src/gromacs/gmxlib/nonbonded/nb_kernel_avx_256_single/nb_kernel_ElecEw_VdwLJ_GeomW4P1_avx_256_single.c

   1 /*
   2  * This file is part of the GROMACS molecular simulation package.
   3  *
   4  * Copyright (c) 2012,2013,2014, 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/legacyheaders/types/simple.h"
  46 #include "gromacs/math/vec.h"
  47 #include "gromacs/legacyheaders/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_ElecEw_VdwLJ_GeomW4P1_VF_avx_256_single
  54  * Electrostatics interaction: Ewald
  55  * VdW interaction:            LennardJones
  56  * Geometry:                   Water4-Particle
  57  * Calculate force/pot:        PotentialAndForce
  58  */
  59 void
  60 nb_kernel_ElecEw_VdwLJ_GeomW4P1_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     real *           vdwioffsetptr3;
  95     __m256           ix3,iy3,iz3,fix3,fiy3,fiz3,iq3,isai3;
  96     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D,vdwjidx0E,vdwjidx0F,vdwjidx0G,vdwjidx0H;
  97     __m256           jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
  98     __m256           dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
  99     __m256           dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
 100     __m256           dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
 101     __m256           dx30,dy30,dz30,rsq30,rinv30,rinvsq30,r30,qq30,c6_30,c12_30;
 102     __m256           velec,felec,velecsum,facel,crf,krf,krf2;
 103     real             *charge;
 104     int              nvdwtype;
 105     __m256           rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
 106     int              *vdwtype;
 107     real             *vdwparam;
 108     __m256           one_sixth   = _mm256_set1_ps(1.0/6.0);
 109     __m256           one_twelfth = _mm256_set1_ps(1.0/12.0);
 110     __m256i          ewitab;
 111     __m128i          ewitab_lo,ewitab_hi;
 112     __m256           ewtabscale,eweps,sh_ewald,ewrt,ewtabhalfspace,ewtabF,ewtabFn,ewtabD,ewtabV;
 113     __m256           beta,beta2,beta3,zeta2,pmecorrF,pmecorrV,rinv3;
 114     real             *ewtab;
 115     __m256           dummy_mask,cutoff_mask;
 116     __m256           signbit = _mm256_castsi256_ps( _mm256_set1_epi32(0x80000000) );
 117     __m256           one     = _mm256_set1_ps(1.0);
 118     __m256           two     = _mm256_set1_ps(2.0);
 119     x                = xx[0];
 120     f                = ff[0];
 121
 122     nri              = nlist->nri;
 123     iinr             = nlist->iinr;
 124     jindex           = nlist->jindex;
 125     jjnr             = nlist->jjnr;
 126     shiftidx         = nlist->shift;
 127     gid              = nlist->gid;
 128     shiftvec         = fr->shift_vec[0];
 129     fshift           = fr->fshift[0];
 130     facel            = _mm256_set1_ps(fr->epsfac);
 131     charge           = mdatoms->chargeA;
 132     nvdwtype         = fr->ntype;
 133     vdwparam         = fr->nbfp;
 134     vdwtype          = mdatoms->typeA;
 135
 136     sh_ewald         = _mm256_set1_ps(fr->ic->sh_ewald);
 137     beta             = _mm256_set1_ps(fr->ic->ewaldcoeff_q);
 138     beta2            = _mm256_mul_ps(beta,beta);
 139     beta3            = _mm256_mul_ps(beta,beta2);
 140
 141     ewtab            = fr->ic->tabq_coul_FDV0;
 142     ewtabscale       = _mm256_set1_ps(fr->ic->tabq_scale);
 143     ewtabhalfspace   = _mm256_set1_ps(0.5/fr->ic->tabq_scale);
 144
 145     /* Setup water-specific parameters */
 146     inr              = nlist->iinr[0];
 147     iq1              = _mm256_mul_ps(facel,_mm256_set1_ps(charge[inr+1]));
 148     iq2              = _mm256_mul_ps(facel,_mm256_set1_ps(charge[inr+2]));
 149     iq3              = _mm256_mul_ps(facel,_mm256_set1_ps(charge[inr+3]));
 150     vdwioffsetptr0   = vdwparam+2*nvdwtype*vdwtype[inr+0];
 151
 152     /* Avoid stupid compiler warnings */
 153     jnrA = jnrB = jnrC = jnrD = jnrE = jnrF = jnrG = jnrH = 0;
 154     j_coord_offsetA = 0;
 155     j_coord_offsetB = 0;
 156     j_coord_offsetC = 0;
 157     j_coord_offsetD = 0;
 158     j_coord_offsetE = 0;
 159     j_coord_offsetF = 0;
 160     j_coord_offsetG = 0;
 161     j_coord_offsetH = 0;
 162
 163     outeriter        = 0;
 164     inneriter        = 0;
 165
 166     for(iidx=0;iidx<4*DIM;iidx++)
 167     {
 168         scratch[iidx] = 0.0;
 169     }
 170
 171     /* Start outer loop over neighborlists */
 172     for(iidx=0; iidx<nri; iidx++)
 173     {
 174         /* Load shift vector for this list */
 175         i_shift_offset   = DIM*shiftidx[iidx];
 176
 177         /* Load limits for loop over neighbors */
 178         j_index_start    = jindex[iidx];
 179         j_index_end      = jindex[iidx+1];
 180
 181         /* Get outer coordinate index */
 182         inr              = iinr[iidx];
 183         i_coord_offset   = DIM*inr;
 184
 185         /* Load i particle coords and add shift vector */
 186         gmx_mm256_load_shift_and_4rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,
 187                                                     &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2,&ix3,&iy3,&iz3);
 188
 189         fix0             = _mm256_setzero_ps();
 190         fiy0             = _mm256_setzero_ps();
 191         fiz0             = _mm256_setzero_ps();
 192         fix1             = _mm256_setzero_ps();
 193         fiy1             = _mm256_setzero_ps();
 194         fiz1             = _mm256_setzero_ps();
 195         fix2             = _mm256_setzero_ps();
 196         fiy2             = _mm256_setzero_ps();
 197         fiz2             = _mm256_setzero_ps();
 198         fix3             = _mm256_setzero_ps();
 199         fiy3             = _mm256_setzero_ps();
 200         fiz3             = _mm256_setzero_ps();
 201
 202         /* Reset potential sums */
 203         velecsum         = _mm256_setzero_ps();
 204         vvdwsum          = _mm256_setzero_ps();
 205
 206         /* Start inner kernel loop */
 207         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+7]>=0; jidx+=8)
 208         {
 209
 210             /* Get j neighbor index, and coordinate index */
 211             jnrA             = jjnr[jidx];
 212             jnrB             = jjnr[jidx+1];
 213             jnrC             = jjnr[jidx+2];
 214             jnrD             = jjnr[jidx+3];
 215             jnrE             = jjnr[jidx+4];
 216             jnrF             = jjnr[jidx+5];
 217             jnrG             = jjnr[jidx+6];
 218             jnrH             = jjnr[jidx+7];
 219             j_coord_offsetA  = DIM*jnrA;
 220             j_coord_offsetB  = DIM*jnrB;
 221             j_coord_offsetC  = DIM*jnrC;
 222             j_coord_offsetD  = DIM*jnrD;
 223             j_coord_offsetE  = DIM*jnrE;
 224             j_coord_offsetF  = DIM*jnrF;
 225             j_coord_offsetG  = DIM*jnrG;
 226             j_coord_offsetH  = DIM*jnrH;
 227
 228             /* load j atom coordinates */
 229             gmx_mm256_load_1rvec_8ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 230                                                  x+j_coord_offsetC,x+j_coord_offsetD,
 231                                                  x+j_coord_offsetE,x+j_coord_offsetF,
 232                                                  x+j_coord_offsetG,x+j_coord_offsetH,
 233                                                  &jx0,&jy0,&jz0);
 234
 235             /* Calculate displacement vector */
 236             dx00             = _mm256_sub_ps(ix0,jx0);
 237             dy00             = _mm256_sub_ps(iy0,jy0);
 238             dz00             = _mm256_sub_ps(iz0,jz0);
 239             dx10             = _mm256_sub_ps(ix1,jx0);
 240             dy10             = _mm256_sub_ps(iy1,jy0);
 241             dz10             = _mm256_sub_ps(iz1,jz0);
 242             dx20             = _mm256_sub_ps(ix2,jx0);
 243             dy20             = _mm256_sub_ps(iy2,jy0);
 244             dz20             = _mm256_sub_ps(iz2,jz0);
 245             dx30             = _mm256_sub_ps(ix3,jx0);
 246             dy30             = _mm256_sub_ps(iy3,jy0);
 247             dz30             = _mm256_sub_ps(iz3,jz0);
 248
 249             /* Calculate squared distance and things based on it */
 250             rsq00            = gmx_mm256_calc_rsq_ps(dx00,dy00,dz00);
 251             rsq10            = gmx_mm256_calc_rsq_ps(dx10,dy10,dz10);
 252             rsq20            = gmx_mm256_calc_rsq_ps(dx20,dy20,dz20);
 253             rsq30            = gmx_mm256_calc_rsq_ps(dx30,dy30,dz30);
 254
 255             rinv10           = gmx_mm256_invsqrt_ps(rsq10);
 256             rinv20           = gmx_mm256_invsqrt_ps(rsq20);
 257             rinv30           = gmx_mm256_invsqrt_ps(rsq30);
 258
 259             rinvsq00         = gmx_mm256_inv_ps(rsq00);
 260             rinvsq10         = _mm256_mul_ps(rinv10,rinv10);
 261             rinvsq20         = _mm256_mul_ps(rinv20,rinv20);
 262             rinvsq30         = _mm256_mul_ps(rinv30,rinv30);
 263
 264             /* Load parameters for j particles */
 265             jq0              = gmx_mm256_load_8real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 266                                                                  charge+jnrC+0,charge+jnrD+0,
 267                                                                  charge+jnrE+0,charge+jnrF+0,
 268                                                                  charge+jnrG+0,charge+jnrH+0);
 269             vdwjidx0A        = 2*vdwtype[jnrA+0];
 270             vdwjidx0B        = 2*vdwtype[jnrB+0];
 271             vdwjidx0C        = 2*vdwtype[jnrC+0];
 272             vdwjidx0D        = 2*vdwtype[jnrD+0];
 273             vdwjidx0E        = 2*vdwtype[jnrE+0];
 274             vdwjidx0F        = 2*vdwtype[jnrF+0];
 275             vdwjidx0G        = 2*vdwtype[jnrG+0];
 276             vdwjidx0H        = 2*vdwtype[jnrH+0];
 277
 278             fjx0             = _mm256_setzero_ps();
 279             fjy0             = _mm256_setzero_ps();
 280             fjz0             = _mm256_setzero_ps();
 281
 282             /**************************
 283              * CALCULATE INTERACTIONS *
 284              **************************/
 285
 286             /* Compute parameters for interactions between i and j atoms */
 287             gmx_mm256_load_8pair_swizzle_ps(vdwioffsetptr0+vdwjidx0A,
 288                                             vdwioffsetptr0+vdwjidx0B,
 289                                             vdwioffsetptr0+vdwjidx0C,
 290                                             vdwioffsetptr0+vdwjidx0D,
 291                                             vdwioffsetptr0+vdwjidx0E,
 292                                             vdwioffsetptr0+vdwjidx0F,
 293                                             vdwioffsetptr0+vdwjidx0G,
 294                                             vdwioffsetptr0+vdwjidx0H,
 295                                             &c6_00,&c12_00);
 296
 297             /* LENNARD-JONES DISPERSION/REPULSION */
 298
 299             rinvsix          = _mm256_mul_ps(_mm256_mul_ps(rinvsq00,rinvsq00),rinvsq00);
 300             vvdw6            = _mm256_mul_ps(c6_00,rinvsix);
 301             vvdw12           = _mm256_mul_ps(c12_00,_mm256_mul_ps(rinvsix,rinvsix));
 302             vvdw             = _mm256_sub_ps( _mm256_mul_ps(vvdw12,one_twelfth) , _mm256_mul_ps(vvdw6,one_sixth) );
 303             fvdw             = _mm256_mul_ps(_mm256_sub_ps(vvdw12,vvdw6),rinvsq00);
 304
 305             /* Update potential sum for this i atom from the interaction with this j atom. */
 306             vvdwsum          = _mm256_add_ps(vvdwsum,vvdw);
 307
 308             fscal            = fvdw;
 309
 310             /* Calculate temporary vectorial force */
 311             tx               = _mm256_mul_ps(fscal,dx00);
 312             ty               = _mm256_mul_ps(fscal,dy00);
 313             tz               = _mm256_mul_ps(fscal,dz00);
 314
 315             /* Update vectorial force */
 316             fix0             = _mm256_add_ps(fix0,tx);
 317             fiy0             = _mm256_add_ps(fiy0,ty);
 318             fiz0             = _mm256_add_ps(fiz0,tz);
 319
 320             fjx0             = _mm256_add_ps(fjx0,tx);
 321             fjy0             = _mm256_add_ps(fjy0,ty);
 322             fjz0             = _mm256_add_ps(fjz0,tz);
 323
 324             /**************************
 325              * CALCULATE INTERACTIONS *
 326              **************************/
 327
 328             r10              = _mm256_mul_ps(rsq10,rinv10);
 329
 330             /* Compute parameters for interactions between i and j atoms */
 331             qq10             = _mm256_mul_ps(iq1,jq0);
 332
 333             /* EWALD ELECTROSTATICS */
 334
 335             /* Analytical PME correction */
 336             zeta2            = _mm256_mul_ps(beta2,rsq10);
 337             rinv3            = _mm256_mul_ps(rinvsq10,rinv10);
 338             pmecorrF         = gmx_mm256_pmecorrF_ps(zeta2);
 339             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
 340             felec            = _mm256_mul_ps(qq10,felec);
 341             pmecorrV         = gmx_mm256_pmecorrV_ps(zeta2);
 342             pmecorrV         = _mm256_mul_ps(pmecorrV,beta);
 343             velec            = _mm256_sub_ps(rinv10,pmecorrV);
 344             velec            = _mm256_mul_ps(qq10,velec);
 345
 346             /* Update potential sum for this i atom from the interaction with this j atom. */
 347             velecsum         = _mm256_add_ps(velecsum,velec);
 348
 349             fscal            = felec;
 350
 351             /* Calculate temporary vectorial force */
 352             tx               = _mm256_mul_ps(fscal,dx10);
 353             ty               = _mm256_mul_ps(fscal,dy10);
 354             tz               = _mm256_mul_ps(fscal,dz10);
 355
 356             /* Update vectorial force */
 357             fix1             = _mm256_add_ps(fix1,tx);
 358             fiy1             = _mm256_add_ps(fiy1,ty);
 359             fiz1             = _mm256_add_ps(fiz1,tz);
 360
 361             fjx0             = _mm256_add_ps(fjx0,tx);
 362             fjy0             = _mm256_add_ps(fjy0,ty);
 363             fjz0             = _mm256_add_ps(fjz0,tz);
 364
 365             /**************************
 366              * CALCULATE INTERACTIONS *
 367              **************************/
 368
 369             r20              = _mm256_mul_ps(rsq20,rinv20);
 370
 371             /* Compute parameters for interactions between i and j atoms */
 372             qq20             = _mm256_mul_ps(iq2,jq0);
 373
 374             /* EWALD ELECTROSTATICS */
 375
 376             /* Analytical PME correction */
 377             zeta2            = _mm256_mul_ps(beta2,rsq20);
 378             rinv3            = _mm256_mul_ps(rinvsq20,rinv20);
 379             pmecorrF         = gmx_mm256_pmecorrF_ps(zeta2);
 380             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
 381             felec            = _mm256_mul_ps(qq20,felec);
 382             pmecorrV         = gmx_mm256_pmecorrV_ps(zeta2);
 383             pmecorrV         = _mm256_mul_ps(pmecorrV,beta);
 384             velec            = _mm256_sub_ps(rinv20,pmecorrV);
 385             velec            = _mm256_mul_ps(qq20,velec);
 386
 387             /* Update potential sum for this i atom from the interaction with this j atom. */
 388             velecsum         = _mm256_add_ps(velecsum,velec);
 389
 390             fscal            = felec;
 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              * CALCULATE INTERACTIONS *
 408              **************************/
 409
 410             r30              = _mm256_mul_ps(rsq30,rinv30);
 411
 412             /* Compute parameters for interactions between i and j atoms */
 413             qq30             = _mm256_mul_ps(iq3,jq0);
 414
 415             /* EWALD ELECTROSTATICS */
 416
 417             /* Analytical PME correction */
 418             zeta2            = _mm256_mul_ps(beta2,rsq30);
 419             rinv3            = _mm256_mul_ps(rinvsq30,rinv30);
 420             pmecorrF         = gmx_mm256_pmecorrF_ps(zeta2);
 421             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
 422             felec            = _mm256_mul_ps(qq30,felec);
 423             pmecorrV         = gmx_mm256_pmecorrV_ps(zeta2);
 424             pmecorrV         = _mm256_mul_ps(pmecorrV,beta);
 425             velec            = _mm256_sub_ps(rinv30,pmecorrV);
 426             velec            = _mm256_mul_ps(qq30,velec);
 427
 428             /* Update potential sum for this i atom from the interaction with this j atom. */
 429             velecsum         = _mm256_add_ps(velecsum,velec);
 430
 431             fscal            = felec;
 432
 433             /* Calculate temporary vectorial force */
 434             tx               = _mm256_mul_ps(fscal,dx30);
 435             ty               = _mm256_mul_ps(fscal,dy30);
 436             tz               = _mm256_mul_ps(fscal,dz30);
 437
 438             /* Update vectorial force */
 439             fix3             = _mm256_add_ps(fix3,tx);
 440             fiy3             = _mm256_add_ps(fiy3,ty);
 441             fiz3             = _mm256_add_ps(fiz3,tz);
 442
 443             fjx0             = _mm256_add_ps(fjx0,tx);
 444             fjy0             = _mm256_add_ps(fjy0,ty);
 445             fjz0             = _mm256_add_ps(fjz0,tz);
 446
 447             fjptrA             = f+j_coord_offsetA;
 448             fjptrB             = f+j_coord_offsetB;
 449             fjptrC             = f+j_coord_offsetC;
 450             fjptrD             = f+j_coord_offsetD;
 451             fjptrE             = f+j_coord_offsetE;
 452             fjptrF             = f+j_coord_offsetF;
 453             fjptrG             = f+j_coord_offsetG;
 454             fjptrH             = f+j_coord_offsetH;
 455
 456             gmx_mm256_decrement_1rvec_8ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjptrE,fjptrF,fjptrG,fjptrH,fjx0,fjy0,fjz0);
 457
 458             /* Inner loop uses 287 flops */
 459         }
 460
 461         if(jidx<j_index_end)
 462         {
 463
 464             /* Get j neighbor index, and coordinate index */
 465             jnrlistA         = jjnr[jidx];
 466             jnrlistB         = jjnr[jidx+1];
 467             jnrlistC         = jjnr[jidx+2];
 468             jnrlistD         = jjnr[jidx+3];
 469             jnrlistE         = jjnr[jidx+4];
 470             jnrlistF         = jjnr[jidx+5];
 471             jnrlistG         = jjnr[jidx+6];
 472             jnrlistH         = jjnr[jidx+7];
 473             /* Sign of each element will be negative for non-real atoms.
 474              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 475              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
 476              */
 477             dummy_mask = gmx_mm256_set_m128(gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx+4)),_mm_setzero_si128())),
 478                                             gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128())));
 479
 480             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
 481             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
 482             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
 483             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
 484             jnrE       = (jnrlistE>=0) ? jnrlistE : 0;
 485             jnrF       = (jnrlistF>=0) ? jnrlistF : 0;
 486             jnrG       = (jnrlistG>=0) ? jnrlistG : 0;
 487             jnrH       = (jnrlistH>=0) ? jnrlistH : 0;
 488             j_coord_offsetA  = DIM*jnrA;
 489             j_coord_offsetB  = DIM*jnrB;
 490             j_coord_offsetC  = DIM*jnrC;
 491             j_coord_offsetD  = DIM*jnrD;
 492             j_coord_offsetE  = DIM*jnrE;
 493             j_coord_offsetF  = DIM*jnrF;
 494             j_coord_offsetG  = DIM*jnrG;
 495             j_coord_offsetH  = DIM*jnrH;
 496
 497             /* load j atom coordinates */
 498             gmx_mm256_load_1rvec_8ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 499                                                  x+j_coord_offsetC,x+j_coord_offsetD,
 500                                                  x+j_coord_offsetE,x+j_coord_offsetF,
 501                                                  x+j_coord_offsetG,x+j_coord_offsetH,
 502                                                  &jx0,&jy0,&jz0);
 503
 504             /* Calculate displacement vector */
 505             dx00             = _mm256_sub_ps(ix0,jx0);
 506             dy00             = _mm256_sub_ps(iy0,jy0);
 507             dz00             = _mm256_sub_ps(iz0,jz0);
 508             dx10             = _mm256_sub_ps(ix1,jx0);
 509             dy10             = _mm256_sub_ps(iy1,jy0);
 510             dz10             = _mm256_sub_ps(iz1,jz0);
 511             dx20             = _mm256_sub_ps(ix2,jx0);
 512             dy20             = _mm256_sub_ps(iy2,jy0);
 513             dz20             = _mm256_sub_ps(iz2,jz0);
 514             dx30             = _mm256_sub_ps(ix3,jx0);
 515             dy30             = _mm256_sub_ps(iy3,jy0);
 516             dz30             = _mm256_sub_ps(iz3,jz0);
 517
 518             /* Calculate squared distance and things based on it */
 519             rsq00            = gmx_mm256_calc_rsq_ps(dx00,dy00,dz00);
 520             rsq10            = gmx_mm256_calc_rsq_ps(dx10,dy10,dz10);
 521             rsq20            = gmx_mm256_calc_rsq_ps(dx20,dy20,dz20);
 522             rsq30            = gmx_mm256_calc_rsq_ps(dx30,dy30,dz30);
 523
 524             rinv10           = gmx_mm256_invsqrt_ps(rsq10);
 525             rinv20           = gmx_mm256_invsqrt_ps(rsq20);
 526             rinv30           = gmx_mm256_invsqrt_ps(rsq30);
 527
 528             rinvsq00         = gmx_mm256_inv_ps(rsq00);
 529             rinvsq10         = _mm256_mul_ps(rinv10,rinv10);
 530             rinvsq20         = _mm256_mul_ps(rinv20,rinv20);
 531             rinvsq30         = _mm256_mul_ps(rinv30,rinv30);
 532
 533             /* Load parameters for j particles */
 534             jq0              = gmx_mm256_load_8real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 535                                                                  charge+jnrC+0,charge+jnrD+0,
 536                                                                  charge+jnrE+0,charge+jnrF+0,
 537                                                                  charge+jnrG+0,charge+jnrH+0);
 538             vdwjidx0A        = 2*vdwtype[jnrA+0];
 539             vdwjidx0B        = 2*vdwtype[jnrB+0];
 540             vdwjidx0C        = 2*vdwtype[jnrC+0];
 541             vdwjidx0D        = 2*vdwtype[jnrD+0];
 542             vdwjidx0E        = 2*vdwtype[jnrE+0];
 543             vdwjidx0F        = 2*vdwtype[jnrF+0];
 544             vdwjidx0G        = 2*vdwtype[jnrG+0];
 545             vdwjidx0H        = 2*vdwtype[jnrH+0];
 546
 547             fjx0             = _mm256_setzero_ps();
 548             fjy0             = _mm256_setzero_ps();
 549             fjz0             = _mm256_setzero_ps();
 550
 551             /**************************
 552              * CALCULATE INTERACTIONS *
 553              **************************/
 554
 555             /* Compute parameters for interactions between i and j atoms */
 556             gmx_mm256_load_8pair_swizzle_ps(vdwioffsetptr0+vdwjidx0A,
 557                                             vdwioffsetptr0+vdwjidx0B,
 558                                             vdwioffsetptr0+vdwjidx0C,
 559                                             vdwioffsetptr0+vdwjidx0D,
 560                                             vdwioffsetptr0+vdwjidx0E,
 561                                             vdwioffsetptr0+vdwjidx0F,
 562                                             vdwioffsetptr0+vdwjidx0G,
 563                                             vdwioffsetptr0+vdwjidx0H,
 564                                             &c6_00,&c12_00);
 565
 566             /* LENNARD-JONES DISPERSION/REPULSION */
 567
 568             rinvsix          = _mm256_mul_ps(_mm256_mul_ps(rinvsq00,rinvsq00),rinvsq00);
 569             vvdw6            = _mm256_mul_ps(c6_00,rinvsix);
 570             vvdw12           = _mm256_mul_ps(c12_00,_mm256_mul_ps(rinvsix,rinvsix));
 571             vvdw             = _mm256_sub_ps( _mm256_mul_ps(vvdw12,one_twelfth) , _mm256_mul_ps(vvdw6,one_sixth) );
 572             fvdw             = _mm256_mul_ps(_mm256_sub_ps(vvdw12,vvdw6),rinvsq00);
 573
 574             /* Update potential sum for this i atom from the interaction with this j atom. */
 575             vvdw             = _mm256_andnot_ps(dummy_mask,vvdw);
 576             vvdwsum          = _mm256_add_ps(vvdwsum,vvdw);
 577
 578             fscal            = fvdw;
 579
 580             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
 581
 582             /* Calculate temporary vectorial force */
 583             tx               = _mm256_mul_ps(fscal,dx00);
 584             ty               = _mm256_mul_ps(fscal,dy00);
 585             tz               = _mm256_mul_ps(fscal,dz00);
 586
 587             /* Update vectorial force */
 588             fix0             = _mm256_add_ps(fix0,tx);
 589             fiy0             = _mm256_add_ps(fiy0,ty);
 590             fiz0             = _mm256_add_ps(fiz0,tz);
 591
 592             fjx0             = _mm256_add_ps(fjx0,tx);
 593             fjy0             = _mm256_add_ps(fjy0,ty);
 594             fjz0             = _mm256_add_ps(fjz0,tz);
 595
 596             /**************************
 597              * CALCULATE INTERACTIONS *
 598              **************************/
 599
 600             r10              = _mm256_mul_ps(rsq10,rinv10);
 601             r10              = _mm256_andnot_ps(dummy_mask,r10);
 602
 603             /* Compute parameters for interactions between i and j atoms */
 604             qq10             = _mm256_mul_ps(iq1,jq0);
 605
 606             /* EWALD ELECTROSTATICS */
 607
 608             /* Analytical PME correction */
 609             zeta2            = _mm256_mul_ps(beta2,rsq10);
 610             rinv3            = _mm256_mul_ps(rinvsq10,rinv10);
 611             pmecorrF         = gmx_mm256_pmecorrF_ps(zeta2);
 612             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
 613             felec            = _mm256_mul_ps(qq10,felec);
 614             pmecorrV         = gmx_mm256_pmecorrV_ps(zeta2);
 615             pmecorrV         = _mm256_mul_ps(pmecorrV,beta);
 616             velec            = _mm256_sub_ps(rinv10,pmecorrV);
 617             velec            = _mm256_mul_ps(qq10,velec);
 618
 619             /* Update potential sum for this i atom from the interaction with this j atom. */
 620             velec            = _mm256_andnot_ps(dummy_mask,velec);
 621             velecsum         = _mm256_add_ps(velecsum,velec);
 622
 623             fscal            = felec;
 624
 625             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
 626
 627             /* Calculate temporary vectorial force */
 628             tx               = _mm256_mul_ps(fscal,dx10);
 629             ty               = _mm256_mul_ps(fscal,dy10);
 630             tz               = _mm256_mul_ps(fscal,dz10);
 631
 632             /* Update vectorial force */
 633             fix1             = _mm256_add_ps(fix1,tx);
 634             fiy1             = _mm256_add_ps(fiy1,ty);
 635             fiz1             = _mm256_add_ps(fiz1,tz);
 636
 637             fjx0             = _mm256_add_ps(fjx0,tx);
 638             fjy0             = _mm256_add_ps(fjy0,ty);
 639             fjz0             = _mm256_add_ps(fjz0,tz);
 640
 641             /**************************
 642              * CALCULATE INTERACTIONS *
 643              **************************/
 644
 645             r20              = _mm256_mul_ps(rsq20,rinv20);
 646             r20              = _mm256_andnot_ps(dummy_mask,r20);
 647
 648             /* Compute parameters for interactions between i and j atoms */
 649             qq20             = _mm256_mul_ps(iq2,jq0);
 650
 651             /* EWALD ELECTROSTATICS */
 652
 653             /* Analytical PME correction */
 654             zeta2            = _mm256_mul_ps(beta2,rsq20);
 655             rinv3            = _mm256_mul_ps(rinvsq20,rinv20);
 656             pmecorrF         = gmx_mm256_pmecorrF_ps(zeta2);
 657             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
 658             felec            = _mm256_mul_ps(qq20,felec);
 659             pmecorrV         = gmx_mm256_pmecorrV_ps(zeta2);
 660             pmecorrV         = _mm256_mul_ps(pmecorrV,beta);
 661             velec            = _mm256_sub_ps(rinv20,pmecorrV);
 662             velec            = _mm256_mul_ps(qq20,velec);
 663
 664             /* Update potential sum for this i atom from the interaction with this j atom. */
 665             velec            = _mm256_andnot_ps(dummy_mask,velec);
 666             velecsum         = _mm256_add_ps(velecsum,velec);
 667
 668             fscal            = felec;
 669
 670             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
 671
 672             /* Calculate temporary vectorial force */
 673             tx               = _mm256_mul_ps(fscal,dx20);
 674             ty               = _mm256_mul_ps(fscal,dy20);
 675             tz               = _mm256_mul_ps(fscal,dz20);
 676
 677             /* Update vectorial force */
 678             fix2             = _mm256_add_ps(fix2,tx);
 679             fiy2             = _mm256_add_ps(fiy2,ty);
 680             fiz2             = _mm256_add_ps(fiz2,tz);
 681
 682             fjx0             = _mm256_add_ps(fjx0,tx);
 683             fjy0             = _mm256_add_ps(fjy0,ty);
 684             fjz0             = _mm256_add_ps(fjz0,tz);
 685
 686             /**************************
 687              * CALCULATE INTERACTIONS *
 688              **************************/
 689
 690             r30              = _mm256_mul_ps(rsq30,rinv30);
 691             r30              = _mm256_andnot_ps(dummy_mask,r30);
 692
 693             /* Compute parameters for interactions between i and j atoms */
 694             qq30             = _mm256_mul_ps(iq3,jq0);
 695
 696             /* EWALD ELECTROSTATICS */
 697
 698             /* Analytical PME correction */
 699             zeta2            = _mm256_mul_ps(beta2,rsq30);
 700             rinv3            = _mm256_mul_ps(rinvsq30,rinv30);
 701             pmecorrF         = gmx_mm256_pmecorrF_ps(zeta2);
 702             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
 703             felec            = _mm256_mul_ps(qq30,felec);
 704             pmecorrV         = gmx_mm256_pmecorrV_ps(zeta2);
 705             pmecorrV         = _mm256_mul_ps(pmecorrV,beta);
 706             velec            = _mm256_sub_ps(rinv30,pmecorrV);
 707             velec            = _mm256_mul_ps(qq30,velec);
 708
 709             /* Update potential sum for this i atom from the interaction with this j atom. */
 710             velec            = _mm256_andnot_ps(dummy_mask,velec);
 711             velecsum         = _mm256_add_ps(velecsum,velec);
 712
 713             fscal            = felec;
 714
 715             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
 716
 717             /* Calculate temporary vectorial force */
 718             tx               = _mm256_mul_ps(fscal,dx30);
 719             ty               = _mm256_mul_ps(fscal,dy30);
 720             tz               = _mm256_mul_ps(fscal,dz30);
 721
 722             /* Update vectorial force */
 723             fix3             = _mm256_add_ps(fix3,tx);
 724             fiy3             = _mm256_add_ps(fiy3,ty);
 725             fiz3             = _mm256_add_ps(fiz3,tz);
 726
 727             fjx0             = _mm256_add_ps(fjx0,tx);
 728             fjy0             = _mm256_add_ps(fjy0,ty);
 729             fjz0             = _mm256_add_ps(fjz0,tz);
 730
 731             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
 732             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
 733             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
 734             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
 735             fjptrE             = (jnrlistE>=0) ? f+j_coord_offsetE : scratch;
 736             fjptrF             = (jnrlistF>=0) ? f+j_coord_offsetF : scratch;
 737             fjptrG             = (jnrlistG>=0) ? f+j_coord_offsetG : scratch;
 738             fjptrH             = (jnrlistH>=0) ? f+j_coord_offsetH : scratch;
 739
 740             gmx_mm256_decrement_1rvec_8ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjptrE,fjptrF,fjptrG,fjptrH,fjx0,fjy0,fjz0);
 741
 742             /* Inner loop uses 290 flops */
 743         }
 744
 745         /* End of innermost loop */
 746
 747         gmx_mm256_update_iforce_4atom_swizzle_ps(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,fix3,fiy3,fiz3,
 748                                                  f+i_coord_offset,fshift+i_shift_offset);
 749
 750         ggid                        = gid[iidx];
 751         /* Update potential energies */
 752         gmx_mm256_update_1pot_ps(velecsum,kernel_data->energygrp_elec+ggid);
 753         gmx_mm256_update_1pot_ps(vvdwsum,kernel_data->energygrp_vdw+ggid);
 754
 755         /* Increment number of inner iterations */
 756         inneriter                  += j_index_end - j_index_start;
 757
 758         /* Outer loop uses 26 flops */
 759     }
 760
 761     /* Increment number of outer iterations */
 762     outeriter        += nri;
 763
 764     /* Update outer/inner flops */
 765
 766     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W4_VF,outeriter*26 + inneriter*290);
 767 }
 768 /*
 769  * Gromacs nonbonded kernel:   nb_kernel_ElecEw_VdwLJ_GeomW4P1_F_avx_256_single
 770  * Electrostatics interaction: Ewald
 771  * VdW interaction:            LennardJones
 772  * Geometry:                   Water4-Particle
 773  * Calculate force/pot:        Force
 774  */
 775 void
 776 nb_kernel_ElecEw_VdwLJ_GeomW4P1_F_avx_256_single
 777                     (t_nblist                    * gmx_restrict       nlist,
 778                      rvec                        * gmx_restrict          xx,
 779                      rvec                        * gmx_restrict          ff,
 780                      t_forcerec                  * gmx_restrict          fr,
 781                      t_mdatoms                   * gmx_restrict     mdatoms,
 782                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
 783                      t_nrnb                      * gmx_restrict        nrnb)
 784 {
 785     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
 786      * just 0 for non-waters.
 787      * Suffixes A,B,C,D,E,F,G,H refer to j loop unrolling done with AVX, e.g. for the eight different
 788      * jnr indices corresponding to data put in the four positions in the SIMD register.
 789      */
 790     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
 791     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
 792     int              jnrA,jnrB,jnrC,jnrD;
 793     int              jnrE,jnrF,jnrG,jnrH;
 794     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
 795     int              jnrlistE,jnrlistF,jnrlistG,jnrlistH;
 796     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
 797     int              j_coord_offsetE,j_coord_offsetF,j_coord_offsetG,j_coord_offsetH;
 798     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
 799     real             rcutoff_scalar;
 800     real             *shiftvec,*fshift,*x,*f;
 801     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD,*fjptrE,*fjptrF,*fjptrG,*fjptrH;
 802     real             scratch[4*DIM];
 803     __m256           tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
 804     real *           vdwioffsetptr0;
 805     __m256           ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
 806     real *           vdwioffsetptr1;
 807     __m256           ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
 808     real *           vdwioffsetptr2;
 809     __m256           ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
 810     real *           vdwioffsetptr3;
 811     __m256           ix3,iy3,iz3,fix3,fiy3,fiz3,iq3,isai3;
 812     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D,vdwjidx0E,vdwjidx0F,vdwjidx0G,vdwjidx0H;
 813     __m256           jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
 814     __m256           dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
 815     __m256           dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
 816     __m256           dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
 817     __m256           dx30,dy30,dz30,rsq30,rinv30,rinvsq30,r30,qq30,c6_30,c12_30;
 818     __m256           velec,felec,velecsum,facel,crf,krf,krf2;
 819     real             *charge;
 820     int              nvdwtype;
 821     __m256           rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
 822     int              *vdwtype;
 823     real             *vdwparam;
 824     __m256           one_sixth   = _mm256_set1_ps(1.0/6.0);
 825     __m256           one_twelfth = _mm256_set1_ps(1.0/12.0);
 826     __m256i          ewitab;
 827     __m128i          ewitab_lo,ewitab_hi;
 828     __m256           ewtabscale,eweps,sh_ewald,ewrt,ewtabhalfspace,ewtabF,ewtabFn,ewtabD,ewtabV;
 829     __m256           beta,beta2,beta3,zeta2,pmecorrF,pmecorrV,rinv3;
 830     real             *ewtab;
 831     __m256           dummy_mask,cutoff_mask;
 832     __m256           signbit = _mm256_castsi256_ps( _mm256_set1_epi32(0x80000000) );
 833     __m256           one     = _mm256_set1_ps(1.0);
 834     __m256           two     = _mm256_set1_ps(2.0);
 835     x                = xx[0];
 836     f                = ff[0];
 837
 838     nri              = nlist->nri;
 839     iinr             = nlist->iinr;
 840     jindex           = nlist->jindex;
 841     jjnr             = nlist->jjnr;
 842     shiftidx         = nlist->shift;
 843     gid              = nlist->gid;
 844     shiftvec         = fr->shift_vec[0];
 845     fshift           = fr->fshift[0];
 846     facel            = _mm256_set1_ps(fr->epsfac);
 847     charge           = mdatoms->chargeA;
 848     nvdwtype         = fr->ntype;
 849     vdwparam         = fr->nbfp;
 850     vdwtype          = mdatoms->typeA;
 851
 852     sh_ewald         = _mm256_set1_ps(fr->ic->sh_ewald);
 853     beta             = _mm256_set1_ps(fr->ic->ewaldcoeff_q);
 854     beta2            = _mm256_mul_ps(beta,beta);
 855     beta3            = _mm256_mul_ps(beta,beta2);
 856
 857     ewtab            = fr->ic->tabq_coul_F;
 858     ewtabscale       = _mm256_set1_ps(fr->ic->tabq_scale);
 859     ewtabhalfspace   = _mm256_set1_ps(0.5/fr->ic->tabq_scale);
 860
 861     /* Setup water-specific parameters */
 862     inr              = nlist->iinr[0];
 863     iq1              = _mm256_mul_ps(facel,_mm256_set1_ps(charge[inr+1]));
 864     iq2              = _mm256_mul_ps(facel,_mm256_set1_ps(charge[inr+2]));
 865     iq3              = _mm256_mul_ps(facel,_mm256_set1_ps(charge[inr+3]));
 866     vdwioffsetptr0   = vdwparam+2*nvdwtype*vdwtype[inr+0];
 867
 868     /* Avoid stupid compiler warnings */
 869     jnrA = jnrB = jnrC = jnrD = jnrE = jnrF = jnrG = jnrH = 0;
 870     j_coord_offsetA = 0;
 871     j_coord_offsetB = 0;
 872     j_coord_offsetC = 0;
 873     j_coord_offsetD = 0;
 874     j_coord_offsetE = 0;
 875     j_coord_offsetF = 0;
 876     j_coord_offsetG = 0;
 877     j_coord_offsetH = 0;
 878
 879     outeriter        = 0;
 880     inneriter        = 0;
 881
 882     for(iidx=0;iidx<4*DIM;iidx++)
 883     {
 884         scratch[iidx] = 0.0;
 885     }
 886
 887     /* Start outer loop over neighborlists */
 888     for(iidx=0; iidx<nri; iidx++)
 889     {
 890         /* Load shift vector for this list */
 891         i_shift_offset   = DIM*shiftidx[iidx];
 892
 893         /* Load limits for loop over neighbors */
 894         j_index_start    = jindex[iidx];
 895         j_index_end      = jindex[iidx+1];
 896
 897         /* Get outer coordinate index */
 898         inr              = iinr[iidx];
 899         i_coord_offset   = DIM*inr;
 900
 901         /* Load i particle coords and add shift vector */
 902         gmx_mm256_load_shift_and_4rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,
 903                                                     &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2,&ix3,&iy3,&iz3);
 904
 905         fix0             = _mm256_setzero_ps();
 906         fiy0             = _mm256_setzero_ps();
 907         fiz0             = _mm256_setzero_ps();
 908         fix1             = _mm256_setzero_ps();
 909         fiy1             = _mm256_setzero_ps();
 910         fiz1             = _mm256_setzero_ps();
 911         fix2             = _mm256_setzero_ps();
 912         fiy2             = _mm256_setzero_ps();
 913         fiz2             = _mm256_setzero_ps();
 914         fix3             = _mm256_setzero_ps();
 915         fiy3             = _mm256_setzero_ps();
 916         fiz3             = _mm256_setzero_ps();
 917
 918         /* Start inner kernel loop */
 919         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+7]>=0; jidx+=8)
 920         {
 921
 922             /* Get j neighbor index, and coordinate index */
 923             jnrA             = jjnr[jidx];
 924             jnrB             = jjnr[jidx+1];
 925             jnrC             = jjnr[jidx+2];
 926             jnrD             = jjnr[jidx+3];
 927             jnrE             = jjnr[jidx+4];
 928             jnrF             = jjnr[jidx+5];
 929             jnrG             = jjnr[jidx+6];
 930             jnrH             = jjnr[jidx+7];
 931             j_coord_offsetA  = DIM*jnrA;
 932             j_coord_offsetB  = DIM*jnrB;
 933             j_coord_offsetC  = DIM*jnrC;
 934             j_coord_offsetD  = DIM*jnrD;
 935             j_coord_offsetE  = DIM*jnrE;
 936             j_coord_offsetF  = DIM*jnrF;
 937             j_coord_offsetG  = DIM*jnrG;
 938             j_coord_offsetH  = DIM*jnrH;
 939
 940             /* load j atom coordinates */
 941             gmx_mm256_load_1rvec_8ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 942                                                  x+j_coord_offsetC,x+j_coord_offsetD,
 943                                                  x+j_coord_offsetE,x+j_coord_offsetF,
 944                                                  x+j_coord_offsetG,x+j_coord_offsetH,
 945                                                  &jx0,&jy0,&jz0);
 946
 947             /* Calculate displacement vector */
 948             dx00             = _mm256_sub_ps(ix0,jx0);
 949             dy00             = _mm256_sub_ps(iy0,jy0);
 950             dz00             = _mm256_sub_ps(iz0,jz0);
 951             dx10             = _mm256_sub_ps(ix1,jx0);
 952             dy10             = _mm256_sub_ps(iy1,jy0);
 953             dz10             = _mm256_sub_ps(iz1,jz0);
 954             dx20             = _mm256_sub_ps(ix2,jx0);
 955             dy20             = _mm256_sub_ps(iy2,jy0);
 956             dz20             = _mm256_sub_ps(iz2,jz0);
 957             dx30             = _mm256_sub_ps(ix3,jx0);
 958             dy30             = _mm256_sub_ps(iy3,jy0);
 959             dz30             = _mm256_sub_ps(iz3,jz0);
 960
 961             /* Calculate squared distance and things based on it */
 962             rsq00            = gmx_mm256_calc_rsq_ps(dx00,dy00,dz00);
 963             rsq10            = gmx_mm256_calc_rsq_ps(dx10,dy10,dz10);
 964             rsq20            = gmx_mm256_calc_rsq_ps(dx20,dy20,dz20);
 965             rsq30            = gmx_mm256_calc_rsq_ps(dx30,dy30,dz30);
 966
 967             rinv10           = gmx_mm256_invsqrt_ps(rsq10);
 968             rinv20           = gmx_mm256_invsqrt_ps(rsq20);
 969             rinv30           = gmx_mm256_invsqrt_ps(rsq30);
 970
 971             rinvsq00         = gmx_mm256_inv_ps(rsq00);
 972             rinvsq10         = _mm256_mul_ps(rinv10,rinv10);
 973             rinvsq20         = _mm256_mul_ps(rinv20,rinv20);
 974             rinvsq30         = _mm256_mul_ps(rinv30,rinv30);
 975
 976             /* Load parameters for j particles */
 977             jq0              = gmx_mm256_load_8real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 978                                                                  charge+jnrC+0,charge+jnrD+0,
 979                                                                  charge+jnrE+0,charge+jnrF+0,
 980                                                                  charge+jnrG+0,charge+jnrH+0);
 981             vdwjidx0A        = 2*vdwtype[jnrA+0];
 982             vdwjidx0B        = 2*vdwtype[jnrB+0];
 983             vdwjidx0C        = 2*vdwtype[jnrC+0];
 984             vdwjidx0D        = 2*vdwtype[jnrD+0];
 985             vdwjidx0E        = 2*vdwtype[jnrE+0];
 986             vdwjidx0F        = 2*vdwtype[jnrF+0];
 987             vdwjidx0G        = 2*vdwtype[jnrG+0];
 988             vdwjidx0H        = 2*vdwtype[jnrH+0];
 989
 990             fjx0             = _mm256_setzero_ps();
 991             fjy0             = _mm256_setzero_ps();
 992             fjz0             = _mm256_setzero_ps();
 993
 994             /**************************
 995              * CALCULATE INTERACTIONS *
 996              **************************/
 997
 998             /* Compute parameters for interactions between i and j atoms */
 999             gmx_mm256_load_8pair_swizzle_ps(vdwioffsetptr0+vdwjidx0A,
1000                                             vdwioffsetptr0+vdwjidx0B,
1001                                             vdwioffsetptr0+vdwjidx0C,
1002                                             vdwioffsetptr0+vdwjidx0D,
1003                                             vdwioffsetptr0+vdwjidx0E,
1004                                             vdwioffsetptr0+vdwjidx0F,
1005                                             vdwioffsetptr0+vdwjidx0G,
1006                                             vdwioffsetptr0+vdwjidx0H,
1007                                             &c6_00,&c12_00);
1008
1009             /* LENNARD-JONES DISPERSION/REPULSION */
1010
1011             rinvsix          = _mm256_mul_ps(_mm256_mul_ps(rinvsq00,rinvsq00),rinvsq00);
1012             fvdw             = _mm256_mul_ps(_mm256_sub_ps(_mm256_mul_ps(c12_00,rinvsix),c6_00),_mm256_mul_ps(rinvsix,rinvsq00));
1013
1014             fscal            = fvdw;
1015
1016             /* Calculate temporary vectorial force */
1017             tx               = _mm256_mul_ps(fscal,dx00);
1018             ty               = _mm256_mul_ps(fscal,dy00);
1019             tz               = _mm256_mul_ps(fscal,dz00);
1020
1021             /* Update vectorial force */
1022             fix0             = _mm256_add_ps(fix0,tx);
1023             fiy0             = _mm256_add_ps(fiy0,ty);
1024             fiz0             = _mm256_add_ps(fiz0,tz);
1025
1026             fjx0             = _mm256_add_ps(fjx0,tx);
1027             fjy0             = _mm256_add_ps(fjy0,ty);
1028             fjz0             = _mm256_add_ps(fjz0,tz);
1029
1030             /**************************
1031              * CALCULATE INTERACTIONS *
1032              **************************/
1033
1034             r10              = _mm256_mul_ps(rsq10,rinv10);
1035
1036             /* Compute parameters for interactions between i and j atoms */
1037             qq10             = _mm256_mul_ps(iq1,jq0);
1038
1039             /* EWALD ELECTROSTATICS */
1040
1041             /* Analytical PME correction */
1042             zeta2            = _mm256_mul_ps(beta2,rsq10);
1043             rinv3            = _mm256_mul_ps(rinvsq10,rinv10);
1044             pmecorrF         = gmx_mm256_pmecorrF_ps(zeta2);
1045             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
1046             felec            = _mm256_mul_ps(qq10,felec);
1047
1048             fscal            = felec;
1049
1050             /* Calculate temporary vectorial force */
1051             tx               = _mm256_mul_ps(fscal,dx10);
1052             ty               = _mm256_mul_ps(fscal,dy10);
1053             tz               = _mm256_mul_ps(fscal,dz10);
1054
1055             /* Update vectorial force */
1056             fix1             = _mm256_add_ps(fix1,tx);
1057             fiy1             = _mm256_add_ps(fiy1,ty);
1058             fiz1             = _mm256_add_ps(fiz1,tz);
1059
1060             fjx0             = _mm256_add_ps(fjx0,tx);
1061             fjy0             = _mm256_add_ps(fjy0,ty);
1062             fjz0             = _mm256_add_ps(fjz0,tz);
1063
1064             /**************************
1065              * CALCULATE INTERACTIONS *
1066              **************************/
1067
1068             r20              = _mm256_mul_ps(rsq20,rinv20);
1069
1070             /* Compute parameters for interactions between i and j atoms */
1071             qq20             = _mm256_mul_ps(iq2,jq0);
1072
1073             /* EWALD ELECTROSTATICS */
1074
1075             /* Analytical PME correction */
1076             zeta2            = _mm256_mul_ps(beta2,rsq20);
1077             rinv3            = _mm256_mul_ps(rinvsq20,rinv20);
1078             pmecorrF         = gmx_mm256_pmecorrF_ps(zeta2);
1079             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
1080             felec            = _mm256_mul_ps(qq20,felec);
1081
1082             fscal            = felec;
1083
1084             /* Calculate temporary vectorial force */
1085             tx               = _mm256_mul_ps(fscal,dx20);
1086             ty               = _mm256_mul_ps(fscal,dy20);
1087             tz               = _mm256_mul_ps(fscal,dz20);
1088
1089             /* Update vectorial force */
1090             fix2             = _mm256_add_ps(fix2,tx);
1091             fiy2             = _mm256_add_ps(fiy2,ty);
1092             fiz2             = _mm256_add_ps(fiz2,tz);
1093
1094             fjx0             = _mm256_add_ps(fjx0,tx);
1095             fjy0             = _mm256_add_ps(fjy0,ty);
1096             fjz0             = _mm256_add_ps(fjz0,tz);
1097
1098             /**************************
1099              * CALCULATE INTERACTIONS *
1100              **************************/
1101
1102             r30              = _mm256_mul_ps(rsq30,rinv30);
1103
1104             /* Compute parameters for interactions between i and j atoms */
1105             qq30             = _mm256_mul_ps(iq3,jq0);
1106
1107             /* EWALD ELECTROSTATICS */
1108
1109             /* Analytical PME correction */
1110             zeta2            = _mm256_mul_ps(beta2,rsq30);
1111             rinv3            = _mm256_mul_ps(rinvsq30,rinv30);
1112             pmecorrF         = gmx_mm256_pmecorrF_ps(zeta2);
1113             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
1114             felec            = _mm256_mul_ps(qq30,felec);
1115
1116             fscal            = felec;
1117
1118             /* Calculate temporary vectorial force */
1119             tx               = _mm256_mul_ps(fscal,dx30);
1120             ty               = _mm256_mul_ps(fscal,dy30);
1121             tz               = _mm256_mul_ps(fscal,dz30);
1122
1123             /* Update vectorial force */
1124             fix3             = _mm256_add_ps(fix3,tx);
1125             fiy3             = _mm256_add_ps(fiy3,ty);
1126             fiz3             = _mm256_add_ps(fiz3,tz);
1127
1128             fjx0             = _mm256_add_ps(fjx0,tx);
1129             fjy0             = _mm256_add_ps(fjy0,ty);
1130             fjz0             = _mm256_add_ps(fjz0,tz);
1131
1132             fjptrA             = f+j_coord_offsetA;
1133             fjptrB             = f+j_coord_offsetB;
1134             fjptrC             = f+j_coord_offsetC;
1135             fjptrD             = f+j_coord_offsetD;
1136             fjptrE             = f+j_coord_offsetE;
1137             fjptrF             = f+j_coord_offsetF;
1138             fjptrG             = f+j_coord_offsetG;
1139             fjptrH             = f+j_coord_offsetH;
1140
1141             gmx_mm256_decrement_1rvec_8ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjptrE,fjptrF,fjptrG,fjptrH,fjx0,fjy0,fjz0);
1142
1143             /* Inner loop uses 198 flops */
1144         }
1145
1146         if(jidx<j_index_end)
1147         {
1148
1149             /* Get j neighbor index, and coordinate index */
1150             jnrlistA         = jjnr[jidx];
1151             jnrlistB         = jjnr[jidx+1];
1152             jnrlistC         = jjnr[jidx+2];
1153             jnrlistD         = jjnr[jidx+3];
1154             jnrlistE         = jjnr[jidx+4];
1155             jnrlistF         = jjnr[jidx+5];
1156             jnrlistG         = jjnr[jidx+6];
1157             jnrlistH         = jjnr[jidx+7];
1158             /* Sign of each element will be negative for non-real atoms.
1159              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
1160              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
1161              */
1162             dummy_mask = gmx_mm256_set_m128(gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx+4)),_mm_setzero_si128())),
1163                                             gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128())));
1164
1165             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
1166             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
1167             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
1168             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
1169             jnrE       = (jnrlistE>=0) ? jnrlistE : 0;
1170             jnrF       = (jnrlistF>=0) ? jnrlistF : 0;
1171             jnrG       = (jnrlistG>=0) ? jnrlistG : 0;
1172             jnrH       = (jnrlistH>=0) ? jnrlistH : 0;
1173             j_coord_offsetA  = DIM*jnrA;
1174             j_coord_offsetB  = DIM*jnrB;
1175             j_coord_offsetC  = DIM*jnrC;
1176             j_coord_offsetD  = DIM*jnrD;
1177             j_coord_offsetE  = DIM*jnrE;
1178             j_coord_offsetF  = DIM*jnrF;
1179             j_coord_offsetG  = DIM*jnrG;
1180             j_coord_offsetH  = DIM*jnrH;
1181
1182             /* load j atom coordinates */
1183             gmx_mm256_load_1rvec_8ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
1184                                                  x+j_coord_offsetC,x+j_coord_offsetD,
1185                                                  x+j_coord_offsetE,x+j_coord_offsetF,
1186                                                  x+j_coord_offsetG,x+j_coord_offsetH,
1187                                                  &jx0,&jy0,&jz0);
1188
1189             /* Calculate displacement vector */
1190             dx00             = _mm256_sub_ps(ix0,jx0);
1191             dy00             = _mm256_sub_ps(iy0,jy0);
1192             dz00             = _mm256_sub_ps(iz0,jz0);
1193             dx10             = _mm256_sub_ps(ix1,jx0);
1194             dy10             = _mm256_sub_ps(iy1,jy0);
1195             dz10             = _mm256_sub_ps(iz1,jz0);
1196             dx20             = _mm256_sub_ps(ix2,jx0);
1197             dy20             = _mm256_sub_ps(iy2,jy0);
1198             dz20             = _mm256_sub_ps(iz2,jz0);
1199             dx30             = _mm256_sub_ps(ix3,jx0);
1200             dy30             = _mm256_sub_ps(iy3,jy0);
1201             dz30             = _mm256_sub_ps(iz3,jz0);
1202
1203             /* Calculate squared distance and things based on it */
1204             rsq00            = gmx_mm256_calc_rsq_ps(dx00,dy00,dz00);
1205             rsq10            = gmx_mm256_calc_rsq_ps(dx10,dy10,dz10);
1206             rsq20            = gmx_mm256_calc_rsq_ps(dx20,dy20,dz20);
1207             rsq30            = gmx_mm256_calc_rsq_ps(dx30,dy30,dz30);
1208
1209             rinv10           = gmx_mm256_invsqrt_ps(rsq10);
1210             rinv20           = gmx_mm256_invsqrt_ps(rsq20);
1211             rinv30           = gmx_mm256_invsqrt_ps(rsq30);
1212
1213             rinvsq00         = gmx_mm256_inv_ps(rsq00);
1214             rinvsq10         = _mm256_mul_ps(rinv10,rinv10);
1215             rinvsq20         = _mm256_mul_ps(rinv20,rinv20);
1216             rinvsq30         = _mm256_mul_ps(rinv30,rinv30);
1217
1218             /* Load parameters for j particles */
1219             jq0              = gmx_mm256_load_8real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
1220                                                                  charge+jnrC+0,charge+jnrD+0,
1221                                                                  charge+jnrE+0,charge+jnrF+0,
1222                                                                  charge+jnrG+0,charge+jnrH+0);
1223             vdwjidx0A        = 2*vdwtype[jnrA+0];
1224             vdwjidx0B        = 2*vdwtype[jnrB+0];
1225             vdwjidx0C        = 2*vdwtype[jnrC+0];
1226             vdwjidx0D        = 2*vdwtype[jnrD+0];
1227             vdwjidx0E        = 2*vdwtype[jnrE+0];
1228             vdwjidx0F        = 2*vdwtype[jnrF+0];
1229             vdwjidx0G        = 2*vdwtype[jnrG+0];
1230             vdwjidx0H        = 2*vdwtype[jnrH+0];
1231
1232             fjx0             = _mm256_setzero_ps();
1233             fjy0             = _mm256_setzero_ps();
1234             fjz0             = _mm256_setzero_ps();
1235
1236             /**************************
1237              * CALCULATE INTERACTIONS *
1238              **************************/
1239
1240             /* Compute parameters for interactions between i and j atoms */
1241             gmx_mm256_load_8pair_swizzle_ps(vdwioffsetptr0+vdwjidx0A,
1242                                             vdwioffsetptr0+vdwjidx0B,
1243                                             vdwioffsetptr0+vdwjidx0C,
1244                                             vdwioffsetptr0+vdwjidx0D,
1245                                             vdwioffsetptr0+vdwjidx0E,
1246                                             vdwioffsetptr0+vdwjidx0F,
1247                                             vdwioffsetptr0+vdwjidx0G,
1248                                             vdwioffsetptr0+vdwjidx0H,
1249                                             &c6_00,&c12_00);
1250
1251             /* LENNARD-JONES DISPERSION/REPULSION */
1252
1253             rinvsix          = _mm256_mul_ps(_mm256_mul_ps(rinvsq00,rinvsq00),rinvsq00);
1254             fvdw             = _mm256_mul_ps(_mm256_sub_ps(_mm256_mul_ps(c12_00,rinvsix),c6_00),_mm256_mul_ps(rinvsix,rinvsq00));
1255
1256             fscal            = fvdw;
1257
1258             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
1259
1260             /* Calculate temporary vectorial force */
1261             tx               = _mm256_mul_ps(fscal,dx00);
1262             ty               = _mm256_mul_ps(fscal,dy00);
1263             tz               = _mm256_mul_ps(fscal,dz00);
1264
1265             /* Update vectorial force */
1266             fix0             = _mm256_add_ps(fix0,tx);
1267             fiy0             = _mm256_add_ps(fiy0,ty);
1268             fiz0             = _mm256_add_ps(fiz0,tz);
1269
1270             fjx0             = _mm256_add_ps(fjx0,tx);
1271             fjy0             = _mm256_add_ps(fjy0,ty);
1272             fjz0             = _mm256_add_ps(fjz0,tz);
1273
1274             /**************************
1275              * CALCULATE INTERACTIONS *
1276              **************************/
1277
1278             r10              = _mm256_mul_ps(rsq10,rinv10);
1279             r10              = _mm256_andnot_ps(dummy_mask,r10);
1280
1281             /* Compute parameters for interactions between i and j atoms */
1282             qq10             = _mm256_mul_ps(iq1,jq0);
1283
1284             /* EWALD ELECTROSTATICS */
1285
1286             /* Analytical PME correction */
1287             zeta2            = _mm256_mul_ps(beta2,rsq10);
1288             rinv3            = _mm256_mul_ps(rinvsq10,rinv10);
1289             pmecorrF         = gmx_mm256_pmecorrF_ps(zeta2);
1290             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
1291             felec            = _mm256_mul_ps(qq10,felec);
1292
1293             fscal            = felec;
1294
1295             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
1296
1297             /* Calculate temporary vectorial force */
1298             tx               = _mm256_mul_ps(fscal,dx10);
1299             ty               = _mm256_mul_ps(fscal,dy10);
1300             tz               = _mm256_mul_ps(fscal,dz10);
1301
1302             /* Update vectorial force */
1303             fix1             = _mm256_add_ps(fix1,tx);
1304             fiy1             = _mm256_add_ps(fiy1,ty);
1305             fiz1             = _mm256_add_ps(fiz1,tz);
1306
1307             fjx0             = _mm256_add_ps(fjx0,tx);
1308             fjy0             = _mm256_add_ps(fjy0,ty);
1309             fjz0             = _mm256_add_ps(fjz0,tz);
1310
1311             /**************************
1312              * CALCULATE INTERACTIONS *
1313              **************************/
1314
1315             r20              = _mm256_mul_ps(rsq20,rinv20);
1316             r20              = _mm256_andnot_ps(dummy_mask,r20);
1317
1318             /* Compute parameters for interactions between i and j atoms */
1319             qq20             = _mm256_mul_ps(iq2,jq0);
1320
1321             /* EWALD ELECTROSTATICS */
1322
1323             /* Analytical PME correction */
1324             zeta2            = _mm256_mul_ps(beta2,rsq20);
1325             rinv3            = _mm256_mul_ps(rinvsq20,rinv20);
1326             pmecorrF         = gmx_mm256_pmecorrF_ps(zeta2);
1327             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
1328             felec            = _mm256_mul_ps(qq20,felec);
1329
1330             fscal            = felec;
1331
1332             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
1333
1334             /* Calculate temporary vectorial force */
1335             tx               = _mm256_mul_ps(fscal,dx20);
1336             ty               = _mm256_mul_ps(fscal,dy20);
1337             tz               = _mm256_mul_ps(fscal,dz20);
1338
1339             /* Update vectorial force */
1340             fix2             = _mm256_add_ps(fix2,tx);
1341             fiy2             = _mm256_add_ps(fiy2,ty);
1342             fiz2             = _mm256_add_ps(fiz2,tz);
1343
1344             fjx0             = _mm256_add_ps(fjx0,tx);
1345             fjy0             = _mm256_add_ps(fjy0,ty);
1346             fjz0             = _mm256_add_ps(fjz0,tz);
1347
1348             /**************************
1349              * CALCULATE INTERACTIONS *
1350              **************************/
1351
1352             r30              = _mm256_mul_ps(rsq30,rinv30);
1353             r30              = _mm256_andnot_ps(dummy_mask,r30);
1354
1355             /* Compute parameters for interactions between i and j atoms */
1356             qq30             = _mm256_mul_ps(iq3,jq0);
1357
1358             /* EWALD ELECTROSTATICS */
1359
1360             /* Analytical PME correction */
1361             zeta2            = _mm256_mul_ps(beta2,rsq30);
1362             rinv3            = _mm256_mul_ps(rinvsq30,rinv30);
1363             pmecorrF         = gmx_mm256_pmecorrF_ps(zeta2);
1364             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
1365             felec            = _mm256_mul_ps(qq30,felec);
1366
1367             fscal            = felec;
1368
1369             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
1370
1371             /* Calculate temporary vectorial force */
1372             tx               = _mm256_mul_ps(fscal,dx30);
1373             ty               = _mm256_mul_ps(fscal,dy30);
1374             tz               = _mm256_mul_ps(fscal,dz30);
1375
1376             /* Update vectorial force */
1377             fix3             = _mm256_add_ps(fix3,tx);
1378             fiy3             = _mm256_add_ps(fiy3,ty);
1379             fiz3             = _mm256_add_ps(fiz3,tz);
1380
1381             fjx0             = _mm256_add_ps(fjx0,tx);
1382             fjy0             = _mm256_add_ps(fjy0,ty);
1383             fjz0             = _mm256_add_ps(fjz0,tz);
1384
1385             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
1386             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
1387             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
1388             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
1389             fjptrE             = (jnrlistE>=0) ? f+j_coord_offsetE : scratch;
1390             fjptrF             = (jnrlistF>=0) ? f+j_coord_offsetF : scratch;
1391             fjptrG             = (jnrlistG>=0) ? f+j_coord_offsetG : scratch;
1392             fjptrH             = (jnrlistH>=0) ? f+j_coord_offsetH : scratch;
1393
1394             gmx_mm256_decrement_1rvec_8ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjptrE,fjptrF,fjptrG,fjptrH,fjx0,fjy0,fjz0);
1395
1396             /* Inner loop uses 201 flops */
1397         }
1398
1399         /* End of innermost loop */
1400
1401         gmx_mm256_update_iforce_4atom_swizzle_ps(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,fix3,fiy3,fiz3,
1402                                                  f+i_coord_offset,fshift+i_shift_offset);
1403
1404         /* Increment number of inner iterations */
1405         inneriter                  += j_index_end - j_index_start;
1406
1407         /* Outer loop uses 24 flops */
1408     }
1409
1410     /* Increment number of outer iterations */
1411     outeriter        += nri;
1412
1413     /* Update outer/inner flops */
1414
1415     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W4_F,outeriter*24 + inneriter*201);
1416 }