src/gromacs/gmxlib/nonbonded/nb_kernel_avx_256_double/nb_kernel_ElecRFCut_VdwLJSh_GeomW4W4_avx_256_double.c

   1 /*
   2  * This file is part of the GROMACS molecular simulation package.
   3  *
   4  * Copyright (c) 2012,2013,2014, by the GROMACS development team, led by
   5  * Mark Abraham, David van der Spoel, Berk Hess, and Erik Lindahl,
   6  * and including many others, as listed in the AUTHORS file in the
   7  * top-level source directory and at http://www.gromacs.org.
   8  *
   9  * GROMACS is free software; you can redistribute it and/or
  10  * modify it under the terms of the GNU Lesser General Public License
  11  * as published by the Free Software Foundation; either version 2.1
  12  * of the License, or (at your option) any later version.
  13  *
  14  * GROMACS is distributed in the hope that it will be useful,
  15  * but WITHOUT ANY WARRANTY; without even the implied warranty of
  16  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
  17  * Lesser General Public License for more details.
  18  *
  19  * You should have received a copy of the GNU Lesser General Public
  20  * License along with GROMACS; if not, see
  21  * http://www.gnu.org/licenses, or write to the Free Software Foundation,
  22  * Inc., 51 Franklin Street, Fifth Floor, Boston, MA  02110-1301  USA.
  23  *
  24  * If you want to redistribute modifications to GROMACS, please
  25  * consider that scientific software is very special. Version
  26  * control is crucial - bugs must be traceable. We will be happy to
  27  * consider code for inclusion in the official distribution, but
  28  * derived work must not be called official GROMACS. Details are found
  29  * in the README & COPYING files - if they are missing, get the
  30  * official version at http://www.gromacs.org.
  31  *
  32  * To help us fund GROMACS development, we humbly ask that you cite
  33  * the research papers on the package. Check out http://www.gromacs.org.
  34  */
  35 /*
  36  * Note: this file was generated by the GROMACS avx_256_double kernel generator.
  37  */
  38 #include "config.h"
  39
  40 #include <math.h>
  41
  42 #include "../nb_kernel.h"
  43 #include "types/simple.h"
  44 #include "gromacs/math/vec.h"
  45 #include "nrnb.h"
  46
  47 #include "gromacs/simd/math_x86_avx_256_double.h"
  48 #include "kernelutil_x86_avx_256_double.h"
  49
  50 /*
  51  * Gromacs nonbonded kernel:   nb_kernel_ElecRFCut_VdwLJSh_GeomW4W4_VF_avx_256_double
  52  * Electrostatics interaction: ReactionField
  53  * VdW interaction:            LennardJones
  54  * Geometry:                   Water4-Water4
  55  * Calculate force/pot:        PotentialAndForce
  56  */
  57 void
  58 nb_kernel_ElecRFCut_VdwLJSh_GeomW4W4_VF_avx_256_double
  59                     (t_nblist                    * gmx_restrict       nlist,
  60                      rvec                        * gmx_restrict          xx,
  61                      rvec                        * gmx_restrict          ff,
  62                      t_forcerec                  * gmx_restrict          fr,
  63                      t_mdatoms                   * gmx_restrict     mdatoms,
  64                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
  65                      t_nrnb                      * gmx_restrict        nrnb)
  66 {
  67     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
  68      * just 0 for non-waters.
  69      * Suffixes A,B,C,D refer to j loop unrolling done with AVX, e.g. for the four different
  70      * jnr indices corresponding to data put in the four positions in the SIMD register.
  71      */
  72     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
  73     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
  74     int              jnrA,jnrB,jnrC,jnrD;
  75     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
  76     int              jnrlistE,jnrlistF,jnrlistG,jnrlistH;
  77     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
  78     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
  79     real             rcutoff_scalar;
  80     real             *shiftvec,*fshift,*x,*f;
  81     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD;
  82     real             scratch[4*DIM];
  83     __m256d          tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
  84     real *           vdwioffsetptr0;
  85     __m256d          ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
  86     real *           vdwioffsetptr1;
  87     __m256d          ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
  88     real *           vdwioffsetptr2;
  89     __m256d          ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
  90     real *           vdwioffsetptr3;
  91     __m256d          ix3,iy3,iz3,fix3,fiy3,fiz3,iq3,isai3;
  92     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
  93     __m256d          jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
  94     int              vdwjidx1A,vdwjidx1B,vdwjidx1C,vdwjidx1D;
  95     __m256d          jx1,jy1,jz1,fjx1,fjy1,fjz1,jq1,isaj1;
  96     int              vdwjidx2A,vdwjidx2B,vdwjidx2C,vdwjidx2D;
  97     __m256d          jx2,jy2,jz2,fjx2,fjy2,fjz2,jq2,isaj2;
  98     int              vdwjidx3A,vdwjidx3B,vdwjidx3C,vdwjidx3D;
  99     __m256d          jx3,jy3,jz3,fjx3,fjy3,fjz3,jq3,isaj3;
 100     __m256d          dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
 101     __m256d          dx11,dy11,dz11,rsq11,rinv11,rinvsq11,r11,qq11,c6_11,c12_11;
 102     __m256d          dx12,dy12,dz12,rsq12,rinv12,rinvsq12,r12,qq12,c6_12,c12_12;
 103     __m256d          dx13,dy13,dz13,rsq13,rinv13,rinvsq13,r13,qq13,c6_13,c12_13;
 104     __m256d          dx21,dy21,dz21,rsq21,rinv21,rinvsq21,r21,qq21,c6_21,c12_21;
 105     __m256d          dx22,dy22,dz22,rsq22,rinv22,rinvsq22,r22,qq22,c6_22,c12_22;
 106     __m256d          dx23,dy23,dz23,rsq23,rinv23,rinvsq23,r23,qq23,c6_23,c12_23;
 107     __m256d          dx31,dy31,dz31,rsq31,rinv31,rinvsq31,r31,qq31,c6_31,c12_31;
 108     __m256d          dx32,dy32,dz32,rsq32,rinv32,rinvsq32,r32,qq32,c6_32,c12_32;
 109     __m256d          dx33,dy33,dz33,rsq33,rinv33,rinvsq33,r33,qq33,c6_33,c12_33;
 110     __m256d          velec,felec,velecsum,facel,crf,krf,krf2;
 111     real             *charge;
 112     int              nvdwtype;
 113     __m256d          rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
 114     int              *vdwtype;
 115     real             *vdwparam;
 116     __m256d          one_sixth   = _mm256_set1_pd(1.0/6.0);
 117     __m256d          one_twelfth = _mm256_set1_pd(1.0/12.0);
 118     __m256d          dummy_mask,cutoff_mask;
 119     __m128           tmpmask0,tmpmask1;
 120     __m256d          signbit = _mm256_castsi256_pd( _mm256_set1_epi32(0x80000000) );
 121     __m256d          one     = _mm256_set1_pd(1.0);
 122     __m256d          two     = _mm256_set1_pd(2.0);
 123     x                = xx[0];
 124     f                = ff[0];
 125
 126     nri              = nlist->nri;
 127     iinr             = nlist->iinr;
 128     jindex           = nlist->jindex;
 129     jjnr             = nlist->jjnr;
 130     shiftidx         = nlist->shift;
 131     gid              = nlist->gid;
 132     shiftvec         = fr->shift_vec[0];
 133     fshift           = fr->fshift[0];
 134     facel            = _mm256_set1_pd(fr->epsfac);
 135     charge           = mdatoms->chargeA;
 136     krf              = _mm256_set1_pd(fr->ic->k_rf);
 137     krf2             = _mm256_set1_pd(fr->ic->k_rf*2.0);
 138     crf              = _mm256_set1_pd(fr->ic->c_rf);
 139     nvdwtype         = fr->ntype;
 140     vdwparam         = fr->nbfp;
 141     vdwtype          = mdatoms->typeA;
 142
 143     /* Setup water-specific parameters */
 144     inr              = nlist->iinr[0];
 145     iq1              = _mm256_mul_pd(facel,_mm256_set1_pd(charge[inr+1]));
 146     iq2              = _mm256_mul_pd(facel,_mm256_set1_pd(charge[inr+2]));
 147     iq3              = _mm256_mul_pd(facel,_mm256_set1_pd(charge[inr+3]));
 148     vdwioffsetptr0   = vdwparam+2*nvdwtype*vdwtype[inr+0];
 149
 150     jq1              = _mm256_set1_pd(charge[inr+1]);
 151     jq2              = _mm256_set1_pd(charge[inr+2]);
 152     jq3              = _mm256_set1_pd(charge[inr+3]);
 153     vdwjidx0A        = 2*vdwtype[inr+0];
 154     c6_00            = _mm256_set1_pd(vdwioffsetptr0[vdwjidx0A]);
 155     c12_00           = _mm256_set1_pd(vdwioffsetptr0[vdwjidx0A+1]);
 156     qq11             = _mm256_mul_pd(iq1,jq1);
 157     qq12             = _mm256_mul_pd(iq1,jq2);
 158     qq13             = _mm256_mul_pd(iq1,jq3);
 159     qq21             = _mm256_mul_pd(iq2,jq1);
 160     qq22             = _mm256_mul_pd(iq2,jq2);
 161     qq23             = _mm256_mul_pd(iq2,jq3);
 162     qq31             = _mm256_mul_pd(iq3,jq1);
 163     qq32             = _mm256_mul_pd(iq3,jq2);
 164     qq33             = _mm256_mul_pd(iq3,jq3);
 165
 166     /* When we use explicit cutoffs the value must be identical for elec and VdW, so use elec as an arbitrary choice */
 167     rcutoff_scalar   = fr->rcoulomb;
 168     rcutoff          = _mm256_set1_pd(rcutoff_scalar);
 169     rcutoff2         = _mm256_mul_pd(rcutoff,rcutoff);
 170
 171     sh_vdw_invrcut6  = _mm256_set1_pd(fr->ic->sh_invrc6);
 172     rvdw             = _mm256_set1_pd(fr->rvdw);
 173
 174     /* Avoid stupid compiler warnings */
 175     jnrA = jnrB = jnrC = jnrD = 0;
 176     j_coord_offsetA = 0;
 177     j_coord_offsetB = 0;
 178     j_coord_offsetC = 0;
 179     j_coord_offsetD = 0;
 180
 181     outeriter        = 0;
 182     inneriter        = 0;
 183
 184     for(iidx=0;iidx<4*DIM;iidx++)
 185     {
 186         scratch[iidx] = 0.0;
 187     }
 188
 189     /* Start outer loop over neighborlists */
 190     for(iidx=0; iidx<nri; iidx++)
 191     {
 192         /* Load shift vector for this list */
 193         i_shift_offset   = DIM*shiftidx[iidx];
 194
 195         /* Load limits for loop over neighbors */
 196         j_index_start    = jindex[iidx];
 197         j_index_end      = jindex[iidx+1];
 198
 199         /* Get outer coordinate index */
 200         inr              = iinr[iidx];
 201         i_coord_offset   = DIM*inr;
 202
 203         /* Load i particle coords and add shift vector */
 204         gmx_mm256_load_shift_and_4rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset,
 205                                                     &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2,&ix3,&iy3,&iz3);
 206
 207         fix0             = _mm256_setzero_pd();
 208         fiy0             = _mm256_setzero_pd();
 209         fiz0             = _mm256_setzero_pd();
 210         fix1             = _mm256_setzero_pd();
 211         fiy1             = _mm256_setzero_pd();
 212         fiz1             = _mm256_setzero_pd();
 213         fix2             = _mm256_setzero_pd();
 214         fiy2             = _mm256_setzero_pd();
 215         fiz2             = _mm256_setzero_pd();
 216         fix3             = _mm256_setzero_pd();
 217         fiy3             = _mm256_setzero_pd();
 218         fiz3             = _mm256_setzero_pd();
 219
 220         /* Reset potential sums */
 221         velecsum         = _mm256_setzero_pd();
 222         vvdwsum          = _mm256_setzero_pd();
 223
 224         /* Start inner kernel loop */
 225         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
 226         {
 227
 228             /* Get j neighbor index, and coordinate index */
 229             jnrA             = jjnr[jidx];
 230             jnrB             = jjnr[jidx+1];
 231             jnrC             = jjnr[jidx+2];
 232             jnrD             = jjnr[jidx+3];
 233             j_coord_offsetA  = DIM*jnrA;
 234             j_coord_offsetB  = DIM*jnrB;
 235             j_coord_offsetC  = DIM*jnrC;
 236             j_coord_offsetD  = DIM*jnrD;
 237
 238             /* load j atom coordinates */
 239             gmx_mm256_load_4rvec_4ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 240                                                  x+j_coord_offsetC,x+j_coord_offsetD,
 241                                                  &jx0,&jy0,&jz0,&jx1,&jy1,&jz1,&jx2,
 242                                                  &jy2,&jz2,&jx3,&jy3,&jz3);
 243
 244             /* Calculate displacement vector */
 245             dx00             = _mm256_sub_pd(ix0,jx0);
 246             dy00             = _mm256_sub_pd(iy0,jy0);
 247             dz00             = _mm256_sub_pd(iz0,jz0);
 248             dx11             = _mm256_sub_pd(ix1,jx1);
 249             dy11             = _mm256_sub_pd(iy1,jy1);
 250             dz11             = _mm256_sub_pd(iz1,jz1);
 251             dx12             = _mm256_sub_pd(ix1,jx2);
 252             dy12             = _mm256_sub_pd(iy1,jy2);
 253             dz12             = _mm256_sub_pd(iz1,jz2);
 254             dx13             = _mm256_sub_pd(ix1,jx3);
 255             dy13             = _mm256_sub_pd(iy1,jy3);
 256             dz13             = _mm256_sub_pd(iz1,jz3);
 257             dx21             = _mm256_sub_pd(ix2,jx1);
 258             dy21             = _mm256_sub_pd(iy2,jy1);
 259             dz21             = _mm256_sub_pd(iz2,jz1);
 260             dx22             = _mm256_sub_pd(ix2,jx2);
 261             dy22             = _mm256_sub_pd(iy2,jy2);
 262             dz22             = _mm256_sub_pd(iz2,jz2);
 263             dx23             = _mm256_sub_pd(ix2,jx3);
 264             dy23             = _mm256_sub_pd(iy2,jy3);
 265             dz23             = _mm256_sub_pd(iz2,jz3);
 266             dx31             = _mm256_sub_pd(ix3,jx1);
 267             dy31             = _mm256_sub_pd(iy3,jy1);
 268             dz31             = _mm256_sub_pd(iz3,jz1);
 269             dx32             = _mm256_sub_pd(ix3,jx2);
 270             dy32             = _mm256_sub_pd(iy3,jy2);
 271             dz32             = _mm256_sub_pd(iz3,jz2);
 272             dx33             = _mm256_sub_pd(ix3,jx3);
 273             dy33             = _mm256_sub_pd(iy3,jy3);
 274             dz33             = _mm256_sub_pd(iz3,jz3);
 275
 276             /* Calculate squared distance and things based on it */
 277             rsq00            = gmx_mm256_calc_rsq_pd(dx00,dy00,dz00);
 278             rsq11            = gmx_mm256_calc_rsq_pd(dx11,dy11,dz11);
 279             rsq12            = gmx_mm256_calc_rsq_pd(dx12,dy12,dz12);
 280             rsq13            = gmx_mm256_calc_rsq_pd(dx13,dy13,dz13);
 281             rsq21            = gmx_mm256_calc_rsq_pd(dx21,dy21,dz21);
 282             rsq22            = gmx_mm256_calc_rsq_pd(dx22,dy22,dz22);
 283             rsq23            = gmx_mm256_calc_rsq_pd(dx23,dy23,dz23);
 284             rsq31            = gmx_mm256_calc_rsq_pd(dx31,dy31,dz31);
 285             rsq32            = gmx_mm256_calc_rsq_pd(dx32,dy32,dz32);
 286             rsq33            = gmx_mm256_calc_rsq_pd(dx33,dy33,dz33);
 287
 288             rinv11           = gmx_mm256_invsqrt_pd(rsq11);
 289             rinv12           = gmx_mm256_invsqrt_pd(rsq12);
 290             rinv13           = gmx_mm256_invsqrt_pd(rsq13);
 291             rinv21           = gmx_mm256_invsqrt_pd(rsq21);
 292             rinv22           = gmx_mm256_invsqrt_pd(rsq22);
 293             rinv23           = gmx_mm256_invsqrt_pd(rsq23);
 294             rinv31           = gmx_mm256_invsqrt_pd(rsq31);
 295             rinv32           = gmx_mm256_invsqrt_pd(rsq32);
 296             rinv33           = gmx_mm256_invsqrt_pd(rsq33);
 297
 298             rinvsq00         = gmx_mm256_inv_pd(rsq00);
 299             rinvsq11         = _mm256_mul_pd(rinv11,rinv11);
 300             rinvsq12         = _mm256_mul_pd(rinv12,rinv12);
 301             rinvsq13         = _mm256_mul_pd(rinv13,rinv13);
 302             rinvsq21         = _mm256_mul_pd(rinv21,rinv21);
 303             rinvsq22         = _mm256_mul_pd(rinv22,rinv22);
 304             rinvsq23         = _mm256_mul_pd(rinv23,rinv23);
 305             rinvsq31         = _mm256_mul_pd(rinv31,rinv31);
 306             rinvsq32         = _mm256_mul_pd(rinv32,rinv32);
 307             rinvsq33         = _mm256_mul_pd(rinv33,rinv33);
 308
 309             fjx0             = _mm256_setzero_pd();
 310             fjy0             = _mm256_setzero_pd();
 311             fjz0             = _mm256_setzero_pd();
 312             fjx1             = _mm256_setzero_pd();
 313             fjy1             = _mm256_setzero_pd();
 314             fjz1             = _mm256_setzero_pd();
 315             fjx2             = _mm256_setzero_pd();
 316             fjy2             = _mm256_setzero_pd();
 317             fjz2             = _mm256_setzero_pd();
 318             fjx3             = _mm256_setzero_pd();
 319             fjy3             = _mm256_setzero_pd();
 320             fjz3             = _mm256_setzero_pd();
 321
 322             /**************************
 323              * CALCULATE INTERACTIONS *
 324              **************************/
 325
 326             if (gmx_mm256_any_lt(rsq00,rcutoff2))
 327             {
 328
 329             /* LENNARD-JONES DISPERSION/REPULSION */
 330
 331             rinvsix          = _mm256_mul_pd(_mm256_mul_pd(rinvsq00,rinvsq00),rinvsq00);
 332             vvdw6            = _mm256_mul_pd(c6_00,rinvsix);
 333             vvdw12           = _mm256_mul_pd(c12_00,_mm256_mul_pd(rinvsix,rinvsix));
 334             vvdw             = _mm256_sub_pd(_mm256_mul_pd( _mm256_sub_pd(vvdw12 , _mm256_mul_pd(c12_00,_mm256_mul_pd(sh_vdw_invrcut6,sh_vdw_invrcut6))), one_twelfth) ,
 335                                           _mm256_mul_pd( _mm256_sub_pd(vvdw6,_mm256_mul_pd(c6_00,sh_vdw_invrcut6)),one_sixth));
 336             fvdw             = _mm256_mul_pd(_mm256_sub_pd(vvdw12,vvdw6),rinvsq00);
 337
 338             cutoff_mask      = _mm256_cmp_pd(rsq00,rcutoff2,_CMP_LT_OQ);
 339
 340             /* Update potential sum for this i atom from the interaction with this j atom. */
 341             vvdw             = _mm256_and_pd(vvdw,cutoff_mask);
 342             vvdwsum          = _mm256_add_pd(vvdwsum,vvdw);
 343
 344             fscal            = fvdw;
 345
 346             fscal            = _mm256_and_pd(fscal,cutoff_mask);
 347
 348             /* Calculate temporary vectorial force */
 349             tx               = _mm256_mul_pd(fscal,dx00);
 350             ty               = _mm256_mul_pd(fscal,dy00);
 351             tz               = _mm256_mul_pd(fscal,dz00);
 352
 353             /* Update vectorial force */
 354             fix0             = _mm256_add_pd(fix0,tx);
 355             fiy0             = _mm256_add_pd(fiy0,ty);
 356             fiz0             = _mm256_add_pd(fiz0,tz);
 357
 358             fjx0             = _mm256_add_pd(fjx0,tx);
 359             fjy0             = _mm256_add_pd(fjy0,ty);
 360             fjz0             = _mm256_add_pd(fjz0,tz);
 361
 362             }
 363
 364             /**************************
 365              * CALCULATE INTERACTIONS *
 366              **************************/
 367
 368             if (gmx_mm256_any_lt(rsq11,rcutoff2))
 369             {
 370
 371             /* REACTION-FIELD ELECTROSTATICS */
 372             velec            = _mm256_mul_pd(qq11,_mm256_sub_pd(_mm256_add_pd(rinv11,_mm256_mul_pd(krf,rsq11)),crf));
 373             felec            = _mm256_mul_pd(qq11,_mm256_sub_pd(_mm256_mul_pd(rinv11,rinvsq11),krf2));
 374
 375             cutoff_mask      = _mm256_cmp_pd(rsq11,rcutoff2,_CMP_LT_OQ);
 376
 377             /* Update potential sum for this i atom from the interaction with this j atom. */
 378             velec            = _mm256_and_pd(velec,cutoff_mask);
 379             velecsum         = _mm256_add_pd(velecsum,velec);
 380
 381             fscal            = felec;
 382
 383             fscal            = _mm256_and_pd(fscal,cutoff_mask);
 384
 385             /* Calculate temporary vectorial force */
 386             tx               = _mm256_mul_pd(fscal,dx11);
 387             ty               = _mm256_mul_pd(fscal,dy11);
 388             tz               = _mm256_mul_pd(fscal,dz11);
 389
 390             /* Update vectorial force */
 391             fix1             = _mm256_add_pd(fix1,tx);
 392             fiy1             = _mm256_add_pd(fiy1,ty);
 393             fiz1             = _mm256_add_pd(fiz1,tz);
 394
 395             fjx1             = _mm256_add_pd(fjx1,tx);
 396             fjy1             = _mm256_add_pd(fjy1,ty);
 397             fjz1             = _mm256_add_pd(fjz1,tz);
 398
 399             }
 400
 401             /**************************
 402              * CALCULATE INTERACTIONS *
 403              **************************/
 404
 405             if (gmx_mm256_any_lt(rsq12,rcutoff2))
 406             {
 407
 408             /* REACTION-FIELD ELECTROSTATICS */
 409             velec            = _mm256_mul_pd(qq12,_mm256_sub_pd(_mm256_add_pd(rinv12,_mm256_mul_pd(krf,rsq12)),crf));
 410             felec            = _mm256_mul_pd(qq12,_mm256_sub_pd(_mm256_mul_pd(rinv12,rinvsq12),krf2));
 411
 412             cutoff_mask      = _mm256_cmp_pd(rsq12,rcutoff2,_CMP_LT_OQ);
 413
 414             /* Update potential sum for this i atom from the interaction with this j atom. */
 415             velec            = _mm256_and_pd(velec,cutoff_mask);
 416             velecsum         = _mm256_add_pd(velecsum,velec);
 417
 418             fscal            = felec;
 419
 420             fscal            = _mm256_and_pd(fscal,cutoff_mask);
 421
 422             /* Calculate temporary vectorial force */
 423             tx               = _mm256_mul_pd(fscal,dx12);
 424             ty               = _mm256_mul_pd(fscal,dy12);
 425             tz               = _mm256_mul_pd(fscal,dz12);
 426
 427             /* Update vectorial force */
 428             fix1             = _mm256_add_pd(fix1,tx);
 429             fiy1             = _mm256_add_pd(fiy1,ty);
 430             fiz1             = _mm256_add_pd(fiz1,tz);
 431
 432             fjx2             = _mm256_add_pd(fjx2,tx);
 433             fjy2             = _mm256_add_pd(fjy2,ty);
 434             fjz2             = _mm256_add_pd(fjz2,tz);
 435
 436             }
 437
 438             /**************************
 439              * CALCULATE INTERACTIONS *
 440              **************************/
 441
 442             if (gmx_mm256_any_lt(rsq13,rcutoff2))
 443             {
 444
 445             /* REACTION-FIELD ELECTROSTATICS */
 446             velec            = _mm256_mul_pd(qq13,_mm256_sub_pd(_mm256_add_pd(rinv13,_mm256_mul_pd(krf,rsq13)),crf));
 447             felec            = _mm256_mul_pd(qq13,_mm256_sub_pd(_mm256_mul_pd(rinv13,rinvsq13),krf2));
 448
 449             cutoff_mask      = _mm256_cmp_pd(rsq13,rcutoff2,_CMP_LT_OQ);
 450
 451             /* Update potential sum for this i atom from the interaction with this j atom. */
 452             velec            = _mm256_and_pd(velec,cutoff_mask);
 453             velecsum         = _mm256_add_pd(velecsum,velec);
 454
 455             fscal            = felec;
 456
 457             fscal            = _mm256_and_pd(fscal,cutoff_mask);
 458
 459             /* Calculate temporary vectorial force */
 460             tx               = _mm256_mul_pd(fscal,dx13);
 461             ty               = _mm256_mul_pd(fscal,dy13);
 462             tz               = _mm256_mul_pd(fscal,dz13);
 463
 464             /* Update vectorial force */
 465             fix1             = _mm256_add_pd(fix1,tx);
 466             fiy1             = _mm256_add_pd(fiy1,ty);
 467             fiz1             = _mm256_add_pd(fiz1,tz);
 468
 469             fjx3             = _mm256_add_pd(fjx3,tx);
 470             fjy3             = _mm256_add_pd(fjy3,ty);
 471             fjz3             = _mm256_add_pd(fjz3,tz);
 472
 473             }
 474
 475             /**************************
 476              * CALCULATE INTERACTIONS *
 477              **************************/
 478
 479             if (gmx_mm256_any_lt(rsq21,rcutoff2))
 480             {
 481
 482             /* REACTION-FIELD ELECTROSTATICS */
 483             velec            = _mm256_mul_pd(qq21,_mm256_sub_pd(_mm256_add_pd(rinv21,_mm256_mul_pd(krf,rsq21)),crf));
 484             felec            = _mm256_mul_pd(qq21,_mm256_sub_pd(_mm256_mul_pd(rinv21,rinvsq21),krf2));
 485
 486             cutoff_mask      = _mm256_cmp_pd(rsq21,rcutoff2,_CMP_LT_OQ);
 487
 488             /* Update potential sum for this i atom from the interaction with this j atom. */
 489             velec            = _mm256_and_pd(velec,cutoff_mask);
 490             velecsum         = _mm256_add_pd(velecsum,velec);
 491
 492             fscal            = felec;
 493
 494             fscal            = _mm256_and_pd(fscal,cutoff_mask);
 495
 496             /* Calculate temporary vectorial force */
 497             tx               = _mm256_mul_pd(fscal,dx21);
 498             ty               = _mm256_mul_pd(fscal,dy21);
 499             tz               = _mm256_mul_pd(fscal,dz21);
 500
 501             /* Update vectorial force */
 502             fix2             = _mm256_add_pd(fix2,tx);
 503             fiy2             = _mm256_add_pd(fiy2,ty);
 504             fiz2             = _mm256_add_pd(fiz2,tz);
 505
 506             fjx1             = _mm256_add_pd(fjx1,tx);
 507             fjy1             = _mm256_add_pd(fjy1,ty);
 508             fjz1             = _mm256_add_pd(fjz1,tz);
 509
 510             }
 511
 512             /**************************
 513              * CALCULATE INTERACTIONS *
 514              **************************/
 515
 516             if (gmx_mm256_any_lt(rsq22,rcutoff2))
 517             {
 518
 519             /* REACTION-FIELD ELECTROSTATICS */
 520             velec            = _mm256_mul_pd(qq22,_mm256_sub_pd(_mm256_add_pd(rinv22,_mm256_mul_pd(krf,rsq22)),crf));
 521             felec            = _mm256_mul_pd(qq22,_mm256_sub_pd(_mm256_mul_pd(rinv22,rinvsq22),krf2));
 522
 523             cutoff_mask      = _mm256_cmp_pd(rsq22,rcutoff2,_CMP_LT_OQ);
 524
 525             /* Update potential sum for this i atom from the interaction with this j atom. */
 526             velec            = _mm256_and_pd(velec,cutoff_mask);
 527             velecsum         = _mm256_add_pd(velecsum,velec);
 528
 529             fscal            = felec;
 530
 531             fscal            = _mm256_and_pd(fscal,cutoff_mask);
 532
 533             /* Calculate temporary vectorial force */
 534             tx               = _mm256_mul_pd(fscal,dx22);
 535             ty               = _mm256_mul_pd(fscal,dy22);
 536             tz               = _mm256_mul_pd(fscal,dz22);
 537
 538             /* Update vectorial force */
 539             fix2             = _mm256_add_pd(fix2,tx);
 540             fiy2             = _mm256_add_pd(fiy2,ty);
 541             fiz2             = _mm256_add_pd(fiz2,tz);
 542
 543             fjx2             = _mm256_add_pd(fjx2,tx);
 544             fjy2             = _mm256_add_pd(fjy2,ty);
 545             fjz2             = _mm256_add_pd(fjz2,tz);
 546
 547             }
 548
 549             /**************************
 550              * CALCULATE INTERACTIONS *
 551              **************************/
 552
 553             if (gmx_mm256_any_lt(rsq23,rcutoff2))
 554             {
 555
 556             /* REACTION-FIELD ELECTROSTATICS */
 557             velec            = _mm256_mul_pd(qq23,_mm256_sub_pd(_mm256_add_pd(rinv23,_mm256_mul_pd(krf,rsq23)),crf));
 558             felec            = _mm256_mul_pd(qq23,_mm256_sub_pd(_mm256_mul_pd(rinv23,rinvsq23),krf2));
 559
 560             cutoff_mask      = _mm256_cmp_pd(rsq23,rcutoff2,_CMP_LT_OQ);
 561
 562             /* Update potential sum for this i atom from the interaction with this j atom. */
 563             velec            = _mm256_and_pd(velec,cutoff_mask);
 564             velecsum         = _mm256_add_pd(velecsum,velec);
 565
 566             fscal            = felec;
 567
 568             fscal            = _mm256_and_pd(fscal,cutoff_mask);
 569
 570             /* Calculate temporary vectorial force */
 571             tx               = _mm256_mul_pd(fscal,dx23);
 572             ty               = _mm256_mul_pd(fscal,dy23);
 573             tz               = _mm256_mul_pd(fscal,dz23);
 574
 575             /* Update vectorial force */
 576             fix2             = _mm256_add_pd(fix2,tx);
 577             fiy2             = _mm256_add_pd(fiy2,ty);
 578             fiz2             = _mm256_add_pd(fiz2,tz);
 579
 580             fjx3             = _mm256_add_pd(fjx3,tx);
 581             fjy3             = _mm256_add_pd(fjy3,ty);
 582             fjz3             = _mm256_add_pd(fjz3,tz);
 583
 584             }
 585
 586             /**************************
 587              * CALCULATE INTERACTIONS *
 588              **************************/
 589
 590             if (gmx_mm256_any_lt(rsq31,rcutoff2))
 591             {
 592
 593             /* REACTION-FIELD ELECTROSTATICS */
 594             velec            = _mm256_mul_pd(qq31,_mm256_sub_pd(_mm256_add_pd(rinv31,_mm256_mul_pd(krf,rsq31)),crf));
 595             felec            = _mm256_mul_pd(qq31,_mm256_sub_pd(_mm256_mul_pd(rinv31,rinvsq31),krf2));
 596
 597             cutoff_mask      = _mm256_cmp_pd(rsq31,rcutoff2,_CMP_LT_OQ);
 598
 599             /* Update potential sum for this i atom from the interaction with this j atom. */
 600             velec            = _mm256_and_pd(velec,cutoff_mask);
 601             velecsum         = _mm256_add_pd(velecsum,velec);
 602
 603             fscal            = felec;
 604
 605             fscal            = _mm256_and_pd(fscal,cutoff_mask);
 606
 607             /* Calculate temporary vectorial force */
 608             tx               = _mm256_mul_pd(fscal,dx31);
 609             ty               = _mm256_mul_pd(fscal,dy31);
 610             tz               = _mm256_mul_pd(fscal,dz31);
 611
 612             /* Update vectorial force */
 613             fix3             = _mm256_add_pd(fix3,tx);
 614             fiy3             = _mm256_add_pd(fiy3,ty);
 615             fiz3             = _mm256_add_pd(fiz3,tz);
 616
 617             fjx1             = _mm256_add_pd(fjx1,tx);
 618             fjy1             = _mm256_add_pd(fjy1,ty);
 619             fjz1             = _mm256_add_pd(fjz1,tz);
 620
 621             }
 622
 623             /**************************
 624              * CALCULATE INTERACTIONS *
 625              **************************/
 626
 627             if (gmx_mm256_any_lt(rsq32,rcutoff2))
 628             {
 629
 630             /* REACTION-FIELD ELECTROSTATICS */
 631             velec            = _mm256_mul_pd(qq32,_mm256_sub_pd(_mm256_add_pd(rinv32,_mm256_mul_pd(krf,rsq32)),crf));
 632             felec            = _mm256_mul_pd(qq32,_mm256_sub_pd(_mm256_mul_pd(rinv32,rinvsq32),krf2));
 633
 634             cutoff_mask      = _mm256_cmp_pd(rsq32,rcutoff2,_CMP_LT_OQ);
 635
 636             /* Update potential sum for this i atom from the interaction with this j atom. */
 637             velec            = _mm256_and_pd(velec,cutoff_mask);
 638             velecsum         = _mm256_add_pd(velecsum,velec);
 639
 640             fscal            = felec;
 641
 642             fscal            = _mm256_and_pd(fscal,cutoff_mask);
 643
 644             /* Calculate temporary vectorial force */
 645             tx               = _mm256_mul_pd(fscal,dx32);
 646             ty               = _mm256_mul_pd(fscal,dy32);
 647             tz               = _mm256_mul_pd(fscal,dz32);
 648
 649             /* Update vectorial force */
 650             fix3             = _mm256_add_pd(fix3,tx);
 651             fiy3             = _mm256_add_pd(fiy3,ty);
 652             fiz3             = _mm256_add_pd(fiz3,tz);
 653
 654             fjx2             = _mm256_add_pd(fjx2,tx);
 655             fjy2             = _mm256_add_pd(fjy2,ty);
 656             fjz2             = _mm256_add_pd(fjz2,tz);
 657
 658             }
 659
 660             /**************************
 661              * CALCULATE INTERACTIONS *
 662              **************************/
 663
 664             if (gmx_mm256_any_lt(rsq33,rcutoff2))
 665             {
 666
 667             /* REACTION-FIELD ELECTROSTATICS */
 668             velec            = _mm256_mul_pd(qq33,_mm256_sub_pd(_mm256_add_pd(rinv33,_mm256_mul_pd(krf,rsq33)),crf));
 669             felec            = _mm256_mul_pd(qq33,_mm256_sub_pd(_mm256_mul_pd(rinv33,rinvsq33),krf2));
 670
 671             cutoff_mask      = _mm256_cmp_pd(rsq33,rcutoff2,_CMP_LT_OQ);
 672
 673             /* Update potential sum for this i atom from the interaction with this j atom. */
 674             velec            = _mm256_and_pd(velec,cutoff_mask);
 675             velecsum         = _mm256_add_pd(velecsum,velec);
 676
 677             fscal            = felec;
 678
 679             fscal            = _mm256_and_pd(fscal,cutoff_mask);
 680
 681             /* Calculate temporary vectorial force */
 682             tx               = _mm256_mul_pd(fscal,dx33);
 683             ty               = _mm256_mul_pd(fscal,dy33);
 684             tz               = _mm256_mul_pd(fscal,dz33);
 685
 686             /* Update vectorial force */
 687             fix3             = _mm256_add_pd(fix3,tx);
 688             fiy3             = _mm256_add_pd(fiy3,ty);
 689             fiz3             = _mm256_add_pd(fiz3,tz);
 690
 691             fjx3             = _mm256_add_pd(fjx3,tx);
 692             fjy3             = _mm256_add_pd(fjy3,ty);
 693             fjz3             = _mm256_add_pd(fjz3,tz);
 694
 695             }
 696
 697             fjptrA             = f+j_coord_offsetA;
 698             fjptrB             = f+j_coord_offsetB;
 699             fjptrC             = f+j_coord_offsetC;
 700             fjptrD             = f+j_coord_offsetD;
 701
 702             gmx_mm256_decrement_4rvec_4ptr_swizzle_pd(fjptrA,fjptrB,fjptrC,fjptrD,
 703                                                       fjx0,fjy0,fjz0,fjx1,fjy1,fjz1,
 704                                                       fjx2,fjy2,fjz2,fjx3,fjy3,fjz3);
 705
 706             /* Inner loop uses 368 flops */
 707         }
 708
 709         if(jidx<j_index_end)
 710         {
 711
 712             /* Get j neighbor index, and coordinate index */
 713             jnrlistA         = jjnr[jidx];
 714             jnrlistB         = jjnr[jidx+1];
 715             jnrlistC         = jjnr[jidx+2];
 716             jnrlistD         = jjnr[jidx+3];
 717             /* Sign of each element will be negative for non-real atoms.
 718              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 719              * so use it as val = _mm_andnot_pd(mask,val) to clear dummy entries.
 720              */
 721             tmpmask0 = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
 722
 723             tmpmask1 = _mm_permute_ps(tmpmask0,_GMX_MM_PERMUTE(3,3,2,2));
 724             tmpmask0 = _mm_permute_ps(tmpmask0,_GMX_MM_PERMUTE(1,1,0,0));
 725             dummy_mask = _mm256_castps_pd(gmx_mm256_set_m128(tmpmask1,tmpmask0));
 726
 727             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
 728             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
 729             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
 730             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
 731             j_coord_offsetA  = DIM*jnrA;
 732             j_coord_offsetB  = DIM*jnrB;
 733             j_coord_offsetC  = DIM*jnrC;
 734             j_coord_offsetD  = DIM*jnrD;
 735
 736             /* load j atom coordinates */
 737             gmx_mm256_load_4rvec_4ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 738                                                  x+j_coord_offsetC,x+j_coord_offsetD,
 739                                                  &jx0,&jy0,&jz0,&jx1,&jy1,&jz1,&jx2,
 740                                                  &jy2,&jz2,&jx3,&jy3,&jz3);
 741
 742             /* Calculate displacement vector */
 743             dx00             = _mm256_sub_pd(ix0,jx0);
 744             dy00             = _mm256_sub_pd(iy0,jy0);
 745             dz00             = _mm256_sub_pd(iz0,jz0);
 746             dx11             = _mm256_sub_pd(ix1,jx1);
 747             dy11             = _mm256_sub_pd(iy1,jy1);
 748             dz11             = _mm256_sub_pd(iz1,jz1);
 749             dx12             = _mm256_sub_pd(ix1,jx2);
 750             dy12             = _mm256_sub_pd(iy1,jy2);
 751             dz12             = _mm256_sub_pd(iz1,jz2);
 752             dx13             = _mm256_sub_pd(ix1,jx3);
 753             dy13             = _mm256_sub_pd(iy1,jy3);
 754             dz13             = _mm256_sub_pd(iz1,jz3);
 755             dx21             = _mm256_sub_pd(ix2,jx1);
 756             dy21             = _mm256_sub_pd(iy2,jy1);
 757             dz21             = _mm256_sub_pd(iz2,jz1);
 758             dx22             = _mm256_sub_pd(ix2,jx2);
 759             dy22             = _mm256_sub_pd(iy2,jy2);
 760             dz22             = _mm256_sub_pd(iz2,jz2);
 761             dx23             = _mm256_sub_pd(ix2,jx3);
 762             dy23             = _mm256_sub_pd(iy2,jy3);
 763             dz23             = _mm256_sub_pd(iz2,jz3);
 764             dx31             = _mm256_sub_pd(ix3,jx1);
 765             dy31             = _mm256_sub_pd(iy3,jy1);
 766             dz31             = _mm256_sub_pd(iz3,jz1);
 767             dx32             = _mm256_sub_pd(ix3,jx2);
 768             dy32             = _mm256_sub_pd(iy3,jy2);
 769             dz32             = _mm256_sub_pd(iz3,jz2);
 770             dx33             = _mm256_sub_pd(ix3,jx3);
 771             dy33             = _mm256_sub_pd(iy3,jy3);
 772             dz33             = _mm256_sub_pd(iz3,jz3);
 773
 774             /* Calculate squared distance and things based on it */
 775             rsq00            = gmx_mm256_calc_rsq_pd(dx00,dy00,dz00);
 776             rsq11            = gmx_mm256_calc_rsq_pd(dx11,dy11,dz11);
 777             rsq12            = gmx_mm256_calc_rsq_pd(dx12,dy12,dz12);
 778             rsq13            = gmx_mm256_calc_rsq_pd(dx13,dy13,dz13);
 779             rsq21            = gmx_mm256_calc_rsq_pd(dx21,dy21,dz21);
 780             rsq22            = gmx_mm256_calc_rsq_pd(dx22,dy22,dz22);
 781             rsq23            = gmx_mm256_calc_rsq_pd(dx23,dy23,dz23);
 782             rsq31            = gmx_mm256_calc_rsq_pd(dx31,dy31,dz31);
 783             rsq32            = gmx_mm256_calc_rsq_pd(dx32,dy32,dz32);
 784             rsq33            = gmx_mm256_calc_rsq_pd(dx33,dy33,dz33);
 785
 786             rinv11           = gmx_mm256_invsqrt_pd(rsq11);
 787             rinv12           = gmx_mm256_invsqrt_pd(rsq12);
 788             rinv13           = gmx_mm256_invsqrt_pd(rsq13);
 789             rinv21           = gmx_mm256_invsqrt_pd(rsq21);
 790             rinv22           = gmx_mm256_invsqrt_pd(rsq22);
 791             rinv23           = gmx_mm256_invsqrt_pd(rsq23);
 792             rinv31           = gmx_mm256_invsqrt_pd(rsq31);
 793             rinv32           = gmx_mm256_invsqrt_pd(rsq32);
 794             rinv33           = gmx_mm256_invsqrt_pd(rsq33);
 795
 796             rinvsq00         = gmx_mm256_inv_pd(rsq00);
 797             rinvsq11         = _mm256_mul_pd(rinv11,rinv11);
 798             rinvsq12         = _mm256_mul_pd(rinv12,rinv12);
 799             rinvsq13         = _mm256_mul_pd(rinv13,rinv13);
 800             rinvsq21         = _mm256_mul_pd(rinv21,rinv21);
 801             rinvsq22         = _mm256_mul_pd(rinv22,rinv22);
 802             rinvsq23         = _mm256_mul_pd(rinv23,rinv23);
 803             rinvsq31         = _mm256_mul_pd(rinv31,rinv31);
 804             rinvsq32         = _mm256_mul_pd(rinv32,rinv32);
 805             rinvsq33         = _mm256_mul_pd(rinv33,rinv33);
 806
 807             fjx0             = _mm256_setzero_pd();
 808             fjy0             = _mm256_setzero_pd();
 809             fjz0             = _mm256_setzero_pd();
 810             fjx1             = _mm256_setzero_pd();
 811             fjy1             = _mm256_setzero_pd();
 812             fjz1             = _mm256_setzero_pd();
 813             fjx2             = _mm256_setzero_pd();
 814             fjy2             = _mm256_setzero_pd();
 815             fjz2             = _mm256_setzero_pd();
 816             fjx3             = _mm256_setzero_pd();
 817             fjy3             = _mm256_setzero_pd();
 818             fjz3             = _mm256_setzero_pd();
 819
 820             /**************************
 821              * CALCULATE INTERACTIONS *
 822              **************************/
 823
 824             if (gmx_mm256_any_lt(rsq00,rcutoff2))
 825             {
 826
 827             /* LENNARD-JONES DISPERSION/REPULSION */
 828
 829             rinvsix          = _mm256_mul_pd(_mm256_mul_pd(rinvsq00,rinvsq00),rinvsq00);
 830             vvdw6            = _mm256_mul_pd(c6_00,rinvsix);
 831             vvdw12           = _mm256_mul_pd(c12_00,_mm256_mul_pd(rinvsix,rinvsix));
 832             vvdw             = _mm256_sub_pd(_mm256_mul_pd( _mm256_sub_pd(vvdw12 , _mm256_mul_pd(c12_00,_mm256_mul_pd(sh_vdw_invrcut6,sh_vdw_invrcut6))), one_twelfth) ,
 833                                           _mm256_mul_pd( _mm256_sub_pd(vvdw6,_mm256_mul_pd(c6_00,sh_vdw_invrcut6)),one_sixth));
 834             fvdw             = _mm256_mul_pd(_mm256_sub_pd(vvdw12,vvdw6),rinvsq00);
 835
 836             cutoff_mask      = _mm256_cmp_pd(rsq00,rcutoff2,_CMP_LT_OQ);
 837
 838             /* Update potential sum for this i atom from the interaction with this j atom. */
 839             vvdw             = _mm256_and_pd(vvdw,cutoff_mask);
 840             vvdw             = _mm256_andnot_pd(dummy_mask,vvdw);
 841             vvdwsum          = _mm256_add_pd(vvdwsum,vvdw);
 842
 843             fscal            = fvdw;
 844
 845             fscal            = _mm256_and_pd(fscal,cutoff_mask);
 846
 847             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
 848
 849             /* Calculate temporary vectorial force */
 850             tx               = _mm256_mul_pd(fscal,dx00);
 851             ty               = _mm256_mul_pd(fscal,dy00);
 852             tz               = _mm256_mul_pd(fscal,dz00);
 853
 854             /* Update vectorial force */
 855             fix0             = _mm256_add_pd(fix0,tx);
 856             fiy0             = _mm256_add_pd(fiy0,ty);
 857             fiz0             = _mm256_add_pd(fiz0,tz);
 858
 859             fjx0             = _mm256_add_pd(fjx0,tx);
 860             fjy0             = _mm256_add_pd(fjy0,ty);
 861             fjz0             = _mm256_add_pd(fjz0,tz);
 862
 863             }
 864
 865             /**************************
 866              * CALCULATE INTERACTIONS *
 867              **************************/
 868
 869             if (gmx_mm256_any_lt(rsq11,rcutoff2))
 870             {
 871
 872             /* REACTION-FIELD ELECTROSTATICS */
 873             velec            = _mm256_mul_pd(qq11,_mm256_sub_pd(_mm256_add_pd(rinv11,_mm256_mul_pd(krf,rsq11)),crf));
 874             felec            = _mm256_mul_pd(qq11,_mm256_sub_pd(_mm256_mul_pd(rinv11,rinvsq11),krf2));
 875
 876             cutoff_mask      = _mm256_cmp_pd(rsq11,rcutoff2,_CMP_LT_OQ);
 877
 878             /* Update potential sum for this i atom from the interaction with this j atom. */
 879             velec            = _mm256_and_pd(velec,cutoff_mask);
 880             velec            = _mm256_andnot_pd(dummy_mask,velec);
 881             velecsum         = _mm256_add_pd(velecsum,velec);
 882
 883             fscal            = felec;
 884
 885             fscal            = _mm256_and_pd(fscal,cutoff_mask);
 886
 887             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
 888
 889             /* Calculate temporary vectorial force */
 890             tx               = _mm256_mul_pd(fscal,dx11);
 891             ty               = _mm256_mul_pd(fscal,dy11);
 892             tz               = _mm256_mul_pd(fscal,dz11);
 893
 894             /* Update vectorial force */
 895             fix1             = _mm256_add_pd(fix1,tx);
 896             fiy1             = _mm256_add_pd(fiy1,ty);
 897             fiz1             = _mm256_add_pd(fiz1,tz);
 898
 899             fjx1             = _mm256_add_pd(fjx1,tx);
 900             fjy1             = _mm256_add_pd(fjy1,ty);
 901             fjz1             = _mm256_add_pd(fjz1,tz);
 902
 903             }
 904
 905             /**************************
 906              * CALCULATE INTERACTIONS *
 907              **************************/
 908
 909             if (gmx_mm256_any_lt(rsq12,rcutoff2))
 910             {
 911
 912             /* REACTION-FIELD ELECTROSTATICS */
 913             velec            = _mm256_mul_pd(qq12,_mm256_sub_pd(_mm256_add_pd(rinv12,_mm256_mul_pd(krf,rsq12)),crf));
 914             felec            = _mm256_mul_pd(qq12,_mm256_sub_pd(_mm256_mul_pd(rinv12,rinvsq12),krf2));
 915
 916             cutoff_mask      = _mm256_cmp_pd(rsq12,rcutoff2,_CMP_LT_OQ);
 917
 918             /* Update potential sum for this i atom from the interaction with this j atom. */
 919             velec            = _mm256_and_pd(velec,cutoff_mask);
 920             velec            = _mm256_andnot_pd(dummy_mask,velec);
 921             velecsum         = _mm256_add_pd(velecsum,velec);
 922
 923             fscal            = felec;
 924
 925             fscal            = _mm256_and_pd(fscal,cutoff_mask);
 926
 927             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
 928
 929             /* Calculate temporary vectorial force */
 930             tx               = _mm256_mul_pd(fscal,dx12);
 931             ty               = _mm256_mul_pd(fscal,dy12);
 932             tz               = _mm256_mul_pd(fscal,dz12);
 933
 934             /* Update vectorial force */
 935             fix1             = _mm256_add_pd(fix1,tx);
 936             fiy1             = _mm256_add_pd(fiy1,ty);
 937             fiz1             = _mm256_add_pd(fiz1,tz);
 938
 939             fjx2             = _mm256_add_pd(fjx2,tx);
 940             fjy2             = _mm256_add_pd(fjy2,ty);
 941             fjz2             = _mm256_add_pd(fjz2,tz);
 942
 943             }
 944
 945             /**************************
 946              * CALCULATE INTERACTIONS *
 947              **************************/
 948
 949             if (gmx_mm256_any_lt(rsq13,rcutoff2))
 950             {
 951
 952             /* REACTION-FIELD ELECTROSTATICS */
 953             velec            = _mm256_mul_pd(qq13,_mm256_sub_pd(_mm256_add_pd(rinv13,_mm256_mul_pd(krf,rsq13)),crf));
 954             felec            = _mm256_mul_pd(qq13,_mm256_sub_pd(_mm256_mul_pd(rinv13,rinvsq13),krf2));
 955
 956             cutoff_mask      = _mm256_cmp_pd(rsq13,rcutoff2,_CMP_LT_OQ);
 957
 958             /* Update potential sum for this i atom from the interaction with this j atom. */
 959             velec            = _mm256_and_pd(velec,cutoff_mask);
 960             velec            = _mm256_andnot_pd(dummy_mask,velec);
 961             velecsum         = _mm256_add_pd(velecsum,velec);
 962
 963             fscal            = felec;
 964
 965             fscal            = _mm256_and_pd(fscal,cutoff_mask);
 966
 967             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
 968
 969             /* Calculate temporary vectorial force */
 970             tx               = _mm256_mul_pd(fscal,dx13);
 971             ty               = _mm256_mul_pd(fscal,dy13);
 972             tz               = _mm256_mul_pd(fscal,dz13);
 973
 974             /* Update vectorial force */
 975             fix1             = _mm256_add_pd(fix1,tx);
 976             fiy1             = _mm256_add_pd(fiy1,ty);
 977             fiz1             = _mm256_add_pd(fiz1,tz);
 978
 979             fjx3             = _mm256_add_pd(fjx3,tx);
 980             fjy3             = _mm256_add_pd(fjy3,ty);
 981             fjz3             = _mm256_add_pd(fjz3,tz);
 982
 983             }
 984
 985             /**************************
 986              * CALCULATE INTERACTIONS *
 987              **************************/
 988
 989             if (gmx_mm256_any_lt(rsq21,rcutoff2))
 990             {
 991
 992             /* REACTION-FIELD ELECTROSTATICS */
 993             velec            = _mm256_mul_pd(qq21,_mm256_sub_pd(_mm256_add_pd(rinv21,_mm256_mul_pd(krf,rsq21)),crf));
 994             felec            = _mm256_mul_pd(qq21,_mm256_sub_pd(_mm256_mul_pd(rinv21,rinvsq21),krf2));
 995
 996             cutoff_mask      = _mm256_cmp_pd(rsq21,rcutoff2,_CMP_LT_OQ);
 997
 998             /* Update potential sum for this i atom from the interaction with this j atom. */
 999             velec            = _mm256_and_pd(velec,cutoff_mask);
1000             velec            = _mm256_andnot_pd(dummy_mask,velec);
1001             velecsum         = _mm256_add_pd(velecsum,velec);
1002
1003             fscal            = felec;
1004
1005             fscal            = _mm256_and_pd(fscal,cutoff_mask);
1006
1007             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
1008
1009             /* Calculate temporary vectorial force */
1010             tx               = _mm256_mul_pd(fscal,dx21);
1011             ty               = _mm256_mul_pd(fscal,dy21);
1012             tz               = _mm256_mul_pd(fscal,dz21);
1013
1014             /* Update vectorial force */
1015             fix2             = _mm256_add_pd(fix2,tx);
1016             fiy2             = _mm256_add_pd(fiy2,ty);
1017             fiz2             = _mm256_add_pd(fiz2,tz);
1018
1019             fjx1             = _mm256_add_pd(fjx1,tx);
1020             fjy1             = _mm256_add_pd(fjy1,ty);
1021             fjz1             = _mm256_add_pd(fjz1,tz);
1022
1023             }
1024
1025             /**************************
1026              * CALCULATE INTERACTIONS *
1027              **************************/
1028
1029             if (gmx_mm256_any_lt(rsq22,rcutoff2))
1030             {
1031
1032             /* REACTION-FIELD ELECTROSTATICS */
1033             velec            = _mm256_mul_pd(qq22,_mm256_sub_pd(_mm256_add_pd(rinv22,_mm256_mul_pd(krf,rsq22)),crf));
1034             felec            = _mm256_mul_pd(qq22,_mm256_sub_pd(_mm256_mul_pd(rinv22,rinvsq22),krf2));
1035
1036             cutoff_mask      = _mm256_cmp_pd(rsq22,rcutoff2,_CMP_LT_OQ);
1037
1038             /* Update potential sum for this i atom from the interaction with this j atom. */
1039             velec            = _mm256_and_pd(velec,cutoff_mask);
1040             velec            = _mm256_andnot_pd(dummy_mask,velec);
1041             velecsum         = _mm256_add_pd(velecsum,velec);
1042
1043             fscal            = felec;
1044
1045             fscal            = _mm256_and_pd(fscal,cutoff_mask);
1046
1047             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
1048
1049             /* Calculate temporary vectorial force */
1050             tx               = _mm256_mul_pd(fscal,dx22);
1051             ty               = _mm256_mul_pd(fscal,dy22);
1052             tz               = _mm256_mul_pd(fscal,dz22);
1053
1054             /* Update vectorial force */
1055             fix2             = _mm256_add_pd(fix2,tx);
1056             fiy2             = _mm256_add_pd(fiy2,ty);
1057             fiz2             = _mm256_add_pd(fiz2,tz);
1058
1059             fjx2             = _mm256_add_pd(fjx2,tx);
1060             fjy2             = _mm256_add_pd(fjy2,ty);
1061             fjz2             = _mm256_add_pd(fjz2,tz);
1062
1063             }
1064
1065             /**************************
1066              * CALCULATE INTERACTIONS *
1067              **************************/
1068
1069             if (gmx_mm256_any_lt(rsq23,rcutoff2))
1070             {
1071
1072             /* REACTION-FIELD ELECTROSTATICS */
1073             velec            = _mm256_mul_pd(qq23,_mm256_sub_pd(_mm256_add_pd(rinv23,_mm256_mul_pd(krf,rsq23)),crf));
1074             felec            = _mm256_mul_pd(qq23,_mm256_sub_pd(_mm256_mul_pd(rinv23,rinvsq23),krf2));
1075
1076             cutoff_mask      = _mm256_cmp_pd(rsq23,rcutoff2,_CMP_LT_OQ);
1077
1078             /* Update potential sum for this i atom from the interaction with this j atom. */
1079             velec            = _mm256_and_pd(velec,cutoff_mask);
1080             velec            = _mm256_andnot_pd(dummy_mask,velec);
1081             velecsum         = _mm256_add_pd(velecsum,velec);
1082
1083             fscal            = felec;
1084
1085             fscal            = _mm256_and_pd(fscal,cutoff_mask);
1086
1087             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
1088
1089             /* Calculate temporary vectorial force */
1090             tx               = _mm256_mul_pd(fscal,dx23);
1091             ty               = _mm256_mul_pd(fscal,dy23);
1092             tz               = _mm256_mul_pd(fscal,dz23);
1093
1094             /* Update vectorial force */
1095             fix2             = _mm256_add_pd(fix2,tx);
1096             fiy2             = _mm256_add_pd(fiy2,ty);
1097             fiz2             = _mm256_add_pd(fiz2,tz);
1098
1099             fjx3             = _mm256_add_pd(fjx3,tx);
1100             fjy3             = _mm256_add_pd(fjy3,ty);
1101             fjz3             = _mm256_add_pd(fjz3,tz);
1102
1103             }
1104
1105             /**************************
1106              * CALCULATE INTERACTIONS *
1107              **************************/
1108
1109             if (gmx_mm256_any_lt(rsq31,rcutoff2))
1110             {
1111
1112             /* REACTION-FIELD ELECTROSTATICS */
1113             velec            = _mm256_mul_pd(qq31,_mm256_sub_pd(_mm256_add_pd(rinv31,_mm256_mul_pd(krf,rsq31)),crf));
1114             felec            = _mm256_mul_pd(qq31,_mm256_sub_pd(_mm256_mul_pd(rinv31,rinvsq31),krf2));
1115
1116             cutoff_mask      = _mm256_cmp_pd(rsq31,rcutoff2,_CMP_LT_OQ);
1117
1118             /* Update potential sum for this i atom from the interaction with this j atom. */
1119             velec            = _mm256_and_pd(velec,cutoff_mask);
1120             velec            = _mm256_andnot_pd(dummy_mask,velec);
1121             velecsum         = _mm256_add_pd(velecsum,velec);
1122
1123             fscal            = felec;
1124
1125             fscal            = _mm256_and_pd(fscal,cutoff_mask);
1126
1127             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
1128
1129             /* Calculate temporary vectorial force */
1130             tx               = _mm256_mul_pd(fscal,dx31);
1131             ty               = _mm256_mul_pd(fscal,dy31);
1132             tz               = _mm256_mul_pd(fscal,dz31);
1133
1134             /* Update vectorial force */
1135             fix3             = _mm256_add_pd(fix3,tx);
1136             fiy3             = _mm256_add_pd(fiy3,ty);
1137             fiz3             = _mm256_add_pd(fiz3,tz);
1138
1139             fjx1             = _mm256_add_pd(fjx1,tx);
1140             fjy1             = _mm256_add_pd(fjy1,ty);
1141             fjz1             = _mm256_add_pd(fjz1,tz);
1142
1143             }
1144
1145             /**************************
1146              * CALCULATE INTERACTIONS *
1147              **************************/
1148
1149             if (gmx_mm256_any_lt(rsq32,rcutoff2))
1150             {
1151
1152             /* REACTION-FIELD ELECTROSTATICS */
1153             velec            = _mm256_mul_pd(qq32,_mm256_sub_pd(_mm256_add_pd(rinv32,_mm256_mul_pd(krf,rsq32)),crf));
1154             felec            = _mm256_mul_pd(qq32,_mm256_sub_pd(_mm256_mul_pd(rinv32,rinvsq32),krf2));
1155
1156             cutoff_mask      = _mm256_cmp_pd(rsq32,rcutoff2,_CMP_LT_OQ);
1157
1158             /* Update potential sum for this i atom from the interaction with this j atom. */
1159             velec            = _mm256_and_pd(velec,cutoff_mask);
1160             velec            = _mm256_andnot_pd(dummy_mask,velec);
1161             velecsum         = _mm256_add_pd(velecsum,velec);
1162
1163             fscal            = felec;
1164
1165             fscal            = _mm256_and_pd(fscal,cutoff_mask);
1166
1167             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
1168
1169             /* Calculate temporary vectorial force */
1170             tx               = _mm256_mul_pd(fscal,dx32);
1171             ty               = _mm256_mul_pd(fscal,dy32);
1172             tz               = _mm256_mul_pd(fscal,dz32);
1173
1174             /* Update vectorial force */
1175             fix3             = _mm256_add_pd(fix3,tx);
1176             fiy3             = _mm256_add_pd(fiy3,ty);
1177             fiz3             = _mm256_add_pd(fiz3,tz);
1178
1179             fjx2             = _mm256_add_pd(fjx2,tx);
1180             fjy2             = _mm256_add_pd(fjy2,ty);
1181             fjz2             = _mm256_add_pd(fjz2,tz);
1182
1183             }
1184
1185             /**************************
1186              * CALCULATE INTERACTIONS *
1187              **************************/
1188
1189             if (gmx_mm256_any_lt(rsq33,rcutoff2))
1190             {
1191
1192             /* REACTION-FIELD ELECTROSTATICS */
1193             velec            = _mm256_mul_pd(qq33,_mm256_sub_pd(_mm256_add_pd(rinv33,_mm256_mul_pd(krf,rsq33)),crf));
1194             felec            = _mm256_mul_pd(qq33,_mm256_sub_pd(_mm256_mul_pd(rinv33,rinvsq33),krf2));
1195
1196             cutoff_mask      = _mm256_cmp_pd(rsq33,rcutoff2,_CMP_LT_OQ);
1197
1198             /* Update potential sum for this i atom from the interaction with this j atom. */
1199             velec            = _mm256_and_pd(velec,cutoff_mask);
1200             velec            = _mm256_andnot_pd(dummy_mask,velec);
1201             velecsum         = _mm256_add_pd(velecsum,velec);
1202
1203             fscal            = felec;
1204
1205             fscal            = _mm256_and_pd(fscal,cutoff_mask);
1206
1207             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
1208
1209             /* Calculate temporary vectorial force */
1210             tx               = _mm256_mul_pd(fscal,dx33);
1211             ty               = _mm256_mul_pd(fscal,dy33);
1212             tz               = _mm256_mul_pd(fscal,dz33);
1213
1214             /* Update vectorial force */
1215             fix3             = _mm256_add_pd(fix3,tx);
1216             fiy3             = _mm256_add_pd(fiy3,ty);
1217             fiz3             = _mm256_add_pd(fiz3,tz);
1218
1219             fjx3             = _mm256_add_pd(fjx3,tx);
1220             fjy3             = _mm256_add_pd(fjy3,ty);
1221             fjz3             = _mm256_add_pd(fjz3,tz);
1222
1223             }
1224
1225             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
1226             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
1227             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
1228             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
1229
1230             gmx_mm256_decrement_4rvec_4ptr_swizzle_pd(fjptrA,fjptrB,fjptrC,fjptrD,
1231                                                       fjx0,fjy0,fjz0,fjx1,fjy1,fjz1,
1232                                                       fjx2,fjy2,fjz2,fjx3,fjy3,fjz3);
1233
1234             /* Inner loop uses 368 flops */
1235         }
1236
1237         /* End of innermost loop */
1238
1239         gmx_mm256_update_iforce_4atom_swizzle_pd(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,fix3,fiy3,fiz3,
1240                                                  f+i_coord_offset,fshift+i_shift_offset);
1241
1242         ggid                        = gid[iidx];
1243         /* Update potential energies */
1244         gmx_mm256_update_1pot_pd(velecsum,kernel_data->energygrp_elec+ggid);
1245         gmx_mm256_update_1pot_pd(vvdwsum,kernel_data->energygrp_vdw+ggid);
1246
1247         /* Increment number of inner iterations */
1248         inneriter                  += j_index_end - j_index_start;
1249
1250         /* Outer loop uses 26 flops */
1251     }
1252
1253     /* Increment number of outer iterations */
1254     outeriter        += nri;
1255
1256     /* Update outer/inner flops */
1257
1258     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W4W4_VF,outeriter*26 + inneriter*368);
1259 }
1260 /*
1261  * Gromacs nonbonded kernel:   nb_kernel_ElecRFCut_VdwLJSh_GeomW4W4_F_avx_256_double
1262  * Electrostatics interaction: ReactionField
1263  * VdW interaction:            LennardJones
1264  * Geometry:                   Water4-Water4
1265  * Calculate force/pot:        Force
1266  */
1267 void
1268 nb_kernel_ElecRFCut_VdwLJSh_GeomW4W4_F_avx_256_double
1269                     (t_nblist                    * gmx_restrict       nlist,
1270                      rvec                        * gmx_restrict          xx,
1271                      rvec                        * gmx_restrict          ff,
1272                      t_forcerec                  * gmx_restrict          fr,
1273                      t_mdatoms                   * gmx_restrict     mdatoms,
1274                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
1275                      t_nrnb                      * gmx_restrict        nrnb)
1276 {
1277     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
1278      * just 0 for non-waters.
1279      * Suffixes A,B,C,D refer to j loop unrolling done with AVX, e.g. for the four different
1280      * jnr indices corresponding to data put in the four positions in the SIMD register.
1281      */
1282     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
1283     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
1284     int              jnrA,jnrB,jnrC,jnrD;
1285     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
1286     int              jnrlistE,jnrlistF,jnrlistG,jnrlistH;
1287     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
1288     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
1289     real             rcutoff_scalar;
1290     real             *shiftvec,*fshift,*x,*f;
1291     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD;
1292     real             scratch[4*DIM];
1293     __m256d          tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
1294     real *           vdwioffsetptr0;
1295     __m256d          ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
1296     real *           vdwioffsetptr1;
1297     __m256d          ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
1298     real *           vdwioffsetptr2;
1299     __m256d          ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
1300     real *           vdwioffsetptr3;
1301     __m256d          ix3,iy3,iz3,fix3,fiy3,fiz3,iq3,isai3;
1302     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
1303     __m256d          jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
1304     int              vdwjidx1A,vdwjidx1B,vdwjidx1C,vdwjidx1D;
1305     __m256d          jx1,jy1,jz1,fjx1,fjy1,fjz1,jq1,isaj1;
1306     int              vdwjidx2A,vdwjidx2B,vdwjidx2C,vdwjidx2D;
1307     __m256d          jx2,jy2,jz2,fjx2,fjy2,fjz2,jq2,isaj2;
1308     int              vdwjidx3A,vdwjidx3B,vdwjidx3C,vdwjidx3D;
1309     __m256d          jx3,jy3,jz3,fjx3,fjy3,fjz3,jq3,isaj3;
1310     __m256d          dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
1311     __m256d          dx11,dy11,dz11,rsq11,rinv11,rinvsq11,r11,qq11,c6_11,c12_11;
1312     __m256d          dx12,dy12,dz12,rsq12,rinv12,rinvsq12,r12,qq12,c6_12,c12_12;
1313     __m256d          dx13,dy13,dz13,rsq13,rinv13,rinvsq13,r13,qq13,c6_13,c12_13;
1314     __m256d          dx21,dy21,dz21,rsq21,rinv21,rinvsq21,r21,qq21,c6_21,c12_21;
1315     __m256d          dx22,dy22,dz22,rsq22,rinv22,rinvsq22,r22,qq22,c6_22,c12_22;
1316     __m256d          dx23,dy23,dz23,rsq23,rinv23,rinvsq23,r23,qq23,c6_23,c12_23;
1317     __m256d          dx31,dy31,dz31,rsq31,rinv31,rinvsq31,r31,qq31,c6_31,c12_31;
1318     __m256d          dx32,dy32,dz32,rsq32,rinv32,rinvsq32,r32,qq32,c6_32,c12_32;
1319     __m256d          dx33,dy33,dz33,rsq33,rinv33,rinvsq33,r33,qq33,c6_33,c12_33;
1320     __m256d          velec,felec,velecsum,facel,crf,krf,krf2;
1321     real             *charge;
1322     int              nvdwtype;
1323     __m256d          rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
1324     int              *vdwtype;
1325     real             *vdwparam;
1326     __m256d          one_sixth   = _mm256_set1_pd(1.0/6.0);
1327     __m256d          one_twelfth = _mm256_set1_pd(1.0/12.0);
1328     __m256d          dummy_mask,cutoff_mask;
1329     __m128           tmpmask0,tmpmask1;
1330     __m256d          signbit = _mm256_castsi256_pd( _mm256_set1_epi32(0x80000000) );
1331     __m256d          one     = _mm256_set1_pd(1.0);
1332     __m256d          two     = _mm256_set1_pd(2.0);
1333     x                = xx[0];
1334     f                = ff[0];
1335
1336     nri              = nlist->nri;
1337     iinr             = nlist->iinr;
1338     jindex           = nlist->jindex;
1339     jjnr             = nlist->jjnr;
1340     shiftidx         = nlist->shift;
1341     gid              = nlist->gid;
1342     shiftvec         = fr->shift_vec[0];
1343     fshift           = fr->fshift[0];
1344     facel            = _mm256_set1_pd(fr->epsfac);
1345     charge           = mdatoms->chargeA;
1346     krf              = _mm256_set1_pd(fr->ic->k_rf);
1347     krf2             = _mm256_set1_pd(fr->ic->k_rf*2.0);
1348     crf              = _mm256_set1_pd(fr->ic->c_rf);
1349     nvdwtype         = fr->ntype;
1350     vdwparam         = fr->nbfp;
1351     vdwtype          = mdatoms->typeA;
1352
1353     /* Setup water-specific parameters */
1354     inr              = nlist->iinr[0];
1355     iq1              = _mm256_mul_pd(facel,_mm256_set1_pd(charge[inr+1]));
1356     iq2              = _mm256_mul_pd(facel,_mm256_set1_pd(charge[inr+2]));
1357     iq3              = _mm256_mul_pd(facel,_mm256_set1_pd(charge[inr+3]));
1358     vdwioffsetptr0   = vdwparam+2*nvdwtype*vdwtype[inr+0];
1359
1360     jq1              = _mm256_set1_pd(charge[inr+1]);
1361     jq2              = _mm256_set1_pd(charge[inr+2]);
1362     jq3              = _mm256_set1_pd(charge[inr+3]);
1363     vdwjidx0A        = 2*vdwtype[inr+0];
1364     c6_00            = _mm256_set1_pd(vdwioffsetptr0[vdwjidx0A]);
1365     c12_00           = _mm256_set1_pd(vdwioffsetptr0[vdwjidx0A+1]);
1366     qq11             = _mm256_mul_pd(iq1,jq1);
1367     qq12             = _mm256_mul_pd(iq1,jq2);
1368     qq13             = _mm256_mul_pd(iq1,jq3);
1369     qq21             = _mm256_mul_pd(iq2,jq1);
1370     qq22             = _mm256_mul_pd(iq2,jq2);
1371     qq23             = _mm256_mul_pd(iq2,jq3);
1372     qq31             = _mm256_mul_pd(iq3,jq1);
1373     qq32             = _mm256_mul_pd(iq3,jq2);
1374     qq33             = _mm256_mul_pd(iq3,jq3);
1375
1376     /* When we use explicit cutoffs the value must be identical for elec and VdW, so use elec as an arbitrary choice */
1377     rcutoff_scalar   = fr->rcoulomb;
1378     rcutoff          = _mm256_set1_pd(rcutoff_scalar);
1379     rcutoff2         = _mm256_mul_pd(rcutoff,rcutoff);
1380
1381     sh_vdw_invrcut6  = _mm256_set1_pd(fr->ic->sh_invrc6);
1382     rvdw             = _mm256_set1_pd(fr->rvdw);
1383
1384     /* Avoid stupid compiler warnings */
1385     jnrA = jnrB = jnrC = jnrD = 0;
1386     j_coord_offsetA = 0;
1387     j_coord_offsetB = 0;
1388     j_coord_offsetC = 0;
1389     j_coord_offsetD = 0;
1390
1391     outeriter        = 0;
1392     inneriter        = 0;
1393
1394     for(iidx=0;iidx<4*DIM;iidx++)
1395     {
1396         scratch[iidx] = 0.0;
1397     }
1398
1399     /* Start outer loop over neighborlists */
1400     for(iidx=0; iidx<nri; iidx++)
1401     {
1402         /* Load shift vector for this list */
1403         i_shift_offset   = DIM*shiftidx[iidx];
1404
1405         /* Load limits for loop over neighbors */
1406         j_index_start    = jindex[iidx];
1407         j_index_end      = jindex[iidx+1];
1408
1409         /* Get outer coordinate index */
1410         inr              = iinr[iidx];
1411         i_coord_offset   = DIM*inr;
1412
1413         /* Load i particle coords and add shift vector */
1414         gmx_mm256_load_shift_and_4rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset,
1415                                                     &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2,&ix3,&iy3,&iz3);
1416
1417         fix0             = _mm256_setzero_pd();
1418         fiy0             = _mm256_setzero_pd();
1419         fiz0             = _mm256_setzero_pd();
1420         fix1             = _mm256_setzero_pd();
1421         fiy1             = _mm256_setzero_pd();
1422         fiz1             = _mm256_setzero_pd();
1423         fix2             = _mm256_setzero_pd();
1424         fiy2             = _mm256_setzero_pd();
1425         fiz2             = _mm256_setzero_pd();
1426         fix3             = _mm256_setzero_pd();
1427         fiy3             = _mm256_setzero_pd();
1428         fiz3             = _mm256_setzero_pd();
1429
1430         /* Start inner kernel loop */
1431         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
1432         {
1433
1434             /* Get j neighbor index, and coordinate index */
1435             jnrA             = jjnr[jidx];
1436             jnrB             = jjnr[jidx+1];
1437             jnrC             = jjnr[jidx+2];
1438             jnrD             = jjnr[jidx+3];
1439             j_coord_offsetA  = DIM*jnrA;
1440             j_coord_offsetB  = DIM*jnrB;
1441             j_coord_offsetC  = DIM*jnrC;
1442             j_coord_offsetD  = DIM*jnrD;
1443
1444             /* load j atom coordinates */
1445             gmx_mm256_load_4rvec_4ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
1446                                                  x+j_coord_offsetC,x+j_coord_offsetD,
1447                                                  &jx0,&jy0,&jz0,&jx1,&jy1,&jz1,&jx2,
1448                                                  &jy2,&jz2,&jx3,&jy3,&jz3);
1449
1450             /* Calculate displacement vector */
1451             dx00             = _mm256_sub_pd(ix0,jx0);
1452             dy00             = _mm256_sub_pd(iy0,jy0);
1453             dz00             = _mm256_sub_pd(iz0,jz0);
1454             dx11             = _mm256_sub_pd(ix1,jx1);
1455             dy11             = _mm256_sub_pd(iy1,jy1);
1456             dz11             = _mm256_sub_pd(iz1,jz1);
1457             dx12             = _mm256_sub_pd(ix1,jx2);
1458             dy12             = _mm256_sub_pd(iy1,jy2);
1459             dz12             = _mm256_sub_pd(iz1,jz2);
1460             dx13             = _mm256_sub_pd(ix1,jx3);
1461             dy13             = _mm256_sub_pd(iy1,jy3);
1462             dz13             = _mm256_sub_pd(iz1,jz3);
1463             dx21             = _mm256_sub_pd(ix2,jx1);
1464             dy21             = _mm256_sub_pd(iy2,jy1);
1465             dz21             = _mm256_sub_pd(iz2,jz1);
1466             dx22             = _mm256_sub_pd(ix2,jx2);
1467             dy22             = _mm256_sub_pd(iy2,jy2);
1468             dz22             = _mm256_sub_pd(iz2,jz2);
1469             dx23             = _mm256_sub_pd(ix2,jx3);
1470             dy23             = _mm256_sub_pd(iy2,jy3);
1471             dz23             = _mm256_sub_pd(iz2,jz3);
1472             dx31             = _mm256_sub_pd(ix3,jx1);
1473             dy31             = _mm256_sub_pd(iy3,jy1);
1474             dz31             = _mm256_sub_pd(iz3,jz1);
1475             dx32             = _mm256_sub_pd(ix3,jx2);
1476             dy32             = _mm256_sub_pd(iy3,jy2);
1477             dz32             = _mm256_sub_pd(iz3,jz2);
1478             dx33             = _mm256_sub_pd(ix3,jx3);
1479             dy33             = _mm256_sub_pd(iy3,jy3);
1480             dz33             = _mm256_sub_pd(iz3,jz3);
1481
1482             /* Calculate squared distance and things based on it */
1483             rsq00            = gmx_mm256_calc_rsq_pd(dx00,dy00,dz00);
1484             rsq11            = gmx_mm256_calc_rsq_pd(dx11,dy11,dz11);
1485             rsq12            = gmx_mm256_calc_rsq_pd(dx12,dy12,dz12);
1486             rsq13            = gmx_mm256_calc_rsq_pd(dx13,dy13,dz13);
1487             rsq21            = gmx_mm256_calc_rsq_pd(dx21,dy21,dz21);
1488             rsq22            = gmx_mm256_calc_rsq_pd(dx22,dy22,dz22);
1489             rsq23            = gmx_mm256_calc_rsq_pd(dx23,dy23,dz23);
1490             rsq31            = gmx_mm256_calc_rsq_pd(dx31,dy31,dz31);
1491             rsq32            = gmx_mm256_calc_rsq_pd(dx32,dy32,dz32);
1492             rsq33            = gmx_mm256_calc_rsq_pd(dx33,dy33,dz33);
1493
1494             rinv11           = gmx_mm256_invsqrt_pd(rsq11);
1495             rinv12           = gmx_mm256_invsqrt_pd(rsq12);
1496             rinv13           = gmx_mm256_invsqrt_pd(rsq13);
1497             rinv21           = gmx_mm256_invsqrt_pd(rsq21);
1498             rinv22           = gmx_mm256_invsqrt_pd(rsq22);
1499             rinv23           = gmx_mm256_invsqrt_pd(rsq23);
1500             rinv31           = gmx_mm256_invsqrt_pd(rsq31);
1501             rinv32           = gmx_mm256_invsqrt_pd(rsq32);
1502             rinv33           = gmx_mm256_invsqrt_pd(rsq33);
1503
1504             rinvsq00         = gmx_mm256_inv_pd(rsq00);
1505             rinvsq11         = _mm256_mul_pd(rinv11,rinv11);
1506             rinvsq12         = _mm256_mul_pd(rinv12,rinv12);
1507             rinvsq13         = _mm256_mul_pd(rinv13,rinv13);
1508             rinvsq21         = _mm256_mul_pd(rinv21,rinv21);
1509             rinvsq22         = _mm256_mul_pd(rinv22,rinv22);
1510             rinvsq23         = _mm256_mul_pd(rinv23,rinv23);
1511             rinvsq31         = _mm256_mul_pd(rinv31,rinv31);
1512             rinvsq32         = _mm256_mul_pd(rinv32,rinv32);
1513             rinvsq33         = _mm256_mul_pd(rinv33,rinv33);
1514
1515             fjx0             = _mm256_setzero_pd();
1516             fjy0             = _mm256_setzero_pd();
1517             fjz0             = _mm256_setzero_pd();
1518             fjx1             = _mm256_setzero_pd();
1519             fjy1             = _mm256_setzero_pd();
1520             fjz1             = _mm256_setzero_pd();
1521             fjx2             = _mm256_setzero_pd();
1522             fjy2             = _mm256_setzero_pd();
1523             fjz2             = _mm256_setzero_pd();
1524             fjx3             = _mm256_setzero_pd();
1525             fjy3             = _mm256_setzero_pd();
1526             fjz3             = _mm256_setzero_pd();
1527
1528             /**************************
1529              * CALCULATE INTERACTIONS *
1530              **************************/
1531
1532             if (gmx_mm256_any_lt(rsq00,rcutoff2))
1533             {
1534
1535             /* LENNARD-JONES DISPERSION/REPULSION */
1536
1537             rinvsix          = _mm256_mul_pd(_mm256_mul_pd(rinvsq00,rinvsq00),rinvsq00);
1538             fvdw             = _mm256_mul_pd(_mm256_sub_pd(_mm256_mul_pd(c12_00,rinvsix),c6_00),_mm256_mul_pd(rinvsix,rinvsq00));
1539
1540             cutoff_mask      = _mm256_cmp_pd(rsq00,rcutoff2,_CMP_LT_OQ);
1541
1542             fscal            = fvdw;
1543
1544             fscal            = _mm256_and_pd(fscal,cutoff_mask);
1545
1546             /* Calculate temporary vectorial force */
1547             tx               = _mm256_mul_pd(fscal,dx00);
1548             ty               = _mm256_mul_pd(fscal,dy00);
1549             tz               = _mm256_mul_pd(fscal,dz00);
1550
1551             /* Update vectorial force */
1552             fix0             = _mm256_add_pd(fix0,tx);
1553             fiy0             = _mm256_add_pd(fiy0,ty);
1554             fiz0             = _mm256_add_pd(fiz0,tz);
1555
1556             fjx0             = _mm256_add_pd(fjx0,tx);
1557             fjy0             = _mm256_add_pd(fjy0,ty);
1558             fjz0             = _mm256_add_pd(fjz0,tz);
1559
1560             }
1561
1562             /**************************
1563              * CALCULATE INTERACTIONS *
1564              **************************/
1565
1566             if (gmx_mm256_any_lt(rsq11,rcutoff2))
1567             {
1568
1569             /* REACTION-FIELD ELECTROSTATICS */
1570             felec            = _mm256_mul_pd(qq11,_mm256_sub_pd(_mm256_mul_pd(rinv11,rinvsq11),krf2));
1571
1572             cutoff_mask      = _mm256_cmp_pd(rsq11,rcutoff2,_CMP_LT_OQ);
1573
1574             fscal            = felec;
1575
1576             fscal            = _mm256_and_pd(fscal,cutoff_mask);
1577
1578             /* Calculate temporary vectorial force */
1579             tx               = _mm256_mul_pd(fscal,dx11);
1580             ty               = _mm256_mul_pd(fscal,dy11);
1581             tz               = _mm256_mul_pd(fscal,dz11);
1582
1583             /* Update vectorial force */
1584             fix1             = _mm256_add_pd(fix1,tx);
1585             fiy1             = _mm256_add_pd(fiy1,ty);
1586             fiz1             = _mm256_add_pd(fiz1,tz);
1587
1588             fjx1             = _mm256_add_pd(fjx1,tx);
1589             fjy1             = _mm256_add_pd(fjy1,ty);
1590             fjz1             = _mm256_add_pd(fjz1,tz);
1591
1592             }
1593
1594             /**************************
1595              * CALCULATE INTERACTIONS *
1596              **************************/
1597
1598             if (gmx_mm256_any_lt(rsq12,rcutoff2))
1599             {
1600
1601             /* REACTION-FIELD ELECTROSTATICS */
1602             felec            = _mm256_mul_pd(qq12,_mm256_sub_pd(_mm256_mul_pd(rinv12,rinvsq12),krf2));
1603
1604             cutoff_mask      = _mm256_cmp_pd(rsq12,rcutoff2,_CMP_LT_OQ);
1605
1606             fscal            = felec;
1607
1608             fscal            = _mm256_and_pd(fscal,cutoff_mask);
1609
1610             /* Calculate temporary vectorial force */
1611             tx               = _mm256_mul_pd(fscal,dx12);
1612             ty               = _mm256_mul_pd(fscal,dy12);
1613             tz               = _mm256_mul_pd(fscal,dz12);
1614
1615             /* Update vectorial force */
1616             fix1             = _mm256_add_pd(fix1,tx);
1617             fiy1             = _mm256_add_pd(fiy1,ty);
1618             fiz1             = _mm256_add_pd(fiz1,tz);
1619
1620             fjx2             = _mm256_add_pd(fjx2,tx);
1621             fjy2             = _mm256_add_pd(fjy2,ty);
1622             fjz2             = _mm256_add_pd(fjz2,tz);
1623
1624             }
1625
1626             /**************************
1627              * CALCULATE INTERACTIONS *
1628              **************************/
1629
1630             if (gmx_mm256_any_lt(rsq13,rcutoff2))
1631             {
1632
1633             /* REACTION-FIELD ELECTROSTATICS */
1634             felec            = _mm256_mul_pd(qq13,_mm256_sub_pd(_mm256_mul_pd(rinv13,rinvsq13),krf2));
1635
1636             cutoff_mask      = _mm256_cmp_pd(rsq13,rcutoff2,_CMP_LT_OQ);
1637
1638             fscal            = felec;
1639
1640             fscal            = _mm256_and_pd(fscal,cutoff_mask);
1641
1642             /* Calculate temporary vectorial force */
1643             tx               = _mm256_mul_pd(fscal,dx13);
1644             ty               = _mm256_mul_pd(fscal,dy13);
1645             tz               = _mm256_mul_pd(fscal,dz13);
1646
1647             /* Update vectorial force */
1648             fix1             = _mm256_add_pd(fix1,tx);
1649             fiy1             = _mm256_add_pd(fiy1,ty);
1650             fiz1             = _mm256_add_pd(fiz1,tz);
1651
1652             fjx3             = _mm256_add_pd(fjx3,tx);
1653             fjy3             = _mm256_add_pd(fjy3,ty);
1654             fjz3             = _mm256_add_pd(fjz3,tz);
1655
1656             }
1657
1658             /**************************
1659              * CALCULATE INTERACTIONS *
1660              **************************/
1661
1662             if (gmx_mm256_any_lt(rsq21,rcutoff2))
1663             {
1664
1665             /* REACTION-FIELD ELECTROSTATICS */
1666             felec            = _mm256_mul_pd(qq21,_mm256_sub_pd(_mm256_mul_pd(rinv21,rinvsq21),krf2));
1667
1668             cutoff_mask      = _mm256_cmp_pd(rsq21,rcutoff2,_CMP_LT_OQ);
1669
1670             fscal            = felec;
1671
1672             fscal            = _mm256_and_pd(fscal,cutoff_mask);
1673
1674             /* Calculate temporary vectorial force */
1675             tx               = _mm256_mul_pd(fscal,dx21);
1676             ty               = _mm256_mul_pd(fscal,dy21);
1677             tz               = _mm256_mul_pd(fscal,dz21);
1678
1679             /* Update vectorial force */
1680             fix2             = _mm256_add_pd(fix2,tx);
1681             fiy2             = _mm256_add_pd(fiy2,ty);
1682             fiz2             = _mm256_add_pd(fiz2,tz);
1683
1684             fjx1             = _mm256_add_pd(fjx1,tx);
1685             fjy1             = _mm256_add_pd(fjy1,ty);
1686             fjz1             = _mm256_add_pd(fjz1,tz);
1687
1688             }
1689
1690             /**************************
1691              * CALCULATE INTERACTIONS *
1692              **************************/
1693
1694             if (gmx_mm256_any_lt(rsq22,rcutoff2))
1695             {
1696
1697             /* REACTION-FIELD ELECTROSTATICS */
1698             felec            = _mm256_mul_pd(qq22,_mm256_sub_pd(_mm256_mul_pd(rinv22,rinvsq22),krf2));
1699
1700             cutoff_mask      = _mm256_cmp_pd(rsq22,rcutoff2,_CMP_LT_OQ);
1701
1702             fscal            = felec;
1703
1704             fscal            = _mm256_and_pd(fscal,cutoff_mask);
1705
1706             /* Calculate temporary vectorial force */
1707             tx               = _mm256_mul_pd(fscal,dx22);
1708             ty               = _mm256_mul_pd(fscal,dy22);
1709             tz               = _mm256_mul_pd(fscal,dz22);
1710
1711             /* Update vectorial force */
1712             fix2             = _mm256_add_pd(fix2,tx);
1713             fiy2             = _mm256_add_pd(fiy2,ty);
1714             fiz2             = _mm256_add_pd(fiz2,tz);
1715
1716             fjx2             = _mm256_add_pd(fjx2,tx);
1717             fjy2             = _mm256_add_pd(fjy2,ty);
1718             fjz2             = _mm256_add_pd(fjz2,tz);
1719
1720             }
1721
1722             /**************************
1723              * CALCULATE INTERACTIONS *
1724              **************************/
1725
1726             if (gmx_mm256_any_lt(rsq23,rcutoff2))
1727             {
1728
1729             /* REACTION-FIELD ELECTROSTATICS */
1730             felec            = _mm256_mul_pd(qq23,_mm256_sub_pd(_mm256_mul_pd(rinv23,rinvsq23),krf2));
1731
1732             cutoff_mask      = _mm256_cmp_pd(rsq23,rcutoff2,_CMP_LT_OQ);
1733
1734             fscal            = felec;
1735
1736             fscal            = _mm256_and_pd(fscal,cutoff_mask);
1737
1738             /* Calculate temporary vectorial force */
1739             tx               = _mm256_mul_pd(fscal,dx23);
1740             ty               = _mm256_mul_pd(fscal,dy23);
1741             tz               = _mm256_mul_pd(fscal,dz23);
1742
1743             /* Update vectorial force */
1744             fix2             = _mm256_add_pd(fix2,tx);
1745             fiy2             = _mm256_add_pd(fiy2,ty);
1746             fiz2             = _mm256_add_pd(fiz2,tz);
1747
1748             fjx3             = _mm256_add_pd(fjx3,tx);
1749             fjy3             = _mm256_add_pd(fjy3,ty);
1750             fjz3             = _mm256_add_pd(fjz3,tz);
1751
1752             }
1753
1754             /**************************
1755              * CALCULATE INTERACTIONS *
1756              **************************/
1757
1758             if (gmx_mm256_any_lt(rsq31,rcutoff2))
1759             {
1760
1761             /* REACTION-FIELD ELECTROSTATICS */
1762             felec            = _mm256_mul_pd(qq31,_mm256_sub_pd(_mm256_mul_pd(rinv31,rinvsq31),krf2));
1763
1764             cutoff_mask      = _mm256_cmp_pd(rsq31,rcutoff2,_CMP_LT_OQ);
1765
1766             fscal            = felec;
1767
1768             fscal            = _mm256_and_pd(fscal,cutoff_mask);
1769
1770             /* Calculate temporary vectorial force */
1771             tx               = _mm256_mul_pd(fscal,dx31);
1772             ty               = _mm256_mul_pd(fscal,dy31);
1773             tz               = _mm256_mul_pd(fscal,dz31);
1774
1775             /* Update vectorial force */
1776             fix3             = _mm256_add_pd(fix3,tx);
1777             fiy3             = _mm256_add_pd(fiy3,ty);
1778             fiz3             = _mm256_add_pd(fiz3,tz);
1779
1780             fjx1             = _mm256_add_pd(fjx1,tx);
1781             fjy1             = _mm256_add_pd(fjy1,ty);
1782             fjz1             = _mm256_add_pd(fjz1,tz);
1783
1784             }
1785
1786             /**************************
1787              * CALCULATE INTERACTIONS *
1788              **************************/
1789
1790             if (gmx_mm256_any_lt(rsq32,rcutoff2))
1791             {
1792
1793             /* REACTION-FIELD ELECTROSTATICS */
1794             felec            = _mm256_mul_pd(qq32,_mm256_sub_pd(_mm256_mul_pd(rinv32,rinvsq32),krf2));
1795
1796             cutoff_mask      = _mm256_cmp_pd(rsq32,rcutoff2,_CMP_LT_OQ);
1797
1798             fscal            = felec;
1799
1800             fscal            = _mm256_and_pd(fscal,cutoff_mask);
1801
1802             /* Calculate temporary vectorial force */
1803             tx               = _mm256_mul_pd(fscal,dx32);
1804             ty               = _mm256_mul_pd(fscal,dy32);
1805             tz               = _mm256_mul_pd(fscal,dz32);
1806
1807             /* Update vectorial force */
1808             fix3             = _mm256_add_pd(fix3,tx);
1809             fiy3             = _mm256_add_pd(fiy3,ty);
1810             fiz3             = _mm256_add_pd(fiz3,tz);
1811
1812             fjx2             = _mm256_add_pd(fjx2,tx);
1813             fjy2             = _mm256_add_pd(fjy2,ty);
1814             fjz2             = _mm256_add_pd(fjz2,tz);
1815
1816             }
1817
1818             /**************************
1819              * CALCULATE INTERACTIONS *
1820              **************************/
1821
1822             if (gmx_mm256_any_lt(rsq33,rcutoff2))
1823             {
1824
1825             /* REACTION-FIELD ELECTROSTATICS */
1826             felec            = _mm256_mul_pd(qq33,_mm256_sub_pd(_mm256_mul_pd(rinv33,rinvsq33),krf2));
1827
1828             cutoff_mask      = _mm256_cmp_pd(rsq33,rcutoff2,_CMP_LT_OQ);
1829
1830             fscal            = felec;
1831
1832             fscal            = _mm256_and_pd(fscal,cutoff_mask);
1833
1834             /* Calculate temporary vectorial force */
1835             tx               = _mm256_mul_pd(fscal,dx33);
1836             ty               = _mm256_mul_pd(fscal,dy33);
1837             tz               = _mm256_mul_pd(fscal,dz33);
1838
1839             /* Update vectorial force */
1840             fix3             = _mm256_add_pd(fix3,tx);
1841             fiy3             = _mm256_add_pd(fiy3,ty);
1842             fiz3             = _mm256_add_pd(fiz3,tz);
1843
1844             fjx3             = _mm256_add_pd(fjx3,tx);
1845             fjy3             = _mm256_add_pd(fjy3,ty);
1846             fjz3             = _mm256_add_pd(fjz3,tz);
1847
1848             }
1849
1850             fjptrA             = f+j_coord_offsetA;
1851             fjptrB             = f+j_coord_offsetB;
1852             fjptrC             = f+j_coord_offsetC;
1853             fjptrD             = f+j_coord_offsetD;
1854
1855             gmx_mm256_decrement_4rvec_4ptr_swizzle_pd(fjptrA,fjptrB,fjptrC,fjptrD,
1856                                                       fjx0,fjy0,fjz0,fjx1,fjy1,fjz1,
1857                                                       fjx2,fjy2,fjz2,fjx3,fjy3,fjz3);
1858
1859             /* Inner loop uses 303 flops */
1860         }
1861
1862         if(jidx<j_index_end)
1863         {
1864
1865             /* Get j neighbor index, and coordinate index */
1866             jnrlistA         = jjnr[jidx];
1867             jnrlistB         = jjnr[jidx+1];
1868             jnrlistC         = jjnr[jidx+2];
1869             jnrlistD         = jjnr[jidx+3];
1870             /* Sign of each element will be negative for non-real atoms.
1871              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
1872              * so use it as val = _mm_andnot_pd(mask,val) to clear dummy entries.
1873              */
1874             tmpmask0 = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
1875
1876             tmpmask1 = _mm_permute_ps(tmpmask0,_GMX_MM_PERMUTE(3,3,2,2));
1877             tmpmask0 = _mm_permute_ps(tmpmask0,_GMX_MM_PERMUTE(1,1,0,0));
1878             dummy_mask = _mm256_castps_pd(gmx_mm256_set_m128(tmpmask1,tmpmask0));
1879
1880             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
1881             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
1882             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
1883             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
1884             j_coord_offsetA  = DIM*jnrA;
1885             j_coord_offsetB  = DIM*jnrB;
1886             j_coord_offsetC  = DIM*jnrC;
1887             j_coord_offsetD  = DIM*jnrD;
1888
1889             /* load j atom coordinates */
1890             gmx_mm256_load_4rvec_4ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
1891                                                  x+j_coord_offsetC,x+j_coord_offsetD,
1892                                                  &jx0,&jy0,&jz0,&jx1,&jy1,&jz1,&jx2,
1893                                                  &jy2,&jz2,&jx3,&jy3,&jz3);
1894
1895             /* Calculate displacement vector */
1896             dx00             = _mm256_sub_pd(ix0,jx0);
1897             dy00             = _mm256_sub_pd(iy0,jy0);
1898             dz00             = _mm256_sub_pd(iz0,jz0);
1899             dx11             = _mm256_sub_pd(ix1,jx1);
1900             dy11             = _mm256_sub_pd(iy1,jy1);
1901             dz11             = _mm256_sub_pd(iz1,jz1);
1902             dx12             = _mm256_sub_pd(ix1,jx2);
1903             dy12             = _mm256_sub_pd(iy1,jy2);
1904             dz12             = _mm256_sub_pd(iz1,jz2);
1905             dx13             = _mm256_sub_pd(ix1,jx3);
1906             dy13             = _mm256_sub_pd(iy1,jy3);
1907             dz13             = _mm256_sub_pd(iz1,jz3);
1908             dx21             = _mm256_sub_pd(ix2,jx1);
1909             dy21             = _mm256_sub_pd(iy2,jy1);
1910             dz21             = _mm256_sub_pd(iz2,jz1);
1911             dx22             = _mm256_sub_pd(ix2,jx2);
1912             dy22             = _mm256_sub_pd(iy2,jy2);
1913             dz22             = _mm256_sub_pd(iz2,jz2);
1914             dx23             = _mm256_sub_pd(ix2,jx3);
1915             dy23             = _mm256_sub_pd(iy2,jy3);
1916             dz23             = _mm256_sub_pd(iz2,jz3);
1917             dx31             = _mm256_sub_pd(ix3,jx1);
1918             dy31             = _mm256_sub_pd(iy3,jy1);
1919             dz31             = _mm256_sub_pd(iz3,jz1);
1920             dx32             = _mm256_sub_pd(ix3,jx2);
1921             dy32             = _mm256_sub_pd(iy3,jy2);
1922             dz32             = _mm256_sub_pd(iz3,jz2);
1923             dx33             = _mm256_sub_pd(ix3,jx3);
1924             dy33             = _mm256_sub_pd(iy3,jy3);
1925             dz33             = _mm256_sub_pd(iz3,jz3);
1926
1927             /* Calculate squared distance and things based on it */
1928             rsq00            = gmx_mm256_calc_rsq_pd(dx00,dy00,dz00);
1929             rsq11            = gmx_mm256_calc_rsq_pd(dx11,dy11,dz11);
1930             rsq12            = gmx_mm256_calc_rsq_pd(dx12,dy12,dz12);
1931             rsq13            = gmx_mm256_calc_rsq_pd(dx13,dy13,dz13);
1932             rsq21            = gmx_mm256_calc_rsq_pd(dx21,dy21,dz21);
1933             rsq22            = gmx_mm256_calc_rsq_pd(dx22,dy22,dz22);
1934             rsq23            = gmx_mm256_calc_rsq_pd(dx23,dy23,dz23);
1935             rsq31            = gmx_mm256_calc_rsq_pd(dx31,dy31,dz31);
1936             rsq32            = gmx_mm256_calc_rsq_pd(dx32,dy32,dz32);
1937             rsq33            = gmx_mm256_calc_rsq_pd(dx33,dy33,dz33);
1938
1939             rinv11           = gmx_mm256_invsqrt_pd(rsq11);
1940             rinv12           = gmx_mm256_invsqrt_pd(rsq12);
1941             rinv13           = gmx_mm256_invsqrt_pd(rsq13);
1942             rinv21           = gmx_mm256_invsqrt_pd(rsq21);
1943             rinv22           = gmx_mm256_invsqrt_pd(rsq22);
1944             rinv23           = gmx_mm256_invsqrt_pd(rsq23);
1945             rinv31           = gmx_mm256_invsqrt_pd(rsq31);
1946             rinv32           = gmx_mm256_invsqrt_pd(rsq32);
1947             rinv33           = gmx_mm256_invsqrt_pd(rsq33);
1948
1949             rinvsq00         = gmx_mm256_inv_pd(rsq00);
1950             rinvsq11         = _mm256_mul_pd(rinv11,rinv11);
1951             rinvsq12         = _mm256_mul_pd(rinv12,rinv12);
1952             rinvsq13         = _mm256_mul_pd(rinv13,rinv13);
1953             rinvsq21         = _mm256_mul_pd(rinv21,rinv21);
1954             rinvsq22         = _mm256_mul_pd(rinv22,rinv22);
1955             rinvsq23         = _mm256_mul_pd(rinv23,rinv23);
1956             rinvsq31         = _mm256_mul_pd(rinv31,rinv31);
1957             rinvsq32         = _mm256_mul_pd(rinv32,rinv32);
1958             rinvsq33         = _mm256_mul_pd(rinv33,rinv33);
1959
1960             fjx0             = _mm256_setzero_pd();
1961             fjy0             = _mm256_setzero_pd();
1962             fjz0             = _mm256_setzero_pd();
1963             fjx1             = _mm256_setzero_pd();
1964             fjy1             = _mm256_setzero_pd();
1965             fjz1             = _mm256_setzero_pd();
1966             fjx2             = _mm256_setzero_pd();
1967             fjy2             = _mm256_setzero_pd();
1968             fjz2             = _mm256_setzero_pd();
1969             fjx3             = _mm256_setzero_pd();
1970             fjy3             = _mm256_setzero_pd();
1971             fjz3             = _mm256_setzero_pd();
1972
1973             /**************************
1974              * CALCULATE INTERACTIONS *
1975              **************************/
1976
1977             if (gmx_mm256_any_lt(rsq00,rcutoff2))
1978             {
1979
1980             /* LENNARD-JONES DISPERSION/REPULSION */
1981
1982             rinvsix          = _mm256_mul_pd(_mm256_mul_pd(rinvsq00,rinvsq00),rinvsq00);
1983             fvdw             = _mm256_mul_pd(_mm256_sub_pd(_mm256_mul_pd(c12_00,rinvsix),c6_00),_mm256_mul_pd(rinvsix,rinvsq00));
1984
1985             cutoff_mask      = _mm256_cmp_pd(rsq00,rcutoff2,_CMP_LT_OQ);
1986
1987             fscal            = fvdw;
1988
1989             fscal            = _mm256_and_pd(fscal,cutoff_mask);
1990
1991             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
1992
1993             /* Calculate temporary vectorial force */
1994             tx               = _mm256_mul_pd(fscal,dx00);
1995             ty               = _mm256_mul_pd(fscal,dy00);
1996             tz               = _mm256_mul_pd(fscal,dz00);
1997
1998             /* Update vectorial force */
1999             fix0             = _mm256_add_pd(fix0,tx);
2000             fiy0             = _mm256_add_pd(fiy0,ty);
2001             fiz0             = _mm256_add_pd(fiz0,tz);
2002
2003             fjx0             = _mm256_add_pd(fjx0,tx);
2004             fjy0             = _mm256_add_pd(fjy0,ty);
2005             fjz0             = _mm256_add_pd(fjz0,tz);
2006
2007             }
2008
2009             /**************************
2010              * CALCULATE INTERACTIONS *
2011              **************************/
2012
2013             if (gmx_mm256_any_lt(rsq11,rcutoff2))
2014             {
2015
2016             /* REACTION-FIELD ELECTROSTATICS */
2017             felec            = _mm256_mul_pd(qq11,_mm256_sub_pd(_mm256_mul_pd(rinv11,rinvsq11),krf2));
2018
2019             cutoff_mask      = _mm256_cmp_pd(rsq11,rcutoff2,_CMP_LT_OQ);
2020
2021             fscal            = felec;
2022
2023             fscal            = _mm256_and_pd(fscal,cutoff_mask);
2024
2025             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
2026
2027             /* Calculate temporary vectorial force */
2028             tx               = _mm256_mul_pd(fscal,dx11);
2029             ty               = _mm256_mul_pd(fscal,dy11);
2030             tz               = _mm256_mul_pd(fscal,dz11);
2031
2032             /* Update vectorial force */
2033             fix1             = _mm256_add_pd(fix1,tx);
2034             fiy1             = _mm256_add_pd(fiy1,ty);
2035             fiz1             = _mm256_add_pd(fiz1,tz);
2036
2037             fjx1             = _mm256_add_pd(fjx1,tx);
2038             fjy1             = _mm256_add_pd(fjy1,ty);
2039             fjz1             = _mm256_add_pd(fjz1,tz);
2040
2041             }
2042
2043             /**************************
2044              * CALCULATE INTERACTIONS *
2045              **************************/
2046
2047             if (gmx_mm256_any_lt(rsq12,rcutoff2))
2048             {
2049
2050             /* REACTION-FIELD ELECTROSTATICS */
2051             felec            = _mm256_mul_pd(qq12,_mm256_sub_pd(_mm256_mul_pd(rinv12,rinvsq12),krf2));
2052
2053             cutoff_mask      = _mm256_cmp_pd(rsq12,rcutoff2,_CMP_LT_OQ);
2054
2055             fscal            = felec;
2056
2057             fscal            = _mm256_and_pd(fscal,cutoff_mask);
2058
2059             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
2060
2061             /* Calculate temporary vectorial force */
2062             tx               = _mm256_mul_pd(fscal,dx12);
2063             ty               = _mm256_mul_pd(fscal,dy12);
2064             tz               = _mm256_mul_pd(fscal,dz12);
2065
2066             /* Update vectorial force */
2067             fix1             = _mm256_add_pd(fix1,tx);
2068             fiy1             = _mm256_add_pd(fiy1,ty);
2069             fiz1             = _mm256_add_pd(fiz1,tz);
2070
2071             fjx2             = _mm256_add_pd(fjx2,tx);
2072             fjy2             = _mm256_add_pd(fjy2,ty);
2073             fjz2             = _mm256_add_pd(fjz2,tz);
2074
2075             }
2076
2077             /**************************
2078              * CALCULATE INTERACTIONS *
2079              **************************/
2080
2081             if (gmx_mm256_any_lt(rsq13,rcutoff2))
2082             {
2083
2084             /* REACTION-FIELD ELECTROSTATICS */
2085             felec            = _mm256_mul_pd(qq13,_mm256_sub_pd(_mm256_mul_pd(rinv13,rinvsq13),krf2));
2086
2087             cutoff_mask      = _mm256_cmp_pd(rsq13,rcutoff2,_CMP_LT_OQ);
2088
2089             fscal            = felec;
2090
2091             fscal            = _mm256_and_pd(fscal,cutoff_mask);
2092
2093             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
2094
2095             /* Calculate temporary vectorial force */
2096             tx               = _mm256_mul_pd(fscal,dx13);
2097             ty               = _mm256_mul_pd(fscal,dy13);
2098             tz               = _mm256_mul_pd(fscal,dz13);
2099
2100             /* Update vectorial force */
2101             fix1             = _mm256_add_pd(fix1,tx);
2102             fiy1             = _mm256_add_pd(fiy1,ty);
2103             fiz1             = _mm256_add_pd(fiz1,tz);
2104
2105             fjx3             = _mm256_add_pd(fjx3,tx);
2106             fjy3             = _mm256_add_pd(fjy3,ty);
2107             fjz3             = _mm256_add_pd(fjz3,tz);
2108
2109             }
2110
2111             /**************************
2112              * CALCULATE INTERACTIONS *
2113              **************************/
2114
2115             if (gmx_mm256_any_lt(rsq21,rcutoff2))
2116             {
2117
2118             /* REACTION-FIELD ELECTROSTATICS */
2119             felec            = _mm256_mul_pd(qq21,_mm256_sub_pd(_mm256_mul_pd(rinv21,rinvsq21),krf2));
2120
2121             cutoff_mask      = _mm256_cmp_pd(rsq21,rcutoff2,_CMP_LT_OQ);
2122
2123             fscal            = felec;
2124
2125             fscal            = _mm256_and_pd(fscal,cutoff_mask);
2126
2127             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
2128
2129             /* Calculate temporary vectorial force */
2130             tx               = _mm256_mul_pd(fscal,dx21);
2131             ty               = _mm256_mul_pd(fscal,dy21);
2132             tz               = _mm256_mul_pd(fscal,dz21);
2133
2134             /* Update vectorial force */
2135             fix2             = _mm256_add_pd(fix2,tx);
2136             fiy2             = _mm256_add_pd(fiy2,ty);
2137             fiz2             = _mm256_add_pd(fiz2,tz);
2138
2139             fjx1             = _mm256_add_pd(fjx1,tx);
2140             fjy1             = _mm256_add_pd(fjy1,ty);
2141             fjz1             = _mm256_add_pd(fjz1,tz);
2142
2143             }
2144
2145             /**************************
2146              * CALCULATE INTERACTIONS *
2147              **************************/
2148
2149             if (gmx_mm256_any_lt(rsq22,rcutoff2))
2150             {
2151
2152             /* REACTION-FIELD ELECTROSTATICS */
2153             felec            = _mm256_mul_pd(qq22,_mm256_sub_pd(_mm256_mul_pd(rinv22,rinvsq22),krf2));
2154
2155             cutoff_mask      = _mm256_cmp_pd(rsq22,rcutoff2,_CMP_LT_OQ);
2156
2157             fscal            = felec;
2158
2159             fscal            = _mm256_and_pd(fscal,cutoff_mask);
2160
2161             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
2162
2163             /* Calculate temporary vectorial force */
2164             tx               = _mm256_mul_pd(fscal,dx22);
2165             ty               = _mm256_mul_pd(fscal,dy22);
2166             tz               = _mm256_mul_pd(fscal,dz22);
2167
2168             /* Update vectorial force */
2169             fix2             = _mm256_add_pd(fix2,tx);
2170             fiy2             = _mm256_add_pd(fiy2,ty);
2171             fiz2             = _mm256_add_pd(fiz2,tz);
2172
2173             fjx2             = _mm256_add_pd(fjx2,tx);
2174             fjy2             = _mm256_add_pd(fjy2,ty);
2175             fjz2             = _mm256_add_pd(fjz2,tz);
2176
2177             }
2178
2179             /**************************
2180              * CALCULATE INTERACTIONS *
2181              **************************/
2182
2183             if (gmx_mm256_any_lt(rsq23,rcutoff2))
2184             {
2185
2186             /* REACTION-FIELD ELECTROSTATICS */
2187             felec            = _mm256_mul_pd(qq23,_mm256_sub_pd(_mm256_mul_pd(rinv23,rinvsq23),krf2));
2188
2189             cutoff_mask      = _mm256_cmp_pd(rsq23,rcutoff2,_CMP_LT_OQ);
2190
2191             fscal            = felec;
2192
2193             fscal            = _mm256_and_pd(fscal,cutoff_mask);
2194
2195             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
2196
2197             /* Calculate temporary vectorial force */
2198             tx               = _mm256_mul_pd(fscal,dx23);
2199             ty               = _mm256_mul_pd(fscal,dy23);
2200             tz               = _mm256_mul_pd(fscal,dz23);
2201
2202             /* Update vectorial force */
2203             fix2             = _mm256_add_pd(fix2,tx);
2204             fiy2             = _mm256_add_pd(fiy2,ty);
2205             fiz2             = _mm256_add_pd(fiz2,tz);
2206
2207             fjx3             = _mm256_add_pd(fjx3,tx);
2208             fjy3             = _mm256_add_pd(fjy3,ty);
2209             fjz3             = _mm256_add_pd(fjz3,tz);
2210
2211             }
2212
2213             /**************************
2214              * CALCULATE INTERACTIONS *
2215              **************************/
2216
2217             if (gmx_mm256_any_lt(rsq31,rcutoff2))
2218             {
2219
2220             /* REACTION-FIELD ELECTROSTATICS */
2221             felec            = _mm256_mul_pd(qq31,_mm256_sub_pd(_mm256_mul_pd(rinv31,rinvsq31),krf2));
2222
2223             cutoff_mask      = _mm256_cmp_pd(rsq31,rcutoff2,_CMP_LT_OQ);
2224
2225             fscal            = felec;
2226
2227             fscal            = _mm256_and_pd(fscal,cutoff_mask);
2228
2229             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
2230
2231             /* Calculate temporary vectorial force */
2232             tx               = _mm256_mul_pd(fscal,dx31);
2233             ty               = _mm256_mul_pd(fscal,dy31);
2234             tz               = _mm256_mul_pd(fscal,dz31);
2235
2236             /* Update vectorial force */
2237             fix3             = _mm256_add_pd(fix3,tx);
2238             fiy3             = _mm256_add_pd(fiy3,ty);
2239             fiz3             = _mm256_add_pd(fiz3,tz);
2240
2241             fjx1             = _mm256_add_pd(fjx1,tx);
2242             fjy1             = _mm256_add_pd(fjy1,ty);
2243             fjz1             = _mm256_add_pd(fjz1,tz);
2244
2245             }
2246
2247             /**************************
2248              * CALCULATE INTERACTIONS *
2249              **************************/
2250
2251             if (gmx_mm256_any_lt(rsq32,rcutoff2))
2252             {
2253
2254             /* REACTION-FIELD ELECTROSTATICS */
2255             felec            = _mm256_mul_pd(qq32,_mm256_sub_pd(_mm256_mul_pd(rinv32,rinvsq32),krf2));
2256
2257             cutoff_mask      = _mm256_cmp_pd(rsq32,rcutoff2,_CMP_LT_OQ);
2258
2259             fscal            = felec;
2260
2261             fscal            = _mm256_and_pd(fscal,cutoff_mask);
2262
2263             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
2264
2265             /* Calculate temporary vectorial force */
2266             tx               = _mm256_mul_pd(fscal,dx32);
2267             ty               = _mm256_mul_pd(fscal,dy32);
2268             tz               = _mm256_mul_pd(fscal,dz32);
2269
2270             /* Update vectorial force */
2271             fix3             = _mm256_add_pd(fix3,tx);
2272             fiy3             = _mm256_add_pd(fiy3,ty);
2273             fiz3             = _mm256_add_pd(fiz3,tz);
2274
2275             fjx2             = _mm256_add_pd(fjx2,tx);
2276             fjy2             = _mm256_add_pd(fjy2,ty);
2277             fjz2             = _mm256_add_pd(fjz2,tz);
2278
2279             }
2280
2281             /**************************
2282              * CALCULATE INTERACTIONS *
2283              **************************/
2284
2285             if (gmx_mm256_any_lt(rsq33,rcutoff2))
2286             {
2287
2288             /* REACTION-FIELD ELECTROSTATICS */
2289             felec            = _mm256_mul_pd(qq33,_mm256_sub_pd(_mm256_mul_pd(rinv33,rinvsq33),krf2));
2290
2291             cutoff_mask      = _mm256_cmp_pd(rsq33,rcutoff2,_CMP_LT_OQ);
2292
2293             fscal            = felec;
2294
2295             fscal            = _mm256_and_pd(fscal,cutoff_mask);
2296
2297             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
2298
2299             /* Calculate temporary vectorial force */
2300             tx               = _mm256_mul_pd(fscal,dx33);
2301             ty               = _mm256_mul_pd(fscal,dy33);
2302             tz               = _mm256_mul_pd(fscal,dz33);
2303
2304             /* Update vectorial force */
2305             fix3             = _mm256_add_pd(fix3,tx);
2306             fiy3             = _mm256_add_pd(fiy3,ty);
2307             fiz3             = _mm256_add_pd(fiz3,tz);
2308
2309             fjx3             = _mm256_add_pd(fjx3,tx);
2310             fjy3             = _mm256_add_pd(fjy3,ty);
2311             fjz3             = _mm256_add_pd(fjz3,tz);
2312
2313             }
2314
2315             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
2316             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
2317             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
2318             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
2319
2320             gmx_mm256_decrement_4rvec_4ptr_swizzle_pd(fjptrA,fjptrB,fjptrC,fjptrD,
2321                                                       fjx0,fjy0,fjz0,fjx1,fjy1,fjz1,
2322                                                       fjx2,fjy2,fjz2,fjx3,fjy3,fjz3);
2323
2324             /* Inner loop uses 303 flops */
2325         }
2326
2327         /* End of innermost loop */
2328
2329         gmx_mm256_update_iforce_4atom_swizzle_pd(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,fix3,fiy3,fiz3,
2330                                                  f+i_coord_offset,fshift+i_shift_offset);
2331
2332         /* Increment number of inner iterations */
2333         inneriter                  += j_index_end - j_index_start;
2334
2335         /* Outer loop uses 24 flops */
2336     }
2337
2338     /* Increment number of outer iterations */
2339     outeriter        += nri;
2340
2341     /* Update outer/inner flops */
2342
2343     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W4W4_F,outeriter*24 + inneriter*303);
2344 }