src/gromacs/gmxlib/nonbonded/nb_kernel_avx_256_double/nb_kernel_ElecEwSw_VdwNone_GeomW3W3_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 #include "config.h"
  39
  40 #include <math.h>
  41
  42 #include "../nb_kernel.h"
  43 #include "gromacs/legacyheaders/types/simple.h"
  44 #include "gromacs/math/vec.h"
  45 #include "gromacs/legacyheaders/nrnb.h"
  46
  47 #include "gromacs/simd/math_x86_avx_256_double.h"
  48 #include "kernelutil_x86_avx_256_double.h"
  49
  50 /*
  51  * Gromacs nonbonded kernel:   nb_kernel_ElecEwSw_VdwNone_GeomW3W3_VF_avx_256_double
  52  * Electrostatics interaction: Ewald
  53  * VdW interaction:            None
  54  * Geometry:                   Water3-Water3
  55  * Calculate force/pot:        PotentialAndForce
  56  */
  57 void
  58 nb_kernel_ElecEwSw_VdwNone_GeomW3W3_VF_avx_256_double
  59                     (t_nblist                    * gmx_restrict       nlist,
  60                      rvec                        * gmx_restrict          xx,
  61                      rvec                        * gmx_restrict          ff,
  62                      t_forcerec                  * gmx_restrict          fr,
  63                      t_mdatoms                   * gmx_restrict     mdatoms,
  64                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
  65                      t_nrnb                      * gmx_restrict        nrnb)
  66 {
  67     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
  68      * just 0 for non-waters.
  69      * Suffixes A,B,C,D refer to j loop unrolling done with AVX, e.g. for the four different
  70      * jnr indices corresponding to data put in the four positions in the SIMD register.
  71      */
  72     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
  73     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
  74     int              jnrA,jnrB,jnrC,jnrD;
  75     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
  76     int              jnrlistE,jnrlistF,jnrlistG,jnrlistH;
  77     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
  78     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
  79     real             rcutoff_scalar;
  80     real             *shiftvec,*fshift,*x,*f;
  81     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD;
  82     real             scratch[4*DIM];
  83     __m256d          tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
  84     real *           vdwioffsetptr0;
  85     __m256d          ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
  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     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
  91     __m256d          jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
  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     __m256d          dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
  97     __m256d          dx01,dy01,dz01,rsq01,rinv01,rinvsq01,r01,qq01,c6_01,c12_01;
  98     __m256d          dx02,dy02,dz02,rsq02,rinv02,rinvsq02,r02,qq02,c6_02,c12_02;
  99     __m256d          dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
 100     __m256d          dx11,dy11,dz11,rsq11,rinv11,rinvsq11,r11,qq11,c6_11,c12_11;
 101     __m256d          dx12,dy12,dz12,rsq12,rinv12,rinvsq12,r12,qq12,c6_12,c12_12;
 102     __m256d          dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
 103     __m256d          dx21,dy21,dz21,rsq21,rinv21,rinvsq21,r21,qq21,c6_21,c12_21;
 104     __m256d          dx22,dy22,dz22,rsq22,rinv22,rinvsq22,r22,qq22,c6_22,c12_22;
 105     __m256d          velec,felec,velecsum,facel,crf,krf,krf2;
 106     real             *charge;
 107     __m128i          ewitab;
 108     __m256d          ewtabscale,eweps,sh_ewald,ewrt,ewtabhalfspace,ewtabF,ewtabFn,ewtabD,ewtabV;
 109     __m256d          beta,beta2,beta3,zeta2,pmecorrF,pmecorrV,rinv3;
 110     real             *ewtab;
 111     __m256d          rswitch,swV3,swV4,swV5,swF2,swF3,swF4,d,d2,sw,dsw;
 112     real             rswitch_scalar,d_scalar;
 113     __m256d          dummy_mask,cutoff_mask;
 114     __m128           tmpmask0,tmpmask1;
 115     __m256d          signbit = _mm256_castsi256_pd( _mm256_set1_epi32(0x80000000) );
 116     __m256d          one     = _mm256_set1_pd(1.0);
 117     __m256d          two     = _mm256_set1_pd(2.0);
 118     x                = xx[0];
 119     f                = ff[0];
 120
 121     nri              = nlist->nri;
 122     iinr             = nlist->iinr;
 123     jindex           = nlist->jindex;
 124     jjnr             = nlist->jjnr;
 125     shiftidx         = nlist->shift;
 126     gid              = nlist->gid;
 127     shiftvec         = fr->shift_vec[0];
 128     fshift           = fr->fshift[0];
 129     facel            = _mm256_set1_pd(fr->epsfac);
 130     charge           = mdatoms->chargeA;
 131
 132     sh_ewald         = _mm256_set1_pd(fr->ic->sh_ewald);
 133     beta             = _mm256_set1_pd(fr->ic->ewaldcoeff_q);
 134     beta2            = _mm256_mul_pd(beta,beta);
 135     beta3            = _mm256_mul_pd(beta,beta2);
 136
 137     ewtab            = fr->ic->tabq_coul_FDV0;
 138     ewtabscale       = _mm256_set1_pd(fr->ic->tabq_scale);
 139     ewtabhalfspace   = _mm256_set1_pd(0.5/fr->ic->tabq_scale);
 140
 141     /* Setup water-specific parameters */
 142     inr              = nlist->iinr[0];
 143     iq0              = _mm256_mul_pd(facel,_mm256_set1_pd(charge[inr+0]));
 144     iq1              = _mm256_mul_pd(facel,_mm256_set1_pd(charge[inr+1]));
 145     iq2              = _mm256_mul_pd(facel,_mm256_set1_pd(charge[inr+2]));
 146
 147     jq0              = _mm256_set1_pd(charge[inr+0]);
 148     jq1              = _mm256_set1_pd(charge[inr+1]);
 149     jq2              = _mm256_set1_pd(charge[inr+2]);
 150     qq00             = _mm256_mul_pd(iq0,jq0);
 151     qq01             = _mm256_mul_pd(iq0,jq1);
 152     qq02             = _mm256_mul_pd(iq0,jq2);
 153     qq10             = _mm256_mul_pd(iq1,jq0);
 154     qq11             = _mm256_mul_pd(iq1,jq1);
 155     qq12             = _mm256_mul_pd(iq1,jq2);
 156     qq20             = _mm256_mul_pd(iq2,jq0);
 157     qq21             = _mm256_mul_pd(iq2,jq1);
 158     qq22             = _mm256_mul_pd(iq2,jq2);
 159
 160     /* When we use explicit cutoffs the value must be identical for elec and VdW, so use elec as an arbitrary choice */
 161     rcutoff_scalar   = fr->rcoulomb;
 162     rcutoff          = _mm256_set1_pd(rcutoff_scalar);
 163     rcutoff2         = _mm256_mul_pd(rcutoff,rcutoff);
 164
 165     rswitch_scalar   = fr->rcoulomb_switch;
 166     rswitch          = _mm256_set1_pd(rswitch_scalar);
 167     /* Setup switch parameters */
 168     d_scalar         = rcutoff_scalar-rswitch_scalar;
 169     d                = _mm256_set1_pd(d_scalar);
 170     swV3             = _mm256_set1_pd(-10.0/(d_scalar*d_scalar*d_scalar));
 171     swV4             = _mm256_set1_pd( 15.0/(d_scalar*d_scalar*d_scalar*d_scalar));
 172     swV5             = _mm256_set1_pd( -6.0/(d_scalar*d_scalar*d_scalar*d_scalar*d_scalar));
 173     swF2             = _mm256_set1_pd(-30.0/(d_scalar*d_scalar*d_scalar));
 174     swF3             = _mm256_set1_pd( 60.0/(d_scalar*d_scalar*d_scalar*d_scalar));
 175     swF4             = _mm256_set1_pd(-30.0/(d_scalar*d_scalar*d_scalar*d_scalar*d_scalar));
 176
 177     /* Avoid stupid compiler warnings */
 178     jnrA = jnrB = jnrC = jnrD = 0;
 179     j_coord_offsetA = 0;
 180     j_coord_offsetB = 0;
 181     j_coord_offsetC = 0;
 182     j_coord_offsetD = 0;
 183
 184     outeriter        = 0;
 185     inneriter        = 0;
 186
 187     for(iidx=0;iidx<4*DIM;iidx++)
 188     {
 189         scratch[iidx] = 0.0;
 190     }
 191
 192     /* Start outer loop over neighborlists */
 193     for(iidx=0; iidx<nri; iidx++)
 194     {
 195         /* Load shift vector for this list */
 196         i_shift_offset   = DIM*shiftidx[iidx];
 197
 198         /* Load limits for loop over neighbors */
 199         j_index_start    = jindex[iidx];
 200         j_index_end      = jindex[iidx+1];
 201
 202         /* Get outer coordinate index */
 203         inr              = iinr[iidx];
 204         i_coord_offset   = DIM*inr;
 205
 206         /* Load i particle coords and add shift vector */
 207         gmx_mm256_load_shift_and_3rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset,
 208                                                     &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2);
 209
 210         fix0             = _mm256_setzero_pd();
 211         fiy0             = _mm256_setzero_pd();
 212         fiz0             = _mm256_setzero_pd();
 213         fix1             = _mm256_setzero_pd();
 214         fiy1             = _mm256_setzero_pd();
 215         fiz1             = _mm256_setzero_pd();
 216         fix2             = _mm256_setzero_pd();
 217         fiy2             = _mm256_setzero_pd();
 218         fiz2             = _mm256_setzero_pd();
 219
 220         /* Reset potential sums */
 221         velecsum         = _mm256_setzero_pd();
 222
 223         /* Start inner kernel loop */
 224         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
 225         {
 226
 227             /* Get j neighbor index, and coordinate index */
 228             jnrA             = jjnr[jidx];
 229             jnrB             = jjnr[jidx+1];
 230             jnrC             = jjnr[jidx+2];
 231             jnrD             = jjnr[jidx+3];
 232             j_coord_offsetA  = DIM*jnrA;
 233             j_coord_offsetB  = DIM*jnrB;
 234             j_coord_offsetC  = DIM*jnrC;
 235             j_coord_offsetD  = DIM*jnrD;
 236
 237             /* load j atom coordinates */
 238             gmx_mm256_load_3rvec_4ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 239                                                  x+j_coord_offsetC,x+j_coord_offsetD,
 240                                               &jx0,&jy0,&jz0,&jx1,&jy1,&jz1,&jx2,&jy2,&jz2);
 241
 242             /* Calculate displacement vector */
 243             dx00             = _mm256_sub_pd(ix0,jx0);
 244             dy00             = _mm256_sub_pd(iy0,jy0);
 245             dz00             = _mm256_sub_pd(iz0,jz0);
 246             dx01             = _mm256_sub_pd(ix0,jx1);
 247             dy01             = _mm256_sub_pd(iy0,jy1);
 248             dz01             = _mm256_sub_pd(iz0,jz1);
 249             dx02             = _mm256_sub_pd(ix0,jx2);
 250             dy02             = _mm256_sub_pd(iy0,jy2);
 251             dz02             = _mm256_sub_pd(iz0,jz2);
 252             dx10             = _mm256_sub_pd(ix1,jx0);
 253             dy10             = _mm256_sub_pd(iy1,jy0);
 254             dz10             = _mm256_sub_pd(iz1,jz0);
 255             dx11             = _mm256_sub_pd(ix1,jx1);
 256             dy11             = _mm256_sub_pd(iy1,jy1);
 257             dz11             = _mm256_sub_pd(iz1,jz1);
 258             dx12             = _mm256_sub_pd(ix1,jx2);
 259             dy12             = _mm256_sub_pd(iy1,jy2);
 260             dz12             = _mm256_sub_pd(iz1,jz2);
 261             dx20             = _mm256_sub_pd(ix2,jx0);
 262             dy20             = _mm256_sub_pd(iy2,jy0);
 263             dz20             = _mm256_sub_pd(iz2,jz0);
 264             dx21             = _mm256_sub_pd(ix2,jx1);
 265             dy21             = _mm256_sub_pd(iy2,jy1);
 266             dz21             = _mm256_sub_pd(iz2,jz1);
 267             dx22             = _mm256_sub_pd(ix2,jx2);
 268             dy22             = _mm256_sub_pd(iy2,jy2);
 269             dz22             = _mm256_sub_pd(iz2,jz2);
 270
 271             /* Calculate squared distance and things based on it */
 272             rsq00            = gmx_mm256_calc_rsq_pd(dx00,dy00,dz00);
 273             rsq01            = gmx_mm256_calc_rsq_pd(dx01,dy01,dz01);
 274             rsq02            = gmx_mm256_calc_rsq_pd(dx02,dy02,dz02);
 275             rsq10            = gmx_mm256_calc_rsq_pd(dx10,dy10,dz10);
 276             rsq11            = gmx_mm256_calc_rsq_pd(dx11,dy11,dz11);
 277             rsq12            = gmx_mm256_calc_rsq_pd(dx12,dy12,dz12);
 278             rsq20            = gmx_mm256_calc_rsq_pd(dx20,dy20,dz20);
 279             rsq21            = gmx_mm256_calc_rsq_pd(dx21,dy21,dz21);
 280             rsq22            = gmx_mm256_calc_rsq_pd(dx22,dy22,dz22);
 281
 282             rinv00           = gmx_mm256_invsqrt_pd(rsq00);
 283             rinv01           = gmx_mm256_invsqrt_pd(rsq01);
 284             rinv02           = gmx_mm256_invsqrt_pd(rsq02);
 285             rinv10           = gmx_mm256_invsqrt_pd(rsq10);
 286             rinv11           = gmx_mm256_invsqrt_pd(rsq11);
 287             rinv12           = gmx_mm256_invsqrt_pd(rsq12);
 288             rinv20           = gmx_mm256_invsqrt_pd(rsq20);
 289             rinv21           = gmx_mm256_invsqrt_pd(rsq21);
 290             rinv22           = gmx_mm256_invsqrt_pd(rsq22);
 291
 292             rinvsq00         = _mm256_mul_pd(rinv00,rinv00);
 293             rinvsq01         = _mm256_mul_pd(rinv01,rinv01);
 294             rinvsq02         = _mm256_mul_pd(rinv02,rinv02);
 295             rinvsq10         = _mm256_mul_pd(rinv10,rinv10);
 296             rinvsq11         = _mm256_mul_pd(rinv11,rinv11);
 297             rinvsq12         = _mm256_mul_pd(rinv12,rinv12);
 298             rinvsq20         = _mm256_mul_pd(rinv20,rinv20);
 299             rinvsq21         = _mm256_mul_pd(rinv21,rinv21);
 300             rinvsq22         = _mm256_mul_pd(rinv22,rinv22);
 301
 302             fjx0             = _mm256_setzero_pd();
 303             fjy0             = _mm256_setzero_pd();
 304             fjz0             = _mm256_setzero_pd();
 305             fjx1             = _mm256_setzero_pd();
 306             fjy1             = _mm256_setzero_pd();
 307             fjz1             = _mm256_setzero_pd();
 308             fjx2             = _mm256_setzero_pd();
 309             fjy2             = _mm256_setzero_pd();
 310             fjz2             = _mm256_setzero_pd();
 311
 312             /**************************
 313              * CALCULATE INTERACTIONS *
 314              **************************/
 315
 316             if (gmx_mm256_any_lt(rsq00,rcutoff2))
 317             {
 318
 319             r00              = _mm256_mul_pd(rsq00,rinv00);
 320
 321             /* EWALD ELECTROSTATICS */
 322
 323             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 324             ewrt             = _mm256_mul_pd(r00,ewtabscale);
 325             ewitab           = _mm256_cvttpd_epi32(ewrt);
 326             eweps            = _mm256_sub_pd(ewrt,_mm256_round_pd(ewrt, _MM_FROUND_FLOOR));
 327             ewitab           = _mm_slli_epi32(ewitab,2);
 328             ewtabF           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
 329             ewtabD           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,1) );
 330             ewtabV           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,2) );
 331             ewtabFn          = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,3) );
 332             GMX_MM256_FULLTRANSPOSE4_PD(ewtabF,ewtabD,ewtabV,ewtabFn);
 333             felec            = _mm256_add_pd(ewtabF,_mm256_mul_pd(eweps,ewtabD));
 334             velec            = _mm256_sub_pd(ewtabV,_mm256_mul_pd(_mm256_mul_pd(ewtabhalfspace,eweps),_mm256_add_pd(ewtabF,felec)));
 335             velec            = _mm256_mul_pd(qq00,_mm256_sub_pd(rinv00,velec));
 336             felec            = _mm256_mul_pd(_mm256_mul_pd(qq00,rinv00),_mm256_sub_pd(rinvsq00,felec));
 337
 338             d                = _mm256_sub_pd(r00,rswitch);
 339             d                = _mm256_max_pd(d,_mm256_setzero_pd());
 340             d2               = _mm256_mul_pd(d,d);
 341             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)))))));
 342
 343             dsw              = _mm256_mul_pd(d2,_mm256_add_pd(swF2,_mm256_mul_pd(d,_mm256_add_pd(swF3,_mm256_mul_pd(d,swF4)))));
 344
 345             /* Evaluate switch function */
 346             /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
 347             felec            = _mm256_sub_pd( _mm256_mul_pd(felec,sw) , _mm256_mul_pd(rinv00,_mm256_mul_pd(velec,dsw)) );
 348             velec            = _mm256_mul_pd(velec,sw);
 349             cutoff_mask      = _mm256_cmp_pd(rsq00,rcutoff2,_CMP_LT_OQ);
 350
 351             /* Update potential sum for this i atom from the interaction with this j atom. */
 352             velec            = _mm256_and_pd(velec,cutoff_mask);
 353             velecsum         = _mm256_add_pd(velecsum,velec);
 354
 355             fscal            = felec;
 356
 357             fscal            = _mm256_and_pd(fscal,cutoff_mask);
 358
 359             /* Calculate temporary vectorial force */
 360             tx               = _mm256_mul_pd(fscal,dx00);
 361             ty               = _mm256_mul_pd(fscal,dy00);
 362             tz               = _mm256_mul_pd(fscal,dz00);
 363
 364             /* Update vectorial force */
 365             fix0             = _mm256_add_pd(fix0,tx);
 366             fiy0             = _mm256_add_pd(fiy0,ty);
 367             fiz0             = _mm256_add_pd(fiz0,tz);
 368
 369             fjx0             = _mm256_add_pd(fjx0,tx);
 370             fjy0             = _mm256_add_pd(fjy0,ty);
 371             fjz0             = _mm256_add_pd(fjz0,tz);
 372
 373             }
 374
 375             /**************************
 376              * CALCULATE INTERACTIONS *
 377              **************************/
 378
 379             if (gmx_mm256_any_lt(rsq01,rcutoff2))
 380             {
 381
 382             r01              = _mm256_mul_pd(rsq01,rinv01);
 383
 384             /* EWALD ELECTROSTATICS */
 385
 386             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 387             ewrt             = _mm256_mul_pd(r01,ewtabscale);
 388             ewitab           = _mm256_cvttpd_epi32(ewrt);
 389             eweps            = _mm256_sub_pd(ewrt,_mm256_round_pd(ewrt, _MM_FROUND_FLOOR));
 390             ewitab           = _mm_slli_epi32(ewitab,2);
 391             ewtabF           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
 392             ewtabD           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,1) );
 393             ewtabV           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,2) );
 394             ewtabFn          = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,3) );
 395             GMX_MM256_FULLTRANSPOSE4_PD(ewtabF,ewtabD,ewtabV,ewtabFn);
 396             felec            = _mm256_add_pd(ewtabF,_mm256_mul_pd(eweps,ewtabD));
 397             velec            = _mm256_sub_pd(ewtabV,_mm256_mul_pd(_mm256_mul_pd(ewtabhalfspace,eweps),_mm256_add_pd(ewtabF,felec)));
 398             velec            = _mm256_mul_pd(qq01,_mm256_sub_pd(rinv01,velec));
 399             felec            = _mm256_mul_pd(_mm256_mul_pd(qq01,rinv01),_mm256_sub_pd(rinvsq01,felec));
 400
 401             d                = _mm256_sub_pd(r01,rswitch);
 402             d                = _mm256_max_pd(d,_mm256_setzero_pd());
 403             d2               = _mm256_mul_pd(d,d);
 404             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)))))));
 405
 406             dsw              = _mm256_mul_pd(d2,_mm256_add_pd(swF2,_mm256_mul_pd(d,_mm256_add_pd(swF3,_mm256_mul_pd(d,swF4)))));
 407
 408             /* Evaluate switch function */
 409             /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
 410             felec            = _mm256_sub_pd( _mm256_mul_pd(felec,sw) , _mm256_mul_pd(rinv01,_mm256_mul_pd(velec,dsw)) );
 411             velec            = _mm256_mul_pd(velec,sw);
 412             cutoff_mask      = _mm256_cmp_pd(rsq01,rcutoff2,_CMP_LT_OQ);
 413
 414             /* Update potential sum for this i atom from the interaction with this j atom. */
 415             velec            = _mm256_and_pd(velec,cutoff_mask);
 416             velecsum         = _mm256_add_pd(velecsum,velec);
 417
 418             fscal            = felec;
 419
 420             fscal            = _mm256_and_pd(fscal,cutoff_mask);
 421
 422             /* Calculate temporary vectorial force */
 423             tx               = _mm256_mul_pd(fscal,dx01);
 424             ty               = _mm256_mul_pd(fscal,dy01);
 425             tz               = _mm256_mul_pd(fscal,dz01);
 426
 427             /* Update vectorial force */
 428             fix0             = _mm256_add_pd(fix0,tx);
 429             fiy0             = _mm256_add_pd(fiy0,ty);
 430             fiz0             = _mm256_add_pd(fiz0,tz);
 431
 432             fjx1             = _mm256_add_pd(fjx1,tx);
 433             fjy1             = _mm256_add_pd(fjy1,ty);
 434             fjz1             = _mm256_add_pd(fjz1,tz);
 435
 436             }
 437
 438             /**************************
 439              * CALCULATE INTERACTIONS *
 440              **************************/
 441
 442             if (gmx_mm256_any_lt(rsq02,rcutoff2))
 443             {
 444
 445             r02              = _mm256_mul_pd(rsq02,rinv02);
 446
 447             /* EWALD ELECTROSTATICS */
 448
 449             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 450             ewrt             = _mm256_mul_pd(r02,ewtabscale);
 451             ewitab           = _mm256_cvttpd_epi32(ewrt);
 452             eweps            = _mm256_sub_pd(ewrt,_mm256_round_pd(ewrt, _MM_FROUND_FLOOR));
 453             ewitab           = _mm_slli_epi32(ewitab,2);
 454             ewtabF           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
 455             ewtabD           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,1) );
 456             ewtabV           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,2) );
 457             ewtabFn          = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,3) );
 458             GMX_MM256_FULLTRANSPOSE4_PD(ewtabF,ewtabD,ewtabV,ewtabFn);
 459             felec            = _mm256_add_pd(ewtabF,_mm256_mul_pd(eweps,ewtabD));
 460             velec            = _mm256_sub_pd(ewtabV,_mm256_mul_pd(_mm256_mul_pd(ewtabhalfspace,eweps),_mm256_add_pd(ewtabF,felec)));
 461             velec            = _mm256_mul_pd(qq02,_mm256_sub_pd(rinv02,velec));
 462             felec            = _mm256_mul_pd(_mm256_mul_pd(qq02,rinv02),_mm256_sub_pd(rinvsq02,felec));
 463
 464             d                = _mm256_sub_pd(r02,rswitch);
 465             d                = _mm256_max_pd(d,_mm256_setzero_pd());
 466             d2               = _mm256_mul_pd(d,d);
 467             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)))))));
 468
 469             dsw              = _mm256_mul_pd(d2,_mm256_add_pd(swF2,_mm256_mul_pd(d,_mm256_add_pd(swF3,_mm256_mul_pd(d,swF4)))));
 470
 471             /* Evaluate switch function */
 472             /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
 473             felec            = _mm256_sub_pd( _mm256_mul_pd(felec,sw) , _mm256_mul_pd(rinv02,_mm256_mul_pd(velec,dsw)) );
 474             velec            = _mm256_mul_pd(velec,sw);
 475             cutoff_mask      = _mm256_cmp_pd(rsq02,rcutoff2,_CMP_LT_OQ);
 476
 477             /* Update potential sum for this i atom from the interaction with this j atom. */
 478             velec            = _mm256_and_pd(velec,cutoff_mask);
 479             velecsum         = _mm256_add_pd(velecsum,velec);
 480
 481             fscal            = felec;
 482
 483             fscal            = _mm256_and_pd(fscal,cutoff_mask);
 484
 485             /* Calculate temporary vectorial force */
 486             tx               = _mm256_mul_pd(fscal,dx02);
 487             ty               = _mm256_mul_pd(fscal,dy02);
 488             tz               = _mm256_mul_pd(fscal,dz02);
 489
 490             /* Update vectorial force */
 491             fix0             = _mm256_add_pd(fix0,tx);
 492             fiy0             = _mm256_add_pd(fiy0,ty);
 493             fiz0             = _mm256_add_pd(fiz0,tz);
 494
 495             fjx2             = _mm256_add_pd(fjx2,tx);
 496             fjy2             = _mm256_add_pd(fjy2,ty);
 497             fjz2             = _mm256_add_pd(fjz2,tz);
 498
 499             }
 500
 501             /**************************
 502              * CALCULATE INTERACTIONS *
 503              **************************/
 504
 505             if (gmx_mm256_any_lt(rsq10,rcutoff2))
 506             {
 507
 508             r10              = _mm256_mul_pd(rsq10,rinv10);
 509
 510             /* EWALD ELECTROSTATICS */
 511
 512             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 513             ewrt             = _mm256_mul_pd(r10,ewtabscale);
 514             ewitab           = _mm256_cvttpd_epi32(ewrt);
 515             eweps            = _mm256_sub_pd(ewrt,_mm256_round_pd(ewrt, _MM_FROUND_FLOOR));
 516             ewitab           = _mm_slli_epi32(ewitab,2);
 517             ewtabF           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
 518             ewtabD           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,1) );
 519             ewtabV           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,2) );
 520             ewtabFn          = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,3) );
 521             GMX_MM256_FULLTRANSPOSE4_PD(ewtabF,ewtabD,ewtabV,ewtabFn);
 522             felec            = _mm256_add_pd(ewtabF,_mm256_mul_pd(eweps,ewtabD));
 523             velec            = _mm256_sub_pd(ewtabV,_mm256_mul_pd(_mm256_mul_pd(ewtabhalfspace,eweps),_mm256_add_pd(ewtabF,felec)));
 524             velec            = _mm256_mul_pd(qq10,_mm256_sub_pd(rinv10,velec));
 525             felec            = _mm256_mul_pd(_mm256_mul_pd(qq10,rinv10),_mm256_sub_pd(rinvsq10,felec));
 526
 527             d                = _mm256_sub_pd(r10,rswitch);
 528             d                = _mm256_max_pd(d,_mm256_setzero_pd());
 529             d2               = _mm256_mul_pd(d,d);
 530             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)))))));
 531
 532             dsw              = _mm256_mul_pd(d2,_mm256_add_pd(swF2,_mm256_mul_pd(d,_mm256_add_pd(swF3,_mm256_mul_pd(d,swF4)))));
 533
 534             /* Evaluate switch function */
 535             /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
 536             felec            = _mm256_sub_pd( _mm256_mul_pd(felec,sw) , _mm256_mul_pd(rinv10,_mm256_mul_pd(velec,dsw)) );
 537             velec            = _mm256_mul_pd(velec,sw);
 538             cutoff_mask      = _mm256_cmp_pd(rsq10,rcutoff2,_CMP_LT_OQ);
 539
 540             /* Update potential sum for this i atom from the interaction with this j atom. */
 541             velec            = _mm256_and_pd(velec,cutoff_mask);
 542             velecsum         = _mm256_add_pd(velecsum,velec);
 543
 544             fscal            = felec;
 545
 546             fscal            = _mm256_and_pd(fscal,cutoff_mask);
 547
 548             /* Calculate temporary vectorial force */
 549             tx               = _mm256_mul_pd(fscal,dx10);
 550             ty               = _mm256_mul_pd(fscal,dy10);
 551             tz               = _mm256_mul_pd(fscal,dz10);
 552
 553             /* Update vectorial force */
 554             fix1             = _mm256_add_pd(fix1,tx);
 555             fiy1             = _mm256_add_pd(fiy1,ty);
 556             fiz1             = _mm256_add_pd(fiz1,tz);
 557
 558             fjx0             = _mm256_add_pd(fjx0,tx);
 559             fjy0             = _mm256_add_pd(fjy0,ty);
 560             fjz0             = _mm256_add_pd(fjz0,tz);
 561
 562             }
 563
 564             /**************************
 565              * CALCULATE INTERACTIONS *
 566              **************************/
 567
 568             if (gmx_mm256_any_lt(rsq11,rcutoff2))
 569             {
 570
 571             r11              = _mm256_mul_pd(rsq11,rinv11);
 572
 573             /* EWALD ELECTROSTATICS */
 574
 575             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 576             ewrt             = _mm256_mul_pd(r11,ewtabscale);
 577             ewitab           = _mm256_cvttpd_epi32(ewrt);
 578             eweps            = _mm256_sub_pd(ewrt,_mm256_round_pd(ewrt, _MM_FROUND_FLOOR));
 579             ewitab           = _mm_slli_epi32(ewitab,2);
 580             ewtabF           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
 581             ewtabD           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,1) );
 582             ewtabV           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,2) );
 583             ewtabFn          = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,3) );
 584             GMX_MM256_FULLTRANSPOSE4_PD(ewtabF,ewtabD,ewtabV,ewtabFn);
 585             felec            = _mm256_add_pd(ewtabF,_mm256_mul_pd(eweps,ewtabD));
 586             velec            = _mm256_sub_pd(ewtabV,_mm256_mul_pd(_mm256_mul_pd(ewtabhalfspace,eweps),_mm256_add_pd(ewtabF,felec)));
 587             velec            = _mm256_mul_pd(qq11,_mm256_sub_pd(rinv11,velec));
 588             felec            = _mm256_mul_pd(_mm256_mul_pd(qq11,rinv11),_mm256_sub_pd(rinvsq11,felec));
 589
 590             d                = _mm256_sub_pd(r11,rswitch);
 591             d                = _mm256_max_pd(d,_mm256_setzero_pd());
 592             d2               = _mm256_mul_pd(d,d);
 593             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)))))));
 594
 595             dsw              = _mm256_mul_pd(d2,_mm256_add_pd(swF2,_mm256_mul_pd(d,_mm256_add_pd(swF3,_mm256_mul_pd(d,swF4)))));
 596
 597             /* Evaluate switch function */
 598             /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
 599             felec            = _mm256_sub_pd( _mm256_mul_pd(felec,sw) , _mm256_mul_pd(rinv11,_mm256_mul_pd(velec,dsw)) );
 600             velec            = _mm256_mul_pd(velec,sw);
 601             cutoff_mask      = _mm256_cmp_pd(rsq11,rcutoff2,_CMP_LT_OQ);
 602
 603             /* Update potential sum for this i atom from the interaction with this j atom. */
 604             velec            = _mm256_and_pd(velec,cutoff_mask);
 605             velecsum         = _mm256_add_pd(velecsum,velec);
 606
 607             fscal            = felec;
 608
 609             fscal            = _mm256_and_pd(fscal,cutoff_mask);
 610
 611             /* Calculate temporary vectorial force */
 612             tx               = _mm256_mul_pd(fscal,dx11);
 613             ty               = _mm256_mul_pd(fscal,dy11);
 614             tz               = _mm256_mul_pd(fscal,dz11);
 615
 616             /* Update vectorial force */
 617             fix1             = _mm256_add_pd(fix1,tx);
 618             fiy1             = _mm256_add_pd(fiy1,ty);
 619             fiz1             = _mm256_add_pd(fiz1,tz);
 620
 621             fjx1             = _mm256_add_pd(fjx1,tx);
 622             fjy1             = _mm256_add_pd(fjy1,ty);
 623             fjz1             = _mm256_add_pd(fjz1,tz);
 624
 625             }
 626
 627             /**************************
 628              * CALCULATE INTERACTIONS *
 629              **************************/
 630
 631             if (gmx_mm256_any_lt(rsq12,rcutoff2))
 632             {
 633
 634             r12              = _mm256_mul_pd(rsq12,rinv12);
 635
 636             /* EWALD ELECTROSTATICS */
 637
 638             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 639             ewrt             = _mm256_mul_pd(r12,ewtabscale);
 640             ewitab           = _mm256_cvttpd_epi32(ewrt);
 641             eweps            = _mm256_sub_pd(ewrt,_mm256_round_pd(ewrt, _MM_FROUND_FLOOR));
 642             ewitab           = _mm_slli_epi32(ewitab,2);
 643             ewtabF           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
 644             ewtabD           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,1) );
 645             ewtabV           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,2) );
 646             ewtabFn          = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,3) );
 647             GMX_MM256_FULLTRANSPOSE4_PD(ewtabF,ewtabD,ewtabV,ewtabFn);
 648             felec            = _mm256_add_pd(ewtabF,_mm256_mul_pd(eweps,ewtabD));
 649             velec            = _mm256_sub_pd(ewtabV,_mm256_mul_pd(_mm256_mul_pd(ewtabhalfspace,eweps),_mm256_add_pd(ewtabF,felec)));
 650             velec            = _mm256_mul_pd(qq12,_mm256_sub_pd(rinv12,velec));
 651             felec            = _mm256_mul_pd(_mm256_mul_pd(qq12,rinv12),_mm256_sub_pd(rinvsq12,felec));
 652
 653             d                = _mm256_sub_pd(r12,rswitch);
 654             d                = _mm256_max_pd(d,_mm256_setzero_pd());
 655             d2               = _mm256_mul_pd(d,d);
 656             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)))))));
 657
 658             dsw              = _mm256_mul_pd(d2,_mm256_add_pd(swF2,_mm256_mul_pd(d,_mm256_add_pd(swF3,_mm256_mul_pd(d,swF4)))));
 659
 660             /* Evaluate switch function */
 661             /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
 662             felec            = _mm256_sub_pd( _mm256_mul_pd(felec,sw) , _mm256_mul_pd(rinv12,_mm256_mul_pd(velec,dsw)) );
 663             velec            = _mm256_mul_pd(velec,sw);
 664             cutoff_mask      = _mm256_cmp_pd(rsq12,rcutoff2,_CMP_LT_OQ);
 665
 666             /* Update potential sum for this i atom from the interaction with this j atom. */
 667             velec            = _mm256_and_pd(velec,cutoff_mask);
 668             velecsum         = _mm256_add_pd(velecsum,velec);
 669
 670             fscal            = felec;
 671
 672             fscal            = _mm256_and_pd(fscal,cutoff_mask);
 673
 674             /* Calculate temporary vectorial force */
 675             tx               = _mm256_mul_pd(fscal,dx12);
 676             ty               = _mm256_mul_pd(fscal,dy12);
 677             tz               = _mm256_mul_pd(fscal,dz12);
 678
 679             /* Update vectorial force */
 680             fix1             = _mm256_add_pd(fix1,tx);
 681             fiy1             = _mm256_add_pd(fiy1,ty);
 682             fiz1             = _mm256_add_pd(fiz1,tz);
 683
 684             fjx2             = _mm256_add_pd(fjx2,tx);
 685             fjy2             = _mm256_add_pd(fjy2,ty);
 686             fjz2             = _mm256_add_pd(fjz2,tz);
 687
 688             }
 689
 690             /**************************
 691              * CALCULATE INTERACTIONS *
 692              **************************/
 693
 694             if (gmx_mm256_any_lt(rsq20,rcutoff2))
 695             {
 696
 697             r20              = _mm256_mul_pd(rsq20,rinv20);
 698
 699             /* EWALD ELECTROSTATICS */
 700
 701             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 702             ewrt             = _mm256_mul_pd(r20,ewtabscale);
 703             ewitab           = _mm256_cvttpd_epi32(ewrt);
 704             eweps            = _mm256_sub_pd(ewrt,_mm256_round_pd(ewrt, _MM_FROUND_FLOOR));
 705             ewitab           = _mm_slli_epi32(ewitab,2);
 706             ewtabF           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
 707             ewtabD           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,1) );
 708             ewtabV           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,2) );
 709             ewtabFn          = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,3) );
 710             GMX_MM256_FULLTRANSPOSE4_PD(ewtabF,ewtabD,ewtabV,ewtabFn);
 711             felec            = _mm256_add_pd(ewtabF,_mm256_mul_pd(eweps,ewtabD));
 712             velec            = _mm256_sub_pd(ewtabV,_mm256_mul_pd(_mm256_mul_pd(ewtabhalfspace,eweps),_mm256_add_pd(ewtabF,felec)));
 713             velec            = _mm256_mul_pd(qq20,_mm256_sub_pd(rinv20,velec));
 714             felec            = _mm256_mul_pd(_mm256_mul_pd(qq20,rinv20),_mm256_sub_pd(rinvsq20,felec));
 715
 716             d                = _mm256_sub_pd(r20,rswitch);
 717             d                = _mm256_max_pd(d,_mm256_setzero_pd());
 718             d2               = _mm256_mul_pd(d,d);
 719             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)))))));
 720
 721             dsw              = _mm256_mul_pd(d2,_mm256_add_pd(swF2,_mm256_mul_pd(d,_mm256_add_pd(swF3,_mm256_mul_pd(d,swF4)))));
 722
 723             /* Evaluate switch function */
 724             /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
 725             felec            = _mm256_sub_pd( _mm256_mul_pd(felec,sw) , _mm256_mul_pd(rinv20,_mm256_mul_pd(velec,dsw)) );
 726             velec            = _mm256_mul_pd(velec,sw);
 727             cutoff_mask      = _mm256_cmp_pd(rsq20,rcutoff2,_CMP_LT_OQ);
 728
 729             /* Update potential sum for this i atom from the interaction with this j atom. */
 730             velec            = _mm256_and_pd(velec,cutoff_mask);
 731             velecsum         = _mm256_add_pd(velecsum,velec);
 732
 733             fscal            = felec;
 734
 735             fscal            = _mm256_and_pd(fscal,cutoff_mask);
 736
 737             /* Calculate temporary vectorial force */
 738             tx               = _mm256_mul_pd(fscal,dx20);
 739             ty               = _mm256_mul_pd(fscal,dy20);
 740             tz               = _mm256_mul_pd(fscal,dz20);
 741
 742             /* Update vectorial force */
 743             fix2             = _mm256_add_pd(fix2,tx);
 744             fiy2             = _mm256_add_pd(fiy2,ty);
 745             fiz2             = _mm256_add_pd(fiz2,tz);
 746
 747             fjx0             = _mm256_add_pd(fjx0,tx);
 748             fjy0             = _mm256_add_pd(fjy0,ty);
 749             fjz0             = _mm256_add_pd(fjz0,tz);
 750
 751             }
 752
 753             /**************************
 754              * CALCULATE INTERACTIONS *
 755              **************************/
 756
 757             if (gmx_mm256_any_lt(rsq21,rcutoff2))
 758             {
 759
 760             r21              = _mm256_mul_pd(rsq21,rinv21);
 761
 762             /* EWALD ELECTROSTATICS */
 763
 764             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 765             ewrt             = _mm256_mul_pd(r21,ewtabscale);
 766             ewitab           = _mm256_cvttpd_epi32(ewrt);
 767             eweps            = _mm256_sub_pd(ewrt,_mm256_round_pd(ewrt, _MM_FROUND_FLOOR));
 768             ewitab           = _mm_slli_epi32(ewitab,2);
 769             ewtabF           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
 770             ewtabD           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,1) );
 771             ewtabV           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,2) );
 772             ewtabFn          = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,3) );
 773             GMX_MM256_FULLTRANSPOSE4_PD(ewtabF,ewtabD,ewtabV,ewtabFn);
 774             felec            = _mm256_add_pd(ewtabF,_mm256_mul_pd(eweps,ewtabD));
 775             velec            = _mm256_sub_pd(ewtabV,_mm256_mul_pd(_mm256_mul_pd(ewtabhalfspace,eweps),_mm256_add_pd(ewtabF,felec)));
 776             velec            = _mm256_mul_pd(qq21,_mm256_sub_pd(rinv21,velec));
 777             felec            = _mm256_mul_pd(_mm256_mul_pd(qq21,rinv21),_mm256_sub_pd(rinvsq21,felec));
 778
 779             d                = _mm256_sub_pd(r21,rswitch);
 780             d                = _mm256_max_pd(d,_mm256_setzero_pd());
 781             d2               = _mm256_mul_pd(d,d);
 782             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)))))));
 783
 784             dsw              = _mm256_mul_pd(d2,_mm256_add_pd(swF2,_mm256_mul_pd(d,_mm256_add_pd(swF3,_mm256_mul_pd(d,swF4)))));
 785
 786             /* Evaluate switch function */
 787             /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
 788             felec            = _mm256_sub_pd( _mm256_mul_pd(felec,sw) , _mm256_mul_pd(rinv21,_mm256_mul_pd(velec,dsw)) );
 789             velec            = _mm256_mul_pd(velec,sw);
 790             cutoff_mask      = _mm256_cmp_pd(rsq21,rcutoff2,_CMP_LT_OQ);
 791
 792             /* Update potential sum for this i atom from the interaction with this j atom. */
 793             velec            = _mm256_and_pd(velec,cutoff_mask);
 794             velecsum         = _mm256_add_pd(velecsum,velec);
 795
 796             fscal            = felec;
 797
 798             fscal            = _mm256_and_pd(fscal,cutoff_mask);
 799
 800             /* Calculate temporary vectorial force */
 801             tx               = _mm256_mul_pd(fscal,dx21);
 802             ty               = _mm256_mul_pd(fscal,dy21);
 803             tz               = _mm256_mul_pd(fscal,dz21);
 804
 805             /* Update vectorial force */
 806             fix2             = _mm256_add_pd(fix2,tx);
 807             fiy2             = _mm256_add_pd(fiy2,ty);
 808             fiz2             = _mm256_add_pd(fiz2,tz);
 809
 810             fjx1             = _mm256_add_pd(fjx1,tx);
 811             fjy1             = _mm256_add_pd(fjy1,ty);
 812             fjz1             = _mm256_add_pd(fjz1,tz);
 813
 814             }
 815
 816             /**************************
 817              * CALCULATE INTERACTIONS *
 818              **************************/
 819
 820             if (gmx_mm256_any_lt(rsq22,rcutoff2))
 821             {
 822
 823             r22              = _mm256_mul_pd(rsq22,rinv22);
 824
 825             /* EWALD ELECTROSTATICS */
 826
 827             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 828             ewrt             = _mm256_mul_pd(r22,ewtabscale);
 829             ewitab           = _mm256_cvttpd_epi32(ewrt);
 830             eweps            = _mm256_sub_pd(ewrt,_mm256_round_pd(ewrt, _MM_FROUND_FLOOR));
 831             ewitab           = _mm_slli_epi32(ewitab,2);
 832             ewtabF           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
 833             ewtabD           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,1) );
 834             ewtabV           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,2) );
 835             ewtabFn          = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,3) );
 836             GMX_MM256_FULLTRANSPOSE4_PD(ewtabF,ewtabD,ewtabV,ewtabFn);
 837             felec            = _mm256_add_pd(ewtabF,_mm256_mul_pd(eweps,ewtabD));
 838             velec            = _mm256_sub_pd(ewtabV,_mm256_mul_pd(_mm256_mul_pd(ewtabhalfspace,eweps),_mm256_add_pd(ewtabF,felec)));
 839             velec            = _mm256_mul_pd(qq22,_mm256_sub_pd(rinv22,velec));
 840             felec            = _mm256_mul_pd(_mm256_mul_pd(qq22,rinv22),_mm256_sub_pd(rinvsq22,felec));
 841
 842             d                = _mm256_sub_pd(r22,rswitch);
 843             d                = _mm256_max_pd(d,_mm256_setzero_pd());
 844             d2               = _mm256_mul_pd(d,d);
 845             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)))))));
 846
 847             dsw              = _mm256_mul_pd(d2,_mm256_add_pd(swF2,_mm256_mul_pd(d,_mm256_add_pd(swF3,_mm256_mul_pd(d,swF4)))));
 848
 849             /* Evaluate switch function */
 850             /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
 851             felec            = _mm256_sub_pd( _mm256_mul_pd(felec,sw) , _mm256_mul_pd(rinv22,_mm256_mul_pd(velec,dsw)) );
 852             velec            = _mm256_mul_pd(velec,sw);
 853             cutoff_mask      = _mm256_cmp_pd(rsq22,rcutoff2,_CMP_LT_OQ);
 854
 855             /* Update potential sum for this i atom from the interaction with this j atom. */
 856             velec            = _mm256_and_pd(velec,cutoff_mask);
 857             velecsum         = _mm256_add_pd(velecsum,velec);
 858
 859             fscal            = felec;
 860
 861             fscal            = _mm256_and_pd(fscal,cutoff_mask);
 862
 863             /* Calculate temporary vectorial force */
 864             tx               = _mm256_mul_pd(fscal,dx22);
 865             ty               = _mm256_mul_pd(fscal,dy22);
 866             tz               = _mm256_mul_pd(fscal,dz22);
 867
 868             /* Update vectorial force */
 869             fix2             = _mm256_add_pd(fix2,tx);
 870             fiy2             = _mm256_add_pd(fiy2,ty);
 871             fiz2             = _mm256_add_pd(fiz2,tz);
 872
 873             fjx2             = _mm256_add_pd(fjx2,tx);
 874             fjy2             = _mm256_add_pd(fjy2,ty);
 875             fjz2             = _mm256_add_pd(fjz2,tz);
 876
 877             }
 878
 879             fjptrA             = f+j_coord_offsetA;
 880             fjptrB             = f+j_coord_offsetB;
 881             fjptrC             = f+j_coord_offsetC;
 882             fjptrD             = f+j_coord_offsetD;
 883
 884             gmx_mm256_decrement_3rvec_4ptr_swizzle_pd(fjptrA,fjptrB,fjptrC,fjptrD,
 885                                                       fjx0,fjy0,fjz0,fjx1,fjy1,fjz1,fjx2,fjy2,fjz2);
 886
 887             /* Inner loop uses 585 flops */
 888         }
 889
 890         if(jidx<j_index_end)
 891         {
 892
 893             /* Get j neighbor index, and coordinate index */
 894             jnrlistA         = jjnr[jidx];
 895             jnrlistB         = jjnr[jidx+1];
 896             jnrlistC         = jjnr[jidx+2];
 897             jnrlistD         = jjnr[jidx+3];
 898             /* Sign of each element will be negative for non-real atoms.
 899              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 900              * so use it as val = _mm_andnot_pd(mask,val) to clear dummy entries.
 901              */
 902             tmpmask0 = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
 903
 904             tmpmask1 = _mm_permute_ps(tmpmask0,_GMX_MM_PERMUTE(3,3,2,2));
 905             tmpmask0 = _mm_permute_ps(tmpmask0,_GMX_MM_PERMUTE(1,1,0,0));
 906             dummy_mask = _mm256_castps_pd(gmx_mm256_set_m128(tmpmask1,tmpmask0));
 907
 908             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
 909             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
 910             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
 911             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
 912             j_coord_offsetA  = DIM*jnrA;
 913             j_coord_offsetB  = DIM*jnrB;
 914             j_coord_offsetC  = DIM*jnrC;
 915             j_coord_offsetD  = DIM*jnrD;
 916
 917             /* load j atom coordinates */
 918             gmx_mm256_load_3rvec_4ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 919                                                  x+j_coord_offsetC,x+j_coord_offsetD,
 920                                               &jx0,&jy0,&jz0,&jx1,&jy1,&jz1,&jx2,&jy2,&jz2);
 921
 922             /* Calculate displacement vector */
 923             dx00             = _mm256_sub_pd(ix0,jx0);
 924             dy00             = _mm256_sub_pd(iy0,jy0);
 925             dz00             = _mm256_sub_pd(iz0,jz0);
 926             dx01             = _mm256_sub_pd(ix0,jx1);
 927             dy01             = _mm256_sub_pd(iy0,jy1);
 928             dz01             = _mm256_sub_pd(iz0,jz1);
 929             dx02             = _mm256_sub_pd(ix0,jx2);
 930             dy02             = _mm256_sub_pd(iy0,jy2);
 931             dz02             = _mm256_sub_pd(iz0,jz2);
 932             dx10             = _mm256_sub_pd(ix1,jx0);
 933             dy10             = _mm256_sub_pd(iy1,jy0);
 934             dz10             = _mm256_sub_pd(iz1,jz0);
 935             dx11             = _mm256_sub_pd(ix1,jx1);
 936             dy11             = _mm256_sub_pd(iy1,jy1);
 937             dz11             = _mm256_sub_pd(iz1,jz1);
 938             dx12             = _mm256_sub_pd(ix1,jx2);
 939             dy12             = _mm256_sub_pd(iy1,jy2);
 940             dz12             = _mm256_sub_pd(iz1,jz2);
 941             dx20             = _mm256_sub_pd(ix2,jx0);
 942             dy20             = _mm256_sub_pd(iy2,jy0);
 943             dz20             = _mm256_sub_pd(iz2,jz0);
 944             dx21             = _mm256_sub_pd(ix2,jx1);
 945             dy21             = _mm256_sub_pd(iy2,jy1);
 946             dz21             = _mm256_sub_pd(iz2,jz1);
 947             dx22             = _mm256_sub_pd(ix2,jx2);
 948             dy22             = _mm256_sub_pd(iy2,jy2);
 949             dz22             = _mm256_sub_pd(iz2,jz2);
 950
 951             /* Calculate squared distance and things based on it */
 952             rsq00            = gmx_mm256_calc_rsq_pd(dx00,dy00,dz00);
 953             rsq01            = gmx_mm256_calc_rsq_pd(dx01,dy01,dz01);
 954             rsq02            = gmx_mm256_calc_rsq_pd(dx02,dy02,dz02);
 955             rsq10            = gmx_mm256_calc_rsq_pd(dx10,dy10,dz10);
 956             rsq11            = gmx_mm256_calc_rsq_pd(dx11,dy11,dz11);
 957             rsq12            = gmx_mm256_calc_rsq_pd(dx12,dy12,dz12);
 958             rsq20            = gmx_mm256_calc_rsq_pd(dx20,dy20,dz20);
 959             rsq21            = gmx_mm256_calc_rsq_pd(dx21,dy21,dz21);
 960             rsq22            = gmx_mm256_calc_rsq_pd(dx22,dy22,dz22);
 961
 962             rinv00           = gmx_mm256_invsqrt_pd(rsq00);
 963             rinv01           = gmx_mm256_invsqrt_pd(rsq01);
 964             rinv02           = gmx_mm256_invsqrt_pd(rsq02);
 965             rinv10           = gmx_mm256_invsqrt_pd(rsq10);
 966             rinv11           = gmx_mm256_invsqrt_pd(rsq11);
 967             rinv12           = gmx_mm256_invsqrt_pd(rsq12);
 968             rinv20           = gmx_mm256_invsqrt_pd(rsq20);
 969             rinv21           = gmx_mm256_invsqrt_pd(rsq21);
 970             rinv22           = gmx_mm256_invsqrt_pd(rsq22);
 971
 972             rinvsq00         = _mm256_mul_pd(rinv00,rinv00);
 973             rinvsq01         = _mm256_mul_pd(rinv01,rinv01);
 974             rinvsq02         = _mm256_mul_pd(rinv02,rinv02);
 975             rinvsq10         = _mm256_mul_pd(rinv10,rinv10);
 976             rinvsq11         = _mm256_mul_pd(rinv11,rinv11);
 977             rinvsq12         = _mm256_mul_pd(rinv12,rinv12);
 978             rinvsq20         = _mm256_mul_pd(rinv20,rinv20);
 979             rinvsq21         = _mm256_mul_pd(rinv21,rinv21);
 980             rinvsq22         = _mm256_mul_pd(rinv22,rinv22);
 981
 982             fjx0             = _mm256_setzero_pd();
 983             fjy0             = _mm256_setzero_pd();
 984             fjz0             = _mm256_setzero_pd();
 985             fjx1             = _mm256_setzero_pd();
 986             fjy1             = _mm256_setzero_pd();
 987             fjz1             = _mm256_setzero_pd();
 988             fjx2             = _mm256_setzero_pd();
 989             fjy2             = _mm256_setzero_pd();
 990             fjz2             = _mm256_setzero_pd();
 991
 992             /**************************
 993              * CALCULATE INTERACTIONS *
 994              **************************/
 995
 996             if (gmx_mm256_any_lt(rsq00,rcutoff2))
 997             {
 998
 999             r00              = _mm256_mul_pd(rsq00,rinv00);
1000             r00              = _mm256_andnot_pd(dummy_mask,r00);
1001
1002             /* EWALD ELECTROSTATICS */
1003
1004             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1005             ewrt             = _mm256_mul_pd(r00,ewtabscale);
1006             ewitab           = _mm256_cvttpd_epi32(ewrt);
1007             eweps            = _mm256_sub_pd(ewrt,_mm256_round_pd(ewrt, _MM_FROUND_FLOOR));
1008             ewitab           = _mm_slli_epi32(ewitab,2);
1009             ewtabF           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
1010             ewtabD           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,1) );
1011             ewtabV           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,2) );
1012             ewtabFn          = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,3) );
1013             GMX_MM256_FULLTRANSPOSE4_PD(ewtabF,ewtabD,ewtabV,ewtabFn);
1014             felec            = _mm256_add_pd(ewtabF,_mm256_mul_pd(eweps,ewtabD));
1015             velec            = _mm256_sub_pd(ewtabV,_mm256_mul_pd(_mm256_mul_pd(ewtabhalfspace,eweps),_mm256_add_pd(ewtabF,felec)));
1016             velec            = _mm256_mul_pd(qq00,_mm256_sub_pd(rinv00,velec));
1017             felec            = _mm256_mul_pd(_mm256_mul_pd(qq00,rinv00),_mm256_sub_pd(rinvsq00,felec));
1018
1019             d                = _mm256_sub_pd(r00,rswitch);
1020             d                = _mm256_max_pd(d,_mm256_setzero_pd());
1021             d2               = _mm256_mul_pd(d,d);
1022             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)))))));
1023
1024             dsw              = _mm256_mul_pd(d2,_mm256_add_pd(swF2,_mm256_mul_pd(d,_mm256_add_pd(swF3,_mm256_mul_pd(d,swF4)))));
1025
1026             /* Evaluate switch function */
1027             /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
1028             felec            = _mm256_sub_pd( _mm256_mul_pd(felec,sw) , _mm256_mul_pd(rinv00,_mm256_mul_pd(velec,dsw)) );
1029             velec            = _mm256_mul_pd(velec,sw);
1030             cutoff_mask      = _mm256_cmp_pd(rsq00,rcutoff2,_CMP_LT_OQ);
1031
1032             /* Update potential sum for this i atom from the interaction with this j atom. */
1033             velec            = _mm256_and_pd(velec,cutoff_mask);
1034             velec            = _mm256_andnot_pd(dummy_mask,velec);
1035             velecsum         = _mm256_add_pd(velecsum,velec);
1036
1037             fscal            = felec;
1038
1039             fscal            = _mm256_and_pd(fscal,cutoff_mask);
1040
1041             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
1042
1043             /* Calculate temporary vectorial force */
1044             tx               = _mm256_mul_pd(fscal,dx00);
1045             ty               = _mm256_mul_pd(fscal,dy00);
1046             tz               = _mm256_mul_pd(fscal,dz00);
1047
1048             /* Update vectorial force */
1049             fix0             = _mm256_add_pd(fix0,tx);
1050             fiy0             = _mm256_add_pd(fiy0,ty);
1051             fiz0             = _mm256_add_pd(fiz0,tz);
1052
1053             fjx0             = _mm256_add_pd(fjx0,tx);
1054             fjy0             = _mm256_add_pd(fjy0,ty);
1055             fjz0             = _mm256_add_pd(fjz0,tz);
1056
1057             }
1058
1059             /**************************
1060              * CALCULATE INTERACTIONS *
1061              **************************/
1062
1063             if (gmx_mm256_any_lt(rsq01,rcutoff2))
1064             {
1065
1066             r01              = _mm256_mul_pd(rsq01,rinv01);
1067             r01              = _mm256_andnot_pd(dummy_mask,r01);
1068
1069             /* EWALD ELECTROSTATICS */
1070
1071             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1072             ewrt             = _mm256_mul_pd(r01,ewtabscale);
1073             ewitab           = _mm256_cvttpd_epi32(ewrt);
1074             eweps            = _mm256_sub_pd(ewrt,_mm256_round_pd(ewrt, _MM_FROUND_FLOOR));
1075             ewitab           = _mm_slli_epi32(ewitab,2);
1076             ewtabF           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
1077             ewtabD           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,1) );
1078             ewtabV           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,2) );
1079             ewtabFn          = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,3) );
1080             GMX_MM256_FULLTRANSPOSE4_PD(ewtabF,ewtabD,ewtabV,ewtabFn);
1081             felec            = _mm256_add_pd(ewtabF,_mm256_mul_pd(eweps,ewtabD));
1082             velec            = _mm256_sub_pd(ewtabV,_mm256_mul_pd(_mm256_mul_pd(ewtabhalfspace,eweps),_mm256_add_pd(ewtabF,felec)));
1083             velec            = _mm256_mul_pd(qq01,_mm256_sub_pd(rinv01,velec));
1084             felec            = _mm256_mul_pd(_mm256_mul_pd(qq01,rinv01),_mm256_sub_pd(rinvsq01,felec));
1085
1086             d                = _mm256_sub_pd(r01,rswitch);
1087             d                = _mm256_max_pd(d,_mm256_setzero_pd());
1088             d2               = _mm256_mul_pd(d,d);
1089             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)))))));
1090
1091             dsw              = _mm256_mul_pd(d2,_mm256_add_pd(swF2,_mm256_mul_pd(d,_mm256_add_pd(swF3,_mm256_mul_pd(d,swF4)))));
1092
1093             /* Evaluate switch function */
1094             /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
1095             felec            = _mm256_sub_pd( _mm256_mul_pd(felec,sw) , _mm256_mul_pd(rinv01,_mm256_mul_pd(velec,dsw)) );
1096             velec            = _mm256_mul_pd(velec,sw);
1097             cutoff_mask      = _mm256_cmp_pd(rsq01,rcutoff2,_CMP_LT_OQ);
1098
1099             /* Update potential sum for this i atom from the interaction with this j atom. */
1100             velec            = _mm256_and_pd(velec,cutoff_mask);
1101             velec            = _mm256_andnot_pd(dummy_mask,velec);
1102             velecsum         = _mm256_add_pd(velecsum,velec);
1103
1104             fscal            = felec;
1105
1106             fscal            = _mm256_and_pd(fscal,cutoff_mask);
1107
1108             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
1109
1110             /* Calculate temporary vectorial force */
1111             tx               = _mm256_mul_pd(fscal,dx01);
1112             ty               = _mm256_mul_pd(fscal,dy01);
1113             tz               = _mm256_mul_pd(fscal,dz01);
1114
1115             /* Update vectorial force */
1116             fix0             = _mm256_add_pd(fix0,tx);
1117             fiy0             = _mm256_add_pd(fiy0,ty);
1118             fiz0             = _mm256_add_pd(fiz0,tz);
1119
1120             fjx1             = _mm256_add_pd(fjx1,tx);
1121             fjy1             = _mm256_add_pd(fjy1,ty);
1122             fjz1             = _mm256_add_pd(fjz1,tz);
1123
1124             }
1125
1126             /**************************
1127              * CALCULATE INTERACTIONS *
1128              **************************/
1129
1130             if (gmx_mm256_any_lt(rsq02,rcutoff2))
1131             {
1132
1133             r02              = _mm256_mul_pd(rsq02,rinv02);
1134             r02              = _mm256_andnot_pd(dummy_mask,r02);
1135
1136             /* EWALD ELECTROSTATICS */
1137
1138             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1139             ewrt             = _mm256_mul_pd(r02,ewtabscale);
1140             ewitab           = _mm256_cvttpd_epi32(ewrt);
1141             eweps            = _mm256_sub_pd(ewrt,_mm256_round_pd(ewrt, _MM_FROUND_FLOOR));
1142             ewitab           = _mm_slli_epi32(ewitab,2);
1143             ewtabF           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
1144             ewtabD           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,1) );
1145             ewtabV           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,2) );
1146             ewtabFn          = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,3) );
1147             GMX_MM256_FULLTRANSPOSE4_PD(ewtabF,ewtabD,ewtabV,ewtabFn);
1148             felec            = _mm256_add_pd(ewtabF,_mm256_mul_pd(eweps,ewtabD));
1149             velec            = _mm256_sub_pd(ewtabV,_mm256_mul_pd(_mm256_mul_pd(ewtabhalfspace,eweps),_mm256_add_pd(ewtabF,felec)));
1150             velec            = _mm256_mul_pd(qq02,_mm256_sub_pd(rinv02,velec));
1151             felec            = _mm256_mul_pd(_mm256_mul_pd(qq02,rinv02),_mm256_sub_pd(rinvsq02,felec));
1152
1153             d                = _mm256_sub_pd(r02,rswitch);
1154             d                = _mm256_max_pd(d,_mm256_setzero_pd());
1155             d2               = _mm256_mul_pd(d,d);
1156             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)))))));
1157
1158             dsw              = _mm256_mul_pd(d2,_mm256_add_pd(swF2,_mm256_mul_pd(d,_mm256_add_pd(swF3,_mm256_mul_pd(d,swF4)))));
1159
1160             /* Evaluate switch function */
1161             /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
1162             felec            = _mm256_sub_pd( _mm256_mul_pd(felec,sw) , _mm256_mul_pd(rinv02,_mm256_mul_pd(velec,dsw)) );
1163             velec            = _mm256_mul_pd(velec,sw);
1164             cutoff_mask      = _mm256_cmp_pd(rsq02,rcutoff2,_CMP_LT_OQ);
1165
1166             /* Update potential sum for this i atom from the interaction with this j atom. */
1167             velec            = _mm256_and_pd(velec,cutoff_mask);
1168             velec            = _mm256_andnot_pd(dummy_mask,velec);
1169             velecsum         = _mm256_add_pd(velecsum,velec);
1170
1171             fscal            = felec;
1172
1173             fscal            = _mm256_and_pd(fscal,cutoff_mask);
1174
1175             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
1176
1177             /* Calculate temporary vectorial force */
1178             tx               = _mm256_mul_pd(fscal,dx02);
1179             ty               = _mm256_mul_pd(fscal,dy02);
1180             tz               = _mm256_mul_pd(fscal,dz02);
1181
1182             /* Update vectorial force */
1183             fix0             = _mm256_add_pd(fix0,tx);
1184             fiy0             = _mm256_add_pd(fiy0,ty);
1185             fiz0             = _mm256_add_pd(fiz0,tz);
1186
1187             fjx2             = _mm256_add_pd(fjx2,tx);
1188             fjy2             = _mm256_add_pd(fjy2,ty);
1189             fjz2             = _mm256_add_pd(fjz2,tz);
1190
1191             }
1192
1193             /**************************
1194              * CALCULATE INTERACTIONS *
1195              **************************/
1196
1197             if (gmx_mm256_any_lt(rsq10,rcutoff2))
1198             {
1199
1200             r10              = _mm256_mul_pd(rsq10,rinv10);
1201             r10              = _mm256_andnot_pd(dummy_mask,r10);
1202
1203             /* EWALD ELECTROSTATICS */
1204
1205             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1206             ewrt             = _mm256_mul_pd(r10,ewtabscale);
1207             ewitab           = _mm256_cvttpd_epi32(ewrt);
1208             eweps            = _mm256_sub_pd(ewrt,_mm256_round_pd(ewrt, _MM_FROUND_FLOOR));
1209             ewitab           = _mm_slli_epi32(ewitab,2);
1210             ewtabF           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
1211             ewtabD           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,1) );
1212             ewtabV           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,2) );
1213             ewtabFn          = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,3) );
1214             GMX_MM256_FULLTRANSPOSE4_PD(ewtabF,ewtabD,ewtabV,ewtabFn);
1215             felec            = _mm256_add_pd(ewtabF,_mm256_mul_pd(eweps,ewtabD));
1216             velec            = _mm256_sub_pd(ewtabV,_mm256_mul_pd(_mm256_mul_pd(ewtabhalfspace,eweps),_mm256_add_pd(ewtabF,felec)));
1217             velec            = _mm256_mul_pd(qq10,_mm256_sub_pd(rinv10,velec));
1218             felec            = _mm256_mul_pd(_mm256_mul_pd(qq10,rinv10),_mm256_sub_pd(rinvsq10,felec));
1219
1220             d                = _mm256_sub_pd(r10,rswitch);
1221             d                = _mm256_max_pd(d,_mm256_setzero_pd());
1222             d2               = _mm256_mul_pd(d,d);
1223             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)))))));
1224
1225             dsw              = _mm256_mul_pd(d2,_mm256_add_pd(swF2,_mm256_mul_pd(d,_mm256_add_pd(swF3,_mm256_mul_pd(d,swF4)))));
1226
1227             /* Evaluate switch function */
1228             /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
1229             felec            = _mm256_sub_pd( _mm256_mul_pd(felec,sw) , _mm256_mul_pd(rinv10,_mm256_mul_pd(velec,dsw)) );
1230             velec            = _mm256_mul_pd(velec,sw);
1231             cutoff_mask      = _mm256_cmp_pd(rsq10,rcutoff2,_CMP_LT_OQ);
1232
1233             /* Update potential sum for this i atom from the interaction with this j atom. */
1234             velec            = _mm256_and_pd(velec,cutoff_mask);
1235             velec            = _mm256_andnot_pd(dummy_mask,velec);
1236             velecsum         = _mm256_add_pd(velecsum,velec);
1237
1238             fscal            = felec;
1239
1240             fscal            = _mm256_and_pd(fscal,cutoff_mask);
1241
1242             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
1243
1244             /* Calculate temporary vectorial force */
1245             tx               = _mm256_mul_pd(fscal,dx10);
1246             ty               = _mm256_mul_pd(fscal,dy10);
1247             tz               = _mm256_mul_pd(fscal,dz10);
1248
1249             /* Update vectorial force */
1250             fix1             = _mm256_add_pd(fix1,tx);
1251             fiy1             = _mm256_add_pd(fiy1,ty);
1252             fiz1             = _mm256_add_pd(fiz1,tz);
1253
1254             fjx0             = _mm256_add_pd(fjx0,tx);
1255             fjy0             = _mm256_add_pd(fjy0,ty);
1256             fjz0             = _mm256_add_pd(fjz0,tz);
1257
1258             }
1259
1260             /**************************
1261              * CALCULATE INTERACTIONS *
1262              **************************/
1263
1264             if (gmx_mm256_any_lt(rsq11,rcutoff2))
1265             {
1266
1267             r11              = _mm256_mul_pd(rsq11,rinv11);
1268             r11              = _mm256_andnot_pd(dummy_mask,r11);
1269
1270             /* EWALD ELECTROSTATICS */
1271
1272             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1273             ewrt             = _mm256_mul_pd(r11,ewtabscale);
1274             ewitab           = _mm256_cvttpd_epi32(ewrt);
1275             eweps            = _mm256_sub_pd(ewrt,_mm256_round_pd(ewrt, _MM_FROUND_FLOOR));
1276             ewitab           = _mm_slli_epi32(ewitab,2);
1277             ewtabF           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
1278             ewtabD           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,1) );
1279             ewtabV           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,2) );
1280             ewtabFn          = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,3) );
1281             GMX_MM256_FULLTRANSPOSE4_PD(ewtabF,ewtabD,ewtabV,ewtabFn);
1282             felec            = _mm256_add_pd(ewtabF,_mm256_mul_pd(eweps,ewtabD));
1283             velec            = _mm256_sub_pd(ewtabV,_mm256_mul_pd(_mm256_mul_pd(ewtabhalfspace,eweps),_mm256_add_pd(ewtabF,felec)));
1284             velec            = _mm256_mul_pd(qq11,_mm256_sub_pd(rinv11,velec));
1285             felec            = _mm256_mul_pd(_mm256_mul_pd(qq11,rinv11),_mm256_sub_pd(rinvsq11,felec));
1286
1287             d                = _mm256_sub_pd(r11,rswitch);
1288             d                = _mm256_max_pd(d,_mm256_setzero_pd());
1289             d2               = _mm256_mul_pd(d,d);
1290             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)))))));
1291
1292             dsw              = _mm256_mul_pd(d2,_mm256_add_pd(swF2,_mm256_mul_pd(d,_mm256_add_pd(swF3,_mm256_mul_pd(d,swF4)))));
1293
1294             /* Evaluate switch function */
1295             /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
1296             felec            = _mm256_sub_pd( _mm256_mul_pd(felec,sw) , _mm256_mul_pd(rinv11,_mm256_mul_pd(velec,dsw)) );
1297             velec            = _mm256_mul_pd(velec,sw);
1298             cutoff_mask      = _mm256_cmp_pd(rsq11,rcutoff2,_CMP_LT_OQ);
1299
1300             /* Update potential sum for this i atom from the interaction with this j atom. */
1301             velec            = _mm256_and_pd(velec,cutoff_mask);
1302             velec            = _mm256_andnot_pd(dummy_mask,velec);
1303             velecsum         = _mm256_add_pd(velecsum,velec);
1304
1305             fscal            = felec;
1306
1307             fscal            = _mm256_and_pd(fscal,cutoff_mask);
1308
1309             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
1310
1311             /* Calculate temporary vectorial force */
1312             tx               = _mm256_mul_pd(fscal,dx11);
1313             ty               = _mm256_mul_pd(fscal,dy11);
1314             tz               = _mm256_mul_pd(fscal,dz11);
1315
1316             /* Update vectorial force */
1317             fix1             = _mm256_add_pd(fix1,tx);
1318             fiy1             = _mm256_add_pd(fiy1,ty);
1319             fiz1             = _mm256_add_pd(fiz1,tz);
1320
1321             fjx1             = _mm256_add_pd(fjx1,tx);
1322             fjy1             = _mm256_add_pd(fjy1,ty);
1323             fjz1             = _mm256_add_pd(fjz1,tz);
1324
1325             }
1326
1327             /**************************
1328              * CALCULATE INTERACTIONS *
1329              **************************/
1330
1331             if (gmx_mm256_any_lt(rsq12,rcutoff2))
1332             {
1333
1334             r12              = _mm256_mul_pd(rsq12,rinv12);
1335             r12              = _mm256_andnot_pd(dummy_mask,r12);
1336
1337             /* EWALD ELECTROSTATICS */
1338
1339             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1340             ewrt             = _mm256_mul_pd(r12,ewtabscale);
1341             ewitab           = _mm256_cvttpd_epi32(ewrt);
1342             eweps            = _mm256_sub_pd(ewrt,_mm256_round_pd(ewrt, _MM_FROUND_FLOOR));
1343             ewitab           = _mm_slli_epi32(ewitab,2);
1344             ewtabF           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
1345             ewtabD           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,1) );
1346             ewtabV           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,2) );
1347             ewtabFn          = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,3) );
1348             GMX_MM256_FULLTRANSPOSE4_PD(ewtabF,ewtabD,ewtabV,ewtabFn);
1349             felec            = _mm256_add_pd(ewtabF,_mm256_mul_pd(eweps,ewtabD));
1350             velec            = _mm256_sub_pd(ewtabV,_mm256_mul_pd(_mm256_mul_pd(ewtabhalfspace,eweps),_mm256_add_pd(ewtabF,felec)));
1351             velec            = _mm256_mul_pd(qq12,_mm256_sub_pd(rinv12,velec));
1352             felec            = _mm256_mul_pd(_mm256_mul_pd(qq12,rinv12),_mm256_sub_pd(rinvsq12,felec));
1353
1354             d                = _mm256_sub_pd(r12,rswitch);
1355             d                = _mm256_max_pd(d,_mm256_setzero_pd());
1356             d2               = _mm256_mul_pd(d,d);
1357             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)))))));
1358
1359             dsw              = _mm256_mul_pd(d2,_mm256_add_pd(swF2,_mm256_mul_pd(d,_mm256_add_pd(swF3,_mm256_mul_pd(d,swF4)))));
1360
1361             /* Evaluate switch function */
1362             /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
1363             felec            = _mm256_sub_pd( _mm256_mul_pd(felec,sw) , _mm256_mul_pd(rinv12,_mm256_mul_pd(velec,dsw)) );
1364             velec            = _mm256_mul_pd(velec,sw);
1365             cutoff_mask      = _mm256_cmp_pd(rsq12,rcutoff2,_CMP_LT_OQ);
1366
1367             /* Update potential sum for this i atom from the interaction with this j atom. */
1368             velec            = _mm256_and_pd(velec,cutoff_mask);
1369             velec            = _mm256_andnot_pd(dummy_mask,velec);
1370             velecsum         = _mm256_add_pd(velecsum,velec);
1371
1372             fscal            = felec;
1373
1374             fscal            = _mm256_and_pd(fscal,cutoff_mask);
1375
1376             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
1377
1378             /* Calculate temporary vectorial force */
1379             tx               = _mm256_mul_pd(fscal,dx12);
1380             ty               = _mm256_mul_pd(fscal,dy12);
1381             tz               = _mm256_mul_pd(fscal,dz12);
1382
1383             /* Update vectorial force */
1384             fix1             = _mm256_add_pd(fix1,tx);
1385             fiy1             = _mm256_add_pd(fiy1,ty);
1386             fiz1             = _mm256_add_pd(fiz1,tz);
1387
1388             fjx2             = _mm256_add_pd(fjx2,tx);
1389             fjy2             = _mm256_add_pd(fjy2,ty);
1390             fjz2             = _mm256_add_pd(fjz2,tz);
1391
1392             }
1393
1394             /**************************
1395              * CALCULATE INTERACTIONS *
1396              **************************/
1397
1398             if (gmx_mm256_any_lt(rsq20,rcutoff2))
1399             {
1400
1401             r20              = _mm256_mul_pd(rsq20,rinv20);
1402             r20              = _mm256_andnot_pd(dummy_mask,r20);
1403
1404             /* EWALD ELECTROSTATICS */
1405
1406             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1407             ewrt             = _mm256_mul_pd(r20,ewtabscale);
1408             ewitab           = _mm256_cvttpd_epi32(ewrt);
1409             eweps            = _mm256_sub_pd(ewrt,_mm256_round_pd(ewrt, _MM_FROUND_FLOOR));
1410             ewitab           = _mm_slli_epi32(ewitab,2);
1411             ewtabF           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
1412             ewtabD           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,1) );
1413             ewtabV           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,2) );
1414             ewtabFn          = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,3) );
1415             GMX_MM256_FULLTRANSPOSE4_PD(ewtabF,ewtabD,ewtabV,ewtabFn);
1416             felec            = _mm256_add_pd(ewtabF,_mm256_mul_pd(eweps,ewtabD));
1417             velec            = _mm256_sub_pd(ewtabV,_mm256_mul_pd(_mm256_mul_pd(ewtabhalfspace,eweps),_mm256_add_pd(ewtabF,felec)));
1418             velec            = _mm256_mul_pd(qq20,_mm256_sub_pd(rinv20,velec));
1419             felec            = _mm256_mul_pd(_mm256_mul_pd(qq20,rinv20),_mm256_sub_pd(rinvsq20,felec));
1420
1421             d                = _mm256_sub_pd(r20,rswitch);
1422             d                = _mm256_max_pd(d,_mm256_setzero_pd());
1423             d2               = _mm256_mul_pd(d,d);
1424             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)))))));
1425
1426             dsw              = _mm256_mul_pd(d2,_mm256_add_pd(swF2,_mm256_mul_pd(d,_mm256_add_pd(swF3,_mm256_mul_pd(d,swF4)))));
1427
1428             /* Evaluate switch function */
1429             /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
1430             felec            = _mm256_sub_pd( _mm256_mul_pd(felec,sw) , _mm256_mul_pd(rinv20,_mm256_mul_pd(velec,dsw)) );
1431             velec            = _mm256_mul_pd(velec,sw);
1432             cutoff_mask      = _mm256_cmp_pd(rsq20,rcutoff2,_CMP_LT_OQ);
1433
1434             /* Update potential sum for this i atom from the interaction with this j atom. */
1435             velec            = _mm256_and_pd(velec,cutoff_mask);
1436             velec            = _mm256_andnot_pd(dummy_mask,velec);
1437             velecsum         = _mm256_add_pd(velecsum,velec);
1438
1439             fscal            = felec;
1440
1441             fscal            = _mm256_and_pd(fscal,cutoff_mask);
1442
1443             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
1444
1445             /* Calculate temporary vectorial force */
1446             tx               = _mm256_mul_pd(fscal,dx20);
1447             ty               = _mm256_mul_pd(fscal,dy20);
1448             tz               = _mm256_mul_pd(fscal,dz20);
1449
1450             /* Update vectorial force */
1451             fix2             = _mm256_add_pd(fix2,tx);
1452             fiy2             = _mm256_add_pd(fiy2,ty);
1453             fiz2             = _mm256_add_pd(fiz2,tz);
1454
1455             fjx0             = _mm256_add_pd(fjx0,tx);
1456             fjy0             = _mm256_add_pd(fjy0,ty);
1457             fjz0             = _mm256_add_pd(fjz0,tz);
1458
1459             }
1460
1461             /**************************
1462              * CALCULATE INTERACTIONS *
1463              **************************/
1464
1465             if (gmx_mm256_any_lt(rsq21,rcutoff2))
1466             {
1467
1468             r21              = _mm256_mul_pd(rsq21,rinv21);
1469             r21              = _mm256_andnot_pd(dummy_mask,r21);
1470
1471             /* EWALD ELECTROSTATICS */
1472
1473             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1474             ewrt             = _mm256_mul_pd(r21,ewtabscale);
1475             ewitab           = _mm256_cvttpd_epi32(ewrt);
1476             eweps            = _mm256_sub_pd(ewrt,_mm256_round_pd(ewrt, _MM_FROUND_FLOOR));
1477             ewitab           = _mm_slli_epi32(ewitab,2);
1478             ewtabF           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
1479             ewtabD           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,1) );
1480             ewtabV           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,2) );
1481             ewtabFn          = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,3) );
1482             GMX_MM256_FULLTRANSPOSE4_PD(ewtabF,ewtabD,ewtabV,ewtabFn);
1483             felec            = _mm256_add_pd(ewtabF,_mm256_mul_pd(eweps,ewtabD));
1484             velec            = _mm256_sub_pd(ewtabV,_mm256_mul_pd(_mm256_mul_pd(ewtabhalfspace,eweps),_mm256_add_pd(ewtabF,felec)));
1485             velec            = _mm256_mul_pd(qq21,_mm256_sub_pd(rinv21,velec));
1486             felec            = _mm256_mul_pd(_mm256_mul_pd(qq21,rinv21),_mm256_sub_pd(rinvsq21,felec));
1487
1488             d                = _mm256_sub_pd(r21,rswitch);
1489             d                = _mm256_max_pd(d,_mm256_setzero_pd());
1490             d2               = _mm256_mul_pd(d,d);
1491             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)))))));
1492
1493             dsw              = _mm256_mul_pd(d2,_mm256_add_pd(swF2,_mm256_mul_pd(d,_mm256_add_pd(swF3,_mm256_mul_pd(d,swF4)))));
1494
1495             /* Evaluate switch function */
1496             /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
1497             felec            = _mm256_sub_pd( _mm256_mul_pd(felec,sw) , _mm256_mul_pd(rinv21,_mm256_mul_pd(velec,dsw)) );
1498             velec            = _mm256_mul_pd(velec,sw);
1499             cutoff_mask      = _mm256_cmp_pd(rsq21,rcutoff2,_CMP_LT_OQ);
1500
1501             /* Update potential sum for this i atom from the interaction with this j atom. */
1502             velec            = _mm256_and_pd(velec,cutoff_mask);
1503             velec            = _mm256_andnot_pd(dummy_mask,velec);
1504             velecsum         = _mm256_add_pd(velecsum,velec);
1505
1506             fscal            = felec;
1507
1508             fscal            = _mm256_and_pd(fscal,cutoff_mask);
1509
1510             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
1511
1512             /* Calculate temporary vectorial force */
1513             tx               = _mm256_mul_pd(fscal,dx21);
1514             ty               = _mm256_mul_pd(fscal,dy21);
1515             tz               = _mm256_mul_pd(fscal,dz21);
1516
1517             /* Update vectorial force */
1518             fix2             = _mm256_add_pd(fix2,tx);
1519             fiy2             = _mm256_add_pd(fiy2,ty);
1520             fiz2             = _mm256_add_pd(fiz2,tz);
1521
1522             fjx1             = _mm256_add_pd(fjx1,tx);
1523             fjy1             = _mm256_add_pd(fjy1,ty);
1524             fjz1             = _mm256_add_pd(fjz1,tz);
1525
1526             }
1527
1528             /**************************
1529              * CALCULATE INTERACTIONS *
1530              **************************/
1531
1532             if (gmx_mm256_any_lt(rsq22,rcutoff2))
1533             {
1534
1535             r22              = _mm256_mul_pd(rsq22,rinv22);
1536             r22              = _mm256_andnot_pd(dummy_mask,r22);
1537
1538             /* EWALD ELECTROSTATICS */
1539
1540             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1541             ewrt             = _mm256_mul_pd(r22,ewtabscale);
1542             ewitab           = _mm256_cvttpd_epi32(ewrt);
1543             eweps            = _mm256_sub_pd(ewrt,_mm256_round_pd(ewrt, _MM_FROUND_FLOOR));
1544             ewitab           = _mm_slli_epi32(ewitab,2);
1545             ewtabF           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
1546             ewtabD           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,1) );
1547             ewtabV           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,2) );
1548             ewtabFn          = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,3) );
1549             GMX_MM256_FULLTRANSPOSE4_PD(ewtabF,ewtabD,ewtabV,ewtabFn);
1550             felec            = _mm256_add_pd(ewtabF,_mm256_mul_pd(eweps,ewtabD));
1551             velec            = _mm256_sub_pd(ewtabV,_mm256_mul_pd(_mm256_mul_pd(ewtabhalfspace,eweps),_mm256_add_pd(ewtabF,felec)));
1552             velec            = _mm256_mul_pd(qq22,_mm256_sub_pd(rinv22,velec));
1553             felec            = _mm256_mul_pd(_mm256_mul_pd(qq22,rinv22),_mm256_sub_pd(rinvsq22,felec));
1554
1555             d                = _mm256_sub_pd(r22,rswitch);
1556             d                = _mm256_max_pd(d,_mm256_setzero_pd());
1557             d2               = _mm256_mul_pd(d,d);
1558             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)))))));
1559
1560             dsw              = _mm256_mul_pd(d2,_mm256_add_pd(swF2,_mm256_mul_pd(d,_mm256_add_pd(swF3,_mm256_mul_pd(d,swF4)))));
1561
1562             /* Evaluate switch function */
1563             /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
1564             felec            = _mm256_sub_pd( _mm256_mul_pd(felec,sw) , _mm256_mul_pd(rinv22,_mm256_mul_pd(velec,dsw)) );
1565             velec            = _mm256_mul_pd(velec,sw);
1566             cutoff_mask      = _mm256_cmp_pd(rsq22,rcutoff2,_CMP_LT_OQ);
1567
1568             /* Update potential sum for this i atom from the interaction with this j atom. */
1569             velec            = _mm256_and_pd(velec,cutoff_mask);
1570             velec            = _mm256_andnot_pd(dummy_mask,velec);
1571             velecsum         = _mm256_add_pd(velecsum,velec);
1572
1573             fscal            = felec;
1574
1575             fscal            = _mm256_and_pd(fscal,cutoff_mask);
1576
1577             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
1578
1579             /* Calculate temporary vectorial force */
1580             tx               = _mm256_mul_pd(fscal,dx22);
1581             ty               = _mm256_mul_pd(fscal,dy22);
1582             tz               = _mm256_mul_pd(fscal,dz22);
1583
1584             /* Update vectorial force */
1585             fix2             = _mm256_add_pd(fix2,tx);
1586             fiy2             = _mm256_add_pd(fiy2,ty);
1587             fiz2             = _mm256_add_pd(fiz2,tz);
1588
1589             fjx2             = _mm256_add_pd(fjx2,tx);
1590             fjy2             = _mm256_add_pd(fjy2,ty);
1591             fjz2             = _mm256_add_pd(fjz2,tz);
1592
1593             }
1594
1595             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
1596             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
1597             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
1598             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
1599
1600             gmx_mm256_decrement_3rvec_4ptr_swizzle_pd(fjptrA,fjptrB,fjptrC,fjptrD,
1601                                                       fjx0,fjy0,fjz0,fjx1,fjy1,fjz1,fjx2,fjy2,fjz2);
1602
1603             /* Inner loop uses 594 flops */
1604         }
1605
1606         /* End of innermost loop */
1607
1608         gmx_mm256_update_iforce_3atom_swizzle_pd(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,
1609                                                  f+i_coord_offset,fshift+i_shift_offset);
1610
1611         ggid                        = gid[iidx];
1612         /* Update potential energies */
1613         gmx_mm256_update_1pot_pd(velecsum,kernel_data->energygrp_elec+ggid);
1614
1615         /* Increment number of inner iterations */
1616         inneriter                  += j_index_end - j_index_start;
1617
1618         /* Outer loop uses 19 flops */
1619     }
1620
1621     /* Increment number of outer iterations */
1622     outeriter        += nri;
1623
1624     /* Update outer/inner flops */
1625
1626     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_W3W3_VF,outeriter*19 + inneriter*594);
1627 }
1628 /*
1629  * Gromacs nonbonded kernel:   nb_kernel_ElecEwSw_VdwNone_GeomW3W3_F_avx_256_double
1630  * Electrostatics interaction: Ewald
1631  * VdW interaction:            None
1632  * Geometry:                   Water3-Water3
1633  * Calculate force/pot:        Force
1634  */
1635 void
1636 nb_kernel_ElecEwSw_VdwNone_GeomW3W3_F_avx_256_double
1637                     (t_nblist                    * gmx_restrict       nlist,
1638                      rvec                        * gmx_restrict          xx,
1639                      rvec                        * gmx_restrict          ff,
1640                      t_forcerec                  * gmx_restrict          fr,
1641                      t_mdatoms                   * gmx_restrict     mdatoms,
1642                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
1643                      t_nrnb                      * gmx_restrict        nrnb)
1644 {
1645     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
1646      * just 0 for non-waters.
1647      * Suffixes A,B,C,D refer to j loop unrolling done with AVX, e.g. for the four different
1648      * jnr indices corresponding to data put in the four positions in the SIMD register.
1649      */
1650     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
1651     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
1652     int              jnrA,jnrB,jnrC,jnrD;
1653     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
1654     int              jnrlistE,jnrlistF,jnrlistG,jnrlistH;
1655     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
1656     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
1657     real             rcutoff_scalar;
1658     real             *shiftvec,*fshift,*x,*f;
1659     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD;
1660     real             scratch[4*DIM];
1661     __m256d          tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
1662     real *           vdwioffsetptr0;
1663     __m256d          ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
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     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
1669     __m256d          jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
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     __m256d          dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
1675     __m256d          dx01,dy01,dz01,rsq01,rinv01,rinvsq01,r01,qq01,c6_01,c12_01;
1676     __m256d          dx02,dy02,dz02,rsq02,rinv02,rinvsq02,r02,qq02,c6_02,c12_02;
1677     __m256d          dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
1678     __m256d          dx11,dy11,dz11,rsq11,rinv11,rinvsq11,r11,qq11,c6_11,c12_11;
1679     __m256d          dx12,dy12,dz12,rsq12,rinv12,rinvsq12,r12,qq12,c6_12,c12_12;
1680     __m256d          dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
1681     __m256d          dx21,dy21,dz21,rsq21,rinv21,rinvsq21,r21,qq21,c6_21,c12_21;
1682     __m256d          dx22,dy22,dz22,rsq22,rinv22,rinvsq22,r22,qq22,c6_22,c12_22;
1683     __m256d          velec,felec,velecsum,facel,crf,krf,krf2;
1684     real             *charge;
1685     __m128i          ewitab;
1686     __m256d          ewtabscale,eweps,sh_ewald,ewrt,ewtabhalfspace,ewtabF,ewtabFn,ewtabD,ewtabV;
1687     __m256d          beta,beta2,beta3,zeta2,pmecorrF,pmecorrV,rinv3;
1688     real             *ewtab;
1689     __m256d          rswitch,swV3,swV4,swV5,swF2,swF3,swF4,d,d2,sw,dsw;
1690     real             rswitch_scalar,d_scalar;
1691     __m256d          dummy_mask,cutoff_mask;
1692     __m128           tmpmask0,tmpmask1;
1693     __m256d          signbit = _mm256_castsi256_pd( _mm256_set1_epi32(0x80000000) );
1694     __m256d          one     = _mm256_set1_pd(1.0);
1695     __m256d          two     = _mm256_set1_pd(2.0);
1696     x                = xx[0];
1697     f                = ff[0];
1698
1699     nri              = nlist->nri;
1700     iinr             = nlist->iinr;
1701     jindex           = nlist->jindex;
1702     jjnr             = nlist->jjnr;
1703     shiftidx         = nlist->shift;
1704     gid              = nlist->gid;
1705     shiftvec         = fr->shift_vec[0];
1706     fshift           = fr->fshift[0];
1707     facel            = _mm256_set1_pd(fr->epsfac);
1708     charge           = mdatoms->chargeA;
1709
1710     sh_ewald         = _mm256_set1_pd(fr->ic->sh_ewald);
1711     beta             = _mm256_set1_pd(fr->ic->ewaldcoeff_q);
1712     beta2            = _mm256_mul_pd(beta,beta);
1713     beta3            = _mm256_mul_pd(beta,beta2);
1714
1715     ewtab            = fr->ic->tabq_coul_FDV0;
1716     ewtabscale       = _mm256_set1_pd(fr->ic->tabq_scale);
1717     ewtabhalfspace   = _mm256_set1_pd(0.5/fr->ic->tabq_scale);
1718
1719     /* Setup water-specific parameters */
1720     inr              = nlist->iinr[0];
1721     iq0              = _mm256_mul_pd(facel,_mm256_set1_pd(charge[inr+0]));
1722     iq1              = _mm256_mul_pd(facel,_mm256_set1_pd(charge[inr+1]));
1723     iq2              = _mm256_mul_pd(facel,_mm256_set1_pd(charge[inr+2]));
1724
1725     jq0              = _mm256_set1_pd(charge[inr+0]);
1726     jq1              = _mm256_set1_pd(charge[inr+1]);
1727     jq2              = _mm256_set1_pd(charge[inr+2]);
1728     qq00             = _mm256_mul_pd(iq0,jq0);
1729     qq01             = _mm256_mul_pd(iq0,jq1);
1730     qq02             = _mm256_mul_pd(iq0,jq2);
1731     qq10             = _mm256_mul_pd(iq1,jq0);
1732     qq11             = _mm256_mul_pd(iq1,jq1);
1733     qq12             = _mm256_mul_pd(iq1,jq2);
1734     qq20             = _mm256_mul_pd(iq2,jq0);
1735     qq21             = _mm256_mul_pd(iq2,jq1);
1736     qq22             = _mm256_mul_pd(iq2,jq2);
1737
1738     /* When we use explicit cutoffs the value must be identical for elec and VdW, so use elec as an arbitrary choice */
1739     rcutoff_scalar   = fr->rcoulomb;
1740     rcutoff          = _mm256_set1_pd(rcutoff_scalar);
1741     rcutoff2         = _mm256_mul_pd(rcutoff,rcutoff);
1742
1743     rswitch_scalar   = fr->rcoulomb_switch;
1744     rswitch          = _mm256_set1_pd(rswitch_scalar);
1745     /* Setup switch parameters */
1746     d_scalar         = rcutoff_scalar-rswitch_scalar;
1747     d                = _mm256_set1_pd(d_scalar);
1748     swV3             = _mm256_set1_pd(-10.0/(d_scalar*d_scalar*d_scalar));
1749     swV4             = _mm256_set1_pd( 15.0/(d_scalar*d_scalar*d_scalar*d_scalar));
1750     swV5             = _mm256_set1_pd( -6.0/(d_scalar*d_scalar*d_scalar*d_scalar*d_scalar));
1751     swF2             = _mm256_set1_pd(-30.0/(d_scalar*d_scalar*d_scalar));
1752     swF3             = _mm256_set1_pd( 60.0/(d_scalar*d_scalar*d_scalar*d_scalar));
1753     swF4             = _mm256_set1_pd(-30.0/(d_scalar*d_scalar*d_scalar*d_scalar*d_scalar));
1754
1755     /* Avoid stupid compiler warnings */
1756     jnrA = jnrB = jnrC = jnrD = 0;
1757     j_coord_offsetA = 0;
1758     j_coord_offsetB = 0;
1759     j_coord_offsetC = 0;
1760     j_coord_offsetD = 0;
1761
1762     outeriter        = 0;
1763     inneriter        = 0;
1764
1765     for(iidx=0;iidx<4*DIM;iidx++)
1766     {
1767         scratch[iidx] = 0.0;
1768     }
1769
1770     /* Start outer loop over neighborlists */
1771     for(iidx=0; iidx<nri; iidx++)
1772     {
1773         /* Load shift vector for this list */
1774         i_shift_offset   = DIM*shiftidx[iidx];
1775
1776         /* Load limits for loop over neighbors */
1777         j_index_start    = jindex[iidx];
1778         j_index_end      = jindex[iidx+1];
1779
1780         /* Get outer coordinate index */
1781         inr              = iinr[iidx];
1782         i_coord_offset   = DIM*inr;
1783
1784         /* Load i particle coords and add shift vector */
1785         gmx_mm256_load_shift_and_3rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset,
1786                                                     &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2);
1787
1788         fix0             = _mm256_setzero_pd();
1789         fiy0             = _mm256_setzero_pd();
1790         fiz0             = _mm256_setzero_pd();
1791         fix1             = _mm256_setzero_pd();
1792         fiy1             = _mm256_setzero_pd();
1793         fiz1             = _mm256_setzero_pd();
1794         fix2             = _mm256_setzero_pd();
1795         fiy2             = _mm256_setzero_pd();
1796         fiz2             = _mm256_setzero_pd();
1797
1798         /* Start inner kernel loop */
1799         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
1800         {
1801
1802             /* Get j neighbor index, and coordinate index */
1803             jnrA             = jjnr[jidx];
1804             jnrB             = jjnr[jidx+1];
1805             jnrC             = jjnr[jidx+2];
1806             jnrD             = jjnr[jidx+3];
1807             j_coord_offsetA  = DIM*jnrA;
1808             j_coord_offsetB  = DIM*jnrB;
1809             j_coord_offsetC  = DIM*jnrC;
1810             j_coord_offsetD  = DIM*jnrD;
1811
1812             /* load j atom coordinates */
1813             gmx_mm256_load_3rvec_4ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
1814                                                  x+j_coord_offsetC,x+j_coord_offsetD,
1815                                               &jx0,&jy0,&jz0,&jx1,&jy1,&jz1,&jx2,&jy2,&jz2);
1816
1817             /* Calculate displacement vector */
1818             dx00             = _mm256_sub_pd(ix0,jx0);
1819             dy00             = _mm256_sub_pd(iy0,jy0);
1820             dz00             = _mm256_sub_pd(iz0,jz0);
1821             dx01             = _mm256_sub_pd(ix0,jx1);
1822             dy01             = _mm256_sub_pd(iy0,jy1);
1823             dz01             = _mm256_sub_pd(iz0,jz1);
1824             dx02             = _mm256_sub_pd(ix0,jx2);
1825             dy02             = _mm256_sub_pd(iy0,jy2);
1826             dz02             = _mm256_sub_pd(iz0,jz2);
1827             dx10             = _mm256_sub_pd(ix1,jx0);
1828             dy10             = _mm256_sub_pd(iy1,jy0);
1829             dz10             = _mm256_sub_pd(iz1,jz0);
1830             dx11             = _mm256_sub_pd(ix1,jx1);
1831             dy11             = _mm256_sub_pd(iy1,jy1);
1832             dz11             = _mm256_sub_pd(iz1,jz1);
1833             dx12             = _mm256_sub_pd(ix1,jx2);
1834             dy12             = _mm256_sub_pd(iy1,jy2);
1835             dz12             = _mm256_sub_pd(iz1,jz2);
1836             dx20             = _mm256_sub_pd(ix2,jx0);
1837             dy20             = _mm256_sub_pd(iy2,jy0);
1838             dz20             = _mm256_sub_pd(iz2,jz0);
1839             dx21             = _mm256_sub_pd(ix2,jx1);
1840             dy21             = _mm256_sub_pd(iy2,jy1);
1841             dz21             = _mm256_sub_pd(iz2,jz1);
1842             dx22             = _mm256_sub_pd(ix2,jx2);
1843             dy22             = _mm256_sub_pd(iy2,jy2);
1844             dz22             = _mm256_sub_pd(iz2,jz2);
1845
1846             /* Calculate squared distance and things based on it */
1847             rsq00            = gmx_mm256_calc_rsq_pd(dx00,dy00,dz00);
1848             rsq01            = gmx_mm256_calc_rsq_pd(dx01,dy01,dz01);
1849             rsq02            = gmx_mm256_calc_rsq_pd(dx02,dy02,dz02);
1850             rsq10            = gmx_mm256_calc_rsq_pd(dx10,dy10,dz10);
1851             rsq11            = gmx_mm256_calc_rsq_pd(dx11,dy11,dz11);
1852             rsq12            = gmx_mm256_calc_rsq_pd(dx12,dy12,dz12);
1853             rsq20            = gmx_mm256_calc_rsq_pd(dx20,dy20,dz20);
1854             rsq21            = gmx_mm256_calc_rsq_pd(dx21,dy21,dz21);
1855             rsq22            = gmx_mm256_calc_rsq_pd(dx22,dy22,dz22);
1856
1857             rinv00           = gmx_mm256_invsqrt_pd(rsq00);
1858             rinv01           = gmx_mm256_invsqrt_pd(rsq01);
1859             rinv02           = gmx_mm256_invsqrt_pd(rsq02);
1860             rinv10           = gmx_mm256_invsqrt_pd(rsq10);
1861             rinv11           = gmx_mm256_invsqrt_pd(rsq11);
1862             rinv12           = gmx_mm256_invsqrt_pd(rsq12);
1863             rinv20           = gmx_mm256_invsqrt_pd(rsq20);
1864             rinv21           = gmx_mm256_invsqrt_pd(rsq21);
1865             rinv22           = gmx_mm256_invsqrt_pd(rsq22);
1866
1867             rinvsq00         = _mm256_mul_pd(rinv00,rinv00);
1868             rinvsq01         = _mm256_mul_pd(rinv01,rinv01);
1869             rinvsq02         = _mm256_mul_pd(rinv02,rinv02);
1870             rinvsq10         = _mm256_mul_pd(rinv10,rinv10);
1871             rinvsq11         = _mm256_mul_pd(rinv11,rinv11);
1872             rinvsq12         = _mm256_mul_pd(rinv12,rinv12);
1873             rinvsq20         = _mm256_mul_pd(rinv20,rinv20);
1874             rinvsq21         = _mm256_mul_pd(rinv21,rinv21);
1875             rinvsq22         = _mm256_mul_pd(rinv22,rinv22);
1876
1877             fjx0             = _mm256_setzero_pd();
1878             fjy0             = _mm256_setzero_pd();
1879             fjz0             = _mm256_setzero_pd();
1880             fjx1             = _mm256_setzero_pd();
1881             fjy1             = _mm256_setzero_pd();
1882             fjz1             = _mm256_setzero_pd();
1883             fjx2             = _mm256_setzero_pd();
1884             fjy2             = _mm256_setzero_pd();
1885             fjz2             = _mm256_setzero_pd();
1886
1887             /**************************
1888              * CALCULATE INTERACTIONS *
1889              **************************/
1890
1891             if (gmx_mm256_any_lt(rsq00,rcutoff2))
1892             {
1893
1894             r00              = _mm256_mul_pd(rsq00,rinv00);
1895
1896             /* EWALD ELECTROSTATICS */
1897
1898             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1899             ewrt             = _mm256_mul_pd(r00,ewtabscale);
1900             ewitab           = _mm256_cvttpd_epi32(ewrt);
1901             eweps            = _mm256_sub_pd(ewrt,_mm256_round_pd(ewrt, _MM_FROUND_FLOOR));
1902             ewitab           = _mm_slli_epi32(ewitab,2);
1903             ewtabF           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
1904             ewtabD           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,1) );
1905             ewtabV           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,2) );
1906             ewtabFn          = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,3) );
1907             GMX_MM256_FULLTRANSPOSE4_PD(ewtabF,ewtabD,ewtabV,ewtabFn);
1908             felec            = _mm256_add_pd(ewtabF,_mm256_mul_pd(eweps,ewtabD));
1909             velec            = _mm256_sub_pd(ewtabV,_mm256_mul_pd(_mm256_mul_pd(ewtabhalfspace,eweps),_mm256_add_pd(ewtabF,felec)));
1910             velec            = _mm256_mul_pd(qq00,_mm256_sub_pd(rinv00,velec));
1911             felec            = _mm256_mul_pd(_mm256_mul_pd(qq00,rinv00),_mm256_sub_pd(rinvsq00,felec));
1912
1913             d                = _mm256_sub_pd(r00,rswitch);
1914             d                = _mm256_max_pd(d,_mm256_setzero_pd());
1915             d2               = _mm256_mul_pd(d,d);
1916             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)))))));
1917
1918             dsw              = _mm256_mul_pd(d2,_mm256_add_pd(swF2,_mm256_mul_pd(d,_mm256_add_pd(swF3,_mm256_mul_pd(d,swF4)))));
1919
1920             /* Evaluate switch function */
1921             /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
1922             felec            = _mm256_sub_pd( _mm256_mul_pd(felec,sw) , _mm256_mul_pd(rinv00,_mm256_mul_pd(velec,dsw)) );
1923             cutoff_mask      = _mm256_cmp_pd(rsq00,rcutoff2,_CMP_LT_OQ);
1924
1925             fscal            = felec;
1926
1927             fscal            = _mm256_and_pd(fscal,cutoff_mask);
1928
1929             /* Calculate temporary vectorial force */
1930             tx               = _mm256_mul_pd(fscal,dx00);
1931             ty               = _mm256_mul_pd(fscal,dy00);
1932             tz               = _mm256_mul_pd(fscal,dz00);
1933
1934             /* Update vectorial force */
1935             fix0             = _mm256_add_pd(fix0,tx);
1936             fiy0             = _mm256_add_pd(fiy0,ty);
1937             fiz0             = _mm256_add_pd(fiz0,tz);
1938
1939             fjx0             = _mm256_add_pd(fjx0,tx);
1940             fjy0             = _mm256_add_pd(fjy0,ty);
1941             fjz0             = _mm256_add_pd(fjz0,tz);
1942
1943             }
1944
1945             /**************************
1946              * CALCULATE INTERACTIONS *
1947              **************************/
1948
1949             if (gmx_mm256_any_lt(rsq01,rcutoff2))
1950             {
1951
1952             r01              = _mm256_mul_pd(rsq01,rinv01);
1953
1954             /* EWALD ELECTROSTATICS */
1955
1956             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1957             ewrt             = _mm256_mul_pd(r01,ewtabscale);
1958             ewitab           = _mm256_cvttpd_epi32(ewrt);
1959             eweps            = _mm256_sub_pd(ewrt,_mm256_round_pd(ewrt, _MM_FROUND_FLOOR));
1960             ewitab           = _mm_slli_epi32(ewitab,2);
1961             ewtabF           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
1962             ewtabD           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,1) );
1963             ewtabV           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,2) );
1964             ewtabFn          = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,3) );
1965             GMX_MM256_FULLTRANSPOSE4_PD(ewtabF,ewtabD,ewtabV,ewtabFn);
1966             felec            = _mm256_add_pd(ewtabF,_mm256_mul_pd(eweps,ewtabD));
1967             velec            = _mm256_sub_pd(ewtabV,_mm256_mul_pd(_mm256_mul_pd(ewtabhalfspace,eweps),_mm256_add_pd(ewtabF,felec)));
1968             velec            = _mm256_mul_pd(qq01,_mm256_sub_pd(rinv01,velec));
1969             felec            = _mm256_mul_pd(_mm256_mul_pd(qq01,rinv01),_mm256_sub_pd(rinvsq01,felec));
1970
1971             d                = _mm256_sub_pd(r01,rswitch);
1972             d                = _mm256_max_pd(d,_mm256_setzero_pd());
1973             d2               = _mm256_mul_pd(d,d);
1974             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)))))));
1975
1976             dsw              = _mm256_mul_pd(d2,_mm256_add_pd(swF2,_mm256_mul_pd(d,_mm256_add_pd(swF3,_mm256_mul_pd(d,swF4)))));
1977
1978             /* Evaluate switch function */
1979             /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
1980             felec            = _mm256_sub_pd( _mm256_mul_pd(felec,sw) , _mm256_mul_pd(rinv01,_mm256_mul_pd(velec,dsw)) );
1981             cutoff_mask      = _mm256_cmp_pd(rsq01,rcutoff2,_CMP_LT_OQ);
1982
1983             fscal            = felec;
1984
1985             fscal            = _mm256_and_pd(fscal,cutoff_mask);
1986
1987             /* Calculate temporary vectorial force */
1988             tx               = _mm256_mul_pd(fscal,dx01);
1989             ty               = _mm256_mul_pd(fscal,dy01);
1990             tz               = _mm256_mul_pd(fscal,dz01);
1991
1992             /* Update vectorial force */
1993             fix0             = _mm256_add_pd(fix0,tx);
1994             fiy0             = _mm256_add_pd(fiy0,ty);
1995             fiz0             = _mm256_add_pd(fiz0,tz);
1996
1997             fjx1             = _mm256_add_pd(fjx1,tx);
1998             fjy1             = _mm256_add_pd(fjy1,ty);
1999             fjz1             = _mm256_add_pd(fjz1,tz);
2000
2001             }
2002
2003             /**************************
2004              * CALCULATE INTERACTIONS *
2005              **************************/
2006
2007             if (gmx_mm256_any_lt(rsq02,rcutoff2))
2008             {
2009
2010             r02              = _mm256_mul_pd(rsq02,rinv02);
2011
2012             /* EWALD ELECTROSTATICS */
2013
2014             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
2015             ewrt             = _mm256_mul_pd(r02,ewtabscale);
2016             ewitab           = _mm256_cvttpd_epi32(ewrt);
2017             eweps            = _mm256_sub_pd(ewrt,_mm256_round_pd(ewrt, _MM_FROUND_FLOOR));
2018             ewitab           = _mm_slli_epi32(ewitab,2);
2019             ewtabF           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
2020             ewtabD           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,1) );
2021             ewtabV           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,2) );
2022             ewtabFn          = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,3) );
2023             GMX_MM256_FULLTRANSPOSE4_PD(ewtabF,ewtabD,ewtabV,ewtabFn);
2024             felec            = _mm256_add_pd(ewtabF,_mm256_mul_pd(eweps,ewtabD));
2025             velec            = _mm256_sub_pd(ewtabV,_mm256_mul_pd(_mm256_mul_pd(ewtabhalfspace,eweps),_mm256_add_pd(ewtabF,felec)));
2026             velec            = _mm256_mul_pd(qq02,_mm256_sub_pd(rinv02,velec));
2027             felec            = _mm256_mul_pd(_mm256_mul_pd(qq02,rinv02),_mm256_sub_pd(rinvsq02,felec));
2028
2029             d                = _mm256_sub_pd(r02,rswitch);
2030             d                = _mm256_max_pd(d,_mm256_setzero_pd());
2031             d2               = _mm256_mul_pd(d,d);
2032             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)))))));
2033
2034             dsw              = _mm256_mul_pd(d2,_mm256_add_pd(swF2,_mm256_mul_pd(d,_mm256_add_pd(swF3,_mm256_mul_pd(d,swF4)))));
2035
2036             /* Evaluate switch function */
2037             /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
2038             felec            = _mm256_sub_pd( _mm256_mul_pd(felec,sw) , _mm256_mul_pd(rinv02,_mm256_mul_pd(velec,dsw)) );
2039             cutoff_mask      = _mm256_cmp_pd(rsq02,rcutoff2,_CMP_LT_OQ);
2040
2041             fscal            = felec;
2042
2043             fscal            = _mm256_and_pd(fscal,cutoff_mask);
2044
2045             /* Calculate temporary vectorial force */
2046             tx               = _mm256_mul_pd(fscal,dx02);
2047             ty               = _mm256_mul_pd(fscal,dy02);
2048             tz               = _mm256_mul_pd(fscal,dz02);
2049
2050             /* Update vectorial force */
2051             fix0             = _mm256_add_pd(fix0,tx);
2052             fiy0             = _mm256_add_pd(fiy0,ty);
2053             fiz0             = _mm256_add_pd(fiz0,tz);
2054
2055             fjx2             = _mm256_add_pd(fjx2,tx);
2056             fjy2             = _mm256_add_pd(fjy2,ty);
2057             fjz2             = _mm256_add_pd(fjz2,tz);
2058
2059             }
2060
2061             /**************************
2062              * CALCULATE INTERACTIONS *
2063              **************************/
2064
2065             if (gmx_mm256_any_lt(rsq10,rcutoff2))
2066             {
2067
2068             r10              = _mm256_mul_pd(rsq10,rinv10);
2069
2070             /* EWALD ELECTROSTATICS */
2071
2072             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
2073             ewrt             = _mm256_mul_pd(r10,ewtabscale);
2074             ewitab           = _mm256_cvttpd_epi32(ewrt);
2075             eweps            = _mm256_sub_pd(ewrt,_mm256_round_pd(ewrt, _MM_FROUND_FLOOR));
2076             ewitab           = _mm_slli_epi32(ewitab,2);
2077             ewtabF           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
2078             ewtabD           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,1) );
2079             ewtabV           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,2) );
2080             ewtabFn          = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,3) );
2081             GMX_MM256_FULLTRANSPOSE4_PD(ewtabF,ewtabD,ewtabV,ewtabFn);
2082             felec            = _mm256_add_pd(ewtabF,_mm256_mul_pd(eweps,ewtabD));
2083             velec            = _mm256_sub_pd(ewtabV,_mm256_mul_pd(_mm256_mul_pd(ewtabhalfspace,eweps),_mm256_add_pd(ewtabF,felec)));
2084             velec            = _mm256_mul_pd(qq10,_mm256_sub_pd(rinv10,velec));
2085             felec            = _mm256_mul_pd(_mm256_mul_pd(qq10,rinv10),_mm256_sub_pd(rinvsq10,felec));
2086
2087             d                = _mm256_sub_pd(r10,rswitch);
2088             d                = _mm256_max_pd(d,_mm256_setzero_pd());
2089             d2               = _mm256_mul_pd(d,d);
2090             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)))))));
2091
2092             dsw              = _mm256_mul_pd(d2,_mm256_add_pd(swF2,_mm256_mul_pd(d,_mm256_add_pd(swF3,_mm256_mul_pd(d,swF4)))));
2093
2094             /* Evaluate switch function */
2095             /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
2096             felec            = _mm256_sub_pd( _mm256_mul_pd(felec,sw) , _mm256_mul_pd(rinv10,_mm256_mul_pd(velec,dsw)) );
2097             cutoff_mask      = _mm256_cmp_pd(rsq10,rcutoff2,_CMP_LT_OQ);
2098
2099             fscal            = felec;
2100
2101             fscal            = _mm256_and_pd(fscal,cutoff_mask);
2102
2103             /* Calculate temporary vectorial force */
2104             tx               = _mm256_mul_pd(fscal,dx10);
2105             ty               = _mm256_mul_pd(fscal,dy10);
2106             tz               = _mm256_mul_pd(fscal,dz10);
2107
2108             /* Update vectorial force */
2109             fix1             = _mm256_add_pd(fix1,tx);
2110             fiy1             = _mm256_add_pd(fiy1,ty);
2111             fiz1             = _mm256_add_pd(fiz1,tz);
2112
2113             fjx0             = _mm256_add_pd(fjx0,tx);
2114             fjy0             = _mm256_add_pd(fjy0,ty);
2115             fjz0             = _mm256_add_pd(fjz0,tz);
2116
2117             }
2118
2119             /**************************
2120              * CALCULATE INTERACTIONS *
2121              **************************/
2122
2123             if (gmx_mm256_any_lt(rsq11,rcutoff2))
2124             {
2125
2126             r11              = _mm256_mul_pd(rsq11,rinv11);
2127
2128             /* EWALD ELECTROSTATICS */
2129
2130             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
2131             ewrt             = _mm256_mul_pd(r11,ewtabscale);
2132             ewitab           = _mm256_cvttpd_epi32(ewrt);
2133             eweps            = _mm256_sub_pd(ewrt,_mm256_round_pd(ewrt, _MM_FROUND_FLOOR));
2134             ewitab           = _mm_slli_epi32(ewitab,2);
2135             ewtabF           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
2136             ewtabD           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,1) );
2137             ewtabV           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,2) );
2138             ewtabFn          = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,3) );
2139             GMX_MM256_FULLTRANSPOSE4_PD(ewtabF,ewtabD,ewtabV,ewtabFn);
2140             felec            = _mm256_add_pd(ewtabF,_mm256_mul_pd(eweps,ewtabD));
2141             velec            = _mm256_sub_pd(ewtabV,_mm256_mul_pd(_mm256_mul_pd(ewtabhalfspace,eweps),_mm256_add_pd(ewtabF,felec)));
2142             velec            = _mm256_mul_pd(qq11,_mm256_sub_pd(rinv11,velec));
2143             felec            = _mm256_mul_pd(_mm256_mul_pd(qq11,rinv11),_mm256_sub_pd(rinvsq11,felec));
2144
2145             d                = _mm256_sub_pd(r11,rswitch);
2146             d                = _mm256_max_pd(d,_mm256_setzero_pd());
2147             d2               = _mm256_mul_pd(d,d);
2148             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)))))));
2149
2150             dsw              = _mm256_mul_pd(d2,_mm256_add_pd(swF2,_mm256_mul_pd(d,_mm256_add_pd(swF3,_mm256_mul_pd(d,swF4)))));
2151
2152             /* Evaluate switch function */
2153             /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
2154             felec            = _mm256_sub_pd( _mm256_mul_pd(felec,sw) , _mm256_mul_pd(rinv11,_mm256_mul_pd(velec,dsw)) );
2155             cutoff_mask      = _mm256_cmp_pd(rsq11,rcutoff2,_CMP_LT_OQ);
2156
2157             fscal            = felec;
2158
2159             fscal            = _mm256_and_pd(fscal,cutoff_mask);
2160
2161             /* Calculate temporary vectorial force */
2162             tx               = _mm256_mul_pd(fscal,dx11);
2163             ty               = _mm256_mul_pd(fscal,dy11);
2164             tz               = _mm256_mul_pd(fscal,dz11);
2165
2166             /* Update vectorial force */
2167             fix1             = _mm256_add_pd(fix1,tx);
2168             fiy1             = _mm256_add_pd(fiy1,ty);
2169             fiz1             = _mm256_add_pd(fiz1,tz);
2170
2171             fjx1             = _mm256_add_pd(fjx1,tx);
2172             fjy1             = _mm256_add_pd(fjy1,ty);
2173             fjz1             = _mm256_add_pd(fjz1,tz);
2174
2175             }
2176
2177             /**************************
2178              * CALCULATE INTERACTIONS *
2179              **************************/
2180
2181             if (gmx_mm256_any_lt(rsq12,rcutoff2))
2182             {
2183
2184             r12              = _mm256_mul_pd(rsq12,rinv12);
2185
2186             /* EWALD ELECTROSTATICS */
2187
2188             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
2189             ewrt             = _mm256_mul_pd(r12,ewtabscale);
2190             ewitab           = _mm256_cvttpd_epi32(ewrt);
2191             eweps            = _mm256_sub_pd(ewrt,_mm256_round_pd(ewrt, _MM_FROUND_FLOOR));
2192             ewitab           = _mm_slli_epi32(ewitab,2);
2193             ewtabF           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
2194             ewtabD           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,1) );
2195             ewtabV           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,2) );
2196             ewtabFn          = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,3) );
2197             GMX_MM256_FULLTRANSPOSE4_PD(ewtabF,ewtabD,ewtabV,ewtabFn);
2198             felec            = _mm256_add_pd(ewtabF,_mm256_mul_pd(eweps,ewtabD));
2199             velec            = _mm256_sub_pd(ewtabV,_mm256_mul_pd(_mm256_mul_pd(ewtabhalfspace,eweps),_mm256_add_pd(ewtabF,felec)));
2200             velec            = _mm256_mul_pd(qq12,_mm256_sub_pd(rinv12,velec));
2201             felec            = _mm256_mul_pd(_mm256_mul_pd(qq12,rinv12),_mm256_sub_pd(rinvsq12,felec));
2202
2203             d                = _mm256_sub_pd(r12,rswitch);
2204             d                = _mm256_max_pd(d,_mm256_setzero_pd());
2205             d2               = _mm256_mul_pd(d,d);
2206             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)))))));
2207
2208             dsw              = _mm256_mul_pd(d2,_mm256_add_pd(swF2,_mm256_mul_pd(d,_mm256_add_pd(swF3,_mm256_mul_pd(d,swF4)))));
2209
2210             /* Evaluate switch function */
2211             /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
2212             felec            = _mm256_sub_pd( _mm256_mul_pd(felec,sw) , _mm256_mul_pd(rinv12,_mm256_mul_pd(velec,dsw)) );
2213             cutoff_mask      = _mm256_cmp_pd(rsq12,rcutoff2,_CMP_LT_OQ);
2214
2215             fscal            = felec;
2216
2217             fscal            = _mm256_and_pd(fscal,cutoff_mask);
2218
2219             /* Calculate temporary vectorial force */
2220             tx               = _mm256_mul_pd(fscal,dx12);
2221             ty               = _mm256_mul_pd(fscal,dy12);
2222             tz               = _mm256_mul_pd(fscal,dz12);
2223
2224             /* Update vectorial force */
2225             fix1             = _mm256_add_pd(fix1,tx);
2226             fiy1             = _mm256_add_pd(fiy1,ty);
2227             fiz1             = _mm256_add_pd(fiz1,tz);
2228
2229             fjx2             = _mm256_add_pd(fjx2,tx);
2230             fjy2             = _mm256_add_pd(fjy2,ty);
2231             fjz2             = _mm256_add_pd(fjz2,tz);
2232
2233             }
2234
2235             /**************************
2236              * CALCULATE INTERACTIONS *
2237              **************************/
2238
2239             if (gmx_mm256_any_lt(rsq20,rcutoff2))
2240             {
2241
2242             r20              = _mm256_mul_pd(rsq20,rinv20);
2243
2244             /* EWALD ELECTROSTATICS */
2245
2246             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
2247             ewrt             = _mm256_mul_pd(r20,ewtabscale);
2248             ewitab           = _mm256_cvttpd_epi32(ewrt);
2249             eweps            = _mm256_sub_pd(ewrt,_mm256_round_pd(ewrt, _MM_FROUND_FLOOR));
2250             ewitab           = _mm_slli_epi32(ewitab,2);
2251             ewtabF           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
2252             ewtabD           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,1) );
2253             ewtabV           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,2) );
2254             ewtabFn          = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,3) );
2255             GMX_MM256_FULLTRANSPOSE4_PD(ewtabF,ewtabD,ewtabV,ewtabFn);
2256             felec            = _mm256_add_pd(ewtabF,_mm256_mul_pd(eweps,ewtabD));
2257             velec            = _mm256_sub_pd(ewtabV,_mm256_mul_pd(_mm256_mul_pd(ewtabhalfspace,eweps),_mm256_add_pd(ewtabF,felec)));
2258             velec            = _mm256_mul_pd(qq20,_mm256_sub_pd(rinv20,velec));
2259             felec            = _mm256_mul_pd(_mm256_mul_pd(qq20,rinv20),_mm256_sub_pd(rinvsq20,felec));
2260
2261             d                = _mm256_sub_pd(r20,rswitch);
2262             d                = _mm256_max_pd(d,_mm256_setzero_pd());
2263             d2               = _mm256_mul_pd(d,d);
2264             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)))))));
2265
2266             dsw              = _mm256_mul_pd(d2,_mm256_add_pd(swF2,_mm256_mul_pd(d,_mm256_add_pd(swF3,_mm256_mul_pd(d,swF4)))));
2267
2268             /* Evaluate switch function */
2269             /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
2270             felec            = _mm256_sub_pd( _mm256_mul_pd(felec,sw) , _mm256_mul_pd(rinv20,_mm256_mul_pd(velec,dsw)) );
2271             cutoff_mask      = _mm256_cmp_pd(rsq20,rcutoff2,_CMP_LT_OQ);
2272
2273             fscal            = felec;
2274
2275             fscal            = _mm256_and_pd(fscal,cutoff_mask);
2276
2277             /* Calculate temporary vectorial force */
2278             tx               = _mm256_mul_pd(fscal,dx20);
2279             ty               = _mm256_mul_pd(fscal,dy20);
2280             tz               = _mm256_mul_pd(fscal,dz20);
2281
2282             /* Update vectorial force */
2283             fix2             = _mm256_add_pd(fix2,tx);
2284             fiy2             = _mm256_add_pd(fiy2,ty);
2285             fiz2             = _mm256_add_pd(fiz2,tz);
2286
2287             fjx0             = _mm256_add_pd(fjx0,tx);
2288             fjy0             = _mm256_add_pd(fjy0,ty);
2289             fjz0             = _mm256_add_pd(fjz0,tz);
2290
2291             }
2292
2293             /**************************
2294              * CALCULATE INTERACTIONS *
2295              **************************/
2296
2297             if (gmx_mm256_any_lt(rsq21,rcutoff2))
2298             {
2299
2300             r21              = _mm256_mul_pd(rsq21,rinv21);
2301
2302             /* EWALD ELECTROSTATICS */
2303
2304             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
2305             ewrt             = _mm256_mul_pd(r21,ewtabscale);
2306             ewitab           = _mm256_cvttpd_epi32(ewrt);
2307             eweps            = _mm256_sub_pd(ewrt,_mm256_round_pd(ewrt, _MM_FROUND_FLOOR));
2308             ewitab           = _mm_slli_epi32(ewitab,2);
2309             ewtabF           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
2310             ewtabD           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,1) );
2311             ewtabV           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,2) );
2312             ewtabFn          = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,3) );
2313             GMX_MM256_FULLTRANSPOSE4_PD(ewtabF,ewtabD,ewtabV,ewtabFn);
2314             felec            = _mm256_add_pd(ewtabF,_mm256_mul_pd(eweps,ewtabD));
2315             velec            = _mm256_sub_pd(ewtabV,_mm256_mul_pd(_mm256_mul_pd(ewtabhalfspace,eweps),_mm256_add_pd(ewtabF,felec)));
2316             velec            = _mm256_mul_pd(qq21,_mm256_sub_pd(rinv21,velec));
2317             felec            = _mm256_mul_pd(_mm256_mul_pd(qq21,rinv21),_mm256_sub_pd(rinvsq21,felec));
2318
2319             d                = _mm256_sub_pd(r21,rswitch);
2320             d                = _mm256_max_pd(d,_mm256_setzero_pd());
2321             d2               = _mm256_mul_pd(d,d);
2322             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)))))));
2323
2324             dsw              = _mm256_mul_pd(d2,_mm256_add_pd(swF2,_mm256_mul_pd(d,_mm256_add_pd(swF3,_mm256_mul_pd(d,swF4)))));
2325
2326             /* Evaluate switch function */
2327             /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
2328             felec            = _mm256_sub_pd( _mm256_mul_pd(felec,sw) , _mm256_mul_pd(rinv21,_mm256_mul_pd(velec,dsw)) );
2329             cutoff_mask      = _mm256_cmp_pd(rsq21,rcutoff2,_CMP_LT_OQ);
2330
2331             fscal            = felec;
2332
2333             fscal            = _mm256_and_pd(fscal,cutoff_mask);
2334
2335             /* Calculate temporary vectorial force */
2336             tx               = _mm256_mul_pd(fscal,dx21);
2337             ty               = _mm256_mul_pd(fscal,dy21);
2338             tz               = _mm256_mul_pd(fscal,dz21);
2339
2340             /* Update vectorial force */
2341             fix2             = _mm256_add_pd(fix2,tx);
2342             fiy2             = _mm256_add_pd(fiy2,ty);
2343             fiz2             = _mm256_add_pd(fiz2,tz);
2344
2345             fjx1             = _mm256_add_pd(fjx1,tx);
2346             fjy1             = _mm256_add_pd(fjy1,ty);
2347             fjz1             = _mm256_add_pd(fjz1,tz);
2348
2349             }
2350
2351             /**************************
2352              * CALCULATE INTERACTIONS *
2353              **************************/
2354
2355             if (gmx_mm256_any_lt(rsq22,rcutoff2))
2356             {
2357
2358             r22              = _mm256_mul_pd(rsq22,rinv22);
2359
2360             /* EWALD ELECTROSTATICS */
2361
2362             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
2363             ewrt             = _mm256_mul_pd(r22,ewtabscale);
2364             ewitab           = _mm256_cvttpd_epi32(ewrt);
2365             eweps            = _mm256_sub_pd(ewrt,_mm256_round_pd(ewrt, _MM_FROUND_FLOOR));
2366             ewitab           = _mm_slli_epi32(ewitab,2);
2367             ewtabF           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
2368             ewtabD           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,1) );
2369             ewtabV           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,2) );
2370             ewtabFn          = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,3) );
2371             GMX_MM256_FULLTRANSPOSE4_PD(ewtabF,ewtabD,ewtabV,ewtabFn);
2372             felec            = _mm256_add_pd(ewtabF,_mm256_mul_pd(eweps,ewtabD));
2373             velec            = _mm256_sub_pd(ewtabV,_mm256_mul_pd(_mm256_mul_pd(ewtabhalfspace,eweps),_mm256_add_pd(ewtabF,felec)));
2374             velec            = _mm256_mul_pd(qq22,_mm256_sub_pd(rinv22,velec));
2375             felec            = _mm256_mul_pd(_mm256_mul_pd(qq22,rinv22),_mm256_sub_pd(rinvsq22,felec));
2376
2377             d                = _mm256_sub_pd(r22,rswitch);
2378             d                = _mm256_max_pd(d,_mm256_setzero_pd());
2379             d2               = _mm256_mul_pd(d,d);
2380             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)))))));
2381
2382             dsw              = _mm256_mul_pd(d2,_mm256_add_pd(swF2,_mm256_mul_pd(d,_mm256_add_pd(swF3,_mm256_mul_pd(d,swF4)))));
2383
2384             /* Evaluate switch function */
2385             /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
2386             felec            = _mm256_sub_pd( _mm256_mul_pd(felec,sw) , _mm256_mul_pd(rinv22,_mm256_mul_pd(velec,dsw)) );
2387             cutoff_mask      = _mm256_cmp_pd(rsq22,rcutoff2,_CMP_LT_OQ);
2388
2389             fscal            = felec;
2390
2391             fscal            = _mm256_and_pd(fscal,cutoff_mask);
2392
2393             /* Calculate temporary vectorial force */
2394             tx               = _mm256_mul_pd(fscal,dx22);
2395             ty               = _mm256_mul_pd(fscal,dy22);
2396             tz               = _mm256_mul_pd(fscal,dz22);
2397
2398             /* Update vectorial force */
2399             fix2             = _mm256_add_pd(fix2,tx);
2400             fiy2             = _mm256_add_pd(fiy2,ty);
2401             fiz2             = _mm256_add_pd(fiz2,tz);
2402
2403             fjx2             = _mm256_add_pd(fjx2,tx);
2404             fjy2             = _mm256_add_pd(fjy2,ty);
2405             fjz2             = _mm256_add_pd(fjz2,tz);
2406
2407             }
2408
2409             fjptrA             = f+j_coord_offsetA;
2410             fjptrB             = f+j_coord_offsetB;
2411             fjptrC             = f+j_coord_offsetC;
2412             fjptrD             = f+j_coord_offsetD;
2413
2414             gmx_mm256_decrement_3rvec_4ptr_swizzle_pd(fjptrA,fjptrB,fjptrC,fjptrD,
2415                                                       fjx0,fjy0,fjz0,fjx1,fjy1,fjz1,fjx2,fjy2,fjz2);
2416
2417             /* Inner loop uses 558 flops */
2418         }
2419
2420         if(jidx<j_index_end)
2421         {
2422
2423             /* Get j neighbor index, and coordinate index */
2424             jnrlistA         = jjnr[jidx];
2425             jnrlistB         = jjnr[jidx+1];
2426             jnrlistC         = jjnr[jidx+2];
2427             jnrlistD         = jjnr[jidx+3];
2428             /* Sign of each element will be negative for non-real atoms.
2429              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
2430              * so use it as val = _mm_andnot_pd(mask,val) to clear dummy entries.
2431              */
2432             tmpmask0 = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
2433
2434             tmpmask1 = _mm_permute_ps(tmpmask0,_GMX_MM_PERMUTE(3,3,2,2));
2435             tmpmask0 = _mm_permute_ps(tmpmask0,_GMX_MM_PERMUTE(1,1,0,0));
2436             dummy_mask = _mm256_castps_pd(gmx_mm256_set_m128(tmpmask1,tmpmask0));
2437
2438             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
2439             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
2440             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
2441             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
2442             j_coord_offsetA  = DIM*jnrA;
2443             j_coord_offsetB  = DIM*jnrB;
2444             j_coord_offsetC  = DIM*jnrC;
2445             j_coord_offsetD  = DIM*jnrD;
2446
2447             /* load j atom coordinates */
2448             gmx_mm256_load_3rvec_4ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
2449                                                  x+j_coord_offsetC,x+j_coord_offsetD,
2450                                               &jx0,&jy0,&jz0,&jx1,&jy1,&jz1,&jx2,&jy2,&jz2);
2451
2452             /* Calculate displacement vector */
2453             dx00             = _mm256_sub_pd(ix0,jx0);
2454             dy00             = _mm256_sub_pd(iy0,jy0);
2455             dz00             = _mm256_sub_pd(iz0,jz0);
2456             dx01             = _mm256_sub_pd(ix0,jx1);
2457             dy01             = _mm256_sub_pd(iy0,jy1);
2458             dz01             = _mm256_sub_pd(iz0,jz1);
2459             dx02             = _mm256_sub_pd(ix0,jx2);
2460             dy02             = _mm256_sub_pd(iy0,jy2);
2461             dz02             = _mm256_sub_pd(iz0,jz2);
2462             dx10             = _mm256_sub_pd(ix1,jx0);
2463             dy10             = _mm256_sub_pd(iy1,jy0);
2464             dz10             = _mm256_sub_pd(iz1,jz0);
2465             dx11             = _mm256_sub_pd(ix1,jx1);
2466             dy11             = _mm256_sub_pd(iy1,jy1);
2467             dz11             = _mm256_sub_pd(iz1,jz1);
2468             dx12             = _mm256_sub_pd(ix1,jx2);
2469             dy12             = _mm256_sub_pd(iy1,jy2);
2470             dz12             = _mm256_sub_pd(iz1,jz2);
2471             dx20             = _mm256_sub_pd(ix2,jx0);
2472             dy20             = _mm256_sub_pd(iy2,jy0);
2473             dz20             = _mm256_sub_pd(iz2,jz0);
2474             dx21             = _mm256_sub_pd(ix2,jx1);
2475             dy21             = _mm256_sub_pd(iy2,jy1);
2476             dz21             = _mm256_sub_pd(iz2,jz1);
2477             dx22             = _mm256_sub_pd(ix2,jx2);
2478             dy22             = _mm256_sub_pd(iy2,jy2);
2479             dz22             = _mm256_sub_pd(iz2,jz2);
2480
2481             /* Calculate squared distance and things based on it */
2482             rsq00            = gmx_mm256_calc_rsq_pd(dx00,dy00,dz00);
2483             rsq01            = gmx_mm256_calc_rsq_pd(dx01,dy01,dz01);
2484             rsq02            = gmx_mm256_calc_rsq_pd(dx02,dy02,dz02);
2485             rsq10            = gmx_mm256_calc_rsq_pd(dx10,dy10,dz10);
2486             rsq11            = gmx_mm256_calc_rsq_pd(dx11,dy11,dz11);
2487             rsq12            = gmx_mm256_calc_rsq_pd(dx12,dy12,dz12);
2488             rsq20            = gmx_mm256_calc_rsq_pd(dx20,dy20,dz20);
2489             rsq21            = gmx_mm256_calc_rsq_pd(dx21,dy21,dz21);
2490             rsq22            = gmx_mm256_calc_rsq_pd(dx22,dy22,dz22);
2491
2492             rinv00           = gmx_mm256_invsqrt_pd(rsq00);
2493             rinv01           = gmx_mm256_invsqrt_pd(rsq01);
2494             rinv02           = gmx_mm256_invsqrt_pd(rsq02);
2495             rinv10           = gmx_mm256_invsqrt_pd(rsq10);
2496             rinv11           = gmx_mm256_invsqrt_pd(rsq11);
2497             rinv12           = gmx_mm256_invsqrt_pd(rsq12);
2498             rinv20           = gmx_mm256_invsqrt_pd(rsq20);
2499             rinv21           = gmx_mm256_invsqrt_pd(rsq21);
2500             rinv22           = gmx_mm256_invsqrt_pd(rsq22);
2501
2502             rinvsq00         = _mm256_mul_pd(rinv00,rinv00);
2503             rinvsq01         = _mm256_mul_pd(rinv01,rinv01);
2504             rinvsq02         = _mm256_mul_pd(rinv02,rinv02);
2505             rinvsq10         = _mm256_mul_pd(rinv10,rinv10);
2506             rinvsq11         = _mm256_mul_pd(rinv11,rinv11);
2507             rinvsq12         = _mm256_mul_pd(rinv12,rinv12);
2508             rinvsq20         = _mm256_mul_pd(rinv20,rinv20);
2509             rinvsq21         = _mm256_mul_pd(rinv21,rinv21);
2510             rinvsq22         = _mm256_mul_pd(rinv22,rinv22);
2511
2512             fjx0             = _mm256_setzero_pd();
2513             fjy0             = _mm256_setzero_pd();
2514             fjz0             = _mm256_setzero_pd();
2515             fjx1             = _mm256_setzero_pd();
2516             fjy1             = _mm256_setzero_pd();
2517             fjz1             = _mm256_setzero_pd();
2518             fjx2             = _mm256_setzero_pd();
2519             fjy2             = _mm256_setzero_pd();
2520             fjz2             = _mm256_setzero_pd();
2521
2522             /**************************
2523              * CALCULATE INTERACTIONS *
2524              **************************/
2525
2526             if (gmx_mm256_any_lt(rsq00,rcutoff2))
2527             {
2528
2529             r00              = _mm256_mul_pd(rsq00,rinv00);
2530             r00              = _mm256_andnot_pd(dummy_mask,r00);
2531
2532             /* EWALD ELECTROSTATICS */
2533
2534             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
2535             ewrt             = _mm256_mul_pd(r00,ewtabscale);
2536             ewitab           = _mm256_cvttpd_epi32(ewrt);
2537             eweps            = _mm256_sub_pd(ewrt,_mm256_round_pd(ewrt, _MM_FROUND_FLOOR));
2538             ewitab           = _mm_slli_epi32(ewitab,2);
2539             ewtabF           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
2540             ewtabD           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,1) );
2541             ewtabV           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,2) );
2542             ewtabFn          = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,3) );
2543             GMX_MM256_FULLTRANSPOSE4_PD(ewtabF,ewtabD,ewtabV,ewtabFn);
2544             felec            = _mm256_add_pd(ewtabF,_mm256_mul_pd(eweps,ewtabD));
2545             velec            = _mm256_sub_pd(ewtabV,_mm256_mul_pd(_mm256_mul_pd(ewtabhalfspace,eweps),_mm256_add_pd(ewtabF,felec)));
2546             velec            = _mm256_mul_pd(qq00,_mm256_sub_pd(rinv00,velec));
2547             felec            = _mm256_mul_pd(_mm256_mul_pd(qq00,rinv00),_mm256_sub_pd(rinvsq00,felec));
2548
2549             d                = _mm256_sub_pd(r00,rswitch);
2550             d                = _mm256_max_pd(d,_mm256_setzero_pd());
2551             d2               = _mm256_mul_pd(d,d);
2552             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)))))));
2553
2554             dsw              = _mm256_mul_pd(d2,_mm256_add_pd(swF2,_mm256_mul_pd(d,_mm256_add_pd(swF3,_mm256_mul_pd(d,swF4)))));
2555
2556             /* Evaluate switch function */
2557             /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
2558             felec            = _mm256_sub_pd( _mm256_mul_pd(felec,sw) , _mm256_mul_pd(rinv00,_mm256_mul_pd(velec,dsw)) );
2559             cutoff_mask      = _mm256_cmp_pd(rsq00,rcutoff2,_CMP_LT_OQ);
2560
2561             fscal            = felec;
2562
2563             fscal            = _mm256_and_pd(fscal,cutoff_mask);
2564
2565             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
2566
2567             /* Calculate temporary vectorial force */
2568             tx               = _mm256_mul_pd(fscal,dx00);
2569             ty               = _mm256_mul_pd(fscal,dy00);
2570             tz               = _mm256_mul_pd(fscal,dz00);
2571
2572             /* Update vectorial force */
2573             fix0             = _mm256_add_pd(fix0,tx);
2574             fiy0             = _mm256_add_pd(fiy0,ty);
2575             fiz0             = _mm256_add_pd(fiz0,tz);
2576
2577             fjx0             = _mm256_add_pd(fjx0,tx);
2578             fjy0             = _mm256_add_pd(fjy0,ty);
2579             fjz0             = _mm256_add_pd(fjz0,tz);
2580
2581             }
2582
2583             /**************************
2584              * CALCULATE INTERACTIONS *
2585              **************************/
2586
2587             if (gmx_mm256_any_lt(rsq01,rcutoff2))
2588             {
2589
2590             r01              = _mm256_mul_pd(rsq01,rinv01);
2591             r01              = _mm256_andnot_pd(dummy_mask,r01);
2592
2593             /* EWALD ELECTROSTATICS */
2594
2595             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
2596             ewrt             = _mm256_mul_pd(r01,ewtabscale);
2597             ewitab           = _mm256_cvttpd_epi32(ewrt);
2598             eweps            = _mm256_sub_pd(ewrt,_mm256_round_pd(ewrt, _MM_FROUND_FLOOR));
2599             ewitab           = _mm_slli_epi32(ewitab,2);
2600             ewtabF           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
2601             ewtabD           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,1) );
2602             ewtabV           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,2) );
2603             ewtabFn          = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,3) );
2604             GMX_MM256_FULLTRANSPOSE4_PD(ewtabF,ewtabD,ewtabV,ewtabFn);
2605             felec            = _mm256_add_pd(ewtabF,_mm256_mul_pd(eweps,ewtabD));
2606             velec            = _mm256_sub_pd(ewtabV,_mm256_mul_pd(_mm256_mul_pd(ewtabhalfspace,eweps),_mm256_add_pd(ewtabF,felec)));
2607             velec            = _mm256_mul_pd(qq01,_mm256_sub_pd(rinv01,velec));
2608             felec            = _mm256_mul_pd(_mm256_mul_pd(qq01,rinv01),_mm256_sub_pd(rinvsq01,felec));
2609
2610             d                = _mm256_sub_pd(r01,rswitch);
2611             d                = _mm256_max_pd(d,_mm256_setzero_pd());
2612             d2               = _mm256_mul_pd(d,d);
2613             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)))))));
2614
2615             dsw              = _mm256_mul_pd(d2,_mm256_add_pd(swF2,_mm256_mul_pd(d,_mm256_add_pd(swF3,_mm256_mul_pd(d,swF4)))));
2616
2617             /* Evaluate switch function */
2618             /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
2619             felec            = _mm256_sub_pd( _mm256_mul_pd(felec,sw) , _mm256_mul_pd(rinv01,_mm256_mul_pd(velec,dsw)) );
2620             cutoff_mask      = _mm256_cmp_pd(rsq01,rcutoff2,_CMP_LT_OQ);
2621
2622             fscal            = felec;
2623
2624             fscal            = _mm256_and_pd(fscal,cutoff_mask);
2625
2626             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
2627
2628             /* Calculate temporary vectorial force */
2629             tx               = _mm256_mul_pd(fscal,dx01);
2630             ty               = _mm256_mul_pd(fscal,dy01);
2631             tz               = _mm256_mul_pd(fscal,dz01);
2632
2633             /* Update vectorial force */
2634             fix0             = _mm256_add_pd(fix0,tx);
2635             fiy0             = _mm256_add_pd(fiy0,ty);
2636             fiz0             = _mm256_add_pd(fiz0,tz);
2637
2638             fjx1             = _mm256_add_pd(fjx1,tx);
2639             fjy1             = _mm256_add_pd(fjy1,ty);
2640             fjz1             = _mm256_add_pd(fjz1,tz);
2641
2642             }
2643
2644             /**************************
2645              * CALCULATE INTERACTIONS *
2646              **************************/
2647
2648             if (gmx_mm256_any_lt(rsq02,rcutoff2))
2649             {
2650
2651             r02              = _mm256_mul_pd(rsq02,rinv02);
2652             r02              = _mm256_andnot_pd(dummy_mask,r02);
2653
2654             /* EWALD ELECTROSTATICS */
2655
2656             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
2657             ewrt             = _mm256_mul_pd(r02,ewtabscale);
2658             ewitab           = _mm256_cvttpd_epi32(ewrt);
2659             eweps            = _mm256_sub_pd(ewrt,_mm256_round_pd(ewrt, _MM_FROUND_FLOOR));
2660             ewitab           = _mm_slli_epi32(ewitab,2);
2661             ewtabF           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
2662             ewtabD           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,1) );
2663             ewtabV           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,2) );
2664             ewtabFn          = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,3) );
2665             GMX_MM256_FULLTRANSPOSE4_PD(ewtabF,ewtabD,ewtabV,ewtabFn);
2666             felec            = _mm256_add_pd(ewtabF,_mm256_mul_pd(eweps,ewtabD));
2667             velec            = _mm256_sub_pd(ewtabV,_mm256_mul_pd(_mm256_mul_pd(ewtabhalfspace,eweps),_mm256_add_pd(ewtabF,felec)));
2668             velec            = _mm256_mul_pd(qq02,_mm256_sub_pd(rinv02,velec));
2669             felec            = _mm256_mul_pd(_mm256_mul_pd(qq02,rinv02),_mm256_sub_pd(rinvsq02,felec));
2670
2671             d                = _mm256_sub_pd(r02,rswitch);
2672             d                = _mm256_max_pd(d,_mm256_setzero_pd());
2673             d2               = _mm256_mul_pd(d,d);
2674             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)))))));
2675
2676             dsw              = _mm256_mul_pd(d2,_mm256_add_pd(swF2,_mm256_mul_pd(d,_mm256_add_pd(swF3,_mm256_mul_pd(d,swF4)))));
2677
2678             /* Evaluate switch function */
2679             /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
2680             felec            = _mm256_sub_pd( _mm256_mul_pd(felec,sw) , _mm256_mul_pd(rinv02,_mm256_mul_pd(velec,dsw)) );
2681             cutoff_mask      = _mm256_cmp_pd(rsq02,rcutoff2,_CMP_LT_OQ);
2682
2683             fscal            = felec;
2684
2685             fscal            = _mm256_and_pd(fscal,cutoff_mask);
2686
2687             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
2688
2689             /* Calculate temporary vectorial force */
2690             tx               = _mm256_mul_pd(fscal,dx02);
2691             ty               = _mm256_mul_pd(fscal,dy02);
2692             tz               = _mm256_mul_pd(fscal,dz02);
2693
2694             /* Update vectorial force */
2695             fix0             = _mm256_add_pd(fix0,tx);
2696             fiy0             = _mm256_add_pd(fiy0,ty);
2697             fiz0             = _mm256_add_pd(fiz0,tz);
2698
2699             fjx2             = _mm256_add_pd(fjx2,tx);
2700             fjy2             = _mm256_add_pd(fjy2,ty);
2701             fjz2             = _mm256_add_pd(fjz2,tz);
2702
2703             }
2704
2705             /**************************
2706              * CALCULATE INTERACTIONS *
2707              **************************/
2708
2709             if (gmx_mm256_any_lt(rsq10,rcutoff2))
2710             {
2711
2712             r10              = _mm256_mul_pd(rsq10,rinv10);
2713             r10              = _mm256_andnot_pd(dummy_mask,r10);
2714
2715             /* EWALD ELECTROSTATICS */
2716
2717             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
2718             ewrt             = _mm256_mul_pd(r10,ewtabscale);
2719             ewitab           = _mm256_cvttpd_epi32(ewrt);
2720             eweps            = _mm256_sub_pd(ewrt,_mm256_round_pd(ewrt, _MM_FROUND_FLOOR));
2721             ewitab           = _mm_slli_epi32(ewitab,2);
2722             ewtabF           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
2723             ewtabD           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,1) );
2724             ewtabV           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,2) );
2725             ewtabFn          = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,3) );
2726             GMX_MM256_FULLTRANSPOSE4_PD(ewtabF,ewtabD,ewtabV,ewtabFn);
2727             felec            = _mm256_add_pd(ewtabF,_mm256_mul_pd(eweps,ewtabD));
2728             velec            = _mm256_sub_pd(ewtabV,_mm256_mul_pd(_mm256_mul_pd(ewtabhalfspace,eweps),_mm256_add_pd(ewtabF,felec)));
2729             velec            = _mm256_mul_pd(qq10,_mm256_sub_pd(rinv10,velec));
2730             felec            = _mm256_mul_pd(_mm256_mul_pd(qq10,rinv10),_mm256_sub_pd(rinvsq10,felec));
2731
2732             d                = _mm256_sub_pd(r10,rswitch);
2733             d                = _mm256_max_pd(d,_mm256_setzero_pd());
2734             d2               = _mm256_mul_pd(d,d);
2735             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)))))));
2736
2737             dsw              = _mm256_mul_pd(d2,_mm256_add_pd(swF2,_mm256_mul_pd(d,_mm256_add_pd(swF3,_mm256_mul_pd(d,swF4)))));
2738
2739             /* Evaluate switch function */
2740             /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
2741             felec            = _mm256_sub_pd( _mm256_mul_pd(felec,sw) , _mm256_mul_pd(rinv10,_mm256_mul_pd(velec,dsw)) );
2742             cutoff_mask      = _mm256_cmp_pd(rsq10,rcutoff2,_CMP_LT_OQ);
2743
2744             fscal            = felec;
2745
2746             fscal            = _mm256_and_pd(fscal,cutoff_mask);
2747
2748             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
2749
2750             /* Calculate temporary vectorial force */
2751             tx               = _mm256_mul_pd(fscal,dx10);
2752             ty               = _mm256_mul_pd(fscal,dy10);
2753             tz               = _mm256_mul_pd(fscal,dz10);
2754
2755             /* Update vectorial force */
2756             fix1             = _mm256_add_pd(fix1,tx);
2757             fiy1             = _mm256_add_pd(fiy1,ty);
2758             fiz1             = _mm256_add_pd(fiz1,tz);
2759
2760             fjx0             = _mm256_add_pd(fjx0,tx);
2761             fjy0             = _mm256_add_pd(fjy0,ty);
2762             fjz0             = _mm256_add_pd(fjz0,tz);
2763
2764             }
2765
2766             /**************************
2767              * CALCULATE INTERACTIONS *
2768              **************************/
2769
2770             if (gmx_mm256_any_lt(rsq11,rcutoff2))
2771             {
2772
2773             r11              = _mm256_mul_pd(rsq11,rinv11);
2774             r11              = _mm256_andnot_pd(dummy_mask,r11);
2775
2776             /* EWALD ELECTROSTATICS */
2777
2778             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
2779             ewrt             = _mm256_mul_pd(r11,ewtabscale);
2780             ewitab           = _mm256_cvttpd_epi32(ewrt);
2781             eweps            = _mm256_sub_pd(ewrt,_mm256_round_pd(ewrt, _MM_FROUND_FLOOR));
2782             ewitab           = _mm_slli_epi32(ewitab,2);
2783             ewtabF           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
2784             ewtabD           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,1) );
2785             ewtabV           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,2) );
2786             ewtabFn          = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,3) );
2787             GMX_MM256_FULLTRANSPOSE4_PD(ewtabF,ewtabD,ewtabV,ewtabFn);
2788             felec            = _mm256_add_pd(ewtabF,_mm256_mul_pd(eweps,ewtabD));
2789             velec            = _mm256_sub_pd(ewtabV,_mm256_mul_pd(_mm256_mul_pd(ewtabhalfspace,eweps),_mm256_add_pd(ewtabF,felec)));
2790             velec            = _mm256_mul_pd(qq11,_mm256_sub_pd(rinv11,velec));
2791             felec            = _mm256_mul_pd(_mm256_mul_pd(qq11,rinv11),_mm256_sub_pd(rinvsq11,felec));
2792
2793             d                = _mm256_sub_pd(r11,rswitch);
2794             d                = _mm256_max_pd(d,_mm256_setzero_pd());
2795             d2               = _mm256_mul_pd(d,d);
2796             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)))))));
2797
2798             dsw              = _mm256_mul_pd(d2,_mm256_add_pd(swF2,_mm256_mul_pd(d,_mm256_add_pd(swF3,_mm256_mul_pd(d,swF4)))));
2799
2800             /* Evaluate switch function */
2801             /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
2802             felec            = _mm256_sub_pd( _mm256_mul_pd(felec,sw) , _mm256_mul_pd(rinv11,_mm256_mul_pd(velec,dsw)) );
2803             cutoff_mask      = _mm256_cmp_pd(rsq11,rcutoff2,_CMP_LT_OQ);
2804
2805             fscal            = felec;
2806
2807             fscal            = _mm256_and_pd(fscal,cutoff_mask);
2808
2809             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
2810
2811             /* Calculate temporary vectorial force */
2812             tx               = _mm256_mul_pd(fscal,dx11);
2813             ty               = _mm256_mul_pd(fscal,dy11);
2814             tz               = _mm256_mul_pd(fscal,dz11);
2815
2816             /* Update vectorial force */
2817             fix1             = _mm256_add_pd(fix1,tx);
2818             fiy1             = _mm256_add_pd(fiy1,ty);
2819             fiz1             = _mm256_add_pd(fiz1,tz);
2820
2821             fjx1             = _mm256_add_pd(fjx1,tx);
2822             fjy1             = _mm256_add_pd(fjy1,ty);
2823             fjz1             = _mm256_add_pd(fjz1,tz);
2824
2825             }
2826
2827             /**************************
2828              * CALCULATE INTERACTIONS *
2829              **************************/
2830
2831             if (gmx_mm256_any_lt(rsq12,rcutoff2))
2832             {
2833
2834             r12              = _mm256_mul_pd(rsq12,rinv12);
2835             r12              = _mm256_andnot_pd(dummy_mask,r12);
2836
2837             /* EWALD ELECTROSTATICS */
2838
2839             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
2840             ewrt             = _mm256_mul_pd(r12,ewtabscale);
2841             ewitab           = _mm256_cvttpd_epi32(ewrt);
2842             eweps            = _mm256_sub_pd(ewrt,_mm256_round_pd(ewrt, _MM_FROUND_FLOOR));
2843             ewitab           = _mm_slli_epi32(ewitab,2);
2844             ewtabF           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
2845             ewtabD           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,1) );
2846             ewtabV           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,2) );
2847             ewtabFn          = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,3) );
2848             GMX_MM256_FULLTRANSPOSE4_PD(ewtabF,ewtabD,ewtabV,ewtabFn);
2849             felec            = _mm256_add_pd(ewtabF,_mm256_mul_pd(eweps,ewtabD));
2850             velec            = _mm256_sub_pd(ewtabV,_mm256_mul_pd(_mm256_mul_pd(ewtabhalfspace,eweps),_mm256_add_pd(ewtabF,felec)));
2851             velec            = _mm256_mul_pd(qq12,_mm256_sub_pd(rinv12,velec));
2852             felec            = _mm256_mul_pd(_mm256_mul_pd(qq12,rinv12),_mm256_sub_pd(rinvsq12,felec));
2853
2854             d                = _mm256_sub_pd(r12,rswitch);
2855             d                = _mm256_max_pd(d,_mm256_setzero_pd());
2856             d2               = _mm256_mul_pd(d,d);
2857             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)))))));
2858
2859             dsw              = _mm256_mul_pd(d2,_mm256_add_pd(swF2,_mm256_mul_pd(d,_mm256_add_pd(swF3,_mm256_mul_pd(d,swF4)))));
2860
2861             /* Evaluate switch function */
2862             /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
2863             felec            = _mm256_sub_pd( _mm256_mul_pd(felec,sw) , _mm256_mul_pd(rinv12,_mm256_mul_pd(velec,dsw)) );
2864             cutoff_mask      = _mm256_cmp_pd(rsq12,rcutoff2,_CMP_LT_OQ);
2865
2866             fscal            = felec;
2867
2868             fscal            = _mm256_and_pd(fscal,cutoff_mask);
2869
2870             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
2871
2872             /* Calculate temporary vectorial force */
2873             tx               = _mm256_mul_pd(fscal,dx12);
2874             ty               = _mm256_mul_pd(fscal,dy12);
2875             tz               = _mm256_mul_pd(fscal,dz12);
2876
2877             /* Update vectorial force */
2878             fix1             = _mm256_add_pd(fix1,tx);
2879             fiy1             = _mm256_add_pd(fiy1,ty);
2880             fiz1             = _mm256_add_pd(fiz1,tz);
2881
2882             fjx2             = _mm256_add_pd(fjx2,tx);
2883             fjy2             = _mm256_add_pd(fjy2,ty);
2884             fjz2             = _mm256_add_pd(fjz2,tz);
2885
2886             }
2887
2888             /**************************
2889              * CALCULATE INTERACTIONS *
2890              **************************/
2891
2892             if (gmx_mm256_any_lt(rsq20,rcutoff2))
2893             {
2894
2895             r20              = _mm256_mul_pd(rsq20,rinv20);
2896             r20              = _mm256_andnot_pd(dummy_mask,r20);
2897
2898             /* EWALD ELECTROSTATICS */
2899
2900             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
2901             ewrt             = _mm256_mul_pd(r20,ewtabscale);
2902             ewitab           = _mm256_cvttpd_epi32(ewrt);
2903             eweps            = _mm256_sub_pd(ewrt,_mm256_round_pd(ewrt, _MM_FROUND_FLOOR));
2904             ewitab           = _mm_slli_epi32(ewitab,2);
2905             ewtabF           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
2906             ewtabD           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,1) );
2907             ewtabV           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,2) );
2908             ewtabFn          = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,3) );
2909             GMX_MM256_FULLTRANSPOSE4_PD(ewtabF,ewtabD,ewtabV,ewtabFn);
2910             felec            = _mm256_add_pd(ewtabF,_mm256_mul_pd(eweps,ewtabD));
2911             velec            = _mm256_sub_pd(ewtabV,_mm256_mul_pd(_mm256_mul_pd(ewtabhalfspace,eweps),_mm256_add_pd(ewtabF,felec)));
2912             velec            = _mm256_mul_pd(qq20,_mm256_sub_pd(rinv20,velec));
2913             felec            = _mm256_mul_pd(_mm256_mul_pd(qq20,rinv20),_mm256_sub_pd(rinvsq20,felec));
2914
2915             d                = _mm256_sub_pd(r20,rswitch);
2916             d                = _mm256_max_pd(d,_mm256_setzero_pd());
2917             d2               = _mm256_mul_pd(d,d);
2918             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)))))));
2919
2920             dsw              = _mm256_mul_pd(d2,_mm256_add_pd(swF2,_mm256_mul_pd(d,_mm256_add_pd(swF3,_mm256_mul_pd(d,swF4)))));
2921
2922             /* Evaluate switch function */
2923             /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
2924             felec            = _mm256_sub_pd( _mm256_mul_pd(felec,sw) , _mm256_mul_pd(rinv20,_mm256_mul_pd(velec,dsw)) );
2925             cutoff_mask      = _mm256_cmp_pd(rsq20,rcutoff2,_CMP_LT_OQ);
2926
2927             fscal            = felec;
2928
2929             fscal            = _mm256_and_pd(fscal,cutoff_mask);
2930
2931             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
2932
2933             /* Calculate temporary vectorial force */
2934             tx               = _mm256_mul_pd(fscal,dx20);
2935             ty               = _mm256_mul_pd(fscal,dy20);
2936             tz               = _mm256_mul_pd(fscal,dz20);
2937
2938             /* Update vectorial force */
2939             fix2             = _mm256_add_pd(fix2,tx);
2940             fiy2             = _mm256_add_pd(fiy2,ty);
2941             fiz2             = _mm256_add_pd(fiz2,tz);
2942
2943             fjx0             = _mm256_add_pd(fjx0,tx);
2944             fjy0             = _mm256_add_pd(fjy0,ty);
2945             fjz0             = _mm256_add_pd(fjz0,tz);
2946
2947             }
2948
2949             /**************************
2950              * CALCULATE INTERACTIONS *
2951              **************************/
2952
2953             if (gmx_mm256_any_lt(rsq21,rcutoff2))
2954             {
2955
2956             r21              = _mm256_mul_pd(rsq21,rinv21);
2957             r21              = _mm256_andnot_pd(dummy_mask,r21);
2958
2959             /* EWALD ELECTROSTATICS */
2960
2961             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
2962             ewrt             = _mm256_mul_pd(r21,ewtabscale);
2963             ewitab           = _mm256_cvttpd_epi32(ewrt);
2964             eweps            = _mm256_sub_pd(ewrt,_mm256_round_pd(ewrt, _MM_FROUND_FLOOR));
2965             ewitab           = _mm_slli_epi32(ewitab,2);
2966             ewtabF           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
2967             ewtabD           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,1) );
2968             ewtabV           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,2) );
2969             ewtabFn          = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,3) );
2970             GMX_MM256_FULLTRANSPOSE4_PD(ewtabF,ewtabD,ewtabV,ewtabFn);
2971             felec            = _mm256_add_pd(ewtabF,_mm256_mul_pd(eweps,ewtabD));
2972             velec            = _mm256_sub_pd(ewtabV,_mm256_mul_pd(_mm256_mul_pd(ewtabhalfspace,eweps),_mm256_add_pd(ewtabF,felec)));
2973             velec            = _mm256_mul_pd(qq21,_mm256_sub_pd(rinv21,velec));
2974             felec            = _mm256_mul_pd(_mm256_mul_pd(qq21,rinv21),_mm256_sub_pd(rinvsq21,felec));
2975
2976             d                = _mm256_sub_pd(r21,rswitch);
2977             d                = _mm256_max_pd(d,_mm256_setzero_pd());
2978             d2               = _mm256_mul_pd(d,d);
2979             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)))))));
2980
2981             dsw              = _mm256_mul_pd(d2,_mm256_add_pd(swF2,_mm256_mul_pd(d,_mm256_add_pd(swF3,_mm256_mul_pd(d,swF4)))));
2982
2983             /* Evaluate switch function */
2984             /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
2985             felec            = _mm256_sub_pd( _mm256_mul_pd(felec,sw) , _mm256_mul_pd(rinv21,_mm256_mul_pd(velec,dsw)) );
2986             cutoff_mask      = _mm256_cmp_pd(rsq21,rcutoff2,_CMP_LT_OQ);
2987
2988             fscal            = felec;
2989
2990             fscal            = _mm256_and_pd(fscal,cutoff_mask);
2991
2992             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
2993
2994             /* Calculate temporary vectorial force */
2995             tx               = _mm256_mul_pd(fscal,dx21);
2996             ty               = _mm256_mul_pd(fscal,dy21);
2997             tz               = _mm256_mul_pd(fscal,dz21);
2998
2999             /* Update vectorial force */
3000             fix2             = _mm256_add_pd(fix2,tx);
3001             fiy2             = _mm256_add_pd(fiy2,ty);
3002             fiz2             = _mm256_add_pd(fiz2,tz);
3003
3004             fjx1             = _mm256_add_pd(fjx1,tx);
3005             fjy1             = _mm256_add_pd(fjy1,ty);
3006             fjz1             = _mm256_add_pd(fjz1,tz);
3007
3008             }
3009
3010             /**************************
3011              * CALCULATE INTERACTIONS *
3012              **************************/
3013
3014             if (gmx_mm256_any_lt(rsq22,rcutoff2))
3015             {
3016
3017             r22              = _mm256_mul_pd(rsq22,rinv22);
3018             r22              = _mm256_andnot_pd(dummy_mask,r22);
3019
3020             /* EWALD ELECTROSTATICS */
3021
3022             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
3023             ewrt             = _mm256_mul_pd(r22,ewtabscale);
3024             ewitab           = _mm256_cvttpd_epi32(ewrt);
3025             eweps            = _mm256_sub_pd(ewrt,_mm256_round_pd(ewrt, _MM_FROUND_FLOOR));
3026             ewitab           = _mm_slli_epi32(ewitab,2);
3027             ewtabF           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
3028             ewtabD           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,1) );
3029             ewtabV           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,2) );
3030             ewtabFn          = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,3) );
3031             GMX_MM256_FULLTRANSPOSE4_PD(ewtabF,ewtabD,ewtabV,ewtabFn);
3032             felec            = _mm256_add_pd(ewtabF,_mm256_mul_pd(eweps,ewtabD));
3033             velec            = _mm256_sub_pd(ewtabV,_mm256_mul_pd(_mm256_mul_pd(ewtabhalfspace,eweps),_mm256_add_pd(ewtabF,felec)));
3034             velec            = _mm256_mul_pd(qq22,_mm256_sub_pd(rinv22,velec));
3035             felec            = _mm256_mul_pd(_mm256_mul_pd(qq22,rinv22),_mm256_sub_pd(rinvsq22,felec));
3036
3037             d                = _mm256_sub_pd(r22,rswitch);
3038             d                = _mm256_max_pd(d,_mm256_setzero_pd());
3039             d2               = _mm256_mul_pd(d,d);
3040             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)))))));
3041
3042             dsw              = _mm256_mul_pd(d2,_mm256_add_pd(swF2,_mm256_mul_pd(d,_mm256_add_pd(swF3,_mm256_mul_pd(d,swF4)))));
3043
3044             /* Evaluate switch function */
3045             /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
3046             felec            = _mm256_sub_pd( _mm256_mul_pd(felec,sw) , _mm256_mul_pd(rinv22,_mm256_mul_pd(velec,dsw)) );
3047             cutoff_mask      = _mm256_cmp_pd(rsq22,rcutoff2,_CMP_LT_OQ);
3048
3049             fscal            = felec;
3050
3051             fscal            = _mm256_and_pd(fscal,cutoff_mask);
3052
3053             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
3054
3055             /* Calculate temporary vectorial force */
3056             tx               = _mm256_mul_pd(fscal,dx22);
3057             ty               = _mm256_mul_pd(fscal,dy22);
3058             tz               = _mm256_mul_pd(fscal,dz22);
3059
3060             /* Update vectorial force */
3061             fix2             = _mm256_add_pd(fix2,tx);
3062             fiy2             = _mm256_add_pd(fiy2,ty);
3063             fiz2             = _mm256_add_pd(fiz2,tz);
3064
3065             fjx2             = _mm256_add_pd(fjx2,tx);
3066             fjy2             = _mm256_add_pd(fjy2,ty);
3067             fjz2             = _mm256_add_pd(fjz2,tz);
3068
3069             }
3070
3071             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
3072             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
3073             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
3074             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
3075
3076             gmx_mm256_decrement_3rvec_4ptr_swizzle_pd(fjptrA,fjptrB,fjptrC,fjptrD,
3077                                                       fjx0,fjy0,fjz0,fjx1,fjy1,fjz1,fjx2,fjy2,fjz2);
3078
3079             /* Inner loop uses 567 flops */
3080         }
3081
3082         /* End of innermost loop */
3083
3084         gmx_mm256_update_iforce_3atom_swizzle_pd(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,
3085                                                  f+i_coord_offset,fshift+i_shift_offset);
3086
3087         /* Increment number of inner iterations */
3088         inneriter                  += j_index_end - j_index_start;
3089
3090         /* Outer loop uses 18 flops */
3091     }
3092
3093     /* Increment number of outer iterations */
3094     outeriter        += nri;
3095
3096     /* Update outer/inner flops */
3097
3098     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_W3W3_F,outeriter*18 + inneriter*567);
3099 }