src/gmxlib/nonbonded/nb_kernel_avx_256_double/nb_kernel_ElecEw_VdwCSTab_GeomW4P1_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_ElecEw_VdwCSTab_GeomW4P1_VF_avx_256_double
  38  * Electrostatics interaction: Ewald
  39  * VdW interaction:            CubicSplineTable
  40  * Geometry:                   Water4-Particle
  41  * Calculate force/pot:        PotentialAndForce
  42  */
  43 void
  44 nb_kernel_ElecEw_VdwCSTab_GeomW4P1_VF_avx_256_double
  45                     (t_nblist * gmx_restrict                nlist,
  46                      rvec * gmx_restrict                    xx,
  47                      rvec * gmx_restrict                    ff,
  48                      t_forcerec * gmx_restrict              fr,
  49                      t_mdatoms * gmx_restrict               mdatoms,
  50                      nb_kernel_data_t * gmx_restrict        kernel_data,
  51                      t_nrnb * gmx_restrict                  nrnb)
  52 {
  53     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
  54      * just 0 for non-waters.
  55      * Suffixes A,B,C,D refer to j loop unrolling done with AVX, e.g. for the four different
  56      * jnr indices corresponding to data put in the four positions in the SIMD register.
  57      */
  58     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
  59     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
  60     int              jnrA,jnrB,jnrC,jnrD;
  61     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
  62     int              jnrlistE,jnrlistF,jnrlistG,jnrlistH;
  63     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
  64     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
  65     real             rcutoff_scalar;
  66     real             *shiftvec,*fshift,*x,*f;
  67     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD;
  68     real             scratch[4*DIM];
  69     __m256d          tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
  70     real *           vdwioffsetptr0;
  71     __m256d          ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
  72     real *           vdwioffsetptr1;
  73     __m256d          ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
  74     real *           vdwioffsetptr2;
  75     __m256d          ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
  76     real *           vdwioffsetptr3;
  77     __m256d          ix3,iy3,iz3,fix3,fiy3,fiz3,iq3,isai3;
  78     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
  79     __m256d          jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
  80     __m256d          dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
  81     __m256d          dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
  82     __m256d          dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
  83     __m256d          dx30,dy30,dz30,rsq30,rinv30,rinvsq30,r30,qq30,c6_30,c12_30;
  84     __m256d          velec,felec,velecsum,facel,crf,krf,krf2;
  85     real             *charge;
  86     int              nvdwtype;
  87     __m256d          rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
  88     int              *vdwtype;
  89     real             *vdwparam;
  90     __m256d          one_sixth   = _mm256_set1_pd(1.0/6.0);
  91     __m256d          one_twelfth = _mm256_set1_pd(1.0/12.0);
  92     __m128i          vfitab;
  93     __m128i          ifour       = _mm_set1_epi32(4);
  94     __m256d          rt,vfeps,vftabscale,Y,F,G,H,Heps,Fp,VV,FF;
  95     real             *vftab;
  96     __m128i          ewitab;
  97     __m256d          ewtabscale,eweps,sh_ewald,ewrt,ewtabhalfspace,ewtabF,ewtabFn,ewtabD,ewtabV;
  98     __m256d          beta,beta2,beta3,zeta2,pmecorrF,pmecorrV,rinv3;
  99     real             *ewtab;
 100     __m256d          dummy_mask,cutoff_mask;
 101     __m128           tmpmask0,tmpmask1;
 102     __m256d          signbit = _mm256_castsi256_pd( _mm256_set1_epi32(0x80000000) );
 103     __m256d          one     = _mm256_set1_pd(1.0);
 104     __m256d          two     = _mm256_set1_pd(2.0);
 105     x                = xx[0];
 106     f                = ff[0];
 107
 108     nri              = nlist->nri;
 109     iinr             = nlist->iinr;
 110     jindex           = nlist->jindex;
 111     jjnr             = nlist->jjnr;
 112     shiftidx         = nlist->shift;
 113     gid              = nlist->gid;
 114     shiftvec         = fr->shift_vec[0];
 115     fshift           = fr->fshift[0];
 116     facel            = _mm256_set1_pd(fr->epsfac);
 117     charge           = mdatoms->chargeA;
 118     nvdwtype         = fr->ntype;
 119     vdwparam         = fr->nbfp;
 120     vdwtype          = mdatoms->typeA;
 121
 122     vftab            = kernel_data->table_vdw->data;
 123     vftabscale       = _mm256_set1_pd(kernel_data->table_vdw->scale);
 124
 125     sh_ewald         = _mm256_set1_pd(fr->ic->sh_ewald);
 126     beta             = _mm256_set1_pd(fr->ic->ewaldcoeff);
 127     beta2            = _mm256_mul_pd(beta,beta);
 128     beta3            = _mm256_mul_pd(beta,beta2);
 129
 130     ewtab            = fr->ic->tabq_coul_FDV0;
 131     ewtabscale       = _mm256_set1_pd(fr->ic->tabq_scale);
 132     ewtabhalfspace   = _mm256_set1_pd(0.5/fr->ic->tabq_scale);
 133
 134     /* Setup water-specific parameters */
 135     inr              = nlist->iinr[0];
 136     iq1              = _mm256_mul_pd(facel,_mm256_set1_pd(charge[inr+1]));
 137     iq2              = _mm256_mul_pd(facel,_mm256_set1_pd(charge[inr+2]));
 138     iq3              = _mm256_mul_pd(facel,_mm256_set1_pd(charge[inr+3]));
 139     vdwioffsetptr0   = vdwparam+2*nvdwtype*vdwtype[inr+0];
 140
 141     /* Avoid stupid compiler warnings */
 142     jnrA = jnrB = jnrC = jnrD = 0;
 143     j_coord_offsetA = 0;
 144     j_coord_offsetB = 0;
 145     j_coord_offsetC = 0;
 146     j_coord_offsetD = 0;
 147
 148     outeriter        = 0;
 149     inneriter        = 0;
 150
 151     for(iidx=0;iidx<4*DIM;iidx++)
 152     {
 153         scratch[iidx] = 0.0;
 154     }
 155
 156     /* Start outer loop over neighborlists */
 157     for(iidx=0; iidx<nri; iidx++)
 158     {
 159         /* Load shift vector for this list */
 160         i_shift_offset   = DIM*shiftidx[iidx];
 161
 162         /* Load limits for loop over neighbors */
 163         j_index_start    = jindex[iidx];
 164         j_index_end      = jindex[iidx+1];
 165
 166         /* Get outer coordinate index */
 167         inr              = iinr[iidx];
 168         i_coord_offset   = DIM*inr;
 169
 170         /* Load i particle coords and add shift vector */
 171         gmx_mm256_load_shift_and_4rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset,
 172                                                     &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2,&ix3,&iy3,&iz3);
 173
 174         fix0             = _mm256_setzero_pd();
 175         fiy0             = _mm256_setzero_pd();
 176         fiz0             = _mm256_setzero_pd();
 177         fix1             = _mm256_setzero_pd();
 178         fiy1             = _mm256_setzero_pd();
 179         fiz1             = _mm256_setzero_pd();
 180         fix2             = _mm256_setzero_pd();
 181         fiy2             = _mm256_setzero_pd();
 182         fiz2             = _mm256_setzero_pd();
 183         fix3             = _mm256_setzero_pd();
 184         fiy3             = _mm256_setzero_pd();
 185         fiz3             = _mm256_setzero_pd();
 186
 187         /* Reset potential sums */
 188         velecsum         = _mm256_setzero_pd();
 189         vvdwsum          = _mm256_setzero_pd();
 190
 191         /* Start inner kernel loop */
 192         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
 193         {
 194
 195             /* Get j neighbor index, and coordinate index */
 196             jnrA             = jjnr[jidx];
 197             jnrB             = jjnr[jidx+1];
 198             jnrC             = jjnr[jidx+2];
 199             jnrD             = jjnr[jidx+3];
 200             j_coord_offsetA  = DIM*jnrA;
 201             j_coord_offsetB  = DIM*jnrB;
 202             j_coord_offsetC  = DIM*jnrC;
 203             j_coord_offsetD  = DIM*jnrD;
 204
 205             /* load j atom coordinates */
 206             gmx_mm256_load_1rvec_4ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 207                                                  x+j_coord_offsetC,x+j_coord_offsetD,
 208                                                  &jx0,&jy0,&jz0);
 209
 210             /* Calculate displacement vector */
 211             dx00             = _mm256_sub_pd(ix0,jx0);
 212             dy00             = _mm256_sub_pd(iy0,jy0);
 213             dz00             = _mm256_sub_pd(iz0,jz0);
 214             dx10             = _mm256_sub_pd(ix1,jx0);
 215             dy10             = _mm256_sub_pd(iy1,jy0);
 216             dz10             = _mm256_sub_pd(iz1,jz0);
 217             dx20             = _mm256_sub_pd(ix2,jx0);
 218             dy20             = _mm256_sub_pd(iy2,jy0);
 219             dz20             = _mm256_sub_pd(iz2,jz0);
 220             dx30             = _mm256_sub_pd(ix3,jx0);
 221             dy30             = _mm256_sub_pd(iy3,jy0);
 222             dz30             = _mm256_sub_pd(iz3,jz0);
 223
 224             /* Calculate squared distance and things based on it */
 225             rsq00            = gmx_mm256_calc_rsq_pd(dx00,dy00,dz00);
 226             rsq10            = gmx_mm256_calc_rsq_pd(dx10,dy10,dz10);
 227             rsq20            = gmx_mm256_calc_rsq_pd(dx20,dy20,dz20);
 228             rsq30            = gmx_mm256_calc_rsq_pd(dx30,dy30,dz30);
 229
 230             rinv00           = gmx_mm256_invsqrt_pd(rsq00);
 231             rinv10           = gmx_mm256_invsqrt_pd(rsq10);
 232             rinv20           = gmx_mm256_invsqrt_pd(rsq20);
 233             rinv30           = gmx_mm256_invsqrt_pd(rsq30);
 234
 235             rinvsq10         = _mm256_mul_pd(rinv10,rinv10);
 236             rinvsq20         = _mm256_mul_pd(rinv20,rinv20);
 237             rinvsq30         = _mm256_mul_pd(rinv30,rinv30);
 238
 239             /* Load parameters for j particles */
 240             jq0              = gmx_mm256_load_4real_swizzle_pd(charge+jnrA+0,charge+jnrB+0,
 241                                                                  charge+jnrC+0,charge+jnrD+0);
 242             vdwjidx0A        = 2*vdwtype[jnrA+0];
 243             vdwjidx0B        = 2*vdwtype[jnrB+0];
 244             vdwjidx0C        = 2*vdwtype[jnrC+0];
 245             vdwjidx0D        = 2*vdwtype[jnrD+0];
 246
 247             fjx0             = _mm256_setzero_pd();
 248             fjy0             = _mm256_setzero_pd();
 249             fjz0             = _mm256_setzero_pd();
 250
 251             /**************************
 252              * CALCULATE INTERACTIONS *
 253              **************************/
 254
 255             r00              = _mm256_mul_pd(rsq00,rinv00);
 256
 257             /* Compute parameters for interactions between i and j atoms */
 258             gmx_mm256_load_4pair_swizzle_pd(vdwioffsetptr0+vdwjidx0A,
 259                                             vdwioffsetptr0+vdwjidx0B,
 260                                             vdwioffsetptr0+vdwjidx0C,
 261                                             vdwioffsetptr0+vdwjidx0D,
 262                                             &c6_00,&c12_00);
 263
 264             /* Calculate table index by multiplying r with table scale and truncate to integer */
 265             rt               = _mm256_mul_pd(r00,vftabscale);
 266             vfitab           = _mm256_cvttpd_epi32(rt);
 267             vfeps            = _mm256_sub_pd(rt,_mm256_round_pd(rt, _MM_FROUND_FLOOR));
 268             vfitab           = _mm_slli_epi32(vfitab,3);
 269
 270             /* CUBIC SPLINE TABLE DISPERSION */
 271             Y                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
 272             F                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
 273             G                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,2) );
 274             H                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,3) );
 275             GMX_MM256_FULLTRANSPOSE4_PD(Y,F,G,H);
 276             Heps             = _mm256_mul_pd(vfeps,H);
 277             Fp               = _mm256_add_pd(F,_mm256_mul_pd(vfeps,_mm256_add_pd(G,Heps)));
 278             VV               = _mm256_add_pd(Y,_mm256_mul_pd(vfeps,Fp));
 279             vvdw6            = _mm256_mul_pd(c6_00,VV);
 280             FF               = _mm256_add_pd(Fp,_mm256_mul_pd(vfeps,_mm256_add_pd(G,_mm256_add_pd(Heps,Heps))));
 281             fvdw6            = _mm256_mul_pd(c6_00,FF);
 282
 283             /* CUBIC SPLINE TABLE REPULSION */
 284             vfitab           = _mm_add_epi32(vfitab,ifour);
 285             Y                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
 286             F                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
 287             G                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,2) );
 288             H                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,3) );
 289             GMX_MM256_FULLTRANSPOSE4_PD(Y,F,G,H);
 290             Heps             = _mm256_mul_pd(vfeps,H);
 291             Fp               = _mm256_add_pd(F,_mm256_mul_pd(vfeps,_mm256_add_pd(G,Heps)));
 292             VV               = _mm256_add_pd(Y,_mm256_mul_pd(vfeps,Fp));
 293             vvdw12           = _mm256_mul_pd(c12_00,VV);
 294             FF               = _mm256_add_pd(Fp,_mm256_mul_pd(vfeps,_mm256_add_pd(G,_mm256_add_pd(Heps,Heps))));
 295             fvdw12           = _mm256_mul_pd(c12_00,FF);
 296             vvdw             = _mm256_add_pd(vvdw12,vvdw6);
 297             fvdw             = _mm256_xor_pd(signbit,_mm256_mul_pd(_mm256_add_pd(fvdw6,fvdw12),_mm256_mul_pd(vftabscale,rinv00)));
 298
 299             /* Update potential sum for this i atom from the interaction with this j atom. */
 300             vvdwsum          = _mm256_add_pd(vvdwsum,vvdw);
 301
 302             fscal            = fvdw;
 303
 304             /* Calculate temporary vectorial force */
 305             tx               = _mm256_mul_pd(fscal,dx00);
 306             ty               = _mm256_mul_pd(fscal,dy00);
 307             tz               = _mm256_mul_pd(fscal,dz00);
 308
 309             /* Update vectorial force */
 310             fix0             = _mm256_add_pd(fix0,tx);
 311             fiy0             = _mm256_add_pd(fiy0,ty);
 312             fiz0             = _mm256_add_pd(fiz0,tz);
 313
 314             fjx0             = _mm256_add_pd(fjx0,tx);
 315             fjy0             = _mm256_add_pd(fjy0,ty);
 316             fjz0             = _mm256_add_pd(fjz0,tz);
 317
 318             /**************************
 319              * CALCULATE INTERACTIONS *
 320              **************************/
 321
 322             r10              = _mm256_mul_pd(rsq10,rinv10);
 323
 324             /* Compute parameters for interactions between i and j atoms */
 325             qq10             = _mm256_mul_pd(iq1,jq0);
 326
 327             /* EWALD ELECTROSTATICS */
 328
 329             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 330             ewrt             = _mm256_mul_pd(r10,ewtabscale);
 331             ewitab           = _mm256_cvttpd_epi32(ewrt);
 332             eweps            = _mm256_sub_pd(ewrt,_mm256_round_pd(ewrt, _MM_FROUND_FLOOR));
 333             ewitab           = _mm_slli_epi32(ewitab,2);
 334             ewtabF           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
 335             ewtabD           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,1) );
 336             ewtabV           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,2) );
 337             ewtabFn          = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,3) );
 338             GMX_MM256_FULLTRANSPOSE4_PD(ewtabF,ewtabD,ewtabV,ewtabFn);
 339             felec            = _mm256_add_pd(ewtabF,_mm256_mul_pd(eweps,ewtabD));
 340             velec            = _mm256_sub_pd(ewtabV,_mm256_mul_pd(_mm256_mul_pd(ewtabhalfspace,eweps),_mm256_add_pd(ewtabF,felec)));
 341             velec            = _mm256_mul_pd(qq10,_mm256_sub_pd(rinv10,velec));
 342             felec            = _mm256_mul_pd(_mm256_mul_pd(qq10,rinv10),_mm256_sub_pd(rinvsq10,felec));
 343
 344             /* Update potential sum for this i atom from the interaction with this j atom. */
 345             velecsum         = _mm256_add_pd(velecsum,velec);
 346
 347             fscal            = felec;
 348
 349             /* Calculate temporary vectorial force */
 350             tx               = _mm256_mul_pd(fscal,dx10);
 351             ty               = _mm256_mul_pd(fscal,dy10);
 352             tz               = _mm256_mul_pd(fscal,dz10);
 353
 354             /* Update vectorial force */
 355             fix1             = _mm256_add_pd(fix1,tx);
 356             fiy1             = _mm256_add_pd(fiy1,ty);
 357             fiz1             = _mm256_add_pd(fiz1,tz);
 358
 359             fjx0             = _mm256_add_pd(fjx0,tx);
 360             fjy0             = _mm256_add_pd(fjy0,ty);
 361             fjz0             = _mm256_add_pd(fjz0,tz);
 362
 363             /**************************
 364              * CALCULATE INTERACTIONS *
 365              **************************/
 366
 367             r20              = _mm256_mul_pd(rsq20,rinv20);
 368
 369             /* Compute parameters for interactions between i and j atoms */
 370             qq20             = _mm256_mul_pd(iq2,jq0);
 371
 372             /* EWALD ELECTROSTATICS */
 373
 374             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 375             ewrt             = _mm256_mul_pd(r20,ewtabscale);
 376             ewitab           = _mm256_cvttpd_epi32(ewrt);
 377             eweps            = _mm256_sub_pd(ewrt,_mm256_round_pd(ewrt, _MM_FROUND_FLOOR));
 378             ewitab           = _mm_slli_epi32(ewitab,2);
 379             ewtabF           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
 380             ewtabD           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,1) );
 381             ewtabV           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,2) );
 382             ewtabFn          = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,3) );
 383             GMX_MM256_FULLTRANSPOSE4_PD(ewtabF,ewtabD,ewtabV,ewtabFn);
 384             felec            = _mm256_add_pd(ewtabF,_mm256_mul_pd(eweps,ewtabD));
 385             velec            = _mm256_sub_pd(ewtabV,_mm256_mul_pd(_mm256_mul_pd(ewtabhalfspace,eweps),_mm256_add_pd(ewtabF,felec)));
 386             velec            = _mm256_mul_pd(qq20,_mm256_sub_pd(rinv20,velec));
 387             felec            = _mm256_mul_pd(_mm256_mul_pd(qq20,rinv20),_mm256_sub_pd(rinvsq20,felec));
 388
 389             /* Update potential sum for this i atom from the interaction with this j atom. */
 390             velecsum         = _mm256_add_pd(velecsum,velec);
 391
 392             fscal            = felec;
 393
 394             /* Calculate temporary vectorial force */
 395             tx               = _mm256_mul_pd(fscal,dx20);
 396             ty               = _mm256_mul_pd(fscal,dy20);
 397             tz               = _mm256_mul_pd(fscal,dz20);
 398
 399             /* Update vectorial force */
 400             fix2             = _mm256_add_pd(fix2,tx);
 401             fiy2             = _mm256_add_pd(fiy2,ty);
 402             fiz2             = _mm256_add_pd(fiz2,tz);
 403
 404             fjx0             = _mm256_add_pd(fjx0,tx);
 405             fjy0             = _mm256_add_pd(fjy0,ty);
 406             fjz0             = _mm256_add_pd(fjz0,tz);
 407
 408             /**************************
 409              * CALCULATE INTERACTIONS *
 410              **************************/
 411
 412             r30              = _mm256_mul_pd(rsq30,rinv30);
 413
 414             /* Compute parameters for interactions between i and j atoms */
 415             qq30             = _mm256_mul_pd(iq3,jq0);
 416
 417             /* EWALD ELECTROSTATICS */
 418
 419             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 420             ewrt             = _mm256_mul_pd(r30,ewtabscale);
 421             ewitab           = _mm256_cvttpd_epi32(ewrt);
 422             eweps            = _mm256_sub_pd(ewrt,_mm256_round_pd(ewrt, _MM_FROUND_FLOOR));
 423             ewitab           = _mm_slli_epi32(ewitab,2);
 424             ewtabF           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
 425             ewtabD           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,1) );
 426             ewtabV           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,2) );
 427             ewtabFn          = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,3) );
 428             GMX_MM256_FULLTRANSPOSE4_PD(ewtabF,ewtabD,ewtabV,ewtabFn);
 429             felec            = _mm256_add_pd(ewtabF,_mm256_mul_pd(eweps,ewtabD));
 430             velec            = _mm256_sub_pd(ewtabV,_mm256_mul_pd(_mm256_mul_pd(ewtabhalfspace,eweps),_mm256_add_pd(ewtabF,felec)));
 431             velec            = _mm256_mul_pd(qq30,_mm256_sub_pd(rinv30,velec));
 432             felec            = _mm256_mul_pd(_mm256_mul_pd(qq30,rinv30),_mm256_sub_pd(rinvsq30,felec));
 433
 434             /* Update potential sum for this i atom from the interaction with this j atom. */
 435             velecsum         = _mm256_add_pd(velecsum,velec);
 436
 437             fscal            = felec;
 438
 439             /* Calculate temporary vectorial force */
 440             tx               = _mm256_mul_pd(fscal,dx30);
 441             ty               = _mm256_mul_pd(fscal,dy30);
 442             tz               = _mm256_mul_pd(fscal,dz30);
 443
 444             /* Update vectorial force */
 445             fix3             = _mm256_add_pd(fix3,tx);
 446             fiy3             = _mm256_add_pd(fiy3,ty);
 447             fiz3             = _mm256_add_pd(fiz3,tz);
 448
 449             fjx0             = _mm256_add_pd(fjx0,tx);
 450             fjy0             = _mm256_add_pd(fjy0,ty);
 451             fjz0             = _mm256_add_pd(fjz0,tz);
 452
 453             fjptrA             = f+j_coord_offsetA;
 454             fjptrB             = f+j_coord_offsetB;
 455             fjptrC             = f+j_coord_offsetC;
 456             fjptrD             = f+j_coord_offsetD;
 457
 458             gmx_mm256_decrement_1rvec_4ptr_swizzle_pd(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
 459
 460             /* Inner loop uses 182 flops */
 461         }
 462
 463         if(jidx<j_index_end)
 464         {
 465
 466             /* Get j neighbor index, and coordinate index */
 467             jnrlistA         = jjnr[jidx];
 468             jnrlistB         = jjnr[jidx+1];
 469             jnrlistC         = jjnr[jidx+2];
 470             jnrlistD         = jjnr[jidx+3];
 471             /* Sign of each element will be negative for non-real atoms.
 472              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 473              * so use it as val = _mm_andnot_pd(mask,val) to clear dummy entries.
 474              */
 475             tmpmask0 = gmx_mm_castsi128_pd(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
 476
 477             tmpmask1 = _mm_permute_ps(tmpmask0,_GMX_MM_PERMUTE(3,3,2,2));
 478             tmpmask0 = _mm_permute_ps(tmpmask0,_GMX_MM_PERMUTE(1,1,0,0));
 479             dummy_mask = _mm256_castps_pd(gmx_mm256_set_m128(tmpmask1,tmpmask0));
 480
 481             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
 482             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
 483             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
 484             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
 485             j_coord_offsetA  = DIM*jnrA;
 486             j_coord_offsetB  = DIM*jnrB;
 487             j_coord_offsetC  = DIM*jnrC;
 488             j_coord_offsetD  = DIM*jnrD;
 489
 490             /* load j atom coordinates */
 491             gmx_mm256_load_1rvec_4ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 492                                                  x+j_coord_offsetC,x+j_coord_offsetD,
 493                                                  &jx0,&jy0,&jz0);
 494
 495             /* Calculate displacement vector */
 496             dx00             = _mm256_sub_pd(ix0,jx0);
 497             dy00             = _mm256_sub_pd(iy0,jy0);
 498             dz00             = _mm256_sub_pd(iz0,jz0);
 499             dx10             = _mm256_sub_pd(ix1,jx0);
 500             dy10             = _mm256_sub_pd(iy1,jy0);
 501             dz10             = _mm256_sub_pd(iz1,jz0);
 502             dx20             = _mm256_sub_pd(ix2,jx0);
 503             dy20             = _mm256_sub_pd(iy2,jy0);
 504             dz20             = _mm256_sub_pd(iz2,jz0);
 505             dx30             = _mm256_sub_pd(ix3,jx0);
 506             dy30             = _mm256_sub_pd(iy3,jy0);
 507             dz30             = _mm256_sub_pd(iz3,jz0);
 508
 509             /* Calculate squared distance and things based on it */
 510             rsq00            = gmx_mm256_calc_rsq_pd(dx00,dy00,dz00);
 511             rsq10            = gmx_mm256_calc_rsq_pd(dx10,dy10,dz10);
 512             rsq20            = gmx_mm256_calc_rsq_pd(dx20,dy20,dz20);
 513             rsq30            = gmx_mm256_calc_rsq_pd(dx30,dy30,dz30);
 514
 515             rinv00           = gmx_mm256_invsqrt_pd(rsq00);
 516             rinv10           = gmx_mm256_invsqrt_pd(rsq10);
 517             rinv20           = gmx_mm256_invsqrt_pd(rsq20);
 518             rinv30           = gmx_mm256_invsqrt_pd(rsq30);
 519
 520             rinvsq10         = _mm256_mul_pd(rinv10,rinv10);
 521             rinvsq20         = _mm256_mul_pd(rinv20,rinv20);
 522             rinvsq30         = _mm256_mul_pd(rinv30,rinv30);
 523
 524             /* Load parameters for j particles */
 525             jq0              = gmx_mm256_load_4real_swizzle_pd(charge+jnrA+0,charge+jnrB+0,
 526                                                                  charge+jnrC+0,charge+jnrD+0);
 527             vdwjidx0A        = 2*vdwtype[jnrA+0];
 528             vdwjidx0B        = 2*vdwtype[jnrB+0];
 529             vdwjidx0C        = 2*vdwtype[jnrC+0];
 530             vdwjidx0D        = 2*vdwtype[jnrD+0];
 531
 532             fjx0             = _mm256_setzero_pd();
 533             fjy0             = _mm256_setzero_pd();
 534             fjz0             = _mm256_setzero_pd();
 535
 536             /**************************
 537              * CALCULATE INTERACTIONS *
 538              **************************/
 539
 540             r00              = _mm256_mul_pd(rsq00,rinv00);
 541             r00              = _mm256_andnot_pd(dummy_mask,r00);
 542
 543             /* Compute parameters for interactions between i and j atoms */
 544             gmx_mm256_load_4pair_swizzle_pd(vdwioffsetptr0+vdwjidx0A,
 545                                             vdwioffsetptr0+vdwjidx0B,
 546                                             vdwioffsetptr0+vdwjidx0C,
 547                                             vdwioffsetptr0+vdwjidx0D,
 548                                             &c6_00,&c12_00);
 549
 550             /* Calculate table index by multiplying r with table scale and truncate to integer */
 551             rt               = _mm256_mul_pd(r00,vftabscale);
 552             vfitab           = _mm256_cvttpd_epi32(rt);
 553             vfeps            = _mm256_sub_pd(rt,_mm256_round_pd(rt, _MM_FROUND_FLOOR));
 554             vfitab           = _mm_slli_epi32(vfitab,3);
 555
 556             /* CUBIC SPLINE TABLE DISPERSION */
 557             Y                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
 558             F                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
 559             G                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,2) );
 560             H                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,3) );
 561             GMX_MM256_FULLTRANSPOSE4_PD(Y,F,G,H);
 562             Heps             = _mm256_mul_pd(vfeps,H);
 563             Fp               = _mm256_add_pd(F,_mm256_mul_pd(vfeps,_mm256_add_pd(G,Heps)));
 564             VV               = _mm256_add_pd(Y,_mm256_mul_pd(vfeps,Fp));
 565             vvdw6            = _mm256_mul_pd(c6_00,VV);
 566             FF               = _mm256_add_pd(Fp,_mm256_mul_pd(vfeps,_mm256_add_pd(G,_mm256_add_pd(Heps,Heps))));
 567             fvdw6            = _mm256_mul_pd(c6_00,FF);
 568
 569             /* CUBIC SPLINE TABLE REPULSION */
 570             vfitab           = _mm_add_epi32(vfitab,ifour);
 571             Y                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
 572             F                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
 573             G                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,2) );
 574             H                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,3) );
 575             GMX_MM256_FULLTRANSPOSE4_PD(Y,F,G,H);
 576             Heps             = _mm256_mul_pd(vfeps,H);
 577             Fp               = _mm256_add_pd(F,_mm256_mul_pd(vfeps,_mm256_add_pd(G,Heps)));
 578             VV               = _mm256_add_pd(Y,_mm256_mul_pd(vfeps,Fp));
 579             vvdw12           = _mm256_mul_pd(c12_00,VV);
 580             FF               = _mm256_add_pd(Fp,_mm256_mul_pd(vfeps,_mm256_add_pd(G,_mm256_add_pd(Heps,Heps))));
 581             fvdw12           = _mm256_mul_pd(c12_00,FF);
 582             vvdw             = _mm256_add_pd(vvdw12,vvdw6);
 583             fvdw             = _mm256_xor_pd(signbit,_mm256_mul_pd(_mm256_add_pd(fvdw6,fvdw12),_mm256_mul_pd(vftabscale,rinv00)));
 584
 585             /* Update potential sum for this i atom from the interaction with this j atom. */
 586             vvdw             = _mm256_andnot_pd(dummy_mask,vvdw);
 587             vvdwsum          = _mm256_add_pd(vvdwsum,vvdw);
 588
 589             fscal            = fvdw;
 590
 591             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
 592
 593             /* Calculate temporary vectorial force */
 594             tx               = _mm256_mul_pd(fscal,dx00);
 595             ty               = _mm256_mul_pd(fscal,dy00);
 596             tz               = _mm256_mul_pd(fscal,dz00);
 597
 598             /* Update vectorial force */
 599             fix0             = _mm256_add_pd(fix0,tx);
 600             fiy0             = _mm256_add_pd(fiy0,ty);
 601             fiz0             = _mm256_add_pd(fiz0,tz);
 602
 603             fjx0             = _mm256_add_pd(fjx0,tx);
 604             fjy0             = _mm256_add_pd(fjy0,ty);
 605             fjz0             = _mm256_add_pd(fjz0,tz);
 606
 607             /**************************
 608              * CALCULATE INTERACTIONS *
 609              **************************/
 610
 611             r10              = _mm256_mul_pd(rsq10,rinv10);
 612             r10              = _mm256_andnot_pd(dummy_mask,r10);
 613
 614             /* Compute parameters for interactions between i and j atoms */
 615             qq10             = _mm256_mul_pd(iq1,jq0);
 616
 617             /* EWALD ELECTROSTATICS */
 618
 619             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 620             ewrt             = _mm256_mul_pd(r10,ewtabscale);
 621             ewitab           = _mm256_cvttpd_epi32(ewrt);
 622             eweps            = _mm256_sub_pd(ewrt,_mm256_round_pd(ewrt, _MM_FROUND_FLOOR));
 623             ewitab           = _mm_slli_epi32(ewitab,2);
 624             ewtabF           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
 625             ewtabD           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,1) );
 626             ewtabV           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,2) );
 627             ewtabFn          = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,3) );
 628             GMX_MM256_FULLTRANSPOSE4_PD(ewtabF,ewtabD,ewtabV,ewtabFn);
 629             felec            = _mm256_add_pd(ewtabF,_mm256_mul_pd(eweps,ewtabD));
 630             velec            = _mm256_sub_pd(ewtabV,_mm256_mul_pd(_mm256_mul_pd(ewtabhalfspace,eweps),_mm256_add_pd(ewtabF,felec)));
 631             velec            = _mm256_mul_pd(qq10,_mm256_sub_pd(rinv10,velec));
 632             felec            = _mm256_mul_pd(_mm256_mul_pd(qq10,rinv10),_mm256_sub_pd(rinvsq10,felec));
 633
 634             /* Update potential sum for this i atom from the interaction with this j atom. */
 635             velec            = _mm256_andnot_pd(dummy_mask,velec);
 636             velecsum         = _mm256_add_pd(velecsum,velec);
 637
 638             fscal            = felec;
 639
 640             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
 641
 642             /* Calculate temporary vectorial force */
 643             tx               = _mm256_mul_pd(fscal,dx10);
 644             ty               = _mm256_mul_pd(fscal,dy10);
 645             tz               = _mm256_mul_pd(fscal,dz10);
 646
 647             /* Update vectorial force */
 648             fix1             = _mm256_add_pd(fix1,tx);
 649             fiy1             = _mm256_add_pd(fiy1,ty);
 650             fiz1             = _mm256_add_pd(fiz1,tz);
 651
 652             fjx0             = _mm256_add_pd(fjx0,tx);
 653             fjy0             = _mm256_add_pd(fjy0,ty);
 654             fjz0             = _mm256_add_pd(fjz0,tz);
 655
 656             /**************************
 657              * CALCULATE INTERACTIONS *
 658              **************************/
 659
 660             r20              = _mm256_mul_pd(rsq20,rinv20);
 661             r20              = _mm256_andnot_pd(dummy_mask,r20);
 662
 663             /* Compute parameters for interactions between i and j atoms */
 664             qq20             = _mm256_mul_pd(iq2,jq0);
 665
 666             /* EWALD ELECTROSTATICS */
 667
 668             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 669             ewrt             = _mm256_mul_pd(r20,ewtabscale);
 670             ewitab           = _mm256_cvttpd_epi32(ewrt);
 671             eweps            = _mm256_sub_pd(ewrt,_mm256_round_pd(ewrt, _MM_FROUND_FLOOR));
 672             ewitab           = _mm_slli_epi32(ewitab,2);
 673             ewtabF           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
 674             ewtabD           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,1) );
 675             ewtabV           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,2) );
 676             ewtabFn          = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,3) );
 677             GMX_MM256_FULLTRANSPOSE4_PD(ewtabF,ewtabD,ewtabV,ewtabFn);
 678             felec            = _mm256_add_pd(ewtabF,_mm256_mul_pd(eweps,ewtabD));
 679             velec            = _mm256_sub_pd(ewtabV,_mm256_mul_pd(_mm256_mul_pd(ewtabhalfspace,eweps),_mm256_add_pd(ewtabF,felec)));
 680             velec            = _mm256_mul_pd(qq20,_mm256_sub_pd(rinv20,velec));
 681             felec            = _mm256_mul_pd(_mm256_mul_pd(qq20,rinv20),_mm256_sub_pd(rinvsq20,felec));
 682
 683             /* Update potential sum for this i atom from the interaction with this j atom. */
 684             velec            = _mm256_andnot_pd(dummy_mask,velec);
 685             velecsum         = _mm256_add_pd(velecsum,velec);
 686
 687             fscal            = felec;
 688
 689             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
 690
 691             /* Calculate temporary vectorial force */
 692             tx               = _mm256_mul_pd(fscal,dx20);
 693             ty               = _mm256_mul_pd(fscal,dy20);
 694             tz               = _mm256_mul_pd(fscal,dz20);
 695
 696             /* Update vectorial force */
 697             fix2             = _mm256_add_pd(fix2,tx);
 698             fiy2             = _mm256_add_pd(fiy2,ty);
 699             fiz2             = _mm256_add_pd(fiz2,tz);
 700
 701             fjx0             = _mm256_add_pd(fjx0,tx);
 702             fjy0             = _mm256_add_pd(fjy0,ty);
 703             fjz0             = _mm256_add_pd(fjz0,tz);
 704
 705             /**************************
 706              * CALCULATE INTERACTIONS *
 707              **************************/
 708
 709             r30              = _mm256_mul_pd(rsq30,rinv30);
 710             r30              = _mm256_andnot_pd(dummy_mask,r30);
 711
 712             /* Compute parameters for interactions between i and j atoms */
 713             qq30             = _mm256_mul_pd(iq3,jq0);
 714
 715             /* EWALD ELECTROSTATICS */
 716
 717             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 718             ewrt             = _mm256_mul_pd(r30,ewtabscale);
 719             ewitab           = _mm256_cvttpd_epi32(ewrt);
 720             eweps            = _mm256_sub_pd(ewrt,_mm256_round_pd(ewrt, _MM_FROUND_FLOOR));
 721             ewitab           = _mm_slli_epi32(ewitab,2);
 722             ewtabF           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
 723             ewtabD           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,1) );
 724             ewtabV           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,2) );
 725             ewtabFn          = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,3) );
 726             GMX_MM256_FULLTRANSPOSE4_PD(ewtabF,ewtabD,ewtabV,ewtabFn);
 727             felec            = _mm256_add_pd(ewtabF,_mm256_mul_pd(eweps,ewtabD));
 728             velec            = _mm256_sub_pd(ewtabV,_mm256_mul_pd(_mm256_mul_pd(ewtabhalfspace,eweps),_mm256_add_pd(ewtabF,felec)));
 729             velec            = _mm256_mul_pd(qq30,_mm256_sub_pd(rinv30,velec));
 730             felec            = _mm256_mul_pd(_mm256_mul_pd(qq30,rinv30),_mm256_sub_pd(rinvsq30,felec));
 731
 732             /* Update potential sum for this i atom from the interaction with this j atom. */
 733             velec            = _mm256_andnot_pd(dummy_mask,velec);
 734             velecsum         = _mm256_add_pd(velecsum,velec);
 735
 736             fscal            = felec;
 737
 738             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
 739
 740             /* Calculate temporary vectorial force */
 741             tx               = _mm256_mul_pd(fscal,dx30);
 742             ty               = _mm256_mul_pd(fscal,dy30);
 743             tz               = _mm256_mul_pd(fscal,dz30);
 744
 745             /* Update vectorial force */
 746             fix3             = _mm256_add_pd(fix3,tx);
 747             fiy3             = _mm256_add_pd(fiy3,ty);
 748             fiz3             = _mm256_add_pd(fiz3,tz);
 749
 750             fjx0             = _mm256_add_pd(fjx0,tx);
 751             fjy0             = _mm256_add_pd(fjy0,ty);
 752             fjz0             = _mm256_add_pd(fjz0,tz);
 753
 754             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
 755             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
 756             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
 757             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
 758
 759             gmx_mm256_decrement_1rvec_4ptr_swizzle_pd(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
 760
 761             /* Inner loop uses 186 flops */
 762         }
 763
 764         /* End of innermost loop */
 765
 766         gmx_mm256_update_iforce_4atom_swizzle_pd(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,fix3,fiy3,fiz3,
 767                                                  f+i_coord_offset,fshift+i_shift_offset);
 768
 769         ggid                        = gid[iidx];
 770         /* Update potential energies */
 771         gmx_mm256_update_1pot_pd(velecsum,kernel_data->energygrp_elec+ggid);
 772         gmx_mm256_update_1pot_pd(vvdwsum,kernel_data->energygrp_vdw+ggid);
 773
 774         /* Increment number of inner iterations */
 775         inneriter                  += j_index_end - j_index_start;
 776
 777         /* Outer loop uses 26 flops */
 778     }
 779
 780     /* Increment number of outer iterations */
 781     outeriter        += nri;
 782
 783     /* Update outer/inner flops */
 784
 785     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W4_VF,outeriter*26 + inneriter*186);
 786 }
 787 /*
 788  * Gromacs nonbonded kernel:   nb_kernel_ElecEw_VdwCSTab_GeomW4P1_F_avx_256_double
 789  * Electrostatics interaction: Ewald
 790  * VdW interaction:            CubicSplineTable
 791  * Geometry:                   Water4-Particle
 792  * Calculate force/pot:        Force
 793  */
 794 void
 795 nb_kernel_ElecEw_VdwCSTab_GeomW4P1_F_avx_256_double
 796                     (t_nblist * gmx_restrict                nlist,
 797                      rvec * gmx_restrict                    xx,
 798                      rvec * gmx_restrict                    ff,
 799                      t_forcerec * gmx_restrict              fr,
 800                      t_mdatoms * gmx_restrict               mdatoms,
 801                      nb_kernel_data_t * gmx_restrict        kernel_data,
 802                      t_nrnb * gmx_restrict                  nrnb)
 803 {
 804     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
 805      * just 0 for non-waters.
 806      * Suffixes A,B,C,D refer to j loop unrolling done with AVX, e.g. for the four different
 807      * jnr indices corresponding to data put in the four positions in the SIMD register.
 808      */
 809     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
 810     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
 811     int              jnrA,jnrB,jnrC,jnrD;
 812     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
 813     int              jnrlistE,jnrlistF,jnrlistG,jnrlistH;
 814     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
 815     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
 816     real             rcutoff_scalar;
 817     real             *shiftvec,*fshift,*x,*f;
 818     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD;
 819     real             scratch[4*DIM];
 820     __m256d          tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
 821     real *           vdwioffsetptr0;
 822     __m256d          ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
 823     real *           vdwioffsetptr1;
 824     __m256d          ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
 825     real *           vdwioffsetptr2;
 826     __m256d          ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
 827     real *           vdwioffsetptr3;
 828     __m256d          ix3,iy3,iz3,fix3,fiy3,fiz3,iq3,isai3;
 829     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
 830     __m256d          jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
 831     __m256d          dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
 832     __m256d          dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
 833     __m256d          dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
 834     __m256d          dx30,dy30,dz30,rsq30,rinv30,rinvsq30,r30,qq30,c6_30,c12_30;
 835     __m256d          velec,felec,velecsum,facel,crf,krf,krf2;
 836     real             *charge;
 837     int              nvdwtype;
 838     __m256d          rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
 839     int              *vdwtype;
 840     real             *vdwparam;
 841     __m256d          one_sixth   = _mm256_set1_pd(1.0/6.0);
 842     __m256d          one_twelfth = _mm256_set1_pd(1.0/12.0);
 843     __m128i          vfitab;
 844     __m128i          ifour       = _mm_set1_epi32(4);
 845     __m256d          rt,vfeps,vftabscale,Y,F,G,H,Heps,Fp,VV,FF;
 846     real             *vftab;
 847     __m128i          ewitab;
 848     __m256d          ewtabscale,eweps,sh_ewald,ewrt,ewtabhalfspace,ewtabF,ewtabFn,ewtabD,ewtabV;
 849     __m256d          beta,beta2,beta3,zeta2,pmecorrF,pmecorrV,rinv3;
 850     real             *ewtab;
 851     __m256d          dummy_mask,cutoff_mask;
 852     __m128           tmpmask0,tmpmask1;
 853     __m256d          signbit = _mm256_castsi256_pd( _mm256_set1_epi32(0x80000000) );
 854     __m256d          one     = _mm256_set1_pd(1.0);
 855     __m256d          two     = _mm256_set1_pd(2.0);
 856     x                = xx[0];
 857     f                = ff[0];
 858
 859     nri              = nlist->nri;
 860     iinr             = nlist->iinr;
 861     jindex           = nlist->jindex;
 862     jjnr             = nlist->jjnr;
 863     shiftidx         = nlist->shift;
 864     gid              = nlist->gid;
 865     shiftvec         = fr->shift_vec[0];
 866     fshift           = fr->fshift[0];
 867     facel            = _mm256_set1_pd(fr->epsfac);
 868     charge           = mdatoms->chargeA;
 869     nvdwtype         = fr->ntype;
 870     vdwparam         = fr->nbfp;
 871     vdwtype          = mdatoms->typeA;
 872
 873     vftab            = kernel_data->table_vdw->data;
 874     vftabscale       = _mm256_set1_pd(kernel_data->table_vdw->scale);
 875
 876     sh_ewald         = _mm256_set1_pd(fr->ic->sh_ewald);
 877     beta             = _mm256_set1_pd(fr->ic->ewaldcoeff);
 878     beta2            = _mm256_mul_pd(beta,beta);
 879     beta3            = _mm256_mul_pd(beta,beta2);
 880
 881     ewtab            = fr->ic->tabq_coul_F;
 882     ewtabscale       = _mm256_set1_pd(fr->ic->tabq_scale);
 883     ewtabhalfspace   = _mm256_set1_pd(0.5/fr->ic->tabq_scale);
 884
 885     /* Setup water-specific parameters */
 886     inr              = nlist->iinr[0];
 887     iq1              = _mm256_mul_pd(facel,_mm256_set1_pd(charge[inr+1]));
 888     iq2              = _mm256_mul_pd(facel,_mm256_set1_pd(charge[inr+2]));
 889     iq3              = _mm256_mul_pd(facel,_mm256_set1_pd(charge[inr+3]));
 890     vdwioffsetptr0   = vdwparam+2*nvdwtype*vdwtype[inr+0];
 891
 892     /* Avoid stupid compiler warnings */
 893     jnrA = jnrB = jnrC = jnrD = 0;
 894     j_coord_offsetA = 0;
 895     j_coord_offsetB = 0;
 896     j_coord_offsetC = 0;
 897     j_coord_offsetD = 0;
 898
 899     outeriter        = 0;
 900     inneriter        = 0;
 901
 902     for(iidx=0;iidx<4*DIM;iidx++)
 903     {
 904         scratch[iidx] = 0.0;
 905     }
 906
 907     /* Start outer loop over neighborlists */
 908     for(iidx=0; iidx<nri; iidx++)
 909     {
 910         /* Load shift vector for this list */
 911         i_shift_offset   = DIM*shiftidx[iidx];
 912
 913         /* Load limits for loop over neighbors */
 914         j_index_start    = jindex[iidx];
 915         j_index_end      = jindex[iidx+1];
 916
 917         /* Get outer coordinate index */
 918         inr              = iinr[iidx];
 919         i_coord_offset   = DIM*inr;
 920
 921         /* Load i particle coords and add shift vector */
 922         gmx_mm256_load_shift_and_4rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset,
 923                                                     &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2,&ix3,&iy3,&iz3);
 924
 925         fix0             = _mm256_setzero_pd();
 926         fiy0             = _mm256_setzero_pd();
 927         fiz0             = _mm256_setzero_pd();
 928         fix1             = _mm256_setzero_pd();
 929         fiy1             = _mm256_setzero_pd();
 930         fiz1             = _mm256_setzero_pd();
 931         fix2             = _mm256_setzero_pd();
 932         fiy2             = _mm256_setzero_pd();
 933         fiz2             = _mm256_setzero_pd();
 934         fix3             = _mm256_setzero_pd();
 935         fiy3             = _mm256_setzero_pd();
 936         fiz3             = _mm256_setzero_pd();
 937
 938         /* Start inner kernel loop */
 939         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
 940         {
 941
 942             /* Get j neighbor index, and coordinate index */
 943             jnrA             = jjnr[jidx];
 944             jnrB             = jjnr[jidx+1];
 945             jnrC             = jjnr[jidx+2];
 946             jnrD             = jjnr[jidx+3];
 947             j_coord_offsetA  = DIM*jnrA;
 948             j_coord_offsetB  = DIM*jnrB;
 949             j_coord_offsetC  = DIM*jnrC;
 950             j_coord_offsetD  = DIM*jnrD;
 951
 952             /* load j atom coordinates */
 953             gmx_mm256_load_1rvec_4ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 954                                                  x+j_coord_offsetC,x+j_coord_offsetD,
 955                                                  &jx0,&jy0,&jz0);
 956
 957             /* Calculate displacement vector */
 958             dx00             = _mm256_sub_pd(ix0,jx0);
 959             dy00             = _mm256_sub_pd(iy0,jy0);
 960             dz00             = _mm256_sub_pd(iz0,jz0);
 961             dx10             = _mm256_sub_pd(ix1,jx0);
 962             dy10             = _mm256_sub_pd(iy1,jy0);
 963             dz10             = _mm256_sub_pd(iz1,jz0);
 964             dx20             = _mm256_sub_pd(ix2,jx0);
 965             dy20             = _mm256_sub_pd(iy2,jy0);
 966             dz20             = _mm256_sub_pd(iz2,jz0);
 967             dx30             = _mm256_sub_pd(ix3,jx0);
 968             dy30             = _mm256_sub_pd(iy3,jy0);
 969             dz30             = _mm256_sub_pd(iz3,jz0);
 970
 971             /* Calculate squared distance and things based on it */
 972             rsq00            = gmx_mm256_calc_rsq_pd(dx00,dy00,dz00);
 973             rsq10            = gmx_mm256_calc_rsq_pd(dx10,dy10,dz10);
 974             rsq20            = gmx_mm256_calc_rsq_pd(dx20,dy20,dz20);
 975             rsq30            = gmx_mm256_calc_rsq_pd(dx30,dy30,dz30);
 976
 977             rinv00           = gmx_mm256_invsqrt_pd(rsq00);
 978             rinv10           = gmx_mm256_invsqrt_pd(rsq10);
 979             rinv20           = gmx_mm256_invsqrt_pd(rsq20);
 980             rinv30           = gmx_mm256_invsqrt_pd(rsq30);
 981
 982             rinvsq10         = _mm256_mul_pd(rinv10,rinv10);
 983             rinvsq20         = _mm256_mul_pd(rinv20,rinv20);
 984             rinvsq30         = _mm256_mul_pd(rinv30,rinv30);
 985
 986             /* Load parameters for j particles */
 987             jq0              = gmx_mm256_load_4real_swizzle_pd(charge+jnrA+0,charge+jnrB+0,
 988                                                                  charge+jnrC+0,charge+jnrD+0);
 989             vdwjidx0A        = 2*vdwtype[jnrA+0];
 990             vdwjidx0B        = 2*vdwtype[jnrB+0];
 991             vdwjidx0C        = 2*vdwtype[jnrC+0];
 992             vdwjidx0D        = 2*vdwtype[jnrD+0];
 993
 994             fjx0             = _mm256_setzero_pd();
 995             fjy0             = _mm256_setzero_pd();
 996             fjz0             = _mm256_setzero_pd();
 997
 998             /**************************
 999              * CALCULATE INTERACTIONS *
1000              **************************/
1001
1002             r00              = _mm256_mul_pd(rsq00,rinv00);
1003
1004             /* Compute parameters for interactions between i and j atoms */
1005             gmx_mm256_load_4pair_swizzle_pd(vdwioffsetptr0+vdwjidx0A,
1006                                             vdwioffsetptr0+vdwjidx0B,
1007                                             vdwioffsetptr0+vdwjidx0C,
1008                                             vdwioffsetptr0+vdwjidx0D,
1009                                             &c6_00,&c12_00);
1010
1011             /* Calculate table index by multiplying r with table scale and truncate to integer */
1012             rt               = _mm256_mul_pd(r00,vftabscale);
1013             vfitab           = _mm256_cvttpd_epi32(rt);
1014             vfeps            = _mm256_sub_pd(rt,_mm256_round_pd(rt, _MM_FROUND_FLOOR));
1015             vfitab           = _mm_slli_epi32(vfitab,3);
1016
1017             /* CUBIC SPLINE TABLE DISPERSION */
1018             Y                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
1019             F                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
1020             G                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,2) );
1021             H                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,3) );
1022             GMX_MM256_FULLTRANSPOSE4_PD(Y,F,G,H);
1023             Heps             = _mm256_mul_pd(vfeps,H);
1024             Fp               = _mm256_add_pd(F,_mm256_mul_pd(vfeps,_mm256_add_pd(G,Heps)));
1025             FF               = _mm256_add_pd(Fp,_mm256_mul_pd(vfeps,_mm256_add_pd(G,_mm256_add_pd(Heps,Heps))));
1026             fvdw6            = _mm256_mul_pd(c6_00,FF);
1027
1028             /* CUBIC SPLINE TABLE REPULSION */
1029             vfitab           = _mm_add_epi32(vfitab,ifour);
1030             Y                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
1031             F                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
1032             G                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,2) );
1033             H                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,3) );
1034             GMX_MM256_FULLTRANSPOSE4_PD(Y,F,G,H);
1035             Heps             = _mm256_mul_pd(vfeps,H);
1036             Fp               = _mm256_add_pd(F,_mm256_mul_pd(vfeps,_mm256_add_pd(G,Heps)));
1037             FF               = _mm256_add_pd(Fp,_mm256_mul_pd(vfeps,_mm256_add_pd(G,_mm256_add_pd(Heps,Heps))));
1038             fvdw12           = _mm256_mul_pd(c12_00,FF);
1039             fvdw             = _mm256_xor_pd(signbit,_mm256_mul_pd(_mm256_add_pd(fvdw6,fvdw12),_mm256_mul_pd(vftabscale,rinv00)));
1040
1041             fscal            = fvdw;
1042
1043             /* Calculate temporary vectorial force */
1044             tx               = _mm256_mul_pd(fscal,dx00);
1045             ty               = _mm256_mul_pd(fscal,dy00);
1046             tz               = _mm256_mul_pd(fscal,dz00);
1047
1048             /* Update vectorial force */
1049             fix0             = _mm256_add_pd(fix0,tx);
1050             fiy0             = _mm256_add_pd(fiy0,ty);
1051             fiz0             = _mm256_add_pd(fiz0,tz);
1052
1053             fjx0             = _mm256_add_pd(fjx0,tx);
1054             fjy0             = _mm256_add_pd(fjy0,ty);
1055             fjz0             = _mm256_add_pd(fjz0,tz);
1056
1057             /**************************
1058              * CALCULATE INTERACTIONS *
1059              **************************/
1060
1061             r10              = _mm256_mul_pd(rsq10,rinv10);
1062
1063             /* Compute parameters for interactions between i and j atoms */
1064             qq10             = _mm256_mul_pd(iq1,jq0);
1065
1066             /* EWALD ELECTROSTATICS */
1067
1068             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1069             ewrt             = _mm256_mul_pd(r10,ewtabscale);
1070             ewitab           = _mm256_cvttpd_epi32(ewrt);
1071             eweps            = _mm256_sub_pd(ewrt,_mm256_round_pd(ewrt, _MM_FROUND_FLOOR));
1072             gmx_mm256_load_4pair_swizzle_pd(ewtab + _mm_extract_epi32(ewitab,0),ewtab + _mm_extract_epi32(ewitab,1),
1073                                             ewtab + _mm_extract_epi32(ewitab,2),ewtab + _mm_extract_epi32(ewitab,3),
1074                                             &ewtabF,&ewtabFn);
1075             felec            = _mm256_add_pd(_mm256_mul_pd( _mm256_sub_pd(one,eweps),ewtabF),_mm256_mul_pd(eweps,ewtabFn));
1076             felec            = _mm256_mul_pd(_mm256_mul_pd(qq10,rinv10),_mm256_sub_pd(rinvsq10,felec));
1077
1078             fscal            = felec;
1079
1080             /* Calculate temporary vectorial force */
1081             tx               = _mm256_mul_pd(fscal,dx10);
1082             ty               = _mm256_mul_pd(fscal,dy10);
1083             tz               = _mm256_mul_pd(fscal,dz10);
1084
1085             /* Update vectorial force */
1086             fix1             = _mm256_add_pd(fix1,tx);
1087             fiy1             = _mm256_add_pd(fiy1,ty);
1088             fiz1             = _mm256_add_pd(fiz1,tz);
1089
1090             fjx0             = _mm256_add_pd(fjx0,tx);
1091             fjy0             = _mm256_add_pd(fjy0,ty);
1092             fjz0             = _mm256_add_pd(fjz0,tz);
1093
1094             /**************************
1095              * CALCULATE INTERACTIONS *
1096              **************************/
1097
1098             r20              = _mm256_mul_pd(rsq20,rinv20);
1099
1100             /* Compute parameters for interactions between i and j atoms */
1101             qq20             = _mm256_mul_pd(iq2,jq0);
1102
1103             /* EWALD ELECTROSTATICS */
1104
1105             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1106             ewrt             = _mm256_mul_pd(r20,ewtabscale);
1107             ewitab           = _mm256_cvttpd_epi32(ewrt);
1108             eweps            = _mm256_sub_pd(ewrt,_mm256_round_pd(ewrt, _MM_FROUND_FLOOR));
1109             gmx_mm256_load_4pair_swizzle_pd(ewtab + _mm_extract_epi32(ewitab,0),ewtab + _mm_extract_epi32(ewitab,1),
1110                                             ewtab + _mm_extract_epi32(ewitab,2),ewtab + _mm_extract_epi32(ewitab,3),
1111                                             &ewtabF,&ewtabFn);
1112             felec            = _mm256_add_pd(_mm256_mul_pd( _mm256_sub_pd(one,eweps),ewtabF),_mm256_mul_pd(eweps,ewtabFn));
1113             felec            = _mm256_mul_pd(_mm256_mul_pd(qq20,rinv20),_mm256_sub_pd(rinvsq20,felec));
1114
1115             fscal            = felec;
1116
1117             /* Calculate temporary vectorial force */
1118             tx               = _mm256_mul_pd(fscal,dx20);
1119             ty               = _mm256_mul_pd(fscal,dy20);
1120             tz               = _mm256_mul_pd(fscal,dz20);
1121
1122             /* Update vectorial force */
1123             fix2             = _mm256_add_pd(fix2,tx);
1124             fiy2             = _mm256_add_pd(fiy2,ty);
1125             fiz2             = _mm256_add_pd(fiz2,tz);
1126
1127             fjx0             = _mm256_add_pd(fjx0,tx);
1128             fjy0             = _mm256_add_pd(fjy0,ty);
1129             fjz0             = _mm256_add_pd(fjz0,tz);
1130
1131             /**************************
1132              * CALCULATE INTERACTIONS *
1133              **************************/
1134
1135             r30              = _mm256_mul_pd(rsq30,rinv30);
1136
1137             /* Compute parameters for interactions between i and j atoms */
1138             qq30             = _mm256_mul_pd(iq3,jq0);
1139
1140             /* EWALD ELECTROSTATICS */
1141
1142             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1143             ewrt             = _mm256_mul_pd(r30,ewtabscale);
1144             ewitab           = _mm256_cvttpd_epi32(ewrt);
1145             eweps            = _mm256_sub_pd(ewrt,_mm256_round_pd(ewrt, _MM_FROUND_FLOOR));
1146             gmx_mm256_load_4pair_swizzle_pd(ewtab + _mm_extract_epi32(ewitab,0),ewtab + _mm_extract_epi32(ewitab,1),
1147                                             ewtab + _mm_extract_epi32(ewitab,2),ewtab + _mm_extract_epi32(ewitab,3),
1148                                             &ewtabF,&ewtabFn);
1149             felec            = _mm256_add_pd(_mm256_mul_pd( _mm256_sub_pd(one,eweps),ewtabF),_mm256_mul_pd(eweps,ewtabFn));
1150             felec            = _mm256_mul_pd(_mm256_mul_pd(qq30,rinv30),_mm256_sub_pd(rinvsq30,felec));
1151
1152             fscal            = felec;
1153
1154             /* Calculate temporary vectorial force */
1155             tx               = _mm256_mul_pd(fscal,dx30);
1156             ty               = _mm256_mul_pd(fscal,dy30);
1157             tz               = _mm256_mul_pd(fscal,dz30);
1158
1159             /* Update vectorial force */
1160             fix3             = _mm256_add_pd(fix3,tx);
1161             fiy3             = _mm256_add_pd(fiy3,ty);
1162             fiz3             = _mm256_add_pd(fiz3,tz);
1163
1164             fjx0             = _mm256_add_pd(fjx0,tx);
1165             fjy0             = _mm256_add_pd(fjy0,ty);
1166             fjz0             = _mm256_add_pd(fjz0,tz);
1167
1168             fjptrA             = f+j_coord_offsetA;
1169             fjptrB             = f+j_coord_offsetB;
1170             fjptrC             = f+j_coord_offsetC;
1171             fjptrD             = f+j_coord_offsetD;
1172
1173             gmx_mm256_decrement_1rvec_4ptr_swizzle_pd(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
1174
1175             /* Inner loop uses 159 flops */
1176         }
1177
1178         if(jidx<j_index_end)
1179         {
1180
1181             /* Get j neighbor index, and coordinate index */
1182             jnrlistA         = jjnr[jidx];
1183             jnrlistB         = jjnr[jidx+1];
1184             jnrlistC         = jjnr[jidx+2];
1185             jnrlistD         = jjnr[jidx+3];
1186             /* Sign of each element will be negative for non-real atoms.
1187              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
1188              * so use it as val = _mm_andnot_pd(mask,val) to clear dummy entries.
1189              */
1190             tmpmask0 = gmx_mm_castsi128_pd(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
1191
1192             tmpmask1 = _mm_permute_ps(tmpmask0,_GMX_MM_PERMUTE(3,3,2,2));
1193             tmpmask0 = _mm_permute_ps(tmpmask0,_GMX_MM_PERMUTE(1,1,0,0));
1194             dummy_mask = _mm256_castps_pd(gmx_mm256_set_m128(tmpmask1,tmpmask0));
1195
1196             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
1197             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
1198             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
1199             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
1200             j_coord_offsetA  = DIM*jnrA;
1201             j_coord_offsetB  = DIM*jnrB;
1202             j_coord_offsetC  = DIM*jnrC;
1203             j_coord_offsetD  = DIM*jnrD;
1204
1205             /* load j atom coordinates */
1206             gmx_mm256_load_1rvec_4ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
1207                                                  x+j_coord_offsetC,x+j_coord_offsetD,
1208                                                  &jx0,&jy0,&jz0);
1209
1210             /* Calculate displacement vector */
1211             dx00             = _mm256_sub_pd(ix0,jx0);
1212             dy00             = _mm256_sub_pd(iy0,jy0);
1213             dz00             = _mm256_sub_pd(iz0,jz0);
1214             dx10             = _mm256_sub_pd(ix1,jx0);
1215             dy10             = _mm256_sub_pd(iy1,jy0);
1216             dz10             = _mm256_sub_pd(iz1,jz0);
1217             dx20             = _mm256_sub_pd(ix2,jx0);
1218             dy20             = _mm256_sub_pd(iy2,jy0);
1219             dz20             = _mm256_sub_pd(iz2,jz0);
1220             dx30             = _mm256_sub_pd(ix3,jx0);
1221             dy30             = _mm256_sub_pd(iy3,jy0);
1222             dz30             = _mm256_sub_pd(iz3,jz0);
1223
1224             /* Calculate squared distance and things based on it */
1225             rsq00            = gmx_mm256_calc_rsq_pd(dx00,dy00,dz00);
1226             rsq10            = gmx_mm256_calc_rsq_pd(dx10,dy10,dz10);
1227             rsq20            = gmx_mm256_calc_rsq_pd(dx20,dy20,dz20);
1228             rsq30            = gmx_mm256_calc_rsq_pd(dx30,dy30,dz30);
1229
1230             rinv00           = gmx_mm256_invsqrt_pd(rsq00);
1231             rinv10           = gmx_mm256_invsqrt_pd(rsq10);
1232             rinv20           = gmx_mm256_invsqrt_pd(rsq20);
1233             rinv30           = gmx_mm256_invsqrt_pd(rsq30);
1234
1235             rinvsq10         = _mm256_mul_pd(rinv10,rinv10);
1236             rinvsq20         = _mm256_mul_pd(rinv20,rinv20);
1237             rinvsq30         = _mm256_mul_pd(rinv30,rinv30);
1238
1239             /* Load parameters for j particles */
1240             jq0              = gmx_mm256_load_4real_swizzle_pd(charge+jnrA+0,charge+jnrB+0,
1241                                                                  charge+jnrC+0,charge+jnrD+0);
1242             vdwjidx0A        = 2*vdwtype[jnrA+0];
1243             vdwjidx0B        = 2*vdwtype[jnrB+0];
1244             vdwjidx0C        = 2*vdwtype[jnrC+0];
1245             vdwjidx0D        = 2*vdwtype[jnrD+0];
1246
1247             fjx0             = _mm256_setzero_pd();
1248             fjy0             = _mm256_setzero_pd();
1249             fjz0             = _mm256_setzero_pd();
1250
1251             /**************************
1252              * CALCULATE INTERACTIONS *
1253              **************************/
1254
1255             r00              = _mm256_mul_pd(rsq00,rinv00);
1256             r00              = _mm256_andnot_pd(dummy_mask,r00);
1257
1258             /* Compute parameters for interactions between i and j atoms */
1259             gmx_mm256_load_4pair_swizzle_pd(vdwioffsetptr0+vdwjidx0A,
1260                                             vdwioffsetptr0+vdwjidx0B,
1261                                             vdwioffsetptr0+vdwjidx0C,
1262                                             vdwioffsetptr0+vdwjidx0D,
1263                                             &c6_00,&c12_00);
1264
1265             /* Calculate table index by multiplying r with table scale and truncate to integer */
1266             rt               = _mm256_mul_pd(r00,vftabscale);
1267             vfitab           = _mm256_cvttpd_epi32(rt);
1268             vfeps            = _mm256_sub_pd(rt,_mm256_round_pd(rt, _MM_FROUND_FLOOR));
1269             vfitab           = _mm_slli_epi32(vfitab,3);
1270
1271             /* CUBIC SPLINE TABLE DISPERSION */
1272             Y                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
1273             F                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
1274             G                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,2) );
1275             H                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,3) );
1276             GMX_MM256_FULLTRANSPOSE4_PD(Y,F,G,H);
1277             Heps             = _mm256_mul_pd(vfeps,H);
1278             Fp               = _mm256_add_pd(F,_mm256_mul_pd(vfeps,_mm256_add_pd(G,Heps)));
1279             FF               = _mm256_add_pd(Fp,_mm256_mul_pd(vfeps,_mm256_add_pd(G,_mm256_add_pd(Heps,Heps))));
1280             fvdw6            = _mm256_mul_pd(c6_00,FF);
1281
1282             /* CUBIC SPLINE TABLE REPULSION */
1283             vfitab           = _mm_add_epi32(vfitab,ifour);
1284             Y                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
1285             F                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
1286             G                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,2) );
1287             H                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,3) );
1288             GMX_MM256_FULLTRANSPOSE4_PD(Y,F,G,H);
1289             Heps             = _mm256_mul_pd(vfeps,H);
1290             Fp               = _mm256_add_pd(F,_mm256_mul_pd(vfeps,_mm256_add_pd(G,Heps)));
1291             FF               = _mm256_add_pd(Fp,_mm256_mul_pd(vfeps,_mm256_add_pd(G,_mm256_add_pd(Heps,Heps))));
1292             fvdw12           = _mm256_mul_pd(c12_00,FF);
1293             fvdw             = _mm256_xor_pd(signbit,_mm256_mul_pd(_mm256_add_pd(fvdw6,fvdw12),_mm256_mul_pd(vftabscale,rinv00)));
1294
1295             fscal            = fvdw;
1296
1297             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
1298
1299             /* Calculate temporary vectorial force */
1300             tx               = _mm256_mul_pd(fscal,dx00);
1301             ty               = _mm256_mul_pd(fscal,dy00);
1302             tz               = _mm256_mul_pd(fscal,dz00);
1303
1304             /* Update vectorial force */
1305             fix0             = _mm256_add_pd(fix0,tx);
1306             fiy0             = _mm256_add_pd(fiy0,ty);
1307             fiz0             = _mm256_add_pd(fiz0,tz);
1308
1309             fjx0             = _mm256_add_pd(fjx0,tx);
1310             fjy0             = _mm256_add_pd(fjy0,ty);
1311             fjz0             = _mm256_add_pd(fjz0,tz);
1312
1313             /**************************
1314              * CALCULATE INTERACTIONS *
1315              **************************/
1316
1317             r10              = _mm256_mul_pd(rsq10,rinv10);
1318             r10              = _mm256_andnot_pd(dummy_mask,r10);
1319
1320             /* Compute parameters for interactions between i and j atoms */
1321             qq10             = _mm256_mul_pd(iq1,jq0);
1322
1323             /* EWALD ELECTROSTATICS */
1324
1325             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1326             ewrt             = _mm256_mul_pd(r10,ewtabscale);
1327             ewitab           = _mm256_cvttpd_epi32(ewrt);
1328             eweps            = _mm256_sub_pd(ewrt,_mm256_round_pd(ewrt, _MM_FROUND_FLOOR));
1329             gmx_mm256_load_4pair_swizzle_pd(ewtab + _mm_extract_epi32(ewitab,0),ewtab + _mm_extract_epi32(ewitab,1),
1330                                             ewtab + _mm_extract_epi32(ewitab,2),ewtab + _mm_extract_epi32(ewitab,3),
1331                                             &ewtabF,&ewtabFn);
1332             felec            = _mm256_add_pd(_mm256_mul_pd( _mm256_sub_pd(one,eweps),ewtabF),_mm256_mul_pd(eweps,ewtabFn));
1333             felec            = _mm256_mul_pd(_mm256_mul_pd(qq10,rinv10),_mm256_sub_pd(rinvsq10,felec));
1334
1335             fscal            = felec;
1336
1337             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
1338
1339             /* Calculate temporary vectorial force */
1340             tx               = _mm256_mul_pd(fscal,dx10);
1341             ty               = _mm256_mul_pd(fscal,dy10);
1342             tz               = _mm256_mul_pd(fscal,dz10);
1343
1344             /* Update vectorial force */
1345             fix1             = _mm256_add_pd(fix1,tx);
1346             fiy1             = _mm256_add_pd(fiy1,ty);
1347             fiz1             = _mm256_add_pd(fiz1,tz);
1348
1349             fjx0             = _mm256_add_pd(fjx0,tx);
1350             fjy0             = _mm256_add_pd(fjy0,ty);
1351             fjz0             = _mm256_add_pd(fjz0,tz);
1352
1353             /**************************
1354              * CALCULATE INTERACTIONS *
1355              **************************/
1356
1357             r20              = _mm256_mul_pd(rsq20,rinv20);
1358             r20              = _mm256_andnot_pd(dummy_mask,r20);
1359
1360             /* Compute parameters for interactions between i and j atoms */
1361             qq20             = _mm256_mul_pd(iq2,jq0);
1362
1363             /* EWALD ELECTROSTATICS */
1364
1365             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1366             ewrt             = _mm256_mul_pd(r20,ewtabscale);
1367             ewitab           = _mm256_cvttpd_epi32(ewrt);
1368             eweps            = _mm256_sub_pd(ewrt,_mm256_round_pd(ewrt, _MM_FROUND_FLOOR));
1369             gmx_mm256_load_4pair_swizzle_pd(ewtab + _mm_extract_epi32(ewitab,0),ewtab + _mm_extract_epi32(ewitab,1),
1370                                             ewtab + _mm_extract_epi32(ewitab,2),ewtab + _mm_extract_epi32(ewitab,3),
1371                                             &ewtabF,&ewtabFn);
1372             felec            = _mm256_add_pd(_mm256_mul_pd( _mm256_sub_pd(one,eweps),ewtabF),_mm256_mul_pd(eweps,ewtabFn));
1373             felec            = _mm256_mul_pd(_mm256_mul_pd(qq20,rinv20),_mm256_sub_pd(rinvsq20,felec));
1374
1375             fscal            = felec;
1376
1377             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
1378
1379             /* Calculate temporary vectorial force */
1380             tx               = _mm256_mul_pd(fscal,dx20);
1381             ty               = _mm256_mul_pd(fscal,dy20);
1382             tz               = _mm256_mul_pd(fscal,dz20);
1383
1384             /* Update vectorial force */
1385             fix2             = _mm256_add_pd(fix2,tx);
1386             fiy2             = _mm256_add_pd(fiy2,ty);
1387             fiz2             = _mm256_add_pd(fiz2,tz);
1388
1389             fjx0             = _mm256_add_pd(fjx0,tx);
1390             fjy0             = _mm256_add_pd(fjy0,ty);
1391             fjz0             = _mm256_add_pd(fjz0,tz);
1392
1393             /**************************
1394              * CALCULATE INTERACTIONS *
1395              **************************/
1396
1397             r30              = _mm256_mul_pd(rsq30,rinv30);
1398             r30              = _mm256_andnot_pd(dummy_mask,r30);
1399
1400             /* Compute parameters for interactions between i and j atoms */
1401             qq30             = _mm256_mul_pd(iq3,jq0);
1402
1403             /* EWALD ELECTROSTATICS */
1404
1405             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1406             ewrt             = _mm256_mul_pd(r30,ewtabscale);
1407             ewitab           = _mm256_cvttpd_epi32(ewrt);
1408             eweps            = _mm256_sub_pd(ewrt,_mm256_round_pd(ewrt, _MM_FROUND_FLOOR));
1409             gmx_mm256_load_4pair_swizzle_pd(ewtab + _mm_extract_epi32(ewitab,0),ewtab + _mm_extract_epi32(ewitab,1),
1410                                             ewtab + _mm_extract_epi32(ewitab,2),ewtab + _mm_extract_epi32(ewitab,3),
1411                                             &ewtabF,&ewtabFn);
1412             felec            = _mm256_add_pd(_mm256_mul_pd( _mm256_sub_pd(one,eweps),ewtabF),_mm256_mul_pd(eweps,ewtabFn));
1413             felec            = _mm256_mul_pd(_mm256_mul_pd(qq30,rinv30),_mm256_sub_pd(rinvsq30,felec));
1414
1415             fscal            = felec;
1416
1417             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
1418
1419             /* Calculate temporary vectorial force */
1420             tx               = _mm256_mul_pd(fscal,dx30);
1421             ty               = _mm256_mul_pd(fscal,dy30);
1422             tz               = _mm256_mul_pd(fscal,dz30);
1423
1424             /* Update vectorial force */
1425             fix3             = _mm256_add_pd(fix3,tx);
1426             fiy3             = _mm256_add_pd(fiy3,ty);
1427             fiz3             = _mm256_add_pd(fiz3,tz);
1428
1429             fjx0             = _mm256_add_pd(fjx0,tx);
1430             fjy0             = _mm256_add_pd(fjy0,ty);
1431             fjz0             = _mm256_add_pd(fjz0,tz);
1432
1433             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
1434             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
1435             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
1436             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
1437
1438             gmx_mm256_decrement_1rvec_4ptr_swizzle_pd(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
1439
1440             /* Inner loop uses 163 flops */
1441         }
1442
1443         /* End of innermost loop */
1444
1445         gmx_mm256_update_iforce_4atom_swizzle_pd(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,fix3,fiy3,fiz3,
1446                                                  f+i_coord_offset,fshift+i_shift_offset);
1447
1448         /* Increment number of inner iterations */
1449         inneriter                  += j_index_end - j_index_start;
1450
1451         /* Outer loop uses 24 flops */
1452     }
1453
1454     /* Increment number of outer iterations */
1455     outeriter        += nri;
1456
1457     /* Update outer/inner flops */
1458
1459     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W4_F,outeriter*24 + inneriter*163);
1460 }