src/gromacs/gmxlib/nonbonded/nb_kernel_avx_256_single/nb_kernel_ElecCSTab_VdwLJ_GeomW3P1_avx_256_single.cpp

   1 /*
   2  * This file is part of the GROMACS molecular simulation package.
   3  *
   4  * Copyright (c) 2012,2013,2014,2015,2017,2018, by the GROMACS development team, led by
   5  * Mark Abraham, David van der Spoel, Berk Hess, and Erik Lindahl,
   6  * and including many others, as listed in the AUTHORS file in the
   7  * top-level source directory and at http://www.gromacs.org.
   8  *
   9  * GROMACS is free software; you can redistribute it and/or
  10  * modify it under the terms of the GNU Lesser General Public License
  11  * as published by the Free Software Foundation; either version 2.1
  12  * of the License, or (at your option) any later version.
  13  *
  14  * GROMACS is distributed in the hope that it will be useful,
  15  * but WITHOUT ANY WARRANTY; without even the implied warranty of
  16  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
  17  * Lesser General Public License for more details.
  18  *
  19  * You should have received a copy of the GNU Lesser General Public
  20  * License along with GROMACS; if not, see
  21  * http://www.gnu.org/licenses, or write to the Free Software Foundation,
  22  * Inc., 51 Franklin Street, Fifth Floor, Boston, MA  02110-1301  USA.
  23  *
  24  * If you want to redistribute modifications to GROMACS, please
  25  * consider that scientific software is very special. Version
  26  * control is crucial - bugs must be traceable. We will be happy to
  27  * consider code for inclusion in the official distribution, but
  28  * derived work must not be called official GROMACS. Details are found
  29  * in the README & COPYING files - if they are missing, get the
  30  * official version at http://www.gromacs.org.
  31  *
  32  * To help us fund GROMACS development, we humbly ask that you cite
  33  * the research papers on the package. Check out http://www.gromacs.org.
  34  */
  35 /*
  36  * Note: this file was generated by the GROMACS avx_256_single kernel generator.
  37  */
  38 #include "gmxpre.h"
  39
  40 #include "config.h"
  41
  42 #include <math.h>
  43
  44 #include "../nb_kernel.h"
  45 #include "gromacs/gmxlib/nrnb.h"
  46
  47 #include "kernelutil_x86_avx_256_single.h"
  48
  49 /*
  50  * Gromacs nonbonded kernel:   nb_kernel_ElecCSTab_VdwLJ_GeomW3P1_VF_avx_256_single
  51  * Electrostatics interaction: CubicSplineTable
  52  * VdW interaction:            LennardJones
  53  * Geometry:                   Water3-Particle
  54  * Calculate force/pot:        PotentialAndForce
  55  */
  56 void
  57 nb_kernel_ElecCSTab_VdwLJ_GeomW3P1_VF_avx_256_single
  58                     (t_nblist                    * gmx_restrict       nlist,
  59                      rvec                        * gmx_restrict          xx,
  60                      rvec                        * gmx_restrict          ff,
  61                      struct t_forcerec           * gmx_restrict          fr,
  62                      t_mdatoms                   * gmx_restrict     mdatoms,
  63                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
  64                      t_nrnb                      * gmx_restrict        nrnb)
  65 {
  66     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
  67      * just 0 for non-waters.
  68      * Suffixes A,B,C,D,E,F,G,H refer to j loop unrolling done with AVX, e.g. for the eight different
  69      * jnr indices corresponding to data put in the four positions in the SIMD register.
  70      */
  71     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
  72     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
  73     int              jnrA,jnrB,jnrC,jnrD;
  74     int              jnrE,jnrF,jnrG,jnrH;
  75     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
  76     int              jnrlistE,jnrlistF,jnrlistG,jnrlistH;
  77     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
  78     int              j_coord_offsetE,j_coord_offsetF,j_coord_offsetG,j_coord_offsetH;
  79     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
  80     real             rcutoff_scalar;
  81     real             *shiftvec,*fshift,*x,*f;
  82     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD,*fjptrE,*fjptrF,*fjptrG,*fjptrH;
  83     real             scratch[4*DIM];
  84     __m256           tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
  85     real *           vdwioffsetptr0;
  86     __m256           ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
  87     real *           vdwioffsetptr1;
  88     __m256           ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
  89     real *           vdwioffsetptr2;
  90     __m256           ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
  91     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D,vdwjidx0E,vdwjidx0F,vdwjidx0G,vdwjidx0H;
  92     __m256           jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
  93     __m256           dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
  94     __m256           dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
  95     __m256           dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
  96     __m256           velec,felec,velecsum,facel,crf,krf,krf2;
  97     real             *charge;
  98     int              nvdwtype;
  99     __m256           rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
 100     int              *vdwtype;
 101     real             *vdwparam;
 102     __m256           one_sixth   = _mm256_set1_ps(1.0/6.0);
 103     __m256           one_twelfth = _mm256_set1_ps(1.0/12.0);
 104     __m256i          vfitab;
 105     __m128i          vfitab_lo,vfitab_hi;
 106     __m128i          ifour       = _mm_set1_epi32(4);
 107     __m256           rt,vfeps,vftabscale,Y,F,G,H,Heps,Fp,VV,FF;
 108     real             *vftab;
 109     __m256           dummy_mask,cutoff_mask;
 110     __m256           signbit = _mm256_castsi256_ps( _mm256_set1_epi32(0x80000000) );
 111     __m256           one     = _mm256_set1_ps(1.0);
 112     __m256           two     = _mm256_set1_ps(2.0);
 113     x                = xx[0];
 114     f                = ff[0];
 115
 116     nri              = nlist->nri;
 117     iinr             = nlist->iinr;
 118     jindex           = nlist->jindex;
 119     jjnr             = nlist->jjnr;
 120     shiftidx         = nlist->shift;
 121     gid              = nlist->gid;
 122     shiftvec         = fr->shift_vec[0];
 123     fshift           = fr->fshift[0];
 124     facel            = _mm256_set1_ps(fr->ic->epsfac);
 125     charge           = mdatoms->chargeA;
 126     nvdwtype         = fr->ntype;
 127     vdwparam         = fr->nbfp;
 128     vdwtype          = mdatoms->typeA;
 129
 130     vftab            = kernel_data->table_elec->data;
 131     vftabscale       = _mm256_set1_ps(kernel_data->table_elec->scale);
 132
 133     /* Setup water-specific parameters */
 134     inr              = nlist->iinr[0];
 135     iq0              = _mm256_mul_ps(facel,_mm256_set1_ps(charge[inr+0]));
 136     iq1              = _mm256_mul_ps(facel,_mm256_set1_ps(charge[inr+1]));
 137     iq2              = _mm256_mul_ps(facel,_mm256_set1_ps(charge[inr+2]));
 138     vdwioffsetptr0   = vdwparam+2*nvdwtype*vdwtype[inr+0];
 139
 140     /* Avoid stupid compiler warnings */
 141     jnrA = jnrB = jnrC = jnrD = jnrE = jnrF = jnrG = jnrH = 0;
 142     j_coord_offsetA = 0;
 143     j_coord_offsetB = 0;
 144     j_coord_offsetC = 0;
 145     j_coord_offsetD = 0;
 146     j_coord_offsetE = 0;
 147     j_coord_offsetF = 0;
 148     j_coord_offsetG = 0;
 149     j_coord_offsetH = 0;
 150
 151     outeriter        = 0;
 152     inneriter        = 0;
 153
 154     for(iidx=0;iidx<4*DIM;iidx++)
 155     {
 156         scratch[iidx] = 0.0;
 157     }
 158
 159     /* Start outer loop over neighborlists */
 160     for(iidx=0; iidx<nri; iidx++)
 161     {
 162         /* Load shift vector for this list */
 163         i_shift_offset   = DIM*shiftidx[iidx];
 164
 165         /* Load limits for loop over neighbors */
 166         j_index_start    = jindex[iidx];
 167         j_index_end      = jindex[iidx+1];
 168
 169         /* Get outer coordinate index */
 170         inr              = iinr[iidx];
 171         i_coord_offset   = DIM*inr;
 172
 173         /* Load i particle coords and add shift vector */
 174         gmx_mm256_load_shift_and_3rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,
 175                                                     &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2);
 176
 177         fix0             = _mm256_setzero_ps();
 178         fiy0             = _mm256_setzero_ps();
 179         fiz0             = _mm256_setzero_ps();
 180         fix1             = _mm256_setzero_ps();
 181         fiy1             = _mm256_setzero_ps();
 182         fiz1             = _mm256_setzero_ps();
 183         fix2             = _mm256_setzero_ps();
 184         fiy2             = _mm256_setzero_ps();
 185         fiz2             = _mm256_setzero_ps();
 186
 187         /* Reset potential sums */
 188         velecsum         = _mm256_setzero_ps();
 189         vvdwsum          = _mm256_setzero_ps();
 190
 191         /* Start inner kernel loop */
 192         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+7]>=0; jidx+=8)
 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             jnrE             = jjnr[jidx+4];
 201             jnrF             = jjnr[jidx+5];
 202             jnrG             = jjnr[jidx+6];
 203             jnrH             = jjnr[jidx+7];
 204             j_coord_offsetA  = DIM*jnrA;
 205             j_coord_offsetB  = DIM*jnrB;
 206             j_coord_offsetC  = DIM*jnrC;
 207             j_coord_offsetD  = DIM*jnrD;
 208             j_coord_offsetE  = DIM*jnrE;
 209             j_coord_offsetF  = DIM*jnrF;
 210             j_coord_offsetG  = DIM*jnrG;
 211             j_coord_offsetH  = DIM*jnrH;
 212
 213             /* load j atom coordinates */
 214             gmx_mm256_load_1rvec_8ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 215                                                  x+j_coord_offsetC,x+j_coord_offsetD,
 216                                                  x+j_coord_offsetE,x+j_coord_offsetF,
 217                                                  x+j_coord_offsetG,x+j_coord_offsetH,
 218                                                  &jx0,&jy0,&jz0);
 219
 220             /* Calculate displacement vector */
 221             dx00             = _mm256_sub_ps(ix0,jx0);
 222             dy00             = _mm256_sub_ps(iy0,jy0);
 223             dz00             = _mm256_sub_ps(iz0,jz0);
 224             dx10             = _mm256_sub_ps(ix1,jx0);
 225             dy10             = _mm256_sub_ps(iy1,jy0);
 226             dz10             = _mm256_sub_ps(iz1,jz0);
 227             dx20             = _mm256_sub_ps(ix2,jx0);
 228             dy20             = _mm256_sub_ps(iy2,jy0);
 229             dz20             = _mm256_sub_ps(iz2,jz0);
 230
 231             /* Calculate squared distance and things based on it */
 232             rsq00            = gmx_mm256_calc_rsq_ps(dx00,dy00,dz00);
 233             rsq10            = gmx_mm256_calc_rsq_ps(dx10,dy10,dz10);
 234             rsq20            = gmx_mm256_calc_rsq_ps(dx20,dy20,dz20);
 235
 236             rinv00           = avx256_invsqrt_f(rsq00);
 237             rinv10           = avx256_invsqrt_f(rsq10);
 238             rinv20           = avx256_invsqrt_f(rsq20);
 239
 240             rinvsq00         = _mm256_mul_ps(rinv00,rinv00);
 241
 242             /* Load parameters for j particles */
 243             jq0              = gmx_mm256_load_8real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 244                                                                  charge+jnrC+0,charge+jnrD+0,
 245                                                                  charge+jnrE+0,charge+jnrF+0,
 246                                                                  charge+jnrG+0,charge+jnrH+0);
 247             vdwjidx0A        = 2*vdwtype[jnrA+0];
 248             vdwjidx0B        = 2*vdwtype[jnrB+0];
 249             vdwjidx0C        = 2*vdwtype[jnrC+0];
 250             vdwjidx0D        = 2*vdwtype[jnrD+0];
 251             vdwjidx0E        = 2*vdwtype[jnrE+0];
 252             vdwjidx0F        = 2*vdwtype[jnrF+0];
 253             vdwjidx0G        = 2*vdwtype[jnrG+0];
 254             vdwjidx0H        = 2*vdwtype[jnrH+0];
 255
 256             fjx0             = _mm256_setzero_ps();
 257             fjy0             = _mm256_setzero_ps();
 258             fjz0             = _mm256_setzero_ps();
 259
 260             /**************************
 261              * CALCULATE INTERACTIONS *
 262              **************************/
 263
 264             r00              = _mm256_mul_ps(rsq00,rinv00);
 265
 266             /* Compute parameters for interactions between i and j atoms */
 267             qq00             = _mm256_mul_ps(iq0,jq0);
 268             gmx_mm256_load_8pair_swizzle_ps(vdwioffsetptr0+vdwjidx0A,
 269                                             vdwioffsetptr0+vdwjidx0B,
 270                                             vdwioffsetptr0+vdwjidx0C,
 271                                             vdwioffsetptr0+vdwjidx0D,
 272                                             vdwioffsetptr0+vdwjidx0E,
 273                                             vdwioffsetptr0+vdwjidx0F,
 274                                             vdwioffsetptr0+vdwjidx0G,
 275                                             vdwioffsetptr0+vdwjidx0H,
 276                                             &c6_00,&c12_00);
 277
 278             /* Calculate table index by multiplying r with table scale and truncate to integer */
 279             rt               = _mm256_mul_ps(r00,vftabscale);
 280             vfitab           = _mm256_cvttps_epi32(rt);
 281             vfeps            = _mm256_sub_ps(rt,_mm256_round_ps(rt, _MM_FROUND_FLOOR));
 282             /*         AVX1 does not support 256-bit integer operations, so now we go to 128-bit mode... */
 283             vfitab_lo        = _mm256_extractf128_si256(vfitab,0x0);
 284             vfitab_hi        = _mm256_extractf128_si256(vfitab,0x1);
 285             vfitab_lo        = _mm_slli_epi32(vfitab_lo,2);
 286             vfitab_hi        = _mm_slli_epi32(vfitab_hi,2);
 287
 288             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 289             Y                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,0)),
 290                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,0)));
 291             F                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,1)),
 292                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,1)));
 293             G                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,2)),
 294                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,2)));
 295             H                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,3)),
 296                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,3)));
 297             GMX_MM256_HALFTRANSPOSE4_PS(Y,F,G,H);
 298             Heps             = _mm256_mul_ps(vfeps,H);
 299             Fp               = _mm256_add_ps(F,_mm256_mul_ps(vfeps,_mm256_add_ps(G,Heps)));
 300             VV               = _mm256_add_ps(Y,_mm256_mul_ps(vfeps,Fp));
 301             velec            = _mm256_mul_ps(qq00,VV);
 302             FF               = _mm256_add_ps(Fp,_mm256_mul_ps(vfeps,_mm256_add_ps(G,_mm256_add_ps(Heps,Heps))));
 303             felec            = _mm256_xor_ps(signbit,_mm256_mul_ps(_mm256_mul_ps(qq00,FF),_mm256_mul_ps(vftabscale,rinv00)));
 304
 305             /* LENNARD-JONES DISPERSION/REPULSION */
 306
 307             rinvsix          = _mm256_mul_ps(_mm256_mul_ps(rinvsq00,rinvsq00),rinvsq00);
 308             vvdw6            = _mm256_mul_ps(c6_00,rinvsix);
 309             vvdw12           = _mm256_mul_ps(c12_00,_mm256_mul_ps(rinvsix,rinvsix));
 310             vvdw             = _mm256_sub_ps( _mm256_mul_ps(vvdw12,one_twelfth) , _mm256_mul_ps(vvdw6,one_sixth) );
 311             fvdw             = _mm256_mul_ps(_mm256_sub_ps(vvdw12,vvdw6),rinvsq00);
 312
 313             /* Update potential sum for this i atom from the interaction with this j atom. */
 314             velecsum         = _mm256_add_ps(velecsum,velec);
 315             vvdwsum          = _mm256_add_ps(vvdwsum,vvdw);
 316
 317             fscal            = _mm256_add_ps(felec,fvdw);
 318
 319             /* Calculate temporary vectorial force */
 320             tx               = _mm256_mul_ps(fscal,dx00);
 321             ty               = _mm256_mul_ps(fscal,dy00);
 322             tz               = _mm256_mul_ps(fscal,dz00);
 323
 324             /* Update vectorial force */
 325             fix0             = _mm256_add_ps(fix0,tx);
 326             fiy0             = _mm256_add_ps(fiy0,ty);
 327             fiz0             = _mm256_add_ps(fiz0,tz);
 328
 329             fjx0             = _mm256_add_ps(fjx0,tx);
 330             fjy0             = _mm256_add_ps(fjy0,ty);
 331             fjz0             = _mm256_add_ps(fjz0,tz);
 332
 333             /**************************
 334              * CALCULATE INTERACTIONS *
 335              **************************/
 336
 337             r10              = _mm256_mul_ps(rsq10,rinv10);
 338
 339             /* Compute parameters for interactions between i and j atoms */
 340             qq10             = _mm256_mul_ps(iq1,jq0);
 341
 342             /* Calculate table index by multiplying r with table scale and truncate to integer */
 343             rt               = _mm256_mul_ps(r10,vftabscale);
 344             vfitab           = _mm256_cvttps_epi32(rt);
 345             vfeps            = _mm256_sub_ps(rt,_mm256_round_ps(rt, _MM_FROUND_FLOOR));
 346             /*         AVX1 does not support 256-bit integer operations, so now we go to 128-bit mode... */
 347             vfitab_lo        = _mm256_extractf128_si256(vfitab,0x0);
 348             vfitab_hi        = _mm256_extractf128_si256(vfitab,0x1);
 349             vfitab_lo        = _mm_slli_epi32(vfitab_lo,2);
 350             vfitab_hi        = _mm_slli_epi32(vfitab_hi,2);
 351
 352             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 353             Y                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,0)),
 354                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,0)));
 355             F                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,1)),
 356                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,1)));
 357             G                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,2)),
 358                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,2)));
 359             H                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,3)),
 360                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,3)));
 361             GMX_MM256_HALFTRANSPOSE4_PS(Y,F,G,H);
 362             Heps             = _mm256_mul_ps(vfeps,H);
 363             Fp               = _mm256_add_ps(F,_mm256_mul_ps(vfeps,_mm256_add_ps(G,Heps)));
 364             VV               = _mm256_add_ps(Y,_mm256_mul_ps(vfeps,Fp));
 365             velec            = _mm256_mul_ps(qq10,VV);
 366             FF               = _mm256_add_ps(Fp,_mm256_mul_ps(vfeps,_mm256_add_ps(G,_mm256_add_ps(Heps,Heps))));
 367             felec            = _mm256_xor_ps(signbit,_mm256_mul_ps(_mm256_mul_ps(qq10,FF),_mm256_mul_ps(vftabscale,rinv10)));
 368
 369             /* Update potential sum for this i atom from the interaction with this j atom. */
 370             velecsum         = _mm256_add_ps(velecsum,velec);
 371
 372             fscal            = felec;
 373
 374             /* Calculate temporary vectorial force */
 375             tx               = _mm256_mul_ps(fscal,dx10);
 376             ty               = _mm256_mul_ps(fscal,dy10);
 377             tz               = _mm256_mul_ps(fscal,dz10);
 378
 379             /* Update vectorial force */
 380             fix1             = _mm256_add_ps(fix1,tx);
 381             fiy1             = _mm256_add_ps(fiy1,ty);
 382             fiz1             = _mm256_add_ps(fiz1,tz);
 383
 384             fjx0             = _mm256_add_ps(fjx0,tx);
 385             fjy0             = _mm256_add_ps(fjy0,ty);
 386             fjz0             = _mm256_add_ps(fjz0,tz);
 387
 388             /**************************
 389              * CALCULATE INTERACTIONS *
 390              **************************/
 391
 392             r20              = _mm256_mul_ps(rsq20,rinv20);
 393
 394             /* Compute parameters for interactions between i and j atoms */
 395             qq20             = _mm256_mul_ps(iq2,jq0);
 396
 397             /* Calculate table index by multiplying r with table scale and truncate to integer */
 398             rt               = _mm256_mul_ps(r20,vftabscale);
 399             vfitab           = _mm256_cvttps_epi32(rt);
 400             vfeps            = _mm256_sub_ps(rt,_mm256_round_ps(rt, _MM_FROUND_FLOOR));
 401             /*         AVX1 does not support 256-bit integer operations, so now we go to 128-bit mode... */
 402             vfitab_lo        = _mm256_extractf128_si256(vfitab,0x0);
 403             vfitab_hi        = _mm256_extractf128_si256(vfitab,0x1);
 404             vfitab_lo        = _mm_slli_epi32(vfitab_lo,2);
 405             vfitab_hi        = _mm_slli_epi32(vfitab_hi,2);
 406
 407             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 408             Y                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,0)),
 409                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,0)));
 410             F                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,1)),
 411                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,1)));
 412             G                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,2)),
 413                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,2)));
 414             H                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,3)),
 415                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,3)));
 416             GMX_MM256_HALFTRANSPOSE4_PS(Y,F,G,H);
 417             Heps             = _mm256_mul_ps(vfeps,H);
 418             Fp               = _mm256_add_ps(F,_mm256_mul_ps(vfeps,_mm256_add_ps(G,Heps)));
 419             VV               = _mm256_add_ps(Y,_mm256_mul_ps(vfeps,Fp));
 420             velec            = _mm256_mul_ps(qq20,VV);
 421             FF               = _mm256_add_ps(Fp,_mm256_mul_ps(vfeps,_mm256_add_ps(G,_mm256_add_ps(Heps,Heps))));
 422             felec            = _mm256_xor_ps(signbit,_mm256_mul_ps(_mm256_mul_ps(qq20,FF),_mm256_mul_ps(vftabscale,rinv20)));
 423
 424             /* Update potential sum for this i atom from the interaction with this j atom. */
 425             velecsum         = _mm256_add_ps(velecsum,velec);
 426
 427             fscal            = felec;
 428
 429             /* Calculate temporary vectorial force */
 430             tx               = _mm256_mul_ps(fscal,dx20);
 431             ty               = _mm256_mul_ps(fscal,dy20);
 432             tz               = _mm256_mul_ps(fscal,dz20);
 433
 434             /* Update vectorial force */
 435             fix2             = _mm256_add_ps(fix2,tx);
 436             fiy2             = _mm256_add_ps(fiy2,ty);
 437             fiz2             = _mm256_add_ps(fiz2,tz);
 438
 439             fjx0             = _mm256_add_ps(fjx0,tx);
 440             fjy0             = _mm256_add_ps(fjy0,ty);
 441             fjz0             = _mm256_add_ps(fjz0,tz);
 442
 443             fjptrA             = f+j_coord_offsetA;
 444             fjptrB             = f+j_coord_offsetB;
 445             fjptrC             = f+j_coord_offsetC;
 446             fjptrD             = f+j_coord_offsetD;
 447             fjptrE             = f+j_coord_offsetE;
 448             fjptrF             = f+j_coord_offsetF;
 449             fjptrG             = f+j_coord_offsetG;
 450             fjptrH             = f+j_coord_offsetH;
 451
 452             gmx_mm256_decrement_1rvec_8ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjptrE,fjptrF,fjptrG,fjptrH,fjx0,fjy0,fjz0);
 453
 454             /* Inner loop uses 145 flops */
 455         }
 456
 457         if(jidx<j_index_end)
 458         {
 459
 460             /* Get j neighbor index, and coordinate index */
 461             jnrlistA         = jjnr[jidx];
 462             jnrlistB         = jjnr[jidx+1];
 463             jnrlistC         = jjnr[jidx+2];
 464             jnrlistD         = jjnr[jidx+3];
 465             jnrlistE         = jjnr[jidx+4];
 466             jnrlistF         = jjnr[jidx+5];
 467             jnrlistG         = jjnr[jidx+6];
 468             jnrlistH         = jjnr[jidx+7];
 469             /* Sign of each element will be negative for non-real atoms.
 470              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 471              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
 472              */
 473             dummy_mask = gmx_mm256_set_m128(gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx+4)),_mm_setzero_si128())),
 474                                             gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128())));
 475
 476             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
 477             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
 478             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
 479             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
 480             jnrE       = (jnrlistE>=0) ? jnrlistE : 0;
 481             jnrF       = (jnrlistF>=0) ? jnrlistF : 0;
 482             jnrG       = (jnrlistG>=0) ? jnrlistG : 0;
 483             jnrH       = (jnrlistH>=0) ? jnrlistH : 0;
 484             j_coord_offsetA  = DIM*jnrA;
 485             j_coord_offsetB  = DIM*jnrB;
 486             j_coord_offsetC  = DIM*jnrC;
 487             j_coord_offsetD  = DIM*jnrD;
 488             j_coord_offsetE  = DIM*jnrE;
 489             j_coord_offsetF  = DIM*jnrF;
 490             j_coord_offsetG  = DIM*jnrG;
 491             j_coord_offsetH  = DIM*jnrH;
 492
 493             /* load j atom coordinates */
 494             gmx_mm256_load_1rvec_8ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 495                                                  x+j_coord_offsetC,x+j_coord_offsetD,
 496                                                  x+j_coord_offsetE,x+j_coord_offsetF,
 497                                                  x+j_coord_offsetG,x+j_coord_offsetH,
 498                                                  &jx0,&jy0,&jz0);
 499
 500             /* Calculate displacement vector */
 501             dx00             = _mm256_sub_ps(ix0,jx0);
 502             dy00             = _mm256_sub_ps(iy0,jy0);
 503             dz00             = _mm256_sub_ps(iz0,jz0);
 504             dx10             = _mm256_sub_ps(ix1,jx0);
 505             dy10             = _mm256_sub_ps(iy1,jy0);
 506             dz10             = _mm256_sub_ps(iz1,jz0);
 507             dx20             = _mm256_sub_ps(ix2,jx0);
 508             dy20             = _mm256_sub_ps(iy2,jy0);
 509             dz20             = _mm256_sub_ps(iz2,jz0);
 510
 511             /* Calculate squared distance and things based on it */
 512             rsq00            = gmx_mm256_calc_rsq_ps(dx00,dy00,dz00);
 513             rsq10            = gmx_mm256_calc_rsq_ps(dx10,dy10,dz10);
 514             rsq20            = gmx_mm256_calc_rsq_ps(dx20,dy20,dz20);
 515
 516             rinv00           = avx256_invsqrt_f(rsq00);
 517             rinv10           = avx256_invsqrt_f(rsq10);
 518             rinv20           = avx256_invsqrt_f(rsq20);
 519
 520             rinvsq00         = _mm256_mul_ps(rinv00,rinv00);
 521
 522             /* Load parameters for j particles */
 523             jq0              = gmx_mm256_load_8real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 524                                                                  charge+jnrC+0,charge+jnrD+0,
 525                                                                  charge+jnrE+0,charge+jnrF+0,
 526                                                                  charge+jnrG+0,charge+jnrH+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             vdwjidx0E        = 2*vdwtype[jnrE+0];
 532             vdwjidx0F        = 2*vdwtype[jnrF+0];
 533             vdwjidx0G        = 2*vdwtype[jnrG+0];
 534             vdwjidx0H        = 2*vdwtype[jnrH+0];
 535
 536             fjx0             = _mm256_setzero_ps();
 537             fjy0             = _mm256_setzero_ps();
 538             fjz0             = _mm256_setzero_ps();
 539
 540             /**************************
 541              * CALCULATE INTERACTIONS *
 542              **************************/
 543
 544             r00              = _mm256_mul_ps(rsq00,rinv00);
 545             r00              = _mm256_andnot_ps(dummy_mask,r00);
 546
 547             /* Compute parameters for interactions between i and j atoms */
 548             qq00             = _mm256_mul_ps(iq0,jq0);
 549             gmx_mm256_load_8pair_swizzle_ps(vdwioffsetptr0+vdwjidx0A,
 550                                             vdwioffsetptr0+vdwjidx0B,
 551                                             vdwioffsetptr0+vdwjidx0C,
 552                                             vdwioffsetptr0+vdwjidx0D,
 553                                             vdwioffsetptr0+vdwjidx0E,
 554                                             vdwioffsetptr0+vdwjidx0F,
 555                                             vdwioffsetptr0+vdwjidx0G,
 556                                             vdwioffsetptr0+vdwjidx0H,
 557                                             &c6_00,&c12_00);
 558
 559             /* Calculate table index by multiplying r with table scale and truncate to integer */
 560             rt               = _mm256_mul_ps(r00,vftabscale);
 561             vfitab           = _mm256_cvttps_epi32(rt);
 562             vfeps            = _mm256_sub_ps(rt,_mm256_round_ps(rt, _MM_FROUND_FLOOR));
 563             /*         AVX1 does not support 256-bit integer operations, so now we go to 128-bit mode... */
 564             vfitab_lo        = _mm256_extractf128_si256(vfitab,0x0);
 565             vfitab_hi        = _mm256_extractf128_si256(vfitab,0x1);
 566             vfitab_lo        = _mm_slli_epi32(vfitab_lo,2);
 567             vfitab_hi        = _mm_slli_epi32(vfitab_hi,2);
 568
 569             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 570             Y                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,0)),
 571                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,0)));
 572             F                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,1)),
 573                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,1)));
 574             G                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,2)),
 575                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,2)));
 576             H                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,3)),
 577                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,3)));
 578             GMX_MM256_HALFTRANSPOSE4_PS(Y,F,G,H);
 579             Heps             = _mm256_mul_ps(vfeps,H);
 580             Fp               = _mm256_add_ps(F,_mm256_mul_ps(vfeps,_mm256_add_ps(G,Heps)));
 581             VV               = _mm256_add_ps(Y,_mm256_mul_ps(vfeps,Fp));
 582             velec            = _mm256_mul_ps(qq00,VV);
 583             FF               = _mm256_add_ps(Fp,_mm256_mul_ps(vfeps,_mm256_add_ps(G,_mm256_add_ps(Heps,Heps))));
 584             felec            = _mm256_xor_ps(signbit,_mm256_mul_ps(_mm256_mul_ps(qq00,FF),_mm256_mul_ps(vftabscale,rinv00)));
 585
 586             /* LENNARD-JONES DISPERSION/REPULSION */
 587
 588             rinvsix          = _mm256_mul_ps(_mm256_mul_ps(rinvsq00,rinvsq00),rinvsq00);
 589             vvdw6            = _mm256_mul_ps(c6_00,rinvsix);
 590             vvdw12           = _mm256_mul_ps(c12_00,_mm256_mul_ps(rinvsix,rinvsix));
 591             vvdw             = _mm256_sub_ps( _mm256_mul_ps(vvdw12,one_twelfth) , _mm256_mul_ps(vvdw6,one_sixth) );
 592             fvdw             = _mm256_mul_ps(_mm256_sub_ps(vvdw12,vvdw6),rinvsq00);
 593
 594             /* Update potential sum for this i atom from the interaction with this j atom. */
 595             velec            = _mm256_andnot_ps(dummy_mask,velec);
 596             velecsum         = _mm256_add_ps(velecsum,velec);
 597             vvdw             = _mm256_andnot_ps(dummy_mask,vvdw);
 598             vvdwsum          = _mm256_add_ps(vvdwsum,vvdw);
 599
 600             fscal            = _mm256_add_ps(felec,fvdw);
 601
 602             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
 603
 604             /* Calculate temporary vectorial force */
 605             tx               = _mm256_mul_ps(fscal,dx00);
 606             ty               = _mm256_mul_ps(fscal,dy00);
 607             tz               = _mm256_mul_ps(fscal,dz00);
 608
 609             /* Update vectorial force */
 610             fix0             = _mm256_add_ps(fix0,tx);
 611             fiy0             = _mm256_add_ps(fiy0,ty);
 612             fiz0             = _mm256_add_ps(fiz0,tz);
 613
 614             fjx0             = _mm256_add_ps(fjx0,tx);
 615             fjy0             = _mm256_add_ps(fjy0,ty);
 616             fjz0             = _mm256_add_ps(fjz0,tz);
 617
 618             /**************************
 619              * CALCULATE INTERACTIONS *
 620              **************************/
 621
 622             r10              = _mm256_mul_ps(rsq10,rinv10);
 623             r10              = _mm256_andnot_ps(dummy_mask,r10);
 624
 625             /* Compute parameters for interactions between i and j atoms */
 626             qq10             = _mm256_mul_ps(iq1,jq0);
 627
 628             /* Calculate table index by multiplying r with table scale and truncate to integer */
 629             rt               = _mm256_mul_ps(r10,vftabscale);
 630             vfitab           = _mm256_cvttps_epi32(rt);
 631             vfeps            = _mm256_sub_ps(rt,_mm256_round_ps(rt, _MM_FROUND_FLOOR));
 632             /*         AVX1 does not support 256-bit integer operations, so now we go to 128-bit mode... */
 633             vfitab_lo        = _mm256_extractf128_si256(vfitab,0x0);
 634             vfitab_hi        = _mm256_extractf128_si256(vfitab,0x1);
 635             vfitab_lo        = _mm_slli_epi32(vfitab_lo,2);
 636             vfitab_hi        = _mm_slli_epi32(vfitab_hi,2);
 637
 638             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 639             Y                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,0)),
 640                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,0)));
 641             F                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,1)),
 642                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,1)));
 643             G                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,2)),
 644                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,2)));
 645             H                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,3)),
 646                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,3)));
 647             GMX_MM256_HALFTRANSPOSE4_PS(Y,F,G,H);
 648             Heps             = _mm256_mul_ps(vfeps,H);
 649             Fp               = _mm256_add_ps(F,_mm256_mul_ps(vfeps,_mm256_add_ps(G,Heps)));
 650             VV               = _mm256_add_ps(Y,_mm256_mul_ps(vfeps,Fp));
 651             velec            = _mm256_mul_ps(qq10,VV);
 652             FF               = _mm256_add_ps(Fp,_mm256_mul_ps(vfeps,_mm256_add_ps(G,_mm256_add_ps(Heps,Heps))));
 653             felec            = _mm256_xor_ps(signbit,_mm256_mul_ps(_mm256_mul_ps(qq10,FF),_mm256_mul_ps(vftabscale,rinv10)));
 654
 655             /* Update potential sum for this i atom from the interaction with this j atom. */
 656             velec            = _mm256_andnot_ps(dummy_mask,velec);
 657             velecsum         = _mm256_add_ps(velecsum,velec);
 658
 659             fscal            = felec;
 660
 661             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
 662
 663             /* Calculate temporary vectorial force */
 664             tx               = _mm256_mul_ps(fscal,dx10);
 665             ty               = _mm256_mul_ps(fscal,dy10);
 666             tz               = _mm256_mul_ps(fscal,dz10);
 667
 668             /* Update vectorial force */
 669             fix1             = _mm256_add_ps(fix1,tx);
 670             fiy1             = _mm256_add_ps(fiy1,ty);
 671             fiz1             = _mm256_add_ps(fiz1,tz);
 672
 673             fjx0             = _mm256_add_ps(fjx0,tx);
 674             fjy0             = _mm256_add_ps(fjy0,ty);
 675             fjz0             = _mm256_add_ps(fjz0,tz);
 676
 677             /**************************
 678              * CALCULATE INTERACTIONS *
 679              **************************/
 680
 681             r20              = _mm256_mul_ps(rsq20,rinv20);
 682             r20              = _mm256_andnot_ps(dummy_mask,r20);
 683
 684             /* Compute parameters for interactions between i and j atoms */
 685             qq20             = _mm256_mul_ps(iq2,jq0);
 686
 687             /* Calculate table index by multiplying r with table scale and truncate to integer */
 688             rt               = _mm256_mul_ps(r20,vftabscale);
 689             vfitab           = _mm256_cvttps_epi32(rt);
 690             vfeps            = _mm256_sub_ps(rt,_mm256_round_ps(rt, _MM_FROUND_FLOOR));
 691             /*         AVX1 does not support 256-bit integer operations, so now we go to 128-bit mode... */
 692             vfitab_lo        = _mm256_extractf128_si256(vfitab,0x0);
 693             vfitab_hi        = _mm256_extractf128_si256(vfitab,0x1);
 694             vfitab_lo        = _mm_slli_epi32(vfitab_lo,2);
 695             vfitab_hi        = _mm_slli_epi32(vfitab_hi,2);
 696
 697             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 698             Y                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,0)),
 699                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,0)));
 700             F                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,1)),
 701                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,1)));
 702             G                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,2)),
 703                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,2)));
 704             H                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,3)),
 705                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,3)));
 706             GMX_MM256_HALFTRANSPOSE4_PS(Y,F,G,H);
 707             Heps             = _mm256_mul_ps(vfeps,H);
 708             Fp               = _mm256_add_ps(F,_mm256_mul_ps(vfeps,_mm256_add_ps(G,Heps)));
 709             VV               = _mm256_add_ps(Y,_mm256_mul_ps(vfeps,Fp));
 710             velec            = _mm256_mul_ps(qq20,VV);
 711             FF               = _mm256_add_ps(Fp,_mm256_mul_ps(vfeps,_mm256_add_ps(G,_mm256_add_ps(Heps,Heps))));
 712             felec            = _mm256_xor_ps(signbit,_mm256_mul_ps(_mm256_mul_ps(qq20,FF),_mm256_mul_ps(vftabscale,rinv20)));
 713
 714             /* Update potential sum for this i atom from the interaction with this j atom. */
 715             velec            = _mm256_andnot_ps(dummy_mask,velec);
 716             velecsum         = _mm256_add_ps(velecsum,velec);
 717
 718             fscal            = felec;
 719
 720             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
 721
 722             /* Calculate temporary vectorial force */
 723             tx               = _mm256_mul_ps(fscal,dx20);
 724             ty               = _mm256_mul_ps(fscal,dy20);
 725             tz               = _mm256_mul_ps(fscal,dz20);
 726
 727             /* Update vectorial force */
 728             fix2             = _mm256_add_ps(fix2,tx);
 729             fiy2             = _mm256_add_ps(fiy2,ty);
 730             fiz2             = _mm256_add_ps(fiz2,tz);
 731
 732             fjx0             = _mm256_add_ps(fjx0,tx);
 733             fjy0             = _mm256_add_ps(fjy0,ty);
 734             fjz0             = _mm256_add_ps(fjz0,tz);
 735
 736             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
 737             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
 738             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
 739             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
 740             fjptrE             = (jnrlistE>=0) ? f+j_coord_offsetE : scratch;
 741             fjptrF             = (jnrlistF>=0) ? f+j_coord_offsetF : scratch;
 742             fjptrG             = (jnrlistG>=0) ? f+j_coord_offsetG : scratch;
 743             fjptrH             = (jnrlistH>=0) ? f+j_coord_offsetH : scratch;
 744
 745             gmx_mm256_decrement_1rvec_8ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjptrE,fjptrF,fjptrG,fjptrH,fjx0,fjy0,fjz0);
 746
 747             /* Inner loop uses 148 flops */
 748         }
 749
 750         /* End of innermost loop */
 751
 752         gmx_mm256_update_iforce_3atom_swizzle_ps(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,
 753                                                  f+i_coord_offset,fshift+i_shift_offset);
 754
 755         ggid                        = gid[iidx];
 756         /* Update potential energies */
 757         gmx_mm256_update_1pot_ps(velecsum,kernel_data->energygrp_elec+ggid);
 758         gmx_mm256_update_1pot_ps(vvdwsum,kernel_data->energygrp_vdw+ggid);
 759
 760         /* Increment number of inner iterations */
 761         inneriter                  += j_index_end - j_index_start;
 762
 763         /* Outer loop uses 20 flops */
 764     }
 765
 766     /* Increment number of outer iterations */
 767     outeriter        += nri;
 768
 769     /* Update outer/inner flops */
 770
 771     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W3_VF,outeriter*20 + inneriter*148);
 772 }
 773 /*
 774  * Gromacs nonbonded kernel:   nb_kernel_ElecCSTab_VdwLJ_GeomW3P1_F_avx_256_single
 775  * Electrostatics interaction: CubicSplineTable
 776  * VdW interaction:            LennardJones
 777  * Geometry:                   Water3-Particle
 778  * Calculate force/pot:        Force
 779  */
 780 void
 781 nb_kernel_ElecCSTab_VdwLJ_GeomW3P1_F_avx_256_single
 782                     (t_nblist                    * gmx_restrict       nlist,
 783                      rvec                        * gmx_restrict          xx,
 784                      rvec                        * gmx_restrict          ff,
 785                      struct t_forcerec           * gmx_restrict          fr,
 786                      t_mdatoms                   * gmx_restrict     mdatoms,
 787                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
 788                      t_nrnb                      * gmx_restrict        nrnb)
 789 {
 790     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
 791      * just 0 for non-waters.
 792      * Suffixes A,B,C,D,E,F,G,H refer to j loop unrolling done with AVX, e.g. for the eight different
 793      * jnr indices corresponding to data put in the four positions in the SIMD register.
 794      */
 795     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
 796     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
 797     int              jnrA,jnrB,jnrC,jnrD;
 798     int              jnrE,jnrF,jnrG,jnrH;
 799     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
 800     int              jnrlistE,jnrlistF,jnrlistG,jnrlistH;
 801     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
 802     int              j_coord_offsetE,j_coord_offsetF,j_coord_offsetG,j_coord_offsetH;
 803     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
 804     real             rcutoff_scalar;
 805     real             *shiftvec,*fshift,*x,*f;
 806     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD,*fjptrE,*fjptrF,*fjptrG,*fjptrH;
 807     real             scratch[4*DIM];
 808     __m256           tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
 809     real *           vdwioffsetptr0;
 810     __m256           ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
 811     real *           vdwioffsetptr1;
 812     __m256           ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
 813     real *           vdwioffsetptr2;
 814     __m256           ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
 815     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D,vdwjidx0E,vdwjidx0F,vdwjidx0G,vdwjidx0H;
 816     __m256           jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
 817     __m256           dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
 818     __m256           dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
 819     __m256           dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
 820     __m256           velec,felec,velecsum,facel,crf,krf,krf2;
 821     real             *charge;
 822     int              nvdwtype;
 823     __m256           rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
 824     int              *vdwtype;
 825     real             *vdwparam;
 826     __m256           one_sixth   = _mm256_set1_ps(1.0/6.0);
 827     __m256           one_twelfth = _mm256_set1_ps(1.0/12.0);
 828     __m256i          vfitab;
 829     __m128i          vfitab_lo,vfitab_hi;
 830     __m128i          ifour       = _mm_set1_epi32(4);
 831     __m256           rt,vfeps,vftabscale,Y,F,G,H,Heps,Fp,VV,FF;
 832     real             *vftab;
 833     __m256           dummy_mask,cutoff_mask;
 834     __m256           signbit = _mm256_castsi256_ps( _mm256_set1_epi32(0x80000000) );
 835     __m256           one     = _mm256_set1_ps(1.0);
 836     __m256           two     = _mm256_set1_ps(2.0);
 837     x                = xx[0];
 838     f                = ff[0];
 839
 840     nri              = nlist->nri;
 841     iinr             = nlist->iinr;
 842     jindex           = nlist->jindex;
 843     jjnr             = nlist->jjnr;
 844     shiftidx         = nlist->shift;
 845     gid              = nlist->gid;
 846     shiftvec         = fr->shift_vec[0];
 847     fshift           = fr->fshift[0];
 848     facel            = _mm256_set1_ps(fr->ic->epsfac);
 849     charge           = mdatoms->chargeA;
 850     nvdwtype         = fr->ntype;
 851     vdwparam         = fr->nbfp;
 852     vdwtype          = mdatoms->typeA;
 853
 854     vftab            = kernel_data->table_elec->data;
 855     vftabscale       = _mm256_set1_ps(kernel_data->table_elec->scale);
 856
 857     /* Setup water-specific parameters */
 858     inr              = nlist->iinr[0];
 859     iq0              = _mm256_mul_ps(facel,_mm256_set1_ps(charge[inr+0]));
 860     iq1              = _mm256_mul_ps(facel,_mm256_set1_ps(charge[inr+1]));
 861     iq2              = _mm256_mul_ps(facel,_mm256_set1_ps(charge[inr+2]));
 862     vdwioffsetptr0   = vdwparam+2*nvdwtype*vdwtype[inr+0];
 863
 864     /* Avoid stupid compiler warnings */
 865     jnrA = jnrB = jnrC = jnrD = jnrE = jnrF = jnrG = jnrH = 0;
 866     j_coord_offsetA = 0;
 867     j_coord_offsetB = 0;
 868     j_coord_offsetC = 0;
 869     j_coord_offsetD = 0;
 870     j_coord_offsetE = 0;
 871     j_coord_offsetF = 0;
 872     j_coord_offsetG = 0;
 873     j_coord_offsetH = 0;
 874
 875     outeriter        = 0;
 876     inneriter        = 0;
 877
 878     for(iidx=0;iidx<4*DIM;iidx++)
 879     {
 880         scratch[iidx] = 0.0;
 881     }
 882
 883     /* Start outer loop over neighborlists */
 884     for(iidx=0; iidx<nri; iidx++)
 885     {
 886         /* Load shift vector for this list */
 887         i_shift_offset   = DIM*shiftidx[iidx];
 888
 889         /* Load limits for loop over neighbors */
 890         j_index_start    = jindex[iidx];
 891         j_index_end      = jindex[iidx+1];
 892
 893         /* Get outer coordinate index */
 894         inr              = iinr[iidx];
 895         i_coord_offset   = DIM*inr;
 896
 897         /* Load i particle coords and add shift vector */
 898         gmx_mm256_load_shift_and_3rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,
 899                                                     &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2);
 900
 901         fix0             = _mm256_setzero_ps();
 902         fiy0             = _mm256_setzero_ps();
 903         fiz0             = _mm256_setzero_ps();
 904         fix1             = _mm256_setzero_ps();
 905         fiy1             = _mm256_setzero_ps();
 906         fiz1             = _mm256_setzero_ps();
 907         fix2             = _mm256_setzero_ps();
 908         fiy2             = _mm256_setzero_ps();
 909         fiz2             = _mm256_setzero_ps();
 910
 911         /* Start inner kernel loop */
 912         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+7]>=0; jidx+=8)
 913         {
 914
 915             /* Get j neighbor index, and coordinate index */
 916             jnrA             = jjnr[jidx];
 917             jnrB             = jjnr[jidx+1];
 918             jnrC             = jjnr[jidx+2];
 919             jnrD             = jjnr[jidx+3];
 920             jnrE             = jjnr[jidx+4];
 921             jnrF             = jjnr[jidx+5];
 922             jnrG             = jjnr[jidx+6];
 923             jnrH             = jjnr[jidx+7];
 924             j_coord_offsetA  = DIM*jnrA;
 925             j_coord_offsetB  = DIM*jnrB;
 926             j_coord_offsetC  = DIM*jnrC;
 927             j_coord_offsetD  = DIM*jnrD;
 928             j_coord_offsetE  = DIM*jnrE;
 929             j_coord_offsetF  = DIM*jnrF;
 930             j_coord_offsetG  = DIM*jnrG;
 931             j_coord_offsetH  = DIM*jnrH;
 932
 933             /* load j atom coordinates */
 934             gmx_mm256_load_1rvec_8ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 935                                                  x+j_coord_offsetC,x+j_coord_offsetD,
 936                                                  x+j_coord_offsetE,x+j_coord_offsetF,
 937                                                  x+j_coord_offsetG,x+j_coord_offsetH,
 938                                                  &jx0,&jy0,&jz0);
 939
 940             /* Calculate displacement vector */
 941             dx00             = _mm256_sub_ps(ix0,jx0);
 942             dy00             = _mm256_sub_ps(iy0,jy0);
 943             dz00             = _mm256_sub_ps(iz0,jz0);
 944             dx10             = _mm256_sub_ps(ix1,jx0);
 945             dy10             = _mm256_sub_ps(iy1,jy0);
 946             dz10             = _mm256_sub_ps(iz1,jz0);
 947             dx20             = _mm256_sub_ps(ix2,jx0);
 948             dy20             = _mm256_sub_ps(iy2,jy0);
 949             dz20             = _mm256_sub_ps(iz2,jz0);
 950
 951             /* Calculate squared distance and things based on it */
 952             rsq00            = gmx_mm256_calc_rsq_ps(dx00,dy00,dz00);
 953             rsq10            = gmx_mm256_calc_rsq_ps(dx10,dy10,dz10);
 954             rsq20            = gmx_mm256_calc_rsq_ps(dx20,dy20,dz20);
 955
 956             rinv00           = avx256_invsqrt_f(rsq00);
 957             rinv10           = avx256_invsqrt_f(rsq10);
 958             rinv20           = avx256_invsqrt_f(rsq20);
 959
 960             rinvsq00         = _mm256_mul_ps(rinv00,rinv00);
 961
 962             /* Load parameters for j particles */
 963             jq0              = gmx_mm256_load_8real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 964                                                                  charge+jnrC+0,charge+jnrD+0,
 965                                                                  charge+jnrE+0,charge+jnrF+0,
 966                                                                  charge+jnrG+0,charge+jnrH+0);
 967             vdwjidx0A        = 2*vdwtype[jnrA+0];
 968             vdwjidx0B        = 2*vdwtype[jnrB+0];
 969             vdwjidx0C        = 2*vdwtype[jnrC+0];
 970             vdwjidx0D        = 2*vdwtype[jnrD+0];
 971             vdwjidx0E        = 2*vdwtype[jnrE+0];
 972             vdwjidx0F        = 2*vdwtype[jnrF+0];
 973             vdwjidx0G        = 2*vdwtype[jnrG+0];
 974             vdwjidx0H        = 2*vdwtype[jnrH+0];
 975
 976             fjx0             = _mm256_setzero_ps();
 977             fjy0             = _mm256_setzero_ps();
 978             fjz0             = _mm256_setzero_ps();
 979
 980             /**************************
 981              * CALCULATE INTERACTIONS *
 982              **************************/
 983
 984             r00              = _mm256_mul_ps(rsq00,rinv00);
 985
 986             /* Compute parameters for interactions between i and j atoms */
 987             qq00             = _mm256_mul_ps(iq0,jq0);
 988             gmx_mm256_load_8pair_swizzle_ps(vdwioffsetptr0+vdwjidx0A,
 989                                             vdwioffsetptr0+vdwjidx0B,
 990                                             vdwioffsetptr0+vdwjidx0C,
 991                                             vdwioffsetptr0+vdwjidx0D,
 992                                             vdwioffsetptr0+vdwjidx0E,
 993                                             vdwioffsetptr0+vdwjidx0F,
 994                                             vdwioffsetptr0+vdwjidx0G,
 995                                             vdwioffsetptr0+vdwjidx0H,
 996                                             &c6_00,&c12_00);
 997
 998             /* Calculate table index by multiplying r with table scale and truncate to integer */
 999             rt               = _mm256_mul_ps(r00,vftabscale);
1000             vfitab           = _mm256_cvttps_epi32(rt);
1001             vfeps            = _mm256_sub_ps(rt,_mm256_round_ps(rt, _MM_FROUND_FLOOR));
1002             /*         AVX1 does not support 256-bit integer operations, so now we go to 128-bit mode... */
1003             vfitab_lo        = _mm256_extractf128_si256(vfitab,0x0);
1004             vfitab_hi        = _mm256_extractf128_si256(vfitab,0x1);
1005             vfitab_lo        = _mm_slli_epi32(vfitab_lo,2);
1006             vfitab_hi        = _mm_slli_epi32(vfitab_hi,2);
1007
1008             /* CUBIC SPLINE TABLE ELECTROSTATICS */
1009             Y                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,0)),
1010                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,0)));
1011             F                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,1)),
1012                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,1)));
1013             G                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,2)),
1014                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,2)));
1015             H                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,3)),
1016                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,3)));
1017             GMX_MM256_HALFTRANSPOSE4_PS(Y,F,G,H);
1018             Heps             = _mm256_mul_ps(vfeps,H);
1019             Fp               = _mm256_add_ps(F,_mm256_mul_ps(vfeps,_mm256_add_ps(G,Heps)));
1020             FF               = _mm256_add_ps(Fp,_mm256_mul_ps(vfeps,_mm256_add_ps(G,_mm256_add_ps(Heps,Heps))));
1021             felec            = _mm256_xor_ps(signbit,_mm256_mul_ps(_mm256_mul_ps(qq00,FF),_mm256_mul_ps(vftabscale,rinv00)));
1022
1023             /* LENNARD-JONES DISPERSION/REPULSION */
1024
1025             rinvsix          = _mm256_mul_ps(_mm256_mul_ps(rinvsq00,rinvsq00),rinvsq00);
1026             fvdw             = _mm256_mul_ps(_mm256_sub_ps(_mm256_mul_ps(c12_00,rinvsix),c6_00),_mm256_mul_ps(rinvsix,rinvsq00));
1027
1028             fscal            = _mm256_add_ps(felec,fvdw);
1029
1030             /* Calculate temporary vectorial force */
1031             tx               = _mm256_mul_ps(fscal,dx00);
1032             ty               = _mm256_mul_ps(fscal,dy00);
1033             tz               = _mm256_mul_ps(fscal,dz00);
1034
1035             /* Update vectorial force */
1036             fix0             = _mm256_add_ps(fix0,tx);
1037             fiy0             = _mm256_add_ps(fiy0,ty);
1038             fiz0             = _mm256_add_ps(fiz0,tz);
1039
1040             fjx0             = _mm256_add_ps(fjx0,tx);
1041             fjy0             = _mm256_add_ps(fjy0,ty);
1042             fjz0             = _mm256_add_ps(fjz0,tz);
1043
1044             /**************************
1045              * CALCULATE INTERACTIONS *
1046              **************************/
1047
1048             r10              = _mm256_mul_ps(rsq10,rinv10);
1049
1050             /* Compute parameters for interactions between i and j atoms */
1051             qq10             = _mm256_mul_ps(iq1,jq0);
1052
1053             /* Calculate table index by multiplying r with table scale and truncate to integer */
1054             rt               = _mm256_mul_ps(r10,vftabscale);
1055             vfitab           = _mm256_cvttps_epi32(rt);
1056             vfeps            = _mm256_sub_ps(rt,_mm256_round_ps(rt, _MM_FROUND_FLOOR));
1057             /*         AVX1 does not support 256-bit integer operations, so now we go to 128-bit mode... */
1058             vfitab_lo        = _mm256_extractf128_si256(vfitab,0x0);
1059             vfitab_hi        = _mm256_extractf128_si256(vfitab,0x1);
1060             vfitab_lo        = _mm_slli_epi32(vfitab_lo,2);
1061             vfitab_hi        = _mm_slli_epi32(vfitab_hi,2);
1062
1063             /* CUBIC SPLINE TABLE ELECTROSTATICS */
1064             Y                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,0)),
1065                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,0)));
1066             F                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,1)),
1067                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,1)));
1068             G                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,2)),
1069                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,2)));
1070             H                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,3)),
1071                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,3)));
1072             GMX_MM256_HALFTRANSPOSE4_PS(Y,F,G,H);
1073             Heps             = _mm256_mul_ps(vfeps,H);
1074             Fp               = _mm256_add_ps(F,_mm256_mul_ps(vfeps,_mm256_add_ps(G,Heps)));
1075             FF               = _mm256_add_ps(Fp,_mm256_mul_ps(vfeps,_mm256_add_ps(G,_mm256_add_ps(Heps,Heps))));
1076             felec            = _mm256_xor_ps(signbit,_mm256_mul_ps(_mm256_mul_ps(qq10,FF),_mm256_mul_ps(vftabscale,rinv10)));
1077
1078             fscal            = felec;
1079
1080             /* Calculate temporary vectorial force */
1081             tx               = _mm256_mul_ps(fscal,dx10);
1082             ty               = _mm256_mul_ps(fscal,dy10);
1083             tz               = _mm256_mul_ps(fscal,dz10);
1084
1085             /* Update vectorial force */
1086             fix1             = _mm256_add_ps(fix1,tx);
1087             fiy1             = _mm256_add_ps(fiy1,ty);
1088             fiz1             = _mm256_add_ps(fiz1,tz);
1089
1090             fjx0             = _mm256_add_ps(fjx0,tx);
1091             fjy0             = _mm256_add_ps(fjy0,ty);
1092             fjz0             = _mm256_add_ps(fjz0,tz);
1093
1094             /**************************
1095              * CALCULATE INTERACTIONS *
1096              **************************/
1097
1098             r20              = _mm256_mul_ps(rsq20,rinv20);
1099
1100             /* Compute parameters for interactions between i and j atoms */
1101             qq20             = _mm256_mul_ps(iq2,jq0);
1102
1103             /* Calculate table index by multiplying r with table scale and truncate to integer */
1104             rt               = _mm256_mul_ps(r20,vftabscale);
1105             vfitab           = _mm256_cvttps_epi32(rt);
1106             vfeps            = _mm256_sub_ps(rt,_mm256_round_ps(rt, _MM_FROUND_FLOOR));
1107             /*         AVX1 does not support 256-bit integer operations, so now we go to 128-bit mode... */
1108             vfitab_lo        = _mm256_extractf128_si256(vfitab,0x0);
1109             vfitab_hi        = _mm256_extractf128_si256(vfitab,0x1);
1110             vfitab_lo        = _mm_slli_epi32(vfitab_lo,2);
1111             vfitab_hi        = _mm_slli_epi32(vfitab_hi,2);
1112
1113             /* CUBIC SPLINE TABLE ELECTROSTATICS */
1114             Y                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,0)),
1115                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,0)));
1116             F                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,1)),
1117                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,1)));
1118             G                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,2)),
1119                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,2)));
1120             H                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,3)),
1121                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,3)));
1122             GMX_MM256_HALFTRANSPOSE4_PS(Y,F,G,H);
1123             Heps             = _mm256_mul_ps(vfeps,H);
1124             Fp               = _mm256_add_ps(F,_mm256_mul_ps(vfeps,_mm256_add_ps(G,Heps)));
1125             FF               = _mm256_add_ps(Fp,_mm256_mul_ps(vfeps,_mm256_add_ps(G,_mm256_add_ps(Heps,Heps))));
1126             felec            = _mm256_xor_ps(signbit,_mm256_mul_ps(_mm256_mul_ps(qq20,FF),_mm256_mul_ps(vftabscale,rinv20)));
1127
1128             fscal            = felec;
1129
1130             /* Calculate temporary vectorial force */
1131             tx               = _mm256_mul_ps(fscal,dx20);
1132             ty               = _mm256_mul_ps(fscal,dy20);
1133             tz               = _mm256_mul_ps(fscal,dz20);
1134
1135             /* Update vectorial force */
1136             fix2             = _mm256_add_ps(fix2,tx);
1137             fiy2             = _mm256_add_ps(fiy2,ty);
1138             fiz2             = _mm256_add_ps(fiz2,tz);
1139
1140             fjx0             = _mm256_add_ps(fjx0,tx);
1141             fjy0             = _mm256_add_ps(fjy0,ty);
1142             fjz0             = _mm256_add_ps(fjz0,tz);
1143
1144             fjptrA             = f+j_coord_offsetA;
1145             fjptrB             = f+j_coord_offsetB;
1146             fjptrC             = f+j_coord_offsetC;
1147             fjptrD             = f+j_coord_offsetD;
1148             fjptrE             = f+j_coord_offsetE;
1149             fjptrF             = f+j_coord_offsetF;
1150             fjptrG             = f+j_coord_offsetG;
1151             fjptrH             = f+j_coord_offsetH;
1152
1153             gmx_mm256_decrement_1rvec_8ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjptrE,fjptrF,fjptrG,fjptrH,fjx0,fjy0,fjz0);
1154
1155             /* Inner loop uses 128 flops */
1156         }
1157
1158         if(jidx<j_index_end)
1159         {
1160
1161             /* Get j neighbor index, and coordinate index */
1162             jnrlistA         = jjnr[jidx];
1163             jnrlistB         = jjnr[jidx+1];
1164             jnrlistC         = jjnr[jidx+2];
1165             jnrlistD         = jjnr[jidx+3];
1166             jnrlistE         = jjnr[jidx+4];
1167             jnrlistF         = jjnr[jidx+5];
1168             jnrlistG         = jjnr[jidx+6];
1169             jnrlistH         = jjnr[jidx+7];
1170             /* Sign of each element will be negative for non-real atoms.
1171              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
1172              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
1173              */
1174             dummy_mask = gmx_mm256_set_m128(gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx+4)),_mm_setzero_si128())),
1175                                             gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128())));
1176
1177             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
1178             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
1179             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
1180             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
1181             jnrE       = (jnrlistE>=0) ? jnrlistE : 0;
1182             jnrF       = (jnrlistF>=0) ? jnrlistF : 0;
1183             jnrG       = (jnrlistG>=0) ? jnrlistG : 0;
1184             jnrH       = (jnrlistH>=0) ? jnrlistH : 0;
1185             j_coord_offsetA  = DIM*jnrA;
1186             j_coord_offsetB  = DIM*jnrB;
1187             j_coord_offsetC  = DIM*jnrC;
1188             j_coord_offsetD  = DIM*jnrD;
1189             j_coord_offsetE  = DIM*jnrE;
1190             j_coord_offsetF  = DIM*jnrF;
1191             j_coord_offsetG  = DIM*jnrG;
1192             j_coord_offsetH  = DIM*jnrH;
1193
1194             /* load j atom coordinates */
1195             gmx_mm256_load_1rvec_8ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
1196                                                  x+j_coord_offsetC,x+j_coord_offsetD,
1197                                                  x+j_coord_offsetE,x+j_coord_offsetF,
1198                                                  x+j_coord_offsetG,x+j_coord_offsetH,
1199                                                  &jx0,&jy0,&jz0);
1200
1201             /* Calculate displacement vector */
1202             dx00             = _mm256_sub_ps(ix0,jx0);
1203             dy00             = _mm256_sub_ps(iy0,jy0);
1204             dz00             = _mm256_sub_ps(iz0,jz0);
1205             dx10             = _mm256_sub_ps(ix1,jx0);
1206             dy10             = _mm256_sub_ps(iy1,jy0);
1207             dz10             = _mm256_sub_ps(iz1,jz0);
1208             dx20             = _mm256_sub_ps(ix2,jx0);
1209             dy20             = _mm256_sub_ps(iy2,jy0);
1210             dz20             = _mm256_sub_ps(iz2,jz0);
1211
1212             /* Calculate squared distance and things based on it */
1213             rsq00            = gmx_mm256_calc_rsq_ps(dx00,dy00,dz00);
1214             rsq10            = gmx_mm256_calc_rsq_ps(dx10,dy10,dz10);
1215             rsq20            = gmx_mm256_calc_rsq_ps(dx20,dy20,dz20);
1216
1217             rinv00           = avx256_invsqrt_f(rsq00);
1218             rinv10           = avx256_invsqrt_f(rsq10);
1219             rinv20           = avx256_invsqrt_f(rsq20);
1220
1221             rinvsq00         = _mm256_mul_ps(rinv00,rinv00);
1222
1223             /* Load parameters for j particles */
1224             jq0              = gmx_mm256_load_8real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
1225                                                                  charge+jnrC+0,charge+jnrD+0,
1226                                                                  charge+jnrE+0,charge+jnrF+0,
1227                                                                  charge+jnrG+0,charge+jnrH+0);
1228             vdwjidx0A        = 2*vdwtype[jnrA+0];
1229             vdwjidx0B        = 2*vdwtype[jnrB+0];
1230             vdwjidx0C        = 2*vdwtype[jnrC+0];
1231             vdwjidx0D        = 2*vdwtype[jnrD+0];
1232             vdwjidx0E        = 2*vdwtype[jnrE+0];
1233             vdwjidx0F        = 2*vdwtype[jnrF+0];
1234             vdwjidx0G        = 2*vdwtype[jnrG+0];
1235             vdwjidx0H        = 2*vdwtype[jnrH+0];
1236
1237             fjx0             = _mm256_setzero_ps();
1238             fjy0             = _mm256_setzero_ps();
1239             fjz0             = _mm256_setzero_ps();
1240
1241             /**************************
1242              * CALCULATE INTERACTIONS *
1243              **************************/
1244
1245             r00              = _mm256_mul_ps(rsq00,rinv00);
1246             r00              = _mm256_andnot_ps(dummy_mask,r00);
1247
1248             /* Compute parameters for interactions between i and j atoms */
1249             qq00             = _mm256_mul_ps(iq0,jq0);
1250             gmx_mm256_load_8pair_swizzle_ps(vdwioffsetptr0+vdwjidx0A,
1251                                             vdwioffsetptr0+vdwjidx0B,
1252                                             vdwioffsetptr0+vdwjidx0C,
1253                                             vdwioffsetptr0+vdwjidx0D,
1254                                             vdwioffsetptr0+vdwjidx0E,
1255                                             vdwioffsetptr0+vdwjidx0F,
1256                                             vdwioffsetptr0+vdwjidx0G,
1257                                             vdwioffsetptr0+vdwjidx0H,
1258                                             &c6_00,&c12_00);
1259
1260             /* Calculate table index by multiplying r with table scale and truncate to integer */
1261             rt               = _mm256_mul_ps(r00,vftabscale);
1262             vfitab           = _mm256_cvttps_epi32(rt);
1263             vfeps            = _mm256_sub_ps(rt,_mm256_round_ps(rt, _MM_FROUND_FLOOR));
1264             /*         AVX1 does not support 256-bit integer operations, so now we go to 128-bit mode... */
1265             vfitab_lo        = _mm256_extractf128_si256(vfitab,0x0);
1266             vfitab_hi        = _mm256_extractf128_si256(vfitab,0x1);
1267             vfitab_lo        = _mm_slli_epi32(vfitab_lo,2);
1268             vfitab_hi        = _mm_slli_epi32(vfitab_hi,2);
1269
1270             /* CUBIC SPLINE TABLE ELECTROSTATICS */
1271             Y                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,0)),
1272                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,0)));
1273             F                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,1)),
1274                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,1)));
1275             G                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,2)),
1276                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,2)));
1277             H                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,3)),
1278                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,3)));
1279             GMX_MM256_HALFTRANSPOSE4_PS(Y,F,G,H);
1280             Heps             = _mm256_mul_ps(vfeps,H);
1281             Fp               = _mm256_add_ps(F,_mm256_mul_ps(vfeps,_mm256_add_ps(G,Heps)));
1282             FF               = _mm256_add_ps(Fp,_mm256_mul_ps(vfeps,_mm256_add_ps(G,_mm256_add_ps(Heps,Heps))));
1283             felec            = _mm256_xor_ps(signbit,_mm256_mul_ps(_mm256_mul_ps(qq00,FF),_mm256_mul_ps(vftabscale,rinv00)));
1284
1285             /* LENNARD-JONES DISPERSION/REPULSION */
1286
1287             rinvsix          = _mm256_mul_ps(_mm256_mul_ps(rinvsq00,rinvsq00),rinvsq00);
1288             fvdw             = _mm256_mul_ps(_mm256_sub_ps(_mm256_mul_ps(c12_00,rinvsix),c6_00),_mm256_mul_ps(rinvsix,rinvsq00));
1289
1290             fscal            = _mm256_add_ps(felec,fvdw);
1291
1292             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
1293
1294             /* Calculate temporary vectorial force */
1295             tx               = _mm256_mul_ps(fscal,dx00);
1296             ty               = _mm256_mul_ps(fscal,dy00);
1297             tz               = _mm256_mul_ps(fscal,dz00);
1298
1299             /* Update vectorial force */
1300             fix0             = _mm256_add_ps(fix0,tx);
1301             fiy0             = _mm256_add_ps(fiy0,ty);
1302             fiz0             = _mm256_add_ps(fiz0,tz);
1303
1304             fjx0             = _mm256_add_ps(fjx0,tx);
1305             fjy0             = _mm256_add_ps(fjy0,ty);
1306             fjz0             = _mm256_add_ps(fjz0,tz);
1307
1308             /**************************
1309              * CALCULATE INTERACTIONS *
1310              **************************/
1311
1312             r10              = _mm256_mul_ps(rsq10,rinv10);
1313             r10              = _mm256_andnot_ps(dummy_mask,r10);
1314
1315             /* Compute parameters for interactions between i and j atoms */
1316             qq10             = _mm256_mul_ps(iq1,jq0);
1317
1318             /* Calculate table index by multiplying r with table scale and truncate to integer */
1319             rt               = _mm256_mul_ps(r10,vftabscale);
1320             vfitab           = _mm256_cvttps_epi32(rt);
1321             vfeps            = _mm256_sub_ps(rt,_mm256_round_ps(rt, _MM_FROUND_FLOOR));
1322             /*         AVX1 does not support 256-bit integer operations, so now we go to 128-bit mode... */
1323             vfitab_lo        = _mm256_extractf128_si256(vfitab,0x0);
1324             vfitab_hi        = _mm256_extractf128_si256(vfitab,0x1);
1325             vfitab_lo        = _mm_slli_epi32(vfitab_lo,2);
1326             vfitab_hi        = _mm_slli_epi32(vfitab_hi,2);
1327
1328             /* CUBIC SPLINE TABLE ELECTROSTATICS */
1329             Y                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,0)),
1330                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,0)));
1331             F                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,1)),
1332                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,1)));
1333             G                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,2)),
1334                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,2)));
1335             H                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,3)),
1336                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,3)));
1337             GMX_MM256_HALFTRANSPOSE4_PS(Y,F,G,H);
1338             Heps             = _mm256_mul_ps(vfeps,H);
1339             Fp               = _mm256_add_ps(F,_mm256_mul_ps(vfeps,_mm256_add_ps(G,Heps)));
1340             FF               = _mm256_add_ps(Fp,_mm256_mul_ps(vfeps,_mm256_add_ps(G,_mm256_add_ps(Heps,Heps))));
1341             felec            = _mm256_xor_ps(signbit,_mm256_mul_ps(_mm256_mul_ps(qq10,FF),_mm256_mul_ps(vftabscale,rinv10)));
1342
1343             fscal            = felec;
1344
1345             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
1346
1347             /* Calculate temporary vectorial force */
1348             tx               = _mm256_mul_ps(fscal,dx10);
1349             ty               = _mm256_mul_ps(fscal,dy10);
1350             tz               = _mm256_mul_ps(fscal,dz10);
1351
1352             /* Update vectorial force */
1353             fix1             = _mm256_add_ps(fix1,tx);
1354             fiy1             = _mm256_add_ps(fiy1,ty);
1355             fiz1             = _mm256_add_ps(fiz1,tz);
1356
1357             fjx0             = _mm256_add_ps(fjx0,tx);
1358             fjy0             = _mm256_add_ps(fjy0,ty);
1359             fjz0             = _mm256_add_ps(fjz0,tz);
1360
1361             /**************************
1362              * CALCULATE INTERACTIONS *
1363              **************************/
1364
1365             r20              = _mm256_mul_ps(rsq20,rinv20);
1366             r20              = _mm256_andnot_ps(dummy_mask,r20);
1367
1368             /* Compute parameters for interactions between i and j atoms */
1369             qq20             = _mm256_mul_ps(iq2,jq0);
1370
1371             /* Calculate table index by multiplying r with table scale and truncate to integer */
1372             rt               = _mm256_mul_ps(r20,vftabscale);
1373             vfitab           = _mm256_cvttps_epi32(rt);
1374             vfeps            = _mm256_sub_ps(rt,_mm256_round_ps(rt, _MM_FROUND_FLOOR));
1375             /*         AVX1 does not support 256-bit integer operations, so now we go to 128-bit mode... */
1376             vfitab_lo        = _mm256_extractf128_si256(vfitab,0x0);
1377             vfitab_hi        = _mm256_extractf128_si256(vfitab,0x1);
1378             vfitab_lo        = _mm_slli_epi32(vfitab_lo,2);
1379             vfitab_hi        = _mm_slli_epi32(vfitab_hi,2);
1380
1381             /* CUBIC SPLINE TABLE ELECTROSTATICS */
1382             Y                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,0)),
1383                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,0)));
1384             F                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,1)),
1385                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,1)));
1386             G                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,2)),
1387                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,2)));
1388             H                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,3)),
1389                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,3)));
1390             GMX_MM256_HALFTRANSPOSE4_PS(Y,F,G,H);
1391             Heps             = _mm256_mul_ps(vfeps,H);
1392             Fp               = _mm256_add_ps(F,_mm256_mul_ps(vfeps,_mm256_add_ps(G,Heps)));
1393             FF               = _mm256_add_ps(Fp,_mm256_mul_ps(vfeps,_mm256_add_ps(G,_mm256_add_ps(Heps,Heps))));
1394             felec            = _mm256_xor_ps(signbit,_mm256_mul_ps(_mm256_mul_ps(qq20,FF),_mm256_mul_ps(vftabscale,rinv20)));
1395
1396             fscal            = felec;
1397
1398             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
1399
1400             /* Calculate temporary vectorial force */
1401             tx               = _mm256_mul_ps(fscal,dx20);
1402             ty               = _mm256_mul_ps(fscal,dy20);
1403             tz               = _mm256_mul_ps(fscal,dz20);
1404
1405             /* Update vectorial force */
1406             fix2             = _mm256_add_ps(fix2,tx);
1407             fiy2             = _mm256_add_ps(fiy2,ty);
1408             fiz2             = _mm256_add_ps(fiz2,tz);
1409
1410             fjx0             = _mm256_add_ps(fjx0,tx);
1411             fjy0             = _mm256_add_ps(fjy0,ty);
1412             fjz0             = _mm256_add_ps(fjz0,tz);
1413
1414             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
1415             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
1416             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
1417             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
1418             fjptrE             = (jnrlistE>=0) ? f+j_coord_offsetE : scratch;
1419             fjptrF             = (jnrlistF>=0) ? f+j_coord_offsetF : scratch;
1420             fjptrG             = (jnrlistG>=0) ? f+j_coord_offsetG : scratch;
1421             fjptrH             = (jnrlistH>=0) ? f+j_coord_offsetH : scratch;
1422
1423             gmx_mm256_decrement_1rvec_8ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjptrE,fjptrF,fjptrG,fjptrH,fjx0,fjy0,fjz0);
1424
1425             /* Inner loop uses 131 flops */
1426         }
1427
1428         /* End of innermost loop */
1429
1430         gmx_mm256_update_iforce_3atom_swizzle_ps(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,
1431                                                  f+i_coord_offset,fshift+i_shift_offset);
1432
1433         /* Increment number of inner iterations */
1434         inneriter                  += j_index_end - j_index_start;
1435
1436         /* Outer loop uses 18 flops */
1437     }
1438
1439     /* Increment number of outer iterations */
1440     outeriter        += nri;
1441
1442     /* Update outer/inner flops */
1443
1444     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W3_F,outeriter*18 + inneriter*131);
1445 }