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

   1 /*
   2  * This file is part of the GROMACS molecular simulation package.
   3  *
   4  * Copyright (c) 2012,2013,2014,2015,2017,2018, 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 #include "gmxpre.h"
  39
  40 #include "config.h"
  41
  42 #include <math.h>
  43
  44 #include "../nb_kernel.h"
  45 #include "gromacs/gmxlib/nrnb.h"
  46
  47 #include "kernelutil_x86_avx_256_single.h"
  48
  49 /*
  50  * Gromacs nonbonded kernel:   nb_kernel_ElecRFCut_VdwLJSh_GeomW3P1_VF_avx_256_single
  51  * Electrostatics interaction: ReactionField
  52  * VdW interaction:            LennardJones
  53  * Geometry:                   Water3-Particle
  54  * Calculate force/pot:        PotentialAndForce
  55  */
  56 void
  57 nb_kernel_ElecRFCut_VdwLJSh_GeomW3P1_VF_avx_256_single
  58                     (t_nblist                    * gmx_restrict       nlist,
  59                      rvec                        * gmx_restrict          xx,
  60                      rvec                        * gmx_restrict          ff,
  61                      struct t_forcerec           * gmx_restrict          fr,
  62                      t_mdatoms                   * gmx_restrict     mdatoms,
  63                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
  64                      t_nrnb                      * gmx_restrict        nrnb)
  65 {
  66     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
  67      * just 0 for non-waters.
  68      * Suffixes A,B,C,D,E,F,G,H refer to j loop unrolling done with AVX, e.g. for the eight different
  69      * jnr indices corresponding to data put in the four positions in the SIMD register.
  70      */
  71     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
  72     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
  73     int              jnrA,jnrB,jnrC,jnrD;
  74     int              jnrE,jnrF,jnrG,jnrH;
  75     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
  76     int              jnrlistE,jnrlistF,jnrlistG,jnrlistH;
  77     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
  78     int              j_coord_offsetE,j_coord_offsetF,j_coord_offsetG,j_coord_offsetH;
  79     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
  80     real             rcutoff_scalar;
  81     real             *shiftvec,*fshift,*x,*f;
  82     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD,*fjptrE,*fjptrF,*fjptrG,*fjptrH;
  83     real             scratch[4*DIM];
  84     __m256           tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
  85     real *           vdwioffsetptr0;
  86     __m256           ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
  87     real *           vdwioffsetptr1;
  88     __m256           ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
  89     real *           vdwioffsetptr2;
  90     __m256           ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
  91     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D,vdwjidx0E,vdwjidx0F,vdwjidx0G,vdwjidx0H;
  92     __m256           jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
  93     __m256           dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
  94     __m256           dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
  95     __m256           dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
  96     __m256           velec,felec,velecsum,facel,crf,krf,krf2;
  97     real             *charge;
  98     int              nvdwtype;
  99     __m256           rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
 100     int              *vdwtype;
 101     real             *vdwparam;
 102     __m256           one_sixth   = _mm256_set1_ps(1.0/6.0);
 103     __m256           one_twelfth = _mm256_set1_ps(1.0/12.0);
 104     __m256           dummy_mask,cutoff_mask;
 105     __m256           signbit = _mm256_castsi256_ps( _mm256_set1_epi32(0x80000000) );
 106     __m256           one     = _mm256_set1_ps(1.0);
 107     __m256           two     = _mm256_set1_ps(2.0);
 108     x                = xx[0];
 109     f                = ff[0];
 110
 111     nri              = nlist->nri;
 112     iinr             = nlist->iinr;
 113     jindex           = nlist->jindex;
 114     jjnr             = nlist->jjnr;
 115     shiftidx         = nlist->shift;
 116     gid              = nlist->gid;
 117     shiftvec         = fr->shift_vec[0];
 118     fshift           = fr->fshift[0];
 119     facel            = _mm256_set1_ps(fr->ic->epsfac);
 120     charge           = mdatoms->chargeA;
 121     krf              = _mm256_set1_ps(fr->ic->k_rf);
 122     krf2             = _mm256_set1_ps(fr->ic->k_rf*2.0);
 123     crf              = _mm256_set1_ps(fr->ic->c_rf);
 124     nvdwtype         = fr->ntype;
 125     vdwparam         = fr->nbfp;
 126     vdwtype          = mdatoms->typeA;
 127
 128     /* Setup water-specific parameters */
 129     inr              = nlist->iinr[0];
 130     iq0              = _mm256_mul_ps(facel,_mm256_set1_ps(charge[inr+0]));
 131     iq1              = _mm256_mul_ps(facel,_mm256_set1_ps(charge[inr+1]));
 132     iq2              = _mm256_mul_ps(facel,_mm256_set1_ps(charge[inr+2]));
 133     vdwioffsetptr0   = vdwparam+2*nvdwtype*vdwtype[inr+0];
 134
 135     /* When we use explicit cutoffs the value must be identical for elec and VdW, so use elec as an arbitrary choice */
 136     rcutoff_scalar   = fr->ic->rcoulomb;
 137     rcutoff          = _mm256_set1_ps(rcutoff_scalar);
 138     rcutoff2         = _mm256_mul_ps(rcutoff,rcutoff);
 139
 140     sh_vdw_invrcut6  = _mm256_set1_ps(fr->ic->sh_invrc6);
 141     rvdw             = _mm256_set1_ps(fr->ic->rvdw);
 142
 143     /* Avoid stupid compiler warnings */
 144     jnrA = jnrB = jnrC = jnrD = jnrE = jnrF = jnrG = jnrH = 0;
 145     j_coord_offsetA = 0;
 146     j_coord_offsetB = 0;
 147     j_coord_offsetC = 0;
 148     j_coord_offsetD = 0;
 149     j_coord_offsetE = 0;
 150     j_coord_offsetF = 0;
 151     j_coord_offsetG = 0;
 152     j_coord_offsetH = 0;
 153
 154     outeriter        = 0;
 155     inneriter        = 0;
 156
 157     for(iidx=0;iidx<4*DIM;iidx++)
 158     {
 159         scratch[iidx] = 0.0;
 160     }
 161
 162     /* Start outer loop over neighborlists */
 163     for(iidx=0; iidx<nri; iidx++)
 164     {
 165         /* Load shift vector for this list */
 166         i_shift_offset   = DIM*shiftidx[iidx];
 167
 168         /* Load limits for loop over neighbors */
 169         j_index_start    = jindex[iidx];
 170         j_index_end      = jindex[iidx+1];
 171
 172         /* Get outer coordinate index */
 173         inr              = iinr[iidx];
 174         i_coord_offset   = DIM*inr;
 175
 176         /* Load i particle coords and add shift vector */
 177         gmx_mm256_load_shift_and_3rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,
 178                                                     &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2);
 179
 180         fix0             = _mm256_setzero_ps();
 181         fiy0             = _mm256_setzero_ps();
 182         fiz0             = _mm256_setzero_ps();
 183         fix1             = _mm256_setzero_ps();
 184         fiy1             = _mm256_setzero_ps();
 185         fiz1             = _mm256_setzero_ps();
 186         fix2             = _mm256_setzero_ps();
 187         fiy2             = _mm256_setzero_ps();
 188         fiz2             = _mm256_setzero_ps();
 189
 190         /* Reset potential sums */
 191         velecsum         = _mm256_setzero_ps();
 192         vvdwsum          = _mm256_setzero_ps();
 193
 194         /* Start inner kernel loop */
 195         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+7]>=0; jidx+=8)
 196         {
 197
 198             /* Get j neighbor index, and coordinate index */
 199             jnrA             = jjnr[jidx];
 200             jnrB             = jjnr[jidx+1];
 201             jnrC             = jjnr[jidx+2];
 202             jnrD             = jjnr[jidx+3];
 203             jnrE             = jjnr[jidx+4];
 204             jnrF             = jjnr[jidx+5];
 205             jnrG             = jjnr[jidx+6];
 206             jnrH             = jjnr[jidx+7];
 207             j_coord_offsetA  = DIM*jnrA;
 208             j_coord_offsetB  = DIM*jnrB;
 209             j_coord_offsetC  = DIM*jnrC;
 210             j_coord_offsetD  = DIM*jnrD;
 211             j_coord_offsetE  = DIM*jnrE;
 212             j_coord_offsetF  = DIM*jnrF;
 213             j_coord_offsetG  = DIM*jnrG;
 214             j_coord_offsetH  = DIM*jnrH;
 215
 216             /* load j atom coordinates */
 217             gmx_mm256_load_1rvec_8ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 218                                                  x+j_coord_offsetC,x+j_coord_offsetD,
 219                                                  x+j_coord_offsetE,x+j_coord_offsetF,
 220                                                  x+j_coord_offsetG,x+j_coord_offsetH,
 221                                                  &jx0,&jy0,&jz0);
 222
 223             /* Calculate displacement vector */
 224             dx00             = _mm256_sub_ps(ix0,jx0);
 225             dy00             = _mm256_sub_ps(iy0,jy0);
 226             dz00             = _mm256_sub_ps(iz0,jz0);
 227             dx10             = _mm256_sub_ps(ix1,jx0);
 228             dy10             = _mm256_sub_ps(iy1,jy0);
 229             dz10             = _mm256_sub_ps(iz1,jz0);
 230             dx20             = _mm256_sub_ps(ix2,jx0);
 231             dy20             = _mm256_sub_ps(iy2,jy0);
 232             dz20             = _mm256_sub_ps(iz2,jz0);
 233
 234             /* Calculate squared distance and things based on it */
 235             rsq00            = gmx_mm256_calc_rsq_ps(dx00,dy00,dz00);
 236             rsq10            = gmx_mm256_calc_rsq_ps(dx10,dy10,dz10);
 237             rsq20            = gmx_mm256_calc_rsq_ps(dx20,dy20,dz20);
 238
 239             rinv00           = avx256_invsqrt_f(rsq00);
 240             rinv10           = avx256_invsqrt_f(rsq10);
 241             rinv20           = avx256_invsqrt_f(rsq20);
 242
 243             rinvsq00         = _mm256_mul_ps(rinv00,rinv00);
 244             rinvsq10         = _mm256_mul_ps(rinv10,rinv10);
 245             rinvsq20         = _mm256_mul_ps(rinv20,rinv20);
 246
 247             /* Load parameters for j particles */
 248             jq0              = gmx_mm256_load_8real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 249                                                                  charge+jnrC+0,charge+jnrD+0,
 250                                                                  charge+jnrE+0,charge+jnrF+0,
 251                                                                  charge+jnrG+0,charge+jnrH+0);
 252             vdwjidx0A        = 2*vdwtype[jnrA+0];
 253             vdwjidx0B        = 2*vdwtype[jnrB+0];
 254             vdwjidx0C        = 2*vdwtype[jnrC+0];
 255             vdwjidx0D        = 2*vdwtype[jnrD+0];
 256             vdwjidx0E        = 2*vdwtype[jnrE+0];
 257             vdwjidx0F        = 2*vdwtype[jnrF+0];
 258             vdwjidx0G        = 2*vdwtype[jnrG+0];
 259             vdwjidx0H        = 2*vdwtype[jnrH+0];
 260
 261             fjx0             = _mm256_setzero_ps();
 262             fjy0             = _mm256_setzero_ps();
 263             fjz0             = _mm256_setzero_ps();
 264
 265             /**************************
 266              * CALCULATE INTERACTIONS *
 267              **************************/
 268
 269             if (gmx_mm256_any_lt(rsq00,rcutoff2))
 270             {
 271
 272             /* Compute parameters for interactions between i and j atoms */
 273             qq00             = _mm256_mul_ps(iq0,jq0);
 274             gmx_mm256_load_8pair_swizzle_ps(vdwioffsetptr0+vdwjidx0A,
 275                                             vdwioffsetptr0+vdwjidx0B,
 276                                             vdwioffsetptr0+vdwjidx0C,
 277                                             vdwioffsetptr0+vdwjidx0D,
 278                                             vdwioffsetptr0+vdwjidx0E,
 279                                             vdwioffsetptr0+vdwjidx0F,
 280                                             vdwioffsetptr0+vdwjidx0G,
 281                                             vdwioffsetptr0+vdwjidx0H,
 282                                             &c6_00,&c12_00);
 283
 284             /* REACTION-FIELD ELECTROSTATICS */
 285             velec            = _mm256_mul_ps(qq00,_mm256_sub_ps(_mm256_add_ps(rinv00,_mm256_mul_ps(krf,rsq00)),crf));
 286             felec            = _mm256_mul_ps(qq00,_mm256_sub_ps(_mm256_mul_ps(rinv00,rinvsq00),krf2));
 287
 288             /* LENNARD-JONES DISPERSION/REPULSION */
 289
 290             rinvsix          = _mm256_mul_ps(_mm256_mul_ps(rinvsq00,rinvsq00),rinvsq00);
 291             vvdw6            = _mm256_mul_ps(c6_00,rinvsix);
 292             vvdw12           = _mm256_mul_ps(c12_00,_mm256_mul_ps(rinvsix,rinvsix));
 293             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) ,
 294                                           _mm256_mul_ps( _mm256_sub_ps(vvdw6,_mm256_mul_ps(c6_00,sh_vdw_invrcut6)),one_sixth));
 295             fvdw             = _mm256_mul_ps(_mm256_sub_ps(vvdw12,vvdw6),rinvsq00);
 296
 297             cutoff_mask      = _mm256_cmp_ps(rsq00,rcutoff2,_CMP_LT_OQ);
 298
 299             /* Update potential sum for this i atom from the interaction with this j atom. */
 300             velec            = _mm256_and_ps(velec,cutoff_mask);
 301             velecsum         = _mm256_add_ps(velecsum,velec);
 302             vvdw             = _mm256_and_ps(vvdw,cutoff_mask);
 303             vvdwsum          = _mm256_add_ps(vvdwsum,vvdw);
 304
 305             fscal            = _mm256_add_ps(felec,fvdw);
 306
 307             fscal            = _mm256_and_ps(fscal,cutoff_mask);
 308
 309             /* Calculate temporary vectorial force */
 310             tx               = _mm256_mul_ps(fscal,dx00);
 311             ty               = _mm256_mul_ps(fscal,dy00);
 312             tz               = _mm256_mul_ps(fscal,dz00);
 313
 314             /* Update vectorial force */
 315             fix0             = _mm256_add_ps(fix0,tx);
 316             fiy0             = _mm256_add_ps(fiy0,ty);
 317             fiz0             = _mm256_add_ps(fiz0,tz);
 318
 319             fjx0             = _mm256_add_ps(fjx0,tx);
 320             fjy0             = _mm256_add_ps(fjy0,ty);
 321             fjz0             = _mm256_add_ps(fjz0,tz);
 322
 323             }
 324
 325             /**************************
 326              * CALCULATE INTERACTIONS *
 327              **************************/
 328
 329             if (gmx_mm256_any_lt(rsq10,rcutoff2))
 330             {
 331
 332             /* Compute parameters for interactions between i and j atoms */
 333             qq10             = _mm256_mul_ps(iq1,jq0);
 334
 335             /* REACTION-FIELD ELECTROSTATICS */
 336             velec            = _mm256_mul_ps(qq10,_mm256_sub_ps(_mm256_add_ps(rinv10,_mm256_mul_ps(krf,rsq10)),crf));
 337             felec            = _mm256_mul_ps(qq10,_mm256_sub_ps(_mm256_mul_ps(rinv10,rinvsq10),krf2));
 338
 339             cutoff_mask      = _mm256_cmp_ps(rsq10,rcutoff2,_CMP_LT_OQ);
 340
 341             /* Update potential sum for this i atom from the interaction with this j atom. */
 342             velec            = _mm256_and_ps(velec,cutoff_mask);
 343             velecsum         = _mm256_add_ps(velecsum,velec);
 344
 345             fscal            = felec;
 346
 347             fscal            = _mm256_and_ps(fscal,cutoff_mask);
 348
 349             /* Calculate temporary vectorial force */
 350             tx               = _mm256_mul_ps(fscal,dx10);
 351             ty               = _mm256_mul_ps(fscal,dy10);
 352             tz               = _mm256_mul_ps(fscal,dz10);
 353
 354             /* Update vectorial force */
 355             fix1             = _mm256_add_ps(fix1,tx);
 356             fiy1             = _mm256_add_ps(fiy1,ty);
 357             fiz1             = _mm256_add_ps(fiz1,tz);
 358
 359             fjx0             = _mm256_add_ps(fjx0,tx);
 360             fjy0             = _mm256_add_ps(fjy0,ty);
 361             fjz0             = _mm256_add_ps(fjz0,tz);
 362
 363             }
 364
 365             /**************************
 366              * CALCULATE INTERACTIONS *
 367              **************************/
 368
 369             if (gmx_mm256_any_lt(rsq20,rcutoff2))
 370             {
 371
 372             /* Compute parameters for interactions between i and j atoms */
 373             qq20             = _mm256_mul_ps(iq2,jq0);
 374
 375             /* REACTION-FIELD ELECTROSTATICS */
 376             velec            = _mm256_mul_ps(qq20,_mm256_sub_ps(_mm256_add_ps(rinv20,_mm256_mul_ps(krf,rsq20)),crf));
 377             felec            = _mm256_mul_ps(qq20,_mm256_sub_ps(_mm256_mul_ps(rinv20,rinvsq20),krf2));
 378
 379             cutoff_mask      = _mm256_cmp_ps(rsq20,rcutoff2,_CMP_LT_OQ);
 380
 381             /* Update potential sum for this i atom from the interaction with this j atom. */
 382             velec            = _mm256_and_ps(velec,cutoff_mask);
 383             velecsum         = _mm256_add_ps(velecsum,velec);
 384
 385             fscal            = felec;
 386
 387             fscal            = _mm256_and_ps(fscal,cutoff_mask);
 388
 389             /* Calculate temporary vectorial force */
 390             tx               = _mm256_mul_ps(fscal,dx20);
 391             ty               = _mm256_mul_ps(fscal,dy20);
 392             tz               = _mm256_mul_ps(fscal,dz20);
 393
 394             /* Update vectorial force */
 395             fix2             = _mm256_add_ps(fix2,tx);
 396             fiy2             = _mm256_add_ps(fiy2,ty);
 397             fiz2             = _mm256_add_ps(fiz2,tz);
 398
 399             fjx0             = _mm256_add_ps(fjx0,tx);
 400             fjy0             = _mm256_add_ps(fjy0,ty);
 401             fjz0             = _mm256_add_ps(fjz0,tz);
 402
 403             }
 404
 405             fjptrA             = f+j_coord_offsetA;
 406             fjptrB             = f+j_coord_offsetB;
 407             fjptrC             = f+j_coord_offsetC;
 408             fjptrD             = f+j_coord_offsetD;
 409             fjptrE             = f+j_coord_offsetE;
 410             fjptrF             = f+j_coord_offsetF;
 411             fjptrG             = f+j_coord_offsetG;
 412             fjptrH             = f+j_coord_offsetH;
 413
 414             gmx_mm256_decrement_1rvec_8ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjptrE,fjptrF,fjptrG,fjptrH,fjx0,fjy0,fjz0);
 415
 416             /* Inner loop uses 129 flops */
 417         }
 418
 419         if(jidx<j_index_end)
 420         {
 421
 422             /* Get j neighbor index, and coordinate index */
 423             jnrlistA         = jjnr[jidx];
 424             jnrlistB         = jjnr[jidx+1];
 425             jnrlistC         = jjnr[jidx+2];
 426             jnrlistD         = jjnr[jidx+3];
 427             jnrlistE         = jjnr[jidx+4];
 428             jnrlistF         = jjnr[jidx+5];
 429             jnrlistG         = jjnr[jidx+6];
 430             jnrlistH         = jjnr[jidx+7];
 431             /* Sign of each element will be negative for non-real atoms.
 432              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 433              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
 434              */
 435             dummy_mask = gmx_mm256_set_m128(gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx+4)),_mm_setzero_si128())),
 436                                             gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128())));
 437
 438             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
 439             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
 440             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
 441             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
 442             jnrE       = (jnrlistE>=0) ? jnrlistE : 0;
 443             jnrF       = (jnrlistF>=0) ? jnrlistF : 0;
 444             jnrG       = (jnrlistG>=0) ? jnrlistG : 0;
 445             jnrH       = (jnrlistH>=0) ? jnrlistH : 0;
 446             j_coord_offsetA  = DIM*jnrA;
 447             j_coord_offsetB  = DIM*jnrB;
 448             j_coord_offsetC  = DIM*jnrC;
 449             j_coord_offsetD  = DIM*jnrD;
 450             j_coord_offsetE  = DIM*jnrE;
 451             j_coord_offsetF  = DIM*jnrF;
 452             j_coord_offsetG  = DIM*jnrG;
 453             j_coord_offsetH  = DIM*jnrH;
 454
 455             /* load j atom coordinates */
 456             gmx_mm256_load_1rvec_8ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 457                                                  x+j_coord_offsetC,x+j_coord_offsetD,
 458                                                  x+j_coord_offsetE,x+j_coord_offsetF,
 459                                                  x+j_coord_offsetG,x+j_coord_offsetH,
 460                                                  &jx0,&jy0,&jz0);
 461
 462             /* Calculate displacement vector */
 463             dx00             = _mm256_sub_ps(ix0,jx0);
 464             dy00             = _mm256_sub_ps(iy0,jy0);
 465             dz00             = _mm256_sub_ps(iz0,jz0);
 466             dx10             = _mm256_sub_ps(ix1,jx0);
 467             dy10             = _mm256_sub_ps(iy1,jy0);
 468             dz10             = _mm256_sub_ps(iz1,jz0);
 469             dx20             = _mm256_sub_ps(ix2,jx0);
 470             dy20             = _mm256_sub_ps(iy2,jy0);
 471             dz20             = _mm256_sub_ps(iz2,jz0);
 472
 473             /* Calculate squared distance and things based on it */
 474             rsq00            = gmx_mm256_calc_rsq_ps(dx00,dy00,dz00);
 475             rsq10            = gmx_mm256_calc_rsq_ps(dx10,dy10,dz10);
 476             rsq20            = gmx_mm256_calc_rsq_ps(dx20,dy20,dz20);
 477
 478             rinv00           = avx256_invsqrt_f(rsq00);
 479             rinv10           = avx256_invsqrt_f(rsq10);
 480             rinv20           = avx256_invsqrt_f(rsq20);
 481
 482             rinvsq00         = _mm256_mul_ps(rinv00,rinv00);
 483             rinvsq10         = _mm256_mul_ps(rinv10,rinv10);
 484             rinvsq20         = _mm256_mul_ps(rinv20,rinv20);
 485
 486             /* Load parameters for j particles */
 487             jq0              = gmx_mm256_load_8real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 488                                                                  charge+jnrC+0,charge+jnrD+0,
 489                                                                  charge+jnrE+0,charge+jnrF+0,
 490                                                                  charge+jnrG+0,charge+jnrH+0);
 491             vdwjidx0A        = 2*vdwtype[jnrA+0];
 492             vdwjidx0B        = 2*vdwtype[jnrB+0];
 493             vdwjidx0C        = 2*vdwtype[jnrC+0];
 494             vdwjidx0D        = 2*vdwtype[jnrD+0];
 495             vdwjidx0E        = 2*vdwtype[jnrE+0];
 496             vdwjidx0F        = 2*vdwtype[jnrF+0];
 497             vdwjidx0G        = 2*vdwtype[jnrG+0];
 498             vdwjidx0H        = 2*vdwtype[jnrH+0];
 499
 500             fjx0             = _mm256_setzero_ps();
 501             fjy0             = _mm256_setzero_ps();
 502             fjz0             = _mm256_setzero_ps();
 503
 504             /**************************
 505              * CALCULATE INTERACTIONS *
 506              **************************/
 507
 508             if (gmx_mm256_any_lt(rsq00,rcutoff2))
 509             {
 510
 511             /* Compute parameters for interactions between i and j atoms */
 512             qq00             = _mm256_mul_ps(iq0,jq0);
 513             gmx_mm256_load_8pair_swizzle_ps(vdwioffsetptr0+vdwjidx0A,
 514                                             vdwioffsetptr0+vdwjidx0B,
 515                                             vdwioffsetptr0+vdwjidx0C,
 516                                             vdwioffsetptr0+vdwjidx0D,
 517                                             vdwioffsetptr0+vdwjidx0E,
 518                                             vdwioffsetptr0+vdwjidx0F,
 519                                             vdwioffsetptr0+vdwjidx0G,
 520                                             vdwioffsetptr0+vdwjidx0H,
 521                                             &c6_00,&c12_00);
 522
 523             /* REACTION-FIELD ELECTROSTATICS */
 524             velec            = _mm256_mul_ps(qq00,_mm256_sub_ps(_mm256_add_ps(rinv00,_mm256_mul_ps(krf,rsq00)),crf));
 525             felec            = _mm256_mul_ps(qq00,_mm256_sub_ps(_mm256_mul_ps(rinv00,rinvsq00),krf2));
 526
 527             /* LENNARD-JONES DISPERSION/REPULSION */
 528
 529             rinvsix          = _mm256_mul_ps(_mm256_mul_ps(rinvsq00,rinvsq00),rinvsq00);
 530             vvdw6            = _mm256_mul_ps(c6_00,rinvsix);
 531             vvdw12           = _mm256_mul_ps(c12_00,_mm256_mul_ps(rinvsix,rinvsix));
 532             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) ,
 533                                           _mm256_mul_ps( _mm256_sub_ps(vvdw6,_mm256_mul_ps(c6_00,sh_vdw_invrcut6)),one_sixth));
 534             fvdw             = _mm256_mul_ps(_mm256_sub_ps(vvdw12,vvdw6),rinvsq00);
 535
 536             cutoff_mask      = _mm256_cmp_ps(rsq00,rcutoff2,_CMP_LT_OQ);
 537
 538             /* Update potential sum for this i atom from the interaction with this j atom. */
 539             velec            = _mm256_and_ps(velec,cutoff_mask);
 540             velec            = _mm256_andnot_ps(dummy_mask,velec);
 541             velecsum         = _mm256_add_ps(velecsum,velec);
 542             vvdw             = _mm256_and_ps(vvdw,cutoff_mask);
 543             vvdw             = _mm256_andnot_ps(dummy_mask,vvdw);
 544             vvdwsum          = _mm256_add_ps(vvdwsum,vvdw);
 545
 546             fscal            = _mm256_add_ps(felec,fvdw);
 547
 548             fscal            = _mm256_and_ps(fscal,cutoff_mask);
 549
 550             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
 551
 552             /* Calculate temporary vectorial force */
 553             tx               = _mm256_mul_ps(fscal,dx00);
 554             ty               = _mm256_mul_ps(fscal,dy00);
 555             tz               = _mm256_mul_ps(fscal,dz00);
 556
 557             /* Update vectorial force */
 558             fix0             = _mm256_add_ps(fix0,tx);
 559             fiy0             = _mm256_add_ps(fiy0,ty);
 560             fiz0             = _mm256_add_ps(fiz0,tz);
 561
 562             fjx0             = _mm256_add_ps(fjx0,tx);
 563             fjy0             = _mm256_add_ps(fjy0,ty);
 564             fjz0             = _mm256_add_ps(fjz0,tz);
 565
 566             }
 567
 568             /**************************
 569              * CALCULATE INTERACTIONS *
 570              **************************/
 571
 572             if (gmx_mm256_any_lt(rsq10,rcutoff2))
 573             {
 574
 575             /* Compute parameters for interactions between i and j atoms */
 576             qq10             = _mm256_mul_ps(iq1,jq0);
 577
 578             /* REACTION-FIELD ELECTROSTATICS */
 579             velec            = _mm256_mul_ps(qq10,_mm256_sub_ps(_mm256_add_ps(rinv10,_mm256_mul_ps(krf,rsq10)),crf));
 580             felec            = _mm256_mul_ps(qq10,_mm256_sub_ps(_mm256_mul_ps(rinv10,rinvsq10),krf2));
 581
 582             cutoff_mask      = _mm256_cmp_ps(rsq10,rcutoff2,_CMP_LT_OQ);
 583
 584             /* Update potential sum for this i atom from the interaction with this j atom. */
 585             velec            = _mm256_and_ps(velec,cutoff_mask);
 586             velec            = _mm256_andnot_ps(dummy_mask,velec);
 587             velecsum         = _mm256_add_ps(velecsum,velec);
 588
 589             fscal            = felec;
 590
 591             fscal            = _mm256_and_ps(fscal,cutoff_mask);
 592
 593             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
 594
 595             /* Calculate temporary vectorial force */
 596             tx               = _mm256_mul_ps(fscal,dx10);
 597             ty               = _mm256_mul_ps(fscal,dy10);
 598             tz               = _mm256_mul_ps(fscal,dz10);
 599
 600             /* Update vectorial force */
 601             fix1             = _mm256_add_ps(fix1,tx);
 602             fiy1             = _mm256_add_ps(fiy1,ty);
 603             fiz1             = _mm256_add_ps(fiz1,tz);
 604
 605             fjx0             = _mm256_add_ps(fjx0,tx);
 606             fjy0             = _mm256_add_ps(fjy0,ty);
 607             fjz0             = _mm256_add_ps(fjz0,tz);
 608
 609             }
 610
 611             /**************************
 612              * CALCULATE INTERACTIONS *
 613              **************************/
 614
 615             if (gmx_mm256_any_lt(rsq20,rcutoff2))
 616             {
 617
 618             /* Compute parameters for interactions between i and j atoms */
 619             qq20             = _mm256_mul_ps(iq2,jq0);
 620
 621             /* REACTION-FIELD ELECTROSTATICS */
 622             velec            = _mm256_mul_ps(qq20,_mm256_sub_ps(_mm256_add_ps(rinv20,_mm256_mul_ps(krf,rsq20)),crf));
 623             felec            = _mm256_mul_ps(qq20,_mm256_sub_ps(_mm256_mul_ps(rinv20,rinvsq20),krf2));
 624
 625             cutoff_mask      = _mm256_cmp_ps(rsq20,rcutoff2,_CMP_LT_OQ);
 626
 627             /* Update potential sum for this i atom from the interaction with this j atom. */
 628             velec            = _mm256_and_ps(velec,cutoff_mask);
 629             velec            = _mm256_andnot_ps(dummy_mask,velec);
 630             velecsum         = _mm256_add_ps(velecsum,velec);
 631
 632             fscal            = felec;
 633
 634             fscal            = _mm256_and_ps(fscal,cutoff_mask);
 635
 636             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
 637
 638             /* Calculate temporary vectorial force */
 639             tx               = _mm256_mul_ps(fscal,dx20);
 640             ty               = _mm256_mul_ps(fscal,dy20);
 641             tz               = _mm256_mul_ps(fscal,dz20);
 642
 643             /* Update vectorial force */
 644             fix2             = _mm256_add_ps(fix2,tx);
 645             fiy2             = _mm256_add_ps(fiy2,ty);
 646             fiz2             = _mm256_add_ps(fiz2,tz);
 647
 648             fjx0             = _mm256_add_ps(fjx0,tx);
 649             fjy0             = _mm256_add_ps(fjy0,ty);
 650             fjz0             = _mm256_add_ps(fjz0,tz);
 651
 652             }
 653
 654             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
 655             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
 656             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
 657             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
 658             fjptrE             = (jnrlistE>=0) ? f+j_coord_offsetE : scratch;
 659             fjptrF             = (jnrlistF>=0) ? f+j_coord_offsetF : scratch;
 660             fjptrG             = (jnrlistG>=0) ? f+j_coord_offsetG : scratch;
 661             fjptrH             = (jnrlistH>=0) ? f+j_coord_offsetH : scratch;
 662
 663             gmx_mm256_decrement_1rvec_8ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjptrE,fjptrF,fjptrG,fjptrH,fjx0,fjy0,fjz0);
 664
 665             /* Inner loop uses 129 flops */
 666         }
 667
 668         /* End of innermost loop */
 669
 670         gmx_mm256_update_iforce_3atom_swizzle_ps(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,
 671                                                  f+i_coord_offset,fshift+i_shift_offset);
 672
 673         ggid                        = gid[iidx];
 674         /* Update potential energies */
 675         gmx_mm256_update_1pot_ps(velecsum,kernel_data->energygrp_elec+ggid);
 676         gmx_mm256_update_1pot_ps(vvdwsum,kernel_data->energygrp_vdw+ggid);
 677
 678         /* Increment number of inner iterations */
 679         inneriter                  += j_index_end - j_index_start;
 680
 681         /* Outer loop uses 20 flops */
 682     }
 683
 684     /* Increment number of outer iterations */
 685     outeriter        += nri;
 686
 687     /* Update outer/inner flops */
 688
 689     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W3_VF,outeriter*20 + inneriter*129);
 690 }
 691 /*
 692  * Gromacs nonbonded kernel:   nb_kernel_ElecRFCut_VdwLJSh_GeomW3P1_F_avx_256_single
 693  * Electrostatics interaction: ReactionField
 694  * VdW interaction:            LennardJones
 695  * Geometry:                   Water3-Particle
 696  * Calculate force/pot:        Force
 697  */
 698 void
 699 nb_kernel_ElecRFCut_VdwLJSh_GeomW3P1_F_avx_256_single
 700                     (t_nblist                    * gmx_restrict       nlist,
 701                      rvec                        * gmx_restrict          xx,
 702                      rvec                        * gmx_restrict          ff,
 703                      struct t_forcerec           * gmx_restrict          fr,
 704                      t_mdatoms                   * gmx_restrict     mdatoms,
 705                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
 706                      t_nrnb                      * gmx_restrict        nrnb)
 707 {
 708     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
 709      * just 0 for non-waters.
 710      * Suffixes A,B,C,D,E,F,G,H refer to j loop unrolling done with AVX, e.g. for the eight different
 711      * jnr indices corresponding to data put in the four positions in the SIMD register.
 712      */
 713     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
 714     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
 715     int              jnrA,jnrB,jnrC,jnrD;
 716     int              jnrE,jnrF,jnrG,jnrH;
 717     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
 718     int              jnrlistE,jnrlistF,jnrlistG,jnrlistH;
 719     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
 720     int              j_coord_offsetE,j_coord_offsetF,j_coord_offsetG,j_coord_offsetH;
 721     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
 722     real             rcutoff_scalar;
 723     real             *shiftvec,*fshift,*x,*f;
 724     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD,*fjptrE,*fjptrF,*fjptrG,*fjptrH;
 725     real             scratch[4*DIM];
 726     __m256           tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
 727     real *           vdwioffsetptr0;
 728     __m256           ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
 729     real *           vdwioffsetptr1;
 730     __m256           ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
 731     real *           vdwioffsetptr2;
 732     __m256           ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
 733     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D,vdwjidx0E,vdwjidx0F,vdwjidx0G,vdwjidx0H;
 734     __m256           jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
 735     __m256           dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
 736     __m256           dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
 737     __m256           dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
 738     __m256           velec,felec,velecsum,facel,crf,krf,krf2;
 739     real             *charge;
 740     int              nvdwtype;
 741     __m256           rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
 742     int              *vdwtype;
 743     real             *vdwparam;
 744     __m256           one_sixth   = _mm256_set1_ps(1.0/6.0);
 745     __m256           one_twelfth = _mm256_set1_ps(1.0/12.0);
 746     __m256           dummy_mask,cutoff_mask;
 747     __m256           signbit = _mm256_castsi256_ps( _mm256_set1_epi32(0x80000000) );
 748     __m256           one     = _mm256_set1_ps(1.0);
 749     __m256           two     = _mm256_set1_ps(2.0);
 750     x                = xx[0];
 751     f                = ff[0];
 752
 753     nri              = nlist->nri;
 754     iinr             = nlist->iinr;
 755     jindex           = nlist->jindex;
 756     jjnr             = nlist->jjnr;
 757     shiftidx         = nlist->shift;
 758     gid              = nlist->gid;
 759     shiftvec         = fr->shift_vec[0];
 760     fshift           = fr->fshift[0];
 761     facel            = _mm256_set1_ps(fr->ic->epsfac);
 762     charge           = mdatoms->chargeA;
 763     krf              = _mm256_set1_ps(fr->ic->k_rf);
 764     krf2             = _mm256_set1_ps(fr->ic->k_rf*2.0);
 765     crf              = _mm256_set1_ps(fr->ic->c_rf);
 766     nvdwtype         = fr->ntype;
 767     vdwparam         = fr->nbfp;
 768     vdwtype          = mdatoms->typeA;
 769
 770     /* Setup water-specific parameters */
 771     inr              = nlist->iinr[0];
 772     iq0              = _mm256_mul_ps(facel,_mm256_set1_ps(charge[inr+0]));
 773     iq1              = _mm256_mul_ps(facel,_mm256_set1_ps(charge[inr+1]));
 774     iq2              = _mm256_mul_ps(facel,_mm256_set1_ps(charge[inr+2]));
 775     vdwioffsetptr0   = vdwparam+2*nvdwtype*vdwtype[inr+0];
 776
 777     /* When we use explicit cutoffs the value must be identical for elec and VdW, so use elec as an arbitrary choice */
 778     rcutoff_scalar   = fr->ic->rcoulomb;
 779     rcutoff          = _mm256_set1_ps(rcutoff_scalar);
 780     rcutoff2         = _mm256_mul_ps(rcutoff,rcutoff);
 781
 782     sh_vdw_invrcut6  = _mm256_set1_ps(fr->ic->sh_invrc6);
 783     rvdw             = _mm256_set1_ps(fr->ic->rvdw);
 784
 785     /* Avoid stupid compiler warnings */
 786     jnrA = jnrB = jnrC = jnrD = jnrE = jnrF = jnrG = jnrH = 0;
 787     j_coord_offsetA = 0;
 788     j_coord_offsetB = 0;
 789     j_coord_offsetC = 0;
 790     j_coord_offsetD = 0;
 791     j_coord_offsetE = 0;
 792     j_coord_offsetF = 0;
 793     j_coord_offsetG = 0;
 794     j_coord_offsetH = 0;
 795
 796     outeriter        = 0;
 797     inneriter        = 0;
 798
 799     for(iidx=0;iidx<4*DIM;iidx++)
 800     {
 801         scratch[iidx] = 0.0;
 802     }
 803
 804     /* Start outer loop over neighborlists */
 805     for(iidx=0; iidx<nri; iidx++)
 806     {
 807         /* Load shift vector for this list */
 808         i_shift_offset   = DIM*shiftidx[iidx];
 809
 810         /* Load limits for loop over neighbors */
 811         j_index_start    = jindex[iidx];
 812         j_index_end      = jindex[iidx+1];
 813
 814         /* Get outer coordinate index */
 815         inr              = iinr[iidx];
 816         i_coord_offset   = DIM*inr;
 817
 818         /* Load i particle coords and add shift vector */
 819         gmx_mm256_load_shift_and_3rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,
 820                                                     &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2);
 821
 822         fix0             = _mm256_setzero_ps();
 823         fiy0             = _mm256_setzero_ps();
 824         fiz0             = _mm256_setzero_ps();
 825         fix1             = _mm256_setzero_ps();
 826         fiy1             = _mm256_setzero_ps();
 827         fiz1             = _mm256_setzero_ps();
 828         fix2             = _mm256_setzero_ps();
 829         fiy2             = _mm256_setzero_ps();
 830         fiz2             = _mm256_setzero_ps();
 831
 832         /* Start inner kernel loop */
 833         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+7]>=0; jidx+=8)
 834         {
 835
 836             /* Get j neighbor index, and coordinate index */
 837             jnrA             = jjnr[jidx];
 838             jnrB             = jjnr[jidx+1];
 839             jnrC             = jjnr[jidx+2];
 840             jnrD             = jjnr[jidx+3];
 841             jnrE             = jjnr[jidx+4];
 842             jnrF             = jjnr[jidx+5];
 843             jnrG             = jjnr[jidx+6];
 844             jnrH             = jjnr[jidx+7];
 845             j_coord_offsetA  = DIM*jnrA;
 846             j_coord_offsetB  = DIM*jnrB;
 847             j_coord_offsetC  = DIM*jnrC;
 848             j_coord_offsetD  = DIM*jnrD;
 849             j_coord_offsetE  = DIM*jnrE;
 850             j_coord_offsetF  = DIM*jnrF;
 851             j_coord_offsetG  = DIM*jnrG;
 852             j_coord_offsetH  = DIM*jnrH;
 853
 854             /* load j atom coordinates */
 855             gmx_mm256_load_1rvec_8ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 856                                                  x+j_coord_offsetC,x+j_coord_offsetD,
 857                                                  x+j_coord_offsetE,x+j_coord_offsetF,
 858                                                  x+j_coord_offsetG,x+j_coord_offsetH,
 859                                                  &jx0,&jy0,&jz0);
 860
 861             /* Calculate displacement vector */
 862             dx00             = _mm256_sub_ps(ix0,jx0);
 863             dy00             = _mm256_sub_ps(iy0,jy0);
 864             dz00             = _mm256_sub_ps(iz0,jz0);
 865             dx10             = _mm256_sub_ps(ix1,jx0);
 866             dy10             = _mm256_sub_ps(iy1,jy0);
 867             dz10             = _mm256_sub_ps(iz1,jz0);
 868             dx20             = _mm256_sub_ps(ix2,jx0);
 869             dy20             = _mm256_sub_ps(iy2,jy0);
 870             dz20             = _mm256_sub_ps(iz2,jz0);
 871
 872             /* Calculate squared distance and things based on it */
 873             rsq00            = gmx_mm256_calc_rsq_ps(dx00,dy00,dz00);
 874             rsq10            = gmx_mm256_calc_rsq_ps(dx10,dy10,dz10);
 875             rsq20            = gmx_mm256_calc_rsq_ps(dx20,dy20,dz20);
 876
 877             rinv00           = avx256_invsqrt_f(rsq00);
 878             rinv10           = avx256_invsqrt_f(rsq10);
 879             rinv20           = avx256_invsqrt_f(rsq20);
 880
 881             rinvsq00         = _mm256_mul_ps(rinv00,rinv00);
 882             rinvsq10         = _mm256_mul_ps(rinv10,rinv10);
 883             rinvsq20         = _mm256_mul_ps(rinv20,rinv20);
 884
 885             /* Load parameters for j particles */
 886             jq0              = gmx_mm256_load_8real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 887                                                                  charge+jnrC+0,charge+jnrD+0,
 888                                                                  charge+jnrE+0,charge+jnrF+0,
 889                                                                  charge+jnrG+0,charge+jnrH+0);
 890             vdwjidx0A        = 2*vdwtype[jnrA+0];
 891             vdwjidx0B        = 2*vdwtype[jnrB+0];
 892             vdwjidx0C        = 2*vdwtype[jnrC+0];
 893             vdwjidx0D        = 2*vdwtype[jnrD+0];
 894             vdwjidx0E        = 2*vdwtype[jnrE+0];
 895             vdwjidx0F        = 2*vdwtype[jnrF+0];
 896             vdwjidx0G        = 2*vdwtype[jnrG+0];
 897             vdwjidx0H        = 2*vdwtype[jnrH+0];
 898
 899             fjx0             = _mm256_setzero_ps();
 900             fjy0             = _mm256_setzero_ps();
 901             fjz0             = _mm256_setzero_ps();
 902
 903             /**************************
 904              * CALCULATE INTERACTIONS *
 905              **************************/
 906
 907             if (gmx_mm256_any_lt(rsq00,rcutoff2))
 908             {
 909
 910             /* Compute parameters for interactions between i and j atoms */
 911             qq00             = _mm256_mul_ps(iq0,jq0);
 912             gmx_mm256_load_8pair_swizzle_ps(vdwioffsetptr0+vdwjidx0A,
 913                                             vdwioffsetptr0+vdwjidx0B,
 914                                             vdwioffsetptr0+vdwjidx0C,
 915                                             vdwioffsetptr0+vdwjidx0D,
 916                                             vdwioffsetptr0+vdwjidx0E,
 917                                             vdwioffsetptr0+vdwjidx0F,
 918                                             vdwioffsetptr0+vdwjidx0G,
 919                                             vdwioffsetptr0+vdwjidx0H,
 920                                             &c6_00,&c12_00);
 921
 922             /* REACTION-FIELD ELECTROSTATICS */
 923             felec            = _mm256_mul_ps(qq00,_mm256_sub_ps(_mm256_mul_ps(rinv00,rinvsq00),krf2));
 924
 925             /* LENNARD-JONES DISPERSION/REPULSION */
 926
 927             rinvsix          = _mm256_mul_ps(_mm256_mul_ps(rinvsq00,rinvsq00),rinvsq00);
 928             fvdw             = _mm256_mul_ps(_mm256_sub_ps(_mm256_mul_ps(c12_00,rinvsix),c6_00),_mm256_mul_ps(rinvsix,rinvsq00));
 929
 930             cutoff_mask      = _mm256_cmp_ps(rsq00,rcutoff2,_CMP_LT_OQ);
 931
 932             fscal            = _mm256_add_ps(felec,fvdw);
 933
 934             fscal            = _mm256_and_ps(fscal,cutoff_mask);
 935
 936             /* Calculate temporary vectorial force */
 937             tx               = _mm256_mul_ps(fscal,dx00);
 938             ty               = _mm256_mul_ps(fscal,dy00);
 939             tz               = _mm256_mul_ps(fscal,dz00);
 940
 941             /* Update vectorial force */
 942             fix0             = _mm256_add_ps(fix0,tx);
 943             fiy0             = _mm256_add_ps(fiy0,ty);
 944             fiz0             = _mm256_add_ps(fiz0,tz);
 945
 946             fjx0             = _mm256_add_ps(fjx0,tx);
 947             fjy0             = _mm256_add_ps(fjy0,ty);
 948             fjz0             = _mm256_add_ps(fjz0,tz);
 949
 950             }
 951
 952             /**************************
 953              * CALCULATE INTERACTIONS *
 954              **************************/
 955
 956             if (gmx_mm256_any_lt(rsq10,rcutoff2))
 957             {
 958
 959             /* Compute parameters for interactions between i and j atoms */
 960             qq10             = _mm256_mul_ps(iq1,jq0);
 961
 962             /* REACTION-FIELD ELECTROSTATICS */
 963             felec            = _mm256_mul_ps(qq10,_mm256_sub_ps(_mm256_mul_ps(rinv10,rinvsq10),krf2));
 964
 965             cutoff_mask      = _mm256_cmp_ps(rsq10,rcutoff2,_CMP_LT_OQ);
 966
 967             fscal            = felec;
 968
 969             fscal            = _mm256_and_ps(fscal,cutoff_mask);
 970
 971             /* Calculate temporary vectorial force */
 972             tx               = _mm256_mul_ps(fscal,dx10);
 973             ty               = _mm256_mul_ps(fscal,dy10);
 974             tz               = _mm256_mul_ps(fscal,dz10);
 975
 976             /* Update vectorial force */
 977             fix1             = _mm256_add_ps(fix1,tx);
 978             fiy1             = _mm256_add_ps(fiy1,ty);
 979             fiz1             = _mm256_add_ps(fiz1,tz);
 980
 981             fjx0             = _mm256_add_ps(fjx0,tx);
 982             fjy0             = _mm256_add_ps(fjy0,ty);
 983             fjz0             = _mm256_add_ps(fjz0,tz);
 984
 985             }
 986
 987             /**************************
 988              * CALCULATE INTERACTIONS *
 989              **************************/
 990
 991             if (gmx_mm256_any_lt(rsq20,rcutoff2))
 992             {
 993
 994             /* Compute parameters for interactions between i and j atoms */
 995             qq20             = _mm256_mul_ps(iq2,jq0);
 996
 997             /* REACTION-FIELD ELECTROSTATICS */
 998             felec            = _mm256_mul_ps(qq20,_mm256_sub_ps(_mm256_mul_ps(rinv20,rinvsq20),krf2));
 999
1000             cutoff_mask      = _mm256_cmp_ps(rsq20,rcutoff2,_CMP_LT_OQ);
1001
1002             fscal            = felec;
1003
1004             fscal            = _mm256_and_ps(fscal,cutoff_mask);
1005
1006             /* Calculate temporary vectorial force */
1007             tx               = _mm256_mul_ps(fscal,dx20);
1008             ty               = _mm256_mul_ps(fscal,dy20);
1009             tz               = _mm256_mul_ps(fscal,dz20);
1010
1011             /* Update vectorial force */
1012             fix2             = _mm256_add_ps(fix2,tx);
1013             fiy2             = _mm256_add_ps(fiy2,ty);
1014             fiz2             = _mm256_add_ps(fiz2,tz);
1015
1016             fjx0             = _mm256_add_ps(fjx0,tx);
1017             fjy0             = _mm256_add_ps(fjy0,ty);
1018             fjz0             = _mm256_add_ps(fjz0,tz);
1019
1020             }
1021
1022             fjptrA             = f+j_coord_offsetA;
1023             fjptrB             = f+j_coord_offsetB;
1024             fjptrC             = f+j_coord_offsetC;
1025             fjptrD             = f+j_coord_offsetD;
1026             fjptrE             = f+j_coord_offsetE;
1027             fjptrF             = f+j_coord_offsetF;
1028             fjptrG             = f+j_coord_offsetG;
1029             fjptrH             = f+j_coord_offsetH;
1030
1031             gmx_mm256_decrement_1rvec_8ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjptrE,fjptrF,fjptrG,fjptrH,fjx0,fjy0,fjz0);
1032
1033             /* Inner loop uses 100 flops */
1034         }
1035
1036         if(jidx<j_index_end)
1037         {
1038
1039             /* Get j neighbor index, and coordinate index */
1040             jnrlistA         = jjnr[jidx];
1041             jnrlistB         = jjnr[jidx+1];
1042             jnrlistC         = jjnr[jidx+2];
1043             jnrlistD         = jjnr[jidx+3];
1044             jnrlistE         = jjnr[jidx+4];
1045             jnrlistF         = jjnr[jidx+5];
1046             jnrlistG         = jjnr[jidx+6];
1047             jnrlistH         = jjnr[jidx+7];
1048             /* Sign of each element will be negative for non-real atoms.
1049              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
1050              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
1051              */
1052             dummy_mask = gmx_mm256_set_m128(gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx+4)),_mm_setzero_si128())),
1053                                             gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128())));
1054
1055             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
1056             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
1057             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
1058             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
1059             jnrE       = (jnrlistE>=0) ? jnrlistE : 0;
1060             jnrF       = (jnrlistF>=0) ? jnrlistF : 0;
1061             jnrG       = (jnrlistG>=0) ? jnrlistG : 0;
1062             jnrH       = (jnrlistH>=0) ? jnrlistH : 0;
1063             j_coord_offsetA  = DIM*jnrA;
1064             j_coord_offsetB  = DIM*jnrB;
1065             j_coord_offsetC  = DIM*jnrC;
1066             j_coord_offsetD  = DIM*jnrD;
1067             j_coord_offsetE  = DIM*jnrE;
1068             j_coord_offsetF  = DIM*jnrF;
1069             j_coord_offsetG  = DIM*jnrG;
1070             j_coord_offsetH  = DIM*jnrH;
1071
1072             /* load j atom coordinates */
1073             gmx_mm256_load_1rvec_8ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
1074                                                  x+j_coord_offsetC,x+j_coord_offsetD,
1075                                                  x+j_coord_offsetE,x+j_coord_offsetF,
1076                                                  x+j_coord_offsetG,x+j_coord_offsetH,
1077                                                  &jx0,&jy0,&jz0);
1078
1079             /* Calculate displacement vector */
1080             dx00             = _mm256_sub_ps(ix0,jx0);
1081             dy00             = _mm256_sub_ps(iy0,jy0);
1082             dz00             = _mm256_sub_ps(iz0,jz0);
1083             dx10             = _mm256_sub_ps(ix1,jx0);
1084             dy10             = _mm256_sub_ps(iy1,jy0);
1085             dz10             = _mm256_sub_ps(iz1,jz0);
1086             dx20             = _mm256_sub_ps(ix2,jx0);
1087             dy20             = _mm256_sub_ps(iy2,jy0);
1088             dz20             = _mm256_sub_ps(iz2,jz0);
1089
1090             /* Calculate squared distance and things based on it */
1091             rsq00            = gmx_mm256_calc_rsq_ps(dx00,dy00,dz00);
1092             rsq10            = gmx_mm256_calc_rsq_ps(dx10,dy10,dz10);
1093             rsq20            = gmx_mm256_calc_rsq_ps(dx20,dy20,dz20);
1094
1095             rinv00           = avx256_invsqrt_f(rsq00);
1096             rinv10           = avx256_invsqrt_f(rsq10);
1097             rinv20           = avx256_invsqrt_f(rsq20);
1098
1099             rinvsq00         = _mm256_mul_ps(rinv00,rinv00);
1100             rinvsq10         = _mm256_mul_ps(rinv10,rinv10);
1101             rinvsq20         = _mm256_mul_ps(rinv20,rinv20);
1102
1103             /* Load parameters for j particles */
1104             jq0              = gmx_mm256_load_8real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
1105                                                                  charge+jnrC+0,charge+jnrD+0,
1106                                                                  charge+jnrE+0,charge+jnrF+0,
1107                                                                  charge+jnrG+0,charge+jnrH+0);
1108             vdwjidx0A        = 2*vdwtype[jnrA+0];
1109             vdwjidx0B        = 2*vdwtype[jnrB+0];
1110             vdwjidx0C        = 2*vdwtype[jnrC+0];
1111             vdwjidx0D        = 2*vdwtype[jnrD+0];
1112             vdwjidx0E        = 2*vdwtype[jnrE+0];
1113             vdwjidx0F        = 2*vdwtype[jnrF+0];
1114             vdwjidx0G        = 2*vdwtype[jnrG+0];
1115             vdwjidx0H        = 2*vdwtype[jnrH+0];
1116
1117             fjx0             = _mm256_setzero_ps();
1118             fjy0             = _mm256_setzero_ps();
1119             fjz0             = _mm256_setzero_ps();
1120
1121             /**************************
1122              * CALCULATE INTERACTIONS *
1123              **************************/
1124
1125             if (gmx_mm256_any_lt(rsq00,rcutoff2))
1126             {
1127
1128             /* Compute parameters for interactions between i and j atoms */
1129             qq00             = _mm256_mul_ps(iq0,jq0);
1130             gmx_mm256_load_8pair_swizzle_ps(vdwioffsetptr0+vdwjidx0A,
1131                                             vdwioffsetptr0+vdwjidx0B,
1132                                             vdwioffsetptr0+vdwjidx0C,
1133                                             vdwioffsetptr0+vdwjidx0D,
1134                                             vdwioffsetptr0+vdwjidx0E,
1135                                             vdwioffsetptr0+vdwjidx0F,
1136                                             vdwioffsetptr0+vdwjidx0G,
1137                                             vdwioffsetptr0+vdwjidx0H,
1138                                             &c6_00,&c12_00);
1139
1140             /* REACTION-FIELD ELECTROSTATICS */
1141             felec            = _mm256_mul_ps(qq00,_mm256_sub_ps(_mm256_mul_ps(rinv00,rinvsq00),krf2));
1142
1143             /* LENNARD-JONES DISPERSION/REPULSION */
1144
1145             rinvsix          = _mm256_mul_ps(_mm256_mul_ps(rinvsq00,rinvsq00),rinvsq00);
1146             fvdw             = _mm256_mul_ps(_mm256_sub_ps(_mm256_mul_ps(c12_00,rinvsix),c6_00),_mm256_mul_ps(rinvsix,rinvsq00));
1147
1148             cutoff_mask      = _mm256_cmp_ps(rsq00,rcutoff2,_CMP_LT_OQ);
1149
1150             fscal            = _mm256_add_ps(felec,fvdw);
1151
1152             fscal            = _mm256_and_ps(fscal,cutoff_mask);
1153
1154             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
1155
1156             /* Calculate temporary vectorial force */
1157             tx               = _mm256_mul_ps(fscal,dx00);
1158             ty               = _mm256_mul_ps(fscal,dy00);
1159             tz               = _mm256_mul_ps(fscal,dz00);
1160
1161             /* Update vectorial force */
1162             fix0             = _mm256_add_ps(fix0,tx);
1163             fiy0             = _mm256_add_ps(fiy0,ty);
1164             fiz0             = _mm256_add_ps(fiz0,tz);
1165
1166             fjx0             = _mm256_add_ps(fjx0,tx);
1167             fjy0             = _mm256_add_ps(fjy0,ty);
1168             fjz0             = _mm256_add_ps(fjz0,tz);
1169
1170             }
1171
1172             /**************************
1173              * CALCULATE INTERACTIONS *
1174              **************************/
1175
1176             if (gmx_mm256_any_lt(rsq10,rcutoff2))
1177             {
1178
1179             /* Compute parameters for interactions between i and j atoms */
1180             qq10             = _mm256_mul_ps(iq1,jq0);
1181
1182             /* REACTION-FIELD ELECTROSTATICS */
1183             felec            = _mm256_mul_ps(qq10,_mm256_sub_ps(_mm256_mul_ps(rinv10,rinvsq10),krf2));
1184
1185             cutoff_mask      = _mm256_cmp_ps(rsq10,rcutoff2,_CMP_LT_OQ);
1186
1187             fscal            = felec;
1188
1189             fscal            = _mm256_and_ps(fscal,cutoff_mask);
1190
1191             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
1192
1193             /* Calculate temporary vectorial force */
1194             tx               = _mm256_mul_ps(fscal,dx10);
1195             ty               = _mm256_mul_ps(fscal,dy10);
1196             tz               = _mm256_mul_ps(fscal,dz10);
1197
1198             /* Update vectorial force */
1199             fix1             = _mm256_add_ps(fix1,tx);
1200             fiy1             = _mm256_add_ps(fiy1,ty);
1201             fiz1             = _mm256_add_ps(fiz1,tz);
1202
1203             fjx0             = _mm256_add_ps(fjx0,tx);
1204             fjy0             = _mm256_add_ps(fjy0,ty);
1205             fjz0             = _mm256_add_ps(fjz0,tz);
1206
1207             }
1208
1209             /**************************
1210              * CALCULATE INTERACTIONS *
1211              **************************/
1212
1213             if (gmx_mm256_any_lt(rsq20,rcutoff2))
1214             {
1215
1216             /* Compute parameters for interactions between i and j atoms */
1217             qq20             = _mm256_mul_ps(iq2,jq0);
1218
1219             /* REACTION-FIELD ELECTROSTATICS */
1220             felec            = _mm256_mul_ps(qq20,_mm256_sub_ps(_mm256_mul_ps(rinv20,rinvsq20),krf2));
1221
1222             cutoff_mask      = _mm256_cmp_ps(rsq20,rcutoff2,_CMP_LT_OQ);
1223
1224             fscal            = felec;
1225
1226             fscal            = _mm256_and_ps(fscal,cutoff_mask);
1227
1228             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
1229
1230             /* Calculate temporary vectorial force */
1231             tx               = _mm256_mul_ps(fscal,dx20);
1232             ty               = _mm256_mul_ps(fscal,dy20);
1233             tz               = _mm256_mul_ps(fscal,dz20);
1234
1235             /* Update vectorial force */
1236             fix2             = _mm256_add_ps(fix2,tx);
1237             fiy2             = _mm256_add_ps(fiy2,ty);
1238             fiz2             = _mm256_add_ps(fiz2,tz);
1239
1240             fjx0             = _mm256_add_ps(fjx0,tx);
1241             fjy0             = _mm256_add_ps(fjy0,ty);
1242             fjz0             = _mm256_add_ps(fjz0,tz);
1243
1244             }
1245
1246             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
1247             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
1248             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
1249             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
1250             fjptrE             = (jnrlistE>=0) ? f+j_coord_offsetE : scratch;
1251             fjptrF             = (jnrlistF>=0) ? f+j_coord_offsetF : scratch;
1252             fjptrG             = (jnrlistG>=0) ? f+j_coord_offsetG : scratch;
1253             fjptrH             = (jnrlistH>=0) ? f+j_coord_offsetH : scratch;
1254
1255             gmx_mm256_decrement_1rvec_8ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjptrE,fjptrF,fjptrG,fjptrH,fjx0,fjy0,fjz0);
1256
1257             /* Inner loop uses 100 flops */
1258         }
1259
1260         /* End of innermost loop */
1261
1262         gmx_mm256_update_iforce_3atom_swizzle_ps(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,
1263                                                  f+i_coord_offset,fshift+i_shift_offset);
1264
1265         /* Increment number of inner iterations */
1266         inneriter                  += j_index_end - j_index_start;
1267
1268         /* Outer loop uses 18 flops */
1269     }
1270
1271     /* Increment number of outer iterations */
1272     outeriter        += nri;
1273
1274     /* Update outer/inner flops */
1275
1276     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W3_F,outeriter*18 + inneriter*100);
1277 }