src/gromacs/gmxlib/nonbonded/nb_kernel_avx_256_double/nb_kernel_ElecCSTab_VdwCSTab_GeomW4W4_avx_256_double.c

   1 /*
   2  * This file is part of the GROMACS molecular simulation package.
   3  *
   4  * Copyright (c) 2012,2013,2014, by the GROMACS development team, led by
   5  * Mark Abraham, David van der Spoel, Berk Hess, and Erik Lindahl,
   6  * and including many others, as listed in the AUTHORS file in the
   7  * top-level source directory and at http://www.gromacs.org.
   8  *
   9  * GROMACS is free software; you can redistribute it and/or
  10  * modify it under the terms of the GNU Lesser General Public License
  11  * as published by the Free Software Foundation; either version 2.1
  12  * of the License, or (at your option) any later version.
  13  *
  14  * GROMACS is distributed in the hope that it will be useful,
  15  * but WITHOUT ANY WARRANTY; without even the implied warranty of
  16  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
  17  * Lesser General Public License for more details.
  18  *
  19  * You should have received a copy of the GNU Lesser General Public
  20  * License along with GROMACS; if not, see
  21  * http://www.gnu.org/licenses, or write to the Free Software Foundation,
  22  * Inc., 51 Franklin Street, Fifth Floor, Boston, MA  02110-1301  USA.
  23  *
  24  * If you want to redistribute modifications to GROMACS, please
  25  * consider that scientific software is very special. Version
  26  * control is crucial - bugs must be traceable. We will be happy to
  27  * consider code for inclusion in the official distribution, but
  28  * derived work must not be called official GROMACS. Details are found
  29  * in the README & COPYING files - if they are missing, get the
  30  * official version at http://www.gromacs.org.
  31  *
  32  * To help us fund GROMACS development, we humbly ask that you cite
  33  * the research papers on the package. Check out http://www.gromacs.org.
  34  */
  35 /*
  36  * Note: this file was generated by the GROMACS avx_256_double kernel generator.
  37  */
  38 #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_ElecCSTab_VdwCSTab_GeomW4W4_VF_avx_256_double
  52  * Electrostatics interaction: CubicSplineTable
  53  * VdW interaction:            CubicSplineTable
  54  * Geometry:                   Water4-Water4
  55  * Calculate force/pot:        PotentialAndForce
  56  */
  57 void
  58 nb_kernel_ElecCSTab_VdwCSTab_GeomW4W4_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     real *           vdwioffsetptr3;
  91     __m256d          ix3,iy3,iz3,fix3,fiy3,fiz3,iq3,isai3;
  92     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
  93     __m256d          jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
  94     int              vdwjidx1A,vdwjidx1B,vdwjidx1C,vdwjidx1D;
  95     __m256d          jx1,jy1,jz1,fjx1,fjy1,fjz1,jq1,isaj1;
  96     int              vdwjidx2A,vdwjidx2B,vdwjidx2C,vdwjidx2D;
  97     __m256d          jx2,jy2,jz2,fjx2,fjy2,fjz2,jq2,isaj2;
  98     int              vdwjidx3A,vdwjidx3B,vdwjidx3C,vdwjidx3D;
  99     __m256d          jx3,jy3,jz3,fjx3,fjy3,fjz3,jq3,isaj3;
 100     __m256d          dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
 101     __m256d          dx11,dy11,dz11,rsq11,rinv11,rinvsq11,r11,qq11,c6_11,c12_11;
 102     __m256d          dx12,dy12,dz12,rsq12,rinv12,rinvsq12,r12,qq12,c6_12,c12_12;
 103     __m256d          dx13,dy13,dz13,rsq13,rinv13,rinvsq13,r13,qq13,c6_13,c12_13;
 104     __m256d          dx21,dy21,dz21,rsq21,rinv21,rinvsq21,r21,qq21,c6_21,c12_21;
 105     __m256d          dx22,dy22,dz22,rsq22,rinv22,rinvsq22,r22,qq22,c6_22,c12_22;
 106     __m256d          dx23,dy23,dz23,rsq23,rinv23,rinvsq23,r23,qq23,c6_23,c12_23;
 107     __m256d          dx31,dy31,dz31,rsq31,rinv31,rinvsq31,r31,qq31,c6_31,c12_31;
 108     __m256d          dx32,dy32,dz32,rsq32,rinv32,rinvsq32,r32,qq32,c6_32,c12_32;
 109     __m256d          dx33,dy33,dz33,rsq33,rinv33,rinvsq33,r33,qq33,c6_33,c12_33;
 110     __m256d          velec,felec,velecsum,facel,crf,krf,krf2;
 111     real             *charge;
 112     int              nvdwtype;
 113     __m256d          rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
 114     int              *vdwtype;
 115     real             *vdwparam;
 116     __m256d          one_sixth   = _mm256_set1_pd(1.0/6.0);
 117     __m256d          one_twelfth = _mm256_set1_pd(1.0/12.0);
 118     __m128i          vfitab;
 119     __m128i          ifour       = _mm_set1_epi32(4);
 120     __m256d          rt,vfeps,vftabscale,Y,F,G,H,Heps,Fp,VV,FF;
 121     real             *vftab;
 122     __m256d          dummy_mask,cutoff_mask;
 123     __m128           tmpmask0,tmpmask1;
 124     __m256d          signbit = _mm256_castsi256_pd( _mm256_set1_epi32(0x80000000) );
 125     __m256d          one     = _mm256_set1_pd(1.0);
 126     __m256d          two     = _mm256_set1_pd(2.0);
 127     x                = xx[0];
 128     f                = ff[0];
 129
 130     nri              = nlist->nri;
 131     iinr             = nlist->iinr;
 132     jindex           = nlist->jindex;
 133     jjnr             = nlist->jjnr;
 134     shiftidx         = nlist->shift;
 135     gid              = nlist->gid;
 136     shiftvec         = fr->shift_vec[0];
 137     fshift           = fr->fshift[0];
 138     facel            = _mm256_set1_pd(fr->epsfac);
 139     charge           = mdatoms->chargeA;
 140     nvdwtype         = fr->ntype;
 141     vdwparam         = fr->nbfp;
 142     vdwtype          = mdatoms->typeA;
 143
 144     vftab            = kernel_data->table_elec_vdw->data;
 145     vftabscale       = _mm256_set1_pd(kernel_data->table_elec_vdw->scale);
 146
 147     /* Setup water-specific parameters */
 148     inr              = nlist->iinr[0];
 149     iq1              = _mm256_mul_pd(facel,_mm256_set1_pd(charge[inr+1]));
 150     iq2              = _mm256_mul_pd(facel,_mm256_set1_pd(charge[inr+2]));
 151     iq3              = _mm256_mul_pd(facel,_mm256_set1_pd(charge[inr+3]));
 152     vdwioffsetptr0   = vdwparam+2*nvdwtype*vdwtype[inr+0];
 153
 154     jq1              = _mm256_set1_pd(charge[inr+1]);
 155     jq2              = _mm256_set1_pd(charge[inr+2]);
 156     jq3              = _mm256_set1_pd(charge[inr+3]);
 157     vdwjidx0A        = 2*vdwtype[inr+0];
 158     c6_00            = _mm256_set1_pd(vdwioffsetptr0[vdwjidx0A]);
 159     c12_00           = _mm256_set1_pd(vdwioffsetptr0[vdwjidx0A+1]);
 160     qq11             = _mm256_mul_pd(iq1,jq1);
 161     qq12             = _mm256_mul_pd(iq1,jq2);
 162     qq13             = _mm256_mul_pd(iq1,jq3);
 163     qq21             = _mm256_mul_pd(iq2,jq1);
 164     qq22             = _mm256_mul_pd(iq2,jq2);
 165     qq23             = _mm256_mul_pd(iq2,jq3);
 166     qq31             = _mm256_mul_pd(iq3,jq1);
 167     qq32             = _mm256_mul_pd(iq3,jq2);
 168     qq33             = _mm256_mul_pd(iq3,jq3);
 169
 170     /* Avoid stupid compiler warnings */
 171     jnrA = jnrB = jnrC = jnrD = 0;
 172     j_coord_offsetA = 0;
 173     j_coord_offsetB = 0;
 174     j_coord_offsetC = 0;
 175     j_coord_offsetD = 0;
 176
 177     outeriter        = 0;
 178     inneriter        = 0;
 179
 180     for(iidx=0;iidx<4*DIM;iidx++)
 181     {
 182         scratch[iidx] = 0.0;
 183     }
 184
 185     /* Start outer loop over neighborlists */
 186     for(iidx=0; iidx<nri; iidx++)
 187     {
 188         /* Load shift vector for this list */
 189         i_shift_offset   = DIM*shiftidx[iidx];
 190
 191         /* Load limits for loop over neighbors */
 192         j_index_start    = jindex[iidx];
 193         j_index_end      = jindex[iidx+1];
 194
 195         /* Get outer coordinate index */
 196         inr              = iinr[iidx];
 197         i_coord_offset   = DIM*inr;
 198
 199         /* Load i particle coords and add shift vector */
 200         gmx_mm256_load_shift_and_4rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset,
 201                                                     &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2,&ix3,&iy3,&iz3);
 202
 203         fix0             = _mm256_setzero_pd();
 204         fiy0             = _mm256_setzero_pd();
 205         fiz0             = _mm256_setzero_pd();
 206         fix1             = _mm256_setzero_pd();
 207         fiy1             = _mm256_setzero_pd();
 208         fiz1             = _mm256_setzero_pd();
 209         fix2             = _mm256_setzero_pd();
 210         fiy2             = _mm256_setzero_pd();
 211         fiz2             = _mm256_setzero_pd();
 212         fix3             = _mm256_setzero_pd();
 213         fiy3             = _mm256_setzero_pd();
 214         fiz3             = _mm256_setzero_pd();
 215
 216         /* Reset potential sums */
 217         velecsum         = _mm256_setzero_pd();
 218         vvdwsum          = _mm256_setzero_pd();
 219
 220         /* Start inner kernel loop */
 221         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
 222         {
 223
 224             /* Get j neighbor index, and coordinate index */
 225             jnrA             = jjnr[jidx];
 226             jnrB             = jjnr[jidx+1];
 227             jnrC             = jjnr[jidx+2];
 228             jnrD             = jjnr[jidx+3];
 229             j_coord_offsetA  = DIM*jnrA;
 230             j_coord_offsetB  = DIM*jnrB;
 231             j_coord_offsetC  = DIM*jnrC;
 232             j_coord_offsetD  = DIM*jnrD;
 233
 234             /* load j atom coordinates */
 235             gmx_mm256_load_4rvec_4ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 236                                                  x+j_coord_offsetC,x+j_coord_offsetD,
 237                                                  &jx0,&jy0,&jz0,&jx1,&jy1,&jz1,&jx2,
 238                                                  &jy2,&jz2,&jx3,&jy3,&jz3);
 239
 240             /* Calculate displacement vector */
 241             dx00             = _mm256_sub_pd(ix0,jx0);
 242             dy00             = _mm256_sub_pd(iy0,jy0);
 243             dz00             = _mm256_sub_pd(iz0,jz0);
 244             dx11             = _mm256_sub_pd(ix1,jx1);
 245             dy11             = _mm256_sub_pd(iy1,jy1);
 246             dz11             = _mm256_sub_pd(iz1,jz1);
 247             dx12             = _mm256_sub_pd(ix1,jx2);
 248             dy12             = _mm256_sub_pd(iy1,jy2);
 249             dz12             = _mm256_sub_pd(iz1,jz2);
 250             dx13             = _mm256_sub_pd(ix1,jx3);
 251             dy13             = _mm256_sub_pd(iy1,jy3);
 252             dz13             = _mm256_sub_pd(iz1,jz3);
 253             dx21             = _mm256_sub_pd(ix2,jx1);
 254             dy21             = _mm256_sub_pd(iy2,jy1);
 255             dz21             = _mm256_sub_pd(iz2,jz1);
 256             dx22             = _mm256_sub_pd(ix2,jx2);
 257             dy22             = _mm256_sub_pd(iy2,jy2);
 258             dz22             = _mm256_sub_pd(iz2,jz2);
 259             dx23             = _mm256_sub_pd(ix2,jx3);
 260             dy23             = _mm256_sub_pd(iy2,jy3);
 261             dz23             = _mm256_sub_pd(iz2,jz3);
 262             dx31             = _mm256_sub_pd(ix3,jx1);
 263             dy31             = _mm256_sub_pd(iy3,jy1);
 264             dz31             = _mm256_sub_pd(iz3,jz1);
 265             dx32             = _mm256_sub_pd(ix3,jx2);
 266             dy32             = _mm256_sub_pd(iy3,jy2);
 267             dz32             = _mm256_sub_pd(iz3,jz2);
 268             dx33             = _mm256_sub_pd(ix3,jx3);
 269             dy33             = _mm256_sub_pd(iy3,jy3);
 270             dz33             = _mm256_sub_pd(iz3,jz3);
 271
 272             /* Calculate squared distance and things based on it */
 273             rsq00            = gmx_mm256_calc_rsq_pd(dx00,dy00,dz00);
 274             rsq11            = gmx_mm256_calc_rsq_pd(dx11,dy11,dz11);
 275             rsq12            = gmx_mm256_calc_rsq_pd(dx12,dy12,dz12);
 276             rsq13            = gmx_mm256_calc_rsq_pd(dx13,dy13,dz13);
 277             rsq21            = gmx_mm256_calc_rsq_pd(dx21,dy21,dz21);
 278             rsq22            = gmx_mm256_calc_rsq_pd(dx22,dy22,dz22);
 279             rsq23            = gmx_mm256_calc_rsq_pd(dx23,dy23,dz23);
 280             rsq31            = gmx_mm256_calc_rsq_pd(dx31,dy31,dz31);
 281             rsq32            = gmx_mm256_calc_rsq_pd(dx32,dy32,dz32);
 282             rsq33            = gmx_mm256_calc_rsq_pd(dx33,dy33,dz33);
 283
 284             rinv00           = gmx_mm256_invsqrt_pd(rsq00);
 285             rinv11           = gmx_mm256_invsqrt_pd(rsq11);
 286             rinv12           = gmx_mm256_invsqrt_pd(rsq12);
 287             rinv13           = gmx_mm256_invsqrt_pd(rsq13);
 288             rinv21           = gmx_mm256_invsqrt_pd(rsq21);
 289             rinv22           = gmx_mm256_invsqrt_pd(rsq22);
 290             rinv23           = gmx_mm256_invsqrt_pd(rsq23);
 291             rinv31           = gmx_mm256_invsqrt_pd(rsq31);
 292             rinv32           = gmx_mm256_invsqrt_pd(rsq32);
 293             rinv33           = gmx_mm256_invsqrt_pd(rsq33);
 294
 295             fjx0             = _mm256_setzero_pd();
 296             fjy0             = _mm256_setzero_pd();
 297             fjz0             = _mm256_setzero_pd();
 298             fjx1             = _mm256_setzero_pd();
 299             fjy1             = _mm256_setzero_pd();
 300             fjz1             = _mm256_setzero_pd();
 301             fjx2             = _mm256_setzero_pd();
 302             fjy2             = _mm256_setzero_pd();
 303             fjz2             = _mm256_setzero_pd();
 304             fjx3             = _mm256_setzero_pd();
 305             fjy3             = _mm256_setzero_pd();
 306             fjz3             = _mm256_setzero_pd();
 307
 308             /**************************
 309              * CALCULATE INTERACTIONS *
 310              **************************/
 311
 312             r00              = _mm256_mul_pd(rsq00,rinv00);
 313
 314             /* Calculate table index by multiplying r with table scale and truncate to integer */
 315             rt               = _mm256_mul_pd(r00,vftabscale);
 316             vfitab           = _mm256_cvttpd_epi32(rt);
 317             vfeps            = _mm256_sub_pd(rt,_mm256_round_pd(rt, _MM_FROUND_FLOOR));
 318             vfitab           = _mm_slli_epi32(_mm_add_epi32(vfitab,_mm_slli_epi32(vfitab,1)),2);
 319
 320             /* CUBIC SPLINE TABLE DISPERSION */
 321             vfitab           = _mm_add_epi32(vfitab,ifour);
 322             Y                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
 323             F                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
 324             G                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,2) );
 325             H                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,3) );
 326             GMX_MM256_FULLTRANSPOSE4_PD(Y,F,G,H);
 327             Heps             = _mm256_mul_pd(vfeps,H);
 328             Fp               = _mm256_add_pd(F,_mm256_mul_pd(vfeps,_mm256_add_pd(G,Heps)));
 329             VV               = _mm256_add_pd(Y,_mm256_mul_pd(vfeps,Fp));
 330             vvdw6            = _mm256_mul_pd(c6_00,VV);
 331             FF               = _mm256_add_pd(Fp,_mm256_mul_pd(vfeps,_mm256_add_pd(G,_mm256_add_pd(Heps,Heps))));
 332             fvdw6            = _mm256_mul_pd(c6_00,FF);
 333
 334             /* CUBIC SPLINE TABLE REPULSION */
 335             vfitab           = _mm_add_epi32(vfitab,ifour);
 336             Y                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
 337             F                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
 338             G                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,2) );
 339             H                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,3) );
 340             GMX_MM256_FULLTRANSPOSE4_PD(Y,F,G,H);
 341             Heps             = _mm256_mul_pd(vfeps,H);
 342             Fp               = _mm256_add_pd(F,_mm256_mul_pd(vfeps,_mm256_add_pd(G,Heps)));
 343             VV               = _mm256_add_pd(Y,_mm256_mul_pd(vfeps,Fp));
 344             vvdw12           = _mm256_mul_pd(c12_00,VV);
 345             FF               = _mm256_add_pd(Fp,_mm256_mul_pd(vfeps,_mm256_add_pd(G,_mm256_add_pd(Heps,Heps))));
 346             fvdw12           = _mm256_mul_pd(c12_00,FF);
 347             vvdw             = _mm256_add_pd(vvdw12,vvdw6);
 348             fvdw             = _mm256_xor_pd(signbit,_mm256_mul_pd(_mm256_add_pd(fvdw6,fvdw12),_mm256_mul_pd(vftabscale,rinv00)));
 349
 350             /* Update potential sum for this i atom from the interaction with this j atom. */
 351             vvdwsum          = _mm256_add_pd(vvdwsum,vvdw);
 352
 353             fscal            = fvdw;
 354
 355             /* Calculate temporary vectorial force */
 356             tx               = _mm256_mul_pd(fscal,dx00);
 357             ty               = _mm256_mul_pd(fscal,dy00);
 358             tz               = _mm256_mul_pd(fscal,dz00);
 359
 360             /* Update vectorial force */
 361             fix0             = _mm256_add_pd(fix0,tx);
 362             fiy0             = _mm256_add_pd(fiy0,ty);
 363             fiz0             = _mm256_add_pd(fiz0,tz);
 364
 365             fjx0             = _mm256_add_pd(fjx0,tx);
 366             fjy0             = _mm256_add_pd(fjy0,ty);
 367             fjz0             = _mm256_add_pd(fjz0,tz);
 368
 369             /**************************
 370              * CALCULATE INTERACTIONS *
 371              **************************/
 372
 373             r11              = _mm256_mul_pd(rsq11,rinv11);
 374
 375             /* Calculate table index by multiplying r with table scale and truncate to integer */
 376             rt               = _mm256_mul_pd(r11,vftabscale);
 377             vfitab           = _mm256_cvttpd_epi32(rt);
 378             vfeps            = _mm256_sub_pd(rt,_mm256_round_pd(rt, _MM_FROUND_FLOOR));
 379             vfitab           = _mm_slli_epi32(_mm_add_epi32(vfitab,_mm_slli_epi32(vfitab,1)),2);
 380
 381             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 382             Y                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
 383             F                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
 384             G                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,2) );
 385             H                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,3) );
 386             GMX_MM256_FULLTRANSPOSE4_PD(Y,F,G,H);
 387             Heps             = _mm256_mul_pd(vfeps,H);
 388             Fp               = _mm256_add_pd(F,_mm256_mul_pd(vfeps,_mm256_add_pd(G,Heps)));
 389             VV               = _mm256_add_pd(Y,_mm256_mul_pd(vfeps,Fp));
 390             velec            = _mm256_mul_pd(qq11,VV);
 391             FF               = _mm256_add_pd(Fp,_mm256_mul_pd(vfeps,_mm256_add_pd(G,_mm256_add_pd(Heps,Heps))));
 392             felec            = _mm256_xor_pd(signbit,_mm256_mul_pd(_mm256_mul_pd(qq11,FF),_mm256_mul_pd(vftabscale,rinv11)));
 393
 394             /* Update potential sum for this i atom from the interaction with this j atom. */
 395             velecsum         = _mm256_add_pd(velecsum,velec);
 396
 397             fscal            = felec;
 398
 399             /* Calculate temporary vectorial force */
 400             tx               = _mm256_mul_pd(fscal,dx11);
 401             ty               = _mm256_mul_pd(fscal,dy11);
 402             tz               = _mm256_mul_pd(fscal,dz11);
 403
 404             /* Update vectorial force */
 405             fix1             = _mm256_add_pd(fix1,tx);
 406             fiy1             = _mm256_add_pd(fiy1,ty);
 407             fiz1             = _mm256_add_pd(fiz1,tz);
 408
 409             fjx1             = _mm256_add_pd(fjx1,tx);
 410             fjy1             = _mm256_add_pd(fjy1,ty);
 411             fjz1             = _mm256_add_pd(fjz1,tz);
 412
 413             /**************************
 414              * CALCULATE INTERACTIONS *
 415              **************************/
 416
 417             r12              = _mm256_mul_pd(rsq12,rinv12);
 418
 419             /* Calculate table index by multiplying r with table scale and truncate to integer */
 420             rt               = _mm256_mul_pd(r12,vftabscale);
 421             vfitab           = _mm256_cvttpd_epi32(rt);
 422             vfeps            = _mm256_sub_pd(rt,_mm256_round_pd(rt, _MM_FROUND_FLOOR));
 423             vfitab           = _mm_slli_epi32(_mm_add_epi32(vfitab,_mm_slli_epi32(vfitab,1)),2);
 424
 425             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 426             Y                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
 427             F                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
 428             G                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,2) );
 429             H                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,3) );
 430             GMX_MM256_FULLTRANSPOSE4_PD(Y,F,G,H);
 431             Heps             = _mm256_mul_pd(vfeps,H);
 432             Fp               = _mm256_add_pd(F,_mm256_mul_pd(vfeps,_mm256_add_pd(G,Heps)));
 433             VV               = _mm256_add_pd(Y,_mm256_mul_pd(vfeps,Fp));
 434             velec            = _mm256_mul_pd(qq12,VV);
 435             FF               = _mm256_add_pd(Fp,_mm256_mul_pd(vfeps,_mm256_add_pd(G,_mm256_add_pd(Heps,Heps))));
 436             felec            = _mm256_xor_pd(signbit,_mm256_mul_pd(_mm256_mul_pd(qq12,FF),_mm256_mul_pd(vftabscale,rinv12)));
 437
 438             /* Update potential sum for this i atom from the interaction with this j atom. */
 439             velecsum         = _mm256_add_pd(velecsum,velec);
 440
 441             fscal            = felec;
 442
 443             /* Calculate temporary vectorial force */
 444             tx               = _mm256_mul_pd(fscal,dx12);
 445             ty               = _mm256_mul_pd(fscal,dy12);
 446             tz               = _mm256_mul_pd(fscal,dz12);
 447
 448             /* Update vectorial force */
 449             fix1             = _mm256_add_pd(fix1,tx);
 450             fiy1             = _mm256_add_pd(fiy1,ty);
 451             fiz1             = _mm256_add_pd(fiz1,tz);
 452
 453             fjx2             = _mm256_add_pd(fjx2,tx);
 454             fjy2             = _mm256_add_pd(fjy2,ty);
 455             fjz2             = _mm256_add_pd(fjz2,tz);
 456
 457             /**************************
 458              * CALCULATE INTERACTIONS *
 459              **************************/
 460
 461             r13              = _mm256_mul_pd(rsq13,rinv13);
 462
 463             /* Calculate table index by multiplying r with table scale and truncate to integer */
 464             rt               = _mm256_mul_pd(r13,vftabscale);
 465             vfitab           = _mm256_cvttpd_epi32(rt);
 466             vfeps            = _mm256_sub_pd(rt,_mm256_round_pd(rt, _MM_FROUND_FLOOR));
 467             vfitab           = _mm_slli_epi32(_mm_add_epi32(vfitab,_mm_slli_epi32(vfitab,1)),2);
 468
 469             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 470             Y                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
 471             F                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
 472             G                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,2) );
 473             H                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,3) );
 474             GMX_MM256_FULLTRANSPOSE4_PD(Y,F,G,H);
 475             Heps             = _mm256_mul_pd(vfeps,H);
 476             Fp               = _mm256_add_pd(F,_mm256_mul_pd(vfeps,_mm256_add_pd(G,Heps)));
 477             VV               = _mm256_add_pd(Y,_mm256_mul_pd(vfeps,Fp));
 478             velec            = _mm256_mul_pd(qq13,VV);
 479             FF               = _mm256_add_pd(Fp,_mm256_mul_pd(vfeps,_mm256_add_pd(G,_mm256_add_pd(Heps,Heps))));
 480             felec            = _mm256_xor_pd(signbit,_mm256_mul_pd(_mm256_mul_pd(qq13,FF),_mm256_mul_pd(vftabscale,rinv13)));
 481
 482             /* Update potential sum for this i atom from the interaction with this j atom. */
 483             velecsum         = _mm256_add_pd(velecsum,velec);
 484
 485             fscal            = felec;
 486
 487             /* Calculate temporary vectorial force */
 488             tx               = _mm256_mul_pd(fscal,dx13);
 489             ty               = _mm256_mul_pd(fscal,dy13);
 490             tz               = _mm256_mul_pd(fscal,dz13);
 491
 492             /* Update vectorial force */
 493             fix1             = _mm256_add_pd(fix1,tx);
 494             fiy1             = _mm256_add_pd(fiy1,ty);
 495             fiz1             = _mm256_add_pd(fiz1,tz);
 496
 497             fjx3             = _mm256_add_pd(fjx3,tx);
 498             fjy3             = _mm256_add_pd(fjy3,ty);
 499             fjz3             = _mm256_add_pd(fjz3,tz);
 500
 501             /**************************
 502              * CALCULATE INTERACTIONS *
 503              **************************/
 504
 505             r21              = _mm256_mul_pd(rsq21,rinv21);
 506
 507             /* Calculate table index by multiplying r with table scale and truncate to integer */
 508             rt               = _mm256_mul_pd(r21,vftabscale);
 509             vfitab           = _mm256_cvttpd_epi32(rt);
 510             vfeps            = _mm256_sub_pd(rt,_mm256_round_pd(rt, _MM_FROUND_FLOOR));
 511             vfitab           = _mm_slli_epi32(_mm_add_epi32(vfitab,_mm_slli_epi32(vfitab,1)),2);
 512
 513             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 514             Y                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
 515             F                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
 516             G                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,2) );
 517             H                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,3) );
 518             GMX_MM256_FULLTRANSPOSE4_PD(Y,F,G,H);
 519             Heps             = _mm256_mul_pd(vfeps,H);
 520             Fp               = _mm256_add_pd(F,_mm256_mul_pd(vfeps,_mm256_add_pd(G,Heps)));
 521             VV               = _mm256_add_pd(Y,_mm256_mul_pd(vfeps,Fp));
 522             velec            = _mm256_mul_pd(qq21,VV);
 523             FF               = _mm256_add_pd(Fp,_mm256_mul_pd(vfeps,_mm256_add_pd(G,_mm256_add_pd(Heps,Heps))));
 524             felec            = _mm256_xor_pd(signbit,_mm256_mul_pd(_mm256_mul_pd(qq21,FF),_mm256_mul_pd(vftabscale,rinv21)));
 525
 526             /* Update potential sum for this i atom from the interaction with this j atom. */
 527             velecsum         = _mm256_add_pd(velecsum,velec);
 528
 529             fscal            = felec;
 530
 531             /* Calculate temporary vectorial force */
 532             tx               = _mm256_mul_pd(fscal,dx21);
 533             ty               = _mm256_mul_pd(fscal,dy21);
 534             tz               = _mm256_mul_pd(fscal,dz21);
 535
 536             /* Update vectorial force */
 537             fix2             = _mm256_add_pd(fix2,tx);
 538             fiy2             = _mm256_add_pd(fiy2,ty);
 539             fiz2             = _mm256_add_pd(fiz2,tz);
 540
 541             fjx1             = _mm256_add_pd(fjx1,tx);
 542             fjy1             = _mm256_add_pd(fjy1,ty);
 543             fjz1             = _mm256_add_pd(fjz1,tz);
 544
 545             /**************************
 546              * CALCULATE INTERACTIONS *
 547              **************************/
 548
 549             r22              = _mm256_mul_pd(rsq22,rinv22);
 550
 551             /* Calculate table index by multiplying r with table scale and truncate to integer */
 552             rt               = _mm256_mul_pd(r22,vftabscale);
 553             vfitab           = _mm256_cvttpd_epi32(rt);
 554             vfeps            = _mm256_sub_pd(rt,_mm256_round_pd(rt, _MM_FROUND_FLOOR));
 555             vfitab           = _mm_slli_epi32(_mm_add_epi32(vfitab,_mm_slli_epi32(vfitab,1)),2);
 556
 557             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 558             Y                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
 559             F                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
 560             G                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,2) );
 561             H                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,3) );
 562             GMX_MM256_FULLTRANSPOSE4_PD(Y,F,G,H);
 563             Heps             = _mm256_mul_pd(vfeps,H);
 564             Fp               = _mm256_add_pd(F,_mm256_mul_pd(vfeps,_mm256_add_pd(G,Heps)));
 565             VV               = _mm256_add_pd(Y,_mm256_mul_pd(vfeps,Fp));
 566             velec            = _mm256_mul_pd(qq22,VV);
 567             FF               = _mm256_add_pd(Fp,_mm256_mul_pd(vfeps,_mm256_add_pd(G,_mm256_add_pd(Heps,Heps))));
 568             felec            = _mm256_xor_pd(signbit,_mm256_mul_pd(_mm256_mul_pd(qq22,FF),_mm256_mul_pd(vftabscale,rinv22)));
 569
 570             /* Update potential sum for this i atom from the interaction with this j atom. */
 571             velecsum         = _mm256_add_pd(velecsum,velec);
 572
 573             fscal            = felec;
 574
 575             /* Calculate temporary vectorial force */
 576             tx               = _mm256_mul_pd(fscal,dx22);
 577             ty               = _mm256_mul_pd(fscal,dy22);
 578             tz               = _mm256_mul_pd(fscal,dz22);
 579
 580             /* Update vectorial force */
 581             fix2             = _mm256_add_pd(fix2,tx);
 582             fiy2             = _mm256_add_pd(fiy2,ty);
 583             fiz2             = _mm256_add_pd(fiz2,tz);
 584
 585             fjx2             = _mm256_add_pd(fjx2,tx);
 586             fjy2             = _mm256_add_pd(fjy2,ty);
 587             fjz2             = _mm256_add_pd(fjz2,tz);
 588
 589             /**************************
 590              * CALCULATE INTERACTIONS *
 591              **************************/
 592
 593             r23              = _mm256_mul_pd(rsq23,rinv23);
 594
 595             /* Calculate table index by multiplying r with table scale and truncate to integer */
 596             rt               = _mm256_mul_pd(r23,vftabscale);
 597             vfitab           = _mm256_cvttpd_epi32(rt);
 598             vfeps            = _mm256_sub_pd(rt,_mm256_round_pd(rt, _MM_FROUND_FLOOR));
 599             vfitab           = _mm_slli_epi32(_mm_add_epi32(vfitab,_mm_slli_epi32(vfitab,1)),2);
 600
 601             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 602             Y                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
 603             F                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
 604             G                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,2) );
 605             H                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,3) );
 606             GMX_MM256_FULLTRANSPOSE4_PD(Y,F,G,H);
 607             Heps             = _mm256_mul_pd(vfeps,H);
 608             Fp               = _mm256_add_pd(F,_mm256_mul_pd(vfeps,_mm256_add_pd(G,Heps)));
 609             VV               = _mm256_add_pd(Y,_mm256_mul_pd(vfeps,Fp));
 610             velec            = _mm256_mul_pd(qq23,VV);
 611             FF               = _mm256_add_pd(Fp,_mm256_mul_pd(vfeps,_mm256_add_pd(G,_mm256_add_pd(Heps,Heps))));
 612             felec            = _mm256_xor_pd(signbit,_mm256_mul_pd(_mm256_mul_pd(qq23,FF),_mm256_mul_pd(vftabscale,rinv23)));
 613
 614             /* Update potential sum for this i atom from the interaction with this j atom. */
 615             velecsum         = _mm256_add_pd(velecsum,velec);
 616
 617             fscal            = felec;
 618
 619             /* Calculate temporary vectorial force */
 620             tx               = _mm256_mul_pd(fscal,dx23);
 621             ty               = _mm256_mul_pd(fscal,dy23);
 622             tz               = _mm256_mul_pd(fscal,dz23);
 623
 624             /* Update vectorial force */
 625             fix2             = _mm256_add_pd(fix2,tx);
 626             fiy2             = _mm256_add_pd(fiy2,ty);
 627             fiz2             = _mm256_add_pd(fiz2,tz);
 628
 629             fjx3             = _mm256_add_pd(fjx3,tx);
 630             fjy3             = _mm256_add_pd(fjy3,ty);
 631             fjz3             = _mm256_add_pd(fjz3,tz);
 632
 633             /**************************
 634              * CALCULATE INTERACTIONS *
 635              **************************/
 636
 637             r31              = _mm256_mul_pd(rsq31,rinv31);
 638
 639             /* Calculate table index by multiplying r with table scale and truncate to integer */
 640             rt               = _mm256_mul_pd(r31,vftabscale);
 641             vfitab           = _mm256_cvttpd_epi32(rt);
 642             vfeps            = _mm256_sub_pd(rt,_mm256_round_pd(rt, _MM_FROUND_FLOOR));
 643             vfitab           = _mm_slli_epi32(_mm_add_epi32(vfitab,_mm_slli_epi32(vfitab,1)),2);
 644
 645             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 646             Y                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
 647             F                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
 648             G                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,2) );
 649             H                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,3) );
 650             GMX_MM256_FULLTRANSPOSE4_PD(Y,F,G,H);
 651             Heps             = _mm256_mul_pd(vfeps,H);
 652             Fp               = _mm256_add_pd(F,_mm256_mul_pd(vfeps,_mm256_add_pd(G,Heps)));
 653             VV               = _mm256_add_pd(Y,_mm256_mul_pd(vfeps,Fp));
 654             velec            = _mm256_mul_pd(qq31,VV);
 655             FF               = _mm256_add_pd(Fp,_mm256_mul_pd(vfeps,_mm256_add_pd(G,_mm256_add_pd(Heps,Heps))));
 656             felec            = _mm256_xor_pd(signbit,_mm256_mul_pd(_mm256_mul_pd(qq31,FF),_mm256_mul_pd(vftabscale,rinv31)));
 657
 658             /* Update potential sum for this i atom from the interaction with this j atom. */
 659             velecsum         = _mm256_add_pd(velecsum,velec);
 660
 661             fscal            = felec;
 662
 663             /* Calculate temporary vectorial force */
 664             tx               = _mm256_mul_pd(fscal,dx31);
 665             ty               = _mm256_mul_pd(fscal,dy31);
 666             tz               = _mm256_mul_pd(fscal,dz31);
 667
 668             /* Update vectorial force */
 669             fix3             = _mm256_add_pd(fix3,tx);
 670             fiy3             = _mm256_add_pd(fiy3,ty);
 671             fiz3             = _mm256_add_pd(fiz3,tz);
 672
 673             fjx1             = _mm256_add_pd(fjx1,tx);
 674             fjy1             = _mm256_add_pd(fjy1,ty);
 675             fjz1             = _mm256_add_pd(fjz1,tz);
 676
 677             /**************************
 678              * CALCULATE INTERACTIONS *
 679              **************************/
 680
 681             r32              = _mm256_mul_pd(rsq32,rinv32);
 682
 683             /* Calculate table index by multiplying r with table scale and truncate to integer */
 684             rt               = _mm256_mul_pd(r32,vftabscale);
 685             vfitab           = _mm256_cvttpd_epi32(rt);
 686             vfeps            = _mm256_sub_pd(rt,_mm256_round_pd(rt, _MM_FROUND_FLOOR));
 687             vfitab           = _mm_slli_epi32(_mm_add_epi32(vfitab,_mm_slli_epi32(vfitab,1)),2);
 688
 689             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 690             Y                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
 691             F                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
 692             G                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,2) );
 693             H                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,3) );
 694             GMX_MM256_FULLTRANSPOSE4_PD(Y,F,G,H);
 695             Heps             = _mm256_mul_pd(vfeps,H);
 696             Fp               = _mm256_add_pd(F,_mm256_mul_pd(vfeps,_mm256_add_pd(G,Heps)));
 697             VV               = _mm256_add_pd(Y,_mm256_mul_pd(vfeps,Fp));
 698             velec            = _mm256_mul_pd(qq32,VV);
 699             FF               = _mm256_add_pd(Fp,_mm256_mul_pd(vfeps,_mm256_add_pd(G,_mm256_add_pd(Heps,Heps))));
 700             felec            = _mm256_xor_pd(signbit,_mm256_mul_pd(_mm256_mul_pd(qq32,FF),_mm256_mul_pd(vftabscale,rinv32)));
 701
 702             /* Update potential sum for this i atom from the interaction with this j atom. */
 703             velecsum         = _mm256_add_pd(velecsum,velec);
 704
 705             fscal            = felec;
 706
 707             /* Calculate temporary vectorial force */
 708             tx               = _mm256_mul_pd(fscal,dx32);
 709             ty               = _mm256_mul_pd(fscal,dy32);
 710             tz               = _mm256_mul_pd(fscal,dz32);
 711
 712             /* Update vectorial force */
 713             fix3             = _mm256_add_pd(fix3,tx);
 714             fiy3             = _mm256_add_pd(fiy3,ty);
 715             fiz3             = _mm256_add_pd(fiz3,tz);
 716
 717             fjx2             = _mm256_add_pd(fjx2,tx);
 718             fjy2             = _mm256_add_pd(fjy2,ty);
 719             fjz2             = _mm256_add_pd(fjz2,tz);
 720
 721             /**************************
 722              * CALCULATE INTERACTIONS *
 723              **************************/
 724
 725             r33              = _mm256_mul_pd(rsq33,rinv33);
 726
 727             /* Calculate table index by multiplying r with table scale and truncate to integer */
 728             rt               = _mm256_mul_pd(r33,vftabscale);
 729             vfitab           = _mm256_cvttpd_epi32(rt);
 730             vfeps            = _mm256_sub_pd(rt,_mm256_round_pd(rt, _MM_FROUND_FLOOR));
 731             vfitab           = _mm_slli_epi32(_mm_add_epi32(vfitab,_mm_slli_epi32(vfitab,1)),2);
 732
 733             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 734             Y                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
 735             F                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
 736             G                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,2) );
 737             H                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,3) );
 738             GMX_MM256_FULLTRANSPOSE4_PD(Y,F,G,H);
 739             Heps             = _mm256_mul_pd(vfeps,H);
 740             Fp               = _mm256_add_pd(F,_mm256_mul_pd(vfeps,_mm256_add_pd(G,Heps)));
 741             VV               = _mm256_add_pd(Y,_mm256_mul_pd(vfeps,Fp));
 742             velec            = _mm256_mul_pd(qq33,VV);
 743             FF               = _mm256_add_pd(Fp,_mm256_mul_pd(vfeps,_mm256_add_pd(G,_mm256_add_pd(Heps,Heps))));
 744             felec            = _mm256_xor_pd(signbit,_mm256_mul_pd(_mm256_mul_pd(qq33,FF),_mm256_mul_pd(vftabscale,rinv33)));
 745
 746             /* Update potential sum for this i atom from the interaction with this j atom. */
 747             velecsum         = _mm256_add_pd(velecsum,velec);
 748
 749             fscal            = felec;
 750
 751             /* Calculate temporary vectorial force */
 752             tx               = _mm256_mul_pd(fscal,dx33);
 753             ty               = _mm256_mul_pd(fscal,dy33);
 754             tz               = _mm256_mul_pd(fscal,dz33);
 755
 756             /* Update vectorial force */
 757             fix3             = _mm256_add_pd(fix3,tx);
 758             fiy3             = _mm256_add_pd(fiy3,ty);
 759             fiz3             = _mm256_add_pd(fiz3,tz);
 760
 761             fjx3             = _mm256_add_pd(fjx3,tx);
 762             fjy3             = _mm256_add_pd(fjy3,ty);
 763             fjz3             = _mm256_add_pd(fjz3,tz);
 764
 765             fjptrA             = f+j_coord_offsetA;
 766             fjptrB             = f+j_coord_offsetB;
 767             fjptrC             = f+j_coord_offsetC;
 768             fjptrD             = f+j_coord_offsetD;
 769
 770             gmx_mm256_decrement_4rvec_4ptr_swizzle_pd(fjptrA,fjptrB,fjptrC,fjptrD,
 771                                                       fjx0,fjy0,fjz0,fjx1,fjy1,fjz1,
 772                                                       fjx2,fjy2,fjz2,fjx3,fjy3,fjz3);
 773
 774             /* Inner loop uses 446 flops */
 775         }
 776
 777         if(jidx<j_index_end)
 778         {
 779
 780             /* Get j neighbor index, and coordinate index */
 781             jnrlistA         = jjnr[jidx];
 782             jnrlistB         = jjnr[jidx+1];
 783             jnrlistC         = jjnr[jidx+2];
 784             jnrlistD         = jjnr[jidx+3];
 785             /* Sign of each element will be negative for non-real atoms.
 786              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 787              * so use it as val = _mm_andnot_pd(mask,val) to clear dummy entries.
 788              */
 789             tmpmask0 = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
 790
 791             tmpmask1 = _mm_permute_ps(tmpmask0,_GMX_MM_PERMUTE(3,3,2,2));
 792             tmpmask0 = _mm_permute_ps(tmpmask0,_GMX_MM_PERMUTE(1,1,0,0));
 793             dummy_mask = _mm256_castps_pd(gmx_mm256_set_m128(tmpmask1,tmpmask0));
 794
 795             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
 796             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
 797             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
 798             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
 799             j_coord_offsetA  = DIM*jnrA;
 800             j_coord_offsetB  = DIM*jnrB;
 801             j_coord_offsetC  = DIM*jnrC;
 802             j_coord_offsetD  = DIM*jnrD;
 803
 804             /* load j atom coordinates */
 805             gmx_mm256_load_4rvec_4ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 806                                                  x+j_coord_offsetC,x+j_coord_offsetD,
 807                                                  &jx0,&jy0,&jz0,&jx1,&jy1,&jz1,&jx2,
 808                                                  &jy2,&jz2,&jx3,&jy3,&jz3);
 809
 810             /* Calculate displacement vector */
 811             dx00             = _mm256_sub_pd(ix0,jx0);
 812             dy00             = _mm256_sub_pd(iy0,jy0);
 813             dz00             = _mm256_sub_pd(iz0,jz0);
 814             dx11             = _mm256_sub_pd(ix1,jx1);
 815             dy11             = _mm256_sub_pd(iy1,jy1);
 816             dz11             = _mm256_sub_pd(iz1,jz1);
 817             dx12             = _mm256_sub_pd(ix1,jx2);
 818             dy12             = _mm256_sub_pd(iy1,jy2);
 819             dz12             = _mm256_sub_pd(iz1,jz2);
 820             dx13             = _mm256_sub_pd(ix1,jx3);
 821             dy13             = _mm256_sub_pd(iy1,jy3);
 822             dz13             = _mm256_sub_pd(iz1,jz3);
 823             dx21             = _mm256_sub_pd(ix2,jx1);
 824             dy21             = _mm256_sub_pd(iy2,jy1);
 825             dz21             = _mm256_sub_pd(iz2,jz1);
 826             dx22             = _mm256_sub_pd(ix2,jx2);
 827             dy22             = _mm256_sub_pd(iy2,jy2);
 828             dz22             = _mm256_sub_pd(iz2,jz2);
 829             dx23             = _mm256_sub_pd(ix2,jx3);
 830             dy23             = _mm256_sub_pd(iy2,jy3);
 831             dz23             = _mm256_sub_pd(iz2,jz3);
 832             dx31             = _mm256_sub_pd(ix3,jx1);
 833             dy31             = _mm256_sub_pd(iy3,jy1);
 834             dz31             = _mm256_sub_pd(iz3,jz1);
 835             dx32             = _mm256_sub_pd(ix3,jx2);
 836             dy32             = _mm256_sub_pd(iy3,jy2);
 837             dz32             = _mm256_sub_pd(iz3,jz2);
 838             dx33             = _mm256_sub_pd(ix3,jx3);
 839             dy33             = _mm256_sub_pd(iy3,jy3);
 840             dz33             = _mm256_sub_pd(iz3,jz3);
 841
 842             /* Calculate squared distance and things based on it */
 843             rsq00            = gmx_mm256_calc_rsq_pd(dx00,dy00,dz00);
 844             rsq11            = gmx_mm256_calc_rsq_pd(dx11,dy11,dz11);
 845             rsq12            = gmx_mm256_calc_rsq_pd(dx12,dy12,dz12);
 846             rsq13            = gmx_mm256_calc_rsq_pd(dx13,dy13,dz13);
 847             rsq21            = gmx_mm256_calc_rsq_pd(dx21,dy21,dz21);
 848             rsq22            = gmx_mm256_calc_rsq_pd(dx22,dy22,dz22);
 849             rsq23            = gmx_mm256_calc_rsq_pd(dx23,dy23,dz23);
 850             rsq31            = gmx_mm256_calc_rsq_pd(dx31,dy31,dz31);
 851             rsq32            = gmx_mm256_calc_rsq_pd(dx32,dy32,dz32);
 852             rsq33            = gmx_mm256_calc_rsq_pd(dx33,dy33,dz33);
 853
 854             rinv00           = gmx_mm256_invsqrt_pd(rsq00);
 855             rinv11           = gmx_mm256_invsqrt_pd(rsq11);
 856             rinv12           = gmx_mm256_invsqrt_pd(rsq12);
 857             rinv13           = gmx_mm256_invsqrt_pd(rsq13);
 858             rinv21           = gmx_mm256_invsqrt_pd(rsq21);
 859             rinv22           = gmx_mm256_invsqrt_pd(rsq22);
 860             rinv23           = gmx_mm256_invsqrt_pd(rsq23);
 861             rinv31           = gmx_mm256_invsqrt_pd(rsq31);
 862             rinv32           = gmx_mm256_invsqrt_pd(rsq32);
 863             rinv33           = gmx_mm256_invsqrt_pd(rsq33);
 864
 865             fjx0             = _mm256_setzero_pd();
 866             fjy0             = _mm256_setzero_pd();
 867             fjz0             = _mm256_setzero_pd();
 868             fjx1             = _mm256_setzero_pd();
 869             fjy1             = _mm256_setzero_pd();
 870             fjz1             = _mm256_setzero_pd();
 871             fjx2             = _mm256_setzero_pd();
 872             fjy2             = _mm256_setzero_pd();
 873             fjz2             = _mm256_setzero_pd();
 874             fjx3             = _mm256_setzero_pd();
 875             fjy3             = _mm256_setzero_pd();
 876             fjz3             = _mm256_setzero_pd();
 877
 878             /**************************
 879              * CALCULATE INTERACTIONS *
 880              **************************/
 881
 882             r00              = _mm256_mul_pd(rsq00,rinv00);
 883             r00              = _mm256_andnot_pd(dummy_mask,r00);
 884
 885             /* Calculate table index by multiplying r with table scale and truncate to integer */
 886             rt               = _mm256_mul_pd(r00,vftabscale);
 887             vfitab           = _mm256_cvttpd_epi32(rt);
 888             vfeps            = _mm256_sub_pd(rt,_mm256_round_pd(rt, _MM_FROUND_FLOOR));
 889             vfitab           = _mm_slli_epi32(_mm_add_epi32(vfitab,_mm_slli_epi32(vfitab,1)),2);
 890
 891             /* CUBIC SPLINE TABLE DISPERSION */
 892             vfitab           = _mm_add_epi32(vfitab,ifour);
 893             Y                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
 894             F                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
 895             G                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,2) );
 896             H                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,3) );
 897             GMX_MM256_FULLTRANSPOSE4_PD(Y,F,G,H);
 898             Heps             = _mm256_mul_pd(vfeps,H);
 899             Fp               = _mm256_add_pd(F,_mm256_mul_pd(vfeps,_mm256_add_pd(G,Heps)));
 900             VV               = _mm256_add_pd(Y,_mm256_mul_pd(vfeps,Fp));
 901             vvdw6            = _mm256_mul_pd(c6_00,VV);
 902             FF               = _mm256_add_pd(Fp,_mm256_mul_pd(vfeps,_mm256_add_pd(G,_mm256_add_pd(Heps,Heps))));
 903             fvdw6            = _mm256_mul_pd(c6_00,FF);
 904
 905             /* CUBIC SPLINE TABLE REPULSION */
 906             vfitab           = _mm_add_epi32(vfitab,ifour);
 907             Y                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
 908             F                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
 909             G                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,2) );
 910             H                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,3) );
 911             GMX_MM256_FULLTRANSPOSE4_PD(Y,F,G,H);
 912             Heps             = _mm256_mul_pd(vfeps,H);
 913             Fp               = _mm256_add_pd(F,_mm256_mul_pd(vfeps,_mm256_add_pd(G,Heps)));
 914             VV               = _mm256_add_pd(Y,_mm256_mul_pd(vfeps,Fp));
 915             vvdw12           = _mm256_mul_pd(c12_00,VV);
 916             FF               = _mm256_add_pd(Fp,_mm256_mul_pd(vfeps,_mm256_add_pd(G,_mm256_add_pd(Heps,Heps))));
 917             fvdw12           = _mm256_mul_pd(c12_00,FF);
 918             vvdw             = _mm256_add_pd(vvdw12,vvdw6);
 919             fvdw             = _mm256_xor_pd(signbit,_mm256_mul_pd(_mm256_add_pd(fvdw6,fvdw12),_mm256_mul_pd(vftabscale,rinv00)));
 920
 921             /* Update potential sum for this i atom from the interaction with this j atom. */
 922             vvdw             = _mm256_andnot_pd(dummy_mask,vvdw);
 923             vvdwsum          = _mm256_add_pd(vvdwsum,vvdw);
 924
 925             fscal            = fvdw;
 926
 927             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
 928
 929             /* Calculate temporary vectorial force */
 930             tx               = _mm256_mul_pd(fscal,dx00);
 931             ty               = _mm256_mul_pd(fscal,dy00);
 932             tz               = _mm256_mul_pd(fscal,dz00);
 933
 934             /* Update vectorial force */
 935             fix0             = _mm256_add_pd(fix0,tx);
 936             fiy0             = _mm256_add_pd(fiy0,ty);
 937             fiz0             = _mm256_add_pd(fiz0,tz);
 938
 939             fjx0             = _mm256_add_pd(fjx0,tx);
 940             fjy0             = _mm256_add_pd(fjy0,ty);
 941             fjz0             = _mm256_add_pd(fjz0,tz);
 942
 943             /**************************
 944              * CALCULATE INTERACTIONS *
 945              **************************/
 946
 947             r11              = _mm256_mul_pd(rsq11,rinv11);
 948             r11              = _mm256_andnot_pd(dummy_mask,r11);
 949
 950             /* Calculate table index by multiplying r with table scale and truncate to integer */
 951             rt               = _mm256_mul_pd(r11,vftabscale);
 952             vfitab           = _mm256_cvttpd_epi32(rt);
 953             vfeps            = _mm256_sub_pd(rt,_mm256_round_pd(rt, _MM_FROUND_FLOOR));
 954             vfitab           = _mm_slli_epi32(_mm_add_epi32(vfitab,_mm_slli_epi32(vfitab,1)),2);
 955
 956             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 957             Y                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
 958             F                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
 959             G                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,2) );
 960             H                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,3) );
 961             GMX_MM256_FULLTRANSPOSE4_PD(Y,F,G,H);
 962             Heps             = _mm256_mul_pd(vfeps,H);
 963             Fp               = _mm256_add_pd(F,_mm256_mul_pd(vfeps,_mm256_add_pd(G,Heps)));
 964             VV               = _mm256_add_pd(Y,_mm256_mul_pd(vfeps,Fp));
 965             velec            = _mm256_mul_pd(qq11,VV);
 966             FF               = _mm256_add_pd(Fp,_mm256_mul_pd(vfeps,_mm256_add_pd(G,_mm256_add_pd(Heps,Heps))));
 967             felec            = _mm256_xor_pd(signbit,_mm256_mul_pd(_mm256_mul_pd(qq11,FF),_mm256_mul_pd(vftabscale,rinv11)));
 968
 969             /* Update potential sum for this i atom from the interaction with this j atom. */
 970             velec            = _mm256_andnot_pd(dummy_mask,velec);
 971             velecsum         = _mm256_add_pd(velecsum,velec);
 972
 973             fscal            = felec;
 974
 975             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
 976
 977             /* Calculate temporary vectorial force */
 978             tx               = _mm256_mul_pd(fscal,dx11);
 979             ty               = _mm256_mul_pd(fscal,dy11);
 980             tz               = _mm256_mul_pd(fscal,dz11);
 981
 982             /* Update vectorial force */
 983             fix1             = _mm256_add_pd(fix1,tx);
 984             fiy1             = _mm256_add_pd(fiy1,ty);
 985             fiz1             = _mm256_add_pd(fiz1,tz);
 986
 987             fjx1             = _mm256_add_pd(fjx1,tx);
 988             fjy1             = _mm256_add_pd(fjy1,ty);
 989             fjz1             = _mm256_add_pd(fjz1,tz);
 990
 991             /**************************
 992              * CALCULATE INTERACTIONS *
 993              **************************/
 994
 995             r12              = _mm256_mul_pd(rsq12,rinv12);
 996             r12              = _mm256_andnot_pd(dummy_mask,r12);
 997
 998             /* Calculate table index by multiplying r with table scale and truncate to integer */
 999             rt               = _mm256_mul_pd(r12,vftabscale);
1000             vfitab           = _mm256_cvttpd_epi32(rt);
1001             vfeps            = _mm256_sub_pd(rt,_mm256_round_pd(rt, _MM_FROUND_FLOOR));
1002             vfitab           = _mm_slli_epi32(_mm_add_epi32(vfitab,_mm_slli_epi32(vfitab,1)),2);
1003
1004             /* CUBIC SPLINE TABLE ELECTROSTATICS */
1005             Y                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
1006             F                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
1007             G                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,2) );
1008             H                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,3) );
1009             GMX_MM256_FULLTRANSPOSE4_PD(Y,F,G,H);
1010             Heps             = _mm256_mul_pd(vfeps,H);
1011             Fp               = _mm256_add_pd(F,_mm256_mul_pd(vfeps,_mm256_add_pd(G,Heps)));
1012             VV               = _mm256_add_pd(Y,_mm256_mul_pd(vfeps,Fp));
1013             velec            = _mm256_mul_pd(qq12,VV);
1014             FF               = _mm256_add_pd(Fp,_mm256_mul_pd(vfeps,_mm256_add_pd(G,_mm256_add_pd(Heps,Heps))));
1015             felec            = _mm256_xor_pd(signbit,_mm256_mul_pd(_mm256_mul_pd(qq12,FF),_mm256_mul_pd(vftabscale,rinv12)));
1016
1017             /* Update potential sum for this i atom from the interaction with this j atom. */
1018             velec            = _mm256_andnot_pd(dummy_mask,velec);
1019             velecsum         = _mm256_add_pd(velecsum,velec);
1020
1021             fscal            = felec;
1022
1023             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
1024
1025             /* Calculate temporary vectorial force */
1026             tx               = _mm256_mul_pd(fscal,dx12);
1027             ty               = _mm256_mul_pd(fscal,dy12);
1028             tz               = _mm256_mul_pd(fscal,dz12);
1029
1030             /* Update vectorial force */
1031             fix1             = _mm256_add_pd(fix1,tx);
1032             fiy1             = _mm256_add_pd(fiy1,ty);
1033             fiz1             = _mm256_add_pd(fiz1,tz);
1034
1035             fjx2             = _mm256_add_pd(fjx2,tx);
1036             fjy2             = _mm256_add_pd(fjy2,ty);
1037             fjz2             = _mm256_add_pd(fjz2,tz);
1038
1039             /**************************
1040              * CALCULATE INTERACTIONS *
1041              **************************/
1042
1043             r13              = _mm256_mul_pd(rsq13,rinv13);
1044             r13              = _mm256_andnot_pd(dummy_mask,r13);
1045
1046             /* Calculate table index by multiplying r with table scale and truncate to integer */
1047             rt               = _mm256_mul_pd(r13,vftabscale);
1048             vfitab           = _mm256_cvttpd_epi32(rt);
1049             vfeps            = _mm256_sub_pd(rt,_mm256_round_pd(rt, _MM_FROUND_FLOOR));
1050             vfitab           = _mm_slli_epi32(_mm_add_epi32(vfitab,_mm_slli_epi32(vfitab,1)),2);
1051
1052             /* CUBIC SPLINE TABLE ELECTROSTATICS */
1053             Y                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
1054             F                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
1055             G                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,2) );
1056             H                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,3) );
1057             GMX_MM256_FULLTRANSPOSE4_PD(Y,F,G,H);
1058             Heps             = _mm256_mul_pd(vfeps,H);
1059             Fp               = _mm256_add_pd(F,_mm256_mul_pd(vfeps,_mm256_add_pd(G,Heps)));
1060             VV               = _mm256_add_pd(Y,_mm256_mul_pd(vfeps,Fp));
1061             velec            = _mm256_mul_pd(qq13,VV);
1062             FF               = _mm256_add_pd(Fp,_mm256_mul_pd(vfeps,_mm256_add_pd(G,_mm256_add_pd(Heps,Heps))));
1063             felec            = _mm256_xor_pd(signbit,_mm256_mul_pd(_mm256_mul_pd(qq13,FF),_mm256_mul_pd(vftabscale,rinv13)));
1064
1065             /* Update potential sum for this i atom from the interaction with this j atom. */
1066             velec            = _mm256_andnot_pd(dummy_mask,velec);
1067             velecsum         = _mm256_add_pd(velecsum,velec);
1068
1069             fscal            = felec;
1070
1071             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
1072
1073             /* Calculate temporary vectorial force */
1074             tx               = _mm256_mul_pd(fscal,dx13);
1075             ty               = _mm256_mul_pd(fscal,dy13);
1076             tz               = _mm256_mul_pd(fscal,dz13);
1077
1078             /* Update vectorial force */
1079             fix1             = _mm256_add_pd(fix1,tx);
1080             fiy1             = _mm256_add_pd(fiy1,ty);
1081             fiz1             = _mm256_add_pd(fiz1,tz);
1082
1083             fjx3             = _mm256_add_pd(fjx3,tx);
1084             fjy3             = _mm256_add_pd(fjy3,ty);
1085             fjz3             = _mm256_add_pd(fjz3,tz);
1086
1087             /**************************
1088              * CALCULATE INTERACTIONS *
1089              **************************/
1090
1091             r21              = _mm256_mul_pd(rsq21,rinv21);
1092             r21              = _mm256_andnot_pd(dummy_mask,r21);
1093
1094             /* Calculate table index by multiplying r with table scale and truncate to integer */
1095             rt               = _mm256_mul_pd(r21,vftabscale);
1096             vfitab           = _mm256_cvttpd_epi32(rt);
1097             vfeps            = _mm256_sub_pd(rt,_mm256_round_pd(rt, _MM_FROUND_FLOOR));
1098             vfitab           = _mm_slli_epi32(_mm_add_epi32(vfitab,_mm_slli_epi32(vfitab,1)),2);
1099
1100             /* CUBIC SPLINE TABLE ELECTROSTATICS */
1101             Y                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
1102             F                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
1103             G                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,2) );
1104             H                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,3) );
1105             GMX_MM256_FULLTRANSPOSE4_PD(Y,F,G,H);
1106             Heps             = _mm256_mul_pd(vfeps,H);
1107             Fp               = _mm256_add_pd(F,_mm256_mul_pd(vfeps,_mm256_add_pd(G,Heps)));
1108             VV               = _mm256_add_pd(Y,_mm256_mul_pd(vfeps,Fp));
1109             velec            = _mm256_mul_pd(qq21,VV);
1110             FF               = _mm256_add_pd(Fp,_mm256_mul_pd(vfeps,_mm256_add_pd(G,_mm256_add_pd(Heps,Heps))));
1111             felec            = _mm256_xor_pd(signbit,_mm256_mul_pd(_mm256_mul_pd(qq21,FF),_mm256_mul_pd(vftabscale,rinv21)));
1112
1113             /* Update potential sum for this i atom from the interaction with this j atom. */
1114             velec            = _mm256_andnot_pd(dummy_mask,velec);
1115             velecsum         = _mm256_add_pd(velecsum,velec);
1116
1117             fscal            = felec;
1118
1119             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
1120
1121             /* Calculate temporary vectorial force */
1122             tx               = _mm256_mul_pd(fscal,dx21);
1123             ty               = _mm256_mul_pd(fscal,dy21);
1124             tz               = _mm256_mul_pd(fscal,dz21);
1125
1126             /* Update vectorial force */
1127             fix2             = _mm256_add_pd(fix2,tx);
1128             fiy2             = _mm256_add_pd(fiy2,ty);
1129             fiz2             = _mm256_add_pd(fiz2,tz);
1130
1131             fjx1             = _mm256_add_pd(fjx1,tx);
1132             fjy1             = _mm256_add_pd(fjy1,ty);
1133             fjz1             = _mm256_add_pd(fjz1,tz);
1134
1135             /**************************
1136              * CALCULATE INTERACTIONS *
1137              **************************/
1138
1139             r22              = _mm256_mul_pd(rsq22,rinv22);
1140             r22              = _mm256_andnot_pd(dummy_mask,r22);
1141
1142             /* Calculate table index by multiplying r with table scale and truncate to integer */
1143             rt               = _mm256_mul_pd(r22,vftabscale);
1144             vfitab           = _mm256_cvttpd_epi32(rt);
1145             vfeps            = _mm256_sub_pd(rt,_mm256_round_pd(rt, _MM_FROUND_FLOOR));
1146             vfitab           = _mm_slli_epi32(_mm_add_epi32(vfitab,_mm_slli_epi32(vfitab,1)),2);
1147
1148             /* CUBIC SPLINE TABLE ELECTROSTATICS */
1149             Y                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
1150             F                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
1151             G                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,2) );
1152             H                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,3) );
1153             GMX_MM256_FULLTRANSPOSE4_PD(Y,F,G,H);
1154             Heps             = _mm256_mul_pd(vfeps,H);
1155             Fp               = _mm256_add_pd(F,_mm256_mul_pd(vfeps,_mm256_add_pd(G,Heps)));
1156             VV               = _mm256_add_pd(Y,_mm256_mul_pd(vfeps,Fp));
1157             velec            = _mm256_mul_pd(qq22,VV);
1158             FF               = _mm256_add_pd(Fp,_mm256_mul_pd(vfeps,_mm256_add_pd(G,_mm256_add_pd(Heps,Heps))));
1159             felec            = _mm256_xor_pd(signbit,_mm256_mul_pd(_mm256_mul_pd(qq22,FF),_mm256_mul_pd(vftabscale,rinv22)));
1160
1161             /* Update potential sum for this i atom from the interaction with this j atom. */
1162             velec            = _mm256_andnot_pd(dummy_mask,velec);
1163             velecsum         = _mm256_add_pd(velecsum,velec);
1164
1165             fscal            = felec;
1166
1167             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
1168
1169             /* Calculate temporary vectorial force */
1170             tx               = _mm256_mul_pd(fscal,dx22);
1171             ty               = _mm256_mul_pd(fscal,dy22);
1172             tz               = _mm256_mul_pd(fscal,dz22);
1173
1174             /* Update vectorial force */
1175             fix2             = _mm256_add_pd(fix2,tx);
1176             fiy2             = _mm256_add_pd(fiy2,ty);
1177             fiz2             = _mm256_add_pd(fiz2,tz);
1178
1179             fjx2             = _mm256_add_pd(fjx2,tx);
1180             fjy2             = _mm256_add_pd(fjy2,ty);
1181             fjz2             = _mm256_add_pd(fjz2,tz);
1182
1183             /**************************
1184              * CALCULATE INTERACTIONS *
1185              **************************/
1186
1187             r23              = _mm256_mul_pd(rsq23,rinv23);
1188             r23              = _mm256_andnot_pd(dummy_mask,r23);
1189
1190             /* Calculate table index by multiplying r with table scale and truncate to integer */
1191             rt               = _mm256_mul_pd(r23,vftabscale);
1192             vfitab           = _mm256_cvttpd_epi32(rt);
1193             vfeps            = _mm256_sub_pd(rt,_mm256_round_pd(rt, _MM_FROUND_FLOOR));
1194             vfitab           = _mm_slli_epi32(_mm_add_epi32(vfitab,_mm_slli_epi32(vfitab,1)),2);
1195
1196             /* CUBIC SPLINE TABLE ELECTROSTATICS */
1197             Y                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
1198             F                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
1199             G                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,2) );
1200             H                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,3) );
1201             GMX_MM256_FULLTRANSPOSE4_PD(Y,F,G,H);
1202             Heps             = _mm256_mul_pd(vfeps,H);
1203             Fp               = _mm256_add_pd(F,_mm256_mul_pd(vfeps,_mm256_add_pd(G,Heps)));
1204             VV               = _mm256_add_pd(Y,_mm256_mul_pd(vfeps,Fp));
1205             velec            = _mm256_mul_pd(qq23,VV);
1206             FF               = _mm256_add_pd(Fp,_mm256_mul_pd(vfeps,_mm256_add_pd(G,_mm256_add_pd(Heps,Heps))));
1207             felec            = _mm256_xor_pd(signbit,_mm256_mul_pd(_mm256_mul_pd(qq23,FF),_mm256_mul_pd(vftabscale,rinv23)));
1208
1209             /* Update potential sum for this i atom from the interaction with this j atom. */
1210             velec            = _mm256_andnot_pd(dummy_mask,velec);
1211             velecsum         = _mm256_add_pd(velecsum,velec);
1212
1213             fscal            = felec;
1214
1215             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
1216
1217             /* Calculate temporary vectorial force */
1218             tx               = _mm256_mul_pd(fscal,dx23);
1219             ty               = _mm256_mul_pd(fscal,dy23);
1220             tz               = _mm256_mul_pd(fscal,dz23);
1221
1222             /* Update vectorial force */
1223             fix2             = _mm256_add_pd(fix2,tx);
1224             fiy2             = _mm256_add_pd(fiy2,ty);
1225             fiz2             = _mm256_add_pd(fiz2,tz);
1226
1227             fjx3             = _mm256_add_pd(fjx3,tx);
1228             fjy3             = _mm256_add_pd(fjy3,ty);
1229             fjz3             = _mm256_add_pd(fjz3,tz);
1230
1231             /**************************
1232              * CALCULATE INTERACTIONS *
1233              **************************/
1234
1235             r31              = _mm256_mul_pd(rsq31,rinv31);
1236             r31              = _mm256_andnot_pd(dummy_mask,r31);
1237
1238             /* Calculate table index by multiplying r with table scale and truncate to integer */
1239             rt               = _mm256_mul_pd(r31,vftabscale);
1240             vfitab           = _mm256_cvttpd_epi32(rt);
1241             vfeps            = _mm256_sub_pd(rt,_mm256_round_pd(rt, _MM_FROUND_FLOOR));
1242             vfitab           = _mm_slli_epi32(_mm_add_epi32(vfitab,_mm_slli_epi32(vfitab,1)),2);
1243
1244             /* CUBIC SPLINE TABLE ELECTROSTATICS */
1245             Y                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
1246             F                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
1247             G                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,2) );
1248             H                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,3) );
1249             GMX_MM256_FULLTRANSPOSE4_PD(Y,F,G,H);
1250             Heps             = _mm256_mul_pd(vfeps,H);
1251             Fp               = _mm256_add_pd(F,_mm256_mul_pd(vfeps,_mm256_add_pd(G,Heps)));
1252             VV               = _mm256_add_pd(Y,_mm256_mul_pd(vfeps,Fp));
1253             velec            = _mm256_mul_pd(qq31,VV);
1254             FF               = _mm256_add_pd(Fp,_mm256_mul_pd(vfeps,_mm256_add_pd(G,_mm256_add_pd(Heps,Heps))));
1255             felec            = _mm256_xor_pd(signbit,_mm256_mul_pd(_mm256_mul_pd(qq31,FF),_mm256_mul_pd(vftabscale,rinv31)));
1256
1257             /* Update potential sum for this i atom from the interaction with this j atom. */
1258             velec            = _mm256_andnot_pd(dummy_mask,velec);
1259             velecsum         = _mm256_add_pd(velecsum,velec);
1260
1261             fscal            = felec;
1262
1263             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
1264
1265             /* Calculate temporary vectorial force */
1266             tx               = _mm256_mul_pd(fscal,dx31);
1267             ty               = _mm256_mul_pd(fscal,dy31);
1268             tz               = _mm256_mul_pd(fscal,dz31);
1269
1270             /* Update vectorial force */
1271             fix3             = _mm256_add_pd(fix3,tx);
1272             fiy3             = _mm256_add_pd(fiy3,ty);
1273             fiz3             = _mm256_add_pd(fiz3,tz);
1274
1275             fjx1             = _mm256_add_pd(fjx1,tx);
1276             fjy1             = _mm256_add_pd(fjy1,ty);
1277             fjz1             = _mm256_add_pd(fjz1,tz);
1278
1279             /**************************
1280              * CALCULATE INTERACTIONS *
1281              **************************/
1282
1283             r32              = _mm256_mul_pd(rsq32,rinv32);
1284             r32              = _mm256_andnot_pd(dummy_mask,r32);
1285
1286             /* Calculate table index by multiplying r with table scale and truncate to integer */
1287             rt               = _mm256_mul_pd(r32,vftabscale);
1288             vfitab           = _mm256_cvttpd_epi32(rt);
1289             vfeps            = _mm256_sub_pd(rt,_mm256_round_pd(rt, _MM_FROUND_FLOOR));
1290             vfitab           = _mm_slli_epi32(_mm_add_epi32(vfitab,_mm_slli_epi32(vfitab,1)),2);
1291
1292             /* CUBIC SPLINE TABLE ELECTROSTATICS */
1293             Y                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
1294             F                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
1295             G                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,2) );
1296             H                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,3) );
1297             GMX_MM256_FULLTRANSPOSE4_PD(Y,F,G,H);
1298             Heps             = _mm256_mul_pd(vfeps,H);
1299             Fp               = _mm256_add_pd(F,_mm256_mul_pd(vfeps,_mm256_add_pd(G,Heps)));
1300             VV               = _mm256_add_pd(Y,_mm256_mul_pd(vfeps,Fp));
1301             velec            = _mm256_mul_pd(qq32,VV);
1302             FF               = _mm256_add_pd(Fp,_mm256_mul_pd(vfeps,_mm256_add_pd(G,_mm256_add_pd(Heps,Heps))));
1303             felec            = _mm256_xor_pd(signbit,_mm256_mul_pd(_mm256_mul_pd(qq32,FF),_mm256_mul_pd(vftabscale,rinv32)));
1304
1305             /* Update potential sum for this i atom from the interaction with this j atom. */
1306             velec            = _mm256_andnot_pd(dummy_mask,velec);
1307             velecsum         = _mm256_add_pd(velecsum,velec);
1308
1309             fscal            = felec;
1310
1311             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
1312
1313             /* Calculate temporary vectorial force */
1314             tx               = _mm256_mul_pd(fscal,dx32);
1315             ty               = _mm256_mul_pd(fscal,dy32);
1316             tz               = _mm256_mul_pd(fscal,dz32);
1317
1318             /* Update vectorial force */
1319             fix3             = _mm256_add_pd(fix3,tx);
1320             fiy3             = _mm256_add_pd(fiy3,ty);
1321             fiz3             = _mm256_add_pd(fiz3,tz);
1322
1323             fjx2             = _mm256_add_pd(fjx2,tx);
1324             fjy2             = _mm256_add_pd(fjy2,ty);
1325             fjz2             = _mm256_add_pd(fjz2,tz);
1326
1327             /**************************
1328              * CALCULATE INTERACTIONS *
1329              **************************/
1330
1331             r33              = _mm256_mul_pd(rsq33,rinv33);
1332             r33              = _mm256_andnot_pd(dummy_mask,r33);
1333
1334             /* Calculate table index by multiplying r with table scale and truncate to integer */
1335             rt               = _mm256_mul_pd(r33,vftabscale);
1336             vfitab           = _mm256_cvttpd_epi32(rt);
1337             vfeps            = _mm256_sub_pd(rt,_mm256_round_pd(rt, _MM_FROUND_FLOOR));
1338             vfitab           = _mm_slli_epi32(_mm_add_epi32(vfitab,_mm_slli_epi32(vfitab,1)),2);
1339
1340             /* CUBIC SPLINE TABLE ELECTROSTATICS */
1341             Y                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
1342             F                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
1343             G                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,2) );
1344             H                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,3) );
1345             GMX_MM256_FULLTRANSPOSE4_PD(Y,F,G,H);
1346             Heps             = _mm256_mul_pd(vfeps,H);
1347             Fp               = _mm256_add_pd(F,_mm256_mul_pd(vfeps,_mm256_add_pd(G,Heps)));
1348             VV               = _mm256_add_pd(Y,_mm256_mul_pd(vfeps,Fp));
1349             velec            = _mm256_mul_pd(qq33,VV);
1350             FF               = _mm256_add_pd(Fp,_mm256_mul_pd(vfeps,_mm256_add_pd(G,_mm256_add_pd(Heps,Heps))));
1351             felec            = _mm256_xor_pd(signbit,_mm256_mul_pd(_mm256_mul_pd(qq33,FF),_mm256_mul_pd(vftabscale,rinv33)));
1352
1353             /* Update potential sum for this i atom from the interaction with this j atom. */
1354             velec            = _mm256_andnot_pd(dummy_mask,velec);
1355             velecsum         = _mm256_add_pd(velecsum,velec);
1356
1357             fscal            = felec;
1358
1359             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
1360
1361             /* Calculate temporary vectorial force */
1362             tx               = _mm256_mul_pd(fscal,dx33);
1363             ty               = _mm256_mul_pd(fscal,dy33);
1364             tz               = _mm256_mul_pd(fscal,dz33);
1365
1366             /* Update vectorial force */
1367             fix3             = _mm256_add_pd(fix3,tx);
1368             fiy3             = _mm256_add_pd(fiy3,ty);
1369             fiz3             = _mm256_add_pd(fiz3,tz);
1370
1371             fjx3             = _mm256_add_pd(fjx3,tx);
1372             fjy3             = _mm256_add_pd(fjy3,ty);
1373             fjz3             = _mm256_add_pd(fjz3,tz);
1374
1375             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
1376             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
1377             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
1378             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
1379
1380             gmx_mm256_decrement_4rvec_4ptr_swizzle_pd(fjptrA,fjptrB,fjptrC,fjptrD,
1381                                                       fjx0,fjy0,fjz0,fjx1,fjy1,fjz1,
1382                                                       fjx2,fjy2,fjz2,fjx3,fjy3,fjz3);
1383
1384             /* Inner loop uses 456 flops */
1385         }
1386
1387         /* End of innermost loop */
1388
1389         gmx_mm256_update_iforce_4atom_swizzle_pd(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,fix3,fiy3,fiz3,
1390                                                  f+i_coord_offset,fshift+i_shift_offset);
1391
1392         ggid                        = gid[iidx];
1393         /* Update potential energies */
1394         gmx_mm256_update_1pot_pd(velecsum,kernel_data->energygrp_elec+ggid);
1395         gmx_mm256_update_1pot_pd(vvdwsum,kernel_data->energygrp_vdw+ggid);
1396
1397         /* Increment number of inner iterations */
1398         inneriter                  += j_index_end - j_index_start;
1399
1400         /* Outer loop uses 26 flops */
1401     }
1402
1403     /* Increment number of outer iterations */
1404     outeriter        += nri;
1405
1406     /* Update outer/inner flops */
1407
1408     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W4W4_VF,outeriter*26 + inneriter*456);
1409 }
1410 /*
1411  * Gromacs nonbonded kernel:   nb_kernel_ElecCSTab_VdwCSTab_GeomW4W4_F_avx_256_double
1412  * Electrostatics interaction: CubicSplineTable
1413  * VdW interaction:            CubicSplineTable
1414  * Geometry:                   Water4-Water4
1415  * Calculate force/pot:        Force
1416  */
1417 void
1418 nb_kernel_ElecCSTab_VdwCSTab_GeomW4W4_F_avx_256_double
1419                     (t_nblist                    * gmx_restrict       nlist,
1420                      rvec                        * gmx_restrict          xx,
1421                      rvec                        * gmx_restrict          ff,
1422                      t_forcerec                  * gmx_restrict          fr,
1423                      t_mdatoms                   * gmx_restrict     mdatoms,
1424                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
1425                      t_nrnb                      * gmx_restrict        nrnb)
1426 {
1427     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
1428      * just 0 for non-waters.
1429      * Suffixes A,B,C,D refer to j loop unrolling done with AVX, e.g. for the four different
1430      * jnr indices corresponding to data put in the four positions in the SIMD register.
1431      */
1432     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
1433     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
1434     int              jnrA,jnrB,jnrC,jnrD;
1435     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
1436     int              jnrlistE,jnrlistF,jnrlistG,jnrlistH;
1437     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
1438     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
1439     real             rcutoff_scalar;
1440     real             *shiftvec,*fshift,*x,*f;
1441     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD;
1442     real             scratch[4*DIM];
1443     __m256d          tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
1444     real *           vdwioffsetptr0;
1445     __m256d          ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
1446     real *           vdwioffsetptr1;
1447     __m256d          ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
1448     real *           vdwioffsetptr2;
1449     __m256d          ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
1450     real *           vdwioffsetptr3;
1451     __m256d          ix3,iy3,iz3,fix3,fiy3,fiz3,iq3,isai3;
1452     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
1453     __m256d          jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
1454     int              vdwjidx1A,vdwjidx1B,vdwjidx1C,vdwjidx1D;
1455     __m256d          jx1,jy1,jz1,fjx1,fjy1,fjz1,jq1,isaj1;
1456     int              vdwjidx2A,vdwjidx2B,vdwjidx2C,vdwjidx2D;
1457     __m256d          jx2,jy2,jz2,fjx2,fjy2,fjz2,jq2,isaj2;
1458     int              vdwjidx3A,vdwjidx3B,vdwjidx3C,vdwjidx3D;
1459     __m256d          jx3,jy3,jz3,fjx3,fjy3,fjz3,jq3,isaj3;
1460     __m256d          dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
1461     __m256d          dx11,dy11,dz11,rsq11,rinv11,rinvsq11,r11,qq11,c6_11,c12_11;
1462     __m256d          dx12,dy12,dz12,rsq12,rinv12,rinvsq12,r12,qq12,c6_12,c12_12;
1463     __m256d          dx13,dy13,dz13,rsq13,rinv13,rinvsq13,r13,qq13,c6_13,c12_13;
1464     __m256d          dx21,dy21,dz21,rsq21,rinv21,rinvsq21,r21,qq21,c6_21,c12_21;
1465     __m256d          dx22,dy22,dz22,rsq22,rinv22,rinvsq22,r22,qq22,c6_22,c12_22;
1466     __m256d          dx23,dy23,dz23,rsq23,rinv23,rinvsq23,r23,qq23,c6_23,c12_23;
1467     __m256d          dx31,dy31,dz31,rsq31,rinv31,rinvsq31,r31,qq31,c6_31,c12_31;
1468     __m256d          dx32,dy32,dz32,rsq32,rinv32,rinvsq32,r32,qq32,c6_32,c12_32;
1469     __m256d          dx33,dy33,dz33,rsq33,rinv33,rinvsq33,r33,qq33,c6_33,c12_33;
1470     __m256d          velec,felec,velecsum,facel,crf,krf,krf2;
1471     real             *charge;
1472     int              nvdwtype;
1473     __m256d          rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
1474     int              *vdwtype;
1475     real             *vdwparam;
1476     __m256d          one_sixth   = _mm256_set1_pd(1.0/6.0);
1477     __m256d          one_twelfth = _mm256_set1_pd(1.0/12.0);
1478     __m128i          vfitab;
1479     __m128i          ifour       = _mm_set1_epi32(4);
1480     __m256d          rt,vfeps,vftabscale,Y,F,G,H,Heps,Fp,VV,FF;
1481     real             *vftab;
1482     __m256d          dummy_mask,cutoff_mask;
1483     __m128           tmpmask0,tmpmask1;
1484     __m256d          signbit = _mm256_castsi256_pd( _mm256_set1_epi32(0x80000000) );
1485     __m256d          one     = _mm256_set1_pd(1.0);
1486     __m256d          two     = _mm256_set1_pd(2.0);
1487     x                = xx[0];
1488     f                = ff[0];
1489
1490     nri              = nlist->nri;
1491     iinr             = nlist->iinr;
1492     jindex           = nlist->jindex;
1493     jjnr             = nlist->jjnr;
1494     shiftidx         = nlist->shift;
1495     gid              = nlist->gid;
1496     shiftvec         = fr->shift_vec[0];
1497     fshift           = fr->fshift[0];
1498     facel            = _mm256_set1_pd(fr->epsfac);
1499     charge           = mdatoms->chargeA;
1500     nvdwtype         = fr->ntype;
1501     vdwparam         = fr->nbfp;
1502     vdwtype          = mdatoms->typeA;
1503
1504     vftab            = kernel_data->table_elec_vdw->data;
1505     vftabscale       = _mm256_set1_pd(kernel_data->table_elec_vdw->scale);
1506
1507     /* Setup water-specific parameters */
1508     inr              = nlist->iinr[0];
1509     iq1              = _mm256_mul_pd(facel,_mm256_set1_pd(charge[inr+1]));
1510     iq2              = _mm256_mul_pd(facel,_mm256_set1_pd(charge[inr+2]));
1511     iq3              = _mm256_mul_pd(facel,_mm256_set1_pd(charge[inr+3]));
1512     vdwioffsetptr0   = vdwparam+2*nvdwtype*vdwtype[inr+0];
1513
1514     jq1              = _mm256_set1_pd(charge[inr+1]);
1515     jq2              = _mm256_set1_pd(charge[inr+2]);
1516     jq3              = _mm256_set1_pd(charge[inr+3]);
1517     vdwjidx0A        = 2*vdwtype[inr+0];
1518     c6_00            = _mm256_set1_pd(vdwioffsetptr0[vdwjidx0A]);
1519     c12_00           = _mm256_set1_pd(vdwioffsetptr0[vdwjidx0A+1]);
1520     qq11             = _mm256_mul_pd(iq1,jq1);
1521     qq12             = _mm256_mul_pd(iq1,jq2);
1522     qq13             = _mm256_mul_pd(iq1,jq3);
1523     qq21             = _mm256_mul_pd(iq2,jq1);
1524     qq22             = _mm256_mul_pd(iq2,jq2);
1525     qq23             = _mm256_mul_pd(iq2,jq3);
1526     qq31             = _mm256_mul_pd(iq3,jq1);
1527     qq32             = _mm256_mul_pd(iq3,jq2);
1528     qq33             = _mm256_mul_pd(iq3,jq3);
1529
1530     /* Avoid stupid compiler warnings */
1531     jnrA = jnrB = jnrC = jnrD = 0;
1532     j_coord_offsetA = 0;
1533     j_coord_offsetB = 0;
1534     j_coord_offsetC = 0;
1535     j_coord_offsetD = 0;
1536
1537     outeriter        = 0;
1538     inneriter        = 0;
1539
1540     for(iidx=0;iidx<4*DIM;iidx++)
1541     {
1542         scratch[iidx] = 0.0;
1543     }
1544
1545     /* Start outer loop over neighborlists */
1546     for(iidx=0; iidx<nri; iidx++)
1547     {
1548         /* Load shift vector for this list */
1549         i_shift_offset   = DIM*shiftidx[iidx];
1550
1551         /* Load limits for loop over neighbors */
1552         j_index_start    = jindex[iidx];
1553         j_index_end      = jindex[iidx+1];
1554
1555         /* Get outer coordinate index */
1556         inr              = iinr[iidx];
1557         i_coord_offset   = DIM*inr;
1558
1559         /* Load i particle coords and add shift vector */
1560         gmx_mm256_load_shift_and_4rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset,
1561                                                     &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2,&ix3,&iy3,&iz3);
1562
1563         fix0             = _mm256_setzero_pd();
1564         fiy0             = _mm256_setzero_pd();
1565         fiz0             = _mm256_setzero_pd();
1566         fix1             = _mm256_setzero_pd();
1567         fiy1             = _mm256_setzero_pd();
1568         fiz1             = _mm256_setzero_pd();
1569         fix2             = _mm256_setzero_pd();
1570         fiy2             = _mm256_setzero_pd();
1571         fiz2             = _mm256_setzero_pd();
1572         fix3             = _mm256_setzero_pd();
1573         fiy3             = _mm256_setzero_pd();
1574         fiz3             = _mm256_setzero_pd();
1575
1576         /* Start inner kernel loop */
1577         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
1578         {
1579
1580             /* Get j neighbor index, and coordinate index */
1581             jnrA             = jjnr[jidx];
1582             jnrB             = jjnr[jidx+1];
1583             jnrC             = jjnr[jidx+2];
1584             jnrD             = jjnr[jidx+3];
1585             j_coord_offsetA  = DIM*jnrA;
1586             j_coord_offsetB  = DIM*jnrB;
1587             j_coord_offsetC  = DIM*jnrC;
1588             j_coord_offsetD  = DIM*jnrD;
1589
1590             /* load j atom coordinates */
1591             gmx_mm256_load_4rvec_4ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
1592                                                  x+j_coord_offsetC,x+j_coord_offsetD,
1593                                                  &jx0,&jy0,&jz0,&jx1,&jy1,&jz1,&jx2,
1594                                                  &jy2,&jz2,&jx3,&jy3,&jz3);
1595
1596             /* Calculate displacement vector */
1597             dx00             = _mm256_sub_pd(ix0,jx0);
1598             dy00             = _mm256_sub_pd(iy0,jy0);
1599             dz00             = _mm256_sub_pd(iz0,jz0);
1600             dx11             = _mm256_sub_pd(ix1,jx1);
1601             dy11             = _mm256_sub_pd(iy1,jy1);
1602             dz11             = _mm256_sub_pd(iz1,jz1);
1603             dx12             = _mm256_sub_pd(ix1,jx2);
1604             dy12             = _mm256_sub_pd(iy1,jy2);
1605             dz12             = _mm256_sub_pd(iz1,jz2);
1606             dx13             = _mm256_sub_pd(ix1,jx3);
1607             dy13             = _mm256_sub_pd(iy1,jy3);
1608             dz13             = _mm256_sub_pd(iz1,jz3);
1609             dx21             = _mm256_sub_pd(ix2,jx1);
1610             dy21             = _mm256_sub_pd(iy2,jy1);
1611             dz21             = _mm256_sub_pd(iz2,jz1);
1612             dx22             = _mm256_sub_pd(ix2,jx2);
1613             dy22             = _mm256_sub_pd(iy2,jy2);
1614             dz22             = _mm256_sub_pd(iz2,jz2);
1615             dx23             = _mm256_sub_pd(ix2,jx3);
1616             dy23             = _mm256_sub_pd(iy2,jy3);
1617             dz23             = _mm256_sub_pd(iz2,jz3);
1618             dx31             = _mm256_sub_pd(ix3,jx1);
1619             dy31             = _mm256_sub_pd(iy3,jy1);
1620             dz31             = _mm256_sub_pd(iz3,jz1);
1621             dx32             = _mm256_sub_pd(ix3,jx2);
1622             dy32             = _mm256_sub_pd(iy3,jy2);
1623             dz32             = _mm256_sub_pd(iz3,jz2);
1624             dx33             = _mm256_sub_pd(ix3,jx3);
1625             dy33             = _mm256_sub_pd(iy3,jy3);
1626             dz33             = _mm256_sub_pd(iz3,jz3);
1627
1628             /* Calculate squared distance and things based on it */
1629             rsq00            = gmx_mm256_calc_rsq_pd(dx00,dy00,dz00);
1630             rsq11            = gmx_mm256_calc_rsq_pd(dx11,dy11,dz11);
1631             rsq12            = gmx_mm256_calc_rsq_pd(dx12,dy12,dz12);
1632             rsq13            = gmx_mm256_calc_rsq_pd(dx13,dy13,dz13);
1633             rsq21            = gmx_mm256_calc_rsq_pd(dx21,dy21,dz21);
1634             rsq22            = gmx_mm256_calc_rsq_pd(dx22,dy22,dz22);
1635             rsq23            = gmx_mm256_calc_rsq_pd(dx23,dy23,dz23);
1636             rsq31            = gmx_mm256_calc_rsq_pd(dx31,dy31,dz31);
1637             rsq32            = gmx_mm256_calc_rsq_pd(dx32,dy32,dz32);
1638             rsq33            = gmx_mm256_calc_rsq_pd(dx33,dy33,dz33);
1639
1640             rinv00           = gmx_mm256_invsqrt_pd(rsq00);
1641             rinv11           = gmx_mm256_invsqrt_pd(rsq11);
1642             rinv12           = gmx_mm256_invsqrt_pd(rsq12);
1643             rinv13           = gmx_mm256_invsqrt_pd(rsq13);
1644             rinv21           = gmx_mm256_invsqrt_pd(rsq21);
1645             rinv22           = gmx_mm256_invsqrt_pd(rsq22);
1646             rinv23           = gmx_mm256_invsqrt_pd(rsq23);
1647             rinv31           = gmx_mm256_invsqrt_pd(rsq31);
1648             rinv32           = gmx_mm256_invsqrt_pd(rsq32);
1649             rinv33           = gmx_mm256_invsqrt_pd(rsq33);
1650
1651             fjx0             = _mm256_setzero_pd();
1652             fjy0             = _mm256_setzero_pd();
1653             fjz0             = _mm256_setzero_pd();
1654             fjx1             = _mm256_setzero_pd();
1655             fjy1             = _mm256_setzero_pd();
1656             fjz1             = _mm256_setzero_pd();
1657             fjx2             = _mm256_setzero_pd();
1658             fjy2             = _mm256_setzero_pd();
1659             fjz2             = _mm256_setzero_pd();
1660             fjx3             = _mm256_setzero_pd();
1661             fjy3             = _mm256_setzero_pd();
1662             fjz3             = _mm256_setzero_pd();
1663
1664             /**************************
1665              * CALCULATE INTERACTIONS *
1666              **************************/
1667
1668             r00              = _mm256_mul_pd(rsq00,rinv00);
1669
1670             /* Calculate table index by multiplying r with table scale and truncate to integer */
1671             rt               = _mm256_mul_pd(r00,vftabscale);
1672             vfitab           = _mm256_cvttpd_epi32(rt);
1673             vfeps            = _mm256_sub_pd(rt,_mm256_round_pd(rt, _MM_FROUND_FLOOR));
1674             vfitab           = _mm_slli_epi32(_mm_add_epi32(vfitab,_mm_slli_epi32(vfitab,1)),2);
1675
1676             /* CUBIC SPLINE TABLE DISPERSION */
1677             vfitab           = _mm_add_epi32(vfitab,ifour);
1678             Y                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
1679             F                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
1680             G                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,2) );
1681             H                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,3) );
1682             GMX_MM256_FULLTRANSPOSE4_PD(Y,F,G,H);
1683             Heps             = _mm256_mul_pd(vfeps,H);
1684             Fp               = _mm256_add_pd(F,_mm256_mul_pd(vfeps,_mm256_add_pd(G,Heps)));
1685             FF               = _mm256_add_pd(Fp,_mm256_mul_pd(vfeps,_mm256_add_pd(G,_mm256_add_pd(Heps,Heps))));
1686             fvdw6            = _mm256_mul_pd(c6_00,FF);
1687
1688             /* CUBIC SPLINE TABLE REPULSION */
1689             vfitab           = _mm_add_epi32(vfitab,ifour);
1690             Y                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
1691             F                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
1692             G                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,2) );
1693             H                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,3) );
1694             GMX_MM256_FULLTRANSPOSE4_PD(Y,F,G,H);
1695             Heps             = _mm256_mul_pd(vfeps,H);
1696             Fp               = _mm256_add_pd(F,_mm256_mul_pd(vfeps,_mm256_add_pd(G,Heps)));
1697             FF               = _mm256_add_pd(Fp,_mm256_mul_pd(vfeps,_mm256_add_pd(G,_mm256_add_pd(Heps,Heps))));
1698             fvdw12           = _mm256_mul_pd(c12_00,FF);
1699             fvdw             = _mm256_xor_pd(signbit,_mm256_mul_pd(_mm256_add_pd(fvdw6,fvdw12),_mm256_mul_pd(vftabscale,rinv00)));
1700
1701             fscal            = fvdw;
1702
1703             /* Calculate temporary vectorial force */
1704             tx               = _mm256_mul_pd(fscal,dx00);
1705             ty               = _mm256_mul_pd(fscal,dy00);
1706             tz               = _mm256_mul_pd(fscal,dz00);
1707
1708             /* Update vectorial force */
1709             fix0             = _mm256_add_pd(fix0,tx);
1710             fiy0             = _mm256_add_pd(fiy0,ty);
1711             fiz0             = _mm256_add_pd(fiz0,tz);
1712
1713             fjx0             = _mm256_add_pd(fjx0,tx);
1714             fjy0             = _mm256_add_pd(fjy0,ty);
1715             fjz0             = _mm256_add_pd(fjz0,tz);
1716
1717             /**************************
1718              * CALCULATE INTERACTIONS *
1719              **************************/
1720
1721             r11              = _mm256_mul_pd(rsq11,rinv11);
1722
1723             /* Calculate table index by multiplying r with table scale and truncate to integer */
1724             rt               = _mm256_mul_pd(r11,vftabscale);
1725             vfitab           = _mm256_cvttpd_epi32(rt);
1726             vfeps            = _mm256_sub_pd(rt,_mm256_round_pd(rt, _MM_FROUND_FLOOR));
1727             vfitab           = _mm_slli_epi32(_mm_add_epi32(vfitab,_mm_slli_epi32(vfitab,1)),2);
1728
1729             /* CUBIC SPLINE TABLE ELECTROSTATICS */
1730             Y                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
1731             F                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
1732             G                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,2) );
1733             H                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,3) );
1734             GMX_MM256_FULLTRANSPOSE4_PD(Y,F,G,H);
1735             Heps             = _mm256_mul_pd(vfeps,H);
1736             Fp               = _mm256_add_pd(F,_mm256_mul_pd(vfeps,_mm256_add_pd(G,Heps)));
1737             FF               = _mm256_add_pd(Fp,_mm256_mul_pd(vfeps,_mm256_add_pd(G,_mm256_add_pd(Heps,Heps))));
1738             felec            = _mm256_xor_pd(signbit,_mm256_mul_pd(_mm256_mul_pd(qq11,FF),_mm256_mul_pd(vftabscale,rinv11)));
1739
1740             fscal            = felec;
1741
1742             /* Calculate temporary vectorial force */
1743             tx               = _mm256_mul_pd(fscal,dx11);
1744             ty               = _mm256_mul_pd(fscal,dy11);
1745             tz               = _mm256_mul_pd(fscal,dz11);
1746
1747             /* Update vectorial force */
1748             fix1             = _mm256_add_pd(fix1,tx);
1749             fiy1             = _mm256_add_pd(fiy1,ty);
1750             fiz1             = _mm256_add_pd(fiz1,tz);
1751
1752             fjx1             = _mm256_add_pd(fjx1,tx);
1753             fjy1             = _mm256_add_pd(fjy1,ty);
1754             fjz1             = _mm256_add_pd(fjz1,tz);
1755
1756             /**************************
1757              * CALCULATE INTERACTIONS *
1758              **************************/
1759
1760             r12              = _mm256_mul_pd(rsq12,rinv12);
1761
1762             /* Calculate table index by multiplying r with table scale and truncate to integer */
1763             rt               = _mm256_mul_pd(r12,vftabscale);
1764             vfitab           = _mm256_cvttpd_epi32(rt);
1765             vfeps            = _mm256_sub_pd(rt,_mm256_round_pd(rt, _MM_FROUND_FLOOR));
1766             vfitab           = _mm_slli_epi32(_mm_add_epi32(vfitab,_mm_slli_epi32(vfitab,1)),2);
1767
1768             /* CUBIC SPLINE TABLE ELECTROSTATICS */
1769             Y                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
1770             F                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
1771             G                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,2) );
1772             H                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,3) );
1773             GMX_MM256_FULLTRANSPOSE4_PD(Y,F,G,H);
1774             Heps             = _mm256_mul_pd(vfeps,H);
1775             Fp               = _mm256_add_pd(F,_mm256_mul_pd(vfeps,_mm256_add_pd(G,Heps)));
1776             FF               = _mm256_add_pd(Fp,_mm256_mul_pd(vfeps,_mm256_add_pd(G,_mm256_add_pd(Heps,Heps))));
1777             felec            = _mm256_xor_pd(signbit,_mm256_mul_pd(_mm256_mul_pd(qq12,FF),_mm256_mul_pd(vftabscale,rinv12)));
1778
1779             fscal            = felec;
1780
1781             /* Calculate temporary vectorial force */
1782             tx               = _mm256_mul_pd(fscal,dx12);
1783             ty               = _mm256_mul_pd(fscal,dy12);
1784             tz               = _mm256_mul_pd(fscal,dz12);
1785
1786             /* Update vectorial force */
1787             fix1             = _mm256_add_pd(fix1,tx);
1788             fiy1             = _mm256_add_pd(fiy1,ty);
1789             fiz1             = _mm256_add_pd(fiz1,tz);
1790
1791             fjx2             = _mm256_add_pd(fjx2,tx);
1792             fjy2             = _mm256_add_pd(fjy2,ty);
1793             fjz2             = _mm256_add_pd(fjz2,tz);
1794
1795             /**************************
1796              * CALCULATE INTERACTIONS *
1797              **************************/
1798
1799             r13              = _mm256_mul_pd(rsq13,rinv13);
1800
1801             /* Calculate table index by multiplying r with table scale and truncate to integer */
1802             rt               = _mm256_mul_pd(r13,vftabscale);
1803             vfitab           = _mm256_cvttpd_epi32(rt);
1804             vfeps            = _mm256_sub_pd(rt,_mm256_round_pd(rt, _MM_FROUND_FLOOR));
1805             vfitab           = _mm_slli_epi32(_mm_add_epi32(vfitab,_mm_slli_epi32(vfitab,1)),2);
1806
1807             /* CUBIC SPLINE TABLE ELECTROSTATICS */
1808             Y                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
1809             F                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
1810             G                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,2) );
1811             H                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,3) );
1812             GMX_MM256_FULLTRANSPOSE4_PD(Y,F,G,H);
1813             Heps             = _mm256_mul_pd(vfeps,H);
1814             Fp               = _mm256_add_pd(F,_mm256_mul_pd(vfeps,_mm256_add_pd(G,Heps)));
1815             FF               = _mm256_add_pd(Fp,_mm256_mul_pd(vfeps,_mm256_add_pd(G,_mm256_add_pd(Heps,Heps))));
1816             felec            = _mm256_xor_pd(signbit,_mm256_mul_pd(_mm256_mul_pd(qq13,FF),_mm256_mul_pd(vftabscale,rinv13)));
1817
1818             fscal            = felec;
1819
1820             /* Calculate temporary vectorial force */
1821             tx               = _mm256_mul_pd(fscal,dx13);
1822             ty               = _mm256_mul_pd(fscal,dy13);
1823             tz               = _mm256_mul_pd(fscal,dz13);
1824
1825             /* Update vectorial force */
1826             fix1             = _mm256_add_pd(fix1,tx);
1827             fiy1             = _mm256_add_pd(fiy1,ty);
1828             fiz1             = _mm256_add_pd(fiz1,tz);
1829
1830             fjx3             = _mm256_add_pd(fjx3,tx);
1831             fjy3             = _mm256_add_pd(fjy3,ty);
1832             fjz3             = _mm256_add_pd(fjz3,tz);
1833
1834             /**************************
1835              * CALCULATE INTERACTIONS *
1836              **************************/
1837
1838             r21              = _mm256_mul_pd(rsq21,rinv21);
1839
1840             /* Calculate table index by multiplying r with table scale and truncate to integer */
1841             rt               = _mm256_mul_pd(r21,vftabscale);
1842             vfitab           = _mm256_cvttpd_epi32(rt);
1843             vfeps            = _mm256_sub_pd(rt,_mm256_round_pd(rt, _MM_FROUND_FLOOR));
1844             vfitab           = _mm_slli_epi32(_mm_add_epi32(vfitab,_mm_slli_epi32(vfitab,1)),2);
1845
1846             /* CUBIC SPLINE TABLE ELECTROSTATICS */
1847             Y                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
1848             F                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
1849             G                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,2) );
1850             H                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,3) );
1851             GMX_MM256_FULLTRANSPOSE4_PD(Y,F,G,H);
1852             Heps             = _mm256_mul_pd(vfeps,H);
1853             Fp               = _mm256_add_pd(F,_mm256_mul_pd(vfeps,_mm256_add_pd(G,Heps)));
1854             FF               = _mm256_add_pd(Fp,_mm256_mul_pd(vfeps,_mm256_add_pd(G,_mm256_add_pd(Heps,Heps))));
1855             felec            = _mm256_xor_pd(signbit,_mm256_mul_pd(_mm256_mul_pd(qq21,FF),_mm256_mul_pd(vftabscale,rinv21)));
1856
1857             fscal            = felec;
1858
1859             /* Calculate temporary vectorial force */
1860             tx               = _mm256_mul_pd(fscal,dx21);
1861             ty               = _mm256_mul_pd(fscal,dy21);
1862             tz               = _mm256_mul_pd(fscal,dz21);
1863
1864             /* Update vectorial force */
1865             fix2             = _mm256_add_pd(fix2,tx);
1866             fiy2             = _mm256_add_pd(fiy2,ty);
1867             fiz2             = _mm256_add_pd(fiz2,tz);
1868
1869             fjx1             = _mm256_add_pd(fjx1,tx);
1870             fjy1             = _mm256_add_pd(fjy1,ty);
1871             fjz1             = _mm256_add_pd(fjz1,tz);
1872
1873             /**************************
1874              * CALCULATE INTERACTIONS *
1875              **************************/
1876
1877             r22              = _mm256_mul_pd(rsq22,rinv22);
1878
1879             /* Calculate table index by multiplying r with table scale and truncate to integer */
1880             rt               = _mm256_mul_pd(r22,vftabscale);
1881             vfitab           = _mm256_cvttpd_epi32(rt);
1882             vfeps            = _mm256_sub_pd(rt,_mm256_round_pd(rt, _MM_FROUND_FLOOR));
1883             vfitab           = _mm_slli_epi32(_mm_add_epi32(vfitab,_mm_slli_epi32(vfitab,1)),2);
1884
1885             /* CUBIC SPLINE TABLE ELECTROSTATICS */
1886             Y                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
1887             F                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
1888             G                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,2) );
1889             H                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,3) );
1890             GMX_MM256_FULLTRANSPOSE4_PD(Y,F,G,H);
1891             Heps             = _mm256_mul_pd(vfeps,H);
1892             Fp               = _mm256_add_pd(F,_mm256_mul_pd(vfeps,_mm256_add_pd(G,Heps)));
1893             FF               = _mm256_add_pd(Fp,_mm256_mul_pd(vfeps,_mm256_add_pd(G,_mm256_add_pd(Heps,Heps))));
1894             felec            = _mm256_xor_pd(signbit,_mm256_mul_pd(_mm256_mul_pd(qq22,FF),_mm256_mul_pd(vftabscale,rinv22)));
1895
1896             fscal            = felec;
1897
1898             /* Calculate temporary vectorial force */
1899             tx               = _mm256_mul_pd(fscal,dx22);
1900             ty               = _mm256_mul_pd(fscal,dy22);
1901             tz               = _mm256_mul_pd(fscal,dz22);
1902
1903             /* Update vectorial force */
1904             fix2             = _mm256_add_pd(fix2,tx);
1905             fiy2             = _mm256_add_pd(fiy2,ty);
1906             fiz2             = _mm256_add_pd(fiz2,tz);
1907
1908             fjx2             = _mm256_add_pd(fjx2,tx);
1909             fjy2             = _mm256_add_pd(fjy2,ty);
1910             fjz2             = _mm256_add_pd(fjz2,tz);
1911
1912             /**************************
1913              * CALCULATE INTERACTIONS *
1914              **************************/
1915
1916             r23              = _mm256_mul_pd(rsq23,rinv23);
1917
1918             /* Calculate table index by multiplying r with table scale and truncate to integer */
1919             rt               = _mm256_mul_pd(r23,vftabscale);
1920             vfitab           = _mm256_cvttpd_epi32(rt);
1921             vfeps            = _mm256_sub_pd(rt,_mm256_round_pd(rt, _MM_FROUND_FLOOR));
1922             vfitab           = _mm_slli_epi32(_mm_add_epi32(vfitab,_mm_slli_epi32(vfitab,1)),2);
1923
1924             /* CUBIC SPLINE TABLE ELECTROSTATICS */
1925             Y                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
1926             F                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
1927             G                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,2) );
1928             H                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,3) );
1929             GMX_MM256_FULLTRANSPOSE4_PD(Y,F,G,H);
1930             Heps             = _mm256_mul_pd(vfeps,H);
1931             Fp               = _mm256_add_pd(F,_mm256_mul_pd(vfeps,_mm256_add_pd(G,Heps)));
1932             FF               = _mm256_add_pd(Fp,_mm256_mul_pd(vfeps,_mm256_add_pd(G,_mm256_add_pd(Heps,Heps))));
1933             felec            = _mm256_xor_pd(signbit,_mm256_mul_pd(_mm256_mul_pd(qq23,FF),_mm256_mul_pd(vftabscale,rinv23)));
1934
1935             fscal            = felec;
1936
1937             /* Calculate temporary vectorial force */
1938             tx               = _mm256_mul_pd(fscal,dx23);
1939             ty               = _mm256_mul_pd(fscal,dy23);
1940             tz               = _mm256_mul_pd(fscal,dz23);
1941
1942             /* Update vectorial force */
1943             fix2             = _mm256_add_pd(fix2,tx);
1944             fiy2             = _mm256_add_pd(fiy2,ty);
1945             fiz2             = _mm256_add_pd(fiz2,tz);
1946
1947             fjx3             = _mm256_add_pd(fjx3,tx);
1948             fjy3             = _mm256_add_pd(fjy3,ty);
1949             fjz3             = _mm256_add_pd(fjz3,tz);
1950
1951             /**************************
1952              * CALCULATE INTERACTIONS *
1953              **************************/
1954
1955             r31              = _mm256_mul_pd(rsq31,rinv31);
1956
1957             /* Calculate table index by multiplying r with table scale and truncate to integer */
1958             rt               = _mm256_mul_pd(r31,vftabscale);
1959             vfitab           = _mm256_cvttpd_epi32(rt);
1960             vfeps            = _mm256_sub_pd(rt,_mm256_round_pd(rt, _MM_FROUND_FLOOR));
1961             vfitab           = _mm_slli_epi32(_mm_add_epi32(vfitab,_mm_slli_epi32(vfitab,1)),2);
1962
1963             /* CUBIC SPLINE TABLE ELECTROSTATICS */
1964             Y                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
1965             F                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
1966             G                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,2) );
1967             H                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,3) );
1968             GMX_MM256_FULLTRANSPOSE4_PD(Y,F,G,H);
1969             Heps             = _mm256_mul_pd(vfeps,H);
1970             Fp               = _mm256_add_pd(F,_mm256_mul_pd(vfeps,_mm256_add_pd(G,Heps)));
1971             FF               = _mm256_add_pd(Fp,_mm256_mul_pd(vfeps,_mm256_add_pd(G,_mm256_add_pd(Heps,Heps))));
1972             felec            = _mm256_xor_pd(signbit,_mm256_mul_pd(_mm256_mul_pd(qq31,FF),_mm256_mul_pd(vftabscale,rinv31)));
1973
1974             fscal            = felec;
1975
1976             /* Calculate temporary vectorial force */
1977             tx               = _mm256_mul_pd(fscal,dx31);
1978             ty               = _mm256_mul_pd(fscal,dy31);
1979             tz               = _mm256_mul_pd(fscal,dz31);
1980
1981             /* Update vectorial force */
1982             fix3             = _mm256_add_pd(fix3,tx);
1983             fiy3             = _mm256_add_pd(fiy3,ty);
1984             fiz3             = _mm256_add_pd(fiz3,tz);
1985
1986             fjx1             = _mm256_add_pd(fjx1,tx);
1987             fjy1             = _mm256_add_pd(fjy1,ty);
1988             fjz1             = _mm256_add_pd(fjz1,tz);
1989
1990             /**************************
1991              * CALCULATE INTERACTIONS *
1992              **************************/
1993
1994             r32              = _mm256_mul_pd(rsq32,rinv32);
1995
1996             /* Calculate table index by multiplying r with table scale and truncate to integer */
1997             rt               = _mm256_mul_pd(r32,vftabscale);
1998             vfitab           = _mm256_cvttpd_epi32(rt);
1999             vfeps            = _mm256_sub_pd(rt,_mm256_round_pd(rt, _MM_FROUND_FLOOR));
2000             vfitab           = _mm_slli_epi32(_mm_add_epi32(vfitab,_mm_slli_epi32(vfitab,1)),2);
2001
2002             /* CUBIC SPLINE TABLE ELECTROSTATICS */
2003             Y                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
2004             F                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
2005             G                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,2) );
2006             H                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,3) );
2007             GMX_MM256_FULLTRANSPOSE4_PD(Y,F,G,H);
2008             Heps             = _mm256_mul_pd(vfeps,H);
2009             Fp               = _mm256_add_pd(F,_mm256_mul_pd(vfeps,_mm256_add_pd(G,Heps)));
2010             FF               = _mm256_add_pd(Fp,_mm256_mul_pd(vfeps,_mm256_add_pd(G,_mm256_add_pd(Heps,Heps))));
2011             felec            = _mm256_xor_pd(signbit,_mm256_mul_pd(_mm256_mul_pd(qq32,FF),_mm256_mul_pd(vftabscale,rinv32)));
2012
2013             fscal            = felec;
2014
2015             /* Calculate temporary vectorial force */
2016             tx               = _mm256_mul_pd(fscal,dx32);
2017             ty               = _mm256_mul_pd(fscal,dy32);
2018             tz               = _mm256_mul_pd(fscal,dz32);
2019
2020             /* Update vectorial force */
2021             fix3             = _mm256_add_pd(fix3,tx);
2022             fiy3             = _mm256_add_pd(fiy3,ty);
2023             fiz3             = _mm256_add_pd(fiz3,tz);
2024
2025             fjx2             = _mm256_add_pd(fjx2,tx);
2026             fjy2             = _mm256_add_pd(fjy2,ty);
2027             fjz2             = _mm256_add_pd(fjz2,tz);
2028
2029             /**************************
2030              * CALCULATE INTERACTIONS *
2031              **************************/
2032
2033             r33              = _mm256_mul_pd(rsq33,rinv33);
2034
2035             /* Calculate table index by multiplying r with table scale and truncate to integer */
2036             rt               = _mm256_mul_pd(r33,vftabscale);
2037             vfitab           = _mm256_cvttpd_epi32(rt);
2038             vfeps            = _mm256_sub_pd(rt,_mm256_round_pd(rt, _MM_FROUND_FLOOR));
2039             vfitab           = _mm_slli_epi32(_mm_add_epi32(vfitab,_mm_slli_epi32(vfitab,1)),2);
2040
2041             /* CUBIC SPLINE TABLE ELECTROSTATICS */
2042             Y                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
2043             F                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
2044             G                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,2) );
2045             H                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,3) );
2046             GMX_MM256_FULLTRANSPOSE4_PD(Y,F,G,H);
2047             Heps             = _mm256_mul_pd(vfeps,H);
2048             Fp               = _mm256_add_pd(F,_mm256_mul_pd(vfeps,_mm256_add_pd(G,Heps)));
2049             FF               = _mm256_add_pd(Fp,_mm256_mul_pd(vfeps,_mm256_add_pd(G,_mm256_add_pd(Heps,Heps))));
2050             felec            = _mm256_xor_pd(signbit,_mm256_mul_pd(_mm256_mul_pd(qq33,FF),_mm256_mul_pd(vftabscale,rinv33)));
2051
2052             fscal            = felec;
2053
2054             /* Calculate temporary vectorial force */
2055             tx               = _mm256_mul_pd(fscal,dx33);
2056             ty               = _mm256_mul_pd(fscal,dy33);
2057             tz               = _mm256_mul_pd(fscal,dz33);
2058
2059             /* Update vectorial force */
2060             fix3             = _mm256_add_pd(fix3,tx);
2061             fiy3             = _mm256_add_pd(fiy3,ty);
2062             fiz3             = _mm256_add_pd(fiz3,tz);
2063
2064             fjx3             = _mm256_add_pd(fjx3,tx);
2065             fjy3             = _mm256_add_pd(fjy3,ty);
2066             fjz3             = _mm256_add_pd(fjz3,tz);
2067
2068             fjptrA             = f+j_coord_offsetA;
2069             fjptrB             = f+j_coord_offsetB;
2070             fjptrC             = f+j_coord_offsetC;
2071             fjptrD             = f+j_coord_offsetD;
2072
2073             gmx_mm256_decrement_4rvec_4ptr_swizzle_pd(fjptrA,fjptrB,fjptrC,fjptrD,
2074                                                       fjx0,fjy0,fjz0,fjx1,fjy1,fjz1,
2075                                                       fjx2,fjy2,fjz2,fjx3,fjy3,fjz3);
2076
2077             /* Inner loop uses 402 flops */
2078         }
2079
2080         if(jidx<j_index_end)
2081         {
2082
2083             /* Get j neighbor index, and coordinate index */
2084             jnrlistA         = jjnr[jidx];
2085             jnrlistB         = jjnr[jidx+1];
2086             jnrlistC         = jjnr[jidx+2];
2087             jnrlistD         = jjnr[jidx+3];
2088             /* Sign of each element will be negative for non-real atoms.
2089              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
2090              * so use it as val = _mm_andnot_pd(mask,val) to clear dummy entries.
2091              */
2092             tmpmask0 = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
2093
2094             tmpmask1 = _mm_permute_ps(tmpmask0,_GMX_MM_PERMUTE(3,3,2,2));
2095             tmpmask0 = _mm_permute_ps(tmpmask0,_GMX_MM_PERMUTE(1,1,0,0));
2096             dummy_mask = _mm256_castps_pd(gmx_mm256_set_m128(tmpmask1,tmpmask0));
2097
2098             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
2099             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
2100             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
2101             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
2102             j_coord_offsetA  = DIM*jnrA;
2103             j_coord_offsetB  = DIM*jnrB;
2104             j_coord_offsetC  = DIM*jnrC;
2105             j_coord_offsetD  = DIM*jnrD;
2106
2107             /* load j atom coordinates */
2108             gmx_mm256_load_4rvec_4ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
2109                                                  x+j_coord_offsetC,x+j_coord_offsetD,
2110                                                  &jx0,&jy0,&jz0,&jx1,&jy1,&jz1,&jx2,
2111                                                  &jy2,&jz2,&jx3,&jy3,&jz3);
2112
2113             /* Calculate displacement vector */
2114             dx00             = _mm256_sub_pd(ix0,jx0);
2115             dy00             = _mm256_sub_pd(iy0,jy0);
2116             dz00             = _mm256_sub_pd(iz0,jz0);
2117             dx11             = _mm256_sub_pd(ix1,jx1);
2118             dy11             = _mm256_sub_pd(iy1,jy1);
2119             dz11             = _mm256_sub_pd(iz1,jz1);
2120             dx12             = _mm256_sub_pd(ix1,jx2);
2121             dy12             = _mm256_sub_pd(iy1,jy2);
2122             dz12             = _mm256_sub_pd(iz1,jz2);
2123             dx13             = _mm256_sub_pd(ix1,jx3);
2124             dy13             = _mm256_sub_pd(iy1,jy3);
2125             dz13             = _mm256_sub_pd(iz1,jz3);
2126             dx21             = _mm256_sub_pd(ix2,jx1);
2127             dy21             = _mm256_sub_pd(iy2,jy1);
2128             dz21             = _mm256_sub_pd(iz2,jz1);
2129             dx22             = _mm256_sub_pd(ix2,jx2);
2130             dy22             = _mm256_sub_pd(iy2,jy2);
2131             dz22             = _mm256_sub_pd(iz2,jz2);
2132             dx23             = _mm256_sub_pd(ix2,jx3);
2133             dy23             = _mm256_sub_pd(iy2,jy3);
2134             dz23             = _mm256_sub_pd(iz2,jz3);
2135             dx31             = _mm256_sub_pd(ix3,jx1);
2136             dy31             = _mm256_sub_pd(iy3,jy1);
2137             dz31             = _mm256_sub_pd(iz3,jz1);
2138             dx32             = _mm256_sub_pd(ix3,jx2);
2139             dy32             = _mm256_sub_pd(iy3,jy2);
2140             dz32             = _mm256_sub_pd(iz3,jz2);
2141             dx33             = _mm256_sub_pd(ix3,jx3);
2142             dy33             = _mm256_sub_pd(iy3,jy3);
2143             dz33             = _mm256_sub_pd(iz3,jz3);
2144
2145             /* Calculate squared distance and things based on it */
2146             rsq00            = gmx_mm256_calc_rsq_pd(dx00,dy00,dz00);
2147             rsq11            = gmx_mm256_calc_rsq_pd(dx11,dy11,dz11);
2148             rsq12            = gmx_mm256_calc_rsq_pd(dx12,dy12,dz12);
2149             rsq13            = gmx_mm256_calc_rsq_pd(dx13,dy13,dz13);
2150             rsq21            = gmx_mm256_calc_rsq_pd(dx21,dy21,dz21);
2151             rsq22            = gmx_mm256_calc_rsq_pd(dx22,dy22,dz22);
2152             rsq23            = gmx_mm256_calc_rsq_pd(dx23,dy23,dz23);
2153             rsq31            = gmx_mm256_calc_rsq_pd(dx31,dy31,dz31);
2154             rsq32            = gmx_mm256_calc_rsq_pd(dx32,dy32,dz32);
2155             rsq33            = gmx_mm256_calc_rsq_pd(dx33,dy33,dz33);
2156
2157             rinv00           = gmx_mm256_invsqrt_pd(rsq00);
2158             rinv11           = gmx_mm256_invsqrt_pd(rsq11);
2159             rinv12           = gmx_mm256_invsqrt_pd(rsq12);
2160             rinv13           = gmx_mm256_invsqrt_pd(rsq13);
2161             rinv21           = gmx_mm256_invsqrt_pd(rsq21);
2162             rinv22           = gmx_mm256_invsqrt_pd(rsq22);
2163             rinv23           = gmx_mm256_invsqrt_pd(rsq23);
2164             rinv31           = gmx_mm256_invsqrt_pd(rsq31);
2165             rinv32           = gmx_mm256_invsqrt_pd(rsq32);
2166             rinv33           = gmx_mm256_invsqrt_pd(rsq33);
2167
2168             fjx0             = _mm256_setzero_pd();
2169             fjy0             = _mm256_setzero_pd();
2170             fjz0             = _mm256_setzero_pd();
2171             fjx1             = _mm256_setzero_pd();
2172             fjy1             = _mm256_setzero_pd();
2173             fjz1             = _mm256_setzero_pd();
2174             fjx2             = _mm256_setzero_pd();
2175             fjy2             = _mm256_setzero_pd();
2176             fjz2             = _mm256_setzero_pd();
2177             fjx3             = _mm256_setzero_pd();
2178             fjy3             = _mm256_setzero_pd();
2179             fjz3             = _mm256_setzero_pd();
2180
2181             /**************************
2182              * CALCULATE INTERACTIONS *
2183              **************************/
2184
2185             r00              = _mm256_mul_pd(rsq00,rinv00);
2186             r00              = _mm256_andnot_pd(dummy_mask,r00);
2187
2188             /* Calculate table index by multiplying r with table scale and truncate to integer */
2189             rt               = _mm256_mul_pd(r00,vftabscale);
2190             vfitab           = _mm256_cvttpd_epi32(rt);
2191             vfeps            = _mm256_sub_pd(rt,_mm256_round_pd(rt, _MM_FROUND_FLOOR));
2192             vfitab           = _mm_slli_epi32(_mm_add_epi32(vfitab,_mm_slli_epi32(vfitab,1)),2);
2193
2194             /* CUBIC SPLINE TABLE DISPERSION */
2195             vfitab           = _mm_add_epi32(vfitab,ifour);
2196             Y                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
2197             F                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
2198             G                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,2) );
2199             H                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,3) );
2200             GMX_MM256_FULLTRANSPOSE4_PD(Y,F,G,H);
2201             Heps             = _mm256_mul_pd(vfeps,H);
2202             Fp               = _mm256_add_pd(F,_mm256_mul_pd(vfeps,_mm256_add_pd(G,Heps)));
2203             FF               = _mm256_add_pd(Fp,_mm256_mul_pd(vfeps,_mm256_add_pd(G,_mm256_add_pd(Heps,Heps))));
2204             fvdw6            = _mm256_mul_pd(c6_00,FF);
2205
2206             /* CUBIC SPLINE TABLE REPULSION */
2207             vfitab           = _mm_add_epi32(vfitab,ifour);
2208             Y                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
2209             F                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
2210             G                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,2) );
2211             H                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,3) );
2212             GMX_MM256_FULLTRANSPOSE4_PD(Y,F,G,H);
2213             Heps             = _mm256_mul_pd(vfeps,H);
2214             Fp               = _mm256_add_pd(F,_mm256_mul_pd(vfeps,_mm256_add_pd(G,Heps)));
2215             FF               = _mm256_add_pd(Fp,_mm256_mul_pd(vfeps,_mm256_add_pd(G,_mm256_add_pd(Heps,Heps))));
2216             fvdw12           = _mm256_mul_pd(c12_00,FF);
2217             fvdw             = _mm256_xor_pd(signbit,_mm256_mul_pd(_mm256_add_pd(fvdw6,fvdw12),_mm256_mul_pd(vftabscale,rinv00)));
2218
2219             fscal            = fvdw;
2220
2221             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
2222
2223             /* Calculate temporary vectorial force */
2224             tx               = _mm256_mul_pd(fscal,dx00);
2225             ty               = _mm256_mul_pd(fscal,dy00);
2226             tz               = _mm256_mul_pd(fscal,dz00);
2227
2228             /* Update vectorial force */
2229             fix0             = _mm256_add_pd(fix0,tx);
2230             fiy0             = _mm256_add_pd(fiy0,ty);
2231             fiz0             = _mm256_add_pd(fiz0,tz);
2232
2233             fjx0             = _mm256_add_pd(fjx0,tx);
2234             fjy0             = _mm256_add_pd(fjy0,ty);
2235             fjz0             = _mm256_add_pd(fjz0,tz);
2236
2237             /**************************
2238              * CALCULATE INTERACTIONS *
2239              **************************/
2240
2241             r11              = _mm256_mul_pd(rsq11,rinv11);
2242             r11              = _mm256_andnot_pd(dummy_mask,r11);
2243
2244             /* Calculate table index by multiplying r with table scale and truncate to integer */
2245             rt               = _mm256_mul_pd(r11,vftabscale);
2246             vfitab           = _mm256_cvttpd_epi32(rt);
2247             vfeps            = _mm256_sub_pd(rt,_mm256_round_pd(rt, _MM_FROUND_FLOOR));
2248             vfitab           = _mm_slli_epi32(_mm_add_epi32(vfitab,_mm_slli_epi32(vfitab,1)),2);
2249
2250             /* CUBIC SPLINE TABLE ELECTROSTATICS */
2251             Y                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
2252             F                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
2253             G                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,2) );
2254             H                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,3) );
2255             GMX_MM256_FULLTRANSPOSE4_PD(Y,F,G,H);
2256             Heps             = _mm256_mul_pd(vfeps,H);
2257             Fp               = _mm256_add_pd(F,_mm256_mul_pd(vfeps,_mm256_add_pd(G,Heps)));
2258             FF               = _mm256_add_pd(Fp,_mm256_mul_pd(vfeps,_mm256_add_pd(G,_mm256_add_pd(Heps,Heps))));
2259             felec            = _mm256_xor_pd(signbit,_mm256_mul_pd(_mm256_mul_pd(qq11,FF),_mm256_mul_pd(vftabscale,rinv11)));
2260
2261             fscal            = felec;
2262
2263             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
2264
2265             /* Calculate temporary vectorial force */
2266             tx               = _mm256_mul_pd(fscal,dx11);
2267             ty               = _mm256_mul_pd(fscal,dy11);
2268             tz               = _mm256_mul_pd(fscal,dz11);
2269
2270             /* Update vectorial force */
2271             fix1             = _mm256_add_pd(fix1,tx);
2272             fiy1             = _mm256_add_pd(fiy1,ty);
2273             fiz1             = _mm256_add_pd(fiz1,tz);
2274
2275             fjx1             = _mm256_add_pd(fjx1,tx);
2276             fjy1             = _mm256_add_pd(fjy1,ty);
2277             fjz1             = _mm256_add_pd(fjz1,tz);
2278
2279             /**************************
2280              * CALCULATE INTERACTIONS *
2281              **************************/
2282
2283             r12              = _mm256_mul_pd(rsq12,rinv12);
2284             r12              = _mm256_andnot_pd(dummy_mask,r12);
2285
2286             /* Calculate table index by multiplying r with table scale and truncate to integer */
2287             rt               = _mm256_mul_pd(r12,vftabscale);
2288             vfitab           = _mm256_cvttpd_epi32(rt);
2289             vfeps            = _mm256_sub_pd(rt,_mm256_round_pd(rt, _MM_FROUND_FLOOR));
2290             vfitab           = _mm_slli_epi32(_mm_add_epi32(vfitab,_mm_slli_epi32(vfitab,1)),2);
2291
2292             /* CUBIC SPLINE TABLE ELECTROSTATICS */
2293             Y                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
2294             F                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
2295             G                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,2) );
2296             H                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,3) );
2297             GMX_MM256_FULLTRANSPOSE4_PD(Y,F,G,H);
2298             Heps             = _mm256_mul_pd(vfeps,H);
2299             Fp               = _mm256_add_pd(F,_mm256_mul_pd(vfeps,_mm256_add_pd(G,Heps)));
2300             FF               = _mm256_add_pd(Fp,_mm256_mul_pd(vfeps,_mm256_add_pd(G,_mm256_add_pd(Heps,Heps))));
2301             felec            = _mm256_xor_pd(signbit,_mm256_mul_pd(_mm256_mul_pd(qq12,FF),_mm256_mul_pd(vftabscale,rinv12)));
2302
2303             fscal            = felec;
2304
2305             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
2306
2307             /* Calculate temporary vectorial force */
2308             tx               = _mm256_mul_pd(fscal,dx12);
2309             ty               = _mm256_mul_pd(fscal,dy12);
2310             tz               = _mm256_mul_pd(fscal,dz12);
2311
2312             /* Update vectorial force */
2313             fix1             = _mm256_add_pd(fix1,tx);
2314             fiy1             = _mm256_add_pd(fiy1,ty);
2315             fiz1             = _mm256_add_pd(fiz1,tz);
2316
2317             fjx2             = _mm256_add_pd(fjx2,tx);
2318             fjy2             = _mm256_add_pd(fjy2,ty);
2319             fjz2             = _mm256_add_pd(fjz2,tz);
2320
2321             /**************************
2322              * CALCULATE INTERACTIONS *
2323              **************************/
2324
2325             r13              = _mm256_mul_pd(rsq13,rinv13);
2326             r13              = _mm256_andnot_pd(dummy_mask,r13);
2327
2328             /* Calculate table index by multiplying r with table scale and truncate to integer */
2329             rt               = _mm256_mul_pd(r13,vftabscale);
2330             vfitab           = _mm256_cvttpd_epi32(rt);
2331             vfeps            = _mm256_sub_pd(rt,_mm256_round_pd(rt, _MM_FROUND_FLOOR));
2332             vfitab           = _mm_slli_epi32(_mm_add_epi32(vfitab,_mm_slli_epi32(vfitab,1)),2);
2333
2334             /* CUBIC SPLINE TABLE ELECTROSTATICS */
2335             Y                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
2336             F                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
2337             G                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,2) );
2338             H                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,3) );
2339             GMX_MM256_FULLTRANSPOSE4_PD(Y,F,G,H);
2340             Heps             = _mm256_mul_pd(vfeps,H);
2341             Fp               = _mm256_add_pd(F,_mm256_mul_pd(vfeps,_mm256_add_pd(G,Heps)));
2342             FF               = _mm256_add_pd(Fp,_mm256_mul_pd(vfeps,_mm256_add_pd(G,_mm256_add_pd(Heps,Heps))));
2343             felec            = _mm256_xor_pd(signbit,_mm256_mul_pd(_mm256_mul_pd(qq13,FF),_mm256_mul_pd(vftabscale,rinv13)));
2344
2345             fscal            = felec;
2346
2347             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
2348
2349             /* Calculate temporary vectorial force */
2350             tx               = _mm256_mul_pd(fscal,dx13);
2351             ty               = _mm256_mul_pd(fscal,dy13);
2352             tz               = _mm256_mul_pd(fscal,dz13);
2353
2354             /* Update vectorial force */
2355             fix1             = _mm256_add_pd(fix1,tx);
2356             fiy1             = _mm256_add_pd(fiy1,ty);
2357             fiz1             = _mm256_add_pd(fiz1,tz);
2358
2359             fjx3             = _mm256_add_pd(fjx3,tx);
2360             fjy3             = _mm256_add_pd(fjy3,ty);
2361             fjz3             = _mm256_add_pd(fjz3,tz);
2362
2363             /**************************
2364              * CALCULATE INTERACTIONS *
2365              **************************/
2366
2367             r21              = _mm256_mul_pd(rsq21,rinv21);
2368             r21              = _mm256_andnot_pd(dummy_mask,r21);
2369
2370             /* Calculate table index by multiplying r with table scale and truncate to integer */
2371             rt               = _mm256_mul_pd(r21,vftabscale);
2372             vfitab           = _mm256_cvttpd_epi32(rt);
2373             vfeps            = _mm256_sub_pd(rt,_mm256_round_pd(rt, _MM_FROUND_FLOOR));
2374             vfitab           = _mm_slli_epi32(_mm_add_epi32(vfitab,_mm_slli_epi32(vfitab,1)),2);
2375
2376             /* CUBIC SPLINE TABLE ELECTROSTATICS */
2377             Y                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
2378             F                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
2379             G                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,2) );
2380             H                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,3) );
2381             GMX_MM256_FULLTRANSPOSE4_PD(Y,F,G,H);
2382             Heps             = _mm256_mul_pd(vfeps,H);
2383             Fp               = _mm256_add_pd(F,_mm256_mul_pd(vfeps,_mm256_add_pd(G,Heps)));
2384             FF               = _mm256_add_pd(Fp,_mm256_mul_pd(vfeps,_mm256_add_pd(G,_mm256_add_pd(Heps,Heps))));
2385             felec            = _mm256_xor_pd(signbit,_mm256_mul_pd(_mm256_mul_pd(qq21,FF),_mm256_mul_pd(vftabscale,rinv21)));
2386
2387             fscal            = felec;
2388
2389             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
2390
2391             /* Calculate temporary vectorial force */
2392             tx               = _mm256_mul_pd(fscal,dx21);
2393             ty               = _mm256_mul_pd(fscal,dy21);
2394             tz               = _mm256_mul_pd(fscal,dz21);
2395
2396             /* Update vectorial force */
2397             fix2             = _mm256_add_pd(fix2,tx);
2398             fiy2             = _mm256_add_pd(fiy2,ty);
2399             fiz2             = _mm256_add_pd(fiz2,tz);
2400
2401             fjx1             = _mm256_add_pd(fjx1,tx);
2402             fjy1             = _mm256_add_pd(fjy1,ty);
2403             fjz1             = _mm256_add_pd(fjz1,tz);
2404
2405             /**************************
2406              * CALCULATE INTERACTIONS *
2407              **************************/
2408
2409             r22              = _mm256_mul_pd(rsq22,rinv22);
2410             r22              = _mm256_andnot_pd(dummy_mask,r22);
2411
2412             /* Calculate table index by multiplying r with table scale and truncate to integer */
2413             rt               = _mm256_mul_pd(r22,vftabscale);
2414             vfitab           = _mm256_cvttpd_epi32(rt);
2415             vfeps            = _mm256_sub_pd(rt,_mm256_round_pd(rt, _MM_FROUND_FLOOR));
2416             vfitab           = _mm_slli_epi32(_mm_add_epi32(vfitab,_mm_slli_epi32(vfitab,1)),2);
2417
2418             /* CUBIC SPLINE TABLE ELECTROSTATICS */
2419             Y                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
2420             F                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
2421             G                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,2) );
2422             H                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,3) );
2423             GMX_MM256_FULLTRANSPOSE4_PD(Y,F,G,H);
2424             Heps             = _mm256_mul_pd(vfeps,H);
2425             Fp               = _mm256_add_pd(F,_mm256_mul_pd(vfeps,_mm256_add_pd(G,Heps)));
2426             FF               = _mm256_add_pd(Fp,_mm256_mul_pd(vfeps,_mm256_add_pd(G,_mm256_add_pd(Heps,Heps))));
2427             felec            = _mm256_xor_pd(signbit,_mm256_mul_pd(_mm256_mul_pd(qq22,FF),_mm256_mul_pd(vftabscale,rinv22)));
2428
2429             fscal            = felec;
2430
2431             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
2432
2433             /* Calculate temporary vectorial force */
2434             tx               = _mm256_mul_pd(fscal,dx22);
2435             ty               = _mm256_mul_pd(fscal,dy22);
2436             tz               = _mm256_mul_pd(fscal,dz22);
2437
2438             /* Update vectorial force */
2439             fix2             = _mm256_add_pd(fix2,tx);
2440             fiy2             = _mm256_add_pd(fiy2,ty);
2441             fiz2             = _mm256_add_pd(fiz2,tz);
2442
2443             fjx2             = _mm256_add_pd(fjx2,tx);
2444             fjy2             = _mm256_add_pd(fjy2,ty);
2445             fjz2             = _mm256_add_pd(fjz2,tz);
2446
2447             /**************************
2448              * CALCULATE INTERACTIONS *
2449              **************************/
2450
2451             r23              = _mm256_mul_pd(rsq23,rinv23);
2452             r23              = _mm256_andnot_pd(dummy_mask,r23);
2453
2454             /* Calculate table index by multiplying r with table scale and truncate to integer */
2455             rt               = _mm256_mul_pd(r23,vftabscale);
2456             vfitab           = _mm256_cvttpd_epi32(rt);
2457             vfeps            = _mm256_sub_pd(rt,_mm256_round_pd(rt, _MM_FROUND_FLOOR));
2458             vfitab           = _mm_slli_epi32(_mm_add_epi32(vfitab,_mm_slli_epi32(vfitab,1)),2);
2459
2460             /* CUBIC SPLINE TABLE ELECTROSTATICS */
2461             Y                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
2462             F                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
2463             G                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,2) );
2464             H                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,3) );
2465             GMX_MM256_FULLTRANSPOSE4_PD(Y,F,G,H);
2466             Heps             = _mm256_mul_pd(vfeps,H);
2467             Fp               = _mm256_add_pd(F,_mm256_mul_pd(vfeps,_mm256_add_pd(G,Heps)));
2468             FF               = _mm256_add_pd(Fp,_mm256_mul_pd(vfeps,_mm256_add_pd(G,_mm256_add_pd(Heps,Heps))));
2469             felec            = _mm256_xor_pd(signbit,_mm256_mul_pd(_mm256_mul_pd(qq23,FF),_mm256_mul_pd(vftabscale,rinv23)));
2470
2471             fscal            = felec;
2472
2473             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
2474
2475             /* Calculate temporary vectorial force */
2476             tx               = _mm256_mul_pd(fscal,dx23);
2477             ty               = _mm256_mul_pd(fscal,dy23);
2478             tz               = _mm256_mul_pd(fscal,dz23);
2479
2480             /* Update vectorial force */
2481             fix2             = _mm256_add_pd(fix2,tx);
2482             fiy2             = _mm256_add_pd(fiy2,ty);
2483             fiz2             = _mm256_add_pd(fiz2,tz);
2484
2485             fjx3             = _mm256_add_pd(fjx3,tx);
2486             fjy3             = _mm256_add_pd(fjy3,ty);
2487             fjz3             = _mm256_add_pd(fjz3,tz);
2488
2489             /**************************
2490              * CALCULATE INTERACTIONS *
2491              **************************/
2492
2493             r31              = _mm256_mul_pd(rsq31,rinv31);
2494             r31              = _mm256_andnot_pd(dummy_mask,r31);
2495
2496             /* Calculate table index by multiplying r with table scale and truncate to integer */
2497             rt               = _mm256_mul_pd(r31,vftabscale);
2498             vfitab           = _mm256_cvttpd_epi32(rt);
2499             vfeps            = _mm256_sub_pd(rt,_mm256_round_pd(rt, _MM_FROUND_FLOOR));
2500             vfitab           = _mm_slli_epi32(_mm_add_epi32(vfitab,_mm_slli_epi32(vfitab,1)),2);
2501
2502             /* CUBIC SPLINE TABLE ELECTROSTATICS */
2503             Y                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
2504             F                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
2505             G                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,2) );
2506             H                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,3) );
2507             GMX_MM256_FULLTRANSPOSE4_PD(Y,F,G,H);
2508             Heps             = _mm256_mul_pd(vfeps,H);
2509             Fp               = _mm256_add_pd(F,_mm256_mul_pd(vfeps,_mm256_add_pd(G,Heps)));
2510             FF               = _mm256_add_pd(Fp,_mm256_mul_pd(vfeps,_mm256_add_pd(G,_mm256_add_pd(Heps,Heps))));
2511             felec            = _mm256_xor_pd(signbit,_mm256_mul_pd(_mm256_mul_pd(qq31,FF),_mm256_mul_pd(vftabscale,rinv31)));
2512
2513             fscal            = felec;
2514
2515             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
2516
2517             /* Calculate temporary vectorial force */
2518             tx               = _mm256_mul_pd(fscal,dx31);
2519             ty               = _mm256_mul_pd(fscal,dy31);
2520             tz               = _mm256_mul_pd(fscal,dz31);
2521
2522             /* Update vectorial force */
2523             fix3             = _mm256_add_pd(fix3,tx);
2524             fiy3             = _mm256_add_pd(fiy3,ty);
2525             fiz3             = _mm256_add_pd(fiz3,tz);
2526
2527             fjx1             = _mm256_add_pd(fjx1,tx);
2528             fjy1             = _mm256_add_pd(fjy1,ty);
2529             fjz1             = _mm256_add_pd(fjz1,tz);
2530
2531             /**************************
2532              * CALCULATE INTERACTIONS *
2533              **************************/
2534
2535             r32              = _mm256_mul_pd(rsq32,rinv32);
2536             r32              = _mm256_andnot_pd(dummy_mask,r32);
2537
2538             /* Calculate table index by multiplying r with table scale and truncate to integer */
2539             rt               = _mm256_mul_pd(r32,vftabscale);
2540             vfitab           = _mm256_cvttpd_epi32(rt);
2541             vfeps            = _mm256_sub_pd(rt,_mm256_round_pd(rt, _MM_FROUND_FLOOR));
2542             vfitab           = _mm_slli_epi32(_mm_add_epi32(vfitab,_mm_slli_epi32(vfitab,1)),2);
2543
2544             /* CUBIC SPLINE TABLE ELECTROSTATICS */
2545             Y                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
2546             F                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
2547             G                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,2) );
2548             H                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,3) );
2549             GMX_MM256_FULLTRANSPOSE4_PD(Y,F,G,H);
2550             Heps             = _mm256_mul_pd(vfeps,H);
2551             Fp               = _mm256_add_pd(F,_mm256_mul_pd(vfeps,_mm256_add_pd(G,Heps)));
2552             FF               = _mm256_add_pd(Fp,_mm256_mul_pd(vfeps,_mm256_add_pd(G,_mm256_add_pd(Heps,Heps))));
2553             felec            = _mm256_xor_pd(signbit,_mm256_mul_pd(_mm256_mul_pd(qq32,FF),_mm256_mul_pd(vftabscale,rinv32)));
2554
2555             fscal            = felec;
2556
2557             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
2558
2559             /* Calculate temporary vectorial force */
2560             tx               = _mm256_mul_pd(fscal,dx32);
2561             ty               = _mm256_mul_pd(fscal,dy32);
2562             tz               = _mm256_mul_pd(fscal,dz32);
2563
2564             /* Update vectorial force */
2565             fix3             = _mm256_add_pd(fix3,tx);
2566             fiy3             = _mm256_add_pd(fiy3,ty);
2567             fiz3             = _mm256_add_pd(fiz3,tz);
2568
2569             fjx2             = _mm256_add_pd(fjx2,tx);
2570             fjy2             = _mm256_add_pd(fjy2,ty);
2571             fjz2             = _mm256_add_pd(fjz2,tz);
2572
2573             /**************************
2574              * CALCULATE INTERACTIONS *
2575              **************************/
2576
2577             r33              = _mm256_mul_pd(rsq33,rinv33);
2578             r33              = _mm256_andnot_pd(dummy_mask,r33);
2579
2580             /* Calculate table index by multiplying r with table scale and truncate to integer */
2581             rt               = _mm256_mul_pd(r33,vftabscale);
2582             vfitab           = _mm256_cvttpd_epi32(rt);
2583             vfeps            = _mm256_sub_pd(rt,_mm256_round_pd(rt, _MM_FROUND_FLOOR));
2584             vfitab           = _mm_slli_epi32(_mm_add_epi32(vfitab,_mm_slli_epi32(vfitab,1)),2);
2585
2586             /* CUBIC SPLINE TABLE ELECTROSTATICS */
2587             Y                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
2588             F                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
2589             G                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,2) );
2590             H                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,3) );
2591             GMX_MM256_FULLTRANSPOSE4_PD(Y,F,G,H);
2592             Heps             = _mm256_mul_pd(vfeps,H);
2593             Fp               = _mm256_add_pd(F,_mm256_mul_pd(vfeps,_mm256_add_pd(G,Heps)));
2594             FF               = _mm256_add_pd(Fp,_mm256_mul_pd(vfeps,_mm256_add_pd(G,_mm256_add_pd(Heps,Heps))));
2595             felec            = _mm256_xor_pd(signbit,_mm256_mul_pd(_mm256_mul_pd(qq33,FF),_mm256_mul_pd(vftabscale,rinv33)));
2596
2597             fscal            = felec;
2598
2599             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
2600
2601             /* Calculate temporary vectorial force */
2602             tx               = _mm256_mul_pd(fscal,dx33);
2603             ty               = _mm256_mul_pd(fscal,dy33);
2604             tz               = _mm256_mul_pd(fscal,dz33);
2605
2606             /* Update vectorial force */
2607             fix3             = _mm256_add_pd(fix3,tx);
2608             fiy3             = _mm256_add_pd(fiy3,ty);
2609             fiz3             = _mm256_add_pd(fiz3,tz);
2610
2611             fjx3             = _mm256_add_pd(fjx3,tx);
2612             fjy3             = _mm256_add_pd(fjy3,ty);
2613             fjz3             = _mm256_add_pd(fjz3,tz);
2614
2615             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
2616             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
2617             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
2618             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
2619
2620             gmx_mm256_decrement_4rvec_4ptr_swizzle_pd(fjptrA,fjptrB,fjptrC,fjptrD,
2621                                                       fjx0,fjy0,fjz0,fjx1,fjy1,fjz1,
2622                                                       fjx2,fjy2,fjz2,fjx3,fjy3,fjz3);
2623
2624             /* Inner loop uses 412 flops */
2625         }
2626
2627         /* End of innermost loop */
2628
2629         gmx_mm256_update_iforce_4atom_swizzle_pd(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,fix3,fiy3,fiz3,
2630                                                  f+i_coord_offset,fshift+i_shift_offset);
2631
2632         /* Increment number of inner iterations */
2633         inneriter                  += j_index_end - j_index_start;
2634
2635         /* Outer loop uses 24 flops */
2636     }
2637
2638     /* Increment number of outer iterations */
2639     outeriter        += nri;
2640
2641     /* Update outer/inner flops */
2642
2643     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W4W4_F,outeriter*24 + inneriter*412);
2644 }