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, 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 #ifdef HAVE_CONFIG_H
  39 #include <config.h>
  40 #endif
  41
  42 #include <math.h>
  43
  44 #include "../nb_kernel.h"
  45 #include "types/simple.h"
  46 #include "vec.h"
  47 #include "nrnb.h"
  48
  49 #include "gromacs/simd/math_x86_avx_256_double.h"
  50 #include "kernelutil_x86_avx_256_double.h"
  51
  52 /*
  53  * Gromacs nonbonded kernel:   nb_kernel_ElecRFCut_VdwCSTab_GeomW4W4_VF_avx_256_double
  54  * Electrostatics interaction: ReactionField
  55  * VdW interaction:            CubicSplineTable
  56  * Geometry:                   Water4-Water4
  57  * Calculate force/pot:        PotentialAndForce
  58  */
  59 void
  60 nb_kernel_ElecRFCut_VdwCSTab_GeomW4W4_VF_avx_256_double
  61                     (t_nblist                    * gmx_restrict       nlist,
  62                      rvec                        * gmx_restrict          xx,
  63                      rvec                        * gmx_restrict          ff,
  64                      t_forcerec                  * gmx_restrict          fr,
  65                      t_mdatoms                   * gmx_restrict     mdatoms,
  66                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
  67                      t_nrnb                      * gmx_restrict        nrnb)
  68 {
  69     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
  70      * just 0 for non-waters.
  71      * Suffixes A,B,C,D refer to j loop unrolling done with AVX, e.g. for the four different
  72      * jnr indices corresponding to data put in the four positions in the SIMD register.
  73      */
  74     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
  75     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
  76     int              jnrA,jnrB,jnrC,jnrD;
  77     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
  78     int              jnrlistE,jnrlistF,jnrlistG,jnrlistH;
  79     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
  80     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
  81     real             rcutoff_scalar;
  82     real             *shiftvec,*fshift,*x,*f;
  83     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD;
  84     real             scratch[4*DIM];
  85     __m256d          tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
  86     real *           vdwioffsetptr0;
  87     __m256d          ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
  88     real *           vdwioffsetptr1;
  89     __m256d          ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
  90     real *           vdwioffsetptr2;
  91     __m256d          ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
  92     real *           vdwioffsetptr3;
  93     __m256d          ix3,iy3,iz3,fix3,fiy3,fiz3,iq3,isai3;
  94     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
  95     __m256d          jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
  96     int              vdwjidx1A,vdwjidx1B,vdwjidx1C,vdwjidx1D;
  97     __m256d          jx1,jy1,jz1,fjx1,fjy1,fjz1,jq1,isaj1;
  98     int              vdwjidx2A,vdwjidx2B,vdwjidx2C,vdwjidx2D;
  99     __m256d          jx2,jy2,jz2,fjx2,fjy2,fjz2,jq2,isaj2;
 100     int              vdwjidx3A,vdwjidx3B,vdwjidx3C,vdwjidx3D;
 101     __m256d          jx3,jy3,jz3,fjx3,fjy3,fjz3,jq3,isaj3;
 102     __m256d          dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
 103     __m256d          dx11,dy11,dz11,rsq11,rinv11,rinvsq11,r11,qq11,c6_11,c12_11;
 104     __m256d          dx12,dy12,dz12,rsq12,rinv12,rinvsq12,r12,qq12,c6_12,c12_12;
 105     __m256d          dx13,dy13,dz13,rsq13,rinv13,rinvsq13,r13,qq13,c6_13,c12_13;
 106     __m256d          dx21,dy21,dz21,rsq21,rinv21,rinvsq21,r21,qq21,c6_21,c12_21;
 107     __m256d          dx22,dy22,dz22,rsq22,rinv22,rinvsq22,r22,qq22,c6_22,c12_22;
 108     __m256d          dx23,dy23,dz23,rsq23,rinv23,rinvsq23,r23,qq23,c6_23,c12_23;
 109     __m256d          dx31,dy31,dz31,rsq31,rinv31,rinvsq31,r31,qq31,c6_31,c12_31;
 110     __m256d          dx32,dy32,dz32,rsq32,rinv32,rinvsq32,r32,qq32,c6_32,c12_32;
 111     __m256d          dx33,dy33,dz33,rsq33,rinv33,rinvsq33,r33,qq33,c6_33,c12_33;
 112     __m256d          velec,felec,velecsum,facel,crf,krf,krf2;
 113     real             *charge;
 114     int              nvdwtype;
 115     __m256d          rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
 116     int              *vdwtype;
 117     real             *vdwparam;
 118     __m256d          one_sixth   = _mm256_set1_pd(1.0/6.0);
 119     __m256d          one_twelfth = _mm256_set1_pd(1.0/12.0);
 120     __m128i          vfitab;
 121     __m128i          ifour       = _mm_set1_epi32(4);
 122     __m256d          rt,vfeps,vftabscale,Y,F,G,H,Heps,Fp,VV,FF;
 123     real             *vftab;
 124     __m256d          dummy_mask,cutoff_mask;
 125     __m128           tmpmask0,tmpmask1;
 126     __m256d          signbit = _mm256_castsi256_pd( _mm256_set1_epi32(0x80000000) );
 127     __m256d          one     = _mm256_set1_pd(1.0);
 128     __m256d          two     = _mm256_set1_pd(2.0);
 129     x                = xx[0];
 130     f                = ff[0];
 131
 132     nri              = nlist->nri;
 133     iinr             = nlist->iinr;
 134     jindex           = nlist->jindex;
 135     jjnr             = nlist->jjnr;
 136     shiftidx         = nlist->shift;
 137     gid              = nlist->gid;
 138     shiftvec         = fr->shift_vec[0];
 139     fshift           = fr->fshift[0];
 140     facel            = _mm256_set1_pd(fr->epsfac);
 141     charge           = mdatoms->chargeA;
 142     krf              = _mm256_set1_pd(fr->ic->k_rf);
 143     krf2             = _mm256_set1_pd(fr->ic->k_rf*2.0);
 144     crf              = _mm256_set1_pd(fr->ic->c_rf);
 145     nvdwtype         = fr->ntype;
 146     vdwparam         = fr->nbfp;
 147     vdwtype          = mdatoms->typeA;
 148
 149     vftab            = kernel_data->table_vdw->data;
 150     vftabscale       = _mm256_set1_pd(kernel_data->table_vdw->scale);
 151
 152     /* Setup water-specific parameters */
 153     inr              = nlist->iinr[0];
 154     iq1              = _mm256_mul_pd(facel,_mm256_set1_pd(charge[inr+1]));
 155     iq2              = _mm256_mul_pd(facel,_mm256_set1_pd(charge[inr+2]));
 156     iq3              = _mm256_mul_pd(facel,_mm256_set1_pd(charge[inr+3]));
 157     vdwioffsetptr0   = vdwparam+2*nvdwtype*vdwtype[inr+0];
 158
 159     jq1              = _mm256_set1_pd(charge[inr+1]);
 160     jq2              = _mm256_set1_pd(charge[inr+2]);
 161     jq3              = _mm256_set1_pd(charge[inr+3]);
 162     vdwjidx0A        = 2*vdwtype[inr+0];
 163     c6_00            = _mm256_set1_pd(vdwioffsetptr0[vdwjidx0A]);
 164     c12_00           = _mm256_set1_pd(vdwioffsetptr0[vdwjidx0A+1]);
 165     qq11             = _mm256_mul_pd(iq1,jq1);
 166     qq12             = _mm256_mul_pd(iq1,jq2);
 167     qq13             = _mm256_mul_pd(iq1,jq3);
 168     qq21             = _mm256_mul_pd(iq2,jq1);
 169     qq22             = _mm256_mul_pd(iq2,jq2);
 170     qq23             = _mm256_mul_pd(iq2,jq3);
 171     qq31             = _mm256_mul_pd(iq3,jq1);
 172     qq32             = _mm256_mul_pd(iq3,jq2);
 173     qq33             = _mm256_mul_pd(iq3,jq3);
 174
 175     /* When we use explicit cutoffs the value must be identical for elec and VdW, so use elec as an arbitrary choice */
 176     rcutoff_scalar   = fr->rcoulomb;
 177     rcutoff          = _mm256_set1_pd(rcutoff_scalar);
 178     rcutoff2         = _mm256_mul_pd(rcutoff,rcutoff);
 179
 180     /* Avoid stupid compiler warnings */
 181     jnrA = jnrB = jnrC = jnrD = 0;
 182     j_coord_offsetA = 0;
 183     j_coord_offsetB = 0;
 184     j_coord_offsetC = 0;
 185     j_coord_offsetD = 0;
 186
 187     outeriter        = 0;
 188     inneriter        = 0;
 189
 190     for(iidx=0;iidx<4*DIM;iidx++)
 191     {
 192         scratch[iidx] = 0.0;
 193     }
 194
 195     /* Start outer loop over neighborlists */
 196     for(iidx=0; iidx<nri; iidx++)
 197     {
 198         /* Load shift vector for this list */
 199         i_shift_offset   = DIM*shiftidx[iidx];
 200
 201         /* Load limits for loop over neighbors */
 202         j_index_start    = jindex[iidx];
 203         j_index_end      = jindex[iidx+1];
 204
 205         /* Get outer coordinate index */
 206         inr              = iinr[iidx];
 207         i_coord_offset   = DIM*inr;
 208
 209         /* Load i particle coords and add shift vector */
 210         gmx_mm256_load_shift_and_4rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset,
 211                                                     &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2,&ix3,&iy3,&iz3);
 212
 213         fix0             = _mm256_setzero_pd();
 214         fiy0             = _mm256_setzero_pd();
 215         fiz0             = _mm256_setzero_pd();
 216         fix1             = _mm256_setzero_pd();
 217         fiy1             = _mm256_setzero_pd();
 218         fiz1             = _mm256_setzero_pd();
 219         fix2             = _mm256_setzero_pd();
 220         fiy2             = _mm256_setzero_pd();
 221         fiz2             = _mm256_setzero_pd();
 222         fix3             = _mm256_setzero_pd();
 223         fiy3             = _mm256_setzero_pd();
 224         fiz3             = _mm256_setzero_pd();
 225
 226         /* Reset potential sums */
 227         velecsum         = _mm256_setzero_pd();
 228         vvdwsum          = _mm256_setzero_pd();
 229
 230         /* Start inner kernel loop */
 231         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
 232         {
 233
 234             /* Get j neighbor index, and coordinate index */
 235             jnrA             = jjnr[jidx];
 236             jnrB             = jjnr[jidx+1];
 237             jnrC             = jjnr[jidx+2];
 238             jnrD             = jjnr[jidx+3];
 239             j_coord_offsetA  = DIM*jnrA;
 240             j_coord_offsetB  = DIM*jnrB;
 241             j_coord_offsetC  = DIM*jnrC;
 242             j_coord_offsetD  = DIM*jnrD;
 243
 244             /* load j atom coordinates */
 245             gmx_mm256_load_4rvec_4ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 246                                                  x+j_coord_offsetC,x+j_coord_offsetD,
 247                                                  &jx0,&jy0,&jz0,&jx1,&jy1,&jz1,&jx2,
 248                                                  &jy2,&jz2,&jx3,&jy3,&jz3);
 249
 250             /* Calculate displacement vector */
 251             dx00             = _mm256_sub_pd(ix0,jx0);
 252             dy00             = _mm256_sub_pd(iy0,jy0);
 253             dz00             = _mm256_sub_pd(iz0,jz0);
 254             dx11             = _mm256_sub_pd(ix1,jx1);
 255             dy11             = _mm256_sub_pd(iy1,jy1);
 256             dz11             = _mm256_sub_pd(iz1,jz1);
 257             dx12             = _mm256_sub_pd(ix1,jx2);
 258             dy12             = _mm256_sub_pd(iy1,jy2);
 259             dz12             = _mm256_sub_pd(iz1,jz2);
 260             dx13             = _mm256_sub_pd(ix1,jx3);
 261             dy13             = _mm256_sub_pd(iy1,jy3);
 262             dz13             = _mm256_sub_pd(iz1,jz3);
 263             dx21             = _mm256_sub_pd(ix2,jx1);
 264             dy21             = _mm256_sub_pd(iy2,jy1);
 265             dz21             = _mm256_sub_pd(iz2,jz1);
 266             dx22             = _mm256_sub_pd(ix2,jx2);
 267             dy22             = _mm256_sub_pd(iy2,jy2);
 268             dz22             = _mm256_sub_pd(iz2,jz2);
 269             dx23             = _mm256_sub_pd(ix2,jx3);
 270             dy23             = _mm256_sub_pd(iy2,jy3);
 271             dz23             = _mm256_sub_pd(iz2,jz3);
 272             dx31             = _mm256_sub_pd(ix3,jx1);
 273             dy31             = _mm256_sub_pd(iy3,jy1);
 274             dz31             = _mm256_sub_pd(iz3,jz1);
 275             dx32             = _mm256_sub_pd(ix3,jx2);
 276             dy32             = _mm256_sub_pd(iy3,jy2);
 277             dz32             = _mm256_sub_pd(iz3,jz2);
 278             dx33             = _mm256_sub_pd(ix3,jx3);
 279             dy33             = _mm256_sub_pd(iy3,jy3);
 280             dz33             = _mm256_sub_pd(iz3,jz3);
 281
 282             /* Calculate squared distance and things based on it */
 283             rsq00            = gmx_mm256_calc_rsq_pd(dx00,dy00,dz00);
 284             rsq11            = gmx_mm256_calc_rsq_pd(dx11,dy11,dz11);
 285             rsq12            = gmx_mm256_calc_rsq_pd(dx12,dy12,dz12);
 286             rsq13            = gmx_mm256_calc_rsq_pd(dx13,dy13,dz13);
 287             rsq21            = gmx_mm256_calc_rsq_pd(dx21,dy21,dz21);
 288             rsq22            = gmx_mm256_calc_rsq_pd(dx22,dy22,dz22);
 289             rsq23            = gmx_mm256_calc_rsq_pd(dx23,dy23,dz23);
 290             rsq31            = gmx_mm256_calc_rsq_pd(dx31,dy31,dz31);
 291             rsq32            = gmx_mm256_calc_rsq_pd(dx32,dy32,dz32);
 292             rsq33            = gmx_mm256_calc_rsq_pd(dx33,dy33,dz33);
 293
 294             rinv00           = gmx_mm256_invsqrt_pd(rsq00);
 295             rinv11           = gmx_mm256_invsqrt_pd(rsq11);
 296             rinv12           = gmx_mm256_invsqrt_pd(rsq12);
 297             rinv13           = gmx_mm256_invsqrt_pd(rsq13);
 298             rinv21           = gmx_mm256_invsqrt_pd(rsq21);
 299             rinv22           = gmx_mm256_invsqrt_pd(rsq22);
 300             rinv23           = gmx_mm256_invsqrt_pd(rsq23);
 301             rinv31           = gmx_mm256_invsqrt_pd(rsq31);
 302             rinv32           = gmx_mm256_invsqrt_pd(rsq32);
 303             rinv33           = gmx_mm256_invsqrt_pd(rsq33);
 304
 305             rinvsq11         = _mm256_mul_pd(rinv11,rinv11);
 306             rinvsq12         = _mm256_mul_pd(rinv12,rinv12);
 307             rinvsq13         = _mm256_mul_pd(rinv13,rinv13);
 308             rinvsq21         = _mm256_mul_pd(rinv21,rinv21);
 309             rinvsq22         = _mm256_mul_pd(rinv22,rinv22);
 310             rinvsq23         = _mm256_mul_pd(rinv23,rinv23);
 311             rinvsq31         = _mm256_mul_pd(rinv31,rinv31);
 312             rinvsq32         = _mm256_mul_pd(rinv32,rinv32);
 313             rinvsq33         = _mm256_mul_pd(rinv33,rinv33);
 314
 315             fjx0             = _mm256_setzero_pd();
 316             fjy0             = _mm256_setzero_pd();
 317             fjz0             = _mm256_setzero_pd();
 318             fjx1             = _mm256_setzero_pd();
 319             fjy1             = _mm256_setzero_pd();
 320             fjz1             = _mm256_setzero_pd();
 321             fjx2             = _mm256_setzero_pd();
 322             fjy2             = _mm256_setzero_pd();
 323             fjz2             = _mm256_setzero_pd();
 324             fjx3             = _mm256_setzero_pd();
 325             fjy3             = _mm256_setzero_pd();
 326             fjz3             = _mm256_setzero_pd();
 327
 328             /**************************
 329              * CALCULATE INTERACTIONS *
 330              **************************/
 331
 332             if (gmx_mm256_any_lt(rsq00,rcutoff2))
 333             {
 334
 335             r00              = _mm256_mul_pd(rsq00,rinv00);
 336
 337             /* Calculate table index by multiplying r with table scale and truncate to integer */
 338             rt               = _mm256_mul_pd(r00,vftabscale);
 339             vfitab           = _mm256_cvttpd_epi32(rt);
 340             vfeps            = _mm256_sub_pd(rt,_mm256_round_pd(rt, _MM_FROUND_FLOOR));
 341             vfitab           = _mm_slli_epi32(vfitab,3);
 342
 343             /* CUBIC SPLINE TABLE DISPERSION */
 344             Y                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
 345             F                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
 346             G                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,2) );
 347             H                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,3) );
 348             GMX_MM256_FULLTRANSPOSE4_PD(Y,F,G,H);
 349             Heps             = _mm256_mul_pd(vfeps,H);
 350             Fp               = _mm256_add_pd(F,_mm256_mul_pd(vfeps,_mm256_add_pd(G,Heps)));
 351             VV               = _mm256_add_pd(Y,_mm256_mul_pd(vfeps,Fp));
 352             vvdw6            = _mm256_mul_pd(c6_00,VV);
 353             FF               = _mm256_add_pd(Fp,_mm256_mul_pd(vfeps,_mm256_add_pd(G,_mm256_add_pd(Heps,Heps))));
 354             fvdw6            = _mm256_mul_pd(c6_00,FF);
 355
 356             /* CUBIC SPLINE TABLE REPULSION */
 357             vfitab           = _mm_add_epi32(vfitab,ifour);
 358             Y                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
 359             F                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
 360             G                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,2) );
 361             H                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,3) );
 362             GMX_MM256_FULLTRANSPOSE4_PD(Y,F,G,H);
 363             Heps             = _mm256_mul_pd(vfeps,H);
 364             Fp               = _mm256_add_pd(F,_mm256_mul_pd(vfeps,_mm256_add_pd(G,Heps)));
 365             VV               = _mm256_add_pd(Y,_mm256_mul_pd(vfeps,Fp));
 366             vvdw12           = _mm256_mul_pd(c12_00,VV);
 367             FF               = _mm256_add_pd(Fp,_mm256_mul_pd(vfeps,_mm256_add_pd(G,_mm256_add_pd(Heps,Heps))));
 368             fvdw12           = _mm256_mul_pd(c12_00,FF);
 369             vvdw             = _mm256_add_pd(vvdw12,vvdw6);
 370             fvdw             = _mm256_xor_pd(signbit,_mm256_mul_pd(_mm256_add_pd(fvdw6,fvdw12),_mm256_mul_pd(vftabscale,rinv00)));
 371
 372             cutoff_mask      = _mm256_cmp_pd(rsq00,rcutoff2,_CMP_LT_OQ);
 373
 374             /* Update potential sum for this i atom from the interaction with this j atom. */
 375             vvdw             = _mm256_and_pd(vvdw,cutoff_mask);
 376             vvdwsum          = _mm256_add_pd(vvdwsum,vvdw);
 377
 378             fscal            = fvdw;
 379
 380             fscal            = _mm256_and_pd(fscal,cutoff_mask);
 381
 382             /* Calculate temporary vectorial force */
 383             tx               = _mm256_mul_pd(fscal,dx00);
 384             ty               = _mm256_mul_pd(fscal,dy00);
 385             tz               = _mm256_mul_pd(fscal,dz00);
 386
 387             /* Update vectorial force */
 388             fix0             = _mm256_add_pd(fix0,tx);
 389             fiy0             = _mm256_add_pd(fiy0,ty);
 390             fiz0             = _mm256_add_pd(fiz0,tz);
 391
 392             fjx0             = _mm256_add_pd(fjx0,tx);
 393             fjy0             = _mm256_add_pd(fjy0,ty);
 394             fjz0             = _mm256_add_pd(fjz0,tz);
 395
 396             }
 397
 398             /**************************
 399              * CALCULATE INTERACTIONS *
 400              **************************/
 401
 402             if (gmx_mm256_any_lt(rsq11,rcutoff2))
 403             {
 404
 405             /* REACTION-FIELD ELECTROSTATICS */
 406             velec            = _mm256_mul_pd(qq11,_mm256_sub_pd(_mm256_add_pd(rinv11,_mm256_mul_pd(krf,rsq11)),crf));
 407             felec            = _mm256_mul_pd(qq11,_mm256_sub_pd(_mm256_mul_pd(rinv11,rinvsq11),krf2));
 408
 409             cutoff_mask      = _mm256_cmp_pd(rsq11,rcutoff2,_CMP_LT_OQ);
 410
 411             /* Update potential sum for this i atom from the interaction with this j atom. */
 412             velec            = _mm256_and_pd(velec,cutoff_mask);
 413             velecsum         = _mm256_add_pd(velecsum,velec);
 414
 415             fscal            = felec;
 416
 417             fscal            = _mm256_and_pd(fscal,cutoff_mask);
 418
 419             /* Calculate temporary vectorial force */
 420             tx               = _mm256_mul_pd(fscal,dx11);
 421             ty               = _mm256_mul_pd(fscal,dy11);
 422             tz               = _mm256_mul_pd(fscal,dz11);
 423
 424             /* Update vectorial force */
 425             fix1             = _mm256_add_pd(fix1,tx);
 426             fiy1             = _mm256_add_pd(fiy1,ty);
 427             fiz1             = _mm256_add_pd(fiz1,tz);
 428
 429             fjx1             = _mm256_add_pd(fjx1,tx);
 430             fjy1             = _mm256_add_pd(fjy1,ty);
 431             fjz1             = _mm256_add_pd(fjz1,tz);
 432
 433             }
 434
 435             /**************************
 436              * CALCULATE INTERACTIONS *
 437              **************************/
 438
 439             if (gmx_mm256_any_lt(rsq12,rcutoff2))
 440             {
 441
 442             /* REACTION-FIELD ELECTROSTATICS */
 443             velec            = _mm256_mul_pd(qq12,_mm256_sub_pd(_mm256_add_pd(rinv12,_mm256_mul_pd(krf,rsq12)),crf));
 444             felec            = _mm256_mul_pd(qq12,_mm256_sub_pd(_mm256_mul_pd(rinv12,rinvsq12),krf2));
 445
 446             cutoff_mask      = _mm256_cmp_pd(rsq12,rcutoff2,_CMP_LT_OQ);
 447
 448             /* Update potential sum for this i atom from the interaction with this j atom. */
 449             velec            = _mm256_and_pd(velec,cutoff_mask);
 450             velecsum         = _mm256_add_pd(velecsum,velec);
 451
 452             fscal            = felec;
 453
 454             fscal            = _mm256_and_pd(fscal,cutoff_mask);
 455
 456             /* Calculate temporary vectorial force */
 457             tx               = _mm256_mul_pd(fscal,dx12);
 458             ty               = _mm256_mul_pd(fscal,dy12);
 459             tz               = _mm256_mul_pd(fscal,dz12);
 460
 461             /* Update vectorial force */
 462             fix1             = _mm256_add_pd(fix1,tx);
 463             fiy1             = _mm256_add_pd(fiy1,ty);
 464             fiz1             = _mm256_add_pd(fiz1,tz);
 465
 466             fjx2             = _mm256_add_pd(fjx2,tx);
 467             fjy2             = _mm256_add_pd(fjy2,ty);
 468             fjz2             = _mm256_add_pd(fjz2,tz);
 469
 470             }
 471
 472             /**************************
 473              * CALCULATE INTERACTIONS *
 474              **************************/
 475
 476             if (gmx_mm256_any_lt(rsq13,rcutoff2))
 477             {
 478
 479             /* REACTION-FIELD ELECTROSTATICS */
 480             velec            = _mm256_mul_pd(qq13,_mm256_sub_pd(_mm256_add_pd(rinv13,_mm256_mul_pd(krf,rsq13)),crf));
 481             felec            = _mm256_mul_pd(qq13,_mm256_sub_pd(_mm256_mul_pd(rinv13,rinvsq13),krf2));
 482
 483             cutoff_mask      = _mm256_cmp_pd(rsq13,rcutoff2,_CMP_LT_OQ);
 484
 485             /* Update potential sum for this i atom from the interaction with this j atom. */
 486             velec            = _mm256_and_pd(velec,cutoff_mask);
 487             velecsum         = _mm256_add_pd(velecsum,velec);
 488
 489             fscal            = felec;
 490
 491             fscal            = _mm256_and_pd(fscal,cutoff_mask);
 492
 493             /* Calculate temporary vectorial force */
 494             tx               = _mm256_mul_pd(fscal,dx13);
 495             ty               = _mm256_mul_pd(fscal,dy13);
 496             tz               = _mm256_mul_pd(fscal,dz13);
 497
 498             /* Update vectorial force */
 499             fix1             = _mm256_add_pd(fix1,tx);
 500             fiy1             = _mm256_add_pd(fiy1,ty);
 501             fiz1             = _mm256_add_pd(fiz1,tz);
 502
 503             fjx3             = _mm256_add_pd(fjx3,tx);
 504             fjy3             = _mm256_add_pd(fjy3,ty);
 505             fjz3             = _mm256_add_pd(fjz3,tz);
 506
 507             }
 508
 509             /**************************
 510              * CALCULATE INTERACTIONS *
 511              **************************/
 512
 513             if (gmx_mm256_any_lt(rsq21,rcutoff2))
 514             {
 515
 516             /* REACTION-FIELD ELECTROSTATICS */
 517             velec            = _mm256_mul_pd(qq21,_mm256_sub_pd(_mm256_add_pd(rinv21,_mm256_mul_pd(krf,rsq21)),crf));
 518             felec            = _mm256_mul_pd(qq21,_mm256_sub_pd(_mm256_mul_pd(rinv21,rinvsq21),krf2));
 519
 520             cutoff_mask      = _mm256_cmp_pd(rsq21,rcutoff2,_CMP_LT_OQ);
 521
 522             /* Update potential sum for this i atom from the interaction with this j atom. */
 523             velec            = _mm256_and_pd(velec,cutoff_mask);
 524             velecsum         = _mm256_add_pd(velecsum,velec);
 525
 526             fscal            = felec;
 527
 528             fscal            = _mm256_and_pd(fscal,cutoff_mask);
 529
 530             /* Calculate temporary vectorial force */
 531             tx               = _mm256_mul_pd(fscal,dx21);
 532             ty               = _mm256_mul_pd(fscal,dy21);
 533             tz               = _mm256_mul_pd(fscal,dz21);
 534
 535             /* Update vectorial force */
 536             fix2             = _mm256_add_pd(fix2,tx);
 537             fiy2             = _mm256_add_pd(fiy2,ty);
 538             fiz2             = _mm256_add_pd(fiz2,tz);
 539
 540             fjx1             = _mm256_add_pd(fjx1,tx);
 541             fjy1             = _mm256_add_pd(fjy1,ty);
 542             fjz1             = _mm256_add_pd(fjz1,tz);
 543
 544             }
 545
 546             /**************************
 547              * CALCULATE INTERACTIONS *
 548              **************************/
 549
 550             if (gmx_mm256_any_lt(rsq22,rcutoff2))
 551             {
 552
 553             /* REACTION-FIELD ELECTROSTATICS */
 554             velec            = _mm256_mul_pd(qq22,_mm256_sub_pd(_mm256_add_pd(rinv22,_mm256_mul_pd(krf,rsq22)),crf));
 555             felec            = _mm256_mul_pd(qq22,_mm256_sub_pd(_mm256_mul_pd(rinv22,rinvsq22),krf2));
 556
 557             cutoff_mask      = _mm256_cmp_pd(rsq22,rcutoff2,_CMP_LT_OQ);
 558
 559             /* Update potential sum for this i atom from the interaction with this j atom. */
 560             velec            = _mm256_and_pd(velec,cutoff_mask);
 561             velecsum         = _mm256_add_pd(velecsum,velec);
 562
 563             fscal            = felec;
 564
 565             fscal            = _mm256_and_pd(fscal,cutoff_mask);
 566
 567             /* Calculate temporary vectorial force */
 568             tx               = _mm256_mul_pd(fscal,dx22);
 569             ty               = _mm256_mul_pd(fscal,dy22);
 570             tz               = _mm256_mul_pd(fscal,dz22);
 571
 572             /* Update vectorial force */
 573             fix2             = _mm256_add_pd(fix2,tx);
 574             fiy2             = _mm256_add_pd(fiy2,ty);
 575             fiz2             = _mm256_add_pd(fiz2,tz);
 576
 577             fjx2             = _mm256_add_pd(fjx2,tx);
 578             fjy2             = _mm256_add_pd(fjy2,ty);
 579             fjz2             = _mm256_add_pd(fjz2,tz);
 580
 581             }
 582
 583             /**************************
 584              * CALCULATE INTERACTIONS *
 585              **************************/
 586
 587             if (gmx_mm256_any_lt(rsq23,rcutoff2))
 588             {
 589
 590             /* REACTION-FIELD ELECTROSTATICS */
 591             velec            = _mm256_mul_pd(qq23,_mm256_sub_pd(_mm256_add_pd(rinv23,_mm256_mul_pd(krf,rsq23)),crf));
 592             felec            = _mm256_mul_pd(qq23,_mm256_sub_pd(_mm256_mul_pd(rinv23,rinvsq23),krf2));
 593
 594             cutoff_mask      = _mm256_cmp_pd(rsq23,rcutoff2,_CMP_LT_OQ);
 595
 596             /* Update potential sum for this i atom from the interaction with this j atom. */
 597             velec            = _mm256_and_pd(velec,cutoff_mask);
 598             velecsum         = _mm256_add_pd(velecsum,velec);
 599
 600             fscal            = felec;
 601
 602             fscal            = _mm256_and_pd(fscal,cutoff_mask);
 603
 604             /* Calculate temporary vectorial force */
 605             tx               = _mm256_mul_pd(fscal,dx23);
 606             ty               = _mm256_mul_pd(fscal,dy23);
 607             tz               = _mm256_mul_pd(fscal,dz23);
 608
 609             /* Update vectorial force */
 610             fix2             = _mm256_add_pd(fix2,tx);
 611             fiy2             = _mm256_add_pd(fiy2,ty);
 612             fiz2             = _mm256_add_pd(fiz2,tz);
 613
 614             fjx3             = _mm256_add_pd(fjx3,tx);
 615             fjy3             = _mm256_add_pd(fjy3,ty);
 616             fjz3             = _mm256_add_pd(fjz3,tz);
 617
 618             }
 619
 620             /**************************
 621              * CALCULATE INTERACTIONS *
 622              **************************/
 623
 624             if (gmx_mm256_any_lt(rsq31,rcutoff2))
 625             {
 626
 627             /* REACTION-FIELD ELECTROSTATICS */
 628             velec            = _mm256_mul_pd(qq31,_mm256_sub_pd(_mm256_add_pd(rinv31,_mm256_mul_pd(krf,rsq31)),crf));
 629             felec            = _mm256_mul_pd(qq31,_mm256_sub_pd(_mm256_mul_pd(rinv31,rinvsq31),krf2));
 630
 631             cutoff_mask      = _mm256_cmp_pd(rsq31,rcutoff2,_CMP_LT_OQ);
 632
 633             /* Update potential sum for this i atom from the interaction with this j atom. */
 634             velec            = _mm256_and_pd(velec,cutoff_mask);
 635             velecsum         = _mm256_add_pd(velecsum,velec);
 636
 637             fscal            = felec;
 638
 639             fscal            = _mm256_and_pd(fscal,cutoff_mask);
 640
 641             /* Calculate temporary vectorial force */
 642             tx               = _mm256_mul_pd(fscal,dx31);
 643             ty               = _mm256_mul_pd(fscal,dy31);
 644             tz               = _mm256_mul_pd(fscal,dz31);
 645
 646             /* Update vectorial force */
 647             fix3             = _mm256_add_pd(fix3,tx);
 648             fiy3             = _mm256_add_pd(fiy3,ty);
 649             fiz3             = _mm256_add_pd(fiz3,tz);
 650
 651             fjx1             = _mm256_add_pd(fjx1,tx);
 652             fjy1             = _mm256_add_pd(fjy1,ty);
 653             fjz1             = _mm256_add_pd(fjz1,tz);
 654
 655             }
 656
 657             /**************************
 658              * CALCULATE INTERACTIONS *
 659              **************************/
 660
 661             if (gmx_mm256_any_lt(rsq32,rcutoff2))
 662             {
 663
 664             /* REACTION-FIELD ELECTROSTATICS */
 665             velec            = _mm256_mul_pd(qq32,_mm256_sub_pd(_mm256_add_pd(rinv32,_mm256_mul_pd(krf,rsq32)),crf));
 666             felec            = _mm256_mul_pd(qq32,_mm256_sub_pd(_mm256_mul_pd(rinv32,rinvsq32),krf2));
 667
 668             cutoff_mask      = _mm256_cmp_pd(rsq32,rcutoff2,_CMP_LT_OQ);
 669
 670             /* Update potential sum for this i atom from the interaction with this j atom. */
 671             velec            = _mm256_and_pd(velec,cutoff_mask);
 672             velecsum         = _mm256_add_pd(velecsum,velec);
 673
 674             fscal            = felec;
 675
 676             fscal            = _mm256_and_pd(fscal,cutoff_mask);
 677
 678             /* Calculate temporary vectorial force */
 679             tx               = _mm256_mul_pd(fscal,dx32);
 680             ty               = _mm256_mul_pd(fscal,dy32);
 681             tz               = _mm256_mul_pd(fscal,dz32);
 682
 683             /* Update vectorial force */
 684             fix3             = _mm256_add_pd(fix3,tx);
 685             fiy3             = _mm256_add_pd(fiy3,ty);
 686             fiz3             = _mm256_add_pd(fiz3,tz);
 687
 688             fjx2             = _mm256_add_pd(fjx2,tx);
 689             fjy2             = _mm256_add_pd(fjy2,ty);
 690             fjz2             = _mm256_add_pd(fjz2,tz);
 691
 692             }
 693
 694             /**************************
 695              * CALCULATE INTERACTIONS *
 696              **************************/
 697
 698             if (gmx_mm256_any_lt(rsq33,rcutoff2))
 699             {
 700
 701             /* REACTION-FIELD ELECTROSTATICS */
 702             velec            = _mm256_mul_pd(qq33,_mm256_sub_pd(_mm256_add_pd(rinv33,_mm256_mul_pd(krf,rsq33)),crf));
 703             felec            = _mm256_mul_pd(qq33,_mm256_sub_pd(_mm256_mul_pd(rinv33,rinvsq33),krf2));
 704
 705             cutoff_mask      = _mm256_cmp_pd(rsq33,rcutoff2,_CMP_LT_OQ);
 706
 707             /* Update potential sum for this i atom from the interaction with this j atom. */
 708             velec            = _mm256_and_pd(velec,cutoff_mask);
 709             velecsum         = _mm256_add_pd(velecsum,velec);
 710
 711             fscal            = felec;
 712
 713             fscal            = _mm256_and_pd(fscal,cutoff_mask);
 714
 715             /* Calculate temporary vectorial force */
 716             tx               = _mm256_mul_pd(fscal,dx33);
 717             ty               = _mm256_mul_pd(fscal,dy33);
 718             tz               = _mm256_mul_pd(fscal,dz33);
 719
 720             /* Update vectorial force */
 721             fix3             = _mm256_add_pd(fix3,tx);
 722             fiy3             = _mm256_add_pd(fiy3,ty);
 723             fiz3             = _mm256_add_pd(fiz3,tz);
 724
 725             fjx3             = _mm256_add_pd(fjx3,tx);
 726             fjy3             = _mm256_add_pd(fjy3,ty);
 727             fjz3             = _mm256_add_pd(fjz3,tz);
 728
 729             }
 730
 731             fjptrA             = f+j_coord_offsetA;
 732             fjptrB             = f+j_coord_offsetB;
 733             fjptrC             = f+j_coord_offsetC;
 734             fjptrD             = f+j_coord_offsetD;
 735
 736             gmx_mm256_decrement_4rvec_4ptr_swizzle_pd(fjptrA,fjptrB,fjptrC,fjptrD,
 737                                                       fjx0,fjy0,fjz0,fjx1,fjy1,fjz1,
 738                                                       fjx2,fjy2,fjz2,fjx3,fjy3,fjz3);
 739
 740             /* Inner loop uses 387 flops */
 741         }
 742
 743         if(jidx<j_index_end)
 744         {
 745
 746             /* Get j neighbor index, and coordinate index */
 747             jnrlistA         = jjnr[jidx];
 748             jnrlistB         = jjnr[jidx+1];
 749             jnrlistC         = jjnr[jidx+2];
 750             jnrlistD         = jjnr[jidx+3];
 751             /* Sign of each element will be negative for non-real atoms.
 752              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 753              * so use it as val = _mm_andnot_pd(mask,val) to clear dummy entries.
 754              */
 755             tmpmask0 = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
 756
 757             tmpmask1 = _mm_permute_ps(tmpmask0,_GMX_MM_PERMUTE(3,3,2,2));
 758             tmpmask0 = _mm_permute_ps(tmpmask0,_GMX_MM_PERMUTE(1,1,0,0));
 759             dummy_mask = _mm256_castps_pd(gmx_mm256_set_m128(tmpmask1,tmpmask0));
 760
 761             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
 762             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
 763             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
 764             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
 765             j_coord_offsetA  = DIM*jnrA;
 766             j_coord_offsetB  = DIM*jnrB;
 767             j_coord_offsetC  = DIM*jnrC;
 768             j_coord_offsetD  = DIM*jnrD;
 769
 770             /* load j atom coordinates */
 771             gmx_mm256_load_4rvec_4ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 772                                                  x+j_coord_offsetC,x+j_coord_offsetD,
 773                                                  &jx0,&jy0,&jz0,&jx1,&jy1,&jz1,&jx2,
 774                                                  &jy2,&jz2,&jx3,&jy3,&jz3);
 775
 776             /* Calculate displacement vector */
 777             dx00             = _mm256_sub_pd(ix0,jx0);
 778             dy00             = _mm256_sub_pd(iy0,jy0);
 779             dz00             = _mm256_sub_pd(iz0,jz0);
 780             dx11             = _mm256_sub_pd(ix1,jx1);
 781             dy11             = _mm256_sub_pd(iy1,jy1);
 782             dz11             = _mm256_sub_pd(iz1,jz1);
 783             dx12             = _mm256_sub_pd(ix1,jx2);
 784             dy12             = _mm256_sub_pd(iy1,jy2);
 785             dz12             = _mm256_sub_pd(iz1,jz2);
 786             dx13             = _mm256_sub_pd(ix1,jx3);
 787             dy13             = _mm256_sub_pd(iy1,jy3);
 788             dz13             = _mm256_sub_pd(iz1,jz3);
 789             dx21             = _mm256_sub_pd(ix2,jx1);
 790             dy21             = _mm256_sub_pd(iy2,jy1);
 791             dz21             = _mm256_sub_pd(iz2,jz1);
 792             dx22             = _mm256_sub_pd(ix2,jx2);
 793             dy22             = _mm256_sub_pd(iy2,jy2);
 794             dz22             = _mm256_sub_pd(iz2,jz2);
 795             dx23             = _mm256_sub_pd(ix2,jx3);
 796             dy23             = _mm256_sub_pd(iy2,jy3);
 797             dz23             = _mm256_sub_pd(iz2,jz3);
 798             dx31             = _mm256_sub_pd(ix3,jx1);
 799             dy31             = _mm256_sub_pd(iy3,jy1);
 800             dz31             = _mm256_sub_pd(iz3,jz1);
 801             dx32             = _mm256_sub_pd(ix3,jx2);
 802             dy32             = _mm256_sub_pd(iy3,jy2);
 803             dz32             = _mm256_sub_pd(iz3,jz2);
 804             dx33             = _mm256_sub_pd(ix3,jx3);
 805             dy33             = _mm256_sub_pd(iy3,jy3);
 806             dz33             = _mm256_sub_pd(iz3,jz3);
 807
 808             /* Calculate squared distance and things based on it */
 809             rsq00            = gmx_mm256_calc_rsq_pd(dx00,dy00,dz00);
 810             rsq11            = gmx_mm256_calc_rsq_pd(dx11,dy11,dz11);
 811             rsq12            = gmx_mm256_calc_rsq_pd(dx12,dy12,dz12);
 812             rsq13            = gmx_mm256_calc_rsq_pd(dx13,dy13,dz13);
 813             rsq21            = gmx_mm256_calc_rsq_pd(dx21,dy21,dz21);
 814             rsq22            = gmx_mm256_calc_rsq_pd(dx22,dy22,dz22);
 815             rsq23            = gmx_mm256_calc_rsq_pd(dx23,dy23,dz23);
 816             rsq31            = gmx_mm256_calc_rsq_pd(dx31,dy31,dz31);
 817             rsq32            = gmx_mm256_calc_rsq_pd(dx32,dy32,dz32);
 818             rsq33            = gmx_mm256_calc_rsq_pd(dx33,dy33,dz33);
 819
 820             rinv00           = gmx_mm256_invsqrt_pd(rsq00);
 821             rinv11           = gmx_mm256_invsqrt_pd(rsq11);
 822             rinv12           = gmx_mm256_invsqrt_pd(rsq12);
 823             rinv13           = gmx_mm256_invsqrt_pd(rsq13);
 824             rinv21           = gmx_mm256_invsqrt_pd(rsq21);
 825             rinv22           = gmx_mm256_invsqrt_pd(rsq22);
 826             rinv23           = gmx_mm256_invsqrt_pd(rsq23);
 827             rinv31           = gmx_mm256_invsqrt_pd(rsq31);
 828             rinv32           = gmx_mm256_invsqrt_pd(rsq32);
 829             rinv33           = gmx_mm256_invsqrt_pd(rsq33);
 830
 831             rinvsq11         = _mm256_mul_pd(rinv11,rinv11);
 832             rinvsq12         = _mm256_mul_pd(rinv12,rinv12);
 833             rinvsq13         = _mm256_mul_pd(rinv13,rinv13);
 834             rinvsq21         = _mm256_mul_pd(rinv21,rinv21);
 835             rinvsq22         = _mm256_mul_pd(rinv22,rinv22);
 836             rinvsq23         = _mm256_mul_pd(rinv23,rinv23);
 837             rinvsq31         = _mm256_mul_pd(rinv31,rinv31);
 838             rinvsq32         = _mm256_mul_pd(rinv32,rinv32);
 839             rinvsq33         = _mm256_mul_pd(rinv33,rinv33);
 840
 841             fjx0             = _mm256_setzero_pd();
 842             fjy0             = _mm256_setzero_pd();
 843             fjz0             = _mm256_setzero_pd();
 844             fjx1             = _mm256_setzero_pd();
 845             fjy1             = _mm256_setzero_pd();
 846             fjz1             = _mm256_setzero_pd();
 847             fjx2             = _mm256_setzero_pd();
 848             fjy2             = _mm256_setzero_pd();
 849             fjz2             = _mm256_setzero_pd();
 850             fjx3             = _mm256_setzero_pd();
 851             fjy3             = _mm256_setzero_pd();
 852             fjz3             = _mm256_setzero_pd();
 853
 854             /**************************
 855              * CALCULATE INTERACTIONS *
 856              **************************/
 857
 858             if (gmx_mm256_any_lt(rsq00,rcutoff2))
 859             {
 860
 861             r00              = _mm256_mul_pd(rsq00,rinv00);
 862             r00              = _mm256_andnot_pd(dummy_mask,r00);
 863
 864             /* Calculate table index by multiplying r with table scale and truncate to integer */
 865             rt               = _mm256_mul_pd(r00,vftabscale);
 866             vfitab           = _mm256_cvttpd_epi32(rt);
 867             vfeps            = _mm256_sub_pd(rt,_mm256_round_pd(rt, _MM_FROUND_FLOOR));
 868             vfitab           = _mm_slli_epi32(vfitab,3);
 869
 870             /* CUBIC SPLINE TABLE DISPERSION */
 871             Y                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
 872             F                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
 873             G                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,2) );
 874             H                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,3) );
 875             GMX_MM256_FULLTRANSPOSE4_PD(Y,F,G,H);
 876             Heps             = _mm256_mul_pd(vfeps,H);
 877             Fp               = _mm256_add_pd(F,_mm256_mul_pd(vfeps,_mm256_add_pd(G,Heps)));
 878             VV               = _mm256_add_pd(Y,_mm256_mul_pd(vfeps,Fp));
 879             vvdw6            = _mm256_mul_pd(c6_00,VV);
 880             FF               = _mm256_add_pd(Fp,_mm256_mul_pd(vfeps,_mm256_add_pd(G,_mm256_add_pd(Heps,Heps))));
 881             fvdw6            = _mm256_mul_pd(c6_00,FF);
 882
 883             /* CUBIC SPLINE TABLE REPULSION */
 884             vfitab           = _mm_add_epi32(vfitab,ifour);
 885             Y                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
 886             F                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
 887             G                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,2) );
 888             H                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,3) );
 889             GMX_MM256_FULLTRANSPOSE4_PD(Y,F,G,H);
 890             Heps             = _mm256_mul_pd(vfeps,H);
 891             Fp               = _mm256_add_pd(F,_mm256_mul_pd(vfeps,_mm256_add_pd(G,Heps)));
 892             VV               = _mm256_add_pd(Y,_mm256_mul_pd(vfeps,Fp));
 893             vvdw12           = _mm256_mul_pd(c12_00,VV);
 894             FF               = _mm256_add_pd(Fp,_mm256_mul_pd(vfeps,_mm256_add_pd(G,_mm256_add_pd(Heps,Heps))));
 895             fvdw12           = _mm256_mul_pd(c12_00,FF);
 896             vvdw             = _mm256_add_pd(vvdw12,vvdw6);
 897             fvdw             = _mm256_xor_pd(signbit,_mm256_mul_pd(_mm256_add_pd(fvdw6,fvdw12),_mm256_mul_pd(vftabscale,rinv00)));
 898
 899             cutoff_mask      = _mm256_cmp_pd(rsq00,rcutoff2,_CMP_LT_OQ);
 900
 901             /* Update potential sum for this i atom from the interaction with this j atom. */
 902             vvdw             = _mm256_and_pd(vvdw,cutoff_mask);
 903             vvdw             = _mm256_andnot_pd(dummy_mask,vvdw);
 904             vvdwsum          = _mm256_add_pd(vvdwsum,vvdw);
 905
 906             fscal            = fvdw;
 907
 908             fscal            = _mm256_and_pd(fscal,cutoff_mask);
 909
 910             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
 911
 912             /* Calculate temporary vectorial force */
 913             tx               = _mm256_mul_pd(fscal,dx00);
 914             ty               = _mm256_mul_pd(fscal,dy00);
 915             tz               = _mm256_mul_pd(fscal,dz00);
 916
 917             /* Update vectorial force */
 918             fix0             = _mm256_add_pd(fix0,tx);
 919             fiy0             = _mm256_add_pd(fiy0,ty);
 920             fiz0             = _mm256_add_pd(fiz0,tz);
 921
 922             fjx0             = _mm256_add_pd(fjx0,tx);
 923             fjy0             = _mm256_add_pd(fjy0,ty);
 924             fjz0             = _mm256_add_pd(fjz0,tz);
 925
 926             }
 927
 928             /**************************
 929              * CALCULATE INTERACTIONS *
 930              **************************/
 931
 932             if (gmx_mm256_any_lt(rsq11,rcutoff2))
 933             {
 934
 935             /* REACTION-FIELD ELECTROSTATICS */
 936             velec            = _mm256_mul_pd(qq11,_mm256_sub_pd(_mm256_add_pd(rinv11,_mm256_mul_pd(krf,rsq11)),crf));
 937             felec            = _mm256_mul_pd(qq11,_mm256_sub_pd(_mm256_mul_pd(rinv11,rinvsq11),krf2));
 938
 939             cutoff_mask      = _mm256_cmp_pd(rsq11,rcutoff2,_CMP_LT_OQ);
 940
 941             /* Update potential sum for this i atom from the interaction with this j atom. */
 942             velec            = _mm256_and_pd(velec,cutoff_mask);
 943             velec            = _mm256_andnot_pd(dummy_mask,velec);
 944             velecsum         = _mm256_add_pd(velecsum,velec);
 945
 946             fscal            = felec;
 947
 948             fscal            = _mm256_and_pd(fscal,cutoff_mask);
 949
 950             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
 951
 952             /* Calculate temporary vectorial force */
 953             tx               = _mm256_mul_pd(fscal,dx11);
 954             ty               = _mm256_mul_pd(fscal,dy11);
 955             tz               = _mm256_mul_pd(fscal,dz11);
 956
 957             /* Update vectorial force */
 958             fix1             = _mm256_add_pd(fix1,tx);
 959             fiy1             = _mm256_add_pd(fiy1,ty);
 960             fiz1             = _mm256_add_pd(fiz1,tz);
 961
 962             fjx1             = _mm256_add_pd(fjx1,tx);
 963             fjy1             = _mm256_add_pd(fjy1,ty);
 964             fjz1             = _mm256_add_pd(fjz1,tz);
 965
 966             }
 967
 968             /**************************
 969              * CALCULATE INTERACTIONS *
 970              **************************/
 971
 972             if (gmx_mm256_any_lt(rsq12,rcutoff2))
 973             {
 974
 975             /* REACTION-FIELD ELECTROSTATICS */
 976             velec            = _mm256_mul_pd(qq12,_mm256_sub_pd(_mm256_add_pd(rinv12,_mm256_mul_pd(krf,rsq12)),crf));
 977             felec            = _mm256_mul_pd(qq12,_mm256_sub_pd(_mm256_mul_pd(rinv12,rinvsq12),krf2));
 978
 979             cutoff_mask      = _mm256_cmp_pd(rsq12,rcutoff2,_CMP_LT_OQ);
 980
 981             /* Update potential sum for this i atom from the interaction with this j atom. */
 982             velec            = _mm256_and_pd(velec,cutoff_mask);
 983             velec            = _mm256_andnot_pd(dummy_mask,velec);
 984             velecsum         = _mm256_add_pd(velecsum,velec);
 985
 986             fscal            = felec;
 987
 988             fscal            = _mm256_and_pd(fscal,cutoff_mask);
 989
 990             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
 991
 992             /* Calculate temporary vectorial force */
 993             tx               = _mm256_mul_pd(fscal,dx12);
 994             ty               = _mm256_mul_pd(fscal,dy12);
 995             tz               = _mm256_mul_pd(fscal,dz12);
 996
 997             /* Update vectorial force */
 998             fix1             = _mm256_add_pd(fix1,tx);
 999             fiy1             = _mm256_add_pd(fiy1,ty);
1000             fiz1             = _mm256_add_pd(fiz1,tz);
1001
1002             fjx2             = _mm256_add_pd(fjx2,tx);
1003             fjy2             = _mm256_add_pd(fjy2,ty);
1004             fjz2             = _mm256_add_pd(fjz2,tz);
1005
1006             }
1007
1008             /**************************
1009              * CALCULATE INTERACTIONS *
1010              **************************/
1011
1012             if (gmx_mm256_any_lt(rsq13,rcutoff2))
1013             {
1014
1015             /* REACTION-FIELD ELECTROSTATICS */
1016             velec            = _mm256_mul_pd(qq13,_mm256_sub_pd(_mm256_add_pd(rinv13,_mm256_mul_pd(krf,rsq13)),crf));
1017             felec            = _mm256_mul_pd(qq13,_mm256_sub_pd(_mm256_mul_pd(rinv13,rinvsq13),krf2));
1018
1019             cutoff_mask      = _mm256_cmp_pd(rsq13,rcutoff2,_CMP_LT_OQ);
1020
1021             /* Update potential sum for this i atom from the interaction with this j atom. */
1022             velec            = _mm256_and_pd(velec,cutoff_mask);
1023             velec            = _mm256_andnot_pd(dummy_mask,velec);
1024             velecsum         = _mm256_add_pd(velecsum,velec);
1025
1026             fscal            = felec;
1027
1028             fscal            = _mm256_and_pd(fscal,cutoff_mask);
1029
1030             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
1031
1032             /* Calculate temporary vectorial force */
1033             tx               = _mm256_mul_pd(fscal,dx13);
1034             ty               = _mm256_mul_pd(fscal,dy13);
1035             tz               = _mm256_mul_pd(fscal,dz13);
1036
1037             /* Update vectorial force */
1038             fix1             = _mm256_add_pd(fix1,tx);
1039             fiy1             = _mm256_add_pd(fiy1,ty);
1040             fiz1             = _mm256_add_pd(fiz1,tz);
1041
1042             fjx3             = _mm256_add_pd(fjx3,tx);
1043             fjy3             = _mm256_add_pd(fjy3,ty);
1044             fjz3             = _mm256_add_pd(fjz3,tz);
1045
1046             }
1047
1048             /**************************
1049              * CALCULATE INTERACTIONS *
1050              **************************/
1051
1052             if (gmx_mm256_any_lt(rsq21,rcutoff2))
1053             {
1054
1055             /* REACTION-FIELD ELECTROSTATICS */
1056             velec            = _mm256_mul_pd(qq21,_mm256_sub_pd(_mm256_add_pd(rinv21,_mm256_mul_pd(krf,rsq21)),crf));
1057             felec            = _mm256_mul_pd(qq21,_mm256_sub_pd(_mm256_mul_pd(rinv21,rinvsq21),krf2));
1058
1059             cutoff_mask      = _mm256_cmp_pd(rsq21,rcutoff2,_CMP_LT_OQ);
1060
1061             /* Update potential sum for this i atom from the interaction with this j atom. */
1062             velec            = _mm256_and_pd(velec,cutoff_mask);
1063             velec            = _mm256_andnot_pd(dummy_mask,velec);
1064             velecsum         = _mm256_add_pd(velecsum,velec);
1065
1066             fscal            = felec;
1067
1068             fscal            = _mm256_and_pd(fscal,cutoff_mask);
1069
1070             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
1071
1072             /* Calculate temporary vectorial force */
1073             tx               = _mm256_mul_pd(fscal,dx21);
1074             ty               = _mm256_mul_pd(fscal,dy21);
1075             tz               = _mm256_mul_pd(fscal,dz21);
1076
1077             /* Update vectorial force */
1078             fix2             = _mm256_add_pd(fix2,tx);
1079             fiy2             = _mm256_add_pd(fiy2,ty);
1080             fiz2             = _mm256_add_pd(fiz2,tz);
1081
1082             fjx1             = _mm256_add_pd(fjx1,tx);
1083             fjy1             = _mm256_add_pd(fjy1,ty);
1084             fjz1             = _mm256_add_pd(fjz1,tz);
1085
1086             }
1087
1088             /**************************
1089              * CALCULATE INTERACTIONS *
1090              **************************/
1091
1092             if (gmx_mm256_any_lt(rsq22,rcutoff2))
1093             {
1094
1095             /* REACTION-FIELD ELECTROSTATICS */
1096             velec            = _mm256_mul_pd(qq22,_mm256_sub_pd(_mm256_add_pd(rinv22,_mm256_mul_pd(krf,rsq22)),crf));
1097             felec            = _mm256_mul_pd(qq22,_mm256_sub_pd(_mm256_mul_pd(rinv22,rinvsq22),krf2));
1098
1099             cutoff_mask      = _mm256_cmp_pd(rsq22,rcutoff2,_CMP_LT_OQ);
1100
1101             /* Update potential sum for this i atom from the interaction with this j atom. */
1102             velec            = _mm256_and_pd(velec,cutoff_mask);
1103             velec            = _mm256_andnot_pd(dummy_mask,velec);
1104             velecsum         = _mm256_add_pd(velecsum,velec);
1105
1106             fscal            = felec;
1107
1108             fscal            = _mm256_and_pd(fscal,cutoff_mask);
1109
1110             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
1111
1112             /* Calculate temporary vectorial force */
1113             tx               = _mm256_mul_pd(fscal,dx22);
1114             ty               = _mm256_mul_pd(fscal,dy22);
1115             tz               = _mm256_mul_pd(fscal,dz22);
1116
1117             /* Update vectorial force */
1118             fix2             = _mm256_add_pd(fix2,tx);
1119             fiy2             = _mm256_add_pd(fiy2,ty);
1120             fiz2             = _mm256_add_pd(fiz2,tz);
1121
1122             fjx2             = _mm256_add_pd(fjx2,tx);
1123             fjy2             = _mm256_add_pd(fjy2,ty);
1124             fjz2             = _mm256_add_pd(fjz2,tz);
1125
1126             }
1127
1128             /**************************
1129              * CALCULATE INTERACTIONS *
1130              **************************/
1131
1132             if (gmx_mm256_any_lt(rsq23,rcutoff2))
1133             {
1134
1135             /* REACTION-FIELD ELECTROSTATICS */
1136             velec            = _mm256_mul_pd(qq23,_mm256_sub_pd(_mm256_add_pd(rinv23,_mm256_mul_pd(krf,rsq23)),crf));
1137             felec            = _mm256_mul_pd(qq23,_mm256_sub_pd(_mm256_mul_pd(rinv23,rinvsq23),krf2));
1138
1139             cutoff_mask      = _mm256_cmp_pd(rsq23,rcutoff2,_CMP_LT_OQ);
1140
1141             /* Update potential sum for this i atom from the interaction with this j atom. */
1142             velec            = _mm256_and_pd(velec,cutoff_mask);
1143             velec            = _mm256_andnot_pd(dummy_mask,velec);
1144             velecsum         = _mm256_add_pd(velecsum,velec);
1145
1146             fscal            = felec;
1147
1148             fscal            = _mm256_and_pd(fscal,cutoff_mask);
1149
1150             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
1151
1152             /* Calculate temporary vectorial force */
1153             tx               = _mm256_mul_pd(fscal,dx23);
1154             ty               = _mm256_mul_pd(fscal,dy23);
1155             tz               = _mm256_mul_pd(fscal,dz23);
1156
1157             /* Update vectorial force */
1158             fix2             = _mm256_add_pd(fix2,tx);
1159             fiy2             = _mm256_add_pd(fiy2,ty);
1160             fiz2             = _mm256_add_pd(fiz2,tz);
1161
1162             fjx3             = _mm256_add_pd(fjx3,tx);
1163             fjy3             = _mm256_add_pd(fjy3,ty);
1164             fjz3             = _mm256_add_pd(fjz3,tz);
1165
1166             }
1167
1168             /**************************
1169              * CALCULATE INTERACTIONS *
1170              **************************/
1171
1172             if (gmx_mm256_any_lt(rsq31,rcutoff2))
1173             {
1174
1175             /* REACTION-FIELD ELECTROSTATICS */
1176             velec            = _mm256_mul_pd(qq31,_mm256_sub_pd(_mm256_add_pd(rinv31,_mm256_mul_pd(krf,rsq31)),crf));
1177             felec            = _mm256_mul_pd(qq31,_mm256_sub_pd(_mm256_mul_pd(rinv31,rinvsq31),krf2));
1178
1179             cutoff_mask      = _mm256_cmp_pd(rsq31,rcutoff2,_CMP_LT_OQ);
1180
1181             /* Update potential sum for this i atom from the interaction with this j atom. */
1182             velec            = _mm256_and_pd(velec,cutoff_mask);
1183             velec            = _mm256_andnot_pd(dummy_mask,velec);
1184             velecsum         = _mm256_add_pd(velecsum,velec);
1185
1186             fscal            = felec;
1187
1188             fscal            = _mm256_and_pd(fscal,cutoff_mask);
1189
1190             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
1191
1192             /* Calculate temporary vectorial force */
1193             tx               = _mm256_mul_pd(fscal,dx31);
1194             ty               = _mm256_mul_pd(fscal,dy31);
1195             tz               = _mm256_mul_pd(fscal,dz31);
1196
1197             /* Update vectorial force */
1198             fix3             = _mm256_add_pd(fix3,tx);
1199             fiy3             = _mm256_add_pd(fiy3,ty);
1200             fiz3             = _mm256_add_pd(fiz3,tz);
1201
1202             fjx1             = _mm256_add_pd(fjx1,tx);
1203             fjy1             = _mm256_add_pd(fjy1,ty);
1204             fjz1             = _mm256_add_pd(fjz1,tz);
1205
1206             }
1207
1208             /**************************
1209              * CALCULATE INTERACTIONS *
1210              **************************/
1211
1212             if (gmx_mm256_any_lt(rsq32,rcutoff2))
1213             {
1214
1215             /* REACTION-FIELD ELECTROSTATICS */
1216             velec            = _mm256_mul_pd(qq32,_mm256_sub_pd(_mm256_add_pd(rinv32,_mm256_mul_pd(krf,rsq32)),crf));
1217             felec            = _mm256_mul_pd(qq32,_mm256_sub_pd(_mm256_mul_pd(rinv32,rinvsq32),krf2));
1218
1219             cutoff_mask      = _mm256_cmp_pd(rsq32,rcutoff2,_CMP_LT_OQ);
1220
1221             /* Update potential sum for this i atom from the interaction with this j atom. */
1222             velec            = _mm256_and_pd(velec,cutoff_mask);
1223             velec            = _mm256_andnot_pd(dummy_mask,velec);
1224             velecsum         = _mm256_add_pd(velecsum,velec);
1225
1226             fscal            = felec;
1227
1228             fscal            = _mm256_and_pd(fscal,cutoff_mask);
1229
1230             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
1231
1232             /* Calculate temporary vectorial force */
1233             tx               = _mm256_mul_pd(fscal,dx32);
1234             ty               = _mm256_mul_pd(fscal,dy32);
1235             tz               = _mm256_mul_pd(fscal,dz32);
1236
1237             /* Update vectorial force */
1238             fix3             = _mm256_add_pd(fix3,tx);
1239             fiy3             = _mm256_add_pd(fiy3,ty);
1240             fiz3             = _mm256_add_pd(fiz3,tz);
1241
1242             fjx2             = _mm256_add_pd(fjx2,tx);
1243             fjy2             = _mm256_add_pd(fjy2,ty);
1244             fjz2             = _mm256_add_pd(fjz2,tz);
1245
1246             }
1247
1248             /**************************
1249              * CALCULATE INTERACTIONS *
1250              **************************/
1251
1252             if (gmx_mm256_any_lt(rsq33,rcutoff2))
1253             {
1254
1255             /* REACTION-FIELD ELECTROSTATICS */
1256             velec            = _mm256_mul_pd(qq33,_mm256_sub_pd(_mm256_add_pd(rinv33,_mm256_mul_pd(krf,rsq33)),crf));
1257             felec            = _mm256_mul_pd(qq33,_mm256_sub_pd(_mm256_mul_pd(rinv33,rinvsq33),krf2));
1258
1259             cutoff_mask      = _mm256_cmp_pd(rsq33,rcutoff2,_CMP_LT_OQ);
1260
1261             /* Update potential sum for this i atom from the interaction with this j atom. */
1262             velec            = _mm256_and_pd(velec,cutoff_mask);
1263             velec            = _mm256_andnot_pd(dummy_mask,velec);
1264             velecsum         = _mm256_add_pd(velecsum,velec);
1265
1266             fscal            = felec;
1267
1268             fscal            = _mm256_and_pd(fscal,cutoff_mask);
1269
1270             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
1271
1272             /* Calculate temporary vectorial force */
1273             tx               = _mm256_mul_pd(fscal,dx33);
1274             ty               = _mm256_mul_pd(fscal,dy33);
1275             tz               = _mm256_mul_pd(fscal,dz33);
1276
1277             /* Update vectorial force */
1278             fix3             = _mm256_add_pd(fix3,tx);
1279             fiy3             = _mm256_add_pd(fiy3,ty);
1280             fiz3             = _mm256_add_pd(fiz3,tz);
1281
1282             fjx3             = _mm256_add_pd(fjx3,tx);
1283             fjy3             = _mm256_add_pd(fjy3,ty);
1284             fjz3             = _mm256_add_pd(fjz3,tz);
1285
1286             }
1287
1288             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
1289             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
1290             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
1291             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
1292
1293             gmx_mm256_decrement_4rvec_4ptr_swizzle_pd(fjptrA,fjptrB,fjptrC,fjptrD,
1294                                                       fjx0,fjy0,fjz0,fjx1,fjy1,fjz1,
1295                                                       fjx2,fjy2,fjz2,fjx3,fjy3,fjz3);
1296
1297             /* Inner loop uses 388 flops */
1298         }
1299
1300         /* End of innermost loop */
1301
1302         gmx_mm256_update_iforce_4atom_swizzle_pd(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,fix3,fiy3,fiz3,
1303                                                  f+i_coord_offset,fshift+i_shift_offset);
1304
1305         ggid                        = gid[iidx];
1306         /* Update potential energies */
1307         gmx_mm256_update_1pot_pd(velecsum,kernel_data->energygrp_elec+ggid);
1308         gmx_mm256_update_1pot_pd(vvdwsum,kernel_data->energygrp_vdw+ggid);
1309
1310         /* Increment number of inner iterations */
1311         inneriter                  += j_index_end - j_index_start;
1312
1313         /* Outer loop uses 26 flops */
1314     }
1315
1316     /* Increment number of outer iterations */
1317     outeriter        += nri;
1318
1319     /* Update outer/inner flops */
1320
1321     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W4W4_VF,outeriter*26 + inneriter*388);
1322 }
1323 /*
1324  * Gromacs nonbonded kernel:   nb_kernel_ElecRFCut_VdwCSTab_GeomW4W4_F_avx_256_double
1325  * Electrostatics interaction: ReactionField
1326  * VdW interaction:            CubicSplineTable
1327  * Geometry:                   Water4-Water4
1328  * Calculate force/pot:        Force
1329  */
1330 void
1331 nb_kernel_ElecRFCut_VdwCSTab_GeomW4W4_F_avx_256_double
1332                     (t_nblist                    * gmx_restrict       nlist,
1333                      rvec                        * gmx_restrict          xx,
1334                      rvec                        * gmx_restrict          ff,
1335                      t_forcerec                  * gmx_restrict          fr,
1336                      t_mdatoms                   * gmx_restrict     mdatoms,
1337                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
1338                      t_nrnb                      * gmx_restrict        nrnb)
1339 {
1340     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
1341      * just 0 for non-waters.
1342      * Suffixes A,B,C,D refer to j loop unrolling done with AVX, e.g. for the four different
1343      * jnr indices corresponding to data put in the four positions in the SIMD register.
1344      */
1345     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
1346     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
1347     int              jnrA,jnrB,jnrC,jnrD;
1348     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
1349     int              jnrlistE,jnrlistF,jnrlistG,jnrlistH;
1350     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
1351     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
1352     real             rcutoff_scalar;
1353     real             *shiftvec,*fshift,*x,*f;
1354     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD;
1355     real             scratch[4*DIM];
1356     __m256d          tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
1357     real *           vdwioffsetptr0;
1358     __m256d          ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
1359     real *           vdwioffsetptr1;
1360     __m256d          ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
1361     real *           vdwioffsetptr2;
1362     __m256d          ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
1363     real *           vdwioffsetptr3;
1364     __m256d          ix3,iy3,iz3,fix3,fiy3,fiz3,iq3,isai3;
1365     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
1366     __m256d          jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
1367     int              vdwjidx1A,vdwjidx1B,vdwjidx1C,vdwjidx1D;
1368     __m256d          jx1,jy1,jz1,fjx1,fjy1,fjz1,jq1,isaj1;
1369     int              vdwjidx2A,vdwjidx2B,vdwjidx2C,vdwjidx2D;
1370     __m256d          jx2,jy2,jz2,fjx2,fjy2,fjz2,jq2,isaj2;
1371     int              vdwjidx3A,vdwjidx3B,vdwjidx3C,vdwjidx3D;
1372     __m256d          jx3,jy3,jz3,fjx3,fjy3,fjz3,jq3,isaj3;
1373     __m256d          dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
1374     __m256d          dx11,dy11,dz11,rsq11,rinv11,rinvsq11,r11,qq11,c6_11,c12_11;
1375     __m256d          dx12,dy12,dz12,rsq12,rinv12,rinvsq12,r12,qq12,c6_12,c12_12;
1376     __m256d          dx13,dy13,dz13,rsq13,rinv13,rinvsq13,r13,qq13,c6_13,c12_13;
1377     __m256d          dx21,dy21,dz21,rsq21,rinv21,rinvsq21,r21,qq21,c6_21,c12_21;
1378     __m256d          dx22,dy22,dz22,rsq22,rinv22,rinvsq22,r22,qq22,c6_22,c12_22;
1379     __m256d          dx23,dy23,dz23,rsq23,rinv23,rinvsq23,r23,qq23,c6_23,c12_23;
1380     __m256d          dx31,dy31,dz31,rsq31,rinv31,rinvsq31,r31,qq31,c6_31,c12_31;
1381     __m256d          dx32,dy32,dz32,rsq32,rinv32,rinvsq32,r32,qq32,c6_32,c12_32;
1382     __m256d          dx33,dy33,dz33,rsq33,rinv33,rinvsq33,r33,qq33,c6_33,c12_33;
1383     __m256d          velec,felec,velecsum,facel,crf,krf,krf2;
1384     real             *charge;
1385     int              nvdwtype;
1386     __m256d          rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
1387     int              *vdwtype;
1388     real             *vdwparam;
1389     __m256d          one_sixth   = _mm256_set1_pd(1.0/6.0);
1390     __m256d          one_twelfth = _mm256_set1_pd(1.0/12.0);
1391     __m128i          vfitab;
1392     __m128i          ifour       = _mm_set1_epi32(4);
1393     __m256d          rt,vfeps,vftabscale,Y,F,G,H,Heps,Fp,VV,FF;
1394     real             *vftab;
1395     __m256d          dummy_mask,cutoff_mask;
1396     __m128           tmpmask0,tmpmask1;
1397     __m256d          signbit = _mm256_castsi256_pd( _mm256_set1_epi32(0x80000000) );
1398     __m256d          one     = _mm256_set1_pd(1.0);
1399     __m256d          two     = _mm256_set1_pd(2.0);
1400     x                = xx[0];
1401     f                = ff[0];
1402
1403     nri              = nlist->nri;
1404     iinr             = nlist->iinr;
1405     jindex           = nlist->jindex;
1406     jjnr             = nlist->jjnr;
1407     shiftidx         = nlist->shift;
1408     gid              = nlist->gid;
1409     shiftvec         = fr->shift_vec[0];
1410     fshift           = fr->fshift[0];
1411     facel            = _mm256_set1_pd(fr->epsfac);
1412     charge           = mdatoms->chargeA;
1413     krf              = _mm256_set1_pd(fr->ic->k_rf);
1414     krf2             = _mm256_set1_pd(fr->ic->k_rf*2.0);
1415     crf              = _mm256_set1_pd(fr->ic->c_rf);
1416     nvdwtype         = fr->ntype;
1417     vdwparam         = fr->nbfp;
1418     vdwtype          = mdatoms->typeA;
1419
1420     vftab            = kernel_data->table_vdw->data;
1421     vftabscale       = _mm256_set1_pd(kernel_data->table_vdw->scale);
1422
1423     /* Setup water-specific parameters */
1424     inr              = nlist->iinr[0];
1425     iq1              = _mm256_mul_pd(facel,_mm256_set1_pd(charge[inr+1]));
1426     iq2              = _mm256_mul_pd(facel,_mm256_set1_pd(charge[inr+2]));
1427     iq3              = _mm256_mul_pd(facel,_mm256_set1_pd(charge[inr+3]));
1428     vdwioffsetptr0   = vdwparam+2*nvdwtype*vdwtype[inr+0];
1429
1430     jq1              = _mm256_set1_pd(charge[inr+1]);
1431     jq2              = _mm256_set1_pd(charge[inr+2]);
1432     jq3              = _mm256_set1_pd(charge[inr+3]);
1433     vdwjidx0A        = 2*vdwtype[inr+0];
1434     c6_00            = _mm256_set1_pd(vdwioffsetptr0[vdwjidx0A]);
1435     c12_00           = _mm256_set1_pd(vdwioffsetptr0[vdwjidx0A+1]);
1436     qq11             = _mm256_mul_pd(iq1,jq1);
1437     qq12             = _mm256_mul_pd(iq1,jq2);
1438     qq13             = _mm256_mul_pd(iq1,jq3);
1439     qq21             = _mm256_mul_pd(iq2,jq1);
1440     qq22             = _mm256_mul_pd(iq2,jq2);
1441     qq23             = _mm256_mul_pd(iq2,jq3);
1442     qq31             = _mm256_mul_pd(iq3,jq1);
1443     qq32             = _mm256_mul_pd(iq3,jq2);
1444     qq33             = _mm256_mul_pd(iq3,jq3);
1445
1446     /* When we use explicit cutoffs the value must be identical for elec and VdW, so use elec as an arbitrary choice */
1447     rcutoff_scalar   = fr->rcoulomb;
1448     rcutoff          = _mm256_set1_pd(rcutoff_scalar);
1449     rcutoff2         = _mm256_mul_pd(rcutoff,rcutoff);
1450
1451     /* Avoid stupid compiler warnings */
1452     jnrA = jnrB = jnrC = jnrD = 0;
1453     j_coord_offsetA = 0;
1454     j_coord_offsetB = 0;
1455     j_coord_offsetC = 0;
1456     j_coord_offsetD = 0;
1457
1458     outeriter        = 0;
1459     inneriter        = 0;
1460
1461     for(iidx=0;iidx<4*DIM;iidx++)
1462     {
1463         scratch[iidx] = 0.0;
1464     }
1465
1466     /* Start outer loop over neighborlists */
1467     for(iidx=0; iidx<nri; iidx++)
1468     {
1469         /* Load shift vector for this list */
1470         i_shift_offset   = DIM*shiftidx[iidx];
1471
1472         /* Load limits for loop over neighbors */
1473         j_index_start    = jindex[iidx];
1474         j_index_end      = jindex[iidx+1];
1475
1476         /* Get outer coordinate index */
1477         inr              = iinr[iidx];
1478         i_coord_offset   = DIM*inr;
1479
1480         /* Load i particle coords and add shift vector */
1481         gmx_mm256_load_shift_and_4rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset,
1482                                                     &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2,&ix3,&iy3,&iz3);
1483
1484         fix0             = _mm256_setzero_pd();
1485         fiy0             = _mm256_setzero_pd();
1486         fiz0             = _mm256_setzero_pd();
1487         fix1             = _mm256_setzero_pd();
1488         fiy1             = _mm256_setzero_pd();
1489         fiz1             = _mm256_setzero_pd();
1490         fix2             = _mm256_setzero_pd();
1491         fiy2             = _mm256_setzero_pd();
1492         fiz2             = _mm256_setzero_pd();
1493         fix3             = _mm256_setzero_pd();
1494         fiy3             = _mm256_setzero_pd();
1495         fiz3             = _mm256_setzero_pd();
1496
1497         /* Start inner kernel loop */
1498         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
1499         {
1500
1501             /* Get j neighbor index, and coordinate index */
1502             jnrA             = jjnr[jidx];
1503             jnrB             = jjnr[jidx+1];
1504             jnrC             = jjnr[jidx+2];
1505             jnrD             = jjnr[jidx+3];
1506             j_coord_offsetA  = DIM*jnrA;
1507             j_coord_offsetB  = DIM*jnrB;
1508             j_coord_offsetC  = DIM*jnrC;
1509             j_coord_offsetD  = DIM*jnrD;
1510
1511             /* load j atom coordinates */
1512             gmx_mm256_load_4rvec_4ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
1513                                                  x+j_coord_offsetC,x+j_coord_offsetD,
1514                                                  &jx0,&jy0,&jz0,&jx1,&jy1,&jz1,&jx2,
1515                                                  &jy2,&jz2,&jx3,&jy3,&jz3);
1516
1517             /* Calculate displacement vector */
1518             dx00             = _mm256_sub_pd(ix0,jx0);
1519             dy00             = _mm256_sub_pd(iy0,jy0);
1520             dz00             = _mm256_sub_pd(iz0,jz0);
1521             dx11             = _mm256_sub_pd(ix1,jx1);
1522             dy11             = _mm256_sub_pd(iy1,jy1);
1523             dz11             = _mm256_sub_pd(iz1,jz1);
1524             dx12             = _mm256_sub_pd(ix1,jx2);
1525             dy12             = _mm256_sub_pd(iy1,jy2);
1526             dz12             = _mm256_sub_pd(iz1,jz2);
1527             dx13             = _mm256_sub_pd(ix1,jx3);
1528             dy13             = _mm256_sub_pd(iy1,jy3);
1529             dz13             = _mm256_sub_pd(iz1,jz3);
1530             dx21             = _mm256_sub_pd(ix2,jx1);
1531             dy21             = _mm256_sub_pd(iy2,jy1);
1532             dz21             = _mm256_sub_pd(iz2,jz1);
1533             dx22             = _mm256_sub_pd(ix2,jx2);
1534             dy22             = _mm256_sub_pd(iy2,jy2);
1535             dz22             = _mm256_sub_pd(iz2,jz2);
1536             dx23             = _mm256_sub_pd(ix2,jx3);
1537             dy23             = _mm256_sub_pd(iy2,jy3);
1538             dz23             = _mm256_sub_pd(iz2,jz3);
1539             dx31             = _mm256_sub_pd(ix3,jx1);
1540             dy31             = _mm256_sub_pd(iy3,jy1);
1541             dz31             = _mm256_sub_pd(iz3,jz1);
1542             dx32             = _mm256_sub_pd(ix3,jx2);
1543             dy32             = _mm256_sub_pd(iy3,jy2);
1544             dz32             = _mm256_sub_pd(iz3,jz2);
1545             dx33             = _mm256_sub_pd(ix3,jx3);
1546             dy33             = _mm256_sub_pd(iy3,jy3);
1547             dz33             = _mm256_sub_pd(iz3,jz3);
1548
1549             /* Calculate squared distance and things based on it */
1550             rsq00            = gmx_mm256_calc_rsq_pd(dx00,dy00,dz00);
1551             rsq11            = gmx_mm256_calc_rsq_pd(dx11,dy11,dz11);
1552             rsq12            = gmx_mm256_calc_rsq_pd(dx12,dy12,dz12);
1553             rsq13            = gmx_mm256_calc_rsq_pd(dx13,dy13,dz13);
1554             rsq21            = gmx_mm256_calc_rsq_pd(dx21,dy21,dz21);
1555             rsq22            = gmx_mm256_calc_rsq_pd(dx22,dy22,dz22);
1556             rsq23            = gmx_mm256_calc_rsq_pd(dx23,dy23,dz23);
1557             rsq31            = gmx_mm256_calc_rsq_pd(dx31,dy31,dz31);
1558             rsq32            = gmx_mm256_calc_rsq_pd(dx32,dy32,dz32);
1559             rsq33            = gmx_mm256_calc_rsq_pd(dx33,dy33,dz33);
1560
1561             rinv00           = gmx_mm256_invsqrt_pd(rsq00);
1562             rinv11           = gmx_mm256_invsqrt_pd(rsq11);
1563             rinv12           = gmx_mm256_invsqrt_pd(rsq12);
1564             rinv13           = gmx_mm256_invsqrt_pd(rsq13);
1565             rinv21           = gmx_mm256_invsqrt_pd(rsq21);
1566             rinv22           = gmx_mm256_invsqrt_pd(rsq22);
1567             rinv23           = gmx_mm256_invsqrt_pd(rsq23);
1568             rinv31           = gmx_mm256_invsqrt_pd(rsq31);
1569             rinv32           = gmx_mm256_invsqrt_pd(rsq32);
1570             rinv33           = gmx_mm256_invsqrt_pd(rsq33);
1571
1572             rinvsq11         = _mm256_mul_pd(rinv11,rinv11);
1573             rinvsq12         = _mm256_mul_pd(rinv12,rinv12);
1574             rinvsq13         = _mm256_mul_pd(rinv13,rinv13);
1575             rinvsq21         = _mm256_mul_pd(rinv21,rinv21);
1576             rinvsq22         = _mm256_mul_pd(rinv22,rinv22);
1577             rinvsq23         = _mm256_mul_pd(rinv23,rinv23);
1578             rinvsq31         = _mm256_mul_pd(rinv31,rinv31);
1579             rinvsq32         = _mm256_mul_pd(rinv32,rinv32);
1580             rinvsq33         = _mm256_mul_pd(rinv33,rinv33);
1581
1582             fjx0             = _mm256_setzero_pd();
1583             fjy0             = _mm256_setzero_pd();
1584             fjz0             = _mm256_setzero_pd();
1585             fjx1             = _mm256_setzero_pd();
1586             fjy1             = _mm256_setzero_pd();
1587             fjz1             = _mm256_setzero_pd();
1588             fjx2             = _mm256_setzero_pd();
1589             fjy2             = _mm256_setzero_pd();
1590             fjz2             = _mm256_setzero_pd();
1591             fjx3             = _mm256_setzero_pd();
1592             fjy3             = _mm256_setzero_pd();
1593             fjz3             = _mm256_setzero_pd();
1594
1595             /**************************
1596              * CALCULATE INTERACTIONS *
1597              **************************/
1598
1599             if (gmx_mm256_any_lt(rsq00,rcutoff2))
1600             {
1601
1602             r00              = _mm256_mul_pd(rsq00,rinv00);
1603
1604             /* Calculate table index by multiplying r with table scale and truncate to integer */
1605             rt               = _mm256_mul_pd(r00,vftabscale);
1606             vfitab           = _mm256_cvttpd_epi32(rt);
1607             vfeps            = _mm256_sub_pd(rt,_mm256_round_pd(rt, _MM_FROUND_FLOOR));
1608             vfitab           = _mm_slli_epi32(vfitab,3);
1609
1610             /* CUBIC SPLINE TABLE DISPERSION */
1611             Y                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
1612             F                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
1613             G                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,2) );
1614             H                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,3) );
1615             GMX_MM256_FULLTRANSPOSE4_PD(Y,F,G,H);
1616             Heps             = _mm256_mul_pd(vfeps,H);
1617             Fp               = _mm256_add_pd(F,_mm256_mul_pd(vfeps,_mm256_add_pd(G,Heps)));
1618             FF               = _mm256_add_pd(Fp,_mm256_mul_pd(vfeps,_mm256_add_pd(G,_mm256_add_pd(Heps,Heps))));
1619             fvdw6            = _mm256_mul_pd(c6_00,FF);
1620
1621             /* CUBIC SPLINE TABLE REPULSION */
1622             vfitab           = _mm_add_epi32(vfitab,ifour);
1623             Y                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
1624             F                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
1625             G                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,2) );
1626             H                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,3) );
1627             GMX_MM256_FULLTRANSPOSE4_PD(Y,F,G,H);
1628             Heps             = _mm256_mul_pd(vfeps,H);
1629             Fp               = _mm256_add_pd(F,_mm256_mul_pd(vfeps,_mm256_add_pd(G,Heps)));
1630             FF               = _mm256_add_pd(Fp,_mm256_mul_pd(vfeps,_mm256_add_pd(G,_mm256_add_pd(Heps,Heps))));
1631             fvdw12           = _mm256_mul_pd(c12_00,FF);
1632             fvdw             = _mm256_xor_pd(signbit,_mm256_mul_pd(_mm256_add_pd(fvdw6,fvdw12),_mm256_mul_pd(vftabscale,rinv00)));
1633
1634             cutoff_mask      = _mm256_cmp_pd(rsq00,rcutoff2,_CMP_LT_OQ);
1635
1636             fscal            = fvdw;
1637
1638             fscal            = _mm256_and_pd(fscal,cutoff_mask);
1639
1640             /* Calculate temporary vectorial force */
1641             tx               = _mm256_mul_pd(fscal,dx00);
1642             ty               = _mm256_mul_pd(fscal,dy00);
1643             tz               = _mm256_mul_pd(fscal,dz00);
1644
1645             /* Update vectorial force */
1646             fix0             = _mm256_add_pd(fix0,tx);
1647             fiy0             = _mm256_add_pd(fiy0,ty);
1648             fiz0             = _mm256_add_pd(fiz0,tz);
1649
1650             fjx0             = _mm256_add_pd(fjx0,tx);
1651             fjy0             = _mm256_add_pd(fjy0,ty);
1652             fjz0             = _mm256_add_pd(fjz0,tz);
1653
1654             }
1655
1656             /**************************
1657              * CALCULATE INTERACTIONS *
1658              **************************/
1659
1660             if (gmx_mm256_any_lt(rsq11,rcutoff2))
1661             {
1662
1663             /* REACTION-FIELD ELECTROSTATICS */
1664             felec            = _mm256_mul_pd(qq11,_mm256_sub_pd(_mm256_mul_pd(rinv11,rinvsq11),krf2));
1665
1666             cutoff_mask      = _mm256_cmp_pd(rsq11,rcutoff2,_CMP_LT_OQ);
1667
1668             fscal            = felec;
1669
1670             fscal            = _mm256_and_pd(fscal,cutoff_mask);
1671
1672             /* Calculate temporary vectorial force */
1673             tx               = _mm256_mul_pd(fscal,dx11);
1674             ty               = _mm256_mul_pd(fscal,dy11);
1675             tz               = _mm256_mul_pd(fscal,dz11);
1676
1677             /* Update vectorial force */
1678             fix1             = _mm256_add_pd(fix1,tx);
1679             fiy1             = _mm256_add_pd(fiy1,ty);
1680             fiz1             = _mm256_add_pd(fiz1,tz);
1681
1682             fjx1             = _mm256_add_pd(fjx1,tx);
1683             fjy1             = _mm256_add_pd(fjy1,ty);
1684             fjz1             = _mm256_add_pd(fjz1,tz);
1685
1686             }
1687
1688             /**************************
1689              * CALCULATE INTERACTIONS *
1690              **************************/
1691
1692             if (gmx_mm256_any_lt(rsq12,rcutoff2))
1693             {
1694
1695             /* REACTION-FIELD ELECTROSTATICS */
1696             felec            = _mm256_mul_pd(qq12,_mm256_sub_pd(_mm256_mul_pd(rinv12,rinvsq12),krf2));
1697
1698             cutoff_mask      = _mm256_cmp_pd(rsq12,rcutoff2,_CMP_LT_OQ);
1699
1700             fscal            = felec;
1701
1702             fscal            = _mm256_and_pd(fscal,cutoff_mask);
1703
1704             /* Calculate temporary vectorial force */
1705             tx               = _mm256_mul_pd(fscal,dx12);
1706             ty               = _mm256_mul_pd(fscal,dy12);
1707             tz               = _mm256_mul_pd(fscal,dz12);
1708
1709             /* Update vectorial force */
1710             fix1             = _mm256_add_pd(fix1,tx);
1711             fiy1             = _mm256_add_pd(fiy1,ty);
1712             fiz1             = _mm256_add_pd(fiz1,tz);
1713
1714             fjx2             = _mm256_add_pd(fjx2,tx);
1715             fjy2             = _mm256_add_pd(fjy2,ty);
1716             fjz2             = _mm256_add_pd(fjz2,tz);
1717
1718             }
1719
1720             /**************************
1721              * CALCULATE INTERACTIONS *
1722              **************************/
1723
1724             if (gmx_mm256_any_lt(rsq13,rcutoff2))
1725             {
1726
1727             /* REACTION-FIELD ELECTROSTATICS */
1728             felec            = _mm256_mul_pd(qq13,_mm256_sub_pd(_mm256_mul_pd(rinv13,rinvsq13),krf2));
1729
1730             cutoff_mask      = _mm256_cmp_pd(rsq13,rcutoff2,_CMP_LT_OQ);
1731
1732             fscal            = felec;
1733
1734             fscal            = _mm256_and_pd(fscal,cutoff_mask);
1735
1736             /* Calculate temporary vectorial force */
1737             tx               = _mm256_mul_pd(fscal,dx13);
1738             ty               = _mm256_mul_pd(fscal,dy13);
1739             tz               = _mm256_mul_pd(fscal,dz13);
1740
1741             /* Update vectorial force */
1742             fix1             = _mm256_add_pd(fix1,tx);
1743             fiy1             = _mm256_add_pd(fiy1,ty);
1744             fiz1             = _mm256_add_pd(fiz1,tz);
1745
1746             fjx3             = _mm256_add_pd(fjx3,tx);
1747             fjy3             = _mm256_add_pd(fjy3,ty);
1748             fjz3             = _mm256_add_pd(fjz3,tz);
1749
1750             }
1751
1752             /**************************
1753              * CALCULATE INTERACTIONS *
1754              **************************/
1755
1756             if (gmx_mm256_any_lt(rsq21,rcutoff2))
1757             {
1758
1759             /* REACTION-FIELD ELECTROSTATICS */
1760             felec            = _mm256_mul_pd(qq21,_mm256_sub_pd(_mm256_mul_pd(rinv21,rinvsq21),krf2));
1761
1762             cutoff_mask      = _mm256_cmp_pd(rsq21,rcutoff2,_CMP_LT_OQ);
1763
1764             fscal            = felec;
1765
1766             fscal            = _mm256_and_pd(fscal,cutoff_mask);
1767
1768             /* Calculate temporary vectorial force */
1769             tx               = _mm256_mul_pd(fscal,dx21);
1770             ty               = _mm256_mul_pd(fscal,dy21);
1771             tz               = _mm256_mul_pd(fscal,dz21);
1772
1773             /* Update vectorial force */
1774             fix2             = _mm256_add_pd(fix2,tx);
1775             fiy2             = _mm256_add_pd(fiy2,ty);
1776             fiz2             = _mm256_add_pd(fiz2,tz);
1777
1778             fjx1             = _mm256_add_pd(fjx1,tx);
1779             fjy1             = _mm256_add_pd(fjy1,ty);
1780             fjz1             = _mm256_add_pd(fjz1,tz);
1781
1782             }
1783
1784             /**************************
1785              * CALCULATE INTERACTIONS *
1786              **************************/
1787
1788             if (gmx_mm256_any_lt(rsq22,rcutoff2))
1789             {
1790
1791             /* REACTION-FIELD ELECTROSTATICS */
1792             felec            = _mm256_mul_pd(qq22,_mm256_sub_pd(_mm256_mul_pd(rinv22,rinvsq22),krf2));
1793
1794             cutoff_mask      = _mm256_cmp_pd(rsq22,rcutoff2,_CMP_LT_OQ);
1795
1796             fscal            = felec;
1797
1798             fscal            = _mm256_and_pd(fscal,cutoff_mask);
1799
1800             /* Calculate temporary vectorial force */
1801             tx               = _mm256_mul_pd(fscal,dx22);
1802             ty               = _mm256_mul_pd(fscal,dy22);
1803             tz               = _mm256_mul_pd(fscal,dz22);
1804
1805             /* Update vectorial force */
1806             fix2             = _mm256_add_pd(fix2,tx);
1807             fiy2             = _mm256_add_pd(fiy2,ty);
1808             fiz2             = _mm256_add_pd(fiz2,tz);
1809
1810             fjx2             = _mm256_add_pd(fjx2,tx);
1811             fjy2             = _mm256_add_pd(fjy2,ty);
1812             fjz2             = _mm256_add_pd(fjz2,tz);
1813
1814             }
1815
1816             /**************************
1817              * CALCULATE INTERACTIONS *
1818              **************************/
1819
1820             if (gmx_mm256_any_lt(rsq23,rcutoff2))
1821             {
1822
1823             /* REACTION-FIELD ELECTROSTATICS */
1824             felec            = _mm256_mul_pd(qq23,_mm256_sub_pd(_mm256_mul_pd(rinv23,rinvsq23),krf2));
1825
1826             cutoff_mask      = _mm256_cmp_pd(rsq23,rcutoff2,_CMP_LT_OQ);
1827
1828             fscal            = felec;
1829
1830             fscal            = _mm256_and_pd(fscal,cutoff_mask);
1831
1832             /* Calculate temporary vectorial force */
1833             tx               = _mm256_mul_pd(fscal,dx23);
1834             ty               = _mm256_mul_pd(fscal,dy23);
1835             tz               = _mm256_mul_pd(fscal,dz23);
1836
1837             /* Update vectorial force */
1838             fix2             = _mm256_add_pd(fix2,tx);
1839             fiy2             = _mm256_add_pd(fiy2,ty);
1840             fiz2             = _mm256_add_pd(fiz2,tz);
1841
1842             fjx3             = _mm256_add_pd(fjx3,tx);
1843             fjy3             = _mm256_add_pd(fjy3,ty);
1844             fjz3             = _mm256_add_pd(fjz3,tz);
1845
1846             }
1847
1848             /**************************
1849              * CALCULATE INTERACTIONS *
1850              **************************/
1851
1852             if (gmx_mm256_any_lt(rsq31,rcutoff2))
1853             {
1854
1855             /* REACTION-FIELD ELECTROSTATICS */
1856             felec            = _mm256_mul_pd(qq31,_mm256_sub_pd(_mm256_mul_pd(rinv31,rinvsq31),krf2));
1857
1858             cutoff_mask      = _mm256_cmp_pd(rsq31,rcutoff2,_CMP_LT_OQ);
1859
1860             fscal            = felec;
1861
1862             fscal            = _mm256_and_pd(fscal,cutoff_mask);
1863
1864             /* Calculate temporary vectorial force */
1865             tx               = _mm256_mul_pd(fscal,dx31);
1866             ty               = _mm256_mul_pd(fscal,dy31);
1867             tz               = _mm256_mul_pd(fscal,dz31);
1868
1869             /* Update vectorial force */
1870             fix3             = _mm256_add_pd(fix3,tx);
1871             fiy3             = _mm256_add_pd(fiy3,ty);
1872             fiz3             = _mm256_add_pd(fiz3,tz);
1873
1874             fjx1             = _mm256_add_pd(fjx1,tx);
1875             fjy1             = _mm256_add_pd(fjy1,ty);
1876             fjz1             = _mm256_add_pd(fjz1,tz);
1877
1878             }
1879
1880             /**************************
1881              * CALCULATE INTERACTIONS *
1882              **************************/
1883
1884             if (gmx_mm256_any_lt(rsq32,rcutoff2))
1885             {
1886
1887             /* REACTION-FIELD ELECTROSTATICS */
1888             felec            = _mm256_mul_pd(qq32,_mm256_sub_pd(_mm256_mul_pd(rinv32,rinvsq32),krf2));
1889
1890             cutoff_mask      = _mm256_cmp_pd(rsq32,rcutoff2,_CMP_LT_OQ);
1891
1892             fscal            = felec;
1893
1894             fscal            = _mm256_and_pd(fscal,cutoff_mask);
1895
1896             /* Calculate temporary vectorial force */
1897             tx               = _mm256_mul_pd(fscal,dx32);
1898             ty               = _mm256_mul_pd(fscal,dy32);
1899             tz               = _mm256_mul_pd(fscal,dz32);
1900
1901             /* Update vectorial force */
1902             fix3             = _mm256_add_pd(fix3,tx);
1903             fiy3             = _mm256_add_pd(fiy3,ty);
1904             fiz3             = _mm256_add_pd(fiz3,tz);
1905
1906             fjx2             = _mm256_add_pd(fjx2,tx);
1907             fjy2             = _mm256_add_pd(fjy2,ty);
1908             fjz2             = _mm256_add_pd(fjz2,tz);
1909
1910             }
1911
1912             /**************************
1913              * CALCULATE INTERACTIONS *
1914              **************************/
1915
1916             if (gmx_mm256_any_lt(rsq33,rcutoff2))
1917             {
1918
1919             /* REACTION-FIELD ELECTROSTATICS */
1920             felec            = _mm256_mul_pd(qq33,_mm256_sub_pd(_mm256_mul_pd(rinv33,rinvsq33),krf2));
1921
1922             cutoff_mask      = _mm256_cmp_pd(rsq33,rcutoff2,_CMP_LT_OQ);
1923
1924             fscal            = felec;
1925
1926             fscal            = _mm256_and_pd(fscal,cutoff_mask);
1927
1928             /* Calculate temporary vectorial force */
1929             tx               = _mm256_mul_pd(fscal,dx33);
1930             ty               = _mm256_mul_pd(fscal,dy33);
1931             tz               = _mm256_mul_pd(fscal,dz33);
1932
1933             /* Update vectorial force */
1934             fix3             = _mm256_add_pd(fix3,tx);
1935             fiy3             = _mm256_add_pd(fiy3,ty);
1936             fiz3             = _mm256_add_pd(fiz3,tz);
1937
1938             fjx3             = _mm256_add_pd(fjx3,tx);
1939             fjy3             = _mm256_add_pd(fjy3,ty);
1940             fjz3             = _mm256_add_pd(fjz3,tz);
1941
1942             }
1943
1944             fjptrA             = f+j_coord_offsetA;
1945             fjptrB             = f+j_coord_offsetB;
1946             fjptrC             = f+j_coord_offsetC;
1947             fjptrD             = f+j_coord_offsetD;
1948
1949             gmx_mm256_decrement_4rvec_4ptr_swizzle_pd(fjptrA,fjptrB,fjptrC,fjptrD,
1950                                                       fjx0,fjy0,fjz0,fjx1,fjy1,fjz1,
1951                                                       fjx2,fjy2,fjz2,fjx3,fjy3,fjz3);
1952
1953             /* Inner loop uses 324 flops */
1954         }
1955
1956         if(jidx<j_index_end)
1957         {
1958
1959             /* Get j neighbor index, and coordinate index */
1960             jnrlistA         = jjnr[jidx];
1961             jnrlistB         = jjnr[jidx+1];
1962             jnrlistC         = jjnr[jidx+2];
1963             jnrlistD         = jjnr[jidx+3];
1964             /* Sign of each element will be negative for non-real atoms.
1965              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
1966              * so use it as val = _mm_andnot_pd(mask,val) to clear dummy entries.
1967              */
1968             tmpmask0 = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
1969
1970             tmpmask1 = _mm_permute_ps(tmpmask0,_GMX_MM_PERMUTE(3,3,2,2));
1971             tmpmask0 = _mm_permute_ps(tmpmask0,_GMX_MM_PERMUTE(1,1,0,0));
1972             dummy_mask = _mm256_castps_pd(gmx_mm256_set_m128(tmpmask1,tmpmask0));
1973
1974             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
1975             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
1976             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
1977             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
1978             j_coord_offsetA  = DIM*jnrA;
1979             j_coord_offsetB  = DIM*jnrB;
1980             j_coord_offsetC  = DIM*jnrC;
1981             j_coord_offsetD  = DIM*jnrD;
1982
1983             /* load j atom coordinates */
1984             gmx_mm256_load_4rvec_4ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
1985                                                  x+j_coord_offsetC,x+j_coord_offsetD,
1986                                                  &jx0,&jy0,&jz0,&jx1,&jy1,&jz1,&jx2,
1987                                                  &jy2,&jz2,&jx3,&jy3,&jz3);
1988
1989             /* Calculate displacement vector */
1990             dx00             = _mm256_sub_pd(ix0,jx0);
1991             dy00             = _mm256_sub_pd(iy0,jy0);
1992             dz00             = _mm256_sub_pd(iz0,jz0);
1993             dx11             = _mm256_sub_pd(ix1,jx1);
1994             dy11             = _mm256_sub_pd(iy1,jy1);
1995             dz11             = _mm256_sub_pd(iz1,jz1);
1996             dx12             = _mm256_sub_pd(ix1,jx2);
1997             dy12             = _mm256_sub_pd(iy1,jy2);
1998             dz12             = _mm256_sub_pd(iz1,jz2);
1999             dx13             = _mm256_sub_pd(ix1,jx3);
2000             dy13             = _mm256_sub_pd(iy1,jy3);
2001             dz13             = _mm256_sub_pd(iz1,jz3);
2002             dx21             = _mm256_sub_pd(ix2,jx1);
2003             dy21             = _mm256_sub_pd(iy2,jy1);
2004             dz21             = _mm256_sub_pd(iz2,jz1);
2005             dx22             = _mm256_sub_pd(ix2,jx2);
2006             dy22             = _mm256_sub_pd(iy2,jy2);
2007             dz22             = _mm256_sub_pd(iz2,jz2);
2008             dx23             = _mm256_sub_pd(ix2,jx3);
2009             dy23             = _mm256_sub_pd(iy2,jy3);
2010             dz23             = _mm256_sub_pd(iz2,jz3);
2011             dx31             = _mm256_sub_pd(ix3,jx1);
2012             dy31             = _mm256_sub_pd(iy3,jy1);
2013             dz31             = _mm256_sub_pd(iz3,jz1);
2014             dx32             = _mm256_sub_pd(ix3,jx2);
2015             dy32             = _mm256_sub_pd(iy3,jy2);
2016             dz32             = _mm256_sub_pd(iz3,jz2);
2017             dx33             = _mm256_sub_pd(ix3,jx3);
2018             dy33             = _mm256_sub_pd(iy3,jy3);
2019             dz33             = _mm256_sub_pd(iz3,jz3);
2020
2021             /* Calculate squared distance and things based on it */
2022             rsq00            = gmx_mm256_calc_rsq_pd(dx00,dy00,dz00);
2023             rsq11            = gmx_mm256_calc_rsq_pd(dx11,dy11,dz11);
2024             rsq12            = gmx_mm256_calc_rsq_pd(dx12,dy12,dz12);
2025             rsq13            = gmx_mm256_calc_rsq_pd(dx13,dy13,dz13);
2026             rsq21            = gmx_mm256_calc_rsq_pd(dx21,dy21,dz21);
2027             rsq22            = gmx_mm256_calc_rsq_pd(dx22,dy22,dz22);
2028             rsq23            = gmx_mm256_calc_rsq_pd(dx23,dy23,dz23);
2029             rsq31            = gmx_mm256_calc_rsq_pd(dx31,dy31,dz31);
2030             rsq32            = gmx_mm256_calc_rsq_pd(dx32,dy32,dz32);
2031             rsq33            = gmx_mm256_calc_rsq_pd(dx33,dy33,dz33);
2032
2033             rinv00           = gmx_mm256_invsqrt_pd(rsq00);
2034             rinv11           = gmx_mm256_invsqrt_pd(rsq11);
2035             rinv12           = gmx_mm256_invsqrt_pd(rsq12);
2036             rinv13           = gmx_mm256_invsqrt_pd(rsq13);
2037             rinv21           = gmx_mm256_invsqrt_pd(rsq21);
2038             rinv22           = gmx_mm256_invsqrt_pd(rsq22);
2039             rinv23           = gmx_mm256_invsqrt_pd(rsq23);
2040             rinv31           = gmx_mm256_invsqrt_pd(rsq31);
2041             rinv32           = gmx_mm256_invsqrt_pd(rsq32);
2042             rinv33           = gmx_mm256_invsqrt_pd(rsq33);
2043
2044             rinvsq11         = _mm256_mul_pd(rinv11,rinv11);
2045             rinvsq12         = _mm256_mul_pd(rinv12,rinv12);
2046             rinvsq13         = _mm256_mul_pd(rinv13,rinv13);
2047             rinvsq21         = _mm256_mul_pd(rinv21,rinv21);
2048             rinvsq22         = _mm256_mul_pd(rinv22,rinv22);
2049             rinvsq23         = _mm256_mul_pd(rinv23,rinv23);
2050             rinvsq31         = _mm256_mul_pd(rinv31,rinv31);
2051             rinvsq32         = _mm256_mul_pd(rinv32,rinv32);
2052             rinvsq33         = _mm256_mul_pd(rinv33,rinv33);
2053
2054             fjx0             = _mm256_setzero_pd();
2055             fjy0             = _mm256_setzero_pd();
2056             fjz0             = _mm256_setzero_pd();
2057             fjx1             = _mm256_setzero_pd();
2058             fjy1             = _mm256_setzero_pd();
2059             fjz1             = _mm256_setzero_pd();
2060             fjx2             = _mm256_setzero_pd();
2061             fjy2             = _mm256_setzero_pd();
2062             fjz2             = _mm256_setzero_pd();
2063             fjx3             = _mm256_setzero_pd();
2064             fjy3             = _mm256_setzero_pd();
2065             fjz3             = _mm256_setzero_pd();
2066
2067             /**************************
2068              * CALCULATE INTERACTIONS *
2069              **************************/
2070
2071             if (gmx_mm256_any_lt(rsq00,rcutoff2))
2072             {
2073
2074             r00              = _mm256_mul_pd(rsq00,rinv00);
2075             r00              = _mm256_andnot_pd(dummy_mask,r00);
2076
2077             /* Calculate table index by multiplying r with table scale and truncate to integer */
2078             rt               = _mm256_mul_pd(r00,vftabscale);
2079             vfitab           = _mm256_cvttpd_epi32(rt);
2080             vfeps            = _mm256_sub_pd(rt,_mm256_round_pd(rt, _MM_FROUND_FLOOR));
2081             vfitab           = _mm_slli_epi32(vfitab,3);
2082
2083             /* CUBIC SPLINE TABLE DISPERSION */
2084             Y                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
2085             F                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
2086             G                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,2) );
2087             H                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,3) );
2088             GMX_MM256_FULLTRANSPOSE4_PD(Y,F,G,H);
2089             Heps             = _mm256_mul_pd(vfeps,H);
2090             Fp               = _mm256_add_pd(F,_mm256_mul_pd(vfeps,_mm256_add_pd(G,Heps)));
2091             FF               = _mm256_add_pd(Fp,_mm256_mul_pd(vfeps,_mm256_add_pd(G,_mm256_add_pd(Heps,Heps))));
2092             fvdw6            = _mm256_mul_pd(c6_00,FF);
2093
2094             /* CUBIC SPLINE TABLE REPULSION */
2095             vfitab           = _mm_add_epi32(vfitab,ifour);
2096             Y                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
2097             F                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
2098             G                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,2) );
2099             H                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,3) );
2100             GMX_MM256_FULLTRANSPOSE4_PD(Y,F,G,H);
2101             Heps             = _mm256_mul_pd(vfeps,H);
2102             Fp               = _mm256_add_pd(F,_mm256_mul_pd(vfeps,_mm256_add_pd(G,Heps)));
2103             FF               = _mm256_add_pd(Fp,_mm256_mul_pd(vfeps,_mm256_add_pd(G,_mm256_add_pd(Heps,Heps))));
2104             fvdw12           = _mm256_mul_pd(c12_00,FF);
2105             fvdw             = _mm256_xor_pd(signbit,_mm256_mul_pd(_mm256_add_pd(fvdw6,fvdw12),_mm256_mul_pd(vftabscale,rinv00)));
2106
2107             cutoff_mask      = _mm256_cmp_pd(rsq00,rcutoff2,_CMP_LT_OQ);
2108
2109             fscal            = fvdw;
2110
2111             fscal            = _mm256_and_pd(fscal,cutoff_mask);
2112
2113             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
2114
2115             /* Calculate temporary vectorial force */
2116             tx               = _mm256_mul_pd(fscal,dx00);
2117             ty               = _mm256_mul_pd(fscal,dy00);
2118             tz               = _mm256_mul_pd(fscal,dz00);
2119
2120             /* Update vectorial force */
2121             fix0             = _mm256_add_pd(fix0,tx);
2122             fiy0             = _mm256_add_pd(fiy0,ty);
2123             fiz0             = _mm256_add_pd(fiz0,tz);
2124
2125             fjx0             = _mm256_add_pd(fjx0,tx);
2126             fjy0             = _mm256_add_pd(fjy0,ty);
2127             fjz0             = _mm256_add_pd(fjz0,tz);
2128
2129             }
2130
2131             /**************************
2132              * CALCULATE INTERACTIONS *
2133              **************************/
2134
2135             if (gmx_mm256_any_lt(rsq11,rcutoff2))
2136             {
2137
2138             /* REACTION-FIELD ELECTROSTATICS */
2139             felec            = _mm256_mul_pd(qq11,_mm256_sub_pd(_mm256_mul_pd(rinv11,rinvsq11),krf2));
2140
2141             cutoff_mask      = _mm256_cmp_pd(rsq11,rcutoff2,_CMP_LT_OQ);
2142
2143             fscal            = felec;
2144
2145             fscal            = _mm256_and_pd(fscal,cutoff_mask);
2146
2147             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
2148
2149             /* Calculate temporary vectorial force */
2150             tx               = _mm256_mul_pd(fscal,dx11);
2151             ty               = _mm256_mul_pd(fscal,dy11);
2152             tz               = _mm256_mul_pd(fscal,dz11);
2153
2154             /* Update vectorial force */
2155             fix1             = _mm256_add_pd(fix1,tx);
2156             fiy1             = _mm256_add_pd(fiy1,ty);
2157             fiz1             = _mm256_add_pd(fiz1,tz);
2158
2159             fjx1             = _mm256_add_pd(fjx1,tx);
2160             fjy1             = _mm256_add_pd(fjy1,ty);
2161             fjz1             = _mm256_add_pd(fjz1,tz);
2162
2163             }
2164
2165             /**************************
2166              * CALCULATE INTERACTIONS *
2167              **************************/
2168
2169             if (gmx_mm256_any_lt(rsq12,rcutoff2))
2170             {
2171
2172             /* REACTION-FIELD ELECTROSTATICS */
2173             felec            = _mm256_mul_pd(qq12,_mm256_sub_pd(_mm256_mul_pd(rinv12,rinvsq12),krf2));
2174
2175             cutoff_mask      = _mm256_cmp_pd(rsq12,rcutoff2,_CMP_LT_OQ);
2176
2177             fscal            = felec;
2178
2179             fscal            = _mm256_and_pd(fscal,cutoff_mask);
2180
2181             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
2182
2183             /* Calculate temporary vectorial force */
2184             tx               = _mm256_mul_pd(fscal,dx12);
2185             ty               = _mm256_mul_pd(fscal,dy12);
2186             tz               = _mm256_mul_pd(fscal,dz12);
2187
2188             /* Update vectorial force */
2189             fix1             = _mm256_add_pd(fix1,tx);
2190             fiy1             = _mm256_add_pd(fiy1,ty);
2191             fiz1             = _mm256_add_pd(fiz1,tz);
2192
2193             fjx2             = _mm256_add_pd(fjx2,tx);
2194             fjy2             = _mm256_add_pd(fjy2,ty);
2195             fjz2             = _mm256_add_pd(fjz2,tz);
2196
2197             }
2198
2199             /**************************
2200              * CALCULATE INTERACTIONS *
2201              **************************/
2202
2203             if (gmx_mm256_any_lt(rsq13,rcutoff2))
2204             {
2205
2206             /* REACTION-FIELD ELECTROSTATICS */
2207             felec            = _mm256_mul_pd(qq13,_mm256_sub_pd(_mm256_mul_pd(rinv13,rinvsq13),krf2));
2208
2209             cutoff_mask      = _mm256_cmp_pd(rsq13,rcutoff2,_CMP_LT_OQ);
2210
2211             fscal            = felec;
2212
2213             fscal            = _mm256_and_pd(fscal,cutoff_mask);
2214
2215             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
2216
2217             /* Calculate temporary vectorial force */
2218             tx               = _mm256_mul_pd(fscal,dx13);
2219             ty               = _mm256_mul_pd(fscal,dy13);
2220             tz               = _mm256_mul_pd(fscal,dz13);
2221
2222             /* Update vectorial force */
2223             fix1             = _mm256_add_pd(fix1,tx);
2224             fiy1             = _mm256_add_pd(fiy1,ty);
2225             fiz1             = _mm256_add_pd(fiz1,tz);
2226
2227             fjx3             = _mm256_add_pd(fjx3,tx);
2228             fjy3             = _mm256_add_pd(fjy3,ty);
2229             fjz3             = _mm256_add_pd(fjz3,tz);
2230
2231             }
2232
2233             /**************************
2234              * CALCULATE INTERACTIONS *
2235              **************************/
2236
2237             if (gmx_mm256_any_lt(rsq21,rcutoff2))
2238             {
2239
2240             /* REACTION-FIELD ELECTROSTATICS */
2241             felec            = _mm256_mul_pd(qq21,_mm256_sub_pd(_mm256_mul_pd(rinv21,rinvsq21),krf2));
2242
2243             cutoff_mask      = _mm256_cmp_pd(rsq21,rcutoff2,_CMP_LT_OQ);
2244
2245             fscal            = felec;
2246
2247             fscal            = _mm256_and_pd(fscal,cutoff_mask);
2248
2249             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
2250
2251             /* Calculate temporary vectorial force */
2252             tx               = _mm256_mul_pd(fscal,dx21);
2253             ty               = _mm256_mul_pd(fscal,dy21);
2254             tz               = _mm256_mul_pd(fscal,dz21);
2255
2256             /* Update vectorial force */
2257             fix2             = _mm256_add_pd(fix2,tx);
2258             fiy2             = _mm256_add_pd(fiy2,ty);
2259             fiz2             = _mm256_add_pd(fiz2,tz);
2260
2261             fjx1             = _mm256_add_pd(fjx1,tx);
2262             fjy1             = _mm256_add_pd(fjy1,ty);
2263             fjz1             = _mm256_add_pd(fjz1,tz);
2264
2265             }
2266
2267             /**************************
2268              * CALCULATE INTERACTIONS *
2269              **************************/
2270
2271             if (gmx_mm256_any_lt(rsq22,rcutoff2))
2272             {
2273
2274             /* REACTION-FIELD ELECTROSTATICS */
2275             felec            = _mm256_mul_pd(qq22,_mm256_sub_pd(_mm256_mul_pd(rinv22,rinvsq22),krf2));
2276
2277             cutoff_mask      = _mm256_cmp_pd(rsq22,rcutoff2,_CMP_LT_OQ);
2278
2279             fscal            = felec;
2280
2281             fscal            = _mm256_and_pd(fscal,cutoff_mask);
2282
2283             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
2284
2285             /* Calculate temporary vectorial force */
2286             tx               = _mm256_mul_pd(fscal,dx22);
2287             ty               = _mm256_mul_pd(fscal,dy22);
2288             tz               = _mm256_mul_pd(fscal,dz22);
2289
2290             /* Update vectorial force */
2291             fix2             = _mm256_add_pd(fix2,tx);
2292             fiy2             = _mm256_add_pd(fiy2,ty);
2293             fiz2             = _mm256_add_pd(fiz2,tz);
2294
2295             fjx2             = _mm256_add_pd(fjx2,tx);
2296             fjy2             = _mm256_add_pd(fjy2,ty);
2297             fjz2             = _mm256_add_pd(fjz2,tz);
2298
2299             }
2300
2301             /**************************
2302              * CALCULATE INTERACTIONS *
2303              **************************/
2304
2305             if (gmx_mm256_any_lt(rsq23,rcutoff2))
2306             {
2307
2308             /* REACTION-FIELD ELECTROSTATICS */
2309             felec            = _mm256_mul_pd(qq23,_mm256_sub_pd(_mm256_mul_pd(rinv23,rinvsq23),krf2));
2310
2311             cutoff_mask      = _mm256_cmp_pd(rsq23,rcutoff2,_CMP_LT_OQ);
2312
2313             fscal            = felec;
2314
2315             fscal            = _mm256_and_pd(fscal,cutoff_mask);
2316
2317             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
2318
2319             /* Calculate temporary vectorial force */
2320             tx               = _mm256_mul_pd(fscal,dx23);
2321             ty               = _mm256_mul_pd(fscal,dy23);
2322             tz               = _mm256_mul_pd(fscal,dz23);
2323
2324             /* Update vectorial force */
2325             fix2             = _mm256_add_pd(fix2,tx);
2326             fiy2             = _mm256_add_pd(fiy2,ty);
2327             fiz2             = _mm256_add_pd(fiz2,tz);
2328
2329             fjx3             = _mm256_add_pd(fjx3,tx);
2330             fjy3             = _mm256_add_pd(fjy3,ty);
2331             fjz3             = _mm256_add_pd(fjz3,tz);
2332
2333             }
2334
2335             /**************************
2336              * CALCULATE INTERACTIONS *
2337              **************************/
2338
2339             if (gmx_mm256_any_lt(rsq31,rcutoff2))
2340             {
2341
2342             /* REACTION-FIELD ELECTROSTATICS */
2343             felec            = _mm256_mul_pd(qq31,_mm256_sub_pd(_mm256_mul_pd(rinv31,rinvsq31),krf2));
2344
2345             cutoff_mask      = _mm256_cmp_pd(rsq31,rcutoff2,_CMP_LT_OQ);
2346
2347             fscal            = felec;
2348
2349             fscal            = _mm256_and_pd(fscal,cutoff_mask);
2350
2351             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
2352
2353             /* Calculate temporary vectorial force */
2354             tx               = _mm256_mul_pd(fscal,dx31);
2355             ty               = _mm256_mul_pd(fscal,dy31);
2356             tz               = _mm256_mul_pd(fscal,dz31);
2357
2358             /* Update vectorial force */
2359             fix3             = _mm256_add_pd(fix3,tx);
2360             fiy3             = _mm256_add_pd(fiy3,ty);
2361             fiz3             = _mm256_add_pd(fiz3,tz);
2362
2363             fjx1             = _mm256_add_pd(fjx1,tx);
2364             fjy1             = _mm256_add_pd(fjy1,ty);
2365             fjz1             = _mm256_add_pd(fjz1,tz);
2366
2367             }
2368
2369             /**************************
2370              * CALCULATE INTERACTIONS *
2371              **************************/
2372
2373             if (gmx_mm256_any_lt(rsq32,rcutoff2))
2374             {
2375
2376             /* REACTION-FIELD ELECTROSTATICS */
2377             felec            = _mm256_mul_pd(qq32,_mm256_sub_pd(_mm256_mul_pd(rinv32,rinvsq32),krf2));
2378
2379             cutoff_mask      = _mm256_cmp_pd(rsq32,rcutoff2,_CMP_LT_OQ);
2380
2381             fscal            = felec;
2382
2383             fscal            = _mm256_and_pd(fscal,cutoff_mask);
2384
2385             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
2386
2387             /* Calculate temporary vectorial force */
2388             tx               = _mm256_mul_pd(fscal,dx32);
2389             ty               = _mm256_mul_pd(fscal,dy32);
2390             tz               = _mm256_mul_pd(fscal,dz32);
2391
2392             /* Update vectorial force */
2393             fix3             = _mm256_add_pd(fix3,tx);
2394             fiy3             = _mm256_add_pd(fiy3,ty);
2395             fiz3             = _mm256_add_pd(fiz3,tz);
2396
2397             fjx2             = _mm256_add_pd(fjx2,tx);
2398             fjy2             = _mm256_add_pd(fjy2,ty);
2399             fjz2             = _mm256_add_pd(fjz2,tz);
2400
2401             }
2402
2403             /**************************
2404              * CALCULATE INTERACTIONS *
2405              **************************/
2406
2407             if (gmx_mm256_any_lt(rsq33,rcutoff2))
2408             {
2409
2410             /* REACTION-FIELD ELECTROSTATICS */
2411             felec            = _mm256_mul_pd(qq33,_mm256_sub_pd(_mm256_mul_pd(rinv33,rinvsq33),krf2));
2412
2413             cutoff_mask      = _mm256_cmp_pd(rsq33,rcutoff2,_CMP_LT_OQ);
2414
2415             fscal            = felec;
2416
2417             fscal            = _mm256_and_pd(fscal,cutoff_mask);
2418
2419             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
2420
2421             /* Calculate temporary vectorial force */
2422             tx               = _mm256_mul_pd(fscal,dx33);
2423             ty               = _mm256_mul_pd(fscal,dy33);
2424             tz               = _mm256_mul_pd(fscal,dz33);
2425
2426             /* Update vectorial force */
2427             fix3             = _mm256_add_pd(fix3,tx);
2428             fiy3             = _mm256_add_pd(fiy3,ty);
2429             fiz3             = _mm256_add_pd(fiz3,tz);
2430
2431             fjx3             = _mm256_add_pd(fjx3,tx);
2432             fjy3             = _mm256_add_pd(fjy3,ty);
2433             fjz3             = _mm256_add_pd(fjz3,tz);
2434
2435             }
2436
2437             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
2438             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
2439             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
2440             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
2441
2442             gmx_mm256_decrement_4rvec_4ptr_swizzle_pd(fjptrA,fjptrB,fjptrC,fjptrD,
2443                                                       fjx0,fjy0,fjz0,fjx1,fjy1,fjz1,
2444                                                       fjx2,fjy2,fjz2,fjx3,fjy3,fjz3);
2445
2446             /* Inner loop uses 325 flops */
2447         }
2448
2449         /* End of innermost loop */
2450
2451         gmx_mm256_update_iforce_4atom_swizzle_pd(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,fix3,fiy3,fiz3,
2452                                                  f+i_coord_offset,fshift+i_shift_offset);
2453
2454         /* Increment number of inner iterations */
2455         inneriter                  += j_index_end - j_index_start;
2456
2457         /* Outer loop uses 24 flops */
2458     }
2459
2460     /* Increment number of outer iterations */
2461     outeriter        += nri;
2462
2463     /* Update outer/inner flops */
2464
2465     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W4W4_F,outeriter*24 + inneriter*325);
2466 }