src/gromacs/gmxlib/nonbonded/nb_kernel_avx_256_double/nb_kernel_ElecCSTab_VdwNone_GeomW4W4_avx_256_double.c

   1 /*
   2  * This file is part of the GROMACS molecular simulation package.
   3  *
   4  * Copyright (c) 2012,2013,2014, by the GROMACS development team, led by
   5  * Mark Abraham, David van der Spoel, Berk Hess, and Erik Lindahl,
   6  * and including many others, as listed in the AUTHORS file in the
   7  * top-level source directory and at http://www.gromacs.org.
   8  *
   9  * GROMACS is free software; you can redistribute it and/or
  10  * modify it under the terms of the GNU Lesser General Public License
  11  * as published by the Free Software Foundation; either version 2.1
  12  * of the License, or (at your option) any later version.
  13  *
  14  * GROMACS is distributed in the hope that it will be useful,
  15  * but WITHOUT ANY WARRANTY; without even the implied warranty of
  16  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
  17  * Lesser General Public License for more details.
  18  *
  19  * You should have received a copy of the GNU Lesser General Public
  20  * License along with GROMACS; if not, see
  21  * http://www.gnu.org/licenses, or write to the Free Software Foundation,
  22  * Inc., 51 Franklin Street, Fifth Floor, Boston, MA  02110-1301  USA.
  23  *
  24  * If you want to redistribute modifications to GROMACS, please
  25  * consider that scientific software is very special. Version
  26  * control is crucial - bugs must be traceable. We will be happy to
  27  * consider code for inclusion in the official distribution, but
  28  * derived work must not be called official GROMACS. Details are found
  29  * in the README & COPYING files - if they are missing, get the
  30  * official version at http://www.gromacs.org.
  31  *
  32  * To help us fund GROMACS development, we humbly ask that you cite
  33  * the research papers on the package. Check out http://www.gromacs.org.
  34  */
  35 /*
  36  * Note: this file was generated by the GROMACS avx_256_double kernel generator.
  37  */
  38 #include "gmxpre.h"
  39
  40 #include "config.h"
  41
  42 #include <math.h>
  43
  44 #include "../nb_kernel.h"
  45 #include "gromacs/legacyheaders/types/simple.h"
  46 #include "gromacs/math/vec.h"
  47 #include "gromacs/legacyheaders/nrnb.h"
  48
  49 #include "gromacs/simd/math_x86_avx_256_double.h"
  50 #include "kernelutil_x86_avx_256_double.h"
  51
  52 /*
  53  * Gromacs nonbonded kernel:   nb_kernel_ElecCSTab_VdwNone_GeomW4W4_VF_avx_256_double
  54  * Electrostatics interaction: CubicSplineTable
  55  * VdW interaction:            None
  56  * Geometry:                   Water4-Water4
  57  * Calculate force/pot:        PotentialAndForce
  58  */
  59 void
  60 nb_kernel_ElecCSTab_VdwNone_GeomW4W4_VF_avx_256_double
  61                     (t_nblist                    * gmx_restrict       nlist,
  62                      rvec                        * gmx_restrict          xx,
  63                      rvec                        * gmx_restrict          ff,
  64                      t_forcerec                  * gmx_restrict          fr,
  65                      t_mdatoms                   * gmx_restrict     mdatoms,
  66                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
  67                      t_nrnb                      * gmx_restrict        nrnb)
  68 {
  69     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
  70      * just 0 for non-waters.
  71      * Suffixes A,B,C,D refer to j loop unrolling done with AVX, e.g. for the four different
  72      * jnr indices corresponding to data put in the four positions in the SIMD register.
  73      */
  74     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
  75     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
  76     int              jnrA,jnrB,jnrC,jnrD;
  77     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
  78     int              jnrlistE,jnrlistF,jnrlistG,jnrlistH;
  79     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
  80     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
  81     real             rcutoff_scalar;
  82     real             *shiftvec,*fshift,*x,*f;
  83     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD;
  84     real             scratch[4*DIM];
  85     __m256d          tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
  86     real *           vdwioffsetptr1;
  87     __m256d          ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
  88     real *           vdwioffsetptr2;
  89     __m256d          ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
  90     real *           vdwioffsetptr3;
  91     __m256d          ix3,iy3,iz3,fix3,fiy3,fiz3,iq3,isai3;
  92     int              vdwjidx1A,vdwjidx1B,vdwjidx1C,vdwjidx1D;
  93     __m256d          jx1,jy1,jz1,fjx1,fjy1,fjz1,jq1,isaj1;
  94     int              vdwjidx2A,vdwjidx2B,vdwjidx2C,vdwjidx2D;
  95     __m256d          jx2,jy2,jz2,fjx2,fjy2,fjz2,jq2,isaj2;
  96     int              vdwjidx3A,vdwjidx3B,vdwjidx3C,vdwjidx3D;
  97     __m256d          jx3,jy3,jz3,fjx3,fjy3,fjz3,jq3,isaj3;
  98     __m256d          dx11,dy11,dz11,rsq11,rinv11,rinvsq11,r11,qq11,c6_11,c12_11;
  99     __m256d          dx12,dy12,dz12,rsq12,rinv12,rinvsq12,r12,qq12,c6_12,c12_12;
 100     __m256d          dx13,dy13,dz13,rsq13,rinv13,rinvsq13,r13,qq13,c6_13,c12_13;
 101     __m256d          dx21,dy21,dz21,rsq21,rinv21,rinvsq21,r21,qq21,c6_21,c12_21;
 102     __m256d          dx22,dy22,dz22,rsq22,rinv22,rinvsq22,r22,qq22,c6_22,c12_22;
 103     __m256d          dx23,dy23,dz23,rsq23,rinv23,rinvsq23,r23,qq23,c6_23,c12_23;
 104     __m256d          dx31,dy31,dz31,rsq31,rinv31,rinvsq31,r31,qq31,c6_31,c12_31;
 105     __m256d          dx32,dy32,dz32,rsq32,rinv32,rinvsq32,r32,qq32,c6_32,c12_32;
 106     __m256d          dx33,dy33,dz33,rsq33,rinv33,rinvsq33,r33,qq33,c6_33,c12_33;
 107     __m256d          velec,felec,velecsum,facel,crf,krf,krf2;
 108     real             *charge;
 109     __m128i          vfitab;
 110     __m128i          ifour       = _mm_set1_epi32(4);
 111     __m256d          rt,vfeps,vftabscale,Y,F,G,H,Heps,Fp,VV,FF;
 112     real             *vftab;
 113     __m256d          dummy_mask,cutoff_mask;
 114     __m128           tmpmask0,tmpmask1;
 115     __m256d          signbit = _mm256_castsi256_pd( _mm256_set1_epi32(0x80000000) );
 116     __m256d          one     = _mm256_set1_pd(1.0);
 117     __m256d          two     = _mm256_set1_pd(2.0);
 118     x                = xx[0];
 119     f                = ff[0];
 120
 121     nri              = nlist->nri;
 122     iinr             = nlist->iinr;
 123     jindex           = nlist->jindex;
 124     jjnr             = nlist->jjnr;
 125     shiftidx         = nlist->shift;
 126     gid              = nlist->gid;
 127     shiftvec         = fr->shift_vec[0];
 128     fshift           = fr->fshift[0];
 129     facel            = _mm256_set1_pd(fr->epsfac);
 130     charge           = mdatoms->chargeA;
 131
 132     vftab            = kernel_data->table_elec->data;
 133     vftabscale       = _mm256_set1_pd(kernel_data->table_elec->scale);
 134
 135     /* Setup water-specific parameters */
 136     inr              = nlist->iinr[0];
 137     iq1              = _mm256_mul_pd(facel,_mm256_set1_pd(charge[inr+1]));
 138     iq2              = _mm256_mul_pd(facel,_mm256_set1_pd(charge[inr+2]));
 139     iq3              = _mm256_mul_pd(facel,_mm256_set1_pd(charge[inr+3]));
 140
 141     jq1              = _mm256_set1_pd(charge[inr+1]);
 142     jq2              = _mm256_set1_pd(charge[inr+2]);
 143     jq3              = _mm256_set1_pd(charge[inr+3]);
 144     qq11             = _mm256_mul_pd(iq1,jq1);
 145     qq12             = _mm256_mul_pd(iq1,jq2);
 146     qq13             = _mm256_mul_pd(iq1,jq3);
 147     qq21             = _mm256_mul_pd(iq2,jq1);
 148     qq22             = _mm256_mul_pd(iq2,jq2);
 149     qq23             = _mm256_mul_pd(iq2,jq3);
 150     qq31             = _mm256_mul_pd(iq3,jq1);
 151     qq32             = _mm256_mul_pd(iq3,jq2);
 152     qq33             = _mm256_mul_pd(iq3,jq3);
 153
 154     /* Avoid stupid compiler warnings */
 155     jnrA = jnrB = jnrC = jnrD = 0;
 156     j_coord_offsetA = 0;
 157     j_coord_offsetB = 0;
 158     j_coord_offsetC = 0;
 159     j_coord_offsetD = 0;
 160
 161     outeriter        = 0;
 162     inneriter        = 0;
 163
 164     for(iidx=0;iidx<4*DIM;iidx++)
 165     {
 166         scratch[iidx] = 0.0;
 167     }
 168
 169     /* Start outer loop over neighborlists */
 170     for(iidx=0; iidx<nri; iidx++)
 171     {
 172         /* Load shift vector for this list */
 173         i_shift_offset   = DIM*shiftidx[iidx];
 174
 175         /* Load limits for loop over neighbors */
 176         j_index_start    = jindex[iidx];
 177         j_index_end      = jindex[iidx+1];
 178
 179         /* Get outer coordinate index */
 180         inr              = iinr[iidx];
 181         i_coord_offset   = DIM*inr;
 182
 183         /* Load i particle coords and add shift vector */
 184         gmx_mm256_load_shift_and_3rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset+DIM,
 185                                                     &ix1,&iy1,&iz1,&ix2,&iy2,&iz2,&ix3,&iy3,&iz3);
 186
 187         fix1             = _mm256_setzero_pd();
 188         fiy1             = _mm256_setzero_pd();
 189         fiz1             = _mm256_setzero_pd();
 190         fix2             = _mm256_setzero_pd();
 191         fiy2             = _mm256_setzero_pd();
 192         fiz2             = _mm256_setzero_pd();
 193         fix3             = _mm256_setzero_pd();
 194         fiy3             = _mm256_setzero_pd();
 195         fiz3             = _mm256_setzero_pd();
 196
 197         /* Reset potential sums */
 198         velecsum         = _mm256_setzero_pd();
 199
 200         /* Start inner kernel loop */
 201         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
 202         {
 203
 204             /* Get j neighbor index, and coordinate index */
 205             jnrA             = jjnr[jidx];
 206             jnrB             = jjnr[jidx+1];
 207             jnrC             = jjnr[jidx+2];
 208             jnrD             = jjnr[jidx+3];
 209             j_coord_offsetA  = DIM*jnrA;
 210             j_coord_offsetB  = DIM*jnrB;
 211             j_coord_offsetC  = DIM*jnrC;
 212             j_coord_offsetD  = DIM*jnrD;
 213
 214             /* load j atom coordinates */
 215             gmx_mm256_load_3rvec_4ptr_swizzle_pd(x+j_coord_offsetA+DIM,x+j_coord_offsetB+DIM,
 216                                                  x+j_coord_offsetC+DIM,x+j_coord_offsetD+DIM,
 217                                                  &jx1,&jy1,&jz1,&jx2,&jy2,&jz2,&jx3,&jy3,&jz3);
 218
 219             /* Calculate displacement vector */
 220             dx11             = _mm256_sub_pd(ix1,jx1);
 221             dy11             = _mm256_sub_pd(iy1,jy1);
 222             dz11             = _mm256_sub_pd(iz1,jz1);
 223             dx12             = _mm256_sub_pd(ix1,jx2);
 224             dy12             = _mm256_sub_pd(iy1,jy2);
 225             dz12             = _mm256_sub_pd(iz1,jz2);
 226             dx13             = _mm256_sub_pd(ix1,jx3);
 227             dy13             = _mm256_sub_pd(iy1,jy3);
 228             dz13             = _mm256_sub_pd(iz1,jz3);
 229             dx21             = _mm256_sub_pd(ix2,jx1);
 230             dy21             = _mm256_sub_pd(iy2,jy1);
 231             dz21             = _mm256_sub_pd(iz2,jz1);
 232             dx22             = _mm256_sub_pd(ix2,jx2);
 233             dy22             = _mm256_sub_pd(iy2,jy2);
 234             dz22             = _mm256_sub_pd(iz2,jz2);
 235             dx23             = _mm256_sub_pd(ix2,jx3);
 236             dy23             = _mm256_sub_pd(iy2,jy3);
 237             dz23             = _mm256_sub_pd(iz2,jz3);
 238             dx31             = _mm256_sub_pd(ix3,jx1);
 239             dy31             = _mm256_sub_pd(iy3,jy1);
 240             dz31             = _mm256_sub_pd(iz3,jz1);
 241             dx32             = _mm256_sub_pd(ix3,jx2);
 242             dy32             = _mm256_sub_pd(iy3,jy2);
 243             dz32             = _mm256_sub_pd(iz3,jz2);
 244             dx33             = _mm256_sub_pd(ix3,jx3);
 245             dy33             = _mm256_sub_pd(iy3,jy3);
 246             dz33             = _mm256_sub_pd(iz3,jz3);
 247
 248             /* Calculate squared distance and things based on it */
 249             rsq11            = gmx_mm256_calc_rsq_pd(dx11,dy11,dz11);
 250             rsq12            = gmx_mm256_calc_rsq_pd(dx12,dy12,dz12);
 251             rsq13            = gmx_mm256_calc_rsq_pd(dx13,dy13,dz13);
 252             rsq21            = gmx_mm256_calc_rsq_pd(dx21,dy21,dz21);
 253             rsq22            = gmx_mm256_calc_rsq_pd(dx22,dy22,dz22);
 254             rsq23            = gmx_mm256_calc_rsq_pd(dx23,dy23,dz23);
 255             rsq31            = gmx_mm256_calc_rsq_pd(dx31,dy31,dz31);
 256             rsq32            = gmx_mm256_calc_rsq_pd(dx32,dy32,dz32);
 257             rsq33            = gmx_mm256_calc_rsq_pd(dx33,dy33,dz33);
 258
 259             rinv11           = gmx_mm256_invsqrt_pd(rsq11);
 260             rinv12           = gmx_mm256_invsqrt_pd(rsq12);
 261             rinv13           = gmx_mm256_invsqrt_pd(rsq13);
 262             rinv21           = gmx_mm256_invsqrt_pd(rsq21);
 263             rinv22           = gmx_mm256_invsqrt_pd(rsq22);
 264             rinv23           = gmx_mm256_invsqrt_pd(rsq23);
 265             rinv31           = gmx_mm256_invsqrt_pd(rsq31);
 266             rinv32           = gmx_mm256_invsqrt_pd(rsq32);
 267             rinv33           = gmx_mm256_invsqrt_pd(rsq33);
 268
 269             fjx1             = _mm256_setzero_pd();
 270             fjy1             = _mm256_setzero_pd();
 271             fjz1             = _mm256_setzero_pd();
 272             fjx2             = _mm256_setzero_pd();
 273             fjy2             = _mm256_setzero_pd();
 274             fjz2             = _mm256_setzero_pd();
 275             fjx3             = _mm256_setzero_pd();
 276             fjy3             = _mm256_setzero_pd();
 277             fjz3             = _mm256_setzero_pd();
 278
 279             /**************************
 280              * CALCULATE INTERACTIONS *
 281              **************************/
 282
 283             r11              = _mm256_mul_pd(rsq11,rinv11);
 284
 285             /* Calculate table index by multiplying r with table scale and truncate to integer */
 286             rt               = _mm256_mul_pd(r11,vftabscale);
 287             vfitab           = _mm256_cvttpd_epi32(rt);
 288             vfeps            = _mm256_sub_pd(rt,_mm256_round_pd(rt, _MM_FROUND_FLOOR));
 289             vfitab           = _mm_slli_epi32(vfitab,2);
 290
 291             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 292             Y                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
 293             F                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
 294             G                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,2) );
 295             H                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,3) );
 296             GMX_MM256_FULLTRANSPOSE4_PD(Y,F,G,H);
 297             Heps             = _mm256_mul_pd(vfeps,H);
 298             Fp               = _mm256_add_pd(F,_mm256_mul_pd(vfeps,_mm256_add_pd(G,Heps)));
 299             VV               = _mm256_add_pd(Y,_mm256_mul_pd(vfeps,Fp));
 300             velec            = _mm256_mul_pd(qq11,VV);
 301             FF               = _mm256_add_pd(Fp,_mm256_mul_pd(vfeps,_mm256_add_pd(G,_mm256_add_pd(Heps,Heps))));
 302             felec            = _mm256_xor_pd(signbit,_mm256_mul_pd(_mm256_mul_pd(qq11,FF),_mm256_mul_pd(vftabscale,rinv11)));
 303
 304             /* Update potential sum for this i atom from the interaction with this j atom. */
 305             velecsum         = _mm256_add_pd(velecsum,velec);
 306
 307             fscal            = felec;
 308
 309             /* Calculate temporary vectorial force */
 310             tx               = _mm256_mul_pd(fscal,dx11);
 311             ty               = _mm256_mul_pd(fscal,dy11);
 312             tz               = _mm256_mul_pd(fscal,dz11);
 313
 314             /* Update vectorial force */
 315             fix1             = _mm256_add_pd(fix1,tx);
 316             fiy1             = _mm256_add_pd(fiy1,ty);
 317             fiz1             = _mm256_add_pd(fiz1,tz);
 318
 319             fjx1             = _mm256_add_pd(fjx1,tx);
 320             fjy1             = _mm256_add_pd(fjy1,ty);
 321             fjz1             = _mm256_add_pd(fjz1,tz);
 322
 323             /**************************
 324              * CALCULATE INTERACTIONS *
 325              **************************/
 326
 327             r12              = _mm256_mul_pd(rsq12,rinv12);
 328
 329             /* Calculate table index by multiplying r with table scale and truncate to integer */
 330             rt               = _mm256_mul_pd(r12,vftabscale);
 331             vfitab           = _mm256_cvttpd_epi32(rt);
 332             vfeps            = _mm256_sub_pd(rt,_mm256_round_pd(rt, _MM_FROUND_FLOOR));
 333             vfitab           = _mm_slli_epi32(vfitab,2);
 334
 335             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 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             velec            = _mm256_mul_pd(qq12,VV);
 345             FF               = _mm256_add_pd(Fp,_mm256_mul_pd(vfeps,_mm256_add_pd(G,_mm256_add_pd(Heps,Heps))));
 346             felec            = _mm256_xor_pd(signbit,_mm256_mul_pd(_mm256_mul_pd(qq12,FF),_mm256_mul_pd(vftabscale,rinv12)));
 347
 348             /* Update potential sum for this i atom from the interaction with this j atom. */
 349             velecsum         = _mm256_add_pd(velecsum,velec);
 350
 351             fscal            = felec;
 352
 353             /* Calculate temporary vectorial force */
 354             tx               = _mm256_mul_pd(fscal,dx12);
 355             ty               = _mm256_mul_pd(fscal,dy12);
 356             tz               = _mm256_mul_pd(fscal,dz12);
 357
 358             /* Update vectorial force */
 359             fix1             = _mm256_add_pd(fix1,tx);
 360             fiy1             = _mm256_add_pd(fiy1,ty);
 361             fiz1             = _mm256_add_pd(fiz1,tz);
 362
 363             fjx2             = _mm256_add_pd(fjx2,tx);
 364             fjy2             = _mm256_add_pd(fjy2,ty);
 365             fjz2             = _mm256_add_pd(fjz2,tz);
 366
 367             /**************************
 368              * CALCULATE INTERACTIONS *
 369              **************************/
 370
 371             r13              = _mm256_mul_pd(rsq13,rinv13);
 372
 373             /* Calculate table index by multiplying r with table scale and truncate to integer */
 374             rt               = _mm256_mul_pd(r13,vftabscale);
 375             vfitab           = _mm256_cvttpd_epi32(rt);
 376             vfeps            = _mm256_sub_pd(rt,_mm256_round_pd(rt, _MM_FROUND_FLOOR));
 377             vfitab           = _mm_slli_epi32(vfitab,2);
 378
 379             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 380             Y                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
 381             F                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
 382             G                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,2) );
 383             H                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,3) );
 384             GMX_MM256_FULLTRANSPOSE4_PD(Y,F,G,H);
 385             Heps             = _mm256_mul_pd(vfeps,H);
 386             Fp               = _mm256_add_pd(F,_mm256_mul_pd(vfeps,_mm256_add_pd(G,Heps)));
 387             VV               = _mm256_add_pd(Y,_mm256_mul_pd(vfeps,Fp));
 388             velec            = _mm256_mul_pd(qq13,VV);
 389             FF               = _mm256_add_pd(Fp,_mm256_mul_pd(vfeps,_mm256_add_pd(G,_mm256_add_pd(Heps,Heps))));
 390             felec            = _mm256_xor_pd(signbit,_mm256_mul_pd(_mm256_mul_pd(qq13,FF),_mm256_mul_pd(vftabscale,rinv13)));
 391
 392             /* Update potential sum for this i atom from the interaction with this j atom. */
 393             velecsum         = _mm256_add_pd(velecsum,velec);
 394
 395             fscal            = felec;
 396
 397             /* Calculate temporary vectorial force */
 398             tx               = _mm256_mul_pd(fscal,dx13);
 399             ty               = _mm256_mul_pd(fscal,dy13);
 400             tz               = _mm256_mul_pd(fscal,dz13);
 401
 402             /* Update vectorial force */
 403             fix1             = _mm256_add_pd(fix1,tx);
 404             fiy1             = _mm256_add_pd(fiy1,ty);
 405             fiz1             = _mm256_add_pd(fiz1,tz);
 406
 407             fjx3             = _mm256_add_pd(fjx3,tx);
 408             fjy3             = _mm256_add_pd(fjy3,ty);
 409             fjz3             = _mm256_add_pd(fjz3,tz);
 410
 411             /**************************
 412              * CALCULATE INTERACTIONS *
 413              **************************/
 414
 415             r21              = _mm256_mul_pd(rsq21,rinv21);
 416
 417             /* Calculate table index by multiplying r with table scale and truncate to integer */
 418             rt               = _mm256_mul_pd(r21,vftabscale);
 419             vfitab           = _mm256_cvttpd_epi32(rt);
 420             vfeps            = _mm256_sub_pd(rt,_mm256_round_pd(rt, _MM_FROUND_FLOOR));
 421             vfitab           = _mm_slli_epi32(vfitab,2);
 422
 423             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 424             Y                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
 425             F                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
 426             G                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,2) );
 427             H                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,3) );
 428             GMX_MM256_FULLTRANSPOSE4_PD(Y,F,G,H);
 429             Heps             = _mm256_mul_pd(vfeps,H);
 430             Fp               = _mm256_add_pd(F,_mm256_mul_pd(vfeps,_mm256_add_pd(G,Heps)));
 431             VV               = _mm256_add_pd(Y,_mm256_mul_pd(vfeps,Fp));
 432             velec            = _mm256_mul_pd(qq21,VV);
 433             FF               = _mm256_add_pd(Fp,_mm256_mul_pd(vfeps,_mm256_add_pd(G,_mm256_add_pd(Heps,Heps))));
 434             felec            = _mm256_xor_pd(signbit,_mm256_mul_pd(_mm256_mul_pd(qq21,FF),_mm256_mul_pd(vftabscale,rinv21)));
 435
 436             /* Update potential sum for this i atom from the interaction with this j atom. */
 437             velecsum         = _mm256_add_pd(velecsum,velec);
 438
 439             fscal            = felec;
 440
 441             /* Calculate temporary vectorial force */
 442             tx               = _mm256_mul_pd(fscal,dx21);
 443             ty               = _mm256_mul_pd(fscal,dy21);
 444             tz               = _mm256_mul_pd(fscal,dz21);
 445
 446             /* Update vectorial force */
 447             fix2             = _mm256_add_pd(fix2,tx);
 448             fiy2             = _mm256_add_pd(fiy2,ty);
 449             fiz2             = _mm256_add_pd(fiz2,tz);
 450
 451             fjx1             = _mm256_add_pd(fjx1,tx);
 452             fjy1             = _mm256_add_pd(fjy1,ty);
 453             fjz1             = _mm256_add_pd(fjz1,tz);
 454
 455             /**************************
 456              * CALCULATE INTERACTIONS *
 457              **************************/
 458
 459             r22              = _mm256_mul_pd(rsq22,rinv22);
 460
 461             /* Calculate table index by multiplying r with table scale and truncate to integer */
 462             rt               = _mm256_mul_pd(r22,vftabscale);
 463             vfitab           = _mm256_cvttpd_epi32(rt);
 464             vfeps            = _mm256_sub_pd(rt,_mm256_round_pd(rt, _MM_FROUND_FLOOR));
 465             vfitab           = _mm_slli_epi32(vfitab,2);
 466
 467             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 468             Y                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
 469             F                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
 470             G                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,2) );
 471             H                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,3) );
 472             GMX_MM256_FULLTRANSPOSE4_PD(Y,F,G,H);
 473             Heps             = _mm256_mul_pd(vfeps,H);
 474             Fp               = _mm256_add_pd(F,_mm256_mul_pd(vfeps,_mm256_add_pd(G,Heps)));
 475             VV               = _mm256_add_pd(Y,_mm256_mul_pd(vfeps,Fp));
 476             velec            = _mm256_mul_pd(qq22,VV);
 477             FF               = _mm256_add_pd(Fp,_mm256_mul_pd(vfeps,_mm256_add_pd(G,_mm256_add_pd(Heps,Heps))));
 478             felec            = _mm256_xor_pd(signbit,_mm256_mul_pd(_mm256_mul_pd(qq22,FF),_mm256_mul_pd(vftabscale,rinv22)));
 479
 480             /* Update potential sum for this i atom from the interaction with this j atom. */
 481             velecsum         = _mm256_add_pd(velecsum,velec);
 482
 483             fscal            = felec;
 484
 485             /* Calculate temporary vectorial force */
 486             tx               = _mm256_mul_pd(fscal,dx22);
 487             ty               = _mm256_mul_pd(fscal,dy22);
 488             tz               = _mm256_mul_pd(fscal,dz22);
 489
 490             /* Update vectorial force */
 491             fix2             = _mm256_add_pd(fix2,tx);
 492             fiy2             = _mm256_add_pd(fiy2,ty);
 493             fiz2             = _mm256_add_pd(fiz2,tz);
 494
 495             fjx2             = _mm256_add_pd(fjx2,tx);
 496             fjy2             = _mm256_add_pd(fjy2,ty);
 497             fjz2             = _mm256_add_pd(fjz2,tz);
 498
 499             /**************************
 500              * CALCULATE INTERACTIONS *
 501              **************************/
 502
 503             r23              = _mm256_mul_pd(rsq23,rinv23);
 504
 505             /* Calculate table index by multiplying r with table scale and truncate to integer */
 506             rt               = _mm256_mul_pd(r23,vftabscale);
 507             vfitab           = _mm256_cvttpd_epi32(rt);
 508             vfeps            = _mm256_sub_pd(rt,_mm256_round_pd(rt, _MM_FROUND_FLOOR));
 509             vfitab           = _mm_slli_epi32(vfitab,2);
 510
 511             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 512             Y                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
 513             F                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
 514             G                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,2) );
 515             H                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,3) );
 516             GMX_MM256_FULLTRANSPOSE4_PD(Y,F,G,H);
 517             Heps             = _mm256_mul_pd(vfeps,H);
 518             Fp               = _mm256_add_pd(F,_mm256_mul_pd(vfeps,_mm256_add_pd(G,Heps)));
 519             VV               = _mm256_add_pd(Y,_mm256_mul_pd(vfeps,Fp));
 520             velec            = _mm256_mul_pd(qq23,VV);
 521             FF               = _mm256_add_pd(Fp,_mm256_mul_pd(vfeps,_mm256_add_pd(G,_mm256_add_pd(Heps,Heps))));
 522             felec            = _mm256_xor_pd(signbit,_mm256_mul_pd(_mm256_mul_pd(qq23,FF),_mm256_mul_pd(vftabscale,rinv23)));
 523
 524             /* Update potential sum for this i atom from the interaction with this j atom. */
 525             velecsum         = _mm256_add_pd(velecsum,velec);
 526
 527             fscal            = felec;
 528
 529             /* Calculate temporary vectorial force */
 530             tx               = _mm256_mul_pd(fscal,dx23);
 531             ty               = _mm256_mul_pd(fscal,dy23);
 532             tz               = _mm256_mul_pd(fscal,dz23);
 533
 534             /* Update vectorial force */
 535             fix2             = _mm256_add_pd(fix2,tx);
 536             fiy2             = _mm256_add_pd(fiy2,ty);
 537             fiz2             = _mm256_add_pd(fiz2,tz);
 538
 539             fjx3             = _mm256_add_pd(fjx3,tx);
 540             fjy3             = _mm256_add_pd(fjy3,ty);
 541             fjz3             = _mm256_add_pd(fjz3,tz);
 542
 543             /**************************
 544              * CALCULATE INTERACTIONS *
 545              **************************/
 546
 547             r31              = _mm256_mul_pd(rsq31,rinv31);
 548
 549             /* Calculate table index by multiplying r with table scale and truncate to integer */
 550             rt               = _mm256_mul_pd(r31,vftabscale);
 551             vfitab           = _mm256_cvttpd_epi32(rt);
 552             vfeps            = _mm256_sub_pd(rt,_mm256_round_pd(rt, _MM_FROUND_FLOOR));
 553             vfitab           = _mm_slli_epi32(vfitab,2);
 554
 555             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 556             Y                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
 557             F                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
 558             G                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,2) );
 559             H                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,3) );
 560             GMX_MM256_FULLTRANSPOSE4_PD(Y,F,G,H);
 561             Heps             = _mm256_mul_pd(vfeps,H);
 562             Fp               = _mm256_add_pd(F,_mm256_mul_pd(vfeps,_mm256_add_pd(G,Heps)));
 563             VV               = _mm256_add_pd(Y,_mm256_mul_pd(vfeps,Fp));
 564             velec            = _mm256_mul_pd(qq31,VV);
 565             FF               = _mm256_add_pd(Fp,_mm256_mul_pd(vfeps,_mm256_add_pd(G,_mm256_add_pd(Heps,Heps))));
 566             felec            = _mm256_xor_pd(signbit,_mm256_mul_pd(_mm256_mul_pd(qq31,FF),_mm256_mul_pd(vftabscale,rinv31)));
 567
 568             /* Update potential sum for this i atom from the interaction with this j atom. */
 569             velecsum         = _mm256_add_pd(velecsum,velec);
 570
 571             fscal            = felec;
 572
 573             /* Calculate temporary vectorial force */
 574             tx               = _mm256_mul_pd(fscal,dx31);
 575             ty               = _mm256_mul_pd(fscal,dy31);
 576             tz               = _mm256_mul_pd(fscal,dz31);
 577
 578             /* Update vectorial force */
 579             fix3             = _mm256_add_pd(fix3,tx);
 580             fiy3             = _mm256_add_pd(fiy3,ty);
 581             fiz3             = _mm256_add_pd(fiz3,tz);
 582
 583             fjx1             = _mm256_add_pd(fjx1,tx);
 584             fjy1             = _mm256_add_pd(fjy1,ty);
 585             fjz1             = _mm256_add_pd(fjz1,tz);
 586
 587             /**************************
 588              * CALCULATE INTERACTIONS *
 589              **************************/
 590
 591             r32              = _mm256_mul_pd(rsq32,rinv32);
 592
 593             /* Calculate table index by multiplying r with table scale and truncate to integer */
 594             rt               = _mm256_mul_pd(r32,vftabscale);
 595             vfitab           = _mm256_cvttpd_epi32(rt);
 596             vfeps            = _mm256_sub_pd(rt,_mm256_round_pd(rt, _MM_FROUND_FLOOR));
 597             vfitab           = _mm_slli_epi32(vfitab,2);
 598
 599             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 600             Y                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
 601             F                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
 602             G                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,2) );
 603             H                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,3) );
 604             GMX_MM256_FULLTRANSPOSE4_PD(Y,F,G,H);
 605             Heps             = _mm256_mul_pd(vfeps,H);
 606             Fp               = _mm256_add_pd(F,_mm256_mul_pd(vfeps,_mm256_add_pd(G,Heps)));
 607             VV               = _mm256_add_pd(Y,_mm256_mul_pd(vfeps,Fp));
 608             velec            = _mm256_mul_pd(qq32,VV);
 609             FF               = _mm256_add_pd(Fp,_mm256_mul_pd(vfeps,_mm256_add_pd(G,_mm256_add_pd(Heps,Heps))));
 610             felec            = _mm256_xor_pd(signbit,_mm256_mul_pd(_mm256_mul_pd(qq32,FF),_mm256_mul_pd(vftabscale,rinv32)));
 611
 612             /* Update potential sum for this i atom from the interaction with this j atom. */
 613             velecsum         = _mm256_add_pd(velecsum,velec);
 614
 615             fscal            = felec;
 616
 617             /* Calculate temporary vectorial force */
 618             tx               = _mm256_mul_pd(fscal,dx32);
 619             ty               = _mm256_mul_pd(fscal,dy32);
 620             tz               = _mm256_mul_pd(fscal,dz32);
 621
 622             /* Update vectorial force */
 623             fix3             = _mm256_add_pd(fix3,tx);
 624             fiy3             = _mm256_add_pd(fiy3,ty);
 625             fiz3             = _mm256_add_pd(fiz3,tz);
 626
 627             fjx2             = _mm256_add_pd(fjx2,tx);
 628             fjy2             = _mm256_add_pd(fjy2,ty);
 629             fjz2             = _mm256_add_pd(fjz2,tz);
 630
 631             /**************************
 632              * CALCULATE INTERACTIONS *
 633              **************************/
 634
 635             r33              = _mm256_mul_pd(rsq33,rinv33);
 636
 637             /* Calculate table index by multiplying r with table scale and truncate to integer */
 638             rt               = _mm256_mul_pd(r33,vftabscale);
 639             vfitab           = _mm256_cvttpd_epi32(rt);
 640             vfeps            = _mm256_sub_pd(rt,_mm256_round_pd(rt, _MM_FROUND_FLOOR));
 641             vfitab           = _mm_slli_epi32(vfitab,2);
 642
 643             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 644             Y                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
 645             F                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
 646             G                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,2) );
 647             H                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,3) );
 648             GMX_MM256_FULLTRANSPOSE4_PD(Y,F,G,H);
 649             Heps             = _mm256_mul_pd(vfeps,H);
 650             Fp               = _mm256_add_pd(F,_mm256_mul_pd(vfeps,_mm256_add_pd(G,Heps)));
 651             VV               = _mm256_add_pd(Y,_mm256_mul_pd(vfeps,Fp));
 652             velec            = _mm256_mul_pd(qq33,VV);
 653             FF               = _mm256_add_pd(Fp,_mm256_mul_pd(vfeps,_mm256_add_pd(G,_mm256_add_pd(Heps,Heps))));
 654             felec            = _mm256_xor_pd(signbit,_mm256_mul_pd(_mm256_mul_pd(qq33,FF),_mm256_mul_pd(vftabscale,rinv33)));
 655
 656             /* Update potential sum for this i atom from the interaction with this j atom. */
 657             velecsum         = _mm256_add_pd(velecsum,velec);
 658
 659             fscal            = felec;
 660
 661             /* Calculate temporary vectorial force */
 662             tx               = _mm256_mul_pd(fscal,dx33);
 663             ty               = _mm256_mul_pd(fscal,dy33);
 664             tz               = _mm256_mul_pd(fscal,dz33);
 665
 666             /* Update vectorial force */
 667             fix3             = _mm256_add_pd(fix3,tx);
 668             fiy3             = _mm256_add_pd(fiy3,ty);
 669             fiz3             = _mm256_add_pd(fiz3,tz);
 670
 671             fjx3             = _mm256_add_pd(fjx3,tx);
 672             fjy3             = _mm256_add_pd(fjy3,ty);
 673             fjz3             = _mm256_add_pd(fjz3,tz);
 674
 675             fjptrA             = f+j_coord_offsetA;
 676             fjptrB             = f+j_coord_offsetB;
 677             fjptrC             = f+j_coord_offsetC;
 678             fjptrD             = f+j_coord_offsetD;
 679
 680             gmx_mm256_decrement_3rvec_4ptr_swizzle_pd(fjptrA+DIM,fjptrB+DIM,fjptrC+DIM,fjptrD+DIM,
 681                                                       fjx1,fjy1,fjz1,fjx2,fjy2,fjz2,fjx3,fjy3,fjz3);
 682
 683             /* Inner loop uses 387 flops */
 684         }
 685
 686         if(jidx<j_index_end)
 687         {
 688
 689             /* Get j neighbor index, and coordinate index */
 690             jnrlistA         = jjnr[jidx];
 691             jnrlistB         = jjnr[jidx+1];
 692             jnrlistC         = jjnr[jidx+2];
 693             jnrlistD         = jjnr[jidx+3];
 694             /* Sign of each element will be negative for non-real atoms.
 695              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 696              * so use it as val = _mm_andnot_pd(mask,val) to clear dummy entries.
 697              */
 698             tmpmask0 = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
 699
 700             tmpmask1 = _mm_permute_ps(tmpmask0,_GMX_MM_PERMUTE(3,3,2,2));
 701             tmpmask0 = _mm_permute_ps(tmpmask0,_GMX_MM_PERMUTE(1,1,0,0));
 702             dummy_mask = _mm256_castps_pd(gmx_mm256_set_m128(tmpmask1,tmpmask0));
 703
 704             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
 705             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
 706             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
 707             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
 708             j_coord_offsetA  = DIM*jnrA;
 709             j_coord_offsetB  = DIM*jnrB;
 710             j_coord_offsetC  = DIM*jnrC;
 711             j_coord_offsetD  = DIM*jnrD;
 712
 713             /* load j atom coordinates */
 714             gmx_mm256_load_3rvec_4ptr_swizzle_pd(x+j_coord_offsetA+DIM,x+j_coord_offsetB+DIM,
 715                                                  x+j_coord_offsetC+DIM,x+j_coord_offsetD+DIM,
 716                                                  &jx1,&jy1,&jz1,&jx2,&jy2,&jz2,&jx3,&jy3,&jz3);
 717
 718             /* Calculate displacement vector */
 719             dx11             = _mm256_sub_pd(ix1,jx1);
 720             dy11             = _mm256_sub_pd(iy1,jy1);
 721             dz11             = _mm256_sub_pd(iz1,jz1);
 722             dx12             = _mm256_sub_pd(ix1,jx2);
 723             dy12             = _mm256_sub_pd(iy1,jy2);
 724             dz12             = _mm256_sub_pd(iz1,jz2);
 725             dx13             = _mm256_sub_pd(ix1,jx3);
 726             dy13             = _mm256_sub_pd(iy1,jy3);
 727             dz13             = _mm256_sub_pd(iz1,jz3);
 728             dx21             = _mm256_sub_pd(ix2,jx1);
 729             dy21             = _mm256_sub_pd(iy2,jy1);
 730             dz21             = _mm256_sub_pd(iz2,jz1);
 731             dx22             = _mm256_sub_pd(ix2,jx2);
 732             dy22             = _mm256_sub_pd(iy2,jy2);
 733             dz22             = _mm256_sub_pd(iz2,jz2);
 734             dx23             = _mm256_sub_pd(ix2,jx3);
 735             dy23             = _mm256_sub_pd(iy2,jy3);
 736             dz23             = _mm256_sub_pd(iz2,jz3);
 737             dx31             = _mm256_sub_pd(ix3,jx1);
 738             dy31             = _mm256_sub_pd(iy3,jy1);
 739             dz31             = _mm256_sub_pd(iz3,jz1);
 740             dx32             = _mm256_sub_pd(ix3,jx2);
 741             dy32             = _mm256_sub_pd(iy3,jy2);
 742             dz32             = _mm256_sub_pd(iz3,jz2);
 743             dx33             = _mm256_sub_pd(ix3,jx3);
 744             dy33             = _mm256_sub_pd(iy3,jy3);
 745             dz33             = _mm256_sub_pd(iz3,jz3);
 746
 747             /* Calculate squared distance and things based on it */
 748             rsq11            = gmx_mm256_calc_rsq_pd(dx11,dy11,dz11);
 749             rsq12            = gmx_mm256_calc_rsq_pd(dx12,dy12,dz12);
 750             rsq13            = gmx_mm256_calc_rsq_pd(dx13,dy13,dz13);
 751             rsq21            = gmx_mm256_calc_rsq_pd(dx21,dy21,dz21);
 752             rsq22            = gmx_mm256_calc_rsq_pd(dx22,dy22,dz22);
 753             rsq23            = gmx_mm256_calc_rsq_pd(dx23,dy23,dz23);
 754             rsq31            = gmx_mm256_calc_rsq_pd(dx31,dy31,dz31);
 755             rsq32            = gmx_mm256_calc_rsq_pd(dx32,dy32,dz32);
 756             rsq33            = gmx_mm256_calc_rsq_pd(dx33,dy33,dz33);
 757
 758             rinv11           = gmx_mm256_invsqrt_pd(rsq11);
 759             rinv12           = gmx_mm256_invsqrt_pd(rsq12);
 760             rinv13           = gmx_mm256_invsqrt_pd(rsq13);
 761             rinv21           = gmx_mm256_invsqrt_pd(rsq21);
 762             rinv22           = gmx_mm256_invsqrt_pd(rsq22);
 763             rinv23           = gmx_mm256_invsqrt_pd(rsq23);
 764             rinv31           = gmx_mm256_invsqrt_pd(rsq31);
 765             rinv32           = gmx_mm256_invsqrt_pd(rsq32);
 766             rinv33           = gmx_mm256_invsqrt_pd(rsq33);
 767
 768             fjx1             = _mm256_setzero_pd();
 769             fjy1             = _mm256_setzero_pd();
 770             fjz1             = _mm256_setzero_pd();
 771             fjx2             = _mm256_setzero_pd();
 772             fjy2             = _mm256_setzero_pd();
 773             fjz2             = _mm256_setzero_pd();
 774             fjx3             = _mm256_setzero_pd();
 775             fjy3             = _mm256_setzero_pd();
 776             fjz3             = _mm256_setzero_pd();
 777
 778             /**************************
 779              * CALCULATE INTERACTIONS *
 780              **************************/
 781
 782             r11              = _mm256_mul_pd(rsq11,rinv11);
 783             r11              = _mm256_andnot_pd(dummy_mask,r11);
 784
 785             /* Calculate table index by multiplying r with table scale and truncate to integer */
 786             rt               = _mm256_mul_pd(r11,vftabscale);
 787             vfitab           = _mm256_cvttpd_epi32(rt);
 788             vfeps            = _mm256_sub_pd(rt,_mm256_round_pd(rt, _MM_FROUND_FLOOR));
 789             vfitab           = _mm_slli_epi32(vfitab,2);
 790
 791             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 792             Y                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
 793             F                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
 794             G                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,2) );
 795             H                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,3) );
 796             GMX_MM256_FULLTRANSPOSE4_PD(Y,F,G,H);
 797             Heps             = _mm256_mul_pd(vfeps,H);
 798             Fp               = _mm256_add_pd(F,_mm256_mul_pd(vfeps,_mm256_add_pd(G,Heps)));
 799             VV               = _mm256_add_pd(Y,_mm256_mul_pd(vfeps,Fp));
 800             velec            = _mm256_mul_pd(qq11,VV);
 801             FF               = _mm256_add_pd(Fp,_mm256_mul_pd(vfeps,_mm256_add_pd(G,_mm256_add_pd(Heps,Heps))));
 802             felec            = _mm256_xor_pd(signbit,_mm256_mul_pd(_mm256_mul_pd(qq11,FF),_mm256_mul_pd(vftabscale,rinv11)));
 803
 804             /* Update potential sum for this i atom from the interaction with this j atom. */
 805             velec            = _mm256_andnot_pd(dummy_mask,velec);
 806             velecsum         = _mm256_add_pd(velecsum,velec);
 807
 808             fscal            = felec;
 809
 810             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
 811
 812             /* Calculate temporary vectorial force */
 813             tx               = _mm256_mul_pd(fscal,dx11);
 814             ty               = _mm256_mul_pd(fscal,dy11);
 815             tz               = _mm256_mul_pd(fscal,dz11);
 816
 817             /* Update vectorial force */
 818             fix1             = _mm256_add_pd(fix1,tx);
 819             fiy1             = _mm256_add_pd(fiy1,ty);
 820             fiz1             = _mm256_add_pd(fiz1,tz);
 821
 822             fjx1             = _mm256_add_pd(fjx1,tx);
 823             fjy1             = _mm256_add_pd(fjy1,ty);
 824             fjz1             = _mm256_add_pd(fjz1,tz);
 825
 826             /**************************
 827              * CALCULATE INTERACTIONS *
 828              **************************/
 829
 830             r12              = _mm256_mul_pd(rsq12,rinv12);
 831             r12              = _mm256_andnot_pd(dummy_mask,r12);
 832
 833             /* Calculate table index by multiplying r with table scale and truncate to integer */
 834             rt               = _mm256_mul_pd(r12,vftabscale);
 835             vfitab           = _mm256_cvttpd_epi32(rt);
 836             vfeps            = _mm256_sub_pd(rt,_mm256_round_pd(rt, _MM_FROUND_FLOOR));
 837             vfitab           = _mm_slli_epi32(vfitab,2);
 838
 839             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 840             Y                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
 841             F                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
 842             G                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,2) );
 843             H                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,3) );
 844             GMX_MM256_FULLTRANSPOSE4_PD(Y,F,G,H);
 845             Heps             = _mm256_mul_pd(vfeps,H);
 846             Fp               = _mm256_add_pd(F,_mm256_mul_pd(vfeps,_mm256_add_pd(G,Heps)));
 847             VV               = _mm256_add_pd(Y,_mm256_mul_pd(vfeps,Fp));
 848             velec            = _mm256_mul_pd(qq12,VV);
 849             FF               = _mm256_add_pd(Fp,_mm256_mul_pd(vfeps,_mm256_add_pd(G,_mm256_add_pd(Heps,Heps))));
 850             felec            = _mm256_xor_pd(signbit,_mm256_mul_pd(_mm256_mul_pd(qq12,FF),_mm256_mul_pd(vftabscale,rinv12)));
 851
 852             /* Update potential sum for this i atom from the interaction with this j atom. */
 853             velec            = _mm256_andnot_pd(dummy_mask,velec);
 854             velecsum         = _mm256_add_pd(velecsum,velec);
 855
 856             fscal            = felec;
 857
 858             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
 859
 860             /* Calculate temporary vectorial force */
 861             tx               = _mm256_mul_pd(fscal,dx12);
 862             ty               = _mm256_mul_pd(fscal,dy12);
 863             tz               = _mm256_mul_pd(fscal,dz12);
 864
 865             /* Update vectorial force */
 866             fix1             = _mm256_add_pd(fix1,tx);
 867             fiy1             = _mm256_add_pd(fiy1,ty);
 868             fiz1             = _mm256_add_pd(fiz1,tz);
 869
 870             fjx2             = _mm256_add_pd(fjx2,tx);
 871             fjy2             = _mm256_add_pd(fjy2,ty);
 872             fjz2             = _mm256_add_pd(fjz2,tz);
 873
 874             /**************************
 875              * CALCULATE INTERACTIONS *
 876              **************************/
 877
 878             r13              = _mm256_mul_pd(rsq13,rinv13);
 879             r13              = _mm256_andnot_pd(dummy_mask,r13);
 880
 881             /* Calculate table index by multiplying r with table scale and truncate to integer */
 882             rt               = _mm256_mul_pd(r13,vftabscale);
 883             vfitab           = _mm256_cvttpd_epi32(rt);
 884             vfeps            = _mm256_sub_pd(rt,_mm256_round_pd(rt, _MM_FROUND_FLOOR));
 885             vfitab           = _mm_slli_epi32(vfitab,2);
 886
 887             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 888             Y                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
 889             F                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
 890             G                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,2) );
 891             H                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,3) );
 892             GMX_MM256_FULLTRANSPOSE4_PD(Y,F,G,H);
 893             Heps             = _mm256_mul_pd(vfeps,H);
 894             Fp               = _mm256_add_pd(F,_mm256_mul_pd(vfeps,_mm256_add_pd(G,Heps)));
 895             VV               = _mm256_add_pd(Y,_mm256_mul_pd(vfeps,Fp));
 896             velec            = _mm256_mul_pd(qq13,VV);
 897             FF               = _mm256_add_pd(Fp,_mm256_mul_pd(vfeps,_mm256_add_pd(G,_mm256_add_pd(Heps,Heps))));
 898             felec            = _mm256_xor_pd(signbit,_mm256_mul_pd(_mm256_mul_pd(qq13,FF),_mm256_mul_pd(vftabscale,rinv13)));
 899
 900             /* Update potential sum for this i atom from the interaction with this j atom. */
 901             velec            = _mm256_andnot_pd(dummy_mask,velec);
 902             velecsum         = _mm256_add_pd(velecsum,velec);
 903
 904             fscal            = felec;
 905
 906             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
 907
 908             /* Calculate temporary vectorial force */
 909             tx               = _mm256_mul_pd(fscal,dx13);
 910             ty               = _mm256_mul_pd(fscal,dy13);
 911             tz               = _mm256_mul_pd(fscal,dz13);
 912
 913             /* Update vectorial force */
 914             fix1             = _mm256_add_pd(fix1,tx);
 915             fiy1             = _mm256_add_pd(fiy1,ty);
 916             fiz1             = _mm256_add_pd(fiz1,tz);
 917
 918             fjx3             = _mm256_add_pd(fjx3,tx);
 919             fjy3             = _mm256_add_pd(fjy3,ty);
 920             fjz3             = _mm256_add_pd(fjz3,tz);
 921
 922             /**************************
 923              * CALCULATE INTERACTIONS *
 924              **************************/
 925
 926             r21              = _mm256_mul_pd(rsq21,rinv21);
 927             r21              = _mm256_andnot_pd(dummy_mask,r21);
 928
 929             /* Calculate table index by multiplying r with table scale and truncate to integer */
 930             rt               = _mm256_mul_pd(r21,vftabscale);
 931             vfitab           = _mm256_cvttpd_epi32(rt);
 932             vfeps            = _mm256_sub_pd(rt,_mm256_round_pd(rt, _MM_FROUND_FLOOR));
 933             vfitab           = _mm_slli_epi32(vfitab,2);
 934
 935             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 936             Y                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
 937             F                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
 938             G                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,2) );
 939             H                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,3) );
 940             GMX_MM256_FULLTRANSPOSE4_PD(Y,F,G,H);
 941             Heps             = _mm256_mul_pd(vfeps,H);
 942             Fp               = _mm256_add_pd(F,_mm256_mul_pd(vfeps,_mm256_add_pd(G,Heps)));
 943             VV               = _mm256_add_pd(Y,_mm256_mul_pd(vfeps,Fp));
 944             velec            = _mm256_mul_pd(qq21,VV);
 945             FF               = _mm256_add_pd(Fp,_mm256_mul_pd(vfeps,_mm256_add_pd(G,_mm256_add_pd(Heps,Heps))));
 946             felec            = _mm256_xor_pd(signbit,_mm256_mul_pd(_mm256_mul_pd(qq21,FF),_mm256_mul_pd(vftabscale,rinv21)));
 947
 948             /* Update potential sum for this i atom from the interaction with this j atom. */
 949             velec            = _mm256_andnot_pd(dummy_mask,velec);
 950             velecsum         = _mm256_add_pd(velecsum,velec);
 951
 952             fscal            = felec;
 953
 954             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
 955
 956             /* Calculate temporary vectorial force */
 957             tx               = _mm256_mul_pd(fscal,dx21);
 958             ty               = _mm256_mul_pd(fscal,dy21);
 959             tz               = _mm256_mul_pd(fscal,dz21);
 960
 961             /* Update vectorial force */
 962             fix2             = _mm256_add_pd(fix2,tx);
 963             fiy2             = _mm256_add_pd(fiy2,ty);
 964             fiz2             = _mm256_add_pd(fiz2,tz);
 965
 966             fjx1             = _mm256_add_pd(fjx1,tx);
 967             fjy1             = _mm256_add_pd(fjy1,ty);
 968             fjz1             = _mm256_add_pd(fjz1,tz);
 969
 970             /**************************
 971              * CALCULATE INTERACTIONS *
 972              **************************/
 973
 974             r22              = _mm256_mul_pd(rsq22,rinv22);
 975             r22              = _mm256_andnot_pd(dummy_mask,r22);
 976
 977             /* Calculate table index by multiplying r with table scale and truncate to integer */
 978             rt               = _mm256_mul_pd(r22,vftabscale);
 979             vfitab           = _mm256_cvttpd_epi32(rt);
 980             vfeps            = _mm256_sub_pd(rt,_mm256_round_pd(rt, _MM_FROUND_FLOOR));
 981             vfitab           = _mm_slli_epi32(vfitab,2);
 982
 983             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 984             Y                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
 985             F                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
 986             G                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,2) );
 987             H                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,3) );
 988             GMX_MM256_FULLTRANSPOSE4_PD(Y,F,G,H);
 989             Heps             = _mm256_mul_pd(vfeps,H);
 990             Fp               = _mm256_add_pd(F,_mm256_mul_pd(vfeps,_mm256_add_pd(G,Heps)));
 991             VV               = _mm256_add_pd(Y,_mm256_mul_pd(vfeps,Fp));
 992             velec            = _mm256_mul_pd(qq22,VV);
 993             FF               = _mm256_add_pd(Fp,_mm256_mul_pd(vfeps,_mm256_add_pd(G,_mm256_add_pd(Heps,Heps))));
 994             felec            = _mm256_xor_pd(signbit,_mm256_mul_pd(_mm256_mul_pd(qq22,FF),_mm256_mul_pd(vftabscale,rinv22)));
 995
 996             /* Update potential sum for this i atom from the interaction with this j atom. */
 997             velec            = _mm256_andnot_pd(dummy_mask,velec);
 998             velecsum         = _mm256_add_pd(velecsum,velec);
 999
1000             fscal            = felec;
1001
1002             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
1003
1004             /* Calculate temporary vectorial force */
1005             tx               = _mm256_mul_pd(fscal,dx22);
1006             ty               = _mm256_mul_pd(fscal,dy22);
1007             tz               = _mm256_mul_pd(fscal,dz22);
1008
1009             /* Update vectorial force */
1010             fix2             = _mm256_add_pd(fix2,tx);
1011             fiy2             = _mm256_add_pd(fiy2,ty);
1012             fiz2             = _mm256_add_pd(fiz2,tz);
1013
1014             fjx2             = _mm256_add_pd(fjx2,tx);
1015             fjy2             = _mm256_add_pd(fjy2,ty);
1016             fjz2             = _mm256_add_pd(fjz2,tz);
1017
1018             /**************************
1019              * CALCULATE INTERACTIONS *
1020              **************************/
1021
1022             r23              = _mm256_mul_pd(rsq23,rinv23);
1023             r23              = _mm256_andnot_pd(dummy_mask,r23);
1024
1025             /* Calculate table index by multiplying r with table scale and truncate to integer */
1026             rt               = _mm256_mul_pd(r23,vftabscale);
1027             vfitab           = _mm256_cvttpd_epi32(rt);
1028             vfeps            = _mm256_sub_pd(rt,_mm256_round_pd(rt, _MM_FROUND_FLOOR));
1029             vfitab           = _mm_slli_epi32(vfitab,2);
1030
1031             /* CUBIC SPLINE TABLE ELECTROSTATICS */
1032             Y                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
1033             F                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
1034             G                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,2) );
1035             H                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,3) );
1036             GMX_MM256_FULLTRANSPOSE4_PD(Y,F,G,H);
1037             Heps             = _mm256_mul_pd(vfeps,H);
1038             Fp               = _mm256_add_pd(F,_mm256_mul_pd(vfeps,_mm256_add_pd(G,Heps)));
1039             VV               = _mm256_add_pd(Y,_mm256_mul_pd(vfeps,Fp));
1040             velec            = _mm256_mul_pd(qq23,VV);
1041             FF               = _mm256_add_pd(Fp,_mm256_mul_pd(vfeps,_mm256_add_pd(G,_mm256_add_pd(Heps,Heps))));
1042             felec            = _mm256_xor_pd(signbit,_mm256_mul_pd(_mm256_mul_pd(qq23,FF),_mm256_mul_pd(vftabscale,rinv23)));
1043
1044             /* Update potential sum for this i atom from the interaction with this j atom. */
1045             velec            = _mm256_andnot_pd(dummy_mask,velec);
1046             velecsum         = _mm256_add_pd(velecsum,velec);
1047
1048             fscal            = felec;
1049
1050             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
1051
1052             /* Calculate temporary vectorial force */
1053             tx               = _mm256_mul_pd(fscal,dx23);
1054             ty               = _mm256_mul_pd(fscal,dy23);
1055             tz               = _mm256_mul_pd(fscal,dz23);
1056
1057             /* Update vectorial force */
1058             fix2             = _mm256_add_pd(fix2,tx);
1059             fiy2             = _mm256_add_pd(fiy2,ty);
1060             fiz2             = _mm256_add_pd(fiz2,tz);
1061
1062             fjx3             = _mm256_add_pd(fjx3,tx);
1063             fjy3             = _mm256_add_pd(fjy3,ty);
1064             fjz3             = _mm256_add_pd(fjz3,tz);
1065
1066             /**************************
1067              * CALCULATE INTERACTIONS *
1068              **************************/
1069
1070             r31              = _mm256_mul_pd(rsq31,rinv31);
1071             r31              = _mm256_andnot_pd(dummy_mask,r31);
1072
1073             /* Calculate table index by multiplying r with table scale and truncate to integer */
1074             rt               = _mm256_mul_pd(r31,vftabscale);
1075             vfitab           = _mm256_cvttpd_epi32(rt);
1076             vfeps            = _mm256_sub_pd(rt,_mm256_round_pd(rt, _MM_FROUND_FLOOR));
1077             vfitab           = _mm_slli_epi32(vfitab,2);
1078
1079             /* CUBIC SPLINE TABLE ELECTROSTATICS */
1080             Y                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
1081             F                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
1082             G                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,2) );
1083             H                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,3) );
1084             GMX_MM256_FULLTRANSPOSE4_PD(Y,F,G,H);
1085             Heps             = _mm256_mul_pd(vfeps,H);
1086             Fp               = _mm256_add_pd(F,_mm256_mul_pd(vfeps,_mm256_add_pd(G,Heps)));
1087             VV               = _mm256_add_pd(Y,_mm256_mul_pd(vfeps,Fp));
1088             velec            = _mm256_mul_pd(qq31,VV);
1089             FF               = _mm256_add_pd(Fp,_mm256_mul_pd(vfeps,_mm256_add_pd(G,_mm256_add_pd(Heps,Heps))));
1090             felec            = _mm256_xor_pd(signbit,_mm256_mul_pd(_mm256_mul_pd(qq31,FF),_mm256_mul_pd(vftabscale,rinv31)));
1091
1092             /* Update potential sum for this i atom from the interaction with this j atom. */
1093             velec            = _mm256_andnot_pd(dummy_mask,velec);
1094             velecsum         = _mm256_add_pd(velecsum,velec);
1095
1096             fscal            = felec;
1097
1098             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
1099
1100             /* Calculate temporary vectorial force */
1101             tx               = _mm256_mul_pd(fscal,dx31);
1102             ty               = _mm256_mul_pd(fscal,dy31);
1103             tz               = _mm256_mul_pd(fscal,dz31);
1104
1105             /* Update vectorial force */
1106             fix3             = _mm256_add_pd(fix3,tx);
1107             fiy3             = _mm256_add_pd(fiy3,ty);
1108             fiz3             = _mm256_add_pd(fiz3,tz);
1109
1110             fjx1             = _mm256_add_pd(fjx1,tx);
1111             fjy1             = _mm256_add_pd(fjy1,ty);
1112             fjz1             = _mm256_add_pd(fjz1,tz);
1113
1114             /**************************
1115              * CALCULATE INTERACTIONS *
1116              **************************/
1117
1118             r32              = _mm256_mul_pd(rsq32,rinv32);
1119             r32              = _mm256_andnot_pd(dummy_mask,r32);
1120
1121             /* Calculate table index by multiplying r with table scale and truncate to integer */
1122             rt               = _mm256_mul_pd(r32,vftabscale);
1123             vfitab           = _mm256_cvttpd_epi32(rt);
1124             vfeps            = _mm256_sub_pd(rt,_mm256_round_pd(rt, _MM_FROUND_FLOOR));
1125             vfitab           = _mm_slli_epi32(vfitab,2);
1126
1127             /* CUBIC SPLINE TABLE ELECTROSTATICS */
1128             Y                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
1129             F                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
1130             G                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,2) );
1131             H                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,3) );
1132             GMX_MM256_FULLTRANSPOSE4_PD(Y,F,G,H);
1133             Heps             = _mm256_mul_pd(vfeps,H);
1134             Fp               = _mm256_add_pd(F,_mm256_mul_pd(vfeps,_mm256_add_pd(G,Heps)));
1135             VV               = _mm256_add_pd(Y,_mm256_mul_pd(vfeps,Fp));
1136             velec            = _mm256_mul_pd(qq32,VV);
1137             FF               = _mm256_add_pd(Fp,_mm256_mul_pd(vfeps,_mm256_add_pd(G,_mm256_add_pd(Heps,Heps))));
1138             felec            = _mm256_xor_pd(signbit,_mm256_mul_pd(_mm256_mul_pd(qq32,FF),_mm256_mul_pd(vftabscale,rinv32)));
1139
1140             /* Update potential sum for this i atom from the interaction with this j atom. */
1141             velec            = _mm256_andnot_pd(dummy_mask,velec);
1142             velecsum         = _mm256_add_pd(velecsum,velec);
1143
1144             fscal            = felec;
1145
1146             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
1147
1148             /* Calculate temporary vectorial force */
1149             tx               = _mm256_mul_pd(fscal,dx32);
1150             ty               = _mm256_mul_pd(fscal,dy32);
1151             tz               = _mm256_mul_pd(fscal,dz32);
1152
1153             /* Update vectorial force */
1154             fix3             = _mm256_add_pd(fix3,tx);
1155             fiy3             = _mm256_add_pd(fiy3,ty);
1156             fiz3             = _mm256_add_pd(fiz3,tz);
1157
1158             fjx2             = _mm256_add_pd(fjx2,tx);
1159             fjy2             = _mm256_add_pd(fjy2,ty);
1160             fjz2             = _mm256_add_pd(fjz2,tz);
1161
1162             /**************************
1163              * CALCULATE INTERACTIONS *
1164              **************************/
1165
1166             r33              = _mm256_mul_pd(rsq33,rinv33);
1167             r33              = _mm256_andnot_pd(dummy_mask,r33);
1168
1169             /* Calculate table index by multiplying r with table scale and truncate to integer */
1170             rt               = _mm256_mul_pd(r33,vftabscale);
1171             vfitab           = _mm256_cvttpd_epi32(rt);
1172             vfeps            = _mm256_sub_pd(rt,_mm256_round_pd(rt, _MM_FROUND_FLOOR));
1173             vfitab           = _mm_slli_epi32(vfitab,2);
1174
1175             /* CUBIC SPLINE TABLE ELECTROSTATICS */
1176             Y                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
1177             F                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
1178             G                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,2) );
1179             H                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,3) );
1180             GMX_MM256_FULLTRANSPOSE4_PD(Y,F,G,H);
1181             Heps             = _mm256_mul_pd(vfeps,H);
1182             Fp               = _mm256_add_pd(F,_mm256_mul_pd(vfeps,_mm256_add_pd(G,Heps)));
1183             VV               = _mm256_add_pd(Y,_mm256_mul_pd(vfeps,Fp));
1184             velec            = _mm256_mul_pd(qq33,VV);
1185             FF               = _mm256_add_pd(Fp,_mm256_mul_pd(vfeps,_mm256_add_pd(G,_mm256_add_pd(Heps,Heps))));
1186             felec            = _mm256_xor_pd(signbit,_mm256_mul_pd(_mm256_mul_pd(qq33,FF),_mm256_mul_pd(vftabscale,rinv33)));
1187
1188             /* Update potential sum for this i atom from the interaction with this j atom. */
1189             velec            = _mm256_andnot_pd(dummy_mask,velec);
1190             velecsum         = _mm256_add_pd(velecsum,velec);
1191
1192             fscal            = felec;
1193
1194             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
1195
1196             /* Calculate temporary vectorial force */
1197             tx               = _mm256_mul_pd(fscal,dx33);
1198             ty               = _mm256_mul_pd(fscal,dy33);
1199             tz               = _mm256_mul_pd(fscal,dz33);
1200
1201             /* Update vectorial force */
1202             fix3             = _mm256_add_pd(fix3,tx);
1203             fiy3             = _mm256_add_pd(fiy3,ty);
1204             fiz3             = _mm256_add_pd(fiz3,tz);
1205
1206             fjx3             = _mm256_add_pd(fjx3,tx);
1207             fjy3             = _mm256_add_pd(fjy3,ty);
1208             fjz3             = _mm256_add_pd(fjz3,tz);
1209
1210             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
1211             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
1212             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
1213             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
1214
1215             gmx_mm256_decrement_3rvec_4ptr_swizzle_pd(fjptrA+DIM,fjptrB+DIM,fjptrC+DIM,fjptrD+DIM,
1216                                                       fjx1,fjy1,fjz1,fjx2,fjy2,fjz2,fjx3,fjy3,fjz3);
1217
1218             /* Inner loop uses 396 flops */
1219         }
1220
1221         /* End of innermost loop */
1222
1223         gmx_mm256_update_iforce_3atom_swizzle_pd(fix1,fiy1,fiz1,fix2,fiy2,fiz2,fix3,fiy3,fiz3,
1224                                                  f+i_coord_offset+DIM,fshift+i_shift_offset);
1225
1226         ggid                        = gid[iidx];
1227         /* Update potential energies */
1228         gmx_mm256_update_1pot_pd(velecsum,kernel_data->energygrp_elec+ggid);
1229
1230         /* Increment number of inner iterations */
1231         inneriter                  += j_index_end - j_index_start;
1232
1233         /* Outer loop uses 19 flops */
1234     }
1235
1236     /* Increment number of outer iterations */
1237     outeriter        += nri;
1238
1239     /* Update outer/inner flops */
1240
1241     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_W4W4_VF,outeriter*19 + inneriter*396);
1242 }
1243 /*
1244  * Gromacs nonbonded kernel:   nb_kernel_ElecCSTab_VdwNone_GeomW4W4_F_avx_256_double
1245  * Electrostatics interaction: CubicSplineTable
1246  * VdW interaction:            None
1247  * Geometry:                   Water4-Water4
1248  * Calculate force/pot:        Force
1249  */
1250 void
1251 nb_kernel_ElecCSTab_VdwNone_GeomW4W4_F_avx_256_double
1252                     (t_nblist                    * gmx_restrict       nlist,
1253                      rvec                        * gmx_restrict          xx,
1254                      rvec                        * gmx_restrict          ff,
1255                      t_forcerec                  * gmx_restrict          fr,
1256                      t_mdatoms                   * gmx_restrict     mdatoms,
1257                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
1258                      t_nrnb                      * gmx_restrict        nrnb)
1259 {
1260     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
1261      * just 0 for non-waters.
1262      * Suffixes A,B,C,D refer to j loop unrolling done with AVX, e.g. for the four different
1263      * jnr indices corresponding to data put in the four positions in the SIMD register.
1264      */
1265     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
1266     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
1267     int              jnrA,jnrB,jnrC,jnrD;
1268     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
1269     int              jnrlistE,jnrlistF,jnrlistG,jnrlistH;
1270     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
1271     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
1272     real             rcutoff_scalar;
1273     real             *shiftvec,*fshift,*x,*f;
1274     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD;
1275     real             scratch[4*DIM];
1276     __m256d          tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
1277     real *           vdwioffsetptr1;
1278     __m256d          ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
1279     real *           vdwioffsetptr2;
1280     __m256d          ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
1281     real *           vdwioffsetptr3;
1282     __m256d          ix3,iy3,iz3,fix3,fiy3,fiz3,iq3,isai3;
1283     int              vdwjidx1A,vdwjidx1B,vdwjidx1C,vdwjidx1D;
1284     __m256d          jx1,jy1,jz1,fjx1,fjy1,fjz1,jq1,isaj1;
1285     int              vdwjidx2A,vdwjidx2B,vdwjidx2C,vdwjidx2D;
1286     __m256d          jx2,jy2,jz2,fjx2,fjy2,fjz2,jq2,isaj2;
1287     int              vdwjidx3A,vdwjidx3B,vdwjidx3C,vdwjidx3D;
1288     __m256d          jx3,jy3,jz3,fjx3,fjy3,fjz3,jq3,isaj3;
1289     __m256d          dx11,dy11,dz11,rsq11,rinv11,rinvsq11,r11,qq11,c6_11,c12_11;
1290     __m256d          dx12,dy12,dz12,rsq12,rinv12,rinvsq12,r12,qq12,c6_12,c12_12;
1291     __m256d          dx13,dy13,dz13,rsq13,rinv13,rinvsq13,r13,qq13,c6_13,c12_13;
1292     __m256d          dx21,dy21,dz21,rsq21,rinv21,rinvsq21,r21,qq21,c6_21,c12_21;
1293     __m256d          dx22,dy22,dz22,rsq22,rinv22,rinvsq22,r22,qq22,c6_22,c12_22;
1294     __m256d          dx23,dy23,dz23,rsq23,rinv23,rinvsq23,r23,qq23,c6_23,c12_23;
1295     __m256d          dx31,dy31,dz31,rsq31,rinv31,rinvsq31,r31,qq31,c6_31,c12_31;
1296     __m256d          dx32,dy32,dz32,rsq32,rinv32,rinvsq32,r32,qq32,c6_32,c12_32;
1297     __m256d          dx33,dy33,dz33,rsq33,rinv33,rinvsq33,r33,qq33,c6_33,c12_33;
1298     __m256d          velec,felec,velecsum,facel,crf,krf,krf2;
1299     real             *charge;
1300     __m128i          vfitab;
1301     __m128i          ifour       = _mm_set1_epi32(4);
1302     __m256d          rt,vfeps,vftabscale,Y,F,G,H,Heps,Fp,VV,FF;
1303     real             *vftab;
1304     __m256d          dummy_mask,cutoff_mask;
1305     __m128           tmpmask0,tmpmask1;
1306     __m256d          signbit = _mm256_castsi256_pd( _mm256_set1_epi32(0x80000000) );
1307     __m256d          one     = _mm256_set1_pd(1.0);
1308     __m256d          two     = _mm256_set1_pd(2.0);
1309     x                = xx[0];
1310     f                = ff[0];
1311
1312     nri              = nlist->nri;
1313     iinr             = nlist->iinr;
1314     jindex           = nlist->jindex;
1315     jjnr             = nlist->jjnr;
1316     shiftidx         = nlist->shift;
1317     gid              = nlist->gid;
1318     shiftvec         = fr->shift_vec[0];
1319     fshift           = fr->fshift[0];
1320     facel            = _mm256_set1_pd(fr->epsfac);
1321     charge           = mdatoms->chargeA;
1322
1323     vftab            = kernel_data->table_elec->data;
1324     vftabscale       = _mm256_set1_pd(kernel_data->table_elec->scale);
1325
1326     /* Setup water-specific parameters */
1327     inr              = nlist->iinr[0];
1328     iq1              = _mm256_mul_pd(facel,_mm256_set1_pd(charge[inr+1]));
1329     iq2              = _mm256_mul_pd(facel,_mm256_set1_pd(charge[inr+2]));
1330     iq3              = _mm256_mul_pd(facel,_mm256_set1_pd(charge[inr+3]));
1331
1332     jq1              = _mm256_set1_pd(charge[inr+1]);
1333     jq2              = _mm256_set1_pd(charge[inr+2]);
1334     jq3              = _mm256_set1_pd(charge[inr+3]);
1335     qq11             = _mm256_mul_pd(iq1,jq1);
1336     qq12             = _mm256_mul_pd(iq1,jq2);
1337     qq13             = _mm256_mul_pd(iq1,jq3);
1338     qq21             = _mm256_mul_pd(iq2,jq1);
1339     qq22             = _mm256_mul_pd(iq2,jq2);
1340     qq23             = _mm256_mul_pd(iq2,jq3);
1341     qq31             = _mm256_mul_pd(iq3,jq1);
1342     qq32             = _mm256_mul_pd(iq3,jq2);
1343     qq33             = _mm256_mul_pd(iq3,jq3);
1344
1345     /* Avoid stupid compiler warnings */
1346     jnrA = jnrB = jnrC = jnrD = 0;
1347     j_coord_offsetA = 0;
1348     j_coord_offsetB = 0;
1349     j_coord_offsetC = 0;
1350     j_coord_offsetD = 0;
1351
1352     outeriter        = 0;
1353     inneriter        = 0;
1354
1355     for(iidx=0;iidx<4*DIM;iidx++)
1356     {
1357         scratch[iidx] = 0.0;
1358     }
1359
1360     /* Start outer loop over neighborlists */
1361     for(iidx=0; iidx<nri; iidx++)
1362     {
1363         /* Load shift vector for this list */
1364         i_shift_offset   = DIM*shiftidx[iidx];
1365
1366         /* Load limits for loop over neighbors */
1367         j_index_start    = jindex[iidx];
1368         j_index_end      = jindex[iidx+1];
1369
1370         /* Get outer coordinate index */
1371         inr              = iinr[iidx];
1372         i_coord_offset   = DIM*inr;
1373
1374         /* Load i particle coords and add shift vector */
1375         gmx_mm256_load_shift_and_3rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset+DIM,
1376                                                     &ix1,&iy1,&iz1,&ix2,&iy2,&iz2,&ix3,&iy3,&iz3);
1377
1378         fix1             = _mm256_setzero_pd();
1379         fiy1             = _mm256_setzero_pd();
1380         fiz1             = _mm256_setzero_pd();
1381         fix2             = _mm256_setzero_pd();
1382         fiy2             = _mm256_setzero_pd();
1383         fiz2             = _mm256_setzero_pd();
1384         fix3             = _mm256_setzero_pd();
1385         fiy3             = _mm256_setzero_pd();
1386         fiz3             = _mm256_setzero_pd();
1387
1388         /* Start inner kernel loop */
1389         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
1390         {
1391
1392             /* Get j neighbor index, and coordinate index */
1393             jnrA             = jjnr[jidx];
1394             jnrB             = jjnr[jidx+1];
1395             jnrC             = jjnr[jidx+2];
1396             jnrD             = jjnr[jidx+3];
1397             j_coord_offsetA  = DIM*jnrA;
1398             j_coord_offsetB  = DIM*jnrB;
1399             j_coord_offsetC  = DIM*jnrC;
1400             j_coord_offsetD  = DIM*jnrD;
1401
1402             /* load j atom coordinates */
1403             gmx_mm256_load_3rvec_4ptr_swizzle_pd(x+j_coord_offsetA+DIM,x+j_coord_offsetB+DIM,
1404                                                  x+j_coord_offsetC+DIM,x+j_coord_offsetD+DIM,
1405                                                  &jx1,&jy1,&jz1,&jx2,&jy2,&jz2,&jx3,&jy3,&jz3);
1406
1407             /* Calculate displacement vector */
1408             dx11             = _mm256_sub_pd(ix1,jx1);
1409             dy11             = _mm256_sub_pd(iy1,jy1);
1410             dz11             = _mm256_sub_pd(iz1,jz1);
1411             dx12             = _mm256_sub_pd(ix1,jx2);
1412             dy12             = _mm256_sub_pd(iy1,jy2);
1413             dz12             = _mm256_sub_pd(iz1,jz2);
1414             dx13             = _mm256_sub_pd(ix1,jx3);
1415             dy13             = _mm256_sub_pd(iy1,jy3);
1416             dz13             = _mm256_sub_pd(iz1,jz3);
1417             dx21             = _mm256_sub_pd(ix2,jx1);
1418             dy21             = _mm256_sub_pd(iy2,jy1);
1419             dz21             = _mm256_sub_pd(iz2,jz1);
1420             dx22             = _mm256_sub_pd(ix2,jx2);
1421             dy22             = _mm256_sub_pd(iy2,jy2);
1422             dz22             = _mm256_sub_pd(iz2,jz2);
1423             dx23             = _mm256_sub_pd(ix2,jx3);
1424             dy23             = _mm256_sub_pd(iy2,jy3);
1425             dz23             = _mm256_sub_pd(iz2,jz3);
1426             dx31             = _mm256_sub_pd(ix3,jx1);
1427             dy31             = _mm256_sub_pd(iy3,jy1);
1428             dz31             = _mm256_sub_pd(iz3,jz1);
1429             dx32             = _mm256_sub_pd(ix3,jx2);
1430             dy32             = _mm256_sub_pd(iy3,jy2);
1431             dz32             = _mm256_sub_pd(iz3,jz2);
1432             dx33             = _mm256_sub_pd(ix3,jx3);
1433             dy33             = _mm256_sub_pd(iy3,jy3);
1434             dz33             = _mm256_sub_pd(iz3,jz3);
1435
1436             /* Calculate squared distance and things based on it */
1437             rsq11            = gmx_mm256_calc_rsq_pd(dx11,dy11,dz11);
1438             rsq12            = gmx_mm256_calc_rsq_pd(dx12,dy12,dz12);
1439             rsq13            = gmx_mm256_calc_rsq_pd(dx13,dy13,dz13);
1440             rsq21            = gmx_mm256_calc_rsq_pd(dx21,dy21,dz21);
1441             rsq22            = gmx_mm256_calc_rsq_pd(dx22,dy22,dz22);
1442             rsq23            = gmx_mm256_calc_rsq_pd(dx23,dy23,dz23);
1443             rsq31            = gmx_mm256_calc_rsq_pd(dx31,dy31,dz31);
1444             rsq32            = gmx_mm256_calc_rsq_pd(dx32,dy32,dz32);
1445             rsq33            = gmx_mm256_calc_rsq_pd(dx33,dy33,dz33);
1446
1447             rinv11           = gmx_mm256_invsqrt_pd(rsq11);
1448             rinv12           = gmx_mm256_invsqrt_pd(rsq12);
1449             rinv13           = gmx_mm256_invsqrt_pd(rsq13);
1450             rinv21           = gmx_mm256_invsqrt_pd(rsq21);
1451             rinv22           = gmx_mm256_invsqrt_pd(rsq22);
1452             rinv23           = gmx_mm256_invsqrt_pd(rsq23);
1453             rinv31           = gmx_mm256_invsqrt_pd(rsq31);
1454             rinv32           = gmx_mm256_invsqrt_pd(rsq32);
1455             rinv33           = gmx_mm256_invsqrt_pd(rsq33);
1456
1457             fjx1             = _mm256_setzero_pd();
1458             fjy1             = _mm256_setzero_pd();
1459             fjz1             = _mm256_setzero_pd();
1460             fjx2             = _mm256_setzero_pd();
1461             fjy2             = _mm256_setzero_pd();
1462             fjz2             = _mm256_setzero_pd();
1463             fjx3             = _mm256_setzero_pd();
1464             fjy3             = _mm256_setzero_pd();
1465             fjz3             = _mm256_setzero_pd();
1466
1467             /**************************
1468              * CALCULATE INTERACTIONS *
1469              **************************/
1470
1471             r11              = _mm256_mul_pd(rsq11,rinv11);
1472
1473             /* Calculate table index by multiplying r with table scale and truncate to integer */
1474             rt               = _mm256_mul_pd(r11,vftabscale);
1475             vfitab           = _mm256_cvttpd_epi32(rt);
1476             vfeps            = _mm256_sub_pd(rt,_mm256_round_pd(rt, _MM_FROUND_FLOOR));
1477             vfitab           = _mm_slli_epi32(vfitab,2);
1478
1479             /* CUBIC SPLINE TABLE ELECTROSTATICS */
1480             Y                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
1481             F                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
1482             G                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,2) );
1483             H                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,3) );
1484             GMX_MM256_FULLTRANSPOSE4_PD(Y,F,G,H);
1485             Heps             = _mm256_mul_pd(vfeps,H);
1486             Fp               = _mm256_add_pd(F,_mm256_mul_pd(vfeps,_mm256_add_pd(G,Heps)));
1487             FF               = _mm256_add_pd(Fp,_mm256_mul_pd(vfeps,_mm256_add_pd(G,_mm256_add_pd(Heps,Heps))));
1488             felec            = _mm256_xor_pd(signbit,_mm256_mul_pd(_mm256_mul_pd(qq11,FF),_mm256_mul_pd(vftabscale,rinv11)));
1489
1490             fscal            = felec;
1491
1492             /* Calculate temporary vectorial force */
1493             tx               = _mm256_mul_pd(fscal,dx11);
1494             ty               = _mm256_mul_pd(fscal,dy11);
1495             tz               = _mm256_mul_pd(fscal,dz11);
1496
1497             /* Update vectorial force */
1498             fix1             = _mm256_add_pd(fix1,tx);
1499             fiy1             = _mm256_add_pd(fiy1,ty);
1500             fiz1             = _mm256_add_pd(fiz1,tz);
1501
1502             fjx1             = _mm256_add_pd(fjx1,tx);
1503             fjy1             = _mm256_add_pd(fjy1,ty);
1504             fjz1             = _mm256_add_pd(fjz1,tz);
1505
1506             /**************************
1507              * CALCULATE INTERACTIONS *
1508              **************************/
1509
1510             r12              = _mm256_mul_pd(rsq12,rinv12);
1511
1512             /* Calculate table index by multiplying r with table scale and truncate to integer */
1513             rt               = _mm256_mul_pd(r12,vftabscale);
1514             vfitab           = _mm256_cvttpd_epi32(rt);
1515             vfeps            = _mm256_sub_pd(rt,_mm256_round_pd(rt, _MM_FROUND_FLOOR));
1516             vfitab           = _mm_slli_epi32(vfitab,2);
1517
1518             /* CUBIC SPLINE TABLE ELECTROSTATICS */
1519             Y                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
1520             F                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
1521             G                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,2) );
1522             H                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,3) );
1523             GMX_MM256_FULLTRANSPOSE4_PD(Y,F,G,H);
1524             Heps             = _mm256_mul_pd(vfeps,H);
1525             Fp               = _mm256_add_pd(F,_mm256_mul_pd(vfeps,_mm256_add_pd(G,Heps)));
1526             FF               = _mm256_add_pd(Fp,_mm256_mul_pd(vfeps,_mm256_add_pd(G,_mm256_add_pd(Heps,Heps))));
1527             felec            = _mm256_xor_pd(signbit,_mm256_mul_pd(_mm256_mul_pd(qq12,FF),_mm256_mul_pd(vftabscale,rinv12)));
1528
1529             fscal            = felec;
1530
1531             /* Calculate temporary vectorial force */
1532             tx               = _mm256_mul_pd(fscal,dx12);
1533             ty               = _mm256_mul_pd(fscal,dy12);
1534             tz               = _mm256_mul_pd(fscal,dz12);
1535
1536             /* Update vectorial force */
1537             fix1             = _mm256_add_pd(fix1,tx);
1538             fiy1             = _mm256_add_pd(fiy1,ty);
1539             fiz1             = _mm256_add_pd(fiz1,tz);
1540
1541             fjx2             = _mm256_add_pd(fjx2,tx);
1542             fjy2             = _mm256_add_pd(fjy2,ty);
1543             fjz2             = _mm256_add_pd(fjz2,tz);
1544
1545             /**************************
1546              * CALCULATE INTERACTIONS *
1547              **************************/
1548
1549             r13              = _mm256_mul_pd(rsq13,rinv13);
1550
1551             /* Calculate table index by multiplying r with table scale and truncate to integer */
1552             rt               = _mm256_mul_pd(r13,vftabscale);
1553             vfitab           = _mm256_cvttpd_epi32(rt);
1554             vfeps            = _mm256_sub_pd(rt,_mm256_round_pd(rt, _MM_FROUND_FLOOR));
1555             vfitab           = _mm_slli_epi32(vfitab,2);
1556
1557             /* CUBIC SPLINE TABLE ELECTROSTATICS */
1558             Y                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
1559             F                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
1560             G                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,2) );
1561             H                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,3) );
1562             GMX_MM256_FULLTRANSPOSE4_PD(Y,F,G,H);
1563             Heps             = _mm256_mul_pd(vfeps,H);
1564             Fp               = _mm256_add_pd(F,_mm256_mul_pd(vfeps,_mm256_add_pd(G,Heps)));
1565             FF               = _mm256_add_pd(Fp,_mm256_mul_pd(vfeps,_mm256_add_pd(G,_mm256_add_pd(Heps,Heps))));
1566             felec            = _mm256_xor_pd(signbit,_mm256_mul_pd(_mm256_mul_pd(qq13,FF),_mm256_mul_pd(vftabscale,rinv13)));
1567
1568             fscal            = felec;
1569
1570             /* Calculate temporary vectorial force */
1571             tx               = _mm256_mul_pd(fscal,dx13);
1572             ty               = _mm256_mul_pd(fscal,dy13);
1573             tz               = _mm256_mul_pd(fscal,dz13);
1574
1575             /* Update vectorial force */
1576             fix1             = _mm256_add_pd(fix1,tx);
1577             fiy1             = _mm256_add_pd(fiy1,ty);
1578             fiz1             = _mm256_add_pd(fiz1,tz);
1579
1580             fjx3             = _mm256_add_pd(fjx3,tx);
1581             fjy3             = _mm256_add_pd(fjy3,ty);
1582             fjz3             = _mm256_add_pd(fjz3,tz);
1583
1584             /**************************
1585              * CALCULATE INTERACTIONS *
1586              **************************/
1587
1588             r21              = _mm256_mul_pd(rsq21,rinv21);
1589
1590             /* Calculate table index by multiplying r with table scale and truncate to integer */
1591             rt               = _mm256_mul_pd(r21,vftabscale);
1592             vfitab           = _mm256_cvttpd_epi32(rt);
1593             vfeps            = _mm256_sub_pd(rt,_mm256_round_pd(rt, _MM_FROUND_FLOOR));
1594             vfitab           = _mm_slli_epi32(vfitab,2);
1595
1596             /* CUBIC SPLINE TABLE ELECTROSTATICS */
1597             Y                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
1598             F                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
1599             G                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,2) );
1600             H                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,3) );
1601             GMX_MM256_FULLTRANSPOSE4_PD(Y,F,G,H);
1602             Heps             = _mm256_mul_pd(vfeps,H);
1603             Fp               = _mm256_add_pd(F,_mm256_mul_pd(vfeps,_mm256_add_pd(G,Heps)));
1604             FF               = _mm256_add_pd(Fp,_mm256_mul_pd(vfeps,_mm256_add_pd(G,_mm256_add_pd(Heps,Heps))));
1605             felec            = _mm256_xor_pd(signbit,_mm256_mul_pd(_mm256_mul_pd(qq21,FF),_mm256_mul_pd(vftabscale,rinv21)));
1606
1607             fscal            = felec;
1608
1609             /* Calculate temporary vectorial force */
1610             tx               = _mm256_mul_pd(fscal,dx21);
1611             ty               = _mm256_mul_pd(fscal,dy21);
1612             tz               = _mm256_mul_pd(fscal,dz21);
1613
1614             /* Update vectorial force */
1615             fix2             = _mm256_add_pd(fix2,tx);
1616             fiy2             = _mm256_add_pd(fiy2,ty);
1617             fiz2             = _mm256_add_pd(fiz2,tz);
1618
1619             fjx1             = _mm256_add_pd(fjx1,tx);
1620             fjy1             = _mm256_add_pd(fjy1,ty);
1621             fjz1             = _mm256_add_pd(fjz1,tz);
1622
1623             /**************************
1624              * CALCULATE INTERACTIONS *
1625              **************************/
1626
1627             r22              = _mm256_mul_pd(rsq22,rinv22);
1628
1629             /* Calculate table index by multiplying r with table scale and truncate to integer */
1630             rt               = _mm256_mul_pd(r22,vftabscale);
1631             vfitab           = _mm256_cvttpd_epi32(rt);
1632             vfeps            = _mm256_sub_pd(rt,_mm256_round_pd(rt, _MM_FROUND_FLOOR));
1633             vfitab           = _mm_slli_epi32(vfitab,2);
1634
1635             /* CUBIC SPLINE TABLE ELECTROSTATICS */
1636             Y                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
1637             F                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
1638             G                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,2) );
1639             H                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,3) );
1640             GMX_MM256_FULLTRANSPOSE4_PD(Y,F,G,H);
1641             Heps             = _mm256_mul_pd(vfeps,H);
1642             Fp               = _mm256_add_pd(F,_mm256_mul_pd(vfeps,_mm256_add_pd(G,Heps)));
1643             FF               = _mm256_add_pd(Fp,_mm256_mul_pd(vfeps,_mm256_add_pd(G,_mm256_add_pd(Heps,Heps))));
1644             felec            = _mm256_xor_pd(signbit,_mm256_mul_pd(_mm256_mul_pd(qq22,FF),_mm256_mul_pd(vftabscale,rinv22)));
1645
1646             fscal            = felec;
1647
1648             /* Calculate temporary vectorial force */
1649             tx               = _mm256_mul_pd(fscal,dx22);
1650             ty               = _mm256_mul_pd(fscal,dy22);
1651             tz               = _mm256_mul_pd(fscal,dz22);
1652
1653             /* Update vectorial force */
1654             fix2             = _mm256_add_pd(fix2,tx);
1655             fiy2             = _mm256_add_pd(fiy2,ty);
1656             fiz2             = _mm256_add_pd(fiz2,tz);
1657
1658             fjx2             = _mm256_add_pd(fjx2,tx);
1659             fjy2             = _mm256_add_pd(fjy2,ty);
1660             fjz2             = _mm256_add_pd(fjz2,tz);
1661
1662             /**************************
1663              * CALCULATE INTERACTIONS *
1664              **************************/
1665
1666             r23              = _mm256_mul_pd(rsq23,rinv23);
1667
1668             /* Calculate table index by multiplying r with table scale and truncate to integer */
1669             rt               = _mm256_mul_pd(r23,vftabscale);
1670             vfitab           = _mm256_cvttpd_epi32(rt);
1671             vfeps            = _mm256_sub_pd(rt,_mm256_round_pd(rt, _MM_FROUND_FLOOR));
1672             vfitab           = _mm_slli_epi32(vfitab,2);
1673
1674             /* CUBIC SPLINE TABLE ELECTROSTATICS */
1675             Y                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
1676             F                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
1677             G                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,2) );
1678             H                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,3) );
1679             GMX_MM256_FULLTRANSPOSE4_PD(Y,F,G,H);
1680             Heps             = _mm256_mul_pd(vfeps,H);
1681             Fp               = _mm256_add_pd(F,_mm256_mul_pd(vfeps,_mm256_add_pd(G,Heps)));
1682             FF               = _mm256_add_pd(Fp,_mm256_mul_pd(vfeps,_mm256_add_pd(G,_mm256_add_pd(Heps,Heps))));
1683             felec            = _mm256_xor_pd(signbit,_mm256_mul_pd(_mm256_mul_pd(qq23,FF),_mm256_mul_pd(vftabscale,rinv23)));
1684
1685             fscal            = felec;
1686
1687             /* Calculate temporary vectorial force */
1688             tx               = _mm256_mul_pd(fscal,dx23);
1689             ty               = _mm256_mul_pd(fscal,dy23);
1690             tz               = _mm256_mul_pd(fscal,dz23);
1691
1692             /* Update vectorial force */
1693             fix2             = _mm256_add_pd(fix2,tx);
1694             fiy2             = _mm256_add_pd(fiy2,ty);
1695             fiz2             = _mm256_add_pd(fiz2,tz);
1696
1697             fjx3             = _mm256_add_pd(fjx3,tx);
1698             fjy3             = _mm256_add_pd(fjy3,ty);
1699             fjz3             = _mm256_add_pd(fjz3,tz);
1700
1701             /**************************
1702              * CALCULATE INTERACTIONS *
1703              **************************/
1704
1705             r31              = _mm256_mul_pd(rsq31,rinv31);
1706
1707             /* Calculate table index by multiplying r with table scale and truncate to integer */
1708             rt               = _mm256_mul_pd(r31,vftabscale);
1709             vfitab           = _mm256_cvttpd_epi32(rt);
1710             vfeps            = _mm256_sub_pd(rt,_mm256_round_pd(rt, _MM_FROUND_FLOOR));
1711             vfitab           = _mm_slli_epi32(vfitab,2);
1712
1713             /* CUBIC SPLINE TABLE ELECTROSTATICS */
1714             Y                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
1715             F                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
1716             G                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,2) );
1717             H                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,3) );
1718             GMX_MM256_FULLTRANSPOSE4_PD(Y,F,G,H);
1719             Heps             = _mm256_mul_pd(vfeps,H);
1720             Fp               = _mm256_add_pd(F,_mm256_mul_pd(vfeps,_mm256_add_pd(G,Heps)));
1721             FF               = _mm256_add_pd(Fp,_mm256_mul_pd(vfeps,_mm256_add_pd(G,_mm256_add_pd(Heps,Heps))));
1722             felec            = _mm256_xor_pd(signbit,_mm256_mul_pd(_mm256_mul_pd(qq31,FF),_mm256_mul_pd(vftabscale,rinv31)));
1723
1724             fscal            = felec;
1725
1726             /* Calculate temporary vectorial force */
1727             tx               = _mm256_mul_pd(fscal,dx31);
1728             ty               = _mm256_mul_pd(fscal,dy31);
1729             tz               = _mm256_mul_pd(fscal,dz31);
1730
1731             /* Update vectorial force */
1732             fix3             = _mm256_add_pd(fix3,tx);
1733             fiy3             = _mm256_add_pd(fiy3,ty);
1734             fiz3             = _mm256_add_pd(fiz3,tz);
1735
1736             fjx1             = _mm256_add_pd(fjx1,tx);
1737             fjy1             = _mm256_add_pd(fjy1,ty);
1738             fjz1             = _mm256_add_pd(fjz1,tz);
1739
1740             /**************************
1741              * CALCULATE INTERACTIONS *
1742              **************************/
1743
1744             r32              = _mm256_mul_pd(rsq32,rinv32);
1745
1746             /* Calculate table index by multiplying r with table scale and truncate to integer */
1747             rt               = _mm256_mul_pd(r32,vftabscale);
1748             vfitab           = _mm256_cvttpd_epi32(rt);
1749             vfeps            = _mm256_sub_pd(rt,_mm256_round_pd(rt, _MM_FROUND_FLOOR));
1750             vfitab           = _mm_slli_epi32(vfitab,2);
1751
1752             /* CUBIC SPLINE TABLE ELECTROSTATICS */
1753             Y                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
1754             F                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
1755             G                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,2) );
1756             H                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,3) );
1757             GMX_MM256_FULLTRANSPOSE4_PD(Y,F,G,H);
1758             Heps             = _mm256_mul_pd(vfeps,H);
1759             Fp               = _mm256_add_pd(F,_mm256_mul_pd(vfeps,_mm256_add_pd(G,Heps)));
1760             FF               = _mm256_add_pd(Fp,_mm256_mul_pd(vfeps,_mm256_add_pd(G,_mm256_add_pd(Heps,Heps))));
1761             felec            = _mm256_xor_pd(signbit,_mm256_mul_pd(_mm256_mul_pd(qq32,FF),_mm256_mul_pd(vftabscale,rinv32)));
1762
1763             fscal            = felec;
1764
1765             /* Calculate temporary vectorial force */
1766             tx               = _mm256_mul_pd(fscal,dx32);
1767             ty               = _mm256_mul_pd(fscal,dy32);
1768             tz               = _mm256_mul_pd(fscal,dz32);
1769
1770             /* Update vectorial force */
1771             fix3             = _mm256_add_pd(fix3,tx);
1772             fiy3             = _mm256_add_pd(fiy3,ty);
1773             fiz3             = _mm256_add_pd(fiz3,tz);
1774
1775             fjx2             = _mm256_add_pd(fjx2,tx);
1776             fjy2             = _mm256_add_pd(fjy2,ty);
1777             fjz2             = _mm256_add_pd(fjz2,tz);
1778
1779             /**************************
1780              * CALCULATE INTERACTIONS *
1781              **************************/
1782
1783             r33              = _mm256_mul_pd(rsq33,rinv33);
1784
1785             /* Calculate table index by multiplying r with table scale and truncate to integer */
1786             rt               = _mm256_mul_pd(r33,vftabscale);
1787             vfitab           = _mm256_cvttpd_epi32(rt);
1788             vfeps            = _mm256_sub_pd(rt,_mm256_round_pd(rt, _MM_FROUND_FLOOR));
1789             vfitab           = _mm_slli_epi32(vfitab,2);
1790
1791             /* CUBIC SPLINE TABLE ELECTROSTATICS */
1792             Y                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
1793             F                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
1794             G                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,2) );
1795             H                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,3) );
1796             GMX_MM256_FULLTRANSPOSE4_PD(Y,F,G,H);
1797             Heps             = _mm256_mul_pd(vfeps,H);
1798             Fp               = _mm256_add_pd(F,_mm256_mul_pd(vfeps,_mm256_add_pd(G,Heps)));
1799             FF               = _mm256_add_pd(Fp,_mm256_mul_pd(vfeps,_mm256_add_pd(G,_mm256_add_pd(Heps,Heps))));
1800             felec            = _mm256_xor_pd(signbit,_mm256_mul_pd(_mm256_mul_pd(qq33,FF),_mm256_mul_pd(vftabscale,rinv33)));
1801
1802             fscal            = felec;
1803
1804             /* Calculate temporary vectorial force */
1805             tx               = _mm256_mul_pd(fscal,dx33);
1806             ty               = _mm256_mul_pd(fscal,dy33);
1807             tz               = _mm256_mul_pd(fscal,dz33);
1808
1809             /* Update vectorial force */
1810             fix3             = _mm256_add_pd(fix3,tx);
1811             fiy3             = _mm256_add_pd(fiy3,ty);
1812             fiz3             = _mm256_add_pd(fiz3,tz);
1813
1814             fjx3             = _mm256_add_pd(fjx3,tx);
1815             fjy3             = _mm256_add_pd(fjy3,ty);
1816             fjz3             = _mm256_add_pd(fjz3,tz);
1817
1818             fjptrA             = f+j_coord_offsetA;
1819             fjptrB             = f+j_coord_offsetB;
1820             fjptrC             = f+j_coord_offsetC;
1821             fjptrD             = f+j_coord_offsetD;
1822
1823             gmx_mm256_decrement_3rvec_4ptr_swizzle_pd(fjptrA+DIM,fjptrB+DIM,fjptrC+DIM,fjptrD+DIM,
1824                                                       fjx1,fjy1,fjz1,fjx2,fjy2,fjz2,fjx3,fjy3,fjz3);
1825
1826             /* Inner loop uses 351 flops */
1827         }
1828
1829         if(jidx<j_index_end)
1830         {
1831
1832             /* Get j neighbor index, and coordinate index */
1833             jnrlistA         = jjnr[jidx];
1834             jnrlistB         = jjnr[jidx+1];
1835             jnrlistC         = jjnr[jidx+2];
1836             jnrlistD         = jjnr[jidx+3];
1837             /* Sign of each element will be negative for non-real atoms.
1838              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
1839              * so use it as val = _mm_andnot_pd(mask,val) to clear dummy entries.
1840              */
1841             tmpmask0 = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
1842
1843             tmpmask1 = _mm_permute_ps(tmpmask0,_GMX_MM_PERMUTE(3,3,2,2));
1844             tmpmask0 = _mm_permute_ps(tmpmask0,_GMX_MM_PERMUTE(1,1,0,0));
1845             dummy_mask = _mm256_castps_pd(gmx_mm256_set_m128(tmpmask1,tmpmask0));
1846
1847             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
1848             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
1849             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
1850             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
1851             j_coord_offsetA  = DIM*jnrA;
1852             j_coord_offsetB  = DIM*jnrB;
1853             j_coord_offsetC  = DIM*jnrC;
1854             j_coord_offsetD  = DIM*jnrD;
1855
1856             /* load j atom coordinates */
1857             gmx_mm256_load_3rvec_4ptr_swizzle_pd(x+j_coord_offsetA+DIM,x+j_coord_offsetB+DIM,
1858                                                  x+j_coord_offsetC+DIM,x+j_coord_offsetD+DIM,
1859                                                  &jx1,&jy1,&jz1,&jx2,&jy2,&jz2,&jx3,&jy3,&jz3);
1860
1861             /* Calculate displacement vector */
1862             dx11             = _mm256_sub_pd(ix1,jx1);
1863             dy11             = _mm256_sub_pd(iy1,jy1);
1864             dz11             = _mm256_sub_pd(iz1,jz1);
1865             dx12             = _mm256_sub_pd(ix1,jx2);
1866             dy12             = _mm256_sub_pd(iy1,jy2);
1867             dz12             = _mm256_sub_pd(iz1,jz2);
1868             dx13             = _mm256_sub_pd(ix1,jx3);
1869             dy13             = _mm256_sub_pd(iy1,jy3);
1870             dz13             = _mm256_sub_pd(iz1,jz3);
1871             dx21             = _mm256_sub_pd(ix2,jx1);
1872             dy21             = _mm256_sub_pd(iy2,jy1);
1873             dz21             = _mm256_sub_pd(iz2,jz1);
1874             dx22             = _mm256_sub_pd(ix2,jx2);
1875             dy22             = _mm256_sub_pd(iy2,jy2);
1876             dz22             = _mm256_sub_pd(iz2,jz2);
1877             dx23             = _mm256_sub_pd(ix2,jx3);
1878             dy23             = _mm256_sub_pd(iy2,jy3);
1879             dz23             = _mm256_sub_pd(iz2,jz3);
1880             dx31             = _mm256_sub_pd(ix3,jx1);
1881             dy31             = _mm256_sub_pd(iy3,jy1);
1882             dz31             = _mm256_sub_pd(iz3,jz1);
1883             dx32             = _mm256_sub_pd(ix3,jx2);
1884             dy32             = _mm256_sub_pd(iy3,jy2);
1885             dz32             = _mm256_sub_pd(iz3,jz2);
1886             dx33             = _mm256_sub_pd(ix3,jx3);
1887             dy33             = _mm256_sub_pd(iy3,jy3);
1888             dz33             = _mm256_sub_pd(iz3,jz3);
1889
1890             /* Calculate squared distance and things based on it */
1891             rsq11            = gmx_mm256_calc_rsq_pd(dx11,dy11,dz11);
1892             rsq12            = gmx_mm256_calc_rsq_pd(dx12,dy12,dz12);
1893             rsq13            = gmx_mm256_calc_rsq_pd(dx13,dy13,dz13);
1894             rsq21            = gmx_mm256_calc_rsq_pd(dx21,dy21,dz21);
1895             rsq22            = gmx_mm256_calc_rsq_pd(dx22,dy22,dz22);
1896             rsq23            = gmx_mm256_calc_rsq_pd(dx23,dy23,dz23);
1897             rsq31            = gmx_mm256_calc_rsq_pd(dx31,dy31,dz31);
1898             rsq32            = gmx_mm256_calc_rsq_pd(dx32,dy32,dz32);
1899             rsq33            = gmx_mm256_calc_rsq_pd(dx33,dy33,dz33);
1900
1901             rinv11           = gmx_mm256_invsqrt_pd(rsq11);
1902             rinv12           = gmx_mm256_invsqrt_pd(rsq12);
1903             rinv13           = gmx_mm256_invsqrt_pd(rsq13);
1904             rinv21           = gmx_mm256_invsqrt_pd(rsq21);
1905             rinv22           = gmx_mm256_invsqrt_pd(rsq22);
1906             rinv23           = gmx_mm256_invsqrt_pd(rsq23);
1907             rinv31           = gmx_mm256_invsqrt_pd(rsq31);
1908             rinv32           = gmx_mm256_invsqrt_pd(rsq32);
1909             rinv33           = gmx_mm256_invsqrt_pd(rsq33);
1910
1911             fjx1             = _mm256_setzero_pd();
1912             fjy1             = _mm256_setzero_pd();
1913             fjz1             = _mm256_setzero_pd();
1914             fjx2             = _mm256_setzero_pd();
1915             fjy2             = _mm256_setzero_pd();
1916             fjz2             = _mm256_setzero_pd();
1917             fjx3             = _mm256_setzero_pd();
1918             fjy3             = _mm256_setzero_pd();
1919             fjz3             = _mm256_setzero_pd();
1920
1921             /**************************
1922              * CALCULATE INTERACTIONS *
1923              **************************/
1924
1925             r11              = _mm256_mul_pd(rsq11,rinv11);
1926             r11              = _mm256_andnot_pd(dummy_mask,r11);
1927
1928             /* Calculate table index by multiplying r with table scale and truncate to integer */
1929             rt               = _mm256_mul_pd(r11,vftabscale);
1930             vfitab           = _mm256_cvttpd_epi32(rt);
1931             vfeps            = _mm256_sub_pd(rt,_mm256_round_pd(rt, _MM_FROUND_FLOOR));
1932             vfitab           = _mm_slli_epi32(vfitab,2);
1933
1934             /* CUBIC SPLINE TABLE ELECTROSTATICS */
1935             Y                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
1936             F                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
1937             G                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,2) );
1938             H                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,3) );
1939             GMX_MM256_FULLTRANSPOSE4_PD(Y,F,G,H);
1940             Heps             = _mm256_mul_pd(vfeps,H);
1941             Fp               = _mm256_add_pd(F,_mm256_mul_pd(vfeps,_mm256_add_pd(G,Heps)));
1942             FF               = _mm256_add_pd(Fp,_mm256_mul_pd(vfeps,_mm256_add_pd(G,_mm256_add_pd(Heps,Heps))));
1943             felec            = _mm256_xor_pd(signbit,_mm256_mul_pd(_mm256_mul_pd(qq11,FF),_mm256_mul_pd(vftabscale,rinv11)));
1944
1945             fscal            = felec;
1946
1947             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
1948
1949             /* Calculate temporary vectorial force */
1950             tx               = _mm256_mul_pd(fscal,dx11);
1951             ty               = _mm256_mul_pd(fscal,dy11);
1952             tz               = _mm256_mul_pd(fscal,dz11);
1953
1954             /* Update vectorial force */
1955             fix1             = _mm256_add_pd(fix1,tx);
1956             fiy1             = _mm256_add_pd(fiy1,ty);
1957             fiz1             = _mm256_add_pd(fiz1,tz);
1958
1959             fjx1             = _mm256_add_pd(fjx1,tx);
1960             fjy1             = _mm256_add_pd(fjy1,ty);
1961             fjz1             = _mm256_add_pd(fjz1,tz);
1962
1963             /**************************
1964              * CALCULATE INTERACTIONS *
1965              **************************/
1966
1967             r12              = _mm256_mul_pd(rsq12,rinv12);
1968             r12              = _mm256_andnot_pd(dummy_mask,r12);
1969
1970             /* Calculate table index by multiplying r with table scale and truncate to integer */
1971             rt               = _mm256_mul_pd(r12,vftabscale);
1972             vfitab           = _mm256_cvttpd_epi32(rt);
1973             vfeps            = _mm256_sub_pd(rt,_mm256_round_pd(rt, _MM_FROUND_FLOOR));
1974             vfitab           = _mm_slli_epi32(vfitab,2);
1975
1976             /* CUBIC SPLINE TABLE ELECTROSTATICS */
1977             Y                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
1978             F                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
1979             G                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,2) );
1980             H                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,3) );
1981             GMX_MM256_FULLTRANSPOSE4_PD(Y,F,G,H);
1982             Heps             = _mm256_mul_pd(vfeps,H);
1983             Fp               = _mm256_add_pd(F,_mm256_mul_pd(vfeps,_mm256_add_pd(G,Heps)));
1984             FF               = _mm256_add_pd(Fp,_mm256_mul_pd(vfeps,_mm256_add_pd(G,_mm256_add_pd(Heps,Heps))));
1985             felec            = _mm256_xor_pd(signbit,_mm256_mul_pd(_mm256_mul_pd(qq12,FF),_mm256_mul_pd(vftabscale,rinv12)));
1986
1987             fscal            = felec;
1988
1989             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
1990
1991             /* Calculate temporary vectorial force */
1992             tx               = _mm256_mul_pd(fscal,dx12);
1993             ty               = _mm256_mul_pd(fscal,dy12);
1994             tz               = _mm256_mul_pd(fscal,dz12);
1995
1996             /* Update vectorial force */
1997             fix1             = _mm256_add_pd(fix1,tx);
1998             fiy1             = _mm256_add_pd(fiy1,ty);
1999             fiz1             = _mm256_add_pd(fiz1,tz);
2000
2001             fjx2             = _mm256_add_pd(fjx2,tx);
2002             fjy2             = _mm256_add_pd(fjy2,ty);
2003             fjz2             = _mm256_add_pd(fjz2,tz);
2004
2005             /**************************
2006              * CALCULATE INTERACTIONS *
2007              **************************/
2008
2009             r13              = _mm256_mul_pd(rsq13,rinv13);
2010             r13              = _mm256_andnot_pd(dummy_mask,r13);
2011
2012             /* Calculate table index by multiplying r with table scale and truncate to integer */
2013             rt               = _mm256_mul_pd(r13,vftabscale);
2014             vfitab           = _mm256_cvttpd_epi32(rt);
2015             vfeps            = _mm256_sub_pd(rt,_mm256_round_pd(rt, _MM_FROUND_FLOOR));
2016             vfitab           = _mm_slli_epi32(vfitab,2);
2017
2018             /* CUBIC SPLINE TABLE ELECTROSTATICS */
2019             Y                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
2020             F                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
2021             G                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,2) );
2022             H                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,3) );
2023             GMX_MM256_FULLTRANSPOSE4_PD(Y,F,G,H);
2024             Heps             = _mm256_mul_pd(vfeps,H);
2025             Fp               = _mm256_add_pd(F,_mm256_mul_pd(vfeps,_mm256_add_pd(G,Heps)));
2026             FF               = _mm256_add_pd(Fp,_mm256_mul_pd(vfeps,_mm256_add_pd(G,_mm256_add_pd(Heps,Heps))));
2027             felec            = _mm256_xor_pd(signbit,_mm256_mul_pd(_mm256_mul_pd(qq13,FF),_mm256_mul_pd(vftabscale,rinv13)));
2028
2029             fscal            = felec;
2030
2031             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
2032
2033             /* Calculate temporary vectorial force */
2034             tx               = _mm256_mul_pd(fscal,dx13);
2035             ty               = _mm256_mul_pd(fscal,dy13);
2036             tz               = _mm256_mul_pd(fscal,dz13);
2037
2038             /* Update vectorial force */
2039             fix1             = _mm256_add_pd(fix1,tx);
2040             fiy1             = _mm256_add_pd(fiy1,ty);
2041             fiz1             = _mm256_add_pd(fiz1,tz);
2042
2043             fjx3             = _mm256_add_pd(fjx3,tx);
2044             fjy3             = _mm256_add_pd(fjy3,ty);
2045             fjz3             = _mm256_add_pd(fjz3,tz);
2046
2047             /**************************
2048              * CALCULATE INTERACTIONS *
2049              **************************/
2050
2051             r21              = _mm256_mul_pd(rsq21,rinv21);
2052             r21              = _mm256_andnot_pd(dummy_mask,r21);
2053
2054             /* Calculate table index by multiplying r with table scale and truncate to integer */
2055             rt               = _mm256_mul_pd(r21,vftabscale);
2056             vfitab           = _mm256_cvttpd_epi32(rt);
2057             vfeps            = _mm256_sub_pd(rt,_mm256_round_pd(rt, _MM_FROUND_FLOOR));
2058             vfitab           = _mm_slli_epi32(vfitab,2);
2059
2060             /* CUBIC SPLINE TABLE ELECTROSTATICS */
2061             Y                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
2062             F                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
2063             G                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,2) );
2064             H                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,3) );
2065             GMX_MM256_FULLTRANSPOSE4_PD(Y,F,G,H);
2066             Heps             = _mm256_mul_pd(vfeps,H);
2067             Fp               = _mm256_add_pd(F,_mm256_mul_pd(vfeps,_mm256_add_pd(G,Heps)));
2068             FF               = _mm256_add_pd(Fp,_mm256_mul_pd(vfeps,_mm256_add_pd(G,_mm256_add_pd(Heps,Heps))));
2069             felec            = _mm256_xor_pd(signbit,_mm256_mul_pd(_mm256_mul_pd(qq21,FF),_mm256_mul_pd(vftabscale,rinv21)));
2070
2071             fscal            = felec;
2072
2073             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
2074
2075             /* Calculate temporary vectorial force */
2076             tx               = _mm256_mul_pd(fscal,dx21);
2077             ty               = _mm256_mul_pd(fscal,dy21);
2078             tz               = _mm256_mul_pd(fscal,dz21);
2079
2080             /* Update vectorial force */
2081             fix2             = _mm256_add_pd(fix2,tx);
2082             fiy2             = _mm256_add_pd(fiy2,ty);
2083             fiz2             = _mm256_add_pd(fiz2,tz);
2084
2085             fjx1             = _mm256_add_pd(fjx1,tx);
2086             fjy1             = _mm256_add_pd(fjy1,ty);
2087             fjz1             = _mm256_add_pd(fjz1,tz);
2088
2089             /**************************
2090              * CALCULATE INTERACTIONS *
2091              **************************/
2092
2093             r22              = _mm256_mul_pd(rsq22,rinv22);
2094             r22              = _mm256_andnot_pd(dummy_mask,r22);
2095
2096             /* Calculate table index by multiplying r with table scale and truncate to integer */
2097             rt               = _mm256_mul_pd(r22,vftabscale);
2098             vfitab           = _mm256_cvttpd_epi32(rt);
2099             vfeps            = _mm256_sub_pd(rt,_mm256_round_pd(rt, _MM_FROUND_FLOOR));
2100             vfitab           = _mm_slli_epi32(vfitab,2);
2101
2102             /* CUBIC SPLINE TABLE ELECTROSTATICS */
2103             Y                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
2104             F                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
2105             G                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,2) );
2106             H                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,3) );
2107             GMX_MM256_FULLTRANSPOSE4_PD(Y,F,G,H);
2108             Heps             = _mm256_mul_pd(vfeps,H);
2109             Fp               = _mm256_add_pd(F,_mm256_mul_pd(vfeps,_mm256_add_pd(G,Heps)));
2110             FF               = _mm256_add_pd(Fp,_mm256_mul_pd(vfeps,_mm256_add_pd(G,_mm256_add_pd(Heps,Heps))));
2111             felec            = _mm256_xor_pd(signbit,_mm256_mul_pd(_mm256_mul_pd(qq22,FF),_mm256_mul_pd(vftabscale,rinv22)));
2112
2113             fscal            = felec;
2114
2115             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
2116
2117             /* Calculate temporary vectorial force */
2118             tx               = _mm256_mul_pd(fscal,dx22);
2119             ty               = _mm256_mul_pd(fscal,dy22);
2120             tz               = _mm256_mul_pd(fscal,dz22);
2121
2122             /* Update vectorial force */
2123             fix2             = _mm256_add_pd(fix2,tx);
2124             fiy2             = _mm256_add_pd(fiy2,ty);
2125             fiz2             = _mm256_add_pd(fiz2,tz);
2126
2127             fjx2             = _mm256_add_pd(fjx2,tx);
2128             fjy2             = _mm256_add_pd(fjy2,ty);
2129             fjz2             = _mm256_add_pd(fjz2,tz);
2130
2131             /**************************
2132              * CALCULATE INTERACTIONS *
2133              **************************/
2134
2135             r23              = _mm256_mul_pd(rsq23,rinv23);
2136             r23              = _mm256_andnot_pd(dummy_mask,r23);
2137
2138             /* Calculate table index by multiplying r with table scale and truncate to integer */
2139             rt               = _mm256_mul_pd(r23,vftabscale);
2140             vfitab           = _mm256_cvttpd_epi32(rt);
2141             vfeps            = _mm256_sub_pd(rt,_mm256_round_pd(rt, _MM_FROUND_FLOOR));
2142             vfitab           = _mm_slli_epi32(vfitab,2);
2143
2144             /* CUBIC SPLINE TABLE ELECTROSTATICS */
2145             Y                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
2146             F                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
2147             G                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,2) );
2148             H                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,3) );
2149             GMX_MM256_FULLTRANSPOSE4_PD(Y,F,G,H);
2150             Heps             = _mm256_mul_pd(vfeps,H);
2151             Fp               = _mm256_add_pd(F,_mm256_mul_pd(vfeps,_mm256_add_pd(G,Heps)));
2152             FF               = _mm256_add_pd(Fp,_mm256_mul_pd(vfeps,_mm256_add_pd(G,_mm256_add_pd(Heps,Heps))));
2153             felec            = _mm256_xor_pd(signbit,_mm256_mul_pd(_mm256_mul_pd(qq23,FF),_mm256_mul_pd(vftabscale,rinv23)));
2154
2155             fscal            = felec;
2156
2157             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
2158
2159             /* Calculate temporary vectorial force */
2160             tx               = _mm256_mul_pd(fscal,dx23);
2161             ty               = _mm256_mul_pd(fscal,dy23);
2162             tz               = _mm256_mul_pd(fscal,dz23);
2163
2164             /* Update vectorial force */
2165             fix2             = _mm256_add_pd(fix2,tx);
2166             fiy2             = _mm256_add_pd(fiy2,ty);
2167             fiz2             = _mm256_add_pd(fiz2,tz);
2168
2169             fjx3             = _mm256_add_pd(fjx3,tx);
2170             fjy3             = _mm256_add_pd(fjy3,ty);
2171             fjz3             = _mm256_add_pd(fjz3,tz);
2172
2173             /**************************
2174              * CALCULATE INTERACTIONS *
2175              **************************/
2176
2177             r31              = _mm256_mul_pd(rsq31,rinv31);
2178             r31              = _mm256_andnot_pd(dummy_mask,r31);
2179
2180             /* Calculate table index by multiplying r with table scale and truncate to integer */
2181             rt               = _mm256_mul_pd(r31,vftabscale);
2182             vfitab           = _mm256_cvttpd_epi32(rt);
2183             vfeps            = _mm256_sub_pd(rt,_mm256_round_pd(rt, _MM_FROUND_FLOOR));
2184             vfitab           = _mm_slli_epi32(vfitab,2);
2185
2186             /* CUBIC SPLINE TABLE ELECTROSTATICS */
2187             Y                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
2188             F                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
2189             G                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,2) );
2190             H                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,3) );
2191             GMX_MM256_FULLTRANSPOSE4_PD(Y,F,G,H);
2192             Heps             = _mm256_mul_pd(vfeps,H);
2193             Fp               = _mm256_add_pd(F,_mm256_mul_pd(vfeps,_mm256_add_pd(G,Heps)));
2194             FF               = _mm256_add_pd(Fp,_mm256_mul_pd(vfeps,_mm256_add_pd(G,_mm256_add_pd(Heps,Heps))));
2195             felec            = _mm256_xor_pd(signbit,_mm256_mul_pd(_mm256_mul_pd(qq31,FF),_mm256_mul_pd(vftabscale,rinv31)));
2196
2197             fscal            = felec;
2198
2199             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
2200
2201             /* Calculate temporary vectorial force */
2202             tx               = _mm256_mul_pd(fscal,dx31);
2203             ty               = _mm256_mul_pd(fscal,dy31);
2204             tz               = _mm256_mul_pd(fscal,dz31);
2205
2206             /* Update vectorial force */
2207             fix3             = _mm256_add_pd(fix3,tx);
2208             fiy3             = _mm256_add_pd(fiy3,ty);
2209             fiz3             = _mm256_add_pd(fiz3,tz);
2210
2211             fjx1             = _mm256_add_pd(fjx1,tx);
2212             fjy1             = _mm256_add_pd(fjy1,ty);
2213             fjz1             = _mm256_add_pd(fjz1,tz);
2214
2215             /**************************
2216              * CALCULATE INTERACTIONS *
2217              **************************/
2218
2219             r32              = _mm256_mul_pd(rsq32,rinv32);
2220             r32              = _mm256_andnot_pd(dummy_mask,r32);
2221
2222             /* Calculate table index by multiplying r with table scale and truncate to integer */
2223             rt               = _mm256_mul_pd(r32,vftabscale);
2224             vfitab           = _mm256_cvttpd_epi32(rt);
2225             vfeps            = _mm256_sub_pd(rt,_mm256_round_pd(rt, _MM_FROUND_FLOOR));
2226             vfitab           = _mm_slli_epi32(vfitab,2);
2227
2228             /* CUBIC SPLINE TABLE ELECTROSTATICS */
2229             Y                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
2230             F                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
2231             G                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,2) );
2232             H                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,3) );
2233             GMX_MM256_FULLTRANSPOSE4_PD(Y,F,G,H);
2234             Heps             = _mm256_mul_pd(vfeps,H);
2235             Fp               = _mm256_add_pd(F,_mm256_mul_pd(vfeps,_mm256_add_pd(G,Heps)));
2236             FF               = _mm256_add_pd(Fp,_mm256_mul_pd(vfeps,_mm256_add_pd(G,_mm256_add_pd(Heps,Heps))));
2237             felec            = _mm256_xor_pd(signbit,_mm256_mul_pd(_mm256_mul_pd(qq32,FF),_mm256_mul_pd(vftabscale,rinv32)));
2238
2239             fscal            = felec;
2240
2241             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
2242
2243             /* Calculate temporary vectorial force */
2244             tx               = _mm256_mul_pd(fscal,dx32);
2245             ty               = _mm256_mul_pd(fscal,dy32);
2246             tz               = _mm256_mul_pd(fscal,dz32);
2247
2248             /* Update vectorial force */
2249             fix3             = _mm256_add_pd(fix3,tx);
2250             fiy3             = _mm256_add_pd(fiy3,ty);
2251             fiz3             = _mm256_add_pd(fiz3,tz);
2252
2253             fjx2             = _mm256_add_pd(fjx2,tx);
2254             fjy2             = _mm256_add_pd(fjy2,ty);
2255             fjz2             = _mm256_add_pd(fjz2,tz);
2256
2257             /**************************
2258              * CALCULATE INTERACTIONS *
2259              **************************/
2260
2261             r33              = _mm256_mul_pd(rsq33,rinv33);
2262             r33              = _mm256_andnot_pd(dummy_mask,r33);
2263
2264             /* Calculate table index by multiplying r with table scale and truncate to integer */
2265             rt               = _mm256_mul_pd(r33,vftabscale);
2266             vfitab           = _mm256_cvttpd_epi32(rt);
2267             vfeps            = _mm256_sub_pd(rt,_mm256_round_pd(rt, _MM_FROUND_FLOOR));
2268             vfitab           = _mm_slli_epi32(vfitab,2);
2269
2270             /* CUBIC SPLINE TABLE ELECTROSTATICS */
2271             Y                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
2272             F                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
2273             G                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,2) );
2274             H                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,3) );
2275             GMX_MM256_FULLTRANSPOSE4_PD(Y,F,G,H);
2276             Heps             = _mm256_mul_pd(vfeps,H);
2277             Fp               = _mm256_add_pd(F,_mm256_mul_pd(vfeps,_mm256_add_pd(G,Heps)));
2278             FF               = _mm256_add_pd(Fp,_mm256_mul_pd(vfeps,_mm256_add_pd(G,_mm256_add_pd(Heps,Heps))));
2279             felec            = _mm256_xor_pd(signbit,_mm256_mul_pd(_mm256_mul_pd(qq33,FF),_mm256_mul_pd(vftabscale,rinv33)));
2280
2281             fscal            = felec;
2282
2283             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
2284
2285             /* Calculate temporary vectorial force */
2286             tx               = _mm256_mul_pd(fscal,dx33);
2287             ty               = _mm256_mul_pd(fscal,dy33);
2288             tz               = _mm256_mul_pd(fscal,dz33);
2289
2290             /* Update vectorial force */
2291             fix3             = _mm256_add_pd(fix3,tx);
2292             fiy3             = _mm256_add_pd(fiy3,ty);
2293             fiz3             = _mm256_add_pd(fiz3,tz);
2294
2295             fjx3             = _mm256_add_pd(fjx3,tx);
2296             fjy3             = _mm256_add_pd(fjy3,ty);
2297             fjz3             = _mm256_add_pd(fjz3,tz);
2298
2299             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
2300             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
2301             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
2302             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
2303
2304             gmx_mm256_decrement_3rvec_4ptr_swizzle_pd(fjptrA+DIM,fjptrB+DIM,fjptrC+DIM,fjptrD+DIM,
2305                                                       fjx1,fjy1,fjz1,fjx2,fjy2,fjz2,fjx3,fjy3,fjz3);
2306
2307             /* Inner loop uses 360 flops */
2308         }
2309
2310         /* End of innermost loop */
2311
2312         gmx_mm256_update_iforce_3atom_swizzle_pd(fix1,fiy1,fiz1,fix2,fiy2,fiz2,fix3,fiy3,fiz3,
2313                                                  f+i_coord_offset+DIM,fshift+i_shift_offset);
2314
2315         /* Increment number of inner iterations */
2316         inneriter                  += j_index_end - j_index_start;
2317
2318         /* Outer loop uses 18 flops */
2319     }
2320
2321     /* Increment number of outer iterations */
2322     outeriter        += nri;
2323
2324     /* Update outer/inner flops */
2325
2326     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_W4W4_F,outeriter*18 + inneriter*360);
2327 }