src/gromacs/gmxlib/nonbonded/nb_kernel_avx_256_double/nb_kernel_ElecCSTab_VdwLJ_GeomW3W3_avx_256_double.c

   1 /*
   2  * This file is part of the GROMACS molecular simulation package.
   3  *
   4  * Copyright (c) 2012,2013,2014, by the GROMACS development team, led by
   5  * Mark Abraham, David van der Spoel, Berk Hess, and Erik Lindahl,
   6  * and including many others, as listed in the AUTHORS file in the
   7  * top-level source directory and at http://www.gromacs.org.
   8  *
   9  * GROMACS is free software; you can redistribute it and/or
  10  * modify it under the terms of the GNU Lesser General Public License
  11  * as published by the Free Software Foundation; either version 2.1
  12  * of the License, or (at your option) any later version.
  13  *
  14  * GROMACS is distributed in the hope that it will be useful,
  15  * but WITHOUT ANY WARRANTY; without even the implied warranty of
  16  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
  17  * Lesser General Public License for more details.
  18  *
  19  * You should have received a copy of the GNU Lesser General Public
  20  * License along with GROMACS; if not, see
  21  * http://www.gnu.org/licenses, or write to the Free Software Foundation,
  22  * Inc., 51 Franklin Street, Fifth Floor, Boston, MA  02110-1301  USA.
  23  *
  24  * If you want to redistribute modifications to GROMACS, please
  25  * consider that scientific software is very special. Version
  26  * control is crucial - bugs must be traceable. We will be happy to
  27  * consider code for inclusion in the official distribution, but
  28  * derived work must not be called official GROMACS. Details are found
  29  * in the README & COPYING files - if they are missing, get the
  30  * official version at http://www.gromacs.org.
  31  *
  32  * To help us fund GROMACS development, we humbly ask that you cite
  33  * the research papers on the package. Check out http://www.gromacs.org.
  34  */
  35 /*
  36  * Note: this file was generated by the GROMACS avx_256_double kernel generator.
  37  */
  38 #include "config.h"
  39
  40 #include <math.h>
  41
  42 #include "../nb_kernel.h"
  43 #include "types/simple.h"
  44 #include "gromacs/math/vec.h"
  45 #include "nrnb.h"
  46
  47 #include "gromacs/simd/math_x86_avx_256_double.h"
  48 #include "kernelutil_x86_avx_256_double.h"
  49
  50 /*
  51  * Gromacs nonbonded kernel:   nb_kernel_ElecCSTab_VdwLJ_GeomW3W3_VF_avx_256_double
  52  * Electrostatics interaction: CubicSplineTable
  53  * VdW interaction:            LennardJones
  54  * Geometry:                   Water3-Water3
  55  * Calculate force/pot:        PotentialAndForce
  56  */
  57 void
  58 nb_kernel_ElecCSTab_VdwLJ_GeomW3W3_VF_avx_256_double
  59                     (t_nblist                    * gmx_restrict       nlist,
  60                      rvec                        * gmx_restrict          xx,
  61                      rvec                        * gmx_restrict          ff,
  62                      t_forcerec                  * gmx_restrict          fr,
  63                      t_mdatoms                   * gmx_restrict     mdatoms,
  64                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
  65                      t_nrnb                      * gmx_restrict        nrnb)
  66 {
  67     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
  68      * just 0 for non-waters.
  69      * Suffixes A,B,C,D refer to j loop unrolling done with AVX, e.g. for the four different
  70      * jnr indices corresponding to data put in the four positions in the SIMD register.
  71      */
  72     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
  73     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
  74     int              jnrA,jnrB,jnrC,jnrD;
  75     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
  76     int              jnrlistE,jnrlistF,jnrlistG,jnrlistH;
  77     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
  78     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
  79     real             rcutoff_scalar;
  80     real             *shiftvec,*fshift,*x,*f;
  81     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD;
  82     real             scratch[4*DIM];
  83     __m256d          tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
  84     real *           vdwioffsetptr0;
  85     __m256d          ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
  86     real *           vdwioffsetptr1;
  87     __m256d          ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
  88     real *           vdwioffsetptr2;
  89     __m256d          ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
  90     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
  91     __m256d          jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
  92     int              vdwjidx1A,vdwjidx1B,vdwjidx1C,vdwjidx1D;
  93     __m256d          jx1,jy1,jz1,fjx1,fjy1,fjz1,jq1,isaj1;
  94     int              vdwjidx2A,vdwjidx2B,vdwjidx2C,vdwjidx2D;
  95     __m256d          jx2,jy2,jz2,fjx2,fjy2,fjz2,jq2,isaj2;
  96     __m256d          dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
  97     __m256d          dx01,dy01,dz01,rsq01,rinv01,rinvsq01,r01,qq01,c6_01,c12_01;
  98     __m256d          dx02,dy02,dz02,rsq02,rinv02,rinvsq02,r02,qq02,c6_02,c12_02;
  99     __m256d          dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
 100     __m256d          dx11,dy11,dz11,rsq11,rinv11,rinvsq11,r11,qq11,c6_11,c12_11;
 101     __m256d          dx12,dy12,dz12,rsq12,rinv12,rinvsq12,r12,qq12,c6_12,c12_12;
 102     __m256d          dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
 103     __m256d          dx21,dy21,dz21,rsq21,rinv21,rinvsq21,r21,qq21,c6_21,c12_21;
 104     __m256d          dx22,dy22,dz22,rsq22,rinv22,rinvsq22,r22,qq22,c6_22,c12_22;
 105     __m256d          velec,felec,velecsum,facel,crf,krf,krf2;
 106     real             *charge;
 107     int              nvdwtype;
 108     __m256d          rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
 109     int              *vdwtype;
 110     real             *vdwparam;
 111     __m256d          one_sixth   = _mm256_set1_pd(1.0/6.0);
 112     __m256d          one_twelfth = _mm256_set1_pd(1.0/12.0);
 113     __m128i          vfitab;
 114     __m128i          ifour       = _mm_set1_epi32(4);
 115     __m256d          rt,vfeps,vftabscale,Y,F,G,H,Heps,Fp,VV,FF;
 116     real             *vftab;
 117     __m256d          dummy_mask,cutoff_mask;
 118     __m128           tmpmask0,tmpmask1;
 119     __m256d          signbit = _mm256_castsi256_pd( _mm256_set1_epi32(0x80000000) );
 120     __m256d          one     = _mm256_set1_pd(1.0);
 121     __m256d          two     = _mm256_set1_pd(2.0);
 122     x                = xx[0];
 123     f                = ff[0];
 124
 125     nri              = nlist->nri;
 126     iinr             = nlist->iinr;
 127     jindex           = nlist->jindex;
 128     jjnr             = nlist->jjnr;
 129     shiftidx         = nlist->shift;
 130     gid              = nlist->gid;
 131     shiftvec         = fr->shift_vec[0];
 132     fshift           = fr->fshift[0];
 133     facel            = _mm256_set1_pd(fr->epsfac);
 134     charge           = mdatoms->chargeA;
 135     nvdwtype         = fr->ntype;
 136     vdwparam         = fr->nbfp;
 137     vdwtype          = mdatoms->typeA;
 138
 139     vftab            = kernel_data->table_elec->data;
 140     vftabscale       = _mm256_set1_pd(kernel_data->table_elec->scale);
 141
 142     /* Setup water-specific parameters */
 143     inr              = nlist->iinr[0];
 144     iq0              = _mm256_mul_pd(facel,_mm256_set1_pd(charge[inr+0]));
 145     iq1              = _mm256_mul_pd(facel,_mm256_set1_pd(charge[inr+1]));
 146     iq2              = _mm256_mul_pd(facel,_mm256_set1_pd(charge[inr+2]));
 147     vdwioffsetptr0   = vdwparam+2*nvdwtype*vdwtype[inr+0];
 148
 149     jq0              = _mm256_set1_pd(charge[inr+0]);
 150     jq1              = _mm256_set1_pd(charge[inr+1]);
 151     jq2              = _mm256_set1_pd(charge[inr+2]);
 152     vdwjidx0A        = 2*vdwtype[inr+0];
 153     qq00             = _mm256_mul_pd(iq0,jq0);
 154     c6_00            = _mm256_set1_pd(vdwioffsetptr0[vdwjidx0A]);
 155     c12_00           = _mm256_set1_pd(vdwioffsetptr0[vdwjidx0A+1]);
 156     qq01             = _mm256_mul_pd(iq0,jq1);
 157     qq02             = _mm256_mul_pd(iq0,jq2);
 158     qq10             = _mm256_mul_pd(iq1,jq0);
 159     qq11             = _mm256_mul_pd(iq1,jq1);
 160     qq12             = _mm256_mul_pd(iq1,jq2);
 161     qq20             = _mm256_mul_pd(iq2,jq0);
 162     qq21             = _mm256_mul_pd(iq2,jq1);
 163     qq22             = _mm256_mul_pd(iq2,jq2);
 164
 165     /* Avoid stupid compiler warnings */
 166     jnrA = jnrB = jnrC = jnrD = 0;
 167     j_coord_offsetA = 0;
 168     j_coord_offsetB = 0;
 169     j_coord_offsetC = 0;
 170     j_coord_offsetD = 0;
 171
 172     outeriter        = 0;
 173     inneriter        = 0;
 174
 175     for(iidx=0;iidx<4*DIM;iidx++)
 176     {
 177         scratch[iidx] = 0.0;
 178     }
 179
 180     /* Start outer loop over neighborlists */
 181     for(iidx=0; iidx<nri; iidx++)
 182     {
 183         /* Load shift vector for this list */
 184         i_shift_offset   = DIM*shiftidx[iidx];
 185
 186         /* Load limits for loop over neighbors */
 187         j_index_start    = jindex[iidx];
 188         j_index_end      = jindex[iidx+1];
 189
 190         /* Get outer coordinate index */
 191         inr              = iinr[iidx];
 192         i_coord_offset   = DIM*inr;
 193
 194         /* Load i particle coords and add shift vector */
 195         gmx_mm256_load_shift_and_3rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset,
 196                                                     &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2);
 197
 198         fix0             = _mm256_setzero_pd();
 199         fiy0             = _mm256_setzero_pd();
 200         fiz0             = _mm256_setzero_pd();
 201         fix1             = _mm256_setzero_pd();
 202         fiy1             = _mm256_setzero_pd();
 203         fiz1             = _mm256_setzero_pd();
 204         fix2             = _mm256_setzero_pd();
 205         fiy2             = _mm256_setzero_pd();
 206         fiz2             = _mm256_setzero_pd();
 207
 208         /* Reset potential sums */
 209         velecsum         = _mm256_setzero_pd();
 210         vvdwsum          = _mm256_setzero_pd();
 211
 212         /* Start inner kernel loop */
 213         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
 214         {
 215
 216             /* Get j neighbor index, and coordinate index */
 217             jnrA             = jjnr[jidx];
 218             jnrB             = jjnr[jidx+1];
 219             jnrC             = jjnr[jidx+2];
 220             jnrD             = jjnr[jidx+3];
 221             j_coord_offsetA  = DIM*jnrA;
 222             j_coord_offsetB  = DIM*jnrB;
 223             j_coord_offsetC  = DIM*jnrC;
 224             j_coord_offsetD  = DIM*jnrD;
 225
 226             /* load j atom coordinates */
 227             gmx_mm256_load_3rvec_4ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 228                                                  x+j_coord_offsetC,x+j_coord_offsetD,
 229                                               &jx0,&jy0,&jz0,&jx1,&jy1,&jz1,&jx2,&jy2,&jz2);
 230
 231             /* Calculate displacement vector */
 232             dx00             = _mm256_sub_pd(ix0,jx0);
 233             dy00             = _mm256_sub_pd(iy0,jy0);
 234             dz00             = _mm256_sub_pd(iz0,jz0);
 235             dx01             = _mm256_sub_pd(ix0,jx1);
 236             dy01             = _mm256_sub_pd(iy0,jy1);
 237             dz01             = _mm256_sub_pd(iz0,jz1);
 238             dx02             = _mm256_sub_pd(ix0,jx2);
 239             dy02             = _mm256_sub_pd(iy0,jy2);
 240             dz02             = _mm256_sub_pd(iz0,jz2);
 241             dx10             = _mm256_sub_pd(ix1,jx0);
 242             dy10             = _mm256_sub_pd(iy1,jy0);
 243             dz10             = _mm256_sub_pd(iz1,jz0);
 244             dx11             = _mm256_sub_pd(ix1,jx1);
 245             dy11             = _mm256_sub_pd(iy1,jy1);
 246             dz11             = _mm256_sub_pd(iz1,jz1);
 247             dx12             = _mm256_sub_pd(ix1,jx2);
 248             dy12             = _mm256_sub_pd(iy1,jy2);
 249             dz12             = _mm256_sub_pd(iz1,jz2);
 250             dx20             = _mm256_sub_pd(ix2,jx0);
 251             dy20             = _mm256_sub_pd(iy2,jy0);
 252             dz20             = _mm256_sub_pd(iz2,jz0);
 253             dx21             = _mm256_sub_pd(ix2,jx1);
 254             dy21             = _mm256_sub_pd(iy2,jy1);
 255             dz21             = _mm256_sub_pd(iz2,jz1);
 256             dx22             = _mm256_sub_pd(ix2,jx2);
 257             dy22             = _mm256_sub_pd(iy2,jy2);
 258             dz22             = _mm256_sub_pd(iz2,jz2);
 259
 260             /* Calculate squared distance and things based on it */
 261             rsq00            = gmx_mm256_calc_rsq_pd(dx00,dy00,dz00);
 262             rsq01            = gmx_mm256_calc_rsq_pd(dx01,dy01,dz01);
 263             rsq02            = gmx_mm256_calc_rsq_pd(dx02,dy02,dz02);
 264             rsq10            = gmx_mm256_calc_rsq_pd(dx10,dy10,dz10);
 265             rsq11            = gmx_mm256_calc_rsq_pd(dx11,dy11,dz11);
 266             rsq12            = gmx_mm256_calc_rsq_pd(dx12,dy12,dz12);
 267             rsq20            = gmx_mm256_calc_rsq_pd(dx20,dy20,dz20);
 268             rsq21            = gmx_mm256_calc_rsq_pd(dx21,dy21,dz21);
 269             rsq22            = gmx_mm256_calc_rsq_pd(dx22,dy22,dz22);
 270
 271             rinv00           = gmx_mm256_invsqrt_pd(rsq00);
 272             rinv01           = gmx_mm256_invsqrt_pd(rsq01);
 273             rinv02           = gmx_mm256_invsqrt_pd(rsq02);
 274             rinv10           = gmx_mm256_invsqrt_pd(rsq10);
 275             rinv11           = gmx_mm256_invsqrt_pd(rsq11);
 276             rinv12           = gmx_mm256_invsqrt_pd(rsq12);
 277             rinv20           = gmx_mm256_invsqrt_pd(rsq20);
 278             rinv21           = gmx_mm256_invsqrt_pd(rsq21);
 279             rinv22           = gmx_mm256_invsqrt_pd(rsq22);
 280
 281             rinvsq00         = _mm256_mul_pd(rinv00,rinv00);
 282
 283             fjx0             = _mm256_setzero_pd();
 284             fjy0             = _mm256_setzero_pd();
 285             fjz0             = _mm256_setzero_pd();
 286             fjx1             = _mm256_setzero_pd();
 287             fjy1             = _mm256_setzero_pd();
 288             fjz1             = _mm256_setzero_pd();
 289             fjx2             = _mm256_setzero_pd();
 290             fjy2             = _mm256_setzero_pd();
 291             fjz2             = _mm256_setzero_pd();
 292
 293             /**************************
 294              * CALCULATE INTERACTIONS *
 295              **************************/
 296
 297             r00              = _mm256_mul_pd(rsq00,rinv00);
 298
 299             /* Calculate table index by multiplying r with table scale and truncate to integer */
 300             rt               = _mm256_mul_pd(r00,vftabscale);
 301             vfitab           = _mm256_cvttpd_epi32(rt);
 302             vfeps            = _mm256_sub_pd(rt,_mm256_round_pd(rt, _MM_FROUND_FLOOR));
 303             vfitab           = _mm_slli_epi32(vfitab,2);
 304
 305             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 306             Y                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
 307             F                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
 308             G                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,2) );
 309             H                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,3) );
 310             GMX_MM256_FULLTRANSPOSE4_PD(Y,F,G,H);
 311             Heps             = _mm256_mul_pd(vfeps,H);
 312             Fp               = _mm256_add_pd(F,_mm256_mul_pd(vfeps,_mm256_add_pd(G,Heps)));
 313             VV               = _mm256_add_pd(Y,_mm256_mul_pd(vfeps,Fp));
 314             velec            = _mm256_mul_pd(qq00,VV);
 315             FF               = _mm256_add_pd(Fp,_mm256_mul_pd(vfeps,_mm256_add_pd(G,_mm256_add_pd(Heps,Heps))));
 316             felec            = _mm256_xor_pd(signbit,_mm256_mul_pd(_mm256_mul_pd(qq00,FF),_mm256_mul_pd(vftabscale,rinv00)));
 317
 318             /* LENNARD-JONES DISPERSION/REPULSION */
 319
 320             rinvsix          = _mm256_mul_pd(_mm256_mul_pd(rinvsq00,rinvsq00),rinvsq00);
 321             vvdw6            = _mm256_mul_pd(c6_00,rinvsix);
 322             vvdw12           = _mm256_mul_pd(c12_00,_mm256_mul_pd(rinvsix,rinvsix));
 323             vvdw             = _mm256_sub_pd( _mm256_mul_pd(vvdw12,one_twelfth) , _mm256_mul_pd(vvdw6,one_sixth) );
 324             fvdw             = _mm256_mul_pd(_mm256_sub_pd(vvdw12,vvdw6),rinvsq00);
 325
 326             /* Update potential sum for this i atom from the interaction with this j atom. */
 327             velecsum         = _mm256_add_pd(velecsum,velec);
 328             vvdwsum          = _mm256_add_pd(vvdwsum,vvdw);
 329
 330             fscal            = _mm256_add_pd(felec,fvdw);
 331
 332             /* Calculate temporary vectorial force */
 333             tx               = _mm256_mul_pd(fscal,dx00);
 334             ty               = _mm256_mul_pd(fscal,dy00);
 335             tz               = _mm256_mul_pd(fscal,dz00);
 336
 337             /* Update vectorial force */
 338             fix0             = _mm256_add_pd(fix0,tx);
 339             fiy0             = _mm256_add_pd(fiy0,ty);
 340             fiz0             = _mm256_add_pd(fiz0,tz);
 341
 342             fjx0             = _mm256_add_pd(fjx0,tx);
 343             fjy0             = _mm256_add_pd(fjy0,ty);
 344             fjz0             = _mm256_add_pd(fjz0,tz);
 345
 346             /**************************
 347              * CALCULATE INTERACTIONS *
 348              **************************/
 349
 350             r01              = _mm256_mul_pd(rsq01,rinv01);
 351
 352             /* Calculate table index by multiplying r with table scale and truncate to integer */
 353             rt               = _mm256_mul_pd(r01,vftabscale);
 354             vfitab           = _mm256_cvttpd_epi32(rt);
 355             vfeps            = _mm256_sub_pd(rt,_mm256_round_pd(rt, _MM_FROUND_FLOOR));
 356             vfitab           = _mm_slli_epi32(vfitab,2);
 357
 358             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 359             Y                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
 360             F                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
 361             G                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,2) );
 362             H                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,3) );
 363             GMX_MM256_FULLTRANSPOSE4_PD(Y,F,G,H);
 364             Heps             = _mm256_mul_pd(vfeps,H);
 365             Fp               = _mm256_add_pd(F,_mm256_mul_pd(vfeps,_mm256_add_pd(G,Heps)));
 366             VV               = _mm256_add_pd(Y,_mm256_mul_pd(vfeps,Fp));
 367             velec            = _mm256_mul_pd(qq01,VV);
 368             FF               = _mm256_add_pd(Fp,_mm256_mul_pd(vfeps,_mm256_add_pd(G,_mm256_add_pd(Heps,Heps))));
 369             felec            = _mm256_xor_pd(signbit,_mm256_mul_pd(_mm256_mul_pd(qq01,FF),_mm256_mul_pd(vftabscale,rinv01)));
 370
 371             /* Update potential sum for this i atom from the interaction with this j atom. */
 372             velecsum         = _mm256_add_pd(velecsum,velec);
 373
 374             fscal            = felec;
 375
 376             /* Calculate temporary vectorial force */
 377             tx               = _mm256_mul_pd(fscal,dx01);
 378             ty               = _mm256_mul_pd(fscal,dy01);
 379             tz               = _mm256_mul_pd(fscal,dz01);
 380
 381             /* Update vectorial force */
 382             fix0             = _mm256_add_pd(fix0,tx);
 383             fiy0             = _mm256_add_pd(fiy0,ty);
 384             fiz0             = _mm256_add_pd(fiz0,tz);
 385
 386             fjx1             = _mm256_add_pd(fjx1,tx);
 387             fjy1             = _mm256_add_pd(fjy1,ty);
 388             fjz1             = _mm256_add_pd(fjz1,tz);
 389
 390             /**************************
 391              * CALCULATE INTERACTIONS *
 392              **************************/
 393
 394             r02              = _mm256_mul_pd(rsq02,rinv02);
 395
 396             /* Calculate table index by multiplying r with table scale and truncate to integer */
 397             rt               = _mm256_mul_pd(r02,vftabscale);
 398             vfitab           = _mm256_cvttpd_epi32(rt);
 399             vfeps            = _mm256_sub_pd(rt,_mm256_round_pd(rt, _MM_FROUND_FLOOR));
 400             vfitab           = _mm_slli_epi32(vfitab,2);
 401
 402             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 403             Y                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
 404             F                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
 405             G                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,2) );
 406             H                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,3) );
 407             GMX_MM256_FULLTRANSPOSE4_PD(Y,F,G,H);
 408             Heps             = _mm256_mul_pd(vfeps,H);
 409             Fp               = _mm256_add_pd(F,_mm256_mul_pd(vfeps,_mm256_add_pd(G,Heps)));
 410             VV               = _mm256_add_pd(Y,_mm256_mul_pd(vfeps,Fp));
 411             velec            = _mm256_mul_pd(qq02,VV);
 412             FF               = _mm256_add_pd(Fp,_mm256_mul_pd(vfeps,_mm256_add_pd(G,_mm256_add_pd(Heps,Heps))));
 413             felec            = _mm256_xor_pd(signbit,_mm256_mul_pd(_mm256_mul_pd(qq02,FF),_mm256_mul_pd(vftabscale,rinv02)));
 414
 415             /* Update potential sum for this i atom from the interaction with this j atom. */
 416             velecsum         = _mm256_add_pd(velecsum,velec);
 417
 418             fscal            = felec;
 419
 420             /* Calculate temporary vectorial force */
 421             tx               = _mm256_mul_pd(fscal,dx02);
 422             ty               = _mm256_mul_pd(fscal,dy02);
 423             tz               = _mm256_mul_pd(fscal,dz02);
 424
 425             /* Update vectorial force */
 426             fix0             = _mm256_add_pd(fix0,tx);
 427             fiy0             = _mm256_add_pd(fiy0,ty);
 428             fiz0             = _mm256_add_pd(fiz0,tz);
 429
 430             fjx2             = _mm256_add_pd(fjx2,tx);
 431             fjy2             = _mm256_add_pd(fjy2,ty);
 432             fjz2             = _mm256_add_pd(fjz2,tz);
 433
 434             /**************************
 435              * CALCULATE INTERACTIONS *
 436              **************************/
 437
 438             r10              = _mm256_mul_pd(rsq10,rinv10);
 439
 440             /* Calculate table index by multiplying r with table scale and truncate to integer */
 441             rt               = _mm256_mul_pd(r10,vftabscale);
 442             vfitab           = _mm256_cvttpd_epi32(rt);
 443             vfeps            = _mm256_sub_pd(rt,_mm256_round_pd(rt, _MM_FROUND_FLOOR));
 444             vfitab           = _mm_slli_epi32(vfitab,2);
 445
 446             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 447             Y                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
 448             F                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
 449             G                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,2) );
 450             H                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,3) );
 451             GMX_MM256_FULLTRANSPOSE4_PD(Y,F,G,H);
 452             Heps             = _mm256_mul_pd(vfeps,H);
 453             Fp               = _mm256_add_pd(F,_mm256_mul_pd(vfeps,_mm256_add_pd(G,Heps)));
 454             VV               = _mm256_add_pd(Y,_mm256_mul_pd(vfeps,Fp));
 455             velec            = _mm256_mul_pd(qq10,VV);
 456             FF               = _mm256_add_pd(Fp,_mm256_mul_pd(vfeps,_mm256_add_pd(G,_mm256_add_pd(Heps,Heps))));
 457             felec            = _mm256_xor_pd(signbit,_mm256_mul_pd(_mm256_mul_pd(qq10,FF),_mm256_mul_pd(vftabscale,rinv10)));
 458
 459             /* Update potential sum for this i atom from the interaction with this j atom. */
 460             velecsum         = _mm256_add_pd(velecsum,velec);
 461
 462             fscal            = felec;
 463
 464             /* Calculate temporary vectorial force */
 465             tx               = _mm256_mul_pd(fscal,dx10);
 466             ty               = _mm256_mul_pd(fscal,dy10);
 467             tz               = _mm256_mul_pd(fscal,dz10);
 468
 469             /* Update vectorial force */
 470             fix1             = _mm256_add_pd(fix1,tx);
 471             fiy1             = _mm256_add_pd(fiy1,ty);
 472             fiz1             = _mm256_add_pd(fiz1,tz);
 473
 474             fjx0             = _mm256_add_pd(fjx0,tx);
 475             fjy0             = _mm256_add_pd(fjy0,ty);
 476             fjz0             = _mm256_add_pd(fjz0,tz);
 477
 478             /**************************
 479              * CALCULATE INTERACTIONS *
 480              **************************/
 481
 482             r11              = _mm256_mul_pd(rsq11,rinv11);
 483
 484             /* Calculate table index by multiplying r with table scale and truncate to integer */
 485             rt               = _mm256_mul_pd(r11,vftabscale);
 486             vfitab           = _mm256_cvttpd_epi32(rt);
 487             vfeps            = _mm256_sub_pd(rt,_mm256_round_pd(rt, _MM_FROUND_FLOOR));
 488             vfitab           = _mm_slli_epi32(vfitab,2);
 489
 490             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 491             Y                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
 492             F                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
 493             G                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,2) );
 494             H                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,3) );
 495             GMX_MM256_FULLTRANSPOSE4_PD(Y,F,G,H);
 496             Heps             = _mm256_mul_pd(vfeps,H);
 497             Fp               = _mm256_add_pd(F,_mm256_mul_pd(vfeps,_mm256_add_pd(G,Heps)));
 498             VV               = _mm256_add_pd(Y,_mm256_mul_pd(vfeps,Fp));
 499             velec            = _mm256_mul_pd(qq11,VV);
 500             FF               = _mm256_add_pd(Fp,_mm256_mul_pd(vfeps,_mm256_add_pd(G,_mm256_add_pd(Heps,Heps))));
 501             felec            = _mm256_xor_pd(signbit,_mm256_mul_pd(_mm256_mul_pd(qq11,FF),_mm256_mul_pd(vftabscale,rinv11)));
 502
 503             /* Update potential sum for this i atom from the interaction with this j atom. */
 504             velecsum         = _mm256_add_pd(velecsum,velec);
 505
 506             fscal            = felec;
 507
 508             /* Calculate temporary vectorial force */
 509             tx               = _mm256_mul_pd(fscal,dx11);
 510             ty               = _mm256_mul_pd(fscal,dy11);
 511             tz               = _mm256_mul_pd(fscal,dz11);
 512
 513             /* Update vectorial force */
 514             fix1             = _mm256_add_pd(fix1,tx);
 515             fiy1             = _mm256_add_pd(fiy1,ty);
 516             fiz1             = _mm256_add_pd(fiz1,tz);
 517
 518             fjx1             = _mm256_add_pd(fjx1,tx);
 519             fjy1             = _mm256_add_pd(fjy1,ty);
 520             fjz1             = _mm256_add_pd(fjz1,tz);
 521
 522             /**************************
 523              * CALCULATE INTERACTIONS *
 524              **************************/
 525
 526             r12              = _mm256_mul_pd(rsq12,rinv12);
 527
 528             /* Calculate table index by multiplying r with table scale and truncate to integer */
 529             rt               = _mm256_mul_pd(r12,vftabscale);
 530             vfitab           = _mm256_cvttpd_epi32(rt);
 531             vfeps            = _mm256_sub_pd(rt,_mm256_round_pd(rt, _MM_FROUND_FLOOR));
 532             vfitab           = _mm_slli_epi32(vfitab,2);
 533
 534             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 535             Y                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
 536             F                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
 537             G                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,2) );
 538             H                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,3) );
 539             GMX_MM256_FULLTRANSPOSE4_PD(Y,F,G,H);
 540             Heps             = _mm256_mul_pd(vfeps,H);
 541             Fp               = _mm256_add_pd(F,_mm256_mul_pd(vfeps,_mm256_add_pd(G,Heps)));
 542             VV               = _mm256_add_pd(Y,_mm256_mul_pd(vfeps,Fp));
 543             velec            = _mm256_mul_pd(qq12,VV);
 544             FF               = _mm256_add_pd(Fp,_mm256_mul_pd(vfeps,_mm256_add_pd(G,_mm256_add_pd(Heps,Heps))));
 545             felec            = _mm256_xor_pd(signbit,_mm256_mul_pd(_mm256_mul_pd(qq12,FF),_mm256_mul_pd(vftabscale,rinv12)));
 546
 547             /* Update potential sum for this i atom from the interaction with this j atom. */
 548             velecsum         = _mm256_add_pd(velecsum,velec);
 549
 550             fscal            = felec;
 551
 552             /* Calculate temporary vectorial force */
 553             tx               = _mm256_mul_pd(fscal,dx12);
 554             ty               = _mm256_mul_pd(fscal,dy12);
 555             tz               = _mm256_mul_pd(fscal,dz12);
 556
 557             /* Update vectorial force */
 558             fix1             = _mm256_add_pd(fix1,tx);
 559             fiy1             = _mm256_add_pd(fiy1,ty);
 560             fiz1             = _mm256_add_pd(fiz1,tz);
 561
 562             fjx2             = _mm256_add_pd(fjx2,tx);
 563             fjy2             = _mm256_add_pd(fjy2,ty);
 564             fjz2             = _mm256_add_pd(fjz2,tz);
 565
 566             /**************************
 567              * CALCULATE INTERACTIONS *
 568              **************************/
 569
 570             r20              = _mm256_mul_pd(rsq20,rinv20);
 571
 572             /* Calculate table index by multiplying r with table scale and truncate to integer */
 573             rt               = _mm256_mul_pd(r20,vftabscale);
 574             vfitab           = _mm256_cvttpd_epi32(rt);
 575             vfeps            = _mm256_sub_pd(rt,_mm256_round_pd(rt, _MM_FROUND_FLOOR));
 576             vfitab           = _mm_slli_epi32(vfitab,2);
 577
 578             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 579             Y                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
 580             F                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
 581             G                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,2) );
 582             H                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,3) );
 583             GMX_MM256_FULLTRANSPOSE4_PD(Y,F,G,H);
 584             Heps             = _mm256_mul_pd(vfeps,H);
 585             Fp               = _mm256_add_pd(F,_mm256_mul_pd(vfeps,_mm256_add_pd(G,Heps)));
 586             VV               = _mm256_add_pd(Y,_mm256_mul_pd(vfeps,Fp));
 587             velec            = _mm256_mul_pd(qq20,VV);
 588             FF               = _mm256_add_pd(Fp,_mm256_mul_pd(vfeps,_mm256_add_pd(G,_mm256_add_pd(Heps,Heps))));
 589             felec            = _mm256_xor_pd(signbit,_mm256_mul_pd(_mm256_mul_pd(qq20,FF),_mm256_mul_pd(vftabscale,rinv20)));
 590
 591             /* Update potential sum for this i atom from the interaction with this j atom. */
 592             velecsum         = _mm256_add_pd(velecsum,velec);
 593
 594             fscal            = felec;
 595
 596             /* Calculate temporary vectorial force */
 597             tx               = _mm256_mul_pd(fscal,dx20);
 598             ty               = _mm256_mul_pd(fscal,dy20);
 599             tz               = _mm256_mul_pd(fscal,dz20);
 600
 601             /* Update vectorial force */
 602             fix2             = _mm256_add_pd(fix2,tx);
 603             fiy2             = _mm256_add_pd(fiy2,ty);
 604             fiz2             = _mm256_add_pd(fiz2,tz);
 605
 606             fjx0             = _mm256_add_pd(fjx0,tx);
 607             fjy0             = _mm256_add_pd(fjy0,ty);
 608             fjz0             = _mm256_add_pd(fjz0,tz);
 609
 610             /**************************
 611              * CALCULATE INTERACTIONS *
 612              **************************/
 613
 614             r21              = _mm256_mul_pd(rsq21,rinv21);
 615
 616             /* Calculate table index by multiplying r with table scale and truncate to integer */
 617             rt               = _mm256_mul_pd(r21,vftabscale);
 618             vfitab           = _mm256_cvttpd_epi32(rt);
 619             vfeps            = _mm256_sub_pd(rt,_mm256_round_pd(rt, _MM_FROUND_FLOOR));
 620             vfitab           = _mm_slli_epi32(vfitab,2);
 621
 622             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 623             Y                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
 624             F                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
 625             G                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,2) );
 626             H                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,3) );
 627             GMX_MM256_FULLTRANSPOSE4_PD(Y,F,G,H);
 628             Heps             = _mm256_mul_pd(vfeps,H);
 629             Fp               = _mm256_add_pd(F,_mm256_mul_pd(vfeps,_mm256_add_pd(G,Heps)));
 630             VV               = _mm256_add_pd(Y,_mm256_mul_pd(vfeps,Fp));
 631             velec            = _mm256_mul_pd(qq21,VV);
 632             FF               = _mm256_add_pd(Fp,_mm256_mul_pd(vfeps,_mm256_add_pd(G,_mm256_add_pd(Heps,Heps))));
 633             felec            = _mm256_xor_pd(signbit,_mm256_mul_pd(_mm256_mul_pd(qq21,FF),_mm256_mul_pd(vftabscale,rinv21)));
 634
 635             /* Update potential sum for this i atom from the interaction with this j atom. */
 636             velecsum         = _mm256_add_pd(velecsum,velec);
 637
 638             fscal            = felec;
 639
 640             /* Calculate temporary vectorial force */
 641             tx               = _mm256_mul_pd(fscal,dx21);
 642             ty               = _mm256_mul_pd(fscal,dy21);
 643             tz               = _mm256_mul_pd(fscal,dz21);
 644
 645             /* Update vectorial force */
 646             fix2             = _mm256_add_pd(fix2,tx);
 647             fiy2             = _mm256_add_pd(fiy2,ty);
 648             fiz2             = _mm256_add_pd(fiz2,tz);
 649
 650             fjx1             = _mm256_add_pd(fjx1,tx);
 651             fjy1             = _mm256_add_pd(fjy1,ty);
 652             fjz1             = _mm256_add_pd(fjz1,tz);
 653
 654             /**************************
 655              * CALCULATE INTERACTIONS *
 656              **************************/
 657
 658             r22              = _mm256_mul_pd(rsq22,rinv22);
 659
 660             /* Calculate table index by multiplying r with table scale and truncate to integer */
 661             rt               = _mm256_mul_pd(r22,vftabscale);
 662             vfitab           = _mm256_cvttpd_epi32(rt);
 663             vfeps            = _mm256_sub_pd(rt,_mm256_round_pd(rt, _MM_FROUND_FLOOR));
 664             vfitab           = _mm_slli_epi32(vfitab,2);
 665
 666             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 667             Y                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
 668             F                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
 669             G                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,2) );
 670             H                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,3) );
 671             GMX_MM256_FULLTRANSPOSE4_PD(Y,F,G,H);
 672             Heps             = _mm256_mul_pd(vfeps,H);
 673             Fp               = _mm256_add_pd(F,_mm256_mul_pd(vfeps,_mm256_add_pd(G,Heps)));
 674             VV               = _mm256_add_pd(Y,_mm256_mul_pd(vfeps,Fp));
 675             velec            = _mm256_mul_pd(qq22,VV);
 676             FF               = _mm256_add_pd(Fp,_mm256_mul_pd(vfeps,_mm256_add_pd(G,_mm256_add_pd(Heps,Heps))));
 677             felec            = _mm256_xor_pd(signbit,_mm256_mul_pd(_mm256_mul_pd(qq22,FF),_mm256_mul_pd(vftabscale,rinv22)));
 678
 679             /* Update potential sum for this i atom from the interaction with this j atom. */
 680             velecsum         = _mm256_add_pd(velecsum,velec);
 681
 682             fscal            = felec;
 683
 684             /* Calculate temporary vectorial force */
 685             tx               = _mm256_mul_pd(fscal,dx22);
 686             ty               = _mm256_mul_pd(fscal,dy22);
 687             tz               = _mm256_mul_pd(fscal,dz22);
 688
 689             /* Update vectorial force */
 690             fix2             = _mm256_add_pd(fix2,tx);
 691             fiy2             = _mm256_add_pd(fiy2,ty);
 692             fiz2             = _mm256_add_pd(fiz2,tz);
 693
 694             fjx2             = _mm256_add_pd(fjx2,tx);
 695             fjy2             = _mm256_add_pd(fjy2,ty);
 696             fjz2             = _mm256_add_pd(fjz2,tz);
 697
 698             fjptrA             = f+j_coord_offsetA;
 699             fjptrB             = f+j_coord_offsetB;
 700             fjptrC             = f+j_coord_offsetC;
 701             fjptrD             = f+j_coord_offsetD;
 702
 703             gmx_mm256_decrement_3rvec_4ptr_swizzle_pd(fjptrA,fjptrB,fjptrC,fjptrD,
 704                                                       fjx0,fjy0,fjz0,fjx1,fjy1,fjz1,fjx2,fjy2,fjz2);
 705
 706             /* Inner loop uses 400 flops */
 707         }
 708
 709         if(jidx<j_index_end)
 710         {
 711
 712             /* Get j neighbor index, and coordinate index */
 713             jnrlistA         = jjnr[jidx];
 714             jnrlistB         = jjnr[jidx+1];
 715             jnrlistC         = jjnr[jidx+2];
 716             jnrlistD         = jjnr[jidx+3];
 717             /* Sign of each element will be negative for non-real atoms.
 718              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 719              * so use it as val = _mm_andnot_pd(mask,val) to clear dummy entries.
 720              */
 721             tmpmask0 = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
 722
 723             tmpmask1 = _mm_permute_ps(tmpmask0,_GMX_MM_PERMUTE(3,3,2,2));
 724             tmpmask0 = _mm_permute_ps(tmpmask0,_GMX_MM_PERMUTE(1,1,0,0));
 725             dummy_mask = _mm256_castps_pd(gmx_mm256_set_m128(tmpmask1,tmpmask0));
 726
 727             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
 728             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
 729             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
 730             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
 731             j_coord_offsetA  = DIM*jnrA;
 732             j_coord_offsetB  = DIM*jnrB;
 733             j_coord_offsetC  = DIM*jnrC;
 734             j_coord_offsetD  = DIM*jnrD;
 735
 736             /* load j atom coordinates */
 737             gmx_mm256_load_3rvec_4ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 738                                                  x+j_coord_offsetC,x+j_coord_offsetD,
 739                                               &jx0,&jy0,&jz0,&jx1,&jy1,&jz1,&jx2,&jy2,&jz2);
 740
 741             /* Calculate displacement vector */
 742             dx00             = _mm256_sub_pd(ix0,jx0);
 743             dy00             = _mm256_sub_pd(iy0,jy0);
 744             dz00             = _mm256_sub_pd(iz0,jz0);
 745             dx01             = _mm256_sub_pd(ix0,jx1);
 746             dy01             = _mm256_sub_pd(iy0,jy1);
 747             dz01             = _mm256_sub_pd(iz0,jz1);
 748             dx02             = _mm256_sub_pd(ix0,jx2);
 749             dy02             = _mm256_sub_pd(iy0,jy2);
 750             dz02             = _mm256_sub_pd(iz0,jz2);
 751             dx10             = _mm256_sub_pd(ix1,jx0);
 752             dy10             = _mm256_sub_pd(iy1,jy0);
 753             dz10             = _mm256_sub_pd(iz1,jz0);
 754             dx11             = _mm256_sub_pd(ix1,jx1);
 755             dy11             = _mm256_sub_pd(iy1,jy1);
 756             dz11             = _mm256_sub_pd(iz1,jz1);
 757             dx12             = _mm256_sub_pd(ix1,jx2);
 758             dy12             = _mm256_sub_pd(iy1,jy2);
 759             dz12             = _mm256_sub_pd(iz1,jz2);
 760             dx20             = _mm256_sub_pd(ix2,jx0);
 761             dy20             = _mm256_sub_pd(iy2,jy0);
 762             dz20             = _mm256_sub_pd(iz2,jz0);
 763             dx21             = _mm256_sub_pd(ix2,jx1);
 764             dy21             = _mm256_sub_pd(iy2,jy1);
 765             dz21             = _mm256_sub_pd(iz2,jz1);
 766             dx22             = _mm256_sub_pd(ix2,jx2);
 767             dy22             = _mm256_sub_pd(iy2,jy2);
 768             dz22             = _mm256_sub_pd(iz2,jz2);
 769
 770             /* Calculate squared distance and things based on it */
 771             rsq00            = gmx_mm256_calc_rsq_pd(dx00,dy00,dz00);
 772             rsq01            = gmx_mm256_calc_rsq_pd(dx01,dy01,dz01);
 773             rsq02            = gmx_mm256_calc_rsq_pd(dx02,dy02,dz02);
 774             rsq10            = gmx_mm256_calc_rsq_pd(dx10,dy10,dz10);
 775             rsq11            = gmx_mm256_calc_rsq_pd(dx11,dy11,dz11);
 776             rsq12            = gmx_mm256_calc_rsq_pd(dx12,dy12,dz12);
 777             rsq20            = gmx_mm256_calc_rsq_pd(dx20,dy20,dz20);
 778             rsq21            = gmx_mm256_calc_rsq_pd(dx21,dy21,dz21);
 779             rsq22            = gmx_mm256_calc_rsq_pd(dx22,dy22,dz22);
 780
 781             rinv00           = gmx_mm256_invsqrt_pd(rsq00);
 782             rinv01           = gmx_mm256_invsqrt_pd(rsq01);
 783             rinv02           = gmx_mm256_invsqrt_pd(rsq02);
 784             rinv10           = gmx_mm256_invsqrt_pd(rsq10);
 785             rinv11           = gmx_mm256_invsqrt_pd(rsq11);
 786             rinv12           = gmx_mm256_invsqrt_pd(rsq12);
 787             rinv20           = gmx_mm256_invsqrt_pd(rsq20);
 788             rinv21           = gmx_mm256_invsqrt_pd(rsq21);
 789             rinv22           = gmx_mm256_invsqrt_pd(rsq22);
 790
 791             rinvsq00         = _mm256_mul_pd(rinv00,rinv00);
 792
 793             fjx0             = _mm256_setzero_pd();
 794             fjy0             = _mm256_setzero_pd();
 795             fjz0             = _mm256_setzero_pd();
 796             fjx1             = _mm256_setzero_pd();
 797             fjy1             = _mm256_setzero_pd();
 798             fjz1             = _mm256_setzero_pd();
 799             fjx2             = _mm256_setzero_pd();
 800             fjy2             = _mm256_setzero_pd();
 801             fjz2             = _mm256_setzero_pd();
 802
 803             /**************************
 804              * CALCULATE INTERACTIONS *
 805              **************************/
 806
 807             r00              = _mm256_mul_pd(rsq00,rinv00);
 808             r00              = _mm256_andnot_pd(dummy_mask,r00);
 809
 810             /* Calculate table index by multiplying r with table scale and truncate to integer */
 811             rt               = _mm256_mul_pd(r00,vftabscale);
 812             vfitab           = _mm256_cvttpd_epi32(rt);
 813             vfeps            = _mm256_sub_pd(rt,_mm256_round_pd(rt, _MM_FROUND_FLOOR));
 814             vfitab           = _mm_slli_epi32(vfitab,2);
 815
 816             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 817             Y                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
 818             F                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
 819             G                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,2) );
 820             H                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,3) );
 821             GMX_MM256_FULLTRANSPOSE4_PD(Y,F,G,H);
 822             Heps             = _mm256_mul_pd(vfeps,H);
 823             Fp               = _mm256_add_pd(F,_mm256_mul_pd(vfeps,_mm256_add_pd(G,Heps)));
 824             VV               = _mm256_add_pd(Y,_mm256_mul_pd(vfeps,Fp));
 825             velec            = _mm256_mul_pd(qq00,VV);
 826             FF               = _mm256_add_pd(Fp,_mm256_mul_pd(vfeps,_mm256_add_pd(G,_mm256_add_pd(Heps,Heps))));
 827             felec            = _mm256_xor_pd(signbit,_mm256_mul_pd(_mm256_mul_pd(qq00,FF),_mm256_mul_pd(vftabscale,rinv00)));
 828
 829             /* LENNARD-JONES DISPERSION/REPULSION */
 830
 831             rinvsix          = _mm256_mul_pd(_mm256_mul_pd(rinvsq00,rinvsq00),rinvsq00);
 832             vvdw6            = _mm256_mul_pd(c6_00,rinvsix);
 833             vvdw12           = _mm256_mul_pd(c12_00,_mm256_mul_pd(rinvsix,rinvsix));
 834             vvdw             = _mm256_sub_pd( _mm256_mul_pd(vvdw12,one_twelfth) , _mm256_mul_pd(vvdw6,one_sixth) );
 835             fvdw             = _mm256_mul_pd(_mm256_sub_pd(vvdw12,vvdw6),rinvsq00);
 836
 837             /* Update potential sum for this i atom from the interaction with this j atom. */
 838             velec            = _mm256_andnot_pd(dummy_mask,velec);
 839             velecsum         = _mm256_add_pd(velecsum,velec);
 840             vvdw             = _mm256_andnot_pd(dummy_mask,vvdw);
 841             vvdwsum          = _mm256_add_pd(vvdwsum,vvdw);
 842
 843             fscal            = _mm256_add_pd(felec,fvdw);
 844
 845             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
 846
 847             /* Calculate temporary vectorial force */
 848             tx               = _mm256_mul_pd(fscal,dx00);
 849             ty               = _mm256_mul_pd(fscal,dy00);
 850             tz               = _mm256_mul_pd(fscal,dz00);
 851
 852             /* Update vectorial force */
 853             fix0             = _mm256_add_pd(fix0,tx);
 854             fiy0             = _mm256_add_pd(fiy0,ty);
 855             fiz0             = _mm256_add_pd(fiz0,tz);
 856
 857             fjx0             = _mm256_add_pd(fjx0,tx);
 858             fjy0             = _mm256_add_pd(fjy0,ty);
 859             fjz0             = _mm256_add_pd(fjz0,tz);
 860
 861             /**************************
 862              * CALCULATE INTERACTIONS *
 863              **************************/
 864
 865             r01              = _mm256_mul_pd(rsq01,rinv01);
 866             r01              = _mm256_andnot_pd(dummy_mask,r01);
 867
 868             /* Calculate table index by multiplying r with table scale and truncate to integer */
 869             rt               = _mm256_mul_pd(r01,vftabscale);
 870             vfitab           = _mm256_cvttpd_epi32(rt);
 871             vfeps            = _mm256_sub_pd(rt,_mm256_round_pd(rt, _MM_FROUND_FLOOR));
 872             vfitab           = _mm_slli_epi32(vfitab,2);
 873
 874             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 875             Y                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
 876             F                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
 877             G                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,2) );
 878             H                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,3) );
 879             GMX_MM256_FULLTRANSPOSE4_PD(Y,F,G,H);
 880             Heps             = _mm256_mul_pd(vfeps,H);
 881             Fp               = _mm256_add_pd(F,_mm256_mul_pd(vfeps,_mm256_add_pd(G,Heps)));
 882             VV               = _mm256_add_pd(Y,_mm256_mul_pd(vfeps,Fp));
 883             velec            = _mm256_mul_pd(qq01,VV);
 884             FF               = _mm256_add_pd(Fp,_mm256_mul_pd(vfeps,_mm256_add_pd(G,_mm256_add_pd(Heps,Heps))));
 885             felec            = _mm256_xor_pd(signbit,_mm256_mul_pd(_mm256_mul_pd(qq01,FF),_mm256_mul_pd(vftabscale,rinv01)));
 886
 887             /* Update potential sum for this i atom from the interaction with this j atom. */
 888             velec            = _mm256_andnot_pd(dummy_mask,velec);
 889             velecsum         = _mm256_add_pd(velecsum,velec);
 890
 891             fscal            = felec;
 892
 893             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
 894
 895             /* Calculate temporary vectorial force */
 896             tx               = _mm256_mul_pd(fscal,dx01);
 897             ty               = _mm256_mul_pd(fscal,dy01);
 898             tz               = _mm256_mul_pd(fscal,dz01);
 899
 900             /* Update vectorial force */
 901             fix0             = _mm256_add_pd(fix0,tx);
 902             fiy0             = _mm256_add_pd(fiy0,ty);
 903             fiz0             = _mm256_add_pd(fiz0,tz);
 904
 905             fjx1             = _mm256_add_pd(fjx1,tx);
 906             fjy1             = _mm256_add_pd(fjy1,ty);
 907             fjz1             = _mm256_add_pd(fjz1,tz);
 908
 909             /**************************
 910              * CALCULATE INTERACTIONS *
 911              **************************/
 912
 913             r02              = _mm256_mul_pd(rsq02,rinv02);
 914             r02              = _mm256_andnot_pd(dummy_mask,r02);
 915
 916             /* Calculate table index by multiplying r with table scale and truncate to integer */
 917             rt               = _mm256_mul_pd(r02,vftabscale);
 918             vfitab           = _mm256_cvttpd_epi32(rt);
 919             vfeps            = _mm256_sub_pd(rt,_mm256_round_pd(rt, _MM_FROUND_FLOOR));
 920             vfitab           = _mm_slli_epi32(vfitab,2);
 921
 922             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 923             Y                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
 924             F                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
 925             G                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,2) );
 926             H                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,3) );
 927             GMX_MM256_FULLTRANSPOSE4_PD(Y,F,G,H);
 928             Heps             = _mm256_mul_pd(vfeps,H);
 929             Fp               = _mm256_add_pd(F,_mm256_mul_pd(vfeps,_mm256_add_pd(G,Heps)));
 930             VV               = _mm256_add_pd(Y,_mm256_mul_pd(vfeps,Fp));
 931             velec            = _mm256_mul_pd(qq02,VV);
 932             FF               = _mm256_add_pd(Fp,_mm256_mul_pd(vfeps,_mm256_add_pd(G,_mm256_add_pd(Heps,Heps))));
 933             felec            = _mm256_xor_pd(signbit,_mm256_mul_pd(_mm256_mul_pd(qq02,FF),_mm256_mul_pd(vftabscale,rinv02)));
 934
 935             /* Update potential sum for this i atom from the interaction with this j atom. */
 936             velec            = _mm256_andnot_pd(dummy_mask,velec);
 937             velecsum         = _mm256_add_pd(velecsum,velec);
 938
 939             fscal            = felec;
 940
 941             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
 942
 943             /* Calculate temporary vectorial force */
 944             tx               = _mm256_mul_pd(fscal,dx02);
 945             ty               = _mm256_mul_pd(fscal,dy02);
 946             tz               = _mm256_mul_pd(fscal,dz02);
 947
 948             /* Update vectorial force */
 949             fix0             = _mm256_add_pd(fix0,tx);
 950             fiy0             = _mm256_add_pd(fiy0,ty);
 951             fiz0             = _mm256_add_pd(fiz0,tz);
 952
 953             fjx2             = _mm256_add_pd(fjx2,tx);
 954             fjy2             = _mm256_add_pd(fjy2,ty);
 955             fjz2             = _mm256_add_pd(fjz2,tz);
 956
 957             /**************************
 958              * CALCULATE INTERACTIONS *
 959              **************************/
 960
 961             r10              = _mm256_mul_pd(rsq10,rinv10);
 962             r10              = _mm256_andnot_pd(dummy_mask,r10);
 963
 964             /* Calculate table index by multiplying r with table scale and truncate to integer */
 965             rt               = _mm256_mul_pd(r10,vftabscale);
 966             vfitab           = _mm256_cvttpd_epi32(rt);
 967             vfeps            = _mm256_sub_pd(rt,_mm256_round_pd(rt, _MM_FROUND_FLOOR));
 968             vfitab           = _mm_slli_epi32(vfitab,2);
 969
 970             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 971             Y                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
 972             F                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
 973             G                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,2) );
 974             H                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,3) );
 975             GMX_MM256_FULLTRANSPOSE4_PD(Y,F,G,H);
 976             Heps             = _mm256_mul_pd(vfeps,H);
 977             Fp               = _mm256_add_pd(F,_mm256_mul_pd(vfeps,_mm256_add_pd(G,Heps)));
 978             VV               = _mm256_add_pd(Y,_mm256_mul_pd(vfeps,Fp));
 979             velec            = _mm256_mul_pd(qq10,VV);
 980             FF               = _mm256_add_pd(Fp,_mm256_mul_pd(vfeps,_mm256_add_pd(G,_mm256_add_pd(Heps,Heps))));
 981             felec            = _mm256_xor_pd(signbit,_mm256_mul_pd(_mm256_mul_pd(qq10,FF),_mm256_mul_pd(vftabscale,rinv10)));
 982
 983             /* Update potential sum for this i atom from the interaction with this j atom. */
 984             velec            = _mm256_andnot_pd(dummy_mask,velec);
 985             velecsum         = _mm256_add_pd(velecsum,velec);
 986
 987             fscal            = felec;
 988
 989             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
 990
 991             /* Calculate temporary vectorial force */
 992             tx               = _mm256_mul_pd(fscal,dx10);
 993             ty               = _mm256_mul_pd(fscal,dy10);
 994             tz               = _mm256_mul_pd(fscal,dz10);
 995
 996             /* Update vectorial force */
 997             fix1             = _mm256_add_pd(fix1,tx);
 998             fiy1             = _mm256_add_pd(fiy1,ty);
 999             fiz1             = _mm256_add_pd(fiz1,tz);
1000
1001             fjx0             = _mm256_add_pd(fjx0,tx);
1002             fjy0             = _mm256_add_pd(fjy0,ty);
1003             fjz0             = _mm256_add_pd(fjz0,tz);
1004
1005             /**************************
1006              * CALCULATE INTERACTIONS *
1007              **************************/
1008
1009             r11              = _mm256_mul_pd(rsq11,rinv11);
1010             r11              = _mm256_andnot_pd(dummy_mask,r11);
1011
1012             /* Calculate table index by multiplying r with table scale and truncate to integer */
1013             rt               = _mm256_mul_pd(r11,vftabscale);
1014             vfitab           = _mm256_cvttpd_epi32(rt);
1015             vfeps            = _mm256_sub_pd(rt,_mm256_round_pd(rt, _MM_FROUND_FLOOR));
1016             vfitab           = _mm_slli_epi32(vfitab,2);
1017
1018             /* CUBIC SPLINE TABLE ELECTROSTATICS */
1019             Y                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
1020             F                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
1021             G                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,2) );
1022             H                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,3) );
1023             GMX_MM256_FULLTRANSPOSE4_PD(Y,F,G,H);
1024             Heps             = _mm256_mul_pd(vfeps,H);
1025             Fp               = _mm256_add_pd(F,_mm256_mul_pd(vfeps,_mm256_add_pd(G,Heps)));
1026             VV               = _mm256_add_pd(Y,_mm256_mul_pd(vfeps,Fp));
1027             velec            = _mm256_mul_pd(qq11,VV);
1028             FF               = _mm256_add_pd(Fp,_mm256_mul_pd(vfeps,_mm256_add_pd(G,_mm256_add_pd(Heps,Heps))));
1029             felec            = _mm256_xor_pd(signbit,_mm256_mul_pd(_mm256_mul_pd(qq11,FF),_mm256_mul_pd(vftabscale,rinv11)));
1030
1031             /* Update potential sum for this i atom from the interaction with this j atom. */
1032             velec            = _mm256_andnot_pd(dummy_mask,velec);
1033             velecsum         = _mm256_add_pd(velecsum,velec);
1034
1035             fscal            = felec;
1036
1037             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
1038
1039             /* Calculate temporary vectorial force */
1040             tx               = _mm256_mul_pd(fscal,dx11);
1041             ty               = _mm256_mul_pd(fscal,dy11);
1042             tz               = _mm256_mul_pd(fscal,dz11);
1043
1044             /* Update vectorial force */
1045             fix1             = _mm256_add_pd(fix1,tx);
1046             fiy1             = _mm256_add_pd(fiy1,ty);
1047             fiz1             = _mm256_add_pd(fiz1,tz);
1048
1049             fjx1             = _mm256_add_pd(fjx1,tx);
1050             fjy1             = _mm256_add_pd(fjy1,ty);
1051             fjz1             = _mm256_add_pd(fjz1,tz);
1052
1053             /**************************
1054              * CALCULATE INTERACTIONS *
1055              **************************/
1056
1057             r12              = _mm256_mul_pd(rsq12,rinv12);
1058             r12              = _mm256_andnot_pd(dummy_mask,r12);
1059
1060             /* Calculate table index by multiplying r with table scale and truncate to integer */
1061             rt               = _mm256_mul_pd(r12,vftabscale);
1062             vfitab           = _mm256_cvttpd_epi32(rt);
1063             vfeps            = _mm256_sub_pd(rt,_mm256_round_pd(rt, _MM_FROUND_FLOOR));
1064             vfitab           = _mm_slli_epi32(vfitab,2);
1065
1066             /* CUBIC SPLINE TABLE ELECTROSTATICS */
1067             Y                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
1068             F                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
1069             G                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,2) );
1070             H                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,3) );
1071             GMX_MM256_FULLTRANSPOSE4_PD(Y,F,G,H);
1072             Heps             = _mm256_mul_pd(vfeps,H);
1073             Fp               = _mm256_add_pd(F,_mm256_mul_pd(vfeps,_mm256_add_pd(G,Heps)));
1074             VV               = _mm256_add_pd(Y,_mm256_mul_pd(vfeps,Fp));
1075             velec            = _mm256_mul_pd(qq12,VV);
1076             FF               = _mm256_add_pd(Fp,_mm256_mul_pd(vfeps,_mm256_add_pd(G,_mm256_add_pd(Heps,Heps))));
1077             felec            = _mm256_xor_pd(signbit,_mm256_mul_pd(_mm256_mul_pd(qq12,FF),_mm256_mul_pd(vftabscale,rinv12)));
1078
1079             /* Update potential sum for this i atom from the interaction with this j atom. */
1080             velec            = _mm256_andnot_pd(dummy_mask,velec);
1081             velecsum         = _mm256_add_pd(velecsum,velec);
1082
1083             fscal            = felec;
1084
1085             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
1086
1087             /* Calculate temporary vectorial force */
1088             tx               = _mm256_mul_pd(fscal,dx12);
1089             ty               = _mm256_mul_pd(fscal,dy12);
1090             tz               = _mm256_mul_pd(fscal,dz12);
1091
1092             /* Update vectorial force */
1093             fix1             = _mm256_add_pd(fix1,tx);
1094             fiy1             = _mm256_add_pd(fiy1,ty);
1095             fiz1             = _mm256_add_pd(fiz1,tz);
1096
1097             fjx2             = _mm256_add_pd(fjx2,tx);
1098             fjy2             = _mm256_add_pd(fjy2,ty);
1099             fjz2             = _mm256_add_pd(fjz2,tz);
1100
1101             /**************************
1102              * CALCULATE INTERACTIONS *
1103              **************************/
1104
1105             r20              = _mm256_mul_pd(rsq20,rinv20);
1106             r20              = _mm256_andnot_pd(dummy_mask,r20);
1107
1108             /* Calculate table index by multiplying r with table scale and truncate to integer */
1109             rt               = _mm256_mul_pd(r20,vftabscale);
1110             vfitab           = _mm256_cvttpd_epi32(rt);
1111             vfeps            = _mm256_sub_pd(rt,_mm256_round_pd(rt, _MM_FROUND_FLOOR));
1112             vfitab           = _mm_slli_epi32(vfitab,2);
1113
1114             /* CUBIC SPLINE TABLE ELECTROSTATICS */
1115             Y                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
1116             F                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
1117             G                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,2) );
1118             H                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,3) );
1119             GMX_MM256_FULLTRANSPOSE4_PD(Y,F,G,H);
1120             Heps             = _mm256_mul_pd(vfeps,H);
1121             Fp               = _mm256_add_pd(F,_mm256_mul_pd(vfeps,_mm256_add_pd(G,Heps)));
1122             VV               = _mm256_add_pd(Y,_mm256_mul_pd(vfeps,Fp));
1123             velec            = _mm256_mul_pd(qq20,VV);
1124             FF               = _mm256_add_pd(Fp,_mm256_mul_pd(vfeps,_mm256_add_pd(G,_mm256_add_pd(Heps,Heps))));
1125             felec            = _mm256_xor_pd(signbit,_mm256_mul_pd(_mm256_mul_pd(qq20,FF),_mm256_mul_pd(vftabscale,rinv20)));
1126
1127             /* Update potential sum for this i atom from the interaction with this j atom. */
1128             velec            = _mm256_andnot_pd(dummy_mask,velec);
1129             velecsum         = _mm256_add_pd(velecsum,velec);
1130
1131             fscal            = felec;
1132
1133             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
1134
1135             /* Calculate temporary vectorial force */
1136             tx               = _mm256_mul_pd(fscal,dx20);
1137             ty               = _mm256_mul_pd(fscal,dy20);
1138             tz               = _mm256_mul_pd(fscal,dz20);
1139
1140             /* Update vectorial force */
1141             fix2             = _mm256_add_pd(fix2,tx);
1142             fiy2             = _mm256_add_pd(fiy2,ty);
1143             fiz2             = _mm256_add_pd(fiz2,tz);
1144
1145             fjx0             = _mm256_add_pd(fjx0,tx);
1146             fjy0             = _mm256_add_pd(fjy0,ty);
1147             fjz0             = _mm256_add_pd(fjz0,tz);
1148
1149             /**************************
1150              * CALCULATE INTERACTIONS *
1151              **************************/
1152
1153             r21              = _mm256_mul_pd(rsq21,rinv21);
1154             r21              = _mm256_andnot_pd(dummy_mask,r21);
1155
1156             /* Calculate table index by multiplying r with table scale and truncate to integer */
1157             rt               = _mm256_mul_pd(r21,vftabscale);
1158             vfitab           = _mm256_cvttpd_epi32(rt);
1159             vfeps            = _mm256_sub_pd(rt,_mm256_round_pd(rt, _MM_FROUND_FLOOR));
1160             vfitab           = _mm_slli_epi32(vfitab,2);
1161
1162             /* CUBIC SPLINE TABLE ELECTROSTATICS */
1163             Y                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
1164             F                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
1165             G                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,2) );
1166             H                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,3) );
1167             GMX_MM256_FULLTRANSPOSE4_PD(Y,F,G,H);
1168             Heps             = _mm256_mul_pd(vfeps,H);
1169             Fp               = _mm256_add_pd(F,_mm256_mul_pd(vfeps,_mm256_add_pd(G,Heps)));
1170             VV               = _mm256_add_pd(Y,_mm256_mul_pd(vfeps,Fp));
1171             velec            = _mm256_mul_pd(qq21,VV);
1172             FF               = _mm256_add_pd(Fp,_mm256_mul_pd(vfeps,_mm256_add_pd(G,_mm256_add_pd(Heps,Heps))));
1173             felec            = _mm256_xor_pd(signbit,_mm256_mul_pd(_mm256_mul_pd(qq21,FF),_mm256_mul_pd(vftabscale,rinv21)));
1174
1175             /* Update potential sum for this i atom from the interaction with this j atom. */
1176             velec            = _mm256_andnot_pd(dummy_mask,velec);
1177             velecsum         = _mm256_add_pd(velecsum,velec);
1178
1179             fscal            = felec;
1180
1181             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
1182
1183             /* Calculate temporary vectorial force */
1184             tx               = _mm256_mul_pd(fscal,dx21);
1185             ty               = _mm256_mul_pd(fscal,dy21);
1186             tz               = _mm256_mul_pd(fscal,dz21);
1187
1188             /* Update vectorial force */
1189             fix2             = _mm256_add_pd(fix2,tx);
1190             fiy2             = _mm256_add_pd(fiy2,ty);
1191             fiz2             = _mm256_add_pd(fiz2,tz);
1192
1193             fjx1             = _mm256_add_pd(fjx1,tx);
1194             fjy1             = _mm256_add_pd(fjy1,ty);
1195             fjz1             = _mm256_add_pd(fjz1,tz);
1196
1197             /**************************
1198              * CALCULATE INTERACTIONS *
1199              **************************/
1200
1201             r22              = _mm256_mul_pd(rsq22,rinv22);
1202             r22              = _mm256_andnot_pd(dummy_mask,r22);
1203
1204             /* Calculate table index by multiplying r with table scale and truncate to integer */
1205             rt               = _mm256_mul_pd(r22,vftabscale);
1206             vfitab           = _mm256_cvttpd_epi32(rt);
1207             vfeps            = _mm256_sub_pd(rt,_mm256_round_pd(rt, _MM_FROUND_FLOOR));
1208             vfitab           = _mm_slli_epi32(vfitab,2);
1209
1210             /* CUBIC SPLINE TABLE ELECTROSTATICS */
1211             Y                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
1212             F                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
1213             G                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,2) );
1214             H                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,3) );
1215             GMX_MM256_FULLTRANSPOSE4_PD(Y,F,G,H);
1216             Heps             = _mm256_mul_pd(vfeps,H);
1217             Fp               = _mm256_add_pd(F,_mm256_mul_pd(vfeps,_mm256_add_pd(G,Heps)));
1218             VV               = _mm256_add_pd(Y,_mm256_mul_pd(vfeps,Fp));
1219             velec            = _mm256_mul_pd(qq22,VV);
1220             FF               = _mm256_add_pd(Fp,_mm256_mul_pd(vfeps,_mm256_add_pd(G,_mm256_add_pd(Heps,Heps))));
1221             felec            = _mm256_xor_pd(signbit,_mm256_mul_pd(_mm256_mul_pd(qq22,FF),_mm256_mul_pd(vftabscale,rinv22)));
1222
1223             /* Update potential sum for this i atom from the interaction with this j atom. */
1224             velec            = _mm256_andnot_pd(dummy_mask,velec);
1225             velecsum         = _mm256_add_pd(velecsum,velec);
1226
1227             fscal            = felec;
1228
1229             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
1230
1231             /* Calculate temporary vectorial force */
1232             tx               = _mm256_mul_pd(fscal,dx22);
1233             ty               = _mm256_mul_pd(fscal,dy22);
1234             tz               = _mm256_mul_pd(fscal,dz22);
1235
1236             /* Update vectorial force */
1237             fix2             = _mm256_add_pd(fix2,tx);
1238             fiy2             = _mm256_add_pd(fiy2,ty);
1239             fiz2             = _mm256_add_pd(fiz2,tz);
1240
1241             fjx2             = _mm256_add_pd(fjx2,tx);
1242             fjy2             = _mm256_add_pd(fjy2,ty);
1243             fjz2             = _mm256_add_pd(fjz2,tz);
1244
1245             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
1246             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
1247             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
1248             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
1249
1250             gmx_mm256_decrement_3rvec_4ptr_swizzle_pd(fjptrA,fjptrB,fjptrC,fjptrD,
1251                                                       fjx0,fjy0,fjz0,fjx1,fjy1,fjz1,fjx2,fjy2,fjz2);
1252
1253             /* Inner loop uses 409 flops */
1254         }
1255
1256         /* End of innermost loop */
1257
1258         gmx_mm256_update_iforce_3atom_swizzle_pd(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,
1259                                                  f+i_coord_offset,fshift+i_shift_offset);
1260
1261         ggid                        = gid[iidx];
1262         /* Update potential energies */
1263         gmx_mm256_update_1pot_pd(velecsum,kernel_data->energygrp_elec+ggid);
1264         gmx_mm256_update_1pot_pd(vvdwsum,kernel_data->energygrp_vdw+ggid);
1265
1266         /* Increment number of inner iterations */
1267         inneriter                  += j_index_end - j_index_start;
1268
1269         /* Outer loop uses 20 flops */
1270     }
1271
1272     /* Increment number of outer iterations */
1273     outeriter        += nri;
1274
1275     /* Update outer/inner flops */
1276
1277     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W3W3_VF,outeriter*20 + inneriter*409);
1278 }
1279 /*
1280  * Gromacs nonbonded kernel:   nb_kernel_ElecCSTab_VdwLJ_GeomW3W3_F_avx_256_double
1281  * Electrostatics interaction: CubicSplineTable
1282  * VdW interaction:            LennardJones
1283  * Geometry:                   Water3-Water3
1284  * Calculate force/pot:        Force
1285  */
1286 void
1287 nb_kernel_ElecCSTab_VdwLJ_GeomW3W3_F_avx_256_double
1288                     (t_nblist                    * gmx_restrict       nlist,
1289                      rvec                        * gmx_restrict          xx,
1290                      rvec                        * gmx_restrict          ff,
1291                      t_forcerec                  * gmx_restrict          fr,
1292                      t_mdatoms                   * gmx_restrict     mdatoms,
1293                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
1294                      t_nrnb                      * gmx_restrict        nrnb)
1295 {
1296     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
1297      * just 0 for non-waters.
1298      * Suffixes A,B,C,D refer to j loop unrolling done with AVX, e.g. for the four different
1299      * jnr indices corresponding to data put in the four positions in the SIMD register.
1300      */
1301     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
1302     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
1303     int              jnrA,jnrB,jnrC,jnrD;
1304     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
1305     int              jnrlistE,jnrlistF,jnrlistG,jnrlistH;
1306     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
1307     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
1308     real             rcutoff_scalar;
1309     real             *shiftvec,*fshift,*x,*f;
1310     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD;
1311     real             scratch[4*DIM];
1312     __m256d          tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
1313     real *           vdwioffsetptr0;
1314     __m256d          ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
1315     real *           vdwioffsetptr1;
1316     __m256d          ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
1317     real *           vdwioffsetptr2;
1318     __m256d          ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
1319     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
1320     __m256d          jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
1321     int              vdwjidx1A,vdwjidx1B,vdwjidx1C,vdwjidx1D;
1322     __m256d          jx1,jy1,jz1,fjx1,fjy1,fjz1,jq1,isaj1;
1323     int              vdwjidx2A,vdwjidx2B,vdwjidx2C,vdwjidx2D;
1324     __m256d          jx2,jy2,jz2,fjx2,fjy2,fjz2,jq2,isaj2;
1325     __m256d          dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
1326     __m256d          dx01,dy01,dz01,rsq01,rinv01,rinvsq01,r01,qq01,c6_01,c12_01;
1327     __m256d          dx02,dy02,dz02,rsq02,rinv02,rinvsq02,r02,qq02,c6_02,c12_02;
1328     __m256d          dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
1329     __m256d          dx11,dy11,dz11,rsq11,rinv11,rinvsq11,r11,qq11,c6_11,c12_11;
1330     __m256d          dx12,dy12,dz12,rsq12,rinv12,rinvsq12,r12,qq12,c6_12,c12_12;
1331     __m256d          dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
1332     __m256d          dx21,dy21,dz21,rsq21,rinv21,rinvsq21,r21,qq21,c6_21,c12_21;
1333     __m256d          dx22,dy22,dz22,rsq22,rinv22,rinvsq22,r22,qq22,c6_22,c12_22;
1334     __m256d          velec,felec,velecsum,facel,crf,krf,krf2;
1335     real             *charge;
1336     int              nvdwtype;
1337     __m256d          rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
1338     int              *vdwtype;
1339     real             *vdwparam;
1340     __m256d          one_sixth   = _mm256_set1_pd(1.0/6.0);
1341     __m256d          one_twelfth = _mm256_set1_pd(1.0/12.0);
1342     __m128i          vfitab;
1343     __m128i          ifour       = _mm_set1_epi32(4);
1344     __m256d          rt,vfeps,vftabscale,Y,F,G,H,Heps,Fp,VV,FF;
1345     real             *vftab;
1346     __m256d          dummy_mask,cutoff_mask;
1347     __m128           tmpmask0,tmpmask1;
1348     __m256d          signbit = _mm256_castsi256_pd( _mm256_set1_epi32(0x80000000) );
1349     __m256d          one     = _mm256_set1_pd(1.0);
1350     __m256d          two     = _mm256_set1_pd(2.0);
1351     x                = xx[0];
1352     f                = ff[0];
1353
1354     nri              = nlist->nri;
1355     iinr             = nlist->iinr;
1356     jindex           = nlist->jindex;
1357     jjnr             = nlist->jjnr;
1358     shiftidx         = nlist->shift;
1359     gid              = nlist->gid;
1360     shiftvec         = fr->shift_vec[0];
1361     fshift           = fr->fshift[0];
1362     facel            = _mm256_set1_pd(fr->epsfac);
1363     charge           = mdatoms->chargeA;
1364     nvdwtype         = fr->ntype;
1365     vdwparam         = fr->nbfp;
1366     vdwtype          = mdatoms->typeA;
1367
1368     vftab            = kernel_data->table_elec->data;
1369     vftabscale       = _mm256_set1_pd(kernel_data->table_elec->scale);
1370
1371     /* Setup water-specific parameters */
1372     inr              = nlist->iinr[0];
1373     iq0              = _mm256_mul_pd(facel,_mm256_set1_pd(charge[inr+0]));
1374     iq1              = _mm256_mul_pd(facel,_mm256_set1_pd(charge[inr+1]));
1375     iq2              = _mm256_mul_pd(facel,_mm256_set1_pd(charge[inr+2]));
1376     vdwioffsetptr0   = vdwparam+2*nvdwtype*vdwtype[inr+0];
1377
1378     jq0              = _mm256_set1_pd(charge[inr+0]);
1379     jq1              = _mm256_set1_pd(charge[inr+1]);
1380     jq2              = _mm256_set1_pd(charge[inr+2]);
1381     vdwjidx0A        = 2*vdwtype[inr+0];
1382     qq00             = _mm256_mul_pd(iq0,jq0);
1383     c6_00            = _mm256_set1_pd(vdwioffsetptr0[vdwjidx0A]);
1384     c12_00           = _mm256_set1_pd(vdwioffsetptr0[vdwjidx0A+1]);
1385     qq01             = _mm256_mul_pd(iq0,jq1);
1386     qq02             = _mm256_mul_pd(iq0,jq2);
1387     qq10             = _mm256_mul_pd(iq1,jq0);
1388     qq11             = _mm256_mul_pd(iq1,jq1);
1389     qq12             = _mm256_mul_pd(iq1,jq2);
1390     qq20             = _mm256_mul_pd(iq2,jq0);
1391     qq21             = _mm256_mul_pd(iq2,jq1);
1392     qq22             = _mm256_mul_pd(iq2,jq2);
1393
1394     /* Avoid stupid compiler warnings */
1395     jnrA = jnrB = jnrC = jnrD = 0;
1396     j_coord_offsetA = 0;
1397     j_coord_offsetB = 0;
1398     j_coord_offsetC = 0;
1399     j_coord_offsetD = 0;
1400
1401     outeriter        = 0;
1402     inneriter        = 0;
1403
1404     for(iidx=0;iidx<4*DIM;iidx++)
1405     {
1406         scratch[iidx] = 0.0;
1407     }
1408
1409     /* Start outer loop over neighborlists */
1410     for(iidx=0; iidx<nri; iidx++)
1411     {
1412         /* Load shift vector for this list */
1413         i_shift_offset   = DIM*shiftidx[iidx];
1414
1415         /* Load limits for loop over neighbors */
1416         j_index_start    = jindex[iidx];
1417         j_index_end      = jindex[iidx+1];
1418
1419         /* Get outer coordinate index */
1420         inr              = iinr[iidx];
1421         i_coord_offset   = DIM*inr;
1422
1423         /* Load i particle coords and add shift vector */
1424         gmx_mm256_load_shift_and_3rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset,
1425                                                     &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2);
1426
1427         fix0             = _mm256_setzero_pd();
1428         fiy0             = _mm256_setzero_pd();
1429         fiz0             = _mm256_setzero_pd();
1430         fix1             = _mm256_setzero_pd();
1431         fiy1             = _mm256_setzero_pd();
1432         fiz1             = _mm256_setzero_pd();
1433         fix2             = _mm256_setzero_pd();
1434         fiy2             = _mm256_setzero_pd();
1435         fiz2             = _mm256_setzero_pd();
1436
1437         /* Start inner kernel loop */
1438         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
1439         {
1440
1441             /* Get j neighbor index, and coordinate index */
1442             jnrA             = jjnr[jidx];
1443             jnrB             = jjnr[jidx+1];
1444             jnrC             = jjnr[jidx+2];
1445             jnrD             = jjnr[jidx+3];
1446             j_coord_offsetA  = DIM*jnrA;
1447             j_coord_offsetB  = DIM*jnrB;
1448             j_coord_offsetC  = DIM*jnrC;
1449             j_coord_offsetD  = DIM*jnrD;
1450
1451             /* load j atom coordinates */
1452             gmx_mm256_load_3rvec_4ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
1453                                                  x+j_coord_offsetC,x+j_coord_offsetD,
1454                                               &jx0,&jy0,&jz0,&jx1,&jy1,&jz1,&jx2,&jy2,&jz2);
1455
1456             /* Calculate displacement vector */
1457             dx00             = _mm256_sub_pd(ix0,jx0);
1458             dy00             = _mm256_sub_pd(iy0,jy0);
1459             dz00             = _mm256_sub_pd(iz0,jz0);
1460             dx01             = _mm256_sub_pd(ix0,jx1);
1461             dy01             = _mm256_sub_pd(iy0,jy1);
1462             dz01             = _mm256_sub_pd(iz0,jz1);
1463             dx02             = _mm256_sub_pd(ix0,jx2);
1464             dy02             = _mm256_sub_pd(iy0,jy2);
1465             dz02             = _mm256_sub_pd(iz0,jz2);
1466             dx10             = _mm256_sub_pd(ix1,jx0);
1467             dy10             = _mm256_sub_pd(iy1,jy0);
1468             dz10             = _mm256_sub_pd(iz1,jz0);
1469             dx11             = _mm256_sub_pd(ix1,jx1);
1470             dy11             = _mm256_sub_pd(iy1,jy1);
1471             dz11             = _mm256_sub_pd(iz1,jz1);
1472             dx12             = _mm256_sub_pd(ix1,jx2);
1473             dy12             = _mm256_sub_pd(iy1,jy2);
1474             dz12             = _mm256_sub_pd(iz1,jz2);
1475             dx20             = _mm256_sub_pd(ix2,jx0);
1476             dy20             = _mm256_sub_pd(iy2,jy0);
1477             dz20             = _mm256_sub_pd(iz2,jz0);
1478             dx21             = _mm256_sub_pd(ix2,jx1);
1479             dy21             = _mm256_sub_pd(iy2,jy1);
1480             dz21             = _mm256_sub_pd(iz2,jz1);
1481             dx22             = _mm256_sub_pd(ix2,jx2);
1482             dy22             = _mm256_sub_pd(iy2,jy2);
1483             dz22             = _mm256_sub_pd(iz2,jz2);
1484
1485             /* Calculate squared distance and things based on it */
1486             rsq00            = gmx_mm256_calc_rsq_pd(dx00,dy00,dz00);
1487             rsq01            = gmx_mm256_calc_rsq_pd(dx01,dy01,dz01);
1488             rsq02            = gmx_mm256_calc_rsq_pd(dx02,dy02,dz02);
1489             rsq10            = gmx_mm256_calc_rsq_pd(dx10,dy10,dz10);
1490             rsq11            = gmx_mm256_calc_rsq_pd(dx11,dy11,dz11);
1491             rsq12            = gmx_mm256_calc_rsq_pd(dx12,dy12,dz12);
1492             rsq20            = gmx_mm256_calc_rsq_pd(dx20,dy20,dz20);
1493             rsq21            = gmx_mm256_calc_rsq_pd(dx21,dy21,dz21);
1494             rsq22            = gmx_mm256_calc_rsq_pd(dx22,dy22,dz22);
1495
1496             rinv00           = gmx_mm256_invsqrt_pd(rsq00);
1497             rinv01           = gmx_mm256_invsqrt_pd(rsq01);
1498             rinv02           = gmx_mm256_invsqrt_pd(rsq02);
1499             rinv10           = gmx_mm256_invsqrt_pd(rsq10);
1500             rinv11           = gmx_mm256_invsqrt_pd(rsq11);
1501             rinv12           = gmx_mm256_invsqrt_pd(rsq12);
1502             rinv20           = gmx_mm256_invsqrt_pd(rsq20);
1503             rinv21           = gmx_mm256_invsqrt_pd(rsq21);
1504             rinv22           = gmx_mm256_invsqrt_pd(rsq22);
1505
1506             rinvsq00         = _mm256_mul_pd(rinv00,rinv00);
1507
1508             fjx0             = _mm256_setzero_pd();
1509             fjy0             = _mm256_setzero_pd();
1510             fjz0             = _mm256_setzero_pd();
1511             fjx1             = _mm256_setzero_pd();
1512             fjy1             = _mm256_setzero_pd();
1513             fjz1             = _mm256_setzero_pd();
1514             fjx2             = _mm256_setzero_pd();
1515             fjy2             = _mm256_setzero_pd();
1516             fjz2             = _mm256_setzero_pd();
1517
1518             /**************************
1519              * CALCULATE INTERACTIONS *
1520              **************************/
1521
1522             r00              = _mm256_mul_pd(rsq00,rinv00);
1523
1524             /* Calculate table index by multiplying r with table scale and truncate to integer */
1525             rt               = _mm256_mul_pd(r00,vftabscale);
1526             vfitab           = _mm256_cvttpd_epi32(rt);
1527             vfeps            = _mm256_sub_pd(rt,_mm256_round_pd(rt, _MM_FROUND_FLOOR));
1528             vfitab           = _mm_slli_epi32(vfitab,2);
1529
1530             /* CUBIC SPLINE TABLE ELECTROSTATICS */
1531             Y                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
1532             F                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
1533             G                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,2) );
1534             H                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,3) );
1535             GMX_MM256_FULLTRANSPOSE4_PD(Y,F,G,H);
1536             Heps             = _mm256_mul_pd(vfeps,H);
1537             Fp               = _mm256_add_pd(F,_mm256_mul_pd(vfeps,_mm256_add_pd(G,Heps)));
1538             FF               = _mm256_add_pd(Fp,_mm256_mul_pd(vfeps,_mm256_add_pd(G,_mm256_add_pd(Heps,Heps))));
1539             felec            = _mm256_xor_pd(signbit,_mm256_mul_pd(_mm256_mul_pd(qq00,FF),_mm256_mul_pd(vftabscale,rinv00)));
1540
1541             /* LENNARD-JONES DISPERSION/REPULSION */
1542
1543             rinvsix          = _mm256_mul_pd(_mm256_mul_pd(rinvsq00,rinvsq00),rinvsq00);
1544             fvdw             = _mm256_mul_pd(_mm256_sub_pd(_mm256_mul_pd(c12_00,rinvsix),c6_00),_mm256_mul_pd(rinvsix,rinvsq00));
1545
1546             fscal            = _mm256_add_pd(felec,fvdw);
1547
1548             /* Calculate temporary vectorial force */
1549             tx               = _mm256_mul_pd(fscal,dx00);
1550             ty               = _mm256_mul_pd(fscal,dy00);
1551             tz               = _mm256_mul_pd(fscal,dz00);
1552
1553             /* Update vectorial force */
1554             fix0             = _mm256_add_pd(fix0,tx);
1555             fiy0             = _mm256_add_pd(fiy0,ty);
1556             fiz0             = _mm256_add_pd(fiz0,tz);
1557
1558             fjx0             = _mm256_add_pd(fjx0,tx);
1559             fjy0             = _mm256_add_pd(fjy0,ty);
1560             fjz0             = _mm256_add_pd(fjz0,tz);
1561
1562             /**************************
1563              * CALCULATE INTERACTIONS *
1564              **************************/
1565
1566             r01              = _mm256_mul_pd(rsq01,rinv01);
1567
1568             /* Calculate table index by multiplying r with table scale and truncate to integer */
1569             rt               = _mm256_mul_pd(r01,vftabscale);
1570             vfitab           = _mm256_cvttpd_epi32(rt);
1571             vfeps            = _mm256_sub_pd(rt,_mm256_round_pd(rt, _MM_FROUND_FLOOR));
1572             vfitab           = _mm_slli_epi32(vfitab,2);
1573
1574             /* CUBIC SPLINE TABLE ELECTROSTATICS */
1575             Y                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
1576             F                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
1577             G                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,2) );
1578             H                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,3) );
1579             GMX_MM256_FULLTRANSPOSE4_PD(Y,F,G,H);
1580             Heps             = _mm256_mul_pd(vfeps,H);
1581             Fp               = _mm256_add_pd(F,_mm256_mul_pd(vfeps,_mm256_add_pd(G,Heps)));
1582             FF               = _mm256_add_pd(Fp,_mm256_mul_pd(vfeps,_mm256_add_pd(G,_mm256_add_pd(Heps,Heps))));
1583             felec            = _mm256_xor_pd(signbit,_mm256_mul_pd(_mm256_mul_pd(qq01,FF),_mm256_mul_pd(vftabscale,rinv01)));
1584
1585             fscal            = felec;
1586
1587             /* Calculate temporary vectorial force */
1588             tx               = _mm256_mul_pd(fscal,dx01);
1589             ty               = _mm256_mul_pd(fscal,dy01);
1590             tz               = _mm256_mul_pd(fscal,dz01);
1591
1592             /* Update vectorial force */
1593             fix0             = _mm256_add_pd(fix0,tx);
1594             fiy0             = _mm256_add_pd(fiy0,ty);
1595             fiz0             = _mm256_add_pd(fiz0,tz);
1596
1597             fjx1             = _mm256_add_pd(fjx1,tx);
1598             fjy1             = _mm256_add_pd(fjy1,ty);
1599             fjz1             = _mm256_add_pd(fjz1,tz);
1600
1601             /**************************
1602              * CALCULATE INTERACTIONS *
1603              **************************/
1604
1605             r02              = _mm256_mul_pd(rsq02,rinv02);
1606
1607             /* Calculate table index by multiplying r with table scale and truncate to integer */
1608             rt               = _mm256_mul_pd(r02,vftabscale);
1609             vfitab           = _mm256_cvttpd_epi32(rt);
1610             vfeps            = _mm256_sub_pd(rt,_mm256_round_pd(rt, _MM_FROUND_FLOOR));
1611             vfitab           = _mm_slli_epi32(vfitab,2);
1612
1613             /* CUBIC SPLINE TABLE ELECTROSTATICS */
1614             Y                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
1615             F                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
1616             G                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,2) );
1617             H                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,3) );
1618             GMX_MM256_FULLTRANSPOSE4_PD(Y,F,G,H);
1619             Heps             = _mm256_mul_pd(vfeps,H);
1620             Fp               = _mm256_add_pd(F,_mm256_mul_pd(vfeps,_mm256_add_pd(G,Heps)));
1621             FF               = _mm256_add_pd(Fp,_mm256_mul_pd(vfeps,_mm256_add_pd(G,_mm256_add_pd(Heps,Heps))));
1622             felec            = _mm256_xor_pd(signbit,_mm256_mul_pd(_mm256_mul_pd(qq02,FF),_mm256_mul_pd(vftabscale,rinv02)));
1623
1624             fscal            = felec;
1625
1626             /* Calculate temporary vectorial force */
1627             tx               = _mm256_mul_pd(fscal,dx02);
1628             ty               = _mm256_mul_pd(fscal,dy02);
1629             tz               = _mm256_mul_pd(fscal,dz02);
1630
1631             /* Update vectorial force */
1632             fix0             = _mm256_add_pd(fix0,tx);
1633             fiy0             = _mm256_add_pd(fiy0,ty);
1634             fiz0             = _mm256_add_pd(fiz0,tz);
1635
1636             fjx2             = _mm256_add_pd(fjx2,tx);
1637             fjy2             = _mm256_add_pd(fjy2,ty);
1638             fjz2             = _mm256_add_pd(fjz2,tz);
1639
1640             /**************************
1641              * CALCULATE INTERACTIONS *
1642              **************************/
1643
1644             r10              = _mm256_mul_pd(rsq10,rinv10);
1645
1646             /* Calculate table index by multiplying r with table scale and truncate to integer */
1647             rt               = _mm256_mul_pd(r10,vftabscale);
1648             vfitab           = _mm256_cvttpd_epi32(rt);
1649             vfeps            = _mm256_sub_pd(rt,_mm256_round_pd(rt, _MM_FROUND_FLOOR));
1650             vfitab           = _mm_slli_epi32(vfitab,2);
1651
1652             /* CUBIC SPLINE TABLE ELECTROSTATICS */
1653             Y                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
1654             F                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
1655             G                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,2) );
1656             H                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,3) );
1657             GMX_MM256_FULLTRANSPOSE4_PD(Y,F,G,H);
1658             Heps             = _mm256_mul_pd(vfeps,H);
1659             Fp               = _mm256_add_pd(F,_mm256_mul_pd(vfeps,_mm256_add_pd(G,Heps)));
1660             FF               = _mm256_add_pd(Fp,_mm256_mul_pd(vfeps,_mm256_add_pd(G,_mm256_add_pd(Heps,Heps))));
1661             felec            = _mm256_xor_pd(signbit,_mm256_mul_pd(_mm256_mul_pd(qq10,FF),_mm256_mul_pd(vftabscale,rinv10)));
1662
1663             fscal            = felec;
1664
1665             /* Calculate temporary vectorial force */
1666             tx               = _mm256_mul_pd(fscal,dx10);
1667             ty               = _mm256_mul_pd(fscal,dy10);
1668             tz               = _mm256_mul_pd(fscal,dz10);
1669
1670             /* Update vectorial force */
1671             fix1             = _mm256_add_pd(fix1,tx);
1672             fiy1             = _mm256_add_pd(fiy1,ty);
1673             fiz1             = _mm256_add_pd(fiz1,tz);
1674
1675             fjx0             = _mm256_add_pd(fjx0,tx);
1676             fjy0             = _mm256_add_pd(fjy0,ty);
1677             fjz0             = _mm256_add_pd(fjz0,tz);
1678
1679             /**************************
1680              * CALCULATE INTERACTIONS *
1681              **************************/
1682
1683             r11              = _mm256_mul_pd(rsq11,rinv11);
1684
1685             /* Calculate table index by multiplying r with table scale and truncate to integer */
1686             rt               = _mm256_mul_pd(r11,vftabscale);
1687             vfitab           = _mm256_cvttpd_epi32(rt);
1688             vfeps            = _mm256_sub_pd(rt,_mm256_round_pd(rt, _MM_FROUND_FLOOR));
1689             vfitab           = _mm_slli_epi32(vfitab,2);
1690
1691             /* CUBIC SPLINE TABLE ELECTROSTATICS */
1692             Y                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
1693             F                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
1694             G                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,2) );
1695             H                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,3) );
1696             GMX_MM256_FULLTRANSPOSE4_PD(Y,F,G,H);
1697             Heps             = _mm256_mul_pd(vfeps,H);
1698             Fp               = _mm256_add_pd(F,_mm256_mul_pd(vfeps,_mm256_add_pd(G,Heps)));
1699             FF               = _mm256_add_pd(Fp,_mm256_mul_pd(vfeps,_mm256_add_pd(G,_mm256_add_pd(Heps,Heps))));
1700             felec            = _mm256_xor_pd(signbit,_mm256_mul_pd(_mm256_mul_pd(qq11,FF),_mm256_mul_pd(vftabscale,rinv11)));
1701
1702             fscal            = felec;
1703
1704             /* Calculate temporary vectorial force */
1705             tx               = _mm256_mul_pd(fscal,dx11);
1706             ty               = _mm256_mul_pd(fscal,dy11);
1707             tz               = _mm256_mul_pd(fscal,dz11);
1708
1709             /* Update vectorial force */
1710             fix1             = _mm256_add_pd(fix1,tx);
1711             fiy1             = _mm256_add_pd(fiy1,ty);
1712             fiz1             = _mm256_add_pd(fiz1,tz);
1713
1714             fjx1             = _mm256_add_pd(fjx1,tx);
1715             fjy1             = _mm256_add_pd(fjy1,ty);
1716             fjz1             = _mm256_add_pd(fjz1,tz);
1717
1718             /**************************
1719              * CALCULATE INTERACTIONS *
1720              **************************/
1721
1722             r12              = _mm256_mul_pd(rsq12,rinv12);
1723
1724             /* Calculate table index by multiplying r with table scale and truncate to integer */
1725             rt               = _mm256_mul_pd(r12,vftabscale);
1726             vfitab           = _mm256_cvttpd_epi32(rt);
1727             vfeps            = _mm256_sub_pd(rt,_mm256_round_pd(rt, _MM_FROUND_FLOOR));
1728             vfitab           = _mm_slli_epi32(vfitab,2);
1729
1730             /* CUBIC SPLINE TABLE ELECTROSTATICS */
1731             Y                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
1732             F                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
1733             G                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,2) );
1734             H                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,3) );
1735             GMX_MM256_FULLTRANSPOSE4_PD(Y,F,G,H);
1736             Heps             = _mm256_mul_pd(vfeps,H);
1737             Fp               = _mm256_add_pd(F,_mm256_mul_pd(vfeps,_mm256_add_pd(G,Heps)));
1738             FF               = _mm256_add_pd(Fp,_mm256_mul_pd(vfeps,_mm256_add_pd(G,_mm256_add_pd(Heps,Heps))));
1739             felec            = _mm256_xor_pd(signbit,_mm256_mul_pd(_mm256_mul_pd(qq12,FF),_mm256_mul_pd(vftabscale,rinv12)));
1740
1741             fscal            = felec;
1742
1743             /* Calculate temporary vectorial force */
1744             tx               = _mm256_mul_pd(fscal,dx12);
1745             ty               = _mm256_mul_pd(fscal,dy12);
1746             tz               = _mm256_mul_pd(fscal,dz12);
1747
1748             /* Update vectorial force */
1749             fix1             = _mm256_add_pd(fix1,tx);
1750             fiy1             = _mm256_add_pd(fiy1,ty);
1751             fiz1             = _mm256_add_pd(fiz1,tz);
1752
1753             fjx2             = _mm256_add_pd(fjx2,tx);
1754             fjy2             = _mm256_add_pd(fjy2,ty);
1755             fjz2             = _mm256_add_pd(fjz2,tz);
1756
1757             /**************************
1758              * CALCULATE INTERACTIONS *
1759              **************************/
1760
1761             r20              = _mm256_mul_pd(rsq20,rinv20);
1762
1763             /* Calculate table index by multiplying r with table scale and truncate to integer */
1764             rt               = _mm256_mul_pd(r20,vftabscale);
1765             vfitab           = _mm256_cvttpd_epi32(rt);
1766             vfeps            = _mm256_sub_pd(rt,_mm256_round_pd(rt, _MM_FROUND_FLOOR));
1767             vfitab           = _mm_slli_epi32(vfitab,2);
1768
1769             /* CUBIC SPLINE TABLE ELECTROSTATICS */
1770             Y                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
1771             F                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
1772             G                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,2) );
1773             H                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,3) );
1774             GMX_MM256_FULLTRANSPOSE4_PD(Y,F,G,H);
1775             Heps             = _mm256_mul_pd(vfeps,H);
1776             Fp               = _mm256_add_pd(F,_mm256_mul_pd(vfeps,_mm256_add_pd(G,Heps)));
1777             FF               = _mm256_add_pd(Fp,_mm256_mul_pd(vfeps,_mm256_add_pd(G,_mm256_add_pd(Heps,Heps))));
1778             felec            = _mm256_xor_pd(signbit,_mm256_mul_pd(_mm256_mul_pd(qq20,FF),_mm256_mul_pd(vftabscale,rinv20)));
1779
1780             fscal            = felec;
1781
1782             /* Calculate temporary vectorial force */
1783             tx               = _mm256_mul_pd(fscal,dx20);
1784             ty               = _mm256_mul_pd(fscal,dy20);
1785             tz               = _mm256_mul_pd(fscal,dz20);
1786
1787             /* Update vectorial force */
1788             fix2             = _mm256_add_pd(fix2,tx);
1789             fiy2             = _mm256_add_pd(fiy2,ty);
1790             fiz2             = _mm256_add_pd(fiz2,tz);
1791
1792             fjx0             = _mm256_add_pd(fjx0,tx);
1793             fjy0             = _mm256_add_pd(fjy0,ty);
1794             fjz0             = _mm256_add_pd(fjz0,tz);
1795
1796             /**************************
1797              * CALCULATE INTERACTIONS *
1798              **************************/
1799
1800             r21              = _mm256_mul_pd(rsq21,rinv21);
1801
1802             /* Calculate table index by multiplying r with table scale and truncate to integer */
1803             rt               = _mm256_mul_pd(r21,vftabscale);
1804             vfitab           = _mm256_cvttpd_epi32(rt);
1805             vfeps            = _mm256_sub_pd(rt,_mm256_round_pd(rt, _MM_FROUND_FLOOR));
1806             vfitab           = _mm_slli_epi32(vfitab,2);
1807
1808             /* CUBIC SPLINE TABLE ELECTROSTATICS */
1809             Y                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
1810             F                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
1811             G                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,2) );
1812             H                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,3) );
1813             GMX_MM256_FULLTRANSPOSE4_PD(Y,F,G,H);
1814             Heps             = _mm256_mul_pd(vfeps,H);
1815             Fp               = _mm256_add_pd(F,_mm256_mul_pd(vfeps,_mm256_add_pd(G,Heps)));
1816             FF               = _mm256_add_pd(Fp,_mm256_mul_pd(vfeps,_mm256_add_pd(G,_mm256_add_pd(Heps,Heps))));
1817             felec            = _mm256_xor_pd(signbit,_mm256_mul_pd(_mm256_mul_pd(qq21,FF),_mm256_mul_pd(vftabscale,rinv21)));
1818
1819             fscal            = felec;
1820
1821             /* Calculate temporary vectorial force */
1822             tx               = _mm256_mul_pd(fscal,dx21);
1823             ty               = _mm256_mul_pd(fscal,dy21);
1824             tz               = _mm256_mul_pd(fscal,dz21);
1825
1826             /* Update vectorial force */
1827             fix2             = _mm256_add_pd(fix2,tx);
1828             fiy2             = _mm256_add_pd(fiy2,ty);
1829             fiz2             = _mm256_add_pd(fiz2,tz);
1830
1831             fjx1             = _mm256_add_pd(fjx1,tx);
1832             fjy1             = _mm256_add_pd(fjy1,ty);
1833             fjz1             = _mm256_add_pd(fjz1,tz);
1834
1835             /**************************
1836              * CALCULATE INTERACTIONS *
1837              **************************/
1838
1839             r22              = _mm256_mul_pd(rsq22,rinv22);
1840
1841             /* Calculate table index by multiplying r with table scale and truncate to integer */
1842             rt               = _mm256_mul_pd(r22,vftabscale);
1843             vfitab           = _mm256_cvttpd_epi32(rt);
1844             vfeps            = _mm256_sub_pd(rt,_mm256_round_pd(rt, _MM_FROUND_FLOOR));
1845             vfitab           = _mm_slli_epi32(vfitab,2);
1846
1847             /* CUBIC SPLINE TABLE ELECTROSTATICS */
1848             Y                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
1849             F                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
1850             G                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,2) );
1851             H                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,3) );
1852             GMX_MM256_FULLTRANSPOSE4_PD(Y,F,G,H);
1853             Heps             = _mm256_mul_pd(vfeps,H);
1854             Fp               = _mm256_add_pd(F,_mm256_mul_pd(vfeps,_mm256_add_pd(G,Heps)));
1855             FF               = _mm256_add_pd(Fp,_mm256_mul_pd(vfeps,_mm256_add_pd(G,_mm256_add_pd(Heps,Heps))));
1856             felec            = _mm256_xor_pd(signbit,_mm256_mul_pd(_mm256_mul_pd(qq22,FF),_mm256_mul_pd(vftabscale,rinv22)));
1857
1858             fscal            = felec;
1859
1860             /* Calculate temporary vectorial force */
1861             tx               = _mm256_mul_pd(fscal,dx22);
1862             ty               = _mm256_mul_pd(fscal,dy22);
1863             tz               = _mm256_mul_pd(fscal,dz22);
1864
1865             /* Update vectorial force */
1866             fix2             = _mm256_add_pd(fix2,tx);
1867             fiy2             = _mm256_add_pd(fiy2,ty);
1868             fiz2             = _mm256_add_pd(fiz2,tz);
1869
1870             fjx2             = _mm256_add_pd(fjx2,tx);
1871             fjy2             = _mm256_add_pd(fjy2,ty);
1872             fjz2             = _mm256_add_pd(fjz2,tz);
1873
1874             fjptrA             = f+j_coord_offsetA;
1875             fjptrB             = f+j_coord_offsetB;
1876             fjptrC             = f+j_coord_offsetC;
1877             fjptrD             = f+j_coord_offsetD;
1878
1879             gmx_mm256_decrement_3rvec_4ptr_swizzle_pd(fjptrA,fjptrB,fjptrC,fjptrD,
1880                                                       fjx0,fjy0,fjz0,fjx1,fjy1,fjz1,fjx2,fjy2,fjz2);
1881
1882             /* Inner loop uses 359 flops */
1883         }
1884
1885         if(jidx<j_index_end)
1886         {
1887
1888             /* Get j neighbor index, and coordinate index */
1889             jnrlistA         = jjnr[jidx];
1890             jnrlistB         = jjnr[jidx+1];
1891             jnrlistC         = jjnr[jidx+2];
1892             jnrlistD         = jjnr[jidx+3];
1893             /* Sign of each element will be negative for non-real atoms.
1894              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
1895              * so use it as val = _mm_andnot_pd(mask,val) to clear dummy entries.
1896              */
1897             tmpmask0 = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
1898
1899             tmpmask1 = _mm_permute_ps(tmpmask0,_GMX_MM_PERMUTE(3,3,2,2));
1900             tmpmask0 = _mm_permute_ps(tmpmask0,_GMX_MM_PERMUTE(1,1,0,0));
1901             dummy_mask = _mm256_castps_pd(gmx_mm256_set_m128(tmpmask1,tmpmask0));
1902
1903             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
1904             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
1905             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
1906             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
1907             j_coord_offsetA  = DIM*jnrA;
1908             j_coord_offsetB  = DIM*jnrB;
1909             j_coord_offsetC  = DIM*jnrC;
1910             j_coord_offsetD  = DIM*jnrD;
1911
1912             /* load j atom coordinates */
1913             gmx_mm256_load_3rvec_4ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
1914                                                  x+j_coord_offsetC,x+j_coord_offsetD,
1915                                               &jx0,&jy0,&jz0,&jx1,&jy1,&jz1,&jx2,&jy2,&jz2);
1916
1917             /* Calculate displacement vector */
1918             dx00             = _mm256_sub_pd(ix0,jx0);
1919             dy00             = _mm256_sub_pd(iy0,jy0);
1920             dz00             = _mm256_sub_pd(iz0,jz0);
1921             dx01             = _mm256_sub_pd(ix0,jx1);
1922             dy01             = _mm256_sub_pd(iy0,jy1);
1923             dz01             = _mm256_sub_pd(iz0,jz1);
1924             dx02             = _mm256_sub_pd(ix0,jx2);
1925             dy02             = _mm256_sub_pd(iy0,jy2);
1926             dz02             = _mm256_sub_pd(iz0,jz2);
1927             dx10             = _mm256_sub_pd(ix1,jx0);
1928             dy10             = _mm256_sub_pd(iy1,jy0);
1929             dz10             = _mm256_sub_pd(iz1,jz0);
1930             dx11             = _mm256_sub_pd(ix1,jx1);
1931             dy11             = _mm256_sub_pd(iy1,jy1);
1932             dz11             = _mm256_sub_pd(iz1,jz1);
1933             dx12             = _mm256_sub_pd(ix1,jx2);
1934             dy12             = _mm256_sub_pd(iy1,jy2);
1935             dz12             = _mm256_sub_pd(iz1,jz2);
1936             dx20             = _mm256_sub_pd(ix2,jx0);
1937             dy20             = _mm256_sub_pd(iy2,jy0);
1938             dz20             = _mm256_sub_pd(iz2,jz0);
1939             dx21             = _mm256_sub_pd(ix2,jx1);
1940             dy21             = _mm256_sub_pd(iy2,jy1);
1941             dz21             = _mm256_sub_pd(iz2,jz1);
1942             dx22             = _mm256_sub_pd(ix2,jx2);
1943             dy22             = _mm256_sub_pd(iy2,jy2);
1944             dz22             = _mm256_sub_pd(iz2,jz2);
1945
1946             /* Calculate squared distance and things based on it */
1947             rsq00            = gmx_mm256_calc_rsq_pd(dx00,dy00,dz00);
1948             rsq01            = gmx_mm256_calc_rsq_pd(dx01,dy01,dz01);
1949             rsq02            = gmx_mm256_calc_rsq_pd(dx02,dy02,dz02);
1950             rsq10            = gmx_mm256_calc_rsq_pd(dx10,dy10,dz10);
1951             rsq11            = gmx_mm256_calc_rsq_pd(dx11,dy11,dz11);
1952             rsq12            = gmx_mm256_calc_rsq_pd(dx12,dy12,dz12);
1953             rsq20            = gmx_mm256_calc_rsq_pd(dx20,dy20,dz20);
1954             rsq21            = gmx_mm256_calc_rsq_pd(dx21,dy21,dz21);
1955             rsq22            = gmx_mm256_calc_rsq_pd(dx22,dy22,dz22);
1956
1957             rinv00           = gmx_mm256_invsqrt_pd(rsq00);
1958             rinv01           = gmx_mm256_invsqrt_pd(rsq01);
1959             rinv02           = gmx_mm256_invsqrt_pd(rsq02);
1960             rinv10           = gmx_mm256_invsqrt_pd(rsq10);
1961             rinv11           = gmx_mm256_invsqrt_pd(rsq11);
1962             rinv12           = gmx_mm256_invsqrt_pd(rsq12);
1963             rinv20           = gmx_mm256_invsqrt_pd(rsq20);
1964             rinv21           = gmx_mm256_invsqrt_pd(rsq21);
1965             rinv22           = gmx_mm256_invsqrt_pd(rsq22);
1966
1967             rinvsq00         = _mm256_mul_pd(rinv00,rinv00);
1968
1969             fjx0             = _mm256_setzero_pd();
1970             fjy0             = _mm256_setzero_pd();
1971             fjz0             = _mm256_setzero_pd();
1972             fjx1             = _mm256_setzero_pd();
1973             fjy1             = _mm256_setzero_pd();
1974             fjz1             = _mm256_setzero_pd();
1975             fjx2             = _mm256_setzero_pd();
1976             fjy2             = _mm256_setzero_pd();
1977             fjz2             = _mm256_setzero_pd();
1978
1979             /**************************
1980              * CALCULATE INTERACTIONS *
1981              **************************/
1982
1983             r00              = _mm256_mul_pd(rsq00,rinv00);
1984             r00              = _mm256_andnot_pd(dummy_mask,r00);
1985
1986             /* Calculate table index by multiplying r with table scale and truncate to integer */
1987             rt               = _mm256_mul_pd(r00,vftabscale);
1988             vfitab           = _mm256_cvttpd_epi32(rt);
1989             vfeps            = _mm256_sub_pd(rt,_mm256_round_pd(rt, _MM_FROUND_FLOOR));
1990             vfitab           = _mm_slli_epi32(vfitab,2);
1991
1992             /* CUBIC SPLINE TABLE ELECTROSTATICS */
1993             Y                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
1994             F                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
1995             G                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,2) );
1996             H                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,3) );
1997             GMX_MM256_FULLTRANSPOSE4_PD(Y,F,G,H);
1998             Heps             = _mm256_mul_pd(vfeps,H);
1999             Fp               = _mm256_add_pd(F,_mm256_mul_pd(vfeps,_mm256_add_pd(G,Heps)));
2000             FF               = _mm256_add_pd(Fp,_mm256_mul_pd(vfeps,_mm256_add_pd(G,_mm256_add_pd(Heps,Heps))));
2001             felec            = _mm256_xor_pd(signbit,_mm256_mul_pd(_mm256_mul_pd(qq00,FF),_mm256_mul_pd(vftabscale,rinv00)));
2002
2003             /* LENNARD-JONES DISPERSION/REPULSION */
2004
2005             rinvsix          = _mm256_mul_pd(_mm256_mul_pd(rinvsq00,rinvsq00),rinvsq00);
2006             fvdw             = _mm256_mul_pd(_mm256_sub_pd(_mm256_mul_pd(c12_00,rinvsix),c6_00),_mm256_mul_pd(rinvsix,rinvsq00));
2007
2008             fscal            = _mm256_add_pd(felec,fvdw);
2009
2010             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
2011
2012             /* Calculate temporary vectorial force */
2013             tx               = _mm256_mul_pd(fscal,dx00);
2014             ty               = _mm256_mul_pd(fscal,dy00);
2015             tz               = _mm256_mul_pd(fscal,dz00);
2016
2017             /* Update vectorial force */
2018             fix0             = _mm256_add_pd(fix0,tx);
2019             fiy0             = _mm256_add_pd(fiy0,ty);
2020             fiz0             = _mm256_add_pd(fiz0,tz);
2021
2022             fjx0             = _mm256_add_pd(fjx0,tx);
2023             fjy0             = _mm256_add_pd(fjy0,ty);
2024             fjz0             = _mm256_add_pd(fjz0,tz);
2025
2026             /**************************
2027              * CALCULATE INTERACTIONS *
2028              **************************/
2029
2030             r01              = _mm256_mul_pd(rsq01,rinv01);
2031             r01              = _mm256_andnot_pd(dummy_mask,r01);
2032
2033             /* Calculate table index by multiplying r with table scale and truncate to integer */
2034             rt               = _mm256_mul_pd(r01,vftabscale);
2035             vfitab           = _mm256_cvttpd_epi32(rt);
2036             vfeps            = _mm256_sub_pd(rt,_mm256_round_pd(rt, _MM_FROUND_FLOOR));
2037             vfitab           = _mm_slli_epi32(vfitab,2);
2038
2039             /* CUBIC SPLINE TABLE ELECTROSTATICS */
2040             Y                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
2041             F                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
2042             G                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,2) );
2043             H                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,3) );
2044             GMX_MM256_FULLTRANSPOSE4_PD(Y,F,G,H);
2045             Heps             = _mm256_mul_pd(vfeps,H);
2046             Fp               = _mm256_add_pd(F,_mm256_mul_pd(vfeps,_mm256_add_pd(G,Heps)));
2047             FF               = _mm256_add_pd(Fp,_mm256_mul_pd(vfeps,_mm256_add_pd(G,_mm256_add_pd(Heps,Heps))));
2048             felec            = _mm256_xor_pd(signbit,_mm256_mul_pd(_mm256_mul_pd(qq01,FF),_mm256_mul_pd(vftabscale,rinv01)));
2049
2050             fscal            = felec;
2051
2052             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
2053
2054             /* Calculate temporary vectorial force */
2055             tx               = _mm256_mul_pd(fscal,dx01);
2056             ty               = _mm256_mul_pd(fscal,dy01);
2057             tz               = _mm256_mul_pd(fscal,dz01);
2058
2059             /* Update vectorial force */
2060             fix0             = _mm256_add_pd(fix0,tx);
2061             fiy0             = _mm256_add_pd(fiy0,ty);
2062             fiz0             = _mm256_add_pd(fiz0,tz);
2063
2064             fjx1             = _mm256_add_pd(fjx1,tx);
2065             fjy1             = _mm256_add_pd(fjy1,ty);
2066             fjz1             = _mm256_add_pd(fjz1,tz);
2067
2068             /**************************
2069              * CALCULATE INTERACTIONS *
2070              **************************/
2071
2072             r02              = _mm256_mul_pd(rsq02,rinv02);
2073             r02              = _mm256_andnot_pd(dummy_mask,r02);
2074
2075             /* Calculate table index by multiplying r with table scale and truncate to integer */
2076             rt               = _mm256_mul_pd(r02,vftabscale);
2077             vfitab           = _mm256_cvttpd_epi32(rt);
2078             vfeps            = _mm256_sub_pd(rt,_mm256_round_pd(rt, _MM_FROUND_FLOOR));
2079             vfitab           = _mm_slli_epi32(vfitab,2);
2080
2081             /* CUBIC SPLINE TABLE ELECTROSTATICS */
2082             Y                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
2083             F                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
2084             G                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,2) );
2085             H                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,3) );
2086             GMX_MM256_FULLTRANSPOSE4_PD(Y,F,G,H);
2087             Heps             = _mm256_mul_pd(vfeps,H);
2088             Fp               = _mm256_add_pd(F,_mm256_mul_pd(vfeps,_mm256_add_pd(G,Heps)));
2089             FF               = _mm256_add_pd(Fp,_mm256_mul_pd(vfeps,_mm256_add_pd(G,_mm256_add_pd(Heps,Heps))));
2090             felec            = _mm256_xor_pd(signbit,_mm256_mul_pd(_mm256_mul_pd(qq02,FF),_mm256_mul_pd(vftabscale,rinv02)));
2091
2092             fscal            = felec;
2093
2094             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
2095
2096             /* Calculate temporary vectorial force */
2097             tx               = _mm256_mul_pd(fscal,dx02);
2098             ty               = _mm256_mul_pd(fscal,dy02);
2099             tz               = _mm256_mul_pd(fscal,dz02);
2100
2101             /* Update vectorial force */
2102             fix0             = _mm256_add_pd(fix0,tx);
2103             fiy0             = _mm256_add_pd(fiy0,ty);
2104             fiz0             = _mm256_add_pd(fiz0,tz);
2105
2106             fjx2             = _mm256_add_pd(fjx2,tx);
2107             fjy2             = _mm256_add_pd(fjy2,ty);
2108             fjz2             = _mm256_add_pd(fjz2,tz);
2109
2110             /**************************
2111              * CALCULATE INTERACTIONS *
2112              **************************/
2113
2114             r10              = _mm256_mul_pd(rsq10,rinv10);
2115             r10              = _mm256_andnot_pd(dummy_mask,r10);
2116
2117             /* Calculate table index by multiplying r with table scale and truncate to integer */
2118             rt               = _mm256_mul_pd(r10,vftabscale);
2119             vfitab           = _mm256_cvttpd_epi32(rt);
2120             vfeps            = _mm256_sub_pd(rt,_mm256_round_pd(rt, _MM_FROUND_FLOOR));
2121             vfitab           = _mm_slli_epi32(vfitab,2);
2122
2123             /* CUBIC SPLINE TABLE ELECTROSTATICS */
2124             Y                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
2125             F                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
2126             G                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,2) );
2127             H                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,3) );
2128             GMX_MM256_FULLTRANSPOSE4_PD(Y,F,G,H);
2129             Heps             = _mm256_mul_pd(vfeps,H);
2130             Fp               = _mm256_add_pd(F,_mm256_mul_pd(vfeps,_mm256_add_pd(G,Heps)));
2131             FF               = _mm256_add_pd(Fp,_mm256_mul_pd(vfeps,_mm256_add_pd(G,_mm256_add_pd(Heps,Heps))));
2132             felec            = _mm256_xor_pd(signbit,_mm256_mul_pd(_mm256_mul_pd(qq10,FF),_mm256_mul_pd(vftabscale,rinv10)));
2133
2134             fscal            = felec;
2135
2136             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
2137
2138             /* Calculate temporary vectorial force */
2139             tx               = _mm256_mul_pd(fscal,dx10);
2140             ty               = _mm256_mul_pd(fscal,dy10);
2141             tz               = _mm256_mul_pd(fscal,dz10);
2142
2143             /* Update vectorial force */
2144             fix1             = _mm256_add_pd(fix1,tx);
2145             fiy1             = _mm256_add_pd(fiy1,ty);
2146             fiz1             = _mm256_add_pd(fiz1,tz);
2147
2148             fjx0             = _mm256_add_pd(fjx0,tx);
2149             fjy0             = _mm256_add_pd(fjy0,ty);
2150             fjz0             = _mm256_add_pd(fjz0,tz);
2151
2152             /**************************
2153              * CALCULATE INTERACTIONS *
2154              **************************/
2155
2156             r11              = _mm256_mul_pd(rsq11,rinv11);
2157             r11              = _mm256_andnot_pd(dummy_mask,r11);
2158
2159             /* Calculate table index by multiplying r with table scale and truncate to integer */
2160             rt               = _mm256_mul_pd(r11,vftabscale);
2161             vfitab           = _mm256_cvttpd_epi32(rt);
2162             vfeps            = _mm256_sub_pd(rt,_mm256_round_pd(rt, _MM_FROUND_FLOOR));
2163             vfitab           = _mm_slli_epi32(vfitab,2);
2164
2165             /* CUBIC SPLINE TABLE ELECTROSTATICS */
2166             Y                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
2167             F                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
2168             G                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,2) );
2169             H                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,3) );
2170             GMX_MM256_FULLTRANSPOSE4_PD(Y,F,G,H);
2171             Heps             = _mm256_mul_pd(vfeps,H);
2172             Fp               = _mm256_add_pd(F,_mm256_mul_pd(vfeps,_mm256_add_pd(G,Heps)));
2173             FF               = _mm256_add_pd(Fp,_mm256_mul_pd(vfeps,_mm256_add_pd(G,_mm256_add_pd(Heps,Heps))));
2174             felec            = _mm256_xor_pd(signbit,_mm256_mul_pd(_mm256_mul_pd(qq11,FF),_mm256_mul_pd(vftabscale,rinv11)));
2175
2176             fscal            = felec;
2177
2178             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
2179
2180             /* Calculate temporary vectorial force */
2181             tx               = _mm256_mul_pd(fscal,dx11);
2182             ty               = _mm256_mul_pd(fscal,dy11);
2183             tz               = _mm256_mul_pd(fscal,dz11);
2184
2185             /* Update vectorial force */
2186             fix1             = _mm256_add_pd(fix1,tx);
2187             fiy1             = _mm256_add_pd(fiy1,ty);
2188             fiz1             = _mm256_add_pd(fiz1,tz);
2189
2190             fjx1             = _mm256_add_pd(fjx1,tx);
2191             fjy1             = _mm256_add_pd(fjy1,ty);
2192             fjz1             = _mm256_add_pd(fjz1,tz);
2193
2194             /**************************
2195              * CALCULATE INTERACTIONS *
2196              **************************/
2197
2198             r12              = _mm256_mul_pd(rsq12,rinv12);
2199             r12              = _mm256_andnot_pd(dummy_mask,r12);
2200
2201             /* Calculate table index by multiplying r with table scale and truncate to integer */
2202             rt               = _mm256_mul_pd(r12,vftabscale);
2203             vfitab           = _mm256_cvttpd_epi32(rt);
2204             vfeps            = _mm256_sub_pd(rt,_mm256_round_pd(rt, _MM_FROUND_FLOOR));
2205             vfitab           = _mm_slli_epi32(vfitab,2);
2206
2207             /* CUBIC SPLINE TABLE ELECTROSTATICS */
2208             Y                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
2209             F                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
2210             G                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,2) );
2211             H                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,3) );
2212             GMX_MM256_FULLTRANSPOSE4_PD(Y,F,G,H);
2213             Heps             = _mm256_mul_pd(vfeps,H);
2214             Fp               = _mm256_add_pd(F,_mm256_mul_pd(vfeps,_mm256_add_pd(G,Heps)));
2215             FF               = _mm256_add_pd(Fp,_mm256_mul_pd(vfeps,_mm256_add_pd(G,_mm256_add_pd(Heps,Heps))));
2216             felec            = _mm256_xor_pd(signbit,_mm256_mul_pd(_mm256_mul_pd(qq12,FF),_mm256_mul_pd(vftabscale,rinv12)));
2217
2218             fscal            = felec;
2219
2220             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
2221
2222             /* Calculate temporary vectorial force */
2223             tx               = _mm256_mul_pd(fscal,dx12);
2224             ty               = _mm256_mul_pd(fscal,dy12);
2225             tz               = _mm256_mul_pd(fscal,dz12);
2226
2227             /* Update vectorial force */
2228             fix1             = _mm256_add_pd(fix1,tx);
2229             fiy1             = _mm256_add_pd(fiy1,ty);
2230             fiz1             = _mm256_add_pd(fiz1,tz);
2231
2232             fjx2             = _mm256_add_pd(fjx2,tx);
2233             fjy2             = _mm256_add_pd(fjy2,ty);
2234             fjz2             = _mm256_add_pd(fjz2,tz);
2235
2236             /**************************
2237              * CALCULATE INTERACTIONS *
2238              **************************/
2239
2240             r20              = _mm256_mul_pd(rsq20,rinv20);
2241             r20              = _mm256_andnot_pd(dummy_mask,r20);
2242
2243             /* Calculate table index by multiplying r with table scale and truncate to integer */
2244             rt               = _mm256_mul_pd(r20,vftabscale);
2245             vfitab           = _mm256_cvttpd_epi32(rt);
2246             vfeps            = _mm256_sub_pd(rt,_mm256_round_pd(rt, _MM_FROUND_FLOOR));
2247             vfitab           = _mm_slli_epi32(vfitab,2);
2248
2249             /* CUBIC SPLINE TABLE ELECTROSTATICS */
2250             Y                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
2251             F                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
2252             G                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,2) );
2253             H                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,3) );
2254             GMX_MM256_FULLTRANSPOSE4_PD(Y,F,G,H);
2255             Heps             = _mm256_mul_pd(vfeps,H);
2256             Fp               = _mm256_add_pd(F,_mm256_mul_pd(vfeps,_mm256_add_pd(G,Heps)));
2257             FF               = _mm256_add_pd(Fp,_mm256_mul_pd(vfeps,_mm256_add_pd(G,_mm256_add_pd(Heps,Heps))));
2258             felec            = _mm256_xor_pd(signbit,_mm256_mul_pd(_mm256_mul_pd(qq20,FF),_mm256_mul_pd(vftabscale,rinv20)));
2259
2260             fscal            = felec;
2261
2262             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
2263
2264             /* Calculate temporary vectorial force */
2265             tx               = _mm256_mul_pd(fscal,dx20);
2266             ty               = _mm256_mul_pd(fscal,dy20);
2267             tz               = _mm256_mul_pd(fscal,dz20);
2268
2269             /* Update vectorial force */
2270             fix2             = _mm256_add_pd(fix2,tx);
2271             fiy2             = _mm256_add_pd(fiy2,ty);
2272             fiz2             = _mm256_add_pd(fiz2,tz);
2273
2274             fjx0             = _mm256_add_pd(fjx0,tx);
2275             fjy0             = _mm256_add_pd(fjy0,ty);
2276             fjz0             = _mm256_add_pd(fjz0,tz);
2277
2278             /**************************
2279              * CALCULATE INTERACTIONS *
2280              **************************/
2281
2282             r21              = _mm256_mul_pd(rsq21,rinv21);
2283             r21              = _mm256_andnot_pd(dummy_mask,r21);
2284
2285             /* Calculate table index by multiplying r with table scale and truncate to integer */
2286             rt               = _mm256_mul_pd(r21,vftabscale);
2287             vfitab           = _mm256_cvttpd_epi32(rt);
2288             vfeps            = _mm256_sub_pd(rt,_mm256_round_pd(rt, _MM_FROUND_FLOOR));
2289             vfitab           = _mm_slli_epi32(vfitab,2);
2290
2291             /* CUBIC SPLINE TABLE ELECTROSTATICS */
2292             Y                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
2293             F                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
2294             G                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,2) );
2295             H                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,3) );
2296             GMX_MM256_FULLTRANSPOSE4_PD(Y,F,G,H);
2297             Heps             = _mm256_mul_pd(vfeps,H);
2298             Fp               = _mm256_add_pd(F,_mm256_mul_pd(vfeps,_mm256_add_pd(G,Heps)));
2299             FF               = _mm256_add_pd(Fp,_mm256_mul_pd(vfeps,_mm256_add_pd(G,_mm256_add_pd(Heps,Heps))));
2300             felec            = _mm256_xor_pd(signbit,_mm256_mul_pd(_mm256_mul_pd(qq21,FF),_mm256_mul_pd(vftabscale,rinv21)));
2301
2302             fscal            = felec;
2303
2304             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
2305
2306             /* Calculate temporary vectorial force */
2307             tx               = _mm256_mul_pd(fscal,dx21);
2308             ty               = _mm256_mul_pd(fscal,dy21);
2309             tz               = _mm256_mul_pd(fscal,dz21);
2310
2311             /* Update vectorial force */
2312             fix2             = _mm256_add_pd(fix2,tx);
2313             fiy2             = _mm256_add_pd(fiy2,ty);
2314             fiz2             = _mm256_add_pd(fiz2,tz);
2315
2316             fjx1             = _mm256_add_pd(fjx1,tx);
2317             fjy1             = _mm256_add_pd(fjy1,ty);
2318             fjz1             = _mm256_add_pd(fjz1,tz);
2319
2320             /**************************
2321              * CALCULATE INTERACTIONS *
2322              **************************/
2323
2324             r22              = _mm256_mul_pd(rsq22,rinv22);
2325             r22              = _mm256_andnot_pd(dummy_mask,r22);
2326
2327             /* Calculate table index by multiplying r with table scale and truncate to integer */
2328             rt               = _mm256_mul_pd(r22,vftabscale);
2329             vfitab           = _mm256_cvttpd_epi32(rt);
2330             vfeps            = _mm256_sub_pd(rt,_mm256_round_pd(rt, _MM_FROUND_FLOOR));
2331             vfitab           = _mm_slli_epi32(vfitab,2);
2332
2333             /* CUBIC SPLINE TABLE ELECTROSTATICS */
2334             Y                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
2335             F                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
2336             G                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,2) );
2337             H                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,3) );
2338             GMX_MM256_FULLTRANSPOSE4_PD(Y,F,G,H);
2339             Heps             = _mm256_mul_pd(vfeps,H);
2340             Fp               = _mm256_add_pd(F,_mm256_mul_pd(vfeps,_mm256_add_pd(G,Heps)));
2341             FF               = _mm256_add_pd(Fp,_mm256_mul_pd(vfeps,_mm256_add_pd(G,_mm256_add_pd(Heps,Heps))));
2342             felec            = _mm256_xor_pd(signbit,_mm256_mul_pd(_mm256_mul_pd(qq22,FF),_mm256_mul_pd(vftabscale,rinv22)));
2343
2344             fscal            = felec;
2345
2346             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
2347
2348             /* Calculate temporary vectorial force */
2349             tx               = _mm256_mul_pd(fscal,dx22);
2350             ty               = _mm256_mul_pd(fscal,dy22);
2351             tz               = _mm256_mul_pd(fscal,dz22);
2352
2353             /* Update vectorial force */
2354             fix2             = _mm256_add_pd(fix2,tx);
2355             fiy2             = _mm256_add_pd(fiy2,ty);
2356             fiz2             = _mm256_add_pd(fiz2,tz);
2357
2358             fjx2             = _mm256_add_pd(fjx2,tx);
2359             fjy2             = _mm256_add_pd(fjy2,ty);
2360             fjz2             = _mm256_add_pd(fjz2,tz);
2361
2362             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
2363             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
2364             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
2365             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
2366
2367             gmx_mm256_decrement_3rvec_4ptr_swizzle_pd(fjptrA,fjptrB,fjptrC,fjptrD,
2368                                                       fjx0,fjy0,fjz0,fjx1,fjy1,fjz1,fjx2,fjy2,fjz2);
2369
2370             /* Inner loop uses 368 flops */
2371         }
2372
2373         /* End of innermost loop */
2374
2375         gmx_mm256_update_iforce_3atom_swizzle_pd(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,
2376                                                  f+i_coord_offset,fshift+i_shift_offset);
2377
2378         /* Increment number of inner iterations */
2379         inneriter                  += j_index_end - j_index_start;
2380
2381         /* Outer loop uses 18 flops */
2382     }
2383
2384     /* Increment number of outer iterations */
2385     outeriter        += nri;
2386
2387     /* Update outer/inner flops */
2388
2389     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W3W3_F,outeriter*18 + inneriter*368);
2390 }