = _mm256_andnot_ps(dummy_mask,vvdw);
 547             vvdwsum          = _mm256_add_ps(vvdwsum,vvdw);
 548
 549             fscal            = _mm256_add_ps(felec,fvdw);
 550
 551             fscal            = _mm256_and_ps(fscal,cutoff_mask);
 552
 553             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
 554
 555             /* Calculate temporary vectorial force */
 556             tx               = _mm256_mul_ps(fscal,dx00);
 557             ty               = _mm256_mul_ps(fscal,dy00);
 558             tz               = _mm256_mul_ps(fscal,dz00);
 559
 560             /* Update vectorial force */
 561             fix0             = _mm256_add_ps(fix0,tx);
 562             fiy0             = _mm256_add_ps(fiy0,ty);
 563             fiz0             = _mm256_add_ps(fiz0,tz);
 564
 565             fjx0             = _mm256_add_ps(fjx0,tx);
 566             fjy0             = _mm256_add_ps(fjy0,ty);
 567             fjz0             = _mm256_add_ps(fjz0,tz);
 568
 569             }
 570
 571             /**************************
 572              * CALCULATE INTERACTIONS *
 573              **************************/
 574
 575             if (gmx_mm256_any_lt(rsq10,rcutoff2))
 576             {
 577
 578             /* Compute parameters for interactions between i and j atoms */
 579             qq10             = _mm256_mul_ps(iq1,jq0);
 580
 581             /* REACTION-FIELD ELECTROSTATICS */
 582             velec            = _mm256_mul_ps(qq10,_mm256_sub_ps(_mm256_add_ps(rinv10,_mm256_mul_ps(krf,rsq10)),crf));
 583             felec            = _mm256_mul_ps(qq10,_mm256_sub_ps(_mm256_mul_ps(rinv10,rinvsq10),krf2));
 584
 585             cutoff_mask      = _mm256_cmp_ps(rsq10,rcutoff2,_CMP_LT_OQ);
 586
 587             /* Update potential sum for this i atom from the interaction with this j atom. */
 588             velec            = _mm256_and_ps(velec,cutoff_mask);
 589             velec            = _mm256_andnot_ps(dummy_mask,velec);
 590             velecsum         = _mm256_add_ps(velecsum,velec);
 591
 592             fscal            = felec;
 593
 594             fscal            = _mm256_and_ps(fscal,cutoff_mask);
 595
 596             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
 597
 598             /* Calculate temporary vectorial force */
 599             tx               = _mm256_mul_ps(fscal,dx10);
 600             ty               = _mm256_mul_ps(fscal,dy10);
 601             tz               = _mm256_mul_ps(fscal,dz10);
 602
 603             /* Update vectorial force */
 604             fix1             = _mm256_add_ps(fix1,tx);
 605             fiy1             = _mm256_add_ps(fiy1,ty);
 606             fiz1             = _mm256_add_ps(fiz1,tz);
 607
 608             fjx0             = _mm256_add_ps(fjx0,tx);
 609             fjy0             = _mm256_add_ps(fjy0,ty);
 610             fjz0             = _mm256_add_ps(fjz0,tz);
 611
 612             }
 613
 614             /**************************
 615              * CALCULATE INTERACTIONS *
 616              **************************/
 617
 618             if (gmx_mm256_any_lt(rsq20,rcutoff2))
 619             {
 620
 621             /* Compute parameters for interactions between i and j atoms */
 622             qq20             = _mm256_mul_ps(iq2,jq0);
 623
 624             /* REACTION-FIELD ELECTROSTATICS */
 625             velec            = _mm256_mul_ps(qq20,_mm256_sub_ps(_mm256_add_ps(rinv20,_mm256_mul_ps(krf,rsq20)),crf));
 626             felec            = _mm256_mul_ps(qq20,_mm256_sub_ps(_mm256_mul_ps(rinv20,rinvsq20),krf2));
 627
 628             cutoff_mask      = _mm256_cmp_ps(rsq20,rcutoff2,_CMP_LT_OQ);
 629
 630             /* Update potential sum for this i atom from the interaction with this j atom. */
 631             velec            = _mm256_and_ps(velec,cutoff_mask);
 632             velec            = _mm256_andnot_ps(dummy_mask,velec);
 633             velecsum         = _mm256_add_ps(velecsum,velec);
 634
 635             fscal            = felec;
 636
 637             fscal            = _mm256_and_ps(fscal,cutoff_mask);
 638
 639             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
 640
 641             /* Calculate temporary vectorial force */
 642             tx               = _mm256_mul_ps(fscal,dx20);
 643             ty               = _mm256_mul_ps(fscal,dy20);
 644             tz               = _mm256_mul_ps(fscal,dz20);
 645
 646             /* Update vectorial force */
 647             fix2             = _mm256_add_ps(fix2,tx);
 648             fiy2             = _mm256_add_ps(fiy2,ty);
 649             fiz2             = _mm256_add_ps(fiz2,tz);
 650
 651             fjx0             = _mm256_add_ps(fjx0,tx);
 652             fjy0             = _mm256_add_ps(fjy0,ty);
 653             fjz0             = _mm256_add_ps(fjz0,tz);
 654
 655             }
 656
 657             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
 658             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
 659             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
 660             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
 661             fjptrE             = (jnrlistE>=0) ? f+j_coord_offsetE : scratch;
 662             fjptrF             = (jnrlistF>=0) ? f+j_coord_offsetF : scratch;
 663             fjptrG             = (jnrlistG>=0) ? f+j_coord_offsetG : scratch;
 664             fjptrH             = (jnrlistH>=0) ? f+j_coord_offsetH : scratch;
 665
 666             gmx_mm256_decrement_1rvec_8ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjptrE,fjptrF,fjptrG,fjptrH,fjx0,fjy0,fjz0);
 667
 668             /* Inner loop uses 129 flops */
 669         }
 670
 671         /* End of innermost loop */
 672
 673         gmx_mm256_update_iforce_3atom_swizzle_ps(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,
 674                                                  f+i_coord_offset,fshift+i_shift_offset);
 675
 676         ggid                        = gid[iidx];
 677         /* Update potential energies */
 678         gmx_mm256_update_1pot_ps(velecsum,kernel_data->energygrp_elec+ggid);
 679         gmx_mm256_update_1pot_ps(vvdwsum,kernel_data->energygrp_vdw+ggid);
 680
 681         /* Increment number of inner iterations */
 682         inneriter                  += j_index_end - j_index_start;
 683
 684         /* Outer loop uses 20 flops */
 685     }
 686
 687     /* Increment number of outer iterations */
 688     outeriter        += nri;
 689
 690     /* Update outer/inner flops */
 691
 692     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W3_VF,outeriter*20 + inneriter*129);
 693 }
 694 /*
 695  * Gromacs nonbonded kernel:   nb_kernel_ElecRFCut_VdwLJSh_GeomW3P1_F_avx_256_single
 696  * Electrostatics interaction: ReactionField
 697  * VdW interaction:            LennardJones
 698  * Geometry:                   Water3-Particle
 699  * Calculate force/pot:        Force
 700  */
 701 void
 702 nb_kernel_ElecRFCut_VdwLJSh_GeomW3P1_F_avx_256_single
 703                     (t_nblist                    * gmx_restrict       nlist,
 704                      rvec                        * gmx_restrict          xx,
 705                      rvec                        * gmx_restrict          ff,
 706                      t_forcerec                  * gmx_restrict          fr,
 707                      t_mdatoms                   * gmx_restrict     mdatoms,
 708                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
 709                      t_nrnb                      * gmx_restrict        nrnb)
 710 {
 711     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
 712      * just 0 for non-waters.
 713      * Suffixes A,B,C,D,E,F,G,H refer to j loop unrolling done with AVX, e.g. for the eight different
 714      * jnr indices corresponding to data put in the four positions in the SIMD register.
 715      */
 716     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
 717     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
 718     int              jnrA,jnrB,jnrC,jnrD;
 719     int              jnrE,jnrF,jnrG,jnrH;
 720     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
 721     int              jnrlistE,jnrlistF,jnrlistG,jnrlistH;
 722     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
 723     int              j_coord_offsetE,j_coord_offsetF,j_coord_offsetG,j_coord_offsetH;
 724     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
 725     real             rcutoff_scalar;
 726     real             *shiftvec,*fshift,*x,*f;
 727     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD,*fjptrE,*fjptrF,*fjptrG,*fjptrH;
 728     real             scratch[4*DIM];
 729     __m256           tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
 730     real *           vdwioffsetptr0;
 731     __m256           ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
 732     real *           vdwioffsetptr1;
 733     __m256           ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
 734     real *           vdwioffsetptr2;
 735     __m256           ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
 736     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D,vdwjidx0E,vdwjidx0F,vdwjidx0G,vdwjidx0H;
 737     __m256           jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
 738     __m256           dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
 739     __m256           dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
 740     __m256           dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
 741     __m256           velec,felec,velecsum,facel,crf,krf,krf2;
 742     real             *charge;
 743     int              nvdwtype;
 744     __m256           rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
 745     int              *vdwtype;
 746     real             *vdwparam;
 747     __m256           one_sixth   = _mm256_set1_ps(1.0/6.0);
 748     __m256           one_twelfth = _mm256_set1_ps(1.0/12.0);
 749     __m256           dummy_mask,cutoff_mask;
 750     __m256           signbit = _mm256_castsi256_ps( _mm256_set1_epi32(0x80000000) );
 751     __m256           one     = _mm256_set1_ps(1.0);
 752     __m256           two     = _mm256_set1_ps(2.0);
 753     x                = xx[0];
 754     f                = ff[0];
 755
 756     nri              = nlist->nri;
 757     iinr             = nlist->iinr;
 758     jindex           = nlist->jindex;
 759     jjnr             = nlist->jjnr;
 760     shiftidx         = nlist->shift;
 761     gid              = nlist->gid;
 762     shiftvec         = fr->shift_vec[0];
 763     fshift           = fr->fshift[0];
 764     facel            = _mm256_set1_ps(fr->epsfac);
 765     charge           = mdatoms->chargeA;
 766     krf              = _mm256_set1_ps(fr->ic->k_rf);
 767     krf2             = _mm256_set1_ps(fr->ic->k_rf*2.0);
 768     crf              = _mm256_set1_ps(fr->ic->c_rf);
 769     nvdwtype         = fr->ntype;
 770     vdwparam         = fr->nbfp;
 771     vdwtype          = mdatoms->typeA;
 772
 773     /* Setup water-specific parameters */
 774     inr              = nlist->iinr[0];
 775     iq0              = _mm256_mul_ps(facel,_mm256_set1_ps(charge[inr+0]));
 776     iq1              = _mm256_mul_ps(facel,_mm256_set1_ps(charge[inr+1]));
 777     iq2              = _mm256_mul_ps(facel,_mm256_set1_ps(charge[inr+2]));
 778     vdwioffsetptr0   = vdwparam+2*nvdwtype*vdwtype[inr+0];
 779
 780     /* When we use explicit cutoffs the value must be identical for elec and VdW, so use elec as an arbitrary choice */
 781     rcutoff_scalar   = fr->rcoulomb;
 782     rcutoff          = _mm256_set1_ps(rcutoff_scalar);
 783     rcutoff2         = _mm256_mul_ps(rcutoff,rcutoff);
 784
 785     sh_vdw_invrcut6  = _mm256_set1_ps(fr->ic->sh_invrc6);
 786     rvdw             = _mm256_set1_ps(fr->rvdw);
 787
 788     /* Avoid stupid compiler warnings */
 789     jnrA = jnrB = jnrC = jnrD = jnrE = jnrF = jnrG = jnrH = 0;
 790     j_coord_offsetA = 0;
 791     j_coord_offsetB = 0;
 792     j_coord_offsetC = 0;
 793     j_coord_offsetD = 0;
 794     j_coord_offsetE = 0;
 795     j_coord_offsetF = 0;
 796     j_coord_offsetG = 0;
 797     j_coord_offsetH = 0;
 798
 799     outeriter        = 0;
 800     inneriter        = 0;
 801
 802     for(iidx=0;iidx<4*DIM;iidx++)
 803     {
 804         scratch[iidx] = 0.0;
 805     }
 806
 807     /* Start outer loop over neighborlists */
 808     for(iidx=0; iidx<nri; iidx++)
 809     {
 810         /* Load shift vector for this list */
 811         i_shift_offset   = DIM*shiftidx[iidx];
 812
 813         /* Load limits for loop over neighbors */
 814         j_index_start    = jindex[iidx];
 815         j_index_end      = jindex[iidx+1];
 816
 817         /* Get outer coordinate index */
 818         inr              = iinr[iidx];
 819         i_coord_offset   = DIM*inr;
 820
 821         /* Load i particle coords and add shift vector */
 822         gmx_mm256_load_shift_and_3rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,
 823                                                     &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2);
 824
 825         fix0             = _mm256_setzero_ps();
 826         fiy0             = _mm256_setzero_ps();
 827         fiz0             = _mm256_setzero_ps();
 828         fix1             = _mm256_setzero_ps();
 829         fiy1             = _mm256_setzero_ps();
 830         fiz1             = _mm256_setzero_ps();
 831         fix2             = _mm256_setzero_ps();
 832         fiy2             = _mm256_setzero_ps();
 833         fiz2             = _mm256_setzero_ps();
 834
 835         /* Start inner kernel loop */
 836         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+7]>=0; jidx+=8)
 837         {
 838
 839             /* Get j neighbor index, and coordinate index */
 840             jnrA             = jjnr[jidx];
 841             jnrB             = jjnr[jidx+1];
 842             jnrC             = jjnr[jidx+2];
 843             jnrD             = jjnr[jidx+3];
 844             jnrE             = jjnr[jidx+4];
 845             jnrF             = jjnr[jidx+5];
 846             jnrG             = jjnr[jidx+6];
 847             jnrH             = jjnr[jidx+7];
 848             j_coord_offsetA  = DIM*jnrA;
 849             j_coord_offsetB  = DIM*jnrB;
 850             j_coord_offsetC  = DIM*jnrC;
 851             j_coord_offsetD  = DIM*jnrD;
 852             j_coord_offsetE  = DIM*jnrE;
 853             j_coord_offsetF  = DIM*jnrF;
 854             j_coord_offsetG  = DIM*jnrG;
 855             j_coord_offsetH  = DIM*jnrH;
 856
 857             /* load j atom coordinates */
 858             gmx_mm256_load_1rvec_8ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 859                                                  x+j_coord_offsetC,x+j_coord_offsetD,
 860                                                  x+j_coord_offsetE,x+j_coord_offsetF,
 861                                                  x+j_coord_offsetG,x+j_coord_offsetH,
 862                                                  &jx0,&jy0,&jz0);
 863
 864             /* Calculate displacement vector */
 865             dx00             = _mm256_sub_ps(ix0,jx0);
 866             dy00             = _mm256_sub_ps(iy0,jy0);
 867             dz00             = _mm256_sub_ps(iz0,jz0);
 868             dx10             = _mm256_sub_ps(ix1,jx0);
 869             dy10             = _mm256_sub_ps(iy1,jy0);
 870             dz10             = _mm256_sub_ps(iz1,jz0);
 871             dx20             = _mm256_sub_ps(ix2,jx0);
 872             dy20             = _mm256_sub_ps(iy2,jy0);
 873             dz20             = _mm256_sub_ps(iz2,jz0);
 874
 875             /* Calculate squared distance and things based on it */
 876             rsq00            = gmx_mm256_calc_rsq_ps(dx00,dy00,dz00);
 877             rsq10            = gmx_mm256_calc_rsq_ps(dx10,dy10,dz10);
 878             rsq20            = gmx_mm256_calc_rsq_ps(dx20,dy20,dz20);
 879
 880             rinv00           = gmx_mm256_invsqrt_ps(rsq00);
 881             rinv10           = gmx_mm256_invsqrt_ps(rsq10);
 882             rinv20           = gmx_mm256_invsqrt_ps(rsq20);
 883
 884             rinvsq00         = _mm256_mul_ps(rinv00,rinv00);
 885             rinvsq10         = _mm256_mul_ps(rinv10,rinv10);
 886             rinvsq20         = _mm256_mul_ps(rinv20,rinv20);
 887
 888             /* Load parameters for j particles */
 889             jq0              = gmx_mm256_load_8real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 890                                                                  charge+jnrC+0,charge+jnrD+0,
 891                                                                  charge+jnrE+0,charge+jnrF+0,
 892                                                                  charge+jnrG+0,charge+jnrH+0);
 893             vdwjidx0A        = 2*vdwtype[jnrA+0];
 894             vdwjidx0B        = 2*vdwtype[jnrB+0];
 895             vdwjidx0C        = 2*vdwtype[jnrC+0];
 896             vdwjidx0D        = 2*vdwtype[jnrD+0];
 897             vdwjidx0E        = 2*vdwtype[jnrE+0];
 898             vdwjidx0F        = 2*vdwtype[jnrF+0];
 899             vdwjidx0G        = 2*vdwtype[jnrG+0];
 900             vdwjidx0H        = 2*vdwtype[jnrH+0];
 901
 902             fjx0             = _mm256_setzero_ps();
 903             fjy0             = _mm256_setzero_ps();
 904             fjz0             = _mm256_setzero_ps();
 905
 906             /**************************
 907              * CALCULATE INTERACTIONS *
 908              **************************/
 909
 910             if (gmx_mm256_any_lt(rsq00,rcutoff2))
 911             {
 912
 913             /* Compute parameters for interactions between i and j atoms */
 914             qq00             = _mm256_mul_ps(iq0,jq0);
 915             gmx_mm256_load_8pair_swizzle_ps(vdwioffsetptr0+vdwjidx0A,
 916                                             vdwioffsetptr0+vdwjidx0B,
 917                                             vdwioffsetptr0+vdwjidx0C,
 918                                             vdwioffsetptr0+vdwjidx0D,
 919                                             vdwioffsetptr0+vdwjidx0E,
 920                                             vdwioffsetptr0+vdwjidx0F,
 921                                             vdwioffsetptr0+vdwjidx0G,
 922                                             vdwioffsetptr0+vdwjidx0H,
 923                                             &c6_00,&c12_00);
 924
 925             /* REACTION-FIELD ELECTROSTATICS */
 926             felec            = _mm256_mul_ps(qq00,_mm256_sub_ps(_mm256_mul_ps(rinv00,rinvsq00),krf2));
 927
 928             /* LENNARD-JONES DISPERSION/REPULSION */
 929
 930             rinvsix          = _mm256_mul_ps(_mm256_mul_ps(rinvsq00,rinvsq00),rinvsq00);
 931             fvdw             = _mm256_mul_ps(_mm256_sub_ps(_mm256_mul_ps(c12_00,rinvsix),c6_00),_mm256_mul_ps(rinvsix,rinvsq00));
 932
 933             cutoff_mask      = _mm256_cmp_ps(rsq00,rcutoff2,_CMP_LT_OQ);
 934
 935             fscal            = _mm256_add_ps(felec,fvdw);
 936
 937             fscal            = _mm256_and_ps(fscal,cutoff_mask);
 938
 939             /* Calculate temporary vectorial force */
 940             tx               = _mm256_mul_ps(fscal,dx00);
 941             ty               = _mm256_mul_ps(fscal,dy00);
 942             tz               = _mm256_mul_ps(fscal,dz00);
 943
 944             /* Update vectorial force */
 945             fix0             = _mm256_add_ps(fix0,tx);
 946             fiy0             = _mm256_add_ps(fiy0,ty);
 947             fiz0             = _mm256_add_ps(fiz0,tz);
 948
 949             fjx0             = _mm256_add_ps(fjx0,tx);
 950             fjy0             = _mm256_add_ps(fjy0,ty);
 951             fjz0             = _mm256_add_ps(fjz0,tz);
 952
 953             }
 954
 955             /**************************
 956              * CALCULATE INTERACTIONS *
 957              **************************/
 958
 959             if (gmx_mm256_any_lt(rsq10,rcutoff2))
 960             {
 961
 962             /* Compute parameters for interactions between i and j atoms */
 963             qq10             = _mm256_mul_ps(iq1,jq0);
 964
 965             /* REACTION-FIELD ELECTROSTATICS */
 966             felec            = _mm256_mul_ps(qq10,_mm256_sub_ps(_mm256_mul_ps(rinv10,rinvsq10),krf2));
 967
 968             cutoff_mask      = _mm256_cmp_ps(rsq10,rcutoff2,_CMP_LT_OQ);
 969
 970             fscal            = felec;
 971
 972             fscal            = _mm256_and_ps(fscal,cutoff_mask);
 973
 974             /* Calculate temporary vectorial force */
 975             tx               = _mm256_mul_ps(fscal,dx10);
 976             ty               = _mm256_mul_ps(fscal,dy10);
 977             tz               = _mm256_mul_ps(fscal,dz10);
 978
 979             /* Update vectorial force */
 980             fix1             = _mm256_add_ps(fix1,tx);
 981             fiy1             = _mm256_add_ps(fiy1,ty);
 982             fiz1             = _mm256_add_ps(fiz1,tz);
 983
 984             fjx0             = _mm256_add_ps(fjx0,tx);
 985             fjy0             = _mm256_add_ps(fjy0,ty);
 986             fjz0             = _mm256_add_ps(fjz0,tz);
 987
 988             }
 989
 990             /**************************
 991              * CALCULATE INTERACTIONS *
 992              **************************/
 993
 994             if (gmx_mm256_any_lt(rsq20,rcutoff2))
 995             {
 996
 997             /* Compute parameters for interactions between i and j atoms */
 998             qq20             = _mm256_mul_ps(iq2,jq0);
 999
1000             /* REACTION-FIELD ELECTROSTATICS */
1001             felec            = _mm256_mul_ps(qq20,_mm256_sub_ps(_mm256_mul_ps(rinv20,rinvsq20),krf2));
1002
1003             cutoff_mask      = _mm256_cmp_ps(rsq20,rcutoff2,_CMP_LT_OQ);
1004
1005             fscal            = felec;
1006
1007             fscal            = _mm256_and_ps(fscal,cutoff_mask);
1008
1009             /* Calculate temporary vectorial force */
1010             tx               = _mm256_mul_ps(fscal,dx20);
1011             ty               = _mm256_mul_ps(fscal,dy20);
1012             tz               = _mm256_mul_ps(fscal,dz20);
1013
1014             /* Update vectorial force */
1015             fix2             = _mm256_add_ps(fix2,tx);
1016             fiy2             = _mm256_add_ps(fiy2,ty);
1017             fiz2             = _mm256_add_ps(fiz2,tz);
1018
1019             fjx0             = _mm256_add_ps(fjx0,tx);
1020             fjy0             = _mm256_add_ps(fjy0,ty);
1021             fjz0             = _mm256_add_ps(fjz0,tz);
1022
1023             }
1024
1025             fjptrA             = f+j_coord_offsetA;
1026             fjptrB             = f+j_coord_offsetB;
1027             fjptrC             = f+j_coord_offsetC;
1028             fjptrD             = f+j_coord_offsetD;
1029             fjptrE             = f+j_coord_offsetE;
1030             fjptrF             = f+j_coord_offsetF;
1031             fjptrG             = f+j_coord_offsetG;
1032             fjptrH             = f+j_coord_offsetH;
1033
1034             gmx_mm256_decrement_1rvec_8ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjptrE,fjptrF,fjptrG,fjptrH,fjx0,fjy0,fjz0);
1035
1036             /* Inner loop uses 100 flops */
1037         }
1038
1039         if(jidx<j_index_end)
1040         {
1041
1042             /* Get j neighbor index, and coordinate index */
1043             jnrlistA         = jjnr[jidx];
1044             jnrlistB         = jjnr[jidx+1];
1045             jnrlistC         = jjnr[jidx+2];
1046             jnrlistD         = jjnr[jidx+3];
1047             jnrlistE         = jjnr[jidx+4];
1048             jnrlistF         = jjnr[jidx+5];
1049             jnrlistG         = jjnr[jidx+6];
1050             jnrlistH         = jjnr[jidx+7];
1051             /* Sign of each element will be negative for non-real atoms.
1052              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
1053              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
1054              */
1055             dummy_mask = gmx_mm256_set_m128(gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx+4)),_mm_setzero_si128())),
1056                                             gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128())));
1057
1058             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
1059             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
1060             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
1061             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
1062             jnrE       = (jnrlistE>=0) ? jnrlistE : 0;
1063             jnrF       = (jnrlistF>=0) ? jnrlistF : 0;
1064             jnrG       = (jnrlistG>=0) ? jnrlistG : 0;
1065             jnrH       = (jnrlistH>=0) ? jnrlistH : 0;
1066             j_coord_offsetA  = DIM*jnrA;
1067             j_coord_offsetB  = DIM*jnrB;
1068             j_coord_offsetC  = DIM*jnrC;
1069             j_coord_offsetD  = DIM*jnrD;
1070             j_coord_offsetE  = DIM*jnrE;
1071             j_coord_offsetF  = DIM*jnrF;
1072             j_coord_offsetG  = DIM*jnrG;
1073             j_coord_offsetH  = DIM*jnrH;
1074
1075             /* load j atom coordinates */
1076             gmx_mm256_load_1rvec_8ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
1077                                                  x+j_coord_offsetC,x+j_coord_offsetD,
1078                                                  x+j_coord_offsetE,x+j_coord_offsetF,
1079                                                  x+j_coord_offsetG,x+j_coord_offsetH,
1080                                                  &jx0,&jy0,&jz0);
1081
1082             /* Calculate displacement vector */
1083             dx00             = _mm256_sub_ps(ix0,jx0);
1084             dy00             = _mm256_sub_ps(iy0,jy0);
1085             dz00             = _mm256_sub_ps(iz0,jz0);
1086             dx10             = _mm256_sub_ps(ix1,jx0);
1087             dy10             = _mm256_sub_ps(iy1,jy0);
1088             dz10             = _mm256_sub_ps(iz1,jz0);
1089             dx20             = _mm256_sub_ps(ix2,jx0);
1090             dy20             = _mm256_sub_ps(iy2,jy0);
1091             dz20             = _mm256_sub_ps(iz2,jz0);
1092
1093             /* Calculate squared distance and things based on it */
1094             rsq00            = gmx_mm256_calc_rsq_ps(dx00,dy00,dz00);
1095             rsq10            = gmx_mm256_calc_rsq_ps(dx10,dy10,dz10);
1096             rsq20            = gmx_mm256_calc_rsq_ps(dx20,dy20,dz20);
1097
1098             rinv00           = gmx_mm256_invsqrt_ps(rsq00);
1099             rinv10           = gmx_mm256_invsqrt_ps(rsq10);
1100             rinv20           = gmx_mm256_invsqrt_ps(rsq20);
1101
1102             rinvsq00         = _mm256_mul_ps(rinv00,rinv00);
1103             rinvsq10         = _mm256_mul_ps(rinv10,rinv10);
1104             rinvsq20         = _mm256_mul_ps(rinv20,rinv20);
1105
1106             /* Load parameters for j particles */
1107             jq0              = gmx_mm256_load_8real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
1108                                                                  charge+jnrC+0,charge+jnrD+0,
1109                                                                  charge+jnrE+0,charge+jnrF+0,
1110                                                                  charge+jnrG+0,charge+jnrH+0);
1111             vdwjidx0A        = 2*vdwtype[jnrA+0];
1112             vdwjidx0B        = 2*vdwtype[jnrB+0];
1113             vdwjidx0C        = 2*vdwtype[jnrC+0];
1114             vdwjidx0D        = 2*vdwtype[jnrD+0];
1115             vdwjidx0E        = 2*vdwtype[jnrE+0];
1116             vdwjidx0F        = 2*vdwtype[jnrF+0];
1117             vdwjidx0G        = 2*vdwtype[jnrG+0];
1118             vdwjidx0H        = 2*vdwtype[jnrH+0];
1119
1120             fjx0             = _mm256_setzero_ps();
1121             fjy0             = _mm256_setzero_ps();
1122             fjz0             = _mm256_setzero_ps();
1123
1124             /**************************
1125              * CALCULATE INTERACTIONS *
1126              **************************/
1127
1128             if (gmx_mm256_any_lt(rsq00,rcutoff2))
1129             {
1130
1131             /* Compute parameters for interactions between i and j atoms */
1132             qq00             = _mm256_mul_ps(iq0,jq0);
1133             gmx_mm256_load_8pair_swizzle_ps(vdwioffsetptr0+vdwjidx0A,
1134                                             vdwioffsetptr0+vdwjidx0B,
1135                                             vdwioffsetptr0+vdwjidx0C,
1136                                             vdwioffsetptr0+vdwjidx0D,
1137                                             vdwioffsetptr0+vdwjidx0E,
1138                                             vdwioffsetptr0+vdwjidx0F,
1139                                             vdwioffsetptr0+vdwjidx0G,
1140                                             vdwioffsetptr0+vdwjidx0H,
1141                                             &c6_00,&c12_00);
1142
1143             /* REACTION-FIELD ELECTROSTATICS */
1144             felec            = _mm256_mul_ps(qq00,_mm256_sub_ps(_mm256_mul_ps(rinv00,rinvsq00),krf2));
1145
1146             /* LENNARD-JONES DISPERSION/REPULSION */
1147
1148             rinvsix          = _mm256_mul_ps(_mm256_mul_ps(rinvsq00,rinvsq00),rinvsq00);
1149             fvdw             = _mm256_mul_ps(_mm256_sub_ps(_mm256_mul_ps(c12_00,rinvsix),c6_00),_mm256_mul_ps(rinvsix,rinvsq00));
1150
1151             cutoff_mask      = _mm256_cmp_ps(rsq00,rcutoff2,_CMP_LT_OQ);
1152
1153             fscal            = _mm256_add_ps(felec,fvdw);
1154
1155             fscal            = _mm256_and_ps(fscal,cutoff_mask);
1156
1157             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
1158
1159             /* Calculate temporary vectorial force */
1160             tx               = _mm256_mul_ps(fscal,dx00);
1161             ty               = _mm256_mul_ps(fscal,dy00);
1162             tz               = _mm256_mul_ps(fscal,dz00);
1163
1164             /* Update vectorial force */
1165             fix0             = _mm256_add_ps(fix0,tx);
1166             fiy0             = _mm256_add_ps(fiy0,ty);
1167             fiz0             = _mm256_add_ps(fiz0,tz);
1168
1169             fjx0             = _mm256_add_ps(fjx0,tx);
1170             fjy0             = _mm256_add_ps(fjy0,ty);
1171             fjz0             = _mm256_add_ps(fjz0,tz);
1172
1173             }
1174
1175             /**************************
1176              * CALCULATE INTERACTIONS *
1177              **************************/
1178
1179             if (gmx_mm256_any_lt(rsq10,rcutoff2))
1180             {
1181
1182             /* Compute parameters for interactions between i and j atoms */
1183             qq10             = _mm256_mul_ps(iq1,jq0);
1184
1185             /* REACTION-FIELD ELECTROSTATICS */
1186             felec            = _mm256_mul_ps(qq10,_mm256_sub_ps(_mm256_mul_ps(rinv10,rinvsq10),krf2));
1187
1188             cutoff_mask      = _mm256_cmp_ps(rsq10,rcutoff2,_CMP_LT_OQ);
1189
1190             fscal            = felec;
1191
1192             fscal            = _mm256_and_ps(fscal,cutoff_mask);
1193
1194             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
1195
1196             /* Calculate temporary vectorial force */
1197             tx               = _mm256_mul_ps(fscal,dx10);
1198             ty               = _mm256_mul_ps(fscal,dy10);
1199             tz               = _mm256_mul_ps(fscal,dz10);
1200
1201             /* Update vectorial force */
1202             fix1             = _mm256_add_ps(fix1,tx);
1203             fiy1             = _mm256_add_ps(fiy1,ty);
1204             fiz1             = _mm256_add_ps(fiz1,tz);
1205
1206             fjx0             = _mm256_add_ps(fjx0,tx);
1207             fjy0             = _mm256_add_ps(fjy0,ty);
1208             fjz0             = _mm256_add_ps(fjz0,tz);
1209
1210             }
1211
1212             /**************************
1213              * CALCULATE INTERACTIONS *
1214              **************************/
1215
1216             if (gmx_mm256_any_lt(rsq20,rcutoff2))
1217             {
1218
1219             /* Compute parameters for interactions between i and j atoms */
1220             qq20             = _mm256_mul_ps(iq2,jq0);
1221
1222             /* REACTION-FIELD ELECTROSTATICS */
1223             felec            = _mm256_mul_ps(qq20,_mm256_sub_ps(_mm256_mul_ps(rinv20,rinvsq20),krf2));
1224
1225             cutoff_mask      = _mm256_cmp_ps(rsq20,rcutoff2,_CMP_LT_OQ);
1226
1227             fscal            = felec;
1228
1229             fscal            = _mm256_and_ps(fscal,cutoff_mask);
1230
1231             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
1232
1233             /* Calculate temporary vectorial force */
1234             tx               = _mm256_mul_ps(fscal,dx20);
1235             ty               = _mm256_mul_ps(fscal,dy20);
1236             tz               = _mm256_mul_ps(fscal,dz20);
1237
1238             /* Update vectorial force */
1239             fix2             = _mm256_add_ps(fix2,tx);
1240             fiy2             = _mm256_add_ps(fiy2,ty);
1241             fiz2             = _mm256_add_ps(fiz2,tz);
1242
1243             fjx0             = _mm256_add_ps(fjx0,tx);
1244             fjy0             = _mm256_add_ps(fjy0,ty);
1245             fjz0             = _mm256_add_ps(fjz0,tz);
1246
1247             }
1248
1249             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
1250             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
1251             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
1252             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
1253             fjptrE             = (jnrlistE>=0) ? f+j_coord_offsetE : scratch;
1254             fjptrF             = (jnrlistF>=0) ? f+j_coord_offsetF : scratch;
1255             fjptrG             = (jnrlistG>=0) ? f+j_coord_offsetG : scratch;
1256             fjptrH             = (jnrlistH>=0) ? f+j_coord_offsetH : scratch;
1257
1258             gmx_mm256_decrement_1rvec_8ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjptrE,fjptrF,fjptrG,fjptrH,fjx0,fjy0,fjz0);
1259
1260             /* Inner loop uses 100 flops */
1261         }
1262
1263         /* End of innermost loop */
1264
1265         gmx_mm256_update_iforce_3atom_swizzle_ps(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,
1266                                                  f+i_coord_offset,fshift+i_shift_offset);
1267
1268         /* Increment number of inner iterations */
1269         inneriter                  += j_index_end - j_index_start;
1270
1271         /* Outer loop uses 18 flops */
1272     }
1273
1274     /* Increment number of outer iterations */
1275     outeriter        += nri;
1276
1277     /* Update outer/inner flops */
1278
1279     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W3_F,outeriter*18 + inneriter*100);
1280 }