src/gmxlib/nonbonded/nb_kernel_avx_256_double/nb_kernel_ElecCSTab_VdwNone_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_ElecCSTab_VdwNone_GeomW4W4_VF_avx_256_double
  38  * Electrostatics interaction: CubicSplineTable
  39  * VdW interaction:            None
  40  * Geometry:                   Water4-Water4
  41  * Calculate force/pot:        PotentialAndForce
  42  */
  43 void
  44 nb_kernel_ElecCSTab_VdwNone_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 *           vdwioffsetptr1;
  71     __m256d          ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
  72     real *           vdwioffsetptr2;
  73     __m256d          ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
  74     real *           vdwioffsetptr3;
  75     __m256d          ix3,iy3,iz3,fix3,fiy3,fiz3,iq3,isai3;
  76     int              vdwjidx1A,vdwjidx1B,vdwjidx1C,vdwjidx1D;
  77     __m256d          jx1,jy1,jz1,fjx1,fjy1,fjz1,jq1,isaj1;
  78     int              vdwjidx2A,vdwjidx2B,vdwjidx2C,vdwjidx2D;
  79     __m256d          jx2,jy2,jz2,fjx2,fjy2,fjz2,jq2,isaj2;
  80     int              vdwjidx3A,vdwjidx3B,vdwjidx3C,vdwjidx3D;
  81     __m256d          jx3,jy3,jz3,fjx3,fjy3,fjz3,jq3,isaj3;
  82     __m256d          dx11,dy11,dz11,rsq11,rinv11,rinvsq11,r11,qq11,c6_11,c12_11;
  83     __m256d          dx12,dy12,dz12,rsq12,rinv12,rinvsq12,r12,qq12,c6_12,c12_12;
  84     __m256d          dx13,dy13,dz13,rsq13,rinv13,rinvsq13,r13,qq13,c6_13,c12_13;
  85     __m256d          dx21,dy21,dz21,rsq21,rinv21,rinvsq21,r21,qq21,c6_21,c12_21;
  86     __m256d          dx22,dy22,dz22,rsq22,rinv22,rinvsq22,r22,qq22,c6_22,c12_22;
  87     __m256d          dx23,dy23,dz23,rsq23,rinv23,rinvsq23,r23,qq23,c6_23,c12_23;
  88     __m256d          dx31,dy31,dz31,rsq31,rinv31,rinvsq31,r31,qq31,c6_31,c12_31;
  89     __m256d          dx32,dy32,dz32,rsq32,rinv32,rinvsq32,r32,qq32,c6_32,c12_32;
  90     __m256d          dx33,dy33,dz33,rsq33,rinv33,rinvsq33,r33,qq33,c6_33,c12_33;
  91     __m256d          velec,felec,velecsum,facel,crf,krf,krf2;
  92     real             *charge;
  93     __m128i          vfitab;
  94     __m128i          ifour       = _mm_set1_epi32(4);
  95     __m256d          rt,vfeps,vftabscale,Y,F,G,H,Heps,Fp,VV,FF;
  96     real             *vftab;
  97     __m256d          dummy_mask,cutoff_mask;
  98     __m128           tmpmask0,tmpmask1;
  99     __m256d          signbit = _mm256_castsi256_pd( _mm256_set1_epi32(0x80000000) );
 100     __m256d          one     = _mm256_set1_pd(1.0);
 101     __m256d          two     = _mm256_set1_pd(2.0);
 102     x                = xx[0];
 103     f                = ff[0];
 104
 105     nri              = nlist->nri;
 106     iinr             = nlist->iinr;
 107     jindex           = nlist->jindex;
 108     jjnr             = nlist->jjnr;
 109     shiftidx         = nlist->shift;
 110     gid              = nlist->gid;
 111     shiftvec         = fr->shift_vec[0];
 112     fshift           = fr->fshift[0];
 113     facel            = _mm256_set1_pd(fr->epsfac);
 114     charge           = mdatoms->chargeA;
 115
 116     vftab            = kernel_data->table_elec->data;
 117     vftabscale       = _mm256_set1_pd(kernel_data->table_elec->scale);
 118
 119     /* Setup water-specific parameters */
 120     inr              = nlist->iinr[0];
 121     iq1              = _mm256_mul_pd(facel,_mm256_set1_pd(charge[inr+1]));
 122     iq2              = _mm256_mul_pd(facel,_mm256_set1_pd(charge[inr+2]));
 123     iq3              = _mm256_mul_pd(facel,_mm256_set1_pd(charge[inr+3]));
 124
 125     jq1              = _mm256_set1_pd(charge[inr+1]);
 126     jq2              = _mm256_set1_pd(charge[inr+2]);
 127     jq3              = _mm256_set1_pd(charge[inr+3]);
 128     qq11             = _mm256_mul_pd(iq1,jq1);
 129     qq12             = _mm256_mul_pd(iq1,jq2);
 130     qq13             = _mm256_mul_pd(iq1,jq3);
 131     qq21             = _mm256_mul_pd(iq2,jq1);
 132     qq22             = _mm256_mul_pd(iq2,jq2);
 133     qq23             = _mm256_mul_pd(iq2,jq3);
 134     qq31             = _mm256_mul_pd(iq3,jq1);
 135     qq32             = _mm256_mul_pd(iq3,jq2);
 136     qq33             = _mm256_mul_pd(iq3,jq3);
 137
 138     /* Avoid stupid compiler warnings */
 139     jnrA = jnrB = jnrC = jnrD = 0;
 140     j_coord_offsetA = 0;
 141     j_coord_offsetB = 0;
 142     j_coord_offsetC = 0;
 143     j_coord_offsetD = 0;
 144
 145     outeriter        = 0;
 146     inneriter        = 0;
 147
 148     for(iidx=0;iidx<4*DIM;iidx++)
 149     {
 150         scratch[iidx] = 0.0;
 151     }
 152
 153     /* Start outer loop over neighborlists */
 154     for(iidx=0; iidx<nri; iidx++)
 155     {
 156         /* Load shift vector for this list */
 157         i_shift_offset   = DIM*shiftidx[iidx];
 158
 159         /* Load limits for loop over neighbors */
 160         j_index_start    = jindex[iidx];
 161         j_index_end      = jindex[iidx+1];
 162
 163         /* Get outer coordinate index */
 164         inr              = iinr[iidx];
 165         i_coord_offset   = DIM*inr;
 166
 167         /* Load i particle coords and add shift vector */
 168         gmx_mm256_load_shift_and_3rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset+DIM,
 169                                                     &ix1,&iy1,&iz1,&ix2,&iy2,&iz2,&ix3,&iy3,&iz3);
 170
 171         fix1             = _mm256_setzero_pd();
 172         fiy1             = _mm256_setzero_pd();
 173         fiz1             = _mm256_setzero_pd();
 174         fix2             = _mm256_setzero_pd();
 175         fiy2             = _mm256_setzero_pd();
 176         fiz2             = _mm256_setzero_pd();
 177         fix3             = _mm256_setzero_pd();
 178         fiy3             = _mm256_setzero_pd();
 179         fiz3             = _mm256_setzero_pd();
 180
 181         /* Reset potential sums */
 182         velecsum         = _mm256_setzero_pd();
 183
 184         /* Start inner kernel loop */
 185         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
 186         {
 187
 188             /* Get j neighbor index, and coordinate index */
 189             jnrA             = jjnr[jidx];
 190             jnrB             = jjnr[jidx+1];
 191             jnrC             = jjnr[jidx+2];
 192             jnrD             = jjnr[jidx+3];
 193             j_coord_offsetA  = DIM*jnrA;
 194             j_coord_offsetB  = DIM*jnrB;
 195             j_coord_offsetC  = DIM*jnrC;
 196             j_coord_offsetD  = DIM*jnrD;
 197
 198             /* load j atom coordinates */
 199             gmx_mm256_load_3rvec_4ptr_swizzle_pd(x+j_coord_offsetA+DIM,x+j_coord_offsetB+DIM,
 200                                                  x+j_coord_offsetC+DIM,x+j_coord_offsetD+DIM,
 201                                                  &jx1,&jy1,&jz1,&jx2,&jy2,&jz2,&jx3,&jy3,&jz3);
 202
 203             /* Calculate displacement vector */
 204             dx11             = _mm256_sub_pd(ix1,jx1);
 205             dy11             = _mm256_sub_pd(iy1,jy1);
 206             dz11             = _mm256_sub_pd(iz1,jz1);
 207             dx12             = _mm256_sub_pd(ix1,jx2);
 208             dy12             = _mm256_sub_pd(iy1,jy2);
 209             dz12             = _mm256_sub_pd(iz1,jz2);
 210             dx13             = _mm256_sub_pd(ix1,jx3);
 211             dy13             = _mm256_sub_pd(iy1,jy3);
 212             dz13             = _mm256_sub_pd(iz1,jz3);
 213             dx21             = _mm256_sub_pd(ix2,jx1);
 214             dy21             = _mm256_sub_pd(iy2,jy1);
 215             dz21             = _mm256_sub_pd(iz2,jz1);
 216             dx22             = _mm256_sub_pd(ix2,jx2);
 217             dy22             = _mm256_sub_pd(iy2,jy2);
 218             dz22             = _mm256_sub_pd(iz2,jz2);
 219             dx23             = _mm256_sub_pd(ix2,jx3);
 220             dy23             = _mm256_sub_pd(iy2,jy3);
 221             dz23             = _mm256_sub_pd(iz2,jz3);
 222             dx31             = _mm256_sub_pd(ix3,jx1);
 223             dy31             = _mm256_sub_pd(iy3,jy1);
 224             dz31             = _mm256_sub_pd(iz3,jz1);
 225             dx32             = _mm256_sub_pd(ix3,jx2);
 226             dy32             = _mm256_sub_pd(iy3,jy2);
 227             dz32             = _mm256_sub_pd(iz3,jz2);
 228             dx33             = _mm256_sub_pd(ix3,jx3);
 229             dy33             = _mm256_sub_pd(iy3,jy3);
 230             dz33             = _mm256_sub_pd(iz3,jz3);
 231
 232             /* Calculate squared distance and things based on it */
 233             rsq11            = gmx_mm256_calc_rsq_pd(dx11,dy11,dz11);
 234             rsq12            = gmx_mm256_calc_rsq_pd(dx12,dy12,dz12);
 235             rsq13            = gmx_mm256_calc_rsq_pd(dx13,dy13,dz13);
 236             rsq21            = gmx_mm256_calc_rsq_pd(dx21,dy21,dz21);
 237             rsq22            = gmx_mm256_calc_rsq_pd(dx22,dy22,dz22);
 238             rsq23            = gmx_mm256_calc_rsq_pd(dx23,dy23,dz23);
 239             rsq31            = gmx_mm256_calc_rsq_pd(dx31,dy31,dz31);
 240             rsq32            = gmx_mm256_calc_rsq_pd(dx32,dy32,dz32);
 241             rsq33            = gmx_mm256_calc_rsq_pd(dx33,dy33,dz33);
 242
 243             rinv11           = gmx_mm256_invsqrt_pd(rsq11);
 244             rinv12           = gmx_mm256_invsqrt_pd(rsq12);
 245             rinv13           = gmx_mm256_invsqrt_pd(rsq13);
 246             rinv21           = gmx_mm256_invsqrt_pd(rsq21);
 247             rinv22           = gmx_mm256_invsqrt_pd(rsq22);
 248             rinv23           = gmx_mm256_invsqrt_pd(rsq23);
 249             rinv31           = gmx_mm256_invsqrt_pd(rsq31);
 250             rinv32           = gmx_mm256_invsqrt_pd(rsq32);
 251             rinv33           = gmx_mm256_invsqrt_pd(rsq33);
 252
 253             fjx1             = _mm256_setzero_pd();
 254             fjy1             = _mm256_setzero_pd();
 255             fjz1             = _mm256_setzero_pd();
 256             fjx2             = _mm256_setzero_pd();
 257             fjy2             = _mm256_setzero_pd();
 258             fjz2             = _mm256_setzero_pd();
 259             fjx3             = _mm256_setzero_pd();
 260             fjy3             = _mm256_setzero_pd();
 261             fjz3             = _mm256_setzero_pd();
 262
 263             /**************************
 264              * CALCULATE INTERACTIONS *
 265              **************************/
 266
 267             r11              = _mm256_mul_pd(rsq11,rinv11);
 268
 269             /* Calculate table index by multiplying r with table scale and truncate to integer */
 270             rt               = _mm256_mul_pd(r11,vftabscale);
 271             vfitab           = _mm256_cvttpd_epi32(rt);
 272             vfeps            = _mm256_sub_pd(rt,_mm256_round_pd(rt, _MM_FROUND_FLOOR));
 273             vfitab           = _mm_slli_epi32(vfitab,2);
 274
 275             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 276             Y                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
 277             F                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
 278             G                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,2) );
 279             H                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,3) );
 280             GMX_MM256_FULLTRANSPOSE4_PD(Y,F,G,H);
 281             Heps             = _mm256_mul_pd(vfeps,H);
 282             Fp               = _mm256_add_pd(F,_mm256_mul_pd(vfeps,_mm256_add_pd(G,Heps)));
 283             VV               = _mm256_add_pd(Y,_mm256_mul_pd(vfeps,Fp));
 284             velec            = _mm256_mul_pd(qq11,VV);
 285             FF               = _mm256_add_pd(Fp,_mm256_mul_pd(vfeps,_mm256_add_pd(G,_mm256_add_pd(Heps,Heps))));
 286             felec            = _mm256_xor_pd(signbit,_mm256_mul_pd(_mm256_mul_pd(qq11,FF),_mm256_mul_pd(vftabscale,rinv11)));
 287
 288             /* Update potential sum for this i atom from the interaction with this j atom. */
 289             velecsum         = _mm256_add_pd(velecsum,velec);
 290
 291             fscal            = felec;
 292
 293             /* Calculate temporary vectorial force */
 294             tx               = _mm256_mul_pd(fscal,dx11);
 295             ty               = _mm256_mul_pd(fscal,dy11);
 296             tz               = _mm256_mul_pd(fscal,dz11);
 297
 298             /* Update vectorial force */
 299             fix1             = _mm256_add_pd(fix1,tx);
 300             fiy1             = _mm256_add_pd(fiy1,ty);
 301             fiz1             = _mm256_add_pd(fiz1,tz);
 302
 303             fjx1             = _mm256_add_pd(fjx1,tx);
 304             fjy1             = _mm256_add_pd(fjy1,ty);
 305             fjz1             = _mm256_add_pd(fjz1,tz);
 306
 307             /**************************
 308              * CALCULATE INTERACTIONS *
 309              **************************/
 310
 311             r12              = _mm256_mul_pd(rsq12,rinv12);
 312
 313             /* Calculate table index by multiplying r with table scale and truncate to integer */
 314             rt               = _mm256_mul_pd(r12,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,2);
 318
 319             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 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             velec            = _mm256_mul_pd(qq12,VV);
 329             FF               = _mm256_add_pd(Fp,_mm256_mul_pd(vfeps,_mm256_add_pd(G,_mm256_add_pd(Heps,Heps))));
 330             felec            = _mm256_xor_pd(signbit,_mm256_mul_pd(_mm256_mul_pd(qq12,FF),_mm256_mul_pd(vftabscale,rinv12)));
 331
 332             /* Update potential sum for this i atom from the interaction with this j atom. */
 333             velecsum         = _mm256_add_pd(velecsum,velec);
 334
 335             fscal            = felec;
 336
 337             /* Calculate temporary vectorial force */
 338             tx               = _mm256_mul_pd(fscal,dx12);
 339             ty               = _mm256_mul_pd(fscal,dy12);
 340             tz               = _mm256_mul_pd(fscal,dz12);
 341
 342             /* Update vectorial force */
 343             fix1             = _mm256_add_pd(fix1,tx);
 344             fiy1             = _mm256_add_pd(fiy1,ty);
 345             fiz1             = _mm256_add_pd(fiz1,tz);
 346
 347             fjx2             = _mm256_add_pd(fjx2,tx);
 348             fjy2             = _mm256_add_pd(fjy2,ty);
 349             fjz2             = _mm256_add_pd(fjz2,tz);
 350
 351             /**************************
 352              * CALCULATE INTERACTIONS *
 353              **************************/
 354
 355             r13              = _mm256_mul_pd(rsq13,rinv13);
 356
 357             /* Calculate table index by multiplying r with table scale and truncate to integer */
 358             rt               = _mm256_mul_pd(r13,vftabscale);
 359             vfitab           = _mm256_cvttpd_epi32(rt);
 360             vfeps            = _mm256_sub_pd(rt,_mm256_round_pd(rt, _MM_FROUND_FLOOR));
 361             vfitab           = _mm_slli_epi32(vfitab,2);
 362
 363             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 364             Y                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
 365             F                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
 366             G                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,2) );
 367             H                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,3) );
 368             GMX_MM256_FULLTRANSPOSE4_PD(Y,F,G,H);
 369             Heps             = _mm256_mul_pd(vfeps,H);
 370             Fp               = _mm256_add_pd(F,_mm256_mul_pd(vfeps,_mm256_add_pd(G,Heps)));
 371             VV               = _mm256_add_pd(Y,_mm256_mul_pd(vfeps,Fp));
 372             velec            = _mm256_mul_pd(qq13,VV);
 373             FF               = _mm256_add_pd(Fp,_mm256_mul_pd(vfeps,_mm256_add_pd(G,_mm256_add_pd(Heps,Heps))));
 374             felec            = _mm256_xor_pd(signbit,_mm256_mul_pd(_mm256_mul_pd(qq13,FF),_mm256_mul_pd(vftabscale,rinv13)));
 375
 376             /* Update potential sum for this i atom from the interaction with this j atom. */
 377             velecsum         = _mm256_add_pd(velecsum,velec);
 378
 379             fscal            = felec;
 380
 381             /* Calculate temporary vectorial force */
 382             tx               = _mm256_mul_pd(fscal,dx13);
 383             ty               = _mm256_mul_pd(fscal,dy13);
 384             tz               = _mm256_mul_pd(fscal,dz13);
 385
 386             /* Update vectorial force */
 387             fix1             = _mm256_add_pd(fix1,tx);
 388             fiy1             = _mm256_add_pd(fiy1,ty);
 389             fiz1             = _mm256_add_pd(fiz1,tz);
 390
 391             fjx3             = _mm256_add_pd(fjx3,tx);
 392             fjy3             = _mm256_add_pd(fjy3,ty);
 393             fjz3             = _mm256_add_pd(fjz3,tz);
 394
 395             /**************************
 396              * CALCULATE INTERACTIONS *
 397              **************************/
 398
 399             r21              = _mm256_mul_pd(rsq21,rinv21);
 400
 401             /* Calculate table index by multiplying r with table scale and truncate to integer */
 402             rt               = _mm256_mul_pd(r21,vftabscale);
 403             vfitab           = _mm256_cvttpd_epi32(rt);
 404             vfeps            = _mm256_sub_pd(rt,_mm256_round_pd(rt, _MM_FROUND_FLOOR));
 405             vfitab           = _mm_slli_epi32(vfitab,2);
 406
 407             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 408             Y                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
 409             F                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
 410             G                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,2) );
 411             H                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,3) );
 412             GMX_MM256_FULLTRANSPOSE4_PD(Y,F,G,H);
 413             Heps             = _mm256_mul_pd(vfeps,H);
 414             Fp               = _mm256_add_pd(F,_mm256_mul_pd(vfeps,_mm256_add_pd(G,Heps)));
 415             VV               = _mm256_add_pd(Y,_mm256_mul_pd(vfeps,Fp));
 416             velec            = _mm256_mul_pd(qq21,VV);
 417             FF               = _mm256_add_pd(Fp,_mm256_mul_pd(vfeps,_mm256_add_pd(G,_mm256_add_pd(Heps,Heps))));
 418             felec            = _mm256_xor_pd(signbit,_mm256_mul_pd(_mm256_mul_pd(qq21,FF),_mm256_mul_pd(vftabscale,rinv21)));
 419
 420             /* Update potential sum for this i atom from the interaction with this j atom. */
 421             velecsum         = _mm256_add_pd(velecsum,velec);
 422
 423             fscal            = felec;
 424
 425             /* Calculate temporary vectorial force */
 426             tx               = _mm256_mul_pd(fscal,dx21);
 427             ty               = _mm256_mul_pd(fscal,dy21);
 428             tz               = _mm256_mul_pd(fscal,dz21);
 429
 430             /* Update vectorial force */
 431             fix2             = _mm256_add_pd(fix2,tx);
 432             fiy2             = _mm256_add_pd(fiy2,ty);
 433             fiz2             = _mm256_add_pd(fiz2,tz);
 434
 435             fjx1             = _mm256_add_pd(fjx1,tx);
 436             fjy1             = _mm256_add_pd(fjy1,ty);
 437             fjz1             = _mm256_add_pd(fjz1,tz);
 438
 439             /**************************
 440              * CALCULATE INTERACTIONS *
 441              **************************/
 442
 443             r22              = _mm256_mul_pd(rsq22,rinv22);
 444
 445             /* Calculate table index by multiplying r with table scale and truncate to integer */
 446             rt               = _mm256_mul_pd(r22,vftabscale);
 447             vfitab           = _mm256_cvttpd_epi32(rt);
 448             vfeps            = _mm256_sub_pd(rt,_mm256_round_pd(rt, _MM_FROUND_FLOOR));
 449             vfitab           = _mm_slli_epi32(vfitab,2);
 450
 451             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 452             Y                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
 453             F                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
 454             G                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,2) );
 455             H                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,3) );
 456             GMX_MM256_FULLTRANSPOSE4_PD(Y,F,G,H);
 457             Heps             = _mm256_mul_pd(vfeps,H);
 458             Fp               = _mm256_add_pd(F,_mm256_mul_pd(vfeps,_mm256_add_pd(G,Heps)));
 459             VV               = _mm256_add_pd(Y,_mm256_mul_pd(vfeps,Fp));
 460             velec            = _mm256_mul_pd(qq22,VV);
 461             FF               = _mm256_add_pd(Fp,_mm256_mul_pd(vfeps,_mm256_add_pd(G,_mm256_add_pd(Heps,Heps))));
 462             felec            = _mm256_xor_pd(signbit,_mm256_mul_pd(_mm256_mul_pd(qq22,FF),_mm256_mul_pd(vftabscale,rinv22)));
 463
 464             /* Update potential sum for this i atom from the interaction with this j atom. */
 465             velecsum         = _mm256_add_pd(velecsum,velec);
 466
 467             fscal            = felec;
 468
 469             /* Calculate temporary vectorial force */
 470             tx               = _mm256_mul_pd(fscal,dx22);
 471             ty               = _mm256_mul_pd(fscal,dy22);
 472             tz               = _mm256_mul_pd(fscal,dz22);
 473
 474             /* Update vectorial force */
 475             fix2             = _mm256_add_pd(fix2,tx);
 476             fiy2             = _mm256_add_pd(fiy2,ty);
 477             fiz2             = _mm256_add_pd(fiz2,tz);
 478
 479             fjx2             = _mm256_add_pd(fjx2,tx);
 480             fjy2             = _mm256_add_pd(fjy2,ty);
 481             fjz2             = _mm256_add_pd(fjz2,tz);
 482
 483             /**************************
 484              * CALCULATE INTERACTIONS *
 485              **************************/
 486
 487             r23              = _mm256_mul_pd(rsq23,rinv23);
 488
 489             /* Calculate table index by multiplying r with table scale and truncate to integer */
 490             rt               = _mm256_mul_pd(r23,vftabscale);
 491             vfitab           = _mm256_cvttpd_epi32(rt);
 492             vfeps            = _mm256_sub_pd(rt,_mm256_round_pd(rt, _MM_FROUND_FLOOR));
 493             vfitab           = _mm_slli_epi32(vfitab,2);
 494
 495             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 496             Y                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
 497             F                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
 498             G                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,2) );
 499             H                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,3) );
 500             GMX_MM256_FULLTRANSPOSE4_PD(Y,F,G,H);
 501             Heps             = _mm256_mul_pd(vfeps,H);
 502             Fp               = _mm256_add_pd(F,_mm256_mul_pd(vfeps,_mm256_add_pd(G,Heps)));
 503             VV               = _mm256_add_pd(Y,_mm256_mul_pd(vfeps,Fp));
 504             velec            = _mm256_mul_pd(qq23,VV);
 505             FF               = _mm256_add_pd(Fp,_mm256_mul_pd(vfeps,_mm256_add_pd(G,_mm256_add_pd(Heps,Heps))));
 506             felec            = _mm256_xor_pd(signbit,_mm256_mul_pd(_mm256_mul_pd(qq23,FF),_mm256_mul_pd(vftabscale,rinv23)));
 507
 508             /* Update potential sum for this i atom from the interaction with this j atom. */
 509             velecsum         = _mm256_add_pd(velecsum,velec);
 510
 511             fscal            = felec;
 512
 513             /* Calculate temporary vectorial force */
 514             tx               = _mm256_mul_pd(fscal,dx23);
 515             ty               = _mm256_mul_pd(fscal,dy23);
 516             tz               = _mm256_mul_pd(fscal,dz23);
 517
 518             /* Update vectorial force */
 519             fix2             = _mm256_add_pd(fix2,tx);
 520             fiy2             = _mm256_add_pd(fiy2,ty);
 521             fiz2             = _mm256_add_pd(fiz2,tz);
 522
 523             fjx3             = _mm256_add_pd(fjx3,tx);
 524             fjy3             = _mm256_add_pd(fjy3,ty);
 525             fjz3             = _mm256_add_pd(fjz3,tz);
 526
 527             /**************************
 528              * CALCULATE INTERACTIONS *
 529              **************************/
 530
 531             r31              = _mm256_mul_pd(rsq31,rinv31);
 532
 533             /* Calculate table index by multiplying r with table scale and truncate to integer */
 534             rt               = _mm256_mul_pd(r31,vftabscale);
 535             vfitab           = _mm256_cvttpd_epi32(rt);
 536             vfeps            = _mm256_sub_pd(rt,_mm256_round_pd(rt, _MM_FROUND_FLOOR));
 537             vfitab           = _mm_slli_epi32(vfitab,2);
 538
 539             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 540             Y                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
 541             F                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
 542             G                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,2) );
 543             H                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,3) );
 544             GMX_MM256_FULLTRANSPOSE4_PD(Y,F,G,H);
 545             Heps             = _mm256_mul_pd(vfeps,H);
 546             Fp               = _mm256_add_pd(F,_mm256_mul_pd(vfeps,_mm256_add_pd(G,Heps)));
 547             VV               = _mm256_add_pd(Y,_mm256_mul_pd(vfeps,Fp));
 548             velec            = _mm256_mul_pd(qq31,VV);
 549             FF               = _mm256_add_pd(Fp,_mm256_mul_pd(vfeps,_mm256_add_pd(G,_mm256_add_pd(Heps,Heps))));
 550             felec            = _mm256_xor_pd(signbit,_mm256_mul_pd(_mm256_mul_pd(qq31,FF),_mm256_mul_pd(vftabscale,rinv31)));
 551
 552             /* Update potential sum for this i atom from the interaction with this j atom. */
 553             velecsum         = _mm256_add_pd(velecsum,velec);
 554
 555             fscal            = felec;
 556
 557             /* Calculate temporary vectorial force */
 558             tx               = _mm256_mul_pd(fscal,dx31);
 559             ty               = _mm256_mul_pd(fscal,dy31);
 560             tz               = _mm256_mul_pd(fscal,dz31);
 561
 562             /* Update vectorial force */
 563             fix3             = _mm256_add_pd(fix3,tx);
 564             fiy3             = _mm256_add_pd(fiy3,ty);
 565             fiz3             = _mm256_add_pd(fiz3,tz);
 566
 567             fjx1             = _mm256_add_pd(fjx1,tx);
 568             fjy1             = _mm256_add_pd(fjy1,ty);
 569             fjz1             = _mm256_add_pd(fjz1,tz);
 570
 571             /**************************
 572              * CALCULATE INTERACTIONS *
 573              **************************/
 574
 575             r32              = _mm256_mul_pd(rsq32,rinv32);
 576
 577             /* Calculate table index by multiplying r with table scale and truncate to integer */
 578             rt               = _mm256_mul_pd(r32,vftabscale);
 579             vfitab           = _mm256_cvttpd_epi32(rt);
 580             vfeps            = _mm256_sub_pd(rt,_mm256_round_pd(rt, _MM_FROUND_FLOOR));
 581             vfitab           = _mm_slli_epi32(vfitab,2);
 582
 583             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 584             Y                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
 585             F                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
 586             G                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,2) );
 587             H                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,3) );
 588             GMX_MM256_FULLTRANSPOSE4_PD(Y,F,G,H);
 589             Heps             = _mm256_mul_pd(vfeps,H);
 590             Fp               = _mm256_add_pd(F,_mm256_mul_pd(vfeps,_mm256_add_pd(G,Heps)));
 591             VV               = _mm256_add_pd(Y,_mm256_mul_pd(vfeps,Fp));
 592             velec            = _mm256_mul_pd(qq32,VV);
 593             FF               = _mm256_add_pd(Fp,_mm256_mul_pd(vfeps,_mm256_add_pd(G,_mm256_add_pd(Heps,Heps))));
 594             felec            = _mm256_xor_pd(signbit,_mm256_mul_pd(_mm256_mul_pd(qq32,FF),_mm256_mul_pd(vftabscale,rinv32)));
 595
 596             /* Update potential sum for this i atom from the interaction with this j atom. */
 597             velecsum         = _mm256_add_pd(velecsum,velec);
 598
 599             fscal            = felec;
 600
 601             /* Calculate temporary vectorial force */
 602             tx               = _mm256_mul_pd(fscal,dx32);
 603             ty               = _mm256_mul_pd(fscal,dy32);
 604             tz               = _mm256_mul_pd(fscal,dz32);
 605
 606             /* Update vectorial force */
 607             fix3             = _mm256_add_pd(fix3,tx);
 608             fiy3             = _mm256_add_pd(fiy3,ty);
 609             fiz3             = _mm256_add_pd(fiz3,tz);
 610
 611             fjx2             = _mm256_add_pd(fjx2,tx);
 612             fjy2             = _mm256_add_pd(fjy2,ty);
 613             fjz2             = _mm256_add_pd(fjz2,tz);
 614
 615             /**************************
 616              * CALCULATE INTERACTIONS *
 617              **************************/
 618
 619             r33              = _mm256_mul_pd(rsq33,rinv33);
 620
 621             /* Calculate table index by multiplying r with table scale and truncate to integer */
 622             rt               = _mm256_mul_pd(r33,vftabscale);
 623             vfitab           = _mm256_cvttpd_epi32(rt);
 624             vfeps            = _mm256_sub_pd(rt,_mm256_round_pd(rt, _MM_FROUND_FLOOR));
 625             vfitab           = _mm_slli_epi32(vfitab,2);
 626
 627             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 628             Y                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
 629             F                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
 630             G                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,2) );
 631             H                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,3) );
 632             GMX_MM256_FULLTRANSPOSE4_PD(Y,F,G,H);
 633             Heps             = _mm256_mul_pd(vfeps,H);
 634             Fp               = _mm256_add_pd(F,_mm256_mul_pd(vfeps,_mm256_add_pd(G,Heps)));
 635             VV               = _mm256_add_pd(Y,_mm256_mul_pd(vfeps,Fp));
 636             velec            = _mm256_mul_pd(qq33,VV);
 637             FF               = _mm256_add_pd(Fp,_mm256_mul_pd(vfeps,_mm256_add_pd(G,_mm256_add_pd(Heps,Heps))));
 638             felec            = _mm256_xor_pd(signbit,_mm256_mul_pd(_mm256_mul_pd(qq33,FF),_mm256_mul_pd(vftabscale,rinv33)));
 639
 640             /* Update potential sum for this i atom from the interaction with this j atom. */
 641             velecsum         = _mm256_add_pd(velecsum,velec);
 642
 643             fscal            = felec;
 644
 645             /* Calculate temporary vectorial force */
 646             tx               = _mm256_mul_pd(fscal,dx33);
 647             ty               = _mm256_mul_pd(fscal,dy33);
 648             tz               = _mm256_mul_pd(fscal,dz33);
 649
 650             /* Update vectorial force */
 651             fix3             = _mm256_add_pd(fix3,tx);
 652             fiy3             = _mm256_add_pd(fiy3,ty);
 653             fiz3             = _mm256_add_pd(fiz3,tz);
 654
 655             fjx3             = _mm256_add_pd(fjx3,tx);
 656             fjy3             = _mm256_add_pd(fjy3,ty);
 657             fjz3             = _mm256_add_pd(fjz3,tz);
 658
 659             fjptrA             = f+j_coord_offsetA;
 660             fjptrB             = f+j_coord_offsetB;
 661             fjptrC             = f+j_coord_offsetC;
 662             fjptrD             = f+j_coord_offsetD;
 663
 664             gmx_mm256_decrement_3rvec_4ptr_swizzle_pd(fjptrA+DIM,fjptrB+DIM,fjptrC+DIM,fjptrD+DIM,
 665                                                       fjx1,fjy1,fjz1,fjx2,fjy2,fjz2,fjx3,fjy3,fjz3);
 666
 667             /* Inner loop uses 387 flops */
 668         }
 669
 670         if(jidx<j_index_end)
 671         {
 672
 673             /* Get j neighbor index, and coordinate index */
 674             jnrlistA         = jjnr[jidx];
 675             jnrlistB         = jjnr[jidx+1];
 676             jnrlistC         = jjnr[jidx+2];
 677             jnrlistD         = jjnr[jidx+3];
 678             /* Sign of each element will be negative for non-real atoms.
 679              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 680              * so use it as val = _mm_andnot_pd(mask,val) to clear dummy entries.
 681              */
 682             tmpmask0 = gmx_mm_castsi128_pd(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
 683
 684             tmpmask1 = _mm_permute_ps(tmpmask0,_GMX_MM_PERMUTE(3,3,2,2));
 685             tmpmask0 = _mm_permute_ps(tmpmask0,_GMX_MM_PERMUTE(1,1,0,0));
 686             dummy_mask = _mm256_castps_pd(gmx_mm256_set_m128(tmpmask1,tmpmask0));
 687
 688             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
 689             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
 690             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
 691             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
 692             j_coord_offsetA  = DIM*jnrA;
 693             j_coord_offsetB  = DIM*jnrB;
 694             j_coord_offsetC  = DIM*jnrC;
 695             j_coord_offsetD  = DIM*jnrD;
 696
 697             /* load j atom coordinates */
 698             gmx_mm256_load_3rvec_4ptr_swizzle_pd(x+j_coord_offsetA+DIM,x+j_coord_offsetB+DIM,
 699                                                  x+j_coord_offsetC+DIM,x+j_coord_offsetD+DIM,
 700                                                  &jx1,&jy1,&jz1,&jx2,&jy2,&jz2,&jx3,&jy3,&jz3);
 701
 702             /* Calculate displacement vector */
 703             dx11             = _mm256_sub_pd(ix1,jx1);
 704             dy11             = _mm256_sub_pd(iy1,jy1);
 705             dz11             = _mm256_sub_pd(iz1,jz1);
 706             dx12             = _mm256_sub_pd(ix1,jx2);
 707             dy12             = _mm256_sub_pd(iy1,jy2);
 708             dz12             = _mm256_sub_pd(iz1,jz2);
 709             dx13             = _mm256_sub_pd(ix1,jx3);
 710             dy13             = _mm256_sub_pd(iy1,jy3);
 711             dz13             = _mm256_sub_pd(iz1,jz3);
 712             dx21             = _mm256_sub_pd(ix2,jx1);
 713             dy21             = _mm256_sub_pd(iy2,jy1);
 714             dz21             = _mm256_sub_pd(iz2,jz1);
 715             dx22             = _mm256_sub_pd(ix2,jx2);
 716             dy22             = _mm256_sub_pd(iy2,jy2);
 717             dz22             = _mm256_sub_pd(iz2,jz2);
 718             dx23             = _mm256_sub_pd(ix2,jx3);
 719             dy23             = _mm256_sub_pd(iy2,jy3);
 720             dz23             = _mm256_sub_pd(iz2,jz3);
 721             dx31             = _mm256_sub_pd(ix3,jx1);
 722             dy31             = _mm256_sub_pd(iy3,jy1);
 723             dz31             = _mm256_sub_pd(iz3,jz1);
 724             dx32             = _mm256_sub_pd(ix3,jx2);
 725             dy32             = _mm256_sub_pd(iy3,jy2);
 726             dz32             = _mm256_sub_pd(iz3,jz2);
 727             dx33             = _mm256_sub_pd(ix3,jx3);
 728             dy33             = _mm256_sub_pd(iy3,jy3);
 729             dz33             = _mm256_sub_pd(iz3,jz3);
 730
 731             /* Calculate squared distance and things based on it */
 732             rsq11            = gmx_mm256_calc_rsq_pd(dx11,dy11,dz11);
 733             rsq12            = gmx_mm256_calc_rsq_pd(dx12,dy12,dz12);
 734             rsq13            = gmx_mm256_calc_rsq_pd(dx13,dy13,dz13);
 735             rsq21            = gmx_mm256_calc_rsq_pd(dx21,dy21,dz21);
 736             rsq22            = gmx_mm256_calc_rsq_pd(dx22,dy22,dz22);
 737             rsq23            = gmx_mm256_calc_rsq_pd(dx23,dy23,dz23);
 738             rsq31            = gmx_mm256_calc_rsq_pd(dx31,dy31,dz31);
 739             rsq32            = gmx_mm256_calc_rsq_pd(dx32,dy32,dz32);
 740             rsq33            = gmx_mm256_calc_rsq_pd(dx33,dy33,dz33);
 741
 742             rinv11           = gmx_mm256_invsqrt_pd(rsq11);
 743             rinv12           = gmx_mm256_invsqrt_pd(rsq12);
 744             rinv13           = gmx_mm256_invsqrt_pd(rsq13);
 745             rinv21           = gmx_mm256_invsqrt_pd(rsq21);
 746             rinv22           = gmx_mm256_invsqrt_pd(rsq22);
 747             rinv23           = gmx_mm256_invsqrt_pd(rsq23);
 748             rinv31           = gmx_mm256_invsqrt_pd(rsq31);
 749             rinv32           = gmx_mm256_invsqrt_pd(rsq32);
 750             rinv33           = gmx_mm256_invsqrt_pd(rsq33);
 751
 752             fjx1             = _mm256_setzero_pd();
 753             fjy1             = _mm256_setzero_pd();
 754             fjz1             = _mm256_setzero_pd();
 755             fjx2             = _mm256_setzero_pd();
 756             fjy2             = _mm256_setzero_pd();
 757             fjz2             = _mm256_setzero_pd();
 758             fjx3             = _mm256_setzero_pd();
 759             fjy3             = _mm256_setzero_pd();
 760             fjz3             = _mm256_setzero_pd();
 761
 762             /**************************
 763              * CALCULATE INTERACTIONS *
 764              **************************/
 765
 766             r11              = _mm256_mul_pd(rsq11,rinv11);
 767             r11              = _mm256_andnot_pd(dummy_mask,r11);
 768
 769             /* Calculate table index by multiplying r with table scale and truncate to integer */
 770             rt               = _mm256_mul_pd(r11,vftabscale);
 771             vfitab           = _mm256_cvttpd_epi32(rt);
 772             vfeps            = _mm256_sub_pd(rt,_mm256_round_pd(rt, _MM_FROUND_FLOOR));
 773             vfitab           = _mm_slli_epi32(vfitab,2);
 774
 775             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 776             Y                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
 777             F                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
 778             G                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,2) );
 779             H                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,3) );
 780             GMX_MM256_FULLTRANSPOSE4_PD(Y,F,G,H);
 781             Heps             = _mm256_mul_pd(vfeps,H);
 782             Fp               = _mm256_add_pd(F,_mm256_mul_pd(vfeps,_mm256_add_pd(G,Heps)));
 783             VV               = _mm256_add_pd(Y,_mm256_mul_pd(vfeps,Fp));
 784             velec            = _mm256_mul_pd(qq11,VV);
 785             FF               = _mm256_add_pd(Fp,_mm256_mul_pd(vfeps,_mm256_add_pd(G,_mm256_add_pd(Heps,Heps))));
 786             felec            = _mm256_xor_pd(signbit,_mm256_mul_pd(_mm256_mul_pd(qq11,FF),_mm256_mul_pd(vftabscale,rinv11)));
 787
 788             /* Update potential sum for this i atom from the interaction with this j atom. */
 789             velec            = _mm256_andnot_pd(dummy_mask,velec);
 790             velecsum         = _mm256_add_pd(velecsum,velec);
 791
 792             fscal            = felec;
 793
 794             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
 795
 796             /* Calculate temporary vectorial force */
 797             tx               = _mm256_mul_pd(fscal,dx11);
 798             ty               = _mm256_mul_pd(fscal,dy11);
 799             tz               = _mm256_mul_pd(fscal,dz11);
 800
 801             /* Update vectorial force */
 802             fix1             = _mm256_add_pd(fix1,tx);
 803             fiy1             = _mm256_add_pd(fiy1,ty);
 804             fiz1             = _mm256_add_pd(fiz1,tz);
 805
 806             fjx1             = _mm256_add_pd(fjx1,tx);
 807             fjy1             = _mm256_add_pd(fjy1,ty);
 808             fjz1             = _mm256_add_pd(fjz1,tz);
 809
 810             /**************************
 811              * CALCULATE INTERACTIONS *
 812              **************************/
 813
 814             r12              = _mm256_mul_pd(rsq12,rinv12);
 815             r12              = _mm256_andnot_pd(dummy_mask,r12);
 816
 817             /* Calculate table index by multiplying r with table scale and truncate to integer */
 818             rt               = _mm256_mul_pd(r12,vftabscale);
 819             vfitab           = _mm256_cvttpd_epi32(rt);
 820             vfeps            = _mm256_sub_pd(rt,_mm256_round_pd(rt, _MM_FROUND_FLOOR));
 821             vfitab           = _mm_slli_epi32(vfitab,2);
 822
 823             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 824             Y                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
 825             F                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
 826             G                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,2) );
 827             H                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,3) );
 828             GMX_MM256_FULLTRANSPOSE4_PD(Y,F,G,H);
 829             Heps             = _mm256_mul_pd(vfeps,H);
 830             Fp               = _mm256_add_pd(F,_mm256_mul_pd(vfeps,_mm256_add_pd(G,Heps)));
 831             VV               = _mm256_add_pd(Y,_mm256_mul_pd(vfeps,Fp));
 832             velec            = _mm256_mul_pd(qq12,VV);
 833             FF               = _mm256_add_pd(Fp,_mm256_mul_pd(vfeps,_mm256_add_pd(G,_mm256_add_pd(Heps,Heps))));
 834             felec            = _mm256_xor_pd(signbit,_mm256_mul_pd(_mm256_mul_pd(qq12,FF),_mm256_mul_pd(vftabscale,rinv12)));
 835
 836             /* Update potential sum for this i atom from the interaction with this j atom. */
 837             velec            = _mm256_andnot_pd(dummy_mask,velec);
 838             velecsum         = _mm256_add_pd(velecsum,velec);
 839
 840             fscal            = felec;
 841
 842             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
 843
 844             /* Calculate temporary vectorial force */
 845             tx               = _mm256_mul_pd(fscal,dx12);
 846             ty               = _mm256_mul_pd(fscal,dy12);
 847             tz               = _mm256_mul_pd(fscal,dz12);
 848
 849             /* Update vectorial force */
 850             fix1             = _mm256_add_pd(fix1,tx);
 851             fiy1             = _mm256_add_pd(fiy1,ty);
 852             fiz1             = _mm256_add_pd(fiz1,tz);
 853
 854             fjx2             = _mm256_add_pd(fjx2,tx);
 855             fjy2             = _mm256_add_pd(fjy2,ty);
 856             fjz2             = _mm256_add_pd(fjz2,tz);
 857
 858             /**************************
 859              * CALCULATE INTERACTIONS *
 860              **************************/
 861
 862             r13              = _mm256_mul_pd(rsq13,rinv13);
 863             r13              = _mm256_andnot_pd(dummy_mask,r13);
 864
 865             /* Calculate table index by multiplying r with table scale and truncate to integer */
 866             rt               = _mm256_mul_pd(r13,vftabscale);
 867             vfitab           = _mm256_cvttpd_epi32(rt);
 868             vfeps            = _mm256_sub_pd(rt,_mm256_round_pd(rt, _MM_FROUND_FLOOR));
 869             vfitab           = _mm_slli_epi32(vfitab,2);
 870
 871             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 872             Y                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
 873             F                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
 874             G                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,2) );
 875             H                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,3) );
 876             GMX_MM256_FULLTRANSPOSE4_PD(Y,F,G,H);
 877             Heps             = _mm256_mul_pd(vfeps,H);
 878             Fp               = _mm256_add_pd(F,_mm256_mul_pd(vfeps,_mm256_add_pd(G,Heps)));
 879             VV               = _mm256_add_pd(Y,_mm256_mul_pd(vfeps,Fp));
 880             velec            = _mm256_mul_pd(qq13,VV);
 881             FF               = _mm256_add_pd(Fp,_mm256_mul_pd(vfeps,_mm256_add_pd(G,_mm256_add_pd(Heps,Heps))));
 882             felec            = _mm256_xor_pd(signbit,_mm256_mul_pd(_mm256_mul_pd(qq13,FF),_mm256_mul_pd(vftabscale,rinv13)));
 883
 884             /* Update potential sum for this i atom from the interaction with this j atom. */
 885             velec            = _mm256_andnot_pd(dummy_mask,velec);
 886             velecsum         = _mm256_add_pd(velecsum,velec);
 887
 888             fscal            = felec;
 889
 890             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
 891
 892             /* Calculate temporary vectorial force */
 893             tx               = _mm256_mul_pd(fscal,dx13);
 894             ty               = _mm256_mul_pd(fscal,dy13);
 895             tz               = _mm256_mul_pd(fscal,dz13);
 896
 897             /* Update vectorial force */
 898             fix1             = _mm256_add_pd(fix1,tx);
 899             fiy1             = _mm256_add_pd(fiy1,ty);
 900             fiz1             = _mm256_add_pd(fiz1,tz);
 901
 902             fjx3             = _mm256_add_pd(fjx3,tx);
 903             fjy3             = _mm256_add_pd(fjy3,ty);
 904             fjz3             = _mm256_add_pd(fjz3,tz);
 905
 906             /**************************
 907              * CALCULATE INTERACTIONS *
 908              **************************/
 909
 910             r21              = _mm256_mul_pd(rsq21,rinv21);
 911             r21              = _mm256_andnot_pd(dummy_mask,r21);
 912
 913             /* Calculate table index by multiplying r with table scale and truncate to integer */
 914             rt               = _mm256_mul_pd(r21,vftabscale);
 915             vfitab           = _mm256_cvttpd_epi32(rt);
 916             vfeps            = _mm256_sub_pd(rt,_mm256_round_pd(rt, _MM_FROUND_FLOOR));
 917             vfitab           = _mm_slli_epi32(vfitab,2);
 918
 919             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 920             Y                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
 921             F                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
 922             G                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,2) );
 923             H                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,3) );
 924             GMX_MM256_FULLTRANSPOSE4_PD(Y,F,G,H);
 925             Heps             = _mm256_mul_pd(vfeps,H);
 926             Fp               = _mm256_add_pd(F,_mm256_mul_pd(vfeps,_mm256_add_pd(G,Heps)));
 927             VV               = _mm256_add_pd(Y,_mm256_mul_pd(vfeps,Fp));
 928             velec            = _mm256_mul_pd(qq21,VV);
 929             FF               = _mm256_add_pd(Fp,_mm256_mul_pd(vfeps,_mm256_add_pd(G,_mm256_add_pd(Heps,Heps))));
 930             felec            = _mm256_xor_pd(signbit,_mm256_mul_pd(_mm256_mul_pd(qq21,FF),_mm256_mul_pd(vftabscale,rinv21)));
 931
 932             /* Update potential sum for this i atom from the interaction with this j atom. */
 933             velec            = _mm256_andnot_pd(dummy_mask,velec);
 934             velecsum         = _mm256_add_pd(velecsum,velec);
 935
 936             fscal            = felec;
 937
 938             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
 939
 940             /* Calculate temporary vectorial force */
 941             tx               = _mm256_mul_pd(fscal,dx21);
 942             ty               = _mm256_mul_pd(fscal,dy21);
 943             tz               = _mm256_mul_pd(fscal,dz21);
 944
 945             /* Update vectorial force */
 946             fix2             = _mm256_add_pd(fix2,tx);
 947             fiy2             = _mm256_add_pd(fiy2,ty);
 948             fiz2             = _mm256_add_pd(fiz2,tz);
 949
 950             fjx1             = _mm256_add_pd(fjx1,tx);
 951             fjy1             = _mm256_add_pd(fjy1,ty);
 952             fjz1             = _mm256_add_pd(fjz1,tz);
 953
 954             /**************************
 955              * CALCULATE INTERACTIONS *
 956              **************************/
 957
 958             r22              = _mm256_mul_pd(rsq22,rinv22);
 959             r22              = _mm256_andnot_pd(dummy_mask,r22);
 960
 961             /* Calculate table index by multiplying r with table scale and truncate to integer */
 962             rt               = _mm256_mul_pd(r22,vftabscale);
 963             vfitab           = _mm256_cvttpd_epi32(rt);
 964             vfeps            = _mm256_sub_pd(rt,_mm256_round_pd(rt, _MM_FROUND_FLOOR));
 965             vfitab           = _mm_slli_epi32(vfitab,2);
 966
 967             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 968             Y                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
 969             F                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
 970             G                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,2) );
 971             H                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,3) );
 972             GMX_MM256_FULLTRANSPOSE4_PD(Y,F,G,H);
 973             Heps             = _mm256_mul_pd(vfeps,H);
 974             Fp               = _mm256_add_pd(F,_mm256_mul_pd(vfeps,_mm256_add_pd(G,Heps)));
 975             VV               = _mm256_add_pd(Y,_mm256_mul_pd(vfeps,Fp));
 976             velec            = _mm256_mul_pd(qq22,VV);
 977             FF               = _mm256_add_pd(Fp,_mm256_mul_pd(vfeps,_mm256_add_pd(G,_mm256_add_pd(Heps,Heps))));
 978             felec            = _mm256_xor_pd(signbit,_mm256_mul_pd(_mm256_mul_pd(qq22,FF),_mm256_mul_pd(vftabscale,rinv22)));
 979
 980             /* Update potential sum for this i atom from the interaction with this j atom. */
 981             velec            = _mm256_andnot_pd(dummy_mask,velec);
 982             velecsum         = _mm256_add_pd(velecsum,velec);
 983
 984             fscal            = felec;
 985
 986             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
 987
 988             /* Calculate temporary vectorial force */
 989             tx               = _mm256_mul_pd(fscal,dx22);
 990             ty               = _mm256_mul_pd(fscal,dy22);
 991             tz               = _mm256_mul_pd(fscal,dz22);
 992
 993             /* Update vectorial force */
 994             fix2             = _mm256_add_pd(fix2,tx);
 995             fiy2             = _mm256_add_pd(fiy2,ty);
 996             fiz2             = _mm256_add_pd(fiz2,tz);
 997
 998             fjx2             = _mm256_add_pd(fjx2,tx);
 999             fjy2             = _mm256_add_pd(fjy2,ty);
1000             fjz2             = _mm256_add_pd(fjz2,tz);
1001
1002             /**************************
1003              * CALCULATE INTERACTIONS *
1004              **************************/
1005
1006             r23              = _mm256_mul_pd(rsq23,rinv23);
1007             r23              = _mm256_andnot_pd(dummy_mask,r23);
1008
1009             /* Calculate table index by multiplying r with table scale and truncate to integer */
1010             rt               = _mm256_mul_pd(r23,vftabscale);
1011             vfitab           = _mm256_cvttpd_epi32(rt);
1012             vfeps            = _mm256_sub_pd(rt,_mm256_round_pd(rt, _MM_FROUND_FLOOR));
1013             vfitab           = _mm_slli_epi32(vfitab,2);
1014
1015             /* CUBIC SPLINE TABLE ELECTROSTATICS */
1016             Y                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
1017             F                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
1018             G                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,2) );
1019             H                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,3) );
1020             GMX_MM256_FULLTRANSPOSE4_PD(Y,F,G,H);
1021             Heps             = _mm256_mul_pd(vfeps,H);
1022             Fp               = _mm256_add_pd(F,_mm256_mul_pd(vfeps,_mm256_add_pd(G,Heps)));
1023             VV               = _mm256_add_pd(Y,_mm256_mul_pd(vfeps,Fp));
1024             velec            = _mm256_mul_pd(qq23,VV);
1025             FF               = _mm256_add_pd(Fp,_mm256_mul_pd(vfeps,_mm256_add_pd(G,_mm256_add_pd(Heps,Heps))));
1026             felec            = _mm256_xor_pd(signbit,_mm256_mul_pd(_mm256_mul_pd(qq23,FF),_mm256_mul_pd(vftabscale,rinv23)));
1027
1028             /* Update potential sum for this i atom from the interaction with this j atom. */
1029             velec            = _mm256_andnot_pd(dummy_mask,velec);
1030             velecsum         = _mm256_add_pd(velecsum,velec);
1031
1032             fscal            = felec;
1033
1034             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
1035
1036             /* Calculate temporary vectorial force */
1037             tx               = _mm256_mul_pd(fscal,dx23);
1038             ty               = _mm256_mul_pd(fscal,dy23);
1039             tz               = _mm256_mul_pd(fscal,dz23);
1040
1041             /* Update vectorial force */
1042             fix2             = _mm256_add_pd(fix2,tx);
1043             fiy2             = _mm256_add_pd(fiy2,ty);
1044             fiz2             = _mm256_add_pd(fiz2,tz);
1045
1046             fjx3             = _mm256_add_pd(fjx3,tx);
1047             fjy3             = _mm256_add_pd(fjy3,ty);
1048             fjz3             = _mm256_add_pd(fjz3,tz);
1049
1050             /**************************
1051              * CALCULATE INTERACTIONS *
1052              **************************/
1053
1054             r31              = _mm256_mul_pd(rsq31,rinv31);
1055             r31              = _mm256_andnot_pd(dummy_mask,r31);
1056
1057             /* Calculate table index by multiplying r with table scale and truncate to integer */
1058             rt               = _mm256_mul_pd(r31,vftabscale);
1059             vfitab           = _mm256_cvttpd_epi32(rt);
1060             vfeps            = _mm256_sub_pd(rt,_mm256_round_pd(rt, _MM_FROUND_FLOOR));
1061             vfitab           = _mm_slli_epi32(vfitab,2);
1062
1063             /* CUBIC SPLINE TABLE ELECTROSTATICS */
1064             Y                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
1065             F                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
1066             G                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,2) );
1067             H                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,3) );
1068             GMX_MM256_FULLTRANSPOSE4_PD(Y,F,G,H);
1069             Heps             = _mm256_mul_pd(vfeps,H);
1070             Fp               = _mm256_add_pd(F,_mm256_mul_pd(vfeps,_mm256_add_pd(G,Heps)));
1071             VV               = _mm256_add_pd(Y,_mm256_mul_pd(vfeps,Fp));
1072             velec            = _mm256_mul_pd(qq31,VV);
1073             FF               = _mm256_add_pd(Fp,_mm256_mul_pd(vfeps,_mm256_add_pd(G,_mm256_add_pd(Heps,Heps))));
1074             felec            = _mm256_xor_pd(signbit,_mm256_mul_pd(_mm256_mul_pd(qq31,FF),_mm256_mul_pd(vftabscale,rinv31)));
1075
1076             /* Update potential sum for this i atom from the interaction with this j atom. */
1077             velec            = _mm256_andnot_pd(dummy_mask,velec);
1078             velecsum         = _mm256_add_pd(velecsum,velec);
1079
1080             fscal            = felec;
1081
1082             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
1083
1084             /* Calculate temporary vectorial force */
1085             tx               = _mm256_mul_pd(fscal,dx31);
1086             ty               = _mm256_mul_pd(fscal,dy31);
1087             tz               = _mm256_mul_pd(fscal,dz31);
1088
1089             /* Update vectorial force */
1090             fix3             = _mm256_add_pd(fix3,tx);
1091             fiy3             = _mm256_add_pd(fiy3,ty);
1092             fiz3             = _mm256_add_pd(fiz3,tz);
1093
1094             fjx1             = _mm256_add_pd(fjx1,tx);
1095             fjy1             = _mm256_add_pd(fjy1,ty);
1096             fjz1             = _mm256_add_pd(fjz1,tz);
1097
1098             /**************************
1099              * CALCULATE INTERACTIONS *
1100              **************************/
1101
1102             r32              = _mm256_mul_pd(rsq32,rinv32);
1103             r32              = _mm256_andnot_pd(dummy_mask,r32);
1104
1105             /* Calculate table index by multiplying r with table scale and truncate to integer */
1106             rt               = _mm256_mul_pd(r32,vftabscale);
1107             vfitab           = _mm256_cvttpd_epi32(rt);
1108             vfeps            = _mm256_sub_pd(rt,_mm256_round_pd(rt, _MM_FROUND_FLOOR));
1109             vfitab           = _mm_slli_epi32(vfitab,2);
1110
1111             /* CUBIC SPLINE TABLE ELECTROSTATICS */
1112             Y                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
1113             F                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
1114             G                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,2) );
1115             H                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,3) );
1116             GMX_MM256_FULLTRANSPOSE4_PD(Y,F,G,H);
1117             Heps             = _mm256_mul_pd(vfeps,H);
1118             Fp               = _mm256_add_pd(F,_mm256_mul_pd(vfeps,_mm256_add_pd(G,Heps)));
1119             VV               = _mm256_add_pd(Y,_mm256_mul_pd(vfeps,Fp));
1120             velec            = _mm256_mul_pd(qq32,VV);
1121             FF               = _mm256_add_pd(Fp,_mm256_mul_pd(vfeps,_mm256_add_pd(G,_mm256_add_pd(Heps,Heps))));
1122             felec            = _mm256_xor_pd(signbit,_mm256_mul_pd(_mm256_mul_pd(qq32,FF),_mm256_mul_pd(vftabscale,rinv32)));
1123
1124             /* Update potential sum for this i atom from the interaction with this j atom. */
1125             velec            = _mm256_andnot_pd(dummy_mask,velec);
1126             velecsum         = _mm256_add_pd(velecsum,velec);
1127
1128             fscal            = felec;
1129
1130             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
1131
1132             /* Calculate temporary vectorial force */
1133             tx               = _mm256_mul_pd(fscal,dx32);
1134             ty               = _mm256_mul_pd(fscal,dy32);
1135             tz               = _mm256_mul_pd(fscal,dz32);
1136
1137             /* Update vectorial force */
1138             fix3             = _mm256_add_pd(fix3,tx);
1139             fiy3             = _mm256_add_pd(fiy3,ty);
1140             fiz3             = _mm256_add_pd(fiz3,tz);
1141
1142             fjx2             = _mm256_add_pd(fjx2,tx);
1143             fjy2             = _mm256_add_pd(fjy2,ty);
1144             fjz2             = _mm256_add_pd(fjz2,tz);
1145
1146             /**************************
1147              * CALCULATE INTERACTIONS *
1148              **************************/
1149
1150             r33              = _mm256_mul_pd(rsq33,rinv33);
1151             r33              = _mm256_andnot_pd(dummy_mask,r33);
1152
1153             /* Calculate table index by multiplying r with table scale and truncate to integer */
1154             rt               = _mm256_mul_pd(r33,vftabscale);
1155             vfitab           = _mm256_cvttpd_epi32(rt);
1156             vfeps            = _mm256_sub_pd(rt,_mm256_round_pd(rt, _MM_FROUND_FLOOR));
1157             vfitab           = _mm_slli_epi32(vfitab,2);
1158
1159             /* CUBIC SPLINE TABLE ELECTROSTATICS */
1160             Y                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
1161             F                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
1162             G                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,2) );
1163             H                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,3) );
1164             GMX_MM256_FULLTRANSPOSE4_PD(Y,F,G,H);
1165             Heps             = _mm256_mul_pd(vfeps,H);
1166             Fp               = _mm256_add_pd(F,_mm256_mul_pd(vfeps,_mm256_add_pd(G,Heps)));
1167             VV               = _mm256_add_pd(Y,_mm256_mul_pd(vfeps,Fp));
1168             velec            = _mm256_mul_pd(qq33,VV);
1169             FF               = _mm256_add_pd(Fp,_mm256_mul_pd(vfeps,_mm256_add_pd(G,_mm256_add_pd(Heps,Heps))));
1170             felec            = _mm256_xor_pd(signbit,_mm256_mul_pd(_mm256_mul_pd(qq33,FF),_mm256_mul_pd(vftabscale,rinv33)));
1171
1172             /* Update potential sum for this i atom from the interaction with this j atom. */
1173             velec            = _mm256_andnot_pd(dummy_mask,velec);
1174             velecsum         = _mm256_add_pd(velecsum,velec);
1175
1176             fscal            = felec;
1177
1178             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
1179
1180             /* Calculate temporary vectorial force */
1181             tx               = _mm256_mul_pd(fscal,dx33);
1182             ty               = _mm256_mul_pd(fscal,dy33);
1183             tz               = _mm256_mul_pd(fscal,dz33);
1184
1185             /* Update vectorial force */
1186             fix3             = _mm256_add_pd(fix3,tx);
1187             fiy3             = _mm256_add_pd(fiy3,ty);
1188             fiz3             = _mm256_add_pd(fiz3,tz);
1189
1190             fjx3             = _mm256_add_pd(fjx3,tx);
1191             fjy3             = _mm256_add_pd(fjy3,ty);
1192             fjz3             = _mm256_add_pd(fjz3,tz);
1193
1194             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
1195             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
1196             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
1197             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
1198
1199             gmx_mm256_decrement_3rvec_4ptr_swizzle_pd(fjptrA+DIM,fjptrB+DIM,fjptrC+DIM,fjptrD+DIM,
1200                                                       fjx1,fjy1,fjz1,fjx2,fjy2,fjz2,fjx3,fjy3,fjz3);
1201
1202             /* Inner loop uses 396 flops */
1203         }
1204
1205         /* End of innermost loop */
1206
1207         gmx_mm256_update_iforce_3atom_swizzle_pd(fix1,fiy1,fiz1,fix2,fiy2,fiz2,fix3,fiy3,fiz3,
1208                                                  f+i_coord_offset+DIM,fshift+i_shift_offset);
1209
1210         ggid                        = gid[iidx];
1211         /* Update potential energies */
1212         gmx_mm256_update_1pot_pd(velecsum,kernel_data->energygrp_elec+ggid);
1213
1214         /* Increment number of inner iterations */
1215         inneriter                  += j_index_end - j_index_start;
1216
1217         /* Outer loop uses 19 flops */
1218     }
1219
1220     /* Increment number of outer iterations */
1221     outeriter        += nri;
1222
1223     /* Update outer/inner flops */
1224
1225     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_W4W4_VF,outeriter*19 + inneriter*396);
1226 }
1227 /*
1228  * Gromacs nonbonded kernel:   nb_kernel_ElecCSTab_VdwNone_GeomW4W4_F_avx_256_double
1229  * Electrostatics interaction: CubicSplineTable
1230  * VdW interaction:            None
1231  * Geometry:                   Water4-Water4
1232  * Calculate force/pot:        Force
1233  */
1234 void
1235 nb_kernel_ElecCSTab_VdwNone_GeomW4W4_F_avx_256_double
1236                     (t_nblist * gmx_restrict                nlist,
1237                      rvec * gmx_restrict                    xx,
1238                      rvec * gmx_restrict                    ff,
1239                      t_forcerec * gmx_restrict              fr,
1240                      t_mdatoms * gmx_restrict               mdatoms,
1241                      nb_kernel_data_t * gmx_restrict        kernel_data,
1242                      t_nrnb * gmx_restrict                  nrnb)
1243 {
1244     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
1245      * just 0 for non-waters.
1246      * Suffixes A,B,C,D refer to j loop unrolling done with AVX, e.g. for the four different
1247      * jnr indices corresponding to data put in the four positions in the SIMD register.
1248      */
1249     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
1250     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
1251     int              jnrA,jnrB,jnrC,jnrD;
1252     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
1253     int              jnrlistE,jnrlistF,jnrlistG,jnrlistH;
1254     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
1255     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
1256     real             rcutoff_scalar;
1257     real             *shiftvec,*fshift,*x,*f;
1258     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD;
1259     real             scratch[4*DIM];
1260     __m256d          tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
1261     real *           vdwioffsetptr1;
1262     __m256d          ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
1263     real *           vdwioffsetptr2;
1264     __m256d          ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
1265     real *           vdwioffsetptr3;
1266     __m256d          ix3,iy3,iz3,fix3,fiy3,fiz3,iq3,isai3;
1267     int              vdwjidx1A,vdwjidx1B,vdwjidx1C,vdwjidx1D;
1268     __m256d          jx1,jy1,jz1,fjx1,fjy1,fjz1,jq1,isaj1;
1269     int              vdwjidx2A,vdwjidx2B,vdwjidx2C,vdwjidx2D;
1270     __m256d          jx2,jy2,jz2,fjx2,fjy2,fjz2,jq2,isaj2;
1271     int              vdwjidx3A,vdwjidx3B,vdwjidx3C,vdwjidx3D;
1272     __m256d          jx3,jy3,jz3,fjx3,fjy3,fjz3,jq3,isaj3;
1273     __m256d          dx11,dy11,dz11,rsq11,rinv11,rinvsq11,r11,qq11,c6_11,c12_11;
1274     __m256d          dx12,dy12,dz12,rsq12,rinv12,rinvsq12,r12,qq12,c6_12,c12_12;
1275     __m256d          dx13,dy13,dz13,rsq13,rinv13,rinvsq13,r13,qq13,c6_13,c12_13;
1276     __m256d          dx21,dy21,dz21,rsq21,rinv21,rinvsq21,r21,qq21,c6_21,c12_21;
1277     __m256d          dx22,dy22,dz22,rsq22,rinv22,rinvsq22,r22,qq22,c6_22,c12_22;
1278     __m256d          dx23,dy23,dz23,rsq23,rinv23,rinvsq23,r23,qq23,c6_23,c12_23;
1279     __m256d          dx31,dy31,dz31,rsq31,rinv31,rinvsq31,r31,qq31,c6_31,c12_31;
1280     __m256d          dx32,dy32,dz32,rsq32,rinv32,rinvsq32,r32,qq32,c6_32,c12_32;
1281     __m256d          dx33,dy33,dz33,rsq33,rinv33,rinvsq33,r33,qq33,c6_33,c12_33;
1282     __m256d          velec,felec,velecsum,facel,crf,krf,krf2;
1283     real             *charge;
1284     __m128i          vfitab;
1285     __m128i          ifour       = _mm_set1_epi32(4);
1286     __m256d          rt,vfeps,vftabscale,Y,F,G,H,Heps,Fp,VV,FF;
1287     real             *vftab;
1288     __m256d          dummy_mask,cutoff_mask;
1289     __m128           tmpmask0,tmpmask1;
1290     __m256d          signbit = _mm256_castsi256_pd( _mm256_set1_epi32(0x80000000) );
1291     __m256d          one     = _mm256_set1_pd(1.0);
1292     __m256d          two     = _mm256_set1_pd(2.0);
1293     x                = xx[0];
1294     f                = ff[0];
1295
1296     nri              = nlist->nri;
1297     iinr             = nlist->iinr;
1298     jindex           = nlist->jindex;
1299     jjnr             = nlist->jjnr;
1300     shiftidx         = nlist->shift;
1301     gid              = nlist->gid;
1302     shiftvec         = fr->shift_vec[0];
1303     fshift           = fr->fshift[0];
1304     facel            = _mm256_set1_pd(fr->epsfac);
1305     charge           = mdatoms->chargeA;
1306
1307     vftab            = kernel_data->table_elec->data;
1308     vftabscale       = _mm256_set1_pd(kernel_data->table_elec->scale);
1309
1310     /* Setup water-specific parameters */
1311     inr              = nlist->iinr[0];
1312     iq1              = _mm256_mul_pd(facel,_mm256_set1_pd(charge[inr+1]));
1313     iq2              = _mm256_mul_pd(facel,_mm256_set1_pd(charge[inr+2]));
1314     iq3              = _mm256_mul_pd(facel,_mm256_set1_pd(charge[inr+3]));
1315
1316     jq1              = _mm256_set1_pd(charge[inr+1]);
1317     jq2              = _mm256_set1_pd(charge[inr+2]);
1318     jq3              = _mm256_set1_pd(charge[inr+3]);
1319     qq11             = _mm256_mul_pd(iq1,jq1);
1320     qq12             = _mm256_mul_pd(iq1,jq2);
1321     qq13             = _mm256_mul_pd(iq1,jq3);
1322     qq21             = _mm256_mul_pd(iq2,jq1);
1323     qq22             = _mm256_mul_pd(iq2,jq2);
1324     qq23             = _mm256_mul_pd(iq2,jq3);
1325     qq31             = _mm256_mul_pd(iq3,jq1);
1326     qq32             = _mm256_mul_pd(iq3,jq2);
1327     qq33             = _mm256_mul_pd(iq3,jq3);
1328
1329     /* Avoid stupid compiler warnings */
1330     jnrA = jnrB = jnrC = jnrD = 0;
1331     j_coord_offsetA = 0;
1332     j_coord_offsetB = 0;
1333     j_coord_offsetC = 0;
1334     j_coord_offsetD = 0;
1335
1336     outeriter        = 0;
1337     inneriter        = 0;
1338
1339     for(iidx=0;iidx<4*DIM;iidx++)
1340     {
1341         scratch[iidx] = 0.0;
1342     }
1343
1344     /* Start outer loop over neighborlists */
1345     for(iidx=0; iidx<nri; iidx++)
1346     {
1347         /* Load shift vector for this list */
1348         i_shift_offset   = DIM*shiftidx[iidx];
1349
1350         /* Load limits for loop over neighbors */
1351         j_index_start    = jindex[iidx];
1352         j_index_end      = jindex[iidx+1];
1353
1354         /* Get outer coordinate index */
1355         inr              = iinr[iidx];
1356         i_coord_offset   = DIM*inr;
1357
1358         /* Load i particle coords and add shift vector */
1359         gmx_mm256_load_shift_and_3rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset+DIM,
1360                                                     &ix1,&iy1,&iz1,&ix2,&iy2,&iz2,&ix3,&iy3,&iz3);
1361
1362         fix1             = _mm256_setzero_pd();
1363         fiy1             = _mm256_setzero_pd();
1364         fiz1             = _mm256_setzero_pd();
1365         fix2             = _mm256_setzero_pd();
1366         fiy2             = _mm256_setzero_pd();
1367         fiz2             = _mm256_setzero_pd();
1368         fix3             = _mm256_setzero_pd();
1369         fiy3             = _mm256_setzero_pd();
1370         fiz3             = _mm256_setzero_pd();
1371
1372         /* Start inner kernel loop */
1373         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
1374         {
1375
1376             /* Get j neighbor index, and coordinate index */
1377             jnrA             = jjnr[jidx];
1378             jnrB             = jjnr[jidx+1];
1379             jnrC             = jjnr[jidx+2];
1380             jnrD             = jjnr[jidx+3];
1381             j_coord_offsetA  = DIM*jnrA;
1382             j_coord_offsetB  = DIM*jnrB;
1383             j_coord_offsetC  = DIM*jnrC;
1384             j_coord_offsetD  = DIM*jnrD;
1385
1386             /* load j atom coordinates */
1387             gmx_mm256_load_3rvec_4ptr_swizzle_pd(x+j_coord_offsetA+DIM,x+j_coord_offsetB+DIM,
1388                                                  x+j_coord_offsetC+DIM,x+j_coord_offsetD+DIM,
1389                                                  &jx1,&jy1,&jz1,&jx2,&jy2,&jz2,&jx3,&jy3,&jz3);
1390
1391             /* Calculate displacement vector */
1392             dx11             = _mm256_sub_pd(ix1,jx1);
1393             dy11             = _mm256_sub_pd(iy1,jy1);
1394             dz11             = _mm256_sub_pd(iz1,jz1);
1395             dx12             = _mm256_sub_pd(ix1,jx2);
1396             dy12             = _mm256_sub_pd(iy1,jy2);
1397             dz12             = _mm256_sub_pd(iz1,jz2);
1398             dx13             = _mm256_sub_pd(ix1,jx3);
1399             dy13             = _mm256_sub_pd(iy1,jy3);
1400             dz13             = _mm256_sub_pd(iz1,jz3);
1401             dx21             = _mm256_sub_pd(ix2,jx1);
1402             dy21             = _mm256_sub_pd(iy2,jy1);
1403             dz21             = _mm256_sub_pd(iz2,jz1);
1404             dx22             = _mm256_sub_pd(ix2,jx2);
1405             dy22             = _mm256_sub_pd(iy2,jy2);
1406             dz22             = _mm256_sub_pd(iz2,jz2);
1407             dx23             = _mm256_sub_pd(ix2,jx3);
1408             dy23             = _mm256_sub_pd(iy2,jy3);
1409             dz23             = _mm256_sub_pd(iz2,jz3);
1410             dx31             = _mm256_sub_pd(ix3,jx1);
1411             dy31             = _mm256_sub_pd(iy3,jy1);
1412             dz31             = _mm256_sub_pd(iz3,jz1);
1413             dx32             = _mm256_sub_pd(ix3,jx2);
1414             dy32             = _mm256_sub_pd(iy3,jy2);
1415             dz32             = _mm256_sub_pd(iz3,jz2);
1416             dx33             = _mm256_sub_pd(ix3,jx3);
1417             dy33             = _mm256_sub_pd(iy3,jy3);
1418             dz33             = _mm256_sub_pd(iz3,jz3);
1419
1420             /* Calculate squared distance and things based on it */
1421             rsq11            = gmx_mm256_calc_rsq_pd(dx11,dy11,dz11);
1422             rsq12            = gmx_mm256_calc_rsq_pd(dx12,dy12,dz12);
1423             rsq13            = gmx_mm256_calc_rsq_pd(dx13,dy13,dz13);
1424             rsq21            = gmx_mm256_calc_rsq_pd(dx21,dy21,dz21);
1425             rsq22            = gmx_mm256_calc_rsq_pd(dx22,dy22,dz22);
1426             rsq23            = gmx_mm256_calc_rsq_pd(dx23,dy23,dz23);
1427             rsq31            = gmx_mm256_calc_rsq_pd(dx31,dy31,dz31);
1428             rsq32            = gmx_mm256_calc_rsq_pd(dx32,dy32,dz32);
1429             rsq33            = gmx_mm256_calc_rsq_pd(dx33,dy33,dz33);
1430
1431             rinv11           = gmx_mm256_invsqrt_pd(rsq11);
1432             rinv12           = gmx_mm256_invsqrt_pd(rsq12);
1433             rinv13           = gmx_mm256_invsqrt_pd(rsq13);
1434             rinv21           = gmx_mm256_invsqrt_pd(rsq21);
1435             rinv22           = gmx_mm256_invsqrt_pd(rsq22);
1436             rinv23           = gmx_mm256_invsqrt_pd(rsq23);
1437             rinv31           = gmx_mm256_invsqrt_pd(rsq31);
1438             rinv32           = gmx_mm256_invsqrt_pd(rsq32);
1439             rinv33           = gmx_mm256_invsqrt_pd(rsq33);
1440
1441             fjx1             = _mm256_setzero_pd();
1442             fjy1             = _mm256_setzero_pd();
1443             fjz1             = _mm256_setzero_pd();
1444             fjx2             = _mm256_setzero_pd();
1445             fjy2             = _mm256_setzero_pd();
1446             fjz2             = _mm256_setzero_pd();
1447             fjx3             = _mm256_setzero_pd();
1448             fjy3             = _mm256_setzero_pd();
1449             fjz3             = _mm256_setzero_pd();
1450
1451             /**************************
1452              * CALCULATE INTERACTIONS *
1453              **************************/
1454
1455             r11              = _mm256_mul_pd(rsq11,rinv11);
1456
1457             /* Calculate table index by multiplying r with table scale and truncate to integer */
1458             rt               = _mm256_mul_pd(r11,vftabscale);
1459             vfitab           = _mm256_cvttpd_epi32(rt);
1460             vfeps            = _mm256_sub_pd(rt,_mm256_round_pd(rt, _MM_FROUND_FLOOR));
1461             vfitab           = _mm_slli_epi32(vfitab,2);
1462
1463             /* CUBIC SPLINE TABLE ELECTROSTATICS */
1464             Y                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
1465             F                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
1466             G                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,2) );
1467             H                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,3) );
1468             GMX_MM256_FULLTRANSPOSE4_PD(Y,F,G,H);
1469             Heps             = _mm256_mul_pd(vfeps,H);
1470             Fp               = _mm256_add_pd(F,_mm256_mul_pd(vfeps,_mm256_add_pd(G,Heps)));
1471             FF               = _mm256_add_pd(Fp,_mm256_mul_pd(vfeps,_mm256_add_pd(G,_mm256_add_pd(Heps,Heps))));
1472             felec            = _mm256_xor_pd(signbit,_mm256_mul_pd(_mm256_mul_pd(qq11,FF),_mm256_mul_pd(vftabscale,rinv11)));
1473
1474             fscal            = felec;
1475
1476             /* Calculate temporary vectorial force */
1477             tx               = _mm256_mul_pd(fscal,dx11);
1478             ty               = _mm256_mul_pd(fscal,dy11);
1479             tz               = _mm256_mul_pd(fscal,dz11);
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             fjx1             = _mm256_add_pd(fjx1,tx);
1487             fjy1             = _mm256_add_pd(fjy1,ty);
1488             fjz1             = _mm256_add_pd(fjz1,tz);
1489
1490             /**************************
1491              * CALCULATE INTERACTIONS *
1492              **************************/
1493
1494             r12              = _mm256_mul_pd(rsq12,rinv12);
1495
1496             /* Calculate table index by multiplying r with table scale and truncate to integer */
1497             rt               = _mm256_mul_pd(r12,vftabscale);
1498             vfitab           = _mm256_cvttpd_epi32(rt);
1499             vfeps            = _mm256_sub_pd(rt,_mm256_round_pd(rt, _MM_FROUND_FLOOR));
1500             vfitab           = _mm_slli_epi32(vfitab,2);
1501
1502             /* CUBIC SPLINE TABLE ELECTROSTATICS */
1503             Y                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
1504             F                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
1505             G                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,2) );
1506             H                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,3) );
1507             GMX_MM256_FULLTRANSPOSE4_PD(Y,F,G,H);
1508             Heps             = _mm256_mul_pd(vfeps,H);
1509             Fp               = _mm256_add_pd(F,_mm256_mul_pd(vfeps,_mm256_add_pd(G,Heps)));
1510             FF               = _mm256_add_pd(Fp,_mm256_mul_pd(vfeps,_mm256_add_pd(G,_mm256_add_pd(Heps,Heps))));
1511             felec            = _mm256_xor_pd(signbit,_mm256_mul_pd(_mm256_mul_pd(qq12,FF),_mm256_mul_pd(vftabscale,rinv12)));
1512
1513             fscal            = felec;
1514
1515             /* Calculate temporary vectorial force */
1516             tx               = _mm256_mul_pd(fscal,dx12);
1517             ty               = _mm256_mul_pd(fscal,dy12);
1518             tz               = _mm256_mul_pd(fscal,dz12);
1519
1520             /* Update vectorial force */
1521             fix1             = _mm256_add_pd(fix1,tx);
1522             fiy1             = _mm256_add_pd(fiy1,ty);
1523             fiz1             = _mm256_add_pd(fiz1,tz);
1524
1525             fjx2             = _mm256_add_pd(fjx2,tx);
1526             fjy2             = _mm256_add_pd(fjy2,ty);
1527             fjz2             = _mm256_add_pd(fjz2,tz);
1528
1529             /**************************
1530              * CALCULATE INTERACTIONS *
1531              **************************/
1532
1533             r13              = _mm256_mul_pd(rsq13,rinv13);
1534
1535             /* Calculate table index by multiplying r with table scale and truncate to integer */
1536             rt               = _mm256_mul_pd(r13,vftabscale);
1537             vfitab           = _mm256_cvttpd_epi32(rt);
1538             vfeps            = _mm256_sub_pd(rt,_mm256_round_pd(rt, _MM_FROUND_FLOOR));
1539             vfitab           = _mm_slli_epi32(vfitab,2);
1540
1541             /* CUBIC SPLINE TABLE ELECTROSTATICS */
1542             Y                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
1543             F                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
1544             G                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,2) );
1545             H                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,3) );
1546             GMX_MM256_FULLTRANSPOSE4_PD(Y,F,G,H);
1547             Heps             = _mm256_mul_pd(vfeps,H);
1548             Fp               = _mm256_add_pd(F,_mm256_mul_pd(vfeps,_mm256_add_pd(G,Heps)));
1549             FF               = _mm256_add_pd(Fp,_mm256_mul_pd(vfeps,_mm256_add_pd(G,_mm256_add_pd(Heps,Heps))));
1550             felec            = _mm256_xor_pd(signbit,_mm256_mul_pd(_mm256_mul_pd(qq13,FF),_mm256_mul_pd(vftabscale,rinv13)));
1551
1552             fscal            = felec;
1553
1554             /* Calculate temporary vectorial force */
1555             tx               = _mm256_mul_pd(fscal,dx13);
1556             ty               = _mm256_mul_pd(fscal,dy13);
1557             tz               = _mm256_mul_pd(fscal,dz13);
1558
1559             /* Update vectorial force */
1560             fix1             = _mm256_add_pd(fix1,tx);
1561             fiy1             = _mm256_add_pd(fiy1,ty);
1562             fiz1             = _mm256_add_pd(fiz1,tz);
1563
1564             fjx3             = _mm256_add_pd(fjx3,tx);
1565             fjy3             = _mm256_add_pd(fjy3,ty);
1566             fjz3             = _mm256_add_pd(fjz3,tz);
1567
1568             /**************************
1569              * CALCULATE INTERACTIONS *
1570              **************************/
1571
1572             r21              = _mm256_mul_pd(rsq21,rinv21);
1573
1574             /* Calculate table index by multiplying r with table scale and truncate to integer */
1575             rt               = _mm256_mul_pd(r21,vftabscale);
1576             vfitab           = _mm256_cvttpd_epi32(rt);
1577             vfeps            = _mm256_sub_pd(rt,_mm256_round_pd(rt, _MM_FROUND_FLOOR));
1578             vfitab           = _mm_slli_epi32(vfitab,2);
1579
1580             /* CUBIC SPLINE TABLE ELECTROSTATICS */
1581             Y                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
1582             F                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
1583             G                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,2) );
1584             H                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,3) );
1585             GMX_MM256_FULLTRANSPOSE4_PD(Y,F,G,H);
1586             Heps             = _mm256_mul_pd(vfeps,H);
1587             Fp               = _mm256_add_pd(F,_mm256_mul_pd(vfeps,_mm256_add_pd(G,Heps)));
1588             FF               = _mm256_add_pd(Fp,_mm256_mul_pd(vfeps,_mm256_add_pd(G,_mm256_add_pd(Heps,Heps))));
1589             felec            = _mm256_xor_pd(signbit,_mm256_mul_pd(_mm256_mul_pd(qq21,FF),_mm256_mul_pd(vftabscale,rinv21)));
1590
1591             fscal            = felec;
1592
1593             /* Calculate temporary vectorial force */
1594             tx               = _mm256_mul_pd(fscal,dx21);
1595             ty               = _mm256_mul_pd(fscal,dy21);
1596             tz               = _mm256_mul_pd(fscal,dz21);
1597
1598             /* Update vectorial force */
1599             fix2             = _mm256_add_pd(fix2,tx);
1600             fiy2             = _mm256_add_pd(fiy2,ty);
1601             fiz2             = _mm256_add_pd(fiz2,tz);
1602
1603             fjx1             = _mm256_add_pd(fjx1,tx);
1604             fjy1             = _mm256_add_pd(fjy1,ty);
1605             fjz1             = _mm256_add_pd(fjz1,tz);
1606
1607             /**************************
1608              * CALCULATE INTERACTIONS *
1609              **************************/
1610
1611             r22              = _mm256_mul_pd(rsq22,rinv22);
1612
1613             /* Calculate table index by multiplying r with table scale and truncate to integer */
1614             rt               = _mm256_mul_pd(r22,vftabscale);
1615             vfitab           = _mm256_cvttpd_epi32(rt);
1616             vfeps            = _mm256_sub_pd(rt,_mm256_round_pd(rt, _MM_FROUND_FLOOR));
1617             vfitab           = _mm_slli_epi32(vfitab,2);
1618
1619             /* CUBIC SPLINE TABLE ELECTROSTATICS */
1620             Y                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
1621             F                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
1622             G                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,2) );
1623             H                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,3) );
1624             GMX_MM256_FULLTRANSPOSE4_PD(Y,F,G,H);
1625             Heps             = _mm256_mul_pd(vfeps,H);
1626             Fp               = _mm256_add_pd(F,_mm256_mul_pd(vfeps,_mm256_add_pd(G,Heps)));
1627             FF               = _mm256_add_pd(Fp,_mm256_mul_pd(vfeps,_mm256_add_pd(G,_mm256_add_pd(Heps,Heps))));
1628             felec            = _mm256_xor_pd(signbit,_mm256_mul_pd(_mm256_mul_pd(qq22,FF),_mm256_mul_pd(vftabscale,rinv22)));
1629
1630             fscal            = felec;
1631
1632             /* Calculate temporary vectorial force */
1633             tx               = _mm256_mul_pd(fscal,dx22);
1634             ty               = _mm256_mul_pd(fscal,dy22);
1635             tz               = _mm256_mul_pd(fscal,dz22);
1636
1637             /* Update vectorial force */
1638             fix2             = _mm256_add_pd(fix2,tx);
1639             fiy2             = _mm256_add_pd(fiy2,ty);
1640             fiz2             = _mm256_add_pd(fiz2,tz);
1641
1642             fjx2             = _mm256_add_pd(fjx2,tx);
1643             fjy2             = _mm256_add_pd(fjy2,ty);
1644             fjz2             = _mm256_add_pd(fjz2,tz);
1645
1646             /**************************
1647              * CALCULATE INTERACTIONS *
1648              **************************/
1649
1650             r23              = _mm256_mul_pd(rsq23,rinv23);
1651
1652             /* Calculate table index by multiplying r with table scale and truncate to integer */
1653             rt               = _mm256_mul_pd(r23,vftabscale);
1654             vfitab           = _mm256_cvttpd_epi32(rt);
1655             vfeps            = _mm256_sub_pd(rt,_mm256_round_pd(rt, _MM_FROUND_FLOOR));
1656             vfitab           = _mm_slli_epi32(vfitab,2);
1657
1658             /* CUBIC SPLINE TABLE ELECTROSTATICS */
1659             Y                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
1660             F                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
1661             G                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,2) );
1662             H                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,3) );
1663             GMX_MM256_FULLTRANSPOSE4_PD(Y,F,G,H);
1664             Heps             = _mm256_mul_pd(vfeps,H);
1665             Fp               = _mm256_add_pd(F,_mm256_mul_pd(vfeps,_mm256_add_pd(G,Heps)));
1666             FF               = _mm256_add_pd(Fp,_mm256_mul_pd(vfeps,_mm256_add_pd(G,_mm256_add_pd(Heps,Heps))));
1667             felec            = _mm256_xor_pd(signbit,_mm256_mul_pd(_mm256_mul_pd(qq23,FF),_mm256_mul_pd(vftabscale,rinv23)));
1668
1669             fscal            = felec;
1670
1671             /* Calculate temporary vectorial force */
1672             tx               = _mm256_mul_pd(fscal,dx23);
1673             ty               = _mm256_mul_pd(fscal,dy23);
1674             tz               = _mm256_mul_pd(fscal,dz23);
1675
1676             /* Update vectorial force */
1677             fix2             = _mm256_add_pd(fix2,tx);
1678             fiy2             = _mm256_add_pd(fiy2,ty);
1679             fiz2             = _mm256_add_pd(fiz2,tz);
1680
1681             fjx3             = _mm256_add_pd(fjx3,tx);
1682             fjy3             = _mm256_add_pd(fjy3,ty);
1683             fjz3             = _mm256_add_pd(fjz3,tz);
1684
1685             /**************************
1686              * CALCULATE INTERACTIONS *
1687              **************************/
1688
1689             r31              = _mm256_mul_pd(rsq31,rinv31);
1690
1691             /* Calculate table index by multiplying r with table scale and truncate to integer */
1692             rt               = _mm256_mul_pd(r31,vftabscale);
1693             vfitab           = _mm256_cvttpd_epi32(rt);
1694             vfeps            = _mm256_sub_pd(rt,_mm256_round_pd(rt, _MM_FROUND_FLOOR));
1695             vfitab           = _mm_slli_epi32(vfitab,2);
1696
1697             /* CUBIC SPLINE TABLE ELECTROSTATICS */
1698             Y                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
1699             F                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
1700             G                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,2) );
1701             H                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,3) );
1702             GMX_MM256_FULLTRANSPOSE4_PD(Y,F,G,H);
1703             Heps             = _mm256_mul_pd(vfeps,H);
1704             Fp               = _mm256_add_pd(F,_mm256_mul_pd(vfeps,_mm256_add_pd(G,Heps)));
1705             FF               = _mm256_add_pd(Fp,_mm256_mul_pd(vfeps,_mm256_add_pd(G,_mm256_add_pd(Heps,Heps))));
1706             felec            = _mm256_xor_pd(signbit,_mm256_mul_pd(_mm256_mul_pd(qq31,FF),_mm256_mul_pd(vftabscale,rinv31)));
1707
1708             fscal            = felec;
1709
1710             /* Calculate temporary vectorial force */
1711             tx               = _mm256_mul_pd(fscal,dx31);
1712             ty               = _mm256_mul_pd(fscal,dy31);
1713             tz               = _mm256_mul_pd(fscal,dz31);
1714
1715             /* Update vectorial force */
1716             fix3             = _mm256_add_pd(fix3,tx);
1717             fiy3             = _mm256_add_pd(fiy3,ty);
1718             fiz3             = _mm256_add_pd(fiz3,tz);
1719
1720             fjx1             = _mm256_add_pd(fjx1,tx);
1721             fjy1             = _mm256_add_pd(fjy1,ty);
1722             fjz1             = _mm256_add_pd(fjz1,tz);
1723
1724             /**************************
1725              * CALCULATE INTERACTIONS *
1726              **************************/
1727
1728             r32              = _mm256_mul_pd(rsq32,rinv32);
1729
1730             /* Calculate table index by multiplying r with table scale and truncate to integer */
1731             rt               = _mm256_mul_pd(r32,vftabscale);
1732             vfitab           = _mm256_cvttpd_epi32(rt);
1733             vfeps            = _mm256_sub_pd(rt,_mm256_round_pd(rt, _MM_FROUND_FLOOR));
1734             vfitab           = _mm_slli_epi32(vfitab,2);
1735
1736             /* CUBIC SPLINE TABLE ELECTROSTATICS */
1737             Y                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
1738             F                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
1739             G                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,2) );
1740             H                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,3) );
1741             GMX_MM256_FULLTRANSPOSE4_PD(Y,F,G,H);
1742             Heps             = _mm256_mul_pd(vfeps,H);
1743             Fp               = _mm256_add_pd(F,_mm256_mul_pd(vfeps,_mm256_add_pd(G,Heps)));
1744             FF               = _mm256_add_pd(Fp,_mm256_mul_pd(vfeps,_mm256_add_pd(G,_mm256_add_pd(Heps,Heps))));
1745             felec            = _mm256_xor_pd(signbit,_mm256_mul_pd(_mm256_mul_pd(qq32,FF),_mm256_mul_pd(vftabscale,rinv32)));
1746
1747             fscal            = felec;
1748
1749             /* Calculate temporary vectorial force */
1750             tx               = _mm256_mul_pd(fscal,dx32);
1751             ty               = _mm256_mul_pd(fscal,dy32);
1752             tz               = _mm256_mul_pd(fscal,dz32);
1753
1754             /* Update vectorial force */
1755             fix3             = _mm256_add_pd(fix3,tx);
1756             fiy3             = _mm256_add_pd(fiy3,ty);
1757             fiz3             = _mm256_add_pd(fiz3,tz);
1758
1759             fjx2             = _mm256_add_pd(fjx2,tx);
1760             fjy2             = _mm256_add_pd(fjy2,ty);
1761             fjz2             = _mm256_add_pd(fjz2,tz);
1762
1763             /**************************
1764              * CALCULATE INTERACTIONS *
1765              **************************/
1766
1767             r33              = _mm256_mul_pd(rsq33,rinv33);
1768
1769             /* Calculate table index by multiplying r with table scale and truncate to integer */
1770             rt               = _mm256_mul_pd(r33,vftabscale);
1771             vfitab           = _mm256_cvttpd_epi32(rt);
1772             vfeps            = _mm256_sub_pd(rt,_mm256_round_pd(rt, _MM_FROUND_FLOOR));
1773             vfitab           = _mm_slli_epi32(vfitab,2);
1774
1775             /* CUBIC SPLINE TABLE ELECTROSTATICS */
1776             Y                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
1777             F                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
1778             G                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,2) );
1779             H                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,3) );
1780             GMX_MM256_FULLTRANSPOSE4_PD(Y,F,G,H);
1781             Heps             = _mm256_mul_pd(vfeps,H);
1782             Fp               = _mm256_add_pd(F,_mm256_mul_pd(vfeps,_mm256_add_pd(G,Heps)));
1783             FF               = _mm256_add_pd(Fp,_mm256_mul_pd(vfeps,_mm256_add_pd(G,_mm256_add_pd(Heps,Heps))));
1784             felec            = _mm256_xor_pd(signbit,_mm256_mul_pd(_mm256_mul_pd(qq33,FF),_mm256_mul_pd(vftabscale,rinv33)));
1785
1786             fscal            = felec;
1787
1788             /* Calculate temporary vectorial force */
1789             tx               = _mm256_mul_pd(fscal,dx33);
1790             ty               = _mm256_mul_pd(fscal,dy33);
1791             tz               = _mm256_mul_pd(fscal,dz33);
1792
1793             /* Update vectorial force */
1794             fix3             = _mm256_add_pd(fix3,tx);
1795             fiy3             = _mm256_add_pd(fiy3,ty);
1796             fiz3             = _mm256_add_pd(fiz3,tz);
1797
1798             fjx3             = _mm256_add_pd(fjx3,tx);
1799             fjy3             = _mm256_add_pd(fjy3,ty);
1800             fjz3             = _mm256_add_pd(fjz3,tz);
1801
1802             fjptrA             = f+j_coord_offsetA;
1803             fjptrB             = f+j_coord_offsetB;
1804             fjptrC             = f+j_coord_offsetC;
1805             fjptrD             = f+j_coord_offsetD;
1806
1807             gmx_mm256_decrement_3rvec_4ptr_swizzle_pd(fjptrA+DIM,fjptrB+DIM,fjptrC+DIM,fjptrD+DIM,
1808                                                       fjx1,fjy1,fjz1,fjx2,fjy2,fjz2,fjx3,fjy3,fjz3);
1809
1810             /* Inner loop uses 351 flops */
1811         }
1812
1813         if(jidx<j_index_end)
1814         {
1815
1816             /* Get j neighbor index, and coordinate index */
1817             jnrlistA         = jjnr[jidx];
1818             jnrlistB         = jjnr[jidx+1];
1819             jnrlistC         = jjnr[jidx+2];
1820             jnrlistD         = jjnr[jidx+3];
1821             /* Sign of each element will be negative for non-real atoms.
1822              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
1823              * so use it as val = _mm_andnot_pd(mask,val) to clear dummy entries.
1824              */
1825             tmpmask0 = gmx_mm_castsi128_pd(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
1826
1827             tmpmask1 = _mm_permute_ps(tmpmask0,_GMX_MM_PERMUTE(3,3,2,2));
1828             tmpmask0 = _mm_permute_ps(tmpmask0,_GMX_MM_PERMUTE(1,1,0,0));
1829             dummy_mask = _mm256_castps_pd(gmx_mm256_set_m128(tmpmask1,tmpmask0));
1830
1831             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
1832             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
1833             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
1834             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
1835             j_coord_offsetA  = DIM*jnrA;
1836             j_coord_offsetB  = DIM*jnrB;
1837             j_coord_offsetC  = DIM*jnrC;
1838             j_coord_offsetD  = DIM*jnrD;
1839
1840             /* load j atom coordinates */
1841             gmx_mm256_load_3rvec_4ptr_swizzle_pd(x+j_coord_offsetA+DIM,x+j_coord_offsetB+DIM,
1842                                                  x+j_coord_offsetC+DIM,x+j_coord_offsetD+DIM,
1843                                                  &jx1,&jy1,&jz1,&jx2,&jy2,&jz2,&jx3,&jy3,&jz3);
1844
1845             /* Calculate displacement vector */
1846             dx11             = _mm256_sub_pd(ix1,jx1);
1847             dy11             = _mm256_sub_pd(iy1,jy1);
1848             dz11             = _mm256_sub_pd(iz1,jz1);
1849             dx12             = _mm256_sub_pd(ix1,jx2);
1850             dy12             = _mm256_sub_pd(iy1,jy2);
1851             dz12             = _mm256_sub_pd(iz1,jz2);
1852             dx13             = _mm256_sub_pd(ix1,jx3);
1853             dy13             = _mm256_sub_pd(iy1,jy3);
1854             dz13             = _mm256_sub_pd(iz1,jz3);
1855             dx21             = _mm256_sub_pd(ix2,jx1);
1856             dy21             = _mm256_sub_pd(iy2,jy1);
1857             dz21             = _mm256_sub_pd(iz2,jz1);
1858             dx22             = _mm256_sub_pd(ix2,jx2);
1859             dy22             = _mm256_sub_pd(iy2,jy2);
1860             dz22             = _mm256_sub_pd(iz2,jz2);
1861             dx23             = _mm256_sub_pd(ix2,jx3);
1862             dy23             = _mm256_sub_pd(iy2,jy3);
1863             dz23             = _mm256_sub_pd(iz2,jz3);
1864             dx31             = _mm256_sub_pd(ix3,jx1);
1865             dy31             = _mm256_sub_pd(iy3,jy1);
1866             dz31             = _mm256_sub_pd(iz3,jz1);
1867             dx32             = _mm256_sub_pd(ix3,jx2);
1868             dy32             = _mm256_sub_pd(iy3,jy2);
1869             dz32             = _mm256_sub_pd(iz3,jz2);
1870             dx33             = _mm256_sub_pd(ix3,jx3);
1871             dy33             = _mm256_sub_pd(iy3,jy3);
1872             dz33             = _mm256_sub_pd(iz3,jz3);
1873
1874             /* Calculate squared distance and things based on it */
1875             rsq11            = gmx_mm256_calc_rsq_pd(dx11,dy11,dz11);
1876             rsq12            = gmx_mm256_calc_rsq_pd(dx12,dy12,dz12);
1877             rsq13            = gmx_mm256_calc_rsq_pd(dx13,dy13,dz13);
1878             rsq21            = gmx_mm256_calc_rsq_pd(dx21,dy21,dz21);
1879             rsq22            = gmx_mm256_calc_rsq_pd(dx22,dy22,dz22);
1880             rsq23            = gmx_mm256_calc_rsq_pd(dx23,dy23,dz23);
1881             rsq31            = gmx_mm256_calc_rsq_pd(dx31,dy31,dz31);
1882             rsq32            = gmx_mm256_calc_rsq_pd(dx32,dy32,dz32);
1883             rsq33            = gmx_mm256_calc_rsq_pd(dx33,dy33,dz33);
1884
1885             rinv11           = gmx_mm256_invsqrt_pd(rsq11);
1886             rinv12           = gmx_mm256_invsqrt_pd(rsq12);
1887             rinv13           = gmx_mm256_invsqrt_pd(rsq13);
1888             rinv21           = gmx_mm256_invsqrt_pd(rsq21);
1889             rinv22           = gmx_mm256_invsqrt_pd(rsq22);
1890             rinv23           = gmx_mm256_invsqrt_pd(rsq23);
1891             rinv31           = gmx_mm256_invsqrt_pd(rsq31);
1892             rinv32           = gmx_mm256_invsqrt_pd(rsq32);
1893             rinv33           = gmx_mm256_invsqrt_pd(rsq33);
1894
1895             fjx1             = _mm256_setzero_pd();
1896             fjy1             = _mm256_setzero_pd();
1897             fjz1             = _mm256_setzero_pd();
1898             fjx2             = _mm256_setzero_pd();
1899             fjy2             = _mm256_setzero_pd();
1900             fjz2             = _mm256_setzero_pd();
1901             fjx3             = _mm256_setzero_pd();
1902             fjy3             = _mm256_setzero_pd();
1903             fjz3             = _mm256_setzero_pd();
1904
1905             /**************************
1906              * CALCULATE INTERACTIONS *
1907              **************************/
1908
1909             r11              = _mm256_mul_pd(rsq11,rinv11);
1910             r11              = _mm256_andnot_pd(dummy_mask,r11);
1911
1912             /* Calculate table index by multiplying r with table scale and truncate to integer */
1913             rt               = _mm256_mul_pd(r11,vftabscale);
1914             vfitab           = _mm256_cvttpd_epi32(rt);
1915             vfeps            = _mm256_sub_pd(rt,_mm256_round_pd(rt, _MM_FROUND_FLOOR));
1916             vfitab           = _mm_slli_epi32(vfitab,2);
1917
1918             /* CUBIC SPLINE TABLE ELECTROSTATICS */
1919             Y                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
1920             F                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
1921             G                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,2) );
1922             H                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,3) );
1923             GMX_MM256_FULLTRANSPOSE4_PD(Y,F,G,H);
1924             Heps             = _mm256_mul_pd(vfeps,H);
1925             Fp               = _mm256_add_pd(F,_mm256_mul_pd(vfeps,_mm256_add_pd(G,Heps)));
1926             FF               = _mm256_add_pd(Fp,_mm256_mul_pd(vfeps,_mm256_add_pd(G,_mm256_add_pd(Heps,Heps))));
1927             felec            = _mm256_xor_pd(signbit,_mm256_mul_pd(_mm256_mul_pd(qq11,FF),_mm256_mul_pd(vftabscale,rinv11)));
1928
1929             fscal            = felec;
1930
1931             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
1932
1933             /* Calculate temporary vectorial force */
1934             tx               = _mm256_mul_pd(fscal,dx11);
1935             ty               = _mm256_mul_pd(fscal,dy11);
1936             tz               = _mm256_mul_pd(fscal,dz11);
1937
1938             /* Update vectorial force */
1939             fix1             = _mm256_add_pd(fix1,tx);
1940             fiy1             = _mm256_add_pd(fiy1,ty);
1941             fiz1             = _mm256_add_pd(fiz1,tz);
1942
1943             fjx1             = _mm256_add_pd(fjx1,tx);
1944             fjy1             = _mm256_add_pd(fjy1,ty);
1945             fjz1             = _mm256_add_pd(fjz1,tz);
1946
1947             /**************************
1948              * CALCULATE INTERACTIONS *
1949              **************************/
1950
1951             r12              = _mm256_mul_pd(rsq12,rinv12);
1952             r12              = _mm256_andnot_pd(dummy_mask,r12);
1953
1954             /* Calculate table index by multiplying r with table scale and truncate to integer */
1955             rt               = _mm256_mul_pd(r12,vftabscale);
1956             vfitab           = _mm256_cvttpd_epi32(rt);
1957             vfeps            = _mm256_sub_pd(rt,_mm256_round_pd(rt, _MM_FROUND_FLOOR));
1958             vfitab           = _mm_slli_epi32(vfitab,2);
1959
1960             /* CUBIC SPLINE TABLE ELECTROSTATICS */
1961             Y                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
1962             F                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
1963             G                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,2) );
1964             H                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,3) );
1965             GMX_MM256_FULLTRANSPOSE4_PD(Y,F,G,H);
1966             Heps             = _mm256_mul_pd(vfeps,H);
1967             Fp               = _mm256_add_pd(F,_mm256_mul_pd(vfeps,_mm256_add_pd(G,Heps)));
1968             FF               = _mm256_add_pd(Fp,_mm256_mul_pd(vfeps,_mm256_add_pd(G,_mm256_add_pd(Heps,Heps))));
1969             felec            = _mm256_xor_pd(signbit,_mm256_mul_pd(_mm256_mul_pd(qq12,FF),_mm256_mul_pd(vftabscale,rinv12)));
1970
1971             fscal            = felec;
1972
1973             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
1974
1975             /* Calculate temporary vectorial force */
1976             tx               = _mm256_mul_pd(fscal,dx12);
1977             ty               = _mm256_mul_pd(fscal,dy12);
1978             tz               = _mm256_mul_pd(fscal,dz12);
1979
1980             /* Update vectorial force */
1981             fix1             = _mm256_add_pd(fix1,tx);
1982             fiy1             = _mm256_add_pd(fiy1,ty);
1983             fiz1             = _mm256_add_pd(fiz1,tz);
1984
1985             fjx2             = _mm256_add_pd(fjx2,tx);
1986             fjy2             = _mm256_add_pd(fjy2,ty);
1987             fjz2             = _mm256_add_pd(fjz2,tz);
1988
1989             /**************************
1990              * CALCULATE INTERACTIONS *
1991              **************************/
1992
1993             r13              = _mm256_mul_pd(rsq13,rinv13);
1994             r13              = _mm256_andnot_pd(dummy_mask,r13);
1995
1996             /* Calculate table index by multiplying r with table scale and truncate to integer */
1997             rt               = _mm256_mul_pd(r13,vftabscale);
1998             vfitab           = _mm256_cvttpd_epi32(rt);
1999             vfeps            = _mm256_sub_pd(rt,_mm256_round_pd(rt, _MM_FROUND_FLOOR));
2000             vfitab           = _mm_slli_epi32(vfitab,2);
2001
2002             /* CUBIC SPLINE TABLE ELECTROSTATICS */
2003             Y                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
2004             F                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
2005             G                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,2) );
2006             H                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,3) );
2007             GMX_MM256_FULLTRANSPOSE4_PD(Y,F,G,H);
2008             Heps             = _mm256_mul_pd(vfeps,H);
2009             Fp               = _mm256_add_pd(F,_mm256_mul_pd(vfeps,_mm256_add_pd(G,Heps)));
2010             FF               = _mm256_add_pd(Fp,_mm256_mul_pd(vfeps,_mm256_add_pd(G,_mm256_add_pd(Heps,Heps))));
2011             felec            = _mm256_xor_pd(signbit,_mm256_mul_pd(_mm256_mul_pd(qq13,FF),_mm256_mul_pd(vftabscale,rinv13)));
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,dx13);
2019             ty               = _mm256_mul_pd(fscal,dy13);
2020             tz               = _mm256_mul_pd(fscal,dz13);
2021
2022             /* Update vectorial force */
2023             fix1             = _mm256_add_pd(fix1,tx);
2024             fiy1             = _mm256_add_pd(fiy1,ty);
2025             fiz1             = _mm256_add_pd(fiz1,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             /**************************
2032              * CALCULATE INTERACTIONS *
2033              **************************/
2034
2035             r21              = _mm256_mul_pd(rsq21,rinv21);
2036             r21              = _mm256_andnot_pd(dummy_mask,r21);
2037
2038             /* Calculate table index by multiplying r with table scale and truncate to integer */
2039             rt               = _mm256_mul_pd(r21,vftabscale);
2040             vfitab           = _mm256_cvttpd_epi32(rt);
2041             vfeps            = _mm256_sub_pd(rt,_mm256_round_pd(rt, _MM_FROUND_FLOOR));
2042             vfitab           = _mm_slli_epi32(vfitab,2);
2043
2044             /* CUBIC SPLINE TABLE ELECTROSTATICS */
2045             Y                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
2046             F                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
2047             G                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,2) );
2048             H                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,3) );
2049             GMX_MM256_FULLTRANSPOSE4_PD(Y,F,G,H);
2050             Heps             = _mm256_mul_pd(vfeps,H);
2051             Fp               = _mm256_add_pd(F,_mm256_mul_pd(vfeps,_mm256_add_pd(G,Heps)));
2052             FF               = _mm256_add_pd(Fp,_mm256_mul_pd(vfeps,_mm256_add_pd(G,_mm256_add_pd(Heps,Heps))));
2053             felec            = _mm256_xor_pd(signbit,_mm256_mul_pd(_mm256_mul_pd(qq21,FF),_mm256_mul_pd(vftabscale,rinv21)));
2054
2055             fscal            = felec;
2056
2057             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
2058
2059             /* Calculate temporary vectorial force */
2060             tx               = _mm256_mul_pd(fscal,dx21);
2061             ty               = _mm256_mul_pd(fscal,dy21);
2062             tz               = _mm256_mul_pd(fscal,dz21);
2063
2064             /* Update vectorial force */
2065             fix2             = _mm256_add_pd(fix2,tx);
2066             fiy2             = _mm256_add_pd(fiy2,ty);
2067             fiz2             = _mm256_add_pd(fiz2,tz);
2068
2069             fjx1             = _mm256_add_pd(fjx1,tx);
2070             fjy1             = _mm256_add_pd(fjy1,ty);
2071             fjz1             = _mm256_add_pd(fjz1,tz);
2072
2073             /**************************
2074              * CALCULATE INTERACTIONS *
2075              **************************/
2076
2077             r22              = _mm256_mul_pd(rsq22,rinv22);
2078             r22              = _mm256_andnot_pd(dummy_mask,r22);
2079
2080             /* Calculate table index by multiplying r with table scale and truncate to integer */
2081             rt               = _mm256_mul_pd(r22,vftabscale);
2082             vfitab           = _mm256_cvttpd_epi32(rt);
2083             vfeps            = _mm256_sub_pd(rt,_mm256_round_pd(rt, _MM_FROUND_FLOOR));
2084             vfitab           = _mm_slli_epi32(vfitab,2);
2085
2086             /* CUBIC SPLINE TABLE ELECTROSTATICS */
2087             Y                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
2088             F                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
2089             G                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,2) );
2090             H                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,3) );
2091             GMX_MM256_FULLTRANSPOSE4_PD(Y,F,G,H);
2092             Heps             = _mm256_mul_pd(vfeps,H);
2093             Fp               = _mm256_add_pd(F,_mm256_mul_pd(vfeps,_mm256_add_pd(G,Heps)));
2094             FF               = _mm256_add_pd(Fp,_mm256_mul_pd(vfeps,_mm256_add_pd(G,_mm256_add_pd(Heps,Heps))));
2095             felec            = _mm256_xor_pd(signbit,_mm256_mul_pd(_mm256_mul_pd(qq22,FF),_mm256_mul_pd(vftabscale,rinv22)));
2096
2097             fscal            = felec;
2098
2099             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
2100
2101             /* Calculate temporary vectorial force */
2102             tx               = _mm256_mul_pd(fscal,dx22);
2103             ty               = _mm256_mul_pd(fscal,dy22);
2104             tz               = _mm256_mul_pd(fscal,dz22);
2105
2106             /* Update vectorial force */
2107             fix2             = _mm256_add_pd(fix2,tx);
2108             fiy2             = _mm256_add_pd(fiy2,ty);
2109             fiz2             = _mm256_add_pd(fiz2,tz);
2110
2111             fjx2             = _mm256_add_pd(fjx2,tx);
2112             fjy2             = _mm256_add_pd(fjy2,ty);
2113             fjz2             = _mm256_add_pd(fjz2,tz);
2114
2115             /**************************
2116              * CALCULATE INTERACTIONS *
2117              **************************/
2118
2119             r23              = _mm256_mul_pd(rsq23,rinv23);
2120             r23              = _mm256_andnot_pd(dummy_mask,r23);
2121
2122             /* Calculate table index by multiplying r with table scale and truncate to integer */
2123             rt               = _mm256_mul_pd(r23,vftabscale);
2124             vfitab           = _mm256_cvttpd_epi32(rt);
2125             vfeps            = _mm256_sub_pd(rt,_mm256_round_pd(rt, _MM_FROUND_FLOOR));
2126             vfitab           = _mm_slli_epi32(vfitab,2);
2127
2128             /* CUBIC SPLINE TABLE ELECTROSTATICS */
2129             Y                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
2130             F                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
2131             G                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,2) );
2132             H                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,3) );
2133             GMX_MM256_FULLTRANSPOSE4_PD(Y,F,G,H);
2134             Heps             = _mm256_mul_pd(vfeps,H);
2135             Fp               = _mm256_add_pd(F,_mm256_mul_pd(vfeps,_mm256_add_pd(G,Heps)));
2136             FF               = _mm256_add_pd(Fp,_mm256_mul_pd(vfeps,_mm256_add_pd(G,_mm256_add_pd(Heps,Heps))));
2137             felec            = _mm256_xor_pd(signbit,_mm256_mul_pd(_mm256_mul_pd(qq23,FF),_mm256_mul_pd(vftabscale,rinv23)));
2138
2139             fscal            = felec;
2140
2141             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
2142
2143             /* Calculate temporary vectorial force */
2144             tx               = _mm256_mul_pd(fscal,dx23);
2145             ty               = _mm256_mul_pd(fscal,dy23);
2146             tz               = _mm256_mul_pd(fscal,dz23);
2147
2148             /* Update vectorial force */
2149             fix2             = _mm256_add_pd(fix2,tx);
2150             fiy2             = _mm256_add_pd(fiy2,ty);
2151             fiz2             = _mm256_add_pd(fiz2,tz);
2152
2153             fjx3             = _mm256_add_pd(fjx3,tx);
2154             fjy3             = _mm256_add_pd(fjy3,ty);
2155             fjz3             = _mm256_add_pd(fjz3,tz);
2156
2157             /**************************
2158              * CALCULATE INTERACTIONS *
2159              **************************/
2160
2161             r31              = _mm256_mul_pd(rsq31,rinv31);
2162             r31              = _mm256_andnot_pd(dummy_mask,r31);
2163
2164             /* Calculate table index by multiplying r with table scale and truncate to integer */
2165             rt               = _mm256_mul_pd(r31,vftabscale);
2166             vfitab           = _mm256_cvttpd_epi32(rt);
2167             vfeps            = _mm256_sub_pd(rt,_mm256_round_pd(rt, _MM_FROUND_FLOOR));
2168             vfitab           = _mm_slli_epi32(vfitab,2);
2169
2170             /* CUBIC SPLINE TABLE ELECTROSTATICS */
2171             Y                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
2172             F                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
2173             G                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,2) );
2174             H                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,3) );
2175             GMX_MM256_FULLTRANSPOSE4_PD(Y,F,G,H);
2176             Heps             = _mm256_mul_pd(vfeps,H);
2177             Fp               = _mm256_add_pd(F,_mm256_mul_pd(vfeps,_mm256_add_pd(G,Heps)));
2178             FF               = _mm256_add_pd(Fp,_mm256_mul_pd(vfeps,_mm256_add_pd(G,_mm256_add_pd(Heps,Heps))));
2179             felec            = _mm256_xor_pd(signbit,_mm256_mul_pd(_mm256_mul_pd(qq31,FF),_mm256_mul_pd(vftabscale,rinv31)));
2180
2181             fscal            = felec;
2182
2183             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
2184
2185             /* Calculate temporary vectorial force */
2186             tx               = _mm256_mul_pd(fscal,dx31);
2187             ty               = _mm256_mul_pd(fscal,dy31);
2188             tz               = _mm256_mul_pd(fscal,dz31);
2189
2190             /* Update vectorial force */
2191             fix3             = _mm256_add_pd(fix3,tx);
2192             fiy3             = _mm256_add_pd(fiy3,ty);
2193             fiz3             = _mm256_add_pd(fiz3,tz);
2194
2195             fjx1             = _mm256_add_pd(fjx1,tx);
2196             fjy1             = _mm256_add_pd(fjy1,ty);
2197             fjz1             = _mm256_add_pd(fjz1,tz);
2198
2199             /**************************
2200              * CALCULATE INTERACTIONS *
2201              **************************/
2202
2203             r32              = _mm256_mul_pd(rsq32,rinv32);
2204             r32              = _mm256_andnot_pd(dummy_mask,r32);
2205
2206             /* Calculate table index by multiplying r with table scale and truncate to integer */
2207             rt               = _mm256_mul_pd(r32,vftabscale);
2208             vfitab           = _mm256_cvttpd_epi32(rt);
2209             vfeps            = _mm256_sub_pd(rt,_mm256_round_pd(rt, _MM_FROUND_FLOOR));
2210             vfitab           = _mm_slli_epi32(vfitab,2);
2211
2212             /* CUBIC SPLINE TABLE ELECTROSTATICS */
2213             Y                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
2214             F                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
2215             G                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,2) );
2216             H                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,3) );
2217             GMX_MM256_FULLTRANSPOSE4_PD(Y,F,G,H);
2218             Heps             = _mm256_mul_pd(vfeps,H);
2219             Fp               = _mm256_add_pd(F,_mm256_mul_pd(vfeps,_mm256_add_pd(G,Heps)));
2220             FF               = _mm256_add_pd(Fp,_mm256_mul_pd(vfeps,_mm256_add_pd(G,_mm256_add_pd(Heps,Heps))));
2221             felec            = _mm256_xor_pd(signbit,_mm256_mul_pd(_mm256_mul_pd(qq32,FF),_mm256_mul_pd(vftabscale,rinv32)));
2222
2223             fscal            = felec;
2224
2225             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
2226
2227             /* Calculate temporary vectorial force */
2228             tx               = _mm256_mul_pd(fscal,dx32);
2229             ty               = _mm256_mul_pd(fscal,dy32);
2230             tz               = _mm256_mul_pd(fscal,dz32);
2231
2232             /* Update vectorial force */
2233             fix3             = _mm256_add_pd(fix3,tx);
2234             fiy3             = _mm256_add_pd(fiy3,ty);
2235             fiz3             = _mm256_add_pd(fiz3,tz);
2236
2237             fjx2             = _mm256_add_pd(fjx2,tx);
2238             fjy2             = _mm256_add_pd(fjy2,ty);
2239             fjz2             = _mm256_add_pd(fjz2,tz);
2240
2241             /**************************
2242              * CALCULATE INTERACTIONS *
2243              **************************/
2244
2245             r33              = _mm256_mul_pd(rsq33,rinv33);
2246             r33              = _mm256_andnot_pd(dummy_mask,r33);
2247
2248             /* Calculate table index by multiplying r with table scale and truncate to integer */
2249             rt               = _mm256_mul_pd(r33,vftabscale);
2250             vfitab           = _mm256_cvttpd_epi32(rt);
2251             vfeps            = _mm256_sub_pd(rt,_mm256_round_pd(rt, _MM_FROUND_FLOOR));
2252             vfitab           = _mm_slli_epi32(vfitab,2);
2253
2254             /* CUBIC SPLINE TABLE ELECTROSTATICS */
2255             Y                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
2256             F                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
2257             G                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,2) );
2258             H                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,3) );
2259             GMX_MM256_FULLTRANSPOSE4_PD(Y,F,G,H);
2260             Heps             = _mm256_mul_pd(vfeps,H);
2261             Fp               = _mm256_add_pd(F,_mm256_mul_pd(vfeps,_mm256_add_pd(G,Heps)));
2262             FF               = _mm256_add_pd(Fp,_mm256_mul_pd(vfeps,_mm256_add_pd(G,_mm256_add_pd(Heps,Heps))));
2263             felec            = _mm256_xor_pd(signbit,_mm256_mul_pd(_mm256_mul_pd(qq33,FF),_mm256_mul_pd(vftabscale,rinv33)));
2264
2265             fscal            = felec;
2266
2267             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
2268
2269             /* Calculate temporary vectorial force */
2270             tx               = _mm256_mul_pd(fscal,dx33);
2271             ty               = _mm256_mul_pd(fscal,dy33);
2272             tz               = _mm256_mul_pd(fscal,dz33);
2273
2274             /* Update vectorial force */
2275             fix3             = _mm256_add_pd(fix3,tx);
2276             fiy3             = _mm256_add_pd(fiy3,ty);
2277             fiz3             = _mm256_add_pd(fiz3,tz);
2278
2279             fjx3             = _mm256_add_pd(fjx3,tx);
2280             fjy3             = _mm256_add_pd(fjy3,ty);
2281             fjz3             = _mm256_add_pd(fjz3,tz);
2282
2283             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
2284             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
2285             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
2286             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
2287
2288             gmx_mm256_decrement_3rvec_4ptr_swizzle_pd(fjptrA+DIM,fjptrB+DIM,fjptrC+DIM,fjptrD+DIM,
2289                                                       fjx1,fjy1,fjz1,fjx2,fjy2,fjz2,fjx3,fjy3,fjz3);
2290
2291             /* Inner loop uses 360 flops */
2292         }
2293
2294         /* End of innermost loop */
2295
2296         gmx_mm256_update_iforce_3atom_swizzle_pd(fix1,fiy1,fiz1,fix2,fiy2,fiz2,fix3,fiy3,fiz3,
2297                                                  f+i_coord_offset+DIM,fshift+i_shift_offset);
2298
2299         /* Increment number of inner iterations */
2300         inneriter                  += j_index_end - j_index_start;
2301
2302         /* Outer loop uses 18 flops */
2303     }
2304
2305     /* Increment number of outer iterations */
2306     outeriter        += nri;
2307
2308     /* Update outer/inner flops */
2309
2310     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_W4W4_F,outeriter*18 + inneriter*360);
2311 }