src/gromacs/gmxlib/nonbonded/nb_kernel_avx_256_double/nb_kernel_ElecEwSw_VdwNone_GeomW4W4_avx_256_double.c

   1 /*
   2  * This file is part of the GROMACS molecular simulation package.
   3  *
   4  * Copyright (c) 2012,2013,2014, by the GROMACS development team, led by
   5  * Mark Abraham, David van der Spoel, Berk Hess, and Erik Lindahl,
   6  * and including many others, as listed in the AUTHORS file in the
   7  * top-level source directory and at http://www.gromacs.org.
   8  *
   9  * GROMACS is free software; you can redistribute it and/or
  10  * modify it under the terms of the GNU Lesser General Public License
  11  * as published by the Free Software Foundation; either version 2.1
  12  * of the License, or (at your option) any later version.
  13  *
  14  * GROMACS is distributed in the hope that it will be useful,
  15  * but WITHOUT ANY WARRANTY; without even the implied warranty of
  16  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
  17  * Lesser General Public License for more details.
  18  *
  19  * You should have received a copy of the GNU Lesser General Public
  20  * License along with GROMACS; if not, see
  21  * http://www.gnu.org/licenses, or write to the Free Software Foundation,
  22  * Inc., 51 Franklin Street, Fifth Floor, Boston, MA  02110-1301  USA.
  23  *
  24  * If you want to redistribute modifications to GROMACS, please
  25  * consider that scientific software is very special. Version
  26  * control is crucial - bugs must be traceable. We will be happy to
  27  * consider code for inclusion in the official distribution, but
  28  * derived work must not be called official GROMACS. Details are found
  29  * in the README & COPYING files - if they are missing, get the
  30  * official version at http://www.gromacs.org.
  31  *
  32  * To help us fund GROMACS development, we humbly ask that you cite
  33  * the research papers on the package. Check out http://www.gromacs.org.
  34  */
  35 /*
  36  * Note: this file was generated by the GROMACS avx_256_double kernel generator.
  37  */
  38 #ifdef HAVE_CONFIG_H
  39 #include <config.h>
  40 #endif
  41
  42 #include <math.h>
  43
  44 #include "../nb_kernel.h"
  45 #include "types/simple.h"
  46 #include "gromacs/math/vec.h"
  47 #include "nrnb.h"
  48
  49 #include "gromacs/simd/math_x86_avx_256_double.h"
  50 #include "kernelutil_x86_avx_256_double.h"
  51
  52 /*
  53  * Gromacs nonbonded kernel:   nb_kernel_ElecEwSw_VdwNone_GeomW4W4_VF_avx_256_double
  54  * Electrostatics interaction: Ewald
  55  * VdW interaction:            None
  56  * Geometry:                   Water4-Water4
  57  * Calculate force/pot:        PotentialAndForce
  58  */
  59 void
  60 nb_kernel_ElecEwSw_VdwNone_GeomW4W4_VF_avx_256_double
  61                     (t_nblist                    * gmx_restrict       nlist,
  62                      rvec                        * gmx_restrict          xx,
  63                      rvec                        * gmx_restrict          ff,
  64                      t_forcerec                  * gmx_restrict          fr,
  65                      t_mdatoms                   * gmx_restrict     mdatoms,
  66                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
  67                      t_nrnb                      * gmx_restrict        nrnb)
  68 {
  69     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
  70      * just 0 for non-waters.
  71      * Suffixes A,B,C,D refer to j loop unrolling done with AVX, e.g. for the four different
  72      * jnr indices corresponding to data put in the four positions in the SIMD register.
  73      */
  74     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
  75     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
  76     int              jnrA,jnrB,jnrC,jnrD;
  77     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
  78     int              jnrlistE,jnrlistF,jnrlistG,jnrlistH;
  79     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
  80     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
  81     real             rcutoff_scalar;
  82     real             *shiftvec,*fshift,*x,*f;
  83     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD;
  84     real             scratch[4*DIM];
  85     __m256d          tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
  86     real *           vdwioffsetptr1;
  87     __m256d          ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
  88     real *           vdwioffsetptr2;
  89     __m256d          ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
  90     real *           vdwioffsetptr3;
  91     __m256d          ix3,iy3,iz3,fix3,fiy3,fiz3,iq3,isai3;
  92     int              vdwjidx1A,vdwjidx1B,vdwjidx1C,vdwjidx1D;
  93     __m256d          jx1,jy1,jz1,fjx1,fjy1,fjz1,jq1,isaj1;
  94     int              vdwjidx2A,vdwjidx2B,vdwjidx2C,vdwjidx2D;
  95     __m256d          jx2,jy2,jz2,fjx2,fjy2,fjz2,jq2,isaj2;
  96     int              vdwjidx3A,vdwjidx3B,vdwjidx3C,vdwjidx3D;
  97     __m256d          jx3,jy3,jz3,fjx3,fjy3,fjz3,jq3,isaj3;
  98     __m256d          dx11,dy11,dz11,rsq11,rinv11,rinvsq11,r11,qq11,c6_11,c12_11;
  99     __m256d          dx12,dy12,dz12,rsq12,rinv12,rinvsq12,r12,qq12,c6_12,c12_12;
 100     __m256d          dx13,dy13,dz13,rsq13,rinv13,rinvsq13,r13,qq13,c6_13,c12_13;
 101     __m256d          dx21,dy21,dz21,rsq21,rinv21,rinvsq21,r21,qq21,c6_21,c12_21;
 102     __m256d          dx22,dy22,dz22,rsq22,rinv22,rinvsq22,r22,qq22,c6_22,c12_22;
 103     __m256d          dx23,dy23,dz23,rsq23,rinv23,rinvsq23,r23,qq23,c6_23,c12_23;
 104     __m256d          dx31,dy31,dz31,rsq31,rinv31,rinvsq31,r31,qq31,c6_31,c12_31;
 105     __m256d          dx32,dy32,dz32,rsq32,rinv32,rinvsq32,r32,qq32,c6_32,c12_32;
 106     __m256d          dx33,dy33,dz33,rsq33,rinv33,rinvsq33,r33,qq33,c6_33,c12_33;
 107     __m256d          velec,felec,velecsum,facel,crf,krf,krf2;
 108     real             *charge;
 109     __m128i          ewitab;
 110     __m256d          ewtabscale,eweps,sh_ewald,ewrt,ewtabhalfspace,ewtabF,ewtabFn,ewtabD,ewtabV;
 111     __m256d          beta,beta2,beta3,zeta2,pmecorrF,pmecorrV,rinv3;
 112     real             *ewtab;
 113     __m256d          rswitch,swV3,swV4,swV5,swF2,swF3,swF4,d,d2,sw,dsw;
 114     real             rswitch_scalar,d_scalar;
 115     __m256d          dummy_mask,cutoff_mask;
 116     __m128           tmpmask0,tmpmask1;
 117     __m256d          signbit = _mm256_castsi256_pd( _mm256_set1_epi32(0x80000000) );
 118     __m256d          one     = _mm256_set1_pd(1.0);
 119     __m256d          two     = _mm256_set1_pd(2.0);
 120     x                = xx[0];
 121     f                = ff[0];
 122
 123     nri              = nlist->nri;
 124     iinr             = nlist->iinr;
 125     jindex           = nlist->jindex;
 126     jjnr             = nlist->jjnr;
 127     shiftidx         = nlist->shift;
 128     gid              = nlist->gid;
 129     shiftvec         = fr->shift_vec[0];
 130     fshift           = fr->fshift[0];
 131     facel            = _mm256_set1_pd(fr->epsfac);
 132     charge           = mdatoms->chargeA;
 133
 134     sh_ewald         = _mm256_set1_pd(fr->ic->sh_ewald);
 135     beta             = _mm256_set1_pd(fr->ic->ewaldcoeff_q);
 136     beta2            = _mm256_mul_pd(beta,beta);
 137     beta3            = _mm256_mul_pd(beta,beta2);
 138
 139     ewtab            = fr->ic->tabq_coul_FDV0;
 140     ewtabscale       = _mm256_set1_pd(fr->ic->tabq_scale);
 141     ewtabhalfspace   = _mm256_set1_pd(0.5/fr->ic->tabq_scale);
 142
 143     /* Setup water-specific parameters */
 144     inr              = nlist->iinr[0];
 145     iq1              = _mm256_mul_pd(facel,_mm256_set1_pd(charge[inr+1]));
 146     iq2              = _mm256_mul_pd(facel,_mm256_set1_pd(charge[inr+2]));
 147     iq3              = _mm256_mul_pd(facel,_mm256_set1_pd(charge[inr+3]));
 148
 149     jq1              = _mm256_set1_pd(charge[inr+1]);
 150     jq2              = _mm256_set1_pd(charge[inr+2]);
 151     jq3              = _mm256_set1_pd(charge[inr+3]);
 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     rswitch_scalar   = fr->rcoulomb_switch;
 168     rswitch          = _mm256_set1_pd(rswitch_scalar);
 169     /* Setup switch parameters */
 170     d_scalar         = rcutoff_scalar-rswitch_scalar;
 171     d                = _mm256_set1_pd(d_scalar);
 172     swV3             = _mm256_set1_pd(-10.0/(d_scalar*d_scalar*d_scalar));
 173     swV4             = _mm256_set1_pd( 15.0/(d_scalar*d_scalar*d_scalar*d_scalar));
 174     swV5             = _mm256_set1_pd( -6.0/(d_scalar*d_scalar*d_scalar*d_scalar*d_scalar));
 175     swF2             = _mm256_set1_pd(-30.0/(d_scalar*d_scalar*d_scalar));
 176     swF3             = _mm256_set1_pd( 60.0/(d_scalar*d_scalar*d_scalar*d_scalar));
 177     swF4             = _mm256_set1_pd(-30.0/(d_scalar*d_scalar*d_scalar*d_scalar*d_scalar));
 178
 179     /* Avoid stupid compiler warnings */
 180     jnrA = jnrB = jnrC = jnrD = 0;
 181     j_coord_offsetA = 0;
 182     j_coord_offsetB = 0;
 183     j_coord_offsetC = 0;
 184     j_coord_offsetD = 0;
 185
 186     outeriter        = 0;
 187     inneriter        = 0;
 188
 189     for(iidx=0;iidx<4*DIM;iidx++)
 190     {
 191         scratch[iidx] = 0.0;
 192     }
 193
 194     /* Start outer loop over neighborlists */
 195     for(iidx=0; iidx<nri; iidx++)
 196     {
 197         /* Load shift vector for this list */
 198         i_shift_offset   = DIM*shiftidx[iidx];
 199
 200         /* Load limits for loop over neighbors */
 201         j_index_start    = jindex[iidx];
 202         j_index_end      = jindex[iidx+1];
 203
 204         /* Get outer coordinate index */
 205         inr              = iinr[iidx];
 206         i_coord_offset   = DIM*inr;
 207
 208         /* Load i particle coords and add shift vector */
 209         gmx_mm256_load_shift_and_3rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset+DIM,
 210                                                     &ix1,&iy1,&iz1,&ix2,&iy2,&iz2,&ix3,&iy3,&iz3);
 211
 212         fix1             = _mm256_setzero_pd();
 213         fiy1             = _mm256_setzero_pd();
 214         fiz1             = _mm256_setzero_pd();
 215         fix2             = _mm256_setzero_pd();
 216         fiy2             = _mm256_setzero_pd();
 217         fiz2             = _mm256_setzero_pd();
 218         fix3             = _mm256_setzero_pd();
 219         fiy3             = _mm256_setzero_pd();
 220         fiz3             = _mm256_setzero_pd();
 221
 222         /* Reset potential sums */
 223         velecsum         = _mm256_setzero_pd();
 224
 225         /* Start inner kernel loop */
 226         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
 227         {
 228
 229             /* Get j neighbor index, and coordinate index */
 230             jnrA             = jjnr[jidx];
 231             jnrB             = jjnr[jidx+1];
 232             jnrC             = jjnr[jidx+2];
 233             jnrD             = jjnr[jidx+3];
 234             j_coord_offsetA  = DIM*jnrA;
 235             j_coord_offsetB  = DIM*jnrB;
 236             j_coord_offsetC  = DIM*jnrC;
 237             j_coord_offsetD  = DIM*jnrD;
 238
 239             /* load j atom coordinates */
 240             gmx_mm256_load_3rvec_4ptr_swizzle_pd(x+j_coord_offsetA+DIM,x+j_coord_offsetB+DIM,
 241                                                  x+j_coord_offsetC+DIM,x+j_coord_offsetD+DIM,
 242                                                  &jx1,&jy1,&jz1,&jx2,&jy2,&jz2,&jx3,&jy3,&jz3);
 243
 244             /* Calculate displacement vector */
 245             dx11             = _mm256_sub_pd(ix1,jx1);
 246             dy11             = _mm256_sub_pd(iy1,jy1);
 247             dz11             = _mm256_sub_pd(iz1,jz1);
 248             dx12             = _mm256_sub_pd(ix1,jx2);
 249             dy12             = _mm256_sub_pd(iy1,jy2);
 250             dz12             = _mm256_sub_pd(iz1,jz2);
 251             dx13             = _mm256_sub_pd(ix1,jx3);
 252             dy13             = _mm256_sub_pd(iy1,jy3);
 253             dz13             = _mm256_sub_pd(iz1,jz3);
 254             dx21             = _mm256_sub_pd(ix2,jx1);
 255             dy21             = _mm256_sub_pd(iy2,jy1);
 256             dz21             = _mm256_sub_pd(iz2,jz1);
 257             dx22             = _mm256_sub_pd(ix2,jx2);
 258             dy22             = _mm256_sub_pd(iy2,jy2);
 259             dz22             = _mm256_sub_pd(iz2,jz2);
 260             dx23             = _mm256_sub_pd(ix2,jx3);
 261             dy23             = _mm256_sub_pd(iy2,jy3);
 262             dz23             = _mm256_sub_pd(iz2,jz3);
 263             dx31             = _mm256_sub_pd(ix3,jx1);
 264             dy31             = _mm256_sub_pd(iy3,jy1);
 265             dz31             = _mm256_sub_pd(iz3,jz1);
 266             dx32             = _mm256_sub_pd(ix3,jx2);
 267             dy32             = _mm256_sub_pd(iy3,jy2);
 268             dz32             = _mm256_sub_pd(iz3,jz2);
 269             dx33             = _mm256_sub_pd(ix3,jx3);
 270             dy33             = _mm256_sub_pd(iy3,jy3);
 271             dz33             = _mm256_sub_pd(iz3,jz3);
 272
 273             /* Calculate squared distance and things based on it */
 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             rinvsq11         = _mm256_mul_pd(rinv11,rinv11);
 295             rinvsq12         = _mm256_mul_pd(rinv12,rinv12);
 296             rinvsq13         = _mm256_mul_pd(rinv13,rinv13);
 297             rinvsq21         = _mm256_mul_pd(rinv21,rinv21);
 298             rinvsq22         = _mm256_mul_pd(rinv22,rinv22);
 299             rinvsq23         = _mm256_mul_pd(rinv23,rinv23);
 300             rinvsq31         = _mm256_mul_pd(rinv31,rinv31);
 301             rinvsq32         = _mm256_mul_pd(rinv32,rinv32);
 302             rinvsq33         = _mm256_mul_pd(rinv33,rinv33);
 303
 304             fjx1             = _mm256_setzero_pd();
 305             fjy1             = _mm256_setzero_pd();
 306             fjz1             = _mm256_setzero_pd();
 307             fjx2             = _mm256_setzero_pd();
 308             fjy2             = _mm256_setzero_pd();
 309             fjz2             = _mm256_setzero_pd();
 310             fjx3             = _mm256_setzero_pd();
 311             fjy3             = _mm256_setzero_pd();
 312             fjz3             = _mm256_setzero_pd();
 313
 314             /**************************
 315              * CALCULATE INTERACTIONS *
 316              **************************/
 317
 318             if (gmx_mm256_any_lt(rsq11,rcutoff2))
 319             {
 320
 321             r11              = _mm256_mul_pd(rsq11,rinv11);
 322
 323             /* EWALD ELECTROSTATICS */
 324
 325             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 326             ewrt             = _mm256_mul_pd(r11,ewtabscale);
 327             ewitab           = _mm256_cvttpd_epi32(ewrt);
 328             eweps            = _mm256_sub_pd(ewrt,_mm256_round_pd(ewrt, _MM_FROUND_FLOOR));
 329             ewitab           = _mm_slli_epi32(ewitab,2);
 330             ewtabF           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
 331             ewtabD           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,1) );
 332             ewtabV           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,2) );
 333             ewtabFn          = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,3) );
 334             GMX_MM256_FULLTRANSPOSE4_PD(ewtabF,ewtabD,ewtabV,ewtabFn);
 335             felec            = _mm256_add_pd(ewtabF,_mm256_mul_pd(eweps,ewtabD));
 336             velec            = _mm256_sub_pd(ewtabV,_mm256_mul_pd(_mm256_mul_pd(ewtabhalfspace,eweps),_mm256_add_pd(ewtabF,felec)));
 337             velec            = _mm256_mul_pd(qq11,_mm256_sub_pd(rinv11,velec));
 338             felec            = _mm256_mul_pd(_mm256_mul_pd(qq11,rinv11),_mm256_sub_pd(rinvsq11,felec));
 339
 340             d                = _mm256_sub_pd(r11,rswitch);
 341             d                = _mm256_max_pd(d,_mm256_setzero_pd());
 342             d2               = _mm256_mul_pd(d,d);
 343             sw               = _mm256_add_pd(one,_mm256_mul_pd(d2,_mm256_mul_pd(d,_mm256_add_pd(swV3,_mm256_mul_pd(d,_mm256_add_pd(swV4,_mm256_mul_pd(d,swV5)))))));
 344
 345             dsw              = _mm256_mul_pd(d2,_mm256_add_pd(swF2,_mm256_mul_pd(d,_mm256_add_pd(swF3,_mm256_mul_pd(d,swF4)))));
 346
 347             /* Evaluate switch function */
 348             /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
 349             felec            = _mm256_sub_pd( _mm256_mul_pd(felec,sw) , _mm256_mul_pd(rinv11,_mm256_mul_pd(velec,dsw)) );
 350             velec            = _mm256_mul_pd(velec,sw);
 351             cutoff_mask      = _mm256_cmp_pd(rsq11,rcutoff2,_CMP_LT_OQ);
 352
 353             /* Update potential sum for this i atom from the interaction with this j atom. */
 354             velec            = _mm256_and_pd(velec,cutoff_mask);
 355             velecsum         = _mm256_add_pd(velecsum,velec);
 356
 357             fscal            = felec;
 358
 359             fscal            = _mm256_and_pd(fscal,cutoff_mask);
 360
 361             /* Calculate temporary vectorial force */
 362             tx               = _mm256_mul_pd(fscal,dx11);
 363             ty               = _mm256_mul_pd(fscal,dy11);
 364             tz               = _mm256_mul_pd(fscal,dz11);
 365
 366             /* Update vectorial force */
 367             fix1             = _mm256_add_pd(fix1,tx);
 368             fiy1             = _mm256_add_pd(fiy1,ty);
 369             fiz1             = _mm256_add_pd(fiz1,tz);
 370
 371             fjx1             = _mm256_add_pd(fjx1,tx);
 372             fjy1             = _mm256_add_pd(fjy1,ty);
 373             fjz1             = _mm256_add_pd(fjz1,tz);
 374
 375             }
 376
 377             /**************************
 378              * CALCULATE INTERACTIONS *
 379              **************************/
 380
 381             if (gmx_mm256_any_lt(rsq12,rcutoff2))
 382             {
 383
 384             r12              = _mm256_mul_pd(rsq12,rinv12);
 385
 386             /* EWALD ELECTROSTATICS */
 387
 388             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 389             ewrt             = _mm256_mul_pd(r12,ewtabscale);
 390             ewitab           = _mm256_cvttpd_epi32(ewrt);
 391             eweps            = _mm256_sub_pd(ewrt,_mm256_round_pd(ewrt, _MM_FROUND_FLOOR));
 392             ewitab           = _mm_slli_epi32(ewitab,2);
 393             ewtabF           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
 394             ewtabD           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,1) );
 395             ewtabV           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,2) );
 396             ewtabFn          = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,3) );
 397             GMX_MM256_FULLTRANSPOSE4_PD(ewtabF,ewtabD,ewtabV,ewtabFn);
 398             felec            = _mm256_add_pd(ewtabF,_mm256_mul_pd(eweps,ewtabD));
 399             velec            = _mm256_sub_pd(ewtabV,_mm256_mul_pd(_mm256_mul_pd(ewtabhalfspace,eweps),_mm256_add_pd(ewtabF,felec)));
 400             velec            = _mm256_mul_pd(qq12,_mm256_sub_pd(rinv12,velec));
 401             felec            = _mm256_mul_pd(_mm256_mul_pd(qq12,rinv12),_mm256_sub_pd(rinvsq12,felec));
 402
 403             d                = _mm256_sub_pd(r12,rswitch);
 404             d                = _mm256_max_pd(d,_mm256_setzero_pd());
 405             d2               = _mm256_mul_pd(d,d);
 406             sw               = _mm256_add_pd(one,_mm256_mul_pd(d2,_mm256_mul_pd(d,_mm256_add_pd(swV3,_mm256_mul_pd(d,_mm256_add_pd(swV4,_mm256_mul_pd(d,swV5)))))));
 407
 408             dsw              = _mm256_mul_pd(d2,_mm256_add_pd(swF2,_mm256_mul_pd(d,_mm256_add_pd(swF3,_mm256_mul_pd(d,swF4)))));
 409
 410             /* Evaluate switch function */
 411             /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
 412             felec            = _mm256_sub_pd( _mm256_mul_pd(felec,sw) , _mm256_mul_pd(rinv12,_mm256_mul_pd(velec,dsw)) );
 413             velec            = _mm256_mul_pd(velec,sw);
 414             cutoff_mask      = _mm256_cmp_pd(rsq12,rcutoff2,_CMP_LT_OQ);
 415
 416             /* Update potential sum for this i atom from the interaction with this j atom. */
 417             velec            = _mm256_and_pd(velec,cutoff_mask);
 418             velecsum         = _mm256_add_pd(velecsum,velec);
 419
 420             fscal            = felec;
 421
 422             fscal            = _mm256_and_pd(fscal,cutoff_mask);
 423
 424             /* Calculate temporary vectorial force */
 425             tx               = _mm256_mul_pd(fscal,dx12);
 426             ty               = _mm256_mul_pd(fscal,dy12);
 427             tz               = _mm256_mul_pd(fscal,dz12);
 428
 429             /* Update vectorial force */
 430             fix1             = _mm256_add_pd(fix1,tx);
 431             fiy1             = _mm256_add_pd(fiy1,ty);
 432             fiz1             = _mm256_add_pd(fiz1,tz);
 433
 434             fjx2             = _mm256_add_pd(fjx2,tx);
 435             fjy2             = _mm256_add_pd(fjy2,ty);
 436             fjz2             = _mm256_add_pd(fjz2,tz);
 437
 438             }
 439
 440             /**************************
 441              * CALCULATE INTERACTIONS *
 442              **************************/
 443
 444             if (gmx_mm256_any_lt(rsq13,rcutoff2))
 445             {
 446
 447             r13              = _mm256_mul_pd(rsq13,rinv13);
 448
 449             /* EWALD ELECTROSTATICS */
 450
 451             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 452             ewrt             = _mm256_mul_pd(r13,ewtabscale);
 453             ewitab           = _mm256_cvttpd_epi32(ewrt);
 454             eweps            = _mm256_sub_pd(ewrt,_mm256_round_pd(ewrt, _MM_FROUND_FLOOR));
 455             ewitab           = _mm_slli_epi32(ewitab,2);
 456             ewtabF           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
 457             ewtabD           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,1) );
 458             ewtabV           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,2) );
 459             ewtabFn          = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,3) );
 460             GMX_MM256_FULLTRANSPOSE4_PD(ewtabF,ewtabD,ewtabV,ewtabFn);
 461             felec            = _mm256_add_pd(ewtabF,_mm256_mul_pd(eweps,ewtabD));
 462             velec            = _mm256_sub_pd(ewtabV,_mm256_mul_pd(_mm256_mul_pd(ewtabhalfspace,eweps),_mm256_add_pd(ewtabF,felec)));
 463             velec            = _mm256_mul_pd(qq13,_mm256_sub_pd(rinv13,velec));
 464             felec            = _mm256_mul_pd(_mm256_mul_pd(qq13,rinv13),_mm256_sub_pd(rinvsq13,felec));
 465
 466             d                = _mm256_sub_pd(r13,rswitch);
 467             d                = _mm256_max_pd(d,_mm256_setzero_pd());
 468             d2               = _mm256_mul_pd(d,d);
 469             sw               = _mm256_add_pd(one,_mm256_mul_pd(d2,_mm256_mul_pd(d,_mm256_add_pd(swV3,_mm256_mul_pd(d,_mm256_add_pd(swV4,_mm256_mul_pd(d,swV5)))))));
 470
 471             dsw              = _mm256_mul_pd(d2,_mm256_add_pd(swF2,_mm256_mul_pd(d,_mm256_add_pd(swF3,_mm256_mul_pd(d,swF4)))));
 472
 473             /* Evaluate switch function */
 474             /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
 475             felec            = _mm256_sub_pd( _mm256_mul_pd(felec,sw) , _mm256_mul_pd(rinv13,_mm256_mul_pd(velec,dsw)) );
 476             velec            = _mm256_mul_pd(velec,sw);
 477             cutoff_mask      = _mm256_cmp_pd(rsq13,rcutoff2,_CMP_LT_OQ);
 478
 479             /* Update potential sum for this i atom from the interaction with this j atom. */
 480             velec            = _mm256_and_pd(velec,cutoff_mask);
 481             velecsum         = _mm256_add_pd(velecsum,velec);
 482
 483             fscal            = felec;
 484
 485             fscal            = _mm256_and_pd(fscal,cutoff_mask);
 486
 487             /* Calculate temporary vectorial force */
 488             tx               = _mm256_mul_pd(fscal,dx13);
 489             ty               = _mm256_mul_pd(fscal,dy13);
 490             tz               = _mm256_mul_pd(fscal,dz13);
 491
 492             /* Update vectorial force */
 493             fix1             = _mm256_add_pd(fix1,tx);
 494             fiy1             = _mm256_add_pd(fiy1,ty);
 495             fiz1             = _mm256_add_pd(fiz1,tz);
 496
 497             fjx3             = _mm256_add_pd(fjx3,tx);
 498             fjy3             = _mm256_add_pd(fjy3,ty);
 499             fjz3             = _mm256_add_pd(fjz3,tz);
 500
 501             }
 502
 503             /**************************
 504              * CALCULATE INTERACTIONS *
 505              **************************/
 506
 507             if (gmx_mm256_any_lt(rsq21,rcutoff2))
 508             {
 509
 510             r21              = _mm256_mul_pd(rsq21,rinv21);
 511
 512             /* EWALD ELECTROSTATICS */
 513
 514             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 515             ewrt             = _mm256_mul_pd(r21,ewtabscale);
 516             ewitab           = _mm256_cvttpd_epi32(ewrt);
 517             eweps            = _mm256_sub_pd(ewrt,_mm256_round_pd(ewrt, _MM_FROUND_FLOOR));
 518             ewitab           = _mm_slli_epi32(ewitab,2);
 519             ewtabF           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
 520             ewtabD           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,1) );
 521             ewtabV           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,2) );
 522             ewtabFn          = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,3) );
 523             GMX_MM256_FULLTRANSPOSE4_PD(ewtabF,ewtabD,ewtabV,ewtabFn);
 524             felec            = _mm256_add_pd(ewtabF,_mm256_mul_pd(eweps,ewtabD));
 525             velec            = _mm256_sub_pd(ewtabV,_mm256_mul_pd(_mm256_mul_pd(ewtabhalfspace,eweps),_mm256_add_pd(ewtabF,felec)));
 526             velec            = _mm256_mul_pd(qq21,_mm256_sub_pd(rinv21,velec));
 527             felec            = _mm256_mul_pd(_mm256_mul_pd(qq21,rinv21),_mm256_sub_pd(rinvsq21,felec));
 528
 529             d                = _mm256_sub_pd(r21,rswitch);
 530             d                = _mm256_max_pd(d,_mm256_setzero_pd());
 531             d2               = _mm256_mul_pd(d,d);
 532             sw               = _mm256_add_pd(one,_mm256_mul_pd(d2,_mm256_mul_pd(d,_mm256_add_pd(swV3,_mm256_mul_pd(d,_mm256_add_pd(swV4,_mm256_mul_pd(d,swV5)))))));
 533
 534             dsw              = _mm256_mul_pd(d2,_mm256_add_pd(swF2,_mm256_mul_pd(d,_mm256_add_pd(swF3,_mm256_mul_pd(d,swF4)))));
 535
 536             /* Evaluate switch function */
 537             /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
 538             felec            = _mm256_sub_pd( _mm256_mul_pd(felec,sw) , _mm256_mul_pd(rinv21,_mm256_mul_pd(velec,dsw)) );
 539             velec            = _mm256_mul_pd(velec,sw);
 540             cutoff_mask      = _mm256_cmp_pd(rsq21,rcutoff2,_CMP_LT_OQ);
 541
 542             /* Update potential sum for this i atom from the interaction with this j atom. */
 543             velec            = _mm256_and_pd(velec,cutoff_mask);
 544             velecsum         = _mm256_add_pd(velecsum,velec);
 545
 546             fscal            = felec;
 547
 548             fscal            = _mm256_and_pd(fscal,cutoff_mask);
 549
 550             /* Calculate temporary vectorial force */
 551             tx               = _mm256_mul_pd(fscal,dx21);
 552             ty               = _mm256_mul_pd(fscal,dy21);
 553             tz               = _mm256_mul_pd(fscal,dz21);
 554
 555             /* Update vectorial force */
 556             fix2             = _mm256_add_pd(fix2,tx);
 557             fiy2             = _mm256_add_pd(fiy2,ty);
 558             fiz2             = _mm256_add_pd(fiz2,tz);
 559
 560             fjx1             = _mm256_add_pd(fjx1,tx);
 561             fjy1             = _mm256_add_pd(fjy1,ty);
 562             fjz1             = _mm256_add_pd(fjz1,tz);
 563
 564             }
 565
 566             /**************************
 567              * CALCULATE INTERACTIONS *
 568              **************************/
 569
 570             if (gmx_mm256_any_lt(rsq22,rcutoff2))
 571             {
 572
 573             r22              = _mm256_mul_pd(rsq22,rinv22);
 574
 575             /* EWALD ELECTROSTATICS */
 576
 577             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 578             ewrt             = _mm256_mul_pd(r22,ewtabscale);
 579             ewitab           = _mm256_cvttpd_epi32(ewrt);
 580             eweps            = _mm256_sub_pd(ewrt,_mm256_round_pd(ewrt, _MM_FROUND_FLOOR));
 581             ewitab           = _mm_slli_epi32(ewitab,2);
 582             ewtabF           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
 583             ewtabD           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,1) );
 584             ewtabV           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,2) );
 585             ewtabFn          = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,3) );
 586             GMX_MM256_FULLTRANSPOSE4_PD(ewtabF,ewtabD,ewtabV,ewtabFn);
 587             felec            = _mm256_add_pd(ewtabF,_mm256_mul_pd(eweps,ewtabD));
 588             velec            = _mm256_sub_pd(ewtabV,_mm256_mul_pd(_mm256_mul_pd(ewtabhalfspace,eweps),_mm256_add_pd(ewtabF,felec)));
 589             velec            = _mm256_mul_pd(qq22,_mm256_sub_pd(rinv22,velec));
 590             felec            = _mm256_mul_pd(_mm256_mul_pd(qq22,rinv22),_mm256_sub_pd(rinvsq22,felec));
 591
 592             d                = _mm256_sub_pd(r22,rswitch);
 593             d                = _mm256_max_pd(d,_mm256_setzero_pd());
 594             d2               = _mm256_mul_pd(d,d);
 595             sw               = _mm256_add_pd(one,_mm256_mul_pd(d2,_mm256_mul_pd(d,_mm256_add_pd(swV3,_mm256_mul_pd(d,_mm256_add_pd(swV4,_mm256_mul_pd(d,swV5)))))));
 596
 597             dsw              = _mm256_mul_pd(d2,_mm256_add_pd(swF2,_mm256_mul_pd(d,_mm256_add_pd(swF3,_mm256_mul_pd(d,swF4)))));
 598
 599             /* Evaluate switch function */
 600             /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
 601             felec            = _mm256_sub_pd( _mm256_mul_pd(felec,sw) , _mm256_mul_pd(rinv22,_mm256_mul_pd(velec,dsw)) );
 602             velec            = _mm256_mul_pd(velec,sw);
 603             cutoff_mask      = _mm256_cmp_pd(rsq22,rcutoff2,_CMP_LT_OQ);
 604
 605             /* Update potential sum for this i atom from the interaction with this j atom. */
 606             velec            = _mm256_and_pd(velec,cutoff_mask);
 607             velecsum         = _mm256_add_pd(velecsum,velec);
 608
 609             fscal            = felec;
 610
 611             fscal            = _mm256_and_pd(fscal,cutoff_mask);
 612
 613             /* Calculate temporary vectorial force */
 614             tx               = _mm256_mul_pd(fscal,dx22);
 615             ty               = _mm256_mul_pd(fscal,dy22);
 616             tz               = _mm256_mul_pd(fscal,dz22);
 617
 618             /* Update vectorial force */
 619             fix2             = _mm256_add_pd(fix2,tx);
 620             fiy2             = _mm256_add_pd(fiy2,ty);
 621             fiz2             = _mm256_add_pd(fiz2,tz);
 622
 623             fjx2             = _mm256_add_pd(fjx2,tx);
 624             fjy2             = _mm256_add_pd(fjy2,ty);
 625             fjz2             = _mm256_add_pd(fjz2,tz);
 626
 627             }
 628
 629             /**************************
 630              * CALCULATE INTERACTIONS *
 631              **************************/
 632
 633             if (gmx_mm256_any_lt(rsq23,rcutoff2))
 634             {
 635
 636             r23              = _mm256_mul_pd(rsq23,rinv23);
 637
 638             /* EWALD ELECTROSTATICS */
 639
 640             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 641             ewrt             = _mm256_mul_pd(r23,ewtabscale);
 642             ewitab           = _mm256_cvttpd_epi32(ewrt);
 643             eweps            = _mm256_sub_pd(ewrt,_mm256_round_pd(ewrt, _MM_FROUND_FLOOR));
 644             ewitab           = _mm_slli_epi32(ewitab,2);
 645             ewtabF           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
 646             ewtabD           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,1) );
 647             ewtabV           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,2) );
 648             ewtabFn          = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,3) );
 649             GMX_MM256_FULLTRANSPOSE4_PD(ewtabF,ewtabD,ewtabV,ewtabFn);
 650             felec            = _mm256_add_pd(ewtabF,_mm256_mul_pd(eweps,ewtabD));
 651             velec            = _mm256_sub_pd(ewtabV,_mm256_mul_pd(_mm256_mul_pd(ewtabhalfspace,eweps),_mm256_add_pd(ewtabF,felec)));
 652             velec            = _mm256_mul_pd(qq23,_mm256_sub_pd(rinv23,velec));
 653             felec            = _mm256_mul_pd(_mm256_mul_pd(qq23,rinv23),_mm256_sub_pd(rinvsq23,felec));
 654
 655             d                = _mm256_sub_pd(r23,rswitch);
 656             d                = _mm256_max_pd(d,_mm256_setzero_pd());
 657             d2               = _mm256_mul_pd(d,d);
 658             sw               = _mm256_add_pd(one,_mm256_mul_pd(d2,_mm256_mul_pd(d,_mm256_add_pd(swV3,_mm256_mul_pd(d,_mm256_add_pd(swV4,_mm256_mul_pd(d,swV5)))))));
 659
 660             dsw              = _mm256_mul_pd(d2,_mm256_add_pd(swF2,_mm256_mul_pd(d,_mm256_add_pd(swF3,_mm256_mul_pd(d,swF4)))));
 661
 662             /* Evaluate switch function */
 663             /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
 664             felec            = _mm256_sub_pd( _mm256_mul_pd(felec,sw) , _mm256_mul_pd(rinv23,_mm256_mul_pd(velec,dsw)) );
 665             velec            = _mm256_mul_pd(velec,sw);
 666             cutoff_mask      = _mm256_cmp_pd(rsq23,rcutoff2,_CMP_LT_OQ);
 667
 668             /* Update potential sum for this i atom from the interaction with this j atom. */
 669             velec            = _mm256_and_pd(velec,cutoff_mask);
 670             velecsum         = _mm256_add_pd(velecsum,velec);
 671
 672             fscal            = felec;
 673
 674             fscal            = _mm256_and_pd(fscal,cutoff_mask);
 675
 676             /* Calculate temporary vectorial force */
 677             tx               = _mm256_mul_pd(fscal,dx23);
 678             ty               = _mm256_mul_pd(fscal,dy23);
 679             tz               = _mm256_mul_pd(fscal,dz23);
 680
 681             /* Update vectorial force */
 682             fix2             = _mm256_add_pd(fix2,tx);
 683             fiy2             = _mm256_add_pd(fiy2,ty);
 684             fiz2             = _mm256_add_pd(fiz2,tz);
 685
 686             fjx3             = _mm256_add_pd(fjx3,tx);
 687             fjy3             = _mm256_add_pd(fjy3,ty);
 688             fjz3             = _mm256_add_pd(fjz3,tz);
 689
 690             }
 691
 692             /**************************
 693              * CALCULATE INTERACTIONS *
 694              **************************/
 695
 696             if (gmx_mm256_any_lt(rsq31,rcutoff2))
 697             {
 698
 699             r31              = _mm256_mul_pd(rsq31,rinv31);
 700
 701             /* EWALD ELECTROSTATICS */
 702
 703             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 704             ewrt             = _mm256_mul_pd(r31,ewtabscale);
 705             ewitab           = _mm256_cvttpd_epi32(ewrt);
 706             eweps            = _mm256_sub_pd(ewrt,_mm256_round_pd(ewrt, _MM_FROUND_FLOOR));
 707             ewitab           = _mm_slli_epi32(ewitab,2);
 708             ewtabF           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
 709             ewtabD           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,1) );
 710             ewtabV           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,2) );
 711             ewtabFn          = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,3) );
 712             GMX_MM256_FULLTRANSPOSE4_PD(ewtabF,ewtabD,ewtabV,ewtabFn);
 713             felec            = _mm256_add_pd(ewtabF,_mm256_mul_pd(eweps,ewtabD));
 714             velec            = _mm256_sub_pd(ewtabV,_mm256_mul_pd(_mm256_mul_pd(ewtabhalfspace,eweps),_mm256_add_pd(ewtabF,felec)));
 715             velec            = _mm256_mul_pd(qq31,_mm256_sub_pd(rinv31,velec));
 716             felec            = _mm256_mul_pd(_mm256_mul_pd(qq31,rinv31),_mm256_sub_pd(rinvsq31,felec));
 717
 718             d                = _mm256_sub_pd(r31,rswitch);
 719             d                = _mm256_max_pd(d,_mm256_setzero_pd());
 720             d2               = _mm256_mul_pd(d,d);
 721             sw               = _mm256_add_pd(one,_mm256_mul_pd(d2,_mm256_mul_pd(d,_mm256_add_pd(swV3,_mm256_mul_pd(d,_mm256_add_pd(swV4,_mm256_mul_pd(d,swV5)))))));
 722
 723             dsw              = _mm256_mul_pd(d2,_mm256_add_pd(swF2,_mm256_mul_pd(d,_mm256_add_pd(swF3,_mm256_mul_pd(d,swF4)))));
 724
 725             /* Evaluate switch function */
 726             /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
 727             felec            = _mm256_sub_pd( _mm256_mul_pd(felec,sw) , _mm256_mul_pd(rinv31,_mm256_mul_pd(velec,dsw)) );
 728             velec            = _mm256_mul_pd(velec,sw);
 729             cutoff_mask      = _mm256_cmp_pd(rsq31,rcutoff2,_CMP_LT_OQ);
 730
 731             /* Update potential sum for this i atom from the interaction with this j atom. */
 732             velec            = _mm256_and_pd(velec,cutoff_mask);
 733             velecsum         = _mm256_add_pd(velecsum,velec);
 734
 735             fscal            = felec;
 736
 737             fscal            = _mm256_and_pd(fscal,cutoff_mask);
 738
 739             /* Calculate temporary vectorial force */
 740             tx               = _mm256_mul_pd(fscal,dx31);
 741             ty               = _mm256_mul_pd(fscal,dy31);
 742             tz               = _mm256_mul_pd(fscal,dz31);
 743
 744             /* Update vectorial force */
 745             fix3             = _mm256_add_pd(fix3,tx);
 746             fiy3             = _mm256_add_pd(fiy3,ty);
 747             fiz3             = _mm256_add_pd(fiz3,tz);
 748
 749             fjx1             = _mm256_add_pd(fjx1,tx);
 750             fjy1             = _mm256_add_pd(fjy1,ty);
 751             fjz1             = _mm256_add_pd(fjz1,tz);
 752
 753             }
 754
 755             /**************************
 756              * CALCULATE INTERACTIONS *
 757              **************************/
 758
 759             if (gmx_mm256_any_lt(rsq32,rcutoff2))
 760             {
 761
 762             r32              = _mm256_mul_pd(rsq32,rinv32);
 763
 764             /* EWALD ELECTROSTATICS */
 765
 766             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 767             ewrt             = _mm256_mul_pd(r32,ewtabscale);
 768             ewitab           = _mm256_cvttpd_epi32(ewrt);
 769             eweps            = _mm256_sub_pd(ewrt,_mm256_round_pd(ewrt, _MM_FROUND_FLOOR));
 770             ewitab           = _mm_slli_epi32(ewitab,2);
 771             ewtabF           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
 772             ewtabD           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,1) );
 773             ewtabV           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,2) );
 774             ewtabFn          = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,3) );
 775             GMX_MM256_FULLTRANSPOSE4_PD(ewtabF,ewtabD,ewtabV,ewtabFn);
 776             felec            = _mm256_add_pd(ewtabF,_mm256_mul_pd(eweps,ewtabD));
 777             velec            = _mm256_sub_pd(ewtabV,_mm256_mul_pd(_mm256_mul_pd(ewtabhalfspace,eweps),_mm256_add_pd(ewtabF,felec)));
 778             velec            = _mm256_mul_pd(qq32,_mm256_sub_pd(rinv32,velec));
 779             felec            = _mm256_mul_pd(_mm256_mul_pd(qq32,rinv32),_mm256_sub_pd(rinvsq32,felec));
 780
 781             d                = _mm256_sub_pd(r32,rswitch);
 782             d                = _mm256_max_pd(d,_mm256_setzero_pd());
 783             d2               = _mm256_mul_pd(d,d);
 784             sw               = _mm256_add_pd(one,_mm256_mul_pd(d2,_mm256_mul_pd(d,_mm256_add_pd(swV3,_mm256_mul_pd(d,_mm256_add_pd(swV4,_mm256_mul_pd(d,swV5)))))));
 785
 786             dsw              = _mm256_mul_pd(d2,_mm256_add_pd(swF2,_mm256_mul_pd(d,_mm256_add_pd(swF3,_mm256_mul_pd(d,swF4)))));
 787
 788             /* Evaluate switch function */
 789             /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
 790             felec            = _mm256_sub_pd( _mm256_mul_pd(felec,sw) , _mm256_mul_pd(rinv32,_mm256_mul_pd(velec,dsw)) );
 791             velec            = _mm256_mul_pd(velec,sw);
 792             cutoff_mask      = _mm256_cmp_pd(rsq32,rcutoff2,_CMP_LT_OQ);
 793
 794             /* Update potential sum for this i atom from the interaction with this j atom. */
 795             velec            = _mm256_and_pd(velec,cutoff_mask);
 796             velecsum         = _mm256_add_pd(velecsum,velec);
 797
 798             fscal            = felec;
 799
 800             fscal            = _mm256_and_pd(fscal,cutoff_mask);
 801
 802             /* Calculate temporary vectorial force */
 803             tx               = _mm256_mul_pd(fscal,dx32);
 804             ty               = _mm256_mul_pd(fscal,dy32);
 805             tz               = _mm256_mul_pd(fscal,dz32);
 806
 807             /* Update vectorial force */
 808             fix3             = _mm256_add_pd(fix3,tx);
 809             fiy3             = _mm256_add_pd(fiy3,ty);
 810             fiz3             = _mm256_add_pd(fiz3,tz);
 811
 812             fjx2             = _mm256_add_pd(fjx2,tx);
 813             fjy2             = _mm256_add_pd(fjy2,ty);
 814             fjz2             = _mm256_add_pd(fjz2,tz);
 815
 816             }
 817
 818             /**************************
 819              * CALCULATE INTERACTIONS *
 820              **************************/
 821
 822             if (gmx_mm256_any_lt(rsq33,rcutoff2))
 823             {
 824
 825             r33              = _mm256_mul_pd(rsq33,rinv33);
 826
 827             /* EWALD ELECTROSTATICS */
 828
 829             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 830             ewrt             = _mm256_mul_pd(r33,ewtabscale);
 831             ewitab           = _mm256_cvttpd_epi32(ewrt);
 832             eweps            = _mm256_sub_pd(ewrt,_mm256_round_pd(ewrt, _MM_FROUND_FLOOR));
 833             ewitab           = _mm_slli_epi32(ewitab,2);
 834             ewtabF           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
 835             ewtabD           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,1) );
 836             ewtabV           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,2) );
 837             ewtabFn          = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,3) );
 838             GMX_MM256_FULLTRANSPOSE4_PD(ewtabF,ewtabD,ewtabV,ewtabFn);
 839             felec            = _mm256_add_pd(ewtabF,_mm256_mul_pd(eweps,ewtabD));
 840             velec            = _mm256_sub_pd(ewtabV,_mm256_mul_pd(_mm256_mul_pd(ewtabhalfspace,eweps),_mm256_add_pd(ewtabF,felec)));
 841             velec            = _mm256_mul_pd(qq33,_mm256_sub_pd(rinv33,velec));
 842             felec            = _mm256_mul_pd(_mm256_mul_pd(qq33,rinv33),_mm256_sub_pd(rinvsq33,felec));
 843
 844             d                = _mm256_sub_pd(r33,rswitch);
 845             d                = _mm256_max_pd(d,_mm256_setzero_pd());
 846             d2               = _mm256_mul_pd(d,d);
 847             sw               = _mm256_add_pd(one,_mm256_mul_pd(d2,_mm256_mul_pd(d,_mm256_add_pd(swV3,_mm256_mul_pd(d,_mm256_add_pd(swV4,_mm256_mul_pd(d,swV5)))))));
 848
 849             dsw              = _mm256_mul_pd(d2,_mm256_add_pd(swF2,_mm256_mul_pd(d,_mm256_add_pd(swF3,_mm256_mul_pd(d,swF4)))));
 850
 851             /* Evaluate switch function */
 852             /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
 853             felec            = _mm256_sub_pd( _mm256_mul_pd(felec,sw) , _mm256_mul_pd(rinv33,_mm256_mul_pd(velec,dsw)) );
 854             velec            = _mm256_mul_pd(velec,sw);
 855             cutoff_mask      = _mm256_cmp_pd(rsq33,rcutoff2,_CMP_LT_OQ);
 856
 857             /* Update potential sum for this i atom from the interaction with this j atom. */
 858             velec            = _mm256_and_pd(velec,cutoff_mask);
 859             velecsum         = _mm256_add_pd(velecsum,velec);
 860
 861             fscal            = felec;
 862
 863             fscal            = _mm256_and_pd(fscal,cutoff_mask);
 864
 865             /* Calculate temporary vectorial force */
 866             tx               = _mm256_mul_pd(fscal,dx33);
 867             ty               = _mm256_mul_pd(fscal,dy33);
 868             tz               = _mm256_mul_pd(fscal,dz33);
 869
 870             /* Update vectorial force */
 871             fix3             = _mm256_add_pd(fix3,tx);
 872             fiy3             = _mm256_add_pd(fiy3,ty);
 873             fiz3             = _mm256_add_pd(fiz3,tz);
 874
 875             fjx3             = _mm256_add_pd(fjx3,tx);
 876             fjy3             = _mm256_add_pd(fjy3,ty);
 877             fjz3             = _mm256_add_pd(fjz3,tz);
 878
 879             }
 880
 881             fjptrA             = f+j_coord_offsetA;
 882             fjptrB             = f+j_coord_offsetB;
 883             fjptrC             = f+j_coord_offsetC;
 884             fjptrD             = f+j_coord_offsetD;
 885
 886             gmx_mm256_decrement_3rvec_4ptr_swizzle_pd(fjptrA+DIM,fjptrB+DIM,fjptrC+DIM,fjptrD+DIM,
 887                                                       fjx1,fjy1,fjz1,fjx2,fjy2,fjz2,fjx3,fjy3,fjz3);
 888
 889             /* Inner loop uses 585 flops */
 890         }
 891
 892         if(jidx<j_index_end)
 893         {
 894
 895             /* Get j neighbor index, and coordinate index */
 896             jnrlistA         = jjnr[jidx];
 897             jnrlistB         = jjnr[jidx+1];
 898             jnrlistC         = jjnr[jidx+2];
 899             jnrlistD         = jjnr[jidx+3];
 900             /* Sign of each element will be negative for non-real atoms.
 901              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 902              * so use it as val = _mm_andnot_pd(mask,val) to clear dummy entries.
 903              */
 904             tmpmask0 = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
 905
 906             tmpmask1 = _mm_permute_ps(tmpmask0,_GMX_MM_PERMUTE(3,3,2,2));
 907             tmpmask0 = _mm_permute_ps(tmpmask0,_GMX_MM_PERMUTE(1,1,0,0));
 908             dummy_mask = _mm256_castps_pd(gmx_mm256_set_m128(tmpmask1,tmpmask0));
 909
 910             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
 911             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
 912             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
 913             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
 914             j_coord_offsetA  = DIM*jnrA;
 915             j_coord_offsetB  = DIM*jnrB;
 916             j_coord_offsetC  = DIM*jnrC;
 917             j_coord_offsetD  = DIM*jnrD;
 918
 919             /* load j atom coordinates */
 920             gmx_mm256_load_3rvec_4ptr_swizzle_pd(x+j_coord_offsetA+DIM,x+j_coord_offsetB+DIM,
 921                                                  x+j_coord_offsetC+DIM,x+j_coord_offsetD+DIM,
 922                                                  &jx1,&jy1,&jz1,&jx2,&jy2,&jz2,&jx3,&jy3,&jz3);
 923
 924             /* Calculate displacement vector */
 925             dx11             = _mm256_sub_pd(ix1,jx1);
 926             dy11             = _mm256_sub_pd(iy1,jy1);
 927             dz11             = _mm256_sub_pd(iz1,jz1);
 928             dx12             = _mm256_sub_pd(ix1,jx2);
 929             dy12             = _mm256_sub_pd(iy1,jy2);
 930             dz12             = _mm256_sub_pd(iz1,jz2);
 931             dx13             = _mm256_sub_pd(ix1,jx3);
 932             dy13             = _mm256_sub_pd(iy1,jy3);
 933             dz13             = _mm256_sub_pd(iz1,jz3);
 934             dx21             = _mm256_sub_pd(ix2,jx1);
 935             dy21             = _mm256_sub_pd(iy2,jy1);
 936             dz21             = _mm256_sub_pd(iz2,jz1);
 937             dx22             = _mm256_sub_pd(ix2,jx2);
 938             dy22             = _mm256_sub_pd(iy2,jy2);
 939             dz22             = _mm256_sub_pd(iz2,jz2);
 940             dx23             = _mm256_sub_pd(ix2,jx3);
 941             dy23             = _mm256_sub_pd(iy2,jy3);
 942             dz23             = _mm256_sub_pd(iz2,jz3);
 943             dx31             = _mm256_sub_pd(ix3,jx1);
 944             dy31             = _mm256_sub_pd(iy3,jy1);
 945             dz31             = _mm256_sub_pd(iz3,jz1);
 946             dx32             = _mm256_sub_pd(ix3,jx2);
 947             dy32             = _mm256_sub_pd(iy3,jy2);
 948             dz32             = _mm256_sub_pd(iz3,jz2);
 949             dx33             = _mm256_sub_pd(ix3,jx3);
 950             dy33             = _mm256_sub_pd(iy3,jy3);
 951             dz33             = _mm256_sub_pd(iz3,jz3);
 952
 953             /* Calculate squared distance and things based on it */
 954             rsq11            = gmx_mm256_calc_rsq_pd(dx11,dy11,dz11);
 955             rsq12            = gmx_mm256_calc_rsq_pd(dx12,dy12,dz12);
 956             rsq13            = gmx_mm256_calc_rsq_pd(dx13,dy13,dz13);
 957             rsq21            = gmx_mm256_calc_rsq_pd(dx21,dy21,dz21);
 958             rsq22            = gmx_mm256_calc_rsq_pd(dx22,dy22,dz22);
 959             rsq23            = gmx_mm256_calc_rsq_pd(dx23,dy23,dz23);
 960             rsq31            = gmx_mm256_calc_rsq_pd(dx31,dy31,dz31);
 961             rsq32            = gmx_mm256_calc_rsq_pd(dx32,dy32,dz32);
 962             rsq33            = gmx_mm256_calc_rsq_pd(dx33,dy33,dz33);
 963
 964             rinv11           = gmx_mm256_invsqrt_pd(rsq11);
 965             rinv12           = gmx_mm256_invsqrt_pd(rsq12);
 966             rinv13           = gmx_mm256_invsqrt_pd(rsq13);
 967             rinv21           = gmx_mm256_invsqrt_pd(rsq21);
 968             rinv22           = gmx_mm256_invsqrt_pd(rsq22);
 969             rinv23           = gmx_mm256_invsqrt_pd(rsq23);
 970             rinv31           = gmx_mm256_invsqrt_pd(rsq31);
 971             rinv32           = gmx_mm256_invsqrt_pd(rsq32);
 972             rinv33           = gmx_mm256_invsqrt_pd(rsq33);
 973
 974             rinvsq11         = _mm256_mul_pd(rinv11,rinv11);
 975             rinvsq12         = _mm256_mul_pd(rinv12,rinv12);
 976             rinvsq13         = _mm256_mul_pd(rinv13,rinv13);
 977             rinvsq21         = _mm256_mul_pd(rinv21,rinv21);
 978             rinvsq22         = _mm256_mul_pd(rinv22,rinv22);
 979             rinvsq23         = _mm256_mul_pd(rinv23,rinv23);
 980             rinvsq31         = _mm256_mul_pd(rinv31,rinv31);
 981             rinvsq32         = _mm256_mul_pd(rinv32,rinv32);
 982             rinvsq33         = _mm256_mul_pd(rinv33,rinv33);
 983
 984             fjx1             = _mm256_setzero_pd();
 985             fjy1             = _mm256_setzero_pd();
 986             fjz1             = _mm256_setzero_pd();
 987             fjx2             = _mm256_setzero_pd();
 988             fjy2             = _mm256_setzero_pd();
 989             fjz2             = _mm256_setzero_pd();
 990             fjx3             = _mm256_setzero_pd();
 991             fjy3             = _mm256_setzero_pd();
 992             fjz3             = _mm256_setzero_pd();
 993
 994             /**************************
 995              * CALCULATE INTERACTIONS *
 996              **************************/
 997
 998             if (gmx_mm256_any_lt(rsq11,rcutoff2))
 999             {
1000
1001             r11              = _mm256_mul_pd(rsq11,rinv11);
1002             r11              = _mm256_andnot_pd(dummy_mask,r11);
1003
1004             /* EWALD ELECTROSTATICS */
1005
1006             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1007             ewrt             = _mm256_mul_pd(r11,ewtabscale);
1008             ewitab           = _mm256_cvttpd_epi32(ewrt);
1009             eweps            = _mm256_sub_pd(ewrt,_mm256_round_pd(ewrt, _MM_FROUND_FLOOR));
1010             ewitab           = _mm_slli_epi32(ewitab,2);
1011             ewtabF           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
1012             ewtabD           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,1) );
1013             ewtabV           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,2) );
1014             ewtabFn          = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,3) );
1015             GMX_MM256_FULLTRANSPOSE4_PD(ewtabF,ewtabD,ewtabV,ewtabFn);
1016             felec            = _mm256_add_pd(ewtabF,_mm256_mul_pd(eweps,ewtabD));
1017             velec            = _mm256_sub_pd(ewtabV,_mm256_mul_pd(_mm256_mul_pd(ewtabhalfspace,eweps),_mm256_add_pd(ewtabF,felec)));
1018             velec            = _mm256_mul_pd(qq11,_mm256_sub_pd(rinv11,velec));
1019             felec            = _mm256_mul_pd(_mm256_mul_pd(qq11,rinv11),_mm256_sub_pd(rinvsq11,felec));
1020
1021             d                = _mm256_sub_pd(r11,rswitch);
1022             d                = _mm256_max_pd(d,_mm256_setzero_pd());
1023             d2               = _mm256_mul_pd(d,d);
1024             sw               = _mm256_add_pd(one,_mm256_mul_pd(d2,_mm256_mul_pd(d,_mm256_add_pd(swV3,_mm256_mul_pd(d,_mm256_add_pd(swV4,_mm256_mul_pd(d,swV5)))))));
1025
1026             dsw              = _mm256_mul_pd(d2,_mm256_add_pd(swF2,_mm256_mul_pd(d,_mm256_add_pd(swF3,_mm256_mul_pd(d,swF4)))));
1027
1028             /* Evaluate switch function */
1029             /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
1030             felec            = _mm256_sub_pd( _mm256_mul_pd(felec,sw) , _mm256_mul_pd(rinv11,_mm256_mul_pd(velec,dsw)) );
1031             velec            = _mm256_mul_pd(velec,sw);
1032             cutoff_mask      = _mm256_cmp_pd(rsq11,rcutoff2,_CMP_LT_OQ);
1033
1034             /* Update potential sum for this i atom from the interaction with this j atom. */
1035             velec            = _mm256_and_pd(velec,cutoff_mask);
1036             velec            = _mm256_andnot_pd(dummy_mask,velec);
1037             velecsum         = _mm256_add_pd(velecsum,velec);
1038
1039             fscal            = felec;
1040
1041             fscal            = _mm256_and_pd(fscal,cutoff_mask);
1042
1043             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
1044
1045             /* Calculate temporary vectorial force */
1046             tx               = _mm256_mul_pd(fscal,dx11);
1047             ty               = _mm256_mul_pd(fscal,dy11);
1048             tz               = _mm256_mul_pd(fscal,dz11);
1049
1050             /* Update vectorial force */
1051             fix1             = _mm256_add_pd(fix1,tx);
1052             fiy1             = _mm256_add_pd(fiy1,ty);
1053             fiz1             = _mm256_add_pd(fiz1,tz);
1054
1055             fjx1             = _mm256_add_pd(fjx1,tx);
1056             fjy1             = _mm256_add_pd(fjy1,ty);
1057             fjz1             = _mm256_add_pd(fjz1,tz);
1058
1059             }
1060
1061             /**************************
1062              * CALCULATE INTERACTIONS *
1063              **************************/
1064
1065             if (gmx_mm256_any_lt(rsq12,rcutoff2))
1066             {
1067
1068             r12              = _mm256_mul_pd(rsq12,rinv12);
1069             r12              = _mm256_andnot_pd(dummy_mask,r12);
1070
1071             /* EWALD ELECTROSTATICS */
1072
1073             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1074             ewrt             = _mm256_mul_pd(r12,ewtabscale);
1075             ewitab           = _mm256_cvttpd_epi32(ewrt);
1076             eweps            = _mm256_sub_pd(ewrt,_mm256_round_pd(ewrt, _MM_FROUND_FLOOR));
1077             ewitab           = _mm_slli_epi32(ewitab,2);
1078             ewtabF           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
1079             ewtabD           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,1) );
1080             ewtabV           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,2) );
1081             ewtabFn          = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,3) );
1082             GMX_MM256_FULLTRANSPOSE4_PD(ewtabF,ewtabD,ewtabV,ewtabFn);
1083             felec            = _mm256_add_pd(ewtabF,_mm256_mul_pd(eweps,ewtabD));
1084             velec            = _mm256_sub_pd(ewtabV,_mm256_mul_pd(_mm256_mul_pd(ewtabhalfspace,eweps),_mm256_add_pd(ewtabF,felec)));
1085             velec            = _mm256_mul_pd(qq12,_mm256_sub_pd(rinv12,velec));
1086             felec            = _mm256_mul_pd(_mm256_mul_pd(qq12,rinv12),_mm256_sub_pd(rinvsq12,felec));
1087
1088             d                = _mm256_sub_pd(r12,rswitch);
1089             d                = _mm256_max_pd(d,_mm256_setzero_pd());
1090             d2               = _mm256_mul_pd(d,d);
1091             sw               = _mm256_add_pd(one,_mm256_mul_pd(d2,_mm256_mul_pd(d,_mm256_add_pd(swV3,_mm256_mul_pd(d,_mm256_add_pd(swV4,_mm256_mul_pd(d,swV5)))))));
1092
1093             dsw              = _mm256_mul_pd(d2,_mm256_add_pd(swF2,_mm256_mul_pd(d,_mm256_add_pd(swF3,_mm256_mul_pd(d,swF4)))));
1094
1095             /* Evaluate switch function */
1096             /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
1097             felec            = _mm256_sub_pd( _mm256_mul_pd(felec,sw) , _mm256_mul_pd(rinv12,_mm256_mul_pd(velec,dsw)) );
1098             velec            = _mm256_mul_pd(velec,sw);
1099             cutoff_mask      = _mm256_cmp_pd(rsq12,rcutoff2,_CMP_LT_OQ);
1100
1101             /* Update potential sum for this i atom from the interaction with this j atom. */
1102             velec            = _mm256_and_pd(velec,cutoff_mask);
1103             velec            = _mm256_andnot_pd(dummy_mask,velec);
1104             velecsum         = _mm256_add_pd(velecsum,velec);
1105
1106             fscal            = felec;
1107
1108             fscal            = _mm256_and_pd(fscal,cutoff_mask);
1109
1110             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
1111
1112             /* Calculate temporary vectorial force */
1113             tx               = _mm256_mul_pd(fscal,dx12);
1114             ty               = _mm256_mul_pd(fscal,dy12);
1115             tz               = _mm256_mul_pd(fscal,dz12);
1116
1117             /* Update vectorial force */
1118             fix1             = _mm256_add_pd(fix1,tx);
1119             fiy1             = _mm256_add_pd(fiy1,ty);
1120             fiz1             = _mm256_add_pd(fiz1,tz);
1121
1122             fjx2             = _mm256_add_pd(fjx2,tx);
1123             fjy2             = _mm256_add_pd(fjy2,ty);
1124             fjz2             = _mm256_add_pd(fjz2,tz);
1125
1126             }
1127
1128             /**************************
1129              * CALCULATE INTERACTIONS *
1130              **************************/
1131
1132             if (gmx_mm256_any_lt(rsq13,rcutoff2))
1133             {
1134
1135             r13              = _mm256_mul_pd(rsq13,rinv13);
1136             r13              = _mm256_andnot_pd(dummy_mask,r13);
1137
1138             /* EWALD ELECTROSTATICS */
1139
1140             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1141             ewrt             = _mm256_mul_pd(r13,ewtabscale);
1142             ewitab           = _mm256_cvttpd_epi32(ewrt);
1143             eweps            = _mm256_sub_pd(ewrt,_mm256_round_pd(ewrt, _MM_FROUND_FLOOR));
1144             ewitab           = _mm_slli_epi32(ewitab,2);
1145             ewtabF           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
1146             ewtabD           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,1) );
1147             ewtabV           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,2) );
1148             ewtabFn          = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,3) );
1149             GMX_MM256_FULLTRANSPOSE4_PD(ewtabF,ewtabD,ewtabV,ewtabFn);
1150             felec            = _mm256_add_pd(ewtabF,_mm256_mul_pd(eweps,ewtabD));
1151             velec            = _mm256_sub_pd(ewtabV,_mm256_mul_pd(_mm256_mul_pd(ewtabhalfspace,eweps),_mm256_add_pd(ewtabF,felec)));
1152             velec            = _mm256_mul_pd(qq13,_mm256_sub_pd(rinv13,velec));
1153             felec            = _mm256_mul_pd(_mm256_mul_pd(qq13,rinv13),_mm256_sub_pd(rinvsq13,felec));
1154
1155             d                = _mm256_sub_pd(r13,rswitch);
1156             d                = _mm256_max_pd(d,_mm256_setzero_pd());
1157             d2               = _mm256_mul_pd(d,d);
1158             sw               = _mm256_add_pd(one,_mm256_mul_pd(d2,_mm256_mul_pd(d,_mm256_add_pd(swV3,_mm256_mul_pd(d,_mm256_add_pd(swV4,_mm256_mul_pd(d,swV5)))))));
1159
1160             dsw              = _mm256_mul_pd(d2,_mm256_add_pd(swF2,_mm256_mul_pd(d,_mm256_add_pd(swF3,_mm256_mul_pd(d,swF4)))));
1161
1162             /* Evaluate switch function */
1163             /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
1164             felec            = _mm256_sub_pd( _mm256_mul_pd(felec,sw) , _mm256_mul_pd(rinv13,_mm256_mul_pd(velec,dsw)) );
1165             velec            = _mm256_mul_pd(velec,sw);
1166             cutoff_mask      = _mm256_cmp_pd(rsq13,rcutoff2,_CMP_LT_OQ);
1167
1168             /* Update potential sum for this i atom from the interaction with this j atom. */
1169             velec            = _mm256_and_pd(velec,cutoff_mask);
1170             velec            = _mm256_andnot_pd(dummy_mask,velec);
1171             velecsum         = _mm256_add_pd(velecsum,velec);
1172
1173             fscal            = felec;
1174
1175             fscal            = _mm256_and_pd(fscal,cutoff_mask);
1176
1177             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
1178
1179             /* Calculate temporary vectorial force */
1180             tx               = _mm256_mul_pd(fscal,dx13);
1181             ty               = _mm256_mul_pd(fscal,dy13);
1182             tz               = _mm256_mul_pd(fscal,dz13);
1183
1184             /* Update vectorial force */
1185             fix1             = _mm256_add_pd(fix1,tx);
1186             fiy1             = _mm256_add_pd(fiy1,ty);
1187             fiz1             = _mm256_add_pd(fiz1,tz);
1188
1189             fjx3             = _mm256_add_pd(fjx3,tx);
1190             fjy3             = _mm256_add_pd(fjy3,ty);
1191             fjz3             = _mm256_add_pd(fjz3,tz);
1192
1193             }
1194
1195             /**************************
1196              * CALCULATE INTERACTIONS *
1197              **************************/
1198
1199             if (gmx_mm256_any_lt(rsq21,rcutoff2))
1200             {
1201
1202             r21              = _mm256_mul_pd(rsq21,rinv21);
1203             r21              = _mm256_andnot_pd(dummy_mask,r21);
1204
1205             /* EWALD ELECTROSTATICS */
1206
1207             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1208             ewrt             = _mm256_mul_pd(r21,ewtabscale);
1209             ewitab           = _mm256_cvttpd_epi32(ewrt);
1210             eweps            = _mm256_sub_pd(ewrt,_mm256_round_pd(ewrt, _MM_FROUND_FLOOR));
1211             ewitab           = _mm_slli_epi32(ewitab,2);
1212             ewtabF           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
1213             ewtabD           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,1) );
1214             ewtabV           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,2) );
1215             ewtabFn          = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,3) );
1216             GMX_MM256_FULLTRANSPOSE4_PD(ewtabF,ewtabD,ewtabV,ewtabFn);
1217             felec            = _mm256_add_pd(ewtabF,_mm256_mul_pd(eweps,ewtabD));
1218             velec            = _mm256_sub_pd(ewtabV,_mm256_mul_pd(_mm256_mul_pd(ewtabhalfspace,eweps),_mm256_add_pd(ewtabF,felec)));
1219             velec            = _mm256_mul_pd(qq21,_mm256_sub_pd(rinv21,velec));
1220             felec            = _mm256_mul_pd(_mm256_mul_pd(qq21,rinv21),_mm256_sub_pd(rinvsq21,felec));
1221
1222             d                = _mm256_sub_pd(r21,rswitch);
1223             d                = _mm256_max_pd(d,_mm256_setzero_pd());
1224             d2               = _mm256_mul_pd(d,d);
1225             sw               = _mm256_add_pd(one,_mm256_mul_pd(d2,_mm256_mul_pd(d,_mm256_add_pd(swV3,_mm256_mul_pd(d,_mm256_add_pd(swV4,_mm256_mul_pd(d,swV5)))))));
1226
1227             dsw              = _mm256_mul_pd(d2,_mm256_add_pd(swF2,_mm256_mul_pd(d,_mm256_add_pd(swF3,_mm256_mul_pd(d,swF4)))));
1228
1229             /* Evaluate switch function */
1230             /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
1231             felec            = _mm256_sub_pd( _mm256_mul_pd(felec,sw) , _mm256_mul_pd(rinv21,_mm256_mul_pd(velec,dsw)) );
1232             velec            = _mm256_mul_pd(velec,sw);
1233             cutoff_mask      = _mm256_cmp_pd(rsq21,rcutoff2,_CMP_LT_OQ);
1234
1235             /* Update potential sum for this i atom from the interaction with this j atom. */
1236             velec            = _mm256_and_pd(velec,cutoff_mask);
1237             velec            = _mm256_andnot_pd(dummy_mask,velec);
1238             velecsum         = _mm256_add_pd(velecsum,velec);
1239
1240             fscal            = felec;
1241
1242             fscal            = _mm256_and_pd(fscal,cutoff_mask);
1243
1244             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
1245
1246             /* Calculate temporary vectorial force */
1247             tx               = _mm256_mul_pd(fscal,dx21);
1248             ty               = _mm256_mul_pd(fscal,dy21);
1249             tz               = _mm256_mul_pd(fscal,dz21);
1250
1251             /* Update vectorial force */
1252             fix2             = _mm256_add_pd(fix2,tx);
1253             fiy2             = _mm256_add_pd(fiy2,ty);
1254             fiz2             = _mm256_add_pd(fiz2,tz);
1255
1256             fjx1             = _mm256_add_pd(fjx1,tx);
1257             fjy1             = _mm256_add_pd(fjy1,ty);
1258             fjz1             = _mm256_add_pd(fjz1,tz);
1259
1260             }
1261
1262             /**************************
1263              * CALCULATE INTERACTIONS *
1264              **************************/
1265
1266             if (gmx_mm256_any_lt(rsq22,rcutoff2))
1267             {
1268
1269             r22              = _mm256_mul_pd(rsq22,rinv22);
1270             r22              = _mm256_andnot_pd(dummy_mask,r22);
1271
1272             /* EWALD ELECTROSTATICS */
1273
1274             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1275             ewrt             = _mm256_mul_pd(r22,ewtabscale);
1276             ewitab           = _mm256_cvttpd_epi32(ewrt);
1277             eweps            = _mm256_sub_pd(ewrt,_mm256_round_pd(ewrt, _MM_FROUND_FLOOR));
1278             ewitab           = _mm_slli_epi32(ewitab,2);
1279             ewtabF           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
1280             ewtabD           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,1) );
1281             ewtabV           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,2) );
1282             ewtabFn          = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,3) );
1283             GMX_MM256_FULLTRANSPOSE4_PD(ewtabF,ewtabD,ewtabV,ewtabFn);
1284             felec            = _mm256_add_pd(ewtabF,_mm256_mul_pd(eweps,ewtabD));
1285             velec            = _mm256_sub_pd(ewtabV,_mm256_mul_pd(_mm256_mul_pd(ewtabhalfspace,eweps),_mm256_add_pd(ewtabF,felec)));
1286             velec            = _mm256_mul_pd(qq22,_mm256_sub_pd(rinv22,velec));
1287             felec            = _mm256_mul_pd(_mm256_mul_pd(qq22,rinv22),_mm256_sub_pd(rinvsq22,felec));
1288
1289             d                = _mm256_sub_pd(r22,rswitch);
1290             d                = _mm256_max_pd(d,_mm256_setzero_pd());
1291             d2               = _mm256_mul_pd(d,d);
1292             sw               = _mm256_add_pd(one,_mm256_mul_pd(d2,_mm256_mul_pd(d,_mm256_add_pd(swV3,_mm256_mul_pd(d,_mm256_add_pd(swV4,_mm256_mul_pd(d,swV5)))))));
1293
1294             dsw              = _mm256_mul_pd(d2,_mm256_add_pd(swF2,_mm256_mul_pd(d,_mm256_add_pd(swF3,_mm256_mul_pd(d,swF4)))));
1295
1296             /* Evaluate switch function */
1297             /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
1298             felec            = _mm256_sub_pd( _mm256_mul_pd(felec,sw) , _mm256_mul_pd(rinv22,_mm256_mul_pd(velec,dsw)) );
1299             velec            = _mm256_mul_pd(velec,sw);
1300             cutoff_mask      = _mm256_cmp_pd(rsq22,rcutoff2,_CMP_LT_OQ);
1301
1302             /* Update potential sum for this i atom from the interaction with this j atom. */
1303             velec            = _mm256_and_pd(velec,cutoff_mask);
1304             velec            = _mm256_andnot_pd(dummy_mask,velec);
1305             velecsum         = _mm256_add_pd(velecsum,velec);
1306
1307             fscal            = felec;
1308
1309             fscal            = _mm256_and_pd(fscal,cutoff_mask);
1310
1311             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
1312
1313             /* Calculate temporary vectorial force */
1314             tx               = _mm256_mul_pd(fscal,dx22);
1315             ty               = _mm256_mul_pd(fscal,dy22);
1316             tz               = _mm256_mul_pd(fscal,dz22);
1317
1318             /* Update vectorial force */
1319             fix2             = _mm256_add_pd(fix2,tx);
1320             fiy2             = _mm256_add_pd(fiy2,ty);
1321             fiz2             = _mm256_add_pd(fiz2,tz);
1322
1323             fjx2             = _mm256_add_pd(fjx2,tx);
1324             fjy2             = _mm256_add_pd(fjy2,ty);
1325             fjz2             = _mm256_add_pd(fjz2,tz);
1326
1327             }
1328
1329             /**************************
1330              * CALCULATE INTERACTIONS *
1331              **************************/
1332
1333             if (gmx_mm256_any_lt(rsq23,rcutoff2))
1334             {
1335
1336             r23              = _mm256_mul_pd(rsq23,rinv23);
1337             r23              = _mm256_andnot_pd(dummy_mask,r23);
1338
1339             /* EWALD ELECTROSTATICS */
1340
1341             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1342             ewrt             = _mm256_mul_pd(r23,ewtabscale);
1343             ewitab           = _mm256_cvttpd_epi32(ewrt);
1344             eweps            = _mm256_sub_pd(ewrt,_mm256_round_pd(ewrt, _MM_FROUND_FLOOR));
1345             ewitab           = _mm_slli_epi32(ewitab,2);
1346             ewtabF           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
1347             ewtabD           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,1) );
1348             ewtabV           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,2) );
1349             ewtabFn          = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,3) );
1350             GMX_MM256_FULLTRANSPOSE4_PD(ewtabF,ewtabD,ewtabV,ewtabFn);
1351             felec            = _mm256_add_pd(ewtabF,_mm256_mul_pd(eweps,ewtabD));
1352             velec            = _mm256_sub_pd(ewtabV,_mm256_mul_pd(_mm256_mul_pd(ewtabhalfspace,eweps),_mm256_add_pd(ewtabF,felec)));
1353             velec            = _mm256_mul_pd(qq23,_mm256_sub_pd(rinv23,velec));
1354             felec            = _mm256_mul_pd(_mm256_mul_pd(qq23,rinv23),_mm256_sub_pd(rinvsq23,felec));
1355
1356             d                = _mm256_sub_pd(r23,rswitch);
1357             d                = _mm256_max_pd(d,_mm256_setzero_pd());
1358             d2               = _mm256_mul_pd(d,d);
1359             sw               = _mm256_add_pd(one,_mm256_mul_pd(d2,_mm256_mul_pd(d,_mm256_add_pd(swV3,_mm256_mul_pd(d,_mm256_add_pd(swV4,_mm256_mul_pd(d,swV5)))))));
1360
1361             dsw              = _mm256_mul_pd(d2,_mm256_add_pd(swF2,_mm256_mul_pd(d,_mm256_add_pd(swF3,_mm256_mul_pd(d,swF4)))));
1362
1363             /* Evaluate switch function */
1364             /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
1365             felec            = _mm256_sub_pd( _mm256_mul_pd(felec,sw) , _mm256_mul_pd(rinv23,_mm256_mul_pd(velec,dsw)) );
1366             velec            = _mm256_mul_pd(velec,sw);
1367             cutoff_mask      = _mm256_cmp_pd(rsq23,rcutoff2,_CMP_LT_OQ);
1368
1369             /* Update potential sum for this i atom from the interaction with this j atom. */
1370             velec            = _mm256_and_pd(velec,cutoff_mask);
1371             velec            = _mm256_andnot_pd(dummy_mask,velec);
1372             velecsum         = _mm256_add_pd(velecsum,velec);
1373
1374             fscal            = felec;
1375
1376             fscal            = _mm256_and_pd(fscal,cutoff_mask);
1377
1378             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
1379
1380             /* Calculate temporary vectorial force */
1381             tx               = _mm256_mul_pd(fscal,dx23);
1382             ty               = _mm256_mul_pd(fscal,dy23);
1383             tz               = _mm256_mul_pd(fscal,dz23);
1384
1385             /* Update vectorial force */
1386             fix2             = _mm256_add_pd(fix2,tx);
1387             fiy2             = _mm256_add_pd(fiy2,ty);
1388             fiz2             = _mm256_add_pd(fiz2,tz);
1389
1390             fjx3             = _mm256_add_pd(fjx3,tx);
1391             fjy3             = _mm256_add_pd(fjy3,ty);
1392             fjz3             = _mm256_add_pd(fjz3,tz);
1393
1394             }
1395
1396             /**************************
1397              * CALCULATE INTERACTIONS *
1398              **************************/
1399
1400             if (gmx_mm256_any_lt(rsq31,rcutoff2))
1401             {
1402
1403             r31              = _mm256_mul_pd(rsq31,rinv31);
1404             r31              = _mm256_andnot_pd(dummy_mask,r31);
1405
1406             /* EWALD ELECTROSTATICS */
1407
1408             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1409             ewrt             = _mm256_mul_pd(r31,ewtabscale);
1410             ewitab           = _mm256_cvttpd_epi32(ewrt);
1411             eweps            = _mm256_sub_pd(ewrt,_mm256_round_pd(ewrt, _MM_FROUND_FLOOR));
1412             ewitab           = _mm_slli_epi32(ewitab,2);
1413             ewtabF           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
1414             ewtabD           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,1) );
1415             ewtabV           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,2) );
1416             ewtabFn          = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,3) );
1417             GMX_MM256_FULLTRANSPOSE4_PD(ewtabF,ewtabD,ewtabV,ewtabFn);
1418             felec            = _mm256_add_pd(ewtabF,_mm256_mul_pd(eweps,ewtabD));
1419             velec            = _mm256_sub_pd(ewtabV,_mm256_mul_pd(_mm256_mul_pd(ewtabhalfspace,eweps),_mm256_add_pd(ewtabF,felec)));
1420             velec            = _mm256_mul_pd(qq31,_mm256_sub_pd(rinv31,velec));
1421             felec            = _mm256_mul_pd(_mm256_mul_pd(qq31,rinv31),_mm256_sub_pd(rinvsq31,felec));
1422
1423             d                = _mm256_sub_pd(r31,rswitch);
1424             d                = _mm256_max_pd(d,_mm256_setzero_pd());
1425             d2               = _mm256_mul_pd(d,d);
1426             sw               = _mm256_add_pd(one,_mm256_mul_pd(d2,_mm256_mul_pd(d,_mm256_add_pd(swV3,_mm256_mul_pd(d,_mm256_add_pd(swV4,_mm256_mul_pd(d,swV5)))))));
1427
1428             dsw              = _mm256_mul_pd(d2,_mm256_add_pd(swF2,_mm256_mul_pd(d,_mm256_add_pd(swF3,_mm256_mul_pd(d,swF4)))));
1429
1430             /* Evaluate switch function */
1431             /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
1432             felec            = _mm256_sub_pd( _mm256_mul_pd(felec,sw) , _mm256_mul_pd(rinv31,_mm256_mul_pd(velec,dsw)) );
1433             velec            = _mm256_mul_pd(velec,sw);
1434             cutoff_mask      = _mm256_cmp_pd(rsq31,rcutoff2,_CMP_LT_OQ);
1435
1436             /* Update potential sum for this i atom from the interaction with this j atom. */
1437             velec            = _mm256_and_pd(velec,cutoff_mask);
1438             velec            = _mm256_andnot_pd(dummy_mask,velec);
1439             velecsum         = _mm256_add_pd(velecsum,velec);
1440
1441             fscal            = felec;
1442
1443             fscal            = _mm256_and_pd(fscal,cutoff_mask);
1444
1445             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
1446
1447             /* Calculate temporary vectorial force */
1448             tx               = _mm256_mul_pd(fscal,dx31);
1449             ty               = _mm256_mul_pd(fscal,dy31);
1450             tz               = _mm256_mul_pd(fscal,dz31);
1451
1452             /* Update vectorial force */
1453             fix3             = _mm256_add_pd(fix3,tx);
1454             fiy3             = _mm256_add_pd(fiy3,ty);
1455             fiz3             = _mm256_add_pd(fiz3,tz);
1456
1457             fjx1             = _mm256_add_pd(fjx1,tx);
1458             fjy1             = _mm256_add_pd(fjy1,ty);
1459             fjz1             = _mm256_add_pd(fjz1,tz);
1460
1461             }
1462
1463             /**************************
1464              * CALCULATE INTERACTIONS *
1465              **************************/
1466
1467             if (gmx_mm256_any_lt(rsq32,rcutoff2))
1468             {
1469
1470             r32              = _mm256_mul_pd(rsq32,rinv32);
1471             r32              = _mm256_andnot_pd(dummy_mask,r32);
1472
1473             /* EWALD ELECTROSTATICS */
1474
1475             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1476             ewrt             = _mm256_mul_pd(r32,ewtabscale);
1477             ewitab           = _mm256_cvttpd_epi32(ewrt);
1478             eweps            = _mm256_sub_pd(ewrt,_mm256_round_pd(ewrt, _MM_FROUND_FLOOR));
1479             ewitab           = _mm_slli_epi32(ewitab,2);
1480             ewtabF           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
1481             ewtabD           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,1) );
1482             ewtabV           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,2) );
1483             ewtabFn          = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,3) );
1484             GMX_MM256_FULLTRANSPOSE4_PD(ewtabF,ewtabD,ewtabV,ewtabFn);
1485             felec            = _mm256_add_pd(ewtabF,_mm256_mul_pd(eweps,ewtabD));
1486             velec            = _mm256_sub_pd(ewtabV,_mm256_mul_pd(_mm256_mul_pd(ewtabhalfspace,eweps),_mm256_add_pd(ewtabF,felec)));
1487             velec            = _mm256_mul_pd(qq32,_mm256_sub_pd(rinv32,velec));
1488             felec            = _mm256_mul_pd(_mm256_mul_pd(qq32,rinv32),_mm256_sub_pd(rinvsq32,felec));
1489
1490             d                = _mm256_sub_pd(r32,rswitch);
1491             d                = _mm256_max_pd(d,_mm256_setzero_pd());
1492             d2               = _mm256_mul_pd(d,d);
1493             sw               = _mm256_add_pd(one,_mm256_mul_pd(d2,_mm256_mul_pd(d,_mm256_add_pd(swV3,_mm256_mul_pd(d,_mm256_add_pd(swV4,_mm256_mul_pd(d,swV5)))))));
1494
1495             dsw              = _mm256_mul_pd(d2,_mm256_add_pd(swF2,_mm256_mul_pd(d,_mm256_add_pd(swF3,_mm256_mul_pd(d,swF4)))));
1496
1497             /* Evaluate switch function */
1498             /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
1499             felec            = _mm256_sub_pd( _mm256_mul_pd(felec,sw) , _mm256_mul_pd(rinv32,_mm256_mul_pd(velec,dsw)) );
1500             velec            = _mm256_mul_pd(velec,sw);
1501             cutoff_mask      = _mm256_cmp_pd(rsq32,rcutoff2,_CMP_LT_OQ);
1502
1503             /* Update potential sum for this i atom from the interaction with this j atom. */
1504             velec            = _mm256_and_pd(velec,cutoff_mask);
1505             velec            = _mm256_andnot_pd(dummy_mask,velec);
1506             velecsum         = _mm256_add_pd(velecsum,velec);
1507
1508             fscal            = felec;
1509
1510             fscal            = _mm256_and_pd(fscal,cutoff_mask);
1511
1512             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
1513
1514             /* Calculate temporary vectorial force */
1515             tx               = _mm256_mul_pd(fscal,dx32);
1516             ty               = _mm256_mul_pd(fscal,dy32);
1517             tz               = _mm256_mul_pd(fscal,dz32);
1518
1519             /* Update vectorial force */
1520             fix3             = _mm256_add_pd(fix3,tx);
1521             fiy3             = _mm256_add_pd(fiy3,ty);
1522             fiz3             = _mm256_add_pd(fiz3,tz);
1523
1524             fjx2             = _mm256_add_pd(fjx2,tx);
1525             fjy2             = _mm256_add_pd(fjy2,ty);
1526             fjz2             = _mm256_add_pd(fjz2,tz);
1527
1528             }
1529
1530             /**************************
1531              * CALCULATE INTERACTIONS *
1532              **************************/
1533
1534             if (gmx_mm256_any_lt(rsq33,rcutoff2))
1535             {
1536
1537             r33              = _mm256_mul_pd(rsq33,rinv33);
1538             r33              = _mm256_andnot_pd(dummy_mask,r33);
1539
1540             /* EWALD ELECTROSTATICS */
1541
1542             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1543             ewrt             = _mm256_mul_pd(r33,ewtabscale);
1544             ewitab           = _mm256_cvttpd_epi32(ewrt);
1545             eweps            = _mm256_sub_pd(ewrt,_mm256_round_pd(ewrt, _MM_FROUND_FLOOR));
1546             ewitab           = _mm_slli_epi32(ewitab,2);
1547             ewtabF           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
1548             ewtabD           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,1) );
1549             ewtabV           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,2) );
1550             ewtabFn          = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,3) );
1551             GMX_MM256_FULLTRANSPOSE4_PD(ewtabF,ewtabD,ewtabV,ewtabFn);
1552             felec            = _mm256_add_pd(ewtabF,_mm256_mul_pd(eweps,ewtabD));
1553             velec            = _mm256_sub_pd(ewtabV,_mm256_mul_pd(_mm256_mul_pd(ewtabhalfspace,eweps),_mm256_add_pd(ewtabF,felec)));
1554             velec            = _mm256_mul_pd(qq33,_mm256_sub_pd(rinv33,velec));
1555             felec            = _mm256_mul_pd(_mm256_mul_pd(qq33,rinv33),_mm256_sub_pd(rinvsq33,felec));
1556
1557             d                = _mm256_sub_pd(r33,rswitch);
1558             d                = _mm256_max_pd(d,_mm256_setzero_pd());
1559             d2               = _mm256_mul_pd(d,d);
1560             sw               = _mm256_add_pd(one,_mm256_mul_pd(d2,_mm256_mul_pd(d,_mm256_add_pd(swV3,_mm256_mul_pd(d,_mm256_add_pd(swV4,_mm256_mul_pd(d,swV5)))))));
1561
1562             dsw              = _mm256_mul_pd(d2,_mm256_add_pd(swF2,_mm256_mul_pd(d,_mm256_add_pd(swF3,_mm256_mul_pd(d,swF4)))));
1563
1564             /* Evaluate switch function */
1565             /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
1566             felec            = _mm256_sub_pd( _mm256_mul_pd(felec,sw) , _mm256_mul_pd(rinv33,_mm256_mul_pd(velec,dsw)) );
1567             velec            = _mm256_mul_pd(velec,sw);
1568             cutoff_mask      = _mm256_cmp_pd(rsq33,rcutoff2,_CMP_LT_OQ);
1569
1570             /* Update potential sum for this i atom from the interaction with this j atom. */
1571             velec            = _mm256_and_pd(velec,cutoff_mask);
1572             velec            = _mm256_andnot_pd(dummy_mask,velec);
1573             velecsum         = _mm256_add_pd(velecsum,velec);
1574
1575             fscal            = felec;
1576
1577             fscal            = _mm256_and_pd(fscal,cutoff_mask);
1578
1579             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
1580
1581             /* Calculate temporary vectorial force */
1582             tx               = _mm256_mul_pd(fscal,dx33);
1583             ty               = _mm256_mul_pd(fscal,dy33);
1584             tz               = _mm256_mul_pd(fscal,dz33);
1585
1586             /* Update vectorial force */
1587             fix3             = _mm256_add_pd(fix3,tx);
1588             fiy3             = _mm256_add_pd(fiy3,ty);
1589             fiz3             = _mm256_add_pd(fiz3,tz);
1590
1591             fjx3             = _mm256_add_pd(fjx3,tx);
1592             fjy3             = _mm256_add_pd(fjy3,ty);
1593             fjz3             = _mm256_add_pd(fjz3,tz);
1594
1595             }
1596
1597             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
1598             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
1599             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
1600             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
1601
1602             gmx_mm256_decrement_3rvec_4ptr_swizzle_pd(fjptrA+DIM,fjptrB+DIM,fjptrC+DIM,fjptrD+DIM,
1603                                                       fjx1,fjy1,fjz1,fjx2,fjy2,fjz2,fjx3,fjy3,fjz3);
1604
1605             /* Inner loop uses 594 flops */
1606         }
1607
1608         /* End of innermost loop */
1609
1610         gmx_mm256_update_iforce_3atom_swizzle_pd(fix1,fiy1,fiz1,fix2,fiy2,fiz2,fix3,fiy3,fiz3,
1611                                                  f+i_coord_offset+DIM,fshift+i_shift_offset);
1612
1613         ggid                        = gid[iidx];
1614         /* Update potential energies */
1615         gmx_mm256_update_1pot_pd(velecsum,kernel_data->energygrp_elec+ggid);
1616
1617         /* Increment number of inner iterations */
1618         inneriter                  += j_index_end - j_index_start;
1619
1620         /* Outer loop uses 19 flops */
1621     }
1622
1623     /* Increment number of outer iterations */
1624     outeriter        += nri;
1625
1626     /* Update outer/inner flops */
1627
1628     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_W4W4_VF,outeriter*19 + inneriter*594);
1629 }
1630 /*
1631  * Gromacs nonbonded kernel:   nb_kernel_ElecEwSw_VdwNone_GeomW4W4_F_avx_256_double
1632  * Electrostatics interaction: Ewald
1633  * VdW interaction:            None
1634  * Geometry:                   Water4-Water4
1635  * Calculate force/pot:        Force
1636  */
1637 void
1638 nb_kernel_ElecEwSw_VdwNone_GeomW4W4_F_avx_256_double
1639                     (t_nblist                    * gmx_restrict       nlist,
1640                      rvec                        * gmx_restrict          xx,
1641                      rvec                        * gmx_restrict          ff,
1642                      t_forcerec                  * gmx_restrict          fr,
1643                      t_mdatoms                   * gmx_restrict     mdatoms,
1644                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
1645                      t_nrnb                      * gmx_restrict        nrnb)
1646 {
1647     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
1648      * just 0 for non-waters.
1649      * Suffixes A,B,C,D refer to j loop unrolling done with AVX, e.g. for the four different
1650      * jnr indices corresponding to data put in the four positions in the SIMD register.
1651      */
1652     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
1653     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
1654     int              jnrA,jnrB,jnrC,jnrD;
1655     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
1656     int              jnrlistE,jnrlistF,jnrlistG,jnrlistH;
1657     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
1658     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
1659     real             rcutoff_scalar;
1660     real             *shiftvec,*fshift,*x,*f;
1661     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD;
1662     real             scratch[4*DIM];
1663     __m256d          tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
1664     real *           vdwioffsetptr1;
1665     __m256d          ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
1666     real *           vdwioffsetptr2;
1667     __m256d          ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
1668     real *           vdwioffsetptr3;
1669     __m256d          ix3,iy3,iz3,fix3,fiy3,fiz3,iq3,isai3;
1670     int              vdwjidx1A,vdwjidx1B,vdwjidx1C,vdwjidx1D;
1671     __m256d          jx1,jy1,jz1,fjx1,fjy1,fjz1,jq1,isaj1;
1672     int              vdwjidx2A,vdwjidx2B,vdwjidx2C,vdwjidx2D;
1673     __m256d          jx2,jy2,jz2,fjx2,fjy2,fjz2,jq2,isaj2;
1674     int              vdwjidx3A,vdwjidx3B,vdwjidx3C,vdwjidx3D;
1675     __m256d          jx3,jy3,jz3,fjx3,fjy3,fjz3,jq3,isaj3;
1676     __m256d          dx11,dy11,dz11,rsq11,rinv11,rinvsq11,r11,qq11,c6_11,c12_11;
1677     __m256d          dx12,dy12,dz12,rsq12,rinv12,rinvsq12,r12,qq12,c6_12,c12_12;
1678     __m256d          dx13,dy13,dz13,rsq13,rinv13,rinvsq13,r13,qq13,c6_13,c12_13;
1679     __m256d          dx21,dy21,dz21,rsq21,rinv21,rinvsq21,r21,qq21,c6_21,c12_21;
1680     __m256d          dx22,dy22,dz22,rsq22,rinv22,rinvsq22,r22,qq22,c6_22,c12_22;
1681     __m256d          dx23,dy23,dz23,rsq23,rinv23,rinvsq23,r23,qq23,c6_23,c12_23;
1682     __m256d          dx31,dy31,dz31,rsq31,rinv31,rinvsq31,r31,qq31,c6_31,c12_31;
1683     __m256d          dx32,dy32,dz32,rsq32,rinv32,rinvsq32,r32,qq32,c6_32,c12_32;
1684     __m256d          dx33,dy33,dz33,rsq33,rinv33,rinvsq33,r33,qq33,c6_33,c12_33;
1685     __m256d          velec,felec,velecsum,facel,crf,krf,krf2;
1686     real             *charge;
1687     __m128i          ewitab;
1688     __m256d          ewtabscale,eweps,sh_ewald,ewrt,ewtabhalfspace,ewtabF,ewtabFn,ewtabD,ewtabV;
1689     __m256d          beta,beta2,beta3,zeta2,pmecorrF,pmecorrV,rinv3;
1690     real             *ewtab;
1691     __m256d          rswitch,swV3,swV4,swV5,swF2,swF3,swF4,d,d2,sw,dsw;
1692     real             rswitch_scalar,d_scalar;
1693     __m256d          dummy_mask,cutoff_mask;
1694     __m128           tmpmask0,tmpmask1;
1695     __m256d          signbit = _mm256_castsi256_pd( _mm256_set1_epi32(0x80000000) );
1696     __m256d          one     = _mm256_set1_pd(1.0);
1697     __m256d          two     = _mm256_set1_pd(2.0);
1698     x                = xx[0];
1699     f                = ff[0];
1700
1701     nri              = nlist->nri;
1702     iinr             = nlist->iinr;
1703     jindex           = nlist->jindex;
1704     jjnr             = nlist->jjnr;
1705     shiftidx         = nlist->shift;
1706     gid              = nlist->gid;
1707     shiftvec         = fr->shift_vec[0];
1708     fshift           = fr->fshift[0];
1709     facel            = _mm256_set1_pd(fr->epsfac);
1710     charge           = mdatoms->chargeA;
1711
1712     sh_ewald         = _mm256_set1_pd(fr->ic->sh_ewald);
1713     beta             = _mm256_set1_pd(fr->ic->ewaldcoeff_q);
1714     beta2            = _mm256_mul_pd(beta,beta);
1715     beta3            = _mm256_mul_pd(beta,beta2);
1716
1717     ewtab            = fr->ic->tabq_coul_FDV0;
1718     ewtabscale       = _mm256_set1_pd(fr->ic->tabq_scale);
1719     ewtabhalfspace   = _mm256_set1_pd(0.5/fr->ic->tabq_scale);
1720
1721     /* Setup water-specific parameters */
1722     inr              = nlist->iinr[0];
1723     iq1              = _mm256_mul_pd(facel,_mm256_set1_pd(charge[inr+1]));
1724     iq2              = _mm256_mul_pd(facel,_mm256_set1_pd(charge[inr+2]));
1725     iq3              = _mm256_mul_pd(facel,_mm256_set1_pd(charge[inr+3]));
1726
1727     jq1              = _mm256_set1_pd(charge[inr+1]);
1728     jq2              = _mm256_set1_pd(charge[inr+2]);
1729     jq3              = _mm256_set1_pd(charge[inr+3]);
1730     qq11             = _mm256_mul_pd(iq1,jq1);
1731     qq12             = _mm256_mul_pd(iq1,jq2);
1732     qq13             = _mm256_mul_pd(iq1,jq3);
1733     qq21             = _mm256_mul_pd(iq2,jq1);
1734     qq22             = _mm256_mul_pd(iq2,jq2);
1735     qq23             = _mm256_mul_pd(iq2,jq3);
1736     qq31             = _mm256_mul_pd(iq3,jq1);
1737     qq32             = _mm256_mul_pd(iq3,jq2);
1738     qq33             = _mm256_mul_pd(iq3,jq3);
1739
1740     /* When we use explicit cutoffs the value must be identical for elec and VdW, so use elec as an arbitrary choice */
1741     rcutoff_scalar   = fr->rcoulomb;
1742     rcutoff          = _mm256_set1_pd(rcutoff_scalar);
1743     rcutoff2         = _mm256_mul_pd(rcutoff,rcutoff);
1744
1745     rswitch_scalar   = fr->rcoulomb_switch;
1746     rswitch          = _mm256_set1_pd(rswitch_scalar);
1747     /* Setup switch parameters */
1748     d_scalar         = rcutoff_scalar-rswitch_scalar;
1749     d                = _mm256_set1_pd(d_scalar);
1750     swV3             = _mm256_set1_pd(-10.0/(d_scalar*d_scalar*d_scalar));
1751     swV4             = _mm256_set1_pd( 15.0/(d_scalar*d_scalar*d_scalar*d_scalar));
1752     swV5             = _mm256_set1_pd( -6.0/(d_scalar*d_scalar*d_scalar*d_scalar*d_scalar));
1753     swF2             = _mm256_set1_pd(-30.0/(d_scalar*d_scalar*d_scalar));
1754     swF3             = _mm256_set1_pd( 60.0/(d_scalar*d_scalar*d_scalar*d_scalar));
1755     swF4             = _mm256_set1_pd(-30.0/(d_scalar*d_scalar*d_scalar*d_scalar*d_scalar));
1756
1757     /* Avoid stupid compiler warnings */
1758     jnrA = jnrB = jnrC = jnrD = 0;
1759     j_coord_offsetA = 0;
1760     j_coord_offsetB = 0;
1761     j_coord_offsetC = 0;
1762     j_coord_offsetD = 0;
1763
1764     outeriter        = 0;
1765     inneriter        = 0;
1766
1767     for(iidx=0;iidx<4*DIM;iidx++)
1768     {
1769         scratch[iidx] = 0.0;
1770     }
1771
1772     /* Start outer loop over neighborlists */
1773     for(iidx=0; iidx<nri; iidx++)
1774     {
1775         /* Load shift vector for this list */
1776         i_shift_offset   = DIM*shiftidx[iidx];
1777
1778         /* Load limits for loop over neighbors */
1779         j_index_start    = jindex[iidx];
1780         j_index_end      = jindex[iidx+1];
1781
1782         /* Get outer coordinate index */
1783         inr              = iinr[iidx];
1784         i_coord_offset   = DIM*inr;
1785
1786         /* Load i particle coords and add shift vector */
1787         gmx_mm256_load_shift_and_3rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset+DIM,
1788                                                     &ix1,&iy1,&iz1,&ix2,&iy2,&iz2,&ix3,&iy3,&iz3);
1789
1790         fix1             = _mm256_setzero_pd();
1791         fiy1             = _mm256_setzero_pd();
1792         fiz1             = _mm256_setzero_pd();
1793         fix2             = _mm256_setzero_pd();
1794         fiy2             = _mm256_setzero_pd();
1795         fiz2             = _mm256_setzero_pd();
1796         fix3             = _mm256_setzero_pd();
1797         fiy3             = _mm256_setzero_pd();
1798         fiz3             = _mm256_setzero_pd();
1799
1800         /* Start inner kernel loop */
1801         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
1802         {
1803
1804             /* Get j neighbor index, and coordinate index */
1805             jnrA             = jjnr[jidx];
1806             jnrB             = jjnr[jidx+1];
1807             jnrC             = jjnr[jidx+2];
1808             jnrD             = jjnr[jidx+3];
1809             j_coord_offsetA  = DIM*jnrA;
1810             j_coord_offsetB  = DIM*jnrB;
1811             j_coord_offsetC  = DIM*jnrC;
1812             j_coord_offsetD  = DIM*jnrD;
1813
1814             /* load j atom coordinates */
1815             gmx_mm256_load_3rvec_4ptr_swizzle_pd(x+j_coord_offsetA+DIM,x+j_coord_offsetB+DIM,
1816                                                  x+j_coord_offsetC+DIM,x+j_coord_offsetD+DIM,
1817                                                  &jx1,&jy1,&jz1,&jx2,&jy2,&jz2,&jx3,&jy3,&jz3);
1818
1819             /* Calculate displacement vector */
1820             dx11             = _mm256_sub_pd(ix1,jx1);
1821             dy11             = _mm256_sub_pd(iy1,jy1);
1822             dz11             = _mm256_sub_pd(iz1,jz1);
1823             dx12             = _mm256_sub_pd(ix1,jx2);
1824             dy12             = _mm256_sub_pd(iy1,jy2);
1825             dz12             = _mm256_sub_pd(iz1,jz2);
1826             dx13             = _mm256_sub_pd(ix1,jx3);
1827             dy13             = _mm256_sub_pd(iy1,jy3);
1828             dz13             = _mm256_sub_pd(iz1,jz3);
1829             dx21             = _mm256_sub_pd(ix2,jx1);
1830             dy21             = _mm256_sub_pd(iy2,jy1);
1831             dz21             = _mm256_sub_pd(iz2,jz1);
1832             dx22             = _mm256_sub_pd(ix2,jx2);
1833             dy22             = _mm256_sub_pd(iy2,jy2);
1834             dz22             = _mm256_sub_pd(iz2,jz2);
1835             dx23             = _mm256_sub_pd(ix2,jx3);
1836             dy23             = _mm256_sub_pd(iy2,jy3);
1837             dz23             = _mm256_sub_pd(iz2,jz3);
1838             dx31             = _mm256_sub_pd(ix3,jx1);
1839             dy31             = _mm256_sub_pd(iy3,jy1);
1840             dz31             = _mm256_sub_pd(iz3,jz1);
1841             dx32             = _mm256_sub_pd(ix3,jx2);
1842             dy32             = _mm256_sub_pd(iy3,jy2);
1843             dz32             = _mm256_sub_pd(iz3,jz2);
1844             dx33             = _mm256_sub_pd(ix3,jx3);
1845             dy33             = _mm256_sub_pd(iy3,jy3);
1846             dz33             = _mm256_sub_pd(iz3,jz3);
1847
1848             /* Calculate squared distance and things based on it */
1849             rsq11            = gmx_mm256_calc_rsq_pd(dx11,dy11,dz11);
1850             rsq12            = gmx_mm256_calc_rsq_pd(dx12,dy12,dz12);
1851             rsq13            = gmx_mm256_calc_rsq_pd(dx13,dy13,dz13);
1852             rsq21            = gmx_mm256_calc_rsq_pd(dx21,dy21,dz21);
1853             rsq22            = gmx_mm256_calc_rsq_pd(dx22,dy22,dz22);
1854             rsq23            = gmx_mm256_calc_rsq_pd(dx23,dy23,dz23);
1855             rsq31            = gmx_mm256_calc_rsq_pd(dx31,dy31,dz31);
1856             rsq32            = gmx_mm256_calc_rsq_pd(dx32,dy32,dz32);
1857             rsq33            = gmx_mm256_calc_rsq_pd(dx33,dy33,dz33);
1858
1859             rinv11           = gmx_mm256_invsqrt_pd(rsq11);
1860             rinv12           = gmx_mm256_invsqrt_pd(rsq12);
1861             rinv13           = gmx_mm256_invsqrt_pd(rsq13);
1862             rinv21           = gmx_mm256_invsqrt_pd(rsq21);
1863             rinv22           = gmx_mm256_invsqrt_pd(rsq22);
1864             rinv23           = gmx_mm256_invsqrt_pd(rsq23);
1865             rinv31           = gmx_mm256_invsqrt_pd(rsq31);
1866             rinv32           = gmx_mm256_invsqrt_pd(rsq32);
1867             rinv33           = gmx_mm256_invsqrt_pd(rsq33);
1868
1869             rinvsq11         = _mm256_mul_pd(rinv11,rinv11);
1870             rinvsq12         = _mm256_mul_pd(rinv12,rinv12);
1871             rinvsq13         = _mm256_mul_pd(rinv13,rinv13);
1872             rinvsq21         = _mm256_mul_pd(rinv21,rinv21);
1873             rinvsq22         = _mm256_mul_pd(rinv22,rinv22);
1874             rinvsq23         = _mm256_mul_pd(rinv23,rinv23);
1875             rinvsq31         = _mm256_mul_pd(rinv31,rinv31);
1876             rinvsq32         = _mm256_mul_pd(rinv32,rinv32);
1877             rinvsq33         = _mm256_mul_pd(rinv33,rinv33);
1878
1879             fjx1             = _mm256_setzero_pd();
1880             fjy1             = _mm256_setzero_pd();
1881             fjz1             = _mm256_setzero_pd();
1882             fjx2             = _mm256_setzero_pd();
1883             fjy2             = _mm256_setzero_pd();
1884             fjz2             = _mm256_setzero_pd();
1885             fjx3             = _mm256_setzero_pd();
1886             fjy3             = _mm256_setzero_pd();
1887             fjz3             = _mm256_setzero_pd();
1888
1889             /**************************
1890              * CALCULATE INTERACTIONS *
1891              **************************/
1892
1893             if (gmx_mm256_any_lt(rsq11,rcutoff2))
1894             {
1895
1896             r11              = _mm256_mul_pd(rsq11,rinv11);
1897
1898             /* EWALD ELECTROSTATICS */
1899
1900             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1901             ewrt             = _mm256_mul_pd(r11,ewtabscale);
1902             ewitab           = _mm256_cvttpd_epi32(ewrt);
1903             eweps            = _mm256_sub_pd(ewrt,_mm256_round_pd(ewrt, _MM_FROUND_FLOOR));
1904             ewitab           = _mm_slli_epi32(ewitab,2);
1905             ewtabF           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
1906             ewtabD           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,1) );
1907             ewtabV           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,2) );
1908             ewtabFn          = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,3) );
1909             GMX_MM256_FULLTRANSPOSE4_PD(ewtabF,ewtabD,ewtabV,ewtabFn);
1910             felec            = _mm256_add_pd(ewtabF,_mm256_mul_pd(eweps,ewtabD));
1911             velec            = _mm256_sub_pd(ewtabV,_mm256_mul_pd(_mm256_mul_pd(ewtabhalfspace,eweps),_mm256_add_pd(ewtabF,felec)));
1912             velec            = _mm256_mul_pd(qq11,_mm256_sub_pd(rinv11,velec));
1913             felec            = _mm256_mul_pd(_mm256_mul_pd(qq11,rinv11),_mm256_sub_pd(rinvsq11,felec));
1914
1915             d                = _mm256_sub_pd(r11,rswitch);
1916             d                = _mm256_max_pd(d,_mm256_setzero_pd());
1917             d2               = _mm256_mul_pd(d,d);
1918             sw               = _mm256_add_pd(one,_mm256_mul_pd(d2,_mm256_mul_pd(d,_mm256_add_pd(swV3,_mm256_mul_pd(d,_mm256_add_pd(swV4,_mm256_mul_pd(d,swV5)))))));
1919
1920             dsw              = _mm256_mul_pd(d2,_mm256_add_pd(swF2,_mm256_mul_pd(d,_mm256_add_pd(swF3,_mm256_mul_pd(d,swF4)))));
1921
1922             /* Evaluate switch function */
1923             /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
1924             felec            = _mm256_sub_pd( _mm256_mul_pd(felec,sw) , _mm256_mul_pd(rinv11,_mm256_mul_pd(velec,dsw)) );
1925             cutoff_mask      = _mm256_cmp_pd(rsq11,rcutoff2,_CMP_LT_OQ);
1926
1927             fscal            = felec;
1928
1929             fscal            = _mm256_and_pd(fscal,cutoff_mask);
1930
1931             /* Calculate temporary vectorial force */
1932             tx               = _mm256_mul_pd(fscal,dx11);
1933             ty               = _mm256_mul_pd(fscal,dy11);
1934             tz               = _mm256_mul_pd(fscal,dz11);
1935
1936             /* Update vectorial force */
1937             fix1             = _mm256_add_pd(fix1,tx);
1938             fiy1             = _mm256_add_pd(fiy1,ty);
1939             fiz1             = _mm256_add_pd(fiz1,tz);
1940
1941             fjx1             = _mm256_add_pd(fjx1,tx);
1942             fjy1             = _mm256_add_pd(fjy1,ty);
1943             fjz1             = _mm256_add_pd(fjz1,tz);
1944
1945             }
1946
1947             /**************************
1948              * CALCULATE INTERACTIONS *
1949              **************************/
1950
1951             if (gmx_mm256_any_lt(rsq12,rcutoff2))
1952             {
1953
1954             r12              = _mm256_mul_pd(rsq12,rinv12);
1955
1956             /* EWALD ELECTROSTATICS */
1957
1958             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1959             ewrt             = _mm256_mul_pd(r12,ewtabscale);
1960             ewitab           = _mm256_cvttpd_epi32(ewrt);
1961             eweps            = _mm256_sub_pd(ewrt,_mm256_round_pd(ewrt, _MM_FROUND_FLOOR));
1962             ewitab           = _mm_slli_epi32(ewitab,2);
1963             ewtabF           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
1964             ewtabD           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,1) );
1965             ewtabV           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,2) );
1966             ewtabFn          = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,3) );
1967             GMX_MM256_FULLTRANSPOSE4_PD(ewtabF,ewtabD,ewtabV,ewtabFn);
1968             felec            = _mm256_add_pd(ewtabF,_mm256_mul_pd(eweps,ewtabD));
1969             velec            = _mm256_sub_pd(ewtabV,_mm256_mul_pd(_mm256_mul_pd(ewtabhalfspace,eweps),_mm256_add_pd(ewtabF,felec)));
1970             velec            = _mm256_mul_pd(qq12,_mm256_sub_pd(rinv12,velec));
1971             felec            = _mm256_mul_pd(_mm256_mul_pd(qq12,rinv12),_mm256_sub_pd(rinvsq12,felec));
1972
1973             d                = _mm256_sub_pd(r12,rswitch);
1974             d                = _mm256_max_pd(d,_mm256_setzero_pd());
1975             d2               = _mm256_mul_pd(d,d);
1976             sw               = _mm256_add_pd(one,_mm256_mul_pd(d2,_mm256_mul_pd(d,_mm256_add_pd(swV3,_mm256_mul_pd(d,_mm256_add_pd(swV4,_mm256_mul_pd(d,swV5)))))));
1977
1978             dsw              = _mm256_mul_pd(d2,_mm256_add_pd(swF2,_mm256_mul_pd(d,_mm256_add_pd(swF3,_mm256_mul_pd(d,swF4)))));
1979
1980             /* Evaluate switch function */
1981             /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
1982             felec            = _mm256_sub_pd( _mm256_mul_pd(felec,sw) , _mm256_mul_pd(rinv12,_mm256_mul_pd(velec,dsw)) );
1983             cutoff_mask      = _mm256_cmp_pd(rsq12,rcutoff2,_CMP_LT_OQ);
1984
1985             fscal            = felec;
1986
1987             fscal            = _mm256_and_pd(fscal,cutoff_mask);
1988
1989             /* Calculate temporary vectorial force */
1990             tx               = _mm256_mul_pd(fscal,dx12);
1991             ty               = _mm256_mul_pd(fscal,dy12);
1992             tz               = _mm256_mul_pd(fscal,dz12);
1993
1994             /* Update vectorial force */
1995             fix1             = _mm256_add_pd(fix1,tx);
1996             fiy1             = _mm256_add_pd(fiy1,ty);
1997             fiz1             = _mm256_add_pd(fiz1,tz);
1998
1999             fjx2             = _mm256_add_pd(fjx2,tx);
2000             fjy2             = _mm256_add_pd(fjy2,ty);
2001             fjz2             = _mm256_add_pd(fjz2,tz);
2002
2003             }
2004
2005             /**************************
2006              * CALCULATE INTERACTIONS *
2007              **************************/
2008
2009             if (gmx_mm256_any_lt(rsq13,rcutoff2))
2010             {
2011
2012             r13              = _mm256_mul_pd(rsq13,rinv13);
2013
2014             /* EWALD ELECTROSTATICS */
2015
2016             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
2017             ewrt             = _mm256_mul_pd(r13,ewtabscale);
2018             ewitab           = _mm256_cvttpd_epi32(ewrt);
2019             eweps            = _mm256_sub_pd(ewrt,_mm256_round_pd(ewrt, _MM_FROUND_FLOOR));
2020             ewitab           = _mm_slli_epi32(ewitab,2);
2021             ewtabF           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
2022             ewtabD           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,1) );
2023             ewtabV           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,2) );
2024             ewtabFn          = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,3) );
2025             GMX_MM256_FULLTRANSPOSE4_PD(ewtabF,ewtabD,ewtabV,ewtabFn);
2026             felec            = _mm256_add_pd(ewtabF,_mm256_mul_pd(eweps,ewtabD));
2027             velec            = _mm256_sub_pd(ewtabV,_mm256_mul_pd(_mm256_mul_pd(ewtabhalfspace,eweps),_mm256_add_pd(ewtabF,felec)));
2028             velec            = _mm256_mul_pd(qq13,_mm256_sub_pd(rinv13,velec));
2029             felec            = _mm256_mul_pd(_mm256_mul_pd(qq13,rinv13),_mm256_sub_pd(rinvsq13,felec));
2030
2031             d                = _mm256_sub_pd(r13,rswitch);
2032             d                = _mm256_max_pd(d,_mm256_setzero_pd());
2033             d2               = _mm256_mul_pd(d,d);
2034             sw               = _mm256_add_pd(one,_mm256_mul_pd(d2,_mm256_mul_pd(d,_mm256_add_pd(swV3,_mm256_mul_pd(d,_mm256_add_pd(swV4,_mm256_mul_pd(d,swV5)))))));
2035
2036             dsw              = _mm256_mul_pd(d2,_mm256_add_pd(swF2,_mm256_mul_pd(d,_mm256_add_pd(swF3,_mm256_mul_pd(d,swF4)))));
2037
2038             /* Evaluate switch function */
2039             /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
2040             felec            = _mm256_sub_pd( _mm256_mul_pd(felec,sw) , _mm256_mul_pd(rinv13,_mm256_mul_pd(velec,dsw)) );
2041             cutoff_mask      = _mm256_cmp_pd(rsq13,rcutoff2,_CMP_LT_OQ);
2042
2043             fscal            = felec;
2044
2045             fscal            = _mm256_and_pd(fscal,cutoff_mask);
2046
2047             /* Calculate temporary vectorial force */
2048             tx               = _mm256_mul_pd(fscal,dx13);
2049             ty               = _mm256_mul_pd(fscal,dy13);
2050             tz               = _mm256_mul_pd(fscal,dz13);
2051
2052             /* Update vectorial force */
2053             fix1             = _mm256_add_pd(fix1,tx);
2054             fiy1             = _mm256_add_pd(fiy1,ty);
2055             fiz1             = _mm256_add_pd(fiz1,tz);
2056
2057             fjx3             = _mm256_add_pd(fjx3,tx);
2058             fjy3             = _mm256_add_pd(fjy3,ty);
2059             fjz3             = _mm256_add_pd(fjz3,tz);
2060
2061             }
2062
2063             /**************************
2064              * CALCULATE INTERACTIONS *
2065              **************************/
2066
2067             if (gmx_mm256_any_lt(rsq21,rcutoff2))
2068             {
2069
2070             r21              = _mm256_mul_pd(rsq21,rinv21);
2071
2072             /* EWALD ELECTROSTATICS */
2073
2074             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
2075             ewrt             = _mm256_mul_pd(r21,ewtabscale);
2076             ewitab           = _mm256_cvttpd_epi32(ewrt);
2077             eweps            = _mm256_sub_pd(ewrt,_mm256_round_pd(ewrt, _MM_FROUND_FLOOR));
2078             ewitab           = _mm_slli_epi32(ewitab,2);
2079             ewtabF           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
2080             ewtabD           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,1) );
2081             ewtabV           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,2) );
2082             ewtabFn          = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,3) );
2083             GMX_MM256_FULLTRANSPOSE4_PD(ewtabF,ewtabD,ewtabV,ewtabFn);
2084             felec            = _mm256_add_pd(ewtabF,_mm256_mul_pd(eweps,ewtabD));
2085             velec            = _mm256_sub_pd(ewtabV,_mm256_mul_pd(_mm256_mul_pd(ewtabhalfspace,eweps),_mm256_add_pd(ewtabF,felec)));
2086             velec            = _mm256_mul_pd(qq21,_mm256_sub_pd(rinv21,velec));
2087             felec            = _mm256_mul_pd(_mm256_mul_pd(qq21,rinv21),_mm256_sub_pd(rinvsq21,felec));
2088
2089             d                = _mm256_sub_pd(r21,rswitch);
2090             d                = _mm256_max_pd(d,_mm256_setzero_pd());
2091             d2               = _mm256_mul_pd(d,d);
2092             sw               = _mm256_add_pd(one,_mm256_mul_pd(d2,_mm256_mul_pd(d,_mm256_add_pd(swV3,_mm256_mul_pd(d,_mm256_add_pd(swV4,_mm256_mul_pd(d,swV5)))))));
2093
2094             dsw              = _mm256_mul_pd(d2,_mm256_add_pd(swF2,_mm256_mul_pd(d,_mm256_add_pd(swF3,_mm256_mul_pd(d,swF4)))));
2095
2096             /* Evaluate switch function */
2097             /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
2098             felec            = _mm256_sub_pd( _mm256_mul_pd(felec,sw) , _mm256_mul_pd(rinv21,_mm256_mul_pd(velec,dsw)) );
2099             cutoff_mask      = _mm256_cmp_pd(rsq21,rcutoff2,_CMP_LT_OQ);
2100
2101             fscal            = felec;
2102
2103             fscal            = _mm256_and_pd(fscal,cutoff_mask);
2104
2105             /* Calculate temporary vectorial force */
2106             tx               = _mm256_mul_pd(fscal,dx21);
2107             ty               = _mm256_mul_pd(fscal,dy21);
2108             tz               = _mm256_mul_pd(fscal,dz21);
2109
2110             /* Update vectorial force */
2111             fix2             = _mm256_add_pd(fix2,tx);
2112             fiy2             = _mm256_add_pd(fiy2,ty);
2113             fiz2             = _mm256_add_pd(fiz2,tz);
2114
2115             fjx1             = _mm256_add_pd(fjx1,tx);
2116             fjy1             = _mm256_add_pd(fjy1,ty);
2117             fjz1             = _mm256_add_pd(fjz1,tz);
2118
2119             }
2120
2121             /**************************
2122              * CALCULATE INTERACTIONS *
2123              **************************/
2124
2125             if (gmx_mm256_any_lt(rsq22,rcutoff2))
2126             {
2127
2128             r22              = _mm256_mul_pd(rsq22,rinv22);
2129
2130             /* EWALD ELECTROSTATICS */
2131
2132             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
2133             ewrt             = _mm256_mul_pd(r22,ewtabscale);
2134             ewitab           = _mm256_cvttpd_epi32(ewrt);
2135             eweps            = _mm256_sub_pd(ewrt,_mm256_round_pd(ewrt, _MM_FROUND_FLOOR));
2136             ewitab           = _mm_slli_epi32(ewitab,2);
2137             ewtabF           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
2138             ewtabD           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,1) );
2139             ewtabV           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,2) );
2140             ewtabFn          = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,3) );
2141             GMX_MM256_FULLTRANSPOSE4_PD(ewtabF,ewtabD,ewtabV,ewtabFn);
2142             felec            = _mm256_add_pd(ewtabF,_mm256_mul_pd(eweps,ewtabD));
2143             velec            = _mm256_sub_pd(ewtabV,_mm256_mul_pd(_mm256_mul_pd(ewtabhalfspace,eweps),_mm256_add_pd(ewtabF,felec)));
2144             velec            = _mm256_mul_pd(qq22,_mm256_sub_pd(rinv22,velec));
2145             felec            = _mm256_mul_pd(_mm256_mul_pd(qq22,rinv22),_mm256_sub_pd(rinvsq22,felec));
2146
2147             d                = _mm256_sub_pd(r22,rswitch);
2148             d                = _mm256_max_pd(d,_mm256_setzero_pd());
2149             d2               = _mm256_mul_pd(d,d);
2150             sw               = _mm256_add_pd(one,_mm256_mul_pd(d2,_mm256_mul_pd(d,_mm256_add_pd(swV3,_mm256_mul_pd(d,_mm256_add_pd(swV4,_mm256_mul_pd(d,swV5)))))));
2151
2152             dsw              = _mm256_mul_pd(d2,_mm256_add_pd(swF2,_mm256_mul_pd(d,_mm256_add_pd(swF3,_mm256_mul_pd(d,swF4)))));
2153
2154             /* Evaluate switch function */
2155             /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
2156             felec            = _mm256_sub_pd( _mm256_mul_pd(felec,sw) , _mm256_mul_pd(rinv22,_mm256_mul_pd(velec,dsw)) );
2157             cutoff_mask      = _mm256_cmp_pd(rsq22,rcutoff2,_CMP_LT_OQ);
2158
2159             fscal            = felec;
2160
2161             fscal            = _mm256_and_pd(fscal,cutoff_mask);
2162
2163             /* Calculate temporary vectorial force */
2164             tx               = _mm256_mul_pd(fscal,dx22);
2165             ty               = _mm256_mul_pd(fscal,dy22);
2166             tz               = _mm256_mul_pd(fscal,dz22);
2167
2168             /* Update vectorial force */
2169             fix2             = _mm256_add_pd(fix2,tx);
2170             fiy2             = _mm256_add_pd(fiy2,ty);
2171             fiz2             = _mm256_add_pd(fiz2,tz);
2172
2173             fjx2             = _mm256_add_pd(fjx2,tx);
2174             fjy2             = _mm256_add_pd(fjy2,ty);
2175             fjz2             = _mm256_add_pd(fjz2,tz);
2176
2177             }
2178
2179             /**************************
2180              * CALCULATE INTERACTIONS *
2181              **************************/
2182
2183             if (gmx_mm256_any_lt(rsq23,rcutoff2))
2184             {
2185
2186             r23              = _mm256_mul_pd(rsq23,rinv23);
2187
2188             /* EWALD ELECTROSTATICS */
2189
2190             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
2191             ewrt             = _mm256_mul_pd(r23,ewtabscale);
2192             ewitab           = _mm256_cvttpd_epi32(ewrt);
2193             eweps            = _mm256_sub_pd(ewrt,_mm256_round_pd(ewrt, _MM_FROUND_FLOOR));
2194             ewitab           = _mm_slli_epi32(ewitab,2);
2195             ewtabF           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
2196             ewtabD           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,1) );
2197             ewtabV           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,2) );
2198             ewtabFn          = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,3) );
2199             GMX_MM256_FULLTRANSPOSE4_PD(ewtabF,ewtabD,ewtabV,ewtabFn);
2200             felec            = _mm256_add_pd(ewtabF,_mm256_mul_pd(eweps,ewtabD));
2201             velec            = _mm256_sub_pd(ewtabV,_mm256_mul_pd(_mm256_mul_pd(ewtabhalfspace,eweps),_mm256_add_pd(ewtabF,felec)));
2202             velec            = _mm256_mul_pd(qq23,_mm256_sub_pd(rinv23,velec));
2203             felec            = _mm256_mul_pd(_mm256_mul_pd(qq23,rinv23),_mm256_sub_pd(rinvsq23,felec));
2204
2205             d                = _mm256_sub_pd(r23,rswitch);
2206             d                = _mm256_max_pd(d,_mm256_setzero_pd());
2207             d2               = _mm256_mul_pd(d,d);
2208             sw               = _mm256_add_pd(one,_mm256_mul_pd(d2,_mm256_mul_pd(d,_mm256_add_pd(swV3,_mm256_mul_pd(d,_mm256_add_pd(swV4,_mm256_mul_pd(d,swV5)))))));
2209
2210             dsw              = _mm256_mul_pd(d2,_mm256_add_pd(swF2,_mm256_mul_pd(d,_mm256_add_pd(swF3,_mm256_mul_pd(d,swF4)))));
2211
2212             /* Evaluate switch function */
2213             /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
2214             felec            = _mm256_sub_pd( _mm256_mul_pd(felec,sw) , _mm256_mul_pd(rinv23,_mm256_mul_pd(velec,dsw)) );
2215             cutoff_mask      = _mm256_cmp_pd(rsq23,rcutoff2,_CMP_LT_OQ);
2216
2217             fscal            = felec;
2218
2219             fscal            = _mm256_and_pd(fscal,cutoff_mask);
2220
2221             /* Calculate temporary vectorial force */
2222             tx               = _mm256_mul_pd(fscal,dx23);
2223             ty               = _mm256_mul_pd(fscal,dy23);
2224             tz               = _mm256_mul_pd(fscal,dz23);
2225
2226             /* Update vectorial force */
2227             fix2             = _mm256_add_pd(fix2,tx);
2228             fiy2             = _mm256_add_pd(fiy2,ty);
2229             fiz2             = _mm256_add_pd(fiz2,tz);
2230
2231             fjx3             = _mm256_add_pd(fjx3,tx);
2232             fjy3             = _mm256_add_pd(fjy3,ty);
2233             fjz3             = _mm256_add_pd(fjz3,tz);
2234
2235             }
2236
2237             /**************************
2238              * CALCULATE INTERACTIONS *
2239              **************************/
2240
2241             if (gmx_mm256_any_lt(rsq31,rcutoff2))
2242             {
2243
2244             r31              = _mm256_mul_pd(rsq31,rinv31);
2245
2246             /* EWALD ELECTROSTATICS */
2247
2248             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
2249             ewrt             = _mm256_mul_pd(r31,ewtabscale);
2250             ewitab           = _mm256_cvttpd_epi32(ewrt);
2251             eweps            = _mm256_sub_pd(ewrt,_mm256_round_pd(ewrt, _MM_FROUND_FLOOR));
2252             ewitab           = _mm_slli_epi32(ewitab,2);
2253             ewtabF           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
2254             ewtabD           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,1) );
2255             ewtabV           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,2) );
2256             ewtabFn          = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,3) );
2257             GMX_MM256_FULLTRANSPOSE4_PD(ewtabF,ewtabD,ewtabV,ewtabFn);
2258             felec            = _mm256_add_pd(ewtabF,_mm256_mul_pd(eweps,ewtabD));
2259             velec            = _mm256_sub_pd(ewtabV,_mm256_mul_pd(_mm256_mul_pd(ewtabhalfspace,eweps),_mm256_add_pd(ewtabF,felec)));
2260             velec            = _mm256_mul_pd(qq31,_mm256_sub_pd(rinv31,velec));
2261             felec            = _mm256_mul_pd(_mm256_mul_pd(qq31,rinv31),_mm256_sub_pd(rinvsq31,felec));
2262
2263             d                = _mm256_sub_pd(r31,rswitch);
2264             d                = _mm256_max_pd(d,_mm256_setzero_pd());
2265             d2               = _mm256_mul_pd(d,d);
2266             sw               = _mm256_add_pd(one,_mm256_mul_pd(d2,_mm256_mul_pd(d,_mm256_add_pd(swV3,_mm256_mul_pd(d,_mm256_add_pd(swV4,_mm256_mul_pd(d,swV5)))))));
2267
2268             dsw              = _mm256_mul_pd(d2,_mm256_add_pd(swF2,_mm256_mul_pd(d,_mm256_add_pd(swF3,_mm256_mul_pd(d,swF4)))));
2269
2270             /* Evaluate switch function */
2271             /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
2272             felec            = _mm256_sub_pd( _mm256_mul_pd(felec,sw) , _mm256_mul_pd(rinv31,_mm256_mul_pd(velec,dsw)) );
2273             cutoff_mask      = _mm256_cmp_pd(rsq31,rcutoff2,_CMP_LT_OQ);
2274
2275             fscal            = felec;
2276
2277             fscal            = _mm256_and_pd(fscal,cutoff_mask);
2278
2279             /* Calculate temporary vectorial force */
2280             tx               = _mm256_mul_pd(fscal,dx31);
2281             ty               = _mm256_mul_pd(fscal,dy31);
2282             tz               = _mm256_mul_pd(fscal,dz31);
2283
2284             /* Update vectorial force */
2285             fix3             = _mm256_add_pd(fix3,tx);
2286             fiy3             = _mm256_add_pd(fiy3,ty);
2287             fiz3             = _mm256_add_pd(fiz3,tz);
2288
2289             fjx1             = _mm256_add_pd(fjx1,tx);
2290             fjy1             = _mm256_add_pd(fjy1,ty);
2291             fjz1             = _mm256_add_pd(fjz1,tz);
2292
2293             }
2294
2295             /**************************
2296              * CALCULATE INTERACTIONS *
2297              **************************/
2298
2299             if (gmx_mm256_any_lt(rsq32,rcutoff2))
2300             {
2301
2302             r32              = _mm256_mul_pd(rsq32,rinv32);
2303
2304             /* EWALD ELECTROSTATICS */
2305
2306             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
2307             ewrt             = _mm256_mul_pd(r32,ewtabscale);
2308             ewitab           = _mm256_cvttpd_epi32(ewrt);
2309             eweps            = _mm256_sub_pd(ewrt,_mm256_round_pd(ewrt, _MM_FROUND_FLOOR));
2310             ewitab           = _mm_slli_epi32(ewitab,2);
2311             ewtabF           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
2312             ewtabD           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,1) );
2313             ewtabV           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,2) );
2314             ewtabFn          = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,3) );
2315             GMX_MM256_FULLTRANSPOSE4_PD(ewtabF,ewtabD,ewtabV,ewtabFn);
2316             felec            = _mm256_add_pd(ewtabF,_mm256_mul_pd(eweps,ewtabD));
2317             velec            = _mm256_sub_pd(ewtabV,_mm256_mul_pd(_mm256_mul_pd(ewtabhalfspace,eweps),_mm256_add_pd(ewtabF,felec)));
2318             velec            = _mm256_mul_pd(qq32,_mm256_sub_pd(rinv32,velec));
2319             felec            = _mm256_mul_pd(_mm256_mul_pd(qq32,rinv32),_mm256_sub_pd(rinvsq32,felec));
2320
2321             d                = _mm256_sub_pd(r32,rswitch);
2322             d                = _mm256_max_pd(d,_mm256_setzero_pd());
2323             d2               = _mm256_mul_pd(d,d);
2324             sw               = _mm256_add_pd(one,_mm256_mul_pd(d2,_mm256_mul_pd(d,_mm256_add_pd(swV3,_mm256_mul_pd(d,_mm256_add_pd(swV4,_mm256_mul_pd(d,swV5)))))));
2325
2326             dsw              = _mm256_mul_pd(d2,_mm256_add_pd(swF2,_mm256_mul_pd(d,_mm256_add_pd(swF3,_mm256_mul_pd(d,swF4)))));
2327
2328             /* Evaluate switch function */
2329             /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
2330             felec            = _mm256_sub_pd( _mm256_mul_pd(felec,sw) , _mm256_mul_pd(rinv32,_mm256_mul_pd(velec,dsw)) );
2331             cutoff_mask      = _mm256_cmp_pd(rsq32,rcutoff2,_CMP_LT_OQ);
2332
2333             fscal            = felec;
2334
2335             fscal            = _mm256_and_pd(fscal,cutoff_mask);
2336
2337             /* Calculate temporary vectorial force */
2338             tx               = _mm256_mul_pd(fscal,dx32);
2339             ty               = _mm256_mul_pd(fscal,dy32);
2340             tz               = _mm256_mul_pd(fscal,dz32);
2341
2342             /* Update vectorial force */
2343             fix3             = _mm256_add_pd(fix3,tx);
2344             fiy3             = _mm256_add_pd(fiy3,ty);
2345             fiz3             = _mm256_add_pd(fiz3,tz);
2346
2347             fjx2             = _mm256_add_pd(fjx2,tx);
2348             fjy2             = _mm256_add_pd(fjy2,ty);
2349             fjz2             = _mm256_add_pd(fjz2,tz);
2350
2351             }
2352
2353             /**************************
2354              * CALCULATE INTERACTIONS *
2355              **************************/
2356
2357             if (gmx_mm256_any_lt(rsq33,rcutoff2))
2358             {
2359
2360             r33              = _mm256_mul_pd(rsq33,rinv33);
2361
2362             /* EWALD ELECTROSTATICS */
2363
2364             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
2365             ewrt             = _mm256_mul_pd(r33,ewtabscale);
2366             ewitab           = _mm256_cvttpd_epi32(ewrt);
2367             eweps            = _mm256_sub_pd(ewrt,_mm256_round_pd(ewrt, _MM_FROUND_FLOOR));
2368             ewitab           = _mm_slli_epi32(ewitab,2);
2369             ewtabF           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
2370             ewtabD           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,1) );
2371             ewtabV           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,2) );
2372             ewtabFn          = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,3) );
2373             GMX_MM256_FULLTRANSPOSE4_PD(ewtabF,ewtabD,ewtabV,ewtabFn);
2374             felec            = _mm256_add_pd(ewtabF,_mm256_mul_pd(eweps,ewtabD));
2375             velec            = _mm256_sub_pd(ewtabV,_mm256_mul_pd(_mm256_mul_pd(ewtabhalfspace,eweps),_mm256_add_pd(ewtabF,felec)));
2376             velec            = _mm256_mul_pd(qq33,_mm256_sub_pd(rinv33,velec));
2377             felec            = _mm256_mul_pd(_mm256_mul_pd(qq33,rinv33),_mm256_sub_pd(rinvsq33,felec));
2378
2379             d                = _mm256_sub_pd(r33,rswitch);
2380             d                = _mm256_max_pd(d,_mm256_setzero_pd());
2381             d2               = _mm256_mul_pd(d,d);
2382             sw               = _mm256_add_pd(one,_mm256_mul_pd(d2,_mm256_mul_pd(d,_mm256_add_pd(swV3,_mm256_mul_pd(d,_mm256_add_pd(swV4,_mm256_mul_pd(d,swV5)))))));
2383
2384             dsw              = _mm256_mul_pd(d2,_mm256_add_pd(swF2,_mm256_mul_pd(d,_mm256_add_pd(swF3,_mm256_mul_pd(d,swF4)))));
2385
2386             /* Evaluate switch function */
2387             /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
2388             felec            = _mm256_sub_pd( _mm256_mul_pd(felec,sw) , _mm256_mul_pd(rinv33,_mm256_mul_pd(velec,dsw)) );
2389             cutoff_mask      = _mm256_cmp_pd(rsq33,rcutoff2,_CMP_LT_OQ);
2390
2391             fscal            = felec;
2392
2393             fscal            = _mm256_and_pd(fscal,cutoff_mask);
2394
2395             /* Calculate temporary vectorial force */
2396             tx               = _mm256_mul_pd(fscal,dx33);
2397             ty               = _mm256_mul_pd(fscal,dy33);
2398             tz               = _mm256_mul_pd(fscal,dz33);
2399
2400             /* Update vectorial force */
2401             fix3             = _mm256_add_pd(fix3,tx);
2402             fiy3             = _mm256_add_pd(fiy3,ty);
2403             fiz3             = _mm256_add_pd(fiz3,tz);
2404
2405             fjx3             = _mm256_add_pd(fjx3,tx);
2406             fjy3             = _mm256_add_pd(fjy3,ty);
2407             fjz3             = _mm256_add_pd(fjz3,tz);
2408
2409             }
2410
2411             fjptrA             = f+j_coord_offsetA;
2412             fjptrB             = f+j_coord_offsetB;
2413             fjptrC             = f+j_coord_offsetC;
2414             fjptrD             = f+j_coord_offsetD;
2415
2416             gmx_mm256_decrement_3rvec_4ptr_swizzle_pd(fjptrA+DIM,fjptrB+DIM,fjptrC+DIM,fjptrD+DIM,
2417                                                       fjx1,fjy1,fjz1,fjx2,fjy2,fjz2,fjx3,fjy3,fjz3);
2418
2419             /* Inner loop uses 558 flops */
2420         }
2421
2422         if(jidx<j_index_end)
2423         {
2424
2425             /* Get j neighbor index, and coordinate index */
2426             jnrlistA         = jjnr[jidx];
2427             jnrlistB         = jjnr[jidx+1];
2428             jnrlistC         = jjnr[jidx+2];
2429             jnrlistD         = jjnr[jidx+3];
2430             /* Sign of each element will be negative for non-real atoms.
2431              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
2432              * so use it as val = _mm_andnot_pd(mask,val) to clear dummy entries.
2433              */
2434             tmpmask0 = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
2435
2436             tmpmask1 = _mm_permute_ps(tmpmask0,_GMX_MM_PERMUTE(3,3,2,2));
2437             tmpmask0 = _mm_permute_ps(tmpmask0,_GMX_MM_PERMUTE(1,1,0,0));
2438             dummy_mask = _mm256_castps_pd(gmx_mm256_set_m128(tmpmask1,tmpmask0));
2439
2440             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
2441             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
2442             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
2443             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
2444             j_coord_offsetA  = DIM*jnrA;
2445             j_coord_offsetB  = DIM*jnrB;
2446             j_coord_offsetC  = DIM*jnrC;
2447             j_coord_offsetD  = DIM*jnrD;
2448
2449             /* load j atom coordinates */
2450             gmx_mm256_load_3rvec_4ptr_swizzle_pd(x+j_coord_offsetA+DIM,x+j_coord_offsetB+DIM,
2451                                                  x+j_coord_offsetC+DIM,x+j_coord_offsetD+DIM,
2452                                                  &jx1,&jy1,&jz1,&jx2,&jy2,&jz2,&jx3,&jy3,&jz3);
2453
2454             /* Calculate displacement vector */
2455             dx11             = _mm256_sub_pd(ix1,jx1);
2456             dy11             = _mm256_sub_pd(iy1,jy1);
2457             dz11             = _mm256_sub_pd(iz1,jz1);
2458             dx12             = _mm256_sub_pd(ix1,jx2);
2459             dy12             = _mm256_sub_pd(iy1,jy2);
2460             dz12             = _mm256_sub_pd(iz1,jz2);
2461             dx13             = _mm256_sub_pd(ix1,jx3);
2462             dy13             = _mm256_sub_pd(iy1,jy3);
2463             dz13             = _mm256_sub_pd(iz1,jz3);
2464             dx21             = _mm256_sub_pd(ix2,jx1);
2465             dy21             = _mm256_sub_pd(iy2,jy1);
2466             dz21             = _mm256_sub_pd(iz2,jz1);
2467             dx22             = _mm256_sub_pd(ix2,jx2);
2468             dy22             = _mm256_sub_pd(iy2,jy2);
2469             dz22             = _mm256_sub_pd(iz2,jz2);
2470             dx23             = _mm256_sub_pd(ix2,jx3);
2471             dy23             = _mm256_sub_pd(iy2,jy3);
2472             dz23             = _mm256_sub_pd(iz2,jz3);
2473             dx31             = _mm256_sub_pd(ix3,jx1);
2474             dy31             = _mm256_sub_pd(iy3,jy1);
2475             dz31             = _mm256_sub_pd(iz3,jz1);
2476             dx32             = _mm256_sub_pd(ix3,jx2);
2477             dy32             = _mm256_sub_pd(iy3,jy2);
2478             dz32             = _mm256_sub_pd(iz3,jz2);
2479             dx33             = _mm256_sub_pd(ix3,jx3);
2480             dy33             = _mm256_sub_pd(iy3,jy3);
2481             dz33             = _mm256_sub_pd(iz3,jz3);
2482
2483             /* Calculate squared distance and things based on it */
2484             rsq11            = gmx_mm256_calc_rsq_pd(dx11,dy11,dz11);
2485             rsq12            = gmx_mm256_calc_rsq_pd(dx12,dy12,dz12);
2486             rsq13            = gmx_mm256_calc_rsq_pd(dx13,dy13,dz13);
2487             rsq21            = gmx_mm256_calc_rsq_pd(dx21,dy21,dz21);
2488             rsq22            = gmx_mm256_calc_rsq_pd(dx22,dy22,dz22);
2489             rsq23            = gmx_mm256_calc_rsq_pd(dx23,dy23,dz23);
2490             rsq31            = gmx_mm256_calc_rsq_pd(dx31,dy31,dz31);
2491             rsq32            = gmx_mm256_calc_rsq_pd(dx32,dy32,dz32);
2492             rsq33            = gmx_mm256_calc_rsq_pd(dx33,dy33,dz33);
2493
2494             rinv11           = gmx_mm256_invsqrt_pd(rsq11);
2495             rinv12           = gmx_mm256_invsqrt_pd(rsq12);
2496             rinv13           = gmx_mm256_invsqrt_pd(rsq13);
2497             rinv21           = gmx_mm256_invsqrt_pd(rsq21);
2498             rinv22           = gmx_mm256_invsqrt_pd(rsq22);
2499             rinv23           = gmx_mm256_invsqrt_pd(rsq23);
2500             rinv31           = gmx_mm256_invsqrt_pd(rsq31);
2501             rinv32           = gmx_mm256_invsqrt_pd(rsq32);
2502             rinv33           = gmx_mm256_invsqrt_pd(rsq33);
2503
2504             rinvsq11         = _mm256_mul_pd(rinv11,rinv11);
2505             rinvsq12         = _mm256_mul_pd(rinv12,rinv12);
2506             rinvsq13         = _mm256_mul_pd(rinv13,rinv13);
2507             rinvsq21         = _mm256_mul_pd(rinv21,rinv21);
2508             rinvsq22         = _mm256_mul_pd(rinv22,rinv22);
2509             rinvsq23         = _mm256_mul_pd(rinv23,rinv23);
2510             rinvsq31         = _mm256_mul_pd(rinv31,rinv31);
2511             rinvsq32         = _mm256_mul_pd(rinv32,rinv32);
2512             rinvsq33         = _mm256_mul_pd(rinv33,rinv33);
2513
2514             fjx1             = _mm256_setzero_pd();
2515             fjy1             = _mm256_setzero_pd();
2516             fjz1             = _mm256_setzero_pd();
2517             fjx2             = _mm256_setzero_pd();
2518             fjy2             = _mm256_setzero_pd();
2519             fjz2             = _mm256_setzero_pd();
2520             fjx3             = _mm256_setzero_pd();
2521             fjy3             = _mm256_setzero_pd();
2522             fjz3             = _mm256_setzero_pd();
2523
2524             /**************************
2525              * CALCULATE INTERACTIONS *
2526              **************************/
2527
2528             if (gmx_mm256_any_lt(rsq11,rcutoff2))
2529             {
2530
2531             r11              = _mm256_mul_pd(rsq11,rinv11);
2532             r11              = _mm256_andnot_pd(dummy_mask,r11);
2533
2534             /* EWALD ELECTROSTATICS */
2535
2536             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
2537             ewrt             = _mm256_mul_pd(r11,ewtabscale);
2538             ewitab           = _mm256_cvttpd_epi32(ewrt);
2539             eweps            = _mm256_sub_pd(ewrt,_mm256_round_pd(ewrt, _MM_FROUND_FLOOR));
2540             ewitab           = _mm_slli_epi32(ewitab,2);
2541             ewtabF           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
2542             ewtabD           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,1) );
2543             ewtabV           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,2) );
2544             ewtabFn          = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,3) );
2545             GMX_MM256_FULLTRANSPOSE4_PD(ewtabF,ewtabD,ewtabV,ewtabFn);
2546             felec            = _mm256_add_pd(ewtabF,_mm256_mul_pd(eweps,ewtabD));
2547             velec            = _mm256_sub_pd(ewtabV,_mm256_mul_pd(_mm256_mul_pd(ewtabhalfspace,eweps),_mm256_add_pd(ewtabF,felec)));
2548             velec            = _mm256_mul_pd(qq11,_mm256_sub_pd(rinv11,velec));
2549             felec            = _mm256_mul_pd(_mm256_mul_pd(qq11,rinv11),_mm256_sub_pd(rinvsq11,felec));
2550
2551             d                = _mm256_sub_pd(r11,rswitch);
2552             d                = _mm256_max_pd(d,_mm256_setzero_pd());
2553             d2               = _mm256_mul_pd(d,d);
2554             sw               = _mm256_add_pd(one,_mm256_mul_pd(d2,_mm256_mul_pd(d,_mm256_add_pd(swV3,_mm256_mul_pd(d,_mm256_add_pd(swV4,_mm256_mul_pd(d,swV5)))))));
2555
2556             dsw              = _mm256_mul_pd(d2,_mm256_add_pd(swF2,_mm256_mul_pd(d,_mm256_add_pd(swF3,_mm256_mul_pd(d,swF4)))));
2557
2558             /* Evaluate switch function */
2559             /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
2560             felec            = _mm256_sub_pd( _mm256_mul_pd(felec,sw) , _mm256_mul_pd(rinv11,_mm256_mul_pd(velec,dsw)) );
2561             cutoff_mask      = _mm256_cmp_pd(rsq11,rcutoff2,_CMP_LT_OQ);
2562
2563             fscal            = felec;
2564
2565             fscal            = _mm256_and_pd(fscal,cutoff_mask);
2566
2567             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
2568
2569             /* Calculate temporary vectorial force */
2570             tx               = _mm256_mul_pd(fscal,dx11);
2571             ty               = _mm256_mul_pd(fscal,dy11);
2572             tz               = _mm256_mul_pd(fscal,dz11);
2573
2574             /* Update vectorial force */
2575             fix1             = _mm256_add_pd(fix1,tx);
2576             fiy1             = _mm256_add_pd(fiy1,ty);
2577             fiz1             = _mm256_add_pd(fiz1,tz);
2578
2579             fjx1             = _mm256_add_pd(fjx1,tx);
2580             fjy1             = _mm256_add_pd(fjy1,ty);
2581             fjz1             = _mm256_add_pd(fjz1,tz);
2582
2583             }
2584
2585             /**************************
2586              * CALCULATE INTERACTIONS *
2587              **************************/
2588
2589             if (gmx_mm256_any_lt(rsq12,rcutoff2))
2590             {
2591
2592             r12              = _mm256_mul_pd(rsq12,rinv12);
2593             r12              = _mm256_andnot_pd(dummy_mask,r12);
2594
2595             /* EWALD ELECTROSTATICS */
2596
2597             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
2598             ewrt             = _mm256_mul_pd(r12,ewtabscale);
2599             ewitab           = _mm256_cvttpd_epi32(ewrt);
2600             eweps            = _mm256_sub_pd(ewrt,_mm256_round_pd(ewrt, _MM_FROUND_FLOOR));
2601             ewitab           = _mm_slli_epi32(ewitab,2);
2602             ewtabF           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
2603             ewtabD           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,1) );
2604             ewtabV           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,2) );
2605             ewtabFn          = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,3) );
2606             GMX_MM256_FULLTRANSPOSE4_PD(ewtabF,ewtabD,ewtabV,ewtabFn);
2607             felec            = _mm256_add_pd(ewtabF,_mm256_mul_pd(eweps,ewtabD));
2608             velec            = _mm256_sub_pd(ewtabV,_mm256_mul_pd(_mm256_mul_pd(ewtabhalfspace,eweps),_mm256_add_pd(ewtabF,felec)));
2609             velec            = _mm256_mul_pd(qq12,_mm256_sub_pd(rinv12,velec));
2610             felec            = _mm256_mul_pd(_mm256_mul_pd(qq12,rinv12),_mm256_sub_pd(rinvsq12,felec));
2611
2612             d                = _mm256_sub_pd(r12,rswitch);
2613             d                = _mm256_max_pd(d,_mm256_setzero_pd());
2614             d2               = _mm256_mul_pd(d,d);
2615             sw               = _mm256_add_pd(one,_mm256_mul_pd(d2,_mm256_mul_pd(d,_mm256_add_pd(swV3,_mm256_mul_pd(d,_mm256_add_pd(swV4,_mm256_mul_pd(d,swV5)))))));
2616
2617             dsw              = _mm256_mul_pd(d2,_mm256_add_pd(swF2,_mm256_mul_pd(d,_mm256_add_pd(swF3,_mm256_mul_pd(d,swF4)))));
2618
2619             /* Evaluate switch function */
2620             /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
2621             felec            = _mm256_sub_pd( _mm256_mul_pd(felec,sw) , _mm256_mul_pd(rinv12,_mm256_mul_pd(velec,dsw)) );
2622             cutoff_mask      = _mm256_cmp_pd(rsq12,rcutoff2,_CMP_LT_OQ);
2623
2624             fscal            = felec;
2625
2626             fscal            = _mm256_and_pd(fscal,cutoff_mask);
2627
2628             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
2629
2630             /* Calculate temporary vectorial force */
2631             tx               = _mm256_mul_pd(fscal,dx12);
2632             ty               = _mm256_mul_pd(fscal,dy12);
2633             tz               = _mm256_mul_pd(fscal,dz12);
2634
2635             /* Update vectorial force */
2636             fix1             = _mm256_add_pd(fix1,tx);
2637             fiy1             = _mm256_add_pd(fiy1,ty);
2638             fiz1             = _mm256_add_pd(fiz1,tz);
2639
2640             fjx2             = _mm256_add_pd(fjx2,tx);
2641             fjy2             = _mm256_add_pd(fjy2,ty);
2642             fjz2             = _mm256_add_pd(fjz2,tz);
2643
2644             }
2645
2646             /**************************
2647              * CALCULATE INTERACTIONS *
2648              **************************/
2649
2650             if (gmx_mm256_any_lt(rsq13,rcutoff2))
2651             {
2652
2653             r13              = _mm256_mul_pd(rsq13,rinv13);
2654             r13              = _mm256_andnot_pd(dummy_mask,r13);
2655
2656             /* EWALD ELECTROSTATICS */
2657
2658             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
2659             ewrt             = _mm256_mul_pd(r13,ewtabscale);
2660             ewitab           = _mm256_cvttpd_epi32(ewrt);
2661             eweps            = _mm256_sub_pd(ewrt,_mm256_round_pd(ewrt, _MM_FROUND_FLOOR));
2662             ewitab           = _mm_slli_epi32(ewitab,2);
2663             ewtabF           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
2664             ewtabD           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,1) );
2665             ewtabV           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,2) );
2666             ewtabFn          = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,3) );
2667             GMX_MM256_FULLTRANSPOSE4_PD(ewtabF,ewtabD,ewtabV,ewtabFn);
2668             felec            = _mm256_add_pd(ewtabF,_mm256_mul_pd(eweps,ewtabD));
2669             velec            = _mm256_sub_pd(ewtabV,_mm256_mul_pd(_mm256_mul_pd(ewtabhalfspace,eweps),_mm256_add_pd(ewtabF,felec)));
2670             velec            = _mm256_mul_pd(qq13,_mm256_sub_pd(rinv13,velec));
2671             felec            = _mm256_mul_pd(_mm256_mul_pd(qq13,rinv13),_mm256_sub_pd(rinvsq13,felec));
2672
2673             d                = _mm256_sub_pd(r13,rswitch);
2674             d                = _mm256_max_pd(d,_mm256_setzero_pd());
2675             d2               = _mm256_mul_pd(d,d);
2676             sw               = _mm256_add_pd(one,_mm256_mul_pd(d2,_mm256_mul_pd(d,_mm256_add_pd(swV3,_mm256_mul_pd(d,_mm256_add_pd(swV4,_mm256_mul_pd(d,swV5)))))));
2677
2678             dsw              = _mm256_mul_pd(d2,_mm256_add_pd(swF2,_mm256_mul_pd(d,_mm256_add_pd(swF3,_mm256_mul_pd(d,swF4)))));
2679
2680             /* Evaluate switch function */
2681             /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
2682             felec            = _mm256_sub_pd( _mm256_mul_pd(felec,sw) , _mm256_mul_pd(rinv13,_mm256_mul_pd(velec,dsw)) );
2683             cutoff_mask      = _mm256_cmp_pd(rsq13,rcutoff2,_CMP_LT_OQ);
2684
2685             fscal            = felec;
2686
2687             fscal            = _mm256_and_pd(fscal,cutoff_mask);
2688
2689             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
2690
2691             /* Calculate temporary vectorial force */
2692             tx               = _mm256_mul_pd(fscal,dx13);
2693             ty               = _mm256_mul_pd(fscal,dy13);
2694             tz               = _mm256_mul_pd(fscal,dz13);
2695
2696             /* Update vectorial force */
2697             fix1             = _mm256_add_pd(fix1,tx);
2698             fiy1             = _mm256_add_pd(fiy1,ty);
2699             fiz1             = _mm256_add_pd(fiz1,tz);
2700
2701             fjx3             = _mm256_add_pd(fjx3,tx);
2702             fjy3             = _mm256_add_pd(fjy3,ty);
2703             fjz3             = _mm256_add_pd(fjz3,tz);
2704
2705             }
2706
2707             /**************************
2708              * CALCULATE INTERACTIONS *
2709              **************************/
2710
2711             if (gmx_mm256_any_lt(rsq21,rcutoff2))
2712             {
2713
2714             r21              = _mm256_mul_pd(rsq21,rinv21);
2715             r21              = _mm256_andnot_pd(dummy_mask,r21);
2716
2717             /* EWALD ELECTROSTATICS */
2718
2719             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
2720             ewrt             = _mm256_mul_pd(r21,ewtabscale);
2721             ewitab           = _mm256_cvttpd_epi32(ewrt);
2722             eweps            = _mm256_sub_pd(ewrt,_mm256_round_pd(ewrt, _MM_FROUND_FLOOR));
2723             ewitab           = _mm_slli_epi32(ewitab,2);
2724             ewtabF           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
2725             ewtabD           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,1) );
2726             ewtabV           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,2) );
2727             ewtabFn          = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,3) );
2728             GMX_MM256_FULLTRANSPOSE4_PD(ewtabF,ewtabD,ewtabV,ewtabFn);
2729             felec            = _mm256_add_pd(ewtabF,_mm256_mul_pd(eweps,ewtabD));
2730             velec            = _mm256_sub_pd(ewtabV,_mm256_mul_pd(_mm256_mul_pd(ewtabhalfspace,eweps),_mm256_add_pd(ewtabF,felec)));
2731             velec            = _mm256_mul_pd(qq21,_mm256_sub_pd(rinv21,velec));
2732             felec            = _mm256_mul_pd(_mm256_mul_pd(qq21,rinv21),_mm256_sub_pd(rinvsq21,felec));
2733
2734             d                = _mm256_sub_pd(r21,rswitch);
2735             d                = _mm256_max_pd(d,_mm256_setzero_pd());
2736             d2               = _mm256_mul_pd(d,d);
2737             sw               = _mm256_add_pd(one,_mm256_mul_pd(d2,_mm256_mul_pd(d,_mm256_add_pd(swV3,_mm256_mul_pd(d,_mm256_add_pd(swV4,_mm256_mul_pd(d,swV5)))))));
2738
2739             dsw              = _mm256_mul_pd(d2,_mm256_add_pd(swF2,_mm256_mul_pd(d,_mm256_add_pd(swF3,_mm256_mul_pd(d,swF4)))));
2740
2741             /* Evaluate switch function */
2742             /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
2743             felec            = _mm256_sub_pd( _mm256_mul_pd(felec,sw) , _mm256_mul_pd(rinv21,_mm256_mul_pd(velec,dsw)) );
2744             cutoff_mask      = _mm256_cmp_pd(rsq21,rcutoff2,_CMP_LT_OQ);
2745
2746             fscal            = felec;
2747
2748             fscal            = _mm256_and_pd(fscal,cutoff_mask);
2749
2750             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
2751
2752             /* Calculate temporary vectorial force */
2753             tx               = _mm256_mul_pd(fscal,dx21);
2754             ty               = _mm256_mul_pd(fscal,dy21);
2755             tz               = _mm256_mul_pd(fscal,dz21);
2756
2757             /* Update vectorial force */
2758             fix2             = _mm256_add_pd(fix2,tx);
2759             fiy2             = _mm256_add_pd(fiy2,ty);
2760             fiz2             = _mm256_add_pd(fiz2,tz);
2761
2762             fjx1             = _mm256_add_pd(fjx1,tx);
2763             fjy1             = _mm256_add_pd(fjy1,ty);
2764             fjz1             = _mm256_add_pd(fjz1,tz);
2765
2766             }
2767
2768             /**************************
2769              * CALCULATE INTERACTIONS *
2770              **************************/
2771
2772             if (gmx_mm256_any_lt(rsq22,rcutoff2))
2773             {
2774
2775             r22              = _mm256_mul_pd(rsq22,rinv22);
2776             r22              = _mm256_andnot_pd(dummy_mask,r22);
2777
2778             /* EWALD ELECTROSTATICS */
2779
2780             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
2781             ewrt             = _mm256_mul_pd(r22,ewtabscale);
2782             ewitab           = _mm256_cvttpd_epi32(ewrt);
2783             eweps            = _mm256_sub_pd(ewrt,_mm256_round_pd(ewrt, _MM_FROUND_FLOOR));
2784             ewitab           = _mm_slli_epi32(ewitab,2);
2785             ewtabF           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
2786             ewtabD           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,1) );
2787             ewtabV           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,2) );
2788             ewtabFn          = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,3) );
2789             GMX_MM256_FULLTRANSPOSE4_PD(ewtabF,ewtabD,ewtabV,ewtabFn);
2790             felec            = _mm256_add_pd(ewtabF,_mm256_mul_pd(eweps,ewtabD));
2791             velec            = _mm256_sub_pd(ewtabV,_mm256_mul_pd(_mm256_mul_pd(ewtabhalfspace,eweps),_mm256_add_pd(ewtabF,felec)));
2792             velec            = _mm256_mul_pd(qq22,_mm256_sub_pd(rinv22,velec));
2793             felec            = _mm256_mul_pd(_mm256_mul_pd(qq22,rinv22),_mm256_sub_pd(rinvsq22,felec));
2794
2795             d                = _mm256_sub_pd(r22,rswitch);
2796             d                = _mm256_max_pd(d,_mm256_setzero_pd());
2797             d2               = _mm256_mul_pd(d,d);
2798             sw               = _mm256_add_pd(one,_mm256_mul_pd(d2,_mm256_mul_pd(d,_mm256_add_pd(swV3,_mm256_mul_pd(d,_mm256_add_pd(swV4,_mm256_mul_pd(d,swV5)))))));
2799
2800             dsw              = _mm256_mul_pd(d2,_mm256_add_pd(swF2,_mm256_mul_pd(d,_mm256_add_pd(swF3,_mm256_mul_pd(d,swF4)))));
2801
2802             /* Evaluate switch function */
2803             /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
2804             felec            = _mm256_sub_pd( _mm256_mul_pd(felec,sw) , _mm256_mul_pd(rinv22,_mm256_mul_pd(velec,dsw)) );
2805             cutoff_mask      = _mm256_cmp_pd(rsq22,rcutoff2,_CMP_LT_OQ);
2806
2807             fscal            = felec;
2808
2809             fscal            = _mm256_and_pd(fscal,cutoff_mask);
2810
2811             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
2812
2813             /* Calculate temporary vectorial force */
2814             tx               = _mm256_mul_pd(fscal,dx22);
2815             ty               = _mm256_mul_pd(fscal,dy22);
2816             tz               = _mm256_mul_pd(fscal,dz22);
2817
2818             /* Update vectorial force */
2819             fix2             = _mm256_add_pd(fix2,tx);
2820             fiy2             = _mm256_add_pd(fiy2,ty);
2821             fiz2             = _mm256_add_pd(fiz2,tz);
2822
2823             fjx2             = _mm256_add_pd(fjx2,tx);
2824             fjy2             = _mm256_add_pd(fjy2,ty);
2825             fjz2             = _mm256_add_pd(fjz2,tz);
2826
2827             }
2828
2829             /**************************
2830              * CALCULATE INTERACTIONS *
2831              **************************/
2832
2833             if (gmx_mm256_any_lt(rsq23,rcutoff2))
2834             {
2835
2836             r23              = _mm256_mul_pd(rsq23,rinv23);
2837             r23              = _mm256_andnot_pd(dummy_mask,r23);
2838
2839             /* EWALD ELECTROSTATICS */
2840
2841             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
2842             ewrt             = _mm256_mul_pd(r23,ewtabscale);
2843             ewitab           = _mm256_cvttpd_epi32(ewrt);
2844             eweps            = _mm256_sub_pd(ewrt,_mm256_round_pd(ewrt, _MM_FROUND_FLOOR));
2845             ewitab           = _mm_slli_epi32(ewitab,2);
2846             ewtabF           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
2847             ewtabD           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,1) );
2848             ewtabV           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,2) );
2849             ewtabFn          = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,3) );
2850             GMX_MM256_FULLTRANSPOSE4_PD(ewtabF,ewtabD,ewtabV,ewtabFn);
2851             felec            = _mm256_add_pd(ewtabF,_mm256_mul_pd(eweps,ewtabD));
2852             velec            = _mm256_sub_pd(ewtabV,_mm256_mul_pd(_mm256_mul_pd(ewtabhalfspace,eweps),_mm256_add_pd(ewtabF,felec)));
2853             velec            = _mm256_mul_pd(qq23,_mm256_sub_pd(rinv23,velec));
2854             felec            = _mm256_mul_pd(_mm256_mul_pd(qq23,rinv23),_mm256_sub_pd(rinvsq23,felec));
2855
2856             d                = _mm256_sub_pd(r23,rswitch);
2857             d                = _mm256_max_pd(d,_mm256_setzero_pd());
2858             d2               = _mm256_mul_pd(d,d);
2859             sw               = _mm256_add_pd(one,_mm256_mul_pd(d2,_mm256_mul_pd(d,_mm256_add_pd(swV3,_mm256_mul_pd(d,_mm256_add_pd(swV4,_mm256_mul_pd(d,swV5)))))));
2860
2861             dsw              = _mm256_mul_pd(d2,_mm256_add_pd(swF2,_mm256_mul_pd(d,_mm256_add_pd(swF3,_mm256_mul_pd(d,swF4)))));
2862
2863             /* Evaluate switch function */
2864             /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
2865             felec            = _mm256_sub_pd( _mm256_mul_pd(felec,sw) , _mm256_mul_pd(rinv23,_mm256_mul_pd(velec,dsw)) );
2866             cutoff_mask      = _mm256_cmp_pd(rsq23,rcutoff2,_CMP_LT_OQ);
2867
2868             fscal            = felec;
2869
2870             fscal            = _mm256_and_pd(fscal,cutoff_mask);
2871
2872             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
2873
2874             /* Calculate temporary vectorial force */
2875             tx               = _mm256_mul_pd(fscal,dx23);
2876             ty               = _mm256_mul_pd(fscal,dy23);
2877             tz               = _mm256_mul_pd(fscal,dz23);
2878
2879             /* Update vectorial force */
2880             fix2             = _mm256_add_pd(fix2,tx);
2881             fiy2             = _mm256_add_pd(fiy2,ty);
2882             fiz2             = _mm256_add_pd(fiz2,tz);
2883
2884             fjx3             = _mm256_add_pd(fjx3,tx);
2885             fjy3             = _mm256_add_pd(fjy3,ty);
2886             fjz3             = _mm256_add_pd(fjz3,tz);
2887
2888             }
2889
2890             /**************************
2891              * CALCULATE INTERACTIONS *
2892              **************************/
2893
2894             if (gmx_mm256_any_lt(rsq31,rcutoff2))
2895             {
2896
2897             r31              = _mm256_mul_pd(rsq31,rinv31);
2898             r31              = _mm256_andnot_pd(dummy_mask,r31);
2899
2900             /* EWALD ELECTROSTATICS */
2901
2902             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
2903             ewrt             = _mm256_mul_pd(r31,ewtabscale);
2904             ewitab           = _mm256_cvttpd_epi32(ewrt);
2905             eweps            = _mm256_sub_pd(ewrt,_mm256_round_pd(ewrt, _MM_FROUND_FLOOR));
2906             ewitab           = _mm_slli_epi32(ewitab,2);
2907             ewtabF           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
2908             ewtabD           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,1) );
2909             ewtabV           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,2) );
2910             ewtabFn          = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,3) );
2911             GMX_MM256_FULLTRANSPOSE4_PD(ewtabF,ewtabD,ewtabV,ewtabFn);
2912             felec            = _mm256_add_pd(ewtabF,_mm256_mul_pd(eweps,ewtabD));
2913             velec            = _mm256_sub_pd(ewtabV,_mm256_mul_pd(_mm256_mul_pd(ewtabhalfspace,eweps),_mm256_add_pd(ewtabF,felec)));
2914             velec            = _mm256_mul_pd(qq31,_mm256_sub_pd(rinv31,velec));
2915             felec            = _mm256_mul_pd(_mm256_mul_pd(qq31,rinv31),_mm256_sub_pd(rinvsq31,felec));
2916
2917             d                = _mm256_sub_pd(r31,rswitch);
2918             d                = _mm256_max_pd(d,_mm256_setzero_pd());
2919             d2               = _mm256_mul_pd(d,d);
2920             sw               = _mm256_add_pd(one,_mm256_mul_pd(d2,_mm256_mul_pd(d,_mm256_add_pd(swV3,_mm256_mul_pd(d,_mm256_add_pd(swV4,_mm256_mul_pd(d,swV5)))))));
2921
2922             dsw              = _mm256_mul_pd(d2,_mm256_add_pd(swF2,_mm256_mul_pd(d,_mm256_add_pd(swF3,_mm256_mul_pd(d,swF4)))));
2923
2924             /* Evaluate switch function */
2925             /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
2926             felec            = _mm256_sub_pd( _mm256_mul_pd(felec,sw) , _mm256_mul_pd(rinv31,_mm256_mul_pd(velec,dsw)) );
2927             cutoff_mask      = _mm256_cmp_pd(rsq31,rcutoff2,_CMP_LT_OQ);
2928
2929             fscal            = felec;
2930
2931             fscal            = _mm256_and_pd(fscal,cutoff_mask);
2932
2933             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
2934
2935             /* Calculate temporary vectorial force */
2936             tx               = _mm256_mul_pd(fscal,dx31);
2937             ty               = _mm256_mul_pd(fscal,dy31);
2938             tz               = _mm256_mul_pd(fscal,dz31);
2939
2940             /* Update vectorial force */
2941             fix3             = _mm256_add_pd(fix3,tx);
2942             fiy3             = _mm256_add_pd(fiy3,ty);
2943             fiz3             = _mm256_add_pd(fiz3,tz);
2944
2945             fjx1             = _mm256_add_pd(fjx1,tx);
2946             fjy1             = _mm256_add_pd(fjy1,ty);
2947             fjz1             = _mm256_add_pd(fjz1,tz);
2948
2949             }
2950
2951             /**************************
2952              * CALCULATE INTERACTIONS *
2953              **************************/
2954
2955             if (gmx_mm256_any_lt(rsq32,rcutoff2))
2956             {
2957
2958             r32              = _mm256_mul_pd(rsq32,rinv32);
2959             r32              = _mm256_andnot_pd(dummy_mask,r32);
2960
2961             /* EWALD ELECTROSTATICS */
2962
2963             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
2964             ewrt             = _mm256_mul_pd(r32,ewtabscale);
2965             ewitab           = _mm256_cvttpd_epi32(ewrt);
2966             eweps            = _mm256_sub_pd(ewrt,_mm256_round_pd(ewrt, _MM_FROUND_FLOOR));
2967             ewitab           = _mm_slli_epi32(ewitab,2);
2968             ewtabF           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
2969             ewtabD           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,1) );
2970             ewtabV           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,2) );
2971             ewtabFn          = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,3) );
2972             GMX_MM256_FULLTRANSPOSE4_PD(ewtabF,ewtabD,ewtabV,ewtabFn);
2973             felec            = _mm256_add_pd(ewtabF,_mm256_mul_pd(eweps,ewtabD));
2974             velec            = _mm256_sub_pd(ewtabV,_mm256_mul_pd(_mm256_mul_pd(ewtabhalfspace,eweps),_mm256_add_pd(ewtabF,felec)));
2975             velec            = _mm256_mul_pd(qq32,_mm256_sub_pd(rinv32,velec));
2976             felec            = _mm256_mul_pd(_mm256_mul_pd(qq32,rinv32),_mm256_sub_pd(rinvsq32,felec));
2977
2978             d                = _mm256_sub_pd(r32,rswitch);
2979             d                = _mm256_max_pd(d,_mm256_setzero_pd());
2980             d2               = _mm256_mul_pd(d,d);
2981             sw               = _mm256_add_pd(one,_mm256_mul_pd(d2,_mm256_mul_pd(d,_mm256_add_pd(swV3,_mm256_mul_pd(d,_mm256_add_pd(swV4,_mm256_mul_pd(d,swV5)))))));
2982
2983             dsw              = _mm256_mul_pd(d2,_mm256_add_pd(swF2,_mm256_mul_pd(d,_mm256_add_pd(swF3,_mm256_mul_pd(d,swF4)))));
2984
2985             /* Evaluate switch function */
2986             /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
2987             felec            = _mm256_sub_pd( _mm256_mul_pd(felec,sw) , _mm256_mul_pd(rinv32,_mm256_mul_pd(velec,dsw)) );
2988             cutoff_mask      = _mm256_cmp_pd(rsq32,rcutoff2,_CMP_LT_OQ);
2989
2990             fscal            = felec;
2991
2992             fscal            = _mm256_and_pd(fscal,cutoff_mask);
2993
2994             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
2995
2996             /* Calculate temporary vectorial force */
2997             tx               = _mm256_mul_pd(fscal,dx32);
2998             ty               = _mm256_mul_pd(fscal,dy32);
2999             tz               = _mm256_mul_pd(fscal,dz32);
3000
3001             /* Update vectorial force */
3002             fix3             = _mm256_add_pd(fix3,tx);
3003             fiy3             = _mm256_add_pd(fiy3,ty);
3004             fiz3             = _mm256_add_pd(fiz3,tz);
3005
3006             fjx2             = _mm256_add_pd(fjx2,tx);
3007             fjy2             = _mm256_add_pd(fjy2,ty);
3008             fjz2             = _mm256_add_pd(fjz2,tz);
3009
3010             }
3011
3012             /**************************
3013              * CALCULATE INTERACTIONS *
3014              **************************/
3015
3016             if (gmx_mm256_any_lt(rsq33,rcutoff2))
3017             {
3018
3019             r33              = _mm256_mul_pd(rsq33,rinv33);
3020             r33              = _mm256_andnot_pd(dummy_mask,r33);
3021
3022             /* EWALD ELECTROSTATICS */
3023
3024             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
3025             ewrt             = _mm256_mul_pd(r33,ewtabscale);
3026             ewitab           = _mm256_cvttpd_epi32(ewrt);
3027             eweps            = _mm256_sub_pd(ewrt,_mm256_round_pd(ewrt, _MM_FROUND_FLOOR));
3028             ewitab           = _mm_slli_epi32(ewitab,2);
3029             ewtabF           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
3030             ewtabD           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,1) );
3031             ewtabV           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,2) );
3032             ewtabFn          = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,3) );
3033             GMX_MM256_FULLTRANSPOSE4_PD(ewtabF,ewtabD,ewtabV,ewtabFn);
3034             felec            = _mm256_add_pd(ewtabF,_mm256_mul_pd(eweps,ewtabD));
3035             velec            = _mm256_sub_pd(ewtabV,_mm256_mul_pd(_mm256_mul_pd(ewtabhalfspace,eweps),_mm256_add_pd(ewtabF,felec)));
3036             velec            = _mm256_mul_pd(qq33,_mm256_sub_pd(rinv33,velec));
3037             felec            = _mm256_mul_pd(_mm256_mul_pd(qq33,rinv33),_mm256_sub_pd(rinvsq33,felec));
3038
3039             d                = _mm256_sub_pd(r33,rswitch);
3040             d                = _mm256_max_pd(d,_mm256_setzero_pd());
3041             d2               = _mm256_mul_pd(d,d);
3042             sw               = _mm256_add_pd(one,_mm256_mul_pd(d2,_mm256_mul_pd(d,_mm256_add_pd(swV3,_mm256_mul_pd(d,_mm256_add_pd(swV4,_mm256_mul_pd(d,swV5)))))));
3043
3044             dsw              = _mm256_mul_pd(d2,_mm256_add_pd(swF2,_mm256_mul_pd(d,_mm256_add_pd(swF3,_mm256_mul_pd(d,swF4)))));
3045
3046             /* Evaluate switch function */
3047             /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
3048             felec            = _mm256_sub_pd( _mm256_mul_pd(felec,sw) , _mm256_mul_pd(rinv33,_mm256_mul_pd(velec,dsw)) );
3049             cutoff_mask      = _mm256_cmp_pd(rsq33,rcutoff2,_CMP_LT_OQ);
3050
3051             fscal            = felec;
3052
3053             fscal            = _mm256_and_pd(fscal,cutoff_mask);
3054
3055             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
3056
3057             /* Calculate temporary vectorial force */
3058             tx               = _mm256_mul_pd(fscal,dx33);
3059             ty               = _mm256_mul_pd(fscal,dy33);
3060             tz               = _mm256_mul_pd(fscal,dz33);
3061
3062             /* Update vectorial force */
3063             fix3             = _mm256_add_pd(fix3,tx);
3064             fiy3             = _mm256_add_pd(fiy3,ty);
3065             fiz3             = _mm256_add_pd(fiz3,tz);
3066
3067             fjx3             = _mm256_add_pd(fjx3,tx);
3068             fjy3             = _mm256_add_pd(fjy3,ty);
3069             fjz3             = _mm256_add_pd(fjz3,tz);
3070
3071             }
3072
3073             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
3074             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
3075             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
3076             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
3077
3078             gmx_mm256_decrement_3rvec_4ptr_swizzle_pd(fjptrA+DIM,fjptrB+DIM,fjptrC+DIM,fjptrD+DIM,
3079                                                       fjx1,fjy1,fjz1,fjx2,fjy2,fjz2,fjx3,fjy3,fjz3);
3080
3081             /* Inner loop uses 567 flops */
3082         }
3083
3084         /* End of innermost loop */
3085
3086         gmx_mm256_update_iforce_3atom_swizzle_pd(fix1,fiy1,fiz1,fix2,fiy2,fiz2,fix3,fiy3,fiz3,
3087                                                  f+i_coord_offset+DIM,fshift+i_shift_offset);
3088
3089         /* Increment number of inner iterations */
3090         inneriter                  += j_index_end - j_index_start;
3091
3092         /* Outer loop uses 18 flops */
3093     }
3094
3095     /* Increment number of outer iterations */
3096     outeriter        += nri;
3097
3098     /* Update outer/inner flops */
3099
3100     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_W4W4_F,outeriter*18 + inneriter*567);
3101 }