src/gromacs/gmxlib/nonbonded/nb_kernel_avx_256_single/nb_kernel_ElecCoul_VdwCSTab_GeomW4P1_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_ElecCoul_VdwCSTab_GeomW4P1_VF_avx_256_single
  51  * Electrostatics interaction: Coulomb
  52  * VdW interaction:            CubicSplineTable
  53  * Geometry:                   Water4-Particle
  54  * Calculate force/pot:        PotentialAndForce
  55  */
  56 void
  57 nb_kernel_ElecCoul_VdwCSTab_GeomW4P1_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     real *           vdwioffsetptr3;
  92     __m256           ix3,iy3,iz3,fix3,fiy3,fiz3,iq3,isai3;
  93     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D,vdwjidx0E,vdwjidx0F,vdwjidx0G,vdwjidx0H;
  94     __m256           jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
  95     __m256           dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
  96     __m256           dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
  97     __m256           dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
  98     __m256           dx30,dy30,dz30,rsq30,rinv30,rinvsq30,r30,qq30,c6_30,c12_30;
  99     __m256           velec,felec,velecsum,facel,crf,krf,krf2;
 100     real             *charge;
 101     int              nvdwtype;
 102     __m256           rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
 103     int              *vdwtype;
 104     real             *vdwparam;
 105     __m256           one_sixth   = _mm256_set1_ps(1.0/6.0);
 106     __m256           one_twelfth = _mm256_set1_ps(1.0/12.0);
 107     __m256i          vfitab;
 108     __m128i          vfitab_lo,vfitab_hi;
 109     __m128i          ifour       = _mm_set1_epi32(4);
 110     __m256           rt,vfeps,vftabscale,Y,F,G,H,Heps,Fp,VV,FF;
 111     real             *vftab;
 112     __m256           dummy_mask,cutoff_mask;
 113     __m256           signbit = _mm256_castsi256_ps( _mm256_set1_epi32(0x80000000) );
 114     __m256           one     = _mm256_set1_ps(1.0);
 115     __m256           two     = _mm256_set1_ps(2.0);
 116     x                = xx[0];
 117     f                = ff[0];
 118
 119     nri              = nlist->nri;
 120     iinr             = nlist->iinr;
 121     jindex           = nlist->jindex;
 122     jjnr             = nlist->jjnr;
 123     shiftidx         = nlist->shift;
 124     gid              = nlist->gid;
 125     shiftvec         = fr->shift_vec[0];
 126     fshift           = fr->fshift[0];
 127     facel            = _mm256_set1_ps(fr->ic->epsfac);
 128     charge           = mdatoms->chargeA;
 129     nvdwtype         = fr->ntype;
 130     vdwparam         = fr->nbfp;
 131     vdwtype          = mdatoms->typeA;
 132
 133     vftab            = kernel_data->table_vdw->data;
 134     vftabscale       = _mm256_set1_ps(kernel_data->table_vdw->scale);
 135
 136     /* Setup water-specific parameters */
 137     inr              = nlist->iinr[0];
 138     iq1              = _mm256_mul_ps(facel,_mm256_set1_ps(charge[inr+1]));
 139     iq2              = _mm256_mul_ps(facel,_mm256_set1_ps(charge[inr+2]));
 140     iq3              = _mm256_mul_ps(facel,_mm256_set1_ps(charge[inr+3]));
 141     vdwioffsetptr0   = vdwparam+2*nvdwtype*vdwtype[inr+0];
 142
 143     /* Avoid stupid compiler warnings */
 144     jnrA = jnrB = jnrC = jnrD = jnrE = jnrF = jnrG = jnrH = 0;
 145     j_coord_offsetA = 0;
 146     j_coord_offsetB = 0;
 147     j_coord_offsetC = 0;
 148     j_coord_offsetD = 0;
 149     j_coord_offsetE = 0;
 150     j_coord_offsetF = 0;
 151     j_coord_offsetG = 0;
 152     j_coord_offsetH = 0;
 153
 154     outeriter        = 0;
 155     inneriter        = 0;
 156
 157     for(iidx=0;iidx<4*DIM;iidx++)
 158     {
 159         scratch[iidx] = 0.0;
 160     }
 161
 162     /* Start outer loop over neighborlists */
 163     for(iidx=0; iidx<nri; iidx++)
 164     {
 165         /* Load shift vector for this list */
 166         i_shift_offset   = DIM*shiftidx[iidx];
 167
 168         /* Load limits for loop over neighbors */
 169         j_index_start    = jindex[iidx];
 170         j_index_end      = jindex[iidx+1];
 171
 172         /* Get outer coordinate index */
 173         inr              = iinr[iidx];
 174         i_coord_offset   = DIM*inr;
 175
 176         /* Load i particle coords and add shift vector */
 177         gmx_mm256_load_shift_and_4rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,
 178                                                     &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2,&ix3,&iy3,&iz3);
 179
 180         fix0             = _mm256_setzero_ps();
 181         fiy0             = _mm256_setzero_ps();
 182         fiz0             = _mm256_setzero_ps();
 183         fix1             = _mm256_setzero_ps();
 184         fiy1             = _mm256_setzero_ps();
 185         fiz1             = _mm256_setzero_ps();
 186         fix2             = _mm256_setzero_ps();
 187         fiy2             = _mm256_setzero_ps();
 188         fiz2             = _mm256_setzero_ps();
 189         fix3             = _mm256_setzero_ps();
 190         fiy3             = _mm256_setzero_ps();
 191         fiz3             = _mm256_setzero_ps();
 192
 193         /* Reset potential sums */
 194         velecsum         = _mm256_setzero_ps();
 195         vvdwsum          = _mm256_setzero_ps();
 196
 197         /* Start inner kernel loop */
 198         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+7]>=0; jidx+=8)
 199         {
 200
 201             /* Get j neighbor index, and coordinate index */
 202             jnrA             = jjnr[jidx];
 203             jnrB             = jjnr[jidx+1];
 204             jnrC             = jjnr[jidx+2];
 205             jnrD             = jjnr[jidx+3];
 206             jnrE             = jjnr[jidx+4];
 207             jnrF             = jjnr[jidx+5];
 208             jnrG             = jjnr[jidx+6];
 209             jnrH             = jjnr[jidx+7];
 210             j_coord_offsetA  = DIM*jnrA;
 211             j_coord_offsetB  = DIM*jnrB;
 212             j_coord_offsetC  = DIM*jnrC;
 213             j_coord_offsetD  = DIM*jnrD;
 214             j_coord_offsetE  = DIM*jnrE;
 215             j_coord_offsetF  = DIM*jnrF;
 216             j_coord_offsetG  = DIM*jnrG;
 217             j_coord_offsetH  = DIM*jnrH;
 218
 219             /* load j atom coordinates */
 220             gmx_mm256_load_1rvec_8ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 221                                                  x+j_coord_offsetC,x+j_coord_offsetD,
 222                                                  x+j_coord_offsetE,x+j_coord_offsetF,
 223                                                  x+j_coord_offsetG,x+j_coord_offsetH,
 224                                                  &jx0,&jy0,&jz0);
 225
 226             /* Calculate displacement vector */
 227             dx00             = _mm256_sub_ps(ix0,jx0);
 228             dy00             = _mm256_sub_ps(iy0,jy0);
 229             dz00             = _mm256_sub_ps(iz0,jz0);
 230             dx10             = _mm256_sub_ps(ix1,jx0);
 231             dy10             = _mm256_sub_ps(iy1,jy0);
 232             dz10             = _mm256_sub_ps(iz1,jz0);
 233             dx20             = _mm256_sub_ps(ix2,jx0);
 234             dy20             = _mm256_sub_ps(iy2,jy0);
 235             dz20             = _mm256_sub_ps(iz2,jz0);
 236             dx30             = _mm256_sub_ps(ix3,jx0);
 237             dy30             = _mm256_sub_ps(iy3,jy0);
 238             dz30             = _mm256_sub_ps(iz3,jz0);
 239
 240             /* Calculate squared distance and things based on it */
 241             rsq00            = gmx_mm256_calc_rsq_ps(dx00,dy00,dz00);
 242             rsq10            = gmx_mm256_calc_rsq_ps(dx10,dy10,dz10);
 243             rsq20            = gmx_mm256_calc_rsq_ps(dx20,dy20,dz20);
 244             rsq30            = gmx_mm256_calc_rsq_ps(dx30,dy30,dz30);
 245
 246             rinv00           = avx256_invsqrt_f(rsq00);
 247             rinv10           = avx256_invsqrt_f(rsq10);
 248             rinv20           = avx256_invsqrt_f(rsq20);
 249             rinv30           = avx256_invsqrt_f(rsq30);
 250
 251             rinvsq10         = _mm256_mul_ps(rinv10,rinv10);
 252             rinvsq20         = _mm256_mul_ps(rinv20,rinv20);
 253             rinvsq30         = _mm256_mul_ps(rinv30,rinv30);
 254
 255             /* Load parameters for j particles */
 256             jq0              = gmx_mm256_load_8real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 257                                                                  charge+jnrC+0,charge+jnrD+0,
 258                                                                  charge+jnrE+0,charge+jnrF+0,
 259                                                                  charge+jnrG+0,charge+jnrH+0);
 260             vdwjidx0A        = 2*vdwtype[jnrA+0];
 261             vdwjidx0B        = 2*vdwtype[jnrB+0];
 262             vdwjidx0C        = 2*vdwtype[jnrC+0];
 263             vdwjidx0D        = 2*vdwtype[jnrD+0];
 264             vdwjidx0E        = 2*vdwtype[jnrE+0];
 265             vdwjidx0F        = 2*vdwtype[jnrF+0];
 266             vdwjidx0G        = 2*vdwtype[jnrG+0];
 267             vdwjidx0H        = 2*vdwtype[jnrH+0];
 268
 269             fjx0             = _mm256_setzero_ps();
 270             fjy0             = _mm256_setzero_ps();
 271             fjz0             = _mm256_setzero_ps();
 272
 273             /**************************
 274              * CALCULATE INTERACTIONS *
 275              **************************/
 276
 277             r00              = _mm256_mul_ps(rsq00,rinv00);
 278
 279             /* Compute parameters for interactions between i and j atoms */
 280             gmx_mm256_load_8pair_swizzle_ps(vdwioffsetptr0+vdwjidx0A,
 281                                             vdwioffsetptr0+vdwjidx0B,
 282                                             vdwioffsetptr0+vdwjidx0C,
 283                                             vdwioffsetptr0+vdwjidx0D,
 284                                             vdwioffsetptr0+vdwjidx0E,
 285                                             vdwioffsetptr0+vdwjidx0F,
 286                                             vdwioffsetptr0+vdwjidx0G,
 287                                             vdwioffsetptr0+vdwjidx0H,
 288                                             &c6_00,&c12_00);
 289
 290             /* Calculate table index by multiplying r with table scale and truncate to integer */
 291             rt               = _mm256_mul_ps(r00,vftabscale);
 292             vfitab           = _mm256_cvttps_epi32(rt);
 293             vfeps            = _mm256_sub_ps(rt,_mm256_round_ps(rt, _MM_FROUND_FLOOR));
 294             /*         AVX1 does not support 256-bit integer operations, so now we go to 128-bit mode... */
 295             vfitab_lo        = _mm256_extractf128_si256(vfitab,0x0);
 296             vfitab_hi        = _mm256_extractf128_si256(vfitab,0x1);
 297             vfitab_lo        = _mm_slli_epi32(vfitab_lo,3);
 298             vfitab_hi        = _mm_slli_epi32(vfitab_hi,3);
 299
 300             /* CUBIC SPLINE TABLE DISPERSION */
 301             Y                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,0)),
 302                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,0)));
 303             F                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,1)),
 304                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,1)));
 305             G                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,2)),
 306                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,2)));
 307             H                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,3)),
 308                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,3)));
 309             GMX_MM256_HALFTRANSPOSE4_PS(Y,F,G,H);
 310             Heps             = _mm256_mul_ps(vfeps,H);
 311             Fp               = _mm256_add_ps(F,_mm256_mul_ps(vfeps,_mm256_add_ps(G,Heps)));
 312             VV               = _mm256_add_ps(Y,_mm256_mul_ps(vfeps,Fp));
 313             vvdw6            = _mm256_mul_ps(c6_00,VV);
 314             FF               = _mm256_add_ps(Fp,_mm256_mul_ps(vfeps,_mm256_add_ps(G,_mm256_add_ps(Heps,Heps))));
 315             fvdw6            = _mm256_mul_ps(c6_00,FF);
 316
 317             /* CUBIC SPLINE TABLE REPULSION */
 318             vfitab_lo        = _mm_add_epi32(vfitab_lo,ifour);
 319             vfitab_hi        = _mm_add_epi32(vfitab_hi,ifour);
 320             Y                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,0)),
 321                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,0)));
 322             F                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,1)),
 323                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,1)));
 324             G                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,2)),
 325                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,2)));
 326             H                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,3)),
 327                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,3)));
 328             GMX_MM256_HALFTRANSPOSE4_PS(Y,F,G,H);
 329             Heps             = _mm256_mul_ps(vfeps,H);
 330             Fp               = _mm256_add_ps(F,_mm256_mul_ps(vfeps,_mm256_add_ps(G,Heps)));
 331             VV               = _mm256_add_ps(Y,_mm256_mul_ps(vfeps,Fp));
 332             vvdw12           = _mm256_mul_ps(c12_00,VV);
 333             FF               = _mm256_add_ps(Fp,_mm256_mul_ps(vfeps,_mm256_add_ps(G,_mm256_add_ps(Heps,Heps))));
 334             fvdw12           = _mm256_mul_ps(c12_00,FF);
 335             vvdw             = _mm256_add_ps(vvdw12,vvdw6);
 336             fvdw             = _mm256_xor_ps(signbit,_mm256_mul_ps(_mm256_add_ps(fvdw6,fvdw12),_mm256_mul_ps(vftabscale,rinv00)));
 337
 338             /* Update potential sum for this i atom from the interaction with this j atom. */
 339             vvdwsum          = _mm256_add_ps(vvdwsum,vvdw);
 340
 341             fscal            = fvdw;
 342
 343             /* Calculate temporary vectorial force */
 344             tx               = _mm256_mul_ps(fscal,dx00);
 345             ty               = _mm256_mul_ps(fscal,dy00);
 346             tz               = _mm256_mul_ps(fscal,dz00);
 347
 348             /* Update vectorial force */
 349             fix0             = _mm256_add_ps(fix0,tx);
 350             fiy0             = _mm256_add_ps(fiy0,ty);
 351             fiz0             = _mm256_add_ps(fiz0,tz);
 352
 353             fjx0             = _mm256_add_ps(fjx0,tx);
 354             fjy0             = _mm256_add_ps(fjy0,ty);
 355             fjz0             = _mm256_add_ps(fjz0,tz);
 356
 357             /**************************
 358              * CALCULATE INTERACTIONS *
 359              **************************/
 360
 361             /* Compute parameters for interactions between i and j atoms */
 362             qq10             = _mm256_mul_ps(iq1,jq0);
 363
 364             /* COULOMB ELECTROSTATICS */
 365             velec            = _mm256_mul_ps(qq10,rinv10);
 366             felec            = _mm256_mul_ps(velec,rinvsq10);
 367
 368             /* Update potential sum for this i atom from the interaction with this j atom. */
 369             velecsum         = _mm256_add_ps(velecsum,velec);
 370
 371             fscal            = felec;
 372
 373             /* Calculate temporary vectorial force */
 374             tx               = _mm256_mul_ps(fscal,dx10);
 375             ty               = _mm256_mul_ps(fscal,dy10);
 376             tz               = _mm256_mul_ps(fscal,dz10);
 377
 378             /* Update vectorial force */
 379             fix1             = _mm256_add_ps(fix1,tx);
 380             fiy1             = _mm256_add_ps(fiy1,ty);
 381             fiz1             = _mm256_add_ps(fiz1,tz);
 382
 383             fjx0             = _mm256_add_ps(fjx0,tx);
 384             fjy0             = _mm256_add_ps(fjy0,ty);
 385             fjz0             = _mm256_add_ps(fjz0,tz);
 386
 387             /**************************
 388              * CALCULATE INTERACTIONS *
 389              **************************/
 390
 391             /* Compute parameters for interactions between i and j atoms */
 392             qq20             = _mm256_mul_ps(iq2,jq0);
 393
 394             /* COULOMB ELECTROSTATICS */
 395             velec            = _mm256_mul_ps(qq20,rinv20);
 396             felec            = _mm256_mul_ps(velec,rinvsq20);
 397
 398             /* Update potential sum for this i atom from the interaction with this j atom. */
 399             velecsum         = _mm256_add_ps(velecsum,velec);
 400
 401             fscal            = felec;
 402
 403             /* Calculate temporary vectorial force */
 404             tx               = _mm256_mul_ps(fscal,dx20);
 405             ty               = _mm256_mul_ps(fscal,dy20);
 406             tz               = _mm256_mul_ps(fscal,dz20);
 407
 408             /* Update vectorial force */
 409             fix2             = _mm256_add_ps(fix2,tx);
 410             fiy2             = _mm256_add_ps(fiy2,ty);
 411             fiz2             = _mm256_add_ps(fiz2,tz);
 412
 413             fjx0             = _mm256_add_ps(fjx0,tx);
 414             fjy0             = _mm256_add_ps(fjy0,ty);
 415             fjz0             = _mm256_add_ps(fjz0,tz);
 416
 417             /**************************
 418              * CALCULATE INTERACTIONS *
 419              **************************/
 420
 421             /* Compute parameters for interactions between i and j atoms */
 422             qq30             = _mm256_mul_ps(iq3,jq0);
 423
 424             /* COULOMB ELECTROSTATICS */
 425             velec            = _mm256_mul_ps(qq30,rinv30);
 426             felec            = _mm256_mul_ps(velec,rinvsq30);
 427
 428             /* Update potential sum for this i atom from the interaction with this j atom. */
 429             velecsum         = _mm256_add_ps(velecsum,velec);
 430
 431             fscal            = felec;
 432
 433             /* Calculate temporary vectorial force */
 434             tx               = _mm256_mul_ps(fscal,dx30);
 435             ty               = _mm256_mul_ps(fscal,dy30);
 436             tz               = _mm256_mul_ps(fscal,dz30);
 437
 438             /* Update vectorial force */
 439             fix3             = _mm256_add_ps(fix3,tx);
 440             fiy3             = _mm256_add_ps(fiy3,ty);
 441             fiz3             = _mm256_add_ps(fiz3,tz);
 442
 443             fjx0             = _mm256_add_ps(fjx0,tx);
 444             fjy0             = _mm256_add_ps(fjy0,ty);
 445             fjz0             = _mm256_add_ps(fjz0,tz);
 446
 447             fjptrA             = f+j_coord_offsetA;
 448             fjptrB             = f+j_coord_offsetB;
 449             fjptrC             = f+j_coord_offsetC;
 450             fjptrD             = f+j_coord_offsetD;
 451             fjptrE             = f+j_coord_offsetE;
 452             fjptrF             = f+j_coord_offsetF;
 453             fjptrG             = f+j_coord_offsetG;
 454             fjptrH             = f+j_coord_offsetH;
 455
 456             gmx_mm256_decrement_1rvec_8ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjptrE,fjptrF,fjptrG,fjptrH,fjx0,fjy0,fjz0);
 457
 458             /* Inner loop uses 140 flops */
 459         }
 460
 461         if(jidx<j_index_end)
 462         {
 463
 464             /* Get j neighbor index, and coordinate index */
 465             jnrlistA         = jjnr[jidx];
 466             jnrlistB         = jjnr[jidx+1];
 467             jnrlistC         = jjnr[jidx+2];
 468             jnrlistD         = jjnr[jidx+3];
 469             jnrlistE         = jjnr[jidx+4];
 470             jnrlistF         = jjnr[jidx+5];
 471             jnrlistG         = jjnr[jidx+6];
 472             jnrlistH         = jjnr[jidx+7];
 473             /* Sign of each element will be negative for non-real atoms.
 474              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 475              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
 476              */
 477             dummy_mask = gmx_mm256_set_m128(gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx+4)),_mm_setzero_si128())),
 478                                             gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128())));
 479
 480             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
 481             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
 482             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
 483             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
 484             jnrE       = (jnrlistE>=0) ? jnrlistE : 0;
 485             jnrF       = (jnrlistF>=0) ? jnrlistF : 0;
 486             jnrG       = (jnrlistG>=0) ? jnrlistG : 0;
 487             jnrH       = (jnrlistH>=0) ? jnrlistH : 0;
 488             j_coord_offsetA  = DIM*jnrA;
 489             j_coord_offsetB  = DIM*jnrB;
 490             j_coord_offsetC  = DIM*jnrC;
 491             j_coord_offsetD  = DIM*jnrD;
 492             j_coord_offsetE  = DIM*jnrE;
 493             j_coord_offsetF  = DIM*jnrF;
 494             j_coord_offsetG  = DIM*jnrG;
 495             j_coord_offsetH  = DIM*jnrH;
 496
 497             /* load j atom coordinates */
 498             gmx_mm256_load_1rvec_8ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 499                                                  x+j_coord_offsetC,x+j_coord_offsetD,
 500                                                  x+j_coord_offsetE,x+j_coord_offsetF,
 501                                                  x+j_coord_offsetG,x+j_coord_offsetH,
 502                                                  &jx0,&jy0,&jz0);
 503
 504             /* Calculate displacement vector */
 505             dx00             = _mm256_sub_ps(ix0,jx0);
 506             dy00             = _mm256_sub_ps(iy0,jy0);
 507             dz00             = _mm256_sub_ps(iz0,jz0);
 508             dx10             = _mm256_sub_ps(ix1,jx0);
 509             dy10             = _mm256_sub_ps(iy1,jy0);
 510             dz10             = _mm256_sub_ps(iz1,jz0);
 511             dx20             = _mm256_sub_ps(ix2,jx0);
 512             dy20             = _mm256_sub_ps(iy2,jy0);
 513             dz20             = _mm256_sub_ps(iz2,jz0);
 514             dx30             = _mm256_sub_ps(ix3,jx0);
 515             dy30             = _mm256_sub_ps(iy3,jy0);
 516             dz30             = _mm256_sub_ps(iz3,jz0);
 517
 518             /* Calculate squared distance and things based on it */
 519             rsq00            = gmx_mm256_calc_rsq_ps(dx00,dy00,dz00);
 520             rsq10            = gmx_mm256_calc_rsq_ps(dx10,dy10,dz10);
 521             rsq20            = gmx_mm256_calc_rsq_ps(dx20,dy20,dz20);
 522             rsq30            = gmx_mm256_calc_rsq_ps(dx30,dy30,dz30);
 523
 524             rinv00           = avx256_invsqrt_f(rsq00);
 525             rinv10           = avx256_invsqrt_f(rsq10);
 526             rinv20           = avx256_invsqrt_f(rsq20);
 527             rinv30           = avx256_invsqrt_f(rsq30);
 528
 529             rinvsq10         = _mm256_mul_ps(rinv10,rinv10);
 530             rinvsq20         = _mm256_mul_ps(rinv20,rinv20);
 531             rinvsq30         = _mm256_mul_ps(rinv30,rinv30);
 532
 533             /* Load parameters for j particles */
 534             jq0              = gmx_mm256_load_8real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 535                                                                  charge+jnrC+0,charge+jnrD+0,
 536                                                                  charge+jnrE+0,charge+jnrF+0,
 537                                                                  charge+jnrG+0,charge+jnrH+0);
 538             vdwjidx0A        = 2*vdwtype[jnrA+0];
 539             vdwjidx0B        = 2*vdwtype[jnrB+0];
 540             vdwjidx0C        = 2*vdwtype[jnrC+0];
 541             vdwjidx0D        = 2*vdwtype[jnrD+0];
 542             vdwjidx0E        = 2*vdwtype[jnrE+0];
 543             vdwjidx0F        = 2*vdwtype[jnrF+0];
 544             vdwjidx0G        = 2*vdwtype[jnrG+0];
 545             vdwjidx0H        = 2*vdwtype[jnrH+0];
 546
 547             fjx0             = _mm256_setzero_ps();
 548             fjy0             = _mm256_setzero_ps();
 549             fjz0             = _mm256_setzero_ps();
 550
 551             /**************************
 552              * CALCULATE INTERACTIONS *
 553              **************************/
 554
 555             r00              = _mm256_mul_ps(rsq00,rinv00);
 556             r00              = _mm256_andnot_ps(dummy_mask,r00);
 557
 558             /* Compute parameters for interactions between i and j atoms */
 559             gmx_mm256_load_8pair_swizzle_ps(vdwioffsetptr0+vdwjidx0A,
 560                                             vdwioffsetptr0+vdwjidx0B,
 561                                             vdwioffsetptr0+vdwjidx0C,
 562                                             vdwioffsetptr0+vdwjidx0D,
 563                                             vdwioffsetptr0+vdwjidx0E,
 564                                             vdwioffsetptr0+vdwjidx0F,
 565                                             vdwioffsetptr0+vdwjidx0G,
 566                                             vdwioffsetptr0+vdwjidx0H,
 567                                             &c6_00,&c12_00);
 568
 569             /* Calculate table index by multiplying r with table scale and truncate to integer */
 570             rt               = _mm256_mul_ps(r00,vftabscale);
 571             vfitab           = _mm256_cvttps_epi32(rt);
 572             vfeps            = _mm256_sub_ps(rt,_mm256_round_ps(rt, _MM_FROUND_FLOOR));
 573             /*         AVX1 does not support 256-bit integer operations, so now we go to 128-bit mode... */
 574             vfitab_lo        = _mm256_extractf128_si256(vfitab,0x0);
 575             vfitab_hi        = _mm256_extractf128_si256(vfitab,0x1);
 576             vfitab_lo        = _mm_slli_epi32(vfitab_lo,3);
 577             vfitab_hi        = _mm_slli_epi32(vfitab_hi,3);
 578
 579             /* CUBIC SPLINE TABLE DISPERSION */
 580             Y                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,0)),
 581                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,0)));
 582             F                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,1)),
 583                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,1)));
 584             G                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,2)),
 585                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,2)));
 586             H                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,3)),
 587                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,3)));
 588             GMX_MM256_HALFTRANSPOSE4_PS(Y,F,G,H);
 589             Heps             = _mm256_mul_ps(vfeps,H);
 590             Fp               = _mm256_add_ps(F,_mm256_mul_ps(vfeps,_mm256_add_ps(G,Heps)));
 591             VV               = _mm256_add_ps(Y,_mm256_mul_ps(vfeps,Fp));
 592             vvdw6            = _mm256_mul_ps(c6_00,VV);
 593             FF               = _mm256_add_ps(Fp,_mm256_mul_ps(vfeps,_mm256_add_ps(G,_mm256_add_ps(Heps,Heps))));
 594             fvdw6            = _mm256_mul_ps(c6_00,FF);
 595
 596             /* CUBIC SPLINE TABLE REPULSION */
 597             vfitab_lo        = _mm_add_epi32(vfitab_lo,ifour);
 598             vfitab_hi        = _mm_add_epi32(vfitab_hi,ifour);
 599             Y                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,0)),
 600                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,0)));
 601             F                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,1)),
 602                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,1)));
 603             G                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,2)),
 604                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,2)));
 605             H                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,3)),
 606                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,3)));
 607             GMX_MM256_HALFTRANSPOSE4_PS(Y,F,G,H);
 608             Heps             = _mm256_mul_ps(vfeps,H);
 609             Fp               = _mm256_add_ps(F,_mm256_mul_ps(vfeps,_mm256_add_ps(G,Heps)));
 610             VV               = _mm256_add_ps(Y,_mm256_mul_ps(vfeps,Fp));
 611             vvdw12           = _mm256_mul_ps(c12_00,VV);
 612             FF               = _mm256_add_ps(Fp,_mm256_mul_ps(vfeps,_mm256_add_ps(G,_mm256_add_ps(Heps,Heps))));
 613             fvdw12           = _mm256_mul_ps(c12_00,FF);
 614             vvdw             = _mm256_add_ps(vvdw12,vvdw6);
 615             fvdw             = _mm256_xor_ps(signbit,_mm256_mul_ps(_mm256_add_ps(fvdw6,fvdw12),_mm256_mul_ps(vftabscale,rinv00)));
 616
 617             /* Update potential sum for this i atom from the interaction with this j atom. */
 618             vvdw             = _mm256_andnot_ps(dummy_mask,vvdw);
 619             vvdwsum          = _mm256_add_ps(vvdwsum,vvdw);
 620
 621             fscal            = fvdw;
 622
 623             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
 624
 625             /* Calculate temporary vectorial force */
 626             tx               = _mm256_mul_ps(fscal,dx00);
 627             ty               = _mm256_mul_ps(fscal,dy00);
 628             tz               = _mm256_mul_ps(fscal,dz00);
 629
 630             /* Update vectorial force */
 631             fix0             = _mm256_add_ps(fix0,tx);
 632             fiy0             = _mm256_add_ps(fiy0,ty);
 633             fiz0             = _mm256_add_ps(fiz0,tz);
 634
 635             fjx0             = _mm256_add_ps(fjx0,tx);
 636             fjy0             = _mm256_add_ps(fjy0,ty);
 637             fjz0             = _mm256_add_ps(fjz0,tz);
 638
 639             /**************************
 640              * CALCULATE INTERACTIONS *
 641              **************************/
 642
 643             /* Compute parameters for interactions between i and j atoms */
 644             qq10             = _mm256_mul_ps(iq1,jq0);
 645
 646             /* COULOMB ELECTROSTATICS */
 647             velec            = _mm256_mul_ps(qq10,rinv10);
 648             felec            = _mm256_mul_ps(velec,rinvsq10);
 649
 650             /* Update potential sum for this i atom from the interaction with this j atom. */
 651             velec            = _mm256_andnot_ps(dummy_mask,velec);
 652             velecsum         = _mm256_add_ps(velecsum,velec);
 653
 654             fscal            = felec;
 655
 656             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
 657
 658             /* Calculate temporary vectorial force */
 659             tx               = _mm256_mul_ps(fscal,dx10);
 660             ty               = _mm256_mul_ps(fscal,dy10);
 661             tz               = _mm256_mul_ps(fscal,dz10);
 662
 663             /* Update vectorial force */
 664             fix1             = _mm256_add_ps(fix1,tx);
 665             fiy1             = _mm256_add_ps(fiy1,ty);
 666             fiz1             = _mm256_add_ps(fiz1,tz);
 667
 668             fjx0             = _mm256_add_ps(fjx0,tx);
 669             fjy0             = _mm256_add_ps(fjy0,ty);
 670             fjz0             = _mm256_add_ps(fjz0,tz);
 671
 672             /**************************
 673              * CALCULATE INTERACTIONS *
 674              **************************/
 675
 676             /* Compute parameters for interactions between i and j atoms */
 677             qq20             = _mm256_mul_ps(iq2,jq0);
 678
 679             /* COULOMB ELECTROSTATICS */
 680             velec            = _mm256_mul_ps(qq20,rinv20);
 681             felec            = _mm256_mul_ps(velec,rinvsq20);
 682
 683             /* Update potential sum for this i atom from the interaction with this j atom. */
 684             velec            = _mm256_andnot_ps(dummy_mask,velec);
 685             velecsum         = _mm256_add_ps(velecsum,velec);
 686
 687             fscal            = felec;
 688
 689             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
 690
 691             /* Calculate temporary vectorial force */
 692             tx               = _mm256_mul_ps(fscal,dx20);
 693             ty               = _mm256_mul_ps(fscal,dy20);
 694             tz               = _mm256_mul_ps(fscal,dz20);
 695
 696             /* Update vectorial force */
 697             fix2             = _mm256_add_ps(fix2,tx);
 698             fiy2             = _mm256_add_ps(fiy2,ty);
 699             fiz2             = _mm256_add_ps(fiz2,tz);
 700
 701             fjx0             = _mm256_add_ps(fjx0,tx);
 702             fjy0             = _mm256_add_ps(fjy0,ty);
 703             fjz0             = _mm256_add_ps(fjz0,tz);
 704
 705             /**************************
 706              * CALCULATE INTERACTIONS *
 707              **************************/
 708
 709             /* Compute parameters for interactions between i and j atoms */
 710             qq30             = _mm256_mul_ps(iq3,jq0);
 711
 712             /* COULOMB ELECTROSTATICS */
 713             velec            = _mm256_mul_ps(qq30,rinv30);
 714             felec            = _mm256_mul_ps(velec,rinvsq30);
 715
 716             /* Update potential sum for this i atom from the interaction with this j atom. */
 717             velec            = _mm256_andnot_ps(dummy_mask,velec);
 718             velecsum         = _mm256_add_ps(velecsum,velec);
 719
 720             fscal            = felec;
 721
 722             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
 723
 724             /* Calculate temporary vectorial force */
 725             tx               = _mm256_mul_ps(fscal,dx30);
 726             ty               = _mm256_mul_ps(fscal,dy30);
 727             tz               = _mm256_mul_ps(fscal,dz30);
 728
 729             /* Update vectorial force */
 730             fix3             = _mm256_add_ps(fix3,tx);
 731             fiy3             = _mm256_add_ps(fiy3,ty);
 732             fiz3             = _mm256_add_ps(fiz3,tz);
 733
 734             fjx0             = _mm256_add_ps(fjx0,tx);
 735             fjy0             = _mm256_add_ps(fjy0,ty);
 736             fjz0             = _mm256_add_ps(fjz0,tz);
 737
 738             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
 739             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
 740             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
 741             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
 742             fjptrE             = (jnrlistE>=0) ? f+j_coord_offsetE : scratch;
 743             fjptrF             = (jnrlistF>=0) ? f+j_coord_offsetF : scratch;
 744             fjptrG             = (jnrlistG>=0) ? f+j_coord_offsetG : scratch;
 745             fjptrH             = (jnrlistH>=0) ? f+j_coord_offsetH : scratch;
 746
 747             gmx_mm256_decrement_1rvec_8ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjptrE,fjptrF,fjptrG,fjptrH,fjx0,fjy0,fjz0);
 748
 749             /* Inner loop uses 141 flops */
 750         }
 751
 752         /* End of innermost loop */
 753
 754         gmx_mm256_update_iforce_4atom_swizzle_ps(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,fix3,fiy3,fiz3,
 755                                                  f+i_coord_offset,fshift+i_shift_offset);
 756
 757         ggid                        = gid[iidx];
 758         /* Update potential energies */
 759         gmx_mm256_update_1pot_ps(velecsum,kernel_data->energygrp_elec+ggid);
 760         gmx_mm256_update_1pot_ps(vvdwsum,kernel_data->energygrp_vdw+ggid);
 761
 762         /* Increment number of inner iterations */
 763         inneriter                  += j_index_end - j_index_start;
 764
 765         /* Outer loop uses 26 flops */
 766     }
 767
 768     /* Increment number of outer iterations */
 769     outeriter        += nri;
 770
 771     /* Update outer/inner flops */
 772
 773     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W4_VF,outeriter*26 + inneriter*141);
 774 }
 775 /*
 776  * Gromacs nonbonded kernel:   nb_kernel_ElecCoul_VdwCSTab_GeomW4P1_F_avx_256_single
 777  * Electrostatics interaction: Coulomb
 778  * VdW interaction:            CubicSplineTable
 779  * Geometry:                   Water4-Particle
 780  * Calculate force/pot:        Force
 781  */
 782 void
 783 nb_kernel_ElecCoul_VdwCSTab_GeomW4P1_F_avx_256_single
 784                     (t_nblist                    * gmx_restrict       nlist,
 785                      rvec                        * gmx_restrict          xx,
 786                      rvec                        * gmx_restrict          ff,
 787                      struct t_forcerec           * gmx_restrict          fr,
 788                      t_mdatoms                   * gmx_restrict     mdatoms,
 789                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
 790                      t_nrnb                      * gmx_restrict        nrnb)
 791 {
 792     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
 793      * just 0 for non-waters.
 794      * Suffixes A,B,C,D,E,F,G,H refer to j loop unrolling done with AVX, e.g. for the eight different
 795      * jnr indices corresponding to data put in the four positions in the SIMD register.
 796      */
 797     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
 798     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
 799     int              jnrA,jnrB,jnrC,jnrD;
 800     int              jnrE,jnrF,jnrG,jnrH;
 801     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
 802     int              jnrlistE,jnrlistF,jnrlistG,jnrlistH;
 803     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
 804     int              j_coord_offsetE,j_coord_offsetF,j_coord_offsetG,j_coord_offsetH;
 805     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
 806     real             rcutoff_scalar;
 807     real             *shiftvec,*fshift,*x,*f;
 808     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD,*fjptrE,*fjptrF,*fjptrG,*fjptrH;
 809     real             scratch[4*DIM];
 810     __m256           tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
 811     real *           vdwioffsetptr0;
 812     __m256           ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
 813     real *           vdwioffsetptr1;
 814     __m256           ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
 815     real *           vdwioffsetptr2;
 816     __m256           ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
 817     real *           vdwioffsetptr3;
 818     __m256           ix3,iy3,iz3,fix3,fiy3,fiz3,iq3,isai3;
 819     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D,vdwjidx0E,vdwjidx0F,vdwjidx0G,vdwjidx0H;
 820     __m256           jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
 821     __m256           dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
 822     __m256           dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
 823     __m256           dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
 824     __m256           dx30,dy30,dz30,rsq30,rinv30,rinvsq30,r30,qq30,c6_30,c12_30;
 825     __m256           velec,felec,velecsum,facel,crf,krf,krf2;
 826     real             *charge;
 827     int              nvdwtype;
 828     __m256           rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
 829     int              *vdwtype;
 830     real             *vdwparam;
 831     __m256           one_sixth   = _mm256_set1_ps(1.0/6.0);
 832     __m256           one_twelfth = _mm256_set1_ps(1.0/12.0);
 833     __m256i          vfitab;
 834     __m128i          vfitab_lo,vfitab_hi;
 835     __m128i          ifour       = _mm_set1_epi32(4);
 836     __m256           rt,vfeps,vftabscale,Y,F,G,H,Heps,Fp,VV,FF;
 837     real             *vftab;
 838     __m256           dummy_mask,cutoff_mask;
 839     __m256           signbit = _mm256_castsi256_ps( _mm256_set1_epi32(0x80000000) );
 840     __m256           one     = _mm256_set1_ps(1.0);
 841     __m256           two     = _mm256_set1_ps(2.0);
 842     x                = xx[0];
 843     f                = ff[0];
 844
 845     nri              = nlist->nri;
 846     iinr             = nlist->iinr;
 847     jindex           = nlist->jindex;
 848     jjnr             = nlist->jjnr;
 849     shiftidx         = nlist->shift;
 850     gid              = nlist->gid;
 851     shiftvec         = fr->shift_vec[0];
 852     fshift           = fr->fshift[0];
 853     facel            = _mm256_set1_ps(fr->ic->epsfac);
 854     charge           = mdatoms->chargeA;
 855     nvdwtype         = fr->ntype;
 856     vdwparam         = fr->nbfp;
 857     vdwtype          = mdatoms->typeA;
 858
 859     vftab            = kernel_data->table_vdw->data;
 860     vftabscale       = _mm256_set1_ps(kernel_data->table_vdw->scale);
 861
 862     /* Setup water-specific parameters */
 863     inr              = nlist->iinr[0];
 864     iq1              = _mm256_mul_ps(facel,_mm256_set1_ps(charge[inr+1]));
 865     iq2              = _mm256_mul_ps(facel,_mm256_set1_ps(charge[inr+2]));
 866     iq3              = _mm256_mul_ps(facel,_mm256_set1_ps(charge[inr+3]));
 867     vdwioffsetptr0   = vdwparam+2*nvdwtype*vdwtype[inr+0];
 868
 869     /* Avoid stupid compiler warnings */
 870     jnrA = jnrB = jnrC = jnrD = jnrE = jnrF = jnrG = jnrH = 0;
 871     j_coord_offsetA = 0;
 872     j_coord_offsetB = 0;
 873     j_coord_offsetC = 0;
 874     j_coord_offsetD = 0;
 875     j_coord_offsetE = 0;
 876     j_coord_offsetF = 0;
 877     j_coord_offsetG = 0;
 878     j_coord_offsetH = 0;
 879
 880     outeriter        = 0;
 881     inneriter        = 0;
 882
 883     for(iidx=0;iidx<4*DIM;iidx++)
 884     {
 885         scratch[iidx] = 0.0;
 886     }
 887
 888     /* Start outer loop over neighborlists */
 889     for(iidx=0; iidx<nri; iidx++)
 890     {
 891         /* Load shift vector for this list */
 892         i_shift_offset   = DIM*shiftidx[iidx];
 893
 894         /* Load limits for loop over neighbors */
 895         j_index_start    = jindex[iidx];
 896         j_index_end      = jindex[iidx+1];
 897
 898         /* Get outer coordinate index */
 899         inr              = iinr[iidx];
 900         i_coord_offset   = DIM*inr;
 901
 902         /* Load i particle coords and add shift vector */
 903         gmx_mm256_load_shift_and_4rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,
 904                                                     &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2,&ix3,&iy3,&iz3);
 905
 906         fix0             = _mm256_setzero_ps();
 907         fiy0             = _mm256_setzero_ps();
 908         fiz0             = _mm256_setzero_ps();
 909         fix1             = _mm256_setzero_ps();
 910         fiy1             = _mm256_setzero_ps();
 911         fiz1             = _mm256_setzero_ps();
 912         fix2             = _mm256_setzero_ps();
 913         fiy2             = _mm256_setzero_ps();
 914         fiz2             = _mm256_setzero_ps();
 915         fix3             = _mm256_setzero_ps();
 916         fiy3             = _mm256_setzero_ps();
 917         fiz3             = _mm256_setzero_ps();
 918
 919         /* Start inner kernel loop */
 920         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+7]>=0; jidx+=8)
 921         {
 922
 923             /* Get j neighbor index, and coordinate index */
 924             jnrA             = jjnr[jidx];
 925             jnrB             = jjnr[jidx+1];
 926             jnrC             = jjnr[jidx+2];
 927             jnrD             = jjnr[jidx+3];
 928             jnrE             = jjnr[jidx+4];
 929             jnrF             = jjnr[jidx+5];
 930             jnrG             = jjnr[jidx+6];
 931             jnrH             = jjnr[jidx+7];
 932             j_coord_offsetA  = DIM*jnrA;
 933             j_coord_offsetB  = DIM*jnrB;
 934             j_coord_offsetC  = DIM*jnrC;
 935             j_coord_offsetD  = DIM*jnrD;
 936             j_coord_offsetE  = DIM*jnrE;
 937             j_coord_offsetF  = DIM*jnrF;
 938             j_coord_offsetG  = DIM*jnrG;
 939             j_coord_offsetH  = DIM*jnrH;
 940
 941             /* load j atom coordinates */
 942             gmx_mm256_load_1rvec_8ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 943                                                  x+j_coord_offsetC,x+j_coord_offsetD,
 944                                                  x+j_coord_offsetE,x+j_coord_offsetF,
 945                                                  x+j_coord_offsetG,x+j_coord_offsetH,
 946                                                  &jx0,&jy0,&jz0);
 947
 948             /* Calculate displacement vector */
 949             dx00             = _mm256_sub_ps(ix0,jx0);
 950             dy00             = _mm256_sub_ps(iy0,jy0);
 951             dz00             = _mm256_sub_ps(iz0,jz0);
 952             dx10             = _mm256_sub_ps(ix1,jx0);
 953             dy10             = _mm256_sub_ps(iy1,jy0);
 954             dz10             = _mm256_sub_ps(iz1,jz0);
 955             dx20             = _mm256_sub_ps(ix2,jx0);
 956             dy20             = _mm256_sub_ps(iy2,jy0);
 957             dz20             = _mm256_sub_ps(iz2,jz0);
 958             dx30             = _mm256_sub_ps(ix3,jx0);
 959             dy30             = _mm256_sub_ps(iy3,jy0);
 960             dz30             = _mm256_sub_ps(iz3,jz0);
 961
 962             /* Calculate squared distance and things based on it */
 963             rsq00            = gmx_mm256_calc_rsq_ps(dx00,dy00,dz00);
 964             rsq10            = gmx_mm256_calc_rsq_ps(dx10,dy10,dz10);
 965             rsq20            = gmx_mm256_calc_rsq_ps(dx20,dy20,dz20);
 966             rsq30            = gmx_mm256_calc_rsq_ps(dx30,dy30,dz30);
 967
 968             rinv00           = avx256_invsqrt_f(rsq00);
 969             rinv10           = avx256_invsqrt_f(rsq10);
 970             rinv20           = avx256_invsqrt_f(rsq20);
 971             rinv30           = avx256_invsqrt_f(rsq30);
 972
 973             rinvsq10         = _mm256_mul_ps(rinv10,rinv10);
 974             rinvsq20         = _mm256_mul_ps(rinv20,rinv20);
 975             rinvsq30         = _mm256_mul_ps(rinv30,rinv30);
 976
 977             /* Load parameters for j particles */
 978             jq0              = gmx_mm256_load_8real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 979                                                                  charge+jnrC+0,charge+jnrD+0,
 980                                                                  charge+jnrE+0,charge+jnrF+0,
 981                                                                  charge+jnrG+0,charge+jnrH+0);
 982             vdwjidx0A        = 2*vdwtype[jnrA+0];
 983             vdwjidx0B        = 2*vdwtype[jnrB+0];
 984             vdwjidx0C        = 2*vdwtype[jnrC+0];
 985             vdwjidx0D        = 2*vdwtype[jnrD+0];
 986             vdwjidx0E        = 2*vdwtype[jnrE+0];
 987             vdwjidx0F        = 2*vdwtype[jnrF+0];
 988             vdwjidx0G        = 2*vdwtype[jnrG+0];
 989             vdwjidx0H        = 2*vdwtype[jnrH+0];
 990
 991             fjx0             = _mm256_setzero_ps();
 992             fjy0             = _mm256_setzero_ps();
 993             fjz0             = _mm256_setzero_ps();
 994
 995             /**************************
 996              * CALCULATE INTERACTIONS *
 997              **************************/
 998
 999             r00              = _mm256_mul_ps(rsq00,rinv00);
1000
1001             /* Compute parameters for interactions between i and j atoms */
1002             gmx_mm256_load_8pair_swizzle_ps(vdwioffsetptr0+vdwjidx0A,
1003                                             vdwioffsetptr0+vdwjidx0B,
1004                                             vdwioffsetptr0+vdwjidx0C,
1005                                             vdwioffsetptr0+vdwjidx0D,
1006                                             vdwioffsetptr0+vdwjidx0E,
1007                                             vdwioffsetptr0+vdwjidx0F,
1008                                             vdwioffsetptr0+vdwjidx0G,
1009                                             vdwioffsetptr0+vdwjidx0H,
1010                                             &c6_00,&c12_00);
1011
1012             /* Calculate table index by multiplying r with table scale and truncate to integer */
1013             rt               = _mm256_mul_ps(r00,vftabscale);
1014             vfitab           = _mm256_cvttps_epi32(rt);
1015             vfeps            = _mm256_sub_ps(rt,_mm256_round_ps(rt, _MM_FROUND_FLOOR));
1016             /*         AVX1 does not support 256-bit integer operations, so now we go to 128-bit mode... */
1017             vfitab_lo        = _mm256_extractf128_si256(vfitab,0x0);
1018             vfitab_hi        = _mm256_extractf128_si256(vfitab,0x1);
1019             vfitab_lo        = _mm_slli_epi32(vfitab_lo,3);
1020             vfitab_hi        = _mm_slli_epi32(vfitab_hi,3);
1021
1022             /* CUBIC SPLINE TABLE DISPERSION */
1023             Y                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,0)),
1024                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,0)));
1025             F                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,1)),
1026                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,1)));
1027             G                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,2)),
1028                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,2)));
1029             H                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,3)),
1030                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,3)));
1031             GMX_MM256_HALFTRANSPOSE4_PS(Y,F,G,H);
1032             Heps             = _mm256_mul_ps(vfeps,H);
1033             Fp               = _mm256_add_ps(F,_mm256_mul_ps(vfeps,_mm256_add_ps(G,Heps)));
1034             FF               = _mm256_add_ps(Fp,_mm256_mul_ps(vfeps,_mm256_add_ps(G,_mm256_add_ps(Heps,Heps))));
1035             fvdw6            = _mm256_mul_ps(c6_00,FF);
1036
1037             /* CUBIC SPLINE TABLE REPULSION */
1038             vfitab_lo        = _mm_add_epi32(vfitab_lo,ifour);
1039             vfitab_hi        = _mm_add_epi32(vfitab_hi,ifour);
1040             Y                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,0)),
1041                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,0)));
1042             F                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,1)),
1043                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,1)));
1044             G                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,2)),
1045                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,2)));
1046             H                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,3)),
1047                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,3)));
1048             GMX_MM256_HALFTRANSPOSE4_PS(Y,F,G,H);
1049             Heps             = _mm256_mul_ps(vfeps,H);
1050             Fp               = _mm256_add_ps(F,_mm256_mul_ps(vfeps,_mm256_add_ps(G,Heps)));
1051             FF               = _mm256_add_ps(Fp,_mm256_mul_ps(vfeps,_mm256_add_ps(G,_mm256_add_ps(Heps,Heps))));
1052             fvdw12           = _mm256_mul_ps(c12_00,FF);
1053             fvdw             = _mm256_xor_ps(signbit,_mm256_mul_ps(_mm256_add_ps(fvdw6,fvdw12),_mm256_mul_ps(vftabscale,rinv00)));
1054
1055             fscal            = fvdw;
1056
1057             /* Calculate temporary vectorial force */
1058             tx               = _mm256_mul_ps(fscal,dx00);
1059             ty               = _mm256_mul_ps(fscal,dy00);
1060             tz               = _mm256_mul_ps(fscal,dz00);
1061
1062             /* Update vectorial force */
1063             fix0             = _mm256_add_ps(fix0,tx);
1064             fiy0             = _mm256_add_ps(fiy0,ty);
1065             fiz0             = _mm256_add_ps(fiz0,tz);
1066
1067             fjx0             = _mm256_add_ps(fjx0,tx);
1068             fjy0             = _mm256_add_ps(fjy0,ty);
1069             fjz0             = _mm256_add_ps(fjz0,tz);
1070
1071             /**************************
1072              * CALCULATE INTERACTIONS *
1073              **************************/
1074
1075             /* Compute parameters for interactions between i and j atoms */
1076             qq10             = _mm256_mul_ps(iq1,jq0);
1077
1078             /* COULOMB ELECTROSTATICS */
1079             velec            = _mm256_mul_ps(qq10,rinv10);
1080             felec            = _mm256_mul_ps(velec,rinvsq10);
1081
1082             fscal            = felec;
1083
1084             /* Calculate temporary vectorial force */
1085             tx               = _mm256_mul_ps(fscal,dx10);
1086             ty               = _mm256_mul_ps(fscal,dy10);
1087             tz               = _mm256_mul_ps(fscal,dz10);
1088
1089             /* Update vectorial force */
1090             fix1             = _mm256_add_ps(fix1,tx);
1091             fiy1             = _mm256_add_ps(fiy1,ty);
1092             fiz1             = _mm256_add_ps(fiz1,tz);
1093
1094             fjx0             = _mm256_add_ps(fjx0,tx);
1095             fjy0             = _mm256_add_ps(fjy0,ty);
1096             fjz0             = _mm256_add_ps(fjz0,tz);
1097
1098             /**************************
1099              * CALCULATE INTERACTIONS *
1100              **************************/
1101
1102             /* Compute parameters for interactions between i and j atoms */
1103             qq20             = _mm256_mul_ps(iq2,jq0);
1104
1105             /* COULOMB ELECTROSTATICS */
1106             velec            = _mm256_mul_ps(qq20,rinv20);
1107             felec            = _mm256_mul_ps(velec,rinvsq20);
1108
1109             fscal            = felec;
1110
1111             /* Calculate temporary vectorial force */
1112             tx               = _mm256_mul_ps(fscal,dx20);
1113             ty               = _mm256_mul_ps(fscal,dy20);
1114             tz               = _mm256_mul_ps(fscal,dz20);
1115
1116             /* Update vectorial force */
1117             fix2             = _mm256_add_ps(fix2,tx);
1118             fiy2             = _mm256_add_ps(fiy2,ty);
1119             fiz2             = _mm256_add_ps(fiz2,tz);
1120
1121             fjx0             = _mm256_add_ps(fjx0,tx);
1122             fjy0             = _mm256_add_ps(fjy0,ty);
1123             fjz0             = _mm256_add_ps(fjz0,tz);
1124
1125             /**************************
1126              * CALCULATE INTERACTIONS *
1127              **************************/
1128
1129             /* Compute parameters for interactions between i and j atoms */
1130             qq30             = _mm256_mul_ps(iq3,jq0);
1131
1132             /* COULOMB ELECTROSTATICS */
1133             velec            = _mm256_mul_ps(qq30,rinv30);
1134             felec            = _mm256_mul_ps(velec,rinvsq30);
1135
1136             fscal            = felec;
1137
1138             /* Calculate temporary vectorial force */
1139             tx               = _mm256_mul_ps(fscal,dx30);
1140             ty               = _mm256_mul_ps(fscal,dy30);
1141             tz               = _mm256_mul_ps(fscal,dz30);
1142
1143             /* Update vectorial force */
1144             fix3             = _mm256_add_ps(fix3,tx);
1145             fiy3             = _mm256_add_ps(fiy3,ty);
1146             fiz3             = _mm256_add_ps(fiz3,tz);
1147
1148             fjx0             = _mm256_add_ps(fjx0,tx);
1149             fjy0             = _mm256_add_ps(fjy0,ty);
1150             fjz0             = _mm256_add_ps(fjz0,tz);
1151
1152             fjptrA             = f+j_coord_offsetA;
1153             fjptrB             = f+j_coord_offsetB;
1154             fjptrC             = f+j_coord_offsetC;
1155             fjptrD             = f+j_coord_offsetD;
1156             fjptrE             = f+j_coord_offsetE;
1157             fjptrF             = f+j_coord_offsetF;
1158             fjptrG             = f+j_coord_offsetG;
1159             fjptrH             = f+j_coord_offsetH;
1160
1161             gmx_mm256_decrement_1rvec_8ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjptrE,fjptrF,fjptrG,fjptrH,fjx0,fjy0,fjz0);
1162
1163             /* Inner loop uses 129 flops */
1164         }
1165
1166         if(jidx<j_index_end)
1167         {
1168
1169             /* Get j neighbor index, and coordinate index */
1170             jnrlistA         = jjnr[jidx];
1171             jnrlistB         = jjnr[jidx+1];
1172             jnrlistC         = jjnr[jidx+2];
1173             jnrlistD         = jjnr[jidx+3];
1174             jnrlistE         = jjnr[jidx+4];
1175             jnrlistF         = jjnr[jidx+5];
1176             jnrlistG         = jjnr[jidx+6];
1177             jnrlistH         = jjnr[jidx+7];
1178             /* Sign of each element will be negative for non-real atoms.
1179              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
1180              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
1181              */
1182             dummy_mask = gmx_mm256_set_m128(gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx+4)),_mm_setzero_si128())),
1183                                             gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128())));
1184
1185             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
1186             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
1187             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
1188             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
1189             jnrE       = (jnrlistE>=0) ? jnrlistE : 0;
1190             jnrF       = (jnrlistF>=0) ? jnrlistF : 0;
1191             jnrG       = (jnrlistG>=0) ? jnrlistG : 0;
1192             jnrH       = (jnrlistH>=0) ? jnrlistH : 0;
1193             j_coord_offsetA  = DIM*jnrA;
1194             j_coord_offsetB  = DIM*jnrB;
1195             j_coord_offsetC  = DIM*jnrC;
1196             j_coord_offsetD  = DIM*jnrD;
1197             j_coord_offsetE  = DIM*jnrE;
1198             j_coord_offsetF  = DIM*jnrF;
1199             j_coord_offsetG  = DIM*jnrG;
1200             j_coord_offsetH  = DIM*jnrH;
1201
1202             /* load j atom coordinates */
1203             gmx_mm256_load_1rvec_8ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
1204                                                  x+j_coord_offsetC,x+j_coord_offsetD,
1205                                                  x+j_coord_offsetE,x+j_coord_offsetF,
1206                                                  x+j_coord_offsetG,x+j_coord_offsetH,
1207                                                  &jx0,&jy0,&jz0);
1208
1209             /* Calculate displacement vector */
1210             dx00             = _mm256_sub_ps(ix0,jx0);
1211             dy00             = _mm256_sub_ps(iy0,jy0);
1212             dz00             = _mm256_sub_ps(iz0,jz0);
1213             dx10             = _mm256_sub_ps(ix1,jx0);
1214             dy10             = _mm256_sub_ps(iy1,jy0);
1215             dz10             = _mm256_sub_ps(iz1,jz0);
1216             dx20             = _mm256_sub_ps(ix2,jx0);
1217             dy20             = _mm256_sub_ps(iy2,jy0);
1218             dz20             = _mm256_sub_ps(iz2,jz0);
1219             dx30             = _mm256_sub_ps(ix3,jx0);
1220             dy30             = _mm256_sub_ps(iy3,jy0);
1221             dz30             = _mm256_sub_ps(iz3,jz0);
1222
1223             /* Calculate squared distance and things based on it */
1224             rsq00            = gmx_mm256_calc_rsq_ps(dx00,dy00,dz00);
1225             rsq10            = gmx_mm256_calc_rsq_ps(dx10,dy10,dz10);
1226             rsq20            = gmx_mm256_calc_rsq_ps(dx20,dy20,dz20);
1227             rsq30            = gmx_mm256_calc_rsq_ps(dx30,dy30,dz30);
1228
1229             rinv00           = avx256_invsqrt_f(rsq00);
1230             rinv10           = avx256_invsqrt_f(rsq10);
1231             rinv20           = avx256_invsqrt_f(rsq20);
1232             rinv30           = avx256_invsqrt_f(rsq30);
1233
1234             rinvsq10         = _mm256_mul_ps(rinv10,rinv10);
1235             rinvsq20         = _mm256_mul_ps(rinv20,rinv20);
1236             rinvsq30         = _mm256_mul_ps(rinv30,rinv30);
1237
1238             /* Load parameters for j particles */
1239             jq0              = gmx_mm256_load_8real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
1240                                                                  charge+jnrC+0,charge+jnrD+0,
1241                                                                  charge+jnrE+0,charge+jnrF+0,
1242                                                                  charge+jnrG+0,charge+jnrH+0);
1243             vdwjidx0A        = 2*vdwtype[jnrA+0];
1244             vdwjidx0B        = 2*vdwtype[jnrB+0];
1245             vdwjidx0C        = 2*vdwtype[jnrC+0];
1246             vdwjidx0D        = 2*vdwtype[jnrD+0];
1247             vdwjidx0E        = 2*vdwtype[jnrE+0];
1248             vdwjidx0F        = 2*vdwtype[jnrF+0];
1249             vdwjidx0G        = 2*vdwtype[jnrG+0];
1250             vdwjidx0H        = 2*vdwtype[jnrH+0];
1251
1252             fjx0             = _mm256_setzero_ps();
1253             fjy0             = _mm256_setzero_ps();
1254             fjz0             = _mm256_setzero_ps();
1255
1256             /**************************
1257              * CALCULATE INTERACTIONS *
1258              **************************/
1259
1260             r00              = _mm256_mul_ps(rsq00,rinv00);
1261             r00              = _mm256_andnot_ps(dummy_mask,r00);
1262
1263             /* Compute parameters for interactions between i and j atoms */
1264             gmx_mm256_load_8pair_swizzle_ps(vdwioffsetptr0+vdwjidx0A,
1265                                             vdwioffsetptr0+vdwjidx0B,
1266                                             vdwioffsetptr0+vdwjidx0C,
1267                                             vdwioffsetptr0+vdwjidx0D,
1268                                             vdwioffsetptr0+vdwjidx0E,
1269                                             vdwioffsetptr0+vdwjidx0F,
1270                                             vdwioffsetptr0+vdwjidx0G,
1271                                             vdwioffsetptr0+vdwjidx0H,
1272                                             &c6_00,&c12_00);
1273
1274             /* Calculate table index by multiplying r with table scale and truncate to integer */
1275             rt               = _mm256_mul_ps(r00,vftabscale);
1276             vfitab           = _mm256_cvttps_epi32(rt);
1277             vfeps            = _mm256_sub_ps(rt,_mm256_round_ps(rt, _MM_FROUND_FLOOR));
1278             /*         AVX1 does not support 256-bit integer operations, so now we go to 128-bit mode... */
1279             vfitab_lo        = _mm256_extractf128_si256(vfitab,0x0);
1280             vfitab_hi        = _mm256_extractf128_si256(vfitab,0x1);
1281             vfitab_lo        = _mm_slli_epi32(vfitab_lo,3);
1282             vfitab_hi        = _mm_slli_epi32(vfitab_hi,3);
1283
1284             /* CUBIC SPLINE TABLE DISPERSION */
1285             Y                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,0)),
1286                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,0)));
1287             F                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,1)),
1288                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,1)));
1289             G                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,2)),
1290                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,2)));
1291             H                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,3)),
1292                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,3)));
1293             GMX_MM256_HALFTRANSPOSE4_PS(Y,F,G,H);
1294             Heps             = _mm256_mul_ps(vfeps,H);
1295             Fp               = _mm256_add_ps(F,_mm256_mul_ps(vfeps,_mm256_add_ps(G,Heps)));
1296             FF               = _mm256_add_ps(Fp,_mm256_mul_ps(vfeps,_mm256_add_ps(G,_mm256_add_ps(Heps,Heps))));
1297             fvdw6            = _mm256_mul_ps(c6_00,FF);
1298
1299             /* CUBIC SPLINE TABLE REPULSION */
1300             vfitab_lo        = _mm_add_epi32(vfitab_lo,ifour);
1301             vfitab_hi        = _mm_add_epi32(vfitab_hi,ifour);
1302             Y                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,0)),
1303                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,0)));
1304             F                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,1)),
1305                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,1)));
1306             G                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,2)),
1307                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,2)));
1308             H                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,3)),
1309                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,3)));
1310             GMX_MM256_HALFTRANSPOSE4_PS(Y,F,G,H);
1311             Heps             = _mm256_mul_ps(vfeps,H);
1312             Fp               = _mm256_add_ps(F,_mm256_mul_ps(vfeps,_mm256_add_ps(G,Heps)));
1313             FF               = _mm256_add_ps(Fp,_mm256_mul_ps(vfeps,_mm256_add_ps(G,_mm256_add_ps(Heps,Heps))));
1314             fvdw12           = _mm256_mul_ps(c12_00,FF);
1315             fvdw             = _mm256_xor_ps(signbit,_mm256_mul_ps(_mm256_add_ps(fvdw6,fvdw12),_mm256_mul_ps(vftabscale,rinv00)));
1316
1317             fscal            = fvdw;
1318
1319             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
1320
1321             /* Calculate temporary vectorial force */
1322             tx               = _mm256_mul_ps(fscal,dx00);
1323             ty               = _mm256_mul_ps(fscal,dy00);
1324             tz               = _mm256_mul_ps(fscal,dz00);
1325
1326             /* Update vectorial force */
1327             fix0             = _mm256_add_ps(fix0,tx);
1328             fiy0             = _mm256_add_ps(fiy0,ty);
1329             fiz0             = _mm256_add_ps(fiz0,tz);
1330
1331             fjx0             = _mm256_add_ps(fjx0,tx);
1332             fjy0             = _mm256_add_ps(fjy0,ty);
1333             fjz0             = _mm256_add_ps(fjz0,tz);
1334
1335             /**************************
1336              * CALCULATE INTERACTIONS *
1337              **************************/
1338
1339             /* Compute parameters for interactions between i and j atoms */
1340             qq10             = _mm256_mul_ps(iq1,jq0);
1341
1342             /* COULOMB ELECTROSTATICS */
1343             velec            = _mm256_mul_ps(qq10,rinv10);
1344             felec            = _mm256_mul_ps(velec,rinvsq10);
1345
1346             fscal            = felec;
1347
1348             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
1349
1350             /* Calculate temporary vectorial force */
1351             tx               = _mm256_mul_ps(fscal,dx10);
1352             ty               = _mm256_mul_ps(fscal,dy10);
1353             tz               = _mm256_mul_ps(fscal,dz10);
1354
1355             /* Update vectorial force */
1356             fix1             = _mm256_add_ps(fix1,tx);
1357             fiy1             = _mm256_add_ps(fiy1,ty);
1358             fiz1             = _mm256_add_ps(fiz1,tz);
1359
1360             fjx0             = _mm256_add_ps(fjx0,tx);
1361             fjy0             = _mm256_add_ps(fjy0,ty);
1362             fjz0             = _mm256_add_ps(fjz0,tz);
1363
1364             /**************************
1365              * CALCULATE INTERACTIONS *
1366              **************************/
1367
1368             /* Compute parameters for interactions between i and j atoms */
1369             qq20             = _mm256_mul_ps(iq2,jq0);
1370
1371             /* COULOMB ELECTROSTATICS */
1372             velec            = _mm256_mul_ps(qq20,rinv20);
1373             felec            = _mm256_mul_ps(velec,rinvsq20);
1374
1375             fscal            = felec;
1376
1377             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
1378
1379             /* Calculate temporary vectorial force */
1380             tx               = _mm256_mul_ps(fscal,dx20);
1381             ty               = _mm256_mul_ps(fscal,dy20);
1382             tz               = _mm256_mul_ps(fscal,dz20);
1383
1384             /* Update vectorial force */
1385             fix2             = _mm256_add_ps(fix2,tx);
1386             fiy2             = _mm256_add_ps(fiy2,ty);
1387             fiz2             = _mm256_add_ps(fiz2,tz);
1388
1389             fjx0             = _mm256_add_ps(fjx0,tx);
1390             fjy0             = _mm256_add_ps(fjy0,ty);
1391             fjz0             = _mm256_add_ps(fjz0,tz);
1392
1393             /**************************
1394              * CALCULATE INTERACTIONS *
1395              **************************/
1396
1397             /* Compute parameters for interactions between i and j atoms */
1398             qq30             = _mm256_mul_ps(iq3,jq0);
1399
1400             /* COULOMB ELECTROSTATICS */
1401             velec            = _mm256_mul_ps(qq30,rinv30);
1402             felec            = _mm256_mul_ps(velec,rinvsq30);
1403
1404             fscal            = felec;
1405
1406             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
1407
1408             /* Calculate temporary vectorial force */
1409             tx               = _mm256_mul_ps(fscal,dx30);
1410             ty               = _mm256_mul_ps(fscal,dy30);
1411             tz               = _mm256_mul_ps(fscal,dz30);
1412
1413             /* Update vectorial force */
1414             fix3             = _mm256_add_ps(fix3,tx);
1415             fiy3             = _mm256_add_ps(fiy3,ty);
1416             fiz3             = _mm256_add_ps(fiz3,tz);
1417
1418             fjx0             = _mm256_add_ps(fjx0,tx);
1419             fjy0             = _mm256_add_ps(fjy0,ty);
1420             fjz0             = _mm256_add_ps(fjz0,tz);
1421
1422             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
1423             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
1424             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
1425             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
1426             fjptrE             = (jnrlistE>=0) ? f+j_coord_offsetE : scratch;
1427             fjptrF             = (jnrlistF>=0) ? f+j_coord_offsetF : scratch;
1428             fjptrG             = (jnrlistG>=0) ? f+j_coord_offsetG : scratch;
1429             fjptrH             = (jnrlistH>=0) ? f+j_coord_offsetH : scratch;
1430
1431             gmx_mm256_decrement_1rvec_8ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjptrE,fjptrF,fjptrG,fjptrH,fjx0,fjy0,fjz0);
1432
1433             /* Inner loop uses 130 flops */
1434         }
1435
1436         /* End of innermost loop */
1437
1438         gmx_mm256_update_iforce_4atom_swizzle_ps(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,fix3,fiy3,fiz3,
1439                                                  f+i_coord_offset,fshift+i_shift_offset);
1440
1441         /* Increment number of inner iterations */
1442         inneriter                  += j_index_end - j_index_start;
1443
1444         /* Outer loop uses 24 flops */
1445     }
1446
1447     /* Increment number of outer iterations */
1448     outeriter        += nri;
1449
1450     /* Update outer/inner flops */
1451
1452     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W4_F,outeriter*24 + inneriter*130);
1453 }