src/gromacs/gmxlib/nonbonded/nb_kernel_avx_256_single/nb_kernel_ElecRFCut_VdwLJSh_GeomW3W3_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 "config.h"
  39
  40 #include <math.h>
  41
  42 #include "../nb_kernel.h"
  43 #include "types/simple.h"
  44 #include "gromacs/math/vec.h"
  45 #include "nrnb.h"
  46
  47 #include "gromacs/simd/math_x86_avx_256_single.h"
  48 #include "kernelutil_x86_avx_256_single.h"
  49
  50 /*
  51  * Gromacs nonbonded kernel:   nb_kernel_ElecRFCut_VdwLJSh_GeomW3W3_VF_avx_256_single
  52  * Electrostatics interaction: ReactionField
  53  * VdW interaction:            LennardJones
  54  * Geometry:                   Water3-Water3
  55  * Calculate force/pot:        PotentialAndForce
  56  */
  57 void
  58 nb_kernel_ElecRFCut_VdwLJSh_GeomW3W3_VF_avx_256_single
  59                     (t_nblist                    * gmx_restrict       nlist,
  60                      rvec                        * gmx_restrict          xx,
  61                      rvec                        * gmx_restrict          ff,
  62                      t_forcerec                  * gmx_restrict          fr,
  63                      t_mdatoms                   * gmx_restrict     mdatoms,
  64                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
  65                      t_nrnb                      * gmx_restrict        nrnb)
  66 {
  67     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
  68      * just 0 for non-waters.
  69      * Suffixes A,B,C,D,E,F,G,H refer to j loop unrolling done with AVX, e.g. for the eight different
  70      * jnr indices corresponding to data put in the four positions in the SIMD register.
  71      */
  72     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
  73     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
  74     int              jnrA,jnrB,jnrC,jnrD;
  75     int              jnrE,jnrF,jnrG,jnrH;
  76     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
  77     int              jnrlistE,jnrlistF,jnrlistG,jnrlistH;
  78     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
  79     int              j_coord_offsetE,j_coord_offsetF,j_coord_offsetG,j_coord_offsetH;
  80     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
  81     real             rcutoff_scalar;
  82     real             *shiftvec,*fshift,*x,*f;
  83     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD,*fjptrE,*fjptrF,*fjptrG,*fjptrH;
  84     real             scratch[4*DIM];
  85     __m256           tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
  86     real *           vdwioffsetptr0;
  87     __m256           ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
  88     real *           vdwioffsetptr1;
  89     __m256           ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
  90     real *           vdwioffsetptr2;
  91     __m256           ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
  92     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D,vdwjidx0E,vdwjidx0F,vdwjidx0G,vdwjidx0H;
  93     __m256           jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
  94     int              vdwjidx1A,vdwjidx1B,vdwjidx1C,vdwjidx1D,vdwjidx1E,vdwjidx1F,vdwjidx1G,vdwjidx1H;
  95     __m256           jx1,jy1,jz1,fjx1,fjy1,fjz1,jq1,isaj1;
  96     int              vdwjidx2A,vdwjidx2B,vdwjidx2C,vdwjidx2D,vdwjidx2E,vdwjidx2F,vdwjidx2G,vdwjidx2H;
  97     __m256           jx2,jy2,jz2,fjx2,fjy2,fjz2,jq2,isaj2;
  98     __m256           dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
  99     __m256           dx01,dy01,dz01,rsq01,rinv01,rinvsq01,r01,qq01,c6_01,c12_01;
 100     __m256           dx02,dy02,dz02,rsq02,rinv02,rinvsq02,r02,qq02,c6_02,c12_02;
 101     __m256           dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
 102     __m256           dx11,dy11,dz11,rsq11,rinv11,rinvsq11,r11,qq11,c6_11,c12_11;
 103     __m256           dx12,dy12,dz12,rsq12,rinv12,rinvsq12,r12,qq12,c6_12,c12_12;
 104     __m256           dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
 105     __m256           dx21,dy21,dz21,rsq21,rinv21,rinvsq21,r21,qq21,c6_21,c12_21;
 106     __m256           dx22,dy22,dz22,rsq22,rinv22,rinvsq22,r22,qq22,c6_22,c12_22;
 107     __m256           velec,felec,velecsum,facel,crf,krf,krf2;
 108     real             *charge;
 109     int              nvdwtype;
 110     __m256           rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
 111     int              *vdwtype;
 112     real             *vdwparam;
 113     __m256           one_sixth   = _mm256_set1_ps(1.0/6.0);
 114     __m256           one_twelfth = _mm256_set1_ps(1.0/12.0);
 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     krf              = _mm256_set1_ps(fr->ic->k_rf);
 133     krf2             = _mm256_set1_ps(fr->ic->k_rf*2.0);
 134     crf              = _mm256_set1_ps(fr->ic->c_rf);
 135     nvdwtype         = fr->ntype;
 136     vdwparam         = fr->nbfp;
 137     vdwtype          = mdatoms->typeA;
 138
 139     /* Setup water-specific parameters */
 140     inr              = nlist->iinr[0];
 141     iq0              = _mm256_mul_ps(facel,_mm256_set1_ps(charge[inr+0]));
 142     iq1              = _mm256_mul_ps(facel,_mm256_set1_ps(charge[inr+1]));
 143     iq2              = _mm256_mul_ps(facel,_mm256_set1_ps(charge[inr+2]));
 144     vdwioffsetptr0   = vdwparam+2*nvdwtype*vdwtype[inr+0];
 145
 146     jq0              = _mm256_set1_ps(charge[inr+0]);
 147     jq1              = _mm256_set1_ps(charge[inr+1]);
 148     jq2              = _mm256_set1_ps(charge[inr+2]);
 149     vdwjidx0A        = 2*vdwtype[inr+0];
 150     qq00             = _mm256_mul_ps(iq0,jq0);
 151     c6_00            = _mm256_set1_ps(vdwioffsetptr0[vdwjidx0A]);
 152     c12_00           = _mm256_set1_ps(vdwioffsetptr0[vdwjidx0A+1]);
 153     qq01             = _mm256_mul_ps(iq0,jq1);
 154     qq02             = _mm256_mul_ps(iq0,jq2);
 155     qq10             = _mm256_mul_ps(iq1,jq0);
 156     qq11             = _mm256_mul_ps(iq1,jq1);
 157     qq12             = _mm256_mul_ps(iq1,jq2);
 158     qq20             = _mm256_mul_ps(iq2,jq0);
 159     qq21             = _mm256_mul_ps(iq2,jq1);
 160     qq22             = _mm256_mul_ps(iq2,jq2);
 161
 162     /* When we use explicit cutoffs the value must be identical for elec and VdW, so use elec as an arbitrary choice */
 163     rcutoff_scalar   = fr->rcoulomb;
 164     rcutoff          = _mm256_set1_ps(rcutoff_scalar);
 165     rcutoff2         = _mm256_mul_ps(rcutoff,rcutoff);
 166
 167     sh_vdw_invrcut6  = _mm256_set1_ps(fr->ic->sh_invrc6);
 168     rvdw             = _mm256_set1_ps(fr->rvdw);
 169
 170     /* Avoid stupid compiler warnings */
 171     jnrA = jnrB = jnrC = jnrD = jnrE = jnrF = jnrG = jnrH = 0;
 172     j_coord_offsetA = 0;
 173     j_coord_offsetB = 0;
 174     j_coord_offsetC = 0;
 175     j_coord_offsetD = 0;
 176     j_coord_offsetE = 0;
 177     j_coord_offsetF = 0;
 178     j_coord_offsetG = 0;
 179     j_coord_offsetH = 0;
 180
 181     outeriter        = 0;
 182     inneriter        = 0;
 183
 184     for(iidx=0;iidx<4*DIM;iidx++)
 185     {
 186         scratch[iidx] = 0.0;
 187     }
 188
 189     /* Start outer loop over neighborlists */
 190     for(iidx=0; iidx<nri; iidx++)
 191     {
 192         /* Load shift vector for this list */
 193         i_shift_offset   = DIM*shiftidx[iidx];
 194
 195         /* Load limits for loop over neighbors */
 196         j_index_start    = jindex[iidx];
 197         j_index_end      = jindex[iidx+1];
 198
 199         /* Get outer coordinate index */
 200         inr              = iinr[iidx];
 201         i_coord_offset   = DIM*inr;
 202
 203         /* Load i particle coords and add shift vector */
 204         gmx_mm256_load_shift_and_3rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,
 205                                                     &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2);
 206
 207         fix0             = _mm256_setzero_ps();
 208         fiy0             = _mm256_setzero_ps();
 209         fiz0             = _mm256_setzero_ps();
 210         fix1             = _mm256_setzero_ps();
 211         fiy1             = _mm256_setzero_ps();
 212         fiz1             = _mm256_setzero_ps();
 213         fix2             = _mm256_setzero_ps();
 214         fiy2             = _mm256_setzero_ps();
 215         fiz2             = _mm256_setzero_ps();
 216
 217         /* Reset potential sums */
 218         velecsum         = _mm256_setzero_ps();
 219         vvdwsum          = _mm256_setzero_ps();
 220
 221         /* Start inner kernel loop */
 222         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+7]>=0; jidx+=8)
 223         {
 224
 225             /* Get j neighbor index, and coordinate index */
 226             jnrA             = jjnr[jidx];
 227             jnrB             = jjnr[jidx+1];
 228             jnrC             = jjnr[jidx+2];
 229             jnrD             = jjnr[jidx+3];
 230             jnrE             = jjnr[jidx+4];
 231             jnrF             = jjnr[jidx+5];
 232             jnrG             = jjnr[jidx+6];
 233             jnrH             = jjnr[jidx+7];
 234             j_coord_offsetA  = DIM*jnrA;
 235             j_coord_offsetB  = DIM*jnrB;
 236             j_coord_offsetC  = DIM*jnrC;
 237             j_coord_offsetD  = DIM*jnrD;
 238             j_coord_offsetE  = DIM*jnrE;
 239             j_coord_offsetF  = DIM*jnrF;
 240             j_coord_offsetG  = DIM*jnrG;
 241             j_coord_offsetH  = DIM*jnrH;
 242
 243             /* load j atom coordinates */
 244             gmx_mm256_load_3rvec_8ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 245                                                  x+j_coord_offsetC,x+j_coord_offsetD,
 246                                                  x+j_coord_offsetE,x+j_coord_offsetF,
 247                                                  x+j_coord_offsetG,x+j_coord_offsetH,
 248                                               &jx0,&jy0,&jz0,&jx1,&jy1,&jz1,&jx2,&jy2,&jz2);
 249
 250             /* Calculate displacement vector */
 251             dx00             = _mm256_sub_ps(ix0,jx0);
 252             dy00             = _mm256_sub_ps(iy0,jy0);
 253             dz00             = _mm256_sub_ps(iz0,jz0);
 254             dx01             = _mm256_sub_ps(ix0,jx1);
 255             dy01             = _mm256_sub_ps(iy0,jy1);
 256             dz01             = _mm256_sub_ps(iz0,jz1);
 257             dx02             = _mm256_sub_ps(ix0,jx2);
 258             dy02             = _mm256_sub_ps(iy0,jy2);
 259             dz02             = _mm256_sub_ps(iz0,jz2);
 260             dx10             = _mm256_sub_ps(ix1,jx0);
 261             dy10             = _mm256_sub_ps(iy1,jy0);
 262             dz10             = _mm256_sub_ps(iz1,jz0);
 263             dx11             = _mm256_sub_ps(ix1,jx1);
 264             dy11             = _mm256_sub_ps(iy1,jy1);
 265             dz11             = _mm256_sub_ps(iz1,jz1);
 266             dx12             = _mm256_sub_ps(ix1,jx2);
 267             dy12             = _mm256_sub_ps(iy1,jy2);
 268             dz12             = _mm256_sub_ps(iz1,jz2);
 269             dx20             = _mm256_sub_ps(ix2,jx0);
 270             dy20             = _mm256_sub_ps(iy2,jy0);
 271             dz20             = _mm256_sub_ps(iz2,jz0);
 272             dx21             = _mm256_sub_ps(ix2,jx1);
 273             dy21             = _mm256_sub_ps(iy2,jy1);
 274             dz21             = _mm256_sub_ps(iz2,jz1);
 275             dx22             = _mm256_sub_ps(ix2,jx2);
 276             dy22             = _mm256_sub_ps(iy2,jy2);
 277             dz22             = _mm256_sub_ps(iz2,jz2);
 278
 279             /* Calculate squared distance and things based on it */
 280             rsq00            = gmx_mm256_calc_rsq_ps(dx00,dy00,dz00);
 281             rsq01            = gmx_mm256_calc_rsq_ps(dx01,dy01,dz01);
 282             rsq02            = gmx_mm256_calc_rsq_ps(dx02,dy02,dz02);
 283             rsq10            = gmx_mm256_calc_rsq_ps(dx10,dy10,dz10);
 284             rsq11            = gmx_mm256_calc_rsq_ps(dx11,dy11,dz11);
 285             rsq12            = gmx_mm256_calc_rsq_ps(dx12,dy12,dz12);
 286             rsq20            = gmx_mm256_calc_rsq_ps(dx20,dy20,dz20);
 287             rsq21            = gmx_mm256_calc_rsq_ps(dx21,dy21,dz21);
 288             rsq22            = gmx_mm256_calc_rsq_ps(dx22,dy22,dz22);
 289
 290             rinv00           = gmx_mm256_invsqrt_ps(rsq00);
 291             rinv01           = gmx_mm256_invsqrt_ps(rsq01);
 292             rinv02           = gmx_mm256_invsqrt_ps(rsq02);
 293             rinv10           = gmx_mm256_invsqrt_ps(rsq10);
 294             rinv11           = gmx_mm256_invsqrt_ps(rsq11);
 295             rinv12           = gmx_mm256_invsqrt_ps(rsq12);
 296             rinv20           = gmx_mm256_invsqrt_ps(rsq20);
 297             rinv21           = gmx_mm256_invsqrt_ps(rsq21);
 298             rinv22           = gmx_mm256_invsqrt_ps(rsq22);
 299
 300             rinvsq00         = _mm256_mul_ps(rinv00,rinv00);
 301             rinvsq01         = _mm256_mul_ps(rinv01,rinv01);
 302             rinvsq02         = _mm256_mul_ps(rinv02,rinv02);
 303             rinvsq10         = _mm256_mul_ps(rinv10,rinv10);
 304             rinvsq11         = _mm256_mul_ps(rinv11,rinv11);
 305             rinvsq12         = _mm256_mul_ps(rinv12,rinv12);
 306             rinvsq20         = _mm256_mul_ps(rinv20,rinv20);
 307             rinvsq21         = _mm256_mul_ps(rinv21,rinv21);
 308             rinvsq22         = _mm256_mul_ps(rinv22,rinv22);
 309
 310             fjx0             = _mm256_setzero_ps();
 311             fjy0             = _mm256_setzero_ps();
 312             fjz0             = _mm256_setzero_ps();
 313             fjx1             = _mm256_setzero_ps();
 314             fjy1             = _mm256_setzero_ps();
 315             fjz1             = _mm256_setzero_ps();
 316             fjx2             = _mm256_setzero_ps();
 317             fjy2             = _mm256_setzero_ps();
 318             fjz2             = _mm256_setzero_ps();
 319
 320             /**************************
 321              * CALCULATE INTERACTIONS *
 322              **************************/
 323
 324             if (gmx_mm256_any_lt(rsq00,rcutoff2))
 325             {
 326
 327             /* REACTION-FIELD ELECTROSTATICS */
 328             velec            = _mm256_mul_ps(qq00,_mm256_sub_ps(_mm256_add_ps(rinv00,_mm256_mul_ps(krf,rsq00)),crf));
 329             felec            = _mm256_mul_ps(qq00,_mm256_sub_ps(_mm256_mul_ps(rinv00,rinvsq00),krf2));
 330
 331             /* LENNARD-JONES DISPERSION/REPULSION */
 332
 333             rinvsix          = _mm256_mul_ps(_mm256_mul_ps(rinvsq00,rinvsq00),rinvsq00);
 334             vvdw6            = _mm256_mul_ps(c6_00,rinvsix);
 335             vvdw12           = _mm256_mul_ps(c12_00,_mm256_mul_ps(rinvsix,rinvsix));
 336             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) ,
 337                                           _mm256_mul_ps( _mm256_sub_ps(vvdw6,_mm256_mul_ps(c6_00,sh_vdw_invrcut6)),one_sixth));
 338             fvdw             = _mm256_mul_ps(_mm256_sub_ps(vvdw12,vvdw6),rinvsq00);
 339
 340             cutoff_mask      = _mm256_cmp_ps(rsq00,rcutoff2,_CMP_LT_OQ);
 341
 342             /* Update potential sum for this i atom from the interaction with this j atom. */
 343             velec            = _mm256_and_ps(velec,cutoff_mask);
 344             velecsum         = _mm256_add_ps(velecsum,velec);
 345             vvdw             = _mm256_and_ps(vvdw,cutoff_mask);
 346             vvdwsum          = _mm256_add_ps(vvdwsum,vvdw);
 347
 348             fscal            = _mm256_add_ps(felec,fvdw);
 349
 350             fscal            = _mm256_and_ps(fscal,cutoff_mask);
 351
 352             /* Calculate temporary vectorial force */
 353             tx               = _mm256_mul_ps(fscal,dx00);
 354             ty               = _mm256_mul_ps(fscal,dy00);
 355             tz               = _mm256_mul_ps(fscal,dz00);
 356
 357             /* Update vectorial force */
 358             fix0             = _mm256_add_ps(fix0,tx);
 359             fiy0             = _mm256_add_ps(fiy0,ty);
 360             fiz0             = _mm256_add_ps(fiz0,tz);
 361
 362             fjx0             = _mm256_add_ps(fjx0,tx);
 363             fjy0             = _mm256_add_ps(fjy0,ty);
 364             fjz0             = _mm256_add_ps(fjz0,tz);
 365
 366             }
 367
 368             /**************************
 369              * CALCULATE INTERACTIONS *
 370              **************************/
 371
 372             if (gmx_mm256_any_lt(rsq01,rcutoff2))
 373             {
 374
 375             /* REACTION-FIELD ELECTROSTATICS */
 376             velec            = _mm256_mul_ps(qq01,_mm256_sub_ps(_mm256_add_ps(rinv01,_mm256_mul_ps(krf,rsq01)),crf));
 377             felec            = _mm256_mul_ps(qq01,_mm256_sub_ps(_mm256_mul_ps(rinv01,rinvsq01),krf2));
 378
 379             cutoff_mask      = _mm256_cmp_ps(rsq01,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,dx01);
 391             ty               = _mm256_mul_ps(fscal,dy01);
 392             tz               = _mm256_mul_ps(fscal,dz01);
 393
 394             /* Update vectorial force */
 395             fix0             = _mm256_add_ps(fix0,tx);
 396             fiy0             = _mm256_add_ps(fiy0,ty);
 397             fiz0             = _mm256_add_ps(fiz0,tz);
 398
 399             fjx1             = _mm256_add_ps(fjx1,tx);
 400             fjy1             = _mm256_add_ps(fjy1,ty);
 401             fjz1             = _mm256_add_ps(fjz1,tz);
 402
 403             }
 404
 405             /**************************
 406              * CALCULATE INTERACTIONS *
 407              **************************/
 408
 409             if (gmx_mm256_any_lt(rsq02,rcutoff2))
 410             {
 411
 412             /* REACTION-FIELD ELECTROSTATICS */
 413             velec            = _mm256_mul_ps(qq02,_mm256_sub_ps(_mm256_add_ps(rinv02,_mm256_mul_ps(krf,rsq02)),crf));
 414             felec            = _mm256_mul_ps(qq02,_mm256_sub_ps(_mm256_mul_ps(rinv02,rinvsq02),krf2));
 415
 416             cutoff_mask      = _mm256_cmp_ps(rsq02,rcutoff2,_CMP_LT_OQ);
 417
 418             /* Update potential sum for this i atom from the interaction with this j atom. */
 419             velec            = _mm256_and_ps(velec,cutoff_mask);
 420             velecsum         = _mm256_add_ps(velecsum,velec);
 421
 422             fscal            = felec;
 423
 424             fscal            = _mm256_and_ps(fscal,cutoff_mask);
 425
 426             /* Calculate temporary vectorial force */
 427             tx               = _mm256_mul_ps(fscal,dx02);
 428             ty               = _mm256_mul_ps(fscal,dy02);
 429             tz               = _mm256_mul_ps(fscal,dz02);
 430
 431             /* Update vectorial force */
 432             fix0             = _mm256_add_ps(fix0,tx);
 433             fiy0             = _mm256_add_ps(fiy0,ty);
 434             fiz0             = _mm256_add_ps(fiz0,tz);
 435
 436             fjx2             = _mm256_add_ps(fjx2,tx);
 437             fjy2             = _mm256_add_ps(fjy2,ty);
 438             fjz2             = _mm256_add_ps(fjz2,tz);
 439
 440             }
 441
 442             /**************************
 443              * CALCULATE INTERACTIONS *
 444              **************************/
 445
 446             if (gmx_mm256_any_lt(rsq10,rcutoff2))
 447             {
 448
 449             /* REACTION-FIELD ELECTROSTATICS */
 450             velec            = _mm256_mul_ps(qq10,_mm256_sub_ps(_mm256_add_ps(rinv10,_mm256_mul_ps(krf,rsq10)),crf));
 451             felec            = _mm256_mul_ps(qq10,_mm256_sub_ps(_mm256_mul_ps(rinv10,rinvsq10),krf2));
 452
 453             cutoff_mask      = _mm256_cmp_ps(rsq10,rcutoff2,_CMP_LT_OQ);
 454
 455             /* Update potential sum for this i atom from the interaction with this j atom. */
 456             velec            = _mm256_and_ps(velec,cutoff_mask);
 457             velecsum         = _mm256_add_ps(velecsum,velec);
 458
 459             fscal            = felec;
 460
 461             fscal            = _mm256_and_ps(fscal,cutoff_mask);
 462
 463             /* Calculate temporary vectorial force */
 464             tx               = _mm256_mul_ps(fscal,dx10);
 465             ty               = _mm256_mul_ps(fscal,dy10);
 466             tz               = _mm256_mul_ps(fscal,dz10);
 467
 468             /* Update vectorial force */
 469             fix1             = _mm256_add_ps(fix1,tx);
 470             fiy1             = _mm256_add_ps(fiy1,ty);
 471             fiz1             = _mm256_add_ps(fiz1,tz);
 472
 473             fjx0             = _mm256_add_ps(fjx0,tx);
 474             fjy0             = _mm256_add_ps(fjy0,ty);
 475             fjz0             = _mm256_add_ps(fjz0,tz);
 476
 477             }
 478
 479             /**************************
 480              * CALCULATE INTERACTIONS *
 481              **************************/
 482
 483             if (gmx_mm256_any_lt(rsq11,rcutoff2))
 484             {
 485
 486             /* REACTION-FIELD ELECTROSTATICS */
 487             velec            = _mm256_mul_ps(qq11,_mm256_sub_ps(_mm256_add_ps(rinv11,_mm256_mul_ps(krf,rsq11)),crf));
 488             felec            = _mm256_mul_ps(qq11,_mm256_sub_ps(_mm256_mul_ps(rinv11,rinvsq11),krf2));
 489
 490             cutoff_mask      = _mm256_cmp_ps(rsq11,rcutoff2,_CMP_LT_OQ);
 491
 492             /* Update potential sum for this i atom from the interaction with this j atom. */
 493             velec            = _mm256_and_ps(velec,cutoff_mask);
 494             velecsum         = _mm256_add_ps(velecsum,velec);
 495
 496             fscal            = felec;
 497
 498             fscal            = _mm256_and_ps(fscal,cutoff_mask);
 499
 500             /* Calculate temporary vectorial force */
 501             tx               = _mm256_mul_ps(fscal,dx11);
 502             ty               = _mm256_mul_ps(fscal,dy11);
 503             tz               = _mm256_mul_ps(fscal,dz11);
 504
 505             /* Update vectorial force */
 506             fix1             = _mm256_add_ps(fix1,tx);
 507             fiy1             = _mm256_add_ps(fiy1,ty);
 508             fiz1             = _mm256_add_ps(fiz1,tz);
 509
 510             fjx1             = _mm256_add_ps(fjx1,tx);
 511             fjy1             = _mm256_add_ps(fjy1,ty);
 512             fjz1             = _mm256_add_ps(fjz1,tz);
 513
 514             }
 515
 516             /**************************
 517              * CALCULATE INTERACTIONS *
 518              **************************/
 519
 520             if (gmx_mm256_any_lt(rsq12,rcutoff2))
 521             {
 522
 523             /* REACTION-FIELD ELECTROSTATICS */
 524             velec            = _mm256_mul_ps(qq12,_mm256_sub_ps(_mm256_add_ps(rinv12,_mm256_mul_ps(krf,rsq12)),crf));
 525             felec            = _mm256_mul_ps(qq12,_mm256_sub_ps(_mm256_mul_ps(rinv12,rinvsq12),krf2));
 526
 527             cutoff_mask      = _mm256_cmp_ps(rsq12,rcutoff2,_CMP_LT_OQ);
 528
 529             /* Update potential sum for this i atom from the interaction with this j atom. */
 530             velec            = _mm256_and_ps(velec,cutoff_mask);
 531             velecsum         = _mm256_add_ps(velecsum,velec);
 532
 533             fscal            = felec;
 534
 535             fscal            = _mm256_and_ps(fscal,cutoff_mask);
 536
 537             /* Calculate temporary vectorial force */
 538             tx               = _mm256_mul_ps(fscal,dx12);
 539             ty               = _mm256_mul_ps(fscal,dy12);
 540             tz               = _mm256_mul_ps(fscal,dz12);
 541
 542             /* Update vectorial force */
 543             fix1             = _mm256_add_ps(fix1,tx);
 544             fiy1             = _mm256_add_ps(fiy1,ty);
 545             fiz1             = _mm256_add_ps(fiz1,tz);
 546
 547             fjx2             = _mm256_add_ps(fjx2,tx);
 548             fjy2             = _mm256_add_ps(fjy2,ty);
 549             fjz2             = _mm256_add_ps(fjz2,tz);
 550
 551             }
 552
 553             /**************************
 554              * CALCULATE INTERACTIONS *
 555              **************************/
 556
 557             if (gmx_mm256_any_lt(rsq20,rcutoff2))
 558             {
 559
 560             /* REACTION-FIELD ELECTROSTATICS */
 561             velec            = _mm256_mul_ps(qq20,_mm256_sub_ps(_mm256_add_ps(rinv20,_mm256_mul_ps(krf,rsq20)),crf));
 562             felec            = _mm256_mul_ps(qq20,_mm256_sub_ps(_mm256_mul_ps(rinv20,rinvsq20),krf2));
 563
 564             cutoff_mask      = _mm256_cmp_ps(rsq20,rcutoff2,_CMP_LT_OQ);
 565
 566             /* Update potential sum for this i atom from the interaction with this j atom. */
 567             velec            = _mm256_and_ps(velec,cutoff_mask);
 568             velecsum         = _mm256_add_ps(velecsum,velec);
 569
 570             fscal            = felec;
 571
 572             fscal            = _mm256_and_ps(fscal,cutoff_mask);
 573
 574             /* Calculate temporary vectorial force */
 575             tx               = _mm256_mul_ps(fscal,dx20);
 576             ty               = _mm256_mul_ps(fscal,dy20);
 577             tz               = _mm256_mul_ps(fscal,dz20);
 578
 579             /* Update vectorial force */
 580             fix2             = _mm256_add_ps(fix2,tx);
 581             fiy2             = _mm256_add_ps(fiy2,ty);
 582             fiz2             = _mm256_add_ps(fiz2,tz);
 583
 584             fjx0             = _mm256_add_ps(fjx0,tx);
 585             fjy0             = _mm256_add_ps(fjy0,ty);
 586             fjz0             = _mm256_add_ps(fjz0,tz);
 587
 588             }
 589
 590             /**************************
 591              * CALCULATE INTERACTIONS *
 592              **************************/
 593
 594             if (gmx_mm256_any_lt(rsq21,rcutoff2))
 595             {
 596
 597             /* REACTION-FIELD ELECTROSTATICS */
 598             velec            = _mm256_mul_ps(qq21,_mm256_sub_ps(_mm256_add_ps(rinv21,_mm256_mul_ps(krf,rsq21)),crf));
 599             felec            = _mm256_mul_ps(qq21,_mm256_sub_ps(_mm256_mul_ps(rinv21,rinvsq21),krf2));
 600
 601             cutoff_mask      = _mm256_cmp_ps(rsq21,rcutoff2,_CMP_LT_OQ);
 602
 603             /* Update potential sum for this i atom from the interaction with this j atom. */
 604             velec            = _mm256_and_ps(velec,cutoff_mask);
 605             velecsum         = _mm256_add_ps(velecsum,velec);
 606
 607             fscal            = felec;
 608
 609             fscal            = _mm256_and_ps(fscal,cutoff_mask);
 610
 611             /* Calculate temporary vectorial force */
 612             tx               = _mm256_mul_ps(fscal,dx21);
 613             ty               = _mm256_mul_ps(fscal,dy21);
 614             tz               = _mm256_mul_ps(fscal,dz21);
 615
 616             /* Update vectorial force */
 617             fix2             = _mm256_add_ps(fix2,tx);
 618             fiy2             = _mm256_add_ps(fiy2,ty);
 619             fiz2             = _mm256_add_ps(fiz2,tz);
 620
 621             fjx1             = _mm256_add_ps(fjx1,tx);
 622             fjy1             = _mm256_add_ps(fjy1,ty);
 623             fjz1             = _mm256_add_ps(fjz1,tz);
 624
 625             }
 626
 627             /**************************
 628              * CALCULATE INTERACTIONS *
 629              **************************/
 630
 631             if (gmx_mm256_any_lt(rsq22,rcutoff2))
 632             {
 633
 634             /* REACTION-FIELD ELECTROSTATICS */
 635             velec            = _mm256_mul_ps(qq22,_mm256_sub_ps(_mm256_add_ps(rinv22,_mm256_mul_ps(krf,rsq22)),crf));
 636             felec            = _mm256_mul_ps(qq22,_mm256_sub_ps(_mm256_mul_ps(rinv22,rinvsq22),krf2));
 637
 638             cutoff_mask      = _mm256_cmp_ps(rsq22,rcutoff2,_CMP_LT_OQ);
 639
 640             /* Update potential sum for this i atom from the interaction with this j atom. */
 641             velec            = _mm256_and_ps(velec,cutoff_mask);
 642             velecsum         = _mm256_add_ps(velecsum,velec);
 643
 644             fscal            = felec;
 645
 646             fscal            = _mm256_and_ps(fscal,cutoff_mask);
 647
 648             /* Calculate temporary vectorial force */
 649             tx               = _mm256_mul_ps(fscal,dx22);
 650             ty               = _mm256_mul_ps(fscal,dy22);
 651             tz               = _mm256_mul_ps(fscal,dz22);
 652
 653             /* Update vectorial force */
 654             fix2             = _mm256_add_ps(fix2,tx);
 655             fiy2             = _mm256_add_ps(fiy2,ty);
 656             fiz2             = _mm256_add_ps(fiz2,tz);
 657
 658             fjx2             = _mm256_add_ps(fjx2,tx);
 659             fjy2             = _mm256_add_ps(fjy2,ty);
 660             fjz2             = _mm256_add_ps(fjz2,tz);
 661
 662             }
 663
 664             fjptrA             = f+j_coord_offsetA;
 665             fjptrB             = f+j_coord_offsetB;
 666             fjptrC             = f+j_coord_offsetC;
 667             fjptrD             = f+j_coord_offsetD;
 668             fjptrE             = f+j_coord_offsetE;
 669             fjptrF             = f+j_coord_offsetF;
 670             fjptrG             = f+j_coord_offsetG;
 671             fjptrH             = f+j_coord_offsetH;
 672
 673             gmx_mm256_decrement_3rvec_8ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjptrE,fjptrF,fjptrG,fjptrH,
 674                                                       fjx0,fjy0,fjz0,fjx1,fjy1,fjz1,fjx2,fjy2,fjz2);
 675
 676             /* Inner loop uses 342 flops */
 677         }
 678
 679         if(jidx<j_index_end)
 680         {
 681
 682             /* Get j neighbor index, and coordinate index */
 683             jnrlistA         = jjnr[jidx];
 684             jnrlistB         = jjnr[jidx+1];
 685             jnrlistC         = jjnr[jidx+2];
 686             jnrlistD         = jjnr[jidx+3];
 687             jnrlistE         = jjnr[jidx+4];
 688             jnrlistF         = jjnr[jidx+5];
 689             jnrlistG         = jjnr[jidx+6];
 690             jnrlistH         = jjnr[jidx+7];
 691             /* Sign of each element will be negative for non-real atoms.
 692              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 693              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
 694              */
 695             dummy_mask = gmx_mm256_set_m128(gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx+4)),_mm_setzero_si128())),
 696                                             gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128())));
 697
 698             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
 699             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
 700             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
 701             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
 702             jnrE       = (jnrlistE>=0) ? jnrlistE : 0;
 703             jnrF       = (jnrlistF>=0) ? jnrlistF : 0;
 704             jnrG       = (jnrlistG>=0) ? jnrlistG : 0;
 705             jnrH       = (jnrlistH>=0) ? jnrlistH : 0;
 706             j_coord_offsetA  = DIM*jnrA;
 707             j_coord_offsetB  = DIM*jnrB;
 708             j_coord_offsetC  = DIM*jnrC;
 709             j_coord_offsetD  = DIM*jnrD;
 710             j_coord_offsetE  = DIM*jnrE;
 711             j_coord_offsetF  = DIM*jnrF;
 712             j_coord_offsetG  = DIM*jnrG;
 713             j_coord_offsetH  = DIM*jnrH;
 714
 715             /* load j atom coordinates */
 716             gmx_mm256_load_3rvec_8ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 717                                                  x+j_coord_offsetC,x+j_coord_offsetD,
 718                                                  x+j_coord_offsetE,x+j_coord_offsetF,
 719                                                  x+j_coord_offsetG,x+j_coord_offsetH,
 720                                               &jx0,&jy0,&jz0,&jx1,&jy1,&jz1,&jx2,&jy2,&jz2);
 721
 722             /* Calculate displacement vector */
 723             dx00             = _mm256_sub_ps(ix0,jx0);
 724             dy00             = _mm256_sub_ps(iy0,jy0);
 725             dz00             = _mm256_sub_ps(iz0,jz0);
 726             dx01             = _mm256_sub_ps(ix0,jx1);
 727             dy01             = _mm256_sub_ps(iy0,jy1);
 728             dz01             = _mm256_sub_ps(iz0,jz1);
 729             dx02             = _mm256_sub_ps(ix0,jx2);
 730             dy02             = _mm256_sub_ps(iy0,jy2);
 731             dz02             = _mm256_sub_ps(iz0,jz2);
 732             dx10             = _mm256_sub_ps(ix1,jx0);
 733             dy10             = _mm256_sub_ps(iy1,jy0);
 734             dz10             = _mm256_sub_ps(iz1,jz0);
 735             dx11             = _mm256_sub_ps(ix1,jx1);
 736             dy11             = _mm256_sub_ps(iy1,jy1);
 737             dz11             = _mm256_sub_ps(iz1,jz1);
 738             dx12             = _mm256_sub_ps(ix1,jx2);
 739             dy12             = _mm256_sub_ps(iy1,jy2);
 740             dz12             = _mm256_sub_ps(iz1,jz2);
 741             dx20             = _mm256_sub_ps(ix2,jx0);
 742             dy20             = _mm256_sub_ps(iy2,jy0);
 743             dz20             = _mm256_sub_ps(iz2,jz0);
 744             dx21             = _mm256_sub_ps(ix2,jx1);
 745             dy21             = _mm256_sub_ps(iy2,jy1);
 746             dz21             = _mm256_sub_ps(iz2,jz1);
 747             dx22             = _mm256_sub_ps(ix2,jx2);
 748             dy22             = _mm256_sub_ps(iy2,jy2);
 749             dz22             = _mm256_sub_ps(iz2,jz2);
 750
 751             /* Calculate squared distance and things based on it */
 752             rsq00            = gmx_mm256_calc_rsq_ps(dx00,dy00,dz00);
 753             rsq01            = gmx_mm256_calc_rsq_ps(dx01,dy01,dz01);
 754             rsq02            = gmx_mm256_calc_rsq_ps(dx02,dy02,dz02);
 755             rsq10            = gmx_mm256_calc_rsq_ps(dx10,dy10,dz10);
 756             rsq11            = gmx_mm256_calc_rsq_ps(dx11,dy11,dz11);
 757             rsq12            = gmx_mm256_calc_rsq_ps(dx12,dy12,dz12);
 758             rsq20            = gmx_mm256_calc_rsq_ps(dx20,dy20,dz20);
 759             rsq21            = gmx_mm256_calc_rsq_ps(dx21,dy21,dz21);
 760             rsq22            = gmx_mm256_calc_rsq_ps(dx22,dy22,dz22);
 761
 762             rinv00           = gmx_mm256_invsqrt_ps(rsq00);
 763             rinv01           = gmx_mm256_invsqrt_ps(rsq01);
 764             rinv02           = gmx_mm256_invsqrt_ps(rsq02);
 765             rinv10           = gmx_mm256_invsqrt_ps(rsq10);
 766             rinv11           = gmx_mm256_invsqrt_ps(rsq11);
 767             rinv12           = gmx_mm256_invsqrt_ps(rsq12);
 768             rinv20           = gmx_mm256_invsqrt_ps(rsq20);
 769             rinv21           = gmx_mm256_invsqrt_ps(rsq21);
 770             rinv22           = gmx_mm256_invsqrt_ps(rsq22);
 771
 772             rinvsq00         = _mm256_mul_ps(rinv00,rinv00);
 773             rinvsq01         = _mm256_mul_ps(rinv01,rinv01);
 774             rinvsq02         = _mm256_mul_ps(rinv02,rinv02);
 775             rinvsq10         = _mm256_mul_ps(rinv10,rinv10);
 776             rinvsq11         = _mm256_mul_ps(rinv11,rinv11);
 777             rinvsq12         = _mm256_mul_ps(rinv12,rinv12);
 778             rinvsq20         = _mm256_mul_ps(rinv20,rinv20);
 779             rinvsq21         = _mm256_mul_ps(rinv21,rinv21);
 780             rinvsq22         = _mm256_mul_ps(rinv22,rinv22);
 781
 782             fjx0             = _mm256_setzero_ps();
 783             fjy0             = _mm256_setzero_ps();
 784             fjz0             = _mm256_setzero_ps();
 785             fjx1             = _mm256_setzero_ps();
 786             fjy1             = _mm256_setzero_ps();
 787             fjz1             = _mm256_setzero_ps();
 788             fjx2             = _mm256_setzero_ps();
 789             fjy2             = _mm256_setzero_ps();
 790             fjz2             = _mm256_setzero_ps();
 791
 792             /**************************
 793              * CALCULATE INTERACTIONS *
 794              **************************/
 795
 796             if (gmx_mm256_any_lt(rsq00,rcutoff2))
 797             {
 798
 799             /* REACTION-FIELD ELECTROSTATICS */
 800             velec            = _mm256_mul_ps(qq00,_mm256_sub_ps(_mm256_add_ps(rinv00,_mm256_mul_ps(krf,rsq00)),crf));
 801             felec            = _mm256_mul_ps(qq00,_mm256_sub_ps(_mm256_mul_ps(rinv00,rinvsq00),krf2));
 802
 803             /* LENNARD-JONES DISPERSION/REPULSION */
 804
 805             rinvsix          = _mm256_mul_ps(_mm256_mul_ps(rinvsq00,rinvsq00),rinvsq00);
 806             vvdw6            = _mm256_mul_ps(c6_00,rinvsix);
 807             vvdw12           = _mm256_mul_ps(c12_00,_mm256_mul_ps(rinvsix,rinvsix));
 808             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) ,
 809                                           _mm256_mul_ps( _mm256_sub_ps(vvdw6,_mm256_mul_ps(c6_00,sh_vdw_invrcut6)),one_sixth));
 810             fvdw             = _mm256_mul_ps(_mm256_sub_ps(vvdw12,vvdw6),rinvsq00);
 811
 812             cutoff_mask      = _mm256_cmp_ps(rsq00,rcutoff2,_CMP_LT_OQ);
 813
 814             /* Update potential sum for this i atom from the interaction with this j atom. */
 815             velec            = _mm256_and_ps(velec,cutoff_mask);
 816             velec            = _mm256_andnot_ps(dummy_mask,velec);
 817             velecsum         = _mm256_add_ps(velecsum,velec);
 818             vvdw             = _mm256_and_ps(vvdw,cutoff_mask);
 819             vvdw             = _mm256_andnot_ps(dummy_mask,vvdw);
 820             vvdwsum          = _mm256_add_ps(vvdwsum,vvdw);
 821
 822             fscal            = _mm256_add_ps(felec,fvdw);
 823
 824             fscal            = _mm256_and_ps(fscal,cutoff_mask);
 825
 826             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
 827
 828             /* Calculate temporary vectorial force */
 829             tx               = _mm256_mul_ps(fscal,dx00);
 830             ty               = _mm256_mul_ps(fscal,dy00);
 831             tz               = _mm256_mul_ps(fscal,dz00);
 832
 833             /* Update vectorial force */
 834             fix0             = _mm256_add_ps(fix0,tx);
 835             fiy0             = _mm256_add_ps(fiy0,ty);
 836             fiz0             = _mm256_add_ps(fiz0,tz);
 837
 838             fjx0             = _mm256_add_ps(fjx0,tx);
 839             fjy0             = _mm256_add_ps(fjy0,ty);
 840             fjz0             = _mm256_add_ps(fjz0,tz);
 841
 842             }
 843
 844             /**************************
 845              * CALCULATE INTERACTIONS *
 846              **************************/
 847
 848             if (gmx_mm256_any_lt(rsq01,rcutoff2))
 849             {
 850
 851             /* REACTION-FIELD ELECTROSTATICS */
 852             velec            = _mm256_mul_ps(qq01,_mm256_sub_ps(_mm256_add_ps(rinv01,_mm256_mul_ps(krf,rsq01)),crf));
 853             felec            = _mm256_mul_ps(qq01,_mm256_sub_ps(_mm256_mul_ps(rinv01,rinvsq01),krf2));
 854
 855             cutoff_mask      = _mm256_cmp_ps(rsq01,rcutoff2,_CMP_LT_OQ);
 856
 857             /* Update potential sum for this i atom from the interaction with this j atom. */
 858             velec            = _mm256_and_ps(velec,cutoff_mask);
 859             velec            = _mm256_andnot_ps(dummy_mask,velec);
 860             velecsum         = _mm256_add_ps(velecsum,velec);
 861
 862             fscal            = felec;
 863
 864             fscal            = _mm256_and_ps(fscal,cutoff_mask);
 865
 866             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
 867
 868             /* Calculate temporary vectorial force */
 869             tx               = _mm256_mul_ps(fscal,dx01);
 870             ty               = _mm256_mul_ps(fscal,dy01);
 871             tz               = _mm256_mul_ps(fscal,dz01);
 872
 873             /* Update vectorial force */
 874             fix0             = _mm256_add_ps(fix0,tx);
 875             fiy0             = _mm256_add_ps(fiy0,ty);
 876             fiz0             = _mm256_add_ps(fiz0,tz);
 877
 878             fjx1             = _mm256_add_ps(fjx1,tx);
 879             fjy1             = _mm256_add_ps(fjy1,ty);
 880             fjz1             = _mm256_add_ps(fjz1,tz);
 881
 882             }
 883
 884             /**************************
 885              * CALCULATE INTERACTIONS *
 886              **************************/
 887
 888             if (gmx_mm256_any_lt(rsq02,rcutoff2))
 889             {
 890
 891             /* REACTION-FIELD ELECTROSTATICS */
 892             velec            = _mm256_mul_ps(qq02,_mm256_sub_ps(_mm256_add_ps(rinv02,_mm256_mul_ps(krf,rsq02)),crf));
 893             felec            = _mm256_mul_ps(qq02,_mm256_sub_ps(_mm256_mul_ps(rinv02,rinvsq02),krf2));
 894
 895             cutoff_mask      = _mm256_cmp_ps(rsq02,rcutoff2,_CMP_LT_OQ);
 896
 897             /* Update potential sum for this i atom from the interaction with this j atom. */
 898             velec            = _mm256_and_ps(velec,cutoff_mask);
 899             velec            = _mm256_andnot_ps(dummy_mask,velec);
 900             velecsum         = _mm256_add_ps(velecsum,velec);
 901
 902             fscal            = felec;
 903
 904             fscal            = _mm256_and_ps(fscal,cutoff_mask);
 905
 906             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
 907
 908             /* Calculate temporary vectorial force */
 909             tx               = _mm256_mul_ps(fscal,dx02);
 910             ty               = _mm256_mul_ps(fscal,dy02);
 911             tz               = _mm256_mul_ps(fscal,dz02);
 912
 913             /* Update vectorial force */
 914             fix0             = _mm256_add_ps(fix0,tx);
 915             fiy0             = _mm256_add_ps(fiy0,ty);
 916             fiz0             = _mm256_add_ps(fiz0,tz);
 917
 918             fjx2             = _mm256_add_ps(fjx2,tx);
 919             fjy2             = _mm256_add_ps(fjy2,ty);
 920             fjz2             = _mm256_add_ps(fjz2,tz);
 921
 922             }
 923
 924             /**************************
 925              * CALCULATE INTERACTIONS *
 926              **************************/
 927
 928             if (gmx_mm256_any_lt(rsq10,rcutoff2))
 929             {
 930
 931             /* REACTION-FIELD ELECTROSTATICS */
 932             velec            = _mm256_mul_ps(qq10,_mm256_sub_ps(_mm256_add_ps(rinv10,_mm256_mul_ps(krf,rsq10)),crf));
 933             felec            = _mm256_mul_ps(qq10,_mm256_sub_ps(_mm256_mul_ps(rinv10,rinvsq10),krf2));
 934
 935             cutoff_mask      = _mm256_cmp_ps(rsq10,rcutoff2,_CMP_LT_OQ);
 936
 937             /* Update potential sum for this i atom from the interaction with this j atom. */
 938             velec            = _mm256_and_ps(velec,cutoff_mask);
 939             velec            = _mm256_andnot_ps(dummy_mask,velec);
 940             velecsum         = _mm256_add_ps(velecsum,velec);
 941
 942             fscal            = felec;
 943
 944             fscal            = _mm256_and_ps(fscal,cutoff_mask);
 945
 946             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
 947
 948             /* Calculate temporary vectorial force */
 949             tx               = _mm256_mul_ps(fscal,dx10);
 950             ty               = _mm256_mul_ps(fscal,dy10);
 951             tz               = _mm256_mul_ps(fscal,dz10);
 952
 953             /* Update vectorial force */
 954             fix1             = _mm256_add_ps(fix1,tx);
 955             fiy1             = _mm256_add_ps(fiy1,ty);
 956             fiz1             = _mm256_add_ps(fiz1,tz);
 957
 958             fjx0             = _mm256_add_ps(fjx0,tx);
 959             fjy0             = _mm256_add_ps(fjy0,ty);
 960             fjz0             = _mm256_add_ps(fjz0,tz);
 961
 962             }
 963
 964             /**************************
 965              * CALCULATE INTERACTIONS *
 966              **************************/
 967
 968             if (gmx_mm256_any_lt(rsq11,rcutoff2))
 969             {
 970
 971             /* REACTION-FIELD ELECTROSTATICS */
 972             velec            = _mm256_mul_ps(qq11,_mm256_sub_ps(_mm256_add_ps(rinv11,_mm256_mul_ps(krf,rsq11)),crf));
 973             felec            = _mm256_mul_ps(qq11,_mm256_sub_ps(_mm256_mul_ps(rinv11,rinvsq11),krf2));
 974
 975             cutoff_mask      = _mm256_cmp_ps(rsq11,rcutoff2,_CMP_LT_OQ);
 976
 977             /* Update potential sum for this i atom from the interaction with this j atom. */
 978             velec            = _mm256_and_ps(velec,cutoff_mask);
 979             velec            = _mm256_andnot_ps(dummy_mask,velec);
 980             velecsum         = _mm256_add_ps(velecsum,velec);
 981
 982             fscal            = felec;
 983
 984             fscal            = _mm256_and_ps(fscal,cutoff_mask);
 985
 986             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
 987
 988             /* Calculate temporary vectorial force */
 989             tx               = _mm256_mul_ps(fscal,dx11);
 990             ty               = _mm256_mul_ps(fscal,dy11);
 991             tz               = _mm256_mul_ps(fscal,dz11);
 992
 993             /* Update vectorial force */
 994             fix1             = _mm256_add_ps(fix1,tx);
 995             fiy1             = _mm256_add_ps(fiy1,ty);
 996             fiz1             = _mm256_add_ps(fiz1,tz);
 997
 998             fjx1             = _mm256_add_ps(fjx1,tx);
 999             fjy1             = _mm256_add_ps(fjy1,ty);
1000             fjz1             = _mm256_add_ps(fjz1,tz);
1001
1002             }
1003
1004             /**************************
1005              * CALCULATE INTERACTIONS *
1006              **************************/
1007
1008             if (gmx_mm256_any_lt(rsq12,rcutoff2))
1009             {
1010
1011             /* REACTION-FIELD ELECTROSTATICS */
1012             velec            = _mm256_mul_ps(qq12,_mm256_sub_ps(_mm256_add_ps(rinv12,_mm256_mul_ps(krf,rsq12)),crf));
1013             felec            = _mm256_mul_ps(qq12,_mm256_sub_ps(_mm256_mul_ps(rinv12,rinvsq12),krf2));
1014
1015             cutoff_mask      = _mm256_cmp_ps(rsq12,rcutoff2,_CMP_LT_OQ);
1016
1017             /* Update potential sum for this i atom from the interaction with this j atom. */
1018             velec            = _mm256_and_ps(velec,cutoff_mask);
1019             velec            = _mm256_andnot_ps(dummy_mask,velec);
1020             velecsum         = _mm256_add_ps(velecsum,velec);
1021
1022             fscal            = felec;
1023
1024             fscal            = _mm256_and_ps(fscal,cutoff_mask);
1025
1026             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
1027
1028             /* Calculate temporary vectorial force */
1029             tx               = _mm256_mul_ps(fscal,dx12);
1030             ty               = _mm256_mul_ps(fscal,dy12);
1031             tz               = _mm256_mul_ps(fscal,dz12);
1032
1033             /* Update vectorial force */
1034             fix1             = _mm256_add_ps(fix1,tx);
1035             fiy1             = _mm256_add_ps(fiy1,ty);
1036             fiz1             = _mm256_add_ps(fiz1,tz);
1037
1038             fjx2             = _mm256_add_ps(fjx2,tx);
1039             fjy2             = _mm256_add_ps(fjy2,ty);
1040             fjz2             = _mm256_add_ps(fjz2,tz);
1041
1042             }
1043
1044             /**************************
1045              * CALCULATE INTERACTIONS *
1046              **************************/
1047
1048             if (gmx_mm256_any_lt(rsq20,rcutoff2))
1049             {
1050
1051             /* REACTION-FIELD ELECTROSTATICS */
1052             velec            = _mm256_mul_ps(qq20,_mm256_sub_ps(_mm256_add_ps(rinv20,_mm256_mul_ps(krf,rsq20)),crf));
1053             felec            = _mm256_mul_ps(qq20,_mm256_sub_ps(_mm256_mul_ps(rinv20,rinvsq20),krf2));
1054
1055             cutoff_mask      = _mm256_cmp_ps(rsq20,rcutoff2,_CMP_LT_OQ);
1056
1057             /* Update potential sum for this i atom from the interaction with this j atom. */
1058             velec            = _mm256_and_ps(velec,cutoff_mask);
1059             velec            = _mm256_andnot_ps(dummy_mask,velec);
1060             velecsum         = _mm256_add_ps(velecsum,velec);
1061
1062             fscal            = felec;
1063
1064             fscal            = _mm256_and_ps(fscal,cutoff_mask);
1065
1066             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
1067
1068             /* Calculate temporary vectorial force */
1069             tx               = _mm256_mul_ps(fscal,dx20);
1070             ty               = _mm256_mul_ps(fscal,dy20);
1071             tz               = _mm256_mul_ps(fscal,dz20);
1072
1073             /* Update vectorial force */
1074             fix2             = _mm256_add_ps(fix2,tx);
1075             fiy2             = _mm256_add_ps(fiy2,ty);
1076             fiz2             = _mm256_add_ps(fiz2,tz);
1077
1078             fjx0             = _mm256_add_ps(fjx0,tx);
1079             fjy0             = _mm256_add_ps(fjy0,ty);
1080             fjz0             = _mm256_add_ps(fjz0,tz);
1081
1082             }
1083
1084             /**************************
1085              * CALCULATE INTERACTIONS *
1086              **************************/
1087
1088             if (gmx_mm256_any_lt(rsq21,rcutoff2))
1089             {
1090
1091             /* REACTION-FIELD ELECTROSTATICS */
1092             velec            = _mm256_mul_ps(qq21,_mm256_sub_ps(_mm256_add_ps(rinv21,_mm256_mul_ps(krf,rsq21)),crf));
1093             felec            = _mm256_mul_ps(qq21,_mm256_sub_ps(_mm256_mul_ps(rinv21,rinvsq21),krf2));
1094
1095             cutoff_mask      = _mm256_cmp_ps(rsq21,rcutoff2,_CMP_LT_OQ);
1096
1097             /* Update potential sum for this i atom from the interaction with this j atom. */
1098             velec            = _mm256_and_ps(velec,cutoff_mask);
1099             velec            = _mm256_andnot_ps(dummy_mask,velec);
1100             velecsum         = _mm256_add_ps(velecsum,velec);
1101
1102             fscal            = felec;
1103
1104             fscal            = _mm256_and_ps(fscal,cutoff_mask);
1105
1106             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
1107
1108             /* Calculate temporary vectorial force */
1109             tx               = _mm256_mul_ps(fscal,dx21);
1110             ty               = _mm256_mul_ps(fscal,dy21);
1111             tz               = _mm256_mul_ps(fscal,dz21);
1112
1113             /* Update vectorial force */
1114             fix2             = _mm256_add_ps(fix2,tx);
1115             fiy2             = _mm256_add_ps(fiy2,ty);
1116             fiz2             = _mm256_add_ps(fiz2,tz);
1117
1118             fjx1             = _mm256_add_ps(fjx1,tx);
1119             fjy1             = _mm256_add_ps(fjy1,ty);
1120             fjz1             = _mm256_add_ps(fjz1,tz);
1121
1122             }
1123
1124             /**************************
1125              * CALCULATE INTERACTIONS *
1126              **************************/
1127
1128             if (gmx_mm256_any_lt(rsq22,rcutoff2))
1129             {
1130
1131             /* REACTION-FIELD ELECTROSTATICS */
1132             velec            = _mm256_mul_ps(qq22,_mm256_sub_ps(_mm256_add_ps(rinv22,_mm256_mul_ps(krf,rsq22)),crf));
1133             felec            = _mm256_mul_ps(qq22,_mm256_sub_ps(_mm256_mul_ps(rinv22,rinvsq22),krf2));
1134
1135             cutoff_mask      = _mm256_cmp_ps(rsq22,rcutoff2,_CMP_LT_OQ);
1136
1137             /* Update potential sum for this i atom from the interaction with this j atom. */
1138             velec            = _mm256_and_ps(velec,cutoff_mask);
1139             velec            = _mm256_andnot_ps(dummy_mask,velec);
1140             velecsum         = _mm256_add_ps(velecsum,velec);
1141
1142             fscal            = felec;
1143
1144             fscal            = _mm256_and_ps(fscal,cutoff_mask);
1145
1146             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
1147
1148             /* Calculate temporary vectorial force */
1149             tx               = _mm256_mul_ps(fscal,dx22);
1150             ty               = _mm256_mul_ps(fscal,dy22);
1151             tz               = _mm256_mul_ps(fscal,dz22);
1152
1153             /* Update vectorial force */
1154             fix2             = _mm256_add_ps(fix2,tx);
1155             fiy2             = _mm256_add_ps(fiy2,ty);
1156             fiz2             = _mm256_add_ps(fiz2,tz);
1157
1158             fjx2             = _mm256_add_ps(fjx2,tx);
1159             fjy2             = _mm256_add_ps(fjy2,ty);
1160             fjz2             = _mm256_add_ps(fjz2,tz);
1161
1162             }
1163
1164             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
1165             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
1166             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
1167             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
1168             fjptrE             = (jnrlistE>=0) ? f+j_coord_offsetE : scratch;
1169             fjptrF             = (jnrlistF>=0) ? f+j_coord_offsetF : scratch;
1170             fjptrG             = (jnrlistG>=0) ? f+j_coord_offsetG : scratch;
1171             fjptrH             = (jnrlistH>=0) ? f+j_coord_offsetH : scratch;
1172
1173             gmx_mm256_decrement_3rvec_8ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjptrE,fjptrF,fjptrG,fjptrH,
1174                                                       fjx0,fjy0,fjz0,fjx1,fjy1,fjz1,fjx2,fjy2,fjz2);
1175
1176             /* Inner loop uses 342 flops */
1177         }
1178
1179         /* End of innermost loop */
1180
1181         gmx_mm256_update_iforce_3atom_swizzle_ps(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,
1182                                                  f+i_coord_offset,fshift+i_shift_offset);
1183
1184         ggid                        = gid[iidx];
1185         /* Update potential energies */
1186         gmx_mm256_update_1pot_ps(velecsum,kernel_data->energygrp_elec+ggid);
1187         gmx_mm256_update_1pot_ps(vvdwsum,kernel_data->energygrp_vdw+ggid);
1188
1189         /* Increment number of inner iterations */
1190         inneriter                  += j_index_end - j_index_start;
1191
1192         /* Outer loop uses 20 flops */
1193     }
1194
1195     /* Increment number of outer iterations */
1196     outeriter        += nri;
1197
1198     /* Update outer/inner flops */
1199
1200     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W3W3_VF,outeriter*20 + inneriter*342);
1201 }
1202 /*
1203  * Gromacs nonbonded kernel:   nb_kernel_ElecRFCut_VdwLJSh_GeomW3W3_F_avx_256_single
1204  * Electrostatics interaction: ReactionField
1205  * VdW interaction:            LennardJones
1206  * Geometry:                   Water3-Water3
1207  * Calculate force/pot:        Force
1208  */
1209 void
1210 nb_kernel_ElecRFCut_VdwLJSh_GeomW3W3_F_avx_256_single
1211                     (t_nblist                    * gmx_restrict       nlist,
1212                      rvec                        * gmx_restrict          xx,
1213                      rvec                        * gmx_restrict          ff,
1214                      t_forcerec                  * gmx_restrict          fr,
1215                      t_mdatoms                   * gmx_restrict     mdatoms,
1216                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
1217                      t_nrnb                      * gmx_restrict        nrnb)
1218 {
1219     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
1220      * just 0 for non-waters.
1221      * Suffixes A,B,C,D,E,F,G,H refer to j loop unrolling done with AVX, e.g. for the eight different
1222      * jnr indices corresponding to data put in the four positions in the SIMD register.
1223      */
1224     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
1225     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
1226     int              jnrA,jnrB,jnrC,jnrD;
1227     int              jnrE,jnrF,jnrG,jnrH;
1228     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
1229     int              jnrlistE,jnrlistF,jnrlistG,jnrlistH;
1230     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
1231     int              j_coord_offsetE,j_coord_offsetF,j_coord_offsetG,j_coord_offsetH;
1232     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
1233     real             rcutoff_scalar;
1234     real             *shiftvec,*fshift,*x,*f;
1235     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD,*fjptrE,*fjptrF,*fjptrG,*fjptrH;
1236     real             scratch[4*DIM];
1237     __m256           tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
1238     real *           vdwioffsetptr0;
1239     __m256           ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
1240     real *           vdwioffsetptr1;
1241     __m256           ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
1242     real *           vdwioffsetptr2;
1243     __m256           ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
1244     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D,vdwjidx0E,vdwjidx0F,vdwjidx0G,vdwjidx0H;
1245     __m256           jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
1246     int              vdwjidx1A,vdwjidx1B,vdwjidx1C,vdwjidx1D,vdwjidx1E,vdwjidx1F,vdwjidx1G,vdwjidx1H;
1247     __m256           jx1,jy1,jz1,fjx1,fjy1,fjz1,jq1,isaj1;
1248     int              vdwjidx2A,vdwjidx2B,vdwjidx2C,vdwjidx2D,vdwjidx2E,vdwjidx2F,vdwjidx2G,vdwjidx2H;
1249     __m256           jx2,jy2,jz2,fjx2,fjy2,fjz2,jq2,isaj2;
1250     __m256           dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
1251     __m256           dx01,dy01,dz01,rsq01,rinv01,rinvsq01,r01,qq01,c6_01,c12_01;
1252     __m256           dx02,dy02,dz02,rsq02,rinv02,rinvsq02,r02,qq02,c6_02,c12_02;
1253     __m256           dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
1254     __m256           dx11,dy11,dz11,rsq11,rinv11,rinvsq11,r11,qq11,c6_11,c12_11;
1255     __m256           dx12,dy12,dz12,rsq12,rinv12,rinvsq12,r12,qq12,c6_12,c12_12;
1256     __m256           dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
1257     __m256           dx21,dy21,dz21,rsq21,rinv21,rinvsq21,r21,qq21,c6_21,c12_21;
1258     __m256           dx22,dy22,dz22,rsq22,rinv22,rinvsq22,r22,qq22,c6_22,c12_22;
1259     __m256           velec,felec,velecsum,facel,crf,krf,krf2;
1260     real             *charge;
1261     int              nvdwtype;
1262     __m256           rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
1263     int              *vdwtype;
1264     real             *vdwparam;
1265     __m256           one_sixth   = _mm256_set1_ps(1.0/6.0);
1266     __m256           one_twelfth = _mm256_set1_ps(1.0/12.0);
1267     __m256           dummy_mask,cutoff_mask;
1268     __m256           signbit = _mm256_castsi256_ps( _mm256_set1_epi32(0x80000000) );
1269     __m256           one     = _mm256_set1_ps(1.0);
1270     __m256           two     = _mm256_set1_ps(2.0);
1271     x                = xx[0];
1272     f                = ff[0];
1273
1274     nri              = nlist->nri;
1275     iinr             = nlist->iinr;
1276     jindex           = nlist->jindex;
1277     jjnr             = nlist->jjnr;
1278     shiftidx         = nlist->shift;
1279     gid              = nlist->gid;
1280     shiftvec         = fr->shift_vec[0];
1281     fshift           = fr->fshift[0];
1282     facel            = _mm256_set1_ps(fr->epsfac);
1283     charge           = mdatoms->chargeA;
1284     krf              = _mm256_set1_ps(fr->ic->k_rf);
1285     krf2             = _mm256_set1_ps(fr->ic->k_rf*2.0);
1286     crf              = _mm256_set1_ps(fr->ic->c_rf);
1287     nvdwtype         = fr->ntype;
1288     vdwparam         = fr->nbfp;
1289     vdwtype          = mdatoms->typeA;
1290
1291     /* Setup water-specific parameters */
1292     inr              = nlist->iinr[0];
1293     iq0              = _mm256_mul_ps(facel,_mm256_set1_ps(charge[inr+0]));
1294     iq1              = _mm256_mul_ps(facel,_mm256_set1_ps(charge[inr+1]));
1295     iq2              = _mm256_mul_ps(facel,_mm256_set1_ps(charge[inr+2]));
1296     vdwioffsetptr0   = vdwparam+2*nvdwtype*vdwtype[inr+0];
1297
1298     jq0              = _mm256_set1_ps(charge[inr+0]);
1299     jq1              = _mm256_set1_ps(charge[inr+1]);
1300     jq2              = _mm256_set1_ps(charge[inr+2]);
1301     vdwjidx0A        = 2*vdwtype[inr+0];
1302     qq00             = _mm256_mul_ps(iq0,jq0);
1303     c6_00            = _mm256_set1_ps(vdwioffsetptr0[vdwjidx0A]);
1304     c12_00           = _mm256_set1_ps(vdwioffsetptr0[vdwjidx0A+1]);
1305     qq01             = _mm256_mul_ps(iq0,jq1);
1306     qq02             = _mm256_mul_ps(iq0,jq2);
1307     qq10             = _mm256_mul_ps(iq1,jq0);
1308     qq11             = _mm256_mul_ps(iq1,jq1);
1309     qq12             = _mm256_mul_ps(iq1,jq2);
1310     qq20             = _mm256_mul_ps(iq2,jq0);
1311     qq21             = _mm256_mul_ps(iq2,jq1);
1312     qq22             = _mm256_mul_ps(iq2,jq2);
1313
1314     /* When we use explicit cutoffs the value must be identical for elec and VdW, so use elec as an arbitrary choice */
1315     rcutoff_scalar   = fr->rcoulomb;
1316     rcutoff          = _mm256_set1_ps(rcutoff_scalar);
1317     rcutoff2         = _mm256_mul_ps(rcutoff,rcutoff);
1318
1319     sh_vdw_invrcut6  = _mm256_set1_ps(fr->ic->sh_invrc6);
1320     rvdw             = _mm256_set1_ps(fr->rvdw);
1321
1322     /* Avoid stupid compiler warnings */
1323     jnrA = jnrB = jnrC = jnrD = jnrE = jnrF = jnrG = jnrH = 0;
1324     j_coord_offsetA = 0;
1325     j_coord_offsetB = 0;
1326     j_coord_offsetC = 0;
1327     j_coord_offsetD = 0;
1328     j_coord_offsetE = 0;
1329     j_coord_offsetF = 0;
1330     j_coord_offsetG = 0;
1331     j_coord_offsetH = 0;
1332
1333     outeriter        = 0;
1334     inneriter        = 0;
1335
1336     for(iidx=0;iidx<4*DIM;iidx++)
1337     {
1338         scratch[iidx] = 0.0;
1339     }
1340
1341     /* Start outer loop over neighborlists */
1342     for(iidx=0; iidx<nri; iidx++)
1343     {
1344         /* Load shift vector for this list */
1345         i_shift_offset   = DIM*shiftidx[iidx];
1346
1347         /* Load limits for loop over neighbors */
1348         j_index_start    = jindex[iidx];
1349         j_index_end      = jindex[iidx+1];
1350
1351         /* Get outer coordinate index */
1352         inr              = iinr[iidx];
1353         i_coord_offset   = DIM*inr;
1354
1355         /* Load i particle coords and add shift vector */
1356         gmx_mm256_load_shift_and_3rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,
1357                                                     &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2);
1358
1359         fix0             = _mm256_setzero_ps();
1360         fiy0             = _mm256_setzero_ps();
1361         fiz0             = _mm256_setzero_ps();
1362         fix1             = _mm256_setzero_ps();
1363         fiy1             = _mm256_setzero_ps();
1364         fiz1             = _mm256_setzero_ps();
1365         fix2             = _mm256_setzero_ps();
1366         fiy2             = _mm256_setzero_ps();
1367         fiz2             = _mm256_setzero_ps();
1368
1369         /* Start inner kernel loop */
1370         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+7]>=0; jidx+=8)
1371         {
1372
1373             /* Get j neighbor index, and coordinate index */
1374             jnrA             = jjnr[jidx];
1375             jnrB             = jjnr[jidx+1];
1376             jnrC             = jjnr[jidx+2];
1377             jnrD             = jjnr[jidx+3];
1378             jnrE             = jjnr[jidx+4];
1379             jnrF             = jjnr[jidx+5];
1380             jnrG             = jjnr[jidx+6];
1381             jnrH             = jjnr[jidx+7];
1382             j_coord_offsetA  = DIM*jnrA;
1383             j_coord_offsetB  = DIM*jnrB;
1384             j_coord_offsetC  = DIM*jnrC;
1385             j_coord_offsetD  = DIM*jnrD;
1386             j_coord_offsetE  = DIM*jnrE;
1387             j_coord_offsetF  = DIM*jnrF;
1388             j_coord_offsetG  = DIM*jnrG;
1389             j_coord_offsetH  = DIM*jnrH;
1390
1391             /* load j atom coordinates */
1392             gmx_mm256_load_3rvec_8ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
1393                                                  x+j_coord_offsetC,x+j_coord_offsetD,
1394                                                  x+j_coord_offsetE,x+j_coord_offsetF,
1395                                                  x+j_coord_offsetG,x+j_coord_offsetH,
1396                                               &jx0,&jy0,&jz0,&jx1,&jy1,&jz1,&jx2,&jy2,&jz2);
1397
1398             /* Calculate displacement vector */
1399             dx00             = _mm256_sub_ps(ix0,jx0);
1400             dy00             = _mm256_sub_ps(iy0,jy0);
1401             dz00             = _mm256_sub_ps(iz0,jz0);
1402             dx01             = _mm256_sub_ps(ix0,jx1);
1403             dy01             = _mm256_sub_ps(iy0,jy1);
1404             dz01             = _mm256_sub_ps(iz0,jz1);
1405             dx02             = _mm256_sub_ps(ix0,jx2);
1406             dy02             = _mm256_sub_ps(iy0,jy2);
1407             dz02             = _mm256_sub_ps(iz0,jz2);
1408             dx10             = _mm256_sub_ps(ix1,jx0);
1409             dy10             = _mm256_sub_ps(iy1,jy0);
1410             dz10             = _mm256_sub_ps(iz1,jz0);
1411             dx11             = _mm256_sub_ps(ix1,jx1);
1412             dy11             = _mm256_sub_ps(iy1,jy1);
1413             dz11             = _mm256_sub_ps(iz1,jz1);
1414             dx12             = _mm256_sub_ps(ix1,jx2);
1415             dy12             = _mm256_sub_ps(iy1,jy2);
1416             dz12             = _mm256_sub_ps(iz1,jz2);
1417             dx20             = _mm256_sub_ps(ix2,jx0);
1418             dy20             = _mm256_sub_ps(iy2,jy0);
1419             dz20             = _mm256_sub_ps(iz2,jz0);
1420             dx21             = _mm256_sub_ps(ix2,jx1);
1421             dy21             = _mm256_sub_ps(iy2,jy1);
1422             dz21             = _mm256_sub_ps(iz2,jz1);
1423             dx22             = _mm256_sub_ps(ix2,jx2);
1424             dy22             = _mm256_sub_ps(iy2,jy2);
1425             dz22             = _mm256_sub_ps(iz2,jz2);
1426
1427             /* Calculate squared distance and things based on it */
1428             rsq00            = gmx_mm256_calc_rsq_ps(dx00,dy00,dz00);
1429             rsq01            = gmx_mm256_calc_rsq_ps(dx01,dy01,dz01);
1430             rsq02            = gmx_mm256_calc_rsq_ps(dx02,dy02,dz02);
1431             rsq10            = gmx_mm256_calc_rsq_ps(dx10,dy10,dz10);
1432             rsq11            = gmx_mm256_calc_rsq_ps(dx11,dy11,dz11);
1433             rsq12            = gmx_mm256_calc_rsq_ps(dx12,dy12,dz12);
1434             rsq20            = gmx_mm256_calc_rsq_ps(dx20,dy20,dz20);
1435             rsq21            = gmx_mm256_calc_rsq_ps(dx21,dy21,dz21);
1436             rsq22            = gmx_mm256_calc_rsq_ps(dx22,dy22,dz22);
1437
1438             rinv00           = gmx_mm256_invsqrt_ps(rsq00);
1439             rinv01           = gmx_mm256_invsqrt_ps(rsq01);
1440             rinv02           = gmx_mm256_invsqrt_ps(rsq02);
1441             rinv10           = gmx_mm256_invsqrt_ps(rsq10);
1442             rinv11           = gmx_mm256_invsqrt_ps(rsq11);
1443             rinv12           = gmx_mm256_invsqrt_ps(rsq12);
1444             rinv20           = gmx_mm256_invsqrt_ps(rsq20);
1445             rinv21           = gmx_mm256_invsqrt_ps(rsq21);
1446             rinv22           = gmx_mm256_invsqrt_ps(rsq22);
1447
1448             rinvsq00         = _mm256_mul_ps(rinv00,rinv00);
1449             rinvsq01         = _mm256_mul_ps(rinv01,rinv01);
1450             rinvsq02         = _mm256_mul_ps(rinv02,rinv02);
1451             rinvsq10         = _mm256_mul_ps(rinv10,rinv10);
1452             rinvsq11         = _mm256_mul_ps(rinv11,rinv11);
1453             rinvsq12         = _mm256_mul_ps(rinv12,rinv12);
1454             rinvsq20         = _mm256_mul_ps(rinv20,rinv20);
1455             rinvsq21         = _mm256_mul_ps(rinv21,rinv21);
1456             rinvsq22         = _mm256_mul_ps(rinv22,rinv22);
1457
1458             fjx0             = _mm256_setzero_ps();
1459             fjy0             = _mm256_setzero_ps();
1460             fjz0             = _mm256_setzero_ps();
1461             fjx1             = _mm256_setzero_ps();
1462             fjy1             = _mm256_setzero_ps();
1463             fjz1             = _mm256_setzero_ps();
1464             fjx2             = _mm256_setzero_ps();
1465             fjy2             = _mm256_setzero_ps();
1466             fjz2             = _mm256_setzero_ps();
1467
1468             /**************************
1469              * CALCULATE INTERACTIONS *
1470              **************************/
1471
1472             if (gmx_mm256_any_lt(rsq00,rcutoff2))
1473             {
1474
1475             /* REACTION-FIELD ELECTROSTATICS */
1476             felec            = _mm256_mul_ps(qq00,_mm256_sub_ps(_mm256_mul_ps(rinv00,rinvsq00),krf2));
1477
1478             /* LENNARD-JONES DISPERSION/REPULSION */
1479
1480             rinvsix          = _mm256_mul_ps(_mm256_mul_ps(rinvsq00,rinvsq00),rinvsq00);
1481             fvdw             = _mm256_mul_ps(_mm256_sub_ps(_mm256_mul_ps(c12_00,rinvsix),c6_00),_mm256_mul_ps(rinvsix,rinvsq00));
1482
1483             cutoff_mask      = _mm256_cmp_ps(rsq00,rcutoff2,_CMP_LT_OQ);
1484
1485             fscal            = _mm256_add_ps(felec,fvdw);
1486
1487             fscal            = _mm256_and_ps(fscal,cutoff_mask);
1488
1489             /* Calculate temporary vectorial force */
1490             tx               = _mm256_mul_ps(fscal,dx00);
1491             ty               = _mm256_mul_ps(fscal,dy00);
1492             tz               = _mm256_mul_ps(fscal,dz00);
1493
1494             /* Update vectorial force */
1495             fix0             = _mm256_add_ps(fix0,tx);
1496             fiy0             = _mm256_add_ps(fiy0,ty);
1497             fiz0             = _mm256_add_ps(fiz0,tz);
1498
1499             fjx0             = _mm256_add_ps(fjx0,tx);
1500             fjy0             = _mm256_add_ps(fjy0,ty);
1501             fjz0             = _mm256_add_ps(fjz0,tz);
1502
1503             }
1504
1505             /**************************
1506              * CALCULATE INTERACTIONS *
1507              **************************/
1508
1509             if (gmx_mm256_any_lt(rsq01,rcutoff2))
1510             {
1511
1512             /* REACTION-FIELD ELECTROSTATICS */
1513             felec            = _mm256_mul_ps(qq01,_mm256_sub_ps(_mm256_mul_ps(rinv01,rinvsq01),krf2));
1514
1515             cutoff_mask      = _mm256_cmp_ps(rsq01,rcutoff2,_CMP_LT_OQ);
1516
1517             fscal            = felec;
1518
1519             fscal            = _mm256_and_ps(fscal,cutoff_mask);
1520
1521             /* Calculate temporary vectorial force */
1522             tx               = _mm256_mul_ps(fscal,dx01);
1523             ty               = _mm256_mul_ps(fscal,dy01);
1524             tz               = _mm256_mul_ps(fscal,dz01);
1525
1526             /* Update vectorial force */
1527             fix0             = _mm256_add_ps(fix0,tx);
1528             fiy0             = _mm256_add_ps(fiy0,ty);
1529             fiz0             = _mm256_add_ps(fiz0,tz);
1530
1531             fjx1             = _mm256_add_ps(fjx1,tx);
1532             fjy1             = _mm256_add_ps(fjy1,ty);
1533             fjz1             = _mm256_add_ps(fjz1,tz);
1534
1535             }
1536
1537             /**************************
1538              * CALCULATE INTERACTIONS *
1539              **************************/
1540
1541             if (gmx_mm256_any_lt(rsq02,rcutoff2))
1542             {
1543
1544             /* REACTION-FIELD ELECTROSTATICS */
1545             felec            = _mm256_mul_ps(qq02,_mm256_sub_ps(_mm256_mul_ps(rinv02,rinvsq02),krf2));
1546
1547             cutoff_mask      = _mm256_cmp_ps(rsq02,rcutoff2,_CMP_LT_OQ);
1548
1549             fscal            = felec;
1550
1551             fscal            = _mm256_and_ps(fscal,cutoff_mask);
1552
1553             /* Calculate temporary vectorial force */
1554             tx               = _mm256_mul_ps(fscal,dx02);
1555             ty               = _mm256_mul_ps(fscal,dy02);
1556             tz               = _mm256_mul_ps(fscal,dz02);
1557
1558             /* Update vectorial force */
1559             fix0             = _mm256_add_ps(fix0,tx);
1560             fiy0             = _mm256_add_ps(fiy0,ty);
1561             fiz0             = _mm256_add_ps(fiz0,tz);
1562
1563             fjx2             = _mm256_add_ps(fjx2,tx);
1564             fjy2             = _mm256_add_ps(fjy2,ty);
1565             fjz2             = _mm256_add_ps(fjz2,tz);
1566
1567             }
1568
1569             /**************************
1570              * CALCULATE INTERACTIONS *
1571              **************************/
1572
1573             if (gmx_mm256_any_lt(rsq10,rcutoff2))
1574             {
1575
1576             /* REACTION-FIELD ELECTROSTATICS */
1577             felec            = _mm256_mul_ps(qq10,_mm256_sub_ps(_mm256_mul_ps(rinv10,rinvsq10),krf2));
1578
1579             cutoff_mask      = _mm256_cmp_ps(rsq10,rcutoff2,_CMP_LT_OQ);
1580
1581             fscal            = felec;
1582
1583             fscal            = _mm256_and_ps(fscal,cutoff_mask);
1584
1585             /* Calculate temporary vectorial force */
1586             tx               = _mm256_mul_ps(fscal,dx10);
1587             ty               = _mm256_mul_ps(fscal,dy10);
1588             tz               = _mm256_mul_ps(fscal,dz10);
1589
1590             /* Update vectorial force */
1591             fix1             = _mm256_add_ps(fix1,tx);
1592             fiy1             = _mm256_add_ps(fiy1,ty);
1593             fiz1             = _mm256_add_ps(fiz1,tz);
1594
1595             fjx0             = _mm256_add_ps(fjx0,tx);
1596             fjy0             = _mm256_add_ps(fjy0,ty);
1597             fjz0             = _mm256_add_ps(fjz0,tz);
1598
1599             }
1600
1601             /**************************
1602              * CALCULATE INTERACTIONS *
1603              **************************/
1604
1605             if (gmx_mm256_any_lt(rsq11,rcutoff2))
1606             {
1607
1608             /* REACTION-FIELD ELECTROSTATICS */
1609             felec            = _mm256_mul_ps(qq11,_mm256_sub_ps(_mm256_mul_ps(rinv11,rinvsq11),krf2));
1610
1611             cutoff_mask      = _mm256_cmp_ps(rsq11,rcutoff2,_CMP_LT_OQ);
1612
1613             fscal            = felec;
1614
1615             fscal            = _mm256_and_ps(fscal,cutoff_mask);
1616
1617             /* Calculate temporary vectorial force */
1618             tx               = _mm256_mul_ps(fscal,dx11);
1619             ty               = _mm256_mul_ps(fscal,dy11);
1620             tz               = _mm256_mul_ps(fscal,dz11);
1621
1622             /* Update vectorial force */
1623             fix1             = _mm256_add_ps(fix1,tx);
1624             fiy1             = _mm256_add_ps(fiy1,ty);
1625             fiz1             = _mm256_add_ps(fiz1,tz);
1626
1627             fjx1             = _mm256_add_ps(fjx1,tx);
1628             fjy1             = _mm256_add_ps(fjy1,ty);
1629             fjz1             = _mm256_add_ps(fjz1,tz);
1630
1631             }
1632
1633             /**************************
1634              * CALCULATE INTERACTIONS *
1635              **************************/
1636
1637             if (gmx_mm256_any_lt(rsq12,rcutoff2))
1638             {
1639
1640             /* REACTION-FIELD ELECTROSTATICS */
1641             felec            = _mm256_mul_ps(qq12,_mm256_sub_ps(_mm256_mul_ps(rinv12,rinvsq12),krf2));
1642
1643             cutoff_mask      = _mm256_cmp_ps(rsq12,rcutoff2,_CMP_LT_OQ);
1644
1645             fscal            = felec;
1646
1647             fscal            = _mm256_and_ps(fscal,cutoff_mask);
1648
1649             /* Calculate temporary vectorial force */
1650             tx               = _mm256_mul_ps(fscal,dx12);
1651             ty               = _mm256_mul_ps(fscal,dy12);
1652             tz               = _mm256_mul_ps(fscal,dz12);
1653
1654             /* Update vectorial force */
1655             fix1             = _mm256_add_ps(fix1,tx);
1656             fiy1             = _mm256_add_ps(fiy1,ty);
1657             fiz1             = _mm256_add_ps(fiz1,tz);
1658
1659             fjx2             = _mm256_add_ps(fjx2,tx);
1660             fjy2             = _mm256_add_ps(fjy2,ty);
1661             fjz2             = _mm256_add_ps(fjz2,tz);
1662
1663             }
1664
1665             /**************************
1666              * CALCULATE INTERACTIONS *
1667              **************************/
1668
1669             if (gmx_mm256_any_lt(rsq20,rcutoff2))
1670             {
1671
1672             /* REACTION-FIELD ELECTROSTATICS */
1673             felec            = _mm256_mul_ps(qq20,_mm256_sub_ps(_mm256_mul_ps(rinv20,rinvsq20),krf2));
1674
1675             cutoff_mask      = _mm256_cmp_ps(rsq20,rcutoff2,_CMP_LT_OQ);
1676
1677             fscal            = felec;
1678
1679             fscal            = _mm256_and_ps(fscal,cutoff_mask);
1680
1681             /* Calculate temporary vectorial force */
1682             tx               = _mm256_mul_ps(fscal,dx20);
1683             ty               = _mm256_mul_ps(fscal,dy20);
1684             tz               = _mm256_mul_ps(fscal,dz20);
1685
1686             /* Update vectorial force */
1687             fix2             = _mm256_add_ps(fix2,tx);
1688             fiy2             = _mm256_add_ps(fiy2,ty);
1689             fiz2             = _mm256_add_ps(fiz2,tz);
1690
1691             fjx0             = _mm256_add_ps(fjx0,tx);
1692             fjy0             = _mm256_add_ps(fjy0,ty);
1693             fjz0             = _mm256_add_ps(fjz0,tz);
1694
1695             }
1696
1697             /**************************
1698              * CALCULATE INTERACTIONS *
1699              **************************/
1700
1701             if (gmx_mm256_any_lt(rsq21,rcutoff2))
1702             {
1703
1704             /* REACTION-FIELD ELECTROSTATICS */
1705             felec            = _mm256_mul_ps(qq21,_mm256_sub_ps(_mm256_mul_ps(rinv21,rinvsq21),krf2));
1706
1707             cutoff_mask      = _mm256_cmp_ps(rsq21,rcutoff2,_CMP_LT_OQ);
1708
1709             fscal            = felec;
1710
1711             fscal            = _mm256_and_ps(fscal,cutoff_mask);
1712
1713             /* Calculate temporary vectorial force */
1714             tx               = _mm256_mul_ps(fscal,dx21);
1715             ty               = _mm256_mul_ps(fscal,dy21);
1716             tz               = _mm256_mul_ps(fscal,dz21);
1717
1718             /* Update vectorial force */
1719             fix2             = _mm256_add_ps(fix2,tx);
1720             fiy2             = _mm256_add_ps(fiy2,ty);
1721             fiz2             = _mm256_add_ps(fiz2,tz);
1722
1723             fjx1             = _mm256_add_ps(fjx1,tx);
1724             fjy1             = _mm256_add_ps(fjy1,ty);
1725             fjz1             = _mm256_add_ps(fjz1,tz);
1726
1727             }
1728
1729             /**************************
1730              * CALCULATE INTERACTIONS *
1731              **************************/
1732
1733             if (gmx_mm256_any_lt(rsq22,rcutoff2))
1734             {
1735
1736             /* REACTION-FIELD ELECTROSTATICS */
1737             felec            = _mm256_mul_ps(qq22,_mm256_sub_ps(_mm256_mul_ps(rinv22,rinvsq22),krf2));
1738
1739             cutoff_mask      = _mm256_cmp_ps(rsq22,rcutoff2,_CMP_LT_OQ);
1740
1741             fscal            = felec;
1742
1743             fscal            = _mm256_and_ps(fscal,cutoff_mask);
1744
1745             /* Calculate temporary vectorial force */
1746             tx               = _mm256_mul_ps(fscal,dx22);
1747             ty               = _mm256_mul_ps(fscal,dy22);
1748             tz               = _mm256_mul_ps(fscal,dz22);
1749
1750             /* Update vectorial force */
1751             fix2             = _mm256_add_ps(fix2,tx);
1752             fiy2             = _mm256_add_ps(fiy2,ty);
1753             fiz2             = _mm256_add_ps(fiz2,tz);
1754
1755             fjx2             = _mm256_add_ps(fjx2,tx);
1756             fjy2             = _mm256_add_ps(fjy2,ty);
1757             fjz2             = _mm256_add_ps(fjz2,tz);
1758
1759             }
1760
1761             fjptrA             = f+j_coord_offsetA;
1762             fjptrB             = f+j_coord_offsetB;
1763             fjptrC             = f+j_coord_offsetC;
1764             fjptrD             = f+j_coord_offsetD;
1765             fjptrE             = f+j_coord_offsetE;
1766             fjptrF             = f+j_coord_offsetF;
1767             fjptrG             = f+j_coord_offsetG;
1768             fjptrH             = f+j_coord_offsetH;
1769
1770             gmx_mm256_decrement_3rvec_8ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjptrE,fjptrF,fjptrG,fjptrH,
1771                                                       fjx0,fjy0,fjz0,fjx1,fjy1,fjz1,fjx2,fjy2,fjz2);
1772
1773             /* Inner loop uses 277 flops */
1774         }
1775
1776         if(jidx<j_index_end)
1777         {
1778
1779             /* Get j neighbor index, and coordinate index */
1780             jnrlistA         = jjnr[jidx];
1781             jnrlistB         = jjnr[jidx+1];
1782             jnrlistC         = jjnr[jidx+2];
1783             jnrlistD         = jjnr[jidx+3];
1784             jnrlistE         = jjnr[jidx+4];
1785             jnrlistF         = jjnr[jidx+5];
1786             jnrlistG         = jjnr[jidx+6];
1787             jnrlistH         = jjnr[jidx+7];
1788             /* Sign of each element will be negative for non-real atoms.
1789              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
1790              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
1791              */
1792             dummy_mask = gmx_mm256_set_m128(gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx+4)),_mm_setzero_si128())),
1793                                             gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128())));
1794
1795             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
1796             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
1797             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
1798             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
1799             jnrE       = (jnrlistE>=0) ? jnrlistE : 0;
1800             jnrF       = (jnrlistF>=0) ? jnrlistF : 0;
1801             jnrG       = (jnrlistG>=0) ? jnrlistG : 0;
1802             jnrH       = (jnrlistH>=0) ? jnrlistH : 0;
1803             j_coord_offsetA  = DIM*jnrA;
1804             j_coord_offsetB  = DIM*jnrB;
1805             j_coord_offsetC  = DIM*jnrC;
1806             j_coord_offsetD  = DIM*jnrD;
1807             j_coord_offsetE  = DIM*jnrE;
1808             j_coord_offsetF  = DIM*jnrF;
1809             j_coord_offsetG  = DIM*jnrG;
1810             j_coord_offsetH  = DIM*jnrH;
1811
1812             /* load j atom coordinates */
1813             gmx_mm256_load_3rvec_8ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
1814                                                  x+j_coord_offsetC,x+j_coord_offsetD,
1815                                                  x+j_coord_offsetE,x+j_coord_offsetF,
1816                                                  x+j_coord_offsetG,x+j_coord_offsetH,
1817                                               &jx0,&jy0,&jz0,&jx1,&jy1,&jz1,&jx2,&jy2,&jz2);
1818
1819             /* Calculate displacement vector */
1820             dx00             = _mm256_sub_ps(ix0,jx0);
1821             dy00             = _mm256_sub_ps(iy0,jy0);
1822             dz00             = _mm256_sub_ps(iz0,jz0);
1823             dx01             = _mm256_sub_ps(ix0,jx1);
1824             dy01             = _mm256_sub_ps(iy0,jy1);
1825             dz01             = _mm256_sub_ps(iz0,jz1);
1826             dx02             = _mm256_sub_ps(ix0,jx2);
1827             dy02             = _mm256_sub_ps(iy0,jy2);
1828             dz02             = _mm256_sub_ps(iz0,jz2);
1829             dx10             = _mm256_sub_ps(ix1,jx0);
1830             dy10             = _mm256_sub_ps(iy1,jy0);
1831             dz10             = _mm256_sub_ps(iz1,jz0);
1832             dx11             = _mm256_sub_ps(ix1,jx1);
1833             dy11             = _mm256_sub_ps(iy1,jy1);
1834             dz11             = _mm256_sub_ps(iz1,jz1);
1835             dx12             = _mm256_sub_ps(ix1,jx2);
1836             dy12             = _mm256_sub_ps(iy1,jy2);
1837             dz12             = _mm256_sub_ps(iz1,jz2);
1838             dx20             = _mm256_sub_ps(ix2,jx0);
1839             dy20             = _mm256_sub_ps(iy2,jy0);
1840             dz20             = _mm256_sub_ps(iz2,jz0);
1841             dx21             = _mm256_sub_ps(ix2,jx1);
1842             dy21             = _mm256_sub_ps(iy2,jy1);
1843             dz21             = _mm256_sub_ps(iz2,jz1);
1844             dx22             = _mm256_sub_ps(ix2,jx2);
1845             dy22             = _mm256_sub_ps(iy2,jy2);
1846             dz22             = _mm256_sub_ps(iz2,jz2);
1847
1848             /* Calculate squared distance and things based on it */
1849             rsq00            = gmx_mm256_calc_rsq_ps(dx00,dy00,dz00);
1850             rsq01            = gmx_mm256_calc_rsq_ps(dx01,dy01,dz01);
1851             rsq02            = gmx_mm256_calc_rsq_ps(dx02,dy02,dz02);
1852             rsq10            = gmx_mm256_calc_rsq_ps(dx10,dy10,dz10);
1853             rsq11            = gmx_mm256_calc_rsq_ps(dx11,dy11,dz11);
1854             rsq12            = gmx_mm256_calc_rsq_ps(dx12,dy12,dz12);
1855             rsq20            = gmx_mm256_calc_rsq_ps(dx20,dy20,dz20);
1856             rsq21            = gmx_mm256_calc_rsq_ps(dx21,dy21,dz21);
1857             rsq22            = gmx_mm256_calc_rsq_ps(dx22,dy22,dz22);
1858
1859             rinv00           = gmx_mm256_invsqrt_ps(rsq00);
1860             rinv01           = gmx_mm256_invsqrt_ps(rsq01);
1861             rinv02           = gmx_mm256_invsqrt_ps(rsq02);
1862             rinv10           = gmx_mm256_invsqrt_ps(rsq10);
1863             rinv11           = gmx_mm256_invsqrt_ps(rsq11);
1864             rinv12           = gmx_mm256_invsqrt_ps(rsq12);
1865             rinv20           = gmx_mm256_invsqrt_ps(rsq20);
1866             rinv21           = gmx_mm256_invsqrt_ps(rsq21);
1867             rinv22           = gmx_mm256_invsqrt_ps(rsq22);
1868
1869             rinvsq00         = _mm256_mul_ps(rinv00,rinv00);
1870             rinvsq01         = _mm256_mul_ps(rinv01,rinv01);
1871             rinvsq02         = _mm256_mul_ps(rinv02,rinv02);
1872             rinvsq10         = _mm256_mul_ps(rinv10,rinv10);
1873             rinvsq11         = _mm256_mul_ps(rinv11,rinv11);
1874             rinvsq12         = _mm256_mul_ps(rinv12,rinv12);
1875             rinvsq20         = _mm256_mul_ps(rinv20,rinv20);
1876             rinvsq21         = _mm256_mul_ps(rinv21,rinv21);
1877             rinvsq22         = _mm256_mul_ps(rinv22,rinv22);
1878
1879             fjx0             = _mm256_setzero_ps();
1880             fjy0             = _mm256_setzero_ps();
1881             fjz0             = _mm256_setzero_ps();
1882             fjx1             = _mm256_setzero_ps();
1883             fjy1             = _mm256_setzero_ps();
1884             fjz1             = _mm256_setzero_ps();
1885             fjx2             = _mm256_setzero_ps();
1886             fjy2             = _mm256_setzero_ps();
1887             fjz2             = _mm256_setzero_ps();
1888
1889             /**************************
1890              * CALCULATE INTERACTIONS *
1891              **************************/
1892
1893             if (gmx_mm256_any_lt(rsq00,rcutoff2))
1894             {
1895
1896             /* REACTION-FIELD ELECTROSTATICS */
1897             felec            = _mm256_mul_ps(qq00,_mm256_sub_ps(_mm256_mul_ps(rinv00,rinvsq00),krf2));
1898
1899             /* LENNARD-JONES DISPERSION/REPULSION */
1900
1901             rinvsix          = _mm256_mul_ps(_mm256_mul_ps(rinvsq00,rinvsq00),rinvsq00);
1902             fvdw             = _mm256_mul_ps(_mm256_sub_ps(_mm256_mul_ps(c12_00,rinvsix),c6_00),_mm256_mul_ps(rinvsix,rinvsq00));
1903
1904             cutoff_mask      = _mm256_cmp_ps(rsq00,rcutoff2,_CMP_LT_OQ);
1905
1906             fscal            = _mm256_add_ps(felec,fvdw);
1907
1908             fscal            = _mm256_and_ps(fscal,cutoff_mask);
1909
1910             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
1911
1912             /* Calculate temporary vectorial force */
1913             tx               = _mm256_mul_ps(fscal,dx00);
1914             ty               = _mm256_mul_ps(fscal,dy00);
1915             tz               = _mm256_mul_ps(fscal,dz00);
1916
1917             /* Update vectorial force */
1918             fix0             = _mm256_add_ps(fix0,tx);
1919             fiy0             = _mm256_add_ps(fiy0,ty);
1920             fiz0             = _mm256_add_ps(fiz0,tz);
1921
1922             fjx0             = _mm256_add_ps(fjx0,tx);
1923             fjy0             = _mm256_add_ps(fjy0,ty);
1924             fjz0             = _mm256_add_ps(fjz0,tz);
1925
1926             }
1927
1928             /**************************
1929              * CALCULATE INTERACTIONS *
1930              **************************/
1931
1932             if (gmx_mm256_any_lt(rsq01,rcutoff2))
1933             {
1934
1935             /* REACTION-FIELD ELECTROSTATICS */
1936             felec            = _mm256_mul_ps(qq01,_mm256_sub_ps(_mm256_mul_ps(rinv01,rinvsq01),krf2));
1937
1938             cutoff_mask      = _mm256_cmp_ps(rsq01,rcutoff2,_CMP_LT_OQ);
1939
1940             fscal            = felec;
1941
1942             fscal            = _mm256_and_ps(fscal,cutoff_mask);
1943
1944             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
1945
1946             /* Calculate temporary vectorial force */
1947             tx               = _mm256_mul_ps(fscal,dx01);
1948             ty               = _mm256_mul_ps(fscal,dy01);
1949             tz               = _mm256_mul_ps(fscal,dz01);
1950
1951             /* Update vectorial force */
1952             fix0             = _mm256_add_ps(fix0,tx);
1953             fiy0             = _mm256_add_ps(fiy0,ty);
1954             fiz0             = _mm256_add_ps(fiz0,tz);
1955
1956             fjx1             = _mm256_add_ps(fjx1,tx);
1957             fjy1             = _mm256_add_ps(fjy1,ty);
1958             fjz1             = _mm256_add_ps(fjz1,tz);
1959
1960             }
1961
1962             /**************************
1963              * CALCULATE INTERACTIONS *
1964              **************************/
1965
1966             if (gmx_mm256_any_lt(rsq02,rcutoff2))
1967             {
1968
1969             /* REACTION-FIELD ELECTROSTATICS */
1970             felec            = _mm256_mul_ps(qq02,_mm256_sub_ps(_mm256_mul_ps(rinv02,rinvsq02),krf2));
1971
1972             cutoff_mask      = _mm256_cmp_ps(rsq02,rcutoff2,_CMP_LT_OQ);
1973
1974             fscal            = felec;
1975
1976             fscal            = _mm256_and_ps(fscal,cutoff_mask);
1977
1978             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
1979
1980             /* Calculate temporary vectorial force */
1981             tx               = _mm256_mul_ps(fscal,dx02);
1982             ty               = _mm256_mul_ps(fscal,dy02);
1983             tz               = _mm256_mul_ps(fscal,dz02);
1984
1985             /* Update vectorial force */
1986             fix0             = _mm256_add_ps(fix0,tx);
1987             fiy0             = _mm256_add_ps(fiy0,ty);
1988             fiz0             = _mm256_add_ps(fiz0,tz);
1989
1990             fjx2             = _mm256_add_ps(fjx2,tx);
1991             fjy2             = _mm256_add_ps(fjy2,ty);
1992             fjz2             = _mm256_add_ps(fjz2,tz);
1993
1994             }
1995
1996             /**************************
1997              * CALCULATE INTERACTIONS *
1998              **************************/
1999
2000             if (gmx_mm256_any_lt(rsq10,rcutoff2))
2001             {
2002
2003             /* REACTION-FIELD ELECTROSTATICS */
2004             felec            = _mm256_mul_ps(qq10,_mm256_sub_ps(_mm256_mul_ps(rinv10,rinvsq10),krf2));
2005
2006             cutoff_mask      = _mm256_cmp_ps(rsq10,rcutoff2,_CMP_LT_OQ);
2007
2008             fscal            = felec;
2009
2010             fscal            = _mm256_and_ps(fscal,cutoff_mask);
2011
2012             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
2013
2014             /* Calculate temporary vectorial force */
2015             tx               = _mm256_mul_ps(fscal,dx10);
2016             ty               = _mm256_mul_ps(fscal,dy10);
2017             tz               = _mm256_mul_ps(fscal,dz10);
2018
2019             /* Update vectorial force */
2020             fix1             = _mm256_add_ps(fix1,tx);
2021             fiy1             = _mm256_add_ps(fiy1,ty);
2022             fiz1             = _mm256_add_ps(fiz1,tz);
2023
2024             fjx0             = _mm256_add_ps(fjx0,tx);
2025             fjy0             = _mm256_add_ps(fjy0,ty);
2026             fjz0             = _mm256_add_ps(fjz0,tz);
2027
2028             }
2029
2030             /**************************
2031              * CALCULATE INTERACTIONS *
2032              **************************/
2033
2034             if (gmx_mm256_any_lt(rsq11,rcutoff2))
2035             {
2036
2037             /* REACTION-FIELD ELECTROSTATICS */
2038             felec            = _mm256_mul_ps(qq11,_mm256_sub_ps(_mm256_mul_ps(rinv11,rinvsq11),krf2));
2039
2040             cutoff_mask      = _mm256_cmp_ps(rsq11,rcutoff2,_CMP_LT_OQ);
2041
2042             fscal            = felec;
2043
2044             fscal            = _mm256_and_ps(fscal,cutoff_mask);
2045
2046             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
2047
2048             /* Calculate temporary vectorial force */
2049             tx               = _mm256_mul_ps(fscal,dx11);
2050             ty               = _mm256_mul_ps(fscal,dy11);
2051             tz               = _mm256_mul_ps(fscal,dz11);
2052
2053             /* Update vectorial force */
2054             fix1             = _mm256_add_ps(fix1,tx);
2055             fiy1             = _mm256_add_ps(fiy1,ty);
2056             fiz1             = _mm256_add_ps(fiz1,tz);
2057
2058             fjx1             = _mm256_add_ps(fjx1,tx);
2059             fjy1             = _mm256_add_ps(fjy1,ty);
2060             fjz1             = _mm256_add_ps(fjz1,tz);
2061
2062             }
2063
2064             /**************************
2065              * CALCULATE INTERACTIONS *
2066              **************************/
2067
2068             if (gmx_mm256_any_lt(rsq12,rcutoff2))
2069             {
2070
2071             /* REACTION-FIELD ELECTROSTATICS */
2072             felec            = _mm256_mul_ps(qq12,_mm256_sub_ps(_mm256_mul_ps(rinv12,rinvsq12),krf2));
2073
2074             cutoff_mask      = _mm256_cmp_ps(rsq12,rcutoff2,_CMP_LT_OQ);
2075
2076             fscal            = felec;
2077
2078             fscal            = _mm256_and_ps(fscal,cutoff_mask);
2079
2080             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
2081
2082             /* Calculate temporary vectorial force */
2083             tx               = _mm256_mul_ps(fscal,dx12);
2084             ty               = _mm256_mul_ps(fscal,dy12);
2085             tz               = _mm256_mul_ps(fscal,dz12);
2086
2087             /* Update vectorial force */
2088             fix1             = _mm256_add_ps(fix1,tx);
2089             fiy1             = _mm256_add_ps(fiy1,ty);
2090             fiz1             = _mm256_add_ps(fiz1,tz);
2091
2092             fjx2             = _mm256_add_ps(fjx2,tx);
2093             fjy2             = _mm256_add_ps(fjy2,ty);
2094             fjz2             = _mm256_add_ps(fjz2,tz);
2095
2096             }
2097
2098             /**************************
2099              * CALCULATE INTERACTIONS *
2100              **************************/
2101
2102             if (gmx_mm256_any_lt(rsq20,rcutoff2))
2103             {
2104
2105             /* REACTION-FIELD ELECTROSTATICS */
2106             felec            = _mm256_mul_ps(qq20,_mm256_sub_ps(_mm256_mul_ps(rinv20,rinvsq20),krf2));
2107
2108             cutoff_mask      = _mm256_cmp_ps(rsq20,rcutoff2,_CMP_LT_OQ);
2109
2110             fscal            = felec;
2111
2112             fscal            = _mm256_and_ps(fscal,cutoff_mask);
2113
2114             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
2115
2116             /* Calculate temporary vectorial force */
2117             tx               = _mm256_mul_ps(fscal,dx20);
2118             ty               = _mm256_mul_ps(fscal,dy20);
2119             tz               = _mm256_mul_ps(fscal,dz20);
2120
2121             /* Update vectorial force */
2122             fix2             = _mm256_add_ps(fix2,tx);
2123             fiy2             = _mm256_add_ps(fiy2,ty);
2124             fiz2             = _mm256_add_ps(fiz2,tz);
2125
2126             fjx0             = _mm256_add_ps(fjx0,tx);
2127             fjy0             = _mm256_add_ps(fjy0,ty);
2128             fjz0             = _mm256_add_ps(fjz0,tz);
2129
2130             }
2131
2132             /**************************
2133              * CALCULATE INTERACTIONS *
2134              **************************/
2135
2136             if (gmx_mm256_any_lt(rsq21,rcutoff2))
2137             {
2138
2139             /* REACTION-FIELD ELECTROSTATICS */
2140             felec            = _mm256_mul_ps(qq21,_mm256_sub_ps(_mm256_mul_ps(rinv21,rinvsq21),krf2));
2141
2142             cutoff_mask      = _mm256_cmp_ps(rsq21,rcutoff2,_CMP_LT_OQ);
2143
2144             fscal            = felec;
2145
2146             fscal            = _mm256_and_ps(fscal,cutoff_mask);
2147
2148             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
2149
2150             /* Calculate temporary vectorial force */
2151             tx               = _mm256_mul_ps(fscal,dx21);
2152             ty               = _mm256_mul_ps(fscal,dy21);
2153             tz               = _mm256_mul_ps(fscal,dz21);
2154
2155             /* Update vectorial force */
2156             fix2             = _mm256_add_ps(fix2,tx);
2157             fiy2             = _mm256_add_ps(fiy2,ty);
2158             fiz2             = _mm256_add_ps(fiz2,tz);
2159
2160             fjx1             = _mm256_add_ps(fjx1,tx);
2161             fjy1             = _mm256_add_ps(fjy1,ty);
2162             fjz1             = _mm256_add_ps(fjz1,tz);
2163
2164             }
2165
2166             /**************************
2167              * CALCULATE INTERACTIONS *
2168              **************************/
2169
2170             if (gmx_mm256_any_lt(rsq22,rcutoff2))
2171             {
2172
2173             /* REACTION-FIELD ELECTROSTATICS */
2174             felec            = _mm256_mul_ps(qq22,_mm256_sub_ps(_mm256_mul_ps(rinv22,rinvsq22),krf2));
2175
2176             cutoff_mask      = _mm256_cmp_ps(rsq22,rcutoff2,_CMP_LT_OQ);
2177
2178             fscal            = felec;
2179
2180             fscal            = _mm256_and_ps(fscal,cutoff_mask);
2181
2182             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
2183
2184             /* Calculate temporary vectorial force */
2185             tx               = _mm256_mul_ps(fscal,dx22);
2186             ty               = _mm256_mul_ps(fscal,dy22);
2187             tz               = _mm256_mul_ps(fscal,dz22);
2188
2189             /* Update vectorial force */
2190             fix2             = _mm256_add_ps(fix2,tx);
2191             fiy2             = _mm256_add_ps(fiy2,ty);
2192             fiz2             = _mm256_add_ps(fiz2,tz);
2193
2194             fjx2             = _mm256_add_ps(fjx2,tx);
2195             fjy2             = _mm256_add_ps(fjy2,ty);
2196             fjz2             = _mm256_add_ps(fjz2,tz);
2197
2198             }
2199
2200             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
2201             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
2202             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
2203             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
2204             fjptrE             = (jnrlistE>=0) ? f+j_coord_offsetE : scratch;
2205             fjptrF             = (jnrlistF>=0) ? f+j_coord_offsetF : scratch;
2206             fjptrG             = (jnrlistG>=0) ? f+j_coord_offsetG : scratch;
2207             fjptrH             = (jnrlistH>=0) ? f+j_coord_offsetH : scratch;
2208
2209             gmx_mm256_decrement_3rvec_8ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjptrE,fjptrF,fjptrG,fjptrH,
2210                                                       fjx0,fjy0,fjz0,fjx1,fjy1,fjz1,fjx2,fjy2,fjz2);
2211
2212             /* Inner loop uses 277 flops */
2213         }
2214
2215         /* End of innermost loop */
2216
2217         gmx_mm256_update_iforce_3atom_swizzle_ps(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,
2218                                                  f+i_coord_offset,fshift+i_shift_offset);
2219
2220         /* Increment number of inner iterations */
2221         inneriter                  += j_index_end - j_index_start;
2222
2223         /* Outer loop uses 18 flops */
2224     }
2225
2226     /* Increment number of outer iterations */
2227     outeriter        += nri;
2228
2229     /* Update outer/inner flops */
2230
2231     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W3W3_F,outeriter*18 + inneriter*277);
2232 }