src/gmxlib/nonbonded/nb_kernel_avx_256_single/nb_kernel_ElecEwSh_VdwLJSh_GeomW4W4_avx_256_single.c

   1 /*
   2  * Note: this file was generated by the Gromacs avx_256_single kernel generator.
   3  *
   4  *                This source code is part of
   5  *
   6  *                 G   R   O   M   A   C   S
   7  *
   8  * Copyright (c) 2001-2012, The GROMACS Development Team
   9  *
  10  * Gromacs is a library for molecular simulation and trajectory analysis,
  11  * written by Erik Lindahl, David van der Spoel, Berk Hess, and others - for
  12  * a full list of developers and information, check out http://www.gromacs.org
  13  *
  14  * This program is free software; you can redistribute it and/or modify it under
  15  * the terms of the GNU Lesser General Public License as published by the Free
  16  * Software Foundation; either version 2 of the License, or (at your option) any
  17  * later version.
  18  *
  19  * To help fund GROMACS development, we humbly ask that you cite
  20  * the papers people have written on it - you can find them on the website.
  21  */
  22 #ifdef HAVE_CONFIG_H
  23 #include <config.h>
  24 #endif
  25
  26 #include <math.h>
  27
  28 #include "../nb_kernel.h"
  29 #include "types/simple.h"
  30 #include "vec.h"
  31 #include "nrnb.h"
  32
  33 #include "gmx_math_x86_avx_256_single.h"
  34 #include "kernelutil_x86_avx_256_single.h"
  35
  36 /*
  37  * Gromacs nonbonded kernel:   nb_kernel_ElecEwSh_VdwLJSh_GeomW4W4_VF_avx_256_single
  38  * Electrostatics interaction: Ewald
  39  * VdW interaction:            LennardJones
  40  * Geometry:                   Water4-Water4
  41  * Calculate force/pot:        PotentialAndForce
  42  */
  43 void
  44 nb_kernel_ElecEwSh_VdwLJSh_GeomW4W4_VF_avx_256_single
  45                     (t_nblist * gmx_restrict                nlist,
  46                      rvec * gmx_restrict                    xx,
  47                      rvec * gmx_restrict                    ff,
  48                      t_forcerec * gmx_restrict              fr,
  49                      t_mdatoms * gmx_restrict               mdatoms,
  50                      nb_kernel_data_t * gmx_restrict        kernel_data,
  51                      t_nrnb * gmx_restrict                  nrnb)
  52 {
  53     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
  54      * just 0 for non-waters.
  55      * Suffixes A,B,C,D,E,F,G,H refer to j loop unrolling done with AVX, e.g. for the eight different
  56      * jnr indices corresponding to data put in the four positions in the SIMD register.
  57      */
  58     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
  59     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
  60     int              jnrA,jnrB,jnrC,jnrD;
  61     int              jnrE,jnrF,jnrG,jnrH;
  62     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
  63     int              jnrlistE,jnrlistF,jnrlistG,jnrlistH;
  64     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
  65     int              j_coord_offsetE,j_coord_offsetF,j_coord_offsetG,j_coord_offsetH;
  66     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
  67     real             rcutoff_scalar;
  68     real             *shiftvec,*fshift,*x,*f;
  69     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD,*fjptrE,*fjptrF,*fjptrG,*fjptrH;
  70     real             scratch[4*DIM];
  71     __m256           tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
  72     real *           vdwioffsetptr0;
  73     __m256           ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
  74     real *           vdwioffsetptr1;
  75     __m256           ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
  76     real *           vdwioffsetptr2;
  77     __m256           ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
  78     real *           vdwioffsetptr3;
  79     __m256           ix3,iy3,iz3,fix3,fiy3,fiz3,iq3,isai3;
  80     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D,vdwjidx0E,vdwjidx0F,vdwjidx0G,vdwjidx0H;
  81     __m256           jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
  82     int              vdwjidx1A,vdwjidx1B,vdwjidx1C,vdwjidx1D,vdwjidx1E,vdwjidx1F,vdwjidx1G,vdwjidx1H;
  83     __m256           jx1,jy1,jz1,fjx1,fjy1,fjz1,jq1,isaj1;
  84     int              vdwjidx2A,vdwjidx2B,vdwjidx2C,vdwjidx2D,vdwjidx2E,vdwjidx2F,vdwjidx2G,vdwjidx2H;
  85     __m256           jx2,jy2,jz2,fjx2,fjy2,fjz2,jq2,isaj2;
  86     int              vdwjidx3A,vdwjidx3B,vdwjidx3C,vdwjidx3D,vdwjidx3E,vdwjidx3F,vdwjidx3G,vdwjidx3H;
  87     __m256           jx3,jy3,jz3,fjx3,fjy3,fjz3,jq3,isaj3;
  88     __m256           dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
  89     __m256           dx11,dy11,dz11,rsq11,rinv11,rinvsq11,r11,qq11,c6_11,c12_11;
  90     __m256           dx12,dy12,dz12,rsq12,rinv12,rinvsq12,r12,qq12,c6_12,c12_12;
  91     __m256           dx13,dy13,dz13,rsq13,rinv13,rinvsq13,r13,qq13,c6_13,c12_13;
  92     __m256           dx21,dy21,dz21,rsq21,rinv21,rinvsq21,r21,qq21,c6_21,c12_21;
  93     __m256           dx22,dy22,dz22,rsq22,rinv22,rinvsq22,r22,qq22,c6_22,c12_22;
  94     __m256           dx23,dy23,dz23,rsq23,rinv23,rinvsq23,r23,qq23,c6_23,c12_23;
  95     __m256           dx31,dy31,dz31,rsq31,rinv31,rinvsq31,r31,qq31,c6_31,c12_31;
  96     __m256           dx32,dy32,dz32,rsq32,rinv32,rinvsq32,r32,qq32,c6_32,c12_32;
  97     __m256           dx33,dy33,dz33,rsq33,rinv33,rinvsq33,r33,qq33,c6_33,c12_33;
  98     __m256           velec,felec,velecsum,facel,crf,krf,krf2;
  99     real             *charge;
 100     int              nvdwtype;
 101     __m256           rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
 102     int              *vdwtype;
 103     real             *vdwparam;
 104     __m256           one_sixth   = _mm256_set1_ps(1.0/6.0);
 105     __m256           one_twelfth = _mm256_set1_ps(1.0/12.0);
 106     __m256i          ewitab;
 107     __m128i          ewitab_lo,ewitab_hi;
 108     __m256           ewtabscale,eweps,sh_ewald,ewrt,ewtabhalfspace,ewtabF,ewtabFn,ewtabD,ewtabV;
 109     __m256           beta,beta2,beta3,zeta2,pmecorrF,pmecorrV,rinv3;
 110     real             *ewtab;
 111     __m256           dummy_mask,cutoff_mask;
 112     __m256           signbit = _mm256_castsi256_ps( _mm256_set1_epi32(0x80000000) );
 113     __m256           one     = _mm256_set1_ps(1.0);
 114     __m256           two     = _mm256_set1_ps(2.0);
 115     x                = xx[0];
 116     f                = ff[0];
 117
 118     nri              = nlist->nri;
 119     iinr             = nlist->iinr;
 120     jindex           = nlist->jindex;
 121     jjnr             = nlist->jjnr;
 122     shiftidx         = nlist->shift;
 123     gid              = nlist->gid;
 124     shiftvec         = fr->shift_vec[0];
 125     fshift           = fr->fshift[0];
 126     facel            = _mm256_set1_ps(fr->epsfac);
 127     charge           = mdatoms->chargeA;
 128     nvdwtype         = fr->ntype;
 129     vdwparam         = fr->nbfp;
 130     vdwtype          = mdatoms->typeA;
 131
 132     sh_ewald         = _mm256_set1_ps(fr->ic->sh_ewald);
 133     beta             = _mm256_set1_ps(fr->ic->ewaldcoeff);
 134     beta2            = _mm256_mul_ps(beta,beta);
 135     beta3            = _mm256_mul_ps(beta,beta2);
 136
 137     ewtab            = fr->ic->tabq_coul_FDV0;
 138     ewtabscale       = _mm256_set1_ps(fr->ic->tabq_scale);
 139     ewtabhalfspace   = _mm256_set1_ps(0.5/fr->ic->tabq_scale);
 140
 141     /* Setup water-specific parameters */
 142     inr              = nlist->iinr[0];
 143     iq1              = _mm256_mul_ps(facel,_mm256_set1_ps(charge[inr+1]));
 144     iq2              = _mm256_mul_ps(facel,_mm256_set1_ps(charge[inr+2]));
 145     iq3              = _mm256_mul_ps(facel,_mm256_set1_ps(charge[inr+3]));
 146     vdwioffsetptr0   = vdwparam+2*nvdwtype*vdwtype[inr+0];
 147
 148     jq1              = _mm256_set1_ps(charge[inr+1]);
 149     jq2              = _mm256_set1_ps(charge[inr+2]);
 150     jq3              = _mm256_set1_ps(charge[inr+3]);
 151     vdwjidx0A        = 2*vdwtype[inr+0];
 152     c6_00            = _mm256_set1_ps(vdwioffsetptr0[vdwjidx0A]);
 153     c12_00           = _mm256_set1_ps(vdwioffsetptr0[vdwjidx0A+1]);
 154     qq11             = _mm256_mul_ps(iq1,jq1);
 155     qq12             = _mm256_mul_ps(iq1,jq2);
 156     qq13             = _mm256_mul_ps(iq1,jq3);
 157     qq21             = _mm256_mul_ps(iq2,jq1);
 158     qq22             = _mm256_mul_ps(iq2,jq2);
 159     qq23             = _mm256_mul_ps(iq2,jq3);
 160     qq31             = _mm256_mul_ps(iq3,jq1);
 161     qq32             = _mm256_mul_ps(iq3,jq2);
 162     qq33             = _mm256_mul_ps(iq3,jq3);
 163
 164     /* When we use explicit cutoffs the value must be identical for elec and VdW, so use elec as an arbitrary choice */
 165     rcutoff_scalar   = fr->rcoulomb;
 166     rcutoff          = _mm256_set1_ps(rcutoff_scalar);
 167     rcutoff2         = _mm256_mul_ps(rcutoff,rcutoff);
 168
 169     sh_vdw_invrcut6  = _mm256_set1_ps(fr->ic->sh_invrc6);
 170     rvdw             = _mm256_set1_ps(fr->rvdw);
 171
 172     /* Avoid stupid compiler warnings */
 173     jnrA = jnrB = jnrC = jnrD = jnrE = jnrF = jnrG = jnrH = 0;
 174     j_coord_offsetA = 0;
 175     j_coord_offsetB = 0;
 176     j_coord_offsetC = 0;
 177     j_coord_offsetD = 0;
 178     j_coord_offsetE = 0;
 179     j_coord_offsetF = 0;
 180     j_coord_offsetG = 0;
 181     j_coord_offsetH = 0;
 182
 183     outeriter        = 0;
 184     inneriter        = 0;
 185
 186     for(iidx=0;iidx<4*DIM;iidx++)
 187     {
 188         scratch[iidx] = 0.0;
 189     }
 190
 191     /* Start outer loop over neighborlists */
 192     for(iidx=0; iidx<nri; iidx++)
 193     {
 194         /* Load shift vector for this list */
 195         i_shift_offset   = DIM*shiftidx[iidx];
 196
 197         /* Load limits for loop over neighbors */
 198         j_index_start    = jindex[iidx];
 199         j_index_end      = jindex[iidx+1];
 200
 201         /* Get outer coordinate index */
 202         inr              = iinr[iidx];
 203         i_coord_offset   = DIM*inr;
 204
 205         /* Load i particle coords and add shift vector */
 206         gmx_mm256_load_shift_and_4rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,
 207                                                     &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2,&ix3,&iy3,&iz3);
 208
 209         fix0             = _mm256_setzero_ps();
 210         fiy0             = _mm256_setzero_ps();
 211         fiz0             = _mm256_setzero_ps();
 212         fix1             = _mm256_setzero_ps();
 213         fiy1             = _mm256_setzero_ps();
 214         fiz1             = _mm256_setzero_ps();
 215         fix2             = _mm256_setzero_ps();
 216         fiy2             = _mm256_setzero_ps();
 217         fiz2             = _mm256_setzero_ps();
 218         fix3             = _mm256_setzero_ps();
 219         fiy3             = _mm256_setzero_ps();
 220         fiz3             = _mm256_setzero_ps();
 221
 222         /* Reset potential sums */
 223         velecsum         = _mm256_setzero_ps();
 224         vvdwsum          = _mm256_setzero_ps();
 225
 226         /* Start inner kernel loop */
 227         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+7]>=0; jidx+=8)
 228         {
 229
 230             /* Get j neighbor index, and coordinate index */
 231             jnrA             = jjnr[jidx];
 232             jnrB             = jjnr[jidx+1];
 233             jnrC             = jjnr[jidx+2];
 234             jnrD             = jjnr[jidx+3];
 235             jnrE             = jjnr[jidx+4];
 236             jnrF             = jjnr[jidx+5];
 237             jnrG             = jjnr[jidx+6];
 238             jnrH             = jjnr[jidx+7];
 239             j_coord_offsetA  = DIM*jnrA;
 240             j_coord_offsetB  = DIM*jnrB;
 241             j_coord_offsetC  = DIM*jnrC;
 242             j_coord_offsetD  = DIM*jnrD;
 243             j_coord_offsetE  = DIM*jnrE;
 244             j_coord_offsetF  = DIM*jnrF;
 245             j_coord_offsetG  = DIM*jnrG;
 246             j_coord_offsetH  = DIM*jnrH;
 247
 248             /* load j atom coordinates */
 249             gmx_mm256_load_4rvec_8ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 250                                                  x+j_coord_offsetC,x+j_coord_offsetD,
 251                                                  x+j_coord_offsetE,x+j_coord_offsetF,
 252                                                  x+j_coord_offsetG,x+j_coord_offsetH,
 253                                                  &jx0,&jy0,&jz0,&jx1,&jy1,&jz1,&jx2,
 254                                                  &jy2,&jz2,&jx3,&jy3,&jz3);
 255
 256             /* Calculate displacement vector */
 257             dx00             = _mm256_sub_ps(ix0,jx0);
 258             dy00             = _mm256_sub_ps(iy0,jy0);
 259             dz00             = _mm256_sub_ps(iz0,jz0);
 260             dx11             = _mm256_sub_ps(ix1,jx1);
 261             dy11             = _mm256_sub_ps(iy1,jy1);
 262             dz11             = _mm256_sub_ps(iz1,jz1);
 263             dx12             = _mm256_sub_ps(ix1,jx2);
 264             dy12             = _mm256_sub_ps(iy1,jy2);
 265             dz12             = _mm256_sub_ps(iz1,jz2);
 266             dx13             = _mm256_sub_ps(ix1,jx3);
 267             dy13             = _mm256_sub_ps(iy1,jy3);
 268             dz13             = _mm256_sub_ps(iz1,jz3);
 269             dx21             = _mm256_sub_ps(ix2,jx1);
 270             dy21             = _mm256_sub_ps(iy2,jy1);
 271             dz21             = _mm256_sub_ps(iz2,jz1);
 272             dx22             = _mm256_sub_ps(ix2,jx2);
 273             dy22             = _mm256_sub_ps(iy2,jy2);
 274             dz22             = _mm256_sub_ps(iz2,jz2);
 275             dx23             = _mm256_sub_ps(ix2,jx3);
 276             dy23             = _mm256_sub_ps(iy2,jy3);
 277             dz23             = _mm256_sub_ps(iz2,jz3);
 278             dx31             = _mm256_sub_ps(ix3,jx1);
 279             dy31             = _mm256_sub_ps(iy3,jy1);
 280             dz31             = _mm256_sub_ps(iz3,jz1);
 281             dx32             = _mm256_sub_ps(ix3,jx2);
 282             dy32             = _mm256_sub_ps(iy3,jy2);
 283             dz32             = _mm256_sub_ps(iz3,jz2);
 284             dx33             = _mm256_sub_ps(ix3,jx3);
 285             dy33             = _mm256_sub_ps(iy3,jy3);
 286             dz33             = _mm256_sub_ps(iz3,jz3);
 287
 288             /* Calculate squared distance and things based on it */
 289             rsq00            = gmx_mm256_calc_rsq_ps(dx00,dy00,dz00);
 290             rsq11            = gmx_mm256_calc_rsq_ps(dx11,dy11,dz11);
 291             rsq12            = gmx_mm256_calc_rsq_ps(dx12,dy12,dz12);
 292             rsq13            = gmx_mm256_calc_rsq_ps(dx13,dy13,dz13);
 293             rsq21            = gmx_mm256_calc_rsq_ps(dx21,dy21,dz21);
 294             rsq22            = gmx_mm256_calc_rsq_ps(dx22,dy22,dz22);
 295             rsq23            = gmx_mm256_calc_rsq_ps(dx23,dy23,dz23);
 296             rsq31            = gmx_mm256_calc_rsq_ps(dx31,dy31,dz31);
 297             rsq32            = gmx_mm256_calc_rsq_ps(dx32,dy32,dz32);
 298             rsq33            = gmx_mm256_calc_rsq_ps(dx33,dy33,dz33);
 299
 300             rinv11           = gmx_mm256_invsqrt_ps(rsq11);
 301             rinv12           = gmx_mm256_invsqrt_ps(rsq12);
 302             rinv13           = gmx_mm256_invsqrt_ps(rsq13);
 303             rinv21           = gmx_mm256_invsqrt_ps(rsq21);
 304             rinv22           = gmx_mm256_invsqrt_ps(rsq22);
 305             rinv23           = gmx_mm256_invsqrt_ps(rsq23);
 306             rinv31           = gmx_mm256_invsqrt_ps(rsq31);
 307             rinv32           = gmx_mm256_invsqrt_ps(rsq32);
 308             rinv33           = gmx_mm256_invsqrt_ps(rsq33);
 309
 310             rinvsq00         = gmx_mm256_inv_ps(rsq00);
 311             rinvsq11         = _mm256_mul_ps(rinv11,rinv11);
 312             rinvsq12         = _mm256_mul_ps(rinv12,rinv12);
 313             rinvsq13         = _mm256_mul_ps(rinv13,rinv13);
 314             rinvsq21         = _mm256_mul_ps(rinv21,rinv21);
 315             rinvsq22         = _mm256_mul_ps(rinv22,rinv22);
 316             rinvsq23         = _mm256_mul_ps(rinv23,rinv23);
 317             rinvsq31         = _mm256_mul_ps(rinv31,rinv31);
 318             rinvsq32         = _mm256_mul_ps(rinv32,rinv32);
 319             rinvsq33         = _mm256_mul_ps(rinv33,rinv33);
 320
 321             fjx0             = _mm256_setzero_ps();
 322             fjy0             = _mm256_setzero_ps();
 323             fjz0             = _mm256_setzero_ps();
 324             fjx1             = _mm256_setzero_ps();
 325             fjy1             = _mm256_setzero_ps();
 326             fjz1             = _mm256_setzero_ps();
 327             fjx2             = _mm256_setzero_ps();
 328             fjy2             = _mm256_setzero_ps();
 329             fjz2             = _mm256_setzero_ps();
 330             fjx3             = _mm256_setzero_ps();
 331             fjy3             = _mm256_setzero_ps();
 332             fjz3             = _mm256_setzero_ps();
 333
 334             /**************************
 335              * CALCULATE INTERACTIONS *
 336              **************************/
 337
 338             if (gmx_mm256_any_lt(rsq00,rcutoff2))
 339             {
 340
 341             /* LENNARD-JONES DISPERSION/REPULSION */
 342
 343             rinvsix          = _mm256_mul_ps(_mm256_mul_ps(rinvsq00,rinvsq00),rinvsq00);
 344             vvdw6            = _mm256_mul_ps(c6_00,rinvsix);
 345             vvdw12           = _mm256_mul_ps(c12_00,_mm256_mul_ps(rinvsix,rinvsix));
 346             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) ,
 347                                           _mm256_mul_ps( _mm256_sub_ps(vvdw6,_mm256_mul_ps(c6_00,sh_vdw_invrcut6)),one_sixth));
 348             fvdw             = _mm256_mul_ps(_mm256_sub_ps(vvdw12,vvdw6),rinvsq00);
 349
 350             cutoff_mask      = _mm256_cmp_ps(rsq00,rcutoff2,_CMP_LT_OQ);
 351
 352             /* Update potential sum for this i atom from the interaction with this j atom. */
 353             vvdw             = _mm256_and_ps(vvdw,cutoff_mask);
 354             vvdwsum          = _mm256_add_ps(vvdwsum,vvdw);
 355
 356             fscal            = fvdw;
 357
 358             fscal            = _mm256_and_ps(fscal,cutoff_mask);
 359
 360             /* Calculate temporary vectorial force */
 361             tx               = _mm256_mul_ps(fscal,dx00);
 362             ty               = _mm256_mul_ps(fscal,dy00);
 363             tz               = _mm256_mul_ps(fscal,dz00);
 364
 365             /* Update vectorial force */
 366             fix0             = _mm256_add_ps(fix0,tx);
 367             fiy0             = _mm256_add_ps(fiy0,ty);
 368             fiz0             = _mm256_add_ps(fiz0,tz);
 369
 370             fjx0             = _mm256_add_ps(fjx0,tx);
 371             fjy0             = _mm256_add_ps(fjy0,ty);
 372             fjz0             = _mm256_add_ps(fjz0,tz);
 373
 374             }
 375
 376             /**************************
 377              * CALCULATE INTERACTIONS *
 378              **************************/
 379
 380             if (gmx_mm256_any_lt(rsq11,rcutoff2))
 381             {
 382
 383             r11              = _mm256_mul_ps(rsq11,rinv11);
 384
 385             /* EWALD ELECTROSTATICS */
 386
 387             /* Analytical PME correction */
 388             zeta2            = _mm256_mul_ps(beta2,rsq11);
 389             rinv3            = _mm256_mul_ps(rinvsq11,rinv11);
 390             pmecorrF         = gmx_mm256_pmecorrF_ps(zeta2);
 391             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
 392             felec            = _mm256_mul_ps(qq11,felec);
 393             pmecorrV         = gmx_mm256_pmecorrV_ps(zeta2);
 394             pmecorrV         = _mm256_mul_ps(pmecorrV,beta);
 395             velec            = _mm256_sub_ps(_mm256_sub_ps(rinv11,sh_ewald),pmecorrV);
 396             velec            = _mm256_mul_ps(qq11,velec);
 397
 398             cutoff_mask      = _mm256_cmp_ps(rsq11,rcutoff2,_CMP_LT_OQ);
 399
 400             /* Update potential sum for this i atom from the interaction with this j atom. */
 401             velec            = _mm256_and_ps(velec,cutoff_mask);
 402             velecsum         = _mm256_add_ps(velecsum,velec);
 403
 404             fscal            = felec;
 405
 406             fscal            = _mm256_and_ps(fscal,cutoff_mask);
 407
 408             /* Calculate temporary vectorial force */
 409             tx               = _mm256_mul_ps(fscal,dx11);
 410             ty               = _mm256_mul_ps(fscal,dy11);
 411             tz               = _mm256_mul_ps(fscal,dz11);
 412
 413             /* Update vectorial force */
 414             fix1             = _mm256_add_ps(fix1,tx);
 415             fiy1             = _mm256_add_ps(fiy1,ty);
 416             fiz1             = _mm256_add_ps(fiz1,tz);
 417
 418             fjx1             = _mm256_add_ps(fjx1,tx);
 419             fjy1             = _mm256_add_ps(fjy1,ty);
 420             fjz1             = _mm256_add_ps(fjz1,tz);
 421
 422             }
 423
 424             /**************************
 425              * CALCULATE INTERACTIONS *
 426              **************************/
 427
 428             if (gmx_mm256_any_lt(rsq12,rcutoff2))
 429             {
 430
 431             r12              = _mm256_mul_ps(rsq12,rinv12);
 432
 433             /* EWALD ELECTROSTATICS */
 434
 435             /* Analytical PME correction */
 436             zeta2            = _mm256_mul_ps(beta2,rsq12);
 437             rinv3            = _mm256_mul_ps(rinvsq12,rinv12);
 438             pmecorrF         = gmx_mm256_pmecorrF_ps(zeta2);
 439             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
 440             felec            = _mm256_mul_ps(qq12,felec);
 441             pmecorrV         = gmx_mm256_pmecorrV_ps(zeta2);
 442             pmecorrV         = _mm256_mul_ps(pmecorrV,beta);
 443             velec            = _mm256_sub_ps(_mm256_sub_ps(rinv12,sh_ewald),pmecorrV);
 444             velec            = _mm256_mul_ps(qq12,velec);
 445
 446             cutoff_mask      = _mm256_cmp_ps(rsq12,rcutoff2,_CMP_LT_OQ);
 447
 448             /* Update potential sum for this i atom from the interaction with this j atom. */
 449             velec            = _mm256_and_ps(velec,cutoff_mask);
 450             velecsum         = _mm256_add_ps(velecsum,velec);
 451
 452             fscal            = felec;
 453
 454             fscal            = _mm256_and_ps(fscal,cutoff_mask);
 455
 456             /* Calculate temporary vectorial force */
 457             tx               = _mm256_mul_ps(fscal,dx12);
 458             ty               = _mm256_mul_ps(fscal,dy12);
 459             tz               = _mm256_mul_ps(fscal,dz12);
 460
 461             /* Update vectorial force */
 462             fix1             = _mm256_add_ps(fix1,tx);
 463             fiy1             = _mm256_add_ps(fiy1,ty);
 464             fiz1             = _mm256_add_ps(fiz1,tz);
 465
 466             fjx2             = _mm256_add_ps(fjx2,tx);
 467             fjy2             = _mm256_add_ps(fjy2,ty);
 468             fjz2             = _mm256_add_ps(fjz2,tz);
 469
 470             }
 471
 472             /**************************
 473              * CALCULATE INTERACTIONS *
 474              **************************/
 475
 476             if (gmx_mm256_any_lt(rsq13,rcutoff2))
 477             {
 478
 479             r13              = _mm256_mul_ps(rsq13,rinv13);
 480
 481             /* EWALD ELECTROSTATICS */
 482
 483             /* Analytical PME correction */
 484             zeta2            = _mm256_mul_ps(beta2,rsq13);
 485             rinv3            = _mm256_mul_ps(rinvsq13,rinv13);
 486             pmecorrF         = gmx_mm256_pmecorrF_ps(zeta2);
 487             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
 488             felec            = _mm256_mul_ps(qq13,felec);
 489             pmecorrV         = gmx_mm256_pmecorrV_ps(zeta2);
 490             pmecorrV         = _mm256_mul_ps(pmecorrV,beta);
 491             velec            = _mm256_sub_ps(_mm256_sub_ps(rinv13,sh_ewald),pmecorrV);
 492             velec            = _mm256_mul_ps(qq13,velec);
 493
 494             cutoff_mask      = _mm256_cmp_ps(rsq13,rcutoff2,_CMP_LT_OQ);
 495
 496             /* Update potential sum for this i atom from the interaction with this j atom. */
 497             velec            = _mm256_and_ps(velec,cutoff_mask);
 498             velecsum         = _mm256_add_ps(velecsum,velec);
 499
 500             fscal            = felec;
 501
 502             fscal            = _mm256_and_ps(fscal,cutoff_mask);
 503
 504             /* Calculate temporary vectorial force */
 505             tx               = _mm256_mul_ps(fscal,dx13);
 506             ty               = _mm256_mul_ps(fscal,dy13);
 507             tz               = _mm256_mul_ps(fscal,dz13);
 508
 509             /* Update vectorial force */
 510             fix1             = _mm256_add_ps(fix1,tx);
 511             fiy1             = _mm256_add_ps(fiy1,ty);
 512             fiz1             = _mm256_add_ps(fiz1,tz);
 513
 514             fjx3             = _mm256_add_ps(fjx3,tx);
 515             fjy3             = _mm256_add_ps(fjy3,ty);
 516             fjz3             = _mm256_add_ps(fjz3,tz);
 517
 518             }
 519
 520             /**************************
 521              * CALCULATE INTERACTIONS *
 522              **************************/
 523
 524             if (gmx_mm256_any_lt(rsq21,rcutoff2))
 525             {
 526
 527             r21              = _mm256_mul_ps(rsq21,rinv21);
 528
 529             /* EWALD ELECTROSTATICS */
 530
 531             /* Analytical PME correction */
 532             zeta2            = _mm256_mul_ps(beta2,rsq21);
 533             rinv3            = _mm256_mul_ps(rinvsq21,rinv21);
 534             pmecorrF         = gmx_mm256_pmecorrF_ps(zeta2);
 535             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
 536             felec            = _mm256_mul_ps(qq21,felec);
 537             pmecorrV         = gmx_mm256_pmecorrV_ps(zeta2);
 538             pmecorrV         = _mm256_mul_ps(pmecorrV,beta);
 539             velec            = _mm256_sub_ps(_mm256_sub_ps(rinv21,sh_ewald),pmecorrV);
 540             velec            = _mm256_mul_ps(qq21,velec);
 541
 542             cutoff_mask      = _mm256_cmp_ps(rsq21,rcutoff2,_CMP_LT_OQ);
 543
 544             /* Update potential sum for this i atom from the interaction with this j atom. */
 545             velec            = _mm256_and_ps(velec,cutoff_mask);
 546             velecsum         = _mm256_add_ps(velecsum,velec);
 547
 548             fscal            = felec;
 549
 550             fscal            = _mm256_and_ps(fscal,cutoff_mask);
 551
 552             /* Calculate temporary vectorial force */
 553             tx               = _mm256_mul_ps(fscal,dx21);
 554             ty               = _mm256_mul_ps(fscal,dy21);
 555             tz               = _mm256_mul_ps(fscal,dz21);
 556
 557             /* Update vectorial force */
 558             fix2             = _mm256_add_ps(fix2,tx);
 559             fiy2             = _mm256_add_ps(fiy2,ty);
 560             fiz2             = _mm256_add_ps(fiz2,tz);
 561
 562             fjx1             = _mm256_add_ps(fjx1,tx);
 563             fjy1             = _mm256_add_ps(fjy1,ty);
 564             fjz1             = _mm256_add_ps(fjz1,tz);
 565
 566             }
 567
 568             /**************************
 569              * CALCULATE INTERACTIONS *
 570              **************************/
 571
 572             if (gmx_mm256_any_lt(rsq22,rcutoff2))
 573             {
 574
 575             r22              = _mm256_mul_ps(rsq22,rinv22);
 576
 577             /* EWALD ELECTROSTATICS */
 578
 579             /* Analytical PME correction */
 580             zeta2            = _mm256_mul_ps(beta2,rsq22);
 581             rinv3            = _mm256_mul_ps(rinvsq22,rinv22);
 582             pmecorrF         = gmx_mm256_pmecorrF_ps(zeta2);
 583             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
 584             felec            = _mm256_mul_ps(qq22,felec);
 585             pmecorrV         = gmx_mm256_pmecorrV_ps(zeta2);
 586             pmecorrV         = _mm256_mul_ps(pmecorrV,beta);
 587             velec            = _mm256_sub_ps(_mm256_sub_ps(rinv22,sh_ewald),pmecorrV);
 588             velec            = _mm256_mul_ps(qq22,velec);
 589
 590             cutoff_mask      = _mm256_cmp_ps(rsq22,rcutoff2,_CMP_LT_OQ);
 591
 592             /* Update potential sum for this i atom from the interaction with this j atom. */
 593             velec            = _mm256_and_ps(velec,cutoff_mask);
 594             velecsum         = _mm256_add_ps(velecsum,velec);
 595
 596             fscal            = felec;
 597
 598             fscal            = _mm256_and_ps(fscal,cutoff_mask);
 599
 600             /* Calculate temporary vectorial force */
 601             tx               = _mm256_mul_ps(fscal,dx22);
 602             ty               = _mm256_mul_ps(fscal,dy22);
 603             tz               = _mm256_mul_ps(fscal,dz22);
 604
 605             /* Update vectorial force */
 606             fix2             = _mm256_add_ps(fix2,tx);
 607             fiy2             = _mm256_add_ps(fiy2,ty);
 608             fiz2             = _mm256_add_ps(fiz2,tz);
 609
 610             fjx2             = _mm256_add_ps(fjx2,tx);
 611             fjy2             = _mm256_add_ps(fjy2,ty);
 612             fjz2             = _mm256_add_ps(fjz2,tz);
 613
 614             }
 615
 616             /**************************
 617              * CALCULATE INTERACTIONS *
 618              **************************/
 619
 620             if (gmx_mm256_any_lt(rsq23,rcutoff2))
 621             {
 622
 623             r23              = _mm256_mul_ps(rsq23,rinv23);
 624
 625             /* EWALD ELECTROSTATICS */
 626
 627             /* Analytical PME correction */
 628             zeta2            = _mm256_mul_ps(beta2,rsq23);
 629             rinv3            = _mm256_mul_ps(rinvsq23,rinv23);
 630             pmecorrF         = gmx_mm256_pmecorrF_ps(zeta2);
 631             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
 632             felec            = _mm256_mul_ps(qq23,felec);
 633             pmecorrV         = gmx_mm256_pmecorrV_ps(zeta2);
 634             pmecorrV         = _mm256_mul_ps(pmecorrV,beta);
 635             velec            = _mm256_sub_ps(_mm256_sub_ps(rinv23,sh_ewald),pmecorrV);
 636             velec            = _mm256_mul_ps(qq23,velec);
 637
 638             cutoff_mask      = _mm256_cmp_ps(rsq23,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,dx23);
 650             ty               = _mm256_mul_ps(fscal,dy23);
 651             tz               = _mm256_mul_ps(fscal,dz23);
 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             fjx3             = _mm256_add_ps(fjx3,tx);
 659             fjy3             = _mm256_add_ps(fjy3,ty);
 660             fjz3             = _mm256_add_ps(fjz3,tz);
 661
 662             }
 663
 664             /**************************
 665              * CALCULATE INTERACTIONS *
 666              **************************/
 667
 668             if (gmx_mm256_any_lt(rsq31,rcutoff2))
 669             {
 670
 671             r31              = _mm256_mul_ps(rsq31,rinv31);
 672
 673             /* EWALD ELECTROSTATICS */
 674
 675             /* Analytical PME correction */
 676             zeta2            = _mm256_mul_ps(beta2,rsq31);
 677             rinv3            = _mm256_mul_ps(rinvsq31,rinv31);
 678             pmecorrF         = gmx_mm256_pmecorrF_ps(zeta2);
 679             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
 680             felec            = _mm256_mul_ps(qq31,felec);
 681             pmecorrV         = gmx_mm256_pmecorrV_ps(zeta2);
 682             pmecorrV         = _mm256_mul_ps(pmecorrV,beta);
 683             velec            = _mm256_sub_ps(_mm256_sub_ps(rinv31,sh_ewald),pmecorrV);
 684             velec            = _mm256_mul_ps(qq31,velec);
 685
 686             cutoff_mask      = _mm256_cmp_ps(rsq31,rcutoff2,_CMP_LT_OQ);
 687
 688             /* Update potential sum for this i atom from the interaction with this j atom. */
 689             velec            = _mm256_and_ps(velec,cutoff_mask);
 690             velecsum         = _mm256_add_ps(velecsum,velec);
 691
 692             fscal            = felec;
 693
 694             fscal            = _mm256_and_ps(fscal,cutoff_mask);
 695
 696             /* Calculate temporary vectorial force */
 697             tx               = _mm256_mul_ps(fscal,dx31);
 698             ty               = _mm256_mul_ps(fscal,dy31);
 699             tz               = _mm256_mul_ps(fscal,dz31);
 700
 701             /* Update vectorial force */
 702             fix3             = _mm256_add_ps(fix3,tx);
 703             fiy3             = _mm256_add_ps(fiy3,ty);
 704             fiz3             = _mm256_add_ps(fiz3,tz);
 705
 706             fjx1             = _mm256_add_ps(fjx1,tx);
 707             fjy1             = _mm256_add_ps(fjy1,ty);
 708             fjz1             = _mm256_add_ps(fjz1,tz);
 709
 710             }
 711
 712             /**************************
 713              * CALCULATE INTERACTIONS *
 714              **************************/
 715
 716             if (gmx_mm256_any_lt(rsq32,rcutoff2))
 717             {
 718
 719             r32              = _mm256_mul_ps(rsq32,rinv32);
 720
 721             /* EWALD ELECTROSTATICS */
 722
 723             /* Analytical PME correction */
 724             zeta2            = _mm256_mul_ps(beta2,rsq32);
 725             rinv3            = _mm256_mul_ps(rinvsq32,rinv32);
 726             pmecorrF         = gmx_mm256_pmecorrF_ps(zeta2);
 727             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
 728             felec            = _mm256_mul_ps(qq32,felec);
 729             pmecorrV         = gmx_mm256_pmecorrV_ps(zeta2);
 730             pmecorrV         = _mm256_mul_ps(pmecorrV,beta);
 731             velec            = _mm256_sub_ps(_mm256_sub_ps(rinv32,sh_ewald),pmecorrV);
 732             velec            = _mm256_mul_ps(qq32,velec);
 733
 734             cutoff_mask      = _mm256_cmp_ps(rsq32,rcutoff2,_CMP_LT_OQ);
 735
 736             /* Update potential sum for this i atom from the interaction with this j atom. */
 737             velec            = _mm256_and_ps(velec,cutoff_mask);
 738             velecsum         = _mm256_add_ps(velecsum,velec);
 739
 740             fscal            = felec;
 741
 742             fscal            = _mm256_and_ps(fscal,cutoff_mask);
 743
 744             /* Calculate temporary vectorial force */
 745             tx               = _mm256_mul_ps(fscal,dx32);
 746             ty               = _mm256_mul_ps(fscal,dy32);
 747             tz               = _mm256_mul_ps(fscal,dz32);
 748
 749             /* Update vectorial force */
 750             fix3             = _mm256_add_ps(fix3,tx);
 751             fiy3             = _mm256_add_ps(fiy3,ty);
 752             fiz3             = _mm256_add_ps(fiz3,tz);
 753
 754             fjx2             = _mm256_add_ps(fjx2,tx);
 755             fjy2             = _mm256_add_ps(fjy2,ty);
 756             fjz2             = _mm256_add_ps(fjz2,tz);
 757
 758             }
 759
 760             /**************************
 761              * CALCULATE INTERACTIONS *
 762              **************************/
 763
 764             if (gmx_mm256_any_lt(rsq33,rcutoff2))
 765             {
 766
 767             r33              = _mm256_mul_ps(rsq33,rinv33);
 768
 769             /* EWALD ELECTROSTATICS */
 770
 771             /* Analytical PME correction */
 772             zeta2            = _mm256_mul_ps(beta2,rsq33);
 773             rinv3            = _mm256_mul_ps(rinvsq33,rinv33);
 774             pmecorrF         = gmx_mm256_pmecorrF_ps(zeta2);
 775             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
 776             felec            = _mm256_mul_ps(qq33,felec);
 777             pmecorrV         = gmx_mm256_pmecorrV_ps(zeta2);
 778             pmecorrV         = _mm256_mul_ps(pmecorrV,beta);
 779             velec            = _mm256_sub_ps(_mm256_sub_ps(rinv33,sh_ewald),pmecorrV);
 780             velec            = _mm256_mul_ps(qq33,velec);
 781
 782             cutoff_mask      = _mm256_cmp_ps(rsq33,rcutoff2,_CMP_LT_OQ);
 783
 784             /* Update potential sum for this i atom from the interaction with this j atom. */
 785             velec            = _mm256_and_ps(velec,cutoff_mask);
 786             velecsum         = _mm256_add_ps(velecsum,velec);
 787
 788             fscal            = felec;
 789
 790             fscal            = _mm256_and_ps(fscal,cutoff_mask);
 791
 792             /* Calculate temporary vectorial force */
 793             tx               = _mm256_mul_ps(fscal,dx33);
 794             ty               = _mm256_mul_ps(fscal,dy33);
 795             tz               = _mm256_mul_ps(fscal,dz33);
 796
 797             /* Update vectorial force */
 798             fix3             = _mm256_add_ps(fix3,tx);
 799             fiy3             = _mm256_add_ps(fiy3,ty);
 800             fiz3             = _mm256_add_ps(fiz3,tz);
 801
 802             fjx3             = _mm256_add_ps(fjx3,tx);
 803             fjy3             = _mm256_add_ps(fjy3,ty);
 804             fjz3             = _mm256_add_ps(fjz3,tz);
 805
 806             }
 807
 808             fjptrA             = f+j_coord_offsetA;
 809             fjptrB             = f+j_coord_offsetB;
 810             fjptrC             = f+j_coord_offsetC;
 811             fjptrD             = f+j_coord_offsetD;
 812             fjptrE             = f+j_coord_offsetE;
 813             fjptrF             = f+j_coord_offsetF;
 814             fjptrG             = f+j_coord_offsetG;
 815             fjptrH             = f+j_coord_offsetH;
 816
 817             gmx_mm256_decrement_4rvec_8ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjptrE,fjptrF,fjptrG,fjptrH,
 818                                                       fjx0,fjy0,fjz0,fjx1,fjy1,fjz1,
 819                                                       fjx2,fjy2,fjz2,fjx3,fjy3,fjz3);
 820
 821             /* Inner loop uses 1025 flops */
 822         }
 823
 824         if(jidx<j_index_end)
 825         {
 826
 827             /* Get j neighbor index, and coordinate index */
 828             jnrlistA         = jjnr[jidx];
 829             jnrlistB         = jjnr[jidx+1];
 830             jnrlistC         = jjnr[jidx+2];
 831             jnrlistD         = jjnr[jidx+3];
 832             jnrlistE         = jjnr[jidx+4];
 833             jnrlistF         = jjnr[jidx+5];
 834             jnrlistG         = jjnr[jidx+6];
 835             jnrlistH         = jjnr[jidx+7];
 836             /* Sign of each element will be negative for non-real atoms.
 837              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 838              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
 839              */
 840             dummy_mask = gmx_mm256_set_m128(gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx+4)),_mm_setzero_si128())),
 841                                             gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128())));
 842
 843             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
 844             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
 845             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
 846             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
 847             jnrE       = (jnrlistE>=0) ? jnrlistE : 0;
 848             jnrF       = (jnrlistF>=0) ? jnrlistF : 0;
 849             jnrG       = (jnrlistG>=0) ? jnrlistG : 0;
 850             jnrH       = (jnrlistH>=0) ? jnrlistH : 0;
 851             j_coord_offsetA  = DIM*jnrA;
 852             j_coord_offsetB  = DIM*jnrB;
 853             j_coord_offsetC  = DIM*jnrC;
 854             j_coord_offsetD  = DIM*jnrD;
 855             j_coord_offsetE  = DIM*jnrE;
 856             j_coord_offsetF  = DIM*jnrF;
 857             j_coord_offsetG  = DIM*jnrG;
 858             j_coord_offsetH  = DIM*jnrH;
 859
 860             /* load j atom coordinates */
 861             gmx_mm256_load_4rvec_8ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 862                                                  x+j_coord_offsetC,x+j_coord_offsetD,
 863                                                  x+j_coord_offsetE,x+j_coord_offsetF,
 864                                                  x+j_coord_offsetG,x+j_coord_offsetH,
 865                                                  &jx0,&jy0,&jz0,&jx1,&jy1,&jz1,&jx2,
 866                                                  &jy2,&jz2,&jx3,&jy3,&jz3);
 867
 868             /* Calculate displacement vector */
 869             dx00             = _mm256_sub_ps(ix0,jx0);
 870             dy00             = _mm256_sub_ps(iy0,jy0);
 871             dz00             = _mm256_sub_ps(iz0,jz0);
 872             dx11             = _mm256_sub_ps(ix1,jx1);
 873             dy11             = _mm256_sub_ps(iy1,jy1);
 874             dz11             = _mm256_sub_ps(iz1,jz1);
 875             dx12             = _mm256_sub_ps(ix1,jx2);
 876             dy12             = _mm256_sub_ps(iy1,jy2);
 877             dz12             = _mm256_sub_ps(iz1,jz2);
 878             dx13             = _mm256_sub_ps(ix1,jx3);
 879             dy13             = _mm256_sub_ps(iy1,jy3);
 880             dz13             = _mm256_sub_ps(iz1,jz3);
 881             dx21             = _mm256_sub_ps(ix2,jx1);
 882             dy21             = _mm256_sub_ps(iy2,jy1);
 883             dz21             = _mm256_sub_ps(iz2,jz1);
 884             dx22             = _mm256_sub_ps(ix2,jx2);
 885             dy22             = _mm256_sub_ps(iy2,jy2);
 886             dz22             = _mm256_sub_ps(iz2,jz2);
 887             dx23             = _mm256_sub_ps(ix2,jx3);
 888             dy23             = _mm256_sub_ps(iy2,jy3);
 889             dz23             = _mm256_sub_ps(iz2,jz3);
 890             dx31             = _mm256_sub_ps(ix3,jx1);
 891             dy31             = _mm256_sub_ps(iy3,jy1);
 892             dz31             = _mm256_sub_ps(iz3,jz1);
 893             dx32             = _mm256_sub_ps(ix3,jx2);
 894             dy32             = _mm256_sub_ps(iy3,jy2);
 895             dz32             = _mm256_sub_ps(iz3,jz2);
 896             dx33             = _mm256_sub_ps(ix3,jx3);
 897             dy33             = _mm256_sub_ps(iy3,jy3);
 898             dz33             = _mm256_sub_ps(iz3,jz3);
 899
 900             /* Calculate squared distance and things based on it */
 901             rsq00            = gmx_mm256_calc_rsq_ps(dx00,dy00,dz00);
 902             rsq11            = gmx_mm256_calc_rsq_ps(dx11,dy11,dz11);
 903             rsq12            = gmx_mm256_calc_rsq_ps(dx12,dy12,dz12);
 904             rsq13            = gmx_mm256_calc_rsq_ps(dx13,dy13,dz13);
 905             rsq21            = gmx_mm256_calc_rsq_ps(dx21,dy21,dz21);
 906             rsq22            = gmx_mm256_calc_rsq_ps(dx22,dy22,dz22);
 907             rsq23            = gmx_mm256_calc_rsq_ps(dx23,dy23,dz23);
 908             rsq31            = gmx_mm256_calc_rsq_ps(dx31,dy31,dz31);
 909             rsq32            = gmx_mm256_calc_rsq_ps(dx32,dy32,dz32);
 910             rsq33            = gmx_mm256_calc_rsq_ps(dx33,dy33,dz33);
 911
 912             rinv11           = gmx_mm256_invsqrt_ps(rsq11);
 913             rinv12           = gmx_mm256_invsqrt_ps(rsq12);
 914             rinv13           = gmx_mm256_invsqrt_ps(rsq13);
 915             rinv21           = gmx_mm256_invsqrt_ps(rsq21);
 916             rinv22           = gmx_mm256_invsqrt_ps(rsq22);
 917             rinv23           = gmx_mm256_invsqrt_ps(rsq23);
 918             rinv31           = gmx_mm256_invsqrt_ps(rsq31);
 919             rinv32           = gmx_mm256_invsqrt_ps(rsq32);
 920             rinv33           = gmx_mm256_invsqrt_ps(rsq33);
 921
 922             rinvsq00         = gmx_mm256_inv_ps(rsq00);
 923             rinvsq11         = _mm256_mul_ps(rinv11,rinv11);
 924             rinvsq12         = _mm256_mul_ps(rinv12,rinv12);
 925             rinvsq13         = _mm256_mul_ps(rinv13,rinv13);
 926             rinvsq21         = _mm256_mul_ps(rinv21,rinv21);
 927             rinvsq22         = _mm256_mul_ps(rinv22,rinv22);
 928             rinvsq23         = _mm256_mul_ps(rinv23,rinv23);
 929             rinvsq31         = _mm256_mul_ps(rinv31,rinv31);
 930             rinvsq32         = _mm256_mul_ps(rinv32,rinv32);
 931             rinvsq33         = _mm256_mul_ps(rinv33,rinv33);
 932
 933             fjx0             = _mm256_setzero_ps();
 934             fjy0             = _mm256_setzero_ps();
 935             fjz0             = _mm256_setzero_ps();
 936             fjx1             = _mm256_setzero_ps();
 937             fjy1             = _mm256_setzero_ps();
 938             fjz1             = _mm256_setzero_ps();
 939             fjx2             = _mm256_setzero_ps();
 940             fjy2             = _mm256_setzero_ps();
 941             fjz2             = _mm256_setzero_ps();
 942             fjx3             = _mm256_setzero_ps();
 943             fjy3             = _mm256_setzero_ps();
 944             fjz3             = _mm256_setzero_ps();
 945
 946             /**************************
 947              * CALCULATE INTERACTIONS *
 948              **************************/
 949
 950             if (gmx_mm256_any_lt(rsq00,rcutoff2))
 951             {
 952
 953             /* LENNARD-JONES DISPERSION/REPULSION */
 954
 955             rinvsix          = _mm256_mul_ps(_mm256_mul_ps(rinvsq00,rinvsq00),rinvsq00);
 956             vvdw6            = _mm256_mul_ps(c6_00,rinvsix);
 957             vvdw12           = _mm256_mul_ps(c12_00,_mm256_mul_ps(rinvsix,rinvsix));
 958             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) ,
 959                                           _mm256_mul_ps( _mm256_sub_ps(vvdw6,_mm256_mul_ps(c6_00,sh_vdw_invrcut6)),one_sixth));
 960             fvdw             = _mm256_mul_ps(_mm256_sub_ps(vvdw12,vvdw6),rinvsq00);
 961
 962             cutoff_mask      = _mm256_cmp_ps(rsq00,rcutoff2,_CMP_LT_OQ);
 963
 964             /* Update potential sum for this i atom from the interaction with this j atom. */
 965             vvdw             = _mm256_and_ps(vvdw,cutoff_mask);
 966             vvdw             = _mm256_andnot_ps(dummy_mask,vvdw);
 967             vvdwsum          = _mm256_add_ps(vvdwsum,vvdw);
 968
 969             fscal            = fvdw;
 970
 971             fscal            = _mm256_and_ps(fscal,cutoff_mask);
 972
 973             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
 974
 975             /* Calculate temporary vectorial force */
 976             tx               = _mm256_mul_ps(fscal,dx00);
 977             ty               = _mm256_mul_ps(fscal,dy00);
 978             tz               = _mm256_mul_ps(fscal,dz00);
 979
 980             /* Update vectorial force */
 981             fix0             = _mm256_add_ps(fix0,tx);
 982             fiy0             = _mm256_add_ps(fiy0,ty);
 983             fiz0             = _mm256_add_ps(fiz0,tz);
 984
 985             fjx0             = _mm256_add_ps(fjx0,tx);
 986             fjy0             = _mm256_add_ps(fjy0,ty);
 987             fjz0             = _mm256_add_ps(fjz0,tz);
 988
 989             }
 990
 991             /**************************
 992              * CALCULATE INTERACTIONS *
 993              **************************/
 994
 995             if (gmx_mm256_any_lt(rsq11,rcutoff2))
 996             {
 997
 998             r11              = _mm256_mul_ps(rsq11,rinv11);
 999             r11              = _mm256_andnot_ps(dummy_mask,r11);
1000
1001             /* EWALD ELECTROSTATICS */
1002
1003             /* Analytical PME correction */
1004             zeta2            = _mm256_mul_ps(beta2,rsq11);
1005             rinv3            = _mm256_mul_ps(rinvsq11,rinv11);
1006             pmecorrF         = gmx_mm256_pmecorrF_ps(zeta2);
1007             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
1008             felec            = _mm256_mul_ps(qq11,felec);
1009             pmecorrV         = gmx_mm256_pmecorrV_ps(zeta2);
1010             pmecorrV         = _mm256_mul_ps(pmecorrV,beta);
1011             velec            = _mm256_sub_ps(_mm256_sub_ps(rinv11,sh_ewald),pmecorrV);
1012             velec            = _mm256_mul_ps(qq11,velec);
1013
1014             cutoff_mask      = _mm256_cmp_ps(rsq11,rcutoff2,_CMP_LT_OQ);
1015
1016             /* Update potential sum for this i atom from the interaction with this j atom. */
1017             velec            = _mm256_and_ps(velec,cutoff_mask);
1018             velec            = _mm256_andnot_ps(dummy_mask,velec);
1019             velecsum         = _mm256_add_ps(velecsum,velec);
1020
1021             fscal            = felec;
1022
1023             fscal            = _mm256_and_ps(fscal,cutoff_mask);
1024
1025             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
1026
1027             /* Calculate temporary vectorial force */
1028             tx               = _mm256_mul_ps(fscal,dx11);
1029             ty               = _mm256_mul_ps(fscal,dy11);
1030             tz               = _mm256_mul_ps(fscal,dz11);
1031
1032             /* Update vectorial force */
1033             fix1             = _mm256_add_ps(fix1,tx);
1034             fiy1             = _mm256_add_ps(fiy1,ty);
1035             fiz1             = _mm256_add_ps(fiz1,tz);
1036
1037             fjx1             = _mm256_add_ps(fjx1,tx);
1038             fjy1             = _mm256_add_ps(fjy1,ty);
1039             fjz1             = _mm256_add_ps(fjz1,tz);
1040
1041             }
1042
1043             /**************************
1044              * CALCULATE INTERACTIONS *
1045              **************************/
1046
1047             if (gmx_mm256_any_lt(rsq12,rcutoff2))
1048             {
1049
1050             r12              = _mm256_mul_ps(rsq12,rinv12);
1051             r12              = _mm256_andnot_ps(dummy_mask,r12);
1052
1053             /* EWALD ELECTROSTATICS */
1054
1055             /* Analytical PME correction */
1056             zeta2            = _mm256_mul_ps(beta2,rsq12);
1057             rinv3            = _mm256_mul_ps(rinvsq12,rinv12);
1058             pmecorrF         = gmx_mm256_pmecorrF_ps(zeta2);
1059             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
1060             felec            = _mm256_mul_ps(qq12,felec);
1061             pmecorrV         = gmx_mm256_pmecorrV_ps(zeta2);
1062             pmecorrV         = _mm256_mul_ps(pmecorrV,beta);
1063             velec            = _mm256_sub_ps(_mm256_sub_ps(rinv12,sh_ewald),pmecorrV);
1064             velec            = _mm256_mul_ps(qq12,velec);
1065
1066             cutoff_mask      = _mm256_cmp_ps(rsq12,rcutoff2,_CMP_LT_OQ);
1067
1068             /* Update potential sum for this i atom from the interaction with this j atom. */
1069             velec            = _mm256_and_ps(velec,cutoff_mask);
1070             velec            = _mm256_andnot_ps(dummy_mask,velec);
1071             velecsum         = _mm256_add_ps(velecsum,velec);
1072
1073             fscal            = felec;
1074
1075             fscal            = _mm256_and_ps(fscal,cutoff_mask);
1076
1077             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
1078
1079             /* Calculate temporary vectorial force */
1080             tx               = _mm256_mul_ps(fscal,dx12);
1081             ty               = _mm256_mul_ps(fscal,dy12);
1082             tz               = _mm256_mul_ps(fscal,dz12);
1083
1084             /* Update vectorial force */
1085             fix1             = _mm256_add_ps(fix1,tx);
1086             fiy1             = _mm256_add_ps(fiy1,ty);
1087             fiz1             = _mm256_add_ps(fiz1,tz);
1088
1089             fjx2             = _mm256_add_ps(fjx2,tx);
1090             fjy2             = _mm256_add_ps(fjy2,ty);
1091             fjz2             = _mm256_add_ps(fjz2,tz);
1092
1093             }
1094
1095             /**************************
1096              * CALCULATE INTERACTIONS *
1097              **************************/
1098
1099             if (gmx_mm256_any_lt(rsq13,rcutoff2))
1100             {
1101
1102             r13              = _mm256_mul_ps(rsq13,rinv13);
1103             r13              = _mm256_andnot_ps(dummy_mask,r13);
1104
1105             /* EWALD ELECTROSTATICS */
1106
1107             /* Analytical PME correction */
1108             zeta2            = _mm256_mul_ps(beta2,rsq13);
1109             rinv3            = _mm256_mul_ps(rinvsq13,rinv13);
1110             pmecorrF         = gmx_mm256_pmecorrF_ps(zeta2);
1111             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
1112             felec            = _mm256_mul_ps(qq13,felec);
1113             pmecorrV         = gmx_mm256_pmecorrV_ps(zeta2);
1114             pmecorrV         = _mm256_mul_ps(pmecorrV,beta);
1115             velec            = _mm256_sub_ps(_mm256_sub_ps(rinv13,sh_ewald),pmecorrV);
1116             velec            = _mm256_mul_ps(qq13,velec);
1117
1118             cutoff_mask      = _mm256_cmp_ps(rsq13,rcutoff2,_CMP_LT_OQ);
1119
1120             /* Update potential sum for this i atom from the interaction with this j atom. */
1121             velec            = _mm256_and_ps(velec,cutoff_mask);
1122             velec            = _mm256_andnot_ps(dummy_mask,velec);
1123             velecsum         = _mm256_add_ps(velecsum,velec);
1124
1125             fscal            = felec;
1126
1127             fscal            = _mm256_and_ps(fscal,cutoff_mask);
1128
1129             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
1130
1131             /* Calculate temporary vectorial force */
1132             tx               = _mm256_mul_ps(fscal,dx13);
1133             ty               = _mm256_mul_ps(fscal,dy13);
1134             tz               = _mm256_mul_ps(fscal,dz13);
1135
1136             /* Update vectorial force */
1137             fix1             = _mm256_add_ps(fix1,tx);
1138             fiy1             = _mm256_add_ps(fiy1,ty);
1139             fiz1             = _mm256_add_ps(fiz1,tz);
1140
1141             fjx3             = _mm256_add_ps(fjx3,tx);
1142             fjy3             = _mm256_add_ps(fjy3,ty);
1143             fjz3             = _mm256_add_ps(fjz3,tz);
1144
1145             }
1146
1147             /**************************
1148              * CALCULATE INTERACTIONS *
1149              **************************/
1150
1151             if (gmx_mm256_any_lt(rsq21,rcutoff2))
1152             {
1153
1154             r21              = _mm256_mul_ps(rsq21,rinv21);
1155             r21              = _mm256_andnot_ps(dummy_mask,r21);
1156
1157             /* EWALD ELECTROSTATICS */
1158
1159             /* Analytical PME correction */
1160             zeta2            = _mm256_mul_ps(beta2,rsq21);
1161             rinv3            = _mm256_mul_ps(rinvsq21,rinv21);
1162             pmecorrF         = gmx_mm256_pmecorrF_ps(zeta2);
1163             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
1164             felec            = _mm256_mul_ps(qq21,felec);
1165             pmecorrV         = gmx_mm256_pmecorrV_ps(zeta2);
1166             pmecorrV         = _mm256_mul_ps(pmecorrV,beta);
1167             velec            = _mm256_sub_ps(_mm256_sub_ps(rinv21,sh_ewald),pmecorrV);
1168             velec            = _mm256_mul_ps(qq21,velec);
1169
1170             cutoff_mask      = _mm256_cmp_ps(rsq21,rcutoff2,_CMP_LT_OQ);
1171
1172             /* Update potential sum for this i atom from the interaction with this j atom. */
1173             velec            = _mm256_and_ps(velec,cutoff_mask);
1174             velec            = _mm256_andnot_ps(dummy_mask,velec);
1175             velecsum         = _mm256_add_ps(velecsum,velec);
1176
1177             fscal            = felec;
1178
1179             fscal            = _mm256_and_ps(fscal,cutoff_mask);
1180
1181             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
1182
1183             /* Calculate temporary vectorial force */
1184             tx               = _mm256_mul_ps(fscal,dx21);
1185             ty               = _mm256_mul_ps(fscal,dy21);
1186             tz               = _mm256_mul_ps(fscal,dz21);
1187
1188             /* Update vectorial force */
1189             fix2             = _mm256_add_ps(fix2,tx);
1190             fiy2             = _mm256_add_ps(fiy2,ty);
1191             fiz2             = _mm256_add_ps(fiz2,tz);
1192
1193             fjx1             = _mm256_add_ps(fjx1,tx);
1194             fjy1             = _mm256_add_ps(fjy1,ty);
1195             fjz1             = _mm256_add_ps(fjz1,tz);
1196
1197             }
1198
1199             /**************************
1200              * CALCULATE INTERACTIONS *
1201              **************************/
1202
1203             if (gmx_mm256_any_lt(rsq22,rcutoff2))
1204             {
1205
1206             r22              = _mm256_mul_ps(rsq22,rinv22);
1207             r22              = _mm256_andnot_ps(dummy_mask,r22);
1208
1209             /* EWALD ELECTROSTATICS */
1210
1211             /* Analytical PME correction */
1212             zeta2            = _mm256_mul_ps(beta2,rsq22);
1213             rinv3            = _mm256_mul_ps(rinvsq22,rinv22);
1214             pmecorrF         = gmx_mm256_pmecorrF_ps(zeta2);
1215             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
1216             felec            = _mm256_mul_ps(qq22,felec);
1217             pmecorrV         = gmx_mm256_pmecorrV_ps(zeta2);
1218             pmecorrV         = _mm256_mul_ps(pmecorrV,beta);
1219             velec            = _mm256_sub_ps(_mm256_sub_ps(rinv22,sh_ewald),pmecorrV);
1220             velec            = _mm256_mul_ps(qq22,velec);
1221
1222             cutoff_mask      = _mm256_cmp_ps(rsq22,rcutoff2,_CMP_LT_OQ);
1223
1224             /* Update potential sum for this i atom from the interaction with this j atom. */
1225             velec            = _mm256_and_ps(velec,cutoff_mask);
1226             velec            = _mm256_andnot_ps(dummy_mask,velec);
1227             velecsum         = _mm256_add_ps(velecsum,velec);
1228
1229             fscal            = felec;
1230
1231             fscal            = _mm256_and_ps(fscal,cutoff_mask);
1232
1233             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
1234
1235             /* Calculate temporary vectorial force */
1236             tx               = _mm256_mul_ps(fscal,dx22);
1237             ty               = _mm256_mul_ps(fscal,dy22);
1238             tz               = _mm256_mul_ps(fscal,dz22);
1239
1240             /* Update vectorial force */
1241             fix2             = _mm256_add_ps(fix2,tx);
1242             fiy2             = _mm256_add_ps(fiy2,ty);
1243             fiz2             = _mm256_add_ps(fiz2,tz);
1244
1245             fjx2             = _mm256_add_ps(fjx2,tx);
1246             fjy2             = _mm256_add_ps(fjy2,ty);
1247             fjz2             = _mm256_add_ps(fjz2,tz);
1248
1249             }
1250
1251             /**************************
1252              * CALCULATE INTERACTIONS *
1253              **************************/
1254
1255             if (gmx_mm256_any_lt(rsq23,rcutoff2))
1256             {
1257
1258             r23              = _mm256_mul_ps(rsq23,rinv23);
1259             r23              = _mm256_andnot_ps(dummy_mask,r23);
1260
1261             /* EWALD ELECTROSTATICS */
1262
1263             /* Analytical PME correction */
1264             zeta2            = _mm256_mul_ps(beta2,rsq23);
1265             rinv3            = _mm256_mul_ps(rinvsq23,rinv23);
1266             pmecorrF         = gmx_mm256_pmecorrF_ps(zeta2);
1267             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
1268             felec            = _mm256_mul_ps(qq23,felec);
1269             pmecorrV         = gmx_mm256_pmecorrV_ps(zeta2);
1270             pmecorrV         = _mm256_mul_ps(pmecorrV,beta);
1271             velec            = _mm256_sub_ps(_mm256_sub_ps(rinv23,sh_ewald),pmecorrV);
1272             velec            = _mm256_mul_ps(qq23,velec);
1273
1274             cutoff_mask      = _mm256_cmp_ps(rsq23,rcutoff2,_CMP_LT_OQ);
1275
1276             /* Update potential sum for this i atom from the interaction with this j atom. */
1277             velec            = _mm256_and_ps(velec,cutoff_mask);
1278             velec            = _mm256_andnot_ps(dummy_mask,velec);
1279             velecsum         = _mm256_add_ps(velecsum,velec);
1280
1281             fscal            = felec;
1282
1283             fscal            = _mm256_and_ps(fscal,cutoff_mask);
1284
1285             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
1286
1287             /* Calculate temporary vectorial force */
1288             tx               = _mm256_mul_ps(fscal,dx23);
1289             ty               = _mm256_mul_ps(fscal,dy23);
1290             tz               = _mm256_mul_ps(fscal,dz23);
1291
1292             /* Update vectorial force */
1293             fix2             = _mm256_add_ps(fix2,tx);
1294             fiy2             = _mm256_add_ps(fiy2,ty);
1295             fiz2             = _mm256_add_ps(fiz2,tz);
1296
1297             fjx3             = _mm256_add_ps(fjx3,tx);
1298             fjy3             = _mm256_add_ps(fjy3,ty);
1299             fjz3             = _mm256_add_ps(fjz3,tz);
1300
1301             }
1302
1303             /**************************
1304              * CALCULATE INTERACTIONS *
1305              **************************/
1306
1307             if (gmx_mm256_any_lt(rsq31,rcutoff2))
1308             {
1309
1310             r31              = _mm256_mul_ps(rsq31,rinv31);
1311             r31              = _mm256_andnot_ps(dummy_mask,r31);
1312
1313             /* EWALD ELECTROSTATICS */
1314
1315             /* Analytical PME correction */
1316             zeta2            = _mm256_mul_ps(beta2,rsq31);
1317             rinv3            = _mm256_mul_ps(rinvsq31,rinv31);
1318             pmecorrF         = gmx_mm256_pmecorrF_ps(zeta2);
1319             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
1320             felec            = _mm256_mul_ps(qq31,felec);
1321             pmecorrV         = gmx_mm256_pmecorrV_ps(zeta2);
1322             pmecorrV         = _mm256_mul_ps(pmecorrV,beta);
1323             velec            = _mm256_sub_ps(_mm256_sub_ps(rinv31,sh_ewald),pmecorrV);
1324             velec            = _mm256_mul_ps(qq31,velec);
1325
1326             cutoff_mask      = _mm256_cmp_ps(rsq31,rcutoff2,_CMP_LT_OQ);
1327
1328             /* Update potential sum for this i atom from the interaction with this j atom. */
1329             velec            = _mm256_and_ps(velec,cutoff_mask);
1330             velec            = _mm256_andnot_ps(dummy_mask,velec);
1331             velecsum         = _mm256_add_ps(velecsum,velec);
1332
1333             fscal            = felec;
1334
1335             fscal            = _mm256_and_ps(fscal,cutoff_mask);
1336
1337             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
1338
1339             /* Calculate temporary vectorial force */
1340             tx               = _mm256_mul_ps(fscal,dx31);
1341             ty               = _mm256_mul_ps(fscal,dy31);
1342             tz               = _mm256_mul_ps(fscal,dz31);
1343
1344             /* Update vectorial force */
1345             fix3             = _mm256_add_ps(fix3,tx);
1346             fiy3             = _mm256_add_ps(fiy3,ty);
1347             fiz3             = _mm256_add_ps(fiz3,tz);
1348
1349             fjx1             = _mm256_add_ps(fjx1,tx);
1350             fjy1             = _mm256_add_ps(fjy1,ty);
1351             fjz1             = _mm256_add_ps(fjz1,tz);
1352
1353             }
1354
1355             /**************************
1356              * CALCULATE INTERACTIONS *
1357              **************************/
1358
1359             if (gmx_mm256_any_lt(rsq32,rcutoff2))
1360             {
1361
1362             r32              = _mm256_mul_ps(rsq32,rinv32);
1363             r32              = _mm256_andnot_ps(dummy_mask,r32);
1364
1365             /* EWALD ELECTROSTATICS */
1366
1367             /* Analytical PME correction */
1368             zeta2            = _mm256_mul_ps(beta2,rsq32);
1369             rinv3            = _mm256_mul_ps(rinvsq32,rinv32);
1370             pmecorrF         = gmx_mm256_pmecorrF_ps(zeta2);
1371             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
1372             felec            = _mm256_mul_ps(qq32,felec);
1373             pmecorrV         = gmx_mm256_pmecorrV_ps(zeta2);
1374             pmecorrV         = _mm256_mul_ps(pmecorrV,beta);
1375             velec            = _mm256_sub_ps(_mm256_sub_ps(rinv32,sh_ewald),pmecorrV);
1376             velec            = _mm256_mul_ps(qq32,velec);
1377
1378             cutoff_mask      = _mm256_cmp_ps(rsq32,rcutoff2,_CMP_LT_OQ);
1379
1380             /* Update potential sum for this i atom from the interaction with this j atom. */
1381             velec            = _mm256_and_ps(velec,cutoff_mask);
1382             velec            = _mm256_andnot_ps(dummy_mask,velec);
1383             velecsum         = _mm256_add_ps(velecsum,velec);
1384
1385             fscal            = felec;
1386
1387             fscal            = _mm256_and_ps(fscal,cutoff_mask);
1388
1389             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
1390
1391             /* Calculate temporary vectorial force */
1392             tx               = _mm256_mul_ps(fscal,dx32);
1393             ty               = _mm256_mul_ps(fscal,dy32);
1394             tz               = _mm256_mul_ps(fscal,dz32);
1395
1396             /* Update vectorial force */
1397             fix3             = _mm256_add_ps(fix3,tx);
1398             fiy3             = _mm256_add_ps(fiy3,ty);
1399             fiz3             = _mm256_add_ps(fiz3,tz);
1400
1401             fjx2             = _mm256_add_ps(fjx2,tx);
1402             fjy2             = _mm256_add_ps(fjy2,ty);
1403             fjz2             = _mm256_add_ps(fjz2,tz);
1404
1405             }
1406
1407             /**************************
1408              * CALCULATE INTERACTIONS *
1409              **************************/
1410
1411             if (gmx_mm256_any_lt(rsq33,rcutoff2))
1412             {
1413
1414             r33              = _mm256_mul_ps(rsq33,rinv33);
1415             r33              = _mm256_andnot_ps(dummy_mask,r33);
1416
1417             /* EWALD ELECTROSTATICS */
1418
1419             /* Analytical PME correction */
1420             zeta2            = _mm256_mul_ps(beta2,rsq33);
1421             rinv3            = _mm256_mul_ps(rinvsq33,rinv33);
1422             pmecorrF         = gmx_mm256_pmecorrF_ps(zeta2);
1423             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
1424             felec            = _mm256_mul_ps(qq33,felec);
1425             pmecorrV         = gmx_mm256_pmecorrV_ps(zeta2);
1426             pmecorrV         = _mm256_mul_ps(pmecorrV,beta);
1427             velec            = _mm256_sub_ps(_mm256_sub_ps(rinv33,sh_ewald),pmecorrV);
1428             velec            = _mm256_mul_ps(qq33,velec);
1429
1430             cutoff_mask      = _mm256_cmp_ps(rsq33,rcutoff2,_CMP_LT_OQ);
1431
1432             /* Update potential sum for this i atom from the interaction with this j atom. */
1433             velec            = _mm256_and_ps(velec,cutoff_mask);
1434             velec            = _mm256_andnot_ps(dummy_mask,velec);
1435             velecsum         = _mm256_add_ps(velecsum,velec);
1436
1437             fscal            = felec;
1438
1439             fscal            = _mm256_and_ps(fscal,cutoff_mask);
1440
1441             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
1442
1443             /* Calculate temporary vectorial force */
1444             tx               = _mm256_mul_ps(fscal,dx33);
1445             ty               = _mm256_mul_ps(fscal,dy33);
1446             tz               = _mm256_mul_ps(fscal,dz33);
1447
1448             /* Update vectorial force */
1449             fix3             = _mm256_add_ps(fix3,tx);
1450             fiy3             = _mm256_add_ps(fiy3,ty);
1451             fiz3             = _mm256_add_ps(fiz3,tz);
1452
1453             fjx3             = _mm256_add_ps(fjx3,tx);
1454             fjy3             = _mm256_add_ps(fjy3,ty);
1455             fjz3             = _mm256_add_ps(fjz3,tz);
1456
1457             }
1458
1459             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
1460             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
1461             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
1462             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
1463             fjptrE             = (jnrlistE>=0) ? f+j_coord_offsetE : scratch;
1464             fjptrF             = (jnrlistF>=0) ? f+j_coord_offsetF : scratch;
1465             fjptrG             = (jnrlistG>=0) ? f+j_coord_offsetG : scratch;
1466             fjptrH             = (jnrlistH>=0) ? f+j_coord_offsetH : scratch;
1467
1468             gmx_mm256_decrement_4rvec_8ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjptrE,fjptrF,fjptrG,fjptrH,
1469                                                       fjx0,fjy0,fjz0,fjx1,fjy1,fjz1,
1470                                                       fjx2,fjy2,fjz2,fjx3,fjy3,fjz3);
1471
1472             /* Inner loop uses 1034 flops */
1473         }
1474
1475         /* End of innermost loop */
1476
1477         gmx_mm256_update_iforce_4atom_swizzle_ps(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,fix3,fiy3,fiz3,
1478                                                  f+i_coord_offset,fshift+i_shift_offset);
1479
1480         ggid                        = gid[iidx];
1481         /* Update potential energies */
1482         gmx_mm256_update_1pot_ps(velecsum,kernel_data->energygrp_elec+ggid);
1483         gmx_mm256_update_1pot_ps(vvdwsum,kernel_data->energygrp_vdw+ggid);
1484
1485         /* Increment number of inner iterations */
1486         inneriter                  += j_index_end - j_index_start;
1487
1488         /* Outer loop uses 26 flops */
1489     }
1490
1491     /* Increment number of outer iterations */
1492     outeriter        += nri;
1493
1494     /* Update outer/inner flops */
1495
1496     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W4W4_VF,outeriter*26 + inneriter*1034);
1497 }
1498 /*
1499  * Gromacs nonbonded kernel:   nb_kernel_ElecEwSh_VdwLJSh_GeomW4W4_F_avx_256_single
1500  * Electrostatics interaction: Ewald
1501  * VdW interaction:            LennardJones
1502  * Geometry:                   Water4-Water4
1503  * Calculate force/pot:        Force
1504  */
1505 void
1506 nb_kernel_ElecEwSh_VdwLJSh_GeomW4W4_F_avx_256_single
1507                     (t_nblist * gmx_restrict                nlist,
1508                      rvec * gmx_restrict                    xx,
1509                      rvec * gmx_restrict                    ff,
1510                      t_forcerec * gmx_restrict              fr,
1511                      t_mdatoms * gmx_restrict               mdatoms,
1512                      nb_kernel_data_t * gmx_restrict        kernel_data,
1513                      t_nrnb * gmx_restrict                  nrnb)
1514 {
1515     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
1516      * just 0 for non-waters.
1517      * Suffixes A,B,C,D,E,F,G,H refer to j loop unrolling done with AVX, e.g. for the eight different
1518      * jnr indices corresponding to data put in the four positions in the SIMD register.
1519      */
1520     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
1521     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
1522     int              jnrA,jnrB,jnrC,jnrD;
1523     int              jnrE,jnrF,jnrG,jnrH;
1524     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
1525     int              jnrlistE,jnrlistF,jnrlistG,jnrlistH;
1526     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
1527     int              j_coord_offsetE,j_coord_offsetF,j_coord_offsetG,j_coord_offsetH;
1528     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
1529     real             rcutoff_scalar;
1530     real             *shiftvec,*fshift,*x,*f;
1531     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD,*fjptrE,*fjptrF,*fjptrG,*fjptrH;
1532     real             scratch[4*DIM];
1533     __m256           tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
1534     real *           vdwioffsetptr0;
1535     __m256           ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
1536     real *           vdwioffsetptr1;
1537     __m256           ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
1538     real *           vdwioffsetptr2;
1539     __m256           ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
1540     real *           vdwioffsetptr3;
1541     __m256           ix3,iy3,iz3,fix3,fiy3,fiz3,iq3,isai3;
1542     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D,vdwjidx0E,vdwjidx0F,vdwjidx0G,vdwjidx0H;
1543     __m256           jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
1544     int              vdwjidx1A,vdwjidx1B,vdwjidx1C,vdwjidx1D,vdwjidx1E,vdwjidx1F,vdwjidx1G,vdwjidx1H;
1545     __m256           jx1,jy1,jz1,fjx1,fjy1,fjz1,jq1,isaj1;
1546     int              vdwjidx2A,vdwjidx2B,vdwjidx2C,vdwjidx2D,vdwjidx2E,vdwjidx2F,vdwjidx2G,vdwjidx2H;
1547     __m256           jx2,jy2,jz2,fjx2,fjy2,fjz2,jq2,isaj2;
1548     int              vdwjidx3A,vdwjidx3B,vdwjidx3C,vdwjidx3D,vdwjidx3E,vdwjidx3F,vdwjidx3G,vdwjidx3H;
1549     __m256           jx3,jy3,jz3,fjx3,fjy3,fjz3,jq3,isaj3;
1550     __m256           dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
1551     __m256           dx11,dy11,dz11,rsq11,rinv11,rinvsq11,r11,qq11,c6_11,c12_11;
1552     __m256           dx12,dy12,dz12,rsq12,rinv12,rinvsq12,r12,qq12,c6_12,c12_12;
1553     __m256           dx13,dy13,dz13,rsq13,rinv13,rinvsq13,r13,qq13,c6_13,c12_13;
1554     __m256           dx21,dy21,dz21,rsq21,rinv21,rinvsq21,r21,qq21,c6_21,c12_21;
1555     __m256           dx22,dy22,dz22,rsq22,rinv22,rinvsq22,r22,qq22,c6_22,c12_22;
1556     __m256           dx23,dy23,dz23,rsq23,rinv23,rinvsq23,r23,qq23,c6_23,c12_23;
1557     __m256           dx31,dy31,dz31,rsq31,rinv31,rinvsq31,r31,qq31,c6_31,c12_31;
1558     __m256           dx32,dy32,dz32,rsq32,rinv32,rinvsq32,r32,qq32,c6_32,c12_32;
1559     __m256           dx33,dy33,dz33,rsq33,rinv33,rinvsq33,r33,qq33,c6_33,c12_33;
1560     __m256           velec,felec,velecsum,facel,crf,krf,krf2;
1561     real             *charge;
1562     int              nvdwtype;
1563     __m256           rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
1564     int              *vdwtype;
1565     real             *vdwparam;
1566     __m256           one_sixth   = _mm256_set1_ps(1.0/6.0);
1567     __m256           one_twelfth = _mm256_set1_ps(1.0/12.0);
1568     __m256i          ewitab;
1569     __m128i          ewitab_lo,ewitab_hi;
1570     __m256           ewtabscale,eweps,sh_ewald,ewrt,ewtabhalfspace,ewtabF,ewtabFn,ewtabD,ewtabV;
1571     __m256           beta,beta2,beta3,zeta2,pmecorrF,pmecorrV,rinv3;
1572     real             *ewtab;
1573     __m256           dummy_mask,cutoff_mask;
1574     __m256           signbit = _mm256_castsi256_ps( _mm256_set1_epi32(0x80000000) );
1575     __m256           one     = _mm256_set1_ps(1.0);
1576     __m256           two     = _mm256_set1_ps(2.0);
1577     x                = xx[0];
1578     f                = ff[0];
1579
1580     nri              = nlist->nri;
1581     iinr             = nlist->iinr;
1582     jindex           = nlist->jindex;
1583     jjnr             = nlist->jjnr;
1584     shiftidx         = nlist->shift;
1585     gid              = nlist->gid;
1586     shiftvec         = fr->shift_vec[0];
1587     fshift           = fr->fshift[0];
1588     facel            = _mm256_set1_ps(fr->epsfac);
1589     charge           = mdatoms->chargeA;
1590     nvdwtype         = fr->ntype;
1591     vdwparam         = fr->nbfp;
1592     vdwtype          = mdatoms->typeA;
1593
1594     sh_ewald         = _mm256_set1_ps(fr->ic->sh_ewald);
1595     beta             = _mm256_set1_ps(fr->ic->ewaldcoeff);
1596     beta2            = _mm256_mul_ps(beta,beta);
1597     beta3            = _mm256_mul_ps(beta,beta2);
1598
1599     ewtab            = fr->ic->tabq_coul_F;
1600     ewtabscale       = _mm256_set1_ps(fr->ic->tabq_scale);
1601     ewtabhalfspace   = _mm256_set1_ps(0.5/fr->ic->tabq_scale);
1602
1603     /* Setup water-specific parameters */
1604     inr              = nlist->iinr[0];
1605     iq1              = _mm256_mul_ps(facel,_mm256_set1_ps(charge[inr+1]));
1606     iq2              = _mm256_mul_ps(facel,_mm256_set1_ps(charge[inr+2]));
1607     iq3              = _mm256_mul_ps(facel,_mm256_set1_ps(charge[inr+3]));
1608     vdwioffsetptr0   = vdwparam+2*nvdwtype*vdwtype[inr+0];
1609
1610     jq1              = _mm256_set1_ps(charge[inr+1]);
1611     jq2              = _mm256_set1_ps(charge[inr+2]);
1612     jq3              = _mm256_set1_ps(charge[inr+3]);
1613     vdwjidx0A        = 2*vdwtype[inr+0];
1614     c6_00            = _mm256_set1_ps(vdwioffsetptr0[vdwjidx0A]);
1615     c12_00           = _mm256_set1_ps(vdwioffsetptr0[vdwjidx0A+1]);
1616     qq11             = _mm256_mul_ps(iq1,jq1);
1617     qq12             = _mm256_mul_ps(iq1,jq2);
1618     qq13             = _mm256_mul_ps(iq1,jq3);
1619     qq21             = _mm256_mul_ps(iq2,jq1);
1620     qq22             = _mm256_mul_ps(iq2,jq2);
1621     qq23             = _mm256_mul_ps(iq2,jq3);
1622     qq31             = _mm256_mul_ps(iq3,jq1);
1623     qq32             = _mm256_mul_ps(iq3,jq2);
1624     qq33             = _mm256_mul_ps(iq3,jq3);
1625
1626     /* When we use explicit cutoffs the value must be identical for elec and VdW, so use elec as an arbitrary choice */
1627     rcutoff_scalar   = fr->rcoulomb;
1628     rcutoff          = _mm256_set1_ps(rcutoff_scalar);
1629     rcutoff2         = _mm256_mul_ps(rcutoff,rcutoff);
1630
1631     sh_vdw_invrcut6  = _mm256_set1_ps(fr->ic->sh_invrc6);
1632     rvdw             = _mm256_set1_ps(fr->rvdw);
1633
1634     /* Avoid stupid compiler warnings */
1635     jnrA = jnrB = jnrC = jnrD = jnrE = jnrF = jnrG = jnrH = 0;
1636     j_coord_offsetA = 0;
1637     j_coord_offsetB = 0;
1638     j_coord_offsetC = 0;
1639     j_coord_offsetD = 0;
1640     j_coord_offsetE = 0;
1641     j_coord_offsetF = 0;
1642     j_coord_offsetG = 0;
1643     j_coord_offsetH = 0;
1644
1645     outeriter        = 0;
1646     inneriter        = 0;
1647
1648     for(iidx=0;iidx<4*DIM;iidx++)
1649     {
1650         scratch[iidx] = 0.0;
1651     }
1652
1653     /* Start outer loop over neighborlists */
1654     for(iidx=0; iidx<nri; iidx++)
1655     {
1656         /* Load shift vector for this list */
1657         i_shift_offset   = DIM*shiftidx[iidx];
1658
1659         /* Load limits for loop over neighbors */
1660         j_index_start    = jindex[iidx];
1661         j_index_end      = jindex[iidx+1];
1662
1663         /* Get outer coordinate index */
1664         inr              = iinr[iidx];
1665         i_coord_offset   = DIM*inr;
1666
1667         /* Load i particle coords and add shift vector */
1668         gmx_mm256_load_shift_and_4rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,
1669                                                     &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2,&ix3,&iy3,&iz3);
1670
1671         fix0             = _mm256_setzero_ps();
1672         fiy0             = _mm256_setzero_ps();
1673         fiz0             = _mm256_setzero_ps();
1674         fix1             = _mm256_setzero_ps();
1675         fiy1             = _mm256_setzero_ps();
1676         fiz1             = _mm256_setzero_ps();
1677         fix2             = _mm256_setzero_ps();
1678         fiy2             = _mm256_setzero_ps();
1679         fiz2             = _mm256_setzero_ps();
1680         fix3             = _mm256_setzero_ps();
1681         fiy3             = _mm256_setzero_ps();
1682         fiz3             = _mm256_setzero_ps();
1683
1684         /* Start inner kernel loop */
1685         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+7]>=0; jidx+=8)
1686         {
1687
1688             /* Get j neighbor index, and coordinate index */
1689             jnrA             = jjnr[jidx];
1690             jnrB             = jjnr[jidx+1];
1691             jnrC             = jjnr[jidx+2];
1692             jnrD             = jjnr[jidx+3];
1693             jnrE             = jjnr[jidx+4];
1694             jnrF             = jjnr[jidx+5];
1695             jnrG             = jjnr[jidx+6];
1696             jnrH             = jjnr[jidx+7];
1697             j_coord_offsetA  = DIM*jnrA;
1698             j_coord_offsetB  = DIM*jnrB;
1699             j_coord_offsetC  = DIM*jnrC;
1700             j_coord_offsetD  = DIM*jnrD;
1701             j_coord_offsetE  = DIM*jnrE;
1702             j_coord_offsetF  = DIM*jnrF;
1703             j_coord_offsetG  = DIM*jnrG;
1704             j_coord_offsetH  = DIM*jnrH;
1705
1706             /* load j atom coordinates */
1707             gmx_mm256_load_4rvec_8ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
1708                                                  x+j_coord_offsetC,x+j_coord_offsetD,
1709                                                  x+j_coord_offsetE,x+j_coord_offsetF,
1710                                                  x+j_coord_offsetG,x+j_coord_offsetH,
1711                                                  &jx0,&jy0,&jz0,&jx1,&jy1,&jz1,&jx2,
1712                                                  &jy2,&jz2,&jx3,&jy3,&jz3);
1713
1714             /* Calculate displacement vector */
1715             dx00             = _mm256_sub_ps(ix0,jx0);
1716             dy00             = _mm256_sub_ps(iy0,jy0);
1717             dz00             = _mm256_sub_ps(iz0,jz0);
1718             dx11             = _mm256_sub_ps(ix1,jx1);
1719             dy11             = _mm256_sub_ps(iy1,jy1);
1720             dz11             = _mm256_sub_ps(iz1,jz1);
1721             dx12             = _mm256_sub_ps(ix1,jx2);
1722             dy12             = _mm256_sub_ps(iy1,jy2);
1723             dz12             = _mm256_sub_ps(iz1,jz2);
1724             dx13             = _mm256_sub_ps(ix1,jx3);
1725             dy13             = _mm256_sub_ps(iy1,jy3);
1726             dz13             = _mm256_sub_ps(iz1,jz3);
1727             dx21             = _mm256_sub_ps(ix2,jx1);
1728             dy21             = _mm256_sub_ps(iy2,jy1);
1729             dz21             = _mm256_sub_ps(iz2,jz1);
1730             dx22             = _mm256_sub_ps(ix2,jx2);
1731             dy22             = _mm256_sub_ps(iy2,jy2);
1732             dz22             = _mm256_sub_ps(iz2,jz2);
1733             dx23             = _mm256_sub_ps(ix2,jx3);
1734             dy23             = _mm256_sub_ps(iy2,jy3);
1735             dz23             = _mm256_sub_ps(iz2,jz3);
1736             dx31             = _mm256_sub_ps(ix3,jx1);
1737             dy31             = _mm256_sub_ps(iy3,jy1);
1738             dz31             = _mm256_sub_ps(iz3,jz1);
1739             dx32             = _mm256_sub_ps(ix3,jx2);
1740             dy32             = _mm256_sub_ps(iy3,jy2);
1741             dz32             = _mm256_sub_ps(iz3,jz2);
1742             dx33             = _mm256_sub_ps(ix3,jx3);
1743             dy33             = _mm256_sub_ps(iy3,jy3);
1744             dz33             = _mm256_sub_ps(iz3,jz3);
1745
1746             /* Calculate squared distance and things based on it */
1747             rsq00            = gmx_mm256_calc_rsq_ps(dx00,dy00,dz00);
1748             rsq11            = gmx_mm256_calc_rsq_ps(dx11,dy11,dz11);
1749             rsq12            = gmx_mm256_calc_rsq_ps(dx12,dy12,dz12);
1750             rsq13            = gmx_mm256_calc_rsq_ps(dx13,dy13,dz13);
1751             rsq21            = gmx_mm256_calc_rsq_ps(dx21,dy21,dz21);
1752             rsq22            = gmx_mm256_calc_rsq_ps(dx22,dy22,dz22);
1753             rsq23            = gmx_mm256_calc_rsq_ps(dx23,dy23,dz23);
1754             rsq31            = gmx_mm256_calc_rsq_ps(dx31,dy31,dz31);
1755             rsq32            = gmx_mm256_calc_rsq_ps(dx32,dy32,dz32);
1756             rsq33            = gmx_mm256_calc_rsq_ps(dx33,dy33,dz33);
1757
1758             rinv11           = gmx_mm256_invsqrt_ps(rsq11);
1759             rinv12           = gmx_mm256_invsqrt_ps(rsq12);
1760             rinv13           = gmx_mm256_invsqrt_ps(rsq13);
1761             rinv21           = gmx_mm256_invsqrt_ps(rsq21);
1762             rinv22           = gmx_mm256_invsqrt_ps(rsq22);
1763             rinv23           = gmx_mm256_invsqrt_ps(rsq23);
1764             rinv31           = gmx_mm256_invsqrt_ps(rsq31);
1765             rinv32           = gmx_mm256_invsqrt_ps(rsq32);
1766             rinv33           = gmx_mm256_invsqrt_ps(rsq33);
1767
1768             rinvsq00         = gmx_mm256_inv_ps(rsq00);
1769             rinvsq11         = _mm256_mul_ps(rinv11,rinv11);
1770             rinvsq12         = _mm256_mul_ps(rinv12,rinv12);
1771             rinvsq13         = _mm256_mul_ps(rinv13,rinv13);
1772             rinvsq21         = _mm256_mul_ps(rinv21,rinv21);
1773             rinvsq22         = _mm256_mul_ps(rinv22,rinv22);
1774             rinvsq23         = _mm256_mul_ps(rinv23,rinv23);
1775             rinvsq31         = _mm256_mul_ps(rinv31,rinv31);
1776             rinvsq32         = _mm256_mul_ps(rinv32,rinv32);
1777             rinvsq33         = _mm256_mul_ps(rinv33,rinv33);
1778
1779             fjx0             = _mm256_setzero_ps();
1780             fjy0             = _mm256_setzero_ps();
1781             fjz0             = _mm256_setzero_ps();
1782             fjx1             = _mm256_setzero_ps();
1783             fjy1             = _mm256_setzero_ps();
1784             fjz1             = _mm256_setzero_ps();
1785             fjx2             = _mm256_setzero_ps();
1786             fjy2             = _mm256_setzero_ps();
1787             fjz2             = _mm256_setzero_ps();
1788             fjx3             = _mm256_setzero_ps();
1789             fjy3             = _mm256_setzero_ps();
1790             fjz3             = _mm256_setzero_ps();
1791
1792             /**************************
1793              * CALCULATE INTERACTIONS *
1794              **************************/
1795
1796             if (gmx_mm256_any_lt(rsq00,rcutoff2))
1797             {
1798
1799             /* LENNARD-JONES DISPERSION/REPULSION */
1800
1801             rinvsix          = _mm256_mul_ps(_mm256_mul_ps(rinvsq00,rinvsq00),rinvsq00);
1802             fvdw             = _mm256_mul_ps(_mm256_sub_ps(_mm256_mul_ps(c12_00,rinvsix),c6_00),_mm256_mul_ps(rinvsix,rinvsq00));
1803
1804             cutoff_mask      = _mm256_cmp_ps(rsq00,rcutoff2,_CMP_LT_OQ);
1805
1806             fscal            = fvdw;
1807
1808             fscal            = _mm256_and_ps(fscal,cutoff_mask);
1809
1810             /* Calculate temporary vectorial force */
1811             tx               = _mm256_mul_ps(fscal,dx00);
1812             ty               = _mm256_mul_ps(fscal,dy00);
1813             tz               = _mm256_mul_ps(fscal,dz00);
1814
1815             /* Update vectorial force */
1816             fix0             = _mm256_add_ps(fix0,tx);
1817             fiy0             = _mm256_add_ps(fiy0,ty);
1818             fiz0             = _mm256_add_ps(fiz0,tz);
1819
1820             fjx0             = _mm256_add_ps(fjx0,tx);
1821             fjy0             = _mm256_add_ps(fjy0,ty);
1822             fjz0             = _mm256_add_ps(fjz0,tz);
1823
1824             }
1825
1826             /**************************
1827              * CALCULATE INTERACTIONS *
1828              **************************/
1829
1830             if (gmx_mm256_any_lt(rsq11,rcutoff2))
1831             {
1832
1833             r11              = _mm256_mul_ps(rsq11,rinv11);
1834
1835             /* EWALD ELECTROSTATICS */
1836
1837             /* Analytical PME correction */
1838             zeta2            = _mm256_mul_ps(beta2,rsq11);
1839             rinv3            = _mm256_mul_ps(rinvsq11,rinv11);
1840             pmecorrF         = gmx_mm256_pmecorrF_ps(zeta2);
1841             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
1842             felec            = _mm256_mul_ps(qq11,felec);
1843
1844             cutoff_mask      = _mm256_cmp_ps(rsq11,rcutoff2,_CMP_LT_OQ);
1845
1846             fscal            = felec;
1847
1848             fscal            = _mm256_and_ps(fscal,cutoff_mask);
1849
1850             /* Calculate temporary vectorial force */
1851             tx               = _mm256_mul_ps(fscal,dx11);
1852             ty               = _mm256_mul_ps(fscal,dy11);
1853             tz               = _mm256_mul_ps(fscal,dz11);
1854
1855             /* Update vectorial force */
1856             fix1             = _mm256_add_ps(fix1,tx);
1857             fiy1             = _mm256_add_ps(fiy1,ty);
1858             fiz1             = _mm256_add_ps(fiz1,tz);
1859
1860             fjx1             = _mm256_add_ps(fjx1,tx);
1861             fjy1             = _mm256_add_ps(fjy1,ty);
1862             fjz1             = _mm256_add_ps(fjz1,tz);
1863
1864             }
1865
1866             /**************************
1867              * CALCULATE INTERACTIONS *
1868              **************************/
1869
1870             if (gmx_mm256_any_lt(rsq12,rcutoff2))
1871             {
1872
1873             r12              = _mm256_mul_ps(rsq12,rinv12);
1874
1875             /* EWALD ELECTROSTATICS */
1876
1877             /* Analytical PME correction */
1878             zeta2            = _mm256_mul_ps(beta2,rsq12);
1879             rinv3            = _mm256_mul_ps(rinvsq12,rinv12);
1880             pmecorrF         = gmx_mm256_pmecorrF_ps(zeta2);
1881             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
1882             felec            = _mm256_mul_ps(qq12,felec);
1883
1884             cutoff_mask      = _mm256_cmp_ps(rsq12,rcutoff2,_CMP_LT_OQ);
1885
1886             fscal            = felec;
1887
1888             fscal            = _mm256_and_ps(fscal,cutoff_mask);
1889
1890             /* Calculate temporary vectorial force */
1891             tx               = _mm256_mul_ps(fscal,dx12);
1892             ty               = _mm256_mul_ps(fscal,dy12);
1893             tz               = _mm256_mul_ps(fscal,dz12);
1894
1895             /* Update vectorial force */
1896             fix1             = _mm256_add_ps(fix1,tx);
1897             fiy1             = _mm256_add_ps(fiy1,ty);
1898             fiz1             = _mm256_add_ps(fiz1,tz);
1899
1900             fjx2             = _mm256_add_ps(fjx2,tx);
1901             fjy2             = _mm256_add_ps(fjy2,ty);
1902             fjz2             = _mm256_add_ps(fjz2,tz);
1903
1904             }
1905
1906             /**************************
1907              * CALCULATE INTERACTIONS *
1908              **************************/
1909
1910             if (gmx_mm256_any_lt(rsq13,rcutoff2))
1911             {
1912
1913             r13              = _mm256_mul_ps(rsq13,rinv13);
1914
1915             /* EWALD ELECTROSTATICS */
1916
1917             /* Analytical PME correction */
1918             zeta2            = _mm256_mul_ps(beta2,rsq13);
1919             rinv3            = _mm256_mul_ps(rinvsq13,rinv13);
1920             pmecorrF         = gmx_mm256_pmecorrF_ps(zeta2);
1921             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
1922             felec            = _mm256_mul_ps(qq13,felec);
1923
1924             cutoff_mask      = _mm256_cmp_ps(rsq13,rcutoff2,_CMP_LT_OQ);
1925
1926             fscal            = felec;
1927
1928             fscal            = _mm256_and_ps(fscal,cutoff_mask);
1929
1930             /* Calculate temporary vectorial force */
1931             tx               = _mm256_mul_ps(fscal,dx13);
1932             ty               = _mm256_mul_ps(fscal,dy13);
1933             tz               = _mm256_mul_ps(fscal,dz13);
1934
1935             /* Update vectorial force */
1936             fix1             = _mm256_add_ps(fix1,tx);
1937             fiy1             = _mm256_add_ps(fiy1,ty);
1938             fiz1             = _mm256_add_ps(fiz1,tz);
1939
1940             fjx3             = _mm256_add_ps(fjx3,tx);
1941             fjy3             = _mm256_add_ps(fjy3,ty);
1942             fjz3             = _mm256_add_ps(fjz3,tz);
1943
1944             }
1945
1946             /**************************
1947              * CALCULATE INTERACTIONS *
1948              **************************/
1949
1950             if (gmx_mm256_any_lt(rsq21,rcutoff2))
1951             {
1952
1953             r21              = _mm256_mul_ps(rsq21,rinv21);
1954
1955             /* EWALD ELECTROSTATICS */
1956
1957             /* Analytical PME correction */
1958             zeta2            = _mm256_mul_ps(beta2,rsq21);
1959             rinv3            = _mm256_mul_ps(rinvsq21,rinv21);
1960             pmecorrF         = gmx_mm256_pmecorrF_ps(zeta2);
1961             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
1962             felec            = _mm256_mul_ps(qq21,felec);
1963
1964             cutoff_mask      = _mm256_cmp_ps(rsq21,rcutoff2,_CMP_LT_OQ);
1965
1966             fscal            = felec;
1967
1968             fscal            = _mm256_and_ps(fscal,cutoff_mask);
1969
1970             /* Calculate temporary vectorial force */
1971             tx               = _mm256_mul_ps(fscal,dx21);
1972             ty               = _mm256_mul_ps(fscal,dy21);
1973             tz               = _mm256_mul_ps(fscal,dz21);
1974
1975             /* Update vectorial force */
1976             fix2             = _mm256_add_ps(fix2,tx);
1977             fiy2             = _mm256_add_ps(fiy2,ty);
1978             fiz2             = _mm256_add_ps(fiz2,tz);
1979
1980             fjx1             = _mm256_add_ps(fjx1,tx);
1981             fjy1             = _mm256_add_ps(fjy1,ty);
1982             fjz1             = _mm256_add_ps(fjz1,tz);
1983
1984             }
1985
1986             /**************************
1987              * CALCULATE INTERACTIONS *
1988              **************************/
1989
1990             if (gmx_mm256_any_lt(rsq22,rcutoff2))
1991             {
1992
1993             r22              = _mm256_mul_ps(rsq22,rinv22);
1994
1995             /* EWALD ELECTROSTATICS */
1996
1997             /* Analytical PME correction */
1998             zeta2            = _mm256_mul_ps(beta2,rsq22);
1999             rinv3            = _mm256_mul_ps(rinvsq22,rinv22);
2000             pmecorrF         = gmx_mm256_pmecorrF_ps(zeta2);
2001             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
2002             felec            = _mm256_mul_ps(qq22,felec);
2003
2004             cutoff_mask      = _mm256_cmp_ps(rsq22,rcutoff2,_CMP_LT_OQ);
2005
2006             fscal            = felec;
2007
2008             fscal            = _mm256_and_ps(fscal,cutoff_mask);
2009
2010             /* Calculate temporary vectorial force */
2011             tx               = _mm256_mul_ps(fscal,dx22);
2012             ty               = _mm256_mul_ps(fscal,dy22);
2013             tz               = _mm256_mul_ps(fscal,dz22);
2014
2015             /* Update vectorial force */
2016             fix2             = _mm256_add_ps(fix2,tx);
2017             fiy2             = _mm256_add_ps(fiy2,ty);
2018             fiz2             = _mm256_add_ps(fiz2,tz);
2019
2020             fjx2             = _mm256_add_ps(fjx2,tx);
2021             fjy2             = _mm256_add_ps(fjy2,ty);
2022             fjz2             = _mm256_add_ps(fjz2,tz);
2023
2024             }
2025
2026             /**************************
2027              * CALCULATE INTERACTIONS *
2028              **************************/
2029
2030             if (gmx_mm256_any_lt(rsq23,rcutoff2))
2031             {
2032
2033             r23              = _mm256_mul_ps(rsq23,rinv23);
2034
2035             /* EWALD ELECTROSTATICS */
2036
2037             /* Analytical PME correction */
2038             zeta2            = _mm256_mul_ps(beta2,rsq23);
2039             rinv3            = _mm256_mul_ps(rinvsq23,rinv23);
2040             pmecorrF         = gmx_mm256_pmecorrF_ps(zeta2);
2041             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
2042             felec            = _mm256_mul_ps(qq23,felec);
2043
2044             cutoff_mask      = _mm256_cmp_ps(rsq23,rcutoff2,_CMP_LT_OQ);
2045
2046             fscal            = felec;
2047
2048             fscal            = _mm256_and_ps(fscal,cutoff_mask);
2049
2050             /* Calculate temporary vectorial force */
2051             tx               = _mm256_mul_ps(fscal,dx23);
2052             ty               = _mm256_mul_ps(fscal,dy23);
2053             tz               = _mm256_mul_ps(fscal,dz23);
2054
2055             /* Update vectorial force */
2056             fix2             = _mm256_add_ps(fix2,tx);
2057             fiy2             = _mm256_add_ps(fiy2,ty);
2058             fiz2             = _mm256_add_ps(fiz2,tz);
2059
2060             fjx3             = _mm256_add_ps(fjx3,tx);
2061             fjy3             = _mm256_add_ps(fjy3,ty);
2062             fjz3             = _mm256_add_ps(fjz3,tz);
2063
2064             }
2065
2066             /**************************
2067              * CALCULATE INTERACTIONS *
2068              **************************/
2069
2070             if (gmx_mm256_any_lt(rsq31,rcutoff2))
2071             {
2072
2073             r31              = _mm256_mul_ps(rsq31,rinv31);
2074
2075             /* EWALD ELECTROSTATICS */
2076
2077             /* Analytical PME correction */
2078             zeta2            = _mm256_mul_ps(beta2,rsq31);
2079             rinv3            = _mm256_mul_ps(rinvsq31,rinv31);
2080             pmecorrF         = gmx_mm256_pmecorrF_ps(zeta2);
2081             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
2082             felec            = _mm256_mul_ps(qq31,felec);
2083
2084             cutoff_mask      = _mm256_cmp_ps(rsq31,rcutoff2,_CMP_LT_OQ);
2085
2086             fscal            = felec;
2087
2088             fscal            = _mm256_and_ps(fscal,cutoff_mask);
2089
2090             /* Calculate temporary vectorial force */
2091             tx               = _mm256_mul_ps(fscal,dx31);
2092             ty               = _mm256_mul_ps(fscal,dy31);
2093             tz               = _mm256_mul_ps(fscal,dz31);
2094
2095             /* Update vectorial force */
2096             fix3             = _mm256_add_ps(fix3,tx);
2097             fiy3             = _mm256_add_ps(fiy3,ty);
2098             fiz3             = _mm256_add_ps(fiz3,tz);
2099
2100             fjx1             = _mm256_add_ps(fjx1,tx);
2101             fjy1             = _mm256_add_ps(fjy1,ty);
2102             fjz1             = _mm256_add_ps(fjz1,tz);
2103
2104             }
2105
2106             /**************************
2107              * CALCULATE INTERACTIONS *
2108              **************************/
2109
2110             if (gmx_mm256_any_lt(rsq32,rcutoff2))
2111             {
2112
2113             r32              = _mm256_mul_ps(rsq32,rinv32);
2114
2115             /* EWALD ELECTROSTATICS */
2116
2117             /* Analytical PME correction */
2118             zeta2            = _mm256_mul_ps(beta2,rsq32);
2119             rinv3            = _mm256_mul_ps(rinvsq32,rinv32);
2120             pmecorrF         = gmx_mm256_pmecorrF_ps(zeta2);
2121             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
2122             felec            = _mm256_mul_ps(qq32,felec);
2123
2124             cutoff_mask      = _mm256_cmp_ps(rsq32,rcutoff2,_CMP_LT_OQ);
2125
2126             fscal            = felec;
2127
2128             fscal            = _mm256_and_ps(fscal,cutoff_mask);
2129
2130             /* Calculate temporary vectorial force */
2131             tx               = _mm256_mul_ps(fscal,dx32);
2132             ty               = _mm256_mul_ps(fscal,dy32);
2133             tz               = _mm256_mul_ps(fscal,dz32);
2134
2135             /* Update vectorial force */
2136             fix3             = _mm256_add_ps(fix3,tx);
2137             fiy3             = _mm256_add_ps(fiy3,ty);
2138             fiz3             = _mm256_add_ps(fiz3,tz);
2139
2140             fjx2             = _mm256_add_ps(fjx2,tx);
2141             fjy2             = _mm256_add_ps(fjy2,ty);
2142             fjz2             = _mm256_add_ps(fjz2,tz);
2143
2144             }
2145
2146             /**************************
2147              * CALCULATE INTERACTIONS *
2148              **************************/
2149
2150             if (gmx_mm256_any_lt(rsq33,rcutoff2))
2151             {
2152
2153             r33              = _mm256_mul_ps(rsq33,rinv33);
2154
2155             /* EWALD ELECTROSTATICS */
2156
2157             /* Analytical PME correction */
2158             zeta2            = _mm256_mul_ps(beta2,rsq33);
2159             rinv3            = _mm256_mul_ps(rinvsq33,rinv33);
2160             pmecorrF         = gmx_mm256_pmecorrF_ps(zeta2);
2161             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
2162             felec            = _mm256_mul_ps(qq33,felec);
2163
2164             cutoff_mask      = _mm256_cmp_ps(rsq33,rcutoff2,_CMP_LT_OQ);
2165
2166             fscal            = felec;
2167
2168             fscal            = _mm256_and_ps(fscal,cutoff_mask);
2169
2170             /* Calculate temporary vectorial force */
2171             tx               = _mm256_mul_ps(fscal,dx33);
2172             ty               = _mm256_mul_ps(fscal,dy33);
2173             tz               = _mm256_mul_ps(fscal,dz33);
2174
2175             /* Update vectorial force */
2176             fix3             = _mm256_add_ps(fix3,tx);
2177             fiy3             = _mm256_add_ps(fiy3,ty);
2178             fiz3             = _mm256_add_ps(fiz3,tz);
2179
2180             fjx3             = _mm256_add_ps(fjx3,tx);
2181             fjy3             = _mm256_add_ps(fjy3,ty);
2182             fjz3             = _mm256_add_ps(fjz3,tz);
2183
2184             }
2185
2186             fjptrA             = f+j_coord_offsetA;
2187             fjptrB             = f+j_coord_offsetB;
2188             fjptrC             = f+j_coord_offsetC;
2189             fjptrD             = f+j_coord_offsetD;
2190             fjptrE             = f+j_coord_offsetE;
2191             fjptrF             = f+j_coord_offsetF;
2192             fjptrG             = f+j_coord_offsetG;
2193             fjptrH             = f+j_coord_offsetH;
2194
2195             gmx_mm256_decrement_4rvec_8ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjptrE,fjptrF,fjptrG,fjptrH,
2196                                                       fjx0,fjy0,fjz0,fjx1,fjy1,fjz1,
2197                                                       fjx2,fjy2,fjz2,fjx3,fjy3,fjz3);
2198
2199             /* Inner loop uses 564 flops */
2200         }
2201
2202         if(jidx<j_index_end)
2203         {
2204
2205             /* Get j neighbor index, and coordinate index */
2206             jnrlistA         = jjnr[jidx];
2207             jnrlistB         = jjnr[jidx+1];
2208             jnrlistC         = jjnr[jidx+2];
2209             jnrlistD         = jjnr[jidx+3];
2210             jnrlistE         = jjnr[jidx+4];
2211             jnrlistF         = jjnr[jidx+5];
2212             jnrlistG         = jjnr[jidx+6];
2213             jnrlistH         = jjnr[jidx+7];
2214             /* Sign of each element will be negative for non-real atoms.
2215              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
2216              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
2217              */
2218             dummy_mask = gmx_mm256_set_m128(gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx+4)),_mm_setzero_si128())),
2219                                             gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128())));
2220
2221             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
2222             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
2223             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
2224             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
2225             jnrE       = (jnrlistE>=0) ? jnrlistE : 0;
2226             jnrF       = (jnrlistF>=0) ? jnrlistF : 0;
2227             jnrG       = (jnrlistG>=0) ? jnrlistG : 0;
2228             jnrH       = (jnrlistH>=0) ? jnrlistH : 0;
2229             j_coord_offsetA  = DIM*jnrA;
2230             j_coord_offsetB  = DIM*jnrB;
2231             j_coord_offsetC  = DIM*jnrC;
2232             j_coord_offsetD  = DIM*jnrD;
2233             j_coord_offsetE  = DIM*jnrE;
2234             j_coord_offsetF  = DIM*jnrF;
2235             j_coord_offsetG  = DIM*jnrG;
2236             j_coord_offsetH  = DIM*jnrH;
2237
2238             /* load j atom coordinates */
2239             gmx_mm256_load_4rvec_8ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
2240                                                  x+j_coord_offsetC,x+j_coord_offsetD,
2241                                                  x+j_coord_offsetE,x+j_coord_offsetF,
2242                                                  x+j_coord_offsetG,x+j_coord_offsetH,
2243                                                  &jx0,&jy0,&jz0,&jx1,&jy1,&jz1,&jx2,
2244                                                  &jy2,&jz2,&jx3,&jy3,&jz3);
2245
2246             /* Calculate displacement vector */
2247             dx00             = _mm256_sub_ps(ix0,jx0);
2248             dy00             = _mm256_sub_ps(iy0,jy0);
2249             dz00             = _mm256_sub_ps(iz0,jz0);
2250             dx11             = _mm256_sub_ps(ix1,jx1);
2251             dy11             = _mm256_sub_ps(iy1,jy1);
2252             dz11             = _mm256_sub_ps(iz1,jz1);
2253             dx12             = _mm256_sub_ps(ix1,jx2);
2254             dy12             = _mm256_sub_ps(iy1,jy2);
2255             dz12             = _mm256_sub_ps(iz1,jz2);
2256             dx13             = _mm256_sub_ps(ix1,jx3);
2257             dy13             = _mm256_sub_ps(iy1,jy3);
2258             dz13             = _mm256_sub_ps(iz1,jz3);
2259             dx21             = _mm256_sub_ps(ix2,jx1);
2260             dy21             = _mm256_sub_ps(iy2,jy1);
2261             dz21             = _mm256_sub_ps(iz2,jz1);
2262             dx22             = _mm256_sub_ps(ix2,jx2);
2263             dy22             = _mm256_sub_ps(iy2,jy2);
2264             dz22             = _mm256_sub_ps(iz2,jz2);
2265             dx23             = _mm256_sub_ps(ix2,jx3);
2266             dy23             = _mm256_sub_ps(iy2,jy3);
2267             dz23             = _mm256_sub_ps(iz2,jz3);
2268             dx31             = _mm256_sub_ps(ix3,jx1);
2269             dy31             = _mm256_sub_ps(iy3,jy1);
2270             dz31             = _mm256_sub_ps(iz3,jz1);
2271             dx32             = _mm256_sub_ps(ix3,jx2);
2272             dy32             = _mm256_sub_ps(iy3,jy2);
2273             dz32             = _mm256_sub_ps(iz3,jz2);
2274             dx33             = _mm256_sub_ps(ix3,jx3);
2275             dy33             = _mm256_sub_ps(iy3,jy3);
2276             dz33             = _mm256_sub_ps(iz3,jz3);
2277
2278             /* Calculate squared distance and things based on it */
2279             rsq00            = gmx_mm256_calc_rsq_ps(dx00,dy00,dz00);
2280             rsq11            = gmx_mm256_calc_rsq_ps(dx11,dy11,dz11);
2281             rsq12            = gmx_mm256_calc_rsq_ps(dx12,dy12,dz12);
2282             rsq13            = gmx_mm256_calc_rsq_ps(dx13,dy13,dz13);
2283             rsq21            = gmx_mm256_calc_rsq_ps(dx21,dy21,dz21);
2284             rsq22            = gmx_mm256_calc_rsq_ps(dx22,dy22,dz22);
2285             rsq23            = gmx_mm256_calc_rsq_ps(dx23,dy23,dz23);
2286             rsq31            = gmx_mm256_calc_rsq_ps(dx31,dy31,dz31);
2287             rsq32            = gmx_mm256_calc_rsq_ps(dx32,dy32,dz32);
2288             rsq33            = gmx_mm256_calc_rsq_ps(dx33,dy33,dz33);
2289
2290             rinv11           = gmx_mm256_invsqrt_ps(rsq11);
2291             rinv12           = gmx_mm256_invsqrt_ps(rsq12);
2292             rinv13           = gmx_mm256_invsqrt_ps(rsq13);
2293             rinv21           = gmx_mm256_invsqrt_ps(rsq21);
2294             rinv22           = gmx_mm256_invsqrt_ps(rsq22);
2295             rinv23           = gmx_mm256_invsqrt_ps(rsq23);
2296             rinv31           = gmx_mm256_invsqrt_ps(rsq31);
2297             rinv32           = gmx_mm256_invsqrt_ps(rsq32);
2298             rinv33           = gmx_mm256_invsqrt_ps(rsq33);
2299
2300             rinvsq00         = gmx_mm256_inv_ps(rsq00);
2301             rinvsq11         = _mm256_mul_ps(rinv11,rinv11);
2302             rinvsq12         = _mm256_mul_ps(rinv12,rinv12);
2303             rinvsq13         = _mm256_mul_ps(rinv13,rinv13);
2304             rinvsq21         = _mm256_mul_ps(rinv21,rinv21);
2305             rinvsq22         = _mm256_mul_ps(rinv22,rinv22);
2306             rinvsq23         = _mm256_mul_ps(rinv23,rinv23);
2307             rinvsq31         = _mm256_mul_ps(rinv31,rinv31);
2308             rinvsq32         = _mm256_mul_ps(rinv32,rinv32);
2309             rinvsq33         = _mm256_mul_ps(rinv33,rinv33);
2310
2311             fjx0             = _mm256_setzero_ps();
2312             fjy0             = _mm256_setzero_ps();
2313             fjz0             = _mm256_setzero_ps();
2314             fjx1             = _mm256_setzero_ps();
2315             fjy1             = _mm256_setzero_ps();
2316             fjz1             = _mm256_setzero_ps();
2317             fjx2             = _mm256_setzero_ps();
2318             fjy2             = _mm256_setzero_ps();
2319             fjz2             = _mm256_setzero_ps();
2320             fjx3             = _mm256_setzero_ps();
2321             fjy3             = _mm256_setzero_ps();
2322             fjz3             = _mm256_setzero_ps();
2323
2324             /**************************
2325              * CALCULATE INTERACTIONS *
2326              **************************/
2327
2328             if (gmx_mm256_any_lt(rsq00,rcutoff2))
2329             {
2330
2331             /* LENNARD-JONES DISPERSION/REPULSION */
2332
2333             rinvsix          = _mm256_mul_ps(_mm256_mul_ps(rinvsq00,rinvsq00),rinvsq00);
2334             fvdw             = _mm256_mul_ps(_mm256_sub_ps(_mm256_mul_ps(c12_00,rinvsix),c6_00),_mm256_mul_ps(rinvsix,rinvsq00));
2335
2336             cutoff_mask      = _mm256_cmp_ps(rsq00,rcutoff2,_CMP_LT_OQ);
2337
2338             fscal            = fvdw;
2339
2340             fscal            = _mm256_and_ps(fscal,cutoff_mask);
2341
2342             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
2343
2344             /* Calculate temporary vectorial force */
2345             tx               = _mm256_mul_ps(fscal,dx00);
2346             ty               = _mm256_mul_ps(fscal,dy00);
2347             tz               = _mm256_mul_ps(fscal,dz00);
2348
2349             /* Update vectorial force */
2350             fix0             = _mm256_add_ps(fix0,tx);
2351             fiy0             = _mm256_add_ps(fiy0,ty);
2352             fiz0             = _mm256_add_ps(fiz0,tz);
2353
2354             fjx0             = _mm256_add_ps(fjx0,tx);
2355             fjy0             = _mm256_add_ps(fjy0,ty);
2356             fjz0             = _mm256_add_ps(fjz0,tz);
2357
2358             }
2359
2360             /**************************
2361              * CALCULATE INTERACTIONS *
2362              **************************/
2363
2364             if (gmx_mm256_any_lt(rsq11,rcutoff2))
2365             {
2366
2367             r11              = _mm256_mul_ps(rsq11,rinv11);
2368             r11              = _mm256_andnot_ps(dummy_mask,r11);
2369
2370             /* EWALD ELECTROSTATICS */
2371
2372             /* Analytical PME correction */
2373             zeta2            = _mm256_mul_ps(beta2,rsq11);
2374             rinv3            = _mm256_mul_ps(rinvsq11,rinv11);
2375             pmecorrF         = gmx_mm256_pmecorrF_ps(zeta2);
2376             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
2377             felec            = _mm256_mul_ps(qq11,felec);
2378
2379             cutoff_mask      = _mm256_cmp_ps(rsq11,rcutoff2,_CMP_LT_OQ);
2380
2381             fscal            = felec;
2382
2383             fscal            = _mm256_and_ps(fscal,cutoff_mask);
2384
2385             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
2386
2387             /* Calculate temporary vectorial force */
2388             tx               = _mm256_mul_ps(fscal,dx11);
2389             ty               = _mm256_mul_ps(fscal,dy11);
2390             tz               = _mm256_mul_ps(fscal,dz11);
2391
2392             /* Update vectorial force */
2393             fix1             = _mm256_add_ps(fix1,tx);
2394             fiy1             = _mm256_add_ps(fiy1,ty);
2395             fiz1             = _mm256_add_ps(fiz1,tz);
2396
2397             fjx1             = _mm256_add_ps(fjx1,tx);
2398             fjy1             = _mm256_add_ps(fjy1,ty);
2399             fjz1             = _mm256_add_ps(fjz1,tz);
2400
2401             }
2402
2403             /**************************
2404              * CALCULATE INTERACTIONS *
2405              **************************/
2406
2407             if (gmx_mm256_any_lt(rsq12,rcutoff2))
2408             {
2409
2410             r12              = _mm256_mul_ps(rsq12,rinv12);
2411             r12              = _mm256_andnot_ps(dummy_mask,r12);
2412
2413             /* EWALD ELECTROSTATICS */
2414
2415             /* Analytical PME correction */
2416             zeta2            = _mm256_mul_ps(beta2,rsq12);
2417             rinv3            = _mm256_mul_ps(rinvsq12,rinv12);
2418             pmecorrF         = gmx_mm256_pmecorrF_ps(zeta2);
2419             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
2420             felec            = _mm256_mul_ps(qq12,felec);
2421
2422             cutoff_mask      = _mm256_cmp_ps(rsq12,rcutoff2,_CMP_LT_OQ);
2423
2424             fscal            = felec;
2425
2426             fscal            = _mm256_and_ps(fscal,cutoff_mask);
2427
2428             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
2429
2430             /* Calculate temporary vectorial force */
2431             tx               = _mm256_mul_ps(fscal,dx12);
2432             ty               = _mm256_mul_ps(fscal,dy12);
2433             tz               = _mm256_mul_ps(fscal,dz12);
2434
2435             /* Update vectorial force */
2436             fix1             = _mm256_add_ps(fix1,tx);
2437             fiy1             = _mm256_add_ps(fiy1,ty);
2438             fiz1             = _mm256_add_ps(fiz1,tz);
2439
2440             fjx2             = _mm256_add_ps(fjx2,tx);
2441             fjy2             = _mm256_add_ps(fjy2,ty);
2442             fjz2             = _mm256_add_ps(fjz2,tz);
2443
2444             }
2445
2446             /**************************
2447              * CALCULATE INTERACTIONS *
2448              **************************/
2449
2450             if (gmx_mm256_any_lt(rsq13,rcutoff2))
2451             {
2452
2453             r13              = _mm256_mul_ps(rsq13,rinv13);
2454             r13              = _mm256_andnot_ps(dummy_mask,r13);
2455
2456             /* EWALD ELECTROSTATICS */
2457
2458             /* Analytical PME correction */
2459             zeta2            = _mm256_mul_ps(beta2,rsq13);
2460             rinv3            = _mm256_mul_ps(rinvsq13,rinv13);
2461             pmecorrF         = gmx_mm256_pmecorrF_ps(zeta2);
2462             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
2463             felec            = _mm256_mul_ps(qq13,felec);
2464
2465             cutoff_mask      = _mm256_cmp_ps(rsq13,rcutoff2,_CMP_LT_OQ);
2466
2467             fscal            = felec;
2468
2469             fscal            = _mm256_and_ps(fscal,cutoff_mask);
2470
2471             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
2472
2473             /* Calculate temporary vectorial force */
2474             tx               = _mm256_mul_ps(fscal,dx13);
2475             ty               = _mm256_mul_ps(fscal,dy13);
2476             tz               = _mm256_mul_ps(fscal,dz13);
2477
2478             /* Update vectorial force */
2479             fix1             = _mm256_add_ps(fix1,tx);
2480             fiy1             = _mm256_add_ps(fiy1,ty);
2481             fiz1             = _mm256_add_ps(fiz1,tz);
2482
2483             fjx3             = _mm256_add_ps(fjx3,tx);
2484             fjy3             = _mm256_add_ps(fjy3,ty);
2485             fjz3             = _mm256_add_ps(fjz3,tz);
2486
2487             }
2488
2489             /**************************
2490              * CALCULATE INTERACTIONS *
2491              **************************/
2492
2493             if (gmx_mm256_any_lt(rsq21,rcutoff2))
2494             {
2495
2496             r21              = _mm256_mul_ps(rsq21,rinv21);
2497             r21              = _mm256_andnot_ps(dummy_mask,r21);
2498
2499             /* EWALD ELECTROSTATICS */
2500
2501             /* Analytical PME correction */
2502             zeta2            = _mm256_mul_ps(beta2,rsq21);
2503             rinv3            = _mm256_mul_ps(rinvsq21,rinv21);
2504             pmecorrF         = gmx_mm256_pmecorrF_ps(zeta2);
2505             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
2506             felec            = _mm256_mul_ps(qq21,felec);
2507
2508             cutoff_mask      = _mm256_cmp_ps(rsq21,rcutoff2,_CMP_LT_OQ);
2509
2510             fscal            = felec;
2511
2512             fscal            = _mm256_and_ps(fscal,cutoff_mask);
2513
2514             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
2515
2516             /* Calculate temporary vectorial force */
2517             tx               = _mm256_mul_ps(fscal,dx21);
2518             ty               = _mm256_mul_ps(fscal,dy21);
2519             tz               = _mm256_mul_ps(fscal,dz21);
2520
2521             /* Update vectorial force */
2522             fix2             = _mm256_add_ps(fix2,tx);
2523             fiy2             = _mm256_add_ps(fiy2,ty);
2524             fiz2             = _mm256_add_ps(fiz2,tz);
2525
2526             fjx1             = _mm256_add_ps(fjx1,tx);
2527             fjy1             = _mm256_add_ps(fjy1,ty);
2528             fjz1             = _mm256_add_ps(fjz1,tz);
2529
2530             }
2531
2532             /**************************
2533              * CALCULATE INTERACTIONS *
2534              **************************/
2535
2536             if (gmx_mm256_any_lt(rsq22,rcutoff2))
2537             {
2538
2539             r22              = _mm256_mul_ps(rsq22,rinv22);
2540             r22              = _mm256_andnot_ps(dummy_mask,r22);
2541
2542             /* EWALD ELECTROSTATICS */
2543
2544             /* Analytical PME correction */
2545             zeta2            = _mm256_mul_ps(beta2,rsq22);
2546             rinv3            = _mm256_mul_ps(rinvsq22,rinv22);
2547             pmecorrF         = gmx_mm256_pmecorrF_ps(zeta2);
2548             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
2549             felec            = _mm256_mul_ps(qq22,felec);
2550
2551             cutoff_mask      = _mm256_cmp_ps(rsq22,rcutoff2,_CMP_LT_OQ);
2552
2553             fscal            = felec;
2554
2555             fscal            = _mm256_and_ps(fscal,cutoff_mask);
2556
2557             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
2558
2559             /* Calculate temporary vectorial force */
2560             tx               = _mm256_mul_ps(fscal,dx22);
2561             ty               = _mm256_mul_ps(fscal,dy22);
2562             tz               = _mm256_mul_ps(fscal,dz22);
2563
2564             /* Update vectorial force */
2565             fix2             = _mm256_add_ps(fix2,tx);
2566             fiy2             = _mm256_add_ps(fiy2,ty);
2567             fiz2             = _mm256_add_ps(fiz2,tz);
2568
2569             fjx2             = _mm256_add_ps(fjx2,tx);
2570             fjy2             = _mm256_add_ps(fjy2,ty);
2571             fjz2             = _mm256_add_ps(fjz2,tz);
2572
2573             }
2574
2575             /**************************
2576              * CALCULATE INTERACTIONS *
2577              **************************/
2578
2579             if (gmx_mm256_any_lt(rsq23,rcutoff2))
2580             {
2581
2582             r23              = _mm256_mul_ps(rsq23,rinv23);
2583             r23              = _mm256_andnot_ps(dummy_mask,r23);
2584
2585             /* EWALD ELECTROSTATICS */
2586
2587             /* Analytical PME correction */
2588             zeta2            = _mm256_mul_ps(beta2,rsq23);
2589             rinv3            = _mm256_mul_ps(rinvsq23,rinv23);
2590             pmecorrF         = gmx_mm256_pmecorrF_ps(zeta2);
2591             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
2592             felec            = _mm256_mul_ps(qq23,felec);
2593
2594             cutoff_mask      = _mm256_cmp_ps(rsq23,rcutoff2,_CMP_LT_OQ);
2595
2596             fscal            = felec;
2597
2598             fscal            = _mm256_and_ps(fscal,cutoff_mask);
2599
2600             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
2601
2602             /* Calculate temporary vectorial force */
2603             tx               = _mm256_mul_ps(fscal,dx23);
2604             ty               = _mm256_mul_ps(fscal,dy23);
2605             tz               = _mm256_mul_ps(fscal,dz23);
2606
2607             /* Update vectorial force */
2608             fix2             = _mm256_add_ps(fix2,tx);
2609             fiy2             = _mm256_add_ps(fiy2,ty);
2610             fiz2             = _mm256_add_ps(fiz2,tz);
2611
2612             fjx3             = _mm256_add_ps(fjx3,tx);
2613             fjy3             = _mm256_add_ps(fjy3,ty);
2614             fjz3             = _mm256_add_ps(fjz3,tz);
2615
2616             }
2617
2618             /**************************
2619              * CALCULATE INTERACTIONS *
2620              **************************/
2621
2622             if (gmx_mm256_any_lt(rsq31,rcutoff2))
2623             {
2624
2625             r31              = _mm256_mul_ps(rsq31,rinv31);
2626             r31              = _mm256_andnot_ps(dummy_mask,r31);
2627
2628             /* EWALD ELECTROSTATICS */
2629
2630             /* Analytical PME correction */
2631             zeta2            = _mm256_mul_ps(beta2,rsq31);
2632             rinv3            = _mm256_mul_ps(rinvsq31,rinv31);
2633             pmecorrF         = gmx_mm256_pmecorrF_ps(zeta2);
2634             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
2635             felec            = _mm256_mul_ps(qq31,felec);
2636
2637             cutoff_mask      = _mm256_cmp_ps(rsq31,rcutoff2,_CMP_LT_OQ);
2638
2639             fscal            = felec;
2640
2641             fscal            = _mm256_and_ps(fscal,cutoff_mask);
2642
2643             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
2644
2645             /* Calculate temporary vectorial force */
2646             tx               = _mm256_mul_ps(fscal,dx31);
2647             ty               = _mm256_mul_ps(fscal,dy31);
2648             tz               = _mm256_mul_ps(fscal,dz31);
2649
2650             /* Update vectorial force */
2651             fix3             = _mm256_add_ps(fix3,tx);
2652             fiy3             = _mm256_add_ps(fiy3,ty);
2653             fiz3             = _mm256_add_ps(fiz3,tz);
2654
2655             fjx1             = _mm256_add_ps(fjx1,tx);
2656             fjy1             = _mm256_add_ps(fjy1,ty);
2657             fjz1             = _mm256_add_ps(fjz1,tz);
2658
2659             }
2660
2661             /**************************
2662              * CALCULATE INTERACTIONS *
2663              **************************/
2664
2665             if (gmx_mm256_any_lt(rsq32,rcutoff2))
2666             {
2667
2668             r32              = _mm256_mul_ps(rsq32,rinv32);
2669             r32              = _mm256_andnot_ps(dummy_mask,r32);
2670
2671             /* EWALD ELECTROSTATICS */
2672
2673             /* Analytical PME correction */
2674             zeta2            = _mm256_mul_ps(beta2,rsq32);
2675             rinv3            = _mm256_mul_ps(rinvsq32,rinv32);
2676             pmecorrF         = gmx_mm256_pmecorrF_ps(zeta2);
2677             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
2678             felec            = _mm256_mul_ps(qq32,felec);
2679
2680             cutoff_mask      = _mm256_cmp_ps(rsq32,rcutoff2,_CMP_LT_OQ);
2681
2682             fscal            = felec;
2683
2684             fscal            = _mm256_and_ps(fscal,cutoff_mask);
2685
2686             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
2687
2688             /* Calculate temporary vectorial force */
2689             tx               = _mm256_mul_ps(fscal,dx32);
2690             ty               = _mm256_mul_ps(fscal,dy32);
2691             tz               = _mm256_mul_ps(fscal,dz32);
2692
2693             /* Update vectorial force */
2694             fix3             = _mm256_add_ps(fix3,tx);
2695             fiy3             = _mm256_add_ps(fiy3,ty);
2696             fiz3             = _mm256_add_ps(fiz3,tz);
2697
2698             fjx2             = _mm256_add_ps(fjx2,tx);
2699             fjy2             = _mm256_add_ps(fjy2,ty);
2700             fjz2             = _mm256_add_ps(fjz2,tz);
2701
2702             }
2703
2704             /**************************
2705              * CALCULATE INTERACTIONS *
2706              **************************/
2707
2708             if (gmx_mm256_any_lt(rsq33,rcutoff2))
2709             {
2710
2711             r33              = _mm256_mul_ps(rsq33,rinv33);
2712             r33              = _mm256_andnot_ps(dummy_mask,r33);
2713
2714             /* EWALD ELECTROSTATICS */
2715
2716             /* Analytical PME correction */
2717             zeta2            = _mm256_mul_ps(beta2,rsq33);
2718             rinv3            = _mm256_mul_ps(rinvsq33,rinv33);
2719             pmecorrF         = gmx_mm256_pmecorrF_ps(zeta2);
2720             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
2721             felec            = _mm256_mul_ps(qq33,felec);
2722
2723             cutoff_mask      = _mm256_cmp_ps(rsq33,rcutoff2,_CMP_LT_OQ);
2724
2725             fscal            = felec;
2726
2727             fscal            = _mm256_and_ps(fscal,cutoff_mask);
2728
2729             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
2730
2731             /* Calculate temporary vectorial force */
2732             tx               = _mm256_mul_ps(fscal,dx33);
2733             ty               = _mm256_mul_ps(fscal,dy33);
2734             tz               = _mm256_mul_ps(fscal,dz33);
2735
2736             /* Update vectorial force */
2737             fix3             = _mm256_add_ps(fix3,tx);
2738             fiy3             = _mm256_add_ps(fiy3,ty);
2739             fiz3             = _mm256_add_ps(fiz3,tz);
2740
2741             fjx3             = _mm256_add_ps(fjx3,tx);
2742             fjy3             = _mm256_add_ps(fjy3,ty);
2743             fjz3             = _mm256_add_ps(fjz3,tz);
2744
2745             }
2746
2747             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
2748             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
2749             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
2750             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
2751             fjptrE             = (jnrlistE>=0) ? f+j_coord_offsetE : scratch;
2752             fjptrF             = (jnrlistF>=0) ? f+j_coord_offsetF : scratch;
2753             fjptrG             = (jnrlistG>=0) ? f+j_coord_offsetG : scratch;
2754             fjptrH             = (jnrlistH>=0) ? f+j_coord_offsetH : scratch;
2755
2756             gmx_mm256_decrement_4rvec_8ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjptrE,fjptrF,fjptrG,fjptrH,
2757                                                       fjx0,fjy0,fjz0,fjx1,fjy1,fjz1,
2758                                                       fjx2,fjy2,fjz2,fjx3,fjy3,fjz3);
2759
2760             /* Inner loop uses 573 flops */
2761         }
2762
2763         /* End of innermost loop */
2764
2765         gmx_mm256_update_iforce_4atom_swizzle_ps(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,fix3,fiy3,fiz3,
2766                                                  f+i_coord_offset,fshift+i_shift_offset);
2767
2768         /* Increment number of inner iterations */
2769         inneriter                  += j_index_end - j_index_start;
2770
2771         /* Outer loop uses 24 flops */
2772     }
2773
2774     /* Increment number of outer iterations */
2775     outeriter        += nri;
2776
2777     /* Update outer/inner flops */
2778
2779     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W4W4_F,outeriter*24 + inneriter*573);
2780 }