src/gromacs/gmxlib/nonbonded/nb_kernel_avx_256_double/nb_kernel_ElecRFCut_VdwCSTab_GeomW4W4_avx_256_double.c

   1 /*
   2  * This file is part of the GROMACS molecular simulation package.
   3  *
   4  * Copyright (c) 2012,2013,2014, by the GROMACS development team, led by
   5  * Mark Abraham, David van der Spoel, Berk Hess, and Erik Lindahl,
   6  * and including many others, as listed in the AUTHORS file in the
   7  * top-level source directory and at http://www.gromacs.org.
   8  *
   9  * GROMACS is free software; you can redistribute it and/or
  10  * modify it under the terms of the GNU Lesser General Public License
  11  * as published by the Free Software Foundation; either version 2.1
  12  * of the License, or (at your option) any later version.
  13  *
  14  * GROMACS is distributed in the hope that it will be useful,
  15  * but WITHOUT ANY WARRANTY; without even the implied warranty of
  16  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
  17  * Lesser General Public License for more details.
  18  *
  19  * You should have received a copy of the GNU Lesser General Public
  20  * License along with GROMACS; if not, see
  21  * http://www.gnu.org/licenses, or write to the Free Software Foundation,
  22  * Inc., 51 Franklin Street, Fifth Floor, Boston, MA  02110-1301  USA.
  23  *
  24  * If you want to redistribute modifications to GROMACS, please
  25  * consider that scientific software is very special. Version
  26  * control is crucial - bugs must be traceable. We will be happy to
  27  * consider code for inclusion in the official distribution, but
  28  * derived work must not be called official GROMACS. Details are found
  29  * in the README & COPYING files - if they are missing, get the
  30  * official version at http://www.gromacs.org.
  31  *
  32  * To help us fund GROMACS development, we humbly ask that you cite
  33  * the research papers on the package. Check out http://www.gromacs.org.
  34  */
  35 /*
  36  * Note: this file was generated by the GROMACS avx_256_double kernel generator.
  37  */
  38 #include "config.h"
  39
  40 #include <math.h>
  41
  42 #include "../nb_kernel.h"
  43 #include "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_ElecRFCut_VdwCSTab_GeomW4W4_VF_avx_256_double
  52  * Electrostatics interaction: ReactionField
  53  * VdW interaction:            CubicSplineTable
  54  * Geometry:                   Water4-Water4
  55  * Calculate force/pot:        PotentialAndForce
  56  */
  57 void
  58 nb_kernel_ElecRFCut_VdwCSTab_GeomW4W4_VF_avx_256_double
  59                     (t_nblist                    * gmx_restrict       nlist,
  60                      rvec                        * gmx_restrict          xx,
  61                      rvec                        * gmx_restrict          ff,
  62                      t_forcerec                  * gmx_restrict          fr,
  63                      t_mdatoms                   * gmx_restrict     mdatoms,
  64                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
  65                      t_nrnb                      * gmx_restrict        nrnb)
  66 {
  67     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
  68      * just 0 for non-waters.
  69      * Suffixes A,B,C,D refer to j loop unrolling done with AVX, e.g. for the four different
  70      * jnr indices corresponding to data put in the four positions in the SIMD register.
  71      */
  72     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
  73     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
  74     int              jnrA,jnrB,jnrC,jnrD;
  75     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
  76     int              jnrlistE,jnrlistF,jnrlistG,jnrlistH;
  77     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
  78     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
  79     real             rcutoff_scalar;
  80     real             *shiftvec,*fshift,*x,*f;
  81     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD;
  82     real             scratch[4*DIM];
  83     __m256d          tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
  84     real *           vdwioffsetptr0;
  85     __m256d          ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
  86     real *           vdwioffsetptr1;
  87     __m256d          ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
  88     real *           vdwioffsetptr2;
  89     __m256d          ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
  90     real *           vdwioffsetptr3;
  91     __m256d          ix3,iy3,iz3,fix3,fiy3,fiz3,iq3,isai3;
  92     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
  93     __m256d          jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
  94     int              vdwjidx1A,vdwjidx1B,vdwjidx1C,vdwjidx1D;
  95     __m256d          jx1,jy1,jz1,fjx1,fjy1,fjz1,jq1,isaj1;
  96     int              vdwjidx2A,vdwjidx2B,vdwjidx2C,vdwjidx2D;
  97     __m256d          jx2,jy2,jz2,fjx2,fjy2,fjz2,jq2,isaj2;
  98     int              vdwjidx3A,vdwjidx3B,vdwjidx3C,vdwjidx3D;
  99     __m256d          jx3,jy3,jz3,fjx3,fjy3,fjz3,jq3,isaj3;
 100     __m256d          dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
 101     __m256d          dx11,dy11,dz11,rsq11,rinv11,rinvsq11,r11,qq11,c6_11,c12_11;
 102     __m256d          dx12,dy12,dz12,rsq12,rinv12,rinvsq12,r12,qq12,c6_12,c12_12;
 103     __m256d          dx13,dy13,dz13,rsq13,rinv13,rinvsq13,r13,qq13,c6_13,c12_13;
 104     __m256d          dx21,dy21,dz21,rsq21,rinv21,rinvsq21,r21,qq21,c6_21,c12_21;
 105     __m256d          dx22,dy22,dz22,rsq22,rinv22,rinvsq22,r22,qq22,c6_22,c12_22;
 106     __m256d          dx23,dy23,dz23,rsq23,rinv23,rinvsq23,r23,qq23,c6_23,c12_23;
 107     __m256d          dx31,dy31,dz31,rsq31,rinv31,rinvsq31,r31,qq31,c6_31,c12_31;
 108     __m256d          dx32,dy32,dz32,rsq32,rinv32,rinvsq32,r32,qq32,c6_32,c12_32;
 109     __m256d          dx33,dy33,dz33,rsq33,rinv33,rinvsq33,r33,qq33,c6_33,c12_33;
 110     __m256d          velec,felec,velecsum,facel,crf,krf,krf2;
 111     real             *charge;
 112     int              nvdwtype;
 113     __m256d          rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
 114     int              *vdwtype;
 115     real             *vdwparam;
 116     __m256d          one_sixth   = _mm256_set1_pd(1.0/6.0);
 117     __m256d          one_twelfth = _mm256_set1_pd(1.0/12.0);
 118     __m128i          vfitab;
 119     __m128i          ifour       = _mm_set1_epi32(4);
 120     __m256d          rt,vfeps,vftabscale,Y,F,G,H,Heps,Fp,VV,FF;
 121     real             *vftab;
 122     __m256d          dummy_mask,cutoff_mask;
 123     __m128           tmpmask0,tmpmask1;
 124     __m256d          signbit = _mm256_castsi256_pd( _mm256_set1_epi32(0x80000000) );
 125     __m256d          one     = _mm256_set1_pd(1.0);
 126     __m256d          two     = _mm256_set1_pd(2.0);
 127     x                = xx[0];
 128     f                = ff[0];
 129
 130     nri              = nlist->nri;
 131     iinr             = nlist->iinr;
 132     jindex           = nlist->jindex;
 133     jjnr             = nlist->jjnr;
 134     shiftidx         = nlist->shift;
 135     gid              = nlist->gid;
 136     shiftvec         = fr->shift_vec[0];
 137     fshift           = fr->fshift[0];
 138     facel            = _mm256_set1_pd(fr->epsfac);
 139     charge           = mdatoms->chargeA;
 140     krf              = _mm256_set1_pd(fr->ic->k_rf);
 141     krf2             = _mm256_set1_pd(fr->ic->k_rf*2.0);
 142     crf              = _mm256_set1_pd(fr->ic->c_rf);
 143     nvdwtype         = fr->ntype;
 144     vdwparam         = fr->nbfp;
 145     vdwtype          = mdatoms->typeA;
 146
 147     vftab            = kernel_data->table_vdw->data;
 148     vftabscale       = _mm256_set1_pd(kernel_data->table_vdw->scale);
 149
 150     /* Setup water-specific parameters */
 151     inr              = nlist->iinr[0];
 152     iq1              = _mm256_mul_pd(facel,_mm256_set1_pd(charge[inr+1]));
 153     iq2              = _mm256_mul_pd(facel,_mm256_set1_pd(charge[inr+2]));
 154     iq3              = _mm256_mul_pd(facel,_mm256_set1_pd(charge[inr+3]));
 155     vdwioffsetptr0   = vdwparam+2*nvdwtype*vdwtype[inr+0];
 156
 157     jq1              = _mm256_set1_pd(charge[inr+1]);
 158     jq2              = _mm256_set1_pd(charge[inr+2]);
 159     jq3              = _mm256_set1_pd(charge[inr+3]);
 160     vdwjidx0A        = 2*vdwtype[inr+0];
 161     c6_00            = _mm256_set1_pd(vdwioffsetptr0[vdwjidx0A]);
 162     c12_00           = _mm256_set1_pd(vdwioffsetptr0[vdwjidx0A+1]);
 163     qq11             = _mm256_mul_pd(iq1,jq1);
 164     qq12             = _mm256_mul_pd(iq1,jq2);
 165     qq13             = _mm256_mul_pd(iq1,jq3);
 166     qq21             = _mm256_mul_pd(iq2,jq1);
 167     qq22             = _mm256_mul_pd(iq2,jq2);
 168     qq23             = _mm256_mul_pd(iq2,jq3);
 169     qq31             = _mm256_mul_pd(iq3,jq1);
 170     qq32             = _mm256_mul_pd(iq3,jq2);
 171     qq33             = _mm256_mul_pd(iq3,jq3);
 172
 173     /* When we use explicit cutoffs the value must be identical for elec and VdW, so use elec as an arbitrary choice */
 174     rcutoff_scalar   = fr->rcoulomb;
 175     rcutoff          = _mm256_set1_pd(rcutoff_scalar);
 176     rcutoff2         = _mm256_mul_pd(rcutoff,rcutoff);
 177
 178     /* Avoid stupid compiler warnings */
 179     jnrA = jnrB = jnrC = jnrD = 0;
 180     j_coord_offsetA = 0;
 181     j_coord_offsetB = 0;
 182     j_coord_offsetC = 0;
 183     j_coord_offsetD = 0;
 184
 185     outeriter        = 0;
 186     inneriter        = 0;
 187
 188     for(iidx=0;iidx<4*DIM;iidx++)
 189     {
 190         scratch[iidx] = 0.0;
 191     }
 192
 193     /* Start outer loop over neighborlists */
 194     for(iidx=0; iidx<nri; iidx++)
 195     {
 196         /* Load shift vector for this list */
 197         i_shift_offset   = DIM*shiftidx[iidx];
 198
 199         /* Load limits for loop over neighbors */
 200         j_index_start    = jindex[iidx];
 201         j_index_end      = jindex[iidx+1];
 202
 203         /* Get outer coordinate index */
 204         inr              = iinr[iidx];
 205         i_coord_offset   = DIM*inr;
 206
 207         /* Load i particle coords and add shift vector */
 208         gmx_mm256_load_shift_and_4rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset,
 209                                                     &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2,&ix3,&iy3,&iz3);
 210
 211         fix0             = _mm256_setzero_pd();
 212         fiy0             = _mm256_setzero_pd();
 213         fiz0             = _mm256_setzero_pd();
 214         fix1             = _mm256_setzero_pd();
 215         fiy1             = _mm256_setzero_pd();
 216         fiz1             = _mm256_setzero_pd();
 217         fix2             = _mm256_setzero_pd();
 218         fiy2             = _mm256_setzero_pd();
 219         fiz2             = _mm256_setzero_pd();
 220         fix3             = _mm256_setzero_pd();
 221         fiy3             = _mm256_setzero_pd();
 222         fiz3             = _mm256_setzero_pd();
 223
 224         /* Reset potential sums */
 225         velecsum         = _mm256_setzero_pd();
 226         vvdwsum          = _mm256_setzero_pd();
 227
 228         /* Start inner kernel loop */
 229         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
 230         {
 231
 232             /* Get j neighbor index, and coordinate index */
 233             jnrA             = jjnr[jidx];
 234             jnrB             = jjnr[jidx+1];
 235             jnrC             = jjnr[jidx+2];
 236             jnrD             = jjnr[jidx+3];
 237             j_coord_offsetA  = DIM*jnrA;
 238             j_coord_offsetB  = DIM*jnrB;
 239             j_coord_offsetC  = DIM*jnrC;
 240             j_coord_offsetD  = DIM*jnrD;
 241
 242             /* load j atom coordinates */
 243             gmx_mm256_load_4rvec_4ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 244                                                  x+j_coord_offsetC,x+j_coord_offsetD,
 245                                                  &jx0,&jy0,&jz0,&jx1,&jy1,&jz1,&jx2,
 246                                                  &jy2,&jz2,&jx3,&jy3,&jz3);
 247
 248             /* Calculate displacement vector */
 249             dx00             = _mm256_sub_pd(ix0,jx0);
 250             dy00             = _mm256_sub_pd(iy0,jy0);
 251             dz00             = _mm256_sub_pd(iz0,jz0);
 252             dx11             = _mm256_sub_pd(ix1,jx1);
 253             dy11             = _mm256_sub_pd(iy1,jy1);
 254             dz11             = _mm256_sub_pd(iz1,jz1);
 255             dx12             = _mm256_sub_pd(ix1,jx2);
 256             dy12             = _mm256_sub_pd(iy1,jy2);
 257             dz12             = _mm256_sub_pd(iz1,jz2);
 258             dx13             = _mm256_sub_pd(ix1,jx3);
 259             dy13             = _mm256_sub_pd(iy1,jy3);
 260             dz13             = _mm256_sub_pd(iz1,jz3);
 261             dx21             = _mm256_sub_pd(ix2,jx1);
 262             dy21             = _mm256_sub_pd(iy2,jy1);
 263             dz21             = _mm256_sub_pd(iz2,jz1);
 264             dx22             = _mm256_sub_pd(ix2,jx2);
 265             dy22             = _mm256_sub_pd(iy2,jy2);
 266             dz22             = _mm256_sub_pd(iz2,jz2);
 267             dx23             = _mm256_sub_pd(ix2,jx3);
 268             dy23             = _mm256_sub_pd(iy2,jy3);
 269             dz23             = _mm256_sub_pd(iz2,jz3);
 270             dx31             = _mm256_sub_pd(ix3,jx1);
 271             dy31             = _mm256_sub_pd(iy3,jy1);
 272             dz31             = _mm256_sub_pd(iz3,jz1);
 273             dx32             = _mm256_sub_pd(ix3,jx2);
 274             dy32             = _mm256_sub_pd(iy3,jy2);
 275             dz32             = _mm256_sub_pd(iz3,jz2);
 276             dx33             = _mm256_sub_pd(ix3,jx3);
 277             dy33             = _mm256_sub_pd(iy3,jy3);
 278             dz33             = _mm256_sub_pd(iz3,jz3);
 279
 280             /* Calculate squared distance and things based on it */
 281             rsq00            = gmx_mm256_calc_rsq_pd(dx00,dy00,dz00);
 282             rsq11            = gmx_mm256_calc_rsq_pd(dx11,dy11,dz11);
 283             rsq12            = gmx_mm256_calc_rsq_pd(dx12,dy12,dz12);
 284             rsq13            = gmx_mm256_calc_rsq_pd(dx13,dy13,dz13);
 285             rsq21            = gmx_mm256_calc_rsq_pd(dx21,dy21,dz21);
 286             rsq22            = gmx_mm256_calc_rsq_pd(dx22,dy22,dz22);
 287             rsq23            = gmx_mm256_calc_rsq_pd(dx23,dy23,dz23);
 288             rsq31            = gmx_mm256_calc_rsq_pd(dx31,dy31,dz31);
 289             rsq32            = gmx_mm256_calc_rsq_pd(dx32,dy32,dz32);
 290             rsq33            = gmx_mm256_calc_rsq_pd(dx33,dy33,dz33);
 291
 292             rinv00           = gmx_mm256_invsqrt_pd(rsq00);
 293             rinv11           = gmx_mm256_invsqrt_pd(rsq11);
 294             rinv12           = gmx_mm256_invsqrt_pd(rsq12);
 295             rinv13           = gmx_mm256_invsqrt_pd(rsq13);
 296             rinv21           = gmx_mm256_invsqrt_pd(rsq21);
 297             rinv22           = gmx_mm256_invsqrt_pd(rsq22);
 298             rinv23           = gmx_mm256_invsqrt_pd(rsq23);
 299             rinv31           = gmx_mm256_invsqrt_pd(rsq31);
 300             rinv32           = gmx_mm256_invsqrt_pd(rsq32);
 301             rinv33           = gmx_mm256_invsqrt_pd(rsq33);
 302
 303             rinvsq11         = _mm256_mul_pd(rinv11,rinv11);
 304             rinvsq12         = _mm256_mul_pd(rinv12,rinv12);
 305             rinvsq13         = _mm256_mul_pd(rinv13,rinv13);
 306             rinvsq21         = _mm256_mul_pd(rinv21,rinv21);
 307             rinvsq22         = _mm256_mul_pd(rinv22,rinv22);
 308             rinvsq23         = _mm256_mul_pd(rinv23,rinv23);
 309             rinvsq31         = _mm256_mul_pd(rinv31,rinv31);
 310             rinvsq32         = _mm256_mul_pd(rinv32,rinv32);
 311             rinvsq33         = _mm256_mul_pd(rinv33,rinv33);
 312
 313             fjx0             = _mm256_setzero_pd();
 314             fjy0             = _mm256_setzero_pd();
 315             fjz0             = _mm256_setzero_pd();
 316             fjx1             = _mm256_setzero_pd();
 317             fjy1             = _mm256_setzero_pd();
 318             fjz1             = _mm256_setzero_pd();
 319             fjx2             = _mm256_setzero_pd();
 320             fjy2             = _mm256_setzero_pd();
 321             fjz2             = _mm256_setzero_pd();
 322             fjx3             = _mm256_setzero_pd();
 323             fjy3             = _mm256_setzero_pd();
 324             fjz3             = _mm256_setzero_pd();
 325
 326             /**************************
 327              * CALCULATE INTERACTIONS *
 328              **************************/
 329
 330             if (gmx_mm256_any_lt(rsq00,rcutoff2))
 331             {
 332
 333             r00              = _mm256_mul_pd(rsq00,rinv00);
 334
 335             /* Calculate table index by multiplying r with table scale and truncate to integer */
 336             rt               = _mm256_mul_pd(r00,vftabscale);
 337             vfitab           = _mm256_cvttpd_epi32(rt);
 338             vfeps            = _mm256_sub_pd(rt,_mm256_round_pd(rt, _MM_FROUND_FLOOR));
 339             vfitab           = _mm_slli_epi32(vfitab,3);
 340
 341             /* CUBIC SPLINE TABLE DISPERSION */
 342             Y                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
 343             F                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
 344             G                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,2) );
 345             H                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,3) );
 346             GMX_MM256_FULLTRANSPOSE4_PD(Y,F,G,H);
 347             Heps             = _mm256_mul_pd(vfeps,H);
 348             Fp               = _mm256_add_pd(F,_mm256_mul_pd(vfeps,_mm256_add_pd(G,Heps)));
 349             VV               = _mm256_add_pd(Y,_mm256_mul_pd(vfeps,Fp));
 350             vvdw6            = _mm256_mul_pd(c6_00,VV);
 351             FF               = _mm256_add_pd(Fp,_mm256_mul_pd(vfeps,_mm256_add_pd(G,_mm256_add_pd(Heps,Heps))));
 352             fvdw6            = _mm256_mul_pd(c6_00,FF);
 353
 354             /* CUBIC SPLINE TABLE REPULSION */
 355             vfitab           = _mm_add_epi32(vfitab,ifour);
 356             Y                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
 357             F                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
 358             G                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,2) );
 359             H                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,3) );
 360             GMX_MM256_FULLTRANSPOSE4_PD(Y,F,G,H);
 361             Heps             = _mm256_mul_pd(vfeps,H);
 362             Fp               = _mm256_add_pd(F,_mm256_mul_pd(vfeps,_mm256_add_pd(G,Heps)));
 363             VV               = _mm256_add_pd(Y,_mm256_mul_pd(vfeps,Fp));
 364             vvdw12           = _mm256_mul_pd(c12_00,VV);
 365             FF               = _mm256_add_pd(Fp,_mm256_mul_pd(vfeps,_mm256_add_pd(G,_mm256_add_pd(Heps,Heps))));
 366             fvdw12           = _mm256_mul_pd(c12_00,FF);
 367             vvdw             = _mm256_add_pd(vvdw12,vvdw6);
 368             fvdw             = _mm256_xor_pd(signbit,_mm256_mul_pd(_mm256_add_pd(fvdw6,fvdw12),_mm256_mul_pd(vftabscale,rinv00)));
 369
 370             cutoff_mask      = _mm256_cmp_pd(rsq00,rcutoff2,_CMP_LT_OQ);
 371
 372             /* Update potential sum for this i atom from the interaction with this j atom. */
 373             vvdw             = _mm256_and_pd(vvdw,cutoff_mask);
 374             vvdwsum          = _mm256_add_pd(vvdwsum,vvdw);
 375
 376             fscal            = fvdw;
 377
 378             fscal            = _mm256_and_pd(fscal,cutoff_mask);
 379
 380             /* Calculate temporary vectorial force */
 381             tx               = _mm256_mul_pd(fscal,dx00);
 382             ty               = _mm256_mul_pd(fscal,dy00);
 383             tz               = _mm256_mul_pd(fscal,dz00);
 384
 385             /* Update vectorial force */
 386             fix0             = _mm256_add_pd(fix0,tx);
 387             fiy0             = _mm256_add_pd(fiy0,ty);
 388             fiz0             = _mm256_add_pd(fiz0,tz);
 389
 390             fjx0             = _mm256_add_pd(fjx0,tx);
 391             fjy0             = _mm256_add_pd(fjy0,ty);
 392             fjz0             = _mm256_add_pd(fjz0,tz);
 393
 394             }
 395
 396             /**************************
 397              * CALCULATE INTERACTIONS *
 398              **************************/
 399
 400             if (gmx_mm256_any_lt(rsq11,rcutoff2))
 401             {
 402
 403             /* REACTION-FIELD ELECTROSTATICS */
 404             velec            = _mm256_mul_pd(qq11,_mm256_sub_pd(_mm256_add_pd(rinv11,_mm256_mul_pd(krf,rsq11)),crf));
 405             felec            = _mm256_mul_pd(qq11,_mm256_sub_pd(_mm256_mul_pd(rinv11,rinvsq11),krf2));
 406
 407             cutoff_mask      = _mm256_cmp_pd(rsq11,rcutoff2,_CMP_LT_OQ);
 408
 409             /* Update potential sum for this i atom from the interaction with this j atom. */
 410             velec            = _mm256_and_pd(velec,cutoff_mask);
 411             velecsum         = _mm256_add_pd(velecsum,velec);
 412
 413             fscal            = felec;
 414
 415             fscal            = _mm256_and_pd(fscal,cutoff_mask);
 416
 417             /* Calculate temporary vectorial force */
 418             tx               = _mm256_mul_pd(fscal,dx11);
 419             ty               = _mm256_mul_pd(fscal,dy11);
 420             tz               = _mm256_mul_pd(fscal,dz11);
 421
 422             /* Update vectorial force */
 423             fix1             = _mm256_add_pd(fix1,tx);
 424             fiy1             = _mm256_add_pd(fiy1,ty);
 425             fiz1             = _mm256_add_pd(fiz1,tz);
 426
 427             fjx1             = _mm256_add_pd(fjx1,tx);
 428             fjy1             = _mm256_add_pd(fjy1,ty);
 429             fjz1             = _mm256_add_pd(fjz1,tz);
 430
 431             }
 432
 433             /**************************
 434              * CALCULATE INTERACTIONS *
 435              **************************/
 436
 437             if (gmx_mm256_any_lt(rsq12,rcutoff2))
 438             {
 439
 440             /* REACTION-FIELD ELECTROSTATICS */
 441             velec            = _mm256_mul_pd(qq12,_mm256_sub_pd(_mm256_add_pd(rinv12,_mm256_mul_pd(krf,rsq12)),crf));
 442             felec            = _mm256_mul_pd(qq12,_mm256_sub_pd(_mm256_mul_pd(rinv12,rinvsq12),krf2));
 443
 444             cutoff_mask      = _mm256_cmp_pd(rsq12,rcutoff2,_CMP_LT_OQ);
 445
 446             /* Update potential sum for this i atom from the interaction with this j atom. */
 447             velec            = _mm256_and_pd(velec,cutoff_mask);
 448             velecsum         = _mm256_add_pd(velecsum,velec);
 449
 450             fscal            = felec;
 451
 452             fscal            = _mm256_and_pd(fscal,cutoff_mask);
 453
 454             /* Calculate temporary vectorial force */
 455             tx               = _mm256_mul_pd(fscal,dx12);
 456             ty               = _mm256_mul_pd(fscal,dy12);
 457             tz               = _mm256_mul_pd(fscal,dz12);
 458
 459             /* Update vectorial force */
 460             fix1             = _mm256_add_pd(fix1,tx);
 461             fiy1             = _mm256_add_pd(fiy1,ty);
 462             fiz1             = _mm256_add_pd(fiz1,tz);
 463
 464             fjx2             = _mm256_add_pd(fjx2,tx);
 465             fjy2             = _mm256_add_pd(fjy2,ty);
 466             fjz2             = _mm256_add_pd(fjz2,tz);
 467
 468             }
 469
 470             /**************************
 471              * CALCULATE INTERACTIONS *
 472              **************************/
 473
 474             if (gmx_mm256_any_lt(rsq13,rcutoff2))
 475             {
 476
 477             /* REACTION-FIELD ELECTROSTATICS */
 478             velec            = _mm256_mul_pd(qq13,_mm256_sub_pd(_mm256_add_pd(rinv13,_mm256_mul_pd(krf,rsq13)),crf));
 479             felec            = _mm256_mul_pd(qq13,_mm256_sub_pd(_mm256_mul_pd(rinv13,rinvsq13),krf2));
 480
 481             cutoff_mask      = _mm256_cmp_pd(rsq13,rcutoff2,_CMP_LT_OQ);
 482
 483             /* Update potential sum for this i atom from the interaction with this j atom. */
 484             velec            = _mm256_and_pd(velec,cutoff_mask);
 485             velecsum         = _mm256_add_pd(velecsum,velec);
 486
 487             fscal            = felec;
 488
 489             fscal            = _mm256_and_pd(fscal,cutoff_mask);
 490
 491             /* Calculate temporary vectorial force */
 492             tx               = _mm256_mul_pd(fscal,dx13);
 493             ty               = _mm256_mul_pd(fscal,dy13);
 494             tz               = _mm256_mul_pd(fscal,dz13);
 495
 496             /* Update vectorial force */
 497             fix1             = _mm256_add_pd(fix1,tx);
 498             fiy1             = _mm256_add_pd(fiy1,ty);
 499             fiz1             = _mm256_add_pd(fiz1,tz);
 500
 501             fjx3             = _mm256_add_pd(fjx3,tx);
 502             fjy3             = _mm256_add_pd(fjy3,ty);
 503             fjz3             = _mm256_add_pd(fjz3,tz);
 504
 505             }
 506
 507             /**************************
 508              * CALCULATE INTERACTIONS *
 509              **************************/
 510
 511             if (gmx_mm256_any_lt(rsq21,rcutoff2))
 512             {
 513
 514             /* REACTION-FIELD ELECTROSTATICS */
 515             velec            = _mm256_mul_pd(qq21,_mm256_sub_pd(_mm256_add_pd(rinv21,_mm256_mul_pd(krf,rsq21)),crf));
 516             felec            = _mm256_mul_pd(qq21,_mm256_sub_pd(_mm256_mul_pd(rinv21,rinvsq21),krf2));
 517
 518             cutoff_mask      = _mm256_cmp_pd(rsq21,rcutoff2,_CMP_LT_OQ);
 519
 520             /* Update potential sum for this i atom from the interaction with this j atom. */
 521             velec            = _mm256_and_pd(velec,cutoff_mask);
 522             velecsum         = _mm256_add_pd(velecsum,velec);
 523
 524             fscal            = felec;
 525
 526             fscal            = _mm256_and_pd(fscal,cutoff_mask);
 527
 528             /* Calculate temporary vectorial force */
 529             tx               = _mm256_mul_pd(fscal,dx21);
 530             ty               = _mm256_mul_pd(fscal,dy21);
 531             tz               = _mm256_mul_pd(fscal,dz21);
 532
 533             /* Update vectorial force */
 534             fix2             = _mm256_add_pd(fix2,tx);
 535             fiy2             = _mm256_add_pd(fiy2,ty);
 536             fiz2             = _mm256_add_pd(fiz2,tz);
 537
 538             fjx1             = _mm256_add_pd(fjx1,tx);
 539             fjy1             = _mm256_add_pd(fjy1,ty);
 540             fjz1             = _mm256_add_pd(fjz1,tz);
 541
 542             }
 543
 544             /**************************
 545              * CALCULATE INTERACTIONS *
 546              **************************/
 547
 548             if (gmx_mm256_any_lt(rsq22,rcutoff2))
 549             {
 550
 551             /* REACTION-FIELD ELECTROSTATICS */
 552             velec            = _mm256_mul_pd(qq22,_mm256_sub_pd(_mm256_add_pd(rinv22,_mm256_mul_pd(krf,rsq22)),crf));
 553             felec            = _mm256_mul_pd(qq22,_mm256_sub_pd(_mm256_mul_pd(rinv22,rinvsq22),krf2));
 554
 555             cutoff_mask      = _mm256_cmp_pd(rsq22,rcutoff2,_CMP_LT_OQ);
 556
 557             /* Update potential sum for this i atom from the interaction with this j atom. */
 558             velec            = _mm256_and_pd(velec,cutoff_mask);
 559             velecsum         = _mm256_add_pd(velecsum,velec);
 560
 561             fscal            = felec;
 562
 563             fscal            = _mm256_and_pd(fscal,cutoff_mask);
 564
 565             /* Calculate temporary vectorial force */
 566             tx               = _mm256_mul_pd(fscal,dx22);
 567             ty               = _mm256_mul_pd(fscal,dy22);
 568             tz               = _mm256_mul_pd(fscal,dz22);
 569
 570             /* Update vectorial force */
 571             fix2             = _mm256_add_pd(fix2,tx);
 572             fiy2             = _mm256_add_pd(fiy2,ty);
 573             fiz2             = _mm256_add_pd(fiz2,tz);
 574
 575             fjx2             = _mm256_add_pd(fjx2,tx);
 576             fjy2             = _mm256_add_pd(fjy2,ty);
 577             fjz2             = _mm256_add_pd(fjz2,tz);
 578
 579             }
 580
 581             /**************************
 582              * CALCULATE INTERACTIONS *
 583              **************************/
 584
 585             if (gmx_mm256_any_lt(rsq23,rcutoff2))
 586             {
 587
 588             /* REACTION-FIELD ELECTROSTATICS */
 589             velec            = _mm256_mul_pd(qq23,_mm256_sub_pd(_mm256_add_pd(rinv23,_mm256_mul_pd(krf,rsq23)),crf));
 590             felec            = _mm256_mul_pd(qq23,_mm256_sub_pd(_mm256_mul_pd(rinv23,rinvsq23),krf2));
 591
 592             cutoff_mask      = _mm256_cmp_pd(rsq23,rcutoff2,_CMP_LT_OQ);
 593
 594             /* Update potential sum for this i atom from the interaction with this j atom. */
 595             velec            = _mm256_and_pd(velec,cutoff_mask);
 596             velecsum         = _mm256_add_pd(velecsum,velec);
 597
 598             fscal            = felec;
 599
 600             fscal            = _mm256_and_pd(fscal,cutoff_mask);
 601
 602             /* Calculate temporary vectorial force */
 603             tx               = _mm256_mul_pd(fscal,dx23);
 604             ty               = _mm256_mul_pd(fscal,dy23);
 605             tz               = _mm256_mul_pd(fscal,dz23);
 606
 607             /* Update vectorial force */
 608             fix2             = _mm256_add_pd(fix2,tx);
 609             fiy2             = _mm256_add_pd(fiy2,ty);
 610             fiz2             = _mm256_add_pd(fiz2,tz);
 611
 612             fjx3             = _mm256_add_pd(fjx3,tx);
 613             fjy3             = _mm256_add_pd(fjy3,ty);
 614             fjz3             = _mm256_add_pd(fjz3,tz);
 615
 616             }
 617
 618             /**************************
 619              * CALCULATE INTERACTIONS *
 620              **************************/
 621
 622             if (gmx_mm256_any_lt(rsq31,rcutoff2))
 623             {
 624
 625             /* REACTION-FIELD ELECTROSTATICS */
 626             velec            = _mm256_mul_pd(qq31,_mm256_sub_pd(_mm256_add_pd(rinv31,_mm256_mul_pd(krf,rsq31)),crf));
 627             felec            = _mm256_mul_pd(qq31,_mm256_sub_pd(_mm256_mul_pd(rinv31,rinvsq31),krf2));
 628
 629             cutoff_mask      = _mm256_cmp_pd(rsq31,rcutoff2,_CMP_LT_OQ);
 630
 631             /* Update potential sum for this i atom from the interaction with this j atom. */
 632             velec            = _mm256_and_pd(velec,cutoff_mask);
 633             velecsum         = _mm256_add_pd(velecsum,velec);
 634
 635             fscal            = felec;
 636
 637             fscal            = _mm256_and_pd(fscal,cutoff_mask);
 638
 639             /* Calculate temporary vectorial force */
 640             tx               = _mm256_mul_pd(fscal,dx31);
 641             ty               = _mm256_mul_pd(fscal,dy31);
 642             tz               = _mm256_mul_pd(fscal,dz31);
 643
 644             /* Update vectorial force */
 645             fix3             = _mm256_add_pd(fix3,tx);
 646             fiy3             = _mm256_add_pd(fiy3,ty);
 647             fiz3             = _mm256_add_pd(fiz3,tz);
 648
 649             fjx1             = _mm256_add_pd(fjx1,tx);
 650             fjy1             = _mm256_add_pd(fjy1,ty);
 651             fjz1             = _mm256_add_pd(fjz1,tz);
 652
 653             }
 654
 655             /**************************
 656              * CALCULATE INTERACTIONS *
 657              **************************/
 658
 659             if (gmx_mm256_any_lt(rsq32,rcutoff2))
 660             {
 661
 662             /* REACTION-FIELD ELECTROSTATICS */
 663             velec            = _mm256_mul_pd(qq32,_mm256_sub_pd(_mm256_add_pd(rinv32,_mm256_mul_pd(krf,rsq32)),crf));
 664             felec            = _mm256_mul_pd(qq32,_mm256_sub_pd(_mm256_mul_pd(rinv32,rinvsq32),krf2));
 665
 666             cutoff_mask      = _mm256_cmp_pd(rsq32,rcutoff2,_CMP_LT_OQ);
 667
 668             /* Update potential sum for this i atom from the interaction with this j atom. */
 669             velec            = _mm256_and_pd(velec,cutoff_mask);
 670             velecsum         = _mm256_add_pd(velecsum,velec);
 671
 672             fscal            = felec;
 673
 674             fscal            = _mm256_and_pd(fscal,cutoff_mask);
 675
 676             /* Calculate temporary vectorial force */
 677             tx               = _mm256_mul_pd(fscal,dx32);
 678             ty               = _mm256_mul_pd(fscal,dy32);
 679             tz               = _mm256_mul_pd(fscal,dz32);
 680
 681             /* Update vectorial force */
 682             fix3             = _mm256_add_pd(fix3,tx);
 683             fiy3             = _mm256_add_pd(fiy3,ty);
 684             fiz3             = _mm256_add_pd(fiz3,tz);
 685
 686             fjx2             = _mm256_add_pd(fjx2,tx);
 687             fjy2             = _mm256_add_pd(fjy2,ty);
 688             fjz2             = _mm256_add_pd(fjz2,tz);
 689
 690             }
 691
 692             /**************************
 693              * CALCULATE INTERACTIONS *
 694              **************************/
 695
 696             if (gmx_mm256_any_lt(rsq33,rcutoff2))
 697             {
 698
 699             /* REACTION-FIELD ELECTROSTATICS */
 700             velec            = _mm256_mul_pd(qq33,_mm256_sub_pd(_mm256_add_pd(rinv33,_mm256_mul_pd(krf,rsq33)),crf));
 701             felec            = _mm256_mul_pd(qq33,_mm256_sub_pd(_mm256_mul_pd(rinv33,rinvsq33),krf2));
 702
 703             cutoff_mask      = _mm256_cmp_pd(rsq33,rcutoff2,_CMP_LT_OQ);
 704
 705             /* Update potential sum for this i atom from the interaction with this j atom. */
 706             velec            = _mm256_and_pd(velec,cutoff_mask);
 707             velecsum         = _mm256_add_pd(velecsum,velec);
 708
 709             fscal            = felec;
 710
 711             fscal            = _mm256_and_pd(fscal,cutoff_mask);
 712
 713             /* Calculate temporary vectorial force */
 714             tx               = _mm256_mul_pd(fscal,dx33);
 715             ty               = _mm256_mul_pd(fscal,dy33);
 716             tz               = _mm256_mul_pd(fscal,dz33);
 717
 718             /* Update vectorial force */
 719             fix3             = _mm256_add_pd(fix3,tx);
 720             fiy3             = _mm256_add_pd(fiy3,ty);
 721             fiz3             = _mm256_add_pd(fiz3,tz);
 722
 723             fjx3             = _mm256_add_pd(fjx3,tx);
 724             fjy3             = _mm256_add_pd(fjy3,ty);
 725             fjz3             = _mm256_add_pd(fjz3,tz);
 726
 727             }
 728
 729             fjptrA             = f+j_coord_offsetA;
 730             fjptrB             = f+j_coord_offsetB;
 731             fjptrC             = f+j_coord_offsetC;
 732             fjptrD             = f+j_coord_offsetD;
 733
 734             gmx_mm256_decrement_4rvec_4ptr_swizzle_pd(fjptrA,fjptrB,fjptrC,fjptrD,
 735                                                       fjx0,fjy0,fjz0,fjx1,fjy1,fjz1,
 736                                                       fjx2,fjy2,fjz2,fjx3,fjy3,fjz3);
 737
 738             /* Inner loop uses 387 flops */
 739         }
 740
 741         if(jidx<j_index_end)
 742         {
 743
 744             /* Get j neighbor index, and coordinate index */
 745             jnrlistA         = jjnr[jidx];
 746             jnrlistB         = jjnr[jidx+1];
 747             jnrlistC         = jjnr[jidx+2];
 748             jnrlistD         = jjnr[jidx+3];
 749             /* Sign of each element will be negative for non-real atoms.
 750              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 751              * so use it as val = _mm_andnot_pd(mask,val) to clear dummy entries.
 752              */
 753             tmpmask0 = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
 754
 755             tmpmask1 = _mm_permute_ps(tmpmask0,_GMX_MM_PERMUTE(3,3,2,2));
 756             tmpmask0 = _mm_permute_ps(tmpmask0,_GMX_MM_PERMUTE(1,1,0,0));
 757             dummy_mask = _mm256_castps_pd(gmx_mm256_set_m128(tmpmask1,tmpmask0));
 758
 759             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
 760             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
 761             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
 762             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
 763             j_coord_offsetA  = DIM*jnrA;
 764             j_coord_offsetB  = DIM*jnrB;
 765             j_coord_offsetC  = DIM*jnrC;
 766             j_coord_offsetD  = DIM*jnrD;
 767
 768             /* load j atom coordinates */
 769             gmx_mm256_load_4rvec_4ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 770                                                  x+j_coord_offsetC,x+j_coord_offsetD,
 771                                                  &jx0,&jy0,&jz0,&jx1,&jy1,&jz1,&jx2,
 772                                                  &jy2,&jz2,&jx3,&jy3,&jz3);
 773
 774             /* Calculate displacement vector */
 775             dx00             = _mm256_sub_pd(ix0,jx0);
 776             dy00             = _mm256_sub_pd(iy0,jy0);
 777             dz00             = _mm256_sub_pd(iz0,jz0);
 778             dx11             = _mm256_sub_pd(ix1,jx1);
 779             dy11             = _mm256_sub_pd(iy1,jy1);
 780             dz11             = _mm256_sub_pd(iz1,jz1);
 781             dx12             = _mm256_sub_pd(ix1,jx2);
 782             dy12             = _mm256_sub_pd(iy1,jy2);
 783             dz12             = _mm256_sub_pd(iz1,jz2);
 784             dx13             = _mm256_sub_pd(ix1,jx3);
 785             dy13             = _mm256_sub_pd(iy1,jy3);
 786             dz13             = _mm256_sub_pd(iz1,jz3);
 787             dx21             = _mm256_sub_pd(ix2,jx1);
 788             dy21             = _mm256_sub_pd(iy2,jy1);
 789             dz21             = _mm256_sub_pd(iz2,jz1);
 790             dx22             = _mm256_sub_pd(ix2,jx2);
 791             dy22             = _mm256_sub_pd(iy2,jy2);
 792             dz22             = _mm256_sub_pd(iz2,jz2);
 793             dx23             = _mm256_sub_pd(ix2,jx3);
 794             dy23             = _mm256_sub_pd(iy2,jy3);
 795             dz23             = _mm256_sub_pd(iz2,jz3);
 796             dx31             = _mm256_sub_pd(ix3,jx1);
 797             dy31             = _mm256_sub_pd(iy3,jy1);
 798             dz31             = _mm256_sub_pd(iz3,jz1);
 799             dx32             = _mm256_sub_pd(ix3,jx2);
 800             dy32             = _mm256_sub_pd(iy3,jy2);
 801             dz32             = _mm256_sub_pd(iz3,jz2);
 802             dx33             = _mm256_sub_pd(ix3,jx3);
 803             dy33             = _mm256_sub_pd(iy3,jy3);
 804             dz33             = _mm256_sub_pd(iz3,jz3);
 805
 806             /* Calculate squared distance and things based on it */
 807             rsq00            = gmx_mm256_calc_rsq_pd(dx00,dy00,dz00);
 808             rsq11            = gmx_mm256_calc_rsq_pd(dx11,dy11,dz11);
 809             rsq12            = gmx_mm256_calc_rsq_pd(dx12,dy12,dz12);
 810             rsq13            = gmx_mm256_calc_rsq_pd(dx13,dy13,dz13);
 811             rsq21            = gmx_mm256_calc_rsq_pd(dx21,dy21,dz21);
 812             rsq22            = gmx_mm256_calc_rsq_pd(dx22,dy22,dz22);
 813             rsq23            = gmx_mm256_calc_rsq_pd(dx23,dy23,dz23);
 814             rsq31            = gmx_mm256_calc_rsq_pd(dx31,dy31,dz31);
 815             rsq32            = gmx_mm256_calc_rsq_pd(dx32,dy32,dz32);
 816             rsq33            = gmx_mm256_calc_rsq_pd(dx33,dy33,dz33);
 817
 818             rinv00           = gmx_mm256_invsqrt_pd(rsq00);
 819             rinv11           = gmx_mm256_invsqrt_pd(rsq11);
 820             rinv12           = gmx_mm256_invsqrt_pd(rsq12);
 821             rinv13           = gmx_mm256_invsqrt_pd(rsq13);
 822             rinv21           = gmx_mm256_invsqrt_pd(rsq21);
 823             rinv22           = gmx_mm256_invsqrt_pd(rsq22);
 824             rinv23           = gmx_mm256_invsqrt_pd(rsq23);
 825             rinv31           = gmx_mm256_invsqrt_pd(rsq31);
 826             rinv32           = gmx_mm256_invsqrt_pd(rsq32);
 827             rinv33           = gmx_mm256_invsqrt_pd(rsq33);
 828
 829             rinvsq11         = _mm256_mul_pd(rinv11,rinv11);
 830             rinvsq12         = _mm256_mul_pd(rinv12,rinv12);
 831             rinvsq13         = _mm256_mul_pd(rinv13,rinv13);
 832             rinvsq21         = _mm256_mul_pd(rinv21,rinv21);
 833             rinvsq22         = _mm256_mul_pd(rinv22,rinv22);
 834             rinvsq23         = _mm256_mul_pd(rinv23,rinv23);
 835             rinvsq31         = _mm256_mul_pd(rinv31,rinv31);
 836             rinvsq32         = _mm256_mul_pd(rinv32,rinv32);
 837             rinvsq33         = _mm256_mul_pd(rinv33,rinv33);
 838
 839             fjx0             = _mm256_setzero_pd();
 840             fjy0             = _mm256_setzero_pd();
 841             fjz0             = _mm256_setzero_pd();
 842             fjx1             = _mm256_setzero_pd();
 843             fjy1             = _mm256_setzero_pd();
 844             fjz1             = _mm256_setzero_pd();
 845             fjx2             = _mm256_setzero_pd();
 846             fjy2             = _mm256_setzero_pd();
 847             fjz2             = _mm256_setzero_pd();
 848             fjx3             = _mm256_setzero_pd();
 849             fjy3             = _mm256_setzero_pd();
 850             fjz3             = _mm256_setzero_pd();
 851
 852             /**************************
 853              * CALCULATE INTERACTIONS *
 854              **************************/
 855
 856             if (gmx_mm256_any_lt(rsq00,rcutoff2))
 857             {
 858
 859             r00              = _mm256_mul_pd(rsq00,rinv00);
 860             r00              = _mm256_andnot_pd(dummy_mask,r00);
 861
 862             /* Calculate table index by multiplying r with table scale and truncate to integer */
 863             rt               = _mm256_mul_pd(r00,vftabscale);
 864             vfitab           = _mm256_cvttpd_epi32(rt);
 865             vfeps            = _mm256_sub_pd(rt,_mm256_round_pd(rt, _MM_FROUND_FLOOR));
 866             vfitab           = _mm_slli_epi32(vfitab,3);
 867
 868             /* CUBIC SPLINE TABLE DISPERSION */
 869             Y                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
 870             F                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
 871             G                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,2) );
 872             H                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,3) );
 873             GMX_MM256_FULLTRANSPOSE4_PD(Y,F,G,H);
 874             Heps             = _mm256_mul_pd(vfeps,H);
 875             Fp               = _mm256_add_pd(F,_mm256_mul_pd(vfeps,_mm256_add_pd(G,Heps)));
 876             VV               = _mm256_add_pd(Y,_mm256_mul_pd(vfeps,Fp));
 877             vvdw6            = _mm256_mul_pd(c6_00,VV);
 878             FF               = _mm256_add_pd(Fp,_mm256_mul_pd(vfeps,_mm256_add_pd(G,_mm256_add_pd(Heps,Heps))));
 879             fvdw6            = _mm256_mul_pd(c6_00,FF);
 880
 881             /* CUBIC SPLINE TABLE REPULSION */
 882             vfitab           = _mm_add_epi32(vfitab,ifour);
 883             Y                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
 884             F                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
 885             G                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,2) );
 886             H                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,3) );
 887             GMX_MM256_FULLTRANSPOSE4_PD(Y,F,G,H);
 888             Heps             = _mm256_mul_pd(vfeps,H);
 889             Fp               = _mm256_add_pd(F,_mm256_mul_pd(vfeps,_mm256_add_pd(G,Heps)));
 890             VV               = _mm256_add_pd(Y,_mm256_mul_pd(vfeps,Fp));
 891             vvdw12           = _mm256_mul_pd(c12_00,VV);
 892             FF               = _mm256_add_pd(Fp,_mm256_mul_pd(vfeps,_mm256_add_pd(G,_mm256_add_pd(Heps,Heps))));
 893             fvdw12           = _mm256_mul_pd(c12_00,FF);
 894             vvdw             = _mm256_add_pd(vvdw12,vvdw6);
 895             fvdw             = _mm256_xor_pd(signbit,_mm256_mul_pd(_mm256_add_pd(fvdw6,fvdw12),_mm256_mul_pd(vftabscale,rinv00)));
 896
 897             cutoff_mask      = _mm256_cmp_pd(rsq00,rcutoff2,_CMP_LT_OQ);
 898
 899             /* Update potential sum for this i atom from the interaction with this j atom. */
 900             vvdw             = _mm256_and_pd(vvdw,cutoff_mask);
 901             vvdw             = _mm256_andnot_pd(dummy_mask,vvdw);
 902             vvdwsum          = _mm256_add_pd(vvdwsum,vvdw);
 903
 904             fscal            = fvdw;
 905
 906             fscal            = _mm256_and_pd(fscal,cutoff_mask);
 907
 908             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
 909
 910             /* Calculate temporary vectorial force */
 911             tx               = _mm256_mul_pd(fscal,dx00);
 912             ty               = _mm256_mul_pd(fscal,dy00);
 913             tz               = _mm256_mul_pd(fscal,dz00);
 914
 915             /* Update vectorial force */
 916             fix0             = _mm256_add_pd(fix0,tx);
 917             fiy0             = _mm256_add_pd(fiy0,ty);
 918             fiz0             = _mm256_add_pd(fiz0,tz);
 919
 920             fjx0             = _mm256_add_pd(fjx0,tx);
 921             fjy0             = _mm256_add_pd(fjy0,ty);
 922             fjz0             = _mm256_add_pd(fjz0,tz);
 923
 924             }
 925
 926             /**************************
 927              * CALCULATE INTERACTIONS *
 928              **************************/
 929
 930             if (gmx_mm256_any_lt(rsq11,rcutoff2))
 931             {
 932
 933             /* REACTION-FIELD ELECTROSTATICS */
 934             velec            = _mm256_mul_pd(qq11,_mm256_sub_pd(_mm256_add_pd(rinv11,_mm256_mul_pd(krf,rsq11)),crf));
 935             felec            = _mm256_mul_pd(qq11,_mm256_sub_pd(_mm256_mul_pd(rinv11,rinvsq11),krf2));
 936
 937             cutoff_mask      = _mm256_cmp_pd(rsq11,rcutoff2,_CMP_LT_OQ);
 938
 939             /* Update potential sum for this i atom from the interaction with this j atom. */
 940             velec            = _mm256_and_pd(velec,cutoff_mask);
 941             velec            = _mm256_andnot_pd(dummy_mask,velec);
 942             velecsum         = _mm256_add_pd(velecsum,velec);
 943
 944             fscal            = felec;
 945
 946             fscal            = _mm256_and_pd(fscal,cutoff_mask);
 947
 948             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
 949
 950             /* Calculate temporary vectorial force */
 951             tx               = _mm256_mul_pd(fscal,dx11);
 952             ty               = _mm256_mul_pd(fscal,dy11);
 953             tz               = _mm256_mul_pd(fscal,dz11);
 954
 955             /* Update vectorial force */
 956             fix1             = _mm256_add_pd(fix1,tx);
 957             fiy1             = _mm256_add_pd(fiy1,ty);
 958             fiz1             = _mm256_add_pd(fiz1,tz);
 959
 960             fjx1             = _mm256_add_pd(fjx1,tx);
 961             fjy1             = _mm256_add_pd(fjy1,ty);
 962             fjz1             = _mm256_add_pd(fjz1,tz);
 963
 964             }
 965
 966             /**************************
 967              * CALCULATE INTERACTIONS *
 968              **************************/
 969
 970             if (gmx_mm256_any_lt(rsq12,rcutoff2))
 971             {
 972
 973             /* REACTION-FIELD ELECTROSTATICS */
 974             velec            = _mm256_mul_pd(qq12,_mm256_sub_pd(_mm256_add_pd(rinv12,_mm256_mul_pd(krf,rsq12)),crf));
 975             felec            = _mm256_mul_pd(qq12,_mm256_sub_pd(_mm256_mul_pd(rinv12,rinvsq12),krf2));
 976
 977             cutoff_mask      = _mm256_cmp_pd(rsq12,rcutoff2,_CMP_LT_OQ);
 978
 979             /* Update potential sum for this i atom from the interaction with this j atom. */
 980             velec            = _mm256_and_pd(velec,cutoff_mask);
 981             velec            = _mm256_andnot_pd(dummy_mask,velec);
 982             velecsum         = _mm256_add_pd(velecsum,velec);
 983
 984             fscal            = felec;
 985
 986             fscal            = _mm256_and_pd(fscal,cutoff_mask);
 987
 988             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
 989
 990             /* Calculate temporary vectorial force */
 991             tx               = _mm256_mul_pd(fscal,dx12);
 992             ty               = _mm256_mul_pd(fscal,dy12);
 993             tz               = _mm256_mul_pd(fscal,dz12);
 994
 995             /* Update vectorial force */
 996             fix1             = _mm256_add_pd(fix1,tx);
 997             fiy1             = _mm256_add_pd(fiy1,ty);
 998             fiz1             = _mm256_add_pd(fiz1,tz);
 999
1000             fjx2             = _mm256_add_pd(fjx2,tx);
1001             fjy2             = _mm256_add_pd(fjy2,ty);
1002             fjz2             = _mm256_add_pd(fjz2,tz);
1003
1004             }
1005
1006             /**************************
1007              * CALCULATE INTERACTIONS *
1008              **************************/
1009
1010             if (gmx_mm256_any_lt(rsq13,rcutoff2))
1011             {
1012
1013             /* REACTION-FIELD ELECTROSTATICS */
1014             velec            = _mm256_mul_pd(qq13,_mm256_sub_pd(_mm256_add_pd(rinv13,_mm256_mul_pd(krf,rsq13)),crf));
1015             felec            = _mm256_mul_pd(qq13,_mm256_sub_pd(_mm256_mul_pd(rinv13,rinvsq13),krf2));
1016
1017             cutoff_mask      = _mm256_cmp_pd(rsq13,rcutoff2,_CMP_LT_OQ);
1018
1019             /* Update potential sum for this i atom from the interaction with this j atom. */
1020             velec            = _mm256_and_pd(velec,cutoff_mask);
1021             velec            = _mm256_andnot_pd(dummy_mask,velec);
1022             velecsum         = _mm256_add_pd(velecsum,velec);
1023
1024             fscal            = felec;
1025
1026             fscal            = _mm256_and_pd(fscal,cutoff_mask);
1027
1028             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
1029
1030             /* Calculate temporary vectorial force */
1031             tx               = _mm256_mul_pd(fscal,dx13);
1032             ty               = _mm256_mul_pd(fscal,dy13);
1033             tz               = _mm256_mul_pd(fscal,dz13);
1034
1035             /* Update vectorial force */
1036             fix1             = _mm256_add_pd(fix1,tx);
1037             fiy1             = _mm256_add_pd(fiy1,ty);
1038             fiz1             = _mm256_add_pd(fiz1,tz);
1039
1040             fjx3             = _mm256_add_pd(fjx3,tx);
1041             fjy3             = _mm256_add_pd(fjy3,ty);
1042             fjz3             = _mm256_add_pd(fjz3,tz);
1043
1044             }
1045
1046             /**************************
1047              * CALCULATE INTERACTIONS *
1048              **************************/
1049
1050             if (gmx_mm256_any_lt(rsq21,rcutoff2))
1051             {
1052
1053             /* REACTION-FIELD ELECTROSTATICS */
1054             velec            = _mm256_mul_pd(qq21,_mm256_sub_pd(_mm256_add_pd(rinv21,_mm256_mul_pd(krf,rsq21)),crf));
1055             felec            = _mm256_mul_pd(qq21,_mm256_sub_pd(_mm256_mul_pd(rinv21,rinvsq21),krf2));
1056
1057             cutoff_mask      = _mm256_cmp_pd(rsq21,rcutoff2,_CMP_LT_OQ);
1058
1059             /* Update potential sum for this i atom from the interaction with this j atom. */
1060             velec            = _mm256_and_pd(velec,cutoff_mask);
1061             velec            = _mm256_andnot_pd(dummy_mask,velec);
1062             velecsum         = _mm256_add_pd(velecsum,velec);
1063
1064             fscal            = felec;
1065
1066             fscal            = _mm256_and_pd(fscal,cutoff_mask);
1067
1068             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
1069
1070             /* Calculate temporary vectorial force */
1071             tx               = _mm256_mul_pd(fscal,dx21);
1072             ty               = _mm256_mul_pd(fscal,dy21);
1073             tz               = _mm256_mul_pd(fscal,dz21);
1074
1075             /* Update vectorial force */
1076             fix2             = _mm256_add_pd(fix2,tx);
1077             fiy2             = _mm256_add_pd(fiy2,ty);
1078             fiz2             = _mm256_add_pd(fiz2,tz);
1079
1080             fjx1             = _mm256_add_pd(fjx1,tx);
1081             fjy1             = _mm256_add_pd(fjy1,ty);
1082             fjz1             = _mm256_add_pd(fjz1,tz);
1083
1084             }
1085
1086             /**************************
1087              * CALCULATE INTERACTIONS *
1088              **************************/
1089
1090             if (gmx_mm256_any_lt(rsq22,rcutoff2))
1091             {
1092
1093             /* REACTION-FIELD ELECTROSTATICS */
1094             velec            = _mm256_mul_pd(qq22,_mm256_sub_pd(_mm256_add_pd(rinv22,_mm256_mul_pd(krf,rsq22)),crf));
1095             felec            = _mm256_mul_pd(qq22,_mm256_sub_pd(_mm256_mul_pd(rinv22,rinvsq22),krf2));
1096
1097             cutoff_mask      = _mm256_cmp_pd(rsq22,rcutoff2,_CMP_LT_OQ);
1098
1099             /* Update potential sum for this i atom from the interaction with this j atom. */
1100             velec            = _mm256_and_pd(velec,cutoff_mask);
1101             velec            = _mm256_andnot_pd(dummy_mask,velec);
1102             velecsum         = _mm256_add_pd(velecsum,velec);
1103
1104             fscal            = felec;
1105
1106             fscal            = _mm256_and_pd(fscal,cutoff_mask);
1107
1108             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
1109
1110             /* Calculate temporary vectorial force */
1111             tx               = _mm256_mul_pd(fscal,dx22);
1112             ty               = _mm256_mul_pd(fscal,dy22);
1113             tz               = _mm256_mul_pd(fscal,dz22);
1114
1115             /* Update vectorial force */
1116             fix2             = _mm256_add_pd(fix2,tx);
1117             fiy2             = _mm256_add_pd(fiy2,ty);
1118             fiz2             = _mm256_add_pd(fiz2,tz);
1119
1120             fjx2             = _mm256_add_pd(fjx2,tx);
1121             fjy2             = _mm256_add_pd(fjy2,ty);
1122             fjz2             = _mm256_add_pd(fjz2,tz);
1123
1124             }
1125
1126             /**************************
1127              * CALCULATE INTERACTIONS *
1128              **************************/
1129
1130             if (gmx_mm256_any_lt(rsq23,rcutoff2))
1131             {
1132
1133             /* REACTION-FIELD ELECTROSTATICS */
1134             velec            = _mm256_mul_pd(qq23,_mm256_sub_pd(_mm256_add_pd(rinv23,_mm256_mul_pd(krf,rsq23)),crf));
1135             felec            = _mm256_mul_pd(qq23,_mm256_sub_pd(_mm256_mul_pd(rinv23,rinvsq23),krf2));
1136
1137             cutoff_mask      = _mm256_cmp_pd(rsq23,rcutoff2,_CMP_LT_OQ);
1138
1139             /* Update potential sum for this i atom from the interaction with this j atom. */
1140             velec            = _mm256_and_pd(velec,cutoff_mask);
1141             velec            = _mm256_andnot_pd(dummy_mask,velec);
1142             velecsum         = _mm256_add_pd(velecsum,velec);
1143
1144             fscal            = felec;
1145
1146             fscal            = _mm256_and_pd(fscal,cutoff_mask);
1147
1148             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
1149
1150             /* Calculate temporary vectorial force */
1151             tx               = _mm256_mul_pd(fscal,dx23);
1152             ty               = _mm256_mul_pd(fscal,dy23);
1153             tz               = _mm256_mul_pd(fscal,dz23);
1154
1155             /* Update vectorial force */
1156             fix2             = _mm256_add_pd(fix2,tx);
1157             fiy2             = _mm256_add_pd(fiy2,ty);
1158             fiz2             = _mm256_add_pd(fiz2,tz);
1159
1160             fjx3             = _mm256_add_pd(fjx3,tx);
1161             fjy3             = _mm256_add_pd(fjy3,ty);
1162             fjz3             = _mm256_add_pd(fjz3,tz);
1163
1164             }
1165
1166             /**************************
1167              * CALCULATE INTERACTIONS *
1168              **************************/
1169
1170             if (gmx_mm256_any_lt(rsq31,rcutoff2))
1171             {
1172
1173             /* REACTION-FIELD ELECTROSTATICS */
1174             velec            = _mm256_mul_pd(qq31,_mm256_sub_pd(_mm256_add_pd(rinv31,_mm256_mul_pd(krf,rsq31)),crf));
1175             felec            = _mm256_mul_pd(qq31,_mm256_sub_pd(_mm256_mul_pd(rinv31,rinvsq31),krf2));
1176
1177             cutoff_mask      = _mm256_cmp_pd(rsq31,rcutoff2,_CMP_LT_OQ);
1178
1179             /* Update potential sum for this i atom from the interaction with this j atom. */
1180             velec            = _mm256_and_pd(velec,cutoff_mask);
1181             velec            = _mm256_andnot_pd(dummy_mask,velec);
1182             velecsum         = _mm256_add_pd(velecsum,velec);
1183
1184             fscal            = felec;
1185
1186             fscal            = _mm256_and_pd(fscal,cutoff_mask);
1187
1188             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
1189
1190             /* Calculate temporary vectorial force */
1191             tx               = _mm256_mul_pd(fscal,dx31);
1192             ty               = _mm256_mul_pd(fscal,dy31);
1193             tz               = _mm256_mul_pd(fscal,dz31);
1194
1195             /* Update vectorial force */
1196             fix3             = _mm256_add_pd(fix3,tx);
1197             fiy3             = _mm256_add_pd(fiy3,ty);
1198             fiz3             = _mm256_add_pd(fiz3,tz);
1199
1200             fjx1             = _mm256_add_pd(fjx1,tx);
1201             fjy1             = _mm256_add_pd(fjy1,ty);
1202             fjz1             = _mm256_add_pd(fjz1,tz);
1203
1204             }
1205
1206             /**************************
1207              * CALCULATE INTERACTIONS *
1208              **************************/
1209
1210             if (gmx_mm256_any_lt(rsq32,rcutoff2))
1211             {
1212
1213             /* REACTION-FIELD ELECTROSTATICS */
1214             velec            = _mm256_mul_pd(qq32,_mm256_sub_pd(_mm256_add_pd(rinv32,_mm256_mul_pd(krf,rsq32)),crf));
1215             felec            = _mm256_mul_pd(qq32,_mm256_sub_pd(_mm256_mul_pd(rinv32,rinvsq32),krf2));
1216
1217             cutoff_mask      = _mm256_cmp_pd(rsq32,rcutoff2,_CMP_LT_OQ);
1218
1219             /* Update potential sum for this i atom from the interaction with this j atom. */
1220             velec            = _mm256_and_pd(velec,cutoff_mask);
1221             velec            = _mm256_andnot_pd(dummy_mask,velec);
1222             velecsum         = _mm256_add_pd(velecsum,velec);
1223
1224             fscal            = felec;
1225
1226             fscal            = _mm256_and_pd(fscal,cutoff_mask);
1227
1228             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
1229
1230             /* Calculate temporary vectorial force */
1231             tx               = _mm256_mul_pd(fscal,dx32);
1232             ty               = _mm256_mul_pd(fscal,dy32);
1233             tz               = _mm256_mul_pd(fscal,dz32);
1234
1235             /* Update vectorial force */
1236             fix3             = _mm256_add_pd(fix3,tx);
1237             fiy3             = _mm256_add_pd(fiy3,ty);
1238             fiz3             = _mm256_add_pd(fiz3,tz);
1239
1240             fjx2             = _mm256_add_pd(fjx2,tx);
1241             fjy2             = _mm256_add_pd(fjy2,ty);
1242             fjz2             = _mm256_add_pd(fjz2,tz);
1243
1244             }
1245
1246             /**************************
1247              * CALCULATE INTERACTIONS *
1248              **************************/
1249
1250             if (gmx_mm256_any_lt(rsq33,rcutoff2))
1251             {
1252
1253             /* REACTION-FIELD ELECTROSTATICS */
1254             velec            = _mm256_mul_pd(qq33,_mm256_sub_pd(_mm256_add_pd(rinv33,_mm256_mul_pd(krf,rsq33)),crf));
1255             felec            = _mm256_mul_pd(qq33,_mm256_sub_pd(_mm256_mul_pd(rinv33,rinvsq33),krf2));
1256
1257             cutoff_mask      = _mm256_cmp_pd(rsq33,rcutoff2,_CMP_LT_OQ);
1258
1259             /* Update potential sum for this i atom from the interaction with this j atom. */
1260             velec            = _mm256_and_pd(velec,cutoff_mask);
1261             velec            = _mm256_andnot_pd(dummy_mask,velec);
1262             velecsum         = _mm256_add_pd(velecsum,velec);
1263
1264             fscal            = felec;
1265
1266             fscal            = _mm256_and_pd(fscal,cutoff_mask);
1267
1268             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
1269
1270             /* Calculate temporary vectorial force */
1271             tx               = _mm256_mul_pd(fscal,dx33);
1272             ty               = _mm256_mul_pd(fscal,dy33);
1273             tz               = _mm256_mul_pd(fscal,dz33);
1274
1275             /* Update vectorial force */
1276             fix3             = _mm256_add_pd(fix3,tx);
1277             fiy3             = _mm256_add_pd(fiy3,ty);
1278             fiz3             = _mm256_add_pd(fiz3,tz);
1279
1280             fjx3             = _mm256_add_pd(fjx3,tx);
1281             fjy3             = _mm256_add_pd(fjy3,ty);
1282             fjz3             = _mm256_add_pd(fjz3,tz);
1283
1284             }
1285
1286             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
1287             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
1288             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
1289             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
1290
1291             gmx_mm256_decrement_4rvec_4ptr_swizzle_pd(fjptrA,fjptrB,fjptrC,fjptrD,
1292                                                       fjx0,fjy0,fjz0,fjx1,fjy1,fjz1,
1293                                                       fjx2,fjy2,fjz2,fjx3,fjy3,fjz3);
1294
1295             /* Inner loop uses 388 flops */
1296         }
1297
1298         /* End of innermost loop */
1299
1300         gmx_mm256_update_iforce_4atom_swizzle_pd(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,fix3,fiy3,fiz3,
1301                                                  f+i_coord_offset,fshift+i_shift_offset);
1302
1303         ggid                        = gid[iidx];
1304         /* Update potential energies */
1305         gmx_mm256_update_1pot_pd(velecsum,kernel_data->energygrp_elec+ggid);
1306         gmx_mm256_update_1pot_pd(vvdwsum,kernel_data->energygrp_vdw+ggid);
1307
1308         /* Increment number of inner iterations */
1309         inneriter                  += j_index_end - j_index_start;
1310
1311         /* Outer loop uses 26 flops */
1312     }
1313
1314     /* Increment number of outer iterations */
1315     outeriter        += nri;
1316
1317     /* Update outer/inner flops */
1318
1319     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W4W4_VF,outeriter*26 + inneriter*388);
1320 }
1321 /*
1322  * Gromacs nonbonded kernel:   nb_kernel_ElecRFCut_VdwCSTab_GeomW4W4_F_avx_256_double
1323  * Electrostatics interaction: ReactionField
1324  * VdW interaction:            CubicSplineTable
1325  * Geometry:                   Water4-Water4
1326  * Calculate force/pot:        Force
1327  */
1328 void
1329 nb_kernel_ElecRFCut_VdwCSTab_GeomW4W4_F_avx_256_double
1330                     (t_nblist                    * gmx_restrict       nlist,
1331                      rvec                        * gmx_restrict          xx,
1332                      rvec                        * gmx_restrict          ff,
1333                      t_forcerec                  * gmx_restrict          fr,
1334                      t_mdatoms                   * gmx_restrict     mdatoms,
1335                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
1336                      t_nrnb                      * gmx_restrict        nrnb)
1337 {
1338     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
1339      * just 0 for non-waters.
1340      * Suffixes A,B,C,D refer to j loop unrolling done with AVX, e.g. for the four different
1341      * jnr indices corresponding to data put in the four positions in the SIMD register.
1342      */
1343     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
1344     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
1345     int              jnrA,jnrB,jnrC,jnrD;
1346     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
1347     int              jnrlistE,jnrlistF,jnrlistG,jnrlistH;
1348     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
1349     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
1350     real             rcutoff_scalar;
1351     real             *shiftvec,*fshift,*x,*f;
1352     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD;
1353     real             scratch[4*DIM];
1354     __m256d          tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
1355     real *           vdwioffsetptr0;
1356     __m256d          ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
1357     real *           vdwioffsetptr1;
1358     __m256d          ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
1359     real *           vdwioffsetptr2;
1360     __m256d          ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
1361     real *           vdwioffsetptr3;
1362     __m256d          ix3,iy3,iz3,fix3,fiy3,fiz3,iq3,isai3;
1363     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
1364     __m256d          jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
1365     int              vdwjidx1A,vdwjidx1B,vdwjidx1C,vdwjidx1D;
1366     __m256d          jx1,jy1,jz1,fjx1,fjy1,fjz1,jq1,isaj1;
1367     int              vdwjidx2A,vdwjidx2B,vdwjidx2C,vdwjidx2D;
1368     __m256d          jx2,jy2,jz2,fjx2,fjy2,fjz2,jq2,isaj2;
1369     int              vdwjidx3A,vdwjidx3B,vdwjidx3C,vdwjidx3D;
1370     __m256d          jx3,jy3,jz3,fjx3,fjy3,fjz3,jq3,isaj3;
1371     __m256d          dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
1372     __m256d          dx11,dy11,dz11,rsq11,rinv11,rinvsq11,r11,qq11,c6_11,c12_11;
1373     __m256d          dx12,dy12,dz12,rsq12,rinv12,rinvsq12,r12,qq12,c6_12,c12_12;
1374     __m256d          dx13,dy13,dz13,rsq13,rinv13,rinvsq13,r13,qq13,c6_13,c12_13;
1375     __m256d          dx21,dy21,dz21,rsq21,rinv21,rinvsq21,r21,qq21,c6_21,c12_21;
1376     __m256d          dx22,dy22,dz22,rsq22,rinv22,rinvsq22,r22,qq22,c6_22,c12_22;
1377     __m256d          dx23,dy23,dz23,rsq23,rinv23,rinvsq23,r23,qq23,c6_23,c12_23;
1378     __m256d          dx31,dy31,dz31,rsq31,rinv31,rinvsq31,r31,qq31,c6_31,c12_31;
1379     __m256d          dx32,dy32,dz32,rsq32,rinv32,rinvsq32,r32,qq32,c6_32,c12_32;
1380     __m256d          dx33,dy33,dz33,rsq33,rinv33,rinvsq33,r33,qq33,c6_33,c12_33;
1381     __m256d          velec,felec,velecsum,facel,crf,krf,krf2;
1382     real             *charge;
1383     int              nvdwtype;
1384     __m256d          rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
1385     int              *vdwtype;
1386     real             *vdwparam;
1387     __m256d          one_sixth   = _mm256_set1_pd(1.0/6.0);
1388     __m256d          one_twelfth = _mm256_set1_pd(1.0/12.0);
1389     __m128i          vfitab;
1390     __m128i          ifour       = _mm_set1_epi32(4);
1391     __m256d          rt,vfeps,vftabscale,Y,F,G,H,Heps,Fp,VV,FF;
1392     real             *vftab;
1393     __m256d          dummy_mask,cutoff_mask;
1394     __m128           tmpmask0,tmpmask1;
1395     __m256d          signbit = _mm256_castsi256_pd( _mm256_set1_epi32(0x80000000) );
1396     __m256d          one     = _mm256_set1_pd(1.0);
1397     __m256d          two     = _mm256_set1_pd(2.0);
1398     x                = xx[0];
1399     f                = ff[0];
1400
1401     nri              = nlist->nri;
1402     iinr             = nlist->iinr;
1403     jindex           = nlist->jindex;
1404     jjnr             = nlist->jjnr;
1405     shiftidx         = nlist->shift;
1406     gid              = nlist->gid;
1407     shiftvec         = fr->shift_vec[0];
1408     fshift           = fr->fshift[0];
1409     facel            = _mm256_set1_pd(fr->epsfac);
1410     charge           = mdatoms->chargeA;
1411     krf              = _mm256_set1_pd(fr->ic->k_rf);
1412     krf2             = _mm256_set1_pd(fr->ic->k_rf*2.0);
1413     crf              = _mm256_set1_pd(fr->ic->c_rf);
1414     nvdwtype         = fr->ntype;
1415     vdwparam         = fr->nbfp;
1416     vdwtype          = mdatoms->typeA;
1417
1418     vftab            = kernel_data->table_vdw->data;
1419     vftabscale       = _mm256_set1_pd(kernel_data->table_vdw->scale);
1420
1421     /* Setup water-specific parameters */
1422     inr              = nlist->iinr[0];
1423     iq1              = _mm256_mul_pd(facel,_mm256_set1_pd(charge[inr+1]));
1424     iq2              = _mm256_mul_pd(facel,_mm256_set1_pd(charge[inr+2]));
1425     iq3              = _mm256_mul_pd(facel,_mm256_set1_pd(charge[inr+3]));
1426     vdwioffsetptr0   = vdwparam+2*nvdwtype*vdwtype[inr+0];
1427
1428     jq1              = _mm256_set1_pd(charge[inr+1]);
1429     jq2              = _mm256_set1_pd(charge[inr+2]);
1430     jq3              = _mm256_set1_pd(charge[inr+3]);
1431     vdwjidx0A        = 2*vdwtype[inr+0];
1432     c6_00            = _mm256_set1_pd(vdwioffsetptr0[vdwjidx0A]);
1433     c12_00           = _mm256_set1_pd(vdwioffsetptr0[vdwjidx0A+1]);
1434     qq11             = _mm256_mul_pd(iq1,jq1);
1435     qq12             = _mm256_mul_pd(iq1,jq2);
1436     qq13             = _mm256_mul_pd(iq1,jq3);
1437     qq21             = _mm256_mul_pd(iq2,jq1);
1438     qq22             = _mm256_mul_pd(iq2,jq2);
1439     qq23             = _mm256_mul_pd(iq2,jq3);
1440     qq31             = _mm256_mul_pd(iq3,jq1);
1441     qq32             = _mm256_mul_pd(iq3,jq2);
1442     qq33             = _mm256_mul_pd(iq3,jq3);
1443
1444     /* When we use explicit cutoffs the value must be identical for elec and VdW, so use elec as an arbitrary choice */
1445     rcutoff_scalar   = fr->rcoulomb;
1446     rcutoff          = _mm256_set1_pd(rcutoff_scalar);
1447     rcutoff2         = _mm256_mul_pd(rcutoff,rcutoff);
1448
1449     /* Avoid stupid compiler warnings */
1450     jnrA = jnrB = jnrC = jnrD = 0;
1451     j_coord_offsetA = 0;
1452     j_coord_offsetB = 0;
1453     j_coord_offsetC = 0;
1454     j_coord_offsetD = 0;
1455
1456     outeriter        = 0;
1457     inneriter        = 0;
1458
1459     for(iidx=0;iidx<4*DIM;iidx++)
1460     {
1461         scratch[iidx] = 0.0;
1462     }
1463
1464     /* Start outer loop over neighborlists */
1465     for(iidx=0; iidx<nri; iidx++)
1466     {
1467         /* Load shift vector for this list */
1468         i_shift_offset   = DIM*shiftidx[iidx];
1469
1470         /* Load limits for loop over neighbors */
1471         j_index_start    = jindex[iidx];
1472         j_index_end      = jindex[iidx+1];
1473
1474         /* Get outer coordinate index */
1475         inr              = iinr[iidx];
1476         i_coord_offset   = DIM*inr;
1477
1478         /* Load i particle coords and add shift vector */
1479         gmx_mm256_load_shift_and_4rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset,
1480                                                     &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2,&ix3,&iy3,&iz3);
1481
1482         fix0             = _mm256_setzero_pd();
1483         fiy0             = _mm256_setzero_pd();
1484         fiz0             = _mm256_setzero_pd();
1485         fix1             = _mm256_setzero_pd();
1486         fiy1             = _mm256_setzero_pd();
1487         fiz1             = _mm256_setzero_pd();
1488         fix2             = _mm256_setzero_pd();
1489         fiy2             = _mm256_setzero_pd();
1490         fiz2             = _mm256_setzero_pd();
1491         fix3             = _mm256_setzero_pd();
1492         fiy3             = _mm256_setzero_pd();
1493         fiz3             = _mm256_setzero_pd();
1494
1495         /* Start inner kernel loop */
1496         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
1497         {
1498
1499             /* Get j neighbor index, and coordinate index */
1500             jnrA             = jjnr[jidx];
1501             jnrB             = jjnr[jidx+1];
1502             jnrC             = jjnr[jidx+2];
1503             jnrD             = jjnr[jidx+3];
1504             j_coord_offsetA  = DIM*jnrA;
1505             j_coord_offsetB  = DIM*jnrB;
1506             j_coord_offsetC  = DIM*jnrC;
1507             j_coord_offsetD  = DIM*jnrD;
1508
1509             /* load j atom coordinates */
1510             gmx_mm256_load_4rvec_4ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
1511                                                  x+j_coord_offsetC,x+j_coord_offsetD,
1512                                                  &jx0,&jy0,&jz0,&jx1,&jy1,&jz1,&jx2,
1513                                                  &jy2,&jz2,&jx3,&jy3,&jz3);
1514
1515             /* Calculate displacement vector */
1516             dx00             = _mm256_sub_pd(ix0,jx0);
1517             dy00             = _mm256_sub_pd(iy0,jy0);
1518             dz00             = _mm256_sub_pd(iz0,jz0);
1519             dx11             = _mm256_sub_pd(ix1,jx1);
1520             dy11             = _mm256_sub_pd(iy1,jy1);
1521             dz11             = _mm256_sub_pd(iz1,jz1);
1522             dx12             = _mm256_sub_pd(ix1,jx2);
1523             dy12             = _mm256_sub_pd(iy1,jy2);
1524             dz12             = _mm256_sub_pd(iz1,jz2);
1525             dx13             = _mm256_sub_pd(ix1,jx3);
1526             dy13             = _mm256_sub_pd(iy1,jy3);
1527             dz13             = _mm256_sub_pd(iz1,jz3);
1528             dx21             = _mm256_sub_pd(ix2,jx1);
1529             dy21             = _mm256_sub_pd(iy2,jy1);
1530             dz21             = _mm256_sub_pd(iz2,jz1);
1531             dx22             = _mm256_sub_pd(ix2,jx2);
1532             dy22             = _mm256_sub_pd(iy2,jy2);
1533             dz22             = _mm256_sub_pd(iz2,jz2);
1534             dx23             = _mm256_sub_pd(ix2,jx3);
1535             dy23             = _mm256_sub_pd(iy2,jy3);
1536             dz23             = _mm256_sub_pd(iz2,jz3);
1537             dx31             = _mm256_sub_pd(ix3,jx1);
1538             dy31             = _mm256_sub_pd(iy3,jy1);
1539             dz31             = _mm256_sub_pd(iz3,jz1);
1540             dx32             = _mm256_sub_pd(ix3,jx2);
1541             dy32             = _mm256_sub_pd(iy3,jy2);
1542             dz32             = _mm256_sub_pd(iz3,jz2);
1543             dx33             = _mm256_sub_pd(ix3,jx3);
1544             dy33             = _mm256_sub_pd(iy3,jy3);
1545             dz33             = _mm256_sub_pd(iz3,jz3);
1546
1547             /* Calculate squared distance and things based on it */
1548             rsq00            = gmx_mm256_calc_rsq_pd(dx00,dy00,dz00);
1549             rsq11            = gmx_mm256_calc_rsq_pd(dx11,dy11,dz11);
1550             rsq12            = gmx_mm256_calc_rsq_pd(dx12,dy12,dz12);
1551             rsq13            = gmx_mm256_calc_rsq_pd(dx13,dy13,dz13);
1552             rsq21            = gmx_mm256_calc_rsq_pd(dx21,dy21,dz21);
1553             rsq22            = gmx_mm256_calc_rsq_pd(dx22,dy22,dz22);
1554             rsq23            = gmx_mm256_calc_rsq_pd(dx23,dy23,dz23);
1555             rsq31            = gmx_mm256_calc_rsq_pd(dx31,dy31,dz31);
1556             rsq32            = gmx_mm256_calc_rsq_pd(dx32,dy32,dz32);
1557             rsq33            = gmx_mm256_calc_rsq_pd(dx33,dy33,dz33);
1558
1559             rinv00           = gmx_mm256_invsqrt_pd(rsq00);
1560             rinv11           = gmx_mm256_invsqrt_pd(rsq11);
1561             rinv12           = gmx_mm256_invsqrt_pd(rsq12);
1562             rinv13           = gmx_mm256_invsqrt_pd(rsq13);
1563             rinv21           = gmx_mm256_invsqrt_pd(rsq21);
1564             rinv22           = gmx_mm256_invsqrt_pd(rsq22);
1565             rinv23           = gmx_mm256_invsqrt_pd(rsq23);
1566             rinv31           = gmx_mm256_invsqrt_pd(rsq31);
1567             rinv32           = gmx_mm256_invsqrt_pd(rsq32);
1568             rinv33           = gmx_mm256_invsqrt_pd(rsq33);
1569
1570             rinvsq11         = _mm256_mul_pd(rinv11,rinv11);
1571             rinvsq12         = _mm256_mul_pd(rinv12,rinv12);
1572             rinvsq13         = _mm256_mul_pd(rinv13,rinv13);
1573             rinvsq21         = _mm256_mul_pd(rinv21,rinv21);
1574             rinvsq22         = _mm256_mul_pd(rinv22,rinv22);
1575             rinvsq23         = _mm256_mul_pd(rinv23,rinv23);
1576             rinvsq31         = _mm256_mul_pd(rinv31,rinv31);
1577             rinvsq32         = _mm256_mul_pd(rinv32,rinv32);
1578             rinvsq33         = _mm256_mul_pd(rinv33,rinv33);
1579
1580             fjx0             = _mm256_setzero_pd();
1581             fjy0             = _mm256_setzero_pd();
1582             fjz0             = _mm256_setzero_pd();
1583             fjx1             = _mm256_setzero_pd();
1584             fjy1             = _mm256_setzero_pd();
1585             fjz1             = _mm256_setzero_pd();
1586             fjx2             = _mm256_setzero_pd();
1587             fjy2             = _mm256_setzero_pd();
1588             fjz2             = _mm256_setzero_pd();
1589             fjx3             = _mm256_setzero_pd();
1590             fjy3             = _mm256_setzero_pd();
1591             fjz3             = _mm256_setzero_pd();
1592
1593             /**************************
1594              * CALCULATE INTERACTIONS *
1595              **************************/
1596
1597             if (gmx_mm256_any_lt(rsq00,rcutoff2))
1598             {
1599
1600             r00              = _mm256_mul_pd(rsq00,rinv00);
1601
1602             /* Calculate table index by multiplying r with table scale and truncate to integer */
1603             rt               = _mm256_mul_pd(r00,vftabscale);
1604             vfitab           = _mm256_cvttpd_epi32(rt);
1605             vfeps            = _mm256_sub_pd(rt,_mm256_round_pd(rt, _MM_FROUND_FLOOR));
1606             vfitab           = _mm_slli_epi32(vfitab,3);
1607
1608             /* CUBIC SPLINE TABLE DISPERSION */
1609             Y                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
1610             F                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
1611             G                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,2) );
1612             H                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,3) );
1613             GMX_MM256_FULLTRANSPOSE4_PD(Y,F,G,H);
1614             Heps             = _mm256_mul_pd(vfeps,H);
1615             Fp               = _mm256_add_pd(F,_mm256_mul_pd(vfeps,_mm256_add_pd(G,Heps)));
1616             FF               = _mm256_add_pd(Fp,_mm256_mul_pd(vfeps,_mm256_add_pd(G,_mm256_add_pd(Heps,Heps))));
1617             fvdw6            = _mm256_mul_pd(c6_00,FF);
1618
1619             /* CUBIC SPLINE TABLE REPULSION */
1620             vfitab           = _mm_add_epi32(vfitab,ifour);
1621             Y                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
1622             F                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
1623             G                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,2) );
1624             H                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,3) );
1625             GMX_MM256_FULLTRANSPOSE4_PD(Y,F,G,H);
1626             Heps             = _mm256_mul_pd(vfeps,H);
1627             Fp               = _mm256_add_pd(F,_mm256_mul_pd(vfeps,_mm256_add_pd(G,Heps)));
1628             FF               = _mm256_add_pd(Fp,_mm256_mul_pd(vfeps,_mm256_add_pd(G,_mm256_add_pd(Heps,Heps))));
1629             fvdw12           = _mm256_mul_pd(c12_00,FF);
1630             fvdw             = _mm256_xor_pd(signbit,_mm256_mul_pd(_mm256_add_pd(fvdw6,fvdw12),_mm256_mul_pd(vftabscale,rinv00)));
1631
1632             cutoff_mask      = _mm256_cmp_pd(rsq00,rcutoff2,_CMP_LT_OQ);
1633
1634             fscal            = fvdw;
1635
1636             fscal            = _mm256_and_pd(fscal,cutoff_mask);
1637
1638             /* Calculate temporary vectorial force */
1639             tx               = _mm256_mul_pd(fscal,dx00);
1640             ty               = _mm256_mul_pd(fscal,dy00);
1641             tz               = _mm256_mul_pd(fscal,dz00);
1642
1643             /* Update vectorial force */
1644             fix0             = _mm256_add_pd(fix0,tx);
1645             fiy0             = _mm256_add_pd(fiy0,ty);
1646             fiz0             = _mm256_add_pd(fiz0,tz);
1647
1648             fjx0             = _mm256_add_pd(fjx0,tx);
1649             fjy0             = _mm256_add_pd(fjy0,ty);
1650             fjz0             = _mm256_add_pd(fjz0,tz);
1651
1652             }
1653
1654             /**************************
1655              * CALCULATE INTERACTIONS *
1656              **************************/
1657
1658             if (gmx_mm256_any_lt(rsq11,rcutoff2))
1659             {
1660
1661             /* REACTION-FIELD ELECTROSTATICS */
1662             felec            = _mm256_mul_pd(qq11,_mm256_sub_pd(_mm256_mul_pd(rinv11,rinvsq11),krf2));
1663
1664             cutoff_mask      = _mm256_cmp_pd(rsq11,rcutoff2,_CMP_LT_OQ);
1665
1666             fscal            = felec;
1667
1668             fscal            = _mm256_and_pd(fscal,cutoff_mask);
1669
1670             /* Calculate temporary vectorial force */
1671             tx               = _mm256_mul_pd(fscal,dx11);
1672             ty               = _mm256_mul_pd(fscal,dy11);
1673             tz               = _mm256_mul_pd(fscal,dz11);
1674
1675             /* Update vectorial force */
1676             fix1             = _mm256_add_pd(fix1,tx);
1677             fiy1             = _mm256_add_pd(fiy1,ty);
1678             fiz1             = _mm256_add_pd(fiz1,tz);
1679
1680             fjx1             = _mm256_add_pd(fjx1,tx);
1681             fjy1             = _mm256_add_pd(fjy1,ty);
1682             fjz1             = _mm256_add_pd(fjz1,tz);
1683
1684             }
1685
1686             /**************************
1687              * CALCULATE INTERACTIONS *
1688              **************************/
1689
1690             if (gmx_mm256_any_lt(rsq12,rcutoff2))
1691             {
1692
1693             /* REACTION-FIELD ELECTROSTATICS */
1694             felec            = _mm256_mul_pd(qq12,_mm256_sub_pd(_mm256_mul_pd(rinv12,rinvsq12),krf2));
1695
1696             cutoff_mask      = _mm256_cmp_pd(rsq12,rcutoff2,_CMP_LT_OQ);
1697
1698             fscal            = felec;
1699
1700             fscal            = _mm256_and_pd(fscal,cutoff_mask);
1701
1702             /* Calculate temporary vectorial force */
1703             tx               = _mm256_mul_pd(fscal,dx12);
1704             ty               = _mm256_mul_pd(fscal,dy12);
1705             tz               = _mm256_mul_pd(fscal,dz12);
1706
1707             /* Update vectorial force */
1708             fix1             = _mm256_add_pd(fix1,tx);
1709             fiy1             = _mm256_add_pd(fiy1,ty);
1710             fiz1             = _mm256_add_pd(fiz1,tz);
1711
1712             fjx2             = _mm256_add_pd(fjx2,tx);
1713             fjy2             = _mm256_add_pd(fjy2,ty);
1714             fjz2             = _mm256_add_pd(fjz2,tz);
1715
1716             }
1717
1718             /**************************
1719              * CALCULATE INTERACTIONS *
1720              **************************/
1721
1722             if (gmx_mm256_any_lt(rsq13,rcutoff2))
1723             {
1724
1725             /* REACTION-FIELD ELECTROSTATICS */
1726             felec            = _mm256_mul_pd(qq13,_mm256_sub_pd(_mm256_mul_pd(rinv13,rinvsq13),krf2));
1727
1728             cutoff_mask      = _mm256_cmp_pd(rsq13,rcutoff2,_CMP_LT_OQ);
1729
1730             fscal            = felec;
1731
1732             fscal            = _mm256_and_pd(fscal,cutoff_mask);
1733
1734             /* Calculate temporary vectorial force */
1735             tx               = _mm256_mul_pd(fscal,dx13);
1736             ty               = _mm256_mul_pd(fscal,dy13);
1737             tz               = _mm256_mul_pd(fscal,dz13);
1738
1739             /* Update vectorial force */
1740             fix1             = _mm256_add_pd(fix1,tx);
1741             fiy1             = _mm256_add_pd(fiy1,ty);
1742             fiz1             = _mm256_add_pd(fiz1,tz);
1743
1744             fjx3             = _mm256_add_pd(fjx3,tx);
1745             fjy3             = _mm256_add_pd(fjy3,ty);
1746             fjz3             = _mm256_add_pd(fjz3,tz);
1747
1748             }
1749
1750             /**************************
1751              * CALCULATE INTERACTIONS *
1752              **************************/
1753
1754             if (gmx_mm256_any_lt(rsq21,rcutoff2))
1755             {
1756
1757             /* REACTION-FIELD ELECTROSTATICS */
1758             felec            = _mm256_mul_pd(qq21,_mm256_sub_pd(_mm256_mul_pd(rinv21,rinvsq21),krf2));
1759
1760             cutoff_mask      = _mm256_cmp_pd(rsq21,rcutoff2,_CMP_LT_OQ);
1761
1762             fscal            = felec;
1763
1764             fscal            = _mm256_and_pd(fscal,cutoff_mask);
1765
1766             /* Calculate temporary vectorial force */
1767             tx               = _mm256_mul_pd(fscal,dx21);
1768             ty               = _mm256_mul_pd(fscal,dy21);
1769             tz               = _mm256_mul_pd(fscal,dz21);
1770
1771             /* Update vectorial force */
1772             fix2             = _mm256_add_pd(fix2,tx);
1773             fiy2             = _mm256_add_pd(fiy2,ty);
1774             fiz2             = _mm256_add_pd(fiz2,tz);
1775
1776             fjx1             = _mm256_add_pd(fjx1,tx);
1777             fjy1             = _mm256_add_pd(fjy1,ty);
1778             fjz1             = _mm256_add_pd(fjz1,tz);
1779
1780             }
1781
1782             /**************************
1783              * CALCULATE INTERACTIONS *
1784              **************************/
1785
1786             if (gmx_mm256_any_lt(rsq22,rcutoff2))
1787             {
1788
1789             /* REACTION-FIELD ELECTROSTATICS */
1790             felec            = _mm256_mul_pd(qq22,_mm256_sub_pd(_mm256_mul_pd(rinv22,rinvsq22),krf2));
1791
1792             cutoff_mask      = _mm256_cmp_pd(rsq22,rcutoff2,_CMP_LT_OQ);
1793
1794             fscal            = felec;
1795
1796             fscal            = _mm256_and_pd(fscal,cutoff_mask);
1797
1798             /* Calculate temporary vectorial force */
1799             tx               = _mm256_mul_pd(fscal,dx22);
1800             ty               = _mm256_mul_pd(fscal,dy22);
1801             tz               = _mm256_mul_pd(fscal,dz22);
1802
1803             /* Update vectorial force */
1804             fix2             = _mm256_add_pd(fix2,tx);
1805             fiy2             = _mm256_add_pd(fiy2,ty);
1806             fiz2             = _mm256_add_pd(fiz2,tz);
1807
1808             fjx2             = _mm256_add_pd(fjx2,tx);
1809             fjy2             = _mm256_add_pd(fjy2,ty);
1810             fjz2             = _mm256_add_pd(fjz2,tz);
1811
1812             }
1813
1814             /**************************
1815              * CALCULATE INTERACTIONS *
1816              **************************/
1817
1818             if (gmx_mm256_any_lt(rsq23,rcutoff2))
1819             {
1820
1821             /* REACTION-FIELD ELECTROSTATICS */
1822             felec            = _mm256_mul_pd(qq23,_mm256_sub_pd(_mm256_mul_pd(rinv23,rinvsq23),krf2));
1823
1824             cutoff_mask      = _mm256_cmp_pd(rsq23,rcutoff2,_CMP_LT_OQ);
1825
1826             fscal            = felec;
1827
1828             fscal            = _mm256_and_pd(fscal,cutoff_mask);
1829
1830             /* Calculate temporary vectorial force */
1831             tx               = _mm256_mul_pd(fscal,dx23);
1832             ty               = _mm256_mul_pd(fscal,dy23);
1833             tz               = _mm256_mul_pd(fscal,dz23);
1834
1835             /* Update vectorial force */
1836             fix2             = _mm256_add_pd(fix2,tx);
1837             fiy2             = _mm256_add_pd(fiy2,ty);
1838             fiz2             = _mm256_add_pd(fiz2,tz);
1839
1840             fjx3             = _mm256_add_pd(fjx3,tx);
1841             fjy3             = _mm256_add_pd(fjy3,ty);
1842             fjz3             = _mm256_add_pd(fjz3,tz);
1843
1844             }
1845
1846             /**************************
1847              * CALCULATE INTERACTIONS *
1848              **************************/
1849
1850             if (gmx_mm256_any_lt(rsq31,rcutoff2))
1851             {
1852
1853             /* REACTION-FIELD ELECTROSTATICS */
1854             felec            = _mm256_mul_pd(qq31,_mm256_sub_pd(_mm256_mul_pd(rinv31,rinvsq31),krf2));
1855
1856             cutoff_mask      = _mm256_cmp_pd(rsq31,rcutoff2,_CMP_LT_OQ);
1857
1858             fscal            = felec;
1859
1860             fscal            = _mm256_and_pd(fscal,cutoff_mask);
1861
1862             /* Calculate temporary vectorial force */
1863             tx               = _mm256_mul_pd(fscal,dx31);
1864             ty               = _mm256_mul_pd(fscal,dy31);
1865             tz               = _mm256_mul_pd(fscal,dz31);
1866
1867             /* Update vectorial force */
1868             fix3             = _mm256_add_pd(fix3,tx);
1869             fiy3             = _mm256_add_pd(fiy3,ty);
1870             fiz3             = _mm256_add_pd(fiz3,tz);
1871
1872             fjx1             = _mm256_add_pd(fjx1,tx);
1873             fjy1             = _mm256_add_pd(fjy1,ty);
1874             fjz1             = _mm256_add_pd(fjz1,tz);
1875
1876             }
1877
1878             /**************************
1879              * CALCULATE INTERACTIONS *
1880              **************************/
1881
1882             if (gmx_mm256_any_lt(rsq32,rcutoff2))
1883             {
1884
1885             /* REACTION-FIELD ELECTROSTATICS */
1886             felec            = _mm256_mul_pd(qq32,_mm256_sub_pd(_mm256_mul_pd(rinv32,rinvsq32),krf2));
1887
1888             cutoff_mask      = _mm256_cmp_pd(rsq32,rcutoff2,_CMP_LT_OQ);
1889
1890             fscal            = felec;
1891
1892             fscal            = _mm256_and_pd(fscal,cutoff_mask);
1893
1894             /* Calculate temporary vectorial force */
1895             tx               = _mm256_mul_pd(fscal,dx32);
1896             ty               = _mm256_mul_pd(fscal,dy32);
1897             tz               = _mm256_mul_pd(fscal,dz32);
1898
1899             /* Update vectorial force */
1900             fix3             = _mm256_add_pd(fix3,tx);
1901             fiy3             = _mm256_add_pd(fiy3,ty);
1902             fiz3             = _mm256_add_pd(fiz3,tz);
1903
1904             fjx2             = _mm256_add_pd(fjx2,tx);
1905             fjy2             = _mm256_add_pd(fjy2,ty);
1906             fjz2             = _mm256_add_pd(fjz2,tz);
1907
1908             }
1909
1910             /**************************
1911              * CALCULATE INTERACTIONS *
1912              **************************/
1913
1914             if (gmx_mm256_any_lt(rsq33,rcutoff2))
1915             {
1916
1917             /* REACTION-FIELD ELECTROSTATICS */
1918             felec            = _mm256_mul_pd(qq33,_mm256_sub_pd(_mm256_mul_pd(rinv33,rinvsq33),krf2));
1919
1920             cutoff_mask      = _mm256_cmp_pd(rsq33,rcutoff2,_CMP_LT_OQ);
1921
1922             fscal            = felec;
1923
1924             fscal            = _mm256_and_pd(fscal,cutoff_mask);
1925
1926             /* Calculate temporary vectorial force */
1927             tx               = _mm256_mul_pd(fscal,dx33);
1928             ty               = _mm256_mul_pd(fscal,dy33);
1929             tz               = _mm256_mul_pd(fscal,dz33);
1930
1931             /* Update vectorial force */
1932             fix3             = _mm256_add_pd(fix3,tx);
1933             fiy3             = _mm256_add_pd(fiy3,ty);
1934             fiz3             = _mm256_add_pd(fiz3,tz);
1935
1936             fjx3             = _mm256_add_pd(fjx3,tx);
1937             fjy3             = _mm256_add_pd(fjy3,ty);
1938             fjz3             = _mm256_add_pd(fjz3,tz);
1939
1940             }
1941
1942             fjptrA             = f+j_coord_offsetA;
1943             fjptrB             = f+j_coord_offsetB;
1944             fjptrC             = f+j_coord_offsetC;
1945             fjptrD             = f+j_coord_offsetD;
1946
1947             gmx_mm256_decrement_4rvec_4ptr_swizzle_pd(fjptrA,fjptrB,fjptrC,fjptrD,
1948                                                       fjx0,fjy0,fjz0,fjx1,fjy1,fjz1,
1949                                                       fjx2,fjy2,fjz2,fjx3,fjy3,fjz3);
1950
1951             /* Inner loop uses 324 flops */
1952         }
1953
1954         if(jidx<j_index_end)
1955         {
1956
1957             /* Get j neighbor index, and coordinate index */
1958             jnrlistA         = jjnr[jidx];
1959             jnrlistB         = jjnr[jidx+1];
1960             jnrlistC         = jjnr[jidx+2];
1961             jnrlistD         = jjnr[jidx+3];
1962             /* Sign of each element will be negative for non-real atoms.
1963              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
1964              * so use it as val = _mm_andnot_pd(mask,val) to clear dummy entries.
1965              */
1966             tmpmask0 = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
1967
1968             tmpmask1 = _mm_permute_ps(tmpmask0,_GMX_MM_PERMUTE(3,3,2,2));
1969             tmpmask0 = _mm_permute_ps(tmpmask0,_GMX_MM_PERMUTE(1,1,0,0));
1970             dummy_mask = _mm256_castps_pd(gmx_mm256_set_m128(tmpmask1,tmpmask0));
1971
1972             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
1973             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
1974             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
1975             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
1976             j_coord_offsetA  = DIM*jnrA;
1977             j_coord_offsetB  = DIM*jnrB;
1978             j_coord_offsetC  = DIM*jnrC;
1979             j_coord_offsetD  = DIM*jnrD;
1980
1981             /* load j atom coordinates */
1982             gmx_mm256_load_4rvec_4ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
1983                                                  x+j_coord_offsetC,x+j_coord_offsetD,
1984                                                  &jx0,&jy0,&jz0,&jx1,&jy1,&jz1,&jx2,
1985                                                  &jy2,&jz2,&jx3,&jy3,&jz3);
1986
1987             /* Calculate displacement vector */
1988             dx00             = _mm256_sub_pd(ix0,jx0);
1989             dy00             = _mm256_sub_pd(iy0,jy0);
1990             dz00             = _mm256_sub_pd(iz0,jz0);
1991             dx11             = _mm256_sub_pd(ix1,jx1);
1992             dy11             = _mm256_sub_pd(iy1,jy1);
1993             dz11             = _mm256_sub_pd(iz1,jz1);
1994             dx12             = _mm256_sub_pd(ix1,jx2);
1995             dy12             = _mm256_sub_pd(iy1,jy2);
1996             dz12             = _mm256_sub_pd(iz1,jz2);
1997             dx13             = _mm256_sub_pd(ix1,jx3);
1998             dy13             = _mm256_sub_pd(iy1,jy3);
1999             dz13             = _mm256_sub_pd(iz1,jz3);
2000             dx21             = _mm256_sub_pd(ix2,jx1);
2001             dy21             = _mm256_sub_pd(iy2,jy1);
2002             dz21             = _mm256_sub_pd(iz2,jz1);
2003             dx22             = _mm256_sub_pd(ix2,jx2);
2004             dy22             = _mm256_sub_pd(iy2,jy2);
2005             dz22             = _mm256_sub_pd(iz2,jz2);
2006             dx23             = _mm256_sub_pd(ix2,jx3);
2007             dy23             = _mm256_sub_pd(iy2,jy3);
2008             dz23             = _mm256_sub_pd(iz2,jz3);
2009             dx31             = _mm256_sub_pd(ix3,jx1);
2010             dy31             = _mm256_sub_pd(iy3,jy1);
2011             dz31             = _mm256_sub_pd(iz3,jz1);
2012             dx32             = _mm256_sub_pd(ix3,jx2);
2013             dy32             = _mm256_sub_pd(iy3,jy2);
2014             dz32             = _mm256_sub_pd(iz3,jz2);
2015             dx33             = _mm256_sub_pd(ix3,jx3);
2016             dy33             = _mm256_sub_pd(iy3,jy3);
2017             dz33             = _mm256_sub_pd(iz3,jz3);
2018
2019             /* Calculate squared distance and things based on it */
2020             rsq00            = gmx_mm256_calc_rsq_pd(dx00,dy00,dz00);
2021             rsq11            = gmx_mm256_calc_rsq_pd(dx11,dy11,dz11);
2022             rsq12            = gmx_mm256_calc_rsq_pd(dx12,dy12,dz12);
2023             rsq13            = gmx_mm256_calc_rsq_pd(dx13,dy13,dz13);
2024             rsq21            = gmx_mm256_calc_rsq_pd(dx21,dy21,dz21);
2025             rsq22            = gmx_mm256_calc_rsq_pd(dx22,dy22,dz22);
2026             rsq23            = gmx_mm256_calc_rsq_pd(dx23,dy23,dz23);
2027             rsq31            = gmx_mm256_calc_rsq_pd(dx31,dy31,dz31);
2028             rsq32            = gmx_mm256_calc_rsq_pd(dx32,dy32,dz32);
2029             rsq33            = gmx_mm256_calc_rsq_pd(dx33,dy33,dz33);
2030
2031             rinv00           = gmx_mm256_invsqrt_pd(rsq00);
2032             rinv11           = gmx_mm256_invsqrt_pd(rsq11);
2033             rinv12           = gmx_mm256_invsqrt_pd(rsq12);
2034             rinv13           = gmx_mm256_invsqrt_pd(rsq13);
2035             rinv21           = gmx_mm256_invsqrt_pd(rsq21);
2036             rinv22           = gmx_mm256_invsqrt_pd(rsq22);
2037             rinv23           = gmx_mm256_invsqrt_pd(rsq23);
2038             rinv31           = gmx_mm256_invsqrt_pd(rsq31);
2039             rinv32           = gmx_mm256_invsqrt_pd(rsq32);
2040             rinv33           = gmx_mm256_invsqrt_pd(rsq33);
2041
2042             rinvsq11         = _mm256_mul_pd(rinv11,rinv11);
2043             rinvsq12         = _mm256_mul_pd(rinv12,rinv12);
2044             rinvsq13         = _mm256_mul_pd(rinv13,rinv13);
2045             rinvsq21         = _mm256_mul_pd(rinv21,rinv21);
2046             rinvsq22         = _mm256_mul_pd(rinv22,rinv22);
2047             rinvsq23         = _mm256_mul_pd(rinv23,rinv23);
2048             rinvsq31         = _mm256_mul_pd(rinv31,rinv31);
2049             rinvsq32         = _mm256_mul_pd(rinv32,rinv32);
2050             rinvsq33         = _mm256_mul_pd(rinv33,rinv33);
2051
2052             fjx0             = _mm256_setzero_pd();
2053             fjy0             = _mm256_setzero_pd();
2054             fjz0             = _mm256_setzero_pd();
2055             fjx1             = _mm256_setzero_pd();
2056             fjy1             = _mm256_setzero_pd();
2057             fjz1             = _mm256_setzero_pd();
2058             fjx2             = _mm256_setzero_pd();
2059             fjy2             = _mm256_setzero_pd();
2060             fjz2             = _mm256_setzero_pd();
2061             fjx3             = _mm256_setzero_pd();
2062             fjy3             = _mm256_setzero_pd();
2063             fjz3             = _mm256_setzero_pd();
2064
2065             /**************************
2066              * CALCULATE INTERACTIONS *
2067              **************************/
2068
2069             if (gmx_mm256_any_lt(rsq00,rcutoff2))
2070             {
2071
2072             r00              = _mm256_mul_pd(rsq00,rinv00);
2073             r00              = _mm256_andnot_pd(dummy_mask,r00);
2074
2075             /* Calculate table index by multiplying r with table scale and truncate to integer */
2076             rt               = _mm256_mul_pd(r00,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,3);
2080
2081             /* CUBIC SPLINE TABLE DISPERSION */
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             fvdw6            = _mm256_mul_pd(c6_00,FF);
2091
2092             /* CUBIC SPLINE TABLE REPULSION */
2093             vfitab           = _mm_add_epi32(vfitab,ifour);
2094             Y                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
2095             F                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
2096             G                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,2) );
2097             H                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,3) );
2098             GMX_MM256_FULLTRANSPOSE4_PD(Y,F,G,H);
2099             Heps             = _mm256_mul_pd(vfeps,H);
2100             Fp               = _mm256_add_pd(F,_mm256_mul_pd(vfeps,_mm256_add_pd(G,Heps)));
2101             FF               = _mm256_add_pd(Fp,_mm256_mul_pd(vfeps,_mm256_add_pd(G,_mm256_add_pd(Heps,Heps))));
2102             fvdw12           = _mm256_mul_pd(c12_00,FF);
2103             fvdw             = _mm256_xor_pd(signbit,_mm256_mul_pd(_mm256_add_pd(fvdw6,fvdw12),_mm256_mul_pd(vftabscale,rinv00)));
2104
2105             cutoff_mask      = _mm256_cmp_pd(rsq00,rcutoff2,_CMP_LT_OQ);
2106
2107             fscal            = fvdw;
2108
2109             fscal            = _mm256_and_pd(fscal,cutoff_mask);
2110
2111             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
2112
2113             /* Calculate temporary vectorial force */
2114             tx               = _mm256_mul_pd(fscal,dx00);
2115             ty               = _mm256_mul_pd(fscal,dy00);
2116             tz               = _mm256_mul_pd(fscal,dz00);
2117
2118             /* Update vectorial force */
2119             fix0             = _mm256_add_pd(fix0,tx);
2120             fiy0             = _mm256_add_pd(fiy0,ty);
2121             fiz0             = _mm256_add_pd(fiz0,tz);
2122
2123             fjx0             = _mm256_add_pd(fjx0,tx);
2124             fjy0             = _mm256_add_pd(fjy0,ty);
2125             fjz0             = _mm256_add_pd(fjz0,tz);
2126
2127             }
2128
2129             /**************************
2130              * CALCULATE INTERACTIONS *
2131              **************************/
2132
2133             if (gmx_mm256_any_lt(rsq11,rcutoff2))
2134             {
2135
2136             /* REACTION-FIELD ELECTROSTATICS */
2137             felec            = _mm256_mul_pd(qq11,_mm256_sub_pd(_mm256_mul_pd(rinv11,rinvsq11),krf2));
2138
2139             cutoff_mask      = _mm256_cmp_pd(rsq11,rcutoff2,_CMP_LT_OQ);
2140
2141             fscal            = felec;
2142
2143             fscal            = _mm256_and_pd(fscal,cutoff_mask);
2144
2145             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
2146
2147             /* Calculate temporary vectorial force */
2148             tx               = _mm256_mul_pd(fscal,dx11);
2149             ty               = _mm256_mul_pd(fscal,dy11);
2150             tz               = _mm256_mul_pd(fscal,dz11);
2151
2152             /* Update vectorial force */
2153             fix1             = _mm256_add_pd(fix1,tx);
2154             fiy1             = _mm256_add_pd(fiy1,ty);
2155             fiz1             = _mm256_add_pd(fiz1,tz);
2156
2157             fjx1             = _mm256_add_pd(fjx1,tx);
2158             fjy1             = _mm256_add_pd(fjy1,ty);
2159             fjz1             = _mm256_add_pd(fjz1,tz);
2160
2161             }
2162
2163             /**************************
2164              * CALCULATE INTERACTIONS *
2165              **************************/
2166
2167             if (gmx_mm256_any_lt(rsq12,rcutoff2))
2168             {
2169
2170             /* REACTION-FIELD ELECTROSTATICS */
2171             felec            = _mm256_mul_pd(qq12,_mm256_sub_pd(_mm256_mul_pd(rinv12,rinvsq12),krf2));
2172
2173             cutoff_mask      = _mm256_cmp_pd(rsq12,rcutoff2,_CMP_LT_OQ);
2174
2175             fscal            = felec;
2176
2177             fscal            = _mm256_and_pd(fscal,cutoff_mask);
2178
2179             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
2180
2181             /* Calculate temporary vectorial force */
2182             tx               = _mm256_mul_pd(fscal,dx12);
2183             ty               = _mm256_mul_pd(fscal,dy12);
2184             tz               = _mm256_mul_pd(fscal,dz12);
2185
2186             /* Update vectorial force */
2187             fix1             = _mm256_add_pd(fix1,tx);
2188             fiy1             = _mm256_add_pd(fiy1,ty);
2189             fiz1             = _mm256_add_pd(fiz1,tz);
2190
2191             fjx2             = _mm256_add_pd(fjx2,tx);
2192             fjy2             = _mm256_add_pd(fjy2,ty);
2193             fjz2             = _mm256_add_pd(fjz2,tz);
2194
2195             }
2196
2197             /**************************
2198              * CALCULATE INTERACTIONS *
2199              **************************/
2200
2201             if (gmx_mm256_any_lt(rsq13,rcutoff2))
2202             {
2203
2204             /* REACTION-FIELD ELECTROSTATICS */
2205             felec            = _mm256_mul_pd(qq13,_mm256_sub_pd(_mm256_mul_pd(rinv13,rinvsq13),krf2));
2206
2207             cutoff_mask      = _mm256_cmp_pd(rsq13,rcutoff2,_CMP_LT_OQ);
2208
2209             fscal            = felec;
2210
2211             fscal            = _mm256_and_pd(fscal,cutoff_mask);
2212
2213             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
2214
2215             /* Calculate temporary vectorial force */
2216             tx               = _mm256_mul_pd(fscal,dx13);
2217             ty               = _mm256_mul_pd(fscal,dy13);
2218             tz               = _mm256_mul_pd(fscal,dz13);
2219
2220             /* Update vectorial force */
2221             fix1             = _mm256_add_pd(fix1,tx);
2222             fiy1             = _mm256_add_pd(fiy1,ty);
2223             fiz1             = _mm256_add_pd(fiz1,tz);
2224
2225             fjx3             = _mm256_add_pd(fjx3,tx);
2226             fjy3             = _mm256_add_pd(fjy3,ty);
2227             fjz3             = _mm256_add_pd(fjz3,tz);
2228
2229             }
2230
2231             /**************************
2232              * CALCULATE INTERACTIONS *
2233              **************************/
2234
2235             if (gmx_mm256_any_lt(rsq21,rcutoff2))
2236             {
2237
2238             /* REACTION-FIELD ELECTROSTATICS */
2239             felec            = _mm256_mul_pd(qq21,_mm256_sub_pd(_mm256_mul_pd(rinv21,rinvsq21),krf2));
2240
2241             cutoff_mask      = _mm256_cmp_pd(rsq21,rcutoff2,_CMP_LT_OQ);
2242
2243             fscal            = felec;
2244
2245             fscal            = _mm256_and_pd(fscal,cutoff_mask);
2246
2247             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
2248
2249             /* Calculate temporary vectorial force */
2250             tx               = _mm256_mul_pd(fscal,dx21);
2251             ty               = _mm256_mul_pd(fscal,dy21);
2252             tz               = _mm256_mul_pd(fscal,dz21);
2253
2254             /* Update vectorial force */
2255             fix2             = _mm256_add_pd(fix2,tx);
2256             fiy2             = _mm256_add_pd(fiy2,ty);
2257             fiz2             = _mm256_add_pd(fiz2,tz);
2258
2259             fjx1             = _mm256_add_pd(fjx1,tx);
2260             fjy1             = _mm256_add_pd(fjy1,ty);
2261             fjz1             = _mm256_add_pd(fjz1,tz);
2262
2263             }
2264
2265             /**************************
2266              * CALCULATE INTERACTIONS *
2267              **************************/
2268
2269             if (gmx_mm256_any_lt(rsq22,rcutoff2))
2270             {
2271
2272             /* REACTION-FIELD ELECTROSTATICS */
2273             felec            = _mm256_mul_pd(qq22,_mm256_sub_pd(_mm256_mul_pd(rinv22,rinvsq22),krf2));
2274
2275             cutoff_mask      = _mm256_cmp_pd(rsq22,rcutoff2,_CMP_LT_OQ);
2276
2277             fscal            = felec;
2278
2279             fscal            = _mm256_and_pd(fscal,cutoff_mask);
2280
2281             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
2282
2283             /* Calculate temporary vectorial force */
2284             tx               = _mm256_mul_pd(fscal,dx22);
2285             ty               = _mm256_mul_pd(fscal,dy22);
2286             tz               = _mm256_mul_pd(fscal,dz22);
2287
2288             /* Update vectorial force */
2289             fix2             = _mm256_add_pd(fix2,tx);
2290             fiy2             = _mm256_add_pd(fiy2,ty);
2291             fiz2             = _mm256_add_pd(fiz2,tz);
2292
2293             fjx2             = _mm256_add_pd(fjx2,tx);
2294             fjy2             = _mm256_add_pd(fjy2,ty);
2295             fjz2             = _mm256_add_pd(fjz2,tz);
2296
2297             }
2298
2299             /**************************
2300              * CALCULATE INTERACTIONS *
2301              **************************/
2302
2303             if (gmx_mm256_any_lt(rsq23,rcutoff2))
2304             {
2305
2306             /* REACTION-FIELD ELECTROSTATICS */
2307             felec            = _mm256_mul_pd(qq23,_mm256_sub_pd(_mm256_mul_pd(rinv23,rinvsq23),krf2));
2308
2309             cutoff_mask      = _mm256_cmp_pd(rsq23,rcutoff2,_CMP_LT_OQ);
2310
2311             fscal            = felec;
2312
2313             fscal            = _mm256_and_pd(fscal,cutoff_mask);
2314
2315             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
2316
2317             /* Calculate temporary vectorial force */
2318             tx               = _mm256_mul_pd(fscal,dx23);
2319             ty               = _mm256_mul_pd(fscal,dy23);
2320             tz               = _mm256_mul_pd(fscal,dz23);
2321
2322             /* Update vectorial force */
2323             fix2             = _mm256_add_pd(fix2,tx);
2324             fiy2             = _mm256_add_pd(fiy2,ty);
2325             fiz2             = _mm256_add_pd(fiz2,tz);
2326
2327             fjx3             = _mm256_add_pd(fjx3,tx);
2328             fjy3             = _mm256_add_pd(fjy3,ty);
2329             fjz3             = _mm256_add_pd(fjz3,tz);
2330
2331             }
2332
2333             /**************************
2334              * CALCULATE INTERACTIONS *
2335              **************************/
2336
2337             if (gmx_mm256_any_lt(rsq31,rcutoff2))
2338             {
2339
2340             /* REACTION-FIELD ELECTROSTATICS */
2341             felec            = _mm256_mul_pd(qq31,_mm256_sub_pd(_mm256_mul_pd(rinv31,rinvsq31),krf2));
2342
2343             cutoff_mask      = _mm256_cmp_pd(rsq31,rcutoff2,_CMP_LT_OQ);
2344
2345             fscal            = felec;
2346
2347             fscal            = _mm256_and_pd(fscal,cutoff_mask);
2348
2349             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
2350
2351             /* Calculate temporary vectorial force */
2352             tx               = _mm256_mul_pd(fscal,dx31);
2353             ty               = _mm256_mul_pd(fscal,dy31);
2354             tz               = _mm256_mul_pd(fscal,dz31);
2355
2356             /* Update vectorial force */
2357             fix3             = _mm256_add_pd(fix3,tx);
2358             fiy3             = _mm256_add_pd(fiy3,ty);
2359             fiz3             = _mm256_add_pd(fiz3,tz);
2360
2361             fjx1             = _mm256_add_pd(fjx1,tx);
2362             fjy1             = _mm256_add_pd(fjy1,ty);
2363             fjz1             = _mm256_add_pd(fjz1,tz);
2364
2365             }
2366
2367             /**************************
2368              * CALCULATE INTERACTIONS *
2369              **************************/
2370
2371             if (gmx_mm256_any_lt(rsq32,rcutoff2))
2372             {
2373
2374             /* REACTION-FIELD ELECTROSTATICS */
2375             felec            = _mm256_mul_pd(qq32,_mm256_sub_pd(_mm256_mul_pd(rinv32,rinvsq32),krf2));
2376
2377             cutoff_mask      = _mm256_cmp_pd(rsq32,rcutoff2,_CMP_LT_OQ);
2378
2379             fscal            = felec;
2380
2381             fscal            = _mm256_and_pd(fscal,cutoff_mask);
2382
2383             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
2384
2385             /* Calculate temporary vectorial force */
2386             tx               = _mm256_mul_pd(fscal,dx32);
2387             ty               = _mm256_mul_pd(fscal,dy32);
2388             tz               = _mm256_mul_pd(fscal,dz32);
2389
2390             /* Update vectorial force */
2391             fix3             = _mm256_add_pd(fix3,tx);
2392             fiy3             = _mm256_add_pd(fiy3,ty);
2393             fiz3             = _mm256_add_pd(fiz3,tz);
2394
2395             fjx2             = _mm256_add_pd(fjx2,tx);
2396             fjy2             = _mm256_add_pd(fjy2,ty);
2397             fjz2             = _mm256_add_pd(fjz2,tz);
2398
2399             }
2400
2401             /**************************
2402              * CALCULATE INTERACTIONS *
2403              **************************/
2404
2405             if (gmx_mm256_any_lt(rsq33,rcutoff2))
2406             {
2407
2408             /* REACTION-FIELD ELECTROSTATICS */
2409             felec            = _mm256_mul_pd(qq33,_mm256_sub_pd(_mm256_mul_pd(rinv33,rinvsq33),krf2));
2410
2411             cutoff_mask      = _mm256_cmp_pd(rsq33,rcutoff2,_CMP_LT_OQ);
2412
2413             fscal            = felec;
2414
2415             fscal            = _mm256_and_pd(fscal,cutoff_mask);
2416
2417             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
2418
2419             /* Calculate temporary vectorial force */
2420             tx               = _mm256_mul_pd(fscal,dx33);
2421             ty               = _mm256_mul_pd(fscal,dy33);
2422             tz               = _mm256_mul_pd(fscal,dz33);
2423
2424             /* Update vectorial force */
2425             fix3             = _mm256_add_pd(fix3,tx);
2426             fiy3             = _mm256_add_pd(fiy3,ty);
2427             fiz3             = _mm256_add_pd(fiz3,tz);
2428
2429             fjx3             = _mm256_add_pd(fjx3,tx);
2430             fjy3             = _mm256_add_pd(fjy3,ty);
2431             fjz3             = _mm256_add_pd(fjz3,tz);
2432
2433             }
2434
2435             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
2436             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
2437             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
2438             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
2439
2440             gmx_mm256_decrement_4rvec_4ptr_swizzle_pd(fjptrA,fjptrB,fjptrC,fjptrD,
2441                                                       fjx0,fjy0,fjz0,fjx1,fjy1,fjz1,
2442                                                       fjx2,fjy2,fjz2,fjx3,fjy3,fjz3);
2443
2444             /* Inner loop uses 325 flops */
2445         }
2446
2447         /* End of innermost loop */
2448
2449         gmx_mm256_update_iforce_4atom_swizzle_pd(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,fix3,fiy3,fiz3,
2450                                                  f+i_coord_offset,fshift+i_shift_offset);
2451
2452         /* Increment number of inner iterations */
2453         inneriter                  += j_index_end - j_index_start;
2454
2455         /* Outer loop uses 24 flops */
2456     }
2457
2458     /* Increment number of outer iterations */
2459     outeriter        += nri;
2460
2461     /* Update outer/inner flops */
2462
2463     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W4W4_F,outeriter*24 + inneriter*325);
2464 }