src/gromacs/gmxlib/nonbonded/nb_kernel_avx_256_double/nb_kernel_ElecRF_VdwCSTab_GeomW4W4_avx_256_double.c

   1 /*
   2  * Note: this file was generated by the Gromacs avx_256_double kernel generator.
   3  *
   4  *                This source code is part of
   5  *
   6  *                 G   R   O   M   A   C   S
   7  *
   8  * Copyright (c) 2001-2012, The GROMACS Development Team
   9  *
  10  * Gromacs is a library for molecular simulation and trajectory analysis,
  11  * written by Erik Lindahl, David van der Spoel, Berk Hess, and others - for
  12  * a full list of developers and information, check out http://www.gromacs.org
  13  *
  14  * This program is free software; you can redistribute it and/or modify it under
  15  * the terms of the GNU Lesser General Public License as published by the Free
  16  * Software Foundation; either version 2 of the License, or (at your option) any
  17  * later version.
  18  *
  19  * To help fund GROMACS development, we humbly ask that you cite
  20  * the papers people have written on it - you can find them on the website.
  21  */
  22 #ifdef HAVE_CONFIG_H
  23 #include <config.h>
  24 #endif
  25
  26 #include <math.h>
  27
  28 #include "../nb_kernel.h"
  29 #include "types/simple.h"
  30 #include "vec.h"
  31 #include "nrnb.h"
  32
  33 #include "gmx_math_x86_avx_256_double.h"
  34 #include "kernelutil_x86_avx_256_double.h"
  35
  36 /*
  37  * Gromacs nonbonded kernel:   nb_kernel_ElecRF_VdwCSTab_GeomW4W4_VF_avx_256_double
  38  * Electrostatics interaction: ReactionField
  39  * VdW interaction:            CubicSplineTable
  40  * Geometry:                   Water4-Water4
  41  * Calculate force/pot:        PotentialAndForce
  42  */
  43 void
  44 nb_kernel_ElecRF_VdwCSTab_GeomW4W4_VF_avx_256_double
  45                     (t_nblist * gmx_restrict                nlist,
  46                      rvec * gmx_restrict                    xx,
  47                      rvec * gmx_restrict                    ff,
  48                      t_forcerec * gmx_restrict              fr,
  49                      t_mdatoms * gmx_restrict               mdatoms,
  50                      nb_kernel_data_t * gmx_restrict        kernel_data,
  51                      t_nrnb * gmx_restrict                  nrnb)
  52 {
  53     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
  54      * just 0 for non-waters.
  55      * Suffixes A,B,C,D refer to j loop unrolling done with AVX, e.g. for the four different
  56      * jnr indices corresponding to data put in the four positions in the SIMD register.
  57      */
  58     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
  59     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
  60     int              jnrA,jnrB,jnrC,jnrD;
  61     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
  62     int              jnrlistE,jnrlistF,jnrlistG,jnrlistH;
  63     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
  64     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
  65     real             rcutoff_scalar;
  66     real             *shiftvec,*fshift,*x,*f;
  67     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD;
  68     real             scratch[4*DIM];
  69     __m256d          tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
  70     real *           vdwioffsetptr0;
  71     __m256d          ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
  72     real *           vdwioffsetptr1;
  73     __m256d          ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
  74     real *           vdwioffsetptr2;
  75     __m256d          ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
  76     real *           vdwioffsetptr3;
  77     __m256d          ix3,iy3,iz3,fix3,fiy3,fiz3,iq3,isai3;
  78     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
  79     __m256d          jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
  80     int              vdwjidx1A,vdwjidx1B,vdwjidx1C,vdwjidx1D;
  81     __m256d          jx1,jy1,jz1,fjx1,fjy1,fjz1,jq1,isaj1;
  82     int              vdwjidx2A,vdwjidx2B,vdwjidx2C,vdwjidx2D;
  83     __m256d          jx2,jy2,jz2,fjx2,fjy2,fjz2,jq2,isaj2;
  84     int              vdwjidx3A,vdwjidx3B,vdwjidx3C,vdwjidx3D;
  85     __m256d          jx3,jy3,jz3,fjx3,fjy3,fjz3,jq3,isaj3;
  86     __m256d          dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
  87     __m256d          dx11,dy11,dz11,rsq11,rinv11,rinvsq11,r11,qq11,c6_11,c12_11;
  88     __m256d          dx12,dy12,dz12,rsq12,rinv12,rinvsq12,r12,qq12,c6_12,c12_12;
  89     __m256d          dx13,dy13,dz13,rsq13,rinv13,rinvsq13,r13,qq13,c6_13,c12_13;
  90     __m256d          dx21,dy21,dz21,rsq21,rinv21,rinvsq21,r21,qq21,c6_21,c12_21;
  91     __m256d          dx22,dy22,dz22,rsq22,rinv22,rinvsq22,r22,qq22,c6_22,c12_22;
  92     __m256d          dx23,dy23,dz23,rsq23,rinv23,rinvsq23,r23,qq23,c6_23,c12_23;
  93     __m256d          dx31,dy31,dz31,rsq31,rinv31,rinvsq31,r31,qq31,c6_31,c12_31;
  94     __m256d          dx32,dy32,dz32,rsq32,rinv32,rinvsq32,r32,qq32,c6_32,c12_32;
  95     __m256d          dx33,dy33,dz33,rsq33,rinv33,rinvsq33,r33,qq33,c6_33,c12_33;
  96     __m256d          velec,felec,velecsum,facel,crf,krf,krf2;
  97     real             *charge;
  98     int              nvdwtype;
  99     __m256d          rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
 100     int              *vdwtype;
 101     real             *vdwparam;
 102     __m256d          one_sixth   = _mm256_set1_pd(1.0/6.0);
 103     __m256d          one_twelfth = _mm256_set1_pd(1.0/12.0);
 104     __m128i          vfitab;
 105     __m128i          ifour       = _mm_set1_epi32(4);
 106     __m256d          rt,vfeps,vftabscale,Y,F,G,H,Heps,Fp,VV,FF;
 107     real             *vftab;
 108     __m256d          dummy_mask,cutoff_mask;
 109     __m128           tmpmask0,tmpmask1;
 110     __m256d          signbit = _mm256_castsi256_pd( _mm256_set1_epi32(0x80000000) );
 111     __m256d          one     = _mm256_set1_pd(1.0);
 112     __m256d          two     = _mm256_set1_pd(2.0);
 113     x                = xx[0];
 114     f                = ff[0];
 115
 116     nri              = nlist->nri;
 117     iinr             = nlist->iinr;
 118     jindex           = nlist->jindex;
 119     jjnr             = nlist->jjnr;
 120     shiftidx         = nlist->shift;
 121     gid              = nlist->gid;
 122     shiftvec         = fr->shift_vec[0];
 123     fshift           = fr->fshift[0];
 124     facel            = _mm256_set1_pd(fr->epsfac);
 125     charge           = mdatoms->chargeA;
 126     krf              = _mm256_set1_pd(fr->ic->k_rf);
 127     krf2             = _mm256_set1_pd(fr->ic->k_rf*2.0);
 128     crf              = _mm256_set1_pd(fr->ic->c_rf);
 129     nvdwtype         = fr->ntype;
 130     vdwparam         = fr->nbfp;
 131     vdwtype          = mdatoms->typeA;
 132
 133     vftab            = kernel_data->table_vdw->data;
 134     vftabscale       = _mm256_set1_pd(kernel_data->table_vdw->scale);
 135
 136     /* Setup water-specific parameters */
 137     inr              = nlist->iinr[0];
 138     iq1              = _mm256_mul_pd(facel,_mm256_set1_pd(charge[inr+1]));
 139     iq2              = _mm256_mul_pd(facel,_mm256_set1_pd(charge[inr+2]));
 140     iq3              = _mm256_mul_pd(facel,_mm256_set1_pd(charge[inr+3]));
 141     vdwioffsetptr0   = vdwparam+2*nvdwtype*vdwtype[inr+0];
 142
 143     jq1              = _mm256_set1_pd(charge[inr+1]);
 144     jq2              = _mm256_set1_pd(charge[inr+2]);
 145     jq3              = _mm256_set1_pd(charge[inr+3]);
 146     vdwjidx0A        = 2*vdwtype[inr+0];
 147     c6_00            = _mm256_set1_pd(vdwioffsetptr0[vdwjidx0A]);
 148     c12_00           = _mm256_set1_pd(vdwioffsetptr0[vdwjidx0A+1]);
 149     qq11             = _mm256_mul_pd(iq1,jq1);
 150     qq12             = _mm256_mul_pd(iq1,jq2);
 151     qq13             = _mm256_mul_pd(iq1,jq3);
 152     qq21             = _mm256_mul_pd(iq2,jq1);
 153     qq22             = _mm256_mul_pd(iq2,jq2);
 154     qq23             = _mm256_mul_pd(iq2,jq3);
 155     qq31             = _mm256_mul_pd(iq3,jq1);
 156     qq32             = _mm256_mul_pd(iq3,jq2);
 157     qq33             = _mm256_mul_pd(iq3,jq3);
 158
 159     /* Avoid stupid compiler warnings */
 160     jnrA = jnrB = jnrC = jnrD = 0;
 161     j_coord_offsetA = 0;
 162     j_coord_offsetB = 0;
 163     j_coord_offsetC = 0;
 164     j_coord_offsetD = 0;
 165
 166     outeriter        = 0;
 167     inneriter        = 0;
 168
 169     for(iidx=0;iidx<4*DIM;iidx++)
 170     {
 171         scratch[iidx] = 0.0;
 172     }
 173
 174     /* Start outer loop over neighborlists */
 175     for(iidx=0; iidx<nri; iidx++)
 176     {
 177         /* Load shift vector for this list */
 178         i_shift_offset   = DIM*shiftidx[iidx];
 179
 180         /* Load limits for loop over neighbors */
 181         j_index_start    = jindex[iidx];
 182         j_index_end      = jindex[iidx+1];
 183
 184         /* Get outer coordinate index */
 185         inr              = iinr[iidx];
 186         i_coord_offset   = DIM*inr;
 187
 188         /* Load i particle coords and add shift vector */
 189         gmx_mm256_load_shift_and_4rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset,
 190                                                     &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2,&ix3,&iy3,&iz3);
 191
 192         fix0             = _mm256_setzero_pd();
 193         fiy0             = _mm256_setzero_pd();
 194         fiz0             = _mm256_setzero_pd();
 195         fix1             = _mm256_setzero_pd();
 196         fiy1             = _mm256_setzero_pd();
 197         fiz1             = _mm256_setzero_pd();
 198         fix2             = _mm256_setzero_pd();
 199         fiy2             = _mm256_setzero_pd();
 200         fiz2             = _mm256_setzero_pd();
 201         fix3             = _mm256_setzero_pd();
 202         fiy3             = _mm256_setzero_pd();
 203         fiz3             = _mm256_setzero_pd();
 204
 205         /* Reset potential sums */
 206         velecsum         = _mm256_setzero_pd();
 207         vvdwsum          = _mm256_setzero_pd();
 208
 209         /* Start inner kernel loop */
 210         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
 211         {
 212
 213             /* Get j neighbor index, and coordinate index */
 214             jnrA             = jjnr[jidx];
 215             jnrB             = jjnr[jidx+1];
 216             jnrC             = jjnr[jidx+2];
 217             jnrD             = jjnr[jidx+3];
 218             j_coord_offsetA  = DIM*jnrA;
 219             j_coord_offsetB  = DIM*jnrB;
 220             j_coord_offsetC  = DIM*jnrC;
 221             j_coord_offsetD  = DIM*jnrD;
 222
 223             /* load j atom coordinates */
 224             gmx_mm256_load_4rvec_4ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 225                                                  x+j_coord_offsetC,x+j_coord_offsetD,
 226                                                  &jx0,&jy0,&jz0,&jx1,&jy1,&jz1,&jx2,
 227                                                  &jy2,&jz2,&jx3,&jy3,&jz3);
 228
 229             /* Calculate displacement vector */
 230             dx00             = _mm256_sub_pd(ix0,jx0);
 231             dy00             = _mm256_sub_pd(iy0,jy0);
 232             dz00             = _mm256_sub_pd(iz0,jz0);
 233             dx11             = _mm256_sub_pd(ix1,jx1);
 234             dy11             = _mm256_sub_pd(iy1,jy1);
 235             dz11             = _mm256_sub_pd(iz1,jz1);
 236             dx12             = _mm256_sub_pd(ix1,jx2);
 237             dy12             = _mm256_sub_pd(iy1,jy2);
 238             dz12             = _mm256_sub_pd(iz1,jz2);
 239             dx13             = _mm256_sub_pd(ix1,jx3);
 240             dy13             = _mm256_sub_pd(iy1,jy3);
 241             dz13             = _mm256_sub_pd(iz1,jz3);
 242             dx21             = _mm256_sub_pd(ix2,jx1);
 243             dy21             = _mm256_sub_pd(iy2,jy1);
 244             dz21             = _mm256_sub_pd(iz2,jz1);
 245             dx22             = _mm256_sub_pd(ix2,jx2);
 246             dy22             = _mm256_sub_pd(iy2,jy2);
 247             dz22             = _mm256_sub_pd(iz2,jz2);
 248             dx23             = _mm256_sub_pd(ix2,jx3);
 249             dy23             = _mm256_sub_pd(iy2,jy3);
 250             dz23             = _mm256_sub_pd(iz2,jz3);
 251             dx31             = _mm256_sub_pd(ix3,jx1);
 252             dy31             = _mm256_sub_pd(iy3,jy1);
 253             dz31             = _mm256_sub_pd(iz3,jz1);
 254             dx32             = _mm256_sub_pd(ix3,jx2);
 255             dy32             = _mm256_sub_pd(iy3,jy2);
 256             dz32             = _mm256_sub_pd(iz3,jz2);
 257             dx33             = _mm256_sub_pd(ix3,jx3);
 258             dy33             = _mm256_sub_pd(iy3,jy3);
 259             dz33             = _mm256_sub_pd(iz3,jz3);
 260
 261             /* Calculate squared distance and things based on it */
 262             rsq00            = gmx_mm256_calc_rsq_pd(dx00,dy00,dz00);
 263             rsq11            = gmx_mm256_calc_rsq_pd(dx11,dy11,dz11);
 264             rsq12            = gmx_mm256_calc_rsq_pd(dx12,dy12,dz12);
 265             rsq13            = gmx_mm256_calc_rsq_pd(dx13,dy13,dz13);
 266             rsq21            = gmx_mm256_calc_rsq_pd(dx21,dy21,dz21);
 267             rsq22            = gmx_mm256_calc_rsq_pd(dx22,dy22,dz22);
 268             rsq23            = gmx_mm256_calc_rsq_pd(dx23,dy23,dz23);
 269             rsq31            = gmx_mm256_calc_rsq_pd(dx31,dy31,dz31);
 270             rsq32            = gmx_mm256_calc_rsq_pd(dx32,dy32,dz32);
 271             rsq33            = gmx_mm256_calc_rsq_pd(dx33,dy33,dz33);
 272
 273             rinv00           = gmx_mm256_invsqrt_pd(rsq00);
 274             rinv11           = gmx_mm256_invsqrt_pd(rsq11);
 275             rinv12           = gmx_mm256_invsqrt_pd(rsq12);
 276             rinv13           = gmx_mm256_invsqrt_pd(rsq13);
 277             rinv21           = gmx_mm256_invsqrt_pd(rsq21);
 278             rinv22           = gmx_mm256_invsqrt_pd(rsq22);
 279             rinv23           = gmx_mm256_invsqrt_pd(rsq23);
 280             rinv31           = gmx_mm256_invsqrt_pd(rsq31);
 281             rinv32           = gmx_mm256_invsqrt_pd(rsq32);
 282             rinv33           = gmx_mm256_invsqrt_pd(rsq33);
 283
 284             rinvsq11         = _mm256_mul_pd(rinv11,rinv11);
 285             rinvsq12         = _mm256_mul_pd(rinv12,rinv12);
 286             rinvsq13         = _mm256_mul_pd(rinv13,rinv13);
 287             rinvsq21         = _mm256_mul_pd(rinv21,rinv21);
 288             rinvsq22         = _mm256_mul_pd(rinv22,rinv22);
 289             rinvsq23         = _mm256_mul_pd(rinv23,rinv23);
 290             rinvsq31         = _mm256_mul_pd(rinv31,rinv31);
 291             rinvsq32         = _mm256_mul_pd(rinv32,rinv32);
 292             rinvsq33         = _mm256_mul_pd(rinv33,rinv33);
 293
 294             fjx0             = _mm256_setzero_pd();
 295             fjy0             = _mm256_setzero_pd();
 296             fjz0             = _mm256_setzero_pd();
 297             fjx1             = _mm256_setzero_pd();
 298             fjy1             = _mm256_setzero_pd();
 299             fjz1             = _mm256_setzero_pd();
 300             fjx2             = _mm256_setzero_pd();
 301             fjy2             = _mm256_setzero_pd();
 302             fjz2             = _mm256_setzero_pd();
 303             fjx3             = _mm256_setzero_pd();
 304             fjy3             = _mm256_setzero_pd();
 305             fjz3             = _mm256_setzero_pd();
 306
 307             /**************************
 308              * CALCULATE INTERACTIONS *
 309              **************************/
 310
 311             r00              = _mm256_mul_pd(rsq00,rinv00);
 312
 313             /* Calculate table index by multiplying r with table scale and truncate to integer */
 314             rt               = _mm256_mul_pd(r00,vftabscale);
 315             vfitab           = _mm256_cvttpd_epi32(rt);
 316             vfeps            = _mm256_sub_pd(rt,_mm256_round_pd(rt, _MM_FROUND_FLOOR));
 317             vfitab           = _mm_slli_epi32(vfitab,3);
 318
 319             /* CUBIC SPLINE TABLE DISPERSION */
 320             Y                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
 321             F                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
 322             G                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,2) );
 323             H                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,3) );
 324             GMX_MM256_FULLTRANSPOSE4_PD(Y,F,G,H);
 325             Heps             = _mm256_mul_pd(vfeps,H);
 326             Fp               = _mm256_add_pd(F,_mm256_mul_pd(vfeps,_mm256_add_pd(G,Heps)));
 327             VV               = _mm256_add_pd(Y,_mm256_mul_pd(vfeps,Fp));
 328             vvdw6            = _mm256_mul_pd(c6_00,VV);
 329             FF               = _mm256_add_pd(Fp,_mm256_mul_pd(vfeps,_mm256_add_pd(G,_mm256_add_pd(Heps,Heps))));
 330             fvdw6            = _mm256_mul_pd(c6_00,FF);
 331
 332             /* CUBIC SPLINE TABLE REPULSION */
 333             vfitab           = _mm_add_epi32(vfitab,ifour);
 334             Y                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
 335             F                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
 336             G                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,2) );
 337             H                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,3) );
 338             GMX_MM256_FULLTRANSPOSE4_PD(Y,F,G,H);
 339             Heps             = _mm256_mul_pd(vfeps,H);
 340             Fp               = _mm256_add_pd(F,_mm256_mul_pd(vfeps,_mm256_add_pd(G,Heps)));
 341             VV               = _mm256_add_pd(Y,_mm256_mul_pd(vfeps,Fp));
 342             vvdw12           = _mm256_mul_pd(c12_00,VV);
 343             FF               = _mm256_add_pd(Fp,_mm256_mul_pd(vfeps,_mm256_add_pd(G,_mm256_add_pd(Heps,Heps))));
 344             fvdw12           = _mm256_mul_pd(c12_00,FF);
 345             vvdw             = _mm256_add_pd(vvdw12,vvdw6);
 346             fvdw             = _mm256_xor_pd(signbit,_mm256_mul_pd(_mm256_add_pd(fvdw6,fvdw12),_mm256_mul_pd(vftabscale,rinv00)));
 347
 348             /* Update potential sum for this i atom from the interaction with this j atom. */
 349             vvdwsum          = _mm256_add_pd(vvdwsum,vvdw);
 350
 351             fscal            = fvdw;
 352
 353             /* Calculate temporary vectorial force */
 354             tx               = _mm256_mul_pd(fscal,dx00);
 355             ty               = _mm256_mul_pd(fscal,dy00);
 356             tz               = _mm256_mul_pd(fscal,dz00);
 357
 358             /* Update vectorial force */
 359             fix0             = _mm256_add_pd(fix0,tx);
 360             fiy0             = _mm256_add_pd(fiy0,ty);
 361             fiz0             = _mm256_add_pd(fiz0,tz);
 362
 363             fjx0             = _mm256_add_pd(fjx0,tx);
 364             fjy0             = _mm256_add_pd(fjy0,ty);
 365             fjz0             = _mm256_add_pd(fjz0,tz);
 366
 367             /**************************
 368              * CALCULATE INTERACTIONS *
 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             /* Update potential sum for this i atom from the interaction with this j atom. */
 376             velecsum         = _mm256_add_pd(velecsum,velec);
 377
 378             fscal            = felec;
 379
 380             /* Calculate temporary vectorial force */
 381             tx               = _mm256_mul_pd(fscal,dx11);
 382             ty               = _mm256_mul_pd(fscal,dy11);
 383             tz               = _mm256_mul_pd(fscal,dz11);
 384
 385             /* Update vectorial force */
 386             fix1             = _mm256_add_pd(fix1,tx);
 387             fiy1             = _mm256_add_pd(fiy1,ty);
 388             fiz1             = _mm256_add_pd(fiz1,tz);
 389
 390             fjx1             = _mm256_add_pd(fjx1,tx);
 391             fjy1             = _mm256_add_pd(fjy1,ty);
 392             fjz1             = _mm256_add_pd(fjz1,tz);
 393
 394             /**************************
 395              * CALCULATE INTERACTIONS *
 396              **************************/
 397
 398             /* REACTION-FIELD ELECTROSTATICS */
 399             velec            = _mm256_mul_pd(qq12,_mm256_sub_pd(_mm256_add_pd(rinv12,_mm256_mul_pd(krf,rsq12)),crf));
 400             felec            = _mm256_mul_pd(qq12,_mm256_sub_pd(_mm256_mul_pd(rinv12,rinvsq12),krf2));
 401
 402             /* Update potential sum for this i atom from the interaction with this j atom. */
 403             velecsum         = _mm256_add_pd(velecsum,velec);
 404
 405             fscal            = felec;
 406
 407             /* Calculate temporary vectorial force */
 408             tx               = _mm256_mul_pd(fscal,dx12);
 409             ty               = _mm256_mul_pd(fscal,dy12);
 410             tz               = _mm256_mul_pd(fscal,dz12);
 411
 412             /* Update vectorial force */
 413             fix1             = _mm256_add_pd(fix1,tx);
 414             fiy1             = _mm256_add_pd(fiy1,ty);
 415             fiz1             = _mm256_add_pd(fiz1,tz);
 416
 417             fjx2             = _mm256_add_pd(fjx2,tx);
 418             fjy2             = _mm256_add_pd(fjy2,ty);
 419             fjz2             = _mm256_add_pd(fjz2,tz);
 420
 421             /**************************
 422              * CALCULATE INTERACTIONS *
 423              **************************/
 424
 425             /* REACTION-FIELD ELECTROSTATICS */
 426             velec            = _mm256_mul_pd(qq13,_mm256_sub_pd(_mm256_add_pd(rinv13,_mm256_mul_pd(krf,rsq13)),crf));
 427             felec            = _mm256_mul_pd(qq13,_mm256_sub_pd(_mm256_mul_pd(rinv13,rinvsq13),krf2));
 428
 429             /* Update potential sum for this i atom from the interaction with this j atom. */
 430             velecsum         = _mm256_add_pd(velecsum,velec);
 431
 432             fscal            = felec;
 433
 434             /* Calculate temporary vectorial force */
 435             tx               = _mm256_mul_pd(fscal,dx13);
 436             ty               = _mm256_mul_pd(fscal,dy13);
 437             tz               = _mm256_mul_pd(fscal,dz13);
 438
 439             /* Update vectorial force */
 440             fix1             = _mm256_add_pd(fix1,tx);
 441             fiy1             = _mm256_add_pd(fiy1,ty);
 442             fiz1             = _mm256_add_pd(fiz1,tz);
 443
 444             fjx3             = _mm256_add_pd(fjx3,tx);
 445             fjy3             = _mm256_add_pd(fjy3,ty);
 446             fjz3             = _mm256_add_pd(fjz3,tz);
 447
 448             /**************************
 449              * CALCULATE INTERACTIONS *
 450              **************************/
 451
 452             /* REACTION-FIELD ELECTROSTATICS */
 453             velec            = _mm256_mul_pd(qq21,_mm256_sub_pd(_mm256_add_pd(rinv21,_mm256_mul_pd(krf,rsq21)),crf));
 454             felec            = _mm256_mul_pd(qq21,_mm256_sub_pd(_mm256_mul_pd(rinv21,rinvsq21),krf2));
 455
 456             /* Update potential sum for this i atom from the interaction with this j atom. */
 457             velecsum         = _mm256_add_pd(velecsum,velec);
 458
 459             fscal            = felec;
 460
 461             /* Calculate temporary vectorial force */
 462             tx               = _mm256_mul_pd(fscal,dx21);
 463             ty               = _mm256_mul_pd(fscal,dy21);
 464             tz               = _mm256_mul_pd(fscal,dz21);
 465
 466             /* Update vectorial force */
 467             fix2             = _mm256_add_pd(fix2,tx);
 468             fiy2             = _mm256_add_pd(fiy2,ty);
 469             fiz2             = _mm256_add_pd(fiz2,tz);
 470
 471             fjx1             = _mm256_add_pd(fjx1,tx);
 472             fjy1             = _mm256_add_pd(fjy1,ty);
 473             fjz1             = _mm256_add_pd(fjz1,tz);
 474
 475             /**************************
 476              * CALCULATE INTERACTIONS *
 477              **************************/
 478
 479             /* REACTION-FIELD ELECTROSTATICS */
 480             velec            = _mm256_mul_pd(qq22,_mm256_sub_pd(_mm256_add_pd(rinv22,_mm256_mul_pd(krf,rsq22)),crf));
 481             felec            = _mm256_mul_pd(qq22,_mm256_sub_pd(_mm256_mul_pd(rinv22,rinvsq22),krf2));
 482
 483             /* Update potential sum for this i atom from the interaction with this j atom. */
 484             velecsum         = _mm256_add_pd(velecsum,velec);
 485
 486             fscal            = felec;
 487
 488             /* Calculate temporary vectorial force */
 489             tx               = _mm256_mul_pd(fscal,dx22);
 490             ty               = _mm256_mul_pd(fscal,dy22);
 491             tz               = _mm256_mul_pd(fscal,dz22);
 492
 493             /* Update vectorial force */
 494             fix2             = _mm256_add_pd(fix2,tx);
 495             fiy2             = _mm256_add_pd(fiy2,ty);
 496             fiz2             = _mm256_add_pd(fiz2,tz);
 497
 498             fjx2             = _mm256_add_pd(fjx2,tx);
 499             fjy2             = _mm256_add_pd(fjy2,ty);
 500             fjz2             = _mm256_add_pd(fjz2,tz);
 501
 502             /**************************
 503              * CALCULATE INTERACTIONS *
 504              **************************/
 505
 506             /* REACTION-FIELD ELECTROSTATICS */
 507             velec            = _mm256_mul_pd(qq23,_mm256_sub_pd(_mm256_add_pd(rinv23,_mm256_mul_pd(krf,rsq23)),crf));
 508             felec            = _mm256_mul_pd(qq23,_mm256_sub_pd(_mm256_mul_pd(rinv23,rinvsq23),krf2));
 509
 510             /* Update potential sum for this i atom from the interaction with this j atom. */
 511             velecsum         = _mm256_add_pd(velecsum,velec);
 512
 513             fscal            = felec;
 514
 515             /* Calculate temporary vectorial force */
 516             tx               = _mm256_mul_pd(fscal,dx23);
 517             ty               = _mm256_mul_pd(fscal,dy23);
 518             tz               = _mm256_mul_pd(fscal,dz23);
 519
 520             /* Update vectorial force */
 521             fix2             = _mm256_add_pd(fix2,tx);
 522             fiy2             = _mm256_add_pd(fiy2,ty);
 523             fiz2             = _mm256_add_pd(fiz2,tz);
 524
 525             fjx3             = _mm256_add_pd(fjx3,tx);
 526             fjy3             = _mm256_add_pd(fjy3,ty);
 527             fjz3             = _mm256_add_pd(fjz3,tz);
 528
 529             /**************************
 530              * CALCULATE INTERACTIONS *
 531              **************************/
 532
 533             /* REACTION-FIELD ELECTROSTATICS */
 534             velec            = _mm256_mul_pd(qq31,_mm256_sub_pd(_mm256_add_pd(rinv31,_mm256_mul_pd(krf,rsq31)),crf));
 535             felec            = _mm256_mul_pd(qq31,_mm256_sub_pd(_mm256_mul_pd(rinv31,rinvsq31),krf2));
 536
 537             /* Update potential sum for this i atom from the interaction with this j atom. */
 538             velecsum         = _mm256_add_pd(velecsum,velec);
 539
 540             fscal            = felec;
 541
 542             /* Calculate temporary vectorial force */
 543             tx               = _mm256_mul_pd(fscal,dx31);
 544             ty               = _mm256_mul_pd(fscal,dy31);
 545             tz               = _mm256_mul_pd(fscal,dz31);
 546
 547             /* Update vectorial force */
 548             fix3             = _mm256_add_pd(fix3,tx);
 549             fiy3             = _mm256_add_pd(fiy3,ty);
 550             fiz3             = _mm256_add_pd(fiz3,tz);
 551
 552             fjx1             = _mm256_add_pd(fjx1,tx);
 553             fjy1             = _mm256_add_pd(fjy1,ty);
 554             fjz1             = _mm256_add_pd(fjz1,tz);
 555
 556             /**************************
 557              * CALCULATE INTERACTIONS *
 558              **************************/
 559
 560             /* REACTION-FIELD ELECTROSTATICS */
 561             velec            = _mm256_mul_pd(qq32,_mm256_sub_pd(_mm256_add_pd(rinv32,_mm256_mul_pd(krf,rsq32)),crf));
 562             felec            = _mm256_mul_pd(qq32,_mm256_sub_pd(_mm256_mul_pd(rinv32,rinvsq32),krf2));
 563
 564             /* Update potential sum for this i atom from the interaction with this j atom. */
 565             velecsum         = _mm256_add_pd(velecsum,velec);
 566
 567             fscal            = felec;
 568
 569             /* Calculate temporary vectorial force */
 570             tx               = _mm256_mul_pd(fscal,dx32);
 571             ty               = _mm256_mul_pd(fscal,dy32);
 572             tz               = _mm256_mul_pd(fscal,dz32);
 573
 574             /* Update vectorial force */
 575             fix3             = _mm256_add_pd(fix3,tx);
 576             fiy3             = _mm256_add_pd(fiy3,ty);
 577             fiz3             = _mm256_add_pd(fiz3,tz);
 578
 579             fjx2             = _mm256_add_pd(fjx2,tx);
 580             fjy2             = _mm256_add_pd(fjy2,ty);
 581             fjz2             = _mm256_add_pd(fjz2,tz);
 582
 583             /**************************
 584              * CALCULATE INTERACTIONS *
 585              **************************/
 586
 587             /* REACTION-FIELD ELECTROSTATICS */
 588             velec            = _mm256_mul_pd(qq33,_mm256_sub_pd(_mm256_add_pd(rinv33,_mm256_mul_pd(krf,rsq33)),crf));
 589             felec            = _mm256_mul_pd(qq33,_mm256_sub_pd(_mm256_mul_pd(rinv33,rinvsq33),krf2));
 590
 591             /* Update potential sum for this i atom from the interaction with this j atom. */
 592             velecsum         = _mm256_add_pd(velecsum,velec);
 593
 594             fscal            = felec;
 595
 596             /* Calculate temporary vectorial force */
 597             tx               = _mm256_mul_pd(fscal,dx33);
 598             ty               = _mm256_mul_pd(fscal,dy33);
 599             tz               = _mm256_mul_pd(fscal,dz33);
 600
 601             /* Update vectorial force */
 602             fix3             = _mm256_add_pd(fix3,tx);
 603             fiy3             = _mm256_add_pd(fiy3,ty);
 604             fiz3             = _mm256_add_pd(fiz3,tz);
 605
 606             fjx3             = _mm256_add_pd(fjx3,tx);
 607             fjy3             = _mm256_add_pd(fjy3,ty);
 608             fjz3             = _mm256_add_pd(fjz3,tz);
 609
 610             fjptrA             = f+j_coord_offsetA;
 611             fjptrB             = f+j_coord_offsetB;
 612             fjptrC             = f+j_coord_offsetC;
 613             fjptrD             = f+j_coord_offsetD;
 614
 615             gmx_mm256_decrement_4rvec_4ptr_swizzle_pd(fjptrA,fjptrB,fjptrC,fjptrD,
 616                                                       fjx0,fjy0,fjz0,fjx1,fjy1,fjz1,
 617                                                       fjx2,fjy2,fjz2,fjx3,fjy3,fjz3);
 618
 619             /* Inner loop uses 347 flops */
 620         }
 621
 622         if(jidx<j_index_end)
 623         {
 624
 625             /* Get j neighbor index, and coordinate index */
 626             jnrlistA         = jjnr[jidx];
 627             jnrlistB         = jjnr[jidx+1];
 628             jnrlistC         = jjnr[jidx+2];
 629             jnrlistD         = jjnr[jidx+3];
 630             /* Sign of each element will be negative for non-real atoms.
 631              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 632              * so use it as val = _mm_andnot_pd(mask,val) to clear dummy entries.
 633              */
 634             tmpmask0 = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
 635
 636             tmpmask1 = _mm_permute_ps(tmpmask0,_GMX_MM_PERMUTE(3,3,2,2));
 637             tmpmask0 = _mm_permute_ps(tmpmask0,_GMX_MM_PERMUTE(1,1,0,0));
 638             dummy_mask = _mm256_castps_pd(gmx_mm256_set_m128(tmpmask1,tmpmask0));
 639
 640             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
 641             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
 642             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
 643             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
 644             j_coord_offsetA  = DIM*jnrA;
 645             j_coord_offsetB  = DIM*jnrB;
 646             j_coord_offsetC  = DIM*jnrC;
 647             j_coord_offsetD  = DIM*jnrD;
 648
 649             /* load j atom coordinates */
 650             gmx_mm256_load_4rvec_4ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 651                                                  x+j_coord_offsetC,x+j_coord_offsetD,
 652                                                  &jx0,&jy0,&jz0,&jx1,&jy1,&jz1,&jx2,
 653                                                  &jy2,&jz2,&jx3,&jy3,&jz3);
 654
 655             /* Calculate displacement vector */
 656             dx00             = _mm256_sub_pd(ix0,jx0);
 657             dy00             = _mm256_sub_pd(iy0,jy0);
 658             dz00             = _mm256_sub_pd(iz0,jz0);
 659             dx11             = _mm256_sub_pd(ix1,jx1);
 660             dy11             = _mm256_sub_pd(iy1,jy1);
 661             dz11             = _mm256_sub_pd(iz1,jz1);
 662             dx12             = _mm256_sub_pd(ix1,jx2);
 663             dy12             = _mm256_sub_pd(iy1,jy2);
 664             dz12             = _mm256_sub_pd(iz1,jz2);
 665             dx13             = _mm256_sub_pd(ix1,jx3);
 666             dy13             = _mm256_sub_pd(iy1,jy3);
 667             dz13             = _mm256_sub_pd(iz1,jz3);
 668             dx21             = _mm256_sub_pd(ix2,jx1);
 669             dy21             = _mm256_sub_pd(iy2,jy1);
 670             dz21             = _mm256_sub_pd(iz2,jz1);
 671             dx22             = _mm256_sub_pd(ix2,jx2);
 672             dy22             = _mm256_sub_pd(iy2,jy2);
 673             dz22             = _mm256_sub_pd(iz2,jz2);
 674             dx23             = _mm256_sub_pd(ix2,jx3);
 675             dy23             = _mm256_sub_pd(iy2,jy3);
 676             dz23             = _mm256_sub_pd(iz2,jz3);
 677             dx31             = _mm256_sub_pd(ix3,jx1);
 678             dy31             = _mm256_sub_pd(iy3,jy1);
 679             dz31             = _mm256_sub_pd(iz3,jz1);
 680             dx32             = _mm256_sub_pd(ix3,jx2);
 681             dy32             = _mm256_sub_pd(iy3,jy2);
 682             dz32             = _mm256_sub_pd(iz3,jz2);
 683             dx33             = _mm256_sub_pd(ix3,jx3);
 684             dy33             = _mm256_sub_pd(iy3,jy3);
 685             dz33             = _mm256_sub_pd(iz3,jz3);
 686
 687             /* Calculate squared distance and things based on it */
 688             rsq00            = gmx_mm256_calc_rsq_pd(dx00,dy00,dz00);
 689             rsq11            = gmx_mm256_calc_rsq_pd(dx11,dy11,dz11);
 690             rsq12            = gmx_mm256_calc_rsq_pd(dx12,dy12,dz12);
 691             rsq13            = gmx_mm256_calc_rsq_pd(dx13,dy13,dz13);
 692             rsq21            = gmx_mm256_calc_rsq_pd(dx21,dy21,dz21);
 693             rsq22            = gmx_mm256_calc_rsq_pd(dx22,dy22,dz22);
 694             rsq23            = gmx_mm256_calc_rsq_pd(dx23,dy23,dz23);
 695             rsq31            = gmx_mm256_calc_rsq_pd(dx31,dy31,dz31);
 696             rsq32            = gmx_mm256_calc_rsq_pd(dx32,dy32,dz32);
 697             rsq33            = gmx_mm256_calc_rsq_pd(dx33,dy33,dz33);
 698
 699             rinv00           = gmx_mm256_invsqrt_pd(rsq00);
 700             rinv11           = gmx_mm256_invsqrt_pd(rsq11);
 701             rinv12           = gmx_mm256_invsqrt_pd(rsq12);
 702             rinv13           = gmx_mm256_invsqrt_pd(rsq13);
 703             rinv21           = gmx_mm256_invsqrt_pd(rsq21);
 704             rinv22           = gmx_mm256_invsqrt_pd(rsq22);
 705             rinv23           = gmx_mm256_invsqrt_pd(rsq23);
 706             rinv31           = gmx_mm256_invsqrt_pd(rsq31);
 707             rinv32           = gmx_mm256_invsqrt_pd(rsq32);
 708             rinv33           = gmx_mm256_invsqrt_pd(rsq33);
 709
 710             rinvsq11         = _mm256_mul_pd(rinv11,rinv11);
 711             rinvsq12         = _mm256_mul_pd(rinv12,rinv12);
 712             rinvsq13         = _mm256_mul_pd(rinv13,rinv13);
 713             rinvsq21         = _mm256_mul_pd(rinv21,rinv21);
 714             rinvsq22         = _mm256_mul_pd(rinv22,rinv22);
 715             rinvsq23         = _mm256_mul_pd(rinv23,rinv23);
 716             rinvsq31         = _mm256_mul_pd(rinv31,rinv31);
 717             rinvsq32         = _mm256_mul_pd(rinv32,rinv32);
 718             rinvsq33         = _mm256_mul_pd(rinv33,rinv33);
 719
 720             fjx0             = _mm256_setzero_pd();
 721             fjy0             = _mm256_setzero_pd();
 722             fjz0             = _mm256_setzero_pd();
 723             fjx1             = _mm256_setzero_pd();
 724             fjy1             = _mm256_setzero_pd();
 725             fjz1             = _mm256_setzero_pd();
 726             fjx2             = _mm256_setzero_pd();
 727             fjy2             = _mm256_setzero_pd();
 728             fjz2             = _mm256_setzero_pd();
 729             fjx3             = _mm256_setzero_pd();
 730             fjy3             = _mm256_setzero_pd();
 731             fjz3             = _mm256_setzero_pd();
 732
 733             /**************************
 734              * CALCULATE INTERACTIONS *
 735              **************************/
 736
 737             r00              = _mm256_mul_pd(rsq00,rinv00);
 738             r00              = _mm256_andnot_pd(dummy_mask,r00);
 739
 740             /* Calculate table index by multiplying r with table scale and truncate to integer */
 741             rt               = _mm256_mul_pd(r00,vftabscale);
 742             vfitab           = _mm256_cvttpd_epi32(rt);
 743             vfeps            = _mm256_sub_pd(rt,_mm256_round_pd(rt, _MM_FROUND_FLOOR));
 744             vfitab           = _mm_slli_epi32(vfitab,3);
 745
 746             /* CUBIC SPLINE TABLE DISPERSION */
 747             Y                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
 748             F                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
 749             G                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,2) );
 750             H                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,3) );
 751             GMX_MM256_FULLTRANSPOSE4_PD(Y,F,G,H);
 752             Heps             = _mm256_mul_pd(vfeps,H);
 753             Fp               = _mm256_add_pd(F,_mm256_mul_pd(vfeps,_mm256_add_pd(G,Heps)));
 754             VV               = _mm256_add_pd(Y,_mm256_mul_pd(vfeps,Fp));
 755             vvdw6            = _mm256_mul_pd(c6_00,VV);
 756             FF               = _mm256_add_pd(Fp,_mm256_mul_pd(vfeps,_mm256_add_pd(G,_mm256_add_pd(Heps,Heps))));
 757             fvdw6            = _mm256_mul_pd(c6_00,FF);
 758
 759             /* CUBIC SPLINE TABLE REPULSION */
 760             vfitab           = _mm_add_epi32(vfitab,ifour);
 761             Y                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
 762             F                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
 763             G                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,2) );
 764             H                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,3) );
 765             GMX_MM256_FULLTRANSPOSE4_PD(Y,F,G,H);
 766             Heps             = _mm256_mul_pd(vfeps,H);
 767             Fp               = _mm256_add_pd(F,_mm256_mul_pd(vfeps,_mm256_add_pd(G,Heps)));
 768             VV               = _mm256_add_pd(Y,_mm256_mul_pd(vfeps,Fp));
 769             vvdw12           = _mm256_mul_pd(c12_00,VV);
 770             FF               = _mm256_add_pd(Fp,_mm256_mul_pd(vfeps,_mm256_add_pd(G,_mm256_add_pd(Heps,Heps))));
 771             fvdw12           = _mm256_mul_pd(c12_00,FF);
 772             vvdw             = _mm256_add_pd(vvdw12,vvdw6);
 773             fvdw             = _mm256_xor_pd(signbit,_mm256_mul_pd(_mm256_add_pd(fvdw6,fvdw12),_mm256_mul_pd(vftabscale,rinv00)));
 774
 775             /* Update potential sum for this i atom from the interaction with this j atom. */
 776             vvdw             = _mm256_andnot_pd(dummy_mask,vvdw);
 777             vvdwsum          = _mm256_add_pd(vvdwsum,vvdw);
 778
 779             fscal            = fvdw;
 780
 781             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
 782
 783             /* Calculate temporary vectorial force */
 784             tx               = _mm256_mul_pd(fscal,dx00);
 785             ty               = _mm256_mul_pd(fscal,dy00);
 786             tz               = _mm256_mul_pd(fscal,dz00);
 787
 788             /* Update vectorial force */
 789             fix0             = _mm256_add_pd(fix0,tx);
 790             fiy0             = _mm256_add_pd(fiy0,ty);
 791             fiz0             = _mm256_add_pd(fiz0,tz);
 792
 793             fjx0             = _mm256_add_pd(fjx0,tx);
 794             fjy0             = _mm256_add_pd(fjy0,ty);
 795             fjz0             = _mm256_add_pd(fjz0,tz);
 796
 797             /**************************
 798              * CALCULATE INTERACTIONS *
 799              **************************/
 800
 801             /* REACTION-FIELD ELECTROSTATICS */
 802             velec            = _mm256_mul_pd(qq11,_mm256_sub_pd(_mm256_add_pd(rinv11,_mm256_mul_pd(krf,rsq11)),crf));
 803             felec            = _mm256_mul_pd(qq11,_mm256_sub_pd(_mm256_mul_pd(rinv11,rinvsq11),krf2));
 804
 805             /* Update potential sum for this i atom from the interaction with this j atom. */
 806             velec            = _mm256_andnot_pd(dummy_mask,velec);
 807             velecsum         = _mm256_add_pd(velecsum,velec);
 808
 809             fscal            = felec;
 810
 811             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
 812
 813             /* Calculate temporary vectorial force */
 814             tx               = _mm256_mul_pd(fscal,dx11);
 815             ty               = _mm256_mul_pd(fscal,dy11);
 816             tz               = _mm256_mul_pd(fscal,dz11);
 817
 818             /* Update vectorial force */
 819             fix1             = _mm256_add_pd(fix1,tx);
 820             fiy1             = _mm256_add_pd(fiy1,ty);
 821             fiz1             = _mm256_add_pd(fiz1,tz);
 822
 823             fjx1             = _mm256_add_pd(fjx1,tx);
 824             fjy1             = _mm256_add_pd(fjy1,ty);
 825             fjz1             = _mm256_add_pd(fjz1,tz);
 826
 827             /**************************
 828              * CALCULATE INTERACTIONS *
 829              **************************/
 830
 831             /* REACTION-FIELD ELECTROSTATICS */
 832             velec            = _mm256_mul_pd(qq12,_mm256_sub_pd(_mm256_add_pd(rinv12,_mm256_mul_pd(krf,rsq12)),crf));
 833             felec            = _mm256_mul_pd(qq12,_mm256_sub_pd(_mm256_mul_pd(rinv12,rinvsq12),krf2));
 834
 835             /* Update potential sum for this i atom from the interaction with this j atom. */
 836             velec            = _mm256_andnot_pd(dummy_mask,velec);
 837             velecsum         = _mm256_add_pd(velecsum,velec);
 838
 839             fscal            = felec;
 840
 841             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
 842
 843             /* Calculate temporary vectorial force */
 844             tx               = _mm256_mul_pd(fscal,dx12);
 845             ty               = _mm256_mul_pd(fscal,dy12);
 846             tz               = _mm256_mul_pd(fscal,dz12);
 847
 848             /* Update vectorial force */
 849             fix1             = _mm256_add_pd(fix1,tx);
 850             fiy1             = _mm256_add_pd(fiy1,ty);
 851             fiz1             = _mm256_add_pd(fiz1,tz);
 852
 853             fjx2             = _mm256_add_pd(fjx2,tx);
 854             fjy2             = _mm256_add_pd(fjy2,ty);
 855             fjz2             = _mm256_add_pd(fjz2,tz);
 856
 857             /**************************
 858              * CALCULATE INTERACTIONS *
 859              **************************/
 860
 861             /* REACTION-FIELD ELECTROSTATICS */
 862             velec            = _mm256_mul_pd(qq13,_mm256_sub_pd(_mm256_add_pd(rinv13,_mm256_mul_pd(krf,rsq13)),crf));
 863             felec            = _mm256_mul_pd(qq13,_mm256_sub_pd(_mm256_mul_pd(rinv13,rinvsq13),krf2));
 864
 865             /* Update potential sum for this i atom from the interaction with this j atom. */
 866             velec            = _mm256_andnot_pd(dummy_mask,velec);
 867             velecsum         = _mm256_add_pd(velecsum,velec);
 868
 869             fscal            = felec;
 870
 871             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
 872
 873             /* Calculate temporary vectorial force */
 874             tx               = _mm256_mul_pd(fscal,dx13);
 875             ty               = _mm256_mul_pd(fscal,dy13);
 876             tz               = _mm256_mul_pd(fscal,dz13);
 877
 878             /* Update vectorial force */
 879             fix1             = _mm256_add_pd(fix1,tx);
 880             fiy1             = _mm256_add_pd(fiy1,ty);
 881             fiz1             = _mm256_add_pd(fiz1,tz);
 882
 883             fjx3             = _mm256_add_pd(fjx3,tx);
 884             fjy3             = _mm256_add_pd(fjy3,ty);
 885             fjz3             = _mm256_add_pd(fjz3,tz);
 886
 887             /**************************
 888              * CALCULATE INTERACTIONS *
 889              **************************/
 890
 891             /* REACTION-FIELD ELECTROSTATICS */
 892             velec            = _mm256_mul_pd(qq21,_mm256_sub_pd(_mm256_add_pd(rinv21,_mm256_mul_pd(krf,rsq21)),crf));
 893             felec            = _mm256_mul_pd(qq21,_mm256_sub_pd(_mm256_mul_pd(rinv21,rinvsq21),krf2));
 894
 895             /* Update potential sum for this i atom from the interaction with this j atom. */
 896             velec            = _mm256_andnot_pd(dummy_mask,velec);
 897             velecsum         = _mm256_add_pd(velecsum,velec);
 898
 899             fscal            = felec;
 900
 901             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
 902
 903             /* Calculate temporary vectorial force */
 904             tx               = _mm256_mul_pd(fscal,dx21);
 905             ty               = _mm256_mul_pd(fscal,dy21);
 906             tz               = _mm256_mul_pd(fscal,dz21);
 907
 908             /* Update vectorial force */
 909             fix2             = _mm256_add_pd(fix2,tx);
 910             fiy2             = _mm256_add_pd(fiy2,ty);
 911             fiz2             = _mm256_add_pd(fiz2,tz);
 912
 913             fjx1             = _mm256_add_pd(fjx1,tx);
 914             fjy1             = _mm256_add_pd(fjy1,ty);
 915             fjz1             = _mm256_add_pd(fjz1,tz);
 916
 917             /**************************
 918              * CALCULATE INTERACTIONS *
 919              **************************/
 920
 921             /* REACTION-FIELD ELECTROSTATICS */
 922             velec            = _mm256_mul_pd(qq22,_mm256_sub_pd(_mm256_add_pd(rinv22,_mm256_mul_pd(krf,rsq22)),crf));
 923             felec            = _mm256_mul_pd(qq22,_mm256_sub_pd(_mm256_mul_pd(rinv22,rinvsq22),krf2));
 924
 925             /* Update potential sum for this i atom from the interaction with this j atom. */
 926             velec            = _mm256_andnot_pd(dummy_mask,velec);
 927             velecsum         = _mm256_add_pd(velecsum,velec);
 928
 929             fscal            = felec;
 930
 931             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
 932
 933             /* Calculate temporary vectorial force */
 934             tx               = _mm256_mul_pd(fscal,dx22);
 935             ty               = _mm256_mul_pd(fscal,dy22);
 936             tz               = _mm256_mul_pd(fscal,dz22);
 937
 938             /* Update vectorial force */
 939             fix2             = _mm256_add_pd(fix2,tx);
 940             fiy2             = _mm256_add_pd(fiy2,ty);
 941             fiz2             = _mm256_add_pd(fiz2,tz);
 942
 943             fjx2             = _mm256_add_pd(fjx2,tx);
 944             fjy2             = _mm256_add_pd(fjy2,ty);
 945             fjz2             = _mm256_add_pd(fjz2,tz);
 946
 947             /**************************
 948              * CALCULATE INTERACTIONS *
 949              **************************/
 950
 951             /* REACTION-FIELD ELECTROSTATICS */
 952             velec            = _mm256_mul_pd(qq23,_mm256_sub_pd(_mm256_add_pd(rinv23,_mm256_mul_pd(krf,rsq23)),crf));
 953             felec            = _mm256_mul_pd(qq23,_mm256_sub_pd(_mm256_mul_pd(rinv23,rinvsq23),krf2));
 954
 955             /* Update potential sum for this i atom from the interaction with this j atom. */
 956             velec            = _mm256_andnot_pd(dummy_mask,velec);
 957             velecsum         = _mm256_add_pd(velecsum,velec);
 958
 959             fscal            = felec;
 960
 961             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
 962
 963             /* Calculate temporary vectorial force */
 964             tx               = _mm256_mul_pd(fscal,dx23);
 965             ty               = _mm256_mul_pd(fscal,dy23);
 966             tz               = _mm256_mul_pd(fscal,dz23);
 967
 968             /* Update vectorial force */
 969             fix2             = _mm256_add_pd(fix2,tx);
 970             fiy2             = _mm256_add_pd(fiy2,ty);
 971             fiz2             = _mm256_add_pd(fiz2,tz);
 972
 973             fjx3             = _mm256_add_pd(fjx3,tx);
 974             fjy3             = _mm256_add_pd(fjy3,ty);
 975             fjz3             = _mm256_add_pd(fjz3,tz);
 976
 977             /**************************
 978              * CALCULATE INTERACTIONS *
 979              **************************/
 980
 981             /* REACTION-FIELD ELECTROSTATICS */
 982             velec            = _mm256_mul_pd(qq31,_mm256_sub_pd(_mm256_add_pd(rinv31,_mm256_mul_pd(krf,rsq31)),crf));
 983             felec            = _mm256_mul_pd(qq31,_mm256_sub_pd(_mm256_mul_pd(rinv31,rinvsq31),krf2));
 984
 985             /* Update potential sum for this i atom from the interaction with this j atom. */
 986             velec            = _mm256_andnot_pd(dummy_mask,velec);
 987             velecsum         = _mm256_add_pd(velecsum,velec);
 988
 989             fscal            = felec;
 990
 991             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
 992
 993             /* Calculate temporary vectorial force */
 994             tx               = _mm256_mul_pd(fscal,dx31);
 995             ty               = _mm256_mul_pd(fscal,dy31);
 996             tz               = _mm256_mul_pd(fscal,dz31);
 997
 998             /* Update vectorial force */
 999             fix3             = _mm256_add_pd(fix3,tx);
1000             fiy3             = _mm256_add_pd(fiy3,ty);
1001             fiz3             = _mm256_add_pd(fiz3,tz);
1002
1003             fjx1             = _mm256_add_pd(fjx1,tx);
1004             fjy1             = _mm256_add_pd(fjy1,ty);
1005             fjz1             = _mm256_add_pd(fjz1,tz);
1006
1007             /**************************
1008              * CALCULATE INTERACTIONS *
1009              **************************/
1010
1011             /* REACTION-FIELD ELECTROSTATICS */
1012             velec            = _mm256_mul_pd(qq32,_mm256_sub_pd(_mm256_add_pd(rinv32,_mm256_mul_pd(krf,rsq32)),crf));
1013             felec            = _mm256_mul_pd(qq32,_mm256_sub_pd(_mm256_mul_pd(rinv32,rinvsq32),krf2));
1014
1015             /* Update potential sum for this i atom from the interaction with this j atom. */
1016             velec            = _mm256_andnot_pd(dummy_mask,velec);
1017             velecsum         = _mm256_add_pd(velecsum,velec);
1018
1019             fscal            = felec;
1020
1021             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
1022
1023             /* Calculate temporary vectorial force */
1024             tx               = _mm256_mul_pd(fscal,dx32);
1025             ty               = _mm256_mul_pd(fscal,dy32);
1026             tz               = _mm256_mul_pd(fscal,dz32);
1027
1028             /* Update vectorial force */
1029             fix3             = _mm256_add_pd(fix3,tx);
1030             fiy3             = _mm256_add_pd(fiy3,ty);
1031             fiz3             = _mm256_add_pd(fiz3,tz);
1032
1033             fjx2             = _mm256_add_pd(fjx2,tx);
1034             fjy2             = _mm256_add_pd(fjy2,ty);
1035             fjz2             = _mm256_add_pd(fjz2,tz);
1036
1037             /**************************
1038              * CALCULATE INTERACTIONS *
1039              **************************/
1040
1041             /* REACTION-FIELD ELECTROSTATICS */
1042             velec            = _mm256_mul_pd(qq33,_mm256_sub_pd(_mm256_add_pd(rinv33,_mm256_mul_pd(krf,rsq33)),crf));
1043             felec            = _mm256_mul_pd(qq33,_mm256_sub_pd(_mm256_mul_pd(rinv33,rinvsq33),krf2));
1044
1045             /* Update potential sum for this i atom from the interaction with this j atom. */
1046             velec            = _mm256_andnot_pd(dummy_mask,velec);
1047             velecsum         = _mm256_add_pd(velecsum,velec);
1048
1049             fscal            = felec;
1050
1051             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
1052
1053             /* Calculate temporary vectorial force */
1054             tx               = _mm256_mul_pd(fscal,dx33);
1055             ty               = _mm256_mul_pd(fscal,dy33);
1056             tz               = _mm256_mul_pd(fscal,dz33);
1057
1058             /* Update vectorial force */
1059             fix3             = _mm256_add_pd(fix3,tx);
1060             fiy3             = _mm256_add_pd(fiy3,ty);
1061             fiz3             = _mm256_add_pd(fiz3,tz);
1062
1063             fjx3             = _mm256_add_pd(fjx3,tx);
1064             fjy3             = _mm256_add_pd(fjy3,ty);
1065             fjz3             = _mm256_add_pd(fjz3,tz);
1066
1067             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
1068             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
1069             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
1070             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
1071
1072             gmx_mm256_decrement_4rvec_4ptr_swizzle_pd(fjptrA,fjptrB,fjptrC,fjptrD,
1073                                                       fjx0,fjy0,fjz0,fjx1,fjy1,fjz1,
1074                                                       fjx2,fjy2,fjz2,fjx3,fjy3,fjz3);
1075
1076             /* Inner loop uses 348 flops */
1077         }
1078
1079         /* End of innermost loop */
1080
1081         gmx_mm256_update_iforce_4atom_swizzle_pd(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,fix3,fiy3,fiz3,
1082                                                  f+i_coord_offset,fshift+i_shift_offset);
1083
1084         ggid                        = gid[iidx];
1085         /* Update potential energies */
1086         gmx_mm256_update_1pot_pd(velecsum,kernel_data->energygrp_elec+ggid);
1087         gmx_mm256_update_1pot_pd(vvdwsum,kernel_data->energygrp_vdw+ggid);
1088
1089         /* Increment number of inner iterations */
1090         inneriter                  += j_index_end - j_index_start;
1091
1092         /* Outer loop uses 26 flops */
1093     }
1094
1095     /* Increment number of outer iterations */
1096     outeriter        += nri;
1097
1098     /* Update outer/inner flops */
1099
1100     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W4W4_VF,outeriter*26 + inneriter*348);
1101 }
1102 /*
1103  * Gromacs nonbonded kernel:   nb_kernel_ElecRF_VdwCSTab_GeomW4W4_F_avx_256_double
1104  * Electrostatics interaction: ReactionField
1105  * VdW interaction:            CubicSplineTable
1106  * Geometry:                   Water4-Water4
1107  * Calculate force/pot:        Force
1108  */
1109 void
1110 nb_kernel_ElecRF_VdwCSTab_GeomW4W4_F_avx_256_double
1111                     (t_nblist * gmx_restrict                nlist,
1112                      rvec * gmx_restrict                    xx,
1113                      rvec * gmx_restrict                    ff,
1114                      t_forcerec * gmx_restrict              fr,
1115                      t_mdatoms * gmx_restrict               mdatoms,
1116                      nb_kernel_data_t * gmx_restrict        kernel_data,
1117                      t_nrnb * gmx_restrict                  nrnb)
1118 {
1119     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
1120      * just 0 for non-waters.
1121      * Suffixes A,B,C,D refer to j loop unrolling done with AVX, e.g. for the four different
1122      * jnr indices corresponding to data put in the four positions in the SIMD register.
1123      */
1124     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
1125     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
1126     int              jnrA,jnrB,jnrC,jnrD;
1127     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
1128     int              jnrlistE,jnrlistF,jnrlistG,jnrlistH;
1129     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
1130     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
1131     real             rcutoff_scalar;
1132     real             *shiftvec,*fshift,*x,*f;
1133     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD;
1134     real             scratch[4*DIM];
1135     __m256d          tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
1136     real *           vdwioffsetptr0;
1137     __m256d          ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
1138     real *           vdwioffsetptr1;
1139     __m256d          ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
1140     real *           vdwioffsetptr2;
1141     __m256d          ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
1142     real *           vdwioffsetptr3;
1143     __m256d          ix3,iy3,iz3,fix3,fiy3,fiz3,iq3,isai3;
1144     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
1145     __m256d          jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
1146     int              vdwjidx1A,vdwjidx1B,vdwjidx1C,vdwjidx1D;
1147     __m256d          jx1,jy1,jz1,fjx1,fjy1,fjz1,jq1,isaj1;
1148     int              vdwjidx2A,vdwjidx2B,vdwjidx2C,vdwjidx2D;
1149     __m256d          jx2,jy2,jz2,fjx2,fjy2,fjz2,jq2,isaj2;
1150     int              vdwjidx3A,vdwjidx3B,vdwjidx3C,vdwjidx3D;
1151     __m256d          jx3,jy3,jz3,fjx3,fjy3,fjz3,jq3,isaj3;
1152     __m256d          dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
1153     __m256d          dx11,dy11,dz11,rsq11,rinv11,rinvsq11,r11,qq11,c6_11,c12_11;
1154     __m256d          dx12,dy12,dz12,rsq12,rinv12,rinvsq12,r12,qq12,c6_12,c12_12;
1155     __m256d          dx13,dy13,dz13,rsq13,rinv13,rinvsq13,r13,qq13,c6_13,c12_13;
1156     __m256d          dx21,dy21,dz21,rsq21,rinv21,rinvsq21,r21,qq21,c6_21,c12_21;
1157     __m256d          dx22,dy22,dz22,rsq22,rinv22,rinvsq22,r22,qq22,c6_22,c12_22;
1158     __m256d          dx23,dy23,dz23,rsq23,rinv23,rinvsq23,r23,qq23,c6_23,c12_23;
1159     __m256d          dx31,dy31,dz31,rsq31,rinv31,rinvsq31,r31,qq31,c6_31,c12_31;
1160     __m256d          dx32,dy32,dz32,rsq32,rinv32,rinvsq32,r32,qq32,c6_32,c12_32;
1161     __m256d          dx33,dy33,dz33,rsq33,rinv33,rinvsq33,r33,qq33,c6_33,c12_33;
1162     __m256d          velec,felec,velecsum,facel,crf,krf,krf2;
1163     real             *charge;
1164     int              nvdwtype;
1165     __m256d          rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
1166     int              *vdwtype;
1167     real             *vdwparam;
1168     __m256d          one_sixth   = _mm256_set1_pd(1.0/6.0);
1169     __m256d          one_twelfth = _mm256_set1_pd(1.0/12.0);
1170     __m128i          vfitab;
1171     __m128i          ifour       = _mm_set1_epi32(4);
1172     __m256d          rt,vfeps,vftabscale,Y,F,G,H,Heps,Fp,VV,FF;
1173     real             *vftab;
1174     __m256d          dummy_mask,cutoff_mask;
1175     __m128           tmpmask0,tmpmask1;
1176     __m256d          signbit = _mm256_castsi256_pd( _mm256_set1_epi32(0x80000000) );
1177     __m256d          one     = _mm256_set1_pd(1.0);
1178     __m256d          two     = _mm256_set1_pd(2.0);
1179     x                = xx[0];
1180     f                = ff[0];
1181
1182     nri              = nlist->nri;
1183     iinr             = nlist->iinr;
1184     jindex           = nlist->jindex;
1185     jjnr             = nlist->jjnr;
1186     shiftidx         = nlist->shift;
1187     gid              = nlist->gid;
1188     shiftvec         = fr->shift_vec[0];
1189     fshift           = fr->fshift[0];
1190     facel            = _mm256_set1_pd(fr->epsfac);
1191     charge           = mdatoms->chargeA;
1192     krf              = _mm256_set1_pd(fr->ic->k_rf);
1193     krf2             = _mm256_set1_pd(fr->ic->k_rf*2.0);
1194     crf              = _mm256_set1_pd(fr->ic->c_rf);
1195     nvdwtype         = fr->ntype;
1196     vdwparam         = fr->nbfp;
1197     vdwtype          = mdatoms->typeA;
1198
1199     vftab            = kernel_data->table_vdw->data;
1200     vftabscale       = _mm256_set1_pd(kernel_data->table_vdw->scale);
1201
1202     /* Setup water-specific parameters */
1203     inr              = nlist->iinr[0];
1204     iq1              = _mm256_mul_pd(facel,_mm256_set1_pd(charge[inr+1]));
1205     iq2              = _mm256_mul_pd(facel,_mm256_set1_pd(charge[inr+2]));
1206     iq3              = _mm256_mul_pd(facel,_mm256_set1_pd(charge[inr+3]));
1207     vdwioffsetptr0   = vdwparam+2*nvdwtype*vdwtype[inr+0];
1208
1209     jq1              = _mm256_set1_pd(charge[inr+1]);
1210     jq2              = _mm256_set1_pd(charge[inr+2]);
1211     jq3              = _mm256_set1_pd(charge[inr+3]);
1212     vdwjidx0A        = 2*vdwtype[inr+0];
1213     c6_00            = _mm256_set1_pd(vdwioffsetptr0[vdwjidx0A]);
1214     c12_00           = _mm256_set1_pd(vdwioffsetptr0[vdwjidx0A+1]);
1215     qq11             = _mm256_mul_pd(iq1,jq1);
1216     qq12             = _mm256_mul_pd(iq1,jq2);
1217     qq13             = _mm256_mul_pd(iq1,jq3);
1218     qq21             = _mm256_mul_pd(iq2,jq1);
1219     qq22             = _mm256_mul_pd(iq2,jq2);
1220     qq23             = _mm256_mul_pd(iq2,jq3);
1221     qq31             = _mm256_mul_pd(iq3,jq1);
1222     qq32             = _mm256_mul_pd(iq3,jq2);
1223     qq33             = _mm256_mul_pd(iq3,jq3);
1224
1225     /* Avoid stupid compiler warnings */
1226     jnrA = jnrB = jnrC = jnrD = 0;
1227     j_coord_offsetA = 0;
1228     j_coord_offsetB = 0;
1229     j_coord_offsetC = 0;
1230     j_coord_offsetD = 0;
1231
1232     outeriter        = 0;
1233     inneriter        = 0;
1234
1235     for(iidx=0;iidx<4*DIM;iidx++)
1236     {
1237         scratch[iidx] = 0.0;
1238     }
1239
1240     /* Start outer loop over neighborlists */
1241     for(iidx=0; iidx<nri; iidx++)
1242     {
1243         /* Load shift vector for this list */
1244         i_shift_offset   = DIM*shiftidx[iidx];
1245
1246         /* Load limits for loop over neighbors */
1247         j_index_start    = jindex[iidx];
1248         j_index_end      = jindex[iidx+1];
1249
1250         /* Get outer coordinate index */
1251         inr              = iinr[iidx];
1252         i_coord_offset   = DIM*inr;
1253
1254         /* Load i particle coords and add shift vector */
1255         gmx_mm256_load_shift_and_4rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset,
1256                                                     &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2,&ix3,&iy3,&iz3);
1257
1258         fix0             = _mm256_setzero_pd();
1259         fiy0             = _mm256_setzero_pd();
1260         fiz0             = _mm256_setzero_pd();
1261         fix1             = _mm256_setzero_pd();
1262         fiy1             = _mm256_setzero_pd();
1263         fiz1             = _mm256_setzero_pd();
1264         fix2             = _mm256_setzero_pd();
1265         fiy2             = _mm256_setzero_pd();
1266         fiz2             = _mm256_setzero_pd();
1267         fix3             = _mm256_setzero_pd();
1268         fiy3             = _mm256_setzero_pd();
1269         fiz3             = _mm256_setzero_pd();
1270
1271         /* Start inner kernel loop */
1272         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
1273         {
1274
1275             /* Get j neighbor index, and coordinate index */
1276             jnrA             = jjnr[jidx];
1277             jnrB             = jjnr[jidx+1];
1278             jnrC             = jjnr[jidx+2];
1279             jnrD             = jjnr[jidx+3];
1280             j_coord_offsetA  = DIM*jnrA;
1281             j_coord_offsetB  = DIM*jnrB;
1282             j_coord_offsetC  = DIM*jnrC;
1283             j_coord_offsetD  = DIM*jnrD;
1284
1285             /* load j atom coordinates */
1286             gmx_mm256_load_4rvec_4ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
1287                                                  x+j_coord_offsetC,x+j_coord_offsetD,
1288                                                  &jx0,&jy0,&jz0,&jx1,&jy1,&jz1,&jx2,
1289                                                  &jy2,&jz2,&jx3,&jy3,&jz3);
1290
1291             /* Calculate displacement vector */
1292             dx00             = _mm256_sub_pd(ix0,jx0);
1293             dy00             = _mm256_sub_pd(iy0,jy0);
1294             dz00             = _mm256_sub_pd(iz0,jz0);
1295             dx11             = _mm256_sub_pd(ix1,jx1);
1296             dy11             = _mm256_sub_pd(iy1,jy1);
1297             dz11             = _mm256_sub_pd(iz1,jz1);
1298             dx12             = _mm256_sub_pd(ix1,jx2);
1299             dy12             = _mm256_sub_pd(iy1,jy2);
1300             dz12             = _mm256_sub_pd(iz1,jz2);
1301             dx13             = _mm256_sub_pd(ix1,jx3);
1302             dy13             = _mm256_sub_pd(iy1,jy3);
1303             dz13             = _mm256_sub_pd(iz1,jz3);
1304             dx21             = _mm256_sub_pd(ix2,jx1);
1305             dy21             = _mm256_sub_pd(iy2,jy1);
1306             dz21             = _mm256_sub_pd(iz2,jz1);
1307             dx22             = _mm256_sub_pd(ix2,jx2);
1308             dy22             = _mm256_sub_pd(iy2,jy2);
1309             dz22             = _mm256_sub_pd(iz2,jz2);
1310             dx23             = _mm256_sub_pd(ix2,jx3);
1311             dy23             = _mm256_sub_pd(iy2,jy3);
1312             dz23             = _mm256_sub_pd(iz2,jz3);
1313             dx31             = _mm256_sub_pd(ix3,jx1);
1314             dy31             = _mm256_sub_pd(iy3,jy1);
1315             dz31             = _mm256_sub_pd(iz3,jz1);
1316             dx32             = _mm256_sub_pd(ix3,jx2);
1317             dy32             = _mm256_sub_pd(iy3,jy2);
1318             dz32             = _mm256_sub_pd(iz3,jz2);
1319             dx33             = _mm256_sub_pd(ix3,jx3);
1320             dy33             = _mm256_sub_pd(iy3,jy3);
1321             dz33             = _mm256_sub_pd(iz3,jz3);
1322
1323             /* Calculate squared distance and things based on it */
1324             rsq00            = gmx_mm256_calc_rsq_pd(dx00,dy00,dz00);
1325             rsq11            = gmx_mm256_calc_rsq_pd(dx11,dy11,dz11);
1326             rsq12            = gmx_mm256_calc_rsq_pd(dx12,dy12,dz12);
1327             rsq13            = gmx_mm256_calc_rsq_pd(dx13,dy13,dz13);
1328             rsq21            = gmx_mm256_calc_rsq_pd(dx21,dy21,dz21);
1329             rsq22            = gmx_mm256_calc_rsq_pd(dx22,dy22,dz22);
1330             rsq23            = gmx_mm256_calc_rsq_pd(dx23,dy23,dz23);
1331             rsq31            = gmx_mm256_calc_rsq_pd(dx31,dy31,dz31);
1332             rsq32            = gmx_mm256_calc_rsq_pd(dx32,dy32,dz32);
1333             rsq33            = gmx_mm256_calc_rsq_pd(dx33,dy33,dz33);
1334
1335             rinv00           = gmx_mm256_invsqrt_pd(rsq00);
1336             rinv11           = gmx_mm256_invsqrt_pd(rsq11);
1337             rinv12           = gmx_mm256_invsqrt_pd(rsq12);
1338             rinv13           = gmx_mm256_invsqrt_pd(rsq13);
1339             rinv21           = gmx_mm256_invsqrt_pd(rsq21);
1340             rinv22           = gmx_mm256_invsqrt_pd(rsq22);
1341             rinv23           = gmx_mm256_invsqrt_pd(rsq23);
1342             rinv31           = gmx_mm256_invsqrt_pd(rsq31);
1343             rinv32           = gmx_mm256_invsqrt_pd(rsq32);
1344             rinv33           = gmx_mm256_invsqrt_pd(rsq33);
1345
1346             rinvsq11         = _mm256_mul_pd(rinv11,rinv11);
1347             rinvsq12         = _mm256_mul_pd(rinv12,rinv12);
1348             rinvsq13         = _mm256_mul_pd(rinv13,rinv13);
1349             rinvsq21         = _mm256_mul_pd(rinv21,rinv21);
1350             rinvsq22         = _mm256_mul_pd(rinv22,rinv22);
1351             rinvsq23         = _mm256_mul_pd(rinv23,rinv23);
1352             rinvsq31         = _mm256_mul_pd(rinv31,rinv31);
1353             rinvsq32         = _mm256_mul_pd(rinv32,rinv32);
1354             rinvsq33         = _mm256_mul_pd(rinv33,rinv33);
1355
1356             fjx0             = _mm256_setzero_pd();
1357             fjy0             = _mm256_setzero_pd();
1358             fjz0             = _mm256_setzero_pd();
1359             fjx1             = _mm256_setzero_pd();
1360             fjy1             = _mm256_setzero_pd();
1361             fjz1             = _mm256_setzero_pd();
1362             fjx2             = _mm256_setzero_pd();
1363             fjy2             = _mm256_setzero_pd();
1364             fjz2             = _mm256_setzero_pd();
1365             fjx3             = _mm256_setzero_pd();
1366             fjy3             = _mm256_setzero_pd();
1367             fjz3             = _mm256_setzero_pd();
1368
1369             /**************************
1370              * CALCULATE INTERACTIONS *
1371              **************************/
1372
1373             r00              = _mm256_mul_pd(rsq00,rinv00);
1374
1375             /* Calculate table index by multiplying r with table scale and truncate to integer */
1376             rt               = _mm256_mul_pd(r00,vftabscale);
1377             vfitab           = _mm256_cvttpd_epi32(rt);
1378             vfeps            = _mm256_sub_pd(rt,_mm256_round_pd(rt, _MM_FROUND_FLOOR));
1379             vfitab           = _mm_slli_epi32(vfitab,3);
1380
1381             /* CUBIC SPLINE TABLE DISPERSION */
1382             Y                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
1383             F                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
1384             G                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,2) );
1385             H                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,3) );
1386             GMX_MM256_FULLTRANSPOSE4_PD(Y,F,G,H);
1387             Heps             = _mm256_mul_pd(vfeps,H);
1388             Fp               = _mm256_add_pd(F,_mm256_mul_pd(vfeps,_mm256_add_pd(G,Heps)));
1389             FF               = _mm256_add_pd(Fp,_mm256_mul_pd(vfeps,_mm256_add_pd(G,_mm256_add_pd(Heps,Heps))));
1390             fvdw6            = _mm256_mul_pd(c6_00,FF);
1391
1392             /* CUBIC SPLINE TABLE REPULSION */
1393             vfitab           = _mm_add_epi32(vfitab,ifour);
1394             Y                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
1395             F                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
1396             G                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,2) );
1397             H                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,3) );
1398             GMX_MM256_FULLTRANSPOSE4_PD(Y,F,G,H);
1399             Heps             = _mm256_mul_pd(vfeps,H);
1400             Fp               = _mm256_add_pd(F,_mm256_mul_pd(vfeps,_mm256_add_pd(G,Heps)));
1401             FF               = _mm256_add_pd(Fp,_mm256_mul_pd(vfeps,_mm256_add_pd(G,_mm256_add_pd(Heps,Heps))));
1402             fvdw12           = _mm256_mul_pd(c12_00,FF);
1403             fvdw             = _mm256_xor_pd(signbit,_mm256_mul_pd(_mm256_add_pd(fvdw6,fvdw12),_mm256_mul_pd(vftabscale,rinv00)));
1404
1405             fscal            = fvdw;
1406
1407             /* Calculate temporary vectorial force */
1408             tx               = _mm256_mul_pd(fscal,dx00);
1409             ty               = _mm256_mul_pd(fscal,dy00);
1410             tz               = _mm256_mul_pd(fscal,dz00);
1411
1412             /* Update vectorial force */
1413             fix0             = _mm256_add_pd(fix0,tx);
1414             fiy0             = _mm256_add_pd(fiy0,ty);
1415             fiz0             = _mm256_add_pd(fiz0,tz);
1416
1417             fjx0             = _mm256_add_pd(fjx0,tx);
1418             fjy0             = _mm256_add_pd(fjy0,ty);
1419             fjz0             = _mm256_add_pd(fjz0,tz);
1420
1421             /**************************
1422              * CALCULATE INTERACTIONS *
1423              **************************/
1424
1425             /* REACTION-FIELD ELECTROSTATICS */
1426             felec            = _mm256_mul_pd(qq11,_mm256_sub_pd(_mm256_mul_pd(rinv11,rinvsq11),krf2));
1427
1428             fscal            = felec;
1429
1430             /* Calculate temporary vectorial force */
1431             tx               = _mm256_mul_pd(fscal,dx11);
1432             ty               = _mm256_mul_pd(fscal,dy11);
1433             tz               = _mm256_mul_pd(fscal,dz11);
1434
1435             /* Update vectorial force */
1436             fix1             = _mm256_add_pd(fix1,tx);
1437             fiy1             = _mm256_add_pd(fiy1,ty);
1438             fiz1             = _mm256_add_pd(fiz1,tz);
1439
1440             fjx1             = _mm256_add_pd(fjx1,tx);
1441             fjy1             = _mm256_add_pd(fjy1,ty);
1442             fjz1             = _mm256_add_pd(fjz1,tz);
1443
1444             /**************************
1445              * CALCULATE INTERACTIONS *
1446              **************************/
1447
1448             /* REACTION-FIELD ELECTROSTATICS */
1449             felec            = _mm256_mul_pd(qq12,_mm256_sub_pd(_mm256_mul_pd(rinv12,rinvsq12),krf2));
1450
1451             fscal            = felec;
1452
1453             /* Calculate temporary vectorial force */
1454             tx               = _mm256_mul_pd(fscal,dx12);
1455             ty               = _mm256_mul_pd(fscal,dy12);
1456             tz               = _mm256_mul_pd(fscal,dz12);
1457
1458             /* Update vectorial force */
1459             fix1             = _mm256_add_pd(fix1,tx);
1460             fiy1             = _mm256_add_pd(fiy1,ty);
1461             fiz1             = _mm256_add_pd(fiz1,tz);
1462
1463             fjx2             = _mm256_add_pd(fjx2,tx);
1464             fjy2             = _mm256_add_pd(fjy2,ty);
1465             fjz2             = _mm256_add_pd(fjz2,tz);
1466
1467             /**************************
1468              * CALCULATE INTERACTIONS *
1469              **************************/
1470
1471             /* REACTION-FIELD ELECTROSTATICS */
1472             felec            = _mm256_mul_pd(qq13,_mm256_sub_pd(_mm256_mul_pd(rinv13,rinvsq13),krf2));
1473
1474             fscal            = felec;
1475
1476             /* Calculate temporary vectorial force */
1477             tx               = _mm256_mul_pd(fscal,dx13);
1478             ty               = _mm256_mul_pd(fscal,dy13);
1479             tz               = _mm256_mul_pd(fscal,dz13);
1480
1481             /* Update vectorial force */
1482             fix1             = _mm256_add_pd(fix1,tx);
1483             fiy1             = _mm256_add_pd(fiy1,ty);
1484             fiz1             = _mm256_add_pd(fiz1,tz);
1485
1486             fjx3             = _mm256_add_pd(fjx3,tx);
1487             fjy3             = _mm256_add_pd(fjy3,ty);
1488             fjz3             = _mm256_add_pd(fjz3,tz);
1489
1490             /**************************
1491              * CALCULATE INTERACTIONS *
1492              **************************/
1493
1494             /* REACTION-FIELD ELECTROSTATICS */
1495             felec            = _mm256_mul_pd(qq21,_mm256_sub_pd(_mm256_mul_pd(rinv21,rinvsq21),krf2));
1496
1497             fscal            = felec;
1498
1499             /* Calculate temporary vectorial force */
1500             tx               = _mm256_mul_pd(fscal,dx21);
1501             ty               = _mm256_mul_pd(fscal,dy21);
1502             tz               = _mm256_mul_pd(fscal,dz21);
1503
1504             /* Update vectorial force */
1505             fix2             = _mm256_add_pd(fix2,tx);
1506             fiy2             = _mm256_add_pd(fiy2,ty);
1507             fiz2             = _mm256_add_pd(fiz2,tz);
1508
1509             fjx1             = _mm256_add_pd(fjx1,tx);
1510             fjy1             = _mm256_add_pd(fjy1,ty);
1511             fjz1             = _mm256_add_pd(fjz1,tz);
1512
1513             /**************************
1514              * CALCULATE INTERACTIONS *
1515              **************************/
1516
1517             /* REACTION-FIELD ELECTROSTATICS */
1518             felec            = _mm256_mul_pd(qq22,_mm256_sub_pd(_mm256_mul_pd(rinv22,rinvsq22),krf2));
1519
1520             fscal            = felec;
1521
1522             /* Calculate temporary vectorial force */
1523             tx               = _mm256_mul_pd(fscal,dx22);
1524             ty               = _mm256_mul_pd(fscal,dy22);
1525             tz               = _mm256_mul_pd(fscal,dz22);
1526
1527             /* Update vectorial force */
1528             fix2             = _mm256_add_pd(fix2,tx);
1529             fiy2             = _mm256_add_pd(fiy2,ty);
1530             fiz2             = _mm256_add_pd(fiz2,tz);
1531
1532             fjx2             = _mm256_add_pd(fjx2,tx);
1533             fjy2             = _mm256_add_pd(fjy2,ty);
1534             fjz2             = _mm256_add_pd(fjz2,tz);
1535
1536             /**************************
1537              * CALCULATE INTERACTIONS *
1538              **************************/
1539
1540             /* REACTION-FIELD ELECTROSTATICS */
1541             felec            = _mm256_mul_pd(qq23,_mm256_sub_pd(_mm256_mul_pd(rinv23,rinvsq23),krf2));
1542
1543             fscal            = felec;
1544
1545             /* Calculate temporary vectorial force */
1546             tx               = _mm256_mul_pd(fscal,dx23);
1547             ty               = _mm256_mul_pd(fscal,dy23);
1548             tz               = _mm256_mul_pd(fscal,dz23);
1549
1550             /* Update vectorial force */
1551             fix2             = _mm256_add_pd(fix2,tx);
1552             fiy2             = _mm256_add_pd(fiy2,ty);
1553             fiz2             = _mm256_add_pd(fiz2,tz);
1554
1555             fjx3             = _mm256_add_pd(fjx3,tx);
1556             fjy3             = _mm256_add_pd(fjy3,ty);
1557             fjz3             = _mm256_add_pd(fjz3,tz);
1558
1559             /**************************
1560              * CALCULATE INTERACTIONS *
1561              **************************/
1562
1563             /* REACTION-FIELD ELECTROSTATICS */
1564             felec            = _mm256_mul_pd(qq31,_mm256_sub_pd(_mm256_mul_pd(rinv31,rinvsq31),krf2));
1565
1566             fscal            = felec;
1567
1568             /* Calculate temporary vectorial force */
1569             tx               = _mm256_mul_pd(fscal,dx31);
1570             ty               = _mm256_mul_pd(fscal,dy31);
1571             tz               = _mm256_mul_pd(fscal,dz31);
1572
1573             /* Update vectorial force */
1574             fix3             = _mm256_add_pd(fix3,tx);
1575             fiy3             = _mm256_add_pd(fiy3,ty);
1576             fiz3             = _mm256_add_pd(fiz3,tz);
1577
1578             fjx1             = _mm256_add_pd(fjx1,tx);
1579             fjy1             = _mm256_add_pd(fjy1,ty);
1580             fjz1             = _mm256_add_pd(fjz1,tz);
1581
1582             /**************************
1583              * CALCULATE INTERACTIONS *
1584              **************************/
1585
1586             /* REACTION-FIELD ELECTROSTATICS */
1587             felec            = _mm256_mul_pd(qq32,_mm256_sub_pd(_mm256_mul_pd(rinv32,rinvsq32),krf2));
1588
1589             fscal            = felec;
1590
1591             /* Calculate temporary vectorial force */
1592             tx               = _mm256_mul_pd(fscal,dx32);
1593             ty               = _mm256_mul_pd(fscal,dy32);
1594             tz               = _mm256_mul_pd(fscal,dz32);
1595
1596             /* Update vectorial force */
1597             fix3             = _mm256_add_pd(fix3,tx);
1598             fiy3             = _mm256_add_pd(fiy3,ty);
1599             fiz3             = _mm256_add_pd(fiz3,tz);
1600
1601             fjx2             = _mm256_add_pd(fjx2,tx);
1602             fjy2             = _mm256_add_pd(fjy2,ty);
1603             fjz2             = _mm256_add_pd(fjz2,tz);
1604
1605             /**************************
1606              * CALCULATE INTERACTIONS *
1607              **************************/
1608
1609             /* REACTION-FIELD ELECTROSTATICS */
1610             felec            = _mm256_mul_pd(qq33,_mm256_sub_pd(_mm256_mul_pd(rinv33,rinvsq33),krf2));
1611
1612             fscal            = felec;
1613
1614             /* Calculate temporary vectorial force */
1615             tx               = _mm256_mul_pd(fscal,dx33);
1616             ty               = _mm256_mul_pd(fscal,dy33);
1617             tz               = _mm256_mul_pd(fscal,dz33);
1618
1619             /* Update vectorial force */
1620             fix3             = _mm256_add_pd(fix3,tx);
1621             fiy3             = _mm256_add_pd(fiy3,ty);
1622             fiz3             = _mm256_add_pd(fiz3,tz);
1623
1624             fjx3             = _mm256_add_pd(fjx3,tx);
1625             fjy3             = _mm256_add_pd(fjy3,ty);
1626             fjz3             = _mm256_add_pd(fjz3,tz);
1627
1628             fjptrA             = f+j_coord_offsetA;
1629             fjptrB             = f+j_coord_offsetB;
1630             fjptrC             = f+j_coord_offsetC;
1631             fjptrD             = f+j_coord_offsetD;
1632
1633             gmx_mm256_decrement_4rvec_4ptr_swizzle_pd(fjptrA,fjptrB,fjptrC,fjptrD,
1634                                                       fjx0,fjy0,fjz0,fjx1,fjy1,fjz1,
1635                                                       fjx2,fjy2,fjz2,fjx3,fjy3,fjz3);
1636
1637             /* Inner loop uses 294 flops */
1638         }
1639
1640         if(jidx<j_index_end)
1641         {
1642
1643             /* Get j neighbor index, and coordinate index */
1644             jnrlistA         = jjnr[jidx];
1645             jnrlistB         = jjnr[jidx+1];
1646             jnrlistC         = jjnr[jidx+2];
1647             jnrlistD         = jjnr[jidx+3];
1648             /* Sign of each element will be negative for non-real atoms.
1649              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
1650              * so use it as val = _mm_andnot_pd(mask,val) to clear dummy entries.
1651              */
1652             tmpmask0 = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
1653
1654             tmpmask1 = _mm_permute_ps(tmpmask0,_GMX_MM_PERMUTE(3,3,2,2));
1655             tmpmask0 = _mm_permute_ps(tmpmask0,_GMX_MM_PERMUTE(1,1,0,0));
1656             dummy_mask = _mm256_castps_pd(gmx_mm256_set_m128(tmpmask1,tmpmask0));
1657
1658             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
1659             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
1660             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
1661             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
1662             j_coord_offsetA  = DIM*jnrA;
1663             j_coord_offsetB  = DIM*jnrB;
1664             j_coord_offsetC  = DIM*jnrC;
1665             j_coord_offsetD  = DIM*jnrD;
1666
1667             /* load j atom coordinates */
1668             gmx_mm256_load_4rvec_4ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
1669                                                  x+j_coord_offsetC,x+j_coord_offsetD,
1670                                                  &jx0,&jy0,&jz0,&jx1,&jy1,&jz1,&jx2,
1671                                                  &jy2,&jz2,&jx3,&jy3,&jz3);
1672
1673             /* Calculate displacement vector */
1674             dx00             = _mm256_sub_pd(ix0,jx0);
1675             dy00             = _mm256_sub_pd(iy0,jy0);
1676             dz00             = _mm256_sub_pd(iz0,jz0);
1677             dx11             = _mm256_sub_pd(ix1,jx1);
1678             dy11             = _mm256_sub_pd(iy1,jy1);
1679             dz11             = _mm256_sub_pd(iz1,jz1);
1680             dx12             = _mm256_sub_pd(ix1,jx2);
1681             dy12             = _mm256_sub_pd(iy1,jy2);
1682             dz12             = _mm256_sub_pd(iz1,jz2);
1683             dx13             = _mm256_sub_pd(ix1,jx3);
1684             dy13             = _mm256_sub_pd(iy1,jy3);
1685             dz13             = _mm256_sub_pd(iz1,jz3);
1686             dx21             = _mm256_sub_pd(ix2,jx1);
1687             dy21             = _mm256_sub_pd(iy2,jy1);
1688             dz21             = _mm256_sub_pd(iz2,jz1);
1689             dx22             = _mm256_sub_pd(ix2,jx2);
1690             dy22             = _mm256_sub_pd(iy2,jy2);
1691             dz22             = _mm256_sub_pd(iz2,jz2);
1692             dx23             = _mm256_sub_pd(ix2,jx3);
1693             dy23             = _mm256_sub_pd(iy2,jy3);
1694             dz23             = _mm256_sub_pd(iz2,jz3);
1695             dx31             = _mm256_sub_pd(ix3,jx1);
1696             dy31             = _mm256_sub_pd(iy3,jy1);
1697             dz31             = _mm256_sub_pd(iz3,jz1);
1698             dx32             = _mm256_sub_pd(ix3,jx2);
1699             dy32             = _mm256_sub_pd(iy3,jy2);
1700             dz32             = _mm256_sub_pd(iz3,jz2);
1701             dx33             = _mm256_sub_pd(ix3,jx3);
1702             dy33             = _mm256_sub_pd(iy3,jy3);
1703             dz33             = _mm256_sub_pd(iz3,jz3);
1704
1705             /* Calculate squared distance and things based on it */
1706             rsq00            = gmx_mm256_calc_rsq_pd(dx00,dy00,dz00);
1707             rsq11            = gmx_mm256_calc_rsq_pd(dx11,dy11,dz11);
1708             rsq12            = gmx_mm256_calc_rsq_pd(dx12,dy12,dz12);
1709             rsq13            = gmx_mm256_calc_rsq_pd(dx13,dy13,dz13);
1710             rsq21            = gmx_mm256_calc_rsq_pd(dx21,dy21,dz21);
1711             rsq22            = gmx_mm256_calc_rsq_pd(dx22,dy22,dz22);
1712             rsq23            = gmx_mm256_calc_rsq_pd(dx23,dy23,dz23);
1713             rsq31            = gmx_mm256_calc_rsq_pd(dx31,dy31,dz31);
1714             rsq32            = gmx_mm256_calc_rsq_pd(dx32,dy32,dz32);
1715             rsq33            = gmx_mm256_calc_rsq_pd(dx33,dy33,dz33);
1716
1717             rinv00           = gmx_mm256_invsqrt_pd(rsq00);
1718             rinv11           = gmx_mm256_invsqrt_pd(rsq11);
1719             rinv12           = gmx_mm256_invsqrt_pd(rsq12);
1720             rinv13           = gmx_mm256_invsqrt_pd(rsq13);
1721             rinv21           = gmx_mm256_invsqrt_pd(rsq21);
1722             rinv22           = gmx_mm256_invsqrt_pd(rsq22);
1723             rinv23           = gmx_mm256_invsqrt_pd(rsq23);
1724             rinv31           = gmx_mm256_invsqrt_pd(rsq31);
1725             rinv32           = gmx_mm256_invsqrt_pd(rsq32);
1726             rinv33           = gmx_mm256_invsqrt_pd(rsq33);
1727
1728             rinvsq11         = _mm256_mul_pd(rinv11,rinv11);
1729             rinvsq12         = _mm256_mul_pd(rinv12,rinv12);
1730             rinvsq13         = _mm256_mul_pd(rinv13,rinv13);
1731             rinvsq21         = _mm256_mul_pd(rinv21,rinv21);
1732             rinvsq22         = _mm256_mul_pd(rinv22,rinv22);
1733             rinvsq23         = _mm256_mul_pd(rinv23,rinv23);
1734             rinvsq31         = _mm256_mul_pd(rinv31,rinv31);
1735             rinvsq32         = _mm256_mul_pd(rinv32,rinv32);
1736             rinvsq33         = _mm256_mul_pd(rinv33,rinv33);
1737
1738             fjx0             = _mm256_setzero_pd();
1739             fjy0             = _mm256_setzero_pd();
1740             fjz0             = _mm256_setzero_pd();
1741             fjx1             = _mm256_setzero_pd();
1742             fjy1             = _mm256_setzero_pd();
1743             fjz1             = _mm256_setzero_pd();
1744             fjx2             = _mm256_setzero_pd();
1745             fjy2             = _mm256_setzero_pd();
1746             fjz2             = _mm256_setzero_pd();
1747             fjx3             = _mm256_setzero_pd();
1748             fjy3             = _mm256_setzero_pd();
1749             fjz3             = _mm256_setzero_pd();
1750
1751             /**************************
1752              * CALCULATE INTERACTIONS *
1753              **************************/
1754
1755             r00              = _mm256_mul_pd(rsq00,rinv00);
1756             r00              = _mm256_andnot_pd(dummy_mask,r00);
1757
1758             /* Calculate table index by multiplying r with table scale and truncate to integer */
1759             rt               = _mm256_mul_pd(r00,vftabscale);
1760             vfitab           = _mm256_cvttpd_epi32(rt);
1761             vfeps            = _mm256_sub_pd(rt,_mm256_round_pd(rt, _MM_FROUND_FLOOR));
1762             vfitab           = _mm_slli_epi32(vfitab,3);
1763
1764             /* CUBIC SPLINE TABLE DISPERSION */
1765             Y                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
1766             F                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
1767             G                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,2) );
1768             H                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,3) );
1769             GMX_MM256_FULLTRANSPOSE4_PD(Y,F,G,H);
1770             Heps             = _mm256_mul_pd(vfeps,H);
1771             Fp               = _mm256_add_pd(F,_mm256_mul_pd(vfeps,_mm256_add_pd(G,Heps)));
1772             FF               = _mm256_add_pd(Fp,_mm256_mul_pd(vfeps,_mm256_add_pd(G,_mm256_add_pd(Heps,Heps))));
1773             fvdw6            = _mm256_mul_pd(c6_00,FF);
1774
1775             /* CUBIC SPLINE TABLE REPULSION */
1776             vfitab           = _mm_add_epi32(vfitab,ifour);
1777             Y                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
1778             F                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
1779             G                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,2) );
1780             H                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,3) );
1781             GMX_MM256_FULLTRANSPOSE4_PD(Y,F,G,H);
1782             Heps             = _mm256_mul_pd(vfeps,H);
1783             Fp               = _mm256_add_pd(F,_mm256_mul_pd(vfeps,_mm256_add_pd(G,Heps)));
1784             FF               = _mm256_add_pd(Fp,_mm256_mul_pd(vfeps,_mm256_add_pd(G,_mm256_add_pd(Heps,Heps))));
1785             fvdw12           = _mm256_mul_pd(c12_00,FF);
1786             fvdw             = _mm256_xor_pd(signbit,_mm256_mul_pd(_mm256_add_pd(fvdw6,fvdw12),_mm256_mul_pd(vftabscale,rinv00)));
1787
1788             fscal            = fvdw;
1789
1790             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
1791
1792             /* Calculate temporary vectorial force */
1793             tx               = _mm256_mul_pd(fscal,dx00);
1794             ty               = _mm256_mul_pd(fscal,dy00);
1795             tz               = _mm256_mul_pd(fscal,dz00);
1796
1797             /* Update vectorial force */
1798             fix0             = _mm256_add_pd(fix0,tx);
1799             fiy0             = _mm256_add_pd(fiy0,ty);
1800             fiz0             = _mm256_add_pd(fiz0,tz);
1801
1802             fjx0             = _mm256_add_pd(fjx0,tx);
1803             fjy0             = _mm256_add_pd(fjy0,ty);
1804             fjz0             = _mm256_add_pd(fjz0,tz);
1805
1806             /**************************
1807              * CALCULATE INTERACTIONS *
1808              **************************/
1809
1810             /* REACTION-FIELD ELECTROSTATICS */
1811             felec            = _mm256_mul_pd(qq11,_mm256_sub_pd(_mm256_mul_pd(rinv11,rinvsq11),krf2));
1812
1813             fscal            = felec;
1814
1815             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
1816
1817             /* Calculate temporary vectorial force */
1818             tx               = _mm256_mul_pd(fscal,dx11);
1819             ty               = _mm256_mul_pd(fscal,dy11);
1820             tz               = _mm256_mul_pd(fscal,dz11);
1821
1822             /* Update vectorial force */
1823             fix1             = _mm256_add_pd(fix1,tx);
1824             fiy1             = _mm256_add_pd(fiy1,ty);
1825             fiz1             = _mm256_add_pd(fiz1,tz);
1826
1827             fjx1             = _mm256_add_pd(fjx1,tx);
1828             fjy1             = _mm256_add_pd(fjy1,ty);
1829             fjz1             = _mm256_add_pd(fjz1,tz);
1830
1831             /**************************
1832              * CALCULATE INTERACTIONS *
1833              **************************/
1834
1835             /* REACTION-FIELD ELECTROSTATICS */
1836             felec            = _mm256_mul_pd(qq12,_mm256_sub_pd(_mm256_mul_pd(rinv12,rinvsq12),krf2));
1837
1838             fscal            = felec;
1839
1840             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
1841
1842             /* Calculate temporary vectorial force */
1843             tx               = _mm256_mul_pd(fscal,dx12);
1844             ty               = _mm256_mul_pd(fscal,dy12);
1845             tz               = _mm256_mul_pd(fscal,dz12);
1846
1847             /* Update vectorial force */
1848             fix1             = _mm256_add_pd(fix1,tx);
1849             fiy1             = _mm256_add_pd(fiy1,ty);
1850             fiz1             = _mm256_add_pd(fiz1,tz);
1851
1852             fjx2             = _mm256_add_pd(fjx2,tx);
1853             fjy2             = _mm256_add_pd(fjy2,ty);
1854             fjz2             = _mm256_add_pd(fjz2,tz);
1855
1856             /**************************
1857              * CALCULATE INTERACTIONS *
1858              **************************/
1859
1860             /* REACTION-FIELD ELECTROSTATICS */
1861             felec            = _mm256_mul_pd(qq13,_mm256_sub_pd(_mm256_mul_pd(rinv13,rinvsq13),krf2));
1862
1863             fscal            = felec;
1864
1865             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
1866
1867             /* Calculate temporary vectorial force */
1868             tx               = _mm256_mul_pd(fscal,dx13);
1869             ty               = _mm256_mul_pd(fscal,dy13);
1870             tz               = _mm256_mul_pd(fscal,dz13);
1871
1872             /* Update vectorial force */
1873             fix1             = _mm256_add_pd(fix1,tx);
1874             fiy1             = _mm256_add_pd(fiy1,ty);
1875             fiz1             = _mm256_add_pd(fiz1,tz);
1876
1877             fjx3             = _mm256_add_pd(fjx3,tx);
1878             fjy3             = _mm256_add_pd(fjy3,ty);
1879             fjz3             = _mm256_add_pd(fjz3,tz);
1880
1881             /**************************
1882              * CALCULATE INTERACTIONS *
1883              **************************/
1884
1885             /* REACTION-FIELD ELECTROSTATICS */
1886             felec            = _mm256_mul_pd(qq21,_mm256_sub_pd(_mm256_mul_pd(rinv21,rinvsq21),krf2));
1887
1888             fscal            = felec;
1889
1890             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
1891
1892             /* Calculate temporary vectorial force */
1893             tx               = _mm256_mul_pd(fscal,dx21);
1894             ty               = _mm256_mul_pd(fscal,dy21);
1895             tz               = _mm256_mul_pd(fscal,dz21);
1896
1897             /* Update vectorial force */
1898             fix2             = _mm256_add_pd(fix2,tx);
1899             fiy2             = _mm256_add_pd(fiy2,ty);
1900             fiz2             = _mm256_add_pd(fiz2,tz);
1901
1902             fjx1             = _mm256_add_pd(fjx1,tx);
1903             fjy1             = _mm256_add_pd(fjy1,ty);
1904             fjz1             = _mm256_add_pd(fjz1,tz);
1905
1906             /**************************
1907              * CALCULATE INTERACTIONS *
1908              **************************/
1909
1910             /* REACTION-FIELD ELECTROSTATICS */
1911             felec            = _mm256_mul_pd(qq22,_mm256_sub_pd(_mm256_mul_pd(rinv22,rinvsq22),krf2));
1912
1913             fscal            = felec;
1914
1915             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
1916
1917             /* Calculate temporary vectorial force */
1918             tx               = _mm256_mul_pd(fscal,dx22);
1919             ty               = _mm256_mul_pd(fscal,dy22);
1920             tz               = _mm256_mul_pd(fscal,dz22);
1921
1922             /* Update vectorial force */
1923             fix2             = _mm256_add_pd(fix2,tx);
1924             fiy2             = _mm256_add_pd(fiy2,ty);
1925             fiz2             = _mm256_add_pd(fiz2,tz);
1926
1927             fjx2             = _mm256_add_pd(fjx2,tx);
1928             fjy2             = _mm256_add_pd(fjy2,ty);
1929             fjz2             = _mm256_add_pd(fjz2,tz);
1930
1931             /**************************
1932              * CALCULATE INTERACTIONS *
1933              **************************/
1934
1935             /* REACTION-FIELD ELECTROSTATICS */
1936             felec            = _mm256_mul_pd(qq23,_mm256_sub_pd(_mm256_mul_pd(rinv23,rinvsq23),krf2));
1937
1938             fscal            = felec;
1939
1940             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
1941
1942             /* Calculate temporary vectorial force */
1943             tx               = _mm256_mul_pd(fscal,dx23);
1944             ty               = _mm256_mul_pd(fscal,dy23);
1945             tz               = _mm256_mul_pd(fscal,dz23);
1946
1947             /* Update vectorial force */
1948             fix2             = _mm256_add_pd(fix2,tx);
1949             fiy2             = _mm256_add_pd(fiy2,ty);
1950             fiz2             = _mm256_add_pd(fiz2,tz);
1951
1952             fjx3             = _mm256_add_pd(fjx3,tx);
1953             fjy3             = _mm256_add_pd(fjy3,ty);
1954             fjz3             = _mm256_add_pd(fjz3,tz);
1955
1956             /**************************
1957              * CALCULATE INTERACTIONS *
1958              **************************/
1959
1960             /* REACTION-FIELD ELECTROSTATICS */
1961             felec            = _mm256_mul_pd(qq31,_mm256_sub_pd(_mm256_mul_pd(rinv31,rinvsq31),krf2));
1962
1963             fscal            = felec;
1964
1965             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
1966
1967             /* Calculate temporary vectorial force */
1968             tx               = _mm256_mul_pd(fscal,dx31);
1969             ty               = _mm256_mul_pd(fscal,dy31);
1970             tz               = _mm256_mul_pd(fscal,dz31);
1971
1972             /* Update vectorial force */
1973             fix3             = _mm256_add_pd(fix3,tx);
1974             fiy3             = _mm256_add_pd(fiy3,ty);
1975             fiz3             = _mm256_add_pd(fiz3,tz);
1976
1977             fjx1             = _mm256_add_pd(fjx1,tx);
1978             fjy1             = _mm256_add_pd(fjy1,ty);
1979             fjz1             = _mm256_add_pd(fjz1,tz);
1980
1981             /**************************
1982              * CALCULATE INTERACTIONS *
1983              **************************/
1984
1985             /* REACTION-FIELD ELECTROSTATICS */
1986             felec            = _mm256_mul_pd(qq32,_mm256_sub_pd(_mm256_mul_pd(rinv32,rinvsq32),krf2));
1987
1988             fscal            = felec;
1989
1990             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
1991
1992             /* Calculate temporary vectorial force */
1993             tx               = _mm256_mul_pd(fscal,dx32);
1994             ty               = _mm256_mul_pd(fscal,dy32);
1995             tz               = _mm256_mul_pd(fscal,dz32);
1996
1997             /* Update vectorial force */
1998             fix3             = _mm256_add_pd(fix3,tx);
1999             fiy3             = _mm256_add_pd(fiy3,ty);
2000             fiz3             = _mm256_add_pd(fiz3,tz);
2001
2002             fjx2             = _mm256_add_pd(fjx2,tx);
2003             fjy2             = _mm256_add_pd(fjy2,ty);
2004             fjz2             = _mm256_add_pd(fjz2,tz);
2005
2006             /**************************
2007              * CALCULATE INTERACTIONS *
2008              **************************/
2009
2010             /* REACTION-FIELD ELECTROSTATICS */
2011             felec            = _mm256_mul_pd(qq33,_mm256_sub_pd(_mm256_mul_pd(rinv33,rinvsq33),krf2));
2012
2013             fscal            = felec;
2014
2015             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
2016
2017             /* Calculate temporary vectorial force */
2018             tx               = _mm256_mul_pd(fscal,dx33);
2019             ty               = _mm256_mul_pd(fscal,dy33);
2020             tz               = _mm256_mul_pd(fscal,dz33);
2021
2022             /* Update vectorial force */
2023             fix3             = _mm256_add_pd(fix3,tx);
2024             fiy3             = _mm256_add_pd(fiy3,ty);
2025             fiz3             = _mm256_add_pd(fiz3,tz);
2026
2027             fjx3             = _mm256_add_pd(fjx3,tx);
2028             fjy3             = _mm256_add_pd(fjy3,ty);
2029             fjz3             = _mm256_add_pd(fjz3,tz);
2030
2031             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
2032             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
2033             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
2034             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
2035
2036             gmx_mm256_decrement_4rvec_4ptr_swizzle_pd(fjptrA,fjptrB,fjptrC,fjptrD,
2037                                                       fjx0,fjy0,fjz0,fjx1,fjy1,fjz1,
2038                                                       fjx2,fjy2,fjz2,fjx3,fjy3,fjz3);
2039
2040             /* Inner loop uses 295 flops */
2041         }
2042
2043         /* End of innermost loop */
2044
2045         gmx_mm256_update_iforce_4atom_swizzle_pd(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,fix3,fiy3,fiz3,
2046                                                  f+i_coord_offset,fshift+i_shift_offset);
2047
2048         /* Increment number of inner iterations */
2049         inneriter                  += j_index_end - j_index_start;
2050
2051         /* Outer loop uses 24 flops */
2052     }
2053
2054     /* Increment number of outer iterations */
2055     outeriter        += nri;
2056
2057     /* Update outer/inner flops */
2058
2059     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W4W4_F,outeriter*24 + inneriter*295);
2060 }