src/gmxlib/nonbonded/nb_kernel_avx_256_double/nb_kernel_ElecEwSh_VdwLJSh_GeomW4W4_avx_256_double.c

   1 /*
   2  * Note: this file was generated by the Gromacs avx_256_double 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_double.h"
  34 #include "kernelutil_x86_avx_256_double.h"
  35
  36 /*
  37  * Gromacs nonbonded kernel:   nb_kernel_ElecEwSh_VdwLJSh_GeomW4W4_VF_avx_256_double
  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_double
  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 refer to j loop unrolling done with AVX, e.g. for the four 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              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
  62     int              jnrlistE,jnrlistF,jnrlistG,jnrlistH;
  63     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
  64     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
  65     real             rcutoff_scalar;
  66     real             *shiftvec,*fshift,*x,*f;
  67     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD;
  68     real             scratch[4*DIM];
  69     __m256d          tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
  70     real *           vdwioffsetptr0;
  71     __m256d          ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
  72     real *           vdwioffsetptr1;
  73     __m256d          ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
  74     real *           vdwioffsetptr2;
  75     __m256d          ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
  76     real *           vdwioffsetptr3;
  77     __m256d          ix3,iy3,iz3,fix3,fiy3,fiz3,iq3,isai3;
  78     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
  79     __m256d          jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
  80     int              vdwjidx1A,vdwjidx1B,vdwjidx1C,vdwjidx1D;
  81     __m256d          jx1,jy1,jz1,fjx1,fjy1,fjz1,jq1,isaj1;
  82     int              vdwjidx2A,vdwjidx2B,vdwjidx2C,vdwjidx2D;
  83     __m256d          jx2,jy2,jz2,fjx2,fjy2,fjz2,jq2,isaj2;
  84     int              vdwjidx3A,vdwjidx3B,vdwjidx3C,vdwjidx3D;
  85     __m256d          jx3,jy3,jz3,fjx3,fjy3,fjz3,jq3,isaj3;
  86     __m256d          dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
  87     __m256d          dx11,dy11,dz11,rsq11,rinv11,rinvsq11,r11,qq11,c6_11,c12_11;
  88     __m256d          dx12,dy12,dz12,rsq12,rinv12,rinvsq12,r12,qq12,c6_12,c12_12;
  89     __m256d          dx13,dy13,dz13,rsq13,rinv13,rinvsq13,r13,qq13,c6_13,c12_13;
  90     __m256d          dx21,dy21,dz21,rsq21,rinv21,rinvsq21,r21,qq21,c6_21,c12_21;
  91     __m256d          dx22,dy22,dz22,rsq22,rinv22,rinvsq22,r22,qq22,c6_22,c12_22;
  92     __m256d          dx23,dy23,dz23,rsq23,rinv23,rinvsq23,r23,qq23,c6_23,c12_23;
  93     __m256d          dx31,dy31,dz31,rsq31,rinv31,rinvsq31,r31,qq31,c6_31,c12_31;
  94     __m256d          dx32,dy32,dz32,rsq32,rinv32,rinvsq32,r32,qq32,c6_32,c12_32;
  95     __m256d          dx33,dy33,dz33,rsq33,rinv33,rinvsq33,r33,qq33,c6_33,c12_33;
  96     __m256d          velec,felec,velecsum,facel,crf,krf,krf2;
  97     real             *charge;
  98     int              nvdwtype;
  99     __m256d          rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
 100     int              *vdwtype;
 101     real             *vdwparam;
 102     __m256d          one_sixth   = _mm256_set1_pd(1.0/6.0);
 103     __m256d          one_twelfth = _mm256_set1_pd(1.0/12.0);
 104     __m128i          ewitab;
 105     __m256d          ewtabscale,eweps,sh_ewald,ewrt,ewtabhalfspace,ewtabF,ewtabFn,ewtabD,ewtabV;
 106     __m256d          beta,beta2,beta3,zeta2,pmecorrF,pmecorrV,rinv3;
 107     real             *ewtab;
 108     __m256d          dummy_mask,cutoff_mask;
 109     __m128           tmpmask0,tmpmask1;
 110     __m256d          signbit = _mm256_castsi256_pd( _mm256_set1_epi32(0x80000000) );
 111     __m256d          one     = _mm256_set1_pd(1.0);
 112     __m256d          two     = _mm256_set1_pd(2.0);
 113     x                = xx[0];
 114     f                = ff[0];
 115
 116     nri              = nlist->nri;
 117     iinr             = nlist->iinr;
 118     jindex           = nlist->jindex;
 119     jjnr             = nlist->jjnr;
 120     shiftidx         = nlist->shift;
 121     gid              = nlist->gid;
 122     shiftvec         = fr->shift_vec[0];
 123     fshift           = fr->fshift[0];
 124     facel            = _mm256_set1_pd(fr->epsfac);
 125     charge           = mdatoms->chargeA;
 126     nvdwtype         = fr->ntype;
 127     vdwparam         = fr->nbfp;
 128     vdwtype          = mdatoms->typeA;
 129
 130     sh_ewald         = _mm256_set1_pd(fr->ic->sh_ewald);
 131     beta             = _mm256_set1_pd(fr->ic->ewaldcoeff);
 132     beta2            = _mm256_mul_pd(beta,beta);
 133     beta3            = _mm256_mul_pd(beta,beta2);
 134
 135     ewtab            = fr->ic->tabq_coul_FDV0;
 136     ewtabscale       = _mm256_set1_pd(fr->ic->tabq_scale);
 137     ewtabhalfspace   = _mm256_set1_pd(0.5/fr->ic->tabq_scale);
 138
 139     /* Setup water-specific parameters */
 140     inr              = nlist->iinr[0];
 141     iq1              = _mm256_mul_pd(facel,_mm256_set1_pd(charge[inr+1]));
 142     iq2              = _mm256_mul_pd(facel,_mm256_set1_pd(charge[inr+2]));
 143     iq3              = _mm256_mul_pd(facel,_mm256_set1_pd(charge[inr+3]));
 144     vdwioffsetptr0   = vdwparam+2*nvdwtype*vdwtype[inr+0];
 145
 146     jq1              = _mm256_set1_pd(charge[inr+1]);
 147     jq2              = _mm256_set1_pd(charge[inr+2]);
 148     jq3              = _mm256_set1_pd(charge[inr+3]);
 149     vdwjidx0A        = 2*vdwtype[inr+0];
 150     c6_00            = _mm256_set1_pd(vdwioffsetptr0[vdwjidx0A]);
 151     c12_00           = _mm256_set1_pd(vdwioffsetptr0[vdwjidx0A+1]);
 152     qq11             = _mm256_mul_pd(iq1,jq1);
 153     qq12             = _mm256_mul_pd(iq1,jq2);
 154     qq13             = _mm256_mul_pd(iq1,jq3);
 155     qq21             = _mm256_mul_pd(iq2,jq1);
 156     qq22             = _mm256_mul_pd(iq2,jq2);
 157     qq23             = _mm256_mul_pd(iq2,jq3);
 158     qq31             = _mm256_mul_pd(iq3,jq1);
 159     qq32             = _mm256_mul_pd(iq3,jq2);
 160     qq33             = _mm256_mul_pd(iq3,jq3);
 161
 162     /* When we use explicit cutoffs the value must be identical for elec and VdW, so use elec as an arbitrary choice */
 163     rcutoff_scalar   = fr->rcoulomb;
 164     rcutoff          = _mm256_set1_pd(rcutoff_scalar);
 165     rcutoff2         = _mm256_mul_pd(rcutoff,rcutoff);
 166
 167     sh_vdw_invrcut6  = _mm256_set1_pd(fr->ic->sh_invrc6);
 168     rvdw             = _mm256_set1_pd(fr->rvdw);
 169
 170     /* Avoid stupid compiler warnings */
 171     jnrA = jnrB = jnrC = jnrD = 0;
 172     j_coord_offsetA = 0;
 173     j_coord_offsetB = 0;
 174     j_coord_offsetC = 0;
 175     j_coord_offsetD = 0;
 176
 177     outeriter        = 0;
 178     inneriter        = 0;
 179
 180     for(iidx=0;iidx<4*DIM;iidx++)
 181     {
 182         scratch[iidx] = 0.0;
 183     }
 184
 185     /* Start outer loop over neighborlists */
 186     for(iidx=0; iidx<nri; iidx++)
 187     {
 188         /* Load shift vector for this list */
 189         i_shift_offset   = DIM*shiftidx[iidx];
 190
 191         /* Load limits for loop over neighbors */
 192         j_index_start    = jindex[iidx];
 193         j_index_end      = jindex[iidx+1];
 194
 195         /* Get outer coordinate index */
 196         inr              = iinr[iidx];
 197         i_coord_offset   = DIM*inr;
 198
 199         /* Load i particle coords and add shift vector */
 200         gmx_mm256_load_shift_and_4rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset,
 201                                                     &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2,&ix3,&iy3,&iz3);
 202
 203         fix0             = _mm256_setzero_pd();
 204         fiy0             = _mm256_setzero_pd();
 205         fiz0             = _mm256_setzero_pd();
 206         fix1             = _mm256_setzero_pd();
 207         fiy1             = _mm256_setzero_pd();
 208         fiz1             = _mm256_setzero_pd();
 209         fix2             = _mm256_setzero_pd();
 210         fiy2             = _mm256_setzero_pd();
 211         fiz2             = _mm256_setzero_pd();
 212         fix3             = _mm256_setzero_pd();
 213         fiy3             = _mm256_setzero_pd();
 214         fiz3             = _mm256_setzero_pd();
 215
 216         /* Reset potential sums */
 217         velecsum         = _mm256_setzero_pd();
 218         vvdwsum          = _mm256_setzero_pd();
 219
 220         /* Start inner kernel loop */
 221         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
 222         {
 223
 224             /* Get j neighbor index, and coordinate index */
 225             jnrA             = jjnr[jidx];
 226             jnrB             = jjnr[jidx+1];
 227             jnrC             = jjnr[jidx+2];
 228             jnrD             = jjnr[jidx+3];
 229             j_coord_offsetA  = DIM*jnrA;
 230             j_coord_offsetB  = DIM*jnrB;
 231             j_coord_offsetC  = DIM*jnrC;
 232             j_coord_offsetD  = DIM*jnrD;
 233
 234             /* load j atom coordinates */
 235             gmx_mm256_load_4rvec_4ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 236                                                  x+j_coord_offsetC,x+j_coord_offsetD,
 237                                                  &jx0,&jy0,&jz0,&jx1,&jy1,&jz1,&jx2,
 238                                                  &jy2,&jz2,&jx3,&jy3,&jz3);
 239
 240             /* Calculate displacement vector */
 241             dx00             = _mm256_sub_pd(ix0,jx0);
 242             dy00             = _mm256_sub_pd(iy0,jy0);
 243             dz00             = _mm256_sub_pd(iz0,jz0);
 244             dx11             = _mm256_sub_pd(ix1,jx1);
 245             dy11             = _mm256_sub_pd(iy1,jy1);
 246             dz11             = _mm256_sub_pd(iz1,jz1);
 247             dx12             = _mm256_sub_pd(ix1,jx2);
 248             dy12             = _mm256_sub_pd(iy1,jy2);
 249             dz12             = _mm256_sub_pd(iz1,jz2);
 250             dx13             = _mm256_sub_pd(ix1,jx3);
 251             dy13             = _mm256_sub_pd(iy1,jy3);
 252             dz13             = _mm256_sub_pd(iz1,jz3);
 253             dx21             = _mm256_sub_pd(ix2,jx1);
 254             dy21             = _mm256_sub_pd(iy2,jy1);
 255             dz21             = _mm256_sub_pd(iz2,jz1);
 256             dx22             = _mm256_sub_pd(ix2,jx2);
 257             dy22             = _mm256_sub_pd(iy2,jy2);
 258             dz22             = _mm256_sub_pd(iz2,jz2);
 259             dx23             = _mm256_sub_pd(ix2,jx3);
 260             dy23             = _mm256_sub_pd(iy2,jy3);
 261             dz23             = _mm256_sub_pd(iz2,jz3);
 262             dx31             = _mm256_sub_pd(ix3,jx1);
 263             dy31             = _mm256_sub_pd(iy3,jy1);
 264             dz31             = _mm256_sub_pd(iz3,jz1);
 265             dx32             = _mm256_sub_pd(ix3,jx2);
 266             dy32             = _mm256_sub_pd(iy3,jy2);
 267             dz32             = _mm256_sub_pd(iz3,jz2);
 268             dx33             = _mm256_sub_pd(ix3,jx3);
 269             dy33             = _mm256_sub_pd(iy3,jy3);
 270             dz33             = _mm256_sub_pd(iz3,jz3);
 271
 272             /* Calculate squared distance and things based on it */
 273             rsq00            = gmx_mm256_calc_rsq_pd(dx00,dy00,dz00);
 274             rsq11            = gmx_mm256_calc_rsq_pd(dx11,dy11,dz11);
 275             rsq12            = gmx_mm256_calc_rsq_pd(dx12,dy12,dz12);
 276             rsq13            = gmx_mm256_calc_rsq_pd(dx13,dy13,dz13);
 277             rsq21            = gmx_mm256_calc_rsq_pd(dx21,dy21,dz21);
 278             rsq22            = gmx_mm256_calc_rsq_pd(dx22,dy22,dz22);
 279             rsq23            = gmx_mm256_calc_rsq_pd(dx23,dy23,dz23);
 280             rsq31            = gmx_mm256_calc_rsq_pd(dx31,dy31,dz31);
 281             rsq32            = gmx_mm256_calc_rsq_pd(dx32,dy32,dz32);
 282             rsq33            = gmx_mm256_calc_rsq_pd(dx33,dy33,dz33);
 283
 284             rinv11           = gmx_mm256_invsqrt_pd(rsq11);
 285             rinv12           = gmx_mm256_invsqrt_pd(rsq12);
 286             rinv13           = gmx_mm256_invsqrt_pd(rsq13);
 287             rinv21           = gmx_mm256_invsqrt_pd(rsq21);
 288             rinv22           = gmx_mm256_invsqrt_pd(rsq22);
 289             rinv23           = gmx_mm256_invsqrt_pd(rsq23);
 290             rinv31           = gmx_mm256_invsqrt_pd(rsq31);
 291             rinv32           = gmx_mm256_invsqrt_pd(rsq32);
 292             rinv33           = gmx_mm256_invsqrt_pd(rsq33);
 293
 294             rinvsq00         = gmx_mm256_inv_pd(rsq00);
 295             rinvsq11         = _mm256_mul_pd(rinv11,rinv11);
 296             rinvsq12         = _mm256_mul_pd(rinv12,rinv12);
 297             rinvsq13         = _mm256_mul_pd(rinv13,rinv13);
 298             rinvsq21         = _mm256_mul_pd(rinv21,rinv21);
 299             rinvsq22         = _mm256_mul_pd(rinv22,rinv22);
 300             rinvsq23         = _mm256_mul_pd(rinv23,rinv23);
 301             rinvsq31         = _mm256_mul_pd(rinv31,rinv31);
 302             rinvsq32         = _mm256_mul_pd(rinv32,rinv32);
 303             rinvsq33         = _mm256_mul_pd(rinv33,rinv33);
 304
 305             fjx0             = _mm256_setzero_pd();
 306             fjy0             = _mm256_setzero_pd();
 307             fjz0             = _mm256_setzero_pd();
 308             fjx1             = _mm256_setzero_pd();
 309             fjy1             = _mm256_setzero_pd();
 310             fjz1             = _mm256_setzero_pd();
 311             fjx2             = _mm256_setzero_pd();
 312             fjy2             = _mm256_setzero_pd();
 313             fjz2             = _mm256_setzero_pd();
 314             fjx3             = _mm256_setzero_pd();
 315             fjy3             = _mm256_setzero_pd();
 316             fjz3             = _mm256_setzero_pd();
 317
 318             /**************************
 319              * CALCULATE INTERACTIONS *
 320              **************************/
 321
 322             if (gmx_mm256_any_lt(rsq00,rcutoff2))
 323             {
 324
 325             /* LENNARD-JONES DISPERSION/REPULSION */
 326
 327             rinvsix          = _mm256_mul_pd(_mm256_mul_pd(rinvsq00,rinvsq00),rinvsq00);
 328             vvdw6            = _mm256_mul_pd(c6_00,rinvsix);
 329             vvdw12           = _mm256_mul_pd(c12_00,_mm256_mul_pd(rinvsix,rinvsix));
 330             vvdw             = _mm256_sub_pd(_mm256_mul_pd( _mm256_sub_pd(vvdw12 , _mm256_mul_pd(c12_00,_mm256_mul_pd(sh_vdw_invrcut6,sh_vdw_invrcut6))), one_twelfth) ,
 331                                           _mm256_mul_pd( _mm256_sub_pd(vvdw6,_mm256_mul_pd(c6_00,sh_vdw_invrcut6)),one_sixth));
 332             fvdw             = _mm256_mul_pd(_mm256_sub_pd(vvdw12,vvdw6),rinvsq00);
 333
 334             cutoff_mask      = _mm256_cmp_pd(rsq00,rcutoff2,_CMP_LT_OQ);
 335
 336             /* Update potential sum for this i atom from the interaction with this j atom. */
 337             vvdw             = _mm256_and_pd(vvdw,cutoff_mask);
 338             vvdwsum          = _mm256_add_pd(vvdwsum,vvdw);
 339
 340             fscal            = fvdw;
 341
 342             fscal            = _mm256_and_pd(fscal,cutoff_mask);
 343
 344             /* Calculate temporary vectorial force */
 345             tx               = _mm256_mul_pd(fscal,dx00);
 346             ty               = _mm256_mul_pd(fscal,dy00);
 347             tz               = _mm256_mul_pd(fscal,dz00);
 348
 349             /* Update vectorial force */
 350             fix0             = _mm256_add_pd(fix0,tx);
 351             fiy0             = _mm256_add_pd(fiy0,ty);
 352             fiz0             = _mm256_add_pd(fiz0,tz);
 353
 354             fjx0             = _mm256_add_pd(fjx0,tx);
 355             fjy0             = _mm256_add_pd(fjy0,ty);
 356             fjz0             = _mm256_add_pd(fjz0,tz);
 357
 358             }
 359
 360             /**************************
 361              * CALCULATE INTERACTIONS *
 362              **************************/
 363
 364             if (gmx_mm256_any_lt(rsq11,rcutoff2))
 365             {
 366
 367             r11              = _mm256_mul_pd(rsq11,rinv11);
 368
 369             /* EWALD ELECTROSTATICS */
 370
 371             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 372             ewrt             = _mm256_mul_pd(r11,ewtabscale);
 373             ewitab           = _mm256_cvttpd_epi32(ewrt);
 374             eweps            = _mm256_sub_pd(ewrt,_mm256_round_pd(ewrt, _MM_FROUND_FLOOR));
 375             ewitab           = _mm_slli_epi32(ewitab,2);
 376             ewtabF           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
 377             ewtabD           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,1) );
 378             ewtabV           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,2) );
 379             ewtabFn          = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,3) );
 380             GMX_MM256_FULLTRANSPOSE4_PD(ewtabF,ewtabD,ewtabV,ewtabFn);
 381             felec            = _mm256_add_pd(ewtabF,_mm256_mul_pd(eweps,ewtabD));
 382             velec            = _mm256_sub_pd(ewtabV,_mm256_mul_pd(_mm256_mul_pd(ewtabhalfspace,eweps),_mm256_add_pd(ewtabF,felec)));
 383             velec            = _mm256_mul_pd(qq11,_mm256_sub_pd(_mm256_sub_pd(rinv11,sh_ewald),velec));
 384             felec            = _mm256_mul_pd(_mm256_mul_pd(qq11,rinv11),_mm256_sub_pd(rinvsq11,felec));
 385
 386             cutoff_mask      = _mm256_cmp_pd(rsq11,rcutoff2,_CMP_LT_OQ);
 387
 388             /* Update potential sum for this i atom from the interaction with this j atom. */
 389             velec            = _mm256_and_pd(velec,cutoff_mask);
 390             velecsum         = _mm256_add_pd(velecsum,velec);
 391
 392             fscal            = felec;
 393
 394             fscal            = _mm256_and_pd(fscal,cutoff_mask);
 395
 396             /* Calculate temporary vectorial force */
 397             tx               = _mm256_mul_pd(fscal,dx11);
 398             ty               = _mm256_mul_pd(fscal,dy11);
 399             tz               = _mm256_mul_pd(fscal,dz11);
 400
 401             /* Update vectorial force */
 402             fix1             = _mm256_add_pd(fix1,tx);
 403             fiy1             = _mm256_add_pd(fiy1,ty);
 404             fiz1             = _mm256_add_pd(fiz1,tz);
 405
 406             fjx1             = _mm256_add_pd(fjx1,tx);
 407             fjy1             = _mm256_add_pd(fjy1,ty);
 408             fjz1             = _mm256_add_pd(fjz1,tz);
 409
 410             }
 411
 412             /**************************
 413              * CALCULATE INTERACTIONS *
 414              **************************/
 415
 416             if (gmx_mm256_any_lt(rsq12,rcutoff2))
 417             {
 418
 419             r12              = _mm256_mul_pd(rsq12,rinv12);
 420
 421             /* EWALD ELECTROSTATICS */
 422
 423             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 424             ewrt             = _mm256_mul_pd(r12,ewtabscale);
 425             ewitab           = _mm256_cvttpd_epi32(ewrt);
 426             eweps            = _mm256_sub_pd(ewrt,_mm256_round_pd(ewrt, _MM_FROUND_FLOOR));
 427             ewitab           = _mm_slli_epi32(ewitab,2);
 428             ewtabF           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
 429             ewtabD           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,1) );
 430             ewtabV           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,2) );
 431             ewtabFn          = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,3) );
 432             GMX_MM256_FULLTRANSPOSE4_PD(ewtabF,ewtabD,ewtabV,ewtabFn);
 433             felec            = _mm256_add_pd(ewtabF,_mm256_mul_pd(eweps,ewtabD));
 434             velec            = _mm256_sub_pd(ewtabV,_mm256_mul_pd(_mm256_mul_pd(ewtabhalfspace,eweps),_mm256_add_pd(ewtabF,felec)));
 435             velec            = _mm256_mul_pd(qq12,_mm256_sub_pd(_mm256_sub_pd(rinv12,sh_ewald),velec));
 436             felec            = _mm256_mul_pd(_mm256_mul_pd(qq12,rinv12),_mm256_sub_pd(rinvsq12,felec));
 437
 438             cutoff_mask      = _mm256_cmp_pd(rsq12,rcutoff2,_CMP_LT_OQ);
 439
 440             /* Update potential sum for this i atom from the interaction with this j atom. */
 441             velec            = _mm256_and_pd(velec,cutoff_mask);
 442             velecsum         = _mm256_add_pd(velecsum,velec);
 443
 444             fscal            = felec;
 445
 446             fscal            = _mm256_and_pd(fscal,cutoff_mask);
 447
 448             /* Calculate temporary vectorial force */
 449             tx               = _mm256_mul_pd(fscal,dx12);
 450             ty               = _mm256_mul_pd(fscal,dy12);
 451             tz               = _mm256_mul_pd(fscal,dz12);
 452
 453             /* Update vectorial force */
 454             fix1             = _mm256_add_pd(fix1,tx);
 455             fiy1             = _mm256_add_pd(fiy1,ty);
 456             fiz1             = _mm256_add_pd(fiz1,tz);
 457
 458             fjx2             = _mm256_add_pd(fjx2,tx);
 459             fjy2             = _mm256_add_pd(fjy2,ty);
 460             fjz2             = _mm256_add_pd(fjz2,tz);
 461
 462             }
 463
 464             /**************************
 465              * CALCULATE INTERACTIONS *
 466              **************************/
 467
 468             if (gmx_mm256_any_lt(rsq13,rcutoff2))
 469             {
 470
 471             r13              = _mm256_mul_pd(rsq13,rinv13);
 472
 473             /* EWALD ELECTROSTATICS */
 474
 475             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 476             ewrt             = _mm256_mul_pd(r13,ewtabscale);
 477             ewitab           = _mm256_cvttpd_epi32(ewrt);
 478             eweps            = _mm256_sub_pd(ewrt,_mm256_round_pd(ewrt, _MM_FROUND_FLOOR));
 479             ewitab           = _mm_slli_epi32(ewitab,2);
 480             ewtabF           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
 481             ewtabD           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,1) );
 482             ewtabV           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,2) );
 483             ewtabFn          = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,3) );
 484             GMX_MM256_FULLTRANSPOSE4_PD(ewtabF,ewtabD,ewtabV,ewtabFn);
 485             felec            = _mm256_add_pd(ewtabF,_mm256_mul_pd(eweps,ewtabD));
 486             velec            = _mm256_sub_pd(ewtabV,_mm256_mul_pd(_mm256_mul_pd(ewtabhalfspace,eweps),_mm256_add_pd(ewtabF,felec)));
 487             velec            = _mm256_mul_pd(qq13,_mm256_sub_pd(_mm256_sub_pd(rinv13,sh_ewald),velec));
 488             felec            = _mm256_mul_pd(_mm256_mul_pd(qq13,rinv13),_mm256_sub_pd(rinvsq13,felec));
 489
 490             cutoff_mask      = _mm256_cmp_pd(rsq13,rcutoff2,_CMP_LT_OQ);
 491
 492             /* Update potential sum for this i atom from the interaction with this j atom. */
 493             velec            = _mm256_and_pd(velec,cutoff_mask);
 494             velecsum         = _mm256_add_pd(velecsum,velec);
 495
 496             fscal            = felec;
 497
 498             fscal            = _mm256_and_pd(fscal,cutoff_mask);
 499
 500             /* Calculate temporary vectorial force */
 501             tx               = _mm256_mul_pd(fscal,dx13);
 502             ty               = _mm256_mul_pd(fscal,dy13);
 503             tz               = _mm256_mul_pd(fscal,dz13);
 504
 505             /* Update vectorial force */
 506             fix1             = _mm256_add_pd(fix1,tx);
 507             fiy1             = _mm256_add_pd(fiy1,ty);
 508             fiz1             = _mm256_add_pd(fiz1,tz);
 509
 510             fjx3             = _mm256_add_pd(fjx3,tx);
 511             fjy3             = _mm256_add_pd(fjy3,ty);
 512             fjz3             = _mm256_add_pd(fjz3,tz);
 513
 514             }
 515
 516             /**************************
 517              * CALCULATE INTERACTIONS *
 518              **************************/
 519
 520             if (gmx_mm256_any_lt(rsq21,rcutoff2))
 521             {
 522
 523             r21              = _mm256_mul_pd(rsq21,rinv21);
 524
 525             /* EWALD ELECTROSTATICS */
 526
 527             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 528             ewrt             = _mm256_mul_pd(r21,ewtabscale);
 529             ewitab           = _mm256_cvttpd_epi32(ewrt);
 530             eweps            = _mm256_sub_pd(ewrt,_mm256_round_pd(ewrt, _MM_FROUND_FLOOR));
 531             ewitab           = _mm_slli_epi32(ewitab,2);
 532             ewtabF           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
 533             ewtabD           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,1) );
 534             ewtabV           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,2) );
 535             ewtabFn          = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,3) );
 536             GMX_MM256_FULLTRANSPOSE4_PD(ewtabF,ewtabD,ewtabV,ewtabFn);
 537             felec            = _mm256_add_pd(ewtabF,_mm256_mul_pd(eweps,ewtabD));
 538             velec            = _mm256_sub_pd(ewtabV,_mm256_mul_pd(_mm256_mul_pd(ewtabhalfspace,eweps),_mm256_add_pd(ewtabF,felec)));
 539             velec            = _mm256_mul_pd(qq21,_mm256_sub_pd(_mm256_sub_pd(rinv21,sh_ewald),velec));
 540             felec            = _mm256_mul_pd(_mm256_mul_pd(qq21,rinv21),_mm256_sub_pd(rinvsq21,felec));
 541
 542             cutoff_mask      = _mm256_cmp_pd(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_pd(velec,cutoff_mask);
 546             velecsum         = _mm256_add_pd(velecsum,velec);
 547
 548             fscal            = felec;
 549
 550             fscal            = _mm256_and_pd(fscal,cutoff_mask);
 551
 552             /* Calculate temporary vectorial force */
 553             tx               = _mm256_mul_pd(fscal,dx21);
 554             ty               = _mm256_mul_pd(fscal,dy21);
 555             tz               = _mm256_mul_pd(fscal,dz21);
 556
 557             /* Update vectorial force */
 558             fix2             = _mm256_add_pd(fix2,tx);
 559             fiy2             = _mm256_add_pd(fiy2,ty);
 560             fiz2             = _mm256_add_pd(fiz2,tz);
 561
 562             fjx1             = _mm256_add_pd(fjx1,tx);
 563             fjy1             = _mm256_add_pd(fjy1,ty);
 564             fjz1             = _mm256_add_pd(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_pd(rsq22,rinv22);
 576
 577             /* EWALD ELECTROSTATICS */
 578
 579             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 580             ewrt             = _mm256_mul_pd(r22,ewtabscale);
 581             ewitab           = _mm256_cvttpd_epi32(ewrt);
 582             eweps            = _mm256_sub_pd(ewrt,_mm256_round_pd(ewrt, _MM_FROUND_FLOOR));
 583             ewitab           = _mm_slli_epi32(ewitab,2);
 584             ewtabF           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
 585             ewtabD           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,1) );
 586             ewtabV           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,2) );
 587             ewtabFn          = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,3) );
 588             GMX_MM256_FULLTRANSPOSE4_PD(ewtabF,ewtabD,ewtabV,ewtabFn);
 589             felec            = _mm256_add_pd(ewtabF,_mm256_mul_pd(eweps,ewtabD));
 590             velec            = _mm256_sub_pd(ewtabV,_mm256_mul_pd(_mm256_mul_pd(ewtabhalfspace,eweps),_mm256_add_pd(ewtabF,felec)));
 591             velec            = _mm256_mul_pd(qq22,_mm256_sub_pd(_mm256_sub_pd(rinv22,sh_ewald),velec));
 592             felec            = _mm256_mul_pd(_mm256_mul_pd(qq22,rinv22),_mm256_sub_pd(rinvsq22,felec));
 593
 594             cutoff_mask      = _mm256_cmp_pd(rsq22,rcutoff2,_CMP_LT_OQ);
 595
 596             /* Update potential sum for this i atom from the interaction with this j atom. */
 597             velec            = _mm256_and_pd(velec,cutoff_mask);
 598             velecsum         = _mm256_add_pd(velecsum,velec);
 599
 600             fscal            = felec;
 601
 602             fscal            = _mm256_and_pd(fscal,cutoff_mask);
 603
 604             /* Calculate temporary vectorial force */
 605             tx               = _mm256_mul_pd(fscal,dx22);
 606             ty               = _mm256_mul_pd(fscal,dy22);
 607             tz               = _mm256_mul_pd(fscal,dz22);
 608
 609             /* Update vectorial force */
 610             fix2             = _mm256_add_pd(fix2,tx);
 611             fiy2             = _mm256_add_pd(fiy2,ty);
 612             fiz2             = _mm256_add_pd(fiz2,tz);
 613
 614             fjx2             = _mm256_add_pd(fjx2,tx);
 615             fjy2             = _mm256_add_pd(fjy2,ty);
 616             fjz2             = _mm256_add_pd(fjz2,tz);
 617
 618             }
 619
 620             /**************************
 621              * CALCULATE INTERACTIONS *
 622              **************************/
 623
 624             if (gmx_mm256_any_lt(rsq23,rcutoff2))
 625             {
 626
 627             r23              = _mm256_mul_pd(rsq23,rinv23);
 628
 629             /* EWALD ELECTROSTATICS */
 630
 631             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 632             ewrt             = _mm256_mul_pd(r23,ewtabscale);
 633             ewitab           = _mm256_cvttpd_epi32(ewrt);
 634             eweps            = _mm256_sub_pd(ewrt,_mm256_round_pd(ewrt, _MM_FROUND_FLOOR));
 635             ewitab           = _mm_slli_epi32(ewitab,2);
 636             ewtabF           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
 637             ewtabD           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,1) );
 638             ewtabV           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,2) );
 639             ewtabFn          = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,3) );
 640             GMX_MM256_FULLTRANSPOSE4_PD(ewtabF,ewtabD,ewtabV,ewtabFn);
 641             felec            = _mm256_add_pd(ewtabF,_mm256_mul_pd(eweps,ewtabD));
 642             velec            = _mm256_sub_pd(ewtabV,_mm256_mul_pd(_mm256_mul_pd(ewtabhalfspace,eweps),_mm256_add_pd(ewtabF,felec)));
 643             velec            = _mm256_mul_pd(qq23,_mm256_sub_pd(_mm256_sub_pd(rinv23,sh_ewald),velec));
 644             felec            = _mm256_mul_pd(_mm256_mul_pd(qq23,rinv23),_mm256_sub_pd(rinvsq23,felec));
 645
 646             cutoff_mask      = _mm256_cmp_pd(rsq23,rcutoff2,_CMP_LT_OQ);
 647
 648             /* Update potential sum for this i atom from the interaction with this j atom. */
 649             velec            = _mm256_and_pd(velec,cutoff_mask);
 650             velecsum         = _mm256_add_pd(velecsum,velec);
 651
 652             fscal            = felec;
 653
 654             fscal            = _mm256_and_pd(fscal,cutoff_mask);
 655
 656             /* Calculate temporary vectorial force */
 657             tx               = _mm256_mul_pd(fscal,dx23);
 658             ty               = _mm256_mul_pd(fscal,dy23);
 659             tz               = _mm256_mul_pd(fscal,dz23);
 660
 661             /* Update vectorial force */
 662             fix2             = _mm256_add_pd(fix2,tx);
 663             fiy2             = _mm256_add_pd(fiy2,ty);
 664             fiz2             = _mm256_add_pd(fiz2,tz);
 665
 666             fjx3             = _mm256_add_pd(fjx3,tx);
 667             fjy3             = _mm256_add_pd(fjy3,ty);
 668             fjz3             = _mm256_add_pd(fjz3,tz);
 669
 670             }
 671
 672             /**************************
 673              * CALCULATE INTERACTIONS *
 674              **************************/
 675
 676             if (gmx_mm256_any_lt(rsq31,rcutoff2))
 677             {
 678
 679             r31              = _mm256_mul_pd(rsq31,rinv31);
 680
 681             /* EWALD ELECTROSTATICS */
 682
 683             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 684             ewrt             = _mm256_mul_pd(r31,ewtabscale);
 685             ewitab           = _mm256_cvttpd_epi32(ewrt);
 686             eweps            = _mm256_sub_pd(ewrt,_mm256_round_pd(ewrt, _MM_FROUND_FLOOR));
 687             ewitab           = _mm_slli_epi32(ewitab,2);
 688             ewtabF           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
 689             ewtabD           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,1) );
 690             ewtabV           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,2) );
 691             ewtabFn          = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,3) );
 692             GMX_MM256_FULLTRANSPOSE4_PD(ewtabF,ewtabD,ewtabV,ewtabFn);
 693             felec            = _mm256_add_pd(ewtabF,_mm256_mul_pd(eweps,ewtabD));
 694             velec            = _mm256_sub_pd(ewtabV,_mm256_mul_pd(_mm256_mul_pd(ewtabhalfspace,eweps),_mm256_add_pd(ewtabF,felec)));
 695             velec            = _mm256_mul_pd(qq31,_mm256_sub_pd(_mm256_sub_pd(rinv31,sh_ewald),velec));
 696             felec            = _mm256_mul_pd(_mm256_mul_pd(qq31,rinv31),_mm256_sub_pd(rinvsq31,felec));
 697
 698             cutoff_mask      = _mm256_cmp_pd(rsq31,rcutoff2,_CMP_LT_OQ);
 699
 700             /* Update potential sum for this i atom from the interaction with this j atom. */
 701             velec            = _mm256_and_pd(velec,cutoff_mask);
 702             velecsum         = _mm256_add_pd(velecsum,velec);
 703
 704             fscal            = felec;
 705
 706             fscal            = _mm256_and_pd(fscal,cutoff_mask);
 707
 708             /* Calculate temporary vectorial force */
 709             tx               = _mm256_mul_pd(fscal,dx31);
 710             ty               = _mm256_mul_pd(fscal,dy31);
 711             tz               = _mm256_mul_pd(fscal,dz31);
 712
 713             /* Update vectorial force */
 714             fix3             = _mm256_add_pd(fix3,tx);
 715             fiy3             = _mm256_add_pd(fiy3,ty);
 716             fiz3             = _mm256_add_pd(fiz3,tz);
 717
 718             fjx1             = _mm256_add_pd(fjx1,tx);
 719             fjy1             = _mm256_add_pd(fjy1,ty);
 720             fjz1             = _mm256_add_pd(fjz1,tz);
 721
 722             }
 723
 724             /**************************
 725              * CALCULATE INTERACTIONS *
 726              **************************/
 727
 728             if (gmx_mm256_any_lt(rsq32,rcutoff2))
 729             {
 730
 731             r32              = _mm256_mul_pd(rsq32,rinv32);
 732
 733             /* EWALD ELECTROSTATICS */
 734
 735             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 736             ewrt             = _mm256_mul_pd(r32,ewtabscale);
 737             ewitab           = _mm256_cvttpd_epi32(ewrt);
 738             eweps            = _mm256_sub_pd(ewrt,_mm256_round_pd(ewrt, _MM_FROUND_FLOOR));
 739             ewitab           = _mm_slli_epi32(ewitab,2);
 740             ewtabF           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
 741             ewtabD           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,1) );
 742             ewtabV           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,2) );
 743             ewtabFn          = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,3) );
 744             GMX_MM256_FULLTRANSPOSE4_PD(ewtabF,ewtabD,ewtabV,ewtabFn);
 745             felec            = _mm256_add_pd(ewtabF,_mm256_mul_pd(eweps,ewtabD));
 746             velec            = _mm256_sub_pd(ewtabV,_mm256_mul_pd(_mm256_mul_pd(ewtabhalfspace,eweps),_mm256_add_pd(ewtabF,felec)));
 747             velec            = _mm256_mul_pd(qq32,_mm256_sub_pd(_mm256_sub_pd(rinv32,sh_ewald),velec));
 748             felec            = _mm256_mul_pd(_mm256_mul_pd(qq32,rinv32),_mm256_sub_pd(rinvsq32,felec));
 749
 750             cutoff_mask      = _mm256_cmp_pd(rsq32,rcutoff2,_CMP_LT_OQ);
 751
 752             /* Update potential sum for this i atom from the interaction with this j atom. */
 753             velec            = _mm256_and_pd(velec,cutoff_mask);
 754             velecsum         = _mm256_add_pd(velecsum,velec);
 755
 756             fscal            = felec;
 757
 758             fscal            = _mm256_and_pd(fscal,cutoff_mask);
 759
 760             /* Calculate temporary vectorial force */
 761             tx               = _mm256_mul_pd(fscal,dx32);
 762             ty               = _mm256_mul_pd(fscal,dy32);
 763             tz               = _mm256_mul_pd(fscal,dz32);
 764
 765             /* Update vectorial force */
 766             fix3             = _mm256_add_pd(fix3,tx);
 767             fiy3             = _mm256_add_pd(fiy3,ty);
 768             fiz3             = _mm256_add_pd(fiz3,tz);
 769
 770             fjx2             = _mm256_add_pd(fjx2,tx);
 771             fjy2             = _mm256_add_pd(fjy2,ty);
 772             fjz2             = _mm256_add_pd(fjz2,tz);
 773
 774             }
 775
 776             /**************************
 777              * CALCULATE INTERACTIONS *
 778              **************************/
 779
 780             if (gmx_mm256_any_lt(rsq33,rcutoff2))
 781             {
 782
 783             r33              = _mm256_mul_pd(rsq33,rinv33);
 784
 785             /* EWALD ELECTROSTATICS */
 786
 787             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 788             ewrt             = _mm256_mul_pd(r33,ewtabscale);
 789             ewitab           = _mm256_cvttpd_epi32(ewrt);
 790             eweps            = _mm256_sub_pd(ewrt,_mm256_round_pd(ewrt, _MM_FROUND_FLOOR));
 791             ewitab           = _mm_slli_epi32(ewitab,2);
 792             ewtabF           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
 793             ewtabD           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,1) );
 794             ewtabV           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,2) );
 795             ewtabFn          = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,3) );
 796             GMX_MM256_FULLTRANSPOSE4_PD(ewtabF,ewtabD,ewtabV,ewtabFn);
 797             felec            = _mm256_add_pd(ewtabF,_mm256_mul_pd(eweps,ewtabD));
 798             velec            = _mm256_sub_pd(ewtabV,_mm256_mul_pd(_mm256_mul_pd(ewtabhalfspace,eweps),_mm256_add_pd(ewtabF,felec)));
 799             velec            = _mm256_mul_pd(qq33,_mm256_sub_pd(_mm256_sub_pd(rinv33,sh_ewald),velec));
 800             felec            = _mm256_mul_pd(_mm256_mul_pd(qq33,rinv33),_mm256_sub_pd(rinvsq33,felec));
 801
 802             cutoff_mask      = _mm256_cmp_pd(rsq33,rcutoff2,_CMP_LT_OQ);
 803
 804             /* Update potential sum for this i atom from the interaction with this j atom. */
 805             velec            = _mm256_and_pd(velec,cutoff_mask);
 806             velecsum         = _mm256_add_pd(velecsum,velec);
 807
 808             fscal            = felec;
 809
 810             fscal            = _mm256_and_pd(fscal,cutoff_mask);
 811
 812             /* Calculate temporary vectorial force */
 813             tx               = _mm256_mul_pd(fscal,dx33);
 814             ty               = _mm256_mul_pd(fscal,dy33);
 815             tz               = _mm256_mul_pd(fscal,dz33);
 816
 817             /* Update vectorial force */
 818             fix3             = _mm256_add_pd(fix3,tx);
 819             fiy3             = _mm256_add_pd(fiy3,ty);
 820             fiz3             = _mm256_add_pd(fiz3,tz);
 821
 822             fjx3             = _mm256_add_pd(fjx3,tx);
 823             fjy3             = _mm256_add_pd(fjy3,ty);
 824             fjz3             = _mm256_add_pd(fjz3,tz);
 825
 826             }
 827
 828             fjptrA             = f+j_coord_offsetA;
 829             fjptrB             = f+j_coord_offsetB;
 830             fjptrC             = f+j_coord_offsetC;
 831             fjptrD             = f+j_coord_offsetD;
 832
 833             gmx_mm256_decrement_4rvec_4ptr_swizzle_pd(fjptrA,fjptrB,fjptrC,fjptrD,
 834                                                       fjx0,fjy0,fjz0,fjx1,fjy1,fjz1,
 835                                                       fjx2,fjy2,fjz2,fjx3,fjy3,fjz3);
 836
 837             /* Inner loop uses 458 flops */
 838         }
 839
 840         if(jidx<j_index_end)
 841         {
 842
 843             /* Get j neighbor index, and coordinate index */
 844             jnrlistA         = jjnr[jidx];
 845             jnrlistB         = jjnr[jidx+1];
 846             jnrlistC         = jjnr[jidx+2];
 847             jnrlistD         = jjnr[jidx+3];
 848             /* Sign of each element will be negative for non-real atoms.
 849              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 850              * so use it as val = _mm_andnot_pd(mask,val) to clear dummy entries.
 851              */
 852             tmpmask0 = gmx_mm_castsi128_pd(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
 853
 854             tmpmask1 = _mm_permute_ps(tmpmask0,_GMX_MM_PERMUTE(3,3,2,2));
 855             tmpmask0 = _mm_permute_ps(tmpmask0,_GMX_MM_PERMUTE(1,1,0,0));
 856             dummy_mask = _mm256_castps_pd(gmx_mm256_set_m128(tmpmask1,tmpmask0));
 857
 858             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
 859             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
 860             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
 861             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
 862             j_coord_offsetA  = DIM*jnrA;
 863             j_coord_offsetB  = DIM*jnrB;
 864             j_coord_offsetC  = DIM*jnrC;
 865             j_coord_offsetD  = DIM*jnrD;
 866
 867             /* load j atom coordinates */
 868             gmx_mm256_load_4rvec_4ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 869                                                  x+j_coord_offsetC,x+j_coord_offsetD,
 870                                                  &jx0,&jy0,&jz0,&jx1,&jy1,&jz1,&jx2,
 871                                                  &jy2,&jz2,&jx3,&jy3,&jz3);
 872
 873             /* Calculate displacement vector */
 874             dx00             = _mm256_sub_pd(ix0,jx0);
 875             dy00             = _mm256_sub_pd(iy0,jy0);
 876             dz00             = _mm256_sub_pd(iz0,jz0);
 877             dx11             = _mm256_sub_pd(ix1,jx1);
 878             dy11             = _mm256_sub_pd(iy1,jy1);
 879             dz11             = _mm256_sub_pd(iz1,jz1);
 880             dx12             = _mm256_sub_pd(ix1,jx2);
 881             dy12             = _mm256_sub_pd(iy1,jy2);
 882             dz12             = _mm256_sub_pd(iz1,jz2);
 883             dx13             = _mm256_sub_pd(ix1,jx3);
 884             dy13             = _mm256_sub_pd(iy1,jy3);
 885             dz13             = _mm256_sub_pd(iz1,jz3);
 886             dx21             = _mm256_sub_pd(ix2,jx1);
 887             dy21             = _mm256_sub_pd(iy2,jy1);
 888             dz21             = _mm256_sub_pd(iz2,jz1);
 889             dx22             = _mm256_sub_pd(ix2,jx2);
 890             dy22             = _mm256_sub_pd(iy2,jy2);
 891             dz22             = _mm256_sub_pd(iz2,jz2);
 892             dx23             = _mm256_sub_pd(ix2,jx3);
 893             dy23             = _mm256_sub_pd(iy2,jy3);
 894             dz23             = _mm256_sub_pd(iz2,jz3);
 895             dx31             = _mm256_sub_pd(ix3,jx1);
 896             dy31             = _mm256_sub_pd(iy3,jy1);
 897             dz31             = _mm256_sub_pd(iz3,jz1);
 898             dx32             = _mm256_sub_pd(ix3,jx2);
 899             dy32             = _mm256_sub_pd(iy3,jy2);
 900             dz32             = _mm256_sub_pd(iz3,jz2);
 901             dx33             = _mm256_sub_pd(ix3,jx3);
 902             dy33             = _mm256_sub_pd(iy3,jy3);
 903             dz33             = _mm256_sub_pd(iz3,jz3);
 904
 905             /* Calculate squared distance and things based on it */
 906             rsq00            = gmx_mm256_calc_rsq_pd(dx00,dy00,dz00);
 907             rsq11            = gmx_mm256_calc_rsq_pd(dx11,dy11,dz11);
 908             rsq12            = gmx_mm256_calc_rsq_pd(dx12,dy12,dz12);
 909             rsq13            = gmx_mm256_calc_rsq_pd(dx13,dy13,dz13);
 910             rsq21            = gmx_mm256_calc_rsq_pd(dx21,dy21,dz21);
 911             rsq22            = gmx_mm256_calc_rsq_pd(dx22,dy22,dz22);
 912             rsq23            = gmx_mm256_calc_rsq_pd(dx23,dy23,dz23);
 913             rsq31            = gmx_mm256_calc_rsq_pd(dx31,dy31,dz31);
 914             rsq32            = gmx_mm256_calc_rsq_pd(dx32,dy32,dz32);
 915             rsq33            = gmx_mm256_calc_rsq_pd(dx33,dy33,dz33);
 916
 917             rinv11           = gmx_mm256_invsqrt_pd(rsq11);
 918             rinv12           = gmx_mm256_invsqrt_pd(rsq12);
 919             rinv13           = gmx_mm256_invsqrt_pd(rsq13);
 920             rinv21           = gmx_mm256_invsqrt_pd(rsq21);
 921             rinv22           = gmx_mm256_invsqrt_pd(rsq22);
 922             rinv23           = gmx_mm256_invsqrt_pd(rsq23);
 923             rinv31           = gmx_mm256_invsqrt_pd(rsq31);
 924             rinv32           = gmx_mm256_invsqrt_pd(rsq32);
 925             rinv33           = gmx_mm256_invsqrt_pd(rsq33);
 926
 927             rinvsq00         = gmx_mm256_inv_pd(rsq00);
 928             rinvsq11         = _mm256_mul_pd(rinv11,rinv11);
 929             rinvsq12         = _mm256_mul_pd(rinv12,rinv12);
 930             rinvsq13         = _mm256_mul_pd(rinv13,rinv13);
 931             rinvsq21         = _mm256_mul_pd(rinv21,rinv21);
 932             rinvsq22         = _mm256_mul_pd(rinv22,rinv22);
 933             rinvsq23         = _mm256_mul_pd(rinv23,rinv23);
 934             rinvsq31         = _mm256_mul_pd(rinv31,rinv31);
 935             rinvsq32         = _mm256_mul_pd(rinv32,rinv32);
 936             rinvsq33         = _mm256_mul_pd(rinv33,rinv33);
 937
 938             fjx0             = _mm256_setzero_pd();
 939             fjy0             = _mm256_setzero_pd();
 940             fjz0             = _mm256_setzero_pd();
 941             fjx1             = _mm256_setzero_pd();
 942             fjy1             = _mm256_setzero_pd();
 943             fjz1             = _mm256_setzero_pd();
 944             fjx2             = _mm256_setzero_pd();
 945             fjy2             = _mm256_setzero_pd();
 946             fjz2             = _mm256_setzero_pd();
 947             fjx3             = _mm256_setzero_pd();
 948             fjy3             = _mm256_setzero_pd();
 949             fjz3             = _mm256_setzero_pd();
 950
 951             /**************************
 952              * CALCULATE INTERACTIONS *
 953              **************************/
 954
 955             if (gmx_mm256_any_lt(rsq00,rcutoff2))
 956             {
 957
 958             /* LENNARD-JONES DISPERSION/REPULSION */
 959
 960             rinvsix          = _mm256_mul_pd(_mm256_mul_pd(rinvsq00,rinvsq00),rinvsq00);
 961             vvdw6            = _mm256_mul_pd(c6_00,rinvsix);
 962             vvdw12           = _mm256_mul_pd(c12_00,_mm256_mul_pd(rinvsix,rinvsix));
 963             vvdw             = _mm256_sub_pd(_mm256_mul_pd( _mm256_sub_pd(vvdw12 , _mm256_mul_pd(c12_00,_mm256_mul_pd(sh_vdw_invrcut6,sh_vdw_invrcut6))), one_twelfth) ,
 964                                           _mm256_mul_pd( _mm256_sub_pd(vvdw6,_mm256_mul_pd(c6_00,sh_vdw_invrcut6)),one_sixth));
 965             fvdw             = _mm256_mul_pd(_mm256_sub_pd(vvdw12,vvdw6),rinvsq00);
 966
 967             cutoff_mask      = _mm256_cmp_pd(rsq00,rcutoff2,_CMP_LT_OQ);
 968
 969             /* Update potential sum for this i atom from the interaction with this j atom. */
 970             vvdw             = _mm256_and_pd(vvdw,cutoff_mask);
 971             vvdw             = _mm256_andnot_pd(dummy_mask,vvdw);
 972             vvdwsum          = _mm256_add_pd(vvdwsum,vvdw);
 973
 974             fscal            = fvdw;
 975
 976             fscal            = _mm256_and_pd(fscal,cutoff_mask);
 977
 978             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
 979
 980             /* Calculate temporary vectorial force */
 981             tx               = _mm256_mul_pd(fscal,dx00);
 982             ty               = _mm256_mul_pd(fscal,dy00);
 983             tz               = _mm256_mul_pd(fscal,dz00);
 984
 985             /* Update vectorial force */
 986             fix0             = _mm256_add_pd(fix0,tx);
 987             fiy0             = _mm256_add_pd(fiy0,ty);
 988             fiz0             = _mm256_add_pd(fiz0,tz);
 989
 990             fjx0             = _mm256_add_pd(fjx0,tx);
 991             fjy0             = _mm256_add_pd(fjy0,ty);
 992             fjz0             = _mm256_add_pd(fjz0,tz);
 993
 994             }
 995
 996             /**************************
 997              * CALCULATE INTERACTIONS *
 998              **************************/
 999
1000             if (gmx_mm256_any_lt(rsq11,rcutoff2))
1001             {
1002
1003             r11              = _mm256_mul_pd(rsq11,rinv11);
1004             r11              = _mm256_andnot_pd(dummy_mask,r11);
1005
1006             /* EWALD ELECTROSTATICS */
1007
1008             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1009             ewrt             = _mm256_mul_pd(r11,ewtabscale);
1010             ewitab           = _mm256_cvttpd_epi32(ewrt);
1011             eweps            = _mm256_sub_pd(ewrt,_mm256_round_pd(ewrt, _MM_FROUND_FLOOR));
1012             ewitab           = _mm_slli_epi32(ewitab,2);
1013             ewtabF           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
1014             ewtabD           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,1) );
1015             ewtabV           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,2) );
1016             ewtabFn          = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,3) );
1017             GMX_MM256_FULLTRANSPOSE4_PD(ewtabF,ewtabD,ewtabV,ewtabFn);
1018             felec            = _mm256_add_pd(ewtabF,_mm256_mul_pd(eweps,ewtabD));
1019             velec            = _mm256_sub_pd(ewtabV,_mm256_mul_pd(_mm256_mul_pd(ewtabhalfspace,eweps),_mm256_add_pd(ewtabF,felec)));
1020             velec            = _mm256_mul_pd(qq11,_mm256_sub_pd(_mm256_sub_pd(rinv11,sh_ewald),velec));
1021             felec            = _mm256_mul_pd(_mm256_mul_pd(qq11,rinv11),_mm256_sub_pd(rinvsq11,felec));
1022
1023             cutoff_mask      = _mm256_cmp_pd(rsq11,rcutoff2,_CMP_LT_OQ);
1024
1025             /* Update potential sum for this i atom from the interaction with this j atom. */
1026             velec            = _mm256_and_pd(velec,cutoff_mask);
1027             velec            = _mm256_andnot_pd(dummy_mask,velec);
1028             velecsum         = _mm256_add_pd(velecsum,velec);
1029
1030             fscal            = felec;
1031
1032             fscal            = _mm256_and_pd(fscal,cutoff_mask);
1033
1034             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
1035
1036             /* Calculate temporary vectorial force */
1037             tx               = _mm256_mul_pd(fscal,dx11);
1038             ty               = _mm256_mul_pd(fscal,dy11);
1039             tz               = _mm256_mul_pd(fscal,dz11);
1040
1041             /* Update vectorial force */
1042             fix1             = _mm256_add_pd(fix1,tx);
1043             fiy1             = _mm256_add_pd(fiy1,ty);
1044             fiz1             = _mm256_add_pd(fiz1,tz);
1045
1046             fjx1             = _mm256_add_pd(fjx1,tx);
1047             fjy1             = _mm256_add_pd(fjy1,ty);
1048             fjz1             = _mm256_add_pd(fjz1,tz);
1049
1050             }
1051
1052             /**************************
1053              * CALCULATE INTERACTIONS *
1054              **************************/
1055
1056             if (gmx_mm256_any_lt(rsq12,rcutoff2))
1057             {
1058
1059             r12              = _mm256_mul_pd(rsq12,rinv12);
1060             r12              = _mm256_andnot_pd(dummy_mask,r12);
1061
1062             /* EWALD ELECTROSTATICS */
1063
1064             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1065             ewrt             = _mm256_mul_pd(r12,ewtabscale);
1066             ewitab           = _mm256_cvttpd_epi32(ewrt);
1067             eweps            = _mm256_sub_pd(ewrt,_mm256_round_pd(ewrt, _MM_FROUND_FLOOR));
1068             ewitab           = _mm_slli_epi32(ewitab,2);
1069             ewtabF           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
1070             ewtabD           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,1) );
1071             ewtabV           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,2) );
1072             ewtabFn          = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,3) );
1073             GMX_MM256_FULLTRANSPOSE4_PD(ewtabF,ewtabD,ewtabV,ewtabFn);
1074             felec            = _mm256_add_pd(ewtabF,_mm256_mul_pd(eweps,ewtabD));
1075             velec            = _mm256_sub_pd(ewtabV,_mm256_mul_pd(_mm256_mul_pd(ewtabhalfspace,eweps),_mm256_add_pd(ewtabF,felec)));
1076             velec            = _mm256_mul_pd(qq12,_mm256_sub_pd(_mm256_sub_pd(rinv12,sh_ewald),velec));
1077             felec            = _mm256_mul_pd(_mm256_mul_pd(qq12,rinv12),_mm256_sub_pd(rinvsq12,felec));
1078
1079             cutoff_mask      = _mm256_cmp_pd(rsq12,rcutoff2,_CMP_LT_OQ);
1080
1081             /* Update potential sum for this i atom from the interaction with this j atom. */
1082             velec            = _mm256_and_pd(velec,cutoff_mask);
1083             velec            = _mm256_andnot_pd(dummy_mask,velec);
1084             velecsum         = _mm256_add_pd(velecsum,velec);
1085
1086             fscal            = felec;
1087
1088             fscal            = _mm256_and_pd(fscal,cutoff_mask);
1089
1090             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
1091
1092             /* Calculate temporary vectorial force */
1093             tx               = _mm256_mul_pd(fscal,dx12);
1094             ty               = _mm256_mul_pd(fscal,dy12);
1095             tz               = _mm256_mul_pd(fscal,dz12);
1096
1097             /* Update vectorial force */
1098             fix1             = _mm256_add_pd(fix1,tx);
1099             fiy1             = _mm256_add_pd(fiy1,ty);
1100             fiz1             = _mm256_add_pd(fiz1,tz);
1101
1102             fjx2             = _mm256_add_pd(fjx2,tx);
1103             fjy2             = _mm256_add_pd(fjy2,ty);
1104             fjz2             = _mm256_add_pd(fjz2,tz);
1105
1106             }
1107
1108             /**************************
1109              * CALCULATE INTERACTIONS *
1110              **************************/
1111
1112             if (gmx_mm256_any_lt(rsq13,rcutoff2))
1113             {
1114
1115             r13              = _mm256_mul_pd(rsq13,rinv13);
1116             r13              = _mm256_andnot_pd(dummy_mask,r13);
1117
1118             /* EWALD ELECTROSTATICS */
1119
1120             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1121             ewrt             = _mm256_mul_pd(r13,ewtabscale);
1122             ewitab           = _mm256_cvttpd_epi32(ewrt);
1123             eweps            = _mm256_sub_pd(ewrt,_mm256_round_pd(ewrt, _MM_FROUND_FLOOR));
1124             ewitab           = _mm_slli_epi32(ewitab,2);
1125             ewtabF           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
1126             ewtabD           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,1) );
1127             ewtabV           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,2) );
1128             ewtabFn          = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,3) );
1129             GMX_MM256_FULLTRANSPOSE4_PD(ewtabF,ewtabD,ewtabV,ewtabFn);
1130             felec            = _mm256_add_pd(ewtabF,_mm256_mul_pd(eweps,ewtabD));
1131             velec            = _mm256_sub_pd(ewtabV,_mm256_mul_pd(_mm256_mul_pd(ewtabhalfspace,eweps),_mm256_add_pd(ewtabF,felec)));
1132             velec            = _mm256_mul_pd(qq13,_mm256_sub_pd(_mm256_sub_pd(rinv13,sh_ewald),velec));
1133             felec            = _mm256_mul_pd(_mm256_mul_pd(qq13,rinv13),_mm256_sub_pd(rinvsq13,felec));
1134
1135             cutoff_mask      = _mm256_cmp_pd(rsq13,rcutoff2,_CMP_LT_OQ);
1136
1137             /* Update potential sum for this i atom from the interaction with this j atom. */
1138             velec            = _mm256_and_pd(velec,cutoff_mask);
1139             velec            = _mm256_andnot_pd(dummy_mask,velec);
1140             velecsum         = _mm256_add_pd(velecsum,velec);
1141
1142             fscal            = felec;
1143
1144             fscal            = _mm256_and_pd(fscal,cutoff_mask);
1145
1146             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
1147
1148             /* Calculate temporary vectorial force */
1149             tx               = _mm256_mul_pd(fscal,dx13);
1150             ty               = _mm256_mul_pd(fscal,dy13);
1151             tz               = _mm256_mul_pd(fscal,dz13);
1152
1153             /* Update vectorial force */
1154             fix1             = _mm256_add_pd(fix1,tx);
1155             fiy1             = _mm256_add_pd(fiy1,ty);
1156             fiz1             = _mm256_add_pd(fiz1,tz);
1157
1158             fjx3             = _mm256_add_pd(fjx3,tx);
1159             fjy3             = _mm256_add_pd(fjy3,ty);
1160             fjz3             = _mm256_add_pd(fjz3,tz);
1161
1162             }
1163
1164             /**************************
1165              * CALCULATE INTERACTIONS *
1166              **************************/
1167
1168             if (gmx_mm256_any_lt(rsq21,rcutoff2))
1169             {
1170
1171             r21              = _mm256_mul_pd(rsq21,rinv21);
1172             r21              = _mm256_andnot_pd(dummy_mask,r21);
1173
1174             /* EWALD ELECTROSTATICS */
1175
1176             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1177             ewrt             = _mm256_mul_pd(r21,ewtabscale);
1178             ewitab           = _mm256_cvttpd_epi32(ewrt);
1179             eweps            = _mm256_sub_pd(ewrt,_mm256_round_pd(ewrt, _MM_FROUND_FLOOR));
1180             ewitab           = _mm_slli_epi32(ewitab,2);
1181             ewtabF           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
1182             ewtabD           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,1) );
1183             ewtabV           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,2) );
1184             ewtabFn          = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,3) );
1185             GMX_MM256_FULLTRANSPOSE4_PD(ewtabF,ewtabD,ewtabV,ewtabFn);
1186             felec            = _mm256_add_pd(ewtabF,_mm256_mul_pd(eweps,ewtabD));
1187             velec            = _mm256_sub_pd(ewtabV,_mm256_mul_pd(_mm256_mul_pd(ewtabhalfspace,eweps),_mm256_add_pd(ewtabF,felec)));
1188             velec            = _mm256_mul_pd(qq21,_mm256_sub_pd(_mm256_sub_pd(rinv21,sh_ewald),velec));
1189             felec            = _mm256_mul_pd(_mm256_mul_pd(qq21,rinv21),_mm256_sub_pd(rinvsq21,felec));
1190
1191             cutoff_mask      = _mm256_cmp_pd(rsq21,rcutoff2,_CMP_LT_OQ);
1192
1193             /* Update potential sum for this i atom from the interaction with this j atom. */
1194             velec            = _mm256_and_pd(velec,cutoff_mask);
1195             velec            = _mm256_andnot_pd(dummy_mask,velec);
1196             velecsum         = _mm256_add_pd(velecsum,velec);
1197
1198             fscal            = felec;
1199
1200             fscal            = _mm256_and_pd(fscal,cutoff_mask);
1201
1202             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
1203
1204             /* Calculate temporary vectorial force */
1205             tx               = _mm256_mul_pd(fscal,dx21);
1206             ty               = _mm256_mul_pd(fscal,dy21);
1207             tz               = _mm256_mul_pd(fscal,dz21);
1208
1209             /* Update vectorial force */
1210             fix2             = _mm256_add_pd(fix2,tx);
1211             fiy2             = _mm256_add_pd(fiy2,ty);
1212             fiz2             = _mm256_add_pd(fiz2,tz);
1213
1214             fjx1             = _mm256_add_pd(fjx1,tx);
1215             fjy1             = _mm256_add_pd(fjy1,ty);
1216             fjz1             = _mm256_add_pd(fjz1,tz);
1217
1218             }
1219
1220             /**************************
1221              * CALCULATE INTERACTIONS *
1222              **************************/
1223
1224             if (gmx_mm256_any_lt(rsq22,rcutoff2))
1225             {
1226
1227             r22              = _mm256_mul_pd(rsq22,rinv22);
1228             r22              = _mm256_andnot_pd(dummy_mask,r22);
1229
1230             /* EWALD ELECTROSTATICS */
1231
1232             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1233             ewrt             = _mm256_mul_pd(r22,ewtabscale);
1234             ewitab           = _mm256_cvttpd_epi32(ewrt);
1235             eweps            = _mm256_sub_pd(ewrt,_mm256_round_pd(ewrt, _MM_FROUND_FLOOR));
1236             ewitab           = _mm_slli_epi32(ewitab,2);
1237             ewtabF           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
1238             ewtabD           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,1) );
1239             ewtabV           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,2) );
1240             ewtabFn          = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,3) );
1241             GMX_MM256_FULLTRANSPOSE4_PD(ewtabF,ewtabD,ewtabV,ewtabFn);
1242             felec            = _mm256_add_pd(ewtabF,_mm256_mul_pd(eweps,ewtabD));
1243             velec            = _mm256_sub_pd(ewtabV,_mm256_mul_pd(_mm256_mul_pd(ewtabhalfspace,eweps),_mm256_add_pd(ewtabF,felec)));
1244             velec            = _mm256_mul_pd(qq22,_mm256_sub_pd(_mm256_sub_pd(rinv22,sh_ewald),velec));
1245             felec            = _mm256_mul_pd(_mm256_mul_pd(qq22,rinv22),_mm256_sub_pd(rinvsq22,felec));
1246
1247             cutoff_mask      = _mm256_cmp_pd(rsq22,rcutoff2,_CMP_LT_OQ);
1248
1249             /* Update potential sum for this i atom from the interaction with this j atom. */
1250             velec            = _mm256_and_pd(velec,cutoff_mask);
1251             velec            = _mm256_andnot_pd(dummy_mask,velec);
1252             velecsum         = _mm256_add_pd(velecsum,velec);
1253
1254             fscal            = felec;
1255
1256             fscal            = _mm256_and_pd(fscal,cutoff_mask);
1257
1258             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
1259
1260             /* Calculate temporary vectorial force */
1261             tx               = _mm256_mul_pd(fscal,dx22);
1262             ty               = _mm256_mul_pd(fscal,dy22);
1263             tz               = _mm256_mul_pd(fscal,dz22);
1264
1265             /* Update vectorial force */
1266             fix2             = _mm256_add_pd(fix2,tx);
1267             fiy2             = _mm256_add_pd(fiy2,ty);
1268             fiz2             = _mm256_add_pd(fiz2,tz);
1269
1270             fjx2             = _mm256_add_pd(fjx2,tx);
1271             fjy2             = _mm256_add_pd(fjy2,ty);
1272             fjz2             = _mm256_add_pd(fjz2,tz);
1273
1274             }
1275
1276             /**************************
1277              * CALCULATE INTERACTIONS *
1278              **************************/
1279
1280             if (gmx_mm256_any_lt(rsq23,rcutoff2))
1281             {
1282
1283             r23              = _mm256_mul_pd(rsq23,rinv23);
1284             r23              = _mm256_andnot_pd(dummy_mask,r23);
1285
1286             /* EWALD ELECTROSTATICS */
1287
1288             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1289             ewrt             = _mm256_mul_pd(r23,ewtabscale);
1290             ewitab           = _mm256_cvttpd_epi32(ewrt);
1291             eweps            = _mm256_sub_pd(ewrt,_mm256_round_pd(ewrt, _MM_FROUND_FLOOR));
1292             ewitab           = _mm_slli_epi32(ewitab,2);
1293             ewtabF           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
1294             ewtabD           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,1) );
1295             ewtabV           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,2) );
1296             ewtabFn          = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,3) );
1297             GMX_MM256_FULLTRANSPOSE4_PD(ewtabF,ewtabD,ewtabV,ewtabFn);
1298             felec            = _mm256_add_pd(ewtabF,_mm256_mul_pd(eweps,ewtabD));
1299             velec            = _mm256_sub_pd(ewtabV,_mm256_mul_pd(_mm256_mul_pd(ewtabhalfspace,eweps),_mm256_add_pd(ewtabF,felec)));
1300             velec            = _mm256_mul_pd(qq23,_mm256_sub_pd(_mm256_sub_pd(rinv23,sh_ewald),velec));
1301             felec            = _mm256_mul_pd(_mm256_mul_pd(qq23,rinv23),_mm256_sub_pd(rinvsq23,felec));
1302
1303             cutoff_mask      = _mm256_cmp_pd(rsq23,rcutoff2,_CMP_LT_OQ);
1304
1305             /* Update potential sum for this i atom from the interaction with this j atom. */
1306             velec            = _mm256_and_pd(velec,cutoff_mask);
1307             velec            = _mm256_andnot_pd(dummy_mask,velec);
1308             velecsum         = _mm256_add_pd(velecsum,velec);
1309
1310             fscal            = felec;
1311
1312             fscal            = _mm256_and_pd(fscal,cutoff_mask);
1313
1314             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
1315
1316             /* Calculate temporary vectorial force */
1317             tx               = _mm256_mul_pd(fscal,dx23);
1318             ty               = _mm256_mul_pd(fscal,dy23);
1319             tz               = _mm256_mul_pd(fscal,dz23);
1320
1321             /* Update vectorial force */
1322             fix2             = _mm256_add_pd(fix2,tx);
1323             fiy2             = _mm256_add_pd(fiy2,ty);
1324             fiz2             = _mm256_add_pd(fiz2,tz);
1325
1326             fjx3             = _mm256_add_pd(fjx3,tx);
1327             fjy3             = _mm256_add_pd(fjy3,ty);
1328             fjz3             = _mm256_add_pd(fjz3,tz);
1329
1330             }
1331
1332             /**************************
1333              * CALCULATE INTERACTIONS *
1334              **************************/
1335
1336             if (gmx_mm256_any_lt(rsq31,rcutoff2))
1337             {
1338
1339             r31              = _mm256_mul_pd(rsq31,rinv31);
1340             r31              = _mm256_andnot_pd(dummy_mask,r31);
1341
1342             /* EWALD ELECTROSTATICS */
1343
1344             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1345             ewrt             = _mm256_mul_pd(r31,ewtabscale);
1346             ewitab           = _mm256_cvttpd_epi32(ewrt);
1347             eweps            = _mm256_sub_pd(ewrt,_mm256_round_pd(ewrt, _MM_FROUND_FLOOR));
1348             ewitab           = _mm_slli_epi32(ewitab,2);
1349             ewtabF           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
1350             ewtabD           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,1) );
1351             ewtabV           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,2) );
1352             ewtabFn          = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,3) );
1353             GMX_MM256_FULLTRANSPOSE4_PD(ewtabF,ewtabD,ewtabV,ewtabFn);
1354             felec            = _mm256_add_pd(ewtabF,_mm256_mul_pd(eweps,ewtabD));
1355             velec            = _mm256_sub_pd(ewtabV,_mm256_mul_pd(_mm256_mul_pd(ewtabhalfspace,eweps),_mm256_add_pd(ewtabF,felec)));
1356             velec            = _mm256_mul_pd(qq31,_mm256_sub_pd(_mm256_sub_pd(rinv31,sh_ewald),velec));
1357             felec            = _mm256_mul_pd(_mm256_mul_pd(qq31,rinv31),_mm256_sub_pd(rinvsq31,felec));
1358
1359             cutoff_mask      = _mm256_cmp_pd(rsq31,rcutoff2,_CMP_LT_OQ);
1360
1361             /* Update potential sum for this i atom from the interaction with this j atom. */
1362             velec            = _mm256_and_pd(velec,cutoff_mask);
1363             velec            = _mm256_andnot_pd(dummy_mask,velec);
1364             velecsum         = _mm256_add_pd(velecsum,velec);
1365
1366             fscal            = felec;
1367
1368             fscal            = _mm256_and_pd(fscal,cutoff_mask);
1369
1370             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
1371
1372             /* Calculate temporary vectorial force */
1373             tx               = _mm256_mul_pd(fscal,dx31);
1374             ty               = _mm256_mul_pd(fscal,dy31);
1375             tz               = _mm256_mul_pd(fscal,dz31);
1376
1377             /* Update vectorial force */
1378             fix3             = _mm256_add_pd(fix3,tx);
1379             fiy3             = _mm256_add_pd(fiy3,ty);
1380             fiz3             = _mm256_add_pd(fiz3,tz);
1381
1382             fjx1             = _mm256_add_pd(fjx1,tx);
1383             fjy1             = _mm256_add_pd(fjy1,ty);
1384             fjz1             = _mm256_add_pd(fjz1,tz);
1385
1386             }
1387
1388             /**************************
1389              * CALCULATE INTERACTIONS *
1390              **************************/
1391
1392             if (gmx_mm256_any_lt(rsq32,rcutoff2))
1393             {
1394
1395             r32              = _mm256_mul_pd(rsq32,rinv32);
1396             r32              = _mm256_andnot_pd(dummy_mask,r32);
1397
1398             /* EWALD ELECTROSTATICS */
1399
1400             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1401             ewrt             = _mm256_mul_pd(r32,ewtabscale);
1402             ewitab           = _mm256_cvttpd_epi32(ewrt);
1403             eweps            = _mm256_sub_pd(ewrt,_mm256_round_pd(ewrt, _MM_FROUND_FLOOR));
1404             ewitab           = _mm_slli_epi32(ewitab,2);
1405             ewtabF           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
1406             ewtabD           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,1) );
1407             ewtabV           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,2) );
1408             ewtabFn          = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,3) );
1409             GMX_MM256_FULLTRANSPOSE4_PD(ewtabF,ewtabD,ewtabV,ewtabFn);
1410             felec            = _mm256_add_pd(ewtabF,_mm256_mul_pd(eweps,ewtabD));
1411             velec            = _mm256_sub_pd(ewtabV,_mm256_mul_pd(_mm256_mul_pd(ewtabhalfspace,eweps),_mm256_add_pd(ewtabF,felec)));
1412             velec            = _mm256_mul_pd(qq32,_mm256_sub_pd(_mm256_sub_pd(rinv32,sh_ewald),velec));
1413             felec            = _mm256_mul_pd(_mm256_mul_pd(qq32,rinv32),_mm256_sub_pd(rinvsq32,felec));
1414
1415             cutoff_mask      = _mm256_cmp_pd(rsq32,rcutoff2,_CMP_LT_OQ);
1416
1417             /* Update potential sum for this i atom from the interaction with this j atom. */
1418             velec            = _mm256_and_pd(velec,cutoff_mask);
1419             velec            = _mm256_andnot_pd(dummy_mask,velec);
1420             velecsum         = _mm256_add_pd(velecsum,velec);
1421
1422             fscal            = felec;
1423
1424             fscal            = _mm256_and_pd(fscal,cutoff_mask);
1425
1426             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
1427
1428             /* Calculate temporary vectorial force */
1429             tx               = _mm256_mul_pd(fscal,dx32);
1430             ty               = _mm256_mul_pd(fscal,dy32);
1431             tz               = _mm256_mul_pd(fscal,dz32);
1432
1433             /* Update vectorial force */
1434             fix3             = _mm256_add_pd(fix3,tx);
1435             fiy3             = _mm256_add_pd(fiy3,ty);
1436             fiz3             = _mm256_add_pd(fiz3,tz);
1437
1438             fjx2             = _mm256_add_pd(fjx2,tx);
1439             fjy2             = _mm256_add_pd(fjy2,ty);
1440             fjz2             = _mm256_add_pd(fjz2,tz);
1441
1442             }
1443
1444             /**************************
1445              * CALCULATE INTERACTIONS *
1446              **************************/
1447
1448             if (gmx_mm256_any_lt(rsq33,rcutoff2))
1449             {
1450
1451             r33              = _mm256_mul_pd(rsq33,rinv33);
1452             r33              = _mm256_andnot_pd(dummy_mask,r33);
1453
1454             /* EWALD ELECTROSTATICS */
1455
1456             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1457             ewrt             = _mm256_mul_pd(r33,ewtabscale);
1458             ewitab           = _mm256_cvttpd_epi32(ewrt);
1459             eweps            = _mm256_sub_pd(ewrt,_mm256_round_pd(ewrt, _MM_FROUND_FLOOR));
1460             ewitab           = _mm_slli_epi32(ewitab,2);
1461             ewtabF           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
1462             ewtabD           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,1) );
1463             ewtabV           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,2) );
1464             ewtabFn          = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,3) );
1465             GMX_MM256_FULLTRANSPOSE4_PD(ewtabF,ewtabD,ewtabV,ewtabFn);
1466             felec            = _mm256_add_pd(ewtabF,_mm256_mul_pd(eweps,ewtabD));
1467             velec            = _mm256_sub_pd(ewtabV,_mm256_mul_pd(_mm256_mul_pd(ewtabhalfspace,eweps),_mm256_add_pd(ewtabF,felec)));
1468             velec            = _mm256_mul_pd(qq33,_mm256_sub_pd(_mm256_sub_pd(rinv33,sh_ewald),velec));
1469             felec            = _mm256_mul_pd(_mm256_mul_pd(qq33,rinv33),_mm256_sub_pd(rinvsq33,felec));
1470
1471             cutoff_mask      = _mm256_cmp_pd(rsq33,rcutoff2,_CMP_LT_OQ);
1472
1473             /* Update potential sum for this i atom from the interaction with this j atom. */
1474             velec            = _mm256_and_pd(velec,cutoff_mask);
1475             velec            = _mm256_andnot_pd(dummy_mask,velec);
1476             velecsum         = _mm256_add_pd(velecsum,velec);
1477
1478             fscal            = felec;
1479
1480             fscal            = _mm256_and_pd(fscal,cutoff_mask);
1481
1482             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
1483
1484             /* Calculate temporary vectorial force */
1485             tx               = _mm256_mul_pd(fscal,dx33);
1486             ty               = _mm256_mul_pd(fscal,dy33);
1487             tz               = _mm256_mul_pd(fscal,dz33);
1488
1489             /* Update vectorial force */
1490             fix3             = _mm256_add_pd(fix3,tx);
1491             fiy3             = _mm256_add_pd(fiy3,ty);
1492             fiz3             = _mm256_add_pd(fiz3,tz);
1493
1494             fjx3             = _mm256_add_pd(fjx3,tx);
1495             fjy3             = _mm256_add_pd(fjy3,ty);
1496             fjz3             = _mm256_add_pd(fjz3,tz);
1497
1498             }
1499
1500             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
1501             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
1502             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
1503             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
1504
1505             gmx_mm256_decrement_4rvec_4ptr_swizzle_pd(fjptrA,fjptrB,fjptrC,fjptrD,
1506                                                       fjx0,fjy0,fjz0,fjx1,fjy1,fjz1,
1507                                                       fjx2,fjy2,fjz2,fjx3,fjy3,fjz3);
1508
1509             /* Inner loop uses 467 flops */
1510         }
1511
1512         /* End of innermost loop */
1513
1514         gmx_mm256_update_iforce_4atom_swizzle_pd(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,fix3,fiy3,fiz3,
1515                                                  f+i_coord_offset,fshift+i_shift_offset);
1516
1517         ggid                        = gid[iidx];
1518         /* Update potential energies */
1519         gmx_mm256_update_1pot_pd(velecsum,kernel_data->energygrp_elec+ggid);
1520         gmx_mm256_update_1pot_pd(vvdwsum,kernel_data->energygrp_vdw+ggid);
1521
1522         /* Increment number of inner iterations */
1523         inneriter                  += j_index_end - j_index_start;
1524
1525         /* Outer loop uses 26 flops */
1526     }
1527
1528     /* Increment number of outer iterations */
1529     outeriter        += nri;
1530
1531     /* Update outer/inner flops */
1532
1533     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W4W4_VF,outeriter*26 + inneriter*467);
1534 }
1535 /*
1536  * Gromacs nonbonded kernel:   nb_kernel_ElecEwSh_VdwLJSh_GeomW4W4_F_avx_256_double
1537  * Electrostatics interaction: Ewald
1538  * VdW interaction:            LennardJones
1539  * Geometry:                   Water4-Water4
1540  * Calculate force/pot:        Force
1541  */
1542 void
1543 nb_kernel_ElecEwSh_VdwLJSh_GeomW4W4_F_avx_256_double
1544                     (t_nblist * gmx_restrict                nlist,
1545                      rvec * gmx_restrict                    xx,
1546                      rvec * gmx_restrict                    ff,
1547                      t_forcerec * gmx_restrict              fr,
1548                      t_mdatoms * gmx_restrict               mdatoms,
1549                      nb_kernel_data_t * gmx_restrict        kernel_data,
1550                      t_nrnb * gmx_restrict                  nrnb)
1551 {
1552     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
1553      * just 0 for non-waters.
1554      * Suffixes A,B,C,D refer to j loop unrolling done with AVX, e.g. for the four different
1555      * jnr indices corresponding to data put in the four positions in the SIMD register.
1556      */
1557     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
1558     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
1559     int              jnrA,jnrB,jnrC,jnrD;
1560     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
1561     int              jnrlistE,jnrlistF,jnrlistG,jnrlistH;
1562     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
1563     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
1564     real             rcutoff_scalar;
1565     real             *shiftvec,*fshift,*x,*f;
1566     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD;
1567     real             scratch[4*DIM];
1568     __m256d          tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
1569     real *           vdwioffsetptr0;
1570     __m256d          ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
1571     real *           vdwioffsetptr1;
1572     __m256d          ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
1573     real *           vdwioffsetptr2;
1574     __m256d          ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
1575     real *           vdwioffsetptr3;
1576     __m256d          ix3,iy3,iz3,fix3,fiy3,fiz3,iq3,isai3;
1577     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
1578     __m256d          jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
1579     int              vdwjidx1A,vdwjidx1B,vdwjidx1C,vdwjidx1D;
1580     __m256d          jx1,jy1,jz1,fjx1,fjy1,fjz1,jq1,isaj1;
1581     int              vdwjidx2A,vdwjidx2B,vdwjidx2C,vdwjidx2D;
1582     __m256d          jx2,jy2,jz2,fjx2,fjy2,fjz2,jq2,isaj2;
1583     int              vdwjidx3A,vdwjidx3B,vdwjidx3C,vdwjidx3D;
1584     __m256d          jx3,jy3,jz3,fjx3,fjy3,fjz3,jq3,isaj3;
1585     __m256d          dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
1586     __m256d          dx11,dy11,dz11,rsq11,rinv11,rinvsq11,r11,qq11,c6_11,c12_11;
1587     __m256d          dx12,dy12,dz12,rsq12,rinv12,rinvsq12,r12,qq12,c6_12,c12_12;
1588     __m256d          dx13,dy13,dz13,rsq13,rinv13,rinvsq13,r13,qq13,c6_13,c12_13;
1589     __m256d          dx21,dy21,dz21,rsq21,rinv21,rinvsq21,r21,qq21,c6_21,c12_21;
1590     __m256d          dx22,dy22,dz22,rsq22,rinv22,rinvsq22,r22,qq22,c6_22,c12_22;
1591     __m256d          dx23,dy23,dz23,rsq23,rinv23,rinvsq23,r23,qq23,c6_23,c12_23;
1592     __m256d          dx31,dy31,dz31,rsq31,rinv31,rinvsq31,r31,qq31,c6_31,c12_31;
1593     __m256d          dx32,dy32,dz32,rsq32,rinv32,rinvsq32,r32,qq32,c6_32,c12_32;
1594     __m256d          dx33,dy33,dz33,rsq33,rinv33,rinvsq33,r33,qq33,c6_33,c12_33;
1595     __m256d          velec,felec,velecsum,facel,crf,krf,krf2;
1596     real             *charge;
1597     int              nvdwtype;
1598     __m256d          rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
1599     int              *vdwtype;
1600     real             *vdwparam;
1601     __m256d          one_sixth   = _mm256_set1_pd(1.0/6.0);
1602     __m256d          one_twelfth = _mm256_set1_pd(1.0/12.0);
1603     __m128i          ewitab;
1604     __m256d          ewtabscale,eweps,sh_ewald,ewrt,ewtabhalfspace,ewtabF,ewtabFn,ewtabD,ewtabV;
1605     __m256d          beta,beta2,beta3,zeta2,pmecorrF,pmecorrV,rinv3;
1606     real             *ewtab;
1607     __m256d          dummy_mask,cutoff_mask;
1608     __m128           tmpmask0,tmpmask1;
1609     __m256d          signbit = _mm256_castsi256_pd( _mm256_set1_epi32(0x80000000) );
1610     __m256d          one     = _mm256_set1_pd(1.0);
1611     __m256d          two     = _mm256_set1_pd(2.0);
1612     x                = xx[0];
1613     f                = ff[0];
1614
1615     nri              = nlist->nri;
1616     iinr             = nlist->iinr;
1617     jindex           = nlist->jindex;
1618     jjnr             = nlist->jjnr;
1619     shiftidx         = nlist->shift;
1620     gid              = nlist->gid;
1621     shiftvec         = fr->shift_vec[0];
1622     fshift           = fr->fshift[0];
1623     facel            = _mm256_set1_pd(fr->epsfac);
1624     charge           = mdatoms->chargeA;
1625     nvdwtype         = fr->ntype;
1626     vdwparam         = fr->nbfp;
1627     vdwtype          = mdatoms->typeA;
1628
1629     sh_ewald         = _mm256_set1_pd(fr->ic->sh_ewald);
1630     beta             = _mm256_set1_pd(fr->ic->ewaldcoeff);
1631     beta2            = _mm256_mul_pd(beta,beta);
1632     beta3            = _mm256_mul_pd(beta,beta2);
1633
1634     ewtab            = fr->ic->tabq_coul_F;
1635     ewtabscale       = _mm256_set1_pd(fr->ic->tabq_scale);
1636     ewtabhalfspace   = _mm256_set1_pd(0.5/fr->ic->tabq_scale);
1637
1638     /* Setup water-specific parameters */
1639     inr              = nlist->iinr[0];
1640     iq1              = _mm256_mul_pd(facel,_mm256_set1_pd(charge[inr+1]));
1641     iq2              = _mm256_mul_pd(facel,_mm256_set1_pd(charge[inr+2]));
1642     iq3              = _mm256_mul_pd(facel,_mm256_set1_pd(charge[inr+3]));
1643     vdwioffsetptr0   = vdwparam+2*nvdwtype*vdwtype[inr+0];
1644
1645     jq1              = _mm256_set1_pd(charge[inr+1]);
1646     jq2              = _mm256_set1_pd(charge[inr+2]);
1647     jq3              = _mm256_set1_pd(charge[inr+3]);
1648     vdwjidx0A        = 2*vdwtype[inr+0];
1649     c6_00            = _mm256_set1_pd(vdwioffsetptr0[vdwjidx0A]);
1650     c12_00           = _mm256_set1_pd(vdwioffsetptr0[vdwjidx0A+1]);
1651     qq11             = _mm256_mul_pd(iq1,jq1);
1652     qq12             = _mm256_mul_pd(iq1,jq2);
1653     qq13             = _mm256_mul_pd(iq1,jq3);
1654     qq21             = _mm256_mul_pd(iq2,jq1);
1655     qq22             = _mm256_mul_pd(iq2,jq2);
1656     qq23             = _mm256_mul_pd(iq2,jq3);
1657     qq31             = _mm256_mul_pd(iq3,jq1);
1658     qq32             = _mm256_mul_pd(iq3,jq2);
1659     qq33             = _mm256_mul_pd(iq3,jq3);
1660
1661     /* When we use explicit cutoffs the value must be identical for elec and VdW, so use elec as an arbitrary choice */
1662     rcutoff_scalar   = fr->rcoulomb;
1663     rcutoff          = _mm256_set1_pd(rcutoff_scalar);
1664     rcutoff2         = _mm256_mul_pd(rcutoff,rcutoff);
1665
1666     sh_vdw_invrcut6  = _mm256_set1_pd(fr->ic->sh_invrc6);
1667     rvdw             = _mm256_set1_pd(fr->rvdw);
1668
1669     /* Avoid stupid compiler warnings */
1670     jnrA = jnrB = jnrC = jnrD = 0;
1671     j_coord_offsetA = 0;
1672     j_coord_offsetB = 0;
1673     j_coord_offsetC = 0;
1674     j_coord_offsetD = 0;
1675
1676     outeriter        = 0;
1677     inneriter        = 0;
1678
1679     for(iidx=0;iidx<4*DIM;iidx++)
1680     {
1681         scratch[iidx] = 0.0;
1682     }
1683
1684     /* Start outer loop over neighborlists */
1685     for(iidx=0; iidx<nri; iidx++)
1686     {
1687         /* Load shift vector for this list */
1688         i_shift_offset   = DIM*shiftidx[iidx];
1689
1690         /* Load limits for loop over neighbors */
1691         j_index_start    = jindex[iidx];
1692         j_index_end      = jindex[iidx+1];
1693
1694         /* Get outer coordinate index */
1695         inr              = iinr[iidx];
1696         i_coord_offset   = DIM*inr;
1697
1698         /* Load i particle coords and add shift vector */
1699         gmx_mm256_load_shift_and_4rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset,
1700                                                     &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2,&ix3,&iy3,&iz3);
1701
1702         fix0             = _mm256_setzero_pd();
1703         fiy0             = _mm256_setzero_pd();
1704         fiz0             = _mm256_setzero_pd();
1705         fix1             = _mm256_setzero_pd();
1706         fiy1             = _mm256_setzero_pd();
1707         fiz1             = _mm256_setzero_pd();
1708         fix2             = _mm256_setzero_pd();
1709         fiy2             = _mm256_setzero_pd();
1710         fiz2             = _mm256_setzero_pd();
1711         fix3             = _mm256_setzero_pd();
1712         fiy3             = _mm256_setzero_pd();
1713         fiz3             = _mm256_setzero_pd();
1714
1715         /* Start inner kernel loop */
1716         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
1717         {
1718
1719             /* Get j neighbor index, and coordinate index */
1720             jnrA             = jjnr[jidx];
1721             jnrB             = jjnr[jidx+1];
1722             jnrC             = jjnr[jidx+2];
1723             jnrD             = jjnr[jidx+3];
1724             j_coord_offsetA  = DIM*jnrA;
1725             j_coord_offsetB  = DIM*jnrB;
1726             j_coord_offsetC  = DIM*jnrC;
1727             j_coord_offsetD  = DIM*jnrD;
1728
1729             /* load j atom coordinates */
1730             gmx_mm256_load_4rvec_4ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
1731                                                  x+j_coord_offsetC,x+j_coord_offsetD,
1732                                                  &jx0,&jy0,&jz0,&jx1,&jy1,&jz1,&jx2,
1733                                                  &jy2,&jz2,&jx3,&jy3,&jz3);
1734
1735             /* Calculate displacement vector */
1736             dx00             = _mm256_sub_pd(ix0,jx0);
1737             dy00             = _mm256_sub_pd(iy0,jy0);
1738             dz00             = _mm256_sub_pd(iz0,jz0);
1739             dx11             = _mm256_sub_pd(ix1,jx1);
1740             dy11             = _mm256_sub_pd(iy1,jy1);
1741             dz11             = _mm256_sub_pd(iz1,jz1);
1742             dx12             = _mm256_sub_pd(ix1,jx2);
1743             dy12             = _mm256_sub_pd(iy1,jy2);
1744             dz12             = _mm256_sub_pd(iz1,jz2);
1745             dx13             = _mm256_sub_pd(ix1,jx3);
1746             dy13             = _mm256_sub_pd(iy1,jy3);
1747             dz13             = _mm256_sub_pd(iz1,jz3);
1748             dx21             = _mm256_sub_pd(ix2,jx1);
1749             dy21             = _mm256_sub_pd(iy2,jy1);
1750             dz21             = _mm256_sub_pd(iz2,jz1);
1751             dx22             = _mm256_sub_pd(ix2,jx2);
1752             dy22             = _mm256_sub_pd(iy2,jy2);
1753             dz22             = _mm256_sub_pd(iz2,jz2);
1754             dx23             = _mm256_sub_pd(ix2,jx3);
1755             dy23             = _mm256_sub_pd(iy2,jy3);
1756             dz23             = _mm256_sub_pd(iz2,jz3);
1757             dx31             = _mm256_sub_pd(ix3,jx1);
1758             dy31             = _mm256_sub_pd(iy3,jy1);
1759             dz31             = _mm256_sub_pd(iz3,jz1);
1760             dx32             = _mm256_sub_pd(ix3,jx2);
1761             dy32             = _mm256_sub_pd(iy3,jy2);
1762             dz32             = _mm256_sub_pd(iz3,jz2);
1763             dx33             = _mm256_sub_pd(ix3,jx3);
1764             dy33             = _mm256_sub_pd(iy3,jy3);
1765             dz33             = _mm256_sub_pd(iz3,jz3);
1766
1767             /* Calculate squared distance and things based on it */
1768             rsq00            = gmx_mm256_calc_rsq_pd(dx00,dy00,dz00);
1769             rsq11            = gmx_mm256_calc_rsq_pd(dx11,dy11,dz11);
1770             rsq12            = gmx_mm256_calc_rsq_pd(dx12,dy12,dz12);
1771             rsq13            = gmx_mm256_calc_rsq_pd(dx13,dy13,dz13);
1772             rsq21            = gmx_mm256_calc_rsq_pd(dx21,dy21,dz21);
1773             rsq22            = gmx_mm256_calc_rsq_pd(dx22,dy22,dz22);
1774             rsq23            = gmx_mm256_calc_rsq_pd(dx23,dy23,dz23);
1775             rsq31            = gmx_mm256_calc_rsq_pd(dx31,dy31,dz31);
1776             rsq32            = gmx_mm256_calc_rsq_pd(dx32,dy32,dz32);
1777             rsq33            = gmx_mm256_calc_rsq_pd(dx33,dy33,dz33);
1778
1779             rinv11           = gmx_mm256_invsqrt_pd(rsq11);
1780             rinv12           = gmx_mm256_invsqrt_pd(rsq12);
1781             rinv13           = gmx_mm256_invsqrt_pd(rsq13);
1782             rinv21           = gmx_mm256_invsqrt_pd(rsq21);
1783             rinv22           = gmx_mm256_invsqrt_pd(rsq22);
1784             rinv23           = gmx_mm256_invsqrt_pd(rsq23);
1785             rinv31           = gmx_mm256_invsqrt_pd(rsq31);
1786             rinv32           = gmx_mm256_invsqrt_pd(rsq32);
1787             rinv33           = gmx_mm256_invsqrt_pd(rsq33);
1788
1789             rinvsq00         = gmx_mm256_inv_pd(rsq00);
1790             rinvsq11         = _mm256_mul_pd(rinv11,rinv11);
1791             rinvsq12         = _mm256_mul_pd(rinv12,rinv12);
1792             rinvsq13         = _mm256_mul_pd(rinv13,rinv13);
1793             rinvsq21         = _mm256_mul_pd(rinv21,rinv21);
1794             rinvsq22         = _mm256_mul_pd(rinv22,rinv22);
1795             rinvsq23         = _mm256_mul_pd(rinv23,rinv23);
1796             rinvsq31         = _mm256_mul_pd(rinv31,rinv31);
1797             rinvsq32         = _mm256_mul_pd(rinv32,rinv32);
1798             rinvsq33         = _mm256_mul_pd(rinv33,rinv33);
1799
1800             fjx0             = _mm256_setzero_pd();
1801             fjy0             = _mm256_setzero_pd();
1802             fjz0             = _mm256_setzero_pd();
1803             fjx1             = _mm256_setzero_pd();
1804             fjy1             = _mm256_setzero_pd();
1805             fjz1             = _mm256_setzero_pd();
1806             fjx2             = _mm256_setzero_pd();
1807             fjy2             = _mm256_setzero_pd();
1808             fjz2             = _mm256_setzero_pd();
1809             fjx3             = _mm256_setzero_pd();
1810             fjy3             = _mm256_setzero_pd();
1811             fjz3             = _mm256_setzero_pd();
1812
1813             /**************************
1814              * CALCULATE INTERACTIONS *
1815              **************************/
1816
1817             if (gmx_mm256_any_lt(rsq00,rcutoff2))
1818             {
1819
1820             /* LENNARD-JONES DISPERSION/REPULSION */
1821
1822             rinvsix          = _mm256_mul_pd(_mm256_mul_pd(rinvsq00,rinvsq00),rinvsq00);
1823             fvdw             = _mm256_mul_pd(_mm256_sub_pd(_mm256_mul_pd(c12_00,rinvsix),c6_00),_mm256_mul_pd(rinvsix,rinvsq00));
1824
1825             cutoff_mask      = _mm256_cmp_pd(rsq00,rcutoff2,_CMP_LT_OQ);
1826
1827             fscal            = fvdw;
1828
1829             fscal            = _mm256_and_pd(fscal,cutoff_mask);
1830
1831             /* Calculate temporary vectorial force */
1832             tx               = _mm256_mul_pd(fscal,dx00);
1833             ty               = _mm256_mul_pd(fscal,dy00);
1834             tz               = _mm256_mul_pd(fscal,dz00);
1835
1836             /* Update vectorial force */
1837             fix0             = _mm256_add_pd(fix0,tx);
1838             fiy0             = _mm256_add_pd(fiy0,ty);
1839             fiz0             = _mm256_add_pd(fiz0,tz);
1840
1841             fjx0             = _mm256_add_pd(fjx0,tx);
1842             fjy0             = _mm256_add_pd(fjy0,ty);
1843             fjz0             = _mm256_add_pd(fjz0,tz);
1844
1845             }
1846
1847             /**************************
1848              * CALCULATE INTERACTIONS *
1849              **************************/
1850
1851             if (gmx_mm256_any_lt(rsq11,rcutoff2))
1852             {
1853
1854             r11              = _mm256_mul_pd(rsq11,rinv11);
1855
1856             /* EWALD ELECTROSTATICS */
1857
1858             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1859             ewrt             = _mm256_mul_pd(r11,ewtabscale);
1860             ewitab           = _mm256_cvttpd_epi32(ewrt);
1861             eweps            = _mm256_sub_pd(ewrt,_mm256_round_pd(ewrt, _MM_FROUND_FLOOR));
1862             gmx_mm256_load_4pair_swizzle_pd(ewtab + _mm_extract_epi32(ewitab,0),ewtab + _mm_extract_epi32(ewitab,1),
1863                                             ewtab + _mm_extract_epi32(ewitab,2),ewtab + _mm_extract_epi32(ewitab,3),
1864                                             &ewtabF,&ewtabFn);
1865             felec            = _mm256_add_pd(_mm256_mul_pd( _mm256_sub_pd(one,eweps),ewtabF),_mm256_mul_pd(eweps,ewtabFn));
1866             felec            = _mm256_mul_pd(_mm256_mul_pd(qq11,rinv11),_mm256_sub_pd(rinvsq11,felec));
1867
1868             cutoff_mask      = _mm256_cmp_pd(rsq11,rcutoff2,_CMP_LT_OQ);
1869
1870             fscal            = felec;
1871
1872             fscal            = _mm256_and_pd(fscal,cutoff_mask);
1873
1874             /* Calculate temporary vectorial force */
1875             tx               = _mm256_mul_pd(fscal,dx11);
1876             ty               = _mm256_mul_pd(fscal,dy11);
1877             tz               = _mm256_mul_pd(fscal,dz11);
1878
1879             /* Update vectorial force */
1880             fix1             = _mm256_add_pd(fix1,tx);
1881             fiy1             = _mm256_add_pd(fiy1,ty);
1882             fiz1             = _mm256_add_pd(fiz1,tz);
1883
1884             fjx1             = _mm256_add_pd(fjx1,tx);
1885             fjy1             = _mm256_add_pd(fjy1,ty);
1886             fjz1             = _mm256_add_pd(fjz1,tz);
1887
1888             }
1889
1890             /**************************
1891              * CALCULATE INTERACTIONS *
1892              **************************/
1893
1894             if (gmx_mm256_any_lt(rsq12,rcutoff2))
1895             {
1896
1897             r12              = _mm256_mul_pd(rsq12,rinv12);
1898
1899             /* EWALD ELECTROSTATICS */
1900
1901             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1902             ewrt             = _mm256_mul_pd(r12,ewtabscale);
1903             ewitab           = _mm256_cvttpd_epi32(ewrt);
1904             eweps            = _mm256_sub_pd(ewrt,_mm256_round_pd(ewrt, _MM_FROUND_FLOOR));
1905             gmx_mm256_load_4pair_swizzle_pd(ewtab + _mm_extract_epi32(ewitab,0),ewtab + _mm_extract_epi32(ewitab,1),
1906                                             ewtab + _mm_extract_epi32(ewitab,2),ewtab + _mm_extract_epi32(ewitab,3),
1907                                             &ewtabF,&ewtabFn);
1908             felec            = _mm256_add_pd(_mm256_mul_pd( _mm256_sub_pd(one,eweps),ewtabF),_mm256_mul_pd(eweps,ewtabFn));
1909             felec            = _mm256_mul_pd(_mm256_mul_pd(qq12,rinv12),_mm256_sub_pd(rinvsq12,felec));
1910
1911             cutoff_mask      = _mm256_cmp_pd(rsq12,rcutoff2,_CMP_LT_OQ);
1912
1913             fscal            = felec;
1914
1915             fscal            = _mm256_and_pd(fscal,cutoff_mask);
1916
1917             /* Calculate temporary vectorial force */
1918             tx               = _mm256_mul_pd(fscal,dx12);
1919             ty               = _mm256_mul_pd(fscal,dy12);
1920             tz               = _mm256_mul_pd(fscal,dz12);
1921
1922             /* Update vectorial force */
1923             fix1             = _mm256_add_pd(fix1,tx);
1924             fiy1             = _mm256_add_pd(fiy1,ty);
1925             fiz1             = _mm256_add_pd(fiz1,tz);
1926
1927             fjx2             = _mm256_add_pd(fjx2,tx);
1928             fjy2             = _mm256_add_pd(fjy2,ty);
1929             fjz2             = _mm256_add_pd(fjz2,tz);
1930
1931             }
1932
1933             /**************************
1934              * CALCULATE INTERACTIONS *
1935              **************************/
1936
1937             if (gmx_mm256_any_lt(rsq13,rcutoff2))
1938             {
1939
1940             r13              = _mm256_mul_pd(rsq13,rinv13);
1941
1942             /* EWALD ELECTROSTATICS */
1943
1944             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1945             ewrt             = _mm256_mul_pd(r13,ewtabscale);
1946             ewitab           = _mm256_cvttpd_epi32(ewrt);
1947             eweps            = _mm256_sub_pd(ewrt,_mm256_round_pd(ewrt, _MM_FROUND_FLOOR));
1948             gmx_mm256_load_4pair_swizzle_pd(ewtab + _mm_extract_epi32(ewitab,0),ewtab + _mm_extract_epi32(ewitab,1),
1949                                             ewtab + _mm_extract_epi32(ewitab,2),ewtab + _mm_extract_epi32(ewitab,3),
1950                                             &ewtabF,&ewtabFn);
1951             felec            = _mm256_add_pd(_mm256_mul_pd( _mm256_sub_pd(one,eweps),ewtabF),_mm256_mul_pd(eweps,ewtabFn));
1952             felec            = _mm256_mul_pd(_mm256_mul_pd(qq13,rinv13),_mm256_sub_pd(rinvsq13,felec));
1953
1954             cutoff_mask      = _mm256_cmp_pd(rsq13,rcutoff2,_CMP_LT_OQ);
1955
1956             fscal            = felec;
1957
1958             fscal            = _mm256_and_pd(fscal,cutoff_mask);
1959
1960             /* Calculate temporary vectorial force */
1961             tx               = _mm256_mul_pd(fscal,dx13);
1962             ty               = _mm256_mul_pd(fscal,dy13);
1963             tz               = _mm256_mul_pd(fscal,dz13);
1964
1965             /* Update vectorial force */
1966             fix1             = _mm256_add_pd(fix1,tx);
1967             fiy1             = _mm256_add_pd(fiy1,ty);
1968             fiz1             = _mm256_add_pd(fiz1,tz);
1969
1970             fjx3             = _mm256_add_pd(fjx3,tx);
1971             fjy3             = _mm256_add_pd(fjy3,ty);
1972             fjz3             = _mm256_add_pd(fjz3,tz);
1973
1974             }
1975
1976             /**************************
1977              * CALCULATE INTERACTIONS *
1978              **************************/
1979
1980             if (gmx_mm256_any_lt(rsq21,rcutoff2))
1981             {
1982
1983             r21              = _mm256_mul_pd(rsq21,rinv21);
1984
1985             /* EWALD ELECTROSTATICS */
1986
1987             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1988             ewrt             = _mm256_mul_pd(r21,ewtabscale);
1989             ewitab           = _mm256_cvttpd_epi32(ewrt);
1990             eweps            = _mm256_sub_pd(ewrt,_mm256_round_pd(ewrt, _MM_FROUND_FLOOR));
1991             gmx_mm256_load_4pair_swizzle_pd(ewtab + _mm_extract_epi32(ewitab,0),ewtab + _mm_extract_epi32(ewitab,1),
1992                                             ewtab + _mm_extract_epi32(ewitab,2),ewtab + _mm_extract_epi32(ewitab,3),
1993                                             &ewtabF,&ewtabFn);
1994             felec            = _mm256_add_pd(_mm256_mul_pd( _mm256_sub_pd(one,eweps),ewtabF),_mm256_mul_pd(eweps,ewtabFn));
1995             felec            = _mm256_mul_pd(_mm256_mul_pd(qq21,rinv21),_mm256_sub_pd(rinvsq21,felec));
1996
1997             cutoff_mask      = _mm256_cmp_pd(rsq21,rcutoff2,_CMP_LT_OQ);
1998
1999             fscal            = felec;
2000
2001             fscal            = _mm256_and_pd(fscal,cutoff_mask);
2002
2003             /* Calculate temporary vectorial force */
2004             tx               = _mm256_mul_pd(fscal,dx21);
2005             ty               = _mm256_mul_pd(fscal,dy21);
2006             tz               = _mm256_mul_pd(fscal,dz21);
2007
2008             /* Update vectorial force */
2009             fix2             = _mm256_add_pd(fix2,tx);
2010             fiy2             = _mm256_add_pd(fiy2,ty);
2011             fiz2             = _mm256_add_pd(fiz2,tz);
2012
2013             fjx1             = _mm256_add_pd(fjx1,tx);
2014             fjy1             = _mm256_add_pd(fjy1,ty);
2015             fjz1             = _mm256_add_pd(fjz1,tz);
2016
2017             }
2018
2019             /**************************
2020              * CALCULATE INTERACTIONS *
2021              **************************/
2022
2023             if (gmx_mm256_any_lt(rsq22,rcutoff2))
2024             {
2025
2026             r22              = _mm256_mul_pd(rsq22,rinv22);
2027
2028             /* EWALD ELECTROSTATICS */
2029
2030             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
2031             ewrt             = _mm256_mul_pd(r22,ewtabscale);
2032             ewitab           = _mm256_cvttpd_epi32(ewrt);
2033             eweps            = _mm256_sub_pd(ewrt,_mm256_round_pd(ewrt, _MM_FROUND_FLOOR));
2034             gmx_mm256_load_4pair_swizzle_pd(ewtab + _mm_extract_epi32(ewitab,0),ewtab + _mm_extract_epi32(ewitab,1),
2035                                             ewtab + _mm_extract_epi32(ewitab,2),ewtab + _mm_extract_epi32(ewitab,3),
2036                                             &ewtabF,&ewtabFn);
2037             felec            = _mm256_add_pd(_mm256_mul_pd( _mm256_sub_pd(one,eweps),ewtabF),_mm256_mul_pd(eweps,ewtabFn));
2038             felec            = _mm256_mul_pd(_mm256_mul_pd(qq22,rinv22),_mm256_sub_pd(rinvsq22,felec));
2039
2040             cutoff_mask      = _mm256_cmp_pd(rsq22,rcutoff2,_CMP_LT_OQ);
2041
2042             fscal            = felec;
2043
2044             fscal            = _mm256_and_pd(fscal,cutoff_mask);
2045
2046             /* Calculate temporary vectorial force */
2047             tx               = _mm256_mul_pd(fscal,dx22);
2048             ty               = _mm256_mul_pd(fscal,dy22);
2049             tz               = _mm256_mul_pd(fscal,dz22);
2050
2051             /* Update vectorial force */
2052             fix2             = _mm256_add_pd(fix2,tx);
2053             fiy2             = _mm256_add_pd(fiy2,ty);
2054             fiz2             = _mm256_add_pd(fiz2,tz);
2055
2056             fjx2             = _mm256_add_pd(fjx2,tx);
2057             fjy2             = _mm256_add_pd(fjy2,ty);
2058             fjz2             = _mm256_add_pd(fjz2,tz);
2059
2060             }
2061
2062             /**************************
2063              * CALCULATE INTERACTIONS *
2064              **************************/
2065
2066             if (gmx_mm256_any_lt(rsq23,rcutoff2))
2067             {
2068
2069             r23              = _mm256_mul_pd(rsq23,rinv23);
2070
2071             /* EWALD ELECTROSTATICS */
2072
2073             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
2074             ewrt             = _mm256_mul_pd(r23,ewtabscale);
2075             ewitab           = _mm256_cvttpd_epi32(ewrt);
2076             eweps            = _mm256_sub_pd(ewrt,_mm256_round_pd(ewrt, _MM_FROUND_FLOOR));
2077             gmx_mm256_load_4pair_swizzle_pd(ewtab + _mm_extract_epi32(ewitab,0),ewtab + _mm_extract_epi32(ewitab,1),
2078                                             ewtab + _mm_extract_epi32(ewitab,2),ewtab + _mm_extract_epi32(ewitab,3),
2079                                             &ewtabF,&ewtabFn);
2080             felec            = _mm256_add_pd(_mm256_mul_pd( _mm256_sub_pd(one,eweps),ewtabF),_mm256_mul_pd(eweps,ewtabFn));
2081             felec            = _mm256_mul_pd(_mm256_mul_pd(qq23,rinv23),_mm256_sub_pd(rinvsq23,felec));
2082
2083             cutoff_mask      = _mm256_cmp_pd(rsq23,rcutoff2,_CMP_LT_OQ);
2084
2085             fscal            = felec;
2086
2087             fscal            = _mm256_and_pd(fscal,cutoff_mask);
2088
2089             /* Calculate temporary vectorial force */
2090             tx               = _mm256_mul_pd(fscal,dx23);
2091             ty               = _mm256_mul_pd(fscal,dy23);
2092             tz               = _mm256_mul_pd(fscal,dz23);
2093
2094             /* Update vectorial force */
2095             fix2             = _mm256_add_pd(fix2,tx);
2096             fiy2             = _mm256_add_pd(fiy2,ty);
2097             fiz2             = _mm256_add_pd(fiz2,tz);
2098
2099             fjx3             = _mm256_add_pd(fjx3,tx);
2100             fjy3             = _mm256_add_pd(fjy3,ty);
2101             fjz3             = _mm256_add_pd(fjz3,tz);
2102
2103             }
2104
2105             /**************************
2106              * CALCULATE INTERACTIONS *
2107              **************************/
2108
2109             if (gmx_mm256_any_lt(rsq31,rcutoff2))
2110             {
2111
2112             r31              = _mm256_mul_pd(rsq31,rinv31);
2113
2114             /* EWALD ELECTROSTATICS */
2115
2116             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
2117             ewrt             = _mm256_mul_pd(r31,ewtabscale);
2118             ewitab           = _mm256_cvttpd_epi32(ewrt);
2119             eweps            = _mm256_sub_pd(ewrt,_mm256_round_pd(ewrt, _MM_FROUND_FLOOR));
2120             gmx_mm256_load_4pair_swizzle_pd(ewtab + _mm_extract_epi32(ewitab,0),ewtab + _mm_extract_epi32(ewitab,1),
2121                                             ewtab + _mm_extract_epi32(ewitab,2),ewtab + _mm_extract_epi32(ewitab,3),
2122                                             &ewtabF,&ewtabFn);
2123             felec            = _mm256_add_pd(_mm256_mul_pd( _mm256_sub_pd(one,eweps),ewtabF),_mm256_mul_pd(eweps,ewtabFn));
2124             felec            = _mm256_mul_pd(_mm256_mul_pd(qq31,rinv31),_mm256_sub_pd(rinvsq31,felec));
2125
2126             cutoff_mask      = _mm256_cmp_pd(rsq31,rcutoff2,_CMP_LT_OQ);
2127
2128             fscal            = felec;
2129
2130             fscal            = _mm256_and_pd(fscal,cutoff_mask);
2131
2132             /* Calculate temporary vectorial force */
2133             tx               = _mm256_mul_pd(fscal,dx31);
2134             ty               = _mm256_mul_pd(fscal,dy31);
2135             tz               = _mm256_mul_pd(fscal,dz31);
2136
2137             /* Update vectorial force */
2138             fix3             = _mm256_add_pd(fix3,tx);
2139             fiy3             = _mm256_add_pd(fiy3,ty);
2140             fiz3             = _mm256_add_pd(fiz3,tz);
2141
2142             fjx1             = _mm256_add_pd(fjx1,tx);
2143             fjy1             = _mm256_add_pd(fjy1,ty);
2144             fjz1             = _mm256_add_pd(fjz1,tz);
2145
2146             }
2147
2148             /**************************
2149              * CALCULATE INTERACTIONS *
2150              **************************/
2151
2152             if (gmx_mm256_any_lt(rsq32,rcutoff2))
2153             {
2154
2155             r32              = _mm256_mul_pd(rsq32,rinv32);
2156
2157             /* EWALD ELECTROSTATICS */
2158
2159             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
2160             ewrt             = _mm256_mul_pd(r32,ewtabscale);
2161             ewitab           = _mm256_cvttpd_epi32(ewrt);
2162             eweps            = _mm256_sub_pd(ewrt,_mm256_round_pd(ewrt, _MM_FROUND_FLOOR));
2163             gmx_mm256_load_4pair_swizzle_pd(ewtab + _mm_extract_epi32(ewitab,0),ewtab + _mm_extract_epi32(ewitab,1),
2164                                             ewtab + _mm_extract_epi32(ewitab,2),ewtab + _mm_extract_epi32(ewitab,3),
2165                                             &ewtabF,&ewtabFn);
2166             felec            = _mm256_add_pd(_mm256_mul_pd( _mm256_sub_pd(one,eweps),ewtabF),_mm256_mul_pd(eweps,ewtabFn));
2167             felec            = _mm256_mul_pd(_mm256_mul_pd(qq32,rinv32),_mm256_sub_pd(rinvsq32,felec));
2168
2169             cutoff_mask      = _mm256_cmp_pd(rsq32,rcutoff2,_CMP_LT_OQ);
2170
2171             fscal            = felec;
2172
2173             fscal            = _mm256_and_pd(fscal,cutoff_mask);
2174
2175             /* Calculate temporary vectorial force */
2176             tx               = _mm256_mul_pd(fscal,dx32);
2177             ty               = _mm256_mul_pd(fscal,dy32);
2178             tz               = _mm256_mul_pd(fscal,dz32);
2179
2180             /* Update vectorial force */
2181             fix3             = _mm256_add_pd(fix3,tx);
2182             fiy3             = _mm256_add_pd(fiy3,ty);
2183             fiz3             = _mm256_add_pd(fiz3,tz);
2184
2185             fjx2             = _mm256_add_pd(fjx2,tx);
2186             fjy2             = _mm256_add_pd(fjy2,ty);
2187             fjz2             = _mm256_add_pd(fjz2,tz);
2188
2189             }
2190
2191             /**************************
2192              * CALCULATE INTERACTIONS *
2193              **************************/
2194
2195             if (gmx_mm256_any_lt(rsq33,rcutoff2))
2196             {
2197
2198             r33              = _mm256_mul_pd(rsq33,rinv33);
2199
2200             /* EWALD ELECTROSTATICS */
2201
2202             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
2203             ewrt             = _mm256_mul_pd(r33,ewtabscale);
2204             ewitab           = _mm256_cvttpd_epi32(ewrt);
2205             eweps            = _mm256_sub_pd(ewrt,_mm256_round_pd(ewrt, _MM_FROUND_FLOOR));
2206             gmx_mm256_load_4pair_swizzle_pd(ewtab + _mm_extract_epi32(ewitab,0),ewtab + _mm_extract_epi32(ewitab,1),
2207                                             ewtab + _mm_extract_epi32(ewitab,2),ewtab + _mm_extract_epi32(ewitab,3),
2208                                             &ewtabF,&ewtabFn);
2209             felec            = _mm256_add_pd(_mm256_mul_pd( _mm256_sub_pd(one,eweps),ewtabF),_mm256_mul_pd(eweps,ewtabFn));
2210             felec            = _mm256_mul_pd(_mm256_mul_pd(qq33,rinv33),_mm256_sub_pd(rinvsq33,felec));
2211
2212             cutoff_mask      = _mm256_cmp_pd(rsq33,rcutoff2,_CMP_LT_OQ);
2213
2214             fscal            = felec;
2215
2216             fscal            = _mm256_and_pd(fscal,cutoff_mask);
2217
2218             /* Calculate temporary vectorial force */
2219             tx               = _mm256_mul_pd(fscal,dx33);
2220             ty               = _mm256_mul_pd(fscal,dy33);
2221             tz               = _mm256_mul_pd(fscal,dz33);
2222
2223             /* Update vectorial force */
2224             fix3             = _mm256_add_pd(fix3,tx);
2225             fiy3             = _mm256_add_pd(fiy3,ty);
2226             fiz3             = _mm256_add_pd(fiz3,tz);
2227
2228             fjx3             = _mm256_add_pd(fjx3,tx);
2229             fjy3             = _mm256_add_pd(fjy3,ty);
2230             fjz3             = _mm256_add_pd(fjz3,tz);
2231
2232             }
2233
2234             fjptrA             = f+j_coord_offsetA;
2235             fjptrB             = f+j_coord_offsetB;
2236             fjptrC             = f+j_coord_offsetC;
2237             fjptrD             = f+j_coord_offsetD;
2238
2239             gmx_mm256_decrement_4rvec_4ptr_swizzle_pd(fjptrA,fjptrB,fjptrC,fjptrD,
2240                                                       fjx0,fjy0,fjz0,fjx1,fjy1,fjz1,
2241                                                       fjx2,fjy2,fjz2,fjx3,fjy3,fjz3);
2242
2243             /* Inner loop uses 384 flops */
2244         }
2245
2246         if(jidx<j_index_end)
2247         {
2248
2249             /* Get j neighbor index, and coordinate index */
2250             jnrlistA         = jjnr[jidx];
2251             jnrlistB         = jjnr[jidx+1];
2252             jnrlistC         = jjnr[jidx+2];
2253             jnrlistD         = jjnr[jidx+3];
2254             /* Sign of each element will be negative for non-real atoms.
2255              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
2256              * so use it as val = _mm_andnot_pd(mask,val) to clear dummy entries.
2257              */
2258             tmpmask0 = gmx_mm_castsi128_pd(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
2259
2260             tmpmask1 = _mm_permute_ps(tmpmask0,_GMX_MM_PERMUTE(3,3,2,2));
2261             tmpmask0 = _mm_permute_ps(tmpmask0,_GMX_MM_PERMUTE(1,1,0,0));
2262             dummy_mask = _mm256_castps_pd(gmx_mm256_set_m128(tmpmask1,tmpmask0));
2263
2264             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
2265             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
2266             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
2267             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
2268             j_coord_offsetA  = DIM*jnrA;
2269             j_coord_offsetB  = DIM*jnrB;
2270             j_coord_offsetC  = DIM*jnrC;
2271             j_coord_offsetD  = DIM*jnrD;
2272
2273             /* load j atom coordinates */
2274             gmx_mm256_load_4rvec_4ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
2275                                                  x+j_coord_offsetC,x+j_coord_offsetD,
2276                                                  &jx0,&jy0,&jz0,&jx1,&jy1,&jz1,&jx2,
2277                                                  &jy2,&jz2,&jx3,&jy3,&jz3);
2278
2279             /* Calculate displacement vector */
2280             dx00             = _mm256_sub_pd(ix0,jx0);
2281             dy00             = _mm256_sub_pd(iy0,jy0);
2282             dz00             = _mm256_sub_pd(iz0,jz0);
2283             dx11             = _mm256_sub_pd(ix1,jx1);
2284             dy11             = _mm256_sub_pd(iy1,jy1);
2285             dz11             = _mm256_sub_pd(iz1,jz1);
2286             dx12             = _mm256_sub_pd(ix1,jx2);
2287             dy12             = _mm256_sub_pd(iy1,jy2);
2288             dz12             = _mm256_sub_pd(iz1,jz2);
2289             dx13             = _mm256_sub_pd(ix1,jx3);
2290             dy13             = _mm256_sub_pd(iy1,jy3);
2291             dz13             = _mm256_sub_pd(iz1,jz3);
2292             dx21             = _mm256_sub_pd(ix2,jx1);
2293             dy21             = _mm256_sub_pd(iy2,jy1);
2294             dz21             = _mm256_sub_pd(iz2,jz1);
2295             dx22             = _mm256_sub_pd(ix2,jx2);
2296             dy22             = _mm256_sub_pd(iy2,jy2);
2297             dz22             = _mm256_sub_pd(iz2,jz2);
2298             dx23             = _mm256_sub_pd(ix2,jx3);
2299             dy23             = _mm256_sub_pd(iy2,jy3);
2300             dz23             = _mm256_sub_pd(iz2,jz3);
2301             dx31             = _mm256_sub_pd(ix3,jx1);
2302             dy31             = _mm256_sub_pd(iy3,jy1);
2303             dz31             = _mm256_sub_pd(iz3,jz1);
2304             dx32             = _mm256_sub_pd(ix3,jx2);
2305             dy32             = _mm256_sub_pd(iy3,jy2);
2306             dz32             = _mm256_sub_pd(iz3,jz2);
2307             dx33             = _mm256_sub_pd(ix3,jx3);
2308             dy33             = _mm256_sub_pd(iy3,jy3);
2309             dz33             = _mm256_sub_pd(iz3,jz3);
2310
2311             /* Calculate squared distance and things based on it */
2312             rsq00            = gmx_mm256_calc_rsq_pd(dx00,dy00,dz00);
2313             rsq11            = gmx_mm256_calc_rsq_pd(dx11,dy11,dz11);
2314             rsq12            = gmx_mm256_calc_rsq_pd(dx12,dy12,dz12);
2315             rsq13            = gmx_mm256_calc_rsq_pd(dx13,dy13,dz13);
2316             rsq21            = gmx_mm256_calc_rsq_pd(dx21,dy21,dz21);
2317             rsq22            = gmx_mm256_calc_rsq_pd(dx22,dy22,dz22);
2318             rsq23            = gmx_mm256_calc_rsq_pd(dx23,dy23,dz23);
2319             rsq31            = gmx_mm256_calc_rsq_pd(dx31,dy31,dz31);
2320             rsq32            = gmx_mm256_calc_rsq_pd(dx32,dy32,dz32);
2321             rsq33            = gmx_mm256_calc_rsq_pd(dx33,dy33,dz33);
2322
2323             rinv11           = gmx_mm256_invsqrt_pd(rsq11);
2324             rinv12           = gmx_mm256_invsqrt_pd(rsq12);
2325             rinv13           = gmx_mm256_invsqrt_pd(rsq13);
2326             rinv21           = gmx_mm256_invsqrt_pd(rsq21);
2327             rinv22           = gmx_mm256_invsqrt_pd(rsq22);
2328             rinv23           = gmx_mm256_invsqrt_pd(rsq23);
2329             rinv31           = gmx_mm256_invsqrt_pd(rsq31);
2330             rinv32           = gmx_mm256_invsqrt_pd(rsq32);
2331             rinv33           = gmx_mm256_invsqrt_pd(rsq33);
2332
2333             rinvsq00         = gmx_mm256_inv_pd(rsq00);
2334             rinvsq11         = _mm256_mul_pd(rinv11,rinv11);
2335             rinvsq12         = _mm256_mul_pd(rinv12,rinv12);
2336             rinvsq13         = _mm256_mul_pd(rinv13,rinv13);
2337             rinvsq21         = _mm256_mul_pd(rinv21,rinv21);
2338             rinvsq22         = _mm256_mul_pd(rinv22,rinv22);
2339             rinvsq23         = _mm256_mul_pd(rinv23,rinv23);
2340             rinvsq31         = _mm256_mul_pd(rinv31,rinv31);
2341             rinvsq32         = _mm256_mul_pd(rinv32,rinv32);
2342             rinvsq33         = _mm256_mul_pd(rinv33,rinv33);
2343
2344             fjx0             = _mm256_setzero_pd();
2345             fjy0             = _mm256_setzero_pd();
2346             fjz0             = _mm256_setzero_pd();
2347             fjx1             = _mm256_setzero_pd();
2348             fjy1             = _mm256_setzero_pd();
2349             fjz1             = _mm256_setzero_pd();
2350             fjx2             = _mm256_setzero_pd();
2351             fjy2             = _mm256_setzero_pd();
2352             fjz2             = _mm256_setzero_pd();
2353             fjx3             = _mm256_setzero_pd();
2354             fjy3             = _mm256_setzero_pd();
2355             fjz3             = _mm256_setzero_pd();
2356
2357             /**************************
2358              * CALCULATE INTERACTIONS *
2359              **************************/
2360
2361             if (gmx_mm256_any_lt(rsq00,rcutoff2))
2362             {
2363
2364             /* LENNARD-JONES DISPERSION/REPULSION */
2365
2366             rinvsix          = _mm256_mul_pd(_mm256_mul_pd(rinvsq00,rinvsq00),rinvsq00);
2367             fvdw             = _mm256_mul_pd(_mm256_sub_pd(_mm256_mul_pd(c12_00,rinvsix),c6_00),_mm256_mul_pd(rinvsix,rinvsq00));
2368
2369             cutoff_mask      = _mm256_cmp_pd(rsq00,rcutoff2,_CMP_LT_OQ);
2370
2371             fscal            = fvdw;
2372
2373             fscal            = _mm256_and_pd(fscal,cutoff_mask);
2374
2375             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
2376
2377             /* Calculate temporary vectorial force */
2378             tx               = _mm256_mul_pd(fscal,dx00);
2379             ty               = _mm256_mul_pd(fscal,dy00);
2380             tz               = _mm256_mul_pd(fscal,dz00);
2381
2382             /* Update vectorial force */
2383             fix0             = _mm256_add_pd(fix0,tx);
2384             fiy0             = _mm256_add_pd(fiy0,ty);
2385             fiz0             = _mm256_add_pd(fiz0,tz);
2386
2387             fjx0             = _mm256_add_pd(fjx0,tx);
2388             fjy0             = _mm256_add_pd(fjy0,ty);
2389             fjz0             = _mm256_add_pd(fjz0,tz);
2390
2391             }
2392
2393             /**************************
2394              * CALCULATE INTERACTIONS *
2395              **************************/
2396
2397             if (gmx_mm256_any_lt(rsq11,rcutoff2))
2398             {
2399
2400             r11              = _mm256_mul_pd(rsq11,rinv11);
2401             r11              = _mm256_andnot_pd(dummy_mask,r11);
2402
2403             /* EWALD ELECTROSTATICS */
2404
2405             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
2406             ewrt             = _mm256_mul_pd(r11,ewtabscale);
2407             ewitab           = _mm256_cvttpd_epi32(ewrt);
2408             eweps            = _mm256_sub_pd(ewrt,_mm256_round_pd(ewrt, _MM_FROUND_FLOOR));
2409             gmx_mm256_load_4pair_swizzle_pd(ewtab + _mm_extract_epi32(ewitab,0),ewtab + _mm_extract_epi32(ewitab,1),
2410                                             ewtab + _mm_extract_epi32(ewitab,2),ewtab + _mm_extract_epi32(ewitab,3),
2411                                             &ewtabF,&ewtabFn);
2412             felec            = _mm256_add_pd(_mm256_mul_pd( _mm256_sub_pd(one,eweps),ewtabF),_mm256_mul_pd(eweps,ewtabFn));
2413             felec            = _mm256_mul_pd(_mm256_mul_pd(qq11,rinv11),_mm256_sub_pd(rinvsq11,felec));
2414
2415             cutoff_mask      = _mm256_cmp_pd(rsq11,rcutoff2,_CMP_LT_OQ);
2416
2417             fscal            = felec;
2418
2419             fscal            = _mm256_and_pd(fscal,cutoff_mask);
2420
2421             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
2422
2423             /* Calculate temporary vectorial force */
2424             tx               = _mm256_mul_pd(fscal,dx11);
2425             ty               = _mm256_mul_pd(fscal,dy11);
2426             tz               = _mm256_mul_pd(fscal,dz11);
2427
2428             /* Update vectorial force */
2429             fix1             = _mm256_add_pd(fix1,tx);
2430             fiy1             = _mm256_add_pd(fiy1,ty);
2431             fiz1             = _mm256_add_pd(fiz1,tz);
2432
2433             fjx1             = _mm256_add_pd(fjx1,tx);
2434             fjy1             = _mm256_add_pd(fjy1,ty);
2435             fjz1             = _mm256_add_pd(fjz1,tz);
2436
2437             }
2438
2439             /**************************
2440              * CALCULATE INTERACTIONS *
2441              **************************/
2442
2443             if (gmx_mm256_any_lt(rsq12,rcutoff2))
2444             {
2445
2446             r12              = _mm256_mul_pd(rsq12,rinv12);
2447             r12              = _mm256_andnot_pd(dummy_mask,r12);
2448
2449             /* EWALD ELECTROSTATICS */
2450
2451             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
2452             ewrt             = _mm256_mul_pd(r12,ewtabscale);
2453             ewitab           = _mm256_cvttpd_epi32(ewrt);
2454             eweps            = _mm256_sub_pd(ewrt,_mm256_round_pd(ewrt, _MM_FROUND_FLOOR));
2455             gmx_mm256_load_4pair_swizzle_pd(ewtab + _mm_extract_epi32(ewitab,0),ewtab + _mm_extract_epi32(ewitab,1),
2456                                             ewtab + _mm_extract_epi32(ewitab,2),ewtab + _mm_extract_epi32(ewitab,3),
2457                                             &ewtabF,&ewtabFn);
2458             felec            = _mm256_add_pd(_mm256_mul_pd( _mm256_sub_pd(one,eweps),ewtabF),_mm256_mul_pd(eweps,ewtabFn));
2459             felec            = _mm256_mul_pd(_mm256_mul_pd(qq12,rinv12),_mm256_sub_pd(rinvsq12,felec));
2460
2461             cutoff_mask      = _mm256_cmp_pd(rsq12,rcutoff2,_CMP_LT_OQ);
2462
2463             fscal            = felec;
2464
2465             fscal            = _mm256_and_pd(fscal,cutoff_mask);
2466
2467             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
2468
2469             /* Calculate temporary vectorial force */
2470             tx               = _mm256_mul_pd(fscal,dx12);
2471             ty               = _mm256_mul_pd(fscal,dy12);
2472             tz               = _mm256_mul_pd(fscal,dz12);
2473
2474             /* Update vectorial force */
2475             fix1             = _mm256_add_pd(fix1,tx);
2476             fiy1             = _mm256_add_pd(fiy1,ty);
2477             fiz1             = _mm256_add_pd(fiz1,tz);
2478
2479             fjx2             = _mm256_add_pd(fjx2,tx);
2480             fjy2             = _mm256_add_pd(fjy2,ty);
2481             fjz2             = _mm256_add_pd(fjz2,tz);
2482
2483             }
2484
2485             /**************************
2486              * CALCULATE INTERACTIONS *
2487              **************************/
2488
2489             if (gmx_mm256_any_lt(rsq13,rcutoff2))
2490             {
2491
2492             r13              = _mm256_mul_pd(rsq13,rinv13);
2493             r13              = _mm256_andnot_pd(dummy_mask,r13);
2494
2495             /* EWALD ELECTROSTATICS */
2496
2497             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
2498             ewrt             = _mm256_mul_pd(r13,ewtabscale);
2499             ewitab           = _mm256_cvttpd_epi32(ewrt);
2500             eweps            = _mm256_sub_pd(ewrt,_mm256_round_pd(ewrt, _MM_FROUND_FLOOR));
2501             gmx_mm256_load_4pair_swizzle_pd(ewtab + _mm_extract_epi32(ewitab,0),ewtab + _mm_extract_epi32(ewitab,1),
2502                                             ewtab + _mm_extract_epi32(ewitab,2),ewtab + _mm_extract_epi32(ewitab,3),
2503                                             &ewtabF,&ewtabFn);
2504             felec            = _mm256_add_pd(_mm256_mul_pd( _mm256_sub_pd(one,eweps),ewtabF),_mm256_mul_pd(eweps,ewtabFn));
2505             felec            = _mm256_mul_pd(_mm256_mul_pd(qq13,rinv13),_mm256_sub_pd(rinvsq13,felec));
2506
2507             cutoff_mask      = _mm256_cmp_pd(rsq13,rcutoff2,_CMP_LT_OQ);
2508
2509             fscal            = felec;
2510
2511             fscal            = _mm256_and_pd(fscal,cutoff_mask);
2512
2513             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
2514
2515             /* Calculate temporary vectorial force */
2516             tx               = _mm256_mul_pd(fscal,dx13);
2517             ty               = _mm256_mul_pd(fscal,dy13);
2518             tz               = _mm256_mul_pd(fscal,dz13);
2519
2520             /* Update vectorial force */
2521             fix1             = _mm256_add_pd(fix1,tx);
2522             fiy1             = _mm256_add_pd(fiy1,ty);
2523             fiz1             = _mm256_add_pd(fiz1,tz);
2524
2525             fjx3             = _mm256_add_pd(fjx3,tx);
2526             fjy3             = _mm256_add_pd(fjy3,ty);
2527             fjz3             = _mm256_add_pd(fjz3,tz);
2528
2529             }
2530
2531             /**************************
2532              * CALCULATE INTERACTIONS *
2533              **************************/
2534
2535             if (gmx_mm256_any_lt(rsq21,rcutoff2))
2536             {
2537
2538             r21              = _mm256_mul_pd(rsq21,rinv21);
2539             r21              = _mm256_andnot_pd(dummy_mask,r21);
2540
2541             /* EWALD ELECTROSTATICS */
2542
2543             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
2544             ewrt             = _mm256_mul_pd(r21,ewtabscale);
2545             ewitab           = _mm256_cvttpd_epi32(ewrt);
2546             eweps            = _mm256_sub_pd(ewrt,_mm256_round_pd(ewrt, _MM_FROUND_FLOOR));
2547             gmx_mm256_load_4pair_swizzle_pd(ewtab + _mm_extract_epi32(ewitab,0),ewtab + _mm_extract_epi32(ewitab,1),
2548                                             ewtab + _mm_extract_epi32(ewitab,2),ewtab + _mm_extract_epi32(ewitab,3),
2549                                             &ewtabF,&ewtabFn);
2550             felec            = _mm256_add_pd(_mm256_mul_pd( _mm256_sub_pd(one,eweps),ewtabF),_mm256_mul_pd(eweps,ewtabFn));
2551             felec            = _mm256_mul_pd(_mm256_mul_pd(qq21,rinv21),_mm256_sub_pd(rinvsq21,felec));
2552
2553             cutoff_mask      = _mm256_cmp_pd(rsq21,rcutoff2,_CMP_LT_OQ);
2554
2555             fscal            = felec;
2556
2557             fscal            = _mm256_and_pd(fscal,cutoff_mask);
2558
2559             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
2560
2561             /* Calculate temporary vectorial force */
2562             tx               = _mm256_mul_pd(fscal,dx21);
2563             ty               = _mm256_mul_pd(fscal,dy21);
2564             tz               = _mm256_mul_pd(fscal,dz21);
2565
2566             /* Update vectorial force */
2567             fix2             = _mm256_add_pd(fix2,tx);
2568             fiy2             = _mm256_add_pd(fiy2,ty);
2569             fiz2             = _mm256_add_pd(fiz2,tz);
2570
2571             fjx1             = _mm256_add_pd(fjx1,tx);
2572             fjy1             = _mm256_add_pd(fjy1,ty);
2573             fjz1             = _mm256_add_pd(fjz1,tz);
2574
2575             }
2576
2577             /**************************
2578              * CALCULATE INTERACTIONS *
2579              **************************/
2580
2581             if (gmx_mm256_any_lt(rsq22,rcutoff2))
2582             {
2583
2584             r22              = _mm256_mul_pd(rsq22,rinv22);
2585             r22              = _mm256_andnot_pd(dummy_mask,r22);
2586
2587             /* EWALD ELECTROSTATICS */
2588
2589             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
2590             ewrt             = _mm256_mul_pd(r22,ewtabscale);
2591             ewitab           = _mm256_cvttpd_epi32(ewrt);
2592             eweps            = _mm256_sub_pd(ewrt,_mm256_round_pd(ewrt, _MM_FROUND_FLOOR));
2593             gmx_mm256_load_4pair_swizzle_pd(ewtab + _mm_extract_epi32(ewitab,0),ewtab + _mm_extract_epi32(ewitab,1),
2594                                             ewtab + _mm_extract_epi32(ewitab,2),ewtab + _mm_extract_epi32(ewitab,3),
2595                                             &ewtabF,&ewtabFn);
2596             felec            = _mm256_add_pd(_mm256_mul_pd( _mm256_sub_pd(one,eweps),ewtabF),_mm256_mul_pd(eweps,ewtabFn));
2597             felec            = _mm256_mul_pd(_mm256_mul_pd(qq22,rinv22),_mm256_sub_pd(rinvsq22,felec));
2598
2599             cutoff_mask      = _mm256_cmp_pd(rsq22,rcutoff2,_CMP_LT_OQ);
2600
2601             fscal            = felec;
2602
2603             fscal            = _mm256_and_pd(fscal,cutoff_mask);
2604
2605             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
2606
2607             /* Calculate temporary vectorial force */
2608             tx               = _mm256_mul_pd(fscal,dx22);
2609             ty               = _mm256_mul_pd(fscal,dy22);
2610             tz               = _mm256_mul_pd(fscal,dz22);
2611
2612             /* Update vectorial force */
2613             fix2             = _mm256_add_pd(fix2,tx);
2614             fiy2             = _mm256_add_pd(fiy2,ty);
2615             fiz2             = _mm256_add_pd(fiz2,tz);
2616
2617             fjx2             = _mm256_add_pd(fjx2,tx);
2618             fjy2             = _mm256_add_pd(fjy2,ty);
2619             fjz2             = _mm256_add_pd(fjz2,tz);
2620
2621             }
2622
2623             /**************************
2624              * CALCULATE INTERACTIONS *
2625              **************************/
2626
2627             if (gmx_mm256_any_lt(rsq23,rcutoff2))
2628             {
2629
2630             r23              = _mm256_mul_pd(rsq23,rinv23);
2631             r23              = _mm256_andnot_pd(dummy_mask,r23);
2632
2633             /* EWALD ELECTROSTATICS */
2634
2635             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
2636             ewrt             = _mm256_mul_pd(r23,ewtabscale);
2637             ewitab           = _mm256_cvttpd_epi32(ewrt);
2638             eweps            = _mm256_sub_pd(ewrt,_mm256_round_pd(ewrt, _MM_FROUND_FLOOR));
2639             gmx_mm256_load_4pair_swizzle_pd(ewtab + _mm_extract_epi32(ewitab,0),ewtab + _mm_extract_epi32(ewitab,1),
2640                                             ewtab + _mm_extract_epi32(ewitab,2),ewtab + _mm_extract_epi32(ewitab,3),
2641                                             &ewtabF,&ewtabFn);
2642             felec            = _mm256_add_pd(_mm256_mul_pd( _mm256_sub_pd(one,eweps),ewtabF),_mm256_mul_pd(eweps,ewtabFn));
2643             felec            = _mm256_mul_pd(_mm256_mul_pd(qq23,rinv23),_mm256_sub_pd(rinvsq23,felec));
2644
2645             cutoff_mask      = _mm256_cmp_pd(rsq23,rcutoff2,_CMP_LT_OQ);
2646
2647             fscal            = felec;
2648
2649             fscal            = _mm256_and_pd(fscal,cutoff_mask);
2650
2651             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
2652
2653             /* Calculate temporary vectorial force */
2654             tx               = _mm256_mul_pd(fscal,dx23);
2655             ty               = _mm256_mul_pd(fscal,dy23);
2656             tz               = _mm256_mul_pd(fscal,dz23);
2657
2658             /* Update vectorial force */
2659             fix2             = _mm256_add_pd(fix2,tx);
2660             fiy2             = _mm256_add_pd(fiy2,ty);
2661             fiz2             = _mm256_add_pd(fiz2,tz);
2662
2663             fjx3             = _mm256_add_pd(fjx3,tx);
2664             fjy3             = _mm256_add_pd(fjy3,ty);
2665             fjz3             = _mm256_add_pd(fjz3,tz);
2666
2667             }
2668
2669             /**************************
2670              * CALCULATE INTERACTIONS *
2671              **************************/
2672
2673             if (gmx_mm256_any_lt(rsq31,rcutoff2))
2674             {
2675
2676             r31              = _mm256_mul_pd(rsq31,rinv31);
2677             r31              = _mm256_andnot_pd(dummy_mask,r31);
2678
2679             /* EWALD ELECTROSTATICS */
2680
2681             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
2682             ewrt             = _mm256_mul_pd(r31,ewtabscale);
2683             ewitab           = _mm256_cvttpd_epi32(ewrt);
2684             eweps            = _mm256_sub_pd(ewrt,_mm256_round_pd(ewrt, _MM_FROUND_FLOOR));
2685             gmx_mm256_load_4pair_swizzle_pd(ewtab + _mm_extract_epi32(ewitab,0),ewtab + _mm_extract_epi32(ewitab,1),
2686                                             ewtab + _mm_extract_epi32(ewitab,2),ewtab + _mm_extract_epi32(ewitab,3),
2687                                             &ewtabF,&ewtabFn);
2688             felec            = _mm256_add_pd(_mm256_mul_pd( _mm256_sub_pd(one,eweps),ewtabF),_mm256_mul_pd(eweps,ewtabFn));
2689             felec            = _mm256_mul_pd(_mm256_mul_pd(qq31,rinv31),_mm256_sub_pd(rinvsq31,felec));
2690
2691             cutoff_mask      = _mm256_cmp_pd(rsq31,rcutoff2,_CMP_LT_OQ);
2692
2693             fscal            = felec;
2694
2695             fscal            = _mm256_and_pd(fscal,cutoff_mask);
2696
2697             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
2698
2699             /* Calculate temporary vectorial force */
2700             tx               = _mm256_mul_pd(fscal,dx31);
2701             ty               = _mm256_mul_pd(fscal,dy31);
2702             tz               = _mm256_mul_pd(fscal,dz31);
2703
2704             /* Update vectorial force */
2705             fix3             = _mm256_add_pd(fix3,tx);
2706             fiy3             = _mm256_add_pd(fiy3,ty);
2707             fiz3             = _mm256_add_pd(fiz3,tz);
2708
2709             fjx1             = _mm256_add_pd(fjx1,tx);
2710             fjy1             = _mm256_add_pd(fjy1,ty);
2711             fjz1             = _mm256_add_pd(fjz1,tz);
2712
2713             }
2714
2715             /**************************
2716              * CALCULATE INTERACTIONS *
2717              **************************/
2718
2719             if (gmx_mm256_any_lt(rsq32,rcutoff2))
2720             {
2721
2722             r32              = _mm256_mul_pd(rsq32,rinv32);
2723             r32              = _mm256_andnot_pd(dummy_mask,r32);
2724
2725             /* EWALD ELECTROSTATICS */
2726
2727             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
2728             ewrt             = _mm256_mul_pd(r32,ewtabscale);
2729             ewitab           = _mm256_cvttpd_epi32(ewrt);
2730             eweps            = _mm256_sub_pd(ewrt,_mm256_round_pd(ewrt, _MM_FROUND_FLOOR));
2731             gmx_mm256_load_4pair_swizzle_pd(ewtab + _mm_extract_epi32(ewitab,0),ewtab + _mm_extract_epi32(ewitab,1),
2732                                             ewtab + _mm_extract_epi32(ewitab,2),ewtab + _mm_extract_epi32(ewitab,3),
2733                                             &ewtabF,&ewtabFn);
2734             felec            = _mm256_add_pd(_mm256_mul_pd( _mm256_sub_pd(one,eweps),ewtabF),_mm256_mul_pd(eweps,ewtabFn));
2735             felec            = _mm256_mul_pd(_mm256_mul_pd(qq32,rinv32),_mm256_sub_pd(rinvsq32,felec));
2736
2737             cutoff_mask      = _mm256_cmp_pd(rsq32,rcutoff2,_CMP_LT_OQ);
2738
2739             fscal            = felec;
2740
2741             fscal            = _mm256_and_pd(fscal,cutoff_mask);
2742
2743             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
2744
2745             /* Calculate temporary vectorial force */
2746             tx               = _mm256_mul_pd(fscal,dx32);
2747             ty               = _mm256_mul_pd(fscal,dy32);
2748             tz               = _mm256_mul_pd(fscal,dz32);
2749
2750             /* Update vectorial force */
2751             fix3             = _mm256_add_pd(fix3,tx);
2752             fiy3             = _mm256_add_pd(fiy3,ty);
2753             fiz3             = _mm256_add_pd(fiz3,tz);
2754
2755             fjx2             = _mm256_add_pd(fjx2,tx);
2756             fjy2             = _mm256_add_pd(fjy2,ty);
2757             fjz2             = _mm256_add_pd(fjz2,tz);
2758
2759             }
2760
2761             /**************************
2762              * CALCULATE INTERACTIONS *
2763              **************************/
2764
2765             if (gmx_mm256_any_lt(rsq33,rcutoff2))
2766             {
2767
2768             r33              = _mm256_mul_pd(rsq33,rinv33);
2769             r33              = _mm256_andnot_pd(dummy_mask,r33);
2770
2771             /* EWALD ELECTROSTATICS */
2772
2773             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
2774             ewrt             = _mm256_mul_pd(r33,ewtabscale);
2775             ewitab           = _mm256_cvttpd_epi32(ewrt);
2776             eweps            = _mm256_sub_pd(ewrt,_mm256_round_pd(ewrt, _MM_FROUND_FLOOR));
2777             gmx_mm256_load_4pair_swizzle_pd(ewtab + _mm_extract_epi32(ewitab,0),ewtab + _mm_extract_epi32(ewitab,1),
2778                                             ewtab + _mm_extract_epi32(ewitab,2),ewtab + _mm_extract_epi32(ewitab,3),
2779                                             &ewtabF,&ewtabFn);
2780             felec            = _mm256_add_pd(_mm256_mul_pd( _mm256_sub_pd(one,eweps),ewtabF),_mm256_mul_pd(eweps,ewtabFn));
2781             felec            = _mm256_mul_pd(_mm256_mul_pd(qq33,rinv33),_mm256_sub_pd(rinvsq33,felec));
2782
2783             cutoff_mask      = _mm256_cmp_pd(rsq33,rcutoff2,_CMP_LT_OQ);
2784
2785             fscal            = felec;
2786
2787             fscal            = _mm256_and_pd(fscal,cutoff_mask);
2788
2789             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
2790
2791             /* Calculate temporary vectorial force */
2792             tx               = _mm256_mul_pd(fscal,dx33);
2793             ty               = _mm256_mul_pd(fscal,dy33);
2794             tz               = _mm256_mul_pd(fscal,dz33);
2795
2796             /* Update vectorial force */
2797             fix3             = _mm256_add_pd(fix3,tx);
2798             fiy3             = _mm256_add_pd(fiy3,ty);
2799             fiz3             = _mm256_add_pd(fiz3,tz);
2800
2801             fjx3             = _mm256_add_pd(fjx3,tx);
2802             fjy3             = _mm256_add_pd(fjy3,ty);
2803             fjz3             = _mm256_add_pd(fjz3,tz);
2804
2805             }
2806
2807             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
2808             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
2809             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
2810             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
2811
2812             gmx_mm256_decrement_4rvec_4ptr_swizzle_pd(fjptrA,fjptrB,fjptrC,fjptrD,
2813                                                       fjx0,fjy0,fjz0,fjx1,fjy1,fjz1,
2814                                                       fjx2,fjy2,fjz2,fjx3,fjy3,fjz3);
2815
2816             /* Inner loop uses 393 flops */
2817         }
2818
2819         /* End of innermost loop */
2820
2821         gmx_mm256_update_iforce_4atom_swizzle_pd(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,fix3,fiy3,fiz3,
2822                                                  f+i_coord_offset,fshift+i_shift_offset);
2823
2824         /* Increment number of inner iterations */
2825         inneriter                  += j_index_end - j_index_start;
2826
2827         /* Outer loop uses 24 flops */
2828     }
2829
2830     /* Increment number of outer iterations */
2831     outeriter        += nri;
2832
2833     /* Update outer/inner flops */
2834
2835     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W4W4_F,outeriter*24 + inneriter*393);
2836 }